xref: /openbmc/qemu/blockdev.c (revision 4c4465ff)
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 "sysemu/replay.h"
63 #include "qemu/cutils.h"
64 #include "qemu/help_option.h"
65 #include "qemu/main-loop.h"
66 #include "qemu/throttle-options.h"
67 
68 QTAILQ_HEAD(, BlockDriverState) monitor_bdrv_states =
69     QTAILQ_HEAD_INITIALIZER(monitor_bdrv_states);
70 
71 void bdrv_set_monitor_owned(BlockDriverState *bs)
72 {
73     QTAILQ_INSERT_TAIL(&monitor_bdrv_states, bs, monitor_list);
74 }
75 
76 static const char *const if_name[IF_COUNT] = {
77     [IF_NONE] = "none",
78     [IF_IDE] = "ide",
79     [IF_SCSI] = "scsi",
80     [IF_FLOPPY] = "floppy",
81     [IF_PFLASH] = "pflash",
82     [IF_MTD] = "mtd",
83     [IF_SD] = "sd",
84     [IF_VIRTIO] = "virtio",
85     [IF_XEN] = "xen",
86 };
87 
88 static int if_max_devs[IF_COUNT] = {
89     /*
90      * Do not change these numbers!  They govern how drive option
91      * index maps to unit and bus.  That mapping is ABI.
92      *
93      * All controllers used to implement if=T drives need to support
94      * if_max_devs[T] units, for any T with if_max_devs[T] != 0.
95      * Otherwise, some index values map to "impossible" bus, unit
96      * values.
97      *
98      * For instance, if you change [IF_SCSI] to 255, -drive
99      * if=scsi,index=12 no longer means bus=1,unit=5, but
100      * bus=0,unit=12.  With an lsi53c895a controller (7 units max),
101      * the drive can't be set up.  Regression.
102      */
103     [IF_IDE] = 2,
104     [IF_SCSI] = 7,
105 };
106 
107 /**
108  * Boards may call this to offer board-by-board overrides
109  * of the default, global values.
110  */
111 void override_max_devs(BlockInterfaceType type, int max_devs)
112 {
113     BlockBackend *blk;
114     DriveInfo *dinfo;
115 
116     if (max_devs <= 0) {
117         return;
118     }
119 
120     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
121         dinfo = blk_legacy_dinfo(blk);
122         if (dinfo->type == type) {
123             fprintf(stderr, "Cannot override units-per-bus property of"
124                     " the %s interface, because a drive of that type has"
125                     " already been added.\n", if_name[type]);
126             g_assert_not_reached();
127         }
128     }
129 
130     if_max_devs[type] = max_devs;
131 }
132 
133 /*
134  * We automatically delete the drive when a device using it gets
135  * unplugged.  Questionable feature, but we can't just drop it.
136  * Device models call blockdev_mark_auto_del() to schedule the
137  * automatic deletion, and generic qdev code calls blockdev_auto_del()
138  * when deletion is actually safe.
139  */
140 void blockdev_mark_auto_del(BlockBackend *blk)
141 {
142     DriveInfo *dinfo = blk_legacy_dinfo(blk);
143     BlockJob *job;
144 
145     if (!dinfo) {
146         return;
147     }
148 
149     for (job = block_job_next(NULL); job; job = block_job_next(job)) {
150         if (block_job_has_bdrv(job, blk_bs(blk))) {
151             AioContext *aio_context = job->job.aio_context;
152             aio_context_acquire(aio_context);
153 
154             job_cancel(&job->job, false);
155 
156             aio_context_release(aio_context);
157         }
158     }
159 
160     dinfo->auto_del = 1;
161 }
162 
163 void blockdev_auto_del(BlockBackend *blk)
164 {
165     DriveInfo *dinfo = blk_legacy_dinfo(blk);
166 
167     if (dinfo && dinfo->auto_del) {
168         monitor_remove_blk(blk);
169         blk_unref(blk);
170     }
171 }
172 
173 /**
174  * Returns the current mapping of how many units per bus
175  * a particular interface can support.
176  *
177  *  A positive integer indicates n units per bus.
178  *  0 implies the mapping has not been established.
179  * -1 indicates an invalid BlockInterfaceType was given.
180  */
181 int drive_get_max_devs(BlockInterfaceType type)
182 {
183     if (type >= IF_IDE && type < IF_COUNT) {
184         return if_max_devs[type];
185     }
186 
187     return -1;
188 }
189 
190 static int drive_index_to_bus_id(BlockInterfaceType type, int index)
191 {
192     int max_devs = if_max_devs[type];
193     return max_devs ? index / max_devs : 0;
194 }
195 
196 static int drive_index_to_unit_id(BlockInterfaceType type, int index)
197 {
198     int max_devs = if_max_devs[type];
199     return max_devs ? index % max_devs : index;
200 }
201 
202 QemuOpts *drive_def(const char *optstr)
203 {
204     return qemu_opts_parse_noisily(qemu_find_opts("drive"), optstr, false);
205 }
206 
207 QemuOpts *drive_add(BlockInterfaceType type, int index, const char *file,
208                     const char *optstr)
209 {
210     QemuOpts *opts;
211 
212     opts = drive_def(optstr);
213     if (!opts) {
214         return NULL;
215     }
216     if (type != IF_DEFAULT) {
217         qemu_opt_set(opts, "if", if_name[type], &error_abort);
218     }
219     if (index >= 0) {
220         qemu_opt_set_number(opts, "index", index, &error_abort);
221     }
222     if (file)
223         qemu_opt_set(opts, "file", file, &error_abort);
224     return opts;
225 }
226 
227 DriveInfo *drive_get(BlockInterfaceType type, int bus, int unit)
228 {
229     BlockBackend *blk;
230     DriveInfo *dinfo;
231 
232     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
233         dinfo = blk_legacy_dinfo(blk);
234         if (dinfo && dinfo->type == type
235             && dinfo->bus == bus && dinfo->unit == unit) {
236             return dinfo;
237         }
238     }
239 
240     return NULL;
241 }
242 
243 void drive_mark_claimed_by_board(void)
244 {
245     BlockBackend *blk;
246     DriveInfo *dinfo;
247 
248     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
249         dinfo = blk_legacy_dinfo(blk);
250         if (dinfo && blk_get_attached_dev(blk)) {
251             dinfo->claimed_by_board = true;
252         }
253     }
254 }
255 
256 void drive_check_orphaned(void)
257 {
258     BlockBackend *blk;
259     DriveInfo *dinfo;
260     Location loc;
261     bool orphans = false;
262 
263     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
264         dinfo = blk_legacy_dinfo(blk);
265         if (dinfo->is_default || dinfo->type == IF_NONE) {
266             continue;
267         }
268         if (!blk_get_attached_dev(blk)) {
269             loc_push_none(&loc);
270             qemu_opts_loc_restore(dinfo->opts);
271             error_report("machine type does not support"
272                          " if=%s,bus=%d,unit=%d",
273                          if_name[dinfo->type], dinfo->bus, dinfo->unit);
274             loc_pop(&loc);
275             orphans = true;
276             continue;
277         }
278         if (!dinfo->claimed_by_board && dinfo->type != IF_VIRTIO) {
279             loc_push_none(&loc);
280             qemu_opts_loc_restore(dinfo->opts);
281             warn_report("bogus if=%s is deprecated, use if=none",
282                         if_name[dinfo->type]);
283             loc_pop(&loc);
284         }
285     }
286 
287     if (orphans) {
288         exit(1);
289     }
290 }
291 
292 DriveInfo *drive_get_by_index(BlockInterfaceType type, int index)
293 {
294     return drive_get(type,
295                      drive_index_to_bus_id(type, index),
296                      drive_index_to_unit_id(type, index));
297 }
298 
299 int drive_get_max_bus(BlockInterfaceType type)
300 {
301     int max_bus;
302     BlockBackend *blk;
303     DriveInfo *dinfo;
304 
305     max_bus = -1;
306     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
307         dinfo = blk_legacy_dinfo(blk);
308         if (dinfo && dinfo->type == type && dinfo->bus > max_bus) {
309             max_bus = dinfo->bus;
310         }
311     }
312     return max_bus;
313 }
314 
315 /* Get a block device.  This should only be used for single-drive devices
316    (e.g. SD/Floppy/MTD).  Multi-disk devices (scsi/ide) should use the
317    appropriate bus.  */
318 DriveInfo *drive_get_next(BlockInterfaceType type)
319 {
320     static int next_block_unit[IF_COUNT];
321 
322     return drive_get(type, 0, next_block_unit[type]++);
323 }
324 
325 static void bdrv_format_print(void *opaque, const char *name)
326 {
327     qemu_printf(" %s", name);
328 }
329 
330 typedef struct {
331     QEMUBH *bh;
332     BlockDriverState *bs;
333 } BDRVPutRefBH;
334 
335 static int parse_block_error_action(const char *buf, bool is_read, Error **errp)
336 {
337     if (!strcmp(buf, "ignore")) {
338         return BLOCKDEV_ON_ERROR_IGNORE;
339     } else if (!is_read && !strcmp(buf, "enospc")) {
340         return BLOCKDEV_ON_ERROR_ENOSPC;
341     } else if (!strcmp(buf, "stop")) {
342         return BLOCKDEV_ON_ERROR_STOP;
343     } else if (!strcmp(buf, "report")) {
344         return BLOCKDEV_ON_ERROR_REPORT;
345     } else {
346         error_setg(errp, "'%s' invalid %s error action",
347                    buf, is_read ? "read" : "write");
348         return -1;
349     }
350 }
351 
352 static bool parse_stats_intervals(BlockAcctStats *stats, QList *intervals,
353                                   Error **errp)
354 {
355     const QListEntry *entry;
356     for (entry = qlist_first(intervals); entry; entry = qlist_next(entry)) {
357         switch (qobject_type(entry->value)) {
358 
359         case QTYPE_QSTRING: {
360             unsigned long long length;
361             const char *str = qstring_get_str(qobject_to(QString,
362                                                          entry->value));
363             if (parse_uint_full(str, &length, 10) == 0 &&
364                 length > 0 && length <= UINT_MAX) {
