xref: /openbmc/qemu/blockdev.c (revision 7200fb21)
1 /*
2  * QEMU host block devices
3  *
4  * Copyright (c) 2003-2008 Fabrice Bellard
5  *
6  * This work is licensed under the terms of the GNU GPL, version 2 or
7  * later.  See the COPYING file in the top-level directory.
8  *
9  * This file incorporates work covered by the following copyright and
10  * permission notice:
11  *
12  * Copyright (c) 2003-2008 Fabrice Bellard
13  *
14  * Permission is hereby granted, free of charge, to any person obtaining a copy
15  * of this software and associated documentation files (the "Software"), to deal
16  * in the Software without restriction, including without limitation the rights
17  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
18  * copies of the Software, and to permit persons to whom the Software is
19  * furnished to do so, subject to the following conditions:
20  *
21  * The above copyright notice and this permission notice shall be included in
22  * all copies or substantial portions of the Software.
23  *
24  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
25  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
26  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
27  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
28  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
29  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
30  * THE SOFTWARE.
31  */
32 
33 #include "qemu/osdep.h"
34 #include "sysemu/block-backend.h"
35 #include "sysemu/blockdev.h"
36 #include "hw/block/block.h"
37 #include "block/blockjob.h"
38 #include "block/dirty-bitmap.h"
39 #include "block/qdict.h"
40 #include "block/throttle-groups.h"
41 #include "monitor/monitor.h"
42 #include "qemu/error-report.h"
43 #include "qemu/option.h"
44 #include "qemu/qemu-print.h"
45 #include "qemu/config-file.h"
46 #include "qapi/qapi-commands-block.h"
47 #include "qapi/qapi-commands-transaction.h"
48 #include "qapi/qapi-visit-block-core.h"
49 #include "qapi/qmp/qdict.h"
50 #include "qapi/qmp/qnum.h"
51 #include "qapi/qmp/qstring.h"
52 #include "qapi/error.h"
53 #include "qapi/qmp/qerror.h"
54 #include "qapi/qmp/qlist.h"
55 #include "qapi/qobject-output-visitor.h"
56 #include "sysemu/sysemu.h"
57 #include "sysemu/iothread.h"
58 #include "block/block_int.h"
59 #include "block/trace.h"
60 #include "sysemu/runstate.h"
61 #include "sysemu/replay.h"
62 #include "qemu/cutils.h"
63 #include "qemu/help_option.h"
64 #include "qemu/main-loop.h"
65 #include "qemu/throttle-options.h"
66 
67 /* Protected by BQL */
68 QTAILQ_HEAD(, BlockDriverState) monitor_bdrv_states =
69     QTAILQ_HEAD_INITIALIZER(monitor_bdrv_states);
70 
71 void bdrv_set_monitor_owned(BlockDriverState *bs)
72 {
73     GLOBAL_STATE_CODE();
74     QTAILQ_INSERT_TAIL(&monitor_bdrv_states, bs, monitor_list);
75 }
76 
77 static const char *const if_name[IF_COUNT] = {
78     [IF_NONE] = "none",
79     [IF_IDE] = "ide",
80     [IF_SCSI] = "scsi",
81     [IF_FLOPPY] = "floppy",
82     [IF_PFLASH] = "pflash",
83     [IF_MTD] = "mtd",
84     [IF_SD] = "sd",
85     [IF_VIRTIO] = "virtio",
86     [IF_XEN] = "xen",
87 };
88 
89 static int if_max_devs[IF_COUNT] = {
90     /*
91      * Do not change these numbers!  They govern how drive option
92      * index maps to unit and bus.  That mapping is ABI.
93      *
94      * All controllers used to implement if=T drives need to support
95      * if_max_devs[T] units, for any T with if_max_devs[T] != 0.
96      * Otherwise, some index values map to "impossible" bus, unit
97      * values.
98      *
99      * For instance, if you change [IF_SCSI] to 255, -drive
100      * if=scsi,index=12 no longer means bus=1,unit=5, but
101      * bus=0,unit=12.  With an lsi53c895a controller (7 units max),
102      * the drive can't be set up.  Regression.
103      */
104     [IF_IDE] = 2,
105     [IF_SCSI] = 7,
106 };
107 
108 /**
109  * Boards may call this to offer board-by-board overrides
110  * of the default, global values.
111  */
112 void override_max_devs(BlockInterfaceType type, int max_devs)
113 {
114     BlockBackend *blk;
115     DriveInfo *dinfo;
116 
117     GLOBAL_STATE_CODE();
118 
119     if (max_devs <= 0) {
120         return;
121     }
122 
123     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
124         dinfo = blk_legacy_dinfo(blk);
125         if (dinfo->type == type) {
126             fprintf(stderr, "Cannot override units-per-bus property of"
127                     " the %s interface, because a drive of that type has"
128                     " already been added.\n", if_name[type]);
129             g_assert_not_reached();
130         }
131     }
132 
133     if_max_devs[type] = max_devs;
134 }
135 
136 /*
137  * We automatically delete the drive when a device using it gets
138  * unplugged.  Questionable feature, but we can't just drop it.
139  * Device models call blockdev_mark_auto_del() to schedule the
140  * automatic deletion, and generic qdev code calls blockdev_auto_del()
141  * when deletion is actually safe.
142  */
143 void blockdev_mark_auto_del(BlockBackend *blk)
144 {
145     DriveInfo *dinfo = blk_legacy_dinfo(blk);
146     BlockJob *job;
147 
148     GLOBAL_STATE_CODE();
149 
150     if (!dinfo) {
151         return;
152     }
153 
154     JOB_LOCK_GUARD();
155 
156     do {
157         job = block_job_next_locked(NULL);
158         while (job && (job->job.cancelled ||
159                        job->job.deferred_to_main_loop ||
160                        !block_job_has_bdrv(job, blk_bs(blk))))
161         {
162             job = block_job_next_locked(job);
163         }
164         if (job) {
165             /*
166              * This drops the job lock temporarily and polls, so we need to
167              * restart processing the list from the start after this.
168              */
169             job_cancel_locked(&job->job, false);
170         }
171     } while (job);
172 
173     dinfo->auto_del = 1;
174 }
175 
176 void blockdev_auto_del(BlockBackend *blk)
177 {
178     DriveInfo *dinfo = blk_legacy_dinfo(blk);
179     GLOBAL_STATE_CODE();
180 
181     if (dinfo && dinfo->auto_del) {
182         monitor_remove_blk(blk);
183         blk_unref(blk);
184     }
185 }
186 
187 static int drive_index_to_bus_id(BlockInterfaceType type, int index)
188 {
189     int max_devs = if_max_devs[type];
190     return max_devs ? index / max_devs : 0;
191 }
192 
193 static int drive_index_to_unit_id(BlockInterfaceType type, int index)
194 {
195     int max_devs = if_max_devs[type];
196     return max_devs ? index % max_devs : index;
197 }
198 
199 QemuOpts *drive_add(BlockInterfaceType type, int index, const char *file,
200                     const char *optstr)
201 {
202     QemuOpts *opts;
203 
204     GLOBAL_STATE_CODE();
205 
206     opts = qemu_opts_parse_noisily(qemu_find_opts("drive"), optstr, false);
207     if (!opts) {
208         return NULL;
209     }
210     if (type != IF_DEFAULT) {
211         qemu_opt_set(opts, "if", if_name[type], &error_abort);
212     }
213     if (index >= 0) {
214         qemu_opt_set_number(opts, "index", index, &error_abort);
215     }
216     if (file)
217         qemu_opt_set(opts, "file", file, &error_abort);
218     return opts;
219 }
220 
221 DriveInfo *drive_get(BlockInterfaceType type, int bus, int unit)
222 {
223     BlockBackend *blk;
224     DriveInfo *dinfo;
225 
226     GLOBAL_STATE_CODE();
227 
228     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
229         dinfo = blk_legacy_dinfo(blk);
230         if (dinfo && dinfo->type == type
231             && dinfo->bus == bus && dinfo->unit == unit) {
232             return dinfo;
233         }
234     }
235 
236     return NULL;
237 }
238 
239 /*
240  * Check board claimed all -drive that are meant to be claimed.
241  * Fatal error if any remain unclaimed.
242  */
243 void drive_check_orphaned(void)
244 {
245     BlockBackend *blk;
246     DriveInfo *dinfo;
247     Location loc;
248     bool orphans = false;
249 
250     GLOBAL_STATE_CODE();
251 
252     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
253         dinfo = blk_legacy_dinfo(blk);
254         /*
255          * Ignore default drives, because we create certain default
256          * drives unconditionally, then leave them unclaimed.  Not the
257          * users fault.
258          * Ignore IF_VIRTIO 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     BlockDriverState *bs;
666     int bdrv_flags = 0;
667 
668     GLOBAL_STATE_CODE();
669     /* bdrv_open() defaults to the values in bdrv_flags (for compatibility
670      * with other callers) rather than what we want as the real defaults.
671      * Apply the defaults here instead. */
672     qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_DIRECT, "off");
673     qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_NO_FLUSH, "off");
674     qdict_set_default_str(bs_opts, BDRV_OPT_READ_ONLY, "off");
675 
676     if (runstate_check(RUN_STATE_INMIGRATE)) {
677         bdrv_flags |= BDRV_O_INACTIVE;
678     }
679 
680     aio_context_acquire(qemu_get_aio_context());
681     bs = bdrv_open(NULL, NULL, bs_opts, bdrv_flags, errp);
682     aio_context_release(qemu_get_aio_context());
683 
684     return bs;
685 }
686 
687 void blockdev_close_all_bdrv_states(void)
688 {
689     BlockDriverState *bs, *next_bs;
690 
691     GLOBAL_STATE_CODE();
692     QTAILQ_FOREACH_SAFE(bs, &monitor_bdrv_states, monitor_list, next_bs) {
693         AioContext *ctx = bdrv_get_aio_context(bs);
694 
695         aio_context_acquire(ctx);
696         bdrv_unref(bs);
697         aio_context_release(ctx);
698     }
699 }
700 
701 /* Iterates over the list of monitor-owned BlockDriverStates */
702 BlockDriverState *bdrv_next_monitor_owned(BlockDriverState *bs)
703 {
704     GLOBAL_STATE_CODE();
705     return bs ? QTAILQ_NEXT(bs, monitor_list)
706               : QTAILQ_FIRST(&monitor_bdrv_states);
707 }
708 
709 static bool qemu_opt_rename(QemuOpts *opts, const char *from, const char *to,
710                             Error **errp)
711 {
712     const char *value;
713 
714     value = qemu_opt_get(opts, from);
715     if (value) {
716         if (qemu_opt_find(opts, to)) {
717             error_setg(errp, "'%s' and its alias '%s' can't be used at the "
718                        "same time", to, from);
719             return false;
720         }
721     }
722 
723     /* rename all items in opts */
724     while ((value = qemu_opt_get(opts, from))) {
725         qemu_opt_set(opts, to, value, &error_abort);
726         qemu_opt_unset(opts, from);
727     }
728     return true;
729 }
730 
731 QemuOptsList qemu_legacy_drive_opts = {
732     .name = "drive",
733     .head = QTAILQ_HEAD_INITIALIZER(qemu_legacy_drive_opts.head),
734     .desc = {
735         {
736             .name = "bus",
737             .type = QEMU_OPT_NUMBER,
738             .help = "bus number",
739         },{
740             .name = "unit",
741             .type = QEMU_OPT_NUMBER,
742             .help = "unit number (i.e. lun for scsi)",
743         },{
744             .name = "index",
745             .type = QEMU_OPT_NUMBER,
746             .help = "index number",
747         },{
748             .name = "media",
749             .type = QEMU_OPT_STRING,
750             .help = "media type (disk, cdrom)",
751         },{
752             .name = "if",
753             .type = QEMU_OPT_STRING,
754             .help = "interface (ide, scsi, sd, mtd, floppy, pflash, virtio)",
755         },{
756             .name = "file",
757             .type = QEMU_OPT_STRING,
758             .help = "file name",
759         },
760 
761         /* Options that are passed on, but have special semantics with -drive */
762         {
763             .name = BDRV_OPT_READ_ONLY,
764             .type = QEMU_OPT_BOOL,
765             .help = "open drive file as read-only",
766         },{
767             .name = "rerror",
768             .type = QEMU_OPT_STRING,
769             .help = "read error action",
770         },{
771             .name = "werror",
772             .type = QEMU_OPT_STRING,
773             .help = "write error action",
774         },{
775             .name = "copy-on-read",
776             .type = QEMU_OPT_BOOL,
777             .help = "copy read data from backing file into image file",
778         },
779 
780         { /* end of list */ }
781     },
782 };
783 
784 DriveInfo *drive_new(QemuOpts *all_opts, BlockInterfaceType block_default_type,
785                      Error **errp)
786 {
787     const char *value;
788     BlockBackend *blk;
789     DriveInfo *dinfo = NULL;
790     QDict *bs_opts;
791     QemuOpts *legacy_opts;
792     DriveMediaType media = MEDIA_DISK;
793     BlockInterfaceType type;
794     int max_devs, bus_id, unit_id, index;
795     const char *werror, *rerror;
796     bool read_only = false;
797     bool copy_on_read;
798     const char *filename;
799     int i;
800 
801     GLOBAL_STATE_CODE();
802 
803     /* Change legacy command line options into QMP ones */
804     static const struct {
805         const char *from;
806         const char *to;
807     } opt_renames[] = {
808         { "iops",           "throttling.iops-total" },
809         { "iops_rd",        "throttling.iops-read" },
810         { "iops_wr",        "throttling.iops-write" },
811 
812         { "bps",            "throttling.bps-total" },
813         { "bps_rd",         "throttling.bps-read" },
814         { "bps_wr",         "throttling.bps-write" },
815 
816         { "iops_max",       "throttling.iops-total-max" },
817         { "iops_rd_max",    "throttling.iops-read-max" },
818         { "iops_wr_max",    "throttling.iops-write-max" },
819 
820         { "bps_max",        "throttling.bps-total-max" },
821         { "bps_rd_max",     "throttling.bps-read-max" },
822         { "bps_wr_max",     "throttling.bps-write-max" },
823 
824         { "iops_size",      "throttling.iops-size" },
825 
826         { "group",          "throttling.group" },
827 
828         { "readonly",       BDRV_OPT_READ_ONLY },
829     };
830 
831     for (i = 0; i < ARRAY_SIZE(opt_renames); i++) {
832         if (!qemu_opt_rename(all_opts, opt_renames[i].from,
833                              opt_renames[i].to, errp)) {
834             return NULL;
835         }
836     }
837 
838     value = qemu_opt_get(all_opts, "cache");
839     if (value) {
840         int flags = 0;
841         bool writethrough;
842 
843         if (bdrv_parse_cache_mode(value, &flags, &writethrough) != 0) {
844             error_setg(errp, "invalid cache option");
845             return NULL;
846         }
847 
848         /* Specific options take precedence */
849         if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_WB)) {
850             qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_WB,
851                               !writethrough, &error_abort);
852         }
853         if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_DIRECT)) {
854             qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_DIRECT,
855                               !!(flags & BDRV_O_NOCACHE), &error_abort);
856         }
857         if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_NO_FLUSH)) {
858             qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_NO_FLUSH,
859                               !!(flags & BDRV_O_NO_FLUSH), &error_abort);
860         }
861         qemu_opt_unset(all_opts, "cache");
862     }
863 
864     /* Get a QDict for processing the options */
865     bs_opts = qdict_new();
866     qemu_opts_to_qdict(all_opts, bs_opts);
867 
868     legacy_opts = qemu_opts_create(&qemu_legacy_drive_opts, NULL, 0,
869                                    &error_abort);
870     if (!qemu_opts_absorb_qdict(legacy_opts, bs_opts, errp)) {
871         goto fail;
872     }
873 
874     /* Media type */
875     value = qemu_opt_get(legacy_opts, "media");
876     if (value) {
877         if (!strcmp(value, "disk")) {
878             media = MEDIA_DISK;
879         } else if (!strcmp(value, "cdrom")) {
880             media = MEDIA_CDROM;
881             read_only = true;
882         } else {
883             error_setg(errp, "'%s' invalid media", value);
884             goto fail;
885         }
886     }
887 
888     /* copy-on-read is disabled with a warning for read-only devices */
889     read_only |= qemu_opt_get_bool(legacy_opts, BDRV_OPT_READ_ONLY, false);
890     copy_on_read = qemu_opt_get_bool(legacy_opts, "copy-on-read", false);
891 
892     if (read_only && copy_on_read) {
893         warn_report("disabling copy-on-read on read-only drive");
894         copy_on_read = false;
895     }
896 
897     qdict_put_str(bs_opts, BDRV_OPT_READ_ONLY, read_only ? "on" : "off");
898     qdict_put_str(bs_opts, "copy-on-read", copy_on_read ? "on" : "off");
899 
900     /* Controller type */
901     value = qemu_opt_get(legacy_opts, "if");
902     if (value) {
903         for (type = 0;
904              type < IF_COUNT && strcmp(value, if_name[type]);
905              type++) {
906         }
907         if (type == IF_COUNT) {
908             error_setg(errp, "unsupported bus type '%s'", value);
909             goto fail;
910         }
911     } else {
912         type = block_default_type;
913     }
914 
915     /* Device address specified by bus/unit or index.
