xref: /openbmc/qemu/blockdev.c (revision 9e9b10c6)
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             monitor_printf(mon, "Device '%s' not found\n", device);
1092             return;
1093         }
1094         if (!blk_is_available(blk)) {
1095             monitor_printf(mon, "Device '%s' has no medium\n", 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         monitor_printf(mon, "'commit' error for '%s': %s\n", device,
1109                        strerror(-ret));
1110     }
1111 }
1112 
1113 static void blockdev_do_action(TransactionAction *action, Error **errp)
1114 {
1115     TransactionActionList list;
1116 
1117     list.value = action;
1118     list.next = NULL;
1119     qmp_transaction(&list, false, NULL, errp);
1120 }
1121 
1122 void qmp_blockdev_snapshot_sync(bool has_device, const char *device,
1123                                 bool has_node_name, const char *node_name,
1124                                 const char *snapshot_file,
1125                                 bool has_snapshot_node_name,
1126                                 const char *snapshot_node_name,
1127                                 bool has_format, const char *format,
1128                                 bool has_mode, NewImageMode mode, Error **errp)
1129 {
1130     BlockdevSnapshotSync snapshot = {
1131         .has_device = has_device,
1132         .device = (char *) device,
1133         .has_node_name = has_node_name,
1134         .node_name = (char *) node_name,
1135         .snapshot_file = (char *) snapshot_file,
1136         .has_snapshot_node_name = has_snapshot_node_name,
1137         .snapshot_node_name = (char *) snapshot_node_name,
1138         .has_format = has_format,
1139         .format = (char *) format,
1140         .has_mode = has_mode,
1141         .mode = mode,
1142     };
1143     TransactionAction action = {
1144         .type = TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC,
1145         .u.blockdev_snapshot_sync.data = &snapshot,
1146     };
1147     blockdev_do_action(&action, errp);
1148 }
1149 
1150 void qmp_blockdev_snapshot(const char *node, const char *overlay,
1151                            Error **errp)
1152 {
1153     BlockdevSnapshot snapshot_data = {
1154         .node = (char *) node,
1155         .overlay = (char *) overlay
1156     };
1157     TransactionAction action = {
1158         .type = TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT,
1159         .u.blockdev_snapshot.data = &snapshot_data,
1160     };
1161     blockdev_do_action(&action, errp);
1162 }
1163 
1164 void qmp_blockdev_snapshot_internal_sync(const char *device,
1165                                          const char *name,
1166                                          Error **errp)
1167 {
1168     BlockdevSnapshotInternal snapshot = {
1169         .device = (char *) device,
1170         .name = (char *) name
1171     };
1172     TransactionAction action = {
1173         .type = TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC,
1174         .u.blockdev_snapshot_internal_sync.data = &snapshot,
1175     };
1176     blockdev_do_action(&action, errp);
1177 }
1178 
1179 SnapshotInfo *qmp_blockdev_snapshot_delete_internal_sync(const char *device,
1180                                                          bool has_id,
1181                                                          const char *id,
1182                                                          bool has_name,
1183                                                          const char *name,
1184                                                          Error **errp)
1185 {
1186     BlockDriverState *bs;
1187     AioContext *aio_context;
1188     QEMUSnapshotInfo sn;
1189     Error *local_err = NULL;
1190     SnapshotInfo *info = NULL;
1191     int ret;
1192 
1193     bs = qmp_get_root_bs(device, errp);
1194     if (!bs) {
1195         return NULL;
1196     }
1197     aio_context = bdrv_get_aio_context(bs);
1198     aio_context_acquire(aio_context);
1199 
1200     if (!has_id) {
1201         id = NULL;
1202     }
1203 
1204     if (!has_name) {
1205         name = NULL;
1206     }
1207 
1208     if (!id && !name) {
1209         error_setg(errp, "Name or id must be provided");
1210         goto out_aio_context;
1211     }
1212 
1213     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_INTERNAL_SNAPSHOT_DELETE, errp)) {
1214         goto out_aio_context;
1215     }
1216 
1217     ret = bdrv_snapshot_find_by_id_and_name(bs, id, name, &sn, &local_err);
1218     if (local_err) {
1219         error_propagate(errp, local_err);
1220         goto out_aio_context;
1221     }
1222     if (!ret) {
1223         error_setg(errp,
1224                    "Snapshot with id '%s' and name '%s' does not exist on "
1225                    "device '%s'",
1226                    STR_OR_NULL(id), STR_OR_NULL(name), device);
1227         goto out_aio_context;
1228     }
1229 
1230     bdrv_snapshot_delete(bs, id, name, &local_err);
1231     if (local_err) {
1232         error_propagate(errp, local_err);
1233         goto out_aio_context;
1234     }
1235 
1236     aio_context_release(aio_context);
1237 
1238     info = g_new0(SnapshotInfo, 1);
1239     info->id = g_strdup(sn.id_str);
1240     info->name = g_strdup(sn.name);
1241     info->date_nsec = sn.date_nsec;
1242     info->date_sec = sn.date_sec;
1243     info->vm_state_size = sn.vm_state_size;
1244     info->vm_clock_nsec = sn.vm_clock_nsec % 1000000000;
1245     info->vm_clock_sec = sn.vm_clock_nsec / 1000000000;
1246 
1247     return info;
1248 
1249 out_aio_context:
1250     aio_context_release(aio_context);
1251     return NULL;
1252 }
1253 
1254 /**
1255  * block_dirty_bitmap_lookup:
1256  * Return a dirty bitmap (if present), after validating
1257  * the node reference and bitmap names.
1258  *
1259  * @node: The name of the BDS node to search for bitmaps
1260  * @name: The name of the bitmap to search for
1261  * @pbs: Output pointer for BDS lookup, if desired. Can be NULL.
1262  * @errp: Output pointer for error information. Can be NULL.
1263  *
1264  * @return: A bitmap object on success, or NULL on failure.
1265  */
1266 static BdrvDirtyBitmap *block_dirty_bitmap_lookup(const char *node,
1267                                                   const char *name,
1268                                                   BlockDriverState **pbs,
1269                                                   Error **errp)
1270 {
1271     BlockDriverState *bs;
1272     BdrvDirtyBitmap *bitmap;
1273 
1274     if (!node) {
1275         error_setg(errp, "Node cannot be NULL");
1276         return NULL;
1277     }
1278     if (!name) {
1279         error_setg(errp, "Bitmap name cannot be NULL");
1280         return NULL;
1281     }
1282     bs = bdrv_lookup_bs(node, node, NULL);
1283     if (!bs) {
1284         error_setg(errp, "Node '%s' not found", node);
1285         return NULL;
1286     }
1287 
1288     bitmap = bdrv_find_dirty_bitmap(bs, name);
1289     if (!bitmap) {
1290         error_setg(errp, "Dirty bitmap '%s' not found", name);
1291         return NULL;
1292     }
1293 
1294     if (pbs) {
1295         *pbs = bs;
1296     }
1297 
1298     return bitmap;
1299 }
1300 
1301 /* New and old BlockDriverState structs for atomic group operations */
1302 
1303 typedef struct BlkActionState BlkActionState;
1304 
1305 /**
1306  * BlkActionOps:
1307  * Table of operations that define an Action.
1308  *
1309  * @instance_size: Size of state struct, in bytes.
1310  * @prepare: Prepare the work, must NOT be NULL.
1311  * @commit: Commit the changes, can be NULL.
1312  * @abort: Abort the changes on fail, can be NULL.
1313  * @clean: Clean up resources after all transaction actions have called
1314  *         commit() or abort(). Can be NULL.
1315  *
1316  * Only prepare() may fail. In a single transaction, only one of commit() or
1317  * abort() will be called. clean() will always be called if it is present.
1318  */
1319 typedef struct BlkActionOps {
1320     size_t instance_size;
1321     void (*prepare)(BlkActionState *common, Error **errp);
1322     void (*commit)(BlkActionState *common);
1323     void (*abort)(BlkActionState *common);
1324     void (*clean)(BlkActionState *common);
1325 } BlkActionOps;
1326 
1327 /**
1328  * BlkActionState:
1329  * Describes one Action's state within a Transaction.
1330  *
1331  * @action: QAPI-defined enum identifying which Action to perform.
1332  * @ops: Table of ActionOps this Action can perform.
1333  * @block_job_txn: Transaction which this action belongs to.
1334  * @entry: List membership for all Actions in this Transaction.
1335  *
1336  * This structure must be arranged as first member in a subclassed type,
1337  * assuming that the compiler will also arrange it to the same offsets as the
1338  * base class.
1339  */
1340 struct BlkActionState {
1341     TransactionAction *action;
1342     const BlkActionOps *ops;
1343     JobTxn *block_job_txn;
1344     TransactionProperties *txn_props;
1345     QTAILQ_ENTRY(BlkActionState) entry;
1346 };
1347 
1348 /* internal snapshot private data */
1349 typedef struct InternalSnapshotState {
1350     BlkActionState common;
1351     BlockDriverState *bs;
1352     QEMUSnapshotInfo sn;
1353     bool created;
1354 } InternalSnapshotState;
1355 
1356 
1357 static int action_check_completion_mode(BlkActionState *s, Error **errp)
1358 {
1359     if (s->txn_props->completion_mode != ACTION_COMPLETION_MODE_INDIVIDUAL) {
1360         error_setg(errp,
1361                    "Action '%s' does not support Transaction property "
1362                    "completion-mode = %s",
1363                    TransactionActionKind_str(s->action->type),
1364                    ActionCompletionMode_str(s->txn_props->completion_mode));
1365         return -1;
1366     }
1367     return 0;
1368 }
1369 
1370 static void internal_snapshot_prepare(BlkActionState *common,
1371                                       Error **errp)
1372 {
1373     Error *local_err = NULL;
1374     const char *device;
1375     const char *name;
1376     BlockDriverState *bs;
1377     QEMUSnapshotInfo old_sn, *sn;
1378     bool ret;
1379     qemu_timeval tv;
1380     BlockdevSnapshotInternal *internal;
1381     InternalSnapshotState *state;
1382     AioContext *aio_context;
1383     int ret1;
1384 
1385     g_assert(common->action->type ==
1386              TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC);
1387     internal = common->action->u.blockdev_snapshot_internal_sync.data;
1388     state = DO_UPCAST(InternalSnapshotState, common, common);
1389 
1390     /* 1. parse input */
1391     device = internal->device;
1392     name = internal->name;
1393 
1394     /* 2. check for validation */
1395     if (action_check_completion_mode(common, errp) < 0) {
1396         return;
1397     }
1398 
1399     bs = qmp_get_root_bs(device, errp);
1400     if (!bs) {
1401         return;
1402     }
1403 
1404     aio_context = bdrv_get_aio_context(bs);
1405     aio_context_acquire(aio_context);
1406 
1407     state->bs = bs;
1408 
1409     /* Paired with .clean() */
1410     bdrv_drained_begin(bs);
1411 
1412     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_INTERNAL_SNAPSHOT, errp)) {
1413         goto out;
1414     }
1415 
1416     if (bdrv_is_read_only(bs)) {
1417         error_setg(errp, "Device '%s' is read only", device);
1418         goto out;
1419     }
1420 
1421     if (!bdrv_can_snapshot(bs)) {
1422         error_setg(errp, "Block format '%s' used by device '%s' "
1423                    "does not support internal snapshots",
1424                    bs->drv->format_name, device);
1425         goto out;
1426     }
1427 
1428     if (!strlen(name)) {
1429         error_setg(errp, "Name is empty");
1430         goto out;
1431     }
1432 
1433     /* check whether a snapshot with name exist */
1434     ret = bdrv_snapshot_find_by_id_and_name(bs, NULL, name, &old_sn,
1435                                             &local_err);
1436     if (local_err) {
1437         error_propagate(errp, local_err);
1438         goto out;
1439     } else if (ret) {
1440         error_setg(errp,
1441                    "Snapshot with name '%s' already exists on device '%s'",
1442                    name, device);
1443         goto out;
1444     }
1445 
1446     /* 3. take the snapshot */
1447     sn = &state->sn;
1448     pstrcpy(sn->name, sizeof(sn->name), name);
1449     qemu_gettimeofday(&tv);
1450     sn->date_sec = tv.tv_sec;
1451     sn->date_nsec = tv.tv_usec * 1000;
1452     sn->vm_clock_nsec = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL);
1453 
1454     ret1 = bdrv_snapshot_create(bs, sn);
1455     if (ret1 < 0) {
1456         error_setg_errno(errp, -ret1,
1457                          "Failed to create snapshot '%s' on device '%s'",
1458                          name, device);
1459         goto out;
1460     }
1461 
1462     /* 4. succeed, mark a snapshot is created */
1463     state->created = true;
1464 
1465 out:
1466     aio_context_release(aio_context);
1467 }
1468 
1469 static void internal_snapshot_abort(BlkActionState *common)
1470 {
1471     InternalSnapshotState *state =
1472                              DO_UPCAST(InternalSnapshotState, common, common);
1473     BlockDriverState *bs = state->bs;
1474     QEMUSnapshotInfo *sn = &state->sn;
1475     AioContext *aio_context;
1476     Error *local_error = NULL;
1477 
1478     if (!state->created) {
1479         return;
1480     }
1481 
1482     aio_context = bdrv_get_aio_context(state->bs);
1483     aio_context_acquire(aio_context);
1484 
1485     if (bdrv_snapshot_delete(bs, sn->id_str, sn->name, &local_error) < 0) {
1486         error_reportf_err(local_error,
1487                           "Failed to delete snapshot with id '%s' and "
1488                           "name '%s' on device '%s' in abort: ",
1489                           sn->id_str, sn->name,
1490                           bdrv_get_device_name(bs));
1491     }
1492 
1493     aio_context_release(aio_context);
1494 }
1495 
1496 static void internal_snapshot_clean(BlkActionState *common)
1497 {
1498     InternalSnapshotState *state = DO_UPCAST(InternalSnapshotState,
1499                                              common, common);
1500     AioContext *aio_context;
1501 
1502     if (!state->bs) {
1503         return;
1504     }
1505 
1506     aio_context = bdrv_get_aio_context(state->bs);
1507     aio_context_acquire(aio_context);
1508 
1509     bdrv_drained_end(state->bs);
1510 
1511     aio_context_release(aio_context);
1512 }
1513 
1514 /* external snapshot private data */
1515 typedef struct ExternalSnapshotState {
1516     BlkActionState common;
1517     BlockDriverState *old_bs;
1518     BlockDriverState *new_bs;
1519     bool overlay_appended;
1520 } ExternalSnapshotState;
1521 
1522 static void external_snapshot_prepare(BlkActionState *common,
1523                                       Error **errp)
1524 {
1525     int flags = 0;
1526     QDict *options = NULL;
1527     Error *local_err = NULL;
1528     /* Device and node name of the image to generate the snapshot from */
1529     const char *device;
1530     const char *node_name;
1531     /* Reference to the new image (for 'blockdev-snapshot') */
1532     const char *snapshot_ref;
1533     /* File name of the new image (for 'blockdev-snapshot-sync') */
1534     const char *new_image_file;
1535     ExternalSnapshotState *state =
1536                              DO_UPCAST(ExternalSnapshotState, common, common);
1537     TransactionAction *action = common->action;
