xref: /openbmc/qemu/blockdev.c (revision 679cb8e1)
1 /*
2  * QEMU host block devices
3  *
4  * Copyright (c) 2003-2008 Fabrice Bellard
5  *
6  * This work is licensed under the terms of the GNU GPL, version 2 or
7  * later.  See the COPYING file in the top-level directory.
8  *
9  * This file incorporates work covered by the following copyright and
10  * permission notice:
11  *
12  * Copyright (c) 2003-2008 Fabrice Bellard
13  *
14  * Permission is hereby granted, free of charge, to any person obtaining a copy
15  * of this software and associated documentation files (the "Software"), to deal
16  * in the Software without restriction, including without limitation the rights
17  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
18  * copies of the Software, and to permit persons to whom the Software is
19  * furnished to do so, subject to the following conditions:
20  *
21  * The above copyright notice and this permission notice shall be included in
22  * all copies or substantial portions of the Software.
23  *
24  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
25  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
26  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
27  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
28  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
29  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
30  * THE SOFTWARE.
31  */
32 
33 #include "qemu/osdep.h"
34 #include "sysemu/block-backend.h"
35 #include "sysemu/blockdev.h"
36 #include "hw/block/block.h"
37 #include "block/blockjob.h"
38 #include "block/qdict.h"
39 #include "block/throttle-groups.h"
40 #include "monitor/monitor.h"
41 #include "qemu/error-report.h"
42 #include "qemu/option.h"
43 #include "qemu/config-file.h"
44 #include "qapi/qapi-commands-block.h"
45 #include "qapi/qapi-commands-transaction.h"
46 #include "qapi/qapi-visit-block-core.h"
47 #include "qapi/qmp/qdict.h"
48 #include "qapi/qmp/qnum.h"
49 #include "qapi/qmp/qstring.h"
50 #include "qapi/error.h"
51 #include "qapi/qmp/qerror.h"
52 #include "qapi/qmp/qlist.h"
53 #include "qapi/qobject-output-visitor.h"
54 #include "sysemu/sysemu.h"
55 #include "sysemu/iothread.h"
56 #include "block/block_int.h"
57 #include "block/trace.h"
58 #include "sysemu/arch_init.h"
59 #include "sysemu/qtest.h"
60 #include "qemu/cutils.h"
61 #include "qemu/help_option.h"
62 #include "qemu/throttle-options.h"
63 
64 static QTAILQ_HEAD(, BlockDriverState) monitor_bdrv_states =
65     QTAILQ_HEAD_INITIALIZER(monitor_bdrv_states);
66 
67 static int do_open_tray(const char *blk_name, const char *qdev_id,
68                         bool force, Error **errp);
69 static void blockdev_remove_medium(bool has_device, const char *device,
70                                    bool has_id, const char *id, Error **errp);
71 static void blockdev_insert_medium(bool has_device, const char *device,
72                                    bool has_id, const char *id,
73                                    const char *node_name, Error **errp);
74 
75 static const char *const if_name[IF_COUNT] = {
76     [IF_NONE] = "none",
77     [IF_IDE] = "ide",
78     [IF_SCSI] = "scsi",
79     [IF_FLOPPY] = "floppy",
80     [IF_PFLASH] = "pflash",
81     [IF_MTD] = "mtd",
82     [IF_SD] = "sd",
83     [IF_VIRTIO] = "virtio",
84     [IF_XEN] = "xen",
85 };
86 
87 static int if_max_devs[IF_COUNT] = {
88     /*
89      * Do not change these numbers!  They govern how drive option
90      * index maps to unit and bus.  That mapping is ABI.
91      *
92      * All controllers used to implement if=T drives need to support
93      * if_max_devs[T] units, for any T with if_max_devs[T] != 0.
94      * Otherwise, some index values map to "impossible" bus, unit
95      * values.
96      *
97      * For instance, if you change [IF_SCSI] to 255, -drive
98      * if=scsi,index=12 no longer means bus=1,unit=5, but
99      * bus=0,unit=12.  With an lsi53c895a controller (7 units max),
100      * the drive can't be set up.  Regression.
101      */
102     [IF_IDE] = 2,
103     [IF_SCSI] = 7,
104 };
105 
106 /**
107  * Boards may call this to offer board-by-board overrides
108  * of the default, global values.
109  */
110 void override_max_devs(BlockInterfaceType type, int max_devs)
111 {
112     BlockBackend *blk;
113     DriveInfo *dinfo;
114 
115     if (max_devs <= 0) {
116         return;
117     }
118 
119     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
120         dinfo = blk_legacy_dinfo(blk);
121         if (dinfo->type == type) {
122             fprintf(stderr, "Cannot override units-per-bus property of"
123                     " the %s interface, because a drive of that type has"
124                     " already been added.\n", if_name[type]);
125             g_assert_not_reached();
126         }
127     }
128 
129     if_max_devs[type] = max_devs;
130 }
131 
132 /*
133  * We automatically delete the drive when a device using it gets
134  * unplugged.  Questionable feature, but we can't just drop it.
135  * Device models call blockdev_mark_auto_del() to schedule the
136  * automatic deletion, and generic qdev code calls blockdev_auto_del()
137  * when deletion is actually safe.
138  */
139 void blockdev_mark_auto_del(BlockBackend *blk)
140 {
141     DriveInfo *dinfo = blk_legacy_dinfo(blk);
142     BlockDriverState *bs = blk_bs(blk);
143     AioContext *aio_context;
144 
145     if (!dinfo) {
146         return;
147     }
148 
149     if (bs) {
150         aio_context = bdrv_get_aio_context(bs);
151         aio_context_acquire(aio_context);
152 
153         if (bs->job) {
154             job_cancel(&bs->job->job, false);
155         }
156 
157         aio_context_release(aio_context);
158     }
159 
160     dinfo->auto_del = 1;
161 }
162 
163 void blockdev_auto_del(BlockBackend *blk)
164 {
165     DriveInfo *dinfo = blk_legacy_dinfo(blk);
166 
167     if (dinfo && dinfo->auto_del) {
168         monitor_remove_blk(blk);
169         blk_unref(blk);
170     }
171 }
172 
173 /**
174  * Returns the current mapping of how many units per bus
175  * a particular interface can support.
176  *
177  *  A positive integer indicates n units per bus.
178  *  0 implies the mapping has not been established.
179  * -1 indicates an invalid BlockInterfaceType was given.
180  */
181 int drive_get_max_devs(BlockInterfaceType type)
182 {
183     if (type >= IF_IDE && type < IF_COUNT) {
184         return if_max_devs[type];
185     }
186 
187     return -1;
188 }
189 
190 static int drive_index_to_bus_id(BlockInterfaceType type, int index)
191 {
192     int max_devs = if_max_devs[type];
193     return max_devs ? index / max_devs : 0;
194 }
195 
196 static int drive_index_to_unit_id(BlockInterfaceType type, int index)
197 {
198     int max_devs = if_max_devs[type];
199     return max_devs ? index % max_devs : index;
200 }
201 
202 QemuOpts *drive_def(const char *optstr)
203 {
204     return qemu_opts_parse_noisily(qemu_find_opts("drive"), optstr, false);
205 }
206 
207 QemuOpts *drive_add(BlockInterfaceType type, int index, const char *file,
208                     const char *optstr)
209 {
210     QemuOpts *opts;
211 
212     opts = drive_def(optstr);
213     if (!opts) {
214         return NULL;
215     }
216     if (type != IF_DEFAULT) {
217         qemu_opt_set(opts, "if", if_name[type], &error_abort);
218     }
219     if (index >= 0) {
220         qemu_opt_set_number(opts, "index", index, &error_abort);
221     }
222     if (file)
223         qemu_opt_set(opts, "file", file, &error_abort);
224     return opts;
225 }
226 
227 DriveInfo *drive_get(BlockInterfaceType type, int bus, int unit)
228 {
229     BlockBackend *blk;
230     DriveInfo *dinfo;
231 
232     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
233         dinfo = blk_legacy_dinfo(blk);
234         if (dinfo && dinfo->type == type
235             && dinfo->bus == bus && dinfo->unit == unit) {
236             return dinfo;
237         }
238     }
239 
240     return NULL;
241 }
242 
243 void drive_check_orphaned(void)
244 {
245     BlockBackend *blk;
246     DriveInfo *dinfo;
247     Location loc;
248     bool orphans = false;
249 
250     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
251         dinfo = blk_legacy_dinfo(blk);
252         if (!blk_get_attached_dev(blk) && !dinfo->is_default &&
253             dinfo->type != IF_NONE) {
254             loc_push_none(&loc);
255             qemu_opts_loc_restore(dinfo->opts);
256             error_report("machine type does not support"
257                          " if=%s,bus=%d,unit=%d",
258                          if_name[dinfo->type], dinfo->bus, dinfo->unit);
259             loc_pop(&loc);
260             orphans = true;
261         }
262     }
263 
264     if (orphans) {
265         exit(1);
266     }
267 }
268 
269 DriveInfo *drive_get_by_index(BlockInterfaceType type, int index)
270 {
271     return drive_get(type,
272                      drive_index_to_bus_id(type, index),
273                      drive_index_to_unit_id(type, index));
274 }
275 
276 int drive_get_max_bus(BlockInterfaceType type)
277 {
278     int max_bus;
279     BlockBackend *blk;
280     DriveInfo *dinfo;
281 
282     max_bus = -1;
283     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
284         dinfo = blk_legacy_dinfo(blk);
285         if (dinfo && dinfo->type == type && dinfo->bus > max_bus) {
286             max_bus = dinfo->bus;
287         }
288     }
289     return max_bus;
290 }
291 
292 /* Get a block device.  This should only be used for single-drive devices
293    (e.g. SD/Floppy/MTD).  Multi-disk devices (scsi/ide) should use the
294    appropriate bus.  */
295 DriveInfo *drive_get_next(BlockInterfaceType type)
296 {
297     static int next_block_unit[IF_COUNT];
298 
299     return drive_get(type, 0, next_block_unit[type]++);
300 }
301 
302 static void bdrv_format_print(void *opaque, const char *name)
303 {
304     error_printf(" %s", name);
305 }
306 
307 typedef struct {
308     QEMUBH *bh;
309     BlockDriverState *bs;
310 } BDRVPutRefBH;
311 
312 static int parse_block_error_action(const char *buf, bool is_read, Error **errp)
313 {
314     if (!strcmp(buf, "ignore")) {
315         return BLOCKDEV_ON_ERROR_IGNORE;
316     } else if (!is_read && !strcmp(buf, "enospc")) {
317         return BLOCKDEV_ON_ERROR_ENOSPC;
318     } else if (!strcmp(buf, "stop")) {
319         return BLOCKDEV_ON_ERROR_STOP;
320     } else if (!strcmp(buf, "report")) {
321         return BLOCKDEV_ON_ERROR_REPORT;
322     } else {
323         error_setg(errp, "'%s' invalid %s error action",
324                    buf, is_read ? "read" : "write");
325         return -1;
326     }
327 }
328 
329 static bool parse_stats_intervals(BlockAcctStats *stats, QList *intervals,
330                                   Error **errp)
331 {
332     const QListEntry *entry;
333     for (entry = qlist_first(intervals); entry; entry = qlist_next(entry)) {
334         switch (qobject_type(entry->value)) {
335 
336         case QTYPE_QSTRING: {
337             unsigned long long length;
338             const char *str = qstring_get_str(qobject_to(QString,
339                                                          entry->value));
340             if (parse_uint_full(str, &length, 10) == 0 &&
341                 length > 0 && length <= UINT_MAX) {
342                 block_acct_add_interval(stats, (unsigned) length);
343             } else {
344                 error_setg(errp, "Invalid interval length: %s", str);
345                 return false;
346             }
347             break;
348         }
349 
350         case QTYPE_QNUM: {
351             int64_t length = qnum_get_int(qobject_to(QNum, entry->value));
352 
353             if (length > 0 && length <= UINT_MAX) {
354                 block_acct_add_interval(stats, (unsigned) length);
355             } else {
356                 error_setg(errp, "Invalid interval length: %" PRId64, length);
357                 return false;
358             }
359             break;
360         }
361 
362         default:
363             error_setg(errp, "The specification of stats-intervals is invalid");
364             return false;
365         }
366     }
367     return true;
368 }
369 
370 typedef enum { MEDIA_DISK, MEDIA_CDROM } DriveMediaType;
371 
372 /* All parameters but @opts are optional and may be set to NULL. */
373 static void extract_common_blockdev_options(QemuOpts *opts, int *bdrv_flags,
374     const char **throttling_group, ThrottleConfig *throttle_cfg,
375     BlockdevDetectZeroesOptions *detect_zeroes, Error **errp)
376 {
377     Error *local_error = NULL;
378     const char *aio;
379 
380     if (bdrv_flags) {
381         if (qemu_opt_get_bool(opts, "copy-on-read", false)) {
382             *bdrv_flags |= BDRV_O_COPY_ON_READ;
383         }
384 
385         if ((aio = qemu_opt_get(opts, "aio")) != NULL) {
386             if (!strcmp(aio, "native")) {
387                 *bdrv_flags |= BDRV_O_NATIVE_AIO;
388             } else if (!strcmp(aio, "threads")) {
389                 /* this is the default */
390             } else {
391                error_setg(errp, "invalid aio option");
392                return;
393             }
394         }
395     }
396 
397     /* disk I/O throttling */
398     if (throttling_group) {
399         *throttling_group = qemu_opt_get(opts, "throttling.group");
400     }
401 
402     if (throttle_cfg) {
403         throttle_config_init(throttle_cfg);
404         throttle_cfg->buckets[THROTTLE_BPS_TOTAL].avg =
405             qemu_opt_get_number(opts, "throttling.bps-total", 0);
406         throttle_cfg->buckets[THROTTLE_BPS_READ].avg  =
407             qemu_opt_get_number(opts, "throttling.bps-read", 0);
408         throttle_cfg->buckets[THROTTLE_BPS_WRITE].avg =
409             qemu_opt_get_number(opts, "throttling.bps-write", 0);
410         throttle_cfg->buckets[THROTTLE_OPS_TOTAL].avg =
411             qemu_opt_get_number(opts, "throttling.iops-total", 0);
412         throttle_cfg->buckets[THROTTLE_OPS_READ].avg =
413             qemu_opt_get_number(opts, "throttling.iops-read", 0);
414         throttle_cfg->buckets[THROTTLE_OPS_WRITE].avg =
415             qemu_opt_get_number(opts, "throttling.iops-write", 0);
416 
417         throttle_cfg->buckets[THROTTLE_BPS_TOTAL].max =
418             qemu_opt_get_number(opts, "throttling.bps-total-max", 0);
419         throttle_cfg->buckets[THROTTLE_BPS_READ].max  =
420             qemu_opt_get_number(opts, "throttling.bps-read-max", 0);
421         throttle_cfg->buckets[THROTTLE_BPS_WRITE].max =
422             qemu_opt_get_number(opts, "throttling.bps-write-max", 0);
423         throttle_cfg->buckets[THROTTLE_OPS_TOTAL].max =
424             qemu_opt_get_number(opts, "throttling.iops-total-max", 0);
425         throttle_cfg->buckets[THROTTLE_OPS_READ].max =
426             qemu_opt_get_number(opts, "throttling.iops-read-max", 0);
427         throttle_cfg->buckets[THROTTLE_OPS_WRITE].max =
428             qemu_opt_get_number(opts, "throttling.iops-write-max", 0);
429 
430         throttle_cfg->buckets[THROTTLE_BPS_TOTAL].burst_length =
431             qemu_opt_get_number(opts, "throttling.bps-total-max-length", 1);
432         throttle_cfg->buckets[THROTTLE_BPS_READ].burst_length  =
433             qemu_opt_get_number(opts, "throttling.bps-read-max-length", 1);
434         throttle_cfg->buckets[THROTTLE_BPS_WRITE].burst_length =
435             qemu_opt_get_number(opts, "throttling.bps-write-max-length", 1);
436         throttle_cfg->buckets[THROTTLE_OPS_TOTAL].burst_length =
437             qemu_opt_get_number(opts, "throttling.iops-total-max-length", 1);
438         throttle_cfg->buckets[THROTTLE_OPS_READ].burst_length =
439             qemu_opt_get_number(opts, "throttling.iops-read-max-length", 1);
440         throttle_cfg->buckets[THROTTLE_OPS_WRITE].burst_length =
441             qemu_opt_get_number(opts, "throttling.iops-write-max-length", 1);
442 
443         throttle_cfg->op_size =
444             qemu_opt_get_number(opts, "throttling.iops-size", 0);
445 
446         if (!throttle_is_valid(throttle_cfg, errp)) {
447             return;
448         }
449     }
450 
451     if (detect_zeroes) {
452         *detect_zeroes =
453             qapi_enum_parse(&BlockdevDetectZeroesOptions_lookup,
454                             qemu_opt_get(opts, "detect-zeroes"),
455                             BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF,
456                             &local_error);
457         if (local_error) {
458             error_propagate(errp, local_error);
459             return;
460         }
461     }
462 }
463 
464 /* Takes the ownership of bs_opts */
465 static BlockBackend *blockdev_init(const char *file, QDict *bs_opts,
466                                    Error **errp)
467 {
468     const char *buf;
469     int bdrv_flags = 0;
470     int on_read_error, on_write_error;
471     bool account_invalid, account_failed;
472     bool writethrough, read_only;
473     BlockBackend *blk;
474     BlockDriverState *bs;
475     ThrottleConfig cfg;
476     int snapshot = 0;
477     Error *error = NULL;
478     QemuOpts *opts;
479     QDict *interval_dict = NULL;
480     QList *interval_list = NULL;
481     const char *id;
482     BlockdevDetectZeroesOptions detect_zeroes =
483         BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF;
484     const char *throttling_group = NULL;
485 
486     /* Check common options by copying from bs_opts to opts, all other options
487      * stay in bs_opts for processing by bdrv_open(). */
488     id = qdict_get_try_str(bs_opts, "id");
489     opts = qemu_opts_create(&qemu_common_drive_opts, id, 1, &error);
490     if (error) {
491         error_propagate(errp, error);
492         goto err_no_opts;
493     }
494 
495     qemu_opts_absorb_qdict(opts, bs_opts, &error);
496     if (error) {
497         error_propagate(errp, error);
498         goto early_err;
499     }
500 
501     if (id) {
502         qdict_del(bs_opts, "id");
503     }
504 
505     /* extract parameters */
506     snapshot = qemu_opt_get_bool(opts, "snapshot", 0);
507 
508     account_invalid = qemu_opt_get_bool(opts, "stats-account-invalid", true);
509     account_failed = qemu_opt_get_bool(opts, "stats-account-failed", true);
510 
511     writethrough = !qemu_opt_get_bool(opts, BDRV_OPT_CACHE_WB, true);
512 
513     id = qemu_opts_id(opts);
514 
515     qdict_extract_subqdict(bs_opts, &interval_dict, "stats-intervals.");
516     qdict_array_split(interval_dict, &interval_list);
517 
518     if (qdict_size(interval_dict) != 0) {
519         error_setg(errp, "Invalid option stats-intervals.%s",
520                    qdict_first(interval_dict)->key);
521         goto early_err;
522     }
523 
524     extract_common_blockdev_options(opts, &bdrv_flags, &throttling_group, &cfg,
525                                     &detect_zeroes, &error);
526     if (error) {
527         error_propagate(errp, error);
528         goto early_err;
529     }
530 
531     if ((buf = qemu_opt_get(opts, "format")) != NULL) {
532         if (is_help_option(buf)) {
533             error_printf("Supported formats:");
534             bdrv_iterate_format(bdrv_format_print, NULL, false);
535             error_printf("\nSupported formats (read-only):");
536             bdrv_iterate_format(bdrv_format_print, NULL, true);
537             error_printf("\n");
538             goto early_err;
539         }
540 
541         if (qdict_haskey(bs_opts, "driver")) {
542             error_setg(errp, "Cannot specify both 'driver' and 'format'");
543             goto early_err;
544         }
545         qdict_put_str(bs_opts, "driver", buf);
546     }
547 
548     on_write_error = BLOCKDEV_ON_ERROR_ENOSPC;
549     if ((buf = qemu_opt_get(opts, "werror")) != NULL) {
550         on_write_error = parse_block_error_action(buf, 0, &error);
551         if (error) {
552             error_propagate(errp, error);
553             goto early_err;
554         }
555     }
556 
557     on_read_error = BLOCKDEV_ON_ERROR_REPORT;
558     if ((buf = qemu_opt_get(opts, "rerror")) != NULL) {
559         on_read_error = parse_block_error_action(buf, 1, &error);
560         if (error) {
561             error_propagate(errp, error);
562             goto early_err;
563         }
564     }
565 
566     if (snapshot) {
567         bdrv_flags |= BDRV_O_SNAPSHOT;
568     }
569 
570     read_only = qemu_opt_get_bool(opts, BDRV_OPT_READ_ONLY, false);
571 
572     /* init */
573     if ((!file || !*file) && !qdict_size(bs_opts)) {
574         BlockBackendRootState *blk_rs;
575 
576         blk = blk_new(0, BLK_PERM_ALL);
577         blk_rs = blk_get_root_state(blk);
578         blk_rs->open_flags    = bdrv_flags;
579         blk_rs->read_only     = read_only;
580         blk_rs->detect_zeroes = detect_zeroes;
581 
582         qobject_unref(bs_opts);
583     } else {
584         if (file && !*file) {
585             file = NULL;
586         }
587 
588         /* bdrv_open() defaults to the values in bdrv_flags (for compatibility
589          * with other callers) rather than what we want as the real defaults.
