xref: /openbmc/qemu/blockdev.c (revision 80adf54e)
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 "qapi/util.h"
48 #include "sysemu/sysemu.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__MAX,
444                             BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF,
445                             &local_error);
446         if (local_error) {
447             error_propagate(errp, local_error);
448             return;
449         }
450     }
451 }
452 
453 /* Takes the ownership of bs_opts */
454 static BlockBackend *blockdev_init(const char *file, QDict *bs_opts,
455                                    Error **errp)
456 {
457     const char *buf;
458     int bdrv_flags = 0;
459     int on_read_error, on_write_error;
460     bool account_invalid, account_failed;
461     bool writethrough, read_only;
462     BlockBackend *blk;
463     BlockDriverState *bs;
464     ThrottleConfig cfg;
465     int snapshot = 0;
466     Error *error = NULL;
467     QemuOpts *opts;
468     QDict *interval_dict = NULL;
469     QList *interval_list = NULL;
470     const char *id;
471     BlockdevDetectZeroesOptions detect_zeroes =
472         BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF;
473     const char *throttling_group = NULL;
474 
475     /* Check common options by copying from bs_opts to opts, all other options
476      * stay in bs_opts for processing by bdrv_open(). */
477     id = qdict_get_try_str(bs_opts, "id");
478     opts = qemu_opts_create(&qemu_common_drive_opts, id, 1, &error);
479     if (error) {
480         error_propagate(errp, error);
481         goto err_no_opts;
482     }
483 
484     qemu_opts_absorb_qdict(opts, bs_opts, &error);
485     if (error) {
486         error_propagate(errp, error);
487         goto early_err;
488     }
489 
490     if (id) {
491         qdict_del(bs_opts, "id");
492     }
493 
494     /* extract parameters */
495     snapshot = qemu_opt_get_bool(opts, "snapshot", 0);
496 
497     account_invalid = qemu_opt_get_bool(opts, "stats-account-invalid", true);
498     account_failed = qemu_opt_get_bool(opts, "stats-account-failed", true);
499 
500     writethrough = !qemu_opt_get_bool(opts, BDRV_OPT_CACHE_WB, true);
501 
502     id = qemu_opts_id(opts);
503 
504     qdict_extract_subqdict(bs_opts, &interval_dict, "stats-intervals.");
505     qdict_array_split(interval_dict, &interval_list);
506 
507     if (qdict_size(interval_dict) != 0) {
508         error_setg(errp, "Invalid option stats-intervals.%s",
509                    qdict_first(interval_dict)->key);
510         goto early_err;
511     }
512 
513     extract_common_blockdev_options(opts, &bdrv_flags, &throttling_group, &cfg,
514                                     &detect_zeroes, &error);
515     if (error) {
516         error_propagate(errp, error);
517         goto early_err;
518     }
519 
520     if ((buf = qemu_opt_get(opts, "format")) != NULL) {
521         if (is_help_option(buf)) {
522             error_printf("Supported formats:");
523             bdrv_iterate_format(bdrv_format_print, NULL);
524             error_printf("\n");
525             goto early_err;
526         }
527 
528         if (qdict_haskey(bs_opts, "driver")) {
529             error_setg(errp, "Cannot specify both 'driver' and 'format'");
530             goto early_err;
531         }
532         qdict_put_str(bs_opts, "driver", buf);
533     }
534 
535     on_write_error = BLOCKDEV_ON_ERROR_ENOSPC;
536     if ((buf = qemu_opt_get(opts, "werror")) != NULL) {
537         on_write_error = parse_block_error_action(buf, 0, &error);
538         if (error) {
539             error_propagate(errp, error);
540             goto early_err;
541         }
542     }
543 
544     on_read_error = BLOCKDEV_ON_ERROR_REPORT;
545     if ((buf = qemu_opt_get(opts, "rerror")) != NULL) {
546         on_read_error = parse_block_error_action(buf, 1, &error);
547         if (error) {
548             error_propagate(errp, error);
549             goto early_err;
550         }
551     }
552 
553     if (snapshot) {
554         bdrv_flags |= BDRV_O_SNAPSHOT;
555     }
556 
557     read_only = qemu_opt_get_bool(opts, BDRV_OPT_READ_ONLY, false);
558 
559     /* init */
560     if ((!file || !*file) && !qdict_size(bs_opts)) {
561         BlockBackendRootState *blk_rs;
562 
563         blk = blk_new(0, BLK_PERM_ALL);
564         blk_rs = blk_get_root_state(blk);
565         blk_rs->open_flags    = bdrv_flags;
566         blk_rs->read_only     = read_only;
567         blk_rs->detect_zeroes = detect_zeroes;
568 
569         QDECREF(bs_opts);
570     } else {
571         if (file && !*file) {
572             file = NULL;
573         }
574 
575         /* bdrv_open() defaults to the values in bdrv_flags (for compatibility
576          * with other callers) rather than what we want as the real defaults.
577          * Apply the defaults here instead. */
578         qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_DIRECT, "off");
579         qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_NO_FLUSH, "off");
580         qdict_set_default_str(bs_opts, BDRV_OPT_READ_ONLY,
581                               read_only ? "on" : "off");
582         assert((bdrv_flags & BDRV_O_CACHE_MASK) == 0);
583 
584         if (runstate_check(RUN_STATE_INMIGRATE)) {
585             bdrv_flags |= BDRV_O_INACTIVE;
586         }
587 
588         blk = blk_new_open(file, NULL, bs_opts, bdrv_flags, errp);
589         if (!blk) {
590             goto err_no_bs_opts;
591         }
592         bs = blk_bs(blk);
593 
594         bs->detect_zeroes = detect_zeroes;
595 
596         block_acct_setup(blk_get_stats(blk), account_invalid, account_failed);
597 
598         if (!parse_stats_intervals(blk_get_stats(blk), interval_list, errp)) {
599             blk_unref(blk);
600             blk = NULL;
601             goto err_no_bs_opts;
602         }
603     }
604 
605     /* disk I/O throttling */
606     if (throttle_enabled(&cfg)) {
607         if (!throttling_group) {
608             throttling_group = id;
609         }
610         blk_io_limits_enable(blk, throttling_group);
611         blk_set_io_limits(blk, &cfg);
612     }
613 
614     blk_set_enable_write_cache(blk, !writethrough);
615     blk_set_on_error(blk, on_read_error, on_write_error);
616 
617     if (!monitor_add_blk(blk, id, errp)) {
618         blk_unref(blk);
619         blk = NULL;
620         goto err_no_bs_opts;
621     }
622 
623 err_no_bs_opts:
624     qemu_opts_del(opts);
625     QDECREF(interval_dict);
626     QDECREF(interval_list);
627     return blk;
628 
629 early_err:
630     qemu_opts_del(opts);
631     QDECREF(interval_dict);
632     QDECREF(interval_list);
633 err_no_opts:
634     QDECREF(bs_opts);
635     return NULL;
636 }
637 
638 /* Takes the ownership of bs_opts */
639 static BlockDriverState *bds_tree_init(QDict *bs_opts, Error **errp)
640 {
641     int bdrv_flags = 0;
642 
643     /* bdrv_open() defaults to the values in bdrv_flags (for compatibility
644      * with other callers) rather than what we want as the real defaults.
645      * Apply the defaults here instead. */
646     qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_DIRECT, "off");
647     qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_NO_FLUSH, "off");
648     qdict_set_default_str(bs_opts, BDRV_OPT_READ_ONLY, "off");
649 
650     if (runstate_check(RUN_STATE_INMIGRATE)) {
651         bdrv_flags |= BDRV_O_INACTIVE;
652     }
653 
654     return bdrv_open(NULL, NULL, bs_opts, bdrv_flags, errp);
655 }
656 
657 void blockdev_close_all_bdrv_states(void)
658 {
659     BlockDriverState *bs, *next_bs;
660 
661     QTAILQ_FOREACH_SAFE(bs, &monitor_bdrv_states, monitor_list, next_bs) {
662         AioContext *ctx = bdrv_get_aio_context(bs);
663 
664         aio_context_acquire(ctx);
665         bdrv_unref(bs);
666         aio_context_release(ctx);
667     }
668 }
669 
670 /* Iterates over the list of monitor-owned BlockDriverStates */
671 BlockDriverState *bdrv_next_monitor_owned(BlockDriverState *bs)
672 {
673     return bs ? QTAILQ_NEXT(bs, monitor_list)
674               : QTAILQ_FIRST(&monitor_bdrv_states);
675 }
676 
677 static void qemu_opt_rename(QemuOpts *opts, const char *from, const char *to,
678                             Error **errp)
679 {
680     const char *value;
681 
682     value = qemu_opt_get(opts, from);
683     if (value) {
684         if (qemu_opt_find(opts, to)) {
685             error_setg(errp, "'%s' and its alias '%s' can't be used at the "
686                        "same time", to, from);
687             return;
688         }
689     }
690 
691     /* rename all items in opts */
692     while ((value = qemu_opt_get(opts, from))) {
693         qemu_opt_set(opts, to, value, &error_abort);
694         qemu_opt_unset(opts, from);
695     }
696 }
697 
698 QemuOptsList qemu_legacy_drive_opts = {
699     .name = "drive",
700     .head = QTAILQ_HEAD_INITIALIZER(qemu_legacy_drive_opts.head),
701     .desc = {
702         {
703             .name = "bus",
704             .type = QEMU_OPT_NUMBER,
705             .help = "bus number",
706         },{
707             .name = "unit",
708             .type = QEMU_OPT_NUMBER,
709             .help = "unit number (i.e. lun for scsi)",
710         },{
711             .name = "index",
712             .type = QEMU_OPT_NUMBER,
713             .help = "index number",
714         },{
715             .name = "media",
716             .type = QEMU_OPT_STRING,
717             .help = "media type (disk, cdrom)",
718         },{
719             .name = "if",
720             .type = QEMU_OPT_STRING,
721             .help = "interface (ide, scsi, sd, mtd, floppy, pflash, virtio)",
722         },{
723             .name = "cyls",
724             .type = QEMU_OPT_NUMBER,
725             .help = "number of cylinders (ide disk geometry)",
726         },{
727             .name = "heads",
728             .type = QEMU_OPT_NUMBER,
729             .help = "number of heads (ide disk geometry)",
730         },{
731             .name = "secs",
732             .type = QEMU_OPT_NUMBER,
733             .help = "number of sectors (ide disk geometry)",
734         },{
735             .name = "trans",
736             .type = QEMU_OPT_STRING,
737             .help = "chs translation (auto, lba, none)",
738         },{
739             .name = "boot",
740             .type = QEMU_OPT_BOOL,
741             .help = "(deprecated, ignored)",
742         },{
743             .name = "addr",
744             .type = QEMU_OPT_STRING,
745             .help = "pci address (virtio only)",
746         },{
747             .name = "serial",
748             .type = QEMU_OPT_STRING,
749             .help = "disk serial number",
750         },{
751             .name = "file",
752             .type = QEMU_OPT_STRING,
753             .help = "file name",
754         },
755 
756         /* Options that are passed on, but have special semantics with -drive */
757         {
758             .name = BDRV_OPT_READ_ONLY,
759             .type = QEMU_OPT_BOOL,
760             .help = "open drive file as read-only",
761         },{
762             .name = "rerror",
763             .type = QEMU_OPT_STRING,
764             .help = "read error action",
765         },{
766             .name = "werror",
767             .type = QEMU_OPT_STRING,
768             .help = "write error action",
769         },{
770             .name = "copy-on-read",
771             .type = QEMU_OPT_BOOL,
772             .help = "copy read data from backing file into image file",
773         },
774 
775         { /* end of list */ }
776     },
777 };
778 
779 DriveInfo *drive_new(QemuOpts *all_opts, BlockInterfaceType block_default_type)
780 {
781     const char *value;
782     BlockBackend *blk;
783     DriveInfo *dinfo = NULL;
784     QDict *bs_opts;
785     QemuOpts *legacy_opts;
786     DriveMediaType media = MEDIA_DISK;
787     BlockInterfaceType type;
788     int cyls, heads, secs, translation;
789     int max_devs, bus_id, unit_id, index;
790     const char *devaddr;
791     const char *werror, *rerror;
792     bool read_only = false;
793     bool copy_on_read;
794     const char *serial;
795     const char *filename;
796     Error *local_err = NULL;
797     int i;
798     const char *deprecated[] = {
799         "serial", "trans", "secs", "heads", "cyls", "addr"
800     };
801 
802     /* Change legacy command line options into QMP ones */
803     static const struct {
804         const char *from;
805         const char *to;
806     } opt_renames[] = {
807         { "iops",           "throttling.iops-total" },
808         { "iops_rd",        "throttling.iops-read" },
809         { "iops_wr",        "throttling.iops-write" },
810 
811         { "bps",            "throttling.bps-total" },
812         { "bps_rd",         "throttling.bps-read" },
813         { "bps_wr",         "throttling.bps-write" },
814 
815         { "iops_max",       "throttling.iops-total-max" },
816         { "iops_rd_max",    "throttling.iops-read-max" },
817         { "iops_wr_max",    "throttling.iops-write-max" },
818 
819         { "bps_max",        "throttling.bps-total-max" },
820         { "bps_rd_max",     "throttling.bps-read-max" },
821         { "bps_wr_max",     "throttling.bps-write-max" },
822 
823         { "iops_size",      "throttling.iops-size" },
824 
825         { "group",          "throttling.group" },
826 
827         { "readonly",       BDRV_OPT_READ_ONLY },
828     };
829 
830     for (i = 0; i < ARRAY_SIZE(opt_renames); i++) {
831         qemu_opt_rename(all_opts, opt_renames[i].from, opt_renames[i].to,
832                         &local_err);
833         if (local_err) {
834             error_report_err(local_err);
835             return NULL;
836         }
837     }
838 
839     value = qemu_opt_get(all_opts, "cache");
840     if (value) {
841         int flags = 0;
842         bool writethrough;
843 
844         if (bdrv_parse_cache_mode(value, &flags, &writethrough) != 0) {
845             error_report("invalid cache option");
846             return NULL;
847         }
848 
849         /* Specific options take precedence */
850         if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_WB)) {
851             qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_WB,
852                               !writethrough, &error_abort);
853         }
854         if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_DIRECT)) {
855             qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_DIRECT,
856                               !!(flags & BDRV_O_NOCACHE), &error_abort);
857         }
858         if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_NO_FLUSH)) {
859             qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_NO_FLUSH,
860                               !!(flags & BDRV_O_NO_FLUSH), &error_abort);
861         }
862         qemu_opt_unset(all_opts, "cache");
863     }
864 
865     /* Get a QDict for processing the options */
866     bs_opts = qdict_new();
867     qemu_opts_to_qdict(all_opts, bs_opts);
868 
869     legacy_opts = qemu_opts_create(&qemu_legacy_drive_opts, NULL, 0,
870                                    &error_abort);
871     qemu_opts_absorb_qdict(legacy_opts, bs_opts, &local_err);
872     if (local_err) {
873         error_report_err(local_err);
874         goto fail;
875     }
876 
877     /* Deprecated option boot=[on|off] */
878     if (qemu_opt_get(legacy_opts, "boot") != NULL) {
879         fprintf(stderr, "qemu-kvm: boot=on|off is deprecated and will be "
880                 "ignored. Future versions will reject this parameter. Please "
881                 "update your scripts.\n");
882     }
883 
884     /* Other deprecated options */
885     if (!qtest_enabled()) {
886         for (i = 0; i < ARRAY_SIZE(deprecated); i++) {
887             if (qemu_opt_get(legacy_opts, deprecated[i]) != NULL) {
888                 error_report("'%s' is deprecated, please use the corresponding "
889                              "option of '-device' instead", deprecated[i]);
890             }
891         }
892     }
893 
894     /* Media type */
895     value = qemu_opt_get(legacy_opts, "media");
896     if (value) {
897         if (!strcmp(value, "disk")) {
898             media = MEDIA_DISK;
899         } else if (!strcmp(value, "cdrom")) {
900             media = MEDIA_CDROM;
901             read_only = true;
902         } else {
903             error_report("'%s' invalid media", value);
904             goto fail;
905         }
906     }
907 
908     /* copy-on-read is disabled with a warning for read-only devices */
909     read_only |= qemu_opt_get_bool(legacy_opts, BDRV_OPT_READ_ONLY, false);
910     copy_on_read = qemu_opt_get_bool(legacy_opts, "copy-on-read", false);
911 
912     if (read_only && copy_on_read) {
913         warn_report("disabling copy-on-read on read-only drive");
914         copy_on_read = false;
915     }
916 
917     qdict_put_str(bs_opts, BDRV_OPT_READ_ONLY, read_only ? "on" : "off");
918     qdict_put_str(bs_opts, "copy-on-read", copy_on_read ? "on" : "off");
919 
920     /* Controller type */
921     value = qemu_opt_get(legacy_opts, "if");
922     if (value) {
923         for (type = 0;
924              type < IF_COUNT && strcmp(value, if_name[type]);
925              type++) {
926         }
927         if (type == IF_COUNT) {
928             error_report("unsupported bus type '%s'", value);
929             goto fail;
930         }
931     } else {
932         type = block_default_type;
933     }
934 
935     /* Geometry */
936     cyls  = qemu_opt_get_number(legacy_opts, "cyls", 0);
937     heads = qemu_opt_get_number(legacy_opts, "heads", 0);
938     secs  = qemu_opt_get_number(legacy_opts, "secs", 0);
939 
940     if (cyls || heads || secs) {
941         if (cyls < 1) {
942             error_report("invalid physical cyls number");
943             goto fail;
944         }
945         if (heads < 1) {
946             error_report("invalid physical heads number");
947             goto fail;
948         }
949         if (secs < 1) {
950             error_report("invalid physical secs number");
951             goto fail;
952         }
953     }
954 
955     translation = BIOS_ATA_TRANSLATION_AUTO;
956     value = qemu_opt_get(legacy_opts, "trans");
957     if (value != NULL) {
958         if (!cyls) {
959             error_report("'%s' trans must be used with cyls, heads and secs",
960                          value);
961             goto fail;
962         }
963         if (!strcmp(value, "none")) {
964             translation = BIOS_ATA_TRANSLATION_NONE;
965         } else if (!strcmp(value, "lba")) {
966             translation = BIOS_ATA_TRANSLATION_LBA;
967         } else if (!strcmp(value, "large")) {
968             translation = BIOS_ATA_TRANSLATION_LARGE;
969         } else if (!strcmp(value, "rechs")) {
970             translation = BIOS_ATA_TRANSLATION_RECHS;
971         } else if (!strcmp(value, "auto")) {
972             translation = BIOS_ATA_TRANSLATION_AUTO;
973         } else {
974             error_report("'%s' invalid translation type", value);
975             goto fail;
976         }
977     }
978 
979     if (media == MEDIA_CDROM) {
980         if (cyls || secs || heads) {
981             error_report("CHS can't be set with media=cdrom");
982             goto fail;
983         }
984     }
985 
986     /* Device address specified by bus/unit or index.