365                 block_acct_add_interval(stats, (unsigned) length);
366             } else {
367                 error_setg(errp, "Invalid interval length: %s", str);
368                 return false;
369             }
370             break;
371         }
372 
373         case QTYPE_QNUM: {
374             int64_t length = qnum_get_int(qobject_to(QNum, entry->value));
375 
376             if (length > 0 && length <= UINT_MAX) {
377                 block_acct_add_interval(stats, (unsigned) length);
378             } else {
379                 error_setg(errp, "Invalid interval length: %" PRId64, length);
380                 return false;
381             }
382             break;
383         }
384 
385         default:
386             error_setg(errp, "The specification of stats-intervals is invalid");
387             return false;
388         }
389     }
390     return true;
391 }
392 
393 typedef enum { MEDIA_DISK, MEDIA_CDROM } DriveMediaType;
394 
395 /* All parameters but @opts are optional and may be set to NULL. */
396 static void extract_common_blockdev_options(QemuOpts *opts, int *bdrv_flags,
397     const char **throttling_group, ThrottleConfig *throttle_cfg,
398     BlockdevDetectZeroesOptions *detect_zeroes, Error **errp)
399 {
400     Error *local_error = NULL;
401     const char *aio;
402 
403     if (bdrv_flags) {
404         if (qemu_opt_get_bool(opts, "copy-on-read", false)) {
405             *bdrv_flags |= BDRV_O_COPY_ON_READ;
406         }
407 
408         if ((aio = qemu_opt_get(opts, "aio")) != NULL) {
409             if (bdrv_parse_aio(aio, bdrv_flags) < 0) {
410                 error_setg(errp, "invalid aio option");
411                 return;
412             }
413         }
414     }
415 
416     /* disk I/O throttling */
417     if (throttling_group) {
418         *throttling_group = qemu_opt_get(opts, "throttling.group");
419     }
420 
421     if (throttle_cfg) {
422         throttle_config_init(throttle_cfg);
423         throttle_cfg->buckets[THROTTLE_BPS_TOTAL].avg =
424             qemu_opt_get_number(opts, "throttling.bps-total", 0);
425         throttle_cfg->buckets[THROTTLE_BPS_READ].avg  =
426             qemu_opt_get_number(opts, "throttling.bps-read", 0);
427         throttle_cfg->buckets[THROTTLE_BPS_WRITE].avg =
428             qemu_opt_get_number(opts, "throttling.bps-write", 0);
429         throttle_cfg->buckets[THROTTLE_OPS_TOTAL].avg =
430             qemu_opt_get_number(opts, "throttling.iops-total", 0);
431         throttle_cfg->buckets[THROTTLE_OPS_READ].avg =
432             qemu_opt_get_number(opts, "throttling.iops-read", 0);
433         throttle_cfg->buckets[THROTTLE_OPS_WRITE].avg =
434             qemu_opt_get_number(opts, "throttling.iops-write", 0);
435 
436         throttle_cfg->buckets[THROTTLE_BPS_TOTAL].max =
437             qemu_opt_get_number(opts, "throttling.bps-total-max", 0);
438         throttle_cfg->buckets[THROTTLE_BPS_READ].max  =
439             qemu_opt_get_number(opts, "throttling.bps-read-max", 0);
440         throttle_cfg->buckets[THROTTLE_BPS_WRITE].max =
441             qemu_opt_get_number(opts, "throttling.bps-write-max", 0);
442         throttle_cfg->buckets[THROTTLE_OPS_TOTAL].max =
443             qemu_opt_get_number(opts, "throttling.iops-total-max", 0);
444         throttle_cfg->buckets[THROTTLE_OPS_READ].max =
445             qemu_opt_get_number(opts, "throttling.iops-read-max", 0);
446         throttle_cfg->buckets[THROTTLE_OPS_WRITE].max =
447             qemu_opt_get_number(opts, "throttling.iops-write-max", 0);
448 
449         throttle_cfg->buckets[THROTTLE_BPS_TOTAL].burst_length =
450             qemu_opt_get_number(opts, "throttling.bps-total-max-length", 1);
451         throttle_cfg->buckets[THROTTLE_BPS_READ].burst_length  =
452             qemu_opt_get_number(opts, "throttling.bps-read-max-length", 1);
453         throttle_cfg->buckets[THROTTLE_BPS_WRITE].burst_length =
454             qemu_opt_get_number(opts, "throttling.bps-write-max-length", 1);
455         throttle_cfg->buckets[THROTTLE_OPS_TOTAL].burst_length =
456             qemu_opt_get_number(opts, "throttling.iops-total-max-length", 1);
457         throttle_cfg->buckets[THROTTLE_OPS_READ].burst_length =
458             qemu_opt_get_number(opts, "throttling.iops-read-max-length", 1);
459         throttle_cfg->buckets[THROTTLE_OPS_WRITE].burst_length =
460             qemu_opt_get_number(opts, "throttling.iops-write-max-length", 1);
461 
462         throttle_cfg->op_size =
463             qemu_opt_get_number(opts, "throttling.iops-size", 0);
464 
465         if (!throttle_is_valid(throttle_cfg, errp)) {
466             return;
467         }
468     }
469 
470     if (detect_zeroes) {
471         *detect_zeroes =
472             qapi_enum_parse(&BlockdevDetectZeroesOptions_lookup,
473                             qemu_opt_get(opts, "detect-zeroes"),
474                             BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF,
475                             &local_error);
476         if (local_error) {
477             error_propagate(errp, local_error);
478             return;
479         }
480     }
481 }
482 
483 /* Takes the ownership of bs_opts */
484 static BlockBackend *blockdev_init(const char *file, QDict *bs_opts,
485                                    Error **errp)
486 {
487     const char *buf;
488     int bdrv_flags = 0;
489     int on_read_error, on_write_error;
490     bool account_invalid, account_failed;
491     bool writethrough, read_only;
492     BlockBackend *blk;
493     BlockDriverState *bs;
494     ThrottleConfig cfg;
495     int snapshot = 0;
496     Error *error = NULL;
497     QemuOpts *opts;
498     QDict *interval_dict = NULL;
499     QList *interval_list = NULL;
500     const char *id;
501     BlockdevDetectZeroesOptions detect_zeroes =
502         BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF;
503     const char *throttling_group = NULL;
504 
505     /* Check common options by copying from bs_opts to opts, all other options
506      * stay in bs_opts for processing by bdrv_open(). */
507     id = qdict_get_try_str(bs_opts, "id");
508     opts = qemu_opts_create(&qemu_common_drive_opts, id, 1, errp);
509     if (!opts) {
510         goto err_no_opts;
511     }
512 
513     if (!qemu_opts_absorb_qdict(opts, bs_opts, errp)) {
514         goto early_err;
515     }
516 
517     if (id) {
518         qdict_del(bs_opts, "id");
519     }
520 
521     /* extract parameters */
522     snapshot = qemu_opt_get_bool(opts, "snapshot", 0);
523 
524     account_invalid = qemu_opt_get_bool(opts, "stats-account-invalid", true);
525     account_failed = qemu_opt_get_bool(opts, "stats-account-failed", true);
526 
527     writethrough = !qemu_opt_get_bool(opts, BDRV_OPT_CACHE_WB, true);
528 
529     id = qemu_opts_id(opts);
530 
531     qdict_extract_subqdict(bs_opts, &interval_dict, "stats-intervals.");
532     qdict_array_split(interval_dict, &interval_list);
533 
534     if (qdict_size(interval_dict) != 0) {
535         error_setg(errp, "Invalid option stats-intervals.%s",
536                    qdict_first(interval_dict)->key);
537         goto early_err;
538     }
539 
540     extract_common_blockdev_options(opts, &bdrv_flags, &throttling_group, &cfg,
541                                     &detect_zeroes, &error);
542     if (error) {
543         error_propagate(errp, error);
544         goto early_err;
545     }
546 
547     if ((buf = qemu_opt_get(opts, "format")) != NULL) {
548         if (is_help_option(buf)) {
549             qemu_printf("Supported formats:");
550             bdrv_iterate_format(bdrv_format_print, NULL, false);
551             qemu_printf("\nSupported formats (read-only):");
552             bdrv_iterate_format(bdrv_format_print, NULL, true);
553             qemu_printf("\n");
554             goto early_err;
555         }
556 
557         if (qdict_haskey(bs_opts, "driver")) {
558             error_setg(errp, "Cannot specify both 'driver' and 'format'");
559             goto early_err;
560         }
561         qdict_put_str(bs_opts, "driver", buf);
562     }
563 
564     on_write_error = BLOCKDEV_ON_ERROR_ENOSPC;
565     if ((buf = qemu_opt_get(opts, "werror")) != NULL) {
566         on_write_error = parse_block_error_action(buf, 0, &error);
567         if (error) {
568             error_propagate(errp, error);
569             goto early_err;
570         }
571     }
572 
573     on_read_error = BLOCKDEV_ON_ERROR_REPORT;
574     if ((buf = qemu_opt_get(opts, "rerror")) != NULL) {
575         on_read_error = parse_block_error_action(buf, 1, &error);
576         if (error) {
577             error_propagate(errp, error);
578             goto early_err;
579         }
580     }
581 
582     if (snapshot) {
583         bdrv_flags |= BDRV_O_SNAPSHOT;
584     }
585 
586     read_only = qemu_opt_get_bool(opts, BDRV_OPT_READ_ONLY, false);
587 
588     /* init */
589     if ((!file || !*file) && !qdict_size(bs_opts)) {
590         BlockBackendRootState *blk_rs;
591 
592         blk = blk_new(qemu_get_aio_context(), 0, BLK_PERM_ALL);
593         blk_rs = blk_get_root_state(blk);
594         blk_rs->open_flags    = bdrv_flags;
595         blk_rs->read_only     = read_only;
596         blk_rs->detect_zeroes = detect_zeroes;
597 
598         qobject_unref(bs_opts);
599     } else {
600         if (file && !*file) {
601             file = NULL;
602         }
603 
604         /* bdrv_open() defaults to the values in bdrv_flags (for compatibility
605          * with other callers) rather than what we want as the real defaults.