916      * If none was specified, try to find the first free one. */
917     bus_id  = qemu_opt_get_number(legacy_opts, "bus", 0);
918     unit_id = qemu_opt_get_number(legacy_opts, "unit", -1);
919     index   = qemu_opt_get_number(legacy_opts, "index", -1);
920 
921     max_devs = if_max_devs[type];
922 
923     if (index != -1) {
924         if (bus_id != 0 || unit_id != -1) {
925             error_setg(errp, "index cannot be used with bus and unit");
926             goto fail;
927         }
928         bus_id = drive_index_to_bus_id(type, index);
929         unit_id = drive_index_to_unit_id(type, index);
930     }
931 
932     if (unit_id == -1) {
933        unit_id = 0;
934        while (drive_get(type, bus_id, unit_id) != NULL) {
935            unit_id++;
936            if (max_devs && unit_id >= max_devs) {
937                unit_id -= max_devs;
938                bus_id++;
939            }
940        }
941     }
942 
943     if (max_devs && unit_id >= max_devs) {
944         error_setg(errp, "unit %d too big (max is %d)", unit_id, max_devs - 1);
945         goto fail;
946     }
947 
948     if (drive_get(type, bus_id, unit_id) != NULL) {
949         error_setg(errp, "drive with bus=%d, unit=%d (index=%d) exists",
950                    bus_id, unit_id, index);
951         goto fail;
952     }
953 
954     /* no id supplied -> create one */
955     if (qemu_opts_id(all_opts) == NULL) {
956         char *new_id;
957         const char *mediastr = "";
958         if (type == IF_IDE || type == IF_SCSI) {
959             mediastr = (media == MEDIA_CDROM) ? "-cd" : "-hd";
960         }
961         if (max_devs) {
962             new_id = g_strdup_printf("%s%i%s%i", if_name[type], bus_id,
963                                      mediastr, unit_id);
964         } else {
965             new_id = g_strdup_printf("%s%s%i", if_name[type],
966                                      mediastr, unit_id);
967         }
968         qdict_put_str(bs_opts, "id", new_id);
969         g_free(new_id);
970     }
971 
972     /* Add virtio block device */
973     if (type == IF_VIRTIO) {
974         QemuOpts *devopts;
975         devopts = qemu_opts_create(qemu_find_opts("device"), NULL, 0,
976                                    &error_abort);
977         qemu_opt_set(devopts, "driver", "virtio-blk", &error_abort);
978         qemu_opt_set(devopts, "drive", qdict_get_str(bs_opts, "id"),
979                      &error_abort);
980     } else if (type == IF_XEN) {
981         QemuOpts *devopts;
982         devopts = qemu_opts_create(qemu_find_opts("device"), NULL, 0,
983                                    &error_abort);
984         qemu_opt_set(devopts, "driver",
985                      (media == MEDIA_CDROM) ? "xen-cdrom" : "xen-disk",
986                      &error_abort);
987         qemu_opt_set(devopts, "drive", qdict_get_str(bs_opts, "id"),
988                      &error_abort);
989     }
990 
991     filename = qemu_opt_get(legacy_opts, "file");
992 
993     /* Check werror/rerror compatibility with if=... */
994     werror = qemu_opt_get(legacy_opts, "werror");
995     if (werror != NULL) {
996         if (type != IF_IDE && type != IF_SCSI && type != IF_VIRTIO &&
997             type != IF_NONE) {
998             error_setg(errp, "werror is not supported by this bus type");
999             goto fail;
1000         }
1001         qdict_put_str(bs_opts, "werror", werror);
1002     }
1003 
1004     rerror = qemu_opt_get(legacy_opts, "rerror");
1005     if (rerror != NULL) {
1006         if (type != IF_IDE && type != IF_VIRTIO && type != IF_SCSI &&
1007             type != IF_NONE) {
1008             error_setg(errp, "rerror is not supported by this bus type");
1009             goto fail;
1010         }
1011         qdict_put_str(bs_opts, "rerror", rerror);
1012     }
1013 
1014     /* Actual block device init: Functionality shared with blockdev-add */
1015     blk = blockdev_init(filename, bs_opts, errp);
1016     bs_opts = NULL;
1017     if (!blk) {
1018         goto fail;
1019     }
1020 
1021     /* Create legacy DriveInfo */
1022     dinfo = g_malloc0(sizeof(*dinfo));
1023     dinfo->opts = all_opts;
1024 
1025     dinfo->type = type;
1026     dinfo->bus = bus_id;
1027     dinfo->unit = unit_id;
1028 
1029     blk_set_legacy_dinfo(blk, dinfo);
1030 
1031     switch(type) {
1032     case IF_IDE:
1033     case IF_SCSI:
1034     case IF_XEN:
1035     case IF_NONE:
1036         dinfo->media_cd = media == MEDIA_CDROM;
1037         break;
1038     default:
1039         break;
1040     }
1041 
1042 fail:
1043     qemu_opts_del(legacy_opts);
1044     qobject_unref(bs_opts);
1045     return dinfo;
1046 }
1047 
1048 static BlockDriverState *qmp_get_root_bs(const char *name, Error **errp)
1049 {
1050     BlockDriverState *bs;
1051     AioContext *aio_context;
1052 
1053     GRAPH_RDLOCK_GUARD_MAINLOOP();
1054 
1055     bs = bdrv_lookup_bs(name, name, errp);
1056     if (bs == NULL) {
1057         return NULL;
1058     }
1059 
1060     if (!bdrv_is_root_node(bs)) {
1061         error_setg(errp, "Need a root block node");
1062         return NULL;
1063     }
1064 
1065     aio_context = bdrv_get_aio_context(bs);
1066     aio_context_acquire(aio_context);
1067 
1068     if (!bdrv_is_inserted(bs)) {
1069         error_setg(errp, "Device has no medium");
1070         bs = NULL;
1071     }
1072 
1073     aio_context_release(aio_context);
1074 
1075     return bs;
1076 }
1077 
1078 static void blockdev_do_action(TransactionAction *action, Error **errp)
1079 {
1080     TransactionActionList list;
1081 
1082     list.value = action;
1083     list.next = NULL;
1084     qmp_transaction(&list, NULL, errp);
1085 }
1086 
1087 void qmp_blockdev_snapshot_sync(const char *device, const char *node_name,
1088                                 const char *snapshot_file,
1089                                 const char *snapshot_node_name,
1090                                 const char *format,
1091                                 bool has_mode, NewImageMode mode, Error **errp)
1092 {
1093     BlockdevSnapshotSync snapshot = {
1094         .device = (char *) device,
1095         .node_name = (char *) node_name,
1096         .snapshot_file = (char *) snapshot_file,
1097         .snapshot_node_name = (char *) snapshot_node_name,
1098         .format = (char *) format,
1099         .has_mode = has_mode,
1100         .mode = mode,
1101     };
1102     TransactionAction action = {
1103         .type = TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC,
1104         .u.blockdev_snapshot_sync.data = &snapshot,
1105     };
1106     blockdev_do_action(&action, errp);
1107 }
1108 
1109 void qmp_blockdev_snapshot(const char *node, const char *overlay,
1110                            Error **errp)
1111 {
1112     BlockdevSnapshot snapshot_data = {
1113         .node = (char *) node,
1114         .overlay = (char *) overlay
1115     };
1116     TransactionAction action = {
1117         .type = TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT,
1118         .u.blockdev_snapshot.data = &snapshot_data,
1119     };
1120     blockdev_do_action(&action, errp);
1121 }
1122 
1123 void qmp_blockdev_snapshot_internal_sync(const char *device,
1124                                          const char *name,
1125                                          Error **errp)
1126 {
1127     BlockdevSnapshotInternal snapshot = {
1128         .device = (char *) device,
1129         .name = (char *) name
1130     };
1131     TransactionAction action = {
1132         .type = TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC,
1133         .u.blockdev_snapshot_internal_sync.data = &snapshot,
1134     };
1135     blockdev_do_action(&action, errp);
1136 }
1137 
1138 SnapshotInfo *qmp_blockdev_snapshot_delete_internal_sync(const char *device,
1139                                                          const char *id,
1140                                                          const char *name,
1141                                                          Error **errp)
1142 {
1143     BlockDriverState *bs;
1144     AioContext *aio_context;
1145     QEMUSnapshotInfo sn;
1146     Error *local_err = NULL;
1147     SnapshotInfo *info = NULL;
1148     int ret;
1149 
1150     GLOBAL_STATE_CODE();
1151     GRAPH_RDLOCK_GUARD_MAINLOOP();
1152 
1153     bs = qmp_get_root_bs(device, errp);
1154     if (!bs) {
1155         return NULL;
1156     }
1157     aio_context = bdrv_get_aio_context(bs);
1158     aio_context_acquire(aio_context);
1159 
1160     if (!id && !name) {
1161         error_setg(errp, "Name or id must be provided");
1162         goto out_aio_context;
1163     }
1164 
1165     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_INTERNAL_SNAPSHOT_DELETE, errp)) {
1166         goto out_aio_context;
1167     }
1168 
1169     ret = bdrv_snapshot_find_by_id_and_name(bs, id, name, &sn, &local_err);
1170     if (local_err) {
1171         error_propagate(errp, local_err);
1172         goto out_aio_context;
1173     }
1174     if (!ret) {
1175         error_setg(errp,
1176                    "Snapshot with id '%s' and name '%s' does not exist on "
1177                    "device '%s'",
1178                    STR_OR_NULL(id), STR_OR_NULL(name), device);
1179         goto out_aio_context;
1180     }
1181 
1182     bdrv_snapshot_delete(bs, id, name, &local_err);
1183     if (local_err) {
1184         error_propagate(errp, local_err);
1185         goto out_aio_context;
1186     }
1187 
1188     aio_context_release(aio_context);
1189 
1190     info = g_new0(SnapshotInfo, 1);
1191     info->id = g_strdup(sn.id_str);
1192     info->name = g_strdup(sn.name);
1193     info->date_nsec = sn.date_nsec;
1194     info->date_sec = sn.date_sec;
1195     info->vm_state_size = sn.vm_state_size;
1196     info->vm_clock_nsec = sn.vm_clock_nsec % 1000000000;
1197     info->vm_clock_sec = sn.vm_clock_nsec / 1000000000;
1198     if (sn.icount != -1ULL) {
1199         info->icount = sn.icount;
1200         info->has_icount = true;
1201     }
1202 
1203     return info;
1204 
1205 out_aio_context:
1206     aio_context_release(aio_context);
1207     return NULL;
1208 }
1209 
1210 /* internal snapshot private data */
1211 typedef struct InternalSnapshotState {
1212     BlockDriverState *bs;
1213     QEMUSnapshotInfo sn;
1214     bool created;
1215 } InternalSnapshotState;
1216 
1217 static void internal_snapshot_abort(void *opaque);
1218 static void internal_snapshot_clean(void *opaque);
1219 TransactionActionDrv internal_snapshot_drv = {
1220     .abort = internal_snapshot_abort,
1221     .clean = internal_snapshot_clean,
1222 };
1223 
1224 static void internal_snapshot_action(BlockdevSnapshotInternal *internal,
1225                                      Transaction *tran, Error **errp)
1226 {
1227     Error *local_err = NULL;
1228     const char *device;
1229     const char *name;
1230     BlockDriverState *bs;
1231     QEMUSnapshotInfo old_sn, *sn;
1232     bool ret;
1233     int64_t rt;
1234     InternalSnapshotState *state = g_new0(InternalSnapshotState, 1);
1235     AioContext *aio_context;
1236     int ret1;
1237 
1238     GLOBAL_STATE_CODE();
1239     GRAPH_RDLOCK_GUARD_MAINLOOP();
1240 
1241     tran_add(tran, &internal_snapshot_drv, state);
1242 
1243     device = internal->device;
1244     name = internal->name;
1245 
1246     bs = qmp_get_root_bs(device, errp);
1247     if (!bs) {
1248         return;
1249     }
1250 
1251     aio_context = bdrv_get_aio_context(bs);
1252     aio_context_acquire(aio_context);
1253 
1254     state->bs = bs;
1255 
1256     /* Paired with .clean() */
1257     bdrv_drained_begin(bs);
1258 
1259     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_INTERNAL_SNAPSHOT, errp)) {
1260         goto out;
1261     }
1262 
1263     if (bdrv_is_read_only(bs)) {
1264         error_setg(errp, "Device '%s' is read only", device);
1265         goto out;
1266     }
1267 
1268     if (!bdrv_can_snapshot(bs)) {
1269         error_setg(errp, "Block format '%s' used by device '%s' "
1270                    "does not support internal snapshots",
1271                    bs->drv->format_name, device);
1272         goto out;
1273     }
1274 
1275     if (!strlen(name)) {
1276         error_setg(errp, "Name is empty");
1277         goto out;
1278     }
1279 
1280     /* check whether a snapshot with name exist */
1281     ret = bdrv_snapshot_find_by_id_and_name(bs, NULL, name, &old_sn,
1282                                             &local_err);
1283     if (local_err) {
1284         error_propagate(errp, local_err);
1285         goto out;
1286     } else if (ret) {
1287         error_setg(errp,
1288                    "Snapshot with name '%s' already exists on device '%s'",
1289                    name, device);
1290         goto out;
1291     }
1292 
1293     /* 3. take the snapshot */
1294     sn = &state->sn;
1295     pstrcpy(sn->name, sizeof(sn->name), name);
1296     rt = g_get_real_time();
1297     sn->date_sec = rt / G_USEC_PER_SEC;
1298     sn->date_nsec = (rt % G_USEC_PER_SEC) * 1000;
1299     sn->vm_clock_nsec = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL);
1300     if (replay_mode != REPLAY_MODE_NONE) {
1301         sn->icount = replay_get_current_icount();
1302     } else {
1303         sn->icount = -1ULL;
1304     }
1305 
1306     ret1 = bdrv_snapshot_create(bs, sn);
1307     if (ret1 < 0) {
1308         error_setg_errno(errp, -ret1,
1309                          "Failed to create snapshot '%s' on device '%s'",
1310                          name, device);
1311         goto out;
1312     }
1313 
1314     /* 4. succeed, mark a snapshot is created */
1315     state->created = true;
1316 
1317 out:
1318     aio_context_release(aio_context);
1319 }
1320 
1321 static void internal_snapshot_abort(void *opaque)
1322 {
1323     InternalSnapshotState *state = opaque;
1324     BlockDriverState *bs = state->bs;
1325     QEMUSnapshotInfo *sn = &state->sn;
1326     AioContext *aio_context;
1327     Error *local_error = NULL;
1328 
1329     GLOBAL_STATE_CODE();