987      * If none was specified, try to find the first free one. */
988     bus_id  = qemu_opt_get_number(legacy_opts, "bus", 0);
989     unit_id = qemu_opt_get_number(legacy_opts, "unit", -1);
990     index   = qemu_opt_get_number(legacy_opts, "index", -1);
991 
992     max_devs = if_max_devs[type];
993 
994     if (index != -1) {
995         if (bus_id != 0 || unit_id != -1) {
996             error_report("index cannot be used with bus and unit");
997             goto fail;
998         }
999         bus_id = drive_index_to_bus_id(type, index);
1000         unit_id = drive_index_to_unit_id(type, index);
1001     }
1002 
1003     if (unit_id == -1) {
1004        unit_id = 0;
1005        while (drive_get(type, bus_id, unit_id) != NULL) {
1006            unit_id++;
1007            if (max_devs && unit_id >= max_devs) {
1008                unit_id -= max_devs;
1009                bus_id++;
1010            }
1011        }
1012     }
1013 
1014     if (max_devs && unit_id >= max_devs) {
1015         error_report("unit %d too big (max is %d)", unit_id, max_devs - 1);
1016         goto fail;
1017     }
1018 
1019     if (drive_get(type, bus_id, unit_id) != NULL) {
1020         error_report("drive with bus=%d, unit=%d (index=%d) exists",
1021                      bus_id, unit_id, index);
1022         goto fail;
1023     }
1024 
1025     /* Serial number */
1026     serial = qemu_opt_get(legacy_opts, "serial");
1027 
1028     /* no id supplied -> create one */
1029     if (qemu_opts_id(all_opts) == NULL) {
1030         char *new_id;
1031         const char *mediastr = "";
1032         if (type == IF_IDE || type == IF_SCSI) {
1033             mediastr = (media == MEDIA_CDROM) ? "-cd" : "-hd";
1034         }
1035         if (max_devs) {
1036             new_id = g_strdup_printf("%s%i%s%i", if_name[type], bus_id,
1037                                      mediastr, unit_id);
1038         } else {
1039             new_id = g_strdup_printf("%s%s%i", if_name[type],
1040                                      mediastr, unit_id);
1041         }
1042         qdict_put_str(bs_opts, "id", new_id);
1043         g_free(new_id);
1044     }
1045 
1046     /* Add virtio block device */
1047     devaddr = qemu_opt_get(legacy_opts, "addr");
1048     if (devaddr && type != IF_VIRTIO) {
1049         error_report("addr is not supported by this bus type");
1050         goto fail;
1051     }
1052 
1053     if (type == IF_VIRTIO) {
1054         QemuOpts *devopts;
1055         devopts = qemu_opts_create(qemu_find_opts("device"), NULL, 0,
1056                                    &error_abort);
1057         if (arch_type == QEMU_ARCH_S390X) {
1058             qemu_opt_set(devopts, "driver", "virtio-blk-ccw", &error_abort);
1059         } else {
1060             qemu_opt_set(devopts, "driver", "virtio-blk-pci", &error_abort);
1061         }
1062         qemu_opt_set(devopts, "drive", qdict_get_str(bs_opts, "id"),
1063                      &error_abort);
1064         if (devaddr) {
1065             qemu_opt_set(devopts, "addr", devaddr, &error_abort);
1066         }
1067     }
1068 
1069     filename = qemu_opt_get(legacy_opts, "file");
1070 
1071     /* Check werror/rerror compatibility with if=... */
1072     werror = qemu_opt_get(legacy_opts, "werror");
1073     if (werror != NULL) {
1074         if (type != IF_IDE && type != IF_SCSI && type != IF_VIRTIO &&
1075             type != IF_NONE) {
1076             error_report("werror is not supported by this bus type");
1077             goto fail;
1078         }
1079         qdict_put_str(bs_opts, "werror", werror);
1080     }
1081 
1082     rerror = qemu_opt_get(legacy_opts, "rerror");
1083     if (rerror != NULL) {
1084         if (type != IF_IDE && type != IF_VIRTIO && type != IF_SCSI &&
1085             type != IF_NONE) {
1086             error_report("rerror is not supported by this bus type");
1087             goto fail;
1088         }
1089         qdict_put_str(bs_opts, "rerror", rerror);
1090     }
1091 
1092     /* Actual block device init: Functionality shared with blockdev-add */
1093     blk = blockdev_init(filename, bs_opts, &local_err);
1094     bs_opts = NULL;
1095     if (!blk) {
1096         if (local_err) {
1097             error_report_err(local_err);
1098         }
1099         goto fail;
1100     } else {
1101         assert(!local_err);
1102     }
1103 
1104     /* Create legacy DriveInfo */
1105     dinfo = g_malloc0(sizeof(*dinfo));
1106     dinfo->opts = all_opts;
1107 
1108     dinfo->cyls = cyls;
1109     dinfo->heads = heads;
1110     dinfo->secs = secs;
1111     dinfo->trans = translation;
1112 
1113     dinfo->type = type;
1114     dinfo->bus = bus_id;
1115     dinfo->unit = unit_id;
1116     dinfo->devaddr = devaddr;
1117     dinfo->serial = g_strdup(serial);
1118 
1119     blk_set_legacy_dinfo(blk, dinfo);
1120 
1121     switch(type) {
1122     case IF_IDE:
1123     case IF_SCSI:
1124     case IF_XEN:
1125     case IF_NONE:
1126         dinfo->media_cd = media == MEDIA_CDROM;
1127         break;
1128     default:
1129         break;
1130     }
1131 
1132 fail:
1133     qemu_opts_del(legacy_opts);
1134     QDECREF(bs_opts);
1135     return dinfo;
1136 }
1137 
1138 static BlockDriverState *qmp_get_root_bs(const char *name, Error **errp)
1139 {
1140     BlockDriverState *bs;
1141 
1142     bs = bdrv_lookup_bs(name, name, errp);
1143     if (bs == NULL) {
1144         return NULL;
1145     }
1146 
1147     if (!bdrv_is_root_node(bs)) {
1148         error_setg(errp, "Need a root block node");
1149         return NULL;
1150     }
1151 
1152     if (!bdrv_is_inserted(bs)) {
1153         error_setg(errp, "Device has no medium");
1154         return NULL;
1155     }
1156 
1157     return bs;
1158 }
1159 
1160 static BlockBackend *qmp_get_blk(const char *blk_name, const char *qdev_id,
1161                                  Error **errp)
1162 {
1163     BlockBackend *blk;
1164 
1165     if (!blk_name == !qdev_id) {
1166         error_setg(errp, "Need exactly one of 'device' and 'id'");
1167         return NULL;
1168     }
1169 
1170     if (qdev_id) {
1171         blk = blk_by_qdev_id(qdev_id, errp);
1172     } else {
1173         blk = blk_by_name(blk_name);
1174         if (blk == NULL) {
1175             error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
1176                       "Device '%s' not found", blk_name);
1177         }
1178     }
1179 
1180     return blk;
1181 }
1182 
1183 void hmp_commit(Monitor *mon, const QDict *qdict)
1184 {
1185     const char *device = qdict_get_str(qdict, "device");
1186     BlockBackend *blk;
1187     int ret;
1188 
1189     if (!strcmp(device, "all")) {
1190         ret = blk_commit_all();
1191     } else {
1192         BlockDriverState *bs;
1193         AioContext *aio_context;
1194 
1195         blk = blk_by_name(device);
1196         if (!blk) {
1197             monitor_printf(mon, "Device '%s' not found\n", device);
1198             return;
1199         }
1200         if (!blk_is_available(blk)) {
1201             monitor_printf(mon, "Device '%s' has no medium\n", device);
1202             return;
1203         }
1204 
1205         bs = blk_bs(blk);
1206         aio_context = bdrv_get_aio_context(bs);
1207         aio_context_acquire(aio_context);
1208 
1209         ret = bdrv_commit(bs);
1210 
1211         aio_context_release(aio_context);
1212     }
1213     if (ret < 0) {
1214         monitor_printf(mon, "'commit' error for '%s': %s\n", device,
1215                        strerror(-ret));
1216     }
1217 }
1218 
1219 static void blockdev_do_action(TransactionAction *action, Error **errp)
1220 {
1221     TransactionActionList list;
1222 
1223     list.value = action;
1224     list.next = NULL;
1225     qmp_transaction(&list, false, NULL, errp);
1226 }
1227 
1228 void qmp_blockdev_snapshot_sync(bool has_device, const char *device,
1229                                 bool has_node_name, const char *node_name,
1230                                 const char *snapshot_file,
1231                                 bool has_snapshot_node_name,
1232                                 const char *snapshot_node_name,
1233                                 bool has_format, const char *format,
1234                                 bool has_mode, NewImageMode mode, Error **errp)
1235 {
1236     BlockdevSnapshotSync snapshot = {
1237         .has_device = has_device,
1238         .device = (char *) device,
1239         .has_node_name = has_node_name,
1240         .node_name = (char *) node_name,
1241         .snapshot_file = (char *) snapshot_file,
1242         .has_snapshot_node_name = has_snapshot_node_name,
1243         .snapshot_node_name = (char *) snapshot_node_name,
1244         .has_format = has_format,
1245         .format = (char *) format,
1246         .has_mode = has_mode,
1247         .mode = mode,
1248     };
1249     TransactionAction action = {
1250         .type = TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC,
1251         .u.blockdev_snapshot_sync.data = &snapshot,
1252     };
1253     blockdev_do_action(&action, errp);
1254 }
1255 
1256 void qmp_blockdev_snapshot(const char *node, const char *overlay,
1257                            Error **errp)
1258 {
1259     BlockdevSnapshot snapshot_data = {
1260         .node = (char *) node,
1261         .overlay = (char *) overlay
1262     };
1263     TransactionAction action = {
1264         .type = TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT,
1265         .u.blockdev_snapshot.data = &snapshot_data,
1266     };
1267     blockdev_do_action(&action, errp);
1268 }
1269 
1270 void qmp_blockdev_snapshot_internal_sync(const char *device,
1271                                          const char *name,
1272                                          Error **errp)
1273 {
1274     BlockdevSnapshotInternal snapshot = {
1275         .device = (char *) device,
1276         .name = (char *) name
1277     };
1278     TransactionAction action = {
1279         .type = TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC,
1280         .u.blockdev_snapshot_internal_sync.data = &snapshot,
1281     };
1282     blockdev_do_action(&action, errp);
1283 }
1284 
1285 SnapshotInfo *qmp_blockdev_snapshot_delete_internal_sync(const char *device,
1286                                                          bool has_id,
1287                                                          const char *id,
1288                                                          bool has_name,
1289                                                          const char *name,
1290                                                          Error **errp)
1291 {
1292     BlockDriverState *bs;
1293     AioContext *aio_context;
1294     QEMUSnapshotInfo sn;
1295     Error *local_err = NULL;
1296     SnapshotInfo *info = NULL;
1297     int ret;
1298 
1299     bs = qmp_get_root_bs(device, errp);
1300     if (!bs) {
1301         return NULL;
1302     }
1303     aio_context = bdrv_get_aio_context(bs);
1304     aio_context_acquire(aio_context);
1305 
1306     if (!has_id) {
1307         id = NULL;
1308     }
1309 
1310     if (!has_name) {
1311         name = NULL;
1312     }
1313 
1314     if (!id && !name) {
1315         error_setg(errp, "Name or id must be provided");
1316         goto out_aio_context;
1317     }
1318 
1319     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_INTERNAL_SNAPSHOT_DELETE, errp)) {
1320         goto out_aio_context;
1321     }
1322 
1323     ret = bdrv_snapshot_find_by_id_and_name(bs, id, name, &sn, &local_err);
1324     if (local_err) {
1325         error_propagate(errp, local_err);
1326         goto out_aio_context;
1327     }
1328     if (!ret) {
1329         error_setg(errp,
1330                    "Snapshot with id '%s' and name '%s' does not exist on "
1331                    "device '%s'",
1332                    STR_OR_NULL(id), STR_OR_NULL(name), device);
1333         goto out_aio_context;
1334     }
1335 
1336     bdrv_snapshot_delete(bs, id, name, &local_err);
1337     if (local_err) {
1338         error_propagate(errp, local_err);
1339         goto out_aio_context;
1340     }
1341 
1342     aio_context_release(aio_context);
1343 
1344     info = g_new0(SnapshotInfo, 1);
1345     info->id = g_strdup(sn.id_str);
1346     info->name = g_strdup(sn.name);
1347     info->date_nsec = sn.date_nsec;
1348     info->date_sec = sn.date_sec;
1349     info->vm_state_size = sn.vm_state_size;
1350     info->vm_clock_nsec = sn.vm_clock_nsec % 1000000000;
1351     info->vm_clock_sec = sn.vm_clock_nsec / 1000000000;
1352 
1353     return info;
1354 
1355 out_aio_context:
1356     aio_context_release(aio_context);
1357     return NULL;
1358 }
1359 
1360 /**
1361  * block_dirty_bitmap_lookup:
1362  * Return a dirty bitmap (if present), after validating
1363  * the node reference and bitmap names.
