xref: /openbmc/qemu/blockdev.c (revision 3f53bc61)
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 "qemu/cutils.h"
54 #include "qemu/help_option.h"
55 #include "qemu/throttle-options.h"
56 
57 static QTAILQ_HEAD(, BlockDriverState) monitor_bdrv_states =
58     QTAILQ_HEAD_INITIALIZER(monitor_bdrv_states);
59 
60 static int do_open_tray(const char *blk_name, const char *qdev_id,
61                         bool force, Error **errp);
62 
63 static const char *const if_name[IF_COUNT] = {
64     [IF_NONE] = "none",
65     [IF_IDE] = "ide",
66     [IF_SCSI] = "scsi",
67     [IF_FLOPPY] = "floppy",
68     [IF_PFLASH] = "pflash",
69     [IF_MTD] = "mtd",
70     [IF_SD] = "sd",
71     [IF_VIRTIO] = "virtio",
72     [IF_XEN] = "xen",
73 };
74 
75 static int if_max_devs[IF_COUNT] = {
76     /*
77      * Do not change these numbers!  They govern how drive option
78      * index maps to unit and bus.  That mapping is ABI.
79      *
80      * All controllers used to implement if=T drives need to support
81      * if_max_devs[T] units, for any T with if_max_devs[T] != 0.
82      * Otherwise, some index values map to "impossible" bus, unit
83      * values.
84      *
85      * For instance, if you change [IF_SCSI] to 255, -drive
86      * if=scsi,index=12 no longer means bus=1,unit=5, but
87      * bus=0,unit=12.  With an lsi53c895a controller (7 units max),
88      * the drive can't be set up.  Regression.
89      */
90     [IF_IDE] = 2,
91     [IF_SCSI] = 7,
92 };
93 
94 /**
95  * Boards may call this to offer board-by-board overrides
96  * of the default, global values.
97  */
98 void override_max_devs(BlockInterfaceType type, int max_devs)
99 {
100     BlockBackend *blk;
101     DriveInfo *dinfo;
102 
103     if (max_devs <= 0) {
104         return;
105     }
106 
107     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
108         dinfo = blk_legacy_dinfo(blk);
109         if (dinfo->type == type) {
110             fprintf(stderr, "Cannot override units-per-bus property of"
111                     " the %s interface, because a drive of that type has"
112                     " already been added.\n", if_name[type]);
113             g_assert_not_reached();
114         }
115     }
116 
117     if_max_devs[type] = max_devs;
118 }
119 
120 /*
121  * We automatically delete the drive when a device using it gets
122  * unplugged.  Questionable feature, but we can't just drop it.
123  * Device models call blockdev_mark_auto_del() to schedule the
124  * automatic deletion, and generic qdev code calls blockdev_auto_del()
125  * when deletion is actually safe.
126  */
127 void blockdev_mark_auto_del(BlockBackend *blk)
128 {
129     DriveInfo *dinfo = blk_legacy_dinfo(blk);
130     BlockDriverState *bs = blk_bs(blk);
131     AioContext *aio_context;
132 
133     if (!dinfo) {
134         return;
135     }
136 
137     if (bs) {
138         aio_context = bdrv_get_aio_context(bs);
139         aio_context_acquire(aio_context);
140 
141         if (bs->job) {
142             block_job_cancel(bs->job);
143         }
144 
145         aio_context_release(aio_context);
146     }
147 
148     dinfo->auto_del = 1;
149 }
150 
151 void blockdev_auto_del(BlockBackend *blk)
152 {
153     DriveInfo *dinfo = blk_legacy_dinfo(blk);
154 
155     if (dinfo && dinfo->auto_del) {
156         monitor_remove_blk(blk);
157         blk_unref(blk);
158     }
159 }
160 
161 /**
162  * Returns the current mapping of how many units per bus
163  * a particular interface can support.
164  *
165  *  A positive integer indicates n units per bus.
166  *  0 implies the mapping has not been established.
167  * -1 indicates an invalid BlockInterfaceType was given.
168  */
169 int drive_get_max_devs(BlockInterfaceType type)
170 {
171     if (type >= IF_IDE && type < IF_COUNT) {
172         return if_max_devs[type];
173     }
174 
175     return -1;
176 }
177 
178 static int drive_index_to_bus_id(BlockInterfaceType type, int index)
179 {
180     int max_devs = if_max_devs[type];
181     return max_devs ? index / max_devs : 0;
182 }
183 
184 static int drive_index_to_unit_id(BlockInterfaceType type, int index)
185 {
186     int max_devs = if_max_devs[type];
187     return max_devs ? index % max_devs : index;
188 }
189 
190 QemuOpts *drive_def(const char *optstr)
191 {
192     return qemu_opts_parse_noisily(qemu_find_opts("drive"), optstr, false);
193 }
194 
195 QemuOpts *drive_add(BlockInterfaceType type, int index, const char *file,
196                     const char *optstr)
197 {
198     QemuOpts *opts;
199 
200     opts = drive_def(optstr);
201     if (!opts) {
202         return NULL;
203     }
204     if (type != IF_DEFAULT) {
205         qemu_opt_set(opts, "if", if_name[type], &error_abort);
206     }
207     if (index >= 0) {
208         qemu_opt_set_number(opts, "index", index, &error_abort);
209     }
210     if (file)
211         qemu_opt_set(opts, "file", file, &error_abort);
212     return opts;
213 }
214 
215 DriveInfo *drive_get(BlockInterfaceType type, int bus, int unit)
216 {
217     BlockBackend *blk;
218     DriveInfo *dinfo;
219 
220     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
221         dinfo = blk_legacy_dinfo(blk);
222         if (dinfo && dinfo->type == type
223             && dinfo->bus == bus && dinfo->unit == unit) {
224             return dinfo;
225         }
226     }
227 
228     return NULL;
229 }
230 
231 void drive_check_orphaned(void)
232 {
233     BlockBackend *blk;
234     DriveInfo *dinfo;
235     Location loc;
236     bool orphans = false;
237 
238     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
239         dinfo = blk_legacy_dinfo(blk);
240         if (!blk_get_attached_dev(blk) && !dinfo->is_default &&
241             dinfo->type != IF_NONE) {
242             loc_push_none(&loc);
243             qemu_opts_loc_restore(dinfo->opts);
244             error_report("machine type does not support"
245                          " if=%s,bus=%d,unit=%d",
246                          if_name[dinfo->type], dinfo->bus, dinfo->unit);
247             loc_pop(&loc);
248             orphans = true;
249         }
250     }
251 
252     if (orphans) {
253         exit(1);
254     }
255 }
256 
257 DriveInfo *drive_get_by_index(BlockInterfaceType type, int index)
258 {
259     return drive_get(type,
260                      drive_index_to_bus_id(type, index),
261                      drive_index_to_unit_id(type, index));
262 }
263 
264 int drive_get_max_bus(BlockInterfaceType type)
265 {
266     int max_bus;
267     BlockBackend *blk;
268     DriveInfo *dinfo;
269 
270     max_bus = -1;
271     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
272         dinfo = blk_legacy_dinfo(blk);
273         if (dinfo && dinfo->type == type && dinfo->bus > max_bus) {
274             max_bus = dinfo->bus;
275         }
276     }
277     return max_bus;
278 }
279 
280 /* Get a block device.  This should only be used for single-drive devices
281    (e.g. SD/Floppy/MTD).  Multi-disk devices (scsi/ide) should use the
282    appropriate bus.  */
283 DriveInfo *drive_get_next(BlockInterfaceType type)
284 {
285     static int next_block_unit[IF_COUNT];
286 
287     return drive_get(type, 0, next_block_unit[type]++);
288 }
289 
290 static void bdrv_format_print(void *opaque, const char *name)
291 {
292     error_printf(" %s", name);
293 }
294 
295 typedef struct {
296     QEMUBH *bh;
297     BlockDriverState *bs;
298 } BDRVPutRefBH;
299 
300 static int parse_block_error_action(const char *buf, bool is_read, Error **errp)
301 {
302     if (!strcmp(buf, "ignore")) {
303         return BLOCKDEV_ON_ERROR_IGNORE;
304     } else if (!is_read && !strcmp(buf, "enospc")) {
305         return BLOCKDEV_ON_ERROR_ENOSPC;
306     } else if (!strcmp(buf, "stop")) {
307         return BLOCKDEV_ON_ERROR_STOP;
308     } else if (!strcmp(buf, "report")) {
309         return BLOCKDEV_ON_ERROR_REPORT;
310     } else {
311         error_setg(errp, "'%s' invalid %s error action",
312                    buf, is_read ? "read" : "write");
313         return -1;
314     }
315 }
316 
317 static bool parse_stats_intervals(BlockAcctStats *stats, QList *intervals,
318                                   Error **errp)
319 {
320     const QListEntry *entry;
321     for (entry = qlist_first(intervals); entry; entry = qlist_next(entry)) {
322         switch (qobject_type(entry->value)) {
323 
324         case QTYPE_QSTRING: {
325             unsigned long long length;
326             const char *str = qstring_get_str(qobject_to_qstring(entry->value));
327             if (parse_uint_full(str, &length, 10) == 0 &&
328                 length > 0 && length <= UINT_MAX) {
329                 block_acct_add_interval(stats, (unsigned) length);
330             } else {
331                 error_setg(errp, "Invalid interval length: %s", str);
332                 return false;
333             }
334             break;
335         }
336 
337         case QTYPE_QINT: {
338             int64_t length = qint_get_int(qobject_to_qint(entry->value));
339             if (length > 0 && length <= UINT_MAX) {
340                 block_acct_add_interval(stats, (unsigned) length);
341             } else {
342                 error_setg(errp, "Invalid interval length: %" PRId64, length);
343                 return false;
344             }
345             break;
346         }
347 
348         default:
349             error_setg(errp, "The specification of stats-intervals is invalid");
350             return false;
351         }
352     }
353     return true;
354 }
355 
356 typedef enum { MEDIA_DISK, MEDIA_CDROM } DriveMediaType;
357 
358 /* All parameters but @opts are optional and may be set to NULL. */
359 static void extract_common_blockdev_options(QemuOpts *opts, int *bdrv_flags,
360     const char **throttling_group, ThrottleConfig *throttle_cfg,
361     BlockdevDetectZeroesOptions *detect_zeroes, Error **errp)
362 {
363     Error *local_error = NULL;
364     const char *aio;
365 
366     if (bdrv_flags) {
367         if (qemu_opt_get_bool(opts, "copy-on-read", false)) {
368             *bdrv_flags |= BDRV_O_COPY_ON_READ;
369         }
370 
371         if ((aio = qemu_opt_get(opts, "aio")) != NULL) {
372             if (!strcmp(aio, "native")) {
373                 *bdrv_flags |= BDRV_O_NATIVE_AIO;
374             } else if (!strcmp(aio, "threads")) {
375                 /* this is the default */
376             } else {
377                error_setg(errp, "invalid aio option");
378                return;
379             }
380         }
381     }
382 
383     /* disk I/O throttling */
384     if (throttling_group) {
385         *throttling_group = qemu_opt_get(opts, "throttling.group");
386     }
387 
388     if (throttle_cfg) {
389         throttle_config_init(throttle_cfg);
390         throttle_cfg->buckets[THROTTLE_BPS_TOTAL].avg =
391             qemu_opt_get_number(opts, "throttling.bps-total", 0);
392         throttle_cfg->buckets[THROTTLE_BPS_READ].avg  =
393             qemu_opt_get_number(opts, "throttling.bps-read", 0);
394         throttle_cfg->buckets[THROTTLE_BPS_WRITE].avg =
395             qemu_opt_get_number(opts, "throttling.bps-write", 0);
396         throttle_cfg->buckets[THROTTLE_OPS_TOTAL].avg =
397             qemu_opt_get_number(opts, "throttling.iops-total", 0);
398         throttle_cfg->buckets[THROTTLE_OPS_READ].avg =
399             qemu_opt_get_number(opts, "throttling.iops-read", 0);
400         throttle_cfg->buckets[THROTTLE_OPS_WRITE].avg =
401             qemu_opt_get_number(opts, "throttling.iops-write", 0);
402 
403         throttle_cfg->buckets[THROTTLE_BPS_TOTAL].max =
404             qemu_opt_get_number(opts, "throttling.bps-total-max", 0);
405         throttle_cfg->buckets[THROTTLE_BPS_READ].max  =
406             qemu_opt_get_number(opts, "throttling.bps-read-max", 0);
407         throttle_cfg->buckets[THROTTLE_BPS_WRITE].max =
408             qemu_opt_get_number(opts, "throttling.bps-write-max", 0);
409         throttle_cfg->buckets[THROTTLE_OPS_TOTAL].max =
410             qemu_opt_get_number(opts, "throttling.iops-total-max", 0);
411         throttle_cfg->buckets[THROTTLE_OPS_READ].max =
412             qemu_opt_get_number(opts, "throttling.iops-read-max", 0);
413         throttle_cfg->buckets[THROTTLE_OPS_WRITE].max =
414             qemu_opt_get_number(opts, "throttling.iops-write-max", 0);
415 
416         throttle_cfg->buckets[THROTTLE_BPS_TOTAL].burst_length =
417             qemu_opt_get_number(opts, "throttling.bps-total-max-length", 1);
418         throttle_cfg->buckets[THROTTLE_BPS_READ].burst_length  =
419             qemu_opt_get_number(opts, "throttling.bps-read-max-length", 1);
420         throttle_cfg->buckets[THROTTLE_BPS_WRITE].burst_length =
421             qemu_opt_get_number(opts, "throttling.bps-write-max-length", 1);
422         throttle_cfg->buckets[THROTTLE_OPS_TOTAL].burst_length =
423             qemu_opt_get_number(opts, "throttling.iops-total-max-length", 1);
424         throttle_cfg->buckets[THROTTLE_OPS_READ].burst_length =
425             qemu_opt_get_number(opts, "throttling.iops-read-max-length", 1);
426         throttle_cfg->buckets[THROTTLE_OPS_WRITE].burst_length =
427             qemu_opt_get_number(opts, "throttling.iops-write-max-length", 1);
428 
429         throttle_cfg->op_size =
430             qemu_opt_get_number(opts, "throttling.iops-size", 0);
431 
432         if (!throttle_is_valid(throttle_cfg, errp)) {
433             return;
434         }
435     }
436 
437     if (detect_zeroes) {
438         *detect_zeroes =
439             qapi_enum_parse(BlockdevDetectZeroesOptions_lookup,
440                             qemu_opt_get(opts, "detect-zeroes"),
441                             BLOCKDEV_DETECT_ZEROES_OPTIONS__MAX,
442                             BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF,
443                             &local_error);
444         if (local_error) {
445             error_propagate(errp, local_error);
446             return;
447         }
448     }
449 }
450 
451 /* Takes the ownership of bs_opts */
452 static BlockBackend *blockdev_init(const char *file, QDict *bs_opts,
453                                    Error **errp)
454 {
455     const char *buf;
456     int bdrv_flags = 0;
457     int on_read_error, on_write_error;
458     bool account_invalid, account_failed;
459     bool writethrough, read_only;
460     BlockBackend *blk;
461     BlockDriverState *bs;
462     ThrottleConfig cfg;
463     int snapshot = 0;
464     Error *error = NULL;
465     QemuOpts *opts;
466     QDict *interval_dict = NULL;
467     QList *interval_list = NULL;
468     const char *id;
469     BlockdevDetectZeroesOptions detect_zeroes =
470         BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF;
471     const char *throttling_group = NULL;
472 
473     /* Check common options by copying from bs_opts to opts, all other options
474      * stay in bs_opts for processing by bdrv_open(). */
475     id = qdict_get_try_str(bs_opts, "id");
476     opts = qemu_opts_create(&qemu_common_drive_opts, id, 1, &error);
477     if (error) {
478         error_propagate(errp, error);
479         goto err_no_opts;
480     }
481 
482     qemu_opts_absorb_qdict(opts, bs_opts, &error);
483     if (error) {
484         error_propagate(errp, error);
485         goto early_err;
486     }
487 
488     if (id) {
489         qdict_del(bs_opts, "id");
490     }
491 
492     /* extract parameters */
493     snapshot = qemu_opt_get_bool(opts, "snapshot", 0);
494 
495     account_invalid = qemu_opt_get_bool(opts, "stats-account-invalid", true);
496     account_failed = qemu_opt_get_bool(opts, "stats-account-failed", true);
497 
498     writethrough = !qemu_opt_get_bool(opts, BDRV_OPT_CACHE_WB, true);
499 
500     id = qemu_opts_id(opts);
501 
502     qdict_extract_subqdict(bs_opts, &interval_dict, "stats-intervals.");
503     qdict_array_split(interval_dict, &interval_list);
504 
505     if (qdict_size(interval_dict) != 0) {
506         error_setg(errp, "Invalid option stats-intervals.%s",
507                    qdict_first(interval_dict)->key);
508         goto early_err;
509     }
510 
511     extract_common_blockdev_options(opts, &bdrv_flags, &throttling_group, &cfg,
512                                     &detect_zeroes, &error);
513     if (error) {
514         error_propagate(errp, error);
515         goto early_err;
516     }
517 
518     if ((buf = qemu_opt_get(opts, "format")) != NULL) {
519         if (is_help_option(buf)) {
520             error_printf("Supported formats:");
521             bdrv_iterate_format(bdrv_format_print, NULL);
522             error_printf("\n");
523             goto early_err;
524         }
525 
526         if (qdict_haskey(bs_opts, "driver")) {
527             error_setg(errp, "Cannot specify both 'driver' and 'format'");
528             goto early_err;
529         }
530         qdict_put(bs_opts, "driver", qstring_from_str(buf));
531     }
532 
533     on_write_error = BLOCKDEV_ON_ERROR_ENOSPC;
534     if ((buf = qemu_opt_get(opts, "werror")) != NULL) {
535         on_write_error = parse_block_error_action(buf, 0, &error);
536         if (error) {
537             error_propagate(errp, error);
538             goto early_err;
539         }
540     }
541 
542     on_read_error = BLOCKDEV_ON_ERROR_REPORT;
543     if ((buf = qemu_opt_get(opts, "rerror")) != NULL) {
544         on_read_error = parse_block_error_action(buf, 1, &error);
545         if (error) {
546             error_propagate(errp, error);
547             goto early_err;
548         }
549     }
550 
551     if (snapshot) {
552         bdrv_flags |= BDRV_O_SNAPSHOT;
553     }
554 
555     read_only = qemu_opt_get_bool(opts, BDRV_OPT_READ_ONLY, false);
556 
557     /* init */
558     if ((!file || !*file) && !qdict_size(bs_opts)) {
559         BlockBackendRootState *blk_rs;
560 
561         blk = blk_new(0, BLK_PERM_ALL);
562         blk_rs = blk_get_root_state(blk);
563         blk_rs->open_flags    = bdrv_flags;
564         blk_rs->read_only     = read_only;
565         blk_rs->detect_zeroes = detect_zeroes;
566 
567         QDECREF(bs_opts);
568     } else {
569         if (file && !*file) {
570             file = NULL;
571         }
572 
573         /* bdrv_open() defaults to the values in bdrv_flags (for compatibility
574          * with other callers) rather than what we want as the real defaults.
