xref: /openbmc/qemu/blockdev.c (revision dc5bd18f)
1 /*
2  * QEMU host block devices
3  *
4  * Copyright (c) 2003-2008 Fabrice Bellard
5  *
6  * This work is licensed under the terms of the GNU GPL, version 2 or
7  * later.  See the COPYING file in the top-level directory.
8  *
9  * This file incorporates work covered by the following copyright and
10  * permission notice:
11  *
12  * Copyright (c) 2003-2008 Fabrice Bellard
13  *
14  * Permission is hereby granted, free of charge, to any person obtaining a copy
15  * of this software and associated documentation files (the "Software"), to deal
16  * in the Software without restriction, including without limitation the rights
17  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
18  * copies of the Software, and to permit persons to whom the Software is
19  * furnished to do so, subject to the following conditions:
20  *
21  * The above copyright notice and this permission notice shall be included in
22  * all copies or substantial portions of the Software.
23  *
24  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
25  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
26  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
27  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
28  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
29  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
30  * THE SOFTWARE.
31  */
32 
33 #include "qemu/osdep.h"
34 #include "sysemu/block-backend.h"
35 #include "sysemu/blockdev.h"
36 #include "hw/block/block.h"
37 #include "block/blockjob.h"
38 #include "block/throttle-groups.h"
39 #include "monitor/monitor.h"
40 #include "qemu/error-report.h"
41 #include "qemu/option.h"
42 #include "qemu/config-file.h"
43 #include "qapi/qapi-commands-block.h"
44 #include "qapi/qapi-commands-transaction.h"
45 #include "qapi/qapi-visit-block-core.h"
46 #include "qapi/qmp/qdict.h"
47 #include "qapi/qmp/qnum.h"
48 #include "qapi/qmp/qstring.h"
49 #include "qapi/error.h"
50 #include "qapi/qmp/qerror.h"
51 #include "qapi/qmp/qlist.h"
52 #include "qapi/qobject-output-visitor.h"
53 #include "sysemu/sysemu.h"
54 #include "sysemu/iothread.h"
55 #include "block/block_int.h"
56 #include "block/trace.h"
57 #include "sysemu/arch_init.h"
58 #include "sysemu/qtest.h"
59 #include "qemu/cutils.h"
60 #include "qemu/help_option.h"
61 #include "qemu/throttle-options.h"
62 
63 static QTAILQ_HEAD(, BlockDriverState) monitor_bdrv_states =
64     QTAILQ_HEAD_INITIALIZER(monitor_bdrv_states);
65 
66 static int do_open_tray(const char *blk_name, const char *qdev_id,
67                         bool force, Error **errp);
68 static void blockdev_remove_medium(bool has_device, const char *device,
69                                    bool has_id, const char *id, Error **errp);
70 static void blockdev_insert_medium(bool has_device, const char *device,
71                                    bool has_id, const char *id,
72                                    const char *node_name, Error **errp);
73 
74 static const char *const if_name[IF_COUNT] = {
75     [IF_NONE] = "none",
76     [IF_IDE] = "ide",
77     [IF_SCSI] = "scsi",
78     [IF_FLOPPY] = "floppy",
79     [IF_PFLASH] = "pflash",
80     [IF_MTD] = "mtd",
81     [IF_SD] = "sd",
82     [IF_VIRTIO] = "virtio",
83     [IF_XEN] = "xen",
84 };
85 
86 static int if_max_devs[IF_COUNT] = {
87     /*
88      * Do not change these numbers!  They govern how drive option
89      * index maps to unit and bus.  That mapping is ABI.
90      *
91      * All controllers used to implement if=T drives need to support
92      * if_max_devs[T] units, for any T with if_max_devs[T] != 0.
93      * Otherwise, some index values map to "impossible" bus, unit
94      * values.
95      *
96      * For instance, if you change [IF_SCSI] to 255, -drive
97      * if=scsi,index=12 no longer means bus=1,unit=5, but
98      * bus=0,unit=12.  With an lsi53c895a controller (7 units max),
99      * the drive can't be set up.  Regression.
100      */
101     [IF_IDE] = 2,
102     [IF_SCSI] = 7,
103 };
104 
105 /**
106  * Boards may call this to offer board-by-board overrides
107  * of the default, global values.
108  */
109 void override_max_devs(BlockInterfaceType type, int max_devs)
110 {
111     BlockBackend *blk;
112     DriveInfo *dinfo;
113 
114     if (max_devs <= 0) {
115         return;
116     }
117 
118     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
119         dinfo = blk_legacy_dinfo(blk);
120         if (dinfo->type == type) {
121             fprintf(stderr, "Cannot override units-per-bus property of"
122                     " the %s interface, because a drive of that type has"
123                     " already been added.\n", if_name[type]);
124             g_assert_not_reached();
125         }
126     }
127 
128     if_max_devs[type] = max_devs;
129 }
130 
131 /*
132  * We automatically delete the drive when a device using it gets
133  * unplugged.  Questionable feature, but we can't just drop it.
134  * Device models call blockdev_mark_auto_del() to schedule the
135  * automatic deletion, and generic qdev code calls blockdev_auto_del()
136  * when deletion is actually safe.
137  */
138 void blockdev_mark_auto_del(BlockBackend *blk)
139 {
140     DriveInfo *dinfo = blk_legacy_dinfo(blk);
141     BlockDriverState *bs = blk_bs(blk);
142     AioContext *aio_context;
143 
144     if (!dinfo) {
145         return;
146     }
147 
148     if (bs) {
149         aio_context = bdrv_get_aio_context(bs);
150         aio_context_acquire(aio_context);
151 
152         if (bs->job) {
153             block_job_cancel(bs->job);
154         }
155 
156         aio_context_release(aio_context);
157     }
158 
159     dinfo->auto_del = 1;
160 }
161 
162 void blockdev_auto_del(BlockBackend *blk)
163 {
164     DriveInfo *dinfo = blk_legacy_dinfo(blk);
165 
166     if (dinfo && dinfo->auto_del) {
167         monitor_remove_blk(blk);
168         blk_unref(blk);
169     }
170 }
171 
172 /**
173  * Returns the current mapping of how many units per bus
174  * a particular interface can support.
175  *
176  *  A positive integer indicates n units per bus.
177  *  0 implies the mapping has not been established.
178  * -1 indicates an invalid BlockInterfaceType was given.
179  */
180 int drive_get_max_devs(BlockInterfaceType type)
181 {
182     if (type >= IF_IDE && type < IF_COUNT) {
183         return if_max_devs[type];
184     }
185 
186     return -1;
187 }
188 
189 static int drive_index_to_bus_id(BlockInterfaceType type, int index)
190 {
191     int max_devs = if_max_devs[type];
192     return max_devs ? index / max_devs : 0;
193 }
194 
195 static int drive_index_to_unit_id(BlockInterfaceType type, int index)
196 {
197     int max_devs = if_max_devs[type];
198     return max_devs ? index % max_devs : index;
199 }
200 
201 QemuOpts *drive_def(const char *optstr)
202 {
203     return qemu_opts_parse_noisily(qemu_find_opts("drive"), optstr, false);
204 }
205 
206 QemuOpts *drive_add(BlockInterfaceType type, int index, const char *file,
207                     const char *optstr)
208 {
209     QemuOpts *opts;
210 
211     opts = drive_def(optstr);
212     if (!opts) {
213         return NULL;
214     }
215     if (type != IF_DEFAULT) {
216         qemu_opt_set(opts, "if", if_name[type], &error_abort);
217     }
218     if (index >= 0) {
219         qemu_opt_set_number(opts, "index", index, &error_abort);
220     }
221     if (file)
222         qemu_opt_set(opts, "file", file, &error_abort);
223     return opts;
224 }
225 
226 DriveInfo *drive_get(BlockInterfaceType type, int bus, int unit)
227 {
228     BlockBackend *blk;
229     DriveInfo *dinfo;
230 
231     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
232         dinfo = blk_legacy_dinfo(blk);
233         if (dinfo && dinfo->type == type
234             && dinfo->bus == bus && dinfo->unit == unit) {
235             return dinfo;
236         }
237     }
238 
239     return NULL;
240 }
241 
242 void drive_check_orphaned(void)
243 {
244     BlockBackend *blk;
245     DriveInfo *dinfo;
246     Location loc;
247     bool orphans = false;
248 
249     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
250         dinfo = blk_legacy_dinfo(blk);
251         if (!blk_get_attached_dev(blk) && !dinfo->is_default &&
252             dinfo->type != IF_NONE) {
253             loc_push_none(&loc);
254             qemu_opts_loc_restore(dinfo->opts);
255             error_report("machine type does not support"
256                          " if=%s,bus=%d,unit=%d",
257                          if_name[dinfo->type], dinfo->bus, dinfo->unit);
258             loc_pop(&loc);
259             orphans = true;
260         }
261     }
262 
263     if (orphans) {
264         exit(1);
265     }
266 }
267 
268 DriveInfo *drive_get_by_index(BlockInterfaceType type, int index)
269 {
270     return drive_get(type,
271                      drive_index_to_bus_id(type, index),
272                      drive_index_to_unit_id(type, index));
273 }
274 
275 int drive_get_max_bus(BlockInterfaceType type)
276 {
277     int max_bus;
278     BlockBackend *blk;
279     DriveInfo *dinfo;
280 
281     max_bus = -1;
282     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
283         dinfo = blk_legacy_dinfo(blk);
284         if (dinfo && dinfo->type == type && dinfo->bus > max_bus) {
285             max_bus = dinfo->bus;
286         }
287     }
288     return max_bus;
289 }
290 
291 /* Get a block device.  This should only be used for single-drive devices
292    (e.g. SD/Floppy/MTD).  Multi-disk devices (scsi/ide) should use the
293    appropriate bus.  */
294 DriveInfo *drive_get_next(BlockInterfaceType type)
295 {
296     static int next_block_unit[IF_COUNT];
297 
298     return drive_get(type, 0, next_block_unit[type]++);
299 }
300 
301 static void bdrv_format_print(void *opaque, const char *name)
302 {
303     error_printf(" %s", name);
304 }
305 
306 typedef struct {
307     QEMUBH *bh;
308     BlockDriverState *bs;
309 } BDRVPutRefBH;
310 
311 static int parse_block_error_action(const char *buf, bool is_read, Error **errp)
312 {
313     if (!strcmp(buf, "ignore")) {
314         return BLOCKDEV_ON_ERROR_IGNORE;
315     } else if (!is_read && !strcmp(buf, "enospc")) {
316         return BLOCKDEV_ON_ERROR_ENOSPC;
317     } else if (!strcmp(buf, "stop")) {
318         return BLOCKDEV_ON_ERROR_STOP;
319     } else if (!strcmp(buf, "report")) {
320         return BLOCKDEV_ON_ERROR_REPORT;
321     } else {
322         error_setg(errp, "'%s' invalid %s error action",
323                    buf, is_read ? "read" : "write");
324         return -1;
325     }
326 }
327 
328 static bool parse_stats_intervals(BlockAcctStats *stats, QList *intervals,
329                                   Error **errp)
330 {
331     const QListEntry *entry;
332     for (entry = qlist_first(intervals); entry; entry = qlist_next(entry)) {
333         switch (qobject_type(entry->value)) {
334 
335         case QTYPE_QSTRING: {
336             unsigned long long length;
337             const char *str = qstring_get_str(qobject_to_qstring(entry->value));
338             if (parse_uint_full(str, &length, 10) == 0 &&
339                 length > 0 && length <= UINT_MAX) {
340                 block_acct_add_interval(stats, (unsigned) length);
341             } else {
342                 error_setg(errp, "Invalid interval length: %s", str);
343                 return false;
344             }
345             break;
346         }
347 
348         case QTYPE_QNUM: {
349             int64_t length = qnum_get_int(qobject_to_qnum(entry->value));
350 
351             if (length > 0 && length <= UINT_MAX) {
352                 block_acct_add_interval(stats, (unsigned) length);
353             } else {
354                 error_setg(errp, "Invalid interval length: %" PRId64, length);
355                 return false;
356             }
357             break;
358         }
359 
360         default:
361             error_setg(errp, "The specification of stats-intervals is invalid");
362             return false;
363         }
364     }
365     return true;
366 }
367 
368 typedef enum { MEDIA_DISK, MEDIA_CDROM } DriveMediaType;
369 
370 /* All parameters but @opts are optional and may be set to NULL. */
371 static void extract_common_blockdev_options(QemuOpts *opts, int *bdrv_flags,
372     const char **throttling_group, ThrottleConfig *throttle_cfg,
373     BlockdevDetectZeroesOptions *detect_zeroes, Error **errp)
374 {
375     Error *local_error = NULL;
376     const char *aio;
377 
378     if (bdrv_flags) {
379         if (qemu_opt_get_bool(opts, "copy-on-read", false)) {
380             *bdrv_flags |= BDRV_O_COPY_ON_READ;
381         }
382 
383         if ((aio = qemu_opt_get(opts, "aio")) != NULL) {
384             if (!strcmp(aio, "native")) {
385                 *bdrv_flags |= BDRV_O_NATIVE_AIO;
386             } else if (!strcmp(aio, "threads")) {
387                 /* this is the default */
388             } else {
389                error_setg(errp, "invalid aio option");
390                return;
391             }
392         }
393     }
394 
395     /* disk I/O throttling */
396     if (throttling_group) {
397         *throttling_group = qemu_opt_get(opts, "throttling.group");
398     }
399 
400     if (throttle_cfg) {
401         throttle_config_init(throttle_cfg);
402         throttle_cfg->buckets[THROTTLE_BPS_TOTAL].avg =
403             qemu_opt_get_number(opts, "throttling.bps-total", 0);
404         throttle_cfg->buckets[THROTTLE_BPS_READ].avg  =
405             qemu_opt_get_number(opts, "throttling.bps-read", 0);
406         throttle_cfg->buckets[THROTTLE_BPS_WRITE].avg =
407             qemu_opt_get_number(opts, "throttling.bps-write", 0);
408         throttle_cfg->buckets[THROTTLE_OPS_TOTAL].avg =
409             qemu_opt_get_number(opts, "throttling.iops-total", 0);
410         throttle_cfg->buckets[THROTTLE_OPS_READ].avg =
411             qemu_opt_get_number(opts, "throttling.iops-read", 0);
412         throttle_cfg->buckets[THROTTLE_OPS_WRITE].avg =
413             qemu_opt_get_number(opts, "throttling.iops-write", 0);
414 
415         throttle_cfg->buckets[THROTTLE_BPS_TOTAL].max =
416             qemu_opt_get_number(opts, "throttling.bps-total-max", 0);
417         throttle_cfg->buckets[THROTTLE_BPS_READ].max  =
418             qemu_opt_get_number(opts, "throttling.bps-read-max", 0);
419         throttle_cfg->buckets[THROTTLE_BPS_WRITE].max =
420             qemu_opt_get_number(opts, "throttling.bps-write-max", 0);
421         throttle_cfg->buckets[THROTTLE_OPS_TOTAL].max =
422             qemu_opt_get_number(opts, "throttling.iops-total-max", 0);
423         throttle_cfg->buckets[THROTTLE_OPS_READ].max =
424             qemu_opt_get_number(opts, "throttling.iops-read-max", 0);
425         throttle_cfg->buckets[THROTTLE_OPS_WRITE].max =
426             qemu_opt_get_number(opts, "throttling.iops-write-max", 0);
427 
428         throttle_cfg->buckets[THROTTLE_BPS_TOTAL].burst_length =
429             qemu_opt_get_number(opts, "throttling.bps-total-max-length", 1);
430         throttle_cfg->buckets[THROTTLE_BPS_READ].burst_length  =
431             qemu_opt_get_number(opts, "throttling.bps-read-max-length", 1);
432         throttle_cfg->buckets[THROTTLE_BPS_WRITE].burst_length =
433             qemu_opt_get_number(opts, "throttling.bps-write-max-length", 1);
434         throttle_cfg->buckets[THROTTLE_OPS_TOTAL].burst_length =
435             qemu_opt_get_number(opts, "throttling.iops-total-max-length", 1);
436         throttle_cfg->buckets[THROTTLE_OPS_READ].burst_length =
437             qemu_opt_get_number(opts, "throttling.iops-read-max-length", 1);
438         throttle_cfg->buckets[THROTTLE_OPS_WRITE].burst_length =
439             qemu_opt_get_number(opts, "throttling.iops-write-max-length", 1);
440 
441         throttle_cfg->op_size =
442             qemu_opt_get_number(opts, "throttling.iops-size", 0);
443 
444         if (!throttle_is_valid(throttle_cfg, errp)) {
445             return;
446         }
447     }
448 
449     if (detect_zeroes) {
450         *detect_zeroes =
451             qapi_enum_parse(&BlockdevDetectZeroesOptions_lookup,
452                             qemu_opt_get(opts, "detect-zeroes"),
453                             BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF,
454                             &local_error);
455         if (local_error) {
456             error_propagate(errp, local_error);
457             return;
458         }
459     }
460 }
461 
462 /* Takes the ownership of bs_opts */
463 static BlockBackend *blockdev_init(const char *file, QDict *bs_opts,
464                                    Error **errp)
465 {
466     const char *buf;
467     int bdrv_flags = 0;
468     int on_read_error, on_write_error;
469     bool account_invalid, account_failed;
470     bool writethrough, read_only;
471     BlockBackend *blk;
472     BlockDriverState *bs;
473     ThrottleConfig cfg;
474     int snapshot = 0;
475     Error *error = NULL;
476     QemuOpts *opts;
477     QDict *interval_dict = NULL;
478     QList *interval_list = NULL;
479     const char *id;
480     BlockdevDetectZeroesOptions detect_zeroes =
481         BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF;
482     const char *throttling_group = NULL;
483 
484     /* Check common options by copying from bs_opts to opts, all other options
485      * stay in bs_opts for processing by bdrv_open(). */
486     id = qdict_get_try_str(bs_opts, "id");
487     opts = qemu_opts_create(&qemu_common_drive_opts, id, 1, &error);
488     if (error) {
489         error_propagate(errp, error);
490         goto err_no_opts;
491     }
492 
493     qemu_opts_absorb_qdict(opts, bs_opts, &error);
494     if (error) {
495         error_propagate(errp, error);
496         goto early_err;
497     }
498 
499     if (id) {
500         qdict_del(bs_opts, "id");
501     }
502 
503     /* extract parameters */
504     snapshot = qemu_opt_get_bool(opts, "snapshot", 0);
505 
506     account_invalid = qemu_opt_get_bool(opts, "stats-account-invalid", true);
507     account_failed = qemu_opt_get_bool(opts, "stats-account-failed", true);
508 
509     writethrough = !qemu_opt_get_bool(opts, BDRV_OPT_CACHE_WB, true);
510 
511     id = qemu_opts_id(opts);
512 
513     qdict_extract_subqdict(bs_opts, &interval_dict, "stats-intervals.");
514     qdict_array_split(interval_dict, &interval_list);
515 
516     if (qdict_size(interval_dict) != 0) {
517         error_setg(errp, "Invalid option stats-intervals.%s",
518                    qdict_first(interval_dict)->key);
519         goto early_err;
520     }
521 
522     extract_common_blockdev_options(opts, &bdrv_flags, &throttling_group, &cfg,
523                                     &detect_zeroes, &error);
524     if (error) {
525         error_propagate(errp, error);
526         goto early_err;
527     }
528 
529     if ((buf = qemu_opt_get(opts, "format")) != NULL) {
530         if (is_help_option(buf)) {
531             error_printf("Supported formats:");
532             bdrv_iterate_format(bdrv_format_print, NULL);
533             error_printf("\n");
534             goto early_err;
535         }
536 
537         if (qdict_haskey(bs_opts, "driver")) {
538             error_setg(errp, "Cannot specify both 'driver' and 'format'");
539             goto early_err;
540         }
541         qdict_put_str(bs_opts, "driver", buf);
542     }
543 
544     on_write_error = BLOCKDEV_ON_ERROR_ENOSPC;
545     if ((buf = qemu_opt_get(opts, "werror")) != NULL) {
546         on_write_error = parse_block_error_action(buf, 0, &error);
547         if (error) {
548             error_propagate(errp, error);
549             goto early_err;
550         }
551     }
552 
553     on_read_error = BLOCKDEV_ON_ERROR_REPORT;
554     if ((buf = qemu_opt_get(opts, "rerror")) != NULL) {
555         on_read_error = parse_block_error_action(buf, 1, &error);
556         if (error) {
557             error_propagate(errp, error);
558             goto early_err;
559         }
560     }
561 
562     if (snapshot) {
563         bdrv_flags |= BDRV_O_SNAPSHOT;
564     }
565 
566     read_only = qemu_opt_get_bool(opts, BDRV_OPT_READ_ONLY, false);
567 
568     /* init */
569     if ((!file || !*file) && !qdict_size(bs_opts)) {
570         BlockBackendRootState *blk_rs;
571 
572         blk = blk_new(0, BLK_PERM_ALL);
573         blk_rs = blk_get_root_state(blk);
574         blk_rs->open_flags    = bdrv_flags;
575         blk_rs->read_only     = read_only;
576         blk_rs->detect_zeroes = detect_zeroes;
577 
578         QDECREF(bs_opts);
579     } else {
580         if (file && !*file) {
581             file = NULL;
582         }
583 
584         /* bdrv_open() defaults to the values in bdrv_flags (for compatibility
585          * with other callers) rather than what we want as the real defaults.
