xref: /openbmc/qemu/blockdev.c (revision 06a47ef5)
1 /*
2  * QEMU host block devices
3  *
4  * Copyright (c) 2003-2008 Fabrice Bellard
5  *
6  * This work is licensed under the terms of the GNU GPL, version 2 or
7  * later.  See the COPYING file in the top-level directory.
8  *
9  * This file incorporates work covered by the following copyright and
10  * permission notice:
11  *
12  * Copyright (c) 2003-2008 Fabrice Bellard
13  *
14  * Permission is hereby granted, free of charge, to any person obtaining a copy
15  * of this software and associated documentation files (the "Software"), to deal
16  * in the Software without restriction, including without limitation the rights
17  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
18  * copies of the Software, and to permit persons to whom the Software is
19  * furnished to do so, subject to the following conditions:
20  *
21  * The above copyright notice and this permission notice shall be included in
22  * all copies or substantial portions of the Software.
23  *
24  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
25  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
26  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
27  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
28  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
29  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
30  * THE SOFTWARE.
31  */
32 
33 #include "qemu/osdep.h"
34 #include "sysemu/block-backend.h"
35 #include "sysemu/blockdev.h"
36 #include "hw/block/block.h"
37 #include "block/blockjob.h"
38 #include "block/qdict.h"
39 #include "block/throttle-groups.h"
40 #include "monitor/monitor.h"
41 #include "qemu/error-report.h"
42 #include "qemu/option.h"
43 #include "qemu/qemu-print.h"
44 #include "qemu/config-file.h"
45 #include "qapi/qapi-commands-block.h"
46 #include "qapi/qapi-commands-transaction.h"
47 #include "qapi/qapi-visit-block-core.h"
48 #include "qapi/qmp/qdict.h"
49 #include "qapi/qmp/qnum.h"
50 #include "qapi/qmp/qstring.h"
51 #include "qapi/error.h"
52 #include "qapi/qmp/qerror.h"
53 #include "qapi/qmp/qlist.h"
54 #include "qapi/qobject-output-visitor.h"
55 #include "sysemu/sysemu.h"
56 #include "sysemu/iothread.h"
57 #include "block/block_int.h"
58 #include "block/trace.h"
59 #include "sysemu/arch_init.h"
60 #include "sysemu/qtest.h"
61 #include "sysemu/runstate.h"
62 #include "qemu/cutils.h"
63 #include "qemu/help_option.h"
64 #include "qemu/main-loop.h"
65 #include "qemu/throttle-options.h"
66 
67 static QTAILQ_HEAD(, BlockDriverState) monitor_bdrv_states =
68     QTAILQ_HEAD_INITIALIZER(monitor_bdrv_states);
69 
70 static int do_open_tray(const char *blk_name, const char *qdev_id,
71                         bool force, Error **errp);
72 static void blockdev_remove_medium(bool has_device, const char *device,
73                                    bool has_id, const char *id, Error **errp);
74 static void blockdev_insert_medium(bool has_device, const char *device,
75                                    bool has_id, const char *id,
76                                    const char *node_name, Error **errp);
77 
78 static const char *const if_name[IF_COUNT] = {
79     [IF_NONE] = "none",
80     [IF_IDE] = "ide",
81     [IF_SCSI] = "scsi",
82     [IF_FLOPPY] = "floppy",
83     [IF_PFLASH] = "pflash",
84     [IF_MTD] = "mtd",
85     [IF_SD] = "sd",
86     [IF_VIRTIO] = "virtio",
87     [IF_XEN] = "xen",
88 };
89 
90 static int if_max_devs[IF_COUNT] = {
91     /*
92      * Do not change these numbers!  They govern how drive option
93      * index maps to unit and bus.  That mapping is ABI.
94      *
95      * All controllers used to implement if=T drives need to support
96      * if_max_devs[T] units, for any T with if_max_devs[T] != 0.
97      * Otherwise, some index values map to "impossible" bus, unit
98      * values.
99      *
100      * For instance, if you change [IF_SCSI] to 255, -drive
101      * if=scsi,index=12 no longer means bus=1,unit=5, but
102      * bus=0,unit=12.  With an lsi53c895a controller (7 units max),
103      * the drive can't be set up.  Regression.
104      */
105     [IF_IDE] = 2,
106     [IF_SCSI] = 7,
107 };
108 
109 /**
110  * Boards may call this to offer board-by-board overrides
111  * of the default, global values.
112  */
113 void override_max_devs(BlockInterfaceType type, int max_devs)
114 {
115     BlockBackend *blk;
116     DriveInfo *dinfo;
117 
118     if (max_devs <= 0) {
119         return;
120     }
121 
122     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
123         dinfo = blk_legacy_dinfo(blk);
124         if (dinfo->type == type) {
125             fprintf(stderr, "Cannot override units-per-bus property of"
126                     " the %s interface, because a drive of that type has"
127                     " already been added.\n", if_name[type]);
128             g_assert_not_reached();
129         }
130     }
131 
132     if_max_devs[type] = max_devs;
133 }
134 
135 /*
136  * We automatically delete the drive when a device using it gets
137  * unplugged.  Questionable feature, but we can't just drop it.
138  * Device models call blockdev_mark_auto_del() to schedule the
139  * automatic deletion, and generic qdev code calls blockdev_auto_del()
140  * when deletion is actually safe.
141  */
142 void blockdev_mark_auto_del(BlockBackend *blk)
143 {
144     DriveInfo *dinfo = blk_legacy_dinfo(blk);
145     BlockJob *job;
146 
147     if (!dinfo) {
148         return;
149     }
150 
151     for (job = block_job_next(NULL); job; job = block_job_next(job)) {
152         if (block_job_has_bdrv(job, blk_bs(blk))) {
153             AioContext *aio_context = job->job.aio_context;
154             aio_context_acquire(aio_context);
155 
156             job_cancel(&job->job, false);
157 
158             aio_context_release(aio_context);
159         }
160     }
161 
162     dinfo->auto_del = 1;
163 }
164 
165 void blockdev_auto_del(BlockBackend *blk)
166 {
167     DriveInfo *dinfo = blk_legacy_dinfo(blk);
168 
169     if (dinfo && dinfo->auto_del) {
170         monitor_remove_blk(blk);
171         blk_unref(blk);
172     }
173 }
174 
175 /**
176  * Returns the current mapping of how many units per bus
177  * a particular interface can support.
178  *
179  *  A positive integer indicates n units per bus.
180  *  0 implies the mapping has not been established.
181  * -1 indicates an invalid BlockInterfaceType was given.
182  */
183 int drive_get_max_devs(BlockInterfaceType type)
184 {
185     if (type >= IF_IDE && type < IF_COUNT) {
186         return if_max_devs[type];
187     }
188 
189     return -1;
190 }
191 
192 static int drive_index_to_bus_id(BlockInterfaceType type, int index)
193 {
194     int max_devs = if_max_devs[type];
195     return max_devs ? index / max_devs : 0;
196 }
197 
198 static int drive_index_to_unit_id(BlockInterfaceType type, int index)
199 {
200     int max_devs = if_max_devs[type];
201     return max_devs ? index % max_devs : index;
202 }
203 
204 QemuOpts *drive_def(const char *optstr)
205 {
206     return qemu_opts_parse_noisily(qemu_find_opts("drive"), optstr, false);
207 }
208 
209 QemuOpts *drive_add(BlockInterfaceType type, int index, const char *file,
210                     const char *optstr)
211 {
212     QemuOpts *opts;
213 
214     opts = drive_def(optstr);
215     if (!opts) {
216         return NULL;
217     }
218     if (type != IF_DEFAULT) {
219         qemu_opt_set(opts, "if", if_name[type], &error_abort);
220     }
221     if (index >= 0) {
222         qemu_opt_set_number(opts, "index", index, &error_abort);
223     }
224     if (file)
225         qemu_opt_set(opts, "file", file, &error_abort);
226     return opts;
227 }
228 
229 DriveInfo *drive_get(BlockInterfaceType type, int bus, int unit)
230 {
231     BlockBackend *blk;
232     DriveInfo *dinfo;
233 
234     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
235         dinfo = blk_legacy_dinfo(blk);
236         if (dinfo && dinfo->type == type
237             && dinfo->bus == bus && dinfo->unit == unit) {
238             return dinfo;
239         }
240     }
241 
242     return NULL;
243 }
244 
245 void drive_check_orphaned(void)
246 {
247     BlockBackend *blk;
248     DriveInfo *dinfo;
249     Location loc;
250     bool orphans = false;
251 
252     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
253         dinfo = blk_legacy_dinfo(blk);
254         if (!blk_get_attached_dev(blk) && !dinfo->is_default &&
255             dinfo->type != IF_NONE) {
256             loc_push_none(&loc);
257             qemu_opts_loc_restore(dinfo->opts);
258             error_report("machine type does not support"
259                          " if=%s,bus=%d,unit=%d",
260                          if_name[dinfo->type], dinfo->bus, dinfo->unit);
261             loc_pop(&loc);
262             orphans = true;
263         }
264     }
265 
266     if (orphans) {
267         exit(1);
268     }
269 }
270 
271 DriveInfo *drive_get_by_index(BlockInterfaceType type, int index)
272 {
273     return drive_get(type,
274                      drive_index_to_bus_id(type, index),
275                      drive_index_to_unit_id(type, index));
276 }
277 
278 int drive_get_max_bus(BlockInterfaceType type)
279 {
280     int max_bus;
281     BlockBackend *blk;
282     DriveInfo *dinfo;
283 
284     max_bus = -1;
285     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
286         dinfo = blk_legacy_dinfo(blk);
287         if (dinfo && dinfo->type == type && dinfo->bus > max_bus) {
288             max_bus = dinfo->bus;
289         }
290     }
291     return max_bus;
292 }
293 
294 /* Get a block device.  This should only be used for single-drive devices
295    (e.g. SD/Floppy/MTD).  Multi-disk devices (scsi/ide) should use the
296    appropriate bus.  */
297 DriveInfo *drive_get_next(BlockInterfaceType type)
298 {
299     static int next_block_unit[IF_COUNT];
300 
301     return drive_get(type, 0, next_block_unit[type]++);
302 }
303 
304 static void bdrv_format_print(void *opaque, const char *name)
305 {
306     qemu_printf(" %s", name);
307 }
308 
309 typedef struct {
310     QEMUBH *bh;
311     BlockDriverState *bs;
312 } BDRVPutRefBH;
313 
314 static int parse_block_error_action(const char *buf, bool is_read, Error **errp)
315 {
316     if (!strcmp(buf, "ignore")) {
317         return BLOCKDEV_ON_ERROR_IGNORE;
318     } else if (!is_read && !strcmp(buf, "enospc")) {
319         return BLOCKDEV_ON_ERROR_ENOSPC;
320     } else if (!strcmp(buf, "stop")) {
321         return BLOCKDEV_ON_ERROR_STOP;
322     } else if (!strcmp(buf, "report")) {
323         return BLOCKDEV_ON_ERROR_REPORT;
324     } else {
325         error_setg(errp, "'%s' invalid %s error action",
326                    buf, is_read ? "read" : "write");
327         return -1;
328     }
329 }
330 
331 static bool parse_stats_intervals(BlockAcctStats *stats, QList *intervals,
332                                   Error **errp)
333 {
334     const QListEntry *entry;
335     for (entry = qlist_first(intervals); entry; entry = qlist_next(entry)) {
336         switch (qobject_type(entry->value)) {
337 
338         case QTYPE_QSTRING: {
339             unsigned long long length;
340             const char *str = qstring_get_str(qobject_to(QString,
341                                                          entry->value));
342             if (parse_uint_full(str, &length, 10) == 0 &&
343                 length > 0 && length <= UINT_MAX) {
344                 block_acct_add_interval(stats, (unsigned) length);
345             } else {
346                 error_setg(errp, "Invalid interval length: %s", str);
347                 return false;
348             }
349             break;
350         }
351 
352         case QTYPE_QNUM: {
353             int64_t length = qnum_get_int(qobject_to(QNum, entry->value));
354 
355             if (length > 0 && length <= UINT_MAX) {
356                 block_acct_add_interval(stats, (unsigned) length);
357             } else {
358                 error_setg(errp, "Invalid interval length: %" PRId64, length);
359                 return false;
360             }
361             break;
362         }
363 
364         default:
365             error_setg(errp, "The specification of stats-intervals is invalid");
366             return false;
367         }
368     }
369     return true;
370 }
371 
372 typedef enum { MEDIA_DISK, MEDIA_CDROM } DriveMediaType;
373 
374 /* All parameters but @opts are optional and may be set to NULL. */
375 static void extract_common_blockdev_options(QemuOpts *opts, int *bdrv_flags,
376     const char **throttling_group, ThrottleConfig *throttle_cfg,
377     BlockdevDetectZeroesOptions *detect_zeroes, Error **errp)
378 {
379     Error *local_error = NULL;
380     const char *aio;
381 
382     if (bdrv_flags) {
383         if (qemu_opt_get_bool(opts, "copy-on-read", false)) {
384             *bdrv_flags |= BDRV_O_COPY_ON_READ;
385         }
386 
387         if ((aio = qemu_opt_get(opts, "aio")) != NULL) {
388             if (!strcmp(aio, "native")) {
389                 *bdrv_flags |= BDRV_O_NATIVE_AIO;
390             } else if (!strcmp(aio, "threads")) {
391                 /* this is the default */
392             } else {
393                error_setg(errp, "invalid aio option");
394                return;
395             }
396         }
397     }
398 
399     /* disk I/O throttling */
400     if (throttling_group) {
401         *throttling_group = qemu_opt_get(opts, "throttling.group");
402     }
403 
404     if (throttle_cfg) {
405         throttle_config_init(throttle_cfg);
406         throttle_cfg->buckets[THROTTLE_BPS_TOTAL].avg =
407             qemu_opt_get_number(opts, "throttling.bps-total", 0);
408         throttle_cfg->buckets[THROTTLE_BPS_READ].avg  =
409             qemu_opt_get_number(opts, "throttling.bps-read", 0);
410         throttle_cfg->buckets[THROTTLE_BPS_WRITE].avg =
411             qemu_opt_get_number(opts, "throttling.bps-write", 0);
412         throttle_cfg->buckets[THROTTLE_OPS_TOTAL].avg =
413             qemu_opt_get_number(opts, "throttling.iops-total", 0);
414         throttle_cfg->buckets[THROTTLE_OPS_READ].avg =
415             qemu_opt_get_number(opts, "throttling.iops-read", 0);
416         throttle_cfg->buckets[THROTTLE_OPS_WRITE].avg =
417             qemu_opt_get_number(opts, "throttling.iops-write", 0);
418 
419         throttle_cfg->buckets[THROTTLE_BPS_TOTAL].max =
420             qemu_opt_get_number(opts, "throttling.bps-total-max", 0);
421         throttle_cfg->buckets[THROTTLE_BPS_READ].max  =
422             qemu_opt_get_number(opts, "throttling.bps-read-max", 0);
423         throttle_cfg->buckets[THROTTLE_BPS_WRITE].max =
424             qemu_opt_get_number(opts, "throttling.bps-write-max", 0);
425         throttle_cfg->buckets[THROTTLE_OPS_TOTAL].max =
426             qemu_opt_get_number(opts, "throttling.iops-total-max", 0);
427         throttle_cfg->buckets[THROTTLE_OPS_READ].max =
428             qemu_opt_get_number(opts, "throttling.iops-read-max", 0);
429         throttle_cfg->buckets[THROTTLE_OPS_WRITE].max =
430             qemu_opt_get_number(opts, "throttling.iops-write-max", 0);
431 
432         throttle_cfg->buckets[THROTTLE_BPS_TOTAL].burst_length =
433             qemu_opt_get_number(opts, "throttling.bps-total-max-length", 1);
434         throttle_cfg->buckets[THROTTLE_BPS_READ].burst_length  =
435             qemu_opt_get_number(opts, "throttling.bps-read-max-length", 1);
436         throttle_cfg->buckets[THROTTLE_BPS_WRITE].burst_length =
437             qemu_opt_get_number(opts, "throttling.bps-write-max-length", 1);
438         throttle_cfg->buckets[THROTTLE_OPS_TOTAL].burst_length =
439             qemu_opt_get_number(opts, "throttling.iops-total-max-length", 1);
440         throttle_cfg->buckets[THROTTLE_OPS_READ].burst_length =
441             qemu_opt_get_number(opts, "throttling.iops-read-max-length", 1);
442         throttle_cfg->buckets[THROTTLE_OPS_WRITE].burst_length =
443             qemu_opt_get_number(opts, "throttling.iops-write-max-length", 1);
444 
445         throttle_cfg->op_size =
446             qemu_opt_get_number(opts, "throttling.iops-size", 0);
447 
448         if (!throttle_is_valid(throttle_cfg, errp)) {
449             return;
450         }
451     }
452 
453     if (detect_zeroes) {
454         *detect_zeroes =
455             qapi_enum_parse(&BlockdevDetectZeroesOptions_lookup,
456                             qemu_opt_get(opts, "detect-zeroes"),
457                             BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF,
458                             &local_error);
459         if (local_error) {
460             error_propagate(errp, local_error);
461             return;
462         }
463     }
464 }
465 
466 /* Takes the ownership of bs_opts */
467 static BlockBackend *blockdev_init(const char *file, QDict *bs_opts,
468                                    Error **errp)
469 {
470     const char *buf;
471     int bdrv_flags = 0;
472     int on_read_error, on_write_error;
473     bool account_invalid, account_failed;
474     bool writethrough, read_only;
475     BlockBackend *blk;
476     BlockDriverState *bs;
477     ThrottleConfig cfg;
478     int snapshot = 0;
479     Error *error = NULL;
480     QemuOpts *opts;
481     QDict *interval_dict = NULL;
482     QList *interval_list = NULL;
483     const char *id;
484     BlockdevDetectZeroesOptions detect_zeroes =
485         BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF;
486     const char *throttling_group = NULL;
487 
488     /* Check common options by copying from bs_opts to opts, all other options
489      * stay in bs_opts for processing by bdrv_open(). */
490     id = qdict_get_try_str(bs_opts, "id");
491     opts = qemu_opts_create(&qemu_common_drive_opts, id, 1, &error);
492     if (error) {
493         error_propagate(errp, error);
494         goto err_no_opts;
495     }
496 
497     qemu_opts_absorb_qdict(opts, bs_opts, &error);
498     if (error) {
499         error_propagate(errp, error);
500         goto early_err;
501     }
502 
503     if (id) {
504         qdict_del(bs_opts, "id");
505     }
506 
507     /* extract parameters */
508     snapshot = qemu_opt_get_bool(opts, "snapshot", 0);
509 
510     account_invalid = qemu_opt_get_bool(opts, "stats-account-invalid", true);
511     account_failed = qemu_opt_get_bool(opts, "stats-account-failed", true);
512 
513     writethrough = !qemu_opt_get_bool(opts, BDRV_OPT_CACHE_WB, true);
514 
515     id = qemu_opts_id(opts);
516 
517     qdict_extract_subqdict(bs_opts, &interval_dict, "stats-intervals.");
518     qdict_array_split(interval_dict, &interval_list);
519 
520     if (qdict_size(interval_dict) != 0) {
521         error_setg(errp, "Invalid option stats-intervals.%s",
522                    qdict_first(interval_dict)->key);
523         goto early_err;
524     }
525 
526     extract_common_blockdev_options(opts, &bdrv_flags, &throttling_group, &cfg,
527                                     &detect_zeroes, &error);
528     if (error) {
529         error_propagate(errp, error);
530         goto early_err;
531     }
532 
533     if ((buf = qemu_opt_get(opts, "format")) != NULL) {
534         if (is_help_option(buf)) {
535             qemu_printf("Supported formats:");
536             bdrv_iterate_format(bdrv_format_print, NULL, false);
537             qemu_printf("\nSupported formats (read-only):");
538             bdrv_iterate_format(bdrv_format_print, NULL, true);
539             qemu_printf("\n");
540             goto early_err;
541         }
542 
543         if (qdict_haskey(bs_opts, "driver")) {
544             error_setg(errp, "Cannot specify both 'driver' and 'format'");
545             goto early_err;
546         }
547         qdict_put_str(bs_opts, "driver", buf);
548     }
549 
550     on_write_error = BLOCKDEV_ON_ERROR_ENOSPC;
551     if ((buf = qemu_opt_get(opts, "werror")) != NULL) {
552         on_write_error = parse_block_error_action(buf, 0, &error);
553         if (error) {
554             error_propagate(errp, error);
555             goto early_err;
556         }
557     }
558 
559     on_read_error = BLOCKDEV_ON_ERROR_REPORT;
560     if ((buf = qemu_opt_get(opts, "rerror")) != NULL) {
561         on_read_error = parse_block_error_action(buf, 1, &error);
562         if (error) {
563             error_propagate(errp, error);
564             goto early_err;
565         }
566     }
567 
568     if (snapshot) {
569         bdrv_flags |= BDRV_O_SNAPSHOT;
570     }
571 
572     read_only = qemu_opt_get_bool(opts, BDRV_OPT_READ_ONLY, false);
573 
574     /* init */
575     if ((!file || !*file) && !qdict_size(bs_opts)) {
576         BlockBackendRootState *blk_rs;
577 
578         blk = blk_new(qemu_get_aio_context(), 0, BLK_PERM_ALL);
579         blk_rs = blk_get_root_state(blk);
580         blk_rs->open_flags    = bdrv_flags;
581         blk_rs->read_only     = read_only;
582         blk_rs->detect_zeroes = detect_zeroes;
583 
584         qobject_unref(bs_opts);
585     } else {
586         if (file && !*file) {
587             file = NULL;
588         }
589 
590         /* bdrv_open() defaults to the values in bdrv_flags (for compatibility
591          * with other callers) rather than what we want as the real defaults.