575          * Apply the defaults here instead. */
576         qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_DIRECT, "off");
577         qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_NO_FLUSH, "off");
578         qdict_set_default_str(bs_opts, BDRV_OPT_READ_ONLY,
579                               read_only ? "on" : "off");
580         assert((bdrv_flags & BDRV_O_CACHE_MASK) == 0);
581 
582         if (runstate_check(RUN_STATE_INMIGRATE)) {
583             bdrv_flags |= BDRV_O_INACTIVE;
584         }
585 
586         blk = blk_new_open(file, NULL, bs_opts, bdrv_flags, errp);
587         if (!blk) {
588             goto err_no_bs_opts;
589         }
590         bs = blk_bs(blk);
591 
592         bs->detect_zeroes = detect_zeroes;
593 
594         if (bdrv_key_required(bs)) {
595             autostart = 0;
596         }
597 
598         block_acct_init(blk_get_stats(blk), account_invalid, account_failed);
599 
600         if (!parse_stats_intervals(blk_get_stats(blk), interval_list, errp)) {
601             blk_unref(blk);
602             blk = NULL;
603             goto err_no_bs_opts;
604         }
605     }
606 
607     /* disk I/O throttling */
608     if (throttle_enabled(&cfg)) {
609         if (!throttling_group) {
610             throttling_group = id;
611         }
612         blk_io_limits_enable(blk, throttling_group);
613         blk_set_io_limits(blk, &cfg);
614     }
615 
616     blk_set_enable_write_cache(blk, !writethrough);
617     blk_set_on_error(blk, on_read_error, on_write_error);
618 
619     if (!monitor_add_blk(blk, id, errp)) {
620         blk_unref(blk);
621         blk = NULL;
622         goto err_no_bs_opts;
623     }
624 
625 err_no_bs_opts:
626     qemu_opts_del(opts);
627     QDECREF(interval_dict);
628     QDECREF(interval_list);
629     return blk;
630 
631 early_err:
632     qemu_opts_del(opts);
633     QDECREF(interval_dict);
634     QDECREF(interval_list);
635 err_no_opts:
636     QDECREF(bs_opts);
637     return NULL;
638 }
639 
640 /* Takes the ownership of bs_opts */
641 static BlockDriverState *bds_tree_init(QDict *bs_opts, Error **errp)
642 {
643     int bdrv_flags = 0;
644 
645     /* bdrv_open() defaults to the values in bdrv_flags (for compatibility
646      * with other callers) rather than what we want as the real defaults.
647      * Apply the defaults here instead. */
648     qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_DIRECT, "off");
649     qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_NO_FLUSH, "off");
650     qdict_set_default_str(bs_opts, BDRV_OPT_READ_ONLY, "off");
651 
652     if (runstate_check(RUN_STATE_INMIGRATE)) {
653         bdrv_flags |= BDRV_O_INACTIVE;
654     }
655 
656     return bdrv_open(NULL, NULL, bs_opts, bdrv_flags, errp);
657 }
658 
659 void blockdev_close_all_bdrv_states(void)
660 {
661     BlockDriverState *bs, *next_bs;
662 
663     QTAILQ_FOREACH_SAFE(bs, &monitor_bdrv_states, monitor_list, next_bs) {
664         AioContext *ctx = bdrv_get_aio_context(bs);
665 
666         aio_context_acquire(ctx);
667         bdrv_unref(bs);
668         aio_context_release(ctx);
669     }
670 }
671 
672 /* Iterates over the list of monitor-owned BlockDriverStates */
673 BlockDriverState *bdrv_next_monitor_owned(BlockDriverState *bs)
674 {
675     return bs ? QTAILQ_NEXT(bs, monitor_list)
676               : QTAILQ_FIRST(&monitor_bdrv_states);
677 }
678 
679 static void qemu_opt_rename(QemuOpts *opts, const char *from, const char *to,
680                             Error **errp)
681 {
682     const char *value;
683 
684     value = qemu_opt_get(opts, from);
685     if (value) {
686         if (qemu_opt_find(opts, to)) {
687             error_setg(errp, "'%s' and its alias '%s' can't be used at the "
688                        "same time", to, from);
689             return;
690         }
691     }
692 
693     /* rename all items in opts */
694     while ((value = qemu_opt_get(opts, from))) {
695         qemu_opt_set(opts, to, value, &error_abort);
696         qemu_opt_unset(opts, from);
697     }
698 }
699 
700 QemuOptsList qemu_legacy_drive_opts = {
701     .name = "drive",
702     .head = QTAILQ_HEAD_INITIALIZER(qemu_legacy_drive_opts.head),
703     .desc = {
704         {
705             .name = "bus",
706             .type = QEMU_OPT_NUMBER,
707             .help = "bus number",
708         },{
709             .name = "unit",
710             .type = QEMU_OPT_NUMBER,
711             .help = "unit number (i.e. lun for scsi)",
712         },{
713             .name = "index",
714             .type = QEMU_OPT_NUMBER,
715             .help = "index number",
716         },{
717             .name = "media",
718             .type = QEMU_OPT_STRING,
719             .help = "media type (disk, cdrom)",
720         },{
721             .name = "if",
722             .type = QEMU_OPT_STRING,
723             .help = "interface (ide, scsi, sd, mtd, floppy, pflash, virtio)",
724         },{
725             .name = "cyls",
726             .type = QEMU_OPT_NUMBER,
727             .help = "number of cylinders (ide disk geometry)",
728         },{
729             .name = "heads",
730             .type = QEMU_OPT_NUMBER,
731             .help = "number of heads (ide disk geometry)",
732         },{
733             .name = "secs",
734             .type = QEMU_OPT_NUMBER,
735             .help = "number of sectors (ide disk geometry)",
736         },{
737             .name = "trans",
738             .type = QEMU_OPT_STRING,
739             .help = "chs translation (auto, lba, none)",
740         },{
741             .name = "boot",
742             .type = QEMU_OPT_BOOL,
743             .help = "(deprecated, ignored)",
744         },{
745             .name = "addr",
746             .type = QEMU_OPT_STRING,
747             .help = "pci address (virtio only)",
748         },{
749             .name = "serial",
750             .type = QEMU_OPT_STRING,
751             .help = "disk serial number",
752         },{
753             .name = "file",
754             .type = QEMU_OPT_STRING,
755             .help = "file name",
756         },
757 
758         /* Options that are passed on, but have special semantics with -drive */
759         {
760             .name = BDRV_OPT_READ_ONLY,
761             .type = QEMU_OPT_BOOL,
762             .help = "open drive file as read-only",
763         },{
764             .name = "rerror",
765             .type = QEMU_OPT_STRING,
766             .help = "read error action",
767         },{
768             .name = "werror",
769             .type = QEMU_OPT_STRING,
770             .help = "write error action",
771         },{
772             .name = "copy-on-read",
773             .type = QEMU_OPT_BOOL,
774             .help = "copy read data from backing file into image file",
775         },
776 
777         { /* end of list */ }
778     },
779 };
780 
781 DriveInfo *drive_new(QemuOpts *all_opts, BlockInterfaceType block_default_type)
782 {
783     const char *value;
784     BlockBackend *blk;
785     DriveInfo *dinfo = NULL;
786     QDict *bs_opts;
787     QemuOpts *legacy_opts;
788     DriveMediaType media = MEDIA_DISK;
789     BlockInterfaceType type;
790     int cyls, heads, secs, translation;
791     int max_devs, bus_id, unit_id, index;
792     const char *devaddr;
793     const char *werror, *rerror;
794     bool read_only = false;
795     bool copy_on_read;
796     const char *serial;
797     const char *filename;
798     Error *local_err = NULL;
799     int i;
800 
801     /* Change legacy command line options into QMP ones */
802     static const struct {
803         const char *from;
804         const char *to;
805     } opt_renames[] = {
806         { "iops",           "throttling.iops-total" },
807         { "iops_rd",        "throttling.iops-read" },
808         { "iops_wr",        "throttling.iops-write" },
809 
810         { "bps",            "throttling.bps-total" },
811         { "bps_rd",         "throttling.bps-read" },
812         { "bps_wr",         "throttling.bps-write" },
813 
814         { "iops_max",       "throttling.iops-total-max" },
815         { "iops_rd_max",    "throttling.iops-read-max" },
816         { "iops_wr_max",    "throttling.iops-write-max" },
817 
818         { "bps_max",        "throttling.bps-total-max" },
819         { "bps_rd_max",     "throttling.bps-read-max" },
820         { "bps_wr_max",     "throttling.bps-write-max" },
821 
822         { "iops_size",      "throttling.iops-size" },
823 
824         { "group",          "throttling.group" },
825 
826         { "readonly",       BDRV_OPT_READ_ONLY },
827     };
828 
829     for (i = 0; i < ARRAY_SIZE(opt_renames); i++) {
830         qemu_opt_rename(all_opts, opt_renames[i].from, opt_renames[i].to,
831                         &local_err);
832         if (local_err) {
833             error_report_err(local_err);
834             return NULL;
835         }
836     }
837 
838     value = qemu_opt_get(all_opts, "cache");
839     if (value) {
840         int flags = 0;
841         bool writethrough;
842 
843         if (bdrv_parse_cache_mode(value, &flags, &writethrough) != 0) {
844             error_report("invalid cache option");
845             return NULL;
846         }
847 
848         /* Specific options take precedence */
849         if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_WB)) {
850             qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_WB,
851                               !writethrough, &error_abort);
852         }
853         if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_DIRECT)) {
854             qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_DIRECT,
855                               !!(flags & BDRV_O_NOCACHE), &error_abort);
856         }
857         if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_NO_FLUSH)) {
858             qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_NO_FLUSH,
859                               !!(flags & BDRV_O_NO_FLUSH), &error_abort);
860         }
861         qemu_opt_unset(all_opts, "cache");
862     }
863 
864     /* Get a QDict for processing the options */
865     bs_opts = qdict_new();
866     qemu_opts_to_qdict(all_opts, bs_opts);
867 
868     legacy_opts = qemu_opts_create(&qemu_legacy_drive_opts, NULL, 0,
869                                    &error_abort);
870     qemu_opts_absorb_qdict(legacy_opts, bs_opts, &local_err);
871     if (local_err) {
872         error_report_err(local_err);
873         goto fail;
874     }
875 
876     /* Deprecated option boot=[on|off] */
877     if (qemu_opt_get(legacy_opts, "boot") != NULL) {
878         fprintf(stderr, "qemu-kvm: boot=on|off is deprecated and will be "
879                 "ignored. Future versions will reject this parameter. Please "
880                 "update your scripts.\n");
881     }
882 
883     /* Media type */
884     value = qemu_opt_get(legacy_opts, "media");
885     if (value) {
886         if (!strcmp(value, "disk")) {
887             media = MEDIA_DISK;
888         } else if (!strcmp(value, "cdrom")) {
889             media = MEDIA_CDROM;
890             read_only = true;
891         } else {
892             error_report("'%s' invalid media", value);
893             goto fail;
894         }
895     }
896 
897     /* copy-on-read is disabled with a warning for read-only devices */
898     read_only |= qemu_opt_get_bool(legacy_opts, BDRV_OPT_READ_ONLY, false);
899     copy_on_read = qemu_opt_get_bool(legacy_opts, "copy-on-read", false);
900 
901     if (read_only && copy_on_read) {
902         error_report("warning: disabling copy-on-read on read-only drive");
903         copy_on_read = false;
904     }
905 
906     qdict_put(bs_opts, BDRV_OPT_READ_ONLY,
907               qstring_from_str(read_only ? "on" : "off"));
908     qdict_put(bs_opts, "copy-on-read",
909               qstring_from_str(copy_on_read ? "on" :"off"));
910 
911     /* Controller type */
912     value = qemu_opt_get(legacy_opts, "if");
913     if (value) {
914         for (type = 0;
915              type < IF_COUNT && strcmp(value, if_name[type]);
916              type++) {
917         }
918         if (type == IF_COUNT) {
919             error_report("unsupported bus type '%s'", value);
920             goto fail;
921         }
922     } else {
923         type = block_default_type;
924     }
925 
926     /* Geometry */
927     cyls  = qemu_opt_get_number(legacy_opts, "cyls", 0);
928     heads = qemu_opt_get_number(legacy_opts, "heads", 0);
929     secs  = qemu_opt_get_number(legacy_opts, "secs", 0);
930 
931     if (cyls || heads || secs) {
932         if (cyls < 1) {
933             error_report("invalid physical cyls number");
934             goto fail;
935         }
936         if (heads < 1) {
937             error_report("invalid physical heads number");
938             goto fail;
939         }
940         if (secs < 1) {
941             error_report("invalid physical secs number");
942             goto fail;
943         }
944     }
945 
946     translation = BIOS_ATA_TRANSLATION_AUTO;
947     value = qemu_opt_get(legacy_opts, "trans");
948     if (value != NULL) {
949         if (!cyls) {
950             error_report("'%s' trans must be used with cyls, heads and secs",
951                          value);
952             goto fail;
953         }
954         if (!strcmp(value, "none")) {
955             translation = BIOS_ATA_TRANSLATION_NONE;
956         } else if (!strcmp(value, "lba")) {
957             translation = BIOS_ATA_TRANSLATION_LBA;
958         } else if (!strcmp(value, "large")) {
959             translation = BIOS_ATA_TRANSLATION_LARGE;
960         } else if (!strcmp(value, "rechs")) {
961             translation = BIOS_ATA_TRANSLATION_RECHS;
962         } else if (!strcmp(value, "auto")) {
963             translation = BIOS_ATA_TRANSLATION_AUTO;
964         } else {
965             error_report("'%s' invalid translation type", value);
966             goto fail;
967         }
968     }
969 
970     if (media == MEDIA_CDROM) {
971         if (cyls || secs || heads) {
972             error_report("CHS can't be set with media=cdrom");
973             goto fail;
974         }
975     }
976 
977     /* Device address specified by bus/unit or index.
