xref: /qemu/blockdev.c (revision 7a4e543d)
1 /*
2  * QEMU host block devices
3  *
4  * Copyright (c) 2003-2008 Fabrice Bellard
5  *
6  * This work is licensed under the terms of the GNU GPL, version 2 or
7  * later.  See the COPYING file in the top-level directory.
8  *
9  * This file incorporates work covered by the following copyright and
10  * permission notice:
11  *
12  * Copyright (c) 2003-2008 Fabrice Bellard
13  *
14  * Permission is hereby granted, free of charge, to any person obtaining a copy
15  * of this software and associated documentation files (the "Software"), to deal
16  * in the Software without restriction, including without limitation the rights
17  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
18  * copies of the Software, and to permit persons to whom the Software is
19  * furnished to do so, subject to the following conditions:
20  *
21  * The above copyright notice and this permission notice shall be included in
22  * all copies or substantial portions of the Software.
23  *
24  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
25  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
26  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
27  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
28  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
29  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
30  * THE SOFTWARE.
31  */
32 
33 #include "qemu/osdep.h"
34 #include "sysemu/block-backend.h"
35 #include "sysemu/blockdev.h"
36 #include "hw/block/block.h"
37 #include "block/blockjob.h"
38 #include "block/throttle-groups.h"
39 #include "monitor/monitor.h"
40 #include "qemu/error-report.h"
41 #include "qemu/option.h"
42 #include "qemu/config-file.h"
43 #include "qapi/qmp/types.h"
44 #include "qapi-visit.h"
45 #include "qapi/qmp/qerror.h"
46 #include "qapi/qmp-output-visitor.h"
47 #include "qapi/util.h"
48 #include "sysemu/sysemu.h"
49 #include "block/block_int.h"
50 #include "qmp-commands.h"
51 #include "trace.h"
52 #include "sysemu/arch_init.h"
53 
54 static QTAILQ_HEAD(, BlockDriverState) monitor_bdrv_states =
55     QTAILQ_HEAD_INITIALIZER(monitor_bdrv_states);
56 
57 static const char *const if_name[IF_COUNT] = {
58     [IF_NONE] = "none",
59     [IF_IDE] = "ide",
60     [IF_SCSI] = "scsi",
61     [IF_FLOPPY] = "floppy",
62     [IF_PFLASH] = "pflash",
63     [IF_MTD] = "mtd",
64     [IF_SD] = "sd",
65     [IF_VIRTIO] = "virtio",
66     [IF_XEN] = "xen",
67 };
68 
69 static int if_max_devs[IF_COUNT] = {
70     /*
71      * Do not change these numbers!  They govern how drive option
72      * index maps to unit and bus.  That mapping is ABI.
73      *
74      * All controllers used to imlement if=T drives need to support
75      * if_max_devs[T] units, for any T with if_max_devs[T] != 0.
76      * Otherwise, some index values map to "impossible" bus, unit
77      * values.
78      *
79      * For instance, if you change [IF_SCSI] to 255, -drive
80      * if=scsi,index=12 no longer means bus=1,unit=5, but
81      * bus=0,unit=12.  With an lsi53c895a controller (7 units max),
82      * the drive can't be set up.  Regression.
83      */
84     [IF_IDE] = 2,
85     [IF_SCSI] = 7,
86 };
87 
88 /**
89  * Boards may call this to offer board-by-board overrides
90  * of the default, global values.
91  */
92 void override_max_devs(BlockInterfaceType type, int max_devs)
93 {
94     BlockBackend *blk;
95     DriveInfo *dinfo;
96 
97     if (max_devs <= 0) {
98         return;
99     }
100 
101     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
102         dinfo = blk_legacy_dinfo(blk);
103         if (dinfo->type == type) {
104             fprintf(stderr, "Cannot override units-per-bus property of"
105                     " the %s interface, because a drive of that type has"
106                     " already been added.\n", if_name[type]);
107             g_assert_not_reached();
108         }
109     }
110 
111     if_max_devs[type] = max_devs;
112 }
113 
114 /*
115  * We automatically delete the drive when a device using it gets
116  * unplugged.  Questionable feature, but we can't just drop it.
117  * Device models call blockdev_mark_auto_del() to schedule the
118  * automatic deletion, and generic qdev code calls blockdev_auto_del()
119  * when deletion is actually safe.
120  */
121 void blockdev_mark_auto_del(BlockBackend *blk)
122 {
123     DriveInfo *dinfo = blk_legacy_dinfo(blk);
124     BlockDriverState *bs = blk_bs(blk);
125     AioContext *aio_context;
126 
127     if (!dinfo) {
128         return;
129     }
130 
131     if (bs) {
132         aio_context = bdrv_get_aio_context(bs);
133         aio_context_acquire(aio_context);
134 
135         if (bs->job) {
136             block_job_cancel(bs->job);
137         }
138 
139         aio_context_release(aio_context);
140     }
141 
142     dinfo->auto_del = 1;
143 }
144 
145 void blockdev_auto_del(BlockBackend *blk)
146 {
147     DriveInfo *dinfo = blk_legacy_dinfo(blk);
148 
149     if (dinfo && dinfo->auto_del) {
150         blk_unref(blk);
151     }
152 }
153 
154 /**
155  * Returns the current mapping of how many units per bus
156  * a particular interface can support.
157  *
158  *  A positive integer indicates n units per bus.
159  *  0 implies the mapping has not been established.
160  * -1 indicates an invalid BlockInterfaceType was given.
161  */
162 int drive_get_max_devs(BlockInterfaceType type)
163 {
164     if (type >= IF_IDE && type < IF_COUNT) {
165         return if_max_devs[type];
166     }
167 
168     return -1;
169 }
170 
171 static int drive_index_to_bus_id(BlockInterfaceType type, int index)
172 {
173     int max_devs = if_max_devs[type];
174     return max_devs ? index / max_devs : 0;
175 }
176 
177 static int drive_index_to_unit_id(BlockInterfaceType type, int index)
178 {
179     int max_devs = if_max_devs[type];
180     return max_devs ? index % max_devs : index;
181 }
182 
183 QemuOpts *drive_def(const char *optstr)
184 {
185     return qemu_opts_parse_noisily(qemu_find_opts("drive"), optstr, false);
186 }
187 
188 QemuOpts *drive_add(BlockInterfaceType type, int index, const char *file,
189                     const char *optstr)
190 {
191     QemuOpts *opts;
192 
193     opts = drive_def(optstr);
194     if (!opts) {
195         return NULL;
196     }
197     if (type != IF_DEFAULT) {
198         qemu_opt_set(opts, "if", if_name[type], &error_abort);
199     }
200     if (index >= 0) {
201         qemu_opt_set_number(opts, "index", index, &error_abort);
202     }
203     if (file)
204         qemu_opt_set(opts, "file", file, &error_abort);
205     return opts;
206 }
207 
208 DriveInfo *drive_get(BlockInterfaceType type, int bus, int unit)
209 {
210     BlockBackend *blk;
211     DriveInfo *dinfo;
212 
213     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
214         dinfo = blk_legacy_dinfo(blk);
215         if (dinfo && dinfo->type == type
216             && dinfo->bus == bus && dinfo->unit == unit) {
217             return dinfo;
218         }
219     }
220 
221     return NULL;
222 }
223 
224 bool drive_check_orphaned(void)
225 {
226     BlockBackend *blk;
227     DriveInfo *dinfo;
228     bool rs = false;
229 
230     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
231         dinfo = blk_legacy_dinfo(blk);
232         /* If dinfo->bdrv->dev is NULL, it has no device attached. */
233         /* Unless this is a default drive, this may be an oversight. */
234         if (!blk_get_attached_dev(blk) && !dinfo->is_default &&
235             dinfo->type != IF_NONE) {
236             fprintf(stderr, "Warning: Orphaned drive without device: "
237                     "id=%s,file=%s,if=%s,bus=%d,unit=%d\n",
238                     blk_name(blk), blk_bs(blk) ? blk_bs(blk)->filename : "",
239                     if_name[dinfo->type], dinfo->bus, dinfo->unit);
240             rs = true;
241         }
242     }
243 
244     return rs;
245 }
246 
247 DriveInfo *drive_get_by_index(BlockInterfaceType type, int index)
248 {
249     return drive_get(type,
250                      drive_index_to_bus_id(type, index),
251                      drive_index_to_unit_id(type, index));
252 }
253 
254 int drive_get_max_bus(BlockInterfaceType type)
255 {
256     int max_bus;
257     BlockBackend *blk;
258     DriveInfo *dinfo;
259 
260     max_bus = -1;
261     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
262         dinfo = blk_legacy_dinfo(blk);
263         if (dinfo && dinfo->type == type && dinfo->bus > max_bus) {
264             max_bus = dinfo->bus;
265         }
266     }
267     return max_bus;
268 }
269 
270 /* Get a block device.  This should only be used for single-drive devices
271    (e.g. SD/Floppy/MTD).  Multi-disk devices (scsi/ide) should use the
272    appropriate bus.  */
273 DriveInfo *drive_get_next(BlockInterfaceType type)
274 {
275     static int next_block_unit[IF_COUNT];
276 
277     return drive_get(type, 0, next_block_unit[type]++);
278 }
279 
280 static void bdrv_format_print(void *opaque, const char *name)
281 {
282     error_printf(" %s", name);
283 }
284 
285 typedef struct {
286     QEMUBH *bh;
287     BlockDriverState *bs;
288 } BDRVPutRefBH;
289 
290 static int parse_block_error_action(const char *buf, bool is_read, Error **errp)
291 {
292     if (!strcmp(buf, "ignore")) {
293         return BLOCKDEV_ON_ERROR_IGNORE;
294     } else if (!is_read && !strcmp(buf, "enospc")) {
295         return BLOCKDEV_ON_ERROR_ENOSPC;
296     } else if (!strcmp(buf, "stop")) {
297         return BLOCKDEV_ON_ERROR_STOP;
298     } else if (!strcmp(buf, "report")) {
299         return BLOCKDEV_ON_ERROR_REPORT;
300     } else {
301         error_setg(errp, "'%s' invalid %s error action",
302                    buf, is_read ? "read" : "write");
303         return -1;
304     }
305 }
306 
307 static bool parse_stats_intervals(BlockAcctStats *stats, QList *intervals,
308                                   Error **errp)
309 {
310     const QListEntry *entry;
311     for (entry = qlist_first(intervals); entry; entry = qlist_next(entry)) {
312         switch (qobject_type(entry->value)) {
313 
314         case QTYPE_QSTRING: {
315             unsigned long long length;
316             const char *str = qstring_get_str(qobject_to_qstring(entry->value));
317             if (parse_uint_full(str, &length, 10) == 0 &&
318                 length > 0 && length <= UINT_MAX) {
319                 block_acct_add_interval(stats, (unsigned) length);
320             } else {
321                 error_setg(errp, "Invalid interval length: %s", str);
322                 return false;
323             }
324             break;
325         }
326 
327         case QTYPE_QINT: {
328             int64_t length = qint_get_int(qobject_to_qint(entry->value));
329             if (length > 0 && length <= UINT_MAX) {
330                 block_acct_add_interval(stats, (unsigned) length);
331             } else {
332                 error_setg(errp, "Invalid interval length: %" PRId64, length);
333                 return false;
334             }
335             break;
336         }
337 
338         default:
339             error_setg(errp, "The specification of stats-intervals is invalid");
340             return false;
341         }
342     }
343     return true;
344 }
345 
346 static bool check_throttle_config(ThrottleConfig *cfg, Error **errp)
347 {
348     if (throttle_conflicting(cfg)) {
349         error_setg(errp, "bps/iops/max total values and read/write values"
350                          " cannot be used at the same time");
351         return false;
352     }
353 
354     if (!throttle_is_valid(cfg)) {
355         error_setg(errp, "bps/iops/max values must be within [0, %lld]",
356                    THROTTLE_VALUE_MAX);
357         return false;
358     }
359 
360     if (throttle_max_is_missing_limit(cfg)) {
361         error_setg(errp, "bps_max/iops_max require corresponding"
362                          " bps/iops values");
363         return false;
364     }
365 
366     return true;
367 }
368 
369 typedef enum { MEDIA_DISK, MEDIA_CDROM } DriveMediaType;
370 
371 /* All parameters but @opts are optional and may be set to NULL. */
372 static void extract_common_blockdev_options(QemuOpts *opts, int *bdrv_flags,
373     const char **throttling_group, ThrottleConfig *throttle_cfg,
374     BlockdevDetectZeroesOptions *detect_zeroes, Error **errp)
375 {
376     const char *discard;
377     Error *local_error = NULL;
378     const char *aio;
379 
380     if (bdrv_flags) {
381         if (!qemu_opt_get_bool(opts, "read-only", false)) {
382             *bdrv_flags |= BDRV_O_RDWR;
383         }
384         if (qemu_opt_get_bool(opts, "copy-on-read", false)) {
385             *bdrv_flags |= BDRV_O_COPY_ON_READ;
386         }
387 
388         if ((discard = qemu_opt_get(opts, "discard")) != NULL) {
389             if (bdrv_parse_discard_flags(discard, bdrv_flags) != 0) {
390                 error_setg(errp, "Invalid discard option");
391                 return;
392             }
393         }
394 
395         if ((aio = qemu_opt_get(opts, "aio")) != NULL) {
396             if (!strcmp(aio, "native")) {
397                 *bdrv_flags |= BDRV_O_NATIVE_AIO;
398             } else if (!strcmp(aio, "threads")) {
399                 /* this is the default */
400             } else {
401                error_setg(errp, "invalid aio option");
402                return;
403             }
404         }
405     }
406 
407     /* disk I/O throttling */
408     if (throttling_group) {
409         *throttling_group = qemu_opt_get(opts, "throttling.group");
410     }
411 
412     if (throttle_cfg) {
413         memset(throttle_cfg, 0, sizeof(*throttle_cfg));
414         throttle_cfg->buckets[THROTTLE_BPS_TOTAL].avg =
415             qemu_opt_get_number(opts, "throttling.bps-total", 0);
416         throttle_cfg->buckets[THROTTLE_BPS_READ].avg  =
417             qemu_opt_get_number(opts, "throttling.bps-read", 0);
418         throttle_cfg->buckets[THROTTLE_BPS_WRITE].avg =
419             qemu_opt_get_number(opts, "throttling.bps-write", 0);
420         throttle_cfg->buckets[THROTTLE_OPS_TOTAL].avg =
421             qemu_opt_get_number(opts, "throttling.iops-total", 0);
422         throttle_cfg->buckets[THROTTLE_OPS_READ].avg =
423             qemu_opt_get_number(opts, "throttling.iops-read", 0);
424         throttle_cfg->buckets[THROTTLE_OPS_WRITE].avg =
425             qemu_opt_get_number(opts, "throttling.iops-write", 0);
426 
427         throttle_cfg->buckets[THROTTLE_BPS_TOTAL].max =
428             qemu_opt_get_number(opts, "throttling.bps-total-max", 0);
429         throttle_cfg->buckets[THROTTLE_BPS_READ].max  =
430             qemu_opt_get_number(opts, "throttling.bps-read-max", 0);
431         throttle_cfg->buckets[THROTTLE_BPS_WRITE].max =
432             qemu_opt_get_number(opts, "throttling.bps-write-max", 0);
433         throttle_cfg->buckets[THROTTLE_OPS_TOTAL].max =
434             qemu_opt_get_number(opts, "throttling.iops-total-max", 0);
435         throttle_cfg->buckets[THROTTLE_OPS_READ].max =
436             qemu_opt_get_number(opts, "throttling.iops-read-max", 0);
437         throttle_cfg->buckets[THROTTLE_OPS_WRITE].max =
438             qemu_opt_get_number(opts, "throttling.iops-write-max", 0);
439 
440         throttle_cfg->op_size =
441             qemu_opt_get_number(opts, "throttling.iops-size", 0);
442 
443         if (!check_throttle_config(throttle_cfg, errp)) {
444             return;
445         }
446     }
447 
448     if (detect_zeroes) {
449         *detect_zeroes =
450             qapi_enum_parse(BlockdevDetectZeroesOptions_lookup,
451                             qemu_opt_get(opts, "detect-zeroes"),
452                             BLOCKDEV_DETECT_ZEROES_OPTIONS__MAX,
453                             BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF,
454                             &local_error);
455         if (local_error) {
456             error_propagate(errp, local_error);
457             return;
458         }
459 
460         if (bdrv_flags &&
461             *detect_zeroes == BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP &&
462             !(*bdrv_flags & BDRV_O_UNMAP))
463         {
464             error_setg(errp, "setting detect-zeroes to unmap is not allowed "
465                              "without setting discard operation to unmap");
466             return;
467         }
468     }
469 }
470 
471 /* Takes the ownership of bs_opts */
472 static BlockBackend *blockdev_init(const char *file, QDict *bs_opts,
473                                    Error **errp)
474 {
475     const char *buf;
476     int bdrv_flags = 0;
477     int on_read_error, on_write_error;
478     bool account_invalid, account_failed;
479     BlockBackend *blk;
480     BlockDriverState *bs;
481     ThrottleConfig cfg;
482     int snapshot = 0;
483     Error *error = NULL;
484     QemuOpts *opts;
485     QDict *interval_dict = NULL;
486     QList *interval_list = NULL;
487     const char *id;
488     BlockdevDetectZeroesOptions detect_zeroes =
489         BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF;
490     const char *throttling_group = NULL;
491 
492     /* Check common options by copying from bs_opts to opts, all other options
493      * stay in bs_opts for processing by bdrv_open(). */
494     id = qdict_get_try_str(bs_opts, "id");
495     opts = qemu_opts_create(&qemu_common_drive_opts, id, 1, &error);
496     if (error) {
497         error_propagate(errp, error);
498         goto err_no_opts;
499     }
500 
501     qemu_opts_absorb_qdict(opts, bs_opts, &error);
502     if (error) {
503         error_propagate(errp, error);
504         goto early_err;
505     }
506 
507     if (id) {
508         qdict_del(bs_opts, "id");
509     }
510 
511     /* extract parameters */
512     snapshot = qemu_opt_get_bool(opts, "snapshot", 0);
513 
514     account_invalid = qemu_opt_get_bool(opts, "stats-account-invalid", true);
515     account_failed = qemu_opt_get_bool(opts, "stats-account-failed", true);
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             error_printf("Supported formats:");
536             bdrv_iterate_format(bdrv_format_print, NULL);
537             error_printf("\n");
538             goto early_err;
539         }
540 
541         if (qdict_haskey(bs_opts, "driver")) {
542             error_setg(errp, "Cannot specify both 'driver' and 'format'");
543             goto early_err;
544         }
545         qdict_put(bs_opts, "driver", qstring_from_str(buf));
546     }
547 
548     on_write_error = BLOCKDEV_ON_ERROR_ENOSPC;
549     if ((buf = qemu_opt_get(opts, "werror")) != NULL) {
550         on_write_error = parse_block_error_action(buf, 0, &error);
551         if (error) {
552             error_propagate(errp, error);
553             goto early_err;
554         }
555     }
556 
557     on_read_error = BLOCKDEV_ON_ERROR_REPORT;
558     if ((buf = qemu_opt_get(opts, "rerror")) != NULL) {
559         on_read_error = parse_block_error_action(buf, 1, &error);
560         if (error) {
561             error_propagate(errp, error);
562             goto early_err;
563         }
564     }
565 
566     if (snapshot) {
567         bdrv_flags |= BDRV_O_SNAPSHOT;
568     }
569 
570     /* init */
571     if ((!file || !*file) && !qdict_size(bs_opts)) {
572         BlockBackendRootState *blk_rs;
573 
574         blk = blk_new(qemu_opts_id(opts), errp);
575         if (!blk) {
576             goto early_err;
577         }
578 
579         blk_rs = blk_get_root_state(blk);
580         blk_rs->open_flags    = bdrv_flags;
581         blk_rs->read_only     = !(bdrv_flags & BDRV_O_RDWR);
582         blk_rs->detect_zeroes = detect_zeroes;
583 
584         if (throttle_enabled(&cfg)) {
585             if (!throttling_group) {
586                 throttling_group = blk_name(blk);
587             }
588             blk_rs->throttle_group = g_strdup(throttling_group);
589             blk_rs->throttle_state = throttle_group_incref(throttling_group);
590             blk_rs->throttle_state->cfg = cfg;
591         }
592 
593         QDECREF(bs_opts);
594     } else {
595         if (file && !*file) {
596             file = NULL;
597         }
598 
599         /* bdrv_open() defaults to the values in bdrv_flags (for compatibility
600          * with other callers) rather than what we want as the real defaults.
