xref: /openbmc/qemu/blockdev.c (revision b14df228)
1 /*
2  * QEMU host block devices
3  *
4  * Copyright (c) 2003-2008 Fabrice Bellard
5  *
6  * This work is licensed under the terms of the GNU GPL, version 2 or
7  * later.  See the COPYING file in the top-level directory.
8  *
9  * This file incorporates work covered by the following copyright and
10  * permission notice:
11  *
12  * Copyright (c) 2003-2008 Fabrice Bellard
13  *
14  * Permission is hereby granted, free of charge, to any person obtaining a copy
15  * of this software and associated documentation files (the "Software"), to deal
16  * in the Software without restriction, including without limitation the rights
17  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
18  * copies of the Software, and to permit persons to whom the Software is
19  * furnished to do so, subject to the following conditions:
20  *
21  * The above copyright notice and this permission notice shall be included in
22  * all copies or substantial portions of the Software.
23  *
24  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
25  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
26  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
27  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
28  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
29  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
30  * THE SOFTWARE.
31  */
32 
33 #include "qemu/osdep.h"
34 #include "sysemu/block-backend.h"
35 #include "sysemu/blockdev.h"
36 #include "hw/block/block.h"
37 #include "block/blockjob.h"
38 #include "block/qdict.h"
39 #include "block/throttle-groups.h"
40 #include "monitor/monitor.h"
41 #include "qemu/error-report.h"
42 #include "qemu/option.h"
43 #include "qemu/qemu-print.h"
44 #include "qemu/config-file.h"
45 #include "qapi/qapi-commands-block.h"
46 #include "qapi/qapi-commands-transaction.h"
47 #include "qapi/qapi-visit-block-core.h"
48 #include "qapi/qmp/qdict.h"
49 #include "qapi/qmp/qnum.h"
50 #include "qapi/qmp/qstring.h"
51 #include "qapi/error.h"
52 #include "qapi/qmp/qerror.h"
53 #include "qapi/qmp/qlist.h"
54 #include "qapi/qobject-output-visitor.h"
55 #include "sysemu/sysemu.h"
56 #include "sysemu/iothread.h"
57 #include "block/block_int.h"
58 #include "block/trace.h"
59 #include "sysemu/runstate.h"
60 #include "sysemu/replay.h"
61 #include "qemu/cutils.h"
62 #include "qemu/help_option.h"
63 #include "qemu/main-loop.h"
64 #include "qemu/throttle-options.h"
65 
66 /* Protected by BQL */
67 QTAILQ_HEAD(, BlockDriverState) monitor_bdrv_states =
68     QTAILQ_HEAD_INITIALIZER(monitor_bdrv_states);
69 
70 void bdrv_set_monitor_owned(BlockDriverState *bs)
71 {
72     GLOBAL_STATE_CODE();
73     QTAILQ_INSERT_TAIL(&monitor_bdrv_states, bs, monitor_list);
74 }
75 
76 static const char *const if_name[IF_COUNT] = {
77     [IF_NONE] = "none",
78     [IF_IDE] = "ide",
79     [IF_SCSI] = "scsi",
80     [IF_FLOPPY] = "floppy",
81     [IF_PFLASH] = "pflash",
82     [IF_MTD] = "mtd",
83     [IF_SD] = "sd",
84     [IF_VIRTIO] = "virtio",
85     [IF_XEN] = "xen",
86 };
87 
88 static int if_max_devs[IF_COUNT] = {
89     /*
90      * Do not change these numbers!  They govern how drive option
91      * index maps to unit and bus.  That mapping is ABI.
92      *
93      * All controllers used to implement if=T drives need to support
94      * if_max_devs[T] units, for any T with if_max_devs[T] != 0.
95      * Otherwise, some index values map to "impossible" bus, unit
96      * values.
97      *
98      * For instance, if you change [IF_SCSI] to 255, -drive
99      * if=scsi,index=12 no longer means bus=1,unit=5, but
100      * bus=0,unit=12.  With an lsi53c895a controller (7 units max),
101      * the drive can't be set up.  Regression.
102      */
103     [IF_IDE] = 2,
104     [IF_SCSI] = 7,
105 };
106 
107 /**
108  * Boards may call this to offer board-by-board overrides
109  * of the default, global values.
110  */
111 void override_max_devs(BlockInterfaceType type, int max_devs)
112 {
113     BlockBackend *blk;
114     DriveInfo *dinfo;
115 
116     GLOBAL_STATE_CODE();
117 
118     if (max_devs <= 0) {
119         return;
120     }
121 
122     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
123         dinfo = blk_legacy_dinfo(blk);
124         if (dinfo->type == type) {
125             fprintf(stderr, "Cannot override units-per-bus property of"
126                     " the %s interface, because a drive of that type has"
127                     " already been added.\n", if_name[type]);
128             g_assert_not_reached();
129         }
130     }
131 
132     if_max_devs[type] = max_devs;
133 }
134 
135 /*
136  * We automatically delete the drive when a device using it gets
137  * unplugged.  Questionable feature, but we can't just drop it.
138  * Device models call blockdev_mark_auto_del() to schedule the
139  * automatic deletion, and generic qdev code calls blockdev_auto_del()
140  * when deletion is actually safe.
141  */
142 void blockdev_mark_auto_del(BlockBackend *blk)
143 {
144     DriveInfo *dinfo = blk_legacy_dinfo(blk);
145     BlockJob *job;
146 
147     GLOBAL_STATE_CODE();
148 
149     if (!dinfo) {
150         return;
151     }
152 
153     for (job = block_job_next(NULL); job; job = block_job_next(job)) {
154         if (block_job_has_bdrv(job, blk_bs(blk))) {
155             AioContext *aio_context = job->job.aio_context;
156             aio_context_acquire(aio_context);
157 
158             job_cancel(&job->job, false);
159 
160             aio_context_release(aio_context);
161         }
162     }
163 
164     dinfo->auto_del = 1;
165 }
166 
167 void blockdev_auto_del(BlockBackend *blk)
168 {
169     DriveInfo *dinfo = blk_legacy_dinfo(blk);
170     GLOBAL_STATE_CODE();
171 
172     if (dinfo && dinfo->auto_del) {
173         monitor_remove_blk(blk);
174         blk_unref(blk);
175     }
176 }
177 
178 static int drive_index_to_bus_id(BlockInterfaceType type, int index)
179 {
180     int max_devs = if_max_devs[type];
181     return max_devs ? index / max_devs : 0;
182 }
183 
184 static int drive_index_to_unit_id(BlockInterfaceType type, int index)
185 {
186     int max_devs = if_max_devs[type];
187     return max_devs ? index % max_devs : index;
188 }
189 
190 QemuOpts *drive_add(BlockInterfaceType type, int index, const char *file,
191                     const char *optstr)
192 {
193     QemuOpts *opts;
194 
195     GLOBAL_STATE_CODE();
196 
197     opts = qemu_opts_parse_noisily(qemu_find_opts("drive"), optstr, false);
198     if (!opts) {
199         return NULL;
200     }
201     if (type != IF_DEFAULT) {
202         qemu_opt_set(opts, "if", if_name[type], &error_abort);
203     }
204     if (index >= 0) {
205         qemu_opt_set_number(opts, "index", index, &error_abort);
206     }
207     if (file)
208         qemu_opt_set(opts, "file", file, &error_abort);
209     return opts;
210 }
211 
212 DriveInfo *drive_get(BlockInterfaceType type, int bus, int unit)
213 {
214     BlockBackend *blk;
215     DriveInfo *dinfo;
216 
217     GLOBAL_STATE_CODE();
218 
219     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
220         dinfo = blk_legacy_dinfo(blk);
221         if (dinfo && dinfo->type == type
222             && dinfo->bus == bus && dinfo->unit == unit) {
223             return dinfo;
224         }
225     }
226 
227     return NULL;
228 }
229 
230 /*
231  * Check board claimed all -drive that are meant to be claimed.
232  * Fatal error if any remain unclaimed.
233  */
234 void drive_check_orphaned(void)
235 {
236     BlockBackend *blk;
237     DriveInfo *dinfo;
238     Location loc;
239     bool orphans = false;
240 
241     GLOBAL_STATE_CODE();
242 
243     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
244         dinfo = blk_legacy_dinfo(blk);
245         /*
246          * Ignore default drives, because we create certain default
247          * drives unconditionally, then leave them unclaimed.  Not the
248          * users fault.
249          * Ignore IF_VIRTIO, because it gets desugared into -device,
250          * so we can leave failing to -device.
251          * Ignore IF_NONE, because leaving unclaimed IF_NONE remains
252          * available for device_add is a feature.
253          */
254         if (dinfo->is_default || dinfo->type == IF_VIRTIO
255             || dinfo->type == IF_NONE) {
256             continue;
257         }
258         if (!blk_get_attached_dev(blk)) {
259             loc_push_none(&loc);
260             qemu_opts_loc_restore(dinfo->opts);
261             error_report("machine type does not support"
262                          " if=%s,bus=%d,unit=%d",
263                          if_name[dinfo->type], dinfo->bus, dinfo->unit);
264             loc_pop(&loc);
265             orphans = true;
266         }
267     }
268 
269     if (orphans) {
270         exit(1);
271     }
272 }
273 
274 DriveInfo *drive_get_by_index(BlockInterfaceType type, int index)
275 {
276     GLOBAL_STATE_CODE();
277     return drive_get(type,
278                      drive_index_to_bus_id(type, index),
279                      drive_index_to_unit_id(type, index));
280 }
281 
282 int drive_get_max_bus(BlockInterfaceType type)
283 {
284     int max_bus;
285     BlockBackend *blk;
286     DriveInfo *dinfo;
287 
288     GLOBAL_STATE_CODE();
289 
290     max_bus = -1;
291     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
292         dinfo = blk_legacy_dinfo(blk);
293         if (dinfo && dinfo->type == type && dinfo->bus > max_bus) {
294             max_bus = dinfo->bus;
295         }
296     }
297     return max_bus;
298 }
299 
300 static void bdrv_format_print(void *opaque, const char *name)
301 {
302     qemu_printf(" %s", name);
303 }
304 
305 typedef struct {
306     QEMUBH *bh;
307     BlockDriverState *bs;
308 } BDRVPutRefBH;
309 
310 static int parse_block_error_action(const char *buf, bool is_read, Error **errp)
311 {
312     if (!strcmp(buf, "ignore")) {
313         return BLOCKDEV_ON_ERROR_IGNORE;
314     } else if (!is_read && !strcmp(buf, "enospc")) {
315         return BLOCKDEV_ON_ERROR_ENOSPC;
316     } else if (!strcmp(buf, "stop")) {
317         return BLOCKDEV_ON_ERROR_STOP;
318     } else if (!strcmp(buf, "report")) {
319         return BLOCKDEV_ON_ERROR_REPORT;
320     } else {
321         error_setg(errp, "'%s' invalid %s error action",
322                    buf, is_read ? "read" : "write");
323         return -1;
324     }
325 }
326 
327 static bool parse_stats_intervals(BlockAcctStats *stats, QList *intervals,
328                                   Error **errp)
329 {
330     const QListEntry *entry;
331     for (entry = qlist_first(intervals); entry; entry = qlist_next(entry)) {
332         switch (qobject_type(entry->value)) {
333 
334         case QTYPE_QSTRING: {
335             unsigned long long length;
336             const char *str = qstring_get_str(qobject_to(QString,
337                                                          entry->value));
338             if (parse_uint_full(str, &length, 10) == 0 &&
339                 length > 0 && length <= UINT_MAX) {
340                 block_acct_add_interval(stats, (unsigned) length);
341             } else {
342                 error_setg(errp, "Invalid interval length: %s", str);
343                 return false;
344             }
345             break;
346         }
347 
348         case QTYPE_QNUM: {
349             int64_t length = qnum_get_int(qobject_to(QNum, entry->value));
350 
351             if (length > 0 && length <= UINT_MAX) {
352                 block_acct_add_interval(stats, (unsigned) length);
353             } else {
354                 error_setg(errp, "Invalid interval length: %" PRId64, length);
355                 return false;
356             }
357             break;
358         }
359 
360         default:
361             error_setg(errp, "The specification of stats-intervals is invalid");
362             return false;
363         }
364     }
365     return true;
366 }
367 
368 typedef enum { MEDIA_DISK, MEDIA_CDROM } DriveMediaType;
369 
370 /* All parameters but @opts are optional and may be set to NULL. */
371 static void extract_common_blockdev_options(QemuOpts *opts, int *bdrv_flags,
372     const char **throttling_group, ThrottleConfig *throttle_cfg,
373     BlockdevDetectZeroesOptions *detect_zeroes, Error **errp)
374 {
375     Error *local_error = NULL;
376     const char *aio;
377 
378     if (bdrv_flags) {
379         if (qemu_opt_get_bool(opts, "copy-on-read", false)) {
380             *bdrv_flags |= BDRV_O_COPY_ON_READ;
381         }
382 
383         if ((aio = qemu_opt_get(opts, "aio")) != NULL) {
384             if (bdrv_parse_aio(aio, bdrv_flags) < 0) {
385                 error_setg(errp, "invalid aio option");
386                 return;
387             }
388         }
389     }
390 
391     /* disk I/O throttling */
392     if (throttling_group) {
393         *throttling_group = qemu_opt_get(opts, "throttling.group");
394     }
395 
396     if (throttle_cfg) {
397         throttle_config_init(throttle_cfg);
398         throttle_cfg->buckets[THROTTLE_BPS_TOTAL].avg =
399             qemu_opt_get_number(opts, "throttling.bps-total", 0);
400         throttle_cfg->buckets[THROTTLE_BPS_READ].avg  =
401             qemu_opt_get_number(opts, "throttling.bps-read", 0);
402         throttle_cfg->buckets[THROTTLE_BPS_WRITE].avg =
403             qemu_opt_get_number(opts, "throttling.bps-write", 0);
404         throttle_cfg->buckets[THROTTLE_OPS_TOTAL].avg =
405             qemu_opt_get_number(opts, "throttling.iops-total", 0);
406         throttle_cfg->buckets[THROTTLE_OPS_READ].avg =
407             qemu_opt_get_number(opts, "throttling.iops-read", 0);
408         throttle_cfg->buckets[THROTTLE_OPS_WRITE].avg =
409             qemu_opt_get_number(opts, "throttling.iops-write", 0);
410 
411         throttle_cfg->buckets[THROTTLE_BPS_TOTAL].max =
412             qemu_opt_get_number(opts, "throttling.bps-total-max", 0);
413         throttle_cfg->buckets[THROTTLE_BPS_READ].max  =
414             qemu_opt_get_number(opts, "throttling.bps-read-max", 0);
415         throttle_cfg->buckets[THROTTLE_BPS_WRITE].max =
416             qemu_opt_get_number(opts, "throttling.bps-write-max", 0);
417         throttle_cfg->buckets[THROTTLE_OPS_TOTAL].max =
418             qemu_opt_get_number(opts, "throttling.iops-total-max", 0);
419         throttle_cfg->buckets[THROTTLE_OPS_READ].max =
420             qemu_opt_get_number(opts, "throttling.iops-read-max", 0);
421         throttle_cfg->buckets[THROTTLE_OPS_WRITE].max =
422             qemu_opt_get_number(opts, "throttling.iops-write-max", 0);
423 
424         throttle_cfg->buckets[THROTTLE_BPS_TOTAL].burst_length =
425             qemu_opt_get_number(opts, "throttling.bps-total-max-length", 1);
426         throttle_cfg->buckets[THROTTLE_BPS_READ].burst_length  =
427             qemu_opt_get_number(opts, "throttling.bps-read-max-length", 1);
428         throttle_cfg->buckets[THROTTLE_BPS_WRITE].burst_length =
429             qemu_opt_get_number(opts, "throttling.bps-write-max-length", 1);
430         throttle_cfg->buckets[THROTTLE_OPS_TOTAL].burst_length =
431             qemu_opt_get_number(opts, "throttling.iops-total-max-length", 1);
432         throttle_cfg->buckets[THROTTLE_OPS_READ].burst_length =
433             qemu_opt_get_number(opts, "throttling.iops-read-max-length", 1);
434         throttle_cfg->buckets[THROTTLE_OPS_WRITE].burst_length =
435             qemu_opt_get_number(opts, "throttling.iops-write-max-length", 1);
436 
437         throttle_cfg->op_size =
438             qemu_opt_get_number(opts, "throttling.iops-size", 0);
439 
440         if (!throttle_is_valid(throttle_cfg, errp)) {
441             return;
442         }
443     }
444 
445     if (detect_zeroes) {
446         *detect_zeroes =
447             qapi_enum_parse(&BlockdevDetectZeroesOptions_lookup,
448                             qemu_opt_get(opts, "detect-zeroes"),
449                             BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF,
450                             &local_error);
451         if (local_error) {
452             error_propagate(errp, local_error);
453             return;
454         }
455     }
456 }
457 
458 /* Takes the ownership of bs_opts */
459 static BlockBackend *blockdev_init(const char *file, QDict *bs_opts,
460                                    Error **errp)
461 {
462     const char *buf;
463     int bdrv_flags = 0;
464     int on_read_error, on_write_error;
465     bool account_invalid, account_failed;
466     bool writethrough, read_only;
467     BlockBackend *blk;
468     BlockDriverState *bs;
469     ThrottleConfig cfg;
470     int snapshot = 0;
471     Error *error = NULL;
472     QemuOpts *opts;
473     QDict *interval_dict = NULL;
474     QList *interval_list = NULL;
475     const char *id;
476     BlockdevDetectZeroesOptions detect_zeroes =
477         BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF;
478     const char *throttling_group = NULL;
479 
480     /* Check common options by copying from bs_opts to opts, all other options
481      * stay in bs_opts for processing by bdrv_open(). */
482     id = qdict_get_try_str(bs_opts, "id");
483     opts = qemu_opts_create(&qemu_common_drive_opts, id, 1, errp);
484     if (!opts) {
485         goto err_no_opts;
486     }
487 
488     if (!qemu_opts_absorb_qdict(opts, bs_opts, errp)) {
489         goto early_err;
490     }
491 
492     if (id) {
493         qdict_del(bs_opts, "id");
494     }
495 
496     /* extract parameters */
497     snapshot = qemu_opt_get_bool(opts, "snapshot", 0);
498 
499     account_invalid = qemu_opt_get_bool(opts, "stats-account-invalid", true);
500     account_failed = qemu_opt_get_bool(opts, "stats-account-failed", true);
501 
502     writethrough = !qemu_opt_get_bool(opts, BDRV_OPT_CACHE_WB, true);
503 
504     id = qemu_opts_id(opts);
505 
506     qdict_extract_subqdict(bs_opts, &interval_dict, "stats-intervals.");
507     qdict_array_split(interval_dict, &interval_list);
508 
509     if (qdict_size(interval_dict) != 0) {
510         error_setg(errp, "Invalid option stats-intervals.%s",
511                    qdict_first(interval_dict)->key);
512         goto early_err;
513     }
514 
515     extract_common_blockdev_options(opts, &bdrv_flags, &throttling_group, &cfg,
516                                     &detect_zeroes, &error);
517     if (error) {
518         error_propagate(errp, error);
519         goto early_err;
520     }
521 
522     if ((buf = qemu_opt_get(opts, "format")) != NULL) {
523         if (is_help_option(buf)) {
524             qemu_printf("Supported formats:");
525             bdrv_iterate_format(bdrv_format_print, NULL, false);
526             qemu_printf("\nSupported formats (read-only):");
527             bdrv_iterate_format(bdrv_format_print, NULL, true);
528             qemu_printf("\n");
529             goto early_err;
530         }
531 
532         if (qdict_haskey(bs_opts, "driver")) {
533             error_setg(errp, "Cannot specify both 'driver' and 'format'");
534             goto early_err;
535         }
536         qdict_put_str(bs_opts, "driver", buf);
537     }
538 
539     on_write_error = BLOCKDEV_ON_ERROR_ENOSPC;
540     if ((buf = qemu_opt_get(opts, "werror")) != NULL) {
541         on_write_error = parse_block_error_action(buf, 0, &error);
542         if (error) {
543             error_propagate(errp, error);
544             goto early_err;
545         }
546     }
547 
548     on_read_error = BLOCKDEV_ON_ERROR_REPORT;
549     if ((buf = qemu_opt_get(opts, "rerror")) != NULL) {
550         on_read_error = parse_block_error_action(buf, 1, &error);
551         if (error) {
552             error_propagate(errp, error);
553             goto early_err;
554         }
555     }
556 
557     if (snapshot) {
558         bdrv_flags |= BDRV_O_SNAPSHOT;
559     }
560 
561     read_only = qemu_opt_get_bool(opts, BDRV_OPT_READ_ONLY, false);
562 
563     /* init */
564     if ((!file || !*file) && !qdict_size(bs_opts)) {
565         BlockBackendRootState *blk_rs;
566 
567         blk = blk_new(qemu_get_aio_context(), 0, BLK_PERM_ALL);
568         blk_rs = blk_get_root_state(blk);
569         blk_rs->open_flags    = bdrv_flags | (read_only ? 0 : BDRV_O_RDWR);
570         blk_rs->detect_zeroes = detect_zeroes;
571 
572         qobject_unref(bs_opts);
573     } else {
574         if (file && !*file) {
575             file = NULL;
576         }
577 
578         /* bdrv_open() defaults to the values in bdrv_flags (for compatibility
579          * with other callers) rather than what we want as the real defaults.