586          * Apply the defaults here instead. */
587         qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_DIRECT, "off");
588         qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_NO_FLUSH, "off");
589         qdict_set_default_str(bs_opts, BDRV_OPT_READ_ONLY,
590                               read_only ? "on" : "off");
591         assert((bdrv_flags & BDRV_O_CACHE_MASK) == 0);
592 
593         if (runstate_check(RUN_STATE_INMIGRATE)) {
594             bdrv_flags |= BDRV_O_INACTIVE;
595         }
596 
597         blk = blk_new_open(file, NULL, bs_opts, bdrv_flags, errp);
598         if (!blk) {
599             goto err_no_bs_opts;
600         }
601         bs = blk_bs(blk);
602 
603         bs->detect_zeroes = detect_zeroes;
604 
605         block_acct_setup(blk_get_stats(blk), account_invalid, account_failed);
606 
607         if (!parse_stats_intervals(blk_get_stats(blk), interval_list, errp)) {
608             blk_unref(blk);
609             blk = NULL;
610             goto err_no_bs_opts;
611         }
612     }
613 
614     /* disk I/O throttling */
615     if (throttle_enabled(&cfg)) {
616         if (!throttling_group) {
617             throttling_group = id;
618         }
619         blk_io_limits_enable(blk, throttling_group);
620         blk_set_io_limits(blk, &cfg);
621     }
622 
623     blk_set_enable_write_cache(blk, !writethrough);
624     blk_set_on_error(blk, on_read_error, on_write_error);
625 
626     if (!monitor_add_blk(blk, id, errp)) {
627         blk_unref(blk);
628         blk = NULL;
629         goto err_no_bs_opts;
630     }
631 
632 err_no_bs_opts:
633     qemu_opts_del(opts);
634     QDECREF(interval_dict);
635     QDECREF(interval_list);
636     return blk;
637 
638 early_err:
639     qemu_opts_del(opts);
640     QDECREF(interval_dict);
641     QDECREF(interval_list);
642 err_no_opts:
643     QDECREF(bs_opts);
644     return NULL;
645 }
646 
647 /* Takes the ownership of bs_opts */
648 static BlockDriverState *bds_tree_init(QDict *bs_opts, Error **errp)
649 {
650     int bdrv_flags = 0;
651 
652     /* bdrv_open() defaults to the values in bdrv_flags (for compatibility
653      * with other callers) rather than what we want as the real defaults.
654      * Apply the defaults here instead. */
655     qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_DIRECT, "off");
656     qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_NO_FLUSH, "off");
657     qdict_set_default_str(bs_opts, BDRV_OPT_READ_ONLY, "off");
658 
659     if (runstate_check(RUN_STATE_INMIGRATE)) {
660         bdrv_flags |= BDRV_O_INACTIVE;
661     }
662 
663     return bdrv_open(NULL, NULL, bs_opts, bdrv_flags, errp);
664 }
665 
666 void blockdev_close_all_bdrv_states(void)
667 {
668     BlockDriverState *bs, *next_bs;
669 
670     QTAILQ_FOREACH_SAFE(bs, &monitor_bdrv_states, monitor_list, next_bs) {
671         AioContext *ctx = bdrv_get_aio_context(bs);
672 
673         aio_context_acquire(ctx);
674         bdrv_unref(bs);
675         aio_context_release(ctx);
676     }
677 }
678 
679 /* Iterates over the list of monitor-owned BlockDriverStates */
680 BlockDriverState *bdrv_next_monitor_owned(BlockDriverState *bs)
681 {
682     return bs ? QTAILQ_NEXT(bs, monitor_list)
683               : QTAILQ_FIRST(&monitor_bdrv_states);
684 }
685 
686 static void qemu_opt_rename(QemuOpts *opts, const char *from, const char *to,
687                             Error **errp)
688 {
689     const char *value;
690 
691     value = qemu_opt_get(opts, from);
692     if (value) {
693         if (qemu_opt_find(opts, to)) {
694             error_setg(errp, "'%s' and its alias '%s' can't be used at the "
695                        "same time", to, from);
696             return;
697         }
698     }
699 
700     /* rename all items in opts */
701     while ((value = qemu_opt_get(opts, from))) {
702         qemu_opt_set(opts, to, value, &error_abort);
703         qemu_opt_unset(opts, from);
704     }
705 }
706 
707 QemuOptsList qemu_legacy_drive_opts = {
708     .name = "drive",
709     .head = QTAILQ_HEAD_INITIALIZER(qemu_legacy_drive_opts.head),
710     .desc = {
711         {
712             .name = "bus",
713             .type = QEMU_OPT_NUMBER,
714             .help = "bus number",
715         },{
716             .name = "unit",
717             .type = QEMU_OPT_NUMBER,
718             .help = "unit number (i.e. lun for scsi)",
719         },{
720             .name = "index",
721             .type = QEMU_OPT_NUMBER,
722             .help = "index number",
723         },{
724             .name = "media",
725             .type = QEMU_OPT_STRING,
726             .help = "media type (disk, cdrom)",
727         },{
728             .name = "if",
729             .type = QEMU_OPT_STRING,
730             .help = "interface (ide, scsi, sd, mtd, floppy, pflash, virtio)",
731         },{
732             .name = "cyls",
733             .type = QEMU_OPT_NUMBER,
734             .help = "number of cylinders (ide disk geometry)",
735         },{
736             .name = "heads",
737             .type = QEMU_OPT_NUMBER,
738             .help = "number of heads (ide disk geometry)",
739         },{
740             .name = "secs",
741             .type = QEMU_OPT_NUMBER,
742             .help = "number of sectors (ide disk geometry)",
743         },{
744             .name = "trans",
745             .type = QEMU_OPT_STRING,
746             .help = "chs translation (auto, lba, none)",
747         },{
748             .name = "addr",
749             .type = QEMU_OPT_STRING,
750             .help = "pci address (virtio only)",
751         },{
752             .name = "serial",
753             .type = QEMU_OPT_STRING,
754             .help = "disk serial number",
755         },{
756             .name = "file",
757             .type = QEMU_OPT_STRING,
758             .help = "file name",
759         },
760 
761         /* Options that are passed on, but have special semantics with -drive */
762         {
763             .name = BDRV_OPT_READ_ONLY,
764             .type = QEMU_OPT_BOOL,
765             .help = "open drive file as read-only",
766         },{
767             .name = "rerror",
768             .type = QEMU_OPT_STRING,
769             .help = "read error action",
770         },{
771             .name = "werror",
772             .type = QEMU_OPT_STRING,
773             .help = "write error action",
774         },{
775             .name = "copy-on-read",
776             .type = QEMU_OPT_BOOL,
777             .help = "copy read data from backing file into image file",
778         },
779 
780         { /* end of list */ }
781     },
782 };
783 
784 DriveInfo *drive_new(QemuOpts *all_opts, BlockInterfaceType block_default_type)
785 {
786     const char *value;
787     BlockBackend *blk;
788     DriveInfo *dinfo = NULL;
789     QDict *bs_opts;
790     QemuOpts *legacy_opts;
791     DriveMediaType media = MEDIA_DISK;
792     BlockInterfaceType type;
793     int cyls, heads, secs, translation;
794     int max_devs, bus_id, unit_id, index;
795     const char *devaddr;
796     const char *werror, *rerror;
797     bool read_only = false;
798     bool copy_on_read;
799     const char *serial;
800     const char *filename;
801     Error *local_err = NULL;
802     int i;
803     const char *deprecated[] = {
804         "serial", "trans", "secs", "heads", "cyls", "addr"
805     };
806 
807     /* Change legacy command line options into QMP ones */
808     static const struct {
809         const char *from;
810         const char *to;
811     } opt_renames[] = {
812         { "iops",           "throttling.iops-total" },
813         { "iops_rd",        "throttling.iops-read" },
814         { "iops_wr",        "throttling.iops-write" },
815 
816         { "bps",            "throttling.bps-total" },
817         { "bps_rd",         "throttling.bps-read" },
818         { "bps_wr",         "throttling.bps-write" },
819 
820         { "iops_max",       "throttling.iops-total-max" },
821         { "iops_rd_max",    "throttling.iops-read-max" },
822         { "iops_wr_max",    "throttling.iops-write-max" },
823 
824         { "bps_max",        "throttling.bps-total-max" },
825         { "bps_rd_max",     "throttling.bps-read-max" },
826         { "bps_wr_max",     "throttling.bps-write-max" },
827 
828         { "iops_size",      "throttling.iops-size" },
829 
830         { "group",          "throttling.group" },
831 
832         { "readonly",       BDRV_OPT_READ_ONLY },
833     };
834 
835     for (i = 0; i < ARRAY_SIZE(opt_renames); i++) {
836         qemu_opt_rename(all_opts, opt_renames[i].from, opt_renames[i].to,
837                         &local_err);
838         if (local_err) {
839             error_report_err(local_err);
840             return NULL;
841         }
842     }
843 
844     value = qemu_opt_get(all_opts, "cache");
845     if (value) {
846         int flags = 0;
847         bool writethrough;
848 
849         if (bdrv_parse_cache_mode(value, &flags, &writethrough) != 0) {
850             error_report("invalid cache option");
851             return NULL;
852         }
853 
854         /* Specific options take precedence */
855         if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_WB)) {
856             qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_WB,
857                               !writethrough, &error_abort);
858         }
859         if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_DIRECT)) {
860             qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_DIRECT,
861                               !!(flags & BDRV_O_NOCACHE), &error_abort);
862         }
863         if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_NO_FLUSH)) {
864             qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_NO_FLUSH,
865                               !!(flags & BDRV_O_NO_FLUSH), &error_abort);
866         }
867         qemu_opt_unset(all_opts, "cache");
868     }
869 
870     /* Get a QDict for processing the options */
871     bs_opts = qdict_new();
872     qemu_opts_to_qdict(all_opts, bs_opts);
873 
874     legacy_opts = qemu_opts_create(&qemu_legacy_drive_opts, NULL, 0,
875                                    &error_abort);
876     qemu_opts_absorb_qdict(legacy_opts, bs_opts, &local_err);
877     if (local_err) {
878         error_report_err(local_err);
879         goto fail;
880     }
881 
882     /* Other deprecated options */
883     if (!qtest_enabled()) {
884         for (i = 0; i < ARRAY_SIZE(deprecated); i++) {
885             if (qemu_opt_get(legacy_opts, deprecated[i]) != NULL) {
886                 error_report("'%s' is deprecated, please use the corresponding "
887                              "option of '-device' instead", deprecated[i]);
888             }
889         }
890     }
891 
892     /* Media type */
893     value = qemu_opt_get(legacy_opts, "media");
894     if (value) {
895         if (!strcmp(value, "disk")) {
896             media = MEDIA_DISK;
897         } else if (!strcmp(value, "cdrom")) {
898             media = MEDIA_CDROM;
899             read_only = true;
900         } else {
901             error_report("'%s' invalid media", value);
902             goto fail;
903         }
904     }
905 
906     /* copy-on-read is disabled with a warning for read-only devices */
907     read_only |= qemu_opt_get_bool(legacy_opts, BDRV_OPT_READ_ONLY, false);
908     copy_on_read = qemu_opt_get_bool(legacy_opts, "copy-on-read", false);
909 
910     if (read_only && copy_on_read) {
911         warn_report("disabling copy-on-read on read-only drive");
912         copy_on_read = false;
913     }
914 
915     qdict_put_str(bs_opts, BDRV_OPT_READ_ONLY, read_only ? "on" : "off");
916     qdict_put_str(bs_opts, "copy-on-read", copy_on_read ? "on" : "off");
917 
918     /* Controller type */
919     value = qemu_opt_get(legacy_opts, "if");
920     if (value) {
921         for (type = 0;
922              type < IF_COUNT && strcmp(value, if_name[type]);
923              type++) {
924         }
925         if (type == IF_COUNT) {
926             error_report("unsupported bus type '%s'", value);
927             goto fail;
928         }
929     } else {
930         type = block_default_type;
931     }
932 
933     /* Geometry */
934     cyls  = qemu_opt_get_number(legacy_opts, "cyls", 0);
935     heads = qemu_opt_get_number(legacy_opts, "heads", 0);
936     secs  = qemu_opt_get_number(legacy_opts, "secs", 0);
937 
938     if (cyls || heads || secs) {
939         if (cyls < 1) {
940             error_report("invalid physical cyls number");
941             goto fail;
942         }
943         if (heads < 1) {
944             error_report("invalid physical heads number");
945             goto fail;
946         }
947         if (secs < 1) {
948             error_report("invalid physical secs number");
949             goto fail;
950         }
951     }
952 
953     translation = BIOS_ATA_TRANSLATION_AUTO;
954     value = qemu_opt_get(legacy_opts, "trans");
955     if (value != NULL) {
956         if (!cyls) {
957             error_report("'%s' trans must be used with cyls, heads and secs",
958                          value);
959             goto fail;
960         }
961         if (!strcmp(value, "none")) {
962             translation = BIOS_ATA_TRANSLATION_NONE;
963         } else if (!strcmp(value, "lba")) {
964             translation = BIOS_ATA_TRANSLATION_LBA;
965         } else if (!strcmp(value, "large")) {
966             translation = BIOS_ATA_TRANSLATION_LARGE;
967         } else if (!strcmp(value, "rechs")) {
968             translation = BIOS_ATA_TRANSLATION_RECHS;
969         } else if (!strcmp(value, "auto")) {
970             translation = BIOS_ATA_TRANSLATION_AUTO;
971         } else {
972             error_report("'%s' invalid translation type", value);
973             goto fail;
974         }
975     }
976 
977     if (media == MEDIA_CDROM) {
978         if (cyls || secs || heads) {
979             error_report("CHS can't be set with media=cdrom");
980             goto fail;
981         }
982     }
983 
984     /* Device address specified by bus/unit or index.
