xref: /openbmc/qemu/blockdev.c (revision ed5abf46)
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 "sysemu/runstate.h"
62 #include "qemu/cutils.h"
63 #include "qemu/help_option.h"
64 #include "qemu/main-loop.h"
65 #include "qemu/throttle-options.h"
66 
67 QTAILQ_HEAD(, BlockDriverState) monitor_bdrv_states =
68     QTAILQ_HEAD_INITIALIZER(monitor_bdrv_states);
69 
70 void bdrv_set_monitor_owned(BlockDriverState *bs)
71 {
72     QTAILQ_INSERT_TAIL(&monitor_bdrv_states, bs, monitor_list);
73 }
74 
75 static const char *const if_name[IF_COUNT] = {
76     [IF_NONE] = "none",
77     [IF_IDE] = "ide",
78     [IF_SCSI] = "scsi",
79     [IF_FLOPPY] = "floppy",
80     [IF_PFLASH] = "pflash",
81     [IF_MTD] = "mtd",
82     [IF_SD] = "sd",
83     [IF_VIRTIO] = "virtio",
84     [IF_XEN] = "xen",
85 };
86 
87 static int if_max_devs[IF_COUNT] = {
88     /*
89      * Do not change these numbers!  They govern how drive option
90      * index maps to unit and bus.  That mapping is ABI.
91      *
92      * All controllers used to implement if=T drives need to support
93      * if_max_devs[T] units, for any T with if_max_devs[T] != 0.
94      * Otherwise, some index values map to "impossible" bus, unit
95      * values.
96      *
97      * For instance, if you change [IF_SCSI] to 255, -drive
98      * if=scsi,index=12 no longer means bus=1,unit=5, but
99      * bus=0,unit=12.  With an lsi53c895a controller (7 units max),
100      * the drive can't be set up.  Regression.
101      */
102     [IF_IDE] = 2,
103     [IF_SCSI] = 7,
104 };
105 
106 /**
107  * Boards may call this to offer board-by-board overrides
108  * of the default, global values.
109  */
110 void override_max_devs(BlockInterfaceType type, int max_devs)
111 {
112     BlockBackend *blk;
113     DriveInfo *dinfo;
114 
115     if (max_devs <= 0) {
116         return;
117     }
118 
119     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
120         dinfo = blk_legacy_dinfo(blk);
121         if (dinfo->type == type) {
122             fprintf(stderr, "Cannot override units-per-bus property of"
123                     " the %s interface, because a drive of that type has"
124                     " already been added.\n", if_name[type]);
125             g_assert_not_reached();
126         }
127     }
128 
129     if_max_devs[type] = max_devs;
130 }
131 
132 /*
133  * We automatically delete the drive when a device using it gets
134  * unplugged.  Questionable feature, but we can't just drop it.
135  * Device models call blockdev_mark_auto_del() to schedule the
136  * automatic deletion, and generic qdev code calls blockdev_auto_del()
137  * when deletion is actually safe.
138  */
139 void blockdev_mark_auto_del(BlockBackend *blk)
140 {
141     DriveInfo *dinfo = blk_legacy_dinfo(blk);
142     BlockJob *job;
143 
144     if (!dinfo) {
145         return;
146     }
147 
148     for (job = block_job_next(NULL); job; job = block_job_next(job)) {
149         if (block_job_has_bdrv(job, blk_bs(blk))) {
150             AioContext *aio_context = job->job.aio_context;
151             aio_context_acquire(aio_context);
152 
153             job_cancel(&job->job, false);
154 
155             aio_context_release(aio_context);
156         }
157     }
158 
159     dinfo->auto_del = 1;
160 }
161 
162 void blockdev_auto_del(BlockBackend *blk)
163 {
164     DriveInfo *dinfo = blk_legacy_dinfo(blk);
165 
166     if (dinfo && dinfo->auto_del) {
167         monitor_remove_blk(blk);
168         blk_unref(blk);
169     }
170 }
171 
172 /**
173  * Returns the current mapping of how many units per bus
174  * a particular interface can support.
175  *
176  *  A positive integer indicates n units per bus.
177  *  0 implies the mapping has not been established.
178  * -1 indicates an invalid BlockInterfaceType was given.
179  */
180 int drive_get_max_devs(BlockInterfaceType type)
181 {
182     if (type >= IF_IDE && type < IF_COUNT) {
183         return if_max_devs[type];
184     }
185 
186     return -1;
187 }
188 
189 static int drive_index_to_bus_id(BlockInterfaceType type, int index)
190 {
191     int max_devs = if_max_devs[type];
192     return max_devs ? index / max_devs : 0;
193 }
194 
195 static int drive_index_to_unit_id(BlockInterfaceType type, int index)
196 {
197     int max_devs = if_max_devs[type];
198     return max_devs ? index % max_devs : index;
199 }
200 
201 QemuOpts *drive_def(const char *optstr)
202 {
203     return qemu_opts_parse_noisily(qemu_find_opts("drive"), optstr, false);
204 }
205 
206 QemuOpts *drive_add(BlockInterfaceType type, int index, const char *file,
207                     const char *optstr)
208 {
209     QemuOpts *opts;
210 
211     opts = drive_def(optstr);
212     if (!opts) {
213         return NULL;
214     }
215     if (type != IF_DEFAULT) {
216         qemu_opt_set(opts, "if", if_name[type], &error_abort);
217     }
218     if (index >= 0) {
219         qemu_opt_set_number(opts, "index", index, &error_abort);
220     }
221     if (file)
222         qemu_opt_set(opts, "file", file, &error_abort);
223     return opts;
224 }
225 
226 DriveInfo *drive_get(BlockInterfaceType type, int bus, int unit)
227 {
228     BlockBackend *blk;
229     DriveInfo *dinfo;
230 
231     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
232         dinfo = blk_legacy_dinfo(blk);
233         if (dinfo && dinfo->type == type
234             && dinfo->bus == bus && dinfo->unit == unit) {
235             return dinfo;
236         }
237     }
238 
239     return NULL;
240 }
241 
242 void drive_check_orphaned(void)
243 {
244     BlockBackend *blk;
245     DriveInfo *dinfo;
246     Location loc;
247     bool orphans = false;
248 
249     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
250         dinfo = blk_legacy_dinfo(blk);
251         if (!blk_get_attached_dev(blk) && !dinfo->is_default &&
252             dinfo->type != IF_NONE) {
253             loc_push_none(&loc);
254             qemu_opts_loc_restore(dinfo->opts);
255             error_report("machine type does not support"
256                          " if=%s,bus=%d,unit=%d",
257                          if_name[dinfo->type], dinfo->bus, dinfo->unit);
258             loc_pop(&loc);
259             orphans = true;
260         }
261     }
262 
263     if (orphans) {
264         exit(1);
265     }
266 }
267 
268 DriveInfo *drive_get_by_index(BlockInterfaceType type, int index)
269 {
270     return drive_get(type,
271                      drive_index_to_bus_id(type, index),
272                      drive_index_to_unit_id(type, index));
273 }
274 
275 int drive_get_max_bus(BlockInterfaceType type)
276 {
277     int max_bus;
278     BlockBackend *blk;
279     DriveInfo *dinfo;
280 
281     max_bus = -1;
282     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
283         dinfo = blk_legacy_dinfo(blk);
284         if (dinfo && dinfo->type == type && dinfo->bus > max_bus) {
285             max_bus = dinfo->bus;
286         }
287     }
288     return max_bus;
289 }
290 
291 /* Get a block device.  This should only be used for single-drive devices
292    (e.g. SD/Floppy/MTD).  Multi-disk devices (scsi/ide) should use the
293    appropriate bus.  */
294 DriveInfo *drive_get_next(BlockInterfaceType type)
295 {
296     static int next_block_unit[IF_COUNT];
297 
298     return drive_get(type, 0, next_block_unit[type]++);
299 }
300 
301 static void bdrv_format_print(void *opaque, const char *name)
302 {
303     qemu_printf(" %s", name);
304 }
305 
306 typedef struct {
307     QEMUBH *bh;
308     BlockDriverState *bs;
309 } BDRVPutRefBH;
310 
311 static int parse_block_error_action(const char *buf, bool is_read, Error **errp)
312 {
313     if (!strcmp(buf, "ignore")) {
314         return BLOCKDEV_ON_ERROR_IGNORE;
315     } else if (!is_read && !strcmp(buf, "enospc")) {
316         return BLOCKDEV_ON_ERROR_ENOSPC;
317     } else if (!strcmp(buf, "stop")) {
318         return BLOCKDEV_ON_ERROR_STOP;
319     } else if (!strcmp(buf, "report")) {
320         return BLOCKDEV_ON_ERROR_REPORT;
321     } else {
322         error_setg(errp, "'%s' invalid %s error action",
323                    buf, is_read ? "read" : "write");
324         return -1;
325     }
326 }
327 
328 static bool parse_stats_intervals(BlockAcctStats *stats, QList *intervals,
329                                   Error **errp)
330 {
331     const QListEntry *entry;
332     for (entry = qlist_first(intervals); entry; entry = qlist_next(entry)) {
333         switch (qobject_type(entry->value)) {
334 
335         case QTYPE_QSTRING: {
336             unsigned long long length;
337             const char *str = qstring_get_str(qobject_to(QString,
338                                                          entry->value));
339             if (parse_uint_full(str, &length, 10) == 0 &&
340                 length > 0 && length <= UINT_MAX) {
341                 block_acct_add_interval(stats, (unsigned) length);
342             } else {
343                 error_setg(errp, "Invalid interval length: %s", str);
344                 return false;
345             }
346             break;
347         }
348 
349         case QTYPE_QNUM: {
350             int64_t length = qnum_get_int(qobject_to(QNum, entry->value));
351 
352             if (length > 0 && length <= UINT_MAX) {
353                 block_acct_add_interval(stats, (unsigned) length);
354             } else {
355                 error_setg(errp, "Invalid interval length: %" PRId64, length);
356                 return false;
357             }
358             break;
359         }
360 
361         default:
362             error_setg(errp, "The specification of stats-intervals is invalid");
363             return false;
364         }
365     }
366     return true;
367 }
368 
369 typedef enum { MEDIA_DISK, MEDIA_CDROM } DriveMediaType;
370 
371 /* All parameters but @opts are optional and may be set to NULL. */
372 static void extract_common_blockdev_options(QemuOpts *opts, int *bdrv_flags,
373     const char **throttling_group, ThrottleConfig *throttle_cfg,
374     BlockdevDetectZeroesOptions *detect_zeroes, Error **errp)
375 {
376     Error *local_error = NULL;
377     const char *aio;
378 
379     if (bdrv_flags) {
380         if (qemu_opt_get_bool(opts, "copy-on-read", false)) {
381             *bdrv_flags |= BDRV_O_COPY_ON_READ;
382         }
383 
384         if ((aio = qemu_opt_get(opts, "aio")) != NULL) {
385             if (bdrv_parse_aio(aio, bdrv_flags) < 0) {
386                 error_setg(errp, "invalid aio option");
387                 return;
388             }
389         }
390     }
391 
392     /* disk I/O throttling */
393     if (throttling_group) {
394         *throttling_group = qemu_opt_get(opts, "throttling.group");
395     }
396 
397     if (throttle_cfg) {
398         throttle_config_init(throttle_cfg);
399         throttle_cfg->buckets[THROTTLE_BPS_TOTAL].avg =
400             qemu_opt_get_number(opts, "throttling.bps-total", 0);
401         throttle_cfg->buckets[THROTTLE_BPS_READ].avg  =
402             qemu_opt_get_number(opts, "throttling.bps-read", 0);
403         throttle_cfg->buckets[THROTTLE_BPS_WRITE].avg =
404             qemu_opt_get_number(opts, "throttling.bps-write", 0);
405         throttle_cfg->buckets[THROTTLE_OPS_TOTAL].avg =
406             qemu_opt_get_number(opts, "throttling.iops-total", 0);
407         throttle_cfg->buckets[THROTTLE_OPS_READ].avg =
408             qemu_opt_get_number(opts, "throttling.iops-read", 0);
409         throttle_cfg->buckets[THROTTLE_OPS_WRITE].avg =
410             qemu_opt_get_number(opts, "throttling.iops-write", 0);
411 
412         throttle_cfg->buckets[THROTTLE_BPS_TOTAL].max =
413             qemu_opt_get_number(opts, "throttling.bps-total-max", 0);
414         throttle_cfg->buckets[THROTTLE_BPS_READ].max  =
415             qemu_opt_get_number(opts, "throttling.bps-read-max", 0);
416         throttle_cfg->buckets[THROTTLE_BPS_WRITE].max =
417             qemu_opt_get_number(opts, "throttling.bps-write-max", 0);
418         throttle_cfg->buckets[THROTTLE_OPS_TOTAL].max =
419             qemu_opt_get_number(opts, "throttling.iops-total-max", 0);
420         throttle_cfg->buckets[THROTTLE_OPS_READ].max =
421             qemu_opt_get_number(opts, "throttling.iops-read-max", 0);
422         throttle_cfg->buckets[THROTTLE_OPS_WRITE].max =
423             qemu_opt_get_number(opts, "throttling.iops-write-max", 0);
424 
425         throttle_cfg->buckets[THROTTLE_BPS_TOTAL].burst_length =
426             qemu_opt_get_number(opts, "throttling.bps-total-max-length", 1);
427         throttle_cfg->buckets[THROTTLE_BPS_READ].burst_length  =
428             qemu_opt_get_number(opts, "throttling.bps-read-max-length", 1);
429         throttle_cfg->buckets[THROTTLE_BPS_WRITE].burst_length =
430             qemu_opt_get_number(opts, "throttling.bps-write-max-length", 1);
431         throttle_cfg->buckets[THROTTLE_OPS_TOTAL].burst_length =
432             qemu_opt_get_number(opts, "throttling.iops-total-max-length", 1);
433         throttle_cfg->buckets[THROTTLE_OPS_READ].burst_length =
434             qemu_opt_get_number(opts, "throttling.iops-read-max-length", 1);
435         throttle_cfg->buckets[THROTTLE_OPS_WRITE].burst_length =
436             qemu_opt_get_number(opts, "throttling.iops-write-max-length", 1);
437 
438         throttle_cfg->op_size =
439             qemu_opt_get_number(opts, "throttling.iops-size", 0);
440 
441         if (!throttle_is_valid(throttle_cfg, errp)) {
442             return;
443         }
444     }
445 
446     if (detect_zeroes) {
447         *detect_zeroes =
448             qapi_enum_parse(&BlockdevDetectZeroesOptions_lookup,
449                             qemu_opt_get(opts, "detect-zeroes"),
450                             BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF,
451                             &local_error);
452         if (local_error) {
453             error_propagate(errp, local_error);
454             return;
455         }
456     }
457 }
458 
459 /* Takes the ownership of bs_opts */
460 static BlockBackend *blockdev_init(const char *file, QDict *bs_opts,
461                                    Error **errp)
462 {
463     const char *buf;
464     int bdrv_flags = 0;
465     int on_read_error, on_write_error;
466     bool account_invalid, account_failed;
467     bool writethrough, read_only;
468     BlockBackend *blk;
469     BlockDriverState *bs;
470     ThrottleConfig cfg;
471     int snapshot = 0;
472     Error *error = NULL;
473     QemuOpts *opts;
474     QDict *interval_dict = NULL;
475     QList *interval_list = NULL;
476     const char *id;
477     BlockdevDetectZeroesOptions detect_zeroes =
478         BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF;
479     const char *throttling_group = NULL;
480 
481     /* Check common options by copying from bs_opts to opts, all other options
482      * stay in bs_opts for processing by bdrv_open(). */
483     id = qdict_get_try_str(bs_opts, "id");
484     opts = qemu_opts_create(&qemu_common_drive_opts, id, 1, &error);
485     if (error) {
486         error_propagate(errp, error);
487         goto err_no_opts;
488     }
489 
490     qemu_opts_absorb_qdict(opts, bs_opts, &error);
491     if (error) {
492         error_propagate(errp, error);
493         goto early_err;
494     }
495 
496     if (id) {
497         qdict_del(bs_opts, "id");
498     }
499 
500     /* extract parameters */
501     snapshot = qemu_opt_get_bool(opts, "snapshot", 0);
502 
503     account_invalid = qemu_opt_get_bool(opts, "stats-account-invalid", true);
504     account_failed = qemu_opt_get_bool(opts, "stats-account-failed", true);
505 
506     writethrough = !qemu_opt_get_bool(opts, BDRV_OPT_CACHE_WB, true);
507 
508     id = qemu_opts_id(opts);
509 
510     qdict_extract_subqdict(bs_opts, &interval_dict, "stats-intervals.");
511     qdict_array_split(interval_dict, &interval_list);
512 
513     if (qdict_size(interval_dict) != 0) {
514         error_setg(errp, "Invalid option stats-intervals.%s",
515                    qdict_first(interval_dict)->key);
516         goto early_err;
517     }
518 
519     extract_common_blockdev_options(opts, &bdrv_flags, &throttling_group, &cfg,
520                                     &detect_zeroes, &error);
521     if (error) {
522         error_propagate(errp, error);
523         goto early_err;
524     }
525 
526     if ((buf = qemu_opt_get(opts, "format")) != NULL) {
527         if (is_help_option(buf)) {
528             qemu_printf("Supported formats:");
529             bdrv_iterate_format(bdrv_format_print, NULL, false);
530             qemu_printf("\nSupported formats (read-only):");
531             bdrv_iterate_format(bdrv_format_print, NULL, true);
532             qemu_printf("\n");
533             goto early_err;
534         }
535 
536         if (qdict_haskey(bs_opts, "driver")) {
537             error_setg(errp, "Cannot specify both 'driver' and 'format'");
538             goto early_err;
539         }
540         qdict_put_str(bs_opts, "driver", buf);
541     }
542 
543     on_write_error = BLOCKDEV_ON_ERROR_ENOSPC;
544     if ((buf = qemu_opt_get(opts, "werror")) != NULL) {
545         on_write_error = parse_block_error_action(buf, 0, &error);
546         if (error) {
547             error_propagate(errp, error);
548             goto early_err;
549         }
550     }
551 
552     on_read_error = BLOCKDEV_ON_ERROR_REPORT;
553     if ((buf = qemu_opt_get(opts, "rerror")) != NULL) {
554         on_read_error = parse_block_error_action(buf, 1, &error);
555         if (error) {
556             error_propagate(errp, error);
557             goto early_err;
558         }
559     }
560 
561     if (snapshot) {
562         bdrv_flags |= BDRV_O_SNAPSHOT;
563     }
564 
565     read_only = qemu_opt_get_bool(opts, BDRV_OPT_READ_ONLY, false);
566 
567     /* init */
568     if ((!file || !*file) && !qdict_size(bs_opts)) {
569         BlockBackendRootState *blk_rs;
570 
571         blk = blk_new(qemu_get_aio_context(), 0, BLK_PERM_ALL);
572         blk_rs = blk_get_root_state(blk);
573         blk_rs->open_flags    = bdrv_flags;
574         blk_rs->read_only     = read_only;
575         blk_rs->detect_zeroes = detect_zeroes;
576 
577         qobject_unref(bs_opts);
578     } else {
579         if (file && !*file) {
580             file = NULL;
581         }
582 
583         /* bdrv_open() defaults to the values in bdrv_flags (for compatibility
584          * with other callers) rather than what we want as the real defaults.