580          * Apply the defaults here instead. */
581         qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_DIRECT, "off");
582         qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_NO_FLUSH, "off");
583         qdict_set_default_str(bs_opts, BDRV_OPT_READ_ONLY,
584                               read_only ? "on" : "off");
585         qdict_set_default_str(bs_opts, BDRV_OPT_AUTO_READ_ONLY, "on");
586         assert((bdrv_flags & BDRV_O_CACHE_MASK) == 0);
587 
588         if (runstate_check(RUN_STATE_INMIGRATE)) {
589             bdrv_flags |= BDRV_O_INACTIVE;
590         }
591 
592         blk = blk_new_open(file, NULL, bs_opts, bdrv_flags, errp);
593         if (!blk) {
594             goto err_no_bs_opts;
595         }
596         bs = blk_bs(blk);
597 
598         bs->detect_zeroes = detect_zeroes;
599 
600         block_acct_setup(blk_get_stats(blk), account_invalid, account_failed);
601 
602         if (!parse_stats_intervals(blk_get_stats(blk), interval_list, errp)) {
603             blk_unref(blk);
604             blk = NULL;
605             goto err_no_bs_opts;
606         }
607     }
608 
609     /* disk I/O throttling */
610     if (throttle_enabled(&cfg)) {
611         if (!throttling_group) {
612             throttling_group = id;
613         }
614         blk_io_limits_enable(blk, throttling_group);
615         blk_set_io_limits(blk, &cfg);
616     }
617 
618     blk_set_enable_write_cache(blk, !writethrough);
619     blk_set_on_error(blk, on_read_error, on_write_error);
620 
621     if (!monitor_add_blk(blk, id, errp)) {
622         blk_unref(blk);
623         blk = NULL;
624         goto err_no_bs_opts;
625     }
626 
627 err_no_bs_opts:
628     qemu_opts_del(opts);
629     qobject_unref(interval_dict);
630     qobject_unref(interval_list);
631     return blk;
632 
633 early_err:
634     qemu_opts_del(opts);
635     qobject_unref(interval_dict);
636     qobject_unref(interval_list);
637 err_no_opts:
638     qobject_unref(bs_opts);
639     return NULL;
640 }
641 
642 /* Takes the ownership of bs_opts */
643 BlockDriverState *bds_tree_init(QDict *bs_opts, Error **errp)
644 {
645     int bdrv_flags = 0;
646 
647     GLOBAL_STATE_CODE();
648     /* bdrv_open() defaults to the values in bdrv_flags (for compatibility
649      * with other callers) rather than what we want as the real defaults.
650      * Apply the defaults here instead. */
651     qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_DIRECT, "off");
652     qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_NO_FLUSH, "off");
653     qdict_set_default_str(bs_opts, BDRV_OPT_READ_ONLY, "off");
654 
655     if (runstate_check(RUN_STATE_INMIGRATE)) {
656         bdrv_flags |= BDRV_O_INACTIVE;
657     }
658 
659     return bdrv_open(NULL, NULL, bs_opts, bdrv_flags, errp);
660 }
661 
662 void blockdev_close_all_bdrv_states(void)
663 {
664     BlockDriverState *bs, *next_bs;
665 
666     GLOBAL_STATE_CODE();
667     QTAILQ_FOREACH_SAFE(bs, &monitor_bdrv_states, monitor_list, next_bs) {
668         AioContext *ctx = bdrv_get_aio_context(bs);
669 
670         aio_context_acquire(ctx);
671         bdrv_unref(bs);
672         aio_context_release(ctx);
673     }
674 }
675 
676 /* Iterates over the list of monitor-owned BlockDriverStates */
677 BlockDriverState *bdrv_next_monitor_owned(BlockDriverState *bs)
678 {
679     GLOBAL_STATE_CODE();
680     return bs ? QTAILQ_NEXT(bs, monitor_list)
681               : QTAILQ_FIRST(&monitor_bdrv_states);
682 }
683 
684 static bool qemu_opt_rename(QemuOpts *opts, const char *from, const char *to,
685                             Error **errp)
686 {
687     const char *value;
688 
689     value = qemu_opt_get(opts, from);
690     if (value) {
691         if (qemu_opt_find(opts, to)) {
692             error_setg(errp, "'%s' and its alias '%s' can't be used at the "
693                        "same time", to, from);
694             return false;
695         }
696     }
697 
698     /* rename all items in opts */
699     while ((value = qemu_opt_get(opts, from))) {
700         qemu_opt_set(opts, to, value, &error_abort);
701         qemu_opt_unset(opts, from);
702     }
703     return true;
704 }
705 
706 QemuOptsList qemu_legacy_drive_opts = {
707     .name = "drive",
708     .head = QTAILQ_HEAD_INITIALIZER(qemu_legacy_drive_opts.head),
709     .desc = {
710         {
711             .name = "bus",
712             .type = QEMU_OPT_NUMBER,
713             .help = "bus number",
714         },{
715             .name = "unit",
716             .type = QEMU_OPT_NUMBER,
717             .help = "unit number (i.e. lun for scsi)",
718         },{
719             .name = "index",
720             .type = QEMU_OPT_NUMBER,
721             .help = "index number",
722         },{
723             .name = "media",
724             .type = QEMU_OPT_STRING,
725             .help = "media type (disk, cdrom)",
726         },{
727             .name = "if",
728             .type = QEMU_OPT_STRING,
729             .help = "interface (ide, scsi, sd, mtd, floppy, pflash, virtio)",
730         },{
731             .name = "file",
732             .type = QEMU_OPT_STRING,
733             .help = "file name",
734         },
735 
736         /* Options that are passed on, but have special semantics with -drive */
737         {
738             .name = BDRV_OPT_READ_ONLY,
739             .type = QEMU_OPT_BOOL,
740             .help = "open drive file as read-only",
741         },{
742             .name = "rerror",
743             .type = QEMU_OPT_STRING,
744             .help = "read error action",
745         },{
746             .name = "werror",
747             .type = QEMU_OPT_STRING,
748             .help = "write error action",
749         },{
750             .name = "copy-on-read",
751             .type = QEMU_OPT_BOOL,
752             .help = "copy read data from backing file into image file",
753         },
754 
755         { /* end of list */ }
756     },
757 };
758 
759 DriveInfo *drive_new(QemuOpts *all_opts, BlockInterfaceType block_default_type,
760                      Error **errp)
761 {
762     const char *value;
763     BlockBackend *blk;
764     DriveInfo *dinfo = NULL;
765     QDict *bs_opts;
766     QemuOpts *legacy_opts;
767     DriveMediaType media = MEDIA_DISK;
768     BlockInterfaceType type;
769     int max_devs, bus_id, unit_id, index;
770     const char *werror, *rerror;
771     bool read_only = false;
772     bool copy_on_read;
773     const char *filename;
774     int i;
775 
776     GLOBAL_STATE_CODE();
777 
778     /* Change legacy command line options into QMP ones */
779     static const struct {
780         const char *from;
781         const char *to;
782     } opt_renames[] = {
783         { "iops",           "throttling.iops-total" },
784         { "iops_rd",        "throttling.iops-read" },
785         { "iops_wr",        "throttling.iops-write" },
786 
787         { "bps",            "throttling.bps-total" },
788         { "bps_rd",         "throttling.bps-read" },
789         { "bps_wr",         "throttling.bps-write" },
790 
791         { "iops_max",       "throttling.iops-total-max" },
792         { "iops_rd_max",    "throttling.iops-read-max" },
793         { "iops_wr_max",    "throttling.iops-write-max" },
794 
795         { "bps_max",        "throttling.bps-total-max" },
796         { "bps_rd_max",     "throttling.bps-read-max" },
797         { "bps_wr_max",     "throttling.bps-write-max" },
798 
799         { "iops_size",      "throttling.iops-size" },
800 
801         { "group",          "throttling.group" },
802 
803         { "readonly",       BDRV_OPT_READ_ONLY },
804     };
805 
806     for (i = 0; i < ARRAY_SIZE(opt_renames); i++) {
807         if (!qemu_opt_rename(all_opts, opt_renames[i].from,
808                              opt_renames[i].to, errp)) {
809             return NULL;
810         }
811     }
812 
813     value = qemu_opt_get(all_opts, "cache");
814     if (value) {
815         int flags = 0;
816         bool writethrough;
817 
818         if (bdrv_parse_cache_mode(value, &flags, &writethrough) != 0) {
819             error_setg(errp, "invalid cache option");
820             return NULL;
821         }
822 
823         /* Specific options take precedence */
824         if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_WB)) {
825             qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_WB,
826                               !writethrough, &error_abort);
827         }
828         if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_DIRECT)) {
829             qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_DIRECT,
830                               !!(flags & BDRV_O_NOCACHE), &error_abort);
831         }
832         if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_NO_FLUSH)) {
833             qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_NO_FLUSH,
834                               !!(flags & BDRV_O_NO_FLUSH), &error_abort);
835         }
836         qemu_opt_unset(all_opts, "cache");
837     }
838 
839     /* Get a QDict for processing the options */
840     bs_opts = qdict_new();
841     qemu_opts_to_qdict(all_opts, bs_opts);
842 
843     legacy_opts = qemu_opts_create(&qemu_legacy_drive_opts, NULL, 0,
844                                    &error_abort);
845     if (!qemu_opts_absorb_qdict(legacy_opts, bs_opts, errp)) {
846         goto fail;
847     }
848 
849     /* Media type */
850     value = qemu_opt_get(legacy_opts, "media");
851     if (value) {
852         if (!strcmp(value, "disk")) {
853             media = MEDIA_DISK;
854         } else if (!strcmp(value, "cdrom")) {
855             media = MEDIA_CDROM;
856             read_only = true;
857         } else {
858             error_setg(errp, "'%s' invalid media", value);
859             goto fail;
860         }
861     }
862 
863     /* copy-on-read is disabled with a warning for read-only devices */
864     read_only |= qemu_opt_get_bool(legacy_opts, BDRV_OPT_READ_ONLY, false);
865     copy_on_read = qemu_opt_get_bool(legacy_opts, "copy-on-read", false);
866 
867     if (read_only && copy_on_read) {
868         warn_report("disabling copy-on-read on read-only drive");
869         copy_on_read = false;
870     }
871 
872     qdict_put_str(bs_opts, BDRV_OPT_READ_ONLY, read_only ? "on" : "off");
873     qdict_put_str(bs_opts, "copy-on-read", copy_on_read ? "on" : "off");
874 
875     /* Controller type */
876     value = qemu_opt_get(legacy_opts, "if");
877     if (value) {
878         for (type = 0;
879              type < IF_COUNT && strcmp(value, if_name[type]);
880              type++) {
881         }
882         if (type == IF_COUNT) {
883             error_setg(errp, "unsupported bus type '%s'", value);
884             goto fail;
885         }
886     } else {
887         type = block_default_type;
888     }
889 
890     /* Device address specified by bus/unit or index.
