xref: /openbmc/qemu/blockdev.c (revision 4a4ff4c5)
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/throttle-groups.h"
39 #include "monitor/monitor.h"
40 #include "qemu/error-report.h"
41 #include "qemu/option.h"
42 #include "qemu/config-file.h"
43 #include "qapi/qapi-commands-block.h"
44 #include "qapi/qapi-commands-transaction.h"
45 #include "qapi/qapi-visit-block-core.h"
46 #include "qapi/qmp/qdict.h"
47 #include "qapi/qmp/qnum.h"
48 #include "qapi/qmp/qstring.h"
49 #include "qapi/error.h"
50 #include "qapi/qmp/qerror.h"
51 #include "qapi/qmp/qlist.h"
52 #include "qapi/qobject-output-visitor.h"
53 #include "sysemu/sysemu.h"
54 #include "sysemu/iothread.h"
55 #include "block/block_int.h"
56 #include "block/trace.h"
57 #include "sysemu/arch_init.h"
58 #include "sysemu/qtest.h"
59 #include "qemu/cutils.h"
60 #include "qemu/help_option.h"
61 #include "qemu/throttle-options.h"
62 
63 static QTAILQ_HEAD(, BlockDriverState) monitor_bdrv_states =
64     QTAILQ_HEAD_INITIALIZER(monitor_bdrv_states);
65 
66 static int do_open_tray(const char *blk_name, const char *qdev_id,
67                         bool force, Error **errp);
68 static void blockdev_remove_medium(bool has_device, const char *device,
69                                    bool has_id, const char *id, Error **errp);
70 static void blockdev_insert_medium(bool has_device, const char *device,
71                                    bool has_id, const char *id,
72                                    const char *node_name, Error **errp);
73 
74 static const char *const if_name[IF_COUNT] = {
75     [IF_NONE] = "none",
76     [IF_IDE] = "ide",
77     [IF_SCSI] = "scsi",
78     [IF_FLOPPY] = "floppy",
79     [IF_PFLASH] = "pflash",
80     [IF_MTD] = "mtd",
81     [IF_SD] = "sd",
82     [IF_VIRTIO] = "virtio",
83     [IF_XEN] = "xen",
84 };
85 
86 static int if_max_devs[IF_COUNT] = {
87     /*
88      * Do not change these numbers!  They govern how drive option
89      * index maps to unit and bus.  That mapping is ABI.
90      *
91      * All controllers used to implement if=T drives need to support
92      * if_max_devs[T] units, for any T with if_max_devs[T] != 0.
93      * Otherwise, some index values map to "impossible" bus, unit
94      * values.
95      *
96      * For instance, if you change [IF_SCSI] to 255, -drive
97      * if=scsi,index=12 no longer means bus=1,unit=5, but
98      * bus=0,unit=12.  With an lsi53c895a controller (7 units max),
99      * the drive can't be set up.  Regression.
100      */
101     [IF_IDE] = 2,
102     [IF_SCSI] = 7,
103 };
104 
105 /**
106  * Boards may call this to offer board-by-board overrides
107  * of the default, global values.
108  */
109 void override_max_devs(BlockInterfaceType type, int max_devs)
110 {
111     BlockBackend *blk;
112     DriveInfo *dinfo;
113 
114     if (max_devs <= 0) {
115         return;
116     }
117 
118     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
119         dinfo = blk_legacy_dinfo(blk);
120         if (dinfo->type == type) {
121             fprintf(stderr, "Cannot override units-per-bus property of"
122                     " the %s interface, because a drive of that type has"
123                     " already been added.\n", if_name[type]);
124             g_assert_not_reached();
125         }
126     }
127 
128     if_max_devs[type] = max_devs;
129 }
130 
131 /*
132  * We automatically delete the drive when a device using it gets
133  * unplugged.  Questionable feature, but we can't just drop it.
134  * Device models call blockdev_mark_auto_del() to schedule the
135  * automatic deletion, and generic qdev code calls blockdev_auto_del()
136  * when deletion is actually safe.
137  */
138 void blockdev_mark_auto_del(BlockBackend *blk)
139 {
140     DriveInfo *dinfo = blk_legacy_dinfo(blk);
141     BlockDriverState *bs = blk_bs(blk);
142     AioContext *aio_context;
143 
144     if (!dinfo) {
145         return;
146     }
147 
148     if (bs) {
149         aio_context = bdrv_get_aio_context(bs);
150         aio_context_acquire(aio_context);
151 
152         if (bs->job) {
153             block_job_cancel(bs->job, false);
154         }
155 
156         aio_context_release(aio_context);
157     }
158 
159     dinfo->auto_del = 1;
160 }
161 
162 void blockdev_auto_del(BlockBackend *blk)
163 {
164     DriveInfo *dinfo = blk_legacy_dinfo(blk);
165 
166     if (dinfo && dinfo->auto_del) {
167         monitor_remove_blk(blk);
168         blk_unref(blk);
169     }
170 }
171 
172 /**
173  * Returns the current mapping of how many units per bus
174  * a particular interface can support.
175  *
176  *  A positive integer indicates n units per bus.
177  *  0 implies the mapping has not been established.
178  * -1 indicates an invalid BlockInterfaceType was given.
179  */
180 int drive_get_max_devs(BlockInterfaceType type)
181 {
182     if (type >= IF_IDE && type < IF_COUNT) {
183         return if_max_devs[type];
184     }
185 
186     return -1;
187 }
188 
189 static int drive_index_to_bus_id(BlockInterfaceType type, int index)
190 {
191     int max_devs = if_max_devs[type];
192     return max_devs ? index / max_devs : 0;
193 }
194 
195 static int drive_index_to_unit_id(BlockInterfaceType type, int index)
196 {
197     int max_devs = if_max_devs[type];
198     return max_devs ? index % max_devs : index;
199 }
200 
201 QemuOpts *drive_def(const char *optstr)
202 {
203     return qemu_opts_parse_noisily(qemu_find_opts("drive"), optstr, false);
204 }
205 
206 QemuOpts *drive_add(BlockInterfaceType type, int index, const char *file,
207                     const char *optstr)
208 {
209     QemuOpts *opts;
210 
211     opts = drive_def(optstr);
212     if (!opts) {
213         return NULL;
214     }
215     if (type != IF_DEFAULT) {
216         qemu_opt_set(opts, "if", if_name[type], &error_abort);
217     }
218     if (index >= 0) {
219         qemu_opt_set_number(opts, "index", index, &error_abort);
220     }
221     if (file)
222         qemu_opt_set(opts, "file", file, &error_abort);
223     return opts;
224 }
225 
226 DriveInfo *drive_get(BlockInterfaceType type, int bus, int unit)
227 {
228     BlockBackend *blk;
229     DriveInfo *dinfo;
230 
231     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
232         dinfo = blk_legacy_dinfo(blk);
233         if (dinfo && dinfo->type == type
234             && dinfo->bus == bus && dinfo->unit == unit) {
235             return dinfo;
236         }
237     }
238 
239     return NULL;
240 }
241 
242 void drive_check_orphaned(void)
243 {
244     BlockBackend *blk;
245     DriveInfo *dinfo;
246     Location loc;
247     bool orphans = false;
248 
249     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
250         dinfo = blk_legacy_dinfo(blk);
251         if (!blk_get_attached_dev(blk) && !dinfo->is_default &&
252             dinfo->type != IF_NONE) {
253             loc_push_none(&loc);
254             qemu_opts_loc_restore(dinfo->opts);
255             error_report("machine type does not support"
256                          " if=%s,bus=%d,unit=%d",
257                          if_name[dinfo->type], dinfo->bus, dinfo->unit);
258             loc_pop(&loc);
259             orphans = true;
260         }
261     }
262 
263     if (orphans) {
264         exit(1);
265     }
266 }
267 
268 DriveInfo *drive_get_by_index(BlockInterfaceType type, int index)
269 {
270     return drive_get(type,
271                      drive_index_to_bus_id(type, index),
272                      drive_index_to_unit_id(type, index));
273 }
274 
275 int drive_get_max_bus(BlockInterfaceType type)
276 {
277     int max_bus;
278     BlockBackend *blk;
279     DriveInfo *dinfo;
280 
281     max_bus = -1;
282     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
283         dinfo = blk_legacy_dinfo(blk);
284         if (dinfo && dinfo->type == type && dinfo->bus > max_bus) {
285             max_bus = dinfo->bus;
286         }
287     }
288     return max_bus;
289 }
290 
291 /* Get a block device.  This should only be used for single-drive devices
292    (e.g. SD/Floppy/MTD).  Multi-disk devices (scsi/ide) should use the
293    appropriate bus.  */
294 DriveInfo *drive_get_next(BlockInterfaceType type)
295 {
296     static int next_block_unit[IF_COUNT];
297 
298     return drive_get(type, 0, next_block_unit[type]++);
299 }
300 
301 static void bdrv_format_print(void *opaque, const char *name)
302 {
303     error_printf(" %s", name);
304 }
305 
306 typedef struct {
307     QEMUBH *bh;
308     BlockDriverState *bs;
309 } BDRVPutRefBH;
310 
311 static int parse_block_error_action(const char *buf, bool is_read, Error **errp)
312 {
313     if (!strcmp(buf, "ignore")) {
314         return BLOCKDEV_ON_ERROR_IGNORE;
315     } else if (!is_read && !strcmp(buf, "enospc")) {
316         return BLOCKDEV_ON_ERROR_ENOSPC;
317     } else if (!strcmp(buf, "stop")) {
318         return BLOCKDEV_ON_ERROR_STOP;
319     } else if (!strcmp(buf, "report")) {
320         return BLOCKDEV_ON_ERROR_REPORT;
321     } else {
322         error_setg(errp, "'%s' invalid %s error action",
323                    buf, is_read ? "read" : "write");
324         return -1;
325     }
326 }
327 
328 static bool parse_stats_intervals(BlockAcctStats *stats, QList *intervals,
329                                   Error **errp)
330 {
331     const QListEntry *entry;
332     for (entry = qlist_first(intervals); entry; entry = qlist_next(entry)) {
333         switch (qobject_type(entry->value)) {
334 
335         case QTYPE_QSTRING: {
336             unsigned long long length;
337             const char *str = qstring_get_str(qobject_to(QString,
338                                                          entry->value));
339             if (parse_uint_full(str, &length, 10) == 0 &&
340                 length > 0 && length <= UINT_MAX) {
341                 block_acct_add_interval(stats, (unsigned) length);
342             } else {
343                 error_setg(errp, "Invalid interval length: %s", str);
344                 return false;
345             }
346             break;
347         }
348 
349         case QTYPE_QNUM: {
350             int64_t length = qnum_get_int(qobject_to(QNum, entry->value));
351 
352             if (length > 0 && length <= UINT_MAX) {
353                 block_acct_add_interval(stats, (unsigned) length);
354             } else {
355                 error_setg(errp, "Invalid interval length: %" PRId64, length);
356                 return false;
357             }
358             break;
359         }
360 
361         default:
362             error_setg(errp, "The specification of stats-intervals is invalid");
363             return false;
364         }
365     }
366     return true;
367 }
368 
369 typedef enum { MEDIA_DISK, MEDIA_CDROM } DriveMediaType;
370 
371 /* All parameters but @opts are optional and may be set to NULL. */
372 static void extract_common_blockdev_options(QemuOpts *opts, int *bdrv_flags,
373     const char **throttling_group, ThrottleConfig *throttle_cfg,
374     BlockdevDetectZeroesOptions *detect_zeroes, Error **errp)
375 {
376     Error *local_error = NULL;
377     const char *aio;
378 
379     if (bdrv_flags) {
380         if (qemu_opt_get_bool(opts, "copy-on-read", false)) {
381             *bdrv_flags |= BDRV_O_COPY_ON_READ;
382         }
383 
384         if ((aio = qemu_opt_get(opts, "aio")) != NULL) {
385             if (!strcmp(aio, "native")) {
386                 *bdrv_flags |= BDRV_O_NATIVE_AIO;
387             } else if (!strcmp(aio, "threads")) {
388                 /* this is the default */
389             } else {
390                error_setg(errp, "invalid aio option");
391                return;
392             }
393         }
394     }
395 
396     /* disk I/O throttling */
397     if (throttling_group) {
398         *throttling_group = qemu_opt_get(opts, "throttling.group");
399     }
400 
401     if (throttle_cfg) {
402         throttle_config_init(throttle_cfg);
403         throttle_cfg->buckets[THROTTLE_BPS_TOTAL].avg =
404             qemu_opt_get_number(opts, "throttling.bps-total", 0);
405         throttle_cfg->buckets[THROTTLE_BPS_READ].avg  =
406             qemu_opt_get_number(opts, "throttling.bps-read", 0);
407         throttle_cfg->buckets[THROTTLE_BPS_WRITE].avg =
408             qemu_opt_get_number(opts, "throttling.bps-write", 0);
409         throttle_cfg->buckets[THROTTLE_OPS_TOTAL].avg =
410             qemu_opt_get_number(opts, "throttling.iops-total", 0);
411         throttle_cfg->buckets[THROTTLE_OPS_READ].avg =
412             qemu_opt_get_number(opts, "throttling.iops-read", 0);
413         throttle_cfg->buckets[THROTTLE_OPS_WRITE].avg =
414             qemu_opt_get_number(opts, "throttling.iops-write", 0);
415 
416         throttle_cfg->buckets[THROTTLE_BPS_TOTAL].max =
417             qemu_opt_get_number(opts, "throttling.bps-total-max", 0);
418         throttle_cfg->buckets[THROTTLE_BPS_READ].max  =
419             qemu_opt_get_number(opts, "throttling.bps-read-max", 0);
420         throttle_cfg->buckets[THROTTLE_BPS_WRITE].max =
421             qemu_opt_get_number(opts, "throttling.bps-write-max", 0);
422         throttle_cfg->buckets[THROTTLE_OPS_TOTAL].max =
423             qemu_opt_get_number(opts, "throttling.iops-total-max", 0);
424         throttle_cfg->buckets[THROTTLE_OPS_READ].max =
425             qemu_opt_get_number(opts, "throttling.iops-read-max", 0);
426         throttle_cfg->buckets[THROTTLE_OPS_WRITE].max =
427             qemu_opt_get_number(opts, "throttling.iops-write-max", 0);
428 
429         throttle_cfg->buckets[THROTTLE_BPS_TOTAL].burst_length =
430             qemu_opt_get_number(opts, "throttling.bps-total-max-length", 1);
431         throttle_cfg->buckets[THROTTLE_BPS_READ].burst_length  =
432             qemu_opt_get_number(opts, "throttling.bps-read-max-length", 1);
433         throttle_cfg->buckets[THROTTLE_BPS_WRITE].burst_length =
434             qemu_opt_get_number(opts, "throttling.bps-write-max-length", 1);
435         throttle_cfg->buckets[THROTTLE_OPS_TOTAL].burst_length =
436             qemu_opt_get_number(opts, "throttling.iops-total-max-length", 1);
437         throttle_cfg->buckets[THROTTLE_OPS_READ].burst_length =
438             qemu_opt_get_number(opts, "throttling.iops-read-max-length", 1);
439         throttle_cfg->buckets[THROTTLE_OPS_WRITE].burst_length =
440             qemu_opt_get_number(opts, "throttling.iops-write-max-length", 1);
441 
442         throttle_cfg->op_size =
443             qemu_opt_get_number(opts, "throttling.iops-size", 0);
444 
445         if (!throttle_is_valid(throttle_cfg, errp)) {
446             return;
447         }
448     }
449 
450     if (detect_zeroes) {
451         *detect_zeroes =
452             qapi_enum_parse(&BlockdevDetectZeroesOptions_lookup,
453                             qemu_opt_get(opts, "detect-zeroes"),
454                             BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF,
455                             &local_error);
456         if (local_error) {
457             error_propagate(errp, local_error);
458             return;
459         }
460     }
461 }
462 
463 /* Takes the ownership of bs_opts */
464 static BlockBackend *blockdev_init(const char *file, QDict *bs_opts,
465                                    Error **errp)
466 {
467     const char *buf;
468     int bdrv_flags = 0;
469     int on_read_error, on_write_error;
470     bool account_invalid, account_failed;
471     bool writethrough, read_only;
472     BlockBackend *blk;
473     BlockDriverState *bs;
474     ThrottleConfig cfg;
475     int snapshot = 0;
476     Error *error = NULL;
477     QemuOpts *opts;
478     QDict *interval_dict = NULL;
479     QList *interval_list = NULL;
480     const char *id;
481     BlockdevDetectZeroesOptions detect_zeroes =
482         BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF;
483     const char *throttling_group = NULL;
484 
485     /* Check common options by copying from bs_opts to opts, all other options
486      * stay in bs_opts for processing by bdrv_open(). */
487     id = qdict_get_try_str(bs_opts, "id");
488     opts = qemu_opts_create(&qemu_common_drive_opts, id, 1, &error);
489     if (error) {
490         error_propagate(errp, error);
491         goto err_no_opts;
492     }
493 
494     qemu_opts_absorb_qdict(opts, bs_opts, &error);
495     if (error) {
496         error_propagate(errp, error);
497         goto early_err;
498     }
499 
500     if (id) {
501         qdict_del(bs_opts, "id");
502     }
503 
504     /* extract parameters */
505     snapshot = qemu_opt_get_bool(opts, "snapshot", 0);
506 
507     account_invalid = qemu_opt_get_bool(opts, "stats-account-invalid", true);
508     account_failed = qemu_opt_get_bool(opts, "stats-account-failed", true);
509 
510     writethrough = !qemu_opt_get_bool(opts, BDRV_OPT_CACHE_WB, true);
511 
512     id = qemu_opts_id(opts);
513 
514     qdict_extract_subqdict(bs_opts, &interval_dict, "stats-intervals.");
515     qdict_array_split(interval_dict, &interval_list);
516 
517     if (qdict_size(interval_dict) != 0) {
518         error_setg(errp, "Invalid option stats-intervals.%s",
519                    qdict_first(interval_dict)->key);
520         goto early_err;
521     }
522 
523     extract_common_blockdev_options(opts, &bdrv_flags, &throttling_group, &cfg,
524                                     &detect_zeroes, &error);
525     if (error) {
526         error_propagate(errp, error);
527         goto early_err;
528     }
529 
530     if ((buf = qemu_opt_get(opts, "format")) != NULL) {
531         if (is_help_option(buf)) {
532             error_printf("Supported formats:");
533             bdrv_iterate_format(bdrv_format_print, NULL);
534             error_printf("\n");
535             goto early_err;
536         }
537 
538         if (qdict_haskey(bs_opts, "driver")) {
539             error_setg(errp, "Cannot specify both 'driver' and 'format'");
540             goto early_err;
541         }
542         qdict_put_str(bs_opts, "driver", buf);
543     }
544 
545     on_write_error = BLOCKDEV_ON_ERROR_ENOSPC;
546     if ((buf = qemu_opt_get(opts, "werror")) != NULL) {
547         on_write_error = parse_block_error_action(buf, 0, &error);
548         if (error) {
549             error_propagate(errp, error);
550             goto early_err;
551         }
552     }
553 
554     on_read_error = BLOCKDEV_ON_ERROR_REPORT;
555     if ((buf = qemu_opt_get(opts, "rerror")) != NULL) {
556         on_read_error = parse_block_error_action(buf, 1, &error);
557         if (error) {
558             error_propagate(errp, error);
559             goto early_err;
560         }
561     }
562 
563     if (snapshot) {
564         bdrv_flags |= BDRV_O_SNAPSHOT;
565     }
566 
567     read_only = qemu_opt_get_bool(opts, BDRV_OPT_READ_ONLY, false);
568 
569     /* init */
570     if ((!file || !*file) && !qdict_size(bs_opts)) {
571         BlockBackendRootState *blk_rs;
572 
573         blk = blk_new(0, BLK_PERM_ALL);
574         blk_rs = blk_get_root_state(blk);
575         blk_rs->open_flags    = bdrv_flags;
576         blk_rs->read_only     = read_only;
577         blk_rs->detect_zeroes = detect_zeroes;
578 
579         qobject_unref(bs_opts);
580     } else {
581         if (file && !*file) {
582             file = NULL;
583         }
584 
585         /* bdrv_open() defaults to the values in bdrv_flags (for compatibility
586          * with other callers) rather than what we want as the real defaults.