601          * Apply the defaults here instead. */
602         qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_WB, "on");
603         qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_DIRECT, "off");
604         qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_NO_FLUSH, "off");
605 
606         if (snapshot) {
607             /* always use cache=unsafe with snapshot */
608             qdict_put(bs_opts, BDRV_OPT_CACHE_WB, qstring_from_str("on"));
609             qdict_put(bs_opts, BDRV_OPT_CACHE_DIRECT, qstring_from_str("off"));
610             qdict_put(bs_opts, BDRV_OPT_CACHE_NO_FLUSH, qstring_from_str("on"));
611         }
612 
613         blk = blk_new_open(qemu_opts_id(opts), file, NULL, bs_opts, bdrv_flags,
614                            errp);
615         if (!blk) {
616             goto err_no_bs_opts;
617         }
618         bs = blk_bs(blk);
619 
620         bs->detect_zeroes = detect_zeroes;
621 
622         /* disk I/O throttling */
623         if (throttle_enabled(&cfg)) {
624             if (!throttling_group) {
625                 throttling_group = blk_name(blk);
626             }
627             bdrv_io_limits_enable(bs, throttling_group);
628             bdrv_set_io_limits(bs, &cfg);
629         }
630 
631         if (bdrv_key_required(bs)) {
632             autostart = 0;
633         }
634 
635         block_acct_init(blk_get_stats(blk), account_invalid, account_failed);
636 
637         if (!parse_stats_intervals(blk_get_stats(blk), interval_list, errp)) {
638             blk_unref(blk);
639             blk = NULL;
640             goto err_no_bs_opts;
641         }
642     }
643 
644     blk_set_on_error(blk, on_read_error, on_write_error);
645 
646 err_no_bs_opts:
647     qemu_opts_del(opts);
648     QDECREF(interval_dict);
649     QDECREF(interval_list);
650     return blk;
651 
652 early_err:
653     qemu_opts_del(opts);
654     QDECREF(interval_dict);
655     QDECREF(interval_list);
656 err_no_opts:
657     QDECREF(bs_opts);
658     return NULL;
659 }
660 
661 static QemuOptsList qemu_root_bds_opts;
662 
663 /* Takes the ownership of bs_opts */
664 static BlockDriverState *bds_tree_init(QDict *bs_opts, Error **errp)
665 {
666     BlockDriverState *bs;
667     QemuOpts *opts;
668     Error *local_error = NULL;
669     BlockdevDetectZeroesOptions detect_zeroes;
670     int ret;
671     int bdrv_flags = 0;
672 
673     opts = qemu_opts_create(&qemu_root_bds_opts, NULL, 1, errp);
674     if (!opts) {
675         goto fail;
676     }
677 
678     qemu_opts_absorb_qdict(opts, bs_opts, &local_error);
679     if (local_error) {
680         error_propagate(errp, local_error);
681         goto fail;
682     }
683 
684     extract_common_blockdev_options(opts, &bdrv_flags, NULL, NULL,
685                                     &detect_zeroes, &local_error);
686     if (local_error) {
687         error_propagate(errp, local_error);
688         goto fail;
689     }
690 
691     bs = NULL;
692     ret = bdrv_open(&bs, NULL, NULL, bs_opts, bdrv_flags, errp);
693     if (ret < 0) {
694         goto fail_no_bs_opts;
695     }
696 
697     bs->detect_zeroes = detect_zeroes;
698 
699 fail_no_bs_opts:
700     qemu_opts_del(opts);
701     return bs;
702 
703 fail:
704     qemu_opts_del(opts);
705     QDECREF(bs_opts);
706     return NULL;
707 }
708 
709 void blockdev_close_all_bdrv_states(void)
710 {
711     BlockDriverState *bs, *next_bs;
712 
713     QTAILQ_FOREACH_SAFE(bs, &monitor_bdrv_states, monitor_list, next_bs) {
714         AioContext *ctx = bdrv_get_aio_context(bs);
715 
716         aio_context_acquire(ctx);
717         bdrv_unref(bs);
718         aio_context_release(ctx);
719     }
720 }
721 
722 static void qemu_opt_rename(QemuOpts *opts, const char *from, const char *to,
723                             Error **errp)
724 {
725     const char *value;
726 
727     value = qemu_opt_get(opts, from);
728     if (value) {
729         if (qemu_opt_find(opts, to)) {
730             error_setg(errp, "'%s' and its alias '%s' can't be used at the "
731                        "same time", to, from);
732             return;
733         }
734     }
735 
736     /* rename all items in opts */
737     while ((value = qemu_opt_get(opts, from))) {
738         qemu_opt_set(opts, to, value, &error_abort);
739         qemu_opt_unset(opts, from);
740     }
741 }
742 
743 QemuOptsList qemu_legacy_drive_opts = {
744     .name = "drive",
745     .head = QTAILQ_HEAD_INITIALIZER(qemu_legacy_drive_opts.head),
746     .desc = {
747         {
748             .name = "bus",
749             .type = QEMU_OPT_NUMBER,
750             .help = "bus number",
751         },{
752             .name = "unit",
753             .type = QEMU_OPT_NUMBER,
754             .help = "unit number (i.e. lun for scsi)",
755         },{
756             .name = "index",
757             .type = QEMU_OPT_NUMBER,
758             .help = "index number",
759         },{
760             .name = "media",
761             .type = QEMU_OPT_STRING,
762             .help = "media type (disk, cdrom)",
763         },{
764             .name = "if",
765             .type = QEMU_OPT_STRING,
766             .help = "interface (ide, scsi, sd, mtd, floppy, pflash, virtio)",
767         },{
768             .name = "cyls",
769             .type = QEMU_OPT_NUMBER,
770             .help = "number of cylinders (ide disk geometry)",
771         },{
772             .name = "heads",
773             .type = QEMU_OPT_NUMBER,
774             .help = "number of heads (ide disk geometry)",
775         },{
776             .name = "secs",
777             .type = QEMU_OPT_NUMBER,
778             .help = "number of sectors (ide disk geometry)",
779         },{
780             .name = "trans",
781             .type = QEMU_OPT_STRING,
782             .help = "chs translation (auto, lba, none)",
783         },{
784             .name = "boot",
785             .type = QEMU_OPT_BOOL,
786             .help = "(deprecated, ignored)",
787         },{
788             .name = "addr",
789             .type = QEMU_OPT_STRING,
790             .help = "pci address (virtio only)",
791         },{
792             .name = "serial",
793             .type = QEMU_OPT_STRING,
794             .help = "disk serial number",
795         },{
796             .name = "file",
797             .type = QEMU_OPT_STRING,
798             .help = "file name",
799         },
800 
801         /* Options that are passed on, but have special semantics with -drive */
802         {
803             .name = "read-only",
804             .type = QEMU_OPT_BOOL,
805             .help = "open drive file as read-only",
806         },{
807             .name = "rerror",
808             .type = QEMU_OPT_STRING,
809             .help = "read error action",
810         },{
811             .name = "werror",
812             .type = QEMU_OPT_STRING,
813             .help = "write error action",
814         },{
815             .name = "copy-on-read",
816             .type = QEMU_OPT_BOOL,
817             .help = "copy read data from backing file into image file",
818         },
819 
820         { /* end of list */ }
821     },
822 };
823 
824 DriveInfo *drive_new(QemuOpts *all_opts, BlockInterfaceType block_default_type)
825 {
826     const char *value;
827     BlockBackend *blk;
828     DriveInfo *dinfo = NULL;
829     QDict *bs_opts;
830     QemuOpts *legacy_opts;
831     DriveMediaType media = MEDIA_DISK;
832     BlockInterfaceType type;
833     int cyls, heads, secs, translation;
834     int max_devs, bus_id, unit_id, index;
835     const char *devaddr;
836     const char *werror, *rerror;
837     bool read_only = false;
838     bool copy_on_read;
839     const char *serial;
840     const char *filename;
841     Error *local_err = NULL;
842     int i;
843 
844     /* Change legacy command line options into QMP ones */
845     static const struct {
846         const char *from;
847         const char *to;
848     } opt_renames[] = {
849         { "iops",           "throttling.iops-total" },
850         { "iops_rd",        "throttling.iops-read" },
851         { "iops_wr",        "throttling.iops-write" },
852 
853         { "bps",            "throttling.bps-total" },
854         { "bps_rd",         "throttling.bps-read" },
855         { "bps_wr",         "throttling.bps-write" },
856 
857         { "iops_max",       "throttling.iops-total-max" },
858         { "iops_rd_max",    "throttling.iops-read-max" },
859         { "iops_wr_max",    "throttling.iops-write-max" },
860 
861         { "bps_max",        "throttling.bps-total-max" },
862         { "bps_rd_max",     "throttling.bps-read-max" },
863         { "bps_wr_max",     "throttling.bps-write-max" },
864 
865         { "iops_size",      "throttling.iops-size" },
866 
867         { "group",          "throttling.group" },
868 
869         { "readonly",       "read-only" },
870     };
871 
872     for (i = 0; i < ARRAY_SIZE(opt_renames); i++) {
873         qemu_opt_rename(all_opts, opt_renames[i].from, opt_renames[i].to,
874                         &local_err);
875         if (local_err) {
876             error_report_err(local_err);
877             return NULL;
878         }
879     }
880 
881     value = qemu_opt_get(all_opts, "cache");
882     if (value) {
883         int flags = 0;
884 
885         if (bdrv_parse_cache_flags(value, &flags) != 0) {
886             error_report("invalid cache option");
887             return NULL;
888         }
889 
890         /* Specific options take precedence */
891         if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_WB)) {
892             qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_WB,
893                               !!(flags & BDRV_O_CACHE_WB), &error_abort);
894         }
895         if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_DIRECT)) {
896             qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_DIRECT,
897                               !!(flags & BDRV_O_NOCACHE), &error_abort);
898         }
899         if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_NO_FLUSH)) {
900             qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_NO_FLUSH,
901                               !!(flags & BDRV_O_NO_FLUSH), &error_abort);
902         }
903         qemu_opt_unset(all_opts, "cache");
904     }
905 
906     /* Get a QDict for processing the options */
907     bs_opts = qdict_new();
908     qemu_opts_to_qdict(all_opts, bs_opts);
909 
910     legacy_opts = qemu_opts_create(&qemu_legacy_drive_opts, NULL, 0,
911                                    &error_abort);
912     qemu_opts_absorb_qdict(legacy_opts, bs_opts, &local_err);
913     if (local_err) {
914         error_report_err(local_err);
915         goto fail;
916     }
917 
918     /* Deprecated option boot=[on|off] */
919     if (qemu_opt_get(legacy_opts, "boot") != NULL) {
920         fprintf(stderr, "qemu-kvm: boot=on|off is deprecated and will be "
921                 "ignored. Future versions will reject this parameter. Please "
922                 "update your scripts.\n");
923     }
924 
925     /* Media type */
926     value = qemu_opt_get(legacy_opts, "media");
927     if (value) {
928         if (!strcmp(value, "disk")) {
929             media = MEDIA_DISK;
930         } else if (!strcmp(value, "cdrom")) {
931             media = MEDIA_CDROM;
932             read_only = true;
933         } else {
934             error_report("'%s' invalid media", value);
935             goto fail;
936         }
937     }
938 
939     /* copy-on-read is disabled with a warning for read-only devices */
940     read_only |= qemu_opt_get_bool(legacy_opts, "read-only", false);
941     copy_on_read = qemu_opt_get_bool(legacy_opts, "copy-on-read", false);
942 
943     if (read_only && copy_on_read) {
944         error_report("warning: disabling copy-on-read on read-only drive");
945         copy_on_read = false;
946     }
947 
948     qdict_put(bs_opts, "read-only",
949               qstring_from_str(read_only ? "on" : "off"));
950     qdict_put(bs_opts, "copy-on-read",
951               qstring_from_str(copy_on_read ? "on" :"off"));
952 
953     /* Controller type */
954     value = qemu_opt_get(legacy_opts, "if");
955     if (value) {
956         for (type = 0;
957              type < IF_COUNT && strcmp(value, if_name[type]);
958              type++) {
959         }
960         if (type == IF_COUNT) {
961             error_report("unsupported bus type '%s'", value);
962             goto fail;
963         }
964     } else {
965         type = block_default_type;
966     }
967 
968     /* Geometry */
969     cyls  = qemu_opt_get_number(legacy_opts, "cyls", 0);
970     heads = qemu_opt_get_number(legacy_opts, "heads", 0);
971     secs  = qemu_opt_get_number(legacy_opts, "secs", 0);
972 
973     if (cyls || heads || secs) {
974         if (cyls < 1) {
975             error_report("invalid physical cyls number");
976             goto fail;
977         }
978         if (heads < 1) {
979             error_report("invalid physical heads number");
980             goto fail;
981         }
982         if (secs < 1) {
983             error_report("invalid physical secs number");
984             goto fail;
985         }
986     }
987 
988     translation = BIOS_ATA_TRANSLATION_AUTO;
989     value = qemu_opt_get(legacy_opts, "trans");
990     if (value != NULL) {
991         if (!cyls) {
992             error_report("'%s' trans must be used with cyls, heads and secs",
993                          value);
994             goto fail;
995         }
996         if (!strcmp(value, "none")) {
997             translation = BIOS_ATA_TRANSLATION_NONE;
998         } else if (!strcmp(value, "lba")) {
999             translation = BIOS_ATA_TRANSLATION_LBA;
1000         } else if (!strcmp(value, "large")) {
1001             translation = BIOS_ATA_TRANSLATION_LARGE;
1002         } else if (!strcmp(value, "rechs")) {
1003             translation = BIOS_ATA_TRANSLATION_RECHS;
1004         } else if (!strcmp(value, "auto")) {
1005             translation = BIOS_ATA_TRANSLATION_AUTO;
1006         } else {
1007             error_report("'%s' invalid translation type", value);
1008             goto fail;
1009         }
1010     }
1011 
1012     if (media == MEDIA_CDROM) {
1013         if (cyls || secs || heads) {
1014             error_report("CHS can't be set with media=cdrom");
1015             goto fail;
1016         }
1017     }
1018 
1019     /* Device address specified by bus/unit or index.