891      * If none was specified, try to find the first free one. */
892     bus_id  = qemu_opt_get_number(legacy_opts, "bus", 0);
893     unit_id = qemu_opt_get_number(legacy_opts, "unit", -1);
894     index   = qemu_opt_get_number(legacy_opts, "index", -1);
895 
896     max_devs = if_max_devs[type];
897 
898     if (index != -1) {
899         if (bus_id != 0 || unit_id != -1) {
900             error_setg(errp, "index cannot be used with bus and unit");
901             goto fail;
902         }
903         bus_id = drive_index_to_bus_id(type, index);
904         unit_id = drive_index_to_unit_id(type, index);
905     }
906 
907     if (unit_id == -1) {
908        unit_id = 0;
909        while (drive_get(type, bus_id, unit_id) != NULL) {
910            unit_id++;
911            if (max_devs && unit_id >= max_devs) {
912                unit_id -= max_devs;
913                bus_id++;
914            }
915        }
916     }
917 
918     if (max_devs && unit_id >= max_devs) {
919         error_setg(errp, "unit %d too big (max is %d)", unit_id, max_devs - 1);
920         goto fail;
921     }
922 
923     if (drive_get(type, bus_id, unit_id) != NULL) {
924         error_setg(errp, "drive with bus=%d, unit=%d (index=%d) exists",
925                    bus_id, unit_id, index);
926         goto fail;
927     }
928 
929     /* no id supplied -> create one */
930     if (qemu_opts_id(all_opts) == NULL) {
931         char *new_id;
932         const char *mediastr = "";
933         if (type == IF_IDE || type == IF_SCSI) {
934             mediastr = (media == MEDIA_CDROM) ? "-cd" : "-hd";
935         }
936         if (max_devs) {
937             new_id = g_strdup_printf("%s%i%s%i", if_name[type], bus_id,
938                                      mediastr, unit_id);
939         } else {
940             new_id = g_strdup_printf("%s%s%i", if_name[type],
941                                      mediastr, unit_id);
942         }
943         qdict_put_str(bs_opts, "id", new_id);
944         g_free(new_id);
945     }
946 
947     /* Add virtio block device */
948     if (type == IF_VIRTIO) {
949         QemuOpts *devopts;
950         devopts = qemu_opts_create(qemu_find_opts("device"), NULL, 0,
951                                    &error_abort);
952         qemu_opt_set(devopts, "driver", "virtio-blk", &error_abort);
953         qemu_opt_set(devopts, "drive", qdict_get_str(bs_opts, "id"),
954                      &error_abort);
955     }
956 
957     filename = qemu_opt_get(legacy_opts, "file");
958 
959     /* Check werror/rerror compatibility with if=... */
960     werror = qemu_opt_get(legacy_opts, "werror");
961     if (werror != NULL) {
962         if (type != IF_IDE && type != IF_SCSI && type != IF_VIRTIO &&
963             type != IF_NONE) {
964             error_setg(errp, "werror is not supported by this bus type");
965             goto fail;
966         }
967         qdict_put_str(bs_opts, "werror", werror);
968     }
969 
970     rerror = qemu_opt_get(legacy_opts, "rerror");
971     if (rerror != NULL) {
972         if (type != IF_IDE && type != IF_VIRTIO && type != IF_SCSI &&
973             type != IF_NONE) {
974             error_setg(errp, "rerror is not supported by this bus type");
975             goto fail;
976         }
977         qdict_put_str(bs_opts, "rerror", rerror);
978     }
979 
980     /* Actual block device init: Functionality shared with blockdev-add */
981     blk = blockdev_init(filename, bs_opts, errp);
982     bs_opts = NULL;
983     if (!blk) {
984         goto fail;
985     }
986 
987     /* Create legacy DriveInfo */
988     dinfo = g_malloc0(sizeof(*dinfo));
989     dinfo->opts = all_opts;
990 
991     dinfo->type = type;
992     dinfo->bus = bus_id;
993     dinfo->unit = unit_id;
994 
995     blk_set_legacy_dinfo(blk, dinfo);
996 
997     switch(type) {
998     case IF_IDE:
999     case IF_SCSI:
1000     case IF_XEN:
1001     case IF_NONE:
1002         dinfo->media_cd = media == MEDIA_CDROM;
1003         break;
1004     default:
1005         break;
1006     }
1007 
1008 fail:
1009     qemu_opts_del(legacy_opts);
1010     qobject_unref(bs_opts);
1011     return dinfo;
1012 }
1013 
1014 static BlockDriverState *qmp_get_root_bs(const char *name, Error **errp)
1015 {
1016     BlockDriverState *bs;
1017 
1018     bs = bdrv_lookup_bs(name, name, errp);
1019     if (bs == NULL) {
1020         return NULL;
1021     }
1022 
1023     if (!bdrv_is_root_node(bs)) {
1024         error_setg(errp, "Need a root block node");
1025         return NULL;
1026     }
1027 
1028     if (!bdrv_is_inserted(bs)) {
1029         error_setg(errp, "Device has no medium");
1030         return NULL;
1031     }
1032 
1033     return bs;
1034 }
1035 
1036 static void blockdev_do_action(TransactionAction *action, Error **errp)
1037 {
1038     TransactionActionList list;
1039 
1040     list.value = action;
1041     list.next = NULL;
1042     qmp_transaction(&list, false, NULL, errp);
1043 }
1044 
1045 void qmp_blockdev_snapshot_sync(bool has_device, const char *device,
1046                                 bool has_node_name, const char *node_name,
1047                                 const char *snapshot_file,
1048                                 bool has_snapshot_node_name,
1049                                 const char *snapshot_node_name,
1050                                 bool has_format, const char *format,
1051                                 bool has_mode, NewImageMode mode, Error **errp)
1052 {
1053     BlockdevSnapshotSync snapshot = {
1054         .has_device = has_device,
1055         .device = (char *) device,
1056         .has_node_name = has_node_name,
1057         .node_name = (char *) node_name,
1058         .snapshot_file = (char *) snapshot_file,
1059         .has_snapshot_node_name = has_snapshot_node_name,
1060         .snapshot_node_name = (char *) snapshot_node_name,
1061         .has_format = has_format,
1062         .format = (char *) format,
1063         .has_mode = has_mode,
1064         .mode = mode,
1065     };
1066     TransactionAction action = {
1067         .type = TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC,
1068         .u.blockdev_snapshot_sync.data = &snapshot,
1069     };
1070     blockdev_do_action(&action, errp);
1071 }
1072 
1073 void qmp_blockdev_snapshot(const char *node, const char *overlay,
1074                            Error **errp)
1075 {
1076     BlockdevSnapshot snapshot_data = {
1077         .node = (char *) node,
1078         .overlay = (char *) overlay
1079     };
1080     TransactionAction action = {
1081         .type = TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT,
1082         .u.blockdev_snapshot.data = &snapshot_data,
1083     };
1084     blockdev_do_action(&action, errp);
1085 }
1086 
1087 void qmp_blockdev_snapshot_internal_sync(const char *device,
1088                                          const char *name,
1089                                          Error **errp)
1090 {
1091     BlockdevSnapshotInternal snapshot = {
1092         .device = (char *) device,
1093         .name = (char *) name
1094     };
1095     TransactionAction action = {
1096         .type = TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC,
1097         .u.blockdev_snapshot_internal_sync.data = &snapshot,
1098     };
1099     blockdev_do_action(&action, errp);
1100 }
1101 
1102 SnapshotInfo *qmp_blockdev_snapshot_delete_internal_sync(const char *device,
1103                                                          bool has_id,
1104                                                          const char *id,
1105                                                          bool has_name,
1106                                                          const char *name,
1107                                                          Error **errp)
1108 {
1109     BlockDriverState *bs;
1110     AioContext *aio_context;
1111     QEMUSnapshotInfo sn;
1112     Error *local_err = NULL;
1113     SnapshotInfo *info = NULL;
1114     int ret;
1115 
1116     bs = qmp_get_root_bs(device, errp);
1117     if (!bs) {
1118         return NULL;
1119     }
1120     aio_context = bdrv_get_aio_context(bs);
1121     aio_context_acquire(aio_context);
1122 
1123     if (!has_id) {
1124         id = NULL;
1125     }
1126 
1127     if (!has_name) {
1128         name = NULL;
1129     }
1130 
1131     if (!id && !name) {
1132         error_setg(errp, "Name or id must be provided");
1133         goto out_aio_context;
1134     }
1135 
1136     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_INTERNAL_SNAPSHOT_DELETE, errp)) {
1137         goto out_aio_context;
1138     }
1139 
1140     ret = bdrv_snapshot_find_by_id_and_name(bs, id, name, &sn, &local_err);
1141     if (local_err) {
1142         error_propagate(errp, local_err);
1143         goto out_aio_context;
1144     }
1145     if (!ret) {
1146         error_setg(errp,
1147                    "Snapshot with id '%s' and name '%s' does not exist on "
1148                    "device '%s'",
1149                    STR_OR_NULL(id), STR_OR_NULL(name), device);
1150         goto out_aio_context;
1151     }
1152 
1153     bdrv_snapshot_delete(bs, id, name, &local_err);
1154     if (local_err) {
1155         error_propagate(errp, local_err);
1156         goto out_aio_context;
1157     }
1158 
1159     aio_context_release(aio_context);
1160 
1161     info = g_new0(SnapshotInfo, 1);
1162     info->id = g_strdup(sn.id_str);
1163     info->name = g_strdup(sn.name);
1164     info->date_nsec = sn.date_nsec;
1165     info->date_sec = sn.date_sec;
1166     info->vm_state_size = sn.vm_state_size;
1167     info->vm_clock_nsec = sn.vm_clock_nsec % 1000000000;
1168     info->vm_clock_sec = sn.vm_clock_nsec / 1000000000;
1169     if (sn.icount != -1ULL) {
1170         info->icount = sn.icount;
1171         info->has_icount = true;
1172     }
1173 
1174     return info;
1175 
1176 out_aio_context:
1177     aio_context_release(aio_context);
1178     return NULL;
1179 }
1180 
1181 /* New and old BlockDriverState structs for atomic group operations */
1182 
1183 typedef struct BlkActionState BlkActionState;
1184 
1185 /**
1186  * BlkActionOps:
1187  * Table of operations that define an Action.
1188  *
1189  * @instance_size: Size of state struct, in bytes.
1190  * @prepare: Prepare the work, must NOT be NULL.
1191  * @commit: Commit the changes, can be NULL.
1192  * @abort: Abort the changes on fail, can be NULL.
1193  * @clean: Clean up resources after all transaction actions have called
1194  *         commit() or abort(). Can be NULL.
1195  *
1196  * Only prepare() may fail. In a single transaction, only one of commit() or
1197  * abort() will be called. clean() will always be called if it is present.
1198  *
1199  * Always run under BQL.
1200  */
1201 typedef struct BlkActionOps {
1202     size_t instance_size;
1203     void (*prepare)(BlkActionState *common, Error **errp);
1204     void (*commit)(BlkActionState *common);
1205     void (*abort)(BlkActionState *common);
1206     void (*clean)(BlkActionState *common);
1207 } BlkActionOps;
1208 
1209 /**
1210  * BlkActionState:
1211  * Describes one Action's state within a Transaction.
1212  *
1213  * @action: QAPI-defined enum identifying which Action to perform.
1214  * @ops: Table of ActionOps this Action can perform.
1215  * @block_job_txn: Transaction which this action belongs to.
1216  * @entry: List membership for all Actions in this Transaction.
1217  *
1218  * This structure must be arranged as first member in a subclassed type,
1219  * assuming that the compiler will also arrange it to the same offsets as the
1220  * base class.
1221  */
1222 struct BlkActionState {
1223     TransactionAction *action;
1224     const BlkActionOps *ops;
1225     JobTxn *block_job_txn;
1226     TransactionProperties *txn_props;
1227     QTAILQ_ENTRY(BlkActionState) entry;
1228 };
1229 
1230 /* internal snapshot private data */
1231 typedef struct InternalSnapshotState {
1232     BlkActionState common;
1233     BlockDriverState *bs;
1234     QEMUSnapshotInfo sn;
1235     bool created;
1236 } InternalSnapshotState;
1237 
1238 
1239 static int action_check_completion_mode(BlkActionState *s, Error **errp)
1240 {
1241     if (s->txn_props->completion_mode != ACTION_COMPLETION_MODE_INDIVIDUAL) {
1242         error_setg(errp,
1243                    "Action '%s' does not support Transaction property "
1244                    "completion-mode = %s",
1245                    TransactionActionKind_str(s->action->type),
1246                    ActionCompletionMode_str(s->txn_props->completion_mode));
1247         return -1;
1248     }
1249     return 0;
1250 }
1251 
1252 static void internal_snapshot_prepare(BlkActionState *common,
1253                                       Error **errp)
1254 {
1255     Error *local_err = NULL;
1256     const char *device;
1257     const char *name;
1258     BlockDriverState *bs;
1259     QEMUSnapshotInfo old_sn, *sn;
1260     bool ret;
1261     int64_t rt;
1262     BlockdevSnapshotInternal *internal;
1263     InternalSnapshotState *state;
1264     AioContext *aio_context;
1265     int ret1;
1266 
1267     g_assert(common->action->type ==
1268              TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC);
1269     internal = common->action->u.blockdev_snapshot_internal_sync.data;
1270     state = DO_UPCAST(InternalSnapshotState, common, common);
1271 
1272     /* 1. parse input */
1273     device = internal->device;
1274     name = internal->name;
1275 
1276     /* 2. check for validation */
1277     if (action_check_completion_mode(common, errp) < 0) {
1278         return;
1279     }
1280 
1281     bs = qmp_get_root_bs(device, errp);
1282     if (!bs) {
1283         return;
1284     }
1285 
1286     aio_context = bdrv_get_aio_context(bs);
1287     aio_context_acquire(aio_context);
1288 
1289     state->bs = bs;
1290 
1291     /* Paired with .clean() */
1292     bdrv_drained_begin(bs);
1293 
1294     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_INTERNAL_SNAPSHOT, errp)) {
1295         goto out;
1296     }
1297 
1298     if (bdrv_is_read_only(bs)) {
1299         error_setg(errp, "Device '%s' is read only", device);
1300         goto out;
1301     }
1302 
1303     if (!bdrv_can_snapshot(bs)) {
1304         error_setg(errp, "Block format '%s' used by device '%s' "
1305                    "does not support internal snapshots",
1306                    bs->drv->format_name, device);
1307         goto out;
1308     }
1309 
1310     if (!strlen(name)) {
1311         error_setg(errp, "Name is empty");
1312         goto out;
1313     }
1314 
1315     /* check whether a snapshot with name exist */
1316     ret = bdrv_snapshot_find_by_id_and_name(bs, NULL, name, &old_sn,
1317                                             &local_err);
1318     if (local_err) {
1319         error_propagate(errp, local_err);
1320         goto out;
1321     } else if (ret) {
1322         error_setg(errp,
1323                    "Snapshot with name '%s' already exists on device '%s'",
1324                    name, device);
1325         goto out;
1326     }
1327 
1328     /* 3. take the snapshot */
1329     sn = &state->sn;
1330     pstrcpy(sn->name, sizeof(sn->name), name);
1331     rt = g_get_real_time();
1332     sn->date_sec = rt / G_USEC_PER_SEC;
1333     sn->date_nsec = (rt % G_USEC_PER_SEC) * 1000;
1334     sn->vm_clock_nsec = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL);
1335     if (replay_mode != REPLAY_MODE_NONE) {
1336         sn->icount = replay_get_current_icount();
1337     } else {
1338         sn->icount = -1ULL;
1339     }
1340 
1341     ret1 = bdrv_snapshot_create(bs, sn);
1342     if (ret1 < 0) {
1343         error_setg_errno(errp, -ret1,
1344                          "Failed to create snapshot '%s' on device '%s'",
1345                          name, device);
1346         goto out;
1347     }
1348 
1349     /* 4. succeed, mark a snapshot is created */
1350     state->created = true;
1351 
1352 out:
1353     aio_context_release(aio_context);
1354 }
1355 
1356 static void internal_snapshot_abort(BlkActionState *common)
1357 {
1358     InternalSnapshotState *state =
1359                              DO_UPCAST(InternalSnapshotState, common, common);
1360     BlockDriverState *bs = state->bs;
1361     QEMUSnapshotInfo *sn = &state->sn;
1362     AioContext *aio_context;
1363     Error *local_error = NULL;
1364 
1365     if (!state->created) {
1366         return;
1367     }
1368 
1369     aio_context = bdrv_get_aio_context(state->bs);
1370     aio_context_acquire(aio_context);
1371 
1372     if (bdrv_snapshot_delete(bs, sn->id_str, sn->name, &local_error) < 0) {
1373         error_reportf_err(local_error,
1374                           "Failed to delete snapshot with id '%s' and "
1375                           "name '%s' on device '%s' in abort: ",
1376                           sn->id_str, sn->name,
1377                           bdrv_get_device_name(bs));
1378     }
1379 
1380     aio_context_release(aio_context);
1381 }
1382 
1383 static void internal_snapshot_clean(BlkActionState *common)
1384 {
1385     InternalSnapshotState *state = DO_UPCAST(InternalSnapshotState,
1386                                              common, common);
1387     AioContext *aio_context;
1388 
1389     if (!state->bs) {
1390         return;
1391     }
1392 
1393     aio_context = bdrv_get_aio_context(state->bs);
1394     aio_context_acquire(aio_context);
1395 
1396     bdrv_drained_end(state->bs);
1397 
1398     aio_context_release(aio_context);
1399 }
1400 
1401 /* external snapshot private data */
1402 typedef struct ExternalSnapshotState {
1403     BlkActionState common;
1404     BlockDriverState *old_bs;
1405     BlockDriverState *new_bs;
1406     bool overlay_appended;
1407 } ExternalSnapshotState;
1408 
1409 static void external_snapshot_prepare(BlkActionState *common,
1410                                       Error **errp)
1411 {
1412     int ret;
1413     int flags = 0;
1414     QDict *options = NULL;
1415     Error *local_err = NULL;
1416     /* Device and node name of the image to generate the snapshot from */
1417     const char *device;
1418     const char *node_name;
1419     /* Reference to the new image (for 'blockdev-snapshot') */
1420     const char *snapshot_ref;
1421     /* File name of the new image (for 'blockdev-snapshot-sync') */
1422     const char *new_image_file;
1423     ExternalSnapshotState *state =
1424                              DO_UPCAST(ExternalSnapshotState, common, common);
1425     TransactionAction *action = common->action;
1426     AioContext *aio_context;
1427     uint64_t perm, shared;
1428 
1429     /* 'blockdev-snapshot' and 'blockdev-snapshot-sync' have similar
1430      * purpose but a different set of parameters */
1431     switch (action->type) {
1432     case TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT:
1433         {
1434             BlockdevSnapshot *s = action->u.blockdev_snapshot.data;
1435             device = s->node;
1436             node_name = s->node;
1437             new_image_file = NULL;
1438             snapshot_ref = s->overlay;
1439         }
1440         break;
1441     case TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC:
1442         {
1443             BlockdevSnapshotSync *s = action->u.blockdev_snapshot_sync.data;
1444             device = s->has_device ? s->device : NULL;
1445             node_name = s->has_node_name ? s->node_name : NULL;
1446             new_image_file = s->snapshot_file;
1447             snapshot_ref = NULL;
1448         }
1449         break;
1450     default:
1451         g_assert_not_reached();
1452     }
1453 
1454     /* start processing */
1455     if (action_check_completion_mode(common, errp) < 0) {
1456         return;
1457     }
1458 
1459     state->old_bs = bdrv_lookup_bs(device, node_name, errp);
1460     if (!state->old_bs) {
1461         return;
1462     }
1463 
1464     aio_context = bdrv_get_aio_context(state->old_bs);
1465     aio_context_acquire(aio_context);
1466 
1467     /* Paired with .clean() */
1468     bdrv_drained_begin(state->old_bs);
1469 
1470     if (!bdrv_is_inserted(state->old_bs)) {
1471         error_setg(errp, QERR_DEVICE_HAS_NO_MEDIUM, device);
1472         goto out;
1473     }
1474 
1475     if (bdrv_op_is_blocked(state->old_bs,
1476                            BLOCK_OP_TYPE_EXTERNAL_SNAPSHOT, errp)) {
1477         goto out;
1478     }
1479 
1480     if (!bdrv_is_read_only(state->old_bs)) {
1481         if (bdrv_flush(state->old_bs)) {
1482             error_setg(errp, QERR_IO_ERROR);
1483             goto out;
1484         }
1485     }
1486 
1487     if (action->type == TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC) {
1488         BlockdevSnapshotSync *s = action->u.blockdev_snapshot_sync.data;
1489         const char *format = s->has_format ? s->format : "qcow2";
1490         enum NewImageMode mode;
1491         const char *snapshot_node_name =
1492             s->has_snapshot_node_name ? s->snapshot_node_name : NULL;
1493 
1494         if (node_name && !snapshot_node_name) {
1495             error_setg(errp, "New overlay node-name missing");
1496             goto out;
1497         }
1498 
1499         if (snapshot_node_name &&
1500             bdrv_lookup_bs(snapshot_node_name, snapshot_node_name, NULL)) {
1501             error_setg(errp, "New overlay node-name already in use");
1502             goto out;
1503         }
1504 
1505         flags = state->old_bs->open_flags;
1506         flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_COPY_ON_READ);
1507         flags |= BDRV_O_NO_BACKING;
1508 
1509         /* create new image w/backing file */
1510         mode = s->has_mode ? s->mode : NEW_IMAGE_MODE_ABSOLUTE_PATHS;
1511         if (mode != NEW_IMAGE_MODE_EXISTING) {
1512             int64_t size = bdrv_getlength(state->old_bs);
1513             if (size < 0) {
1514                 error_setg_errno(errp, -size, "bdrv_getlength failed");
1515                 goto out;
1516             }
1517             bdrv_refresh_filename(state->old_bs);
1518             bdrv_img_create(new_image_file, format,
1519                             state->old_bs->filename,
1520                             state->old_bs->drv->format_name,
1521                             NULL, size, flags, false, &local_err);
1522             if (local_err) {
1523                 error_propagate(errp, local_err);
1524                 goto out;
1525             }
1526         }
1527 
1528         options = qdict_new();
1529         if (snapshot_node_name) {
1530             qdict_put_str(options, "node-name", snapshot_node_name);
1531         }
1532         qdict_put_str(options, "driver", format);
1533     }
1534 
1535     state->new_bs = bdrv_open(new_image_file, snapshot_ref, options, flags,
1536                               errp);
1537     /* We will manually add the backing_hd field to the bs later */
1538     if (!state->new_bs) {
1539         goto out;
1540     }
1541 
1542     /*
1543      * Allow attaching a backing file to an overlay that's already in use only
1544      * if the parents don't assume that they are already seeing a valid image.
