xref: /qemu/blockdev.c (revision b15c0f7d)
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/qdict.h"
39 #include "block/throttle-groups.h"
40 #include "monitor/monitor.h"
41 #include "qemu/error-report.h"
42 #include "qemu/option.h"
43 #include "qemu/qemu-print.h"
44 #include "qemu/config-file.h"
45 #include "qapi/qapi-commands-block.h"
46 #include "qapi/qapi-commands-transaction.h"
47 #include "qapi/qapi-visit-block-core.h"
48 #include "qapi/qmp/qdict.h"
49 #include "qapi/qmp/qnum.h"
50 #include "qapi/qmp/qstring.h"
51 #include "qapi/error.h"
52 #include "qapi/qmp/qerror.h"
53 #include "qapi/qmp/qlist.h"
54 #include "qapi/qobject-output-visitor.h"
55 #include "sysemu/sysemu.h"
56 #include "sysemu/iothread.h"
57 #include "block/block_int.h"
58 #include "block/trace.h"
59 #include "sysemu/arch_init.h"
60 #include "sysemu/qtest.h"
61 #include "qemu/cutils.h"
62 #include "qemu/help_option.h"
63 #include "qemu/throttle-options.h"
64 
65 static QTAILQ_HEAD(, BlockDriverState) monitor_bdrv_states =
66     QTAILQ_HEAD_INITIALIZER(monitor_bdrv_states);
67 
68 static int do_open_tray(const char *blk_name, const char *qdev_id,
69                         bool force, Error **errp);
70 static void blockdev_remove_medium(bool has_device, const char *device,
71                                    bool has_id, const char *id, Error **errp);
72 static void blockdev_insert_medium(bool has_device, const char *device,
73                                    bool has_id, const char *id,
74                                    const char *node_name, Error **errp);
75 
76 static const char *const if_name[IF_COUNT] = {
77     [IF_NONE] = "none",
78     [IF_IDE] = "ide",
79     [IF_SCSI] = "scsi",
80     [IF_FLOPPY] = "floppy",
81     [IF_PFLASH] = "pflash",
82     [IF_MTD] = "mtd",
83     [IF_SD] = "sd",
84     [IF_VIRTIO] = "virtio",
85     [IF_XEN] = "xen",
86 };
87 
88 static int if_max_devs[IF_COUNT] = {
89     /*
90      * Do not change these numbers!  They govern how drive option
91      * index maps to unit and bus.  That mapping is ABI.
92      *
93      * All controllers used to implement if=T drives need to support
94      * if_max_devs[T] units, for any T with if_max_devs[T] != 0.
95      * Otherwise, some index values map to "impossible" bus, unit
96      * values.
97      *
98      * For instance, if you change [IF_SCSI] to 255, -drive
99      * if=scsi,index=12 no longer means bus=1,unit=5, but
100      * bus=0,unit=12.  With an lsi53c895a controller (7 units max),
101      * the drive can't be set up.  Regression.
102      */
103     [IF_IDE] = 2,
104     [IF_SCSI] = 7,
105 };
106 
107 /**
108  * Boards may call this to offer board-by-board overrides
109  * of the default, global values.
110  */
111 void override_max_devs(BlockInterfaceType type, int max_devs)
112 {
113     BlockBackend *blk;
114     DriveInfo *dinfo;
115 
116     if (max_devs <= 0) {
117         return;
118     }
119 
120     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
121         dinfo = blk_legacy_dinfo(blk);
122         if (dinfo->type == type) {
123             fprintf(stderr, "Cannot override units-per-bus property of"
124                     " the %s interface, because a drive of that type has"
125                     " already been added.\n", if_name[type]);
126             g_assert_not_reached();
127         }
128     }
129 
130     if_max_devs[type] = max_devs;
131 }
132 
133 /*
134  * We automatically delete the drive when a device using it gets
135  * unplugged.  Questionable feature, but we can't just drop it.
136  * Device models call blockdev_mark_auto_del() to schedule the
137  * automatic deletion, and generic qdev code calls blockdev_auto_del()
138  * when deletion is actually safe.
139  */
140 void blockdev_mark_auto_del(BlockBackend *blk)
141 {
142     DriveInfo *dinfo = blk_legacy_dinfo(blk);
143     BlockDriverState *bs = blk_bs(blk);
144     AioContext *aio_context;
145 
146     if (!dinfo) {
147         return;
148     }
149 
150     if (bs) {
151         aio_context = bdrv_get_aio_context(bs);
152         aio_context_acquire(aio_context);
153 
154         if (bs->job) {
155             job_cancel(&bs->job->job, false);
156         }
157 
158         aio_context_release(aio_context);
159     }
160 
161     dinfo->auto_del = 1;
162 }
163 
164 void blockdev_auto_del(BlockBackend *blk)
165 {
166     DriveInfo *dinfo = blk_legacy_dinfo(blk);
167 
168     if (dinfo && dinfo->auto_del) {
169         monitor_remove_blk(blk);
170         blk_unref(blk);
171     }
172 }
173 
174 /**
175  * Returns the current mapping of how many units per bus
176  * a particular interface can support.
177  *
178  *  A positive integer indicates n units per bus.
179  *  0 implies the mapping has not been established.
180  * -1 indicates an invalid BlockInterfaceType was given.
181  */
182 int drive_get_max_devs(BlockInterfaceType type)
183 {
184     if (type >= IF_IDE && type < IF_COUNT) {
185         return if_max_devs[type];
186     }
187 
188     return -1;
189 }
190 
191 static int drive_index_to_bus_id(BlockInterfaceType type, int index)
192 {
193     int max_devs = if_max_devs[type];
194     return max_devs ? index / max_devs : 0;
195 }
196 
197 static int drive_index_to_unit_id(BlockInterfaceType type, int index)
198 {
199     int max_devs = if_max_devs[type];
200     return max_devs ? index % max_devs : index;
201 }
202 
203 QemuOpts *drive_def(const char *optstr)
204 {
205     return qemu_opts_parse_noisily(qemu_find_opts("drive"), optstr, false);
206 }
207 
208 QemuOpts *drive_add(BlockInterfaceType type, int index, const char *file,
209                     const char *optstr)
210 {
211     QemuOpts *opts;
212 
213     opts = drive_def(optstr);
214     if (!opts) {
215         return NULL;
216     }
217     if (type != IF_DEFAULT) {
218         qemu_opt_set(opts, "if", if_name[type], &error_abort);
219     }
220     if (index >= 0) {
221         qemu_opt_set_number(opts, "index", index, &error_abort);
222     }
223     if (file)
224         qemu_opt_set(opts, "file", file, &error_abort);
225     return opts;
226 }
227 
228 DriveInfo *drive_get(BlockInterfaceType type, int bus, int unit)
229 {
230     BlockBackend *blk;
231     DriveInfo *dinfo;
232 
233     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
234         dinfo = blk_legacy_dinfo(blk);
235         if (dinfo && dinfo->type == type
236             && dinfo->bus == bus && dinfo->unit == unit) {
237             return dinfo;
238         }
239     }
240 
241     return NULL;
242 }
243 
244 void drive_check_orphaned(void)
245 {
246     BlockBackend *blk;
247     DriveInfo *dinfo;
248     Location loc;
249     bool orphans = false;
250 
251     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
252         dinfo = blk_legacy_dinfo(blk);
253         if (!blk_get_attached_dev(blk) && !dinfo->is_default &&
254             dinfo->type != IF_NONE) {
255             loc_push_none(&loc);
256             qemu_opts_loc_restore(dinfo->opts);
257             error_report("machine type does not support"
258                          " if=%s,bus=%d,unit=%d",
259                          if_name[dinfo->type], dinfo->bus, dinfo->unit);
260             loc_pop(&loc);
261             orphans = true;
262         }
263     }
264 
265     if (orphans) {
266         exit(1);
267     }
268 }
269 
270 DriveInfo *drive_get_by_index(BlockInterfaceType type, int index)
271 {
272     return drive_get(type,
273                      drive_index_to_bus_id(type, index),
274                      drive_index_to_unit_id(type, index));
275 }
276 
277 int drive_get_max_bus(BlockInterfaceType type)
278 {
279     int max_bus;
280     BlockBackend *blk;
281     DriveInfo *dinfo;
282 
283     max_bus = -1;
284     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
285         dinfo = blk_legacy_dinfo(blk);
286         if (dinfo && dinfo->type == type && dinfo->bus > max_bus) {
287             max_bus = dinfo->bus;
288         }
289     }
290     return max_bus;
291 }
292 
293 /* Get a block device.  This should only be used for single-drive devices
294    (e.g. SD/Floppy/MTD).  Multi-disk devices (scsi/ide) should use the
295    appropriate bus.  */
296 DriveInfo *drive_get_next(BlockInterfaceType type)
297 {
298     static int next_block_unit[IF_COUNT];
299 
300     return drive_get(type, 0, next_block_unit[type]++);
301 }
302 
303 static void bdrv_format_print(void *opaque, const char *name)
304 {
305     qemu_printf(" %s", name);
306 }
307 
308 typedef struct {
309     QEMUBH *bh;
310     BlockDriverState *bs;
311 } BDRVPutRefBH;
312 
313 static int parse_block_error_action(const char *buf, bool is_read, Error **errp)
314 {
315     if (!strcmp(buf, "ignore")) {
316         return BLOCKDEV_ON_ERROR_IGNORE;
317     } else if (!is_read && !strcmp(buf, "enospc")) {
318         return BLOCKDEV_ON_ERROR_ENOSPC;
319     } else if (!strcmp(buf, "stop")) {
320         return BLOCKDEV_ON_ERROR_STOP;
321     } else if (!strcmp(buf, "report")) {
322         return BLOCKDEV_ON_ERROR_REPORT;
323     } else {
324         error_setg(errp, "'%s' invalid %s error action",
325                    buf, is_read ? "read" : "write");
326         return -1;
327     }
328 }
329 
330 static bool parse_stats_intervals(BlockAcctStats *stats, QList *intervals,
331                                   Error **errp)
332 {
333     const QListEntry *entry;
334     for (entry = qlist_first(intervals); entry; entry = qlist_next(entry)) {
335         switch (qobject_type(entry->value)) {
336 
337         case QTYPE_QSTRING: {
338             unsigned long long length;
339             const char *str = qstring_get_str(qobject_to(QString,
340                                                          entry->value));
341             if (parse_uint_full(str, &length, 10) == 0 &&
342                 length > 0 && length <= UINT_MAX) {
343                 block_acct_add_interval(stats, (unsigned) length);
344             } else {
345                 error_setg(errp, "Invalid interval length: %s", str);
346                 return false;
347             }
348             break;
349         }
350 
351         case QTYPE_QNUM: {
352             int64_t length = qnum_get_int(qobject_to(QNum, entry->value));
353 
354             if (length > 0 && length <= UINT_MAX) {
355                 block_acct_add_interval(stats, (unsigned) length);
356             } else {
357                 error_setg(errp, "Invalid interval length: %" PRId64, length);
358                 return false;
359             }
360             break;
361         }
362 
363         default:
364             error_setg(errp, "The specification of stats-intervals is invalid");
365             return false;
366         }
367     }
368     return true;
369 }
370 
371 typedef enum { MEDIA_DISK, MEDIA_CDROM } DriveMediaType;
372 
373 /* All parameters but @opts are optional and may be set to NULL. */
374 static void extract_common_blockdev_options(QemuOpts *opts, int *bdrv_flags,
375     const char **throttling_group, ThrottleConfig *throttle_cfg,
376     BlockdevDetectZeroesOptions *detect_zeroes, Error **errp)
377 {
378     Error *local_error = NULL;
379     const char *aio;
380 
381     if (bdrv_flags) {
382         if (qemu_opt_get_bool(opts, "copy-on-read", false)) {
383             *bdrv_flags |= BDRV_O_COPY_ON_READ;
384         }
385 
386         if ((aio = qemu_opt_get(opts, "aio")) != NULL) {
387             if (!strcmp(aio, "native")) {
388                 *bdrv_flags |= BDRV_O_NATIVE_AIO;
389             } else if (!strcmp(aio, "threads")) {
390                 /* this is the default */
391             } else {
392                error_setg(errp, "invalid aio option");
393                return;
394             }
395         }
396     }
397 
398     /* disk I/O throttling */
399     if (throttling_group) {
400         *throttling_group = qemu_opt_get(opts, "throttling.group");
401     }
402 
403     if (throttle_cfg) {
404         throttle_config_init(throttle_cfg);
405         throttle_cfg->buckets[THROTTLE_BPS_TOTAL].avg =
406             qemu_opt_get_number(opts, "throttling.bps-total", 0);
407         throttle_cfg->buckets[THROTTLE_BPS_READ].avg  =
408             qemu_opt_get_number(opts, "throttling.bps-read", 0);
409         throttle_cfg->buckets[THROTTLE_BPS_WRITE].avg =
410             qemu_opt_get_number(opts, "throttling.bps-write", 0);
411         throttle_cfg->buckets[THROTTLE_OPS_TOTAL].avg =
412             qemu_opt_get_number(opts, "throttling.iops-total", 0);
413         throttle_cfg->buckets[THROTTLE_OPS_READ].avg =
414             qemu_opt_get_number(opts, "throttling.iops-read", 0);
415         throttle_cfg->buckets[THROTTLE_OPS_WRITE].avg =
416             qemu_opt_get_number(opts, "throttling.iops-write", 0);
417 
418         throttle_cfg->buckets[THROTTLE_BPS_TOTAL].max =
419             qemu_opt_get_number(opts, "throttling.bps-total-max", 0);
420         throttle_cfg->buckets[THROTTLE_BPS_READ].max  =
421             qemu_opt_get_number(opts, "throttling.bps-read-max", 0);
422         throttle_cfg->buckets[THROTTLE_BPS_WRITE].max =
423             qemu_opt_get_number(opts, "throttling.bps-write-max", 0);
424         throttle_cfg->buckets[THROTTLE_OPS_TOTAL].max =
425             qemu_opt_get_number(opts, "throttling.iops-total-max", 0);
426         throttle_cfg->buckets[THROTTLE_OPS_READ].max =
427             qemu_opt_get_number(opts, "throttling.iops-read-max", 0);
428         throttle_cfg->buckets[THROTTLE_OPS_WRITE].max =
429             qemu_opt_get_number(opts, "throttling.iops-write-max", 0);
430 
431         throttle_cfg->buckets[THROTTLE_BPS_TOTAL].burst_length =
432             qemu_opt_get_number(opts, "throttling.bps-total-max-length", 1);
433         throttle_cfg->buckets[THROTTLE_BPS_READ].burst_length  =
434             qemu_opt_get_number(opts, "throttling.bps-read-max-length", 1);
435         throttle_cfg->buckets[THROTTLE_BPS_WRITE].burst_length =
436             qemu_opt_get_number(opts, "throttling.bps-write-max-length", 1);
437         throttle_cfg->buckets[THROTTLE_OPS_TOTAL].burst_length =
438             qemu_opt_get_number(opts, "throttling.iops-total-max-length", 1);
439         throttle_cfg->buckets[THROTTLE_OPS_READ].burst_length =
440             qemu_opt_get_number(opts, "throttling.iops-read-max-length", 1);
441         throttle_cfg->buckets[THROTTLE_OPS_WRITE].burst_length =
442             qemu_opt_get_number(opts, "throttling.iops-write-max-length", 1);
443 
444         throttle_cfg->op_size =
445             qemu_opt_get_number(opts, "throttling.iops-size", 0);
446 
447         if (!throttle_is_valid(throttle_cfg, errp)) {
448             return;
449         }
450     }
451 
452     if (detect_zeroes) {
453         *detect_zeroes =
454             qapi_enum_parse(&BlockdevDetectZeroesOptions_lookup,
455                             qemu_opt_get(opts, "detect-zeroes"),
456                             BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF,
457                             &local_error);
458         if (local_error) {
459             error_propagate(errp, local_error);
460             return;
461         }
462     }
463 }
464 
465 /* Takes the ownership of bs_opts */
466 static BlockBackend *blockdev_init(const char *file, QDict *bs_opts,
467                                    Error **errp)
468 {
469     const char *buf;
470     int bdrv_flags = 0;
471     int on_read_error, on_write_error;
472     bool account_invalid, account_failed;
473     bool writethrough, read_only;
474     BlockBackend *blk;
475     BlockDriverState *bs;
476     ThrottleConfig cfg;
477     int snapshot = 0;
478     Error *error = NULL;
479     QemuOpts *opts;
480     QDict *interval_dict = NULL;
481     QList *interval_list = NULL;
482     const char *id;
483     BlockdevDetectZeroesOptions detect_zeroes =
484         BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF;
485     const char *throttling_group = NULL;
486 
487     /* Check common options by copying from bs_opts to opts, all other options
488      * stay in bs_opts for processing by bdrv_open(). */
489     id = qdict_get_try_str(bs_opts, "id");
490     opts = qemu_opts_create(&qemu_common_drive_opts, id, 1, &error);
491     if (error) {
492         error_propagate(errp, error);
493         goto err_no_opts;
494     }
495 
496     qemu_opts_absorb_qdict(opts, bs_opts, &error);
497     if (error) {
498         error_propagate(errp, error);
499         goto early_err;
500     }
501 
502     if (id) {
503         qdict_del(bs_opts, "id");
504     }
505 
506     /* extract parameters */
507     snapshot = qemu_opt_get_bool(opts, "snapshot", 0);
508 
509     account_invalid = qemu_opt_get_bool(opts, "stats-account-invalid", true);
510     account_failed = qemu_opt_get_bool(opts, "stats-account-failed", true);
511 
512     writethrough = !qemu_opt_get_bool(opts, BDRV_OPT_CACHE_WB, true);
513 
514     id = qemu_opts_id(opts);
515 
516     qdict_extract_subqdict(bs_opts, &interval_dict, "stats-intervals.");
517     qdict_array_split(interval_dict, &interval_list);
518 
519     if (qdict_size(interval_dict) != 0) {
520         error_setg(errp, "Invalid option stats-intervals.%s",
521                    qdict_first(interval_dict)->key);
522         goto early_err;
523     }
524 
525     extract_common_blockdev_options(opts, &bdrv_flags, &throttling_group, &cfg,
526                                     &detect_zeroes, &error);
527     if (error) {
528         error_propagate(errp, error);
529         goto early_err;
530     }
531 
532     if ((buf = qemu_opt_get(opts, "format")) != NULL) {
533         if (is_help_option(buf)) {
534             qemu_printf("Supported formats:");
535             bdrv_iterate_format(bdrv_format_print, NULL, false);
536             qemu_printf("\nSupported formats (read-only):");
537             bdrv_iterate_format(bdrv_format_print, NULL, true);
538             qemu_printf("\n");
539             goto early_err;
540         }
541 
542         if (qdict_haskey(bs_opts, "driver")) {
543             error_setg(errp, "Cannot specify both 'driver' and 'format'");
544             goto early_err;
545         }
546         qdict_put_str(bs_opts, "driver", buf);
547     }
548 
549     on_write_error = BLOCKDEV_ON_ERROR_ENOSPC;
550     if ((buf = qemu_opt_get(opts, "werror")) != NULL) {
551         on_write_error = parse_block_error_action(buf, 0, &error);
552         if (error) {
553             error_propagate(errp, error);
554             goto early_err;
555         }
556     }
557 
558     on_read_error = BLOCKDEV_ON_ERROR_REPORT;
559     if ((buf = qemu_opt_get(opts, "rerror")) != NULL) {
560         on_read_error = parse_block_error_action(buf, 1, &error);
561         if (error) {
562             error_propagate(errp, error);
563             goto early_err;
564         }
565     }
566 
567     if (snapshot) {
568         bdrv_flags |= BDRV_O_SNAPSHOT;
569     }
570 
571     read_only = qemu_opt_get_bool(opts, BDRV_OPT_READ_ONLY, false);
572 
573     /* init */
574     if ((!file || !*file) && !qdict_size(bs_opts)) {
575         BlockBackendRootState *blk_rs;
576 
577         blk = blk_new(0, BLK_PERM_ALL);
578         blk_rs = blk_get_root_state(blk);
579         blk_rs->open_flags    = bdrv_flags;
580         blk_rs->read_only     = read_only;
581         blk_rs->detect_zeroes = detect_zeroes;
582 
583         qobject_unref(bs_opts);
584     } else {
585         if (file && !*file) {
586             file = NULL;
587         }
588 
589         /* bdrv_open() defaults to the values in bdrv_flags (for compatibility
590          * with other callers) rather than what we want as the real defaults.