1330     GRAPH_RDLOCK_GUARD_MAINLOOP();
1331 
1332     if (!state->created) {
1333         return;
1334     }
1335 
1336     aio_context = bdrv_get_aio_context(state->bs);
1337     aio_context_acquire(aio_context);
1338 
1339     if (bdrv_snapshot_delete(bs, sn->id_str, sn->name, &local_error) < 0) {
1340         error_reportf_err(local_error,
1341                           "Failed to delete snapshot with id '%s' and "
1342                           "name '%s' on device '%s' in abort: ",
1343                           sn->id_str, sn->name,
1344                           bdrv_get_device_name(bs));
1345     }
1346 
1347     aio_context_release(aio_context);
1348 }
1349 
1350 static void internal_snapshot_clean(void *opaque)
1351 {
1352     g_autofree InternalSnapshotState *state = opaque;
1353     AioContext *aio_context;
1354 
1355     if (!state->bs) {
1356         return;
1357     }
1358 
1359     aio_context = bdrv_get_aio_context(state->bs);
1360     aio_context_acquire(aio_context);
1361 
1362     bdrv_drained_end(state->bs);
1363 
1364     aio_context_release(aio_context);
1365 }
1366 
1367 /* external snapshot private data */
1368 typedef struct ExternalSnapshotState {
1369     BlockDriverState *old_bs;
1370     BlockDriverState *new_bs;
1371     bool overlay_appended;
1372 } ExternalSnapshotState;
1373 
1374 static void external_snapshot_commit(void *opaque);
1375 static void external_snapshot_abort(void *opaque);
1376 static void external_snapshot_clean(void *opaque);
1377 TransactionActionDrv external_snapshot_drv = {
1378     .commit = external_snapshot_commit,
1379     .abort = external_snapshot_abort,
1380     .clean = external_snapshot_clean,
1381 };
1382 
1383 static void external_snapshot_action(TransactionAction *action,
1384                                      Transaction *tran, Error **errp)
1385 {
1386     int ret;
1387     int flags = 0;
1388     QDict *options = NULL;
1389     Error *local_err = NULL;
1390     /* Device and node name of the image to generate the snapshot from */
1391     const char *device;
1392     const char *node_name;
1393     /* Reference to the new image (for 'blockdev-snapshot') */
1394     const char *snapshot_ref;
1395     /* File name of the new image (for 'blockdev-snapshot-sync') */
1396     const char *new_image_file;
1397     ExternalSnapshotState *state = g_new0(ExternalSnapshotState, 1);
1398     AioContext *aio_context;
1399     uint64_t perm, shared;
1400 
1401     /* TODO We'll eventually have to take a writer lock in this function */
1402     GRAPH_RDLOCK_GUARD_MAINLOOP();
1403 
1404     tran_add(tran, &external_snapshot_drv, state);
1405 
1406     /* 'blockdev-snapshot' and 'blockdev-snapshot-sync' have similar
1407      * purpose but a different set of parameters */
1408     switch (action->type) {
1409     case TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT:
1410         {
1411             BlockdevSnapshot *s = action->u.blockdev_snapshot.data;
1412             device = s->node;
1413             node_name = s->node;
1414             new_image_file = NULL;
1415             snapshot_ref = s->overlay;
1416         }
1417         break;
1418     case TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC:
1419         {
1420             BlockdevSnapshotSync *s = action->u.blockdev_snapshot_sync.data;
1421             device = s->device;
1422             node_name = s->node_name;
1423             new_image_file = s->snapshot_file;
1424             snapshot_ref = NULL;
1425         }
1426         break;
1427     default:
1428         g_assert_not_reached();
1429     }
1430 
1431     /* start processing */
1432 
1433     state->old_bs = bdrv_lookup_bs(device, node_name, errp);
1434     if (!state->old_bs) {
1435         return;
1436     }
1437 
1438     aio_context = bdrv_get_aio_context(state->old_bs);
1439     aio_context_acquire(aio_context);
1440 
1441     /* Paired with .clean() */
1442     bdrv_drained_begin(state->old_bs);
1443 
1444     if (!bdrv_is_inserted(state->old_bs)) {
1445         error_setg(errp, QERR_DEVICE_HAS_NO_MEDIUM, device);
1446         goto out;
1447     }
1448 
1449     if (bdrv_op_is_blocked(state->old_bs,
1450                            BLOCK_OP_TYPE_EXTERNAL_SNAPSHOT, errp)) {
1451         goto out;
1452     }
1453 
1454     if (!bdrv_is_read_only(state->old_bs)) {
1455         if (bdrv_flush(state->old_bs)) {
1456             error_setg(errp, QERR_IO_ERROR);
1457             goto out;
1458         }
1459     }
1460 
1461     if (action->type == TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC) {
1462         BlockdevSnapshotSync *s = action->u.blockdev_snapshot_sync.data;
1463         const char *format = s->format ?: "qcow2";
1464         enum NewImageMode mode;
1465         const char *snapshot_node_name = s->snapshot_node_name;
1466 
1467         if (node_name && !snapshot_node_name) {
1468             error_setg(errp, "New overlay node-name missing");
1469             goto out;
1470         }
1471 
1472         if (snapshot_node_name &&
1473             bdrv_lookup_bs(snapshot_node_name, snapshot_node_name, NULL)) {
1474             error_setg(errp, "New overlay node-name already in use");
1475             goto out;
1476         }
1477 
1478         flags = state->old_bs->open_flags;
1479         flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_COPY_ON_READ);
1480         flags |= BDRV_O_NO_BACKING;
1481 
1482         /* create new image w/backing file */
1483         mode = s->has_mode ? s->mode : NEW_IMAGE_MODE_ABSOLUTE_PATHS;
1484         if (mode != NEW_IMAGE_MODE_EXISTING) {
1485             int64_t size = bdrv_getlength(state->old_bs);
1486             if (size < 0) {
1487                 error_setg_errno(errp, -size, "bdrv_getlength failed");
1488                 goto out;
1489             }
1490             bdrv_refresh_filename(state->old_bs);
1491 
1492             aio_context_release(aio_context);
1493             bdrv_img_create(new_image_file, format,
1494                             state->old_bs->filename,
1495                             state->old_bs->drv->format_name,
1496                             NULL, size, flags, false, &local_err);
1497             aio_context_acquire(aio_context);
1498 
1499             if (local_err) {
1500                 error_propagate(errp, local_err);
1501                 goto out;
1502             }
1503         }
1504 
1505         options = qdict_new();
1506         if (snapshot_node_name) {
1507             qdict_put_str(options, "node-name", snapshot_node_name);
1508         }
1509         qdict_put_str(options, "driver", format);
1510     }
1511     aio_context_release(aio_context);
1512 
1513     aio_context_acquire(qemu_get_aio_context());
1514     state->new_bs = bdrv_open(new_image_file, snapshot_ref, options, flags,
1515                               errp);
1516     aio_context_release(qemu_get_aio_context());
1517 
1518     /* We will manually add the backing_hd field to the bs later */
1519     if (!state->new_bs) {
1520         return;
1521     }
1522 
1523     aio_context_acquire(aio_context);
1524 
1525     /*
1526      * Allow attaching a backing file to an overlay that's already in use only
1527      * if the parents don't assume that they are already seeing a valid image.
1528      * (Specifically, allow it as a mirror target, which is write-only access.)
1529      */
1530     bdrv_get_cumulative_perm(state->new_bs, &perm, &shared);
1531     if (perm & BLK_PERM_CONSISTENT_READ) {
1532         error_setg(errp, "The overlay is already in use");
1533         goto out;
1534     }
1535 
1536     if (state->new_bs->drv->is_filter) {
1537         error_setg(errp, "Filters cannot be used as overlays");
1538         goto out;
1539     }
1540 
1541     if (bdrv_cow_child(state->new_bs)) {
1542         error_setg(errp, "The overlay already has a backing image");
1543         goto out;
1544     }
1545 
1546     if (!state->new_bs->drv->supports_backing) {
1547         error_setg(errp, "The overlay does not support backing images");
1548         goto out;
1549     }
1550 
1551     ret = bdrv_append(state->new_bs, state->old_bs, errp);
1552     if (ret < 0) {
1553         goto out;
1554     }
1555     state->overlay_appended = true;
1556 
1557 out:
1558     aio_context_release(aio_context);
1559 }
1560 
1561 static void external_snapshot_commit(void *opaque)
1562 {
1563     ExternalSnapshotState *state = opaque;
1564     AioContext *aio_context;
1565 
1566     aio_context = bdrv_get_aio_context(state->old_bs);
1567     aio_context_acquire(aio_context);
1568 
1569     /* We don't need (or want) to use the transactional
1570      * bdrv_reopen_multiple() across all the entries at once, because we
1571      * don't want to abort all of them if one of them fails the reopen */
1572     if (!qatomic_read(&state->old_bs->copy_on_read)) {
1573         bdrv_reopen_set_read_only(state->old_bs, true, NULL);
1574     }
1575 
1576     aio_context_release(aio_context);
1577 }
1578 
1579 static void external_snapshot_abort(void *opaque)
1580 {
1581     ExternalSnapshotState *state = opaque;
1582     if (state->new_bs) {
1583         if (state->overlay_appended) {
1584             AioContext *aio_context;
1585             AioContext *tmp_context;
1586             int ret;
1587 
1588             aio_context = bdrv_get_aio_context(state->old_bs);
1589             aio_context_acquire(aio_context);
1590 
1591             bdrv_ref(state->old_bs);   /* we can't let bdrv_set_backind_hd()
1592                                           close state->old_bs; we need it */
1593             bdrv_set_backing_hd(state->new_bs, NULL, &error_abort);
1594 
1595             /*
1596              * The call to bdrv_set_backing_hd() above returns state->old_bs to
1597              * the main AioContext. As we're still going to be using it, return
1598              * it to the AioContext it was before.
1599              */
1600             tmp_context = bdrv_get_aio_context(state->old_bs);
1601             if (aio_context != tmp_context) {
1602                 aio_context_release(aio_context);
1603                 aio_context_acquire(tmp_context);
1604 
1605                 ret = bdrv_try_change_aio_context(state->old_bs,
1606                                                   aio_context, NULL, NULL);
1607                 assert(ret == 0);
1608 
1609                 aio_context_release(tmp_context);
1610                 aio_context_acquire(aio_context);
1611             }
1612 
1613             bdrv_drained_begin(state->new_bs);
1614             bdrv_graph_wrlock(state->old_bs);
1615             bdrv_replace_node(state->new_bs, state->old_bs, &error_abort);
1616             bdrv_graph_wrunlock();
1617             bdrv_drained_end(state->new_bs);
1618 
1619             bdrv_unref(state->old_bs); /* bdrv_replace_node() ref'ed old_bs */
1620 
1621             aio_context_release(aio_context);
1622         }
1623     }
1624 }
1625 
1626 static void external_snapshot_clean(void *opaque)
1627 {
1628     g_autofree ExternalSnapshotState *state = opaque;
1629     AioContext *aio_context;
1630 
1631     if (!state->old_bs) {
1632         return;
1633     }
1634 
1635     aio_context = bdrv_get_aio_context(state->old_bs);
1636     aio_context_acquire(aio_context);
1637 
1638     bdrv_drained_end(state->old_bs);
1639     bdrv_unref(state->new_bs);
1640 
1641     aio_context_release(aio_context);
1642 }
1643 
1644 typedef struct DriveBackupState {
1645     BlockDriverState *bs;
1646     BlockJob *job;
1647 } DriveBackupState;
1648 
1649 static BlockJob *do_backup_common(BackupCommon *backup,
1650                                   BlockDriverState *bs,
1651                                   BlockDriverState *target_bs,
1652                                   AioContext *aio_context,
1653                                   JobTxn *txn, Error **errp);
1654 
1655 static void drive_backup_commit(void *opaque);
1656 static void drive_backup_abort(void *opaque);
1657 static void drive_backup_clean(void *opaque);
1658 TransactionActionDrv drive_backup_drv = {
1659     .commit = drive_backup_commit,
1660     .abort = drive_backup_abort,
1661     .clean = drive_backup_clean,
1662 };
1663 
1664 static void drive_backup_action(DriveBackup *backup,
1665                                 JobTxn *block_job_txn,
1666                                 Transaction *tran, Error **errp)
1667 {
1668     DriveBackupState *state = g_new0(DriveBackupState, 1);
1669     BlockDriverState *bs;
1670     BlockDriverState *target_bs;
1671     BlockDriverState *source = NULL;
1672     AioContext *aio_context;
1673     AioContext *old_context;
1674     const char *format;
1675     QDict *options;
1676     Error *local_err = NULL;
1677     int flags;
1678     int64_t size;
1679     bool set_backing_hd = false;
1680     int ret;
1681 
1682     GLOBAL_STATE_CODE();
1683 
1684     tran_add(tran, &drive_backup_drv, state);
1685 
1686     if (!backup->has_mode) {
1687         backup->mode = NEW_IMAGE_MODE_ABSOLUTE_PATHS;
1688     }
1689 
1690     bs = bdrv_lookup_bs(backup->device, backup->device, errp);
1691     if (!bs) {
1692         return;
1693     }
1694 
1695     if (!bs->drv) {
1696         error_setg(errp, "Device has no medium");
1697         return;
1698     }
1699 
1700     aio_context = bdrv_get_aio_context(bs);
1701     aio_context_acquire(aio_context);
1702 
1703     state->bs = bs;
1704     /* Paired with .clean() */
1705     bdrv_drained_begin(bs);
1706 
1707     format = backup->format;
1708     if (!format && backup->mode != NEW_IMAGE_MODE_EXISTING) {
1709         format = bs->drv->format_name;
1710     }
1711 
1712     /* Early check to avoid creating target */
1713     bdrv_graph_rdlock_main_loop();
1714     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_BACKUP_SOURCE, errp)) {
1715         bdrv_graph_rdunlock_main_loop();
1716         goto out;
1717     }
1718 
1719     flags = bs->open_flags | BDRV_O_RDWR;
1720 
1721     /*
1722      * See if we have a backing HD we can use to create our new image
1723      * on top of.