1545      * (Specifically, allow it as a mirror target, which is write-only access.)
1546      */
1547     bdrv_get_cumulative_perm(state->new_bs, &perm, &shared);
1548     if (perm & BLK_PERM_CONSISTENT_READ) {
1549         error_setg(errp, "The overlay is already in use");
1550         goto out;
1551     }
1552 
1553     if (state->new_bs->drv->is_filter) {
1554         error_setg(errp, "Filters cannot be used as overlays");
1555         goto out;
1556     }
1557 
1558     if (bdrv_cow_child(state->new_bs)) {
1559         error_setg(errp, "The overlay already has a backing image");
1560         goto out;
1561     }
1562 
1563     if (!state->new_bs->drv->supports_backing) {
1564         error_setg(errp, "The overlay does not support backing images");
1565         goto out;
1566     }
1567 
1568     ret = bdrv_append(state->new_bs, state->old_bs, errp);
1569     if (ret < 0) {
1570         goto out;
1571     }
1572     state->overlay_appended = true;
1573 
1574 out:
1575     aio_context_release(aio_context);
1576 }
1577 
1578 static void external_snapshot_commit(BlkActionState *common)
1579 {
1580     ExternalSnapshotState *state =
1581                              DO_UPCAST(ExternalSnapshotState, common, common);
1582     AioContext *aio_context;
1583 
1584     aio_context = bdrv_get_aio_context(state->old_bs);
1585     aio_context_acquire(aio_context);
1586 
1587     /* We don't need (or want) to use the transactional
1588      * bdrv_reopen_multiple() across all the entries at once, because we
1589      * don't want to abort all of them if one of them fails the reopen */
1590     if (!qatomic_read(&state->old_bs->copy_on_read)) {
1591         bdrv_reopen_set_read_only(state->old_bs, true, NULL);
1592     }
1593 
1594     aio_context_release(aio_context);
1595 }
1596 
1597 static void external_snapshot_abort(BlkActionState *common)
1598 {
1599     ExternalSnapshotState *state =
1600                              DO_UPCAST(ExternalSnapshotState, common, common);
1601     if (state->new_bs) {
1602         if (state->overlay_appended) {
1603             AioContext *aio_context;
1604             AioContext *tmp_context;
1605             int ret;
1606 
1607             aio_context = bdrv_get_aio_context(state->old_bs);
1608             aio_context_acquire(aio_context);
1609 
1610             bdrv_ref(state->old_bs);   /* we can't let bdrv_set_backind_hd()
1611                                           close state->old_bs; we need it */
1612             bdrv_set_backing_hd(state->new_bs, NULL, &error_abort);
1613 
1614             /*
1615              * The call to bdrv_set_backing_hd() above returns state->old_bs to
1616              * the main AioContext. As we're still going to be using it, return
1617              * it to the AioContext it was before.
1618              */
1619             tmp_context = bdrv_get_aio_context(state->old_bs);
1620             if (aio_context != tmp_context) {
1621                 aio_context_release(aio_context);
1622                 aio_context_acquire(tmp_context);
1623 
1624                 ret = bdrv_try_set_aio_context(state->old_bs,
1625                                                aio_context, NULL);
1626                 assert(ret == 0);
1627 
1628                 aio_context_release(tmp_context);
1629                 aio_context_acquire(aio_context);
1630             }
1631 
1632             bdrv_replace_node(state->new_bs, state->old_bs, &error_abort);
1633             bdrv_unref(state->old_bs); /* bdrv_replace_node() ref'ed old_bs */
1634 
1635             aio_context_release(aio_context);
1636         }
1637     }
1638 }
1639 
1640 static void external_snapshot_clean(BlkActionState *common)
1641 {
1642     ExternalSnapshotState *state =
1643                              DO_UPCAST(ExternalSnapshotState, common, common);
1644     AioContext *aio_context;
1645 
1646     if (!state->old_bs) {
1647         return;
1648     }
1649 
1650     aio_context = bdrv_get_aio_context(state->old_bs);
1651     aio_context_acquire(aio_context);
1652 
1653     bdrv_drained_end(state->old_bs);
1654     bdrv_unref(state->new_bs);
1655 
1656     aio_context_release(aio_context);
1657 }
1658 
1659 typedef struct DriveBackupState {
1660     BlkActionState common;
1661     BlockDriverState *bs;
1662     BlockJob *job;
1663 } DriveBackupState;
1664 
1665 static BlockJob *do_backup_common(BackupCommon *backup,
1666                                   BlockDriverState *bs,
1667                                   BlockDriverState *target_bs,
1668                                   AioContext *aio_context,
1669                                   JobTxn *txn, Error **errp);
1670 
1671 static void drive_backup_prepare(BlkActionState *common, Error **errp)
1672 {
1673     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1674     DriveBackup *backup;
1675     BlockDriverState *bs;
1676     BlockDriverState *target_bs;
1677     BlockDriverState *source = NULL;
1678     AioContext *aio_context;
1679     AioContext *old_context;
1680     QDict *options;
1681     Error *local_err = NULL;
1682     int flags;
1683     int64_t size;
1684     bool set_backing_hd = false;
1685     int ret;
1686 
1687     assert(common->action->type == TRANSACTION_ACTION_KIND_DRIVE_BACKUP);
1688     backup = common->action->u.drive_backup.data;
1689 
1690     if (!backup->has_mode) {
1691         backup->mode = NEW_IMAGE_MODE_ABSOLUTE_PATHS;
1692     }
1693 
1694     bs = bdrv_lookup_bs(backup->device, backup->device, errp);
1695     if (!bs) {
1696         return;
1697     }
1698 
1699     if (!bs->drv) {
1700         error_setg(errp, "Device has no medium");
1701         return;
1702     }
1703 
1704     aio_context = bdrv_get_aio_context(bs);
1705     aio_context_acquire(aio_context);
1706 
1707     state->bs = bs;
1708     /* Paired with .clean() */
1709     bdrv_drained_begin(bs);
1710 
1711     if (!backup->has_format) {
1712         backup->format = backup->mode == NEW_IMAGE_MODE_EXISTING ?
1713                          NULL : (char *) bs->drv->format_name;
1714     }
1715 
1716     /* Early check to avoid creating target */
1717     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_BACKUP_SOURCE, errp)) {
1718         goto out;
1719     }
1720 
1721     flags = bs->open_flags | BDRV_O_RDWR;
1722 
1723     /*
1724      * See if we have a backing HD we can use to create our new image
1725      * on top of.
1726      */
1727     if (backup->sync == MIRROR_SYNC_MODE_TOP) {
1728         /*
1729          * Backup will not replace the source by the target, so none
1730          * of the filters skipped here will be removed (in contrast to
1731          * mirror).  Therefore, we can skip all of them when looking
1732          * for the first COW relationship.
1733          */
1734         source = bdrv_cow_bs(bdrv_skip_filters(bs));
1735         if (!source) {
1736             backup->sync = MIRROR_SYNC_MODE_FULL;
1737         }
1738     }
1739     if (backup->sync == MIRROR_SYNC_MODE_NONE) {
1740         source = bs;
1741         flags |= BDRV_O_NO_BACKING;
1742         set_backing_hd = true;
1743     }
1744 
1745     size = bdrv_getlength(bs);
1746     if (size < 0) {
1747         error_setg_errno(errp, -size, "bdrv_getlength failed");
1748         goto out;
1749     }
1750 
1751     if (backup->mode != NEW_IMAGE_MODE_EXISTING) {
1752         assert(backup->format);
1753         if (source) {
1754             /* Implicit filters should not appear in the filename */
1755             BlockDriverState *explicit_backing =
1756                 bdrv_skip_implicit_filters(source);
1757 
1758             bdrv_refresh_filename(explicit_backing);
1759             bdrv_img_create(backup->target, backup->format,
1760                             explicit_backing->filename,
1761                             explicit_backing->drv->format_name, NULL,
1762                             size, flags, false, &local_err);
1763         } else {
1764             bdrv_img_create(backup->target, backup->format, NULL, NULL, NULL,
1765                             size, flags, false, &local_err);
1766         }
1767     }
1768 
1769     if (local_err) {
1770         error_propagate(errp, local_err);
1771         goto out;
1772     }
1773 
1774     options = qdict_new();
1775     qdict_put_str(options, "discard", "unmap");
1776     qdict_put_str(options, "detect-zeroes", "unmap");
1777     if (backup->format) {
1778         qdict_put_str(options, "driver", backup->format);
1779     }
1780 
1781     target_bs = bdrv_open(backup->target, NULL, options, flags, errp);
1782     if (!target_bs) {
1783         goto out;
1784     }
1785 
1786     /* Honor bdrv_try_set_aio_context() context acquisition requirements. */
1787     old_context = bdrv_get_aio_context(target_bs);
1788     aio_context_release(aio_context);
1789     aio_context_acquire(old_context);
1790 
1791     ret = bdrv_try_set_aio_context(target_bs, aio_context, errp);
1792     if (ret < 0) {
1793         bdrv_unref(target_bs);
1794         aio_context_release(old_context);
1795         return;
1796     }
1797 
1798     aio_context_release(old_context);
1799     aio_context_acquire(aio_context);
1800 
1801     if (set_backing_hd) {
1802         if (bdrv_set_backing_hd(target_bs, source, errp) < 0) {
1803             goto unref;
1804         }
1805     }
1806 
1807     state->job = do_backup_common(qapi_DriveBackup_base(backup),
1808                                   bs, target_bs, aio_context,
1809                                   common->block_job_txn, errp);
1810 
1811 unref:
1812     bdrv_unref(target_bs);
1813 out:
1814     aio_context_release(aio_context);
1815 }
1816 
1817 static void drive_backup_commit(BlkActionState *common)
1818 {
1819     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1820     AioContext *aio_context;
1821 
1822     aio_context = bdrv_get_aio_context(state->bs);
1823     aio_context_acquire(aio_context);
1824 
1825     assert(state->job);
1826     job_start(&state->job->job);
1827 
1828     aio_context_release(aio_context);
1829 }
1830 
1831 static void drive_backup_abort(BlkActionState *common)
1832 {
1833     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1834 
1835     if (state->job) {
1836         AioContext *aio_context;
1837 
1838         aio_context = bdrv_get_aio_context(state->bs);
1839         aio_context_acquire(aio_context);
1840 
1841         job_cancel_sync(&state->job->job, true);
1842 
1843         aio_context_release(aio_context);
1844     }
1845 }
1846 
1847 static void drive_backup_clean(BlkActionState *common)
1848 {
1849     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1850     AioContext *aio_context;
1851 
1852     if (!state->bs) {
1853         return;
1854     }
1855 
1856     aio_context = bdrv_get_aio_context(state->bs);
1857     aio_context_acquire(aio_context);
1858 
1859     bdrv_drained_end(state->bs);
1860 
1861     aio_context_release(aio_context);
1862 }
1863 
1864 typedef struct BlockdevBackupState {
1865     BlkActionState common;
1866     BlockDriverState *bs;
1867     BlockJob *job;
1868 } BlockdevBackupState;
1869 
1870 static void blockdev_backup_prepare(BlkActionState *common, Error **errp)
1871 {
1872     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1873     BlockdevBackup *backup;
1874     BlockDriverState *bs;
1875     BlockDriverState *target_bs;
1876     AioContext *aio_context;
1877     AioContext *old_context;
1878     int ret;
1879 
1880     assert(common->action->type == TRANSACTION_ACTION_KIND_BLOCKDEV_BACKUP);
1881     backup = common->action->u.blockdev_backup.data;
1882 
1883     bs = bdrv_lookup_bs(backup->device, backup->device, errp);
1884     if (!bs) {
1885         return;
1886     }
1887 
1888     target_bs = bdrv_lookup_bs(backup->target, backup->target, errp);
1889     if (!target_bs) {
1890         return;
1891     }
1892 
1893     /* Honor bdrv_try_set_aio_context() context acquisition requirements. */
1894     aio_context = bdrv_get_aio_context(bs);
1895     old_context = bdrv_get_aio_context(target_bs);
1896     aio_context_acquire(old_context);
1897 
1898     ret = bdrv_try_set_aio_context(target_bs, aio_context, errp);
1899     if (ret < 0) {
1900         aio_context_release(old_context);
1901         return;
1902     }
1903 
1904     aio_context_release(old_context);
1905     aio_context_acquire(aio_context);
1906     state->bs = bs;
1907 
1908     /* Paired with .clean() */
1909     bdrv_drained_begin(state->bs);
1910 
1911     state->job = do_backup_common(qapi_BlockdevBackup_base(backup),
1912                                   bs, target_bs, aio_context,
1913                                   common->block_job_txn, errp);
1914 
1915     aio_context_release(aio_context);
1916 }
1917 
1918 static void blockdev_backup_commit(BlkActionState *common)
1919 {
1920     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1921     AioContext *aio_context;
1922 
1923     aio_context = bdrv_get_aio_context(state->bs);
1924     aio_context_acquire(aio_context);
1925 
1926     assert(state->job);
1927     job_start(&state->job->job);
1928 
1929     aio_context_release(aio_context);
1930 }
1931 
1932 static void blockdev_backup_abort(BlkActionState *common)
1933 {
1934     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1935 
1936     if (state->job) {
1937         AioContext *aio_context;
1938 
1939         aio_context = bdrv_get_aio_context(state->bs);
1940         aio_context_acquire(aio_context);
1941 
1942         job_cancel_sync(&state->job->job, true);
1943 
1944         aio_context_release(aio_context);
1945     }
1946 }
1947 
1948 static void blockdev_backup_clean(BlkActionState *common)
1949 {
1950     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1951     AioContext *aio_context;
1952 
1953     if (!state->bs) {
1954         return;
1955     }
1956 
1957     aio_context = bdrv_get_aio_context(state->bs);
1958     aio_context_acquire(aio_context);
1959 
1960     bdrv_drained_end(state->bs);
1961 
1962     aio_context_release(aio_context);
1963 }
1964 
1965 typedef struct BlockDirtyBitmapState {
1966     BlkActionState common;
1967     BdrvDirtyBitmap *bitmap;
1968     BlockDriverState *bs;
1969     HBitmap *backup;
1970     bool prepared;
1971     bool was_enabled;
1972 } BlockDirtyBitmapState;
1973 
1974 static void block_dirty_bitmap_add_prepare(BlkActionState *common,
1975                                            Error **errp)
1976 {
1977     Error *local_err = NULL;
1978     BlockDirtyBitmapAdd *action;
1979     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
1980                                              common, common);
1981 
1982     if (action_check_completion_mode(common, errp) < 0) {
1983         return;
1984     }
1985 
1986     action = common->action->u.block_dirty_bitmap_add.data;
1987     /* AIO context taken and released within qmp_block_dirty_bitmap_add */
1988     qmp_block_dirty_bitmap_add(action->node, action->name,
1989                                action->has_granularity, action->granularity,
1990                                action->has_persistent, action->persistent,
1991                                action->has_disabled, action->disabled,
1992                                &local_err);
1993 
1994     if (!local_err) {
1995         state->prepared = true;
1996     } else {
1997         error_propagate(errp, local_err);
1998     }
1999 }
2000 
2001 static void block_dirty_bitmap_add_abort(BlkActionState *common)
2002 {
2003     BlockDirtyBitmapAdd *action;
2004     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2005                                              common, common);
2006 
2007     action = common->action->u.block_dirty_bitmap_add.data;
2008     /* Should not be able to fail: IF the bitmap was added via .prepare(),
2009      * then the node reference and bitmap name must have been valid.
