xref: /openbmc/qemu/blockdev.c (revision 9cbb6362)
1 /*
2  * QEMU host block devices
3  *
4  * Copyright (c) 2003-2008 Fabrice Bellard
5  *
6  * This work is licensed under the terms of the GNU GPL, version 2 or
7  * later.  See the COPYING file in the top-level directory.
8  *
9  * This file incorporates work covered by the following copyright and
10  * permission notice:
11  *
12  * Copyright (c) 2003-2008 Fabrice Bellard
13  *
14  * Permission is hereby granted, free of charge, to any person obtaining a copy
15  * of this software and associated documentation files (the "Software"), to deal
16  * in the Software without restriction, including without limitation the rights
17  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
18  * copies of the Software, and to permit persons to whom the Software is
19  * furnished to do so, subject to the following conditions:
20  *
21  * The above copyright notice and this permission notice shall be included in
22  * all copies or substantial portions of the Software.
23  *
24  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
25  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
26  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
27  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
28  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
29  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
30  * THE SOFTWARE.
31  */
32 
33 #include "qemu/osdep.h"
34 #include "sysemu/block-backend.h"
35 #include "sysemu/blockdev.h"
36 #include "hw/block/block.h"
37 #include "block/blockjob.h"
38 #include "block/throttle-groups.h"
39 #include "monitor/monitor.h"
40 #include "qemu/error-report.h"
41 #include "qemu/option.h"
42 #include "qemu/config-file.h"
43 #include "qapi/qmp/types.h"
44 #include "qapi-visit.h"
45 #include "qapi/qmp/qerror.h"
46 #include "qapi/qobject-output-visitor.h"
47 #include "sysemu/sysemu.h"
48 #include "sysemu/iothread.h"
49 #include "block/block_int.h"
50 #include "qmp-commands.h"
51 #include "block/trace.h"
52 #include "sysemu/arch_init.h"
53 #include "sysemu/qtest.h"
54 #include "qemu/cutils.h"
55 #include "qemu/help_option.h"
56 #include "qemu/throttle-options.h"
57 
58 static QTAILQ_HEAD(, BlockDriverState) monitor_bdrv_states =
59     QTAILQ_HEAD_INITIALIZER(monitor_bdrv_states);
60 
61 static int do_open_tray(const char *blk_name, const char *qdev_id,
62                         bool force, Error **errp);
63 
64 static const char *const if_name[IF_COUNT] = {
65     [IF_NONE] = "none",
66     [IF_IDE] = "ide",
67     [IF_SCSI] = "scsi",
68     [IF_FLOPPY] = "floppy",
69     [IF_PFLASH] = "pflash",
70     [IF_MTD] = "mtd",
71     [IF_SD] = "sd",
72     [IF_VIRTIO] = "virtio",
73     [IF_XEN] = "xen",
74 };
75 
76 static int if_max_devs[IF_COUNT] = {
77     /*
78      * Do not change these numbers!  They govern how drive option
79      * index maps to unit and bus.  That mapping is ABI.
80      *
81      * All controllers used to implement if=T drives need to support
82      * if_max_devs[T] units, for any T with if_max_devs[T] != 0.
83      * Otherwise, some index values map to "impossible" bus, unit
84      * values.
85      *
86      * For instance, if you change [IF_SCSI] to 255, -drive
87      * if=scsi,index=12 no longer means bus=1,unit=5, but
88      * bus=0,unit=12.  With an lsi53c895a controller (7 units max),
89      * the drive can't be set up.  Regression.
90      */
91     [IF_IDE] = 2,
92     [IF_SCSI] = 7,
93 };
94 
95 /**
96  * Boards may call this to offer board-by-board overrides
97  * of the default, global values.
98  */
99 void override_max_devs(BlockInterfaceType type, int max_devs)
100 {
101     BlockBackend *blk;
102     DriveInfo *dinfo;
103 
104     if (max_devs <= 0) {
105         return;
106     }
107 
108     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
109         dinfo = blk_legacy_dinfo(blk);
110         if (dinfo->type == type) {
111             fprintf(stderr, "Cannot override units-per-bus property of"
112                     " the %s interface, because a drive of that type has"
113                     " already been added.\n", if_name[type]);
114             g_assert_not_reached();
115         }
116     }
117 
118     if_max_devs[type] = max_devs;
119 }
120 
121 /*
122  * We automatically delete the drive when a device using it gets
123  * unplugged.  Questionable feature, but we can't just drop it.
124  * Device models call blockdev_mark_auto_del() to schedule the
125  * automatic deletion, and generic qdev code calls blockdev_auto_del()
126  * when deletion is actually safe.
127  */
128 void blockdev_mark_auto_del(BlockBackend *blk)
129 {
130     DriveInfo *dinfo = blk_legacy_dinfo(blk);
131     BlockDriverState *bs = blk_bs(blk);
132     AioContext *aio_context;
133 
134     if (!dinfo) {
135         return;
136     }
137 
138     if (bs) {
139         aio_context = bdrv_get_aio_context(bs);
140         aio_context_acquire(aio_context);
141 
142         if (bs->job) {
143             block_job_cancel(bs->job);
144         }
145 
146         aio_context_release(aio_context);
147     }
148 
149     dinfo->auto_del = 1;
150 }
151 
152 void blockdev_auto_del(BlockBackend *blk)
153 {
154     DriveInfo *dinfo = blk_legacy_dinfo(blk);
155 
156     if (dinfo && dinfo->auto_del) {
157         monitor_remove_blk(blk);
158         blk_unref(blk);
159     }
160 }
161 
162 /**
163  * Returns the current mapping of how many units per bus
164  * a particular interface can support.
165  *
166  *  A positive integer indicates n units per bus.
167  *  0 implies the mapping has not been established.
168  * -1 indicates an invalid BlockInterfaceType was given.
169  */
170 int drive_get_max_devs(BlockInterfaceType type)
171 {
172     if (type >= IF_IDE && type < IF_COUNT) {
173         return if_max_devs[type];
174     }
175 
176     return -1;
177 }
178 
179 static int drive_index_to_bus_id(BlockInterfaceType type, int index)
180 {
181     int max_devs = if_max_devs[type];
182     return max_devs ? index / max_devs : 0;
183 }
184 
185 static int drive_index_to_unit_id(BlockInterfaceType type, int index)
186 {
187     int max_devs = if_max_devs[type];
188     return max_devs ? index % max_devs : index;
189 }
190 
191 QemuOpts *drive_def(const char *optstr)
192 {
193     return qemu_opts_parse_noisily(qemu_find_opts("drive"), optstr, false);
194 }
195 
196 QemuOpts *drive_add(BlockInterfaceType type, int index, const char *file,
197                     const char *optstr)
198 {
199     QemuOpts *opts;
200 
201     opts = drive_def(optstr);
202     if (!opts) {
203         return NULL;
204     }
205     if (type != IF_DEFAULT) {
206         qemu_opt_set(opts, "if", if_name[type], &error_abort);
207     }
208     if (index >= 0) {
209         qemu_opt_set_number(opts, "index", index, &error_abort);
210     }
211     if (file)
212         qemu_opt_set(opts, "file", file, &error_abort);
213     return opts;
214 }
215 
216 DriveInfo *drive_get(BlockInterfaceType type, int bus, int unit)
217 {
218     BlockBackend *blk;
219     DriveInfo *dinfo;
220 
221     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
222         dinfo = blk_legacy_dinfo(blk);
223         if (dinfo && dinfo->type == type
224             && dinfo->bus == bus && dinfo->unit == unit) {
225             return dinfo;
226         }
227     }
228 
229     return NULL;
230 }
231 
232 void drive_check_orphaned(void)
233 {
234     BlockBackend *blk;
235     DriveInfo *dinfo;
236     Location loc;
237     bool orphans = false;
238 
239     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
240         dinfo = blk_legacy_dinfo(blk);
241         if (!blk_get_attached_dev(blk) && !dinfo->is_default &&
242             dinfo->type != IF_NONE) {
243             loc_push_none(&loc);
244             qemu_opts_loc_restore(dinfo->opts);
245             error_report("machine type does not support"
246                          " if=%s,bus=%d,unit=%d",
247                          if_name[dinfo->type], dinfo->bus, dinfo->unit);
248             loc_pop(&loc);
249             orphans = true;
250         }
251     }
252 
253     if (orphans) {
254         exit(1);
255     }
256 }
257 
258 DriveInfo *drive_get_by_index(BlockInterfaceType type, int index)
259 {
260     return drive_get(type,
261                      drive_index_to_bus_id(type, index),
262                      drive_index_to_unit_id(type, index));
263 }
264 
265 int drive_get_max_bus(BlockInterfaceType type)
266 {
267     int max_bus;
268     BlockBackend *blk;
269     DriveInfo *dinfo;
270 
271     max_bus = -1;
272     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
273         dinfo = blk_legacy_dinfo(blk);
274         if (dinfo && dinfo->type == type && dinfo->bus > max_bus) {
275             max_bus = dinfo->bus;
276         }
277     }
278     return max_bus;
279 }
280 
281 /* Get a block device.  This should only be used for single-drive devices
282    (e.g. SD/Floppy/MTD).  Multi-disk devices (scsi/ide) should use the
283    appropriate bus.  */
284 DriveInfo *drive_get_next(BlockInterfaceType type)
285 {
286     static int next_block_unit[IF_COUNT];
287 
288     return drive_get(type, 0, next_block_unit[type]++);
289 }
290 
291 static void bdrv_format_print(void *opaque, const char *name)
292 {
293     error_printf(" %s", name);
294 }
295 
296 typedef struct {
297     QEMUBH *bh;
298     BlockDriverState *bs;
299 } BDRVPutRefBH;
300 
301 static int parse_block_error_action(const char *buf, bool is_read, Error **errp)
302 {
303     if (!strcmp(buf, "ignore")) {
304         return BLOCKDEV_ON_ERROR_IGNORE;
305     } else if (!is_read && !strcmp(buf, "enospc")) {
306         return BLOCKDEV_ON_ERROR_ENOSPC;
307     } else if (!strcmp(buf, "stop")) {
308         return BLOCKDEV_ON_ERROR_STOP;
309     } else if (!strcmp(buf, "report")) {
310         return BLOCKDEV_ON_ERROR_REPORT;
311     } else {
312         error_setg(errp, "'%s' invalid %s error action",
313                    buf, is_read ? "read" : "write");
314         return -1;
315     }
316 }
317 
318 static bool parse_stats_intervals(BlockAcctStats *stats, QList *intervals,
319                                   Error **errp)
320 {
321     const QListEntry *entry;
322     for (entry = qlist_first(intervals); entry; entry = qlist_next(entry)) {
323         switch (qobject_type(entry->value)) {
324 
325         case QTYPE_QSTRING: {
326             unsigned long long length;
327             const char *str = qstring_get_str(qobject_to_qstring(entry->value));
328             if (parse_uint_full(str, &length, 10) == 0 &&
329                 length > 0 && length <= UINT_MAX) {
330                 block_acct_add_interval(stats, (unsigned) length);
331             } else {
332                 error_setg(errp, "Invalid interval length: %s", str);
333                 return false;
334             }
335             break;
336         }
337 
338         case QTYPE_QNUM: {
339             int64_t length = qnum_get_int(qobject_to_qnum(entry->value));
340 
341             if (length > 0 && length <= UINT_MAX) {
342                 block_acct_add_interval(stats, (unsigned) length);
343             } else {
344                 error_setg(errp, "Invalid interval length: %" PRId64, length);
345                 return false;
346             }
347             break;
348         }
349 
350         default:
351             error_setg(errp, "The specification of stats-intervals is invalid");
352             return false;
353         }
354     }
355     return true;
356 }
357 
358 typedef enum { MEDIA_DISK, MEDIA_CDROM } DriveMediaType;
359 
360 /* All parameters but @opts are optional and may be set to NULL. */
361 static void extract_common_blockdev_options(QemuOpts *opts, int *bdrv_flags,
362     const char **throttling_group, ThrottleConfig *throttle_cfg,
363     BlockdevDetectZeroesOptions *detect_zeroes, Error **errp)
364 {
365     Error *local_error = NULL;
366     const char *aio;
367 
368     if (bdrv_flags) {
369         if (qemu_opt_get_bool(opts, "copy-on-read", false)) {
370             *bdrv_flags |= BDRV_O_COPY_ON_READ;
371         }
372 
373         if ((aio = qemu_opt_get(opts, "aio")) != NULL) {
374             if (!strcmp(aio, "native")) {
375                 *bdrv_flags |= BDRV_O_NATIVE_AIO;
376             } else if (!strcmp(aio, "threads")) {
377                 /* this is the default */
378             } else {
379                error_setg(errp, "invalid aio option");
380                return;
381             }
382         }
383     }
384 
385     /* disk I/O throttling */
386     if (throttling_group) {
387         *throttling_group = qemu_opt_get(opts, "throttling.group");
388     }
389 
390     if (throttle_cfg) {
391         throttle_config_init(throttle_cfg);
392         throttle_cfg->buckets[THROTTLE_BPS_TOTAL].avg =
393             qemu_opt_get_number(opts, "throttling.bps-total", 0);
394         throttle_cfg->buckets[THROTTLE_BPS_READ].avg  =
395             qemu_opt_get_number(opts, "throttling.bps-read", 0);
396         throttle_cfg->buckets[THROTTLE_BPS_WRITE].avg =
397             qemu_opt_get_number(opts, "throttling.bps-write", 0);
398         throttle_cfg->buckets[THROTTLE_OPS_TOTAL].avg =
399             qemu_opt_get_number(opts, "throttling.iops-total", 0);
400         throttle_cfg->buckets[THROTTLE_OPS_READ].avg =
401             qemu_opt_get_number(opts, "throttling.iops-read", 0);
402         throttle_cfg->buckets[THROTTLE_OPS_WRITE].avg =
403             qemu_opt_get_number(opts, "throttling.iops-write", 0);
404 
405         throttle_cfg->buckets[THROTTLE_BPS_TOTAL].max =
406             qemu_opt_get_number(opts, "throttling.bps-total-max", 0);
407         throttle_cfg->buckets[THROTTLE_BPS_READ].max  =
408             qemu_opt_get_number(opts, "throttling.bps-read-max", 0);
409         throttle_cfg->buckets[THROTTLE_BPS_WRITE].max =
410             qemu_opt_get_number(opts, "throttling.bps-write-max", 0);
411         throttle_cfg->buckets[THROTTLE_OPS_TOTAL].max =
412             qemu_opt_get_number(opts, "throttling.iops-total-max", 0);
413         throttle_cfg->buckets[THROTTLE_OPS_READ].max =
414             qemu_opt_get_number(opts, "throttling.iops-read-max", 0);
415         throttle_cfg->buckets[THROTTLE_OPS_WRITE].max =
416             qemu_opt_get_number(opts, "throttling.iops-write-max", 0);
417 
418         throttle_cfg->buckets[THROTTLE_BPS_TOTAL].burst_length =
419             qemu_opt_get_number(opts, "throttling.bps-total-max-length", 1);
420         throttle_cfg->buckets[THROTTLE_BPS_READ].burst_length  =
421             qemu_opt_get_number(opts, "throttling.bps-read-max-length", 1);
422         throttle_cfg->buckets[THROTTLE_BPS_WRITE].burst_length =
423             qemu_opt_get_number(opts, "throttling.bps-write-max-length", 1);
424         throttle_cfg->buckets[THROTTLE_OPS_TOTAL].burst_length =
425             qemu_opt_get_number(opts, "throttling.iops-total-max-length", 1);
426         throttle_cfg->buckets[THROTTLE_OPS_READ].burst_length =
427             qemu_opt_get_number(opts, "throttling.iops-read-max-length", 1);
428         throttle_cfg->buckets[THROTTLE_OPS_WRITE].burst_length =
429             qemu_opt_get_number(opts, "throttling.iops-write-max-length", 1);
430 
431         throttle_cfg->op_size =
432             qemu_opt_get_number(opts, "throttling.iops-size", 0);
433 
434         if (!throttle_is_valid(throttle_cfg, errp)) {
435             return;
436         }
437     }
438 
439     if (detect_zeroes) {
440         *detect_zeroes =
441             qapi_enum_parse(&BlockdevDetectZeroesOptions_lookup,
442                             qemu_opt_get(opts, "detect-zeroes"),
443                             BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF,
444                             &local_error);
445         if (local_error) {
446             error_propagate(errp, local_error);
447             return;
448         }
449     }
450 }
451 
452 /* Takes the ownership of bs_opts */
453 static BlockBackend *blockdev_init(const char *file, QDict *bs_opts,
454                                    Error **errp)
455 {
456     const char *buf;
457     int bdrv_flags = 0;
458     int on_read_error, on_write_error;
459     bool account_invalid, account_failed;
460     bool writethrough, read_only;
461     BlockBackend *blk;
462     BlockDriverState *bs;
463     ThrottleConfig cfg;
464     int snapshot = 0;
465     Error *error = NULL;
466     QemuOpts *opts;
467     QDict *interval_dict = NULL;
468     QList *interval_list = NULL;
469     const char *id;
470     BlockdevDetectZeroesOptions detect_zeroes =
471         BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF;
472     const char *throttling_group = NULL;
473 
474     /* Check common options by copying from bs_opts to opts, all other options
475      * stay in bs_opts for processing by bdrv_open(). */
476     id = qdict_get_try_str(bs_opts, "id");
477     opts = qemu_opts_create(&qemu_common_drive_opts, id, 1, &error);
478     if (error) {
479         error_propagate(errp, error);
480         goto err_no_opts;
481     }
482 
483     qemu_opts_absorb_qdict(opts, bs_opts, &error);
484     if (error) {
485         error_propagate(errp, error);
486         goto early_err;
487     }
488 
489     if (id) {
490         qdict_del(bs_opts, "id");
491     }
492 
493     /* extract parameters */
494     snapshot = qemu_opt_get_bool(opts, "snapshot", 0);
495 
496     account_invalid = qemu_opt_get_bool(opts, "stats-account-invalid", true);
497     account_failed = qemu_opt_get_bool(opts, "stats-account-failed", true);
498 
499     writethrough = !qemu_opt_get_bool(opts, BDRV_OPT_CACHE_WB, true);
500 
501     id = qemu_opts_id(opts);
502 
503     qdict_extract_subqdict(bs_opts, &interval_dict, "stats-intervals.");
504     qdict_array_split(interval_dict, &interval_list);
505 
506     if (qdict_size(interval_dict) != 0) {
507         error_setg(errp, "Invalid option stats-intervals.%s",
508                    qdict_first(interval_dict)->key);
509         goto early_err;
510     }
511 
512     extract_common_blockdev_options(opts, &bdrv_flags, &throttling_group, &cfg,
513                                     &detect_zeroes, &error);
514     if (error) {
515         error_propagate(errp, error);
516         goto early_err;
517     }
518 
519     if ((buf = qemu_opt_get(opts, "format")) != NULL) {
520         if (is_help_option(buf)) {
521             error_printf("Supported formats:");
522             bdrv_iterate_format(bdrv_format_print, NULL);
523             error_printf("\n");
524             goto early_err;
525         }
526 
527         if (qdict_haskey(bs_opts, "driver")) {
528             error_setg(errp, "Cannot specify both 'driver' and 'format'");
529             goto early_err;
530         }
531         qdict_put_str(bs_opts, "driver", buf);
532     }
533 
534     on_write_error = BLOCKDEV_ON_ERROR_ENOSPC;
535     if ((buf = qemu_opt_get(opts, "werror")) != NULL) {
536         on_write_error = parse_block_error_action(buf, 0, &error);
537         if (error) {
538             error_propagate(errp, error);
539             goto early_err;
540         }
541     }
542 
543     on_read_error = BLOCKDEV_ON_ERROR_REPORT;
544     if ((buf = qemu_opt_get(opts, "rerror")) != NULL) {
545         on_read_error = parse_block_error_action(buf, 1, &error);
546         if (error) {
547             error_propagate(errp, error);
548             goto early_err;
549         }
550     }
551 
552     if (snapshot) {
553         bdrv_flags |= BDRV_O_SNAPSHOT;
554     }
555 
556     read_only = qemu_opt_get_bool(opts, BDRV_OPT_READ_ONLY, false);
557 
558     /* init */
559     if ((!file || !*file) && !qdict_size(bs_opts)) {
560         BlockBackendRootState *blk_rs;
561 
562         blk = blk_new(0, BLK_PERM_ALL);
563         blk_rs = blk_get_root_state(blk);
564         blk_rs->open_flags    = bdrv_flags;
565         blk_rs->read_only     = read_only;
566         blk_rs->detect_zeroes = detect_zeroes;
567 
568         QDECREF(bs_opts);
569     } else {
570         if (file && !*file) {
571             file = NULL;
572         }
573 
574         /* bdrv_open() defaults to the values in bdrv_flags (for compatibility
575          * with other callers) rather than what we want as the real defaults.