985      * If none was specified, try to find the first free one. */
986     bus_id  = qemu_opt_get_number(legacy_opts, "bus", 0);
987     unit_id = qemu_opt_get_number(legacy_opts, "unit", -1);
988     index   = qemu_opt_get_number(legacy_opts, "index", -1);
989 
990     max_devs = if_max_devs[type];
991 
992     if (index != -1) {
993         if (bus_id != 0 || unit_id != -1) {
994             error_report("index cannot be used with bus and unit");
995             goto fail;
996         }
997         bus_id = drive_index_to_bus_id(type, index);
998         unit_id = drive_index_to_unit_id(type, index);
999     }
1000 
1001     if (unit_id == -1) {
1002        unit_id = 0;
1003        while (drive_get(type, bus_id, unit_id) != NULL) {
1004            unit_id++;
1005            if (max_devs && unit_id >= max_devs) {
1006                unit_id -= max_devs;
1007                bus_id++;
1008            }
1009        }
1010     }
1011 
1012     if (max_devs && unit_id >= max_devs) {
1013         error_report("unit %d too big (max is %d)", unit_id, max_devs - 1);
1014         goto fail;
1015     }
1016 
1017     if (drive_get(type, bus_id, unit_id) != NULL) {
1018         error_report("drive with bus=%d, unit=%d (index=%d) exists",
1019                      bus_id, unit_id, index);
1020         goto fail;
1021     }
1022 
1023     /* Serial number */
1024     serial = qemu_opt_get(legacy_opts, "serial");
1025 
1026     /* no id supplied -> create one */
1027     if (qemu_opts_id(all_opts) == NULL) {
1028         char *new_id;
1029         const char *mediastr = "";
1030         if (type == IF_IDE || type == IF_SCSI) {
1031             mediastr = (media == MEDIA_CDROM) ? "-cd" : "-hd";
1032         }
1033         if (max_devs) {
1034             new_id = g_strdup_printf("%s%i%s%i", if_name[type], bus_id,
1035                                      mediastr, unit_id);
1036         } else {
1037             new_id = g_strdup_printf("%s%s%i", if_name[type],
1038                                      mediastr, unit_id);
1039         }
1040         qdict_put_str(bs_opts, "id", new_id);
1041         g_free(new_id);
1042     }
1043 
1044     /* Add virtio block device */
1045     devaddr = qemu_opt_get(legacy_opts, "addr");
1046     if (devaddr && type != IF_VIRTIO) {
1047         error_report("addr is not supported by this bus type");
1048         goto fail;
1049     }
1050 
1051     if (type == IF_VIRTIO) {
1052         QemuOpts *devopts;
1053         devopts = qemu_opts_create(qemu_find_opts("device"), NULL, 0,
1054                                    &error_abort);
1055         if (arch_type == QEMU_ARCH_S390X) {
1056             qemu_opt_set(devopts, "driver", "virtio-blk-ccw", &error_abort);
1057         } else {
1058             qemu_opt_set(devopts, "driver", "virtio-blk-pci", &error_abort);
1059         }
1060         qemu_opt_set(devopts, "drive", qdict_get_str(bs_opts, "id"),
1061                      &error_abort);
1062         if (devaddr) {
1063             qemu_opt_set(devopts, "addr", devaddr, &error_abort);
1064         }
1065     }
1066 
1067     filename = qemu_opt_get(legacy_opts, "file");
1068 
1069     /* Check werror/rerror compatibility with if=... */
1070     werror = qemu_opt_get(legacy_opts, "werror");
1071     if (werror != NULL) {
1072         if (type != IF_IDE && type != IF_SCSI && type != IF_VIRTIO &&
1073             type != IF_NONE) {
1074             error_report("werror is not supported by this bus type");
1075             goto fail;
1076         }
1077         qdict_put_str(bs_opts, "werror", werror);
1078     }
1079 
1080     rerror = qemu_opt_get(legacy_opts, "rerror");
1081     if (rerror != NULL) {
1082         if (type != IF_IDE && type != IF_VIRTIO && type != IF_SCSI &&
1083             type != IF_NONE) {
1084             error_report("rerror is not supported by this bus type");
1085             goto fail;
1086         }
1087         qdict_put_str(bs_opts, "rerror", rerror);
1088     }
1089 
1090     /* Actual block device init: Functionality shared with blockdev-add */
1091     blk = blockdev_init(filename, bs_opts, &local_err);
1092     bs_opts = NULL;
1093     if (!blk) {
1094         if (local_err) {
1095             error_report_err(local_err);
1096         }
1097         goto fail;
1098     } else {
1099         assert(!local_err);
1100     }
1101 
1102     /* Create legacy DriveInfo */
1103     dinfo = g_malloc0(sizeof(*dinfo));
1104     dinfo->opts = all_opts;
1105 
1106     dinfo->cyls = cyls;
1107     dinfo->heads = heads;
1108     dinfo->secs = secs;
1109     dinfo->trans = translation;
1110 
1111     dinfo->type = type;
1112     dinfo->bus = bus_id;
1113     dinfo->unit = unit_id;
1114     dinfo->devaddr = devaddr;
1115     dinfo->serial = g_strdup(serial);
1116 
1117     blk_set_legacy_dinfo(blk, dinfo);
1118 
1119     switch(type) {
1120     case IF_IDE:
1121     case IF_SCSI:
1122     case IF_XEN:
1123     case IF_NONE:
1124         dinfo->media_cd = media == MEDIA_CDROM;
1125         break;
1126     default:
1127         break;
1128     }
1129 
1130 fail:
1131     qemu_opts_del(legacy_opts);
1132     QDECREF(bs_opts);
1133     return dinfo;
1134 }
1135 
1136 static BlockDriverState *qmp_get_root_bs(const char *name, Error **errp)
1137 {
1138     BlockDriverState *bs;
1139 
1140     bs = bdrv_lookup_bs(name, name, errp);
1141     if (bs == NULL) {
1142         return NULL;
1143     }
1144 
1145     if (!bdrv_is_root_node(bs)) {
1146         error_setg(errp, "Need a root block node");
1147         return NULL;
1148     }
1149 
1150     if (!bdrv_is_inserted(bs)) {
1151         error_setg(errp, "Device has no medium");
1152         return NULL;
1153     }
1154 
1155     return bs;
1156 }
1157 
1158 static BlockBackend *qmp_get_blk(const char *blk_name, const char *qdev_id,
1159                                  Error **errp)
1160 {
1161     BlockBackend *blk;
1162 
1163     if (!blk_name == !qdev_id) {
1164         error_setg(errp, "Need exactly one of 'device' and 'id'");
1165         return NULL;
1166     }
1167 
1168     if (qdev_id) {
1169         blk = blk_by_qdev_id(qdev_id, errp);
1170     } else {
1171         blk = blk_by_name(blk_name);
1172         if (blk == NULL) {
1173             error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
1174                       "Device '%s' not found", blk_name);
1175         }
1176     }
1177 
1178     return blk;
1179 }
1180 
1181 void hmp_commit(Monitor *mon, const QDict *qdict)
1182 {
1183     const char *device = qdict_get_str(qdict, "device");
1184     BlockBackend *blk;
1185     int ret;
1186 
1187     if (!strcmp(device, "all")) {
1188         ret = blk_commit_all();
1189     } else {
1190         BlockDriverState *bs;
1191         AioContext *aio_context;
1192 
1193         blk = blk_by_name(device);
1194         if (!blk) {
1195             monitor_printf(mon, "Device '%s' not found\n", device);
1196             return;
1197         }
1198         if (!blk_is_available(blk)) {
1199             monitor_printf(mon, "Device '%s' has no medium\n", device);
1200             return;
1201         }
1202 
1203         bs = blk_bs(blk);
1204         aio_context = bdrv_get_aio_context(bs);
1205         aio_context_acquire(aio_context);
1206 
1207         ret = bdrv_commit(bs);
1208 
1209         aio_context_release(aio_context);
1210     }
1211     if (ret < 0) {
1212         monitor_printf(mon, "'commit' error for '%s': %s\n", device,
1213                        strerror(-ret));
1214     }
1215 }
1216 
1217 static void blockdev_do_action(TransactionAction *action, Error **errp)
1218 {
1219     TransactionActionList list;
1220 
1221     list.value = action;
1222     list.next = NULL;
1223     qmp_transaction(&list, false, NULL, errp);
1224 }
1225 
1226 void qmp_blockdev_snapshot_sync(bool has_device, const char *device,
1227                                 bool has_node_name, const char *node_name,
1228                                 const char *snapshot_file,
1229                                 bool has_snapshot_node_name,
1230                                 const char *snapshot_node_name,
1231                                 bool has_format, const char *format,
1232                                 bool has_mode, NewImageMode mode, Error **errp)
1233 {
1234     BlockdevSnapshotSync snapshot = {
1235         .has_device = has_device,
1236         .device = (char *) device,
1237         .has_node_name = has_node_name,
1238         .node_name = (char *) node_name,
1239         .snapshot_file = (char *) snapshot_file,
1240         .has_snapshot_node_name = has_snapshot_node_name,
1241         .snapshot_node_name = (char *) snapshot_node_name,
1242         .has_format = has_format,
1243         .format = (char *) format,
1244         .has_mode = has_mode,
1245         .mode = mode,
1246     };
1247     TransactionAction action = {
1248         .type = TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC,
1249         .u.blockdev_snapshot_sync.data = &snapshot,
1250     };
1251     blockdev_do_action(&action, errp);
1252 }
1253 
1254 void qmp_blockdev_snapshot(const char *node, const char *overlay,
1255                            Error **errp)
1256 {
1257     BlockdevSnapshot snapshot_data = {
1258         .node = (char *) node,
1259         .overlay = (char *) overlay
1260     };
1261     TransactionAction action = {
1262         .type = TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT,
1263         .u.blockdev_snapshot.data = &snapshot_data,
1264     };
1265     blockdev_do_action(&action, errp);
1266 }
1267 
1268 void qmp_blockdev_snapshot_internal_sync(const char *device,
1269                                          const char *name,
1270                                          Error **errp)
1271 {
1272     BlockdevSnapshotInternal snapshot = {
1273         .device = (char *) device,
1274         .name = (char *) name
1275     };
1276     TransactionAction action = {
1277         .type = TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC,
1278         .u.blockdev_snapshot_internal_sync.data = &snapshot,
1279     };
1280     blockdev_do_action(&action, errp);
1281 }
1282 
1283 SnapshotInfo *qmp_blockdev_snapshot_delete_internal_sync(const char *device,
1284                                                          bool has_id,
1285                                                          const char *id,
1286                                                          bool has_name,
1287                                                          const char *name,
1288                                                          Error **errp)
1289 {
1290     BlockDriverState *bs;
1291     AioContext *aio_context;
1292     QEMUSnapshotInfo sn;
1293     Error *local_err = NULL;
1294     SnapshotInfo *info = NULL;
1295     int ret;
1296 
1297     bs = qmp_get_root_bs(device, errp);
1298     if (!bs) {
1299         return NULL;
1300     }
1301     aio_context = bdrv_get_aio_context(bs);
1302     aio_context_acquire(aio_context);
1303 
1304     if (!has_id) {
1305         id = NULL;
1306     }
1307 
1308     if (!has_name) {
1309         name = NULL;
1310     }
1311 
1312     if (!id && !name) {
1313         error_setg(errp, "Name or id must be provided");
1314         goto out_aio_context;
1315     }
1316 
1317     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_INTERNAL_SNAPSHOT_DELETE, errp)) {
1318         goto out_aio_context;
1319     }
1320 
1321     ret = bdrv_snapshot_find_by_id_and_name(bs, id, name, &sn, &local_err);
1322     if (local_err) {
1323         error_propagate(errp, local_err);
1324         goto out_aio_context;
1325     }
1326     if (!ret) {
1327         error_setg(errp,
1328                    "Snapshot with id '%s' and name '%s' does not exist on "
1329                    "device '%s'",
1330                    STR_OR_NULL(id), STR_OR_NULL(name), device);
1331         goto out_aio_context;
1332     }
1333 
1334     bdrv_snapshot_delete(bs, id, name, &local_err);
1335     if (local_err) {
1336         error_propagate(errp, local_err);
1337         goto out_aio_context;
1338     }
1339 
1340     aio_context_release(aio_context);
1341 
1342     info = g_new0(SnapshotInfo, 1);
1343     info->id = g_strdup(sn.id_str);
1344     info->name = g_strdup(sn.name);
1345     info->date_nsec = sn.date_nsec;
1346     info->date_sec = sn.date_sec;
1347     info->vm_state_size = sn.vm_state_size;
1348     info->vm_clock_nsec = sn.vm_clock_nsec % 1000000000;
1349     info->vm_clock_sec = sn.vm_clock_nsec / 1000000000;
1350 
1351     return info;
1352 
1353 out_aio_context:
1354     aio_context_release(aio_context);
1355     return NULL;
1356 }
1357 
1358 /**
1359  * block_dirty_bitmap_lookup:
1360  * Return a dirty bitmap (if present), after validating
1361  * the node reference and bitmap names.
1362  *
1363  * @node: The name of the BDS node to search for bitmaps
1364  * @name: The name of the bitmap to search for
1365  * @pbs: Output pointer for BDS lookup, if desired. Can be NULL.
1366  * @paio: Output pointer for aio_context acquisition, if desired. Can be NULL.
1367  * @errp: Output pointer for error information. Can be NULL.
1368  *
1369  * @return: A bitmap object on success, or NULL on failure.
1370  */
1371 static BdrvDirtyBitmap *block_dirty_bitmap_lookup(const char *node,
1372                                                   const char *name,
1373                                                   BlockDriverState **pbs,
1374                                                   Error **errp)
1375 {
1376     BlockDriverState *bs;
1377     BdrvDirtyBitmap *bitmap;
1378 
1379     if (!node) {
1380         error_setg(errp, "Node cannot be NULL");
1381         return NULL;
1382     }
1383     if (!name) {
1384         error_setg(errp, "Bitmap name cannot be NULL");
1385         return NULL;
1386     }
1387     bs = bdrv_lookup_bs(node, node, NULL);
1388     if (!bs) {
1389         error_setg(errp, "Node '%s' not found", node);
1390         return NULL;
1391     }
1392 
1393     bitmap = bdrv_find_dirty_bitmap(bs, name);
1394     if (!bitmap) {
1395         error_setg(errp, "Dirty bitmap '%s' not found", name);
1396         return NULL;
1397     }
1398 
1399     if (pbs) {
1400         *pbs = bs;
1401     }
1402 
1403     return bitmap;
1404 }
1405 
1406 /* New and old BlockDriverState structs for atomic group operations */
1407 
1408 typedef struct BlkActionState BlkActionState;
1409 
1410 /**
1411  * BlkActionOps:
1412  * Table of operations that define an Action.
1413  *
1414  * @instance_size: Size of state struct, in bytes.
1415  * @prepare: Prepare the work, must NOT be NULL.
1416  * @commit: Commit the changes, can be NULL.
1417  * @abort: Abort the changes on fail, can be NULL.
1418  * @clean: Clean up resources after all transaction actions have called
1419  *         commit() or abort(). Can be NULL.
1420  *
1421  * Only prepare() may fail. In a single transaction, only one of commit() or
1422  * abort() will be called. clean() will always be called if it is present.
1423  */
1424 typedef struct BlkActionOps {
1425     size_t instance_size;
1426     void (*prepare)(BlkActionState *common, Error **errp);
1427     void (*commit)(BlkActionState *common);
1428     void (*abort)(BlkActionState *common);
1429     void (*clean)(BlkActionState *common);
1430 } BlkActionOps;
1431 
1432 /**
1433  * BlkActionState:
1434  * Describes one Action's state within a Transaction.
1435  *
1436  * @action: QAPI-defined enum identifying which Action to perform.
1437  * @ops: Table of ActionOps this Action can perform.
1438  * @block_job_txn: Transaction which this action belongs to.
1439  * @entry: List membership for all Actions in this Transaction.
1440  *
1441  * This structure must be arranged as first member in a subclassed type,
1442  * assuming that the compiler will also arrange it to the same offsets as the
1443  * base class.