2010      */
2011     if (state->prepared) {
2012         qmp_block_dirty_bitmap_remove(action->node, action->name, &error_abort);
2013     }
2014 }
2015 
2016 static void block_dirty_bitmap_clear_prepare(BlkActionState *common,
2017                                              Error **errp)
2018 {
2019     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2020                                              common, common);
2021     BlockDirtyBitmap *action;
2022 
2023     if (action_check_completion_mode(common, errp) < 0) {
2024         return;
2025     }
2026 
2027     action = common->action->u.block_dirty_bitmap_clear.data;
2028     state->bitmap = block_dirty_bitmap_lookup(action->node,
2029                                               action->name,
2030                                               &state->bs,
2031                                               errp);
2032     if (!state->bitmap) {
2033         return;
2034     }
2035 
2036     if (bdrv_dirty_bitmap_check(state->bitmap, BDRV_BITMAP_DEFAULT, errp)) {
2037         return;
2038     }
2039 
2040     bdrv_clear_dirty_bitmap(state->bitmap, &state->backup);
2041 }
2042 
2043 static void block_dirty_bitmap_restore(BlkActionState *common)
2044 {
2045     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2046                                              common, common);
2047 
2048     if (state->backup) {
2049         bdrv_restore_dirty_bitmap(state->bitmap, state->backup);
2050     }
2051 }
2052 
2053 static void block_dirty_bitmap_free_backup(BlkActionState *common)
2054 {
2055     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2056                                              common, common);
2057 
2058     hbitmap_free(state->backup);
2059 }
2060 
2061 static void block_dirty_bitmap_enable_prepare(BlkActionState *common,
2062                                               Error **errp)
2063 {
2064     BlockDirtyBitmap *action;
2065     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2066                                              common, common);
2067 
2068     if (action_check_completion_mode(common, errp) < 0) {
2069         return;
2070     }
2071 
2072     action = common->action->u.block_dirty_bitmap_enable.data;
2073     state->bitmap = block_dirty_bitmap_lookup(action->node,
2074                                               action->name,
2075                                               NULL,
2076                                               errp);
2077     if (!state->bitmap) {
2078         return;
2079     }
2080 
2081     if (bdrv_dirty_bitmap_check(state->bitmap, BDRV_BITMAP_ALLOW_RO, errp)) {
2082         return;
2083     }
2084 
2085     state->was_enabled = bdrv_dirty_bitmap_enabled(state->bitmap);
2086     bdrv_enable_dirty_bitmap(state->bitmap);
2087 }
2088 
2089 static void block_dirty_bitmap_enable_abort(BlkActionState *common)
2090 {
2091     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2092                                              common, common);
2093 
2094     if (!state->was_enabled) {
2095         bdrv_disable_dirty_bitmap(state->bitmap);
2096     }
2097 }
2098 
2099 static void block_dirty_bitmap_disable_prepare(BlkActionState *common,
2100                                                Error **errp)
2101 {
2102     BlockDirtyBitmap *action;
2103     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2104                                              common, common);
2105 
2106     if (action_check_completion_mode(common, errp) < 0) {
2107         return;
2108     }
2109 
2110     action = common->action->u.block_dirty_bitmap_disable.data;
2111     state->bitmap = block_dirty_bitmap_lookup(action->node,
2112                                               action->name,
2113                                               NULL,
2114                                               errp);
2115     if (!state->bitmap) {
2116         return;
2117     }
2118 
2119     if (bdrv_dirty_bitmap_check(state->bitmap, BDRV_BITMAP_ALLOW_RO, errp)) {
2120         return;
2121     }
2122 
2123     state->was_enabled = bdrv_dirty_bitmap_enabled(state->bitmap);
2124     bdrv_disable_dirty_bitmap(state->bitmap);
2125 }
2126 
2127 static void block_dirty_bitmap_disable_abort(BlkActionState *common)
2128 {
2129     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2130                                              common, common);
2131 
2132     if (state->was_enabled) {
2133         bdrv_enable_dirty_bitmap(state->bitmap);
2134     }
2135 }
2136 
2137 static void block_dirty_bitmap_merge_prepare(BlkActionState *common,
2138                                              Error **errp)
2139 {
2140     BlockDirtyBitmapMerge *action;
2141     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2142                                              common, common);
2143 
2144     if (action_check_completion_mode(common, errp) < 0) {
2145         return;
2146     }
2147 
2148     action = common->action->u.block_dirty_bitmap_merge.data;
2149 
2150     state->bitmap = block_dirty_bitmap_merge(action->node, action->target,
2151                                              action->bitmaps, &state->backup,
2152                                              errp);
2153 }
2154 
2155 static void block_dirty_bitmap_remove_prepare(BlkActionState *common,
2156                                               Error **errp)
2157 {
2158     BlockDirtyBitmap *action;
2159     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2160                                              common, common);
2161 
2162     if (action_check_completion_mode(common, errp) < 0) {
2163         return;
2164     }
2165 
2166     action = common->action->u.block_dirty_bitmap_remove.data;
2167 
2168     state->bitmap = block_dirty_bitmap_remove(action->node, action->name,
2169                                               false, &state->bs, errp);
2170     if (state->bitmap) {
2171         bdrv_dirty_bitmap_skip_store(state->bitmap, true);
2172         bdrv_dirty_bitmap_set_busy(state->bitmap, true);
2173     }
2174 }
2175 
2176 static void block_dirty_bitmap_remove_abort(BlkActionState *common)
2177 {
2178     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2179                                              common, common);
2180 
2181     if (state->bitmap) {
2182         bdrv_dirty_bitmap_skip_store(state->bitmap, false);
2183         bdrv_dirty_bitmap_set_busy(state->bitmap, false);
2184     }
2185 }
2186 
2187 static void block_dirty_bitmap_remove_commit(BlkActionState *common)
2188 {
2189     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2190                                              common, common);
2191 
2192     bdrv_dirty_bitmap_set_busy(state->bitmap, false);
2193     bdrv_release_dirty_bitmap(state->bitmap);
2194 }
2195 
2196 static void abort_prepare(BlkActionState *common, Error **errp)
2197 {
2198     error_setg(errp, "Transaction aborted using Abort action");
2199 }
2200 
2201 static void abort_commit(BlkActionState *common)
2202 {
2203     g_assert_not_reached(); /* this action never succeeds */
2204 }
2205 
2206 static const BlkActionOps actions[] = {
2207     [TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT] = {
2208         .instance_size = sizeof(ExternalSnapshotState),
2209         .prepare  = external_snapshot_prepare,
2210         .commit   = external_snapshot_commit,
2211         .abort = external_snapshot_abort,
2212         .clean = external_snapshot_clean,
2213     },
2214     [TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC] = {
2215         .instance_size = sizeof(ExternalSnapshotState),
2216         .prepare  = external_snapshot_prepare,
2217         .commit   = external_snapshot_commit,
2218         .abort = external_snapshot_abort,
2219         .clean = external_snapshot_clean,
2220     },
2221     [TRANSACTION_ACTION_KIND_DRIVE_BACKUP] = {
2222         .instance_size = sizeof(DriveBackupState),
2223         .prepare = drive_backup_prepare,
2224         .commit = drive_backup_commit,
2225         .abort = drive_backup_abort,
2226         .clean = drive_backup_clean,
2227     },
2228     [TRANSACTION_ACTION_KIND_BLOCKDEV_BACKUP] = {
2229         .instance_size = sizeof(BlockdevBackupState),
2230         .prepare = blockdev_backup_prepare,
2231         .commit = blockdev_backup_commit,
2232         .abort = blockdev_backup_abort,
2233         .clean = blockdev_backup_clean,
2234     },
2235     [TRANSACTION_ACTION_KIND_ABORT] = {
2236         .instance_size = sizeof(BlkActionState),
2237         .prepare = abort_prepare,
2238         .commit = abort_commit,
2239     },
2240     [TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC] = {
2241         .instance_size = sizeof(InternalSnapshotState),
2242         .prepare  = internal_snapshot_prepare,
2243         .abort = internal_snapshot_abort,
2244         .clean = internal_snapshot_clean,
2245     },
2246     [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_ADD] = {
2247         .instance_size = sizeof(BlockDirtyBitmapState),
2248         .prepare = block_dirty_bitmap_add_prepare,
2249         .abort = block_dirty_bitmap_add_abort,
2250     },
2251     [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_CLEAR] = {
2252         .instance_size = sizeof(BlockDirtyBitmapState),
2253         .prepare = block_dirty_bitmap_clear_prepare,
2254         .commit = block_dirty_bitmap_free_backup,
2255         .abort = block_dirty_bitmap_restore,
2256     },
2257     [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_ENABLE] = {
2258         .instance_size = sizeof(BlockDirtyBitmapState),
2259         .prepare = block_dirty_bitmap_enable_prepare,
2260         .abort = block_dirty_bitmap_enable_abort,
2261     },
2262     [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_DISABLE] = {
2263         .instance_size = sizeof(BlockDirtyBitmapState),
2264         .prepare = block_dirty_bitmap_disable_prepare,
2265         .abort = block_dirty_bitmap_disable_abort,
2266     },
2267     [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_MERGE] = {
2268         .instance_size = sizeof(BlockDirtyBitmapState),
2269         .prepare = block_dirty_bitmap_merge_prepare,
2270         .commit = block_dirty_bitmap_free_backup,
2271         .abort = block_dirty_bitmap_restore,
2272     },
2273     [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_REMOVE] = {
2274         .instance_size = sizeof(BlockDirtyBitmapState),
2275         .prepare = block_dirty_bitmap_remove_prepare,
2276         .commit = block_dirty_bitmap_remove_commit,
2277         .abort = block_dirty_bitmap_remove_abort,
2278     },
2279     /* Where are transactions for MIRROR, COMMIT and STREAM?
2280      * Although these blockjobs use transaction callbacks like the backup job,
2281      * these jobs do not necessarily adhere to transaction semantics.
2282      * These jobs may not fully undo all of their actions on abort, nor do they
2283      * necessarily work in transactions with more than one job in them.
2284      */
2285 };
2286 
2287 /**
2288  * Allocate a TransactionProperties structure if necessary, and fill
2289  * that structure with desired defaults if they are unset.
2290  */
2291 static TransactionProperties *get_transaction_properties(
2292     TransactionProperties *props)
2293 {
2294     if (!props) {
2295         props = g_new0(TransactionProperties, 1);
2296     }
2297 
2298     if (!props->has_completion_mode) {
2299         props->has_completion_mode = true;
2300         props->completion_mode = ACTION_COMPLETION_MODE_INDIVIDUAL;
2301     }
2302 
2303     return props;
2304 }
2305 
2306 /*
2307  * 'Atomic' group operations.  The operations are performed as a set, and if
2308  * any fail then we roll back all operations in the group.
2309  *
2310  * Always run under BQL.
2311  */
2312 void qmp_transaction(TransactionActionList *dev_list,
2313                      bool has_props,
2314                      struct TransactionProperties *props,
2315                      Error **errp)
2316 {
2317     TransactionActionList *dev_entry = dev_list;
2318     JobTxn *block_job_txn = NULL;
2319     BlkActionState *state, *next;
2320     Error *local_err = NULL;
2321 
2322     GLOBAL_STATE_CODE();
2323 
2324     QTAILQ_HEAD(, BlkActionState) snap_bdrv_states;
2325     QTAILQ_INIT(&snap_bdrv_states);
2326 
2327     /* Does this transaction get canceled as a group on failure?
2328      * If not, we don't really need to make a JobTxn.