606          * Apply the defaults here instead. */
607         qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_DIRECT, "off");
608         qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_NO_FLUSH, "off");
609         qdict_set_default_str(bs_opts, BDRV_OPT_READ_ONLY,
610                               read_only ? "on" : "off");
611         qdict_set_default_str(bs_opts, BDRV_OPT_AUTO_READ_ONLY, "on");
612         assert((bdrv_flags & BDRV_O_CACHE_MASK) == 0);
613 
614         if (runstate_check(RUN_STATE_INMIGRATE)) {
615             bdrv_flags |= BDRV_O_INACTIVE;
616         }
617 
618         blk = blk_new_open(file, NULL, bs_opts, bdrv_flags, errp);
619         if (!blk) {
620             goto err_no_bs_opts;
621         }
622         bs = blk_bs(blk);
623 
624         bs->detect_zeroes = detect_zeroes;
625 
626         block_acct_setup(blk_get_stats(blk), account_invalid, account_failed);
627 
628         if (!parse_stats_intervals(blk_get_stats(blk), interval_list, errp)) {
629             blk_unref(blk);
630             blk = NULL;
631             goto err_no_bs_opts;
632         }
633     }
634 
635     /* disk I/O throttling */
636     if (throttle_enabled(&cfg)) {
637         if (!throttling_group) {
638             throttling_group = id;
639         }
640         blk_io_limits_enable(blk, throttling_group);
641         blk_set_io_limits(blk, &cfg);
642     }
643 
644     blk_set_enable_write_cache(blk, !writethrough);
645     blk_set_on_error(blk, on_read_error, on_write_error);
646 
647     if (!monitor_add_blk(blk, id, errp)) {
648         blk_unref(blk);
649         blk = NULL;
650         goto err_no_bs_opts;
651     }
652 
653 err_no_bs_opts:
654     qemu_opts_del(opts);
655     qobject_unref(interval_dict);
656     qobject_unref(interval_list);
657     return blk;
658 
659 early_err:
660     qemu_opts_del(opts);
661     qobject_unref(interval_dict);
662     qobject_unref(interval_list);
663 err_no_opts:
664     qobject_unref(bs_opts);
665     return NULL;
666 }
667 
668 /* Takes the ownership of bs_opts */
669 BlockDriverState *bds_tree_init(QDict *bs_opts, Error **errp)
670 {
671     int bdrv_flags = 0;
672 
673     /* bdrv_open() defaults to the values in bdrv_flags (for compatibility
674      * with other callers) rather than what we want as the real defaults.
675      * Apply the defaults here instead. */
676     qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_DIRECT, "off");
677     qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_NO_FLUSH, "off");
678     qdict_set_default_str(bs_opts, BDRV_OPT_READ_ONLY, "off");
679 
680     if (runstate_check(RUN_STATE_INMIGRATE)) {
681         bdrv_flags |= BDRV_O_INACTIVE;
682     }
683 
684     return bdrv_open(NULL, NULL, bs_opts, bdrv_flags, errp);
685 }
686 
687 void blockdev_close_all_bdrv_states(void)
688 {
689     BlockDriverState *bs, *next_bs;
690 
691     QTAILQ_FOREACH_SAFE(bs, &monitor_bdrv_states, monitor_list, next_bs) {
692         AioContext *ctx = bdrv_get_aio_context(bs);
693 
694         aio_context_acquire(ctx);
695         bdrv_unref(bs);
696         aio_context_release(ctx);
697     }
698 }
699 
700 /* Iterates over the list of monitor-owned BlockDriverStates */
701 BlockDriverState *bdrv_next_monitor_owned(BlockDriverState *bs)
702 {
703     return bs ? QTAILQ_NEXT(bs, monitor_list)
704               : QTAILQ_FIRST(&monitor_bdrv_states);
705 }
706 
707 static bool qemu_opt_rename(QemuOpts *opts, const char *from, const char *to,
708                             Error **errp)
709 {
710     const char *value;
711 
712     value = qemu_opt_get(opts, from);
713     if (value) {
714         if (qemu_opt_find(opts, to)) {
715             error_setg(errp, "'%s' and its alias '%s' can't be used at the "
716                        "same time", to, from);
717             return false;
718         }
719     }
720 
721     /* rename all items in opts */
722     while ((value = qemu_opt_get(opts, from))) {
723         qemu_opt_set(opts, to, value, &error_abort);
724         qemu_opt_unset(opts, from);
725     }
726     return true;
727 }
728 
729 QemuOptsList qemu_legacy_drive_opts = {
730     .name = "drive",
731     .head = QTAILQ_HEAD_INITIALIZER(qemu_legacy_drive_opts.head),
732     .desc = {
733         {
734             .name = "bus",
735             .type = QEMU_OPT_NUMBER,
736             .help = "bus number",
737         },{
738             .name = "unit",
739             .type = QEMU_OPT_NUMBER,
740             .help = "unit number (i.e. lun for scsi)",
741         },{
742             .name = "index",
743             .type = QEMU_OPT_NUMBER,
744             .help = "index number",
745         },{
746             .name = "media",
747             .type = QEMU_OPT_STRING,
748             .help = "media type (disk, cdrom)",
749         },{
750             .name = "if",
751             .type = QEMU_OPT_STRING,
752             .help = "interface (ide, scsi, sd, mtd, floppy, pflash, virtio)",
753         },{
754             .name = "file",
755             .type = QEMU_OPT_STRING,
756             .help = "file name",
757         },
758 
759         /* Options that are passed on, but have special semantics with -drive */
760         {
761             .name = BDRV_OPT_READ_ONLY,
762             .type = QEMU_OPT_BOOL,
763             .help = "open drive file as read-only",
764         },{
765             .name = "rerror",
766             .type = QEMU_OPT_STRING,
767             .help = "read error action",
768         },{
769             .name = "werror",
770             .type = QEMU_OPT_STRING,
771             .help = "write error action",
772         },{
773             .name = "copy-on-read",
774             .type = QEMU_OPT_BOOL,
775             .help = "copy read data from backing file into image file",
776         },
777 
778         { /* end of list */ }
779     },
780 };
781 
782 DriveInfo *drive_new(QemuOpts *all_opts, BlockInterfaceType block_default_type,
783                      Error **errp)
784 {
785     const char *value;
786     BlockBackend *blk;
787     DriveInfo *dinfo = NULL;
788     QDict *bs_opts;
789     QemuOpts *legacy_opts;
790     DriveMediaType media = MEDIA_DISK;
791     BlockInterfaceType type;
792     int max_devs, bus_id, unit_id, index;
793     const char *werror, *rerror;
794     bool read_only = false;
795     bool copy_on_read;
796     const char *filename;
797     int i;
798 
799     /* Change legacy command line options into QMP ones */
800     static const struct {
801         const char *from;
802         const char *to;
803     } opt_renames[] = {
804         { "iops",           "throttling.iops-total" },
805         { "iops_rd",        "throttling.iops-read" },
806         { "iops_wr",        "throttling.iops-write" },
807 
808         { "bps",            "throttling.bps-total" },
809         { "bps_rd",         "throttling.bps-read" },
810         { "bps_wr",         "throttling.bps-write" },
811 
812         { "iops_max",       "throttling.iops-total-max" },
813         { "iops_rd_max",    "throttling.iops-read-max" },
814         { "iops_wr_max",    "throttling.iops-write-max" },
815 
816         { "bps_max",        "throttling.bps-total-max" },
817         { "bps_rd_max",     "throttling.bps-read-max" },
818         { "bps_wr_max",     "throttling.bps-write-max" },
819 
820         { "iops_size",      "throttling.iops-size" },
821 
822         { "group",          "throttling.group" },
823 
824         { "readonly",       BDRV_OPT_READ_ONLY },
825     };
826 
827     for (i = 0; i < ARRAY_SIZE(opt_renames); i++) {
828         if (!qemu_opt_rename(all_opts, opt_renames[i].from,
829                              opt_renames[i].to, errp)) {
830             return NULL;
831         }
832     }
833 
834     value = qemu_opt_get(all_opts, "cache");
835     if (value) {
836         int flags = 0;
837         bool writethrough;
838 
839         if (bdrv_parse_cache_mode(value, &flags, &writethrough) != 0) {
840             error_setg(errp, "invalid cache option");
841             return NULL;
842         }
843 
844         /* Specific options take precedence */
845         if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_WB)) {
846             qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_WB,
847                               !writethrough, &error_abort);
848         }
849         if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_DIRECT)) {
850             qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_DIRECT,
851                               !!(flags & BDRV_O_NOCACHE), &error_abort);
852         }
853         if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_NO_FLUSH)) {
854             qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_NO_FLUSH,
855                               !!(flags & BDRV_O_NO_FLUSH), &error_abort);
856         }
857         qemu_opt_unset(all_opts, "cache");
858     }
859 
860     /* Get a QDict for processing the options */
861     bs_opts = qdict_new();
862     qemu_opts_to_qdict(all_opts, bs_opts);
863 
864     legacy_opts = qemu_opts_create(&qemu_legacy_drive_opts, NULL, 0,
865                                    &error_abort);
866     if (!qemu_opts_absorb_qdict(legacy_opts, bs_opts, errp)) {
867         goto fail;
868     }
869 
870     /* Media type */
871     value = qemu_opt_get(legacy_opts, "media");
872     if (value) {
873         if (!strcmp(value, "disk")) {
874             media = MEDIA_DISK;
875         } else if (!strcmp(value, "cdrom")) {
876             media = MEDIA_CDROM;
877             read_only = true;
878         } else {
879             error_setg(errp, "'%s' invalid media", value);
880             goto fail;
881         }
882     }
883 
884     /* copy-on-read is disabled with a warning for read-only devices */
885     read_only |= qemu_opt_get_bool(legacy_opts, BDRV_OPT_READ_ONLY, false);
886     copy_on_read = qemu_opt_get_bool(legacy_opts, "copy-on-read", false);
887 
888     if (read_only && copy_on_read) {
889         warn_report("disabling copy-on-read on read-only drive");
890         copy_on_read = false;
891     }
892 
893     qdict_put_str(bs_opts, BDRV_OPT_READ_ONLY, read_only ? "on" : "off");
894     qdict_put_str(bs_opts, "copy-on-read", copy_on_read ? "on" : "off");
895 
896     /* Controller type */
897     value = qemu_opt_get(legacy_opts, "if");
898     if (value) {
899         for (type = 0;
900              type < IF_COUNT && strcmp(value, if_name[type]);
901              type++) {
902         }
903         if (type == IF_COUNT) {
904             error_setg(errp, "unsupported bus type '%s'", value);
905             goto fail;
906         }
907     } else {
908         type = block_default_type;
909     }
910 
911     /* Device address specified by bus/unit or index.