1020      * If none was specified, try to find the first free one. */
1021     bus_id  = qemu_opt_get_number(legacy_opts, "bus", 0);
1022     unit_id = qemu_opt_get_number(legacy_opts, "unit", -1);
1023     index   = qemu_opt_get_number(legacy_opts, "index", -1);
1024 
1025     max_devs = if_max_devs[type];
1026 
1027     if (index != -1) {
1028         if (bus_id != 0 || unit_id != -1) {
1029             error_report("index cannot be used with bus and unit");
1030             goto fail;
1031         }
1032         bus_id = drive_index_to_bus_id(type, index);
1033         unit_id = drive_index_to_unit_id(type, index);
1034     }
1035 
1036     if (unit_id == -1) {
1037        unit_id = 0;
1038        while (drive_get(type, bus_id, unit_id) != NULL) {
1039            unit_id++;
1040            if (max_devs && unit_id >= max_devs) {
1041                unit_id -= max_devs;
1042                bus_id++;
1043            }
1044        }
1045     }
1046 
1047     if (max_devs && unit_id >= max_devs) {
1048         error_report("unit %d too big (max is %d)", unit_id, max_devs - 1);
1049         goto fail;
1050     }
1051 
1052     if (drive_get(type, bus_id, unit_id) != NULL) {
1053         error_report("drive with bus=%d, unit=%d (index=%d) exists",
1054                      bus_id, unit_id, index);
1055         goto fail;
1056     }
1057 
1058     /* Serial number */
1059     serial = qemu_opt_get(legacy_opts, "serial");
1060 
1061     /* no id supplied -> create one */
1062     if (qemu_opts_id(all_opts) == NULL) {
1063         char *new_id;
1064         const char *mediastr = "";
1065         if (type == IF_IDE || type == IF_SCSI) {
1066             mediastr = (media == MEDIA_CDROM) ? "-cd" : "-hd";
1067         }
1068         if (max_devs) {
1069             new_id = g_strdup_printf("%s%i%s%i", if_name[type], bus_id,
1070                                      mediastr, unit_id);
1071         } else {
1072             new_id = g_strdup_printf("%s%s%i", if_name[type],
1073                                      mediastr, unit_id);
1074         }
1075         qdict_put(bs_opts, "id", qstring_from_str(new_id));
1076         g_free(new_id);
1077     }
1078 
1079     /* Add virtio block device */
1080     devaddr = qemu_opt_get(legacy_opts, "addr");
1081     if (devaddr && type != IF_VIRTIO) {
1082         error_report("addr is not supported by this bus type");
1083         goto fail;
1084     }
1085 
1086     if (type == IF_VIRTIO) {
1087         QemuOpts *devopts;
1088         devopts = qemu_opts_create(qemu_find_opts("device"), NULL, 0,
1089                                    &error_abort);
1090         if (arch_type == QEMU_ARCH_S390X) {
1091             qemu_opt_set(devopts, "driver", "virtio-blk-ccw", &error_abort);
1092         } else {
1093             qemu_opt_set(devopts, "driver", "virtio-blk-pci", &error_abort);
1094         }
1095         qemu_opt_set(devopts, "drive", qdict_get_str(bs_opts, "id"),
1096                      &error_abort);
1097         if (devaddr) {
1098             qemu_opt_set(devopts, "addr", devaddr, &error_abort);
1099         }
1100     }
1101 
1102     filename = qemu_opt_get(legacy_opts, "file");
1103 
1104     /* Check werror/rerror compatibility with if=... */
1105     werror = qemu_opt_get(legacy_opts, "werror");
1106     if (werror != NULL) {
1107         if (type != IF_IDE && type != IF_SCSI && type != IF_VIRTIO &&
1108             type != IF_NONE) {
1109             error_report("werror is not supported by this bus type");
1110             goto fail;
1111         }
1112         qdict_put(bs_opts, "werror", qstring_from_str(werror));
1113     }
1114 
1115     rerror = qemu_opt_get(legacy_opts, "rerror");
1116     if (rerror != NULL) {
1117         if (type != IF_IDE && type != IF_VIRTIO && type != IF_SCSI &&
1118             type != IF_NONE) {
1119             error_report("rerror is not supported by this bus type");
1120             goto fail;
1121         }
1122         qdict_put(bs_opts, "rerror", qstring_from_str(rerror));
1123     }
1124 
1125     /* Actual block device init: Functionality shared with blockdev-add */
1126     blk = blockdev_init(filename, bs_opts, &local_err);
1127     bs_opts = NULL;
1128     if (!blk) {
1129         if (local_err) {
1130             error_report_err(local_err);
1131         }
1132         goto fail;
1133     } else {
1134         assert(!local_err);
1135     }
1136 
1137     /* Create legacy DriveInfo */
1138     dinfo = g_malloc0(sizeof(*dinfo));
1139     dinfo->opts = all_opts;
1140 
1141     dinfo->cyls = cyls;
1142     dinfo->heads = heads;
1143     dinfo->secs = secs;
1144     dinfo->trans = translation;
1145 
1146     dinfo->type = type;
1147     dinfo->bus = bus_id;
1148     dinfo->unit = unit_id;
1149     dinfo->devaddr = devaddr;
1150     dinfo->serial = g_strdup(serial);
1151 
1152     blk_set_legacy_dinfo(blk, dinfo);
1153 
1154     switch(type) {
1155     case IF_IDE:
1156     case IF_SCSI:
1157     case IF_XEN:
1158     case IF_NONE:
1159         dinfo->media_cd = media == MEDIA_CDROM;
1160         break;
1161     default:
1162         break;
1163     }
1164 
1165 fail:
1166     qemu_opts_del(legacy_opts);
1167     QDECREF(bs_opts);
1168     return dinfo;
1169 }
1170 
1171 void hmp_commit(Monitor *mon, const QDict *qdict)
1172 {
1173     const char *device = qdict_get_str(qdict, "device");
1174     BlockBackend *blk;
1175     int ret;
1176 
1177     if (!strcmp(device, "all")) {
1178         ret = bdrv_commit_all();
1179     } else {
1180         BlockDriverState *bs;
1181         AioContext *aio_context;
1182 
1183         blk = blk_by_name(device);
1184         if (!blk) {
1185             monitor_printf(mon, "Device '%s' not found\n", device);
1186             return;
1187         }
1188         if (!blk_is_available(blk)) {
1189             monitor_printf(mon, "Device '%s' has no medium\n", device);
1190             return;
1191         }
1192 
1193         bs = blk_bs(blk);
1194         aio_context = bdrv_get_aio_context(bs);
1195         aio_context_acquire(aio_context);
1196 
1197         ret = bdrv_commit(bs);
1198 
1199         aio_context_release(aio_context);
1200     }
1201     if (ret < 0) {
1202         monitor_printf(mon, "'commit' error for '%s': %s\n", device,
1203                        strerror(-ret));
1204     }
1205 }
1206 
1207 static void blockdev_do_action(TransactionActionKind type, void *data,
1208                                Error **errp)
1209 {
1210     TransactionAction action;
1211     TransactionActionList list;
1212 
1213     action.type = type;
1214     action.u.data = data;
1215     list.value = &action;
1216     list.next = NULL;
1217     qmp_transaction(&list, false, NULL, errp);
1218 }
1219 
1220 void qmp_blockdev_snapshot_sync(bool has_device, const char *device,
1221                                 bool has_node_name, const char *node_name,
1222                                 const char *snapshot_file,
1223                                 bool has_snapshot_node_name,
1224                                 const char *snapshot_node_name,
1225                                 bool has_format, const char *format,
1226                                 bool has_mode, NewImageMode mode, Error **errp)
1227 {
1228     BlockdevSnapshotSync snapshot = {
1229         .has_device = has_device,
1230         .device = (char *) device,
1231         .has_node_name = has_node_name,
1232         .node_name = (char *) node_name,
1233         .snapshot_file = (char *) snapshot_file,
1234         .has_snapshot_node_name = has_snapshot_node_name,
1235         .snapshot_node_name = (char *) snapshot_node_name,
1236         .has_format = has_format,
1237         .format = (char *) format,
1238         .has_mode = has_mode,
1239         .mode = mode,
1240     };
1241     blockdev_do_action(TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC,
1242                        &snapshot, errp);
1243 }
1244 
1245 void qmp_blockdev_snapshot(const char *node, const char *overlay,
1246                            Error **errp)
1247 {
1248     BlockdevSnapshot snapshot_data = {
1249         .node = (char *) node,
1250         .overlay = (char *) overlay
1251     };
1252 
1253     blockdev_do_action(TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT,
1254                        &snapshot_data, errp);
1255 }
1256 
1257 void qmp_blockdev_snapshot_internal_sync(const char *device,
1258                                          const char *name,
1259                                          Error **errp)
1260 {
1261     BlockdevSnapshotInternal snapshot = {
1262         .device = (char *) device,
1263         .name = (char *) name
1264     };
1265 
1266     blockdev_do_action(TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC,
1267                        &snapshot, errp);
1268 }
1269 
1270 SnapshotInfo *qmp_blockdev_snapshot_delete_internal_sync(const char *device,
1271                                                          bool has_id,
1272                                                          const char *id,
1273                                                          bool has_name,
1274                                                          const char *name,
1275                                                          Error **errp)
1276 {
1277     BlockDriverState *bs;
1278     BlockBackend *blk;
1279     AioContext *aio_context;
1280     QEMUSnapshotInfo sn;
1281     Error *local_err = NULL;
1282     SnapshotInfo *info = NULL;
1283     int ret;
1284 
1285     blk = blk_by_name(device);
1286     if (!blk) {
1287         error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
1288                   "Device '%s' not found", device);
1289         return NULL;
1290     }
1291 
1292     aio_context = blk_get_aio_context(blk);
1293     aio_context_acquire(aio_context);
1294 
1295     if (!has_id) {
1296         id = NULL;
1297     }
1298 
1299     if (!has_name) {
1300         name = NULL;
1301     }
1302 
1303     if (!id && !name) {
1304         error_setg(errp, "Name or id must be provided");
1305         goto out_aio_context;
1306     }
1307 
1308     if (!blk_is_available(blk)) {
1309         error_setg(errp, "Device '%s' has no medium", device);
1310         goto out_aio_context;
1311     }
1312     bs = blk_bs(blk);
1313 
1314     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_INTERNAL_SNAPSHOT_DELETE, errp)) {
1315         goto out_aio_context;
1316     }
1317 
1318     ret = bdrv_snapshot_find_by_id_and_name(bs, id, name, &sn, &local_err);
1319     if (local_err) {
1320         error_propagate(errp, local_err);
1321         goto out_aio_context;
1322     }
1323     if (!ret) {
1324         error_setg(errp,
1325                    "Snapshot with id '%s' and name '%s' does not exist on "
1326                    "device '%s'",
1327                    STR_OR_NULL(id), STR_OR_NULL(name), device);
1328         goto out_aio_context;
1329     }
1330 
1331     bdrv_snapshot_delete(bs, id, name, &local_err);
1332     if (local_err) {
1333         error_propagate(errp, local_err);
1334         goto out_aio_context;
1335     }
1336 
1337     aio_context_release(aio_context);
1338 
1339     info = g_new0(SnapshotInfo, 1);
1340     info->id = g_strdup(sn.id_str);
1341     info->name = g_strdup(sn.name);
1342     info->date_nsec = sn.date_nsec;
1343     info->date_sec = sn.date_sec;
1344     info->vm_state_size = sn.vm_state_size;
1345     info->vm_clock_nsec = sn.vm_clock_nsec % 1000000000;
1346     info->vm_clock_sec = sn.vm_clock_nsec / 1000000000;
1347 
1348     return info;
1349 
1350 out_aio_context:
1351     aio_context_release(aio_context);
1352     return NULL;
1353 }
1354 
1355 /**
1356  * block_dirty_bitmap_lookup:
1357  * Return a dirty bitmap (if present), after validating
1358  * the node reference and bitmap names.
1359  *
1360  * @node: The name of the BDS node to search for bitmaps
1361  * @name: The name of the bitmap to search for
1362  * @pbs: Output pointer for BDS lookup, if desired. Can be NULL.
1363  * @paio: Output pointer for aio_context acquisition, if desired. Can be NULL.
1364  * @errp: Output pointer for error information. Can be NULL.
1365  *
1366  * @return: A bitmap object on success, or NULL on failure.
1367  */
1368 static BdrvDirtyBitmap *block_dirty_bitmap_lookup(const char *node,
1369                                                   const char *name,
1370                                                   BlockDriverState **pbs,
1371                                                   AioContext **paio,
1372                                                   Error **errp)
1373 {
1374     BlockDriverState *bs;
1375     BdrvDirtyBitmap *bitmap;
1376     AioContext *aio_context;
1377 
1378     if (!node) {
1379         error_setg(errp, "Node cannot be NULL");
1380         return NULL;
1381     }
1382     if (!name) {
1383         error_setg(errp, "Bitmap name cannot be NULL");
1384         return NULL;
1385     }
1386     bs = bdrv_lookup_bs(node, node, NULL);
1387     if (!bs) {
1388         error_setg(errp, "Node '%s' not found", node);
1389         return NULL;
1390     }
1391 
1392     aio_context = bdrv_get_aio_context(bs);
1393     aio_context_acquire(aio_context);
1394 
1395     bitmap = bdrv_find_dirty_bitmap(bs, name);
1396     if (!bitmap) {
1397         error_setg(errp, "Dirty bitmap '%s' not found", name);
1398         goto fail;
1399     }
1400 
1401     if (pbs) {
1402         *pbs = bs;
1403     }
1404     if (paio) {
1405         *paio = aio_context;
1406     } else {
1407         aio_context_release(aio_context);
1408     }
1409 
1410     return bitmap;
1411 
1412  fail:
1413     aio_context_release(aio_context);
1414     return NULL;
1415 }
1416 
1417 /* New and old BlockDriverState structs for atomic group operations */
1418 
1419 typedef struct BlkActionState BlkActionState;
1420 
1421 /**
1422  * BlkActionOps:
1423  * Table of operations that define an Action.
1424  *
1425  * @instance_size: Size of state struct, in bytes.
1426  * @prepare: Prepare the work, must NOT be NULL.
1427  * @commit: Commit the changes, can be NULL.
1428  * @abort: Abort the changes on fail, can be NULL.
1429  * @clean: Clean up resources after all transaction actions have called
1430  *         commit() or abort(). Can be NULL.
1431  *
1432  * Only prepare() may fail. In a single transaction, only one of commit() or
1433  * abort() will be called. clean() will always be called if it is present.
1434  */
1435 typedef struct BlkActionOps {
1436     size_t instance_size;
1437     void (*prepare)(BlkActionState *common, Error **errp);
1438     void (*commit)(BlkActionState *common);
1439     void (*abort)(BlkActionState *common);
1440     void (*clean)(BlkActionState *common);
1441 } BlkActionOps;
1442 
1443 /**
1444  * BlkActionState:
1445  * Describes one Action's state within a Transaction.
1446  *
1447  * @action: QAPI-defined enum identifying which Action to perform.
1448  * @ops: Table of ActionOps this Action can perform.
1449  * @block_job_txn: Transaction which this action belongs to.
1450  * @entry: List membership for all Actions in this Transaction.
1451  *
1452  * This structure must be arranged as first member in a subclassed type,
1453  * assuming that the compiler will also arrange it to the same offsets as the
1454  * base class.