585          * Apply the defaults here instead. */
586         qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_DIRECT, "off");
587         qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_NO_FLUSH, "off");
588         qdict_set_default_str(bs_opts, BDRV_OPT_READ_ONLY,
589                               read_only ? "on" : "off");
590         qdict_set_default_str(bs_opts, BDRV_OPT_AUTO_READ_ONLY, "on");
591         assert((bdrv_flags & BDRV_O_CACHE_MASK) == 0);
592 
593         if (runstate_check(RUN_STATE_INMIGRATE)) {
594             bdrv_flags |= BDRV_O_INACTIVE;
595         }
596 
597         blk = blk_new_open(file, NULL, bs_opts, bdrv_flags, errp);
598         if (!blk) {
599             goto err_no_bs_opts;
600         }
601         bs = blk_bs(blk);
602 
603         bs->detect_zeroes = detect_zeroes;
604 
605         block_acct_setup(blk_get_stats(blk), account_invalid, account_failed);
606 
607         if (!parse_stats_intervals(blk_get_stats(blk), interval_list, errp)) {
608             blk_unref(blk);
609             blk = NULL;
610             goto err_no_bs_opts;
611         }
612     }
613 
614     /* disk I/O throttling */
615     if (throttle_enabled(&cfg)) {
616         if (!throttling_group) {
617             throttling_group = id;
618         }
619         blk_io_limits_enable(blk, throttling_group);
620         blk_set_io_limits(blk, &cfg);
621     }
622 
623     blk_set_enable_write_cache(blk, !writethrough);
624     blk_set_on_error(blk, on_read_error, on_write_error);
625 
626     if (!monitor_add_blk(blk, id, errp)) {
627         blk_unref(blk);
628         blk = NULL;
629         goto err_no_bs_opts;
630     }
631 
632 err_no_bs_opts:
633     qemu_opts_del(opts);
634     qobject_unref(interval_dict);
635     qobject_unref(interval_list);
636     return blk;
637 
638 early_err:
639     qemu_opts_del(opts);
640     qobject_unref(interval_dict);
641     qobject_unref(interval_list);
642 err_no_opts:
643     qobject_unref(bs_opts);
644     return NULL;
645 }
646 
647 /* Takes the ownership of bs_opts */
648 BlockDriverState *bds_tree_init(QDict *bs_opts, Error **errp)
649 {
650     int bdrv_flags = 0;
651 
652     /* bdrv_open() defaults to the values in bdrv_flags (for compatibility
653      * with other callers) rather than what we want as the real defaults.
654      * Apply the defaults here instead. */
655     qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_DIRECT, "off");
656     qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_NO_FLUSH, "off");
657     qdict_set_default_str(bs_opts, BDRV_OPT_READ_ONLY, "off");
658 
659     if (runstate_check(RUN_STATE_INMIGRATE)) {
660         bdrv_flags |= BDRV_O_INACTIVE;
661     }
662 
663     return bdrv_open(NULL, NULL, bs_opts, bdrv_flags, errp);
664 }
665 
666 void blockdev_close_all_bdrv_states(void)
667 {
668     BlockDriverState *bs, *next_bs;
669 
670     QTAILQ_FOREACH_SAFE(bs, &monitor_bdrv_states, monitor_list, next_bs) {
671         AioContext *ctx = bdrv_get_aio_context(bs);
672 
673         aio_context_acquire(ctx);
674         bdrv_unref(bs);
675         aio_context_release(ctx);
676     }
677 }
678 
679 /* Iterates over the list of monitor-owned BlockDriverStates */
680 BlockDriverState *bdrv_next_monitor_owned(BlockDriverState *bs)
681 {
682     return bs ? QTAILQ_NEXT(bs, monitor_list)
683               : QTAILQ_FIRST(&monitor_bdrv_states);
684 }
685 
686 static void qemu_opt_rename(QemuOpts *opts, const char *from, const char *to,
687                             Error **errp)
688 {
689     const char *value;
690 
691     value = qemu_opt_get(opts, from);
692     if (value) {
693         if (qemu_opt_find(opts, to)) {
694             error_setg(errp, "'%s' and its alias '%s' can't be used at the "
695                        "same time", to, from);
696             return;
697         }
698     }
699 
700     /* rename all items in opts */
701     while ((value = qemu_opt_get(opts, from))) {
702         qemu_opt_set(opts, to, value, &error_abort);
703         qemu_opt_unset(opts, from);
704     }
705 }
706 
707 QemuOptsList qemu_legacy_drive_opts = {
708     .name = "drive",
709     .head = QTAILQ_HEAD_INITIALIZER(qemu_legacy_drive_opts.head),
710     .desc = {
711         {
712             .name = "bus",
713             .type = QEMU_OPT_NUMBER,
714             .help = "bus number",
715         },{
716             .name = "unit",
717             .type = QEMU_OPT_NUMBER,
718             .help = "unit number (i.e. lun for scsi)",
719         },{
720             .name = "index",
721             .type = QEMU_OPT_NUMBER,
722             .help = "index number",
723         },{
724             .name = "media",
725             .type = QEMU_OPT_STRING,
726             .help = "media type (disk, cdrom)",
727         },{
728             .name = "if",
729             .type = QEMU_OPT_STRING,
730             .help = "interface (ide, scsi, sd, mtd, floppy, pflash, virtio)",
731         },{
732             .name = "file",
733             .type = QEMU_OPT_STRING,
734             .help = "file name",
735         },
736 
737         /* Options that are passed on, but have special semantics with -drive */
738         {
739             .name = BDRV_OPT_READ_ONLY,
740             .type = QEMU_OPT_BOOL,
741             .help = "open drive file as read-only",
742         },{
743             .name = "rerror",
744             .type = QEMU_OPT_STRING,
745             .help = "read error action",
746         },{
747             .name = "werror",
748             .type = QEMU_OPT_STRING,
749             .help = "write error action",
750         },{
751             .name = "copy-on-read",
752             .type = QEMU_OPT_BOOL,
753             .help = "copy read data from backing file into image file",
754         },
755 
756         { /* end of list */ }
757     },
758 };
759 
760 DriveInfo *drive_new(QemuOpts *all_opts, BlockInterfaceType block_default_type,
761                      Error **errp)
762 {
763     const char *value;
764     BlockBackend *blk;
765     DriveInfo *dinfo = NULL;
766     QDict *bs_opts;
767     QemuOpts *legacy_opts;
768     DriveMediaType media = MEDIA_DISK;
769     BlockInterfaceType type;
770     int max_devs, bus_id, unit_id, index;
771     const char *werror, *rerror;
772     bool read_only = false;
773     bool copy_on_read;
774     const char *filename;
775     Error *local_err = NULL;
776     int i;
777 
778     /* Change legacy command line options into QMP ones */
779     static const struct {
780         const char *from;
781         const char *to;
782     } opt_renames[] = {
783         { "iops",           "throttling.iops-total" },
784         { "iops_rd",        "throttling.iops-read" },
785         { "iops_wr",        "throttling.iops-write" },
786 
787         { "bps",            "throttling.bps-total" },
788         { "bps_rd",         "throttling.bps-read" },
789         { "bps_wr",         "throttling.bps-write" },
790 
791         { "iops_max",       "throttling.iops-total-max" },
792         { "iops_rd_max",    "throttling.iops-read-max" },
793         { "iops_wr_max",    "throttling.iops-write-max" },
794 
795         { "bps_max",        "throttling.bps-total-max" },
796         { "bps_rd_max",     "throttling.bps-read-max" },
797         { "bps_wr_max",     "throttling.bps-write-max" },
798 
799         { "iops_size",      "throttling.iops-size" },
800 
801         { "group",          "throttling.group" },
802 
803         { "readonly",       BDRV_OPT_READ_ONLY },
804     };
805 
806     for (i = 0; i < ARRAY_SIZE(opt_renames); i++) {
807         qemu_opt_rename(all_opts, opt_renames[i].from, opt_renames[i].to,
808                         &local_err);
809         if (local_err) {
810             error_propagate(errp, local_err);
811             return NULL;
812         }
813     }
814 
815     value = qemu_opt_get(all_opts, "cache");
816     if (value) {
817         int flags = 0;
818         bool writethrough;
819 
820         if (bdrv_parse_cache_mode(value, &flags, &writethrough) != 0) {
821             error_setg(errp, "invalid cache option");
822             return NULL;
823         }
824 
825         /* Specific options take precedence */
826         if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_WB)) {
827             qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_WB,
828                               !writethrough, &error_abort);
829         }
830         if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_DIRECT)) {
831             qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_DIRECT,
832                               !!(flags & BDRV_O_NOCACHE), &error_abort);
833         }
834         if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_NO_FLUSH)) {
835             qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_NO_FLUSH,
836                               !!(flags & BDRV_O_NO_FLUSH), &error_abort);
837         }
838         qemu_opt_unset(all_opts, "cache");
839     }
840 
841     /* Get a QDict for processing the options */
842     bs_opts = qdict_new();
843     qemu_opts_to_qdict(all_opts, bs_opts);
844 
845     legacy_opts = qemu_opts_create(&qemu_legacy_drive_opts, NULL, 0,
846                                    &error_abort);
847     qemu_opts_absorb_qdict(legacy_opts, bs_opts, &local_err);
848     if (local_err) {
849         error_propagate(errp, local_err);
850         goto fail;
851     }
852 
853     /* Media type */
854     value = qemu_opt_get(legacy_opts, "media");
855     if (value) {
856         if (!strcmp(value, "disk")) {
857             media = MEDIA_DISK;
858         } else if (!strcmp(value, "cdrom")) {
859             media = MEDIA_CDROM;
860             read_only = true;
861         } else {
862             error_setg(errp, "'%s' invalid media", value);
863             goto fail;
864         }
865     }
866 
867     /* copy-on-read is disabled with a warning for read-only devices */
868     read_only |= qemu_opt_get_bool(legacy_opts, BDRV_OPT_READ_ONLY, false);
869     copy_on_read = qemu_opt_get_bool(legacy_opts, "copy-on-read", false);
870 
871     if (read_only && copy_on_read) {
872         warn_report("disabling copy-on-read on read-only drive");
873         copy_on_read = false;
874     }
875 
876     qdict_put_str(bs_opts, BDRV_OPT_READ_ONLY, read_only ? "on" : "off");
877     qdict_put_str(bs_opts, "copy-on-read", copy_on_read ? "on" : "off");
878 
879     /* Controller type */
880     value = qemu_opt_get(legacy_opts, "if");
881     if (value) {
882         for (type = 0;
883              type < IF_COUNT && strcmp(value, if_name[type]);
884              type++) {
885         }
886         if (type == IF_COUNT) {
887             error_setg(errp, "unsupported bus type '%s'", value);
888             goto fail;
889         }
890     } else {
891         type = block_default_type;
892     }
893 
894     /* Device address specified by bus/unit or index.