1444  */
1445 struct BlkActionState {
1446     TransactionAction *action;
1447     const BlkActionOps *ops;
1448     BlockJobTxn *block_job_txn;
1449     TransactionProperties *txn_props;
1450     QSIMPLEQ_ENTRY(BlkActionState) entry;
1451 };
1452 
1453 /* internal snapshot private data */
1454 typedef struct InternalSnapshotState {
1455     BlkActionState common;
1456     BlockDriverState *bs;
1457     QEMUSnapshotInfo sn;
1458     bool created;
1459 } InternalSnapshotState;
1460 
1461 
1462 static int action_check_completion_mode(BlkActionState *s, Error **errp)
1463 {
1464     if (s->txn_props->completion_mode != ACTION_COMPLETION_MODE_INDIVIDUAL) {
1465         error_setg(errp,
1466                    "Action '%s' does not support Transaction property "
1467                    "completion-mode = %s",
1468                    TransactionActionKind_str(s->action->type),
1469                    ActionCompletionMode_str(s->txn_props->completion_mode));
1470         return -1;
1471     }
1472     return 0;
1473 }
1474 
1475 static void internal_snapshot_prepare(BlkActionState *common,
1476                                       Error **errp)
1477 {
1478     Error *local_err = NULL;
1479     const char *device;
1480     const char *name;
1481     BlockDriverState *bs;
1482     QEMUSnapshotInfo old_sn, *sn;
1483     bool ret;
1484     qemu_timeval tv;
1485     BlockdevSnapshotInternal *internal;
1486     InternalSnapshotState *state;
1487     AioContext *aio_context;
1488     int ret1;
1489 
1490     g_assert(common->action->type ==
1491              TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC);
1492     internal = common->action->u.blockdev_snapshot_internal_sync.data;
1493     state = DO_UPCAST(InternalSnapshotState, common, common);
1494 
1495     /* 1. parse input */
1496     device = internal->device;
1497     name = internal->name;
1498 
1499     /* 2. check for validation */
1500     if (action_check_completion_mode(common, errp) < 0) {
1501         return;
1502     }
1503 
1504     bs = qmp_get_root_bs(device, errp);
1505     if (!bs) {
1506         return;
1507     }
1508 
1509     aio_context = bdrv_get_aio_context(bs);
1510     aio_context_acquire(aio_context);
1511 
1512     state->bs = bs;
1513 
1514     /* Paired with .clean() */
1515     bdrv_drained_begin(bs);
1516 
1517     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_INTERNAL_SNAPSHOT, errp)) {
1518         goto out;
1519     }
1520 
1521     if (bdrv_is_read_only(bs)) {
1522         error_setg(errp, "Device '%s' is read only", device);
1523         goto out;
1524     }
1525 
1526     if (!bdrv_can_snapshot(bs)) {
1527         error_setg(errp, "Block format '%s' used by device '%s' "
1528                    "does not support internal snapshots",
1529                    bs->drv->format_name, device);
1530         goto out;
1531     }
1532 
1533     if (!strlen(name)) {
1534         error_setg(errp, "Name is empty");
1535         goto out;
1536     }
1537 
1538     /* check whether a snapshot with name exist */
1539     ret = bdrv_snapshot_find_by_id_and_name(bs, NULL, name, &old_sn,
1540                                             &local_err);
1541     if (local_err) {
1542         error_propagate(errp, local_err);
1543         goto out;
1544     } else if (ret) {
1545         error_setg(errp,
1546                    "Snapshot with name '%s' already exists on device '%s'",
1547                    name, device);
1548         goto out;
1549     }
1550 
1551     /* 3. take the snapshot */
1552     sn = &state->sn;
1553     pstrcpy(sn->name, sizeof(sn->name), name);
1554     qemu_gettimeofday(&tv);
1555     sn->date_sec = tv.tv_sec;
1556     sn->date_nsec = tv.tv_usec * 1000;
1557     sn->vm_clock_nsec = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL);
1558 
1559     ret1 = bdrv_snapshot_create(bs, sn);
1560     if (ret1 < 0) {
1561         error_setg_errno(errp, -ret1,
1562                          "Failed to create snapshot '%s' on device '%s'",
1563                          name, device);
1564         goto out;
1565     }
1566 
1567     /* 4. succeed, mark a snapshot is created */
1568     state->created = true;
1569 
1570 out:
1571     aio_context_release(aio_context);
1572 }
1573 
1574 static void internal_snapshot_abort(BlkActionState *common)
1575 {
1576     InternalSnapshotState *state =
1577                              DO_UPCAST(InternalSnapshotState, common, common);
1578     BlockDriverState *bs = state->bs;
1579     QEMUSnapshotInfo *sn = &state->sn;
1580     AioContext *aio_context;
1581     Error *local_error = NULL;
1582 
1583     if (!state->created) {
1584         return;
1585     }
1586 
1587     aio_context = bdrv_get_aio_context(state->bs);
1588     aio_context_acquire(aio_context);
1589 
1590     if (bdrv_snapshot_delete(bs, sn->id_str, sn->name, &local_error) < 0) {
1591         error_reportf_err(local_error,
1592                           "Failed to delete snapshot with id '%s' and "
1593                           "name '%s' on device '%s' in abort: ",
1594                           sn->id_str, sn->name,
1595                           bdrv_get_device_name(bs));
1596     }
1597 
1598     aio_context_release(aio_context);
1599 }
1600 
1601 static void internal_snapshot_clean(BlkActionState *common)
1602 {
1603     InternalSnapshotState *state = DO_UPCAST(InternalSnapshotState,
1604                                              common, common);
1605     AioContext *aio_context;
1606 
1607     if (!state->bs) {
1608         return;
1609     }
1610 
1611     aio_context = bdrv_get_aio_context(state->bs);
1612     aio_context_acquire(aio_context);
1613 
1614     bdrv_drained_end(state->bs);
1615 
1616     aio_context_release(aio_context);
1617 }
1618 
1619 /* external snapshot private data */
1620 typedef struct ExternalSnapshotState {
1621     BlkActionState common;
1622     BlockDriverState *old_bs;
1623     BlockDriverState *new_bs;
1624     bool overlay_appended;
1625 } ExternalSnapshotState;
1626 
1627 static void external_snapshot_prepare(BlkActionState *common,
1628                                       Error **errp)
1629 {
1630     int flags = 0;
1631     QDict *options = NULL;
1632     Error *local_err = NULL;
1633     /* Device and node name of the image to generate the snapshot from */
1634     const char *device;
1635     const char *node_name;
1636     /* Reference to the new image (for 'blockdev-snapshot') */
1637     const char *snapshot_ref;
1638     /* File name of the new image (for 'blockdev-snapshot-sync') */
1639     const char *new_image_file;
1640     ExternalSnapshotState *state =
1641                              DO_UPCAST(ExternalSnapshotState, common, common);
1642     TransactionAction *action = common->action;
1643     AioContext *aio_context;
1644 
1645     /* 'blockdev-snapshot' and 'blockdev-snapshot-sync' have similar
1646      * purpose but a different set of parameters */
1647     switch (action->type) {
1648     case TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT:
1649         {
1650             BlockdevSnapshot *s = action->u.blockdev_snapshot.data;
1651             device = s->node;
1652             node_name = s->node;
1653             new_image_file = NULL;
1654             snapshot_ref = s->overlay;
1655         }
1656         break;
1657     case TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC:
1658         {
1659             BlockdevSnapshotSync *s = action->u.blockdev_snapshot_sync.data;
1660             device = s->has_device ? s->device : NULL;
1661             node_name = s->has_node_name ? s->node_name : NULL;
1662             new_image_file = s->snapshot_file;
1663             snapshot_ref = NULL;
1664         }
1665         break;
1666     default:
1667         g_assert_not_reached();
1668     }
1669 
1670     /* start processing */
1671     if (action_check_completion_mode(common, errp) < 0) {
1672         return;
1673     }
1674 
1675     state->old_bs = bdrv_lookup_bs(device, node_name, errp);
1676     if (!state->old_bs) {
1677         return;
1678     }
1679 
1680     aio_context = bdrv_get_aio_context(state->old_bs);
1681     aio_context_acquire(aio_context);
1682 
1683     /* Paired with .clean() */
1684     bdrv_drained_begin(state->old_bs);
1685 
1686     if (!bdrv_is_inserted(state->old_bs)) {
1687         error_setg(errp, QERR_DEVICE_HAS_NO_MEDIUM, device);
1688         goto out;
1689     }
1690 
1691     if (bdrv_op_is_blocked(state->old_bs,
1692                            BLOCK_OP_TYPE_EXTERNAL_SNAPSHOT, errp)) {
1693         goto out;
1694     }
1695 
1696     if (!bdrv_is_read_only(state->old_bs)) {
1697         if (bdrv_flush(state->old_bs)) {
1698             error_setg(errp, QERR_IO_ERROR);
1699             goto out;
1700         }
1701     }
1702 
1703     if (!bdrv_is_first_non_filter(state->old_bs)) {
1704         error_setg(errp, QERR_FEATURE_DISABLED, "snapshot");
1705         goto out;
1706     }
1707 
1708     if (action->type == TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC) {
1709         BlockdevSnapshotSync *s = action->u.blockdev_snapshot_sync.data;
1710         const char *format = s->has_format ? s->format : "qcow2";
1711         enum NewImageMode mode;
1712         const char *snapshot_node_name =
1713             s->has_snapshot_node_name ? s->snapshot_node_name : NULL;
1714 
1715         if (node_name && !snapshot_node_name) {
1716             error_setg(errp, "New snapshot node name missing");
1717             goto out;
1718         }
1719 
1720         if (snapshot_node_name &&
1721             bdrv_lookup_bs(snapshot_node_name, snapshot_node_name, NULL)) {
1722             error_setg(errp, "New snapshot node name already in use");
1723             goto out;
1724         }
1725 
1726         flags = state->old_bs->open_flags;
1727         flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_COPY_ON_READ);
1728         flags |= BDRV_O_NO_BACKING;
1729 
1730         /* create new image w/backing file */
1731         mode = s->has_mode ? s->mode : NEW_IMAGE_MODE_ABSOLUTE_PATHS;
1732         if (mode != NEW_IMAGE_MODE_EXISTING) {
1733             int64_t size = bdrv_getlength(state->old_bs);
1734             if (size < 0) {
1735                 error_setg_errno(errp, -size, "bdrv_getlength failed");
1736                 goto out;
1737             }
1738             bdrv_img_create(new_image_file, format,
1739                             state->old_bs->filename,
1740                             state->old_bs->drv->format_name,
1741                             NULL, size, flags, false, &local_err);
1742             if (local_err) {
1743                 error_propagate(errp, local_err);
1744                 goto out;
1745             }
1746         }
1747 
1748         options = qdict_new();
1749         if (s->has_snapshot_node_name) {
1750             qdict_put_str(options, "node-name", snapshot_node_name);
1751         }
1752         qdict_put_str(options, "driver", format);
1753     }
1754 
1755     state->new_bs = bdrv_open(new_image_file, snapshot_ref, options, flags,
1756                               errp);
1757     /* We will manually add the backing_hd field to the bs later */
1758     if (!state->new_bs) {
1759         goto out;
1760     }
1761 
1762     if (bdrv_has_blk(state->new_bs)) {
1763         error_setg(errp, "The snapshot is already in use");
1764         goto out;
1765     }
1766 
1767     if (bdrv_op_is_blocked(state->new_bs, BLOCK_OP_TYPE_EXTERNAL_SNAPSHOT,
1768                            errp)) {
1769         goto out;
1770     }
1771 
1772     if (state->new_bs->backing != NULL) {
1773         error_setg(errp, "The snapshot already has a backing image");
1774         goto out;
1775     }
1776 
1777     if (!state->new_bs->drv->supports_backing) {
1778         error_setg(errp, "The snapshot does not support backing images");
1779         goto out;
1780     }
1781 
1782     bdrv_set_aio_context(state->new_bs, aio_context);
1783 
1784     /* This removes our old bs and adds the new bs. This is an operation that
1785      * can fail, so we need to do it in .prepare; undoing it for abort is
1786      * always possible. */
1787     bdrv_ref(state->new_bs);
1788     bdrv_append(state->new_bs, state->old_bs, &local_err);
1789     if (local_err) {
1790         error_propagate(errp, local_err);
1791         goto out;
1792     }
1793     state->overlay_appended = true;
1794 
1795 out:
1796     aio_context_release(aio_context);
1797 }
1798 
1799 static void external_snapshot_commit(BlkActionState *common)
1800 {
1801     ExternalSnapshotState *state =
1802                              DO_UPCAST(ExternalSnapshotState, common, common);
1803     AioContext *aio_context;
1804 
1805     aio_context = bdrv_get_aio_context(state->old_bs);
1806     aio_context_acquire(aio_context);
1807 
1808     /* We don't need (or want) to use the transactional
1809      * bdrv_reopen_multiple() across all the entries at once, because we
1810      * don't want to abort all of them if one of them fails the reopen */
1811     if (!atomic_read(&state->old_bs->copy_on_read)) {
1812         bdrv_reopen(state->old_bs, state->old_bs->open_flags & ~BDRV_O_RDWR,
1813                     NULL);
1814     }
1815 
1816     aio_context_release(aio_context);
1817 }
1818 
1819 static void external_snapshot_abort(BlkActionState *common)
1820 {
1821     ExternalSnapshotState *state =
1822                              DO_UPCAST(ExternalSnapshotState, common, common);
1823     if (state->new_bs) {
1824         if (state->overlay_appended) {
1825             AioContext *aio_context;
1826 
1827             aio_context = bdrv_get_aio_context(state->old_bs);
1828             aio_context_acquire(aio_context);
1829 
1830             bdrv_ref(state->old_bs);   /* we can't let bdrv_set_backind_hd()
1831                                           close state->old_bs; we need it */
1832             bdrv_set_backing_hd(state->new_bs, NULL, &error_abort);
1833             bdrv_replace_node(state->new_bs, state->old_bs, &error_abort);
1834             bdrv_unref(state->old_bs); /* bdrv_replace_node() ref'ed old_bs */
1835 
1836             aio_context_release(aio_context);
1837         }
1838     }
1839 }
1840 
1841 static void external_snapshot_clean(BlkActionState *common)
1842 {
1843     ExternalSnapshotState *state =
1844                              DO_UPCAST(ExternalSnapshotState, common, common);
1845     AioContext *aio_context;
1846 
1847     if (!state->old_bs) {
1848         return;
1849     }
1850 
1851     aio_context = bdrv_get_aio_context(state->old_bs);
1852     aio_context_acquire(aio_context);
1853 
1854     bdrv_drained_end(state->old_bs);
1855     bdrv_unref(state->new_bs);
1856 
1857     aio_context_release(aio_context);
1858 }
1859 
1860 typedef struct DriveBackupState {
1861     BlkActionState common;
1862     BlockDriverState *bs;
1863     BlockJob *job;
1864 } DriveBackupState;
1865 
1866 static BlockJob *do_drive_backup(DriveBackup *backup, BlockJobTxn *txn,
1867                             Error **errp);
1868 
1869 static void drive_backup_prepare(BlkActionState *common, Error **errp)
1870 {
1871     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1872     BlockDriverState *bs;
1873     DriveBackup *backup;
1874     AioContext *aio_context;
1875     Error *local_err = NULL;
1876 
1877     assert(common->action->type == TRANSACTION_ACTION_KIND_DRIVE_BACKUP);
1878     backup = common->action->u.drive_backup.data;
1879 
1880     bs = qmp_get_root_bs(backup->device, errp);
1881     if (!bs) {
1882         return;
1883     }
1884 
1885     aio_context = bdrv_get_aio_context(bs);
1886     aio_context_acquire(aio_context);
1887 
1888     /* Paired with .clean() */
1889     bdrv_drained_begin(bs);
1890 
1891     state->bs = bs;
1892 
1893     state->job = do_drive_backup(backup, common->block_job_txn, &local_err);
1894     if (local_err) {
1895         error_propagate(errp, local_err);
1896         goto out;
1897     }
1898 
1899 out:
1900     aio_context_release(aio_context);
1901 }
1902 
1903 static void drive_backup_commit(BlkActionState *common)
1904 {
1905     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1906     AioContext *aio_context;
1907 
1908     aio_context = bdrv_get_aio_context(state->bs);
1909     aio_context_acquire(aio_context);
1910 
1911     assert(state->job);
1912     block_job_start(state->job);
1913 
1914     aio_context_release(aio_context);
1915 }
1916 
1917 static void drive_backup_abort(BlkActionState *common)
1918 {
1919     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1920 
1921     if (state->job) {
1922         AioContext *aio_context;
1923 
1924         aio_context = bdrv_get_aio_context(state->bs);
1925         aio_context_acquire(aio_context);
1926 
1927         block_job_cancel_sync(state->job);
1928 
1929         aio_context_release(aio_context);
1930     }
1931 }
1932 
1933 static void drive_backup_clean(BlkActionState *common)
1934 {
1935     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1936     AioContext *aio_context;
1937 
1938     if (!state->bs) {
1939         return;
1940     }
1941 
1942     aio_context = bdrv_get_aio_context(state->bs);
1943     aio_context_acquire(aio_context);
1944 
1945     bdrv_drained_end(state->bs);
1946 
1947     aio_context_release(aio_context);
1948 }
1949 
1950 typedef struct BlockdevBackupState {
1951     BlkActionState common;
1952     BlockDriverState *bs;
1953     BlockJob *job;
1954 } BlockdevBackupState;
1955 
1956 static BlockJob *do_blockdev_backup(BlockdevBackup *backup, BlockJobTxn *txn,
1957                                     Error **errp);
1958 
1959 static void blockdev_backup_prepare(BlkActionState *common, Error **errp)
1960 {
1961     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1962     BlockdevBackup *backup;
1963     BlockDriverState *bs, *target;
1964     AioContext *aio_context;
1965     Error *local_err = NULL;
1966 
1967     assert(common->action->type == TRANSACTION_ACTION_KIND_BLOCKDEV_BACKUP);
1968     backup = common->action->u.blockdev_backup.data;
1969 
1970     bs = qmp_get_root_bs(backup->device, errp);
1971     if (!bs) {
1972         return;
1973     }
1974 
1975     target = bdrv_lookup_bs(backup->target, backup->target, errp);
1976     if (!target) {
1977         return;
1978     }
1979 
1980     aio_context = bdrv_get_aio_context(bs);
1981     if (aio_context != bdrv_get_aio_context(target)) {
1982         error_setg(errp, "Backup between two IO threads is not implemented");
1983         return;
1984     }
1985     aio_context_acquire(aio_context);
1986     state->bs = bs;
1987 
1988     /* Paired with .clean() */
1989     bdrv_drained_begin(state->bs);
1990 
1991     state->job = do_blockdev_backup(backup, common->block_job_txn, &local_err);
1992     if (local_err) {
1993         error_propagate(errp, local_err);
1994         goto out;
1995     }
1996 
1997 out:
1998     aio_context_release(aio_context);
1999 }
2000 
2001 static void blockdev_backup_commit(BlkActionState *common)
2002 {
2003     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
2004     AioContext *aio_context;
2005 
2006     aio_context = bdrv_get_aio_context(state->bs);
2007     aio_context_acquire(aio_context);
2008 
2009     assert(state->job);
2010     block_job_start(state->job);
2011 
2012     aio_context_release(aio_context);
2013 }
2014 
2015 static void blockdev_backup_abort(BlkActionState *common)
2016 {
2017     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
2018 
2019     if (state->job) {
2020         AioContext *aio_context;
2021 
2022         aio_context = bdrv_get_aio_context(state->bs);
2023         aio_context_acquire(aio_context);
2024 
2025         block_job_cancel_sync(state->job);
2026 
2027         aio_context_release(aio_context);
2028     }
2029 }
2030 
2031 static void blockdev_backup_clean(BlkActionState *common)
2032 {
2033     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
2034     AioContext *aio_context;
2035 
2036     if (!state->bs) {
2037         return;
2038     }
2039 
2040     aio_context = bdrv_get_aio_context(state->bs);
2041     aio_context_acquire(aio_context);
2042 
2043     bdrv_drained_end(state->bs);
2044 
2045     aio_context_release(aio_context);
2046 }
2047 
2048 typedef struct BlockDirtyBitmapState {
2049     BlkActionState common;
2050     BdrvDirtyBitmap *bitmap;
2051     BlockDriverState *bs;
2052     HBitmap *backup;
2053     bool prepared;
2054 } BlockDirtyBitmapState;
2055 
2056 static void block_dirty_bitmap_add_prepare(BlkActionState *common,
2057                                            Error **errp)
2058 {
2059     Error *local_err = NULL;
2060     BlockDirtyBitmapAdd *action;
2061     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2062                                              common, common);
2063 
2064     if (action_check_completion_mode(common, errp) < 0) {
2065         return;
2066     }
2067 
2068     action = common->action->u.block_dirty_bitmap_add.data;
2069     /* AIO context taken and released within qmp_block_dirty_bitmap_add */
2070     qmp_block_dirty_bitmap_add(action->node, action->name,
2071                                action->has_granularity, action->granularity,
2072                                action->has_persistent, action->persistent,
2073                                action->has_autoload, action->autoload,
2074                                &local_err);
2075 
2076     if (!local_err) {
2077         state->prepared = true;
2078     } else {
2079         error_propagate(errp, local_err);
2080     }
2081 }
2082 
2083 static void block_dirty_bitmap_add_abort(BlkActionState *common)
2084 {
2085     BlockDirtyBitmapAdd *action;
2086     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2087                                              common, common);
2088 
2089     action = common->action->u.block_dirty_bitmap_add.data;
2090     /* Should not be able to fail: IF the bitmap was added via .prepare(),
2091      * then the node reference and bitmap name must have been valid.