1724      */
1725     if (backup->sync == MIRROR_SYNC_MODE_TOP) {
1726         /*
1727          * Backup will not replace the source by the target, so none
1728          * of the filters skipped here will be removed (in contrast to
1729          * mirror).  Therefore, we can skip all of them when looking
1730          * for the first COW relationship.
1731          */
1732         source = bdrv_cow_bs(bdrv_skip_filters(bs));
1733         if (!source) {
1734             backup->sync = MIRROR_SYNC_MODE_FULL;
1735         }
1736     }
1737     if (backup->sync == MIRROR_SYNC_MODE_NONE) {
1738         source = bs;
1739         flags |= BDRV_O_NO_BACKING;
1740         set_backing_hd = true;
1741     }
1742     bdrv_graph_rdunlock_main_loop();
1743 
1744     size = bdrv_getlength(bs);
1745     if (size < 0) {
1746         error_setg_errno(errp, -size, "bdrv_getlength failed");
1747         goto out;
1748     }
1749 
1750     if (backup->mode != NEW_IMAGE_MODE_EXISTING) {
1751         assert(format);
1752         if (source) {
1753             /* Implicit filters should not appear in the filename */
1754             BlockDriverState *explicit_backing;
1755 
1756             bdrv_graph_rdlock_main_loop();
1757             explicit_backing = bdrv_skip_implicit_filters(source);
1758             bdrv_refresh_filename(explicit_backing);
1759             bdrv_graph_rdunlock_main_loop();
1760 
1761             bdrv_img_create(backup->target, format,
1762                             explicit_backing->filename,
1763                             explicit_backing->drv->format_name, NULL,
1764                             size, flags, false, &local_err);
1765         } else {
1766             bdrv_img_create(backup->target, format, NULL, NULL, NULL,
1767                             size, flags, false, &local_err);
1768         }
1769     }
1770 
1771     if (local_err) {
1772         error_propagate(errp, local_err);
1773         goto out;
1774     }
1775 
1776     options = qdict_new();
1777     qdict_put_str(options, "discard", "unmap");
1778     qdict_put_str(options, "detect-zeroes", "unmap");
1779     if (format) {
1780         qdict_put_str(options, "driver", format);
1781     }
1782     aio_context_release(aio_context);
1783 
1784     aio_context_acquire(qemu_get_aio_context());
1785     target_bs = bdrv_open(backup->target, NULL, options, flags, errp);
1786     aio_context_release(qemu_get_aio_context());
1787 
1788     if (!target_bs) {
1789         return;
1790     }
1791 
1792     /* Honor bdrv_try_change_aio_context() context acquisition requirements. */
1793     old_context = bdrv_get_aio_context(target_bs);
1794     aio_context_acquire(old_context);
1795 
1796     ret = bdrv_try_change_aio_context(target_bs, aio_context, NULL, errp);
1797     if (ret < 0) {
1798         bdrv_unref(target_bs);
1799         aio_context_release(old_context);
1800         return;
1801     }
1802 
1803     aio_context_release(old_context);
1804     aio_context_acquire(aio_context);
1805 
1806     if (set_backing_hd) {
1807         if (bdrv_set_backing_hd(target_bs, source, errp) < 0) {
1808             goto unref;
1809         }
1810     }
1811 
1812     state->job = do_backup_common(qapi_DriveBackup_base(backup),
1813                                   bs, target_bs, aio_context,
1814                                   block_job_txn, errp);
1815 
1816 unref:
1817     bdrv_unref(target_bs);
1818 out:
1819     aio_context_release(aio_context);
1820 }
1821 
1822 static void drive_backup_commit(void *opaque)
1823 {
1824     DriveBackupState *state = opaque;
1825     AioContext *aio_context;
1826 
1827     aio_context = bdrv_get_aio_context(state->bs);
1828     aio_context_acquire(aio_context);
1829 
1830     assert(state->job);
1831     job_start(&state->job->job);
1832 
1833     aio_context_release(aio_context);
1834 }
1835 
1836 static void drive_backup_abort(void *opaque)
1837 {
1838     DriveBackupState *state = opaque;
1839 
1840     if (state->job) {
1841         job_cancel_sync(&state->job->job, true);
1842     }
1843 }
1844 
1845 static void drive_backup_clean(void *opaque)
1846 {
1847     g_autofree DriveBackupState *state = opaque;
1848     AioContext *aio_context;
1849 
1850     if (!state->bs) {
1851         return;
1852     }
1853 
1854     aio_context = bdrv_get_aio_context(state->bs);
1855     aio_context_acquire(aio_context);
1856 
1857     bdrv_drained_end(state->bs);
1858 
1859     aio_context_release(aio_context);
1860 }
1861 
1862 typedef struct BlockdevBackupState {
1863     BlockDriverState *bs;
1864     BlockJob *job;
1865 } BlockdevBackupState;
1866 
1867 static void blockdev_backup_commit(void *opaque);
1868 static void blockdev_backup_abort(void *opaque);
1869 static void blockdev_backup_clean(void *opaque);
1870 TransactionActionDrv blockdev_backup_drv = {
1871     .commit = blockdev_backup_commit,
1872     .abort = blockdev_backup_abort,
1873     .clean = blockdev_backup_clean,
1874 };
1875 
1876 static void blockdev_backup_action(BlockdevBackup *backup,
1877                                    JobTxn *block_job_txn,
1878                                    Transaction *tran, Error **errp)
1879 {
1880     BlockdevBackupState *state = g_new0(BlockdevBackupState, 1);
1881     BlockDriverState *bs;
1882     BlockDriverState *target_bs;
1883     AioContext *aio_context;
1884     AioContext *old_context;
1885     int ret;
1886 
1887     tran_add(tran, &blockdev_backup_drv, state);
1888 
1889     bs = bdrv_lookup_bs(backup->device, backup->device, errp);
1890     if (!bs) {
1891         return;
1892     }
1893 
1894     target_bs = bdrv_lookup_bs(backup->target, backup->target, errp);
1895     if (!target_bs) {
1896         return;
1897     }
1898 
1899     /* Honor bdrv_try_change_aio_context() context acquisition requirements. */
1900     aio_context = bdrv_get_aio_context(bs);
1901     old_context = bdrv_get_aio_context(target_bs);
1902     aio_context_acquire(old_context);
1903 
1904     ret = bdrv_try_change_aio_context(target_bs, aio_context, NULL, errp);
1905     if (ret < 0) {
1906         aio_context_release(old_context);
1907         return;
1908     }
1909 
1910     aio_context_release(old_context);
1911     aio_context_acquire(aio_context);
1912     state->bs = bs;
1913 
1914     /* Paired with .clean() */
1915     bdrv_drained_begin(state->bs);
1916 
1917     state->job = do_backup_common(qapi_BlockdevBackup_base(backup),
1918                                   bs, target_bs, aio_context,
1919                                   block_job_txn, errp);
1920 
1921     aio_context_release(aio_context);
1922 }
1923 
1924 static void blockdev_backup_commit(void *opaque)
1925 {
1926     BlockdevBackupState *state = opaque;
1927     AioContext *aio_context;
1928 
1929     aio_context = bdrv_get_aio_context(state->bs);
1930     aio_context_acquire(aio_context);
1931 
1932     assert(state->job);
1933     job_start(&state->job->job);
1934 
1935     aio_context_release(aio_context);
1936 }
1937 
1938 static void blockdev_backup_abort(void *opaque)
1939 {
1940     BlockdevBackupState *state = opaque;
1941 
1942     if (state->job) {
1943         job_cancel_sync(&state->job->job, true);
1944     }
1945 }
1946 
1947 static void blockdev_backup_clean(void *opaque)
1948 {
1949     g_autofree BlockdevBackupState *state = opaque;
1950     AioContext *aio_context;
1951 
1952     if (!state->bs) {
1953         return;
1954     }
1955 
1956     aio_context = bdrv_get_aio_context(state->bs);
1957     aio_context_acquire(aio_context);
1958 
1959     bdrv_drained_end(state->bs);
1960 
1961     aio_context_release(aio_context);
1962 }
1963 
1964 typedef struct BlockDirtyBitmapState {
1965     BdrvDirtyBitmap *bitmap;
1966     BlockDriverState *bs;
1967     HBitmap *backup;
1968     bool was_enabled;
1969 } BlockDirtyBitmapState;
1970 
1971 static void block_dirty_bitmap_add_abort(void *opaque);
1972 TransactionActionDrv block_dirty_bitmap_add_drv = {
1973     .abort = block_dirty_bitmap_add_abort,
1974     .clean = g_free,
1975 };
1976 
1977 static void block_dirty_bitmap_add_action(BlockDirtyBitmapAdd *action,
1978                                           Transaction *tran, Error **errp)
1979 {
1980     Error *local_err = NULL;
1981     BlockDirtyBitmapState *state = g_new0(BlockDirtyBitmapState, 1);
1982 
1983     tran_add(tran, &block_dirty_bitmap_add_drv, state);
1984 
1985     /* AIO context taken and released within qmp_block_dirty_bitmap_add */
1986     qmp_block_dirty_bitmap_add(action->node, action->name,
1987                                action->has_granularity, action->granularity,
1988                                action->has_persistent, action->persistent,
1989                                action->has_disabled, action->disabled,
1990                                &local_err);
1991 
1992     if (!local_err) {
1993         state->bitmap = block_dirty_bitmap_lookup(action->node, action->name,
1994                                                   NULL, &error_abort);
1995     } else {
1996         error_propagate(errp, local_err);
1997     }
1998 }
1999 
2000 static void block_dirty_bitmap_add_abort(void *opaque)
2001 {
2002     BlockDirtyBitmapState *state = opaque;
2003 
2004     if (state->bitmap) {
2005         bdrv_release_dirty_bitmap(state->bitmap);
2006     }
2007 }
2008 
2009 static void block_dirty_bitmap_restore(void *opaque);
2010 static void block_dirty_bitmap_free_backup(void *opaque);
2011 TransactionActionDrv block_dirty_bitmap_clear_drv = {
2012     .abort = block_dirty_bitmap_restore,
2013     .commit = block_dirty_bitmap_free_backup,
2014     .clean = g_free,
2015 };
2016 
2017 static void block_dirty_bitmap_clear_action(BlockDirtyBitmap *action,
2018                                             Transaction *tran, Error **errp)
2019 {
2020     BlockDirtyBitmapState *state = g_new0(BlockDirtyBitmapState, 1);
2021 
2022     tran_add(tran, &block_dirty_bitmap_clear_drv, state);
2023 
2024     state->bitmap = block_dirty_bitmap_lookup(action->node,
2025                                               action->name,
2026                                               &state->bs,
2027                                               errp);
2028     if (!state->bitmap) {
2029         return;
2030     }
2031 
2032     if (bdrv_dirty_bitmap_check(state->bitmap, BDRV_BITMAP_DEFAULT, errp)) {
2033         return;
2034     }
2035 
2036     bdrv_clear_dirty_bitmap(state->bitmap, &state->backup);
2037 }
2038 
2039 static void block_dirty_bitmap_restore(void *opaque)
2040 {
2041     BlockDirtyBitmapState *state = opaque;
2042 
2043     if (state->backup) {
2044         bdrv_restore_dirty_bitmap(state->bitmap, state->backup);
2045     }
2046 }
2047 
2048 static void block_dirty_bitmap_free_backup(void *opaque)
2049 {
2050     BlockDirtyBitmapState *state = opaque;
2051 
2052     hbitmap_free(state->backup);
2053 }
2054 
2055 static void block_dirty_bitmap_enable_abort(void *opaque);
2056 TransactionActionDrv block_dirty_bitmap_enable_drv = {
2057     .abort = block_dirty_bitmap_enable_abort,
2058     .clean = g_free,
2059 };
2060 
2061 static void block_dirty_bitmap_enable_action(BlockDirtyBitmap *action,
2062                                              Transaction *tran, Error **errp)
2063 {
2064     BlockDirtyBitmapState *state = g_new0(BlockDirtyBitmapState, 1);
2065 
2066     tran_add(tran, &block_dirty_bitmap_enable_drv, state);
2067 
2068     state->bitmap = block_dirty_bitmap_lookup(action->node,
2069                                               action->name,
2070                                               NULL,
2071                                               errp);
2072     if (!state->bitmap) {
2073         return;
2074     }
2075 
2076     if (bdrv_dirty_bitmap_check(state->bitmap, BDRV_BITMAP_ALLOW_RO, errp)) {
2077         return;
2078     }
2079 
2080     state->was_enabled = bdrv_dirty_bitmap_enabled(state->bitmap);
2081     bdrv_enable_dirty_bitmap(state->bitmap);
2082 }
2083 
2084 static void block_dirty_bitmap_enable_abort(void *opaque)
2085 {
2086     BlockDirtyBitmapState *state = opaque;
2087 
2088     if (!state->was_enabled) {
2089         bdrv_disable_dirty_bitmap(state->bitmap);
2090     }
2091 }
2092 
2093 static void block_dirty_bitmap_disable_abort(void *opaque);
2094 TransactionActionDrv block_dirty_bitmap_disable_drv = {
2095     .abort = block_dirty_bitmap_disable_abort,
2096     .clean = g_free,
2097 };
2098 
2099 static void block_dirty_bitmap_disable_action(BlockDirtyBitmap *action,
2100                                               Transaction *tran, Error **errp)
2101 {
2102     BlockDirtyBitmapState *state = g_new0(BlockDirtyBitmapState, 1);
2103 
2104     tran_add(tran, &block_dirty_bitmap_disable_drv, state);
2105 
2106     state->bitmap = block_dirty_bitmap_lookup(action->node,
2107                                               action->name,
2108                                               NULL,
2109                                               errp);
2110     if (!state->bitmap) {
2111         return;
2112     }
2113 
2114     if (bdrv_dirty_bitmap_check(state->bitmap, BDRV_BITMAP_ALLOW_RO, errp)) {
2115         return;
2116     }
2117 
2118     state->was_enabled = bdrv_dirty_bitmap_enabled(state->bitmap);
2119     bdrv_disable_dirty_bitmap(state->bitmap);
2120 }
2121 
2122 static void block_dirty_bitmap_disable_abort(void *opaque)
2123 {
2124     BlockDirtyBitmapState *state = opaque;
2125 
2126     if (state->was_enabled) {
2127         bdrv_enable_dirty_bitmap(state->bitmap);
2128     }
2129 }
2130 
2131 TransactionActionDrv block_dirty_bitmap_merge_drv = {
2132     .commit = block_dirty_bitmap_free_backup,
2133     .abort = block_dirty_bitmap_restore,
2134     .clean = g_free,
2135 };
2136 
2137 static void block_dirty_bitmap_merge_action(BlockDirtyBitmapMerge *action,
2138                                             Transaction *tran, Error **errp)
2139 {
2140     BlockDirtyBitmapState *state = g_new0(BlockDirtyBitmapState, 1);
2141 
2142     tran_add(tran, &block_dirty_bitmap_merge_drv, state);
2143 
2144     state->bitmap = block_dirty_bitmap_merge(action->node, action->target,
2145                                              action->bitmaps, &state->backup,
2146                                              errp);
2147 }
2148 
2149 static void block_dirty_bitmap_remove_commit(void *opaque);
2150 static void block_dirty_bitmap_remove_abort(void *opaque);
2151 TransactionActionDrv block_dirty_bitmap_remove_drv = {
2152     .commit = block_dirty_bitmap_remove_commit,
2153     .abort = block_dirty_bitmap_remove_abort,
2154     .clean = g_free,
2155 };
2156 
2157 static void block_dirty_bitmap_remove_action(BlockDirtyBitmap *action,
2158                                              Transaction *tran, Error **errp)
2159 {
2160     BlockDirtyBitmapState *state = g_new0(BlockDirtyBitmapState, 1);
2161 
2162     tran_add(tran, &block_dirty_bitmap_remove_drv, state);
2163 
2164 
2165     state->bitmap = block_dirty_bitmap_remove(action->node, action->name,
2166                                               false, &state->bs, errp);
2167     if (state->bitmap) {
2168         bdrv_dirty_bitmap_skip_store(state->bitmap, true);
2169         bdrv_dirty_bitmap_set_busy(state->bitmap, true);
2170     }
2171 }
2172 
2173 static void block_dirty_bitmap_remove_abort(void *opaque)
2174 {
2175     BlockDirtyBitmapState *state = opaque;
2176 
2177     if (state->bitmap) {
2178         bdrv_dirty_bitmap_skip_store(state->bitmap, false);
2179         bdrv_dirty_bitmap_set_busy(state->bitmap, false);
2180     }
2181 }
2182 
2183 static void block_dirty_bitmap_remove_commit(void *opaque)
2184 {
2185     BlockDirtyBitmapState *state = opaque;
2186 
2187     bdrv_dirty_bitmap_set_busy(state->bitmap, false);
2188     bdrv_release_dirty_bitmap(state->bitmap);
2189 }
2190 
2191 static void abort_commit(void *opaque);
2192 TransactionActionDrv abort_drv = {
2193     .commit = abort_commit,
2194 };
2195 
2196 static void abort_action(Transaction *tran, Error **errp)
2197 {
2198     tran_add(tran, &abort_drv, NULL);
2199     error_setg(errp, "Transaction aborted using Abort action");
2200 }
2201 
2202 static void abort_commit(void *opaque)
2203 {
2204     g_assert_not_reached(); /* this action never succeeds */
2205 }
2206 
2207 static void transaction_action(TransactionAction *act, JobTxn *block_job_txn,
2208                                Transaction *tran, Error **errp)
2209 {
2210     switch (act->type) {
2211     case TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT:
2212     case TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC:
2213         external_snapshot_action(act, tran, errp);
2214         return;
2215     case TRANSACTION_ACTION_KIND_DRIVE_BACKUP:
2216         drive_backup_action(act->u.drive_backup.data,
2217                             block_job_txn, tran, errp);
2218         return;
2219     case TRANSACTION_ACTION_KIND_BLOCKDEV_BACKUP:
2220         blockdev_backup_action(act->u.blockdev_backup.data,
2221                                block_job_txn, tran, errp);
2222         return;
2223     case TRANSACTION_ACTION_KIND_ABORT:
2224         abort_action(tran, errp);
2225         return;
2226     case TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC:
2227         internal_snapshot_action(act->u.blockdev_snapshot_internal_sync.data,
2228                                  tran, errp);
2229         return;
2230     case TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_ADD:
2231         block_dirty_bitmap_add_action(act->u.block_dirty_bitmap_add.data,
2232                                       tran, errp);
2233         return;
2234     case TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_CLEAR:
2235         block_dirty_bitmap_clear_action(act->u.block_dirty_bitmap_clear.data,
2236                                         tran, errp);
2237         return;
2238     case TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_ENABLE:
2239         block_dirty_bitmap_enable_action(act->u.block_dirty_bitmap_enable.data,
2240                                          tran, errp);
2241         return;
2242     case TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_DISABLE:
2243         block_dirty_bitmap_disable_action(
2244                 act->u.block_dirty_bitmap_disable.data, tran, errp);
2245         return;
2246     case TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_MERGE:
2247         block_dirty_bitmap_merge_action(act->u.block_dirty_bitmap_merge.data,
2248                                         tran, errp);
2249         return;
2250     case TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_REMOVE:
2251         block_dirty_bitmap_remove_action(act->u.block_dirty_bitmap_remove.data,
2252                                          tran, errp);
2253         return;
2254     /*
2255      * Where are transactions for MIRROR, COMMIT and STREAM?