895      * If none was specified, try to find the first free one. */
896     bus_id  = qemu_opt_get_number(legacy_opts, "bus", 0);
897     unit_id = qemu_opt_get_number(legacy_opts, "unit", -1);
898     index   = qemu_opt_get_number(legacy_opts, "index", -1);
899 
900     max_devs = if_max_devs[type];
901 
902     if (index != -1) {
903         if (bus_id != 0 || unit_id != -1) {
904             error_setg(errp, "index cannot be used with bus and unit");
905             goto fail;
906         }
907         bus_id = drive_index_to_bus_id(type, index);
908         unit_id = drive_index_to_unit_id(type, index);
909     }
910 
911     if (unit_id == -1) {
912        unit_id = 0;
913        while (drive_get(type, bus_id, unit_id) != NULL) {
914            unit_id++;
915            if (max_devs && unit_id >= max_devs) {
916                unit_id -= max_devs;
917                bus_id++;
918            }
919        }
920     }
921 
922     if (max_devs && unit_id >= max_devs) {
923         error_setg(errp, "unit %d too big (max is %d)", unit_id, max_devs - 1);
924         goto fail;
925     }
926 
927     if (drive_get(type, bus_id, unit_id) != NULL) {
928         error_setg(errp, "drive with bus=%d, unit=%d (index=%d) exists",
929                    bus_id, unit_id, index);
930         goto fail;
931     }
932 
933     /* no id supplied -> create one */
934     if (qemu_opts_id(all_opts) == NULL) {
935         char *new_id;
936         const char *mediastr = "";
937         if (type == IF_IDE || type == IF_SCSI) {
938             mediastr = (media == MEDIA_CDROM) ? "-cd" : "-hd";
939         }
940         if (max_devs) {
941             new_id = g_strdup_printf("%s%i%s%i", if_name[type], bus_id,
942                                      mediastr, unit_id);
943         } else {
944             new_id = g_strdup_printf("%s%s%i", if_name[type],
945                                      mediastr, unit_id);
946         }
947         qdict_put_str(bs_opts, "id", new_id);
948         g_free(new_id);
949     }
950 
951     /* Add virtio block device */
952     if (type == IF_VIRTIO) {
953         QemuOpts *devopts;
954         devopts = qemu_opts_create(qemu_find_opts("device"), NULL, 0,
955                                    &error_abort);
956         if (arch_type == QEMU_ARCH_S390X) {
957             qemu_opt_set(devopts, "driver", "virtio-blk-ccw", &error_abort);
958         } else {
959             qemu_opt_set(devopts, "driver", "virtio-blk-pci", &error_abort);
960         }
961         qemu_opt_set(devopts, "drive", qdict_get_str(bs_opts, "id"),
962                      &error_abort);
963     }
964 
965     filename = qemu_opt_get(legacy_opts, "file");
966 
967     /* Check werror/rerror compatibility with if=... */
968     werror = qemu_opt_get(legacy_opts, "werror");
969     if (werror != NULL) {
970         if (type != IF_IDE && type != IF_SCSI && type != IF_VIRTIO &&
971             type != IF_NONE) {
972             error_setg(errp, "werror is not supported by this bus type");
973             goto fail;
974         }
975         qdict_put_str(bs_opts, "werror", werror);
976     }
977 
978     rerror = qemu_opt_get(legacy_opts, "rerror");
979     if (rerror != NULL) {
980         if (type != IF_IDE && type != IF_VIRTIO && type != IF_SCSI &&
981             type != IF_NONE) {
982             error_setg(errp, "rerror is not supported by this bus type");
983             goto fail;
984         }
985         qdict_put_str(bs_opts, "rerror", rerror);
986     }
987 
988     /* Actual block device init: Functionality shared with blockdev-add */
989     blk = blockdev_init(filename, bs_opts, &local_err);
990     bs_opts = NULL;
991     if (!blk) {
992         error_propagate(errp, local_err);
993         goto fail;
994     } else {
995         assert(!local_err);
996     }
997 
998     /* Create legacy DriveInfo */
999     dinfo = g_malloc0(sizeof(*dinfo));
1000     dinfo->opts = all_opts;
1001 
1002     dinfo->type = type;
1003     dinfo->bus = bus_id;
1004     dinfo->unit = unit_id;
1005 
1006     blk_set_legacy_dinfo(blk, dinfo);
1007 
1008     switch(type) {
1009     case IF_IDE:
1010     case IF_SCSI:
1011     case IF_XEN:
1012     case IF_NONE:
1013         dinfo->media_cd = media == MEDIA_CDROM;
1014         break;
1015     default:
1016         break;
1017     }
1018 
1019 fail:
1020     qemu_opts_del(legacy_opts);
1021     qobject_unref(bs_opts);
1022     return dinfo;
1023 }
1024 
1025 static BlockDriverState *qmp_get_root_bs(const char *name, Error **errp)
1026 {
1027     BlockDriverState *bs;
1028 
1029     bs = bdrv_lookup_bs(name, name, errp);
1030     if (bs == NULL) {
1031         return NULL;
1032     }
1033 
1034     if (!bdrv_is_root_node(bs)) {
1035         error_setg(errp, "Need a root block node");
1036         return NULL;
1037     }
1038 
1039     if (!bdrv_is_inserted(bs)) {
1040         error_setg(errp, "Device has no medium");
1041         return NULL;
1042     }
1043 
1044     return bs;
1045 }
1046 
1047 static void blockdev_do_action(TransactionAction *action, Error **errp)
1048 {
1049     TransactionActionList list;
1050 
1051     list.value = action;
1052     list.next = NULL;
1053     qmp_transaction(&list, false, NULL, errp);
1054 }
1055 
1056 void qmp_blockdev_snapshot_sync(bool has_device, const char *device,
1057                                 bool has_node_name, const char *node_name,
1058                                 const char *snapshot_file,
1059                                 bool has_snapshot_node_name,
1060                                 const char *snapshot_node_name,
1061                                 bool has_format, const char *format,
1062                                 bool has_mode, NewImageMode mode, Error **errp)
1063 {
1064     BlockdevSnapshotSync snapshot = {
1065         .has_device = has_device,
1066         .device = (char *) device,
1067         .has_node_name = has_node_name,
1068         .node_name = (char *) node_name,
1069         .snapshot_file = (char *) snapshot_file,
1070         .has_snapshot_node_name = has_snapshot_node_name,
1071         .snapshot_node_name = (char *) snapshot_node_name,
1072         .has_format = has_format,
1073         .format = (char *) format,
1074         .has_mode = has_mode,
1075         .mode = mode,
1076     };
1077     TransactionAction action = {
1078         .type = TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC,
1079         .u.blockdev_snapshot_sync.data = &snapshot,
1080     };
1081     blockdev_do_action(&action, errp);
1082 }
1083 
1084 void qmp_blockdev_snapshot(const char *node, const char *overlay,
1085                            Error **errp)
1086 {
1087     BlockdevSnapshot snapshot_data = {
1088         .node = (char *) node,
1089         .overlay = (char *) overlay
1090     };
1091     TransactionAction action = {
1092         .type = TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT,
1093         .u.blockdev_snapshot.data = &snapshot_data,
1094     };
1095     blockdev_do_action(&action, errp);
1096 }
1097 
1098 void qmp_blockdev_snapshot_internal_sync(const char *device,
1099                                          const char *name,
1100                                          Error **errp)
1101 {
1102     BlockdevSnapshotInternal snapshot = {
1103         .device = (char *) device,
1104         .name = (char *) name
1105     };
1106     TransactionAction action = {
1107         .type = TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC,
1108         .u.blockdev_snapshot_internal_sync.data = &snapshot,
1109     };
1110     blockdev_do_action(&action, errp);
1111 }
1112 
1113 SnapshotInfo *qmp_blockdev_snapshot_delete_internal_sync(const char *device,
1114                                                          bool has_id,
1115                                                          const char *id,
1116                                                          bool has_name,
1117                                                          const char *name,
1118                                                          Error **errp)
1119 {
1120     BlockDriverState *bs;
1121     AioContext *aio_context;
1122     QEMUSnapshotInfo sn;
1123     Error *local_err = NULL;
1124     SnapshotInfo *info = NULL;
1125     int ret;
1126 
1127     bs = qmp_get_root_bs(device, errp);
1128     if (!bs) {
1129         return NULL;
1130     }
1131     aio_context = bdrv_get_aio_context(bs);
1132     aio_context_acquire(aio_context);
1133 
1134     if (!has_id) {
1135         id = NULL;
1136     }
1137 
1138     if (!has_name) {
1139         name = NULL;
1140     }
1141 
1142     if (!id && !name) {
1143         error_setg(errp, "Name or id must be provided");
1144         goto out_aio_context;
1145     }
1146 
1147     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_INTERNAL_SNAPSHOT_DELETE, errp)) {
1148         goto out_aio_context;
1149     }
1150 
1151     ret = bdrv_snapshot_find_by_id_and_name(bs, id, name, &sn, &local_err);
1152     if (local_err) {
1153         error_propagate(errp, local_err);
1154         goto out_aio_context;
1155     }
1156     if (!ret) {
1157         error_setg(errp,
1158                    "Snapshot with id '%s' and name '%s' does not exist on "
1159                    "device '%s'",
1160                    STR_OR_NULL(id), STR_OR_NULL(name), device);
1161         goto out_aio_context;
1162     }
1163 
1164     bdrv_snapshot_delete(bs, id, name, &local_err);
1165     if (local_err) {
1166         error_propagate(errp, local_err);
1167         goto out_aio_context;
1168     }
1169 
1170     aio_context_release(aio_context);
1171 
1172     info = g_new0(SnapshotInfo, 1);
1173     info->id = g_strdup(sn.id_str);
1174     info->name = g_strdup(sn.name);
1175     info->date_nsec = sn.date_nsec;
1176     info->date_sec = sn.date_sec;
1177     info->vm_state_size = sn.vm_state_size;
1178     info->vm_clock_nsec = sn.vm_clock_nsec % 1000000000;
1179     info->vm_clock_sec = sn.vm_clock_nsec / 1000000000;
1180 
1181     return info;
1182 
1183 out_aio_context:
1184     aio_context_release(aio_context);
1185     return NULL;
1186 }
1187 
1188 /**
1189  * block_dirty_bitmap_lookup:
1190  * Return a dirty bitmap (if present), after validating
1191  * the node reference and bitmap names.
1192  *
1193  * @node: The name of the BDS node to search for bitmaps
1194  * @name: The name of the bitmap to search for
1195  * @pbs: Output pointer for BDS lookup, if desired. Can be NULL.
1196  * @errp: Output pointer for error information. Can be NULL.
1197  *
1198  * @return: A bitmap object on success, or NULL on failure.
1199  */
1200 static BdrvDirtyBitmap *block_dirty_bitmap_lookup(const char *node,
1201                                                   const char *name,
1202                                                   BlockDriverState **pbs,
1203                                                   Error **errp)
1204 {
1205     BlockDriverState *bs;
1206     BdrvDirtyBitmap *bitmap;
1207 
1208     if (!node) {
1209         error_setg(errp, "Node cannot be NULL");
1210         return NULL;
1211     }
1212     if (!name) {
1213         error_setg(errp, "Bitmap name cannot be NULL");
1214         return NULL;
1215     }
1216     bs = bdrv_lookup_bs(node, node, NULL);
1217     if (!bs) {
1218         error_setg(errp, "Node '%s' not found", node);
1219         return NULL;
1220     }
1221 
1222     bitmap = bdrv_find_dirty_bitmap(bs, name);
1223     if (!bitmap) {
1224         error_setg(errp, "Dirty bitmap '%s' not found", name);
1225         return NULL;
1226     }
1227 
1228     if (pbs) {
1229         *pbs = bs;
1230     }
1231 
1232     return bitmap;
1233 }
1234 
1235 /* New and old BlockDriverState structs for atomic group operations */
1236 
1237 typedef struct BlkActionState BlkActionState;
1238 
1239 /**
1240  * BlkActionOps:
1241  * Table of operations that define an Action.
1242  *
1243  * @instance_size: Size of state struct, in bytes.
1244  * @prepare: Prepare the work, must NOT be NULL.
1245  * @commit: Commit the changes, can be NULL.
1246  * @abort: Abort the changes on fail, can be NULL.
1247  * @clean: Clean up resources after all transaction actions have called
1248  *         commit() or abort(). Can be NULL.
1249  *
1250  * Only prepare() may fail. In a single transaction, only one of commit() or
1251  * abort() will be called. clean() will always be called if it is present.
1252  */
1253 typedef struct BlkActionOps {
1254     size_t instance_size;
1255     void (*prepare)(BlkActionState *common, Error **errp);
1256     void (*commit)(BlkActionState *common);
1257     void (*abort)(BlkActionState *common);
1258     void (*clean)(BlkActionState *common);
1259 } BlkActionOps;
1260 
1261 /**
1262  * BlkActionState:
1263  * Describes one Action's state within a Transaction.
1264  *
1265  * @action: QAPI-defined enum identifying which Action to perform.
1266  * @ops: Table of ActionOps this Action can perform.
1267  * @block_job_txn: Transaction which this action belongs to.
1268  * @entry: List membership for all Actions in this Transaction.
1269  *
1270  * This structure must be arranged as first member in a subclassed type,
1271  * assuming that the compiler will also arrange it to the same offsets as the
1272  * base class.
1273  */
1274 struct BlkActionState {
1275     TransactionAction *action;
1276     const BlkActionOps *ops;
1277     JobTxn *block_job_txn;
1278     TransactionProperties *txn_props;
1279     QTAILQ_ENTRY(BlkActionState) entry;
1280 };
1281 
1282 /* internal snapshot private data */
1283 typedef struct InternalSnapshotState {
1284     BlkActionState common;
1285     BlockDriverState *bs;
1286     QEMUSnapshotInfo sn;
1287     bool created;
1288 } InternalSnapshotState;
1289 
1290 
1291 static int action_check_completion_mode(BlkActionState *s, Error **errp)
1292 {
1293     if (s->txn_props->completion_mode != ACTION_COMPLETION_MODE_INDIVIDUAL) {
1294         error_setg(errp,
1295                    "Action '%s' does not support Transaction property "
1296                    "completion-mode = %s",
1297                    TransactionActionKind_str(s->action->type),
1298                    ActionCompletionMode_str(s->txn_props->completion_mode));
1299         return -1;
1300     }
1301     return 0;
1302 }
1303 
1304 static void internal_snapshot_prepare(BlkActionState *common,
1305                                       Error **errp)
1306 {
1307     Error *local_err = NULL;
1308     const char *device;
1309     const char *name;
1310     BlockDriverState *bs;
1311     QEMUSnapshotInfo old_sn, *sn;
1312     bool ret;
1313     qemu_timeval tv;
1314     BlockdevSnapshotInternal *internal;
1315     InternalSnapshotState *state;
1316     AioContext *aio_context;
1317     int ret1;
1318 
1319     g_assert(common->action->type ==
1320              TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC);
1321     internal = common->action->u.blockdev_snapshot_internal_sync.data;
1322     state = DO_UPCAST(InternalSnapshotState, common, common);
1323 
1324     /* 1. parse input */
1325     device = internal->device;
1326     name = internal->name;
1327 
1328     /* 2. check for validation */
1329     if (action_check_completion_mode(common, errp) < 0) {
1330         return;
1331     }
1332 
1333     bs = qmp_get_root_bs(device, errp);
1334     if (!bs) {
1335         return;
1336     }
1337 
1338     aio_context = bdrv_get_aio_context(bs);
1339     aio_context_acquire(aio_context);
1340 
1341     state->bs = bs;
1342 
1343     /* Paired with .clean() */
1344     bdrv_drained_begin(bs);
1345 
1346     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_INTERNAL_SNAPSHOT, errp)) {
1347         goto out;
1348     }
1349 
1350     if (bdrv_is_read_only(bs)) {
1351         error_setg(errp, "Device '%s' is read only", device);
1352         goto out;
1353     }
1354 
1355     if (!bdrv_can_snapshot(bs)) {
1356         error_setg(errp, "Block format '%s' used by device '%s' "
1357                    "does not support internal snapshots",
1358                    bs->drv->format_name, device);
1359         goto out;
1360     }
1361 
1362     if (!strlen(name)) {
1363         error_setg(errp, "Name is empty");
1364         goto out;
1365     }
1366 
1367     /* check whether a snapshot with name exist */
1368     ret = bdrv_snapshot_find_by_id_and_name(bs, NULL, name, &old_sn,
1369                                             &local_err);
1370     if (local_err) {
1371         error_propagate(errp, local_err);
1372         goto out;
1373     } else if (ret) {
1374         error_setg(errp,
1375                    "Snapshot with name '%s' already exists on device '%s'",
1376                    name, device);
1377         goto out;
1378     }
1379 
1380     /* 3. take the snapshot */
1381     sn = &state->sn;
1382     pstrcpy(sn->name, sizeof(sn->name), name);
1383     qemu_gettimeofday(&tv);
1384     sn->date_sec = tv.tv_sec;
1385     sn->date_nsec = tv.tv_usec * 1000;
1386     sn->vm_clock_nsec = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL);
1387 
1388     ret1 = bdrv_snapshot_create(bs, sn);
1389     if (ret1 < 0) {
1390         error_setg_errno(errp, -ret1,
1391                          "Failed to create snapshot '%s' on device '%s'",
1392                          name, device);
1393         goto out;
1394     }
1395 
1396     /* 4. succeed, mark a snapshot is created */
1397     state->created = true;
1398 
1399 out:
1400     aio_context_release(aio_context);
1401 }
1402 
1403 static void internal_snapshot_abort(BlkActionState *common)
1404 {
1405     InternalSnapshotState *state =
1406                              DO_UPCAST(InternalSnapshotState, common, common);
1407     BlockDriverState *bs = state->bs;
1408     QEMUSnapshotInfo *sn = &state->sn;
1409     AioContext *aio_context;
1410     Error *local_error = NULL;
1411 
1412     if (!state->created) {
1413         return;
1414     }
1415 
1416     aio_context = bdrv_get_aio_context(state->bs);
1417     aio_context_acquire(aio_context);
1418 
1419     if (bdrv_snapshot_delete(bs, sn->id_str, sn->name, &local_error) < 0) {
1420         error_reportf_err(local_error,
1421                           "Failed to delete snapshot with id '%s' and "
1422                           "name '%s' on device '%s' in abort: ",
1423                           sn->id_str, sn->name,
1424                           bdrv_get_device_name(bs));
1425     }
1426 
1427     aio_context_release(aio_context);
1428 }
1429 
1430 static void internal_snapshot_clean(BlkActionState *common)
1431 {
1432     InternalSnapshotState *state = DO_UPCAST(InternalSnapshotState,
1433                                              common, common);
1434     AioContext *aio_context;
1435 
1436     if (!state->bs) {
1437         return;
1438     }
1439 
1440     aio_context = bdrv_get_aio_context(state->bs);
1441     aio_context_acquire(aio_context);
1442 
1443     bdrv_drained_end(state->bs);
1444 
1445     aio_context_release(aio_context);
1446 }
1447 
1448 /* external snapshot private data */
1449 typedef struct ExternalSnapshotState {
1450     BlkActionState common;
1451     BlockDriverState *old_bs;
1452     BlockDriverState *new_bs;
1453     bool overlay_appended;
1454 } ExternalSnapshotState;
1455 
1456 static void external_snapshot_prepare(BlkActionState *common,
1457                                       Error **errp)
1458 {
1459     int flags = 0;
1460     QDict *options = NULL;
1461     Error *local_err = NULL;
1462     /* Device and node name of the image to generate the snapshot from */
1463     const char *device;
1464     const char *node_name;
1465     /* Reference to the new image (for 'blockdev-snapshot') */
1466     const char *snapshot_ref;
1467     /* File name of the new image (for 'blockdev-snapshot-sync') */
1468     const char *new_image_file;
1469     ExternalSnapshotState *state =
1470                              DO_UPCAST(ExternalSnapshotState, common, common);
1471     TransactionAction *action = common->action;
1472     AioContext *aio_context;
1473     uint64_t perm, shared;
1474 
1475     /* 'blockdev-snapshot' and 'blockdev-snapshot-sync' have similar
1476      * purpose but a different set of parameters */
1477     switch (action->type) {
1478     case TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT:
1479         {
1480             BlockdevSnapshot *s = action->u.blockdev_snapshot.data;
1481             device = s->node;
1482             node_name = s->node;
1483             new_image_file = NULL;
1484             snapshot_ref = s->overlay;
1485         }
1486         break;
1487     case TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC:
1488         {
1489             BlockdevSnapshotSync *s = action->u.blockdev_snapshot_sync.data;
1490             device = s->has_device ? s->device : NULL;
1491             node_name = s->has_node_name ? s->node_name : NULL;
1492             new_image_file = s->snapshot_file;
1493             snapshot_ref = NULL;
1494         }
1495         break;
1496     default:
1497         g_assert_not_reached();
1498     }
1499 
1500     /* start processing */
1501     if (action_check_completion_mode(common, errp) < 0) {
1502         return;
1503     }
1504 
1505     state->old_bs = bdrv_lookup_bs(device, node_name, errp);
1506     if (!state->old_bs) {
1507         return;
1508     }
1509 
1510     aio_context = bdrv_get_aio_context(state->old_bs);
1511     aio_context_acquire(aio_context);
1512 
1513     /* Paired with .clean() */
1514     bdrv_drained_begin(state->old_bs);
1515 
1516     if (!bdrv_is_inserted(state->old_bs)) {
1517         error_setg(errp, QERR_DEVICE_HAS_NO_MEDIUM, device);
1518         goto out;
1519     }
1520 
1521     if (bdrv_op_is_blocked(state->old_bs,
1522                            BLOCK_OP_TYPE_EXTERNAL_SNAPSHOT, errp)) {
1523         goto out;
1524     }
1525 
1526     if (!bdrv_is_read_only(state->old_bs)) {
1527         if (bdrv_flush(state->old_bs)) {
1528             error_setg(errp, QERR_IO_ERROR);
1529             goto out;
1530         }
1531     }
1532 
1533     if (action->type == TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC) {
1534         BlockdevSnapshotSync *s = action->u.blockdev_snapshot_sync.data;
1535         const char *format = s->has_format ? s->format : "qcow2";
1536         enum NewImageMode mode;
1537         const char *snapshot_node_name =
1538             s->has_snapshot_node_name ? s->snapshot_node_name : NULL;
1539 
1540         if (node_name && !snapshot_node_name) {
1541             error_setg(errp, "New overlay node name missing");
1542             goto out;
1543         }
1544 
1545         if (snapshot_node_name &&
1546             bdrv_lookup_bs(snapshot_node_name, snapshot_node_name, NULL)) {
1547             error_setg(errp, "New overlay node name already in use");
1548             goto out;
1549         }
1550 
1551         flags = state->old_bs->open_flags;
1552         flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_COPY_ON_READ);
1553         flags |= BDRV_O_NO_BACKING;
1554 
1555         /* create new image w/backing file */
1556         mode = s->has_mode ? s->mode : NEW_IMAGE_MODE_ABSOLUTE_PATHS;
1557         if (mode != NEW_IMAGE_MODE_EXISTING) {
1558             int64_t size = bdrv_getlength(state->old_bs);
1559             if (size < 0) {
1560                 error_setg_errno(errp, -size, "bdrv_getlength failed");
1561                 goto out;
1562             }
1563             bdrv_refresh_filename(state->old_bs);
1564             bdrv_img_create(new_image_file, format,
1565                             state->old_bs->filename,
1566                             state->old_bs->drv->format_name,
1567                             NULL, size, flags, false, &local_err);
1568             if (local_err) {
1569                 error_propagate(errp, local_err);
1570                 goto out;
1571             }
1572         }
1573 
1574         options = qdict_new();
1575         if (snapshot_node_name) {
1576             qdict_put_str(options, "node-name", snapshot_node_name);
1577         }
1578         qdict_put_str(options, "driver", format);
1579     }
1580 
1581     state->new_bs = bdrv_open(new_image_file, snapshot_ref, options, flags,
1582                               errp);
1583     /* We will manually add the backing_hd field to the bs later */
1584     if (!state->new_bs) {
1585         goto out;
1586     }
1587 
1588     /*
1589      * Allow attaching a backing file to an overlay that's already in use only
1590      * if the parents don't assume that they are already seeing a valid image.