1538     AioContext *aio_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     ret = bdrv_try_set_aio_context(state->new_bs, aio_context, errp);
1680     if (ret < 0) {
1681         goto out;
1682     }
1683 
1684     /* This removes our old bs and adds the new bs. This is an operation that
1685      * can fail, so we need to do it in .prepare; undoing it for abort is
1686      * always possible. */
1687     bdrv_ref(state->new_bs);
1688     bdrv_append(state->new_bs, state->old_bs, &local_err);
1689     if (local_err) {
1690         error_propagate(errp, local_err);
1691         goto out;
1692     }
1693     state->overlay_appended = true;
1694 
1695 out:
1696     aio_context_release(aio_context);
1697 }
1698 
1699 static void external_snapshot_commit(BlkActionState *common)
1700 {
1701     ExternalSnapshotState *state =
1702                              DO_UPCAST(ExternalSnapshotState, common, common);
1703     AioContext *aio_context;
1704 
1705     aio_context = bdrv_get_aio_context(state->old_bs);
1706     aio_context_acquire(aio_context);
1707 
1708     /* We don't need (or want) to use the transactional
1709      * bdrv_reopen_multiple() across all the entries at once, because we
1710      * don't want to abort all of them if one of them fails the reopen */
1711     if (!atomic_read(&state->old_bs->copy_on_read)) {
1712         bdrv_reopen_set_read_only(state->old_bs, true, NULL);
1713     }
1714 
1715     aio_context_release(aio_context);
1716 }
1717 
1718 static void external_snapshot_abort(BlkActionState *common)
1719 {
1720     ExternalSnapshotState *state =
1721                              DO_UPCAST(ExternalSnapshotState, common, common);
1722     if (state->new_bs) {
1723         if (state->overlay_appended) {
1724             AioContext *aio_context;
1725 
1726             aio_context = bdrv_get_aio_context(state->old_bs);
1727             aio_context_acquire(aio_context);
1728 
1729             bdrv_ref(state->old_bs);   /* we can't let bdrv_set_backind_hd()
1730                                           close state->old_bs; we need it */
1731             bdrv_set_backing_hd(state->new_bs, NULL, &error_abort);
1732             bdrv_replace_node(state->new_bs, state->old_bs, &error_abort);
1733             bdrv_unref(state->old_bs); /* bdrv_replace_node() ref'ed old_bs */
1734 
1735             aio_context_release(aio_context);
1736         }
1737     }
1738 }
1739 
1740 static void external_snapshot_clean(BlkActionState *common)
1741 {
1742     ExternalSnapshotState *state =
1743                              DO_UPCAST(ExternalSnapshotState, common, common);
1744     AioContext *aio_context;
1745 
1746     if (!state->old_bs) {
1747         return;
1748     }
1749 
1750     aio_context = bdrv_get_aio_context(state->old_bs);
1751     aio_context_acquire(aio_context);
1752 
1753     bdrv_drained_end(state->old_bs);
1754     bdrv_unref(state->new_bs);
1755 
1756     aio_context_release(aio_context);
1757 }
1758 
1759 typedef struct DriveBackupState {
1760     BlkActionState common;
1761     BlockDriverState *bs;
1762     BlockJob *job;
1763 } DriveBackupState;
1764 
1765 static BlockJob *do_drive_backup(DriveBackup *backup, JobTxn *txn,
1766                             Error **errp);
1767 
1768 static void drive_backup_prepare(BlkActionState *common, Error **errp)
1769 {
1770     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1771     BlockDriverState *bs;
1772     DriveBackup *backup;
1773     AioContext *aio_context;
1774     Error *local_err = NULL;
1775 
1776     assert(common->action->type == TRANSACTION_ACTION_KIND_DRIVE_BACKUP);
1777     backup = common->action->u.drive_backup.data;
1778 
1779     bs = bdrv_lookup_bs(backup->device, backup->device, errp);
1780     if (!bs) {
1781         return;
1782     }
1783 
1784     aio_context = bdrv_get_aio_context(bs);
1785     aio_context_acquire(aio_context);
1786 
1787     /* Paired with .clean() */
1788     bdrv_drained_begin(bs);
1789 
1790     state->bs = bs;
1791 
1792     state->job = do_drive_backup(backup, common->block_job_txn, &local_err);
1793     if (local_err) {
1794         error_propagate(errp, local_err);
1795         goto out;
1796     }
1797 
1798 out:
1799     aio_context_release(aio_context);
1800 }
1801 
1802 static void drive_backup_commit(BlkActionState *common)
1803 {
1804     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1805     AioContext *aio_context;
1806 
1807     aio_context = bdrv_get_aio_context(state->bs);
1808     aio_context_acquire(aio_context);
1809 
1810     assert(state->job);
1811     job_start(&state->job->job);
1812 
1813     aio_context_release(aio_context);
1814 }
1815 
1816 static void drive_backup_abort(BlkActionState *common)
1817 {
1818     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1819 
1820     if (state->job) {
1821         AioContext *aio_context;
1822 
1823         aio_context = bdrv_get_aio_context(state->bs);
1824         aio_context_acquire(aio_context);
1825 
1826         job_cancel_sync(&state->job->job);
1827 
1828         aio_context_release(aio_context);
1829     }
1830 }
1831 
1832 static void drive_backup_clean(BlkActionState *common)
1833 {
1834     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1835     AioContext *aio_context;
1836 
1837     if (!state->bs) {
1838         return;
1839     }
1840 
1841     aio_context = bdrv_get_aio_context(state->bs);
1842     aio_context_acquire(aio_context);
1843 
1844     bdrv_drained_end(state->bs);
1845 
1846     aio_context_release(aio_context);
1847 }
1848 
1849 typedef struct BlockdevBackupState {
1850     BlkActionState common;
1851     BlockDriverState *bs;
1852     BlockJob *job;
1853 } BlockdevBackupState;
1854 
1855 static BlockJob *do_blockdev_backup(BlockdevBackup *backup, JobTxn *txn,
1856                                     Error **errp);
1857 
1858 static void blockdev_backup_prepare(BlkActionState *common, Error **errp)
1859 {
1860     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1861     BlockdevBackup *backup;
1862     BlockDriverState *bs, *target;
1863     AioContext *aio_context;
1864     Error *local_err = NULL;
1865 
1866     assert(common->action->type == TRANSACTION_ACTION_KIND_BLOCKDEV_BACKUP);
1867     backup = common->action->u.blockdev_backup.data;
1868 
1869     bs = bdrv_lookup_bs(backup->device, backup->device, errp);
1870     if (!bs) {
1871         return;
1872     }
1873 
1874     target = bdrv_lookup_bs(backup->target, backup->target, errp);
1875     if (!target) {
1876         return;
1877     }
1878 
1879     aio_context = bdrv_get_aio_context(bs);
1880     aio_context_acquire(aio_context);
1881     state->bs = bs;
1882 
1883     /* Paired with .clean() */
1884     bdrv_drained_begin(state->bs);
1885 
1886     state->job = do_blockdev_backup(backup, common->block_job_txn, &local_err);
1887     if (local_err) {
1888         error_propagate(errp, local_err);
1889         goto out;
1890     }
1891 
1892 out:
1893     aio_context_release(aio_context);
1894 }
1895 
1896 static void blockdev_backup_commit(BlkActionState *common)
1897 {
1898     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1899     AioContext *aio_context;
1900 
1901     aio_context = bdrv_get_aio_context(state->bs);
1902     aio_context_acquire(aio_context);
1903 
1904     assert(state->job);
1905     job_start(&state->job->job);
1906 
1907     aio_context_release(aio_context);
1908 }
1909 
1910 static void blockdev_backup_abort(BlkActionState *common)
1911 {
1912     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1913 
1914     if (state->job) {
1915         AioContext *aio_context;
1916 
1917         aio_context = bdrv_get_aio_context(state->bs);
1918         aio_context_acquire(aio_context);
1919 
1920         job_cancel_sync(&state->job->job);
1921 
1922         aio_context_release(aio_context);
1923     }
1924 }
1925 
1926 static void blockdev_backup_clean(BlkActionState *common)
1927 {
1928     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1929     AioContext *aio_context;
1930 
1931     if (!state->bs) {
1932         return;
1933     }
1934 
1935     aio_context = bdrv_get_aio_context(state->bs);
1936     aio_context_acquire(aio_context);
1937 
1938     bdrv_drained_end(state->bs);
1939 
1940     aio_context_release(aio_context);
1941 }
1942 
1943 typedef struct BlockDirtyBitmapState {
1944     BlkActionState common;
1945     BdrvDirtyBitmap *bitmap;
1946     BlockDriverState *bs;
1947     HBitmap *backup;
1948     bool prepared;
1949     bool was_enabled;
1950 } BlockDirtyBitmapState;
1951 
1952 static void block_dirty_bitmap_add_prepare(BlkActionState *common,
1953                                            Error **errp)
1954 {
1955     Error *local_err = NULL;
1956     BlockDirtyBitmapAdd *action;
1957     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
1958                                              common, common);
1959 
1960     if (action_check_completion_mode(common, errp) < 0) {
1961         return;
1962     }
1963 
1964     action = common->action->u.block_dirty_bitmap_add.data;
1965     /* AIO context taken and released within qmp_block_dirty_bitmap_add */
1966     qmp_block_dirty_bitmap_add(action->node, action->name,
1967                                action->has_granularity, action->granularity,
1968                                action->has_persistent, action->persistent,
1969                                action->has_autoload, action->autoload,
1970                                action->has_disabled, action->disabled,
1971                                &local_err);
1972 
1973     if (!local_err) {
1974         state->prepared = true;
1975     } else {
1976         error_propagate(errp, local_err);
1977     }
1978 }
1979 
1980 static void block_dirty_bitmap_add_abort(BlkActionState *common)
1981 {
1982     BlockDirtyBitmapAdd *action;
1983     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
1984                                              common, common);
1985 
1986     action = common->action->u.block_dirty_bitmap_add.data;
1987     /* Should not be able to fail: IF the bitmap was added via .prepare(),
1988      * then the node reference and bitmap name must have been valid.
1989      */
1990     if (state->prepared) {
1991         qmp_block_dirty_bitmap_remove(action->node, action->name, &error_abort);
1992     }
1993 }
1994 
1995 static void block_dirty_bitmap_clear_prepare(BlkActionState *common,
1996                                              Error **errp)
1997 {
1998     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
1999                                              common, common);
2000     BlockDirtyBitmap *action;
2001 
2002     if (action_check_completion_mode(common, errp) < 0) {
2003         return;
2004     }
2005 
2006     action = common->action->u.block_dirty_bitmap_clear.data;
2007     state->bitmap = block_dirty_bitmap_lookup(action->node,
2008                                               action->name,
2009                                               &state->bs,
2010                                               errp);
2011     if (!state->bitmap) {
2012         return;
2013     }
2014 
2015     if (bdrv_dirty_bitmap_check(state->bitmap, BDRV_BITMAP_DEFAULT, errp)) {
2016         return;
2017     }
2018 
2019     bdrv_clear_dirty_bitmap(state->bitmap, &state->backup);
2020 }
2021 
2022 static void block_dirty_bitmap_restore(BlkActionState *common)
2023 {
2024     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2025                                              common, common);
2026 
2027     if (state->backup) {
2028         bdrv_restore_dirty_bitmap(state->bitmap, state->backup);
2029     }
2030 }
2031 
2032 static void block_dirty_bitmap_free_backup(BlkActionState *common)
2033 {
2034     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2035                                              common, common);
2036 
2037     hbitmap_free(state->backup);
2038 }
2039 
2040 static void block_dirty_bitmap_enable_prepare(BlkActionState *common,
2041                                               Error **errp)
2042 {
2043     BlockDirtyBitmap *action;
2044     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2045                                              common, common);
2046 
2047     if (action_check_completion_mode(common, errp) < 0) {
2048         return;
2049     }
2050 
2051     action = common->action->u.block_dirty_bitmap_enable.data;
2052     state->bitmap = block_dirty_bitmap_lookup(action->node,
2053                                               action->name,
2054                                               NULL,
2055                                               errp);
2056     if (!state->bitmap) {
2057         return;
2058     }
2059 
2060     if (bdrv_dirty_bitmap_check(state->bitmap, BDRV_BITMAP_ALLOW_RO, errp)) {
2061         return;
2062     }
2063 
2064     state->was_enabled = bdrv_dirty_bitmap_enabled(state->bitmap);
2065     bdrv_enable_dirty_bitmap(state->bitmap);
2066 }
2067 
2068 static void block_dirty_bitmap_enable_abort(BlkActionState *common)
2069 {
2070     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2071                                              common, common);
2072 
2073     if (!state->was_enabled) {
2074         bdrv_disable_dirty_bitmap(state->bitmap);
2075     }
2076 }
2077 
2078 static void block_dirty_bitmap_disable_prepare(BlkActionState *common,
2079                                                Error **errp)
2080 {
2081     BlockDirtyBitmap *action;
2082     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2083                                              common, common);
2084 
2085     if (action_check_completion_mode(common, errp) < 0) {
2086         return;
2087     }
2088 
2089     action = common->action->u.block_dirty_bitmap_disable.data;
2090     state->bitmap = block_dirty_bitmap_lookup(action->node,
2091                                               action->name,
2092                                               NULL,
2093                                               errp);
2094     if (!state->bitmap) {
2095         return;
2096     }
2097 
2098     if (bdrv_dirty_bitmap_check(state->bitmap, BDRV_BITMAP_ALLOW_RO, errp)) {
2099         return;
2100     }
2101 
2102     state->was_enabled = bdrv_dirty_bitmap_enabled(state->bitmap);
2103     bdrv_disable_dirty_bitmap(state->bitmap);
2104 }
2105 
2106 static void block_dirty_bitmap_disable_abort(BlkActionState *common)
2107 {
2108     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2109                                              common, common);
2110 
2111     if (state->was_enabled) {
2112         bdrv_enable_dirty_bitmap(state->bitmap);
2113     }
2114 }
2115 
2116 static BdrvDirtyBitmap *do_block_dirty_bitmap_merge(
2117         const char *node, const char *target,
2118         BlockDirtyBitmapMergeSourceList *bitmaps,
2119         HBitmap **backup, Error **errp);
2120 
2121 static void block_dirty_bitmap_merge_prepare(BlkActionState *common,
2122                                              Error **errp)
2123 {
2124     BlockDirtyBitmapMerge *action;
2125     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2126                                              common, common);
2127 
2128     if (action_check_completion_mode(common, errp) < 0) {
2129         return;
2130     }
2131 
2132     action = common->action->u.block_dirty_bitmap_merge.data;
2133 
2134     state->bitmap = do_block_dirty_bitmap_merge(action->node, action->target,
2135                                                 action->bitmaps, &state->backup,