576          * Apply the defaults here instead. */
577         qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_DIRECT, "off");
578         qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_NO_FLUSH, "off");
579         qdict_set_default_str(bs_opts, BDRV_OPT_READ_ONLY,
580                               read_only ? "on" : "off");
581         assert((bdrv_flags & BDRV_O_CACHE_MASK) == 0);
582 
583         if (runstate_check(RUN_STATE_INMIGRATE)) {
584             bdrv_flags |= BDRV_O_INACTIVE;
585         }
586 
587         blk = blk_new_open(file, NULL, bs_opts, bdrv_flags, errp);
588         if (!blk) {
589             goto err_no_bs_opts;
590         }
591         bs = blk_bs(blk);
592 
593         bs->detect_zeroes = detect_zeroes;
594 
595         block_acct_setup(blk_get_stats(blk), account_invalid, account_failed);
596 
597         if (!parse_stats_intervals(blk_get_stats(blk), interval_list, errp)) {
598             blk_unref(blk);
599             blk = NULL;
600             goto err_no_bs_opts;
601         }
602     }
603 
604     /* disk I/O throttling */
605     if (throttle_enabled(&cfg)) {
606         if (!throttling_group) {
607             throttling_group = id;
608         }
609         blk_io_limits_enable(blk, throttling_group);
610         blk_set_io_limits(blk, &cfg);
611     }
612 
613     blk_set_enable_write_cache(blk, !writethrough);
614     blk_set_on_error(blk, on_read_error, on_write_error);
615 
616     if (!monitor_add_blk(blk, id, errp)) {
617         blk_unref(blk);
618         blk = NULL;
619         goto err_no_bs_opts;
620     }
621 
622 err_no_bs_opts:
623     qemu_opts_del(opts);
624     QDECREF(interval_dict);
625     QDECREF(interval_list);
626     return blk;
627 
628 early_err:
629     qemu_opts_del(opts);
630     QDECREF(interval_dict);
631     QDECREF(interval_list);
632 err_no_opts:
633     QDECREF(bs_opts);
634     return NULL;
635 }
636 
637 /* Takes the ownership of bs_opts */
638 static BlockDriverState *bds_tree_init(QDict *bs_opts, Error **errp)
639 {
640     int bdrv_flags = 0;
641 
642     /* bdrv_open() defaults to the values in bdrv_flags (for compatibility
643      * with other callers) rather than what we want as the real defaults.
644      * Apply the defaults here instead. */
645     qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_DIRECT, "off");
646     qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_NO_FLUSH, "off");
647     qdict_set_default_str(bs_opts, BDRV_OPT_READ_ONLY, "off");
648 
649     if (runstate_check(RUN_STATE_INMIGRATE)) {
650         bdrv_flags |= BDRV_O_INACTIVE;
651     }
652 
653     return bdrv_open(NULL, NULL, bs_opts, bdrv_flags, errp);
654 }
655 
656 void blockdev_close_all_bdrv_states(void)
657 {
658     BlockDriverState *bs, *next_bs;
659 
660     QTAILQ_FOREACH_SAFE(bs, &monitor_bdrv_states, monitor_list, next_bs) {
661         AioContext *ctx = bdrv_get_aio_context(bs);
662 
663         aio_context_acquire(ctx);
664         bdrv_unref(bs);
665         aio_context_release(ctx);
666     }
667 }
668 
669 /* Iterates over the list of monitor-owned BlockDriverStates */
670 BlockDriverState *bdrv_next_monitor_owned(BlockDriverState *bs)
671 {
672     return bs ? QTAILQ_NEXT(bs, monitor_list)
673               : QTAILQ_FIRST(&monitor_bdrv_states);
674 }
675 
676 static void qemu_opt_rename(QemuOpts *opts, const char *from, const char *to,
677                             Error **errp)
678 {
679     const char *value;
680 
681     value = qemu_opt_get(opts, from);
682     if (value) {
683         if (qemu_opt_find(opts, to)) {
684             error_setg(errp, "'%s' and its alias '%s' can't be used at the "
685                        "same time", to, from);
686             return;
687         }
688     }
689 
690     /* rename all items in opts */
691     while ((value = qemu_opt_get(opts, from))) {
692         qemu_opt_set(opts, to, value, &error_abort);
693         qemu_opt_unset(opts, from);
694     }
695 }
696 
697 QemuOptsList qemu_legacy_drive_opts = {
698     .name = "drive",
699     .head = QTAILQ_HEAD_INITIALIZER(qemu_legacy_drive_opts.head),
700     .desc = {
701         {
702             .name = "bus",
703             .type = QEMU_OPT_NUMBER,
704             .help = "bus number",
705         },{
706             .name = "unit",
707             .type = QEMU_OPT_NUMBER,
708             .help = "unit number (i.e. lun for scsi)",
709         },{
710             .name = "index",
711             .type = QEMU_OPT_NUMBER,
712             .help = "index number",
713         },{
714             .name = "media",
715             .type = QEMU_OPT_STRING,
716             .help = "media type (disk, cdrom)",
717         },{
718             .name = "if",
719             .type = QEMU_OPT_STRING,
720             .help = "interface (ide, scsi, sd, mtd, floppy, pflash, virtio)",
721         },{
722             .name = "cyls",
723             .type = QEMU_OPT_NUMBER,
724             .help = "number of cylinders (ide disk geometry)",
725         },{
726             .name = "heads",
727             .type = QEMU_OPT_NUMBER,
728             .help = "number of heads (ide disk geometry)",
729         },{
730             .name = "secs",
731             .type = QEMU_OPT_NUMBER,
732             .help = "number of sectors (ide disk geometry)",
733         },{
734             .name = "trans",
735             .type = QEMU_OPT_STRING,
736             .help = "chs translation (auto, lba, none)",
737         },{
738             .name = "addr",
739             .type = QEMU_OPT_STRING,
740             .help = "pci address (virtio only)",
741         },{
742             .name = "serial",
743             .type = QEMU_OPT_STRING,
744             .help = "disk serial number",
745         },{
746             .name = "file",
747             .type = QEMU_OPT_STRING,
748             .help = "file name",
749         },
750 
751         /* Options that are passed on, but have special semantics with -drive */
752         {
753             .name = BDRV_OPT_READ_ONLY,
754             .type = QEMU_OPT_BOOL,
755             .help = "open drive file as read-only",
756         },{
757             .name = "rerror",
758             .type = QEMU_OPT_STRING,
759             .help = "read error action",
760         },{
761             .name = "werror",
762             .type = QEMU_OPT_STRING,
763             .help = "write error action",
764         },{
765             .name = "copy-on-read",
766             .type = QEMU_OPT_BOOL,
767             .help = "copy read data from backing file into image file",
768         },
769 
770         { /* end of list */ }
771     },
772 };
773 
774 DriveInfo *drive_new(QemuOpts *all_opts, BlockInterfaceType block_default_type)
775 {
776     const char *value;
777     BlockBackend *blk;
778     DriveInfo *dinfo = NULL;
779     QDict *bs_opts;
780     QemuOpts *legacy_opts;
781     DriveMediaType media = MEDIA_DISK;
782     BlockInterfaceType type;
783     int cyls, heads, secs, translation;
784     int max_devs, bus_id, unit_id, index;
785     const char *devaddr;
786     const char *werror, *rerror;
787     bool read_only = false;
788     bool copy_on_read;
789     const char *serial;
790     const char *filename;
791     Error *local_err = NULL;
792     int i;
793     const char *deprecated[] = {
794         "serial", "trans", "secs", "heads", "cyls", "addr"
795     };
796 
797     /* Change legacy command line options into QMP ones */
798     static const struct {
799         const char *from;
800         const char *to;
801     } opt_renames[] = {
802         { "iops",           "throttling.iops-total" },
803         { "iops_rd",        "throttling.iops-read" },
804         { "iops_wr",        "throttling.iops-write" },
805 
806         { "bps",            "throttling.bps-total" },
807         { "bps_rd",         "throttling.bps-read" },
808         { "bps_wr",         "throttling.bps-write" },
809 
810         { "iops_max",       "throttling.iops-total-max" },
811         { "iops_rd_max",    "throttling.iops-read-max" },
812         { "iops_wr_max",    "throttling.iops-write-max" },
813 
814         { "bps_max",        "throttling.bps-total-max" },
815         { "bps_rd_max",     "throttling.bps-read-max" },
816         { "bps_wr_max",     "throttling.bps-write-max" },
817 
818         { "iops_size",      "throttling.iops-size" },
819 
820         { "group",          "throttling.group" },
821 
822         { "readonly",       BDRV_OPT_READ_ONLY },
823     };
824 
825     for (i = 0; i < ARRAY_SIZE(opt_renames); i++) {
826         qemu_opt_rename(all_opts, opt_renames[i].from, opt_renames[i].to,
827                         &local_err);
828         if (local_err) {
829             error_report_err(local_err);
830             return NULL;
831         }
832     }
833 
834     value = qemu_opt_get(all_opts, "cache");
835     if (value) {
836         int flags = 0;
837         bool writethrough;
838 
839         if (bdrv_parse_cache_mode(value, &flags, &writethrough) != 0) {
840             error_report("invalid cache option");
841             return NULL;
842         }
843 
844         /* Specific options take precedence */
845         if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_WB)) {
846             qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_WB,
847                               !writethrough, &error_abort);
848         }
849         if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_DIRECT)) {
850             qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_DIRECT,
851                               !!(flags & BDRV_O_NOCACHE), &error_abort);
852         }
853         if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_NO_FLUSH)) {
854             qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_NO_FLUSH,
855                               !!(flags & BDRV_O_NO_FLUSH), &error_abort);
856         }
857         qemu_opt_unset(all_opts, "cache");
858     }
859 
860     /* Get a QDict for processing the options */
861     bs_opts = qdict_new();
862     qemu_opts_to_qdict(all_opts, bs_opts);
863 
864     legacy_opts = qemu_opts_create(&qemu_legacy_drive_opts, NULL, 0,
865                                    &error_abort);
866     qemu_opts_absorb_qdict(legacy_opts, bs_opts, &local_err);
867     if (local_err) {
868         error_report_err(local_err);
869         goto fail;
870     }
871 
872     /* Other deprecated options */
873     if (!qtest_enabled()) {
874         for (i = 0; i < ARRAY_SIZE(deprecated); i++) {
875             if (qemu_opt_get(legacy_opts, deprecated[i]) != NULL) {
876                 error_report("'%s' is deprecated, please use the corresponding "
877                              "option of '-device' instead", deprecated[i]);
878             }
879         }
880     }
881 
882     /* Media type */
883     value = qemu_opt_get(legacy_opts, "media");
884     if (value) {
885         if (!strcmp(value, "disk")) {
886             media = MEDIA_DISK;
887         } else if (!strcmp(value, "cdrom")) {
888             media = MEDIA_CDROM;
889             read_only = true;
890         } else {
891             error_report("'%s' invalid media", value);
892             goto fail;
893         }
894     }
895 
896     /* copy-on-read is disabled with a warning for read-only devices */
897     read_only |= qemu_opt_get_bool(legacy_opts, BDRV_OPT_READ_ONLY, false);
898     copy_on_read = qemu_opt_get_bool(legacy_opts, "copy-on-read", false);
899 
900     if (read_only && copy_on_read) {
901         warn_report("disabling copy-on-read on read-only drive");
902         copy_on_read = false;
903     }
904 
905     qdict_put_str(bs_opts, BDRV_OPT_READ_ONLY, read_only ? "on" : "off");
906     qdict_put_str(bs_opts, "copy-on-read", copy_on_read ? "on" : "off");
907 
908     /* Controller type */
909     value = qemu_opt_get(legacy_opts, "if");
910     if (value) {
911         for (type = 0;
912              type < IF_COUNT && strcmp(value, if_name[type]);
913              type++) {
914         }
915         if (type == IF_COUNT) {
916             error_report("unsupported bus type '%s'", value);
917             goto fail;
918         }
919     } else {
920         type = block_default_type;
921     }
922 
923     /* Geometry */
924     cyls  = qemu_opt_get_number(legacy_opts, "cyls", 0);
925     heads = qemu_opt_get_number(legacy_opts, "heads", 0);
926     secs  = qemu_opt_get_number(legacy_opts, "secs", 0);
927 
928     if (cyls || heads || secs) {
929         if (cyls < 1) {
930             error_report("invalid physical cyls number");
931             goto fail;
932         }
933         if (heads < 1) {
934             error_report("invalid physical heads number");
935             goto fail;
936         }
937         if (secs < 1) {
938             error_report("invalid physical secs number");
939             goto fail;
940         }
941     }
942 
943     translation = BIOS_ATA_TRANSLATION_AUTO;
944     value = qemu_opt_get(legacy_opts, "trans");
945     if (value != NULL) {
946         if (!cyls) {
947             error_report("'%s' trans must be used with cyls, heads and secs",
948                          value);
949             goto fail;
950         }
951         if (!strcmp(value, "none")) {
952             translation = BIOS_ATA_TRANSLATION_NONE;
953         } else if (!strcmp(value, "lba")) {
954             translation = BIOS_ATA_TRANSLATION_LBA;
955         } else if (!strcmp(value, "large")) {
956             translation = BIOS_ATA_TRANSLATION_LARGE;
957         } else if (!strcmp(value, "rechs")) {
958             translation = BIOS_ATA_TRANSLATION_RECHS;
959         } else if (!strcmp(value, "auto")) {
960             translation = BIOS_ATA_TRANSLATION_AUTO;
961         } else {
962             error_report("'%s' invalid translation type", value);
963             goto fail;
964         }
965     }
966 
967     if (media == MEDIA_CDROM) {
968         if (cyls || secs || heads) {
969             error_report("CHS can't be set with media=cdrom");
970             goto fail;
971         }
972     }
973 
974     /* Device address specified by bus/unit or index.