587          * Apply the defaults here instead. */
588         qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_DIRECT, "off");
589         qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_NO_FLUSH, "off");
590         qdict_set_default_str(bs_opts, BDRV_OPT_READ_ONLY,
591                               read_only ? "on" : "off");
592         assert((bdrv_flags & BDRV_O_CACHE_MASK) == 0);
593 
594         if (runstate_check(RUN_STATE_INMIGRATE)) {
595             bdrv_flags |= BDRV_O_INACTIVE;
596         }
597 
598         blk = blk_new_open(file, NULL, bs_opts, bdrv_flags, errp);
599         if (!blk) {
600             goto err_no_bs_opts;
601         }
602         bs = blk_bs(blk);
603 
604         bs->detect_zeroes = detect_zeroes;
605 
606         block_acct_setup(blk_get_stats(blk), account_invalid, account_failed);
607 
608         if (!parse_stats_intervals(blk_get_stats(blk), interval_list, errp)) {
609             blk_unref(blk);
610             blk = NULL;
611             goto err_no_bs_opts;
612         }
613     }
614 
615     /* disk I/O throttling */
616     if (throttle_enabled(&cfg)) {
617         if (!throttling_group) {
618             throttling_group = id;
619         }
620         blk_io_limits_enable(blk, throttling_group);
621         blk_set_io_limits(blk, &cfg);
622     }
623 
624     blk_set_enable_write_cache(blk, !writethrough);
625     blk_set_on_error(blk, on_read_error, on_write_error);
626 
627     if (!monitor_add_blk(blk, id, errp)) {
628         blk_unref(blk);
629         blk = NULL;
630         goto err_no_bs_opts;
631     }
632 
633 err_no_bs_opts:
634     qemu_opts_del(opts);
635     qobject_unref(interval_dict);
636     qobject_unref(interval_list);
637     return blk;
638 
639 early_err:
640     qemu_opts_del(opts);
641     qobject_unref(interval_dict);
642     qobject_unref(interval_list);
643 err_no_opts:
644     qobject_unref(bs_opts);
645     return NULL;
646 }
647 
648 /* Takes the ownership of bs_opts */
649 static BlockDriverState *bds_tree_init(QDict *bs_opts, Error **errp)
650 {
651     int bdrv_flags = 0;
652 
653     /* bdrv_open() defaults to the values in bdrv_flags (for compatibility
654      * with other callers) rather than what we want as the real defaults.
655      * Apply the defaults here instead. */
656     qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_DIRECT, "off");
657     qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_NO_FLUSH, "off");
658     qdict_set_default_str(bs_opts, BDRV_OPT_READ_ONLY, "off");
659 
660     if (runstate_check(RUN_STATE_INMIGRATE)) {
661         bdrv_flags |= BDRV_O_INACTIVE;
662     }
663 
664     return bdrv_open(NULL, NULL, bs_opts, bdrv_flags, errp);
665 }
666 
667 void blockdev_close_all_bdrv_states(void)
668 {
669     BlockDriverState *bs, *next_bs;
670 
671     QTAILQ_FOREACH_SAFE(bs, &monitor_bdrv_states, monitor_list, next_bs) {
672         AioContext *ctx = bdrv_get_aio_context(bs);
673 
674         aio_context_acquire(ctx);
675         bdrv_unref(bs);
676         aio_context_release(ctx);
677     }
678 }
679 
680 /* Iterates over the list of monitor-owned BlockDriverStates */
681 BlockDriverState *bdrv_next_monitor_owned(BlockDriverState *bs)
682 {
683     return bs ? QTAILQ_NEXT(bs, monitor_list)
684               : QTAILQ_FIRST(&monitor_bdrv_states);
685 }
686 
687 static void qemu_opt_rename(QemuOpts *opts, const char *from, const char *to,
688                             Error **errp)
689 {
690     const char *value;
691 
692     value = qemu_opt_get(opts, from);
693     if (value) {
694         if (qemu_opt_find(opts, to)) {
695             error_setg(errp, "'%s' and its alias '%s' can't be used at the "
696                        "same time", to, from);
697             return;
698         }
699     }
700 
701     /* rename all items in opts */
702     while ((value = qemu_opt_get(opts, from))) {
703         qemu_opt_set(opts, to, value, &error_abort);
704         qemu_opt_unset(opts, from);
705     }
706 }
707 
708 QemuOptsList qemu_legacy_drive_opts = {
709     .name = "drive",
710     .head = QTAILQ_HEAD_INITIALIZER(qemu_legacy_drive_opts.head),
711     .desc = {
712         {
713             .name = "bus",
714             .type = QEMU_OPT_NUMBER,
715             .help = "bus number",
716         },{
717             .name = "unit",
718             .type = QEMU_OPT_NUMBER,
719             .help = "unit number (i.e. lun for scsi)",
720         },{
721             .name = "index",
722             .type = QEMU_OPT_NUMBER,
723             .help = "index number",
724         },{
725             .name = "media",
726             .type = QEMU_OPT_STRING,
727             .help = "media type (disk, cdrom)",
728         },{
729             .name = "if",
730             .type = QEMU_OPT_STRING,
731             .help = "interface (ide, scsi, sd, mtd, floppy, pflash, virtio)",
732         },{
733             .name = "cyls",
734             .type = QEMU_OPT_NUMBER,
735             .help = "number of cylinders (ide disk geometry)",
736         },{
737             .name = "heads",
738             .type = QEMU_OPT_NUMBER,
739             .help = "number of heads (ide disk geometry)",
740         },{
741             .name = "secs",
742             .type = QEMU_OPT_NUMBER,
743             .help = "number of sectors (ide disk geometry)",
744         },{
745             .name = "trans",
746             .type = QEMU_OPT_STRING,
747             .help = "chs translation (auto, lba, none)",
748         },{
749             .name = "addr",
750             .type = QEMU_OPT_STRING,
751             .help = "pci address (virtio only)",
752         },{
753             .name = "serial",
754             .type = QEMU_OPT_STRING,
755             .help = "disk serial number",
756         },{
757             .name = "file",
758             .type = QEMU_OPT_STRING,
759             .help = "file name",
760         },
761 
762         /* Options that are passed on, but have special semantics with -drive */
763         {
764             .name = BDRV_OPT_READ_ONLY,
765             .type = QEMU_OPT_BOOL,
766             .help = "open drive file as read-only",
767         },{
768             .name = "rerror",
769             .type = QEMU_OPT_STRING,
770             .help = "read error action",
771         },{
772             .name = "werror",
773             .type = QEMU_OPT_STRING,
774             .help = "write error action",
775         },{
776             .name = "copy-on-read",
777             .type = QEMU_OPT_BOOL,
778             .help = "copy read data from backing file into image file",
779         },
780 
781         { /* end of list */ }
782     },
783 };
784 
785 DriveInfo *drive_new(QemuOpts *all_opts, BlockInterfaceType block_default_type)
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 cyls, heads, secs, translation;
795     int max_devs, bus_id, unit_id, index;
796     const char *devaddr;
797     const char *werror, *rerror;
798     bool read_only = false;
799     bool copy_on_read;
800     const char *serial;
801     const char *filename;
802     Error *local_err = NULL;
803     int i;
804     const char *deprecated[] = {
805         "serial", "trans", "secs", "heads", "cyls", "addr"
806     };
807 
808     /* Change legacy command line options into QMP ones */
809     static const struct {
810         const char *from;
811         const char *to;
812     } opt_renames[] = {
813         { "iops",           "throttling.iops-total" },
814         { "iops_rd",        "throttling.iops-read" },
815         { "iops_wr",        "throttling.iops-write" },
816 
817         { "bps",            "throttling.bps-total" },
818         { "bps_rd",         "throttling.bps-read" },
819         { "bps_wr",         "throttling.bps-write" },
820 
821         { "iops_max",       "throttling.iops-total-max" },
822         { "iops_rd_max",    "throttling.iops-read-max" },
823         { "iops_wr_max",    "throttling.iops-write-max" },
824 
825         { "bps_max",        "throttling.bps-total-max" },
826         { "bps_rd_max",     "throttling.bps-read-max" },
827         { "bps_wr_max",     "throttling.bps-write-max" },
828 
829         { "iops_size",      "throttling.iops-size" },
830 
831         { "group",          "throttling.group" },
832 
833         { "readonly",       BDRV_OPT_READ_ONLY },
834     };
835 
836     for (i = 0; i < ARRAY_SIZE(opt_renames); i++) {
837         qemu_opt_rename(all_opts, opt_renames[i].from, opt_renames[i].to,
838                         &local_err);
839         if (local_err) {
840             error_report_err(local_err);
841             return NULL;
842         }
843     }
844 
845     value = qemu_opt_get(all_opts, "cache");
846     if (value) {
847         int flags = 0;
848         bool writethrough;
849 
850         if (bdrv_parse_cache_mode(value, &flags, &writethrough) != 0) {
851             error_report("invalid cache option");
852             return NULL;
853         }
854 
855         /* Specific options take precedence */
856         if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_WB)) {
857             qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_WB,
858                               !writethrough, &error_abort);
859         }
860         if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_DIRECT)) {
861             qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_DIRECT,
862                               !!(flags & BDRV_O_NOCACHE), &error_abort);
863         }
864         if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_NO_FLUSH)) {
865             qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_NO_FLUSH,
866                               !!(flags & BDRV_O_NO_FLUSH), &error_abort);
867         }
868         qemu_opt_unset(all_opts, "cache");
869     }
870 
871     /* Get a QDict for processing the options */
872     bs_opts = qdict_new();
873     qemu_opts_to_qdict(all_opts, bs_opts);
874 
875     legacy_opts = qemu_opts_create(&qemu_legacy_drive_opts, NULL, 0,
876                                    &error_abort);
877     qemu_opts_absorb_qdict(legacy_opts, bs_opts, &local_err);
878     if (local_err) {
879         error_report_err(local_err);
880         goto fail;
881     }
882 
883     /* Other deprecated options */
884     if (!qtest_enabled()) {
885         for (i = 0; i < ARRAY_SIZE(deprecated); i++) {
886             if (qemu_opt_get(legacy_opts, deprecated[i]) != NULL) {
887                 error_report("'%s' is deprecated, please use the corresponding "
888                              "option of '-device' instead", deprecated[i]);
889             }
890         }
891     }
892 
893     /* Media type */
894     value = qemu_opt_get(legacy_opts, "media");
895     if (value) {
896         if (!strcmp(value, "disk")) {
897             media = MEDIA_DISK;
898         } else if (!strcmp(value, "cdrom")) {
899             media = MEDIA_CDROM;
900             read_only = true;
901         } else {
902             error_report("'%s' invalid media", value);
903             goto fail;
904         }
905     }
906 
907     /* copy-on-read is disabled with a warning for read-only devices */
908     read_only |= qemu_opt_get_bool(legacy_opts, BDRV_OPT_READ_ONLY, false);
909     copy_on_read = qemu_opt_get_bool(legacy_opts, "copy-on-read", false);
910 
911     if (read_only && copy_on_read) {
912         warn_report("disabling copy-on-read on read-only drive");
913         copy_on_read = false;
914     }
915 
916     qdict_put_str(bs_opts, BDRV_OPT_READ_ONLY, read_only ? "on" : "off");
917     qdict_put_str(bs_opts, "copy-on-read", copy_on_read ? "on" : "off");
918 
919     /* Controller type */
920     value = qemu_opt_get(legacy_opts, "if");
921     if (value) {
922         for (type = 0;
923              type < IF_COUNT && strcmp(value, if_name[type]);
924              type++) {
925         }
926         if (type == IF_COUNT) {
927             error_report("unsupported bus type '%s'", value);
928             goto fail;
929         }
930     } else {
931         type = block_default_type;
932     }
933 
934     /* Geometry */
935     cyls  = qemu_opt_get_number(legacy_opts, "cyls", 0);
936     heads = qemu_opt_get_number(legacy_opts, "heads", 0);
937     secs  = qemu_opt_get_number(legacy_opts, "secs", 0);
938 
939     if (cyls || heads || secs) {
940         if (cyls < 1) {
941             error_report("invalid physical cyls number");
942             goto fail;
943         }
944         if (heads < 1) {
945             error_report("invalid physical heads number");
946             goto fail;
947         }
948         if (secs < 1) {
949             error_report("invalid physical secs number");
950             goto fail;
951         }
952     }
953 
954     translation = BIOS_ATA_TRANSLATION_AUTO;
955     value = qemu_opt_get(legacy_opts, "trans");
956     if (value != NULL) {
957         if (!cyls) {
958             error_report("'%s' trans must be used with cyls, heads and secs",
959                          value);
960             goto fail;
961         }
962         if (!strcmp(value, "none")) {
963             translation = BIOS_ATA_TRANSLATION_NONE;
964         } else if (!strcmp(value, "lba")) {
965             translation = BIOS_ATA_TRANSLATION_LBA;
966         } else if (!strcmp(value, "large")) {
967             translation = BIOS_ATA_TRANSLATION_LARGE;
968         } else if (!strcmp(value, "rechs")) {
969             translation = BIOS_ATA_TRANSLATION_RECHS;
970         } else if (!strcmp(value, "auto")) {
971             translation = BIOS_ATA_TRANSLATION_AUTO;
972         } else {
973             error_report("'%s' invalid translation type", value);
974             goto fail;
975         }
976     }
977 
978     if (media == MEDIA_CDROM) {
979         if (cyls || secs || heads) {
980             error_report("CHS can't be set with media=cdrom");
981             goto fail;
982         }
983     }
984 
985     /* Device address specified by bus/unit or index.