2136                                                 errp);
2137 }
2138 
2139 static BdrvDirtyBitmap *do_block_dirty_bitmap_remove(
2140         const char *node, const char *name, bool release,
2141         BlockDriverState **bitmap_bs, Error **errp);
2142 
2143 static void block_dirty_bitmap_remove_prepare(BlkActionState *common,
2144                                               Error **errp)
2145 {
2146     BlockDirtyBitmap *action;
2147     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2148                                              common, common);
2149 
2150     if (action_check_completion_mode(common, errp) < 0) {
2151         return;
2152     }
2153 
2154     action = common->action->u.block_dirty_bitmap_remove.data;
2155 
2156     state->bitmap = do_block_dirty_bitmap_remove(action->node, action->name,
2157                                                  false, &state->bs, errp);
2158     if (state->bitmap) {
2159         bdrv_dirty_bitmap_skip_store(state->bitmap, true);
2160         bdrv_dirty_bitmap_set_busy(state->bitmap, true);
2161     }
2162 }
2163 
2164 static void block_dirty_bitmap_remove_abort(BlkActionState *common)
2165 {
2166     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2167                                              common, common);
2168 
2169     if (state->bitmap) {
2170         bdrv_dirty_bitmap_skip_store(state->bitmap, false);
2171         bdrv_dirty_bitmap_set_busy(state->bitmap, false);
2172     }
2173 }
2174 
2175 static void block_dirty_bitmap_remove_commit(BlkActionState *common)
2176 {
2177     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2178                                              common, common);
2179 
2180     bdrv_dirty_bitmap_set_busy(state->bitmap, false);
2181     bdrv_release_dirty_bitmap(state->bs, state->bitmap);
2182 }
2183 
2184 static void abort_prepare(BlkActionState *common, Error **errp)
2185 {
2186     error_setg(errp, "Transaction aborted using Abort action");
2187 }
2188 
2189 static void abort_commit(BlkActionState *common)
2190 {
2191     g_assert_not_reached(); /* this action never succeeds */
2192 }
2193 
2194 static const BlkActionOps actions[] = {
2195     [TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT] = {
2196         .instance_size = sizeof(ExternalSnapshotState),
2197         .prepare  = external_snapshot_prepare,
2198         .commit   = external_snapshot_commit,
2199         .abort = external_snapshot_abort,
2200         .clean = external_snapshot_clean,
2201     },
2202     [TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC] = {
2203         .instance_size = sizeof(ExternalSnapshotState),
2204         .prepare  = external_snapshot_prepare,
2205         .commit   = external_snapshot_commit,
2206         .abort = external_snapshot_abort,
2207         .clean = external_snapshot_clean,
2208     },
2209     [TRANSACTION_ACTION_KIND_DRIVE_BACKUP] = {
2210         .instance_size = sizeof(DriveBackupState),
2211         .prepare = drive_backup_prepare,
2212         .commit = drive_backup_commit,
2213         .abort = drive_backup_abort,
2214         .clean = drive_backup_clean,
2215     },
2216     [TRANSACTION_ACTION_KIND_BLOCKDEV_BACKUP] = {
2217         .instance_size = sizeof(BlockdevBackupState),
2218         .prepare = blockdev_backup_prepare,
2219         .commit = blockdev_backup_commit,
2220         .abort = blockdev_backup_abort,
2221         .clean = blockdev_backup_clean,
2222     },
2223     [TRANSACTION_ACTION_KIND_ABORT] = {
2224         .instance_size = sizeof(BlkActionState),
2225         .prepare = abort_prepare,
2226         .commit = abort_commit,
2227     },
2228     [TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC] = {
2229         .instance_size = sizeof(InternalSnapshotState),
2230         .prepare  = internal_snapshot_prepare,
2231         .abort = internal_snapshot_abort,
2232         .clean = internal_snapshot_clean,
2233     },
2234     [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_ADD] = {
2235         .instance_size = sizeof(BlockDirtyBitmapState),
2236         .prepare = block_dirty_bitmap_add_prepare,
2237         .abort = block_dirty_bitmap_add_abort,
2238     },
2239     [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_CLEAR] = {
2240         .instance_size = sizeof(BlockDirtyBitmapState),
2241         .prepare = block_dirty_bitmap_clear_prepare,
2242         .commit = block_dirty_bitmap_free_backup,
2243         .abort = block_dirty_bitmap_restore,
2244     },
2245     [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_ENABLE] = {
2246         .instance_size = sizeof(BlockDirtyBitmapState),
2247         .prepare = block_dirty_bitmap_enable_prepare,
2248         .abort = block_dirty_bitmap_enable_abort,
2249     },
2250     [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_DISABLE] = {
2251         .instance_size = sizeof(BlockDirtyBitmapState),
2252         .prepare = block_dirty_bitmap_disable_prepare,
2253         .abort = block_dirty_bitmap_disable_abort,
2254     },
2255     [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_MERGE] = {
2256         .instance_size = sizeof(BlockDirtyBitmapState),
2257         .prepare = block_dirty_bitmap_merge_prepare,
2258         .commit = block_dirty_bitmap_free_backup,
2259         .abort = block_dirty_bitmap_restore,
2260     },
2261     [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_REMOVE] = {
2262         .instance_size = sizeof(BlockDirtyBitmapState),
2263         .prepare = block_dirty_bitmap_remove_prepare,
2264         .commit = block_dirty_bitmap_remove_commit,
2265         .abort = block_dirty_bitmap_remove_abort,
2266     },
2267     /* Where are transactions for MIRROR, COMMIT and STREAM?
2268      * Although these blockjobs use transaction callbacks like the backup job,
2269      * these jobs do not necessarily adhere to transaction semantics.
2270      * These jobs may not fully undo all of their actions on abort, nor do they
2271      * necessarily work in transactions with more than one job in them.
2272      */
2273 };
2274 
2275 /**
2276  * Allocate a TransactionProperties structure if necessary, and fill
2277  * that structure with desired defaults if they are unset.
2278  */
2279 static TransactionProperties *get_transaction_properties(
2280     TransactionProperties *props)
2281 {
2282     if (!props) {
2283         props = g_new0(TransactionProperties, 1);
2284     }
2285 
2286     if (!props->has_completion_mode) {
2287         props->has_completion_mode = true;
2288         props->completion_mode = ACTION_COMPLETION_MODE_INDIVIDUAL;
2289     }
2290 
2291     return props;
2292 }
2293 
2294 /*
2295  * 'Atomic' group operations.  The operations are performed as a set, and if
2296  * any fail then we roll back all operations in the group.
2297  */
2298 void qmp_transaction(TransactionActionList *dev_list,
2299                      bool has_props,
2300                      struct TransactionProperties *props,
2301                      Error **errp)
2302 {
2303     TransactionActionList *dev_entry = dev_list;
2304     JobTxn *block_job_txn = NULL;
2305     BlkActionState *state, *next;
2306     Error *local_err = NULL;
2307 
2308     QTAILQ_HEAD(, BlkActionState) snap_bdrv_states;
2309     QTAILQ_INIT(&snap_bdrv_states);
2310 
2311     /* Does this transaction get canceled as a group on failure?
2312      * If not, we don't really need to make a JobTxn.
2313      */
2314     props = get_transaction_properties(props);
2315     if (props->completion_mode != ACTION_COMPLETION_MODE_INDIVIDUAL) {
2316         block_job_txn = job_txn_new();
2317     }
2318 
2319     /* drain all i/o before any operations */
2320     bdrv_drain_all();
2321 
2322     /* We don't do anything in this loop that commits us to the operations */
2323     while (NULL != dev_entry) {
2324         TransactionAction *dev_info = NULL;
2325         const BlkActionOps *ops;
2326 
2327         dev_info = dev_entry->value;
2328         dev_entry = dev_entry->next;
2329 
2330         assert(dev_info->type < ARRAY_SIZE(actions));
2331 
2332         ops = &actions[dev_info->type];
2333         assert(ops->instance_size > 0);
2334 
2335         state = g_malloc0(ops->instance_size);
2336         state->ops = ops;
2337         state->action = dev_info;
2338         state->block_job_txn = block_job_txn;
2339         state->txn_props = props;
2340         QTAILQ_INSERT_TAIL(&snap_bdrv_states, state, entry);
2341 
2342         state->ops->prepare(state, &local_err);
2343         if (local_err) {
2344             error_propagate(errp, local_err);
2345             goto delete_and_fail;
2346         }
2347     }
2348 
2349     QTAILQ_FOREACH(state, &snap_bdrv_states, entry) {
2350         if (state->ops->commit) {
2351             state->ops->commit(state);
2352         }
2353     }
2354 
2355     /* success */
2356     goto exit;
2357 
2358 delete_and_fail:
2359     /* failure, and it is all-or-none; roll back all operations */
2360     QTAILQ_FOREACH_REVERSE(state, &snap_bdrv_states, entry) {
2361         if (state->ops->abort) {
2362             state->ops->abort(state);
2363         }
2364     }
2365 exit:
2366     QTAILQ_FOREACH_SAFE(state, &snap_bdrv_states, entry, next) {
2367         if (state->ops->clean) {
2368             state->ops->clean(state);
2369         }
2370         g_free(state);
2371     }
2372     if (!has_props) {
2373         qapi_free_TransactionProperties(props);
2374     }
2375     job_txn_unref(block_job_txn);
2376 }
2377 
2378 void qmp_eject(bool has_device, const char *device,
2379                bool has_id, const char *id,
2380                bool has_force, bool force, Error **errp)
2381 {
2382     Error *local_err = NULL;
2383     int rc;
2384 
2385     if (!has_force) {
2386         force = false;
2387     }
2388 
2389     rc = do_open_tray(has_device ? device : NULL,
2390                       has_id ? id : NULL,
2391                       force, &local_err);
2392     if (rc && rc != -ENOSYS) {
2393         error_propagate(errp, local_err);
2394         return;
2395     }
2396     error_free(local_err);
2397 
2398     blockdev_remove_medium(has_device, device, has_id, id, errp);
2399 }
2400 
2401 void qmp_block_passwd(bool has_device, const char *device,
2402                       bool has_node_name, const char *node_name,
2403                       const char *password, Error **errp)
2404 {
2405     error_setg(errp,
2406                "Setting block passwords directly is no longer supported");
2407 }
2408 
2409 /*
2410  * Attempt to open the tray of @device.
2411  * If @force, ignore its tray lock.
2412  * Else, if the tray is locked, don't open it, but ask the guest to open it.
2413  * On error, store an error through @errp and return -errno.
2414  * If @device does not exist, return -ENODEV.
2415  * If it has no removable media, return -ENOTSUP.
2416  * If it has no tray, return -ENOSYS.
2417  * If the guest was asked to open the tray, return -EINPROGRESS.
2418  * Else, return 0.
2419  */
2420 static int do_open_tray(const char *blk_name, const char *qdev_id,
2421                         bool force, Error **errp)
2422 {
2423     BlockBackend *blk;
2424     const char *device = qdev_id ?: blk_name;
2425     bool locked;
2426 
2427     blk = qmp_get_blk(blk_name, qdev_id, errp);
2428     if (!blk) {
2429         return -ENODEV;
2430     }
2431 
2432     if (!blk_dev_has_removable_media(blk)) {
2433         error_setg(errp, "Device '%s' is not removable", device);
2434         return -ENOTSUP;
2435     }
2436 
2437     if (!blk_dev_has_tray(blk)) {
2438         error_setg(errp, "Device '%s' does not have a tray", device);
2439         return -ENOSYS;
2440     }
2441 
2442     if (blk_dev_is_tray_open(blk)) {
2443         return 0;
2444     }
2445 
2446     locked = blk_dev_is_medium_locked(blk);
2447     if (locked) {
2448         blk_dev_eject_request(blk, force);
2449     }
2450 
2451     if (!locked || force) {
2452         blk_dev_change_media_cb(blk, false, &error_abort);
2453     }
2454 
2455     if (locked && !force) {
2456         error_setg(errp, "Device '%s' is locked and force was not specified, "
2457                    "wait for tray to open and try again", device);
2458         return -EINPROGRESS;
2459     }
2460 
2461     return 0;
2462 }
2463 
2464 void qmp_blockdev_open_tray(bool has_device, const char *device,
2465                             bool has_id, const char *id,
2466                             bool has_force, bool force,
2467                             Error **errp)
2468 {
2469     Error *local_err = NULL;
2470     int rc;
2471 
2472     if (!has_force) {
2473         force = false;
2474     }
2475     rc = do_open_tray(has_device ? device : NULL,
2476                       has_id ? id : NULL,
2477                       force, &local_err);
2478     if (rc && rc != -ENOSYS && rc != -EINPROGRESS) {
2479         error_propagate(errp, local_err);
2480         return;
2481     }
2482     error_free(local_err);
2483 }
2484 
2485 void qmp_blockdev_close_tray(bool has_device, const char *device,
2486                              bool has_id, const char *id,
2487                              Error **errp)
2488 {
2489     BlockBackend *blk;
2490     Error *local_err = NULL;
2491 
2492     device = has_device ? device : NULL;
2493     id = has_id ? id : NULL;
2494 
2495     blk = qmp_get_blk(device, id, errp);
2496     if (!blk) {
2497         return;
2498     }
2499 
2500     if (!blk_dev_has_removable_media(blk)) {
2501         error_setg(errp, "Device '%s' is not removable", device ?: id);
2502         return;
2503     }
2504 
2505     if (!blk_dev_has_tray(blk)) {
2506         /* Ignore this command on tray-less devices */
2507         return;
2508     }
2509 
2510     if (!blk_dev_is_tray_open(blk)) {
2511         return;
2512     }
2513 
2514     blk_dev_change_media_cb(blk, true, &local_err);
2515     if (local_err) {
2516         error_propagate(errp, local_err);
2517         return;
2518     }
2519 }
2520 
2521 static void blockdev_remove_medium(bool has_device, const char *device,
2522                                    bool has_id, const char *id, Error **errp)
2523 {
2524     BlockBackend *blk;
2525     BlockDriverState *bs;
2526     AioContext *aio_context;
2527     bool has_attached_device;
2528 
2529     device = has_device ? device : NULL;
2530     id = has_id ? id : NULL;
2531 
2532     blk = qmp_get_blk(device, id, errp);
2533     if (!blk) {
2534         return;
2535     }
2536 
2537     /* For BBs without a device, we can exchange the BDS tree at will */
2538     has_attached_device = blk_get_attached_dev(blk);
2539 
2540     if (has_attached_device && !blk_dev_has_removable_media(blk)) {
2541         error_setg(errp, "Device '%s' is not removable", device ?: id);
2542         return;
2543     }
2544 
2545     if (has_attached_device && blk_dev_has_tray(blk) &&
2546         !blk_dev_is_tray_open(blk))
2547     {
2548         error_setg(errp, "Tray of device '%s' is not open", device ?: id);
2549         return;
2550     }
2551 
2552     bs = blk_bs(blk);
2553     if (!bs) {
2554         return;
2555     }
2556 
2557     aio_context = bdrv_get_aio_context(bs);
2558     aio_context_acquire(aio_context);
2559 
2560     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_EJECT, errp)) {
2561         goto out;
2562     }
2563 
2564     blk_remove_bs(blk);
2565 
2566     if (!blk_dev_has_tray(blk)) {
2567         /* For tray-less devices, blockdev-open-tray is a no-op (or may not be
2568          * called at all); therefore, the medium needs to be ejected here.