591          * Apply the defaults here instead. */
592         qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_DIRECT, "off");
593         qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_NO_FLUSH, "off");
594         qdict_set_default_str(bs_opts, BDRV_OPT_READ_ONLY,
595                               read_only ? "on" : "off");
596         qdict_set_default_str(bs_opts, BDRV_OPT_AUTO_READ_ONLY, "on");
597         assert((bdrv_flags & BDRV_O_CACHE_MASK) == 0);
598 
599         if (runstate_check(RUN_STATE_INMIGRATE)) {
600             bdrv_flags |= BDRV_O_INACTIVE;
601         }
602 
603         blk = blk_new_open(file, NULL, bs_opts, bdrv_flags, errp);
604         if (!blk) {
605             goto err_no_bs_opts;
606         }
607         bs = blk_bs(blk);
608 
609         bs->detect_zeroes = detect_zeroes;
610 
611         block_acct_setup(blk_get_stats(blk), account_invalid, account_failed);
612 
613         if (!parse_stats_intervals(blk_get_stats(blk), interval_list, errp)) {
614             blk_unref(blk);
615             blk = NULL;
616             goto err_no_bs_opts;
617         }
618     }
619 
620     /* disk I/O throttling */
621     if (throttle_enabled(&cfg)) {
622         if (!throttling_group) {
623             throttling_group = id;
624         }
625         blk_io_limits_enable(blk, throttling_group);
626         blk_set_io_limits(blk, &cfg);
627     }
628 
629     blk_set_enable_write_cache(blk, !writethrough);
630     blk_set_on_error(blk, on_read_error, on_write_error);
631 
632     if (!monitor_add_blk(blk, id, errp)) {
633         blk_unref(blk);
634         blk = NULL;
635         goto err_no_bs_opts;
636     }
637 
638 err_no_bs_opts:
639     qemu_opts_del(opts);
640     qobject_unref(interval_dict);
641     qobject_unref(interval_list);
642     return blk;
643 
644 early_err:
645     qemu_opts_del(opts);
646     qobject_unref(interval_dict);
647     qobject_unref(interval_list);
648 err_no_opts:
649     qobject_unref(bs_opts);
650     return NULL;
651 }
652 
653 /* Takes the ownership of bs_opts */
654 static BlockDriverState *bds_tree_init(QDict *bs_opts, Error **errp)
655 {
656     int bdrv_flags = 0;
657 
658     /* bdrv_open() defaults to the values in bdrv_flags (for compatibility
659      * with other callers) rather than what we want as the real defaults.
660      * Apply the defaults here instead. */
661     qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_DIRECT, "off");
662     qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_NO_FLUSH, "off");
663     qdict_set_default_str(bs_opts, BDRV_OPT_READ_ONLY, "off");
664 
665     if (runstate_check(RUN_STATE_INMIGRATE)) {
666         bdrv_flags |= BDRV_O_INACTIVE;
667     }
668 
669     return bdrv_open(NULL, NULL, bs_opts, bdrv_flags, errp);
670 }
671 
672 void blockdev_close_all_bdrv_states(void)
673 {
674     BlockDriverState *bs, *next_bs;
675 
676     QTAILQ_FOREACH_SAFE(bs, &monitor_bdrv_states, monitor_list, next_bs) {
677         AioContext *ctx = bdrv_get_aio_context(bs);
678 
679         aio_context_acquire(ctx);
680         bdrv_unref(bs);
681         aio_context_release(ctx);
682     }
683 }
684 
685 /* Iterates over the list of monitor-owned BlockDriverStates */
686 BlockDriverState *bdrv_next_monitor_owned(BlockDriverState *bs)
687 {
688     return bs ? QTAILQ_NEXT(bs, monitor_list)
689               : QTAILQ_FIRST(&monitor_bdrv_states);
690 }
691 
692 static void qemu_opt_rename(QemuOpts *opts, const char *from, const char *to,
693                             Error **errp)
694 {
695     const char *value;
696 
697     value = qemu_opt_get(opts, from);
698     if (value) {
699         if (qemu_opt_find(opts, to)) {
700             error_setg(errp, "'%s' and its alias '%s' can't be used at the "
701                        "same time", to, from);
702             return;
703         }
704     }
705 
706     /* rename all items in opts */
707     while ((value = qemu_opt_get(opts, from))) {
708         qemu_opt_set(opts, to, value, &error_abort);
709         qemu_opt_unset(opts, from);
710     }
711 }
712 
713 QemuOptsList qemu_legacy_drive_opts = {
714     .name = "drive",
715     .head = QTAILQ_HEAD_INITIALIZER(qemu_legacy_drive_opts.head),
716     .desc = {
717         {
718             .name = "bus",
719             .type = QEMU_OPT_NUMBER,
720             .help = "bus number",
721         },{
722             .name = "unit",
723             .type = QEMU_OPT_NUMBER,
724             .help = "unit number (i.e. lun for scsi)",
725         },{
726             .name = "index",
727             .type = QEMU_OPT_NUMBER,
728             .help = "index number",
729         },{
730             .name = "media",
731             .type = QEMU_OPT_STRING,
732             .help = "media type (disk, cdrom)",
733         },{
734             .name = "if",
735             .type = QEMU_OPT_STRING,
736             .help = "interface (ide, scsi, sd, mtd, floppy, pflash, virtio)",
737         },{
738             .name = "file",
739             .type = QEMU_OPT_STRING,
740             .help = "file name",
741         },
742 
743         /* Options that are passed on, but have special semantics with -drive */
744         {
745             .name = BDRV_OPT_READ_ONLY,
746             .type = QEMU_OPT_BOOL,
747             .help = "open drive file as read-only",
748         },{
749             .name = "rerror",
750             .type = QEMU_OPT_STRING,
751             .help = "read error action",
752         },{
753             .name = "werror",
754             .type = QEMU_OPT_STRING,
755             .help = "write error action",
756         },{
757             .name = "copy-on-read",
758             .type = QEMU_OPT_BOOL,
759             .help = "copy read data from backing file into image file",
760         },
761 
762         { /* end of list */ }
763     },
764 };
765 
766 DriveInfo *drive_new(QemuOpts *all_opts, BlockInterfaceType block_default_type,
767                      Error **errp)
768 {
769     const char *value;
770     BlockBackend *blk;
771     DriveInfo *dinfo = NULL;
772     QDict *bs_opts;
773     QemuOpts *legacy_opts;
774     DriveMediaType media = MEDIA_DISK;
775     BlockInterfaceType type;
776     int max_devs, bus_id, unit_id, index;
777     const char *werror, *rerror;
778     bool read_only = false;
779     bool copy_on_read;
780     const char *filename;
781     Error *local_err = NULL;
782     int i;
783 
784     /* Change legacy command line options into QMP ones */
785     static const struct {
786         const char *from;
787         const char *to;
788     } opt_renames[] = {
789         { "iops",           "throttling.iops-total" },
790         { "iops_rd",        "throttling.iops-read" },
791         { "iops_wr",        "throttling.iops-write" },
792 
793         { "bps",            "throttling.bps-total" },
794         { "bps_rd",         "throttling.bps-read" },
795         { "bps_wr",         "throttling.bps-write" },
796 
797         { "iops_max",       "throttling.iops-total-max" },
798         { "iops_rd_max",    "throttling.iops-read-max" },
799         { "iops_wr_max",    "throttling.iops-write-max" },
800 
801         { "bps_max",        "throttling.bps-total-max" },
802         { "bps_rd_max",     "throttling.bps-read-max" },
803         { "bps_wr_max",     "throttling.bps-write-max" },
804 
805         { "iops_size",      "throttling.iops-size" },
806 
807         { "group",          "throttling.group" },
808 
809         { "readonly",       BDRV_OPT_READ_ONLY },
810     };
811 
812     for (i = 0; i < ARRAY_SIZE(opt_renames); i++) {
813         qemu_opt_rename(all_opts, opt_renames[i].from, opt_renames[i].to,
814                         &local_err);
815         if (local_err) {
816             error_propagate(errp, local_err);
817             return NULL;
818         }
819     }
820 
821     value = qemu_opt_get(all_opts, "cache");
822     if (value) {
823         int flags = 0;
824         bool writethrough;
825 
826         if (bdrv_parse_cache_mode(value, &flags, &writethrough) != 0) {
827             error_setg(errp, "invalid cache option");
828             return NULL;
829         }
830 
831         /* Specific options take precedence */
832         if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_WB)) {
833             qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_WB,
834                               !writethrough, &error_abort);
835         }
836         if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_DIRECT)) {
837             qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_DIRECT,
838                               !!(flags & BDRV_O_NOCACHE), &error_abort);
839         }
840         if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_NO_FLUSH)) {
841             qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_NO_FLUSH,
842                               !!(flags & BDRV_O_NO_FLUSH), &error_abort);
843         }
844         qemu_opt_unset(all_opts, "cache");
845     }
846 
847     /* Get a QDict for processing the options */
848     bs_opts = qdict_new();
849     qemu_opts_to_qdict(all_opts, bs_opts);
850 
851     legacy_opts = qemu_opts_create(&qemu_legacy_drive_opts, NULL, 0,
852                                    &error_abort);
853     qemu_opts_absorb_qdict(legacy_opts, bs_opts, &local_err);
854     if (local_err) {
855         error_propagate(errp, local_err);
856         goto fail;
857     }
858 
859     /* Media type */
860     value = qemu_opt_get(legacy_opts, "media");
861     if (value) {
862         if (!strcmp(value, "disk")) {
863             media = MEDIA_DISK;
864         } else if (!strcmp(value, "cdrom")) {
865             media = MEDIA_CDROM;
866             read_only = true;
867         } else {
868             error_setg(errp, "'%s' invalid media", value);
869             goto fail;
870         }
871     }
872 
873     /* copy-on-read is disabled with a warning for read-only devices */
874     read_only |= qemu_opt_get_bool(legacy_opts, BDRV_OPT_READ_ONLY, false);
875     copy_on_read = qemu_opt_get_bool(legacy_opts, "copy-on-read", false);
876 
877     if (read_only && copy_on_read) {
878         warn_report("disabling copy-on-read on read-only drive");
879         copy_on_read = false;
880     }
881 
882     qdict_put_str(bs_opts, BDRV_OPT_READ_ONLY, read_only ? "on" : "off");
883     qdict_put_str(bs_opts, "copy-on-read", copy_on_read ? "on" : "off");
884 
885     /* Controller type */
886     value = qemu_opt_get(legacy_opts, "if");
887     if (value) {
888         for (type = 0;
889              type < IF_COUNT && strcmp(value, if_name[type]);
890              type++) {
891         }
892         if (type == IF_COUNT) {
893             error_setg(errp, "unsupported bus type '%s'", value);
894             goto fail;
895         }
896     } else {
897         type = block_default_type;
898     }
899 
900     /* Device address specified by bus/unit or index.
901      * If none was specified, try to find the first free one. */
902     bus_id  = qemu_opt_get_number(legacy_opts, "bus", 0);
903     unit_id = qemu_opt_get_number(legacy_opts, "unit", -1);
904     index   = qemu_opt_get_number(legacy_opts, "index", -1);
905 
906     max_devs = if_max_devs[type];
907 
908     if (index != -1) {
909         if (bus_id != 0 || unit_id != -1) {
910             error_setg(errp, "index cannot be used with bus and unit");
911             goto fail;
912         }
913         bus_id = drive_index_to_bus_id(type, index);
914         unit_id = drive_index_to_unit_id(type, index);
915     }
916 
917     if (unit_id == -1) {
918        unit_id = 0;
919        while (drive_get(type, bus_id, unit_id) != NULL) {
920            unit_id++;
921            if (max_devs && unit_id >= max_devs) {
922                unit_id -= max_devs;
923                bus_id++;
924            }
925        }
926     }
927 
928     if (max_devs && unit_id >= max_devs) {
929         error_setg(errp, "unit %d too big (max is %d)", unit_id, max_devs - 1);
930         goto fail;
931     }
932 
933     if (drive_get(type, bus_id, unit_id) != NULL) {
934         error_setg(errp, "drive with bus=%d, unit=%d (index=%d) exists",
935                    bus_id, unit_id, index);
936         goto fail;
937     }
938 
939     /* no id supplied -> create one */
940     if (qemu_opts_id(all_opts) == NULL) {
941         char *new_id;
942         const char *mediastr = "";
943         if (type == IF_IDE || type == IF_SCSI) {
944             mediastr = (media == MEDIA_CDROM) ? "-cd" : "-hd";
945         }
946         if (max_devs) {
947             new_id = g_strdup_printf("%s%i%s%i", if_name[type], bus_id,
948                                      mediastr, unit_id);
949         } else {
950             new_id = g_strdup_printf("%s%s%i", if_name[type],
951                                      mediastr, unit_id);
952         }
953         qdict_put_str(bs_opts, "id", new_id);
954         g_free(new_id);
955     }
956 
957     /* Add virtio block device */
958     if (type == IF_VIRTIO) {
959         QemuOpts *devopts;
960         devopts = qemu_opts_create(qemu_find_opts("device"), NULL, 0,
961                                    &error_abort);
962         if (arch_type == QEMU_ARCH_S390X) {
963             qemu_opt_set(devopts, "driver", "virtio-blk-ccw", &error_abort);
964         } else {
965             qemu_opt_set(devopts, "driver", "virtio-blk-pci", &error_abort);
966         }
967         qemu_opt_set(devopts, "drive", qdict_get_str(bs_opts, "id"),
968                      &error_abort);
969     }
970 
971     filename = qemu_opt_get(legacy_opts, "file");
972 
973     /* Check werror/rerror compatibility with if=... */
974     werror = qemu_opt_get(legacy_opts, "werror");
975     if (werror != NULL) {
976         if (type != IF_IDE && type != IF_SCSI && type != IF_VIRTIO &&
977             type != IF_NONE) {
978             error_setg(errp, "werror is not supported by this bus type");
979             goto fail;
980         }
981         qdict_put_str(bs_opts, "werror", werror);
982     }
983 
984     rerror = qemu_opt_get(legacy_opts, "rerror");
985     if (rerror != NULL) {
986         if (type != IF_IDE && type != IF_VIRTIO && type != IF_SCSI &&
987             type != IF_NONE) {
988             error_setg(errp, "rerror is not supported by this bus type");
989             goto fail;
990         }
991         qdict_put_str(bs_opts, "rerror", rerror);
992     }
993 
994     /* Actual block device init: Functionality shared with blockdev-add */
995     blk = blockdev_init(filename, bs_opts, &local_err);
996     bs_opts = NULL;
997     if (!blk) {
998         error_propagate(errp, local_err);
999         goto fail;
1000     } else {
1001         assert(!local_err);
1002     }
1003 
1004     /* Create legacy DriveInfo */
1005     dinfo = g_malloc0(sizeof(*dinfo));
1006     dinfo->opts = all_opts;
1007 
1008     dinfo->type = type;
1009     dinfo->bus = bus_id;
1010     dinfo->unit = unit_id;
1011 
1012     blk_set_legacy_dinfo(blk, dinfo);
1013 
1014     switch(type) {
1015     case IF_IDE:
1016     case IF_SCSI:
1017     case IF_XEN:
1018     case IF_NONE:
1019         dinfo->media_cd = media == MEDIA_CDROM;
1020         break;
1021     default:
1022         break;
1023     }
1024 
1025 fail:
1026     qemu_opts_del(legacy_opts);
1027     qobject_unref(bs_opts);
1028     return dinfo;
1029 }
1030 
1031 static BlockDriverState *qmp_get_root_bs(const char *name, Error **errp)
1032 {
1033     BlockDriverState *bs;
1034 
1035     bs = bdrv_lookup_bs(name, name, errp);
1036     if (bs == NULL) {
1037         return NULL;
1038     }
1039 
1040     if (!bdrv_is_root_node(bs)) {
1041         error_setg(errp, "Need a root block node");
1042         return NULL;
1043     }
1044 
1045     if (!bdrv_is_inserted(bs)) {
1046         error_setg(errp, "Device has no medium");
1047         return NULL;
1048     }
1049 
1050     return bs;
1051 }
1052 
1053 static BlockBackend *qmp_get_blk(const char *blk_name, const char *qdev_id,
1054                                  Error **errp)
1055 {
1056     BlockBackend *blk;
1057 
1058     if (!blk_name == !qdev_id) {
1059         error_setg(errp, "Need exactly one of 'device' and 'id'");
1060         return NULL;
1061     }
1062 
1063     if (qdev_id) {
1064         blk = blk_by_qdev_id(qdev_id, errp);
1065     } else {
1066         blk = blk_by_name(blk_name);
1067         if (blk == NULL) {
1068             error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
1069                       "Device '%s' not found", blk_name);
1070         }
1071     }
1072 
1073     return blk;
1074 }
1075 
1076 void hmp_commit(Monitor *mon, const QDict *qdict)
1077 {
1078     const char *device = qdict_get_str(qdict, "device");
1079     BlockBackend *blk;
1080     int ret;
1081 
1082     if (!strcmp(device, "all")) {
1083         ret = blk_commit_all();
1084     } else {
1085         BlockDriverState *bs;
1086         AioContext *aio_context;
1087 
1088         blk = blk_by_name(device);
1089         if (!blk) {
1090             monitor_printf(mon, "Device '%s' not found\n", device);
1091             return;
1092         }
1093         if (!blk_is_available(blk)) {
1094             monitor_printf(mon, "Device '%s' has no medium\n", device);
1095             return;
1096         }
1097 
1098         bs = blk_bs(blk);
1099         aio_context = bdrv_get_aio_context(bs);
1100         aio_context_acquire(aio_context);
1101 
1102         ret = bdrv_commit(bs);
1103 
1104         aio_context_release(aio_context);
1105     }
1106     if (ret < 0) {
1107         monitor_printf(mon, "'commit' error for '%s': %s\n", device,
1108                        strerror(-ret));
1109     }
1110 }
1111 
1112 static void blockdev_do_action(TransactionAction *action, Error **errp)
1113 {
1114     TransactionActionList list;
1115 
1116     list.value = action;
1117     list.next = NULL;
1118     qmp_transaction(&list, false, NULL, errp);
1119 }
1120 
1121 void qmp_blockdev_snapshot_sync(bool has_device, const char *device,
1122                                 bool has_node_name, const char *node_name,
1123                                 const char *snapshot_file,
1124                                 bool has_snapshot_node_name,
1125                                 const char *snapshot_node_name,
1126                                 bool has_format, const char *format,
1127                                 bool has_mode, NewImageMode mode, Error **errp)
1128 {
1129     BlockdevSnapshotSync snapshot = {
1130         .has_device = has_device,
1131         .device = (char *) device,
1132         .has_node_name = has_node_name,
1133         .node_name = (char *) node_name,
1134         .snapshot_file = (char *) snapshot_file,
1135         .has_snapshot_node_name = has_snapshot_node_name,
1136         .snapshot_node_name = (char *) snapshot_node_name,
1137         .has_format = has_format,
1138         .format = (char *) format,
1139         .has_mode = has_mode,
1140         .mode = mode,
1141     };
1142     TransactionAction action = {
1143         .type = TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC,
1144         .u.blockdev_snapshot_sync.data = &snapshot,
1145     };
1146     blockdev_do_action(&action, errp);
1147 }
1148 
1149 void qmp_blockdev_snapshot(const char *node, const char *overlay,
1150                            Error **errp)
1151 {
1152     BlockdevSnapshot snapshot_data = {
1153         .node = (char *) node,
1154         .overlay = (char *) overlay
1155     };
1156     TransactionAction action = {
1157         .type = TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT,
1158         .u.blockdev_snapshot.data = &snapshot_data,
1159     };
1160     blockdev_do_action(&action, errp);
1161 }
1162 
1163 void qmp_blockdev_snapshot_internal_sync(const char *device,
1164                                          const char *name,
1165                                          Error **errp)
1166 {
1167     BlockdevSnapshotInternal snapshot = {
1168         .device = (char *) device,
1169         .name = (char *) name
1170     };
1171     TransactionAction action = {
1172         .type = TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC,
1173         .u.blockdev_snapshot_internal_sync.data = &snapshot,
1174     };
1175     blockdev_do_action(&action, errp);
1176 }
1177 
1178 SnapshotInfo *qmp_blockdev_snapshot_delete_internal_sync(const char *device,
1179                                                          bool has_id,
1180                                                          const char *id,
1181                                                          bool has_name,
1182                                                          const char *name,
1183                                                          Error **errp)
1184 {
1185     BlockDriverState *bs;
1186     AioContext *aio_context;
1187     QEMUSnapshotInfo sn;
1188     Error *local_err = NULL;
1189     SnapshotInfo *info = NULL;
1190     int ret;
1191 
1192     bs = qmp_get_root_bs(device, errp);
1193     if (!bs) {
1194         return NULL;
1195     }
1196     aio_context = bdrv_get_aio_context(bs);
1197     aio_context_acquire(aio_context);
1198 
1199     if (!has_id) {
1200         id = NULL;
1201     }
1202 
1203     if (!has_name) {
1204         name = NULL;
1205     }
1206 
1207     if (!id && !name) {
1208         error_setg(errp, "Name or id must be provided");
1209         goto out_aio_context;
1210     }
1211 
1212     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_INTERNAL_SNAPSHOT_DELETE, errp)) {
1213         goto out_aio_context;
1214     }
1215 
1216     ret = bdrv_snapshot_find_by_id_and_name(bs, id, name, &sn, &local_err);
1217     if (local_err) {
1218         error_propagate(errp, local_err);
1219         goto out_aio_context;
1220     }
1221     if (!ret) {
1222         error_setg(errp,
1223                    "Snapshot with id '%s' and name '%s' does not exist on "
1224                    "device '%s'",
1225                    STR_OR_NULL(id), STR_OR_NULL(name), device);
1226         goto out_aio_context;
1227     }
1228 
1229     bdrv_snapshot_delete(bs, id, name, &local_err);
1230     if (local_err) {
1231         error_propagate(errp, local_err);
1232         goto out_aio_context;
1233     }
1234 
1235     aio_context_release(aio_context);
1236 
1237     info = g_new0(SnapshotInfo, 1);
1238     info->id = g_strdup(sn.id_str);
1239     info->name = g_strdup(sn.name);
1240     info->date_nsec = sn.date_nsec;
1241     info->date_sec = sn.date_sec;
1242     info->vm_state_size = sn.vm_state_size;
1243     info->vm_clock_nsec = sn.vm_clock_nsec % 1000000000;
1244     info->vm_clock_sec = sn.vm_clock_nsec / 1000000000;
1245 
1246     return info;
1247 
1248 out_aio_context:
1249     aio_context_release(aio_context);
1250     return NULL;
1251 }
1252 
1253 /**
1254  * block_dirty_bitmap_lookup:
1255  * Return a dirty bitmap (if present), after validating
1256  * the node reference and bitmap names.