912      * If none was specified, try to find the first free one. */
913     bus_id  = qemu_opt_get_number(legacy_opts, "bus", 0);
914     unit_id = qemu_opt_get_number(legacy_opts, "unit", -1);
915     index   = qemu_opt_get_number(legacy_opts, "index", -1);
916 
917     max_devs = if_max_devs[type];
918 
919     if (index != -1) {
920         if (bus_id != 0 || unit_id != -1) {
921             error_setg(errp, "index cannot be used with bus and unit");
922             goto fail;
923         }
924         bus_id = drive_index_to_bus_id(type, index);
925         unit_id = drive_index_to_unit_id(type, index);
926     }
927 
928     if (unit_id == -1) {
929        unit_id = 0;
930        while (drive_get(type, bus_id, unit_id) != NULL) {
931            unit_id++;
932            if (max_devs && unit_id >= max_devs) {
933                unit_id -= max_devs;
934                bus_id++;
935            }
936        }
937     }
938 
939     if (max_devs && unit_id >= max_devs) {
940         error_setg(errp, "unit %d too big (max is %d)", unit_id, max_devs - 1);
941         goto fail;
942     }
943 
944     if (drive_get(type, bus_id, unit_id) != NULL) {
945         error_setg(errp, "drive with bus=%d, unit=%d (index=%d) exists",
946                    bus_id, unit_id, index);
947         goto fail;
948     }
949 
950     /* no id supplied -> create one */
951     if (qemu_opts_id(all_opts) == NULL) {
952         char *new_id;
953         const char *mediastr = "";
954         if (type == IF_IDE || type == IF_SCSI) {
955             mediastr = (media == MEDIA_CDROM) ? "-cd" : "-hd";
956         }
957         if (max_devs) {
958             new_id = g_strdup_printf("%s%i%s%i", if_name[type], bus_id,
959                                      mediastr, unit_id);
960         } else {
961             new_id = g_strdup_printf("%s%s%i", if_name[type],
962                                      mediastr, unit_id);
963         }
964         qdict_put_str(bs_opts, "id", new_id);
965         g_free(new_id);
966     }
967 
968     /* Add virtio block device */
969     if (type == IF_VIRTIO) {
970         QemuOpts *devopts;
971         devopts = qemu_opts_create(qemu_find_opts("device"), NULL, 0,
972                                    &error_abort);
973         if (arch_type == QEMU_ARCH_S390X) {
974             qemu_opt_set(devopts, "driver", "virtio-blk-ccw", &error_abort);
975         } else {
976             qemu_opt_set(devopts, "driver", "virtio-blk-pci", &error_abort);
977         }
978         qemu_opt_set(devopts, "drive", qdict_get_str(bs_opts, "id"),
979                      &error_abort);
980     }
981 
982     filename = qemu_opt_get(legacy_opts, "file");
983 
984     /* Check werror/rerror compatibility with if=... */
985     werror = qemu_opt_get(legacy_opts, "werror");
986     if (werror != NULL) {
987         if (type != IF_IDE && type != IF_SCSI && type != IF_VIRTIO &&
988             type != IF_NONE) {
989             error_setg(errp, "werror is not supported by this bus type");
990             goto fail;
991         }
992         qdict_put_str(bs_opts, "werror", werror);
993     }
994 
995     rerror = qemu_opt_get(legacy_opts, "rerror");
996     if (rerror != NULL) {
997         if (type != IF_IDE && type != IF_VIRTIO && type != IF_SCSI &&
998             type != IF_NONE) {
999             error_setg(errp, "rerror is not supported by this bus type");
1000             goto fail;
1001         }
1002         qdict_put_str(bs_opts, "rerror", rerror);
1003     }
1004 
1005     /* Actual block device init: Functionality shared with blockdev-add */
1006     blk = blockdev_init(filename, bs_opts, errp);
1007     bs_opts = NULL;
1008     if (!blk) {
1009         goto fail;
1010     }
1011 
1012     /* Create legacy DriveInfo */
1013     dinfo = g_malloc0(sizeof(*dinfo));
1014     dinfo->opts = all_opts;
1015 
1016     dinfo->type = type;
1017     dinfo->bus = bus_id;
1018     dinfo->unit = unit_id;
1019 
1020     blk_set_legacy_dinfo(blk, dinfo);
1021 
1022     switch(type) {
1023     case IF_IDE:
1024     case IF_SCSI:
1025     case IF_XEN:
1026     case IF_NONE:
1027         dinfo->media_cd = media == MEDIA_CDROM;
1028         break;
1029     default:
1030         break;
1031     }
1032 
1033 fail:
1034     qemu_opts_del(legacy_opts);
1035     qobject_unref(bs_opts);
1036     return dinfo;
1037 }
1038 
1039 static BlockDriverState *qmp_get_root_bs(const char *name, Error **errp)
1040 {
1041     BlockDriverState *bs;
1042 
1043     bs = bdrv_lookup_bs(name, name, errp);
1044     if (bs == NULL) {
1045         return NULL;
1046     }
1047 
1048     if (!bdrv_is_root_node(bs)) {
1049         error_setg(errp, "Need a root block node");
1050         return NULL;
1051     }
1052 
1053     if (!bdrv_is_inserted(bs)) {
1054         error_setg(errp, "Device has no medium");
1055         return NULL;
1056     }
1057 
1058     return bs;
1059 }
1060 
1061 static void blockdev_do_action(TransactionAction *action, Error **errp)
1062 {
1063     TransactionActionList list;
1064 
1065     list.value = action;
1066     list.next = NULL;
1067     qmp_transaction(&list, false, NULL, errp);
1068 }
1069 
1070 void qmp_blockdev_snapshot_sync(bool has_device, const char *device,
1071                                 bool has_node_name, const char *node_name,
1072                                 const char *snapshot_file,
1073                                 bool has_snapshot_node_name,
1074                                 const char *snapshot_node_name,
1075                                 bool has_format, const char *format,
1076                                 bool has_mode, NewImageMode mode, Error **errp)
1077 {
1078     BlockdevSnapshotSync snapshot = {
1079         .has_device = has_device,
1080         .device = (char *) device,
1081         .has_node_name = has_node_name,
1082         .node_name = (char *) node_name,
1083         .snapshot_file = (char *) snapshot_file,
1084         .has_snapshot_node_name = has_snapshot_node_name,
1085         .snapshot_node_name = (char *) snapshot_node_name,
1086         .has_format = has_format,
1087         .format = (char *) format,
1088         .has_mode = has_mode,
1089         .mode = mode,
1090     };
1091     TransactionAction action = {
1092         .type = TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC,
1093         .u.blockdev_snapshot_sync.data = &snapshot,
1094     };
1095     blockdev_do_action(&action, errp);
1096 }
1097 
1098 void qmp_blockdev_snapshot(const char *node, const char *overlay,
1099                            Error **errp)
1100 {
1101     BlockdevSnapshot snapshot_data = {
1102         .node = (char *) node,
1103         .overlay = (char *) overlay
1104     };
1105     TransactionAction action = {
1106         .type = TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT,
1107         .u.blockdev_snapshot.data = &snapshot_data,
1108     };
1109     blockdev_do_action(&action, errp);
1110 }
1111 
1112 void qmp_blockdev_snapshot_internal_sync(const char *device,
1113                                          const char *name,
1114                                          Error **errp)
1115 {
1116     BlockdevSnapshotInternal snapshot = {
1117         .device = (char *) device,
1118         .name = (char *) name
1119     };
1120     TransactionAction action = {
1121         .type = TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC,
1122         .u.blockdev_snapshot_internal_sync.data = &snapshot,
1123     };
1124     blockdev_do_action(&action, errp);
1125 }
1126 
1127 SnapshotInfo *qmp_blockdev_snapshot_delete_internal_sync(const char *device,
1128                                                          bool has_id,
1129                                                          const char *id,
1130                                                          bool has_name,
1131                                                          const char *name,
1132                                                          Error **errp)
1133 {
1134     BlockDriverState *bs;
1135     AioContext *aio_context;
1136     QEMUSnapshotInfo sn;
1137     Error *local_err = NULL;
1138     SnapshotInfo *info = NULL;
1139     int ret;
1140 
1141     bs = qmp_get_root_bs(device, errp);
1142     if (!bs) {
1143         return NULL;
1144     }
1145     aio_context = bdrv_get_aio_context(bs);
1146     aio_context_acquire(aio_context);
1147 
1148     if (!has_id) {
1149         id = NULL;
1150     }
1151 
1152     if (!has_name) {
1153         name = NULL;
1154     }
1155 
1156     if (!id && !name) {
1157         error_setg(errp, "Name or id must be provided");
1158         goto out_aio_context;
1159     }
1160 
1161     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_INTERNAL_SNAPSHOT_DELETE, errp)) {
1162         goto out_aio_context;
1163     }
1164 
1165     ret = bdrv_snapshot_find_by_id_and_name(bs, id, name, &sn, &local_err);
1166     if (local_err) {
1167         error_propagate(errp, local_err);
1168         goto out_aio_context;
1169     }
1170     if (!ret) {
1171         error_setg(errp,
1172                    "Snapshot with id '%s' and name '%s' does not exist on "
1173                    "device '%s'",
1174                    STR_OR_NULL(id), STR_OR_NULL(name), device);
1175         goto out_aio_context;
1176     }
1177 
1178     bdrv_snapshot_delete(bs, id, name, &local_err);
1179     if (local_err) {
1180         error_propagate(errp, local_err);
1181         goto out_aio_context;
1182     }
1183 
1184     aio_context_release(aio_context);
1185 
1186     info = g_new0(SnapshotInfo, 1);
1187     info->id = g_strdup(sn.id_str);
1188     info->name = g_strdup(sn.name);
1189     info->date_nsec = sn.date_nsec;
1190     info->date_sec = sn.date_sec;
1191     info->vm_state_size = sn.vm_state_size;
1192     info->vm_clock_nsec = sn.vm_clock_nsec % 1000000000;
1193     info->vm_clock_sec = sn.vm_clock_nsec / 1000000000;
1194     if (sn.icount != -1ULL) {
1195         info->icount = sn.icount;
1196         info->has_icount = true;
1197     }
1198 
1199     return info;
1200 
1201 out_aio_context:
1202     aio_context_release(aio_context);
1203     return NULL;
1204 }
1205 
1206 /* New and old BlockDriverState structs for atomic group operations */
1207 
1208 typedef struct BlkActionState BlkActionState;
1209 
1210 /**
1211  * BlkActionOps:
1212  * Table of operations that define an Action.
1213  *
1214  * @instance_size: Size of state struct, in bytes.
1215  * @prepare: Prepare the work, must NOT be NULL.
1216  * @commit: Commit the changes, can be NULL.
1217  * @abort: Abort the changes on fail, can be NULL.
1218  * @clean: Clean up resources after all transaction actions have called
1219  *         commit() or abort(). Can be NULL.
1220  *
1221  * Only prepare() may fail. In a single transaction, only one of commit() or
1222  * abort() will be called. clean() will always be called if it is present.
1223  */
1224 typedef struct BlkActionOps {
1225     size_t instance_size;
1226     void (*prepare)(BlkActionState *common, Error **errp);
1227     void (*commit)(BlkActionState *common);
1228     void (*abort)(BlkActionState *common);
1229     void (*clean)(BlkActionState *common);
1230 } BlkActionOps;
1231 
1232 /**
1233  * BlkActionState:
1234  * Describes one Action's state within a Transaction.
1235  *
1236  * @action: QAPI-defined enum identifying which Action to perform.
1237  * @ops: Table of ActionOps this Action can perform.
1238  * @block_job_txn: Transaction which this action belongs to.