975      * If none was specified, try to find the first free one. */
976     bus_id  = qemu_opt_get_number(legacy_opts, "bus", 0);
977     unit_id = qemu_opt_get_number(legacy_opts, "unit", -1);
978     index   = qemu_opt_get_number(legacy_opts, "index", -1);
979 
980     max_devs = if_max_devs[type];
981 
982     if (index != -1) {
983         if (bus_id != 0 || unit_id != -1) {
984             error_report("index cannot be used with bus and unit");
985             goto fail;
986         }
987         bus_id = drive_index_to_bus_id(type, index);
988         unit_id = drive_index_to_unit_id(type, index);
989     }
990 
991     if (unit_id == -1) {
992        unit_id = 0;
993        while (drive_get(type, bus_id, unit_id) != NULL) {
994            unit_id++;
995            if (max_devs && unit_id >= max_devs) {
996                unit_id -= max_devs;
997                bus_id++;
998            }
999        }
1000     }
1001 
1002     if (max_devs && unit_id >= max_devs) {
1003         error_report("unit %d too big (max is %d)", unit_id, max_devs - 1);
1004         goto fail;
1005     }
1006 
1007     if (drive_get(type, bus_id, unit_id) != NULL) {
1008         error_report("drive with bus=%d, unit=%d (index=%d) exists",
1009                      bus_id, unit_id, index);
1010         goto fail;
1011     }
1012 
1013     /* Serial number */
1014     serial = qemu_opt_get(legacy_opts, "serial");
1015 
1016     /* no id supplied -> create one */
1017     if (qemu_opts_id(all_opts) == NULL) {
1018         char *new_id;
1019         const char *mediastr = "";
1020         if (type == IF_IDE || type == IF_SCSI) {
1021             mediastr = (media == MEDIA_CDROM) ? "-cd" : "-hd";
1022         }
1023         if (max_devs) {
1024             new_id = g_strdup_printf("%s%i%s%i", if_name[type], bus_id,
1025                                      mediastr, unit_id);
1026         } else {
1027             new_id = g_strdup_printf("%s%s%i", if_name[type],
1028                                      mediastr, unit_id);
1029         }
1030         qdict_put_str(bs_opts, "id", new_id);
1031         g_free(new_id);
1032     }
1033 
1034     /* Add virtio block device */
1035     devaddr = qemu_opt_get(legacy_opts, "addr");
1036     if (devaddr && type != IF_VIRTIO) {
1037         error_report("addr is not supported by this bus type");
1038         goto fail;
1039     }
1040 
1041     if (type == IF_VIRTIO) {
1042         QemuOpts *devopts;
1043         devopts = qemu_opts_create(qemu_find_opts("device"), NULL, 0,
1044                                    &error_abort);
1045         if (arch_type == QEMU_ARCH_S390X) {
1046             qemu_opt_set(devopts, "driver", "virtio-blk-ccw", &error_abort);
1047         } else {
1048             qemu_opt_set(devopts, "driver", "virtio-blk-pci", &error_abort);
1049         }
1050         qemu_opt_set(devopts, "drive", qdict_get_str(bs_opts, "id"),
1051                      &error_abort);
1052         if (devaddr) {
1053             qemu_opt_set(devopts, "addr", devaddr, &error_abort);
1054         }
1055     }
1056 
1057     filename = qemu_opt_get(legacy_opts, "file");
1058 
1059     /* Check werror/rerror compatibility with if=... */
1060     werror = qemu_opt_get(legacy_opts, "werror");
1061     if (werror != NULL) {
1062         if (type != IF_IDE && type != IF_SCSI && type != IF_VIRTIO &&
1063             type != IF_NONE) {
1064             error_report("werror is not supported by this bus type");
1065             goto fail;
1066         }
1067         qdict_put_str(bs_opts, "werror", werror);
1068     }
1069 
1070     rerror = qemu_opt_get(legacy_opts, "rerror");
1071     if (rerror != NULL) {
1072         if (type != IF_IDE && type != IF_VIRTIO && type != IF_SCSI &&
1073             type != IF_NONE) {
1074             error_report("rerror is not supported by this bus type");
1075             goto fail;
1076         }
1077         qdict_put_str(bs_opts, "rerror", rerror);
1078     }
1079 
1080     /* Actual block device init: Functionality shared with blockdev-add */
1081     blk = blockdev_init(filename, bs_opts, &local_err);
1082     bs_opts = NULL;
1083     if (!blk) {
1084         if (local_err) {
1085             error_report_err(local_err);
1086         }
1087         goto fail;
1088     } else {
1089         assert(!local_err);
1090     }
1091 
1092     /* Create legacy DriveInfo */
1093     dinfo = g_malloc0(sizeof(*dinfo));
1094     dinfo->opts = all_opts;
1095 
1096     dinfo->cyls = cyls;
1097     dinfo->heads = heads;
1098     dinfo->secs = secs;
1099     dinfo->trans = translation;
1100 
1101     dinfo->type = type;
1102     dinfo->bus = bus_id;
1103     dinfo->unit = unit_id;
1104     dinfo->devaddr = devaddr;
1105     dinfo->serial = g_strdup(serial);
1106 
1107     blk_set_legacy_dinfo(blk, dinfo);
1108 
1109     switch(type) {
1110     case IF_IDE:
1111     case IF_SCSI:
1112     case IF_XEN:
1113     case IF_NONE:
1114         dinfo->media_cd = media == MEDIA_CDROM;
1115         break;
1116     default:
1117         break;
1118     }
1119 
1120 fail:
1121     qemu_opts_del(legacy_opts);
1122     QDECREF(bs_opts);
1123     return dinfo;
1124 }
1125 
1126 static BlockDriverState *qmp_get_root_bs(const char *name, Error **errp)
1127 {
1128     BlockDriverState *bs;
1129 
1130     bs = bdrv_lookup_bs(name, name, errp);
1131     if (bs == NULL) {
1132         return NULL;
1133     }
1134 
1135     if (!bdrv_is_root_node(bs)) {
1136         error_setg(errp, "Need a root block node");
1137         return NULL;
1138     }
1139 
1140     if (!bdrv_is_inserted(bs)) {
1141         error_setg(errp, "Device has no medium");
1142         return NULL;
1143     }
1144 
1145     return bs;
1146 }
1147 
1148 static BlockBackend *qmp_get_blk(const char *blk_name, const char *qdev_id,
1149                                  Error **errp)
1150 {
1151     BlockBackend *blk;
1152 
1153     if (!blk_name == !qdev_id) {
1154         error_setg(errp, "Need exactly one of 'device' and 'id'");
1155         return NULL;
1156     }
1157 
1158     if (qdev_id) {
1159         blk = blk_by_qdev_id(qdev_id, errp);
1160     } else {
1161         blk = blk_by_name(blk_name);
1162         if (blk == NULL) {
1163             error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
1164                       "Device '%s' not found", blk_name);
1165         }
1166     }
1167 
1168     return blk;
1169 }
1170 
1171 void hmp_commit(Monitor *mon, const QDict *qdict)
1172 {
1173     const char *device = qdict_get_str(qdict, "device");
1174     BlockBackend *blk;
1175     int ret;
1176 
1177     if (!strcmp(device, "all")) {
1178         ret = blk_commit_all();
1179     } else {
1180         BlockDriverState *bs;
1181         AioContext *aio_context;
1182 
1183         blk = blk_by_name(device);
1184         if (!blk) {
1185             monitor_printf(mon, "Device '%s' not found\n", device);
1186             return;
1187         }
1188         if (!blk_is_available(blk)) {
1189             monitor_printf(mon, "Device '%s' has no medium\n", device);
1190             return;
1191         }
1192 
1193         bs = blk_bs(blk);
1194         aio_context = bdrv_get_aio_context(bs);
1195         aio_context_acquire(aio_context);
1196 
1197         ret = bdrv_commit(bs);
1198 
1199         aio_context_release(aio_context);
1200     }
1201     if (ret < 0) {
1202         monitor_printf(mon, "'commit' error for '%s': %s\n", device,
1203                        strerror(-ret));
1204     }
1205 }
1206 
1207 static void blockdev_do_action(TransactionAction *action, Error **errp)
1208 {
1209     TransactionActionList list;
1210 
1211     list.value = action;
1212     list.next = NULL;
1213     qmp_transaction(&list, false, NULL, errp);
1214 }
1215 
1216 void qmp_blockdev_snapshot_sync(bool has_device, const char *device,
1217                                 bool has_node_name, const char *node_name,
1218                                 const char *snapshot_file,
1219                                 bool has_snapshot_node_name,
1220                                 const char *snapshot_node_name,
1221                                 bool has_format, const char *format,
1222                                 bool has_mode, NewImageMode mode, Error **errp)
1223 {
1224     BlockdevSnapshotSync snapshot = {
1225         .has_device = has_device,
1226         .device = (char *) device,
1227         .has_node_name = has_node_name,
1228         .node_name = (char *) node_name,
1229         .snapshot_file = (char *) snapshot_file,
1230         .has_snapshot_node_name = has_snapshot_node_name,
1231         .snapshot_node_name = (char *) snapshot_node_name,
1232         .has_format = has_format,
1233         .format = (char *) format,
1234         .has_mode = has_mode,
1235         .mode = mode,
1236     };
1237     TransactionAction action = {
1238         .type = TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC,
1239         .u.blockdev_snapshot_sync.data = &snapshot,
1240     };
1241     blockdev_do_action(&action, errp);
1242 }
1243 
1244 void qmp_blockdev_snapshot(const char *node, const char *overlay,
1245                            Error **errp)
1246 {
1247     BlockdevSnapshot snapshot_data = {
1248         .node = (char *) node,
1249         .overlay = (char *) overlay
1250     };
1251     TransactionAction action = {
1252         .type = TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT,
1253         .u.blockdev_snapshot.data = &snapshot_data,
1254     };
1255     blockdev_do_action(&action, errp);
1256 }
1257 
1258 void qmp_blockdev_snapshot_internal_sync(const char *device,
1259                                          const char *name,
1260                                          Error **errp)
1261 {
1262     BlockdevSnapshotInternal snapshot = {
1263         .device = (char *) device,
1264         .name = (char *) name
1265     };
1266     TransactionAction action = {
1267         .type = TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC,
1268         .u.blockdev_snapshot_internal_sync.data = &snapshot,
1269     };
1270     blockdev_do_action(&action, errp);
1271 }
1272 
1273 SnapshotInfo *qmp_blockdev_snapshot_delete_internal_sync(const char *device,
1274                                                          bool has_id,
1275                                                          const char *id,
1276                                                          bool has_name,
1277                                                          const char *name,
1278                                                          Error **errp)
1279 {
1280     BlockDriverState *bs;
1281     AioContext *aio_context;
1282     QEMUSnapshotInfo sn;
1283     Error *local_err = NULL;
1284     SnapshotInfo *info = NULL;
1285     int ret;
1286 
1287     bs = qmp_get_root_bs(device, errp);
1288     if (!bs) {
1289         return NULL;
1290     }
1291     aio_context = bdrv_get_aio_context(bs);
1292     aio_context_acquire(aio_context);
1293 
1294     if (!has_id) {
1295         id = NULL;
1296     }
1297 
1298     if (!has_name) {
1299         name = NULL;
1300     }
1301 
1302     if (!id && !name) {
1303         error_setg(errp, "Name or id must be provided");
1304         goto out_aio_context;
1305     }
1306 
1307     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_INTERNAL_SNAPSHOT_DELETE, errp)) {
1308         goto out_aio_context;
1309     }
1310 
1311     ret = bdrv_snapshot_find_by_id_and_name(bs, id, name, &sn, &local_err);
1312     if (local_err) {
1313         error_propagate(errp, local_err);
1314         goto out_aio_context;
1315     }
1316     if (!ret) {
1317         error_setg(errp,
1318                    "Snapshot with id '%s' and name '%s' does not exist on "
1319                    "device '%s'",
1320                    STR_OR_NULL(id), STR_OR_NULL(name), device);
1321         goto out_aio_context;
1322     }
1323 
1324     bdrv_snapshot_delete(bs, id, name, &local_err);
1325     if (local_err) {
1326         error_propagate(errp, local_err);
1327         goto out_aio_context;
1328     }
1329 
1330     aio_context_release(aio_context);
1331 
1332     info = g_new0(SnapshotInfo, 1);
1333     info->id = g_strdup(sn.id_str);
1334     info->name = g_strdup(sn.name);
1335     info->date_nsec = sn.date_nsec;
1336     info->date_sec = sn.date_sec;
1337     info->vm_state_size = sn.vm_state_size;
1338     info->vm_clock_nsec = sn.vm_clock_nsec % 1000000000;
1339     info->vm_clock_sec = sn.vm_clock_nsec / 1000000000;
1340 
1341     return info;
1342 
1343 out_aio_context:
1344     aio_context_release(aio_context);
1345     return NULL;
1346 }
1347 
1348 /**
1349  * block_dirty_bitmap_lookup:
1350  * Return a dirty bitmap (if present), after validating
1351  * the node reference and bitmap names.
1352  *
1353  * @node: The name of the BDS node to search for bitmaps
1354  * @name: The name of the bitmap to search for
1355  * @pbs: Output pointer for BDS lookup, if desired. Can be NULL.
1356  * @paio: Output pointer for aio_context acquisition, if desired. Can be NULL.
1357  * @errp: Output pointer for error information. Can be NULL.
1358  *
1359  * @return: A bitmap object on success, or NULL on failure.
1360  */
1361 static BdrvDirtyBitmap *block_dirty_bitmap_lookup(const char *node,
1362                                                   const char *name,
1363                                                   BlockDriverState **pbs,
1364                                                   Error **errp)
1365 {
1366     BlockDriverState *bs;
1367     BdrvDirtyBitmap *bitmap;
1368 
1369     if (!node) {
1370         error_setg(errp, "Node cannot be NULL");
1371         return NULL;
1372     }
1373     if (!name) {
1374         error_setg(errp, "Bitmap name cannot be NULL");
1375         return NULL;
1376     }
1377     bs = bdrv_lookup_bs(node, node, NULL);
1378     if (!bs) {
1379         error_setg(errp, "Node '%s' not found", node);
1380         return NULL;
1381     }
1382 
1383     bitmap = bdrv_find_dirty_bitmap(bs, name);
1384     if (!bitmap) {
1385         error_setg(errp, "Dirty bitmap '%s' not found", name);
1386         return NULL;
1387     }
1388 
1389     if (pbs) {
1390         *pbs = bs;
1391     }
1392 
1393     return bitmap;
1394 }
1395 
1396 /* New and old BlockDriverState structs for atomic group operations */
1397 
1398 typedef struct BlkActionState BlkActionState;
1399 
1400 /**
1401  * BlkActionOps:
1402  * Table of operations that define an Action.
1403  *
1404  * @instance_size: Size of state struct, in bytes.
1405  * @prepare: Prepare the work, must NOT be NULL.
1406  * @commit: Commit the changes, can be NULL.
1407  * @abort: Abort the changes on fail, can be NULL.
1408  * @clean: Clean up resources after all transaction actions have called
1409  *         commit() or abort(). Can be NULL.
1410  *
1411  * Only prepare() may fail. In a single transaction, only one of commit() or
1412  * abort() will be called. clean() will always be called if it is present.
1413  */
1414 typedef struct BlkActionOps {
1415     size_t instance_size;
1416     void (*prepare)(BlkActionState *common, Error **errp);
1417     void (*commit)(BlkActionState *common);
1418     void (*abort)(BlkActionState *common);
1419     void (*clean)(BlkActionState *common);
1420 } BlkActionOps;
1421 
1422 /**
1423  * BlkActionState:
1424  * Describes one Action's state within a Transaction.
1425  *
1426  * @action: QAPI-defined enum identifying which Action to perform.
1427  * @ops: Table of ActionOps this Action can perform.
1428  * @block_job_txn: Transaction which this action belongs to.
1429  * @entry: List membership for all Actions in this Transaction.
1430  *
1431  * This structure must be arranged as first member in a subclassed type,
1432  * assuming that the compiler will also arrange it to the same offsets as the
1433  * base class.