1591      * (Specifically, allow it as a mirror target, which is write-only access.)
1592      */
1593     bdrv_get_cumulative_perm(state->new_bs, &perm, &shared);
1594     if (perm & BLK_PERM_CONSISTENT_READ) {
1595         error_setg(errp, "The overlay is already in use");
1596         goto out;
1597     }
1598 
1599     if (state->new_bs->backing != NULL) {
1600         error_setg(errp, "The overlay already has a backing image");
1601         goto out;
1602     }
1603 
1604     if (!state->new_bs->drv->supports_backing) {
1605         error_setg(errp, "The overlay does not support backing images");
1606         goto out;
1607     }
1608 
1609     /* This removes our old bs and adds the new bs. This is an operation that
1610      * can fail, so we need to do it in .prepare; undoing it for abort is
1611      * always possible. */
1612     bdrv_ref(state->new_bs);
1613     bdrv_append(state->new_bs, state->old_bs, &local_err);
1614     if (local_err) {
1615         error_propagate(errp, local_err);
1616         goto out;
1617     }
1618     state->overlay_appended = true;
1619 
1620 out:
1621     aio_context_release(aio_context);
1622 }
1623 
1624 static void external_snapshot_commit(BlkActionState *common)
1625 {
1626     ExternalSnapshotState *state =
1627                              DO_UPCAST(ExternalSnapshotState, common, common);
1628     AioContext *aio_context;
1629 
1630     aio_context = bdrv_get_aio_context(state->old_bs);
1631     aio_context_acquire(aio_context);
1632 
1633     /* We don't need (or want) to use the transactional
1634      * bdrv_reopen_multiple() across all the entries at once, because we
1635      * don't want to abort all of them if one of them fails the reopen */
1636     if (!atomic_read(&state->old_bs->copy_on_read)) {
1637         bdrv_reopen_set_read_only(state->old_bs, true, NULL);
1638     }
1639 
1640     aio_context_release(aio_context);
1641 }
1642 
1643 static void external_snapshot_abort(BlkActionState *common)
1644 {
1645     ExternalSnapshotState *state =
1646                              DO_UPCAST(ExternalSnapshotState, common, common);
1647     if (state->new_bs) {
1648         if (state->overlay_appended) {
1649             AioContext *aio_context;
1650             AioContext *tmp_context;
1651             int ret;
1652 
1653             aio_context = bdrv_get_aio_context(state->old_bs);
1654             aio_context_acquire(aio_context);
1655 
1656             bdrv_ref(state->old_bs);   /* we can't let bdrv_set_backind_hd()
1657                                           close state->old_bs; we need it */
1658             bdrv_set_backing_hd(state->new_bs, NULL, &error_abort);
1659 
1660             /*
1661              * The call to bdrv_set_backing_hd() above returns state->old_bs to
1662              * the main AioContext. As we're still going to be using it, return
1663              * it to the AioContext it was before.
1664              */
1665             tmp_context = bdrv_get_aio_context(state->old_bs);
1666             if (aio_context != tmp_context) {
1667                 aio_context_release(aio_context);
1668                 aio_context_acquire(tmp_context);
1669 
1670                 ret = bdrv_try_set_aio_context(state->old_bs,
1671                                                aio_context, NULL);
1672                 assert(ret == 0);
1673 
1674                 aio_context_release(tmp_context);
1675                 aio_context_acquire(aio_context);
1676             }
1677 
1678             bdrv_replace_node(state->new_bs, state->old_bs, &error_abort);
1679             bdrv_unref(state->old_bs); /* bdrv_replace_node() ref'ed old_bs */
1680 
1681             aio_context_release(aio_context);
1682         }
1683     }
1684 }
1685 
1686 static void external_snapshot_clean(BlkActionState *common)
1687 {
1688     ExternalSnapshotState *state =
1689                              DO_UPCAST(ExternalSnapshotState, common, common);
1690     AioContext *aio_context;
1691 
1692     if (!state->old_bs) {
1693         return;
1694     }
1695 
1696     aio_context = bdrv_get_aio_context(state->old_bs);
1697     aio_context_acquire(aio_context);
1698 
1699     bdrv_drained_end(state->old_bs);
1700     bdrv_unref(state->new_bs);
1701 
1702     aio_context_release(aio_context);
1703 }
1704 
1705 typedef struct DriveBackupState {
1706     BlkActionState common;
1707     BlockDriverState *bs;
1708     BlockJob *job;
1709 } DriveBackupState;
1710 
1711 static BlockJob *do_backup_common(BackupCommon *backup,
1712                                   BlockDriverState *bs,
1713                                   BlockDriverState *target_bs,
1714                                   AioContext *aio_context,
1715                                   JobTxn *txn, Error **errp);
1716 
1717 static void drive_backup_prepare(BlkActionState *common, Error **errp)
1718 {
1719     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1720     DriveBackup *backup;
1721     BlockDriverState *bs;
1722     BlockDriverState *target_bs;
1723     BlockDriverState *source = NULL;
1724     AioContext *aio_context;
1725     AioContext *old_context;
1726     QDict *options;
1727     Error *local_err = NULL;
1728     int flags;
1729     int64_t size;
1730     bool set_backing_hd = false;
1731     int ret;
1732 
1733     assert(common->action->type == TRANSACTION_ACTION_KIND_DRIVE_BACKUP);
1734     backup = common->action->u.drive_backup.data;
1735 
1736     if (!backup->has_mode) {
1737         backup->mode = NEW_IMAGE_MODE_ABSOLUTE_PATHS;
1738     }
1739 
1740     bs = bdrv_lookup_bs(backup->device, backup->device, errp);
1741     if (!bs) {
1742         return;
1743     }
1744 
1745     if (!bs->drv) {
1746         error_setg(errp, "Device has no medium");
1747         return;
1748     }
1749 
1750     aio_context = bdrv_get_aio_context(bs);
1751     aio_context_acquire(aio_context);
1752 
1753     /* Paired with .clean() */
1754     bdrv_drained_begin(bs);
1755 
1756     if (!backup->has_format) {
1757         backup->format = backup->mode == NEW_IMAGE_MODE_EXISTING ?
1758                          NULL : (char *) bs->drv->format_name;
1759     }
1760 
1761     /* Early check to avoid creating target */
1762     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_BACKUP_SOURCE, errp)) {
1763         goto out;
1764     }
1765 
1766     flags = bs->open_flags | BDRV_O_RDWR;
1767 
1768     /*
1769      * See if we have a backing HD we can use to create our new image
1770      * on top of.
1771      */
1772     if (backup->sync == MIRROR_SYNC_MODE_TOP) {
1773         source = backing_bs(bs);
1774         if (!source) {
1775             backup->sync = MIRROR_SYNC_MODE_FULL;
1776         }
1777     }
1778     if (backup->sync == MIRROR_SYNC_MODE_NONE) {
1779         source = bs;
1780         flags |= BDRV_O_NO_BACKING;
1781         set_backing_hd = true;
1782     }
1783 
1784     size = bdrv_getlength(bs);
1785     if (size < 0) {
1786         error_setg_errno(errp, -size, "bdrv_getlength failed");
1787         goto out;
1788     }
1789 
1790     if (backup->mode != NEW_IMAGE_MODE_EXISTING) {
1791         assert(backup->format);
1792         if (source) {
1793             bdrv_refresh_filename(source);
1794             bdrv_img_create(backup->target, backup->format, source->filename,
1795                             source->drv->format_name, NULL,
1796                             size, flags, false, &local_err);
1797         } else {
1798             bdrv_img_create(backup->target, backup->format, NULL, NULL, NULL,
1799                             size, flags, false, &local_err);
1800         }
1801     }
1802 
1803     if (local_err) {
1804         error_propagate(errp, local_err);
1805         goto out;
1806     }
1807 
1808     options = qdict_new();
1809     qdict_put_str(options, "discard", "unmap");
1810     qdict_put_str(options, "detect-zeroes", "unmap");
1811     if (backup->format) {
1812         qdict_put_str(options, "driver", backup->format);
1813     }
1814 
1815     target_bs = bdrv_open(backup->target, NULL, options, flags, errp);
1816     if (!target_bs) {
1817         goto out;
1818     }
1819 
1820     /* Honor bdrv_try_set_aio_context() context acquisition requirements. */
1821     old_context = bdrv_get_aio_context(target_bs);
1822     aio_context_release(aio_context);
1823     aio_context_acquire(old_context);
1824 
1825     ret = bdrv_try_set_aio_context(target_bs, aio_context, errp);
1826     if (ret < 0) {
1827         bdrv_unref(target_bs);
1828         aio_context_release(old_context);
1829         return;
1830     }
1831 
1832     aio_context_release(old_context);
1833     aio_context_acquire(aio_context);
1834 
1835     if (set_backing_hd) {
1836         bdrv_set_backing_hd(target_bs, source, &local_err);
1837         if (local_err) {
1838             goto unref;
1839         }
1840     }
1841 
1842     state->bs = bs;
1843 
1844     state->job = do_backup_common(qapi_DriveBackup_base(backup),
1845                                   bs, target_bs, aio_context,
1846                                   common->block_job_txn, errp);
1847 
1848 unref:
1849     bdrv_unref(target_bs);
1850 out:
1851     aio_context_release(aio_context);
1852 }
1853 
1854 static void drive_backup_commit(BlkActionState *common)
1855 {
1856     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1857     AioContext *aio_context;
1858 
1859     aio_context = bdrv_get_aio_context(state->bs);
1860     aio_context_acquire(aio_context);
1861 
1862     assert(state->job);
1863     job_start(&state->job->job);
1864 
1865     aio_context_release(aio_context);
1866 }
1867 
1868 static void drive_backup_abort(BlkActionState *common)
1869 {
1870     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1871 
1872     if (state->job) {
1873         AioContext *aio_context;
1874 
1875         aio_context = bdrv_get_aio_context(state->bs);
1876         aio_context_acquire(aio_context);
1877 
1878         job_cancel_sync(&state->job->job);
1879 
1880         aio_context_release(aio_context);
1881     }
1882 }
1883 
1884 static void drive_backup_clean(BlkActionState *common)
1885 {
1886     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1887     AioContext *aio_context;
1888 
1889     if (!state->bs) {
1890         return;
1891     }
1892 
1893     aio_context = bdrv_get_aio_context(state->bs);
1894     aio_context_acquire(aio_context);
1895 
1896     bdrv_drained_end(state->bs);
1897 
1898     aio_context_release(aio_context);
1899 }
1900 
1901 typedef struct BlockdevBackupState {
1902     BlkActionState common;
1903     BlockDriverState *bs;
1904     BlockJob *job;
1905 } BlockdevBackupState;
1906 
1907 static void blockdev_backup_prepare(BlkActionState *common, Error **errp)
1908 {
1909     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1910     BlockdevBackup *backup;
1911     BlockDriverState *bs;
1912     BlockDriverState *target_bs;
1913     AioContext *aio_context;
1914     AioContext *old_context;
1915     int ret;
1916 
1917     assert(common->action->type == TRANSACTION_ACTION_KIND_BLOCKDEV_BACKUP);
1918     backup = common->action->u.blockdev_backup.data;
1919 
1920     bs = bdrv_lookup_bs(backup->device, backup->device, errp);
1921     if (!bs) {
1922         return;
1923     }
1924 
1925     target_bs = bdrv_lookup_bs(backup->target, backup->target, errp);
1926     if (!target_bs) {
1927         return;
1928     }
1929 
1930     /* Honor bdrv_try_set_aio_context() context acquisition requirements. */
1931     aio_context = bdrv_get_aio_context(bs);
1932     old_context = bdrv_get_aio_context(target_bs);
1933     aio_context_acquire(old_context);
1934 
1935     ret = bdrv_try_set_aio_context(target_bs, aio_context, errp);
1936     if (ret < 0) {
1937         aio_context_release(old_context);
1938         return;
1939     }
1940 
1941     aio_context_release(old_context);
1942     aio_context_acquire(aio_context);
1943     state->bs = bs;
1944 
1945     /* Paired with .clean() */
1946     bdrv_drained_begin(state->bs);
1947 
1948     state->job = do_backup_common(qapi_BlockdevBackup_base(backup),
1949                                   bs, target_bs, aio_context,
1950                                   common->block_job_txn, errp);
1951 
1952     aio_context_release(aio_context);
1953 }
1954 
1955 static void blockdev_backup_commit(BlkActionState *common)
1956 {
1957     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1958     AioContext *aio_context;
1959 
1960     aio_context = bdrv_get_aio_context(state->bs);
1961     aio_context_acquire(aio_context);
1962 
1963     assert(state->job);
1964     job_start(&state->job->job);
1965 
1966     aio_context_release(aio_context);
1967 }
1968 
1969 static void blockdev_backup_abort(BlkActionState *common)
1970 {
1971     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1972 
1973     if (state->job) {
1974         AioContext *aio_context;
1975 
1976         aio_context = bdrv_get_aio_context(state->bs);
1977         aio_context_acquire(aio_context);
1978 
1979         job_cancel_sync(&state->job->job);
1980 
1981         aio_context_release(aio_context);
1982     }
1983 }
1984 
1985 static void blockdev_backup_clean(BlkActionState *common)
1986 {
1987     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1988     AioContext *aio_context;
1989 
1990     if (!state->bs) {
1991         return;
1992     }
1993 
1994     aio_context = bdrv_get_aio_context(state->bs);
1995     aio_context_acquire(aio_context);
1996 
1997     bdrv_drained_end(state->bs);
1998 
1999     aio_context_release(aio_context);
2000 }
2001 
2002 typedef struct BlockDirtyBitmapState {
2003     BlkActionState common;
2004     BdrvDirtyBitmap *bitmap;
2005     BlockDriverState *bs;
2006     HBitmap *backup;
2007     bool prepared;
2008     bool was_enabled;
2009 } BlockDirtyBitmapState;
2010 
2011 static void block_dirty_bitmap_add_prepare(BlkActionState *common,
2012                                            Error **errp)
2013 {
2014     Error *local_err = NULL;
2015     BlockDirtyBitmapAdd *action;
2016     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2017                                              common, common);
2018 
2019     if (action_check_completion_mode(common, errp) < 0) {
2020         return;
2021     }
2022 
2023     action = common->action->u.block_dirty_bitmap_add.data;
2024     /* AIO context taken and released within qmp_block_dirty_bitmap_add */
2025     qmp_block_dirty_bitmap_add(action->node, action->name,
2026                                action->has_granularity, action->granularity,
2027                                action->has_persistent, action->persistent,
2028                                action->has_disabled, action->disabled,
2029                                &local_err);
2030 
2031     if (!local_err) {
2032         state->prepared = true;
2033     } else {
2034         error_propagate(errp, local_err);
2035     }
2036 }
2037 
2038 static void block_dirty_bitmap_add_abort(BlkActionState *common)
2039 {
2040     BlockDirtyBitmapAdd *action;
2041     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2042                                              common, common);
2043 
2044     action = common->action->u.block_dirty_bitmap_add.data;
2045     /* Should not be able to fail: IF the bitmap was added via .prepare(),
2046      * then the node reference and bitmap name must have been valid.