1257  *
1258  * @node: The name of the BDS node to search for bitmaps
1259  * @name: The name of the bitmap to search for
1260  * @pbs: Output pointer for BDS lookup, if desired. Can be NULL.
1261  * @errp: Output pointer for error information. Can be NULL.
1262  *
1263  * @return: A bitmap object on success, or NULL on failure.
1264  */
1265 static BdrvDirtyBitmap *block_dirty_bitmap_lookup(const char *node,
1266                                                   const char *name,
1267                                                   BlockDriverState **pbs,
1268                                                   Error **errp)
1269 {
1270     BlockDriverState *bs;
1271     BdrvDirtyBitmap *bitmap;
1272 
1273     if (!node) {
1274         error_setg(errp, "Node cannot be NULL");
1275         return NULL;
1276     }
1277     if (!name) {
1278         error_setg(errp, "Bitmap name cannot be NULL");
1279         return NULL;
1280     }
1281     bs = bdrv_lookup_bs(node, node, NULL);
1282     if (!bs) {
1283         error_setg(errp, "Node '%s' not found", node);
1284         return NULL;
1285     }
1286 
1287     bitmap = bdrv_find_dirty_bitmap(bs, name);
1288     if (!bitmap) {
1289         error_setg(errp, "Dirty bitmap '%s' not found", name);
1290         return NULL;
1291     }
1292 
1293     if (pbs) {
1294         *pbs = bs;
1295     }
1296 
1297     return bitmap;
1298 }
1299 
1300 /* New and old BlockDriverState structs for atomic group operations */
1301 
1302 typedef struct BlkActionState BlkActionState;
1303 
1304 /**
1305  * BlkActionOps:
1306  * Table of operations that define an Action.
1307  *
1308  * @instance_size: Size of state struct, in bytes.
1309  * @prepare: Prepare the work, must NOT be NULL.
1310  * @commit: Commit the changes, can be NULL.
1311  * @abort: Abort the changes on fail, can be NULL.
1312  * @clean: Clean up resources after all transaction actions have called
1313  *         commit() or abort(). Can be NULL.
1314  *
1315  * Only prepare() may fail. In a single transaction, only one of commit() or
1316  * abort() will be called. clean() will always be called if it is present.
1317  */
1318 typedef struct BlkActionOps {
1319     size_t instance_size;
1320     void (*prepare)(BlkActionState *common, Error **errp);
1321     void (*commit)(BlkActionState *common);
1322     void (*abort)(BlkActionState *common);
1323     void (*clean)(BlkActionState *common);
1324 } BlkActionOps;
1325 
1326 /**
1327  * BlkActionState:
1328  * Describes one Action's state within a Transaction.
1329  *
1330  * @action: QAPI-defined enum identifying which Action to perform.
1331  * @ops: Table of ActionOps this Action can perform.
1332  * @block_job_txn: Transaction which this action belongs to.
1333  * @entry: List membership for all Actions in this Transaction.
1334  *
1335  * This structure must be arranged as first member in a subclassed type,
1336  * assuming that the compiler will also arrange it to the same offsets as the
1337  * base class.
1338  */
1339 struct BlkActionState {
1340     TransactionAction *action;
1341     const BlkActionOps *ops;
1342     JobTxn *block_job_txn;
1343     TransactionProperties *txn_props;
1344     QTAILQ_ENTRY(BlkActionState) entry;
1345 };
1346 
1347 /* internal snapshot private data */
1348 typedef struct InternalSnapshotState {
1349     BlkActionState common;
1350     BlockDriverState *bs;
1351     QEMUSnapshotInfo sn;
1352     bool created;
1353 } InternalSnapshotState;
1354 
1355 
1356 static int action_check_completion_mode(BlkActionState *s, Error **errp)
1357 {
1358     if (s->txn_props->completion_mode != ACTION_COMPLETION_MODE_INDIVIDUAL) {
1359         error_setg(errp,
1360                    "Action '%s' does not support Transaction property "
1361                    "completion-mode = %s",
1362                    TransactionActionKind_str(s->action->type),
1363                    ActionCompletionMode_str(s->txn_props->completion_mode));
1364         return -1;
1365     }
1366     return 0;
1367 }
1368 
1369 static void internal_snapshot_prepare(BlkActionState *common,
1370                                       Error **errp)
1371 {
1372     Error *local_err = NULL;
1373     const char *device;
1374     const char *name;
1375     BlockDriverState *bs;
1376     QEMUSnapshotInfo old_sn, *sn;
1377     bool ret;
1378     qemu_timeval tv;
1379     BlockdevSnapshotInternal *internal;
1380     InternalSnapshotState *state;
1381     AioContext *aio_context;
1382     int ret1;
1383 
1384     g_assert(common->action->type ==
1385              TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC);
1386     internal = common->action->u.blockdev_snapshot_internal_sync.data;
1387     state = DO_UPCAST(InternalSnapshotState, common, common);
1388 
1389     /* 1. parse input */
1390     device = internal->device;
1391     name = internal->name;
1392 
1393     /* 2. check for validation */
1394     if (action_check_completion_mode(common, errp) < 0) {
1395         return;
1396     }
1397 
1398     bs = qmp_get_root_bs(device, errp);
1399     if (!bs) {
1400         return;
1401     }
1402 
1403     aio_context = bdrv_get_aio_context(bs);
1404     aio_context_acquire(aio_context);
1405 
1406     state->bs = bs;
1407 
1408     /* Paired with .clean() */
1409     bdrv_drained_begin(bs);
1410 
1411     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_INTERNAL_SNAPSHOT, errp)) {
1412         goto out;
1413     }
1414 
1415     if (bdrv_is_read_only(bs)) {
1416         error_setg(errp, "Device '%s' is read only", device);
1417         goto out;
1418     }
1419 
1420     if (!bdrv_can_snapshot(bs)) {
1421         error_setg(errp, "Block format '%s' used by device '%s' "
1422                    "does not support internal snapshots",
1423                    bs->drv->format_name, device);
1424         goto out;
1425     }
1426 
1427     if (!strlen(name)) {
1428         error_setg(errp, "Name is empty");
1429         goto out;
1430     }
1431 
1432     /* check whether a snapshot with name exist */
1433     ret = bdrv_snapshot_find_by_id_and_name(bs, NULL, name, &old_sn,
1434                                             &local_err);
1435     if (local_err) {
1436         error_propagate(errp, local_err);
1437         goto out;
1438     } else if (ret) {
1439         error_setg(errp,
1440                    "Snapshot with name '%s' already exists on device '%s'",
1441                    name, device);
1442         goto out;
1443     }
1444 
1445     /* 3. take the snapshot */
1446     sn = &state->sn;
1447     pstrcpy(sn->name, sizeof(sn->name), name);
1448     qemu_gettimeofday(&tv);
1449     sn->date_sec = tv.tv_sec;
1450     sn->date_nsec = tv.tv_usec * 1000;
1451     sn->vm_clock_nsec = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL);
1452 
1453     ret1 = bdrv_snapshot_create(bs, sn);
1454     if (ret1 < 0) {
1455         error_setg_errno(errp, -ret1,
1456                          "Failed to create snapshot '%s' on device '%s'",
1457                          name, device);
1458         goto out;
1459     }
1460 
1461     /* 4. succeed, mark a snapshot is created */
1462     state->created = true;
1463 
1464 out:
1465     aio_context_release(aio_context);
1466 }
1467 
1468 static void internal_snapshot_abort(BlkActionState *common)
1469 {
1470     InternalSnapshotState *state =
1471                              DO_UPCAST(InternalSnapshotState, common, common);
1472     BlockDriverState *bs = state->bs;
1473     QEMUSnapshotInfo *sn = &state->sn;
1474     AioContext *aio_context;
1475     Error *local_error = NULL;
1476 
1477     if (!state->created) {
1478         return;
1479     }
1480 
1481     aio_context = bdrv_get_aio_context(state->bs);
1482     aio_context_acquire(aio_context);
1483 
1484     if (bdrv_snapshot_delete(bs, sn->id_str, sn->name, &local_error) < 0) {
1485         error_reportf_err(local_error,
1486                           "Failed to delete snapshot with id '%s' and "
1487                           "name '%s' on device '%s' in abort: ",
1488                           sn->id_str, sn->name,
1489                           bdrv_get_device_name(bs));
1490     }
1491 
1492     aio_context_release(aio_context);
1493 }
1494 
1495 static void internal_snapshot_clean(BlkActionState *common)
1496 {
1497     InternalSnapshotState *state = DO_UPCAST(InternalSnapshotState,
1498                                              common, common);
1499     AioContext *aio_context;
1500 
1501     if (!state->bs) {
1502         return;
1503     }
1504 
1505     aio_context = bdrv_get_aio_context(state->bs);
1506     aio_context_acquire(aio_context);
1507 
1508     bdrv_drained_end(state->bs);
1509 
1510     aio_context_release(aio_context);
1511 }
1512 
1513 /* external snapshot private data */
1514 typedef struct ExternalSnapshotState {
1515     BlkActionState common;
1516     BlockDriverState *old_bs;
1517     BlockDriverState *new_bs;
1518     bool overlay_appended;
1519 } ExternalSnapshotState;
1520 
1521 static void external_snapshot_prepare(BlkActionState *common,
1522                                       Error **errp)
1523 {
1524     int flags = 0;
1525     QDict *options = NULL;
1526     Error *local_err = NULL;
1527     /* Device and node name of the image to generate the snapshot from */
1528     const char *device;
1529     const char *node_name;
1530     /* Reference to the new image (for 'blockdev-snapshot') */
1531     const char *snapshot_ref;
1532     /* File name of the new image (for 'blockdev-snapshot-sync') */
1533     const char *new_image_file;
1534     ExternalSnapshotState *state =
1535                              DO_UPCAST(ExternalSnapshotState, common, common);
1536     TransactionAction *action = common->action;
1537     AioContext *aio_context;
1538 
1539     /* 'blockdev-snapshot' and 'blockdev-snapshot-sync' have similar
1540      * purpose but a different set of parameters */
1541     switch (action->type) {
1542     case TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT:
1543         {
1544             BlockdevSnapshot *s = action->u.blockdev_snapshot.data;
1545             device = s->node;
1546             node_name = s->node;
1547             new_image_file = NULL;
1548             snapshot_ref = s->overlay;
1549         }
1550         break;
1551     case TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC:
1552         {
1553             BlockdevSnapshotSync *s = action->u.blockdev_snapshot_sync.data;
1554             device = s->has_device ? s->device : NULL;
1555             node_name = s->has_node_name ? s->node_name : NULL;
1556             new_image_file = s->snapshot_file;
1557             snapshot_ref = NULL;
1558         }
1559         break;
1560     default:
1561         g_assert_not_reached();
1562     }
1563 
1564     /* start processing */
1565     if (action_check_completion_mode(common, errp) < 0) {
1566         return;
1567     }
1568 
1569     state->old_bs = bdrv_lookup_bs(device, node_name, errp);
1570     if (!state->old_bs) {
1571         return;
1572     }
1573 
1574     aio_context = bdrv_get_aio_context(state->old_bs);
1575     aio_context_acquire(aio_context);
1576 
1577     /* Paired with .clean() */
1578     bdrv_drained_begin(state->old_bs);
1579 
1580     if (!bdrv_is_inserted(state->old_bs)) {
1581         error_setg(errp, QERR_DEVICE_HAS_NO_MEDIUM, device);
1582         goto out;
1583     }
1584 
1585     if (bdrv_op_is_blocked(state->old_bs,
1586                            BLOCK_OP_TYPE_EXTERNAL_SNAPSHOT, errp)) {
1587         goto out;
1588     }
1589 
1590     if (!bdrv_is_read_only(state->old_bs)) {
1591         if (bdrv_flush(state->old_bs)) {
1592             error_setg(errp, QERR_IO_ERROR);
1593             goto out;
1594         }
1595     }
1596 
1597     if (!bdrv_is_first_non_filter(state->old_bs)) {
1598         error_setg(errp, QERR_FEATURE_DISABLED, "snapshot");
1599         goto out;
1600     }
1601 
1602     if (action->type == TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC) {
1603         BlockdevSnapshotSync *s = action->u.blockdev_snapshot_sync.data;
1604         const char *format = s->has_format ? s->format : "qcow2";
1605         enum NewImageMode mode;
1606         const char *snapshot_node_name =
1607             s->has_snapshot_node_name ? s->snapshot_node_name : NULL;
1608 
1609         if (node_name && !snapshot_node_name) {
1610             error_setg(errp, "New snapshot node name missing");
1611             goto out;
1612         }
1613 
1614         if (snapshot_node_name &&
1615             bdrv_lookup_bs(snapshot_node_name, snapshot_node_name, NULL)) {
1616             error_setg(errp, "New snapshot node name already in use");
1617             goto out;
1618         }
1619 
1620         flags = state->old_bs->open_flags;
1621         flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_COPY_ON_READ);
1622         flags |= BDRV_O_NO_BACKING;
1623 
1624         /* create new image w/backing file */
1625         mode = s->has_mode ? s->mode : NEW_IMAGE_MODE_ABSOLUTE_PATHS;
1626         if (mode != NEW_IMAGE_MODE_EXISTING) {
1627             int64_t size = bdrv_getlength(state->old_bs);
1628             if (size < 0) {
1629                 error_setg_errno(errp, -size, "bdrv_getlength failed");
1630                 goto out;
1631             }
1632             bdrv_refresh_filename(state->old_bs);
1633             bdrv_img_create(new_image_file, format,
1634                             state->old_bs->filename,
1635                             state->old_bs->drv->format_name,
1636                             NULL, size, flags, false, &local_err);
1637             if (local_err) {
1638                 error_propagate(errp, local_err);
1639                 goto out;
1640             }
1641         }
1642 
1643         options = qdict_new();
1644         if (snapshot_node_name) {
1645             qdict_put_str(options, "node-name", snapshot_node_name);
1646         }
1647         qdict_put_str(options, "driver", format);
1648     }
1649 
1650     state->new_bs = bdrv_open(new_image_file, snapshot_ref, options, flags,
1651                               errp);
1652     /* We will manually add the backing_hd field to the bs later */
1653     if (!state->new_bs) {
1654         goto out;
1655     }
1656 
1657     if (bdrv_has_blk(state->new_bs)) {
1658         error_setg(errp, "The snapshot is already in use");
1659         goto out;
1660     }
1661 
1662     if (bdrv_op_is_blocked(state->new_bs, BLOCK_OP_TYPE_EXTERNAL_SNAPSHOT,
1663                            errp)) {
1664         goto out;
1665     }
1666 
1667     if (state->new_bs->backing != NULL) {
1668         error_setg(errp, "The snapshot already has a backing image");
1669         goto out;
1670     }
1671 
1672     if (!state->new_bs->drv->supports_backing) {
1673         error_setg(errp, "The snapshot does not support backing images");
1674         goto out;
1675     }
1676 
1677     bdrv_set_aio_context(state->new_bs, aio_context);
1678 
1679     /* This removes our old bs and adds the new bs. This is an operation that
1680      * can fail, so we need to do it in .prepare; undoing it for abort is
1681      * always possible. */
1682     bdrv_ref(state->new_bs);
1683     bdrv_append(state->new_bs, state->old_bs, &local_err);
1684     if (local_err) {
1685         error_propagate(errp, local_err);
1686         goto out;
1687     }
1688     state->overlay_appended = true;
1689 
1690 out:
1691     aio_context_release(aio_context);
1692 }
1693 
1694 static void external_snapshot_commit(BlkActionState *common)
1695 {
1696     ExternalSnapshotState *state =
1697                              DO_UPCAST(ExternalSnapshotState, common, common);
1698     AioContext *aio_context;
1699 
1700     aio_context = bdrv_get_aio_context(state->old_bs);
1701     aio_context_acquire(aio_context);
1702 
1703     /* We don't need (or want) to use the transactional
1704      * bdrv_reopen_multiple() across all the entries at once, because we
1705      * don't want to abort all of them if one of them fails the reopen */
1706     if (!atomic_read(&state->old_bs->copy_on_read)) {
1707         bdrv_reopen_set_read_only(state->old_bs, true, NULL);
1708     }
1709 
1710     aio_context_release(aio_context);
1711 }
1712 
1713 static void external_snapshot_abort(BlkActionState *common)
1714 {
1715     ExternalSnapshotState *state =
1716                              DO_UPCAST(ExternalSnapshotState, common, common);
1717     if (state->new_bs) {
1718         if (state->overlay_appended) {
1719             AioContext *aio_context;
1720 
1721             aio_context = bdrv_get_aio_context(state->old_bs);
1722             aio_context_acquire(aio_context);
1723 
1724             bdrv_ref(state->old_bs);   /* we can't let bdrv_set_backind_hd()
1725                                           close state->old_bs; we need it */
1726             bdrv_set_backing_hd(state->new_bs, NULL, &error_abort);
1727             bdrv_replace_node(state->new_bs, state->old_bs, &error_abort);
1728             bdrv_unref(state->old_bs); /* bdrv_replace_node() ref'ed old_bs */
1729 
1730             aio_context_release(aio_context);
1731         }
1732     }
1733 }
1734 
1735 static void external_snapshot_clean(BlkActionState *common)
1736 {
1737     ExternalSnapshotState *state =
1738                              DO_UPCAST(ExternalSnapshotState, common, common);
1739     AioContext *aio_context;
1740 
1741     if (!state->old_bs) {
1742         return;
1743     }
1744 
1745     aio_context = bdrv_get_aio_context(state->old_bs);
1746     aio_context_acquire(aio_context);
1747 
1748     bdrv_drained_end(state->old_bs);
1749     bdrv_unref(state->new_bs);
1750 
1751     aio_context_release(aio_context);
1752 }
1753 
1754 typedef struct DriveBackupState {
1755     BlkActionState common;
1756     BlockDriverState *bs;
1757     BlockJob *job;
1758 } DriveBackupState;
1759 
1760 static BlockJob *do_drive_backup(DriveBackup *backup, JobTxn *txn,
1761                             Error **errp);
1762 
1763 static void drive_backup_prepare(BlkActionState *common, Error **errp)
1764 {
1765     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1766     BlockDriverState *bs;
1767     DriveBackup *backup;
1768     AioContext *aio_context;
1769     Error *local_err = NULL;
1770 
1771     assert(common->action->type == TRANSACTION_ACTION_KIND_DRIVE_BACKUP);
1772     backup = common->action->u.drive_backup.data;
1773 
1774     bs = qmp_get_root_bs(backup->device, errp);
1775     if (!bs) {
1776         return;
1777     }
1778 
1779     aio_context = bdrv_get_aio_context(bs);
1780     aio_context_acquire(aio_context);
1781 
1782     /* Paired with .clean() */
1783     bdrv_drained_begin(bs);
1784 
1785     state->bs = bs;
1786 
1787     state->job = do_drive_backup(backup, common->block_job_txn, &local_err);
1788     if (local_err) {
1789         error_propagate(errp, local_err);
1790         goto out;
1791     }
1792 
1793 out:
1794     aio_context_release(aio_context);
1795 }
1796 
1797 static void drive_backup_commit(BlkActionState *common)
1798 {
1799     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1800     AioContext *aio_context;
1801 
1802     aio_context = bdrv_get_aio_context(state->bs);
1803     aio_context_acquire(aio_context);
1804 
1805     assert(state->job);
1806     job_start(&state->job->job);
1807 
1808     aio_context_release(aio_context);
1809 }
1810 
1811 static void drive_backup_abort(BlkActionState *common)
1812 {
1813     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1814 
1815     if (state->job) {