1434  */
1435 struct BlkActionState {
1436     TransactionAction *action;
1437     const BlkActionOps *ops;
1438     BlockJobTxn *block_job_txn;
1439     TransactionProperties *txn_props;
1440     QSIMPLEQ_ENTRY(BlkActionState) entry;
1441 };
1442 
1443 /* internal snapshot private data */
1444 typedef struct InternalSnapshotState {
1445     BlkActionState common;
1446     BlockDriverState *bs;
1447     QEMUSnapshotInfo sn;
1448     bool created;
1449 } InternalSnapshotState;
1450 
1451 
1452 static int action_check_completion_mode(BlkActionState *s, Error **errp)
1453 {
1454     if (s->txn_props->completion_mode != ACTION_COMPLETION_MODE_INDIVIDUAL) {
1455         error_setg(errp,
1456                    "Action '%s' does not support Transaction property "
1457                    "completion-mode = %s",
1458                    TransactionActionKind_str(s->action->type),
1459                    ActionCompletionMode_str(s->txn_props->completion_mode));
1460         return -1;
1461     }
1462     return 0;
1463 }
1464 
1465 static void internal_snapshot_prepare(BlkActionState *common,
1466                                       Error **errp)
1467 {
1468     Error *local_err = NULL;
1469     const char *device;
1470     const char *name;
1471     BlockDriverState *bs;
1472     QEMUSnapshotInfo old_sn, *sn;
1473     bool ret;
1474     qemu_timeval tv;
1475     BlockdevSnapshotInternal *internal;
1476     InternalSnapshotState *state;
1477     AioContext *aio_context;
1478     int ret1;
1479 
1480     g_assert(common->action->type ==
1481              TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC);
1482     internal = common->action->u.blockdev_snapshot_internal_sync.data;
1483     state = DO_UPCAST(InternalSnapshotState, common, common);
1484 
1485     /* 1. parse input */
1486     device = internal->device;
1487     name = internal->name;
1488 
1489     /* 2. check for validation */
1490     if (action_check_completion_mode(common, errp) < 0) {
1491         return;
1492     }
1493 
1494     bs = qmp_get_root_bs(device, errp);
1495     if (!bs) {
1496         return;
1497     }
1498 
1499     aio_context = bdrv_get_aio_context(bs);
1500     aio_context_acquire(aio_context);
1501 
1502     state->bs = bs;
1503 
1504     /* Paired with .clean() */
1505     bdrv_drained_begin(bs);
1506 
1507     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_INTERNAL_SNAPSHOT, errp)) {
1508         goto out;
1509     }
1510 
1511     if (bdrv_is_read_only(bs)) {
1512         error_setg(errp, "Device '%s' is read only", device);
1513         goto out;
1514     }
1515 
1516     if (!bdrv_can_snapshot(bs)) {
1517         error_setg(errp, "Block format '%s' used by device '%s' "
1518                    "does not support internal snapshots",
1519                    bs->drv->format_name, device);
1520         goto out;
1521     }
1522 
1523     if (!strlen(name)) {
1524         error_setg(errp, "Name is empty");
1525         goto out;
1526     }
1527 
1528     /* check whether a snapshot with name exist */
1529     ret = bdrv_snapshot_find_by_id_and_name(bs, NULL, name, &old_sn,
1530                                             &local_err);
1531     if (local_err) {
1532         error_propagate(errp, local_err);
1533         goto out;
1534     } else if (ret) {
1535         error_setg(errp,
1536                    "Snapshot with name '%s' already exists on device '%s'",
1537                    name, device);
1538         goto out;
1539     }
1540 
1541     /* 3. take the snapshot */
1542     sn = &state->sn;
1543     pstrcpy(sn->name, sizeof(sn->name), name);
1544     qemu_gettimeofday(&tv);
1545     sn->date_sec = tv.tv_sec;
1546     sn->date_nsec = tv.tv_usec * 1000;
1547     sn->vm_clock_nsec = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL);
1548 
1549     ret1 = bdrv_snapshot_create(bs, sn);
1550     if (ret1 < 0) {
1551         error_setg_errno(errp, -ret1,
1552                          "Failed to create snapshot '%s' on device '%s'",
1553                          name, device);
1554         goto out;
1555     }
1556 
1557     /* 4. succeed, mark a snapshot is created */
1558     state->created = true;
1559 
1560 out:
1561     aio_context_release(aio_context);
1562 }
1563 
1564 static void internal_snapshot_abort(BlkActionState *common)
1565 {
1566     InternalSnapshotState *state =
1567                              DO_UPCAST(InternalSnapshotState, common, common);
1568     BlockDriverState *bs = state->bs;
1569     QEMUSnapshotInfo *sn = &state->sn;
1570     AioContext *aio_context;
1571     Error *local_error = NULL;
1572 
1573     if (!state->created) {
1574         return;
1575     }
1576 
1577     aio_context = bdrv_get_aio_context(state->bs);
1578     aio_context_acquire(aio_context);
1579 
1580     if (bdrv_snapshot_delete(bs, sn->id_str, sn->name, &local_error) < 0) {
1581         error_reportf_err(local_error,
1582                           "Failed to delete snapshot with id '%s' and "
1583                           "name '%s' on device '%s' in abort: ",
1584                           sn->id_str, sn->name,
1585                           bdrv_get_device_name(bs));
1586     }
1587 
1588     aio_context_release(aio_context);
1589 }
1590 
1591 static void internal_snapshot_clean(BlkActionState *common)
1592 {
1593     InternalSnapshotState *state = DO_UPCAST(InternalSnapshotState,
1594                                              common, common);
1595     AioContext *aio_context;
1596 
1597     if (!state->bs) {
1598         return;
1599     }
1600 
1601     aio_context = bdrv_get_aio_context(state->bs);
1602     aio_context_acquire(aio_context);
1603 
1604     bdrv_drained_end(state->bs);
1605 
1606     aio_context_release(aio_context);
1607 }
1608 
1609 /* external snapshot private data */
1610 typedef struct ExternalSnapshotState {
1611     BlkActionState common;
1612     BlockDriverState *old_bs;
1613     BlockDriverState *new_bs;
1614     bool overlay_appended;
1615 } ExternalSnapshotState;
1616 
1617 static void external_snapshot_prepare(BlkActionState *common,
1618                                       Error **errp)
1619 {
1620     int flags = 0;
1621     QDict *options = NULL;
1622     Error *local_err = NULL;
1623     /* Device and node name of the image to generate the snapshot from */
1624     const char *device;
1625     const char *node_name;
1626     /* Reference to the new image (for 'blockdev-snapshot') */
1627     const char *snapshot_ref;
1628     /* File name of the new image (for 'blockdev-snapshot-sync') */
1629     const char *new_image_file;
1630     ExternalSnapshotState *state =
1631                              DO_UPCAST(ExternalSnapshotState, common, common);
1632     TransactionAction *action = common->action;
1633     AioContext *aio_context;
1634 
1635     /* 'blockdev-snapshot' and 'blockdev-snapshot-sync' have similar
1636      * purpose but a different set of parameters */
1637     switch (action->type) {
1638     case TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT:
1639         {
1640             BlockdevSnapshot *s = action->u.blockdev_snapshot.data;
1641             device = s->node;
1642             node_name = s->node;
1643             new_image_file = NULL;
1644             snapshot_ref = s->overlay;
1645         }
1646         break;
1647     case TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC:
1648         {
1649             BlockdevSnapshotSync *s = action->u.blockdev_snapshot_sync.data;
1650             device = s->has_device ? s->device : NULL;
1651             node_name = s->has_node_name ? s->node_name : NULL;
1652             new_image_file = s->snapshot_file;
1653             snapshot_ref = NULL;
1654         }
1655         break;
1656     default:
1657         g_assert_not_reached();
1658     }
1659 
1660     /* start processing */
1661     if (action_check_completion_mode(common, errp) < 0) {
1662         return;
1663     }
1664 
1665     state->old_bs = bdrv_lookup_bs(device, node_name, errp);
1666     if (!state->old_bs) {
1667         return;
1668     }
1669 
1670     aio_context = bdrv_get_aio_context(state->old_bs);
1671     aio_context_acquire(aio_context);
1672 
1673     /* Paired with .clean() */
1674     bdrv_drained_begin(state->old_bs);
1675 
1676     if (!bdrv_is_inserted(state->old_bs)) {
1677         error_setg(errp, QERR_DEVICE_HAS_NO_MEDIUM, device);
1678         goto out;
1679     }
1680 
1681     if (bdrv_op_is_blocked(state->old_bs,
1682                            BLOCK_OP_TYPE_EXTERNAL_SNAPSHOT, errp)) {
1683         goto out;
1684     }
1685 
1686     if (!bdrv_is_read_only(state->old_bs)) {
1687         if (bdrv_flush(state->old_bs)) {
1688             error_setg(errp, QERR_IO_ERROR);
1689             goto out;
1690         }
1691     }
1692 
1693     if (!bdrv_is_first_non_filter(state->old_bs)) {
1694         error_setg(errp, QERR_FEATURE_DISABLED, "snapshot");
1695         goto out;
1696     }
1697 
1698     if (action->type == TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC) {
1699         BlockdevSnapshotSync *s = action->u.blockdev_snapshot_sync.data;
1700         const char *format = s->has_format ? s->format : "qcow2";
1701         enum NewImageMode mode;
1702         const char *snapshot_node_name =
1703             s->has_snapshot_node_name ? s->snapshot_node_name : NULL;
1704 
1705         if (node_name && !snapshot_node_name) {
1706             error_setg(errp, "New snapshot node name missing");
1707             goto out;
1708         }
1709 
1710         if (snapshot_node_name &&
1711             bdrv_lookup_bs(snapshot_node_name, snapshot_node_name, NULL)) {
1712             error_setg(errp, "New snapshot node name already in use");
1713             goto out;
1714         }
1715 
1716         flags = state->old_bs->open_flags;
1717         flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_COPY_ON_READ);
1718         flags |= BDRV_O_NO_BACKING;
1719 
1720         /* create new image w/backing file */
1721         mode = s->has_mode ? s->mode : NEW_IMAGE_MODE_ABSOLUTE_PATHS;
1722         if (mode != NEW_IMAGE_MODE_EXISTING) {
1723             int64_t size = bdrv_getlength(state->old_bs);
1724             if (size < 0) {
1725                 error_setg_errno(errp, -size, "bdrv_getlength failed");
1726                 goto out;
1727             }
1728             bdrv_img_create(new_image_file, format,
1729                             state->old_bs->filename,
1730                             state->old_bs->drv->format_name,
1731                             NULL, size, flags, false, &local_err);
1732             if (local_err) {
1733                 error_propagate(errp, local_err);
1734                 goto out;
1735             }
1736         }
1737 
1738         options = qdict_new();
1739         if (s->has_snapshot_node_name) {
1740             qdict_put_str(options, "node-name", snapshot_node_name);
1741         }
1742         qdict_put_str(options, "driver", format);
1743     }
1744 
1745     state->new_bs = bdrv_open(new_image_file, snapshot_ref, options, flags,
1746                               errp);
1747     /* We will manually add the backing_hd field to the bs later */
1748     if (!state->new_bs) {
1749         goto out;
1750     }
1751 
1752     if (bdrv_has_blk(state->new_bs)) {
1753         error_setg(errp, "The snapshot is already in use");
1754         goto out;
1755     }
1756 
1757     if (bdrv_op_is_blocked(state->new_bs, BLOCK_OP_TYPE_EXTERNAL_SNAPSHOT,
1758                            errp)) {
1759         goto out;
1760     }
1761 
1762     if (state->new_bs->backing != NULL) {
1763         error_setg(errp, "The snapshot already has a backing image");
1764         goto out;
1765     }
1766 
1767     if (!state->new_bs->drv->supports_backing) {
1768         error_setg(errp, "The snapshot does not support backing images");
1769         goto out;
1770     }
1771 
1772     bdrv_set_aio_context(state->new_bs, aio_context);
1773 
1774     /* This removes our old bs and adds the new bs. This is an operation that
1775      * can fail, so we need to do it in .prepare; undoing it for abort is
1776      * always possible. */
1777     bdrv_ref(state->new_bs);
1778     bdrv_append(state->new_bs, state->old_bs, &local_err);
1779     if (local_err) {
1780         error_propagate(errp, local_err);
1781         goto out;
1782     }
1783     state->overlay_appended = true;
1784 
1785 out:
1786     aio_context_release(aio_context);
1787 }
1788 
1789 static void external_snapshot_commit(BlkActionState *common)
1790 {
1791     ExternalSnapshotState *state =
1792                              DO_UPCAST(ExternalSnapshotState, common, common);
1793     AioContext *aio_context;
1794 
1795     aio_context = bdrv_get_aio_context(state->old_bs);
1796     aio_context_acquire(aio_context);
1797 
1798     /* We don't need (or want) to use the transactional
1799      * bdrv_reopen_multiple() across all the entries at once, because we
1800      * don't want to abort all of them if one of them fails the reopen */
1801     if (!atomic_read(&state->old_bs->copy_on_read)) {
1802         bdrv_reopen(state->old_bs, state->old_bs->open_flags & ~BDRV_O_RDWR,
1803                     NULL);
1804     }
1805 
1806     aio_context_release(aio_context);
1807 }
1808 
1809 static void external_snapshot_abort(BlkActionState *common)
1810 {
1811     ExternalSnapshotState *state =
1812                              DO_UPCAST(ExternalSnapshotState, common, common);
1813     if (state->new_bs) {
1814         if (state->overlay_appended) {
1815             AioContext *aio_context;
1816 
1817             aio_context = bdrv_get_aio_context(state->old_bs);
1818             aio_context_acquire(aio_context);
1819 
1820             bdrv_ref(state->old_bs);   /* we can't let bdrv_set_backind_hd()
1821                                           close state->old_bs; we need it */
1822             bdrv_set_backing_hd(state->new_bs, NULL, &error_abort);
1823             bdrv_replace_node(state->new_bs, state->old_bs, &error_abort);
1824             bdrv_unref(state->old_bs); /* bdrv_replace_node() ref'ed old_bs */
1825 
1826             aio_context_release(aio_context);
1827         }
1828     }
1829 }
1830 
1831 static void external_snapshot_clean(BlkActionState *common)
1832 {
1833     ExternalSnapshotState *state =
1834                              DO_UPCAST(ExternalSnapshotState, common, common);
1835     AioContext *aio_context;
1836 
1837     if (!state->old_bs) {
1838         return;
1839     }
1840 
1841     aio_context = bdrv_get_aio_context(state->old_bs);
1842     aio_context_acquire(aio_context);
1843 
1844     bdrv_drained_end(state->old_bs);
1845     bdrv_unref(state->new_bs);
1846 
1847     aio_context_release(aio_context);
1848 }
1849 
1850 typedef struct DriveBackupState {
1851     BlkActionState common;
1852     BlockDriverState *bs;
1853     BlockJob *job;
1854 } DriveBackupState;
1855 
1856 static BlockJob *do_drive_backup(DriveBackup *backup, BlockJobTxn *txn,
1857                             Error **errp);
1858 
1859 static void drive_backup_prepare(BlkActionState *common, Error **errp)
1860 {
1861     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1862     BlockDriverState *bs;
1863     DriveBackup *backup;
1864     AioContext *aio_context;
1865     Error *local_err = NULL;
1866 
1867     assert(common->action->type == TRANSACTION_ACTION_KIND_DRIVE_BACKUP);
1868     backup = common->action->u.drive_backup.data;
1869 
1870     bs = qmp_get_root_bs(backup->device, errp);
1871     if (!bs) {
1872         return;
1873     }
1874 
1875     aio_context = bdrv_get_aio_context(bs);
1876     aio_context_acquire(aio_context);
1877 
1878     /* Paired with .clean() */
1879     bdrv_drained_begin(bs);
1880 
1881     state->bs = bs;
1882 
1883     state->job = do_drive_backup(backup, common->block_job_txn, &local_err);
1884     if (local_err) {
1885         error_propagate(errp, local_err);
1886         goto out;
1887     }
1888 
1889 out:
1890     aio_context_release(aio_context);
1891 }
1892 
1893 static void drive_backup_commit(BlkActionState *common)
1894 {
1895     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1896     AioContext *aio_context;
1897 
1898     aio_context = bdrv_get_aio_context(state->bs);
1899     aio_context_acquire(aio_context);
1900 
1901     assert(state->job);
1902     block_job_start(state->job);
1903 
1904     aio_context_release(aio_context);
1905 }
1906 
1907 static void drive_backup_abort(BlkActionState *common)
1908 {
1909     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1910 
1911     if (state->job) {
1912         AioContext *aio_context;
1913 
1914         aio_context = bdrv_get_aio_context(state->bs);
1915         aio_context_acquire(aio_context);
1916 
1917         block_job_cancel_sync(state->job);
1918 
1919         aio_context_release(aio_context);
1920     }
1921 }
1922 
1923 static void drive_backup_clean(BlkActionState *common)
1924 {
1925     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1926     AioContext *aio_context;
1927 
1928     if (!state->bs) {
1929         return;
1930     }
1931 
1932     aio_context = bdrv_get_aio_context(state->bs);
1933     aio_context_acquire(aio_context);
1934 
1935     bdrv_drained_end(state->bs);
1936 
1937     aio_context_release(aio_context);
1938 }
1939 
1940 typedef struct BlockdevBackupState {
1941     BlkActionState common;
1942     BlockDriverState *bs;
1943     BlockJob *job;
1944 } BlockdevBackupState;
1945 
1946 static BlockJob *do_blockdev_backup(BlockdevBackup *backup, BlockJobTxn *txn,
1947                                     Error **errp);
1948 
1949 static void blockdev_backup_prepare(BlkActionState *common, Error **errp)
1950 {
1951     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1952     BlockdevBackup *backup;
1953     BlockDriverState *bs, *target;
1954     AioContext *aio_context;
1955     Error *local_err = NULL;
1956 
1957     assert(common->action->type == TRANSACTION_ACTION_KIND_BLOCKDEV_BACKUP);
1958     backup = common->action->u.blockdev_backup.data;
1959 
1960     bs = qmp_get_root_bs(backup->device, errp);
1961     if (!bs) {
1962         return;
1963     }
1964 
1965     target = bdrv_lookup_bs(backup->target, backup->target, errp);
1966     if (!target) {
1967         return;
1968     }
1969 
1970     aio_context = bdrv_get_aio_context(bs);
1971     if (aio_context != bdrv_get_aio_context(target)) {
1972         error_setg(errp, "Backup between two IO threads is not implemented");
1973         return;
1974     }
1975     aio_context_acquire(aio_context);
1976     state->bs = bs;
1977 
1978     /* Paired with .clean() */
1979     bdrv_drained_begin(state->bs);
1980 
1981     state->job = do_blockdev_backup(backup, common->block_job_txn, &local_err);
1982     if (local_err) {
1983         error_propagate(errp, local_err);
1984         goto out;
1985     }
1986 
1987 out:
1988     aio_context_release(aio_context);
1989 }
1990 
1991 static void blockdev_backup_commit(BlkActionState *common)
1992 {
1993     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1994     AioContext *aio_context;
1995 
1996     aio_context = bdrv_get_aio_context(state->bs);
1997     aio_context_acquire(aio_context);
1998 
1999     assert(state->job);
2000     block_job_start(state->job);
2001 
2002     aio_context_release(aio_context);
2003 }
2004 
2005 static void blockdev_backup_abort(BlkActionState *common)
2006 {
2007     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
2008 
2009     if (state->job) {
2010         AioContext *aio_context;
2011 
2012         aio_context = bdrv_get_aio_context(state->bs);
2013         aio_context_acquire(aio_context);
2014 
2015         block_job_cancel_sync(state->job);
2016 
2017         aio_context_release(aio_context);
2018     }
2019 }
2020 
2021 static void blockdev_backup_clean(BlkActionState *common)
2022 {
2023     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
2024     AioContext *aio_context;
2025 
2026     if (!state->bs) {
2027         return;
2028     }
2029 
2030     aio_context = bdrv_get_aio_context(state->bs);
2031     aio_context_acquire(aio_context);
2032 
2033     bdrv_drained_end(state->bs);
2034 
2035     aio_context_release(aio_context);
2036 }
2037 
2038 typedef struct BlockDirtyBitmapState {
2039     BlkActionState common;
2040     BdrvDirtyBitmap *bitmap;
2041     BlockDriverState *bs;
2042     HBitmap *backup;
2043     bool prepared;
2044 } BlockDirtyBitmapState;
2045 
2046 static void block_dirty_bitmap_add_prepare(BlkActionState *common,
2047                                            Error **errp)
2048 {
2049     Error *local_err = NULL;
2050     BlockDirtyBitmapAdd *action;
2051     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2052                                              common, common);
2053 
2054     if (action_check_completion_mode(common, errp) < 0) {
2055         return;
2056     }
2057 
2058     action = common->action->u.block_dirty_bitmap_add.data;
2059     /* AIO context taken and released within qmp_block_dirty_bitmap_add */
2060     qmp_block_dirty_bitmap_add(action->node, action->name,
2061                                action->has_granularity, action->granularity,
2062                                action->has_persistent, action->persistent,
2063                                action->has_autoload, action->autoload,
2064                                &local_err);
2065 
2066     if (!local_err) {
2067         state->prepared = true;
2068     } else {
2069         error_propagate(errp, local_err);
2070     }
2071 }
2072 
2073 static void block_dirty_bitmap_add_abort(BlkActionState *common)
2074 {
2075     BlockDirtyBitmapAdd *action;
2076     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2077                                              common, common);
2078 
2079     action = common->action->u.block_dirty_bitmap_add.data;
2080     /* Should not be able to fail: IF the bitmap was added via .prepare(),
2081      * then the node reference and bitmap name must have been valid.
