xref: /openbmc/qemu/blockdev.c (revision 971febb8)
1 /*
2  * QEMU host block devices
3  *
4  * Copyright (c) 2003-2008 Fabrice Bellard
5  *
6  * This work is licensed under the terms of the GNU GPL, version 2 or
7  * later.  See the COPYING file in the top-level directory.
8  *
9  * This file incorporates work covered by the following copyright and
10  * permission notice:
11  *
12  * Copyright (c) 2003-2008 Fabrice Bellard
13  *
14  * Permission is hereby granted, free of charge, to any person obtaining a copy
15  * of this software and associated documentation files (the "Software"), to deal
16  * in the Software without restriction, including without limitation the rights
17  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
18  * copies of the Software, and to permit persons to whom the Software is
19  * furnished to do so, subject to the following conditions:
20  *
21  * The above copyright notice and this permission notice shall be included in
22  * all copies or substantial portions of the Software.
23  *
24  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
25  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
26  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
27  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
28  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
29  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
30  * THE SOFTWARE.
31  */
32 
33 #include "qemu/osdep.h"
34 #include "sysemu/block-backend.h"
35 #include "sysemu/blockdev.h"
36 #include "hw/block/block.h"
37 #include "block/blockjob.h"
38 #include "block/dirty-bitmap.h"
39 #include "block/qdict.h"
40 #include "block/throttle-groups.h"
41 #include "monitor/monitor.h"
42 #include "qemu/error-report.h"
43 #include "qemu/option.h"
44 #include "qemu/qemu-print.h"
45 #include "qemu/config-file.h"
46 #include "qapi/qapi-commands-block.h"
47 #include "qapi/qapi-commands-transaction.h"
48 #include "qapi/qapi-visit-block-core.h"
49 #include "qapi/qmp/qdict.h"
50 #include "qapi/qmp/qnum.h"
51 #include "qapi/qmp/qstring.h"
52 #include "qapi/error.h"
53 #include "qapi/qmp/qerror.h"
54 #include "qapi/qmp/qlist.h"
55 #include "qapi/qobject-output-visitor.h"
56 #include "sysemu/sysemu.h"
57 #include "sysemu/iothread.h"
58 #include "block/block_int.h"
59 #include "block/trace.h"
60 #include "sysemu/runstate.h"
61 #include "sysemu/replay.h"
62 #include "qemu/cutils.h"
63 #include "qemu/help_option.h"
64 #include "qemu/main-loop.h"
65 #include "qemu/throttle-options.h"
66 
67 /* Protected by BQL */
68 QTAILQ_HEAD(, BlockDriverState) monitor_bdrv_states =
69     QTAILQ_HEAD_INITIALIZER(monitor_bdrv_states);
70 
71 void bdrv_set_monitor_owned(BlockDriverState *bs)
72 {
73     GLOBAL_STATE_CODE();
74     QTAILQ_INSERT_TAIL(&monitor_bdrv_states, bs, monitor_list);
75 }
76 
77 static const char *const if_name[IF_COUNT] = {
78     [IF_NONE] = "none",
79     [IF_IDE] = "ide",
80     [IF_SCSI] = "scsi",
81     [IF_FLOPPY] = "floppy",
82     [IF_PFLASH] = "pflash",
83     [IF_MTD] = "mtd",
84     [IF_SD] = "sd",
85     [IF_VIRTIO] = "virtio",
86     [IF_XEN] = "xen",
87 };
88 
89 static int if_max_devs[IF_COUNT] = {
90     /*
91      * Do not change these numbers!  They govern how drive option
92      * index maps to unit and bus.  That mapping is ABI.
93      *
94      * All controllers used to implement if=T drives need to support
95      * if_max_devs[T] units, for any T with if_max_devs[T] != 0.
96      * Otherwise, some index values map to "impossible" bus, unit
97      * values.
98      *
99      * For instance, if you change [IF_SCSI] to 255, -drive
100      * if=scsi,index=12 no longer means bus=1,unit=5, but
101      * bus=0,unit=12.  With an lsi53c895a controller (7 units max),
102      * the drive can't be set up.  Regression.
103      */
104     [IF_IDE] = 2,
105     [IF_SCSI] = 7,
106 };
107 
108 /**
109  * Boards may call this to offer board-by-board overrides
110  * of the default, global values.
111  */
112 void override_max_devs(BlockInterfaceType type, int max_devs)
113 {
114     BlockBackend *blk;
115     DriveInfo *dinfo;
116 
117     GLOBAL_STATE_CODE();
118 
119     if (max_devs <= 0) {
120         return;
121     }
122 
123     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
124         dinfo = blk_legacy_dinfo(blk);
125         if (dinfo->type == type) {
126             fprintf(stderr, "Cannot override units-per-bus property of"
127                     " the %s interface, because a drive of that type has"
128                     " already been added.\n", if_name[type]);
129             g_assert_not_reached();
130         }
131     }
132 
133     if_max_devs[type] = max_devs;
134 }
135 
136 /*
137  * We automatically delete the drive when a device using it gets
138  * unplugged.  Questionable feature, but we can't just drop it.
139  * Device models call blockdev_mark_auto_del() to schedule the
140  * automatic deletion, and generic qdev code calls blockdev_auto_del()
141  * when deletion is actually safe.
142  */
143 void blockdev_mark_auto_del(BlockBackend *blk)
144 {
145     DriveInfo *dinfo = blk_legacy_dinfo(blk);
146     BlockJob *job;
147 
148     GLOBAL_STATE_CODE();
149 
150     if (!dinfo) {
151         return;
152     }
153 
154     JOB_LOCK_GUARD();
155 
156     do {
157         job = block_job_next_locked(NULL);
158         while (job && (job->job.cancelled ||
159                        job->job.deferred_to_main_loop ||
160                        !block_job_has_bdrv(job, blk_bs(blk))))
161         {
162             job = block_job_next_locked(job);
163         }
164         if (job) {
165             /*
166              * This drops the job lock temporarily and polls, so we need to
167              * restart processing the list from the start after this.
168              */
169             job_cancel_locked(&job->job, false);
170         }
171     } while (job);
172 
173     dinfo->auto_del = 1;
174 }
175 
176 void blockdev_auto_del(BlockBackend *blk)
177 {
178     DriveInfo *dinfo = blk_legacy_dinfo(blk);
179     GLOBAL_STATE_CODE();
180 
181     if (dinfo && dinfo->auto_del) {
182         monitor_remove_blk(blk);
183         blk_unref(blk);
184     }
185 }
186 
187 static int drive_index_to_bus_id(BlockInterfaceType type, int index)
188 {
189     int max_devs = if_max_devs[type];
190     return max_devs ? index / max_devs : 0;
191 }
192 
193 static int drive_index_to_unit_id(BlockInterfaceType type, int index)
194 {
195     int max_devs = if_max_devs[type];
196     return max_devs ? index % max_devs : index;
197 }
198 
199 QemuOpts *drive_add(BlockInterfaceType type, int index, const char *file,
200                     const char *optstr)
201 {
202     QemuOpts *opts;
203 
204     GLOBAL_STATE_CODE();
205 
206     opts = qemu_opts_parse_noisily(qemu_find_opts("drive"), optstr, false);
207     if (!opts) {
208         return NULL;
209     }
210     if (type != IF_DEFAULT) {
211         qemu_opt_set(opts, "if", if_name[type], &error_abort);
212     }
213     if (index >= 0) {
214         qemu_opt_set_number(opts, "index", index, &error_abort);
215     }
216     if (file)
217         qemu_opt_set(opts, "file", file, &error_abort);
218     return opts;
219 }
220 
221 DriveInfo *drive_get(BlockInterfaceType type, int bus, int unit)
222 {
223     BlockBackend *blk;
224     DriveInfo *dinfo;
225 
226     GLOBAL_STATE_CODE();
227 
228     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
229         dinfo = blk_legacy_dinfo(blk);
230         if (dinfo && dinfo->type == type
231             && dinfo->bus == bus && dinfo->unit == unit) {
232             return dinfo;
233         }
234     }
235 
236     return NULL;
237 }
238 
239 /*
240  * Check board claimed all -drive that are meant to be claimed.
241  * Fatal error if any remain unclaimed.
242  */
243 void drive_check_orphaned(void)
244 {
245     BlockBackend *blk;
246     DriveInfo *dinfo;
247     Location loc;
248     bool orphans = false;
249 
250     GLOBAL_STATE_CODE();
251 
252     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
253         dinfo = blk_legacy_dinfo(blk);
254         /*
255          * Ignore default drives, because we create certain default
256          * drives unconditionally, then leave them unclaimed.  Not the
257          * users fault.
258          * Ignore IF_VIRTIO or IF_XEN, because it gets desugared into
259          * -device, so we can leave failing to -device.
260          * Ignore IF_NONE, because leaving unclaimed IF_NONE remains
261          * available for device_add is a feature.
262          */
263         if (dinfo->is_default || dinfo->type == IF_VIRTIO
264             || dinfo->type == IF_XEN || dinfo->type == IF_NONE) {
265             continue;
266         }
267         if (!blk_get_attached_dev(blk)) {
268             loc_push_none(&loc);
269             qemu_opts_loc_restore(dinfo->opts);
270             error_report("machine type does not support"
271                          " if=%s,bus=%d,unit=%d",
272                          if_name[dinfo->type], dinfo->bus, dinfo->unit);
273             loc_pop(&loc);
274             orphans = true;
275         }
276     }
277 
278     if (orphans) {
279         exit(1);
280     }
281 }
282 
283 DriveInfo *drive_get_by_index(BlockInterfaceType type, int index)
284 {
285     GLOBAL_STATE_CODE();
286     return drive_get(type,
287                      drive_index_to_bus_id(type, index),
288                      drive_index_to_unit_id(type, index));
289 }
290 
291 int drive_get_max_bus(BlockInterfaceType type)
292 {
293     int max_bus;
294     BlockBackend *blk;
295     DriveInfo *dinfo;
296 
297     GLOBAL_STATE_CODE();
298 
299     max_bus = -1;
300     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
301         dinfo = blk_legacy_dinfo(blk);
302         if (dinfo && dinfo->type == type && dinfo->bus > max_bus) {
303             max_bus = dinfo->bus;
304         }
305     }
306     return max_bus;
307 }
308 
309 static void bdrv_format_print(void *opaque, const char *name)
310 {
311     qemu_printf(" %s", name);
312 }
313 
314 typedef struct {
315     QEMUBH *bh;
316     BlockDriverState *bs;
317 } BDRVPutRefBH;
318 
319 static int parse_block_error_action(const char *buf, bool is_read, Error **errp)
320 {
321     if (!strcmp(buf, "ignore")) {
322         return BLOCKDEV_ON_ERROR_IGNORE;
323     } else if (!is_read && !strcmp(buf, "enospc")) {
324         return BLOCKDEV_ON_ERROR_ENOSPC;
325     } else if (!strcmp(buf, "stop")) {
326         return BLOCKDEV_ON_ERROR_STOP;
327     } else if (!strcmp(buf, "report")) {
328         return BLOCKDEV_ON_ERROR_REPORT;
329     } else {
330         error_setg(errp, "'%s' invalid %s error action",
331                    buf, is_read ? "read" : "write");
332         return -1;
333     }
334 }
335 
336 static bool parse_stats_intervals(BlockAcctStats *stats, QList *intervals,
337                                   Error **errp)
338 {
339     const QListEntry *entry;
340     for (entry = qlist_first(intervals); entry; entry = qlist_next(entry)) {
341         switch (qobject_type(entry->value)) {
342 
343         case QTYPE_QSTRING: {
344             uint64_t length;
345             const char *str = qstring_get_str(qobject_to(QString,
346                                                          entry->value));
347             if (parse_uint_full(str, 10, &length) == 0 &&
348                 length > 0 && length <= UINT_MAX) {
349                 block_acct_add_interval(stats, (unsigned) length);
350             } else {
351                 error_setg(errp, "Invalid interval length: %s", str);
352                 return false;
353             }
354             break;
355         }
356 
357         case QTYPE_QNUM: {
358             int64_t length = qnum_get_int(qobject_to(QNum, entry->value));
359 
360             if (length > 0 && length <= UINT_MAX) {
361                 block_acct_add_interval(stats, (unsigned) length);
362             } else {
363                 error_setg(errp, "Invalid interval length: %" PRId64, length);
364                 return false;
365             }
366             break;
367         }
368 
369         default:
370             error_setg(errp, "The specification of stats-intervals is invalid");
371             return false;
372         }
373     }
374     return true;
375 }
376 
377 typedef enum { MEDIA_DISK, MEDIA_CDROM } DriveMediaType;
378 
379 /* All parameters but @opts are optional and may be set to NULL. */
380 static void extract_common_blockdev_options(QemuOpts *opts, int *bdrv_flags,
381     const char **throttling_group, ThrottleConfig *throttle_cfg,
382     BlockdevDetectZeroesOptions *detect_zeroes, Error **errp)
383 {
384     Error *local_error = NULL;
385     const char *aio;
386 
387     if (bdrv_flags) {
388         if (qemu_opt_get_bool(opts, "copy-on-read", false)) {
389             *bdrv_flags |= BDRV_O_COPY_ON_READ;
390         }
391 
392         if ((aio = qemu_opt_get(opts, "aio")) != NULL) {
393             if (bdrv_parse_aio(aio, bdrv_flags) < 0) {
394                 error_setg(errp, "invalid aio option");
395                 return;
396             }
397         }
398     }
399 
400     /* disk I/O throttling */
401     if (throttling_group) {
402         *throttling_group = qemu_opt_get(opts, "throttling.group");
403     }
404 
405     if (throttle_cfg) {
406         throttle_config_init(throttle_cfg);
407         throttle_cfg->buckets[THROTTLE_BPS_TOTAL].avg =
408             qemu_opt_get_number(opts, "throttling.bps-total", 0);
409         throttle_cfg->buckets[THROTTLE_BPS_READ].avg  =
410             qemu_opt_get_number(opts, "throttling.bps-read", 0);
411         throttle_cfg->buckets[THROTTLE_BPS_WRITE].avg =
412             qemu_opt_get_number(opts, "throttling.bps-write", 0);
413         throttle_cfg->buckets[THROTTLE_OPS_TOTAL].avg =
414             qemu_opt_get_number(opts, "throttling.iops-total", 0);
415         throttle_cfg->buckets[THROTTLE_OPS_READ].avg =
416             qemu_opt_get_number(opts, "throttling.iops-read", 0);
417         throttle_cfg->buckets[THROTTLE_OPS_WRITE].avg =
418             qemu_opt_get_number(opts, "throttling.iops-write", 0);
419 
420         throttle_cfg->buckets[THROTTLE_BPS_TOTAL].max =
421             qemu_opt_get_number(opts, "throttling.bps-total-max", 0);
422         throttle_cfg->buckets[THROTTLE_BPS_READ].max  =
423             qemu_opt_get_number(opts, "throttling.bps-read-max", 0);
424         throttle_cfg->buckets[THROTTLE_BPS_WRITE].max =
425             qemu_opt_get_number(opts, "throttling.bps-write-max", 0);
426         throttle_cfg->buckets[THROTTLE_OPS_TOTAL].max =
427             qemu_opt_get_number(opts, "throttling.iops-total-max", 0);
428         throttle_cfg->buckets[THROTTLE_OPS_READ].max =
429             qemu_opt_get_number(opts, "throttling.iops-read-max", 0);
430         throttle_cfg->buckets[THROTTLE_OPS_WRITE].max =
431             qemu_opt_get_number(opts, "throttling.iops-write-max", 0);
432 
433         throttle_cfg->buckets[THROTTLE_BPS_TOTAL].burst_length =
434             qemu_opt_get_number(opts, "throttling.bps-total-max-length", 1);
435         throttle_cfg->buckets[THROTTLE_BPS_READ].burst_length  =
436             qemu_opt_get_number(opts, "throttling.bps-read-max-length", 1);
437         throttle_cfg->buckets[THROTTLE_BPS_WRITE].burst_length =
438             qemu_opt_get_number(opts, "throttling.bps-write-max-length", 1);
439         throttle_cfg->buckets[THROTTLE_OPS_TOTAL].burst_length =
440             qemu_opt_get_number(opts, "throttling.iops-total-max-length", 1);
441         throttle_cfg->buckets[THROTTLE_OPS_READ].burst_length =
442             qemu_opt_get_number(opts, "throttling.iops-read-max-length", 1);
443         throttle_cfg->buckets[THROTTLE_OPS_WRITE].burst_length =
444             qemu_opt_get_number(opts, "throttling.iops-write-max-length", 1);
445 
446         throttle_cfg->op_size =
447             qemu_opt_get_number(opts, "throttling.iops-size", 0);
448 