1816         AioContext *aio_context;
1817 
1818         aio_context = bdrv_get_aio_context(state->bs);
1819         aio_context_acquire(aio_context);
1820 
1821         job_cancel_sync(&state->job->job);
1822 
1823         aio_context_release(aio_context);
1824     }
1825 }
1826 
1827 static void drive_backup_clean(BlkActionState *common)
1828 {
1829     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1830     AioContext *aio_context;
1831 
1832     if (!state->bs) {
1833         return;
1834     }
1835 
1836     aio_context = bdrv_get_aio_context(state->bs);
1837     aio_context_acquire(aio_context);
1838 
1839     bdrv_drained_end(state->bs);
1840 
1841     aio_context_release(aio_context);
1842 }
1843 
1844 typedef struct BlockdevBackupState {
1845     BlkActionState common;
1846     BlockDriverState *bs;
1847     BlockJob *job;
1848 } BlockdevBackupState;
1849 
1850 static BlockJob *do_blockdev_backup(BlockdevBackup *backup, JobTxn *txn,
1851                                     Error **errp);
1852 
1853 static void blockdev_backup_prepare(BlkActionState *common, Error **errp)
1854 {
1855     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1856     BlockdevBackup *backup;
1857     BlockDriverState *bs, *target;
1858     AioContext *aio_context;
1859     Error *local_err = NULL;
1860 
1861     assert(common->action->type == TRANSACTION_ACTION_KIND_BLOCKDEV_BACKUP);
1862     backup = common->action->u.blockdev_backup.data;
1863 
1864     bs = bdrv_lookup_bs(backup->device, backup->device, errp);
1865     if (!bs) {
1866         return;
1867     }
1868 
1869     target = bdrv_lookup_bs(backup->target, backup->target, errp);
1870     if (!target) {
1871         return;
1872     }
1873 
1874     aio_context = bdrv_get_aio_context(bs);
1875     if (aio_context != bdrv_get_aio_context(target)) {
1876         error_setg(errp, "Backup between two IO threads is not implemented");
1877         return;
1878     }
1879     aio_context_acquire(aio_context);
1880     state->bs = bs;
1881 
1882     /* Paired with .clean() */
1883     bdrv_drained_begin(state->bs);
1884 
1885     state->job = do_blockdev_backup(backup, common->block_job_txn, &local_err);
1886     if (local_err) {
1887         error_propagate(errp, local_err);
1888         goto out;
1889     }
1890 
1891 out:
1892     aio_context_release(aio_context);
1893 }
1894 
1895 static void blockdev_backup_commit(BlkActionState *common)
1896 {
1897     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1898     AioContext *aio_context;
1899 
1900     aio_context = bdrv_get_aio_context(state->bs);
1901     aio_context_acquire(aio_context);
1902 
1903     assert(state->job);
1904     job_start(&state->job->job);
1905 
1906     aio_context_release(aio_context);
1907 }
1908 
1909 static void blockdev_backup_abort(BlkActionState *common)
1910 {
1911     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1912 
1913     if (state->job) {
1914         AioContext *aio_context;
1915 
1916         aio_context = bdrv_get_aio_context(state->bs);
1917         aio_context_acquire(aio_context);
1918 
1919         job_cancel_sync(&state->job->job);
1920 
1921         aio_context_release(aio_context);
1922     }
1923 }
1924 
1925 static void blockdev_backup_clean(BlkActionState *common)
1926 {
1927     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1928     AioContext *aio_context;
1929 
1930     if (!state->bs) {
1931         return;
1932     }
1933 
1934     aio_context = bdrv_get_aio_context(state->bs);
1935     aio_context_acquire(aio_context);
1936 
1937     bdrv_drained_end(state->bs);
1938 
1939     aio_context_release(aio_context);
1940 }
1941 
1942 typedef struct BlockDirtyBitmapState {
1943     BlkActionState common;
1944     BdrvDirtyBitmap *bitmap;
1945     BlockDriverState *bs;
1946     HBitmap *backup;
1947     bool prepared;
1948     bool was_enabled;
1949 } BlockDirtyBitmapState;
1950 
1951 static void block_dirty_bitmap_add_prepare(BlkActionState *common,
1952                                            Error **errp)
1953 {
1954     Error *local_err = NULL;
1955     BlockDirtyBitmapAdd *action;
1956     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
1957                                              common, common);
1958 
1959     if (action_check_completion_mode(common, errp) < 0) {
1960         return;
1961     }
1962 
1963     action = common->action->u.block_dirty_bitmap_add.data;
1964     /* AIO context taken and released within qmp_block_dirty_bitmap_add */
1965     qmp_block_dirty_bitmap_add(action->node, action->name,
1966                                action->has_granularity, action->granularity,
1967                                action->has_persistent, action->persistent,
1968                                action->has_autoload, action->autoload,
1969                                action->has_disabled, action->disabled,
1970                                &local_err);
1971 
1972     if (!local_err) {
1973         state->prepared = true;
1974     } else {
1975         error_propagate(errp, local_err);
1976     }
1977 }
1978 
1979 static void block_dirty_bitmap_add_abort(BlkActionState *common)
1980 {
1981     BlockDirtyBitmapAdd *action;
1982     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
1983                                              common, common);
1984 
1985     action = common->action->u.block_dirty_bitmap_add.data;
1986     /* Should not be able to fail: IF the bitmap was added via .prepare(),
1987      * then the node reference and bitmap name must have been valid.
1988      */
1989     if (state->prepared) {
1990         qmp_block_dirty_bitmap_remove(action->node, action->name, &error_abort);
1991     }
1992 }
1993 
1994 static void block_dirty_bitmap_clear_prepare(BlkActionState *common,
1995                                              Error **errp)
1996 {
1997     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
1998                                              common, common);
1999     BlockDirtyBitmap *action;
2000 
2001     if (action_check_completion_mode(common, errp) < 0) {
2002         return;
2003     }
2004 
2005     action = common->action->u.block_dirty_bitmap_clear.data;
2006     state->bitmap = block_dirty_bitmap_lookup(action->node,
2007                                               action->name,
2008                                               &state->bs,
2009                                               errp);
2010     if (!state->bitmap) {
2011         return;
2012     }
2013 
2014     if (bdrv_dirty_bitmap_check(state->bitmap, BDRV_BITMAP_DEFAULT, errp)) {
2015         return;
2016     }
2017 
2018     bdrv_clear_dirty_bitmap(state->bitmap, &state->backup);
2019 }
2020 
2021 static void block_dirty_bitmap_restore(BlkActionState *common)
2022 {
2023     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2024                                              common, common);
2025 
2026     if (state->backup) {
2027         bdrv_restore_dirty_bitmap(state->bitmap, state->backup);
2028     }
2029 }
2030 
2031 static void block_dirty_bitmap_free_backup(BlkActionState *common)
2032 {
2033     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2034                                              common, common);
2035 
2036     hbitmap_free(state->backup);
2037 }
2038 
2039 static void block_dirty_bitmap_enable_prepare(BlkActionState *common,
2040                                               Error **errp)
2041 {
2042     BlockDirtyBitmap *action;
2043     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2044                                              common, common);
2045 
2046     if (action_check_completion_mode(common, errp) < 0) {
2047         return;
2048     }
2049 
2050     action = common->action->u.block_dirty_bitmap_enable.data;
2051     state->bitmap = block_dirty_bitmap_lookup(action->node,
2052                                               action->name,
2053                                               NULL,
2054                                               errp);
2055     if (!state->bitmap) {
2056         return;
2057     }
2058 
2059     if (bdrv_dirty_bitmap_check(state->bitmap, BDRV_BITMAP_ALLOW_RO, errp)) {
2060         return;
2061     }
2062 
2063     state->was_enabled = bdrv_dirty_bitmap_enabled(state->bitmap);
2064     bdrv_enable_dirty_bitmap(state->bitmap);
2065 }
2066 
2067 static void block_dirty_bitmap_enable_abort(BlkActionState *common)
2068 {
2069     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2070                                              common, common);
2071 
2072     if (!state->was_enabled) {
2073         bdrv_disable_dirty_bitmap(state->bitmap);
2074     }
2075 }
2076 
2077 static void block_dirty_bitmap_disable_prepare(BlkActionState *common,
2078                                                Error **errp)
2079 {
2080     BlockDirtyBitmap *action;
2081     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2082                                              common, common);
2083 
2084     if (action_check_completion_mode(common, errp) < 0) {
2085         return;
2086     }
2087 
2088     action = common->action->u.block_dirty_bitmap_disable.data;
2089     state->bitmap = block_dirty_bitmap_lookup(action->node,
2090                                               action->name,
2091                                               NULL,
2092                                               errp);
2093     if (!state->bitmap) {
2094         return;
2095     }
2096 
2097     if (bdrv_dirty_bitmap_check(state->bitmap, BDRV_BITMAP_ALLOW_RO, errp)) {
2098         return;
2099     }
2100 
2101     state->was_enabled = bdrv_dirty_bitmap_enabled(state->bitmap);
2102     bdrv_disable_dirty_bitmap(state->bitmap);
2103 }
2104 
2105 static void block_dirty_bitmap_disable_abort(BlkActionState *common)
2106 {
2107     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2108                                              common, common);
2109 
2110     if (state->was_enabled) {
2111         bdrv_enable_dirty_bitmap(state->bitmap);
2112     }
2113 }
2114 
2115 static BdrvDirtyBitmap *do_block_dirty_bitmap_merge(const char *node,
2116                                                     const char *target,
2117                                                     strList *bitmaps,
2118                                                     HBitmap **backup,
2119                                                     Error **errp);
2120 
2121 static void block_dirty_bitmap_merge_prepare(BlkActionState *common,
2122                                              Error **errp)
2123 {
2124     BlockDirtyBitmapMerge *action;
2125     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2126                                              common, common);
2127 
2128     if (action_check_completion_mode(common, errp) < 0) {
2129         return;
2130     }
2131 
2132     action = common->action->u.block_dirty_bitmap_merge.data;
2133 
2134     state->bitmap = do_block_dirty_bitmap_merge(action->node, action->target,
2135                                                 action->bitmaps, &state->backup,
2136                                                 errp);
2137 }
2138 
2139 static void abort_prepare(BlkActionState *common, Error **errp)
2140 {
2141     error_setg(errp, "Transaction aborted using Abort action");
2142 }
2143 
2144 static void abort_commit(BlkActionState *common)
2145 {
2146     g_assert_not_reached(); /* this action never succeeds */
2147 }
2148 
2149 static const BlkActionOps actions[] = {
2150     [TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT] = {
2151         .instance_size = sizeof(ExternalSnapshotState),
2152         .prepare  = external_snapshot_prepare,
2153         .commit   = external_snapshot_commit,
2154         .abort = external_snapshot_abort,
2155         .clean = external_snapshot_clean,
2156     },
2157     [TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC] = {
2158         .instance_size = sizeof(ExternalSnapshotState),
2159         .prepare  = external_snapshot_prepare,
2160         .commit   = external_snapshot_commit,
2161         .abort = external_snapshot_abort,
2162         .clean = external_snapshot_clean,
2163     },
2164     [TRANSACTION_ACTION_KIND_DRIVE_BACKUP] = {
2165         .instance_size = sizeof(DriveBackupState),
2166         .prepare = drive_backup_prepare,
2167         .commit = drive_backup_commit,
2168         .abort = drive_backup_abort,
2169         .clean = drive_backup_clean,
2170     },
2171     [TRANSACTION_ACTION_KIND_BLOCKDEV_BACKUP] = {
2172         .instance_size = sizeof(BlockdevBackupState),
2173         .prepare = blockdev_backup_prepare,
2174         .commit = blockdev_backup_commit,
2175         .abort = blockdev_backup_abort,
2176         .clean = blockdev_backup_clean,
2177     },
2178     [TRANSACTION_ACTION_KIND_ABORT] = {
2179         .instance_size = sizeof(BlkActionState),
2180         .prepare = abort_prepare,
2181         .commit = abort_commit,
2182     },
2183     [TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC] = {
2184         .instance_size = sizeof(InternalSnapshotState),
2185         .prepare  = internal_snapshot_prepare,
2186         .abort = internal_snapshot_abort,
2187         .clean = internal_snapshot_clean,
2188     },
2189     [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_ADD] = {
2190         .instance_size = sizeof(BlockDirtyBitmapState),
2191         .prepare = block_dirty_bitmap_add_prepare,
2192         .abort = block_dirty_bitmap_add_abort,
2193     },
2194     [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_CLEAR] = {
2195         .instance_size = sizeof(BlockDirtyBitmapState),
2196         .prepare = block_dirty_bitmap_clear_prepare,
2197         .commit = block_dirty_bitmap_free_backup,
2198         .abort = block_dirty_bitmap_restore,
2199     },
2200     [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_ENABLE] = {
2201         .instance_size = sizeof(BlockDirtyBitmapState),
2202         .prepare = block_dirty_bitmap_enable_prepare,
2203         .abort = block_dirty_bitmap_enable_abort,
2204     },
2205     [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_DISABLE] = {
2206         .instance_size = sizeof(BlockDirtyBitmapState),
2207         .prepare = block_dirty_bitmap_disable_prepare,
2208         .abort = block_dirty_bitmap_disable_abort,
2209     },
2210     [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_MERGE] = {
2211         .instance_size = sizeof(BlockDirtyBitmapState),
2212         .prepare = block_dirty_bitmap_merge_prepare,
2213         .commit = block_dirty_bitmap_free_backup,
2214         .abort = block_dirty_bitmap_restore,
2215     },
2216     /* Where are transactions for MIRROR, COMMIT and STREAM?
2217      * Although these blockjobs use transaction callbacks like the backup job,
2218      * these jobs do not necessarily adhere to transaction semantics.
2219      * These jobs may not fully undo all of their actions on abort, nor do they
2220      * necessarily work in transactions with more than one job in them.
2221      */
2222 };
2223 
2224 /**
2225  * Allocate a TransactionProperties structure if necessary, and fill
2226  * that structure with desired defaults if they are unset.
2227  */
2228 static TransactionProperties *get_transaction_properties(
2229     TransactionProperties *props)
2230 {
2231     if (!props) {
2232         props = g_new0(TransactionProperties, 1);
2233     }
2234 
2235     if (!props->has_completion_mode) {
2236         props->has_completion_mode = true;
2237         props->completion_mode = ACTION_COMPLETION_MODE_INDIVIDUAL;
2238     }
2239 
2240     return props;
2241 }
2242 
2243 /*
2244  * 'Atomic' group operations.  The operations are performed as a set, and if
2245  * any fail then we roll back all operations in the group.
2246  */
2247 void qmp_transaction(TransactionActionList *dev_list,
2248                      bool has_props,
2249                      struct TransactionProperties *props,
2250                      Error **errp)
2251 {
2252     TransactionActionList *dev_entry = dev_list;
2253     JobTxn *block_job_txn = NULL;
2254     BlkActionState *state, *next;
2255     Error *local_err = NULL;
2256 
2257     QTAILQ_HEAD(, BlkActionState) snap_bdrv_states;
2258     QTAILQ_INIT(&snap_bdrv_states);
2259 
2260     /* Does this transaction get canceled as a group on failure?
2261      * If not, we don't really need to make a JobTxn.
2262      */
2263     props = get_transaction_properties(props);
2264     if (props->completion_mode != ACTION_COMPLETION_MODE_INDIVIDUAL) {
2265         block_job_txn = job_txn_new();
2266     }
2267 
2268     /* drain all i/o before any operations */
2269     bdrv_drain_all();
2270 
2271     /* We don't do anything in this loop that commits us to the operations */
2272     while (NULL != dev_entry) {
2273         TransactionAction *dev_info = NULL;
2274         const BlkActionOps *ops;
2275 
2276         dev_info = dev_entry->value;
2277         dev_entry = dev_entry->next;
2278 
2279         assert(dev_info->type < ARRAY_SIZE(actions));
2280 
2281         ops = &actions[dev_info->type];
2282         assert(ops->instance_size > 0);
2283 
2284         state = g_malloc0(ops->instance_size);
2285         state->ops = ops;
2286         state->action = dev_info;
2287         state->block_job_txn = block_job_txn;
2288         state->txn_props = props;
2289         QTAILQ_INSERT_TAIL(&snap_bdrv_states, state, entry);
2290 
2291         state->ops->prepare(state, &local_err);
2292         if (local_err) {
2293             error_propagate(errp, local_err);
2294             goto delete_and_fail;
2295         }
2296     }
2297 
2298     QTAILQ_FOREACH(state, &snap_bdrv_states, entry) {
2299         if (state->ops->commit) {
2300             state->ops->commit(state);
2301         }
2302     }
2303 
2304     /* success */
2305     goto exit;
2306 
2307 delete_and_fail:
2308     /* failure, and it is all-or-none; roll back all operations */
2309     QTAILQ_FOREACH_REVERSE(state, &snap_bdrv_states, entry) {
2310         if (state->ops->abort) {
2311             state->ops->abort(state);
2312         }
2313     }
2314 exit:
2315     QTAILQ_FOREACH_SAFE(state, &snap_bdrv_states, entry, next) {
2316         if (state->ops->clean) {
2317             state->ops->clean(state);
2318         }
2319         g_free(state);
2320     }
2321     if (!has_props) {
2322         qapi_free_TransactionProperties(props);
2323     }
2324     job_txn_unref(block_job_txn);
2325 }
2326 
2327 void qmp_eject(bool has_device, const char *device,
2328                bool has_id, const char *id,
2329                bool has_force, bool force, Error **errp)
2330 {
2331     Error *local_err = NULL;
2332     int rc;
2333 
2334     if (!has_force) {
2335         force = false;
2336     }
2337 
2338     rc = do_open_tray(has_device ? device : NULL,
2339                       has_id ? id : NULL,
2340                       force, &local_err);
2341     if (rc && rc != -ENOSYS) {
2342         error_propagate(errp, local_err);
2343         return;
2344     }
2345     error_free(local_err);
2346 
2347     blockdev_remove_medium(has_device, device, has_id, id, errp);
2348 }
2349 
2350 void qmp_block_passwd(bool has_device, const char *device,
2351                       bool has_node_name, const char *node_name,
2352                       const char *password, Error **errp)
2353 {
2354     error_setg(errp,
2355                "Setting block passwords directly is no longer supported");
2356 }
2357 
2358 /*
2359  * Attempt to open the tray of @device.