1455  */
1456 struct BlkActionState {
1457     TransactionAction *action;
1458     const BlkActionOps *ops;
1459     BlockJobTxn *block_job_txn;
1460     TransactionProperties *txn_props;
1461     QSIMPLEQ_ENTRY(BlkActionState) entry;
1462 };
1463 
1464 /* internal snapshot private data */
1465 typedef struct InternalSnapshotState {
1466     BlkActionState common;
1467     BlockDriverState *bs;
1468     AioContext *aio_context;
1469     QEMUSnapshotInfo sn;
1470     bool created;
1471 } InternalSnapshotState;
1472 
1473 
1474 static int action_check_completion_mode(BlkActionState *s, Error **errp)
1475 {
1476     if (s->txn_props->completion_mode != ACTION_COMPLETION_MODE_INDIVIDUAL) {
1477         error_setg(errp,
1478                    "Action '%s' does not support Transaction property "
1479                    "completion-mode = %s",
1480                    TransactionActionKind_lookup[s->action->type],
1481                    ActionCompletionMode_lookup[s->txn_props->completion_mode]);
1482         return -1;
1483     }
1484     return 0;
1485 }
1486 
1487 static void internal_snapshot_prepare(BlkActionState *common,
1488                                       Error **errp)
1489 {
1490     Error *local_err = NULL;
1491     const char *device;
1492     const char *name;
1493     BlockBackend *blk;
1494     BlockDriverState *bs;
1495     QEMUSnapshotInfo old_sn, *sn;
1496     bool ret;
1497     qemu_timeval tv;
1498     BlockdevSnapshotInternal *internal;
1499     InternalSnapshotState *state;
1500     int ret1;
1501 
1502     g_assert(common->action->type ==
1503              TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC);
1504     internal = common->action->u.blockdev_snapshot_internal_sync;
1505     state = DO_UPCAST(InternalSnapshotState, common, common);
1506 
1507     /* 1. parse input */
1508     device = internal->device;
1509     name = internal->name;
1510 
1511     /* 2. check for validation */
1512     if (action_check_completion_mode(common, errp) < 0) {
1513         return;
1514     }
1515 
1516     blk = blk_by_name(device);
1517     if (!blk) {
1518         error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
1519                   "Device '%s' not found", device);
1520         return;
1521     }
1522 
1523     /* AioContext is released in .clean() */
1524     state->aio_context = blk_get_aio_context(blk);
1525     aio_context_acquire(state->aio_context);
1526 
1527     if (!blk_is_available(blk)) {
1528         error_setg(errp, QERR_DEVICE_HAS_NO_MEDIUM, device);
1529         return;
1530     }
1531     bs = blk_bs(blk);
1532 
1533     state->bs = bs;
1534     bdrv_drained_begin(bs);
1535 
1536     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_INTERNAL_SNAPSHOT, errp)) {
1537         return;
1538     }
1539 
1540     if (bdrv_is_read_only(bs)) {
1541         error_setg(errp, "Device '%s' is read only", device);
1542         return;
1543     }
1544 
1545     if (!bdrv_can_snapshot(bs)) {
1546         error_setg(errp, "Block format '%s' used by device '%s' "
1547                    "does not support internal snapshots",
1548                    bs->drv->format_name, device);
1549         return;
1550     }
1551 
1552     if (!strlen(name)) {
1553         error_setg(errp, "Name is empty");
1554         return;
1555     }
1556 
1557     /* check whether a snapshot with name exist */
1558     ret = bdrv_snapshot_find_by_id_and_name(bs, NULL, name, &old_sn,
1559                                             &local_err);
1560     if (local_err) {
1561         error_propagate(errp, local_err);
1562         return;
1563     } else if (ret) {
1564         error_setg(errp,
1565                    "Snapshot with name '%s' already exists on device '%s'",
1566                    name, device);
1567         return;
1568     }
1569 
1570     /* 3. take the snapshot */
1571     sn = &state->sn;
1572     pstrcpy(sn->name, sizeof(sn->name), name);
1573     qemu_gettimeofday(&tv);
1574     sn->date_sec = tv.tv_sec;
1575     sn->date_nsec = tv.tv_usec * 1000;
1576     sn->vm_clock_nsec = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL);
1577 
1578     ret1 = bdrv_snapshot_create(bs, sn);
1579     if (ret1 < 0) {
1580         error_setg_errno(errp, -ret1,
1581                          "Failed to create snapshot '%s' on device '%s'",
1582                          name, device);
1583         return;
1584     }
1585 
1586     /* 4. succeed, mark a snapshot is created */
1587     state->created = true;
1588 }
1589 
1590 static void internal_snapshot_abort(BlkActionState *common)
1591 {
1592     InternalSnapshotState *state =
1593                              DO_UPCAST(InternalSnapshotState, common, common);
1594     BlockDriverState *bs = state->bs;
1595     QEMUSnapshotInfo *sn = &state->sn;
1596     Error *local_error = NULL;
1597 
1598     if (!state->created) {
1599         return;
1600     }
1601 
1602     if (bdrv_snapshot_delete(bs, sn->id_str, sn->name, &local_error) < 0) {
1603         error_reportf_err(local_error,
1604                           "Failed to delete snapshot with id '%s' and "
1605                           "name '%s' on device '%s' in abort: ",
1606                           sn->id_str, sn->name,
1607                           bdrv_get_device_name(bs));
1608     }
1609 }
1610 
1611 static void internal_snapshot_clean(BlkActionState *common)
1612 {
1613     InternalSnapshotState *state = DO_UPCAST(InternalSnapshotState,
1614                                              common, common);
1615 
1616     if (state->aio_context) {
1617         if (state->bs) {
1618             bdrv_drained_end(state->bs);
1619         }
1620         aio_context_release(state->aio_context);
1621     }
1622 }
1623 
1624 /* external snapshot private data */
1625 typedef struct ExternalSnapshotState {
1626     BlkActionState common;
1627     BlockDriverState *old_bs;
1628     BlockDriverState *new_bs;
1629     AioContext *aio_context;
1630 } ExternalSnapshotState;
1631 
1632 static void external_snapshot_prepare(BlkActionState *common,
1633                                       Error **errp)
1634 {
1635     int flags = 0, ret;
1636     QDict *options = NULL;
1637     Error *local_err = NULL;
1638     /* Device and node name of the image to generate the snapshot from */
1639     const char *device;
1640     const char *node_name;
1641     /* Reference to the new image (for 'blockdev-snapshot') */
1642     const char *snapshot_ref;
1643     /* File name of the new image (for 'blockdev-snapshot-sync') */
1644     const char *new_image_file;
1645     ExternalSnapshotState *state =
1646                              DO_UPCAST(ExternalSnapshotState, common, common);
1647     TransactionAction *action = common->action;
1648 
1649     /* 'blockdev-snapshot' and 'blockdev-snapshot-sync' have similar
1650      * purpose but a different set of parameters */
1651     switch (action->type) {
1652     case TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT:
1653         {
1654             BlockdevSnapshot *s = action->u.blockdev_snapshot;
1655             device = s->node;
1656             node_name = s->node;
1657             new_image_file = NULL;
1658             snapshot_ref = s->overlay;
1659         }
1660         break;
1661     case TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC:
1662         {
1663             BlockdevSnapshotSync *s = action->u.blockdev_snapshot_sync;
1664             device = s->has_device ? s->device : NULL;
1665             node_name = s->has_node_name ? s->node_name : NULL;
1666             new_image_file = s->snapshot_file;
1667             snapshot_ref = NULL;
1668         }
1669         break;
1670     default:
1671         g_assert_not_reached();
1672     }
1673 
1674     /* start processing */
1675     if (action_check_completion_mode(common, errp) < 0) {
1676         return;
1677     }
1678 
1679     state->old_bs = bdrv_lookup_bs(device, node_name, errp);
1680     if (!state->old_bs) {
1681         return;
1682     }
1683 
1684     /* Acquire AioContext now so any threads operating on old_bs stop */
1685     state->aio_context = bdrv_get_aio_context(state->old_bs);
1686     aio_context_acquire(state->aio_context);
1687     bdrv_drained_begin(state->old_bs);
1688 
1689     if (!bdrv_is_inserted(state->old_bs)) {
1690         error_setg(errp, QERR_DEVICE_HAS_NO_MEDIUM, device);
1691         return;
1692     }
1693 
1694     if (bdrv_op_is_blocked(state->old_bs,
1695                            BLOCK_OP_TYPE_EXTERNAL_SNAPSHOT, errp)) {
1696         return;
1697     }
1698 
1699     if (!bdrv_is_read_only(state->old_bs)) {
1700         if (bdrv_flush(state->old_bs)) {
1701             error_setg(errp, QERR_IO_ERROR);
1702             return;
1703         }
1704     }
1705 
1706     if (!bdrv_is_first_non_filter(state->old_bs)) {
1707         error_setg(errp, QERR_FEATURE_DISABLED, "snapshot");
1708         return;
1709     }
1710 
1711     if (action->type == TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC) {
1712         BlockdevSnapshotSync *s = action->u.blockdev_snapshot_sync;
1713         const char *format = s->has_format ? s->format : "qcow2";
1714         enum NewImageMode mode;
1715         const char *snapshot_node_name =
1716             s->has_snapshot_node_name ? s->snapshot_node_name : NULL;
1717 
1718         if (node_name && !snapshot_node_name) {
1719             error_setg(errp, "New snapshot node name missing");
1720             return;
1721         }
1722 
1723         if (snapshot_node_name &&
1724             bdrv_lookup_bs(snapshot_node_name, snapshot_node_name, NULL)) {
1725             error_setg(errp, "New snapshot node name already in use");
1726             return;
1727         }
1728 
1729         flags = state->old_bs->open_flags;
1730 
1731         /* create new image w/backing file */
1732         mode = s->has_mode ? s->mode : NEW_IMAGE_MODE_ABSOLUTE_PATHS;
1733         if (mode != NEW_IMAGE_MODE_EXISTING) {
1734             bdrv_img_create(new_image_file, format,
1735                             state->old_bs->filename,
1736                             state->old_bs->drv->format_name,
1737                             NULL, -1, flags, &local_err, false);
1738             if (local_err) {
1739                 error_propagate(errp, local_err);
1740                 return;
1741             }
1742         }
1743 
1744         options = qdict_new();
1745         if (s->has_snapshot_node_name) {
1746             qdict_put(options, "node-name",
1747                       qstring_from_str(snapshot_node_name));
1748         }
1749         qdict_put(options, "driver", qstring_from_str(format));
1750 
1751         flags |= BDRV_O_NO_BACKING;
1752     }
1753 
1754     assert(state->new_bs == NULL);
1755     ret = bdrv_open(&state->new_bs, new_image_file, snapshot_ref, options,
1756                     flags, errp);
1757     /* We will manually add the backing_hd field to the bs later */
1758     if (ret != 0) {
1759         return;
1760     }
1761 
1762     if (state->new_bs->blk != NULL) {
1763         error_setg(errp, "The snapshot is already in use by %s",
1764                    blk_name(state->new_bs->blk));
1765         return;
1766     }
1767 
1768     if (bdrv_op_is_blocked(state->new_bs, BLOCK_OP_TYPE_EXTERNAL_SNAPSHOT,
1769                            errp)) {
1770         return;
1771     }
1772 
1773     if (state->new_bs->backing != NULL) {
1774         error_setg(errp, "The snapshot already has a backing image");
1775         return;
1776     }
1777 
1778     if (!state->new_bs->drv->supports_backing) {
1779         error_setg(errp, "The snapshot does not support backing images");
1780     }
1781 }
1782 
1783 static void external_snapshot_commit(BlkActionState *common)
1784 {
1785     ExternalSnapshotState *state =
1786                              DO_UPCAST(ExternalSnapshotState, common, common);
1787 
1788     bdrv_set_aio_context(state->new_bs, state->aio_context);
1789 
1790     /* This removes our old bs and adds the new bs */
1791     bdrv_append(state->new_bs, state->old_bs);
1792     /* We don't need (or want) to use the transactional
1793      * bdrv_reopen_multiple() across all the entries at once, because we
1794      * don't want to abort all of them if one of them fails the reopen */
1795     bdrv_reopen(state->old_bs, state->old_bs->open_flags & ~BDRV_O_RDWR,
1796                 NULL);
1797 }
1798 
1799 static void external_snapshot_abort(BlkActionState *common)
1800 {
1801     ExternalSnapshotState *state =
1802                              DO_UPCAST(ExternalSnapshotState, common, common);
1803     if (state->new_bs) {
1804         bdrv_unref(state->new_bs);
1805     }
1806 }
1807 
1808 static void external_snapshot_clean(BlkActionState *common)
1809 {
1810     ExternalSnapshotState *state =
1811                              DO_UPCAST(ExternalSnapshotState, common, common);
1812     if (state->aio_context) {
1813         bdrv_drained_end(state->old_bs);
1814         aio_context_release(state->aio_context);
1815     }
1816 }
1817 
1818 typedef struct DriveBackupState {
1819     BlkActionState common;
1820     BlockDriverState *bs;
1821     AioContext *aio_context;
1822     BlockJob *job;
1823 } DriveBackupState;
1824 
1825 static void do_drive_backup(const char *device, const char *target,
1826                             bool has_format, const char *format,
1827                             enum MirrorSyncMode sync,
1828                             bool has_mode, enum NewImageMode mode,
1829                             bool has_speed, int64_t speed,
1830                             bool has_bitmap, const char *bitmap,
1831                             bool has_on_source_error,
1832                             BlockdevOnError on_source_error,
1833                             bool has_on_target_error,
1834                             BlockdevOnError on_target_error,
1835                             BlockJobTxn *txn, Error **errp);
1836 
1837 static void drive_backup_prepare(BlkActionState *common, Error **errp)
1838 {
1839     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1840     BlockBackend *blk;
1841     DriveBackup *backup;
1842     Error *local_err = NULL;
1843 
1844     assert(common->action->type == TRANSACTION_ACTION_KIND_DRIVE_BACKUP);
1845     backup = common->action->u.drive_backup;
1846 
1847     blk = blk_by_name(backup->device);
1848     if (!blk) {
1849         error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
1850                   "Device '%s' not found", backup->device);
1851         return;
1852     }
1853 
1854     if (!blk_is_available(blk)) {
1855         error_setg(errp, QERR_DEVICE_HAS_NO_MEDIUM, backup->device);
1856         return;
1857     }
1858 
1859     /* AioContext is released in .clean() */
1860     state->aio_context = blk_get_aio_context(blk);
1861     aio_context_acquire(state->aio_context);
1862     bdrv_drained_begin(blk_bs(blk));
1863     state->bs = blk_bs(blk);
1864 
1865     do_drive_backup(backup->device, backup->target,
1866                     backup->has_format, backup->format,
1867                     backup->sync,
1868                     backup->has_mode, backup->mode,
1869                     backup->has_speed, backup->speed,
1870                     backup->has_bitmap, backup->bitmap,
1871                     backup->has_on_source_error, backup->on_source_error,
1872                     backup->has_on_target_error, backup->on_target_error,
1873                     common->block_job_txn, &local_err);
1874     if (local_err) {
1875         error_propagate(errp, local_err);
1876         return;
1877     }
1878 
1879     state->job = state->bs->job;
1880 }
1881 
1882 static void drive_backup_abort(BlkActionState *common)
1883 {
1884     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1885     BlockDriverState *bs = state->bs;
1886 
1887     /* Only cancel if it's the job we started */
1888     if (bs && bs->job && bs->job == state->job) {
1889         block_job_cancel_sync(bs->job);
1890     }
1891 }
1892 
1893 static void drive_backup_clean(BlkActionState *common)
1894 {
1895     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1896 
1897     if (state->aio_context) {
1898         bdrv_drained_end(state->bs);
1899         aio_context_release(state->aio_context);
1900     }
1901 }
1902 
1903 typedef struct BlockdevBackupState {
1904     BlkActionState common;
1905     BlockDriverState *bs;
1906     BlockJob *job;
1907     AioContext *aio_context;
1908 } BlockdevBackupState;
1909 
1910 static void do_blockdev_backup(const char *device, const char *target,
1911                                enum MirrorSyncMode sync,
1912                                bool has_speed, int64_t speed,
1913                                bool has_on_source_error,
1914                                BlockdevOnError on_source_error,
1915                                bool has_on_target_error,
1916                                BlockdevOnError on_target_error,
1917                                BlockJobTxn *txn, Error **errp);
1918 
1919 static void blockdev_backup_prepare(BlkActionState *common, Error **errp)
1920 {
1921     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1922     BlockdevBackup *backup;
1923     BlockBackend *blk, *target;
1924     Error *local_err = NULL;
1925 
1926     assert(common->action->type == TRANSACTION_ACTION_KIND_BLOCKDEV_BACKUP);
1927     backup = common->action->u.blockdev_backup;
1928 
1929     blk = blk_by_name(backup->device);
1930     if (!blk) {
1931         error_setg(errp, "Device '%s' not found", backup->device);
1932         return;
1933     }
1934 
1935     if (!blk_is_available(blk)) {
1936         error_setg(errp, QERR_DEVICE_HAS_NO_MEDIUM, backup->device);
1937         return;
1938     }
1939 
1940     target = blk_by_name(backup->target);
1941     if (!target) {
1942         error_setg(errp, "Device '%s' not found", backup->target);
1943         return;
1944     }
1945 
1946     /* AioContext is released in .clean() */
1947     state->aio_context = blk_get_aio_context(blk);
1948     if (state->aio_context != blk_get_aio_context(target)) {
1949         state->aio_context = NULL;
1950         error_setg(errp, "Backup between two IO threads is not implemented");
1951         return;
1952     }
1953     aio_context_acquire(state->aio_context);
1954     state->bs = blk_bs(blk);
1955     bdrv_drained_begin(state->bs);
1956 
1957     do_blockdev_backup(backup->device, backup->target,
1958                        backup->sync,
1959                        backup->has_speed, backup->speed,
1960                        backup->has_on_source_error, backup->on_source_error,
1961                        backup->has_on_target_error, backup->on_target_error,
1962                        common->block_job_txn, &local_err);
1963     if (local_err) {
1964         error_propagate(errp, local_err);
1965         return;
1966     }
1967 
1968     state->job = state->bs->job;
1969 }
1970 
1971 static void blockdev_backup_abort(BlkActionState *common)
1972 {
1973     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1974     BlockDriverState *bs = state->bs;
1975 
1976     /* Only cancel if it's the job we started */
1977     if (bs && bs->job && bs->job == state->job) {
1978         block_job_cancel_sync(bs->job);
1979     }
1980 }
1981 
1982 static void blockdev_backup_clean(BlkActionState *common)
1983 {
1984     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1985 
1986     if (state->aio_context) {
1987         bdrv_drained_end(state->bs);
1988         aio_context_release(state->aio_context);
1989     }
1990 }
1991 
1992 typedef struct BlockDirtyBitmapState {
1993     BlkActionState common;
1994     BdrvDirtyBitmap *bitmap;
1995     BlockDriverState *bs;
1996     AioContext *aio_context;
1997     HBitmap *backup;
1998     bool prepared;
1999 } BlockDirtyBitmapState;
2000 
2001 static void block_dirty_bitmap_add_prepare(BlkActionState *common,
2002                                            Error **errp)
2003 {
2004     Error *local_err = NULL;
2005     BlockDirtyBitmapAdd *action;
2006     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2007                                              common, common);
2008 
2009     if (action_check_completion_mode(common, errp) < 0) {
2010         return;
2011     }
2012 
2013     action = common->action->u.block_dirty_bitmap_add;
2014     /* AIO context taken and released within qmp_block_dirty_bitmap_add */
2015     qmp_block_dirty_bitmap_add(action->node, action->name,
2016                                action->has_granularity, action->granularity,
2017                                &local_err);
2018 
2019     if (!local_err) {
2020         state->prepared = true;
2021     } else {
2022         error_propagate(errp, local_err);
2023     }
2024 }
2025 
2026 static void block_dirty_bitmap_add_abort(BlkActionState *common)
2027 {
2028     BlockDirtyBitmapAdd *action;
2029     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2030                                              common, common);
2031 
2032     action = common->action->u.block_dirty_bitmap_add;
2033     /* Should not be able to fail: IF the bitmap was added via .prepare(),
2034      * then the node reference and bitmap name must have been valid.