449         if (!throttle_is_valid(throttle_cfg, errp)) {
450             return;
451         }
452     }
453 
454     if (detect_zeroes) {
455         *detect_zeroes =
456             qapi_enum_parse(&BlockdevDetectZeroesOptions_lookup,
457                             qemu_opt_get(opts, "detect-zeroes"),
458                             BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF,
459                             &local_error);
460         if (local_error) {
461             error_propagate(errp, local_error);
462             return;
463         }
464     }
465 }
466 
467 static OnOffAuto account_get_opt(QemuOpts *opts, const char *name)
468 {
469     if (!qemu_opt_find(opts, name)) {
470         return ON_OFF_AUTO_AUTO;
471     }
472     if (qemu_opt_get_bool(opts, name, true)) {
473         return ON_OFF_AUTO_ON;
474     }
475     return ON_OFF_AUTO_OFF;
476 }
477 
478 /* Takes the ownership of bs_opts */
479 static BlockBackend *blockdev_init(const char *file, QDict *bs_opts,
480                                    Error **errp)
481 {
482     const char *buf;
483     int bdrv_flags = 0;
484     int on_read_error, on_write_error;
485     OnOffAuto account_invalid, account_failed;
486     bool writethrough, read_only;
487     BlockBackend *blk;
488     BlockDriverState *bs;
489     ThrottleConfig cfg;
490     int snapshot = 0;
491     Error *error = NULL;
492     QemuOpts *opts;
493     QDict *interval_dict = NULL;
494     QList *interval_list = NULL;
495     const char *id;
496     BlockdevDetectZeroesOptions detect_zeroes =
497         BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF;
498     const char *throttling_group = NULL;
499 
500     /* Check common options by copying from bs_opts to opts, all other options
501      * stay in bs_opts for processing by bdrv_open(). */
502     id = qdict_get_try_str(bs_opts, "id");
503     opts = qemu_opts_create(&qemu_common_drive_opts, id, 1, errp);
504     if (!opts) {
505         goto err_no_opts;
506     }
507 
508     if (!qemu_opts_absorb_qdict(opts, bs_opts, errp)) {
509         goto early_err;
510     }
511 
512     if (id) {
513         qdict_del(bs_opts, "id");
514     }
515 
516     /* extract parameters */
517     snapshot = qemu_opt_get_bool(opts, "snapshot", 0);
518 
519     account_invalid = account_get_opt(opts, "stats-account-invalid");
520     account_failed = account_get_opt(opts, "stats-account-failed");
521 
522     writethrough = !qemu_opt_get_bool(opts, BDRV_OPT_CACHE_WB, true);
523 
524     id = qemu_opts_id(opts);
525 
526     qdict_extract_subqdict(bs_opts, &interval_dict, "stats-intervals.");
527     qdict_array_split(interval_dict, &interval_list);
528 
529     if (qdict_size(interval_dict) != 0) {
530         error_setg(errp, "Invalid option stats-intervals.%s",
531                    qdict_first(interval_dict)->key);
532         goto early_err;
533     }
534 
535     extract_common_blockdev_options(opts, &bdrv_flags, &throttling_group, &cfg,
536                                     &detect_zeroes, &error);
537     if (error) {
538         error_propagate(errp, error);
539         goto early_err;
540     }
541 
542     if ((buf = qemu_opt_get(opts, "format")) != NULL) {
543         if (is_help_option(buf)) {
544             qemu_printf("Supported formats:");
545             bdrv_iterate_format(bdrv_format_print, NULL, false);
546             qemu_printf("\nSupported formats (read-only):");
547             bdrv_iterate_format(bdrv_format_print, NULL, true);
548             qemu_printf("\n");
549             goto early_err;
550         }
551 
552         if (qdict_haskey(bs_opts, "driver")) {
553             error_setg(errp, "Cannot specify both 'driver' and 'format'");
554             goto early_err;
555         }
556         qdict_put_str(bs_opts, "driver", buf);
557     }
558 
559     on_write_error = BLOCKDEV_ON_ERROR_ENOSPC;
560     if ((buf = qemu_opt_get(opts, "werror")) != NULL) {
561         on_write_error = parse_block_error_action(buf, 0, &error);
562         if (error) {
563             error_propagate(errp, error);
564             goto early_err;
565         }
566     }
567 
568     on_read_error = BLOCKDEV_ON_ERROR_REPORT;
569     if ((buf = qemu_opt_get(opts, "rerror")) != NULL) {
570         on_read_error = parse_block_error_action(buf, 1, &error);
571         if (error) {
572             error_propagate(errp, error);
573             goto early_err;
574         }
575     }
576 
577     if (snapshot) {
578         bdrv_flags |= BDRV_O_SNAPSHOT;
579     }
580 
581     read_only = qemu_opt_get_bool(opts, BDRV_OPT_READ_ONLY, false);
582 
583     /* init */
584     if ((!file || !*file) && !qdict_size(bs_opts)) {
585         BlockBackendRootState *blk_rs;
586 
587         blk = blk_new(qemu_get_aio_context(), 0, BLK_PERM_ALL);
588         blk_rs = blk_get_root_state(blk);
589         blk_rs->open_flags    = bdrv_flags | (read_only ? 0 : BDRV_O_RDWR);
590         blk_rs->detect_zeroes = detect_zeroes;
591 
592         qobject_unref(bs_opts);
593     } else {
594         if (file && !*file) {
595             file = NULL;
596         }
597 
598         /* bdrv_open() defaults to the values in bdrv_flags (for compatibility
599          * with other callers) rather than what we want as the real defaults.
600          * Apply the defaults here instead. */
601         qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_DIRECT, "off");
602         qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_NO_FLUSH, "off");
603         qdict_set_default_str(bs_opts, BDRV_OPT_READ_ONLY,
604                               read_only ? "on" : "off");
605         qdict_set_default_str(bs_opts, BDRV_OPT_AUTO_READ_ONLY, "on");
606         assert((bdrv_flags & BDRV_O_CACHE_MASK) == 0);
607 
608         if (runstate_check(RUN_STATE_INMIGRATE)) {
609             bdrv_flags |= BDRV_O_INACTIVE;
610         }
611 
612         blk = blk_new_open(file, NULL, bs_opts, bdrv_flags, errp);
613         if (!blk) {
614             goto err_no_bs_opts;
615         }
616         bs = blk_bs(blk);
617 
618         bs->detect_zeroes = detect_zeroes;
619 
620         block_acct_setup(blk_get_stats(blk), account_invalid, account_failed);
621 
622         if (!parse_stats_intervals(blk_get_stats(blk), interval_list, errp)) {
623             blk_unref(blk);
624             blk = NULL;
625             goto err_no_bs_opts;
626         }
627     }
628 
629     /* disk I/O throttling */
630     if (throttle_enabled(&cfg)) {
631         if (!throttling_group) {
632             throttling_group = id;
633         }
634         blk_io_limits_enable(blk, throttling_group);
635         blk_set_io_limits(blk, &cfg);
636     }
637 
638     blk_set_enable_write_cache(blk, !writethrough);
639     blk_set_on_error(blk, on_read_error, on_write_error);
640 
641     if (!monitor_add_blk(blk, id, errp)) {
642         blk_unref(blk);
643         blk = NULL;
644         goto err_no_bs_opts;
645     }
646 
647 err_no_bs_opts:
648     qemu_opts_del(opts);
649     qobject_unref(interval_dict);
650     qobject_unref(interval_list);
651     return blk;
652 
653 early_err:
654     qemu_opts_del(opts);
655     qobject_unref(interval_dict);
656     qobject_unref(interval_list);
657 err_no_opts:
658     qobject_unref(bs_opts);
659     return NULL;
660 }
661 
662 /* Takes the ownership of bs_opts */
663 BlockDriverState *bds_tree_init(QDict *bs_opts, Error **errp)
664 {
665     int bdrv_flags = 0;
666 
667     GLOBAL_STATE_CODE();
668     /* bdrv_open() defaults to the values in bdrv_flags (for compatibility
669      * with other callers) rather than what we want as the real defaults.
670      * Apply the defaults here instead. */
671     qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_DIRECT, "off");
672     qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_NO_FLUSH, "off");
673     qdict_set_default_str(bs_opts, BDRV_OPT_READ_ONLY, "off");
674 
675     if (runstate_check(RUN_STATE_INMIGRATE)) {
676         bdrv_flags |= BDRV_O_INACTIVE;
677     }
678 
679     return bdrv_open(NULL, NULL, bs_opts, bdrv_flags, errp);
680 }
681 
682 void blockdev_close_all_bdrv_states(void)
683 {
684     BlockDriverState *bs, *next_bs;
685 
686     GLOBAL_STATE_CODE();
687     QTAILQ_FOREACH_SAFE(bs, &monitor_bdrv_states, monitor_list, next_bs) {
688         bdrv_unref(bs);
689     }
690 }
691 
692 /* Iterates over the list of monitor-owned BlockDriverStates */
693 BlockDriverState *bdrv_next_monitor_owned(BlockDriverState *bs)
694 {
695     GLOBAL_STATE_CODE();
696     return bs ? QTAILQ_NEXT(bs, monitor_list)
697               : QTAILQ_FIRST(&monitor_bdrv_states);
698 }
699 
700 static bool qemu_opt_rename(QemuOpts *opts, const char *from, const char *to,
701                             Error **errp)
702 {
703     const char *value;
704 
705     value = qemu_opt_get(opts, from);
706     if (value) {
707         if (qemu_opt_find(opts, to)) {
708             error_setg(errp, "'%s' and its alias '%s' can't be used at the "
709                        "same time", to, from);
710             return false;
711         }
712     }
713 
714     /* rename all items in opts */
715     while ((value = qemu_opt_get(opts, from))) {
716         qemu_opt_set(opts, to, value, &error_abort);
717         qemu_opt_unset(opts, from);
718     }
719     return true;
720 }
721 
722 QemuOptsList qemu_legacy_drive_opts = {
723     .name = "drive",
724     .head = QTAILQ_HEAD_INITIALIZER(qemu_legacy_drive_opts.head),
725     .desc = {
726         {
727             .name = "bus",
728             .type = QEMU_OPT_NUMBER,
729             .help = "bus number",
730         },{
731             .name = "unit",
732             .type = QEMU_OPT_NUMBER,
733             .help = "unit number (i.e. lun for scsi)",
734         },{
735             .name = "index",
736             .type = QEMU_OPT_NUMBER,
737             .help = "index number",
738         },{
739             .name = "media",
740             .type = QEMU_OPT_STRING,
741             .help = "media type (disk, cdrom)",
742         },{
743             .name = "if",
744             .type = QEMU_OPT_STRING,
745             .help = "interface (ide, scsi, sd, mtd, floppy, pflash, virtio)",
746         },{
747             .name = "file",
748             .type = QEMU_OPT_STRING,
749             .help = "file name",
750         },
751 
752         /* Options that are passed on, but have special semantics with -drive */
753         {
754             .name = BDRV_OPT_READ_ONLY,
755             .type = QEMU_OPT_BOOL,
756             .help = "open drive file as read-only",
757         },{
758             .name = "rerror",
759             .type = QEMU_OPT_STRING,
760             .help = "read error action",
761         },{
762             .name = "werror",
763             .type = QEMU_OPT_STRING,
764             .help = "write error action",
765         },{
766             .name = "copy-on-read",
767             .type = QEMU_OPT_BOOL,
768             .help = "copy read data from backing file into image file",
769         },
770 
771         { /* end of list */ }
772     },
773 };
774 
775 DriveInfo *drive_new(QemuOpts *all_opts, BlockInterfaceType block_default_type,
776                      Error **errp)
777 {
778     const char *value;
779     BlockBackend *blk;
780     DriveInfo *dinfo = NULL;
781     QDict *bs_opts;
782     QemuOpts *legacy_opts;
783     DriveMediaType media = MEDIA_DISK;
784     BlockInterfaceType type;
785     int max_devs, bus_id, unit_id, index;
786     const char *werror, *rerror;
787     bool read_only = false;
788     bool copy_on_read;
789     const char *filename;
790     int i;
791 
792     GLOBAL_STATE_CODE();
793 
794     /* Change legacy command line options into QMP ones */
795     static const struct {
796         const char *from;
797         const char *to;
798     } opt_renames[] = {
799         { "iops",           "throttling.iops-total" },
800         { "iops_rd",        "throttling.iops-read" },
801         { "iops_wr",        "throttling.iops-write" },
802 
803         { "bps",            "throttling.bps-total" },
804         { "bps_rd",         "throttling.bps-read" },
805         { "bps_wr",         "throttling.bps-write" },
806 
807         { "iops_max",       "throttling.iops-total-max" },
808         { "iops_rd_max",    "throttling.iops-read-max" },
809         { "iops_wr_max",    "throttling.iops-write-max" },
810 
811         { "bps_max",        "throttling.bps-total-max" },
812         { "bps_rd_max",     "throttling.bps-read-max" },
813         { "bps_wr_max",     "throttling.bps-write-max" },
814 
815         { "iops_size",      "throttling.iops-size" },
816 
817         { "group",          "throttling.group" },
818 
819         { "readonly",       BDRV_OPT_READ_ONLY },
820     };
821 
822     for (i = 0; i < ARRAY_SIZE(opt_renames); i++) {
823         if (!qemu_opt_rename(all_opts, opt_renames[i].from,
824                              opt_renames[i].to, errp)) {
825             return NULL;
826         }
827     }
828 
829     value = qemu_opt_get(all_opts, "cache");
830     if (value) {
831         int flags = 0;
832         bool writethrough;
833 
834         if (bdrv_parse_cache_mode(value, &flags, &writethrough) != 0) {
835             error_setg(errp, "invalid cache option");
836             return NULL;
837         }
838 
839         /* Specific options take precedence */
840         if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_WB)) {
841             qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_WB,
842                               !writethrough, &error_abort);
843         }
844         if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_DIRECT)) {
845             qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_DIRECT,
846                               !!(flags & BDRV_O_NOCACHE), &error_abort);
847         }
848         if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_NO_FLUSH)) {
849             qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_NO_FLUSH,
850                               !!(flags & BDRV_O_NO_FLUSH), &error_abort);
851         }
852         qemu_opt_unset(all_opts, "cache");
853     }
854 
855     /* Get a QDict for processing the options */
856     bs_opts = qdict_new();
857     qemu_opts_to_qdict(all_opts, bs_opts);
858 
859     legacy_opts = qemu_opts_create(&qemu_legacy_drive_opts, NULL, 0,
860                                    &error_abort);
861     if (!qemu_opts_absorb_qdict(legacy_opts, bs_opts, errp)) {
862         goto fail;
863     }
864 
865     /* Media type */
866     value = qemu_opt_get(legacy_opts, "media");
867     if (value) {
868         if (!strcmp(value, "disk")) {
869             media = MEDIA_DISK;
870         } else if (!strcmp(value, "cdrom")) {
871             media = MEDIA_CDROM;
872             read_only = true;
873         } else {
874             error_setg(errp, "'%s' invalid media", value);
875             goto fail;
876         }
877     }
878 
879     /* copy-on-read is disabled with a warning for read-only devices */
880     read_only |= qemu_opt_get_bool(legacy_opts, BDRV_OPT_READ_ONLY, false);
881     copy_on_read = qemu_opt_get_bool(legacy_opts, "copy-on-read", false);
882 
883     if (read_only && copy_on_read) {
884         warn_report("disabling copy-on-read on read-only drive");
885         copy_on_read = false;
886     }
887 
888     qdict_put_str(bs_opts, BDRV_OPT_READ_ONLY, read_only ? "on" : "off");
889     qdict_put_str(bs_opts, "copy-on-read", copy_on_read ? "on" : "off");
890 
891     /* Controller type */
892     value = qemu_opt_get(legacy_opts, "if");
893     if (value) {
894         for (type = 0;
895              type < IF_COUNT && strcmp(value, if_name[type]);
896              type++) {
897         }
898         if (type == IF_COUNT) {
899             error_setg(errp, "unsupported bus type '%s'", value);
900             goto fail;
901         }
902     } else {
903         type = block_default_type;
904     }
905 
906     /* Device address specified by bus/unit or index.