1364  *
1365  * @node: The name of the BDS node to search for bitmaps
1366  * @name: The name of the bitmap to search for
1367  * @pbs: Output pointer for BDS lookup, if desired. Can be NULL.
1368  * @paio: Output pointer for aio_context acquisition, if desired. Can be NULL.
1369  * @errp: Output pointer for error information. Can be NULL.
1370  *
1371  * @return: A bitmap object on success, or NULL on failure.
1372  */
1373 static BdrvDirtyBitmap *block_dirty_bitmap_lookup(const char *node,
1374                                                   const char *name,
1375                                                   BlockDriverState **pbs,
1376                                                   Error **errp)
1377 {
1378     BlockDriverState *bs;
1379     BdrvDirtyBitmap *bitmap;
1380 
1381     if (!node) {
1382         error_setg(errp, "Node cannot be NULL");
1383         return NULL;
1384     }
1385     if (!name) {
1386         error_setg(errp, "Bitmap name cannot be NULL");
1387         return NULL;
1388     }
1389     bs = bdrv_lookup_bs(node, node, NULL);
1390     if (!bs) {
1391         error_setg(errp, "Node '%s' not found", node);
1392         return NULL;
1393     }
1394 
1395     bitmap = bdrv_find_dirty_bitmap(bs, name);
1396     if (!bitmap) {
1397         error_setg(errp, "Dirty bitmap '%s' not found", name);
1398         return NULL;
1399     }
1400 
1401     if (pbs) {
1402         *pbs = bs;
1403     }
1404 
1405     return bitmap;
1406 }
1407 
1408 /* New and old BlockDriverState structs for atomic group operations */
1409 
1410 typedef struct BlkActionState BlkActionState;
1411 
1412 /**
1413  * BlkActionOps:
1414  * Table of operations that define an Action.
1415  *
1416  * @instance_size: Size of state struct, in bytes.
1417  * @prepare: Prepare the work, must NOT be NULL.
1418  * @commit: Commit the changes, can be NULL.
1419  * @abort: Abort the changes on fail, can be NULL.
1420  * @clean: Clean up resources after all transaction actions have called
1421  *         commit() or abort(). Can be NULL.
1422  *
1423  * Only prepare() may fail. In a single transaction, only one of commit() or
1424  * abort() will be called. clean() will always be called if it is present.
1425  */
1426 typedef struct BlkActionOps {
1427     size_t instance_size;
1428     void (*prepare)(BlkActionState *common, Error **errp);
1429     void (*commit)(BlkActionState *common);
1430     void (*abort)(BlkActionState *common);
1431     void (*clean)(BlkActionState *common);
1432 } BlkActionOps;
1433 
1434 /**
1435  * BlkActionState:
1436  * Describes one Action's state within a Transaction.
1437  *
1438  * @action: QAPI-defined enum identifying which Action to perform.
1439  * @ops: Table of ActionOps this Action can perform.
1440  * @block_job_txn: Transaction which this action belongs to.
1441  * @entry: List membership for all Actions in this Transaction.
1442  *
1443  * This structure must be arranged as first member in a subclassed type,
1444  * assuming that the compiler will also arrange it to the same offsets as the
1445  * base class.
1446  */
1447 struct BlkActionState {
1448     TransactionAction *action;
1449     const BlkActionOps *ops;
1450     BlockJobTxn *block_job_txn;
1451     TransactionProperties *txn_props;
1452     QSIMPLEQ_ENTRY(BlkActionState) entry;
1453 };
1454 
1455 /* internal snapshot private data */
1456 typedef struct InternalSnapshotState {
1457     BlkActionState common;
1458     BlockDriverState *bs;
1459     AioContext *aio_context;
1460     QEMUSnapshotInfo sn;
1461     bool created;
1462 } InternalSnapshotState;
1463 
1464 
1465 static int action_check_completion_mode(BlkActionState *s, Error **errp)
1466 {
1467     if (s->txn_props->completion_mode != ACTION_COMPLETION_MODE_INDIVIDUAL) {
1468         error_setg(errp,
1469                    "Action '%s' does not support Transaction property "
1470                    "completion-mode = %s",
1471                    TransactionActionKind_lookup[s->action->type],
1472                    ActionCompletionMode_lookup[s->txn_props->completion_mode]);
1473         return -1;
1474     }
1475     return 0;
1476 }
1477 
1478 static void internal_snapshot_prepare(BlkActionState *common,
1479                                       Error **errp)
1480 {
1481     Error *local_err = NULL;
1482     const char *device;
1483     const char *name;
1484     BlockDriverState *bs;
1485     QEMUSnapshotInfo old_sn, *sn;
1486     bool ret;
1487     qemu_timeval tv;
1488     BlockdevSnapshotInternal *internal;
1489     InternalSnapshotState *state;
1490     int ret1;
1491 
1492     g_assert(common->action->type ==
1493              TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC);
1494     internal = common->action->u.blockdev_snapshot_internal_sync.data;
1495     state = DO_UPCAST(InternalSnapshotState, common, common);
1496 
1497     /* 1. parse input */
1498     device = internal->device;
1499     name = internal->name;
1500 
1501     /* 2. check for validation */
1502     if (action_check_completion_mode(common, errp) < 0) {
1503         return;
1504     }
1505 
1506     bs = qmp_get_root_bs(device, errp);
1507     if (!bs) {
1508         return;
1509     }
1510 
1511     /* AioContext is released in .clean() */
1512     state->aio_context = bdrv_get_aio_context(bs);
1513     aio_context_acquire(state->aio_context);
1514 
1515     state->bs = bs;
1516     bdrv_drained_begin(bs);
1517 
1518     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_INTERNAL_SNAPSHOT, errp)) {
1519         return;
1520     }
1521 
1522     if (bdrv_is_read_only(bs)) {
1523         error_setg(errp, "Device '%s' is read only", device);
1524         return;
1525     }
1526 
1527     if (!bdrv_can_snapshot(bs)) {
1528         error_setg(errp, "Block format '%s' used by device '%s' "
1529                    "does not support internal snapshots",
1530                    bs->drv->format_name, device);
1531         return;
1532     }
1533 
1534     if (!strlen(name)) {
1535         error_setg(errp, "Name is empty");
1536         return;
1537     }
1538 
1539     /* check whether a snapshot with name exist */
1540     ret = bdrv_snapshot_find_by_id_and_name(bs, NULL, name, &old_sn,
1541                                             &local_err);
1542     if (local_err) {
1543         error_propagate(errp, local_err);
1544         return;
1545     } else if (ret) {
1546         error_setg(errp,
1547                    "Snapshot with name '%s' already exists on device '%s'",
1548                    name, device);
1549         return;
1550     }
1551 
1552     /* 3. take the snapshot */
1553     sn = &state->sn;
1554     pstrcpy(sn->name, sizeof(sn->name), name);
1555     qemu_gettimeofday(&tv);
1556     sn->date_sec = tv.tv_sec;
1557     sn->date_nsec = tv.tv_usec * 1000;
1558     sn->vm_clock_nsec = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL);
1559 
1560     ret1 = bdrv_snapshot_create(bs, sn);
1561     if (ret1 < 0) {
1562         error_setg_errno(errp, -ret1,
1563                          "Failed to create snapshot '%s' on device '%s'",
1564                          name, device);
1565         return;
1566     }
1567 
1568     /* 4. succeed, mark a snapshot is created */
1569     state->created = true;
1570 }
1571 
1572 static void internal_snapshot_abort(BlkActionState *common)
1573 {
1574     InternalSnapshotState *state =
1575                              DO_UPCAST(InternalSnapshotState, common, common);
1576     BlockDriverState *bs = state->bs;
1577     QEMUSnapshotInfo *sn = &state->sn;
1578     Error *local_error = NULL;
1579 
1580     if (!state->created) {
1581         return;
1582     }
1583 
1584     if (bdrv_snapshot_delete(bs, sn->id_str, sn->name, &local_error) < 0) {
1585         error_reportf_err(local_error,
1586                           "Failed to delete snapshot with id '%s' and "
1587                           "name '%s' on device '%s' in abort: ",
1588                           sn->id_str, sn->name,
1589                           bdrv_get_device_name(bs));
1590     }
1591 }
1592 
1593 static void internal_snapshot_clean(BlkActionState *common)
1594 {
1595     InternalSnapshotState *state = DO_UPCAST(InternalSnapshotState,
1596                                              common, common);
1597 
1598     if (state->aio_context) {
1599         if (state->bs) {
1600             bdrv_drained_end(state->bs);
1601         }
1602         aio_context_release(state->aio_context);
1603     }
1604 }
1605 
1606 /* external snapshot private data */
1607 typedef struct ExternalSnapshotState {
1608     BlkActionState common;
1609     BlockDriverState *old_bs;
1610     BlockDriverState *new_bs;
1611     AioContext *aio_context;
1612     bool overlay_appended;
1613 } ExternalSnapshotState;
1614 
1615 static void external_snapshot_prepare(BlkActionState *common,
1616                                       Error **errp)
1617 {
1618     int flags = 0;
1619     QDict *options = NULL;
1620     Error *local_err = NULL;
1621     /* Device and node name of the image to generate the snapshot from */
1622     const char *device;
1623     const char *node_name;
1624     /* Reference to the new image (for 'blockdev-snapshot') */
1625     const char *snapshot_ref;
1626     /* File name of the new image (for 'blockdev-snapshot-sync') */
1627     const char *new_image_file;
1628     ExternalSnapshotState *state =
1629                              DO_UPCAST(ExternalSnapshotState, common, common);
1630     TransactionAction *action = common->action;
1631 
1632     /* 'blockdev-snapshot' and 'blockdev-snapshot-sync' have similar
1633      * purpose but a different set of parameters */
1634     switch (action->type) {
1635     case TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT:
1636         {
1637             BlockdevSnapshot *s = action->u.blockdev_snapshot.data;
1638             device = s->node;
1639             node_name = s->node;
1640             new_image_file = NULL;
1641             snapshot_ref = s->overlay;
1642         }
1643         break;
1644     case TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC:
1645         {
1646             BlockdevSnapshotSync *s = action->u.blockdev_snapshot_sync.data;
1647             device = s->has_device ? s->device : NULL;
1648             node_name = s->has_node_name ? s->node_name : NULL;
1649             new_image_file = s->snapshot_file;
1650             snapshot_ref = NULL;
1651         }
1652         break;
1653     default:
1654         g_assert_not_reached();
1655     }
1656 
1657     /* start processing */
1658     if (action_check_completion_mode(common, errp) < 0) {
1659         return;
1660     }
1661 
1662     state->old_bs = bdrv_lookup_bs(device, node_name, errp);
1663     if (!state->old_bs) {
1664         return;
1665     }
1666 
1667     /* Acquire AioContext now so any threads operating on old_bs stop */
1668     state->aio_context = bdrv_get_aio_context(state->old_bs);
1669     aio_context_acquire(state->aio_context);
1670     bdrv_drained_begin(state->old_bs);
1671 
1672     if (!bdrv_is_inserted(state->old_bs)) {
1673         error_setg(errp, QERR_DEVICE_HAS_NO_MEDIUM, device);
1674         return;
1675     }
1676 
1677     if (bdrv_op_is_blocked(state->old_bs,
1678                            BLOCK_OP_TYPE_EXTERNAL_SNAPSHOT, errp)) {
1679         return;
1680     }
1681 
1682     if (!bdrv_is_read_only(state->old_bs)) {
1683         if (bdrv_flush(state->old_bs)) {
1684             error_setg(errp, QERR_IO_ERROR);
1685             return;
1686         }
1687     }
1688 
1689     if (!bdrv_is_first_non_filter(state->old_bs)) {
1690         error_setg(errp, QERR_FEATURE_DISABLED, "snapshot");
1691         return;
1692     }
1693 
1694     if (action->type == TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC) {
1695         BlockdevSnapshotSync *s = action->u.blockdev_snapshot_sync.data;
1696         const char *format = s->has_format ? s->format : "qcow2";
1697         enum NewImageMode mode;
1698         const char *snapshot_node_name =
1699             s->has_snapshot_node_name ? s->snapshot_node_name : NULL;
1700 
1701         if (node_name && !snapshot_node_name) {
1702             error_setg(errp, "New snapshot node name missing");
1703             return;
1704         }
1705 
1706         if (snapshot_node_name &&
1707             bdrv_lookup_bs(snapshot_node_name, snapshot_node_name, NULL)) {
1708             error_setg(errp, "New snapshot node name already in use");
1709             return;
1710         }
1711 
1712         flags = state->old_bs->open_flags;
1713         flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_COPY_ON_READ);
1714         flags |= BDRV_O_NO_BACKING;
1715 
1716         /* create new image w/backing file */
1717         mode = s->has_mode ? s->mode : NEW_IMAGE_MODE_ABSOLUTE_PATHS;
1718         if (mode != NEW_IMAGE_MODE_EXISTING) {
1719             int64_t size = bdrv_getlength(state->old_bs);
1720             if (size < 0) {
1721                 error_setg_errno(errp, -size, "bdrv_getlength failed");
1722                 return;
1723             }
1724             bdrv_img_create(new_image_file, format,
1725                             state->old_bs->filename,
1726                             state->old_bs->drv->format_name,
1727                             NULL, size, flags, false, &local_err);
1728             if (local_err) {
1729                 error_propagate(errp, local_err);
1730                 return;
1731             }
1732         }
1733 
1734         options = qdict_new();
1735         if (s->has_snapshot_node_name) {
1736             qdict_put_str(options, "node-name", snapshot_node_name);
1737         }
1738         qdict_put_str(options, "driver", format);
1739     }
1740 
1741     state->new_bs = bdrv_open(new_image_file, snapshot_ref, options, flags,
1742                               errp);
1743     /* We will manually add the backing_hd field to the bs later */
1744     if (!state->new_bs) {
1745         return;
1746     }
1747 
1748     if (bdrv_has_blk(state->new_bs)) {
1749         error_setg(errp, "The snapshot is already in use");
1750         return;
1751     }
1752 
1753     if (bdrv_op_is_blocked(state->new_bs, BLOCK_OP_TYPE_EXTERNAL_SNAPSHOT,
1754                            errp)) {
1755         return;
1756     }
1757 
1758     if (state->new_bs->backing != NULL) {
1759         error_setg(errp, "The snapshot already has a backing image");
1760         return;
1761     }
1762 
1763     if (!state->new_bs->drv->supports_backing) {
1764         error_setg(errp, "The snapshot does not support backing images");
1765         return;
1766     }
1767 
1768     bdrv_set_aio_context(state->new_bs, state->aio_context);
1769 
1770     /* This removes our old bs and adds the new bs. This is an operation that
1771      * can fail, so we need to do it in .prepare; undoing it for abort is
1772      * always possible. */
1773     bdrv_ref(state->new_bs);
1774     bdrv_append(state->new_bs, state->old_bs, &local_err);
1775     if (local_err) {
1776         error_propagate(errp, local_err);
1777         return;
1778     }
1779     state->overlay_appended = true;
1780 }
1781 
1782 static void external_snapshot_commit(BlkActionState *common)
1783 {
1784     ExternalSnapshotState *state =
1785                              DO_UPCAST(ExternalSnapshotState, common, common);
1786 
1787     /* We don't need (or want) to use the transactional
1788      * bdrv_reopen_multiple() across all the entries at once, because we
1789      * don't want to abort all of them if one of them fails the reopen */
1790     if (!atomic_read(&state->old_bs->copy_on_read)) {
1791         bdrv_reopen(state->old_bs, state->old_bs->open_flags & ~BDRV_O_RDWR,
1792                     NULL);
1793     }
1794 }
1795 
1796 static void external_snapshot_abort(BlkActionState *common)
1797 {
1798     ExternalSnapshotState *state =
1799                              DO_UPCAST(ExternalSnapshotState, common, common);
1800     if (state->new_bs) {
1801         if (state->overlay_appended) {
1802             bdrv_ref(state->old_bs);   /* we can't let bdrv_set_backind_hd()
1803                                           close state->old_bs; we need it */
1804             bdrv_set_backing_hd(state->new_bs, NULL, &error_abort);
1805             bdrv_replace_node(state->new_bs, state->old_bs, &error_abort);
1806             bdrv_unref(state->old_bs); /* bdrv_replace_node() ref'ed old_bs */
1807         }
1808     }
1809 }
1810 