2082      */
2083     if (state->prepared) {
2084         qmp_block_dirty_bitmap_remove(action->node, action->name, &error_abort);
2085     }
2086 }
2087 
2088 static void block_dirty_bitmap_clear_prepare(BlkActionState *common,
2089                                              Error **errp)
2090 {
2091     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2092                                              common, common);
2093     BlockDirtyBitmap *action;
2094 
2095     if (action_check_completion_mode(common, errp) < 0) {
2096         return;
2097     }
2098 
2099     action = common->action->u.block_dirty_bitmap_clear.data;
2100     state->bitmap = block_dirty_bitmap_lookup(action->node,
2101                                               action->name,
2102                                               &state->bs,
2103                                               errp);
2104     if (!state->bitmap) {
2105         return;
2106     }
2107 
2108     if (bdrv_dirty_bitmap_frozen(state->bitmap)) {
2109         error_setg(errp, "Cannot modify a frozen bitmap");
2110         return;
2111     } else if (!bdrv_dirty_bitmap_enabled(state->bitmap)) {
2112         error_setg(errp, "Cannot clear a disabled bitmap");
2113         return;
2114     } else if (bdrv_dirty_bitmap_readonly(state->bitmap)) {
2115         error_setg(errp, "Cannot clear a readonly bitmap");
2116         return;
2117     }
2118 
2119     bdrv_clear_dirty_bitmap(state->bitmap, &state->backup);
2120 }
2121 
2122 static void block_dirty_bitmap_clear_abort(BlkActionState *common)
2123 {
2124     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2125                                              common, common);
2126 
2127     if (state->backup) {
2128         bdrv_undo_clear_dirty_bitmap(state->bitmap, state->backup);
2129     }
2130 }
2131 
2132 static void block_dirty_bitmap_clear_commit(BlkActionState *common)
2133 {
2134     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2135                                              common, common);
2136 
2137     hbitmap_free(state->backup);
2138 }
2139 
2140 static void abort_prepare(BlkActionState *common, Error **errp)
2141 {
2142     error_setg(errp, "Transaction aborted using Abort action");
2143 }
2144 
2145 static void abort_commit(BlkActionState *common)
2146 {
2147     g_assert_not_reached(); /* this action never succeeds */
2148 }
2149 
2150 static const BlkActionOps actions[] = {
2151     [TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT] = {
2152         .instance_size = sizeof(ExternalSnapshotState),
2153         .prepare  = external_snapshot_prepare,
2154         .commit   = external_snapshot_commit,
2155         .abort = external_snapshot_abort,
2156         .clean = external_snapshot_clean,
2157     },
2158     [TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC] = {
2159         .instance_size = sizeof(ExternalSnapshotState),
2160         .prepare  = external_snapshot_prepare,
2161         .commit   = external_snapshot_commit,
2162         .abort = external_snapshot_abort,
2163         .clean = external_snapshot_clean,
2164     },
2165     [TRANSACTION_ACTION_KIND_DRIVE_BACKUP] = {
2166         .instance_size = sizeof(DriveBackupState),
2167         .prepare = drive_backup_prepare,
2168         .commit = drive_backup_commit,
2169         .abort = drive_backup_abort,
2170         .clean = drive_backup_clean,
2171     },
2172     [TRANSACTION_ACTION_KIND_BLOCKDEV_BACKUP] = {
2173         .instance_size = sizeof(BlockdevBackupState),
2174         .prepare = blockdev_backup_prepare,
2175         .commit = blockdev_backup_commit,
2176         .abort = blockdev_backup_abort,
2177         .clean = blockdev_backup_clean,
2178     },
2179     [TRANSACTION_ACTION_KIND_ABORT] = {
2180         .instance_size = sizeof(BlkActionState),
2181         .prepare = abort_prepare,
2182         .commit = abort_commit,
2183     },
2184     [TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC] = {
2185         .instance_size = sizeof(InternalSnapshotState),
2186         .prepare  = internal_snapshot_prepare,
2187         .abort = internal_snapshot_abort,
2188         .clean = internal_snapshot_clean,
2189     },
2190     [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_ADD] = {
2191         .instance_size = sizeof(BlockDirtyBitmapState),
2192         .prepare = block_dirty_bitmap_add_prepare,
2193         .abort = block_dirty_bitmap_add_abort,
2194     },
2195     [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_CLEAR] = {
2196         .instance_size = sizeof(BlockDirtyBitmapState),
2197         .prepare = block_dirty_bitmap_clear_prepare,
2198         .commit = block_dirty_bitmap_clear_commit,
2199         .abort = block_dirty_bitmap_clear_abort,
2200     }
2201 };
2202 
2203 /**
2204  * Allocate a TransactionProperties structure if necessary, and fill
2205  * that structure with desired defaults if they are unset.
2206  */
2207 static TransactionProperties *get_transaction_properties(
2208     TransactionProperties *props)
2209 {
2210     if (!props) {
2211         props = g_new0(TransactionProperties, 1);
2212     }
2213 
2214     if (!props->has_completion_mode) {
2215         props->has_completion_mode = true;
2216         props->completion_mode = ACTION_COMPLETION_MODE_INDIVIDUAL;
2217     }
2218 
2219     return props;
2220 }
2221 
2222 /*
2223  * 'Atomic' group operations.  The operations are performed as a set, and if
2224  * any fail then we roll back all operations in the group.
2225  */
2226 void qmp_transaction(TransactionActionList *dev_list,
2227                      bool has_props,
2228                      struct TransactionProperties *props,
2229                      Error **errp)
2230 {
2231     TransactionActionList *dev_entry = dev_list;
2232     BlockJobTxn *block_job_txn = NULL;
2233     BlkActionState *state, *next;
2234     Error *local_err = NULL;
2235 
2236     QSIMPLEQ_HEAD(snap_bdrv_states, BlkActionState) snap_bdrv_states;
2237     QSIMPLEQ_INIT(&snap_bdrv_states);
2238 
2239     /* Does this transaction get canceled as a group on failure?
2240      * If not, we don't really need to make a BlockJobTxn.
2241      */
2242     props = get_transaction_properties(props);
2243     if (props->completion_mode != ACTION_COMPLETION_MODE_INDIVIDUAL) {
2244         block_job_txn = block_job_txn_new();
2245     }
2246 
2247     /* drain all i/o before any operations */
2248     bdrv_drain_all();
2249 
2250     /* We don't do anything in this loop that commits us to the operations */
2251     while (NULL != dev_entry) {
2252         TransactionAction *dev_info = NULL;
2253         const BlkActionOps *ops;
2254 
2255         dev_info = dev_entry->value;
2256         dev_entry = dev_entry->next;
2257 
2258         assert(dev_info->type < ARRAY_SIZE(actions));
2259 
2260         ops = &actions[dev_info->type];
2261         assert(ops->instance_size > 0);
2262 
2263         state = g_malloc0(ops->instance_size);
2264         state->ops = ops;
2265         state->action = dev_info;
2266         state->block_job_txn = block_job_txn;
2267         state->txn_props = props;
2268         QSIMPLEQ_INSERT_TAIL(&snap_bdrv_states, state, entry);
2269 
2270         state->ops->prepare(state, &local_err);
2271         if (local_err) {
2272             error_propagate(errp, local_err);
2273             goto delete_and_fail;
2274         }
2275     }
2276 
2277     QSIMPLEQ_FOREACH(state, &snap_bdrv_states, entry) {
2278         if (state->ops->commit) {
2279             state->ops->commit(state);
2280         }
2281     }
2282 
2283     /* success */
2284     goto exit;
2285 
2286 delete_and_fail:
2287     /* failure, and it is all-or-none; roll back all operations */
2288     QSIMPLEQ_FOREACH(state, &snap_bdrv_states, entry) {
2289         if (state->ops->abort) {
2290             state->ops->abort(state);
2291         }
2292     }
2293 exit:
2294     QSIMPLEQ_FOREACH_SAFE(state, &snap_bdrv_states, entry, next) {
2295         if (state->ops->clean) {
2296             state->ops->clean(state);
2297         }
2298         g_free(state);
2299     }
2300     if (!has_props) {
2301         qapi_free_TransactionProperties(props);
2302     }
2303     block_job_txn_unref(block_job_txn);
2304 }
2305 
2306 void qmp_eject(bool has_device, const char *device,
2307                bool has_id, const char *id,
2308                bool has_force, bool force, Error **errp)
2309 {
2310     Error *local_err = NULL;
2311     int rc;
2312 
2313     if (!has_force) {
2314         force = false;
2315     }
2316 
2317     rc = do_open_tray(has_device ? device : NULL,
2318                       has_id ? id : NULL,
2319                       force, &local_err);
2320     if (rc && rc != -ENOSYS) {
2321         error_propagate(errp, local_err);
2322         return;
2323     }
2324     error_free(local_err);
2325 
2326     qmp_x_blockdev_remove_medium(has_device, device, has_id, id, errp);
2327 }
2328 
2329 void qmp_block_passwd(bool has_device, const char *device,
2330                       bool has_node_name, const char *node_name,
2331                       const char *password, Error **errp)
2332 {
2333     error_setg(errp,
2334                "Setting block passwords directly is no longer supported");
2335 }
2336 
2337 /*
2338  * Attempt to open the tray of @device.
2339  * If @force, ignore its tray lock.
2340  * Else, if the tray is locked, don't open it, but ask the guest to open it.
2341  * On error, store an error through @errp and return -errno.
2342  * If @device does not exist, return -ENODEV.
2343  * If it has no removable media, return -ENOTSUP.
2344  * If it has no tray, return -ENOSYS.
2345  * If the guest was asked to open the tray, return -EINPROGRESS.
2346  * Else, return 0.
2347  */
2348 static int do_open_tray(const char *blk_name, const char *qdev_id,
2349                         bool force, Error **errp)
2350 {
2351     BlockBackend *blk;
2352     const char *device = qdev_id ?: blk_name;
2353     bool locked;
2354 
2355     blk = qmp_get_blk(blk_name, qdev_id, errp);
2356     if (!blk) {
2357         return -ENODEV;
2358     }
2359 
2360     if (!blk_dev_has_removable_media(blk)) {
2361         error_setg(errp, "Device '%s' is not removable", device);
2362         return -ENOTSUP;
2363     }
2364 
2365     if (!blk_dev_has_tray(blk)) {
2366         error_setg(errp, "Device '%s' does not have a tray", device);
2367         return -ENOSYS;
2368     }
2369 
2370     if (blk_dev_is_tray_open(blk)) {
2371         return 0;
2372     }
2373 
2374     locked = blk_dev_is_medium_locked(blk);
2375     if (locked) {
2376         blk_dev_eject_request(blk, force);
2377     }
2378 
2379     if (!locked || force) {
2380         blk_dev_change_media_cb(blk, false, &error_abort);
2381     }
2382 
2383     if (locked && !force) {
2384         error_setg(errp, "Device '%s' is locked and force was not specified, "
2385                    "wait for tray to open and try again", device);
2386         return -EINPROGRESS;
2387     }
2388 
2389     return 0;
2390 }
2391 
2392 void qmp_blockdev_open_tray(bool has_device, const char *device,
2393                             bool has_id, const char *id,
2394                             bool has_force, bool force,
2395                             Error **errp)
2396 {
2397     Error *local_err = NULL;
2398     int rc;
2399 
2400     if (!has_force) {
2401         force = false;
2402     }
2403     rc = do_open_tray(has_device ? device : NULL,
2404                       has_id ? id : NULL,
2405                       force, &local_err);
2406     if (rc && rc != -ENOSYS && rc != -EINPROGRESS) {
2407         error_propagate(errp, local_err);
2408         return;
2409     }
2410     error_free(local_err);
2411 }
2412 
2413 void qmp_blockdev_close_tray(bool has_device, const char *device,
2414                              bool has_id, const char *id,
2415                              Error **errp)
2416 {
2417     BlockBackend *blk;
2418     Error *local_err = NULL;
2419 
2420     device = has_device ? device : NULL;
2421     id = has_id ? id : NULL;
2422 
2423     blk = qmp_get_blk(device, id, errp);
2424     if (!blk) {
2425         return;
2426     }
2427 
2428     if (!blk_dev_has_removable_media(blk)) {
2429         error_setg(errp, "Device '%s' is not removable", device ?: id);
2430         return;
2431     }
2432 
2433     if (!blk_dev_has_tray(blk)) {
2434         /* Ignore this command on tray-less devices */
2435         return;
2436     }
2437 
2438     if (!blk_dev_is_tray_open(blk)) {
2439         return;
2440     }
2441 
2442     blk_dev_change_media_cb(blk, true, &local_err);
2443     if (local_err) {
2444         error_propagate(errp, local_err);
2445         return;
2446     }
2447 }
2448 
2449 void qmp_x_blockdev_remove_medium(bool has_device, const char *device,
2450                                   bool has_id, const char *id, Error **errp)
2451 {
2452     BlockBackend *blk;
2453     BlockDriverState *bs;
2454     AioContext *aio_context;
2455     bool has_attached_device;
2456 
2457     device = has_device ? device : NULL;
2458     id = has_id ? id : NULL;
2459 
2460     blk = qmp_get_blk(device, id, errp);
2461     if (!blk) {
2462         return;
2463     }
2464 
2465     /* For BBs without a device, we can exchange the BDS tree at will */
2466     has_attached_device = blk_get_attached_dev(blk);
2467 
2468     if (has_attached_device && !blk_dev_has_removable_media(blk)) {
2469         error_setg(errp, "Device '%s' is not removable", device ?: id);
2470         return;
2471     }
2472 
2473     if (has_attached_device && blk_dev_has_tray(blk) &&
2474         !blk_dev_is_tray_open(blk))
2475     {
2476         error_setg(errp, "Tray of device '%s' is not open", device ?: id);
2477         return;
2478     }
2479 
2480     bs = blk_bs(blk);
2481     if (!bs) {
2482         return;
2483     }
2484 
2485     aio_context = bdrv_get_aio_context(bs);
2486     aio_context_acquire(aio_context);
2487 
2488     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_EJECT, errp)) {
2489         goto out;
2490     }
2491 
2492     blk_remove_bs(blk);
2493 
2494     if (!blk_dev_has_tray(blk)) {
2495         /* For tray-less devices, blockdev-open-tray is a no-op (or may not be
2496          * called at all); therefore, the medium needs to be ejected here.
2497          * Do it after blk_remove_bs() so blk_is_inserted(blk) returns the @load
2498          * value passed here (i.e. false). */
2499         blk_dev_change_media_cb(blk, false, &error_abort);
2500     }
2501 
2502 out:
2503     aio_context_release(aio_context);
2504 }
2505 
2506 static void qmp_blockdev_insert_anon_medium(BlockBackend *blk,
2507                                             BlockDriverState *bs, Error **errp)
2508 {
2509     Error *local_err = NULL;
2510     bool has_device;
2511     int ret;
2512 
2513     /* For BBs without a device, we can exchange the BDS tree at will */
2514     has_device = blk_get_attached_dev(blk);
2515 
2516     if (has_device && !blk_dev_has_removable_media(blk)) {
2517         error_setg(errp, "Device is not removable");
2518         return;
2519     }
2520 
2521     if (has_device && blk_dev_has_tray(blk) && !blk_dev_is_tray_open(blk)) {
2522         error_setg(errp, "Tray of the device is not open");
2523         return;
2524     }
2525 
2526     if (blk_bs(blk)) {
2527         error_setg(errp, "There already is a medium in the device");
2528         return;
2529     }
2530 
2531     ret = blk_insert_bs(blk, bs, errp);
2532     if (ret < 0) {
2533         return;
2534     }
2535 
2536     if (!blk_dev_has_tray(blk)) {
2537         /* For tray-less devices, blockdev-close-tray is a no-op (or may not be
2538          * called at all); therefore, the medium needs to be pushed into the
2539          * slot here.