2047      */
2048     if (state->prepared) {
2049         qmp_block_dirty_bitmap_remove(action->node, action->name, &error_abort);
2050     }
2051 }
2052 
2053 static void block_dirty_bitmap_clear_prepare(BlkActionState *common,
2054                                              Error **errp)
2055 {
2056     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2057                                              common, common);
2058     BlockDirtyBitmap *action;
2059 
2060     if (action_check_completion_mode(common, errp) < 0) {
2061         return;
2062     }
2063 
2064     action = common->action->u.block_dirty_bitmap_clear.data;
2065     state->bitmap = block_dirty_bitmap_lookup(action->node,
2066                                               action->name,
2067                                               &state->bs,
2068                                               errp);
2069     if (!state->bitmap) {
2070         return;
2071     }
2072 
2073     if (bdrv_dirty_bitmap_check(state->bitmap, BDRV_BITMAP_DEFAULT, errp)) {
2074         return;
2075     }
2076 
2077     bdrv_clear_dirty_bitmap(state->bitmap, &state->backup);
2078 }
2079 
2080 static void block_dirty_bitmap_restore(BlkActionState *common)
2081 {
2082     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2083                                              common, common);
2084 
2085     if (state->backup) {
2086         bdrv_restore_dirty_bitmap(state->bitmap, state->backup);
2087     }
2088 }
2089 
2090 static void block_dirty_bitmap_free_backup(BlkActionState *common)
2091 {
2092     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2093                                              common, common);
2094 
2095     hbitmap_free(state->backup);
2096 }
2097 
2098 static void block_dirty_bitmap_enable_prepare(BlkActionState *common,
2099                                               Error **errp)
2100 {
2101     BlockDirtyBitmap *action;
2102     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2103                                              common, common);
2104 
2105     if (action_check_completion_mode(common, errp) < 0) {
2106         return;
2107     }
2108 
2109     action = common->action->u.block_dirty_bitmap_enable.data;
2110     state->bitmap = block_dirty_bitmap_lookup(action->node,
2111                                               action->name,
2112                                               NULL,
2113                                               errp);
2114     if (!state->bitmap) {
2115         return;
2116     }
2117 
2118     if (bdrv_dirty_bitmap_check(state->bitmap, BDRV_BITMAP_ALLOW_RO, errp)) {
2119         return;
2120     }
2121 
2122     state->was_enabled = bdrv_dirty_bitmap_enabled(state->bitmap);
2123     bdrv_enable_dirty_bitmap(state->bitmap);
2124 }
2125 
2126 static void block_dirty_bitmap_enable_abort(BlkActionState *common)
2127 {
2128     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2129                                              common, common);
2130 
2131     if (!state->was_enabled) {
2132         bdrv_disable_dirty_bitmap(state->bitmap);
2133     }
2134 }
2135 
2136 static void block_dirty_bitmap_disable_prepare(BlkActionState *common,
2137                                                Error **errp)
2138 {
2139     BlockDirtyBitmap *action;
2140     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2141                                              common, common);
2142 
2143     if (action_check_completion_mode(common, errp) < 0) {
2144         return;
2145     }
2146 
2147     action = common->action->u.block_dirty_bitmap_disable.data;
2148     state->bitmap = block_dirty_bitmap_lookup(action->node,
2149                                               action->name,
2150                                               NULL,
2151                                               errp);
2152     if (!state->bitmap) {
2153         return;
2154     }
2155 
2156     if (bdrv_dirty_bitmap_check(state->bitmap, BDRV_BITMAP_ALLOW_RO, errp)) {
2157         return;
2158     }
2159 
2160     state->was_enabled = bdrv_dirty_bitmap_enabled(state->bitmap);
2161     bdrv_disable_dirty_bitmap(state->bitmap);
2162 }
2163 
2164 static void block_dirty_bitmap_disable_abort(BlkActionState *common)
2165 {
2166     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2167                                              common, common);
2168 
2169     if (state->was_enabled) {
2170         bdrv_enable_dirty_bitmap(state->bitmap);
2171     }
2172 }
2173 
2174 static BdrvDirtyBitmap *do_block_dirty_bitmap_merge(
2175         const char *node, const char *target,
2176         BlockDirtyBitmapMergeSourceList *bitmaps,
2177         HBitmap **backup, Error **errp);
2178 
2179 static void block_dirty_bitmap_merge_prepare(BlkActionState *common,
2180                                              Error **errp)
2181 {
2182     BlockDirtyBitmapMerge *action;
2183     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2184                                              common, common);
2185 
2186     if (action_check_completion_mode(common, errp) < 0) {
2187         return;
2188     }
2189 
2190     action = common->action->u.block_dirty_bitmap_merge.data;
2191 
2192     state->bitmap = do_block_dirty_bitmap_merge(action->node, action->target,
2193                                                 action->bitmaps, &state->backup,
2194                                                 errp);
2195 }
2196 
2197 static BdrvDirtyBitmap *do_block_dirty_bitmap_remove(
2198         const char *node, const char *name, bool release,
2199         BlockDriverState **bitmap_bs, Error **errp);
2200 
2201 static void block_dirty_bitmap_remove_prepare(BlkActionState *common,
2202                                               Error **errp)
2203 {
2204     BlockDirtyBitmap *action;
2205     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2206                                              common, common);
2207 
2208     if (action_check_completion_mode(common, errp) < 0) {
2209         return;
2210     }
2211 
2212     action = common->action->u.block_dirty_bitmap_remove.data;
2213 
2214     state->bitmap = do_block_dirty_bitmap_remove(action->node, action->name,
2215                                                  false, &state->bs, errp);
2216     if (state->bitmap) {
2217         bdrv_dirty_bitmap_skip_store(state->bitmap, true);
2218         bdrv_dirty_bitmap_set_busy(state->bitmap, true);
2219     }
2220 }
2221 
2222 static void block_dirty_bitmap_remove_abort(BlkActionState *common)
2223 {
2224     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2225                                              common, common);
2226 
2227     if (state->bitmap) {
2228         bdrv_dirty_bitmap_skip_store(state->bitmap, false);
2229         bdrv_dirty_bitmap_set_busy(state->bitmap, false);
2230     }
2231 }
2232 
2233 static void block_dirty_bitmap_remove_commit(BlkActionState *common)
2234 {
2235     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2236                                              common, common);
2237 
2238     bdrv_dirty_bitmap_set_busy(state->bitmap, false);
2239     bdrv_release_dirty_bitmap(state->bitmap);
2240 }
2241 
2242 static void abort_prepare(BlkActionState *common, Error **errp)
2243 {
2244     error_setg(errp, "Transaction aborted using Abort action");
2245 }
2246 
2247 static void abort_commit(BlkActionState *common)
2248 {
2249     g_assert_not_reached(); /* this action never succeeds */
2250 }
2251 
2252 static const BlkActionOps actions[] = {
2253     [TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT] = {
2254         .instance_size = sizeof(ExternalSnapshotState),
2255         .prepare  = external_snapshot_prepare,
2256         .commit   = external_snapshot_commit,
2257         .abort = external_snapshot_abort,
2258         .clean = external_snapshot_clean,
2259     },
2260     [TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC] = {
2261         .instance_size = sizeof(ExternalSnapshotState),
2262         .prepare  = external_snapshot_prepare,
2263         .commit   = external_snapshot_commit,
2264         .abort = external_snapshot_abort,
2265         .clean = external_snapshot_clean,
2266     },
2267     [TRANSACTION_ACTION_KIND_DRIVE_BACKUP] = {
2268         .instance_size = sizeof(DriveBackupState),
2269         .prepare = drive_backup_prepare,
2270         .commit = drive_backup_commit,
2271         .abort = drive_backup_abort,
2272         .clean = drive_backup_clean,
2273     },
2274     [TRANSACTION_ACTION_KIND_BLOCKDEV_BACKUP] = {
2275         .instance_size = sizeof(BlockdevBackupState),
2276         .prepare = blockdev_backup_prepare,
2277         .commit = blockdev_backup_commit,
2278         .abort = blockdev_backup_abort,
2279         .clean = blockdev_backup_clean,
2280     },
2281     [TRANSACTION_ACTION_KIND_ABORT] = {
2282         .instance_size = sizeof(BlkActionState),
2283         .prepare = abort_prepare,
2284         .commit = abort_commit,
2285     },
2286     [TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC] = {
2287         .instance_size = sizeof(InternalSnapshotState),
2288         .prepare  = internal_snapshot_prepare,
2289         .abort = internal_snapshot_abort,
2290         .clean = internal_snapshot_clean,
2291     },
2292     [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_ADD] = {
2293         .instance_size = sizeof(BlockDirtyBitmapState),
2294         .prepare = block_dirty_bitmap_add_prepare,
2295         .abort = block_dirty_bitmap_add_abort,
2296     },
2297     [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_CLEAR] = {
2298         .instance_size = sizeof(BlockDirtyBitmapState),
2299         .prepare = block_dirty_bitmap_clear_prepare,
2300         .commit = block_dirty_bitmap_free_backup,
2301         .abort = block_dirty_bitmap_restore,
2302     },
2303     [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_ENABLE] = {
2304         .instance_size = sizeof(BlockDirtyBitmapState),
2305         .prepare = block_dirty_bitmap_enable_prepare,
2306         .abort = block_dirty_bitmap_enable_abort,
2307     },
2308     [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_DISABLE] = {
2309         .instance_size = sizeof(BlockDirtyBitmapState),
2310         .prepare = block_dirty_bitmap_disable_prepare,
2311         .abort = block_dirty_bitmap_disable_abort,
2312     },
2313     [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_MERGE] = {
2314         .instance_size = sizeof(BlockDirtyBitmapState),
2315         .prepare = block_dirty_bitmap_merge_prepare,
2316         .commit = block_dirty_bitmap_free_backup,
2317         .abort = block_dirty_bitmap_restore,
2318     },
2319     [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_REMOVE] = {
2320         .instance_size = sizeof(BlockDirtyBitmapState),
2321         .prepare = block_dirty_bitmap_remove_prepare,
2322         .commit = block_dirty_bitmap_remove_commit,
2323         .abort = block_dirty_bitmap_remove_abort,
2324     },
2325     /* Where are transactions for MIRROR, COMMIT and STREAM?
2326      * Although these blockjobs use transaction callbacks like the backup job,
2327      * these jobs do not necessarily adhere to transaction semantics.
2328      * These jobs may not fully undo all of their actions on abort, nor do they
2329      * necessarily work in transactions with more than one job in them.
2330      */
2331 };
2332 
2333 /**
2334  * Allocate a TransactionProperties structure if necessary, and fill
2335  * that structure with desired defaults if they are unset.
2336  */
2337 static TransactionProperties *get_transaction_properties(
2338     TransactionProperties *props)
2339 {
2340     if (!props) {
2341         props = g_new0(TransactionProperties, 1);
2342     }
2343 
2344     if (!props->has_completion_mode) {
2345         props->has_completion_mode = true;
2346         props->completion_mode = ACTION_COMPLETION_MODE_INDIVIDUAL;
2347     }
2348 
2349     return props;
2350 }
2351 
2352 /*
2353  * 'Atomic' group operations.  The operations are performed as a set, and if
2354  * any fail then we roll back all operations in the group.
2355  */
2356 void qmp_transaction(TransactionActionList *dev_list,
2357                      bool has_props,
2358                      struct TransactionProperties *props,
2359                      Error **errp)
2360 {
2361     TransactionActionList *dev_entry = dev_list;
2362     JobTxn *block_job_txn = NULL;
2363     BlkActionState *state, *next;
2364     Error *local_err = NULL;
2365 
2366     QTAILQ_HEAD(, BlkActionState) snap_bdrv_states;
2367     QTAILQ_INIT(&snap_bdrv_states);
2368 
2369     /* Does this transaction get canceled as a group on failure?
2370      * If not, we don't really need to make a JobTxn.
2371      */
2372     props = get_transaction_properties(props);
2373     if (props->completion_mode != ACTION_COMPLETION_MODE_INDIVIDUAL) {
2374         block_job_txn = job_txn_new();
2375     }
2376 
2377     /* drain all i/o before any operations */
2378     bdrv_drain_all();
2379 
2380     /* We don't do anything in this loop that commits us to the operations */
2381     while (NULL != dev_entry) {
2382         TransactionAction *dev_info = NULL;
2383         const BlkActionOps *ops;
2384 
2385         dev_info = dev_entry->value;
2386         dev_entry = dev_entry->next;
2387 
2388         assert(dev_info->type < ARRAY_SIZE(actions));
2389 
2390         ops = &actions[dev_info->type];
2391         assert(ops->instance_size > 0);
2392 
2393         state = g_malloc0(ops->instance_size);
2394         state->ops = ops;
2395         state->action = dev_info;
2396         state->block_job_txn = block_job_txn;
2397         state->txn_props = props;
2398         QTAILQ_INSERT_TAIL(&snap_bdrv_states, state, entry);
2399 
2400         state->ops->prepare(state, &local_err);
2401         if (local_err) {
2402             error_propagate(errp, local_err);
2403             goto delete_and_fail;
2404         }
2405     }
2406 
2407     QTAILQ_FOREACH(state, &snap_bdrv_states, entry) {
2408         if (state->ops->commit) {
2409             state->ops->commit(state);
2410         }
2411     }
2412 
2413     /* success */
2414     goto exit;
2415 
2416 delete_and_fail:
2417     /* failure, and it is all-or-none; roll back all operations */
2418     QTAILQ_FOREACH_REVERSE(state, &snap_bdrv_states, entry) {
2419         if (state->ops->abort) {
2420             state->ops->abort(state);
2421         }
2422     }
2423 exit:
2424     QTAILQ_FOREACH_SAFE(state, &snap_bdrv_states, entry, next) {
2425         if (state->ops->clean) {
2426             state->ops->clean(state);
2427         }
2428         g_free(state);
2429     }
2430     if (!has_props) {
2431         qapi_free_TransactionProperties(props);
2432     }
2433     job_txn_unref(block_job_txn);
2434 }
2435 
2436 void qmp_block_passwd(bool has_device, const char *device,
2437                       bool has_node_name, const char *node_name,
2438                       const char *password, Error **errp)
2439 {
2440     error_setg(errp,
2441                "Setting block passwords directly is no longer supported");
2442 }
2443 
2444 void qmp_block_dirty_bitmap_add(const char *node, const char *name,
2445                                 bool has_granularity, uint32_t granularity,
2446                                 bool has_persistent, bool persistent,
2447                                 bool has_disabled, bool disabled,
2448                                 Error **errp)
2449 {
2450     BlockDriverState *bs;
2451     BdrvDirtyBitmap *bitmap;
2452     AioContext *aio_context;
2453 
2454     if (!name || name[0] == '\0') {
2455         error_setg(errp, "Bitmap name cannot be empty");
2456         return;
2457     }
2458 
2459     bs = bdrv_lookup_bs(node, node, errp);
2460     if (!bs) {
2461         return;
2462     }
2463 
2464     aio_context = bdrv_get_aio_context(bs);
2465     aio_context_acquire(aio_context);
2466 
2467     if (has_granularity) {
2468         if (granularity < 512 || !is_power_of_2(granularity)) {
2469             error_setg(errp, "Granularity must be power of 2 "
2470                              "and at least 512");
2471             goto out;
2472         }
2473     } else {
2474         /* Default to cluster size, if available: */
2475         granularity = bdrv_get_default_bitmap_granularity(bs);
2476     }
2477 
2478     if (!has_persistent) {
2479         persistent = false;
2480     }
2481 
2482     if (!has_disabled) {
2483         disabled = false;
2484     }
2485 
2486     if (persistent &&
2487         !bdrv_can_store_new_dirty_bitmap(bs, name, granularity, errp))
2488     {
2489         goto out;
2490     }
2491 
2492     bitmap = bdrv_create_dirty_bitmap(bs, granularity, name, errp);
2493     if (bitmap == NULL) {
2494         goto out;
2495     }
2496 
2497     if (disabled) {
2498         bdrv_disable_dirty_bitmap(bitmap);
2499     }
2500 
2501     bdrv_dirty_bitmap_set_persistence(bitmap, persistent);
2502 
2503 out:
2504     aio_context_release(aio_context);
2505 }
2506 
2507 static BdrvDirtyBitmap *do_block_dirty_bitmap_remove(
2508         const char *node, const char *name, bool release,
2509         BlockDriverState **bitmap_bs, Error **errp)
2510 {
2511     BlockDriverState *bs;
2512     BdrvDirtyBitmap *bitmap;
2513     AioContext *aio_context;
2514 
2515     bitmap = block_dirty_bitmap_lookup(node, name, &bs, errp);
2516     if (!bitmap || !bs) {
2517         return NULL;
2518     }
2519 
2520     aio_context = bdrv_get_aio_context(bs);
2521     aio_context_acquire(aio_context);
2522 
2523     if (bdrv_dirty_bitmap_check(bitmap, BDRV_BITMAP_BUSY | BDRV_BITMAP_RO,
2524                                 errp)) {
2525         aio_context_release(aio_context);
2526         return NULL;
2527     }
2528 
2529     if (bdrv_dirty_bitmap_get_persistence(bitmap) &&
2530         bdrv_remove_persistent_dirty_bitmap(bs, name, errp) < 0)
2531     {
2532         aio_context_release(aio_context);
2533         return NULL;
2534     }
2535 
2536     if (release) {
2537         bdrv_release_dirty_bitmap(bitmap);
2538     }
2539 
2540     if (bitmap_bs) {
2541         *bitmap_bs = bs;
2542     }
2543 
2544     aio_context_release(aio_context);
2545     return release ? NULL : bitmap;
2546 }
2547 
2548 void qmp_block_dirty_bitmap_remove(const char *node, const char *name,
2549                                    Error **errp)
2550 {
2551     do_block_dirty_bitmap_remove(node, name, true, NULL, errp);
2552 }
2553 
2554 /**
2555  * Completely clear a bitmap, for the purposes of synchronizing a bitmap
2556  * immediately after a full backup operation.