592          * Apply the defaults here instead. */
593         qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_DIRECT, "off");
594         qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_NO_FLUSH, "off");
595         qdict_set_default_str(bs_opts, BDRV_OPT_READ_ONLY,
596                               read_only ? "on" : "off");
597         qdict_set_default_str(bs_opts, BDRV_OPT_AUTO_READ_ONLY, "on");
598         assert((bdrv_flags & BDRV_O_CACHE_MASK) == 0);
599 
600         if (runstate_check(RUN_STATE_INMIGRATE)) {
601             bdrv_flags |= BDRV_O_INACTIVE;
602         }
603 
604         blk = blk_new_open(file, NULL, bs_opts, bdrv_flags, errp);
605         if (!blk) {
606             goto err_no_bs_opts;
607         }
608         bs = blk_bs(blk);
609 
610         bs->detect_zeroes = detect_zeroes;
611 
612         block_acct_setup(blk_get_stats(blk), account_invalid, account_failed);
613 
614         if (!parse_stats_intervals(blk_get_stats(blk), interval_list, errp)) {
615             blk_unref(blk);
616             blk = NULL;
617             goto err_no_bs_opts;
618         }
619     }
620 
621     /* disk I/O throttling */
622     if (throttle_enabled(&cfg)) {
623         if (!throttling_group) {
624             throttling_group = id;
625         }
626         blk_io_limits_enable(blk, throttling_group);
627         blk_set_io_limits(blk, &cfg);
628     }
629 
630     blk_set_enable_write_cache(blk, !writethrough);
631     blk_set_on_error(blk, on_read_error, on_write_error);
632 
633     if (!monitor_add_blk(blk, id, errp)) {
634         blk_unref(blk);
635         blk = NULL;
636         goto err_no_bs_opts;
637     }
638 
639 err_no_bs_opts:
640     qemu_opts_del(opts);
641     qobject_unref(interval_dict);
642     qobject_unref(interval_list);
643     return blk;
644 
645 early_err:
646     qemu_opts_del(opts);
647     qobject_unref(interval_dict);
648     qobject_unref(interval_list);
649 err_no_opts:
650     qobject_unref(bs_opts);
651     return NULL;
652 }
653 
654 /* Takes the ownership of bs_opts */
655 static BlockDriverState *bds_tree_init(QDict *bs_opts, Error **errp)
656 {
657     int bdrv_flags = 0;
658 
659     /* bdrv_open() defaults to the values in bdrv_flags (for compatibility
660      * with other callers) rather than what we want as the real defaults.
661      * Apply the defaults here instead. */
662     qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_DIRECT, "off");
663     qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_NO_FLUSH, "off");
664     qdict_set_default_str(bs_opts, BDRV_OPT_READ_ONLY, "off");
665 
666     if (runstate_check(RUN_STATE_INMIGRATE)) {
667         bdrv_flags |= BDRV_O_INACTIVE;
668     }
669 
670     return bdrv_open(NULL, NULL, bs_opts, bdrv_flags, errp);
671 }
672 
673 void blockdev_close_all_bdrv_states(void)
674 {
675     BlockDriverState *bs, *next_bs;
676 
677     QTAILQ_FOREACH_SAFE(bs, &monitor_bdrv_states, monitor_list, next_bs) {
678         AioContext *ctx = bdrv_get_aio_context(bs);
679 
680         aio_context_acquire(ctx);
681         bdrv_unref(bs);
682         aio_context_release(ctx);
683     }
684 }
685 
686 /* Iterates over the list of monitor-owned BlockDriverStates */
687 BlockDriverState *bdrv_next_monitor_owned(BlockDriverState *bs)
688 {
689     return bs ? QTAILQ_NEXT(bs, monitor_list)
690               : QTAILQ_FIRST(&monitor_bdrv_states);
691 }
692 
693 static void qemu_opt_rename(QemuOpts *opts, const char *from, const char *to,
694                             Error **errp)
695 {
696     const char *value;
697 
698     value = qemu_opt_get(opts, from);
699     if (value) {
700         if (qemu_opt_find(opts, to)) {
701             error_setg(errp, "'%s' and its alias '%s' can't be used at the "
702                        "same time", to, from);
703             return;
704         }
705     }
706 
707     /* rename all items in opts */
708     while ((value = qemu_opt_get(opts, from))) {
709         qemu_opt_set(opts, to, value, &error_abort);
710         qemu_opt_unset(opts, from);
711     }
712 }
713 
714 QemuOptsList qemu_legacy_drive_opts = {
715     .name = "drive",
716     .head = QTAILQ_HEAD_INITIALIZER(qemu_legacy_drive_opts.head),
717     .desc = {
718         {
719             .name = "bus",
720             .type = QEMU_OPT_NUMBER,
721             .help = "bus number",
722         },{
723             .name = "unit",
724             .type = QEMU_OPT_NUMBER,
725             .help = "unit number (i.e. lun for scsi)",
726         },{
727             .name = "index",
728             .type = QEMU_OPT_NUMBER,
729             .help = "index number",
730         },{
731             .name = "media",
732             .type = QEMU_OPT_STRING,
733             .help = "media type (disk, cdrom)",
734         },{
735             .name = "if",
736             .type = QEMU_OPT_STRING,
737             .help = "interface (ide, scsi, sd, mtd, floppy, pflash, virtio)",
738         },{
739             .name = "file",
740             .type = QEMU_OPT_STRING,
741             .help = "file name",
742         },
743 
744         /* Options that are passed on, but have special semantics with -drive */
745         {
746             .name = BDRV_OPT_READ_ONLY,
747             .type = QEMU_OPT_BOOL,
748             .help = "open drive file as read-only",
749         },{
750             .name = "rerror",
751             .type = QEMU_OPT_STRING,
752             .help = "read error action",
753         },{
754             .name = "werror",
755             .type = QEMU_OPT_STRING,
756             .help = "write error action",
757         },{
758             .name = "copy-on-read",
759             .type = QEMU_OPT_BOOL,
760             .help = "copy read data from backing file into image file",
761         },
762 
763         { /* end of list */ }
764     },
765 };
766 
767 DriveInfo *drive_new(QemuOpts *all_opts, BlockInterfaceType block_default_type,
768                      Error **errp)
769 {
770     const char *value;
771     BlockBackend *blk;
772     DriveInfo *dinfo = NULL;
773     QDict *bs_opts;
774     QemuOpts *legacy_opts;
775     DriveMediaType media = MEDIA_DISK;
776     BlockInterfaceType type;
777     int max_devs, bus_id, unit_id, index;
778     const char *werror, *rerror;
779     bool read_only = false;
780     bool copy_on_read;
781     const char *filename;
782     Error *local_err = NULL;
783     int i;
784 
785     /* Change legacy command line options into QMP ones */
786     static const struct {
787         const char *from;
788         const char *to;
789     } opt_renames[] = {
790         { "iops",           "throttling.iops-total" },
791         { "iops_rd",        "throttling.iops-read" },
792         { "iops_wr",        "throttling.iops-write" },
793 
794         { "bps",            "throttling.bps-total" },
795         { "bps_rd",         "throttling.bps-read" },
796         { "bps_wr",         "throttling.bps-write" },
797 
798         { "iops_max",       "throttling.iops-total-max" },
799         { "iops_rd_max",    "throttling.iops-read-max" },
800         { "iops_wr_max",    "throttling.iops-write-max" },
801 
802         { "bps_max",        "throttling.bps-total-max" },
803         { "bps_rd_max",     "throttling.bps-read-max" },
804         { "bps_wr_max",     "throttling.bps-write-max" },
805 
806         { "iops_size",      "throttling.iops-size" },
807 
808         { "group",          "throttling.group" },
809 
810         { "readonly",       BDRV_OPT_READ_ONLY },
811     };
812 
813     for (i = 0; i < ARRAY_SIZE(opt_renames); i++) {
814         qemu_opt_rename(all_opts, opt_renames[i].from, opt_renames[i].to,
815                         &local_err);
816         if (local_err) {
817             error_propagate(errp, local_err);
818             return NULL;
819         }
820     }
821 
822     value = qemu_opt_get(all_opts, "cache");
823     if (value) {
824         int flags = 0;
825         bool writethrough;
826 
827         if (bdrv_parse_cache_mode(value, &flags, &writethrough) != 0) {
828             error_setg(errp, "invalid cache option");
829             return NULL;
830         }
831 
832         /* Specific options take precedence */
833         if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_WB)) {
834             qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_WB,
835                               !writethrough, &error_abort);
836         }
837         if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_DIRECT)) {
838             qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_DIRECT,
839                               !!(flags & BDRV_O_NOCACHE), &error_abort);
840         }
841         if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_NO_FLUSH)) {
842             qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_NO_FLUSH,
843                               !!(flags & BDRV_O_NO_FLUSH), &error_abort);
844         }
845         qemu_opt_unset(all_opts, "cache");
846     }
847 
848     /* Get a QDict for processing the options */
849     bs_opts = qdict_new();
850     qemu_opts_to_qdict(all_opts, bs_opts);
851 
852     legacy_opts = qemu_opts_create(&qemu_legacy_drive_opts, NULL, 0,
853                                    &error_abort);
854     qemu_opts_absorb_qdict(legacy_opts, bs_opts, &local_err);
855     if (local_err) {
856         error_propagate(errp, local_err);
857         goto fail;
858     }
859 
860     /* Media type */
861     value = qemu_opt_get(legacy_opts, "media");
862     if (value) {
863         if (!strcmp(value, "disk")) {
864             media = MEDIA_DISK;
865         } else if (!strcmp(value, "cdrom")) {
866             media = MEDIA_CDROM;
867             read_only = true;
868         } else {
869             error_setg(errp, "'%s' invalid media", value);
870             goto fail;
871         }
872     }
873 
874     /* copy-on-read is disabled with a warning for read-only devices */
875     read_only |= qemu_opt_get_bool(legacy_opts, BDRV_OPT_READ_ONLY, false);
876     copy_on_read = qemu_opt_get_bool(legacy_opts, "copy-on-read", false);
877 
878     if (read_only && copy_on_read) {
879         warn_report("disabling copy-on-read on read-only drive");
880         copy_on_read = false;
881     }
882 
883     qdict_put_str(bs_opts, BDRV_OPT_READ_ONLY, read_only ? "on" : "off");
884     qdict_put_str(bs_opts, "copy-on-read", copy_on_read ? "on" : "off");
885 
886     /* Controller type */
887     value = qemu_opt_get(legacy_opts, "if");
888     if (value) {
889         for (type = 0;
890              type < IF_COUNT && strcmp(value, if_name[type]);
891              type++) {
892         }
893         if (type == IF_COUNT) {
894             error_setg(errp, "unsupported bus type '%s'", value);
895             goto fail;
896         }
897     } else {
898         type = block_default_type;
899     }
900 
901     /* Device address specified by bus/unit or index.
902      * If none was specified, try to find the first free one. */
903     bus_id  = qemu_opt_get_number(legacy_opts, "bus", 0);
904     unit_id = qemu_opt_get_number(legacy_opts, "unit", -1);
905     index   = qemu_opt_get_number(legacy_opts, "index", -1);
906 
907     max_devs = if_max_devs[type];
908 
909     if (index != -1) {
910         if (bus_id != 0 || unit_id != -1) {
911             error_setg(errp, "index cannot be used with bus and unit");
912             goto fail;
913         }
914         bus_id = drive_index_to_bus_id(type, index);
915         unit_id = drive_index_to_unit_id(type, index);
916     }
917 
918     if (unit_id == -1) {
919        unit_id = 0;
920        while (drive_get(type, bus_id, unit_id) != NULL) {
921            unit_id++;
922            if (max_devs && unit_id >= max_devs) {
923                unit_id -= max_devs;
924                bus_id++;
925            }
926        }
927     }
928 
929     if (max_devs && unit_id >= max_devs) {
930         error_setg(errp, "unit %d too big (max is %d)", unit_id, max_devs - 1);
931         goto fail;
932     }
933 
934     if (drive_get(type, bus_id, unit_id) != NULL) {
935         error_setg(errp, "drive with bus=%d, unit=%d (index=%d) exists",
936                    bus_id, unit_id, index);
937         goto fail;
938     }
939 
940     /* no id supplied -> create one */
941     if (qemu_opts_id(all_opts) == NULL) {
942         char *new_id;
943         const char *mediastr = "";
944         if (type == IF_IDE || type == IF_SCSI) {
945             mediastr = (media == MEDIA_CDROM) ? "-cd" : "-hd";
946         }
947         if (max_devs) {
948             new_id = g_strdup_printf("%s%i%s%i", if_name[type], bus_id,
949                                      mediastr, unit_id);
950         } else {
951             new_id = g_strdup_printf("%s%s%i", if_name[type],
952                                      mediastr, unit_id);
953         }
954         qdict_put_str(bs_opts, "id", new_id);
955         g_free(new_id);
956     }
957 
958     /* Add virtio block device */
959     if (type == IF_VIRTIO) {
960         QemuOpts *devopts;
961         devopts = qemu_opts_create(qemu_find_opts("device"), NULL, 0,
962                                    &error_abort);
963         if (arch_type == QEMU_ARCH_S390X) {
964             qemu_opt_set(devopts, "driver", "virtio-blk-ccw", &error_abort);
965         } else {
966             qemu_opt_set(devopts, "driver", "virtio-blk-pci", &error_abort);
967         }
968         qemu_opt_set(devopts, "drive", qdict_get_str(bs_opts, "id"),
969                      &error_abort);
970     }
971 
972     filename = qemu_opt_get(legacy_opts, "file");
973 
974     /* Check werror/rerror compatibility with if=... */
975     werror = qemu_opt_get(legacy_opts, "werror");
976     if (werror != NULL) {
977         if (type != IF_IDE && type != IF_SCSI && type != IF_VIRTIO &&
978             type != IF_NONE) {
979             error_setg(errp, "werror is not supported by this bus type");
980             goto fail;
981         }
982         qdict_put_str(bs_opts, "werror", werror);
983     }
984 
985     rerror = qemu_opt_get(legacy_opts, "rerror");
986     if (rerror != NULL) {
987         if (type != IF_IDE && type != IF_VIRTIO && type != IF_SCSI &&
988             type != IF_NONE) {
989             error_setg(errp, "rerror is not supported by this bus type");
990             goto fail;
991         }
992         qdict_put_str(bs_opts, "rerror", rerror);
993     }
994 
995     /* Actual block device init: Functionality shared with blockdev-add */
996     blk = blockdev_init(filename, bs_opts, &local_err);
997     bs_opts = NULL;
998     if (!blk) {
999         error_propagate(errp, local_err);
1000         goto fail;
1001     } else {
1002         assert(!local_err);
1003     }
1004 
1005     /* Create legacy DriveInfo */
1006     dinfo = g_malloc0(sizeof(*dinfo));
1007     dinfo->opts = all_opts;
1008 
1009     dinfo->type = type;
1010     dinfo->bus = bus_id;
1011     dinfo->unit = unit_id;
1012 
1013     blk_set_legacy_dinfo(blk, dinfo);
1014 
1015     switch(type) {
1016     case IF_IDE:
1017     case IF_SCSI:
1018     case IF_XEN:
1019     case IF_NONE:
1020         dinfo->media_cd = media == MEDIA_CDROM;
1021         break;
1022     default:
1023         break;
1024     }
1025 
1026 fail:
1027     qemu_opts_del(legacy_opts);
1028     qobject_unref(bs_opts);
1029     return dinfo;
1030 }
1031 
1032 static BlockDriverState *qmp_get_root_bs(const char *name, Error **errp)
1033 {
1034     BlockDriverState *bs;
1035 
1036     bs = bdrv_lookup_bs(name, name, errp);
1037     if (bs == NULL) {
1038         return NULL;
1039     }
1040 
1041     if (!bdrv_is_root_node(bs)) {
1042         error_setg(errp, "Need a root block node");
1043         return NULL;
1044     }
1045 
1046     if (!bdrv_is_inserted(bs)) {
1047         error_setg(errp, "Device has no medium");
1048         return NULL;
1049     }
1050 
1051     return bs;
1052 }
1053 
1054 static BlockBackend *qmp_get_blk(const char *blk_name, const char *qdev_id,
1055                                  Error **errp)
1056 {
1057     BlockBackend *blk;
1058 
1059     if (!blk_name == !qdev_id) {
1060         error_setg(errp, "Need exactly one of 'device' and 'id'");
1061         return NULL;
1062     }
1063 
1064     if (qdev_id) {
1065         blk = blk_by_qdev_id(qdev_id, errp);
1066     } else {
1067         blk = blk_by_name(blk_name);
1068         if (blk == NULL) {
1069             error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
1070                       "Device '%s' not found", blk_name);
1071         }
1072     }
1073 
1074     return blk;
1075 }
1076 
1077 void hmp_commit(Monitor *mon, const QDict *qdict)
1078 {
1079     const char *device = qdict_get_str(qdict, "device");
1080     BlockBackend *blk;
1081     int ret;
1082 
1083     if (!strcmp(device, "all")) {
1084         ret = blk_commit_all();
1085     } else {
1086         BlockDriverState *bs;
1087         AioContext *aio_context;
1088 
1089         blk = blk_by_name(device);
1090         if (!blk) {
1091             error_report("Device '%s' not found", device);
1092             return;
1093         }
1094         if (!blk_is_available(blk)) {
1095             error_report("Device '%s' has no medium", device);
1096             return;
1097         }
1098 
1099         bs = blk_bs(blk);
1100         aio_context = bdrv_get_aio_context(bs);
1101         aio_context_acquire(aio_context);
1102 
1103         ret = bdrv_commit(bs);
1104 
1105         aio_context_release(aio_context);
1106     }
1107     if (ret < 0) {
1108         error_report("'commit' error for '%s': %s", device, strerror(-ret));
1109     }
1110 }
1111 
1112 static void blockdev_do_action(TransactionAction *action, Error **errp)
1113 {
1114     TransactionActionList list;
1115 
1116     list.value = action;
1117     list.next = NULL;
1118     qmp_transaction(&list, false, NULL, errp);
1119 }
1120 
1121 void qmp_blockdev_snapshot_sync(bool has_device, const char *device,
1122                                 bool has_node_name, const char *node_name,
1123                                 const char *snapshot_file,
1124                                 bool has_snapshot_node_name,
1125                                 const char *snapshot_node_name,
1126                                 bool has_format, const char *format,
1127                                 bool has_mode, NewImageMode mode, Error **errp)
1128 {
1129     BlockdevSnapshotSync snapshot = {
1130         .has_device = has_device,
1131         .device = (char *) device,
1132         .has_node_name = has_node_name,
1133         .node_name = (char *) node_name,
1134         .snapshot_file = (char *) snapshot_file,
1135         .has_snapshot_node_name = has_snapshot_node_name,
1136         .snapshot_node_name = (char *) snapshot_node_name,
1137         .has_format = has_format,
1138         .format = (char *) format,
1139         .has_mode = has_mode,
1140         .mode = mode,
1141     };
1142     TransactionAction action = {
1143         .type = TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC,
1144         .u.blockdev_snapshot_sync.data = &snapshot,
1145     };
1146     blockdev_do_action(&action, errp);
1147 }
1148 
1149 void qmp_blockdev_snapshot(const char *node, const char *overlay,
1150                            Error **errp)
1151 {
1152     BlockdevSnapshot snapshot_data = {
1153         .node = (char *) node,
1154         .overlay = (char *) overlay
1155     };
1156     TransactionAction action = {
1157         .type = TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT,
1158         .u.blockdev_snapshot.data = &snapshot_data,
1159     };
1160     blockdev_do_action(&action, errp);
1161 }
1162 
1163 void qmp_blockdev_snapshot_internal_sync(const char *device,
1164                                          const char *name,
1165                                          Error **errp)
1166 {
1167     BlockdevSnapshotInternal snapshot = {
1168         .device = (char *) device,
1169         .name = (char *) name
1170     };
1171     TransactionAction action = {
1172         .type = TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC,
1173         .u.blockdev_snapshot_internal_sync.data = &snapshot,
1174     };
1175     blockdev_do_action(&action, errp);
1176 }
1177 
1178 SnapshotInfo *qmp_blockdev_snapshot_delete_internal_sync(const char *device,
1179                                                          bool has_id,
1180                                                          const char *id,
1181                                                          bool has_name,
1182                                                          const char *name,
1183                                                          Error **errp)
1184 {
1185     BlockDriverState *bs;
1186     AioContext *aio_context;
1187     QEMUSnapshotInfo sn;
1188     Error *local_err = NULL;
1189     SnapshotInfo *info = NULL;
1190     int ret;
1191 
1192     bs = qmp_get_root_bs(device, errp);
1193     if (!bs) {
1194         return NULL;
1195     }
1196     aio_context = bdrv_get_aio_context(bs);
1197     aio_context_acquire(aio_context);
1198 
1199     if (!has_id) {
1200         id = NULL;
1201     }
1202 
1203     if (!has_name) {
1204         name = NULL;
1205     }
1206 
1207     if (!id && !name) {
1208         error_setg(errp, "Name or id must be provided");
1209         goto out_aio_context;
1210     }
1211 
1212     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_INTERNAL_SNAPSHOT_DELETE, errp)) {
1213         goto out_aio_context;
1214     }
1215 
1216     ret = bdrv_snapshot_find_by_id_and_name(bs, id, name, &sn, &local_err);
1217     if (local_err) {
1218         error_propagate(errp, local_err);
1219         goto out_aio_context;
1220     }
1221     if (!ret) {
1222         error_setg(errp,
1223                    "Snapshot with id '%s' and name '%s' does not exist on "
1224                    "device '%s'",
1225                    STR_OR_NULL(id), STR_OR_NULL(name), device);
1226         goto out_aio_context;
1227     }
1228 
1229     bdrv_snapshot_delete(bs, id, name, &local_err);
1230     if (local_err) {
1231         error_propagate(errp, local_err);
1232         goto out_aio_context;
1233     }
1234 
1235     aio_context_release(aio_context);
1236 
1237     info = g_new0(SnapshotInfo, 1);
1238     info->id = g_strdup(sn.id_str);
1239     info->name = g_strdup(sn.name);
1240     info->date_nsec = sn.date_nsec;
1241     info->date_sec = sn.date_sec;
1242     info->vm_state_size = sn.vm_state_size;
1243     info->vm_clock_nsec = sn.vm_clock_nsec % 1000000000;
1244     info->vm_clock_sec = sn.vm_clock_nsec / 1000000000;
1245 
1246     return info;
1247 
1248 out_aio_context:
1249     aio_context_release(aio_context);
1250     return NULL;
1251 }
1252 
1253 /**
1254  * block_dirty_bitmap_lookup:
1255  * Return a dirty bitmap (if present), after validating
1256  * the node reference and bitmap names.