2540          * Do it after blk_insert_bs() so blk_is_inserted(blk) returns the @load
2541          * value passed here (i.e. true). */
2542         blk_dev_change_media_cb(blk, true, &local_err);
2543         if (local_err) {
2544             error_propagate(errp, local_err);
2545             blk_remove_bs(blk);
2546             return;
2547         }
2548     }
2549 }
2550 
2551 void qmp_x_blockdev_insert_medium(bool has_device, const char *device,
2552                                   bool has_id, const char *id,
2553                                   const char *node_name, Error **errp)
2554 {
2555     BlockBackend *blk;
2556     BlockDriverState *bs;
2557 
2558     blk = qmp_get_blk(has_device ? device : NULL,
2559                       has_id ? id : NULL,
2560                       errp);
2561     if (!blk) {
2562         return;
2563     }
2564 
2565     bs = bdrv_find_node(node_name);
2566     if (!bs) {
2567         error_setg(errp, "Node '%s' not found", node_name);
2568         return;
2569     }
2570 
2571     if (bdrv_has_blk(bs)) {
2572         error_setg(errp, "Node '%s' is already in use", node_name);
2573         return;
2574     }
2575 
2576     qmp_blockdev_insert_anon_medium(blk, bs, errp);
2577 }
2578 
2579 void qmp_blockdev_change_medium(bool has_device, const char *device,
2580                                 bool has_id, const char *id,
2581                                 const char *filename,
2582                                 bool has_format, const char *format,
2583                                 bool has_read_only,
2584                                 BlockdevChangeReadOnlyMode read_only,
2585                                 Error **errp)
2586 {
2587     BlockBackend *blk;
2588     BlockDriverState *medium_bs = NULL;
2589     int bdrv_flags;
2590     bool detect_zeroes;
2591     int rc;
2592     QDict *options = NULL;
2593     Error *err = NULL;
2594 
2595     blk = qmp_get_blk(has_device ? device : NULL,
2596                       has_id ? id : NULL,
2597                       errp);
2598     if (!blk) {
2599         goto fail;
2600     }
2601 
2602     if (blk_bs(blk)) {
2603         blk_update_root_state(blk);
2604     }
2605 
2606     bdrv_flags = blk_get_open_flags_from_root_state(blk);
2607     bdrv_flags &= ~(BDRV_O_TEMPORARY | BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING |
2608         BDRV_O_PROTOCOL);
2609 
2610     if (!has_read_only) {
2611         read_only = BLOCKDEV_CHANGE_READ_ONLY_MODE_RETAIN;
2612     }
2613 
2614     switch (read_only) {
2615     case BLOCKDEV_CHANGE_READ_ONLY_MODE_RETAIN:
2616         break;
2617 
2618     case BLOCKDEV_CHANGE_READ_ONLY_MODE_READ_ONLY:
2619         bdrv_flags &= ~BDRV_O_RDWR;
2620         break;
2621 
2622     case BLOCKDEV_CHANGE_READ_ONLY_MODE_READ_WRITE:
2623         bdrv_flags |= BDRV_O_RDWR;
2624         break;
2625 
2626     default:
2627         abort();
2628     }
2629 
2630     options = qdict_new();
2631     detect_zeroes = blk_get_detect_zeroes_from_root_state(blk);
2632     qdict_put_str(options, "detect-zeroes", detect_zeroes ? "on" : "off");
2633 
2634     if (has_format) {
2635         qdict_put_str(options, "driver", format);
2636     }
2637 
2638     medium_bs = bdrv_open(filename, NULL, options, bdrv_flags, errp);
2639     if (!medium_bs) {
2640         goto fail;
2641     }
2642 
2643     rc = do_open_tray(has_device ? device : NULL,
2644                       has_id ? id : NULL,
2645                       false, &err);
2646     if (rc && rc != -ENOSYS) {
2647         error_propagate(errp, err);
2648         goto fail;
2649     }
2650     error_free(err);
2651     err = NULL;
2652 
2653     qmp_x_blockdev_remove_medium(has_device, device, has_id, id, &err);
2654     if (err) {
2655         error_propagate(errp, err);
2656         goto fail;
2657     }
2658 
2659     qmp_blockdev_insert_anon_medium(blk, medium_bs, &err);
2660     if (err) {
2661         error_propagate(errp, err);
2662         goto fail;
2663     }
2664 
2665     qmp_blockdev_close_tray(has_device, device, has_id, id, errp);
2666 
2667 fail:
2668     /* If the medium has been inserted, the device has its own reference, so
2669      * ours must be relinquished; and if it has not been inserted successfully,
2670      * the reference must be relinquished anyway */
2671     bdrv_unref(medium_bs);
2672 }
2673 
2674 /* throttling disk I/O limits */
2675 void qmp_block_set_io_throttle(BlockIOThrottle *arg, Error **errp)
2676 {
2677     ThrottleConfig cfg;
2678     BlockDriverState *bs;
2679     BlockBackend *blk;
2680     AioContext *aio_context;
2681 
2682     blk = qmp_get_blk(arg->has_device ? arg->device : NULL,
2683                       arg->has_id ? arg->id : NULL,
2684                       errp);
2685     if (!blk) {
2686         return;
2687     }
2688 
2689     aio_context = blk_get_aio_context(blk);
2690     aio_context_acquire(aio_context);
2691 
2692     bs = blk_bs(blk);
2693     if (!bs) {
2694         error_setg(errp, "Device has no medium");
2695         goto out;
2696     }
2697 
2698     throttle_config_init(&cfg);
2699     cfg.buckets[THROTTLE_BPS_TOTAL].avg = arg->bps;
2700     cfg.buckets[THROTTLE_BPS_READ].avg  = arg->bps_rd;
2701     cfg.buckets[THROTTLE_BPS_WRITE].avg = arg->bps_wr;
2702 
2703     cfg.buckets[THROTTLE_OPS_TOTAL].avg = arg->iops;
2704     cfg.buckets[THROTTLE_OPS_READ].avg  = arg->iops_rd;
2705     cfg.buckets[THROTTLE_OPS_WRITE].avg = arg->iops_wr;
2706 
2707     if (arg->has_bps_max) {
2708         cfg.buckets[THROTTLE_BPS_TOTAL].max = arg->bps_max;
2709     }
2710     if (arg->has_bps_rd_max) {
2711         cfg.buckets[THROTTLE_BPS_READ].max = arg->bps_rd_max;
2712     }
2713     if (arg->has_bps_wr_max) {
2714         cfg.buckets[THROTTLE_BPS_WRITE].max = arg->bps_wr_max;
2715     }
2716     if (arg->has_iops_max) {
2717         cfg.buckets[THROTTLE_OPS_TOTAL].max = arg->iops_max;
2718     }
2719     if (arg->has_iops_rd_max) {
2720         cfg.buckets[THROTTLE_OPS_READ].max = arg->iops_rd_max;
2721     }
2722     if (arg->has_iops_wr_max) {
2723         cfg.buckets[THROTTLE_OPS_WRITE].max = arg->iops_wr_max;
2724     }
2725 
2726     if (arg->has_bps_max_length) {
2727         cfg.buckets[THROTTLE_BPS_TOTAL].burst_length = arg->bps_max_length;
2728     }
2729     if (arg->has_bps_rd_max_length) {
2730         cfg.buckets[THROTTLE_BPS_READ].burst_length = arg->bps_rd_max_length;
2731     }
2732     if (arg->has_bps_wr_max_length) {
2733         cfg.buckets[THROTTLE_BPS_WRITE].burst_length = arg->bps_wr_max_length;
2734     }
2735     if (arg->has_iops_max_length) {
2736         cfg.buckets[THROTTLE_OPS_TOTAL].burst_length = arg->iops_max_length;
2737     }
2738     if (arg->has_iops_rd_max_length) {
2739         cfg.buckets[THROTTLE_OPS_READ].burst_length = arg->iops_rd_max_length;
2740     }
2741     if (arg->has_iops_wr_max_length) {
2742         cfg.buckets[THROTTLE_OPS_WRITE].burst_length = arg->iops_wr_max_length;
2743     }
2744 
2745     if (arg->has_iops_size) {
2746         cfg.op_size = arg->iops_size;
2747     }
2748 
2749     if (!throttle_is_valid(&cfg, errp)) {
2750         goto out;
2751     }
2752 
2753     if (throttle_enabled(&cfg)) {
2754         /* Enable I/O limits if they're not enabled yet, otherwise
2755          * just update the throttling group. */
2756         if (!blk_get_public(blk)->throttle_group_member.throttle_state) {
2757             blk_io_limits_enable(blk,
2758                                  arg->has_group ? arg->group :
2759                                  arg->has_device ? arg->device :
2760                                  arg->id);
2761         } else if (arg->has_group) {
2762             blk_io_limits_update_group(blk, arg->group);
2763         }
2764         /* Set the new throttling configuration */
2765         blk_set_io_limits(blk, &cfg);
2766     } else if (blk_get_public(blk)->throttle_group_member.throttle_state) {
2767         /* If all throttling settings are set to 0, disable I/O limits */
2768         blk_io_limits_disable(blk);
2769     }
2770 
2771 out:
2772     aio_context_release(aio_context);
2773 }
2774 
2775 void qmp_block_dirty_bitmap_add(const char *node, const char *name,
2776                                 bool has_granularity, uint32_t granularity,
2777                                 bool has_persistent, bool persistent,
2778                                 bool has_autoload, bool autoload,
2779                                 Error **errp)
2780 {
2781     BlockDriverState *bs;
2782     BdrvDirtyBitmap *bitmap;
2783 
2784     if (!name || name[0] == '\0') {
2785         error_setg(errp, "Bitmap name cannot be empty");
2786         return;
2787     }
2788 
2789     bs = bdrv_lookup_bs(node, node, errp);
2790     if (!bs) {
2791         return;
2792     }
2793 
2794     if (has_granularity) {
2795         if (granularity < 512 || !is_power_of_2(granularity)) {
2796             error_setg(errp, "Granularity must be power of 2 "
2797                              "and at least 512");
2798             return;
2799         }
2800     } else {
2801         /* Default to cluster size, if available: */
2802         granularity = bdrv_get_default_bitmap_granularity(bs);
2803     }
2804 
2805     if (!has_persistent) {
2806         persistent = false;
2807     }
2808     if (!has_autoload) {
2809         autoload = false;
2810     }
2811 
2812     if (has_autoload && !persistent) {
2813         error_setg(errp, "Autoload flag must be used only for persistent "
2814                          "bitmaps");
2815         return;
2816     }
2817 
2818     if (persistent &&
2819         !bdrv_can_store_new_dirty_bitmap(bs, name, granularity, errp))
2820     {
2821         return;
2822     }
2823 
2824     bitmap = bdrv_create_dirty_bitmap(bs, granularity, name, errp);
2825     if (bitmap == NULL) {
2826         return;
2827     }
2828 
2829     bdrv_dirty_bitmap_set_persistance(bitmap, persistent);
2830     bdrv_dirty_bitmap_set_autoload(bitmap, autoload);
2831 }
2832 
2833 void qmp_block_dirty_bitmap_remove(const char *node, const char *name,
2834                                    Error **errp)
2835 {
2836     BlockDriverState *bs;
2837     BdrvDirtyBitmap *bitmap;
2838     Error *local_err = NULL;
2839 
2840     bitmap = block_dirty_bitmap_lookup(node, name, &bs, errp);
2841     if (!bitmap || !bs) {
2842         return;
2843     }
2844 
2845     if (bdrv_dirty_bitmap_frozen(bitmap)) {
2846         error_setg(errp,
2847                    "Bitmap '%s' is currently frozen and cannot be removed",
2848                    name);
2849         return;
2850     }
2851 
2852     if (bdrv_dirty_bitmap_get_persistance(bitmap)) {
2853         bdrv_remove_persistent_dirty_bitmap(bs, name, &local_err);
2854         if (local_err != NULL) {
2855             error_propagate(errp, local_err);
2856             return;
2857         }
2858     }
2859 
2860     bdrv_dirty_bitmap_make_anon(bitmap);
2861     bdrv_release_dirty_bitmap(bs, bitmap);
2862 }
2863 
2864 /**
2865  * Completely clear a bitmap, for the purposes of synchronizing a bitmap
2866  * immediately after a full backup operation.
2867  */
2868 void qmp_block_dirty_bitmap_clear(const char *node, const char *name,
2869                                   Error **errp)
2870 {
2871     BdrvDirtyBitmap *bitmap;
2872     BlockDriverState *bs;
2873 
2874     bitmap = block_dirty_bitmap_lookup(node, name, &bs, errp);
2875     if (!bitmap || !bs) {
2876         return;
2877     }
2878 
2879     if (bdrv_dirty_bitmap_frozen(bitmap)) {
2880         error_setg(errp,
2881                    "Bitmap '%s' is currently frozen and cannot be modified",
2882                    name);
2883         return;
2884     } else if (!bdrv_dirty_bitmap_enabled(bitmap)) {
2885         error_setg(errp,
2886                    "Bitmap '%s' is currently disabled and cannot be cleared",
2887                    name);
2888         return;
2889     } else if (bdrv_dirty_bitmap_readonly(bitmap)) {
2890         error_setg(errp, "Bitmap '%s' is readonly and cannot be cleared", name);
2891         return;
2892     }
2893 
2894     bdrv_clear_dirty_bitmap(bitmap, NULL);
2895 }
2896 
2897 BlockDirtyBitmapSha256 *qmp_x_debug_block_dirty_bitmap_sha256(const char *node,
2898                                                               const char *name,
2899                                                               Error **errp)
2900 {
2901     BdrvDirtyBitmap *bitmap;
2902     BlockDriverState *bs;
2903     BlockDirtyBitmapSha256 *ret = NULL;
2904     char *sha256;
2905 
2906     bitmap = block_dirty_bitmap_lookup(node, name, &bs, errp);
2907     if (!bitmap || !bs) {
2908         return NULL;
2909     }
2910 
2911     sha256 = bdrv_dirty_bitmap_sha256(bitmap, errp);
2912     if (sha256 == NULL) {
2913         return NULL;
2914     }
2915 
2916     ret = g_new(BlockDirtyBitmapSha256, 1);
2917     ret->sha256 = sha256;
2918 
2919     return ret;
2920 }
2921 
2922 void hmp_drive_del(Monitor *mon, const QDict *qdict)
2923 {
2924     const char *id = qdict_get_str(qdict, "id");
2925     BlockBackend *blk;
2926     BlockDriverState *bs;
2927     AioContext *aio_context;
2928     Error *local_err = NULL;
2929 
2930     bs = bdrv_find_node(id);
2931     if (bs) {
2932         qmp_blockdev_del(id, &local_err);
2933         if (local_err) {
2934             error_report_err(local_err);
2935         }
2936         return;
2937     }
2938 
2939     blk = blk_by_name(id);
2940     if (!blk) {
2941         error_report("Device '%s' not found", id);
2942         return;
2943     }
2944 
2945     if (!blk_legacy_dinfo(blk)) {
2946         error_report("Deleting device added with blockdev-add"
2947                      " is not supported");
2948         return;
2949     }
2950 
2951     aio_context = blk_get_aio_context(blk);
2952     aio_context_acquire(aio_context);
2953 
2954     bs = blk_bs(blk);
2955     if (bs) {
2956         if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_DRIVE_DEL, &local_err)) {
2957             error_report_err(local_err);
2958             aio_context_release(aio_context);
2959             return;
2960         }
2961 
2962         blk_remove_bs(blk);
2963     }
2964 
2965     /* Make the BlockBackend and the attached BlockDriverState anonymous */
2966     monitor_remove_blk(blk);
2967 
2968     /* If this BlockBackend has a device attached to it, its refcount will be
2969      * decremented when the device is removed; otherwise we have to do so here.
2970      */
2971     if (blk_get_attached_dev(blk)) {
2972         /* Further I/O must not pause the guest */
2973         blk_set_on_error(blk, BLOCKDEV_ON_ERROR_REPORT,
2974                          BLOCKDEV_ON_ERROR_REPORT);
2975     } else {
2976         blk_unref(blk);
2977     }
2978 
2979     aio_context_release(aio_context);
2980 }
2981 
2982 void qmp_block_resize(bool has_device, const char *device,
2983                       bool has_node_name, const char *node_name,
2984                       int64_t size, Error **errp)
2985 {
2986     Error *local_err = NULL;
2987     BlockBackend *blk = NULL;
2988     BlockDriverState *bs;
2989     AioContext *aio_context;
2990     int ret;
2991 
2992     bs = bdrv_lookup_bs(has_device ? device : NULL,
2993                         has_node_name ? node_name : NULL,
2994                         &local_err);
2995     if (local_err) {
2996         error_propagate(errp, local_err);
2997         return;
2998     }
2999 
3000     aio_context = bdrv_get_aio_context(bs);
3001     aio_context_acquire(aio_context);
3002 
3003     if (!bdrv_is_first_non_filter(bs)) {
3004         error_setg(errp, QERR_FEATURE_DISABLED, "resize");
3005         goto out;
3006     }
3007 
3008     if (size < 0) {
3009         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "size", "a >0 size");
3010         goto out;
3011     }
3012 
3013     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_RESIZE, NULL)) {
3014         error_setg(errp, QERR_DEVICE_IN_USE, device);
3015         goto out;
3016     }
3017 
3018     blk = blk_new(BLK_PERM_RESIZE, BLK_PERM_ALL);
3019     ret = blk_insert_bs(blk, bs, errp);
3020     if (ret < 0) {
3021         goto out;
3022     }
3023 
3024     bdrv_drained_begin(bs);
3025     ret = blk_truncate(blk, size, PREALLOC_MODE_OFF, errp);
3026     bdrv_drained_end(bs);
3027 
3028 out:
3029     blk_unref(blk);
3030     aio_context_release(aio_context);
3031 }
3032 
3033 void qmp_block_stream(bool has_job_id, const char *job_id, const char *device,
3034                       bool has_base, const char *base,
3035                       bool has_base_node, const char *base_node,
3036                       bool has_backing_file, const char *backing_file,
3037                       bool has_speed, int64_t speed,
3038                       bool has_on_error, BlockdevOnError on_error,
3039                       Error **errp)
3040 {
3041     BlockDriverState *bs, *iter;
3042     BlockDriverState *base_bs = NULL;
3043     AioContext *aio_context;
3044     Error *local_err = NULL;
3045     const char *base_name = NULL;
3046 
3047     if (!has_on_error) {
3048         on_error = BLOCKDEV_ON_ERROR_REPORT;
3049     }
3050 
3051     bs = bdrv_lookup_bs(device, device, errp);
3052     if (!bs) {
3053         return;
3054     }
3055 
3056     aio_context = bdrv_get_aio_context(bs);
3057     aio_context_acquire(aio_context);
3058 
3059     if (has_base && has_base_node) {
3060         error_setg(errp, "'base' and 'base-node' cannot be specified "
3061                    "at the same time");
3062         goto out;
3063     }
3064 
3065     if (has_base) {
3066         base_bs = bdrv_find_backing_image(bs, base);
3067         if (base_bs == NULL) {
3068             error_setg(errp, QERR_BASE_NOT_FOUND, base);
3069             goto out;
3070         }
3071         assert(bdrv_get_aio_context(base_bs) == aio_context);
3072         base_name = base;
3073     }
3074 
3075     if (has_base_node) {
3076         base_bs = bdrv_lookup_bs(NULL, base_node, errp);
3077         if (!base_bs) {
3078             goto out;
3079         }
3080         if (bs == base_bs || !bdrv_chain_contains(bs, base_bs)) {
3081             error_setg(errp, "Node '%s' is not a backing image of '%s'",
3082                        base_node, device);
3083             goto out;
3084         }
3085         assert(bdrv_get_aio_context(base_bs) == aio_context);
3086         base_name = base_bs->filename;
3087     }
3088 
3089     /* Check for op blockers in the whole chain between bs and base */
3090     for (iter = bs; iter && iter != base_bs; iter = backing_bs(iter)) {
3091         if (bdrv_op_is_blocked(iter, BLOCK_OP_TYPE_STREAM, errp)) {
3092             goto out;
3093         }
3094     }
3095 
3096     /* if we are streaming the entire chain, the result will have no backing
3097      * file, and specifying one is therefore an error */
3098     if (base_bs == NULL && has_backing_file) {
3099         error_setg(errp, "backing file specified, but streaming the "
3100                          "entire chain");
3101         goto out;
3102     }
3103 
3104     /* backing_file string overrides base bs filename */
3105     base_name = has_backing_file ? backing_file : base_name;
3106 
3107     stream_start(has_job_id ? job_id : NULL, bs, base_bs, base_name,
3108                  has_speed ? speed : 0, on_error, &local_err);
3109     if (local_err) {
3110         error_propagate(errp, local_err);
3111         goto out;
3112     }
3113 
3114     trace_qmp_block_stream(bs, bs->job);
3115 
3116 out:
3117     aio_context_release(aio_context);
3118 }
3119 
3120 void qmp_block_commit(bool has_job_id, const char *job_id, const char *device,
3121                       bool has_base, const char *base,
3122                       bool has_top, const char *top,
3123                       bool has_backing_file, const char *backing_file,
3124                       bool has_speed, int64_t speed,
3125                       bool has_filter_node_name, const char *filter_node_name,
3126                       Error **errp)
3127 {
3128     BlockDriverState *bs;
3129     BlockDriverState *iter;
3130     BlockDriverState *base_bs, *top_bs;
3131     AioContext *aio_context;
3132     Error *local_err = NULL;
3133     /* This will be part of the QMP command, if/when the
3134      * BlockdevOnError change for blkmirror makes it in
3135      */
3136     BlockdevOnError on_error = BLOCKDEV_ON_ERROR_REPORT;
3137 
3138     if (!has_speed) {
3139         speed = 0;
3140     }
3141     if (!has_filter_node_name) {
3142         filter_node_name = NULL;
3143     }
3144 
3145     /* Important Note:
3146      *  libvirt relies on the DeviceNotFound error class in order to probe for
3147      *  live commit feature versions; for this to work, we must make sure to
3148      *  perform the device lookup before any generic errors that may occur in a
3149      *  scenario in which all optional arguments are omitted. */
3150     bs = qmp_get_root_bs(device, &local_err);
3151     if (!bs) {
3152         bs = bdrv_lookup_bs(device, device, NULL);
3153         if (!bs) {
3154             error_free(local_err);
3155             error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
3156                       "Device '%s' not found", device);
3157         } else {
3158             error_propagate(errp, local_err);
3159         }
3160         return;
3161     }
3162 
3163     aio_context = bdrv_get_aio_context(bs);
3164     aio_context_acquire(aio_context);
3165 
3166     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_COMMIT_SOURCE, errp)) {
3167         goto out;
3168     }
3169 
3170     /* default top_bs is the active layer */
3171     top_bs = bs;
3172 
3173     if (has_top && top) {
3174         if (strcmp(bs->filename, top) != 0) {
3175             top_bs = bdrv_find_backing_image(bs, top);
3176         }
3177     }
3178 
3179     if (top_bs == NULL) {
3180         error_setg(errp, "Top image file %s not found", top ? top : "NULL");
3181         goto out;
3182     }
3183 
3184     assert(bdrv_get_aio_context(top_bs) == aio_context);
3185 
3186     if (has_base && base) {
3187         base_bs = bdrv_find_backing_image(top_bs, base);
3188     } else {
3189         base_bs = bdrv_find_base(top_bs);
3190     }
3191 
3192     if (base_bs == NULL) {
3193         error_setg(errp, QERR_BASE_NOT_FOUND, base ? base : "NULL");
3194         goto out;
3195     }
3196 
3197     assert(bdrv_get_aio_context(base_bs) == aio_context);
3198 
3199     for (iter = top_bs; iter != backing_bs(base_bs); iter = backing_bs(iter)) {
3200         if (bdrv_op_is_blocked(iter, BLOCK_OP_TYPE_COMMIT_TARGET, errp)) {
3201             goto out;
3202         }
3203     }
3204 
3205     /* Do not allow attempts to commit an image into itself */
3206     if (top_bs == base_bs) {
3207         error_setg(errp, "cannot commit an image into itself");
3208         goto out;
3209     }
3210 
3211     if (top_bs == bs) {
3212         if (has_backing_file) {
3213             error_setg(errp, "'backing-file' specified,"
3214                              " but 'top' is the active layer");
3215             goto out;
3216         }
3217         commit_active_start(has_job_id ? job_id : NULL, bs, base_bs,
3218                             BLOCK_JOB_DEFAULT, speed, on_error,
3219                             filter_node_name, NULL, NULL, false, &local_err);
3220     } else {
3221         BlockDriverState *overlay_bs = bdrv_find_overlay(bs, top_bs);
3222         if (bdrv_op_is_blocked(overlay_bs, BLOCK_OP_TYPE_COMMIT_TARGET, errp)) {
3223             goto out;
3224         }
3225         commit_start(has_job_id ? job_id : NULL, bs, base_bs, top_bs, speed,
3226                      on_error, has_backing_file ? backing_file : NULL,
3227                      filter_node_name, &local_err);
3228     }
3229     if (local_err != NULL) {
3230         error_propagate(errp, local_err);
3231         goto out;
3232     }
3233 
3234 out:
3235     aio_context_release(aio_context);
3236 }
3237 
3238 static BlockJob *do_drive_backup(DriveBackup *backup, BlockJobTxn *txn,
3239                                  Error **errp)
3240 {
3241     BlockDriverState *bs;
3242     BlockDriverState *target_bs;
3243     BlockDriverState *source = NULL;
3244     BlockJob *job = NULL;
3245     BdrvDirtyBitmap *bmap = NULL;
3246     AioContext *aio_context;
3247     QDict *options = NULL;
3248     Error *local_err = NULL;
3249     int flags;
3250     int64_t size;
3251     bool set_backing_hd = false;
3252 
3253     if (!backup->has_speed) {
3254         backup->speed = 0;
3255     }
3256     if (!backup->has_on_source_error) {
3257         backup->on_source_error = BLOCKDEV_ON_ERROR_REPORT;
3258     }
3259     if (!backup->has_on_target_error) {
3260         backup->on_target_error = BLOCKDEV_ON_ERROR_REPORT;
3261     }
3262     if (!backup->has_mode) {
3263         backup->mode = NEW_IMAGE_MODE_ABSOLUTE_PATHS;
3264     }
3265     if (!backup->has_job_id) {
3266         backup->job_id = NULL;
3267     }
3268     if (!backup->has_compress) {
3269         backup->compress = false;
3270     }
3271 
3272     bs = qmp_get_root_bs(backup->device, errp);
3273     if (!bs) {
3274         return NULL;
3275     }
3276 
3277     aio_context = bdrv_get_aio_context(bs);
3278     aio_context_acquire(aio_context);
3279 
3280     if (!backup->has_format) {
3281         backup->format = backup->mode == NEW_IMAGE_MODE_EXISTING ?