2557  */
2558 void qmp_block_dirty_bitmap_clear(const char *node, const char *name,
2559                                   Error **errp)
2560 {
2561     BdrvDirtyBitmap *bitmap;
2562     BlockDriverState *bs;
2563 
2564     bitmap = block_dirty_bitmap_lookup(node, name, &bs, errp);
2565     if (!bitmap || !bs) {
2566         return;
2567     }
2568 
2569     if (bdrv_dirty_bitmap_check(bitmap, BDRV_BITMAP_DEFAULT, errp)) {
2570         return;
2571     }
2572 
2573     bdrv_clear_dirty_bitmap(bitmap, NULL);
2574 }
2575 
2576 void qmp_block_dirty_bitmap_enable(const char *node, const char *name,
2577                                    Error **errp)
2578 {
2579     BlockDriverState *bs;
2580     BdrvDirtyBitmap *bitmap;
2581 
2582     bitmap = block_dirty_bitmap_lookup(node, name, &bs, errp);
2583     if (!bitmap) {
2584         return;
2585     }
2586 
2587     if (bdrv_dirty_bitmap_check(bitmap, BDRV_BITMAP_ALLOW_RO, errp)) {
2588         return;
2589     }
2590 
2591     bdrv_enable_dirty_bitmap(bitmap);
2592 }
2593 
2594 void qmp_block_dirty_bitmap_disable(const char *node, const char *name,
2595                                     Error **errp)
2596 {
2597     BlockDriverState *bs;
2598     BdrvDirtyBitmap *bitmap;
2599 
2600     bitmap = block_dirty_bitmap_lookup(node, name, &bs, errp);
2601     if (!bitmap) {
2602         return;
2603     }
2604 
2605     if (bdrv_dirty_bitmap_check(bitmap, BDRV_BITMAP_ALLOW_RO, errp)) {
2606         return;
2607     }
2608 
2609     bdrv_disable_dirty_bitmap(bitmap);
2610 }
2611 
2612 static BdrvDirtyBitmap *do_block_dirty_bitmap_merge(
2613         const char *node, const char *target,
2614         BlockDirtyBitmapMergeSourceList *bitmaps,
2615         HBitmap **backup, Error **errp)
2616 {
2617     BlockDriverState *bs;
2618     BdrvDirtyBitmap *dst, *src, *anon;
2619     BlockDirtyBitmapMergeSourceList *lst;
2620     Error *local_err = NULL;
2621 
2622     dst = block_dirty_bitmap_lookup(node, target, &bs, errp);
2623     if (!dst) {
2624         return NULL;
2625     }
2626 
2627     anon = bdrv_create_dirty_bitmap(bs, bdrv_dirty_bitmap_granularity(dst),
2628                                     NULL, errp);
2629     if (!anon) {
2630         return NULL;
2631     }
2632 
2633     for (lst = bitmaps; lst; lst = lst->next) {
2634         switch (lst->value->type) {
2635             const char *name, *node;
2636         case QTYPE_QSTRING:
2637             name = lst->value->u.local;
2638             src = bdrv_find_dirty_bitmap(bs, name);
2639             if (!src) {
2640                 error_setg(errp, "Dirty bitmap '%s' not found", name);
2641                 dst = NULL;
2642                 goto out;
2643             }
2644             break;
2645         case QTYPE_QDICT:
2646             node = lst->value->u.external.node;
2647             name = lst->value->u.external.name;
2648             src = block_dirty_bitmap_lookup(node, name, NULL, errp);
2649             if (!src) {
2650                 dst = NULL;
2651                 goto out;
2652             }
2653             break;
2654         default:
2655             abort();
2656         }
2657 
2658         bdrv_merge_dirty_bitmap(anon, src, NULL, &local_err);
2659         if (local_err) {
2660             error_propagate(errp, local_err);
2661             dst = NULL;
2662             goto out;
2663         }
2664     }
2665 
2666     /* Merge into dst; dst is unchanged on failure. */
2667     bdrv_merge_dirty_bitmap(dst, anon, backup, errp);
2668 
2669  out:
2670     bdrv_release_dirty_bitmap(anon);
2671     return dst;
2672 }
2673 
2674 void qmp_block_dirty_bitmap_merge(const char *node, const char *target,
2675                                   BlockDirtyBitmapMergeSourceList *bitmaps,
2676                                   Error **errp)
2677 {
2678     do_block_dirty_bitmap_merge(node, target, bitmaps, NULL, errp);
2679 }
2680 
2681 BlockDirtyBitmapSha256 *qmp_x_debug_block_dirty_bitmap_sha256(const char *node,
2682                                                               const char *name,
2683                                                               Error **errp)
2684 {
2685     BdrvDirtyBitmap *bitmap;
2686     BlockDriverState *bs;
2687     BlockDirtyBitmapSha256 *ret = NULL;
2688     char *sha256;
2689 
2690     bitmap = block_dirty_bitmap_lookup(node, name, &bs, errp);
2691     if (!bitmap || !bs) {
2692         return NULL;
2693     }
2694 
2695     sha256 = bdrv_dirty_bitmap_sha256(bitmap, errp);
2696     if (sha256 == NULL) {
2697         return NULL;
2698     }
2699 
2700     ret = g_new(BlockDirtyBitmapSha256, 1);
2701     ret->sha256 = sha256;
2702 
2703     return ret;
2704 }
2705 
2706 void qmp_block_resize(bool has_device, const char *device,
2707                       bool has_node_name, const char *node_name,
2708                       int64_t size, Error **errp)
2709 {
2710     Error *local_err = NULL;
2711     BlockBackend *blk = NULL;
2712     BlockDriverState *bs;
2713     AioContext *aio_context;
2714     int ret;
2715 
2716     bs = bdrv_lookup_bs(has_device ? device : NULL,
2717                         has_node_name ? node_name : NULL,
2718                         &local_err);
2719     if (local_err) {
2720         error_propagate(errp, local_err);
2721         return;
2722     }
2723 
2724     aio_context = bdrv_get_aio_context(bs);
2725     aio_context_acquire(aio_context);
2726 
2727     if (size < 0) {
2728         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "size", "a >0 size");
2729         goto out;
2730     }
2731 
2732     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_RESIZE, NULL)) {
2733         error_setg(errp, QERR_DEVICE_IN_USE, device);
2734         goto out;
2735     }
2736 
2737     blk = blk_new(bdrv_get_aio_context(bs), BLK_PERM_RESIZE, BLK_PERM_ALL);
2738     ret = blk_insert_bs(blk, bs, errp);
2739     if (ret < 0) {
2740         goto out;
2741     }
2742 
2743     bdrv_drained_begin(bs);
2744     ret = blk_truncate(blk, size, false, PREALLOC_MODE_OFF, errp);
2745     bdrv_drained_end(bs);
2746 
2747 out:
2748     blk_unref(blk);
2749     aio_context_release(aio_context);
2750 }
2751 
2752 void qmp_block_stream(bool has_job_id, const char *job_id, const char *device,
2753                       bool has_base, const char *base,
2754                       bool has_base_node, const char *base_node,
2755                       bool has_backing_file, const char *backing_file,
2756                       bool has_speed, int64_t speed,
2757                       bool has_on_error, BlockdevOnError on_error,
2758                       bool has_auto_finalize, bool auto_finalize,
2759                       bool has_auto_dismiss, bool auto_dismiss,
2760                       Error **errp)
2761 {
2762     BlockDriverState *bs, *iter;
2763     BlockDriverState *base_bs = NULL;
2764     AioContext *aio_context;
2765     Error *local_err = NULL;
2766     const char *base_name = NULL;
2767     int job_flags = JOB_DEFAULT;
2768 
2769     if (!has_on_error) {
2770         on_error = BLOCKDEV_ON_ERROR_REPORT;
2771     }
2772 
2773     bs = bdrv_lookup_bs(device, device, errp);
2774     if (!bs) {
2775         return;
2776     }
2777 
2778     aio_context = bdrv_get_aio_context(bs);
2779     aio_context_acquire(aio_context);
2780 
2781     if (has_base && has_base_node) {
2782         error_setg(errp, "'base' and 'base-node' cannot be specified "
2783                    "at the same time");
2784         goto out;
2785     }
2786 
2787     if (has_base) {
2788         base_bs = bdrv_find_backing_image(bs, base);
2789         if (base_bs == NULL) {
2790             error_setg(errp, QERR_BASE_NOT_FOUND, base);
2791             goto out;
2792         }
2793         assert(bdrv_get_aio_context(base_bs) == aio_context);
2794         base_name = base;
2795     }
2796 
2797     if (has_base_node) {
2798         base_bs = bdrv_lookup_bs(NULL, base_node, errp);
2799         if (!base_bs) {
2800             goto out;
2801         }
2802         if (bs == base_bs || !bdrv_chain_contains(bs, base_bs)) {
2803             error_setg(errp, "Node '%s' is not a backing image of '%s'",
2804                        base_node, device);
2805             goto out;
2806         }
2807         assert(bdrv_get_aio_context(base_bs) == aio_context);
2808         bdrv_refresh_filename(base_bs);
2809         base_name = base_bs->filename;
2810     }
2811 
2812     /* Check for op blockers in the whole chain between bs and base */
2813     for (iter = bs; iter && iter != base_bs; iter = backing_bs(iter)) {
2814         if (bdrv_op_is_blocked(iter, BLOCK_OP_TYPE_STREAM, errp)) {
2815             goto out;
2816         }
2817     }
2818 
2819     /* if we are streaming the entire chain, the result will have no backing
2820      * file, and specifying one is therefore an error */
2821     if (base_bs == NULL && has_backing_file) {
2822         error_setg(errp, "backing file specified, but streaming the "
2823                          "entire chain");
2824         goto out;
2825     }
2826 
2827     /* backing_file string overrides base bs filename */
2828     base_name = has_backing_file ? backing_file : base_name;
2829 
2830     if (has_auto_finalize && !auto_finalize) {
2831         job_flags |= JOB_MANUAL_FINALIZE;
2832     }
2833     if (has_auto_dismiss && !auto_dismiss) {
2834         job_flags |= JOB_MANUAL_DISMISS;
2835     }
2836 
2837     stream_start(has_job_id ? job_id : NULL, bs, base_bs, base_name,
2838                  job_flags, has_speed ? speed : 0, on_error, &local_err);
2839     if (local_err) {
2840         error_propagate(errp, local_err);
2841         goto out;
2842     }
2843 
2844     trace_qmp_block_stream(bs);
2845 
2846 out:
2847     aio_context_release(aio_context);
2848 }
2849 
2850 void qmp_block_commit(bool has_job_id, const char *job_id, const char *device,
2851                       bool has_base_node, const char *base_node,
2852                       bool has_base, const char *base,
2853                       bool has_top_node, const char *top_node,
2854                       bool has_top, const char *top,
2855                       bool has_backing_file, const char *backing_file,
2856                       bool has_speed, int64_t speed,
2857                       bool has_on_error, BlockdevOnError on_error,
2858                       bool has_filter_node_name, const char *filter_node_name,
2859                       bool has_auto_finalize, bool auto_finalize,
2860                       bool has_auto_dismiss, bool auto_dismiss,
2861                       Error **errp)
2862 {
2863     BlockDriverState *bs;
2864     BlockDriverState *iter;
2865     BlockDriverState *base_bs, *top_bs;
2866     AioContext *aio_context;
2867     Error *local_err = NULL;
2868     int job_flags = JOB_DEFAULT;
2869 
2870     if (!has_speed) {
2871         speed = 0;
2872     }
2873     if (!has_on_error) {
2874         on_error = BLOCKDEV_ON_ERROR_REPORT;
2875     }
2876     if (!has_filter_node_name) {
2877         filter_node_name = NULL;
2878     }
2879     if (has_auto_finalize && !auto_finalize) {
2880         job_flags |= JOB_MANUAL_FINALIZE;
2881     }
2882     if (has_auto_dismiss && !auto_dismiss) {
2883         job_flags |= JOB_MANUAL_DISMISS;
2884     }
2885 
2886     /* Important Note:
2887      *  libvirt relies on the DeviceNotFound error class in order to probe for
2888      *  live commit feature versions; for this to work, we must make sure to
2889      *  perform the device lookup before any generic errors that may occur in a
2890      *  scenario in which all optional arguments are omitted. */
2891     bs = qmp_get_root_bs(device, &local_err);
2892     if (!bs) {
2893         bs = bdrv_lookup_bs(device, device, NULL);
2894         if (!bs) {
2895             error_free(local_err);
2896             error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
2897                       "Device '%s' not found", device);
2898         } else {
2899             error_propagate(errp, local_err);
2900         }
2901         return;
2902     }
2903 
2904     aio_context = bdrv_get_aio_context(bs);
2905     aio_context_acquire(aio_context);
2906 
2907     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_COMMIT_SOURCE, errp)) {
2908         goto out;
2909     }
2910 
2911     /* default top_bs is the active layer */
2912     top_bs = bs;
2913 
2914     if (has_top_node && has_top) {
2915         error_setg(errp, "'top-node' and 'top' are mutually exclusive");
2916         goto out;
2917     } else if (has_top_node) {
2918         top_bs = bdrv_lookup_bs(NULL, top_node, errp);
2919         if (top_bs == NULL) {
2920             goto out;
2921         }
2922         if (!bdrv_chain_contains(bs, top_bs)) {
2923             error_setg(errp, "'%s' is not in this backing file chain",
2924                        top_node);
2925             goto out;
2926         }
2927     } else if (has_top && top) {
2928         /* This strcmp() is just a shortcut, there is no need to
2929          * refresh @bs's filename.  If it mismatches,
2930          * bdrv_find_backing_image() will do the refresh and may still
2931          * return @bs. */
2932         if (strcmp(bs->filename, top) != 0) {
2933             top_bs = bdrv_find_backing_image(bs, top);
2934         }
2935     }
2936 
2937     if (top_bs == NULL) {
2938         error_setg(errp, "Top image file %s not found", top ? top : "NULL");
2939         goto out;
2940     }
2941 
2942     assert(bdrv_get_aio_context(top_bs) == aio_context);
2943 
2944     if (has_base_node && has_base) {
2945         error_setg(errp, "'base-node' and 'base' are mutually exclusive");
2946         goto out;
2947     } else if (has_base_node) {
2948         base_bs = bdrv_lookup_bs(NULL, base_node, errp);
2949         if (base_bs == NULL) {
2950             goto out;
2951         }
2952         if (!bdrv_chain_contains(top_bs, base_bs)) {
2953             error_setg(errp, "'%s' is not in this backing file chain",
2954                        base_node);
2955             goto out;
2956         }
2957     } else if (has_base && base) {
2958         base_bs = bdrv_find_backing_image(top_bs, base);
2959     } else {
2960         base_bs = bdrv_find_base(top_bs);
2961     }
2962 
2963     if (base_bs == NULL) {
2964         error_setg(errp, QERR_BASE_NOT_FOUND, base ? base : "NULL");
2965         goto out;
2966     }
2967 
2968     assert(bdrv_get_aio_context(base_bs) == aio_context);
2969 
2970     for (iter = top_bs; iter != backing_bs(base_bs); iter = backing_bs(iter)) {
2971         if (bdrv_op_is_blocked(iter, BLOCK_OP_TYPE_COMMIT_TARGET, errp)) {
2972             goto out;
2973         }
2974     }
2975 
2976     /* Do not allow attempts to commit an image into itself */
2977     if (top_bs == base_bs) {
2978         error_setg(errp, "cannot commit an image into itself");
2979         goto out;
2980     }
2981 
2982     if (top_bs == bs) {
2983         if (has_backing_file) {
2984             error_setg(errp, "'backing-file' specified,"
2985                              " but 'top' is the active layer");
2986             goto out;
2987         }
2988         commit_active_start(has_job_id ? job_id : NULL, bs, base_bs,
2989                             job_flags, speed, on_error,
2990                             filter_node_name, NULL, NULL, false, &local_err);
2991     } else {
2992         BlockDriverState *overlay_bs = bdrv_find_overlay(bs, top_bs);
2993         if (bdrv_op_is_blocked(overlay_bs, BLOCK_OP_TYPE_COMMIT_TARGET, errp)) {
2994             goto out;
2995         }
2996         commit_start(has_job_id ? job_id : NULL, bs, base_bs, top_bs, job_flags,
2997                      speed, on_error, has_backing_file ? backing_file : NULL,
2998                      filter_node_name, &local_err);
2999     }
3000     if (local_err != NULL) {
3001         error_propagate(errp, local_err);
3002         goto out;
3003     }
3004 
3005 out:
3006     aio_context_release(aio_context);
3007 }
3008 
3009 /* Common QMP interface for drive-backup and blockdev-backup */
3010 static BlockJob *do_backup_common(BackupCommon *backup,
3011                                   BlockDriverState *bs,
3012                                   BlockDriverState *target_bs,
3013                                   AioContext *aio_context,
3014                                   JobTxn *txn, Error **errp)
3015 {
3016     BlockJob *job = NULL;
3017     BdrvDirtyBitmap *bmap = NULL;
3018     int job_flags = JOB_DEFAULT;
3019 
3020     if (!backup->has_speed) {
3021         backup->speed = 0;
3022     }
3023     if (!backup->has_on_source_error) {
3024         backup->on_source_error = BLOCKDEV_ON_ERROR_REPORT;
3025     }
3026     if (!backup->has_on_target_error) {
3027         backup->on_target_error = BLOCKDEV_ON_ERROR_REPORT;
3028     }
3029     if (!backup->has_job_id) {
3030         backup->job_id = NULL;
3031     }
3032     if (!backup->has_auto_finalize) {
3033         backup->auto_finalize = true;
3034     }
3035     if (!backup->has_auto_dismiss) {
3036         backup->auto_dismiss = true;
3037     }
3038     if (!backup->has_compress) {
3039         backup->compress = false;
3040     }
3041 
3042     if ((backup->sync == MIRROR_SYNC_MODE_BITMAP) ||
3043         (backup->sync == MIRROR_SYNC_MODE_INCREMENTAL)) {
3044         /* done before desugaring 'incremental' to print the right message */
3045         if (!backup->has_bitmap) {
3046             error_setg(errp, "must provide a valid bitmap name for "
3047                        "'%s' sync mode", MirrorSyncMode_str(backup->sync));
3048             return NULL;
3049         }
3050     }
3051 
3052     if (backup->sync == MIRROR_SYNC_MODE_INCREMENTAL) {
3053         if (backup->has_bitmap_mode &&
3054             backup->bitmap_mode != BITMAP_SYNC_MODE_ON_SUCCESS) {
3055             error_setg(errp, "Bitmap sync mode must be '%s' "
3056                        "when using sync mode '%s'",
3057                        BitmapSyncMode_str(BITMAP_SYNC_MODE_ON_SUCCESS),
3058                        MirrorSyncMode_str(backup->sync));
3059             return NULL;
3060         }
3061         backup->has_bitmap_mode = true;
3062         backup->sync = MIRROR_SYNC_MODE_BITMAP;
3063         backup->bitmap_mode = BITMAP_SYNC_MODE_ON_SUCCESS;
3064     }
3065 
3066     if (backup->has_bitmap) {
3067         bmap = bdrv_find_dirty_bitmap(bs, backup->bitmap);
3068         if (!bmap) {
3069             error_setg(errp, "Bitmap '%s' could not be found", backup->bitmap);
3070             return NULL;
3071         }
3072         if (!backup->has_bitmap_mode) {
3073             error_setg(errp, "Bitmap sync mode must be given "
3074                        "when providing a bitmap");
3075             return NULL;
3076         }
3077         if (bdrv_dirty_bitmap_check(bmap, BDRV_BITMAP_ALLOW_RO, errp)) {
3078             return NULL;
3079         }
3080 
3081         /* This does not produce a useful bitmap artifact: */
3082         if (backup->sync == MIRROR_SYNC_MODE_NONE) {
3083             error_setg(errp, "sync mode '%s' does not produce meaningful bitmap"
3084                        " outputs", MirrorSyncMode_str(backup->sync));
3085             return NULL;
3086         }
3087 
3088         /* If the bitmap isn't used for input or output, this is useless: */
3089         if (backup->bitmap_mode == BITMAP_SYNC_MODE_NEVER &&
3090             backup->sync != MIRROR_SYNC_MODE_BITMAP) {
3091             error_setg(errp, "Bitmap sync mode '%s' has no meaningful effect"
3092                        " when combined with sync mode '%s'",
3093                        BitmapSyncMode_str(backup->bitmap_mode),
3094                        MirrorSyncMode_str(backup->sync));
3095             return NULL;
3096         }
3097     }
3098 
3099     if (!backup->has_bitmap && backup->has_bitmap_mode) {
3100         error_setg(errp, "Cannot specify bitmap sync mode without a bitmap");
3101         return NULL;
3102     }
3103 
3104     if (!backup->auto_finalize) {
3105         job_flags |= JOB_MANUAL_FINALIZE;
3106     }
3107     if (!backup->auto_dismiss) {
3108         job_flags |= JOB_MANUAL_DISMISS;
3109     }
3110 
3111     job = backup_job_create(backup->job_id, bs, target_bs, backup->speed,
3112                             backup->sync, bmap, backup->bitmap_mode,
3113                             backup->compress,
3114                             backup->filter_node_name,
3115                             backup->on_source_error,
3116                             backup->on_target_error,
3117                             job_flags, NULL, NULL, txn, errp);
3118     return job;
3119 }
3120 
3121 void qmp_drive_backup(DriveBackup *backup, Error **errp)
3122 {
3123     TransactionAction action = {
3124         .type = TRANSACTION_ACTION_KIND_DRIVE_BACKUP,
3125         .u.drive_backup.data = backup,
3126     };
3127     blockdev_do_action(&action, errp);
3128 }
3129 
3130 BlockDeviceInfoList *qmp_query_named_block_nodes(bool has_flat,
3131                                                  bool flat,
3132                                                  Error **errp)
3133 {
3134     bool return_flat = has_flat && flat;
3135 
3136     return bdrv_named_nodes_list(return_flat, errp);
3137 }
3138 
3139 XDbgBlockGraph *qmp_x_debug_query_block_graph(Error **errp)
3140 {
3141     return bdrv_get_xdbg_block_graph(errp);
3142 }
3143 
3144 void qmp_blockdev_backup(BlockdevBackup *backup, Error **errp)
3145 {
3146     TransactionAction action = {
3147         .type = TRANSACTION_ACTION_KIND_BLOCKDEV_BACKUP,
3148         .u.blockdev_backup.data = backup,
3149     };
3150     blockdev_do_action(&action, errp);
3151 }
3152 
3153 /* Parameter check and block job starting for drive mirroring.