2092      */
2093     if (state->prepared) {
2094         qmp_block_dirty_bitmap_remove(action->node, action->name, &error_abort);
2095     }
2096 }
2097 
2098 static void block_dirty_bitmap_clear_prepare(BlkActionState *common,
2099                                              Error **errp)
2100 {
2101     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2102                                              common, common);
2103     BlockDirtyBitmap *action;
2104 
2105     if (action_check_completion_mode(common, errp) < 0) {
2106         return;
2107     }
2108 
2109     action = common->action->u.block_dirty_bitmap_clear.data;
2110     state->bitmap = block_dirty_bitmap_lookup(action->node,
2111                                               action->name,
2112                                               &state->bs,
2113                                               errp);
2114     if (!state->bitmap) {
2115         return;
2116     }
2117 
2118     if (bdrv_dirty_bitmap_frozen(state->bitmap)) {
2119         error_setg(errp, "Cannot modify a frozen bitmap");
2120         return;
2121     } else if (!bdrv_dirty_bitmap_enabled(state->bitmap)) {
2122         error_setg(errp, "Cannot clear a disabled bitmap");
2123         return;
2124     } else if (bdrv_dirty_bitmap_readonly(state->bitmap)) {
2125         error_setg(errp, "Cannot clear a readonly bitmap");
2126         return;
2127     }
2128 
2129     bdrv_clear_dirty_bitmap(state->bitmap, &state->backup);
2130 }
2131 
2132 static void block_dirty_bitmap_clear_abort(BlkActionState *common)
2133 {
2134     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2135                                              common, common);
2136 
2137     if (state->backup) {
2138         bdrv_undo_clear_dirty_bitmap(state->bitmap, state->backup);
2139     }
2140 }
2141 
2142 static void block_dirty_bitmap_clear_commit(BlkActionState *common)
2143 {
2144     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2145                                              common, common);
2146 
2147     hbitmap_free(state->backup);
2148 }
2149 
2150 static void abort_prepare(BlkActionState *common, Error **errp)
2151 {
2152     error_setg(errp, "Transaction aborted using Abort action");
2153 }
2154 
2155 static void abort_commit(BlkActionState *common)
2156 {
2157     g_assert_not_reached(); /* this action never succeeds */
2158 }
2159 
2160 static const BlkActionOps actions[] = {
2161     [TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT] = {
2162         .instance_size = sizeof(ExternalSnapshotState),
2163         .prepare  = external_snapshot_prepare,
2164         .commit   = external_snapshot_commit,
2165         .abort = external_snapshot_abort,
2166         .clean = external_snapshot_clean,
2167     },
2168     [TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC] = {
2169         .instance_size = sizeof(ExternalSnapshotState),
2170         .prepare  = external_snapshot_prepare,
2171         .commit   = external_snapshot_commit,
2172         .abort = external_snapshot_abort,
2173         .clean = external_snapshot_clean,
2174     },
2175     [TRANSACTION_ACTION_KIND_DRIVE_BACKUP] = {
2176         .instance_size = sizeof(DriveBackupState),
2177         .prepare = drive_backup_prepare,
2178         .commit = drive_backup_commit,
2179         .abort = drive_backup_abort,
2180         .clean = drive_backup_clean,
2181     },
2182     [TRANSACTION_ACTION_KIND_BLOCKDEV_BACKUP] = {
2183         .instance_size = sizeof(BlockdevBackupState),
2184         .prepare = blockdev_backup_prepare,
2185         .commit = blockdev_backup_commit,
2186         .abort = blockdev_backup_abort,
2187         .clean = blockdev_backup_clean,
2188     },
2189     [TRANSACTION_ACTION_KIND_ABORT] = {
2190         .instance_size = sizeof(BlkActionState),
2191         .prepare = abort_prepare,
2192         .commit = abort_commit,
2193     },
2194     [TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC] = {
2195         .instance_size = sizeof(InternalSnapshotState),
2196         .prepare  = internal_snapshot_prepare,
2197         .abort = internal_snapshot_abort,
2198         .clean = internal_snapshot_clean,
2199     },
2200     [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_ADD] = {
2201         .instance_size = sizeof(BlockDirtyBitmapState),
2202         .prepare = block_dirty_bitmap_add_prepare,
2203         .abort = block_dirty_bitmap_add_abort,
2204     },
2205     [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_CLEAR] = {
2206         .instance_size = sizeof(BlockDirtyBitmapState),
2207         .prepare = block_dirty_bitmap_clear_prepare,
2208         .commit = block_dirty_bitmap_clear_commit,
2209         .abort = block_dirty_bitmap_clear_abort,
2210     }
2211 };
2212 
2213 /**
2214  * Allocate a TransactionProperties structure if necessary, and fill
2215  * that structure with desired defaults if they are unset.
2216  */
2217 static TransactionProperties *get_transaction_properties(
2218     TransactionProperties *props)
2219 {
2220     if (!props) {
2221         props = g_new0(TransactionProperties, 1);
2222     }
2223 
2224     if (!props->has_completion_mode) {
2225         props->has_completion_mode = true;
2226         props->completion_mode = ACTION_COMPLETION_MODE_INDIVIDUAL;
2227     }
2228 
2229     return props;
2230 }
2231 
2232 /*
2233  * 'Atomic' group operations.  The operations are performed as a set, and if
2234  * any fail then we roll back all operations in the group.
2235  */
2236 void qmp_transaction(TransactionActionList *dev_list,
2237                      bool has_props,
2238                      struct TransactionProperties *props,
2239                      Error **errp)
2240 {
2241     TransactionActionList *dev_entry = dev_list;
2242     BlockJobTxn *block_job_txn = NULL;
2243     BlkActionState *state, *next;
2244     Error *local_err = NULL;
2245 
2246     QSIMPLEQ_HEAD(snap_bdrv_states, BlkActionState) snap_bdrv_states;
2247     QSIMPLEQ_INIT(&snap_bdrv_states);
2248 
2249     /* Does this transaction get canceled as a group on failure?
2250      * If not, we don't really need to make a BlockJobTxn.
2251      */
2252     props = get_transaction_properties(props);
2253     if (props->completion_mode != ACTION_COMPLETION_MODE_INDIVIDUAL) {
2254         block_job_txn = block_job_txn_new();
2255     }
2256 
2257     /* drain all i/o before any operations */
2258     bdrv_drain_all();
2259 
2260     /* We don't do anything in this loop that commits us to the operations */
2261     while (NULL != dev_entry) {
2262         TransactionAction *dev_info = NULL;
2263         const BlkActionOps *ops;
2264 
2265         dev_info = dev_entry->value;
2266         dev_entry = dev_entry->next;
2267 
2268         assert(dev_info->type < ARRAY_SIZE(actions));
2269 
2270         ops = &actions[dev_info->type];
2271         assert(ops->instance_size > 0);
2272 
2273         state = g_malloc0(ops->instance_size);
2274         state->ops = ops;
2275         state->action = dev_info;
2276         state->block_job_txn = block_job_txn;
2277         state->txn_props = props;
2278         QSIMPLEQ_INSERT_TAIL(&snap_bdrv_states, state, entry);
2279 
2280         state->ops->prepare(state, &local_err);
2281         if (local_err) {
2282             error_propagate(errp, local_err);
2283             goto delete_and_fail;
2284         }
2285     }
2286 
2287     QSIMPLEQ_FOREACH(state, &snap_bdrv_states, entry) {
2288         if (state->ops->commit) {
2289             state->ops->commit(state);
2290         }
2291     }
2292 
2293     /* success */
2294     goto exit;
2295 
2296 delete_and_fail:
2297     /* failure, and it is all-or-none; roll back all operations */
2298     QSIMPLEQ_FOREACH(state, &snap_bdrv_states, entry) {
2299         if (state->ops->abort) {
2300             state->ops->abort(state);
2301         }
2302     }
2303 exit:
2304     QSIMPLEQ_FOREACH_SAFE(state, &snap_bdrv_states, entry, next) {
2305         if (state->ops->clean) {
2306             state->ops->clean(state);
2307         }
2308         g_free(state);
2309     }
2310     if (!has_props) {
2311         qapi_free_TransactionProperties(props);
2312     }
2313     block_job_txn_unref(block_job_txn);
2314 }
2315 
2316 void qmp_eject(bool has_device, const char *device,
2317                bool has_id, const char *id,
2318                bool has_force, bool force, Error **errp)
2319 {
2320     Error *local_err = NULL;
2321     int rc;
2322 
2323     if (!has_force) {
2324         force = false;
2325     }
2326 
2327     rc = do_open_tray(has_device ? device : NULL,
2328                       has_id ? id : NULL,
2329                       force, &local_err);
2330     if (rc && rc != -ENOSYS) {
2331         error_propagate(errp, local_err);
2332         return;
2333     }
2334     error_free(local_err);
2335 
2336     blockdev_remove_medium(has_device, device, has_id, id, errp);
2337 }
2338 
2339 void qmp_block_passwd(bool has_device, const char *device,
2340                       bool has_node_name, const char *node_name,
2341                       const char *password, Error **errp)
2342 {
2343     error_setg(errp,
2344                "Setting block passwords directly is no longer supported");
2345 }
2346 
2347 /*
2348  * Attempt to open the tray of @device.
2349  * If @force, ignore its tray lock.
2350  * Else, if the tray is locked, don't open it, but ask the guest to open it.
2351  * On error, store an error through @errp and return -errno.
2352  * If @device does not exist, return -ENODEV.
2353  * If it has no removable media, return -ENOTSUP.
2354  * If it has no tray, return -ENOSYS.
2355  * If the guest was asked to open the tray, return -EINPROGRESS.
2356  * Else, return 0.
2357  */
2358 static int do_open_tray(const char *blk_name, const char *qdev_id,
2359                         bool force, Error **errp)
2360 {
2361     BlockBackend *blk;
2362     const char *device = qdev_id ?: blk_name;
2363     bool locked;
2364 
2365     blk = qmp_get_blk(blk_name, qdev_id, errp);
2366     if (!blk) {
2367         return -ENODEV;
2368     }
2369 
2370     if (!blk_dev_has_removable_media(blk)) {
2371         error_setg(errp, "Device '%s' is not removable", device);
2372         return -ENOTSUP;
2373     }
2374 
2375     if (!blk_dev_has_tray(blk)) {
2376         error_setg(errp, "Device '%s' does not have a tray", device);
2377         return -ENOSYS;
2378     }
2379 
2380     if (blk_dev_is_tray_open(blk)) {
2381         return 0;
2382     }
2383 
2384     locked = blk_dev_is_medium_locked(blk);
2385     if (locked) {
2386         blk_dev_eject_request(blk, force);
2387     }
2388 
2389     if (!locked || force) {
2390         blk_dev_change_media_cb(blk, false, &error_abort);
2391     }
2392 
2393     if (locked && !force) {
2394         error_setg(errp, "Device '%s' is locked and force was not specified, "
2395                    "wait for tray to open and try again", device);
2396         return -EINPROGRESS;
2397     }
2398 
2399     return 0;
2400 }
2401 
2402 void qmp_blockdev_open_tray(bool has_device, const char *device,
2403                             bool has_id, const char *id,
2404                             bool has_force, bool force,
2405                             Error **errp)
2406 {
2407     Error *local_err = NULL;
2408     int rc;
2409 
2410     if (!has_force) {
2411         force = false;
2412     }
2413     rc = do_open_tray(has_device ? device : NULL,
2414                       has_id ? id : NULL,
2415                       force, &local_err);
2416     if (rc && rc != -ENOSYS && rc != -EINPROGRESS) {
2417         error_propagate(errp, local_err);
2418         return;
2419     }
2420     error_free(local_err);
2421 }
2422 
2423 void qmp_blockdev_close_tray(bool has_device, const char *device,
2424                              bool has_id, const char *id,
2425                              Error **errp)
2426 {
2427     BlockBackend *blk;
2428     Error *local_err = NULL;
2429 
2430     device = has_device ? device : NULL;
2431     id = has_id ? id : NULL;
2432 
2433     blk = qmp_get_blk(device, id, errp);
2434     if (!blk) {
2435         return;
2436     }
2437 
2438     if (!blk_dev_has_removable_media(blk)) {
2439         error_setg(errp, "Device '%s' is not removable", device ?: id);
2440         return;
2441     }
2442 
2443     if (!blk_dev_has_tray(blk)) {
2444         /* Ignore this command on tray-less devices */
2445         return;
2446     }
2447 
2448     if (!blk_dev_is_tray_open(blk)) {
2449         return;
2450     }
2451 
2452     blk_dev_change_media_cb(blk, true, &local_err);
2453     if (local_err) {
2454         error_propagate(errp, local_err);
2455         return;
2456     }
2457 }
2458 
2459 static void blockdev_remove_medium(bool has_device, const char *device,
2460                                    bool has_id, const char *id, Error **errp)
2461 {
2462     BlockBackend *blk;
2463     BlockDriverState *bs;
2464     AioContext *aio_context;
2465     bool has_attached_device;
2466 
2467     device = has_device ? device : NULL;
2468     id = has_id ? id : NULL;
2469 
2470     blk = qmp_get_blk(device, id, errp);
2471     if (!blk) {
2472         return;
2473     }
2474 
2475     /* For BBs without a device, we can exchange the BDS tree at will */
2476     has_attached_device = blk_get_attached_dev(blk);
2477 
2478     if (has_attached_device && !blk_dev_has_removable_media(blk)) {
2479         error_setg(errp, "Device '%s' is not removable", device ?: id);
2480         return;
2481     }
2482 
2483     if (has_attached_device && blk_dev_has_tray(blk) &&
2484         !blk_dev_is_tray_open(blk))
2485     {
2486         error_setg(errp, "Tray of device '%s' is not open", device ?: id);
2487         return;
2488     }
2489 
2490     bs = blk_bs(blk);
2491     if (!bs) {
2492         return;
2493     }
2494 
2495     aio_context = bdrv_get_aio_context(bs);
2496     aio_context_acquire(aio_context);
2497 
2498     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_EJECT, errp)) {
2499         goto out;
2500     }
2501 
2502     blk_remove_bs(blk);
2503 
2504     if (!blk_dev_has_tray(blk)) {
2505         /* For tray-less devices, blockdev-open-tray is a no-op (or may not be
2506          * called at all); therefore, the medium needs to be ejected here.
2507          * Do it after blk_remove_bs() so blk_is_inserted(blk) returns the @load
2508          * value passed here (i.e. false). */
2509         blk_dev_change_media_cb(blk, false, &error_abort);
2510     }
2511 
2512 out:
2513     aio_context_release(aio_context);
2514 }
2515 
2516 void qmp_blockdev_remove_medium(const char *id, Error **errp)
2517 {
2518     blockdev_remove_medium(false, NULL, true, id, errp);
2519 }
2520 
2521 static void qmp_blockdev_insert_anon_medium(BlockBackend *blk,
2522                                             BlockDriverState *bs, Error **errp)
2523 {
2524     Error *local_err = NULL;
2525     bool has_device;
2526     int ret;
2527 
2528     /* For BBs without a device, we can exchange the BDS tree at will */
2529     has_device = blk_get_attached_dev(blk);
2530 
2531     if (has_device && !blk_dev_has_removable_media(blk)) {
2532         error_setg(errp, "Device is not removable");
2533         return;
2534     }
2535 
2536     if (has_device && blk_dev_has_tray(blk) && !blk_dev_is_tray_open(blk)) {
2537         error_setg(errp, "Tray of the device is not open");
2538         return;
2539     }
2540 
2541     if (blk_bs(blk)) {
2542         error_setg(errp, "There already is a medium in the device");
2543         return;
2544     }
2545 
2546     ret = blk_insert_bs(blk, bs, errp);
2547     if (ret < 0) {
2548         return;
2549     }
2550 
2551     if (!blk_dev_has_tray(blk)) {
2552         /* For tray-less devices, blockdev-close-tray is a no-op (or may not be
2553          * called at all); therefore, the medium needs to be pushed into the
2554          * slot here.