978      * If none was specified, try to find the first free one. */
979     bus_id  = qemu_opt_get_number(legacy_opts, "bus", 0);
980     unit_id = qemu_opt_get_number(legacy_opts, "unit", -1);
981     index   = qemu_opt_get_number(legacy_opts, "index", -1);
982 
983     max_devs = if_max_devs[type];
984 
985     if (index != -1) {
986         if (bus_id != 0 || unit_id != -1) {
987             error_report("index cannot be used with bus and unit");
988             goto fail;
989         }
990         bus_id = drive_index_to_bus_id(type, index);
991         unit_id = drive_index_to_unit_id(type, index);
992     }
993 
994     if (unit_id == -1) {
995        unit_id = 0;
996        while (drive_get(type, bus_id, unit_id) != NULL) {
997            unit_id++;
998            if (max_devs && unit_id >= max_devs) {
999                unit_id -= max_devs;
1000                bus_id++;
1001            }
1002        }
1003     }
1004 
1005     if (max_devs && unit_id >= max_devs) {
1006         error_report("unit %d too big (max is %d)", unit_id, max_devs - 1);
1007         goto fail;
1008     }
1009 
1010     if (drive_get(type, bus_id, unit_id) != NULL) {
1011         error_report("drive with bus=%d, unit=%d (index=%d) exists",
1012                      bus_id, unit_id, index);
1013         goto fail;
1014     }
1015 
1016     /* Serial number */
1017     serial = qemu_opt_get(legacy_opts, "serial");
1018 
1019     /* no id supplied -> create one */
1020     if (qemu_opts_id(all_opts) == NULL) {
1021         char *new_id;
1022         const char *mediastr = "";
1023         if (type == IF_IDE || type == IF_SCSI) {
1024             mediastr = (media == MEDIA_CDROM) ? "-cd" : "-hd";
1025         }
1026         if (max_devs) {
1027             new_id = g_strdup_printf("%s%i%s%i", if_name[type], bus_id,
1028                                      mediastr, unit_id);
1029         } else {
1030             new_id = g_strdup_printf("%s%s%i", if_name[type],
1031                                      mediastr, unit_id);
1032         }
1033         qdict_put(bs_opts, "id", qstring_from_str(new_id));
1034         g_free(new_id);
1035     }
1036 
1037     /* Add virtio block device */
1038     devaddr = qemu_opt_get(legacy_opts, "addr");
1039     if (devaddr && type != IF_VIRTIO) {
1040         error_report("addr is not supported by this bus type");
1041         goto fail;
1042     }
1043 
1044     if (type == IF_VIRTIO) {
1045         QemuOpts *devopts;
1046         devopts = qemu_opts_create(qemu_find_opts("device"), NULL, 0,
1047                                    &error_abort);
1048         if (arch_type == QEMU_ARCH_S390X) {
1049             qemu_opt_set(devopts, "driver", "virtio-blk-ccw", &error_abort);
1050         } else {
1051             qemu_opt_set(devopts, "driver", "virtio-blk-pci", &error_abort);
1052         }
1053         qemu_opt_set(devopts, "drive", qdict_get_str(bs_opts, "id"),
1054                      &error_abort);
1055         if (devaddr) {
1056             qemu_opt_set(devopts, "addr", devaddr, &error_abort);
1057         }
1058     }
1059 
1060     filename = qemu_opt_get(legacy_opts, "file");
1061 
1062     /* Check werror/rerror compatibility with if=... */
1063     werror = qemu_opt_get(legacy_opts, "werror");
1064     if (werror != NULL) {
1065         if (type != IF_IDE && type != IF_SCSI && type != IF_VIRTIO &&
1066             type != IF_NONE) {
1067             error_report("werror is not supported by this bus type");
1068             goto fail;
1069         }
1070         qdict_put(bs_opts, "werror", qstring_from_str(werror));
1071     }
1072 
1073     rerror = qemu_opt_get(legacy_opts, "rerror");
1074     if (rerror != NULL) {
1075         if (type != IF_IDE && type != IF_VIRTIO && type != IF_SCSI &&
1076             type != IF_NONE) {
1077             error_report("rerror is not supported by this bus type");
1078             goto fail;
1079         }
1080         qdict_put(bs_opts, "rerror", qstring_from_str(rerror));
1081     }
1082 
1083     /* Actual block device init: Functionality shared with blockdev-add */
1084     blk = blockdev_init(filename, bs_opts, &local_err);
1085     bs_opts = NULL;
1086     if (!blk) {
1087         if (local_err) {
1088             error_report_err(local_err);
1089         }
1090         goto fail;
1091     } else {
1092         assert(!local_err);
1093     }
1094 
1095     /* Create legacy DriveInfo */
1096     dinfo = g_malloc0(sizeof(*dinfo));
1097     dinfo->opts = all_opts;
1098 
1099     dinfo->cyls = cyls;
1100     dinfo->heads = heads;
1101     dinfo->secs = secs;
1102     dinfo->trans = translation;
1103 
1104     dinfo->type = type;
1105     dinfo->bus = bus_id;
1106     dinfo->unit = unit_id;
1107     dinfo->devaddr = devaddr;
1108     dinfo->serial = g_strdup(serial);
1109 
1110     blk_set_legacy_dinfo(blk, dinfo);
1111 
1112     switch(type) {
1113     case IF_IDE:
1114     case IF_SCSI:
1115     case IF_XEN:
1116     case IF_NONE:
1117         dinfo->media_cd = media == MEDIA_CDROM;
1118         break;
1119     default:
1120         break;
1121     }
1122 
1123 fail:
1124     qemu_opts_del(legacy_opts);
1125     QDECREF(bs_opts);
1126     return dinfo;
1127 }
1128 
1129 static BlockDriverState *qmp_get_root_bs(const char *name, Error **errp)
1130 {
1131     BlockDriverState *bs;
1132 
1133     bs = bdrv_lookup_bs(name, name, errp);
1134     if (bs == NULL) {
1135         return NULL;
1136     }
1137 
1138     if (!bdrv_is_root_node(bs)) {
1139         error_setg(errp, "Need a root block node");
1140         return NULL;
1141     }
1142 
1143     if (!bdrv_is_inserted(bs)) {
1144         error_setg(errp, "Device has no medium");
1145         return NULL;
1146     }
1147 
1148     return bs;
1149 }
1150 
1151 static BlockBackend *qmp_get_blk(const char *blk_name, const char *qdev_id,
1152                                  Error **errp)
1153 {
1154     BlockBackend *blk;
1155 
1156     if (!blk_name == !qdev_id) {
1157         error_setg(errp, "Need exactly one of 'device' and 'id'");
1158         return NULL;
1159     }
1160 
1161     if (qdev_id) {
1162         blk = blk_by_qdev_id(qdev_id, errp);
1163     } else {
1164         blk = blk_by_name(blk_name);
1165         if (blk == NULL) {
1166             error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
1167                       "Device '%s' not found", blk_name);
1168         }
1169     }
1170 
1171     return blk;
1172 }
1173 
1174 void hmp_commit(Monitor *mon, const QDict *qdict)
1175 {
1176     const char *device = qdict_get_str(qdict, "device");
1177     BlockBackend *blk;
1178     int ret;
1179 
1180     if (!strcmp(device, "all")) {
1181         ret = blk_commit_all();
1182     } else {
1183         BlockDriverState *bs;
1184         AioContext *aio_context;
1185 
1186         blk = blk_by_name(device);
1187         if (!blk) {
1188             monitor_printf(mon, "Device '%s' not found\n", device);
1189             return;
1190         }
1191         if (!blk_is_available(blk)) {
1192             monitor_printf(mon, "Device '%s' has no medium\n", device);
1193             return;
1194         }
1195 
1196         bs = blk_bs(blk);
1197         aio_context = bdrv_get_aio_context(bs);
1198         aio_context_acquire(aio_context);
1199 
1200         ret = bdrv_commit(bs);
1201 
1202         aio_context_release(aio_context);
1203     }
1204     if (ret < 0) {
1205         monitor_printf(mon, "'commit' error for '%s': %s\n", device,
1206                        strerror(-ret));
1207     }
1208 }
1209 
1210 static void blockdev_do_action(TransactionAction *action, Error **errp)
1211 {
1212     TransactionActionList list;
1213 
1214     list.value = action;
1215     list.next = NULL;
1216     qmp_transaction(&list, false, NULL, errp);
1217 }
1218 
1219 void qmp_blockdev_snapshot_sync(bool has_device, const char *device,
1220                                 bool has_node_name, const char *node_name,
1221                                 const char *snapshot_file,
1222                                 bool has_snapshot_node_name,
1223                                 const char *snapshot_node_name,
1224                                 bool has_format, const char *format,
1225                                 bool has_mode, NewImageMode mode, Error **errp)
1226 {
1227     BlockdevSnapshotSync snapshot = {
1228         .has_device = has_device,
1229         .device = (char *) device,
1230         .has_node_name = has_node_name,
1231         .node_name = (char *) node_name,
1232         .snapshot_file = (char *) snapshot_file,
1233         .has_snapshot_node_name = has_snapshot_node_name,
1234         .snapshot_node_name = (char *) snapshot_node_name,
1235         .has_format = has_format,
1236         .format = (char *) format,
1237         .has_mode = has_mode,
1238         .mode = mode,
1239     };
1240     TransactionAction action = {
1241         .type = TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC,
1242         .u.blockdev_snapshot_sync.data = &snapshot,
1243     };
1244     blockdev_do_action(&action, errp);
1245 }
1246 
1247 void qmp_blockdev_snapshot(const char *node, const char *overlay,
1248                            Error **errp)
1249 {
1250     BlockdevSnapshot snapshot_data = {
1251         .node = (char *) node,
1252         .overlay = (char *) overlay
1253     };
1254     TransactionAction action = {
1255         .type = TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT,
1256         .u.blockdev_snapshot.data = &snapshot_data,
1257     };
1258     blockdev_do_action(&action, errp);
1259 }
1260 
1261 void qmp_blockdev_snapshot_internal_sync(const char *device,
1262                                          const char *name,
1263                                          Error **errp)
1264 {
1265     BlockdevSnapshotInternal snapshot = {
1266         .device = (char *) device,
1267         .name = (char *) name
1268     };
1269     TransactionAction action = {
1270         .type = TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC,
1271         .u.blockdev_snapshot_internal_sync.data = &snapshot,
1272     };
1273     blockdev_do_action(&action, errp);
1274 }
1275 
1276 SnapshotInfo *qmp_blockdev_snapshot_delete_internal_sync(const char *device,
1277                                                          bool has_id,
1278                                                          const char *id,
1279                                                          bool has_name,
1280                                                          const char *name,
1281                                                          Error **errp)
1282 {
1283     BlockDriverState *bs;
1284     AioContext *aio_context;
1285     QEMUSnapshotInfo sn;
1286     Error *local_err = NULL;
1287     SnapshotInfo *info = NULL;
1288     int ret;
1289 
1290     bs = qmp_get_root_bs(device, errp);
1291     if (!bs) {
1292         return NULL;
1293     }
1294     aio_context = bdrv_get_aio_context(bs);
1295     aio_context_acquire(aio_context);
1296 
1297     if (!has_id) {
1298         id = NULL;
1299     }
1300 
1301     if (!has_name) {
1302         name = NULL;
1303     }
1304 
1305     if (!id && !name) {
1306         error_setg(errp, "Name or id must be provided");
1307         goto out_aio_context;
1308     }
1309 
1310     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_INTERNAL_SNAPSHOT_DELETE, errp)) {
1311         goto out_aio_context;
1312     }
1313 
1314     ret = bdrv_snapshot_find_by_id_and_name(bs, id, name, &sn, &local_err);
1315     if (local_err) {
1316         error_propagate(errp, local_err);
1317         goto out_aio_context;
1318     }
1319     if (!ret) {
1320         error_setg(errp,
1321                    "Snapshot with id '%s' and name '%s' does not exist on "
1322                    "device '%s'",
1323                    STR_OR_NULL(id), STR_OR_NULL(name), device);
1324         goto out_aio_context;
1325     }
1326 
1327     bdrv_snapshot_delete(bs, id, name, &local_err);
1328     if (local_err) {
1329         error_propagate(errp, local_err);
1330         goto out_aio_context;
1331     }
1332 
1333     aio_context_release(aio_context);
1334 
1335     info = g_new0(SnapshotInfo, 1);
1336     info->id = g_strdup(sn.id_str);
1337     info->name = g_strdup(sn.name);
1338     info->date_nsec = sn.date_nsec;
1339     info->date_sec = sn.date_sec;
1340     info->vm_state_size = sn.vm_state_size;
1341     info->vm_clock_nsec = sn.vm_clock_nsec % 1000000000;
1342     info->vm_clock_sec = sn.vm_clock_nsec / 1000000000;
1343 
1344     return info;
1345 
1346 out_aio_context:
1347     aio_context_release(aio_context);
1348     return NULL;
1349 }
1350 
1351 /**
1352  * block_dirty_bitmap_lookup:
1353  * Return a dirty bitmap (if present), after validating
1354  * the node reference and bitmap names.
1355  *
1356  * @node: The name of the BDS node to search for bitmaps
1357  * @name: The name of the bitmap to search for
1358  * @pbs: Output pointer for BDS lookup, if desired. Can be NULL.
1359  * @paio: Output pointer for aio_context acquisition, if desired. Can be NULL.
1360  * @errp: Output pointer for error information. Can be NULL.
1361  *
1362  * @return: A bitmap object on success, or NULL on failure.
1363  */
1364 static BdrvDirtyBitmap *block_dirty_bitmap_lookup(const char *node,
1365                                                   const char *name,
1366                                                   BlockDriverState **pbs,
1367                                                   AioContext **paio,
1368                                                   Error **errp)
1369 {
1370     BlockDriverState *bs;
1371     BdrvDirtyBitmap *bitmap;
1372     AioContext *aio_context;
1373 
1374     if (!node) {
1375         error_setg(errp, "Node cannot be NULL");
1376         return NULL;
1377     }
1378     if (!name) {
1379         error_setg(errp, "Bitmap name cannot be NULL");
1380         return NULL;
1381     }
1382     bs = bdrv_lookup_bs(node, node, NULL);
1383     if (!bs) {
1384         error_setg(errp, "Node '%s' not found", node);
1385         return NULL;
1386     }
1387 
1388     aio_context = bdrv_get_aio_context(bs);
1389     aio_context_acquire(aio_context);
1390 
1391     bitmap = bdrv_find_dirty_bitmap(bs, name);
1392     if (!bitmap) {
1393         error_setg(errp, "Dirty bitmap '%s' not found", name);
1394         goto fail;
1395     }
1396 
1397     if (pbs) {
1398         *pbs = bs;
1399     }
1400     if (paio) {
1401         *paio = aio_context;
1402     } else {
1403         aio_context_release(aio_context);
1404     }
1405 
1406     return bitmap;
1407 
1408  fail:
1409     aio_context_release(aio_context);
1410     return NULL;
1411 }
1412 
1413 /* New and old BlockDriverState structs for atomic group operations */
1414 
1415 typedef struct BlkActionState BlkActionState;
1416 
1417 /**
1418  * BlkActionOps:
1419  * Table of operations that define an Action.
1420  *
1421  * @instance_size: Size of state struct, in bytes.
1422  * @prepare: Prepare the work, must NOT be NULL.
1423  * @commit: Commit the changes, can be NULL.
1424  * @abort: Abort the changes on fail, can be NULL.
1425  * @clean: Clean up resources after all transaction actions have called
1426  *         commit() or abort(). Can be NULL.
1427  *
1428  * Only prepare() may fail. In a single transaction, only one of commit() or
1429  * abort() will be called. clean() will always be called if it is present.
1430  */
1431 typedef struct BlkActionOps {
1432     size_t instance_size;
1433     void (*prepare)(BlkActionState *common, Error **errp);
1434     void (*commit)(BlkActionState *common);
1435     void (*abort)(BlkActionState *common);
1436     void (*clean)(BlkActionState *common);
1437 } BlkActionOps;
1438 
1439 /**
1440  * BlkActionState:
1441  * Describes one Action's state within a Transaction.
1442  *
1443  * @action: QAPI-defined enum identifying which Action to perform.
1444  * @ops: Table of ActionOps this Action can perform.
1445  * @block_job_txn: Transaction which this action belongs to.
1446  * @entry: List membership for all Actions in this Transaction.
1447  *
1448  * This structure must be arranged as first member in a subclassed type,
1449  * assuming that the compiler will also arrange it to the same offsets as the
1450  * base class.