3154  * Caller should hold @device and @target's aio context (must be the same).
3155  **/
3156 static void blockdev_mirror_common(const char *job_id, BlockDriverState *bs,
3157                                    BlockDriverState *target,
3158                                    bool has_replaces, const char *replaces,
3159                                    enum MirrorSyncMode sync,
3160                                    BlockMirrorBackingMode backing_mode,
3161                                    bool zero_target,
3162                                    bool has_speed, int64_t speed,
3163                                    bool has_granularity, uint32_t granularity,
3164                                    bool has_buf_size, int64_t buf_size,
3165                                    bool has_on_source_error,
3166                                    BlockdevOnError on_source_error,
3167                                    bool has_on_target_error,
3168                                    BlockdevOnError on_target_error,
3169                                    bool has_unmap, bool unmap,
3170                                    bool has_filter_node_name,
3171                                    const char *filter_node_name,
3172                                    bool has_copy_mode, MirrorCopyMode copy_mode,
3173                                    bool has_auto_finalize, bool auto_finalize,
3174                                    bool has_auto_dismiss, bool auto_dismiss,
3175                                    Error **errp)
3176 {
3177     int job_flags = JOB_DEFAULT;
3178 
3179     if (!has_speed) {
3180         speed = 0;
3181     }
3182     if (!has_on_source_error) {
3183         on_source_error = BLOCKDEV_ON_ERROR_REPORT;
3184     }
3185     if (!has_on_target_error) {
3186         on_target_error = BLOCKDEV_ON_ERROR_REPORT;
3187     }
3188     if (!has_granularity) {
3189         granularity = 0;
3190     }
3191     if (!has_buf_size) {
3192         buf_size = 0;
3193     }
3194     if (!has_unmap) {
3195         unmap = true;
3196     }
3197     if (!has_filter_node_name) {
3198         filter_node_name = NULL;
3199     }
3200     if (!has_copy_mode) {
3201         copy_mode = MIRROR_COPY_MODE_BACKGROUND;
3202     }
3203     if (has_auto_finalize && !auto_finalize) {
3204         job_flags |= JOB_MANUAL_FINALIZE;
3205     }
3206     if (has_auto_dismiss && !auto_dismiss) {
3207         job_flags |= JOB_MANUAL_DISMISS;
3208     }
3209 
3210     if (granularity != 0 && (granularity < 512 || granularity > 1048576 * 64)) {
3211         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "granularity",
3212                    "a value in range [512B, 64MB]");
3213         return;
3214     }
3215     if (granularity & (granularity - 1)) {
3216         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "granularity",
3217                    "power of 2");
3218         return;
3219     }
3220 
3221     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_MIRROR_SOURCE, errp)) {
3222         return;
3223     }
3224     if (bdrv_op_is_blocked(target, BLOCK_OP_TYPE_MIRROR_TARGET, errp)) {
3225         return;
3226     }
3227 
3228     if (!bs->backing && sync == MIRROR_SYNC_MODE_TOP) {
3229         sync = MIRROR_SYNC_MODE_FULL;
3230     }
3231 
3232     if (has_replaces) {
3233         BlockDriverState *to_replace_bs;
3234         AioContext *replace_aio_context;
3235         int64_t bs_size, replace_size;
3236 
3237         bs_size = bdrv_getlength(bs);
3238         if (bs_size < 0) {
3239             error_setg_errno(errp, -bs_size, "Failed to query device's size");
3240             return;
3241         }
3242 
3243         to_replace_bs = check_to_replace_node(bs, replaces, errp);
3244         if (!to_replace_bs) {
3245             return;
3246         }
3247 
3248         replace_aio_context = bdrv_get_aio_context(to_replace_bs);
3249         aio_context_acquire(replace_aio_context);
3250         replace_size = bdrv_getlength(to_replace_bs);
3251         aio_context_release(replace_aio_context);
3252 
3253         if (replace_size < 0) {
3254             error_setg_errno(errp, -replace_size,
3255                              "Failed to query the replacement node's size");
3256             return;
3257         }
3258         if (bs_size != replace_size) {
3259             error_setg(errp, "cannot replace image with a mirror image of "
3260                              "different size");
3261             return;
3262         }
3263     }
3264 
3265     /* pass the node name to replace to mirror start since it's loose coupling
3266      * and will allow to check whether the node still exist at mirror completion
3267      */
3268     mirror_start(job_id, bs, target,
3269                  has_replaces ? replaces : NULL, job_flags,
3270                  speed, granularity, buf_size, sync, backing_mode, zero_target,
3271                  on_source_error, on_target_error, unmap, filter_node_name,
3272                  copy_mode, errp);
3273 }
3274 
3275 void qmp_drive_mirror(DriveMirror *arg, Error **errp)
3276 {
3277     BlockDriverState *bs;
3278     BlockDriverState *source, *target_bs;
3279     AioContext *aio_context;
3280     AioContext *old_context;
3281     BlockMirrorBackingMode backing_mode;
3282     Error *local_err = NULL;
3283     QDict *options = NULL;
3284     int flags;
3285     int64_t size;
3286     const char *format = arg->format;
3287     bool zero_target;
3288     int ret;
3289 
3290     bs = qmp_get_root_bs(arg->device, errp);
3291     if (!bs) {
3292         return;
3293     }
3294 
3295     /* Early check to avoid creating target */
3296     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_MIRROR_SOURCE, errp)) {
3297         return;
3298     }
3299 
3300     aio_context = bdrv_get_aio_context(bs);
3301     aio_context_acquire(aio_context);
3302 
3303     if (!arg->has_mode) {
3304         arg->mode = NEW_IMAGE_MODE_ABSOLUTE_PATHS;
3305     }
3306 
3307     if (!arg->has_format) {
3308         format = (arg->mode == NEW_IMAGE_MODE_EXISTING
3309                   ? NULL : bs->drv->format_name);
3310     }
3311 
3312     flags = bs->open_flags | BDRV_O_RDWR;
3313     source = backing_bs(bs);
3314     if (!source && arg->sync == MIRROR_SYNC_MODE_TOP) {
3315         arg->sync = MIRROR_SYNC_MODE_FULL;
3316     }
3317     if (arg->sync == MIRROR_SYNC_MODE_NONE) {
3318         source = bs;
3319     }
3320 
3321     size = bdrv_getlength(bs);
3322     if (size < 0) {
3323         error_setg_errno(errp, -size, "bdrv_getlength failed");
3324         goto out;
3325     }
3326 
3327     if (arg->has_replaces) {
3328         if (!arg->has_node_name) {
3329             error_setg(errp, "a node-name must be provided when replacing a"
3330                              " named node of the graph");
3331             goto out;
3332         }
3333     }
3334 
3335     if (arg->mode == NEW_IMAGE_MODE_ABSOLUTE_PATHS) {
3336         backing_mode = MIRROR_SOURCE_BACKING_CHAIN;
3337     } else {
3338         backing_mode = MIRROR_OPEN_BACKING_CHAIN;
3339     }
3340 
3341     /* Don't open backing image in create() */
3342     flags |= BDRV_O_NO_BACKING;
3343 
3344     if ((arg->sync == MIRROR_SYNC_MODE_FULL || !source)
3345         && arg->mode != NEW_IMAGE_MODE_EXISTING)
3346     {
3347         /* create new image w/o backing file */
3348         assert(format);
3349         bdrv_img_create(arg->target, format,
3350                         NULL, NULL, NULL, size, flags, false, &local_err);
3351     } else {
3352         switch (arg->mode) {
3353         case NEW_IMAGE_MODE_EXISTING:
3354             break;
3355         case NEW_IMAGE_MODE_ABSOLUTE_PATHS:
3356             /* create new image with backing file */
3357             bdrv_refresh_filename(source);
3358             bdrv_img_create(arg->target, format,
3359                             source->filename,
3360                             source->drv->format_name,
3361                             NULL, size, flags, false, &local_err);
3362             break;
3363         default:
3364             abort();
3365         }
3366     }
3367 
3368     if (local_err) {
3369         error_propagate(errp, local_err);
3370         goto out;
3371     }
3372 
3373     options = qdict_new();
3374     if (arg->has_node_name) {
3375         qdict_put_str(options, "node-name", arg->node_name);
3376     }
3377     if (format) {
3378         qdict_put_str(options, "driver", format);
3379     }
3380 
3381     /* Mirroring takes care of copy-on-write using the source's backing
3382      * file.