907      * If none was specified, try to find the first free one. */
908     bus_id  = qemu_opt_get_number(legacy_opts, "bus", 0);
909     unit_id = qemu_opt_get_number(legacy_opts, "unit", -1);
910     index   = qemu_opt_get_number(legacy_opts, "index", -1);
911 
912     max_devs = if_max_devs[type];
913 
914     if (index != -1) {
915         if (bus_id != 0 || unit_id != -1) {
916             error_setg(errp, "index cannot be used with bus and unit");
917             goto fail;
918         }
919         bus_id = drive_index_to_bus_id(type, index);
920         unit_id = drive_index_to_unit_id(type, index);
921     }
922 
923     if (unit_id == -1) {
924        unit_id = 0;
925        while (drive_get(type, bus_id, unit_id) != NULL) {
926            unit_id++;
927            if (max_devs && unit_id >= max_devs) {
928                unit_id -= max_devs;
929                bus_id++;
930            }
931        }
932     }
933 
934     if (max_devs && unit_id >= max_devs) {
935         error_setg(errp, "unit %d too big (max is %d)", unit_id, max_devs - 1);
936         goto fail;
937     }
938 
939     if (drive_get(type, bus_id, unit_id) != NULL) {
940         error_setg(errp, "drive with bus=%d, unit=%d (index=%d) exists",
941                    bus_id, unit_id, index);
942         goto fail;
943     }
944 
945     /* no id supplied -> create one */
946     if (qemu_opts_id(all_opts) == NULL) {
947         char *new_id;
948         const char *mediastr = "";
949         if (type == IF_IDE || type == IF_SCSI) {
950             mediastr = (media == MEDIA_CDROM) ? "-cd" : "-hd";
951         }
952         if (max_devs) {
953             new_id = g_strdup_printf("%s%i%s%i", if_name[type], bus_id,
954                                      mediastr, unit_id);
955         } else {
956             new_id = g_strdup_printf("%s%s%i", if_name[type],
957                                      mediastr, unit_id);
958         }
959         qdict_put_str(bs_opts, "id", new_id);
960         g_free(new_id);
961     }
962 
963     /* Add virtio block device */
964     if (type == IF_VIRTIO) {
965         QemuOpts *devopts;
966         devopts = qemu_opts_create(qemu_find_opts("device"), NULL, 0,
967                                    &error_abort);
968         qemu_opt_set(devopts, "driver", "virtio-blk", &error_abort);
969         qemu_opt_set(devopts, "drive", qdict_get_str(bs_opts, "id"),
970                      &error_abort);
971     } else if (type == IF_XEN) {
972         QemuOpts *devopts;
973         devopts = qemu_opts_create(qemu_find_opts("device"), NULL, 0,
974                                    &error_abort);
975         qemu_opt_set(devopts, "driver",
976                      (media == MEDIA_CDROM) ? "xen-cdrom" : "xen-disk",
977                      &error_abort);
978         qemu_opt_set(devopts, "drive", qdict_get_str(bs_opts, "id"),
979                      &error_abort);
980     }
981 
982     filename = qemu_opt_get(legacy_opts, "file");
983 
984     /* Check werror/rerror compatibility with if=... */
985     werror = qemu_opt_get(legacy_opts, "werror");
986     if (werror != NULL) {
987         if (type != IF_IDE && type != IF_SCSI && type != IF_VIRTIO &&
988             type != IF_NONE) {
989             error_setg(errp, "werror is not supported by this bus type");
990             goto fail;
991         }
992         qdict_put_str(bs_opts, "werror", werror);
993     }
994 
995     rerror = qemu_opt_get(legacy_opts, "rerror");
996     if (rerror != NULL) {
997         if (type != IF_IDE && type != IF_VIRTIO && type != IF_SCSI &&
998             type != IF_NONE) {
999             error_setg(errp, "rerror is not supported by this bus type");
1000             goto fail;
1001         }
1002         qdict_put_str(bs_opts, "rerror", rerror);
1003     }
1004 
1005     /* Actual block device init: Functionality shared with blockdev-add */
1006     blk = blockdev_init(filename, bs_opts, errp);
1007     bs_opts = NULL;
1008     if (!blk) {
1009         goto fail;
1010     }
1011 
1012     /* Create legacy DriveInfo */
1013     dinfo = g_malloc0(sizeof(*dinfo));
1014     dinfo->opts = all_opts;
1015 
1016     dinfo->type = type;
1017     dinfo->bus = bus_id;
1018     dinfo->unit = unit_id;
1019 
1020     blk_set_legacy_dinfo(blk, dinfo);
1021 
1022     switch(type) {
1023     case IF_IDE:
1024     case IF_SCSI:
1025     case IF_XEN:
1026     case IF_NONE:
1027         dinfo->media_cd = media == MEDIA_CDROM;
1028         break;
1029     default:
1030         break;
1031     }
1032 
1033 fail:
1034     qemu_opts_del(legacy_opts);
1035     qobject_unref(bs_opts);
1036     return dinfo;
1037 }
1038 
1039 static BlockDriverState *qmp_get_root_bs(const char *name, Error **errp)
1040 {
1041     BlockDriverState *bs;
1042 
1043     GRAPH_RDLOCK_GUARD_MAINLOOP();
1044 
1045     bs = bdrv_lookup_bs(name, name, errp);
1046     if (bs == NULL) {
1047         return NULL;
1048     }
1049 
1050     if (!bdrv_is_root_node(bs)) {
1051         error_setg(errp, "Need a root block node");
1052         return NULL;
1053     }
1054 
1055     if (!bdrv_is_inserted(bs)) {
1056         error_setg(errp, "Device has no medium");
1057         bs = NULL;
1058     }
1059 
1060     return bs;
1061 }
1062 
1063 static void blockdev_do_action(TransactionAction *action, Error **errp)
1064 {
1065     TransactionActionList list;
1066 
1067     list.value = action;
1068     list.next = NULL;
1069     qmp_transaction(&list, NULL, errp);
1070 }
1071 
1072 void qmp_blockdev_snapshot_sync(const char *device, const char *node_name,
1073                                 const char *snapshot_file,
1074                                 const char *snapshot_node_name,
1075                                 const char *format,
1076                                 bool has_mode, NewImageMode mode, Error **errp)
1077 {
1078     BlockdevSnapshotSync snapshot = {
1079         .device = (char *) device,
1080         .node_name = (char *) node_name,
1081         .snapshot_file = (char *) snapshot_file,
1082         .snapshot_node_name = (char *) snapshot_node_name,
1083         .format = (char *) format,
1084         .has_mode = has_mode,
1085         .mode = mode,
1086     };
1087     TransactionAction action = {
1088         .type = TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC,
1089         .u.blockdev_snapshot_sync.data = &snapshot,
1090     };
1091     blockdev_do_action(&action, errp);
1092 }
1093 
1094 void qmp_blockdev_snapshot(const char *node, const char *overlay,
1095                            Error **errp)
1096 {
1097     BlockdevSnapshot snapshot_data = {
1098         .node = (char *) node,
1099         .overlay = (char *) overlay
1100     };
1101     TransactionAction action = {
1102         .type = TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT,
1103         .u.blockdev_snapshot.data = &snapshot_data,
1104     };
1105     blockdev_do_action(&action, errp);
1106 }
1107 
1108 void qmp_blockdev_snapshot_internal_sync(const char *device,
1109                                          const char *name,
1110                                          Error **errp)
1111 {
1112     BlockdevSnapshotInternal snapshot = {
1113         .device = (char *) device,
1114         .name = (char *) name
1115     };
1116     TransactionAction action = {
1117         .type = TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC,
1118         .u.blockdev_snapshot_internal_sync.data = &snapshot,
1119     };
1120     blockdev_do_action(&action, errp);
1121 }
1122 
1123 SnapshotInfo *qmp_blockdev_snapshot_delete_internal_sync(const char *device,
1124                                                          const char *id,
1125                                                          const char *name,
1126                                                          Error **errp)
1127 {
1128     BlockDriverState *bs;
1129     QEMUSnapshotInfo sn;
1130     Error *local_err = NULL;
1131     SnapshotInfo *info = NULL;
1132     int ret;
1133 
1134     GLOBAL_STATE_CODE();
1135     GRAPH_RDLOCK_GUARD_MAINLOOP();
1136 
1137     bs = qmp_get_root_bs(device, errp);
1138     if (!bs) {
1139         return NULL;
1140     }
1141 
1142     if (!id && !name) {
1143         error_setg(errp, "Name or id must be provided");
1144         return NULL;
1145     }
1146 
1147     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_INTERNAL_SNAPSHOT_DELETE, errp)) {
1148         return NULL;
1149     }
1150 
1151     ret = bdrv_snapshot_find_by_id_and_name(bs, id, name, &sn, &local_err);
1152     if (local_err) {
1153         error_propagate(errp, local_err);
1154         return NULL;
1155     }
1156     if (!ret) {
1157         error_setg(errp,
1158                    "Snapshot with id '%s' and name '%s' does not exist on "
1159                    "device '%s'",
1160                    STR_OR_NULL(id), STR_OR_NULL(name), device);
1161         return NULL;
1162     }
1163 
1164     bdrv_snapshot_delete(bs, id, name, &local_err);
1165     if (local_err) {
1166         error_propagate(errp, local_err);
1167         return NULL;
1168     }
1169 
1170     info = g_new0(SnapshotInfo, 1);
1171     info->id = g_strdup(sn.id_str);
1172     info->name = g_strdup(sn.name);
1173     info->date_nsec = sn.date_nsec;
1174     info->date_sec = sn.date_sec;
1175     info->vm_state_size = sn.vm_state_size;
1176     info->vm_clock_nsec = sn.vm_clock_nsec % 1000000000;
1177     info->vm_clock_sec = sn.vm_clock_nsec / 1000000000;
1178     if (sn.icount != -1ULL) {
1179         info->icount = sn.icount;
1180         info->has_icount = true;
1181     }
1182 
1183     return info;
1184 }
1185 
1186 /* internal snapshot private data */
1187 typedef struct InternalSnapshotState {
1188     BlockDriverState *bs;
1189     QEMUSnapshotInfo sn;
1190     bool created;
1191 } InternalSnapshotState;
1192 
1193 static void internal_snapshot_abort(void *opaque);
1194 static void internal_snapshot_clean(void *opaque);
1195 TransactionActionDrv internal_snapshot_drv = {
1196     .abort = internal_snapshot_abort,
1197     .clean = internal_snapshot_clean,
1198 };
1199 
1200 static void internal_snapshot_action(BlockdevSnapshotInternal *internal,
1201                                      Transaction *tran, Error **errp)
1202 {
1203     Error *local_err = NULL;
1204     const char *device;
1205     const char *name;
1206     BlockDriverState *bs;
1207     QEMUSnapshotInfo old_sn, *sn;
1208     bool ret;
1209     int64_t rt;
1210     InternalSnapshotState *state = g_new0(InternalSnapshotState, 1);
1211     int ret1;
1212 
1213     GLOBAL_STATE_CODE();
1214     GRAPH_RDLOCK_GUARD_MAINLOOP();
1215 
1216     tran_add(tran, &internal_snapshot_drv, state);
1217 
1218     device = internal->device;
1219     name = internal->name;
1220 
1221     bs = qmp_get_root_bs(device, errp);
1222     if (!bs) {
1223         return;
1224     }
1225 
1226     state->bs = bs;
1227 
1228     /* Paired with .clean() */
1229     bdrv_drained_begin(bs);
1230 
1231     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_INTERNAL_SNAPSHOT, errp)) {
1232         return;
1233     }
1234 
1235     if (bdrv_is_read_only(bs)) {
1236         error_setg(errp, "Device '%s' is read only", device);
1237         return;
1238     }
1239 
1240     if (!bdrv_can_snapshot(bs)) {
1241         error_setg(errp, "Block format '%s' used by device '%s' "
1242                    "does not support internal snapshots",
1243                    bs->drv->format_name, device);
1244         return;
1245     }
1246 
1247     if (!strlen(name)) {
1248         error_setg(errp, "Name is empty");
1249         return;
1250     }
1251 
1252     /* check whether a snapshot with name exist */
1253     ret = bdrv_snapshot_find_by_id_and_name(bs, NULL, name, &old_sn,
1254                                             &local_err);
1255     if (local_err) {
1256         error_propagate(errp, local_err);
1257         return;
1258     } else if (ret) {
1259         error_setg(errp,
1260                    "Snapshot with name '%s' already exists on device '%s'",
1261                    name, device);
1262         return;
1263     }
1264 
1265     /* 3. take the snapshot */
1266     sn = &state->sn;
1267     pstrcpy(sn->name, sizeof(sn->name), name);
1268     rt = g_get_real_time();
1269     sn->date_sec = rt / G_USEC_PER_SEC;
1270     sn->date_nsec = (rt % G_USEC_PER_SEC) * 1000;
1271     sn->vm_clock_nsec = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL);
1272     if (replay_mode != REPLAY_MODE_NONE) {
1273         sn->icount = replay_get_current_icount();
1274     } else {
1275         sn->icount = -1ULL;
1276     }
1277 
1278     ret1 = bdrv_snapshot_create(bs, sn);
1279     if (ret1 < 0) {
1280         error_setg_errno(errp, -ret1,
1281                          "Failed to create snapshot '%s' on device '%s'",
1282                          name, device);
1283         return;
1284     }
1285 
1286     /* 4. succeed, mark a snapshot is created */
1287     state->created = true;
1288 }
1289 
1290 static void internal_snapshot_abort(void *opaque)
1291 {
1292     InternalSnapshotState *state = opaque;
1293     BlockDriverState *bs = state->bs;
1294     QEMUSnapshotInfo *sn = &state->sn;
1295     Error *local_error = NULL;
1296 
1297     GLOBAL_STATE_CODE();
1298     GRAPH_RDLOCK_GUARD_MAINLOOP();
1299 
1300     if (!state->created) {
1301         return;
1302     }
1303 
1304     if (bdrv_snapshot_delete(bs, sn->id_str, sn->name, &local_error) < 0) {
1305         error_reportf_err(local_error,
1306                           "Failed to delete snapshot with id '%s' and "
1307                           "name '%s' on device '%s' in abort: ",
1308                           sn->id_str, sn->name,
1309                           bdrv_get_device_name(bs));
1310     }
1311 }
1312 
1313 static void internal_snapshot_clean(void *opaque)
1314 {
1315     g_autofree InternalSnapshotState *state = opaque;
1316 
1317     if (!state->bs) {
1318         return;
1319     }
1320 
1321     bdrv_drained_end(state->bs);
1322 }
1323 
1324 /* external snapshot private data */
1325 typedef struct ExternalSnapshotState {
1326     BlockDriverState *old_bs;
1327     BlockDriverState *new_bs;
1328     bool overlay_appended;
1329 } ExternalSnapshotState;
1330 
1331 static void external_snapshot_commit(void *opaque);
1332 static void external_snapshot_abort(void *opaque);
1333 static void external_snapshot_clean(void *opaque);
1334 TransactionActionDrv external_snapshot_drv = {
1335     .commit = external_snapshot_commit,
1336     .abort = external_snapshot_abort,
1337     .clean = external_snapshot_clean,
1338 };
1339 
1340 static void external_snapshot_action(TransactionAction *action,
1341                                      Transaction *tran, Error **errp)
1342 {
1343     int ret;
1344     int flags = 0;
1345     QDict *options = NULL;
1346     Error *local_err = NULL;
1347     /* Device and node name of the image to generate the snapshot from */
1348     const char *device;
1349     const char *node_name;
1350     /* Reference to the new image (for 'blockdev-snapshot') */
1351     const char *snapshot_ref;
1352     /* File name of the new image (for 'blockdev-snapshot-sync') */
1353     const char *new_image_file;
1354     ExternalSnapshotState *state = g_new0(ExternalSnapshotState, 1);
1355     uint64_t perm, shared;
1356 
1357     /* TODO We'll eventually have to take a writer lock in this function */
1358     GRAPH_RDLOCK_GUARD_MAINLOOP();
1359 
1360     tran_add(tran, &external_snapshot_drv, state);
1361 
1362     /* 'blockdev-snapshot' and 'blockdev-snapshot-sync' have similar
1363      * purpose but a different set of parameters */
1364     switch (action->type) {
1365     case TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT:
1366         {
1367             BlockdevSnapshot *s = action->u.blockdev_snapshot.data;
1368             device = s->node;
1369             node_name = s->node;
1370             new_image_file = NULL;
1371             snapshot_ref = s->overlay;
1372         }
1373         break;
1374     case TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC:
1375         {
1376             BlockdevSnapshotSync *s = action->u.blockdev_snapshot_sync.data;
1377             device = s->device;
1378             node_name = s->node_name;
1379             new_image_file = s->snapshot_file;
1380             snapshot_ref = NULL;
1381         }
1382         break;
1383     default:
1384         g_assert_not_reached();
1385     }
1386 
1387     /* start processing */
1388 
1389     state->old_bs = bdrv_lookup_bs(device, node_name, errp);
1390     if (!state->old_bs) {
1391         return;
1392     }
1393 
1394     /* Paired with .clean() */
1395     bdrv_drained_begin(state->old_bs);
1396 
1397     if (!bdrv_is_inserted(state->old_bs)) {
1398         error_setg(errp, QERR_DEVICE_HAS_NO_MEDIUM,
1399                    bdrv_get_device_or_node_name(state->old_bs));
1400         return;
1401     }
1402 
1403     if (bdrv_op_is_blocked(state->old_bs,
1404                            BLOCK_OP_TYPE_EXTERNAL_SNAPSHOT, errp)) {
1405         return;
1406     }
1407 
1408     if (!bdrv_is_read_only(state->old_bs)) {
1409         if (bdrv_flush(state->old_bs)) {
1410             error_setg(errp, QERR_IO_ERROR);
1411             return;
1412         }
1413     }
1414 
1415     if (action->type == TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC) {
1416         BlockdevSnapshotSync *s = action->u.blockdev_snapshot_sync.data;
1417         const char *format = s->format ?: "qcow2";
1418         enum NewImageMode mode;
1419         const char *snapshot_node_name = s->snapshot_node_name;
1420 
1421         if (node_name && !snapshot_node_name) {
1422             error_setg(errp, "New overlay node-name missing");
1423             return;
1424         }
1425 
1426         if (snapshot_node_name &&
1427             bdrv_lookup_bs(snapshot_node_name, snapshot_node_name, NULL)) {
1428             error_setg(errp, "New overlay node-name already in use");
1429             return;
1430         }
1431 
1432         flags = state->old_bs->open_flags;
1433         flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_COPY_ON_READ);
1434         flags |= BDRV_O_NO_BACKING;
1435 
1436         /* create new image w/backing file */
1437         mode = s->has_mode ? s->mode : NEW_IMAGE_MODE_ABSOLUTE_PATHS;
1438         if (mode != NEW_IMAGE_MODE_EXISTING) {
1439             int64_t size = bdrv_getlength(state->old_bs);
1440             if (size < 0) {
1441                 error_setg_errno(errp, -size, "bdrv_getlength failed");
1442                 return;
1443             }
1444             bdrv_refresh_filename(state->old_bs);
1445 
1446             bdrv_img_create(new_image_file, format,
1447                             state->old_bs->filename,
1448                             state->old_bs->drv->format_name,
1449                             NULL, size, flags, false, &local_err);
1450 
1451             if (local_err) {
1452                 error_propagate(errp, local_err);
1453                 return;
1454             }
1455         }
1456 
1457         options = qdict_new();
1458         if (snapshot_node_name) {
1459             qdict_put_str(options, "node-name", snapshot_node_name);
1460         }
1461         qdict_put_str(options, "driver", format);
1462     }
1463 
1464     state->new_bs = bdrv_open(new_image_file, snapshot_ref, options, flags,
1465                               errp);
1466 
1467     /* We will manually add the backing_hd field to the bs later */
1468     if (!state->new_bs) {
1469         return;
1470     }
1471 
1472     /*
1473      * Allow attaching a backing file to an overlay that's already in use only
1474      * if the parents don't assume that they are already seeing a valid image.