2035      */
2036     if (state->prepared) {
2037         qmp_block_dirty_bitmap_remove(action->node, action->name, &error_abort);
2038     }
2039 }
2040 
2041 static void block_dirty_bitmap_clear_prepare(BlkActionState *common,
2042                                              Error **errp)
2043 {
2044     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2045                                              common, common);
2046     BlockDirtyBitmap *action;
2047 
2048     if (action_check_completion_mode(common, errp) < 0) {
2049         return;
2050     }
2051 
2052     action = common->action->u.block_dirty_bitmap_clear;
2053     state->bitmap = block_dirty_bitmap_lookup(action->node,
2054                                               action->name,
2055                                               &state->bs,
2056                                               &state->aio_context,
2057                                               errp);
2058     if (!state->bitmap) {
2059         return;
2060     }
2061 
2062     if (bdrv_dirty_bitmap_frozen(state->bitmap)) {
2063         error_setg(errp, "Cannot modify a frozen bitmap");
2064         return;
2065     } else if (!bdrv_dirty_bitmap_enabled(state->bitmap)) {
2066         error_setg(errp, "Cannot clear a disabled bitmap");
2067         return;
2068     }
2069 
2070     bdrv_clear_dirty_bitmap(state->bitmap, &state->backup);
2071     /* AioContext is released in .clean() */
2072 }
2073 
2074 static void block_dirty_bitmap_clear_abort(BlkActionState *common)
2075 {
2076     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2077                                              common, common);
2078 
2079     bdrv_undo_clear_dirty_bitmap(state->bitmap, state->backup);
2080 }
2081 
2082 static void block_dirty_bitmap_clear_commit(BlkActionState *common)
2083 {
2084     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2085                                              common, common);
2086 
2087     hbitmap_free(state->backup);
2088 }
2089 
2090 static void block_dirty_bitmap_clear_clean(BlkActionState *common)
2091 {
2092     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2093                                              common, common);
2094 
2095     if (state->aio_context) {
2096         aio_context_release(state->aio_context);
2097     }
2098 }
2099 
2100 static void abort_prepare(BlkActionState *common, Error **errp)
2101 {
2102     error_setg(errp, "Transaction aborted using Abort action");
2103 }
2104 
2105 static void abort_commit(BlkActionState *common)
2106 {
2107     g_assert_not_reached(); /* this action never succeeds */
2108 }
2109 
2110 static const BlkActionOps actions[] = {
2111     [TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT] = {
2112         .instance_size = sizeof(ExternalSnapshotState),
2113         .prepare  = external_snapshot_prepare,
2114         .commit   = external_snapshot_commit,
2115         .abort = external_snapshot_abort,
2116         .clean = external_snapshot_clean,
2117     },
2118     [TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC] = {
2119         .instance_size = sizeof(ExternalSnapshotState),
2120         .prepare  = external_snapshot_prepare,
2121         .commit   = external_snapshot_commit,
2122         .abort = external_snapshot_abort,
2123         .clean = external_snapshot_clean,
2124     },
2125     [TRANSACTION_ACTION_KIND_DRIVE_BACKUP] = {
2126         .instance_size = sizeof(DriveBackupState),
2127         .prepare = drive_backup_prepare,
2128         .abort = drive_backup_abort,
2129         .clean = drive_backup_clean,
2130     },
2131     [TRANSACTION_ACTION_KIND_BLOCKDEV_BACKUP] = {
2132         .instance_size = sizeof(BlockdevBackupState),
2133         .prepare = blockdev_backup_prepare,
2134         .abort = blockdev_backup_abort,
2135         .clean = blockdev_backup_clean,
2136     },
2137     [TRANSACTION_ACTION_KIND_ABORT] = {
2138         .instance_size = sizeof(BlkActionState),
2139         .prepare = abort_prepare,
2140         .commit = abort_commit,
2141     },
2142     [TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC] = {
2143         .instance_size = sizeof(InternalSnapshotState),
2144         .prepare  = internal_snapshot_prepare,
2145         .abort = internal_snapshot_abort,
2146         .clean = internal_snapshot_clean,
2147     },
2148     [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_ADD] = {
2149         .instance_size = sizeof(BlockDirtyBitmapState),
2150         .prepare = block_dirty_bitmap_add_prepare,
2151         .abort = block_dirty_bitmap_add_abort,
2152     },
2153     [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_CLEAR] = {
2154         .instance_size = sizeof(BlockDirtyBitmapState),
2155         .prepare = block_dirty_bitmap_clear_prepare,
2156         .commit = block_dirty_bitmap_clear_commit,
2157         .abort = block_dirty_bitmap_clear_abort,
2158         .clean = block_dirty_bitmap_clear_clean,
2159     }
2160 };
2161 
2162 /**
2163  * Allocate a TransactionProperties structure if necessary, and fill
2164  * that structure with desired defaults if they are unset.
2165  */
2166 static TransactionProperties *get_transaction_properties(
2167     TransactionProperties *props)
2168 {
2169     if (!props) {
2170         props = g_new0(TransactionProperties, 1);
2171     }
2172 
2173     if (!props->has_completion_mode) {
2174         props->has_completion_mode = true;
2175         props->completion_mode = ACTION_COMPLETION_MODE_INDIVIDUAL;
2176     }
2177 
2178     return props;
2179 }
2180 
2181 /*
2182  * 'Atomic' group operations.  The operations are performed as a set, and if
2183  * any fail then we roll back all operations in the group.
2184  */
2185 void qmp_transaction(TransactionActionList *dev_list,
2186                      bool has_props,
2187                      struct TransactionProperties *props,
2188                      Error **errp)
2189 {
2190     TransactionActionList *dev_entry = dev_list;
2191     BlockJobTxn *block_job_txn = NULL;
2192     BlkActionState *state, *next;
2193     Error *local_err = NULL;
2194 
2195     QSIMPLEQ_HEAD(snap_bdrv_states, BlkActionState) snap_bdrv_states;
2196     QSIMPLEQ_INIT(&snap_bdrv_states);
2197 
2198     /* Does this transaction get canceled as a group on failure?
2199      * If not, we don't really need to make a BlockJobTxn.
2200      */
2201     props = get_transaction_properties(props);
2202     if (props->completion_mode != ACTION_COMPLETION_MODE_INDIVIDUAL) {
2203         block_job_txn = block_job_txn_new();
2204     }
2205 
2206     /* drain all i/o before any operations */
2207     bdrv_drain_all();
2208 
2209     /* We don't do anything in this loop that commits us to the operations */
2210     while (NULL != dev_entry) {
2211         TransactionAction *dev_info = NULL;
2212         const BlkActionOps *ops;
2213 
2214         dev_info = dev_entry->value;
2215         dev_entry = dev_entry->next;
2216 
2217         assert(dev_info->type < ARRAY_SIZE(actions));
2218 
2219         ops = &actions[dev_info->type];
2220         assert(ops->instance_size > 0);
2221 
2222         state = g_malloc0(ops->instance_size);
2223         state->ops = ops;
2224         state->action = dev_info;
2225         state->block_job_txn = block_job_txn;
2226         state->txn_props = props;
2227         QSIMPLEQ_INSERT_TAIL(&snap_bdrv_states, state, entry);
2228 
2229         state->ops->prepare(state, &local_err);
2230         if (local_err) {
2231             error_propagate(errp, local_err);
2232             goto delete_and_fail;
2233         }
2234     }
2235 
2236     QSIMPLEQ_FOREACH(state, &snap_bdrv_states, entry) {
2237         if (state->ops->commit) {
2238             state->ops->commit(state);
2239         }
2240     }
2241 
2242     /* success */
2243     goto exit;
2244 
2245 delete_and_fail:
2246     /* failure, and it is all-or-none; roll back all operations */
2247     QSIMPLEQ_FOREACH(state, &snap_bdrv_states, entry) {
2248         if (state->ops->abort) {
2249             state->ops->abort(state);
2250         }
2251     }
2252 exit:
2253     QSIMPLEQ_FOREACH_SAFE(state, &snap_bdrv_states, entry, next) {
2254         if (state->ops->clean) {
2255             state->ops->clean(state);
2256         }
2257         g_free(state);
2258     }
2259     if (!has_props) {
2260         qapi_free_TransactionProperties(props);
2261     }
2262     block_job_txn_unref(block_job_txn);
2263 }
2264 
2265 void qmp_eject(const char *device, bool has_force, bool force, Error **errp)
2266 {
2267     Error *local_err = NULL;
2268 
2269     qmp_blockdev_open_tray(device, has_force, force, &local_err);
2270     if (local_err) {
2271         error_propagate(errp, local_err);
2272         return;
2273     }
2274 
2275     qmp_x_blockdev_remove_medium(device, errp);
2276 }
2277 
2278 void qmp_block_passwd(bool has_device, const char *device,
2279                       bool has_node_name, const char *node_name,
2280                       const char *password, Error **errp)
2281 {
2282     Error *local_err = NULL;
2283     BlockDriverState *bs;
2284     AioContext *aio_context;
2285 
2286     bs = bdrv_lookup_bs(has_device ? device : NULL,
2287                         has_node_name ? node_name : NULL,
2288                         &local_err);
2289     if (local_err) {
2290         error_propagate(errp, local_err);
2291         return;
2292     }
2293 
2294     aio_context = bdrv_get_aio_context(bs);
2295     aio_context_acquire(aio_context);
2296 
2297     bdrv_add_key(bs, password, errp);
2298 
2299     aio_context_release(aio_context);
2300 }
2301 
2302 void qmp_blockdev_open_tray(const char *device, bool has_force, bool force,
2303                             Error **errp)
2304 {
2305     BlockBackend *blk;
2306     bool locked;
2307 
2308     if (!has_force) {
2309         force = false;
2310     }
2311 
2312     blk = blk_by_name(device);
2313     if (!blk) {
2314         error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
2315                   "Device '%s' not found", device);
2316         return;
2317     }
2318 
2319     if (!blk_dev_has_removable_media(blk)) {
2320         error_setg(errp, "Device '%s' is not removable", device);
2321         return;
2322     }
2323 
2324     if (!blk_dev_has_tray(blk)) {
2325         /* Ignore this command on tray-less devices */
2326         return;
2327     }
2328 
2329     if (blk_dev_is_tray_open(blk)) {
2330         return;
2331     }
2332 
2333     locked = blk_dev_is_medium_locked(blk);
2334     if (locked) {
2335         blk_dev_eject_request(blk, force);
2336     }
2337 
2338     if (!locked || force) {
2339         blk_dev_change_media_cb(blk, false);
2340     }
2341 }
2342 
2343 void qmp_blockdev_close_tray(const char *device, Error **errp)
2344 {
2345     BlockBackend *blk;
2346 
2347     blk = blk_by_name(device);
2348     if (!blk) {
2349         error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
2350                   "Device '%s' not found", device);
2351         return;
2352     }
2353 
2354     if (!blk_dev_has_removable_media(blk)) {
2355         error_setg(errp, "Device '%s' is not removable", device);
2356         return;
2357     }
2358 
2359     if (!blk_dev_has_tray(blk)) {
2360         /* Ignore this command on tray-less devices */
2361         return;
2362     }
2363 
2364     if (!blk_dev_is_tray_open(blk)) {
2365         return;
2366     }
2367 
2368     blk_dev_change_media_cb(blk, true);
2369 }
2370 
2371 void qmp_x_blockdev_remove_medium(const char *device, Error **errp)
2372 {
2373     BlockBackend *blk;
2374     BlockDriverState *bs;
2375     AioContext *aio_context;
2376     bool has_device;
2377 
2378     blk = blk_by_name(device);
2379     if (!blk) {
2380         error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
2381                   "Device '%s' not found", device);
2382         return;
2383     }
2384 
2385     /* For BBs without a device, we can exchange the BDS tree at will */
2386     has_device = blk_get_attached_dev(blk);
2387 
2388     if (has_device && !blk_dev_has_removable_media(blk)) {
2389         error_setg(errp, "Device '%s' is not removable", device);
2390         return;
2391     }
2392 
2393     if (has_device && blk_dev_has_tray(blk) && !blk_dev_is_tray_open(blk)) {
2394         error_setg(errp, "Tray of device '%s' is not open", device);
2395         return;
2396     }
2397 
2398     bs = blk_bs(blk);
2399     if (!bs) {
2400         return;
2401     }
2402 
2403     aio_context = bdrv_get_aio_context(bs);
2404     aio_context_acquire(aio_context);
2405 
2406     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_EJECT, errp)) {
2407         goto out;
2408     }
2409 
2410     /* This follows the convention established by bdrv_make_anon() */
2411     if (bs->device_list.tqe_prev) {
2412         bdrv_device_remove(bs);
2413     }
2414 
2415     blk_remove_bs(blk);
2416 
2417     if (!blk_dev_has_tray(blk)) {
2418         /* For tray-less devices, blockdev-open-tray is a no-op (or may not be
2419          * called at all); therefore, the medium needs to be ejected here.
2420          * Do it after blk_remove_bs() so blk_is_inserted(blk) returns the @load
2421          * value passed here (i.e. false). */
2422         blk_dev_change_media_cb(blk, false);
2423     }
2424 
2425 out:
2426     aio_context_release(aio_context);
2427 }
2428 
2429 static void qmp_blockdev_insert_anon_medium(const char *device,
2430                                             BlockDriverState *bs, Error **errp)
2431 {
2432     BlockBackend *blk;
2433     bool has_device;
2434 
2435     blk = blk_by_name(device);
2436     if (!blk) {
2437         error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
2438                   "Device '%s' not found", device);
2439         return;
2440     }
2441 
2442     /* For BBs without a device, we can exchange the BDS tree at will */
2443     has_device = blk_get_attached_dev(blk);
2444 
2445     if (has_device && !blk_dev_has_removable_media(blk)) {
2446         error_setg(errp, "Device '%s' is not removable", device);
2447         return;
2448     }
2449 
2450     if (has_device && blk_dev_has_tray(blk) && !blk_dev_is_tray_open(blk)) {
2451         error_setg(errp, "Tray of device '%s' is not open", device);
2452         return;
2453     }
2454 
2455     if (blk_bs(blk)) {
2456         error_setg(errp, "There already is a medium in device '%s'", device);
2457         return;
2458     }
2459 
2460     blk_insert_bs(blk, bs);
2461 
2462     QTAILQ_INSERT_TAIL(&bdrv_states, bs, device_list);
2463 
2464     if (!blk_dev_has_tray(blk)) {
2465         /* For tray-less devices, blockdev-close-tray is a no-op (or may not be
2466          * called at all); therefore, the medium needs to be pushed into the
2467          * slot here.