986      * If none was specified, try to find the first free one. */
987     bus_id  = qemu_opt_get_number(legacy_opts, "bus", 0);
988     unit_id = qemu_opt_get_number(legacy_opts, "unit", -1);
989     index   = qemu_opt_get_number(legacy_opts, "index", -1);
990 
991     max_devs = if_max_devs[type];
992 
993     if (index != -1) {
994         if (bus_id != 0 || unit_id != -1) {
995             error_report("index cannot be used with bus and unit");
996             goto fail;
997         }
998         bus_id = drive_index_to_bus_id(type, index);
999         unit_id = drive_index_to_unit_id(type, index);
1000     }
1001 
1002     if (unit_id == -1) {
1003        unit_id = 0;
1004        while (drive_get(type, bus_id, unit_id) != NULL) {
1005            unit_id++;
1006            if (max_devs && unit_id >= max_devs) {
1007                unit_id -= max_devs;
1008                bus_id++;
1009            }
1010        }
1011     }
1012 
1013     if (max_devs && unit_id >= max_devs) {
1014         error_report("unit %d too big (max is %d)", unit_id, max_devs - 1);
1015         goto fail;
1016     }
1017 
1018     if (drive_get(type, bus_id, unit_id) != NULL) {
1019         error_report("drive with bus=%d, unit=%d (index=%d) exists",
1020                      bus_id, unit_id, index);
1021         goto fail;
1022     }
1023 
1024     /* Serial number */
1025     serial = qemu_opt_get(legacy_opts, "serial");
1026 
1027     /* no id supplied -> create one */
1028     if (qemu_opts_id(all_opts) == NULL) {
1029         char *new_id;
1030         const char *mediastr = "";
1031         if (type == IF_IDE || type == IF_SCSI) {
1032             mediastr = (media == MEDIA_CDROM) ? "-cd" : "-hd";
1033         }
1034         if (max_devs) {
1035             new_id = g_strdup_printf("%s%i%s%i", if_name[type], bus_id,
1036                                      mediastr, unit_id);
1037         } else {
1038             new_id = g_strdup_printf("%s%s%i", if_name[type],
1039                                      mediastr, unit_id);
1040         }
1041         qdict_put_str(bs_opts, "id", new_id);
1042         g_free(new_id);
1043     }
1044 
1045     /* Add virtio block device */
1046     devaddr = qemu_opt_get(legacy_opts, "addr");
1047     if (devaddr && type != IF_VIRTIO) {
1048         error_report("addr is not supported by this bus type");
1049         goto fail;
1050     }
1051 
1052     if (type == IF_VIRTIO) {
1053         QemuOpts *devopts;
1054         devopts = qemu_opts_create(qemu_find_opts("device"), NULL, 0,
1055                                    &error_abort);
1056         if (arch_type == QEMU_ARCH_S390X) {
1057             qemu_opt_set(devopts, "driver", "virtio-blk-ccw", &error_abort);
1058         } else {
1059             qemu_opt_set(devopts, "driver", "virtio-blk-pci", &error_abort);
1060         }
1061         qemu_opt_set(devopts, "drive", qdict_get_str(bs_opts, "id"),
1062                      &error_abort);
1063         if (devaddr) {
1064             qemu_opt_set(devopts, "addr", devaddr, &error_abort);
1065         }
1066     }
1067 
1068     filename = qemu_opt_get(legacy_opts, "file");
1069 
1070     /* Check werror/rerror compatibility with if=... */
1071     werror = qemu_opt_get(legacy_opts, "werror");
1072     if (werror != NULL) {
1073         if (type != IF_IDE && type != IF_SCSI && type != IF_VIRTIO &&
1074             type != IF_NONE) {
1075             error_report("werror is not supported by this bus type");
1076             goto fail;
1077         }
1078         qdict_put_str(bs_opts, "werror", werror);
1079     }
1080 
1081     rerror = qemu_opt_get(legacy_opts, "rerror");
1082     if (rerror != NULL) {
1083         if (type != IF_IDE && type != IF_VIRTIO && type != IF_SCSI &&
1084             type != IF_NONE) {
1085             error_report("rerror is not supported by this bus type");
1086             goto fail;
1087         }
1088         qdict_put_str(bs_opts, "rerror", rerror);
1089     }
1090 
1091     /* Actual block device init: Functionality shared with blockdev-add */
1092     blk = blockdev_init(filename, bs_opts, &local_err);
1093     bs_opts = NULL;
1094     if (!blk) {
1095         if (local_err) {
1096             error_report_err(local_err);
1097         }
1098         goto fail;
1099     } else {
1100         assert(!local_err);
1101     }
1102 
1103     /* Create legacy DriveInfo */
1104     dinfo = g_malloc0(sizeof(*dinfo));
1105     dinfo->opts = all_opts;
1106 
1107     dinfo->cyls = cyls;
1108     dinfo->heads = heads;
1109     dinfo->secs = secs;
1110     dinfo->trans = translation;
1111 
1112     dinfo->type = type;
1113     dinfo->bus = bus_id;
1114     dinfo->unit = unit_id;
1115     dinfo->devaddr = devaddr;
1116     dinfo->serial = g_strdup(serial);
1117 
1118     blk_set_legacy_dinfo(blk, dinfo);
1119 
1120     switch(type) {
1121     case IF_IDE:
1122     case IF_SCSI:
1123     case IF_XEN:
1124     case IF_NONE:
1125         dinfo->media_cd = media == MEDIA_CDROM;
1126         break;
1127     default:
1128         break;
1129     }
1130 
1131 fail:
1132     qemu_opts_del(legacy_opts);
1133     qobject_unref(bs_opts);
1134     return dinfo;
1135 }
1136 
1137 static BlockDriverState *qmp_get_root_bs(const char *name, Error **errp)
1138 {
1139     BlockDriverState *bs;
1140 
1141     bs = bdrv_lookup_bs(name, name, errp);
1142     if (bs == NULL) {
1143         return NULL;
1144     }
1145 
1146     if (!bdrv_is_root_node(bs)) {
1147         error_setg(errp, "Need a root block node");
1148         return NULL;
1149     }
1150 
1151     if (!bdrv_is_inserted(bs)) {
1152         error_setg(errp, "Device has no medium");
1153         return NULL;
1154     }
1155 
1156     return bs;
1157 }
1158 
1159 static BlockBackend *qmp_get_blk(const char *blk_name, const char *qdev_id,
1160                                  Error **errp)
1161 {
1162     BlockBackend *blk;
1163 
1164     if (!blk_name == !qdev_id) {
1165         error_setg(errp, "Need exactly one of 'device' and 'id'");
1166         return NULL;
1167     }
1168 
1169     if (qdev_id) {
1170         blk = blk_by_qdev_id(qdev_id, errp);
1171     } else {
1172         blk = blk_by_name(blk_name);
1173         if (blk == NULL) {
1174             error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
1175                       "Device '%s' not found", blk_name);
1176         }
1177     }
1178 
1179     return blk;
1180 }
1181 
1182 void hmp_commit(Monitor *mon, const QDict *qdict)
1183 {
1184     const char *device = qdict_get_str(qdict, "device");
1185     BlockBackend *blk;
1186     int ret;
1187 
1188     if (!strcmp(device, "all")) {
1189         ret = blk_commit_all();
1190     } else {
1191         BlockDriverState *bs;
1192         AioContext *aio_context;
1193 
1194         blk = blk_by_name(device);
1195         if (!blk) {
1196             monitor_printf(mon, "Device '%s' not found\n", device);
1197             return;
1198         }
1199         if (!blk_is_available(blk)) {
1200             monitor_printf(mon, "Device '%s' has no medium\n", device);
1201             return;
1202         }
1203 
1204         bs = blk_bs(blk);
1205         aio_context = bdrv_get_aio_context(bs);
1206         aio_context_acquire(aio_context);
1207 
1208         ret = bdrv_commit(bs);
1209 
1210         aio_context_release(aio_context);
1211     }
1212     if (ret < 0) {
1213         monitor_printf(mon, "'commit' error for '%s': %s\n", device,
1214                        strerror(-ret));
1215     }
1216 }
1217 
1218 static void blockdev_do_action(TransactionAction *action, Error **errp)
1219 {
1220     TransactionActionList list;
1221 
1222     list.value = action;
1223     list.next = NULL;
1224     qmp_transaction(&list, false, NULL, errp);
1225 }
1226 
1227 void qmp_blockdev_snapshot_sync(bool has_device, const char *device,
1228                                 bool has_node_name, const char *node_name,
1229                                 const char *snapshot_file,
1230                                 bool has_snapshot_node_name,
1231                                 const char *snapshot_node_name,
1232                                 bool has_format, const char *format,
1233                                 bool has_mode, NewImageMode mode, Error **errp)
1234 {
1235     BlockdevSnapshotSync snapshot = {
1236         .has_device = has_device,
1237         .device = (char *) device,
1238         .has_node_name = has_node_name,
1239         .node_name = (char *) node_name,
1240         .snapshot_file = (char *) snapshot_file,
1241         .has_snapshot_node_name = has_snapshot_node_name,
1242         .snapshot_node_name = (char *) snapshot_node_name,
1243         .has_format = has_format,
1244         .format = (char *) format,
1245         .has_mode = has_mode,
1246         .mode = mode,
1247     };
1248     TransactionAction action = {
1249         .type = TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC,
1250         .u.blockdev_snapshot_sync.data = &snapshot,
1251     };
1252     blockdev_do_action(&action, errp);
1253 }
1254 
1255 void qmp_blockdev_snapshot(const char *node, const char *overlay,
1256                            Error **errp)
1257 {
1258     BlockdevSnapshot snapshot_data = {
1259         .node = (char *) node,
1260         .overlay = (char *) overlay
1261     };
1262     TransactionAction action = {
1263         .type = TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT,
1264         .u.blockdev_snapshot.data = &snapshot_data,
1265     };
1266     blockdev_do_action(&action, errp);
1267 }
1268 
1269 void qmp_blockdev_snapshot_internal_sync(const char *device,
1270                                          const char *name,
1271                                          Error **errp)
1272 {
1273     BlockdevSnapshotInternal snapshot = {
1274         .device = (char *) device,
1275         .name = (char *) name
1276     };
1277     TransactionAction action = {
1278         .type = TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC,
1279         .u.blockdev_snapshot_internal_sync.data = &snapshot,
1280     };
1281     blockdev_do_action(&action, errp);
1282 }
1283 
1284 SnapshotInfo *qmp_blockdev_snapshot_delete_internal_sync(const char *device,
1285                                                          bool has_id,
1286                                                          const char *id,
1287                                                          bool has_name,
1288                                                          const char *name,
1289                                                          Error **errp)
1290 {
1291     BlockDriverState *bs;
1292     AioContext *aio_context;
1293     QEMUSnapshotInfo sn;
1294     Error *local_err = NULL;
1295     SnapshotInfo *info = NULL;
1296     int ret;
1297 
1298     bs = qmp_get_root_bs(device, errp);
1299     if (!bs) {
1300         return NULL;
1301     }
1302     aio_context = bdrv_get_aio_context(bs);
1303     aio_context_acquire(aio_context);
1304 
1305     if (!has_id) {
1306         id = NULL;
1307     }
1308 
1309     if (!has_name) {
1310         name = NULL;
1311     }
1312 
1313     if (!id && !name) {
1314         error_setg(errp, "Name or id must be provided");
1315         goto out_aio_context;
1316     }
1317 
1318     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_INTERNAL_SNAPSHOT_DELETE, errp)) {
1319         goto out_aio_context;
1320     }
1321 
1322     ret = bdrv_snapshot_find_by_id_and_name(bs, id, name, &sn, &local_err);
1323     if (local_err) {
1324         error_propagate(errp, local_err);
1325         goto out_aio_context;
1326     }
1327     if (!ret) {
1328         error_setg(errp,
1329                    "Snapshot with id '%s' and name '%s' does not exist on "
1330                    "device '%s'",
1331                    STR_OR_NULL(id), STR_OR_NULL(name), device);
1332         goto out_aio_context;
1333     }
1334 
1335     bdrv_snapshot_delete(bs, id, name, &local_err);
1336     if (local_err) {
1337         error_propagate(errp, local_err);
1338         goto out_aio_context;
1339     }
1340 
1341     aio_context_release(aio_context);
1342 
1343     info = g_new0(SnapshotInfo, 1);
1344     info->id = g_strdup(sn.id_str);
1345     info->name = g_strdup(sn.name);
1346     info->date_nsec = sn.date_nsec;
1347     info->date_sec = sn.date_sec;
1348     info->vm_state_size = sn.vm_state_size;
1349     info->vm_clock_nsec = sn.vm_clock_nsec % 1000000000;
1350     info->vm_clock_sec = sn.vm_clock_nsec / 1000000000;
1351 
1352     return info;
1353 
1354 out_aio_context:
1355     aio_context_release(aio_context);
1356     return NULL;
1357 }
1358 
1359 /**
1360  * block_dirty_bitmap_lookup:
1361  * Return a dirty bitmap (if present), after validating
1362  * the node reference and bitmap names.
1363  *
1364  * @node: The name of the BDS node to search for bitmaps
1365  * @name: The name of the bitmap to search for
1366  * @pbs: Output pointer for BDS lookup, if desired. Can be NULL.
1367  * @paio: Output pointer for aio_context acquisition, if desired. Can be NULL.
1368  * @errp: Output pointer for error information. Can be NULL.
1369  *
1370  * @return: A bitmap object on success, or NULL on failure.
1371  */
1372 static BdrvDirtyBitmap *block_dirty_bitmap_lookup(const char *node,
1373                                                   const char *name,
1374                                                   BlockDriverState **pbs,
1375                                                   Error **errp)
1376 {
1377     BlockDriverState *bs;
1378     BdrvDirtyBitmap *bitmap;
1379 
1380     if (!node) {
1381         error_setg(errp, "Node cannot be NULL");
1382         return NULL;
1383     }
1384     if (!name) {
1385         error_setg(errp, "Bitmap name cannot be NULL");
1386         return NULL;
1387     }
1388     bs = bdrv_lookup_bs(node, node, NULL);
1389     if (!bs) {
1390         error_setg(errp, "Node '%s' not found", node);
1391         return NULL;
1392     }
1393 
1394     bitmap = bdrv_find_dirty_bitmap(bs, name);
1395     if (!bitmap) {
1396         error_setg(errp, "Dirty bitmap '%s' not found", name);
1397         return NULL;
1398     }
1399 
1400     if (pbs) {
1401         *pbs = bs;
1402     }
1403 
1404     return bitmap;
1405 }
1406 
1407 /* New and old BlockDriverState structs for atomic group operations */
1408 
1409 typedef struct BlkActionState BlkActionState;
1410 
1411 /**
1412  * BlkActionOps:
1413  * Table of operations that define an Action.
1414  *
1415  * @instance_size: Size of state struct, in bytes.
1416  * @prepare: Prepare the work, must NOT be NULL.
1417  * @commit: Commit the changes, can be NULL.
1418  * @abort: Abort the changes on fail, can be NULL.
1419  * @clean: Clean up resources after all transaction actions have called
1420  *         commit() or abort(). Can be NULL.
1421  *
1422  * Only prepare() may fail. In a single transaction, only one of commit() or
1423  * abort() will be called. clean() will always be called if it is present.
1424  */
1425 typedef struct BlkActionOps {
1426     size_t instance_size;
1427     void (*prepare)(BlkActionState *common, Error **errp);
1428     void (*commit)(BlkActionState *common);
1429     void (*abort)(BlkActionState *common);
1430     void (*clean)(BlkActionState *common);
1431 } BlkActionOps;
1432 
1433 /**
1434  * BlkActionState:
1435  * Describes one Action's state within a Transaction.
1436  *
1437  * @action: QAPI-defined enum identifying which Action to perform.
1438  * @ops: Table of ActionOps this Action can perform.
1439  * @block_job_txn: Transaction which this action belongs to.
1440  * @entry: List membership for all Actions in this Transaction.
1441  *
1442  * This structure must be arranged as first member in a subclassed type,
1443  * assuming that the compiler will also arrange it to the same offsets as the
1444  * base class.
1445  */
1446 struct BlkActionState {
1447     TransactionAction *action;
1448     const BlkActionOps *ops;
1449     BlockJobTxn *block_job_txn;
1450     TransactionProperties *txn_props;
1451     QSIMPLEQ_ENTRY(BlkActionState) entry;
1452 };
1453 
1454 /* internal snapshot private data */
1455 typedef struct InternalSnapshotState {
1456     BlkActionState common;
1457     BlockDriverState *bs;
1458     QEMUSnapshotInfo sn;
1459     bool created;
1460 } InternalSnapshotState;
1461 
1462 
1463 static int action_check_completion_mode(BlkActionState *s, Error **errp)
1464 {
1465     if (s->txn_props->completion_mode != ACTION_COMPLETION_MODE_INDIVIDUAL) {
1466         error_setg(errp,
1467                    "Action '%s' does not support Transaction property "
1468                    "completion-mode = %s",
1469                    TransactionActionKind_str(s->action->type),
1470                    ActionCompletionMode_str(s->txn_props->completion_mode));
1471         return -1;
1472     }
1473     return 0;
1474 }
1475 
1476 static void internal_snapshot_prepare(BlkActionState *common,
1477                                       Error **errp)
1478 {
1479     Error *local_err = NULL;
1480     const char *device;
1481     const char *name;
1482     BlockDriverState *bs;
1483     QEMUSnapshotInfo old_sn, *sn;
1484     bool ret;
1485     qemu_timeval tv;
1486     BlockdevSnapshotInternal *internal;
1487     InternalSnapshotState *state;
1488     AioContext *aio_context;
1489     int ret1;
1490 
1491     g_assert(common->action->type ==
1492              TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC);
1493     internal = common->action->u.blockdev_snapshot_internal_sync.data;
1494     state = DO_UPCAST(InternalSnapshotState, common, common);
1495 
1496     /* 1. parse input */
1497     device = internal->device;
1498     name = internal->name;
1499 
1500     /* 2. check for validation */
1501     if (action_check_completion_mode(common, errp) < 0) {
1502         return;
1503     }
1504 
1505     bs = qmp_get_root_bs(device, errp);
1506     if (!bs) {
1507         return;
1508     }
1509 
1510     aio_context = bdrv_get_aio_context(bs);
1511     aio_context_acquire(aio_context);
1512 
1513     state->bs = bs;
1514 
1515     /* Paired with .clean() */
1516     bdrv_drained_begin(bs);
1517 
1518     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_INTERNAL_SNAPSHOT, errp)) {
1519         goto out;
1520     }
1521 
1522     if (bdrv_is_read_only(bs)) {
1523         error_setg(errp, "Device '%s' is read only", device);
1524         goto out;
1525     }
1526 
1527     if (!bdrv_can_snapshot(bs)) {
1528         error_setg(errp, "Block format '%s' used by device '%s' "
1529                    "does not support internal snapshots",
1530                    bs->drv->format_name, device);
1531         goto out;
1532     }
1533 
1534     if (!strlen(name)) {
1535         error_setg(errp, "Name is empty");
1536         goto out;
1537     }
1538 
1539     /* check whether a snapshot with name exist */
1540     ret = bdrv_snapshot_find_by_id_and_name(bs, NULL, name, &old_sn,
1541                                             &local_err);
1542     if (local_err) {
1543         error_propagate(errp, local_err);
1544         goto out;
1545     } else if (ret) {
1546         error_setg(errp,
1547                    "Snapshot with name '%s' already exists on device '%s'",
1548                    name, device);
1549         goto out;
1550     }
1551 
1552     /* 3. take the snapshot */
1553     sn = &state->sn;
1554     pstrcpy(sn->name, sizeof(sn->name), name);
1555     qemu_gettimeofday(&tv);
1556     sn->date_sec = tv.tv_sec;
1557     sn->date_nsec = tv.tv_usec * 1000;
1558     sn->vm_clock_nsec = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL);
1559 
1560     ret1 = bdrv_snapshot_create(bs, sn);
1561     if (ret1 < 0) {
1562         error_setg_errno(errp, -ret1,
1563                          "Failed to create snapshot '%s' on device '%s'",
1564                          name, device);
1565         goto out;
1566     }
1567 
1568     /* 4. succeed, mark a snapshot is created */
1569     state->created = true;
1570 
1571 out:
1572     aio_context_release(aio_context);
1573 }
1574 
1575 static void internal_snapshot_abort(BlkActionState *common)
1576 {
1577     InternalSnapshotState *state =
1578                              DO_UPCAST(InternalSnapshotState, common, common);
1579     BlockDriverState *bs = state->bs;
1580     QEMUSnapshotInfo *sn = &state->sn;
1581     AioContext *aio_context;
1582     Error *local_error = NULL;
1583 
1584     if (!state->created) {
1585         return;
1586     }
1587 
1588     aio_context = bdrv_get_aio_context(state->bs);
1589     aio_context_acquire(aio_context);
1590 
1591     if (bdrv_snapshot_delete(bs, sn->id_str, sn->name, &local_error) < 0) {
1592         error_reportf_err(local_error,
1593                           "Failed to delete snapshot with id '%s' and "
1594                           "name '%s' on device '%s' in abort: ",
1595                           sn->id_str, sn->name,
1596                           bdrv_get_device_name(bs));
1597     }
1598 
1599     aio_context_release(aio_context);
1600 }
1601 
1602 static void internal_snapshot_clean(BlkActionState *common)
1603 {
1604     InternalSnapshotState *state = DO_UPCAST(InternalSnapshotState,
1605                                              common, common);
1606     AioContext *aio_context;
1607 
1608     if (!state->bs) {
1609         return;
1610     }
1611 
1612     aio_context = bdrv_get_aio_context(state->bs);