1475      * (Specifically, allow it as a mirror target, which is write-only access.)
1476      */
1477     bdrv_get_cumulative_perm(state->new_bs, &perm, &shared);
1478     if (perm & BLK_PERM_CONSISTENT_READ) {
1479         error_setg(errp, "The overlay is already in use");
1480         return;
1481     }
1482 
1483     if (state->new_bs->drv->is_filter) {
1484         error_setg(errp, "Filters cannot be used as overlays");
1485         return;
1486     }
1487 
1488     if (bdrv_cow_child(state->new_bs)) {
1489         error_setg(errp, "The overlay already has a backing image");
1490         return;
1491     }
1492 
1493     if (!state->new_bs->drv->supports_backing) {
1494         error_setg(errp, "The overlay does not support backing images");
1495         return;
1496     }
1497 
1498     ret = bdrv_append(state->new_bs, state->old_bs, errp);
1499     if (ret < 0) {
1500         return;
1501     }
1502     state->overlay_appended = true;
1503 }
1504 
1505 static void external_snapshot_commit(void *opaque)
1506 {
1507     ExternalSnapshotState *state = opaque;
1508 
1509     /* We don't need (or want) to use the transactional
1510      * bdrv_reopen_multiple() across all the entries at once, because we
1511      * don't want to abort all of them if one of them fails the reopen */
1512     if (!qatomic_read(&state->old_bs->copy_on_read)) {
1513         bdrv_reopen_set_read_only(state->old_bs, true, NULL);
1514     }
1515 }
1516 
1517 static void external_snapshot_abort(void *opaque)
1518 {
1519     ExternalSnapshotState *state = opaque;
1520     if (state->new_bs) {
1521         if (state->overlay_appended) {
1522             AioContext *aio_context;
1523             AioContext *tmp_context;
1524             int ret;
1525 
1526             aio_context = bdrv_get_aio_context(state->old_bs);
1527 
1528             bdrv_ref(state->old_bs);   /* we can't let bdrv_set_backind_hd()
1529                                           close state->old_bs; we need it */
1530             bdrv_set_backing_hd(state->new_bs, NULL, &error_abort);
1531 
1532             /*
1533              * The call to bdrv_set_backing_hd() above returns state->old_bs to
1534              * the main AioContext. As we're still going to be using it, return
1535              * it to the AioContext it was before.
1536              */
1537             tmp_context = bdrv_get_aio_context(state->old_bs);
1538             if (aio_context != tmp_context) {
1539                 ret = bdrv_try_change_aio_context(state->old_bs,
1540                                                   aio_context, NULL, NULL);
1541                 assert(ret == 0);
1542             }
1543 
1544             bdrv_drained_begin(state->new_bs);
1545             bdrv_graph_wrlock();
1546             bdrv_replace_node(state->new_bs, state->old_bs, &error_abort);
1547             bdrv_graph_wrunlock();
1548             bdrv_drained_end(state->new_bs);
1549 
1550             bdrv_unref(state->old_bs); /* bdrv_replace_node() ref'ed old_bs */
1551         }
1552     }
1553 }
1554 
1555 static void external_snapshot_clean(void *opaque)
1556 {
1557     g_autofree ExternalSnapshotState *state = opaque;
1558 
1559     if (!state->old_bs) {
1560         return;
1561     }
1562 
1563     bdrv_drained_end(state->old_bs);
1564     bdrv_unref(state->new_bs);
1565 }
1566 
1567 typedef struct DriveBackupState {
1568     BlockDriverState *bs;
1569     BlockJob *job;
1570 } DriveBackupState;
1571 
1572 static BlockJob *do_backup_common(BackupCommon *backup,
1573                                   BlockDriverState *bs,
1574                                   BlockDriverState *target_bs,
1575                                   AioContext *aio_context,
1576                                   JobTxn *txn, Error **errp);
1577 
1578 static void drive_backup_commit(void *opaque);
1579 static void drive_backup_abort(void *opaque);
1580 static void drive_backup_clean(void *opaque);
1581 TransactionActionDrv drive_backup_drv = {
1582     .commit = drive_backup_commit,
1583     .abort = drive_backup_abort,
1584     .clean = drive_backup_clean,
1585 };
1586 
1587 static void drive_backup_action(DriveBackup *backup,
1588                                 JobTxn *block_job_txn,
1589                                 Transaction *tran, Error **errp)
1590 {
1591     DriveBackupState *state = g_new0(DriveBackupState, 1);
1592     BlockDriverState *bs;
1593     BlockDriverState *target_bs;
1594     BlockDriverState *source = NULL;
1595     AioContext *aio_context;
1596     const char *format;
1597     QDict *options;
1598     Error *local_err = NULL;
1599     int flags;
1600     int64_t size;
1601     bool set_backing_hd = false;
1602     int ret;
1603 
1604     GLOBAL_STATE_CODE();
1605 
1606     tran_add(tran, &drive_backup_drv, state);
1607 
1608     if (!backup->has_mode) {
1609         backup->mode = NEW_IMAGE_MODE_ABSOLUTE_PATHS;
1610     }
1611 
1612     bs = bdrv_lookup_bs(backup->device, backup->device, errp);
1613     if (!bs) {
1614         return;
1615     }
1616 
1617     if (!bs->drv) {
1618         error_setg(errp, "Device has no medium");
1619         return;
1620     }
1621 
1622     aio_context = bdrv_get_aio_context(bs);
1623 
1624     state->bs = bs;
1625     /* Paired with .clean() */
1626     bdrv_drained_begin(bs);
1627 
1628     format = backup->format;
1629     if (!format && backup->mode != NEW_IMAGE_MODE_EXISTING) {
1630         format = bs->drv->format_name;
1631     }
1632 
1633     /* Early check to avoid creating target */
1634     bdrv_graph_rdlock_main_loop();
1635     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_BACKUP_SOURCE, errp)) {
1636         bdrv_graph_rdunlock_main_loop();
1637         return;
1638     }
1639 
1640     flags = bs->open_flags | BDRV_O_RDWR;
1641 
1642     /*
1643      * See if we have a backing HD we can use to create our new image
1644      * on top of.
1645      */
1646     if (backup->sync == MIRROR_SYNC_MODE_TOP) {
1647         /*
1648          * Backup will not replace the source by the target, so none
1649          * of the filters skipped here will be removed (in contrast to
1650          * mirror).  Therefore, we can skip all of them when looking
1651          * for the first COW relationship.
1652          */
1653         source = bdrv_cow_bs(bdrv_skip_filters(bs));
1654         if (!source) {
1655             backup->sync = MIRROR_SYNC_MODE_FULL;
1656         }
1657     }
1658     if (backup->sync == MIRROR_SYNC_MODE_NONE) {
1659         source = bs;
1660         flags |= BDRV_O_NO_BACKING;
1661         set_backing_hd = true;
1662     }
1663     bdrv_graph_rdunlock_main_loop();
1664 
1665     size = bdrv_getlength(bs);
1666     if (size < 0) {
1667         error_setg_errno(errp, -size, "bdrv_getlength failed");
1668         return;
1669     }
1670 
1671     if (backup->mode != NEW_IMAGE_MODE_EXISTING) {
1672         assert(format);
1673         if (source) {
1674             /* Implicit filters should not appear in the filename */
1675             BlockDriverState *explicit_backing;
1676 
1677             bdrv_graph_rdlock_main_loop();
1678             explicit_backing = bdrv_skip_implicit_filters(source);
1679             bdrv_refresh_filename(explicit_backing);
1680             bdrv_graph_rdunlock_main_loop();
1681 
1682             bdrv_img_create(backup->target, format,
1683                             explicit_backing->filename,
1684                             explicit_backing->drv->format_name, NULL,
1685                             size, flags, false, &local_err);
1686         } else {
1687             bdrv_img_create(backup->target, format, NULL, NULL, NULL,
1688                             size, flags, false, &local_err);
1689         }
1690     }
1691 
1692     if (local_err) {
1693         error_propagate(errp, local_err);
1694         return;
1695     }
1696 
1697     options = qdict_new();
1698     qdict_put_str(options, "discard", "unmap");
1699     qdict_put_str(options, "detect-zeroes", "unmap");
1700     if (format) {
1701         qdict_put_str(options, "driver", format);
1702     }
1703 
1704     target_bs = bdrv_open(backup->target, NULL, options, flags, errp);
1705     if (!target_bs) {
1706         return;
1707     }
1708 
1709     ret = bdrv_try_change_aio_context(target_bs, aio_context, NULL, errp);
1710     if (ret < 0) {
1711         bdrv_unref(target_bs);
1712         return;
1713     }
1714 
1715     if (set_backing_hd) {
1716         if (bdrv_set_backing_hd(target_bs, source, errp) < 0) {
1717             goto unref;
1718         }
1719     }
1720 
1721     state->job = do_backup_common(qapi_DriveBackup_base(backup),
1722                                   bs, target_bs, aio_context,
1723                                   block_job_txn, errp);
1724 
1725 unref:
1726     bdrv_unref(target_bs);
1727 }
1728 
1729 static void drive_backup_commit(void *opaque)
1730 {
1731     DriveBackupState *state = opaque;
1732 
1733     assert(state->job);
1734     job_start(&state->job->job);
1735 }
1736 
1737 static void drive_backup_abort(void *opaque)
1738 {
1739     DriveBackupState *state = opaque;
1740 
1741     if (state->job) {
1742         job_cancel_sync(&state->job->job, true);
1743     }
1744 }
1745 
1746 static void drive_backup_clean(void *opaque)
1747 {
1748     g_autofree DriveBackupState *state = opaque;
1749 
1750     if (!state->bs) {
1751         return;
1752     }
1753 
1754     bdrv_drained_end(state->bs);
1755 }
1756 
1757 typedef struct BlockdevBackupState {
1758     BlockDriverState *bs;
1759     BlockJob *job;
1760 } BlockdevBackupState;
1761 
1762 static void blockdev_backup_commit(void *opaque);
1763 static void blockdev_backup_abort(void *opaque);
1764 static void blockdev_backup_clean(void *opaque);
1765 TransactionActionDrv blockdev_backup_drv = {
1766     .commit = blockdev_backup_commit,
1767     .abort = blockdev_backup_abort,
1768     .clean = blockdev_backup_clean,
1769 };
1770 
1771 static void blockdev_backup_action(BlockdevBackup *backup,
1772                                    JobTxn *block_job_txn,
1773                                    Transaction *tran, Error **errp)
1774 {
1775     BlockdevBackupState *state = g_new0(BlockdevBackupState, 1);
1776     BlockDriverState *bs;
1777     BlockDriverState *target_bs;
1778     AioContext *aio_context;
1779     int ret;
1780 
1781     tran_add(tran, &blockdev_backup_drv, state);
1782 
1783     bs = bdrv_lookup_bs(backup->device, backup->device, errp);
1784     if (!bs) {
1785         return;
1786     }
1787 
1788     target_bs = bdrv_lookup_bs(backup->target, backup->target, errp);
1789     if (!target_bs) {
1790         return;
1791     }
1792 
1793     /* Honor bdrv_try_change_aio_context() context acquisition requirements. */
1794     aio_context = bdrv_get_aio_context(bs);
1795 
1796     ret = bdrv_try_change_aio_context(target_bs, aio_context, NULL, errp);
1797     if (ret < 0) {
1798         return;
1799     }
1800 
1801     state->bs = bs;
1802 
1803     /* Paired with .clean() */
1804     bdrv_drained_begin(state->bs);
1805 
1806     state->job = do_backup_common(qapi_BlockdevBackup_base(backup),
1807                                   bs, target_bs, aio_context,
1808                                   block_job_txn, errp);
1809 }
1810 
1811 static void blockdev_backup_commit(void *opaque)
1812 {
1813     BlockdevBackupState *state = opaque;
1814 
1815     assert(state->job);
1816     job_start(&state->job->job);
1817 }
1818 
1819 static void blockdev_backup_abort(void *opaque)
1820 {
1821     BlockdevBackupState *state = opaque;
1822 
1823     if (state->job) {
1824         job_cancel_sync(&state->job->job, true);
1825     }
1826 }
1827 
1828 static void blockdev_backup_clean(void *opaque)
1829 {
1830     g_autofree BlockdevBackupState *state = opaque;
1831 
1832     if (!state->bs) {
1833         return;
1834     }
1835 
1836     bdrv_drained_end(state->bs);
1837 }
1838 
1839 typedef struct BlockDirtyBitmapState {
1840     BdrvDirtyBitmap *bitmap;
1841     BlockDriverState *bs;
1842     HBitmap *backup;
1843     bool was_enabled;
1844 } BlockDirtyBitmapState;
1845 
1846 static void block_dirty_bitmap_add_abort(void *opaque);
1847 TransactionActionDrv block_dirty_bitmap_add_drv = {