2360  * If @force, ignore its tray lock.
2361  * Else, if the tray is locked, don't open it, but ask the guest to open it.
2362  * On error, store an error through @errp and return -errno.
2363  * If @device does not exist, return -ENODEV.
2364  * If it has no removable media, return -ENOTSUP.
2365  * If it has no tray, return -ENOSYS.
2366  * If the guest was asked to open the tray, return -EINPROGRESS.
2367  * Else, return 0.
2368  */
2369 static int do_open_tray(const char *blk_name, const char *qdev_id,
2370                         bool force, Error **errp)
2371 {
2372     BlockBackend *blk;
2373     const char *device = qdev_id ?: blk_name;
2374     bool locked;
2375 
2376     blk = qmp_get_blk(blk_name, qdev_id, errp);
2377     if (!blk) {
2378         return -ENODEV;
2379     }
2380 
2381     if (!blk_dev_has_removable_media(blk)) {
2382         error_setg(errp, "Device '%s' is not removable", device);
2383         return -ENOTSUP;
2384     }
2385 
2386     if (!blk_dev_has_tray(blk)) {
2387         error_setg(errp, "Device '%s' does not have a tray", device);
2388         return -ENOSYS;
2389     }
2390 
2391     if (blk_dev_is_tray_open(blk)) {
2392         return 0;
2393     }
2394 
2395     locked = blk_dev_is_medium_locked(blk);
2396     if (locked) {
2397         blk_dev_eject_request(blk, force);
2398     }
2399 
2400     if (!locked || force) {
2401         blk_dev_change_media_cb(blk, false, &error_abort);
2402     }
2403 
2404     if (locked && !force) {
2405         error_setg(errp, "Device '%s' is locked and force was not specified, "
2406                    "wait for tray to open and try again", device);
2407         return -EINPROGRESS;
2408     }
2409 
2410     return 0;
2411 }
2412 
2413 void qmp_blockdev_open_tray(bool has_device, const char *device,
2414                             bool has_id, const char *id,
2415                             bool has_force, bool force,
2416                             Error **errp)
2417 {
2418     Error *local_err = NULL;
2419     int rc;
2420 
2421     if (!has_force) {
2422         force = false;
2423     }
2424     rc = do_open_tray(has_device ? device : NULL,
2425                       has_id ? id : NULL,
2426                       force, &local_err);
2427     if (rc && rc != -ENOSYS && rc != -EINPROGRESS) {
2428         error_propagate(errp, local_err);
2429         return;
2430     }
2431     error_free(local_err);
2432 }
2433 
2434 void qmp_blockdev_close_tray(bool has_device, const char *device,
2435                              bool has_id, const char *id,
2436                              Error **errp)
2437 {
2438     BlockBackend *blk;
2439     Error *local_err = NULL;
2440 
2441     device = has_device ? device : NULL;
2442     id = has_id ? id : NULL;
2443 
2444     blk = qmp_get_blk(device, id, errp);
2445     if (!blk) {
2446         return;
2447     }
2448 
2449     if (!blk_dev_has_removable_media(blk)) {
2450         error_setg(errp, "Device '%s' is not removable", device ?: id);
2451         return;
2452     }
2453 
2454     if (!blk_dev_has_tray(blk)) {
2455         /* Ignore this command on tray-less devices */
2456         return;
2457     }
2458 
2459     if (!blk_dev_is_tray_open(blk)) {
2460         return;
2461     }
2462 
2463     blk_dev_change_media_cb(blk, true, &local_err);
2464     if (local_err) {
2465         error_propagate(errp, local_err);
2466         return;
2467     }
2468 }
2469 
2470 static void blockdev_remove_medium(bool has_device, const char *device,
2471                                    bool has_id, const char *id, Error **errp)
2472 {
2473     BlockBackend *blk;
2474     BlockDriverState *bs;
2475     AioContext *aio_context;
2476     bool has_attached_device;
2477 
2478     device = has_device ? device : NULL;
2479     id = has_id ? id : NULL;
2480 
2481     blk = qmp_get_blk(device, id, errp);
2482     if (!blk) {
2483         return;
2484     }
2485 
2486     /* For BBs without a device, we can exchange the BDS tree at will */
2487     has_attached_device = blk_get_attached_dev(blk);
2488 
2489     if (has_attached_device && !blk_dev_has_removable_media(blk)) {
2490         error_setg(errp, "Device '%s' is not removable", device ?: id);
2491         return;
2492     }
2493 
2494     if (has_attached_device && blk_dev_has_tray(blk) &&
2495         !blk_dev_is_tray_open(blk))
2496     {
2497         error_setg(errp, "Tray of device '%s' is not open", device ?: id);
2498         return;
2499     }
2500 
2501     bs = blk_bs(blk);
2502     if (!bs) {
2503         return;
2504     }
2505 
2506     aio_context = bdrv_get_aio_context(bs);
2507     aio_context_acquire(aio_context);
2508 
2509     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_EJECT, errp)) {
2510         goto out;
2511     }
2512 
2513     blk_remove_bs(blk);
2514 
2515     if (!blk_dev_has_tray(blk)) {
2516         /* For tray-less devices, blockdev-open-tray is a no-op (or may not be
2517          * called at all); therefore, the medium needs to be ejected here.
2518          * Do it after blk_remove_bs() so blk_is_inserted(blk) returns the @load
2519          * value passed here (i.e. false). */
2520         blk_dev_change_media_cb(blk, false, &error_abort);
2521     }
2522 
2523 out:
2524     aio_context_release(aio_context);
2525 }
2526 
2527 void qmp_blockdev_remove_medium(const char *id, Error **errp)
2528 {
2529     blockdev_remove_medium(false, NULL, true, id, errp);
2530 }
2531 
2532 static void qmp_blockdev_insert_anon_medium(BlockBackend *blk,
2533                                             BlockDriverState *bs, Error **errp)
2534 {
2535     Error *local_err = NULL;
2536     bool has_device;
2537     int ret;
2538 
2539     /* For BBs without a device, we can exchange the BDS tree at will */
2540     has_device = blk_get_attached_dev(blk);
2541 
2542     if (has_device && !blk_dev_has_removable_media(blk)) {
2543         error_setg(errp, "Device is not removable");
2544         return;
2545     }
2546 
2547     if (has_device && blk_dev_has_tray(blk) && !blk_dev_is_tray_open(blk)) {
2548         error_setg(errp, "Tray of the device is not open");
2549         return;
2550     }
2551 
2552     if (blk_bs(blk)) {
2553         error_setg(errp, "There already is a medium in the device");
2554         return;
2555     }
2556 
2557     ret = blk_insert_bs(blk, bs, errp);
2558     if (ret < 0) {
2559         return;
2560     }
2561 
2562     if (!blk_dev_has_tray(blk)) {
2563         /* For tray-less devices, blockdev-close-tray is a no-op (or may not be
2564          * called at all); therefore, the medium needs to be pushed into the
2565          * slot here.
2566          * Do it after blk_insert_bs() so blk_is_inserted(blk) returns the @load
2567          * value passed here (i.e. true). */
2568         blk_dev_change_media_cb(blk, true, &local_err);
2569         if (local_err) {
2570             error_propagate(errp, local_err);
2571             blk_remove_bs(blk);
2572             return;
2573         }
2574     }
2575 }
2576 
2577 static void blockdev_insert_medium(bool has_device, const char *device,
2578                                    bool has_id, const char *id,
2579                                    const char *node_name, Error **errp)
2580 {
2581     BlockBackend *blk;
2582     BlockDriverState *bs;
2583 
2584     blk = qmp_get_blk(has_device ? device : NULL,
2585                       has_id ? id : NULL,
2586                       errp);
2587     if (!blk) {
2588         return;
2589     }
2590 
2591     bs = bdrv_find_node(node_name);
2592     if (!bs) {
2593         error_setg(errp, "Node '%s' not found", node_name);
2594         return;
2595     }
2596 
2597     if (bdrv_has_blk(bs)) {
2598         error_setg(errp, "Node '%s' is already in use", node_name);
2599         return;
2600     }
2601 
2602     qmp_blockdev_insert_anon_medium(blk, bs, errp);
2603 }
2604 
2605 void qmp_blockdev_insert_medium(const char *id, const char *node_name,
2606                                 Error **errp)
2607 {
2608     blockdev_insert_medium(false, NULL, true, id, node_name, errp);
2609 }
2610 
2611 void qmp_blockdev_change_medium(bool has_device, const char *device,
2612                                 bool has_id, const char *id,
2613                                 const char *filename,
2614                                 bool has_format, const char *format,
2615                                 bool has_read_only,
2616                                 BlockdevChangeReadOnlyMode read_only,
2617                                 Error **errp)
2618 {
2619     BlockBackend *blk;
2620     BlockDriverState *medium_bs = NULL;
2621     int bdrv_flags;
2622     bool detect_zeroes;
2623     int rc;
2624     QDict *options = NULL;
2625     Error *err = NULL;
2626 
2627     blk = qmp_get_blk(has_device ? device : NULL,
2628                       has_id ? id : NULL,
2629                       errp);
2630     if (!blk) {
2631         goto fail;
2632     }
2633 
2634     if (blk_bs(blk)) {
2635         blk_update_root_state(blk);
2636     }
2637 
2638     bdrv_flags = blk_get_open_flags_from_root_state(blk);
2639     bdrv_flags &= ~(BDRV_O_TEMPORARY | BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING |
2640         BDRV_O_PROTOCOL | BDRV_O_AUTO_RDONLY);
2641 
2642     if (!has_read_only) {
2643         read_only = BLOCKDEV_CHANGE_READ_ONLY_MODE_RETAIN;
2644     }
2645 
2646     switch (read_only) {
2647     case BLOCKDEV_CHANGE_READ_ONLY_MODE_RETAIN:
2648         break;
2649 
2650     case BLOCKDEV_CHANGE_READ_ONLY_MODE_READ_ONLY:
2651         bdrv_flags &= ~BDRV_O_RDWR;
2652         break;
2653 
2654     case BLOCKDEV_CHANGE_READ_ONLY_MODE_READ_WRITE:
2655         bdrv_flags |= BDRV_O_RDWR;
2656         break;
2657 
2658     default:
2659         abort();
2660     }
2661 
2662     options = qdict_new();
2663     detect_zeroes = blk_get_detect_zeroes_from_root_state(blk);
2664     qdict_put_str(options, "detect-zeroes", detect_zeroes ? "on" : "off");
2665 
2666     if (has_format) {
2667         qdict_put_str(options, "driver", format);
2668     }
2669 
2670     medium_bs = bdrv_open(filename, NULL, options, bdrv_flags, errp);
2671     if (!medium_bs) {
2672         goto fail;
2673     }
2674 
2675     rc = do_open_tray(has_device ? device : NULL,
2676                       has_id ? id : NULL,
2677                       false, &err);
2678     if (rc && rc != -ENOSYS) {
2679         error_propagate(errp, err);
2680         goto fail;
2681     }
2682     error_free(err);
2683     err = NULL;
2684 
2685     blockdev_remove_medium(has_device, device, has_id, id, &err);
2686     if (err) {
2687         error_propagate(errp, err);
2688         goto fail;
2689     }
2690 
2691     qmp_blockdev_insert_anon_medium(blk, medium_bs, &err);
2692     if (err) {
2693         error_propagate(errp, err);
2694         goto fail;
2695     }
2696 
2697     qmp_blockdev_close_tray(has_device, device, has_id, id, errp);
2698 
2699 fail:
2700     /* If the medium has been inserted, the device has its own reference, so
2701      * ours must be relinquished; and if it has not been inserted successfully,
2702      * the reference must be relinquished anyway */
2703     bdrv_unref(medium_bs);
2704 }
2705 
2706 /* throttling disk I/O limits */
2707 void qmp_block_set_io_throttle(BlockIOThrottle *arg, Error **errp)
2708 {
2709     ThrottleConfig cfg;
2710     BlockDriverState *bs;
2711     BlockBackend *blk;
2712     AioContext *aio_context;
2713 
2714     blk = qmp_get_blk(arg->has_device ? arg->device : NULL,
2715                       arg->has_id ? arg->id : NULL,
2716                       errp);
2717     if (!blk) {
2718         return;
2719     }
2720 
2721     aio_context = blk_get_aio_context(blk);
2722     aio_context_acquire(aio_context);
2723 
2724     bs = blk_bs(blk);
2725     if (!bs) {
2726         error_setg(errp, "Device has no medium");
2727         goto out;
2728     }
2729 
2730     throttle_config_init(&cfg);
2731     cfg.buckets[THROTTLE_BPS_TOTAL].avg = arg->bps;
2732     cfg.buckets[THROTTLE_BPS_READ].avg  = arg->bps_rd;
2733     cfg.buckets[THROTTLE_BPS_WRITE].avg = arg->bps_wr;
2734 
2735     cfg.buckets[THROTTLE_OPS_TOTAL].avg = arg->iops;
2736     cfg.buckets[THROTTLE_OPS_READ].avg  = arg->iops_rd;
2737     cfg.buckets[THROTTLE_OPS_WRITE].avg = arg->iops_wr;
2738 
2739     if (arg->has_bps_max) {
2740         cfg.buckets[THROTTLE_BPS_TOTAL].max = arg->bps_max;
2741     }
2742     if (arg->has_bps_rd_max) {
2743         cfg.buckets[THROTTLE_BPS_READ].max = arg->bps_rd_max;
2744     }
2745     if (arg->has_bps_wr_max) {
2746         cfg.buckets[THROTTLE_BPS_WRITE].max = arg->bps_wr_max;
2747     }
2748     if (arg->has_iops_max) {
2749         cfg.buckets[THROTTLE_OPS_TOTAL].max = arg->iops_max;
2750     }
2751     if (arg->has_iops_rd_max) {
2752         cfg.buckets[THROTTLE_OPS_READ].max = arg->iops_rd_max;
2753     }
2754     if (arg->has_iops_wr_max) {
2755         cfg.buckets[THROTTLE_OPS_WRITE].max = arg->iops_wr_max;
2756     }
2757 
2758     if (arg->has_bps_max_length) {
2759         cfg.buckets[THROTTLE_BPS_TOTAL].burst_length = arg->bps_max_length;
2760     }
2761     if (arg->has_bps_rd_max_length) {
2762         cfg.buckets[THROTTLE_BPS_READ].burst_length = arg->bps_rd_max_length;
2763     }
2764     if (arg->has_bps_wr_max_length) {
2765         cfg.buckets[THROTTLE_BPS_WRITE].burst_length = arg->bps_wr_max_length;
2766     }
2767     if (arg->has_iops_max_length) {
2768         cfg.buckets[THROTTLE_OPS_TOTAL].burst_length = arg->iops_max_length;
2769     }
2770     if (arg->has_iops_rd_max_length) {
2771         cfg.buckets[THROTTLE_OPS_READ].burst_length = arg->iops_rd_max_length;
2772     }
2773     if (arg->has_iops_wr_max_length) {
2774         cfg.buckets[THROTTLE_OPS_WRITE].burst_length = arg->iops_wr_max_length;
2775     }
2776 
2777     if (arg->has_iops_size) {
2778         cfg.op_size = arg->iops_size;
2779     }
2780 
2781     if (!throttle_is_valid(&cfg, errp)) {
2782         goto out;
2783     }
2784 
2785     if (throttle_enabled(&cfg)) {
2786         /* Enable I/O limits if they're not enabled yet, otherwise
2787          * just update the throttling group. */
2788         if (!blk_get_public(blk)->throttle_group_member.throttle_state) {
2789             blk_io_limits_enable(blk,
2790                                  arg->has_group ? arg->group :
2791                                  arg->has_device ? arg->device :
2792                                  arg->id);
2793         } else if (arg->has_group) {
2794             blk_io_limits_update_group(blk, arg->group);
2795         }
2796         /* Set the new throttling configuration */
2797         blk_set_io_limits(blk, &cfg);
2798     } else if (blk_get_public(blk)->throttle_group_member.throttle_state) {
2799         /* If all throttling settings are set to 0, disable I/O limits */
2800         blk_io_limits_disable(blk);
2801     }
2802 
2803 out:
2804     aio_context_release(aio_context);
2805 }
2806 
2807 void qmp_block_dirty_bitmap_add(const char *node, const char *name,
2808                                 bool has_granularity, uint32_t granularity,
2809                                 bool has_persistent, bool persistent,
2810                                 bool has_autoload, bool autoload,
2811                                 bool has_disabled, bool disabled,
2812                                 Error **errp)
2813 {
2814     BlockDriverState *bs;
2815     BdrvDirtyBitmap *bitmap;
2816     AioContext *aio_context = NULL;
2817 
2818     if (!name || name[0] == '\0') {
2819         error_setg(errp, "Bitmap name cannot be empty");
2820         return;
2821     }
2822 
2823     bs = bdrv_lookup_bs(node, node, errp);
2824     if (!bs) {
2825         return;
2826     }
2827 
2828     if (has_granularity) {
2829         if (granularity < 512 || !is_power_of_2(granularity)) {
2830             error_setg(errp, "Granularity must be power of 2 "
2831                              "and at least 512");
2832             return;
2833         }
2834     } else {
2835         /* Default to cluster size, if available: */
2836         granularity = bdrv_get_default_bitmap_granularity(bs);
2837     }
2838 
2839     if (!has_persistent) {
2840         persistent = false;
2841     }
2842 
2843     if (has_autoload) {
2844         warn_report("Autoload option is deprecated and its value is ignored");
2845     }
2846 
2847     if (!has_disabled) {
2848         disabled = false;
2849     }
2850 
2851     if (persistent) {
2852         aio_context = bdrv_get_aio_context(bs);
2853         aio_context_acquire(aio_context);
2854         if (!bdrv_can_store_new_dirty_bitmap(bs, name, granularity, errp)) {
2855             goto out;
2856         }
2857     }
2858 
2859     bitmap = bdrv_create_dirty_bitmap(bs, granularity, name, errp);
2860     if (bitmap == NULL) {
2861         goto out;
2862     }
2863 
2864     if (disabled) {
2865         bdrv_disable_dirty_bitmap(bitmap);
2866     }
2867 
2868     bdrv_dirty_bitmap_set_persistence(bitmap, persistent);
2869  out:
2870     if (aio_context) {
2871         aio_context_release(aio_context);
2872     }
2873 }
2874 
2875 void qmp_block_dirty_bitmap_remove(const char *node, const char *name,
2876                                    Error **errp)
2877 {
2878     BlockDriverState *bs;
2879     BdrvDirtyBitmap *bitmap;
2880     Error *local_err = NULL;
2881     AioContext *aio_context = NULL;
2882 
2883     bitmap = block_dirty_bitmap_lookup(node, name, &bs, errp);
2884     if (!bitmap || !bs) {
2885         return;
2886     }
2887 
2888     if (bdrv_dirty_bitmap_check(bitmap, BDRV_BITMAP_BUSY | BDRV_BITMAP_RO,
2889                                 errp)) {
2890         return;
2891     }
2892 
2893     if (bdrv_dirty_bitmap_get_persistence(bitmap)) {
2894         aio_context = bdrv_get_aio_context(bs);
2895         aio_context_acquire(aio_context);
2896         bdrv_remove_persistent_dirty_bitmap(bs, name, &local_err);
2897         if (local_err != NULL) {
2898             error_propagate(errp, local_err);
2899             goto out;
2900         }
2901     }
2902 
2903     bdrv_release_dirty_bitmap(bs, bitmap);
2904  out:
2905     if (aio_context) {
2906         aio_context_release(aio_context);
2907     }
2908 }
2909 
2910 /**
2911  * Completely clear a bitmap, for the purposes of synchronizing a bitmap
2912  * immediately after a full backup operation.