1257  *
1258  * @node: The name of the BDS node to search for bitmaps
1259  * @name: The name of the bitmap to search for
1260  * @pbs: Output pointer for BDS lookup, if desired. Can be NULL.
1261  * @errp: Output pointer for error information. Can be NULL.
1262  *
1263  * @return: A bitmap object on success, or NULL on failure.
1264  */
1265 static BdrvDirtyBitmap *block_dirty_bitmap_lookup(const char *node,
1266                                                   const char *name,
1267                                                   BlockDriverState **pbs,
1268                                                   Error **errp)
1269 {
1270     BlockDriverState *bs;
1271     BdrvDirtyBitmap *bitmap;
1272 
1273     if (!node) {
1274         error_setg(errp, "Node cannot be NULL");
1275         return NULL;
1276     }
1277     if (!name) {
1278         error_setg(errp, "Bitmap name cannot be NULL");
1279         return NULL;
1280     }
1281     bs = bdrv_lookup_bs(node, node, NULL);
1282     if (!bs) {
1283         error_setg(errp, "Node '%s' not found", node);
1284         return NULL;
1285     }
1286 
1287     bitmap = bdrv_find_dirty_bitmap(bs, name);
1288     if (!bitmap) {
1289         error_setg(errp, "Dirty bitmap '%s' not found", name);
1290         return NULL;
1291     }
1292 
1293     if (pbs) {
1294         *pbs = bs;
1295     }
1296 
1297     return bitmap;
1298 }
1299 
1300 /* New and old BlockDriverState structs for atomic group operations */
1301 
1302 typedef struct BlkActionState BlkActionState;
1303 
1304 /**
1305  * BlkActionOps:
1306  * Table of operations that define an Action.
1307  *
1308  * @instance_size: Size of state struct, in bytes.
1309  * @prepare: Prepare the work, must NOT be NULL.
1310  * @commit: Commit the changes, can be NULL.
1311  * @abort: Abort the changes on fail, can be NULL.
1312  * @clean: Clean up resources after all transaction actions have called
1313  *         commit() or abort(). Can be NULL.
1314  *
1315  * Only prepare() may fail. In a single transaction, only one of commit() or
1316  * abort() will be called. clean() will always be called if it is present.
1317  */
1318 typedef struct BlkActionOps {
1319     size_t instance_size;
1320     void (*prepare)(BlkActionState *common, Error **errp);
1321     void (*commit)(BlkActionState *common);
1322     void (*abort)(BlkActionState *common);
1323     void (*clean)(BlkActionState *common);
1324 } BlkActionOps;
1325 
1326 /**
1327  * BlkActionState:
1328  * Describes one Action's state within a Transaction.
1329  *
1330  * @action: QAPI-defined enum identifying which Action to perform.
1331  * @ops: Table of ActionOps this Action can perform.
1332  * @block_job_txn: Transaction which this action belongs to.
1333  * @entry: List membership for all Actions in this Transaction.
1334  *
1335  * This structure must be arranged as first member in a subclassed type,
1336  * assuming that the compiler will also arrange it to the same offsets as the
1337  * base class.
1338  */
1339 struct BlkActionState {
1340     TransactionAction *action;
1341     const BlkActionOps *ops;
1342     JobTxn *block_job_txn;
1343     TransactionProperties *txn_props;
1344     QTAILQ_ENTRY(BlkActionState) entry;
1345 };
1346 
1347 /* internal snapshot private data */
1348 typedef struct InternalSnapshotState {
1349     BlkActionState common;
1350     BlockDriverState *bs;
1351     QEMUSnapshotInfo sn;
1352     bool created;
1353 } InternalSnapshotState;
1354 
1355 
1356 static int action_check_completion_mode(BlkActionState *s, Error **errp)
1357 {
1358     if (s->txn_props->completion_mode != ACTION_COMPLETION_MODE_INDIVIDUAL) {
1359         error_setg(errp,
1360                    "Action '%s' does not support Transaction property "
1361                    "completion-mode = %s",
1362                    TransactionActionKind_str(s->action->type),
1363                    ActionCompletionMode_str(s->txn_props->completion_mode));
1364         return -1;
1365     }
1366     return 0;
1367 }
1368 
1369 static void internal_snapshot_prepare(BlkActionState *common,
1370                                       Error **errp)
1371 {
1372     Error *local_err = NULL;
1373     const char *device;
1374     const char *name;
1375     BlockDriverState *bs;
1376     QEMUSnapshotInfo old_sn, *sn;
1377     bool ret;
1378     qemu_timeval tv;
1379     BlockdevSnapshotInternal *internal;
1380     InternalSnapshotState *state;
1381     AioContext *aio_context;
1382     int ret1;
1383 
1384     g_assert(common->action->type ==
1385              TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC);
1386     internal = common->action->u.blockdev_snapshot_internal_sync.data;
1387     state = DO_UPCAST(InternalSnapshotState, common, common);
1388 
1389     /* 1. parse input */
1390     device = internal->device;
1391     name = internal->name;
1392 
1393     /* 2. check for validation */
1394     if (action_check_completion_mode(common, errp) < 0) {
1395         return;
1396     }
1397 
1398     bs = qmp_get_root_bs(device, errp);
1399     if (!bs) {
1400         return;
1401     }
1402 
1403     aio_context = bdrv_get_aio_context(bs);
1404     aio_context_acquire(aio_context);
1405 
1406     state->bs = bs;
1407 
1408     /* Paired with .clean() */
1409     bdrv_drained_begin(bs);
1410 
1411     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_INTERNAL_SNAPSHOT, errp)) {
1412         goto out;
1413     }
1414 
1415     if (bdrv_is_read_only(bs)) {
1416         error_setg(errp, "Device '%s' is read only", device);
1417         goto out;
1418     }
1419 
1420     if (!bdrv_can_snapshot(bs)) {
1421         error_setg(errp, "Block format '%s' used by device '%s' "
1422                    "does not support internal snapshots",
1423                    bs->drv->format_name, device);
1424         goto out;
1425     }
1426 
1427     if (!strlen(name)) {
1428         error_setg(errp, "Name is empty");
1429         goto out;
1430     }
1431 
1432     /* check whether a snapshot with name exist */
1433     ret = bdrv_snapshot_find_by_id_and_name(bs, NULL, name, &old_sn,
1434                                             &local_err);
1435     if (local_err) {
1436         error_propagate(errp, local_err);
1437         goto out;
1438     } else if (ret) {
1439         error_setg(errp,
1440                    "Snapshot with name '%s' already exists on device '%s'",
1441                    name, device);
1442         goto out;
1443     }
1444 
1445     /* 3. take the snapshot */
1446     sn = &state->sn;
1447     pstrcpy(sn->name, sizeof(sn->name), name);
1448     qemu_gettimeofday(&tv);
1449     sn->date_sec = tv.tv_sec;
1450     sn->date_nsec = tv.tv_usec * 1000;
1451     sn->vm_clock_nsec = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL);
1452 
1453     ret1 = bdrv_snapshot_create(bs, sn);
1454     if (ret1 < 0) {
1455         error_setg_errno(errp, -ret1,
1456                          "Failed to create snapshot '%s' on device '%s'",
1457                          name, device);
1458         goto out;
1459     }
1460 
1461     /* 4. succeed, mark a snapshot is created */
1462     state->created = true;
1463 
1464 out:
1465     aio_context_release(aio_context);
1466 }
1467 
1468 static void internal_snapshot_abort(BlkActionState *common)
1469 {
1470     InternalSnapshotState *state =
1471                              DO_UPCAST(InternalSnapshotState, common, common);
1472     BlockDriverState *bs = state->bs;
1473     QEMUSnapshotInfo *sn = &state->sn;
1474     AioContext *aio_context;
1475     Error *local_error = NULL;
1476 
1477     if (!state->created) {
1478         return;
1479     }
1480 
1481     aio_context = bdrv_get_aio_context(state->bs);
1482     aio_context_acquire(aio_context);
1483 
1484     if (bdrv_snapshot_delete(bs, sn->id_str, sn->name, &local_error) < 0) {
1485         error_reportf_err(local_error,
1486                           "Failed to delete snapshot with id '%s' and "
1487                           "name '%s' on device '%s' in abort: ",
1488                           sn->id_str, sn->name,
1489                           bdrv_get_device_name(bs));
1490     }
1491 
1492     aio_context_release(aio_context);
1493 }
1494 
1495 static void internal_snapshot_clean(BlkActionState *common)
1496 {
1497     InternalSnapshotState *state = DO_UPCAST(InternalSnapshotState,
1498                                              common, common);
1499     AioContext *aio_context;
1500 
1501     if (!state->bs) {
1502         return;
1503     }
1504 
1505     aio_context = bdrv_get_aio_context(state->bs);
1506     aio_context_acquire(aio_context);
1507 
1508     bdrv_drained_end(state->bs);
1509 
1510     aio_context_release(aio_context);
1511 }
1512 
1513 /* external snapshot private data */
1514 typedef struct ExternalSnapshotState {
1515     BlkActionState common;
1516     BlockDriverState *old_bs;
1517     BlockDriverState *new_bs;
1518     bool overlay_appended;
1519 } ExternalSnapshotState;
1520 
1521 static void external_snapshot_prepare(BlkActionState *common,
1522                                       Error **errp)
1523 {
1524     int flags = 0;
1525     QDict *options = NULL;
1526     Error *local_err = NULL;
1527     /* Device and node name of the image to generate the snapshot from */
1528     const char *device;
1529     const char *node_name;
1530     /* Reference to the new image (for 'blockdev-snapshot') */
1531     const char *snapshot_ref;
1532     /* File name of the new image (for 'blockdev-snapshot-sync') */
1533     const char *new_image_file;
1534     ExternalSnapshotState *state =
1535                              DO_UPCAST(ExternalSnapshotState, common, common);
1536     TransactionAction *action = common->action;
1537     AioContext *aio_context;
1538     AioContext *old_context;
1539     int ret;
1540 
1541     /* 'blockdev-snapshot' and 'blockdev-snapshot-sync' have similar
1542      * purpose but a different set of parameters */
1543     switch (action->type) {
1544     case TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT:
1545         {
1546             BlockdevSnapshot *s = action->u.blockdev_snapshot.data;
1547             device = s->node;
1548             node_name = s->node;
1549             new_image_file = NULL;
1550             snapshot_ref = s->overlay;
1551         }
1552         break;
1553     case TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC:
1554         {
1555             BlockdevSnapshotSync *s = action->u.blockdev_snapshot_sync.data;
1556             device = s->has_device ? s->device : NULL;
1557             node_name = s->has_node_name ? s->node_name : NULL;
1558             new_image_file = s->snapshot_file;
1559             snapshot_ref = NULL;
1560         }
1561         break;
1562     default:
1563         g_assert_not_reached();
1564     }
1565 
1566     /* start processing */
1567     if (action_check_completion_mode(common, errp) < 0) {
1568         return;
1569     }
1570 
1571     state->old_bs = bdrv_lookup_bs(device, node_name, errp);
1572     if (!state->old_bs) {
1573         return;
1574     }
1575 
1576     aio_context = bdrv_get_aio_context(state->old_bs);
1577     aio_context_acquire(aio_context);
1578 
1579     /* Paired with .clean() */
1580     bdrv_drained_begin(state->old_bs);
1581 
1582     if (!bdrv_is_inserted(state->old_bs)) {
1583         error_setg(errp, QERR_DEVICE_HAS_NO_MEDIUM, device);
1584         goto out;
1585     }
1586 
1587     if (bdrv_op_is_blocked(state->old_bs,
1588                            BLOCK_OP_TYPE_EXTERNAL_SNAPSHOT, errp)) {
1589         goto out;
1590     }
1591 
1592     if (!bdrv_is_read_only(state->old_bs)) {
1593         if (bdrv_flush(state->old_bs)) {
1594             error_setg(errp, QERR_IO_ERROR);
1595             goto out;
1596         }
1597     }
1598 
1599     if (!bdrv_is_first_non_filter(state->old_bs)) {
1600         error_setg(errp, QERR_FEATURE_DISABLED, "snapshot");
1601         goto out;
1602     }
1603 
1604     if (action->type == TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC) {
1605         BlockdevSnapshotSync *s = action->u.blockdev_snapshot_sync.data;
1606         const char *format = s->has_format ? s->format : "qcow2";
1607         enum NewImageMode mode;
1608         const char *snapshot_node_name =
1609             s->has_snapshot_node_name ? s->snapshot_node_name : NULL;
1610 
1611         if (node_name && !snapshot_node_name) {
1612             error_setg(errp, "New overlay node name missing");
1613             goto out;
1614         }
1615 
1616         if (snapshot_node_name &&
1617             bdrv_lookup_bs(snapshot_node_name, snapshot_node_name, NULL)) {
1618             error_setg(errp, "New overlay node name already in use");
1619             goto out;
1620         }
1621 
1622         flags = state->old_bs->open_flags;
1623         flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_COPY_ON_READ);
1624         flags |= BDRV_O_NO_BACKING;
1625 
1626         /* create new image w/backing file */
1627         mode = s->has_mode ? s->mode : NEW_IMAGE_MODE_ABSOLUTE_PATHS;
1628         if (mode != NEW_IMAGE_MODE_EXISTING) {
1629             int64_t size = bdrv_getlength(state->old_bs);
1630             if (size < 0) {
1631                 error_setg_errno(errp, -size, "bdrv_getlength failed");
1632                 goto out;
1633             }
1634             bdrv_refresh_filename(state->old_bs);
1635             bdrv_img_create(new_image_file, format,
1636                             state->old_bs->filename,
1637                             state->old_bs->drv->format_name,
1638                             NULL, size, flags, false, &local_err);
1639             if (local_err) {
1640                 error_propagate(errp, local_err);
1641                 goto out;
1642             }
1643         }
1644 
1645         options = qdict_new();
1646         if (snapshot_node_name) {
1647             qdict_put_str(options, "node-name", snapshot_node_name);
1648         }
1649         qdict_put_str(options, "driver", format);
1650     }
1651 
1652     state->new_bs = bdrv_open(new_image_file, snapshot_ref, options, flags,
1653                               errp);
1654     /* We will manually add the backing_hd field to the bs later */
1655     if (!state->new_bs) {
1656         goto out;
1657     }
1658 
1659     if (bdrv_has_blk(state->new_bs)) {
1660         error_setg(errp, "The overlay is already in use");
1661         goto out;
1662     }
1663 
1664     if (bdrv_op_is_blocked(state->new_bs, BLOCK_OP_TYPE_EXTERNAL_SNAPSHOT,
1665                            errp)) {
1666         goto out;
1667     }
1668 
1669     if (state->new_bs->backing != NULL) {
1670         error_setg(errp, "The overlay already has a backing image");
1671         goto out;
1672     }
1673 
1674     if (!state->new_bs->drv->supports_backing) {
1675         error_setg(errp, "The overlay does not support backing images");
1676         goto out;
1677     }
1678 
1679     /* Honor bdrv_try_set_aio_context() context acquisition requirements. */
1680     old_context = bdrv_get_aio_context(state->new_bs);
1681     aio_context_release(aio_context);
1682     aio_context_acquire(old_context);
1683 
1684     ret = bdrv_try_set_aio_context(state->new_bs, aio_context, errp);
1685 
1686     aio_context_release(old_context);
1687     aio_context_acquire(aio_context);
1688 
1689     if (ret < 0) {
1690         goto out;
1691     }
1692 
1693     /* This removes our old bs and adds the new bs. This is an operation that
1694      * can fail, so we need to do it in .prepare; undoing it for abort is
1695      * always possible. */
1696     bdrv_ref(state->new_bs);
1697     bdrv_append(state->new_bs, state->old_bs, &local_err);
1698     if (local_err) {
1699         error_propagate(errp, local_err);
1700         goto out;
1701     }
1702     state->overlay_appended = true;
1703 
1704 out:
1705     aio_context_release(aio_context);
1706 }
1707 
1708 static void external_snapshot_commit(BlkActionState *common)
1709 {
1710     ExternalSnapshotState *state =
1711                              DO_UPCAST(ExternalSnapshotState, common, common);
1712     AioContext *aio_context;
1713 
1714     aio_context = bdrv_get_aio_context(state->old_bs);
1715     aio_context_acquire(aio_context);
1716 
1717     /* We don't need (or want) to use the transactional
1718      * bdrv_reopen_multiple() across all the entries at once, because we
1719      * don't want to abort all of them if one of them fails the reopen */
1720     if (!atomic_read(&state->old_bs->copy_on_read)) {
1721         bdrv_reopen_set_read_only(state->old_bs, true, NULL);
1722     }
1723 
1724     aio_context_release(aio_context);
1725 }
1726 
1727 static void external_snapshot_abort(BlkActionState *common)
1728 {
1729     ExternalSnapshotState *state =
1730                              DO_UPCAST(ExternalSnapshotState, common, common);
1731     if (state->new_bs) {
1732         if (state->overlay_appended) {
1733             AioContext *aio_context;
1734             AioContext *tmp_context;
1735             int ret;
1736 
1737             aio_context = bdrv_get_aio_context(state->old_bs);
1738             aio_context_acquire(aio_context);
1739 
1740             bdrv_ref(state->old_bs);   /* we can't let bdrv_set_backind_hd()
1741                                           close state->old_bs; we need it */
1742             bdrv_set_backing_hd(state->new_bs, NULL, &error_abort);
1743 
1744             /*
1745              * The call to bdrv_set_backing_hd() above returns state->old_bs to
1746              * the main AioContext. As we're still going to be using it, return
1747              * it to the AioContext it was before.