2468          * Do it after blk_insert_bs() so blk_is_inserted(blk) returns the @load
2469          * value passed here (i.e. true). */
2470         blk_dev_change_media_cb(blk, true);
2471     }
2472 }
2473 
2474 void qmp_x_blockdev_insert_medium(const char *device, const char *node_name,
2475                                   Error **errp)
2476 {
2477     BlockDriverState *bs;
2478 
2479     bs = bdrv_find_node(node_name);
2480     if (!bs) {
2481         error_setg(errp, "Node '%s' not found", node_name);
2482         return;
2483     }
2484 
2485     if (bs->blk) {
2486         error_setg(errp, "Node '%s' is already in use by '%s'", node_name,
2487                    blk_name(bs->blk));
2488         return;
2489     }
2490 
2491     qmp_blockdev_insert_anon_medium(device, bs, errp);
2492 }
2493 
2494 void qmp_blockdev_change_medium(const char *device, const char *filename,
2495                                 bool has_format, const char *format,
2496                                 bool has_read_only,
2497                                 BlockdevChangeReadOnlyMode read_only,
2498                                 Error **errp)
2499 {
2500     BlockBackend *blk;
2501     BlockDriverState *medium_bs = NULL;
2502     int bdrv_flags, ret;
2503     QDict *options = NULL;
2504     Error *err = NULL;
2505 
2506     blk = blk_by_name(device);
2507     if (!blk) {
2508         error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
2509                   "Device '%s' not found", device);
2510         goto fail;
2511     }
2512 
2513     if (blk_bs(blk)) {
2514         blk_update_root_state(blk);
2515     }
2516 
2517     bdrv_flags = blk_get_open_flags_from_root_state(blk);
2518 
2519     if (!has_read_only) {
2520         read_only = BLOCKDEV_CHANGE_READ_ONLY_MODE_RETAIN;
2521     }
2522 
2523     switch (read_only) {
2524     case BLOCKDEV_CHANGE_READ_ONLY_MODE_RETAIN:
2525         break;
2526 
2527     case BLOCKDEV_CHANGE_READ_ONLY_MODE_READ_ONLY:
2528         bdrv_flags &= ~BDRV_O_RDWR;
2529         break;
2530 
2531     case BLOCKDEV_CHANGE_READ_ONLY_MODE_READ_WRITE:
2532         bdrv_flags |= BDRV_O_RDWR;
2533         break;
2534 
2535     default:
2536         abort();
2537     }
2538 
2539     if (has_format) {
2540         options = qdict_new();
2541         qdict_put(options, "driver", qstring_from_str(format));
2542     }
2543 
2544     assert(!medium_bs);
2545     ret = bdrv_open(&medium_bs, filename, NULL, options, bdrv_flags, errp);
2546     if (ret < 0) {
2547         goto fail;
2548     }
2549 
2550     blk_apply_root_state(blk, medium_bs);
2551 
2552     bdrv_add_key(medium_bs, NULL, &err);
2553     if (err) {
2554         error_propagate(errp, err);
2555         goto fail;
2556     }
2557 
2558     qmp_blockdev_open_tray(device, false, false, &err);
2559     if (err) {
2560         error_propagate(errp, err);
2561         goto fail;
2562     }
2563 
2564     qmp_x_blockdev_remove_medium(device, &err);
2565     if (err) {
2566         error_propagate(errp, err);
2567         goto fail;
2568     }
2569 
2570     qmp_blockdev_insert_anon_medium(device, medium_bs, &err);
2571     if (err) {
2572         error_propagate(errp, err);
2573         goto fail;
2574     }
2575 
2576     qmp_blockdev_close_tray(device, errp);
2577 
2578 fail:
2579     /* If the medium has been inserted, the device has its own reference, so
2580      * ours must be relinquished; and if it has not been inserted successfully,
2581      * the reference must be relinquished anyway */
2582     bdrv_unref(medium_bs);
2583 }
2584 
2585 /* throttling disk I/O limits */
2586 void qmp_block_set_io_throttle(const char *device, int64_t bps, int64_t bps_rd,
2587                                int64_t bps_wr,
2588                                int64_t iops,
2589                                int64_t iops_rd,
2590                                int64_t iops_wr,
2591                                bool has_bps_max,
2592                                int64_t bps_max,
2593                                bool has_bps_rd_max,
2594                                int64_t bps_rd_max,
2595                                bool has_bps_wr_max,
2596                                int64_t bps_wr_max,
2597                                bool has_iops_max,
2598                                int64_t iops_max,
2599                                bool has_iops_rd_max,
2600                                int64_t iops_rd_max,
2601                                bool has_iops_wr_max,
2602                                int64_t iops_wr_max,
2603                                bool has_iops_size,
2604                                int64_t iops_size,
2605                                bool has_group,
2606                                const char *group, Error **errp)
2607 {
2608     ThrottleConfig cfg;
2609     BlockDriverState *bs;
2610     BlockBackend *blk;
2611     AioContext *aio_context;
2612 
2613     blk = blk_by_name(device);
2614     if (!blk) {
2615         error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
2616                   "Device '%s' not found", device);
2617         return;
2618     }
2619 
2620     aio_context = blk_get_aio_context(blk);
2621     aio_context_acquire(aio_context);
2622 
2623     bs = blk_bs(blk);
2624     if (!bs) {
2625         error_setg(errp, "Device '%s' has no medium", device);
2626         goto out;
2627     }
2628 
2629     memset(&cfg, 0, sizeof(cfg));
2630     cfg.buckets[THROTTLE_BPS_TOTAL].avg = bps;
2631     cfg.buckets[THROTTLE_BPS_READ].avg  = bps_rd;
2632     cfg.buckets[THROTTLE_BPS_WRITE].avg = bps_wr;
2633 
2634     cfg.buckets[THROTTLE_OPS_TOTAL].avg = iops;
2635     cfg.buckets[THROTTLE_OPS_READ].avg  = iops_rd;
2636     cfg.buckets[THROTTLE_OPS_WRITE].avg = iops_wr;
2637 
2638     if (has_bps_max) {
2639         cfg.buckets[THROTTLE_BPS_TOTAL].max = bps_max;
2640     }
2641     if (has_bps_rd_max) {
2642         cfg.buckets[THROTTLE_BPS_READ].max = bps_rd_max;
2643     }
2644     if (has_bps_wr_max) {
2645         cfg.buckets[THROTTLE_BPS_WRITE].max = bps_wr_max;
2646     }
2647     if (has_iops_max) {
2648         cfg.buckets[THROTTLE_OPS_TOTAL].max = iops_max;
2649     }
2650     if (has_iops_rd_max) {
2651         cfg.buckets[THROTTLE_OPS_READ].max = iops_rd_max;
2652     }
2653     if (has_iops_wr_max) {
2654         cfg.buckets[THROTTLE_OPS_WRITE].max = iops_wr_max;
2655     }
2656 
2657     if (has_iops_size) {
2658         cfg.op_size = iops_size;
2659     }
2660 
2661     if (!check_throttle_config(&cfg, errp)) {
2662         goto out;
2663     }
2664 
2665     if (throttle_enabled(&cfg)) {
2666         /* Enable I/O limits if they're not enabled yet, otherwise
2667          * just update the throttling group. */
2668         if (!bs->throttle_state) {
2669             bdrv_io_limits_enable(bs, has_group ? group : device);
2670         } else if (has_group) {
2671             bdrv_io_limits_update_group(bs, group);
2672         }
2673         /* Set the new throttling configuration */
2674         bdrv_set_io_limits(bs, &cfg);
2675     } else if (bs->throttle_state) {
2676         /* If all throttling settings are set to 0, disable I/O limits */
2677         bdrv_io_limits_disable(bs);
2678     }
2679 
2680 out:
2681     aio_context_release(aio_context);
2682 }
2683 
2684 void qmp_block_dirty_bitmap_add(const char *node, const char *name,
2685                                 bool has_granularity, uint32_t granularity,
2686                                 Error **errp)
2687 {
2688     AioContext *aio_context;
2689     BlockDriverState *bs;
2690 
2691     if (!name || name[0] == '\0') {
2692         error_setg(errp, "Bitmap name cannot be empty");
2693         return;
2694     }
2695 
2696     bs = bdrv_lookup_bs(node, node, errp);
2697     if (!bs) {
2698         return;
2699     }
2700 
2701     aio_context = bdrv_get_aio_context(bs);
2702     aio_context_acquire(aio_context);
2703 
2704     if (has_granularity) {
2705         if (granularity < 512 || !is_power_of_2(granularity)) {
2706             error_setg(errp, "Granularity must be power of 2 "
2707                              "and at least 512");
2708             goto out;
2709         }
2710     } else {
2711         /* Default to cluster size, if available: */
2712         granularity = bdrv_get_default_bitmap_granularity(bs);
2713     }
2714 
2715     bdrv_create_dirty_bitmap(bs, granularity, name, errp);
2716 
2717  out:
2718     aio_context_release(aio_context);
2719 }
2720 
2721 void qmp_block_dirty_bitmap_remove(const char *node, const char *name,
2722                                    Error **errp)
2723 {
2724     AioContext *aio_context;
2725     BlockDriverState *bs;
2726     BdrvDirtyBitmap *bitmap;
2727 
2728     bitmap = block_dirty_bitmap_lookup(node, name, &bs, &aio_context, errp);
2729     if (!bitmap || !bs) {
2730         return;
2731     }
2732 
2733     if (bdrv_dirty_bitmap_frozen(bitmap)) {
2734         error_setg(errp,
2735                    "Bitmap '%s' is currently frozen and cannot be removed",
2736                    name);
2737         goto out;
2738     }
2739     bdrv_dirty_bitmap_make_anon(bitmap);
2740     bdrv_release_dirty_bitmap(bs, bitmap);
2741 
2742  out:
2743     aio_context_release(aio_context);
2744 }
2745 
2746 /**
2747  * Completely clear a bitmap, for the purposes of synchronizing a bitmap
2748  * immediately after a full backup operation.
2749  */
2750 void qmp_block_dirty_bitmap_clear(const char *node, const char *name,
2751                                   Error **errp)
2752 {
2753     AioContext *aio_context;
2754     BdrvDirtyBitmap *bitmap;
2755     BlockDriverState *bs;
2756 
2757     bitmap = block_dirty_bitmap_lookup(node, name, &bs, &aio_context, errp);
2758     if (!bitmap || !bs) {
2759         return;
2760     }
2761 
2762     if (bdrv_dirty_bitmap_frozen(bitmap)) {
2763         error_setg(errp,
2764                    "Bitmap '%s' is currently frozen and cannot be modified",
2765                    name);
2766         goto out;
2767     } else if (!bdrv_dirty_bitmap_enabled(bitmap)) {
2768         error_setg(errp,
2769                    "Bitmap '%s' is currently disabled and cannot be cleared",
2770                    name);
2771         goto out;
2772     }
2773 
2774     bdrv_clear_dirty_bitmap(bitmap, NULL);
2775 
2776  out:
2777     aio_context_release(aio_context);
2778 }
2779 
2780 void hmp_drive_del(Monitor *mon, const QDict *qdict)
2781 {
2782     const char *id = qdict_get_str(qdict, "id");
2783     BlockBackend *blk;
2784     BlockDriverState *bs;
2785     AioContext *aio_context;
2786     Error *local_err = NULL;
2787 
2788     blk = blk_by_name(id);
2789     if (!blk) {
2790         error_report("Device '%s' not found", id);
2791         return;
2792     }
2793 
2794     if (!blk_legacy_dinfo(blk)) {
2795         error_report("Deleting device added with blockdev-add"
2796                      " is not supported");
2797         return;
2798     }
2799 
2800     aio_context = blk_get_aio_context(blk);
2801     aio_context_acquire(aio_context);
2802 
2803     bs = blk_bs(blk);
2804     if (bs) {
2805         if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_DRIVE_DEL, &local_err)) {
2806             error_report_err(local_err);
2807             aio_context_release(aio_context);
2808             return;
2809         }
2810 
2811         blk_remove_bs(blk);
2812     }
2813 
2814     /* if we have a device attached to this BlockDriverState
2815      * then we need to make the drive anonymous until the device
2816      * can be removed.  If this is a drive with no device backing
2817      * then we can just get rid of the block driver state right here.
2818      */
2819     if (blk_get_attached_dev(blk)) {
2820         blk_hide_on_behalf_of_hmp_drive_del(blk);
2821         /* Further I/O must not pause the guest */
2822         blk_set_on_error(blk, BLOCKDEV_ON_ERROR_REPORT,
2823                          BLOCKDEV_ON_ERROR_REPORT);
2824     } else {
2825         blk_unref(blk);
2826     }
2827 
2828     aio_context_release(aio_context);
2829 }
2830 
2831 void qmp_block_resize(bool has_device, const char *device,
2832                       bool has_node_name, const char *node_name,
2833                       int64_t size, Error **errp)
2834 {
2835     Error *local_err = NULL;
2836     BlockDriverState *bs;
2837     AioContext *aio_context;
2838     int ret;
2839 
2840     bs = bdrv_lookup_bs(has_device ? device : NULL,
2841                         has_node_name ? node_name : NULL,
2842                         &local_err);
2843     if (local_err) {
2844         error_propagate(errp, local_err);
2845         return;
2846     }
2847 
2848     aio_context = bdrv_get_aio_context(bs);
2849     aio_context_acquire(aio_context);
2850 
2851     if (!bdrv_is_first_non_filter(bs)) {
2852         error_setg(errp, QERR_FEATURE_DISABLED, "resize");
2853         goto out;
2854     }
2855 
2856     if (size < 0) {
2857         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "size", "a >0 size");
2858         goto out;
2859     }
2860 
2861     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_RESIZE, NULL)) {
2862         error_setg(errp, QERR_DEVICE_IN_USE, device);
2863         goto out;
2864     }
2865 
2866     /* complete all in-flight operations before resizing the device */
2867     bdrv_drain_all();
2868 
2869     ret = bdrv_truncate(bs, size);
2870     switch (ret) {
2871     case 0:
2872         break;
2873     case -ENOMEDIUM:
2874         error_setg(errp, QERR_DEVICE_HAS_NO_MEDIUM, device);
2875         break;
2876     case -ENOTSUP:
2877         error_setg(errp, QERR_UNSUPPORTED);
2878         break;
2879     case -EACCES:
2880         error_setg(errp, "Device '%s' is read only", device);
2881         break;
2882     case -EBUSY:
2883         error_setg(errp, QERR_DEVICE_IN_USE, device);
2884         break;
2885     default:
2886         error_setg_errno(errp, -ret, "Could not resize");
2887         break;
2888     }
2889 
2890 out:
2891     aio_context_release(aio_context);
2892 }
2893 
2894 static void block_job_cb(void *opaque, int ret)
2895 {
2896     /* Note that this function may be executed from another AioContext besides
2897      * the QEMU main loop.  If you need to access anything that assumes the
2898      * QEMU global mutex, use a BH or introduce a mutex.
2899      */
2900 
2901     BlockDriverState *bs = opaque;
2902     const char *msg = NULL;
2903 
2904     trace_block_job_cb(bs, bs->job, ret);
2905 
2906     assert(bs->job);
2907 
2908     if (ret < 0) {
2909         msg = strerror(-ret);
2910     }
2911 
2912     if (block_job_is_cancelled(bs->job)) {
2913         block_job_event_cancelled(bs->job);
2914     } else {
2915         block_job_event_completed(bs->job, msg);
2916     }
2917 }
2918 
2919 void qmp_block_stream(const char *device,
2920                       bool has_base, const char *base,
2921                       bool has_backing_file, const char *backing_file,
2922                       bool has_speed, int64_t speed,
2923                       bool has_on_error, BlockdevOnError on_error,
2924                       Error **errp)
2925 {
2926     BlockBackend *blk;
2927     BlockDriverState *bs;
2928     BlockDriverState *base_bs = NULL;
2929     AioContext *aio_context;
2930     Error *local_err = NULL;
2931     const char *base_name = NULL;
2932 
2933     if (!has_on_error) {
2934         on_error = BLOCKDEV_ON_ERROR_REPORT;
2935     }
2936 
2937     blk = blk_by_name(device);
2938     if (!blk) {
2939         error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
2940                   "Device '%s' not found", device);
2941         return;
2942     }
2943 
2944     aio_context = blk_get_aio_context(blk);
2945     aio_context_acquire(aio_context);
2946 
2947     if (!blk_is_available(blk)) {
2948         error_setg(errp, "Device '%s' has no medium", device);
2949         goto out;
2950     }
2951     bs = blk_bs(blk);
2952 
2953     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_STREAM, errp)) {
2954         goto out;
2955     }
2956 
2957     if (has_base) {
2958         base_bs = bdrv_find_backing_image(bs, base);
2959         if (base_bs == NULL) {
2960             error_setg(errp, QERR_BASE_NOT_FOUND, base);
2961             goto out;
2962         }
2963         assert(bdrv_get_aio_context(base_bs) == aio_context);
2964         base_name = base;
2965     }
2966 
2967     /* if we are streaming the entire chain, the result will have no backing
2968      * file, and specifying one is therefore an error */
2969     if (base_bs == NULL && has_backing_file) {
2970         error_setg(errp, "backing file specified, but streaming the "
2971                          "entire chain");
2972         goto out;
2973     }
2974 
2975     /* backing_file string overrides base bs filename */
2976     base_name = has_backing_file ? backing_file : base_name;
2977 
2978     stream_start(bs, base_bs, base_name, has_speed ? speed : 0,
2979                  on_error, block_job_cb, bs, &local_err);
2980     if (local_err) {
2981         error_propagate(errp, local_err);
2982         goto out;
2983     }
2984 
2985     trace_qmp_block_stream(bs, bs->job);
2986 
2987 out:
2988     aio_context_release(aio_context);
2989 }
2990 
2991 void qmp_block_commit(const char *device,
2992                       bool has_base, const char *base,
2993                       bool has_top, const char *top,
2994                       bool has_backing_file, const char *backing_file,
2995                       bool has_speed, int64_t speed,
2996                       Error **errp)
2997 {
2998     BlockBackend *blk;
2999     BlockDriverState *bs;
3000     BlockDriverState *base_bs, *top_bs;
3001     AioContext *aio_context;
3002     Error *local_err = NULL;
3003     /* This will be part of the QMP command, if/when the
3004      * BlockdevOnError change for blkmirror makes it in
3005      */
3006     BlockdevOnError on_error = BLOCKDEV_ON_ERROR_REPORT;
3007 
3008     if (!has_speed) {
3009         speed = 0;
3010     }
3011 
3012     /* Important Note:
3013      *  libvirt relies on the DeviceNotFound error class in order to probe for
3014      *  live commit feature versions; for this to work, we must make sure to
3015      *  perform the device lookup before any generic errors that may occur in a
3016      *  scenario in which all optional arguments are omitted. */
3017     blk = blk_by_name(device);
3018     if (!blk) {
3019         error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
3020                   "Device '%s' not found", device);
3021         return;
3022     }
3023 