2913  */
2914 void qmp_block_dirty_bitmap_clear(const char *node, const char *name,
2915                                   Error **errp)
2916 {
2917     BdrvDirtyBitmap *bitmap;
2918     BlockDriverState *bs;
2919 
2920     bitmap = block_dirty_bitmap_lookup(node, name, &bs, errp);
2921     if (!bitmap || !bs) {
2922         return;
2923     }
2924 
2925     if (bdrv_dirty_bitmap_check(bitmap, BDRV_BITMAP_DEFAULT, errp)) {
2926         return;
2927     }
2928 
2929     bdrv_clear_dirty_bitmap(bitmap, NULL);
2930 }
2931 
2932 void qmp_block_dirty_bitmap_enable(const char *node, const char *name,
2933                                    Error **errp)
2934 {
2935     BlockDriverState *bs;
2936     BdrvDirtyBitmap *bitmap;
2937 
2938     bitmap = block_dirty_bitmap_lookup(node, name, &bs, errp);
2939     if (!bitmap) {
2940         return;
2941     }
2942 
2943     if (bdrv_dirty_bitmap_check(bitmap, BDRV_BITMAP_ALLOW_RO, errp)) {
2944         return;
2945     }
2946 
2947     bdrv_enable_dirty_bitmap(bitmap);
2948 }
2949 
2950 void qmp_block_dirty_bitmap_disable(const char *node, const char *name,
2951                                     Error **errp)
2952 {
2953     BlockDriverState *bs;
2954     BdrvDirtyBitmap *bitmap;
2955 
2956     bitmap = block_dirty_bitmap_lookup(node, name, &bs, errp);
2957     if (!bitmap) {
2958         return;
2959     }
2960 
2961     if (bdrv_dirty_bitmap_check(bitmap, BDRV_BITMAP_ALLOW_RO, errp)) {
2962         return;
2963     }
2964 
2965     bdrv_disable_dirty_bitmap(bitmap);
2966 }
2967 
2968 static BdrvDirtyBitmap *do_block_dirty_bitmap_merge(const char *node,
2969                                                     const char *target,
2970                                                     strList *bitmaps,
2971                                                     HBitmap **backup,
2972                                                     Error **errp)
2973 {
2974     BlockDriverState *bs;
2975     BdrvDirtyBitmap *dst, *src, *anon;
2976     strList *lst;
2977     Error *local_err = NULL;
2978 
2979     dst = block_dirty_bitmap_lookup(node, target, &bs, errp);
2980     if (!dst) {
2981         return NULL;
2982     }
2983 
2984     anon = bdrv_create_dirty_bitmap(bs, bdrv_dirty_bitmap_granularity(dst),
2985                                     NULL, errp);
2986     if (!anon) {
2987         return NULL;
2988     }
2989 
2990     for (lst = bitmaps; lst; lst = lst->next) {
2991         src = bdrv_find_dirty_bitmap(bs, lst->value);
2992         if (!src) {
2993             error_setg(errp, "Dirty bitmap '%s' not found", lst->value);
2994             dst = NULL;
2995             goto out;
2996         }
2997 
2998         bdrv_merge_dirty_bitmap(anon, src, NULL, &local_err);
2999         if (local_err) {
3000             error_propagate(errp, local_err);
3001             dst = NULL;
3002             goto out;
3003         }
3004     }
3005 
3006     /* Merge into dst; dst is unchanged on failure. */
3007     bdrv_merge_dirty_bitmap(dst, anon, backup, errp);
3008 
3009  out:
3010     bdrv_release_dirty_bitmap(bs, anon);
3011     return dst;
3012 }
3013 
3014 void qmp_block_dirty_bitmap_merge(const char *node, const char *target,
3015                                   strList *bitmaps, Error **errp)
3016 {
3017     do_block_dirty_bitmap_merge(node, target, bitmaps, NULL, errp);
3018 }
3019 
3020 BlockDirtyBitmapSha256 *qmp_x_debug_block_dirty_bitmap_sha256(const char *node,
3021                                                               const char *name,
3022                                                               Error **errp)
3023 {
3024     BdrvDirtyBitmap *bitmap;
3025     BlockDriverState *bs;
3026     BlockDirtyBitmapSha256 *ret = NULL;
3027     char *sha256;
3028 
3029     bitmap = block_dirty_bitmap_lookup(node, name, &bs, errp);
3030     if (!bitmap || !bs) {
3031         return NULL;
3032     }
3033 
3034     sha256 = bdrv_dirty_bitmap_sha256(bitmap, errp);
3035     if (sha256 == NULL) {
3036         return NULL;
3037     }
3038 
3039     ret = g_new(BlockDirtyBitmapSha256, 1);
3040     ret->sha256 = sha256;
3041 
3042     return ret;
3043 }
3044 
3045 void hmp_drive_del(Monitor *mon, const QDict *qdict)
3046 {
3047     const char *id = qdict_get_str(qdict, "id");
3048     BlockBackend *blk;
3049     BlockDriverState *bs;
3050     AioContext *aio_context;
3051     Error *local_err = NULL;
3052 
3053     bs = bdrv_find_node(id);
3054     if (bs) {
3055         qmp_blockdev_del(id, &local_err);
3056         if (local_err) {
3057             error_report_err(local_err);
3058         }
3059         return;
3060     }
3061 
3062     blk = blk_by_name(id);
3063     if (!blk) {
3064         error_report("Device '%s' not found", id);
3065         return;
3066     }
3067 
3068     if (!blk_legacy_dinfo(blk)) {
3069         error_report("Deleting device added with blockdev-add"
3070                      " is not supported");
3071         return;
3072     }
3073 
3074     aio_context = blk_get_aio_context(blk);
3075     aio_context_acquire(aio_context);
3076 
3077     bs = blk_bs(blk);
3078     if (bs) {
3079         if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_DRIVE_DEL, &local_err)) {
3080             error_report_err(local_err);
3081             aio_context_release(aio_context);
3082             return;
3083         }
3084 
3085         blk_remove_bs(blk);
3086     }
3087 
3088     /* Make the BlockBackend and the attached BlockDriverState anonymous */
3089     monitor_remove_blk(blk);
3090 
3091     /* If this BlockBackend has a device attached to it, its refcount will be
3092      * decremented when the device is removed; otherwise we have to do so here.
3093      */
3094     if (blk_get_attached_dev(blk)) {
3095         /* Further I/O must not pause the guest */
3096         blk_set_on_error(blk, BLOCKDEV_ON_ERROR_REPORT,
3097                          BLOCKDEV_ON_ERROR_REPORT);
3098     } else {
3099         blk_unref(blk);
3100     }
3101 
3102     aio_context_release(aio_context);
3103 }
3104 
3105 void qmp_block_resize(bool has_device, const char *device,
3106                       bool has_node_name, const char *node_name,
3107                       int64_t size, Error **errp)
3108 {
3109     Error *local_err = NULL;
3110     BlockBackend *blk = NULL;
3111     BlockDriverState *bs;
3112     AioContext *aio_context;
3113     int ret;
3114 
3115     bs = bdrv_lookup_bs(has_device ? device : NULL,
3116                         has_node_name ? node_name : NULL,
3117                         &local_err);
3118     if (local_err) {
3119         error_propagate(errp, local_err);
3120         return;
3121     }
3122 
3123     aio_context = bdrv_get_aio_context(bs);
3124     aio_context_acquire(aio_context);
3125 
3126     if (!bdrv_is_first_non_filter(bs)) {
3127         error_setg(errp, QERR_FEATURE_DISABLED, "resize");
3128         goto out;
3129     }
3130 
3131     if (size < 0) {
3132         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "size", "a >0 size");
3133         goto out;
3134     }
3135 
3136     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_RESIZE, NULL)) {
3137         error_setg(errp, QERR_DEVICE_IN_USE, device);
3138         goto out;
3139     }
3140 
3141     blk = blk_new(BLK_PERM_RESIZE, BLK_PERM_ALL);
3142     ret = blk_insert_bs(blk, bs, errp);
3143     if (ret < 0) {
3144         goto out;
3145     }
3146 
3147     bdrv_drained_begin(bs);
3148     ret = blk_truncate(blk, size, PREALLOC_MODE_OFF, errp);
3149     bdrv_drained_end(bs);
3150 
3151 out:
3152     blk_unref(blk);
3153     aio_context_release(aio_context);
3154 }
3155 
3156 void qmp_block_stream(bool has_job_id, const char *job_id, const char *device,
3157                       bool has_base, const char *base,
3158                       bool has_base_node, const char *base_node,
3159                       bool has_backing_file, const char *backing_file,
3160                       bool has_speed, int64_t speed,
3161                       bool has_on_error, BlockdevOnError on_error,
3162                       bool has_auto_finalize, bool auto_finalize,
3163                       bool has_auto_dismiss, bool auto_dismiss,
3164                       Error **errp)
3165 {
3166     BlockDriverState *bs, *iter;
3167     BlockDriverState *base_bs = NULL;
3168     AioContext *aio_context;
3169     Error *local_err = NULL;
3170     const char *base_name = NULL;
3171     int job_flags = JOB_DEFAULT;
3172 
3173     if (!has_on_error) {
3174         on_error = BLOCKDEV_ON_ERROR_REPORT;
3175     }
3176 
3177     bs = bdrv_lookup_bs(device, device, errp);
3178     if (!bs) {
3179         return;
3180     }
3181 
3182     aio_context = bdrv_get_aio_context(bs);
3183     aio_context_acquire(aio_context);
3184 
3185     if (has_base && has_base_node) {
3186         error_setg(errp, "'base' and 'base-node' cannot be specified "
3187                    "at the same time");
3188         goto out;
3189     }
3190 
3191     if (has_base) {
3192         base_bs = bdrv_find_backing_image(bs, base);
3193         if (base_bs == NULL) {
3194             error_setg(errp, QERR_BASE_NOT_FOUND, base);
3195             goto out;
3196         }
3197         assert(bdrv_get_aio_context(base_bs) == aio_context);
3198         base_name = base;
3199     }
3200 
3201     if (has_base_node) {
3202         base_bs = bdrv_lookup_bs(NULL, base_node, errp);
3203         if (!base_bs) {
3204             goto out;
3205         }
3206         if (bs == base_bs || !bdrv_chain_contains(bs, base_bs)) {
3207             error_setg(errp, "Node '%s' is not a backing image of '%s'",
3208                        base_node, device);
3209             goto out;
3210         }
3211         assert(bdrv_get_aio_context(base_bs) == aio_context);
3212         bdrv_refresh_filename(base_bs);
3213         base_name = base_bs->filename;
3214     }
3215 
3216     /* Check for op blockers in the whole chain between bs and base */
3217     for (iter = bs; iter && iter != base_bs; iter = backing_bs(iter)) {
3218         if (bdrv_op_is_blocked(iter, BLOCK_OP_TYPE_STREAM, errp)) {
3219             goto out;
3220         }
3221     }
3222 
3223     /* if we are streaming the entire chain, the result will have no backing
3224      * file, and specifying one is therefore an error */
3225     if (base_bs == NULL && has_backing_file) {
3226         error_setg(errp, "backing file specified, but streaming the "
3227                          "entire chain");
3228         goto out;
3229     }
3230 
3231     /* backing_file string overrides base bs filename */
3232     base_name = has_backing_file ? backing_file : base_name;
3233 
3234     if (has_auto_finalize && !auto_finalize) {
3235         job_flags |= JOB_MANUAL_FINALIZE;
3236     }
3237     if (has_auto_dismiss && !auto_dismiss) {
3238         job_flags |= JOB_MANUAL_DISMISS;
3239     }
3240 
3241     stream_start(has_job_id ? job_id : NULL, bs, base_bs, base_name,
3242                  job_flags, has_speed ? speed : 0, on_error, &local_err);
3243     if (local_err) {
3244         error_propagate(errp, local_err);
3245         goto out;
3246     }
3247 
3248     trace_qmp_block_stream(bs, bs->job);
3249 
3250 out:
3251     aio_context_release(aio_context);
3252 }
3253 
3254 void qmp_block_commit(bool has_job_id, const char *job_id, const char *device,
3255                       bool has_base_node, const char *base_node,
3256                       bool has_base, const char *base,
3257                       bool has_top_node, const char *top_node,
3258                       bool has_top, const char *top,
3259                       bool has_backing_file, const char *backing_file,
3260                       bool has_speed, int64_t speed,
3261                       bool has_filter_node_name, const char *filter_node_name,
3262                       bool has_auto_finalize, bool auto_finalize,
3263                       bool has_auto_dismiss, bool auto_dismiss,
3264                       Error **errp)
3265 {
3266     BlockDriverState *bs;
3267     BlockDriverState *iter;
3268     BlockDriverState *base_bs, *top_bs;
3269     AioContext *aio_context;
3270     Error *local_err = NULL;
3271     /* This will be part of the QMP command, if/when the
3272      * BlockdevOnError change for blkmirror makes it in
3273      */
3274     BlockdevOnError on_error = BLOCKDEV_ON_ERROR_REPORT;
3275     int job_flags = JOB_DEFAULT;
3276 
3277     if (!has_speed) {
3278         speed = 0;
3279     }
3280     if (!has_filter_node_name) {
3281         filter_node_name = NULL;
3282     }
3283     if (has_auto_finalize && !auto_finalize) {
3284         job_flags |= JOB_MANUAL_FINALIZE;
3285     }
3286     if (has_auto_dismiss && !auto_dismiss) {
3287         job_flags |= JOB_MANUAL_DISMISS;
3288     }
3289 
3290     /* Important Note:
3291      *  libvirt relies on the DeviceNotFound error class in order to probe for
3292      *  live commit feature versions; for this to work, we must make sure to
3293      *  perform the device lookup before any generic errors that may occur in a
3294      *  scenario in which all optional arguments are omitted. */
3295     bs = qmp_get_root_bs(device, &local_err);
3296     if (!bs) {
3297         bs = bdrv_lookup_bs(device, device, NULL);
3298         if (!bs) {
3299             error_free(local_err);
3300             error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
3301                       "Device '%s' not found", device);
3302         } else {
3303             error_propagate(errp, local_err);
3304         }
3305         return;
3306     }
3307 
3308     aio_context = bdrv_get_aio_context(bs);
3309     aio_context_acquire(aio_context);
3310 
3311     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_COMMIT_SOURCE, errp)) {
3312         goto out;
3313     }
3314 
3315     /* default top_bs is the active layer */
3316     top_bs = bs;
3317 
3318     if (has_top_node && has_top) {
3319         error_setg(errp, "'top-node' and 'top' are mutually exclusive");
3320         goto out;
3321     } else if (has_top_node) {
3322         top_bs = bdrv_lookup_bs(NULL, top_node, errp);
3323         if (top_bs == NULL) {
3324             goto out;
3325         }
3326         if (!bdrv_chain_contains(bs, top_bs)) {
3327             error_setg(errp, "'%s' is not in this backing file chain",
3328                        top_node);
3329             goto out;
3330         }
3331     } else if (has_top && top) {
3332         /* This strcmp() is just a shortcut, there is no need to
3333          * refresh @bs's filename.  If it mismatches,
3334          * bdrv_find_backing_image() will do the refresh and may still
3335          * return @bs. */
3336         if (strcmp(bs->filename, top) != 0) {
3337             top_bs = bdrv_find_backing_image(bs, top);
3338         }
3339     }
3340 
3341     if (top_bs == NULL) {
3342         error_setg(errp, "Top image file %s not found", top ? top : "NULL");
3343         goto out;
3344     }
3345 
3346     assert(bdrv_get_aio_context(top_bs) == aio_context);
3347 
3348     if (has_base_node && has_base) {
3349         error_setg(errp, "'base-node' and 'base' are mutually exclusive");
3350         goto out;
3351     } else if (has_base_node) {
3352         base_bs = bdrv_lookup_bs(NULL, base_node, errp);
3353         if (base_bs == NULL) {
3354             goto out;
3355         }
3356         if (!bdrv_chain_contains(top_bs, base_bs)) {
3357             error_setg(errp, "'%s' is not in this backing file chain",
3358                        base_node);
3359             goto out;
3360         }
3361     } else if (has_base && base) {
3362         base_bs = bdrv_find_backing_image(top_bs, base);
3363     } else {
3364         base_bs = bdrv_find_base(top_bs);
3365     }
3366 
3367     if (base_bs == NULL) {
3368         error_setg(errp, QERR_BASE_NOT_FOUND, base ? base : "NULL");
3369         goto out;
3370     }
3371 
3372     assert(bdrv_get_aio_context(base_bs) == aio_context);
3373 
3374     for (iter = top_bs; iter != backing_bs(base_bs); iter = backing_bs(iter)) {
3375         if (bdrv_op_is_blocked(iter, BLOCK_OP_TYPE_COMMIT_TARGET, errp)) {
3376             goto out;
3377         }
3378     }
3379 
3380     /* Do not allow attempts to commit an image into itself */
3381     if (top_bs == base_bs) {
3382         error_setg(errp, "cannot commit an image into itself");
3383         goto out;
3384     }
3385 
3386     if (top_bs == bs) {
3387         if (has_backing_file) {
3388             error_setg(errp, "'backing-file' specified,"
3389                              " but 'top' is the active layer");
3390             goto out;
3391         }
3392         commit_active_start(has_job_id ? job_id : NULL, bs, base_bs,
3393                             job_flags, speed, on_error,
3394                             filter_node_name, NULL, NULL, false, &local_err);
3395     } else {
3396         BlockDriverState *overlay_bs = bdrv_find_overlay(bs, top_bs);
3397         if (bdrv_op_is_blocked(overlay_bs, BLOCK_OP_TYPE_COMMIT_TARGET, errp)) {
3398             goto out;
3399         }
3400         commit_start(has_job_id ? job_id : NULL, bs, base_bs, top_bs, job_flags,
3401                      speed, on_error, has_backing_file ? backing_file : NULL,
3402                      filter_node_name, &local_err);
3403     }
3404     if (local_err != NULL) {
3405         error_propagate(errp, local_err);
3406         goto out;
3407     }
3408 
3409 out:
3410     aio_context_release(aio_context);
3411 }
3412 
3413 static BlockJob *do_drive_backup(DriveBackup *backup, JobTxn *txn,
3414                                  Error **errp)
3415 {
3416     BlockDriverState *bs;
3417     BlockDriverState *target_bs;
3418     BlockDriverState *source = NULL;
3419     BlockJob *job = NULL;
3420     BdrvDirtyBitmap *bmap = NULL;
3421     AioContext *aio_context;
3422     QDict *options = NULL;
3423     Error *local_err = NULL;
3424     int flags, job_flags = JOB_DEFAULT;
3425     int64_t size;
3426     bool set_backing_hd = false;
3427 
3428     if (!backup->has_speed) {
3429         backup->speed = 0;
3430     }
3431     if (!backup->has_on_source_error) {
3432         backup->on_source_error = BLOCKDEV_ON_ERROR_REPORT;
3433     }
3434     if (!backup->has_on_target_error) {
3435         backup->on_target_error = BLOCKDEV_ON_ERROR_REPORT;
3436     }
3437     if (!backup->has_mode) {
3438         backup->mode = NEW_IMAGE_MODE_ABSOLUTE_PATHS;
3439     }
3440     if (!backup->has_job_id) {
3441         backup->job_id = NULL;
3442     }
3443     if (!backup->has_auto_finalize) {
3444         backup->auto_finalize = true;
3445     }
3446     if (!backup->has_auto_dismiss) {
3447         backup->auto_dismiss = true;
3448     }
3449     if (!backup->has_compress) {
3450         backup->compress = false;
3451     }
3452 
3453     bs = qmp_get_root_bs(backup->device, errp);
3454     if (!bs) {
3455         return NULL;
3456     }
3457 
3458     aio_context = bdrv_get_aio_context(bs);
3459     aio_context_acquire(aio_context);
3460 
3461     if (!backup->has_format) {
3462         backup->format = backup->mode == NEW_IMAGE_MODE_EXISTING ?