590          * Apply the defaults here instead. */
591         qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_DIRECT, "off");
592         qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_NO_FLUSH, "off");
593         qdict_set_default_str(bs_opts, BDRV_OPT_READ_ONLY,
594                               read_only ? "on" : "off");
595         qdict_set_default_str(bs_opts, BDRV_OPT_AUTO_READ_ONLY, "on");
596         assert((bdrv_flags & BDRV_O_CACHE_MASK) == 0);
597 
598         if (runstate_check(RUN_STATE_INMIGRATE)) {
599             bdrv_flags |= BDRV_O_INACTIVE;
600         }
601 
602         blk = blk_new_open(file, NULL, bs_opts, bdrv_flags, errp);
603         if (!blk) {
604             goto err_no_bs_opts;
605         }
606         bs = blk_bs(blk);
607 
608         bs->detect_zeroes = detect_zeroes;
609 
610         block_acct_setup(blk_get_stats(blk), account_invalid, account_failed);
611 
612         if (!parse_stats_intervals(blk_get_stats(blk), interval_list, errp)) {
613             blk_unref(blk);
614             blk = NULL;
615             goto err_no_bs_opts;
616         }
617     }
618 
619     /* disk I/O throttling */
620     if (throttle_enabled(&cfg)) {
621         if (!throttling_group) {
622             throttling_group = id;
623         }
624         blk_io_limits_enable(blk, throttling_group);
625         blk_set_io_limits(blk, &cfg);
626     }
627 
628     blk_set_enable_write_cache(blk, !writethrough);
629     blk_set_on_error(blk, on_read_error, on_write_error);
630 
631     if (!monitor_add_blk(blk, id, errp)) {
632         blk_unref(blk);
633         blk = NULL;
634         goto err_no_bs_opts;
635     }
636 
637 err_no_bs_opts:
638     qemu_opts_del(opts);
639     qobject_unref(interval_dict);
640     qobject_unref(interval_list);
641     return blk;
642 
643 early_err:
644     qemu_opts_del(opts);
645     qobject_unref(interval_dict);
646     qobject_unref(interval_list);
647 err_no_opts:
648     qobject_unref(bs_opts);
649     return NULL;
650 }
651 
652 /* Takes the ownership of bs_opts */
653 static BlockDriverState *bds_tree_init(QDict *bs_opts, Error **errp)
654 {
655     int bdrv_flags = 0;
656 
657     /* bdrv_open() defaults to the values in bdrv_flags (for compatibility
658      * with other callers) rather than what we want as the real defaults.
659      * Apply the defaults here instead. */
660     qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_DIRECT, "off");
661     qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_NO_FLUSH, "off");
662     qdict_set_default_str(bs_opts, BDRV_OPT_READ_ONLY, "off");
663 
664     if (runstate_check(RUN_STATE_INMIGRATE)) {
665         bdrv_flags |= BDRV_O_INACTIVE;
666     }
667 
668     return bdrv_open(NULL, NULL, bs_opts, bdrv_flags, errp);
669 }
670 
671 void blockdev_close_all_bdrv_states(void)
672 {
673     BlockDriverState *bs, *next_bs;
674 
675     QTAILQ_FOREACH_SAFE(bs, &monitor_bdrv_states, monitor_list, next_bs) {
676         AioContext *ctx = bdrv_get_aio_context(bs);
677 
678         aio_context_acquire(ctx);
679         bdrv_unref(bs);
680         aio_context_release(ctx);
681     }
682 }
683 
684 /* Iterates over the list of monitor-owned BlockDriverStates */
685 BlockDriverState *bdrv_next_monitor_owned(BlockDriverState *bs)
686 {
687     return bs ? QTAILQ_NEXT(bs, monitor_list)
688               : QTAILQ_FIRST(&monitor_bdrv_states);
689 }
690 
691 static void qemu_opt_rename(QemuOpts *opts, const char *from, const char *to,
692                             Error **errp)
693 {
694     const char *value;
695 
696     value = qemu_opt_get(opts, from);
697     if (value) {
698         if (qemu_opt_find(opts, to)) {
699             error_setg(errp, "'%s' and its alias '%s' can't be used at the "
700                        "same time", to, from);
701             return;
702         }
703     }
704 
705     /* rename all items in opts */
706     while ((value = qemu_opt_get(opts, from))) {
707         qemu_opt_set(opts, to, value, &error_abort);
708         qemu_opt_unset(opts, from);
709     }
710 }
711 
712 QemuOptsList qemu_legacy_drive_opts = {
713     .name = "drive",
714     .head = QTAILQ_HEAD_INITIALIZER(qemu_legacy_drive_opts.head),
715     .desc = {
716         {
717             .name = "bus",
718             .type = QEMU_OPT_NUMBER,
719             .help = "bus number",
720         },{
721             .name = "unit",
722             .type = QEMU_OPT_NUMBER,
723             .help = "unit number (i.e. lun for scsi)",
724         },{
725             .name = "index",
726             .type = QEMU_OPT_NUMBER,
727             .help = "index number",
728         },{
729             .name = "media",
730             .type = QEMU_OPT_STRING,
731             .help = "media type (disk, cdrom)",
732         },{
733             .name = "if",
734             .type = QEMU_OPT_STRING,
735             .help = "interface (ide, scsi, sd, mtd, floppy, pflash, virtio)",
736         },{
737             .name = "file",
738             .type = QEMU_OPT_STRING,
739             .help = "file name",
740         },
741 
742         /* Options that are passed on, but have special semantics with -drive */
743         {
744             .name = BDRV_OPT_READ_ONLY,
745             .type = QEMU_OPT_BOOL,
746             .help = "open drive file as read-only",
747         },{
748             .name = "rerror",
749             .type = QEMU_OPT_STRING,
750             .help = "read error action",
751         },{
752             .name = "werror",
753             .type = QEMU_OPT_STRING,
754             .help = "write error action",
755         },{
756             .name = "copy-on-read",
757             .type = QEMU_OPT_BOOL,
758             .help = "copy read data from backing file into image file",
759         },
760 
761         { /* end of list */ }
762     },
763 };
764 
765 DriveInfo *drive_new(QemuOpts *all_opts, BlockInterfaceType block_default_type,
766                      Error **errp)
767 {
768     const char *value;
769     BlockBackend *blk;
770     DriveInfo *dinfo = NULL;
771     QDict *bs_opts;
772     QemuOpts *legacy_opts;
773     DriveMediaType media = MEDIA_DISK;
774     BlockInterfaceType type;
775     int max_devs, bus_id, unit_id, index;
776     const char *werror, *rerror;
777     bool read_only = false;
778     bool copy_on_read;
779     const char *filename;
780     Error *local_err = NULL;
781     int i;
782 
783     /* Change legacy command line options into QMP ones */
784     static const struct {
785         const char *from;
786         const char *to;
787     } opt_renames[] = {
788         { "iops",           "throttling.iops-total" },
789         { "iops_rd",        "throttling.iops-read" },
790         { "iops_wr",        "throttling.iops-write" },
791 
792         { "bps",            "throttling.bps-total" },
793         { "bps_rd",         "throttling.bps-read" },
794         { "bps_wr",         "throttling.bps-write" },
795 
796         { "iops_max",       "throttling.iops-total-max" },
797         { "iops_rd_max",    "throttling.iops-read-max" },
798         { "iops_wr_max",    "throttling.iops-write-max" },
799 
800         { "bps_max",        "throttling.bps-total-max" },
801         { "bps_rd_max",     "throttling.bps-read-max" },
802         { "bps_wr_max",     "throttling.bps-write-max" },
803 
804         { "iops_size",      "throttling.iops-size" },
805 
806         { "group",          "throttling.group" },
807 
808         { "readonly",       BDRV_OPT_READ_ONLY },
809     };
810 
811     for (i = 0; i < ARRAY_SIZE(opt_renames); i++) {
812         qemu_opt_rename(all_opts, opt_renames[i].from, opt_renames[i].to,
813                         &local_err);
814         if (local_err) {
815             error_propagate(errp, local_err);
816             return NULL;
817         }
818     }
819 
820     value = qemu_opt_get(all_opts, "cache");
821     if (value) {
822         int flags = 0;
823         bool writethrough;
824 
825         if (bdrv_parse_cache_mode(value, &flags, &writethrough) != 0) {
826             error_setg(errp, "invalid cache option");
827             return NULL;
828         }
829 
830         /* Specific options take precedence */
831         if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_WB)) {
832             qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_WB,
833                               !writethrough, &error_abort);
834         }
835         if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_DIRECT)) {
836             qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_DIRECT,
837                               !!(flags & BDRV_O_NOCACHE), &error_abort);
838         }
839         if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_NO_FLUSH)) {
840             qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_NO_FLUSH,
841                               !!(flags & BDRV_O_NO_FLUSH), &error_abort);
842         }
843         qemu_opt_unset(all_opts, "cache");
844     }
845 
846     /* Get a QDict for processing the options */
847     bs_opts = qdict_new();
848     qemu_opts_to_qdict(all_opts, bs_opts);
849 
850     legacy_opts = qemu_opts_create(&qemu_legacy_drive_opts, NULL, 0,
851                                    &error_abort);
852     qemu_opts_absorb_qdict(legacy_opts, bs_opts, &local_err);
853     if (local_err) {
854         error_propagate(errp, local_err);
855         goto fail;
856     }
857 
858     /* Media type */
859     value = qemu_opt_get(legacy_opts, "media");
860     if (value) {
861         if (!strcmp(value, "disk")) {
862             media = MEDIA_DISK;
863         } else if (!strcmp(value, "cdrom")) {
864             media = MEDIA_CDROM;
865             read_only = true;
866         } else {
867             error_setg(errp, "'%s' invalid media", value);
868             goto fail;
869         }
870     }
871 
872     /* copy-on-read is disabled with a warning for read-only devices */
873     read_only |= qemu_opt_get_bool(legacy_opts, BDRV_OPT_READ_ONLY, false);
874     copy_on_read = qemu_opt_get_bool(legacy_opts, "copy-on-read", false);
875 
876     if (read_only && copy_on_read) {
877         warn_report("disabling copy-on-read on read-only drive");
878         copy_on_read = false;
879     }
880 
881     qdict_put_str(bs_opts, BDRV_OPT_READ_ONLY, read_only ? "on" : "off");
882     qdict_put_str(bs_opts, "copy-on-read", copy_on_read ? "on" : "off");
883 
884     /* Controller type */
885     value = qemu_opt_get(legacy_opts, "if");
886     if (value) {
887         for (type = 0;
888              type < IF_COUNT && strcmp(value, if_name[type]);
889              type++) {
890         }
891         if (type == IF_COUNT) {
892             error_setg(errp, "unsupported bus type '%s'", value);
893             goto fail;
894         }
895     } else {
896         type = block_default_type;
897     }
898 
899     /* Device address specified by bus/unit or index.
900      * If none was specified, try to find the first free one. */
901     bus_id  = qemu_opt_get_number(legacy_opts, "bus", 0);
902     unit_id = qemu_opt_get_number(legacy_opts, "unit", -1);
903     index   = qemu_opt_get_number(legacy_opts, "index", -1);
904 
905     max_devs = if_max_devs[type];
906 
907     if (index != -1) {
908         if (bus_id != 0 || unit_id != -1) {
909             error_setg(errp, "index cannot be used with bus and unit");
910             goto fail;
911         }
912         bus_id = drive_index_to_bus_id(type, index);
913         unit_id = drive_index_to_unit_id(type, index);
914     }
915 
916     if (unit_id == -1) {
917        unit_id = 0;
918        while (drive_get(type, bus_id, unit_id) != NULL) {
919            unit_id++;
920            if (max_devs && unit_id >= max_devs) {
921                unit_id -= max_devs;
922                bus_id++;
923            }
924        }
925     }
926 
927     if (max_devs && unit_id >= max_devs) {
928         error_setg(errp, "unit %d too big (max is %d)", unit_id, max_devs - 1);
929         goto fail;
930     }
931 
932     if (drive_get(type, bus_id, unit_id) != NULL) {
933         error_setg(errp, "drive with bus=%d, unit=%d (index=%d) exists",
934                    bus_id, unit_id, index);
935         goto fail;
936     }
937 
938     /* no id supplied -> create one */
939     if (qemu_opts_id(all_opts) == NULL) {
940         char *new_id;
941         const char *mediastr = "";
942         if (type == IF_IDE || type == IF_SCSI) {
943             mediastr = (media == MEDIA_CDROM) ? "-cd" : "-hd";
944         }
945         if (max_devs) {
946             new_id = g_strdup_printf("%s%i%s%i", if_name[type], bus_id,
947                                      mediastr, unit_id);
948         } else {
949             new_id = g_strdup_printf("%s%s%i", if_name[type],
950                                      mediastr, unit_id);
951         }
952         qdict_put_str(bs_opts, "id", new_id);
953         g_free(new_id);
954     }
955 
956     /* Add virtio block device */
957     if (type == IF_VIRTIO) {
958         QemuOpts *devopts;
959         devopts = qemu_opts_create(qemu_find_opts("device"), NULL, 0,
960                                    &error_abort);
961         if (arch_type == QEMU_ARCH_S390X) {
962             qemu_opt_set(devopts, "driver", "virtio-blk-ccw", &error_abort);
963         } else {
964             qemu_opt_set(devopts, "driver", "virtio-blk-pci", &error_abort);
965         }
966         qemu_opt_set(devopts, "drive", qdict_get_str(bs_opts, "id"),
967                      &error_abort);
968     }
969 
970     filename = qemu_opt_get(legacy_opts, "file");
971 
972     /* Check werror/rerror compatibility with if=... */
973     werror = qemu_opt_get(legacy_opts, "werror");
974     if (werror != NULL) {
975         if (type != IF_IDE && type != IF_SCSI && type != IF_VIRTIO &&
976             type != IF_NONE) {
977             error_setg(errp, "werror is not supported by this bus type");
978             goto fail;
979         }
980         qdict_put_str(bs_opts, "werror", werror);
981     }
982 
983     rerror = qemu_opt_get(legacy_opts, "rerror");
984     if (rerror != NULL) {
985         if (type != IF_IDE && type != IF_VIRTIO && type != IF_SCSI &&
986             type != IF_NONE) {
987             error_setg(errp, "rerror is not supported by this bus type");
988             goto fail;
989         }
990         qdict_put_str(bs_opts, "rerror", rerror);
991     }
992 
993     /* Actual block device init: Functionality shared with blockdev-add */
994     blk = blockdev_init(filename, bs_opts, &local_err);
995     bs_opts = NULL;
996     if (!blk) {
997         error_propagate(errp, local_err);
998         goto fail;
999     } else {
1000         assert(!local_err);
1001     }
1002 
1003     /* Create legacy DriveInfo */
1004     dinfo = g_malloc0(sizeof(*dinfo));
1005     dinfo->opts = all_opts;
1006 
1007     dinfo->type = type;
1008     dinfo->bus = bus_id;
1009     dinfo->unit = unit_id;
1010 
1011     blk_set_legacy_dinfo(blk, dinfo);
1012 
1013     switch(type) {
1014     case IF_IDE:
1015     case IF_SCSI:
1016     case IF_XEN:
1017     case IF_NONE:
1018         dinfo->media_cd = media == MEDIA_CDROM;
1019         break;
1020     default:
1021         break;
1022     }
1023 
1024 fail:
1025     qemu_opts_del(legacy_opts);
1026     qobject_unref(bs_opts);
1027     return dinfo;
1028 }
1029 
1030 static BlockDriverState *qmp_get_root_bs(const char *name, Error **errp)
1031 {
1032     BlockDriverState *bs;
1033 
1034     bs = bdrv_lookup_bs(name, name, errp);
1035     if (bs == NULL) {
1036         return NULL;
1037     }
1038 
1039     if (!bdrv_is_root_node(bs)) {
1040         error_setg(errp, "Need a root block node");
1041         return NULL;
1042     }
1043 
1044     if (!bdrv_is_inserted(bs)) {
1045         error_setg(errp, "Device has no medium");
1046         return NULL;
1047     }
1048 
1049     return bs;
1050 }
1051 
1052 static BlockBackend *qmp_get_blk(const char *blk_name, const char *qdev_id,
1053                                  Error **errp)
1054 {
1055     BlockBackend *blk;
1056 
1057     if (!blk_name == !qdev_id) {
1058         error_setg(errp, "Need exactly one of 'device' and 'id'");
1059         return NULL;
1060     }
1061 
1062     if (qdev_id) {
1063         blk = blk_by_qdev_id(qdev_id, errp);
1064     } else {
1065         blk = blk_by_name(blk_name);
1066         if (blk == NULL) {
1067             error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
1068                       "Device '%s' not found", blk_name);
1069         }
1070     }
1071 
1072     return blk;
1073 }
1074 
1075 void hmp_commit(Monitor *mon, const QDict *qdict)
1076 {
1077     const char *device = qdict_get_str(qdict, "device");
1078     BlockBackend *blk;
1079     int ret;
1080 
1081     if (!strcmp(device, "all")) {
1082         ret = blk_commit_all();
1083     } else {
1084         BlockDriverState *bs;
1085         AioContext *aio_context;
1086 
1087         blk = blk_by_name(device);
1088         if (!blk) {
1089             monitor_printf(mon, "Device '%s' not found\n", device);
1090             return;
1091         }
1092         if (!blk_is_available(blk)) {
1093             monitor_printf(mon, "Device '%s' has no medium\n", device);
1094             return;
1095         }
1096 
1097         bs = blk_bs(blk);
1098         aio_context = bdrv_get_aio_context(bs);
1099         aio_context_acquire(aio_context);
1100 
1101         ret = bdrv_commit(bs);
1102 
1103         aio_context_release(aio_context);
1104     }
1105     if (ret < 0) {
1106         monitor_printf(mon, "'commit' error for '%s': %s\n", device,
1107                        strerror(-ret));
1108     }
1109 }
1110 
1111 static void blockdev_do_action(TransactionAction *action, Error **errp)
1112 {
1113     TransactionActionList list;
1114 
1115     list.value = action;
1116     list.next = NULL;
1117     qmp_transaction(&list, false, NULL, errp);
1118 }
1119 
1120 void qmp_blockdev_snapshot_sync(bool has_device, const char *device,
1121                                 bool has_node_name, const char *node_name,
1122                                 const char *snapshot_file,
1123                                 bool has_snapshot_node_name,
1124                                 const char *snapshot_node_name,
1125                                 bool has_format, const char *format,
1126                                 bool has_mode, NewImageMode mode, Error **errp)
1127 {
1128     BlockdevSnapshotSync snapshot = {
1129         .has_device = has_device,
1130         .device = (char *) device,
1131         .has_node_name = has_node_name,
1132         .node_name = (char *) node_name,
1133         .snapshot_file = (char *) snapshot_file,
1134         .has_snapshot_node_name = has_snapshot_node_name,
1135         .snapshot_node_name = (char *) snapshot_node_name,
1136         .has_format = has_format,
1137         .format = (char *) format,
1138         .has_mode = has_mode,
1139         .mode = mode,
1140     };
1141     TransactionAction action = {
1142         .type = TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC,
1143         .u.blockdev_snapshot_sync.data = &snapshot,
1144     };
1145     blockdev_do_action(&action, errp);
1146 }
1147 
1148 void qmp_blockdev_snapshot(const char *node, const char *overlay,
1149                            Error **errp)
1150 {
1151     BlockdevSnapshot snapshot_data = {
1152         .node = (char *) node,
1153         .overlay = (char *) overlay
1154     };
1155     TransactionAction action = {
1156         .type = TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT,
1157         .u.blockdev_snapshot.data = &snapshot_data,
1158     };
1159     blockdev_do_action(&action, errp);
1160 }
1161 
1162 void qmp_blockdev_snapshot_internal_sync(const char *device,
1163                                          const char *name,
1164                                          Error **errp)
1165 {
1166     BlockdevSnapshotInternal snapshot = {
1167         .device = (char *) device,
1168         .name = (char *) name
1169     };
1170     TransactionAction action = {
1171         .type = TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC,
1172         .u.blockdev_snapshot_internal_sync.data = &snapshot,
1173     };
1174     blockdev_do_action(&action, errp);
1175 }
1176 
1177 SnapshotInfo *qmp_blockdev_snapshot_delete_internal_sync(const char *device,
1178                                                          bool has_id,
1179                                                          const char *id,
1180                                                          bool has_name,
1181                                                          const char *name,
1182                                                          Error **errp)
1183 {
1184     BlockDriverState *bs;
1185     AioContext *aio_context;
1186     QEMUSnapshotInfo sn;
1187     Error *local_err = NULL;
1188     SnapshotInfo *info = NULL;
1189     int ret;
1190 
1191     bs = qmp_get_root_bs(device, errp);
1192     if (!bs) {
1193         return NULL;
1194     }
1195     aio_context = bdrv_get_aio_context(bs);
1196     aio_context_acquire(aio_context);
1197 
1198     if (!has_id) {
1199         id = NULL;
1200     }
1201 
1202     if (!has_name) {
1203         name = NULL;
1204     }
1205 
1206     if (!id && !name) {
1207         error_setg(errp, "Name or id must be provided");
1208         goto out_aio_context;
1209     }
1210 
1211     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_INTERNAL_SNAPSHOT_DELETE, errp)) {
1212         goto out_aio_context;
1213     }
1214 
1215     ret = bdrv_snapshot_find_by_id_and_name(bs, id, name, &sn, &local_err);
1216     if (local_err) {
1217         error_propagate(errp, local_err);
1218         goto out_aio_context;
1219     }
1220     if (!ret) {
1221         error_setg(errp,
1222                    "Snapshot with id '%s' and name '%s' does not exist on "
1223                    "device '%s'",
1224                    STR_OR_NULL(id), STR_OR_NULL(name), device);
1225         goto out_aio_context;
1226     }
1227 
1228     bdrv_snapshot_delete(bs, id, name, &local_err);
1229     if (local_err) {
1230         error_propagate(errp, local_err);
1231         goto out_aio_context;
1232     }
1233 
1234     aio_context_release(aio_context);
1235 
1236     info = g_new0(SnapshotInfo, 1);
1237     info->id = g_strdup(sn.id_str);
1238     info->name = g_strdup(sn.name);
1239     info->date_nsec = sn.date_nsec;
1240     info->date_sec = sn.date_sec;
1241     info->vm_state_size = sn.vm_state_size;
1242     info->vm_clock_nsec = sn.vm_clock_nsec % 1000000000;
1243     info->vm_clock_sec = sn.vm_clock_nsec / 1000000000;
1244 
1245     return info;
1246 
1247 out_aio_context:
1248     aio_context_release(aio_context);
1249     return NULL;
1250 }
1251 
1252 /**
1253  * block_dirty_bitmap_lookup:
1254  * Return a dirty bitmap (if present), after validating
1255  * the node reference and bitmap names.