1239  * @entry: List membership for all Actions in this Transaction.
1240  *
1241  * This structure must be arranged as first member in a subclassed type,
1242  * assuming that the compiler will also arrange it to the same offsets as the
1243  * base class.
1244  */
1245 struct BlkActionState {
1246     TransactionAction *action;
1247     const BlkActionOps *ops;
1248     JobTxn *block_job_txn;
1249     TransactionProperties *txn_props;
1250     QTAILQ_ENTRY(BlkActionState) entry;
1251 };
1252 
1253 /* internal snapshot private data */
1254 typedef struct InternalSnapshotState {
1255     BlkActionState common;
1256     BlockDriverState *bs;
1257     QEMUSnapshotInfo sn;
1258     bool created;
1259 } InternalSnapshotState;
1260 
1261 
1262 static int action_check_completion_mode(BlkActionState *s, Error **errp)
1263 {
1264     if (s->txn_props->completion_mode != ACTION_COMPLETION_MODE_INDIVIDUAL) {
1265         error_setg(errp,
1266                    "Action '%s' does not support Transaction property "
1267                    "completion-mode = %s",
1268                    TransactionActionKind_str(s->action->type),
1269                    ActionCompletionMode_str(s->txn_props->completion_mode));
1270         return -1;
1271     }
1272     return 0;
1273 }
1274 
1275 static void internal_snapshot_prepare(BlkActionState *common,
1276                                       Error **errp)
1277 {
1278     Error *local_err = NULL;
1279     const char *device;
1280     const char *name;
1281     BlockDriverState *bs;
1282     QEMUSnapshotInfo old_sn, *sn;
1283     bool ret;
1284     qemu_timeval tv;
1285     BlockdevSnapshotInternal *internal;
1286     InternalSnapshotState *state;
1287     AioContext *aio_context;
1288     int ret1;
1289 
1290     g_assert(common->action->type ==
1291              TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC);
1292     internal = common->action->u.blockdev_snapshot_internal_sync.data;
1293     state = DO_UPCAST(InternalSnapshotState, common, common);
1294 
1295     /* 1. parse input */
1296     device = internal->device;
1297     name = internal->name;
1298 
1299     /* 2. check for validation */
1300     if (action_check_completion_mode(common, errp) < 0) {
1301         return;
1302     }
1303 
1304     bs = qmp_get_root_bs(device, errp);
1305     if (!bs) {
1306         return;
1307     }
1308 
1309     aio_context = bdrv_get_aio_context(bs);
1310     aio_context_acquire(aio_context);
1311 
1312     state->bs = bs;
1313 
1314     /* Paired with .clean() */
1315     bdrv_drained_begin(bs);
1316 
1317     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_INTERNAL_SNAPSHOT, errp)) {
1318         goto out;
1319     }
1320 
1321     if (bdrv_is_read_only(bs)) {
1322         error_setg(errp, "Device '%s' is read only", device);
1323         goto out;
1324     }
1325 
1326     if (!bdrv_can_snapshot(bs)) {
1327         error_setg(errp, "Block format '%s' used by device '%s' "
1328                    "does not support internal snapshots",
1329                    bs->drv->format_name, device);
1330         goto out;
1331     }
1332 
1333     if (!strlen(name)) {
1334         error_setg(errp, "Name is empty");
1335         goto out;
1336     }
1337 
1338     /* check whether a snapshot with name exist */
1339     ret = bdrv_snapshot_find_by_id_and_name(bs, NULL, name, &old_sn,
1340                                             &local_err);
1341     if (local_err) {
1342         error_propagate(errp, local_err);
1343         goto out;
1344     } else if (ret) {
1345         error_setg(errp,
1346                    "Snapshot with name '%s' already exists on device '%s'",
1347                    name, device);
1348         goto out;
1349     }
1350 
1351     /* 3. take the snapshot */
1352     sn = &state->sn;
1353     pstrcpy(sn->name, sizeof(sn->name), name);
1354     qemu_gettimeofday(&tv);
1355     sn->date_sec = tv.tv_sec;
1356     sn->date_nsec = tv.tv_usec * 1000;
1357     sn->vm_clock_nsec = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL);
1358     if (replay_mode != REPLAY_MODE_NONE) {
1359         sn->icount = replay_get_current_icount();
1360     } else {
1361         sn->icount = -1ULL;
1362     }
1363 
1364     ret1 = bdrv_snapshot_create(bs, sn);
1365     if (ret1 < 0) {
1366         error_setg_errno(errp, -ret1,
1367                          "Failed to create snapshot '%s' on device '%s'",
1368                          name, device);
1369         goto out;
1370     }
1371 
1372     /* 4. succeed, mark a snapshot is created */
1373     state->created = true;
1374 
1375 out:
1376     aio_context_release(aio_context);
1377 }
1378 
1379 static void internal_snapshot_abort(BlkActionState *common)
1380 {
1381     InternalSnapshotState *state =
1382                              DO_UPCAST(InternalSnapshotState, common, common);
1383     BlockDriverState *bs = state->bs;
1384     QEMUSnapshotInfo *sn = &state->sn;
1385     AioContext *aio_context;
1386     Error *local_error = NULL;
1387 
1388     if (!state->created) {
1389         return;
1390     }
1391 
1392     aio_context = bdrv_get_aio_context(state->bs);
1393     aio_context_acquire(aio_context);
1394 
1395     if (bdrv_snapshot_delete(bs, sn->id_str, sn->name, &local_error) < 0) {
1396         error_reportf_err(local_error,
1397                           "Failed to delete snapshot with id '%s' and "
1398                           "name '%s' on device '%s' in abort: ",
1399                           sn->id_str, sn->name,
1400                           bdrv_get_device_name(bs));
1401     }
1402 
1403     aio_context_release(aio_context);
1404 }
1405 
1406 static void internal_snapshot_clean(BlkActionState *common)
1407 {
1408     InternalSnapshotState *state = DO_UPCAST(InternalSnapshotState,
1409                                              common, common);
1410     AioContext *aio_context;
1411 
1412     if (!state->bs) {
1413         return;
1414     }
1415 
1416     aio_context = bdrv_get_aio_context(state->bs);
1417     aio_context_acquire(aio_context);
1418 
1419     bdrv_drained_end(state->bs);
1420 
1421     aio_context_release(aio_context);
1422 }
1423 
1424 /* external snapshot private data */
1425 typedef struct ExternalSnapshotState {
1426     BlkActionState common;
1427     BlockDriverState *old_bs;
1428     BlockDriverState *new_bs;
1429     bool overlay_appended;
1430 } ExternalSnapshotState;
1431 
1432 static void external_snapshot_prepare(BlkActionState *common,
1433                                       Error **errp)
1434 {
1435     int flags = 0;
1436     QDict *options = NULL;
1437     Error *local_err = NULL;
1438     /* Device and node name of the image to generate the snapshot from */
1439     const char *device;
1440     const char *node_name;
1441     /* Reference to the new image (for 'blockdev-snapshot') */
1442     const char *snapshot_ref;
1443     /* File name of the new image (for 'blockdev-snapshot-sync') */
1444     const char *new_image_file;
1445     ExternalSnapshotState *state =
1446                              DO_UPCAST(ExternalSnapshotState, common, common);
1447     TransactionAction *action = common->action;
1448     AioContext *aio_context;
1449     uint64_t perm, shared;
1450 
1451     /* 'blockdev-snapshot' and 'blockdev-snapshot-sync' have similar
1452      * purpose but a different set of parameters */
1453     switch (action->type) {
1454     case TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT:
1455         {
1456             BlockdevSnapshot *s = action->u.blockdev_snapshot.data;
1457             device = s->node;
1458             node_name = s->node;
1459             new_image_file = NULL;
1460             snapshot_ref = s->overlay;
1461         }
1462         break;
1463     case TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC:
1464         {
1465             BlockdevSnapshotSync *s = action->u.blockdev_snapshot_sync.data;
1466             device = s->has_device ? s->device : NULL;
1467             node_name = s->has_node_name ? s->node_name : NULL;
1468             new_image_file = s->snapshot_file;
1469             snapshot_ref = NULL;
1470         }
1471         break;
1472     default:
1473         g_assert_not_reached();
1474     }
1475 
1476     /* start processing */
1477     if (action_check_completion_mode(common, errp) < 0) {
1478         return;
1479     }
1480 
1481     state->old_bs = bdrv_lookup_bs(device, node_name, errp);
1482     if (!state->old_bs) {
1483         return;
1484     }
1485 
1486     aio_context = bdrv_get_aio_context(state->old_bs);
1487     aio_context_acquire(aio_context);
1488 
1489     /* Paired with .clean() */
1490     bdrv_drained_begin(state->old_bs);
1491 
1492     if (!bdrv_is_inserted(state->old_bs)) {
1493         error_setg(errp, QERR_DEVICE_HAS_NO_MEDIUM, device);
1494         goto out;
1495     }
1496 
1497     if (bdrv_op_is_blocked(state->old_bs,
1498                            BLOCK_OP_TYPE_EXTERNAL_SNAPSHOT, errp)) {
1499         goto out;
1500     }
1501 
1502     if (!bdrv_is_read_only(state->old_bs)) {
1503         if (bdrv_flush(state->old_bs)) {
1504             error_setg(errp, QERR_IO_ERROR);
1505             goto out;
1506         }
1507     }
1508 
1509     if (action->type == TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC) {
1510         BlockdevSnapshotSync *s = action->u.blockdev_snapshot_sync.data;
1511         const char *format = s->has_format ? s->format : "qcow2";
1512         enum NewImageMode mode;
1513         const char *snapshot_node_name =
1514             s->has_snapshot_node_name ? s->snapshot_node_name : NULL;
1515 
1516         if (node_name && !snapshot_node_name) {
1517             error_setg(errp, "New overlay node name missing");
1518             goto out;
1519         }
1520 
1521         if (snapshot_node_name &&
1522             bdrv_lookup_bs(snapshot_node_name, snapshot_node_name, NULL)) {
1523             error_setg(errp, "New overlay node name already in use");
1524             goto out;
1525         }
1526 
1527         flags = state->old_bs->open_flags;
1528         flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_COPY_ON_READ);
1529         flags |= BDRV_O_NO_BACKING;
1530 
1531         /* create new image w/backing file */
1532         mode = s->has_mode ? s->mode : NEW_IMAGE_MODE_ABSOLUTE_PATHS;
1533         if (mode != NEW_IMAGE_MODE_EXISTING) {
1534             int64_t size = bdrv_getlength(state->old_bs);
1535             if (size < 0) {
1536                 error_setg_errno(errp, -size, "bdrv_getlength failed");
1537                 goto out;
1538             }
1539             bdrv_refresh_filename(state->old_bs);
1540             bdrv_img_create(new_image_file, format,
1541                             state->old_bs->filename,
1542                             state->old_bs->drv->format_name,
1543                             NULL, size, flags, false, &local_err);
1544             if (local_err) {
1545                 error_propagate(errp, local_err);
1546                 goto out;
1547             }
1548         }
1549 
1550         options = qdict_new();
1551         if (snapshot_node_name) {
1552             qdict_put_str(options, "node-name", snapshot_node_name);
1553         }
1554         qdict_put_str(options, "driver", format);
1555     }
1556 
1557     state->new_bs = bdrv_open(new_image_file, snapshot_ref, options, flags,
1558                               errp);
1559     /* We will manually add the backing_hd field to the bs later */
1560     if (!state->new_bs) {
1561         goto out;
1562     }
1563 
1564     /*
1565      * Allow attaching a backing file to an overlay that's already in use only
1566      * if the parents don't assume that they are already seeing a valid image.