1451  */
1452 struct BlkActionState {
1453     TransactionAction *action;
1454     const BlkActionOps *ops;
1455     BlockJobTxn *block_job_txn;
1456     TransactionProperties *txn_props;
1457     QSIMPLEQ_ENTRY(BlkActionState) entry;
1458 };
1459 
1460 /* internal snapshot private data */
1461 typedef struct InternalSnapshotState {
1462     BlkActionState common;
1463     BlockDriverState *bs;
1464     AioContext *aio_context;
1465     QEMUSnapshotInfo sn;
1466     bool created;
1467 } InternalSnapshotState;
1468 
1469 
1470 static int action_check_completion_mode(BlkActionState *s, Error **errp)
1471 {
1472     if (s->txn_props->completion_mode != ACTION_COMPLETION_MODE_INDIVIDUAL) {
1473         error_setg(errp,
1474                    "Action '%s' does not support Transaction property "
1475                    "completion-mode = %s",
1476                    TransactionActionKind_lookup[s->action->type],
1477                    ActionCompletionMode_lookup[s->txn_props->completion_mode]);
1478         return -1;
1479     }
1480     return 0;
1481 }
1482 
1483 static void internal_snapshot_prepare(BlkActionState *common,
1484                                       Error **errp)
1485 {
1486     Error *local_err = NULL;
1487     const char *device;
1488     const char *name;
1489     BlockDriverState *bs;
1490     QEMUSnapshotInfo old_sn, *sn;
1491     bool ret;
1492     qemu_timeval tv;
1493     BlockdevSnapshotInternal *internal;
1494     InternalSnapshotState *state;
1495     int ret1;
1496 
1497     g_assert(common->action->type ==
1498              TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC);
1499     internal = common->action->u.blockdev_snapshot_internal_sync.data;
1500     state = DO_UPCAST(InternalSnapshotState, common, common);
1501 
1502     /* 1. parse input */
1503     device = internal->device;
1504     name = internal->name;
1505 
1506     /* 2. check for validation */
1507     if (action_check_completion_mode(common, errp) < 0) {
1508         return;
1509     }
1510 
1511     bs = qmp_get_root_bs(device, errp);
1512     if (!bs) {
1513         return;
1514     }
1515 
1516     /* AioContext is released in .clean() */
1517     state->aio_context = bdrv_get_aio_context(bs);
1518     aio_context_acquire(state->aio_context);
1519 
1520     state->bs = bs;
1521     bdrv_drained_begin(bs);
1522 
1523     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_INTERNAL_SNAPSHOT, errp)) {
1524         return;
1525     }
1526 
1527     if (bdrv_is_read_only(bs)) {
1528         error_setg(errp, "Device '%s' is read only", device);
1529         return;
1530     }
1531 
1532     if (!bdrv_can_snapshot(bs)) {
1533         error_setg(errp, "Block format '%s' used by device '%s' "
1534                    "does not support internal snapshots",
1535                    bs->drv->format_name, device);
1536         return;
1537     }
1538 
1539     if (!strlen(name)) {
1540         error_setg(errp, "Name is empty");
1541         return;
1542     }
1543 
1544     /* check whether a snapshot with name exist */
1545     ret = bdrv_snapshot_find_by_id_and_name(bs, NULL, name, &old_sn,
1546                                             &local_err);
1547     if (local_err) {
1548         error_propagate(errp, local_err);
1549         return;
1550     } else if (ret) {
1551         error_setg(errp,
1552                    "Snapshot with name '%s' already exists on device '%s'",
1553                    name, device);
1554         return;
1555     }
1556 
1557     /* 3. take the snapshot */
1558     sn = &state->sn;
1559     pstrcpy(sn->name, sizeof(sn->name), name);
1560     qemu_gettimeofday(&tv);
1561     sn->date_sec = tv.tv_sec;
1562     sn->date_nsec = tv.tv_usec * 1000;
1563     sn->vm_clock_nsec = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL);
1564 
1565     ret1 = bdrv_snapshot_create(bs, sn);
1566     if (ret1 < 0) {
1567         error_setg_errno(errp, -ret1,
1568                          "Failed to create snapshot '%s' on device '%s'",
1569                          name, device);
1570         return;
1571     }
1572 
1573     /* 4. succeed, mark a snapshot is created */
1574     state->created = true;
1575 }
1576 
1577 static void internal_snapshot_abort(BlkActionState *common)
1578 {
1579     InternalSnapshotState *state =
1580                              DO_UPCAST(InternalSnapshotState, common, common);
1581     BlockDriverState *bs = state->bs;
1582     QEMUSnapshotInfo *sn = &state->sn;
1583     Error *local_error = NULL;
1584 
1585     if (!state->created) {
1586         return;
1587     }
1588 
1589     if (bdrv_snapshot_delete(bs, sn->id_str, sn->name, &local_error) < 0) {
1590         error_reportf_err(local_error,
1591                           "Failed to delete snapshot with id '%s' and "
1592                           "name '%s' on device '%s' in abort: ",
1593                           sn->id_str, sn->name,
1594                           bdrv_get_device_name(bs));
1595     }
1596 }
1597 
1598 static void internal_snapshot_clean(BlkActionState *common)
1599 {
1600     InternalSnapshotState *state = DO_UPCAST(InternalSnapshotState,
1601                                              common, common);
1602 
1603     if (state->aio_context) {
1604         if (state->bs) {
1605             bdrv_drained_end(state->bs);
1606         }
1607         aio_context_release(state->aio_context);
1608     }
1609 }
1610 
1611 /* external snapshot private data */
1612 typedef struct ExternalSnapshotState {
1613     BlkActionState common;
1614     BlockDriverState *old_bs;
1615     BlockDriverState *new_bs;
1616     AioContext *aio_context;
1617     bool overlay_appended;
1618 } ExternalSnapshotState;
1619 
1620 static void external_snapshot_prepare(BlkActionState *common,
1621                                       Error **errp)
1622 {
1623     int flags = 0;
1624     QDict *options = NULL;
1625     Error *local_err = NULL;
1626     /* Device and node name of the image to generate the snapshot from */
1627     const char *device;
1628     const char *node_name;
1629     /* Reference to the new image (for 'blockdev-snapshot') */
1630     const char *snapshot_ref;
1631     /* File name of the new image (for 'blockdev-snapshot-sync') */
1632     const char *new_image_file;
1633     ExternalSnapshotState *state =
1634                              DO_UPCAST(ExternalSnapshotState, common, common);
1635     TransactionAction *action = common->action;
1636 
1637     /* 'blockdev-snapshot' and 'blockdev-snapshot-sync' have similar
1638      * purpose but a different set of parameters */
1639     switch (action->type) {
1640     case TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT:
1641         {
1642             BlockdevSnapshot *s = action->u.blockdev_snapshot.data;
1643             device = s->node;
1644             node_name = s->node;
1645             new_image_file = NULL;
1646             snapshot_ref = s->overlay;
1647         }
1648         break;
1649     case TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC:
1650         {
1651             BlockdevSnapshotSync *s = action->u.blockdev_snapshot_sync.data;
1652             device = s->has_device ? s->device : NULL;
1653             node_name = s->has_node_name ? s->node_name : NULL;
1654             new_image_file = s->snapshot_file;
1655             snapshot_ref = NULL;
1656         }
1657         break;
1658     default:
1659         g_assert_not_reached();
1660     }
1661 
1662     /* start processing */
1663     if (action_check_completion_mode(common, errp) < 0) {
1664         return;
1665     }
1666 
1667     state->old_bs = bdrv_lookup_bs(device, node_name, errp);
1668     if (!state->old_bs) {
1669         return;
1670     }
1671 
1672     /* Acquire AioContext now so any threads operating on old_bs stop */
1673     state->aio_context = bdrv_get_aio_context(state->old_bs);
1674     aio_context_acquire(state->aio_context);
1675     bdrv_drained_begin(state->old_bs);
1676 
1677     if (!bdrv_is_inserted(state->old_bs)) {
1678         error_setg(errp, QERR_DEVICE_HAS_NO_MEDIUM, device);
1679         return;
1680     }
1681 
1682     if (bdrv_op_is_blocked(state->old_bs,
1683                            BLOCK_OP_TYPE_EXTERNAL_SNAPSHOT, errp)) {
1684         return;
1685     }
1686 
1687     if (!bdrv_is_read_only(state->old_bs)) {
1688         if (bdrv_flush(state->old_bs)) {
1689             error_setg(errp, QERR_IO_ERROR);
1690             return;
1691         }
1692     }
1693 
1694     if (!bdrv_is_first_non_filter(state->old_bs)) {
1695         error_setg(errp, QERR_FEATURE_DISABLED, "snapshot");
1696         return;
1697     }
1698 
1699     if (action->type == TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC) {
1700         BlockdevSnapshotSync *s = action->u.blockdev_snapshot_sync.data;
1701         const char *format = s->has_format ? s->format : "qcow2";
1702         enum NewImageMode mode;
1703         const char *snapshot_node_name =
1704             s->has_snapshot_node_name ? s->snapshot_node_name : NULL;
1705 
1706         if (node_name && !snapshot_node_name) {
1707             error_setg(errp, "New snapshot node name missing");
1708             return;
1709         }
1710 
1711         if (snapshot_node_name &&
1712             bdrv_lookup_bs(snapshot_node_name, snapshot_node_name, NULL)) {
1713             error_setg(errp, "New snapshot node name already in use");
1714             return;
1715         }
1716 
1717         flags = state->old_bs->open_flags;
1718         flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_COPY_ON_READ);
1719 
1720         /* create new image w/backing file */
1721         mode = s->has_mode ? s->mode : NEW_IMAGE_MODE_ABSOLUTE_PATHS;
1722         if (mode != NEW_IMAGE_MODE_EXISTING) {
1723             int64_t size = bdrv_getlength(state->old_bs);
1724             if (size < 0) {
1725                 error_setg_errno(errp, -size, "bdrv_getlength failed");
1726                 return;
1727             }
1728             bdrv_img_create(new_image_file, format,
1729                             state->old_bs->filename,
1730                             state->old_bs->drv->format_name,
1731                             NULL, size, flags, &local_err, false);
1732             if (local_err) {
1733                 error_propagate(errp, local_err);
1734                 return;
1735             }
1736         }
1737 
1738         options = qdict_new();
1739         if (s->has_snapshot_node_name) {
1740             qdict_put(options, "node-name",
1741                       qstring_from_str(snapshot_node_name));
1742         }
1743         qdict_put(options, "driver", qstring_from_str(format));
1744 
1745         flags |= BDRV_O_NO_BACKING;
1746     }
1747 
1748     state->new_bs = bdrv_open(new_image_file, snapshot_ref, options, flags,
1749                               errp);
1750     /* We will manually add the backing_hd field to the bs later */
1751     if (!state->new_bs) {
1752         return;
1753     }
1754 
1755     if (bdrv_has_blk(state->new_bs)) {
1756         error_setg(errp, "The snapshot is already in use");
1757         return;
1758     }
1759 
1760     if (bdrv_op_is_blocked(state->new_bs, BLOCK_OP_TYPE_EXTERNAL_SNAPSHOT,
1761                            errp)) {
1762         return;
1763     }
1764 
1765     if (state->new_bs->backing != NULL) {
1766         error_setg(errp, "The snapshot already has a backing image");
1767         return;
1768     }
1769 
1770     if (!state->new_bs->drv->supports_backing) {
1771         error_setg(errp, "The snapshot does not support backing images");
1772         return;
1773     }
1774 
1775     /* This removes our old bs and adds the new bs. This is an operation that
1776      * can fail, so we need to do it in .prepare; undoing it for abort is
1777      * always possible. */
1778     bdrv_ref(state->new_bs);
1779     bdrv_append(state->new_bs, state->old_bs, &local_err);
1780     if (local_err) {
1781         error_propagate(errp, local_err);
1782         return;
1783     }
1784     state->overlay_appended = true;
1785 }
1786 
1787 static void external_snapshot_commit(BlkActionState *common)
1788 {
1789     ExternalSnapshotState *state =
1790                              DO_UPCAST(ExternalSnapshotState, common, common);
1791 
1792     bdrv_set_aio_context(state->new_bs, state->aio_context);
1793 
1794     /* We don't need (or want) to use the transactional
1795      * bdrv_reopen_multiple() across all the entries at once, because we
1796      * don't want to abort all of them if one of them fails the reopen */
1797     if (!state->old_bs->copy_on_read) {
1798         bdrv_reopen(state->old_bs, state->old_bs->open_flags & ~BDRV_O_RDWR,
1799                     NULL);
1800     }
1801 }
1802 
1803 static void external_snapshot_abort(BlkActionState *common)
1804 {
1805     ExternalSnapshotState *state =
1806                              DO_UPCAST(ExternalSnapshotState, common, common);
1807     if (state->new_bs) {
1808         if (state->overlay_appended) {
1809             bdrv_replace_node(state->new_bs, state->old_bs, &error_abort);
1810         }
1811     }
1812 }
1813 
1814 static void external_snapshot_clean(BlkActionState *common)
1815 {
1816     ExternalSnapshotState *state =
1817                              DO_UPCAST(ExternalSnapshotState, common, common);
1818     if (state->aio_context) {
1819         bdrv_drained_end(state->old_bs);
1820         aio_context_release(state->aio_context);
1821         bdrv_unref(state->new_bs);
1822     }
1823 }
1824 
1825 typedef struct DriveBackupState {
1826     BlkActionState common;
1827     BlockDriverState *bs;
1828     AioContext *aio_context;
1829     BlockJob *job;
1830 } DriveBackupState;
1831 
1832 static BlockJob *do_drive_backup(DriveBackup *backup, BlockJobTxn *txn,
1833                             Error **errp);
1834 
1835 static void drive_backup_prepare(BlkActionState *common, Error **errp)
1836 {
1837     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1838     BlockDriverState *bs;
1839     DriveBackup *backup;
1840     Error *local_err = NULL;
1841 
1842     assert(common->action->type == TRANSACTION_ACTION_KIND_DRIVE_BACKUP);
1843     backup = common->action->u.drive_backup.data;
1844 
1845     bs = qmp_get_root_bs(backup->device, errp);
1846     if (!bs) {
1847         return;
1848     }
1849 
1850     /* AioContext is released in .clean() */
1851     state->aio_context = bdrv_get_aio_context(bs);
1852     aio_context_acquire(state->aio_context);
1853     bdrv_drained_begin(bs);
1854     state->bs = bs;
1855 
1856     state->job = do_drive_backup(backup, common->block_job_txn, &local_err);
1857     if (local_err) {
1858         error_propagate(errp, local_err);
1859         return;
1860     }
1861 }
1862 
1863 static void drive_backup_commit(BlkActionState *common)
1864 {
1865     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1866     assert(state->job);
1867     block_job_start(state->job);
1868 }
1869 
1870 static void drive_backup_abort(BlkActionState *common)
1871 {
1872     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1873 
1874     if (state->job) {
1875         block_job_cancel_sync(state->job);
1876     }
1877 }
1878 
1879 static void drive_backup_clean(BlkActionState *common)
1880 {
1881     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1882 
1883     if (state->aio_context) {
1884         bdrv_drained_end(state->bs);
1885         aio_context_release(state->aio_context);
1886     }
1887 }
1888 
1889 typedef struct BlockdevBackupState {
1890     BlkActionState common;
1891     BlockDriverState *bs;
1892     BlockJob *job;
1893     AioContext *aio_context;
1894 } BlockdevBackupState;
1895 
1896 static BlockJob *do_blockdev_backup(BlockdevBackup *backup, BlockJobTxn *txn,
1897                                     Error **errp);
1898 
1899 static void blockdev_backup_prepare(BlkActionState *common, Error **errp)
1900 {
1901     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1902     BlockdevBackup *backup;
1903     BlockDriverState *bs, *target;
1904     Error *local_err = NULL;
1905 
1906     assert(common->action->type == TRANSACTION_ACTION_KIND_BLOCKDEV_BACKUP);
1907     backup = common->action->u.blockdev_backup.data;
1908 
1909     bs = qmp_get_root_bs(backup->device, errp);
1910     if (!bs) {
1911         return;
1912     }
1913 
1914     target = bdrv_lookup_bs(backup->target, backup->target, errp);
1915     if (!target) {
1916         return;
1917     }
1918 
1919     /* AioContext is released in .clean() */
1920     state->aio_context = bdrv_get_aio_context(bs);
1921     if (state->aio_context != bdrv_get_aio_context(target)) {
1922         state->aio_context = NULL;
1923         error_setg(errp, "Backup between two IO threads is not implemented");
1924         return;
1925     }
1926     aio_context_acquire(state->aio_context);
1927     state->bs = bs;
1928     bdrv_drained_begin(state->bs);
1929 
1930     state->job = do_blockdev_backup(backup, common->block_job_txn, &local_err);
1931     if (local_err) {
1932         error_propagate(errp, local_err);
1933         return;
1934     }
1935 }
1936 
1937 static void blockdev_backup_commit(BlkActionState *common)
1938 {
1939     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1940     assert(state->job);
1941     block_job_start(state->job);
1942 }
1943 
1944 static void blockdev_backup_abort(BlkActionState *common)
1945 {
1946     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1947 
1948     if (state->job) {
1949         block_job_cancel_sync(state->job);
1950     }
1951 }
1952 
1953 static void blockdev_backup_clean(BlkActionState *common)
1954 {
1955     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1956 
1957     if (state->aio_context) {
1958         bdrv_drained_end(state->bs);
1959         aio_context_release(state->aio_context);
1960     }
1961 }
1962 
1963 typedef struct BlockDirtyBitmapState {
1964     BlkActionState common;
1965     BdrvDirtyBitmap *bitmap;
1966     BlockDriverState *bs;
1967     AioContext *aio_context;
1968     HBitmap *backup;
1969     bool prepared;
1970 } BlockDirtyBitmapState;
1971 
1972 static void block_dirty_bitmap_add_prepare(BlkActionState *common,
1973                                            Error **errp)
1974 {
1975     Error *local_err = NULL;
1976     BlockDirtyBitmapAdd *action;
1977     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
1978                                              common, common);
1979 
1980     if (action_check_completion_mode(common, errp) < 0) {
1981         return;
1982     }
1983 
1984     action = common->action->u.block_dirty_bitmap_add.data;
1985     /* AIO context taken and released within qmp_block_dirty_bitmap_add */
1986     qmp_block_dirty_bitmap_add(action->node, action->name,
1987                                action->has_granularity, action->granularity,
1988                                &local_err);
1989 
1990     if (!local_err) {
1991         state->prepared = true;
1992     } else {
1993         error_propagate(errp, local_err);
1994     }
1995 }
1996 
1997 static void block_dirty_bitmap_add_abort(BlkActionState *common)
1998 {
1999     BlockDirtyBitmapAdd *action;
2000     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2001                                              common, common);
2002 
2003     action = common->action->u.block_dirty_bitmap_add.data;
2004     /* Should not be able to fail: IF the bitmap was added via .prepare(),
2005      * then the node reference and bitmap name must have been valid.