1811 static void external_snapshot_clean(BlkActionState *common)
1812 {
1813     ExternalSnapshotState *state =
1814                              DO_UPCAST(ExternalSnapshotState, common, common);
1815     if (state->aio_context) {
1816         bdrv_drained_end(state->old_bs);
1817         aio_context_release(state->aio_context);
1818         bdrv_unref(state->new_bs);
1819     }
1820 }
1821 
1822 typedef struct DriveBackupState {
1823     BlkActionState common;
1824     BlockDriverState *bs;
1825     AioContext *aio_context;
1826     BlockJob *job;
1827 } DriveBackupState;
1828 
1829 static BlockJob *do_drive_backup(DriveBackup *backup, BlockJobTxn *txn,
1830                             Error **errp);
1831 
1832 static void drive_backup_prepare(BlkActionState *common, Error **errp)
1833 {
1834     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1835     BlockDriverState *bs;
1836     DriveBackup *backup;
1837     Error *local_err = NULL;
1838 
1839     assert(common->action->type == TRANSACTION_ACTION_KIND_DRIVE_BACKUP);
1840     backup = common->action->u.drive_backup.data;
1841 
1842     bs = qmp_get_root_bs(backup->device, errp);
1843     if (!bs) {
1844         return;
1845     }
1846 
1847     /* AioContext is released in .clean() */
1848     state->aio_context = bdrv_get_aio_context(bs);
1849     aio_context_acquire(state->aio_context);
1850     bdrv_drained_begin(bs);
1851     state->bs = bs;
1852 
1853     state->job = do_drive_backup(backup, common->block_job_txn, &local_err);
1854     if (local_err) {
1855         error_propagate(errp, local_err);
1856         return;
1857     }
1858 }
1859 
1860 static void drive_backup_commit(BlkActionState *common)
1861 {
1862     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1863     assert(state->job);
1864     block_job_start(state->job);
1865 }
1866 
1867 static void drive_backup_abort(BlkActionState *common)
1868 {
1869     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1870 
1871     if (state->job) {
1872         block_job_cancel_sync(state->job);
1873     }
1874 }
1875 
1876 static void drive_backup_clean(BlkActionState *common)
1877 {
1878     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1879 
1880     if (state->aio_context) {
1881         bdrv_drained_end(state->bs);
1882         aio_context_release(state->aio_context);
1883     }
1884 }
1885 
1886 typedef struct BlockdevBackupState {
1887     BlkActionState common;
1888     BlockDriverState *bs;
1889     BlockJob *job;
1890     AioContext *aio_context;
1891 } BlockdevBackupState;
1892 
1893 static BlockJob *do_blockdev_backup(BlockdevBackup *backup, BlockJobTxn *txn,
1894                                     Error **errp);
1895 
1896 static void blockdev_backup_prepare(BlkActionState *common, Error **errp)
1897 {
1898     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1899     BlockdevBackup *backup;
1900     BlockDriverState *bs, *target;
1901     Error *local_err = NULL;
1902 
1903     assert(common->action->type == TRANSACTION_ACTION_KIND_BLOCKDEV_BACKUP);
1904     backup = common->action->u.blockdev_backup.data;
1905 
1906     bs = qmp_get_root_bs(backup->device, errp);
1907     if (!bs) {
1908         return;
1909     }
1910 
1911     target = bdrv_lookup_bs(backup->target, backup->target, errp);
1912     if (!target) {
1913         return;
1914     }
1915 
1916     /* AioContext is released in .clean() */
1917     state->aio_context = bdrv_get_aio_context(bs);
1918     if (state->aio_context != bdrv_get_aio_context(target)) {
1919         state->aio_context = NULL;
1920         error_setg(errp, "Backup between two IO threads is not implemented");
1921         return;
1922     }
1923     aio_context_acquire(state->aio_context);
1924     state->bs = bs;
1925     bdrv_drained_begin(state->bs);
1926 
1927     state->job = do_blockdev_backup(backup, common->block_job_txn, &local_err);
1928     if (local_err) {
1929         error_propagate(errp, local_err);
1930         return;
1931     }
1932 }
1933 
1934 static void blockdev_backup_commit(BlkActionState *common)
1935 {
1936     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1937     assert(state->job);
1938     block_job_start(state->job);
1939 }
1940 
1941 static void blockdev_backup_abort(BlkActionState *common)
1942 {
1943     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1944 
1945     if (state->job) {
1946         block_job_cancel_sync(state->job);
1947     }
1948 }
1949 
1950 static void blockdev_backup_clean(BlkActionState *common)
1951 {
1952     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1953 
1954     if (state->aio_context) {
1955         bdrv_drained_end(state->bs);
1956         aio_context_release(state->aio_context);
1957     }
1958 }
1959 
1960 typedef struct BlockDirtyBitmapState {
1961     BlkActionState common;
1962     BdrvDirtyBitmap *bitmap;
1963     BlockDriverState *bs;
1964     AioContext *aio_context;
1965     HBitmap *backup;
1966     bool prepared;
1967 } BlockDirtyBitmapState;
1968 
1969 static void block_dirty_bitmap_add_prepare(BlkActionState *common,
1970                                            Error **errp)
1971 {
1972     Error *local_err = NULL;
1973     BlockDirtyBitmapAdd *action;
1974     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
1975                                              common, common);
1976 
1977     if (action_check_completion_mode(common, errp) < 0) {
1978         return;
1979     }
1980 
1981     action = common->action->u.block_dirty_bitmap_add.data;
1982     /* AIO context taken and released within qmp_block_dirty_bitmap_add */
1983     qmp_block_dirty_bitmap_add(action->node, action->name,
1984                                action->has_granularity, action->granularity,
1985                                action->has_persistent, action->persistent,
1986                                action->has_autoload, action->autoload,
1987                                &local_err);
1988 
1989     if (!local_err) {
1990         state->prepared = true;
1991     } else {
1992         error_propagate(errp, local_err);
1993     }
1994 }
1995 
1996 static void block_dirty_bitmap_add_abort(BlkActionState *common)
1997 {
1998     BlockDirtyBitmapAdd *action;
1999     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2000                                              common, common);
2001 
2002     action = common->action->u.block_dirty_bitmap_add.data;
2003     /* Should not be able to fail: IF the bitmap was added via .prepare(),
2004      * then the node reference and bitmap name must have been valid.
2005      */
2006     if (state->prepared) {
2007         qmp_block_dirty_bitmap_remove(action->node, action->name, &error_abort);
2008     }
2009 }
2010 
2011 static void block_dirty_bitmap_clear_prepare(BlkActionState *common,
2012                                              Error **errp)
2013 {
2014     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2015                                              common, common);
2016     BlockDirtyBitmap *action;
2017 
2018     if (action_check_completion_mode(common, errp) < 0) {
2019         return;
2020     }
2021 
2022     action = common->action->u.block_dirty_bitmap_clear.data;
2023     state->bitmap = block_dirty_bitmap_lookup(action->node,
2024                                               action->name,
2025                                               &state->bs,
2026                                               errp);
2027     if (!state->bitmap) {
2028         return;
2029     }
2030 
2031     if (bdrv_dirty_bitmap_frozen(state->bitmap)) {
2032         error_setg(errp, "Cannot modify a frozen bitmap");
2033         return;
2034     } else if (!bdrv_dirty_bitmap_enabled(state->bitmap)) {
2035         error_setg(errp, "Cannot clear a disabled bitmap");
2036         return;
2037     } else if (bdrv_dirty_bitmap_readonly(state->bitmap)) {
2038         error_setg(errp, "Cannot clear a readonly bitmap");
2039         return;
2040     }
2041 
2042     bdrv_clear_dirty_bitmap(state->bitmap, &state->backup);
2043     /* AioContext is released in .clean() */
2044 }
2045 
2046 static void block_dirty_bitmap_clear_abort(BlkActionState *common)
2047 {
2048     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2049                                              common, common);
2050 
2051     if (state->backup) {
2052         bdrv_undo_clear_dirty_bitmap(state->bitmap, state->backup);
2053     }
2054 }
2055 
2056 static void block_dirty_bitmap_clear_commit(BlkActionState *common)
2057 {
2058     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2059                                              common, common);
2060 
2061     hbitmap_free(state->backup);
2062 }
2063 
2064 static void block_dirty_bitmap_clear_clean(BlkActionState *common)
2065 {
2066     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2067                                              common, common);
2068 
2069     if (state->aio_context) {
2070         aio_context_release(state->aio_context);
2071     }
2072 }
2073 
2074 static void abort_prepare(BlkActionState *common, Error **errp)
2075 {
2076     error_setg(errp, "Transaction aborted using Abort action");
2077 }
2078 
2079 static void abort_commit(BlkActionState *common)
2080 {
2081     g_assert_not_reached(); /* this action never succeeds */
2082 }
2083 
2084 static const BlkActionOps actions[] = {
2085     [TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT] = {
2086         .instance_size = sizeof(ExternalSnapshotState),
2087         .prepare  = external_snapshot_prepare,
2088         .commit   = external_snapshot_commit,
2089         .abort = external_snapshot_abort,
2090         .clean = external_snapshot_clean,
2091     },
2092     [TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC] = {
2093         .instance_size = sizeof(ExternalSnapshotState),
2094         .prepare  = external_snapshot_prepare,
2095         .commit   = external_snapshot_commit,
2096         .abort = external_snapshot_abort,
2097         .clean = external_snapshot_clean,
2098     },
2099     [TRANSACTION_ACTION_KIND_DRIVE_BACKUP] = {
2100         .instance_size = sizeof(DriveBackupState),
2101         .prepare = drive_backup_prepare,
2102         .commit = drive_backup_commit,
2103         .abort = drive_backup_abort,
2104         .clean = drive_backup_clean,
2105     },
2106     [TRANSACTION_ACTION_KIND_BLOCKDEV_BACKUP] = {
2107         .instance_size = sizeof(BlockdevBackupState),
2108         .prepare = blockdev_backup_prepare,
2109         .commit = blockdev_backup_commit,
2110         .abort = blockdev_backup_abort,
2111         .clean = blockdev_backup_clean,
2112     },
2113     [TRANSACTION_ACTION_KIND_ABORT] = {
2114         .instance_size = sizeof(BlkActionState),
2115         .prepare = abort_prepare,
2116         .commit = abort_commit,
2117     },
2118     [TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC] = {
2119         .instance_size = sizeof(InternalSnapshotState),
2120         .prepare  = internal_snapshot_prepare,
2121         .abort = internal_snapshot_abort,
2122         .clean = internal_snapshot_clean,
2123     },
2124     [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_ADD] = {
2125         .instance_size = sizeof(BlockDirtyBitmapState),
2126         .prepare = block_dirty_bitmap_add_prepare,
2127         .abort = block_dirty_bitmap_add_abort,
2128     },
2129     [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_CLEAR] = {
2130         .instance_size = sizeof(BlockDirtyBitmapState),
2131         .prepare = block_dirty_bitmap_clear_prepare,
2132         .commit = block_dirty_bitmap_clear_commit,
2133         .abort = block_dirty_bitmap_clear_abort,
2134         .clean = block_dirty_bitmap_clear_clean,
2135     }
2136 };
2137 
2138 /**
2139  * Allocate a TransactionProperties structure if necessary, and fill
2140  * that structure with desired defaults if they are unset.
2141  */
2142 static TransactionProperties *get_transaction_properties(
2143     TransactionProperties *props)
2144 {
2145     if (!props) {
2146         props = g_new0(TransactionProperties, 1);
2147     }
2148 
2149     if (!props->has_completion_mode) {
2150         props->has_completion_mode = true;
2151         props->completion_mode = ACTION_COMPLETION_MODE_INDIVIDUAL;
2152     }
2153 
2154     return props;
2155 }
2156 
2157 /*
2158  * 'Atomic' group operations.  The operations are performed as a set, and if
2159  * any fail then we roll back all operations in the group.
2160  */
2161 void qmp_transaction(TransactionActionList *dev_list,
2162                      bool has_props,
2163                      struct TransactionProperties *props,
2164                      Error **errp)
2165 {
2166     TransactionActionList *dev_entry = dev_list;
2167     BlockJobTxn *block_job_txn = NULL;
2168     BlkActionState *state, *next;
2169     Error *local_err = NULL;
2170 
2171     QSIMPLEQ_HEAD(snap_bdrv_states, BlkActionState) snap_bdrv_states;
2172     QSIMPLEQ_INIT(&snap_bdrv_states);
2173 
2174     /* Does this transaction get canceled as a group on failure?
2175      * If not, we don't really need to make a BlockJobTxn.
2176      */
2177     props = get_transaction_properties(props);
2178     if (props->completion_mode != ACTION_COMPLETION_MODE_INDIVIDUAL) {
2179         block_job_txn = block_job_txn_new();
2180     }
2181 
2182     /* drain all i/o before any operations */
2183     bdrv_drain_all();
2184 
2185     /* We don't do anything in this loop that commits us to the operations */
2186     while (NULL != dev_entry) {
2187         TransactionAction *dev_info = NULL;
2188         const BlkActionOps *ops;
2189 
2190         dev_info = dev_entry->value;
2191         dev_entry = dev_entry->next;
2192 
2193         assert(dev_info->type < ARRAY_SIZE(actions));
2194 
2195         ops = &actions[dev_info->type];
2196         assert(ops->instance_size > 0);
2197 
2198         state = g_malloc0(ops->instance_size);
2199         state->ops = ops;
2200         state->action = dev_info;
2201         state->block_job_txn = block_job_txn;
2202         state->txn_props = props;
2203         QSIMPLEQ_INSERT_TAIL(&snap_bdrv_states, state, entry);
2204 
2205         state->ops->prepare(state, &local_err);
2206         if (local_err) {
2207             error_propagate(errp, local_err);
2208             goto delete_and_fail;
2209         }
2210     }
2211 
2212     QSIMPLEQ_FOREACH(state, &snap_bdrv_states, entry) {
2213         if (state->ops->commit) {
2214             state->ops->commit(state);
2215         }
2216     }
2217 
2218     /* success */
2219     goto exit;
2220 
2221 delete_and_fail:
2222     /* failure, and it is all-or-none; roll back all operations */
2223     QSIMPLEQ_FOREACH(state, &snap_bdrv_states, entry) {
2224         if (state->ops->abort) {
2225             state->ops->abort(state);
2226         }
2227     }
2228 exit:
2229     QSIMPLEQ_FOREACH_SAFE(state, &snap_bdrv_states, entry, next) {
2230         if (state->ops->clean) {
2231             state->ops->clean(state);
2232         }
2233         g_free(state);
2234     }
2235     if (!has_props) {
2236         qapi_free_TransactionProperties(props);
2237     }
2238     block_job_txn_unref(block_job_txn);
2239 }
2240 
2241 void qmp_eject(bool has_device, const char *device,
2242                bool has_id, const char *id,
2243                bool has_force, bool force, Error **errp)
2244 {
2245     Error *local_err = NULL;
2246     int rc;
2247 
2248     if (!has_force) {
2249         force = false;
2250     }
2251 
2252     rc = do_open_tray(has_device ? device : NULL,
2253                       has_id ? id : NULL,
2254                       force, &local_err);
2255     if (rc && rc != -ENOSYS) {
2256         error_propagate(errp, local_err);
2257         return;
2258     }
2259     error_free(local_err);
2260 
2261     qmp_x_blockdev_remove_medium(has_device, device, has_id, id, errp);
2262 }
2263 
2264 void qmp_block_passwd(bool has_device, const char *device,
2265                       bool has_node_name, const char *node_name,
2266                       const char *password, Error **errp)
2267 {
2268     error_setg(errp,
2269                "Setting block passwords directly is no longer supported");
2270 }
2271 
2272 /*
2273  * Attempt to open the tray of @device.
2274  * If @force, ignore its tray lock.
2275  * Else, if the tray is locked, don't open it, but ask the guest to open it.
2276  * On error, store an error through @errp and return -errno.