1567      * (Specifically, allow it as a mirror target, which is write-only access.)
1568      */
1569     bdrv_get_cumulative_perm(state->new_bs, &perm, &shared);
1570     if (perm & BLK_PERM_CONSISTENT_READ) {
1571         error_setg(errp, "The overlay is already in use");
1572         goto out;
1573     }
1574 
1575     if (state->new_bs->drv->is_filter) {
1576         error_setg(errp, "Filters cannot be used as overlays");
1577         goto out;
1578     }
1579 
1580     if (bdrv_cow_child(state->new_bs)) {
1581         error_setg(errp, "The overlay already has a backing image");
1582         goto out;
1583     }
1584 
1585     if (!state->new_bs->drv->supports_backing) {
1586         error_setg(errp, "The overlay does not support backing images");
1587         goto out;
1588     }
1589 
1590     /* This removes our old bs and adds the new bs. This is an operation that
1591      * can fail, so we need to do it in .prepare; undoing it for abort is
1592      * always possible. */
1593     bdrv_ref(state->new_bs);
1594     bdrv_append(state->new_bs, state->old_bs, &local_err);
1595     if (local_err) {
1596         error_propagate(errp, local_err);
1597         goto out;
1598     }
1599     state->overlay_appended = true;
1600 
1601 out:
1602     aio_context_release(aio_context);
1603 }
1604 
1605 static void external_snapshot_commit(BlkActionState *common)
1606 {
1607     ExternalSnapshotState *state =
1608                              DO_UPCAST(ExternalSnapshotState, common, common);
1609     AioContext *aio_context;
1610 
1611     aio_context = bdrv_get_aio_context(state->old_bs);
1612     aio_context_acquire(aio_context);
1613 
1614     /* We don't need (or want) to use the transactional
1615      * bdrv_reopen_multiple() across all the entries at once, because we
1616      * don't want to abort all of them if one of them fails the reopen */
1617     if (!qatomic_read(&state->old_bs->copy_on_read)) {
1618         bdrv_reopen_set_read_only(state->old_bs, true, NULL);
1619     }
1620 
1621     aio_context_release(aio_context);
1622 }
1623 
1624 static void external_snapshot_abort(BlkActionState *common)
1625 {
1626     ExternalSnapshotState *state =
1627                              DO_UPCAST(ExternalSnapshotState, common, common);
1628     if (state->new_bs) {
1629         if (state->overlay_appended) {
1630             AioContext *aio_context;
1631             AioContext *tmp_context;
1632             int ret;
1633 
1634             aio_context = bdrv_get_aio_context(state->old_bs);
1635             aio_context_acquire(aio_context);
1636 
1637             bdrv_ref(state->old_bs);   /* we can't let bdrv_set_backind_hd()
1638                                           close state->old_bs; we need it */
1639             bdrv_set_backing_hd(state->new_bs, NULL, &error_abort);
1640 
1641             /*
1642              * The call to bdrv_set_backing_hd() above returns state->old_bs to
1643              * the main AioContext. As we're still going to be using it, return
1644              * it to the AioContext it was before.
1645              */
1646             tmp_context = bdrv_get_aio_context(state->old_bs);
1647             if (aio_context != tmp_context) {
1648                 aio_context_release(aio_context);
1649                 aio_context_acquire(tmp_context);
1650 
1651                 ret = bdrv_try_set_aio_context(state->old_bs,
1652                                                aio_context, NULL);
1653                 assert(ret == 0);
1654 
1655                 aio_context_release(tmp_context);
1656                 aio_context_acquire(aio_context);
1657             }
1658 
1659             bdrv_replace_node(state->new_bs, state->old_bs, &error_abort);
1660             bdrv_unref(state->old_bs); /* bdrv_replace_node() ref'ed old_bs */
1661 
1662             aio_context_release(aio_context);
1663         }
1664     }
1665 }
1666 
1667 static void external_snapshot_clean(BlkActionState *common)
1668 {
1669     ExternalSnapshotState *state =
1670                              DO_UPCAST(ExternalSnapshotState, common, common);
1671     AioContext *aio_context;
1672 
1673     if (!state->old_bs) {
1674         return;
1675     }
1676 
1677     aio_context = bdrv_get_aio_context(state->old_bs);
1678     aio_context_acquire(aio_context);
1679 
1680     bdrv_drained_end(state->old_bs);
1681     bdrv_unref(state->new_bs);
1682 
1683     aio_context_release(aio_context);
1684 }
1685 
1686 typedef struct DriveBackupState {
1687     BlkActionState common;
1688     BlockDriverState *bs;
1689     BlockJob *job;
1690 } DriveBackupState;
1691 
1692 static BlockJob *do_backup_common(BackupCommon *backup,
1693                                   BlockDriverState *bs,
1694                                   BlockDriverState *target_bs,
1695                                   AioContext *aio_context,
1696                                   JobTxn *txn, Error **errp);
1697 
1698 static void drive_backup_prepare(BlkActionState *common, Error **errp)
1699 {
1700     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1701     DriveBackup *backup;
1702     BlockDriverState *bs;
1703     BlockDriverState *target_bs;
1704     BlockDriverState *source = NULL;
1705     AioContext *aio_context;
1706     AioContext *old_context;
1707     QDict *options;
1708     Error *local_err = NULL;
1709     int flags;
1710     int64_t size;
1711     bool set_backing_hd = false;
1712     int ret;
1713 
1714     assert(common->action->type == TRANSACTION_ACTION_KIND_DRIVE_BACKUP);
1715     backup = common->action->u.drive_backup.data;
1716 
1717     if (!backup->has_mode) {
1718         backup->mode = NEW_IMAGE_MODE_ABSOLUTE_PATHS;
1719     }
1720 
1721     bs = bdrv_lookup_bs(backup->device, backup->device, errp);
1722     if (!bs) {
1723         return;
1724     }
1725 
1726     if (!bs->drv) {
1727         error_setg(errp, "Device has no medium");
1728         return;
1729     }
1730 
1731     aio_context = bdrv_get_aio_context(bs);
1732     aio_context_acquire(aio_context);
1733 
1734     /* Paired with .clean() */
1735     bdrv_drained_begin(bs);
1736 
1737     if (!backup->has_format) {
1738         backup->format = backup->mode == NEW_IMAGE_MODE_EXISTING ?
1739                          NULL : (char *) bs->drv->format_name;
1740     }
1741 
1742     /* Early check to avoid creating target */
1743     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_BACKUP_SOURCE, errp)) {
1744         goto out;
1745     }
1746 
1747     flags = bs->open_flags | BDRV_O_RDWR;
1748 
1749     /*
1750      * See if we have a backing HD we can use to create our new image
1751      * on top of.
1752      */
1753     if (backup->sync == MIRROR_SYNC_MODE_TOP) {
1754         /*
1755          * Backup will not replace the source by the target, so none
1756          * of the filters skipped here will be removed (in contrast to
1757          * mirror).  Therefore, we can skip all of them when looking
1758          * for the first COW relationship.