2569          * Do it after blk_remove_bs() so blk_is_inserted(blk) returns the @load
2570          * value passed here (i.e. false). */
2571         blk_dev_change_media_cb(blk, false, &error_abort);
2572     }
2573 
2574 out:
2575     aio_context_release(aio_context);
2576 }
2577 
2578 void qmp_blockdev_remove_medium(const char *id, Error **errp)
2579 {
2580     blockdev_remove_medium(false, NULL, true, id, errp);
2581 }
2582 
2583 static void qmp_blockdev_insert_anon_medium(BlockBackend *blk,
2584                                             BlockDriverState *bs, Error **errp)
2585 {
2586     Error *local_err = NULL;
2587     bool has_device;
2588     int ret;
2589 
2590     /* For BBs without a device, we can exchange the BDS tree at will */
2591     has_device = blk_get_attached_dev(blk);
2592 
2593     if (has_device && !blk_dev_has_removable_media(blk)) {
2594         error_setg(errp, "Device is not removable");
2595         return;
2596     }
2597 
2598     if (has_device && blk_dev_has_tray(blk) && !blk_dev_is_tray_open(blk)) {
2599         error_setg(errp, "Tray of the device is not open");
2600         return;
2601     }
2602 
2603     if (blk_bs(blk)) {
2604         error_setg(errp, "There already is a medium in the device");
2605         return;
2606     }
2607 
2608     ret = blk_insert_bs(blk, bs, errp);
2609     if (ret < 0) {
2610         return;
2611     }
2612 
2613     if (!blk_dev_has_tray(blk)) {
2614         /* For tray-less devices, blockdev-close-tray is a no-op (or may not be
2615          * called at all); therefore, the medium needs to be pushed into the
2616          * slot here.
2617          * Do it after blk_insert_bs() so blk_is_inserted(blk) returns the @load
2618          * value passed here (i.e. true). */
2619         blk_dev_change_media_cb(blk, true, &local_err);
2620         if (local_err) {
2621             error_propagate(errp, local_err);
2622             blk_remove_bs(blk);
2623             return;
2624         }
2625     }
2626 }
2627 
2628 static void blockdev_insert_medium(bool has_device, const char *device,
2629                                    bool has_id, const char *id,
2630                                    const char *node_name, Error **errp)
2631 {
2632     BlockBackend *blk;
2633     BlockDriverState *bs;
2634 
2635     blk = qmp_get_blk(has_device ? device : NULL,
2636                       has_id ? id : NULL,
2637                       errp);
2638     if (!blk) {
2639         return;
2640     }
2641 
2642     bs = bdrv_find_node(node_name);
2643     if (!bs) {
2644         error_setg(errp, "Node '%s' not found", node_name);
2645         return;
2646     }
2647 
2648     if (bdrv_has_blk(bs)) {
2649         error_setg(errp, "Node '%s' is already in use", node_name);
2650         return;
2651     }
2652 
2653     qmp_blockdev_insert_anon_medium(blk, bs, errp);
2654 }
2655 
2656 void qmp_blockdev_insert_medium(const char *id, const char *node_name,
2657                                 Error **errp)
2658 {
2659     blockdev_insert_medium(false, NULL, true, id, node_name, errp);
2660 }
2661 
2662 void qmp_blockdev_change_medium(bool has_device, const char *device,
2663                                 bool has_id, const char *id,
2664                                 const char *filename,
2665                                 bool has_format, const char *format,
2666                                 bool has_read_only,
2667                                 BlockdevChangeReadOnlyMode read_only,
2668                                 Error **errp)
2669 {
2670     BlockBackend *blk;
2671     BlockDriverState *medium_bs = NULL;
2672     int bdrv_flags;
2673     bool detect_zeroes;
2674     int rc;
2675     QDict *options = NULL;
2676     Error *err = NULL;
2677 
2678     blk = qmp_get_blk(has_device ? device : NULL,
2679                       has_id ? id : NULL,
2680                       errp);
2681     if (!blk) {
2682         goto fail;
2683     }
2684 
2685     if (blk_bs(blk)) {
2686         blk_update_root_state(blk);
2687     }
2688 
2689     bdrv_flags = blk_get_open_flags_from_root_state(blk);
2690     bdrv_flags &= ~(BDRV_O_TEMPORARY | BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING |
2691         BDRV_O_PROTOCOL | BDRV_O_AUTO_RDONLY);
2692 
2693     if (!has_read_only) {
2694         read_only = BLOCKDEV_CHANGE_READ_ONLY_MODE_RETAIN;
2695     }
2696 
2697     switch (read_only) {
2698     case BLOCKDEV_CHANGE_READ_ONLY_MODE_RETAIN:
2699         break;
2700 
2701     case BLOCKDEV_CHANGE_READ_ONLY_MODE_READ_ONLY:
2702         bdrv_flags &= ~BDRV_O_RDWR;
2703         break;
2704 
2705     case BLOCKDEV_CHANGE_READ_ONLY_MODE_READ_WRITE:
2706         bdrv_flags |= BDRV_O_RDWR;
2707         break;
2708 
2709     default:
2710         abort();
2711     }
2712 
2713     options = qdict_new();
2714     detect_zeroes = blk_get_detect_zeroes_from_root_state(blk);
2715     qdict_put_str(options, "detect-zeroes", detect_zeroes ? "on" : "off");
2716 
2717     if (has_format) {
2718         qdict_put_str(options, "driver", format);
2719     }
2720 
2721     medium_bs = bdrv_open(filename, NULL, options, bdrv_flags, errp);
2722     if (!medium_bs) {
2723         goto fail;
2724     }
2725 
2726     rc = do_open_tray(has_device ? device : NULL,
2727                       has_id ? id : NULL,
2728                       false, &err);
2729     if (rc && rc != -ENOSYS) {
2730         error_propagate(errp, err);
2731         goto fail;
2732     }
2733     error_free(err);
2734     err = NULL;
2735 
2736     blockdev_remove_medium(has_device, device, has_id, id, &err);
2737     if (err) {
2738         error_propagate(errp, err);
2739         goto fail;
2740     }
2741 
2742     qmp_blockdev_insert_anon_medium(blk, medium_bs, &err);
2743     if (err) {
2744         error_propagate(errp, err);
2745         goto fail;
2746     }
2747 
2748     qmp_blockdev_close_tray(has_device, device, has_id, id, errp);
2749 
2750 fail:
2751     /* If the medium has been inserted, the device has its own reference, so
2752      * ours must be relinquished; and if it has not been inserted successfully,
2753      * the reference must be relinquished anyway */
2754     bdrv_unref(medium_bs);
2755 }
2756 
2757 /* throttling disk I/O limits */
2758 void qmp_block_set_io_throttle(BlockIOThrottle *arg, Error **errp)
2759 {
2760     ThrottleConfig cfg;
2761     BlockDriverState *bs;
2762     BlockBackend *blk;
2763     AioContext *aio_context;
2764 
2765     blk = qmp_get_blk(arg->has_device ? arg->device : NULL,
2766                       arg->has_id ? arg->id : NULL,
2767                       errp);
2768     if (!blk) {
2769         return;
2770     }
2771 
2772     aio_context = blk_get_aio_context(blk);
2773     aio_context_acquire(aio_context);
2774 
2775     bs = blk_bs(blk);
2776     if (!bs) {
2777         error_setg(errp, "Device has no medium");
2778         goto out;
2779     }
2780 
2781     throttle_config_init(&cfg);
2782     cfg.buckets[THROTTLE_BPS_TOTAL].avg = arg->bps;
2783     cfg.buckets[THROTTLE_BPS_READ].avg  = arg->bps_rd;
2784     cfg.buckets[THROTTLE_BPS_WRITE].avg = arg->bps_wr;
2785 
2786     cfg.buckets[THROTTLE_OPS_TOTAL].avg = arg->iops;
2787     cfg.buckets[THROTTLE_OPS_READ].avg  = arg->iops_rd;
2788     cfg.buckets[THROTTLE_OPS_WRITE].avg = arg->iops_wr;
2789 
2790     if (arg->has_bps_max) {
2791         cfg.buckets[THROTTLE_BPS_TOTAL].max = arg->bps_max;
2792     }
2793     if (arg->has_bps_rd_max) {
2794         cfg.buckets[THROTTLE_BPS_READ].max = arg->bps_rd_max;
2795     }
2796     if (arg->has_bps_wr_max) {
2797         cfg.buckets[THROTTLE_BPS_WRITE].max = arg->bps_wr_max;
2798     }
2799     if (arg->has_iops_max) {
2800         cfg.buckets[THROTTLE_OPS_TOTAL].max = arg->iops_max;
2801     }
2802     if (arg->has_iops_rd_max) {
2803         cfg.buckets[THROTTLE_OPS_READ].max = arg->iops_rd_max;
2804     }
2805     if (arg->has_iops_wr_max) {
2806         cfg.buckets[THROTTLE_OPS_WRITE].max = arg->iops_wr_max;
2807     }
2808 
2809     if (arg->has_bps_max_length) {
2810         cfg.buckets[THROTTLE_BPS_TOTAL].burst_length = arg->bps_max_length;
2811     }
2812     if (arg->has_bps_rd_max_length) {
2813         cfg.buckets[THROTTLE_BPS_READ].burst_length = arg->bps_rd_max_length;
2814     }
2815     if (arg->has_bps_wr_max_length) {
2816         cfg.buckets[THROTTLE_BPS_WRITE].burst_length = arg->bps_wr_max_length;
2817     }
2818     if (arg->has_iops_max_length) {
2819         cfg.buckets[THROTTLE_OPS_TOTAL].burst_length = arg->iops_max_length;
2820     }
2821     if (arg->has_iops_rd_max_length) {
2822         cfg.buckets[THROTTLE_OPS_READ].burst_length = arg->iops_rd_max_length;
2823     }
2824     if (arg->has_iops_wr_max_length) {
2825         cfg.buckets[THROTTLE_OPS_WRITE].burst_length = arg->iops_wr_max_length;
2826     }
2827 
2828     if (arg->has_iops_size) {
2829         cfg.op_size = arg->iops_size;
2830     }
2831 
2832     if (!throttle_is_valid(&cfg, errp)) {
2833         goto out;
2834     }
2835 
2836     if (throttle_enabled(&cfg)) {
2837         /* Enable I/O limits if they're not enabled yet, otherwise
2838          * just update the throttling group. */
2839         if (!blk_get_public(blk)->throttle_group_member.throttle_state) {
2840             blk_io_limits_enable(blk,
2841                                  arg->has_group ? arg->group :
2842                                  arg->has_device ? arg->device :
2843                                  arg->id);
2844         } else if (arg->has_group) {
2845             blk_io_limits_update_group(blk, arg->group);
2846         }
2847         /* Set the new throttling configuration */
2848         blk_set_io_limits(blk, &cfg);
2849     } else if (blk_get_public(blk)->throttle_group_member.throttle_state) {
2850         /* If all throttling settings are set to 0, disable I/O limits */
2851         blk_io_limits_disable(blk);
2852     }
2853 
2854 out:
2855     aio_context_release(aio_context);
2856 }
2857 
2858 void qmp_block_dirty_bitmap_add(const char *node, const char *name,
2859                                 bool has_granularity, uint32_t granularity,
2860                                 bool has_persistent, bool persistent,
2861                                 bool has_autoload, bool autoload,
2862                                 bool has_disabled, bool disabled,
2863                                 Error **errp)
2864 {
2865     BlockDriverState *bs;
2866     BdrvDirtyBitmap *bitmap;
2867 
2868     if (!name || name[0] == '\0') {
2869         error_setg(errp, "Bitmap name cannot be empty");
2870         return;
2871     }
2872 
2873     bs = bdrv_lookup_bs(node, node, errp);
2874     if (!bs) {
2875         return;
2876     }
2877 
2878     if (has_granularity) {
2879         if (granularity < 512 || !is_power_of_2(granularity)) {
2880             error_setg(errp, "Granularity must be power of 2 "
2881                              "and at least 512");
2882             return;
2883         }
2884     } else {
2885         /* Default to cluster size, if available: */
2886         granularity = bdrv_get_default_bitmap_granularity(bs);
2887     }
2888 
2889     if (!has_persistent) {
2890         persistent = false;
2891     }
2892 
2893     if (has_autoload) {
2894         warn_report("Autoload option is deprecated and its value is ignored");
2895     }
2896 
2897     if (!has_disabled) {
2898         disabled = false;
2899     }
2900 
2901     if (persistent) {
2902         AioContext *aio_context = bdrv_get_aio_context(bs);
2903         bool ok;
2904 
2905         aio_context_acquire(aio_context);
2906         ok = bdrv_can_store_new_dirty_bitmap(bs, name, granularity, errp);
2907         aio_context_release(aio_context);
2908         if (!ok) {
2909             return;
2910         }
2911     }
2912 
2913     bitmap = bdrv_create_dirty_bitmap(bs, granularity, name, errp);
2914     if (bitmap == NULL) {
2915         return;
2916     }
2917 
2918     if (disabled) {
2919         bdrv_disable_dirty_bitmap(bitmap);
2920     }
2921 
2922     bdrv_dirty_bitmap_set_persistence(bitmap, persistent);
2923 }
2924 
2925 static BdrvDirtyBitmap *do_block_dirty_bitmap_remove(
2926         const char *node, const char *name, bool release,
2927         BlockDriverState **bitmap_bs, Error **errp)
2928 {
2929     BlockDriverState *bs;
2930     BdrvDirtyBitmap *bitmap;
2931 
2932     bitmap = block_dirty_bitmap_lookup(node, name, &bs, errp);
2933     if (!bitmap || !bs) {
2934         return NULL;
2935     }
2936 
2937     if (bdrv_dirty_bitmap_check(bitmap, BDRV_BITMAP_BUSY | BDRV_BITMAP_RO,
2938                                 errp)) {
2939         return NULL;
2940     }
2941 
2942     if (bdrv_dirty_bitmap_get_persistence(bitmap)) {
2943         AioContext *aio_context = bdrv_get_aio_context(bs);
2944         Error *local_err = NULL;
2945 
2946         aio_context_acquire(aio_context);
2947         bdrv_remove_persistent_dirty_bitmap(bs, name, &local_err);
2948         aio_context_release(aio_context);
2949 
2950         if (local_err != NULL) {
2951             error_propagate(errp, local_err);
2952             return NULL;
2953         }
2954     }
2955 
2956     if (release) {
2957         bdrv_release_dirty_bitmap(bs, bitmap);
2958     }
2959 
2960     if (bitmap_bs) {
2961         *bitmap_bs = bs;
2962     }
2963 
2964     return release ? NULL : bitmap;
2965 }
2966 
2967 void qmp_block_dirty_bitmap_remove(const char *node, const char *name,
2968                                    Error **errp)
2969 {
2970     do_block_dirty_bitmap_remove(node, name, true, NULL, errp);
2971 }
2972 
2973 /**
2974  * Completely clear a bitmap, for the purposes of synchronizing a bitmap
2975  * immediately after a full backup operation.