1748              */
1749             tmp_context = bdrv_get_aio_context(state->old_bs);
1750             if (aio_context != tmp_context) {
1751                 aio_context_release(aio_context);
1752                 aio_context_acquire(tmp_context);
1753 
1754                 ret = bdrv_try_set_aio_context(state->old_bs,
1755                                                aio_context, NULL);
1756                 assert(ret == 0);
1757 
1758                 aio_context_release(tmp_context);
1759                 aio_context_acquire(aio_context);
1760             }
1761 
1762             bdrv_replace_node(state->new_bs, state->old_bs, &error_abort);
1763             bdrv_unref(state->old_bs); /* bdrv_replace_node() ref'ed old_bs */
1764 
1765             aio_context_release(aio_context);
1766         }
1767     }
1768 }
1769 
1770 static void external_snapshot_clean(BlkActionState *common)
1771 {
1772     ExternalSnapshotState *state =
1773                              DO_UPCAST(ExternalSnapshotState, common, common);
1774     AioContext *aio_context;
1775 
1776     if (!state->old_bs) {
1777         return;
1778     }
1779 
1780     aio_context = bdrv_get_aio_context(state->old_bs);
1781     aio_context_acquire(aio_context);
1782 
1783     bdrv_drained_end(state->old_bs);
1784     bdrv_unref(state->new_bs);
1785 
1786     aio_context_release(aio_context);
1787 }
1788 
1789 typedef struct DriveBackupState {
1790     BlkActionState common;
1791     BlockDriverState *bs;
1792     BlockJob *job;
1793 } DriveBackupState;
1794 
1795 static BlockJob *do_backup_common(BackupCommon *backup,
1796                                   BlockDriverState *bs,
1797                                   BlockDriverState *target_bs,
1798                                   AioContext *aio_context,
1799                                   JobTxn *txn, Error **errp);
1800 
1801 static void drive_backup_prepare(BlkActionState *common, Error **errp)
1802 {
1803     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1804     DriveBackup *backup;
1805     BlockDriverState *bs;
1806     BlockDriverState *target_bs;
1807     BlockDriverState *source = NULL;
1808     AioContext *aio_context;
1809     AioContext *old_context;
1810     QDict *options;
1811     Error *local_err = NULL;
1812     int flags;
1813     int64_t size;
1814     bool set_backing_hd = false;
1815     int ret;
1816 
1817     assert(common->action->type == TRANSACTION_ACTION_KIND_DRIVE_BACKUP);
1818     backup = common->action->u.drive_backup.data;
1819 
1820     if (!backup->has_mode) {
1821         backup->mode = NEW_IMAGE_MODE_ABSOLUTE_PATHS;
1822     }
1823 
1824     bs = bdrv_lookup_bs(backup->device, backup->device, errp);
1825     if (!bs) {
1826         return;
1827     }
1828 
1829     if (!bs->drv) {
1830         error_setg(errp, "Device has no medium");
1831         return;
1832     }
1833 
1834     aio_context = bdrv_get_aio_context(bs);
1835     aio_context_acquire(aio_context);
1836 
1837     /* Paired with .clean() */
1838     bdrv_drained_begin(bs);
1839 
1840     if (!backup->has_format) {
1841         backup->format = backup->mode == NEW_IMAGE_MODE_EXISTING ?
1842                          NULL : (char *) bs->drv->format_name;
1843     }
1844 
1845     /* Early check to avoid creating target */
1846     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_BACKUP_SOURCE, errp)) {
1847         goto out;
1848     }
1849 
1850     flags = bs->open_flags | BDRV_O_RDWR;
1851 
1852     /*
1853      * See if we have a backing HD we can use to create our new image
1854      * on top of.
1855      */
1856     if (backup->sync == MIRROR_SYNC_MODE_TOP) {
1857         source = backing_bs(bs);
1858         if (!source) {
1859             backup->sync = MIRROR_SYNC_MODE_FULL;
1860         }
1861     }
1862     if (backup->sync == MIRROR_SYNC_MODE_NONE) {
1863         source = bs;
1864         flags |= BDRV_O_NO_BACKING;
1865         set_backing_hd = true;
1866     }
1867 
1868     size = bdrv_getlength(bs);
1869     if (size < 0) {
1870         error_setg_errno(errp, -size, "bdrv_getlength failed");
1871         goto out;
1872     }
1873 
1874     if (backup->mode != NEW_IMAGE_MODE_EXISTING) {
1875         assert(backup->format);
1876         if (source) {
1877             bdrv_refresh_filename(source);
1878             bdrv_img_create(backup->target, backup->format, source->filename,
1879                             source->drv->format_name, NULL,
1880                             size, flags, false, &local_err);
1881         } else {
1882             bdrv_img_create(backup->target, backup->format, NULL, NULL, NULL,
1883                             size, flags, false, &local_err);
1884         }
1885     }
1886 
1887     if (local_err) {
1888         error_propagate(errp, local_err);
1889         goto out;
1890     }
1891 
1892     options = qdict_new();
1893     qdict_put_str(options, "discard", "unmap");
1894     qdict_put_str(options, "detect-zeroes", "unmap");
1895     if (backup->format) {
1896         qdict_put_str(options, "driver", backup->format);
1897     }
1898 
1899     target_bs = bdrv_open(backup->target, NULL, options, flags, errp);
1900     if (!target_bs) {
1901         goto out;
1902     }
1903 
1904     /* Honor bdrv_try_set_aio_context() context acquisition requirements. */
1905     old_context = bdrv_get_aio_context(target_bs);
1906     aio_context_release(aio_context);
1907     aio_context_acquire(old_context);
1908 
1909     ret = bdrv_try_set_aio_context(target_bs, aio_context, errp);
1910     if (ret < 0) {
1911         bdrv_unref(target_bs);
1912         aio_context_release(old_context);
1913         return;
1914     }
1915 
1916     aio_context_release(old_context);
1917     aio_context_acquire(aio_context);
1918 
1919     if (set_backing_hd) {
1920         bdrv_set_backing_hd(target_bs, source, &local_err);
1921         if (local_err) {
1922             goto unref;
1923         }
1924     }
1925 
1926     state->bs = bs;
1927 
1928     state->job = do_backup_common(qapi_DriveBackup_base(backup),
1929                                   bs, target_bs, aio_context,
1930                                   common->block_job_txn, errp);
1931 
1932 unref:
1933     bdrv_unref(target_bs);
1934 out:
1935     aio_context_release(aio_context);
1936 }
1937 
1938 static void drive_backup_commit(BlkActionState *common)
1939 {
1940     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1941     AioContext *aio_context;
1942 
1943     aio_context = bdrv_get_aio_context(state->bs);
1944     aio_context_acquire(aio_context);
1945 
1946     assert(state->job);
1947     job_start(&state->job->job);
1948 
1949     aio_context_release(aio_context);
1950 }
1951 
1952 static void drive_backup_abort(BlkActionState *common)
1953 {
1954     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1955 
1956     if (state->job) {
1957         AioContext *aio_context;
1958 
1959         aio_context = bdrv_get_aio_context(state->bs);
1960         aio_context_acquire(aio_context);
1961 
1962         job_cancel_sync(&state->job->job);
1963 
1964         aio_context_release(aio_context);
1965     }
1966 }
1967 
1968 static void drive_backup_clean(BlkActionState *common)
1969 {
1970     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1971     AioContext *aio_context;
1972 
1973     if (!state->bs) {
1974         return;
1975     }
1976 
1977     aio_context = bdrv_get_aio_context(state->bs);
1978     aio_context_acquire(aio_context);
1979 
1980     bdrv_drained_end(state->bs);
1981 
1982     aio_context_release(aio_context);
1983 }
1984 
1985 typedef struct BlockdevBackupState {
1986     BlkActionState common;
1987     BlockDriverState *bs;
1988     BlockJob *job;
1989 } BlockdevBackupState;
1990 
1991 static void blockdev_backup_prepare(BlkActionState *common, Error **errp)
1992 {
1993     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1994     BlockdevBackup *backup;
1995     BlockDriverState *bs;
1996     BlockDriverState *target_bs;
1997     AioContext *aio_context;
1998     AioContext *old_context;
1999     int ret;
2000 
2001     assert(common->action->type == TRANSACTION_ACTION_KIND_BLOCKDEV_BACKUP);
2002     backup = common->action->u.blockdev_backup.data;
2003 
2004     bs = bdrv_lookup_bs(backup->device, backup->device, errp);
2005     if (!bs) {
2006         return;
2007     }
2008 
2009     target_bs = bdrv_lookup_bs(backup->target, backup->target, errp);
2010     if (!target_bs) {
2011         return;
2012     }
2013 
2014     /* Honor bdrv_try_set_aio_context() context acquisition requirements. */
2015     aio_context = bdrv_get_aio_context(bs);
2016     old_context = bdrv_get_aio_context(target_bs);
2017     aio_context_acquire(old_context);
2018 
2019     ret = bdrv_try_set_aio_context(target_bs, aio_context, errp);
2020     if (ret < 0) {
2021         aio_context_release(old_context);
2022         return;
2023     }
2024 
2025     aio_context_release(old_context);
2026     aio_context_acquire(aio_context);
2027     state->bs = bs;
2028 
2029     /* Paired with .clean() */
2030     bdrv_drained_begin(state->bs);
2031 
2032     state->job = do_backup_common(qapi_BlockdevBackup_base(backup),
2033                                   bs, target_bs, aio_context,
2034                                   common->block_job_txn, errp);
2035 
2036     aio_context_release(aio_context);
2037 }
2038 
2039 static void blockdev_backup_commit(BlkActionState *common)
2040 {
2041     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
2042     AioContext *aio_context;
2043 
2044     aio_context = bdrv_get_aio_context(state->bs);
2045     aio_context_acquire(aio_context);
2046 
2047     assert(state->job);
2048     job_start(&state->job->job);
2049 
2050     aio_context_release(aio_context);
2051 }
2052 
2053 static void blockdev_backup_abort(BlkActionState *common)
2054 {
2055     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
2056 
2057     if (state->job) {
2058         AioContext *aio_context;
2059 
2060         aio_context = bdrv_get_aio_context(state->bs);
2061         aio_context_acquire(aio_context);
2062 
2063         job_cancel_sync(&state->job->job);
2064 
2065         aio_context_release(aio_context);
2066     }
2067 }
2068 
2069 static void blockdev_backup_clean(BlkActionState *common)
2070 {
2071     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
2072     AioContext *aio_context;
2073 
2074     if (!state->bs) {
2075         return;
2076     }
2077 
2078     aio_context = bdrv_get_aio_context(state->bs);
2079     aio_context_acquire(aio_context);
2080 
2081     bdrv_drained_end(state->bs);
2082 
2083     aio_context_release(aio_context);
2084 }
2085 
2086 typedef struct BlockDirtyBitmapState {
2087     BlkActionState common;
2088     BdrvDirtyBitmap *bitmap;
2089     BlockDriverState *bs;
2090     HBitmap *backup;
2091     bool prepared;
2092     bool was_enabled;
2093 } BlockDirtyBitmapState;
2094 
2095 static void block_dirty_bitmap_add_prepare(BlkActionState *common,
2096                                            Error **errp)
2097 {
2098     Error *local_err = NULL;
2099     BlockDirtyBitmapAdd *action;
2100     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2101                                              common, common);
2102 
2103     if (action_check_completion_mode(common, errp) < 0) {
2104         return;
2105     }
2106 
2107     action = common->action->u.block_dirty_bitmap_add.data;
2108     /* AIO context taken and released within qmp_block_dirty_bitmap_add */
2109     qmp_block_dirty_bitmap_add(action->node, action->name,
2110                                action->has_granularity, action->granularity,
2111                                action->has_persistent, action->persistent,
2112                                action->has_disabled, action->disabled,
2113                                &local_err);
2114 
2115     if (!local_err) {
2116         state->prepared = true;
2117     } else {
2118         error_propagate(errp, local_err);
2119     }
2120 }
2121 
2122 static void block_dirty_bitmap_add_abort(BlkActionState *common)
2123 {
2124     BlockDirtyBitmapAdd *action;
2125     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2126                                              common, common);
2127 
2128     action = common->action->u.block_dirty_bitmap_add.data;
2129     /* Should not be able to fail: IF the bitmap was added via .prepare(),
2130      * then the node reference and bitmap name must have been valid.
2131      */
2132     if (state->prepared) {
2133         qmp_block_dirty_bitmap_remove(action->node, action->name, &error_abort);
2134     }
2135 }
2136 
2137 static void block_dirty_bitmap_clear_prepare(BlkActionState *common,
2138                                              Error **errp)
2139 {
2140     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2141                                              common, common);
2142     BlockDirtyBitmap *action;
2143 
2144     if (action_check_completion_mode(common, errp) < 0) {
2145         return;
2146     }
2147 
2148     action = common->action->u.block_dirty_bitmap_clear.data;
2149     state->bitmap = block_dirty_bitmap_lookup(action->node,
2150                                               action->name,
2151                                               &state->bs,
2152                                               errp);
2153     if (!state->bitmap) {
2154         return;
2155     }
2156 
2157     if (bdrv_dirty_bitmap_check(state->bitmap, BDRV_BITMAP_DEFAULT, errp)) {
2158         return;
2159     }
2160 
2161     bdrv_clear_dirty_bitmap(state->bitmap, &state->backup);
2162 }
2163 
2164 static void block_dirty_bitmap_restore(BlkActionState *common)
2165 {
2166     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2167                                              common, common);
2168 
2169     if (state->backup) {
2170         bdrv_restore_dirty_bitmap(state->bitmap, state->backup);
2171     }
2172 }
2173 
2174 static void block_dirty_bitmap_free_backup(BlkActionState *common)
2175 {
2176     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2177                                              common, common);
2178 
2179     hbitmap_free(state->backup);
2180 }
2181 
2182 static void block_dirty_bitmap_enable_prepare(BlkActionState *common,
2183                                               Error **errp)
2184 {
2185     BlockDirtyBitmap *action;
2186     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2187                                              common, common);
2188 
2189     if (action_check_completion_mode(common, errp) < 0) {
2190         return;
2191     }
2192 
2193     action = common->action->u.block_dirty_bitmap_enable.data;
2194     state->bitmap = block_dirty_bitmap_lookup(action->node,
2195                                               action->name,
2196                                               NULL,
2197                                               errp);
2198     if (!state->bitmap) {
2199         return;
2200     }
2201 
2202     if (bdrv_dirty_bitmap_check(state->bitmap, BDRV_BITMAP_ALLOW_RO, errp)) {
2203         return;
2204     }
2205 
2206     state->was_enabled = bdrv_dirty_bitmap_enabled(state->bitmap);
2207     bdrv_enable_dirty_bitmap(state->bitmap);
2208 }
2209 
2210 static void block_dirty_bitmap_enable_abort(BlkActionState *common)
2211 {
2212     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2213                                              common, common);
2214 
2215     if (!state->was_enabled) {
2216         bdrv_disable_dirty_bitmap(state->bitmap);
2217     }
2218 }
2219 
2220 static void block_dirty_bitmap_disable_prepare(BlkActionState *common,
2221                                                Error **errp)
2222 {
2223     BlockDirtyBitmap *action;
2224     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2225                                              common, common);
2226 
2227     if (action_check_completion_mode(common, errp) < 0) {
2228         return;
2229     }
2230 
2231     action = common->action->u.block_dirty_bitmap_disable.data;
2232     state->bitmap = block_dirty_bitmap_lookup(action->node,
2233                                               action->name,
2234                                               NULL,
2235                                               errp);
2236     if (!state->bitmap) {
2237         return;
2238     }
2239 
2240     if (bdrv_dirty_bitmap_check(state->bitmap, BDRV_BITMAP_ALLOW_RO, errp)) {
2241         return;
2242     }
2243 
2244     state->was_enabled = bdrv_dirty_bitmap_enabled(state->bitmap);
2245     bdrv_disable_dirty_bitmap(state->bitmap);
2246 }
2247 
2248 static void block_dirty_bitmap_disable_abort(BlkActionState *common)
2249 {
2250     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2251                                              common, common);
2252 
2253     if (state->was_enabled) {
2254         bdrv_enable_dirty_bitmap(state->bitmap);
2255     }
2256 }
2257 
2258 static BdrvDirtyBitmap *do_block_dirty_bitmap_merge(
2259         const char *node, const char *target,
2260         BlockDirtyBitmapMergeSourceList *bitmaps,
2261         HBitmap **backup, Error **errp);
2262 
2263 static void block_dirty_bitmap_merge_prepare(BlkActionState *common,
2264                                              Error **errp)
2265 {
2266     BlockDirtyBitmapMerge *action;
2267     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2268                                              common, common);
2269 
2270     if (action_check_completion_mode(common, errp) < 0) {
2271         return;
2272     }
2273 
2274     action = common->action->u.block_dirty_bitmap_merge.data;
2275 
2276     state->bitmap = do_block_dirty_bitmap_merge(action->node, action->target,
2277                                                 action->bitmaps, &state->backup,
2278                                                 errp);
2279 }
2280 
2281 static BdrvDirtyBitmap *do_block_dirty_bitmap_remove(
2282         const char *node, const char *name, bool release,
2283         BlockDriverState **bitmap_bs, Error **errp);
2284 
2285 static void block_dirty_bitmap_remove_prepare(BlkActionState *common,
2286                                               Error **errp)
2287 {
2288     BlockDirtyBitmap *action;
2289     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2290                                              common, common);
2291 
2292     if (action_check_completion_mode(common, errp) < 0) {
2293         return;
2294     }
2295 
2296     action = common->action->u.block_dirty_bitmap_remove.data;
2297 
2298     state->bitmap = do_block_dirty_bitmap_remove(action->node, action->name,
2299                                                  false, &state->bs, errp);
2300     if (state->bitmap) {
2301         bdrv_dirty_bitmap_skip_store(state->bitmap, true);
2302         bdrv_dirty_bitmap_set_busy(state->bitmap, true);
2303     }
2304 }
2305 
2306 static void block_dirty_bitmap_remove_abort(BlkActionState *common)
2307 {
2308     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2309                                              common, common);
2310 
2311     if (state->bitmap) {
2312         bdrv_dirty_bitmap_skip_store(state->bitmap, false);
2313         bdrv_dirty_bitmap_set_busy(state->bitmap, false);
2314     }
2315 }
2316 
2317 static void block_dirty_bitmap_remove_commit(BlkActionState *common)
2318 {
2319     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2320                                              common, common);
2321 
2322     bdrv_dirty_bitmap_set_busy(state->bitmap, false);
2323     bdrv_release_dirty_bitmap(state->bitmap);
2324 }
2325 
2326 static void abort_prepare(BlkActionState *common, Error **errp)
2327 {
2328     error_setg(errp, "Transaction aborted using Abort action");
2329 }
2330 
2331 static void abort_commit(BlkActionState *common)
2332 {
2333     g_assert_not_reached(); /* this action never succeeds */
2334 }
2335 
2336 static const BlkActionOps actions[] = {
2337     [TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT] = {
2338         .instance_size = sizeof(ExternalSnapshotState),
2339         .prepare  = external_snapshot_prepare,
2340         .commit   = external_snapshot_commit,
2341         .abort = external_snapshot_abort,
2342         .clean = external_snapshot_clean,
2343     },
2344     [TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC] = {
2345         .instance_size = sizeof(ExternalSnapshotState),
2346         .prepare  = external_snapshot_prepare,
2347         .commit   = external_snapshot_commit,
2348         .abort = external_snapshot_abort,
2349         .clean = external_snapshot_clean,
2350     },
2351     [TRANSACTION_ACTION_KIND_DRIVE_BACKUP] = {
2352         .instance_size = sizeof(DriveBackupState),
2353         .prepare = drive_backup_prepare,
2354         .commit = drive_backup_commit,
2355         .abort = drive_backup_abort,
2356         .clean = drive_backup_clean,
2357     },
2358     [TRANSACTION_ACTION_KIND_BLOCKDEV_BACKUP] = {
2359         .instance_size = sizeof(BlockdevBackupState),
2360         .prepare = blockdev_backup_prepare,
2361         .commit = blockdev_backup_commit,
2362         .abort = blockdev_backup_abort,
2363         .clean = blockdev_backup_clean,
2364     },
2365     [TRANSACTION_ACTION_KIND_ABORT] = {
2366         .instance_size = sizeof(BlkActionState),
2367         .prepare = abort_prepare,
2368         .commit = abort_commit,
2369     },
2370     [TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC] = {
2371         .instance_size = sizeof(InternalSnapshotState),
2372         .prepare  = internal_snapshot_prepare,
2373         .abort = internal_snapshot_abort,
2374         .clean = internal_snapshot_clean,
2375     },
2376     [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_ADD] = {
2377         .instance_size = sizeof(BlockDirtyBitmapState),
2378         .prepare = block_dirty_bitmap_add_prepare,
2379         .abort = block_dirty_bitmap_add_abort,
2380     },
2381     [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_CLEAR] = {
2382         .instance_size = sizeof(BlockDirtyBitmapState),
2383         .prepare = block_dirty_bitmap_clear_prepare,
2384         .commit = block_dirty_bitmap_free_backup,
2385         .abort = block_dirty_bitmap_restore,
2386     },
2387     [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_ENABLE] = {
2388         .instance_size = sizeof(BlockDirtyBitmapState),
2389         .prepare = block_dirty_bitmap_enable_prepare,
2390         .abort = block_dirty_bitmap_enable_abort,
2391     },
2392     [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_DISABLE] = {
2393         .instance_size = sizeof(BlockDirtyBitmapState),
2394         .prepare = block_dirty_bitmap_disable_prepare,
2395         .abort = block_dirty_bitmap_disable_abort,
2396     },
2397     [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_MERGE] = {
2398         .instance_size = sizeof(BlockDirtyBitmapState),
2399         .prepare = block_dirty_bitmap_merge_prepare,
2400         .commit = block_dirty_bitmap_free_backup,
2401         .abort = block_dirty_bitmap_restore,
2402     },
2403     [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_REMOVE] = {
2404         .instance_size = sizeof(BlockDirtyBitmapState),
2405         .prepare = block_dirty_bitmap_remove_prepare,
2406         .commit = block_dirty_bitmap_remove_commit,
2407         .abort = block_dirty_bitmap_remove_abort,
2408     },
2409     /* Where are transactions for MIRROR, COMMIT and STREAM?