3383      */
3384     target_bs = bdrv_open(arg->target, NULL, options, flags, errp);
3385     if (!target_bs) {
3386         goto out;
3387     }
3388 
3389     zero_target = (arg->sync == MIRROR_SYNC_MODE_FULL &&
3390                    (arg->mode == NEW_IMAGE_MODE_EXISTING ||
3391                     !bdrv_has_zero_init(target_bs)));
3392 
3393 
3394     /* Honor bdrv_try_set_aio_context() context acquisition requirements. */
3395     old_context = bdrv_get_aio_context(target_bs);
3396     aio_context_release(aio_context);
3397     aio_context_acquire(old_context);
3398 
3399     ret = bdrv_try_set_aio_context(target_bs, aio_context, errp);
3400     if (ret < 0) {
3401         bdrv_unref(target_bs);
3402         aio_context_release(old_context);
3403         return;
3404     }
3405 
3406     aio_context_release(old_context);
3407     aio_context_acquire(aio_context);
3408 
3409     blockdev_mirror_common(arg->has_job_id ? arg->job_id : NULL, bs, target_bs,
3410                            arg->has_replaces, arg->replaces, arg->sync,
3411                            backing_mode, zero_target,
3412                            arg->has_speed, arg->speed,
3413                            arg->has_granularity, arg->granularity,
3414                            arg->has_buf_size, arg->buf_size,
3415                            arg->has_on_source_error, arg->on_source_error,
3416                            arg->has_on_target_error, arg->on_target_error,
3417                            arg->has_unmap, arg->unmap,
3418                            false, NULL,
3419                            arg->has_copy_mode, arg->copy_mode,
3420                            arg->has_auto_finalize, arg->auto_finalize,
3421                            arg->has_auto_dismiss, arg->auto_dismiss,
3422                            &local_err);
3423     bdrv_unref(target_bs);
3424     error_propagate(errp, local_err);
3425 out:
3426     aio_context_release(aio_context);
3427 }
3428 
3429 void qmp_blockdev_mirror(bool has_job_id, const char *job_id,
3430                          const char *device, const char *target,
3431                          bool has_replaces, const char *replaces,
3432                          MirrorSyncMode sync,
3433                          bool has_speed, int64_t speed,
3434                          bool has_granularity, uint32_t granularity,
3435                          bool has_buf_size, int64_t buf_size,
3436                          bool has_on_source_error,
3437                          BlockdevOnError on_source_error,
3438                          bool has_on_target_error,
3439                          BlockdevOnError on_target_error,
3440                          bool has_filter_node_name,
3441                          const char *filter_node_name,
3442                          bool has_copy_mode, MirrorCopyMode copy_mode,
3443                          bool has_auto_finalize, bool auto_finalize,
3444                          bool has_auto_dismiss, bool auto_dismiss,
3445                          Error **errp)
3446 {
3447     BlockDriverState *bs;
3448     BlockDriverState *target_bs;
3449     AioContext *aio_context;
3450     AioContext *old_context;
3451     BlockMirrorBackingMode backing_mode = MIRROR_LEAVE_BACKING_CHAIN;
3452     Error *local_err = NULL;
3453     bool zero_target;
3454     int ret;
3455 
3456     bs = qmp_get_root_bs(device, errp);
3457     if (!bs) {
3458         return;
3459     }
3460 
3461     target_bs = bdrv_lookup_bs(target, target, errp);
3462     if (!target_bs) {
3463         return;
3464     }
3465 
3466     zero_target = (sync == MIRROR_SYNC_MODE_FULL);
3467 
3468     /* Honor bdrv_try_set_aio_context() context acquisition requirements. */
3469     old_context = bdrv_get_aio_context(target_bs);
3470     aio_context = bdrv_get_aio_context(bs);
3471     aio_context_acquire(old_context);
3472 
3473     ret = bdrv_try_set_aio_context(target_bs, aio_context, errp);
3474 
3475     aio_context_release(old_context);
3476     aio_context_acquire(aio_context);
3477 
3478     if (ret < 0) {
3479         goto out;
3480     }
3481 
3482     blockdev_mirror_common(has_job_id ? job_id : NULL, bs, target_bs,
3483                            has_replaces, replaces, sync, backing_mode,
3484                            zero_target, has_speed, speed,
3485                            has_granularity, granularity,
3486                            has_buf_size, buf_size,
3487                            has_on_source_error, on_source_error,
3488                            has_on_target_error, on_target_error,
3489                            true, true,
3490                            has_filter_node_name, filter_node_name,
3491                            has_copy_mode, copy_mode,
3492                            has_auto_finalize, auto_finalize,
3493                            has_auto_dismiss, auto_dismiss,
3494                            &local_err);
3495     error_propagate(errp, local_err);
3496 out:
3497     aio_context_release(aio_context);
3498 }
3499 
3500 /* Get a block job using its ID and acquire its AioContext */
3501 static BlockJob *find_block_job(const char *id, AioContext **aio_context,
3502                                 Error **errp)
3503 {
3504     BlockJob *job;
3505 
3506     assert(id != NULL);
3507 
3508     *aio_context = NULL;
3509 
3510     job = block_job_get(id);
3511 
3512     if (!job) {
3513         error_set(errp, ERROR_CLASS_DEVICE_NOT_ACTIVE,
3514                   "Block job '%s' not found", id);
3515         return NULL;
3516     }
3517 
3518     *aio_context = blk_get_aio_context(job->blk);
3519     aio_context_acquire(*aio_context);
3520 
3521     return job;
3522 }
3523 
3524 void qmp_block_job_set_speed(const char *device, int64_t speed, Error **errp)
3525 {
3526     AioContext *aio_context;
3527     BlockJob *job = find_block_job(device, &aio_context, errp);
3528 
3529     if (!job) {
3530         return;
3531     }
3532 
3533     block_job_set_speed(job, speed, errp);
3534     aio_context_release(aio_context);
3535 }
3536 
3537 void qmp_block_job_cancel(const char *device,
3538                           bool has_force, bool force, Error **errp)
3539 {
3540     AioContext *aio_context;
3541     BlockJob *job = find_block_job(device, &aio_context, errp);
3542 
3543     if (!job) {
3544         return;
3545     }
3546 
3547     if (!has_force) {
3548         force = false;
3549     }
3550 
3551     if (job_user_paused(&job->job) && !force) {
3552         error_setg(errp, "The block job for device '%s' is currently paused",
3553                    device);
3554         goto out;
3555     }
3556 
3557     trace_qmp_block_job_cancel(job);
3558     job_user_cancel(&job->job, force, errp);
3559 out:
3560     aio_context_release(aio_context);
3561 }
3562 
3563 void qmp_block_job_pause(const char *device, Error **errp)
3564 {
3565     AioContext *aio_context;
3566     BlockJob *job = find_block_job(device, &aio_context, errp);
3567 
3568     if (!job) {
3569         return;
3570     }
3571 
3572     trace_qmp_block_job_pause(job);
3573     job_user_pause(&job->job, errp);
3574     aio_context_release(aio_context);
3575 }
3576 
3577 void qmp_block_job_resume(const char *device, Error **errp)
3578 {
3579     AioContext *aio_context;
3580     BlockJob *job = find_block_job(device, &aio_context, errp);
3581 
3582     if (!job) {
3583         return;
3584     }
3585 
3586     trace_qmp_block_job_resume(job);
3587     job_user_resume(&job->job, errp);
3588     aio_context_release(aio_context);
3589 }
3590 
3591 void qmp_block_job_complete(const char *device, Error **errp)
3592 {
3593     AioContext *aio_context;
3594     BlockJob *job = find_block_job(device, &aio_context, errp);
3595 
3596     if (!job) {
3597         return;
3598     }
3599 
3600     trace_qmp_block_job_complete(job);
3601     job_complete(&job->job, errp);
3602     aio_context_release(aio_context);
3603 }
3604 
3605 void qmp_block_job_finalize(const char *id, Error **errp)
3606 {
3607     AioContext *aio_context;
3608     BlockJob *job = find_block_job(id, &aio_context, errp);
3609 
3610     if (!job) {
3611         return;
3612     }
3613 
3614     trace_qmp_block_job_finalize(job);
3615     job_finalize(&job->job, errp);
3616     aio_context_release(aio_context);
3617 }
3618 
3619 void qmp_block_job_dismiss(const char *id, Error **errp)
3620 {
3621     AioContext *aio_context;
3622     BlockJob *bjob = find_block_job(id, &aio_context, errp);
3623     Job *job;
3624 
3625     if (!bjob) {
3626         return;
3627     }
3628 
3629     trace_qmp_block_job_dismiss(bjob);
3630     job = &bjob->job;
3631     job_dismiss(&job, errp);
3632     aio_context_release(aio_context);
3633 }
3634 
3635 void qmp_change_backing_file(const char *device,
3636                              const char *image_node_name,
3637                              const char *backing_file,
3638                              Error **errp)
3639 {
3640     BlockDriverState *bs = NULL;
3641     AioContext *aio_context;
3642     BlockDriverState *image_bs = NULL;
3643     Error *local_err = NULL;
3644     bool ro;
3645     int ret;
3646 
3647     bs = qmp_get_root_bs(device, errp);
3648     if (!bs) {
3649         return;
3650     }
3651 
3652     aio_context = bdrv_get_aio_context(bs);
3653     aio_context_acquire(aio_context);
3654 
3655     image_bs = bdrv_lookup_bs(NULL, image_node_name, &local_err);
3656     if (local_err) {
3657         error_propagate(errp, local_err);
3658         goto out;
3659     }
3660 
3661     if (!image_bs) {
3662         error_setg(errp, "image file not found");
3663         goto out;
3664     }
3665 
3666     if (bdrv_find_base(image_bs) == image_bs) {
3667         error_setg(errp, "not allowing backing file change on an image "
3668                          "without a backing file");
3669         goto out;
3670     }
3671 
3672     /* even though we are not necessarily operating on bs, we need it to
3673      * determine if block ops are currently prohibited on the chain */
3674     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_CHANGE, errp)) {
3675         goto out;
3676     }
3677 
3678     /* final sanity check */
3679     if (!bdrv_chain_contains(bs, image_bs)) {
3680         error_setg(errp, "'%s' and image file are not in the same chain",
3681                    device);
3682         goto out;
3683     }
3684 
3685     /* if not r/w, reopen to make r/w */
3686     ro = bdrv_is_read_only(image_bs);
3687 
3688     if (ro) {
3689         if (bdrv_reopen_set_read_only(image_bs, false, errp) != 0) {
3690             goto out;
3691         }
3692     }
3693 
3694     ret = bdrv_change_backing_file(image_bs, backing_file,
3695                                image_bs->drv ? image_bs->drv->format_name : "");
3696 
3697     if (ret < 0) {
3698         error_setg_errno(errp, -ret, "Could not change backing file to '%s'",
3699                          backing_file);
3700         /* don't exit here, so we can try to restore open flags if
3701          * appropriate */
3702     }
3703 
3704     if (ro) {
3705         bdrv_reopen_set_read_only(image_bs, true, &local_err);
3706         error_propagate(errp, local_err);
3707     }
3708 
3709 out:
3710     aio_context_release(aio_context);
3711 }
3712 
3713 void qmp_blockdev_add(BlockdevOptions *options, Error **errp)
3714 {
3715     BlockDriverState *bs;
3716     QObject *obj;
3717     Visitor *v = qobject_output_visitor_new(&obj);
3718     QDict *qdict;
3719     Error *local_err = NULL;
3720 
3721     visit_type_BlockdevOptions(v, NULL, &options, &local_err);
3722     if (local_err) {
3723         error_propagate(errp, local_err);
3724         goto fail;
3725     }
3726 
3727     visit_complete(v, &obj);
3728     qdict = qobject_to(QDict, obj);
3729 
3730     qdict_flatten(qdict);
3731 
3732     if (!qdict_get_try_str(qdict, "node-name")) {
3733         error_setg(errp, "'node-name' must be specified for the root node");
3734         goto fail;
3735     }
3736 
3737     bs = bds_tree_init(qdict, errp);
3738     if (!bs) {
3739         goto fail;
3740     }
3741 
3742     bdrv_set_monitor_owned(bs);
3743 
3744 fail:
3745     visit_free(v);
3746 }
3747 
3748 void qmp_x_blockdev_reopen(BlockdevOptions *options, Error **errp)
3749 {
3750     BlockDriverState *bs;
3751     AioContext *ctx;
3752     QObject *obj;
3753     Visitor *v = qobject_output_visitor_new(&obj);
3754     Error *local_err = NULL;
3755     BlockReopenQueue *queue;
3756     QDict *qdict;
3757 
3758     /* Check for the selected node name */
3759     if (!options->has_node_name) {
3760         error_setg(errp, "Node name not specified");
3761         goto fail;
3762     }
3763 
3764     bs = bdrv_find_node(options->node_name);
3765     if (!bs) {
3766         error_setg(errp, "Cannot find node named '%s'", options->node_name);
3767         goto fail;
3768     }
3769 
3770     /* Put all options in a QDict and flatten it */
3771     visit_type_BlockdevOptions(v, NULL, &options, &local_err);
3772     if (local_err) {
3773         error_propagate(errp, local_err);
3774         goto fail;
3775     }
3776 
3777     visit_complete(v, &obj);
3778     qdict = qobject_to(QDict, obj);
3779 
3780     qdict_flatten(qdict);
3781 
3782     /* Perform the reopen operation */
3783     ctx = bdrv_get_aio_context(bs);
3784     aio_context_acquire(ctx);
3785     bdrv_subtree_drained_begin(bs);
3786     queue = bdrv_reopen_queue(NULL, bs, qdict, false);
3787     bdrv_reopen_multiple(queue, errp);
3788     bdrv_subtree_drained_end(bs);
3789     aio_context_release(ctx);
3790 
3791 fail:
3792     visit_free(v);
3793 }
3794 
3795 void qmp_blockdev_del(const char *node_name, Error **errp)
3796 {
3797     AioContext *aio_context;
3798     BlockDriverState *bs;
3799 
3800     bs = bdrv_find_node(node_name);
3801     if (!bs) {
3802         error_setg(errp, "Cannot find node %s", node_name);
3803         return;
3804     }
3805     if (bdrv_has_blk(bs)) {
3806         error_setg(errp, "Node %s is in use", node_name);
3807         return;
3808     }
3809     aio_context = bdrv_get_aio_context(bs);
3810     aio_context_acquire(aio_context);
3811 
3812     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_DRIVE_DEL, errp)) {
3813         goto out;
3814     }
3815 
3816     if (!QTAILQ_IN_USE(bs, monitor_list)) {
3817         error_setg(errp, "Node %s is not owned by the monitor",
3818                    bs->node_name);
3819         goto out;
3820     }
3821 
3822     if (bs->refcnt > 1) {
3823         error_setg(errp, "Block device %s is in use",
3824                    bdrv_get_device_or_node_name(bs));
3825         goto out;
3826     }
3827 
3828     QTAILQ_REMOVE(&monitor_bdrv_states, bs, monitor_list);
3829     bdrv_unref(bs);
3830 
3831 out:
3832     aio_context_release(aio_context);
3833 }
3834 
3835 static BdrvChild *bdrv_find_child(BlockDriverState *parent_bs,
3836                                   const char *child_name)
3837 {
3838     BdrvChild *child;
3839 
3840     QLIST_FOREACH(child, &parent_bs->children, next) {
3841         if (strcmp(child->name, child_name) == 0) {
3842             return child;
3843         }
3844     }
3845 
3846     return NULL;
3847 }
3848 
3849 void qmp_x_blockdev_change(const char *parent, bool has_child,
3850                            const char *child, bool has_node,
3851                            const char *node, Error **errp)
3852 {
3853     BlockDriverState *parent_bs, *new_bs = NULL;
3854     BdrvChild *p_child;
3855 
3856     parent_bs = bdrv_lookup_bs(parent, parent, errp);
3857     if (!parent_bs) {
3858         return;
3859     }
3860 
3861     if (has_child == has_node) {
3862         if (has_child) {
3863             error_setg(errp, "The parameters child and node are in conflict");
3864         } else {
3865             error_setg(errp, "Either child or node must be specified");
3866         }
3867         return;
3868     }
3869 
3870     if (has_child) {
3871         p_child = bdrv_find_child(parent_bs, child);
3872         if (!p_child) {
3873             error_setg(errp, "Node '%s' does not have child '%s'",
3874                        parent, child);
3875             return;
3876         }
3877         bdrv_del_child(parent_bs, p_child, errp);
3878     }
3879 
3880     if (has_node) {
3881         new_bs = bdrv_find_node(node);
3882         if (!new_bs) {
3883             error_setg(errp, "Node '%s' not found", node);
3884             return;
3885         }
3886         bdrv_add_child(parent_bs, new_bs, errp);
3887     }
3888 }
3889 
3890 BlockJobInfoList *qmp_query_block_jobs(Error **errp)
3891 {
3892     BlockJobInfoList *head = NULL, **p_next = &head;
3893     BlockJob *job;
3894 
3895     for (job = block_job_next(NULL); job; job = block_job_next(job)) {
3896         BlockJobInfoList *elem;
3897         AioContext *aio_context;
3898 
3899         if (block_job_is_internal(job)) {
3900             continue;
3901         }
3902         elem = g_new0(BlockJobInfoList, 1);
3903         aio_context = blk_get_aio_context(job->blk);
3904         aio_context_acquire(aio_context);
3905         elem->value = block_job_query(job, errp);
3906         aio_context_release(aio_context);
3907         if (!elem->value) {
3908             g_free(elem);
3909             qapi_free_BlockJobInfoList(head);
3910             return NULL;
3911         }
3912         *p_next = elem;
3913         p_next = &elem->next;
3914     }
3915 
3916     return head;
3917 }
3918 
3919 void qmp_x_blockdev_set_iothread(const char *node_name, StrOrNull *iothread,
3920                                  bool has_force, bool force, Error **errp)
3921 {
3922     AioContext *old_context;
3923     AioContext *new_context;
3924     BlockDriverState *bs;
3925 
3926     bs = bdrv_find_node(node_name);
3927     if (!bs) {
3928         error_setg(errp, "Cannot find node %s", node_name);
3929         return;
3930     }
3931 
3932     /* Protects against accidents. */
3933     if (!(has_force && force) && bdrv_has_blk(bs)) {
3934         error_setg(errp, "Node %s is associated with a BlockBackend and could "
3935                          "be in use (use force=true to override this check)",
3936                          node_name);
3937         return;
3938     }
3939 
3940     if (iothread->type == QTYPE_QSTRING) {
3941         IOThread *obj = iothread_by_id(iothread->u.s);
3942         if (!obj) {
3943             error_setg(errp, "Cannot find iothread %s", iothread->u.s);
3944             return;
3945         }
3946 
3947         new_context = iothread_get_aio_context(obj);
3948     } else {
3949         new_context = qemu_get_aio_context();
3950     }
3951 
3952     old_context = bdrv_get_aio_context(bs);
3953     aio_context_acquire(old_context);
3954 
3955     bdrv_try_set_aio_context(bs, new_context, errp);
3956 
3957     aio_context_release(old_context);
3958 }
3959 
3960 QemuOptsList qemu_common_drive_opts = {
3961     .name = "drive",
3962     .head = QTAILQ_HEAD_INITIALIZER(qemu_common_drive_opts.head),
3963     .desc = {
3964         {
3965             .name = "snapshot",
3966             .type = QEMU_OPT_BOOL,
3967             .help = "enable/disable snapshot mode",
3968         },{
3969             .name = "aio",
3970             .type = QEMU_OPT_STRING,
3971             .help = "host AIO implementation (threads, native, io_uring)",
3972         },{
3973             .name = BDRV_OPT_CACHE_WB,
3974             .type = QEMU_OPT_BOOL,
3975             .help = "Enable writeback mode",
3976         },{
3977             .name = "format",
3978             .type = QEMU_OPT_STRING,
3979             .help = "disk format (raw, qcow2, ...)",
3980         },{
3981             .name = "rerror",
3982             .type = QEMU_OPT_STRING,
3983             .help = "read error action",
3984         },{
3985             .name = "werror",
3986             .type = QEMU_OPT_STRING,
3987             .help = "write error action",
3988         },{
3989             .name = BDRV_OPT_READ_ONLY,
3990             .type = QEMU_OPT_BOOL,
3991             .help = "open drive file as read-only",
3992         },
3993 
3994         THROTTLE_OPTS,
3995 
3996         {
3997             .name = "throttling.group",
3998             .type = QEMU_OPT_STRING,
3999             .help = "name of the block throttling group",
4000         },{
4001             .name = "copy-on-read",
4002             .type = QEMU_OPT_BOOL,
4003             .help = "copy read data from backing file into image file",
4004         },{
4005             .name = "detect-zeroes",
4006             .type = QEMU_OPT_STRING,
4007             .help = "try to optimize zero writes (off, on, unmap)",
4008         },{
4009             .name = "stats-account-invalid",
4010             .type = QEMU_OPT_BOOL,
4011             .help = "whether to account for invalid I/O operations "
4012                     "in the statistics",
4013         },{
4014             .name = "stats-account-failed",
4015             .type = QEMU_OPT_BOOL,
4016             .help = "whether to account for failed I/O operations "
4017                     "in the statistics",
4018         },
4019         { /* end of list */ }
4020     },
4021 };
4022 
4023 QemuOptsList qemu_drive_opts = {
4024     .name = "drive",
4025     .head = QTAILQ_HEAD_INITIALIZER(qemu_drive_opts.head),
4026     .desc = {
4027         /*
4028          * no elements => accept any params
4029          * validation will happen later
4030          */
4031         { /* end of list */ }
4032     },
4033 };
4034