2256      * Although these blockjobs use transaction callbacks like the backup job,
2257      * these jobs do not necessarily adhere to transaction semantics.
2258      * These jobs may not fully undo all of their actions on abort, nor do they
2259      * necessarily work in transactions with more than one job in them.
2260      */
2261     case TRANSACTION_ACTION_KIND__MAX:
2262     default:
2263         g_assert_not_reached();
2264     };
2265 }
2266 
2267 
2268 /*
2269  * 'Atomic' group operations.  The operations are performed as a set, and if
2270  * any fail then we roll back all operations in the group.
2271  *
2272  * Always run under BQL.
2273  */
2274 void qmp_transaction(TransactionActionList *actions,
2275                      struct TransactionProperties *properties,
2276                      Error **errp)
2277 {
2278     TransactionActionList *act;
2279     JobTxn *block_job_txn = NULL;
2280     Error *local_err = NULL;
2281     Transaction *tran;
2282     ActionCompletionMode comp_mode =
2283         properties ? properties->completion_mode :
2284         ACTION_COMPLETION_MODE_INDIVIDUAL;
2285 
2286     GLOBAL_STATE_CODE();
2287 
2288     /* Does this transaction get canceled as a group on failure?
2289      * If not, we don't really need to make a JobTxn.
2290      */
2291     if (comp_mode != ACTION_COMPLETION_MODE_INDIVIDUAL) {
2292         for (act = actions; act; act = act->next) {
2293             TransactionActionKind type = act->value->type;
2294 
2295             if (type != TRANSACTION_ACTION_KIND_BLOCKDEV_BACKUP &&
2296                 type != TRANSACTION_ACTION_KIND_DRIVE_BACKUP)
2297             {
2298                 error_setg(errp,
2299                            "Action '%s' does not support transaction property "
2300                            "completion-mode = %s",
2301                            TransactionActionKind_str(type),
2302                            ActionCompletionMode_str(comp_mode));
2303                 return;
2304             }
2305         }
2306 
2307         block_job_txn = job_txn_new();
2308     }
2309 
2310     /* drain all i/o before any operations */
2311     bdrv_drain_all();
2312 
2313     tran = tran_new();
2314 
2315     /* We don't do anything in this loop that commits us to the operations */
2316     for (act = actions; act; act = act->next) {
2317         transaction_action(act->value, block_job_txn, tran, &local_err);
2318         if (local_err) {
2319             error_propagate(errp, local_err);
2320             goto delete_and_fail;
2321         }
2322     }
2323 
2324     tran_commit(tran);
2325 
2326     /* success */
2327     goto exit;
2328 
2329 delete_and_fail:
2330     /* failure, and it is all-or-none; roll back all operations */
2331     tran_abort(tran);
2332 exit:
2333     job_txn_unref(block_job_txn);
2334 }
2335 
2336 BlockDirtyBitmapSha256 *qmp_x_debug_block_dirty_bitmap_sha256(const char *node,
2337                                                               const char *name,
2338                                                               Error **errp)
2339 {
2340     BdrvDirtyBitmap *bitmap;
2341     BlockDriverState *bs;
2342     BlockDirtyBitmapSha256 *ret = NULL;
2343     char *sha256;
2344 
2345     bitmap = block_dirty_bitmap_lookup(node, name, &bs, errp);
2346     if (!bitmap || !bs) {
2347         return NULL;
2348     }
2349 
2350     sha256 = bdrv_dirty_bitmap_sha256(bitmap, errp);
2351     if (sha256 == NULL) {
2352         return NULL;
2353     }
2354 
2355     ret = g_new(BlockDirtyBitmapSha256, 1);
2356     ret->sha256 = sha256;
2357 
2358     return ret;
2359 }
2360 
2361 void coroutine_fn qmp_block_resize(const char *device, const char *node_name,
2362                                    int64_t size, Error **errp)
2363 {
2364     Error *local_err = NULL;
2365     BlockBackend *blk;
2366     BlockDriverState *bs;
2367     AioContext *old_ctx;
2368 
2369     bs = bdrv_lookup_bs(device, node_name, &local_err);
2370     if (local_err) {
2371         error_propagate(errp, local_err);
2372         return;
2373     }
2374 
2375     if (size < 0) {
2376         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "size", "a >0 size");
2377         return;
2378     }
2379 
2380     bdrv_graph_co_rdlock();
2381     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_RESIZE, NULL)) {
2382         error_setg(errp, QERR_DEVICE_IN_USE, device);
2383         bdrv_graph_co_rdunlock();
2384         return;
2385     }
2386     bdrv_graph_co_rdunlock();
2387 
2388     blk = blk_co_new_with_bs(bs, BLK_PERM_RESIZE, BLK_PERM_ALL, errp);
2389     if (!blk) {
2390         return;
2391     }
2392 
2393     bdrv_co_lock(bs);
2394     bdrv_drained_begin(bs);
2395     bdrv_co_unlock(bs);
2396 
2397     old_ctx = bdrv_co_enter(bs);
2398     blk_co_truncate(blk, size, false, PREALLOC_MODE_OFF, 0, errp);
2399     bdrv_co_leave(bs, old_ctx);
2400 
2401     bdrv_co_lock(bs);
2402     bdrv_drained_end(bs);
2403     blk_co_unref(blk);
2404     bdrv_co_unlock(bs);
2405 }
2406 
2407 void qmp_block_stream(const char *job_id, const char *device,
2408                       const char *base,
2409                       const char *base_node,
2410                       const char *backing_file,
2411                       const char *bottom,
2412                       bool has_speed, int64_t speed,
2413                       bool has_on_error, BlockdevOnError on_error,
2414                       const char *filter_node_name,
2415                       bool has_auto_finalize, bool auto_finalize,
2416                       bool has_auto_dismiss, bool auto_dismiss,
2417                       Error **errp)
2418 {
2419     BlockDriverState *bs, *iter, *iter_end;
2420     BlockDriverState *base_bs = NULL;
2421     BlockDriverState *bottom_bs = NULL;
2422     AioContext *aio_context;
2423     Error *local_err = NULL;
2424     int job_flags = JOB_DEFAULT;
2425 
2426     GLOBAL_STATE_CODE();
2427 
2428     if (base && base_node) {
2429         error_setg(errp, "'base' and 'base-node' cannot be specified "
2430                    "at the same time");
2431         return;
2432     }
2433 
2434     if (base && bottom) {
2435         error_setg(errp, "'base' and 'bottom' cannot be specified "
2436                    "at the same time");
2437         return;
2438     }
2439 
2440     if (bottom && base_node) {
2441         error_setg(errp, "'bottom' and 'base-node' cannot be specified "
2442                    "at the same time");
2443         return;
2444     }
2445 
2446     if (!has_on_error) {
2447         on_error = BLOCKDEV_ON_ERROR_REPORT;
2448     }
2449 
2450     bs = bdrv_lookup_bs(device, device, errp);
2451     if (!bs) {
2452         return;
2453     }
2454 
2455     aio_context = bdrv_get_aio_context(bs);
2456     aio_context_acquire(aio_context);
2457 
2458     bdrv_graph_rdlock_main_loop();
2459     if (base) {
2460         base_bs = bdrv_find_backing_image(bs, base);
2461         if (base_bs == NULL) {
2462             error_setg(errp, "Can't find '%s' in the backing chain", base);
2463             goto out_rdlock;
2464         }
2465         assert(bdrv_get_aio_context(base_bs) == aio_context);
2466     }
2467 
2468     if (base_node) {
2469         base_bs = bdrv_lookup_bs(NULL, base_node, errp);
2470         if (!base_bs) {
2471             goto out_rdlock;
2472         }
2473         if (bs == base_bs || !bdrv_chain_contains(bs, base_bs)) {
2474             error_setg(errp, "Node '%s' is not a backing image of '%s'",
2475                        base_node, device);
2476             goto out_rdlock;
2477         }
2478         assert(bdrv_get_aio_context(base_bs) == aio_context);
2479 
2480         bdrv_refresh_filename(base_bs);
2481     }
2482 
2483     if (bottom) {
2484         bottom_bs = bdrv_lookup_bs(NULL, bottom, errp);
2485         if (!bottom_bs) {
2486             goto out_rdlock;
2487         }
2488         if (!bottom_bs->drv) {
2489             error_setg(errp, "Node '%s' is not open", bottom);
2490             goto out_rdlock;
2491         }
2492         if (bottom_bs->drv->is_filter) {
2493             error_setg(errp, "Node '%s' is a filter, use a non-filter node "
2494                        "as 'bottom'", bottom);
2495             goto out_rdlock;
2496         }
2497         if (!bdrv_chain_contains(bs, bottom_bs)) {
2498             error_setg(errp, "Node '%s' is not in a chain starting from '%s'",
2499                        bottom, device);
2500             goto out_rdlock;
2501         }
2502         assert(bdrv_get_aio_context(bottom_bs) == aio_context);
2503     }
2504 
2505     /*
2506      * Check for op blockers in the whole chain between bs and base (or bottom)
2507      */
2508     iter_end = bottom ? bdrv_filter_or_cow_bs(bottom_bs) : base_bs;
2509     for (iter = bs; iter && iter != iter_end;
2510          iter = bdrv_filter_or_cow_bs(iter))
2511     {
2512         if (bdrv_op_is_blocked(iter, BLOCK_OP_TYPE_STREAM, errp)) {
2513             goto out_rdlock;
2514         }
2515     }
2516     bdrv_graph_rdunlock_main_loop();
2517 
2518     /* if we are streaming the entire chain, the result will have no backing
2519      * file, and specifying one is therefore an error */
2520     if (!base_bs && backing_file) {
2521         error_setg(errp, "backing file specified, but streaming the "
2522                          "entire chain");
2523         goto out;
2524     }
2525 
2526     if (has_auto_finalize && !auto_finalize) {
2527         job_flags |= JOB_MANUAL_FINALIZE;
2528     }
2529     if (has_auto_dismiss && !auto_dismiss) {
2530         job_flags |= JOB_MANUAL_DISMISS;
2531     }
2532 
2533     stream_start(job_id, bs, base_bs, backing_file,
2534                  bottom_bs, job_flags, has_speed ? speed : 0, on_error,
2535                  filter_node_name, &local_err);
2536     if (local_err) {
2537         error_propagate(errp, local_err);
2538         goto out;
2539     }
2540 
2541     trace_qmp_block_stream(bs);
2542 
2543 out:
2544     aio_context_release(aio_context);
2545     return;
2546 
2547 out_rdlock:
2548     bdrv_graph_rdunlock_main_loop();
2549     aio_context_release(aio_context);
2550 }
2551 
2552 void qmp_block_commit(const char *job_id, const char *device,
2553                       const char *base_node,
2554                       const char *base,
2555                       const char *top_node,
2556                       const char *top,
2557                       const char *backing_file,
2558                       bool has_speed, int64_t speed,
2559                       bool has_on_error, BlockdevOnError on_error,
2560                       const char *filter_node_name,
2561                       bool has_auto_finalize, bool auto_finalize,
2562                       bool has_auto_dismiss, bool auto_dismiss,
2563                       Error **errp)
2564 {
2565     BlockDriverState *bs;
2566     BlockDriverState *iter;
2567     BlockDriverState *base_bs, *top_bs;
2568     AioContext *aio_context;
2569     Error *local_err = NULL;
2570     int job_flags = JOB_DEFAULT;
2571     uint64_t top_perm, top_shared;
2572 
2573     /* TODO We'll eventually have to take a writer lock in this function */
2574     GRAPH_RDLOCK_GUARD_MAINLOOP();
2575 
2576     if (!has_speed) {
2577         speed = 0;
2578     }
2579     if (!has_on_error) {
2580         on_error = BLOCKDEV_ON_ERROR_REPORT;
2581     }
2582     if (has_auto_finalize && !auto_finalize) {
2583         job_flags |= JOB_MANUAL_FINALIZE;
2584     }
2585     if (has_auto_dismiss && !auto_dismiss) {
2586         job_flags |= JOB_MANUAL_DISMISS;
2587     }
2588 
2589     /* Important Note:
2590      *  libvirt relies on the DeviceNotFound error class in order to probe for
2591      *  live commit feature versions; for this to work, we must make sure to
2592      *  perform the device lookup before any generic errors that may occur in a
2593      *  scenario in which all optional arguments are omitted. */