1613     aio_context_acquire(aio_context);
1614 
1615     bdrv_drained_end(state->bs);
1616 
1617     aio_context_release(aio_context);
1618 }
1619 
1620 /* external snapshot private data */
1621 typedef struct ExternalSnapshotState {
1622     BlkActionState common;
1623     BlockDriverState *old_bs;
1624     BlockDriverState *new_bs;
1625     bool overlay_appended;
1626 } ExternalSnapshotState;
1627 
1628 static void external_snapshot_prepare(BlkActionState *common,
1629                                       Error **errp)
1630 {
1631     int flags = 0;
1632     QDict *options = NULL;
1633     Error *local_err = NULL;
1634     /* Device and node name of the image to generate the snapshot from */
1635     const char *device;
1636     const char *node_name;
1637     /* Reference to the new image (for 'blockdev-snapshot') */
1638     const char *snapshot_ref;
1639     /* File name of the new image (for 'blockdev-snapshot-sync') */
1640     const char *new_image_file;
1641     ExternalSnapshotState *state =
1642                              DO_UPCAST(ExternalSnapshotState, common, common);
1643     TransactionAction *action = common->action;
1644     AioContext *aio_context;
1645 
1646     /* 'blockdev-snapshot' and 'blockdev-snapshot-sync' have similar
1647      * purpose but a different set of parameters */
1648     switch (action->type) {
1649     case TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT:
1650         {
1651             BlockdevSnapshot *s = action->u.blockdev_snapshot.data;
1652             device = s->node;
1653             node_name = s->node;
1654             new_image_file = NULL;
1655             snapshot_ref = s->overlay;
1656         }
1657         break;
1658     case TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC:
1659         {
1660             BlockdevSnapshotSync *s = action->u.blockdev_snapshot_sync.data;
1661             device = s->has_device ? s->device : NULL;
1662             node_name = s->has_node_name ? s->node_name : NULL;
1663             new_image_file = s->snapshot_file;
1664             snapshot_ref = NULL;
1665         }
1666         break;
1667     default:
1668         g_assert_not_reached();
1669     }
1670 
1671     /* start processing */
1672     if (action_check_completion_mode(common, errp) < 0) {
1673         return;
1674     }
1675 
1676     state->old_bs = bdrv_lookup_bs(device, node_name, errp);
1677     if (!state->old_bs) {
1678         return;
1679     }
1680 
1681     aio_context = bdrv_get_aio_context(state->old_bs);
1682     aio_context_acquire(aio_context);
1683 
1684     /* Paired with .clean() */
1685     bdrv_drained_begin(state->old_bs);
1686 
1687     if (!bdrv_is_inserted(state->old_bs)) {
1688         error_setg(errp, QERR_DEVICE_HAS_NO_MEDIUM, device);
1689         goto out;
1690     }
1691 
1692     if (bdrv_op_is_blocked(state->old_bs,
1693                            BLOCK_OP_TYPE_EXTERNAL_SNAPSHOT, errp)) {
1694         goto out;
1695     }
1696 
1697     if (!bdrv_is_read_only(state->old_bs)) {
1698         if (bdrv_flush(state->old_bs)) {
1699             error_setg(errp, QERR_IO_ERROR);
1700             goto out;
1701         }
1702     }
1703 
1704     if (!bdrv_is_first_non_filter(state->old_bs)) {
1705         error_setg(errp, QERR_FEATURE_DISABLED, "snapshot");
1706         goto out;
1707     }
1708 
1709     if (action->type == TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC) {
1710         BlockdevSnapshotSync *s = action->u.blockdev_snapshot_sync.data;
1711         const char *format = s->has_format ? s->format : "qcow2";
1712         enum NewImageMode mode;
1713         const char *snapshot_node_name =
1714             s->has_snapshot_node_name ? s->snapshot_node_name : NULL;
1715 
1716         if (node_name && !snapshot_node_name) {
1717             error_setg(errp, "New snapshot node name missing");
1718             goto out;
1719         }
1720 
1721         if (snapshot_node_name &&
1722             bdrv_lookup_bs(snapshot_node_name, snapshot_node_name, NULL)) {
1723             error_setg(errp, "New snapshot node name already in use");
1724             goto out;
1725         }
1726 
1727         flags = state->old_bs->open_flags;
1728         flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_COPY_ON_READ);
1729         flags |= BDRV_O_NO_BACKING;
1730 
1731         /* create new image w/backing file */
1732         mode = s->has_mode ? s->mode : NEW_IMAGE_MODE_ABSOLUTE_PATHS;
1733         if (mode != NEW_IMAGE_MODE_EXISTING) {
1734             int64_t size = bdrv_getlength(state->old_bs);
1735             if (size < 0) {
1736                 error_setg_errno(errp, -size, "bdrv_getlength failed");
1737                 goto out;
1738             }
1739             bdrv_img_create(new_image_file, format,
1740                             state->old_bs->filename,
1741                             state->old_bs->drv->format_name,
1742                             NULL, size, flags, false, &local_err);
1743             if (local_err) {
1744                 error_propagate(errp, local_err);
1745                 goto out;
1746             }
1747         }
1748 
1749         options = qdict_new();
1750         if (s->has_snapshot_node_name) {
1751             qdict_put_str(options, "node-name", snapshot_node_name);
1752         }
1753         qdict_put_str(options, "driver", format);
1754     }
1755 
1756     state->new_bs = bdrv_open(new_image_file, snapshot_ref, options, flags,
1757                               errp);
1758     /* We will manually add the backing_hd field to the bs later */
1759     if (!state->new_bs) {
1760         goto out;
1761     }
1762 
1763     if (bdrv_has_blk(state->new_bs)) {
1764         error_setg(errp, "The snapshot is already in use");
1765         goto out;
1766     }
1767 
1768     if (bdrv_op_is_blocked(state->new_bs, BLOCK_OP_TYPE_EXTERNAL_SNAPSHOT,
1769                            errp)) {
1770         goto out;
1771     }
1772 
1773     if (state->new_bs->backing != NULL) {
1774         error_setg(errp, "The snapshot already has a backing image");
1775         goto out;
1776     }
1777 
1778     if (!state->new_bs->drv->supports_backing) {
1779         error_setg(errp, "The snapshot does not support backing images");
1780         goto out;
1781     }
1782 
1783     bdrv_set_aio_context(state->new_bs, aio_context);
1784 
1785     /* This removes our old bs and adds the new bs. This is an operation that
1786      * can fail, so we need to do it in .prepare; undoing it for abort is
1787      * always possible. */
1788     bdrv_ref(state->new_bs);
1789     bdrv_append(state->new_bs, state->old_bs, &local_err);
1790     if (local_err) {
1791         error_propagate(errp, local_err);
1792         goto out;
1793     }
1794     state->overlay_appended = true;
1795 
1796 out:
1797     aio_context_release(aio_context);
1798 }
1799 
1800 static void external_snapshot_commit(BlkActionState *common)
1801 {
1802     ExternalSnapshotState *state =
1803                              DO_UPCAST(ExternalSnapshotState, common, common);
1804     AioContext *aio_context;
1805 
1806     aio_context = bdrv_get_aio_context(state->old_bs);
1807     aio_context_acquire(aio_context);
1808 
1809     /* We don't need (or want) to use the transactional
1810      * bdrv_reopen_multiple() across all the entries at once, because we
1811      * don't want to abort all of them if one of them fails the reopen */
1812     if (!atomic_read(&state->old_bs->copy_on_read)) {
1813         bdrv_reopen(state->old_bs, state->old_bs->open_flags & ~BDRV_O_RDWR,
1814                     NULL);
1815     }
1816 
1817     aio_context_release(aio_context);
1818 }
1819 
1820 static void external_snapshot_abort(BlkActionState *common)
1821 {
1822     ExternalSnapshotState *state =
1823                              DO_UPCAST(ExternalSnapshotState, common, common);
1824     if (state->new_bs) {
1825         if (state->overlay_appended) {
1826             AioContext *aio_context;
1827 
1828             aio_context = bdrv_get_aio_context(state->old_bs);
1829             aio_context_acquire(aio_context);
1830 
1831             bdrv_ref(state->old_bs);   /* we can't let bdrv_set_backind_hd()
1832                                           close state->old_bs; we need it */
1833             bdrv_set_backing_hd(state->new_bs, NULL, &error_abort);
1834             bdrv_replace_node(state->new_bs, state->old_bs, &error_abort);
1835             bdrv_unref(state->old_bs); /* bdrv_replace_node() ref'ed old_bs */
1836 
1837             aio_context_release(aio_context);
1838         }
1839     }
1840 }
1841 
1842 static void external_snapshot_clean(BlkActionState *common)
1843 {
1844     ExternalSnapshotState *state =
1845                              DO_UPCAST(ExternalSnapshotState, common, common);
1846     AioContext *aio_context;
1847 
1848     if (!state->old_bs) {
1849         return;
1850     }
1851 
1852     aio_context = bdrv_get_aio_context(state->old_bs);
1853     aio_context_acquire(aio_context);
1854 
1855     bdrv_drained_end(state->old_bs);
1856     bdrv_unref(state->new_bs);
1857 
1858     aio_context_release(aio_context);
1859 }
1860 
1861 typedef struct DriveBackupState {
1862     BlkActionState common;
1863     BlockDriverState *bs;
1864     BlockJob *job;
1865 } DriveBackupState;
1866 
1867 static BlockJob *do_drive_backup(DriveBackup *backup, BlockJobTxn *txn,
1868                             Error **errp);
1869 
1870 static void drive_backup_prepare(BlkActionState *common, Error **errp)
1871 {
1872     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1873     BlockDriverState *bs;
1874     DriveBackup *backup;
1875     AioContext *aio_context;
1876     Error *local_err = NULL;
1877 
1878     assert(common->action->type == TRANSACTION_ACTION_KIND_DRIVE_BACKUP);
1879     backup = common->action->u.drive_backup.data;
1880 
1881     bs = qmp_get_root_bs(backup->device, errp);
1882     if (!bs) {
1883         return;
1884     }
1885 
1886     aio_context = bdrv_get_aio_context(bs);
1887     aio_context_acquire(aio_context);
1888 
1889     /* Paired with .clean() */
1890     bdrv_drained_begin(bs);
1891 
1892     state->bs = bs;
1893 
1894     state->job = do_drive_backup(backup, common->block_job_txn, &local_err);
1895     if (local_err) {
1896         error_propagate(errp, local_err);
1897         goto out;
1898     }
1899 
1900 out:
1901     aio_context_release(aio_context);
1902 }
1903 
1904 static void drive_backup_commit(BlkActionState *common)
1905 {
1906     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1907     AioContext *aio_context;
1908 
1909     aio_context = bdrv_get_aio_context(state->bs);
1910     aio_context_acquire(aio_context);
1911 
1912     assert(state->job);
1913     block_job_start(state->job);
1914 
1915     aio_context_release(aio_context);
1916 }
1917 
1918 static void drive_backup_abort(BlkActionState *common)
1919 {
1920     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1921 
1922     if (state->job) {
1923         AioContext *aio_context;
1924 
1925         aio_context = bdrv_get_aio_context(state->bs);
1926         aio_context_acquire(aio_context);
1927 
1928         block_job_cancel_sync(state->job);
1929 
1930         aio_context_release(aio_context);
1931     }
1932 }
1933 
1934 static void drive_backup_clean(BlkActionState *common)
1935 {
1936     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1937     AioContext *aio_context;
1938 
1939     if (!state->bs) {
1940         return;
1941     }
1942 
1943     aio_context = bdrv_get_aio_context(state->bs);
1944     aio_context_acquire(aio_context);
1945 
1946     bdrv_drained_end(state->bs);
1947 
1948     aio_context_release(aio_context);
1949 }
1950 
1951 typedef struct BlockdevBackupState {
1952     BlkActionState common;
1953     BlockDriverState *bs;
1954     BlockJob *job;
1955 } BlockdevBackupState;
1956 
1957 static BlockJob *do_blockdev_backup(BlockdevBackup *backup, BlockJobTxn *txn,
1958                                     Error **errp);
1959 
1960 static void blockdev_backup_prepare(BlkActionState *common, Error **errp)
1961 {
1962     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1963     BlockdevBackup *backup;
1964     BlockDriverState *bs, *target;
1965     AioContext *aio_context;
1966     Error *local_err = NULL;
1967 
1968     assert(common->action->type == TRANSACTION_ACTION_KIND_BLOCKDEV_BACKUP);
1969     backup = common->action->u.blockdev_backup.data;
1970 
1971     bs = qmp_get_root_bs(backup->device, errp);
1972     if (!bs) {
1973         return;
1974     }
1975 
1976     target = bdrv_lookup_bs(backup->target, backup->target, errp);
1977     if (!target) {
1978         return;
1979     }
1980 
1981     aio_context = bdrv_get_aio_context(bs);
1982     if (aio_context != bdrv_get_aio_context(target)) {
1983         error_setg(errp, "Backup between two IO threads is not implemented");
1984         return;
1985     }
1986     aio_context_acquire(aio_context);
1987     state->bs = bs;
1988 
1989     /* Paired with .clean() */
1990     bdrv_drained_begin(state->bs);
1991 
1992     state->job = do_blockdev_backup(backup, common->block_job_txn, &local_err);
1993     if (local_err) {
1994         error_propagate(errp, local_err);
1995         goto out;
1996     }
1997 
1998 out:
1999     aio_context_release(aio_context);
2000 }
2001 
2002 static void blockdev_backup_commit(BlkActionState *common)
2003 {
2004     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
2005     AioContext *aio_context;
2006 
2007     aio_context = bdrv_get_aio_context(state->bs);
2008     aio_context_acquire(aio_context);
2009 
2010     assert(state->job);
2011     block_job_start(state->job);
2012 
2013     aio_context_release(aio_context);
2014 }
2015 
2016 static void blockdev_backup_abort(BlkActionState *common)
2017 {
2018     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
2019 
2020     if (state->job) {
2021         AioContext *aio_context;
2022 
2023         aio_context = bdrv_get_aio_context(state->bs);
2024         aio_context_acquire(aio_context);
2025 
2026         block_job_cancel_sync(state->job);
2027 
2028         aio_context_release(aio_context);
2029     }
2030 }
2031 
2032 static void blockdev_backup_clean(BlkActionState *common)
2033 {
2034     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
2035     AioContext *aio_context;
2036 
2037     if (!state->bs) {
2038         return;
2039     }
2040 
2041     aio_context = bdrv_get_aio_context(state->bs);
2042     aio_context_acquire(aio_context);
2043 
2044     bdrv_drained_end(state->bs);
2045 
2046     aio_context_release(aio_context);
2047 }
2048 
2049 typedef struct BlockDirtyBitmapState {
2050     BlkActionState common;
2051     BdrvDirtyBitmap *bitmap;
2052     BlockDriverState *bs;
2053     HBitmap *backup;
2054     bool prepared;
2055 } BlockDirtyBitmapState;
2056 
2057 static void block_dirty_bitmap_add_prepare(BlkActionState *common,
2058                                            Error **errp)
2059 {
2060     Error *local_err = NULL;
2061     BlockDirtyBitmapAdd *action;
2062     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2063                                              common, common);
2064 
2065     if (action_check_completion_mode(common, errp) < 0) {
2066         return;
2067     }
2068 
2069     action = common->action->u.block_dirty_bitmap_add.data;
2070     /* AIO context taken and released within qmp_block_dirty_bitmap_add */
2071     qmp_block_dirty_bitmap_add(action->node, action->name,
2072                                action->has_granularity, action->granularity,
2073                                action->has_persistent, action->persistent,
2074                                action->has_autoload, action->autoload,
2075                                &local_err);
2076 
2077     if (!local_err) {
2078         state->prepared = true;
2079     } else {
2080         error_propagate(errp, local_err);
2081     }
2082 }
2083 
2084 static void block_dirty_bitmap_add_abort(BlkActionState *common)
2085 {
2086     BlockDirtyBitmapAdd *action;
2087     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2088                                              common, common);
2089 
2090     action = common->action->u.block_dirty_bitmap_add.data;
2091     /* Should not be able to fail: IF the bitmap was added via .prepare(),
2092      * then the node reference and bitmap name must have been valid.
2093      */
2094     if (state->prepared) {
2095         qmp_block_dirty_bitmap_remove(action->node, action->name, &error_abort);
2096     }
2097 }
2098 
2099 static void block_dirty_bitmap_clear_prepare(BlkActionState *common,
2100                                              Error **errp)
2101 {
2102     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2103                                              common, common);
2104     BlockDirtyBitmap *action;
2105 
2106     if (action_check_completion_mode(common, errp) < 0) {
2107         return;
2108     }
2109 
2110     action = common->action->u.block_dirty_bitmap_clear.data;
2111     state->bitmap = block_dirty_bitmap_lookup(action->node,
2112                                               action->name,
2113                                               &state->bs,
2114                                               errp);
2115     if (!state->bitmap) {
2116         return;
2117     }
2118 
2119     if (bdrv_dirty_bitmap_frozen(state->bitmap)) {
2120         error_setg(errp, "Cannot modify a frozen bitmap");
2121         return;
2122     } else if (bdrv_dirty_bitmap_qmp_locked(state->bitmap)) {
2123         error_setg(errp, "Cannot modify a locked bitmap");
2124         return;
2125     } else if (!bdrv_dirty_bitmap_enabled(state->bitmap)) {
2126         error_setg(errp, "Cannot clear a disabled bitmap");
2127         return;
2128     } else if (bdrv_dirty_bitmap_readonly(state->bitmap)) {
2129         error_setg(errp, "Cannot clear a readonly bitmap");
2130         return;
2131     }
2132 
2133     bdrv_clear_dirty_bitmap(state->bitmap, &state->backup);
2134 }
2135 
2136 static void block_dirty_bitmap_clear_abort(BlkActionState *common)
2137 {
2138     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2139                                              common, common);
2140 
2141     if (state->backup) {
2142         bdrv_undo_clear_dirty_bitmap(state->bitmap, state->backup);
2143     }
2144 }
2145 
2146 static void block_dirty_bitmap_clear_commit(BlkActionState *common)
2147 {
2148     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2149                                              common, common);
2150 
2151     hbitmap_free(state->backup);
2152 }
2153 
2154 static void abort_prepare(BlkActionState *common, Error **errp)
2155 {
2156     error_setg(errp, "Transaction aborted using Abort action");
2157 }
2158 
2159 static void abort_commit(BlkActionState *common)
2160 {
2161     g_assert_not_reached(); /* this action never succeeds */
2162 }
2163 
2164 static const BlkActionOps actions[] = {
2165     [TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT] = {
2166         .instance_size = sizeof(ExternalSnapshotState),
2167         .prepare  = external_snapshot_prepare,
2168         .commit   = external_snapshot_commit,
2169         .abort = external_snapshot_abort,
2170         .clean = external_snapshot_clean,
2171     },
2172     [TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC] = {
2173         .instance_size = sizeof(ExternalSnapshotState),
2174         .prepare  = external_snapshot_prepare,
2175         .commit   = external_snapshot_commit,
2176         .abort = external_snapshot_abort,
2177         .clean = external_snapshot_clean,
2178     },
2179     [TRANSACTION_ACTION_KIND_DRIVE_BACKUP] = {
2180         .instance_size = sizeof(DriveBackupState),
2181         .prepare = drive_backup_prepare,
2182         .commit = drive_backup_commit,
2183         .abort = drive_backup_abort,
2184         .clean = drive_backup_clean,
2185     },
2186     [TRANSACTION_ACTION_KIND_BLOCKDEV_BACKUP] = {
2187         .instance_size = sizeof(BlockdevBackupState),
2188         .prepare = blockdev_backup_prepare,
2189         .commit = blockdev_backup_commit,
2190         .abort = blockdev_backup_abort,
2191         .clean = blockdev_backup_clean,
2192     },
2193     [TRANSACTION_ACTION_KIND_ABORT] = {
2194         .instance_size = sizeof(BlkActionState),
2195         .prepare = abort_prepare,
2196         .commit = abort_commit,
2197     },
2198     [TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC] = {
2199         .instance_size = sizeof(InternalSnapshotState),
2200         .prepare  = internal_snapshot_prepare,
2201         .abort = internal_snapshot_abort,
2202         .clean = internal_snapshot_clean,
2203     },
2204     [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_ADD] = {
2205         .instance_size = sizeof(BlockDirtyBitmapState),
2206         .prepare = block_dirty_bitmap_add_prepare,
2207         .abort = block_dirty_bitmap_add_abort,
2208     },
2209     [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_CLEAR] = {
2210         .instance_size = sizeof(BlockDirtyBitmapState),
2211         .prepare = block_dirty_bitmap_clear_prepare,
2212         .commit = block_dirty_bitmap_clear_commit,
2213         .abort = block_dirty_bitmap_clear_abort,
2214     }
2215 };
2216 
2217 /**
2218  * Allocate a TransactionProperties structure if necessary, and fill
2219  * that structure with desired defaults if they are unset.