2006      */
2007     if (state->prepared) {
2008         qmp_block_dirty_bitmap_remove(action->node, action->name, &error_abort);
2009     }
2010 }
2011 
2012 static void block_dirty_bitmap_clear_prepare(BlkActionState *common,
2013                                              Error **errp)
2014 {
2015     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2016                                              common, common);
2017     BlockDirtyBitmap *action;
2018 
2019     if (action_check_completion_mode(common, errp) < 0) {
2020         return;
2021     }
2022 
2023     action = common->action->u.block_dirty_bitmap_clear.data;
2024     state->bitmap = block_dirty_bitmap_lookup(action->node,
2025                                               action->name,
2026                                               &state->bs,
2027                                               &state->aio_context,
2028                                               errp);
2029     if (!state->bitmap) {
2030         return;
2031     }
2032 
2033     if (bdrv_dirty_bitmap_frozen(state->bitmap)) {
2034         error_setg(errp, "Cannot modify a frozen bitmap");
2035         return;
2036     } else if (!bdrv_dirty_bitmap_enabled(state->bitmap)) {
2037         error_setg(errp, "Cannot clear a disabled bitmap");
2038         return;
2039     }
2040 
2041     bdrv_clear_dirty_bitmap(state->bitmap, &state->backup);
2042     /* AioContext is released in .clean() */
2043 }
2044 
2045 static void block_dirty_bitmap_clear_abort(BlkActionState *common)
2046 {
2047     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2048                                              common, common);
2049 
2050     bdrv_undo_clear_dirty_bitmap(state->bitmap, state->backup);
2051 }
2052 
2053 static void block_dirty_bitmap_clear_commit(BlkActionState *common)
2054 {
2055     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2056                                              common, common);
2057 
2058     hbitmap_free(state->backup);
2059 }
2060 
2061 static void block_dirty_bitmap_clear_clean(BlkActionState *common)
2062 {
2063     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2064                                              common, common);
2065 
2066     if (state->aio_context) {
2067         aio_context_release(state->aio_context);
2068     }
2069 }
2070 
2071 static void abort_prepare(BlkActionState *common, Error **errp)
2072 {
2073     error_setg(errp, "Transaction aborted using Abort action");
2074 }
2075 
2076 static void abort_commit(BlkActionState *common)
2077 {
2078     g_assert_not_reached(); /* this action never succeeds */
2079 }
2080 
2081 static const BlkActionOps actions[] = {
2082     [TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT] = {
2083         .instance_size = sizeof(ExternalSnapshotState),
2084         .prepare  = external_snapshot_prepare,
2085         .commit   = external_snapshot_commit,
2086         .abort = external_snapshot_abort,
2087         .clean = external_snapshot_clean,
2088     },
2089     [TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC] = {
2090         .instance_size = sizeof(ExternalSnapshotState),
2091         .prepare  = external_snapshot_prepare,
2092         .commit   = external_snapshot_commit,
2093         .abort = external_snapshot_abort,
2094         .clean = external_snapshot_clean,
2095     },
2096     [TRANSACTION_ACTION_KIND_DRIVE_BACKUP] = {
2097         .instance_size = sizeof(DriveBackupState),
2098         .prepare = drive_backup_prepare,
2099         .commit = drive_backup_commit,
2100         .abort = drive_backup_abort,
2101         .clean = drive_backup_clean,
2102     },
2103     [TRANSACTION_ACTION_KIND_BLOCKDEV_BACKUP] = {
2104         .instance_size = sizeof(BlockdevBackupState),
2105         .prepare = blockdev_backup_prepare,
2106         .commit = blockdev_backup_commit,
2107         .abort = blockdev_backup_abort,
2108         .clean = blockdev_backup_clean,
2109     },
2110     [TRANSACTION_ACTION_KIND_ABORT] = {
2111         .instance_size = sizeof(BlkActionState),
2112         .prepare = abort_prepare,
2113         .commit = abort_commit,
2114     },
2115     [TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC] = {
2116         .instance_size = sizeof(InternalSnapshotState),
2117         .prepare  = internal_snapshot_prepare,
2118         .abort = internal_snapshot_abort,
2119         .clean = internal_snapshot_clean,
2120     },
2121     [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_ADD] = {
2122         .instance_size = sizeof(BlockDirtyBitmapState),
2123         .prepare = block_dirty_bitmap_add_prepare,
2124         .abort = block_dirty_bitmap_add_abort,
2125     },
2126     [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_CLEAR] = {
2127         .instance_size = sizeof(BlockDirtyBitmapState),
2128         .prepare = block_dirty_bitmap_clear_prepare,
2129         .commit = block_dirty_bitmap_clear_commit,
2130         .abort = block_dirty_bitmap_clear_abort,
2131         .clean = block_dirty_bitmap_clear_clean,
2132     }
2133 };
2134 
2135 /**
2136  * Allocate a TransactionProperties structure if necessary, and fill
2137  * that structure with desired defaults if they are unset.
2138  */
2139 static TransactionProperties *get_transaction_properties(
2140     TransactionProperties *props)
2141 {
2142     if (!props) {
2143         props = g_new0(TransactionProperties, 1);
2144     }
2145 
2146     if (!props->has_completion_mode) {
2147         props->has_completion_mode = true;
2148         props->completion_mode = ACTION_COMPLETION_MODE_INDIVIDUAL;
2149     }
2150 
2151     return props;
2152 }
2153 
2154 /*
2155  * 'Atomic' group operations.  The operations are performed as a set, and if
2156  * any fail then we roll back all operations in the group.
2157  */
2158 void qmp_transaction(TransactionActionList *dev_list,
2159                      bool has_props,
2160                      struct TransactionProperties *props,
2161                      Error **errp)
2162 {
2163     TransactionActionList *dev_entry = dev_list;
2164     BlockJobTxn *block_job_txn = NULL;
2165     BlkActionState *state, *next;
2166     Error *local_err = NULL;
2167 
2168     QSIMPLEQ_HEAD(snap_bdrv_states, BlkActionState) snap_bdrv_states;
2169     QSIMPLEQ_INIT(&snap_bdrv_states);
2170 
2171     /* Does this transaction get canceled as a group on failure?
2172      * If not, we don't really need to make a BlockJobTxn.
2173      */
2174     props = get_transaction_properties(props);
2175     if (props->completion_mode != ACTION_COMPLETION_MODE_INDIVIDUAL) {
2176         block_job_txn = block_job_txn_new();
2177     }
2178 
2179     /* drain all i/o before any operations */
2180     bdrv_drain_all();
2181 
2182     /* We don't do anything in this loop that commits us to the operations */
2183     while (NULL != dev_entry) {
2184         TransactionAction *dev_info = NULL;
2185         const BlkActionOps *ops;
2186 
2187         dev_info = dev_entry->value;
2188         dev_entry = dev_entry->next;
2189 
2190         assert(dev_info->type < ARRAY_SIZE(actions));
2191 
2192         ops = &actions[dev_info->type];
2193         assert(ops->instance_size > 0);
2194 
2195         state = g_malloc0(ops->instance_size);
2196         state->ops = ops;
2197         state->action = dev_info;
2198         state->block_job_txn = block_job_txn;
2199         state->txn_props = props;
2200         QSIMPLEQ_INSERT_TAIL(&snap_bdrv_states, state, entry);
2201 
2202         state->ops->prepare(state, &local_err);
2203         if (local_err) {
2204             error_propagate(errp, local_err);
2205             goto delete_and_fail;
2206         }
2207     }
2208 
2209     QSIMPLEQ_FOREACH(state, &snap_bdrv_states, entry) {
2210         if (state->ops->commit) {
2211             state->ops->commit(state);
2212         }
2213     }
2214 
2215     /* success */
2216     goto exit;
2217 
2218 delete_and_fail:
2219     /* failure, and it is all-or-none; roll back all operations */
2220     QSIMPLEQ_FOREACH(state, &snap_bdrv_states, entry) {
2221         if (state->ops->abort) {
2222             state->ops->abort(state);
2223         }
2224     }
2225 exit:
2226     QSIMPLEQ_FOREACH_SAFE(state, &snap_bdrv_states, entry, next) {
2227         if (state->ops->clean) {
2228             state->ops->clean(state);
2229         }
2230         g_free(state);
2231     }
2232     if (!has_props) {
2233         qapi_free_TransactionProperties(props);
2234     }
2235     block_job_txn_unref(block_job_txn);
2236 }
2237 
2238 void qmp_eject(bool has_device, const char *device,
2239                bool has_id, const char *id,
2240                bool has_force, bool force, Error **errp)
2241 {
2242     Error *local_err = NULL;
2243     int rc;
2244 
2245     if (!has_force) {
2246         force = false;
2247     }
2248 
2249     rc = do_open_tray(has_device ? device : NULL,
2250                       has_id ? id : NULL,
2251                       force, &local_err);
2252     if (rc && rc != -ENOSYS) {
2253         error_propagate(errp, local_err);
2254         return;
2255     }
2256     error_free(local_err);
2257 
2258     qmp_x_blockdev_remove_medium(has_device, device, has_id, id, errp);
2259 }
2260 
2261 void qmp_block_passwd(bool has_device, const char *device,
2262                       bool has_node_name, const char *node_name,
2263                       const char *password, Error **errp)
2264 {
2265     Error *local_err = NULL;
2266     BlockDriverState *bs;
2267     AioContext *aio_context;
2268 
2269     bs = bdrv_lookup_bs(has_device ? device : NULL,
2270                         has_node_name ? node_name : NULL,
2271                         &local_err);
2272     if (local_err) {
2273         error_propagate(errp, local_err);
2274         return;
2275     }
2276 
2277     aio_context = bdrv_get_aio_context(bs);
2278     aio_context_acquire(aio_context);
2279 
2280     bdrv_add_key(bs, password, errp);
2281 
2282     aio_context_release(aio_context);
2283 }
2284 
2285 /*
2286  * Attempt to open the tray of @device.
2287  * If @force, ignore its tray lock.
2288  * Else, if the tray is locked, don't open it, but ask the guest to open it.
2289  * On error, store an error through @errp and return -errno.
2290  * If @device does not exist, return -ENODEV.
2291  * If it has no removable media, return -ENOTSUP.
2292  * If it has no tray, return -ENOSYS.
2293  * If the guest was asked to open the tray, return -EINPROGRESS.
2294  * Else, return 0.
2295  */
2296 static int do_open_tray(const char *blk_name, const char *qdev_id,
2297                         bool force, Error **errp)
2298 {
2299     BlockBackend *blk;
2300     const char *device = qdev_id ?: blk_name;
2301     bool locked;
2302 
2303     blk = qmp_get_blk(blk_name, qdev_id, errp);
2304     if (!blk) {
2305         return -ENODEV;
2306     }
2307 
2308     if (!blk_dev_has_removable_media(blk)) {
2309         error_setg(errp, "Device '%s' is not removable", device);
2310         return -ENOTSUP;
2311     }
2312 
2313     if (!blk_dev_has_tray(blk)) {
2314         error_setg(errp, "Device '%s' does not have a tray", device);
2315         return -ENOSYS;
2316     }
2317 
2318     if (blk_dev_is_tray_open(blk)) {
2319         return 0;
2320     }
2321 
2322     locked = blk_dev_is_medium_locked(blk);
2323     if (locked) {
2324         blk_dev_eject_request(blk, force);
2325     }
2326 
2327     if (!locked || force) {
2328         blk_dev_change_media_cb(blk, false, &error_abort);
2329     }
2330 
2331     if (locked && !force) {
2332         error_setg(errp, "Device '%s' is locked and force was not specified, "
2333                    "wait for tray to open and try again", device);
2334         return -EINPROGRESS;
2335     }
2336 
2337     return 0;
2338 }
2339 
2340 void qmp_blockdev_open_tray(bool has_device, const char *device,
2341                             bool has_id, const char *id,
2342                             bool has_force, bool force,
2343                             Error **errp)
2344 {
2345     Error *local_err = NULL;
2346     int rc;
2347 
2348     if (!has_force) {
2349         force = false;
2350     }
2351     rc = do_open_tray(has_device ? device : NULL,
2352                       has_id ? id : NULL,
2353                       force, &local_err);
2354     if (rc && rc != -ENOSYS && rc != -EINPROGRESS) {
2355         error_propagate(errp, local_err);
2356         return;
2357     }
2358     error_free(local_err);
2359 }
2360 
2361 void qmp_blockdev_close_tray(bool has_device, const char *device,
2362                              bool has_id, const char *id,
2363                              Error **errp)
2364 {
2365     BlockBackend *blk;
2366     Error *local_err = NULL;
2367 
2368     device = has_device ? device : NULL;
2369     id = has_id ? id : NULL;
2370 
2371     blk = qmp_get_blk(device, id, errp);
2372     if (!blk) {
2373         return;
2374     }
2375 
2376     if (!blk_dev_has_removable_media(blk)) {
2377         error_setg(errp, "Device '%s' is not removable", device ?: id);
2378         return;
2379     }
2380 
2381     if (!blk_dev_has_tray(blk)) {
2382         /* Ignore this command on tray-less devices */
2383         return;
2384     }
2385 
2386     if (!blk_dev_is_tray_open(blk)) {
2387         return;
2388     }
2389 
2390     blk_dev_change_media_cb(blk, true, &local_err);
2391     if (local_err) {
2392         error_propagate(errp, local_err);
2393         return;
2394     }
2395 }
2396 
2397 void qmp_x_blockdev_remove_medium(bool has_device, const char *device,
2398                                   bool has_id, const char *id, Error **errp)
2399 {
2400     BlockBackend *blk;
2401     BlockDriverState *bs;
2402     AioContext *aio_context;
2403     bool has_attached_device;
2404 
2405     device = has_device ? device : NULL;
2406     id = has_id ? id : NULL;
2407 
2408     blk = qmp_get_blk(device, id, errp);
2409     if (!blk) {
2410         return;
2411     }
2412 
2413     /* For BBs without a device, we can exchange the BDS tree at will */
2414     has_attached_device = blk_get_attached_dev(blk);
2415 
2416     if (has_attached_device && !blk_dev_has_removable_media(blk)) {
2417         error_setg(errp, "Device '%s' is not removable", device ?: id);
2418         return;
2419     }
2420 
2421     if (has_attached_device && blk_dev_has_tray(blk) &&
2422         !blk_dev_is_tray_open(blk))
2423     {
2424         error_setg(errp, "Tray of device '%s' is not open", device ?: id);
2425         return;
2426     }
2427 
2428     bs = blk_bs(blk);
2429     if (!bs) {
2430         return;
2431     }
2432 
2433     aio_context = bdrv_get_aio_context(bs);
2434     aio_context_acquire(aio_context);
2435 
2436     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_EJECT, errp)) {
2437         goto out;
2438     }
2439 
2440     blk_remove_bs(blk);
2441 
2442     if (!blk_dev_has_tray(blk)) {
2443         /* For tray-less devices, blockdev-open-tray is a no-op (or may not be
2444          * called at all); therefore, the medium needs to be ejected here.
2445          * Do it after blk_remove_bs() so blk_is_inserted(blk) returns the @load
2446          * value passed here (i.e. false). */
2447         blk_dev_change_media_cb(blk, false, &error_abort);
2448     }
2449 
2450 out:
2451     aio_context_release(aio_context);
2452 }
2453 
2454 static void qmp_blockdev_insert_anon_medium(BlockBackend *blk,
2455                                             BlockDriverState *bs, Error **errp)
2456 {
2457     Error *local_err = NULL;
2458     bool has_device;
2459     int ret;
2460 
2461     /* For BBs without a device, we can exchange the BDS tree at will */
2462     has_device = blk_get_attached_dev(blk);
2463 
2464     if (has_device && !blk_dev_has_removable_media(blk)) {
2465         error_setg(errp, "Device is not removable");
2466         return;
2467     }
2468 
2469     if (has_device && blk_dev_has_tray(blk) && !blk_dev_is_tray_open(blk)) {
2470         error_setg(errp, "Tray of the device is not open");
2471         return;
2472     }
2473 
2474     if (blk_bs(blk)) {
2475         error_setg(errp, "There already is a medium in the device");
2476         return;
2477     }
2478 
2479     ret = blk_insert_bs(blk, bs, errp);
2480     if (ret < 0) {
2481         return;
2482     }
2483 
2484     if (!blk_dev_has_tray(blk)) {
2485         /* For tray-less devices, blockdev-close-tray is a no-op (or may not be
2486          * called at all); therefore, the medium needs to be pushed into the
2487          * slot here.