1256  *
1257  * @node: The name of the BDS node to search for bitmaps
1258  * @name: The name of the bitmap to search for
1259  * @pbs: Output pointer for BDS lookup, if desired. Can be NULL.
1260  * @errp: Output pointer for error information. Can be NULL.
1261  *
1262  * @return: A bitmap object on success, or NULL on failure.
1263  */
1264 static BdrvDirtyBitmap *block_dirty_bitmap_lookup(const char *node,
1265                                                   const char *name,
1266                                                   BlockDriverState **pbs,
1267                                                   Error **errp)
1268 {
1269     BlockDriverState *bs;
1270     BdrvDirtyBitmap *bitmap;
1271 
1272     if (!node) {
1273         error_setg(errp, "Node cannot be NULL");
1274         return NULL;
1275     }
1276     if (!name) {
1277         error_setg(errp, "Bitmap name cannot be NULL");
1278         return NULL;
1279     }
1280     bs = bdrv_lookup_bs(node, node, NULL);
1281     if (!bs) {
1282         error_setg(errp, "Node '%s' not found", node);
1283         return NULL;
1284     }
1285 
1286     bitmap = bdrv_find_dirty_bitmap(bs, name);
1287     if (!bitmap) {
1288         error_setg(errp, "Dirty bitmap '%s' not found", name);
1289         return NULL;
1290     }
1291 
1292     if (pbs) {
1293         *pbs = bs;
1294     }
1295 
1296     return bitmap;
1297 }
1298 
1299 /* New and old BlockDriverState structs for atomic group operations */
1300 
1301 typedef struct BlkActionState BlkActionState;
1302 
1303 /**
1304  * BlkActionOps:
1305  * Table of operations that define an Action.
1306  *
1307  * @instance_size: Size of state struct, in bytes.
1308  * @prepare: Prepare the work, must NOT be NULL.
1309  * @commit: Commit the changes, can be NULL.
1310  * @abort: Abort the changes on fail, can be NULL.
1311  * @clean: Clean up resources after all transaction actions have called
1312  *         commit() or abort(). Can be NULL.
1313  *
1314  * Only prepare() may fail. In a single transaction, only one of commit() or
1315  * abort() will be called. clean() will always be called if it is present.
1316  */
1317 typedef struct BlkActionOps {
1318     size_t instance_size;
1319     void (*prepare)(BlkActionState *common, Error **errp);
1320     void (*commit)(BlkActionState *common);
1321     void (*abort)(BlkActionState *common);
1322     void (*clean)(BlkActionState *common);
1323 } BlkActionOps;
1324 
1325 /**
1326  * BlkActionState:
1327  * Describes one Action's state within a Transaction.
1328  *
1329  * @action: QAPI-defined enum identifying which Action to perform.
1330  * @ops: Table of ActionOps this Action can perform.
1331  * @block_job_txn: Transaction which this action belongs to.
1332  * @entry: List membership for all Actions in this Transaction.
1333  *
1334  * This structure must be arranged as first member in a subclassed type,
1335  * assuming that the compiler will also arrange it to the same offsets as the
1336  * base class.
1337  */
1338 struct BlkActionState {
1339     TransactionAction *action;
1340     const BlkActionOps *ops;
1341     JobTxn *block_job_txn;
1342     TransactionProperties *txn_props;
1343     QTAILQ_ENTRY(BlkActionState) entry;
1344 };
1345 
1346 /* internal snapshot private data */
1347 typedef struct InternalSnapshotState {
1348     BlkActionState common;
1349     BlockDriverState *bs;
1350     QEMUSnapshotInfo sn;
1351     bool created;
1352 } InternalSnapshotState;
1353 
1354 
1355 static int action_check_completion_mode(BlkActionState *s, Error **errp)
1356 {
1357     if (s->txn_props->completion_mode != ACTION_COMPLETION_MODE_INDIVIDUAL) {
1358         error_setg(errp,
1359                    "Action '%s' does not support Transaction property "
1360                    "completion-mode = %s",
1361                    TransactionActionKind_str(s->action->type),
1362                    ActionCompletionMode_str(s->txn_props->completion_mode));
1363         return -1;
1364     }
1365     return 0;
1366 }
1367 
1368 static void internal_snapshot_prepare(BlkActionState *common,
1369                                       Error **errp)
1370 {
1371     Error *local_err = NULL;
1372     const char *device;
1373     const char *name;
1374     BlockDriverState *bs;
1375     QEMUSnapshotInfo old_sn, *sn;
1376     bool ret;
1377     qemu_timeval tv;
1378     BlockdevSnapshotInternal *internal;
1379     InternalSnapshotState *state;
1380     AioContext *aio_context;
1381     int ret1;
1382 
1383     g_assert(common->action->type ==
1384              TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC);
1385     internal = common->action->u.blockdev_snapshot_internal_sync.data;
1386     state = DO_UPCAST(InternalSnapshotState, common, common);
1387 
1388     /* 1. parse input */
1389     device = internal->device;
1390     name = internal->name;
1391 
1392     /* 2. check for validation */
1393     if (action_check_completion_mode(common, errp) < 0) {
1394         return;
1395     }
1396 
1397     bs = qmp_get_root_bs(device, errp);
1398     if (!bs) {
1399         return;
1400     }
1401 
1402     aio_context = bdrv_get_aio_context(bs);
1403     aio_context_acquire(aio_context);
1404 
1405     state->bs = bs;
1406 
1407     /* Paired with .clean() */
1408     bdrv_drained_begin(bs);
1409 
1410     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_INTERNAL_SNAPSHOT, errp)) {
1411         goto out;
1412     }
1413 
1414     if (bdrv_is_read_only(bs)) {
1415         error_setg(errp, "Device '%s' is read only", device);
1416         goto out;
1417     }
1418 
1419     if (!bdrv_can_snapshot(bs)) {
1420         error_setg(errp, "Block format '%s' used by device '%s' "
1421                    "does not support internal snapshots",
1422                    bs->drv->format_name, device);
1423         goto out;
1424     }
1425 
1426     if (!strlen(name)) {
1427         error_setg(errp, "Name is empty");
1428         goto out;
1429     }
1430 
1431     /* check whether a snapshot with name exist */
1432     ret = bdrv_snapshot_find_by_id_and_name(bs, NULL, name, &old_sn,
1433                                             &local_err);
1434     if (local_err) {
1435         error_propagate(errp, local_err);
1436         goto out;
1437     } else if (ret) {
1438         error_setg(errp,
1439                    "Snapshot with name '%s' already exists on device '%s'",
1440                    name, device);
1441         goto out;
1442     }
1443 
1444     /* 3. take the snapshot */
1445     sn = &state->sn;
1446     pstrcpy(sn->name, sizeof(sn->name), name);
1447     qemu_gettimeofday(&tv);
1448     sn->date_sec = tv.tv_sec;
1449     sn->date_nsec = tv.tv_usec * 1000;
1450     sn->vm_clock_nsec = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL);
1451 
1452     ret1 = bdrv_snapshot_create(bs, sn);
1453     if (ret1 < 0) {
1454         error_setg_errno(errp, -ret1,
1455                          "Failed to create snapshot '%s' on device '%s'",
1456                          name, device);
1457         goto out;
1458     }
1459 
1460     /* 4. succeed, mark a snapshot is created */
1461     state->created = true;
1462 
1463 out:
1464     aio_context_release(aio_context);
1465 }
1466 
1467 static void internal_snapshot_abort(BlkActionState *common)
1468 {
1469     InternalSnapshotState *state =
1470                              DO_UPCAST(InternalSnapshotState, common, common);
1471     BlockDriverState *bs = state->bs;
1472     QEMUSnapshotInfo *sn = &state->sn;
1473     AioContext *aio_context;
1474     Error *local_error = NULL;
1475 
1476     if (!state->created) {
1477         return;
1478     }
1479 
1480     aio_context = bdrv_get_aio_context(state->bs);
1481     aio_context_acquire(aio_context);
1482 
1483     if (bdrv_snapshot_delete(bs, sn->id_str, sn->name, &local_error) < 0) {
1484         error_reportf_err(local_error,
1485                           "Failed to delete snapshot with id '%s' and "
1486                           "name '%s' on device '%s' in abort: ",
1487                           sn->id_str, sn->name,
1488                           bdrv_get_device_name(bs));
1489     }
1490 
1491     aio_context_release(aio_context);
1492 }
1493 
1494 static void internal_snapshot_clean(BlkActionState *common)
1495 {
1496     InternalSnapshotState *state = DO_UPCAST(InternalSnapshotState,
1497                                              common, common);
1498     AioContext *aio_context;
1499 
1500     if (!state->bs) {
1501         return;
1502     }
1503 
1504     aio_context = bdrv_get_aio_context(state->bs);
1505     aio_context_acquire(aio_context);
1506 
1507     bdrv_drained_end(state->bs);
1508 
1509     aio_context_release(aio_context);
1510 }
1511 
1512 /* external snapshot private data */
1513 typedef struct ExternalSnapshotState {
1514     BlkActionState common;
1515     BlockDriverState *old_bs;
1516     BlockDriverState *new_bs;
1517     bool overlay_appended;
1518 } ExternalSnapshotState;
1519 
1520 static void external_snapshot_prepare(BlkActionState *common,
1521                                       Error **errp)
1522 {
1523     int flags = 0;
1524     QDict *options = NULL;
1525     Error *local_err = NULL;
1526     /* Device and node name of the image to generate the snapshot from */
1527     const char *device;
1528     const char *node_name;
1529     /* Reference to the new image (for 'blockdev-snapshot') */
1530     const char *snapshot_ref;
1531     /* File name of the new image (for 'blockdev-snapshot-sync') */
1532     const char *new_image_file;
1533     ExternalSnapshotState *state =
1534                              DO_UPCAST(ExternalSnapshotState, common, common);
1535     TransactionAction *action = common->action;
1536     AioContext *aio_context;
1537 
1538     /* 'blockdev-snapshot' and 'blockdev-snapshot-sync' have similar
1539      * purpose but a different set of parameters */
1540     switch (action->type) {
1541     case TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT:
1542         {
1543             BlockdevSnapshot *s = action->u.blockdev_snapshot.data;
1544             device = s->node;
1545             node_name = s->node;
1546             new_image_file = NULL;
1547             snapshot_ref = s->overlay;
1548         }
1549         break;
1550     case TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC:
1551         {
1552             BlockdevSnapshotSync *s = action->u.blockdev_snapshot_sync.data;
1553             device = s->has_device ? s->device : NULL;
1554             node_name = s->has_node_name ? s->node_name : NULL;
1555             new_image_file = s->snapshot_file;
1556             snapshot_ref = NULL;
1557         }
1558         break;
1559     default:
1560         g_assert_not_reached();
1561     }
1562 
1563     /* start processing */
1564     if (action_check_completion_mode(common, errp) < 0) {
1565         return;
1566     }
1567 
1568     state->old_bs = bdrv_lookup_bs(device, node_name, errp);
1569     if (!state->old_bs) {
1570         return;
1571     }
1572 
1573     aio_context = bdrv_get_aio_context(state->old_bs);
1574     aio_context_acquire(aio_context);
1575 
1576     /* Paired with .clean() */
1577     bdrv_drained_begin(state->old_bs);
1578 
1579     if (!bdrv_is_inserted(state->old_bs)) {
1580         error_setg(errp, QERR_DEVICE_HAS_NO_MEDIUM, device);
1581         goto out;
1582     }
1583 
1584     if (bdrv_op_is_blocked(state->old_bs,
1585                            BLOCK_OP_TYPE_EXTERNAL_SNAPSHOT, errp)) {
1586         goto out;
1587     }
1588 
1589     if (!bdrv_is_read_only(state->old_bs)) {
1590         if (bdrv_flush(state->old_bs)) {
1591             error_setg(errp, QERR_IO_ERROR);
1592             goto out;
1593         }
1594     }
1595 
1596     if (!bdrv_is_first_non_filter(state->old_bs)) {
1597         error_setg(errp, QERR_FEATURE_DISABLED, "snapshot");
1598         goto out;
1599     }
1600 
1601     if (action->type == TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC) {
1602         BlockdevSnapshotSync *s = action->u.blockdev_snapshot_sync.data;
1603         const char *format = s->has_format ? s->format : "qcow2";
1604         enum NewImageMode mode;
1605         const char *snapshot_node_name =
1606             s->has_snapshot_node_name ? s->snapshot_node_name : NULL;
1607 
1608         if (node_name && !snapshot_node_name) {
1609             error_setg(errp, "New snapshot node name missing");
1610             goto out;
1611         }
1612 
1613         if (snapshot_node_name &&
1614             bdrv_lookup_bs(snapshot_node_name, snapshot_node_name, NULL)) {
1615             error_setg(errp, "New snapshot node name already in use");
1616             goto out;
1617         }
1618 
1619         flags = state->old_bs->open_flags;
1620         flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_COPY_ON_READ);
1621         flags |= BDRV_O_NO_BACKING;
1622 
1623         /* create new image w/backing file */
1624         mode = s->has_mode ? s->mode : NEW_IMAGE_MODE_ABSOLUTE_PATHS;
1625         if (mode != NEW_IMAGE_MODE_EXISTING) {
1626             int64_t size = bdrv_getlength(state->old_bs);
1627             if (size < 0) {
1628                 error_setg_errno(errp, -size, "bdrv_getlength failed");
1629                 goto out;
1630             }
1631             bdrv_refresh_filename(state->old_bs);
1632             bdrv_img_create(new_image_file, format,
1633                             state->old_bs->filename,
1634                             state->old_bs->drv->format_name,
1635                             NULL, size, flags, false, &local_err);
1636             if (local_err) {
1637                 error_propagate(errp, local_err);
1638                 goto out;
1639             }
1640         }
1641 
1642         options = qdict_new();
1643         if (snapshot_node_name) {
1644             qdict_put_str(options, "node-name", snapshot_node_name);
1645         }
1646         qdict_put_str(options, "driver", format);
1647     }
1648 
1649     state->new_bs = bdrv_open(new_image_file, snapshot_ref, options, flags,
1650                               errp);
1651     /* We will manually add the backing_hd field to the bs later */
1652     if (!state->new_bs) {
1653         goto out;
1654     }
1655 
1656     if (bdrv_has_blk(state->new_bs)) {
1657         error_setg(errp, "The snapshot is already in use");
1658         goto out;
1659     }
1660 
1661     if (bdrv_op_is_blocked(state->new_bs, BLOCK_OP_TYPE_EXTERNAL_SNAPSHOT,
1662                            errp)) {
1663         goto out;
1664     }
1665 
1666     if (state->new_bs->backing != NULL) {
1667         error_setg(errp, "The snapshot already has a backing image");
1668         goto out;
1669     }
1670 
1671     if (!state->new_bs->drv->supports_backing) {
1672         error_setg(errp, "The snapshot does not support backing images");
1673         goto out;
1674     }
1675 
1676     bdrv_set_aio_context(state->new_bs, aio_context);
1677 
1678     /* This removes our old bs and adds the new bs. This is an operation that
1679      * can fail, so we need to do it in .prepare; undoing it for abort is
1680      * always possible. */
1681     bdrv_ref(state->new_bs);
1682     bdrv_append(state->new_bs, state->old_bs, &local_err);
1683     if (local_err) {
1684         error_propagate(errp, local_err);
1685         goto out;
1686     }
1687     state->overlay_appended = true;
1688 
1689 out:
1690     aio_context_release(aio_context);
1691 }
1692 
1693 static void external_snapshot_commit(BlkActionState *common)
1694 {
1695     ExternalSnapshotState *state =
1696                              DO_UPCAST(ExternalSnapshotState, common, common);
1697     AioContext *aio_context;
1698 
1699     aio_context = bdrv_get_aio_context(state->old_bs);
1700     aio_context_acquire(aio_context);
1701 
1702     /* We don't need (or want) to use the transactional
1703      * bdrv_reopen_multiple() across all the entries at once, because we
1704      * don't want to abort all of them if one of them fails the reopen */
1705     if (!atomic_read(&state->old_bs->copy_on_read)) {
1706         bdrv_reopen_set_read_only(state->old_bs, true, NULL);
1707     }
1708 
1709     aio_context_release(aio_context);
1710 }
1711 
1712 static void external_snapshot_abort(BlkActionState *common)
1713 {
1714     ExternalSnapshotState *state =
1715                              DO_UPCAST(ExternalSnapshotState, common, common);
1716     if (state->new_bs) {
1717         if (state->overlay_appended) {
1718             AioContext *aio_context;
1719 
1720             aio_context = bdrv_get_aio_context(state->old_bs);
1721             aio_context_acquire(aio_context);
1722 
1723             bdrv_ref(state->old_bs);   /* we can't let bdrv_set_backind_hd()
1724                                           close state->old_bs; we need it */
1725             bdrv_set_backing_hd(state->new_bs, NULL, &error_abort);
1726             bdrv_replace_node(state->new_bs, state->old_bs, &error_abort);
1727             bdrv_unref(state->old_bs); /* bdrv_replace_node() ref'ed old_bs */
1728 
1729             aio_context_release(aio_context);
1730         }
1731     }
1732 }
1733 
1734 static void external_snapshot_clean(BlkActionState *common)
1735 {
1736     ExternalSnapshotState *state =
1737                              DO_UPCAST(ExternalSnapshotState, common, common);
1738     AioContext *aio_context;
1739 
1740     if (!state->old_bs) {
1741         return;
1742     }
1743 
1744     aio_context = bdrv_get_aio_context(state->old_bs);
1745     aio_context_acquire(aio_context);
1746 
1747     bdrv_drained_end(state->old_bs);
1748     bdrv_unref(state->new_bs);
1749 
1750     aio_context_release(aio_context);
1751 }
1752 
1753 typedef struct DriveBackupState {
1754     BlkActionState common;
1755     BlockDriverState *bs;
1756     BlockJob *job;
1757 } DriveBackupState;
1758 
1759 static BlockJob *do_drive_backup(DriveBackup *backup, JobTxn *txn,
1760                             Error **errp);