2220  */
2221 static TransactionProperties *get_transaction_properties(
2222     TransactionProperties *props)
2223 {
2224     if (!props) {
2225         props = g_new0(TransactionProperties, 1);
2226     }
2227 
2228     if (!props->has_completion_mode) {
2229         props->has_completion_mode = true;
2230         props->completion_mode = ACTION_COMPLETION_MODE_INDIVIDUAL;
2231     }
2232 
2233     return props;
2234 }
2235 
2236 /*
2237  * 'Atomic' group operations.  The operations are performed as a set, and if
2238  * any fail then we roll back all operations in the group.
2239  */
2240 void qmp_transaction(TransactionActionList *dev_list,
2241                      bool has_props,
2242                      struct TransactionProperties *props,
2243                      Error **errp)
2244 {
2245     TransactionActionList *dev_entry = dev_list;
2246     BlockJobTxn *block_job_txn = NULL;
2247     BlkActionState *state, *next;
2248     Error *local_err = NULL;
2249 
2250     QSIMPLEQ_HEAD(snap_bdrv_states, BlkActionState) snap_bdrv_states;
2251     QSIMPLEQ_INIT(&snap_bdrv_states);
2252 
2253     /* Does this transaction get canceled as a group on failure?
2254      * If not, we don't really need to make a BlockJobTxn.
2255      */
2256     props = get_transaction_properties(props);
2257     if (props->completion_mode != ACTION_COMPLETION_MODE_INDIVIDUAL) {
2258         block_job_txn = block_job_txn_new();
2259     }
2260 
2261     /* drain all i/o before any operations */
2262     bdrv_drain_all();
2263 
2264     /* We don't do anything in this loop that commits us to the operations */
2265     while (NULL != dev_entry) {
2266         TransactionAction *dev_info = NULL;
2267         const BlkActionOps *ops;
2268 
2269         dev_info = dev_entry->value;
2270         dev_entry = dev_entry->next;
2271 
2272         assert(dev_info->type < ARRAY_SIZE(actions));
2273 
2274         ops = &actions[dev_info->type];
2275         assert(ops->instance_size > 0);
2276 
2277         state = g_malloc0(ops->instance_size);
2278         state->ops = ops;
2279         state->action = dev_info;
2280         state->block_job_txn = block_job_txn;
2281         state->txn_props = props;
2282         QSIMPLEQ_INSERT_TAIL(&snap_bdrv_states, state, entry);
2283 
2284         state->ops->prepare(state, &local_err);
2285         if (local_err) {
2286             error_propagate(errp, local_err);
2287             goto delete_and_fail;
2288         }
2289     }
2290 
2291     QSIMPLEQ_FOREACH(state, &snap_bdrv_states, entry) {
2292         if (state->ops->commit) {
2293             state->ops->commit(state);
2294         }
2295     }
2296 
2297     /* success */
2298     goto exit;
2299 
2300 delete_and_fail:
2301     /* failure, and it is all-or-none; roll back all operations */
2302     QSIMPLEQ_FOREACH(state, &snap_bdrv_states, entry) {
2303         if (state->ops->abort) {
2304             state->ops->abort(state);
2305         }
2306     }
2307 exit:
2308     QSIMPLEQ_FOREACH_SAFE(state, &snap_bdrv_states, entry, next) {
2309         if (state->ops->clean) {
2310             state->ops->clean(state);
2311         }
2312         g_free(state);
2313     }
2314     if (!has_props) {
2315         qapi_free_TransactionProperties(props);
2316     }
2317     block_job_txn_unref(block_job_txn);
2318 }
2319 
2320 void qmp_eject(bool has_device, const char *device,
2321                bool has_id, const char *id,
2322                bool has_force, bool force, Error **errp)
2323 {
2324     Error *local_err = NULL;
2325     int rc;
2326 
2327     if (!has_force) {
2328         force = false;
2329     }
2330 
2331     rc = do_open_tray(has_device ? device : NULL,
2332                       has_id ? id : NULL,
2333                       force, &local_err);
2334     if (rc && rc != -ENOSYS) {
2335         error_propagate(errp, local_err);
2336         return;
2337     }
2338     error_free(local_err);
2339 
2340     blockdev_remove_medium(has_device, device, has_id, id, errp);
2341 }
2342 
2343 void qmp_block_passwd(bool has_device, const char *device,
2344                       bool has_node_name, const char *node_name,
2345                       const char *password, Error **errp)
2346 {
2347     error_setg(errp,
2348                "Setting block passwords directly is no longer supported");
2349 }
2350 
2351 /*
2352  * Attempt to open the tray of @device.
2353  * If @force, ignore its tray lock.
2354  * Else, if the tray is locked, don't open it, but ask the guest to open it.
2355  * On error, store an error through @errp and return -errno.
2356  * If @device does not exist, return -ENODEV.
2357  * If it has no removable media, return -ENOTSUP.
2358  * If it has no tray, return -ENOSYS.
2359  * If the guest was asked to open the tray, return -EINPROGRESS.
2360  * Else, return 0.
2361  */
2362 static int do_open_tray(const char *blk_name, const char *qdev_id,
2363                         bool force, Error **errp)
2364 {
2365     BlockBackend *blk;
2366     const char *device = qdev_id ?: blk_name;
2367     bool locked;
2368 
2369     blk = qmp_get_blk(blk_name, qdev_id, errp);
2370     if (!blk) {
2371         return -ENODEV;
2372     }
2373 
2374     if (!blk_dev_has_removable_media(blk)) {
2375         error_setg(errp, "Device '%s' is not removable", device);
2376         return -ENOTSUP;
2377     }
2378 
2379     if (!blk_dev_has_tray(blk)) {
2380         error_setg(errp, "Device '%s' does not have a tray", device);
2381         return -ENOSYS;
2382     }
2383 
2384     if (blk_dev_is_tray_open(blk)) {
2385         return 0;
2386     }
2387 
2388     locked = blk_dev_is_medium_locked(blk);
2389     if (locked) {
2390         blk_dev_eject_request(blk, force);
2391     }
2392 
2393     if (!locked || force) {
2394         blk_dev_change_media_cb(blk, false, &error_abort);
2395     }
2396 
2397     if (locked && !force) {
2398         error_setg(errp, "Device '%s' is locked and force was not specified, "
2399                    "wait for tray to open and try again", device);
2400         return -EINPROGRESS;
2401     }
2402 
2403     return 0;
2404 }
2405 
2406 void qmp_blockdev_open_tray(bool has_device, const char *device,
2407                             bool has_id, const char *id,
2408                             bool has_force, bool force,
2409                             Error **errp)
2410 {
2411     Error *local_err = NULL;
2412     int rc;
2413 
2414     if (!has_force) {
2415         force = false;
2416     }
2417     rc = do_open_tray(has_device ? device : NULL,
2418                       has_id ? id : NULL,
2419                       force, &local_err);
2420     if (rc && rc != -ENOSYS && rc != -EINPROGRESS) {
2421         error_propagate(errp, local_err);
2422         return;
2423     }
2424     error_free(local_err);
2425 }
2426 
2427 void qmp_blockdev_close_tray(bool has_device, const char *device,
2428                              bool has_id, const char *id,
2429                              Error **errp)
2430 {
2431     BlockBackend *blk;
2432     Error *local_err = NULL;
2433 
2434     device = has_device ? device : NULL;
2435     id = has_id ? id : NULL;
2436 
2437     blk = qmp_get_blk(device, id, errp);
2438     if (!blk) {
2439         return;
2440     }
2441 
2442     if (!blk_dev_has_removable_media(blk)) {
2443         error_setg(errp, "Device '%s' is not removable", device ?: id);
2444         return;
2445     }
2446 
2447     if (!blk_dev_has_tray(blk)) {
2448         /* Ignore this command on tray-less devices */
2449         return;
2450     }
2451 
2452     if (!blk_dev_is_tray_open(blk)) {
2453         return;
2454     }
2455 
2456     blk_dev_change_media_cb(blk, true, &local_err);
2457     if (local_err) {
2458         error_propagate(errp, local_err);
2459         return;
2460     }
2461 }
2462 
2463 static void blockdev_remove_medium(bool has_device, const char *device,
2464                                    bool has_id, const char *id, Error **errp)
2465 {
2466     BlockBackend *blk;
2467     BlockDriverState *bs;
2468     AioContext *aio_context;
2469     bool has_attached_device;
2470 
2471     device = has_device ? device : NULL;
2472     id = has_id ? id : NULL;
2473 
2474     blk = qmp_get_blk(device, id, errp);
2475     if (!blk) {
2476         return;
2477     }
2478 
2479     /* For BBs without a device, we can exchange the BDS tree at will */
2480     has_attached_device = blk_get_attached_dev(blk);
2481 
2482     if (has_attached_device && !blk_dev_has_removable_media(blk)) {
2483         error_setg(errp, "Device '%s' is not removable", device ?: id);
2484         return;
2485     }
2486 
2487     if (has_attached_device && blk_dev_has_tray(blk) &&
2488         !blk_dev_is_tray_open(blk))
2489     {
2490         error_setg(errp, "Tray of device '%s' is not open", device ?: id);
2491         return;
2492     }
2493 
2494     bs = blk_bs(blk);
2495     if (!bs) {
2496         return;
2497     }
2498 
2499     aio_context = bdrv_get_aio_context(bs);
2500     aio_context_acquire(aio_context);
2501 
2502     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_EJECT, errp)) {
2503         goto out;
2504     }
2505 
2506     blk_remove_bs(blk);
2507 
2508     if (!blk_dev_has_tray(blk)) {
2509         /* For tray-less devices, blockdev-open-tray is a no-op (or may not be
2510          * called at all); therefore, the medium needs to be ejected here.
2511          * Do it after blk_remove_bs() so blk_is_inserted(blk) returns the @load
2512          * value passed here (i.e. false). */
2513         blk_dev_change_media_cb(blk, false, &error_abort);
2514     }
2515 
2516 out:
2517     aio_context_release(aio_context);
2518 }
2519 
2520 void qmp_blockdev_remove_medium(const char *id, Error **errp)
2521 {
2522     blockdev_remove_medium(false, NULL, true, id, errp);
2523 }
2524 
2525 static void qmp_blockdev_insert_anon_medium(BlockBackend *blk,
2526                                             BlockDriverState *bs, Error **errp)
2527 {
2528     Error *local_err = NULL;
2529     bool has_device;
2530     int ret;
2531 
2532     /* For BBs without a device, we can exchange the BDS tree at will */
2533     has_device = blk_get_attached_dev(blk);
2534 
2535     if (has_device && !blk_dev_has_removable_media(blk)) {
2536         error_setg(errp, "Device is not removable");
2537         return;
2538     }
2539 
2540     if (has_device && blk_dev_has_tray(blk) && !blk_dev_is_tray_open(blk)) {
2541         error_setg(errp, "Tray of the device is not open");
2542         return;
2543     }
2544 
2545     if (blk_bs(blk)) {
2546         error_setg(errp, "There already is a medium in the device");
2547         return;
2548     }
2549 
2550     ret = blk_insert_bs(blk, bs, errp);
2551     if (ret < 0) {
2552         return;
2553     }
2554 
2555     if (!blk_dev_has_tray(blk)) {
2556         /* For tray-less devices, blockdev-close-tray is a no-op (or may not be
2557          * called at all); therefore, the medium needs to be pushed into the
2558          * slot here.
2559          * Do it after blk_insert_bs() so blk_is_inserted(blk) returns the @load
2560          * value passed here (i.e. true). */
2561         blk_dev_change_media_cb(blk, true, &local_err);
2562         if (local_err) {
2563             error_propagate(errp, local_err);
2564             blk_remove_bs(blk);
2565             return;
2566         }
2567     }
2568 }
2569 
2570 static void blockdev_insert_medium(bool has_device, const char *device,
2571                                    bool has_id, const char *id,
2572                                    const char *node_name, Error **errp)
2573 {
2574     BlockBackend *blk;
2575     BlockDriverState *bs;
2576 
2577     blk = qmp_get_blk(has_device ? device : NULL,
2578                       has_id ? id : NULL,
2579                       errp);
2580     if (!blk) {
2581         return;
2582     }
2583 
2584     bs = bdrv_find_node(node_name);
2585     if (!bs) {
2586         error_setg(errp, "Node '%s' not found", node_name);
2587         return;
2588     }
2589 
2590     if (bdrv_has_blk(bs)) {
2591         error_setg(errp, "Node '%s' is already in use", node_name);
2592         return;
2593     }
2594 
2595     qmp_blockdev_insert_anon_medium(blk, bs, errp);
2596 }
2597 
2598 void qmp_blockdev_insert_medium(const char *id, const char *node_name,
2599                                 Error **errp)
2600 {
2601     blockdev_insert_medium(false, NULL, true, id, node_name, errp);
2602 }
2603 
2604 void qmp_blockdev_change_medium(bool has_device, const char *device,
2605                                 bool has_id, const char *id,
2606                                 const char *filename,
2607                                 bool has_format, const char *format,
2608                                 bool has_read_only,
2609                                 BlockdevChangeReadOnlyMode read_only,
2610                                 Error **errp)
2611 {
2612     BlockBackend *blk;
2613     BlockDriverState *medium_bs = NULL;
2614     int bdrv_flags;
2615     bool detect_zeroes;
2616     int rc;
2617     QDict *options = NULL;
2618     Error *err = NULL;
2619 
2620     blk = qmp_get_blk(has_device ? device : NULL,
2621                       has_id ? id : NULL,
2622                       errp);
2623     if (!blk) {
2624         goto fail;
2625     }
2626 
2627     if (blk_bs(blk)) {
2628         blk_update_root_state(blk);
2629     }
2630 
2631     bdrv_flags = blk_get_open_flags_from_root_state(blk);
2632     bdrv_flags &= ~(BDRV_O_TEMPORARY | BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING |
2633         BDRV_O_PROTOCOL);
2634 
2635     if (!has_read_only) {
2636         read_only = BLOCKDEV_CHANGE_READ_ONLY_MODE_RETAIN;
2637     }
2638 
2639     switch (read_only) {
2640     case BLOCKDEV_CHANGE_READ_ONLY_MODE_RETAIN:
2641         break;
2642 
2643     case BLOCKDEV_CHANGE_READ_ONLY_MODE_READ_ONLY:
2644         bdrv_flags &= ~BDRV_O_RDWR;
2645         break;
2646 
2647     case BLOCKDEV_CHANGE_READ_ONLY_MODE_READ_WRITE:
2648         bdrv_flags |= BDRV_O_RDWR;
2649         break;
2650 
2651     default:
2652         abort();
2653     }
2654 
2655     options = qdict_new();
2656     detect_zeroes = blk_get_detect_zeroes_from_root_state(blk);
2657     qdict_put_str(options, "detect-zeroes", detect_zeroes ? "on" : "off");
2658 
2659     if (has_format) {
2660         qdict_put_str(options, "driver", format);
2661     }
2662 
2663     medium_bs = bdrv_open(filename, NULL, options, bdrv_flags, errp);
2664     if (!medium_bs) {
2665         goto fail;
2666     }
2667 
2668     rc = do_open_tray(has_device ? device : NULL,
2669                       has_id ? id : NULL,
2670                       false, &err);
2671     if (rc && rc != -ENOSYS) {
2672         error_propagate(errp, err);
2673         goto fail;
2674     }
2675     error_free(err);
2676     err = NULL;
2677 
2678     blockdev_remove_medium(has_device, device, has_id, id, &err);
2679     if (err) {
2680         error_propagate(errp, err);
2681         goto fail;
2682     }
2683 
2684     qmp_blockdev_insert_anon_medium(blk, medium_bs, &err);
2685     if (err) {
2686         error_propagate(errp, err);
2687         goto fail;
2688     }
2689 
2690     qmp_blockdev_close_tray(has_device, device, has_id, id, errp);
2691 
2692 fail:
2693     /* If the medium has been inserted, the device has its own reference, so
2694      * ours must be relinquished; and if it has not been inserted successfully,
2695      * the reference must be relinquished anyway */
2696     bdrv_unref(medium_bs);
2697 }
2698 
2699 /* throttling disk I/O limits */
2700 void qmp_block_set_io_throttle(BlockIOThrottle *arg, Error **errp)
2701 {
2702     ThrottleConfig cfg;
2703     BlockDriverState *bs;
2704     BlockBackend *blk;
2705     AioContext *aio_context;
2706 
2707     blk = qmp_get_blk(arg->has_device ? arg->device : NULL,
2708                       arg->has_id ? arg->id : NULL,
2709                       errp);
2710     if (!blk) {
2711         return;
2712     }
2713 
2714     aio_context = blk_get_aio_context(blk);
2715     aio_context_acquire(aio_context);
2716 
2717     bs = blk_bs(blk);
2718     if (!bs) {
2719         error_setg(errp, "Device has no medium");
2720         goto out;
2721     }
2722 
2723     throttle_config_init(&cfg);
2724     cfg.buckets[THROTTLE_BPS_TOTAL].avg = arg->bps;
2725     cfg.buckets[THROTTLE_BPS_READ].avg  = arg->bps_rd;
2726     cfg.buckets[THROTTLE_BPS_WRITE].avg = arg->bps_wr;
2727 
2728     cfg.buckets[THROTTLE_OPS_TOTAL].avg = arg->iops;
2729     cfg.buckets[THROTTLE_OPS_READ].avg  = arg->iops_rd;
2730     cfg.buckets[THROTTLE_OPS_WRITE].avg = arg->iops_wr;
2731 
2732     if (arg->has_bps_max) {
2733         cfg.buckets[THROTTLE_BPS_TOTAL].max = arg->bps_max;
2734     }
2735     if (arg->has_bps_rd_max) {
2736         cfg.buckets[THROTTLE_BPS_READ].max = arg->bps_rd_max;
2737     }
2738     if (arg->has_bps_wr_max) {
2739         cfg.buckets[THROTTLE_BPS_WRITE].max = arg->bps_wr_max;
2740     }
2741     if (arg->has_iops_max) {
2742         cfg.buckets[THROTTLE_OPS_TOTAL].max = arg->iops_max;
2743     }
2744     if (arg->has_iops_rd_max) {
2745         cfg.buckets[THROTTLE_OPS_READ].max = arg->iops_rd_max;
2746     }
2747     if (arg->has_iops_wr_max) {
2748         cfg.buckets[THROTTLE_OPS_WRITE].max = arg->iops_wr_max;
2749     }
2750 
2751     if (arg->has_bps_max_length) {
2752         cfg.buckets[THROTTLE_BPS_TOTAL].burst_length = arg->bps_max_length;
2753     }
2754     if (arg->has_bps_rd_max_length) {
2755         cfg.buckets[THROTTLE_BPS_READ].burst_length = arg->bps_rd_max_length;
2756     }
2757     if (arg->has_bps_wr_max_length) {
2758         cfg.buckets[THROTTLE_BPS_WRITE].burst_length = arg->bps_wr_max_length;
2759     }
2760     if (arg->has_iops_max_length) {
2761         cfg.buckets[THROTTLE_OPS_TOTAL].burst_length = arg->iops_max_length;
2762     }
2763     if (arg->has_iops_rd_max_length) {
2764         cfg.buckets[THROTTLE_OPS_READ].burst_length = arg->iops_rd_max_length;
2765     }
2766     if (arg->has_iops_wr_max_length) {
2767         cfg.buckets[THROTTLE_OPS_WRITE].burst_length = arg->iops_wr_max_length;
2768     }
2769 
2770     if (arg->has_iops_size) {
2771         cfg.op_size = arg->iops_size;
2772     }
2773 
2774     if (!throttle_is_valid(&cfg, errp)) {
2775         goto out;
2776     }
2777 
2778     if (throttle_enabled(&cfg)) {
2779         /* Enable I/O limits if they're not enabled yet, otherwise
2780          * just update the throttling group. */