2976  */
2977 void qmp_block_dirty_bitmap_clear(const char *node, const char *name,
2978                                   Error **errp)
2979 {
2980     BdrvDirtyBitmap *bitmap;
2981     BlockDriverState *bs;
2982 
2983     bitmap = block_dirty_bitmap_lookup(node, name, &bs, errp);
2984     if (!bitmap || !bs) {
2985         return;
2986     }
2987 
2988     if (bdrv_dirty_bitmap_check(bitmap, BDRV_BITMAP_DEFAULT, errp)) {
2989         return;
2990     }
2991 
2992     bdrv_clear_dirty_bitmap(bitmap, NULL);
2993 }
2994 
2995 void qmp_block_dirty_bitmap_enable(const char *node, const char *name,
2996                                    Error **errp)
2997 {
2998     BlockDriverState *bs;
2999     BdrvDirtyBitmap *bitmap;
3000 
3001     bitmap = block_dirty_bitmap_lookup(node, name, &bs, errp);
3002     if (!bitmap) {
3003         return;
3004     }
3005 
3006     if (bdrv_dirty_bitmap_check(bitmap, BDRV_BITMAP_ALLOW_RO, errp)) {
3007         return;
3008     }
3009 
3010     bdrv_enable_dirty_bitmap(bitmap);
3011 }
3012 
3013 void qmp_block_dirty_bitmap_disable(const char *node, const char *name,
3014                                     Error **errp)
3015 {
3016     BlockDriverState *bs;
3017     BdrvDirtyBitmap *bitmap;
3018 
3019     bitmap = block_dirty_bitmap_lookup(node, name, &bs, errp);
3020     if (!bitmap) {
3021         return;
3022     }
3023 
3024     if (bdrv_dirty_bitmap_check(bitmap, BDRV_BITMAP_ALLOW_RO, errp)) {
3025         return;
3026     }
3027 
3028     bdrv_disable_dirty_bitmap(bitmap);
3029 }
3030 
3031 static BdrvDirtyBitmap *do_block_dirty_bitmap_merge(
3032         const char *node, const char *target,
3033         BlockDirtyBitmapMergeSourceList *bitmaps,
3034         HBitmap **backup, Error **errp)
3035 {
3036     BlockDriverState *bs;
3037     BdrvDirtyBitmap *dst, *src, *anon;
3038     BlockDirtyBitmapMergeSourceList *lst;
3039     Error *local_err = NULL;
3040 
3041     dst = block_dirty_bitmap_lookup(node, target, &bs, errp);
3042     if (!dst) {
3043         return NULL;
3044     }
3045 
3046     anon = bdrv_create_dirty_bitmap(bs, bdrv_dirty_bitmap_granularity(dst),
3047                                     NULL, errp);
3048     if (!anon) {
3049         return NULL;
3050     }
3051 
3052     for (lst = bitmaps; lst; lst = lst->next) {
3053         switch (lst->value->type) {
3054             const char *name, *node;
3055         case QTYPE_QSTRING:
3056             name = lst->value->u.local;
3057             src = bdrv_find_dirty_bitmap(bs, name);
3058             if (!src) {
3059                 error_setg(errp, "Dirty bitmap '%s' not found", name);
3060                 dst = NULL;
3061                 goto out;
3062             }
3063             break;
3064         case QTYPE_QDICT:
3065             node = lst->value->u.external.node;
3066             name = lst->value->u.external.name;
3067             src = block_dirty_bitmap_lookup(node, name, NULL, errp);
3068             if (!src) {
3069                 dst = NULL;
3070                 goto out;
3071             }
3072             break;
3073         default:
3074             abort();
3075         }
3076 
3077         bdrv_merge_dirty_bitmap(anon, src, NULL, &local_err);
3078         if (local_err) {
3079             error_propagate(errp, local_err);
3080             dst = NULL;
3081             goto out;
3082         }
3083     }
3084 
3085     /* Merge into dst; dst is unchanged on failure. */
3086     bdrv_merge_dirty_bitmap(dst, anon, backup, errp);
3087 
3088  out:
3089     bdrv_release_dirty_bitmap(bs, anon);
3090     return dst;
3091 }
3092 
3093 void qmp_block_dirty_bitmap_merge(const char *node, const char *target,
3094                                   BlockDirtyBitmapMergeSourceList *bitmaps,
3095                                   Error **errp)
3096 {
3097     do_block_dirty_bitmap_merge(node, target, bitmaps, NULL, errp);
3098 }
3099 
3100 BlockDirtyBitmapSha256 *qmp_x_debug_block_dirty_bitmap_sha256(const char *node,
3101                                                               const char *name,
3102                                                               Error **errp)
3103 {
3104     BdrvDirtyBitmap *bitmap;
3105     BlockDriverState *bs;
3106     BlockDirtyBitmapSha256 *ret = NULL;
3107     char *sha256;
3108 
3109     bitmap = block_dirty_bitmap_lookup(node, name, &bs, errp);
3110     if (!bitmap || !bs) {
3111         return NULL;
3112     }
3113 
3114     sha256 = bdrv_dirty_bitmap_sha256(bitmap, errp);
3115     if (sha256 == NULL) {
3116         return NULL;
3117     }
3118 
3119     ret = g_new(BlockDirtyBitmapSha256, 1);
3120     ret->sha256 = sha256;
3121 
3122     return ret;
3123 }
3124 
3125 void hmp_drive_del(Monitor *mon, const QDict *qdict)
3126 {
3127     const char *id = qdict_get_str(qdict, "id");
3128     BlockBackend *blk;
3129     BlockDriverState *bs;
3130     AioContext *aio_context;
3131     Error *local_err = NULL;
3132 
3133     bs = bdrv_find_node(id);
3134     if (bs) {
3135         qmp_blockdev_del(id, &local_err);
3136         if (local_err) {
3137             error_report_err(local_err);
3138         }
3139         return;
3140     }
3141 
3142     blk = blk_by_name(id);
3143     if (!blk) {
3144         error_report("Device '%s' not found", id);
3145         return;
3146     }
3147 
3148     if (!blk_legacy_dinfo(blk)) {
3149         error_report("Deleting device added with blockdev-add"
3150                      " is not supported");
3151         return;
3152     }
3153 
3154     aio_context = blk_get_aio_context(blk);
3155     aio_context_acquire(aio_context);
3156 
3157     bs = blk_bs(blk);
3158     if (bs) {
3159         if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_DRIVE_DEL, &local_err)) {
3160             error_report_err(local_err);
3161             aio_context_release(aio_context);
3162             return;
3163         }
3164 
3165         blk_remove_bs(blk);
3166     }
3167 
3168     /* Make the BlockBackend and the attached BlockDriverState anonymous */
3169     monitor_remove_blk(blk);
3170 
3171     /* If this BlockBackend has a device attached to it, its refcount will be
3172      * decremented when the device is removed; otherwise we have to do so here.
3173      */
3174     if (blk_get_attached_dev(blk)) {
3175         /* Further I/O must not pause the guest */
3176         blk_set_on_error(blk, BLOCKDEV_ON_ERROR_REPORT,
3177                          BLOCKDEV_ON_ERROR_REPORT);
3178     } else {
3179         blk_unref(blk);
3180     }
3181 
3182     aio_context_release(aio_context);
3183 }
3184 
3185 void qmp_block_resize(bool has_device, const char *device,
3186                       bool has_node_name, const char *node_name,
3187                       int64_t size, Error **errp)
3188 {
3189     Error *local_err = NULL;
3190     BlockBackend *blk = NULL;
3191     BlockDriverState *bs;
3192     AioContext *aio_context;
3193     int ret;
3194 
3195     bs = bdrv_lookup_bs(has_device ? device : NULL,
3196                         has_node_name ? node_name : NULL,
3197                         &local_err);
3198     if (local_err) {
3199         error_propagate(errp, local_err);
3200         return;
3201     }
3202 
3203     aio_context = bdrv_get_aio_context(bs);
3204     aio_context_acquire(aio_context);
3205 
3206     if (!bdrv_is_first_non_filter(bs)) {
3207         error_setg(errp, QERR_FEATURE_DISABLED, "resize");
3208         goto out;
3209     }
3210 
3211     if (size < 0) {
3212         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "size", "a >0 size");
3213         goto out;
3214     }
3215 
3216     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_RESIZE, NULL)) {
3217         error_setg(errp, QERR_DEVICE_IN_USE, device);
3218         goto out;
3219     }
3220 
3221     blk = blk_new(bdrv_get_aio_context(bs), BLK_PERM_RESIZE, BLK_PERM_ALL);
3222     ret = blk_insert_bs(blk, bs, errp);
3223     if (ret < 0) {
3224         goto out;
3225     }
3226 
3227     bdrv_drained_begin(bs);
3228     ret = blk_truncate(blk, size, PREALLOC_MODE_OFF, errp);
3229     bdrv_drained_end(bs);
3230 
3231 out:
3232     blk_unref(blk);
3233     aio_context_release(aio_context);
3234 }
3235 
3236 void qmp_block_stream(bool has_job_id, const char *job_id, const char *device,
3237                       bool has_base, const char *base,
3238                       bool has_base_node, const char *base_node,
3239                       bool has_backing_file, const char *backing_file,
3240                       bool has_speed, int64_t speed,
3241                       bool has_on_error, BlockdevOnError on_error,
3242                       bool has_auto_finalize, bool auto_finalize,
3243                       bool has_auto_dismiss, bool auto_dismiss,
3244                       Error **errp)
3245 {
3246     BlockDriverState *bs, *iter;
3247     BlockDriverState *base_bs = NULL;
3248     AioContext *aio_context;
3249     Error *local_err = NULL;
3250     const char *base_name = NULL;
3251     int job_flags = JOB_DEFAULT;
3252 
3253     if (!has_on_error) {
3254         on_error = BLOCKDEV_ON_ERROR_REPORT;
3255     }
3256 
3257     bs = bdrv_lookup_bs(device, device, errp);
3258     if (!bs) {
3259         return;
3260     }
3261 
3262     aio_context = bdrv_get_aio_context(bs);
3263     aio_context_acquire(aio_context);
3264 
3265     if (has_base && has_base_node) {
3266         error_setg(errp, "'base' and 'base-node' cannot be specified "
3267                    "at the same time");
3268         goto out;
3269     }
3270 
3271     if (has_base) {
3272         base_bs = bdrv_find_backing_image(bs, base);
3273         if (base_bs == NULL) {
3274             error_setg(errp, QERR_BASE_NOT_FOUND, base);
3275             goto out;
3276         }
3277         assert(bdrv_get_aio_context(base_bs) == aio_context);
3278         base_name = base;
3279     }
3280 
3281     if (has_base_node) {
3282         base_bs = bdrv_lookup_bs(NULL, base_node, errp);
3283         if (!base_bs) {
3284             goto out;
3285         }
3286         if (bs == base_bs || !bdrv_chain_contains(bs, base_bs)) {
3287             error_setg(errp, "Node '%s' is not a backing image of '%s'",
3288                        base_node, device);
3289             goto out;
3290         }
3291         assert(bdrv_get_aio_context(base_bs) == aio_context);
3292         bdrv_refresh_filename(base_bs);
3293         base_name = base_bs->filename;
3294     }
3295 
3296     /* Check for op blockers in the whole chain between bs and base */
3297     for (iter = bs; iter && iter != base_bs; iter = backing_bs(iter)) {
3298         if (bdrv_op_is_blocked(iter, BLOCK_OP_TYPE_STREAM, errp)) {
3299             goto out;
3300         }
3301     }
3302 
3303     /* if we are streaming the entire chain, the result will have no backing
3304      * file, and specifying one is therefore an error */
3305     if (base_bs == NULL && has_backing_file) {
3306         error_setg(errp, "backing file specified, but streaming the "
3307                          "entire chain");
3308         goto out;
3309     }
3310 
3311     /* backing_file string overrides base bs filename */
3312     base_name = has_backing_file ? backing_file : base_name;
3313 
3314     if (has_auto_finalize && !auto_finalize) {
3315         job_flags |= JOB_MANUAL_FINALIZE;
3316     }
3317     if (has_auto_dismiss && !auto_dismiss) {
3318         job_flags |= JOB_MANUAL_DISMISS;
3319     }
3320 
3321     stream_start(has_job_id ? job_id : NULL, bs, base_bs, base_name,
3322                  job_flags, has_speed ? speed : 0, on_error, &local_err);
3323     if (local_err) {
3324         error_propagate(errp, local_err);
3325         goto out;
3326     }
3327 
3328     trace_qmp_block_stream(bs);
3329 
3330 out:
3331     aio_context_release(aio_context);
3332 }
3333 
3334 void qmp_block_commit(bool has_job_id, const char *job_id, const char *device,
3335                       bool has_base_node, const char *base_node,
3336                       bool has_base, const char *base,
3337                       bool has_top_node, const char *top_node,
3338                       bool has_top, const char *top,
3339                       bool has_backing_file, const char *backing_file,
3340                       bool has_speed, int64_t speed,
3341                       bool has_filter_node_name, const char *filter_node_name,
3342                       bool has_auto_finalize, bool auto_finalize,
3343                       bool has_auto_dismiss, bool auto_dismiss,
3344                       Error **errp)
3345 {
3346     BlockDriverState *bs;
3347     BlockDriverState *iter;
3348     BlockDriverState *base_bs, *top_bs;
3349     AioContext *aio_context;
3350     Error *local_err = NULL;
3351     /* This will be part of the QMP command, if/when the
3352      * BlockdevOnError change for blkmirror makes it in
3353      */
3354     BlockdevOnError on_error = BLOCKDEV_ON_ERROR_REPORT;
3355     int job_flags = JOB_DEFAULT;
3356 
3357     if (!has_speed) {
3358         speed = 0;
3359     }
3360     if (!has_filter_node_name) {
3361         filter_node_name = NULL;
3362     }
3363     if (has_auto_finalize && !auto_finalize) {
3364         job_flags |= JOB_MANUAL_FINALIZE;
3365     }
3366     if (has_auto_dismiss && !auto_dismiss) {
3367         job_flags |= JOB_MANUAL_DISMISS;
3368     }
3369 
3370     /* Important Note:
3371      *  libvirt relies on the DeviceNotFound error class in order to probe for
3372      *  live commit feature versions; for this to work, we must make sure to
3373      *  perform the device lookup before any generic errors that may occur in a
3374      *  scenario in which all optional arguments are omitted. */
3375     bs = qmp_get_root_bs(device, &local_err);
3376     if (!bs) {
3377         bs = bdrv_lookup_bs(device, device, NULL);
3378         if (!bs) {
3379             error_free(local_err);
3380             error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
3381                       "Device '%s' not found", device);
3382         } else {