3463                          NULL : (char*) bs->drv->format_name;
3464     }
3465 
3466     /* Early check to avoid creating target */
3467     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_BACKUP_SOURCE, errp)) {
3468         goto out;
3469     }
3470 
3471     flags = bs->open_flags | BDRV_O_RDWR;
3472 
3473     /* See if we have a backing HD we can use to create our new image
3474      * on top of. */
3475     if (backup->sync == MIRROR_SYNC_MODE_TOP) {
3476         source = backing_bs(bs);
3477         if (!source) {
3478             backup->sync = MIRROR_SYNC_MODE_FULL;
3479         }
3480     }
3481     if (backup->sync == MIRROR_SYNC_MODE_NONE) {
3482         source = bs;
3483         flags |= BDRV_O_NO_BACKING;
3484         set_backing_hd = true;
3485     }
3486 
3487     size = bdrv_getlength(bs);
3488     if (size < 0) {
3489         error_setg_errno(errp, -size, "bdrv_getlength failed");
3490         goto out;
3491     }
3492 
3493     if (backup->mode != NEW_IMAGE_MODE_EXISTING) {
3494         assert(backup->format);
3495         if (source) {
3496             bdrv_refresh_filename(source);
3497             bdrv_img_create(backup->target, backup->format, source->filename,
3498                             source->drv->format_name, NULL,
3499                             size, flags, false, &local_err);
3500         } else {
3501             bdrv_img_create(backup->target, backup->format, NULL, NULL, NULL,
3502                             size, flags, false, &local_err);
3503         }
3504     }
3505 
3506     if (local_err) {
3507         error_propagate(errp, local_err);
3508         goto out;
3509     }
3510 
3511     if (backup->format) {
3512         if (!options) {
3513             options = qdict_new();
3514         }
3515         qdict_put_str(options, "driver", backup->format);
3516     }
3517 
3518     target_bs = bdrv_open(backup->target, NULL, options, flags, errp);
3519     if (!target_bs) {
3520         goto out;
3521     }
3522 
3523     bdrv_set_aio_context(target_bs, aio_context);
3524 
3525     if (set_backing_hd) {
3526         bdrv_set_backing_hd(target_bs, source, &local_err);
3527         if (local_err) {
3528             bdrv_unref(target_bs);
3529             goto out;
3530         }
3531     }
3532 
3533     if (backup->has_bitmap) {
3534         bmap = bdrv_find_dirty_bitmap(bs, backup->bitmap);
3535         if (!bmap) {
3536             error_setg(errp, "Bitmap '%s' could not be found", backup->bitmap);
3537             bdrv_unref(target_bs);
3538             goto out;
3539         }
3540         if (bdrv_dirty_bitmap_check(bmap, BDRV_BITMAP_DEFAULT, errp)) {
3541             goto out;
3542         }
3543     }
3544     if (!backup->auto_finalize) {
3545         job_flags |= JOB_MANUAL_FINALIZE;
3546     }
3547     if (!backup->auto_dismiss) {
3548         job_flags |= JOB_MANUAL_DISMISS;
3549     }
3550 
3551     job = backup_job_create(backup->job_id, bs, target_bs, backup->speed,
3552                             backup->sync, bmap, backup->compress,
3553                             backup->on_source_error, backup->on_target_error,
3554                             job_flags, NULL, NULL, txn, &local_err);
3555     bdrv_unref(target_bs);
3556     if (local_err != NULL) {
3557         error_propagate(errp, local_err);
3558         goto out;
3559     }
3560 
3561 out:
3562     aio_context_release(aio_context);
3563     return job;
3564 }
3565 
3566 void qmp_drive_backup(DriveBackup *arg, Error **errp)
3567 {
3568 
3569     BlockJob *job;
3570     job = do_drive_backup(arg, NULL, errp);
3571     if (job) {
3572         job_start(&job->job);
3573     }
3574 }
3575 
3576 BlockDeviceInfoList *qmp_query_named_block_nodes(Error **errp)
3577 {
3578     return bdrv_named_nodes_list(errp);
3579 }
3580 
3581 XDbgBlockGraph *qmp_x_debug_query_block_graph(Error **errp)
3582 {
3583     return bdrv_get_xdbg_block_graph(errp);
3584 }
3585 
3586 BlockJob *do_blockdev_backup(BlockdevBackup *backup, JobTxn *txn,
3587                              Error **errp)
3588 {
3589     BlockDriverState *bs;
3590     BlockDriverState *target_bs;
3591     Error *local_err = NULL;
3592     BdrvDirtyBitmap *bmap = NULL;
3593     AioContext *aio_context;
3594     BlockJob *job = NULL;
3595     int job_flags = JOB_DEFAULT;
3596 
3597     if (!backup->has_speed) {
3598         backup->speed = 0;
3599     }
3600     if (!backup->has_on_source_error) {
3601         backup->on_source_error = BLOCKDEV_ON_ERROR_REPORT;
3602     }
3603     if (!backup->has_on_target_error) {
3604         backup->on_target_error = BLOCKDEV_ON_ERROR_REPORT;
3605     }
3606     if (!backup->has_job_id) {
3607         backup->job_id = NULL;
3608     }
3609     if (!backup->has_auto_finalize) {
3610         backup->auto_finalize = true;
3611     }
3612     if (!backup->has_auto_dismiss) {
3613         backup->auto_dismiss = true;
3614     }
3615     if (!backup->has_compress) {
3616         backup->compress = false;
3617     }
3618 
3619     bs = bdrv_lookup_bs(backup->device, backup->device, errp);
3620     if (!bs) {
3621         return NULL;
3622     }
3623 
3624     aio_context = bdrv_get_aio_context(bs);
3625     aio_context_acquire(aio_context);
3626 
3627     target_bs = bdrv_lookup_bs(backup->target, backup->target, errp);
3628     if (!target_bs) {
3629         goto out;
3630     }
3631 
3632     if (bdrv_get_aio_context(target_bs) != aio_context) {
3633         if (!bdrv_has_blk(target_bs)) {
3634             /* The target BDS is not attached, we can safely move it to another
3635              * AioContext. */
3636             bdrv_set_aio_context(target_bs, aio_context);
3637         } else {
3638             error_setg(errp, "Target is attached to a different thread from "
3639                              "source.");
3640             goto out;
3641         }
3642     }
3643 
3644     if (backup->has_bitmap) {
3645         bmap = bdrv_find_dirty_bitmap(bs, backup->bitmap);
3646         if (!bmap) {
3647             error_setg(errp, "Bitmap '%s' could not be found", backup->bitmap);
3648             goto out;
3649         }
3650         if (bdrv_dirty_bitmap_check(bmap, BDRV_BITMAP_DEFAULT, errp)) {
3651             goto out;
3652         }
3653     }
3654 
3655     if (!backup->auto_finalize) {
3656         job_flags |= JOB_MANUAL_FINALIZE;
3657     }
3658     if (!backup->auto_dismiss) {
3659         job_flags |= JOB_MANUAL_DISMISS;
3660     }
3661     job = backup_job_create(backup->job_id, bs, target_bs, backup->speed,
3662                             backup->sync, bmap, backup->compress,
3663                             backup->on_source_error, backup->on_target_error,
3664                             job_flags, NULL, NULL, txn, &local_err);
3665     if (local_err != NULL) {
3666         error_propagate(errp, local_err);
3667     }
3668 out:
3669     aio_context_release(aio_context);
3670     return job;
3671 }
3672 
3673 void qmp_blockdev_backup(BlockdevBackup *arg, Error **errp)
3674 {
3675     BlockJob *job;
3676     job = do_blockdev_backup(arg, NULL, errp);
3677     if (job) {
3678         job_start(&job->job);
3679     }
3680 }
3681 
3682 /* Parameter check and block job starting for drive mirroring.
3683  * Caller should hold @device and @target's aio context (must be the same).
3684  **/
3685 static void blockdev_mirror_common(const char *job_id, BlockDriverState *bs,
3686                                    BlockDriverState *target,
3687                                    bool has_replaces, const char *replaces,
3688                                    enum MirrorSyncMode sync,
3689                                    BlockMirrorBackingMode backing_mode,
3690                                    bool has_speed, int64_t speed,
3691                                    bool has_granularity, uint32_t granularity,
3692                                    bool has_buf_size, int64_t buf_size,
3693                                    bool has_on_source_error,
3694                                    BlockdevOnError on_source_error,
3695                                    bool has_on_target_error,
3696                                    BlockdevOnError on_target_error,
3697                                    bool has_unmap, bool unmap,
3698                                    bool has_filter_node_name,
3699                                    const char *filter_node_name,
3700                                    bool has_copy_mode, MirrorCopyMode copy_mode,
3701                                    bool has_auto_finalize, bool auto_finalize,
3702                                    bool has_auto_dismiss, bool auto_dismiss,
3703                                    Error **errp)
3704 {
3705     int job_flags = JOB_DEFAULT;
3706 
3707     if (!has_speed) {
3708         speed = 0;
3709     }
3710     if (!has_on_source_error) {
3711         on_source_error = BLOCKDEV_ON_ERROR_REPORT;
3712     }
3713     if (!has_on_target_error) {
3714         on_target_error = BLOCKDEV_ON_ERROR_REPORT;
3715     }
3716     if (!has_granularity) {
3717         granularity = 0;
3718     }
3719     if (!has_buf_size) {
3720         buf_size = 0;
3721     }
3722     if (!has_unmap) {
3723         unmap = true;
3724     }
3725     if (!has_filter_node_name) {
3726         filter_node_name = NULL;
3727     }
3728     if (!has_copy_mode) {
3729         copy_mode = MIRROR_COPY_MODE_BACKGROUND;
3730     }
3731     if (has_auto_finalize && !auto_finalize) {
3732         job_flags |= JOB_MANUAL_FINALIZE;
3733     }
3734     if (has_auto_dismiss && !auto_dismiss) {
3735         job_flags |= JOB_MANUAL_DISMISS;
3736     }
3737 
3738     if (granularity != 0 && (granularity < 512 || granularity > 1048576 * 64)) {
3739         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "granularity",
3740                    "a value in range [512B, 64MB]");
3741         return;
3742     }
3743     if (granularity & (granularity - 1)) {
3744         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "granularity",
3745                    "power of 2");
3746         return;
3747     }
3748 
3749     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_MIRROR_SOURCE, errp)) {
3750         return;
3751     }
3752     if (bdrv_op_is_blocked(target, BLOCK_OP_TYPE_MIRROR_TARGET, errp)) {
3753         return;
3754     }
3755 
3756     if (!bs->backing && sync == MIRROR_SYNC_MODE_TOP) {
3757         sync = MIRROR_SYNC_MODE_FULL;
3758     }
3759 
3760     if (has_replaces) {
3761         BlockDriverState *to_replace_bs;
3762         AioContext *replace_aio_context;
3763         int64_t bs_size, replace_size;
3764 
3765         bs_size = bdrv_getlength(bs);
3766         if (bs_size < 0) {
3767             error_setg_errno(errp, -bs_size, "Failed to query device's size");
3768             return;
3769         }
3770 
3771         to_replace_bs = check_to_replace_node(bs, replaces, errp);
3772         if (!to_replace_bs) {
3773             return;
3774         }
3775 
3776         replace_aio_context = bdrv_get_aio_context(to_replace_bs);
3777         aio_context_acquire(replace_aio_context);
3778         replace_size = bdrv_getlength(to_replace_bs);
3779         aio_context_release(replace_aio_context);
3780 
3781         if (replace_size < 0) {
3782             error_setg_errno(errp, -replace_size,
3783                              "Failed to query the replacement node's size");
3784             return;
3785         }
3786         if (bs_size != replace_size) {
3787             error_setg(errp, "cannot replace image with a mirror image of "
3788                              "different size");
3789             return;
3790         }
3791     }
3792 
3793     /* pass the node name to replace to mirror start since it's loose coupling
3794      * and will allow to check whether the node still exist at mirror completion
3795      */
3796     mirror_start(job_id, bs, target,
3797                  has_replaces ? replaces : NULL, job_flags,
3798                  speed, granularity, buf_size, sync, backing_mode,
3799                  on_source_error, on_target_error, unmap, filter_node_name,
3800                  copy_mode, errp);
3801 }
3802 
3803 void qmp_drive_mirror(DriveMirror *arg, Error **errp)
3804 {
3805     BlockDriverState *bs;
3806     BlockDriverState *source, *target_bs;
3807     AioContext *aio_context;
3808     BlockMirrorBackingMode backing_mode;
3809     Error *local_err = NULL;
3810     QDict *options = NULL;
3811     int flags;
3812     int64_t size;
3813     const char *format = arg->format;
3814 
3815     bs = qmp_get_root_bs(arg->device, errp);
3816     if (!bs) {
3817         return;
3818     }
3819 
3820     /* Early check to avoid creating target */
3821     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_MIRROR_SOURCE, errp)) {
3822         return;
3823     }
3824 
3825     aio_context = bdrv_get_aio_context(bs);
3826     aio_context_acquire(aio_context);
3827 
3828     if (!arg->has_mode) {
3829         arg->mode = NEW_IMAGE_MODE_ABSOLUTE_PATHS;
3830     }
3831 
3832     if (!arg->has_format) {
3833         format = (arg->mode == NEW_IMAGE_MODE_EXISTING
3834                   ? NULL : bs->drv->format_name);
3835     }
3836 
3837     flags = bs->open_flags | BDRV_O_RDWR;
3838     source = backing_bs(bs);
3839     if (!source && arg->sync == MIRROR_SYNC_MODE_TOP) {
3840         arg->sync = MIRROR_SYNC_MODE_FULL;
3841     }
3842     if (arg->sync == MIRROR_SYNC_MODE_NONE) {
3843         source = bs;
3844     }
3845 
3846     size = bdrv_getlength(bs);
3847     if (size < 0) {
3848         error_setg_errno(errp, -size, "bdrv_getlength failed");
3849         goto out;
3850     }
3851 
3852     if (arg->has_replaces) {
3853         if (!arg->has_node_name) {
3854             error_setg(errp, "a node-name must be provided when replacing a"
3855                              " named node of the graph");
3856             goto out;
3857         }
3858     }
3859 
3860     if (arg->mode == NEW_IMAGE_MODE_ABSOLUTE_PATHS) {
3861         backing_mode = MIRROR_SOURCE_BACKING_CHAIN;
3862     } else {
3863         backing_mode = MIRROR_OPEN_BACKING_CHAIN;
3864     }
3865 
3866     /* Don't open backing image in create() */
3867     flags |= BDRV_O_NO_BACKING;
3868 
3869     if ((arg->sync == MIRROR_SYNC_MODE_FULL || !source)
3870         && arg->mode != NEW_IMAGE_MODE_EXISTING)
3871     {
3872         /* create new image w/o backing file */
3873         assert(format);
3874         bdrv_img_create(arg->target, format,
3875                         NULL, NULL, NULL, size, flags, false, &local_err);
3876     } else {
3877         switch (arg->mode) {
3878         case NEW_IMAGE_MODE_EXISTING:
3879             break;
3880         case NEW_IMAGE_MODE_ABSOLUTE_PATHS:
3881             /* create new image with backing file */
3882             bdrv_refresh_filename(source);
3883             bdrv_img_create(arg->target, format,
3884                             source->filename,
3885                             source->drv->format_name,
3886                             NULL, size, flags, false, &local_err);
3887             break;
3888         default:
3889             abort();
3890         }
3891     }
3892 
3893     if (local_err) {
3894         error_propagate(errp, local_err);
3895         goto out;
3896     }
3897 
3898     options = qdict_new();
3899     if (arg->has_node_name) {
3900         qdict_put_str(options, "node-name", arg->node_name);
3901     }
3902     if (format) {
3903         qdict_put_str(options, "driver", format);
3904     }
3905 
3906     /* Mirroring takes care of copy-on-write using the source's backing
3907      * file.