2410      * Although these blockjobs use transaction callbacks like the backup job,
2411      * these jobs do not necessarily adhere to transaction semantics.
2412      * These jobs may not fully undo all of their actions on abort, nor do they
2413      * necessarily work in transactions with more than one job in them.
2414      */
2415 };
2416 
2417 /**
2418  * Allocate a TransactionProperties structure if necessary, and fill
2419  * that structure with desired defaults if they are unset.
2420  */
2421 static TransactionProperties *get_transaction_properties(
2422     TransactionProperties *props)
2423 {
2424     if (!props) {
2425         props = g_new0(TransactionProperties, 1);
2426     }
2427 
2428     if (!props->has_completion_mode) {
2429         props->has_completion_mode = true;
2430         props->completion_mode = ACTION_COMPLETION_MODE_INDIVIDUAL;
2431     }
2432 
2433     return props;
2434 }
2435 
2436 /*
2437  * 'Atomic' group operations.  The operations are performed as a set, and if
2438  * any fail then we roll back all operations in the group.
2439  */
2440 void qmp_transaction(TransactionActionList *dev_list,
2441                      bool has_props,
2442                      struct TransactionProperties *props,
2443                      Error **errp)
2444 {
2445     TransactionActionList *dev_entry = dev_list;
2446     JobTxn *block_job_txn = NULL;
2447     BlkActionState *state, *next;
2448     Error *local_err = NULL;
2449 
2450     QTAILQ_HEAD(, BlkActionState) snap_bdrv_states;
2451     QTAILQ_INIT(&snap_bdrv_states);
2452 
2453     /* Does this transaction get canceled as a group on failure?
2454      * If not, we don't really need to make a JobTxn.
2455      */
2456     props = get_transaction_properties(props);
2457     if (props->completion_mode != ACTION_COMPLETION_MODE_INDIVIDUAL) {
2458         block_job_txn = job_txn_new();
2459     }
2460 
2461     /* drain all i/o before any operations */
2462     bdrv_drain_all();
2463 
2464     /* We don't do anything in this loop that commits us to the operations */
2465     while (NULL != dev_entry) {
2466         TransactionAction *dev_info = NULL;
2467         const BlkActionOps *ops;
2468 
2469         dev_info = dev_entry->value;
2470         dev_entry = dev_entry->next;
2471 
2472         assert(dev_info->type < ARRAY_SIZE(actions));
2473 
2474         ops = &actions[dev_info->type];
2475         assert(ops->instance_size > 0);
2476 
2477         state = g_malloc0(ops->instance_size);
2478         state->ops = ops;
2479         state->action = dev_info;
2480         state->block_job_txn = block_job_txn;
2481         state->txn_props = props;
2482         QTAILQ_INSERT_TAIL(&snap_bdrv_states, state, entry);
2483 
2484         state->ops->prepare(state, &local_err);
2485         if (local_err) {
2486             error_propagate(errp, local_err);
2487             goto delete_and_fail;
2488         }
2489     }
2490 
2491     QTAILQ_FOREACH(state, &snap_bdrv_states, entry) {
2492         if (state->ops->commit) {
2493             state->ops->commit(state);
2494         }
2495     }
2496 
2497     /* success */
2498     goto exit;
2499 
2500 delete_and_fail:
2501     /* failure, and it is all-or-none; roll back all operations */
2502     QTAILQ_FOREACH_REVERSE(state, &snap_bdrv_states, entry) {
2503         if (state->ops->abort) {
2504             state->ops->abort(state);
2505         }
2506     }
2507 exit:
2508     QTAILQ_FOREACH_SAFE(state, &snap_bdrv_states, entry, next) {
2509         if (state->ops->clean) {
2510             state->ops->clean(state);
2511         }
2512         g_free(state);
2513     }
2514     if (!has_props) {
2515         qapi_free_TransactionProperties(props);
2516     }
2517     job_txn_unref(block_job_txn);
2518 }
2519 
2520 void qmp_eject(bool has_device, const char *device,
2521                bool has_id, const char *id,
2522                bool has_force, bool force, Error **errp)
2523 {
2524     Error *local_err = NULL;
2525     int rc;
2526 
2527     if (!has_force) {
2528         force = false;
2529     }
2530 
2531     rc = do_open_tray(has_device ? device : NULL,
2532                       has_id ? id : NULL,
2533                       force, &local_err);
2534     if (rc && rc != -ENOSYS) {
2535         error_propagate(errp, local_err);
2536         return;
2537     }
2538     error_free(local_err);
2539 
2540     blockdev_remove_medium(has_device, device, has_id, id, errp);
2541 }
2542 
2543 void qmp_block_passwd(bool has_device, const char *device,
2544                       bool has_node_name, const char *node_name,
2545                       const char *password, Error **errp)
2546 {
2547     error_setg(errp,
2548                "Setting block passwords directly is no longer supported");
2549 }
2550 
2551 /*
2552  * Attempt to open the tray of @device.
2553  * If @force, ignore its tray lock.
2554  * Else, if the tray is locked, don't open it, but ask the guest to open it.
2555  * On error, store an error through @errp and return -errno.
2556  * If @device does not exist, return -ENODEV.
2557  * If it has no removable media, return -ENOTSUP.
2558  * If it has no tray, return -ENOSYS.
2559  * If the guest was asked to open the tray, return -EINPROGRESS.
2560  * Else, return 0.
2561  */
2562 static int do_open_tray(const char *blk_name, const char *qdev_id,
2563                         bool force, Error **errp)
2564 {
2565     BlockBackend *blk;
2566     const char *device = qdev_id ?: blk_name;
2567     bool locked;
2568 
2569     blk = qmp_get_blk(blk_name, qdev_id, errp);
2570     if (!blk) {
2571         return -ENODEV;
2572     }
2573 
2574     if (!blk_dev_has_removable_media(blk)) {
2575         error_setg(errp, "Device '%s' is not removable", device);
2576         return -ENOTSUP;
2577     }
2578 
2579     if (!blk_dev_has_tray(blk)) {
2580         error_setg(errp, "Device '%s' does not have a tray", device);
2581         return -ENOSYS;
2582     }
2583 
2584     if (blk_dev_is_tray_open(blk)) {
2585         return 0;
2586     }
2587 
2588     locked = blk_dev_is_medium_locked(blk);
2589     if (locked) {
2590         blk_dev_eject_request(blk, force);
2591     }
2592 
2593     if (!locked || force) {
2594         blk_dev_change_media_cb(blk, false, &error_abort);
2595     }
2596 
2597     if (locked && !force) {
2598         error_setg(errp, "Device '%s' is locked and force was not specified, "
2599                    "wait for tray to open and try again", device);
2600         return -EINPROGRESS;
2601     }
2602 
2603     return 0;
2604 }
2605 
2606 void qmp_blockdev_open_tray(bool has_device, const char *device,
2607                             bool has_id, const char *id,
2608                             bool has_force, bool force,
2609                             Error **errp)
2610 {
2611     Error *local_err = NULL;
2612     int rc;
2613 
2614     if (!has_force) {
2615         force = false;
2616     }
2617     rc = do_open_tray(has_device ? device : NULL,
2618                       has_id ? id : NULL,
2619                       force, &local_err);
2620     if (rc && rc != -ENOSYS && rc != -EINPROGRESS) {
2621         error_propagate(errp, local_err);
2622         return;
2623     }
2624     error_free(local_err);
2625 }
2626 
2627 void qmp_blockdev_close_tray(bool has_device, const char *device,
2628                              bool has_id, const char *id,
2629                              Error **errp)
2630 {
2631     BlockBackend *blk;
2632     Error *local_err = NULL;
2633 
2634     device = has_device ? device : NULL;
2635     id = has_id ? id : NULL;
2636 
2637     blk = qmp_get_blk(device, id, errp);
2638     if (!blk) {
2639         return;
2640     }
2641 
2642     if (!blk_dev_has_removable_media(blk)) {
2643         error_setg(errp, "Device '%s' is not removable", device ?: id);
2644         return;
2645     }
2646 
2647     if (!blk_dev_has_tray(blk)) {
2648         /* Ignore this command on tray-less devices */
2649         return;
2650     }
2651 
2652     if (!blk_dev_is_tray_open(blk)) {
2653         return;
2654     }
2655 
2656     blk_dev_change_media_cb(blk, true, &local_err);
2657     if (local_err) {
2658         error_propagate(errp, local_err);
2659         return;
2660     }
2661 }
2662 
2663 static void blockdev_remove_medium(bool has_device, const char *device,
2664                                    bool has_id, const char *id, Error **errp)
2665 {
2666     BlockBackend *blk;
2667     BlockDriverState *bs;
2668     AioContext *aio_context;
2669     bool has_attached_device;
2670 
2671     device = has_device ? device : NULL;
2672     id = has_id ? id : NULL;
2673 
2674     blk = qmp_get_blk(device, id, errp);
2675     if (!blk) {
2676         return;
2677     }
2678 
2679     /* For BBs without a device, we can exchange the BDS tree at will */
2680     has_attached_device = blk_get_attached_dev(blk);
2681 
2682     if (has_attached_device && !blk_dev_has_removable_media(blk)) {
2683         error_setg(errp, "Device '%s' is not removable", device ?: id);
2684         return;
2685     }
2686 
2687     if (has_attached_device && blk_dev_has_tray(blk) &&
2688         !blk_dev_is_tray_open(blk))
2689     {
2690         error_setg(errp, "Tray of device '%s' is not open", device ?: id);
2691         return;
2692     }
2693 
2694     bs = blk_bs(blk);
2695     if (!bs) {
2696         return;
2697     }
2698 
2699     aio_context = bdrv_get_aio_context(bs);
2700     aio_context_acquire(aio_context);
2701 
2702     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_EJECT, errp)) {
2703         goto out;
2704     }
2705 
2706     blk_remove_bs(blk);
2707 
2708     if (!blk_dev_has_tray(blk)) {
2709         /* For tray-less devices, blockdev-open-tray is a no-op (or may not be
2710          * called at all); therefore, the medium needs to be ejected here.
2711          * Do it after blk_remove_bs() so blk_is_inserted(blk) returns the @load
2712          * value passed here (i.e. false). */
2713         blk_dev_change_media_cb(blk, false, &error_abort);
2714     }
2715 
2716 out:
2717     aio_context_release(aio_context);
2718 }
2719 
2720 void qmp_blockdev_remove_medium(const char *id, Error **errp)
2721 {
2722     blockdev_remove_medium(false, NULL, true, id, errp);
2723 }
2724 
2725 static void qmp_blockdev_insert_anon_medium(BlockBackend *blk,
2726                                             BlockDriverState *bs, Error **errp)
2727 {
2728     Error *local_err = NULL;
2729     bool has_device;
2730     int ret;
2731 
2732     /* For BBs without a device, we can exchange the BDS tree at will */
2733     has_device = blk_get_attached_dev(blk);
2734 
2735     if (has_device && !blk_dev_has_removable_media(blk)) {
2736         error_setg(errp, "Device is not removable");
2737         return;
2738     }
2739 
2740     if (has_device && blk_dev_has_tray(blk) && !blk_dev_is_tray_open(blk)) {
2741         error_setg(errp, "Tray of the device is not open");
2742         return;
2743     }
2744 
2745     if (blk_bs(blk)) {
2746         error_setg(errp, "There already is a medium in the device");
2747         return;
2748     }
2749 
2750     ret = blk_insert_bs(blk, bs, errp);
2751     if (ret < 0) {
2752         return;
2753     }
2754 
2755     if (!blk_dev_has_tray(blk)) {
2756         /* For tray-less devices, blockdev-close-tray is a no-op (or may not be
2757          * called at all); therefore, the medium needs to be pushed into the
2758          * slot here.
2759          * Do it after blk_insert_bs() so blk_is_inserted(blk) returns the @load
2760          * value passed here (i.e. true). */
2761         blk_dev_change_media_cb(blk, true, &local_err);
2762         if (local_err) {
2763             error_propagate(errp, local_err);
2764             blk_remove_bs(blk);
2765             return;
2766         }
2767     }
2768 }
2769 
2770 static void blockdev_insert_medium(bool has_device, const char *device,
2771                                    bool has_id, const char *id,
2772                                    const char *node_name, Error **errp)
2773 {
2774     BlockBackend *blk;
2775     BlockDriverState *bs;
2776 
2777     blk = qmp_get_blk(has_device ? device : NULL,
2778                       has_id ? id : NULL,
2779                       errp);
2780     if (!blk) {
2781         return;
2782     }
2783 
2784     bs = bdrv_find_node(node_name);
2785     if (!bs) {
2786         error_setg(errp, "Node '%s' not found", node_name);
2787         return;
2788     }
2789 
2790     if (bdrv_has_blk(bs)) {
2791         error_setg(errp, "Node '%s' is already in use", node_name);
2792         return;
2793     }
2794 
2795     qmp_blockdev_insert_anon_medium(blk, bs, errp);
2796 }
2797 
2798 void qmp_blockdev_insert_medium(const char *id, const char *node_name,
2799                                 Error **errp)
2800 {
2801     blockdev_insert_medium(false, NULL, true, id, node_name, errp);
2802 }
2803 
2804 void qmp_blockdev_change_medium(bool has_device, const char *device,
2805                                 bool has_id, const char *id,
2806                                 const char *filename,
2807                                 bool has_format, const char *format,
2808                                 bool has_read_only,
2809                                 BlockdevChangeReadOnlyMode read_only,
2810                                 Error **errp)
2811 {
2812     BlockBackend *blk;
2813     BlockDriverState *medium_bs = NULL;
2814     int bdrv_flags;
2815     bool detect_zeroes;
2816     int rc;
2817     QDict *options = NULL;
2818     Error *err = NULL;
2819 
2820     blk = qmp_get_blk(has_device ? device : NULL,
2821                       has_id ? id : NULL,
2822                       errp);
2823     if (!blk) {
2824         goto fail;
2825     }
2826 
2827     if (blk_bs(blk)) {
2828         blk_update_root_state(blk);
2829     }
2830 
2831     bdrv_flags = blk_get_open_flags_from_root_state(blk);
2832     bdrv_flags &= ~(BDRV_O_TEMPORARY | BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING |
2833         BDRV_O_PROTOCOL | BDRV_O_AUTO_RDONLY);
2834 
2835     if (!has_read_only) {
2836         read_only = BLOCKDEV_CHANGE_READ_ONLY_MODE_RETAIN;
2837     }
2838 
2839     switch (read_only) {
2840     case BLOCKDEV_CHANGE_READ_ONLY_MODE_RETAIN:
2841         break;
2842 
2843     case BLOCKDEV_CHANGE_READ_ONLY_MODE_READ_ONLY:
2844         bdrv_flags &= ~BDRV_O_RDWR;
2845         break;
2846 
2847     case BLOCKDEV_CHANGE_READ_ONLY_MODE_READ_WRITE:
2848         bdrv_flags |= BDRV_O_RDWR;
2849         break;
2850 
2851     default:
2852         abort();
2853     }
2854 
2855     options = qdict_new();
2856     detect_zeroes = blk_get_detect_zeroes_from_root_state(blk);
2857     qdict_put_str(options, "detect-zeroes", detect_zeroes ? "on" : "off");
2858 
2859     if (has_format) {
2860         qdict_put_str(options, "driver", format);
2861     }
2862 
2863     medium_bs = bdrv_open(filename, NULL, options, bdrv_flags, errp);
2864     if (!medium_bs) {
2865         goto fail;
2866     }
2867 
2868     rc = do_open_tray(has_device ? device : NULL,
2869                       has_id ? id : NULL,
2870                       false, &err);
2871     if (rc && rc != -ENOSYS) {
2872         error_propagate(errp, err);
2873         goto fail;
2874     }
2875     error_free(err);
2876     err = NULL;
2877 
2878     blockdev_remove_medium(has_device, device, has_id, id, &err);
2879     if (err) {
2880         error_propagate(errp, err);
2881         goto fail;
2882     }
2883 
2884     qmp_blockdev_insert_anon_medium(blk, medium_bs, &err);
2885     if (err) {
2886         error_propagate(errp, err);
2887         goto fail;
2888     }
2889 
2890     qmp_blockdev_close_tray(has_device, device, has_id, id, errp);
2891 
2892 fail:
2893     /* If the medium has been inserted, the device has its own reference, so
2894      * ours must be relinquished; and if it has not been inserted successfully,
2895      * the reference must be relinquished anyway */
2896     bdrv_unref(medium_bs);
2897 }
2898 
2899 /* throttling disk I/O limits */
2900 void qmp_block_set_io_throttle(BlockIOThrottle *arg, Error **errp)
2901 {
2902     ThrottleConfig cfg;
2903     BlockDriverState *bs;
2904     BlockBackend *blk;
2905     AioContext *aio_context;
2906 
2907     blk = qmp_get_blk(arg->has_device ? arg->device : NULL,
2908                       arg->has_id ? arg->id : NULL,
2909                       errp);
2910     if (!blk) {
2911         return;
2912     }
2913 
2914     aio_context = blk_get_aio_context(blk);
2915     aio_context_acquire(aio_context);
2916 
2917     bs = blk_bs(blk);
2918     if (!bs) {
2919         error_setg(errp, "Device has no medium");
2920         goto out;
2921     }
2922 
2923     throttle_config_init(&cfg);
2924     cfg.buckets[THROTTLE_BPS_TOTAL].avg = arg->bps;
2925     cfg.buckets[THROTTLE_BPS_READ].avg  = arg->bps_rd;
2926     cfg.buckets[THROTTLE_BPS_WRITE].avg = arg->bps_wr;
2927 
2928     cfg.buckets[THROTTLE_OPS_TOTAL].avg = arg->iops;
2929     cfg.buckets[THROTTLE_OPS_READ].avg  = arg->iops_rd;
2930     cfg.buckets[THROTTLE_OPS_WRITE].avg = arg->iops_wr;
2931 
2932     if (arg->has_bps_max) {
2933         cfg.buckets[THROTTLE_BPS_TOTAL].max = arg->bps_max;
2934     }
2935     if (arg->has_bps_rd_max) {
2936         cfg.buckets[THROTTLE_BPS_READ].max = arg->bps_rd_max;
2937     }
2938     if (arg->has_bps_wr_max) {
2939         cfg.buckets[THROTTLE_BPS_WRITE].max = arg->bps_wr_max;
2940     }
2941     if (arg->has_iops_max) {
2942         cfg.buckets[THROTTLE_OPS_TOTAL].max = arg->iops_max;
2943     }
2944     if (arg->has_iops_rd_max) {
2945         cfg.buckets[THROTTLE_OPS_READ].max = arg->iops_rd_max;
2946     }
2947     if (arg->has_iops_wr_max) {
2948         cfg.buckets[THROTTLE_OPS_WRITE].max = arg->iops_wr_max;
2949     }
2950 
2951     if (arg->has_bps_max_length) {
2952         cfg.buckets[THROTTLE_BPS_TOTAL].burst_length = arg->bps_max_length;
2953     }
2954     if (arg->has_bps_rd_max_length) {
2955         cfg.buckets[THROTTLE_BPS_READ].burst_length = arg->bps_rd_max_length;
2956     }
2957     if (arg->has_bps_wr_max_length) {
2958         cfg.buckets[THROTTLE_BPS_WRITE].burst_length = arg->bps_wr_max_length;
2959     }
2960     if (arg->has_iops_max_length) {
2961         cfg.buckets[THROTTLE_OPS_TOTAL].burst_length = arg->iops_max_length;
2962     }
2963     if (arg->has_iops_rd_max_length) {
2964         cfg.buckets[THROTTLE_OPS_READ].burst_length = arg->iops_rd_max_length;
2965     }
2966     if (arg->has_iops_wr_max_length) {
2967         cfg.buckets[THROTTLE_OPS_WRITE].burst_length = arg->iops_wr_max_length;
2968     }
2969 
2970     if (arg->has_iops_size) {
2971         cfg.op_size = arg->iops_size;
2972     }
2973 
2974     if (!throttle_is_valid(&cfg, errp)) {
2975         goto out;
2976     }
2977 
2978     if (throttle_enabled(&cfg)) {
2979         /* Enable I/O limits if they're not enabled yet, otherwise
2980          * just update the throttling group. */
2981         if (!blk_get_public(blk)->throttle_group_member.throttle_state) {
2982             blk_io_limits_enable(blk,
2983                                  arg->has_group ? arg->group :
2984                                  arg->has_device ? arg->device :
2985                                  arg->id);
2986         } else if (arg->has_group) {
2987             blk_io_limits_update_group(blk, arg->group);
2988         }
2989         /* Set the new throttling configuration */
2990         blk_set_io_limits(blk, &cfg);
2991     } else if (blk_get_public(blk)->throttle_group_member.throttle_state) {
2992         /* If all throttling settings are set to 0, disable I/O limits */
2993         blk_io_limits_disable(blk);
2994     }
2995 
2996 out:
2997     aio_context_release(aio_context);
2998 }
2999 
3000 void qmp_block_dirty_bitmap_add(const char *node, const char *name,
3001                                 bool has_granularity, uint32_t granularity,
3002                                 bool has_persistent, bool persistent,
3003                                 bool has_disabled, bool disabled,
3004                                 Error **errp)
3005 {
3006     BlockDriverState *bs;
3007     BdrvDirtyBitmap *bitmap;
3008     AioContext *aio_context;
3009 
3010     if (!name || name[0] == '\0') {
3011         error_setg(errp, "Bitmap name cannot be empty");
3012         return;
3013     }
3014 
3015     bs = bdrv_lookup_bs(node, node, errp);
3016     if (!bs) {
3017         return;
3018     }
3019 
3020     aio_context = bdrv_get_aio_context(bs);
3021     aio_context_acquire(aio_context);
3022 
3023     if (has_granularity) {
3024         if (granularity < 512 || !is_power_of_2(granularity)) {
3025             error_setg(errp, "Granularity must be power of 2 "
3026                              "and at least 512");
3027             goto out;
3028         }
3029     } else {
3030         /* Default to cluster size, if available: */
3031         granularity = bdrv_get_default_bitmap_granularity(bs);
3032     }
3033 
3034     if (!has_persistent) {
3035         persistent = false;
3036     }
3037 
3038     if (!has_disabled) {
3039         disabled = false;
3040     }
3041 
3042     if (persistent &&
3043         !bdrv_can_store_new_dirty_bitmap(bs, name, granularity, errp))
3044     {
3045         goto out;
3046     }
3047 
3048     bitmap = bdrv_create_dirty_bitmap(bs, granularity, name, errp);
3049     if (bitmap == NULL) {
3050         goto out;
3051     }
3052 
3053     if (disabled) {
3054         bdrv_disable_dirty_bitmap(bitmap);
3055     }
3056 
3057     bdrv_dirty_bitmap_set_persistence(bitmap, persistent);
3058 
3059 out:
3060     aio_context_release(aio_context);
3061 }
3062 
3063 static BdrvDirtyBitmap *do_block_dirty_bitmap_remove(
3064         const char *node, const char *name, bool release,
3065         BlockDriverState **bitmap_bs, Error **errp)
3066 {
3067     BlockDriverState *bs;
3068     BdrvDirtyBitmap *bitmap;
3069     AioContext *aio_context;
3070 
3071     bitmap = block_dirty_bitmap_lookup(node, name, &bs, errp);
3072     if (!bitmap || !bs) {
3073         return NULL;
3074     }
3075 
3076     aio_context = bdrv_get_aio_context(bs);
3077     aio_context_acquire(aio_context);
3078 
3079     if (bdrv_dirty_bitmap_check(bitmap, BDRV_BITMAP_BUSY | BDRV_BITMAP_RO,
3080                                 errp)) {
3081         aio_context_release(aio_context);
3082         return NULL;
3083     }
3084 
3085     if (bdrv_dirty_bitmap_get_persistence(bitmap) &&
3086         bdrv_remove_persistent_dirty_bitmap(bs, name, errp) < 0)
3087     {
3088         aio_context_release(aio_context);
3089         return NULL;
3090     }
3091 
3092     if (release) {
3093         bdrv_release_dirty_bitmap(bitmap);
3094     }
3095 
3096     if (bitmap_bs) {
3097         *bitmap_bs = bs;
3098     }
3099 
3100     aio_context_release(aio_context);
3101     return release ? NULL : bitmap;
3102 }
3103 
3104 void qmp_block_dirty_bitmap_remove(const char *node, const char *name,
3105                                    Error **errp)
3106 {
3107     do_block_dirty_bitmap_remove(node, name, true, NULL, errp);
3108 }
3109 
3110 /**
3111  * Completely clear a bitmap, for the purposes of synchronizing a bitmap
3112  * immediately after a full backup operation.