1759          */
1760         source = bdrv_cow_bs(bdrv_skip_filters(bs));
1761         if (!source) {
1762             backup->sync = MIRROR_SYNC_MODE_FULL;
1763         }
1764     }
1765     if (backup->sync == MIRROR_SYNC_MODE_NONE) {
1766         source = bs;
1767         flags |= BDRV_O_NO_BACKING;
1768         set_backing_hd = true;
1769     }
1770 
1771     size = bdrv_getlength(bs);
1772     if (size < 0) {
1773         error_setg_errno(errp, -size, "bdrv_getlength failed");
1774         goto out;
1775     }
1776 
1777     if (backup->mode != NEW_IMAGE_MODE_EXISTING) {
1778         assert(backup->format);
1779         if (source) {
1780             /* Implicit filters should not appear in the filename */
1781             BlockDriverState *explicit_backing =
1782                 bdrv_skip_implicit_filters(source);
1783 
1784             bdrv_refresh_filename(explicit_backing);
1785             bdrv_img_create(backup->target, backup->format,
1786                             explicit_backing->filename,
1787                             explicit_backing->drv->format_name, NULL,
1788                             size, flags, false, &local_err);
1789         } else {
1790             bdrv_img_create(backup->target, backup->format, NULL, NULL, NULL,
1791                             size, flags, false, &local_err);
1792         }
1793     }
1794 
1795     if (local_err) {
1796         error_propagate(errp, local_err);
1797         goto out;
1798     }
1799 
1800     options = qdict_new();
1801     qdict_put_str(options, "discard", "unmap");
1802     qdict_put_str(options, "detect-zeroes", "unmap");
1803     if (backup->format) {
1804         qdict_put_str(options, "driver", backup->format);
1805     }
1806 
1807     target_bs = bdrv_open(backup->target, NULL, options, flags, errp);
1808     if (!target_bs) {
1809         goto out;
1810     }
1811 
1812     /* Honor bdrv_try_set_aio_context() context acquisition requirements. */
1813     old_context = bdrv_get_aio_context(target_bs);
1814     aio_context_release(aio_context);
1815     aio_context_acquire(old_context);
1816 
1817     ret = bdrv_try_set_aio_context(target_bs, aio_context, errp);
1818     if (ret < 0) {
1819         bdrv_unref(target_bs);
1820         aio_context_release(old_context);
1821         return;
1822     }
1823 
1824     aio_context_release(old_context);
1825     aio_context_acquire(aio_context);
1826 
1827     if (set_backing_hd) {
1828         bdrv_set_backing_hd(target_bs, source, &local_err);
1829         if (local_err) {
1830             error_propagate(errp, local_err);
1831             goto unref;
1832         }
1833     }
1834 
1835     state->bs = bs;
1836 
1837     state->job = do_backup_common(qapi_DriveBackup_base(backup),
1838                                   bs, target_bs, aio_context,
1839                                   common->block_job_txn, errp);
1840 
1841 unref:
1842     bdrv_unref(target_bs);
1843 out:
1844     aio_context_release(aio_context);
1845 }
1846 
1847 static void drive_backup_commit(BlkActionState *common)
1848 {
1849     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1850     AioContext *aio_context;
1851 
1852     aio_context = bdrv_get_aio_context(state->bs);
1853     aio_context_acquire(aio_context);
1854 
1855     assert(state->job);
1856     job_start(&state->job->job);
1857 
1858     aio_context_release(aio_context);
1859 }
1860 
1861 static void drive_backup_abort(BlkActionState *common)
1862 {
1863     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1864 
1865     if (state->job) {
1866         AioContext *aio_context;
1867 
1868         aio_context = bdrv_get_aio_context(state->bs);
1869         aio_context_acquire(aio_context);
1870 
1871         job_cancel_sync(&state->job->job);
1872 
1873         aio_context_release(aio_context);
1874     }
1875 }
1876 
1877 static void drive_backup_clean(BlkActionState *common)
1878 {
1879     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1880     AioContext *aio_context;
1881 
1882     if (!state->bs) {
1883         return;
1884     }
1885 
1886     aio_context = bdrv_get_aio_context(state->bs);
1887     aio_context_acquire(aio_context);
1888 
1889     bdrv_drained_end(state->bs);
1890 
1891     aio_context_release(aio_context);
1892 }
1893 
1894 typedef struct BlockdevBackupState {
1895     BlkActionState common;
1896     BlockDriverState *bs;
1897     BlockJob *job;
1898 } BlockdevBackupState;
1899 
1900 static void blockdev_backup_prepare(BlkActionState *common, Error **errp)
1901 {
1902     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1903     BlockdevBackup *backup;
1904     BlockDriverState *bs;
1905     BlockDriverState *target_bs;
1906     AioContext *aio_context;
1907     AioContext *old_context;
1908     int ret;
1909 
1910     assert(common->action->type == TRANSACTION_ACTION_KIND_BLOCKDEV_BACKUP);
1911     backup = common->action->u.blockdev_backup.data;
1912 
1913     bs = bdrv_lookup_bs(backup->device, backup->device, errp);
1914     if (!bs) {
1915         return;
1916     }
1917 
1918     target_bs = bdrv_lookup_bs(backup->target, backup->target, errp);
1919     if (!target_bs) {
1920         return;
1921     }
1922 
1923     /* Honor bdrv_try_set_aio_context() context acquisition requirements. */
1924     aio_context = bdrv_get_aio_context(bs);
1925     old_context = bdrv_get_aio_context(target_bs);
1926     aio_context_acquire(old_context);
1927 
1928     ret = bdrv_try_set_aio_context(target_bs, aio_context, errp);
1929     if (ret < 0) {
1930         aio_context_release(old_context);
1931         return;
1932     }
1933 
1934     aio_context_release(old_context);
1935     aio_context_acquire(aio_context);
1936     state->bs = bs;
1937 
1938     /* Paired with .clean() */
1939     bdrv_drained_begin(state->bs);
1940 
1941     state->job = do_backup_common(qapi_BlockdevBackup_base(backup),
1942                                   bs, target_bs, aio_context,
1943                                   common->block_job_txn, errp);
1944 
1945     aio_context_release(aio_context);
1946 }
1947 
1948 static void blockdev_backup_commit(BlkActionState *common)
1949 {
1950     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1951     AioContext *aio_context;
1952 
1953     aio_context = bdrv_get_aio_context(state->bs);
1954     aio_context_acquire(aio_context);
1955 
1956     assert(state->job);
1957     job_start(&state->job->job);
1958 
1959     aio_context_release(aio_context);
1960 }
1961 
1962 static void blockdev_backup_abort(BlkActionState *common)
1963 {
1964     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1965 
1966     if (state->job) {
1967         AioContext *aio_context;
1968 
1969         aio_context = bdrv_get_aio_context(state->bs);
1970         aio_context_acquire(aio_context);
1971 
1972         job_cancel_sync(&state->job->job);
1973 
1974         aio_context_release(aio_context);
1975     }
1976 }
1977 
1978 static void blockdev_backup_clean(BlkActionState *common)
1979 {
1980     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1981     AioContext *aio_context;
1982 
1983     if (!state->bs) {
1984         return;
1985     }
1986 
1987     aio_context = bdrv_get_aio_context(state->bs);
1988     aio_context_acquire(aio_context);
1989 
1990     bdrv_drained_end(state->bs);
1991 
1992     aio_context_release(aio_context);
1993 }
1994 
1995 typedef struct BlockDirtyBitmapState {
1996     BlkActionState common;
1997     BdrvDirtyBitmap *bitmap;
1998     BlockDriverState *bs;
1999     HBitmap *backup;
2000     bool prepared;
2001     bool was_enabled;
2002 } BlockDirtyBitmapState;
2003 
2004 static void block_dirty_bitmap_add_prepare(BlkActionState *common,
2005                                            Error **errp)
2006 {
2007     Error *local_err = NULL;
2008     BlockDirtyBitmapAdd *action;
2009     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2010                                              common, common);
2011 
2012     if (action_check_completion_mode(common, errp) < 0) {
2013         return;
2014     }
2015 
2016     action = common->action->u.block_dirty_bitmap_add.data;
2017     /* AIO context taken and released within qmp_block_dirty_bitmap_add */
2018     qmp_block_dirty_bitmap_add(action->node, action->name,
2019                                action->has_granularity, action->granularity,
2020                                action->has_persistent, action->persistent,
2021                                action->has_disabled, action->disabled,
2022                                &local_err);
2023 
2024     if (!local_err) {
2025         state->prepared = true;
2026     } else {
2027         error_propagate(errp, local_err);
2028     }
2029 }
2030 
2031 static void block_dirty_bitmap_add_abort(BlkActionState *common)
2032 {
2033     BlockDirtyBitmapAdd *action;
2034     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2035                                              common, common);
2036 
2037     action = common->action->u.block_dirty_bitmap_add.data;
2038     /* Should not be able to fail: IF the bitmap was added via .prepare(),
2039      * then the node reference and bitmap name must have been valid.
2040      */
2041     if (state->prepared) {
2042         qmp_block_dirty_bitmap_remove(action->node, action->name, &error_abort);
2043     }
2044 }
2045 
2046 static void block_dirty_bitmap_clear_prepare(BlkActionState *common,
2047                                              Error **errp)
2048 {
2049     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2050                                              common, common);
2051     BlockDirtyBitmap *action;
2052 
2053     if (action_check_completion_mode(common, errp) < 0) {
2054         return;
2055     }
2056 
2057     action = common->action->u.block_dirty_bitmap_clear.data;
2058     state->bitmap = block_dirty_bitmap_lookup(action->node,
2059                                               action->name,
2060                                               &state->bs,
2061                                               errp);
2062     if (!state->bitmap) {
2063         return;
2064     }
2065 
2066     if (bdrv_dirty_bitmap_check(state->bitmap, BDRV_BITMAP_DEFAULT, errp)) {
2067         return;
2068     }
2069 
2070     bdrv_clear_dirty_bitmap(state->bitmap, &state->backup);
2071 }
2072 
2073 static void block_dirty_bitmap_restore(BlkActionState *common)
2074 {
2075     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2076                                              common, common);
2077 
2078     if (state->backup) {
2079         bdrv_restore_dirty_bitmap(state->bitmap, state->backup);
2080     }
2081 }
2082 
2083 static void block_dirty_bitmap_free_backup(BlkActionState *common)
2084 {
2085     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2086                                              common, common);
2087 
2088     hbitmap_free(state->backup);
2089 }
2090 
2091 static void block_dirty_bitmap_enable_prepare(BlkActionState *common,
2092                                               Error **errp)
2093 {
2094     BlockDirtyBitmap *action;
2095     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2096                                              common, common);
2097 
2098     if (action_check_completion_mode(common, errp) < 0) {
2099         return;
2100     }
2101 
2102     action = common->action->u.block_dirty_bitmap_enable.data;
2103     state->bitmap = block_dirty_bitmap_lookup(action->node,
2104                                               action->name,
2105                                               NULL,
2106                                               errp);
2107     if (!state->bitmap) {
2108         return;
2109     }
2110 
2111     if (bdrv_dirty_bitmap_check(state->bitmap, BDRV_BITMAP_ALLOW_RO, errp)) {
2112         return;
2113     }
2114 
2115     state->was_enabled = bdrv_dirty_bitmap_enabled(state->bitmap);
2116     bdrv_enable_dirty_bitmap(state->bitmap);
2117 }
2118 
2119 static void block_dirty_bitmap_enable_abort(BlkActionState *common)
2120 {
2121     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2122                                              common, common);
2123 
2124     if (!state->was_enabled) {
2125         bdrv_disable_dirty_bitmap(state->bitmap);
2126     }
2127 }
2128 
2129 static void block_dirty_bitmap_disable_prepare(BlkActionState *common,
2130                                                Error **errp)
2131 {
2132     BlockDirtyBitmap *action;