3282                          NULL : (char*) bs->drv->format_name;
3283     }
3284 
3285     /* Early check to avoid creating target */
3286     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_BACKUP_SOURCE, errp)) {
3287         goto out;
3288     }
3289 
3290     flags = bs->open_flags | BDRV_O_RDWR;
3291 
3292     /* See if we have a backing HD we can use to create our new image
3293      * on top of. */
3294     if (backup->sync == MIRROR_SYNC_MODE_TOP) {
3295         source = backing_bs(bs);
3296         if (!source) {
3297             backup->sync = MIRROR_SYNC_MODE_FULL;
3298         }
3299     }
3300     if (backup->sync == MIRROR_SYNC_MODE_NONE) {
3301         source = bs;
3302         flags |= BDRV_O_NO_BACKING;
3303         set_backing_hd = true;
3304     }
3305 
3306     size = bdrv_getlength(bs);
3307     if (size < 0) {
3308         error_setg_errno(errp, -size, "bdrv_getlength failed");
3309         goto out;
3310     }
3311 
3312     if (backup->mode != NEW_IMAGE_MODE_EXISTING) {
3313         assert(backup->format);
3314         if (source) {
3315             bdrv_img_create(backup->target, backup->format, source->filename,
3316                             source->drv->format_name, NULL,
3317                             size, flags, false, &local_err);
3318         } else {
3319             bdrv_img_create(backup->target, backup->format, NULL, NULL, NULL,
3320                             size, flags, false, &local_err);
3321         }
3322     }
3323 
3324     if (local_err) {
3325         error_propagate(errp, local_err);
3326         goto out;
3327     }
3328 
3329     if (backup->format) {
3330         if (!options) {
3331             options = qdict_new();
3332         }
3333         qdict_put_str(options, "driver", backup->format);
3334     }
3335 
3336     target_bs = bdrv_open(backup->target, NULL, options, flags, errp);
3337     if (!target_bs) {
3338         goto out;
3339     }
3340 
3341     bdrv_set_aio_context(target_bs, aio_context);
3342 
3343     if (set_backing_hd) {
3344         bdrv_set_backing_hd(target_bs, source, &local_err);
3345         if (local_err) {
3346             bdrv_unref(target_bs);
3347             goto out;
3348         }
3349     }
3350 
3351     if (backup->has_bitmap) {
3352         bmap = bdrv_find_dirty_bitmap(bs, backup->bitmap);
3353         if (!bmap) {
3354             error_setg(errp, "Bitmap '%s' could not be found", backup->bitmap);
3355             bdrv_unref(target_bs);
3356             goto out;
3357         }
3358     }
3359 
3360     job = backup_job_create(backup->job_id, bs, target_bs, backup->speed,
3361                             backup->sync, bmap, backup->compress,
3362                             backup->on_source_error, backup->on_target_error,
3363                             BLOCK_JOB_DEFAULT, NULL, NULL, txn, &local_err);
3364     bdrv_unref(target_bs);
3365     if (local_err != NULL) {
3366         error_propagate(errp, local_err);
3367         goto out;
3368     }
3369 
3370 out:
3371     aio_context_release(aio_context);
3372     return job;
3373 }
3374 
3375 void qmp_drive_backup(DriveBackup *arg, Error **errp)
3376 {
3377 
3378     BlockJob *job;
3379     job = do_drive_backup(arg, NULL, errp);
3380     if (job) {
3381         block_job_start(job);
3382     }
3383 }
3384 
3385 BlockDeviceInfoList *qmp_query_named_block_nodes(Error **errp)
3386 {
3387     return bdrv_named_nodes_list(errp);
3388 }
3389 
3390 BlockJob *do_blockdev_backup(BlockdevBackup *backup, BlockJobTxn *txn,
3391                              Error **errp)
3392 {
3393     BlockDriverState *bs;
3394     BlockDriverState *target_bs;
3395     Error *local_err = NULL;
3396     AioContext *aio_context;
3397     BlockJob *job = NULL;
3398 
3399     if (!backup->has_speed) {
3400         backup->speed = 0;
3401     }
3402     if (!backup->has_on_source_error) {
3403         backup->on_source_error = BLOCKDEV_ON_ERROR_REPORT;
3404     }
3405     if (!backup->has_on_target_error) {
3406         backup->on_target_error = BLOCKDEV_ON_ERROR_REPORT;
3407     }
3408     if (!backup->has_job_id) {
3409         backup->job_id = NULL;
3410     }
3411     if (!backup->has_compress) {
3412         backup->compress = false;
3413     }
3414 
3415     bs = qmp_get_root_bs(backup->device, errp);
3416     if (!bs) {
3417         return NULL;
3418     }
3419 
3420     aio_context = bdrv_get_aio_context(bs);
3421     aio_context_acquire(aio_context);
3422 
3423     target_bs = bdrv_lookup_bs(backup->target, backup->target, errp);
3424     if (!target_bs) {
3425         goto out;
3426     }
3427 
3428     if (bdrv_get_aio_context(target_bs) != aio_context) {
3429         if (!bdrv_has_blk(target_bs)) {
3430             /* The target BDS is not attached, we can safely move it to another
3431              * AioContext. */
3432             bdrv_set_aio_context(target_bs, aio_context);
3433         } else {
3434             error_setg(errp, "Target is attached to a different thread from "
3435                              "source.");
3436             goto out;
3437         }
3438     }
3439     job = backup_job_create(backup->job_id, bs, target_bs, backup->speed,
3440                             backup->sync, NULL, backup->compress,
3441                             backup->on_source_error, backup->on_target_error,
3442                             BLOCK_JOB_DEFAULT, NULL, NULL, txn, &local_err);
3443     if (local_err != NULL) {
3444         error_propagate(errp, local_err);
3445     }
3446 out:
3447     aio_context_release(aio_context);
3448     return job;
3449 }
3450 
3451 void qmp_blockdev_backup(BlockdevBackup *arg, Error **errp)
3452 {
3453     BlockJob *job;
3454     job = do_blockdev_backup(arg, NULL, errp);
3455     if (job) {
3456         block_job_start(job);
3457     }
3458 }
3459 
3460 /* Parameter check and block job starting for drive mirroring.
3461  * Caller should hold @device and @target's aio context (must be the same).
3462  **/
3463 static void blockdev_mirror_common(const char *job_id, BlockDriverState *bs,
3464                                    BlockDriverState *target,
3465                                    bool has_replaces, const char *replaces,
3466                                    enum MirrorSyncMode sync,
3467                                    BlockMirrorBackingMode backing_mode,
3468                                    bool has_speed, int64_t speed,
3469                                    bool has_granularity, uint32_t granularity,
3470                                    bool has_buf_size, int64_t buf_size,
3471                                    bool has_on_source_error,
3472                                    BlockdevOnError on_source_error,
3473                                    bool has_on_target_error,
3474                                    BlockdevOnError on_target_error,
3475                                    bool has_unmap, bool unmap,
3476                                    bool has_filter_node_name,
3477                                    const char *filter_node_name,
3478                                    Error **errp)
3479 {
3480 
3481     if (!has_speed) {
3482         speed = 0;
3483     }
3484     if (!has_on_source_error) {
3485         on_source_error = BLOCKDEV_ON_ERROR_REPORT;
3486     }
3487     if (!has_on_target_error) {
3488         on_target_error = BLOCKDEV_ON_ERROR_REPORT;
3489     }
3490     if (!has_granularity) {
3491         granularity = 0;
3492     }
3493     if (!has_buf_size) {
3494         buf_size = 0;
3495     }
3496     if (!has_unmap) {
3497         unmap = true;
3498     }
3499     if (!has_filter_node_name) {
3500         filter_node_name = NULL;
3501     }
3502 
3503     if (granularity != 0 && (granularity < 512 || granularity > 1048576 * 64)) {
3504         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "granularity",
3505                    "a value in range [512B, 64MB]");
3506         return;
3507     }
3508     if (granularity & (granularity - 1)) {
3509         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "granularity",
3510                    "power of 2");
3511         return;
3512     }
3513 
3514     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_MIRROR_SOURCE, errp)) {
3515         return;
3516     }
3517     if (bdrv_op_is_blocked(target, BLOCK_OP_TYPE_MIRROR_TARGET, errp)) {
3518         return;
3519     }
3520 
3521     if (!bs->backing && sync == MIRROR_SYNC_MODE_TOP) {
3522         sync = MIRROR_SYNC_MODE_FULL;
3523     }
3524 
3525     /* pass the node name to replace to mirror start since it's loose coupling
3526      * and will allow to check whether the node still exist at mirror completion
3527      */
3528     mirror_start(job_id, bs, target,
3529                  has_replaces ? replaces : NULL,
3530                  speed, granularity, buf_size, sync, backing_mode,
3531                  on_source_error, on_target_error, unmap, filter_node_name,
3532                  errp);
3533 }
3534 
3535 void qmp_drive_mirror(DriveMirror *arg, Error **errp)
3536 {
3537     BlockDriverState *bs;
3538     BlockDriverState *source, *target_bs;
3539     AioContext *aio_context;
3540     BlockMirrorBackingMode backing_mode;
3541     Error *local_err = NULL;
3542     QDict *options = NULL;
3543     int flags;
3544     int64_t size;
3545     const char *format = arg->format;
3546 
3547     bs = qmp_get_root_bs(arg->device, errp);
3548     if (!bs) {
3549         return;
3550     }
3551 
3552     aio_context = bdrv_get_aio_context(bs);
3553     aio_context_acquire(aio_context);
3554 
3555     if (!arg->has_mode) {
3556         arg->mode = NEW_IMAGE_MODE_ABSOLUTE_PATHS;
3557     }
3558 
3559     if (!arg->has_format) {
3560         format = (arg->mode == NEW_IMAGE_MODE_EXISTING
3561                   ? NULL : bs->drv->format_name);
3562     }
3563 
3564     flags = bs->open_flags | BDRV_O_RDWR;
3565     source = backing_bs(bs);
3566     if (!source && arg->sync == MIRROR_SYNC_MODE_TOP) {
3567         arg->sync = MIRROR_SYNC_MODE_FULL;
3568     }
3569     if (arg->sync == MIRROR_SYNC_MODE_NONE) {
3570         source = bs;
3571     }
3572 
3573     size = bdrv_getlength(bs);
3574     if (size < 0) {
3575         error_setg_errno(errp, -size, "bdrv_getlength failed");
3576         goto out;
3577     }
3578 
3579     if (arg->has_replaces) {
3580         BlockDriverState *to_replace_bs;
3581         AioContext *replace_aio_context;
3582         int64_t replace_size;
3583 
3584         if (!arg->has_node_name) {
3585             error_setg(errp, "a node-name must be provided when replacing a"
3586                              " named node of the graph");
3587             goto out;
3588         }
3589 
3590         to_replace_bs = check_to_replace_node(bs, arg->replaces, &local_err);
3591 
3592         if (!to_replace_bs) {
3593             error_propagate(errp, local_err);
3594             goto out;
3595         }
3596 
3597         replace_aio_context = bdrv_get_aio_context(to_replace_bs);
3598         aio_context_acquire(replace_aio_context);
3599         replace_size = bdrv_getlength(to_replace_bs);
3600         aio_context_release(replace_aio_context);
3601 
3602         if (size != replace_size) {
3603             error_setg(errp, "cannot replace image with a mirror image of "
3604                              "different size");
3605             goto out;
3606         }
3607     }
3608 
3609     if (arg->mode == NEW_IMAGE_MODE_ABSOLUTE_PATHS) {
3610         backing_mode = MIRROR_SOURCE_BACKING_CHAIN;
3611     } else {
3612         backing_mode = MIRROR_OPEN_BACKING_CHAIN;
3613     }
3614 
3615     /* Don't open backing image in create() */
3616     flags |= BDRV_O_NO_BACKING;
3617 
3618     if ((arg->sync == MIRROR_SYNC_MODE_FULL || !source)
3619         && arg->mode != NEW_IMAGE_MODE_EXISTING)
3620     {
3621         /* create new image w/o backing file */
3622         assert(format);
3623         bdrv_img_create(arg->target, format,
3624                         NULL, NULL, NULL, size, flags, false, &local_err);
3625     } else {
3626         switch (arg->mode) {
3627         case NEW_IMAGE_MODE_EXISTING:
3628             break;
3629         case NEW_IMAGE_MODE_ABSOLUTE_PATHS:
3630             /* create new image with backing file */
3631             bdrv_img_create(arg->target, format,
3632                             source->filename,
3633                             source->drv->format_name,
3634                             NULL, size, flags, false, &local_err);
3635             break;
3636         default:
3637             abort();
3638         }
3639     }
3640 
3641     if (local_err) {
3642         error_propagate(errp, local_err);
3643         goto out;
3644     }
3645 
3646     options = qdict_new();
3647     if (arg->has_node_name) {
3648         qdict_put_str(options, "node-name", arg->node_name);
3649     }
3650     if (format) {
3651         qdict_put_str(options, "driver", format);
3652     }
3653 
3654     /* Mirroring takes care of copy-on-write using the source's backing
3655      * file.