2555          * Do it after blk_insert_bs() so blk_is_inserted(blk) returns the @load
2556          * value passed here (i.e. true). */
2557         blk_dev_change_media_cb(blk, true, &local_err);
2558         if (local_err) {
2559             error_propagate(errp, local_err);
2560             blk_remove_bs(blk);
2561             return;
2562         }
2563     }
2564 }
2565 
2566 static void blockdev_insert_medium(bool has_device, const char *device,
2567                                    bool has_id, const char *id,
2568                                    const char *node_name, Error **errp)
2569 {
2570     BlockBackend *blk;
2571     BlockDriverState *bs;
2572 
2573     blk = qmp_get_blk(has_device ? device : NULL,
2574                       has_id ? id : NULL,
2575                       errp);
2576     if (!blk) {
2577         return;
2578     }
2579 
2580     bs = bdrv_find_node(node_name);
2581     if (!bs) {
2582         error_setg(errp, "Node '%s' not found", node_name);
2583         return;
2584     }
2585 
2586     if (bdrv_has_blk(bs)) {
2587         error_setg(errp, "Node '%s' is already in use", node_name);
2588         return;
2589     }
2590 
2591     qmp_blockdev_insert_anon_medium(blk, bs, errp);
2592 }
2593 
2594 void qmp_blockdev_insert_medium(const char *id, const char *node_name,
2595                                 Error **errp)
2596 {
2597     blockdev_insert_medium(false, NULL, true, id, node_name, errp);
2598 }
2599 
2600 void qmp_blockdev_change_medium(bool has_device, const char *device,
2601                                 bool has_id, const char *id,
2602                                 const char *filename,
2603                                 bool has_format, const char *format,
2604                                 bool has_read_only,
2605                                 BlockdevChangeReadOnlyMode read_only,
2606                                 Error **errp)
2607 {
2608     BlockBackend *blk;
2609     BlockDriverState *medium_bs = NULL;
2610     int bdrv_flags;
2611     bool detect_zeroes;
2612     int rc;
2613     QDict *options = NULL;
2614     Error *err = NULL;
2615 
2616     blk = qmp_get_blk(has_device ? device : NULL,
2617                       has_id ? id : NULL,
2618                       errp);
2619     if (!blk) {
2620         goto fail;
2621     }
2622 
2623     if (blk_bs(blk)) {
2624         blk_update_root_state(blk);
2625     }
2626 
2627     bdrv_flags = blk_get_open_flags_from_root_state(blk);
2628     bdrv_flags &= ~(BDRV_O_TEMPORARY | BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING |
2629         BDRV_O_PROTOCOL);
2630 
2631     if (!has_read_only) {
2632         read_only = BLOCKDEV_CHANGE_READ_ONLY_MODE_RETAIN;
2633     }
2634 
2635     switch (read_only) {
2636     case BLOCKDEV_CHANGE_READ_ONLY_MODE_RETAIN:
2637         break;
2638 
2639     case BLOCKDEV_CHANGE_READ_ONLY_MODE_READ_ONLY:
2640         bdrv_flags &= ~BDRV_O_RDWR;
2641         break;
2642 
2643     case BLOCKDEV_CHANGE_READ_ONLY_MODE_READ_WRITE:
2644         bdrv_flags |= BDRV_O_RDWR;
2645         break;
2646 
2647     default:
2648         abort();
2649     }
2650 
2651     options = qdict_new();
2652     detect_zeroes = blk_get_detect_zeroes_from_root_state(blk);
2653     qdict_put_str(options, "detect-zeroes", detect_zeroes ? "on" : "off");
2654 
2655     if (has_format) {
2656         qdict_put_str(options, "driver", format);
2657     }
2658 
2659     medium_bs = bdrv_open(filename, NULL, options, bdrv_flags, errp);
2660     if (!medium_bs) {
2661         goto fail;
2662     }
2663 
2664     rc = do_open_tray(has_device ? device : NULL,
2665                       has_id ? id : NULL,
2666                       false, &err);
2667     if (rc && rc != -ENOSYS) {
2668         error_propagate(errp, err);
2669         goto fail;
2670     }
2671     error_free(err);
2672     err = NULL;
2673 
2674     blockdev_remove_medium(has_device, device, has_id, id, &err);
2675     if (err) {
2676         error_propagate(errp, err);
2677         goto fail;
2678     }
2679 
2680     qmp_blockdev_insert_anon_medium(blk, medium_bs, &err);
2681     if (err) {
2682         error_propagate(errp, err);
2683         goto fail;
2684     }
2685 
2686     qmp_blockdev_close_tray(has_device, device, has_id, id, errp);
2687 
2688 fail:
2689     /* If the medium has been inserted, the device has its own reference, so
2690      * ours must be relinquished; and if it has not been inserted successfully,
2691      * the reference must be relinquished anyway */
2692     bdrv_unref(medium_bs);
2693 }
2694 
2695 /* throttling disk I/O limits */
2696 void qmp_block_set_io_throttle(BlockIOThrottle *arg, Error **errp)
2697 {
2698     ThrottleConfig cfg;
2699     BlockDriverState *bs;
2700     BlockBackend *blk;
2701     AioContext *aio_context;
2702 
2703     blk = qmp_get_blk(arg->has_device ? arg->device : NULL,
2704                       arg->has_id ? arg->id : NULL,
2705                       errp);
2706     if (!blk) {
2707         return;
2708     }
2709 
2710     aio_context = blk_get_aio_context(blk);
2711     aio_context_acquire(aio_context);
2712 
2713     bs = blk_bs(blk);
2714     if (!bs) {
2715         error_setg(errp, "Device has no medium");
2716         goto out;
2717     }
2718 
2719     throttle_config_init(&cfg);
2720     cfg.buckets[THROTTLE_BPS_TOTAL].avg = arg->bps;
2721     cfg.buckets[THROTTLE_BPS_READ].avg  = arg->bps_rd;
2722     cfg.buckets[THROTTLE_BPS_WRITE].avg = arg->bps_wr;
2723 
2724     cfg.buckets[THROTTLE_OPS_TOTAL].avg = arg->iops;
2725     cfg.buckets[THROTTLE_OPS_READ].avg  = arg->iops_rd;
2726     cfg.buckets[THROTTLE_OPS_WRITE].avg = arg->iops_wr;
2727 
2728     if (arg->has_bps_max) {
2729         cfg.buckets[THROTTLE_BPS_TOTAL].max = arg->bps_max;
2730     }
2731     if (arg->has_bps_rd_max) {
2732         cfg.buckets[THROTTLE_BPS_READ].max = arg->bps_rd_max;
2733     }
2734     if (arg->has_bps_wr_max) {
2735         cfg.buckets[THROTTLE_BPS_WRITE].max = arg->bps_wr_max;
2736     }
2737     if (arg->has_iops_max) {
2738         cfg.buckets[THROTTLE_OPS_TOTAL].max = arg->iops_max;
2739     }
2740     if (arg->has_iops_rd_max) {
2741         cfg.buckets[THROTTLE_OPS_READ].max = arg->iops_rd_max;
2742     }
2743     if (arg->has_iops_wr_max) {
2744         cfg.buckets[THROTTLE_OPS_WRITE].max = arg->iops_wr_max;
2745     }
2746 
2747     if (arg->has_bps_max_length) {
2748         cfg.buckets[THROTTLE_BPS_TOTAL].burst_length = arg->bps_max_length;
2749     }
2750     if (arg->has_bps_rd_max_length) {
2751         cfg.buckets[THROTTLE_BPS_READ].burst_length = arg->bps_rd_max_length;
2752     }
2753     if (arg->has_bps_wr_max_length) {
2754         cfg.buckets[THROTTLE_BPS_WRITE].burst_length = arg->bps_wr_max_length;
2755     }
2756     if (arg->has_iops_max_length) {
2757         cfg.buckets[THROTTLE_OPS_TOTAL].burst_length = arg->iops_max_length;
2758     }
2759     if (arg->has_iops_rd_max_length) {
2760         cfg.buckets[THROTTLE_OPS_READ].burst_length = arg->iops_rd_max_length;
2761     }
2762     if (arg->has_iops_wr_max_length) {
2763         cfg.buckets[THROTTLE_OPS_WRITE].burst_length = arg->iops_wr_max_length;
2764     }
2765 
2766     if (arg->has_iops_size) {
2767         cfg.op_size = arg->iops_size;
2768     }
2769 
2770     if (!throttle_is_valid(&cfg, errp)) {
2771         goto out;
2772     }
2773 
2774     if (throttle_enabled(&cfg)) {
2775         /* Enable I/O limits if they're not enabled yet, otherwise
2776          * just update the throttling group. */
2777         if (!blk_get_public(blk)->throttle_group_member.throttle_state) {
2778             blk_io_limits_enable(blk,
2779                                  arg->has_group ? arg->group :
2780                                  arg->has_device ? arg->device :
2781                                  arg->id);
2782         } else if (arg->has_group) {
2783             blk_io_limits_update_group(blk, arg->group);
2784         }
2785         /* Set the new throttling configuration */
2786         blk_set_io_limits(blk, &cfg);
2787     } else if (blk_get_public(blk)->throttle_group_member.throttle_state) {
2788         /* If all throttling settings are set to 0, disable I/O limits */
2789         blk_io_limits_disable(blk);
2790     }
2791 
2792 out:
2793     aio_context_release(aio_context);
2794 }
2795 
2796 void qmp_block_dirty_bitmap_add(const char *node, const char *name,
2797                                 bool has_granularity, uint32_t granularity,
2798                                 bool has_persistent, bool persistent,
2799                                 bool has_autoload, bool autoload,
2800                                 Error **errp)
2801 {
2802     BlockDriverState *bs;
2803     BdrvDirtyBitmap *bitmap;
2804 
2805     if (!name || name[0] == '\0') {
2806         error_setg(errp, "Bitmap name cannot be empty");
2807         return;
2808     }
2809 
2810     bs = bdrv_lookup_bs(node, node, errp);
2811     if (!bs) {
2812         return;
2813     }
2814 
2815     if (has_granularity) {
2816         if (granularity < 512 || !is_power_of_2(granularity)) {
2817             error_setg(errp, "Granularity must be power of 2 "
2818                              "and at least 512");
2819             return;
2820         }
2821     } else {
2822         /* Default to cluster size, if available: */
2823         granularity = bdrv_get_default_bitmap_granularity(bs);
2824     }
2825 
2826     if (!has_persistent) {
2827         persistent = false;
2828     }
2829 
2830     if (has_autoload) {
2831         warn_report("Autoload option is deprecated and its value is ignored");
2832     }
2833 
2834     if (persistent &&
2835         !bdrv_can_store_new_dirty_bitmap(bs, name, granularity, errp))
2836     {
2837         return;
2838     }
2839 
2840     bitmap = bdrv_create_dirty_bitmap(bs, granularity, name, errp);
2841     if (bitmap == NULL) {
2842         return;
2843     }
2844 
2845     bdrv_dirty_bitmap_set_persistance(bitmap, persistent);
2846 }
2847 
2848 void qmp_block_dirty_bitmap_remove(const char *node, const char *name,
2849                                    Error **errp)
2850 {
2851     BlockDriverState *bs;
2852     BdrvDirtyBitmap *bitmap;
2853     Error *local_err = NULL;
2854 
2855     bitmap = block_dirty_bitmap_lookup(node, name, &bs, errp);
2856     if (!bitmap || !bs) {
2857         return;
2858     }
2859 
2860     if (bdrv_dirty_bitmap_frozen(bitmap)) {
2861         error_setg(errp,
2862                    "Bitmap '%s' is currently frozen and cannot be removed",
2863                    name);
2864         return;
2865     }
2866 
2867     if (bdrv_dirty_bitmap_get_persistance(bitmap)) {
2868         bdrv_remove_persistent_dirty_bitmap(bs, name, &local_err);
2869         if (local_err != NULL) {
2870             error_propagate(errp, local_err);
2871             return;
2872         }
2873     }
2874 
2875     bdrv_dirty_bitmap_make_anon(bitmap);
2876     bdrv_release_dirty_bitmap(bs, bitmap);
2877 }
2878 
2879 /**
2880  * Completely clear a bitmap, for the purposes of synchronizing a bitmap
2881  * immediately after a full backup operation.
2882  */
2883 void qmp_block_dirty_bitmap_clear(const char *node, const char *name,
2884                                   Error **errp)
2885 {
2886     BdrvDirtyBitmap *bitmap;
2887     BlockDriverState *bs;
2888 
2889     bitmap = block_dirty_bitmap_lookup(node, name, &bs, errp);
2890     if (!bitmap || !bs) {
2891         return;
2892     }
2893 
2894     if (bdrv_dirty_bitmap_frozen(bitmap)) {
2895         error_setg(errp,
2896                    "Bitmap '%s' is currently frozen and cannot be modified",
2897                    name);
2898         return;
2899     } else if (!bdrv_dirty_bitmap_enabled(bitmap)) {
2900         error_setg(errp,
2901                    "Bitmap '%s' is currently disabled and cannot be cleared",
2902                    name);
2903         return;
2904     } else if (bdrv_dirty_bitmap_readonly(bitmap)) {
2905         error_setg(errp, "Bitmap '%s' is readonly and cannot be cleared", name);
2906         return;
2907     }
2908 
2909     bdrv_clear_dirty_bitmap(bitmap, NULL);
2910 }
2911 
2912 BlockDirtyBitmapSha256 *qmp_x_debug_block_dirty_bitmap_sha256(const char *node,
2913                                                               const char *name,
2914                                                               Error **errp)
2915 {
2916     BdrvDirtyBitmap *bitmap;
2917     BlockDriverState *bs;
2918     BlockDirtyBitmapSha256 *ret = NULL;
2919     char *sha256;
2920 
2921     bitmap = block_dirty_bitmap_lookup(node, name, &bs, errp);
2922     if (!bitmap || !bs) {
2923         return NULL;
2924     }
2925 
2926     sha256 = bdrv_dirty_bitmap_sha256(bitmap, errp);
2927     if (sha256 == NULL) {
2928         return NULL;
2929     }
2930 
2931     ret = g_new(BlockDirtyBitmapSha256, 1);
2932     ret->sha256 = sha256;
2933 
2934     return ret;
2935 }
2936 
2937 void hmp_drive_del(Monitor *mon, const QDict *qdict)
2938 {
2939     const char *id = qdict_get_str(qdict, "id");
2940     BlockBackend *blk;
2941     BlockDriverState *bs;
2942     AioContext *aio_context;
2943     Error *local_err = NULL;
2944 
2945     bs = bdrv_find_node(id);
2946     if (bs) {
2947         qmp_blockdev_del(id, &local_err);
2948         if (local_err) {
2949             error_report_err(local_err);
2950         }
2951         return;
2952     }
2953 
2954     blk = blk_by_name(id);
2955     if (!blk) {
2956         error_report("Device '%s' not found", id);
2957         return;
2958     }
2959 
2960     if (!blk_legacy_dinfo(blk)) {
2961         error_report("Deleting device added with blockdev-add"
2962                      " is not supported");
2963         return;
2964     }
2965 
2966     aio_context = blk_get_aio_context(blk);
2967     aio_context_acquire(aio_context);
2968 
2969     bs = blk_bs(blk);
2970     if (bs) {
2971         if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_DRIVE_DEL, &local_err)) {
2972             error_report_err(local_err);
2973             aio_context_release(aio_context);
2974             return;
2975         }
2976 
2977         blk_remove_bs(blk);
2978     }
2979 
2980     /* Make the BlockBackend and the attached BlockDriverState anonymous */
2981     monitor_remove_blk(blk);
2982 
2983     /* If this BlockBackend has a device attached to it, its refcount will be
2984      * decremented when the device is removed; otherwise we have to do so here.
2985      */
2986     if (blk_get_attached_dev(blk)) {
2987         /* Further I/O must not pause the guest */
2988         blk_set_on_error(blk, BLOCKDEV_ON_ERROR_REPORT,
2989                          BLOCKDEV_ON_ERROR_REPORT);
2990     } else {
2991         blk_unref(blk);
2992     }
2993 
2994     aio_context_release(aio_context);
2995 }
2996 
2997 void qmp_block_resize(bool has_device, const char *device,
2998                       bool has_node_name, const char *node_name,
2999                       int64_t size, Error **errp)
3000 {
3001     Error *local_err = NULL;
3002     BlockBackend *blk = NULL;
3003     BlockDriverState *bs;
3004     AioContext *aio_context;
3005     int ret;
3006 
3007     bs = bdrv_lookup_bs(has_device ? device : NULL,
3008                         has_node_name ? node_name : NULL,
3009                         &local_err);
3010     if (local_err) {
3011         error_propagate(errp, local_err);
3012         return;
3013     }
3014 
3015     aio_context = bdrv_get_aio_context(bs);
3016     aio_context_acquire(aio_context);
3017 
3018     if (!bdrv_is_first_non_filter(bs)) {
3019         error_setg(errp, QERR_FEATURE_DISABLED, "resize");
3020         goto out;
3021     }
3022 
3023     if (size < 0) {
3024         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "size", "a >0 size");
3025         goto out;
3026     }
3027 
3028     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_RESIZE, NULL)) {
3029         error_setg(errp, QERR_DEVICE_IN_USE, device);
3030         goto out;
3031     }
3032 
3033     blk = blk_new(BLK_PERM_RESIZE, BLK_PERM_ALL);
3034     ret = blk_insert_bs(blk, bs, errp);
3035     if (ret < 0) {
3036         goto out;
3037     }
3038 
3039     bdrv_drained_begin(bs);
3040     ret = blk_truncate(blk, size, PREALLOC_MODE_OFF, errp);
3041     bdrv_drained_end(bs);
3042 
3043 out:
3044     blk_unref(blk);
3045     aio_context_release(aio_context);
3046 }
3047 
3048 void qmp_block_stream(bool has_job_id, const char *job_id, const char *device,
3049                       bool has_base, const char *base,
3050                       bool has_base_node, const char *base_node,
3051                       bool has_backing_file, const char *backing_file,
3052                       bool has_speed, int64_t speed,
3053                       bool has_on_error, BlockdevOnError on_error,
3054                       Error **errp)
3055 {
3056     BlockDriverState *bs, *iter;
3057     BlockDriverState *base_bs = NULL;
3058     AioContext *aio_context;
3059     Error *local_err = NULL;
3060     const char *base_name = NULL;
3061 
3062     if (!has_on_error) {
3063         on_error = BLOCKDEV_ON_ERROR_REPORT;
3064     }
3065 
3066     bs = bdrv_lookup_bs(device, device, errp);
3067     if (!bs) {
3068         return;
3069     }
3070 
3071     aio_context = bdrv_get_aio_context(bs);
3072     aio_context_acquire(aio_context);
3073 
3074     if (has_base && has_base_node) {
3075         error_setg(errp, "'base' and 'base-node' cannot be specified "
3076                    "at the same time");
3077         goto out;
3078     }
3079 
3080     if (has_base) {
3081         base_bs = bdrv_find_backing_image(bs, base);
3082         if (base_bs == NULL) {
3083             error_setg(errp, QERR_BASE_NOT_FOUND, base);
3084             goto out;
3085         }
3086         assert(bdrv_get_aio_context(base_bs) == aio_context);
3087         base_name = base;
3088     }
3089 
3090     if (has_base_node) {
3091         base_bs = bdrv_lookup_bs(NULL, base_node, errp);
3092         if (!base_bs) {
3093             goto out;
3094         }
3095         if (bs == base_bs || !bdrv_chain_contains(bs, base_bs)) {
3096             error_setg(errp, "Node '%s' is not a backing image of '%s'",
3097                        base_node, device);
3098             goto out;
3099         }
3100         assert(bdrv_get_aio_context(base_bs) == aio_context);
3101         base_name = base_bs->filename;
3102     }
3103 
3104     /* Check for op blockers in the whole chain between bs and base */
3105     for (iter = bs; iter && iter != base_bs; iter = backing_bs(iter)) {
3106         if (bdrv_op_is_blocked(iter, BLOCK_OP_TYPE_STREAM, errp)) {
3107             goto out;
3108         }
3109     }
3110 
3111     /* if we are streaming the entire chain, the result will have no backing
3112      * file, and specifying one is therefore an error */
3113     if (base_bs == NULL && has_backing_file) {
3114         error_setg(errp, "backing file specified, but streaming the "
3115                          "entire chain");
3116         goto out;
3117     }
3118 
3119     /* backing_file string overrides base bs filename */
3120     base_name = has_backing_file ? backing_file : base_name;
3121 
3122     stream_start(has_job_id ? job_id : NULL, bs, base_bs, base_name,
3123                  has_speed ? speed : 0, on_error, &local_err);
3124     if (local_err) {
3125         error_propagate(errp, local_err);
3126         goto out;
3127     }
3128 
3129     trace_qmp_block_stream(bs, bs->job);
3130 
3131 out:
3132     aio_context_release(aio_context);
3133 }
3134 
3135 void qmp_block_commit(bool has_job_id, const char *job_id, const char *device,
3136                       bool has_base, const char *base,
3137                       bool has_top, const char *top,
3138                       bool has_backing_file, const char *backing_file,
3139                       bool has_speed, int64_t speed,
3140                       bool has_filter_node_name, const char *filter_node_name,
3141                       Error **errp)
3142 {
3143     BlockDriverState *bs;
3144     BlockDriverState *iter;
3145     BlockDriverState *base_bs, *top_bs;
3146     AioContext *aio_context;
3147     Error *local_err = NULL;
3148     /* This will be part of the QMP command, if/when the
3149      * BlockdevOnError change for blkmirror makes it in
3150      */
3151     BlockdevOnError on_error = BLOCKDEV_ON_ERROR_REPORT;
3152 
3153     if (!has_speed) {
3154         speed = 0;
3155     }
3156     if (!has_filter_node_name) {
3157         filter_node_name = NULL;
3158     }
3159 
3160     /* Important Note:
3161      *  libvirt relies on the DeviceNotFound error class in order to probe for
3162      *  live commit feature versions; for this to work, we must make sure to
3163      *  perform the device lookup before any generic errors that may occur in a
3164      *  scenario in which all optional arguments are omitted. */
3165     bs = qmp_get_root_bs(device, &local_err);
3166     if (!bs) {
3167         bs = bdrv_lookup_bs(device, device, NULL);
3168         if (!bs) {
3169             error_free(local_err);
3170             error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
3171                       "Device '%s' not found", device);
3172         } else {
3173             error_propagate(errp, local_err);
3174         }
3175         return;
3176     }
3177 
3178     aio_context = bdrv_get_aio_context(bs);
3179     aio_context_acquire(aio_context);
3180 
3181     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_COMMIT_SOURCE, errp)) {
3182         goto out;
3183     }
3184 
3185     /* default top_bs is the active layer */
3186     top_bs = bs;
3187 
3188     if (has_top && top) {
3189         if (strcmp(bs->filename, top) != 0) {
3190             top_bs = bdrv_find_backing_image(bs, top);
3191         }
3192     }
3193 
3194     if (top_bs == NULL) {
3195         error_setg(errp, "Top image file %s not found", top ? top : "NULL");
3196         goto out;
3197     }
3198 
3199     assert(bdrv_get_aio_context(top_bs) == aio_context);
3200 
3201     if (has_base && base) {
3202         base_bs = bdrv_find_backing_image(top_bs, base);
3203     } else {
3204         base_bs = bdrv_find_base(top_bs);
3205     }
3206 
3207     if (base_bs == NULL) {
3208         error_setg(errp, QERR_BASE_NOT_FOUND, base ? base : "NULL");
3209         goto out;
3210     }
3211 
3212     assert(bdrv_get_aio_context(base_bs) == aio_context);
3213 
3214     for (iter = top_bs; iter != backing_bs(base_bs); iter = backing_bs(iter)) {
3215         if (bdrv_op_is_blocked(iter, BLOCK_OP_TYPE_COMMIT_TARGET, errp)) {
3216             goto out;
3217         }
3218     }
3219 
3220     /* Do not allow attempts to commit an image into itself */
3221     if (top_bs == base_bs) {
3222         error_setg(errp, "cannot commit an image into itself");
3223         goto out;
3224     }
3225 
3226     if (top_bs == bs) {
3227         if (has_backing_file) {
3228             error_setg(errp, "'backing-file' specified,"
3229                              " but 'top' is the active layer");
3230             goto out;
3231         }
3232         commit_active_start(has_job_id ? job_id : NULL, bs, base_bs,
3233                             BLOCK_JOB_DEFAULT, speed, on_error,
3234                             filter_node_name, NULL, NULL, false, &local_err);
3235     } else {
3236         BlockDriverState *overlay_bs = bdrv_find_overlay(bs, top_bs);
3237         if (bdrv_op_is_blocked(overlay_bs, BLOCK_OP_TYPE_COMMIT_TARGET, errp)) {
3238             goto out;
3239         }
3240         commit_start(has_job_id ? job_id : NULL, bs, base_bs, top_bs, speed,
3241                      on_error, has_backing_file ? backing_file : NULL,
3242                      filter_node_name, &local_err);
3243     }
3244     if (local_err != NULL) {
3245         error_propagate(errp, local_err);
3246         goto out;
3247     }
3248 
3249 out:
3250     aio_context_release(aio_context);
3251 }
3252 
3253 static BlockJob *do_drive_backup(DriveBackup *backup, BlockJobTxn *txn,
3254                                  Error **errp)
3255 {
3256     BlockDriverState *bs;
3257     BlockDriverState *target_bs;
3258     BlockDriverState *source = NULL;
3259     BlockJob *job = NULL;
3260     BdrvDirtyBitmap *bmap = NULL;
3261     AioContext *aio_context;
3262     QDict *options = NULL;
3263     Error *local_err = NULL;
3264     int flags;
3265     int64_t size;
3266     bool set_backing_hd = false;
3267 
3268     if (!backup->has_speed) {
3269         backup->speed = 0;
3270     }
3271     if (!backup->has_on_source_error) {
3272         backup->on_source_error = BLOCKDEV_ON_ERROR_REPORT;
3273     }
3274     if (!backup->has_on_target_error) {
3275         backup->on_target_error = BLOCKDEV_ON_ERROR_REPORT;
3276     }
3277     if (!backup->has_mode) {
3278         backup->mode = NEW_IMAGE_MODE_ABSOLUTE_PATHS;
3279     }
3280     if (!backup->has_job_id) {
3281         backup->job_id = NULL;
3282     }
3283     if (!backup->has_compress) {
3284         backup->compress = false;
3285     }
3286 
3287     bs = qmp_get_root_bs(backup->device, errp);
3288     if (!bs) {
3289         return NULL;
3290     }
3291 
3292     aio_context = bdrv_get_aio_context(bs);
3293     aio_context_acquire(aio_context);
3294 
3295     if (!backup->has_format) {
3296         backup->format = backup->mode == NEW_IMAGE_MODE_EXISTING ?