2277  * If @device does not exist, return -ENODEV.
2278  * If it has no removable media, return -ENOTSUP.
2279  * If it has no tray, return -ENOSYS.
2280  * If the guest was asked to open the tray, return -EINPROGRESS.
2281  * Else, return 0.
2282  */
2283 static int do_open_tray(const char *blk_name, const char *qdev_id,
2284                         bool force, Error **errp)
2285 {
2286     BlockBackend *blk;
2287     const char *device = qdev_id ?: blk_name;
2288     bool locked;
2289 
2290     blk = qmp_get_blk(blk_name, qdev_id, errp);
2291     if (!blk) {
2292         return -ENODEV;
2293     }
2294 
2295     if (!blk_dev_has_removable_media(blk)) {
2296         error_setg(errp, "Device '%s' is not removable", device);
2297         return -ENOTSUP;
2298     }
2299 
2300     if (!blk_dev_has_tray(blk)) {
2301         error_setg(errp, "Device '%s' does not have a tray", device);
2302         return -ENOSYS;
2303     }
2304 
2305     if (blk_dev_is_tray_open(blk)) {
2306         return 0;
2307     }
2308 
2309     locked = blk_dev_is_medium_locked(blk);
2310     if (locked) {
2311         blk_dev_eject_request(blk, force);
2312     }
2313 
2314     if (!locked || force) {
2315         blk_dev_change_media_cb(blk, false, &error_abort);
2316     }
2317 
2318     if (locked && !force) {
2319         error_setg(errp, "Device '%s' is locked and force was not specified, "
2320                    "wait for tray to open and try again", device);
2321         return -EINPROGRESS;
2322     }
2323 
2324     return 0;
2325 }
2326 
2327 void qmp_blockdev_open_tray(bool has_device, const char *device,
2328                             bool has_id, const char *id,
2329                             bool has_force, bool force,
2330                             Error **errp)
2331 {
2332     Error *local_err = NULL;
2333     int rc;
2334 
2335     if (!has_force) {
2336         force = false;
2337     }
2338     rc = do_open_tray(has_device ? device : NULL,
2339                       has_id ? id : NULL,
2340                       force, &local_err);
2341     if (rc && rc != -ENOSYS && rc != -EINPROGRESS) {
2342         error_propagate(errp, local_err);
2343         return;
2344     }
2345     error_free(local_err);
2346 }
2347 
2348 void qmp_blockdev_close_tray(bool has_device, const char *device,
2349                              bool has_id, const char *id,
2350                              Error **errp)
2351 {
2352     BlockBackend *blk;
2353     Error *local_err = NULL;
2354 
2355     device = has_device ? device : NULL;
2356     id = has_id ? id : NULL;
2357 
2358     blk = qmp_get_blk(device, id, errp);
2359     if (!blk) {
2360         return;
2361     }
2362 
2363     if (!blk_dev_has_removable_media(blk)) {
2364         error_setg(errp, "Device '%s' is not removable", device ?: id);
2365         return;
2366     }
2367 
2368     if (!blk_dev_has_tray(blk)) {
2369         /* Ignore this command on tray-less devices */
2370         return;
2371     }
2372 
2373     if (!blk_dev_is_tray_open(blk)) {
2374         return;
2375     }
2376 
2377     blk_dev_change_media_cb(blk, true, &local_err);
2378     if (local_err) {
2379         error_propagate(errp, local_err);
2380         return;
2381     }
2382 }
2383 
2384 void qmp_x_blockdev_remove_medium(bool has_device, const char *device,
2385                                   bool has_id, const char *id, Error **errp)
2386 {
2387     BlockBackend *blk;
2388     BlockDriverState *bs;
2389     AioContext *aio_context;
2390     bool has_attached_device;
2391 
2392     device = has_device ? device : NULL;
2393     id = has_id ? id : NULL;
2394 
2395     blk = qmp_get_blk(device, id, errp);
2396     if (!blk) {
2397         return;
2398     }
2399 
2400     /* For BBs without a device, we can exchange the BDS tree at will */
2401     has_attached_device = blk_get_attached_dev(blk);
2402 
2403     if (has_attached_device && !blk_dev_has_removable_media(blk)) {
2404         error_setg(errp, "Device '%s' is not removable", device ?: id);
2405         return;
2406     }
2407 
2408     if (has_attached_device && blk_dev_has_tray(blk) &&
2409         !blk_dev_is_tray_open(blk))
2410     {
2411         error_setg(errp, "Tray of device '%s' is not open", device ?: id);
2412         return;
2413     }
2414 
2415     bs = blk_bs(blk);
2416     if (!bs) {
2417         return;
2418     }
2419 
2420     aio_context = bdrv_get_aio_context(bs);
2421     aio_context_acquire(aio_context);
2422 
2423     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_EJECT, errp)) {
2424         goto out;
2425     }
2426 
2427     blk_remove_bs(blk);
2428 
2429     if (!blk_dev_has_tray(blk)) {
2430         /* For tray-less devices, blockdev-open-tray is a no-op (or may not be
2431          * called at all); therefore, the medium needs to be ejected here.
2432          * Do it after blk_remove_bs() so blk_is_inserted(blk) returns the @load
2433          * value passed here (i.e. false). */
2434         blk_dev_change_media_cb(blk, false, &error_abort);
2435     }
2436 
2437 out:
2438     aio_context_release(aio_context);
2439 }
2440 
2441 static void qmp_blockdev_insert_anon_medium(BlockBackend *blk,
2442                                             BlockDriverState *bs, Error **errp)
2443 {
2444     Error *local_err = NULL;
2445     bool has_device;
2446     int ret;
2447 
2448     /* For BBs without a device, we can exchange the BDS tree at will */
2449     has_device = blk_get_attached_dev(blk);
2450 
2451     if (has_device && !blk_dev_has_removable_media(blk)) {
2452         error_setg(errp, "Device is not removable");
2453         return;
2454     }
2455 
2456     if (has_device && blk_dev_has_tray(blk) && !blk_dev_is_tray_open(blk)) {
2457         error_setg(errp, "Tray of the device is not open");
2458         return;
2459     }
2460 
2461     if (blk_bs(blk)) {
2462         error_setg(errp, "There already is a medium in the device");
2463         return;
2464     }
2465 
2466     ret = blk_insert_bs(blk, bs, errp);
2467     if (ret < 0) {
2468         return;
2469     }
2470 
2471     if (!blk_dev_has_tray(blk)) {
2472         /* For tray-less devices, blockdev-close-tray is a no-op (or may not be
2473          * called at all); therefore, the medium needs to be pushed into the
2474          * slot here.
2475          * Do it after blk_insert_bs() so blk_is_inserted(blk) returns the @load
2476          * value passed here (i.e. true). */
2477         blk_dev_change_media_cb(blk, true, &local_err);
2478         if (local_err) {
2479             error_propagate(errp, local_err);
2480             blk_remove_bs(blk);
2481             return;
2482         }
2483     }
2484 }
2485 
2486 void qmp_x_blockdev_insert_medium(bool has_device, const char *device,
2487                                   bool has_id, const char *id,
2488                                   const char *node_name, Error **errp)
2489 {
2490     BlockBackend *blk;
2491     BlockDriverState *bs;
2492 
2493     blk = qmp_get_blk(has_device ? device : NULL,
2494                       has_id ? id : NULL,
2495                       errp);
2496     if (!blk) {
2497         return;
2498     }
2499 
2500     bs = bdrv_find_node(node_name);
2501     if (!bs) {
2502         error_setg(errp, "Node '%s' not found", node_name);
2503         return;
2504     }
2505 
2506     if (bdrv_has_blk(bs)) {
2507         error_setg(errp, "Node '%s' is already in use", node_name);
2508         return;
2509     }
2510 
2511     qmp_blockdev_insert_anon_medium(blk, bs, errp);
2512 }
2513 
2514 void qmp_blockdev_change_medium(bool has_device, const char *device,
2515                                 bool has_id, const char *id,
2516                                 const char *filename,
2517                                 bool has_format, const char *format,
2518                                 bool has_read_only,
2519                                 BlockdevChangeReadOnlyMode read_only,
2520                                 Error **errp)
2521 {
2522     BlockBackend *blk;
2523     BlockDriverState *medium_bs = NULL;
2524     int bdrv_flags;
2525     bool detect_zeroes;
2526     int rc;
2527     QDict *options = NULL;
2528     Error *err = NULL;
2529 
2530     blk = qmp_get_blk(has_device ? device : NULL,
2531                       has_id ? id : NULL,
2532                       errp);
2533     if (!blk) {
2534         goto fail;
2535     }
2536 
2537     if (blk_bs(blk)) {
2538         blk_update_root_state(blk);
2539     }
2540 
2541     bdrv_flags = blk_get_open_flags_from_root_state(blk);
2542     bdrv_flags &= ~(BDRV_O_TEMPORARY | BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING |
2543         BDRV_O_PROTOCOL);
2544 
2545     if (!has_read_only) {
2546         read_only = BLOCKDEV_CHANGE_READ_ONLY_MODE_RETAIN;
2547     }
2548 
2549     switch (read_only) {
2550     case BLOCKDEV_CHANGE_READ_ONLY_MODE_RETAIN:
2551         break;
2552 
2553     case BLOCKDEV_CHANGE_READ_ONLY_MODE_READ_ONLY:
2554         bdrv_flags &= ~BDRV_O_RDWR;
2555         break;
2556 
2557     case BLOCKDEV_CHANGE_READ_ONLY_MODE_READ_WRITE:
2558         bdrv_flags |= BDRV_O_RDWR;
2559         break;
2560 
2561     default:
2562         abort();
2563     }
2564 
2565     options = qdict_new();
2566     detect_zeroes = blk_get_detect_zeroes_from_root_state(blk);
2567     qdict_put_str(options, "detect-zeroes", detect_zeroes ? "on" : "off");
2568 
2569     if (has_format) {
2570         qdict_put_str(options, "driver", format);
2571     }
2572 
2573     medium_bs = bdrv_open(filename, NULL, options, bdrv_flags, errp);
2574     if (!medium_bs) {
2575         goto fail;
2576     }
2577 
2578     rc = do_open_tray(has_device ? device : NULL,
2579                       has_id ? id : NULL,
2580                       false, &err);
2581     if (rc && rc != -ENOSYS) {
2582         error_propagate(errp, err);
2583         goto fail;
2584     }
2585     error_free(err);
2586     err = NULL;
2587 
2588     qmp_x_blockdev_remove_medium(has_device, device, has_id, id, &err);
2589     if (err) {
2590         error_propagate(errp, err);
2591         goto fail;
2592     }
2593 
2594     qmp_blockdev_insert_anon_medium(blk, medium_bs, &err);
2595     if (err) {
2596         error_propagate(errp, err);
2597         goto fail;
2598     }
2599 
2600     qmp_blockdev_close_tray(has_device, device, has_id, id, errp);
2601 
2602 fail:
2603     /* If the medium has been inserted, the device has its own reference, so
2604      * ours must be relinquished; and if it has not been inserted successfully,
2605      * the reference must be relinquished anyway */
2606     bdrv_unref(medium_bs);
2607 }
2608 
2609 /* throttling disk I/O limits */
2610 void qmp_block_set_io_throttle(BlockIOThrottle *arg, Error **errp)
2611 {
2612     ThrottleConfig cfg;
2613     BlockDriverState *bs;
2614     BlockBackend *blk;
2615     AioContext *aio_context;
2616 
2617     blk = qmp_get_blk(arg->has_device ? arg->device : NULL,
2618                       arg->has_id ? arg->id : NULL,
2619                       errp);
2620     if (!blk) {
2621         return;
2622     }
2623 
2624     aio_context = blk_get_aio_context(blk);
2625     aio_context_acquire(aio_context);
2626 
2627     bs = blk_bs(blk);
2628     if (!bs) {
2629         error_setg(errp, "Device has no medium");
2630         goto out;
2631     }
2632 
2633     throttle_config_init(&cfg);
2634     cfg.buckets[THROTTLE_BPS_TOTAL].avg = arg->bps;
2635     cfg.buckets[THROTTLE_BPS_READ].avg  = arg->bps_rd;
2636     cfg.buckets[THROTTLE_BPS_WRITE].avg = arg->bps_wr;
2637 
2638     cfg.buckets[THROTTLE_OPS_TOTAL].avg = arg->iops;
2639     cfg.buckets[THROTTLE_OPS_READ].avg  = arg->iops_rd;
2640     cfg.buckets[THROTTLE_OPS_WRITE].avg = arg->iops_wr;
2641 
2642     if (arg->has_bps_max) {
2643         cfg.buckets[THROTTLE_BPS_TOTAL].max = arg->bps_max;
2644     }
2645     if (arg->has_bps_rd_max) {
2646         cfg.buckets[THROTTLE_BPS_READ].max = arg->bps_rd_max;
2647     }
2648     if (arg->has_bps_wr_max) {
2649         cfg.buckets[THROTTLE_BPS_WRITE].max = arg->bps_wr_max;
2650     }
2651     if (arg->has_iops_max) {
2652         cfg.buckets[THROTTLE_OPS_TOTAL].max = arg->iops_max;
2653     }
2654     if (arg->has_iops_rd_max) {
2655         cfg.buckets[THROTTLE_OPS_READ].max = arg->iops_rd_max;
2656     }
2657     if (arg->has_iops_wr_max) {
2658         cfg.buckets[THROTTLE_OPS_WRITE].max = arg->iops_wr_max;
2659     }
2660 
2661     if (arg->has_bps_max_length) {
2662         cfg.buckets[THROTTLE_BPS_TOTAL].burst_length = arg->bps_max_length;
2663     }
2664     if (arg->has_bps_rd_max_length) {
2665         cfg.buckets[THROTTLE_BPS_READ].burst_length = arg->bps_rd_max_length;
2666     }
2667     if (arg->has_bps_wr_max_length) {
2668         cfg.buckets[THROTTLE_BPS_WRITE].burst_length = arg->bps_wr_max_length;
2669     }
2670     if (arg->has_iops_max_length) {
2671         cfg.buckets[THROTTLE_OPS_TOTAL].burst_length = arg->iops_max_length;
2672     }
2673     if (arg->has_iops_rd_max_length) {
2674         cfg.buckets[THROTTLE_OPS_READ].burst_length = arg->iops_rd_max_length;
2675     }
2676     if (arg->has_iops_wr_max_length) {
2677         cfg.buckets[THROTTLE_OPS_WRITE].burst_length = arg->iops_wr_max_length;
2678     }
2679 
2680     if (arg->has_iops_size) {
2681         cfg.op_size = arg->iops_size;
2682     }
2683 
2684     if (!throttle_is_valid(&cfg, errp)) {
2685         goto out;
2686     }
2687 
2688     if (throttle_enabled(&cfg)) {
2689         /* Enable I/O limits if they're not enabled yet, otherwise
2690          * just update the throttling group. */
2691         if (!blk_get_public(blk)->throttle_state) {
2692             blk_io_limits_enable(blk,
2693                                  arg->has_group ? arg->group :
2694                                  arg->has_device ? arg->device :
2695                                  arg->id);
2696         } else if (arg->has_group) {
2697             blk_io_limits_update_group(blk, arg->group);
2698         }
2699         /* Set the new throttling configuration */
2700         blk_set_io_limits(blk, &cfg);
2701     } else if (blk_get_public(blk)->throttle_state) {
2702         /* If all throttling settings are set to 0, disable I/O limits */
2703         blk_io_limits_disable(blk);
2704     }
2705 
2706 out:
2707     aio_context_release(aio_context);
2708 }
2709 
2710 void qmp_block_dirty_bitmap_add(const char *node, const char *name,
2711                                 bool has_granularity, uint32_t granularity,
2712                                 bool has_persistent, bool persistent,
2713                                 bool has_autoload, bool autoload,
2714                                 Error **errp)
2715 {
2716     BlockDriverState *bs;
2717     BdrvDirtyBitmap *bitmap;
2718 
2719     if (!name || name[0] == '\0') {
2720         error_setg(errp, "Bitmap name cannot be empty");
2721         return;
2722     }
2723 
2724     bs = bdrv_lookup_bs(node, node, errp);
2725     if (!bs) {
2726         return;
2727     }
2728 
2729     if (has_granularity) {
2730         if (granularity < 512 || !is_power_of_2(granularity)) {
2731             error_setg(errp, "Granularity must be power of 2 "
2732                              "and at least 512");
2733             return;
2734         }
2735     } else {
2736         /* Default to cluster size, if available: */
2737         granularity = bdrv_get_default_bitmap_granularity(bs);
2738     }
2739 
2740     if (!has_persistent) {
2741         persistent = false;
2742     }
2743     if (!has_autoload) {
2744         autoload = false;
2745     }
2746 
2747     if (has_autoload && !persistent) {
2748         error_setg(errp, "Autoload flag must be used only for persistent "
2749                          "bitmaps");
2750         return;
2751     }
2752 
2753     if (persistent &&
2754         !bdrv_can_store_new_dirty_bitmap(bs, name, granularity, errp))
2755     {
2756         return;
2757     }
2758 
2759     bitmap = bdrv_create_dirty_bitmap(bs, granularity, name, errp);
2760     if (bitmap == NULL) {
2761         return;
2762     }
2763 
2764     bdrv_dirty_bitmap_set_persistance(bitmap, persistent);
2765     bdrv_dirty_bitmap_set_autoload(bitmap, autoload);
2766 }
2767 
2768 void qmp_block_dirty_bitmap_remove(const char *node, const char *name,
2769                                    Error **errp)
2770 {
2771     BlockDriverState *bs;
2772     BdrvDirtyBitmap *bitmap;
2773     Error *local_err = NULL;
2774 
2775     bitmap = block_dirty_bitmap_lookup(node, name, &bs, errp);
2776     if (!bitmap || !bs) {
2777         return;
2778     }
2779 
2780     if (bdrv_dirty_bitmap_frozen(bitmap)) {
2781         error_setg(errp,
2782                    "Bitmap '%s' is currently frozen and cannot be removed",
2783                    name);
2784         return;
2785     }
2786 
2787     if (bdrv_dirty_bitmap_get_persistance(bitmap)) {
2788         bdrv_remove_persistent_dirty_bitmap(bs, name, &local_err);
2789         if (local_err != NULL) {
2790             error_propagate(errp, local_err);
2791             return;
2792         }
2793     }
2794 
2795     bdrv_dirty_bitmap_make_anon(bitmap);
2796     bdrv_release_dirty_bitmap(bs, bitmap);
2797 }
2798 
2799 /**
2800  * Completely clear a bitmap, for the purposes of synchronizing a bitmap
2801  * immediately after a full backup operation.