3113  */
3114 void qmp_block_dirty_bitmap_clear(const char *node, const char *name,
3115                                   Error **errp)
3116 {
3117     BdrvDirtyBitmap *bitmap;
3118     BlockDriverState *bs;
3119 
3120     bitmap = block_dirty_bitmap_lookup(node, name, &bs, errp);
3121     if (!bitmap || !bs) {
3122         return;
3123     }
3124 
3125     if (bdrv_dirty_bitmap_check(bitmap, BDRV_BITMAP_DEFAULT, errp)) {
3126         return;
3127     }
3128 
3129     bdrv_clear_dirty_bitmap(bitmap, NULL);
3130 }
3131 
3132 void qmp_block_dirty_bitmap_enable(const char *node, const char *name,
3133                                    Error **errp)
3134 {
3135     BlockDriverState *bs;
3136     BdrvDirtyBitmap *bitmap;
3137 
3138     bitmap = block_dirty_bitmap_lookup(node, name, &bs, errp);
3139     if (!bitmap) {
3140         return;
3141     }
3142 
3143     if (bdrv_dirty_bitmap_check(bitmap, BDRV_BITMAP_ALLOW_RO, errp)) {
3144         return;
3145     }
3146 
3147     bdrv_enable_dirty_bitmap(bitmap);
3148 }
3149 
3150 void qmp_block_dirty_bitmap_disable(const char *node, const char *name,
3151                                     Error **errp)
3152 {
3153     BlockDriverState *bs;
3154     BdrvDirtyBitmap *bitmap;
3155 
3156     bitmap = block_dirty_bitmap_lookup(node, name, &bs, errp);
3157     if (!bitmap) {
3158         return;
3159     }
3160 
3161     if (bdrv_dirty_bitmap_check(bitmap, BDRV_BITMAP_ALLOW_RO, errp)) {
3162         return;
3163     }
3164 
3165     bdrv_disable_dirty_bitmap(bitmap);
3166 }
3167 
3168 static BdrvDirtyBitmap *do_block_dirty_bitmap_merge(
3169         const char *node, const char *target,
3170         BlockDirtyBitmapMergeSourceList *bitmaps,
3171         HBitmap **backup, Error **errp)
3172 {
3173     BlockDriverState *bs;
3174     BdrvDirtyBitmap *dst, *src, *anon;
3175     BlockDirtyBitmapMergeSourceList *lst;
3176     Error *local_err = NULL;
3177 
3178     dst = block_dirty_bitmap_lookup(node, target, &bs, errp);
3179     if (!dst) {
3180         return NULL;
3181     }
3182 
3183     anon = bdrv_create_dirty_bitmap(bs, bdrv_dirty_bitmap_granularity(dst),
3184                                     NULL, errp);
3185     if (!anon) {
3186         return NULL;
3187     }
3188 
3189     for (lst = bitmaps; lst; lst = lst->next) {
3190         switch (lst->value->type) {
3191             const char *name, *node;
3192         case QTYPE_QSTRING:
3193             name = lst->value->u.local;
3194             src = bdrv_find_dirty_bitmap(bs, name);
3195             if (!src) {
3196                 error_setg(errp, "Dirty bitmap '%s' not found", name);
3197                 dst = NULL;
3198                 goto out;
3199             }
3200             break;
3201         case QTYPE_QDICT:
3202             node = lst->value->u.external.node;
3203             name = lst->value->u.external.name;
3204             src = block_dirty_bitmap_lookup(node, name, NULL, errp);
3205             if (!src) {
3206                 dst = NULL;
3207                 goto out;
3208             }
3209             break;
3210         default:
3211             abort();
3212         }
3213 
3214         bdrv_merge_dirty_bitmap(anon, src, NULL, &local_err);
3215         if (local_err) {
3216             error_propagate(errp, local_err);
3217             dst = NULL;
3218             goto out;
3219         }
3220     }
3221 
3222     /* Merge into dst; dst is unchanged on failure. */
3223     bdrv_merge_dirty_bitmap(dst, anon, backup, errp);
3224 
3225  out:
3226     bdrv_release_dirty_bitmap(anon);
3227     return dst;
3228 }
3229 
3230 void qmp_block_dirty_bitmap_merge(const char *node, const char *target,
3231                                   BlockDirtyBitmapMergeSourceList *bitmaps,
3232                                   Error **errp)
3233 {
3234     do_block_dirty_bitmap_merge(node, target, bitmaps, NULL, errp);
3235 }
3236 
3237 BlockDirtyBitmapSha256 *qmp_x_debug_block_dirty_bitmap_sha256(const char *node,
3238                                                               const char *name,
3239                                                               Error **errp)
3240 {
3241     BdrvDirtyBitmap *bitmap;
3242     BlockDriverState *bs;
3243     BlockDirtyBitmapSha256 *ret = NULL;
3244     char *sha256;
3245 
3246     bitmap = block_dirty_bitmap_lookup(node, name, &bs, errp);
3247     if (!bitmap || !bs) {
3248         return NULL;
3249     }
3250 
3251     sha256 = bdrv_dirty_bitmap_sha256(bitmap, errp);
3252     if (sha256 == NULL) {
3253         return NULL;
3254     }
3255 
3256     ret = g_new(BlockDirtyBitmapSha256, 1);
3257     ret->sha256 = sha256;
3258 
3259     return ret;
3260 }
3261 
3262 void hmp_drive_del(Monitor *mon, const QDict *qdict)
3263 {
3264     const char *id = qdict_get_str(qdict, "id");
3265     BlockBackend *blk;
3266     BlockDriverState *bs;
3267     AioContext *aio_context;
3268     Error *local_err = NULL;
3269 
3270     bs = bdrv_find_node(id);
3271     if (bs) {
3272         qmp_blockdev_del(id, &local_err);
3273         if (local_err) {
3274             error_report_err(local_err);
3275         }
3276         return;
3277     }
3278 
3279     blk = blk_by_name(id);
3280     if (!blk) {
3281         error_report("Device '%s' not found", id);
3282         return;
3283     }
3284 
3285     if (!blk_legacy_dinfo(blk)) {
3286         error_report("Deleting device added with blockdev-add"
3287                      " is not supported");
3288         return;
3289     }
3290 
3291     aio_context = blk_get_aio_context(blk);
3292     aio_context_acquire(aio_context);
3293 
3294     bs = blk_bs(blk);
3295     if (bs) {
3296         if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_DRIVE_DEL, &local_err)) {
3297             error_report_err(local_err);
3298             aio_context_release(aio_context);
3299             return;
3300         }
3301 
3302         blk_remove_bs(blk);
3303     }
3304 
3305     /* Make the BlockBackend and the attached BlockDriverState anonymous */
3306     monitor_remove_blk(blk);
3307 
3308     /* If this BlockBackend has a device attached to it, its refcount will be
3309      * decremented when the device is removed; otherwise we have to do so here.
3310      */
3311     if (blk_get_attached_dev(blk)) {
3312         /* Further I/O must not pause the guest */
3313         blk_set_on_error(blk, BLOCKDEV_ON_ERROR_REPORT,
3314                          BLOCKDEV_ON_ERROR_REPORT);
3315     } else {
3316         blk_unref(blk);
3317     }
3318 
3319     aio_context_release(aio_context);
3320 }
3321 
3322 void qmp_block_resize(bool has_device, const char *device,
3323                       bool has_node_name, const char *node_name,
3324                       int64_t size, Error **errp)
3325 {
3326     Error *local_err = NULL;
3327     BlockBackend *blk = NULL;
3328     BlockDriverState *bs;
3329     AioContext *aio_context;
3330     int ret;
3331 
3332     bs = bdrv_lookup_bs(has_device ? device : NULL,
3333                         has_node_name ? node_name : NULL,
3334                         &local_err);
3335     if (local_err) {
3336         error_propagate(errp, local_err);
3337         return;
3338     }
3339 
3340     aio_context = bdrv_get_aio_context(bs);
3341     aio_context_acquire(aio_context);
3342 
3343     if (!bdrv_is_first_non_filter(bs)) {
3344         error_setg(errp, QERR_FEATURE_DISABLED, "resize");
3345         goto out;
3346     }
3347 
3348     if (size < 0) {
3349         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "size", "a >0 size");
3350         goto out;
3351     }
3352 
3353     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_RESIZE, NULL)) {
3354         error_setg(errp, QERR_DEVICE_IN_USE, device);
3355         goto out;
3356     }
3357 
3358     blk = blk_new(bdrv_get_aio_context(bs), BLK_PERM_RESIZE, BLK_PERM_ALL);
3359     ret = blk_insert_bs(blk, bs, errp);
3360     if (ret < 0) {
3361         goto out;
3362     }
3363 
3364     bdrv_drained_begin(bs);
3365     ret = blk_truncate(blk, size, false, PREALLOC_MODE_OFF, errp);
3366     bdrv_drained_end(bs);
3367 
3368 out:
3369     blk_unref(blk);
3370     aio_context_release(aio_context);
3371 }
3372 
3373 void qmp_block_stream(bool has_job_id, const char *job_id, const char *device,
3374                       bool has_base, const char *base,
3375                       bool has_base_node, const char *base_node,
3376                       bool has_backing_file, const char *backing_file,
3377                       bool has_speed, int64_t speed,
3378                       bool has_on_error, BlockdevOnError on_error,
3379                       bool has_auto_finalize, bool auto_finalize,
3380                       bool has_auto_dismiss, bool auto_dismiss,
3381                       Error **errp)
3382 {
3383     BlockDriverState *bs, *iter;
3384     BlockDriverState *base_bs = NULL;
3385     AioContext *aio_context;
3386     Error *local_err = NULL;
3387     const char *base_name = NULL;
3388     int job_flags = JOB_DEFAULT;
3389 
3390     if (!has_on_error) {
3391         on_error = BLOCKDEV_ON_ERROR_REPORT;
3392     }
3393 
3394     bs = bdrv_lookup_bs(device, device, errp);
3395     if (!bs) {
3396         return;
3397     }
3398 
3399     aio_context = bdrv_get_aio_context(bs);
3400     aio_context_acquire(aio_context);
3401 
3402     if (has_base && has_base_node) {
3403         error_setg(errp, "'base' and 'base-node' cannot be specified "
3404                    "at the same time");
3405         goto out;
3406     }
3407 
3408     if (has_base) {
3409         base_bs = bdrv_find_backing_image(bs, base);
3410         if (base_bs == NULL) {
3411             error_setg(errp, QERR_BASE_NOT_FOUND, base);
3412             goto out;
3413         }
3414         assert(bdrv_get_aio_context(base_bs) == aio_context);
3415         base_name = base;
3416     }
3417 
3418     if (has_base_node) {
3419         base_bs = bdrv_lookup_bs(NULL, base_node, errp);
3420         if (!base_bs) {
3421             goto out;
3422         }
3423         if (bs == base_bs || !bdrv_chain_contains(bs, base_bs)) {
3424             error_setg(errp, "Node '%s' is not a backing image of '%s'",
3425                        base_node, device);
3426             goto out;
3427         }
3428         assert(bdrv_get_aio_context(base_bs) == aio_context);
3429         bdrv_refresh_filename(base_bs);
3430         base_name = base_bs->filename;
3431     }
3432 
3433     /* Check for op blockers in the whole chain between bs and base */
3434     for (iter = bs; iter && iter != base_bs; iter = backing_bs(iter)) {
3435         if (bdrv_op_is_blocked(iter, BLOCK_OP_TYPE_STREAM, errp)) {
3436             goto out;
3437         }
3438     }
3439 
3440     /* if we are streaming the entire chain, the result will have no backing
3441      * file, and specifying one is therefore an error */
3442     if (base_bs == NULL && has_backing_file) {
3443         error_setg(errp, "backing file specified, but streaming the "
3444                          "entire chain");
3445         goto out;
3446     }
3447 
3448     /* backing_file string overrides base bs filename */
3449     base_name = has_backing_file ? backing_file : base_name;
3450 
3451     if (has_auto_finalize && !auto_finalize) {
3452         job_flags |= JOB_MANUAL_FINALIZE;
3453     }
3454     if (has_auto_dismiss && !auto_dismiss) {
3455         job_flags |= JOB_MANUAL_DISMISS;
3456     }
3457 
3458     stream_start(has_job_id ? job_id : NULL, bs, base_bs, base_name,
3459                  job_flags, has_speed ? speed : 0, on_error, &local_err);
3460     if (local_err) {
3461         error_propagate(errp, local_err);
3462         goto out;
3463     }
3464 
3465     trace_qmp_block_stream(bs);
3466 
3467 out:
3468     aio_context_release(aio_context);
3469 }
3470 
3471 void qmp_block_commit(bool has_job_id, const char *job_id, const char *device,
3472                       bool has_base_node, const char *base_node,
3473                       bool has_base, const char *base,
3474                       bool has_top_node, const char *top_node,
3475                       bool has_top, const char *top,
3476                       bool has_backing_file, const char *backing_file,
3477                       bool has_speed, int64_t speed,
3478                       bool has_filter_node_name, const char *filter_node_name,
3479                       bool has_auto_finalize, bool auto_finalize,
3480                       bool has_auto_dismiss, bool auto_dismiss,
3481                       Error **errp)
3482 {
3483     BlockDriverState *bs;
3484     BlockDriverState *iter;
3485     BlockDriverState *base_bs, *top_bs;
3486     AioContext *aio_context;
3487     Error *local_err = NULL;
3488     /* This will be part of the QMP command, if/when the
3489      * BlockdevOnError change for blkmirror makes it in
3490      */
3491     BlockdevOnError on_error = BLOCKDEV_ON_ERROR_REPORT;
3492     int job_flags = JOB_DEFAULT;
3493 
3494     if (!has_speed) {
3495         speed = 0;
3496     }
3497     if (!has_filter_node_name) {
3498         filter_node_name = NULL;
3499     }
3500     if (has_auto_finalize && !auto_finalize) {
3501         job_flags |= JOB_MANUAL_FINALIZE;
3502     }
3503     if (has_auto_dismiss && !auto_dismiss) {
3504         job_flags |= JOB_MANUAL_DISMISS;
3505     }
3506 
3507     /* Important Note:
3508      *  libvirt relies on the DeviceNotFound error class in order to probe for
3509      *  live commit feature versions; for this to work, we must make sure to
3510      *  perform the device lookup before any generic errors that may occur in a
3511      *  scenario in which all optional arguments are omitted. */
3512     bs = qmp_get_root_bs(device, &local_err);
3513     if (!bs) {
3514         bs = bdrv_lookup_bs(device, device, NULL);
3515         if (!bs) {
3516             error_free(local_err);
3517             error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
3518                       "Device '%s' not found", device);
3519         } else {
3520             error_propagate(errp, local_err);
3521         }
3522         return;
3523     }
3524 
3525     aio_context = bdrv_get_aio_context(bs);
3526     aio_context_acquire(aio_context);
3527 
3528     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_COMMIT_SOURCE, errp)) {
3529         goto out;
3530     }
3531 
3532     /* default top_bs is the active layer */
3533     top_bs = bs;
3534 
3535     if (has_top_node && has_top) {
3536         error_setg(errp, "'top-node' and 'top' are mutually exclusive");
3537         goto out;
3538     } else if (has_top_node) {
3539         top_bs = bdrv_lookup_bs(NULL, top_node, errp);
3540         if (top_bs == NULL) {
3541             goto out;
3542         }
3543         if (!bdrv_chain_contains(bs, top_bs)) {
3544             error_setg(errp, "'%s' is not in this backing file chain",
3545                        top_node);
3546             goto out;
3547         }
3548     } else if (has_top && top) {
3549         /* This strcmp() is just a shortcut, there is no need to
3550          * refresh @bs's filename.  If it mismatches,
3551          * bdrv_find_backing_image() will do the refresh and may still
3552          * return @bs. */
3553         if (strcmp(bs->filename, top) != 0) {
3554             top_bs = bdrv_find_backing_image(bs, top);
3555         }
3556     }
3557 
3558     if (top_bs == NULL) {
3559         error_setg(errp, "Top image file %s not found", top ? top : "NULL");
3560         goto out;
3561     }
3562 
3563     assert(bdrv_get_aio_context(top_bs) == aio_context);
3564 
3565     if (has_base_node && has_base) {
3566         error_setg(errp, "'base-node' and 'base' are mutually exclusive");
3567         goto out;
3568     } else if (has_base_node) {
3569         base_bs = bdrv_lookup_bs(NULL, base_node, errp);
3570         if (base_bs == NULL) {
3571             goto out;
3572         }
3573         if (!bdrv_chain_contains(top_bs, base_bs)) {
3574             error_setg(errp, "'%s' is not in this backing file chain",
3575                        base_node);