3656      */
3657     target_bs = bdrv_open(arg->target, NULL, options, flags, errp);
3658     if (!target_bs) {
3659         goto out;
3660     }
3661 
3662     bdrv_set_aio_context(target_bs, aio_context);
3663 
3664     blockdev_mirror_common(arg->has_job_id ? arg->job_id : NULL, bs, target_bs,
3665                            arg->has_replaces, arg->replaces, arg->sync,
3666                            backing_mode, arg->has_speed, arg->speed,
3667                            arg->has_granularity, arg->granularity,
3668                            arg->has_buf_size, arg->buf_size,
3669                            arg->has_on_source_error, arg->on_source_error,
3670                            arg->has_on_target_error, arg->on_target_error,
3671                            arg->has_unmap, arg->unmap,
3672                            false, NULL,
3673                            &local_err);
3674     bdrv_unref(target_bs);
3675     error_propagate(errp, local_err);
3676 out:
3677     aio_context_release(aio_context);
3678 }
3679 
3680 void qmp_blockdev_mirror(bool has_job_id, const char *job_id,
3681                          const char *device, const char *target,
3682                          bool has_replaces, const char *replaces,
3683                          MirrorSyncMode sync,
3684                          bool has_speed, int64_t speed,
3685                          bool has_granularity, uint32_t granularity,
3686                          bool has_buf_size, int64_t buf_size,
3687                          bool has_on_source_error,
3688                          BlockdevOnError on_source_error,
3689                          bool has_on_target_error,
3690                          BlockdevOnError on_target_error,
3691                          bool has_filter_node_name,
3692                          const char *filter_node_name,
3693                          Error **errp)
3694 {
3695     BlockDriverState *bs;
3696     BlockDriverState *target_bs;
3697     AioContext *aio_context;
3698     BlockMirrorBackingMode backing_mode = MIRROR_LEAVE_BACKING_CHAIN;
3699     Error *local_err = NULL;
3700 
3701     bs = qmp_get_root_bs(device, errp);
3702     if (!bs) {
3703         return;
3704     }
3705 
3706     target_bs = bdrv_lookup_bs(target, target, errp);
3707     if (!target_bs) {
3708         return;
3709     }
3710 
3711     aio_context = bdrv_get_aio_context(bs);
3712     aio_context_acquire(aio_context);
3713 
3714     bdrv_set_aio_context(target_bs, aio_context);
3715 
3716     blockdev_mirror_common(has_job_id ? job_id : NULL, bs, target_bs,
3717                            has_replaces, replaces, sync, backing_mode,
3718                            has_speed, speed,
3719                            has_granularity, granularity,
3720                            has_buf_size, buf_size,
3721                            has_on_source_error, on_source_error,
3722                            has_on_target_error, on_target_error,
3723                            true, true,
3724                            has_filter_node_name, filter_node_name,
3725                            &local_err);
3726     error_propagate(errp, local_err);
3727 
3728     aio_context_release(aio_context);
3729 }
3730 
3731 /* Get a block job using its ID and acquire its AioContext */
3732 static BlockJob *find_block_job(const char *id, AioContext **aio_context,
3733                                 Error **errp)
3734 {
3735     BlockJob *job;
3736 
3737     assert(id != NULL);
3738 
3739     *aio_context = NULL;
3740 
3741     job = block_job_get(id);
3742 
3743     if (!job) {
3744         error_set(errp, ERROR_CLASS_DEVICE_NOT_ACTIVE,
3745                   "Block job '%s' not found", id);
3746         return NULL;
3747     }
3748 
3749     *aio_context = blk_get_aio_context(job->blk);
3750     aio_context_acquire(*aio_context);
3751 
3752     return job;
3753 }
3754 
3755 void qmp_block_job_set_speed(const char *device, int64_t speed, Error **errp)
3756 {
3757     AioContext *aio_context;
3758     BlockJob *job = find_block_job(device, &aio_context, errp);
3759 
3760     if (!job) {
3761         return;
3762     }
3763 
3764     block_job_set_speed(job, speed, errp);
3765     aio_context_release(aio_context);
3766 }
3767 
3768 void qmp_block_job_cancel(const char *device,
3769                           bool has_force, bool force, Error **errp)
3770 {
3771     AioContext *aio_context;
3772     BlockJob *job = find_block_job(device, &aio_context, errp);
3773 
3774     if (!job) {
3775         return;
3776     }
3777 
3778     if (!has_force) {
3779         force = false;
3780     }
3781 
3782     if (block_job_user_paused(job) && !force) {
3783         error_setg(errp, "The block job for device '%s' is currently paused",
3784                    device);
3785         goto out;
3786     }
3787 
3788     trace_qmp_block_job_cancel(job);
3789     block_job_cancel(job);
3790 out:
3791     aio_context_release(aio_context);
3792 }
3793 
3794 void qmp_block_job_pause(const char *device, Error **errp)
3795 {
3796     AioContext *aio_context;
3797     BlockJob *job = find_block_job(device, &aio_context, errp);
3798 
3799     if (!job || block_job_user_paused(job)) {
3800         return;
3801     }
3802 
3803     trace_qmp_block_job_pause(job);
3804     block_job_user_pause(job);
3805     aio_context_release(aio_context);
3806 }
3807 
3808 void qmp_block_job_resume(const char *device, Error **errp)
3809 {
3810     AioContext *aio_context;
3811     BlockJob *job = find_block_job(device, &aio_context, errp);
3812 
3813     if (!job || !block_job_user_paused(job)) {
3814         return;
3815     }
3816 
3817     trace_qmp_block_job_resume(job);
3818     block_job_user_resume(job);
3819     aio_context_release(aio_context);
3820 }
3821 
3822 void qmp_block_job_complete(const char *device, Error **errp)
3823 {
3824     AioContext *aio_context;
3825     BlockJob *job = find_block_job(device, &aio_context, errp);
3826 
3827     if (!job) {
3828         return;
3829     }
3830 
3831     trace_qmp_block_job_complete(job);
3832     block_job_complete(job, errp);
3833     aio_context_release(aio_context);
3834 }
3835 
3836 void qmp_change_backing_file(const char *device,
3837                              const char *image_node_name,
3838                              const char *backing_file,
3839                              Error **errp)
3840 {
3841     BlockDriverState *bs = NULL;
3842     AioContext *aio_context;
3843     BlockDriverState *image_bs = NULL;
3844     Error *local_err = NULL;
3845     bool ro;
3846     int open_flags;
3847     int ret;
3848 
3849     bs = qmp_get_root_bs(device, errp);
3850     if (!bs) {
3851         return;
3852     }
3853 
3854     aio_context = bdrv_get_aio_context(bs);
3855     aio_context_acquire(aio_context);
3856 
3857     image_bs = bdrv_lookup_bs(NULL, image_node_name, &local_err);
3858     if (local_err) {
3859         error_propagate(errp, local_err);
3860         goto out;
3861     }
3862 
3863     if (!image_bs) {
3864         error_setg(errp, "image file not found");
3865         goto out;
3866     }
3867 
3868     if (bdrv_find_base(image_bs) == image_bs) {
3869         error_setg(errp, "not allowing backing file change on an image "
3870                          "without a backing file");
3871         goto out;
3872     }
3873 
3874     /* even though we are not necessarily operating on bs, we need it to
3875      * determine if block ops are currently prohibited on the chain */
3876     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_CHANGE, errp)) {
3877         goto out;
3878     }
3879 
3880     /* final sanity check */
3881     if (!bdrv_chain_contains(bs, image_bs)) {
3882         error_setg(errp, "'%s' and image file are not in the same chain",
3883                    device);
3884         goto out;
3885     }
3886 
3887     /* if not r/w, reopen to make r/w */
3888     open_flags = image_bs->open_flags;
3889     ro = bdrv_is_read_only(image_bs);
3890 
3891     if (ro) {
3892         bdrv_reopen(image_bs, open_flags | BDRV_O_RDWR, &local_err);
3893         if (local_err) {
3894             error_propagate(errp, local_err);
3895             goto out;
3896         }
3897     }
3898 
3899     ret = bdrv_change_backing_file(image_bs, backing_file,
3900                                image_bs->drv ? image_bs->drv->format_name : "");
3901 
3902     if (ret < 0) {
3903         error_setg_errno(errp, -ret, "Could not change backing file to '%s'",
3904                          backing_file);
3905         /* don't exit here, so we can try to restore open flags if
3906          * appropriate */
3907     }
3908 
3909     if (ro) {
3910         bdrv_reopen(image_bs, open_flags, &local_err);
3911         error_propagate(errp, local_err);
3912     }
3913 
3914 out:
3915     aio_context_release(aio_context);
3916 }
3917 
3918 void hmp_drive_add_node(Monitor *mon, const char *optstr)
3919 {
3920     QemuOpts *opts;
3921     QDict *qdict;
3922     Error *local_err = NULL;
3923 
3924     opts = qemu_opts_parse_noisily(&qemu_drive_opts, optstr, false);
3925     if (!opts) {
3926         return;
3927     }
3928 
3929     qdict = qemu_opts_to_qdict(opts, NULL);
3930 
3931     if (!qdict_get_try_str(qdict, "node-name")) {
3932         QDECREF(qdict);
3933         error_report("'node-name' needs to be specified");
3934         goto out;
3935     }
3936 
3937     BlockDriverState *bs = bds_tree_init(qdict, &local_err);
3938     if (!bs) {
3939         error_report_err(local_err);
3940         goto out;
3941     }
3942 
3943     QTAILQ_INSERT_TAIL(&monitor_bdrv_states, bs, monitor_list);
3944 
3945 out:
3946     qemu_opts_del(opts);
3947 }
3948 
3949 void qmp_blockdev_add(BlockdevOptions *options, Error **errp)
3950 {
3951     BlockDriverState *bs;
3952     QObject *obj;
3953     Visitor *v = qobject_output_visitor_new(&obj);
3954     QDict *qdict;
3955     const QDictEntry *ent;
3956     Error *local_err = NULL;
3957 
3958     visit_type_BlockdevOptions(v, NULL, &options, &local_err);
3959     if (local_err) {
3960         error_propagate(errp, local_err);
3961         goto fail;
3962     }
3963 
3964     visit_complete(v, &obj);
3965     qdict = qobject_to_qdict(obj);
3966 
3967     qdict_flatten(qdict);
3968 
3969     /*
3970      * Rewrite "backing": null to "backing": ""
3971      * TODO Rewrite "" to null instead, and perhaps not even here
3972      */
3973     for (ent = qdict_first(qdict); ent; ent = qdict_next(qdict, ent)) {
3974         char *dot = strrchr(ent->key, '.');
3975 
3976         if (!strcmp(dot ? dot + 1 : ent->key, "backing")
3977             && qobject_type(ent->value) == QTYPE_QNULL) {
3978             qdict_put(qdict, ent->key, qstring_new());
3979         }
3980     }
3981 
3982     if (!qdict_get_try_str(qdict, "node-name")) {
3983         error_setg(errp, "'node-name' must be specified for the root node");
3984         goto fail;
3985     }
3986 
3987     bs = bds_tree_init(qdict, errp);
3988     if (!bs) {
3989         goto fail;
3990     }
3991 
3992     QTAILQ_INSERT_TAIL(&monitor_bdrv_states, bs, monitor_list);
3993 
3994 fail:
3995     visit_free(v);
3996 }
3997 
3998 void qmp_blockdev_del(const char *node_name, Error **errp)
3999 {
4000     AioContext *aio_context;
4001     BlockDriverState *bs;
4002 
4003     bs = bdrv_find_node(node_name);
4004     if (!bs) {
4005         error_setg(errp, "Cannot find node %s", node_name);
4006         return;
4007     }
4008     if (bdrv_has_blk(bs)) {
4009         error_setg(errp, "Node %s is in use", node_name);
4010         return;
4011     }
4012     aio_context = bdrv_get_aio_context(bs);
4013     aio_context_acquire(aio_context);
4014 
4015     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_DRIVE_DEL, errp)) {
4016         goto out;
4017     }
4018 
4019     if (!bs->monitor_list.tqe_prev) {
4020         error_setg(errp, "Node %s is not owned by the monitor",
4021                    bs->node_name);
4022         goto out;
4023     }
4024 
4025     if (bs->refcnt > 1) {
4026         error_setg(errp, "Block device %s is in use",
4027                    bdrv_get_device_or_node_name(bs));
4028         goto out;
4029     }
4030 
4031     QTAILQ_REMOVE(&monitor_bdrv_states, bs, monitor_list);
4032     bdrv_unref(bs);
4033 
4034 out:
4035     aio_context_release(aio_context);
4036 }
4037 
4038 static BdrvChild *bdrv_find_child(BlockDriverState *parent_bs,
4039                                   const char *child_name)
4040 {
4041     BdrvChild *child;
4042 
4043     QLIST_FOREACH(child, &parent_bs->children, next) {
4044         if (strcmp(child->name, child_name) == 0) {
4045             return child;
4046         }
4047     }
4048 
4049     return NULL;
4050 }
4051 
4052 void qmp_x_blockdev_change(const char *parent, bool has_child,
4053                            const char *child, bool has_node,
4054                            const char *node, Error **errp)
4055 {
4056     BlockDriverState *parent_bs, *new_bs = NULL;
4057     BdrvChild *p_child;
4058 
4059     parent_bs = bdrv_lookup_bs(parent, parent, errp);
4060     if (!parent_bs) {
4061         return;
4062     }
4063 
4064     if (has_child == has_node) {
4065         if (has_child) {
4066             error_setg(errp, "The parameters child and node are in conflict");
4067         } else {
4068             error_setg(errp, "Either child or node must be specified");
4069         }
4070         return;
4071     }
4072 
4073     if (has_child) {
4074         p_child = bdrv_find_child(parent_bs, child);
4075         if (!p_child) {
4076             error_setg(errp, "Node '%s' does not have child '%s'",
4077                        parent, child);
4078             return;
4079         }
4080         bdrv_del_child(parent_bs, p_child, errp);
4081     }
4082 
4083     if (has_node) {
4084         new_bs = bdrv_find_node(node);
4085         if (!new_bs) {
4086             error_setg(errp, "Node '%s' not found", node);
4087             return;
4088         }
4089         bdrv_add_child(parent_bs, new_bs, errp);
4090     }
4091 }
4092 
4093 BlockJobInfoList *qmp_query_block_jobs(Error **errp)
4094 {
4095     BlockJobInfoList *head = NULL, **p_next = &head;
4096     BlockJob *job;
4097 
4098     for (job = block_job_next(NULL); job; job = block_job_next(job)) {
4099         BlockJobInfoList *elem;
4100         AioContext *aio_context;
4101 
4102         if (block_job_is_internal(job)) {
4103             continue;
4104         }
4105         elem = g_new0(BlockJobInfoList, 1);
4106         aio_context = blk_get_aio_context(job->blk);
4107         aio_context_acquire(aio_context);
4108         elem->value = block_job_query(job, errp);
4109         aio_context_release(aio_context);
4110         if (!elem->value) {
4111             g_free(elem);
4112             qapi_free_BlockJobInfoList(head);
4113             return NULL;
4114         }
4115         *p_next = elem;
4116         p_next = &elem->next;
4117     }
4118 
4119     return head;
4120 }
4121 
4122 void qmp_x_blockdev_set_iothread(const char *node_name, StrOrNull *iothread,
4123                                  bool has_force, bool force, Error **errp)
4124 {
4125     AioContext *old_context;
4126     AioContext *new_context;
4127     BlockDriverState *bs;
4128 
4129     bs = bdrv_find_node(node_name);
4130     if (!bs) {
4131         error_setg(errp, "Cannot find node %s", node_name);
4132         return;
4133     }
4134 
4135     /* Protects against accidents. */
4136     if (!(has_force && force) && bdrv_has_blk(bs)) {
4137         error_setg(errp, "Node %s is associated with a BlockBackend and could "
4138                          "be in use (use force=true to override this check)",
4139                          node_name);
4140         return;
4141     }
4142 
4143     if (iothread->type == QTYPE_QSTRING) {
4144         IOThread *obj = iothread_by_id(iothread->u.s);
4145         if (!obj) {
4146             error_setg(errp, "Cannot find iothread %s", iothread->u.s);
4147             return;
4148         }
4149 
4150         new_context = iothread_get_aio_context(obj);
4151     } else {
4152         new_context = qemu_get_aio_context();
4153     }
4154 
4155     old_context = bdrv_get_aio_context(bs);
4156     aio_context_acquire(old_context);
4157 
4158     bdrv_set_aio_context(bs, new_context);
4159 
4160     aio_context_release(old_context);
4161 }
4162 
4163 QemuOptsList qemu_common_drive_opts = {
4164     .name = "drive",
4165     .head = QTAILQ_HEAD_INITIALIZER(qemu_common_drive_opts.head),
4166     .desc = {
4167         {
4168             .name = "snapshot",
4169             .type = QEMU_OPT_BOOL,
4170             .help = "enable/disable snapshot mode",
4171         },{
4172             .name = "aio",
4173             .type = QEMU_OPT_STRING,
4174             .help = "host AIO implementation (threads, native)",
4175         },{
4176             .name = BDRV_OPT_CACHE_WB,
4177             .type = QEMU_OPT_BOOL,
4178             .help = "Enable writeback mode",
4179         },{
4180             .name = "format",
4181             .type = QEMU_OPT_STRING,
4182             .help = "disk format (raw, qcow2, ...)",
4183         },{
4184             .name = "rerror",
4185             .type = QEMU_OPT_STRING,
4186             .help = "read error action",
4187         },{
4188             .name = "werror",
4189             .type = QEMU_OPT_STRING,
4190             .help = "write error action",
4191         },{
4192             .name = BDRV_OPT_READ_ONLY,
4193             .type = QEMU_OPT_BOOL,
4194             .help = "open drive file as read-only",
4195         },
4196 
4197         THROTTLE_OPTS,
4198 
4199         {
4200             .name = "throttling.group",
4201             .type = QEMU_OPT_STRING,
4202             .help = "name of the block throttling group",
4203         },{
4204             .name = "copy-on-read",
4205             .type = QEMU_OPT_BOOL,
4206             .help = "copy read data from backing file into image file",
4207         },{
4208             .name = "detect-zeroes",
4209             .type = QEMU_OPT_STRING,
4210             .help = "try to optimize zero writes (off, on, unmap)",
4211         },{
4212             .name = "stats-account-invalid",
4213             .type = QEMU_OPT_BOOL,
4214             .help = "whether to account for invalid I/O operations "
4215                     "in the statistics",
4216         },{
4217             .name = "stats-account-failed",
4218             .type = QEMU_OPT_BOOL,
4219             .help = "whether to account for failed I/O operations "
4220                     "in the statistics",
4221         },
4222         { /* end of list */ }
4223     },
4224 };
4225 
4226 QemuOptsList qemu_drive_opts = {
4227     .name = "drive",
4228     .head = QTAILQ_HEAD_INITIALIZER(qemu_drive_opts.head),
4229     .desc = {
4230         /*
4231          * no elements => accept any params
4232          * validation will happen later
4233          */
4234         { /* end of list */ }
4235     },
4236 };
4237