2488          * Do it after blk_insert_bs() so blk_is_inserted(blk) returns the @load
2489          * value passed here (i.e. true). */
2490         blk_dev_change_media_cb(blk, true, &local_err);
2491         if (local_err) {
2492             error_propagate(errp, local_err);
2493             blk_remove_bs(blk);
2494             return;
2495         }
2496     }
2497 }
2498 
2499 void qmp_x_blockdev_insert_medium(bool has_device, const char *device,
2500                                   bool has_id, const char *id,
2501                                   const char *node_name, Error **errp)
2502 {
2503     BlockBackend *blk;
2504     BlockDriverState *bs;
2505 
2506     blk = qmp_get_blk(has_device ? device : NULL,
2507                       has_id ? id : NULL,
2508                       errp);
2509     if (!blk) {
2510         return;
2511     }
2512 
2513     bs = bdrv_find_node(node_name);
2514     if (!bs) {
2515         error_setg(errp, "Node '%s' not found", node_name);
2516         return;
2517     }
2518 
2519     if (bdrv_has_blk(bs)) {
2520         error_setg(errp, "Node '%s' is already in use", node_name);
2521         return;
2522     }
2523 
2524     qmp_blockdev_insert_anon_medium(blk, bs, errp);
2525 }
2526 
2527 void qmp_blockdev_change_medium(bool has_device, const char *device,
2528                                 bool has_id, const char *id,
2529                                 const char *filename,
2530                                 bool has_format, const char *format,
2531                                 bool has_read_only,
2532                                 BlockdevChangeReadOnlyMode read_only,
2533                                 Error **errp)
2534 {
2535     BlockBackend *blk;
2536     BlockDriverState *medium_bs = NULL;
2537     int bdrv_flags;
2538     bool detect_zeroes;
2539     int rc;
2540     QDict *options = NULL;
2541     Error *err = NULL;
2542 
2543     blk = qmp_get_blk(has_device ? device : NULL,
2544                       has_id ? id : NULL,
2545                       errp);
2546     if (!blk) {
2547         goto fail;
2548     }
2549 
2550     if (blk_bs(blk)) {
2551         blk_update_root_state(blk);
2552     }
2553 
2554     bdrv_flags = blk_get_open_flags_from_root_state(blk);
2555     bdrv_flags &= ~(BDRV_O_TEMPORARY | BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING |
2556         BDRV_O_PROTOCOL);
2557 
2558     if (!has_read_only) {
2559         read_only = BLOCKDEV_CHANGE_READ_ONLY_MODE_RETAIN;
2560     }
2561 
2562     switch (read_only) {
2563     case BLOCKDEV_CHANGE_READ_ONLY_MODE_RETAIN:
2564         break;
2565 
2566     case BLOCKDEV_CHANGE_READ_ONLY_MODE_READ_ONLY:
2567         bdrv_flags &= ~BDRV_O_RDWR;
2568         break;
2569 
2570     case BLOCKDEV_CHANGE_READ_ONLY_MODE_READ_WRITE:
2571         bdrv_flags |= BDRV_O_RDWR;
2572         break;
2573 
2574     default:
2575         abort();
2576     }
2577 
2578     options = qdict_new();
2579     detect_zeroes = blk_get_detect_zeroes_from_root_state(blk);
2580     qdict_put(options, "detect-zeroes",
2581               qstring_from_str(detect_zeroes ? "on" : "off"));
2582 
2583     if (has_format) {
2584         qdict_put(options, "driver", qstring_from_str(format));
2585     }
2586 
2587     medium_bs = bdrv_open(filename, NULL, options, bdrv_flags, errp);
2588     if (!medium_bs) {
2589         goto fail;
2590     }
2591 
2592     bdrv_add_key(medium_bs, NULL, &err);
2593     if (err) {
2594         error_propagate(errp, err);
2595         goto fail;
2596     }
2597 
2598     rc = do_open_tray(has_device ? device : NULL,
2599                       has_id ? id : NULL,
2600                       false, &err);
2601     if (rc && rc != -ENOSYS) {
2602         error_propagate(errp, err);
2603         goto fail;
2604     }
2605     error_free(err);
2606     err = NULL;
2607 
2608     qmp_x_blockdev_remove_medium(has_device, device, has_id, id, &err);
2609     if (err) {
2610         error_propagate(errp, err);
2611         goto fail;
2612     }
2613 
2614     qmp_blockdev_insert_anon_medium(blk, medium_bs, &err);
2615     if (err) {
2616         error_propagate(errp, err);
2617         goto fail;
2618     }
2619 
2620     qmp_blockdev_close_tray(has_device, device, has_id, id, errp);
2621 
2622 fail:
2623     /* If the medium has been inserted, the device has its own reference, so
2624      * ours must be relinquished; and if it has not been inserted successfully,
2625      * the reference must be relinquished anyway */
2626     bdrv_unref(medium_bs);
2627 }
2628 
2629 /* throttling disk I/O limits */
2630 void qmp_block_set_io_throttle(BlockIOThrottle *arg, Error **errp)
2631 {
2632     ThrottleConfig cfg;
2633     BlockDriverState *bs;
2634     BlockBackend *blk;
2635     AioContext *aio_context;
2636 
2637     blk = qmp_get_blk(arg->has_device ? arg->device : NULL,
2638                       arg->has_id ? arg->id : NULL,
2639                       errp);
2640     if (!blk) {
2641         return;
2642     }
2643 
2644     aio_context = blk_get_aio_context(blk);
2645     aio_context_acquire(aio_context);
2646 
2647     bs = blk_bs(blk);
2648     if (!bs) {
2649         error_setg(errp, "Device has no medium");
2650         goto out;
2651     }
2652 
2653     throttle_config_init(&cfg);
2654     cfg.buckets[THROTTLE_BPS_TOTAL].avg = arg->bps;
2655     cfg.buckets[THROTTLE_BPS_READ].avg  = arg->bps_rd;
2656     cfg.buckets[THROTTLE_BPS_WRITE].avg = arg->bps_wr;
2657 
2658     cfg.buckets[THROTTLE_OPS_TOTAL].avg = arg->iops;
2659     cfg.buckets[THROTTLE_OPS_READ].avg  = arg->iops_rd;
2660     cfg.buckets[THROTTLE_OPS_WRITE].avg = arg->iops_wr;
2661 
2662     if (arg->has_bps_max) {
2663         cfg.buckets[THROTTLE_BPS_TOTAL].max = arg->bps_max;
2664     }
2665     if (arg->has_bps_rd_max) {
2666         cfg.buckets[THROTTLE_BPS_READ].max = arg->bps_rd_max;
2667     }
2668     if (arg->has_bps_wr_max) {
2669         cfg.buckets[THROTTLE_BPS_WRITE].max = arg->bps_wr_max;
2670     }
2671     if (arg->has_iops_max) {
2672         cfg.buckets[THROTTLE_OPS_TOTAL].max = arg->iops_max;
2673     }
2674     if (arg->has_iops_rd_max) {
2675         cfg.buckets[THROTTLE_OPS_READ].max = arg->iops_rd_max;
2676     }
2677     if (arg->has_iops_wr_max) {
2678         cfg.buckets[THROTTLE_OPS_WRITE].max = arg->iops_wr_max;
2679     }
2680 
2681     if (arg->has_bps_max_length) {
2682         cfg.buckets[THROTTLE_BPS_TOTAL].burst_length = arg->bps_max_length;
2683     }
2684     if (arg->has_bps_rd_max_length) {
2685         cfg.buckets[THROTTLE_BPS_READ].burst_length = arg->bps_rd_max_length;
2686     }
2687     if (arg->has_bps_wr_max_length) {
2688         cfg.buckets[THROTTLE_BPS_WRITE].burst_length = arg->bps_wr_max_length;
2689     }
2690     if (arg->has_iops_max_length) {
2691         cfg.buckets[THROTTLE_OPS_TOTAL].burst_length = arg->iops_max_length;
2692     }
2693     if (arg->has_iops_rd_max_length) {
2694         cfg.buckets[THROTTLE_OPS_READ].burst_length = arg->iops_rd_max_length;
2695     }
2696     if (arg->has_iops_wr_max_length) {
2697         cfg.buckets[THROTTLE_OPS_WRITE].burst_length = arg->iops_wr_max_length;
2698     }
2699 
2700     if (arg->has_iops_size) {
2701         cfg.op_size = arg->iops_size;
2702     }
2703 
2704     if (!throttle_is_valid(&cfg, errp)) {
2705         goto out;
2706     }
2707 
2708     if (throttle_enabled(&cfg)) {
2709         /* Enable I/O limits if they're not enabled yet, otherwise
2710          * just update the throttling group. */
2711         if (!blk_get_public(blk)->throttle_state) {
2712             blk_io_limits_enable(blk,
2713                                  arg->has_group ? arg->group :
2714                                  arg->has_device ? arg->device :
2715                                  arg->id);
2716         } else if (arg->has_group) {
2717             blk_io_limits_update_group(blk, arg->group);
2718         }
2719         /* Set the new throttling configuration */
2720         blk_set_io_limits(blk, &cfg);
2721     } else if (blk_get_public(blk)->throttle_state) {
2722         /* If all throttling settings are set to 0, disable I/O limits */
2723         blk_io_limits_disable(blk);
2724     }
2725 
2726 out:
2727     aio_context_release(aio_context);
2728 }
2729 
2730 void qmp_block_dirty_bitmap_add(const char *node, const char *name,
2731                                 bool has_granularity, uint32_t granularity,
2732                                 Error **errp)
2733 {
2734     AioContext *aio_context;
2735     BlockDriverState *bs;
2736 
2737     if (!name || name[0] == '\0') {
2738         error_setg(errp, "Bitmap name cannot be empty");
2739         return;
2740     }
2741 
2742     bs = bdrv_lookup_bs(node, node, errp);
2743     if (!bs) {
2744         return;
2745     }
2746 
2747     aio_context = bdrv_get_aio_context(bs);
2748     aio_context_acquire(aio_context);
2749 
2750     if (has_granularity) {
2751         if (granularity < 512 || !is_power_of_2(granularity)) {
2752             error_setg(errp, "Granularity must be power of 2 "
2753                              "and at least 512");
2754             goto out;
2755         }
2756     } else {
2757         /* Default to cluster size, if available: */
2758         granularity = bdrv_get_default_bitmap_granularity(bs);
2759     }
2760 
2761     bdrv_create_dirty_bitmap(bs, granularity, name, errp);
2762 
2763  out:
2764     aio_context_release(aio_context);
2765 }
2766 
2767 void qmp_block_dirty_bitmap_remove(const char *node, const char *name,
2768                                    Error **errp)
2769 {
2770     AioContext *aio_context;
2771     BlockDriverState *bs;
2772     BdrvDirtyBitmap *bitmap;
2773 
2774     bitmap = block_dirty_bitmap_lookup(node, name, &bs, &aio_context, errp);
2775     if (!bitmap || !bs) {
2776         return;
2777     }
2778 
2779     if (bdrv_dirty_bitmap_frozen(bitmap)) {
2780         error_setg(errp,
2781                    "Bitmap '%s' is currently frozen and cannot be removed",
2782                    name);
2783         goto out;
2784     }
2785     bdrv_dirty_bitmap_make_anon(bitmap);
2786     bdrv_release_dirty_bitmap(bs, bitmap);
2787 
2788  out:
2789     aio_context_release(aio_context);
2790 }
2791 
2792 /**
2793  * Completely clear a bitmap, for the purposes of synchronizing a bitmap
2794  * immediately after a full backup operation.
2795  */
2796 void qmp_block_dirty_bitmap_clear(const char *node, const char *name,
2797                                   Error **errp)
2798 {
2799     AioContext *aio_context;
2800     BdrvDirtyBitmap *bitmap;
2801     BlockDriverState *bs;
2802 
2803     bitmap = block_dirty_bitmap_lookup(node, name, &bs, &aio_context, errp);
2804     if (!bitmap || !bs) {
2805         return;
2806     }
2807 
2808     if (bdrv_dirty_bitmap_frozen(bitmap)) {
2809         error_setg(errp,
2810                    "Bitmap '%s' is currently frozen and cannot be modified",
2811                    name);
2812         goto out;
2813     } else if (!bdrv_dirty_bitmap_enabled(bitmap)) {
2814         error_setg(errp,
2815                    "Bitmap '%s' is currently disabled and cannot be cleared",
2816                    name);
2817         goto out;
2818     }
2819 
2820     bdrv_clear_dirty_bitmap(bitmap, NULL);
2821 
2822  out:
2823     aio_context_release(aio_context);
2824 }
2825 
2826 void hmp_drive_del(Monitor *mon, const QDict *qdict)
2827 {
2828     const char *id = qdict_get_str(qdict, "id");
2829     BlockBackend *blk;
2830     BlockDriverState *bs;
2831     AioContext *aio_context;
2832     Error *local_err = NULL;
2833 
2834     bs = bdrv_find_node(id);
2835     if (bs) {
2836         qmp_x_blockdev_del(id, &local_err);
2837         if (local_err) {
2838             error_report_err(local_err);
2839         }
2840         return;
2841     }
2842 
2843     blk = blk_by_name(id);
2844     if (!blk) {
2845         error_report("Device '%s' not found", id);
2846         return;
2847     }
2848 
2849     if (!blk_legacy_dinfo(blk)) {
2850         error_report("Deleting device added with blockdev-add"
2851                      " is not supported");
2852         return;
2853     }
2854 
2855     aio_context = blk_get_aio_context(blk);
2856     aio_context_acquire(aio_context);
2857 
2858     bs = blk_bs(blk);
2859     if (bs) {
2860         if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_DRIVE_DEL, &local_err)) {
2861             error_report_err(local_err);
2862             aio_context_release(aio_context);
2863             return;
2864         }
2865 
2866         blk_remove_bs(blk);
2867     }
2868 
2869     /* Make the BlockBackend and the attached BlockDriverState anonymous */
2870     monitor_remove_blk(blk);
2871 
2872     /* If this BlockBackend has a device attached to it, its refcount will be
2873      * decremented when the device is removed; otherwise we have to do so here.
2874      */
2875     if (blk_get_attached_dev(blk)) {
2876         /* Further I/O must not pause the guest */
2877         blk_set_on_error(blk, BLOCKDEV_ON_ERROR_REPORT,
2878                          BLOCKDEV_ON_ERROR_REPORT);
2879     } else {
2880         blk_unref(blk);
2881     }
2882 
2883     aio_context_release(aio_context);
2884 }
2885 
2886 void qmp_block_resize(bool has_device, const char *device,
2887                       bool has_node_name, const char *node_name,
2888                       int64_t size, Error **errp)
2889 {
2890     Error *local_err = NULL;
2891     BlockBackend *blk = NULL;
2892     BlockDriverState *bs;
2893     AioContext *aio_context;
2894     int ret;
2895 
2896     bs = bdrv_lookup_bs(has_device ? device : NULL,
2897                         has_node_name ? node_name : NULL,
2898                         &local_err);
2899     if (local_err) {
2900         error_propagate(errp, local_err);
2901         return;
2902     }
2903 
2904     aio_context = bdrv_get_aio_context(bs);
2905     aio_context_acquire(aio_context);
2906 
2907     if (!bdrv_is_first_non_filter(bs)) {
2908         error_setg(errp, QERR_FEATURE_DISABLED, "resize");
2909         goto out;
2910     }
2911 
2912     if (size < 0) {
2913         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "size", "a >0 size");
2914         goto out;
2915     }
2916 
2917     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_RESIZE, NULL)) {
2918         error_setg(errp, QERR_DEVICE_IN_USE, device);
2919         goto out;
2920     }
2921 
2922     blk = blk_new(BLK_PERM_RESIZE, BLK_PERM_ALL);
2923     ret = blk_insert_bs(blk, bs, errp);
2924     if (ret < 0) {
2925         goto out;
2926     }
2927 
2928     /* complete all in-flight operations before resizing the device */
2929     bdrv_drain_all();
2930 
2931     ret = blk_truncate(blk, size);
2932     switch (ret) {
2933     case 0:
2934         break;
2935     case -ENOMEDIUM:
2936         error_setg(errp, QERR_DEVICE_HAS_NO_MEDIUM, device);
2937         break;
2938     case -ENOTSUP:
2939         error_setg(errp, QERR_UNSUPPORTED);
2940         break;
2941     case -EACCES:
2942         error_setg(errp, "Device '%s' is read only", device);
2943         break;
2944     case -EBUSY:
2945         error_setg(errp, QERR_DEVICE_IN_USE, device);
2946         break;
2947     default:
2948         error_setg_errno(errp, -ret, "Could not resize");
2949         break;
2950     }
2951 
2952 out:
2953     blk_unref(blk);
2954     aio_context_release(aio_context);
2955 }
2956 
2957 void qmp_block_stream(bool has_job_id, const char *job_id, const char *device,
2958                       bool has_base, const char *base,
2959                       bool has_base_node, const char *base_node,
2960                       bool has_backing_file, const char *backing_file,
2961                       bool has_speed, int64_t speed,
2962                       bool has_on_error, BlockdevOnError on_error,
2963                       Error **errp)
2964 {
2965     BlockDriverState *bs, *iter;
2966     BlockDriverState *base_bs = NULL;
2967     AioContext *aio_context;
2968     Error *local_err = NULL;
2969     const char *base_name = NULL;
2970 
2971     if (!has_on_error) {
2972         on_error = BLOCKDEV_ON_ERROR_REPORT;
2973     }
2974 
2975     bs = bdrv_lookup_bs(device, device, errp);
2976     if (!bs) {
2977         return;
2978     }
2979 
2980     aio_context = bdrv_get_aio_context(bs);
2981     aio_context_acquire(aio_context);
2982 
2983     if (has_base && has_base_node) {
2984         error_setg(errp, "'base' and 'base-node' cannot be specified "
2985                    "at the same time");
2986         goto out;
2987     }
2988 
2989     if (has_base) {
2990         base_bs = bdrv_find_backing_image(bs, base);
2991         if (base_bs == NULL) {
2992             error_setg(errp, QERR_BASE_NOT_FOUND, base);
2993             goto out;
2994         }
2995         assert(bdrv_get_aio_context(base_bs) == aio_context);
2996         base_name = base;
2997     }
2998 
2999     if (has_base_node) {
3000         base_bs = bdrv_lookup_bs(NULL, base_node, errp);
3001         if (!base_bs) {
3002             goto out;
3003         }
3004         if (bs == base_bs || !bdrv_chain_contains(bs, base_bs)) {
3005             error_setg(errp, "Node '%s' is not a backing image of '%s'",
3006                        base_node, device);
3007             goto out;
3008         }
3009         assert(bdrv_get_aio_context(base_bs) == aio_context);
3010         base_name = base_bs->filename;
3011     }
3012 
3013     /* Check for op blockers in the whole chain between bs and base */
3014     for (iter = bs; iter && iter != base_bs; iter = backing_bs(iter)) {
3015         if (bdrv_op_is_blocked(iter, BLOCK_OP_TYPE_STREAM, errp)) {
3016             goto out;
3017         }
3018     }
3019 
3020     /* if we are streaming the entire chain, the result will have no backing
3021      * file, and specifying one is therefore an error */
3022     if (base_bs == NULL && has_backing_file) {
3023         error_setg(errp, "backing file specified, but streaming the "
3024                          "entire chain");
3025         goto out;
3026     }
3027 
3028     /* backing_file string overrides base bs filename */
3029     base_name = has_backing_file ? backing_file : base_name;
3030 
3031     stream_start(has_job_id ? job_id : NULL, bs, base_bs, base_name,
3032                  has_speed ? speed : 0, on_error, &local_err);
3033     if (local_err) {
3034         error_propagate(errp, local_err);
3035         goto out;
3036     }
3037 
3038     trace_qmp_block_stream(bs, bs->job);
3039 
3040 out:
3041     aio_context_release(aio_context);
3042 }
3043 
3044 void qmp_block_commit(bool has_job_id, const char *job_id, const char *device,
3045                       bool has_base, const char *base,
3046                       bool has_top, const char *top,
3047                       bool has_backing_file, const char *backing_file,
3048                       bool has_speed, int64_t speed,
3049                       bool has_filter_node_name, const char *filter_node_name,
3050                       Error **errp)
3051 {
3052     BlockDriverState *bs;
3053     BlockDriverState *iter;
3054     BlockDriverState *base_bs, *top_bs;
3055     AioContext *aio_context;
3056     Error *local_err = NULL;
3057     /* This will be part of the QMP command, if/when the
3058      * BlockdevOnError change for blkmirror makes it in
3059      */
3060     BlockdevOnError on_error = BLOCKDEV_ON_ERROR_REPORT;
3061 
3062     if (!has_speed) {
3063         speed = 0;
3064     }
3065     if (!has_filter_node_name) {
3066         filter_node_name = NULL;
3067     }
3068 
3069     /* Important Note:
3070      *  libvirt relies on the DeviceNotFound error class in order to probe for
3071      *  live commit feature versions; for this to work, we must make sure to
3072      *  perform the device lookup before any generic errors that may occur in a
3073      *  scenario in which all optional arguments are omitted. */
3074     bs = qmp_get_root_bs(device, &local_err);
3075     if (!bs) {
3076         bs = bdrv_lookup_bs(device, device, NULL);
3077         if (!bs) {
3078             error_free(local_err);
3079             error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
3080                       "Device '%s' not found", device);
3081         } else {
3082             error_propagate(errp, local_err);
3083         }
3084         return;
3085     }
3086 
3087     aio_context = bdrv_get_aio_context(bs);
3088     aio_context_acquire(aio_context);
3089 
3090     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_COMMIT_SOURCE, errp)) {
3091         goto out;
3092     }
3093 
3094     /* default top_bs is the active layer */
3095     top_bs = bs;
3096 
3097     if (has_top && top) {
3098         if (strcmp(bs->filename, top) != 0) {
3099             top_bs = bdrv_find_backing_image(bs, top);
3100         }
3101     }
3102 
3103     if (top_bs == NULL) {
3104         error_setg(errp, "Top image file %s not found", top ? top : "NULL");
3105         goto out;
3106     }
3107 
3108     assert(bdrv_get_aio_context(top_bs) == aio_context);
3109 
3110     if (has_base && base) {
3111         base_bs = bdrv_find_backing_image(top_bs, base);
3112     } else {
3113         base_bs = bdrv_find_base(top_bs);
3114     }
3115 
3116     if (base_bs == NULL) {
3117         error_setg(errp, QERR_BASE_NOT_FOUND, base ? base : "NULL");
3118         goto out;
3119     }
3120 
3121     assert(bdrv_get_aio_context(base_bs) == aio_context);
3122 
3123     for (iter = top_bs; iter != backing_bs(base_bs); iter = backing_bs(iter)) {
3124         if (bdrv_op_is_blocked(iter, BLOCK_OP_TYPE_COMMIT_TARGET, errp)) {
3125             goto out;
3126         }
3127     }
3128 
3129     /* Do not allow attempts to commit an image into itself */
3130     if (top_bs == base_bs) {
3131         error_setg(errp, "cannot commit an image into itself");
3132         goto out;
3133     }
3134 
3135     if (top_bs == bs) {
3136         if (has_backing_file) {
3137             error_setg(errp, "'backing-file' specified,"
3138                              " but 'top' is the active layer");
3139             goto out;
3140         }
3141         commit_active_start(has_job_id ? job_id : NULL, bs, base_bs,
3142                             BLOCK_JOB_DEFAULT, speed, on_error,
3143                             filter_node_name, NULL, NULL, &local_err, false);
3144     } else {
3145         BlockDriverState *overlay_bs = bdrv_find_overlay(bs, top_bs);
3146         if (bdrv_op_is_blocked(overlay_bs, BLOCK_OP_TYPE_COMMIT_TARGET, errp)) {
3147             goto out;
3148         }
3149         commit_start(has_job_id ? job_id : NULL, bs, base_bs, top_bs, speed,
3150                      on_error, has_backing_file ? backing_file : NULL,
3151                      filter_node_name, &local_err);
3152     }
3153     if (local_err != NULL) {
3154         error_propagate(errp, local_err);
3155         goto out;
3156     }
3157 
3158 out:
3159     aio_context_release(aio_context);
3160 }
3161 
3162 static BlockJob *do_drive_backup(DriveBackup *backup, BlockJobTxn *txn,
3163                                  Error **errp)
3164 {
3165     BlockDriverState *bs;
3166     BlockDriverState *target_bs;
3167     BlockDriverState *source = NULL;
3168     BlockJob *job = NULL;
3169     BdrvDirtyBitmap *bmap = NULL;
3170     AioContext *aio_context;
3171     QDict *options = NULL;
3172     Error *local_err = NULL;
3173     int flags;
3174     int64_t size;
3175 
3176     if (!backup->has_speed) {
3177         backup->speed = 0;
3178     }
3179     if (!backup->has_on_source_error) {
3180         backup->on_source_error = BLOCKDEV_ON_ERROR_REPORT;
3181     }
3182     if (!backup->has_on_target_error) {
3183         backup->on_target_error = BLOCKDEV_ON_ERROR_REPORT;
3184     }
3185     if (!backup->has_mode) {
3186         backup->mode = NEW_IMAGE_MODE_ABSOLUTE_PATHS;
3187     }
3188     if (!backup->has_job_id) {
3189         backup->job_id = NULL;
3190     }
3191     if (!backup->has_compress) {
3192         backup->compress = false;
3193     }
3194 
3195     bs = qmp_get_root_bs(backup->device, errp);
3196     if (!bs) {
3197         return NULL;
3198     }
3199 
3200     aio_context = bdrv_get_aio_context(bs);
3201     aio_context_acquire(aio_context);
3202 
3203     if (!backup->has_format) {
3204         backup->format = backup->mode == NEW_IMAGE_MODE_EXISTING ?