3297                          NULL : (char*) bs->drv->format_name;
3298     }
3299 
3300     /* Early check to avoid creating target */
3301     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_BACKUP_SOURCE, errp)) {
3302         goto out;
3303     }
3304 
3305     flags = bs->open_flags | BDRV_O_RDWR;
3306 
3307     /* See if we have a backing HD we can use to create our new image
3308      * on top of. */
3309     if (backup->sync == MIRROR_SYNC_MODE_TOP) {
3310         source = backing_bs(bs);
3311         if (!source) {
3312             backup->sync = MIRROR_SYNC_MODE_FULL;
3313         }
3314     }
3315     if (backup->sync == MIRROR_SYNC_MODE_NONE) {
3316         source = bs;
3317         flags |= BDRV_O_NO_BACKING;
3318         set_backing_hd = true;
3319     }
3320 
3321     size = bdrv_getlength(bs);
3322     if (size < 0) {
3323         error_setg_errno(errp, -size, "bdrv_getlength failed");
3324         goto out;
3325     }
3326 
3327     if (backup->mode != NEW_IMAGE_MODE_EXISTING) {
3328         assert(backup->format);
3329         if (source) {
3330             bdrv_img_create(backup->target, backup->format, source->filename,
3331                             source->drv->format_name, NULL,
3332                             size, flags, false, &local_err);
3333         } else {
3334             bdrv_img_create(backup->target, backup->format, NULL, NULL, NULL,
3335                             size, flags, false, &local_err);
3336         }
3337     }
3338 
3339     if (local_err) {
3340         error_propagate(errp, local_err);
3341         goto out;
3342     }
3343 
3344     if (backup->format) {
3345         if (!options) {
3346             options = qdict_new();
3347         }
3348         qdict_put_str(options, "driver", backup->format);
3349     }
3350 
3351     target_bs = bdrv_open(backup->target, NULL, options, flags, errp);
3352     if (!target_bs) {
3353         goto out;
3354     }
3355 
3356     bdrv_set_aio_context(target_bs, aio_context);
3357 
3358     if (set_backing_hd) {
3359         bdrv_set_backing_hd(target_bs, source, &local_err);
3360         if (local_err) {
3361             bdrv_unref(target_bs);
3362             goto out;
3363         }
3364     }
3365 
3366     if (backup->has_bitmap) {
3367         bmap = bdrv_find_dirty_bitmap(bs, backup->bitmap);
3368         if (!bmap) {
3369             error_setg(errp, "Bitmap '%s' could not be found", backup->bitmap);
3370             bdrv_unref(target_bs);
3371             goto out;
3372         }
3373     }
3374 
3375     job = backup_job_create(backup->job_id, bs, target_bs, backup->speed,
3376                             backup->sync, bmap, backup->compress,
3377                             backup->on_source_error, backup->on_target_error,
3378                             BLOCK_JOB_DEFAULT, NULL, NULL, txn, &local_err);
3379     bdrv_unref(target_bs);
3380     if (local_err != NULL) {
3381         error_propagate(errp, local_err);
3382         goto out;
3383     }
3384 
3385 out:
3386     aio_context_release(aio_context);
3387     return job;
3388 }
3389 
3390 void qmp_drive_backup(DriveBackup *arg, Error **errp)
3391 {
3392 
3393     BlockJob *job;
3394     job = do_drive_backup(arg, NULL, errp);
3395     if (job) {
3396         block_job_start(job);
3397     }
3398 }
3399 
3400 BlockDeviceInfoList *qmp_query_named_block_nodes(Error **errp)
3401 {
3402     return bdrv_named_nodes_list(errp);
3403 }
3404 
3405 BlockJob *do_blockdev_backup(BlockdevBackup *backup, BlockJobTxn *txn,
3406                              Error **errp)
3407 {
3408     BlockDriverState *bs;
3409     BlockDriverState *target_bs;
3410     Error *local_err = NULL;
3411     AioContext *aio_context;
3412     BlockJob *job = NULL;
3413 
3414     if (!backup->has_speed) {
3415         backup->speed = 0;
3416     }
3417     if (!backup->has_on_source_error) {
3418         backup->on_source_error = BLOCKDEV_ON_ERROR_REPORT;
3419     }
3420     if (!backup->has_on_target_error) {
3421         backup->on_target_error = BLOCKDEV_ON_ERROR_REPORT;
3422     }
3423     if (!backup->has_job_id) {
3424         backup->job_id = NULL;
3425     }
3426     if (!backup->has_compress) {
3427         backup->compress = false;
3428     }
3429 
3430     bs = qmp_get_root_bs(backup->device, errp);
3431     if (!bs) {
3432         return NULL;
3433     }
3434 
3435     aio_context = bdrv_get_aio_context(bs);
3436     aio_context_acquire(aio_context);
3437 
3438     target_bs = bdrv_lookup_bs(backup->target, backup->target, errp);
3439     if (!target_bs) {
3440         goto out;
3441     }
3442 
3443     if (bdrv_get_aio_context(target_bs) != aio_context) {
3444         if (!bdrv_has_blk(target_bs)) {
3445             /* The target BDS is not attached, we can safely move it to another
3446              * AioContext. */
3447             bdrv_set_aio_context(target_bs, aio_context);
3448         } else {
3449             error_setg(errp, "Target is attached to a different thread from "
3450                              "source.");
3451             goto out;
3452         }
3453     }
3454     job = backup_job_create(backup->job_id, bs, target_bs, backup->speed,
3455                             backup->sync, NULL, backup->compress,
3456                             backup->on_source_error, backup->on_target_error,
3457                             BLOCK_JOB_DEFAULT, NULL, NULL, txn, &local_err);
3458     if (local_err != NULL) {
3459         error_propagate(errp, local_err);
3460     }
3461 out:
3462     aio_context_release(aio_context);
3463     return job;
3464 }
3465 
3466 void qmp_blockdev_backup(BlockdevBackup *arg, Error **errp)
3467 {
3468     BlockJob *job;
3469     job = do_blockdev_backup(arg, NULL, errp);
3470     if (job) {
3471         block_job_start(job);
3472     }
3473 }
3474 
3475 /* Parameter check and block job starting for drive mirroring.
3476  * Caller should hold @device and @target's aio context (must be the same).
3477  **/
3478 static void blockdev_mirror_common(const char *job_id, BlockDriverState *bs,
3479                                    BlockDriverState *target,
3480                                    bool has_replaces, const char *replaces,
3481                                    enum MirrorSyncMode sync,
3482                                    BlockMirrorBackingMode backing_mode,
3483                                    bool has_speed, int64_t speed,
3484                                    bool has_granularity, uint32_t granularity,
3485                                    bool has_buf_size, int64_t buf_size,
3486                                    bool has_on_source_error,
3487                                    BlockdevOnError on_source_error,
3488                                    bool has_on_target_error,
3489                                    BlockdevOnError on_target_error,
3490                                    bool has_unmap, bool unmap,
3491                                    bool has_filter_node_name,
3492                                    const char *filter_node_name,
3493                                    Error **errp)
3494 {
3495 
3496     if (!has_speed) {
3497         speed = 0;
3498     }
3499     if (!has_on_source_error) {
3500         on_source_error = BLOCKDEV_ON_ERROR_REPORT;
3501     }
3502     if (!has_on_target_error) {
3503         on_target_error = BLOCKDEV_ON_ERROR_REPORT;
3504     }
3505     if (!has_granularity) {
3506         granularity = 0;
3507     }
3508     if (!has_buf_size) {
3509         buf_size = 0;
3510     }
3511     if (!has_unmap) {
3512         unmap = true;
3513     }
3514     if (!has_filter_node_name) {
3515         filter_node_name = NULL;
3516     }
3517 
3518     if (granularity != 0 && (granularity < 512 || granularity > 1048576 * 64)) {
3519         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "granularity",
3520                    "a value in range [512B, 64MB]");
3521         return;
3522     }
3523     if (granularity & (granularity - 1)) {
3524         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "granularity",
3525                    "power of 2");
3526         return;
3527     }
3528 
3529     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_MIRROR_SOURCE, errp)) {
3530         return;
3531     }
3532     if (bdrv_op_is_blocked(target, BLOCK_OP_TYPE_MIRROR_TARGET, errp)) {
3533         return;
3534     }
3535 
3536     if (!bs->backing && sync == MIRROR_SYNC_MODE_TOP) {
3537         sync = MIRROR_SYNC_MODE_FULL;
3538     }
3539 
3540     /* pass the node name to replace to mirror start since it's loose coupling
3541      * and will allow to check whether the node still exist at mirror completion
3542      */
3543     mirror_start(job_id, bs, target,
3544                  has_replaces ? replaces : NULL,
3545                  speed, granularity, buf_size, sync, backing_mode,
3546                  on_source_error, on_target_error, unmap, filter_node_name,
3547                  errp);
3548 }
3549 
3550 void qmp_drive_mirror(DriveMirror *arg, Error **errp)
3551 {
3552     BlockDriverState *bs;
3553     BlockDriverState *source, *target_bs;
3554     AioContext *aio_context;
3555     BlockMirrorBackingMode backing_mode;
3556     Error *local_err = NULL;
3557     QDict *options = NULL;
3558     int flags;
3559     int64_t size;
3560     const char *format = arg->format;
3561 
3562     bs = qmp_get_root_bs(arg->device, errp);
3563     if (!bs) {
3564         return;
3565     }
3566 
3567     /* Early check to avoid creating target */
3568     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_MIRROR_SOURCE, errp)) {
3569         return;
3570     }
3571 
3572     aio_context = bdrv_get_aio_context(bs);
3573     aio_context_acquire(aio_context);
3574 
3575     if (!arg->has_mode) {
3576         arg->mode = NEW_IMAGE_MODE_ABSOLUTE_PATHS;
3577     }
3578 
3579     if (!arg->has_format) {
3580         format = (arg->mode == NEW_IMAGE_MODE_EXISTING
3581                   ? NULL : bs->drv->format_name);
3582     }
3583 
3584     flags = bs->open_flags | BDRV_O_RDWR;
3585     source = backing_bs(bs);
3586     if (!source && arg->sync == MIRROR_SYNC_MODE_TOP) {
3587         arg->sync = MIRROR_SYNC_MODE_FULL;
3588     }
3589     if (arg->sync == MIRROR_SYNC_MODE_NONE) {
3590         source = bs;
3591     }
3592 
3593     size = bdrv_getlength(bs);
3594     if (size < 0) {
3595         error_setg_errno(errp, -size, "bdrv_getlength failed");
3596         goto out;
3597     }
3598 
3599     if (arg->has_replaces) {
3600         BlockDriverState *to_replace_bs;
3601         AioContext *replace_aio_context;
3602         int64_t replace_size;
3603 
3604         if (!arg->has_node_name) {
3605             error_setg(errp, "a node-name must be provided when replacing a"
3606                              " named node of the graph");
3607             goto out;
3608         }
3609 
3610         to_replace_bs = check_to_replace_node(bs, arg->replaces, &local_err);
3611 
3612         if (!to_replace_bs) {
3613             error_propagate(errp, local_err);
3614             goto out;
3615         }
3616 
3617         replace_aio_context = bdrv_get_aio_context(to_replace_bs);
3618         aio_context_acquire(replace_aio_context);
3619         replace_size = bdrv_getlength(to_replace_bs);
3620         aio_context_release(replace_aio_context);
3621 
3622         if (size != replace_size) {
3623             error_setg(errp, "cannot replace image with a mirror image of "
3624                              "different size");
3625             goto out;
3626         }
3627     }
3628 
3629     if (arg->mode == NEW_IMAGE_MODE_ABSOLUTE_PATHS) {
3630         backing_mode = MIRROR_SOURCE_BACKING_CHAIN;
3631     } else {
3632         backing_mode = MIRROR_OPEN_BACKING_CHAIN;
3633     }
3634 
3635     /* Don't open backing image in create() */
3636     flags |= BDRV_O_NO_BACKING;
3637 
3638     if ((arg->sync == MIRROR_SYNC_MODE_FULL || !source)
3639         && arg->mode != NEW_IMAGE_MODE_EXISTING)
3640     {
3641         /* create new image w/o backing file */
3642         assert(format);
3643         bdrv_img_create(arg->target, format,
3644                         NULL, NULL, NULL, size, flags, false, &local_err);
3645     } else {
3646         switch (arg->mode) {
3647         case NEW_IMAGE_MODE_EXISTING:
3648             break;
3649         case NEW_IMAGE_MODE_ABSOLUTE_PATHS:
3650             /* create new image with backing file */
3651             bdrv_img_create(arg->target, format,
3652                             source->filename,
3653                             source->drv->format_name,
3654                             NULL, size, flags, false, &local_err);
3655             break;
3656         default:
3657             abort();
3658         }
3659     }
3660 
3661     if (local_err) {
3662         error_propagate(errp, local_err);
3663         goto out;
3664     }
3665 
3666     options = qdict_new();
3667     if (arg->has_node_name) {
3668         qdict_put_str(options, "node-name", arg->node_name);
3669     }
3670     if (format) {
3671         qdict_put_str(options, "driver", format);
3672     }
3673 
3674     /* Mirroring takes care of copy-on-write using the source's backing
3675      * file.