2329      */
2330     props = get_transaction_properties(props);
2331     if (props->completion_mode != ACTION_COMPLETION_MODE_INDIVIDUAL) {
2332         block_job_txn = job_txn_new();
2333     }
2334 
2335     /* drain all i/o before any operations */
2336     bdrv_drain_all();
2337 
2338     /* We don't do anything in this loop that commits us to the operations */
2339     while (NULL != dev_entry) {
2340         TransactionAction *dev_info = NULL;
2341         const BlkActionOps *ops;
2342 
2343         dev_info = dev_entry->value;
2344         dev_entry = dev_entry->next;
2345 
2346         assert(dev_info->type < ARRAY_SIZE(actions));
2347 
2348         ops = &actions[dev_info->type];
2349         assert(ops->instance_size > 0);
2350 
2351         state = g_malloc0(ops->instance_size);
2352         state->ops = ops;
2353         state->action = dev_info;
2354         state->block_job_txn = block_job_txn;
2355         state->txn_props = props;
2356         QTAILQ_INSERT_TAIL(&snap_bdrv_states, state, entry);
2357 
2358         state->ops->prepare(state, &local_err);
2359         if (local_err) {
2360             error_propagate(errp, local_err);
2361             goto delete_and_fail;
2362         }
2363     }
2364 
2365     QTAILQ_FOREACH(state, &snap_bdrv_states, entry) {
2366         if (state->ops->commit) {
2367             state->ops->commit(state);
2368         }
2369     }
2370 
2371     /* success */
2372     goto exit;
2373 
2374 delete_and_fail:
2375     /* failure, and it is all-or-none; roll back all operations */
2376     QTAILQ_FOREACH_REVERSE(state, &snap_bdrv_states, entry) {
2377         if (state->ops->abort) {
2378             state->ops->abort(state);
2379         }
2380     }
2381 exit:
2382     QTAILQ_FOREACH_SAFE(state, &snap_bdrv_states, entry, next) {
2383         if (state->ops->clean) {
2384             state->ops->clean(state);
2385         }
2386         g_free(state);
2387     }
2388     if (!has_props) {
2389         qapi_free_TransactionProperties(props);
2390     }
2391     job_txn_unref(block_job_txn);
2392 }
2393 
2394 BlockDirtyBitmapSha256 *qmp_x_debug_block_dirty_bitmap_sha256(const char *node,
2395                                                               const char *name,
2396                                                               Error **errp)
2397 {
2398     BdrvDirtyBitmap *bitmap;
2399     BlockDriverState *bs;
2400     BlockDirtyBitmapSha256 *ret = NULL;
2401     char *sha256;
2402 
2403     bitmap = block_dirty_bitmap_lookup(node, name, &bs, errp);
2404     if (!bitmap || !bs) {
2405         return NULL;
2406     }
2407 
2408     sha256 = bdrv_dirty_bitmap_sha256(bitmap, errp);
2409     if (sha256 == NULL) {
2410         return NULL;
2411     }
2412 
2413     ret = g_new(BlockDirtyBitmapSha256, 1);
2414     ret->sha256 = sha256;
2415 
2416     return ret;
2417 }
2418 
2419 void coroutine_fn qmp_block_resize(bool has_device, const char *device,
2420                                    bool has_node_name, const char *node_name,
2421                                    int64_t size, Error **errp)
2422 {
2423     Error *local_err = NULL;
2424     BlockBackend *blk;
2425     BlockDriverState *bs;
2426     AioContext *old_ctx;
2427 
2428     bs = bdrv_lookup_bs(has_device ? device : NULL,
2429                         has_node_name ? node_name : NULL,
2430                         &local_err);
2431     if (local_err) {
2432         error_propagate(errp, local_err);
2433         return;
2434     }
2435 
2436     if (size < 0) {
2437         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "size", "a >0 size");
2438         return;
2439     }
2440 
2441     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_RESIZE, NULL)) {
2442         error_setg(errp, QERR_DEVICE_IN_USE, device);
2443         return;
2444     }
2445 
2446     blk = blk_new_with_bs(bs, BLK_PERM_RESIZE, BLK_PERM_ALL, errp);
2447     if (!blk) {
2448         return;
2449     }
2450 
2451     bdrv_co_lock(bs);
2452     bdrv_drained_begin(bs);
2453     bdrv_co_unlock(bs);
2454 
2455     old_ctx = bdrv_co_enter(bs);
2456     blk_truncate(blk, size, false, PREALLOC_MODE_OFF, 0, errp);
2457     bdrv_co_leave(bs, old_ctx);
2458 
2459     bdrv_co_lock(bs);
2460     bdrv_drained_end(bs);
2461     blk_unref(blk);
2462     bdrv_co_unlock(bs);
2463 }
2464 
2465 void qmp_block_stream(bool has_job_id, const char *job_id, const char *device,
2466                       bool has_base, const char *base,
2467                       bool has_base_node, const char *base_node,
2468                       bool has_backing_file, const char *backing_file,
2469                       bool has_bottom, const char *bottom,
2470                       bool has_speed, int64_t speed,
2471                       bool has_on_error, BlockdevOnError on_error,
2472                       bool has_filter_node_name, const char *filter_node_name,
2473                       bool has_auto_finalize, bool auto_finalize,
2474                       bool has_auto_dismiss, bool auto_dismiss,
2475                       Error **errp)
2476 {
2477     BlockDriverState *bs, *iter, *iter_end;
2478     BlockDriverState *base_bs = NULL;
2479     BlockDriverState *bottom_bs = NULL;
2480     AioContext *aio_context;
2481     Error *local_err = NULL;
2482     int job_flags = JOB_DEFAULT;
2483 
2484     if (has_base && has_base_node) {
2485         error_setg(errp, "'base' and 'base-node' cannot be specified "
2486                    "at the same time");
2487         return;
2488     }
2489 
2490     if (has_base && has_bottom) {
2491         error_setg(errp, "'base' and 'bottom' cannot be specified "
2492                    "at the same time");
2493         return;
2494     }
2495 
2496     if (has_bottom && has_base_node) {
2497         error_setg(errp, "'bottom' and 'base-node' cannot be specified "
2498                    "at the same time");
2499         return;
2500     }
2501 
2502     if (!has_on_error) {
2503         on_error = BLOCKDEV_ON_ERROR_REPORT;
2504     }
2505 
2506     bs = bdrv_lookup_bs(device, device, errp);
2507     if (!bs) {
2508         return;
2509     }
2510 
2511     aio_context = bdrv_get_aio_context(bs);
2512     aio_context_acquire(aio_context);
2513 
2514     if (has_base) {
2515         base_bs = bdrv_find_backing_image(bs, base);
2516         if (base_bs == NULL) {
2517             error_setg(errp, "Can't find '%s' in the backing chain", base);
2518             goto out;
2519         }
2520         assert(bdrv_get_aio_context(base_bs) == aio_context);
2521     }
2522 
2523     if (has_base_node) {
2524         base_bs = bdrv_lookup_bs(NULL, base_node, errp);
2525         if (!base_bs) {
2526             goto out;
2527         }
2528         if (bs == base_bs || !bdrv_chain_contains(bs, base_bs)) {
2529             error_setg(errp, "Node '%s' is not a backing image of '%s'",
2530                        base_node, device);
2531             goto out;
2532         }
2533         assert(bdrv_get_aio_context(base_bs) == aio_context);
2534         bdrv_refresh_filename(base_bs);
2535     }
2536 
2537     if (has_bottom) {
2538         bottom_bs = bdrv_lookup_bs(NULL, bottom, errp);
2539         if (!bottom_bs) {
2540             goto out;
2541         }
2542         if (!bottom_bs->drv) {
2543             error_setg(errp, "Node '%s' is not open", bottom);
2544             goto out;
2545         }
2546         if (bottom_bs->drv->is_filter) {
2547             error_setg(errp, "Node '%s' is a filter, use a non-filter node "
2548                        "as 'bottom'", bottom);
2549             goto out;
2550         }
2551         if (!bdrv_chain_contains(bs, bottom_bs)) {
2552             error_setg(errp, "Node '%s' is not in a chain starting from '%s'",
2553                        bottom, device);
2554             goto out;
2555         }
2556         assert(bdrv_get_aio_context(bottom_bs) == aio_context);
2557     }
2558 
2559     /*
2560      * Check for op blockers in the whole chain between bs and base (or bottom)
2561      */
2562     iter_end = has_bottom ? bdrv_filter_or_cow_bs(bottom_bs) : base_bs;
2563     for (iter = bs; iter && iter != iter_end;
2564          iter = bdrv_filter_or_cow_bs(iter))
2565     {
2566         if (bdrv_op_is_blocked(iter, BLOCK_OP_TYPE_STREAM, errp)) {
2567             goto out;
2568         }
2569     }
2570 
2571     /* if we are streaming the entire chain, the result will have no backing
2572      * file, and specifying one is therefore an error */
2573     if (base_bs == NULL && has_backing_file) {
2574         error_setg(errp, "backing file specified, but streaming the "
2575                          "entire chain");
2576         goto out;
2577     }
2578 
2579     if (has_auto_finalize && !auto_finalize) {
2580         job_flags |= JOB_MANUAL_FINALIZE;
2581     }
2582     if (has_auto_dismiss && !auto_dismiss) {
2583         job_flags |= JOB_MANUAL_DISMISS;
2584     }
2585 
2586     stream_start(has_job_id ? job_id : NULL, bs, base_bs, backing_file,
2587                  bottom_bs, job_flags, has_speed ? speed : 0, on_error,
2588                  filter_node_name, &local_err);
2589     if (local_err) {
2590         error_propagate(errp, local_err);
2591         goto out;
2592     }
2593 
2594     trace_qmp_block_stream(bs);
2595 
2596 out:
2597     aio_context_release(aio_context);
2598 }
2599 
2600 void qmp_block_commit(bool has_job_id, const char *job_id, const char *device,
2601                       bool has_base_node, const char *base_node,
2602                       bool has_base, const char *base,
2603                       bool has_top_node, const char *top_node,
2604                       bool has_top, const char *top,
2605                       bool has_backing_file, const char *backing_file,
2606                       bool has_speed, int64_t speed,
2607                       bool has_on_error, BlockdevOnError on_error,
2608                       bool has_filter_node_name, const char *filter_node_name,
2609                       bool has_auto_finalize, bool auto_finalize,
2610                       bool has_auto_dismiss, bool auto_dismiss,
2611                       Error **errp)
2612 {
2613     BlockDriverState *bs;
2614     BlockDriverState *iter;
2615     BlockDriverState *base_bs, *top_bs;
2616     AioContext *aio_context;
2617     Error *local_err = NULL;
2618     int job_flags = JOB_DEFAULT;
2619     uint64_t top_perm, top_shared;
2620 
2621     if (!has_speed) {
2622         speed = 0;
2623     }
2624     if (!has_on_error) {
2625         on_error = BLOCKDEV_ON_ERROR_REPORT;
2626     }
2627     if (!has_filter_node_name) {
2628         filter_node_name = NULL;
2629     }
2630     if (has_auto_finalize && !auto_finalize) {
2631         job_flags |= JOB_MANUAL_FINALIZE;
2632     }
2633     if (has_auto_dismiss && !auto_dismiss) {
2634         job_flags |= JOB_MANUAL_DISMISS;
2635     }
2636 
2637     /* Important Note:
2638      *  libvirt relies on the DeviceNotFound error class in order to probe for
2639      *  live commit feature versions; for this to work, we must make sure to
2640      *  perform the device lookup before any generic errors that may occur in a
2641      *  scenario in which all optional arguments are omitted. */
2642     bs = qmp_get_root_bs(device, &local_err);
2643     if (!bs) {
2644         bs = bdrv_lookup_bs(device, device, NULL);
2645         if (!bs) {
2646             error_free(local_err);
2647             error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
2648                       "Device '%s' not found", device);
2649         } else {
2650             error_propagate(errp, local_err);
2651         }
2652         return;
2653     }
2654 
2655     aio_context = bdrv_get_aio_context(bs);
2656     aio_context_acquire(aio_context);
2657 
2658     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_COMMIT_SOURCE, errp)) {
2659         goto out;
2660     }
2661 
2662     /* default top_bs is the active layer */
2663     top_bs = bs;
2664 
2665     if (has_top_node && has_top) {
2666         error_setg(errp, "'top-node' and 'top' are mutually exclusive");
2667         goto out;
2668     } else if (has_top_node) {
2669         top_bs = bdrv_lookup_bs(NULL, top_node, errp);
2670         if (top_bs == NULL) {
2671             goto out;
2672         }
2673         if (!bdrv_chain_contains(bs, top_bs)) {
2674             error_setg(errp, "'%s' is not in this backing file chain",
2675                        top_node);
2676             goto out;
2677         }
2678     } else if (has_top && top) {
2679         /* This strcmp() is just a shortcut, there is no need to
2680          * refresh @bs's filename.  If it mismatches,
2681          * bdrv_find_backing_image() will do the refresh and may still
2682          * return @bs. */
2683         if (strcmp(bs->filename, top) != 0) {
2684             top_bs = bdrv_find_backing_image(bs, top);
2685         }
2686     }
2687 
2688     if (top_bs == NULL) {
2689         error_setg(errp, "Top image file %s not found", top ? top : "NULL");
2690         goto out;
2691     }
2692 
2693     assert(bdrv_get_aio_context(top_bs) == aio_context);
2694 
2695     if (has_base_node && has_base) {
2696         error_setg(errp, "'base-node' and 'base' are mutually exclusive");
2697         goto out;
2698     } else if (has_base_node) {
2699         base_bs = bdrv_lookup_bs(NULL, base_node, errp);
2700         if (base_bs == NULL) {
2701             goto out;
2702         }
2703         if (!bdrv_chain_contains(top_bs, base_bs)) {
2704             error_setg(errp, "'%s' is not in this backing file chain",
2705                        base_node);
2706             goto out;
2707         }
2708     } else if (has_base && base) {
2709         base_bs = bdrv_find_backing_image(top_bs, base);
2710         if (base_bs == NULL) {
2711             error_setg(errp, "Can't find '%s' in the backing chain", base);
2712             goto out;
2713         }
2714     } else {
2715         base_bs = bdrv_find_base(top_bs);
2716         if (base_bs == NULL) {
2717             error_setg(errp, "There is no backimg image");
2718             goto out;
2719         }
2720     }
2721 
2722     assert(bdrv_get_aio_context(base_bs) == aio_context);
2723 
2724     for (iter = top_bs; iter != bdrv_filter_or_cow_bs(base_bs);
2725          iter = bdrv_filter_or_cow_bs(iter))
2726     {
2727         if (bdrv_op_is_blocked(iter, BLOCK_OP_TYPE_COMMIT_TARGET, errp)) {
2728             goto out;
2729         }
2730     }
2731 
2732     /* Do not allow attempts to commit an image into itself */
2733     if (top_bs == base_bs) {
2734         error_setg(errp, "cannot commit an image into itself");
2735         goto out;
2736     }
2737 
2738     /*
2739      * Active commit is required if and only if someone has taken a
2740      * WRITE permission on the top node.  Historically, we have always
2741      * used active commit for top nodes, so continue that practice
2742      * lest we possibly break clients that rely on this behavior, e.g.
2743      * to later attach this node to a writing parent.
2744      * (Active commit is never really wrong.)
2745      */
2746     bdrv_get_cumulative_perm(top_bs, &top_perm, &top_shared);
2747     if (top_perm & BLK_PERM_WRITE ||
2748         bdrv_skip_filters(top_bs) == bdrv_skip_filters(bs))
2749     {
2750         if (has_backing_file) {
2751             if (bdrv_skip_filters(top_bs) == bdrv_skip_filters(bs)) {
2752                 error_setg(errp, "'backing-file' specified,"
2753                                  " but 'top' is the active layer");
2754             } else {
2755                 error_setg(errp, "'backing-file' specified, but 'top' has a "
2756                                  "writer on it");
2757             }
2758             goto out;
2759         }
2760         if (!has_job_id) {
2761             /*
2762              * Emulate here what block_job_create() does, because it
2763              * is possible that @bs != @top_bs (the block job should
2764              * be named after @bs, even if @top_bs is the actual
2765              * source)
2766              */
2767             job_id = bdrv_get_device_name(bs);
2768         }
2769         commit_active_start(job_id, top_bs, base_bs, job_flags, speed, on_error,
2770                             filter_node_name, NULL, NULL, false, &local_err);
2771     } else {
2772         BlockDriverState *overlay_bs = bdrv_find_overlay(bs, top_bs);
2773         if (bdrv_op_is_blocked(overlay_bs, BLOCK_OP_TYPE_COMMIT_TARGET, errp)) {
2774             goto out;
2775         }
2776         commit_start(has_job_id ? job_id : NULL, bs, base_bs, top_bs, job_flags,
2777                      speed, on_error, has_backing_file ? backing_file : NULL,
2778                      filter_node_name, &local_err);
2779     }
2780     if (local_err != NULL) {
2781         error_propagate(errp, local_err);
2782         goto out;
2783     }
2784 
2785 out:
2786     aio_context_release(aio_context);
2787 }
2788 
2789 /* Common QMP interface for drive-backup and blockdev-backup */
2790 static BlockJob *do_backup_common(BackupCommon *backup,
2791                                   BlockDriverState *bs,
2792                                   BlockDriverState *target_bs,
2793                                   AioContext *aio_context,
2794                                   JobTxn *txn, Error **errp)
2795 {
2796     BlockJob *job = NULL;
2797     BdrvDirtyBitmap *bmap = NULL;
2798     BackupPerf perf = { .max_workers = 64 };
2799     int job_flags = JOB_DEFAULT;
2800 
2801     if (!backup->has_speed) {
2802         backup->speed = 0;
2803     }
2804     if (!backup->has_on_source_error) {
2805         backup->on_source_error = BLOCKDEV_ON_ERROR_REPORT;
2806     }
2807     if (!backup->has_on_target_error) {
2808         backup->on_target_error = BLOCKDEV_ON_ERROR_REPORT;
2809     }
2810     if (!backup->has_job_id) {
2811         backup->job_id = NULL;
2812     }
2813     if (!backup->has_auto_finalize) {
2814         backup->auto_finalize = true;
2815     }
2816     if (!backup->has_auto_dismiss) {
2817         backup->auto_dismiss = true;
2818     }
2819     if (!backup->has_compress) {
2820         backup->compress = false;
2821     }
2822 
2823     if (backup->x_perf) {
2824         if (backup->x_perf->has_use_copy_range) {
2825             perf.use_copy_range = backup->x_perf->use_copy_range;
2826         }
2827         if (backup->x_perf->has_max_workers) {
2828             perf.max_workers = backup->x_perf->max_workers;
2829         }
2830         if (backup->x_perf->has_max_chunk) {
2831             perf.max_chunk = backup->x_perf->max_chunk;
2832         }
2833     }
2834 
2835     if ((backup->sync == MIRROR_SYNC_MODE_BITMAP) ||
2836         (backup->sync == MIRROR_SYNC_MODE_INCREMENTAL)) {
2837         /* done before desugaring 'incremental' to print the right message */
2838         if (!backup->has_bitmap) {
2839             error_setg(errp, "must provide a valid bitmap name for "
2840                        "'%s' sync mode", MirrorSyncMode_str(backup->sync));
2841             return NULL;
2842         }
2843     }
2844 
2845     if (backup->sync == MIRROR_SYNC_MODE_INCREMENTAL) {
2846         if (backup->has_bitmap_mode &&
2847             backup->bitmap_mode != BITMAP_SYNC_MODE_ON_SUCCESS) {
2848             error_setg(errp, "Bitmap sync mode must be '%s' "
2849                        "when using sync mode '%s'",
2850                        BitmapSyncMode_str(BITMAP_SYNC_MODE_ON_SUCCESS),
2851                        MirrorSyncMode_str(backup->sync));
2852             return NULL;
2853         }
2854         backup->has_bitmap_mode = true;
2855         backup->sync = MIRROR_SYNC_MODE_BITMAP;
2856         backup->bitmap_mode = BITMAP_SYNC_MODE_ON_SUCCESS;
2857     }
2858 
2859     if (backup->has_bitmap) {
2860         bmap = bdrv_find_dirty_bitmap(bs, backup->bitmap);
2861         if (!bmap) {
2862             error_setg(errp, "Bitmap '%s' could not be found", backup->bitmap);
2863             return NULL;
2864         }
2865         if (!backup->has_bitmap_mode) {
2866             error_setg(errp, "Bitmap sync mode must be given "
2867                        "when providing a bitmap");
2868             return NULL;
2869         }
2870         if (bdrv_dirty_bitmap_check(bmap, BDRV_BITMAP_ALLOW_RO, errp)) {
2871             return NULL;
2872         }
2873 
2874         /* This does not produce a useful bitmap artifact: */
2875         if (backup->sync == MIRROR_SYNC_MODE_NONE) {
2876             error_setg(errp, "sync mode '%s' does not produce meaningful bitmap"
2877                        " outputs", MirrorSyncMode_str(backup->sync));
2878             return NULL;
2879         }
2880 
2881         /* If the bitmap isn't used for input or output, this is useless: */
2882         if (backup->bitmap_mode == BITMAP_SYNC_MODE_NEVER &&
2883             backup->sync != MIRROR_SYNC_MODE_BITMAP) {
2884             error_setg(errp, "Bitmap sync mode '%s' has no meaningful effect"
2885                        " when combined with sync mode '%s'",
2886                        BitmapSyncMode_str(backup->bitmap_mode),
2887                        MirrorSyncMode_str(backup->sync));
2888             return NULL;
2889         }
2890     }
2891 
2892     if (!backup->has_bitmap && backup->has_bitmap_mode) {
2893         error_setg(errp, "Cannot specify bitmap sync mode without a bitmap");
2894         return NULL;
2895     }
2896 
2897     if (!backup->auto_finalize) {
2898         job_flags |= JOB_MANUAL_FINALIZE;
2899     }
2900     if (!backup->auto_dismiss) {
2901         job_flags |= JOB_MANUAL_DISMISS;
2902     }
2903 
2904     job = backup_job_create(backup->job_id, bs, target_bs, backup->speed,
2905                             backup->sync, bmap, backup->bitmap_mode,
2906                             backup->compress,
2907                             backup->filter_node_name,
2908                             &perf,
2909                             backup->on_source_error,
2910                             backup->on_target_error,
2911                             job_flags, NULL, NULL, txn, errp);
2912     return job;
2913 }
2914 
2915 void qmp_drive_backup(DriveBackup *backup, Error **errp)
2916 {
2917     TransactionAction action = {
2918         .type = TRANSACTION_ACTION_KIND_DRIVE_BACKUP,
2919         .u.drive_backup.data = backup,
2920     };
2921     blockdev_do_action(&action, errp);
2922 }
2923 
2924 BlockDeviceInfoList *qmp_query_named_block_nodes(bool has_flat,
2925                                                  bool flat,
2926                                                  Error **errp)
2927 {
2928     bool return_flat = has_flat && flat;
2929 
2930     return bdrv_named_nodes_list(return_flat, errp);
2931 }
2932 
2933 XDbgBlockGraph *qmp_x_debug_query_block_graph(Error **errp)
2934 {
2935     return bdrv_get_xdbg_block_graph(errp);
2936 }
2937 
2938 void qmp_blockdev_backup(BlockdevBackup *backup, Error **errp)
2939 {
2940     TransactionAction action = {
2941         .type = TRANSACTION_ACTION_KIND_BLOCKDEV_BACKUP,
2942         .u.blockdev_backup.data = backup,
2943     };
2944     blockdev_do_action(&action, errp);
2945 }
2946 
2947 /* Parameter check and block job starting for drive mirroring.