3383             error_propagate(errp, local_err);
3384         }
3385         return;
3386     }
3387 
3388     aio_context = bdrv_get_aio_context(bs);
3389     aio_context_acquire(aio_context);
3390 
3391     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_COMMIT_SOURCE, errp)) {
3392         goto out;
3393     }
3394 
3395     /* default top_bs is the active layer */
3396     top_bs = bs;
3397 
3398     if (has_top_node && has_top) {
3399         error_setg(errp, "'top-node' and 'top' are mutually exclusive");
3400         goto out;
3401     } else if (has_top_node) {
3402         top_bs = bdrv_lookup_bs(NULL, top_node, errp);
3403         if (top_bs == NULL) {
3404             goto out;
3405         }
3406         if (!bdrv_chain_contains(bs, top_bs)) {
3407             error_setg(errp, "'%s' is not in this backing file chain",
3408                        top_node);
3409             goto out;
3410         }
3411     } else if (has_top && top) {
3412         /* This strcmp() is just a shortcut, there is no need to
3413          * refresh @bs's filename.  If it mismatches,
3414          * bdrv_find_backing_image() will do the refresh and may still
3415          * return @bs. */
3416         if (strcmp(bs->filename, top) != 0) {
3417             top_bs = bdrv_find_backing_image(bs, top);
3418         }
3419     }
3420 
3421     if (top_bs == NULL) {
3422         error_setg(errp, "Top image file %s not found", top ? top : "NULL");
3423         goto out;
3424     }
3425 
3426     assert(bdrv_get_aio_context(top_bs) == aio_context);
3427 
3428     if (has_base_node && has_base) {
3429         error_setg(errp, "'base-node' and 'base' are mutually exclusive");
3430         goto out;
3431     } else if (has_base_node) {
3432         base_bs = bdrv_lookup_bs(NULL, base_node, errp);
3433         if (base_bs == NULL) {
3434             goto out;
3435         }
3436         if (!bdrv_chain_contains(top_bs, base_bs)) {
3437             error_setg(errp, "'%s' is not in this backing file chain",
3438                        base_node);
3439             goto out;
3440         }
3441     } else if (has_base && base) {
3442         base_bs = bdrv_find_backing_image(top_bs, base);
3443     } else {
3444         base_bs = bdrv_find_base(top_bs);
3445     }
3446 
3447     if (base_bs == NULL) {
3448         error_setg(errp, QERR_BASE_NOT_FOUND, base ? base : "NULL");
3449         goto out;
3450     }
3451 
3452     assert(bdrv_get_aio_context(base_bs) == aio_context);
3453 
3454     for (iter = top_bs; iter != backing_bs(base_bs); iter = backing_bs(iter)) {
3455         if (bdrv_op_is_blocked(iter, BLOCK_OP_TYPE_COMMIT_TARGET, errp)) {
3456             goto out;
3457         }
3458     }
3459 
3460     /* Do not allow attempts to commit an image into itself */
3461     if (top_bs == base_bs) {
3462         error_setg(errp, "cannot commit an image into itself");
3463         goto out;
3464     }
3465 
3466     if (top_bs == bs) {
3467         if (has_backing_file) {
3468             error_setg(errp, "'backing-file' specified,"
3469                              " but 'top' is the active layer");
3470             goto out;
3471         }
3472         commit_active_start(has_job_id ? job_id : NULL, bs, base_bs,
3473                             job_flags, speed, on_error,
3474                             filter_node_name, NULL, NULL, false, &local_err);
3475     } else {
3476         BlockDriverState *overlay_bs = bdrv_find_overlay(bs, top_bs);
3477         if (bdrv_op_is_blocked(overlay_bs, BLOCK_OP_TYPE_COMMIT_TARGET, errp)) {
3478             goto out;
3479         }
3480         commit_start(has_job_id ? job_id : NULL, bs, base_bs, top_bs, job_flags,
3481                      speed, on_error, has_backing_file ? backing_file : NULL,
3482                      filter_node_name, &local_err);
3483     }
3484     if (local_err != NULL) {
3485         error_propagate(errp, local_err);
3486         goto out;
3487     }
3488 
3489 out:
3490     aio_context_release(aio_context);
3491 }
3492 
3493 /* Common QMP interface for drive-backup and blockdev-backup */
3494 static BlockJob *do_backup_common(BackupCommon *backup,
3495                                   BlockDriverState *bs,
3496                                   BlockDriverState *target_bs,
3497                                   AioContext *aio_context,
3498                                   JobTxn *txn, Error **errp)
3499 {
3500     BlockJob *job = NULL;
3501     BdrvDirtyBitmap *bmap = NULL;
3502     int job_flags = JOB_DEFAULT;
3503     int ret;
3504 
3505     if (!backup->has_speed) {
3506         backup->speed = 0;
3507     }
3508     if (!backup->has_on_source_error) {
3509         backup->on_source_error = BLOCKDEV_ON_ERROR_REPORT;
3510     }
3511     if (!backup->has_on_target_error) {
3512         backup->on_target_error = BLOCKDEV_ON_ERROR_REPORT;
3513     }
3514     if (!backup->has_job_id) {
3515         backup->job_id = NULL;
3516     }
3517     if (!backup->has_auto_finalize) {
3518         backup->auto_finalize = true;
3519     }
3520     if (!backup->has_auto_dismiss) {
3521         backup->auto_dismiss = true;
3522     }
3523     if (!backup->has_compress) {
3524         backup->compress = false;
3525     }
3526 
3527     ret = bdrv_try_set_aio_context(target_bs, aio_context, errp);
3528     if (ret < 0) {
3529         return NULL;
3530     }
3531 
3532     if ((backup->sync == MIRROR_SYNC_MODE_BITMAP) ||
3533         (backup->sync == MIRROR_SYNC_MODE_INCREMENTAL)) {
3534         /* done before desugaring 'incremental' to print the right message */
3535         if (!backup->has_bitmap) {
3536             error_setg(errp, "must provide a valid bitmap name for "
3537                        "'%s' sync mode", MirrorSyncMode_str(backup->sync));
3538             return NULL;
3539         }
3540     }
3541 
3542     if (backup->sync == MIRROR_SYNC_MODE_INCREMENTAL) {
3543         if (backup->has_bitmap_mode &&
3544             backup->bitmap_mode != BITMAP_SYNC_MODE_ON_SUCCESS) {
3545             error_setg(errp, "Bitmap sync mode must be '%s' "
3546                        "when using sync mode '%s'",
3547                        BitmapSyncMode_str(BITMAP_SYNC_MODE_ON_SUCCESS),
3548                        MirrorSyncMode_str(backup->sync));
3549             return NULL;
3550         }
3551         backup->has_bitmap_mode = true;
3552         backup->sync = MIRROR_SYNC_MODE_BITMAP;
3553         backup->bitmap_mode = BITMAP_SYNC_MODE_ON_SUCCESS;
3554     }
3555 
3556     if (backup->has_bitmap) {
3557         bmap = bdrv_find_dirty_bitmap(bs, backup->bitmap);
3558         if (!bmap) {
3559             error_setg(errp, "Bitmap '%s' could not be found", backup->bitmap);
3560             return NULL;
3561         }
3562         if (!backup->has_bitmap_mode) {
3563             error_setg(errp, "Bitmap sync mode must be given "
3564                        "when providing a bitmap");
3565             return NULL;
3566         }
3567         if (bdrv_dirty_bitmap_check(bmap, BDRV_BITMAP_ALLOW_RO, errp)) {
3568             return NULL;
3569         }
3570 
3571         /* This does not produce a useful bitmap artifact: */
3572         if (backup->sync == MIRROR_SYNC_MODE_NONE) {
3573             error_setg(errp, "sync mode '%s' does not produce meaningful bitmap"
3574                        " outputs", MirrorSyncMode_str(backup->sync));
3575             return NULL;
3576         }
3577 
3578         /* If the bitmap isn't used for input or output, this is useless: */
3579         if (backup->bitmap_mode == BITMAP_SYNC_MODE_NEVER &&
3580             backup->sync != MIRROR_SYNC_MODE_BITMAP) {
3581             error_setg(errp, "Bitmap sync mode '%s' has no meaningful effect"
3582                        " when combined with sync mode '%s'",
3583                        BitmapSyncMode_str(backup->bitmap_mode),
3584                        MirrorSyncMode_str(backup->sync));
3585             return NULL;
3586         }
3587     }
3588 
3589     if (!backup->has_bitmap && backup->has_bitmap_mode) {
3590         error_setg(errp, "Cannot specify bitmap sync mode without a bitmap");
3591         return NULL;
3592     }
3593 
3594     if (!backup->auto_finalize) {
3595         job_flags |= JOB_MANUAL_FINALIZE;
3596     }
3597     if (!backup->auto_dismiss) {
3598         job_flags |= JOB_MANUAL_DISMISS;
3599     }
3600 
3601     job = backup_job_create(backup->job_id, bs, target_bs, backup->speed,
3602                             backup->sync, bmap, backup->bitmap_mode,
3603                             backup->compress,
3604                             backup->on_source_error,
3605                             backup->on_target_error,
3606                             job_flags, NULL, NULL, txn, errp);
3607     return job;
3608 }
3609 
3610 static BlockJob *do_drive_backup(DriveBackup *backup, JobTxn *txn,
3611                                  Error **errp)
3612 {
3613     BlockDriverState *bs;
3614     BlockDriverState *target_bs;
3615     BlockDriverState *source = NULL;
3616     BlockJob *job = NULL;
3617     AioContext *aio_context;
3618     QDict *options;
3619     Error *local_err = NULL;
3620     int flags;
3621     int64_t size;
3622     bool set_backing_hd = false;
3623 
3624     if (!backup->has_mode) {
3625         backup->mode = NEW_IMAGE_MODE_ABSOLUTE_PATHS;
3626     }
3627 
3628     bs = bdrv_lookup_bs(backup->device, backup->device, errp);
3629     if (!bs) {
3630         return NULL;
3631     }
3632 
3633     if (!bs->drv) {
3634         error_setg(errp, "Device has no medium");
3635         return NULL;
3636     }
3637 
3638     aio_context = bdrv_get_aio_context(bs);
3639     aio_context_acquire(aio_context);
3640 
3641     if (!backup->has_format) {
3642         backup->format = backup->mode == NEW_IMAGE_MODE_EXISTING ?
3643                          NULL : (char*) bs->drv->format_name;
3644     }
3645 
3646     /* Early check to avoid creating target */
3647     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_BACKUP_SOURCE, errp)) {
3648         goto out;
3649     }
3650 
3651     flags = bs->open_flags | BDRV_O_RDWR;
3652 
3653     /* See if we have a backing HD we can use to create our new image
3654      * on top of. */
3655     if (backup->sync == MIRROR_SYNC_MODE_TOP) {
3656         source = backing_bs(bs);
3657         if (!source) {
3658             backup->sync = MIRROR_SYNC_MODE_FULL;
3659         }
3660     }
3661     if (backup->sync == MIRROR_SYNC_MODE_NONE) {
3662         source = bs;
3663         flags |= BDRV_O_NO_BACKING;
3664         set_backing_hd = true;
3665     }
3666 
3667     size = bdrv_getlength(bs);
3668     if (size < 0) {
3669         error_setg_errno(errp, -size, "bdrv_getlength failed");
3670         goto out;
3671     }
3672 
3673     if (backup->mode != NEW_IMAGE_MODE_EXISTING) {
3674         assert(backup->format);
3675         if (source) {
3676             bdrv_refresh_filename(source);
3677             bdrv_img_create(backup->target, backup->format, source->filename,
3678                             source->drv->format_name, NULL,
3679                             size, flags, false, &local_err);
3680         } else {
3681             bdrv_img_create(backup->target, backup->format, NULL, NULL, NULL,
3682                             size, flags, false, &local_err);
3683         }
3684     }
3685 
3686     if (local_err) {
3687         error_propagate(errp, local_err);
3688         goto out;
3689     }
3690 
3691     options = qdict_new();
3692     qdict_put_str(options, "discard", "unmap");
3693     qdict_put_str(options, "detect-zeroes", "unmap");
3694     if (backup->format) {
3695         qdict_put_str(options, "driver", backup->format);
3696     }
3697 
3698     target_bs = bdrv_open(backup->target, NULL, options, flags, errp);
3699     if (!target_bs) {
3700         goto out;
3701     }
3702 
3703     if (set_backing_hd) {
3704         bdrv_set_backing_hd(target_bs, source, &local_err);
3705         if (local_err) {
3706             goto unref;
3707         }
3708     }
3709 
3710     job = do_backup_common(qapi_DriveBackup_base(backup),
3711                            bs, target_bs, aio_context, txn, errp);
3712 
3713 unref:
3714     bdrv_unref(target_bs);
3715 out:
3716     aio_context_release(aio_context);
3717     return job;
3718 }
3719 
3720 void qmp_drive_backup(DriveBackup *arg, Error **errp)
3721 {
3722 
3723     BlockJob *job;
3724     job = do_drive_backup(arg, NULL, errp);
3725     if (job) {
3726         job_start(&job->job);
3727     }
3728 }
3729 
3730 BlockDeviceInfoList *qmp_query_named_block_nodes(Error **errp)
3731 {
3732     return bdrv_named_nodes_list(errp);
3733 }
3734 
3735 XDbgBlockGraph *qmp_x_debug_query_block_graph(Error **errp)
3736 {
3737     return bdrv_get_xdbg_block_graph(errp);
3738 }
3739 
3740 BlockJob *do_blockdev_backup(BlockdevBackup *backup, JobTxn *txn,
3741                              Error **errp)
3742 {
3743     BlockDriverState *bs;
3744     BlockDriverState *target_bs;
3745     AioContext *aio_context;
3746     BlockJob *job;
3747 
3748     bs = bdrv_lookup_bs(backup->device, backup->device, errp);
3749     if (!bs) {
3750         return NULL;
3751     }
3752 
3753     target_bs = bdrv_lookup_bs(backup->target, backup->target, errp);
3754     if (!target_bs) {
3755         return NULL;
3756     }
3757 
3758     aio_context = bdrv_get_aio_context(bs);
3759     aio_context_acquire(aio_context);
3760 
3761     job = do_backup_common(qapi_BlockdevBackup_base(backup),
3762                            bs, target_bs, aio_context, txn, errp);
3763 
3764     aio_context_release(aio_context);
3765     return job;
3766 }
3767 
3768 void qmp_blockdev_backup(BlockdevBackup *arg, Error **errp)
3769 {
3770     BlockJob *job;
3771     job = do_blockdev_backup(arg, NULL, errp);
3772     if (job) {
3773         job_start(&job->job);
3774     }
3775 }
3776 
3777 /* Parameter check and block job starting for drive mirroring.
3778  * Caller should hold @device and @target's aio context (must be the same).