2594     bs = qmp_get_root_bs(device, &local_err);
2595     if (!bs) {
2596         bs = bdrv_lookup_bs(device, device, NULL);
2597         if (!bs) {
2598             error_free(local_err);
2599             error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
2600                       "Device '%s' not found", device);
2601         } else {
2602             error_propagate(errp, local_err);
2603         }
2604         return;
2605     }
2606 
2607     aio_context = bdrv_get_aio_context(bs);
2608     aio_context_acquire(aio_context);
2609 
2610     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_COMMIT_SOURCE, errp)) {
2611         goto out;
2612     }
2613 
2614     /* default top_bs is the active layer */
2615     top_bs = bs;
2616 
2617     if (top_node && top) {
2618         error_setg(errp, "'top-node' and 'top' are mutually exclusive");
2619         goto out;
2620     } else if (top_node) {
2621         top_bs = bdrv_lookup_bs(NULL, top_node, errp);
2622         if (top_bs == NULL) {
2623             goto out;
2624         }
2625         if (!bdrv_chain_contains(bs, top_bs)) {
2626             error_setg(errp, "'%s' is not in this backing file chain",
2627                        top_node);
2628             goto out;
2629         }
2630     } else if (top) {
2631         /* This strcmp() is just a shortcut, there is no need to
2632          * refresh @bs's filename.  If it mismatches,
2633          * bdrv_find_backing_image() will do the refresh and may still
2634          * return @bs. */
2635         if (strcmp(bs->filename, top) != 0) {
2636             top_bs = bdrv_find_backing_image(bs, top);
2637         }
2638     }
2639 
2640     if (top_bs == NULL) {
2641         error_setg(errp, "Top image file %s not found", top ? top : "NULL");
2642         goto out;
2643     }
2644 
2645     assert(bdrv_get_aio_context(top_bs) == aio_context);
2646 
2647     if (base_node && base) {
2648         error_setg(errp, "'base-node' and 'base' are mutually exclusive");
2649         goto out;
2650     } else if (base_node) {
2651         base_bs = bdrv_lookup_bs(NULL, base_node, errp);
2652         if (base_bs == NULL) {
2653             goto out;
2654         }
2655         if (!bdrv_chain_contains(top_bs, base_bs)) {
2656             error_setg(errp, "'%s' is not in this backing file chain",
2657                        base_node);
2658             goto out;
2659         }
2660     } else if (base) {
2661         base_bs = bdrv_find_backing_image(top_bs, base);
2662         if (base_bs == NULL) {
2663             error_setg(errp, "Can't find '%s' in the backing chain", base);
2664             goto out;
2665         }
2666     } else {
2667         base_bs = bdrv_find_base(top_bs);
2668         if (base_bs == NULL) {
2669             error_setg(errp, "There is no backimg image");
2670             goto out;
2671         }
2672     }
2673 
2674     assert(bdrv_get_aio_context(base_bs) == aio_context);
2675 
2676     for (iter = top_bs; iter != bdrv_filter_or_cow_bs(base_bs);
2677          iter = bdrv_filter_or_cow_bs(iter))
2678     {
2679         if (bdrv_op_is_blocked(iter, BLOCK_OP_TYPE_COMMIT_TARGET, errp)) {
2680             goto out;
2681         }
2682     }
2683 
2684     /* Do not allow attempts to commit an image into itself */
2685     if (top_bs == base_bs) {
2686         error_setg(errp, "cannot commit an image into itself");
2687         goto out;
2688     }
2689 
2690     /*
2691      * Active commit is required if and only if someone has taken a
2692      * WRITE permission on the top node.  Historically, we have always
2693      * used active commit for top nodes, so continue that practice
2694      * lest we possibly break clients that rely on this behavior, e.g.
2695      * to later attach this node to a writing parent.
2696      * (Active commit is never really wrong.)
2697      */
2698     bdrv_get_cumulative_perm(top_bs, &top_perm, &top_shared);
2699     if (top_perm & BLK_PERM_WRITE ||
2700         bdrv_skip_filters(top_bs) == bdrv_skip_filters(bs))
2701     {
2702         if (backing_file) {
2703             if (bdrv_skip_filters(top_bs) == bdrv_skip_filters(bs)) {
2704                 error_setg(errp, "'backing-file' specified,"
2705                                  " but 'top' is the active layer");
2706             } else {
2707                 error_setg(errp, "'backing-file' specified, but 'top' has a "
2708                                  "writer on it");
2709             }
2710             goto out;
2711         }
2712         if (!job_id) {
2713             /*
2714              * Emulate here what block_job_create() does, because it
2715              * is possible that @bs != @top_bs (the block job should
2716              * be named after @bs, even if @top_bs is the actual
2717              * source)
2718              */
2719             job_id = bdrv_get_device_name(bs);
2720         }
2721         commit_active_start(job_id, top_bs, base_bs, job_flags, speed, on_error,
2722                             filter_node_name, NULL, NULL, false, &local_err);
2723     } else {
2724         BlockDriverState *overlay_bs = bdrv_find_overlay(bs, top_bs);
2725         if (bdrv_op_is_blocked(overlay_bs, BLOCK_OP_TYPE_COMMIT_TARGET, errp)) {
2726             goto out;
2727         }
2728         commit_start(job_id, bs, base_bs, top_bs, job_flags,
2729                      speed, on_error, backing_file,
2730                      filter_node_name, &local_err);
2731     }
2732     if (local_err != NULL) {
2733         error_propagate(errp, local_err);
2734         goto out;
2735     }
2736 
2737 out:
2738     aio_context_release(aio_context);
2739 }
2740 
2741 /* Common QMP interface for drive-backup and blockdev-backup */
2742 static BlockJob *do_backup_common(BackupCommon *backup,
2743                                   BlockDriverState *bs,
2744                                   BlockDriverState *target_bs,
2745                                   AioContext *aio_context,
2746                                   JobTxn *txn, Error **errp)
2747 {
2748     BlockJob *job = NULL;
2749     BdrvDirtyBitmap *bmap = NULL;
2750     BackupPerf perf = { .max_workers = 64 };
2751     int job_flags = JOB_DEFAULT;
2752 
2753     if (!backup->has_speed) {
2754         backup->speed = 0;
2755     }
2756     if (!backup->has_on_source_error) {
2757         backup->on_source_error = BLOCKDEV_ON_ERROR_REPORT;
2758     }
2759     if (!backup->has_on_target_error) {
2760         backup->on_target_error = BLOCKDEV_ON_ERROR_REPORT;
2761     }
2762     if (!backup->has_auto_finalize) {
2763         backup->auto_finalize = true;
2764     }
2765     if (!backup->has_auto_dismiss) {
2766         backup->auto_dismiss = true;
2767     }
2768     if (!backup->has_compress) {
2769         backup->compress = false;
2770     }
2771 
2772     if (backup->x_perf) {
2773         if (backup->x_perf->has_use_copy_range) {
2774             perf.use_copy_range = backup->x_perf->use_copy_range;
2775         }
2776         if (backup->x_perf->has_max_workers) {
2777             perf.max_workers = backup->x_perf->max_workers;
2778         }
2779         if (backup->x_perf->has_max_chunk) {
2780             perf.max_chunk = backup->x_perf->max_chunk;
2781         }
2782     }
2783 
2784     if ((backup->sync == MIRROR_SYNC_MODE_BITMAP) ||
2785         (backup->sync == MIRROR_SYNC_MODE_INCREMENTAL)) {
2786         /* done before desugaring 'incremental' to print the right message */
2787         if (!backup->bitmap) {
2788             error_setg(errp, "must provide a valid bitmap name for "
2789                        "'%s' sync mode", MirrorSyncMode_str(backup->sync));
2790             return NULL;
2791         }
2792     }
2793 
2794     if (backup->sync == MIRROR_SYNC_MODE_INCREMENTAL) {
2795         if (backup->has_bitmap_mode &&
2796             backup->bitmap_mode != BITMAP_SYNC_MODE_ON_SUCCESS) {
2797             error_setg(errp, "Bitmap sync mode must be '%s' "
2798                        "when using sync mode '%s'",
2799                        BitmapSyncMode_str(BITMAP_SYNC_MODE_ON_SUCCESS),
2800                        MirrorSyncMode_str(backup->sync));
2801             return NULL;
2802         }
2803         backup->has_bitmap_mode = true;
2804         backup->sync = MIRROR_SYNC_MODE_BITMAP;
2805         backup->bitmap_mode = BITMAP_SYNC_MODE_ON_SUCCESS;
2806     }
2807 
2808     if (backup->bitmap) {
2809         bmap = bdrv_find_dirty_bitmap(bs, backup->bitmap);
2810         if (!bmap) {
2811             error_setg(errp, "Bitmap '%s' could not be found", backup->bitmap);
2812             return NULL;
2813         }
2814         if (!backup->has_bitmap_mode) {
2815             error_setg(errp, "Bitmap sync mode must be given "
2816                        "when providing a bitmap");
2817             return NULL;
2818         }
2819         if (bdrv_dirty_bitmap_check(bmap, BDRV_BITMAP_ALLOW_RO, errp)) {
2820             return NULL;
2821         }
2822 
2823         /* This does not produce a useful bitmap artifact: */
2824         if (backup->sync == MIRROR_SYNC_MODE_NONE) {
2825             error_setg(errp, "sync mode '%s' does not produce meaningful bitmap"
2826                        " outputs", MirrorSyncMode_str(backup->sync));
2827             return NULL;
2828         }
2829 
2830         /* If the bitmap isn't used for input or output, this is useless: */
2831         if (backup->bitmap_mode == BITMAP_SYNC_MODE_NEVER &&
2832             backup->sync != MIRROR_SYNC_MODE_BITMAP) {
2833             error_setg(errp, "Bitmap sync mode '%s' has no meaningful effect"
2834                        " when combined with sync mode '%s'",
2835                        BitmapSyncMode_str(backup->bitmap_mode),
2836                        MirrorSyncMode_str(backup->sync));
2837             return NULL;
2838         }
2839     }
2840 
2841     if (!backup->bitmap && backup->has_bitmap_mode) {
2842         error_setg(errp, "Cannot specify bitmap sync mode without a bitmap");
2843         return NULL;
2844     }
2845 
2846     if (!backup->auto_finalize) {
2847         job_flags |= JOB_MANUAL_FINALIZE;
2848     }
2849     if (!backup->auto_dismiss) {
2850         job_flags |= JOB_MANUAL_DISMISS;
2851     }
2852 
2853     job = backup_job_create(backup->job_id, bs, target_bs, backup->speed,
2854                             backup->sync, bmap, backup->bitmap_mode,
2855                             backup->compress,
2856                             backup->filter_node_name,
2857                             &perf,
2858                             backup->on_source_error,
2859                             backup->on_target_error,
2860                             job_flags, NULL, NULL, txn, errp);
2861     return job;
2862 }
2863 
2864 void qmp_drive_backup(DriveBackup *backup, Error **errp)
2865 {
2866     TransactionAction action = {
2867         .type = TRANSACTION_ACTION_KIND_DRIVE_BACKUP,
2868         .u.drive_backup.data = backup,
2869     };
2870     blockdev_do_action(&action, errp);
2871 }
2872 
2873 BlockDeviceInfoList *qmp_query_named_block_nodes(bool has_flat,
2874                                                  bool flat,
2875                                                  Error **errp)
2876 {
2877     bool return_flat = has_flat && flat;
2878 
2879     return bdrv_named_nodes_list(return_flat, errp);
2880 }
2881 
2882 XDbgBlockGraph *qmp_x_debug_query_block_graph(Error **errp)
2883 {
2884     GRAPH_RDLOCK_GUARD_MAINLOOP();
2885 
2886     return bdrv_get_xdbg_block_graph(errp);
2887 }
2888 
2889 void qmp_blockdev_backup(BlockdevBackup *backup, Error **errp)
2890 {
2891     TransactionAction action = {
2892         .type = TRANSACTION_ACTION_KIND_BLOCKDEV_BACKUP,
2893         .u.blockdev_backup.data = backup,
2894     };
2895     blockdev_do_action(&action, errp);
2896 }
2897 
2898 /* Parameter check and block job starting for drive mirroring.
2899  * Caller should hold @device and @target's aio context (must be the same).