2948  * Caller should hold @device and @target's aio context (must be the same).
2949  **/
2950 static void blockdev_mirror_common(const char *job_id, BlockDriverState *bs,
2951                                    BlockDriverState *target,
2952                                    bool has_replaces, const char *replaces,
2953                                    enum MirrorSyncMode sync,
2954                                    BlockMirrorBackingMode backing_mode,
2955                                    bool zero_target,
2956                                    bool has_speed, int64_t speed,
2957                                    bool has_granularity, uint32_t granularity,
2958                                    bool has_buf_size, int64_t buf_size,
2959                                    bool has_on_source_error,
2960                                    BlockdevOnError on_source_error,
2961                                    bool has_on_target_error,
2962                                    BlockdevOnError on_target_error,
2963                                    bool has_unmap, bool unmap,
2964                                    bool has_filter_node_name,
2965                                    const char *filter_node_name,
2966                                    bool has_copy_mode, MirrorCopyMode copy_mode,
2967                                    bool has_auto_finalize, bool auto_finalize,
2968                                    bool has_auto_dismiss, bool auto_dismiss,
2969                                    Error **errp)
2970 {
2971     BlockDriverState *unfiltered_bs;
2972     int job_flags = JOB_DEFAULT;
2973 
2974     if (!has_speed) {
2975         speed = 0;
2976     }
2977     if (!has_on_source_error) {
2978         on_source_error = BLOCKDEV_ON_ERROR_REPORT;
2979     }
2980     if (!has_on_target_error) {
2981         on_target_error = BLOCKDEV_ON_ERROR_REPORT;
2982     }
2983     if (!has_granularity) {
2984         granularity = 0;
2985     }
2986     if (!has_buf_size) {
2987         buf_size = 0;
2988     }
2989     if (!has_unmap) {
2990         unmap = true;
2991     }
2992     if (!has_filter_node_name) {
2993         filter_node_name = NULL;
2994     }
2995     if (!has_copy_mode) {
2996         copy_mode = MIRROR_COPY_MODE_BACKGROUND;
2997     }
2998     if (has_auto_finalize && !auto_finalize) {
2999         job_flags |= JOB_MANUAL_FINALIZE;
3000     }
3001     if (has_auto_dismiss && !auto_dismiss) {
3002         job_flags |= JOB_MANUAL_DISMISS;
3003     }
3004 
3005     if (granularity != 0 && (granularity < 512 || granularity > 1048576 * 64)) {
3006         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "granularity",
3007                    "a value in range [512B, 64MB]");
3008         return;
3009     }
3010     if (granularity & (granularity - 1)) {
3011         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "granularity",
3012                    "a power of 2");
3013         return;
3014     }
3015 
3016     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_MIRROR_SOURCE, errp)) {
3017         return;
3018     }
3019     if (bdrv_op_is_blocked(target, BLOCK_OP_TYPE_MIRROR_TARGET, errp)) {
3020         return;
3021     }
3022 
3023     if (!bdrv_backing_chain_next(bs) && sync == MIRROR_SYNC_MODE_TOP) {
3024         sync = MIRROR_SYNC_MODE_FULL;
3025     }
3026 
3027     if (!has_replaces) {
3028         /* We want to mirror from @bs, but keep implicit filters on top */
3029         unfiltered_bs = bdrv_skip_implicit_filters(bs);
3030         if (unfiltered_bs != bs) {
3031             replaces = unfiltered_bs->node_name;
3032             has_replaces = true;
3033         }
3034     }
3035 
3036     if (has_replaces) {
3037         BlockDriverState *to_replace_bs;
3038         AioContext *replace_aio_context;
3039         int64_t bs_size, replace_size;
3040 
3041         bs_size = bdrv_getlength(bs);
3042         if (bs_size < 0) {
3043             error_setg_errno(errp, -bs_size, "Failed to query device's size");
3044             return;
3045         }
3046 
3047         to_replace_bs = check_to_replace_node(bs, replaces, errp);
3048         if (!to_replace_bs) {
3049             return;
3050         }
3051 
3052         replace_aio_context = bdrv_get_aio_context(to_replace_bs);
3053         aio_context_acquire(replace_aio_context);
3054         replace_size = bdrv_getlength(to_replace_bs);
3055         aio_context_release(replace_aio_context);
3056 
3057         if (replace_size < 0) {
3058             error_setg_errno(errp, -replace_size,
3059                              "Failed to query the replacement node's size");
3060             return;
3061         }
3062         if (bs_size != replace_size) {
3063             error_setg(errp, "cannot replace image with a mirror image of "
3064                              "different size");
3065             return;
3066         }
3067     }
3068 
3069     /* pass the node name to replace to mirror start since it's loose coupling
3070      * and will allow to check whether the node still exist at mirror completion
3071      */
3072     mirror_start(job_id, bs, target,
3073                  has_replaces ? replaces : NULL, job_flags,
3074                  speed, granularity, buf_size, sync, backing_mode, zero_target,
3075                  on_source_error, on_target_error, unmap, filter_node_name,
3076                  copy_mode, errp);
3077 }
3078 
3079 void qmp_drive_mirror(DriveMirror *arg, Error **errp)
3080 {
3081     BlockDriverState *bs;
3082     BlockDriverState *target_backing_bs, *target_bs;
3083     AioContext *aio_context;
3084     AioContext *old_context;
3085     BlockMirrorBackingMode backing_mode;
3086     Error *local_err = NULL;
3087     QDict *options = NULL;
3088     int flags;
3089     int64_t size;
3090     const char *format = arg->format;
3091     bool zero_target;
3092     int ret;
3093 
3094     bs = qmp_get_root_bs(arg->device, errp);
3095     if (!bs) {
3096         return;
3097     }
3098 
3099     /* Early check to avoid creating target */
3100     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_MIRROR_SOURCE, errp)) {
3101         return;
3102     }
3103 
3104     aio_context = bdrv_get_aio_context(bs);
3105     aio_context_acquire(aio_context);
3106 
3107     if (!arg->has_mode) {
3108         arg->mode = NEW_IMAGE_MODE_ABSOLUTE_PATHS;
3109     }
3110 
3111     if (!arg->has_format) {
3112         format = (arg->mode == NEW_IMAGE_MODE_EXISTING
3113                   ? NULL : bs->drv->format_name);
3114     }
3115 
3116     flags = bs->open_flags | BDRV_O_RDWR;
3117     target_backing_bs = bdrv_cow_bs(bdrv_skip_filters(bs));
3118     if (!target_backing_bs && arg->sync == MIRROR_SYNC_MODE_TOP) {
3119         arg->sync = MIRROR_SYNC_MODE_FULL;
3120     }
3121     if (arg->sync == MIRROR_SYNC_MODE_NONE) {
3122         target_backing_bs = bs;
3123     }
3124 
3125     size = bdrv_getlength(bs);
3126     if (size < 0) {
3127         error_setg_errno(errp, -size, "bdrv_getlength failed");
3128         goto out;
3129     }
3130 
3131     if (arg->has_replaces) {
3132         if (!arg->has_node_name) {
3133             error_setg(errp, "a node-name must be provided when replacing a"
3134                              " named node of the graph");
3135             goto out;
3136         }
3137     }
3138 
3139     if (arg->mode == NEW_IMAGE_MODE_ABSOLUTE_PATHS) {
3140         backing_mode = MIRROR_SOURCE_BACKING_CHAIN;
3141     } else {
3142         backing_mode = MIRROR_OPEN_BACKING_CHAIN;
3143     }
3144 
3145     /* Don't open backing image in create() */
3146     flags |= BDRV_O_NO_BACKING;
3147 
3148     if ((arg->sync == MIRROR_SYNC_MODE_FULL || !target_backing_bs)
3149         && arg->mode != NEW_IMAGE_MODE_EXISTING)
3150     {
3151         /* create new image w/o backing file */
3152         assert(format);
3153         bdrv_img_create(arg->target, format,
3154                         NULL, NULL, NULL, size, flags, false, &local_err);
3155     } else {
3156         /* Implicit filters should not appear in the filename */
3157         BlockDriverState *explicit_backing =
3158             bdrv_skip_implicit_filters(target_backing_bs);
3159 
3160         switch (arg->mode) {
3161         case NEW_IMAGE_MODE_EXISTING:
3162             break;
3163         case NEW_IMAGE_MODE_ABSOLUTE_PATHS:
3164             /* create new image with backing file */
3165             bdrv_refresh_filename(explicit_backing);
3166             bdrv_img_create(arg->target, format,
3167                             explicit_backing->filename,
3168                             explicit_backing->drv->format_name,
3169                             NULL, size, flags, false, &local_err);
3170             break;
3171         default:
3172             abort();
3173         }
3174     }
3175 
3176     if (local_err) {
3177         error_propagate(errp, local_err);
3178         goto out;
3179     }
3180 
3181     options = qdict_new();
3182     if (arg->has_node_name) {
3183         qdict_put_str(options, "node-name", arg->node_name);
3184     }
3185     if (format) {
3186         qdict_put_str(options, "driver", format);
3187     }
3188 
3189     /* Mirroring takes care of copy-on-write using the source's backing
3190      * file.
3191      */
3192     target_bs = bdrv_open(arg->target, NULL, options, flags, errp);
3193     if (!target_bs) {
3194         goto out;
3195     }
3196 
3197     zero_target = (arg->sync == MIRROR_SYNC_MODE_FULL &&
3198                    (arg->mode == NEW_IMAGE_MODE_EXISTING ||
3199                     !bdrv_has_zero_init(target_bs)));
3200 
3201 
3202     /* Honor bdrv_try_set_aio_context() context acquisition requirements. */
3203     old_context = bdrv_get_aio_context(target_bs);
3204     aio_context_release(aio_context);
3205     aio_context_acquire(old_context);
3206 
3207     ret = bdrv_try_set_aio_context(target_bs, aio_context, errp);
3208     if (ret < 0) {
3209         bdrv_unref(target_bs);
3210         aio_context_release(old_context);
3211         return;
3212     }
3213 
3214     aio_context_release(old_context);
3215     aio_context_acquire(aio_context);
3216 
3217     blockdev_mirror_common(arg->has_job_id ? arg->job_id : NULL, bs, target_bs,
3218                            arg->has_replaces, arg->replaces, arg->sync,
3219                            backing_mode, zero_target,
3220                            arg->has_speed, arg->speed,
3221                            arg->has_granularity, arg->granularity,
3222                            arg->has_buf_size, arg->buf_size,
3223                            arg->has_on_source_error, arg->on_source_error,
3224                            arg->has_on_target_error, arg->on_target_error,
3225                            arg->has_unmap, arg->unmap,
3226                            false, NULL,
3227                            arg->has_copy_mode, arg->copy_mode,
3228                            arg->has_auto_finalize, arg->auto_finalize,
3229                            arg->has_auto_dismiss, arg->auto_dismiss,
3230                            errp);
3231     bdrv_unref(target_bs);
3232 out:
3233     aio_context_release(aio_context);
3234 }
3235 
3236 void qmp_blockdev_mirror(bool has_job_id, const char *job_id,
3237                          const char *device, const char *target,
3238                          bool has_replaces, const char *replaces,
3239                          MirrorSyncMode sync,
3240                          bool has_speed, int64_t speed,
3241                          bool has_granularity, uint32_t granularity,
3242                          bool has_buf_size, int64_t buf_size,
3243                          bool has_on_source_error,
3244                          BlockdevOnError on_source_error,
3245                          bool has_on_target_error,
3246                          BlockdevOnError on_target_error,
3247                          bool has_filter_node_name,
3248                          const char *filter_node_name,
3249                          bool has_copy_mode, MirrorCopyMode copy_mode,
3250                          bool has_auto_finalize, bool auto_finalize,
3251                          bool has_auto_dismiss, bool auto_dismiss,
3252                          Error **errp)
3253 {
3254     BlockDriverState *bs;
3255     BlockDriverState *target_bs;
3256     AioContext *aio_context;
3257     AioContext *old_context;
3258     BlockMirrorBackingMode backing_mode = MIRROR_LEAVE_BACKING_CHAIN;
3259     bool zero_target;
3260     int ret;
3261 
3262     bs = qmp_get_root_bs(device, errp);
3263     if (!bs) {
3264         return;
3265     }
3266 
3267     target_bs = bdrv_lookup_bs(target, target, errp);
3268     if (!target_bs) {
3269         return;
3270     }
3271 
3272     zero_target = (sync == MIRROR_SYNC_MODE_FULL);
3273 
3274     /* Honor bdrv_try_set_aio_context() context acquisition requirements. */
3275     old_context = bdrv_get_aio_context(target_bs);
3276     aio_context = bdrv_get_aio_context(bs);
3277     aio_context_acquire(old_context);
3278 
3279     ret = bdrv_try_set_aio_context(target_bs, aio_context, errp);
3280 
3281     aio_context_release(old_context);
3282     aio_context_acquire(aio_context);
3283 
3284     if (ret < 0) {
3285         goto out;
3286     }
3287 
3288     blockdev_mirror_common(has_job_id ? job_id : NULL, bs, target_bs,
3289                            has_replaces, replaces, sync, backing_mode,
3290                            zero_target, has_speed, speed,
3291                            has_granularity, granularity,
3292                            has_buf_size, buf_size,
3293                            has_on_source_error, on_source_error,
3294                            has_on_target_error, on_target_error,
3295                            true, true,
3296                            has_filter_node_name, filter_node_name,
3297                            has_copy_mode, copy_mode,
3298                            has_auto_finalize, auto_finalize,
3299                            has_auto_dismiss, auto_dismiss,
3300                            errp);
3301 out:
3302     aio_context_release(aio_context);
3303 }
3304 
3305 /* Get a block job using its ID and acquire its AioContext */
3306 static BlockJob *find_block_job(const char *id, AioContext **aio_context,
3307                                 Error **errp)
3308 {
3309     BlockJob *job;
3310 
3311     assert(id != NULL);
3312 
3313     *aio_context = NULL;
3314 
3315     job = block_job_get(id);
3316 
3317     if (!job) {
3318         error_set(errp, ERROR_CLASS_DEVICE_NOT_ACTIVE,
3319                   "Block job '%s' not found", id);
3320         return NULL;
3321     }
3322 
3323     *aio_context = block_job_get_aio_context(job);
3324     aio_context_acquire(*aio_context);
3325 
3326     return job;
3327 }
3328 
3329 void qmp_block_job_set_speed(const char *device, int64_t speed, Error **errp)
3330 {
3331     AioContext *aio_context;
3332     BlockJob *job = find_block_job(device, &aio_context, errp);
3333 
3334     if (!job) {
3335         return;
3336     }
3337 
3338     block_job_set_speed(job, speed, errp);
3339     aio_context_release(aio_context);
3340 }
3341 
3342 void qmp_block_job_cancel(const char *device,
3343                           bool has_force, bool force, Error **errp)
3344 {
3345     AioContext *aio_context;
3346     BlockJob *job = find_block_job(device, &aio_context, errp);
3347 
3348     if (!job) {
3349         return;
3350     }
3351 
3352     if (!has_force) {
3353         force = false;
3354     }
3355 
3356     if (job_user_paused(&job->job) && !force) {
3357         error_setg(errp, "The block job for device '%s' is currently paused",
3358                    device);
3359         goto out;
3360     }
3361 
3362     trace_qmp_block_job_cancel(job);
3363     job_user_cancel(&job->job, force, errp);
3364 out:
3365     aio_context_release(aio_context);
3366 }
3367 
3368 void qmp_block_job_pause(const char *device, Error **errp)
3369 {
3370     AioContext *aio_context;
3371     BlockJob *job = find_block_job(device, &aio_context, errp);
3372 
3373     if (!job) {
3374         return;
3375     }
3376 
3377     trace_qmp_block_job_pause(job);
3378     job_user_pause(&job->job, errp);
3379     aio_context_release(aio_context);
3380 }
3381 
3382 void qmp_block_job_resume(const char *device, Error **errp)
3383 {
3384     AioContext *aio_context;
3385     BlockJob *job = find_block_job(device, &aio_context, errp);
3386 
3387     if (!job) {
3388         return;
3389     }
3390 
3391     trace_qmp_block_job_resume(job);
3392     job_user_resume(&job->job, errp);
3393     aio_context_release(aio_context);
3394 }
3395 
3396 void qmp_block_job_complete(const char *device, Error **errp)
3397 {
3398     AioContext *aio_context;
3399     BlockJob *job = find_block_job(device, &aio_context, errp);
3400 
3401     if (!job) {
3402         return;
3403     }
3404 
3405     trace_qmp_block_job_complete(job);
3406     job_complete(&job->job, errp);
3407     aio_context_release(aio_context);
3408 }
3409 
3410 void qmp_block_job_finalize(const char *id, Error **errp)
3411 {
3412     AioContext *aio_context;
3413     BlockJob *job = find_block_job(id, &aio_context, errp);
3414 
3415     if (!job) {
3416         return;
3417     }
3418 
3419     trace_qmp_block_job_finalize(job);
3420     job_ref(&job->job);
3421     job_finalize(&job->job, errp);
3422 
3423     /*
3424      * Job's context might have changed via job_finalize (and job_txn_apply
3425      * automatically acquires the new one), so make sure we release the correct
3426      * one.