3205                          NULL : (char*) bs->drv->format_name;
3206     }
3207 
3208     /* Early check to avoid creating target */
3209     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_BACKUP_SOURCE, errp)) {
3210         goto out;
3211     }
3212 
3213     flags = bs->open_flags | BDRV_O_RDWR;
3214 
3215     /* See if we have a backing HD we can use to create our new image
3216      * on top of. */
3217     if (backup->sync == MIRROR_SYNC_MODE_TOP) {
3218         source = backing_bs(bs);
3219         if (!source) {
3220             backup->sync = MIRROR_SYNC_MODE_FULL;
3221         }
3222     }
3223     if (backup->sync == MIRROR_SYNC_MODE_NONE) {
3224         source = bs;
3225     }
3226 
3227     size = bdrv_getlength(bs);
3228     if (size < 0) {
3229         error_setg_errno(errp, -size, "bdrv_getlength failed");
3230         goto out;
3231     }
3232 
3233     if (backup->mode != NEW_IMAGE_MODE_EXISTING) {
3234         assert(backup->format);
3235         if (source) {
3236             bdrv_img_create(backup->target, backup->format, source->filename,
3237                             source->drv->format_name, NULL,
3238                             size, flags, &local_err, false);
3239         } else {
3240             bdrv_img_create(backup->target, backup->format, NULL, NULL, NULL,
3241                             size, flags, &local_err, false);
3242         }
3243     }
3244 
3245     if (local_err) {
3246         error_propagate(errp, local_err);
3247         goto out;
3248     }
3249 
3250     if (backup->format) {
3251         options = qdict_new();
3252         qdict_put(options, "driver", qstring_from_str(backup->format));
3253     }
3254 
3255     target_bs = bdrv_open(backup->target, NULL, options, flags, errp);
3256     if (!target_bs) {
3257         goto out;
3258     }
3259 
3260     bdrv_set_aio_context(target_bs, aio_context);
3261 
3262     if (backup->has_bitmap) {
3263         bmap = bdrv_find_dirty_bitmap(bs, backup->bitmap);
3264         if (!bmap) {
3265             error_setg(errp, "Bitmap '%s' could not be found", backup->bitmap);
3266             bdrv_unref(target_bs);
3267             goto out;
3268         }
3269     }
3270 
3271     job = backup_job_create(backup->job_id, bs, target_bs, backup->speed,
3272                             backup->sync, bmap, backup->compress,
3273                             backup->on_source_error, backup->on_target_error,
3274                             BLOCK_JOB_DEFAULT, NULL, NULL, txn, &local_err);
3275     bdrv_unref(target_bs);
3276     if (local_err != NULL) {
3277         error_propagate(errp, local_err);
3278         goto out;
3279     }
3280 
3281 out:
3282     aio_context_release(aio_context);
3283     return job;
3284 }
3285 
3286 void qmp_drive_backup(DriveBackup *arg, Error **errp)
3287 {
3288 
3289     BlockJob *job;
3290     job = do_drive_backup(arg, NULL, errp);
3291     if (job) {
3292         block_job_start(job);
3293     }
3294 }
3295 
3296 BlockDeviceInfoList *qmp_query_named_block_nodes(Error **errp)
3297 {
3298     return bdrv_named_nodes_list(errp);
3299 }
3300 
3301 BlockJob *do_blockdev_backup(BlockdevBackup *backup, BlockJobTxn *txn,
3302                              Error **errp)
3303 {
3304     BlockDriverState *bs;
3305     BlockDriverState *target_bs;
3306     Error *local_err = NULL;
3307     AioContext *aio_context;
3308     BlockJob *job = NULL;
3309 
3310     if (!backup->has_speed) {
3311         backup->speed = 0;
3312     }
3313     if (!backup->has_on_source_error) {
3314         backup->on_source_error = BLOCKDEV_ON_ERROR_REPORT;
3315     }
3316     if (!backup->has_on_target_error) {
3317         backup->on_target_error = BLOCKDEV_ON_ERROR_REPORT;
3318     }
3319     if (!backup->has_job_id) {
3320         backup->job_id = NULL;
3321     }
3322     if (!backup->has_compress) {
3323         backup->compress = false;
3324     }
3325 
3326     bs = qmp_get_root_bs(backup->device, errp);
3327     if (!bs) {
3328         return NULL;
3329     }
3330 
3331     aio_context = bdrv_get_aio_context(bs);
3332     aio_context_acquire(aio_context);
3333 
3334     target_bs = bdrv_lookup_bs(backup->target, backup->target, errp);
3335     if (!target_bs) {
3336         goto out;
3337     }
3338 
3339     if (bdrv_get_aio_context(target_bs) != aio_context) {
3340         if (!bdrv_has_blk(target_bs)) {
3341             /* The target BDS is not attached, we can safely move it to another
3342              * AioContext. */
3343             bdrv_set_aio_context(target_bs, aio_context);
3344         } else {
3345             error_setg(errp, "Target is attached to a different thread from "
3346                              "source.");
3347             goto out;
3348         }
3349     }
3350     job = backup_job_create(backup->job_id, bs, target_bs, backup->speed,
3351                             backup->sync, NULL, backup->compress,
3352                             backup->on_source_error, backup->on_target_error,
3353                             BLOCK_JOB_DEFAULT, NULL, NULL, txn, &local_err);
3354     if (local_err != NULL) {
3355         error_propagate(errp, local_err);
3356     }
3357 out:
3358     aio_context_release(aio_context);
3359     return job;
3360 }
3361 
3362 void qmp_blockdev_backup(BlockdevBackup *arg, Error **errp)
3363 {
3364     BlockJob *job;
3365     job = do_blockdev_backup(arg, NULL, errp);
3366     if (job) {
3367         block_job_start(job);
3368     }
3369 }
3370 
3371 /* Parameter check and block job starting for drive mirroring.
3372  * Caller should hold @device and @target's aio context (must be the same).
3373  **/
3374 static void blockdev_mirror_common(const char *job_id, BlockDriverState *bs,
3375                                    BlockDriverState *target,
3376                                    bool has_replaces, const char *replaces,
3377                                    enum MirrorSyncMode sync,
3378                                    BlockMirrorBackingMode backing_mode,
3379                                    bool has_speed, int64_t speed,
3380                                    bool has_granularity, uint32_t granularity,
3381                                    bool has_buf_size, int64_t buf_size,
3382                                    bool has_on_source_error,
3383                                    BlockdevOnError on_source_error,
3384                                    bool has_on_target_error,
3385                                    BlockdevOnError on_target_error,
3386                                    bool has_unmap, bool unmap,
3387                                    bool has_filter_node_name,
3388                                    const char *filter_node_name,
3389                                    Error **errp)
3390 {
3391 
3392     if (!has_speed) {
3393         speed = 0;
3394     }
3395     if (!has_on_source_error) {
3396         on_source_error = BLOCKDEV_ON_ERROR_REPORT;
3397     }
3398     if (!has_on_target_error) {
3399         on_target_error = BLOCKDEV_ON_ERROR_REPORT;
3400     }
3401     if (!has_granularity) {
3402         granularity = 0;
3403     }
3404     if (!has_buf_size) {
3405         buf_size = 0;
3406     }
3407     if (!has_unmap) {
3408         unmap = true;
3409     }
3410     if (!has_filter_node_name) {
3411         filter_node_name = NULL;
3412     }
3413 
3414     if (granularity != 0 && (granularity < 512 || granularity > 1048576 * 64)) {
3415         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "granularity",
3416                    "a value in range [512B, 64MB]");
3417         return;
3418     }
3419     if (granularity & (granularity - 1)) {
3420         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "granularity",
3421                    "power of 2");
3422         return;
3423     }
3424 
3425     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_MIRROR_SOURCE, errp)) {
3426         return;
3427     }
3428     if (bdrv_op_is_blocked(target, BLOCK_OP_TYPE_MIRROR_TARGET, errp)) {
3429         return;
3430     }
3431 
3432     if (!bs->backing && sync == MIRROR_SYNC_MODE_TOP) {
3433         sync = MIRROR_SYNC_MODE_FULL;
3434     }
3435 
3436     /* pass the node name to replace to mirror start since it's loose coupling
3437      * and will allow to check whether the node still exist at mirror completion
3438      */
3439     mirror_start(job_id, bs, target,
3440                  has_replaces ? replaces : NULL,
3441                  speed, granularity, buf_size, sync, backing_mode,
3442                  on_source_error, on_target_error, unmap, filter_node_name,
3443                  errp);
3444 }
3445 
3446 void qmp_drive_mirror(DriveMirror *arg, Error **errp)
3447 {
3448     BlockDriverState *bs;
3449     BlockDriverState *source, *target_bs;
3450     AioContext *aio_context;
3451     BlockMirrorBackingMode backing_mode;
3452     Error *local_err = NULL;
3453     QDict *options = NULL;
3454     int flags;
3455     int64_t size;
3456     const char *format = arg->format;
3457 
3458     bs = qmp_get_root_bs(arg->device, errp);
3459     if (!bs) {
3460         return;
3461     }
3462 
3463     aio_context = bdrv_get_aio_context(bs);
3464     aio_context_acquire(aio_context);
3465 
3466     if (!arg->has_mode) {
3467         arg->mode = NEW_IMAGE_MODE_ABSOLUTE_PATHS;
3468     }
3469 
3470     if (!arg->has_format) {
3471         format = (arg->mode == NEW_IMAGE_MODE_EXISTING
3472                   ? NULL : bs->drv->format_name);
3473     }
3474 
3475     flags = bs->open_flags | BDRV_O_RDWR;
3476     source = backing_bs(bs);
3477     if (!source && arg->sync == MIRROR_SYNC_MODE_TOP) {
3478         arg->sync = MIRROR_SYNC_MODE_FULL;
3479     }
3480     if (arg->sync == MIRROR_SYNC_MODE_NONE) {
3481         source = bs;
3482     }
3483 
3484     size = bdrv_getlength(bs);
3485     if (size < 0) {
3486         error_setg_errno(errp, -size, "bdrv_getlength failed");
3487         goto out;
3488     }
3489 
3490     if (arg->has_replaces) {
3491         BlockDriverState *to_replace_bs;
3492         AioContext *replace_aio_context;
3493         int64_t replace_size;
3494 
3495         if (!arg->has_node_name) {
3496             error_setg(errp, "a node-name must be provided when replacing a"
3497                              " named node of the graph");
3498             goto out;
3499         }
3500 
3501         to_replace_bs = check_to_replace_node(bs, arg->replaces, &local_err);
3502 
3503         if (!to_replace_bs) {
3504             error_propagate(errp, local_err);
3505             goto out;
3506         }
3507 
3508         replace_aio_context = bdrv_get_aio_context(to_replace_bs);
3509         aio_context_acquire(replace_aio_context);
3510         replace_size = bdrv_getlength(to_replace_bs);
3511         aio_context_release(replace_aio_context);
3512 
3513         if (size != replace_size) {
3514             error_setg(errp, "cannot replace image with a mirror image of "
3515                              "different size");
3516             goto out;
3517         }
3518     }
3519 
3520     if (arg->mode == NEW_IMAGE_MODE_ABSOLUTE_PATHS) {
3521         backing_mode = MIRROR_SOURCE_BACKING_CHAIN;
3522     } else {
3523         backing_mode = MIRROR_OPEN_BACKING_CHAIN;
3524     }
3525 
3526     if ((arg->sync == MIRROR_SYNC_MODE_FULL || !source)
3527         && arg->mode != NEW_IMAGE_MODE_EXISTING)
3528     {
3529         /* create new image w/o backing file */
3530         assert(format);
3531         bdrv_img_create(arg->target, format,
3532                         NULL, NULL, NULL, size, flags, &local_err, false);
3533     } else {
3534         switch (arg->mode) {
3535         case NEW_IMAGE_MODE_EXISTING:
3536             break;
3537         case NEW_IMAGE_MODE_ABSOLUTE_PATHS:
3538             /* create new image with backing file */
3539             bdrv_img_create(arg->target, format,
3540                             source->filename,
3541                             source->drv->format_name,
3542                             NULL, size, flags, &local_err, false);
3543             break;
3544         default:
3545             abort();
3546         }
3547     }
3548 
3549     if (local_err) {
3550         error_propagate(errp, local_err);
3551         goto out;
3552     }
3553 
3554     options = qdict_new();
3555     if (arg->has_node_name) {
3556         qdict_put(options, "node-name", qstring_from_str(arg->node_name));
3557     }
3558     if (format) {
3559         qdict_put(options, "driver", qstring_from_str(format));
3560     }
3561 
3562     /* Mirroring takes care of copy-on-write using the source's backing
3563      * file.