3779  **/
3780 static void blockdev_mirror_common(const char *job_id, BlockDriverState *bs,
3781                                    BlockDriverState *target,
3782                                    bool has_replaces, const char *replaces,
3783                                    enum MirrorSyncMode sync,
3784                                    BlockMirrorBackingMode backing_mode,
3785                                    bool zero_target,
3786                                    bool has_speed, int64_t speed,
3787                                    bool has_granularity, uint32_t granularity,
3788                                    bool has_buf_size, int64_t buf_size,
3789                                    bool has_on_source_error,
3790                                    BlockdevOnError on_source_error,
3791                                    bool has_on_target_error,
3792                                    BlockdevOnError on_target_error,
3793                                    bool has_unmap, bool unmap,
3794                                    bool has_filter_node_name,
3795                                    const char *filter_node_name,
3796                                    bool has_copy_mode, MirrorCopyMode copy_mode,
3797                                    bool has_auto_finalize, bool auto_finalize,
3798                                    bool has_auto_dismiss, bool auto_dismiss,
3799                                    Error **errp)
3800 {
3801     int job_flags = JOB_DEFAULT;
3802 
3803     if (!has_speed) {
3804         speed = 0;
3805     }
3806     if (!has_on_source_error) {
3807         on_source_error = BLOCKDEV_ON_ERROR_REPORT;
3808     }
3809     if (!has_on_target_error) {
3810         on_target_error = BLOCKDEV_ON_ERROR_REPORT;
3811     }
3812     if (!has_granularity) {
3813         granularity = 0;
3814     }
3815     if (!has_buf_size) {
3816         buf_size = 0;
3817     }
3818     if (!has_unmap) {
3819         unmap = true;
3820     }
3821     if (!has_filter_node_name) {
3822         filter_node_name = NULL;
3823     }
3824     if (!has_copy_mode) {
3825         copy_mode = MIRROR_COPY_MODE_BACKGROUND;
3826     }
3827     if (has_auto_finalize && !auto_finalize) {
3828         job_flags |= JOB_MANUAL_FINALIZE;
3829     }
3830     if (has_auto_dismiss && !auto_dismiss) {
3831         job_flags |= JOB_MANUAL_DISMISS;
3832     }
3833 
3834     if (granularity != 0 && (granularity < 512 || granularity > 1048576 * 64)) {
3835         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "granularity",
3836                    "a value in range [512B, 64MB]");
3837         return;
3838     }
3839     if (granularity & (granularity - 1)) {
3840         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "granularity",
3841                    "power of 2");
3842         return;
3843     }
3844 
3845     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_MIRROR_SOURCE, errp)) {
3846         return;
3847     }
3848     if (bdrv_op_is_blocked(target, BLOCK_OP_TYPE_MIRROR_TARGET, errp)) {
3849         return;
3850     }
3851 
3852     if (!bs->backing && sync == MIRROR_SYNC_MODE_TOP) {
3853         sync = MIRROR_SYNC_MODE_FULL;
3854     }
3855 
3856     if (has_replaces) {
3857         BlockDriverState *to_replace_bs;
3858         AioContext *replace_aio_context;
3859         int64_t bs_size, replace_size;
3860 
3861         bs_size = bdrv_getlength(bs);
3862         if (bs_size < 0) {
3863             error_setg_errno(errp, -bs_size, "Failed to query device's size");
3864             return;
3865         }
3866 
3867         to_replace_bs = check_to_replace_node(bs, replaces, errp);
3868         if (!to_replace_bs) {
3869             return;
3870         }
3871 
3872         replace_aio_context = bdrv_get_aio_context(to_replace_bs);
3873         aio_context_acquire(replace_aio_context);
3874         replace_size = bdrv_getlength(to_replace_bs);
3875         aio_context_release(replace_aio_context);
3876 
3877         if (replace_size < 0) {
3878             error_setg_errno(errp, -replace_size,
3879                              "Failed to query the replacement node's size");
3880             return;
3881         }
3882         if (bs_size != replace_size) {
3883             error_setg(errp, "cannot replace image with a mirror image of "
3884                              "different size");
3885             return;
3886         }
3887     }
3888 
3889     /* pass the node name to replace to mirror start since it's loose coupling
3890      * and will allow to check whether the node still exist at mirror completion
3891      */
3892     mirror_start(job_id, bs, target,
3893                  has_replaces ? replaces : NULL, job_flags,
3894                  speed, granularity, buf_size, sync, backing_mode, zero_target,
3895                  on_source_error, on_target_error, unmap, filter_node_name,
3896                  copy_mode, errp);
3897 }
3898 
3899 void qmp_drive_mirror(DriveMirror *arg, Error **errp)
3900 {
3901     BlockDriverState *bs;
3902     BlockDriverState *source, *target_bs;
3903     AioContext *aio_context;
3904     BlockMirrorBackingMode backing_mode;
3905     Error *local_err = NULL;
3906     QDict *options = NULL;
3907     int flags;
3908     int64_t size;
3909     const char *format = arg->format;
3910     bool zero_target;
3911     int ret;
3912 
3913     bs = qmp_get_root_bs(arg->device, errp);
3914     if (!bs) {
3915         return;
3916     }
3917 
3918     /* Early check to avoid creating target */
3919     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_MIRROR_SOURCE, errp)) {
3920         return;
3921     }
3922 
3923     aio_context = bdrv_get_aio_context(bs);
3924     aio_context_acquire(aio_context);
3925 
3926     if (!arg->has_mode) {
3927         arg->mode = NEW_IMAGE_MODE_ABSOLUTE_PATHS;
3928     }
3929 
3930     if (!arg->has_format) {
3931         format = (arg->mode == NEW_IMAGE_MODE_EXISTING
3932                   ? NULL : bs->drv->format_name);
3933     }
3934 
3935     flags = bs->open_flags | BDRV_O_RDWR;
3936     source = backing_bs(bs);
3937     if (!source && arg->sync == MIRROR_SYNC_MODE_TOP) {
3938         arg->sync = MIRROR_SYNC_MODE_FULL;
3939     }
3940     if (arg->sync == MIRROR_SYNC_MODE_NONE) {
3941         source = bs;
3942     }
3943 
3944     size = bdrv_getlength(bs);
3945     if (size < 0) {
3946         error_setg_errno(errp, -size, "bdrv_getlength failed");
3947         goto out;
3948     }
3949 
3950     if (arg->has_replaces) {
3951         if (!arg->has_node_name) {
3952             error_setg(errp, "a node-name must be provided when replacing a"
3953                              " named node of the graph");
3954             goto out;
3955         }
3956     }
3957 
3958     if (arg->mode == NEW_IMAGE_MODE_ABSOLUTE_PATHS) {
3959         backing_mode = MIRROR_SOURCE_BACKING_CHAIN;
3960     } else {
3961         backing_mode = MIRROR_OPEN_BACKING_CHAIN;
3962     }
3963 
3964     /* Don't open backing image in create() */
3965     flags |= BDRV_O_NO_BACKING;
3966 
3967     if ((arg->sync == MIRROR_SYNC_MODE_FULL || !source)
3968         && arg->mode != NEW_IMAGE_MODE_EXISTING)
3969     {
3970         /* create new image w/o backing file */
3971         assert(format);
3972         bdrv_img_create(arg->target, format,
3973                         NULL, NULL, NULL, size, flags, false, &local_err);
3974     } else {
3975         switch (arg->mode) {
3976         case NEW_IMAGE_MODE_EXISTING:
3977             break;
3978         case NEW_IMAGE_MODE_ABSOLUTE_PATHS:
3979             /* create new image with backing file */
3980             bdrv_refresh_filename(source);
3981             bdrv_img_create(arg->target, format,
3982                             source->filename,
3983                             source->drv->format_name,
3984                             NULL, size, flags, false, &local_err);
3985             break;
3986         default:
3987             abort();
3988         }
3989     }
3990 
3991     if (local_err) {
3992         error_propagate(errp, local_err);
3993         goto out;
3994     }
3995 
3996     options = qdict_new();
3997     if (arg->has_node_name) {
3998         qdict_put_str(options, "node-name", arg->node_name);
3999     }
4000     if (format) {
4001         qdict_put_str(options, "driver", format);
4002     }
4003 
4004     /* Mirroring takes care of copy-on-write using the source's backing
4005      * file.
4006      */
4007     target_bs = bdrv_open(arg->target, NULL, options, flags, errp);
4008     if (!target_bs) {
4009         goto out;
4010     }
4011 
4012     zero_target = (arg->sync == MIRROR_SYNC_MODE_FULL &&
4013                    (arg->mode == NEW_IMAGE_MODE_EXISTING ||
4014                     !bdrv_has_zero_init(target_bs)));
4015 
4016     ret = bdrv_try_set_aio_context(target_bs, aio_context, errp);
4017     if (ret < 0) {
4018         bdrv_unref(target_bs);
4019         goto out;
4020     }
4021 
4022     blockdev_mirror_common(arg->has_job_id ? arg->job_id : NULL, bs, target_bs,
4023                            arg->has_replaces, arg->replaces, arg->sync,
4024                            backing_mode, zero_target,
4025                            arg->has_speed, arg->speed,
4026                            arg->has_granularity, arg->granularity,
4027                            arg->has_buf_size, arg->buf_size,
4028                            arg->has_on_source_error, arg->on_source_error,
4029                            arg->has_on_target_error, arg->on_target_error,
4030                            arg->has_unmap, arg->unmap,
4031                            false, NULL,
4032                            arg->has_copy_mode, arg->copy_mode,
4033                            arg->has_auto_finalize, arg->auto_finalize,
4034                            arg->has_auto_dismiss, arg->auto_dismiss,
4035                            &local_err);
4036     bdrv_unref(target_bs);
4037     error_propagate(errp, local_err);
4038 out:
4039     aio_context_release(aio_context);
4040 }
4041 
4042 void qmp_blockdev_mirror(bool has_job_id, const char *job_id,
4043                          const char *device, const char *target,
4044                          bool has_replaces, const char *replaces,
4045                          MirrorSyncMode sync,
4046                          bool has_speed, int64_t speed,
4047                          bool has_granularity, uint32_t granularity,
4048                          bool has_buf_size, int64_t buf_size,
4049                          bool has_on_source_error,
4050                          BlockdevOnError on_source_error,
4051                          bool has_on_target_error,
4052                          BlockdevOnError on_target_error,
4053                          bool has_filter_node_name,
4054                          const char *filter_node_name,
4055                          bool has_copy_mode, MirrorCopyMode copy_mode,
4056                          bool has_auto_finalize, bool auto_finalize,
4057                          bool has_auto_dismiss, bool auto_dismiss,
4058                          Error **errp)
4059 {
4060     BlockDriverState *bs;
4061     BlockDriverState *target_bs;
4062     AioContext *aio_context;
4063     BlockMirrorBackingMode backing_mode = MIRROR_LEAVE_BACKING_CHAIN;
4064     Error *local_err = NULL;
4065     bool zero_target;
4066     int ret;
4067 
4068     bs = qmp_get_root_bs(device, errp);
4069     if (!bs) {
4070         return;
4071     }
4072 
4073     target_bs = bdrv_lookup_bs(target, target, errp);
4074     if (!target_bs) {
4075         return;
4076     }
4077 
4078     zero_target = (sync == MIRROR_SYNC_MODE_FULL);
4079 
4080     aio_context = bdrv_get_aio_context(bs);
4081     aio_context_acquire(aio_context);
4082 
4083     ret = bdrv_try_set_aio_context(target_bs, aio_context, errp);
4084     if (ret < 0) {
4085         goto out;
4086     }
4087 
4088     blockdev_mirror_common(has_job_id ? job_id : NULL, bs, target_bs,
4089                            has_replaces, replaces, sync, backing_mode,
4090                            zero_target, has_speed, speed,
4091                            has_granularity, granularity,
4092                            has_buf_size, buf_size,
4093                            has_on_source_error, on_source_error,
4094                            has_on_target_error, on_target_error,
4095                            true, true,
4096                            has_filter_node_name, filter_node_name,
4097                            has_copy_mode, copy_mode,
4098                            has_auto_finalize, auto_finalize,
4099                            has_auto_dismiss, auto_dismiss,
4100                            &local_err);
4101     error_propagate(errp, local_err);
4102 out:
4103     aio_context_release(aio_context);
4104 }
4105 
4106 /* Get a block job using its ID and acquire its AioContext */
4107 static BlockJob *find_block_job(const char *id, AioContext **aio_context,
4108                                 Error **errp)
4109 {
4110     BlockJob *job;
4111 
4112     assert(id != NULL);
4113 
4114     *aio_context = NULL;
4115 
4116     job = block_job_get(id);
4117 
4118     if (!job) {
4119         error_set(errp, ERROR_CLASS_DEVICE_NOT_ACTIVE,
4120                   "Block job '%s' not found", id);
4121         return NULL;
4122     }
4123 
4124     *aio_context = blk_get_aio_context(job->blk);
4125     aio_context_acquire(*aio_context);
4126 
4127     return job;
4128 }
4129 
4130 void qmp_block_job_set_speed(const char *device, int64_t speed, Error **errp)
4131 {
4132     AioContext *aio_context;
4133     BlockJob *job = find_block_job(device, &aio_context, errp);
4134 
4135     if (!job) {
4136         return;
4137     }
4138 
4139     block_job_set_speed(job, speed, errp);
4140     aio_context_release(aio_context);
4141 }
4142 
4143 void qmp_block_job_cancel(const char *device,
4144                           bool has_force, bool force, Error **errp)
4145 {
4146     AioContext *aio_context;
4147     BlockJob *job = find_block_job(device, &aio_context, errp);
4148 
4149     if (!job) {
4150         return;
4151     }
4152 
4153     if (!has_force) {
4154         force = false;
4155     }
4156 
4157     if (job_user_paused(&job->job) && !force) {
4158         error_setg(errp, "The block job for device '%s' is currently paused",
4159                    device);
4160         goto out;
4161     }
4162 
4163     trace_qmp_block_job_cancel(job);
4164     job_user_cancel(&job->job, force, errp);
4165 out:
4166     aio_context_release(aio_context);
4167 }
4168 
4169 void qmp_block_job_pause(const char *device, Error **errp)
4170 {
4171     AioContext *aio_context;
4172     BlockJob *job = find_block_job(device, &aio_context, errp);
4173 
4174     if (!job) {
4175         return;
4176     }
4177 
4178     trace_qmp_block_job_pause(job);
4179     job_user_pause(&job->job, errp);
4180     aio_context_release(aio_context);
4181 }
4182 
4183 void qmp_block_job_resume(const char *device, Error **errp)
4184 {
4185     AioContext *aio_context;
4186     BlockJob *job = find_block_job(device, &aio_context, errp);
4187 
4188     if (!job) {
4189         return;
4190     }
4191 
4192     trace_qmp_block_job_resume(job);
4193     job_user_resume(&job->job, errp);
4194     aio_context_release(aio_context);
4195 }
4196 
4197 void qmp_block_job_complete(const char *device, Error **errp)
4198 {
4199     AioContext *aio_context;
4200     BlockJob *job = find_block_job(device, &aio_context, errp);
4201 
4202     if (!job) {
4203         return;
4204     }
4205 
4206     trace_qmp_block_job_complete(job);
4207     job_complete(&job->job, errp);
4208     aio_context_release(aio_context);
4209 }
4210 
4211 void qmp_block_job_finalize(const char *id, Error **errp)
4212 {
4213     AioContext *aio_context;
4214     BlockJob *job = find_block_job(id, &aio_context, errp);
4215 
4216     if (!job) {
4217         return;
4218     }
4219 
4220     trace_qmp_block_job_finalize(job);
4221     job_finalize(&job->job, errp);