2781         if (!blk_get_public(blk)->throttle_group_member.throttle_state) {
2782             blk_io_limits_enable(blk,
2783                                  arg->has_group ? arg->group :
2784                                  arg->has_device ? arg->device :
2785                                  arg->id);
2786         } else if (arg->has_group) {
2787             blk_io_limits_update_group(blk, arg->group);
2788         }
2789         /* Set the new throttling configuration */
2790         blk_set_io_limits(blk, &cfg);
2791     } else if (blk_get_public(blk)->throttle_group_member.throttle_state) {
2792         /* If all throttling settings are set to 0, disable I/O limits */
2793         blk_io_limits_disable(blk);
2794     }
2795 
2796 out:
2797     aio_context_release(aio_context);
2798 }
2799 
2800 void qmp_block_dirty_bitmap_add(const char *node, const char *name,
2801                                 bool has_granularity, uint32_t granularity,
2802                                 bool has_persistent, bool persistent,
2803                                 bool has_autoload, bool autoload,
2804                                 Error **errp)
2805 {
2806     BlockDriverState *bs;
2807     BdrvDirtyBitmap *bitmap;
2808 
2809     if (!name || name[0] == '\0') {
2810         error_setg(errp, "Bitmap name cannot be empty");
2811         return;
2812     }
2813 
2814     bs = bdrv_lookup_bs(node, node, errp);
2815     if (!bs) {
2816         return;
2817     }
2818 
2819     if (has_granularity) {
2820         if (granularity < 512 || !is_power_of_2(granularity)) {
2821             error_setg(errp, "Granularity must be power of 2 "
2822                              "and at least 512");
2823             return;
2824         }
2825     } else {
2826         /* Default to cluster size, if available: */
2827         granularity = bdrv_get_default_bitmap_granularity(bs);
2828     }
2829 
2830     if (!has_persistent) {
2831         persistent = false;
2832     }
2833 
2834     if (has_autoload) {
2835         warn_report("Autoload option is deprecated and its value is ignored");
2836     }
2837 
2838     if (persistent &&
2839         !bdrv_can_store_new_dirty_bitmap(bs, name, granularity, errp))
2840     {
2841         return;
2842     }
2843 
2844     bitmap = bdrv_create_dirty_bitmap(bs, granularity, name, errp);
2845     if (bitmap == NULL) {
2846         return;
2847     }
2848 
2849     bdrv_dirty_bitmap_set_persistance(bitmap, persistent);
2850 }
2851 
2852 void qmp_block_dirty_bitmap_remove(const char *node, const char *name,
2853                                    Error **errp)
2854 {
2855     BlockDriverState *bs;
2856     BdrvDirtyBitmap *bitmap;
2857     Error *local_err = NULL;
2858 
2859     bitmap = block_dirty_bitmap_lookup(node, name, &bs, errp);
2860     if (!bitmap || !bs) {
2861         return;
2862     }
2863 
2864     if (bdrv_dirty_bitmap_frozen(bitmap)) {
2865         error_setg(errp,
2866                    "Bitmap '%s' is currently frozen and cannot be removed",
2867                    name);
2868         return;
2869     } else if (bdrv_dirty_bitmap_qmp_locked(bitmap)) {
2870         error_setg(errp,
2871                    "Bitmap '%s' is currently locked and cannot be removed",
2872                    name);
2873         return;
2874     }
2875 
2876     if (bdrv_dirty_bitmap_get_persistance(bitmap)) {
2877         bdrv_remove_persistent_dirty_bitmap(bs, name, &local_err);
2878         if (local_err != NULL) {
2879             error_propagate(errp, local_err);
2880             return;
2881         }
2882     }
2883 
2884     bdrv_dirty_bitmap_make_anon(bitmap);
2885     bdrv_release_dirty_bitmap(bs, bitmap);
2886 }
2887 
2888 /**
2889  * Completely clear a bitmap, for the purposes of synchronizing a bitmap
2890  * immediately after a full backup operation.
2891  */
2892 void qmp_block_dirty_bitmap_clear(const char *node, const char *name,
2893                                   Error **errp)
2894 {
2895     BdrvDirtyBitmap *bitmap;
2896     BlockDriverState *bs;
2897 
2898     bitmap = block_dirty_bitmap_lookup(node, name, &bs, errp);
2899     if (!bitmap || !bs) {
2900         return;
2901     }
2902 
2903     if (bdrv_dirty_bitmap_frozen(bitmap)) {
2904         error_setg(errp,
2905                    "Bitmap '%s' is currently frozen and cannot be modified",
2906                    name);
2907         return;
2908     } else if (bdrv_dirty_bitmap_qmp_locked(bitmap)) {
2909         error_setg(errp,
2910                    "Bitmap '%s' is currently locked and cannot be modified",
2911                    name);
2912         return;
2913     } else if (!bdrv_dirty_bitmap_enabled(bitmap)) {
2914         error_setg(errp,
2915                    "Bitmap '%s' is currently disabled and cannot be cleared",
2916                    name);
2917         return;
2918     } else if (bdrv_dirty_bitmap_readonly(bitmap)) {
2919         error_setg(errp, "Bitmap '%s' is readonly and cannot be cleared", name);
2920         return;
2921     }
2922 
2923     bdrv_clear_dirty_bitmap(bitmap, NULL);
2924 }
2925 
2926 BlockDirtyBitmapSha256 *qmp_x_debug_block_dirty_bitmap_sha256(const char *node,
2927                                                               const char *name,
2928                                                               Error **errp)
2929 {
2930     BdrvDirtyBitmap *bitmap;
2931     BlockDriverState *bs;
2932     BlockDirtyBitmapSha256 *ret = NULL;
2933     char *sha256;
2934 
2935     bitmap = block_dirty_bitmap_lookup(node, name, &bs, errp);
2936     if (!bitmap || !bs) {
2937         return NULL;
2938     }
2939 
2940     sha256 = bdrv_dirty_bitmap_sha256(bitmap, errp);
2941     if (sha256 == NULL) {
2942         return NULL;
2943     }
2944 
2945     ret = g_new(BlockDirtyBitmapSha256, 1);
2946     ret->sha256 = sha256;
2947 
2948     return ret;
2949 }
2950 
2951 void hmp_drive_del(Monitor *mon, const QDict *qdict)
2952 {
2953     const char *id = qdict_get_str(qdict, "id");
2954     BlockBackend *blk;
2955     BlockDriverState *bs;
2956     AioContext *aio_context;
2957     Error *local_err = NULL;
2958 
2959     bs = bdrv_find_node(id);
2960     if (bs) {
2961         qmp_blockdev_del(id, &local_err);
2962         if (local_err) {
2963             error_report_err(local_err);
2964         }
2965         return;
2966     }
2967 
2968     blk = blk_by_name(id);
2969     if (!blk) {
2970         error_report("Device '%s' not found", id);
2971         return;
2972     }
2973 
2974     if (!blk_legacy_dinfo(blk)) {
2975         error_report("Deleting device added with blockdev-add"
2976                      " is not supported");
2977         return;
2978     }
2979 
2980     aio_context = blk_get_aio_context(blk);
2981     aio_context_acquire(aio_context);
2982 
2983     bs = blk_bs(blk);
2984     if (bs) {
2985         if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_DRIVE_DEL, &local_err)) {
2986             error_report_err(local_err);
2987             aio_context_release(aio_context);
2988             return;
2989         }
2990 
2991         blk_remove_bs(blk);
2992     }
2993 
2994     /* Make the BlockBackend and the attached BlockDriverState anonymous */
2995     monitor_remove_blk(blk);
2996 
2997     /* If this BlockBackend has a device attached to it, its refcount will be
2998      * decremented when the device is removed; otherwise we have to do so here.
2999      */
3000     if (blk_get_attached_dev(blk)) {
3001         /* Further I/O must not pause the guest */
3002         blk_set_on_error(blk, BLOCKDEV_ON_ERROR_REPORT,
3003                          BLOCKDEV_ON_ERROR_REPORT);
3004     } else {
3005         blk_unref(blk);
3006     }
3007 
3008     aio_context_release(aio_context);
3009 }
3010 
3011 void qmp_block_resize(bool has_device, const char *device,
3012                       bool has_node_name, const char *node_name,
3013                       int64_t size, Error **errp)
3014 {
3015     Error *local_err = NULL;
3016     BlockBackend *blk = NULL;
3017     BlockDriverState *bs;
3018     AioContext *aio_context;
3019     int ret;
3020 
3021     bs = bdrv_lookup_bs(has_device ? device : NULL,
3022                         has_node_name ? node_name : NULL,
3023                         &local_err);
3024     if (local_err) {
3025         error_propagate(errp, local_err);
3026         return;
3027     }
3028 
3029     aio_context = bdrv_get_aio_context(bs);
3030     aio_context_acquire(aio_context);
3031 
3032     if (!bdrv_is_first_non_filter(bs)) {
3033         error_setg(errp, QERR_FEATURE_DISABLED, "resize");
3034         goto out;
3035     }
3036 
3037     if (size < 0) {
3038         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "size", "a >0 size");
3039         goto out;
3040     }
3041 
3042     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_RESIZE, NULL)) {
3043         error_setg(errp, QERR_DEVICE_IN_USE, device);
3044         goto out;
3045     }
3046 
3047     blk = blk_new(BLK_PERM_RESIZE, BLK_PERM_ALL);
3048     ret = blk_insert_bs(blk, bs, errp);
3049     if (ret < 0) {
3050         goto out;
3051     }
3052 
3053     bdrv_drained_begin(bs);
3054     ret = blk_truncate(blk, size, PREALLOC_MODE_OFF, errp);
3055     bdrv_drained_end(bs);
3056 
3057 out:
3058     blk_unref(blk);
3059     aio_context_release(aio_context);
3060 }
3061 
3062 void qmp_block_stream(bool has_job_id, const char *job_id, const char *device,
3063                       bool has_base, const char *base,
3064                       bool has_base_node, const char *base_node,
3065                       bool has_backing_file, const char *backing_file,
3066                       bool has_speed, int64_t speed,
3067                       bool has_on_error, BlockdevOnError on_error,
3068                       Error **errp)
3069 {
3070     BlockDriverState *bs, *iter;
3071     BlockDriverState *base_bs = NULL;
3072     AioContext *aio_context;
3073     Error *local_err = NULL;
3074     const char *base_name = NULL;
3075 
3076     if (!has_on_error) {
3077         on_error = BLOCKDEV_ON_ERROR_REPORT;
3078     }
3079 
3080     bs = bdrv_lookup_bs(device, device, errp);
3081     if (!bs) {
3082         return;
3083     }
3084 
3085     aio_context = bdrv_get_aio_context(bs);
3086     aio_context_acquire(aio_context);
3087 
3088     if (has_base && has_base_node) {
3089         error_setg(errp, "'base' and 'base-node' cannot be specified "
3090                    "at the same time");
3091         goto out;
3092     }
3093 
3094     if (has_base) {
3095         base_bs = bdrv_find_backing_image(bs, base);
3096         if (base_bs == NULL) {
3097             error_setg(errp, QERR_BASE_NOT_FOUND, base);
3098             goto out;
3099         }
3100         assert(bdrv_get_aio_context(base_bs) == aio_context);
3101         base_name = base;
3102     }
3103 
3104     if (has_base_node) {
3105         base_bs = bdrv_lookup_bs(NULL, base_node, errp);
3106         if (!base_bs) {
3107             goto out;
3108         }
3109         if (bs == base_bs || !bdrv_chain_contains(bs, base_bs)) {
3110             error_setg(errp, "Node '%s' is not a backing image of '%s'",
3111                        base_node, device);
3112             goto out;
3113         }
3114         assert(bdrv_get_aio_context(base_bs) == aio_context);
3115         base_name = base_bs->filename;
3116     }
3117 
3118     /* Check for op blockers in the whole chain between bs and base */
3119     for (iter = bs; iter && iter != base_bs; iter = backing_bs(iter)) {
3120         if (bdrv_op_is_blocked(iter, BLOCK_OP_TYPE_STREAM, errp)) {
3121             goto out;
3122         }
3123     }
3124 
3125     /* if we are streaming the entire chain, the result will have no backing
3126      * file, and specifying one is therefore an error */
3127     if (base_bs == NULL && has_backing_file) {
3128         error_setg(errp, "backing file specified, but streaming the "
3129                          "entire chain");
3130         goto out;
3131     }
3132 
3133     /* backing_file string overrides base bs filename */
3134     base_name = has_backing_file ? backing_file : base_name;
3135 
3136     stream_start(has_job_id ? job_id : NULL, bs, base_bs, base_name,
3137                  has_speed ? speed : 0, on_error, &local_err);
3138     if (local_err) {
3139         error_propagate(errp, local_err);
3140         goto out;
3141     }
3142 
3143     trace_qmp_block_stream(bs, bs->job);
3144 
3145 out:
3146     aio_context_release(aio_context);
3147 }
3148 
3149 void qmp_block_commit(bool has_job_id, const char *job_id, const char *device,
3150                       bool has_base, const char *base,
3151                       bool has_top, const char *top,
3152                       bool has_backing_file, const char *backing_file,
3153                       bool has_speed, int64_t speed,
3154                       bool has_filter_node_name, const char *filter_node_name,
3155                       Error **errp)
3156 {
3157     BlockDriverState *bs;
3158     BlockDriverState *iter;
3159     BlockDriverState *base_bs, *top_bs;
3160     AioContext *aio_context;
3161     Error *local_err = NULL;
3162     /* This will be part of the QMP command, if/when the
3163      * BlockdevOnError change for blkmirror makes it in
3164      */
3165     BlockdevOnError on_error = BLOCKDEV_ON_ERROR_REPORT;
3166 
3167     if (!has_speed) {
3168         speed = 0;
3169     }
3170     if (!has_filter_node_name) {
3171         filter_node_name = NULL;
3172     }
3173 
3174     /* Important Note:
3175      *  libvirt relies on the DeviceNotFound error class in order to probe for
3176      *  live commit feature versions; for this to work, we must make sure to
3177      *  perform the device lookup before any generic errors that may occur in a
3178      *  scenario in which all optional arguments are omitted. */
3179     bs = qmp_get_root_bs(device, &local_err);
3180     if (!bs) {
3181         bs = bdrv_lookup_bs(device, device, NULL);
3182         if (!bs) {
3183             error_free(local_err);
3184             error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
3185                       "Device '%s' not found", device);
3186         } else {
3187             error_propagate(errp, local_err);
3188         }
3189         return;
3190     }
3191 
3192     aio_context = bdrv_get_aio_context(bs);
3193     aio_context_acquire(aio_context);
3194 
3195     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_COMMIT_SOURCE, errp)) {
3196         goto out;
3197     }
3198 
3199     /* default top_bs is the active layer */
3200     top_bs = bs;
3201 
3202     if (has_top && top) {
3203         if (strcmp(bs->filename, top) != 0) {
3204             top_bs = bdrv_find_backing_image(bs, top);
3205         }
3206     }
3207 
3208     if (top_bs == NULL) {
3209         error_setg(errp, "Top image file %s not found", top ? top : "NULL");
3210         goto out;
3211     }
3212 
3213     assert(bdrv_get_aio_context(top_bs) == aio_context);
3214 
3215     if (has_base && base) {
3216         base_bs = bdrv_find_backing_image(top_bs, base);
3217     } else {
3218         base_bs = bdrv_find_base(top_bs);
3219     }
3220 
3221     if (base_bs == NULL) {
3222         error_setg(errp, QERR_BASE_NOT_FOUND, base ? base : "NULL");
3223         goto out;
3224     }
3225 
3226     assert(bdrv_get_aio_context(base_bs) == aio_context);
3227 
3228     for (iter = top_bs; iter != backing_bs(base_bs); iter = backing_bs(iter)) {
3229         if (bdrv_op_is_blocked(iter, BLOCK_OP_TYPE_COMMIT_TARGET, errp)) {
3230             goto out;
3231         }
3232     }
3233 
3234     /* Do not allow attempts to commit an image into itself */
3235     if (top_bs == base_bs) {
3236         error_setg(errp, "cannot commit an image into itself");
3237         goto out;
3238     }
3239 
3240     if (top_bs == bs) {
3241         if (has_backing_file) {
3242             error_setg(errp, "'backing-file' specified,"
3243                              " but 'top' is the active layer");
3244             goto out;
3245         }
3246         commit_active_start(has_job_id ? job_id : NULL, bs, base_bs,
3247                             BLOCK_JOB_DEFAULT, speed, on_error,
3248                             filter_node_name, NULL, NULL, false, &local_err);
3249     } else {
3250         BlockDriverState *overlay_bs = bdrv_find_overlay(bs, top_bs);
3251         if (bdrv_op_is_blocked(overlay_bs, BLOCK_OP_TYPE_COMMIT_TARGET, errp)) {
3252             goto out;
3253         }
3254         commit_start(has_job_id ? job_id : NULL, bs, base_bs, top_bs, speed,
3255                      on_error, has_backing_file ? backing_file : NULL,
3256                      filter_node_name, &local_err);
3257     }
3258     if (local_err != NULL) {
3259         error_propagate(errp, local_err);
3260         goto out;
3261     }
3262 
3263 out:
3264     aio_context_release(aio_context);
3265 }
3266 
3267 static BlockJob *do_drive_backup(DriveBackup *backup, BlockJobTxn *txn,
3268                                  Error **errp)
3269 {
3270     BlockDriverState *bs;
3271     BlockDriverState *target_bs;
3272     BlockDriverState *source = NULL;
3273     BlockJob *job = NULL;
3274     BdrvDirtyBitmap *bmap = NULL;
3275     AioContext *aio_context;
3276     QDict *options = NULL;
3277     Error *local_err = NULL;
3278     int flags, job_flags = BLOCK_JOB_DEFAULT;
3279     int64_t size;
3280     bool set_backing_hd = false;
3281 
3282     if (!backup->has_speed) {
3283         backup->speed = 0;
3284     }
3285     if (!backup->has_on_source_error) {
3286         backup->on_source_error = BLOCKDEV_ON_ERROR_REPORT;
3287     }
3288     if (!backup->has_on_target_error) {
3289         backup->on_target_error = BLOCKDEV_ON_ERROR_REPORT;
3290     }
3291     if (!backup->has_mode) {
3292         backup->mode = NEW_IMAGE_MODE_ABSOLUTE_PATHS;
3293     }
3294     if (!backup->has_job_id) {
3295         backup->job_id = NULL;
3296     }
3297     if (!backup->has_auto_finalize) {
3298         backup->auto_finalize = true;
3299     }
3300     if (!backup->has_auto_dismiss) {
3301         backup->auto_dismiss = true;
3302     }
3303     if (!backup->has_compress) {
3304         backup->compress = false;
3305     }
3306 
3307     bs = qmp_get_root_bs(backup->device, errp);
3308     if (!bs) {
3309         return NULL;
3310     }
3311 
3312     aio_context = bdrv_get_aio_context(bs);
3313     aio_context_acquire(aio_context);
3314 
3315     if (!backup->has_format) {
3316         backup->format = backup->mode == NEW_IMAGE_MODE_EXISTING ?