1848     .abort = block_dirty_bitmap_add_abort,
1849     .clean = g_free,
1850 };
1851 
1852 static void block_dirty_bitmap_add_action(BlockDirtyBitmapAdd *action,
1853                                           Transaction *tran, Error **errp)
1854 {
1855     Error *local_err = NULL;
1856     BlockDirtyBitmapState *state = g_new0(BlockDirtyBitmapState, 1);
1857 
1858     tran_add(tran, &block_dirty_bitmap_add_drv, state);
1859 
1860     /* AIO context taken and released within qmp_block_dirty_bitmap_add */
1861     qmp_block_dirty_bitmap_add(action->node, action->name,
1862                                action->has_granularity, action->granularity,
1863                                action->has_persistent, action->persistent,
1864                                action->has_disabled, action->disabled,
1865                                &local_err);
1866 
1867     if (!local_err) {
1868         state->bitmap = block_dirty_bitmap_lookup(action->node, action->name,
1869                                                   NULL, &error_abort);
1870     } else {
1871         error_propagate(errp, local_err);
1872     }
1873 }
1874 
1875 static void block_dirty_bitmap_add_abort(void *opaque)
1876 {
1877     BlockDirtyBitmapState *state = opaque;
1878 
1879     if (state->bitmap) {
1880         bdrv_release_dirty_bitmap(state->bitmap);
1881     }
1882 }
1883 
1884 static void block_dirty_bitmap_restore(void *opaque);
1885 static void block_dirty_bitmap_free_backup(void *opaque);
1886 TransactionActionDrv block_dirty_bitmap_clear_drv = {
1887     .abort = block_dirty_bitmap_restore,
1888     .commit = block_dirty_bitmap_free_backup,
1889     .clean = g_free,
1890 };
1891 
1892 static void block_dirty_bitmap_clear_action(BlockDirtyBitmap *action,
1893                                             Transaction *tran, Error **errp)
1894 {
1895     BlockDirtyBitmapState *state = g_new0(BlockDirtyBitmapState, 1);
1896 
1897     tran_add(tran, &block_dirty_bitmap_clear_drv, state);
1898 
1899     state->bitmap = block_dirty_bitmap_lookup(action->node,
1900                                               action->name,
1901                                               &state->bs,
1902                                               errp);
1903     if (!state->bitmap) {
1904         return;
1905     }
1906 
1907     if (bdrv_dirty_bitmap_check(state->bitmap, BDRV_BITMAP_DEFAULT, errp)) {
1908         return;
1909     }
1910 
1911     bdrv_clear_dirty_bitmap(state->bitmap, &state->backup);
1912 }
1913 
1914 static void block_dirty_bitmap_restore(void *opaque)
1915 {
1916     BlockDirtyBitmapState *state = opaque;
1917 
1918     if (state->backup) {
1919         bdrv_restore_dirty_bitmap(state->bitmap, state->backup);
1920     }
1921 }
1922 
1923 static void block_dirty_bitmap_free_backup(void *opaque)
1924 {
1925     BlockDirtyBitmapState *state = opaque;
1926 
1927     hbitmap_free(state->backup);
1928 }
1929 
1930 static void block_dirty_bitmap_enable_abort(void *opaque);
1931 TransactionActionDrv block_dirty_bitmap_enable_drv = {
1932     .abort = block_dirty_bitmap_enable_abort,
1933     .clean = g_free,
1934 };
1935 
1936 static void block_dirty_bitmap_enable_action(BlockDirtyBitmap *action,
1937                                              Transaction *tran, Error **errp)
1938 {
1939     BlockDirtyBitmapState *state = g_new0(BlockDirtyBitmapState, 1);
1940 
1941     tran_add(tran, &block_dirty_bitmap_enable_drv, state);
1942 
1943     state->bitmap = block_dirty_bitmap_lookup(action->node,
1944                                               action->name,
1945                                               NULL,
1946                                               errp);
1947     if (!state->bitmap) {
1948         return;
1949     }
1950 
1951     if (bdrv_dirty_bitmap_check(state->bitmap, BDRV_BITMAP_ALLOW_RO, errp)) {
1952         return;
1953     }
1954 
1955     state->was_enabled = bdrv_dirty_bitmap_enabled(state->bitmap);
1956     bdrv_enable_dirty_bitmap(state->bitmap);
1957 }
1958 
1959 static void block_dirty_bitmap_enable_abort(void *opaque)
1960 {
1961     BlockDirtyBitmapState *state = opaque;
1962 
1963     if (!state->was_enabled) {
1964         bdrv_disable_dirty_bitmap(state->bitmap);
1965     }
1966 }
1967 
1968 static void block_dirty_bitmap_disable_abort(void *opaque);
1969 TransactionActionDrv block_dirty_bitmap_disable_drv = {
1970     .abort = block_dirty_bitmap_disable_abort,
1971     .clean = g_free,
1972 };
1973 
1974 static void block_dirty_bitmap_disable_action(BlockDirtyBitmap *action,
1975                                               Transaction *tran, Error **errp)
1976 {
1977     BlockDirtyBitmapState *state = g_new0(BlockDirtyBitmapState, 1);
1978 
1979     tran_add(tran, &block_dirty_bitmap_disable_drv, state);
1980 
1981     state->bitmap = block_dirty_bitmap_lookup(action->node,
1982                                               action->name,
1983                                               NULL,
1984                                               errp);
1985     if (!state->bitmap) {
1986         return;
1987     }
1988 
1989     if (bdrv_dirty_bitmap_check(state->bitmap, BDRV_BITMAP_ALLOW_RO, errp)) {
1990         return;
1991     }
1992 
1993     state->was_enabled = bdrv_dirty_bitmap_enabled(state->bitmap);
1994     bdrv_disable_dirty_bitmap(state->bitmap);
1995 }
1996 
1997 static void block_dirty_bitmap_disable_abort(void *opaque)
1998 {
1999     BlockDirtyBitmapState *state = opaque;
2000 
2001     if (state->was_enabled) {
2002         bdrv_enable_dirty_bitmap(state->bitmap);
2003     }
2004 }
2005 
2006 TransactionActionDrv block_dirty_bitmap_merge_drv = {
2007     .commit = block_dirty_bitmap_free_backup,
2008     .abort = block_dirty_bitmap_restore,
2009     .clean = g_free,
2010 };
2011 
2012 static void block_dirty_bitmap_merge_action(BlockDirtyBitmapMerge *action,
2013                                             Transaction *tran, Error **errp)
2014 {
2015     BlockDirtyBitmapState *state = g_new0(BlockDirtyBitmapState, 1);
2016 
2017     tran_add(tran, &block_dirty_bitmap_merge_drv, state);
2018 
2019     state->bitmap = block_dirty_bitmap_merge(action->node, action->target,
2020                                              action->bitmaps, &state->backup,
2021                                              errp);
2022 }
2023 
2024 static void block_dirty_bitmap_remove_commit(void *opaque);
2025 static void block_dirty_bitmap_remove_abort(void *opaque);
2026 TransactionActionDrv block_dirty_bitmap_remove_drv = {
2027     .commit = block_dirty_bitmap_remove_commit,
2028     .abort = block_dirty_bitmap_remove_abort,
2029     .clean = g_free,
2030 };
2031 
2032 static void block_dirty_bitmap_remove_action(BlockDirtyBitmap *action,
2033                                              Transaction *tran, Error **errp)
2034 {
2035     BlockDirtyBitmapState *state = g_new0(BlockDirtyBitmapState, 1);
2036 
2037     tran_add(tran, &block_dirty_bitmap_remove_drv, state);
2038 
2039 
2040     state->bitmap = block_dirty_bitmap_remove(action->node, action->name,
2041                                               false, &state->bs, errp);
2042     if (state->bitmap) {
2043         bdrv_dirty_bitmap_skip_store(state->bitmap, true);
2044         bdrv_dirty_bitmap_set_busy(state->bitmap, true);
2045     }
2046 }
2047 
2048 static void block_dirty_bitmap_remove_abort(void *opaque)
2049 {
2050     BlockDirtyBitmapState *state = opaque;
2051 
2052     if (state->bitmap) {
2053         bdrv_dirty_bitmap_skip_store(state->bitmap, false);
2054         bdrv_dirty_bitmap_set_busy(state->bitmap, false);
2055     }
2056 }
2057 
2058 static void block_dirty_bitmap_remove_commit(void *opaque)
2059 {
2060     BlockDirtyBitmapState *state = opaque;
2061 
2062     bdrv_dirty_bitmap_set_busy(state->bitmap, false);
2063     bdrv_release_dirty_bitmap(state->bitmap);
2064 }
2065 
2066 static void abort_commit(void *opaque);
2067 TransactionActionDrv abort_drv = {
2068     .commit = abort_commit,
2069 };
2070 
2071 static void abort_action(Transaction *tran, Error **errp)
2072 {
2073     tran_add(tran, &abort_drv, NULL);
2074     error_setg(errp, "Transaction aborted using Abort action");
2075 }
2076 
2077 static void abort_commit(void *opaque)
2078 {
2079     g_assert_not_reached(); /* this action never succeeds */
2080 }
2081 
2082 static void transaction_action(TransactionAction *act, JobTxn *block_job_txn,
2083                                Transaction *tran, Error **errp)
2084 {
2085     switch (act->type) {
2086     case TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT:
2087     case TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC:
2088         external_snapshot_action(act, tran, errp);
2089         return;
2090     case TRANSACTION_ACTION_KIND_DRIVE_BACKUP:
2091         drive_backup_action(act->u.drive_backup.data,
2092                             block_job_txn, tran, errp);
2093         return;
2094     case TRANSACTION_ACTION_KIND_BLOCKDEV_BACKUP:
2095         blockdev_backup_action(act->u.blockdev_backup.data,
2096                                block_job_txn, tran, errp);
2097         return;
2098     case TRANSACTION_ACTION_KIND_ABORT:
2099         abort_action(tran, errp);
2100         return;
2101     case TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC:
2102         internal_snapshot_action(act->u.blockdev_snapshot_internal_sync.data,
2103                                  tran, errp);
2104         return;
2105     case TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_ADD:
2106         block_dirty_bitmap_add_action(act->u.block_dirty_bitmap_add.data,
2107                                       tran, errp);
2108         return;
2109     case TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_CLEAR:
2110         block_dirty_bitmap_clear_action(act->u.block_dirty_bitmap_clear.data,
2111                                         tran, errp);
2112         return;
2113     case TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_ENABLE:
2114         block_dirty_bitmap_enable_action(act->u.block_dirty_bitmap_enable.data,
2115                                          tran, errp);
2116         return;
2117     case TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_DISABLE:
2118         block_dirty_bitmap_disable_action(
2119                 act->u.block_dirty_bitmap_disable.data, tran, errp);
2120         return;
2121     case TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_MERGE:
2122         block_dirty_bitmap_merge_action(act->u.block_dirty_bitmap_merge.data,
2123                                         tran, errp);
2124         return;
2125     case TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_REMOVE:
2126         block_dirty_bitmap_remove_action(act->u.block_dirty_bitmap_remove.data,
2127                                          tran, errp);
2128         return;
2129     /*
2130      * Where are transactions for MIRROR, COMMIT and STREAM?
2131      * Although these blockjobs use transaction callbacks like the backup job,
2132      * these jobs do not necessarily adhere to transaction semantics.
2133      * These jobs may not fully undo all of their actions on abort, nor do they
2134      * necessarily work in transactions with more than one job in them.
2135      */
2136     case TRANSACTION_ACTION_KIND__MAX:
2137     default:
2138         g_assert_not_reached();
2139     };
2140 }
2141 
2142 
2143 /*
2144  * 'Atomic' group operations.  The operations are performed as a set, and if
2145  * any fail then we roll back all operations in the group.
2146  *
2147  * Always run under BQL.
2148  */
2149 void qmp_transaction(TransactionActionList *actions,
2150                      struct TransactionProperties *properties,
2151                      Error **errp)
2152 {
2153     TransactionActionList *act;
2154     JobTxn *block_job_txn = NULL;
2155     Error *local_err = NULL;
2156     Transaction *tran;
2157     ActionCompletionMode comp_mode =
2158         properties ? properties->completion_mode :
2159         ACTION_COMPLETION_MODE_INDIVIDUAL;
2160 
2161     GLOBAL_STATE_CODE();
2162 
2163     /* Does this transaction get canceled as a group on failure?
2164      * If not, we don't really need to make a JobTxn.