1761 
1762 static void drive_backup_prepare(BlkActionState *common, Error **errp)
1763 {
1764     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1765     BlockDriverState *bs;
1766     DriveBackup *backup;
1767     AioContext *aio_context;
1768     Error *local_err = NULL;
1769 
1770     assert(common->action->type == TRANSACTION_ACTION_KIND_DRIVE_BACKUP);
1771     backup = common->action->u.drive_backup.data;
1772 
1773     bs = qmp_get_root_bs(backup->device, errp);
1774     if (!bs) {
1775         return;
1776     }
1777 
1778     aio_context = bdrv_get_aio_context(bs);
1779     aio_context_acquire(aio_context);
1780 
1781     /* Paired with .clean() */
1782     bdrv_drained_begin(bs);
1783 
1784     state->bs = bs;
1785 
1786     state->job = do_drive_backup(backup, common->block_job_txn, &local_err);
1787     if (local_err) {
1788         error_propagate(errp, local_err);
1789         goto out;
1790     }
1791 
1792 out:
1793     aio_context_release(aio_context);
1794 }
1795 
1796 static void drive_backup_commit(BlkActionState *common)
1797 {
1798     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1799     AioContext *aio_context;
1800 
1801     aio_context = bdrv_get_aio_context(state->bs);
1802     aio_context_acquire(aio_context);
1803 
1804     assert(state->job);
1805     job_start(&state->job->job);
1806 
1807     aio_context_release(aio_context);
1808 }
1809 
1810 static void drive_backup_abort(BlkActionState *common)
1811 {
1812     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1813 
1814     if (state->job) {
1815         AioContext *aio_context;
1816 
1817         aio_context = bdrv_get_aio_context(state->bs);
1818         aio_context_acquire(aio_context);
1819 
1820         job_cancel_sync(&state->job->job);
1821 
1822         aio_context_release(aio_context);
1823     }
1824 }
1825 
1826 static void drive_backup_clean(BlkActionState *common)
1827 {
1828     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1829     AioContext *aio_context;
1830 
1831     if (!state->bs) {
1832         return;
1833     }
1834 
1835     aio_context = bdrv_get_aio_context(state->bs);
1836     aio_context_acquire(aio_context);
1837 
1838     bdrv_drained_end(state->bs);
1839 
1840     aio_context_release(aio_context);
1841 }
1842 
1843 typedef struct BlockdevBackupState {
1844     BlkActionState common;
1845     BlockDriverState *bs;
1846     BlockJob *job;
1847 } BlockdevBackupState;
1848 
1849 static BlockJob *do_blockdev_backup(BlockdevBackup *backup, JobTxn *txn,
1850                                     Error **errp);
1851 
1852 static void blockdev_backup_prepare(BlkActionState *common, Error **errp)
1853 {
1854     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1855     BlockdevBackup *backup;
1856     BlockDriverState *bs, *target;
1857     AioContext *aio_context;
1858     Error *local_err = NULL;
1859 
1860     assert(common->action->type == TRANSACTION_ACTION_KIND_BLOCKDEV_BACKUP);
1861     backup = common->action->u.blockdev_backup.data;
1862 
1863     bs = bdrv_lookup_bs(backup->device, backup->device, errp);
1864     if (!bs) {
1865         return;
1866     }
1867 
1868     target = bdrv_lookup_bs(backup->target, backup->target, errp);
1869     if (!target) {
1870         return;
1871     }
1872 
1873     aio_context = bdrv_get_aio_context(bs);
1874     if (aio_context != bdrv_get_aio_context(target)) {
1875         error_setg(errp, "Backup between two IO threads is not implemented");
1876         return;
1877     }
1878     aio_context_acquire(aio_context);
1879     state->bs = bs;
1880 
1881     /* Paired with .clean() */
1882     bdrv_drained_begin(state->bs);
1883 
1884     state->job = do_blockdev_backup(backup, common->block_job_txn, &local_err);
1885     if (local_err) {
1886         error_propagate(errp, local_err);
1887         goto out;
1888     }
1889 
1890 out:
1891     aio_context_release(aio_context);
1892 }
1893 
1894 static void blockdev_backup_commit(BlkActionState *common)
1895 {
1896     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1897     AioContext *aio_context;
1898 
1899     aio_context = bdrv_get_aio_context(state->bs);
1900     aio_context_acquire(aio_context);
1901 
1902     assert(state->job);
1903     job_start(&state->job->job);
1904 
1905     aio_context_release(aio_context);
1906 }
1907 
1908 static void blockdev_backup_abort(BlkActionState *common)
1909 {
1910     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1911 
1912     if (state->job) {
1913         AioContext *aio_context;
1914 
1915         aio_context = bdrv_get_aio_context(state->bs);
1916         aio_context_acquire(aio_context);
1917 
1918         job_cancel_sync(&state->job->job);
1919 
1920         aio_context_release(aio_context);
1921     }
1922 }
1923 
1924 static void blockdev_backup_clean(BlkActionState *common)
1925 {
1926     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1927     AioContext *aio_context;
1928 
1929     if (!state->bs) {
1930         return;
1931     }
1932 
1933     aio_context = bdrv_get_aio_context(state->bs);
1934     aio_context_acquire(aio_context);
1935 
1936     bdrv_drained_end(state->bs);
1937 
1938     aio_context_release(aio_context);
1939 }
1940 
1941 typedef struct BlockDirtyBitmapState {
1942     BlkActionState common;
1943     BdrvDirtyBitmap *bitmap;
1944     BlockDriverState *bs;
1945     HBitmap *backup;
1946     bool prepared;
1947     bool was_enabled;
1948 } BlockDirtyBitmapState;
1949 
1950 static void block_dirty_bitmap_add_prepare(BlkActionState *common,
1951                                            Error **errp)
1952 {
1953     Error *local_err = NULL;
1954     BlockDirtyBitmapAdd *action;
1955     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
1956                                              common, common);
1957 
1958     if (action_check_completion_mode(common, errp) < 0) {
1959         return;
1960     }
1961 
1962     action = common->action->u.block_dirty_bitmap_add.data;
1963     /* AIO context taken and released within qmp_block_dirty_bitmap_add */
1964     qmp_block_dirty_bitmap_add(action->node, action->name,
1965                                action->has_granularity, action->granularity,
1966                                action->has_persistent, action->persistent,
1967                                action->has_autoload, action->autoload,
1968                                action->has_disabled, action->disabled,
1969                                &local_err);
1970 
1971     if (!local_err) {
1972         state->prepared = true;
1973     } else {
1974         error_propagate(errp, local_err);
1975     }
1976 }
1977 
1978 static void block_dirty_bitmap_add_abort(BlkActionState *common)
1979 {
1980     BlockDirtyBitmapAdd *action;
1981     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
1982                                              common, common);
1983 
1984     action = common->action->u.block_dirty_bitmap_add.data;
1985     /* Should not be able to fail: IF the bitmap was added via .prepare(),
1986      * then the node reference and bitmap name must have been valid.
1987      */
1988     if (state->prepared) {
1989         qmp_block_dirty_bitmap_remove(action->node, action->name, &error_abort);
1990     }
1991 }
1992 
1993 static void block_dirty_bitmap_clear_prepare(BlkActionState *common,
1994                                              Error **errp)
1995 {
1996     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
1997                                              common, common);
1998     BlockDirtyBitmap *action;
1999 
2000     if (action_check_completion_mode(common, errp) < 0) {
2001         return;
2002     }
2003 
2004     action = common->action->u.block_dirty_bitmap_clear.data;
2005     state->bitmap = block_dirty_bitmap_lookup(action->node,
2006                                               action->name,
2007                                               &state->bs,
2008                                               errp);
2009     if (!state->bitmap) {
2010         return;
2011     }
2012 
2013     if (bdrv_dirty_bitmap_check(state->bitmap, BDRV_BITMAP_DEFAULT, errp)) {
2014         return;
2015     }
2016 
2017     bdrv_clear_dirty_bitmap(state->bitmap, &state->backup);
2018 }
2019 
2020 static void block_dirty_bitmap_restore(BlkActionState *common)
2021 {
2022     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2023                                              common, common);
2024 
2025     if (state->backup) {
2026         bdrv_restore_dirty_bitmap(state->bitmap, state->backup);
2027     }
2028 }
2029 
2030 static void block_dirty_bitmap_free_backup(BlkActionState *common)
2031 {
2032     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2033                                              common, common);
2034 
2035     hbitmap_free(state->backup);
2036 }
2037 
2038 static void block_dirty_bitmap_enable_prepare(BlkActionState *common,
2039                                               Error **errp)
2040 {
2041     BlockDirtyBitmap *action;
2042     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2043                                              common, common);
2044 
2045     if (action_check_completion_mode(common, errp) < 0) {
2046         return;
2047     }
2048 
2049     action = common->action->u.block_dirty_bitmap_enable.data;
2050     state->bitmap = block_dirty_bitmap_lookup(action->node,
2051                                               action->name,
2052                                               NULL,
2053                                               errp);
2054     if (!state->bitmap) {
2055         return;
2056     }
2057 
2058     if (bdrv_dirty_bitmap_check(state->bitmap, BDRV_BITMAP_ALLOW_RO, errp)) {
2059         return;
2060     }
2061 
2062     state->was_enabled = bdrv_dirty_bitmap_enabled(state->bitmap);
2063     bdrv_enable_dirty_bitmap(state->bitmap);
2064 }
2065 
2066 static void block_dirty_bitmap_enable_abort(BlkActionState *common)
2067 {
2068     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2069                                              common, common);
2070 
2071     if (!state->was_enabled) {
2072         bdrv_disable_dirty_bitmap(state->bitmap);
2073     }
2074 }
2075 
2076 static void block_dirty_bitmap_disable_prepare(BlkActionState *common,
2077                                                Error **errp)
2078 {
2079     BlockDirtyBitmap *action;
2080     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2081                                              common, common);
2082 
2083     if (action_check_completion_mode(common, errp) < 0) {
2084         return;
2085     }
2086 
2087     action = common->action->u.block_dirty_bitmap_disable.data;
2088     state->bitmap = block_dirty_bitmap_lookup(action->node,
2089                                               action->name,
2090                                               NULL,
2091                                               errp);
2092     if (!state->bitmap) {
2093         return;
2094     }
2095 
2096     if (bdrv_dirty_bitmap_check(state->bitmap, BDRV_BITMAP_ALLOW_RO, errp)) {
2097         return;
2098     }
2099 
2100     state->was_enabled = bdrv_dirty_bitmap_enabled(state->bitmap);
2101     bdrv_disable_dirty_bitmap(state->bitmap);
2102 }
2103 
2104 static void block_dirty_bitmap_disable_abort(BlkActionState *common)
2105 {
2106     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2107                                              common, common);
2108 
2109     if (state->was_enabled) {
2110         bdrv_enable_dirty_bitmap(state->bitmap);
2111     }
2112 }
2113 
2114 static BdrvDirtyBitmap *do_block_dirty_bitmap_merge(const char *node,
2115                                                     const char *target,
2116                                                     strList *bitmaps,
2117                                                     HBitmap **backup,
2118                                                     Error **errp);
2119 
2120 static void block_dirty_bitmap_merge_prepare(BlkActionState *common,
2121                                              Error **errp)
2122 {
2123     BlockDirtyBitmapMerge *action;
2124     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2125                                              common, common);
2126 
2127     if (action_check_completion_mode(common, errp) < 0) {
2128         return;
2129     }
2130 
2131     action = common->action->u.block_dirty_bitmap_merge.data;
2132 
2133     state->bitmap = do_block_dirty_bitmap_merge(action->node, action->target,
2134                                                 action->bitmaps, &state->backup,
2135                                                 errp);
2136 }
2137 
2138 static void abort_prepare(BlkActionState *common, Error **errp)
2139 {
2140     error_setg(errp, "Transaction aborted using Abort action");
2141 }
2142 
2143 static void abort_commit(BlkActionState *common)
2144 {
2145     g_assert_not_reached(); /* this action never succeeds */
2146 }
2147 
2148 static const BlkActionOps actions[] = {
2149     [TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT] = {
2150         .instance_size = sizeof(ExternalSnapshotState),
2151         .prepare  = external_snapshot_prepare,
2152         .commit   = external_snapshot_commit,
2153         .abort = external_snapshot_abort,
2154         .clean = external_snapshot_clean,
2155     },
2156     [TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC] = {
2157         .instance_size = sizeof(ExternalSnapshotState),
2158         .prepare  = external_snapshot_prepare,
2159         .commit   = external_snapshot_commit,
2160         .abort = external_snapshot_abort,
2161         .clean = external_snapshot_clean,
2162     },
2163     [TRANSACTION_ACTION_KIND_DRIVE_BACKUP] = {
2164         .instance_size = sizeof(DriveBackupState),
2165         .prepare = drive_backup_prepare,
2166         .commit = drive_backup_commit,
2167         .abort = drive_backup_abort,
2168         .clean = drive_backup_clean,
2169     },
2170     [TRANSACTION_ACTION_KIND_BLOCKDEV_BACKUP] = {
2171         .instance_size = sizeof(BlockdevBackupState),
2172         .prepare = blockdev_backup_prepare,
2173         .commit = blockdev_backup_commit,
2174         .abort = blockdev_backup_abort,
2175         .clean = blockdev_backup_clean,
2176     },
2177     [TRANSACTION_ACTION_KIND_ABORT] = {
2178         .instance_size = sizeof(BlkActionState),
2179         .prepare = abort_prepare,
2180         .commit = abort_commit,
2181     },
2182     [TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC] = {
2183         .instance_size = sizeof(InternalSnapshotState),
2184         .prepare  = internal_snapshot_prepare,
2185         .abort = internal_snapshot_abort,
2186         .clean = internal_snapshot_clean,
2187     },
2188     [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_ADD] = {
2189         .instance_size = sizeof(BlockDirtyBitmapState),
2190         .prepare = block_dirty_bitmap_add_prepare,
2191         .abort = block_dirty_bitmap_add_abort,
2192     },
2193     [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_CLEAR] = {
2194         .instance_size = sizeof(BlockDirtyBitmapState),
2195         .prepare = block_dirty_bitmap_clear_prepare,
2196         .commit = block_dirty_bitmap_free_backup,
2197         .abort = block_dirty_bitmap_restore,
2198     },
2199     [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_ENABLE] = {
2200         .instance_size = sizeof(BlockDirtyBitmapState),
2201         .prepare = block_dirty_bitmap_enable_prepare,
2202         .abort = block_dirty_bitmap_enable_abort,
2203     },
2204     [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_DISABLE] = {
2205         .instance_size = sizeof(BlockDirtyBitmapState),
2206         .prepare = block_dirty_bitmap_disable_prepare,
2207         .abort = block_dirty_bitmap_disable_abort,
2208     },
2209     [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_MERGE] = {
2210         .instance_size = sizeof(BlockDirtyBitmapState),
2211         .prepare = block_dirty_bitmap_merge_prepare,
2212         .commit = block_dirty_bitmap_free_backup,
2213         .abort = block_dirty_bitmap_restore,
2214     },
2215     /* Where are transactions for MIRROR, COMMIT and STREAM?
2216      * Although these blockjobs use transaction callbacks like the backup job,
2217      * these jobs do not necessarily adhere to transaction semantics.
2218      * These jobs may not fully undo all of their actions on abort, nor do they
2219      * necessarily work in transactions with more than one job in them.
2220      */
2221 };
2222 
2223 /**
2224  * Allocate a TransactionProperties structure if necessary, and fill
2225  * that structure with desired defaults if they are unset.
2226  */
2227 static TransactionProperties *get_transaction_properties(
2228     TransactionProperties *props)
2229 {
2230     if (!props) {
2231         props = g_new0(TransactionProperties, 1);
2232     }
2233 
2234     if (!props->has_completion_mode) {
2235         props->has_completion_mode = true;
2236         props->completion_mode = ACTION_COMPLETION_MODE_INDIVIDUAL;
2237     }
2238 
2239     return props;
2240 }
2241 
2242 /*
2243  * 'Atomic' group operations.  The operations are performed as a set, and if
2244  * any fail then we roll back all operations in the group.
2245  */
2246 void qmp_transaction(TransactionActionList *dev_list,
2247                      bool has_props,
2248                      struct TransactionProperties *props,
2249                      Error **errp)
2250 {
2251     TransactionActionList *dev_entry = dev_list;
2252     JobTxn *block_job_txn = NULL;
2253     BlkActionState *state, *next;
2254     Error *local_err = NULL;
2255 
2256     QTAILQ_HEAD(, BlkActionState) snap_bdrv_states;
2257     QTAILQ_INIT(&snap_bdrv_states);
2258 
2259     /* Does this transaction get canceled as a group on failure?
2260      * If not, we don't really need to make a JobTxn.