3317                          NULL : (char*) bs->drv->format_name;
3318     }
3319 
3320     /* Early check to avoid creating target */
3321     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_BACKUP_SOURCE, errp)) {
3322         goto out;
3323     }
3324 
3325     flags = bs->open_flags | BDRV_O_RDWR;
3326 
3327     /* See if we have a backing HD we can use to create our new image
3328      * on top of. */
3329     if (backup->sync == MIRROR_SYNC_MODE_TOP) {
3330         source = backing_bs(bs);
3331         if (!source) {
3332             backup->sync = MIRROR_SYNC_MODE_FULL;
3333         }
3334     }
3335     if (backup->sync == MIRROR_SYNC_MODE_NONE) {
3336         source = bs;
3337         flags |= BDRV_O_NO_BACKING;
3338         set_backing_hd = true;
3339     }
3340 
3341     size = bdrv_getlength(bs);
3342     if (size < 0) {
3343         error_setg_errno(errp, -size, "bdrv_getlength failed");
3344         goto out;
3345     }
3346 
3347     if (backup->mode != NEW_IMAGE_MODE_EXISTING) {
3348         assert(backup->format);
3349         if (source) {
3350             bdrv_img_create(backup->target, backup->format, source->filename,
3351                             source->drv->format_name, NULL,
3352                             size, flags, false, &local_err);
3353         } else {
3354             bdrv_img_create(backup->target, backup->format, NULL, NULL, NULL,
3355                             size, flags, false, &local_err);
3356         }
3357     }
3358 
3359     if (local_err) {
3360         error_propagate(errp, local_err);
3361         goto out;
3362     }
3363 
3364     if (backup->format) {
3365         if (!options) {
3366             options = qdict_new();
3367         }
3368         qdict_put_str(options, "driver", backup->format);
3369     }
3370 
3371     target_bs = bdrv_open(backup->target, NULL, options, flags, errp);
3372     if (!target_bs) {
3373         goto out;
3374     }
3375 
3376     bdrv_set_aio_context(target_bs, aio_context);
3377 
3378     if (set_backing_hd) {
3379         bdrv_set_backing_hd(target_bs, source, &local_err);
3380         if (local_err) {
3381             bdrv_unref(target_bs);
3382             goto out;
3383         }
3384     }
3385 
3386     if (backup->has_bitmap) {
3387         bmap = bdrv_find_dirty_bitmap(bs, backup->bitmap);
3388         if (!bmap) {
3389             error_setg(errp, "Bitmap '%s' could not be found", backup->bitmap);
3390             bdrv_unref(target_bs);
3391             goto out;
3392         }
3393         if (bdrv_dirty_bitmap_qmp_locked(bmap)) {
3394             error_setg(errp,
3395                        "Bitmap '%s' is currently locked and cannot be used for "
3396                        "backup", backup->bitmap);
3397             goto out;
3398         }
3399     }
3400     if (!backup->auto_finalize) {
3401         job_flags |= BLOCK_JOB_MANUAL_FINALIZE;
3402     }
3403     if (!backup->auto_dismiss) {
3404         job_flags |= BLOCK_JOB_MANUAL_DISMISS;
3405     }
3406 
3407     job = backup_job_create(backup->job_id, bs, target_bs, backup->speed,
3408                             backup->sync, bmap, backup->compress,
3409                             backup->on_source_error, backup->on_target_error,
3410                             job_flags, NULL, NULL, txn, &local_err);
3411     bdrv_unref(target_bs);
3412     if (local_err != NULL) {
3413         error_propagate(errp, local_err);
3414         goto out;
3415     }
3416 
3417 out:
3418     aio_context_release(aio_context);
3419     return job;
3420 }
3421 
3422 void qmp_drive_backup(DriveBackup *arg, Error **errp)
3423 {
3424 
3425     BlockJob *job;
3426     job = do_drive_backup(arg, NULL, errp);
3427     if (job) {
3428         block_job_start(job);
3429     }
3430 }
3431 
3432 BlockDeviceInfoList *qmp_query_named_block_nodes(Error **errp)
3433 {
3434     return bdrv_named_nodes_list(errp);
3435 }
3436 
3437 BlockJob *do_blockdev_backup(BlockdevBackup *backup, BlockJobTxn *txn,
3438                              Error **errp)
3439 {
3440     BlockDriverState *bs;
3441     BlockDriverState *target_bs;
3442     Error *local_err = NULL;
3443     AioContext *aio_context;
3444     BlockJob *job = NULL;
3445     int job_flags = BLOCK_JOB_DEFAULT;
3446 
3447     if (!backup->has_speed) {
3448         backup->speed = 0;
3449     }
3450     if (!backup->has_on_source_error) {
3451         backup->on_source_error = BLOCKDEV_ON_ERROR_REPORT;
3452     }
3453     if (!backup->has_on_target_error) {
3454         backup->on_target_error = BLOCKDEV_ON_ERROR_REPORT;
3455     }
3456     if (!backup->has_job_id) {
3457         backup->job_id = NULL;
3458     }
3459     if (!backup->has_auto_finalize) {
3460         backup->auto_finalize = true;
3461     }
3462     if (!backup->has_auto_dismiss) {
3463         backup->auto_dismiss = true;
3464     }
3465     if (!backup->has_compress) {
3466         backup->compress = false;
3467     }
3468 
3469     bs = qmp_get_root_bs(backup->device, errp);
3470     if (!bs) {
3471         return NULL;
3472     }
3473 
3474     aio_context = bdrv_get_aio_context(bs);
3475     aio_context_acquire(aio_context);
3476 
3477     target_bs = bdrv_lookup_bs(backup->target, backup->target, errp);
3478     if (!target_bs) {
3479         goto out;
3480     }
3481 
3482     if (bdrv_get_aio_context(target_bs) != aio_context) {
3483         if (!bdrv_has_blk(target_bs)) {
3484             /* The target BDS is not attached, we can safely move it to another
3485              * AioContext. */
3486             bdrv_set_aio_context(target_bs, aio_context);
3487         } else {
3488             error_setg(errp, "Target is attached to a different thread from "
3489                              "source.");
3490             goto out;
3491         }
3492     }
3493     if (!backup->auto_finalize) {
3494         job_flags |= BLOCK_JOB_MANUAL_FINALIZE;
3495     }
3496     if (!backup->auto_dismiss) {
3497         job_flags |= BLOCK_JOB_MANUAL_DISMISS;
3498     }
3499     job = backup_job_create(backup->job_id, bs, target_bs, backup->speed,
3500                             backup->sync, NULL, backup->compress,
3501                             backup->on_source_error, backup->on_target_error,
3502                             job_flags, NULL, NULL, txn, &local_err);
3503     if (local_err != NULL) {
3504         error_propagate(errp, local_err);
3505     }
3506 out:
3507     aio_context_release(aio_context);
3508     return job;
3509 }
3510 
3511 void qmp_blockdev_backup(BlockdevBackup *arg, Error **errp)
3512 {
3513     BlockJob *job;
3514     job = do_blockdev_backup(arg, NULL, errp);
3515     if (job) {
3516         block_job_start(job);
3517     }
3518 }
3519 
3520 /* Parameter check and block job starting for drive mirroring.
3521  * Caller should hold @device and @target's aio context (must be the same).
3522  **/
3523 static void blockdev_mirror_common(const char *job_id, BlockDriverState *bs,
3524                                    BlockDriverState *target,
3525                                    bool has_replaces, const char *replaces,
3526                                    enum MirrorSyncMode sync,
3527                                    BlockMirrorBackingMode backing_mode,
3528                                    bool has_speed, int64_t speed,
3529                                    bool has_granularity, uint32_t granularity,
3530                                    bool has_buf_size, int64_t buf_size,
3531                                    bool has_on_source_error,
3532                                    BlockdevOnError on_source_error,
3533                                    bool has_on_target_error,
3534                                    BlockdevOnError on_target_error,
3535                                    bool has_unmap, bool unmap,
3536                                    bool has_filter_node_name,
3537                                    const char *filter_node_name,
3538                                    Error **errp)
3539 {
3540 
3541     if (!has_speed) {
3542         speed = 0;
3543     }
3544     if (!has_on_source_error) {
3545         on_source_error = BLOCKDEV_ON_ERROR_REPORT;
3546     }
3547     if (!has_on_target_error) {
3548         on_target_error = BLOCKDEV_ON_ERROR_REPORT;
3549     }
3550     if (!has_granularity) {
3551         granularity = 0;
3552     }
3553     if (!has_buf_size) {
3554         buf_size = 0;
3555     }
3556     if (!has_unmap) {
3557         unmap = true;
3558     }
3559     if (!has_filter_node_name) {
3560         filter_node_name = NULL;
3561     }
3562 
3563     if (granularity != 0 && (granularity < 512 || granularity > 1048576 * 64)) {
3564         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "granularity",
3565                    "a value in range [512B, 64MB]");
3566         return;
3567     }
3568     if (granularity & (granularity - 1)) {
3569         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "granularity",
3570                    "power of 2");
3571         return;
3572     }
3573 
3574     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_MIRROR_SOURCE, errp)) {
3575         return;
3576     }
3577     if (bdrv_op_is_blocked(target, BLOCK_OP_TYPE_MIRROR_TARGET, errp)) {
3578         return;
3579     }
3580 
3581     if (!bs->backing && sync == MIRROR_SYNC_MODE_TOP) {
3582         sync = MIRROR_SYNC_MODE_FULL;
3583     }
3584 
3585     /* pass the node name to replace to mirror start since it's loose coupling
3586      * and will allow to check whether the node still exist at mirror completion
3587      */
3588     mirror_start(job_id, bs, target,
3589                  has_replaces ? replaces : NULL,
3590                  speed, granularity, buf_size, sync, backing_mode,
3591                  on_source_error, on_target_error, unmap, filter_node_name,
3592                  errp);
3593 }
3594 
3595 void qmp_drive_mirror(DriveMirror *arg, Error **errp)
3596 {
3597     BlockDriverState *bs;
3598     BlockDriverState *source, *target_bs;
3599     AioContext *aio_context;
3600     BlockMirrorBackingMode backing_mode;
3601     Error *local_err = NULL;
3602     QDict *options = NULL;
3603     int flags;
3604     int64_t size;
3605     const char *format = arg->format;
3606 
3607     bs = qmp_get_root_bs(arg->device, errp);
3608     if (!bs) {
3609         return;
3610     }
3611 
3612     /* Early check to avoid creating target */
3613     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_MIRROR_SOURCE, errp)) {
3614         return;
3615     }
3616 
3617     aio_context = bdrv_get_aio_context(bs);
3618     aio_context_acquire(aio_context);
3619 
3620     if (!arg->has_mode) {
3621         arg->mode = NEW_IMAGE_MODE_ABSOLUTE_PATHS;
3622     }
3623 
3624     if (!arg->has_format) {
3625         format = (arg->mode == NEW_IMAGE_MODE_EXISTING
3626                   ? NULL : bs->drv->format_name);
3627     }
3628 
3629     flags = bs->open_flags | BDRV_O_RDWR;
3630     source = backing_bs(bs);
3631     if (!source && arg->sync == MIRROR_SYNC_MODE_TOP) {
3632         arg->sync = MIRROR_SYNC_MODE_FULL;
3633     }
3634     if (arg->sync == MIRROR_SYNC_MODE_NONE) {
3635         source = bs;
3636     }
3637 
3638     size = bdrv_getlength(bs);
3639     if (size < 0) {
3640         error_setg_errno(errp, -size, "bdrv_getlength failed");
3641         goto out;
3642     }
3643 
3644     if (arg->has_replaces) {
3645         BlockDriverState *to_replace_bs;
3646         AioContext *replace_aio_context;
3647         int64_t replace_size;
3648 
3649         if (!arg->has_node_name) {
3650             error_setg(errp, "a node-name must be provided when replacing a"
3651                              " named node of the graph");
3652             goto out;
3653         }
3654 
3655         to_replace_bs = check_to_replace_node(bs, arg->replaces, &local_err);
3656 
3657         if (!to_replace_bs) {
3658             error_propagate(errp, local_err);
3659             goto out;
3660         }
3661 
3662         replace_aio_context = bdrv_get_aio_context(to_replace_bs);
3663         aio_context_acquire(replace_aio_context);
3664         replace_size = bdrv_getlength(to_replace_bs);
3665         aio_context_release(replace_aio_context);
3666 
3667         if (size != replace_size) {
3668             error_setg(errp, "cannot replace image with a mirror image of "
3669                              "different size");
3670             goto out;
3671         }
3672     }
3673 
3674     if (arg->mode == NEW_IMAGE_MODE_ABSOLUTE_PATHS) {
3675         backing_mode = MIRROR_SOURCE_BACKING_CHAIN;
3676     } else {
3677         backing_mode = MIRROR_OPEN_BACKING_CHAIN;
3678     }
3679 
3680     /* Don't open backing image in create() */
3681     flags |= BDRV_O_NO_BACKING;
3682 
3683     if ((arg->sync == MIRROR_SYNC_MODE_FULL || !source)
3684         && arg->mode != NEW_IMAGE_MODE_EXISTING)
3685     {
3686         /* create new image w/o backing file */
3687         assert(format);
3688         bdrv_img_create(arg->target, format,
3689                         NULL, NULL, NULL, size, flags, false, &local_err);
3690     } else {
3691         switch (arg->mode) {
3692         case NEW_IMAGE_MODE_EXISTING:
3693             break;
3694         case NEW_IMAGE_MODE_ABSOLUTE_PATHS:
3695             /* create new image with backing file */
3696             bdrv_img_create(arg->target, format,
3697                             source->filename,
3698                             source->drv->format_name,
3699                             NULL, size, flags, false, &local_err);
3700             break;
3701         default:
3702             abort();
3703         }
3704     }
3705 
3706     if (local_err) {
3707         error_propagate(errp, local_err);
3708         goto out;
3709     }
3710 
3711     options = qdict_new();
3712     if (arg->has_node_name) {
3713         qdict_put_str(options, "node-name", arg->node_name);
3714     }
3715     if (format) {
3716         qdict_put_str(options, "driver", format);
3717     }
3718 
3719     /* Mirroring takes care of copy-on-write using the source's backing
3720      * file.