2802  */
2803 void qmp_block_dirty_bitmap_clear(const char *node, const char *name,
2804                                   Error **errp)
2805 {
2806     BdrvDirtyBitmap *bitmap;
2807     BlockDriverState *bs;
2808 
2809     bitmap = block_dirty_bitmap_lookup(node, name, &bs, errp);
2810     if (!bitmap || !bs) {
2811         return;
2812     }
2813 
2814     if (bdrv_dirty_bitmap_frozen(bitmap)) {
2815         error_setg(errp,
2816                    "Bitmap '%s' is currently frozen and cannot be modified",
2817                    name);
2818         return;
2819     } else if (!bdrv_dirty_bitmap_enabled(bitmap)) {
2820         error_setg(errp,
2821                    "Bitmap '%s' is currently disabled and cannot be cleared",
2822                    name);
2823         return;
2824     } else if (bdrv_dirty_bitmap_readonly(bitmap)) {
2825         error_setg(errp, "Bitmap '%s' is readonly and cannot be cleared", name);
2826         return;
2827     }
2828 
2829     bdrv_clear_dirty_bitmap(bitmap, NULL);
2830 }
2831 
2832 BlockDirtyBitmapSha256 *qmp_x_debug_block_dirty_bitmap_sha256(const char *node,
2833                                                               const char *name,
2834                                                               Error **errp)
2835 {
2836     BdrvDirtyBitmap *bitmap;
2837     BlockDriverState *bs;
2838     BlockDirtyBitmapSha256 *ret = NULL;
2839     char *sha256;
2840 
2841     bitmap = block_dirty_bitmap_lookup(node, name, &bs, errp);
2842     if (!bitmap || !bs) {
2843         return NULL;
2844     }
2845 
2846     sha256 = bdrv_dirty_bitmap_sha256(bitmap, errp);
2847     if (sha256 == NULL) {
2848         return NULL;
2849     }
2850 
2851     ret = g_new(BlockDirtyBitmapSha256, 1);
2852     ret->sha256 = sha256;
2853 
2854     return ret;
2855 }
2856 
2857 void hmp_drive_del(Monitor *mon, const QDict *qdict)
2858 {
2859     const char *id = qdict_get_str(qdict, "id");
2860     BlockBackend *blk;
2861     BlockDriverState *bs;
2862     AioContext *aio_context;
2863     Error *local_err = NULL;
2864 
2865     bs = bdrv_find_node(id);
2866     if (bs) {
2867         qmp_blockdev_del(id, &local_err);
2868         if (local_err) {
2869             error_report_err(local_err);
2870         }
2871         return;
2872     }
2873 
2874     blk = blk_by_name(id);
2875     if (!blk) {
2876         error_report("Device '%s' not found", id);
2877         return;
2878     }
2879 
2880     if (!blk_legacy_dinfo(blk)) {
2881         error_report("Deleting device added with blockdev-add"
2882                      " is not supported");
2883         return;
2884     }
2885 
2886     aio_context = blk_get_aio_context(blk);
2887     aio_context_acquire(aio_context);
2888 
2889     bs = blk_bs(blk);
2890     if (bs) {
2891         if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_DRIVE_DEL, &local_err)) {
2892             error_report_err(local_err);
2893             aio_context_release(aio_context);
2894             return;
2895         }
2896 
2897         blk_remove_bs(blk);
2898     }
2899 
2900     /* Make the BlockBackend and the attached BlockDriverState anonymous */
2901     monitor_remove_blk(blk);
2902 
2903     /* If this BlockBackend has a device attached to it, its refcount will be
2904      * decremented when the device is removed; otherwise we have to do so here.
2905      */
2906     if (blk_get_attached_dev(blk)) {
2907         /* Further I/O must not pause the guest */
2908         blk_set_on_error(blk, BLOCKDEV_ON_ERROR_REPORT,
2909                          BLOCKDEV_ON_ERROR_REPORT);
2910     } else {
2911         blk_unref(blk);
2912     }
2913 
2914     aio_context_release(aio_context);
2915 }
2916 
2917 void qmp_block_resize(bool has_device, const char *device,
2918                       bool has_node_name, const char *node_name,
2919                       int64_t size, Error **errp)
2920 {
2921     Error *local_err = NULL;
2922     BlockBackend *blk = NULL;
2923     BlockDriverState *bs;
2924     AioContext *aio_context;
2925     int ret;
2926 
2927     bs = bdrv_lookup_bs(has_device ? device : NULL,
2928                         has_node_name ? node_name : NULL,
2929                         &local_err);
2930     if (local_err) {
2931         error_propagate(errp, local_err);
2932         return;
2933     }
2934 
2935     aio_context = bdrv_get_aio_context(bs);
2936     aio_context_acquire(aio_context);
2937 
2938     if (!bdrv_is_first_non_filter(bs)) {
2939         error_setg(errp, QERR_FEATURE_DISABLED, "resize");
2940         goto out;
2941     }
2942 
2943     if (size < 0) {
2944         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "size", "a >0 size");
2945         goto out;
2946     }
2947 
2948     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_RESIZE, NULL)) {
2949         error_setg(errp, QERR_DEVICE_IN_USE, device);
2950         goto out;
2951     }
2952 
2953     blk = blk_new(BLK_PERM_RESIZE, BLK_PERM_ALL);
2954     ret = blk_insert_bs(blk, bs, errp);
2955     if (ret < 0) {
2956         goto out;
2957     }
2958 
2959     bdrv_drained_begin(bs);
2960     ret = blk_truncate(blk, size, PREALLOC_MODE_OFF, errp);
2961     bdrv_drained_end(bs);
2962 
2963 out:
2964     blk_unref(blk);
2965     aio_context_release(aio_context);
2966 }
2967 
2968 void qmp_block_stream(bool has_job_id, const char *job_id, const char *device,
2969                       bool has_base, const char *base,
2970                       bool has_base_node, const char *base_node,
2971                       bool has_backing_file, const char *backing_file,
2972                       bool has_speed, int64_t speed,
2973                       bool has_on_error, BlockdevOnError on_error,
2974                       Error **errp)
2975 {
2976     BlockDriverState *bs, *iter;
2977     BlockDriverState *base_bs = NULL;
2978     AioContext *aio_context;
2979     Error *local_err = NULL;
2980     const char *base_name = NULL;
2981 
2982     if (!has_on_error) {
2983         on_error = BLOCKDEV_ON_ERROR_REPORT;
2984     }
2985 
2986     bs = bdrv_lookup_bs(device, device, errp);
2987     if (!bs) {
2988         return;
2989     }
2990 
2991     aio_context = bdrv_get_aio_context(bs);
2992     aio_context_acquire(aio_context);
2993 
2994     if (has_base && has_base_node) {
2995         error_setg(errp, "'base' and 'base-node' cannot be specified "
2996                    "at the same time");
2997         goto out;
2998     }
2999 
3000     if (has_base) {
3001         base_bs = bdrv_find_backing_image(bs, base);
3002         if (base_bs == NULL) {
3003             error_setg(errp, QERR_BASE_NOT_FOUND, base);
3004             goto out;
3005         }
3006         assert(bdrv_get_aio_context(base_bs) == aio_context);
3007         base_name = base;
3008     }
3009 
3010     if (has_base_node) {
3011         base_bs = bdrv_lookup_bs(NULL, base_node, errp);
3012         if (!base_bs) {
3013             goto out;
3014         }
3015         if (bs == base_bs || !bdrv_chain_contains(bs, base_bs)) {
3016             error_setg(errp, "Node '%s' is not a backing image of '%s'",
3017                        base_node, device);
3018             goto out;
3019         }
3020         assert(bdrv_get_aio_context(base_bs) == aio_context);
3021         base_name = base_bs->filename;
3022     }
3023 
3024     /* Check for op blockers in the whole chain between bs and base */
3025     for (iter = bs; iter && iter != base_bs; iter = backing_bs(iter)) {
3026         if (bdrv_op_is_blocked(iter, BLOCK_OP_TYPE_STREAM, errp)) {
3027             goto out;
3028         }
3029     }
3030 
3031     /* if we are streaming the entire chain, the result will have no backing
3032      * file, and specifying one is therefore an error */
3033     if (base_bs == NULL && has_backing_file) {
3034         error_setg(errp, "backing file specified, but streaming the "
3035                          "entire chain");
3036         goto out;
3037     }
3038 
3039     /* backing_file string overrides base bs filename */
3040     base_name = has_backing_file ? backing_file : base_name;
3041 
3042     stream_start(has_job_id ? job_id : NULL, bs, base_bs, base_name,
3043                  has_speed ? speed : 0, on_error, &local_err);
3044     if (local_err) {
3045         error_propagate(errp, local_err);
3046         goto out;
3047     }
3048 
3049     trace_qmp_block_stream(bs, bs->job);
3050 
3051 out:
3052     aio_context_release(aio_context);
3053 }
3054 
3055 void qmp_block_commit(bool has_job_id, const char *job_id, const char *device,
3056                       bool has_base, const char *base,
3057                       bool has_top, const char *top,
3058                       bool has_backing_file, const char *backing_file,
3059                       bool has_speed, int64_t speed,
3060                       bool has_filter_node_name, const char *filter_node_name,
3061                       Error **errp)
3062 {
3063     BlockDriverState *bs;
3064     BlockDriverState *iter;
3065     BlockDriverState *base_bs, *top_bs;
3066     AioContext *aio_context;
3067     Error *local_err = NULL;
3068     /* This will be part of the QMP command, if/when the
3069      * BlockdevOnError change for blkmirror makes it in
3070      */
3071     BlockdevOnError on_error = BLOCKDEV_ON_ERROR_REPORT;
3072 
3073     if (!has_speed) {
3074         speed = 0;
3075     }
3076     if (!has_filter_node_name) {
3077         filter_node_name = NULL;
3078     }
3079 
3080     /* Important Note:
3081      *  libvirt relies on the DeviceNotFound error class in order to probe for
3082      *  live commit feature versions; for this to work, we must make sure to
3083      *  perform the device lookup before any generic errors that may occur in a
3084      *  scenario in which all optional arguments are omitted. */
3085     bs = qmp_get_root_bs(device, &local_err);
3086     if (!bs) {
3087         bs = bdrv_lookup_bs(device, device, NULL);
3088         if (!bs) {
3089             error_free(local_err);
3090             error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
3091                       "Device '%s' not found", device);
3092         } else {
3093             error_propagate(errp, local_err);
3094         }
3095         return;
3096     }
3097 
3098     aio_context = bdrv_get_aio_context(bs);
3099     aio_context_acquire(aio_context);
3100 
3101     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_COMMIT_SOURCE, errp)) {
3102         goto out;
3103     }
3104 
3105     /* default top_bs is the active layer */
3106     top_bs = bs;
3107 
3108     if (has_top && top) {
3109         if (strcmp(bs->filename, top) != 0) {
3110             top_bs = bdrv_find_backing_image(bs, top);
3111         }
3112     }
3113 
3114     if (top_bs == NULL) {
3115         error_setg(errp, "Top image file %s not found", top ? top : "NULL");
3116         goto out;
3117     }
3118 
3119     assert(bdrv_get_aio_context(top_bs) == aio_context);
3120 
3121     if (has_base && base) {
3122         base_bs = bdrv_find_backing_image(top_bs, base);
3123     } else {
3124         base_bs = bdrv_find_base(top_bs);
3125     }
3126 
3127     if (base_bs == NULL) {
3128         error_setg(errp, QERR_BASE_NOT_FOUND, base ? base : "NULL");
3129         goto out;
3130     }
3131 
3132     assert(bdrv_get_aio_context(base_bs) == aio_context);
3133 
3134     for (iter = top_bs; iter != backing_bs(base_bs); iter = backing_bs(iter)) {
3135         if (bdrv_op_is_blocked(iter, BLOCK_OP_TYPE_COMMIT_TARGET, errp)) {
3136             goto out;
3137         }
3138     }
3139 
3140     /* Do not allow attempts to commit an image into itself */
3141     if (top_bs == base_bs) {
3142         error_setg(errp, "cannot commit an image into itself");
3143         goto out;
3144     }
3145 
3146     if (top_bs == bs) {
3147         if (has_backing_file) {
3148             error_setg(errp, "'backing-file' specified,"
3149                              " but 'top' is the active layer");
3150             goto out;
3151         }
3152         commit_active_start(has_job_id ? job_id : NULL, bs, base_bs,
3153                             BLOCK_JOB_DEFAULT, speed, on_error,
3154                             filter_node_name, NULL, NULL, false, &local_err);
3155     } else {
3156         BlockDriverState *overlay_bs = bdrv_find_overlay(bs, top_bs);
3157         if (bdrv_op_is_blocked(overlay_bs, BLOCK_OP_TYPE_COMMIT_TARGET, errp)) {
3158             goto out;
3159         }
3160         commit_start(has_job_id ? job_id : NULL, bs, base_bs, top_bs, speed,
3161                      on_error, has_backing_file ? backing_file : NULL,
3162                      filter_node_name, &local_err);
3163     }
3164     if (local_err != NULL) {
3165         error_propagate(errp, local_err);
3166         goto out;
3167     }
3168 
3169 out:
3170     aio_context_release(aio_context);
3171 }
3172 
3173 static BlockJob *do_drive_backup(DriveBackup *backup, BlockJobTxn *txn,
3174                                  Error **errp)
3175 {
3176     BlockDriverState *bs;
3177     BlockDriverState *target_bs;
3178     BlockDriverState *source = NULL;
3179     BlockJob *job = NULL;
3180     BdrvDirtyBitmap *bmap = NULL;
3181     AioContext *aio_context;
3182     QDict *options = NULL;
3183     Error *local_err = NULL;
3184     int flags;
3185     int64_t size;
3186     bool set_backing_hd = false;
3187 
3188     if (!backup->has_speed) {
3189         backup->speed = 0;
3190     }
3191     if (!backup->has_on_source_error) {
3192         backup->on_source_error = BLOCKDEV_ON_ERROR_REPORT;
3193     }
3194     if (!backup->has_on_target_error) {
3195         backup->on_target_error = BLOCKDEV_ON_ERROR_REPORT;
3196     }
3197     if (!backup->has_mode) {
3198         backup->mode = NEW_IMAGE_MODE_ABSOLUTE_PATHS;
3199     }
3200     if (!backup->has_job_id) {
3201         backup->job_id = NULL;
3202     }
3203     if (!backup->has_compress) {
3204         backup->compress = false;
3205     }
3206 
3207     bs = qmp_get_root_bs(backup->device, errp);
3208     if (!bs) {
3209         return NULL;
3210     }
3211 
3212     aio_context = bdrv_get_aio_context(bs);
3213     aio_context_acquire(aio_context);
3214 
3215     if (!backup->has_format) {
3216         backup->format = backup->mode == NEW_IMAGE_MODE_EXISTING ?