2261      */
2262     props = get_transaction_properties(props);
2263     if (props->completion_mode != ACTION_COMPLETION_MODE_INDIVIDUAL) {
2264         block_job_txn = job_txn_new();
2265     }
2266 
2267     /* drain all i/o before any operations */
2268     bdrv_drain_all();
2269 
2270     /* We don't do anything in this loop that commits us to the operations */
2271     while (NULL != dev_entry) {
2272         TransactionAction *dev_info = NULL;
2273         const BlkActionOps *ops;
2274 
2275         dev_info = dev_entry->value;
2276         dev_entry = dev_entry->next;
2277 
2278         assert(dev_info->type < ARRAY_SIZE(actions));
2279 
2280         ops = &actions[dev_info->type];
2281         assert(ops->instance_size > 0);
2282 
2283         state = g_malloc0(ops->instance_size);
2284         state->ops = ops;
2285         state->action = dev_info;
2286         state->block_job_txn = block_job_txn;
2287         state->txn_props = props;
2288         QTAILQ_INSERT_TAIL(&snap_bdrv_states, state, entry);
2289 
2290         state->ops->prepare(state, &local_err);
2291         if (local_err) {
2292             error_propagate(errp, local_err);
2293             goto delete_and_fail;
2294         }
2295     }
2296 
2297     QTAILQ_FOREACH(state, &snap_bdrv_states, entry) {
2298         if (state->ops->commit) {
2299             state->ops->commit(state);
2300         }
2301     }
2302 
2303     /* success */
2304     goto exit;
2305 
2306 delete_and_fail:
2307     /* failure, and it is all-or-none; roll back all operations */
2308     QTAILQ_FOREACH_REVERSE(state, &snap_bdrv_states, entry) {
2309         if (state->ops->abort) {
2310             state->ops->abort(state);
2311         }
2312     }
2313 exit:
2314     QTAILQ_FOREACH_SAFE(state, &snap_bdrv_states, entry, next) {
2315         if (state->ops->clean) {
2316             state->ops->clean(state);
2317         }
2318         g_free(state);
2319     }
2320     if (!has_props) {
2321         qapi_free_TransactionProperties(props);
2322     }
2323     job_txn_unref(block_job_txn);
2324 }
2325 
2326 void qmp_eject(bool has_device, const char *device,
2327                bool has_id, const char *id,
2328                bool has_force, bool force, Error **errp)
2329 {
2330     Error *local_err = NULL;
2331     int rc;
2332 
2333     if (!has_force) {
2334         force = false;
2335     }
2336 
2337     rc = do_open_tray(has_device ? device : NULL,
2338                       has_id ? id : NULL,
2339                       force, &local_err);
2340     if (rc && rc != -ENOSYS) {
2341         error_propagate(errp, local_err);
2342         return;
2343     }
2344     error_free(local_err);
2345 
2346     blockdev_remove_medium(has_device, device, has_id, id, errp);
2347 }
2348 
2349 void qmp_block_passwd(bool has_device, const char *device,
2350                       bool has_node_name, const char *node_name,
2351                       const char *password, Error **errp)
2352 {
2353     error_setg(errp,
2354                "Setting block passwords directly is no longer supported");
2355 }
2356 
2357 /*
2358  * Attempt to open the tray of @device.
2359  * If @force, ignore its tray lock.
2360  * Else, if the tray is locked, don't open it, but ask the guest to open it.
2361  * On error, store an error through @errp and return -errno.
2362  * If @device does not exist, return -ENODEV.
2363  * If it has no removable media, return -ENOTSUP.
2364  * If it has no tray, return -ENOSYS.
2365  * If the guest was asked to open the tray, return -EINPROGRESS.
2366  * Else, return 0.
2367  */
2368 static int do_open_tray(const char *blk_name, const char *qdev_id,
2369                         bool force, Error **errp)
2370 {
2371     BlockBackend *blk;
2372     const char *device = qdev_id ?: blk_name;
2373     bool locked;
2374 
2375     blk = qmp_get_blk(blk_name, qdev_id, errp);
2376     if (!blk) {
2377         return -ENODEV;
2378     }
2379 
2380     if (!blk_dev_has_removable_media(blk)) {
2381         error_setg(errp, "Device '%s' is not removable", device);
2382         return -ENOTSUP;
2383     }
2384 
2385     if (!blk_dev_has_tray(blk)) {
2386         error_setg(errp, "Device '%s' does not have a tray", device);
2387         return -ENOSYS;
2388     }
2389 
2390     if (blk_dev_is_tray_open(blk)) {
2391         return 0;
2392     }
2393 
2394     locked = blk_dev_is_medium_locked(blk);
2395     if (locked) {
2396         blk_dev_eject_request(blk, force);
2397     }
2398 
2399     if (!locked || force) {
2400         blk_dev_change_media_cb(blk, false, &error_abort);
2401     }
2402 
2403     if (locked && !force) {
2404         error_setg(errp, "Device '%s' is locked and force was not specified, "
2405                    "wait for tray to open and try again", device);
2406         return -EINPROGRESS;
2407     }
2408 
2409     return 0;
2410 }
2411 
2412 void qmp_blockdev_open_tray(bool has_device, const char *device,
2413                             bool has_id, const char *id,
2414                             bool has_force, bool force,
2415                             Error **errp)
2416 {
2417     Error *local_err = NULL;
2418     int rc;
2419 
2420     if (!has_force) {
2421         force = false;
2422     }
2423     rc = do_open_tray(has_device ? device : NULL,
2424                       has_id ? id : NULL,
2425                       force, &local_err);
2426     if (rc && rc != -ENOSYS && rc != -EINPROGRESS) {
2427         error_propagate(errp, local_err);
2428         return;
2429     }
2430     error_free(local_err);
2431 }
2432 
2433 void qmp_blockdev_close_tray(bool has_device, const char *device,
2434                              bool has_id, const char *id,
2435                              Error **errp)
2436 {
2437     BlockBackend *blk;
2438     Error *local_err = NULL;
2439 
2440     device = has_device ? device : NULL;
2441     id = has_id ? id : NULL;
2442 
2443     blk = qmp_get_blk(device, id, errp);
2444     if (!blk) {
2445         return;
2446     }
2447 
2448     if (!blk_dev_has_removable_media(blk)) {
2449         error_setg(errp, "Device '%s' is not removable", device ?: id);
2450         return;
2451     }
2452 
2453     if (!blk_dev_has_tray(blk)) {
2454         /* Ignore this command on tray-less devices */
2455         return;
2456     }
2457 
2458     if (!blk_dev_is_tray_open(blk)) {
2459         return;
2460     }
2461 
2462     blk_dev_change_media_cb(blk, true, &local_err);
2463     if (local_err) {
2464         error_propagate(errp, local_err);
2465         return;
2466     }
2467 }
2468 
2469 static void blockdev_remove_medium(bool has_device, const char *device,
2470                                    bool has_id, const char *id, Error **errp)
2471 {
2472     BlockBackend *blk;
2473     BlockDriverState *bs;
2474     AioContext *aio_context;
2475     bool has_attached_device;
2476 
2477     device = has_device ? device : NULL;
2478     id = has_id ? id : NULL;
2479 
2480     blk = qmp_get_blk(device, id, errp);
2481     if (!blk) {
2482         return;
2483     }
2484 
2485     /* For BBs without a device, we can exchange the BDS tree at will */
2486     has_attached_device = blk_get_attached_dev(blk);
2487 
2488     if (has_attached_device && !blk_dev_has_removable_media(blk)) {
2489         error_setg(errp, "Device '%s' is not removable", device ?: id);
2490         return;
2491     }
2492 
2493     if (has_attached_device && blk_dev_has_tray(blk) &&
2494         !blk_dev_is_tray_open(blk))
2495     {
2496         error_setg(errp, "Tray of device '%s' is not open", device ?: id);
2497         return;
2498     }
2499 
2500     bs = blk_bs(blk);
2501     if (!bs) {
2502         return;
2503     }
2504 
2505     aio_context = bdrv_get_aio_context(bs);
2506     aio_context_acquire(aio_context);
2507 
2508     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_EJECT, errp)) {
2509         goto out;
2510     }
2511 
2512     blk_remove_bs(blk);
2513 
2514     if (!blk_dev_has_tray(blk)) {
2515         /* For tray-less devices, blockdev-open-tray is a no-op (or may not be
2516          * called at all); therefore, the medium needs to be ejected here.
2517          * Do it after blk_remove_bs() so blk_is_inserted(blk) returns the @load
2518          * value passed here (i.e. false). */
2519         blk_dev_change_media_cb(blk, false, &error_abort);
2520     }
2521 
2522 out:
2523     aio_context_release(aio_context);
2524 }
2525 
2526 void qmp_blockdev_remove_medium(const char *id, Error **errp)
2527 {
2528     blockdev_remove_medium(false, NULL, true, id, errp);
2529 }
2530 
2531 static void qmp_blockdev_insert_anon_medium(BlockBackend *blk,
2532                                             BlockDriverState *bs, Error **errp)
2533 {
2534     Error *local_err = NULL;
2535     bool has_device;
2536     int ret;
2537 
2538     /* For BBs without a device, we can exchange the BDS tree at will */
2539     has_device = blk_get_attached_dev(blk);
2540 
2541     if (has_device && !blk_dev_has_removable_media(blk)) {
2542         error_setg(errp, "Device is not removable");
2543         return;
2544     }
2545 
2546     if (has_device && blk_dev_has_tray(blk) && !blk_dev_is_tray_open(blk)) {
2547         error_setg(errp, "Tray of the device is not open");
2548         return;
2549     }
2550 
2551     if (blk_bs(blk)) {
2552         error_setg(errp, "There already is a medium in the device");
2553         return;
2554     }
2555 
2556     ret = blk_insert_bs(blk, bs, errp);
2557     if (ret < 0) {
2558         return;
2559     }
2560 
2561     if (!blk_dev_has_tray(blk)) {
2562         /* For tray-less devices, blockdev-close-tray is a no-op (or may not be
2563          * called at all); therefore, the medium needs to be pushed into the
2564          * slot here.
2565          * Do it after blk_insert_bs() so blk_is_inserted(blk) returns the @load
2566          * value passed here (i.e. true). */
2567         blk_dev_change_media_cb(blk, true, &local_err);
2568         if (local_err) {
2569             error_propagate(errp, local_err);
2570             blk_remove_bs(blk);
2571             return;
2572         }
2573     }
2574 }
2575 
2576 static void blockdev_insert_medium(bool has_device, const char *device,
2577                                    bool has_id, const char *id,
2578                                    const char *node_name, Error **errp)
2579 {
2580     BlockBackend *blk;
2581     BlockDriverState *bs;
2582 
2583     blk = qmp_get_blk(has_device ? device : NULL,
2584                       has_id ? id : NULL,
2585                       errp);
2586     if (!blk) {
2587         return;
2588     }
2589 
2590     bs = bdrv_find_node(node_name);
2591     if (!bs) {
2592         error_setg(errp, "Node '%s' not found", node_name);
2593         return;
2594     }
2595 
2596     if (bdrv_has_blk(bs)) {
2597         error_setg(errp, "Node '%s' is already in use", node_name);
2598         return;
2599     }
2600 
2601     qmp_blockdev_insert_anon_medium(blk, bs, errp);
2602 }
2603 
2604 void qmp_blockdev_insert_medium(const char *id, const char *node_name,
2605                                 Error **errp)
2606 {
2607     blockdev_insert_medium(false, NULL, true, id, node_name, errp);
2608 }
2609 
2610 void qmp_blockdev_change_medium(bool has_device, const char *device,
2611                                 bool has_id, const char *id,
2612                                 const char *filename,
2613                                 bool has_format, const char *format,
2614                                 bool has_read_only,
2615                                 BlockdevChangeReadOnlyMode read_only,
2616                                 Error **errp)
2617 {
2618     BlockBackend *blk;
2619     BlockDriverState *medium_bs = NULL;
2620     int bdrv_flags;
2621     bool detect_zeroes;
2622     int rc;
2623     QDict *options = NULL;
2624     Error *err = NULL;
2625 
2626     blk = qmp_get_blk(has_device ? device : NULL,
2627                       has_id ? id : NULL,
2628                       errp);
2629     if (!blk) {
2630         goto fail;
2631     }
2632 
2633     if (blk_bs(blk)) {
2634         blk_update_root_state(blk);
2635     }
2636 
2637     bdrv_flags = blk_get_open_flags_from_root_state(blk);
2638     bdrv_flags &= ~(BDRV_O_TEMPORARY | BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING |
2639         BDRV_O_PROTOCOL | BDRV_O_AUTO_RDONLY);
2640 
2641     if (!has_read_only) {
2642         read_only = BLOCKDEV_CHANGE_READ_ONLY_MODE_RETAIN;
2643     }
2644 
2645     switch (read_only) {
2646     case BLOCKDEV_CHANGE_READ_ONLY_MODE_RETAIN:
2647         break;
2648 
2649     case BLOCKDEV_CHANGE_READ_ONLY_MODE_READ_ONLY:
2650         bdrv_flags &= ~BDRV_O_RDWR;
2651         break;
2652 
2653     case BLOCKDEV_CHANGE_READ_ONLY_MODE_READ_WRITE:
2654         bdrv_flags |= BDRV_O_RDWR;
2655         break;
2656 
2657     default:
2658         abort();
2659     }
2660 
2661     options = qdict_new();
2662     detect_zeroes = blk_get_detect_zeroes_from_root_state(blk);
2663     qdict_put_str(options, "detect-zeroes", detect_zeroes ? "on" : "off");
2664 
2665     if (has_format) {
2666         qdict_put_str(options, "driver", format);
2667     }
2668 
2669     medium_bs = bdrv_open(filename, NULL, options, bdrv_flags, errp);
2670     if (!medium_bs) {
2671         goto fail;
2672     }
2673 
2674     rc = do_open_tray(has_device ? device : NULL,
2675                       has_id ? id : NULL,
2676                       false, &err);
2677     if (rc && rc != -ENOSYS) {
2678         error_propagate(errp, err);
2679         goto fail;
2680     }
2681     error_free(err);
2682     err = NULL;
2683 
2684     blockdev_remove_medium(has_device, device, has_id, id, &err);
2685     if (err) {
2686         error_propagate(errp, err);
2687         goto fail;
2688     }
2689 
2690     qmp_blockdev_insert_anon_medium(blk, medium_bs, &err);
2691     if (err) {
2692         error_propagate(errp, err);
2693         goto fail;
2694     }
2695 
2696     qmp_blockdev_close_tray(has_device, device, has_id, id, errp);
2697 
2698 fail:
2699     /* If the medium has been inserted, the device has its own reference, so
2700      * ours must be relinquished; and if it has not been inserted successfully,
2701      * the reference must be relinquished anyway */
2702     bdrv_unref(medium_bs);
2703 }
2704 
2705 /* throttling disk I/O limits */
2706 void qmp_block_set_io_throttle(BlockIOThrottle *arg, Error **errp)
2707 {
2708     ThrottleConfig cfg;
2709     BlockDriverState *bs;
2710     BlockBackend *blk;
2711     AioContext *aio_context;
2712 
2713     blk = qmp_get_blk(arg->has_device ? arg->device : NULL,
2714                       arg->has_id ? arg->id : NULL,
2715                       errp);
2716     if (!blk) {
2717         return;
2718     }
2719 
2720     aio_context = blk_get_aio_context(blk);
2721     aio_context_acquire(aio_context);
2722 
2723     bs = blk_bs(blk);
2724     if (!bs) {
2725         error_setg(errp, "Device has no medium");
2726         goto out;
2727     }
2728 
2729     throttle_config_init(&cfg);
2730     cfg.buckets[THROTTLE_BPS_TOTAL].avg = arg->bps;
2731     cfg.buckets[THROTTLE_BPS_READ].avg  = arg->bps_rd;
2732     cfg.buckets[THROTTLE_BPS_WRITE].avg = arg->bps_wr;
2733 
2734     cfg.buckets[THROTTLE_OPS_TOTAL].avg = arg->iops;
2735     cfg.buckets[THROTTLE_OPS_READ].avg  = arg->iops_rd;
2736     cfg.buckets[THROTTLE_OPS_WRITE].avg = arg->iops_wr;
2737 
2738     if (arg->has_bps_max) {
2739         cfg.buckets[THROTTLE_BPS_TOTAL].max = arg->bps_max;
2740     }
2741     if (arg->has_bps_rd_max) {
2742         cfg.buckets[THROTTLE_BPS_READ].max = arg->bps_rd_max;
2743     }
2744     if (arg->has_bps_wr_max) {
2745         cfg.buckets[THROTTLE_BPS_WRITE].max = arg->bps_wr_max;
2746     }
2747     if (arg->has_iops_max) {
2748         cfg.buckets[THROTTLE_OPS_TOTAL].max = arg->iops_max;
2749     }
2750     if (arg->has_iops_rd_max) {
2751         cfg.buckets[THROTTLE_OPS_READ].max = arg->iops_rd_max;
2752     }
2753     if (arg->has_iops_wr_max) {
2754         cfg.buckets[THROTTLE_OPS_WRITE].max = arg->iops_wr_max;
2755     }
2756 
2757     if (arg->has_bps_max_length) {
2758         cfg.buckets[THROTTLE_BPS_TOTAL].burst_length = arg->bps_max_length;
2759     }
2760     if (arg->has_bps_rd_max_length) {
2761         cfg.buckets[THROTTLE_BPS_READ].burst_length = arg->bps_rd_max_length;
2762     }
2763     if (arg->has_bps_wr_max_length) {
2764         cfg.buckets[THROTTLE_BPS_WRITE].burst_length = arg->bps_wr_max_length;
2765     }
2766     if (arg->has_iops_max_length) {
2767         cfg.buckets[THROTTLE_OPS_TOTAL].burst_length = arg->iops_max_length;
2768     }
2769     if (arg->has_iops_rd_max_length) {
2770         cfg.buckets[THROTTLE_OPS_READ].burst_length = arg->iops_rd_max_length;
2771     }
2772     if (arg->has_iops_wr_max_length) {
2773         cfg.buckets[THROTTLE_OPS_WRITE].burst_length = arg->iops_wr_max_length;
2774     }
2775 
2776     if (arg->has_iops_size) {
2777         cfg.op_size = arg->iops_size;
2778     }
2779 
2780     if (!throttle_is_valid(&cfg, errp)) {
2781         goto out;
2782     }
2783 
2784     if (throttle_enabled(&cfg)) {
2785         /* Enable I/O limits if they're not enabled yet, otherwise
2786          * just update the throttling group. */
2787         if (!blk_get_public(blk)->throttle_group_member.throttle_state) {
2788             blk_io_limits_enable(blk,
2789                                  arg->has_group ? arg->group :
2790                                  arg->has_device ? arg->device :
2791                                  arg->id);
2792         } else if (arg->has_group) {
2793             blk_io_limits_update_group(blk, arg->group);
2794         }
2795         /* Set the new throttling configuration */
2796         blk_set_io_limits(blk, &cfg);
2797     } else if (blk_get_public(blk)->throttle_group_member.throttle_state) {
2798         /* If all throttling settings are set to 0, disable I/O limits */
2799         blk_io_limits_disable(blk);
2800     }
2801 
2802 out:
2803     aio_context_release(aio_context);
2804 }
2805 
2806 void qmp_block_dirty_bitmap_add(const char *node, const char *name,
2807                                 bool has_granularity, uint32_t granularity,
2808                                 bool has_persistent, bool persistent,
2809                                 bool has_autoload, bool autoload,
2810                                 bool has_disabled, bool disabled,
2811                                 Error **errp)
2812 {
2813     BlockDriverState *bs;
2814     BdrvDirtyBitmap *bitmap;
2815     AioContext *aio_context = NULL;
2816 
2817     if (!name || name[0] == '\0') {
2818         error_setg(errp, "Bitmap name cannot be empty");
2819         return;
2820     }
2821 
2822     bs = bdrv_lookup_bs(node, node, errp);
2823     if (!bs) {
2824         return;
2825     }
2826 
2827     if (has_granularity) {
2828         if (granularity < 512 || !is_power_of_2(granularity)) {
2829             error_setg(errp, "Granularity must be power of 2 "
2830                              "and at least 512");
2831             return;
2832         }
2833     } else {
2834         /* Default to cluster size, if available: */
2835         granularity = bdrv_get_default_bitmap_granularity(bs);
2836     }
2837 
2838     if (!has_persistent) {
2839         persistent = false;
2840     }
2841 
2842     if (has_autoload) {
2843         warn_report("Autoload option is deprecated and its value is ignored");
2844     }
2845 
2846     if (!has_disabled) {
2847         disabled = false;
2848     }
2849 
2850     if (persistent) {
2851         aio_context = bdrv_get_aio_context(bs);
2852         aio_context_acquire(aio_context);
2853         if (!bdrv_can_store_new_dirty_bitmap(bs, name, granularity, errp)) {
2854             goto out;
2855         }
2856     }
2857 
2858     bitmap = bdrv_create_dirty_bitmap(bs, granularity, name, errp);
2859     if (bitmap == NULL) {
2860         goto out;
2861     }
2862 
2863     if (disabled) {
2864         bdrv_disable_dirty_bitmap(bitmap);
2865     }
2866 
2867     bdrv_dirty_bitmap_set_persistence(bitmap, persistent);
2868  out:
2869     if (aio_context) {
2870         aio_context_release(aio_context);
2871     }
2872 }
2873 
2874 void qmp_block_dirty_bitmap_remove(const char *node, const char *name,
2875                                    Error **errp)
2876 {
2877     BlockDriverState *bs;
2878     BdrvDirtyBitmap *bitmap;
2879     Error *local_err = NULL;
2880     AioContext *aio_context = NULL;
2881 
2882     bitmap = block_dirty_bitmap_lookup(node, name, &bs, errp);
2883     if (!bitmap || !bs) {
2884         return;
2885     }
2886 
2887     if (bdrv_dirty_bitmap_check(bitmap, BDRV_BITMAP_BUSY | BDRV_BITMAP_RO,
2888                                 errp)) {
2889         return;
2890     }
2891 
2892     if (bdrv_dirty_bitmap_get_persistence(bitmap)) {
2893         aio_context = bdrv_get_aio_context(bs);
2894         aio_context_acquire(aio_context);
2895         bdrv_remove_persistent_dirty_bitmap(bs, name, &local_err);
2896         if (local_err != NULL) {
2897             error_propagate(errp, local_err);
2898             goto out;
2899         }
2900     }
2901 
2902     bdrv_release_dirty_bitmap(bs, bitmap);
2903  out:
2904     if (aio_context) {
2905         aio_context_release(aio_context);
2906     }
2907 }
2908 
2909 /**
2910  * Completely clear a bitmap, for the purposes of synchronizing a bitmap
2911  * immediately after a full backup operation.