3024     aio_context = blk_get_aio_context(blk);
3025     aio_context_acquire(aio_context);
3026 
3027     if (!blk_is_available(blk)) {
3028         error_setg(errp, "Device '%s' has no medium", device);
3029         goto out;
3030     }
3031     bs = blk_bs(blk);
3032 
3033     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_COMMIT_SOURCE, errp)) {
3034         goto out;
3035     }
3036 
3037     /* default top_bs is the active layer */
3038     top_bs = bs;
3039 
3040     if (has_top && top) {
3041         if (strcmp(bs->filename, top) != 0) {
3042             top_bs = bdrv_find_backing_image(bs, top);
3043         }
3044     }
3045 
3046     if (top_bs == NULL) {
3047         error_setg(errp, "Top image file %s not found", top ? top : "NULL");
3048         goto out;
3049     }
3050 
3051     assert(bdrv_get_aio_context(top_bs) == aio_context);
3052 
3053     if (has_base && base) {
3054         base_bs = bdrv_find_backing_image(top_bs, base);
3055     } else {
3056         base_bs = bdrv_find_base(top_bs);
3057     }
3058 
3059     if (base_bs == NULL) {
3060         error_setg(errp, QERR_BASE_NOT_FOUND, base ? base : "NULL");
3061         goto out;
3062     }
3063 
3064     assert(bdrv_get_aio_context(base_bs) == aio_context);
3065 
3066     if (bdrv_op_is_blocked(base_bs, BLOCK_OP_TYPE_COMMIT_TARGET, errp)) {
3067         goto out;
3068     }
3069 
3070     /* Do not allow attempts to commit an image into itself */
3071     if (top_bs == base_bs) {
3072         error_setg(errp, "cannot commit an image into itself");
3073         goto out;
3074     }
3075 
3076     if (top_bs == bs) {
3077         if (has_backing_file) {
3078             error_setg(errp, "'backing-file' specified,"
3079                              " but 'top' is the active layer");
3080             goto out;
3081         }
3082         commit_active_start(bs, base_bs, speed, on_error, block_job_cb,
3083                             bs, &local_err);
3084     } else {
3085         commit_start(bs, base_bs, top_bs, speed, on_error, block_job_cb, bs,
3086                      has_backing_file ? backing_file : NULL, &local_err);
3087     }
3088     if (local_err != NULL) {
3089         error_propagate(errp, local_err);
3090         goto out;
3091     }
3092 
3093 out:
3094     aio_context_release(aio_context);
3095 }
3096 
3097 static void do_drive_backup(const char *device, const char *target,
3098                             bool has_format, const char *format,
3099                             enum MirrorSyncMode sync,
3100                             bool has_mode, enum NewImageMode mode,
3101                             bool has_speed, int64_t speed,
3102                             bool has_bitmap, const char *bitmap,
3103                             bool has_on_source_error,
3104                             BlockdevOnError on_source_error,
3105                             bool has_on_target_error,
3106                             BlockdevOnError on_target_error,
3107                             BlockJobTxn *txn, Error **errp)
3108 {
3109     BlockBackend *blk;
3110     BlockDriverState *bs;
3111     BlockDriverState *target_bs;
3112     BlockDriverState *source = NULL;
3113     BdrvDirtyBitmap *bmap = NULL;
3114     AioContext *aio_context;
3115     QDict *options = NULL;
3116     Error *local_err = NULL;
3117     int flags;
3118     int64_t size;
3119     int ret;
3120 
3121     if (!has_speed) {
3122         speed = 0;
3123     }
3124     if (!has_on_source_error) {
3125         on_source_error = BLOCKDEV_ON_ERROR_REPORT;
3126     }
3127     if (!has_on_target_error) {
3128         on_target_error = BLOCKDEV_ON_ERROR_REPORT;
3129     }
3130     if (!has_mode) {
3131         mode = NEW_IMAGE_MODE_ABSOLUTE_PATHS;
3132     }
3133 
3134     blk = blk_by_name(device);
3135     if (!blk) {
3136         error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
3137                   "Device '%s' not found", device);
3138         return;
3139     }
3140 
3141     aio_context = blk_get_aio_context(blk);
3142     aio_context_acquire(aio_context);
3143 
3144     /* Although backup_run has this check too, we need to use bs->drv below, so
3145      * do an early check redundantly. */
3146     if (!blk_is_available(blk)) {
3147         error_setg(errp, QERR_DEVICE_HAS_NO_MEDIUM, device);
3148         goto out;
3149     }
3150     bs = blk_bs(blk);
3151 
3152     if (!has_format) {
3153         format = mode == NEW_IMAGE_MODE_EXISTING ? NULL : bs->drv->format_name;
3154     }
3155 
3156     /* Early check to avoid creating target */
3157     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_BACKUP_SOURCE, errp)) {
3158         goto out;
3159     }
3160 
3161     flags = bs->open_flags | BDRV_O_RDWR;
3162 
3163     /* See if we have a backing HD we can use to create our new image
3164      * on top of. */
3165     if (sync == MIRROR_SYNC_MODE_TOP) {
3166         source = backing_bs(bs);
3167         if (!source) {
3168             sync = MIRROR_SYNC_MODE_FULL;
3169         }
3170     }
3171     if (sync == MIRROR_SYNC_MODE_NONE) {
3172         source = bs;
3173     }
3174 
3175     size = bdrv_getlength(bs);
3176     if (size < 0) {
3177         error_setg_errno(errp, -size, "bdrv_getlength failed");
3178         goto out;
3179     }
3180 
3181     if (mode != NEW_IMAGE_MODE_EXISTING) {
3182         assert(format);
3183         if (source) {
3184             bdrv_img_create(target, format, source->filename,
3185                             source->drv->format_name, NULL,
3186                             size, flags, &local_err, false);
3187         } else {
3188             bdrv_img_create(target, format, NULL, NULL, NULL,
3189                             size, flags, &local_err, false);
3190         }
3191     }
3192 
3193     if (local_err) {
3194         error_propagate(errp, local_err);
3195         goto out;
3196     }
3197 
3198     if (format) {
3199         options = qdict_new();
3200         qdict_put(options, "driver", qstring_from_str(format));
3201     }
3202 
3203     target_bs = NULL;
3204     ret = bdrv_open(&target_bs, target, NULL, options, flags, &local_err);
3205     if (ret < 0) {
3206         error_propagate(errp, local_err);
3207         goto out;
3208     }
3209 
3210     bdrv_set_aio_context(target_bs, aio_context);
3211 
3212     if (has_bitmap) {
3213         bmap = bdrv_find_dirty_bitmap(bs, bitmap);
3214         if (!bmap) {
3215             error_setg(errp, "Bitmap '%s' could not be found", bitmap);
3216             bdrv_unref(target_bs);
3217             goto out;
3218         }
3219     }
3220 
3221     backup_start(bs, target_bs, speed, sync, bmap,
3222                  on_source_error, on_target_error,
3223                  block_job_cb, bs, txn, &local_err);
3224     if (local_err != NULL) {
3225         bdrv_unref(target_bs);
3226         error_propagate(errp, local_err);
3227         goto out;
3228     }
3229 
3230 out:
3231     aio_context_release(aio_context);
3232 }
3233 
3234 void qmp_drive_backup(const char *device, const char *target,
3235                       bool has_format, const char *format,
3236                       enum MirrorSyncMode sync,
3237                       bool has_mode, enum NewImageMode mode,
3238                       bool has_speed, int64_t speed,
3239                       bool has_bitmap, const char *bitmap,
3240                       bool has_on_source_error, BlockdevOnError on_source_error,
3241                       bool has_on_target_error, BlockdevOnError on_target_error,
3242                       Error **errp)
3243 {
3244     return do_drive_backup(device, target, has_format, format, sync,
3245                            has_mode, mode, has_speed, speed,
3246                            has_bitmap, bitmap,
3247                            has_on_source_error, on_source_error,
3248                            has_on_target_error, on_target_error,
3249                            NULL, errp);
3250 }
3251 
3252 BlockDeviceInfoList *qmp_query_named_block_nodes(Error **errp)
3253 {
3254     return bdrv_named_nodes_list(errp);
3255 }
3256 
3257 void do_blockdev_backup(const char *device, const char *target,
3258                          enum MirrorSyncMode sync,
3259                          bool has_speed, int64_t speed,
3260                          bool has_on_source_error,
3261                          BlockdevOnError on_source_error,
3262                          bool has_on_target_error,
3263                          BlockdevOnError on_target_error,
3264                          BlockJobTxn *txn, Error **errp)
3265 {
3266     BlockBackend *blk, *target_blk;
3267     BlockDriverState *bs;
3268     BlockDriverState *target_bs;
3269     Error *local_err = NULL;
3270     AioContext *aio_context;
3271 
3272     if (!has_speed) {
3273         speed = 0;
3274     }
3275     if (!has_on_source_error) {
3276         on_source_error = BLOCKDEV_ON_ERROR_REPORT;
3277     }
3278     if (!has_on_target_error) {
3279         on_target_error = BLOCKDEV_ON_ERROR_REPORT;
3280     }
3281 
3282     blk = blk_by_name(device);
3283     if (!blk) {
3284         error_setg(errp, "Device '%s' not found", device);
3285         return;
3286     }
3287 
3288     aio_context = blk_get_aio_context(blk);
3289     aio_context_acquire(aio_context);
3290 
3291     if (!blk_is_available(blk)) {
3292         error_setg(errp, "Device '%s' has no medium", device);
3293         goto out;
3294     }
3295     bs = blk_bs(blk);
3296 
3297     target_blk = blk_by_name(target);
3298     if (!target_blk) {
3299         error_setg(errp, "Device '%s' not found", target);
3300         goto out;
3301     }
3302 
3303     if (!blk_is_available(target_blk)) {
3304         error_setg(errp, "Device '%s' has no medium", target);
3305         goto out;
3306     }
3307     target_bs = blk_bs(target_blk);
3308 
3309     bdrv_ref(target_bs);
3310     bdrv_set_aio_context(target_bs, aio_context);
3311     backup_start(bs, target_bs, speed, sync, NULL, on_source_error,
3312                  on_target_error, block_job_cb, bs, txn, &local_err);
3313     if (local_err != NULL) {
3314         bdrv_unref(target_bs);
3315         error_propagate(errp, local_err);
3316     }
3317 out:
3318     aio_context_release(aio_context);
3319 }
3320 
3321 void qmp_blockdev_backup(const char *device, const char *target,
3322                          enum MirrorSyncMode sync,
3323                          bool has_speed, int64_t speed,
3324                          bool has_on_source_error,
3325                          BlockdevOnError on_source_error,
3326                          bool has_on_target_error,
3327                          BlockdevOnError on_target_error,
3328                          Error **errp)
3329 {
3330     do_blockdev_backup(device, target, sync, has_speed, speed,
3331                        has_on_source_error, on_source_error,
3332                        has_on_target_error, on_target_error,
3333                        NULL, errp);
3334 }
3335 
3336 /* Parameter check and block job starting for drive mirroring.
3337  * Caller should hold @device and @target's aio context (must be the same).
3338  **/
3339 static void blockdev_mirror_common(BlockDriverState *bs,
3340                                    BlockDriverState *target,
3341                                    bool has_replaces, const char *replaces,
3342                                    enum MirrorSyncMode sync,
3343                                    bool has_speed, int64_t speed,
3344                                    bool has_granularity, uint32_t granularity,
3345                                    bool has_buf_size, int64_t buf_size,
3346                                    bool has_on_source_error,
3347                                    BlockdevOnError on_source_error,
3348                                    bool has_on_target_error,
3349                                    BlockdevOnError on_target_error,
3350                                    bool has_unmap, bool unmap,
3351                                    Error **errp)
3352 {
3353 
3354     if (!has_speed) {
3355         speed = 0;
3356     }
3357     if (!has_on_source_error) {
3358         on_source_error = BLOCKDEV_ON_ERROR_REPORT;
3359     }
3360     if (!has_on_target_error) {
3361         on_target_error = BLOCKDEV_ON_ERROR_REPORT;
3362     }
3363     if (!has_granularity) {
3364         granularity = 0;
3365     }
3366     if (!has_buf_size) {
3367         buf_size = 0;
3368     }
3369     if (!has_unmap) {
3370         unmap = true;
3371     }
3372 
3373     if (granularity != 0 && (granularity < 512 || granularity > 1048576 * 64)) {
3374         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "granularity",
3375                    "a value in range [512B, 64MB]");
3376         return;
3377     }
3378     if (granularity & (granularity - 1)) {
3379         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "granularity",
3380                    "power of 2");
3381         return;
3382     }
3383 
3384     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_MIRROR_SOURCE, errp)) {
3385         return;
3386     }
3387     if (bdrv_op_is_blocked(target, BLOCK_OP_TYPE_MIRROR_TARGET, errp)) {
3388         return;
3389     }
3390     if (target->blk) {
3391         error_setg(errp, "Cannot mirror to an attached block device");
3392         return;
3393     }
3394 
3395     if (!bs->backing && sync == MIRROR_SYNC_MODE_TOP) {
3396         sync = MIRROR_SYNC_MODE_FULL;
3397     }
3398 
3399     /* pass the node name to replace to mirror start since it's loose coupling
3400      * and will allow to check whether the node still exist at mirror completion
3401      */
3402     mirror_start(bs, target,
3403                  has_replaces ? replaces : NULL,
3404                  speed, granularity, buf_size, sync,
3405                  on_source_error, on_target_error, unmap,
3406                  block_job_cb, bs, errp);
3407 }
3408 
3409 void qmp_drive_mirror(const char *device, const char *target,
3410                       bool has_format, const char *format,
3411                       bool has_node_name, const char *node_name,
3412                       bool has_replaces, const char *replaces,
3413                       enum MirrorSyncMode sync,
3414                       bool has_mode, enum NewImageMode mode,
3415                       bool has_speed, int64_t speed,
3416                       bool has_granularity, uint32_t granularity,
3417                       bool has_buf_size, int64_t buf_size,
3418                       bool has_on_source_error, BlockdevOnError on_source_error,
3419                       bool has_on_target_error, BlockdevOnError on_target_error,
3420                       bool has_unmap, bool unmap,
3421                       Error **errp)
3422 {
3423     BlockDriverState *bs;
3424     BlockBackend *blk;
3425     BlockDriverState *source, *target_bs;
3426     AioContext *aio_context;
3427     Error *local_err = NULL;
3428     QDict *options = NULL;
3429     int flags;
3430     int64_t size;
3431     int ret;
3432 
3433     blk = blk_by_name(device);
3434     if (!blk) {
3435         error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
3436                   "Device '%s' not found", device);
3437         return;
3438     }
3439 
3440     aio_context = blk_get_aio_context(blk);
3441     aio_context_acquire(aio_context);
3442 
3443     if (!blk_is_available(blk)) {
3444         error_setg(errp, QERR_DEVICE_HAS_NO_MEDIUM, device);
3445         goto out;
3446     }
3447     bs = blk_bs(blk);
3448     if (!has_mode) {
3449         mode = NEW_IMAGE_MODE_ABSOLUTE_PATHS;
3450     }
3451 
3452     if (!has_format) {
3453         format = mode == NEW_IMAGE_MODE_EXISTING ? NULL : bs->drv->format_name;
3454     }
3455 
3456     flags = bs->open_flags | BDRV_O_RDWR;
3457     source = backing_bs(bs);
3458     if (!source && sync == MIRROR_SYNC_MODE_TOP) {
3459         sync = MIRROR_SYNC_MODE_FULL;
3460     }
3461     if (sync == MIRROR_SYNC_MODE_NONE) {
3462         source = bs;
3463     }
3464 
3465     size = bdrv_getlength(bs);
3466     if (size < 0) {
3467         error_setg_errno(errp, -size, "bdrv_getlength failed");
3468         goto out;
3469     }
3470 
3471     if (has_replaces) {
3472         BlockDriverState *to_replace_bs;
3473         AioContext *replace_aio_context;
3474         int64_t replace_size;
3475 
3476         if (!has_node_name) {
3477             error_setg(errp, "a node-name must be provided when replacing a"
3478                              " named node of the graph");
3479             goto out;
3480         }
3481 
3482         to_replace_bs = check_to_replace_node(bs, replaces, &local_err);
3483 
3484         if (!to_replace_bs) {
3485             error_propagate(errp, local_err);
3486             goto out;
3487         }
3488 
3489         replace_aio_context = bdrv_get_aio_context(to_replace_bs);
3490         aio_context_acquire(replace_aio_context);
3491         replace_size = bdrv_getlength(to_replace_bs);
3492         aio_context_release(replace_aio_context);
3493 
3494         if (size != replace_size) {
3495             error_setg(errp, "cannot replace image with a mirror image of "
3496                              "different size");
3497             goto out;
3498         }
3499     }
3500 
3501     if ((sync == MIRROR_SYNC_MODE_FULL || !source)
3502         && mode != NEW_IMAGE_MODE_EXISTING)
3503     {
3504         /* create new image w/o backing file */
3505         assert(format);
3506         bdrv_img_create(target, format,
3507                         NULL, NULL, NULL, size, flags, &local_err, false);
3508     } else {
3509         switch (mode) {
3510         case NEW_IMAGE_MODE_EXISTING:
3511             break;
3512         case NEW_IMAGE_MODE_ABSOLUTE_PATHS:
3513             /* create new image with backing file */
3514             bdrv_img_create(target, format,
3515                             source->filename,
3516                             source->drv->format_name,
3517                             NULL, size, flags, &local_err, false);
3518             break;
3519         default:
3520             abort();
3521         }
3522     }
3523 
3524     if (local_err) {
3525         error_propagate(errp, local_err);
3526         goto out;
3527     }
3528 
3529     options = qdict_new();
3530     if (has_node_name) {
3531         qdict_put(options, "node-name", qstring_from_str(node_name));
3532     }
3533     if (format) {
3534         qdict_put(options, "driver", qstring_from_str(format));
3535     }
3536 
3537     /* Mirroring takes care of copy-on-write using the source's backing
3538      * file.