2165      */
2166     if (comp_mode != ACTION_COMPLETION_MODE_INDIVIDUAL) {
2167         for (act = actions; act; act = act->next) {
2168             TransactionActionKind type = act->value->type;
2169 
2170             if (type != TRANSACTION_ACTION_KIND_BLOCKDEV_BACKUP &&
2171                 type != TRANSACTION_ACTION_KIND_DRIVE_BACKUP)
2172             {
2173                 error_setg(errp,
2174                            "Action '%s' does not support transaction property "
2175                            "completion-mode = %s",
2176                            TransactionActionKind_str(type),
2177                            ActionCompletionMode_str(comp_mode));
2178                 return;
2179             }
2180         }
2181 
2182         block_job_txn = job_txn_new();
2183     }
2184 
2185     /* drain all i/o before any operations */
2186     bdrv_drain_all();
2187 
2188     tran = tran_new();
2189 
2190     /* We don't do anything in this loop that commits us to the operations */
2191     for (act = actions; act; act = act->next) {
2192         transaction_action(act->value, block_job_txn, tran, &local_err);
2193         if (local_err) {
2194             error_propagate(errp, local_err);
2195             goto delete_and_fail;
2196         }
2197     }
2198 
2199     tran_commit(tran);
2200 
2201     /* success */
2202     goto exit;
2203 
2204 delete_and_fail:
2205     /* failure, and it is all-or-none; roll back all operations */
2206     tran_abort(tran);
2207 exit:
2208     job_txn_unref(block_job_txn);
2209 }
2210 
2211 BlockDirtyBitmapSha256 *qmp_x_debug_block_dirty_bitmap_sha256(const char *node,
2212                                                               const char *name,
2213                                                               Error **errp)
2214 {
2215     BdrvDirtyBitmap *bitmap;
2216     BlockDriverState *bs;
2217     BlockDirtyBitmapSha256 *ret = NULL;
2218     char *sha256;
2219 
2220     bitmap = block_dirty_bitmap_lookup(node, name, &bs, errp);
2221     if (!bitmap || !bs) {
2222         return NULL;
2223     }
2224 
2225     sha256 = bdrv_dirty_bitmap_sha256(bitmap, errp);
2226     if (sha256 == NULL) {
2227         return NULL;
2228     }
2229 
2230     ret = g_new(BlockDirtyBitmapSha256, 1);
2231     ret->sha256 = sha256;
2232 
2233     return ret;
2234 }
2235 
2236 void coroutine_fn qmp_block_resize(const char *device, const char *node_name,
2237                                    int64_t size, Error **errp)
2238 {
2239     Error *local_err = NULL;
2240     BlockBackend *blk;
2241     BlockDriverState *bs;
2242     AioContext *old_ctx;
2243 
2244     bs = bdrv_lookup_bs(device, node_name, &local_err);
2245     if (local_err) {
2246         error_propagate(errp, local_err);
2247         return;
2248     }
2249 
2250     if (size < 0) {
2251         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "size", "a >0 size");
2252         return;
2253     }
2254 
2255     bdrv_graph_co_rdlock();
2256     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_RESIZE, errp)) {
2257         bdrv_graph_co_rdunlock();
2258         return;
2259     }
2260     bdrv_graph_co_rdunlock();
2261 
2262     blk = blk_co_new_with_bs(bs, BLK_PERM_RESIZE, BLK_PERM_ALL, errp);
2263     if (!blk) {
2264         return;
2265     }
2266 
2267     bdrv_drained_begin(bs);
2268 
2269     old_ctx = bdrv_co_enter(bs);
2270     blk_co_truncate(blk, size, false, PREALLOC_MODE_OFF, 0, errp);
2271     bdrv_co_leave(bs, old_ctx);
2272 
2273     bdrv_drained_end(bs);
2274     blk_co_unref(blk);
2275 }
2276 
2277 void qmp_block_stream(const char *job_id, const char *device,
2278                       const char *base,
2279                       const char *base_node,
2280                       const char *backing_file,
2281                       bool has_backing_mask_protocol,
2282                       bool backing_mask_protocol,
2283                       const char *bottom,
2284                       bool has_speed, int64_t speed,
2285                       bool has_on_error, BlockdevOnError on_error,
2286                       const char *filter_node_name,
2287                       bool has_auto_finalize, bool auto_finalize,
2288                       bool has_auto_dismiss, bool auto_dismiss,
2289                       Error **errp)
2290 {
2291     BlockDriverState *bs, *iter, *iter_end;
2292     BlockDriverState *base_bs = NULL;
2293     BlockDriverState *bottom_bs = NULL;
2294     AioContext *aio_context;
2295     Error *local_err = NULL;
2296     int job_flags = JOB_DEFAULT;
2297 
2298     GLOBAL_STATE_CODE();
2299 
2300     if (base && base_node) {
2301         error_setg(errp, "'base' and 'base-node' cannot be specified "
2302                    "at the same time");
2303         return;
2304     }
2305 
2306     if (base && bottom) {
2307         error_setg(errp, "'base' and 'bottom' cannot be specified "
2308                    "at the same time");
2309         return;
2310     }
2311 
2312     if (bottom && base_node) {
2313         error_setg(errp, "'bottom' and 'base-node' cannot be specified "
2314                    "at the same time");
2315         return;
2316     }
2317 
2318     if (!has_backing_mask_protocol) {
2319         backing_mask_protocol = false;
2320     }
2321 
2322     if (!has_on_error) {
2323         on_error = BLOCKDEV_ON_ERROR_REPORT;
2324     }
2325 
2326     bs = bdrv_lookup_bs(device, device, errp);
2327     if (!bs) {
2328         return;
2329     }
2330 
2331     aio_context = bdrv_get_aio_context(bs);
2332 
2333     bdrv_graph_rdlock_main_loop();
2334     if (base) {
2335         base_bs = bdrv_find_backing_image(bs, base);
2336         if (base_bs == NULL) {
2337             error_setg(errp, "Can't find '%s' in the backing chain", base);
2338             goto out_rdlock;
2339         }
2340         assert(bdrv_get_aio_context(base_bs) == aio_context);
2341     }
2342 
2343     if (base_node) {
2344         base_bs = bdrv_lookup_bs(NULL, base_node, errp);
2345         if (!base_bs) {
2346             goto out_rdlock;
2347         }
2348         if (bs == base_bs || !bdrv_chain_contains(bs, base_bs)) {
2349             error_setg(errp, "Node '%s' is not a backing image of '%s'",
2350                        base_node, device);
2351             goto out_rdlock;
2352         }
2353         assert(bdrv_get_aio_context(base_bs) == aio_context);
2354 
2355         bdrv_refresh_filename(base_bs);
2356     }
2357 
2358     if (bottom) {
2359         bottom_bs = bdrv_lookup_bs(NULL, bottom, errp);
2360         if (!bottom_bs) {
2361             goto out_rdlock;
2362         }
2363         if (!bottom_bs->drv) {
2364             error_setg(errp, "Node '%s' is not open", bottom);
2365             goto out_rdlock;
2366         }
2367         if (bottom_bs->drv->is_filter) {
2368             error_setg(errp, "Node '%s' is a filter, use a non-filter node "
2369                        "as 'bottom'", bottom);
2370             goto out_rdlock;
2371         }
2372         if (!bdrv_chain_contains(bs, bottom_bs)) {
2373             error_setg(errp, "Node '%s' is not in a chain starting from '%s'",
2374                        bottom, device);
2375             goto out_rdlock;
2376         }
2377         assert(bdrv_get_aio_context(bottom_bs) == aio_context);
2378     }
2379 
2380     /*
2381      * Check for op blockers in the whole chain between bs and base (or bottom)
2382      */
2383     iter_end = bottom ? bdrv_filter_or_cow_bs(bottom_bs) : base_bs;
2384     for (iter = bs; iter && iter != iter_end;
2385          iter = bdrv_filter_or_cow_bs(iter))
2386     {
2387         if (bdrv_op_is_blocked(iter, BLOCK_OP_TYPE_STREAM, errp)) {
2388             goto out_rdlock;
2389         }
2390     }
2391     bdrv_graph_rdunlock_main_loop();
2392 
2393     /* if we are streaming the entire chain, the result will have no backing
2394      * file, and specifying one is therefore an error */
2395     if (!base_bs && backing_file) {
2396         error_setg(errp, "backing file specified, but streaming the "
2397                          "entire chain");
2398         return;
2399     }
2400 
2401     if (has_auto_finalize && !auto_finalize) {
2402         job_flags |= JOB_MANUAL_FINALIZE;
2403     }
2404     if (has_auto_dismiss && !auto_dismiss) {
2405         job_flags |= JOB_MANUAL_DISMISS;
2406     }
2407 
2408     stream_start(job_id, bs, base_bs, backing_file,
2409                  backing_mask_protocol,
2410                  bottom_bs, job_flags, has_speed ? speed : 0, on_error,
2411                  filter_node_name, &local_err);
2412     if (local_err) {
2413         error_propagate(errp, local_err);
2414         return;
2415     }
2416 
2417     trace_qmp_block_stream(bs);
2418     return;
2419 
2420 out_rdlock:
2421     bdrv_graph_rdunlock_main_loop();
2422 }
2423 
2424 void qmp_block_commit(const char *job_id, const char *device,
2425                       const char *base_node,
2426                       const char *base,
2427                       const char *top_node,
2428                       const char *top,
2429                       const char *backing_file,
2430                       bool has_backing_mask_protocol,
2431                       bool backing_mask_protocol,
2432                       bool has_speed, int64_t speed,
2433                       bool has_on_error, BlockdevOnError on_error,
2434                       const char *filter_node_name,
2435                       bool has_auto_finalize, bool auto_finalize,
2436                       bool has_auto_dismiss, bool auto_dismiss,
2437                       Error **errp)
2438 {
2439     BlockDriverState *bs;
2440     BlockDriverState *iter;
2441     BlockDriverState *base_bs, *top_bs;
2442     AioContext *aio_context;
2443     Error *local_err = NULL;
2444     int job_flags = JOB_DEFAULT;
2445     uint64_t top_perm, top_shared;
2446 
2447     /* TODO We'll eventually have to take a writer lock in this function */
2448     GRAPH_RDLOCK_GUARD_MAINLOOP();
2449 
2450     if (!has_speed) {
2451         speed = 0;
2452     }
2453     if (!has_on_error) {
2454         on_error = BLOCKDEV_ON_ERROR_REPORT;
2455     }
2456     if (has_auto_finalize && !auto_finalize) {
2457         job_flags |= JOB_MANUAL_FINALIZE;
2458     }
2459     if (has_auto_dismiss && !auto_dismiss) {
2460         job_flags |= JOB_MANUAL_DISMISS;
2461     }
2462     if (!has_backing_mask_protocol) {
2463         backing_mask_protocol = false;
2464     }
2465 
2466     /* Important Note:
2467      *  libvirt relies on the DeviceNotFound error class in order to probe for
2468      *  live commit feature versions; for this to work, we must make sure to
2469      *  perform the device lookup before any generic errors that may occur in a
2470      *  scenario in which all optional arguments are omitted. */
2471     bs = qmp_get_root_bs(device, &local_err);
2472     if (!bs) {
2473         bs = bdrv_lookup_bs(device, device, NULL);
2474         if (!bs) {
2475             error_free(local_err);
2476             error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
2477                       "Device '%s' not found", device);
2478         } else {
2479             error_propagate(errp, local_err);
2480         }
2481         return;
2482     }
2483 
2484     aio_context = bdrv_get_aio_context(bs);
2485 
2486     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_COMMIT_SOURCE, errp)) {
2487         return;
2488     }
2489 
2490     /* default top_bs is the active layer */
2491     top_bs = bs;
2492 
2493     if (top_node && top) {
2494         error_setg(errp, "'top-node' and 'top' are mutually exclusive");
2495         return;
2496     } else if (top_node) {
2497         top_bs = bdrv_lookup_bs(NULL, top_node, errp);
2498         if (top_bs == NULL) {
2499             return;
2500         }
2501         if (!bdrv_chain_contains(bs, top_bs)) {
2502             error_setg(errp, "'%s' is not in this backing file chain",
2503                        top_node);
2504             return;
2505         }
2506     } else if (top) {
2507         /* This strcmp() is just a shortcut, there is no need to
2508          * refresh @bs's filename.  If it mismatches,
2509          * bdrv_find_backing_image() will do the refresh and may still
2510          * return @bs. */
2511         if (strcmp(bs->filename, top) != 0) {
2512             top_bs = bdrv_find_backing_image(bs, top);
2513         }
2514     }
2515 
2516     if (top_bs == NULL) {
2517         error_setg(errp, "Top image file %s not found", top ? top : "NULL");
2518         return;
2519     }
2520 
2521     assert(bdrv_get_aio_context(top_bs) == aio_context);
2522 
2523     if (base_node && base) {
2524         error_setg(errp, "'base-node' and 'base' are mutually exclusive");
2525         return;
2526     } else if (base_node) {
2527         base_bs = bdrv_lookup_bs(NULL, base_node, errp);
2528         if (base_bs == NULL) {
2529             return;
2530         }
2531         if (!bdrv_chain_contains(top_bs, base_bs)) {
2532             error_setg(errp, "'%s' is not in this backing file chain",
2533                        base_node);
2534             return;
2535         }
2536     } else if (base) {
2537         base_bs = bdrv_find_backing_image(top_bs, base);
2538         if (base_bs == NULL) {
2539             error_setg(errp, "Can't find '%s' in the backing chain", base);
2540             return;
2541         }
2542     } else {
2543         base_bs = bdrv_find_base(top_bs);
2544         if (base_bs == NULL) {
2545             error_setg(errp, "There is no backimg image");
2546             return;
2547         }
2548     }
2549 
2550     assert(bdrv_get_aio_context(base_bs) == aio_context);
2551 
2552     for (iter = top_bs; iter != bdrv_filter_or_cow_bs(base_bs);
2553          iter = bdrv_filter_or_cow_bs(iter))
2554     {
2555         if (bdrv_op_is_blocked(iter, BLOCK_OP_TYPE_COMMIT_TARGET, errp)) {
2556             return;
2557         }
2558     }
2559 
2560     /* Do not allow attempts to commit an image into itself */
2561     if (top_bs == base_bs) {
2562         error_setg(errp, "cannot commit an image into itself");
2563         return;
2564     }
2565 
2566     /*
2567      * Active commit is required if and only if someone has taken a
2568      * WRITE permission on the top node.  Historically, we have always
2569      * used active commit for top nodes, so continue that practice
2570      * lest we possibly break clients that rely on this behavior, e.g.
2571      * to later attach this node to a writing parent.
2572      * (Active commit is never really wrong.)
2573      */
2574     bdrv_get_cumulative_perm(top_bs, &top_perm, &top_shared);
2575     if (top_perm & BLK_PERM_WRITE ||
2576         bdrv_skip_filters(top_bs) == bdrv_skip_filters(bs))
2577     {
2578         if (backing_file) {
2579             if (bdrv_skip_filters(top_bs) == bdrv_skip_filters(bs)) {
2580                 error_setg(errp, "'backing-file' specified,"
2581                                  " but 'top' is the active layer");
2582             } else {
2583                 error_setg(errp, "'backing-file' specified, but 'top' has a "
2584                                  "writer on it");
2585             }
2586             return;
2587         }
2588         if (!job_id) {
2589             /*
2590              * Emulate here what block_job_create() does, because it
2591              * is possible that @bs != @top_bs (the block job should
2592              * be named after @bs, even if @top_bs is the actual
2593              * source)
2594              */
2595             job_id = bdrv_get_device_name(bs);
2596         }
2597         commit_active_start(job_id, top_bs, base_bs, job_flags, speed, on_error,
2598                             filter_node_name, NULL, NULL, false, &local_err);
2599     } else {
2600         BlockDriverState *overlay_bs = bdrv_find_overlay(bs, top_bs);
2601         if (bdrv_op_is_blocked(overlay_bs, BLOCK_OP_TYPE_COMMIT_TARGET, errp)) {
2602             return;
2603         }
2604         commit_start(job_id, bs, base_bs, top_bs, job_flags,
2605                      speed, on_error, backing_file,
2606                      backing_mask_protocol,
2607                      filter_node_name, &local_err);
2608     }
2609     if (local_err != NULL) {
2610         error_propagate(errp, local_err);
2611         return;
2612     }
2613 }
2614 
2615 /* Common QMP interface for drive-backup and blockdev-backup */
2616 static BlockJob *do_backup_common(BackupCommon *backup,
2617                                   BlockDriverState *bs,
2618                                   BlockDriverState *target_bs,
2619                                   AioContext *aio_context,
2620                                   JobTxn *txn, Error **errp)
2621 {
2622     BlockJob *job = NULL;
2623     BdrvDirtyBitmap *bmap = NULL;
2624     BackupPerf perf = { .max_workers = 64 };
2625     int job_flags = JOB_DEFAULT;
2626 
2627     if (!backup->has_speed) {
2628         backup->speed = 0;
2629     }
2630     if (!backup->has_on_source_error) {
2631         backup->on_source_error = BLOCKDEV_ON_ERROR_REPORT;
2632     }
2633     if (!backup->has_on_target_error) {
2634         backup->on_target_error = BLOCKDEV_ON_ERROR_REPORT;
2635     }
2636     if (!backup->has_auto_finalize) {
2637         backup->auto_finalize = true;
2638     }
2639     if (!backup->has_auto_dismiss) {
2640         backup->auto_dismiss = true;
2641     }
2642     if (!backup->has_compress) {
2643         backup->compress = false;
2644     }
2645 
2646     if (backup->x_perf) {
2647         if (backup->x_perf->has_use_copy_range) {
2648             perf.use_copy_range = backup->x_perf->use_copy_range;
2649         }
2650         if (backup->x_perf->has_max_workers) {
2651             perf.max_workers = backup->x_perf->max_workers;
2652         }
2653         if (backup->x_perf->has_max_chunk) {
2654             perf.max_chunk = backup->x_perf->max_chunk;
2655         }
2656     }
2657 
2658     if ((backup->sync == MIRROR_SYNC_MODE_BITMAP) ||
2659         (backup->sync == MIRROR_SYNC_MODE_INCREMENTAL)) {
2660         /* done before desugaring 'incremental' to print the right message */
2661         if (!backup->bitmap) {
2662             error_setg(errp, "must provide a valid bitmap name for "
2663                        "'%s' sync mode", MirrorSyncMode_str(backup->sync));
2664             return NULL;
2665         }
2666     }
2667 
2668     if (backup->sync == MIRROR_SYNC_MODE_INCREMENTAL) {
2669         if (backup->has_bitmap_mode &&
2670             backup->bitmap_mode != BITMAP_SYNC_MODE_ON_SUCCESS) {
2671             error_setg(errp, "Bitmap sync mode must be '%s' "
2672                        "when using sync mode '%s'",
2673                        BitmapSyncMode_str(BITMAP_SYNC_MODE_ON_SUCCESS),
2674                        MirrorSyncMode_str(backup->sync));
2675             return NULL;
2676         }
2677         backup->has_bitmap_mode = true;
2678         backup->sync = MIRROR_SYNC_MODE_BITMAP;
2679         backup->bitmap_mode = BITMAP_SYNC_MODE_ON_SUCCESS;
2680     }
2681 
2682     if (backup->bitmap) {
2683         bmap = bdrv_find_dirty_bitmap(bs, backup->bitmap);
2684         if (!bmap) {
2685             error_setg(errp, "Bitmap '%s' could not be found", backup->bitmap);
2686             return NULL;
2687         }
2688         if (!backup->has_bitmap_mode) {
2689             error_setg(errp, "Bitmap sync mode must be given "
2690                        "when providing a bitmap");
2691             return NULL;
2692         }
2693         if (bdrv_dirty_bitmap_check(bmap, BDRV_BITMAP_ALLOW_RO, errp)) {
2694             return NULL;
2695         }
2696 
2697         /* This does not produce a useful bitmap artifact: */
2698         if (backup->sync == MIRROR_SYNC_MODE_NONE) {
2699             error_setg(errp, "sync mode '%s' does not produce meaningful bitmap"
2700                        " outputs", MirrorSyncMode_str(backup->sync));
2701             return NULL;
2702         }
2703 
2704         /* If the bitmap isn't used for input or output, this is useless: */
2705         if (backup->bitmap_mode == BITMAP_SYNC_MODE_NEVER &&
2706             backup->sync != MIRROR_SYNC_MODE_BITMAP) {
2707             error_setg(errp, "Bitmap sync mode '%s' has no meaningful effect"
2708                        " when combined with sync mode '%s'",
2709                        BitmapSyncMode_str(backup->bitmap_mode),
2710                        MirrorSyncMode_str(backup->sync));
2711             return NULL;
2712         }
2713     }
2714 
2715     if (!backup->bitmap && backup->has_bitmap_mode) {
2716         error_setg(errp, "Cannot specify bitmap sync mode without a bitmap");
2717         return NULL;
2718     }
2719 
2720     if (!backup->auto_finalize) {
2721         job_flags |= JOB_MANUAL_FINALIZE;
2722     }
2723     if (!backup->auto_dismiss) {
2724         job_flags |= JOB_MANUAL_DISMISS;
2725     }
2726 
2727     job = backup_job_create(backup->job_id, bs, target_bs, backup->speed,
2728                             backup->sync, bmap, backup->bitmap_mode,
2729                             backup->compress,
2730                             backup->filter_node_name,
2731                             &perf,
2732                             backup->on_source_error,
2733                             backup->on_target_error,
2734                             job_flags, NULL, NULL, txn, errp);
2735     return job;
2736 }
2737 
2738 void qmp_drive_backup(DriveBackup *backup, Error **errp)
2739 {
2740     TransactionAction action = {
2741         .type = TRANSACTION_ACTION_KIND_DRIVE_BACKUP,
2742         .u.drive_backup.data = backup,
2743     };
2744     blockdev_do_action(&action, errp);
2745 }
2746 
2747 BlockDeviceInfoList *qmp_query_named_block_nodes(bool has_flat,
2748                                                  bool flat,
2749                                                  Error **errp)
2750 {
2751     bool return_flat = has_flat && flat;
2752 
2753     return bdrv_named_nodes_list(return_flat, errp);
2754 }
2755 
2756 XDbgBlockGraph *qmp_x_debug_query_block_graph(Error **errp)
2757 {
2758     GRAPH_RDLOCK_GUARD_MAINLOOP();
2759 
2760     return bdrv_get_xdbg_block_graph(errp);
2761 }
2762 
2763 void qmp_blockdev_backup(BlockdevBackup *backup, Error **errp)
2764 {
2765     TransactionAction action = {
2766         .type = TRANSACTION_ACTION_KIND_BLOCKDEV_BACKUP,
2767         .u.blockdev_backup.data = backup,
2768     };
2769     blockdev_do_action(&action, errp);
2770 }
2771 
2772 /* Parameter check and block job starting for drive mirroring.