3721      */
3722     target_bs = bdrv_open(arg->target, NULL, options, flags, errp);
3723     if (!target_bs) {
3724         goto out;
3725     }
3726 
3727     bdrv_set_aio_context(target_bs, aio_context);
3728 
3729     blockdev_mirror_common(arg->has_job_id ? arg->job_id : NULL, bs, target_bs,
3730                            arg->has_replaces, arg->replaces, arg->sync,
3731                            backing_mode, arg->has_speed, arg->speed,
3732                            arg->has_granularity, arg->granularity,
3733                            arg->has_buf_size, arg->buf_size,
3734                            arg->has_on_source_error, arg->on_source_error,
3735                            arg->has_on_target_error, arg->on_target_error,
3736                            arg->has_unmap, arg->unmap,
3737                            false, NULL,
3738                            &local_err);
3739     bdrv_unref(target_bs);
3740     error_propagate(errp, local_err);
3741 out:
3742     aio_context_release(aio_context);
3743 }
3744 
3745 void qmp_blockdev_mirror(bool has_job_id, const char *job_id,
3746                          const char *device, const char *target,
3747                          bool has_replaces, const char *replaces,
3748                          MirrorSyncMode sync,
3749                          bool has_speed, int64_t speed,
3750                          bool has_granularity, uint32_t granularity,
3751                          bool has_buf_size, int64_t buf_size,
3752                          bool has_on_source_error,
3753                          BlockdevOnError on_source_error,
3754                          bool has_on_target_error,
3755                          BlockdevOnError on_target_error,
3756                          bool has_filter_node_name,
3757                          const char *filter_node_name,
3758                          Error **errp)
3759 {
3760     BlockDriverState *bs;
3761     BlockDriverState *target_bs;
3762     AioContext *aio_context;
3763     BlockMirrorBackingMode backing_mode = MIRROR_LEAVE_BACKING_CHAIN;
3764     Error *local_err = NULL;
3765 
3766     bs = qmp_get_root_bs(device, errp);
3767     if (!bs) {
3768         return;
3769     }
3770 
3771     target_bs = bdrv_lookup_bs(target, target, errp);
3772     if (!target_bs) {
3773         return;
3774     }
3775 
3776     aio_context = bdrv_get_aio_context(bs);
3777     aio_context_acquire(aio_context);
3778 
3779     bdrv_set_aio_context(target_bs, aio_context);
3780 
3781     blockdev_mirror_common(has_job_id ? job_id : NULL, bs, target_bs,
3782                            has_replaces, replaces, sync, backing_mode,
3783                            has_speed, speed,
3784                            has_granularity, granularity,
3785                            has_buf_size, buf_size,
3786                            has_on_source_error, on_source_error,
3787                            has_on_target_error, on_target_error,
3788                            true, true,
3789                            has_filter_node_name, filter_node_name,
3790                            &local_err);
3791     error_propagate(errp, local_err);
3792 
3793     aio_context_release(aio_context);
3794 }
3795 
3796 /* Get a block job using its ID and acquire its AioContext */
3797 static BlockJob *find_block_job(const char *id, AioContext **aio_context,
3798                                 Error **errp)
3799 {
3800     BlockJob *job;
3801 
3802     assert(id != NULL);
3803 
3804     *aio_context = NULL;
3805 
3806     job = block_job_get(id);
3807 
3808     if (!job) {
3809         error_set(errp, ERROR_CLASS_DEVICE_NOT_ACTIVE,
3810                   "Block job '%s' not found", id);
3811         return NULL;
3812     }
3813 
3814     *aio_context = blk_get_aio_context(job->blk);
3815     aio_context_acquire(*aio_context);
3816 
3817     return job;
3818 }
3819 
3820 void qmp_block_job_set_speed(const char *device, int64_t speed, Error **errp)
3821 {
3822     AioContext *aio_context;
3823     BlockJob *job = find_block_job(device, &aio_context, errp);
3824 
3825     if (!job) {
3826         return;
3827     }
3828 
3829     block_job_set_speed(job, speed, errp);
3830     aio_context_release(aio_context);
3831 }
3832 
3833 void qmp_block_job_cancel(const char *device,
3834                           bool has_force, bool force, Error **errp)
3835 {
3836     AioContext *aio_context;
3837     BlockJob *job = find_block_job(device, &aio_context, errp);
3838 
3839     if (!job) {
3840         return;
3841     }
3842 
3843     if (!has_force) {
3844         force = false;
3845     }
3846 
3847     if (block_job_user_paused(job) && !force) {
3848         error_setg(errp, "The block job for device '%s' is currently paused",
3849                    device);
3850         goto out;
3851     }
3852 
3853     trace_qmp_block_job_cancel(job);
3854     block_job_user_cancel(job, force, errp);
3855 out:
3856     aio_context_release(aio_context);
3857 }
3858 
3859 void qmp_block_job_pause(const char *device, Error **errp)
3860 {
3861     AioContext *aio_context;
3862     BlockJob *job = find_block_job(device, &aio_context, errp);
3863 
3864     if (!job) {
3865         return;
3866     }
3867 
3868     trace_qmp_block_job_pause(job);
3869     block_job_user_pause(job, errp);
3870     aio_context_release(aio_context);
3871 }
3872 
3873 void qmp_block_job_resume(const char *device, Error **errp)
3874 {
3875     AioContext *aio_context;
3876     BlockJob *job = find_block_job(device, &aio_context, errp);
3877 
3878     if (!job) {
3879         return;
3880     }
3881 
3882     trace_qmp_block_job_resume(job);
3883     block_job_user_resume(job, errp);
3884     aio_context_release(aio_context);
3885 }
3886 
3887 void qmp_block_job_complete(const char *device, Error **errp)
3888 {
3889     AioContext *aio_context;
3890     BlockJob *job = find_block_job(device, &aio_context, errp);
3891 
3892     if (!job) {
3893         return;
3894     }
3895 
3896     trace_qmp_block_job_complete(job);
3897     block_job_complete(job, errp);
3898     aio_context_release(aio_context);
3899 }
3900 
3901 void qmp_block_job_finalize(const char *id, Error **errp)
3902 {
3903     AioContext *aio_context;
3904     BlockJob *job = find_block_job(id, &aio_context, errp);
3905 
3906     if (!job) {
3907         return;
3908     }
3909 
3910     trace_qmp_block_job_finalize(job);
3911     block_job_finalize(job, errp);
3912     aio_context_release(aio_context);
3913 }
3914 
3915 void qmp_block_job_dismiss(const char *id, Error **errp)
3916 {
3917     AioContext *aio_context;
3918     BlockJob *job = find_block_job(id, &aio_context, errp);
3919 
3920     if (!job) {
3921         return;
3922     }
3923 
3924     trace_qmp_block_job_dismiss(job);
3925     block_job_dismiss(&job, errp);
3926     aio_context_release(aio_context);
3927 }
3928 
3929 void qmp_change_backing_file(const char *device,
3930                              const char *image_node_name,
3931                              const char *backing_file,
3932                              Error **errp)
3933 {
3934     BlockDriverState *bs = NULL;
3935     AioContext *aio_context;
3936     BlockDriverState *image_bs = NULL;
3937     Error *local_err = NULL;
3938     bool ro;
3939     int open_flags;
3940     int ret;
3941 
3942     bs = qmp_get_root_bs(device, errp);
3943     if (!bs) {
3944         return;
3945     }
3946 
3947     aio_context = bdrv_get_aio_context(bs);
3948     aio_context_acquire(aio_context);
3949 
3950     image_bs = bdrv_lookup_bs(NULL, image_node_name, &local_err);
3951     if (local_err) {
3952         error_propagate(errp, local_err);
3953         goto out;
3954     }
3955 
3956     if (!image_bs) {
3957         error_setg(errp, "image file not found");
3958         goto out;
3959     }
3960 
3961     if (bdrv_find_base(image_bs) == image_bs) {
3962         error_setg(errp, "not allowing backing file change on an image "
3963                          "without a backing file");
3964         goto out;
3965     }
3966 
3967     /* even though we are not necessarily operating on bs, we need it to
3968      * determine if block ops are currently prohibited on the chain */
3969     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_CHANGE, errp)) {
3970         goto out;
3971     }
3972 
3973     /* final sanity check */
3974     if (!bdrv_chain_contains(bs, image_bs)) {
3975         error_setg(errp, "'%s' and image file are not in the same chain",
3976                    device);
3977         goto out;
3978     }
3979 
3980     /* if not r/w, reopen to make r/w */
3981     open_flags = image_bs->open_flags;
3982     ro = bdrv_is_read_only(image_bs);
3983 
3984     if (ro) {
3985         bdrv_reopen(image_bs, open_flags | BDRV_O_RDWR, &local_err);
3986         if (local_err) {
3987             error_propagate(errp, local_err);
3988             goto out;
3989         }
3990     }
3991 
3992     ret = bdrv_change_backing_file(image_bs, backing_file,
3993                                image_bs->drv ? image_bs->drv->format_name : "");
3994 
3995     if (ret < 0) {
3996         error_setg_errno(errp, -ret, "Could not change backing file to '%s'",
3997                          backing_file);
3998         /* don't exit here, so we can try to restore open flags if
3999          * appropriate */
4000     }
4001 
4002     if (ro) {
4003         bdrv_reopen(image_bs, open_flags, &local_err);
4004         error_propagate(errp, local_err);
4005     }
4006 
4007 out:
4008     aio_context_release(aio_context);
4009 }
4010 
4011 void hmp_drive_add_node(Monitor *mon, const char *optstr)
4012 {
4013     QemuOpts *opts;
4014     QDict *qdict;
4015     Error *local_err = NULL;
4016 
4017     opts = qemu_opts_parse_noisily(&qemu_drive_opts, optstr, false);
4018     if (!opts) {
4019         return;
4020     }
4021 
4022     qdict = qemu_opts_to_qdict(opts, NULL);
4023 
4024     if (!qdict_get_try_str(qdict, "node-name")) {
4025         qobject_unref(qdict);
4026         error_report("'node-name' needs to be specified");
4027         goto out;
4028     }
4029 
4030     BlockDriverState *bs = bds_tree_init(qdict, &local_err);
4031     if (!bs) {
4032         error_report_err(local_err);
4033         goto out;
4034     }
4035 
4036     QTAILQ_INSERT_TAIL(&monitor_bdrv_states, bs, monitor_list);
4037 
4038 out:
4039     qemu_opts_del(opts);
4040 }
4041 
4042 void qmp_blockdev_add(BlockdevOptions *options, Error **errp)
4043 {
4044     BlockDriverState *bs;
4045     QObject *obj;
4046     Visitor *v = qobject_output_visitor_new(&obj);
4047     QDict *qdict;
4048     Error *local_err = NULL;
4049 
4050     visit_type_BlockdevOptions(v, NULL, &options, &local_err);
4051     if (local_err) {
4052         error_propagate(errp, local_err);
4053         goto fail;
4054     }
4055 
4056     visit_complete(v, &obj);
4057     qdict = qobject_to(QDict, obj);
4058 
4059     qdict_flatten(qdict);
4060 
4061     if (!qdict_get_try_str(qdict, "node-name")) {
4062         error_setg(errp, "'node-name' must be specified for the root node");
4063         goto fail;
4064     }
4065 
4066     bs = bds_tree_init(qdict, errp);
4067     if (!bs) {
4068         goto fail;
4069     }
4070 
4071     QTAILQ_INSERT_TAIL(&monitor_bdrv_states, bs, monitor_list);
4072 
4073 fail:
4074     visit_free(v);
4075 }
4076 
4077 void qmp_blockdev_del(const char *node_name, Error **errp)
4078 {
4079     AioContext *aio_context;
4080     BlockDriverState *bs;
4081 
4082     bs = bdrv_find_node(node_name);
4083     if (!bs) {
4084         error_setg(errp, "Cannot find node %s", node_name);
4085         return;
4086     }
4087     if (bdrv_has_blk(bs)) {
4088         error_setg(errp, "Node %s is in use", node_name);
4089         return;
4090     }
4091     aio_context = bdrv_get_aio_context(bs);
4092     aio_context_acquire(aio_context);
4093 
4094     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_DRIVE_DEL, errp)) {
4095         goto out;
4096     }
4097 
4098     if (!bs->monitor_list.tqe_prev) {
4099         error_setg(errp, "Node %s is not owned by the monitor",
4100                    bs->node_name);
4101         goto out;
4102     }
4103 
4104     if (bs->refcnt > 1) {
4105         error_setg(errp, "Block device %s is in use",
4106                    bdrv_get_device_or_node_name(bs));
4107         goto out;
4108     }
4109 
4110     QTAILQ_REMOVE(&monitor_bdrv_states, bs, monitor_list);
4111     bdrv_unref(bs);
4112 
4113 out:
4114     aio_context_release(aio_context);
4115 }
4116 
4117 static BdrvChild *bdrv_find_child(BlockDriverState *parent_bs,
4118                                   const char *child_name)
4119 {
4120     BdrvChild *child;
4121 
4122     QLIST_FOREACH(child, &parent_bs->children, next) {
4123         if (strcmp(child->name, child_name) == 0) {
4124             return child;
4125         }
4126     }
4127 
4128     return NULL;
4129 }
4130 
4131 void qmp_x_blockdev_change(const char *parent, bool has_child,
4132                            const char *child, bool has_node,
4133                            const char *node, Error **errp)
4134 {
4135     BlockDriverState *parent_bs, *new_bs = NULL;
4136     BdrvChild *p_child;
4137 
4138     parent_bs = bdrv_lookup_bs(parent, parent, errp);
4139     if (!parent_bs) {
4140         return;
4141     }
4142 
4143     if (has_child == has_node) {
4144         if (has_child) {
4145             error_setg(errp, "The parameters child and node are in conflict");
4146         } else {
4147             error_setg(errp, "Either child or node must be specified");
4148         }
4149         return;
4150     }
4151 
4152     if (has_child) {
4153         p_child = bdrv_find_child(parent_bs, child);
4154         if (!p_child) {
4155             error_setg(errp, "Node '%s' does not have child '%s'",
4156                        parent, child);
4157             return;
4158         }
4159         bdrv_del_child(parent_bs, p_child, errp);
4160     }
4161 
4162     if (has_node) {
4163         new_bs = bdrv_find_node(node);
4164         if (!new_bs) {
4165             error_setg(errp, "Node '%s' not found", node);
4166             return;
4167         }
4168         bdrv_add_child(parent_bs, new_bs, errp);
4169     }
4170 }
4171 
4172 BlockJobInfoList *qmp_query_block_jobs(Error **errp)
4173 {
4174     BlockJobInfoList *head = NULL, **p_next = &head;
4175     BlockJob *job;
4176 
4177     for (job = block_job_next(NULL); job; job = block_job_next(job)) {
4178         BlockJobInfoList *elem;
4179         AioContext *aio_context;
4180 
4181         if (block_job_is_internal(job)) {
4182             continue;
4183         }
4184         elem = g_new0(BlockJobInfoList, 1);
4185         aio_context = blk_get_aio_context(job->blk);
4186         aio_context_acquire(aio_context);
4187         elem->value = block_job_query(job, errp);
4188         aio_context_release(aio_context);
4189         if (!elem->value) {
4190             g_free(elem);
4191             qapi_free_BlockJobInfoList(head);
4192             return NULL;
4193         }
4194         *p_next = elem;
4195         p_next = &elem->next;
4196     }
4197 
4198     return head;
4199 }
4200 
4201 void qmp_x_blockdev_set_iothread(const char *node_name, StrOrNull *iothread,
4202                                  bool has_force, bool force, Error **errp)
4203 {
4204     AioContext *old_context;
4205     AioContext *new_context;
4206     BlockDriverState *bs;
4207 
4208     bs = bdrv_find_node(node_name);
4209     if (!bs) {
4210         error_setg(errp, "Cannot find node %s", node_name);
4211         return;
4212     }
4213 
4214     /* Protects against accidents. */
4215     if (!(has_force && force) && bdrv_has_blk(bs)) {
4216         error_setg(errp, "Node %s is associated with a BlockBackend and could "
4217                          "be in use (use force=true to override this check)",
4218                          node_name);
4219         return;
4220     }
4221 
4222     if (iothread->type == QTYPE_QSTRING) {
4223         IOThread *obj = iothread_by_id(iothread->u.s);
4224         if (!obj) {
4225             error_setg(errp, "Cannot find iothread %s", iothread->u.s);
4226             return;
4227         }
4228 
4229         new_context = iothread_get_aio_context(obj);
4230     } else {
4231         new_context = qemu_get_aio_context();
4232     }
4233 
4234     old_context = bdrv_get_aio_context(bs);
4235     aio_context_acquire(old_context);
4236 
4237     bdrv_set_aio_context(bs, new_context);
4238 
4239     aio_context_release(old_context);
4240 }
4241 
4242 void qmp_x_block_latency_histogram_set(
4243     const char *device,
4244     bool has_boundaries, uint64List *boundaries,
4245     bool has_boundaries_read, uint64List *boundaries_read,
4246     bool has_boundaries_write, uint64List *boundaries_write,
4247     bool has_boundaries_flush, uint64List *boundaries_flush,
4248     Error **errp)
4249 {
4250     BlockBackend *blk = blk_by_name(device);
4251     BlockAcctStats *stats;
4252 
4253     if (!blk) {
4254         error_setg(errp, "Device '%s' not found", device);
4255         return;
4256     }
4257     stats = blk_get_stats(blk);
4258 
4259     if (!has_boundaries && !has_boundaries_read && !has_boundaries_write &&
4260         !has_boundaries_flush)
4261     {
4262         block_latency_histograms_clear(stats);
4263         return;
4264     }
4265 
4266     if (has_boundaries || has_boundaries_read) {
4267         block_latency_histogram_set(
4268             stats, BLOCK_ACCT_READ,
4269             has_boundaries_read ? boundaries_read : boundaries);
4270     }
4271 
4272     if (has_boundaries || has_boundaries_write) {
4273         block_latency_histogram_set(
4274             stats, BLOCK_ACCT_WRITE,
4275             has_boundaries_write ? boundaries_write : boundaries);
4276     }
4277 
4278     if (has_boundaries || has_boundaries_flush) {
4279         block_latency_histogram_set(
4280             stats, BLOCK_ACCT_FLUSH,
4281             has_boundaries_flush ? boundaries_flush : boundaries);
4282     }
4283 }
4284 
4285 QemuOptsList qemu_common_drive_opts = {
4286     .name = "drive",
4287     .head = QTAILQ_HEAD_INITIALIZER(qemu_common_drive_opts.head),
4288     .desc = {
4289         {
4290             .name = "snapshot",
4291             .type = QEMU_OPT_BOOL,
4292             .help = "enable/disable snapshot mode",
4293         },{
4294             .name = "aio",
4295             .type = QEMU_OPT_STRING,
4296             .help = "host AIO implementation (threads, native)",
4297         },{
4298             .name = BDRV_OPT_CACHE_WB,
4299             .type = QEMU_OPT_BOOL,
4300             .help = "Enable writeback mode",
4301         },{
4302             .name = "format",
4303             .type = QEMU_OPT_STRING,
4304             .help = "disk format (raw, qcow2, ...)",
4305         },{
4306             .name = "rerror",
4307             .type = QEMU_OPT_STRING,
4308             .help = "read error action",
4309         },{
4310             .name = "werror",
4311             .type = QEMU_OPT_STRING,
4312             .help = "write error action",
4313         },{
4314             .name = BDRV_OPT_READ_ONLY,
4315             .type = QEMU_OPT_BOOL,
4316             .help = "open drive file as read-only",
4317         },
4318 
4319         THROTTLE_OPTS,
4320 
4321         {
4322             .name = "throttling.group",
4323             .type = QEMU_OPT_STRING,
4324             .help = "name of the block throttling group",
4325         },{
4326             .name = "copy-on-read",
4327             .type = QEMU_OPT_BOOL,
4328             .help = "copy read data from backing file into image file",
4329         },{
4330             .name = "detect-zeroes",
4331             .type = QEMU_OPT_STRING,
4332             .help = "try to optimize zero writes (off, on, unmap)",
4333         },{
4334             .name = "stats-account-invalid",
4335             .type = QEMU_OPT_BOOL,
4336             .help = "whether to account for invalid I/O operations "
4337                     "in the statistics",
4338         },{
4339             .name = "stats-account-failed",
4340             .type = QEMU_OPT_BOOL,
4341             .help = "whether to account for failed I/O operations "
4342                     "in the statistics",
4343         },
4344         { /* end of list */ }
4345     },
4346 };
4347 
4348 QemuOptsList qemu_drive_opts = {
4349     .name = "drive",
4350     .head = QTAILQ_HEAD_INITIALIZER(qemu_drive_opts.head),
4351     .desc = {
4352         /*
4353          * no elements => accept any params
4354          * validation will happen later
4355          */
4356         { /* end of list */ }
4357     },
4358 };
4359