3539      */
3540     target_bs = NULL;
3541     ret = bdrv_open(&target_bs, target, NULL, options,
3542                     flags | BDRV_O_NO_BACKING, &local_err);
3543     if (ret < 0) {
3544         error_propagate(errp, local_err);
3545         goto out;
3546     }
3547 
3548     bdrv_set_aio_context(target_bs, aio_context);
3549 
3550     blockdev_mirror_common(bs, target_bs,
3551                            has_replaces, replaces, sync,
3552                            has_speed, speed,
3553                            has_granularity, granularity,
3554                            has_buf_size, buf_size,
3555                            has_on_source_error, on_source_error,
3556                            has_on_target_error, on_target_error,
3557                            has_unmap, unmap,
3558                            &local_err);
3559     if (local_err) {
3560         error_propagate(errp, local_err);
3561         bdrv_unref(target_bs);
3562     }
3563 out:
3564     aio_context_release(aio_context);
3565 }
3566 
3567 void qmp_blockdev_mirror(const char *device, const char *target,
3568                          bool has_replaces, const char *replaces,
3569                          MirrorSyncMode sync,
3570                          bool has_speed, int64_t speed,
3571                          bool has_granularity, uint32_t granularity,
3572                          bool has_buf_size, int64_t buf_size,
3573                          bool has_on_source_error,
3574                          BlockdevOnError on_source_error,
3575                          bool has_on_target_error,
3576                          BlockdevOnError on_target_error,
3577                          Error **errp)
3578 {
3579     BlockDriverState *bs;
3580     BlockBackend *blk;
3581     BlockDriverState *target_bs;
3582     AioContext *aio_context;
3583     Error *local_err = NULL;
3584 
3585     blk = blk_by_name(device);
3586     if (!blk) {
3587         error_setg(errp, "Device '%s' not found", device);
3588         return;
3589     }
3590     bs = blk_bs(blk);
3591 
3592     if (!bs) {
3593         error_setg(errp, "Device '%s' has no media", device);
3594         return;
3595     }
3596 
3597     target_bs = bdrv_lookup_bs(target, target, errp);
3598     if (!target_bs) {
3599         return;
3600     }
3601 
3602     aio_context = bdrv_get_aio_context(bs);
3603     aio_context_acquire(aio_context);
3604 
3605     bdrv_ref(target_bs);
3606     bdrv_set_aio_context(target_bs, aio_context);
3607 
3608     blockdev_mirror_common(bs, target_bs,
3609                            has_replaces, replaces, sync,
3610                            has_speed, speed,
3611                            has_granularity, granularity,
3612                            has_buf_size, buf_size,
3613                            has_on_source_error, on_source_error,
3614                            has_on_target_error, on_target_error,
3615                            true, true,
3616                            &local_err);
3617     if (local_err) {
3618         error_propagate(errp, local_err);
3619         bdrv_unref(target_bs);
3620     }
3621 
3622     aio_context_release(aio_context);
3623 }
3624 
3625 /* Get the block job for a given device name and acquire its AioContext */
3626 static BlockJob *find_block_job(const char *device, AioContext **aio_context,
3627                                 Error **errp)
3628 {
3629     BlockBackend *blk;
3630     BlockDriverState *bs;
3631 
3632     *aio_context = NULL;
3633 
3634     blk = blk_by_name(device);
3635     if (!blk) {
3636         goto notfound;
3637     }
3638 
3639     *aio_context = blk_get_aio_context(blk);
3640     aio_context_acquire(*aio_context);
3641 
3642     if (!blk_is_available(blk)) {
3643         goto notfound;
3644     }
3645     bs = blk_bs(blk);
3646 
3647     if (!bs->job) {
3648         goto notfound;
3649     }
3650 
3651     return bs->job;
3652 
3653 notfound:
3654     error_set(errp, ERROR_CLASS_DEVICE_NOT_ACTIVE,
3655               "No active block job on device '%s'", device);
3656     if (*aio_context) {
3657         aio_context_release(*aio_context);
3658         *aio_context = NULL;
3659     }
3660     return NULL;
3661 }
3662 
3663 void qmp_block_job_set_speed(const char *device, int64_t speed, Error **errp)
3664 {
3665     AioContext *aio_context;
3666     BlockJob *job = find_block_job(device, &aio_context, errp);
3667 
3668     if (!job) {
3669         return;
3670     }
3671 
3672     block_job_set_speed(job, speed, errp);
3673     aio_context_release(aio_context);
3674 }
3675 
3676 void qmp_block_job_cancel(const char *device,
3677                           bool has_force, bool force, Error **errp)
3678 {
3679     AioContext *aio_context;
3680     BlockJob *job = find_block_job(device, &aio_context, errp);
3681 
3682     if (!job) {
3683         return;
3684     }
3685 
3686     if (!has_force) {
3687         force = false;
3688     }
3689 
3690     if (job->user_paused && !force) {
3691         error_setg(errp, "The block job for device '%s' is currently paused",
3692                    device);
3693         goto out;
3694     }
3695 
3696     trace_qmp_block_job_cancel(job);
3697     block_job_cancel(job);
3698 out:
3699     aio_context_release(aio_context);
3700 }
3701 
3702 void qmp_block_job_pause(const char *device, Error **errp)
3703 {
3704     AioContext *aio_context;
3705     BlockJob *job = find_block_job(device, &aio_context, errp);
3706 
3707     if (!job || job->user_paused) {
3708         return;
3709     }
3710 
3711     job->user_paused = true;
3712     trace_qmp_block_job_pause(job);
3713     block_job_pause(job);
3714     aio_context_release(aio_context);
3715 }
3716 
3717 void qmp_block_job_resume(const char *device, Error **errp)
3718 {
3719     AioContext *aio_context;
3720     BlockJob *job = find_block_job(device, &aio_context, errp);
3721 
3722     if (!job || !job->user_paused) {
3723         return;
3724     }
3725 
3726     job->user_paused = false;
3727     trace_qmp_block_job_resume(job);
3728     block_job_resume(job);
3729     aio_context_release(aio_context);
3730 }
3731 
3732 void qmp_block_job_complete(const char *device, Error **errp)
3733 {
3734     AioContext *aio_context;
3735     BlockJob *job = find_block_job(device, &aio_context, errp);
3736 
3737     if (!job) {
3738         return;
3739     }
3740 
3741     trace_qmp_block_job_complete(job);
3742     block_job_complete(job, errp);
3743     aio_context_release(aio_context);
3744 }
3745 
3746 void qmp_change_backing_file(const char *device,
3747                              const char *image_node_name,
3748                              const char *backing_file,
3749                              Error **errp)
3750 {
3751     BlockBackend *blk;
3752     BlockDriverState *bs = NULL;
3753     AioContext *aio_context;
3754     BlockDriverState *image_bs = NULL;
3755     Error *local_err = NULL;
3756     bool ro;
3757     int open_flags;
3758     int ret;
3759 
3760     blk = blk_by_name(device);
3761     if (!blk) {
3762         error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
3763                   "Device '%s' not found", device);
3764         return;
3765     }
3766 
3767     aio_context = blk_get_aio_context(blk);
3768     aio_context_acquire(aio_context);
3769 
3770     if (!blk_is_available(blk)) {
3771         error_setg(errp, "Device '%s' has no medium", device);
3772         goto out;
3773     }
3774     bs = blk_bs(blk);
3775 
3776     image_bs = bdrv_lookup_bs(NULL, image_node_name, &local_err);
3777     if (local_err) {
3778         error_propagate(errp, local_err);
3779         goto out;
3780     }
3781 
3782     if (!image_bs) {
3783         error_setg(errp, "image file not found");
3784         goto out;
3785     }
3786 
3787     if (bdrv_find_base(image_bs) == image_bs) {
3788         error_setg(errp, "not allowing backing file change on an image "
3789                          "without a backing file");
3790         goto out;
3791     }
3792 
3793     /* even though we are not necessarily operating on bs, we need it to
3794      * determine if block ops are currently prohibited on the chain */
3795     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_CHANGE, errp)) {
3796         goto out;
3797     }
3798 
3799     /* final sanity check */
3800     if (!bdrv_chain_contains(bs, image_bs)) {
3801         error_setg(errp, "'%s' and image file are not in the same chain",
3802                    device);
3803         goto out;
3804     }
3805 
3806     /* if not r/w, reopen to make r/w */
3807     open_flags = image_bs->open_flags;
3808     ro = bdrv_is_read_only(image_bs);
3809 
3810     if (ro) {
3811         bdrv_reopen(image_bs, open_flags | BDRV_O_RDWR, &local_err);
3812         if (local_err) {
3813             error_propagate(errp, local_err);
3814             goto out;
3815         }
3816     }
3817 
3818     ret = bdrv_change_backing_file(image_bs, backing_file,
3819                                image_bs->drv ? image_bs->drv->format_name : "");
3820 
3821     if (ret < 0) {
3822         error_setg_errno(errp, -ret, "Could not change backing file to '%s'",
3823                          backing_file);
3824         /* don't exit here, so we can try to restore open flags if
3825          * appropriate */
3826     }
3827 
3828     if (ro) {
3829         bdrv_reopen(image_bs, open_flags, &local_err);
3830         if (local_err) {
3831             error_propagate(errp, local_err); /* will preserve prior errp */
3832         }
3833     }
3834 
3835 out:
3836     aio_context_release(aio_context);
3837 }
3838 
3839 void qmp_blockdev_add(BlockdevOptions *options, Error **errp)
3840 {
3841     QmpOutputVisitor *ov = qmp_output_visitor_new();
3842     BlockDriverState *bs;
3843     BlockBackend *blk = NULL;
3844     QObject *obj;
3845     QDict *qdict;
3846     Error *local_err = NULL;
3847 
3848     /* TODO Sort it out in raw-posix and drive_new(): Reject aio=native with
3849      * cache.direct=false instead of silently switching to aio=threads, except
3850      * when called from drive_new().
3851      *
3852      * For now, simply forbidding the combination for all drivers will do. */
3853     if (options->has_aio && options->aio == BLOCKDEV_AIO_OPTIONS_NATIVE) {
3854         bool direct = options->has_cache &&
3855                       options->cache->has_direct &&
3856                       options->cache->direct;
3857         if (!direct) {
3858             error_setg(errp, "aio=native requires cache.direct=true");
3859             goto fail;
3860         }
3861     }
3862 
3863     visit_type_BlockdevOptions(qmp_output_get_visitor(ov), NULL, &options,
3864                                &local_err);
3865     if (local_err) {
3866         error_propagate(errp, local_err);
3867         goto fail;
3868     }
3869 
3870     obj = qmp_output_get_qobject(ov);
3871     qdict = qobject_to_qdict(obj);
3872 
3873     qdict_flatten(qdict);
3874 
3875     if (options->has_id) {
3876         blk = blockdev_init(NULL, qdict, &local_err);
3877         if (local_err) {
3878             error_propagate(errp, local_err);
3879             goto fail;
3880         }
3881 
3882         bs = blk_bs(blk);
3883     } else {
3884         if (!qdict_get_try_str(qdict, "node-name")) {
3885             error_setg(errp, "'id' and/or 'node-name' need to be specified for "
3886                        "the root node");
3887             goto fail;
3888         }
3889 
3890         bs = bds_tree_init(qdict, errp);
3891         if (!bs) {
3892             goto fail;
3893         }
3894 
3895         QTAILQ_INSERT_TAIL(&monitor_bdrv_states, bs, monitor_list);
3896     }
3897 
3898     if (bs && bdrv_key_required(bs)) {
3899         if (blk) {
3900             blk_unref(blk);
3901         } else {
3902             QTAILQ_REMOVE(&monitor_bdrv_states, bs, monitor_list);
3903             bdrv_unref(bs);
3904         }
3905         error_setg(errp, "blockdev-add doesn't support encrypted devices");
3906         goto fail;
3907     }
3908 
3909 fail:
3910     qmp_output_visitor_cleanup(ov);
3911 }
3912 
3913 void qmp_x_blockdev_del(bool has_id, const char *id,
3914                         bool has_node_name, const char *node_name, Error **errp)
3915 {
3916     AioContext *aio_context;
3917     BlockBackend *blk;
3918     BlockDriverState *bs;
3919 
3920     if (has_id && has_node_name) {
3921         error_setg(errp, "Only one of id and node-name must be specified");
3922         return;
3923     } else if (!has_id && !has_node_name) {
3924         error_setg(errp, "No block device specified");
3925         return;
3926     }
3927 
3928     if (has_id) {
3929         blk = blk_by_name(id);
3930         if (!blk) {
3931             error_setg(errp, "Cannot find block backend %s", id);
3932             return;
3933         }
3934         if (blk_get_refcnt(blk) > 1) {
3935             error_setg(errp, "Block backend %s is in use", id);
3936             return;
3937         }
3938         bs = blk_bs(blk);
3939         aio_context = blk_get_aio_context(blk);
3940     } else {
3941         bs = bdrv_find_node(node_name);
3942         if (!bs) {
3943             error_setg(errp, "Cannot find node %s", node_name);
3944             return;
3945         }
3946         blk = bs->blk;
3947         if (blk) {
3948             error_setg(errp, "Node %s is in use by %s",
3949                        node_name, blk_name(blk));
3950             return;
3951         }
3952         aio_context = bdrv_get_aio_context(bs);
3953     }
3954 
3955     aio_context_acquire(aio_context);
3956 
3957     if (bs) {
3958         if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_DRIVE_DEL, errp)) {
3959             goto out;
3960         }
3961 
3962         if (!blk && !bs->monitor_list.tqe_prev) {
3963             error_setg(errp, "Node %s is not owned by the monitor",
3964                        bs->node_name);
3965             goto out;
3966         }
3967 
3968         if (bs->refcnt > 1) {
3969             error_setg(errp, "Block device %s is in use",
3970                        bdrv_get_device_or_node_name(bs));
3971             goto out;
3972         }
3973     }
3974 
3975     if (blk) {
3976         blk_unref(blk);
3977     } else {
3978         QTAILQ_REMOVE(&monitor_bdrv_states, bs, monitor_list);
3979         bdrv_unref(bs);
3980     }
3981 
3982 out:
3983     aio_context_release(aio_context);
3984 }
3985 
3986 BlockJobInfoList *qmp_query_block_jobs(Error **errp)
3987 {
3988     BlockJobInfoList *head = NULL, **p_next = &head;
3989     BlockDriverState *bs;
3990 
3991     for (bs = bdrv_next(NULL); bs; bs = bdrv_next(bs)) {
3992         AioContext *aio_context = bdrv_get_aio_context(bs);
3993 
3994         aio_context_acquire(aio_context);
3995 
3996         if (bs->job) {
3997             BlockJobInfoList *elem = g_new0(BlockJobInfoList, 1);
3998             elem->value = block_job_query(bs->job);
3999             *p_next = elem;
4000             p_next = &elem->next;
4001         }
4002 
4003         aio_context_release(aio_context);
4004     }
4005 
4006     return head;
4007 }
4008 
4009 QemuOptsList qemu_common_drive_opts = {
4010     .name = "drive",
4011     .head = QTAILQ_HEAD_INITIALIZER(qemu_common_drive_opts.head),
4012     .desc = {
4013         {
4014             .name = "snapshot",
4015             .type = QEMU_OPT_BOOL,
4016             .help = "enable/disable snapshot mode",
4017         },{
4018             .name = "discard",
4019             .type = QEMU_OPT_STRING,
4020             .help = "discard operation (ignore/off, unmap/on)",
4021         },{
4022             .name = "aio",
4023             .type = QEMU_OPT_STRING,
4024             .help = "host AIO implementation (threads, native)",
4025         },{
4026             .name = "format",
4027             .type = QEMU_OPT_STRING,
4028             .help = "disk format (raw, qcow2, ...)",
4029         },{
4030             .name = "rerror",
4031             .type = QEMU_OPT_STRING,
4032             .help = "read error action",
4033         },{
4034             .name = "werror",
4035             .type = QEMU_OPT_STRING,
4036             .help = "write error action",
4037         },{
4038             .name = "read-only",
4039             .type = QEMU_OPT_BOOL,
4040             .help = "open drive file as read-only",
4041         },{
4042             .name = "throttling.iops-total",
4043             .type = QEMU_OPT_NUMBER,
4044             .help = "limit total I/O operations per second",
4045         },{
4046             .name = "throttling.iops-read",
4047             .type = QEMU_OPT_NUMBER,
4048             .help = "limit read operations per second",
4049         },{
4050             .name = "throttling.iops-write",
4051             .type = QEMU_OPT_NUMBER,
4052             .help = "limit write operations per second",
4053         },{
4054             .name = "throttling.bps-total",
4055             .type = QEMU_OPT_NUMBER,
4056             .help = "limit total bytes per second",
4057         },{
4058             .name = "throttling.bps-read",
4059             .type = QEMU_OPT_NUMBER,
4060             .help = "limit read bytes per second",
4061         },{
4062             .name = "throttling.bps-write",
4063             .type = QEMU_OPT_NUMBER,
4064             .help = "limit write bytes per second",
4065         },{
4066             .name = "throttling.iops-total-max",
4067             .type = QEMU_OPT_NUMBER,
4068             .help = "I/O operations burst",
4069         },{
4070             .name = "throttling.iops-read-max",
4071             .type = QEMU_OPT_NUMBER,
4072             .help = "I/O operations read burst",
4073         },{
4074             .name = "throttling.iops-write-max",
4075             .type = QEMU_OPT_NUMBER,
4076             .help = "I/O operations write burst",
4077         },{
4078             .name = "throttling.bps-total-max",
4079             .type = QEMU_OPT_NUMBER,
4080             .help = "total bytes burst",
4081         },{
4082             .name = "throttling.bps-read-max",
4083             .type = QEMU_OPT_NUMBER,
4084             .help = "total bytes read burst",
4085         },{
4086             .name = "throttling.bps-write-max",
4087             .type = QEMU_OPT_NUMBER,
4088             .help = "total bytes write burst",
4089         },{
4090             .name = "throttling.iops-size",
4091             .type = QEMU_OPT_NUMBER,
4092             .help = "when limiting by iops max size of an I/O in bytes",
4093         },{
4094             .name = "throttling.group",
4095             .type = QEMU_OPT_STRING,
4096             .help = "name of the block throttling group",
4097         },{
4098             .name = "copy-on-read",
4099             .type = QEMU_OPT_BOOL,
4100             .help = "copy read data from backing file into image file",
4101         },{
4102             .name = "detect-zeroes",
4103             .type = QEMU_OPT_STRING,
4104             .help = "try to optimize zero writes (off, on, unmap)",
4105         },{
4106             .name = "stats-account-invalid",
4107             .type = QEMU_OPT_BOOL,
4108             .help = "whether to account for invalid I/O operations "
4109                     "in the statistics",
4110         },{
4111             .name = "stats-account-failed",
4112             .type = QEMU_OPT_BOOL,
4113             .help = "whether to account for failed I/O operations "
4114                     "in the statistics",
4115         },
4116         { /* end of list */ }
4117     },
4118 };
4119 
4120 static QemuOptsList qemu_root_bds_opts = {
4121     .name = "root-bds",
4122     .head = QTAILQ_HEAD_INITIALIZER(qemu_common_drive_opts.head),
4123     .desc = {
4124         {
4125             .name = "discard",
4126             .type = QEMU_OPT_STRING,
4127             .help = "discard operation (ignore/off, unmap/on)",
4128         },{
4129             .name = "aio",
4130             .type = QEMU_OPT_STRING,
4131             .help = "host AIO implementation (threads, native)",
4132         },{
4133             .name = "read-only",
4134             .type = QEMU_OPT_BOOL,
4135             .help = "open drive file as read-only",
4136         },{
4137             .name = "copy-on-read",
4138             .type = QEMU_OPT_BOOL,
4139             .help = "copy read data from backing file into image file",
4140         },{
4141             .name = "detect-zeroes",
4142             .type = QEMU_OPT_STRING,
4143             .help = "try to optimize zero writes (off, on, unmap)",
4144         },
4145         { /* end of list */ }
4146     },
4147 };
4148 
4149 QemuOptsList qemu_drive_opts = {
4150     .name = "drive",
4151     .head = QTAILQ_HEAD_INITIALIZER(qemu_drive_opts.head),
4152     .desc = {
4153         /*
4154          * no elements => accept any params
4155          * validation will happen later
4156          */
4157         { /* end of list */ }
4158     },
4159 };
4160