3564      */
3565     target_bs = bdrv_open(arg->target, NULL, options,
3566                           flags | BDRV_O_NO_BACKING, errp);
3567     if (!target_bs) {
3568         goto out;
3569     }
3570 
3571     bdrv_set_aio_context(target_bs, aio_context);
3572 
3573     blockdev_mirror_common(arg->has_job_id ? arg->job_id : NULL, bs, target_bs,
3574                            arg->has_replaces, arg->replaces, arg->sync,
3575                            backing_mode, arg->has_speed, arg->speed,
3576                            arg->has_granularity, arg->granularity,
3577                            arg->has_buf_size, arg->buf_size,
3578                            arg->has_on_source_error, arg->on_source_error,
3579                            arg->has_on_target_error, arg->on_target_error,
3580                            arg->has_unmap, arg->unmap,
3581                            false, NULL,
3582                            &local_err);
3583     bdrv_unref(target_bs);
3584     error_propagate(errp, local_err);
3585 out:
3586     aio_context_release(aio_context);
3587 }
3588 
3589 void qmp_blockdev_mirror(bool has_job_id, const char *job_id,
3590                          const char *device, const char *target,
3591                          bool has_replaces, const char *replaces,
3592                          MirrorSyncMode sync,
3593                          bool has_speed, int64_t speed,
3594                          bool has_granularity, uint32_t granularity,
3595                          bool has_buf_size, int64_t buf_size,
3596                          bool has_on_source_error,
3597                          BlockdevOnError on_source_error,
3598                          bool has_on_target_error,
3599                          BlockdevOnError on_target_error,
3600                          bool has_filter_node_name,
3601                          const char *filter_node_name,
3602                          Error **errp)
3603 {
3604     BlockDriverState *bs;
3605     BlockDriverState *target_bs;
3606     AioContext *aio_context;
3607     BlockMirrorBackingMode backing_mode = MIRROR_LEAVE_BACKING_CHAIN;
3608     Error *local_err = NULL;
3609 
3610     bs = qmp_get_root_bs(device, errp);
3611     if (!bs) {
3612         return;
3613     }
3614 
3615     target_bs = bdrv_lookup_bs(target, target, errp);
3616     if (!target_bs) {
3617         return;
3618     }
3619 
3620     aio_context = bdrv_get_aio_context(bs);
3621     aio_context_acquire(aio_context);
3622 
3623     bdrv_set_aio_context(target_bs, aio_context);
3624 
3625     blockdev_mirror_common(has_job_id ? job_id : NULL, bs, target_bs,
3626                            has_replaces, replaces, sync, backing_mode,
3627                            has_speed, speed,
3628                            has_granularity, granularity,
3629                            has_buf_size, buf_size,
3630                            has_on_source_error, on_source_error,
3631                            has_on_target_error, on_target_error,
3632                            true, true,
3633                            has_filter_node_name, filter_node_name,
3634                            &local_err);
3635     error_propagate(errp, local_err);
3636 
3637     aio_context_release(aio_context);
3638 }
3639 
3640 /* Get a block job using its ID and acquire its AioContext */
3641 static BlockJob *find_block_job(const char *id, AioContext **aio_context,
3642                                 Error **errp)
3643 {
3644     BlockJob *job;
3645 
3646     assert(id != NULL);
3647 
3648     *aio_context = NULL;
3649 
3650     job = block_job_get(id);
3651 
3652     if (!job) {
3653         error_set(errp, ERROR_CLASS_DEVICE_NOT_ACTIVE,
3654                   "Block job '%s' not found", id);
3655         return NULL;
3656     }
3657 
3658     *aio_context = blk_get_aio_context(job->blk);
3659     aio_context_acquire(*aio_context);
3660 
3661     return job;
3662 }
3663 
3664 void qmp_block_job_set_speed(const char *device, int64_t speed, Error **errp)
3665 {
3666     AioContext *aio_context;
3667     BlockJob *job = find_block_job(device, &aio_context, errp);
3668 
3669     if (!job) {
3670         return;
3671     }
3672 
3673     block_job_set_speed(job, speed, errp);
3674     aio_context_release(aio_context);
3675 }
3676 
3677 void qmp_block_job_cancel(const char *device,
3678                           bool has_force, bool force, Error **errp)
3679 {
3680     AioContext *aio_context;
3681     BlockJob *job = find_block_job(device, &aio_context, errp);
3682 
3683     if (!job) {
3684         return;
3685     }
3686 
3687     if (!has_force) {
3688         force = false;
3689     }
3690 
3691     if (block_job_user_paused(job) && !force) {
3692         error_setg(errp, "The block job for device '%s' is currently paused",
3693                    device);
3694         goto out;
3695     }
3696 
3697     trace_qmp_block_job_cancel(job);
3698     block_job_cancel(job);
3699 out:
3700     aio_context_release(aio_context);
3701 }
3702 
3703 void qmp_block_job_pause(const char *device, Error **errp)
3704 {
3705     AioContext *aio_context;
3706     BlockJob *job = find_block_job(device, &aio_context, errp);
3707 
3708     if (!job || block_job_user_paused(job)) {
3709         return;
3710     }
3711 
3712     trace_qmp_block_job_pause(job);
3713     block_job_user_pause(job);
3714     aio_context_release(aio_context);
3715 }
3716 
3717 void qmp_block_job_resume(const char *device, Error **errp)
3718 {
3719     AioContext *aio_context;
3720     BlockJob *job = find_block_job(device, &aio_context, errp);
3721 
3722     if (!job || !block_job_user_paused(job)) {
3723         return;
3724     }
3725 
3726     trace_qmp_block_job_resume(job);
3727     block_job_iostatus_reset(job);
3728     block_job_user_resume(job);
3729     aio_context_release(aio_context);
3730 }
3731 
3732 void qmp_block_job_complete(const char *device, Error **errp)
3733 {
3734     AioContext *aio_context;
3735     BlockJob *job = find_block_job(device, &aio_context, errp);
3736 
3737     if (!job) {
3738         return;
3739     }
3740 
3741     trace_qmp_block_job_complete(job);
3742     block_job_complete(job, errp);
3743     aio_context_release(aio_context);
3744 }
3745 
3746 void qmp_change_backing_file(const char *device,
3747                              const char *image_node_name,
3748                              const char *backing_file,
3749                              Error **errp)
3750 {
3751     BlockDriverState *bs = NULL;
3752     AioContext *aio_context;
3753     BlockDriverState *image_bs = NULL;
3754     Error *local_err = NULL;
3755     bool ro;
3756     int open_flags;
3757     int ret;
3758 
3759     bs = qmp_get_root_bs(device, errp);
3760     if (!bs) {
3761         return;
3762     }
3763 
3764     aio_context = bdrv_get_aio_context(bs);
3765     aio_context_acquire(aio_context);
3766 
3767     image_bs = bdrv_lookup_bs(NULL, image_node_name, &local_err);
3768     if (local_err) {
3769         error_propagate(errp, local_err);
3770         goto out;
3771     }
3772 
3773     if (!image_bs) {
3774         error_setg(errp, "image file not found");
3775         goto out;
3776     }
3777 
3778     if (bdrv_find_base(image_bs) == image_bs) {
3779         error_setg(errp, "not allowing backing file change on an image "
3780                          "without a backing file");
3781         goto out;
3782     }
3783 
3784     /* even though we are not necessarily operating on bs, we need it to
3785      * determine if block ops are currently prohibited on the chain */
3786     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_CHANGE, errp)) {
3787         goto out;
3788     }
3789 
3790     /* final sanity check */
3791     if (!bdrv_chain_contains(bs, image_bs)) {
3792         error_setg(errp, "'%s' and image file are not in the same chain",
3793                    device);
3794         goto out;
3795     }
3796 
3797     /* if not r/w, reopen to make r/w */
3798     open_flags = image_bs->open_flags;
3799     ro = bdrv_is_read_only(image_bs);
3800 
3801     if (ro) {
3802         bdrv_reopen(image_bs, open_flags | BDRV_O_RDWR, &local_err);
3803         if (local_err) {
3804             error_propagate(errp, local_err);
3805             goto out;
3806         }
3807     }
3808 
3809     ret = bdrv_change_backing_file(image_bs, backing_file,
3810                                image_bs->drv ? image_bs->drv->format_name : "");
3811 
3812     if (ret < 0) {
3813         error_setg_errno(errp, -ret, "Could not change backing file to '%s'",
3814                          backing_file);
3815         /* don't exit here, so we can try to restore open flags if
3816          * appropriate */
3817     }
3818 
3819     if (ro) {
3820         bdrv_reopen(image_bs, open_flags, &local_err);
3821         error_propagate(errp, local_err);
3822     }
3823 
3824 out:
3825     aio_context_release(aio_context);
3826 }
3827 
3828 void hmp_drive_add_node(Monitor *mon, const char *optstr)
3829 {
3830     QemuOpts *opts;
3831     QDict *qdict;
3832     Error *local_err = NULL;
3833 
3834     opts = qemu_opts_parse_noisily(&qemu_drive_opts, optstr, false);
3835     if (!opts) {
3836         return;
3837     }
3838 
3839     qdict = qemu_opts_to_qdict(opts, NULL);
3840 
3841     if (!qdict_get_try_str(qdict, "node-name")) {
3842         QDECREF(qdict);
3843         error_report("'node-name' needs to be specified");
3844         goto out;
3845     }
3846 
3847     BlockDriverState *bs = bds_tree_init(qdict, &local_err);
3848     if (!bs) {
3849         error_report_err(local_err);
3850         goto out;
3851     }
3852 
3853     QTAILQ_INSERT_TAIL(&monitor_bdrv_states, bs, monitor_list);
3854 
3855 out:
3856     qemu_opts_del(opts);
3857 }
3858 
3859 void qmp_blockdev_add(BlockdevOptions *options, Error **errp)
3860 {
3861     BlockDriverState *bs;
3862     QObject *obj;
3863     Visitor *v = qobject_output_visitor_new(&obj);
3864     QDict *qdict;
3865     Error *local_err = NULL;
3866 
3867     visit_type_BlockdevOptions(v, NULL, &options, &local_err);
3868     if (local_err) {
3869         error_propagate(errp, local_err);
3870         goto fail;
3871     }
3872 
3873     visit_complete(v, &obj);
3874     qdict = qobject_to_qdict(obj);
3875 
3876     qdict_flatten(qdict);
3877 
3878     if (!qdict_get_try_str(qdict, "node-name")) {
3879         error_setg(errp, "'node-name' must be specified for the root node");
3880         goto fail;
3881     }
3882 
3883     bs = bds_tree_init(qdict, errp);
3884     if (!bs) {
3885         goto fail;
3886     }
3887 
3888     QTAILQ_INSERT_TAIL(&monitor_bdrv_states, bs, monitor_list);
3889 
3890     if (bs && bdrv_key_required(bs)) {
3891         QTAILQ_REMOVE(&monitor_bdrv_states, bs, monitor_list);
3892         bdrv_unref(bs);
3893         error_setg(errp, "blockdev-add doesn't support encrypted devices");
3894         goto fail;
3895     }
3896 
3897 fail:
3898     visit_free(v);
3899 }
3900 
3901 void qmp_x_blockdev_del(const char *node_name, Error **errp)
3902 {
3903     AioContext *aio_context;
3904     BlockDriverState *bs;
3905 
3906     bs = bdrv_find_node(node_name);
3907     if (!bs) {
3908         error_setg(errp, "Cannot find node %s", node_name);
3909         return;
3910     }
3911     if (bdrv_has_blk(bs)) {
3912         error_setg(errp, "Node %s is in use", node_name);
3913         return;
3914     }
3915     aio_context = bdrv_get_aio_context(bs);
3916     aio_context_acquire(aio_context);
3917 
3918     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_DRIVE_DEL, errp)) {
3919         goto out;
3920     }
3921 
3922     if (!bs->monitor_list.tqe_prev) {
3923         error_setg(errp, "Node %s is not owned by the monitor",
3924                    bs->node_name);
3925         goto out;
3926     }
3927 
3928     if (bs->refcnt > 1) {
3929         error_setg(errp, "Block device %s is in use",
3930                    bdrv_get_device_or_node_name(bs));
3931         goto out;
3932     }
3933 
3934     QTAILQ_REMOVE(&monitor_bdrv_states, bs, monitor_list);
3935     bdrv_unref(bs);
3936 
3937 out:
3938     aio_context_release(aio_context);
3939 }
3940 
3941 static BdrvChild *bdrv_find_child(BlockDriverState *parent_bs,
3942                                   const char *child_name)
3943 {
3944     BdrvChild *child;
3945 
3946     QLIST_FOREACH(child, &parent_bs->children, next) {
3947         if (strcmp(child->name, child_name) == 0) {
3948             return child;
3949         }
3950     }
3951 
3952     return NULL;
3953 }
3954 
3955 void qmp_x_blockdev_change(const char *parent, bool has_child,
3956                            const char *child, bool has_node,
3957                            const char *node, Error **errp)
3958 {
3959     BlockDriverState *parent_bs, *new_bs = NULL;
3960     BdrvChild *p_child;
3961 
3962     parent_bs = bdrv_lookup_bs(parent, parent, errp);
3963     if (!parent_bs) {
3964         return;
3965     }
3966 
3967     if (has_child == has_node) {
3968         if (has_child) {
3969             error_setg(errp, "The parameters child and node are in conflict");
3970         } else {
3971             error_setg(errp, "Either child or node must be specified");
3972         }
3973         return;
3974     }
3975 
3976     if (has_child) {
3977         p_child = bdrv_find_child(parent_bs, child);
3978         if (!p_child) {
3979             error_setg(errp, "Node '%s' does not have child '%s'",
3980                        parent, child);
3981             return;
3982         }
3983         bdrv_del_child(parent_bs, p_child, errp);
3984     }
3985 
3986     if (has_node) {
3987         new_bs = bdrv_find_node(node);
3988         if (!new_bs) {
3989             error_setg(errp, "Node '%s' not found", node);
3990             return;
3991         }
3992         bdrv_add_child(parent_bs, new_bs, errp);
3993     }
3994 }
3995 
3996 BlockJobInfoList *qmp_query_block_jobs(Error **errp)
3997 {
3998     BlockJobInfoList *head = NULL, **p_next = &head;
3999     BlockJob *job;
4000 
4001     for (job = block_job_next(NULL); job; job = block_job_next(job)) {
4002         BlockJobInfoList *elem;
4003         AioContext *aio_context;
4004 
4005         if (block_job_is_internal(job)) {
4006             continue;
4007         }
4008         elem = g_new0(BlockJobInfoList, 1);
4009         aio_context = blk_get_aio_context(job->blk);
4010         aio_context_acquire(aio_context);
4011         elem->value = block_job_query(job, errp);
4012         aio_context_release(aio_context);
4013         if (!elem->value) {
4014             g_free(elem);
4015             qapi_free_BlockJobInfoList(head);
4016             return NULL;
4017         }
4018         *p_next = elem;
4019         p_next = &elem->next;
4020     }
4021 
4022     return head;
4023 }
4024 
4025 QemuOptsList qemu_common_drive_opts = {
4026     .name = "drive",
4027     .head = QTAILQ_HEAD_INITIALIZER(qemu_common_drive_opts.head),
4028     .desc = {
4029         {
4030             .name = "snapshot",
4031             .type = QEMU_OPT_BOOL,
4032             .help = "enable/disable snapshot mode",
4033         },{
4034             .name = "aio",
4035             .type = QEMU_OPT_STRING,
4036             .help = "host AIO implementation (threads, native)",
4037         },{
4038             .name = BDRV_OPT_CACHE_WB,
4039             .type = QEMU_OPT_BOOL,
4040             .help = "Enable writeback mode",
4041         },{
4042             .name = "format",
4043             .type = QEMU_OPT_STRING,
4044             .help = "disk format (raw, qcow2, ...)",
4045         },{
4046             .name = "rerror",
4047             .type = QEMU_OPT_STRING,
4048             .help = "read error action",
4049         },{
4050             .name = "werror",
4051             .type = QEMU_OPT_STRING,
4052             .help = "write error action",
4053         },{
4054             .name = BDRV_OPT_READ_ONLY,
4055             .type = QEMU_OPT_BOOL,
4056             .help = "open drive file as read-only",
4057         },
4058 
4059         THROTTLE_OPTS,
4060 
4061         {
4062             .name = "throttling.group",
4063             .type = QEMU_OPT_STRING,
4064             .help = "name of the block throttling group",
4065         },{
4066             .name = "copy-on-read",
4067             .type = QEMU_OPT_BOOL,
4068             .help = "copy read data from backing file into image file",
4069         },{
4070             .name = "detect-zeroes",
4071             .type = QEMU_OPT_STRING,
4072             .help = "try to optimize zero writes (off, on, unmap)",
4073         },{
4074             .name = "stats-account-invalid",
4075             .type = QEMU_OPT_BOOL,
4076             .help = "whether to account for invalid I/O operations "
4077                     "in the statistics",
4078         },{
4079             .name = "stats-account-failed",
4080             .type = QEMU_OPT_BOOL,
4081             .help = "whether to account for failed I/O operations "
4082                     "in the statistics",
4083         },
4084         { /* end of list */ }
4085     },
4086 };
4087 
4088 QemuOptsList qemu_drive_opts = {
4089     .name = "drive",
4090     .head = QTAILQ_HEAD_INITIALIZER(qemu_drive_opts.head),
4091     .desc = {
4092         /*
4093          * no elements => accept any params
4094          * validation will happen later
4095          */
4096         { /* end of list */ }
4097     },
4098 };
4099