3217                          NULL : (char*) bs->drv->format_name;
3218     }
3219 
3220     /* Early check to avoid creating target */
3221     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_BACKUP_SOURCE, errp)) {
3222         goto out;
3223     }
3224 
3225     flags = bs->open_flags | BDRV_O_RDWR;
3226 
3227     /* See if we have a backing HD we can use to create our new image
3228      * on top of. */
3229     if (backup->sync == MIRROR_SYNC_MODE_TOP) {
3230         source = backing_bs(bs);
3231         if (!source) {
3232             backup->sync = MIRROR_SYNC_MODE_FULL;
3233         }
3234     }
3235     if (backup->sync == MIRROR_SYNC_MODE_NONE) {
3236         source = bs;
3237         flags |= BDRV_O_NO_BACKING;
3238         set_backing_hd = true;
3239     }
3240 
3241     size = bdrv_getlength(bs);
3242     if (size < 0) {
3243         error_setg_errno(errp, -size, "bdrv_getlength failed");
3244         goto out;
3245     }
3246 
3247     if (backup->mode != NEW_IMAGE_MODE_EXISTING) {
3248         assert(backup->format);
3249         if (source) {
3250             bdrv_img_create(backup->target, backup->format, source->filename,
3251                             source->drv->format_name, NULL,
3252                             size, flags, false, &local_err);
3253         } else {
3254             bdrv_img_create(backup->target, backup->format, NULL, NULL, NULL,
3255                             size, flags, false, &local_err);
3256         }
3257     }
3258 
3259     if (local_err) {
3260         error_propagate(errp, local_err);
3261         goto out;
3262     }
3263 
3264     if (backup->format) {
3265         if (!options) {
3266             options = qdict_new();
3267         }
3268         qdict_put_str(options, "driver", backup->format);
3269     }
3270 
3271     target_bs = bdrv_open(backup->target, NULL, options, flags, errp);
3272     if (!target_bs) {
3273         goto out;
3274     }
3275 
3276     bdrv_set_aio_context(target_bs, aio_context);
3277 
3278     if (set_backing_hd) {
3279         bdrv_set_backing_hd(target_bs, source, &local_err);
3280         if (local_err) {
3281             bdrv_unref(target_bs);
3282             goto out;
3283         }
3284     }
3285 
3286     if (backup->has_bitmap) {
3287         bmap = bdrv_find_dirty_bitmap(bs, backup->bitmap);
3288         if (!bmap) {
3289             error_setg(errp, "Bitmap '%s' could not be found", backup->bitmap);
3290             bdrv_unref(target_bs);
3291             goto out;
3292         }
3293     }
3294 
3295     job = backup_job_create(backup->job_id, bs, target_bs, backup->speed,
3296                             backup->sync, bmap, backup->compress,
3297                             backup->on_source_error, backup->on_target_error,
3298                             BLOCK_JOB_DEFAULT, NULL, NULL, txn, &local_err);
3299     bdrv_unref(target_bs);
3300     if (local_err != NULL) {
3301         error_propagate(errp, local_err);
3302         goto out;
3303     }
3304 
3305 out:
3306     aio_context_release(aio_context);
3307     return job;
3308 }
3309 
3310 void qmp_drive_backup(DriveBackup *arg, Error **errp)
3311 {
3312 
3313     BlockJob *job;
3314     job = do_drive_backup(arg, NULL, errp);
3315     if (job) {
3316         block_job_start(job);
3317     }
3318 }
3319 
3320 BlockDeviceInfoList *qmp_query_named_block_nodes(Error **errp)
3321 {
3322     return bdrv_named_nodes_list(errp);
3323 }
3324 
3325 BlockJob *do_blockdev_backup(BlockdevBackup *backup, BlockJobTxn *txn,
3326                              Error **errp)
3327 {
3328     BlockDriverState *bs;
3329     BlockDriverState *target_bs;
3330     Error *local_err = NULL;
3331     AioContext *aio_context;
3332     BlockJob *job = NULL;
3333 
3334     if (!backup->has_speed) {
3335         backup->speed = 0;
3336     }
3337     if (!backup->has_on_source_error) {
3338         backup->on_source_error = BLOCKDEV_ON_ERROR_REPORT;
3339     }
3340     if (!backup->has_on_target_error) {
3341         backup->on_target_error = BLOCKDEV_ON_ERROR_REPORT;
3342     }
3343     if (!backup->has_job_id) {
3344         backup->job_id = NULL;
3345     }
3346     if (!backup->has_compress) {
3347         backup->compress = false;
3348     }
3349 
3350     bs = qmp_get_root_bs(backup->device, errp);
3351     if (!bs) {
3352         return NULL;
3353     }
3354 
3355     aio_context = bdrv_get_aio_context(bs);
3356     aio_context_acquire(aio_context);
3357 
3358     target_bs = bdrv_lookup_bs(backup->target, backup->target, errp);
3359     if (!target_bs) {
3360         goto out;
3361     }
3362 
3363     if (bdrv_get_aio_context(target_bs) != aio_context) {
3364         if (!bdrv_has_blk(target_bs)) {
3365             /* The target BDS is not attached, we can safely move it to another
3366              * AioContext. */
3367             bdrv_set_aio_context(target_bs, aio_context);
3368         } else {
3369             error_setg(errp, "Target is attached to a different thread from "
3370                              "source.");
3371             goto out;
3372         }
3373     }
3374     job = backup_job_create(backup->job_id, bs, target_bs, backup->speed,
3375                             backup->sync, NULL, backup->compress,
3376                             backup->on_source_error, backup->on_target_error,
3377                             BLOCK_JOB_DEFAULT, NULL, NULL, txn, &local_err);
3378     if (local_err != NULL) {
3379         error_propagate(errp, local_err);
3380     }
3381 out:
3382     aio_context_release(aio_context);
3383     return job;
3384 }
3385 
3386 void qmp_blockdev_backup(BlockdevBackup *arg, Error **errp)
3387 {
3388     BlockJob *job;
3389     job = do_blockdev_backup(arg, NULL, errp);
3390     if (job) {
3391         block_job_start(job);
3392     }
3393 }
3394 
3395 /* Parameter check and block job starting for drive mirroring.
3396  * Caller should hold @device and @target's aio context (must be the same).
3397  **/
3398 static void blockdev_mirror_common(const char *job_id, BlockDriverState *bs,
3399                                    BlockDriverState *target,
3400                                    bool has_replaces, const char *replaces,
3401                                    enum MirrorSyncMode sync,
3402                                    BlockMirrorBackingMode backing_mode,
3403                                    bool has_speed, int64_t speed,
3404                                    bool has_granularity, uint32_t granularity,
3405                                    bool has_buf_size, int64_t buf_size,
3406                                    bool has_on_source_error,
3407                                    BlockdevOnError on_source_error,
3408                                    bool has_on_target_error,
3409                                    BlockdevOnError on_target_error,
3410                                    bool has_unmap, bool unmap,
3411                                    bool has_filter_node_name,
3412                                    const char *filter_node_name,
3413                                    Error **errp)
3414 {
3415 
3416     if (!has_speed) {
3417         speed = 0;
3418     }
3419     if (!has_on_source_error) {
3420         on_source_error = BLOCKDEV_ON_ERROR_REPORT;
3421     }
3422     if (!has_on_target_error) {
3423         on_target_error = BLOCKDEV_ON_ERROR_REPORT;
3424     }
3425     if (!has_granularity) {
3426         granularity = 0;
3427     }
3428     if (!has_buf_size) {
3429         buf_size = 0;
3430     }
3431     if (!has_unmap) {
3432         unmap = true;
3433     }
3434     if (!has_filter_node_name) {
3435         filter_node_name = NULL;
3436     }
3437 
3438     if (granularity != 0 && (granularity < 512 || granularity > 1048576 * 64)) {
3439         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "granularity",
3440                    "a value in range [512B, 64MB]");
3441         return;
3442     }
3443     if (granularity & (granularity - 1)) {
3444         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "granularity",
3445                    "power of 2");
3446         return;
3447     }
3448 
3449     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_MIRROR_SOURCE, errp)) {
3450         return;
3451     }
3452     if (bdrv_op_is_blocked(target, BLOCK_OP_TYPE_MIRROR_TARGET, errp)) {
3453         return;
3454     }
3455 
3456     if (!bs->backing && sync == MIRROR_SYNC_MODE_TOP) {
3457         sync = MIRROR_SYNC_MODE_FULL;
3458     }
3459 
3460     /* pass the node name to replace to mirror start since it's loose coupling
3461      * and will allow to check whether the node still exist at mirror completion
3462      */
3463     mirror_start(job_id, bs, target,
3464                  has_replaces ? replaces : NULL,
3465                  speed, granularity, buf_size, sync, backing_mode,
3466                  on_source_error, on_target_error, unmap, filter_node_name,
3467                  errp);
3468 }
3469 
3470 void qmp_drive_mirror(DriveMirror *arg, Error **errp)
3471 {
3472     BlockDriverState *bs;
3473     BlockDriverState *source, *target_bs;
3474     AioContext *aio_context;
3475     BlockMirrorBackingMode backing_mode;
3476     Error *local_err = NULL;
3477     QDict *options = NULL;
3478     int flags;
3479     int64_t size;
3480     const char *format = arg->format;
3481 
3482     bs = qmp_get_root_bs(arg->device, errp);
3483     if (!bs) {
3484         return;
3485     }
3486 
3487     aio_context = bdrv_get_aio_context(bs);
3488     aio_context_acquire(aio_context);
3489 
3490     if (!arg->has_mode) {
3491         arg->mode = NEW_IMAGE_MODE_ABSOLUTE_PATHS;
3492     }
3493 
3494     if (!arg->has_format) {
3495         format = (arg->mode == NEW_IMAGE_MODE_EXISTING
3496                   ? NULL : bs->drv->format_name);
3497     }
3498 
3499     flags = bs->open_flags | BDRV_O_RDWR;
3500     source = backing_bs(bs);
3501     if (!source && arg->sync == MIRROR_SYNC_MODE_TOP) {
3502         arg->sync = MIRROR_SYNC_MODE_FULL;
3503     }
3504     if (arg->sync == MIRROR_SYNC_MODE_NONE) {
3505         source = bs;
3506     }
3507 
3508     size = bdrv_getlength(bs);
3509     if (size < 0) {
3510         error_setg_errno(errp, -size, "bdrv_getlength failed");
3511         goto out;
3512     }
3513 
3514     if (arg->has_replaces) {
3515         BlockDriverState *to_replace_bs;
3516         AioContext *replace_aio_context;
3517         int64_t replace_size;
3518 
3519         if (!arg->has_node_name) {
3520             error_setg(errp, "a node-name must be provided when replacing a"
3521                              " named node of the graph");
3522             goto out;
3523         }
3524 
3525         to_replace_bs = check_to_replace_node(bs, arg->replaces, &local_err);
3526 
3527         if (!to_replace_bs) {
3528             error_propagate(errp, local_err);
3529             goto out;
3530         }
3531 
3532         replace_aio_context = bdrv_get_aio_context(to_replace_bs);
3533         aio_context_acquire(replace_aio_context);
3534         replace_size = bdrv_getlength(to_replace_bs);
3535         aio_context_release(replace_aio_context);
3536 
3537         if (size != replace_size) {
3538             error_setg(errp, "cannot replace image with a mirror image of "
3539                              "different size");
3540             goto out;
3541         }
3542     }
3543 
3544     if (arg->mode == NEW_IMAGE_MODE_ABSOLUTE_PATHS) {
3545         backing_mode = MIRROR_SOURCE_BACKING_CHAIN;
3546     } else {
3547         backing_mode = MIRROR_OPEN_BACKING_CHAIN;
3548     }
3549 
3550     /* Don't open backing image in create() */
3551     flags |= BDRV_O_NO_BACKING;
3552 
3553     if ((arg->sync == MIRROR_SYNC_MODE_FULL || !source)
3554         && arg->mode != NEW_IMAGE_MODE_EXISTING)
3555     {
3556         /* create new image w/o backing file */
3557         assert(format);
3558         bdrv_img_create(arg->target, format,
3559                         NULL, NULL, NULL, size, flags, false, &local_err);
3560     } else {
3561         switch (arg->mode) {
3562         case NEW_IMAGE_MODE_EXISTING:
3563             break;
3564         case NEW_IMAGE_MODE_ABSOLUTE_PATHS:
3565             /* create new image with backing file */
3566             bdrv_img_create(arg->target, format,
3567                             source->filename,
3568                             source->drv->format_name,
3569                             NULL, size, flags, false, &local_err);
3570             break;
3571         default:
3572             abort();
3573         }
3574     }
3575 
3576     if (local_err) {
3577         error_propagate(errp, local_err);
3578         goto out;
3579     }
3580 
3581     options = qdict_new();
3582     if (arg->has_node_name) {
3583         qdict_put_str(options, "node-name", arg->node_name);
3584     }
3585     if (format) {
3586         qdict_put_str(options, "driver", format);
3587     }
3588 
3589     /* Mirroring takes care of copy-on-write using the source's backing
3590      * file.