3676      */
3677     target_bs = bdrv_open(arg->target, NULL, options, flags, errp);
3678     if (!target_bs) {
3679         goto out;
3680     }
3681 
3682     bdrv_set_aio_context(target_bs, aio_context);
3683 
3684     blockdev_mirror_common(arg->has_job_id ? arg->job_id : NULL, bs, target_bs,
3685                            arg->has_replaces, arg->replaces, arg->sync,
3686                            backing_mode, arg->has_speed, arg->speed,
3687                            arg->has_granularity, arg->granularity,
3688                            arg->has_buf_size, arg->buf_size,
3689                            arg->has_on_source_error, arg->on_source_error,
3690                            arg->has_on_target_error, arg->on_target_error,
3691                            arg->has_unmap, arg->unmap,
3692                            false, NULL,
3693                            &local_err);
3694     bdrv_unref(target_bs);
3695     error_propagate(errp, local_err);
3696 out:
3697     aio_context_release(aio_context);
3698 }
3699 
3700 void qmp_blockdev_mirror(bool has_job_id, const char *job_id,
3701                          const char *device, const char *target,
3702                          bool has_replaces, const char *replaces,
3703                          MirrorSyncMode sync,
3704                          bool has_speed, int64_t speed,
3705                          bool has_granularity, uint32_t granularity,
3706                          bool has_buf_size, int64_t buf_size,
3707                          bool has_on_source_error,
3708                          BlockdevOnError on_source_error,
3709                          bool has_on_target_error,
3710                          BlockdevOnError on_target_error,
3711                          bool has_filter_node_name,
3712                          const char *filter_node_name,
3713                          Error **errp)
3714 {
3715     BlockDriverState *bs;
3716     BlockDriverState *target_bs;
3717     AioContext *aio_context;
3718     BlockMirrorBackingMode backing_mode = MIRROR_LEAVE_BACKING_CHAIN;
3719     Error *local_err = NULL;
3720 
3721     bs = qmp_get_root_bs(device, errp);
3722     if (!bs) {
3723         return;
3724     }
3725 
3726     target_bs = bdrv_lookup_bs(target, target, errp);
3727     if (!target_bs) {
3728         return;
3729     }
3730 
3731     aio_context = bdrv_get_aio_context(bs);
3732     aio_context_acquire(aio_context);
3733 
3734     bdrv_set_aio_context(target_bs, aio_context);
3735 
3736     blockdev_mirror_common(has_job_id ? job_id : NULL, bs, target_bs,
3737                            has_replaces, replaces, sync, backing_mode,
3738                            has_speed, speed,
3739                            has_granularity, granularity,
3740                            has_buf_size, buf_size,
3741                            has_on_source_error, on_source_error,
3742                            has_on_target_error, on_target_error,
3743                            true, true,
3744                            has_filter_node_name, filter_node_name,
3745                            &local_err);
3746     error_propagate(errp, local_err);
3747 
3748     aio_context_release(aio_context);
3749 }
3750 
3751 /* Get a block job using its ID and acquire its AioContext */
3752 static BlockJob *find_block_job(const char *id, AioContext **aio_context,
3753                                 Error **errp)
3754 {
3755     BlockJob *job;
3756 
3757     assert(id != NULL);
3758 
3759     *aio_context = NULL;
3760 
3761     job = block_job_get(id);
3762 
3763     if (!job) {
3764         error_set(errp, ERROR_CLASS_DEVICE_NOT_ACTIVE,
3765                   "Block job '%s' not found", id);
3766         return NULL;
3767     }
3768 
3769     *aio_context = blk_get_aio_context(job->blk);
3770     aio_context_acquire(*aio_context);
3771 
3772     return job;
3773 }
3774 
3775 void qmp_block_job_set_speed(const char *device, int64_t speed, Error **errp)
3776 {
3777     AioContext *aio_context;
3778     BlockJob *job = find_block_job(device, &aio_context, errp);
3779 
3780     if (!job) {
3781         return;
3782     }
3783 
3784     block_job_set_speed(job, speed, errp);
3785     aio_context_release(aio_context);
3786 }
3787 
3788 void qmp_block_job_cancel(const char *device,
3789                           bool has_force, bool force, Error **errp)
3790 {
3791     AioContext *aio_context;
3792     BlockJob *job = find_block_job(device, &aio_context, errp);
3793 
3794     if (!job) {
3795         return;
3796     }
3797 
3798     if (!has_force) {
3799         force = false;
3800     }
3801 
3802     if (block_job_user_paused(job) && !force) {
3803         error_setg(errp, "The block job for device '%s' is currently paused",
3804                    device);
3805         goto out;
3806     }
3807 
3808     trace_qmp_block_job_cancel(job);
3809     block_job_cancel(job);
3810 out:
3811     aio_context_release(aio_context);
3812 }
3813 
3814 void qmp_block_job_pause(const char *device, Error **errp)
3815 {
3816     AioContext *aio_context;
3817     BlockJob *job = find_block_job(device, &aio_context, errp);
3818 
3819     if (!job || block_job_user_paused(job)) {
3820         return;
3821     }
3822 
3823     trace_qmp_block_job_pause(job);
3824     block_job_user_pause(job);
3825     aio_context_release(aio_context);
3826 }
3827 
3828 void qmp_block_job_resume(const char *device, Error **errp)
3829 {
3830     AioContext *aio_context;
3831     BlockJob *job = find_block_job(device, &aio_context, errp);
3832 
3833     if (!job || !block_job_user_paused(job)) {
3834         return;
3835     }
3836 
3837     trace_qmp_block_job_resume(job);
3838     block_job_user_resume(job);
3839     aio_context_release(aio_context);
3840 }
3841 
3842 void qmp_block_job_complete(const char *device, Error **errp)
3843 {
3844     AioContext *aio_context;
3845     BlockJob *job = find_block_job(device, &aio_context, errp);
3846 
3847     if (!job) {
3848         return;
3849     }
3850 
3851     trace_qmp_block_job_complete(job);
3852     block_job_complete(job, errp);
3853     aio_context_release(aio_context);
3854 }
3855 
3856 void qmp_change_backing_file(const char *device,
3857                              const char *image_node_name,
3858                              const char *backing_file,
3859                              Error **errp)
3860 {
3861     BlockDriverState *bs = NULL;
3862     AioContext *aio_context;
3863     BlockDriverState *image_bs = NULL;
3864     Error *local_err = NULL;
3865     bool ro;
3866     int open_flags;
3867     int ret;
3868 
3869     bs = qmp_get_root_bs(device, errp);
3870     if (!bs) {
3871         return;
3872     }
3873 
3874     aio_context = bdrv_get_aio_context(bs);
3875     aio_context_acquire(aio_context);
3876 
3877     image_bs = bdrv_lookup_bs(NULL, image_node_name, &local_err);
3878     if (local_err) {
3879         error_propagate(errp, local_err);
3880         goto out;
3881     }
3882 
3883     if (!image_bs) {
3884         error_setg(errp, "image file not found");
3885         goto out;
3886     }
3887 
3888     if (bdrv_find_base(image_bs) == image_bs) {
3889         error_setg(errp, "not allowing backing file change on an image "
3890                          "without a backing file");
3891         goto out;
3892     }
3893 
3894     /* even though we are not necessarily operating on bs, we need it to
3895      * determine if block ops are currently prohibited on the chain */
3896     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_CHANGE, errp)) {
3897         goto out;
3898     }
3899 
3900     /* final sanity check */
3901     if (!bdrv_chain_contains(bs, image_bs)) {
3902         error_setg(errp, "'%s' and image file are not in the same chain",
3903                    device);
3904         goto out;
3905     }
3906 
3907     /* if not r/w, reopen to make r/w */
3908     open_flags = image_bs->open_flags;
3909     ro = bdrv_is_read_only(image_bs);
3910 
3911     if (ro) {
3912         bdrv_reopen(image_bs, open_flags | BDRV_O_RDWR, &local_err);
3913         if (local_err) {
3914             error_propagate(errp, local_err);
3915             goto out;
3916         }
3917     }
3918 
3919     ret = bdrv_change_backing_file(image_bs, backing_file,
3920                                image_bs->drv ? image_bs->drv->format_name : "");
3921 
3922     if (ret < 0) {
3923         error_setg_errno(errp, -ret, "Could not change backing file to '%s'",
3924                          backing_file);
3925         /* don't exit here, so we can try to restore open flags if
3926          * appropriate */
3927     }
3928 
3929     if (ro) {
3930         bdrv_reopen(image_bs, open_flags, &local_err);
3931         error_propagate(errp, local_err);
3932     }
3933 
3934 out:
3935     aio_context_release(aio_context);
3936 }
3937 
3938 void hmp_drive_add_node(Monitor *mon, const char *optstr)
3939 {
3940     QemuOpts *opts;
3941     QDict *qdict;
3942     Error *local_err = NULL;
3943 
3944     opts = qemu_opts_parse_noisily(&qemu_drive_opts, optstr, false);
3945     if (!opts) {
3946         return;
3947     }
3948 
3949     qdict = qemu_opts_to_qdict(opts, NULL);
3950 
3951     if (!qdict_get_try_str(qdict, "node-name")) {
3952         QDECREF(qdict);
3953         error_report("'node-name' needs to be specified");
3954         goto out;
3955     }
3956 
3957     BlockDriverState *bs = bds_tree_init(qdict, &local_err);
3958     if (!bs) {
3959         error_report_err(local_err);
3960         goto out;
3961     }
3962 
3963     QTAILQ_INSERT_TAIL(&monitor_bdrv_states, bs, monitor_list);
3964 
3965 out:
3966     qemu_opts_del(opts);
3967 }
3968 
3969 void qmp_blockdev_add(BlockdevOptions *options, Error **errp)
3970 {
3971     BlockDriverState *bs;
3972     QObject *obj;
3973     Visitor *v = qobject_output_visitor_new(&obj);
3974     QDict *qdict;
3975     const QDictEntry *ent;
3976     Error *local_err = NULL;
3977 
3978     visit_type_BlockdevOptions(v, NULL, &options, &local_err);
3979     if (local_err) {
3980         error_propagate(errp, local_err);
3981         goto fail;
3982     }
3983 
3984     visit_complete(v, &obj);
3985     qdict = qobject_to_qdict(obj);
3986 
3987     qdict_flatten(qdict);
3988 
3989     /*
3990      * Rewrite "backing": null to "backing": ""
3991      * TODO Rewrite "" to null instead, and perhaps not even here
3992      */
3993     for (ent = qdict_first(qdict); ent; ent = qdict_next(qdict, ent)) {
3994         char *dot = strrchr(ent->key, '.');
3995 
3996         if (!strcmp(dot ? dot + 1 : ent->key, "backing")
3997             && qobject_type(ent->value) == QTYPE_QNULL) {
3998             qdict_put(qdict, ent->key, qstring_new());
3999         }
4000     }
4001 
4002     if (!qdict_get_try_str(qdict, "node-name")) {
4003         error_setg(errp, "'node-name' must be specified for the root node");
4004         goto fail;
4005     }
4006 
4007     bs = bds_tree_init(qdict, errp);
4008     if (!bs) {
4009         goto fail;
4010     }
4011 
4012     QTAILQ_INSERT_TAIL(&monitor_bdrv_states, bs, monitor_list);
4013 
4014 fail:
4015     visit_free(v);
4016 }
4017 
4018 void qmp_blockdev_del(const char *node_name, Error **errp)
4019 {
4020     AioContext *aio_context;
4021     BlockDriverState *bs;
4022 
4023     bs = bdrv_find_node(node_name);
4024     if (!bs) {
4025         error_setg(errp, "Cannot find node %s", node_name);
4026         return;
4027     }
4028     if (bdrv_has_blk(bs)) {
4029         error_setg(errp, "Node %s is in use", node_name);
4030         return;
4031     }
4032     aio_context = bdrv_get_aio_context(bs);
4033     aio_context_acquire(aio_context);
4034 
4035     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_DRIVE_DEL, errp)) {
4036         goto out;
4037     }
4038 
4039     if (!bs->monitor_list.tqe_prev) {
4040         error_setg(errp, "Node %s is not owned by the monitor",
4041                    bs->node_name);
4042         goto out;
4043     }
4044 
4045     if (bs->refcnt > 1) {
4046         error_setg(errp, "Block device %s is in use",
4047                    bdrv_get_device_or_node_name(bs));
4048         goto out;
4049     }
4050 
4051     QTAILQ_REMOVE(&monitor_bdrv_states, bs, monitor_list);
4052     bdrv_unref(bs);
4053 
4054 out:
4055     aio_context_release(aio_context);
4056 }
4057 
4058 static BdrvChild *bdrv_find_child(BlockDriverState *parent_bs,
4059                                   const char *child_name)
4060 {
4061     BdrvChild *child;
4062 
4063     QLIST_FOREACH(child, &parent_bs->children, next) {
4064         if (strcmp(child->name, child_name) == 0) {
4065             return child;
4066         }
4067     }
4068 
4069     return NULL;
4070 }
4071 
4072 void qmp_x_blockdev_change(const char *parent, bool has_child,
4073                            const char *child, bool has_node,
4074                            const char *node, Error **errp)
4075 {
4076     BlockDriverState *parent_bs, *new_bs = NULL;
4077     BdrvChild *p_child;
4078 
4079     parent_bs = bdrv_lookup_bs(parent, parent, errp);
4080     if (!parent_bs) {
4081         return;
4082     }
4083 
4084     if (has_child == has_node) {
4085         if (has_child) {
4086             error_setg(errp, "The parameters child and node are in conflict");
4087         } else {
4088             error_setg(errp, "Either child or node must be specified");
4089         }
4090         return;
4091     }
4092 
4093     if (has_child) {
4094         p_child = bdrv_find_child(parent_bs, child);
4095         if (!p_child) {
4096             error_setg(errp, "Node '%s' does not have child '%s'",
4097                        parent, child);
4098             return;
4099         }
4100         bdrv_del_child(parent_bs, p_child, errp);
4101     }
4102 
4103     if (has_node) {
4104         new_bs = bdrv_find_node(node);
4105         if (!new_bs) {
4106             error_setg(errp, "Node '%s' not found", node);
4107             return;
4108         }
4109         bdrv_add_child(parent_bs, new_bs, errp);
4110     }
4111 }
4112 
4113 BlockJobInfoList *qmp_query_block_jobs(Error **errp)
4114 {
4115     BlockJobInfoList *head = NULL, **p_next = &head;
4116     BlockJob *job;
4117 
4118     for (job = block_job_next(NULL); job; job = block_job_next(job)) {
4119         BlockJobInfoList *elem;
4120         AioContext *aio_context;
4121 
4122         if (block_job_is_internal(job)) {
4123             continue;
4124         }
4125         elem = g_new0(BlockJobInfoList, 1);
4126         aio_context = blk_get_aio_context(job->blk);
4127         aio_context_acquire(aio_context);
4128         elem->value = block_job_query(job, errp);
4129         aio_context_release(aio_context);
4130         if (!elem->value) {
4131             g_free(elem);
4132             qapi_free_BlockJobInfoList(head);
4133             return NULL;
4134         }
4135         *p_next = elem;
4136         p_next = &elem->next;
4137     }
4138 
4139     return head;
4140 }
4141 
4142 void qmp_x_blockdev_set_iothread(const char *node_name, StrOrNull *iothread,
4143                                  bool has_force, bool force, Error **errp)
4144 {
4145     AioContext *old_context;
4146     AioContext *new_context;
4147     BlockDriverState *bs;
4148 
4149     bs = bdrv_find_node(node_name);
4150     if (!bs) {
4151         error_setg(errp, "Cannot find node %s", node_name);
4152         return;
4153     }
4154 
4155     /* Protects against accidents. */
4156     if (!(has_force && force) && bdrv_has_blk(bs)) {
4157         error_setg(errp, "Node %s is associated with a BlockBackend and could "
4158                          "be in use (use force=true to override this check)",
4159                          node_name);
4160         return;
4161     }
4162 
4163     if (iothread->type == QTYPE_QSTRING) {
4164         IOThread *obj = iothread_by_id(iothread->u.s);
4165         if (!obj) {
4166             error_setg(errp, "Cannot find iothread %s", iothread->u.s);
4167             return;
4168         }
4169 
4170         new_context = iothread_get_aio_context(obj);
4171     } else {
4172         new_context = qemu_get_aio_context();
4173     }
4174 
4175     old_context = bdrv_get_aio_context(bs);
4176     aio_context_acquire(old_context);
4177 
4178     bdrv_set_aio_context(bs, new_context);
4179 
4180     aio_context_release(old_context);
4181 }
4182 
4183 QemuOptsList qemu_common_drive_opts = {
4184     .name = "drive",
4185     .head = QTAILQ_HEAD_INITIALIZER(qemu_common_drive_opts.head),
4186     .desc = {
4187         {
4188             .name = "snapshot",
4189             .type = QEMU_OPT_BOOL,
4190             .help = "enable/disable snapshot mode",
4191         },{
4192             .name = "aio",
4193             .type = QEMU_OPT_STRING,
4194             .help = "host AIO implementation (threads, native)",
4195         },{
4196             .name = BDRV_OPT_CACHE_WB,
4197             .type = QEMU_OPT_BOOL,
4198             .help = "Enable writeback mode",
4199         },{
4200             .name = "format",
4201             .type = QEMU_OPT_STRING,
4202             .help = "disk format (raw, qcow2, ...)",
4203         },{
4204             .name = "rerror",
4205             .type = QEMU_OPT_STRING,
4206             .help = "read error action",
4207         },{
4208             .name = "werror",
4209             .type = QEMU_OPT_STRING,
4210             .help = "write error action",
4211         },{
4212             .name = BDRV_OPT_READ_ONLY,
4213             .type = QEMU_OPT_BOOL,
4214             .help = "open drive file as read-only",
4215         },
4216 
4217         THROTTLE_OPTS,
4218 
4219         {
4220             .name = "throttling.group",
4221             .type = QEMU_OPT_STRING,
4222             .help = "name of the block throttling group",
4223         },{
4224             .name = "copy-on-read",
4225             .type = QEMU_OPT_BOOL,
4226             .help = "copy read data from backing file into image file",
4227         },{
4228             .name = "detect-zeroes",
4229             .type = QEMU_OPT_STRING,
4230             .help = "try to optimize zero writes (off, on, unmap)",
4231         },{
4232             .name = "stats-account-invalid",
4233             .type = QEMU_OPT_BOOL,
4234             .help = "whether to account for invalid I/O operations "
4235                     "in the statistics",
4236         },{
4237             .name = "stats-account-failed",
4238             .type = QEMU_OPT_BOOL,
4239             .help = "whether to account for failed I/O operations "
4240                     "in the statistics",
4241         },
4242         { /* end of list */ }
4243     },
4244 };
4245 
4246 QemuOptsList qemu_drive_opts = {
4247     .name = "drive",
4248     .head = QTAILQ_HEAD_INITIALIZER(qemu_drive_opts.head),
4249     .desc = {
4250         /*
4251          * no elements => accept any params
4252          * validation will happen later
4253          */
4254         { /* end of list */ }
4255     },
4256 };
4257