2773  * Caller should hold @device and @target's aio context (must be the same).
2774  **/
2775 static void blockdev_mirror_common(const char *job_id, BlockDriverState *bs,
2776                                    BlockDriverState *target,
2777                                    const char *replaces,
2778                                    enum MirrorSyncMode sync,
2779                                    BlockMirrorBackingMode backing_mode,
2780                                    bool zero_target,
2781                                    bool has_speed, int64_t speed,
2782                                    bool has_granularity, uint32_t granularity,
2783                                    bool has_buf_size, int64_t buf_size,
2784                                    bool has_on_source_error,
2785                                    BlockdevOnError on_source_error,
2786                                    bool has_on_target_error,
2787                                    BlockdevOnError on_target_error,
2788                                    bool has_unmap, bool unmap,
2789                                    const char *filter_node_name,
2790                                    bool has_copy_mode, MirrorCopyMode copy_mode,
2791                                    bool has_auto_finalize, bool auto_finalize,
2792                                    bool has_auto_dismiss, bool auto_dismiss,
2793                                    Error **errp)
2794 {
2795     BlockDriverState *unfiltered_bs;
2796     int job_flags = JOB_DEFAULT;
2797 
2798     GLOBAL_STATE_CODE();
2799     GRAPH_RDLOCK_GUARD_MAINLOOP();
2800 
2801     if (!has_speed) {
2802         speed = 0;
2803     }
2804     if (!has_on_source_error) {
2805         on_source_error = BLOCKDEV_ON_ERROR_REPORT;
2806     }
2807     if (!has_on_target_error) {
2808         on_target_error = BLOCKDEV_ON_ERROR_REPORT;
2809     }
2810     if (!has_granularity) {
2811         granularity = 0;
2812     }
2813     if (!has_buf_size) {
2814         buf_size = 0;
2815     }
2816     if (!has_unmap) {
2817         unmap = true;
2818     }
2819     if (!has_copy_mode) {
2820         copy_mode = MIRROR_COPY_MODE_BACKGROUND;
2821     }
2822     if (has_auto_finalize && !auto_finalize) {
2823         job_flags |= JOB_MANUAL_FINALIZE;
2824     }
2825     if (has_auto_dismiss && !auto_dismiss) {
2826         job_flags |= JOB_MANUAL_DISMISS;
2827     }
2828 
2829     if (granularity != 0 && (granularity < 512 || granularity > 1048576 * 64)) {
2830         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "granularity",
2831                    "a value in range [512B, 64MB]");
2832         return;
2833     }
2834     if (granularity & (granularity - 1)) {
2835         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "granularity",
2836                    "a power of 2");
2837         return;
2838     }
2839 
2840     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_MIRROR_SOURCE, errp)) {
2841         return;
2842     }
2843     if (bdrv_op_is_blocked(target, BLOCK_OP_TYPE_MIRROR_TARGET, errp)) {
2844         return;
2845     }
2846 
2847     if (!bdrv_backing_chain_next(bs) && sync == MIRROR_SYNC_MODE_TOP) {
2848         sync = MIRROR_SYNC_MODE_FULL;
2849     }
2850 
2851     if (!replaces) {
2852         /* We want to mirror from @bs, but keep implicit filters on top */
2853         unfiltered_bs = bdrv_skip_implicit_filters(bs);
2854         if (unfiltered_bs != bs) {
2855             replaces = unfiltered_bs->node_name;
2856         }
2857     }
2858 
2859     if (replaces) {
2860         BlockDriverState *to_replace_bs;
2861         int64_t bs_size, replace_size;
2862 
2863         bs_size = bdrv_getlength(bs);
2864         if (bs_size < 0) {
2865             error_setg_errno(errp, -bs_size, "Failed to query device's size");
2866             return;
2867         }
2868 
2869         to_replace_bs = check_to_replace_node(bs, replaces, errp);
2870         if (!to_replace_bs) {
2871             return;
2872         }
2873 
2874         replace_size = bdrv_getlength(to_replace_bs);
2875 
2876         if (replace_size < 0) {
2877             error_setg_errno(errp, -replace_size,
2878                              "Failed to query the replacement node's size");
2879             return;
2880         }
2881         if (bs_size != replace_size) {
2882             error_setg(errp, "cannot replace image with a mirror image of "
2883                              "different size");
2884             return;
2885         }
2886     }
2887 
2888     /* pass the node name to replace to mirror start since it's loose coupling
2889      * and will allow to check whether the node still exist at mirror completion
2890      */
2891     mirror_start(job_id, bs, target,
2892                  replaces, job_flags,
2893                  speed, granularity, buf_size, sync, backing_mode, zero_target,
2894                  on_source_error, on_target_error, unmap, filter_node_name,
2895                  copy_mode, errp);
2896 }
2897 
2898 void qmp_drive_mirror(DriveMirror *arg, Error **errp)
2899 {
2900     BlockDriverState *bs;
2901     BlockDriverState *target_backing_bs, *target_bs;
2902     AioContext *aio_context;
2903     BlockMirrorBackingMode backing_mode;
2904     Error *local_err = NULL;
2905     QDict *options = NULL;
2906     int flags;
2907     int64_t size;
2908     const char *format = arg->format;
2909     bool zero_target;
2910     int ret;
2911 
2912     bs = qmp_get_root_bs(arg->device, errp);
2913     if (!bs) {
2914         return;
2915     }
2916 
2917     /* Early check to avoid creating target */
2918     bdrv_graph_rdlock_main_loop();
2919     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_MIRROR_SOURCE, errp)) {
2920         bdrv_graph_rdunlock_main_loop();
2921         return;
2922     }
2923 
2924     aio_context = bdrv_get_aio_context(bs);
2925 
2926     if (!arg->has_mode) {
2927         arg->mode = NEW_IMAGE_MODE_ABSOLUTE_PATHS;
2928     }
2929 
2930     if (!arg->format) {
2931         format = (arg->mode == NEW_IMAGE_MODE_EXISTING
2932                   ? NULL : bs->drv->format_name);
2933     }
2934 
2935     flags = bs->open_flags | BDRV_O_RDWR;
2936     target_backing_bs = bdrv_cow_bs(bdrv_skip_filters(bs));
2937     if (!target_backing_bs && arg->sync == MIRROR_SYNC_MODE_TOP) {
2938         arg->sync = MIRROR_SYNC_MODE_FULL;
2939     }
2940     if (arg->sync == MIRROR_SYNC_MODE_NONE) {
2941         target_backing_bs = bs;
2942     }
2943     bdrv_graph_rdunlock_main_loop();
2944 
2945     size = bdrv_getlength(bs);
2946     if (size < 0) {
2947         error_setg_errno(errp, -size, "bdrv_getlength failed");
2948         return;
2949     }
2950 
2951     if (arg->replaces) {
2952         if (!arg->node_name) {
2953             error_setg(errp, "a node-name must be provided when replacing a"
2954                              " named node of the graph");
2955             return;
2956         }
2957     }
2958 
2959     if (arg->mode == NEW_IMAGE_MODE_ABSOLUTE_PATHS) {
2960         backing_mode = MIRROR_SOURCE_BACKING_CHAIN;
2961     } else {
2962         backing_mode = MIRROR_OPEN_BACKING_CHAIN;
2963     }
2964 
2965     /* Don't open backing image in create() */
2966     flags |= BDRV_O_NO_BACKING;
2967 
2968     if ((arg->sync == MIRROR_SYNC_MODE_FULL || !target_backing_bs)
2969         && arg->mode != NEW_IMAGE_MODE_EXISTING)
2970     {
2971         /* create new image w/o backing file */
2972         assert(format);
2973         bdrv_img_create(arg->target, format,
2974                         NULL, NULL, NULL, size, flags, false, &local_err);
2975     } else {
2976         BlockDriverState *explicit_backing;
2977 
2978         switch (arg->mode) {
2979         case NEW_IMAGE_MODE_EXISTING:
2980             break;
2981         case NEW_IMAGE_MODE_ABSOLUTE_PATHS:
2982             /*
2983              * Create new image with backing file.
2984              * Implicit filters should not appear in the filename.
2985              */
2986             bdrv_graph_rdlock_main_loop();
2987             explicit_backing = bdrv_skip_implicit_filters(target_backing_bs);
2988             bdrv_refresh_filename(explicit_backing);
2989             bdrv_graph_rdunlock_main_loop();
2990 
2991             bdrv_img_create(arg->target, format,
2992                             explicit_backing->filename,
2993                             explicit_backing->drv->format_name,
2994                             NULL, size, flags, false, &local_err);
2995             break;
2996         default:
2997             abort();
2998         }
2999     }
3000 
3001     if (local_err) {
3002         error_propagate(errp, local_err);
3003         return;
3004     }
3005 
3006     options = qdict_new();
3007     if (arg->node_name) {
3008         qdict_put_str(options, "node-name", arg->node_name);
3009     }
3010     if (format) {
3011         qdict_put_str(options, "driver", format);
3012     }
3013 
3014     /* Mirroring takes care of copy-on-write using the source's backing
3015      * file.
3016      */
3017     target_bs = bdrv_open(arg->target, NULL, options, flags, errp);
3018     if (!target_bs) {
3019         return;
3020     }
3021 
3022     bdrv_graph_rdlock_main_loop();
3023     zero_target = (arg->sync == MIRROR_SYNC_MODE_FULL &&
3024                    (arg->mode == NEW_IMAGE_MODE_EXISTING ||
3025                     !bdrv_has_zero_init(target_bs)));
3026     bdrv_graph_rdunlock_main_loop();
3027 
3028 
3029     ret = bdrv_try_change_aio_context(target_bs, aio_context, NULL, errp);
3030     if (ret < 0) {
3031         bdrv_unref(target_bs);
3032         return;
3033     }
3034 
3035     blockdev_mirror_common(arg->job_id, bs, target_bs,
3036                            arg->replaces, arg->sync,
3037                            backing_mode, zero_target,
3038                            arg->has_speed, arg->speed,
3039                            arg->has_granularity, arg->granularity,
3040                            arg->has_buf_size, arg->buf_size,
3041                            arg->has_on_source_error, arg->on_source_error,
3042                            arg->has_on_target_error, arg->on_target_error,
3043                            arg->has_unmap, arg->unmap,
3044                            NULL,
3045                            arg->has_copy_mode, arg->copy_mode,
3046                            arg->has_auto_finalize, arg->auto_finalize,
3047                            arg->has_auto_dismiss, arg->auto_dismiss,
3048                            errp);
3049     bdrv_unref(target_bs);
3050 }
3051 
3052 void qmp_blockdev_mirror(const char *job_id,
3053                          const char *device, const char *target,
3054                          const char *replaces,
3055                          MirrorSyncMode sync,
3056                          bool has_speed, int64_t speed,
3057                          bool has_granularity, uint32_t granularity,
3058                          bool has_buf_size, int64_t buf_size,
3059                          bool has_on_source_error,
3060                          BlockdevOnError on_source_error,
3061                          bool has_on_target_error,
3062                          BlockdevOnError on_target_error,
3063                          const char *filter_node_name,
3064                          bool has_copy_mode, MirrorCopyMode copy_mode,
3065                          bool has_auto_finalize, bool auto_finalize,
3066                          bool has_auto_dismiss, bool auto_dismiss,
3067                          Error **errp)
3068 {
3069     BlockDriverState *bs;
3070     BlockDriverState *target_bs;
3071     AioContext *aio_context;
3072     BlockMirrorBackingMode backing_mode = MIRROR_LEAVE_BACKING_CHAIN;
3073     bool zero_target;
3074     int ret;
3075 
3076     bs = qmp_get_root_bs(device, errp);
3077     if (!bs) {
3078         return;
3079     }
3080 
3081     target_bs = bdrv_lookup_bs(target, target, errp);
3082     if (!target_bs) {
3083         return;
3084     }
3085 
3086     zero_target = (sync == MIRROR_SYNC_MODE_FULL);
3087 
3088     aio_context = bdrv_get_aio_context(bs);
3089 
3090     ret = bdrv_try_change_aio_context(target_bs, aio_context, NULL, errp);
3091     if (ret < 0) {
3092         return;
3093     }
3094 
3095     blockdev_mirror_common(job_id, bs, target_bs,
3096                            replaces, sync, backing_mode,
3097                            zero_target, has_speed, speed,
3098                            has_granularity, granularity,
3099                            has_buf_size, buf_size,
3100                            has_on_source_error, on_source_error,
3101                            has_on_target_error, on_target_error,
3102                            true, true, filter_node_name,
3103                            has_copy_mode, copy_mode,
3104                            has_auto_finalize, auto_finalize,
3105                            has_auto_dismiss, auto_dismiss,
3106                            errp);
3107 }
3108 
3109 /*
3110  * Get a block job using its ID. Called with job_mutex held.