3576             goto out;
3577         }
3578     } else if (has_base && base) {
3579         base_bs = bdrv_find_backing_image(top_bs, base);
3580     } else {
3581         base_bs = bdrv_find_base(top_bs);
3582     }
3583 
3584     if (base_bs == NULL) {
3585         error_setg(errp, QERR_BASE_NOT_FOUND, base ? base : "NULL");
3586         goto out;
3587     }
3588 
3589     assert(bdrv_get_aio_context(base_bs) == aio_context);
3590 
3591     for (iter = top_bs; iter != backing_bs(base_bs); iter = backing_bs(iter)) {
3592         if (bdrv_op_is_blocked(iter, BLOCK_OP_TYPE_COMMIT_TARGET, errp)) {
3593             goto out;
3594         }
3595     }
3596 
3597     /* Do not allow attempts to commit an image into itself */
3598     if (top_bs == base_bs) {
3599         error_setg(errp, "cannot commit an image into itself");
3600         goto out;
3601     }
3602 
3603     if (top_bs == bs) {
3604         if (has_backing_file) {
3605             error_setg(errp, "'backing-file' specified,"
3606                              " but 'top' is the active layer");
3607             goto out;
3608         }
3609         commit_active_start(has_job_id ? job_id : NULL, bs, base_bs,
3610                             job_flags, speed, on_error,
3611                             filter_node_name, NULL, NULL, false, &local_err);
3612     } else {
3613         BlockDriverState *overlay_bs = bdrv_find_overlay(bs, top_bs);
3614         if (bdrv_op_is_blocked(overlay_bs, BLOCK_OP_TYPE_COMMIT_TARGET, errp)) {
3615             goto out;
3616         }
3617         commit_start(has_job_id ? job_id : NULL, bs, base_bs, top_bs, job_flags,
3618                      speed, on_error, has_backing_file ? backing_file : NULL,
3619                      filter_node_name, &local_err);
3620     }
3621     if (local_err != NULL) {
3622         error_propagate(errp, local_err);
3623         goto out;
3624     }
3625 
3626 out:
3627     aio_context_release(aio_context);
3628 }
3629 
3630 /* Common QMP interface for drive-backup and blockdev-backup */
3631 static BlockJob *do_backup_common(BackupCommon *backup,
3632                                   BlockDriverState *bs,
3633                                   BlockDriverState *target_bs,
3634                                   AioContext *aio_context,
3635                                   JobTxn *txn, Error **errp)
3636 {
3637     BlockJob *job = NULL;
3638     BdrvDirtyBitmap *bmap = NULL;
3639     int job_flags = JOB_DEFAULT;
3640 
3641     if (!backup->has_speed) {
3642         backup->speed = 0;
3643     }
3644     if (!backup->has_on_source_error) {
3645         backup->on_source_error = BLOCKDEV_ON_ERROR_REPORT;
3646     }
3647     if (!backup->has_on_target_error) {
3648         backup->on_target_error = BLOCKDEV_ON_ERROR_REPORT;
3649     }
3650     if (!backup->has_job_id) {
3651         backup->job_id = NULL;
3652     }
3653     if (!backup->has_auto_finalize) {
3654         backup->auto_finalize = true;
3655     }
3656     if (!backup->has_auto_dismiss) {
3657         backup->auto_dismiss = true;
3658     }
3659     if (!backup->has_compress) {
3660         backup->compress = false;
3661     }
3662 
3663     if ((backup->sync == MIRROR_SYNC_MODE_BITMAP) ||
3664         (backup->sync == MIRROR_SYNC_MODE_INCREMENTAL)) {
3665         /* done before desugaring 'incremental' to print the right message */
3666         if (!backup->has_bitmap) {
3667             error_setg(errp, "must provide a valid bitmap name for "
3668                        "'%s' sync mode", MirrorSyncMode_str(backup->sync));
3669             return NULL;
3670         }
3671     }
3672 
3673     if (backup->sync == MIRROR_SYNC_MODE_INCREMENTAL) {
3674         if (backup->has_bitmap_mode &&
3675             backup->bitmap_mode != BITMAP_SYNC_MODE_ON_SUCCESS) {
3676             error_setg(errp, "Bitmap sync mode must be '%s' "
3677                        "when using sync mode '%s'",
3678                        BitmapSyncMode_str(BITMAP_SYNC_MODE_ON_SUCCESS),
3679                        MirrorSyncMode_str(backup->sync));
3680             return NULL;
3681         }
3682         backup->has_bitmap_mode = true;
3683         backup->sync = MIRROR_SYNC_MODE_BITMAP;
3684         backup->bitmap_mode = BITMAP_SYNC_MODE_ON_SUCCESS;
3685     }
3686 
3687     if (backup->has_bitmap) {
3688         bmap = bdrv_find_dirty_bitmap(bs, backup->bitmap);
3689         if (!bmap) {
3690             error_setg(errp, "Bitmap '%s' could not be found", backup->bitmap);
3691             return NULL;
3692         }
3693         if (!backup->has_bitmap_mode) {
3694             error_setg(errp, "Bitmap sync mode must be given "
3695                        "when providing a bitmap");
3696             return NULL;
3697         }
3698         if (bdrv_dirty_bitmap_check(bmap, BDRV_BITMAP_ALLOW_RO, errp)) {
3699             return NULL;
3700         }
3701 
3702         /* This does not produce a useful bitmap artifact: */
3703         if (backup->sync == MIRROR_SYNC_MODE_NONE) {
3704             error_setg(errp, "sync mode '%s' does not produce meaningful bitmap"
3705                        " outputs", MirrorSyncMode_str(backup->sync));
3706             return NULL;
3707         }
3708 
3709         /* If the bitmap isn't used for input or output, this is useless: */
3710         if (backup->bitmap_mode == BITMAP_SYNC_MODE_NEVER &&
3711             backup->sync != MIRROR_SYNC_MODE_BITMAP) {
3712             error_setg(errp, "Bitmap sync mode '%s' has no meaningful effect"
3713                        " when combined with sync mode '%s'",
3714                        BitmapSyncMode_str(backup->bitmap_mode),
3715                        MirrorSyncMode_str(backup->sync));
3716             return NULL;
3717         }
3718     }
3719 
3720     if (!backup->has_bitmap && backup->has_bitmap_mode) {
3721         error_setg(errp, "Cannot specify bitmap sync mode without a bitmap");
3722         return NULL;
3723     }
3724 
3725     if (!backup->auto_finalize) {
3726         job_flags |= JOB_MANUAL_FINALIZE;
3727     }
3728     if (!backup->auto_dismiss) {
3729         job_flags |= JOB_MANUAL_DISMISS;
3730     }
3731 
3732     job = backup_job_create(backup->job_id, bs, target_bs, backup->speed,
3733                             backup->sync, bmap, backup->bitmap_mode,
3734                             backup->compress,
3735                             backup->filter_node_name,
3736                             backup->on_source_error,
3737                             backup->on_target_error,
3738                             job_flags, NULL, NULL, txn, errp);
3739     return job;
3740 }
3741 
3742 void qmp_drive_backup(DriveBackup *backup, Error **errp)
3743 {
3744     TransactionAction action = {
3745         .type = TRANSACTION_ACTION_KIND_DRIVE_BACKUP,
3746         .u.drive_backup.data = backup,
3747     };
3748     blockdev_do_action(&action, errp);
3749 }
3750 
3751 BlockDeviceInfoList *qmp_query_named_block_nodes(Error **errp)
3752 {
3753     return bdrv_named_nodes_list(errp);
3754 }
3755 
3756 XDbgBlockGraph *qmp_x_debug_query_block_graph(Error **errp)
3757 {
3758     return bdrv_get_xdbg_block_graph(errp);
3759 }
3760 
3761 void qmp_blockdev_backup(BlockdevBackup *backup, Error **errp)
3762 {
3763     TransactionAction action = {
3764         .type = TRANSACTION_ACTION_KIND_BLOCKDEV_BACKUP,
3765         .u.blockdev_backup.data = backup,
3766     };
3767     blockdev_do_action(&action, errp);
3768 }
3769 
3770 /* Parameter check and block job starting for drive mirroring.
3771  * Caller should hold @device and @target's aio context (must be the same).
3772  **/
3773 static void blockdev_mirror_common(const char *job_id, BlockDriverState *bs,
3774                                    BlockDriverState *target,
3775                                    bool has_replaces, const char *replaces,
3776                                    enum MirrorSyncMode sync,
3777                                    BlockMirrorBackingMode backing_mode,
3778                                    bool zero_target,
3779                                    bool has_speed, int64_t speed,
3780                                    bool has_granularity, uint32_t granularity,
3781                                    bool has_buf_size, int64_t buf_size,
3782                                    bool has_on_source_error,
3783                                    BlockdevOnError on_source_error,
3784                                    bool has_on_target_error,
3785                                    BlockdevOnError on_target_error,
3786                                    bool has_unmap, bool unmap,
3787                                    bool has_filter_node_name,
3788                                    const char *filter_node_name,
3789                                    bool has_copy_mode, MirrorCopyMode copy_mode,
3790                                    bool has_auto_finalize, bool auto_finalize,
3791                                    bool has_auto_dismiss, bool auto_dismiss,
3792                                    Error **errp)
3793 {
3794     int job_flags = JOB_DEFAULT;
3795 
3796     if (!has_speed) {
3797         speed = 0;
3798     }
3799     if (!has_on_source_error) {
3800         on_source_error = BLOCKDEV_ON_ERROR_REPORT;
3801     }
3802     if (!has_on_target_error) {
3803         on_target_error = BLOCKDEV_ON_ERROR_REPORT;
3804     }
3805     if (!has_granularity) {
3806         granularity = 0;
3807     }
3808     if (!has_buf_size) {
3809         buf_size = 0;
3810     }
3811     if (!has_unmap) {
3812         unmap = true;
3813     }
3814     if (!has_filter_node_name) {
3815         filter_node_name = NULL;
3816     }
3817     if (!has_copy_mode) {
3818         copy_mode = MIRROR_COPY_MODE_BACKGROUND;
3819     }
3820     if (has_auto_finalize && !auto_finalize) {
3821         job_flags |= JOB_MANUAL_FINALIZE;
3822     }
3823     if (has_auto_dismiss && !auto_dismiss) {
3824         job_flags |= JOB_MANUAL_DISMISS;
3825     }
3826 
3827     if (granularity != 0 && (granularity < 512 || granularity > 1048576 * 64)) {
3828         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "granularity",
3829                    "a value in range [512B, 64MB]");
3830         return;
3831     }
3832     if (granularity & (granularity - 1)) {
3833         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "granularity",
3834                    "power of 2");
3835         return;
3836     }
3837 
3838     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_MIRROR_SOURCE, errp)) {
3839         return;
3840     }
3841     if (bdrv_op_is_blocked(target, BLOCK_OP_TYPE_MIRROR_TARGET, errp)) {
3842         return;
3843     }
3844 
3845     if (!bs->backing && sync == MIRROR_SYNC_MODE_TOP) {
3846         sync = MIRROR_SYNC_MODE_FULL;
3847     }
3848 
3849     if (has_replaces) {
3850         BlockDriverState *to_replace_bs;
3851         AioContext *replace_aio_context;
3852         int64_t bs_size, replace_size;
3853 
3854         bs_size = bdrv_getlength(bs);
3855         if (bs_size < 0) {
3856             error_setg_errno(errp, -bs_size, "Failed to query device's size");
3857             return;
3858         }
3859 
3860         to_replace_bs = check_to_replace_node(bs, replaces, errp);
3861         if (!to_replace_bs) {
3862             return;
3863         }
3864 
3865         replace_aio_context = bdrv_get_aio_context(to_replace_bs);
3866         aio_context_acquire(replace_aio_context);
3867         replace_size = bdrv_getlength(to_replace_bs);
3868         aio_context_release(replace_aio_context);
3869 
3870         if (replace_size < 0) {
3871             error_setg_errno(errp, -replace_size,
3872                              "Failed to query the replacement node's size");
3873             return;
3874         }
3875         if (bs_size != replace_size) {
3876             error_setg(errp, "cannot replace image with a mirror image of "
3877                              "different size");
3878             return;
3879         }
3880     }
3881 
3882     /* pass the node name to replace to mirror start since it's loose coupling
3883      * and will allow to check whether the node still exist at mirror completion
3884      */
3885     mirror_start(job_id, bs, target,
3886                  has_replaces ? replaces : NULL, job_flags,
3887                  speed, granularity, buf_size, sync, backing_mode, zero_target,
3888                  on_source_error, on_target_error, unmap, filter_node_name,
3889                  copy_mode, errp);
3890 }
3891 
3892 void qmp_drive_mirror(DriveMirror *arg, Error **errp)
3893 {
3894     BlockDriverState *bs;
3895     BlockDriverState *source, *target_bs;
3896     AioContext *aio_context;
3897     AioContext *old_context;
3898     BlockMirrorBackingMode backing_mode;
3899     Error *local_err = NULL;
3900     QDict *options = NULL;
3901     int flags;
3902     int64_t size;
3903     const char *format = arg->format;
3904     bool zero_target;
3905     int ret;
3906 
3907     bs = qmp_get_root_bs(arg->device, errp);
3908     if (!bs) {
3909         return;
3910     }
3911 
3912     /* Early check to avoid creating target */
3913     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_MIRROR_SOURCE, errp)) {
3914         return;
3915     }
3916 
3917     aio_context = bdrv_get_aio_context(bs);
3918     aio_context_acquire(aio_context);
3919 
3920     if (!arg->has_mode) {
3921         arg->mode = NEW_IMAGE_MODE_ABSOLUTE_PATHS;
3922     }
3923 
3924     if (!arg->has_format) {
3925         format = (arg->mode == NEW_IMAGE_MODE_EXISTING
3926                   ? NULL : bs->drv->format_name);
3927     }
3928 
3929     flags = bs->open_flags | BDRV_O_RDWR;
3930     source = backing_bs(bs);
3931     if (!source && arg->sync == MIRROR_SYNC_MODE_TOP) {
3932         arg->sync = MIRROR_SYNC_MODE_FULL;
3933     }
3934     if (arg->sync == MIRROR_SYNC_MODE_NONE) {
3935         source = bs;
3936     }
3937 
3938     size = bdrv_getlength(bs);
3939     if (size < 0) {
3940         error_setg_errno(errp, -size, "bdrv_getlength failed");
3941         goto out;
3942     }
3943 
3944     if (arg->has_replaces) {
3945         if (!arg->has_node_name) {
3946             error_setg(errp, "a node-name must be provided when replacing a"
3947                              " named node of the graph");
3948             goto out;
3949         }
3950     }
3951 
3952     if (arg->mode == NEW_IMAGE_MODE_ABSOLUTE_PATHS) {
3953         backing_mode = MIRROR_SOURCE_BACKING_CHAIN;
3954     } else {
3955         backing_mode = MIRROR_OPEN_BACKING_CHAIN;
3956     }
3957 
3958     /* Don't open backing image in create() */
3959     flags |= BDRV_O_NO_BACKING;
3960 
3961     if ((arg->sync == MIRROR_SYNC_MODE_FULL || !source)
3962         && arg->mode != NEW_IMAGE_MODE_EXISTING)
3963     {
3964         /* create new image w/o backing file */
3965         assert(format);
3966         bdrv_img_create(arg->target, format,
3967                         NULL, NULL, NULL, size, flags, false, &local_err);
3968     } else {
3969         switch (arg->mode) {
3970         case NEW_IMAGE_MODE_EXISTING:
3971             break;
3972         case NEW_IMAGE_MODE_ABSOLUTE_PATHS:
3973             /* create new image with backing file */
3974             bdrv_refresh_filename(source);
3975             bdrv_img_create(arg->target, format,
3976                             source->filename,
3977                             source->drv->format_name,
3978                             NULL, size, flags, false, &local_err);
3979             break;
3980         default:
3981             abort();
3982         }
3983     }
3984 
3985     if (local_err) {
3986         error_propagate(errp, local_err);
3987         goto out;
3988     }
3989 
3990     options = qdict_new();
3991     if (arg->has_node_name) {
3992         qdict_put_str(options, "node-name", arg->node_name);
3993     }
3994     if (format) {
3995         qdict_put_str(options, "driver", format);
3996     }
3997 
3998     /* Mirroring takes care of copy-on-write using the source's backing
3999      * file.