3427      */
3428     aio_context = block_job_get_aio_context(job);
3429     job_unref(&job->job);
3430     aio_context_release(aio_context);
3431 }
3432 
3433 void qmp_block_job_dismiss(const char *id, Error **errp)
3434 {
3435     AioContext *aio_context;
3436     BlockJob *bjob = find_block_job(id, &aio_context, errp);
3437     Job *job;
3438 
3439     if (!bjob) {
3440         return;
3441     }
3442 
3443     trace_qmp_block_job_dismiss(bjob);
3444     job = &bjob->job;
3445     job_dismiss(&job, errp);
3446     aio_context_release(aio_context);
3447 }
3448 
3449 void qmp_change_backing_file(const char *device,
3450                              const char *image_node_name,
3451                              const char *backing_file,
3452                              Error **errp)
3453 {
3454     BlockDriverState *bs = NULL;
3455     AioContext *aio_context;
3456     BlockDriverState *image_bs = NULL;
3457     Error *local_err = NULL;
3458     bool ro;
3459     int ret;
3460 
3461     bs = qmp_get_root_bs(device, errp);
3462     if (!bs) {
3463         return;
3464     }
3465 
3466     aio_context = bdrv_get_aio_context(bs);
3467     aio_context_acquire(aio_context);
3468 
3469     image_bs = bdrv_lookup_bs(NULL, image_node_name, &local_err);
3470     if (local_err) {
3471         error_propagate(errp, local_err);
3472         goto out;
3473     }
3474 
3475     if (!image_bs) {
3476         error_setg(errp, "image file not found");
3477         goto out;
3478     }
3479 
3480     if (bdrv_find_base(image_bs) == image_bs) {
3481         error_setg(errp, "not allowing backing file change on an image "
3482                          "without a backing file");
3483         goto out;
3484     }
3485 
3486     /* even though we are not necessarily operating on bs, we need it to
3487      * determine if block ops are currently prohibited on the chain */
3488     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_CHANGE, errp)) {
3489         goto out;
3490     }
3491 
3492     /* final sanity check */
3493     if (!bdrv_chain_contains(bs, image_bs)) {
3494         error_setg(errp, "'%s' and image file are not in the same chain",
3495                    device);
3496         goto out;
3497     }
3498 
3499     /* if not r/w, reopen to make r/w */
3500     ro = bdrv_is_read_only(image_bs);
3501 
3502     if (ro) {
3503         if (bdrv_reopen_set_read_only(image_bs, false, errp) != 0) {
3504             goto out;
3505         }
3506     }
3507 
3508     ret = bdrv_change_backing_file(image_bs, backing_file,
3509                                    image_bs->drv ? image_bs->drv->format_name : "",
3510                                    false);
3511 
3512     if (ret < 0) {
3513         error_setg_errno(errp, -ret, "Could not change backing file to '%s'",
3514                          backing_file);
3515         /* don't exit here, so we can try to restore open flags if
3516          * appropriate */
3517     }
3518 
3519     if (ro) {
3520         bdrv_reopen_set_read_only(image_bs, true, errp);
3521     }
3522 
3523 out:
3524     aio_context_release(aio_context);
3525 }
3526 
3527 void qmp_blockdev_add(BlockdevOptions *options, Error **errp)
3528 {
3529     BlockDriverState *bs;
3530     QObject *obj;
3531     Visitor *v = qobject_output_visitor_new(&obj);
3532     QDict *qdict;
3533 
3534     visit_type_BlockdevOptions(v, NULL, &options, &error_abort);
3535     visit_complete(v, &obj);
3536     qdict = qobject_to(QDict, obj);
3537 
3538     qdict_flatten(qdict);
3539 
3540     if (!qdict_get_try_str(qdict, "node-name")) {
3541         error_setg(errp, "'node-name' must be specified for the root node");
3542         goto fail;
3543     }
3544 
3545     bs = bds_tree_init(qdict, errp);
3546     if (!bs) {
3547         goto fail;
3548     }
3549 
3550     bdrv_set_monitor_owned(bs);
3551 
3552 fail:
3553     visit_free(v);
3554 }
3555 
3556 void qmp_blockdev_reopen(BlockdevOptionsList *reopen_list, Error **errp)
3557 {
3558     BlockReopenQueue *queue = NULL;
3559     GSList *drained = NULL;
3560     GSList *p;
3561 
3562     /* Add each one of the BDS that we want to reopen to the queue */
3563     for (; reopen_list != NULL; reopen_list = reopen_list->next) {
3564         BlockdevOptions *options = reopen_list->value;
3565         BlockDriverState *bs;
3566         AioContext *ctx;
3567         QObject *obj;
3568         Visitor *v;
3569         QDict *qdict;
3570 
3571         /* Check for the selected node name */
3572         if (!options->has_node_name) {
3573             error_setg(errp, "node-name not specified");
3574             goto fail;
3575         }
3576 
3577         bs = bdrv_find_node(options->node_name);
3578         if (!bs) {
3579             error_setg(errp, "Failed to find node with node-name='%s'",
3580                        options->node_name);
3581             goto fail;
3582         }
3583 
3584         /* Put all options in a QDict and flatten it */
3585         v = qobject_output_visitor_new(&obj);
3586         visit_type_BlockdevOptions(v, NULL, &options, &error_abort);
3587         visit_complete(v, &obj);
3588         visit_free(v);
3589 
3590         qdict = qobject_to(QDict, obj);
3591 
3592         qdict_flatten(qdict);
3593 
3594         ctx = bdrv_get_aio_context(bs);
3595         aio_context_acquire(ctx);
3596 
3597         bdrv_subtree_drained_begin(bs);
3598         queue = bdrv_reopen_queue(queue, bs, qdict, false);
3599         drained = g_slist_prepend(drained, bs);
3600 
3601         aio_context_release(ctx);
3602     }
3603 
3604     /* Perform the reopen operation */
3605     bdrv_reopen_multiple(queue, errp);
3606     queue = NULL;
3607 
3608 fail:
3609     bdrv_reopen_queue_free(queue);
3610     for (p = drained; p; p = p->next) {
3611         BlockDriverState *bs = p->data;
3612         AioContext *ctx = bdrv_get_aio_context(bs);
3613 
3614         aio_context_acquire(ctx);
3615         bdrv_subtree_drained_end(bs);
3616         aio_context_release(ctx);
3617     }
3618     g_slist_free(drained);
3619 }
3620 
3621 void qmp_blockdev_del(const char *node_name, Error **errp)
3622 {
3623     AioContext *aio_context;
3624     BlockDriverState *bs;
3625 
3626     GLOBAL_STATE_CODE();
3627 
3628     bs = bdrv_find_node(node_name);
3629     if (!bs) {
3630         error_setg(errp, "Failed to find node with node-name='%s'", node_name);
3631         return;
3632     }
3633     if (bdrv_has_blk(bs)) {
3634         error_setg(errp, "Node %s is in use", node_name);
3635         return;
3636     }
3637     aio_context = bdrv_get_aio_context(bs);
3638     aio_context_acquire(aio_context);
3639 
3640     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_DRIVE_DEL, errp)) {
3641         goto out;
3642     }
3643 
3644     if (!QTAILQ_IN_USE(bs, monitor_list)) {
3645         error_setg(errp, "Node %s is not owned by the monitor",
3646                    bs->node_name);
3647         goto out;
3648     }
3649 
3650     if (bs->refcnt > 1) {
3651         error_setg(errp, "Block device %s is in use",
3652                    bdrv_get_device_or_node_name(bs));
3653         goto out;
3654     }
3655 
3656     QTAILQ_REMOVE(&monitor_bdrv_states, bs, monitor_list);
3657     bdrv_unref(bs);
3658 
3659 out:
3660     aio_context_release(aio_context);
3661 }
3662 
3663 static BdrvChild *bdrv_find_child(BlockDriverState *parent_bs,
3664                                   const char *child_name)
3665 {
3666     BdrvChild *child;
3667 
3668     QLIST_FOREACH(child, &parent_bs->children, next) {
3669         if (strcmp(child->name, child_name) == 0) {
3670             return child;
3671         }
3672     }
3673 
3674     return NULL;
3675 }
3676 
3677 void qmp_x_blockdev_change(const char *parent, bool has_child,
3678                            const char *child, bool has_node,
3679                            const char *node, Error **errp)
3680 {
3681     BlockDriverState *parent_bs, *new_bs = NULL;
3682     BdrvChild *p_child;
3683 
3684     parent_bs = bdrv_lookup_bs(parent, parent, errp);
3685     if (!parent_bs) {
3686         return;
3687     }
3688 
3689     if (has_child == has_node) {
3690         if (has_child) {
3691             error_setg(errp, "The parameters child and node are in conflict");
3692         } else {
3693             error_setg(errp, "Either child or node must be specified");
3694         }
3695         return;
3696     }
3697 
3698     if (has_child) {
3699         p_child = bdrv_find_child(parent_bs, child);
3700         if (!p_child) {
3701             error_setg(errp, "Node '%s' does not have child '%s'",
3702                        parent, child);
3703             return;
3704         }
3705         bdrv_del_child(parent_bs, p_child, errp);
3706     }
3707 
3708     if (has_node) {
3709         new_bs = bdrv_find_node(node);
3710         if (!new_bs) {
3711             error_setg(errp, "Node '%s' not found", node);
3712             return;
3713         }
3714         bdrv_add_child(parent_bs, new_bs, errp);
3715     }
3716 }
3717 
3718 BlockJobInfoList *qmp_query_block_jobs(Error **errp)
3719 {
3720     BlockJobInfoList *head = NULL, **tail = &head;
3721     BlockJob *job;
3722 
3723     for (job = block_job_next(NULL); job; job = block_job_next(job)) {
3724         BlockJobInfo *value;
3725         AioContext *aio_context;
3726 
3727         if (block_job_is_internal(job)) {
3728             continue;
3729         }
3730         aio_context = block_job_get_aio_context(job);
3731         aio_context_acquire(aio_context);
3732         value = block_job_query(job, errp);
3733         aio_context_release(aio_context);
3734         if (!value) {
3735             qapi_free_BlockJobInfoList(head);
3736             return NULL;
3737         }
3738         QAPI_LIST_APPEND(tail, value);
3739     }
3740 
3741     return head;
3742 }
3743 
3744 void qmp_x_blockdev_set_iothread(const char *node_name, StrOrNull *iothread,
3745                                  bool has_force, bool force, Error **errp)
3746 {
3747     AioContext *old_context;
3748     AioContext *new_context;
3749     BlockDriverState *bs;
3750 
3751     bs = bdrv_find_node(node_name);
3752     if (!bs) {
3753         error_setg(errp, "Failed to find node with node-name='%s'", node_name);
3754         return;
3755     }
3756 
3757     /* Protects against accidents. */
3758     if (!(has_force && force) && bdrv_has_blk(bs)) {
3759         error_setg(errp, "Node %s is associated with a BlockBackend and could "
3760                          "be in use (use force=true to override this check)",
3761                          node_name);
3762         return;
3763     }
3764 
3765     if (iothread->type == QTYPE_QSTRING) {
3766         IOThread *obj = iothread_by_id(iothread->u.s);
3767         if (!obj) {
3768             error_setg(errp, "Cannot find iothread %s", iothread->u.s);
3769             return;
3770         }
3771 
3772         new_context = iothread_get_aio_context(obj);
3773     } else {
3774         new_context = qemu_get_aio_context();
3775     }
3776 
3777     old_context = bdrv_get_aio_context(bs);
3778     aio_context_acquire(old_context);
3779 
3780     bdrv_try_set_aio_context(bs, new_context, errp);
3781 
3782     aio_context_release(old_context);
3783 }
3784 
3785 QemuOptsList qemu_common_drive_opts = {
3786     .name = "drive",
3787     .head = QTAILQ_HEAD_INITIALIZER(qemu_common_drive_opts.head),
3788     .desc = {
3789         {
3790             .name = "snapshot",
3791             .type = QEMU_OPT_BOOL,
3792             .help = "enable/disable snapshot mode",
3793         },{
3794             .name = "aio",
3795             .type = QEMU_OPT_STRING,
3796             .help = "host AIO implementation (threads, native, io_uring)",
3797         },{
3798             .name = BDRV_OPT_CACHE_WB,
3799             .type = QEMU_OPT_BOOL,
3800             .help = "Enable writeback mode",
3801         },{
3802             .name = "format",
3803             .type = QEMU_OPT_STRING,
3804             .help = "disk format (raw, qcow2, ...)",
3805         },{
3806             .name = "rerror",
3807             .type = QEMU_OPT_STRING,
3808             .help = "read error action",
3809         },{
3810             .name = "werror",
3811             .type = QEMU_OPT_STRING,
3812             .help = "write error action",
3813         },{
3814             .name = BDRV_OPT_READ_ONLY,
3815             .type = QEMU_OPT_BOOL,
3816             .help = "open drive file as read-only",
3817         },
3818 
3819         THROTTLE_OPTS,
3820 
3821         {
3822             .name = "throttling.group",
3823             .type = QEMU_OPT_STRING,
3824             .help = "name of the block throttling group",
3825         },{
3826             .name = "copy-on-read",
3827             .type = QEMU_OPT_BOOL,
3828             .help = "copy read data from backing file into image file",
3829         },{
3830             .name = "detect-zeroes",
3831             .type = QEMU_OPT_STRING,
3832             .help = "try to optimize zero writes (off, on, unmap)",
3833         },{
3834             .name = "stats-account-invalid",
3835             .type = QEMU_OPT_BOOL,
3836             .help = "whether to account for invalid I/O operations "
3837                     "in the statistics",
3838         },{
3839             .name = "stats-account-failed",
3840             .type = QEMU_OPT_BOOL,
3841             .help = "whether to account for failed I/O operations "
3842                     "in the statistics",
3843         },
3844         { /* end of list */ }
3845     },
3846 };
3847 
3848 QemuOptsList qemu_drive_opts = {
3849     .name = "drive",
3850     .head = QTAILQ_HEAD_INITIALIZER(qemu_drive_opts.head),
3851     .desc = {
3852         /*
3853          * no elements => accept any params
3854          * validation will happen later
3855          */
3856         { /* end of list */ }
3857     },
3858 };
3859