2912  */
2913 void qmp_block_dirty_bitmap_clear(const char *node, const char *name,
2914                                   Error **errp)
2915 {
2916     BdrvDirtyBitmap *bitmap;
2917     BlockDriverState *bs;
2918 
2919     bitmap = block_dirty_bitmap_lookup(node, name, &bs, errp);
2920     if (!bitmap || !bs) {
2921         return;
2922     }
2923 
2924     if (bdrv_dirty_bitmap_check(bitmap, BDRV_BITMAP_DEFAULT, errp)) {
2925         return;
2926     }
2927 
2928     bdrv_clear_dirty_bitmap(bitmap, NULL);
2929 }
2930 
2931 void qmp_block_dirty_bitmap_enable(const char *node, const char *name,
2932                                    Error **errp)
2933 {
2934     BlockDriverState *bs;
2935     BdrvDirtyBitmap *bitmap;
2936 
2937     bitmap = block_dirty_bitmap_lookup(node, name, &bs, errp);
2938     if (!bitmap) {
2939         return;
2940     }
2941 
2942     if (bdrv_dirty_bitmap_check(bitmap, BDRV_BITMAP_ALLOW_RO, errp)) {
2943         return;
2944     }
2945 
2946     bdrv_enable_dirty_bitmap(bitmap);
2947 }
2948 
2949 void qmp_block_dirty_bitmap_disable(const char *node, const char *name,
2950                                     Error **errp)
2951 {
2952     BlockDriverState *bs;
2953     BdrvDirtyBitmap *bitmap;
2954 
2955     bitmap = block_dirty_bitmap_lookup(node, name, &bs, errp);
2956     if (!bitmap) {
2957         return;
2958     }
2959 
2960     if (bdrv_dirty_bitmap_check(bitmap, BDRV_BITMAP_ALLOW_RO, errp)) {
2961         return;
2962     }
2963 
2964     bdrv_disable_dirty_bitmap(bitmap);
2965 }
2966 
2967 static BdrvDirtyBitmap *do_block_dirty_bitmap_merge(const char *node,
2968                                                     const char *target,
2969                                                     strList *bitmaps,
2970                                                     HBitmap **backup,
2971                                                     Error **errp)
2972 {
2973     BlockDriverState *bs;
2974     BdrvDirtyBitmap *dst, *src, *anon;
2975     strList *lst;
2976     Error *local_err = NULL;
2977 
2978     dst = block_dirty_bitmap_lookup(node, target, &bs, errp);
2979     if (!dst) {
2980         return NULL;
2981     }
2982 
2983     anon = bdrv_create_dirty_bitmap(bs, bdrv_dirty_bitmap_granularity(dst),
2984                                     NULL, errp);
2985     if (!anon) {
2986         return NULL;
2987     }
2988 
2989     for (lst = bitmaps; lst; lst = lst->next) {
2990         src = bdrv_find_dirty_bitmap(bs, lst->value);
2991         if (!src) {
2992             error_setg(errp, "Dirty bitmap '%s' not found", lst->value);
2993             dst = NULL;
2994             goto out;
2995         }
2996 
2997         bdrv_merge_dirty_bitmap(anon, src, NULL, &local_err);
2998         if (local_err) {
2999             error_propagate(errp, local_err);
3000             dst = NULL;
3001             goto out;
3002         }
3003     }
3004 
3005     /* Merge into dst; dst is unchanged on failure. */
3006     bdrv_merge_dirty_bitmap(dst, anon, backup, errp);
3007 
3008  out:
3009     bdrv_release_dirty_bitmap(bs, anon);
3010     return dst;
3011 }
3012 
3013 void qmp_block_dirty_bitmap_merge(const char *node, const char *target,
3014                                   strList *bitmaps, Error **errp)
3015 {
3016     do_block_dirty_bitmap_merge(node, target, bitmaps, NULL, errp);
3017 }
3018 
3019 BlockDirtyBitmapSha256 *qmp_x_debug_block_dirty_bitmap_sha256(const char *node,
3020                                                               const char *name,
3021                                                               Error **errp)
3022 {
3023     BdrvDirtyBitmap *bitmap;
3024     BlockDriverState *bs;
3025     BlockDirtyBitmapSha256 *ret = NULL;
3026     char *sha256;
3027 
3028     bitmap = block_dirty_bitmap_lookup(node, name, &bs, errp);
3029     if (!bitmap || !bs) {
3030         return NULL;
3031     }
3032 
3033     sha256 = bdrv_dirty_bitmap_sha256(bitmap, errp);
3034     if (sha256 == NULL) {
3035         return NULL;
3036     }
3037 
3038     ret = g_new(BlockDirtyBitmapSha256, 1);
3039     ret->sha256 = sha256;
3040 
3041     return ret;
3042 }
3043 
3044 void hmp_drive_del(Monitor *mon, const QDict *qdict)
3045 {
3046     const char *id = qdict_get_str(qdict, "id");
3047     BlockBackend *blk;
3048     BlockDriverState *bs;
3049     AioContext *aio_context;
3050     Error *local_err = NULL;
3051 
3052     bs = bdrv_find_node(id);
3053     if (bs) {
3054         qmp_blockdev_del(id, &local_err);
3055         if (local_err) {
3056             error_report_err(local_err);
3057         }
3058         return;
3059     }
3060 
3061     blk = blk_by_name(id);
3062     if (!blk) {
3063         error_report("Device '%s' not found", id);
3064         return;
3065     }
3066 
3067     if (!blk_legacy_dinfo(blk)) {
3068         error_report("Deleting device added with blockdev-add"
3069                      " is not supported");
3070         return;
3071     }
3072 
3073     aio_context = blk_get_aio_context(blk);
3074     aio_context_acquire(aio_context);
3075 
3076     bs = blk_bs(blk);
3077     if (bs) {
3078         if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_DRIVE_DEL, &local_err)) {
3079             error_report_err(local_err);
3080             aio_context_release(aio_context);
3081             return;
3082         }
3083 
3084         blk_remove_bs(blk);
3085     }
3086 
3087     /* Make the BlockBackend and the attached BlockDriverState anonymous */
3088     monitor_remove_blk(blk);
3089 
3090     /* If this BlockBackend has a device attached to it, its refcount will be
3091      * decremented when the device is removed; otherwise we have to do so here.
3092      */
3093     if (blk_get_attached_dev(blk)) {
3094         /* Further I/O must not pause the guest */
3095         blk_set_on_error(blk, BLOCKDEV_ON_ERROR_REPORT,
3096                          BLOCKDEV_ON_ERROR_REPORT);
3097     } else {
3098         blk_unref(blk);
3099     }
3100 
3101     aio_context_release(aio_context);
3102 }
3103 
3104 void qmp_block_resize(bool has_device, const char *device,
3105                       bool has_node_name, const char *node_name,
3106                       int64_t size, Error **errp)
3107 {
3108     Error *local_err = NULL;
3109     BlockBackend *blk = NULL;
3110     BlockDriverState *bs;
3111     AioContext *aio_context;
3112     int ret;
3113 
3114     bs = bdrv_lookup_bs(has_device ? device : NULL,
3115                         has_node_name ? node_name : NULL,
3116                         &local_err);
3117     if (local_err) {
3118         error_propagate(errp, local_err);
3119         return;
3120     }
3121 
3122     aio_context = bdrv_get_aio_context(bs);
3123     aio_context_acquire(aio_context);
3124 
3125     if (!bdrv_is_first_non_filter(bs)) {
3126         error_setg(errp, QERR_FEATURE_DISABLED, "resize");
3127         goto out;
3128     }
3129 
3130     if (size < 0) {
3131         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "size", "a >0 size");
3132         goto out;
3133     }
3134 
3135     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_RESIZE, NULL)) {
3136         error_setg(errp, QERR_DEVICE_IN_USE, device);
3137         goto out;
3138     }
3139 
3140     blk = blk_new(BLK_PERM_RESIZE, BLK_PERM_ALL);
3141     ret = blk_insert_bs(blk, bs, errp);
3142     if (ret < 0) {
3143         goto out;
3144     }
3145 
3146     bdrv_drained_begin(bs);
3147     ret = blk_truncate(blk, size, PREALLOC_MODE_OFF, errp);
3148     bdrv_drained_end(bs);
3149 
3150 out:
3151     blk_unref(blk);
3152     aio_context_release(aio_context);
3153 }
3154 
3155 void qmp_block_stream(bool has_job_id, const char *job_id, const char *device,
3156                       bool has_base, const char *base,
3157                       bool has_base_node, const char *base_node,
3158                       bool has_backing_file, const char *backing_file,
3159                       bool has_speed, int64_t speed,
3160                       bool has_on_error, BlockdevOnError on_error,
3161                       bool has_auto_finalize, bool auto_finalize,
3162                       bool has_auto_dismiss, bool auto_dismiss,
3163                       Error **errp)
3164 {
3165     BlockDriverState *bs, *iter;
3166     BlockDriverState *base_bs = NULL;
3167     AioContext *aio_context;
3168     Error *local_err = NULL;
3169     const char *base_name = NULL;
3170     int job_flags = JOB_DEFAULT;
3171 
3172     if (!has_on_error) {
3173         on_error = BLOCKDEV_ON_ERROR_REPORT;
3174     }
3175 
3176     bs = bdrv_lookup_bs(device, device, errp);
3177     if (!bs) {
3178         return;
3179     }
3180 
3181     aio_context = bdrv_get_aio_context(bs);
3182     aio_context_acquire(aio_context);
3183 
3184     if (has_base && has_base_node) {
3185         error_setg(errp, "'base' and 'base-node' cannot be specified "
3186                    "at the same time");
3187         goto out;
3188     }
3189 
3190     if (has_base) {
3191         base_bs = bdrv_find_backing_image(bs, base);
3192         if (base_bs == NULL) {
3193             error_setg(errp, QERR_BASE_NOT_FOUND, base);
3194             goto out;
3195         }
3196         assert(bdrv_get_aio_context(base_bs) == aio_context);
3197         base_name = base;
3198     }
3199 
3200     if (has_base_node) {
3201         base_bs = bdrv_lookup_bs(NULL, base_node, errp);
3202         if (!base_bs) {
3203             goto out;
3204         }
3205         if (bs == base_bs || !bdrv_chain_contains(bs, base_bs)) {
3206             error_setg(errp, "Node '%s' is not a backing image of '%s'",
3207                        base_node, device);
3208             goto out;
3209         }
3210         assert(bdrv_get_aio_context(base_bs) == aio_context);
3211         bdrv_refresh_filename(base_bs);
3212         base_name = base_bs->filename;
3213     }
3214 
3215     /* Check for op blockers in the whole chain between bs and base */
3216     for (iter = bs; iter && iter != base_bs; iter = backing_bs(iter)) {
3217         if (bdrv_op_is_blocked(iter, BLOCK_OP_TYPE_STREAM, errp)) {
3218             goto out;
3219         }
3220     }
3221 
3222     /* if we are streaming the entire chain, the result will have no backing
3223      * file, and specifying one is therefore an error */
3224     if (base_bs == NULL && has_backing_file) {
3225         error_setg(errp, "backing file specified, but streaming the "
3226                          "entire chain");
3227         goto out;
3228     }
3229 
3230     /* backing_file string overrides base bs filename */
3231     base_name = has_backing_file ? backing_file : base_name;
3232 
3233     if (has_auto_finalize && !auto_finalize) {
3234         job_flags |= JOB_MANUAL_FINALIZE;
3235     }
3236     if (has_auto_dismiss && !auto_dismiss) {
3237         job_flags |= JOB_MANUAL_DISMISS;
3238     }
3239 
3240     stream_start(has_job_id ? job_id : NULL, bs, base_bs, base_name,
3241                  job_flags, has_speed ? speed : 0, on_error, &local_err);
3242     if (local_err) {
3243         error_propagate(errp, local_err);
3244         goto out;
3245     }
3246 
3247     trace_qmp_block_stream(bs, bs->job);
3248 
3249 out:
3250     aio_context_release(aio_context);
3251 }
3252 
3253 void qmp_block_commit(bool has_job_id, const char *job_id, const char *device,
3254                       bool has_base_node, const char *base_node,
3255                       bool has_base, const char *base,
3256                       bool has_top_node, const char *top_node,
3257                       bool has_top, const char *top,
3258                       bool has_backing_file, const char *backing_file,
3259                       bool has_speed, int64_t speed,
3260                       bool has_filter_node_name, const char *filter_node_name,
3261                       bool has_auto_finalize, bool auto_finalize,
3262                       bool has_auto_dismiss, bool auto_dismiss,
3263                       Error **errp)
3264 {
3265     BlockDriverState *bs;
3266     BlockDriverState *iter;
3267     BlockDriverState *base_bs, *top_bs;
3268     AioContext *aio_context;
3269     Error *local_err = NULL;
3270     /* This will be part of the QMP command, if/when the
3271      * BlockdevOnError change for blkmirror makes it in
3272      */
3273     BlockdevOnError on_error = BLOCKDEV_ON_ERROR_REPORT;
3274     int job_flags = JOB_DEFAULT;
3275 
3276     if (!has_speed) {
3277         speed = 0;
3278     }
3279     if (!has_filter_node_name) {
3280         filter_node_name = NULL;
3281     }
3282     if (has_auto_finalize && !auto_finalize) {
3283         job_flags |= JOB_MANUAL_FINALIZE;
3284     }
3285     if (has_auto_dismiss && !auto_dismiss) {
3286         job_flags |= JOB_MANUAL_DISMISS;
3287     }
3288 
3289     /* Important Note:
3290      *  libvirt relies on the DeviceNotFound error class in order to probe for
3291      *  live commit feature versions; for this to work, we must make sure to
3292      *  perform the device lookup before any generic errors that may occur in a
3293      *  scenario in which all optional arguments are omitted. */
3294     bs = qmp_get_root_bs(device, &local_err);
3295     if (!bs) {
3296         bs = bdrv_lookup_bs(device, device, NULL);
3297         if (!bs) {
3298             error_free(local_err);
3299             error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
3300                       "Device '%s' not found", device);
3301         } else {
3302             error_propagate(errp, local_err);
3303         }
3304         return;
3305     }
3306 
3307     aio_context = bdrv_get_aio_context(bs);
3308     aio_context_acquire(aio_context);
3309 
3310     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_COMMIT_SOURCE, errp)) {
3311         goto out;
3312     }
3313 
3314     /* default top_bs is the active layer */
3315     top_bs = bs;
3316 
3317     if (has_top_node && has_top) {
3318         error_setg(errp, "'top-node' and 'top' are mutually exclusive");
3319         goto out;
3320     } else if (has_top_node) {
3321         top_bs = bdrv_lookup_bs(NULL, top_node, errp);
3322         if (top_bs == NULL) {
3323             goto out;
3324         }
3325         if (!bdrv_chain_contains(bs, top_bs)) {
3326             error_setg(errp, "'%s' is not in this backing file chain",
3327                        top_node);
3328             goto out;
3329         }
3330     } else if (has_top && top) {
3331         /* This strcmp() is just a shortcut, there is no need to
3332          * refresh @bs's filename.  If it mismatches,
3333          * bdrv_find_backing_image() will do the refresh and may still
3334          * return @bs. */
3335         if (strcmp(bs->filename, top) != 0) {
3336             top_bs = bdrv_find_backing_image(bs, top);
3337         }
3338     }
3339 
3340     if (top_bs == NULL) {
3341         error_setg(errp, "Top image file %s not found", top ? top : "NULL");
3342         goto out;
3343     }
3344 
3345     assert(bdrv_get_aio_context(top_bs) == aio_context);
3346 
3347     if (has_base_node && has_base) {
3348         error_setg(errp, "'base-node' and 'base' are mutually exclusive");
3349         goto out;
3350     } else if (has_base_node) {
3351         base_bs = bdrv_lookup_bs(NULL, base_node, errp);
3352         if (base_bs == NULL) {
3353             goto out;
3354         }
3355         if (!bdrv_chain_contains(top_bs, base_bs)) {
3356             error_setg(errp, "'%s' is not in this backing file chain",
3357                        base_node);
3358             goto out;
3359         }
3360     } else if (has_base && base) {
3361         base_bs = bdrv_find_backing_image(top_bs, base);
3362     } else {
3363         base_bs = bdrv_find_base(top_bs);
3364     }
3365 
3366     if (base_bs == NULL) {
3367         error_setg(errp, QERR_BASE_NOT_FOUND, base ? base : "NULL");
3368         goto out;
3369     }
3370 
3371     assert(bdrv_get_aio_context(base_bs) == aio_context);
3372 
3373     for (iter = top_bs; iter != backing_bs(base_bs); iter = backing_bs(iter)) {
3374         if (bdrv_op_is_blocked(iter, BLOCK_OP_TYPE_COMMIT_TARGET, errp)) {
3375             goto out;
3376         }
3377     }
3378 
3379     /* Do not allow attempts to commit an image into itself */
3380     if (top_bs == base_bs) {
3381         error_setg(errp, "cannot commit an image into itself");
3382         goto out;
3383     }
3384 
3385     if (top_bs == bs) {
3386         if (has_backing_file) {
3387             error_setg(errp, "'backing-file' specified,"
3388                              " but 'top' is the active layer");
3389             goto out;
3390         }
3391         commit_active_start(has_job_id ? job_id : NULL, bs, base_bs,
3392                             job_flags, speed, on_error,
3393                             filter_node_name, NULL, NULL, false, &local_err);
3394     } else {
3395         BlockDriverState *overlay_bs = bdrv_find_overlay(bs, top_bs);
3396         if (bdrv_op_is_blocked(overlay_bs, BLOCK_OP_TYPE_COMMIT_TARGET, errp)) {
3397             goto out;
3398         }
3399         commit_start(has_job_id ? job_id : NULL, bs, base_bs, top_bs, job_flags,
3400                      speed, on_error, has_backing_file ? backing_file : NULL,
3401                      filter_node_name, &local_err);
3402     }
3403     if (local_err != NULL) {
3404         error_propagate(errp, local_err);
3405         goto out;
3406     }
3407 
3408 out:
3409     aio_context_release(aio_context);
3410 }
3411 
3412 static BlockJob *do_drive_backup(DriveBackup *backup, JobTxn *txn,
3413                                  Error **errp)
3414 {
3415     BlockDriverState *bs;
3416     BlockDriverState *target_bs;
3417     BlockDriverState *source = NULL;
3418     BlockJob *job = NULL;
3419     BdrvDirtyBitmap *bmap = NULL;
3420     AioContext *aio_context;
3421     QDict *options = NULL;
3422     Error *local_err = NULL;
3423     int flags, job_flags = JOB_DEFAULT;
3424     int64_t size;
3425     bool set_backing_hd = false;
3426 
3427     if (!backup->has_speed) {
3428         backup->speed = 0;
3429     }
3430     if (!backup->has_on_source_error) {
3431         backup->on_source_error = BLOCKDEV_ON_ERROR_REPORT;
3432     }
3433     if (!backup->has_on_target_error) {
3434         backup->on_target_error = BLOCKDEV_ON_ERROR_REPORT;
3435     }
3436     if (!backup->has_mode) {
3437         backup->mode = NEW_IMAGE_MODE_ABSOLUTE_PATHS;
3438     }
3439     if (!backup->has_job_id) {
3440         backup->job_id = NULL;
3441     }
3442     if (!backup->has_auto_finalize) {
3443         backup->auto_finalize = true;
3444     }
3445     if (!backup->has_auto_dismiss) {
3446         backup->auto_dismiss = true;
3447     }
3448     if (!backup->has_compress) {
3449         backup->compress = false;
3450     }
3451 
3452     bs = qmp_get_root_bs(backup->device, errp);
3453     if (!bs) {
3454         return NULL;
3455     }
3456 
3457     aio_context = bdrv_get_aio_context(bs);
3458     aio_context_acquire(aio_context);
3459 
3460     if (!backup->has_format) {
3461         backup->format = backup->mode == NEW_IMAGE_MODE_EXISTING ?