3111  */
3112 static BlockJob *find_block_job_locked(const char *id, Error **errp)
3113 {
3114     BlockJob *job;
3115 
3116     assert(id != NULL);
3117 
3118     job = block_job_get_locked(id);
3119 
3120     if (!job) {
3121         error_set(errp, ERROR_CLASS_DEVICE_NOT_ACTIVE,
3122                   "Block job '%s' not found", id);
3123         return NULL;
3124     }
3125 
3126     return job;
3127 }
3128 
3129 void qmp_block_job_set_speed(const char *device, int64_t speed, Error **errp)
3130 {
3131     BlockJob *job;
3132 
3133     JOB_LOCK_GUARD();
3134     job = find_block_job_locked(device, errp);
3135 
3136     if (!job) {
3137         return;
3138     }
3139 
3140     block_job_set_speed_locked(job, speed, errp);
3141 }
3142 
3143 void qmp_block_job_cancel(const char *device,
3144                           bool has_force, bool force, Error **errp)
3145 {
3146     BlockJob *job;
3147 
3148     JOB_LOCK_GUARD();
3149     job = find_block_job_locked(device, errp);
3150 
3151     if (!job) {
3152         return;
3153     }
3154 
3155     if (!has_force) {
3156         force = false;
3157     }
3158 
3159     if (job_user_paused_locked(&job->job) && !force) {
3160         error_setg(errp, "The block job for device '%s' is currently paused",
3161                    device);
3162         return;
3163     }
3164 
3165     trace_qmp_block_job_cancel(job);
3166     job_user_cancel_locked(&job->job, force, errp);
3167 }
3168 
3169 void qmp_block_job_pause(const char *device, Error **errp)
3170 {
3171     BlockJob *job;
3172 
3173     JOB_LOCK_GUARD();
3174     job = find_block_job_locked(device, errp);
3175 
3176     if (!job) {
3177         return;
3178     }
3179 
3180     trace_qmp_block_job_pause(job);
3181     job_user_pause_locked(&job->job, errp);
3182 }
3183 
3184 void qmp_block_job_resume(const char *device, Error **errp)
3185 {
3186     BlockJob *job;
3187 
3188     JOB_LOCK_GUARD();
3189     job = find_block_job_locked(device, errp);
3190 
3191     if (!job) {
3192         return;
3193     }
3194 
3195     trace_qmp_block_job_resume(job);
3196     job_user_resume_locked(&job->job, errp);
3197 }
3198 
3199 void qmp_block_job_complete(const char *device, Error **errp)
3200 {
3201     BlockJob *job;
3202 
3203     JOB_LOCK_GUARD();
3204     job = find_block_job_locked(device, errp);
3205 
3206     if (!job) {
3207         return;
3208     }
3209 
3210     trace_qmp_block_job_complete(job);
3211     job_complete_locked(&job->job, errp);
3212 }
3213 
3214 void qmp_block_job_finalize(const char *id, Error **errp)
3215 {
3216     BlockJob *job;
3217 
3218     JOB_LOCK_GUARD();
3219     job = find_block_job_locked(id, errp);
3220 
3221     if (!job) {
3222         return;
3223     }
3224 
3225     trace_qmp_block_job_finalize(job);
3226     job_ref_locked(&job->job);
3227     job_finalize_locked(&job->job, errp);
3228 
3229     job_unref_locked(&job->job);
3230 }
3231 
3232 void qmp_block_job_dismiss(const char *id, Error **errp)
3233 {
3234     BlockJob *bjob;
3235     Job *job;
3236 
3237     JOB_LOCK_GUARD();
3238     bjob = find_block_job_locked(id, errp);
3239 
3240     if (!bjob) {
3241         return;
3242     }
3243 
3244     trace_qmp_block_job_dismiss(bjob);
3245     job = &bjob->job;
3246     job_dismiss_locked(&job, errp);
3247 }
3248 
3249 void qmp_block_job_change(BlockJobChangeOptions *opts, Error **errp)
3250 {
3251     BlockJob *job;
3252 
3253     JOB_LOCK_GUARD();
3254     job = find_block_job_locked(opts->id, errp);
3255 
3256     if (!job) {
3257         return;
3258     }
3259 
3260     block_job_change_locked(job, opts, errp);
3261 }
3262 
3263 void qmp_change_backing_file(const char *device,
3264                              const char *image_node_name,
3265                              const char *backing_file,
3266                              Error **errp)
3267 {
3268     BlockDriverState *bs = NULL;
3269     BlockDriverState *image_bs = NULL;
3270     Error *local_err = NULL;
3271     bool ro;
3272     int ret;
3273 
3274     bs = qmp_get_root_bs(device, errp);
3275     if (!bs) {
3276         return;
3277     }
3278 
3279     bdrv_graph_rdlock_main_loop();
3280 
3281     image_bs = bdrv_lookup_bs(NULL, image_node_name, &local_err);
3282     if (local_err) {
3283         error_propagate(errp, local_err);
3284         goto out_rdlock;
3285     }
3286 
3287     if (!image_bs) {
3288         error_setg(errp, "image file not found");
3289         goto out_rdlock;
3290     }
3291 
3292     if (bdrv_find_base(image_bs) == image_bs) {
3293         error_setg(errp, "not allowing backing file change on an image "
3294                          "without a backing file");
3295         goto out_rdlock;
3296     }
3297 
3298     /* even though we are not necessarily operating on bs, we need it to
3299      * determine if block ops are currently prohibited on the chain */
3300     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_CHANGE, errp)) {
3301         goto out_rdlock;
3302     }
3303 
3304     /* final sanity check */
3305     if (!bdrv_chain_contains(bs, image_bs)) {
3306         error_setg(errp, "'%s' and image file are not in the same chain",
3307                    device);
3308         goto out_rdlock;
3309     }
3310     bdrv_graph_rdunlock_main_loop();
3311 
3312     /* if not r/w, reopen to make r/w */
3313     ro = bdrv_is_read_only(image_bs);
3314 
3315     if (ro) {
3316         if (bdrv_reopen_set_read_only(image_bs, false, errp) != 0) {
3317             return;
3318         }
3319     }
3320 
3321     ret = bdrv_change_backing_file(image_bs, backing_file,
3322                                    image_bs->drv ? image_bs->drv->format_name : "",
3323                                    false);
3324 
3325     if (ret < 0) {
3326         error_setg_errno(errp, -ret, "Could not change backing file to '%s'",
3327                          backing_file);
3328         /* don't exit here, so we can try to restore open flags if
3329          * appropriate */
3330     }
3331 
3332     if (ro) {
3333         bdrv_reopen_set_read_only(image_bs, true, errp);
3334     }
3335     return;
3336 
3337 out_rdlock:
3338     bdrv_graph_rdunlock_main_loop();
3339 }
3340 
3341 void qmp_blockdev_add(BlockdevOptions *options, Error **errp)
3342 {
3343     BlockDriverState *bs;
3344     QObject *obj;
3345     Visitor *v = qobject_output_visitor_new(&obj);
3346     QDict *qdict;
3347 
3348     visit_type_BlockdevOptions(v, NULL, &options, &error_abort);
3349     visit_complete(v, &obj);
3350     qdict = qobject_to(QDict, obj);
3351 
3352     qdict_flatten(qdict);
3353 
3354     if (!qdict_get_try_str(qdict, "node-name")) {
3355         error_setg(errp, "'node-name' must be specified for the root node");
3356         goto fail;
3357     }
3358 
3359     bs = bds_tree_init(qdict, errp);
3360     if (!bs) {
3361         goto fail;
3362     }
3363 
3364     bdrv_set_monitor_owned(bs);
3365 
3366 fail:
3367     visit_free(v);
3368 }
3369 
3370 void qmp_blockdev_reopen(BlockdevOptionsList *reopen_list, Error **errp)
3371 {
3372     BlockReopenQueue *queue = NULL;
3373 
3374     /* Add each one of the BDS that we want to reopen to the queue */
3375     for (; reopen_list != NULL; reopen_list = reopen_list->next) {
3376         BlockdevOptions *options = reopen_list->value;
3377         BlockDriverState *bs;
3378         QObject *obj;
3379         Visitor *v;
3380         QDict *qdict;
3381 
3382         /* Check for the selected node name */
3383         if (!options->node_name) {
3384             error_setg(errp, "node-name not specified");
3385             goto fail;
3386         }
3387 
3388         bs = bdrv_find_node(options->node_name);
3389         if (!bs) {
3390             error_setg(errp, "Failed to find node with node-name='%s'",
3391                        options->node_name);
3392             goto fail;
3393         }
3394 
3395         /* Put all options in a QDict and flatten it */
3396         v = qobject_output_visitor_new(&obj);
3397         visit_type_BlockdevOptions(v, NULL, &options, &error_abort);
3398         visit_complete(v, &obj);
3399         visit_free(v);
3400 
3401         qdict = qobject_to(QDict, obj);
3402 
3403         qdict_flatten(qdict);
3404 
3405         queue = bdrv_reopen_queue(queue, bs, qdict, false);
3406     }
3407 
3408     /* Perform the reopen operation */
3409     bdrv_reopen_multiple(queue, errp);
3410     queue = NULL;
3411 
3412 fail:
3413     bdrv_reopen_queue_free(queue);
3414 }
3415 
3416 void qmp_blockdev_del(const char *node_name, Error **errp)
3417 {
3418     BlockDriverState *bs;
3419 
3420     GLOBAL_STATE_CODE();
3421     GRAPH_RDLOCK_GUARD_MAINLOOP();
3422 
3423     bs = bdrv_find_node(node_name);
3424     if (!bs) {
3425         error_setg(errp, "Failed to find node with node-name='%s'", node_name);
3426         return;
3427     }
3428     if (bdrv_has_blk(bs)) {
3429         error_setg(errp, "Node %s is in use", node_name);
3430         return;
3431     }
3432 
3433     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_DRIVE_DEL, errp)) {
3434         return;
3435     }
3436 
3437     if (!QTAILQ_IN_USE(bs, monitor_list)) {
3438         error_setg(errp, "Node %s is not owned by the monitor",
3439                    bs->node_name);
3440         return;
3441     }
3442 
3443     if (bs->refcnt > 1) {
3444         error_setg(errp, "Block device %s is in use",
3445                    bdrv_get_device_or_node_name(bs));
3446         return;
3447     }
3448 
3449     QTAILQ_REMOVE(&monitor_bdrv_states, bs, monitor_list);
3450     bdrv_unref(bs);
3451 }
3452 
3453 static BdrvChild * GRAPH_RDLOCK
3454 bdrv_find_child(BlockDriverState *parent_bs, const char *child_name)
3455 {
3456     BdrvChild *child;
3457 
3458     QLIST_FOREACH(child, &parent_bs->children, next) {
3459         if (strcmp(child->name, child_name) == 0) {
3460             return child;
3461         }
3462     }
3463 
3464     return NULL;
3465 }
3466 
3467 void qmp_x_blockdev_change(const char *parent, const char *child,
3468                            const char *node, Error **errp)
3469 {
3470     BlockDriverState *parent_bs, *new_bs = NULL;
3471     BdrvChild *p_child;
3472 
3473     bdrv_graph_wrlock();
3474 
3475     parent_bs = bdrv_lookup_bs(parent, parent, errp);
3476     if (!parent_bs) {
3477         goto out;
3478     }
3479 
3480     if (!child == !node) {
3481         if (child) {
3482             error_setg(errp, "The parameters child and node are in conflict");
3483         } else {
3484             error_setg(errp, "Either child or node must be specified");
3485         }
3486         goto out;
3487     }
3488 
3489     if (child) {
3490         p_child = bdrv_find_child(parent_bs, child);
3491         if (!p_child) {
3492             error_setg(errp, "Node '%s' does not have child '%s'",
3493                        parent, child);
3494             goto out;
3495         }
3496         bdrv_del_child(parent_bs, p_child, errp);
3497     }
3498 
3499     if (node) {
3500         new_bs = bdrv_find_node(node);
3501         if (!new_bs) {
3502             error_setg(errp, "Node '%s' not found", node);
3503             goto out;
3504         }
3505         bdrv_add_child(parent_bs, new_bs, errp);
3506     }
3507 
3508 out:
3509     bdrv_graph_wrunlock();
3510 }
3511 
3512 BlockJobInfoList *qmp_query_block_jobs(Error **errp)
3513 {
3514     BlockJobInfoList *head = NULL, **tail = &head;
3515     BlockJob *job;
3516 
3517     JOB_LOCK_GUARD();
3518 
3519     for (job = block_job_next_locked(NULL); job;
3520          job = block_job_next_locked(job)) {
3521         BlockJobInfo *value;
3522 
3523         if (block_job_is_internal(job)) {
3524             continue;
3525         }
3526         value = block_job_query_locked(job, errp);
3527         if (!value) {
3528             qapi_free_BlockJobInfoList(head);
3529             return NULL;
3530         }
3531         QAPI_LIST_APPEND(tail, value);
3532     }
3533 
3534     return head;
3535 }
3536 
3537 void qmp_x_blockdev_set_iothread(const char *node_name, StrOrNull *iothread,
3538                                  bool has_force, bool force, Error **errp)
3539 {
3540     AioContext *new_context;
3541     BlockDriverState *bs;
3542 
3543     GRAPH_RDLOCK_GUARD_MAINLOOP();
3544 
3545     bs = bdrv_find_node(node_name);
3546     if (!bs) {
3547         error_setg(errp, "Failed to find node with node-name='%s'", node_name);
3548         return;
3549     }
3550 
3551     /* Protects against accidents. */
3552     if (!(has_force && force) && bdrv_has_blk(bs)) {
3553         error_setg(errp, "Node %s is associated with a BlockBackend and could "
3554                          "be in use (use force=true to override this check)",
3555                          node_name);
3556         return;
3557     }
3558 
3559     if (iothread->type == QTYPE_QSTRING) {
3560         IOThread *obj = iothread_by_id(iothread->u.s);
3561         if (!obj) {
3562             error_setg(errp, "Cannot find iothread %s", iothread->u.s);
3563             return;
3564         }
3565 
3566         new_context = iothread_get_aio_context(obj);
3567     } else {
3568         new_context = qemu_get_aio_context();
3569     }
3570 
3571     bdrv_try_change_aio_context(bs, new_context, NULL, errp);
3572 }
3573 
3574 QemuOptsList qemu_common_drive_opts = {
3575     .name = "drive",
3576     .head = QTAILQ_HEAD_INITIALIZER(qemu_common_drive_opts.head),
3577     .desc = {
3578         {
3579             .name = "snapshot",
3580             .type = QEMU_OPT_BOOL,
3581             .help = "enable/disable snapshot mode",
3582         },{
3583             .name = "aio",
3584             .type = QEMU_OPT_STRING,
3585             .help = "host AIO implementation (threads, native, io_uring)",
3586         },{
3587             .name = BDRV_OPT_CACHE_WB,
3588             .type = QEMU_OPT_BOOL,
3589             .help = "Enable writeback mode",
3590         },{
3591             .name = "format",
3592             .type = QEMU_OPT_STRING,
3593             .help = "disk format (raw, qcow2, ...)",
3594         },{
3595             .name = "rerror",
3596             .type = QEMU_OPT_STRING,
3597             .help = "read error action",
3598         },{
3599             .name = "werror",
3600             .type = QEMU_OPT_STRING,
3601             .help = "write error action",
3602         },{
3603             .name = BDRV_OPT_READ_ONLY,
3604             .type = QEMU_OPT_BOOL,
3605             .help = "open drive file as read-only",
3606         },
3607 
3608         THROTTLE_OPTS,
3609 
3610         {
3611             .name = "throttling.group",
3612             .type = QEMU_OPT_STRING,
3613             .help = "name of the block throttling group",
3614         },{
3615             .name = "copy-on-read",
3616             .type = QEMU_OPT_BOOL,
3617             .help = "copy read data from backing file into image file",
3618         },{
3619             .name = "detect-zeroes",
3620             .type = QEMU_OPT_STRING,
3621             .help = "try to optimize zero writes (off, on, unmap)",
3622         },{
3623             .name = "stats-account-invalid",
3624             .type = QEMU_OPT_BOOL,
3625             .help = "whether to account for invalid I/O operations "
3626                     "in the statistics",
3627         },{
3628             .name = "stats-account-failed",
3629             .type = QEMU_OPT_BOOL,
3630             .help = "whether to account for failed I/O operations "
3631                     "in the statistics",
3632         },
3633         { /* end of list */ }
3634     },
3635 };
3636 
3637 QemuOptsList qemu_drive_opts = {
3638     .name = "drive",
3639     .head = QTAILQ_HEAD_INITIALIZER(qemu_drive_opts.head),
3640     .desc = {
3641         /*
3642          * no elements => accept any params
3643          * validation will happen later
3644          */
3645         { /* end of list */ }
3646     },
3647 };
3648