4000      */
4001     target_bs = bdrv_open(arg->target, NULL, options, flags, errp);
4002     if (!target_bs) {
4003         goto out;
4004     }
4005 
4006     zero_target = (arg->sync == MIRROR_SYNC_MODE_FULL &&
4007                    (arg->mode == NEW_IMAGE_MODE_EXISTING ||
4008                     !bdrv_has_zero_init(target_bs)));
4009 
4010 
4011     /* Honor bdrv_try_set_aio_context() context acquisition requirements. */
4012     old_context = bdrv_get_aio_context(target_bs);
4013     aio_context_release(aio_context);
4014     aio_context_acquire(old_context);
4015 
4016     ret = bdrv_try_set_aio_context(target_bs, aio_context, errp);
4017     if (ret < 0) {
4018         bdrv_unref(target_bs);
4019         aio_context_release(old_context);
4020         return;
4021     }
4022 
4023     aio_context_release(old_context);
4024     aio_context_acquire(aio_context);
4025 
4026     blockdev_mirror_common(arg->has_job_id ? arg->job_id : NULL, bs, target_bs,
4027                            arg->has_replaces, arg->replaces, arg->sync,
4028                            backing_mode, zero_target,
4029                            arg->has_speed, arg->speed,
4030                            arg->has_granularity, arg->granularity,
4031                            arg->has_buf_size, arg->buf_size,
4032                            arg->has_on_source_error, arg->on_source_error,
4033                            arg->has_on_target_error, arg->on_target_error,
4034                            arg->has_unmap, arg->unmap,
4035                            false, NULL,
4036                            arg->has_copy_mode, arg->copy_mode,
4037                            arg->has_auto_finalize, arg->auto_finalize,
4038                            arg->has_auto_dismiss, arg->auto_dismiss,
4039                            &local_err);
4040     bdrv_unref(target_bs);
4041     error_propagate(errp, local_err);
4042 out:
4043     aio_context_release(aio_context);
4044 }
4045 
4046 void qmp_blockdev_mirror(bool has_job_id, const char *job_id,
4047                          const char *device, const char *target,
4048                          bool has_replaces, const char *replaces,
4049                          MirrorSyncMode sync,
4050                          bool has_speed, int64_t speed,
4051                          bool has_granularity, uint32_t granularity,
4052                          bool has_buf_size, int64_t buf_size,
4053                          bool has_on_source_error,
4054                          BlockdevOnError on_source_error,
4055                          bool has_on_target_error,
4056                          BlockdevOnError on_target_error,
4057                          bool has_filter_node_name,
4058                          const char *filter_node_name,
4059                          bool has_copy_mode, MirrorCopyMode copy_mode,
4060                          bool has_auto_finalize, bool auto_finalize,
4061                          bool has_auto_dismiss, bool auto_dismiss,
4062                          Error **errp)
4063 {
4064     BlockDriverState *bs;
4065     BlockDriverState *target_bs;
4066     AioContext *aio_context;
4067     AioContext *old_context;
4068     BlockMirrorBackingMode backing_mode = MIRROR_LEAVE_BACKING_CHAIN;
4069     Error *local_err = NULL;
4070     bool zero_target;
4071     int ret;
4072 
4073     bs = qmp_get_root_bs(device, errp);
4074     if (!bs) {
4075         return;
4076     }
4077 
4078     target_bs = bdrv_lookup_bs(target, target, errp);
4079     if (!target_bs) {
4080         return;
4081     }
4082 
4083     zero_target = (sync == MIRROR_SYNC_MODE_FULL);
4084 
4085     /* Honor bdrv_try_set_aio_context() context acquisition requirements. */
4086     old_context = bdrv_get_aio_context(target_bs);
4087     aio_context = bdrv_get_aio_context(bs);
4088     aio_context_acquire(old_context);
4089 
4090     ret = bdrv_try_set_aio_context(target_bs, aio_context, errp);
4091 
4092     aio_context_release(old_context);
4093     aio_context_acquire(aio_context);
4094 
4095     if (ret < 0) {
4096         goto out;
4097     }
4098 
4099     blockdev_mirror_common(has_job_id ? job_id : NULL, bs, target_bs,
4100                            has_replaces, replaces, sync, backing_mode,
4101                            zero_target, has_speed, speed,
4102                            has_granularity, granularity,
4103                            has_buf_size, buf_size,
4104                            has_on_source_error, on_source_error,
4105                            has_on_target_error, on_target_error,
4106                            true, true,
4107                            has_filter_node_name, filter_node_name,
4108                            has_copy_mode, copy_mode,
4109                            has_auto_finalize, auto_finalize,
4110                            has_auto_dismiss, auto_dismiss,
4111                            &local_err);
4112     error_propagate(errp, local_err);
4113 out:
4114     aio_context_release(aio_context);
4115 }
4116 
4117 /* Get a block job using its ID and acquire its AioContext */
4118 static BlockJob *find_block_job(const char *id, AioContext **aio_context,
4119                                 Error **errp)
4120 {
4121     BlockJob *job;
4122 
4123     assert(id != NULL);
4124 
4125     *aio_context = NULL;
4126 
4127     job = block_job_get(id);
4128 
4129     if (!job) {
4130         error_set(errp, ERROR_CLASS_DEVICE_NOT_ACTIVE,
4131                   "Block job '%s' not found", id);
4132         return NULL;
4133     }
4134 
4135     *aio_context = blk_get_aio_context(job->blk);
4136     aio_context_acquire(*aio_context);
4137 
4138     return job;
4139 }
4140 
4141 void qmp_block_job_set_speed(const char *device, int64_t speed, Error **errp)
4142 {
4143     AioContext *aio_context;
4144     BlockJob *job = find_block_job(device, &aio_context, errp);
4145 
4146     if (!job) {
4147         return;
4148     }
4149 
4150     block_job_set_speed(job, speed, errp);
4151     aio_context_release(aio_context);
4152 }
4153 
4154 void qmp_block_job_cancel(const char *device,
4155                           bool has_force, bool force, Error **errp)
4156 {
4157     AioContext *aio_context;
4158     BlockJob *job = find_block_job(device, &aio_context, errp);
4159 
4160     if (!job) {
4161         return;
4162     }
4163 
4164     if (!has_force) {
4165         force = false;
4166     }
4167 
4168     if (job_user_paused(&job->job) && !force) {
4169         error_setg(errp, "The block job for device '%s' is currently paused",
4170                    device);
4171         goto out;
4172     }
4173 
4174     trace_qmp_block_job_cancel(job);
4175     job_user_cancel(&job->job, force, errp);
4176 out:
4177     aio_context_release(aio_context);
4178 }
4179 
4180 void qmp_block_job_pause(const char *device, Error **errp)
4181 {
4182     AioContext *aio_context;
4183     BlockJob *job = find_block_job(device, &aio_context, errp);
4184 
4185     if (!job) {
4186         return;
4187     }
4188 
4189     trace_qmp_block_job_pause(job);
4190     job_user_pause(&job->job, errp);
4191     aio_context_release(aio_context);
4192 }
4193 
4194 void qmp_block_job_resume(const char *device, Error **errp)
4195 {
4196     AioContext *aio_context;
4197     BlockJob *job = find_block_job(device, &aio_context, errp);
4198 
4199     if (!job) {
4200         return;
4201     }
4202 
4203     trace_qmp_block_job_resume(job);
4204     job_user_resume(&job->job, errp);
4205     aio_context_release(aio_context);
4206 }
4207 
4208 void qmp_block_job_complete(const char *device, Error **errp)
4209 {
4210     AioContext *aio_context;
4211     BlockJob *job = find_block_job(device, &aio_context, errp);
4212 
4213     if (!job) {
4214         return;
4215     }
4216 
4217     trace_qmp_block_job_complete(job);
4218     job_complete(&job->job, errp);
4219     aio_context_release(aio_context);
4220 }
4221 
4222 void qmp_block_job_finalize(const char *id, Error **errp)
4223 {
4224     AioContext *aio_context;
4225     BlockJob *job = find_block_job(id, &aio_context, errp);
4226 
4227     if (!job) {
4228         return;
4229     }
4230 
4231     trace_qmp_block_job_finalize(job);
4232     job_finalize(&job->job, errp);
4233     aio_context_release(aio_context);
4234 }
4235 
4236 void qmp_block_job_dismiss(const char *id, Error **errp)
4237 {
4238     AioContext *aio_context;
4239     BlockJob *bjob = find_block_job(id, &aio_context, errp);
4240     Job *job;
4241 
4242     if (!bjob) {
4243         return;
4244     }
4245 
4246     trace_qmp_block_job_dismiss(bjob);
4247     job = &bjob->job;
4248     job_dismiss(&job, errp);
4249     aio_context_release(aio_context);
4250 }
4251 
4252 void qmp_change_backing_file(const char *device,
4253                              const char *image_node_name,
4254                              const char *backing_file,
4255                              Error **errp)
4256 {
4257     BlockDriverState *bs = NULL;
4258     AioContext *aio_context;
4259     BlockDriverState *image_bs = NULL;
4260     Error *local_err = NULL;
4261     bool ro;
4262     int ret;
4263 
4264     bs = qmp_get_root_bs(device, errp);
4265     if (!bs) {
4266         return;
4267     }
4268 
4269     aio_context = bdrv_get_aio_context(bs);
4270     aio_context_acquire(aio_context);
4271 
4272     image_bs = bdrv_lookup_bs(NULL, image_node_name, &local_err);
4273     if (local_err) {
4274         error_propagate(errp, local_err);
4275         goto out;
4276     }
4277 
4278     if (!image_bs) {
4279         error_setg(errp, "image file not found");
4280         goto out;
4281     }
4282 
4283     if (bdrv_find_base(image_bs) == image_bs) {
4284         error_setg(errp, "not allowing backing file change on an image "
4285                          "without a backing file");
4286         goto out;
4287     }
4288 
4289     /* even though we are not necessarily operating on bs, we need it to
4290      * determine if block ops are currently prohibited on the chain */
4291     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_CHANGE, errp)) {
4292         goto out;
4293     }
4294 
4295     /* final sanity check */
4296     if (!bdrv_chain_contains(bs, image_bs)) {
4297         error_setg(errp, "'%s' and image file are not in the same chain",
4298                    device);
4299         goto out;
4300     }
4301 
4302     /* if not r/w, reopen to make r/w */
4303     ro = bdrv_is_read_only(image_bs);
4304 
4305     if (ro) {
4306         if (bdrv_reopen_set_read_only(image_bs, false, errp) != 0) {
4307             goto out;
4308         }
4309     }
4310 
4311     ret = bdrv_change_backing_file(image_bs, backing_file,
4312                                image_bs->drv ? image_bs->drv->format_name : "");
4313 
4314     if (ret < 0) {
4315         error_setg_errno(errp, -ret, "Could not change backing file to '%s'",
4316                          backing_file);
4317         /* don't exit here, so we can try to restore open flags if
4318          * appropriate */
4319     }
4320 
4321     if (ro) {
4322         bdrv_reopen_set_read_only(image_bs, true, &local_err);
4323         error_propagate(errp, local_err);
4324     }
4325 
4326 out:
4327     aio_context_release(aio_context);
4328 }
4329 
4330 void hmp_drive_add_node(Monitor *mon, const char *optstr)
4331 {
4332     QemuOpts *opts;
4333     QDict *qdict;
4334     Error *local_err = NULL;
4335 
4336     opts = qemu_opts_parse_noisily(&qemu_drive_opts, optstr, false);
4337     if (!opts) {
4338         return;
4339     }
4340 
4341     qdict = qemu_opts_to_qdict(opts, NULL);
4342 
4343     if (!qdict_get_try_str(qdict, "node-name")) {
4344         qobject_unref(qdict);
4345         error_report("'node-name' needs to be specified");
4346         goto out;
4347     }
4348 
4349     BlockDriverState *bs = bds_tree_init(qdict, &local_err);
4350     if (!bs) {
4351         error_report_err(local_err);
4352         goto out;
4353     }
4354 
4355     QTAILQ_INSERT_TAIL(&monitor_bdrv_states, bs, monitor_list);
4356 
4357 out:
4358     qemu_opts_del(opts);
4359 }
4360 
4361 void qmp_blockdev_add(BlockdevOptions *options, Error **errp)
4362 {
4363     BlockDriverState *bs;
4364     QObject *obj;
4365     Visitor *v = qobject_output_visitor_new(&obj);
4366     QDict *qdict;
4367     Error *local_err = NULL;
4368 
4369     visit_type_BlockdevOptions(v, NULL, &options, &local_err);
4370     if (local_err) {
4371         error_propagate(errp, local_err);
4372         goto fail;
4373     }
4374 
4375     visit_complete(v, &obj);
4376     qdict = qobject_to(QDict, obj);
4377 
4378     qdict_flatten(qdict);
4379 
4380     if (!qdict_get_try_str(qdict, "node-name")) {
4381         error_setg(errp, "'node-name' must be specified for the root node");
4382         goto fail;
4383     }
4384 
4385     bs = bds_tree_init(qdict, errp);
4386     if (!bs) {
4387         goto fail;
4388     }
4389 
4390     QTAILQ_INSERT_TAIL(&monitor_bdrv_states, bs, monitor_list);
4391 
4392 fail:
4393     visit_free(v);
4394 }
4395 
4396 void qmp_x_blockdev_reopen(BlockdevOptions *options, Error **errp)
4397 {
4398     BlockDriverState *bs;
4399     AioContext *ctx;
4400     QObject *obj;
4401     Visitor *v = qobject_output_visitor_new(&obj);
4402     Error *local_err = NULL;
4403     BlockReopenQueue *queue;
4404     QDict *qdict;
4405 
4406     /* Check for the selected node name */
4407     if (!options->has_node_name) {
4408         error_setg(errp, "Node name not specified");
4409         goto fail;
4410     }
4411 
4412     bs = bdrv_find_node(options->node_name);
4413     if (!bs) {
4414         error_setg(errp, "Cannot find node named '%s'", options->node_name);
4415         goto fail;
4416     }
4417 
4418     /* Put all options in a QDict and flatten it */
4419     visit_type_BlockdevOptions(v, NULL, &options, &local_err);
4420     if (local_err) {
4421         error_propagate(errp, local_err);
4422         goto fail;
4423     }
4424 
4425     visit_complete(v, &obj);
4426     qdict = qobject_to(QDict, obj);
4427 
4428     qdict_flatten(qdict);
4429 
4430     /* Perform the reopen operation */
4431     ctx = bdrv_get_aio_context(bs);
4432     aio_context_acquire(ctx);
4433     bdrv_subtree_drained_begin(bs);
4434     queue = bdrv_reopen_queue(NULL, bs, qdict, false);
4435     bdrv_reopen_multiple(queue, errp);
4436     bdrv_subtree_drained_end(bs);
4437     aio_context_release(ctx);
4438 
4439 fail:
4440     visit_free(v);
4441 }
4442 
4443 void qmp_blockdev_del(const char *node_name, Error **errp)
4444 {
4445     AioContext *aio_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     if (bdrv_has_blk(bs)) {
4454         error_setg(errp, "Node %s is in use", node_name);
4455         return;
4456     }
4457     aio_context = bdrv_get_aio_context(bs);
4458     aio_context_acquire(aio_context);
4459 
4460     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_DRIVE_DEL, errp)) {
4461         goto out;
4462     }
4463 
4464     if (!QTAILQ_IN_USE(bs, monitor_list)) {
4465         error_setg(errp, "Node %s is not owned by the monitor",
4466                    bs->node_name);
4467         goto out;
4468     }
4469 
4470     if (bs->refcnt > 1) {
4471         error_setg(errp, "Block device %s is in use",
4472                    bdrv_get_device_or_node_name(bs));
4473         goto out;
4474     }
4475 
4476     QTAILQ_REMOVE(&monitor_bdrv_states, bs, monitor_list);
4477     bdrv_unref(bs);
4478 
4479 out:
4480     aio_context_release(aio_context);
4481 }
4482 
4483 static BdrvChild *bdrv_find_child(BlockDriverState *parent_bs,
4484                                   const char *child_name)
4485 {
4486     BdrvChild *child;
4487 
4488     QLIST_FOREACH(child, &parent_bs->children, next) {
4489         if (strcmp(child->name, child_name) == 0) {
4490             return child;
4491         }
4492     }
4493 
4494     return NULL;
4495 }
4496 
4497 void qmp_x_blockdev_change(const char *parent, bool has_child,
4498                            const char *child, bool has_node,
4499                            const char *node, Error **errp)
4500 {
4501     BlockDriverState *parent_bs, *new_bs = NULL;
4502     BdrvChild *p_child;
4503 
4504     parent_bs = bdrv_lookup_bs(parent, parent, errp);
4505     if (!parent_bs) {
4506         return;
4507     }
4508 
4509     if (has_child == has_node) {
4510         if (has_child) {
4511             error_setg(errp, "The parameters child and node are in conflict");
4512         } else {
4513             error_setg(errp, "Either child or node must be specified");
4514         }
4515         return;
4516     }
4517 
4518     if (has_child) {
4519         p_child = bdrv_find_child(parent_bs, child);
4520         if (!p_child) {
4521             error_setg(errp, "Node '%s' does not have child '%s'",
4522                        parent, child);
4523             return;
4524         }
4525         bdrv_del_child(parent_bs, p_child, errp);
4526     }
4527 
4528     if (has_node) {
4529         new_bs = bdrv_find_node(node);
4530         if (!new_bs) {
4531             error_setg(errp, "Node '%s' not found", node);
4532             return;
4533         }
4534         bdrv_add_child(parent_bs, new_bs, errp);
4535     }
4536 }
4537 
4538 BlockJobInfoList *qmp_query_block_jobs(Error **errp)
4539 {
4540     BlockJobInfoList *head = NULL, **p_next = &head;
4541     BlockJob *job;
4542 
4543     for (job = block_job_next(NULL); job; job = block_job_next(job)) {
4544         BlockJobInfoList *elem;
4545         AioContext *aio_context;
4546 
4547         if (block_job_is_internal(job)) {
4548             continue;
4549         }
4550         elem = g_new0(BlockJobInfoList, 1);
4551         aio_context = blk_get_aio_context(job->blk);
4552         aio_context_acquire(aio_context);
4553         elem->value = block_job_query(job, errp);
4554         aio_context_release(aio_context);
4555         if (!elem->value) {
4556             g_free(elem);
4557             qapi_free_BlockJobInfoList(head);
4558             return NULL;
4559         }
4560         *p_next = elem;
4561         p_next = &elem->next;
4562     }
4563 
4564     return head;
4565 }
4566 
4567 void qmp_x_blockdev_set_iothread(const char *node_name, StrOrNull *iothread,
4568                                  bool has_force, bool force, Error **errp)
4569 {
4570     AioContext *old_context;
4571     AioContext *new_context;
4572     BlockDriverState *bs;
4573 
4574     bs = bdrv_find_node(node_name);
4575     if (!bs) {
4576         error_setg(errp, "Cannot find node %s", node_name);
4577         return;
4578     }
4579 
4580     /* Protects against accidents. */
4581     if (!(has_force && force) && bdrv_has_blk(bs)) {
4582         error_setg(errp, "Node %s is associated with a BlockBackend and could "
4583                          "be in use (use force=true to override this check)",
4584                          node_name);
4585         return;
4586     }
4587 
4588     if (iothread->type == QTYPE_QSTRING) {
4589         IOThread *obj = iothread_by_id(iothread->u.s);
4590         if (!obj) {
4591             error_setg(errp, "Cannot find iothread %s", iothread->u.s);
4592             return;
4593         }
4594 
4595         new_context = iothread_get_aio_context(obj);
4596     } else {
4597         new_context = qemu_get_aio_context();
4598     }
4599 
4600     old_context = bdrv_get_aio_context(bs);
4601     aio_context_acquire(old_context);
4602 
4603     bdrv_try_set_aio_context(bs, new_context, errp);
4604 
4605     aio_context_release(old_context);
4606 }
4607 
4608 void qmp_block_latency_histogram_set(
4609     const char *id,
4610     bool has_boundaries, uint64List *boundaries,
4611     bool has_boundaries_read, uint64List *boundaries_read,
4612     bool has_boundaries_write, uint64List *boundaries_write,
4613     bool has_boundaries_flush, uint64List *boundaries_flush,
4614     Error **errp)
4615 {
4616     BlockBackend *blk = qmp_get_blk(NULL, id, errp);
4617     BlockAcctStats *stats;
4618     int ret;
4619 
4620     if (!blk) {
4621         return;
4622     }
4623 
4624     stats = blk_get_stats(blk);
4625 
4626     if (!has_boundaries && !has_boundaries_read && !has_boundaries_write &&
4627         !has_boundaries_flush)
4628     {
4629         block_latency_histograms_clear(stats);
4630         return;
4631     }
4632 
4633     if (has_boundaries || has_boundaries_read) {
4634         ret = block_latency_histogram_set(
4635             stats, BLOCK_ACCT_READ,
4636             has_boundaries_read ? boundaries_read : boundaries);
4637         if (ret) {
4638             error_setg(errp, "Device '%s' set read boundaries fail", id);
4639             return;
4640         }
4641     }
4642 
4643     if (has_boundaries || has_boundaries_write) {
4644         ret = block_latency_histogram_set(
4645             stats, BLOCK_ACCT_WRITE,
4646             has_boundaries_write ? boundaries_write : boundaries);
4647         if (ret) {
4648             error_setg(errp, "Device '%s' set write boundaries fail", id);
4649             return;
4650         }
4651     }
4652 
4653     if (has_boundaries || has_boundaries_flush) {
4654         ret = block_latency_histogram_set(
4655             stats, BLOCK_ACCT_FLUSH,
4656             has_boundaries_flush ? boundaries_flush : boundaries);
4657         if (ret) {
4658             error_setg(errp, "Device '%s' set flush boundaries fail", id);
4659             return;
4660         }
4661     }
4662 }
4663 
4664 QemuOptsList qemu_common_drive_opts = {
4665     .name = "drive",
4666     .head = QTAILQ_HEAD_INITIALIZER(qemu_common_drive_opts.head),
4667     .desc = {
4668         {
4669             .name = "snapshot",
4670             .type = QEMU_OPT_BOOL,
4671             .help = "enable/disable snapshot mode",
4672         },{
4673             .name = "aio",
4674             .type = QEMU_OPT_STRING,
4675             .help = "host AIO implementation (threads, native)",
4676         },{
4677             .name = BDRV_OPT_CACHE_WB,
4678             .type = QEMU_OPT_BOOL,
4679             .help = "Enable writeback mode",
4680         },{
4681             .name = "format",
4682             .type = QEMU_OPT_STRING,
4683             .help = "disk format (raw, qcow2, ...)",
4684         },{
4685             .name = "rerror",
4686             .type = QEMU_OPT_STRING,
4687             .help = "read error action",
4688         },{
4689             .name = "werror",
4690             .type = QEMU_OPT_STRING,
4691             .help = "write error action",
4692         },{
4693             .name = BDRV_OPT_READ_ONLY,
4694             .type = QEMU_OPT_BOOL,
4695             .help = "open drive file as read-only",
4696         },
4697 
4698         THROTTLE_OPTS,
4699 
4700         {
4701             .name = "throttling.group",
4702             .type = QEMU_OPT_STRING,
4703             .help = "name of the block throttling group",
4704         },{
4705             .name = "copy-on-read",
4706             .type = QEMU_OPT_BOOL,
4707             .help = "copy read data from backing file into image file",
4708         },{
4709             .name = "detect-zeroes",
4710             .type = QEMU_OPT_STRING,
4711             .help = "try to optimize zero writes (off, on, unmap)",
4712         },{
4713             .name = "stats-account-invalid",
4714             .type = QEMU_OPT_BOOL,
4715             .help = "whether to account for invalid I/O operations "
4716                     "in the statistics",
4717         },{
4718             .name = "stats-account-failed",
4719             .type = QEMU_OPT_BOOL,
4720             .help = "whether to account for failed I/O operations "
4721                     "in the statistics",
4722         },
4723         { /* end of list */ }
4724     },
4725 };
4726 
4727 QemuOptsList qemu_drive_opts = {
4728     .name = "drive",
4729     .head = QTAILQ_HEAD_INITIALIZER(qemu_drive_opts.head),
4730     .desc = {
4731         /*
4732          * no elements => accept any params
4733          * validation will happen later
4734          */
4735         { /* end of list */ }
4736     },
4737 };
4738