2900  **/
2901 static void blockdev_mirror_common(const char *job_id, BlockDriverState *bs,
2902                                    BlockDriverState *target,
2903                                    const char *replaces,
2904                                    enum MirrorSyncMode sync,
2905                                    BlockMirrorBackingMode backing_mode,
2906                                    bool zero_target,
2907                                    bool has_speed, int64_t speed,
2908                                    bool has_granularity, uint32_t granularity,
2909                                    bool has_buf_size, int64_t buf_size,
2910                                    bool has_on_source_error,
2911                                    BlockdevOnError on_source_error,
2912                                    bool has_on_target_error,
2913                                    BlockdevOnError on_target_error,
2914                                    bool has_unmap, bool unmap,
2915                                    const char *filter_node_name,
2916                                    bool has_copy_mode, MirrorCopyMode copy_mode,
2917                                    bool has_auto_finalize, bool auto_finalize,
2918                                    bool has_auto_dismiss, bool auto_dismiss,
2919                                    Error **errp)
2920 {
2921     BlockDriverState *unfiltered_bs;
2922     int job_flags = JOB_DEFAULT;
2923 
2924     GLOBAL_STATE_CODE();
2925     GRAPH_RDLOCK_GUARD_MAINLOOP();
2926 
2927     if (!has_speed) {
2928         speed = 0;
2929     }
2930     if (!has_on_source_error) {
2931         on_source_error = BLOCKDEV_ON_ERROR_REPORT;
2932     }
2933     if (!has_on_target_error) {
2934         on_target_error = BLOCKDEV_ON_ERROR_REPORT;
2935     }
2936     if (!has_granularity) {
2937         granularity = 0;
2938     }
2939     if (!has_buf_size) {
2940         buf_size = 0;
2941     }
2942     if (!has_unmap) {
2943         unmap = true;
2944     }
2945     if (!has_copy_mode) {
2946         copy_mode = MIRROR_COPY_MODE_BACKGROUND;
2947     }
2948     if (has_auto_finalize && !auto_finalize) {
2949         job_flags |= JOB_MANUAL_FINALIZE;
2950     }
2951     if (has_auto_dismiss && !auto_dismiss) {
2952         job_flags |= JOB_MANUAL_DISMISS;
2953     }
2954 
2955     if (granularity != 0 && (granularity < 512 || granularity > 1048576 * 64)) {
2956         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "granularity",
2957                    "a value in range [512B, 64MB]");
2958         return;
2959     }
2960     if (granularity & (granularity - 1)) {
2961         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "granularity",
2962                    "a power of 2");
2963         return;
2964     }
2965 
2966     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_MIRROR_SOURCE, errp)) {
2967         return;
2968     }
2969     if (bdrv_op_is_blocked(target, BLOCK_OP_TYPE_MIRROR_TARGET, errp)) {
2970         return;
2971     }
2972 
2973     if (!bdrv_backing_chain_next(bs) && sync == MIRROR_SYNC_MODE_TOP) {
2974         sync = MIRROR_SYNC_MODE_FULL;
2975     }
2976 
2977     if (!replaces) {
2978         /* We want to mirror from @bs, but keep implicit filters on top */
2979         unfiltered_bs = bdrv_skip_implicit_filters(bs);
2980         if (unfiltered_bs != bs) {
2981             replaces = unfiltered_bs->node_name;
2982         }
2983     }
2984 
2985     if (replaces) {
2986         BlockDriverState *to_replace_bs;
2987         AioContext *aio_context;
2988         AioContext *replace_aio_context;
2989         int64_t bs_size, replace_size;
2990 
2991         bs_size = bdrv_getlength(bs);
2992         if (bs_size < 0) {
2993             error_setg_errno(errp, -bs_size, "Failed to query device's size");
2994             return;
2995         }
2996 
2997         to_replace_bs = check_to_replace_node(bs, replaces, errp);
2998         if (!to_replace_bs) {
2999             return;
3000         }
3001 
3002         aio_context = bdrv_get_aio_context(bs);
3003         replace_aio_context = bdrv_get_aio_context(to_replace_bs);
3004         /*
3005          * bdrv_getlength() is a co-wrapper and uses AIO_WAIT_WHILE. Be sure not
3006          * to acquire the same AioContext twice.
3007          */
3008         if (replace_aio_context != aio_context) {
3009             aio_context_acquire(replace_aio_context);
3010         }
3011         replace_size = bdrv_getlength(to_replace_bs);
3012         if (replace_aio_context != aio_context) {
3013             aio_context_release(replace_aio_context);
3014         }
3015 
3016         if (replace_size < 0) {
3017             error_setg_errno(errp, -replace_size,
3018                              "Failed to query the replacement node's size");
3019             return;
3020         }
3021         if (bs_size != replace_size) {
3022             error_setg(errp, "cannot replace image with a mirror image of "
3023                              "different size");
3024             return;
3025         }
3026     }
3027 
3028     /* pass the node name to replace to mirror start since it's loose coupling
3029      * and will allow to check whether the node still exist at mirror completion
3030      */
3031     mirror_start(job_id, bs, target,
3032                  replaces, job_flags,
3033                  speed, granularity, buf_size, sync, backing_mode, zero_target,
3034                  on_source_error, on_target_error, unmap, filter_node_name,
3035                  copy_mode, errp);
3036 }
3037 
3038 void qmp_drive_mirror(DriveMirror *arg, Error **errp)
3039 {
3040     BlockDriverState *bs;
3041     BlockDriverState *target_backing_bs, *target_bs;
3042     AioContext *aio_context;
3043     AioContext *old_context;
3044     BlockMirrorBackingMode backing_mode;
3045     Error *local_err = NULL;
3046     QDict *options = NULL;
3047     int flags;
3048     int64_t size;
3049     const char *format = arg->format;
3050     bool zero_target;
3051     int ret;
3052 
3053     bs = qmp_get_root_bs(arg->device, errp);
3054     if (!bs) {
3055         return;
3056     }
3057 
3058     /* Early check to avoid creating target */
3059     bdrv_graph_rdlock_main_loop();
3060     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_MIRROR_SOURCE, errp)) {
3061         bdrv_graph_rdunlock_main_loop();
3062         return;
3063     }
3064 
3065     aio_context = bdrv_get_aio_context(bs);
3066     aio_context_acquire(aio_context);
3067 
3068     if (!arg->has_mode) {
3069         arg->mode = NEW_IMAGE_MODE_ABSOLUTE_PATHS;
3070     }
3071 
3072     if (!arg->format) {
3073         format = (arg->mode == NEW_IMAGE_MODE_EXISTING
3074                   ? NULL : bs->drv->format_name);
3075     }
3076 
3077     flags = bs->open_flags | BDRV_O_RDWR;
3078     target_backing_bs = bdrv_cow_bs(bdrv_skip_filters(bs));
3079     if (!target_backing_bs && arg->sync == MIRROR_SYNC_MODE_TOP) {
3080         arg->sync = MIRROR_SYNC_MODE_FULL;
3081     }
3082     if (arg->sync == MIRROR_SYNC_MODE_NONE) {
3083         target_backing_bs = bs;
3084     }
3085     bdrv_graph_rdunlock_main_loop();
3086 
3087     size = bdrv_getlength(bs);
3088     if (size < 0) {
3089         error_setg_errno(errp, -size, "bdrv_getlength failed");
3090         goto out;
3091     }
3092 
3093     if (arg->replaces) {
3094         if (!arg->node_name) {
3095             error_setg(errp, "a node-name must be provided when replacing a"
3096                              " named node of the graph");
3097             goto out;
3098         }
3099     }
3100 
3101     if (arg->mode == NEW_IMAGE_MODE_ABSOLUTE_PATHS) {
3102         backing_mode = MIRROR_SOURCE_BACKING_CHAIN;
3103     } else {
3104         backing_mode = MIRROR_OPEN_BACKING_CHAIN;
3105     }
3106 
3107     /* Don't open backing image in create() */
3108     flags |= BDRV_O_NO_BACKING;
3109 
3110     if ((arg->sync == MIRROR_SYNC_MODE_FULL || !target_backing_bs)
3111         && arg->mode != NEW_IMAGE_MODE_EXISTING)
3112     {
3113         /* create new image w/o backing file */
3114         assert(format);
3115         bdrv_img_create(arg->target, format,
3116                         NULL, NULL, NULL, size, flags, false, &local_err);
3117     } else {
3118         BlockDriverState *explicit_backing;
3119 
3120         switch (arg->mode) {
3121         case NEW_IMAGE_MODE_EXISTING:
3122             break;
3123         case NEW_IMAGE_MODE_ABSOLUTE_PATHS:
3124             /*
3125              * Create new image with backing file.
3126              * Implicit filters should not appear in the filename.
3127              */
3128             bdrv_graph_rdlock_main_loop();
3129             explicit_backing = bdrv_skip_implicit_filters(target_backing_bs);
3130             bdrv_refresh_filename(explicit_backing);
3131             bdrv_graph_rdunlock_main_loop();
3132 
3133             bdrv_img_create(arg->target, format,
3134                             explicit_backing->filename,
3135                             explicit_backing->drv->format_name,
3136                             NULL, size, flags, false, &local_err);
3137             break;
3138         default:
3139             abort();
3140         }
3141     }
3142 
3143     if (local_err) {
3144         error_propagate(errp, local_err);
3145         goto out;
3146     }
3147 
3148     options = qdict_new();
3149     if (arg->node_name) {
3150         qdict_put_str(options, "node-name", arg->node_name);
3151     }
3152     if (format) {
3153         qdict_put_str(options, "driver", format);
3154     }
3155     aio_context_release(aio_context);
3156 
3157     /* Mirroring takes care of copy-on-write using the source's backing
3158      * file.
3159      */
3160     aio_context_acquire(qemu_get_aio_context());
3161     target_bs = bdrv_open(arg->target, NULL, options, flags, errp);
3162     aio_context_release(qemu_get_aio_context());
3163 
3164     if (!target_bs) {
3165         return;
3166     }
3167 
3168     bdrv_graph_rdlock_main_loop();
3169     zero_target = (arg->sync == MIRROR_SYNC_MODE_FULL &&
3170                    (arg->mode == NEW_IMAGE_MODE_EXISTING ||
3171                     !bdrv_has_zero_init(target_bs)));
3172     bdrv_graph_rdunlock_main_loop();
3173 
3174 
3175     /* Honor bdrv_try_change_aio_context() context acquisition requirements. */
3176     old_context = bdrv_get_aio_context(target_bs);
3177     aio_context_acquire(old_context);
3178 
3179     ret = bdrv_try_change_aio_context(target_bs, aio_context, NULL, errp);
3180     if (ret < 0) {
3181         bdrv_unref(target_bs);
3182         aio_context_release(old_context);
3183         return;
3184     }
3185 
3186     aio_context_release(old_context);
3187     aio_context_acquire(aio_context);
3188 
3189     blockdev_mirror_common(arg->job_id, bs, target_bs,
3190                            arg->replaces, arg->sync,
3191                            backing_mode, zero_target,
3192                            arg->has_speed, arg->speed,
3193                            arg->has_granularity, arg->granularity,
3194                            arg->has_buf_size, arg->buf_size,
3195                            arg->has_on_source_error, arg->on_source_error,
3196                            arg->has_on_target_error, arg->on_target_error,
3197                            arg->has_unmap, arg->unmap,
3198                            NULL,
3199                            arg->has_copy_mode, arg->copy_mode,
3200                            arg->has_auto_finalize, arg->auto_finalize,
3201                            arg->has_auto_dismiss, arg->auto_dismiss,
3202                            errp);
3203     bdrv_unref(target_bs);
3204 out:
3205     aio_context_release(aio_context);
3206 }
3207 
3208 void qmp_blockdev_mirror(const char *job_id,
3209                          const char *device, const char *target,
3210                          const char *replaces,
3211                          MirrorSyncMode sync,
3212                          bool has_speed, int64_t speed,
3213                          bool has_granularity, uint32_t granularity,
3214                          bool has_buf_size, int64_t buf_size,
3215                          bool has_on_source_error,
3216                          BlockdevOnError on_source_error,
3217                          bool has_on_target_error,
3218                          BlockdevOnError on_target_error,
3219                          const char *filter_node_name,
3220                          bool has_copy_mode, MirrorCopyMode copy_mode,
3221                          bool has_auto_finalize, bool auto_finalize,
3222                          bool has_auto_dismiss, bool auto_dismiss,
3223                          Error **errp)
3224 {
3225     BlockDriverState *bs;
3226     BlockDriverState *target_bs;
3227     AioContext *aio_context;
3228     AioContext *old_context;
3229     BlockMirrorBackingMode backing_mode = MIRROR_LEAVE_BACKING_CHAIN;
3230     bool zero_target;
3231     int ret;
3232 
3233     bs = qmp_get_root_bs(device, errp);
3234     if (!bs) {
3235         return;
3236     }
3237 
3238     target_bs = bdrv_lookup_bs(target, target, errp);
3239     if (!target_bs) {
3240         return;
3241     }
3242 
3243     zero_target = (sync == MIRROR_SYNC_MODE_FULL);
3244 
3245     /* Honor bdrv_try_change_aio_context() context acquisition requirements. */
3246     old_context = bdrv_get_aio_context(target_bs);
3247     aio_context = bdrv_get_aio_context(bs);
3248     aio_context_acquire(old_context);
3249 
3250     ret = bdrv_try_change_aio_context(target_bs, aio_context, NULL, errp);
3251 
3252     aio_context_release(old_context);
3253     aio_context_acquire(aio_context);
3254 
3255     if (ret < 0) {
3256         goto out;
3257     }
3258 
3259     blockdev_mirror_common(job_id, bs, target_bs,
3260                            replaces, sync, backing_mode,
3261                            zero_target, has_speed, speed,
3262                            has_granularity, granularity,
3263                            has_buf_size, buf_size,
3264                            has_on_source_error, on_source_error,
3265                            has_on_target_error, on_target_error,
3266                            true, true, filter_node_name,
3267                            has_copy_mode, copy_mode,
3268                            has_auto_finalize, auto_finalize,
3269                            has_auto_dismiss, auto_dismiss,
3270                            errp);
3271 out:
3272     aio_context_release(aio_context);
3273 }
3274 
3275 /*
3276  * Get a block job using its ID. Called with job_mutex held.