3462                          NULL : (char*) bs->drv->format_name;
3463     }
3464 
3465     /* Early check to avoid creating target */
3466     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_BACKUP_SOURCE, errp)) {
3467         goto out;
3468     }
3469 
3470     flags = bs->open_flags | BDRV_O_RDWR;
3471 
3472     /* See if we have a backing HD we can use to create our new image
3473      * on top of. */
3474     if (backup->sync == MIRROR_SYNC_MODE_TOP) {
3475         source = backing_bs(bs);
3476         if (!source) {
3477             backup->sync = MIRROR_SYNC_MODE_FULL;
3478         }
3479     }
3480     if (backup->sync == MIRROR_SYNC_MODE_NONE) {
3481         source = bs;
3482         flags |= BDRV_O_NO_BACKING;
3483         set_backing_hd = true;
3484     }
3485 
3486     size = bdrv_getlength(bs);
3487     if (size < 0) {
3488         error_setg_errno(errp, -size, "bdrv_getlength failed");
3489         goto out;
3490     }
3491 
3492     if (backup->mode != NEW_IMAGE_MODE_EXISTING) {
3493         assert(backup->format);
3494         if (source) {
3495             bdrv_refresh_filename(source);
3496             bdrv_img_create(backup->target, backup->format, source->filename,
3497                             source->drv->format_name, NULL,
3498                             size, flags, false, &local_err);
3499         } else {
3500             bdrv_img_create(backup->target, backup->format, NULL, NULL, NULL,
3501                             size, flags, false, &local_err);
3502         }
3503     }
3504 
3505     if (local_err) {
3506         error_propagate(errp, local_err);
3507         goto out;
3508     }
3509 
3510     if (backup->format) {
3511         if (!options) {
3512             options = qdict_new();
3513         }
3514         qdict_put_str(options, "driver", backup->format);
3515     }
3516 
3517     target_bs = bdrv_open(backup->target, NULL, options, flags, errp);
3518     if (!target_bs) {
3519         goto out;
3520     }
3521 
3522     bdrv_set_aio_context(target_bs, aio_context);
3523 
3524     if (set_backing_hd) {
3525         bdrv_set_backing_hd(target_bs, source, &local_err);
3526         if (local_err) {
3527             bdrv_unref(target_bs);
3528             goto out;
3529         }
3530     }
3531 
3532     if (backup->has_bitmap) {
3533         bmap = bdrv_find_dirty_bitmap(bs, backup->bitmap);
3534         if (!bmap) {
3535             error_setg(errp, "Bitmap '%s' could not be found", backup->bitmap);
3536             bdrv_unref(target_bs);
3537             goto out;
3538         }
3539         if (bdrv_dirty_bitmap_check(bmap, BDRV_BITMAP_DEFAULT, errp)) {
3540             goto out;
3541         }
3542     }
3543     if (!backup->auto_finalize) {
3544         job_flags |= JOB_MANUAL_FINALIZE;
3545     }
3546     if (!backup->auto_dismiss) {
3547         job_flags |= JOB_MANUAL_DISMISS;
3548     }
3549 
3550     job = backup_job_create(backup->job_id, bs, target_bs, backup->speed,
3551                             backup->sync, bmap, backup->compress,
3552                             backup->on_source_error, backup->on_target_error,
3553                             job_flags, NULL, NULL, txn, &local_err);
3554     bdrv_unref(target_bs);
3555     if (local_err != NULL) {
3556         error_propagate(errp, local_err);
3557         goto out;
3558     }
3559 
3560 out:
3561     aio_context_release(aio_context);
3562     return job;
3563 }
3564 
3565 void qmp_drive_backup(DriveBackup *arg, Error **errp)
3566 {
3567 
3568     BlockJob *job;
3569     job = do_drive_backup(arg, NULL, errp);
3570     if (job) {
3571         job_start(&job->job);
3572     }
3573 }
3574 
3575 BlockDeviceInfoList *qmp_query_named_block_nodes(Error **errp)
3576 {
3577     return bdrv_named_nodes_list(errp);
3578 }
3579 
3580 XDbgBlockGraph *qmp_x_debug_query_block_graph(Error **errp)
3581 {
3582     return bdrv_get_xdbg_block_graph(errp);
3583 }
3584 
3585 BlockJob *do_blockdev_backup(BlockdevBackup *backup, JobTxn *txn,
3586                              Error **errp)
3587 {
3588     BlockDriverState *bs;
3589     BlockDriverState *target_bs;
3590     Error *local_err = NULL;
3591     BdrvDirtyBitmap *bmap = NULL;
3592     AioContext *aio_context;
3593     BlockJob *job = NULL;
3594     int job_flags = JOB_DEFAULT;
3595 
3596     if (!backup->has_speed) {
3597         backup->speed = 0;
3598     }
3599     if (!backup->has_on_source_error) {
3600         backup->on_source_error = BLOCKDEV_ON_ERROR_REPORT;
3601     }
3602     if (!backup->has_on_target_error) {
3603         backup->on_target_error = BLOCKDEV_ON_ERROR_REPORT;
3604     }
3605     if (!backup->has_job_id) {
3606         backup->job_id = NULL;
3607     }
3608     if (!backup->has_auto_finalize) {
3609         backup->auto_finalize = true;
3610     }
3611     if (!backup->has_auto_dismiss) {
3612         backup->auto_dismiss = true;
3613     }
3614     if (!backup->has_compress) {
3615         backup->compress = false;
3616     }
3617 
3618     bs = bdrv_lookup_bs(backup->device, backup->device, errp);
3619     if (!bs) {
3620         return NULL;
3621     }
3622 
3623     aio_context = bdrv_get_aio_context(bs);
3624     aio_context_acquire(aio_context);
3625 
3626     target_bs = bdrv_lookup_bs(backup->target, backup->target, errp);
3627     if (!target_bs) {
3628         goto out;
3629     }
3630 
3631     if (bdrv_get_aio_context(target_bs) != aio_context) {
3632         if (!bdrv_has_blk(target_bs)) {
3633             /* The target BDS is not attached, we can safely move it to another
3634              * AioContext. */
3635             bdrv_set_aio_context(target_bs, aio_context);
3636         } else {
3637             error_setg(errp, "Target is attached to a different thread from "
3638                              "source.");
3639             goto out;
3640         }
3641     }
3642 
3643     if (backup->has_bitmap) {
3644         bmap = bdrv_find_dirty_bitmap(bs, backup->bitmap);
3645         if (!bmap) {
3646             error_setg(errp, "Bitmap '%s' could not be found", backup->bitmap);
3647             goto out;
3648         }
3649         if (bdrv_dirty_bitmap_check(bmap, BDRV_BITMAP_DEFAULT, errp)) {
3650             goto out;
3651         }
3652     }
3653 
3654     if (!backup->auto_finalize) {
3655         job_flags |= JOB_MANUAL_FINALIZE;
3656     }
3657     if (!backup->auto_dismiss) {
3658         job_flags |= JOB_MANUAL_DISMISS;
3659     }
3660     job = backup_job_create(backup->job_id, bs, target_bs, backup->speed,
3661                             backup->sync, bmap, backup->compress,
3662                             backup->on_source_error, backup->on_target_error,
3663                             job_flags, NULL, NULL, txn, &local_err);
3664     if (local_err != NULL) {
3665         error_propagate(errp, local_err);
3666     }
3667 out:
3668     aio_context_release(aio_context);
3669     return job;
3670 }
3671 
3672 void qmp_blockdev_backup(BlockdevBackup *arg, Error **errp)
3673 {
3674     BlockJob *job;
3675     job = do_blockdev_backup(arg, NULL, errp);
3676     if (job) {
3677         job_start(&job->job);
3678     }
3679 }
3680 
3681 /* Parameter check and block job starting for drive mirroring.
3682  * Caller should hold @device and @target's aio context (must be the same).
3683  **/
3684 static void blockdev_mirror_common(const char *job_id, BlockDriverState *bs,
3685                                    BlockDriverState *target,
3686                                    bool has_replaces, const char *replaces,
3687                                    enum MirrorSyncMode sync,
3688                                    BlockMirrorBackingMode backing_mode,
3689                                    bool has_speed, int64_t speed,
3690                                    bool has_granularity, uint32_t granularity,
3691                                    bool has_buf_size, int64_t buf_size,
3692                                    bool has_on_source_error,
3693                                    BlockdevOnError on_source_error,
3694                                    bool has_on_target_error,
3695                                    BlockdevOnError on_target_error,
3696                                    bool has_unmap, bool unmap,
3697                                    bool has_filter_node_name,
3698                                    const char *filter_node_name,
3699                                    bool has_copy_mode, MirrorCopyMode copy_mode,
3700                                    bool has_auto_finalize, bool auto_finalize,
3701                                    bool has_auto_dismiss, bool auto_dismiss,
3702                                    Error **errp)
3703 {
3704     int job_flags = JOB_DEFAULT;
3705 
3706     if (!has_speed) {
3707         speed = 0;
3708     }
3709     if (!has_on_source_error) {
3710         on_source_error = BLOCKDEV_ON_ERROR_REPORT;
3711     }
3712     if (!has_on_target_error) {
3713         on_target_error = BLOCKDEV_ON_ERROR_REPORT;
3714     }
3715     if (!has_granularity) {
3716         granularity = 0;
3717     }
3718     if (!has_buf_size) {
3719         buf_size = 0;
3720     }
3721     if (!has_unmap) {
3722         unmap = true;
3723     }
3724     if (!has_filter_node_name) {
3725         filter_node_name = NULL;
3726     }
3727     if (!has_copy_mode) {
3728         copy_mode = MIRROR_COPY_MODE_BACKGROUND;
3729     }
3730     if (has_auto_finalize && !auto_finalize) {
3731         job_flags |= JOB_MANUAL_FINALIZE;
3732     }
3733     if (has_auto_dismiss && !auto_dismiss) {
3734         job_flags |= JOB_MANUAL_DISMISS;
3735     }
3736 
3737     if (granularity != 0 && (granularity < 512 || granularity > 1048576 * 64)) {
3738         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "granularity",
3739                    "a value in range [512B, 64MB]");
3740         return;
3741     }
3742     if (granularity & (granularity - 1)) {
3743         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "granularity",
3744                    "power of 2");
3745         return;
3746     }
3747 
3748     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_MIRROR_SOURCE, errp)) {
3749         return;
3750     }
3751     if (bdrv_op_is_blocked(target, BLOCK_OP_TYPE_MIRROR_TARGET, errp)) {
3752         return;
3753     }
3754 
3755     if (!bs->backing && sync == MIRROR_SYNC_MODE_TOP) {
3756         sync = MIRROR_SYNC_MODE_FULL;
3757     }
3758 
3759     if (has_replaces) {
3760         BlockDriverState *to_replace_bs;
3761         AioContext *replace_aio_context;
3762         int64_t bs_size, replace_size;
3763 
3764         bs_size = bdrv_getlength(bs);
3765         if (bs_size < 0) {
3766             error_setg_errno(errp, -bs_size, "Failed to query device's size");
3767             return;
3768         }
3769 
3770         to_replace_bs = check_to_replace_node(bs, replaces, errp);
3771         if (!to_replace_bs) {
3772             return;
3773         }
3774 
3775         replace_aio_context = bdrv_get_aio_context(to_replace_bs);
3776         aio_context_acquire(replace_aio_context);
3777         replace_size = bdrv_getlength(to_replace_bs);
3778         aio_context_release(replace_aio_context);
3779 
3780         if (replace_size < 0) {
3781             error_setg_errno(errp, -replace_size,
3782                              "Failed to query the replacement node's size");
3783             return;
3784         }
3785         if (bs_size != replace_size) {
3786             error_setg(errp, "cannot replace image with a mirror image of "
3787                              "different size");
3788             return;
3789         }
3790     }
3791 
3792     /* pass the node name to replace to mirror start since it's loose coupling
3793      * and will allow to check whether the node still exist at mirror completion
3794      */
3795     mirror_start(job_id, bs, target,
3796                  has_replaces ? replaces : NULL, job_flags,
3797                  speed, granularity, buf_size, sync, backing_mode,
3798                  on_source_error, on_target_error, unmap, filter_node_name,
3799                  copy_mode, errp);
3800 }
3801 
3802 void qmp_drive_mirror(DriveMirror *arg, Error **errp)
3803 {
3804     BlockDriverState *bs;
3805     BlockDriverState *source, *target_bs;
3806     AioContext *aio_context;
3807     BlockMirrorBackingMode backing_mode;
3808     Error *local_err = NULL;
3809     QDict *options = NULL;
3810     int flags;
3811     int64_t size;
3812     const char *format = arg->format;
3813 
3814     bs = qmp_get_root_bs(arg->device, errp);
3815     if (!bs) {
3816         return;
3817     }
3818 
3819     /* Early check to avoid creating target */
3820     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_MIRROR_SOURCE, errp)) {
3821         return;
3822     }
3823 
3824     aio_context = bdrv_get_aio_context(bs);
3825     aio_context_acquire(aio_context);
3826 
3827     if (!arg->has_mode) {
3828         arg->mode = NEW_IMAGE_MODE_ABSOLUTE_PATHS;
3829     }
3830 
3831     if (!arg->has_format) {
3832         format = (arg->mode == NEW_IMAGE_MODE_EXISTING
3833                   ? NULL : bs->drv->format_name);
3834     }
3835 
3836     flags = bs->open_flags | BDRV_O_RDWR;
3837     source = backing_bs(bs);
3838     if (!source && arg->sync == MIRROR_SYNC_MODE_TOP) {
3839         arg->sync = MIRROR_SYNC_MODE_FULL;
3840     }
3841     if (arg->sync == MIRROR_SYNC_MODE_NONE) {
3842         source = bs;
3843     }
3844 
3845     size = bdrv_getlength(bs);
3846     if (size < 0) {
3847         error_setg_errno(errp, -size, "bdrv_getlength failed");
3848         goto out;
3849     }
3850 
3851     if (arg->has_replaces) {
3852         if (!arg->has_node_name) {
3853             error_setg(errp, "a node-name must be provided when replacing a"
3854                              " named node of the graph");
3855             goto out;
3856         }
3857     }
3858 
3859     if (arg->mode == NEW_IMAGE_MODE_ABSOLUTE_PATHS) {
3860         backing_mode = MIRROR_SOURCE_BACKING_CHAIN;
3861     } else {
3862         backing_mode = MIRROR_OPEN_BACKING_CHAIN;
3863     }
3864 
3865     /* Don't open backing image in create() */
3866     flags |= BDRV_O_NO_BACKING;
3867 
3868     if ((arg->sync == MIRROR_SYNC_MODE_FULL || !source)
3869         && arg->mode != NEW_IMAGE_MODE_EXISTING)
3870     {
3871         /* create new image w/o backing file */
3872         assert(format);
3873         bdrv_img_create(arg->target, format,
3874                         NULL, NULL, NULL, size, flags, false, &local_err);
3875     } else {
3876         switch (arg->mode) {
3877         case NEW_IMAGE_MODE_EXISTING:
3878             break;
3879         case NEW_IMAGE_MODE_ABSOLUTE_PATHS:
3880             /* create new image with backing file */
3881             bdrv_refresh_filename(source);
3882             bdrv_img_create(arg->target, format,
3883                             source->filename,
3884                             source->drv->format_name,
3885                             NULL, size, flags, false, &local_err);
3886             break;
3887         default:
3888             abort();
3889         }
3890     }
3891 
3892     if (local_err) {
3893         error_propagate(errp, local_err);
3894         goto out;
3895     }
3896 
3897     options = qdict_new();
3898     if (arg->has_node_name) {
3899         qdict_put_str(options, "node-name", arg->node_name);
3900     }
3901     if (format) {
3902         qdict_put_str(options, "driver", format);
3903     }
3904 
3905     /* Mirroring takes care of copy-on-write using the source's backing
3906      * file.