3591      */
3592     target_bs = bdrv_open(arg->target, NULL, options, flags, errp);
3593     if (!target_bs) {
3594         goto out;
3595     }
3596 
3597     bdrv_set_aio_context(target_bs, aio_context);
3598 
3599     blockdev_mirror_common(arg->has_job_id ? arg->job_id : NULL, bs, target_bs,
3600                            arg->has_replaces, arg->replaces, arg->sync,
3601                            backing_mode, arg->has_speed, arg->speed,
3602                            arg->has_granularity, arg->granularity,
3603                            arg->has_buf_size, arg->buf_size,
3604                            arg->has_on_source_error, arg->on_source_error,
3605                            arg->has_on_target_error, arg->on_target_error,
3606                            arg->has_unmap, arg->unmap,
3607                            false, NULL,
3608                            &local_err);
3609     bdrv_unref(target_bs);
3610     error_propagate(errp, local_err);
3611 out:
3612     aio_context_release(aio_context);
3613 }
3614 
3615 void qmp_blockdev_mirror(bool has_job_id, const char *job_id,
3616                          const char *device, const char *target,
3617                          bool has_replaces, const char *replaces,
3618                          MirrorSyncMode sync,
3619                          bool has_speed, int64_t speed,
3620                          bool has_granularity, uint32_t granularity,
3621                          bool has_buf_size, int64_t buf_size,
3622                          bool has_on_source_error,
3623                          BlockdevOnError on_source_error,
3624                          bool has_on_target_error,
3625                          BlockdevOnError on_target_error,
3626                          bool has_filter_node_name,
3627                          const char *filter_node_name,
3628                          Error **errp)
3629 {
3630     BlockDriverState *bs;
3631     BlockDriverState *target_bs;
3632     AioContext *aio_context;
3633     BlockMirrorBackingMode backing_mode = MIRROR_LEAVE_BACKING_CHAIN;
3634     Error *local_err = NULL;
3635 
3636     bs = qmp_get_root_bs(device, errp);
3637     if (!bs) {
3638         return;
3639     }
3640 
3641     target_bs = bdrv_lookup_bs(target, target, errp);
3642     if (!target_bs) {
3643         return;
3644     }
3645 
3646     aio_context = bdrv_get_aio_context(bs);
3647     aio_context_acquire(aio_context);
3648 
3649     bdrv_set_aio_context(target_bs, aio_context);
3650 
3651     blockdev_mirror_common(has_job_id ? job_id : NULL, bs, target_bs,
3652                            has_replaces, replaces, sync, backing_mode,
3653                            has_speed, speed,
3654                            has_granularity, granularity,
3655                            has_buf_size, buf_size,
3656                            has_on_source_error, on_source_error,
3657                            has_on_target_error, on_target_error,
3658                            true, true,
3659                            has_filter_node_name, filter_node_name,
3660                            &local_err);
3661     error_propagate(errp, local_err);
3662 
3663     aio_context_release(aio_context);
3664 }
3665 
3666 /* Get a block job using its ID and acquire its AioContext */
3667 static BlockJob *find_block_job(const char *id, AioContext **aio_context,
3668                                 Error **errp)
3669 {
3670     BlockJob *job;
3671 
3672     assert(id != NULL);
3673 
3674     *aio_context = NULL;
3675 
3676     job = block_job_get(id);
3677 
3678     if (!job) {
3679         error_set(errp, ERROR_CLASS_DEVICE_NOT_ACTIVE,
3680                   "Block job '%s' not found", id);
3681         return NULL;
3682     }
3683 
3684     *aio_context = blk_get_aio_context(job->blk);
3685     aio_context_acquire(*aio_context);
3686 
3687     return job;
3688 }
3689 
3690 void qmp_block_job_set_speed(const char *device, int64_t speed, Error **errp)
3691 {
3692     AioContext *aio_context;
3693     BlockJob *job = find_block_job(device, &aio_context, errp);
3694 
3695     if (!job) {
3696         return;
3697     }
3698 
3699     block_job_set_speed(job, speed, errp);
3700     aio_context_release(aio_context);
3701 }
3702 
3703 void qmp_block_job_cancel(const char *device,
3704                           bool has_force, bool force, Error **errp)
3705 {
3706     AioContext *aio_context;
3707     BlockJob *job = find_block_job(device, &aio_context, errp);
3708 
3709     if (!job) {
3710         return;
3711     }
3712 
3713     if (!has_force) {
3714         force = false;
3715     }
3716 
3717     if (block_job_user_paused(job) && !force) {
3718         error_setg(errp, "The block job for device '%s' is currently paused",
3719                    device);
3720         goto out;
3721     }
3722 
3723     trace_qmp_block_job_cancel(job);
3724     block_job_cancel(job);
3725 out:
3726     aio_context_release(aio_context);
3727 }
3728 
3729 void qmp_block_job_pause(const char *device, Error **errp)
3730 {
3731     AioContext *aio_context;
3732     BlockJob *job = find_block_job(device, &aio_context, errp);
3733 
3734     if (!job || block_job_user_paused(job)) {
3735         return;
3736     }
3737 
3738     trace_qmp_block_job_pause(job);
3739     block_job_user_pause(job);
3740     aio_context_release(aio_context);
3741 }
3742 
3743 void qmp_block_job_resume(const char *device, Error **errp)
3744 {
3745     AioContext *aio_context;
3746     BlockJob *job = find_block_job(device, &aio_context, errp);
3747 
3748     if (!job || !block_job_user_paused(job)) {
3749         return;
3750     }
3751 
3752     trace_qmp_block_job_resume(job);
3753     block_job_user_resume(job);
3754     aio_context_release(aio_context);
3755 }
3756 
3757 void qmp_block_job_complete(const char *device, Error **errp)
3758 {
3759     AioContext *aio_context;
3760     BlockJob *job = find_block_job(device, &aio_context, errp);
3761 
3762     if (!job) {
3763         return;
3764     }
3765 
3766     trace_qmp_block_job_complete(job);
3767     block_job_complete(job, errp);
3768     aio_context_release(aio_context);
3769 }
3770 
3771 void qmp_change_backing_file(const char *device,
3772                              const char *image_node_name,
3773                              const char *backing_file,
3774                              Error **errp)
3775 {
3776     BlockDriverState *bs = NULL;
3777     AioContext *aio_context;
3778     BlockDriverState *image_bs = NULL;
3779     Error *local_err = NULL;
3780     bool ro;
3781     int open_flags;
3782     int ret;
3783 
3784     bs = qmp_get_root_bs(device, errp);
3785     if (!bs) {
3786         return;
3787     }
3788 
3789     aio_context = bdrv_get_aio_context(bs);
3790     aio_context_acquire(aio_context);
3791 
3792     image_bs = bdrv_lookup_bs(NULL, image_node_name, &local_err);
3793     if (local_err) {
3794         error_propagate(errp, local_err);
3795         goto out;
3796     }
3797 
3798     if (!image_bs) {
3799         error_setg(errp, "image file not found");
3800         goto out;
3801     }
3802 
3803     if (bdrv_find_base(image_bs) == image_bs) {
3804         error_setg(errp, "not allowing backing file change on an image "
3805                          "without a backing file");
3806         goto out;
3807     }
3808 
3809     /* even though we are not necessarily operating on bs, we need it to
3810      * determine if block ops are currently prohibited on the chain */
3811     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_CHANGE, errp)) {
3812         goto out;
3813     }
3814 
3815     /* final sanity check */
3816     if (!bdrv_chain_contains(bs, image_bs)) {
3817         error_setg(errp, "'%s' and image file are not in the same chain",
3818                    device);
3819         goto out;
3820     }
3821 
3822     /* if not r/w, reopen to make r/w */
3823     open_flags = image_bs->open_flags;
3824     ro = bdrv_is_read_only(image_bs);
3825 
3826     if (ro) {
3827         bdrv_reopen(image_bs, open_flags | BDRV_O_RDWR, &local_err);
3828         if (local_err) {
3829             error_propagate(errp, local_err);
3830             goto out;
3831         }
3832     }
3833 
3834     ret = bdrv_change_backing_file(image_bs, backing_file,
3835                                image_bs->drv ? image_bs->drv->format_name : "");
3836 
3837     if (ret < 0) {
3838         error_setg_errno(errp, -ret, "Could not change backing file to '%s'",
3839                          backing_file);
3840         /* don't exit here, so we can try to restore open flags if
3841          * appropriate */
3842     }
3843 
3844     if (ro) {
3845         bdrv_reopen(image_bs, open_flags, &local_err);
3846         error_propagate(errp, local_err);
3847     }
3848 
3849 out:
3850     aio_context_release(aio_context);
3851 }
3852 
3853 void hmp_drive_add_node(Monitor *mon, const char *optstr)
3854 {
3855     QemuOpts *opts;
3856     QDict *qdict;
3857     Error *local_err = NULL;
3858 
3859     opts = qemu_opts_parse_noisily(&qemu_drive_opts, optstr, false);
3860     if (!opts) {
3861         return;
3862     }
3863 
3864     qdict = qemu_opts_to_qdict(opts, NULL);
3865 
3866     if (!qdict_get_try_str(qdict, "node-name")) {
3867         QDECREF(qdict);
3868         error_report("'node-name' needs to be specified");
3869         goto out;
3870     }
3871 
3872     BlockDriverState *bs = bds_tree_init(qdict, &local_err);
3873     if (!bs) {
3874         error_report_err(local_err);
3875         goto out;
3876     }
3877 
3878     QTAILQ_INSERT_TAIL(&monitor_bdrv_states, bs, monitor_list);
3879 
3880 out:
3881     qemu_opts_del(opts);
3882 }
3883 
3884 void qmp_blockdev_add(BlockdevOptions *options, Error **errp)
3885 {
3886     BlockDriverState *bs;
3887     QObject *obj;
3888     Visitor *v = qobject_output_visitor_new(&obj);
3889     QDict *qdict;
3890     const QDictEntry *ent;
3891     Error *local_err = NULL;
3892 
3893     visit_type_BlockdevOptions(v, NULL, &options, &local_err);
3894     if (local_err) {
3895         error_propagate(errp, local_err);
3896         goto fail;
3897     }
3898 
3899     visit_complete(v, &obj);
3900     qdict = qobject_to_qdict(obj);
3901 
3902     qdict_flatten(qdict);
3903 
3904     /*
3905      * Rewrite "backing": null to "backing": ""
3906      * TODO Rewrite "" to null instead, and perhaps not even here
3907      */
3908     for (ent = qdict_first(qdict); ent; ent = qdict_next(qdict, ent)) {
3909         char *dot = strrchr(ent->key, '.');
3910 
3911         if (!strcmp(dot ? dot + 1 : ent->key, "backing")
3912             && qobject_type(ent->value) == QTYPE_QNULL) {
3913             qdict_put(qdict, ent->key, qstring_new());
3914         }
3915     }
3916 
3917     if (!qdict_get_try_str(qdict, "node-name")) {
3918         error_setg(errp, "'node-name' must be specified for the root node");
3919         goto fail;
3920     }
3921 
3922     bs = bds_tree_init(qdict, errp);
3923     if (!bs) {
3924         goto fail;
3925     }
3926 
3927     QTAILQ_INSERT_TAIL(&monitor_bdrv_states, bs, monitor_list);
3928 
3929 fail:
3930     visit_free(v);
3931 }
3932 
3933 void qmp_blockdev_del(const char *node_name, Error **errp)
3934 {
3935     AioContext *aio_context;
3936     BlockDriverState *bs;
3937 
3938     bs = bdrv_find_node(node_name);
3939     if (!bs) {
3940         error_setg(errp, "Cannot find node %s", node_name);
3941         return;
3942     }
3943     if (bdrv_has_blk(bs)) {
3944         error_setg(errp, "Node %s is in use", node_name);
3945         return;
3946     }
3947     aio_context = bdrv_get_aio_context(bs);
3948     aio_context_acquire(aio_context);
3949 
3950     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_DRIVE_DEL, errp)) {
3951         goto out;
3952     }
3953 
3954     if (!bs->monitor_list.tqe_prev) {
3955         error_setg(errp, "Node %s is not owned by the monitor",
3956                    bs->node_name);
3957         goto out;
3958     }
3959 
3960     if (bs->refcnt > 1) {
3961         error_setg(errp, "Block device %s is in use",
3962                    bdrv_get_device_or_node_name(bs));
3963         goto out;
3964     }
3965 
3966     QTAILQ_REMOVE(&monitor_bdrv_states, bs, monitor_list);
3967     bdrv_unref(bs);
3968 
3969 out:
3970     aio_context_release(aio_context);
3971 }
3972 
3973 static BdrvChild *bdrv_find_child(BlockDriverState *parent_bs,
3974                                   const char *child_name)
3975 {
3976     BdrvChild *child;
3977 
3978     QLIST_FOREACH(child, &parent_bs->children, next) {
3979         if (strcmp(child->name, child_name) == 0) {
3980             return child;
3981         }
3982     }
3983 
3984     return NULL;
3985 }
3986 
3987 void qmp_x_blockdev_change(const char *parent, bool has_child,
3988                            const char *child, bool has_node,
3989                            const char *node, Error **errp)
3990 {
3991     BlockDriverState *parent_bs, *new_bs = NULL;
3992     BdrvChild *p_child;
3993 
3994     parent_bs = bdrv_lookup_bs(parent, parent, errp);
3995     if (!parent_bs) {
3996         return;
3997     }
3998 
3999     if (has_child == has_node) {
4000         if (has_child) {
4001             error_setg(errp, "The parameters child and node are in conflict");
4002         } else {
4003             error_setg(errp, "Either child or node must be specified");
4004         }
4005         return;
4006     }
4007 
4008     if (has_child) {
4009         p_child = bdrv_find_child(parent_bs, child);
4010         if (!p_child) {
4011             error_setg(errp, "Node '%s' does not have child '%s'",
4012                        parent, child);
4013             return;
4014         }
4015         bdrv_del_child(parent_bs, p_child, errp);
4016     }
4017 
4018     if (has_node) {
4019         new_bs = bdrv_find_node(node);
4020         if (!new_bs) {
4021             error_setg(errp, "Node '%s' not found", node);
4022             return;
4023         }
4024         bdrv_add_child(parent_bs, new_bs, errp);
4025     }
4026 }
4027 
4028 BlockJobInfoList *qmp_query_block_jobs(Error **errp)
4029 {
4030     BlockJobInfoList *head = NULL, **p_next = &head;
4031     BlockJob *job;
4032 
4033     for (job = block_job_next(NULL); job; job = block_job_next(job)) {
4034         BlockJobInfoList *elem;
4035         AioContext *aio_context;
4036 
4037         if (block_job_is_internal(job)) {
4038             continue;
4039         }
4040         elem = g_new0(BlockJobInfoList, 1);
4041         aio_context = blk_get_aio_context(job->blk);
4042         aio_context_acquire(aio_context);
4043         elem->value = block_job_query(job, errp);
4044         aio_context_release(aio_context);
4045         if (!elem->value) {
4046             g_free(elem);
4047             qapi_free_BlockJobInfoList(head);
4048             return NULL;
4049         }
4050         *p_next = elem;
4051         p_next = &elem->next;
4052     }
4053 
4054     return head;
4055 }
4056 
4057 QemuOptsList qemu_common_drive_opts = {
4058     .name = "drive",
4059     .head = QTAILQ_HEAD_INITIALIZER(qemu_common_drive_opts.head),
4060     .desc = {
4061         {
4062             .name = "snapshot",
4063             .type = QEMU_OPT_BOOL,
4064             .help = "enable/disable snapshot mode",
4065         },{
4066             .name = "aio",
4067             .type = QEMU_OPT_STRING,
4068             .help = "host AIO implementation (threads, native)",
4069         },{
4070             .name = BDRV_OPT_CACHE_WB,
4071             .type = QEMU_OPT_BOOL,
4072             .help = "Enable writeback mode",
4073         },{
4074             .name = "format",
4075             .type = QEMU_OPT_STRING,
4076             .help = "disk format (raw, qcow2, ...)",
4077         },{
4078             .name = "rerror",
4079             .type = QEMU_OPT_STRING,
4080             .help = "read error action",
4081         },{
4082             .name = "werror",
4083             .type = QEMU_OPT_STRING,
4084             .help = "write error action",
4085         },{
4086             .name = BDRV_OPT_READ_ONLY,
4087             .type = QEMU_OPT_BOOL,
4088             .help = "open drive file as read-only",
4089         },
4090 
4091         THROTTLE_OPTS,
4092 
4093         {
4094             .name = "throttling.group",
4095             .type = QEMU_OPT_STRING,
4096             .help = "name of the block throttling group",
4097         },{
4098             .name = "copy-on-read",
4099             .type = QEMU_OPT_BOOL,
4100             .help = "copy read data from backing file into image file",
4101         },{
4102             .name = "detect-zeroes",
4103             .type = QEMU_OPT_STRING,
4104             .help = "try to optimize zero writes (off, on, unmap)",
4105         },{
4106             .name = "stats-account-invalid",
4107             .type = QEMU_OPT_BOOL,
4108             .help = "whether to account for invalid I/O operations "
4109                     "in the statistics",
4110         },{
4111             .name = "stats-account-failed",
4112             .type = QEMU_OPT_BOOL,
4113             .help = "whether to account for failed I/O operations "
4114                     "in the statistics",
4115         },
4116         { /* end of list */ }
4117     },
4118 };
4119 
4120 QemuOptsList qemu_drive_opts = {
4121     .name = "drive",
4122     .head = QTAILQ_HEAD_INITIALIZER(qemu_drive_opts.head),
4123     .desc = {
4124         /*
4125          * no elements => accept any params
4126          * validation will happen later
4127          */
4128         { /* end of list */ }
4129     },
4130 };
4131