3277  */
3278 static BlockJob *find_block_job_locked(const char *id, Error **errp)
3279 {
3280     BlockJob *job;
3281 
3282     assert(id != NULL);
3283 
3284     job = block_job_get_locked(id);
3285 
3286     if (!job) {
3287         error_set(errp, ERROR_CLASS_DEVICE_NOT_ACTIVE,
3288                   "Block job '%s' not found", id);
3289         return NULL;
3290     }
3291 
3292     return job;
3293 }
3294 
3295 void qmp_block_job_set_speed(const char *device, int64_t speed, Error **errp)
3296 {
3297     BlockJob *job;
3298 
3299     JOB_LOCK_GUARD();
3300     job = find_block_job_locked(device, errp);
3301 
3302     if (!job) {
3303         return;
3304     }
3305 
3306     block_job_set_speed_locked(job, speed, errp);
3307 }
3308 
3309 void qmp_block_job_cancel(const char *device,
3310                           bool has_force, bool force, Error **errp)
3311 {
3312     BlockJob *job;
3313 
3314     JOB_LOCK_GUARD();
3315     job = find_block_job_locked(device, errp);
3316 
3317     if (!job) {
3318         return;
3319     }
3320 
3321     if (!has_force) {
3322         force = false;
3323     }
3324 
3325     if (job_user_paused_locked(&job->job) && !force) {
3326         error_setg(errp, "The block job for device '%s' is currently paused",
3327                    device);
3328         return;
3329     }
3330 
3331     trace_qmp_block_job_cancel(job);
3332     job_user_cancel_locked(&job->job, force, errp);
3333 }
3334 
3335 void qmp_block_job_pause(const char *device, Error **errp)
3336 {
3337     BlockJob *job;
3338 
3339     JOB_LOCK_GUARD();
3340     job = find_block_job_locked(device, errp);
3341 
3342     if (!job) {
3343         return;
3344     }
3345 
3346     trace_qmp_block_job_pause(job);
3347     job_user_pause_locked(&job->job, errp);
3348 }
3349 
3350 void qmp_block_job_resume(const char *device, Error **errp)
3351 {
3352     BlockJob *job;
3353 
3354     JOB_LOCK_GUARD();
3355     job = find_block_job_locked(device, errp);
3356 
3357     if (!job) {
3358         return;
3359     }
3360 
3361     trace_qmp_block_job_resume(job);
3362     job_user_resume_locked(&job->job, errp);
3363 }
3364 
3365 void qmp_block_job_complete(const char *device, Error **errp)
3366 {
3367     BlockJob *job;
3368 
3369     JOB_LOCK_GUARD();
3370     job = find_block_job_locked(device, errp);
3371 
3372     if (!job) {
3373         return;
3374     }
3375 
3376     trace_qmp_block_job_complete(job);
3377     job_complete_locked(&job->job, errp);
3378 }
3379 
3380 void qmp_block_job_finalize(const char *id, Error **errp)
3381 {
3382     BlockJob *job;
3383 
3384     JOB_LOCK_GUARD();
3385     job = find_block_job_locked(id, errp);
3386 
3387     if (!job) {
3388         return;
3389     }
3390 
3391     trace_qmp_block_job_finalize(job);
3392     job_ref_locked(&job->job);
3393     job_finalize_locked(&job->job, errp);
3394 
3395     job_unref_locked(&job->job);
3396 }
3397 
3398 void qmp_block_job_dismiss(const char *id, Error **errp)
3399 {
3400     BlockJob *bjob;
3401     Job *job;
3402 
3403     JOB_LOCK_GUARD();
3404     bjob = find_block_job_locked(id, errp);
3405 
3406     if (!bjob) {
3407         return;
3408     }
3409 
3410     trace_qmp_block_job_dismiss(bjob);
3411     job = &bjob->job;
3412     job_dismiss_locked(&job, errp);
3413 }
3414 
3415 void qmp_block_job_change(BlockJobChangeOptions *opts, Error **errp)
3416 {
3417     BlockJob *job;
3418 
3419     JOB_LOCK_GUARD();
3420     job = find_block_job_locked(opts->id, errp);
3421 
3422     if (!job) {
3423         return;
3424     }
3425 
3426     block_job_change_locked(job, opts, errp);
3427 }
3428 
3429 void qmp_change_backing_file(const char *device,
3430                              const char *image_node_name,
3431                              const char *backing_file,
3432                              Error **errp)
3433 {
3434     BlockDriverState *bs = NULL;
3435     AioContext *aio_context;
3436     BlockDriverState *image_bs = NULL;
3437     Error *local_err = NULL;
3438     bool ro;
3439     int ret;
3440 
3441     bs = qmp_get_root_bs(device, errp);
3442     if (!bs) {
3443         return;
3444     }
3445 
3446     aio_context = bdrv_get_aio_context(bs);
3447     aio_context_acquire(aio_context);
3448 
3449     bdrv_graph_rdlock_main_loop();
3450 
3451     image_bs = bdrv_lookup_bs(NULL, image_node_name, &local_err);
3452     if (local_err) {
3453         error_propagate(errp, local_err);
3454         goto out_rdlock;
3455     }
3456 
3457     if (!image_bs) {
3458         error_setg(errp, "image file not found");
3459         goto out_rdlock;
3460     }
3461 
3462     if (bdrv_find_base(image_bs) == image_bs) {
3463         error_setg(errp, "not allowing backing file change on an image "
3464                          "without a backing file");
3465         goto out_rdlock;
3466     }
3467 
3468     /* even though we are not necessarily operating on bs, we need it to
3469      * determine if block ops are currently prohibited on the chain */
3470     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_CHANGE, errp)) {
3471         goto out_rdlock;
3472     }
3473 
3474     /* final sanity check */
3475     if (!bdrv_chain_contains(bs, image_bs)) {
3476         error_setg(errp, "'%s' and image file are not in the same chain",
3477                    device);
3478         goto out_rdlock;
3479     }
3480     bdrv_graph_rdunlock_main_loop();
3481 
3482     /* if not r/w, reopen to make r/w */
3483     ro = bdrv_is_read_only(image_bs);
3484 
3485     if (ro) {
3486         if (bdrv_reopen_set_read_only(image_bs, false, errp) != 0) {
3487             goto out;
3488         }
3489     }
3490 
3491     ret = bdrv_change_backing_file(image_bs, backing_file,
3492                                    image_bs->drv ? image_bs->drv->format_name : "",
3493                                    false);
3494 
3495     if (ret < 0) {
3496         error_setg_errno(errp, -ret, "Could not change backing file to '%s'",
3497                          backing_file);
3498         /* don't exit here, so we can try to restore open flags if
3499          * appropriate */
3500     }
3501 
3502     if (ro) {
3503         bdrv_reopen_set_read_only(image_bs, true, errp);
3504     }
3505 
3506 out:
3507     aio_context_release(aio_context);
3508     return;
3509 
3510 out_rdlock:
3511     bdrv_graph_rdunlock_main_loop();
3512     aio_context_release(aio_context);
3513 }
3514 
3515 void qmp_blockdev_add(BlockdevOptions *options, Error **errp)
3516 {
3517     BlockDriverState *bs;
3518     QObject *obj;
3519     Visitor *v = qobject_output_visitor_new(&obj);
3520     QDict *qdict;
3521 
3522     visit_type_BlockdevOptions(v, NULL, &options, &error_abort);
3523     visit_complete(v, &obj);
3524     qdict = qobject_to(QDict, obj);
3525 
3526     qdict_flatten(qdict);
3527 
3528     if (!qdict_get_try_str(qdict, "node-name")) {
3529         error_setg(errp, "'node-name' must be specified for the root node");
3530         goto fail;
3531     }
3532 
3533     bs = bds_tree_init(qdict, errp);
3534     if (!bs) {
3535         goto fail;
3536     }
3537 
3538     bdrv_set_monitor_owned(bs);
3539 
3540 fail:
3541     visit_free(v);
3542 }
3543 
3544 void qmp_blockdev_reopen(BlockdevOptionsList *reopen_list, Error **errp)
3545 {
3546     BlockReopenQueue *queue = NULL;
3547 
3548     /* Add each one of the BDS that we want to reopen to the queue */
3549     for (; reopen_list != NULL; reopen_list = reopen_list->next) {
3550         BlockdevOptions *options = reopen_list->value;
3551         BlockDriverState *bs;
3552         AioContext *ctx;
3553         QObject *obj;
3554         Visitor *v;
3555         QDict *qdict;
3556 
3557         /* Check for the selected node name */
3558         if (!options->node_name) {
3559             error_setg(errp, "node-name not specified");
3560             goto fail;
3561         }
3562 
3563         bs = bdrv_find_node(options->node_name);
3564         if (!bs) {
3565             error_setg(errp, "Failed to find node with node-name='%s'",
3566                        options->node_name);
3567             goto fail;
3568         }
3569 
3570         /* Put all options in a QDict and flatten it */
3571         v = qobject_output_visitor_new(&obj);
3572         visit_type_BlockdevOptions(v, NULL, &options, &error_abort);
3573         visit_complete(v, &obj);
3574         visit_free(v);
3575 
3576         qdict = qobject_to(QDict, obj);
3577 
3578         qdict_flatten(qdict);
3579 
3580         ctx = bdrv_get_aio_context(bs);
3581         aio_context_acquire(ctx);
3582 
3583         queue = bdrv_reopen_queue(queue, bs, qdict, false);
3584 
3585         aio_context_release(ctx);
3586     }
3587 
3588     /* Perform the reopen operation */
3589     bdrv_reopen_multiple(queue, errp);
3590     queue = NULL;
3591 
3592 fail:
3593     bdrv_reopen_queue_free(queue);
3594 }
3595 
3596 void qmp_blockdev_del(const char *node_name, Error **errp)
3597 {
3598     AioContext *aio_context;
3599     BlockDriverState *bs;
3600 
3601     GLOBAL_STATE_CODE();
3602     GRAPH_RDLOCK_GUARD_MAINLOOP();
3603 
3604     bs = bdrv_find_node(node_name);
3605     if (!bs) {
3606         error_setg(errp, "Failed to find node with node-name='%s'", node_name);
3607         return;
3608     }
3609     if (bdrv_has_blk(bs)) {
3610         error_setg(errp, "Node %s is in use", node_name);
3611         return;
3612     }
3613     aio_context = bdrv_get_aio_context(bs);
3614     aio_context_acquire(aio_context);
3615 
3616     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_DRIVE_DEL, errp)) {
3617         goto out;
3618     }
3619 
3620     if (!QTAILQ_IN_USE(bs, monitor_list)) {
3621         error_setg(errp, "Node %s is not owned by the monitor",
3622                    bs->node_name);
3623         goto out;
3624     }
3625 
3626     if (bs->refcnt > 1) {
3627         error_setg(errp, "Block device %s is in use",
3628                    bdrv_get_device_or_node_name(bs));
3629         goto out;
3630     }
3631 
3632     QTAILQ_REMOVE(&monitor_bdrv_states, bs, monitor_list);
3633     bdrv_unref(bs);
3634 
3635 out:
3636     aio_context_release(aio_context);
3637 }
3638 
3639 static BdrvChild * GRAPH_RDLOCK
3640 bdrv_find_child(BlockDriverState *parent_bs, const char *child_name)
3641 {
3642     BdrvChild *child;
3643 
3644     QLIST_FOREACH(child, &parent_bs->children, next) {
3645         if (strcmp(child->name, child_name) == 0) {
3646             return child;
3647         }
3648     }
3649 
3650     return NULL;
3651 }
3652 
3653 void qmp_x_blockdev_change(const char *parent, const char *child,
3654                            const char *node, Error **errp)
3655 {
3656     BlockDriverState *parent_bs, *new_bs = NULL;
3657     BdrvChild *p_child;
3658 
3659     bdrv_graph_wrlock(NULL);
3660 
3661     parent_bs = bdrv_lookup_bs(parent, parent, errp);
3662     if (!parent_bs) {
3663         goto out;
3664     }
3665 
3666     if (!child == !node) {
3667         if (child) {
3668             error_setg(errp, "The parameters child and node are in conflict");
3669         } else {
3670             error_setg(errp, "Either child or node must be specified");
3671         }
3672         goto out;
3673     }
3674 
3675     if (child) {
3676         p_child = bdrv_find_child(parent_bs, child);
3677         if (!p_child) {
3678             error_setg(errp, "Node '%s' does not have child '%s'",
3679                        parent, child);
3680             goto out;
3681         }
3682         bdrv_del_child(parent_bs, p_child, errp);
3683     }
3684 
3685     if (node) {
3686         new_bs = bdrv_find_node(node);
3687         if (!new_bs) {
3688             error_setg(errp, "Node '%s' not found", node);
3689             goto out;
3690         }
3691         bdrv_add_child(parent_bs, new_bs, errp);
3692     }
3693 
3694 out:
3695     bdrv_graph_wrunlock();
3696 }
3697 
3698 BlockJobInfoList *qmp_query_block_jobs(Error **errp)
3699 {
3700     BlockJobInfoList *head = NULL, **tail = &head;
3701     BlockJob *job;
3702 
3703     JOB_LOCK_GUARD();
3704 
3705     for (job = block_job_next_locked(NULL); job;
3706          job = block_job_next_locked(job)) {
3707         BlockJobInfo *value;
3708 
3709         if (block_job_is_internal(job)) {
3710             continue;
3711         }
3712         value = block_job_query_locked(job, errp);
3713         if (!value) {
3714             qapi_free_BlockJobInfoList(head);
3715             return NULL;
3716         }
3717         QAPI_LIST_APPEND(tail, value);
3718     }
3719 
3720     return head;
3721 }
3722 
3723 void qmp_x_blockdev_set_iothread(const char *node_name, StrOrNull *iothread,
3724                                  bool has_force, bool force, Error **errp)
3725 {
3726     AioContext *old_context;
3727     AioContext *new_context;
3728     BlockDriverState *bs;
3729 
3730     GRAPH_RDLOCK_GUARD_MAINLOOP();
3731 
3732     bs = bdrv_find_node(node_name);
3733     if (!bs) {
3734         error_setg(errp, "Failed to find node with node-name='%s'", node_name);
3735         return;
3736     }
3737 
3738     /* Protects against accidents. */
3739     if (!(has_force && force) && bdrv_has_blk(bs)) {
3740         error_setg(errp, "Node %s is associated with a BlockBackend and could "
3741                          "be in use (use force=true to override this check)",
3742                          node_name);
3743         return;
3744     }
3745 
3746     if (iothread->type == QTYPE_QSTRING) {
3747         IOThread *obj = iothread_by_id(iothread->u.s);
3748         if (!obj) {
3749             error_setg(errp, "Cannot find iothread %s", iothread->u.s);
3750             return;
3751         }
3752 
3753         new_context = iothread_get_aio_context(obj);
3754     } else {
3755         new_context = qemu_get_aio_context();
3756     }
3757 
3758     old_context = bdrv_get_aio_context(bs);
3759     aio_context_acquire(old_context);
3760 
3761     bdrv_try_change_aio_context(bs, new_context, NULL, errp);
3762 
3763     aio_context_release(old_context);
3764 }
3765 
3766 QemuOptsList qemu_common_drive_opts = {
3767     .name = "drive",
3768     .head = QTAILQ_HEAD_INITIALIZER(qemu_common_drive_opts.head),
3769     .desc = {
3770         {
3771             .name = "snapshot",
3772             .type = QEMU_OPT_BOOL,
3773             .help = "enable/disable snapshot mode",
3774         },{
3775             .name = "aio",
3776             .type = QEMU_OPT_STRING,
3777             .help = "host AIO implementation (threads, native, io_uring)",
3778         },{
3779             .name = BDRV_OPT_CACHE_WB,
3780             .type = QEMU_OPT_BOOL,
3781             .help = "Enable writeback mode",
3782         },{
3783             .name = "format",
3784             .type = QEMU_OPT_STRING,
3785             .help = "disk format (raw, qcow2, ...)",
3786         },{
3787             .name = "rerror",
3788             .type = QEMU_OPT_STRING,
3789             .help = "read error action",
3790         },{
3791             .name = "werror",
3792             .type = QEMU_OPT_STRING,
3793             .help = "write error action",
3794         },{
3795             .name = BDRV_OPT_READ_ONLY,
3796             .type = QEMU_OPT_BOOL,
3797             .help = "open drive file as read-only",
3798         },
3799 
3800         THROTTLE_OPTS,
3801 
3802         {
3803             .name = "throttling.group",
3804             .type = QEMU_OPT_STRING,
3805             .help = "name of the block throttling group",
3806         },{
3807             .name = "copy-on-read",
3808             .type = QEMU_OPT_BOOL,
3809             .help = "copy read data from backing file into image file",
3810         },{
3811             .name = "detect-zeroes",
3812             .type = QEMU_OPT_STRING,
3813             .help = "try to optimize zero writes (off, on, unmap)",
3814         },{
3815             .name = "stats-account-invalid",
3816             .type = QEMU_OPT_BOOL,
3817             .help = "whether to account for invalid I/O operations "
3818                     "in the statistics",
3819         },{
3820             .name = "stats-account-failed",
3821             .type = QEMU_OPT_BOOL,
3822             .help = "whether to account for failed I/O operations "
3823                     "in the statistics",
3824         },
3825         { /* end of list */ }
3826     },
3827 };
3828 
3829 QemuOptsList qemu_drive_opts = {
3830     .name = "drive",
3831     .head = QTAILQ_HEAD_INITIALIZER(qemu_drive_opts.head),
3832     .desc = {
3833         /*
3834          * no elements => accept any params
3835          * validation will happen later
3836          */
3837         { /* end of list */ }
3838     },
3839 };
3840