3908      */
3909     target_bs = bdrv_open(arg->target, NULL, options, flags, errp);
3910     if (!target_bs) {
3911         goto out;
3912     }
3913 
3914     bdrv_set_aio_context(target_bs, aio_context);
3915 
3916     blockdev_mirror_common(arg->has_job_id ? arg->job_id : NULL, bs, target_bs,
3917                            arg->has_replaces, arg->replaces, arg->sync,
3918                            backing_mode, arg->has_speed, arg->speed,
3919                            arg->has_granularity, arg->granularity,
3920                            arg->has_buf_size, arg->buf_size,
3921                            arg->has_on_source_error, arg->on_source_error,
3922                            arg->has_on_target_error, arg->on_target_error,
3923                            arg->has_unmap, arg->unmap,
3924                            false, NULL,
3925                            arg->has_copy_mode, arg->copy_mode,
3926                            arg->has_auto_finalize, arg->auto_finalize,
3927                            arg->has_auto_dismiss, arg->auto_dismiss,
3928                            &local_err);
3929     bdrv_unref(target_bs);
3930     error_propagate(errp, local_err);
3931 out:
3932     aio_context_release(aio_context);
3933 }
3934 
3935 void qmp_blockdev_mirror(bool has_job_id, const char *job_id,
3936                          const char *device, const char *target,
3937                          bool has_replaces, const char *replaces,
3938                          MirrorSyncMode sync,
3939                          bool has_speed, int64_t speed,
3940                          bool has_granularity, uint32_t granularity,
3941                          bool has_buf_size, int64_t buf_size,
3942                          bool has_on_source_error,
3943                          BlockdevOnError on_source_error,
3944                          bool has_on_target_error,
3945                          BlockdevOnError on_target_error,
3946                          bool has_filter_node_name,
3947                          const char *filter_node_name,
3948                          bool has_copy_mode, MirrorCopyMode copy_mode,
3949                          bool has_auto_finalize, bool auto_finalize,
3950                          bool has_auto_dismiss, bool auto_dismiss,
3951                          Error **errp)
3952 {
3953     BlockDriverState *bs;
3954     BlockDriverState *target_bs;
3955     AioContext *aio_context;
3956     BlockMirrorBackingMode backing_mode = MIRROR_LEAVE_BACKING_CHAIN;
3957     Error *local_err = NULL;
3958 
3959     bs = qmp_get_root_bs(device, errp);
3960     if (!bs) {
3961         return;
3962     }
3963 
3964     target_bs = bdrv_lookup_bs(target, target, errp);
3965     if (!target_bs) {
3966         return;
3967     }
3968 
3969     aio_context = bdrv_get_aio_context(bs);
3970     aio_context_acquire(aio_context);
3971 
3972     bdrv_set_aio_context(target_bs, aio_context);
3973 
3974     blockdev_mirror_common(has_job_id ? job_id : NULL, bs, target_bs,
3975                            has_replaces, replaces, sync, backing_mode,
3976                            has_speed, speed,
3977                            has_granularity, granularity,
3978                            has_buf_size, buf_size,
3979                            has_on_source_error, on_source_error,
3980                            has_on_target_error, on_target_error,
3981                            true, true,
3982                            has_filter_node_name, filter_node_name,
3983                            has_copy_mode, copy_mode,
3984                            has_auto_finalize, auto_finalize,
3985                            has_auto_dismiss, auto_dismiss,
3986                            &local_err);
3987     error_propagate(errp, local_err);
3988 
3989     aio_context_release(aio_context);
3990 }
3991 
3992 /* Get a block job using its ID and acquire its AioContext */
3993 static BlockJob *find_block_job(const char *id, AioContext **aio_context,
3994                                 Error **errp)
3995 {
3996     BlockJob *job;
3997 
3998     assert(id != NULL);
3999 
4000     *aio_context = NULL;
4001 
4002     job = block_job_get(id);
4003 
4004     if (!job) {
4005         error_set(errp, ERROR_CLASS_DEVICE_NOT_ACTIVE,
4006                   "Block job '%s' not found", id);
4007         return NULL;
4008     }
4009 
4010     *aio_context = blk_get_aio_context(job->blk);
4011     aio_context_acquire(*aio_context);
4012 
4013     return job;
4014 }
4015 
4016 void qmp_block_job_set_speed(const char *device, int64_t speed, Error **errp)
4017 {
4018     AioContext *aio_context;
4019     BlockJob *job = find_block_job(device, &aio_context, errp);
4020 
4021     if (!job) {
4022         return;
4023     }
4024 
4025     block_job_set_speed(job, speed, errp);
4026     aio_context_release(aio_context);
4027 }
4028 
4029 void qmp_block_job_cancel(const char *device,
4030                           bool has_force, bool force, Error **errp)
4031 {
4032     AioContext *aio_context;
4033     BlockJob *job = find_block_job(device, &aio_context, errp);
4034 
4035     if (!job) {
4036         return;
4037     }
4038 
4039     if (!has_force) {
4040         force = false;
4041     }
4042 
4043     if (job_user_paused(&job->job) && !force) {
4044         error_setg(errp, "The block job for device '%s' is currently paused",
4045                    device);
4046         goto out;
4047     }
4048 
4049     trace_qmp_block_job_cancel(job);
4050     job_user_cancel(&job->job, force, errp);
4051 out:
4052     aio_context_release(aio_context);
4053 }
4054 
4055 void qmp_block_job_pause(const char *device, Error **errp)
4056 {
4057     AioContext *aio_context;
4058     BlockJob *job = find_block_job(device, &aio_context, errp);
4059 
4060     if (!job) {
4061         return;
4062     }
4063 
4064     trace_qmp_block_job_pause(job);
4065     job_user_pause(&job->job, errp);
4066     aio_context_release(aio_context);
4067 }
4068 
4069 void qmp_block_job_resume(const char *device, Error **errp)
4070 {
4071     AioContext *aio_context;
4072     BlockJob *job = find_block_job(device, &aio_context, errp);
4073 
4074     if (!job) {
4075         return;
4076     }
4077 
4078     trace_qmp_block_job_resume(job);
4079     job_user_resume(&job->job, errp);
4080     aio_context_release(aio_context);
4081 }
4082 
4083 void qmp_block_job_complete(const char *device, Error **errp)
4084 {
4085     AioContext *aio_context;
4086     BlockJob *job = find_block_job(device, &aio_context, errp);
4087 
4088     if (!job) {
4089         return;
4090     }
4091 
4092     trace_qmp_block_job_complete(job);
4093     job_complete(&job->job, errp);
4094     aio_context_release(aio_context);
4095 }
4096 
4097 void qmp_block_job_finalize(const char *id, Error **errp)
4098 {
4099     AioContext *aio_context;
4100     BlockJob *job = find_block_job(id, &aio_context, errp);
4101 
4102     if (!job) {
4103         return;
4104     }
4105 
4106     trace_qmp_block_job_finalize(job);
4107     job_finalize(&job->job, errp);
4108     aio_context_release(aio_context);
4109 }
4110 
4111 void qmp_block_job_dismiss(const char *id, Error **errp)
4112 {
4113     AioContext *aio_context;
4114     BlockJob *bjob = find_block_job(id, &aio_context, errp);
4115     Job *job;
4116 
4117     if (!bjob) {
4118         return;
4119     }
4120 
4121     trace_qmp_block_job_dismiss(bjob);
4122     job = &bjob->job;
4123     job_dismiss(&job, errp);
4124     aio_context_release(aio_context);
4125 }
4126 
4127 void qmp_change_backing_file(const char *device,
4128                              const char *image_node_name,
4129                              const char *backing_file,
4130                              Error **errp)
4131 {
4132     BlockDriverState *bs = NULL;
4133     AioContext *aio_context;
4134     BlockDriverState *image_bs = NULL;
4135     Error *local_err = NULL;
4136     bool ro;
4137     int ret;
4138 
4139     bs = qmp_get_root_bs(device, errp);
4140     if (!bs) {
4141         return;
4142     }
4143 
4144     aio_context = bdrv_get_aio_context(bs);
4145     aio_context_acquire(aio_context);
4146 
4147     image_bs = bdrv_lookup_bs(NULL, image_node_name, &local_err);
4148     if (local_err) {
4149         error_propagate(errp, local_err);
4150         goto out;
4151     }
4152 
4153     if (!image_bs) {
4154         error_setg(errp, "image file not found");
4155         goto out;
4156     }
4157 
4158     if (bdrv_find_base(image_bs) == image_bs) {
4159         error_setg(errp, "not allowing backing file change on an image "
4160                          "without a backing file");
4161         goto out;
4162     }
4163 
4164     /* even though we are not necessarily operating on bs, we need it to
4165      * determine if block ops are currently prohibited on the chain */
4166     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_CHANGE, errp)) {
4167         goto out;
4168     }
4169 
4170     /* final sanity check */
4171     if (!bdrv_chain_contains(bs, image_bs)) {
4172         error_setg(errp, "'%s' and image file are not in the same chain",
4173                    device);
4174         goto out;
4175     }
4176 
4177     /* if not r/w, reopen to make r/w */
4178     ro = bdrv_is_read_only(image_bs);
4179 
4180     if (ro) {
4181         if (bdrv_reopen_set_read_only(image_bs, false, errp) != 0) {
4182             goto out;
4183         }
4184     }
4185 
4186     ret = bdrv_change_backing_file(image_bs, backing_file,
4187                                image_bs->drv ? image_bs->drv->format_name : "");
4188 
4189     if (ret < 0) {
4190         error_setg_errno(errp, -ret, "Could not change backing file to '%s'",
4191                          backing_file);
4192         /* don't exit here, so we can try to restore open flags if
4193          * appropriate */
4194     }
4195 
4196     if (ro) {
4197         bdrv_reopen_set_read_only(image_bs, true, &local_err);
4198         error_propagate(errp, local_err);
4199     }
4200 
4201 out:
4202     aio_context_release(aio_context);
4203 }
4204 
4205 void hmp_drive_add_node(Monitor *mon, const char *optstr)
4206 {
4207     QemuOpts *opts;
4208     QDict *qdict;
4209     Error *local_err = NULL;
4210 
4211     opts = qemu_opts_parse_noisily(&qemu_drive_opts, optstr, false);
4212     if (!opts) {
4213         return;
4214     }
4215 
4216     qdict = qemu_opts_to_qdict(opts, NULL);
4217 
4218     if (!qdict_get_try_str(qdict, "node-name")) {
4219         qobject_unref(qdict);
4220         error_report("'node-name' needs to be specified");
4221         goto out;
4222     }
4223 
4224     BlockDriverState *bs = bds_tree_init(qdict, &local_err);
4225     if (!bs) {
4226         error_report_err(local_err);
4227         goto out;
4228     }
4229 
4230     QTAILQ_INSERT_TAIL(&monitor_bdrv_states, bs, monitor_list);
4231 
4232 out:
4233     qemu_opts_del(opts);
4234 }
4235 
4236 void qmp_blockdev_add(BlockdevOptions *options, Error **errp)
4237 {
4238     BlockDriverState *bs;
4239     QObject *obj;
4240     Visitor *v = qobject_output_visitor_new(&obj);
4241     QDict *qdict;
4242     Error *local_err = NULL;
4243 
4244     visit_type_BlockdevOptions(v, NULL, &options, &local_err);
4245     if (local_err) {
4246         error_propagate(errp, local_err);
4247         goto fail;
4248     }
4249 
4250     visit_complete(v, &obj);
4251     qdict = qobject_to(QDict, obj);
4252 
4253     qdict_flatten(qdict);
4254 
4255     if (!qdict_get_try_str(qdict, "node-name")) {
4256         error_setg(errp, "'node-name' must be specified for the root node");
4257         goto fail;
4258     }
4259 
4260     bs = bds_tree_init(qdict, errp);
4261     if (!bs) {
4262         goto fail;
4263     }
4264 
4265     QTAILQ_INSERT_TAIL(&monitor_bdrv_states, bs, monitor_list);
4266 
4267 fail:
4268     visit_free(v);
4269 }
4270 
4271 void qmp_x_blockdev_reopen(BlockdevOptions *options, Error **errp)
4272 {
4273     BlockDriverState *bs;
4274     AioContext *ctx;
4275     QObject *obj;
4276     Visitor *v = qobject_output_visitor_new(&obj);
4277     Error *local_err = NULL;
4278     BlockReopenQueue *queue;
4279     QDict *qdict;
4280 
4281     /* Check for the selected node name */
4282     if (!options->has_node_name) {
4283         error_setg(errp, "Node name not specified");
4284         goto fail;
4285     }
4286 
4287     bs = bdrv_find_node(options->node_name);
4288     if (!bs) {
4289         error_setg(errp, "Cannot find node named '%s'", options->node_name);
4290         goto fail;
4291     }
4292 
4293     /* Put all options in a QDict and flatten it */
4294     visit_type_BlockdevOptions(v, NULL, &options, &local_err);
4295     if (local_err) {
4296         error_propagate(errp, local_err);
4297         goto fail;
4298     }
4299 
4300     visit_complete(v, &obj);
4301     qdict = qobject_to(QDict, obj);
4302 
4303     qdict_flatten(qdict);
4304 
4305     /* Perform the reopen operation */
4306     ctx = bdrv_get_aio_context(bs);
4307     aio_context_acquire(ctx);
4308     bdrv_subtree_drained_begin(bs);
4309     queue = bdrv_reopen_queue(NULL, bs, qdict, false);
4310     bdrv_reopen_multiple(queue, errp);
4311     bdrv_subtree_drained_end(bs);
4312     aio_context_release(ctx);
4313 
4314 fail:
4315     visit_free(v);
4316 }
4317 
4318 void qmp_blockdev_del(const char *node_name, Error **errp)
4319 {
4320     AioContext *aio_context;
4321     BlockDriverState *bs;
4322 
4323     bs = bdrv_find_node(node_name);
4324     if (!bs) {
4325         error_setg(errp, "Cannot find node %s", node_name);
4326         return;
4327     }
4328     if (bdrv_has_blk(bs)) {
4329         error_setg(errp, "Node %s is in use", node_name);
4330         return;
4331     }
4332     aio_context = bdrv_get_aio_context(bs);
4333     aio_context_acquire(aio_context);
4334 
4335     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_DRIVE_DEL, errp)) {
4336         goto out;
4337     }
4338 
4339     if (!QTAILQ_IN_USE(bs, monitor_list)) {
4340         error_setg(errp, "Node %s is not owned by the monitor",
4341                    bs->node_name);
4342         goto out;
4343     }
4344 
4345     if (bs->refcnt > 1) {
4346         error_setg(errp, "Block device %s is in use",
4347                    bdrv_get_device_or_node_name(bs));
4348         goto out;
4349     }
4350 
4351     QTAILQ_REMOVE(&monitor_bdrv_states, bs, monitor_list);
4352     bdrv_unref(bs);
4353 
4354 out:
4355     aio_context_release(aio_context);
4356 }
4357 
4358 static BdrvChild *bdrv_find_child(BlockDriverState *parent_bs,
4359                                   const char *child_name)
4360 {
4361     BdrvChild *child;
4362 
4363     QLIST_FOREACH(child, &parent_bs->children, next) {
4364         if (strcmp(child->name, child_name) == 0) {
4365             return child;
4366         }
4367     }
4368 
4369     return NULL;
4370 }
4371 
4372 void qmp_x_blockdev_change(const char *parent, bool has_child,
4373                            const char *child, bool has_node,
4374                            const char *node, Error **errp)
4375 {
4376     BlockDriverState *parent_bs, *new_bs = NULL;
4377     BdrvChild *p_child;
4378 
4379     parent_bs = bdrv_lookup_bs(parent, parent, errp);
4380     if (!parent_bs) {
4381         return;
4382     }
4383 
4384     if (has_child == has_node) {
4385         if (has_child) {
4386             error_setg(errp, "The parameters child and node are in conflict");
4387         } else {
4388             error_setg(errp, "Either child or node must be specified");
4389         }
4390         return;
4391     }
4392 
4393     if (has_child) {
4394         p_child = bdrv_find_child(parent_bs, child);
4395         if (!p_child) {
4396             error_setg(errp, "Node '%s' does not have child '%s'",
4397                        parent, child);
4398             return;
4399         }
4400         bdrv_del_child(parent_bs, p_child, errp);
4401     }
4402 
4403     if (has_node) {
4404         new_bs = bdrv_find_node(node);
4405         if (!new_bs) {
4406             error_setg(errp, "Node '%s' not found", node);
4407             return;
4408         }
4409         bdrv_add_child(parent_bs, new_bs, errp);
4410     }
4411 }
4412 
4413 BlockJobInfoList *qmp_query_block_jobs(Error **errp)
4414 {
4415     BlockJobInfoList *head = NULL, **p_next = &head;
4416     BlockJob *job;
4417 
4418     for (job = block_job_next(NULL); job; job = block_job_next(job)) {
4419         BlockJobInfoList *elem;
4420         AioContext *aio_context;
4421 
4422         if (block_job_is_internal(job)) {
4423             continue;
4424         }
4425         elem = g_new0(BlockJobInfoList, 1);
4426         aio_context = blk_get_aio_context(job->blk);
4427         aio_context_acquire(aio_context);
4428         elem->value = block_job_query(job, errp);
4429         aio_context_release(aio_context);
4430         if (!elem->value) {
4431             g_free(elem);
4432             qapi_free_BlockJobInfoList(head);
4433             return NULL;
4434         }
4435         *p_next = elem;
4436         p_next = &elem->next;
4437     }
4438 
4439     return head;
4440 }
4441 
4442 void qmp_x_blockdev_set_iothread(const char *node_name, StrOrNull *iothread,
4443                                  bool has_force, bool force, Error **errp)
4444 {
4445     AioContext *old_context;
4446     AioContext *new_context;
4447     BlockDriverState *bs;
4448 
4449     bs = bdrv_find_node(node_name);
4450     if (!bs) {
4451         error_setg(errp, "Cannot find node %s", node_name);
4452         return;
4453     }
4454 
4455     /* Protects against accidents. */
4456     if (!(has_force && force) && bdrv_has_blk(bs)) {
4457         error_setg(errp, "Node %s is associated with a BlockBackend and could "
4458                          "be in use (use force=true to override this check)",
4459                          node_name);
4460         return;
4461     }
4462 
4463     if (iothread->type == QTYPE_QSTRING) {
4464         IOThread *obj = iothread_by_id(iothread->u.s);
4465         if (!obj) {
4466             error_setg(errp, "Cannot find iothread %s", iothread->u.s);
4467             return;
4468         }
4469 
4470         new_context = iothread_get_aio_context(obj);
4471     } else {
4472         new_context = qemu_get_aio_context();
4473     }
4474 
4475     old_context = bdrv_get_aio_context(bs);
4476     aio_context_acquire(old_context);
4477 
4478     bdrv_set_aio_context(bs, new_context);
4479 
4480     aio_context_release(old_context);
4481 }
4482 
4483 void qmp_block_latency_histogram_set(
4484     const char *id,
4485     bool has_boundaries, uint64List *boundaries,
4486     bool has_boundaries_read, uint64List *boundaries_read,
4487     bool has_boundaries_write, uint64List *boundaries_write,
4488     bool has_boundaries_flush, uint64List *boundaries_flush,
4489     Error **errp)
4490 {
4491     BlockBackend *blk = qmp_get_blk(NULL, id, errp);
4492     BlockAcctStats *stats;
4493     int ret;
4494 
4495     if (!blk) {
4496         return;
4497     }
4498 
4499     stats = blk_get_stats(blk);
4500 
4501     if (!has_boundaries && !has_boundaries_read && !has_boundaries_write &&
4502         !has_boundaries_flush)
4503     {
4504         block_latency_histograms_clear(stats);
4505         return;
4506     }
4507 
4508     if (has_boundaries || has_boundaries_read) {
4509         ret = block_latency_histogram_set(
4510             stats, BLOCK_ACCT_READ,
4511             has_boundaries_read ? boundaries_read : boundaries);
4512         if (ret) {
4513             error_setg(errp, "Device '%s' set read boundaries fail", id);
4514             return;
4515         }
4516     }
4517 
4518     if (has_boundaries || has_boundaries_write) {
4519         ret = block_latency_histogram_set(
4520             stats, BLOCK_ACCT_WRITE,
4521             has_boundaries_write ? boundaries_write : boundaries);
4522         if (ret) {
4523             error_setg(errp, "Device '%s' set write boundaries fail", id);
4524             return;
4525         }
4526     }
4527 
4528     if (has_boundaries || has_boundaries_flush) {
4529         ret = block_latency_histogram_set(
4530             stats, BLOCK_ACCT_FLUSH,
4531             has_boundaries_flush ? boundaries_flush : boundaries);
4532         if (ret) {
4533             error_setg(errp, "Device '%s' set flush boundaries fail", id);
4534             return;
4535         }
4536     }
4537 }
4538 
4539 QemuOptsList qemu_common_drive_opts = {
4540     .name = "drive",
4541     .head = QTAILQ_HEAD_INITIALIZER(qemu_common_drive_opts.head),
4542     .desc = {
4543         {
4544             .name = "snapshot",
4545             .type = QEMU_OPT_BOOL,
4546             .help = "enable/disable snapshot mode",
4547         },{
4548             .name = "aio",
4549             .type = QEMU_OPT_STRING,
4550             .help = "host AIO implementation (threads, native)",
4551         },{
4552             .name = BDRV_OPT_CACHE_WB,
4553             .type = QEMU_OPT_BOOL,
4554             .help = "Enable writeback mode",
4555         },{
4556             .name = "format",
4557             .type = QEMU_OPT_STRING,
4558             .help = "disk format (raw, qcow2, ...)",
4559         },{
4560             .name = "rerror",
4561             .type = QEMU_OPT_STRING,
4562             .help = "read error action",
4563         },{
4564             .name = "werror",
4565             .type = QEMU_OPT_STRING,
4566             .help = "write error action",
4567         },{
4568             .name = BDRV_OPT_READ_ONLY,
4569             .type = QEMU_OPT_BOOL,
4570             .help = "open drive file as read-only",
4571         },
4572 
4573         THROTTLE_OPTS,
4574 
4575         {
4576             .name = "throttling.group",
4577             .type = QEMU_OPT_STRING,
4578             .help = "name of the block throttling group",
4579         },{
4580             .name = "copy-on-read",
4581             .type = QEMU_OPT_BOOL,
4582             .help = "copy read data from backing file into image file",
4583         },{
4584             .name = "detect-zeroes",
4585             .type = QEMU_OPT_STRING,
4586             .help = "try to optimize zero writes (off, on, unmap)",
4587         },{
4588             .name = "stats-account-invalid",
4589             .type = QEMU_OPT_BOOL,
4590             .help = "whether to account for invalid I/O operations "
4591                     "in the statistics",
4592         },{
4593             .name = "stats-account-failed",
4594             .type = QEMU_OPT_BOOL,
4595             .help = "whether to account for failed I/O operations "
4596                     "in the statistics",
4597         },
4598         { /* end of list */ }
4599     },
4600 };
4601 
4602 QemuOptsList qemu_drive_opts = {
4603     .name = "drive",
4604     .head = QTAILQ_HEAD_INITIALIZER(qemu_drive_opts.head),
4605     .desc = {
4606         /*
4607          * no elements => accept any params
4608          * validation will happen later
4609          */
4610         { /* end of list */ }
4611     },
4612 };
4613