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