3907      */
3908     target_bs = bdrv_open(arg->target, NULL, options, flags, errp);
3909     if (!target_bs) {
3910         goto out;
3911     }
3912 
3913     bdrv_set_aio_context(target_bs, aio_context);
3914 
3915     blockdev_mirror_common(arg->has_job_id ? arg->job_id : NULL, bs, target_bs,
3916                            arg->has_replaces, arg->replaces, arg->sync,
3917                            backing_mode, arg->has_speed, arg->speed,
3918                            arg->has_granularity, arg->granularity,
3919                            arg->has_buf_size, arg->buf_size,
3920                            arg->has_on_source_error, arg->on_source_error,
3921                            arg->has_on_target_error, arg->on_target_error,
3922                            arg->has_unmap, arg->unmap,
3923                            false, NULL,
3924                            arg->has_copy_mode, arg->copy_mode,
3925                            arg->has_auto_finalize, arg->auto_finalize,
3926                            arg->has_auto_dismiss, arg->auto_dismiss,
3927                            &local_err);
3928     bdrv_unref(target_bs);
3929     error_propagate(errp, local_err);
3930 out:
3931     aio_context_release(aio_context);
3932 }
3933 
3934 void qmp_blockdev_mirror(bool has_job_id, const char *job_id,
3935                          const char *device, const char *target,
3936                          bool has_replaces, const char *replaces,
3937                          MirrorSyncMode sync,
3938                          bool has_speed, int64_t speed,
3939                          bool has_granularity, uint32_t granularity,
3940                          bool has_buf_size, int64_t buf_size,
3941                          bool has_on_source_error,
3942                          BlockdevOnError on_source_error,
3943                          bool has_on_target_error,
3944                          BlockdevOnError on_target_error,
3945                          bool has_filter_node_name,
3946                          const char *filter_node_name,
3947                          bool has_copy_mode, MirrorCopyMode copy_mode,
3948                          bool has_auto_finalize, bool auto_finalize,
3949                          bool has_auto_dismiss, bool auto_dismiss,
3950                          Error **errp)
3951 {
3952     BlockDriverState *bs;
3953     BlockDriverState *target_bs;
3954     AioContext *aio_context;
3955     BlockMirrorBackingMode backing_mode = MIRROR_LEAVE_BACKING_CHAIN;
3956     Error *local_err = NULL;
3957 
3958     bs = qmp_get_root_bs(device, errp);
3959     if (!bs) {
3960         return;
3961     }
3962 
3963     target_bs = bdrv_lookup_bs(target, target, errp);
3964     if (!target_bs) {
3965         return;
3966     }
3967 
3968     aio_context = bdrv_get_aio_context(bs);
3969     aio_context_acquire(aio_context);
3970 
3971     bdrv_set_aio_context(target_bs, aio_context);
3972 
3973     blockdev_mirror_common(has_job_id ? job_id : NULL, bs, target_bs,
3974                            has_replaces, replaces, sync, backing_mode,
3975                            has_speed, speed,
3976                            has_granularity, granularity,
3977                            has_buf_size, buf_size,
3978                            has_on_source_error, on_source_error,
3979                            has_on_target_error, on_target_error,
3980                            true, true,
3981                            has_filter_node_name, filter_node_name,
3982                            has_copy_mode, copy_mode,
3983                            has_auto_finalize, auto_finalize,
3984                            has_auto_dismiss, auto_dismiss,
3985                            &local_err);
3986     error_propagate(errp, local_err);
3987 
3988     aio_context_release(aio_context);
3989 }
3990 
3991 /* Get a block job using its ID and acquire its AioContext */
3992 static BlockJob *find_block_job(const char *id, AioContext **aio_context,
3993                                 Error **errp)
3994 {
3995     BlockJob *job;
3996 
3997     assert(id != NULL);
3998 
3999     *aio_context = NULL;
4000 
4001     job = block_job_get(id);
4002 
4003     if (!job) {
4004         error_set(errp, ERROR_CLASS_DEVICE_NOT_ACTIVE,
4005                   "Block job '%s' not found", id);
4006         return NULL;
4007     }
4008 
4009     *aio_context = blk_get_aio_context(job->blk);
4010     aio_context_acquire(*aio_context);
4011 
4012     return job;
4013 }
4014 
4015 void qmp_block_job_set_speed(const char *device, int64_t speed, Error **errp)
4016 {
4017     AioContext *aio_context;
4018     BlockJob *job = find_block_job(device, &aio_context, errp);
4019 
4020     if (!job) {
4021         return;
4022     }
4023 
4024     block_job_set_speed(job, speed, errp);
4025     aio_context_release(aio_context);
4026 }
4027 
4028 void qmp_block_job_cancel(const char *device,
4029                           bool has_force, bool force, Error **errp)
4030 {
4031     AioContext *aio_context;
4032     BlockJob *job = find_block_job(device, &aio_context, errp);
4033 
4034     if (!job) {
4035         return;
4036     }
4037 
4038     if (!has_force) {
4039         force = false;
4040     }
4041 
4042     if (job_user_paused(&job->job) && !force) {
4043         error_setg(errp, "The block job for device '%s' is currently paused",
4044                    device);
4045         goto out;
4046     }
4047 
4048     trace_qmp_block_job_cancel(job);
4049     job_user_cancel(&job->job, force, errp);
4050 out:
4051     aio_context_release(aio_context);
4052 }
4053 
4054 void qmp_block_job_pause(const char *device, Error **errp)
4055 {
4056     AioContext *aio_context;
4057     BlockJob *job = find_block_job(device, &aio_context, errp);
4058 
4059     if (!job) {
4060         return;
4061     }
4062 
4063     trace_qmp_block_job_pause(job);
4064     job_user_pause(&job->job, errp);
4065     aio_context_release(aio_context);
4066 }
4067 
4068 void qmp_block_job_resume(const char *device, Error **errp)
4069 {
4070     AioContext *aio_context;
4071     BlockJob *job = find_block_job(device, &aio_context, errp);
4072 
4073     if (!job) {
4074         return;
4075     }
4076 
4077     trace_qmp_block_job_resume(job);
4078     job_user_resume(&job->job, errp);
4079     aio_context_release(aio_context);
4080 }
4081 
4082 void qmp_block_job_complete(const char *device, Error **errp)
4083 {
4084     AioContext *aio_context;
4085     BlockJob *job = find_block_job(device, &aio_context, errp);
4086 
4087     if (!job) {
4088         return;
4089     }
4090 
4091     trace_qmp_block_job_complete(job);
4092     job_complete(&job->job, errp);
4093     aio_context_release(aio_context);
4094 }
4095 
4096 void qmp_block_job_finalize(const char *id, Error **errp)
4097 {
4098     AioContext *aio_context;
4099     BlockJob *job = find_block_job(id, &aio_context, errp);
4100 
4101     if (!job) {
4102         return;
4103     }
4104 
4105     trace_qmp_block_job_finalize(job);
4106     job_finalize(&job->job, errp);
4107     aio_context_release(aio_context);
4108 }
4109 
4110 void qmp_block_job_dismiss(const char *id, Error **errp)
4111 {
4112     AioContext *aio_context;
4113     BlockJob *bjob = find_block_job(id, &aio_context, errp);
4114     Job *job;
4115 
4116     if (!bjob) {
4117         return;
4118     }
4119 
4120     trace_qmp_block_job_dismiss(bjob);
4121     job = &bjob->job;
4122     job_dismiss(&job, errp);
4123     aio_context_release(aio_context);
4124 }
4125 
4126 void qmp_change_backing_file(const char *device,
4127                              const char *image_node_name,
4128                              const char *backing_file,
4129                              Error **errp)
4130 {
4131     BlockDriverState *bs = NULL;
4132     AioContext *aio_context;
4133     BlockDriverState *image_bs = NULL;
4134     Error *local_err = NULL;
4135     bool ro;
4136     int ret;
4137 
4138     bs = qmp_get_root_bs(device, errp);
4139     if (!bs) {
4140         return;
4141     }
4142 
4143     aio_context = bdrv_get_aio_context(bs);
4144     aio_context_acquire(aio_context);
4145 
4146     image_bs = bdrv_lookup_bs(NULL, image_node_name, &local_err);
4147     if (local_err) {
4148         error_propagate(errp, local_err);
4149         goto out;
4150     }
4151 
4152     if (!image_bs) {
4153         error_setg(errp, "image file not found");
4154         goto out;
4155     }
4156 
4157     if (bdrv_find_base(image_bs) == image_bs) {
4158         error_setg(errp, "not allowing backing file change on an image "
4159                          "without a backing file");
4160         goto out;
4161     }
4162 
4163     /* even though we are not necessarily operating on bs, we need it to
4164      * determine if block ops are currently prohibited on the chain */
4165     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_CHANGE, errp)) {
4166         goto out;
4167     }
4168 
4169     /* final sanity check */
4170     if (!bdrv_chain_contains(bs, image_bs)) {
4171         error_setg(errp, "'%s' and image file are not in the same chain",
4172                    device);
4173         goto out;
4174     }
4175 
4176     /* if not r/w, reopen to make r/w */
4177     ro = bdrv_is_read_only(image_bs);
4178 
4179     if (ro) {
4180         if (bdrv_reopen_set_read_only(image_bs, false, errp) != 0) {
4181             goto out;
4182         }
4183     }
4184 
4185     ret = bdrv_change_backing_file(image_bs, backing_file,
4186                                image_bs->drv ? image_bs->drv->format_name : "");
4187 
4188     if (ret < 0) {
4189         error_setg_errno(errp, -ret, "Could not change backing file to '%s'",
4190                          backing_file);
4191         /* don't exit here, so we can try to restore open flags if
4192          * appropriate */
4193     }
4194 
4195     if (ro) {
4196         bdrv_reopen_set_read_only(image_bs, true, &local_err);
4197         error_propagate(errp, local_err);
4198     }
4199 
4200 out:
4201     aio_context_release(aio_context);
4202 }
4203 
4204 void hmp_drive_add_node(Monitor *mon, const char *optstr)
4205 {
4206     QemuOpts *opts;
4207     QDict *qdict;
4208     Error *local_err = NULL;
4209 
4210     opts = qemu_opts_parse_noisily(&qemu_drive_opts, optstr, false);
4211     if (!opts) {
4212         return;
4213     }
4214 
4215     qdict = qemu_opts_to_qdict(opts, NULL);
4216 
4217     if (!qdict_get_try_str(qdict, "node-name")) {
4218         qobject_unref(qdict);
4219         error_report("'node-name' needs to be specified");
4220         goto out;
4221     }
4222 
4223     BlockDriverState *bs = bds_tree_init(qdict, &local_err);
4224     if (!bs) {
4225         error_report_err(local_err);
4226         goto out;
4227     }
4228 
4229     QTAILQ_INSERT_TAIL(&monitor_bdrv_states, bs, monitor_list);
4230 
4231 out:
4232     qemu_opts_del(opts);
4233 }
4234 
4235 void qmp_blockdev_add(BlockdevOptions *options, Error **errp)
4236 {
4237     BlockDriverState *bs;
4238     QObject *obj;
4239     Visitor *v = qobject_output_visitor_new(&obj);
4240     QDict *qdict;
4241     Error *local_err = NULL;
4242 
4243     visit_type_BlockdevOptions(v, NULL, &options, &local_err);
4244     if (local_err) {
4245         error_propagate(errp, local_err);
4246         goto fail;
4247     }
4248 
4249     visit_complete(v, &obj);
4250     qdict = qobject_to(QDict, obj);
4251 
4252     qdict_flatten(qdict);
4253 
4254     if (!qdict_get_try_str(qdict, "node-name")) {
4255         error_setg(errp, "'node-name' must be specified for the root node");
4256         goto fail;
4257     }
4258 
4259     bs = bds_tree_init(qdict, errp);
4260     if (!bs) {
4261         goto fail;
4262     }
4263 
4264     QTAILQ_INSERT_TAIL(&monitor_bdrv_states, bs, monitor_list);
4265 
4266 fail:
4267     visit_free(v);
4268 }
4269 
4270 void qmp_x_blockdev_reopen(BlockdevOptions *options, Error **errp)
4271 {
4272     BlockDriverState *bs;
4273     AioContext *ctx;
4274     QObject *obj;
4275     Visitor *v = qobject_output_visitor_new(&obj);
4276     Error *local_err = NULL;
4277     BlockReopenQueue *queue;
4278     QDict *qdict;
4279 
4280     /* Check for the selected node name */
4281     if (!options->has_node_name) {
4282         error_setg(errp, "Node name not specified");
4283         goto fail;
4284     }
4285 
4286     bs = bdrv_find_node(options->node_name);
4287     if (!bs) {
4288         error_setg(errp, "Cannot find node named '%s'", options->node_name);
4289         goto fail;
4290     }
4291 
4292     /* Put all options in a QDict and flatten it */
4293     visit_type_BlockdevOptions(v, NULL, &options, &local_err);
4294     if (local_err) {
4295         error_propagate(errp, local_err);
4296         goto fail;
4297     }
4298 
4299     visit_complete(v, &obj);
4300     qdict = qobject_to(QDict, obj);
4301 
4302     qdict_flatten(qdict);
4303 
4304     /* Perform the reopen operation */
4305     ctx = bdrv_get_aio_context(bs);
4306     aio_context_acquire(ctx);
4307     bdrv_subtree_drained_begin(bs);
4308     queue = bdrv_reopen_queue(NULL, bs, qdict, false);
4309     bdrv_reopen_multiple(queue, errp);
4310     bdrv_subtree_drained_end(bs);
4311     aio_context_release(ctx);
4312 
4313 fail:
4314     visit_free(v);
4315 }
4316 
4317 void qmp_blockdev_del(const char *node_name, Error **errp)
4318 {
4319     AioContext *aio_context;
4320     BlockDriverState *bs;
4321 
4322     bs = bdrv_find_node(node_name);
4323     if (!bs) {
4324         error_setg(errp, "Cannot find node %s", node_name);
4325         return;
4326     }
4327     if (bdrv_has_blk(bs)) {
4328         error_setg(errp, "Node %s is in use", node_name);
4329         return;
4330     }
4331     aio_context = bdrv_get_aio_context(bs);
4332     aio_context_acquire(aio_context);
4333 
4334     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_DRIVE_DEL, errp)) {
4335         goto out;
4336     }
4337 
4338     if (!QTAILQ_IN_USE(bs, monitor_list)) {
4339         error_setg(errp, "Node %s is not owned by the monitor",
4340                    bs->node_name);
4341         goto out;
4342     }
4343 
4344     if (bs->refcnt > 1) {
4345         error_setg(errp, "Block device %s is in use",
4346                    bdrv_get_device_or_node_name(bs));
4347         goto out;
4348     }
4349 
4350     QTAILQ_REMOVE(&monitor_bdrv_states, bs, monitor_list);
4351     bdrv_unref(bs);
4352 
4353 out:
4354     aio_context_release(aio_context);
4355 }
4356 
4357 static BdrvChild *bdrv_find_child(BlockDriverState *parent_bs,
4358                                   const char *child_name)
4359 {
4360     BdrvChild *child;
4361 
4362     QLIST_FOREACH(child, &parent_bs->children, next) {
4363         if (strcmp(child->name, child_name) == 0) {
4364             return child;
4365         }
4366     }
4367 
4368     return NULL;
4369 }
4370 
4371 void qmp_x_blockdev_change(const char *parent, bool has_child,
4372                            const char *child, bool has_node,
4373                            const char *node, Error **errp)
4374 {
4375     BlockDriverState *parent_bs, *new_bs = NULL;
4376     BdrvChild *p_child;
4377 
4378     parent_bs = bdrv_lookup_bs(parent, parent, errp);
4379     if (!parent_bs) {
4380         return;
4381     }
4382 
4383     if (has_child == has_node) {
4384         if (has_child) {
4385             error_setg(errp, "The parameters child and node are in conflict");
4386         } else {
4387             error_setg(errp, "Either child or node must be specified");
4388         }
4389         return;
4390     }
4391 
4392     if (has_child) {
4393         p_child = bdrv_find_child(parent_bs, child);
4394         if (!p_child) {
4395             error_setg(errp, "Node '%s' does not have child '%s'",
4396                        parent, child);
4397             return;
4398         }
4399         bdrv_del_child(parent_bs, p_child, errp);
4400     }
4401 
4402     if (has_node) {
4403         new_bs = bdrv_find_node(node);
4404         if (!new_bs) {
4405             error_setg(errp, "Node '%s' not found", node);
4406             return;
4407         }
4408         bdrv_add_child(parent_bs, new_bs, errp);
4409     }
4410 }
4411 
4412 BlockJobInfoList *qmp_query_block_jobs(Error **errp)
4413 {
4414     BlockJobInfoList *head = NULL, **p_next = &head;
4415     BlockJob *job;
4416 
4417     for (job = block_job_next(NULL); job; job = block_job_next(job)) {
4418         BlockJobInfoList *elem;
4419         AioContext *aio_context;
4420 
4421         if (block_job_is_internal(job)) {
4422             continue;
4423         }
4424         elem = g_new0(BlockJobInfoList, 1);
4425         aio_context = blk_get_aio_context(job->blk);
4426         aio_context_acquire(aio_context);
4427         elem->value = block_job_query(job, errp);
4428         aio_context_release(aio_context);
4429         if (!elem->value) {
4430             g_free(elem);
4431             qapi_free_BlockJobInfoList(head);
4432             return NULL;
4433         }
4434         *p_next = elem;
4435         p_next = &elem->next;
4436     }
4437 
4438     return head;
4439 }
4440 
4441 void qmp_x_blockdev_set_iothread(const char *node_name, StrOrNull *iothread,
4442                                  bool has_force, bool force, Error **errp)
4443 {
4444     AioContext *old_context;
4445     AioContext *new_context;
4446     BlockDriverState *bs;
4447 
4448     bs = bdrv_find_node(node_name);
4449     if (!bs) {
4450         error_setg(errp, "Cannot find node %s", node_name);
4451         return;
4452     }
4453 
4454     /* Protects against accidents. */
4455     if (!(has_force && force) && bdrv_has_blk(bs)) {
4456         error_setg(errp, "Node %s is associated with a BlockBackend and could "
4457                          "be in use (use force=true to override this check)",
4458                          node_name);
4459         return;
4460     }
4461 
4462     if (iothread->type == QTYPE_QSTRING) {
4463         IOThread *obj = iothread_by_id(iothread->u.s);
4464         if (!obj) {
4465             error_setg(errp, "Cannot find iothread %s", iothread->u.s);
4466             return;
4467         }
4468 
4469         new_context = iothread_get_aio_context(obj);
4470     } else {
4471         new_context = qemu_get_aio_context();
4472     }
4473 
4474     old_context = bdrv_get_aio_context(bs);
4475     aio_context_acquire(old_context);
4476 
4477     bdrv_set_aio_context(bs, new_context);
4478 
4479     aio_context_release(old_context);
4480 }
4481 
4482 void qmp_block_latency_histogram_set(
4483     const char *id,
4484     bool has_boundaries, uint64List *boundaries,
4485     bool has_boundaries_read, uint64List *boundaries_read,
4486     bool has_boundaries_write, uint64List *boundaries_write,
4487     bool has_boundaries_flush, uint64List *boundaries_flush,
4488     Error **errp)
4489 {
4490     BlockBackend *blk = qmp_get_blk(NULL, id, errp);
4491     BlockAcctStats *stats;
4492     int ret;
4493 
4494     if (!blk) {
4495         return;
4496     }
4497 
4498     stats = blk_get_stats(blk);
4499 
4500     if (!has_boundaries && !has_boundaries_read && !has_boundaries_write &&
4501         !has_boundaries_flush)
4502     {
4503         block_latency_histograms_clear(stats);
4504         return;
4505     }
4506 
4507     if (has_boundaries || has_boundaries_read) {
4508         ret = block_latency_histogram_set(
4509             stats, BLOCK_ACCT_READ,
4510             has_boundaries_read ? boundaries_read : boundaries);
4511         if (ret) {
4512             error_setg(errp, "Device '%s' set read boundaries fail", id);
4513             return;
4514         }
4515     }
4516 
4517     if (has_boundaries || has_boundaries_write) {
4518         ret = block_latency_histogram_set(
4519             stats, BLOCK_ACCT_WRITE,
4520             has_boundaries_write ? boundaries_write : boundaries);
4521         if (ret) {
4522             error_setg(errp, "Device '%s' set write boundaries fail", id);
4523             return;
4524         }
4525     }
4526 
4527     if (has_boundaries || has_boundaries_flush) {
4528         ret = block_latency_histogram_set(
4529             stats, BLOCK_ACCT_FLUSH,
4530             has_boundaries_flush ? boundaries_flush : boundaries);
4531         if (ret) {
4532             error_setg(errp, "Device '%s' set flush boundaries fail", id);
4533             return;
4534         }
4535     }
4536 }
4537 
4538 QemuOptsList qemu_common_drive_opts = {
4539     .name = "drive",
4540     .head = QTAILQ_HEAD_INITIALIZER(qemu_common_drive_opts.head),
4541     .desc = {
4542         {
4543             .name = "snapshot",
4544             .type = QEMU_OPT_BOOL,
4545             .help = "enable/disable snapshot mode",
4546         },{
4547             .name = "aio",
4548             .type = QEMU_OPT_STRING,
4549             .help = "host AIO implementation (threads, native)",
4550         },{
4551             .name = BDRV_OPT_CACHE_WB,
4552             .type = QEMU_OPT_BOOL,
4553             .help = "Enable writeback mode",
4554         },{
4555             .name = "format",
4556             .type = QEMU_OPT_STRING,
4557             .help = "disk format (raw, qcow2, ...)",
4558         },{
4559             .name = "rerror",
4560             .type = QEMU_OPT_STRING,
4561             .help = "read error action",
4562         },{
4563             .name = "werror",
4564             .type = QEMU_OPT_STRING,
4565             .help = "write error action",
4566         },{
4567             .name = BDRV_OPT_READ_ONLY,
4568             .type = QEMU_OPT_BOOL,
4569             .help = "open drive file as read-only",
4570         },
4571 
4572         THROTTLE_OPTS,
4573 
4574         {
4575             .name = "throttling.group",
4576             .type = QEMU_OPT_STRING,
4577             .help = "name of the block throttling group",
4578         },{
4579             .name = "copy-on-read",
4580             .type = QEMU_OPT_BOOL,
4581             .help = "copy read data from backing file into image file",
4582         },{
4583             .name = "detect-zeroes",
4584             .type = QEMU_OPT_STRING,
4585             .help = "try to optimize zero writes (off, on, unmap)",
4586         },{
4587             .name = "stats-account-invalid",
4588             .type = QEMU_OPT_BOOL,
4589             .help = "whether to account for invalid I/O operations "
4590                     "in the statistics",
4591         },{
4592             .name = "stats-account-failed",
4593             .type = QEMU_OPT_BOOL,
4594             .help = "whether to account for failed I/O operations "
4595                     "in the statistics",
4596         },
4597         { /* end of list */ }
4598     },
4599 };
4600 
4601 QemuOptsList qemu_drive_opts = {
4602     .name = "drive",
4603     .head = QTAILQ_HEAD_INITIALIZER(qemu_drive_opts.head),
4604     .desc = {
4605         /*
4606          * no elements => accept any params
4607          * validation will happen later
4608          */
4609         { /* end of list */ }
4610     },
4611 };
4612