4222     aio_context_release(aio_context);
4223 }
4224 
4225 void qmp_block_job_dismiss(const char *id, Error **errp)
4226 {
4227     AioContext *aio_context;
4228     BlockJob *bjob = find_block_job(id, &aio_context, errp);
4229     Job *job;
4230 
4231     if (!bjob) {
4232         return;
4233     }
4234 
4235     trace_qmp_block_job_dismiss(bjob);
4236     job = &bjob->job;
4237     job_dismiss(&job, errp);
4238     aio_context_release(aio_context);
4239 }
4240 
4241 void qmp_change_backing_file(const char *device,
4242                              const char *image_node_name,
4243                              const char *backing_file,
4244                              Error **errp)
4245 {
4246     BlockDriverState *bs = NULL;
4247     AioContext *aio_context;
4248     BlockDriverState *image_bs = NULL;
4249     Error *local_err = NULL;
4250     bool ro;
4251     int ret;
4252 
4253     bs = qmp_get_root_bs(device, errp);
4254     if (!bs) {
4255         return;
4256     }
4257 
4258     aio_context = bdrv_get_aio_context(bs);
4259     aio_context_acquire(aio_context);
4260 
4261     image_bs = bdrv_lookup_bs(NULL, image_node_name, &local_err);
4262     if (local_err) {
4263         error_propagate(errp, local_err);
4264         goto out;
4265     }
4266 
4267     if (!image_bs) {
4268         error_setg(errp, "image file not found");
4269         goto out;
4270     }
4271 
4272     if (bdrv_find_base(image_bs) == image_bs) {
4273         error_setg(errp, "not allowing backing file change on an image "
4274                          "without a backing file");
4275         goto out;
4276     }
4277 
4278     /* even though we are not necessarily operating on bs, we need it to
4279      * determine if block ops are currently prohibited on the chain */
4280     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_CHANGE, errp)) {
4281         goto out;
4282     }
4283 
4284     /* final sanity check */
4285     if (!bdrv_chain_contains(bs, image_bs)) {
4286         error_setg(errp, "'%s' and image file are not in the same chain",
4287                    device);
4288         goto out;
4289     }
4290 
4291     /* if not r/w, reopen to make r/w */
4292     ro = bdrv_is_read_only(image_bs);
4293 
4294     if (ro) {
4295         if (bdrv_reopen_set_read_only(image_bs, false, errp) != 0) {
4296             goto out;
4297         }
4298     }
4299 
4300     ret = bdrv_change_backing_file(image_bs, backing_file,
4301                                image_bs->drv ? image_bs->drv->format_name : "");
4302 
4303     if (ret < 0) {
4304         error_setg_errno(errp, -ret, "Could not change backing file to '%s'",
4305                          backing_file);
4306         /* don't exit here, so we can try to restore open flags if
4307          * appropriate */
4308     }
4309 
4310     if (ro) {
4311         bdrv_reopen_set_read_only(image_bs, true, &local_err);
4312         error_propagate(errp, local_err);
4313     }
4314 
4315 out:
4316     aio_context_release(aio_context);
4317 }
4318 
4319 void hmp_drive_add_node(Monitor *mon, const char *optstr)
4320 {
4321     QemuOpts *opts;
4322     QDict *qdict;
4323     Error *local_err = NULL;
4324 
4325     opts = qemu_opts_parse_noisily(&qemu_drive_opts, optstr, false);
4326     if (!opts) {
4327         return;
4328     }
4329 
4330     qdict = qemu_opts_to_qdict(opts, NULL);
4331 
4332     if (!qdict_get_try_str(qdict, "node-name")) {
4333         qobject_unref(qdict);
4334         error_report("'node-name' needs to be specified");
4335         goto out;
4336     }
4337 
4338     BlockDriverState *bs = bds_tree_init(qdict, &local_err);
4339     if (!bs) {
4340         error_report_err(local_err);
4341         goto out;
4342     }
4343 
4344     QTAILQ_INSERT_TAIL(&monitor_bdrv_states, bs, monitor_list);
4345 
4346 out:
4347     qemu_opts_del(opts);
4348 }
4349 
4350 void qmp_blockdev_add(BlockdevOptions *options, Error **errp)
4351 {
4352     BlockDriverState *bs;
4353     QObject *obj;
4354     Visitor *v = qobject_output_visitor_new(&obj);
4355     QDict *qdict;
4356     Error *local_err = NULL;
4357 
4358     visit_type_BlockdevOptions(v, NULL, &options, &local_err);
4359     if (local_err) {
4360         error_propagate(errp, local_err);
4361         goto fail;
4362     }
4363 
4364     visit_complete(v, &obj);
4365     qdict = qobject_to(QDict, obj);
4366 
4367     qdict_flatten(qdict);
4368 
4369     if (!qdict_get_try_str(qdict, "node-name")) {
4370         error_setg(errp, "'node-name' must be specified for the root node");
4371         goto fail;
4372     }
4373 
4374     bs = bds_tree_init(qdict, errp);
4375     if (!bs) {
4376         goto fail;
4377     }
4378 
4379     QTAILQ_INSERT_TAIL(&monitor_bdrv_states, bs, monitor_list);
4380 
4381 fail:
4382     visit_free(v);
4383 }
4384 
4385 void qmp_x_blockdev_reopen(BlockdevOptions *options, Error **errp)
4386 {
4387     BlockDriverState *bs;
4388     AioContext *ctx;
4389     QObject *obj;
4390     Visitor *v = qobject_output_visitor_new(&obj);
4391     Error *local_err = NULL;
4392     BlockReopenQueue *queue;
4393     QDict *qdict;
4394 
4395     /* Check for the selected node name */
4396     if (!options->has_node_name) {
4397         error_setg(errp, "Node name not specified");
4398         goto fail;
4399     }
4400 
4401     bs = bdrv_find_node(options->node_name);
4402     if (!bs) {
4403         error_setg(errp, "Cannot find node named '%s'", options->node_name);
4404         goto fail;
4405     }
4406 
4407     /* Put all options in a QDict and flatten it */
4408     visit_type_BlockdevOptions(v, NULL, &options, &local_err);
4409     if (local_err) {
4410         error_propagate(errp, local_err);
4411         goto fail;
4412     }
4413 
4414     visit_complete(v, &obj);
4415     qdict = qobject_to(QDict, obj);
4416 
4417     qdict_flatten(qdict);
4418 
4419     /* Perform the reopen operation */
4420     ctx = bdrv_get_aio_context(bs);
4421     aio_context_acquire(ctx);
4422     bdrv_subtree_drained_begin(bs);
4423     queue = bdrv_reopen_queue(NULL, bs, qdict, false);
4424     bdrv_reopen_multiple(queue, errp);
4425     bdrv_subtree_drained_end(bs);
4426     aio_context_release(ctx);
4427 
4428 fail:
4429     visit_free(v);
4430 }
4431 
4432 void qmp_blockdev_del(const char *node_name, Error **errp)
4433 {
4434     AioContext *aio_context;
4435     BlockDriverState *bs;
4436 
4437     bs = bdrv_find_node(node_name);
4438     if (!bs) {
4439         error_setg(errp, "Cannot find node %s", node_name);
4440         return;
4441     }
4442     if (bdrv_has_blk(bs)) {
4443         error_setg(errp, "Node %s is in use", node_name);
4444         return;
4445     }
4446     aio_context = bdrv_get_aio_context(bs);
4447     aio_context_acquire(aio_context);
4448 
4449     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_DRIVE_DEL, errp)) {
4450         goto out;
4451     }
4452 
4453     if (!QTAILQ_IN_USE(bs, monitor_list)) {
4454         error_setg(errp, "Node %s is not owned by the monitor",
4455                    bs->node_name);
4456         goto out;
4457     }
4458 
4459     if (bs->refcnt > 1) {
4460         error_setg(errp, "Block device %s is in use",
4461                    bdrv_get_device_or_node_name(bs));
4462         goto out;
4463     }
4464 
4465     QTAILQ_REMOVE(&monitor_bdrv_states, bs, monitor_list);
4466     bdrv_unref(bs);
4467 
4468 out:
4469     aio_context_release(aio_context);
4470 }
4471 
4472 static BdrvChild *bdrv_find_child(BlockDriverState *parent_bs,
4473                                   const char *child_name)
4474 {
4475     BdrvChild *child;
4476 
4477     QLIST_FOREACH(child, &parent_bs->children, next) {
4478         if (strcmp(child->name, child_name) == 0) {
4479             return child;
4480         }
4481     }
4482 
4483     return NULL;
4484 }
4485 
4486 void qmp_x_blockdev_change(const char *parent, bool has_child,
4487                            const char *child, bool has_node,
4488                            const char *node, Error **errp)
4489 {
4490     BlockDriverState *parent_bs, *new_bs = NULL;
4491     BdrvChild *p_child;
4492 
4493     parent_bs = bdrv_lookup_bs(parent, parent, errp);
4494     if (!parent_bs) {
4495         return;
4496     }
4497 
4498     if (has_child == has_node) {
4499         if (has_child) {
4500             error_setg(errp, "The parameters child and node are in conflict");
4501         } else {
4502             error_setg(errp, "Either child or node must be specified");
4503         }
4504         return;
4505     }
4506 
4507     if (has_child) {
4508         p_child = bdrv_find_child(parent_bs, child);
4509         if (!p_child) {
4510             error_setg(errp, "Node '%s' does not have child '%s'",
4511                        parent, child);
4512             return;
4513         }
4514         bdrv_del_child(parent_bs, p_child, errp);
4515     }
4516 
4517     if (has_node) {
4518         new_bs = bdrv_find_node(node);
4519         if (!new_bs) {
4520             error_setg(errp, "Node '%s' not found", node);
4521             return;
4522         }
4523         bdrv_add_child(parent_bs, new_bs, errp);
4524     }
4525 }
4526 
4527 BlockJobInfoList *qmp_query_block_jobs(Error **errp)
4528 {
4529     BlockJobInfoList *head = NULL, **p_next = &head;
4530     BlockJob *job;
4531 
4532     for (job = block_job_next(NULL); job; job = block_job_next(job)) {
4533         BlockJobInfoList *elem;
4534         AioContext *aio_context;
4535 
4536         if (block_job_is_internal(job)) {
4537             continue;
4538         }
4539         elem = g_new0(BlockJobInfoList, 1);
4540         aio_context = blk_get_aio_context(job->blk);
4541         aio_context_acquire(aio_context);
4542         elem->value = block_job_query(job, errp);
4543         aio_context_release(aio_context);
4544         if (!elem->value) {
4545             g_free(elem);
4546             qapi_free_BlockJobInfoList(head);
4547             return NULL;
4548         }
4549         *p_next = elem;
4550         p_next = &elem->next;
4551     }
4552 
4553     return head;
4554 }
4555 
4556 void qmp_x_blockdev_set_iothread(const char *node_name, StrOrNull *iothread,
4557                                  bool has_force, bool force, Error **errp)
4558 {
4559     AioContext *old_context;
4560     AioContext *new_context;
4561     BlockDriverState *bs;
4562 
4563     bs = bdrv_find_node(node_name);
4564     if (!bs) {
4565         error_setg(errp, "Cannot find node %s", node_name);
4566         return;
4567     }
4568 
4569     /* Protects against accidents. */
4570     if (!(has_force && force) && bdrv_has_blk(bs)) {
4571         error_setg(errp, "Node %s is associated with a BlockBackend and could "
4572                          "be in use (use force=true to override this check)",
4573                          node_name);
4574         return;
4575     }
4576 
4577     if (iothread->type == QTYPE_QSTRING) {
4578         IOThread *obj = iothread_by_id(iothread->u.s);
4579         if (!obj) {
4580             error_setg(errp, "Cannot find iothread %s", iothread->u.s);
4581             return;
4582         }
4583 
4584         new_context = iothread_get_aio_context(obj);
4585     } else {
4586         new_context = qemu_get_aio_context();
4587     }
4588 
4589     old_context = bdrv_get_aio_context(bs);
4590     aio_context_acquire(old_context);
4591 
4592     bdrv_try_set_aio_context(bs, new_context, errp);
4593 
4594     aio_context_release(old_context);
4595 }
4596 
4597 void qmp_block_latency_histogram_set(
4598     const char *id,
4599     bool has_boundaries, uint64List *boundaries,
4600     bool has_boundaries_read, uint64List *boundaries_read,
4601     bool has_boundaries_write, uint64List *boundaries_write,
4602     bool has_boundaries_flush, uint64List *boundaries_flush,
4603     Error **errp)
4604 {
4605     BlockBackend *blk = qmp_get_blk(NULL, id, errp);
4606     BlockAcctStats *stats;
4607     int ret;
4608 
4609     if (!blk) {
4610         return;
4611     }
4612 
4613     stats = blk_get_stats(blk);
4614 
4615     if (!has_boundaries && !has_boundaries_read && !has_boundaries_write &&
4616         !has_boundaries_flush)
4617     {
4618         block_latency_histograms_clear(stats);
4619         return;
4620     }
4621 
4622     if (has_boundaries || has_boundaries_read) {
4623         ret = block_latency_histogram_set(
4624             stats, BLOCK_ACCT_READ,
4625             has_boundaries_read ? boundaries_read : boundaries);
4626         if (ret) {
4627             error_setg(errp, "Device '%s' set read boundaries fail", id);
4628             return;
4629         }
4630     }
4631 
4632     if (has_boundaries || has_boundaries_write) {
4633         ret = block_latency_histogram_set(
4634             stats, BLOCK_ACCT_WRITE,
4635             has_boundaries_write ? boundaries_write : boundaries);
4636         if (ret) {
4637             error_setg(errp, "Device '%s' set write boundaries fail", id);
4638             return;
4639         }
4640     }
4641 
4642     if (has_boundaries || has_boundaries_flush) {
4643         ret = block_latency_histogram_set(
4644             stats, BLOCK_ACCT_FLUSH,
4645             has_boundaries_flush ? boundaries_flush : boundaries);
4646         if (ret) {
4647             error_setg(errp, "Device '%s' set flush boundaries fail", id);
4648             return;
4649         }
4650     }
4651 }
4652 
4653 QemuOptsList qemu_common_drive_opts = {
4654     .name = "drive",
4655     .head = QTAILQ_HEAD_INITIALIZER(qemu_common_drive_opts.head),
4656     .desc = {
4657         {
4658             .name = "snapshot",
4659             .type = QEMU_OPT_BOOL,
4660             .help = "enable/disable snapshot mode",
4661         },{
4662             .name = "aio",
4663             .type = QEMU_OPT_STRING,
4664             .help = "host AIO implementation (threads, native)",
4665         },{
4666             .name = BDRV_OPT_CACHE_WB,
4667             .type = QEMU_OPT_BOOL,
4668             .help = "Enable writeback mode",
4669         },{
4670             .name = "format",
4671             .type = QEMU_OPT_STRING,
4672             .help = "disk format (raw, qcow2, ...)",
4673         },{
4674             .name = "rerror",
4675             .type = QEMU_OPT_STRING,
4676             .help = "read error action",
4677         },{
4678             .name = "werror",
4679             .type = QEMU_OPT_STRING,
4680             .help = "write error action",
4681         },{
4682             .name = BDRV_OPT_READ_ONLY,
4683             .type = QEMU_OPT_BOOL,
4684             .help = "open drive file as read-only",
4685         },
4686 
4687         THROTTLE_OPTS,
4688 
4689         {
4690             .name = "throttling.group",
4691             .type = QEMU_OPT_STRING,
4692             .help = "name of the block throttling group",
4693         },{
4694             .name = "copy-on-read",
4695             .type = QEMU_OPT_BOOL,
4696             .help = "copy read data from backing file into image file",
4697         },{
4698             .name = "detect-zeroes",
4699             .type = QEMU_OPT_STRING,
4700             .help = "try to optimize zero writes (off, on, unmap)",
4701         },{
4702             .name = "stats-account-invalid",
4703             .type = QEMU_OPT_BOOL,
4704             .help = "whether to account for invalid I/O operations "
4705                     "in the statistics",
4706         },{
4707             .name = "stats-account-failed",
4708             .type = QEMU_OPT_BOOL,
4709             .help = "whether to account for failed I/O operations "
4710                     "in the statistics",
4711         },
4712         { /* end of list */ }
4713     },
4714 };
4715 
4716 QemuOptsList qemu_drive_opts = {
4717     .name = "drive",
4718     .head = QTAILQ_HEAD_INITIALIZER(qemu_drive_opts.head),
4719     .desc = {
4720         /*
4721          * no elements => accept any params
4722          * validation will happen later
4723          */
4724         { /* end of list */ }
4725     },
4726 };
4727