2133     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2134                                              common, common);
2135 
2136     if (action_check_completion_mode(common, errp) < 0) {
2137         return;
2138     }
2139 
2140     action = common->action->u.block_dirty_bitmap_disable.data;
2141     state->bitmap = block_dirty_bitmap_lookup(action->node,
2142                                               action->name,
2143                                               NULL,
2144                                               errp);
2145     if (!state->bitmap) {
2146         return;
2147     }
2148 
2149     if (bdrv_dirty_bitmap_check(state->bitmap, BDRV_BITMAP_ALLOW_RO, errp)) {
2150         return;
2151     }
2152 
2153     state->was_enabled = bdrv_dirty_bitmap_enabled(state->bitmap);
2154     bdrv_disable_dirty_bitmap(state->bitmap);
2155 }
2156 
2157 static void block_dirty_bitmap_disable_abort(BlkActionState *common)
2158 {
2159     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2160                                              common, common);
2161 
2162     if (state->was_enabled) {
2163         bdrv_enable_dirty_bitmap(state->bitmap);
2164     }
2165 }
2166 
2167 static void block_dirty_bitmap_merge_prepare(BlkActionState *common,
2168                                              Error **errp)
2169 {
2170     BlockDirtyBitmapMerge *action;
2171     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2172                                              common, common);
2173 
2174     if (action_check_completion_mode(common, errp) < 0) {
2175         return;
2176     }
2177 
2178     action = common->action->u.block_dirty_bitmap_merge.data;
2179 
2180     state->bitmap = block_dirty_bitmap_merge(action->node, action->target,
2181                                              action->bitmaps, &state->backup,
2182                                              errp);
2183 }
2184 
2185 static void block_dirty_bitmap_remove_prepare(BlkActionState *common,
2186                                               Error **errp)
2187 {
2188     BlockDirtyBitmap *action;
2189     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2190                                              common, common);
2191 
2192     if (action_check_completion_mode(common, errp) < 0) {
2193         return;
2194     }
2195 
2196     action = common->action->u.block_dirty_bitmap_remove.data;
2197 
2198     state->bitmap = block_dirty_bitmap_remove(action->node, action->name,
2199                                               false, &state->bs, errp);
2200     if (state->bitmap) {
2201         bdrv_dirty_bitmap_skip_store(state->bitmap, true);
2202         bdrv_dirty_bitmap_set_busy(state->bitmap, true);
2203     }
2204 }
2205 
2206 static void block_dirty_bitmap_remove_abort(BlkActionState *common)
2207 {
2208     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2209                                              common, common);
2210 
2211     if (state->bitmap) {
2212         bdrv_dirty_bitmap_skip_store(state->bitmap, false);
2213         bdrv_dirty_bitmap_set_busy(state->bitmap, false);
2214     }
2215 }
2216 
2217 static void block_dirty_bitmap_remove_commit(BlkActionState *common)
2218 {
2219     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2220                                              common, common);
2221 
2222     bdrv_dirty_bitmap_set_busy(state->bitmap, false);
2223     bdrv_release_dirty_bitmap(state->bitmap);
2224 }
2225 
2226 static void abort_prepare(BlkActionState *common, Error **errp)
2227 {
2228     error_setg(errp, "Transaction aborted using Abort action");
2229 }
2230 
2231 static void abort_commit(BlkActionState *common)
2232 {
2233     g_assert_not_reached(); /* this action never succeeds */
2234 }
2235 
2236 static const BlkActionOps actions[] = {
2237     [TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT] = {
2238         .instance_size = sizeof(ExternalSnapshotState),
2239         .prepare  = external_snapshot_prepare,
2240         .commit   = external_snapshot_commit,
2241         .abort = external_snapshot_abort,
2242         .clean = external_snapshot_clean,
2243     },
2244     [TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC] = {
2245         .instance_size = sizeof(ExternalSnapshotState),
2246         .prepare  = external_snapshot_prepare,
2247         .commit   = external_snapshot_commit,
2248         .abort = external_snapshot_abort,
2249         .clean = external_snapshot_clean,
2250     },
2251     [TRANSACTION_ACTION_KIND_DRIVE_BACKUP] = {
2252         .instance_size = sizeof(DriveBackupState),
2253         .prepare = drive_backup_prepare,
2254         .commit = drive_backup_commit,
2255         .abort = drive_backup_abort,
2256         .clean = drive_backup_clean,
2257     },
2258     [TRANSACTION_ACTION_KIND_BLOCKDEV_BACKUP] = {
2259         .instance_size = sizeof(BlockdevBackupState),
2260         .prepare = blockdev_backup_prepare,
2261         .commit = blockdev_backup_commit,
2262         .abort = blockdev_backup_abort,
2263         .clean = blockdev_backup_clean,
2264     },
2265     [TRANSACTION_ACTION_KIND_ABORT] = {
2266         .instance_size = sizeof(BlkActionState),
2267         .prepare = abort_prepare,
2268         .commit = abort_commit,
2269     },
2270     [TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC] = {
2271         .instance_size = sizeof(InternalSnapshotState),
2272         .prepare  = internal_snapshot_prepare,
2273         .abort = internal_snapshot_abort,
2274         .clean = internal_snapshot_clean,
2275     },
2276     [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_ADD] = {
2277         .instance_size = sizeof(BlockDirtyBitmapState),
2278         .prepare = block_dirty_bitmap_add_prepare,
2279         .abort = block_dirty_bitmap_add_abort,
2280     },
2281     [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_CLEAR] = {
2282         .instance_size = sizeof(BlockDirtyBitmapState),
2283         .prepare = block_dirty_bitmap_clear_prepare,
2284         .commit = block_dirty_bitmap_free_backup,
2285         .abort = block_dirty_bitmap_restore,
2286     },
2287     [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_ENABLE] = {
2288         .instance_size = sizeof(BlockDirtyBitmapState),
2289         .prepare = block_dirty_bitmap_enable_prepare,
2290         .abort = block_dirty_bitmap_enable_abort,
2291     },
2292     [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_DISABLE] = {
2293         .instance_size = sizeof(BlockDirtyBitmapState),
2294         .prepare = block_dirty_bitmap_disable_prepare,
2295         .abort = block_dirty_bitmap_disable_abort,
2296     },
2297     [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_MERGE] = {
2298         .instance_size = sizeof(BlockDirtyBitmapState),
2299         .prepare = block_dirty_bitmap_merge_prepare,
2300         .commit = block_dirty_bitmap_free_backup,
2301         .abort = block_dirty_bitmap_restore,
2302     },
2303     [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_REMOVE] = {
2304         .instance_size = sizeof(BlockDirtyBitmapState),
2305         .prepare = block_dirty_bitmap_remove_prepare,
2306         .commit = block_dirty_bitmap_remove_commit,
2307         .abort = block_dirty_bitmap_remove_abort,
2308     },
2309     /* Where are transactions for MIRROR, COMMIT and STREAM?
2310      * Although these blockjobs use transaction callbacks like the backup job,
2311      * these jobs do not necessarily adhere to transaction semantics.
2312      * These jobs may not fully undo all of their actions on abort, nor do they
2313      * necessarily work in transactions with more than one job in them.
2314      */
2315 };
2316 
2317 /**
2318  * Allocate a TransactionProperties structure if necessary, and fill
2319  * that structure with desired defaults if they are unset.
2320  */
2321 static TransactionProperties *get_transaction_properties(
2322     TransactionProperties *props)
2323 {
2324     if (!props) {
2325         props = g_new0(TransactionProperties, 1);
2326     }
2327 
2328     if (!props->has_completion_mode) {
2329         props->has_completion_mode = true;
2330         props->completion_mode = ACTION_COMPLETION_MODE_INDIVIDUAL;
2331     }
2332 
2333     return props;
2334 }
2335 
2336 /*
2337  * 'Atomic' group operations.  The operations are performed as a set, and if
2338  * any fail then we roll back all operations in the group.
2339  */
2340 void qmp_transaction(TransactionActionList *dev_list,
2341                      bool has_props,
2342                      struct TransactionProperties *props,
2343                      Error **errp)
2344 {
2345     TransactionActionList *dev_entry = dev_list;
2346     JobTxn *block_job_txn = NULL;
2347     BlkActionState *state, *next;
2348     Error *local_err = NULL;
2349 
2350     QTAILQ_HEAD(, BlkActionState) snap_bdrv_states;
2351     QTAILQ_INIT(&snap_bdrv_states);
2352 
2353     /* Does this transaction get canceled as a group on failure?
2354      * If not, we don't really need to make a JobTxn.
2355      */
2356     props = get_transaction_properties(props);
2357     if (props->completion_mode != ACTION_COMPLETION_MODE_INDIVIDUAL) {
2358         block_job_txn = job_txn_new();
2359     }
2360 
2361     /* drain all i/o before any operations */
2362     bdrv_drain_all();
2363 
2364     /* We don't do anything in this loop that commits us to the operations */
2365     while (NULL != dev_entry) {
2366         TransactionAction *dev_info = NULL;
2367         const BlkActionOps *ops;
2368 
2369         dev_info = dev_entry->value;
2370         dev_entry = dev_entry->next;
2371 
2372         assert(dev_info->type < ARRAY_SIZE(actions));
2373 
2374         ops = &actions[dev_info->type];
2375         assert(ops->instance_size > 0);
2376 
2377         state = g_malloc0(ops->instance_size);
2378         state->ops = ops;
2379         state->action = dev_info;
2380         state->block_job_txn = block_job_txn;
2381         state->txn_props = props;
2382         QTAILQ_INSERT_TAIL(&snap_bdrv_states, state, entry);
2383 
2384         state->ops->prepare(state, &local_err);
2385         if (local_err) {
2386             error_propagate(errp, local_err);
2387             goto delete_and_fail;
2388         }
2389     }
2390 
2391     QTAILQ_FOREACH(state, &snap_bdrv_states, entry) {
2392         if (state->ops->commit) {
2393             state->ops->commit(state);
2394         }
2395     }
2396 
2397     /* success */
2398     goto exit;
2399 
2400 delete_and_fail:
2401     /* failure, and it is all-or-none; roll back all operations */
2402     QTAILQ_FOREACH_REVERSE(state, &snap_bdrv_states, entry) {
2403         if (state->ops->abort) {
2404             state->ops->abort(state);
2405         }
2406     }
2407 exit:
2408     QTAILQ_FOREACH_SAFE(state, &snap_bdrv_states, entry, next) {
2409         if (state->ops->clean) {
2410             state->ops->clean(state);
2411         }
2412         g_free(state);
2413     }
2414     if (!has_props) {
2415         qapi_free_TransactionProperties(props);
2416     }
2417     job_txn_unref(block_job_txn);
2418 }
2419 
2420 void qmp_block_passwd(bool has_device, const char *device,
2421                       bool has_node_name, const char *node_name,
2422                       const char *password, Error **errp)
2423 {
2424     error_setg(errp,
2425                "Setting block passwords directly is no longer supported");
2426 }
2427 
2428 BlockDirtyBitmapSha256 *qmp_x_debug_block_dirty_bitmap_sha256(const char *node,
2429                                                               const char *name,
2430                                                               Error **errp)
2431 {
2432     BdrvDirtyBitmap *bitmap;
2433     BlockDriverState *bs;
2434     BlockDirtyBitmapSha256 *ret = NULL;
2435     char *sha256;
2436 
2437     bitmap = block_dirty_bitmap_lookup(node, name, &bs, errp);
2438     if (!bitmap || !bs) {
2439         return NULL;
2440     }
2441 
2442     sha256 = bdrv_dirty_bitmap_sha256(bitmap, errp);
2443     if (sha256 == NULL) {
2444         return NULL;
2445     }
2446 
2447     ret = g_new(BlockDirtyBitmapSha256, 1);
2448     ret->sha256 = sha256;
2449 
2450     return ret;
2451 }
2452 
2453 void coroutine_fn qmp_block_resize(bool has_device, const char *device,
2454                                    bool has_node_name, const char *node_name,
2455                                    int64_t size, Error **errp)
2456 {
2457     Error *local_err = NULL;
2458     BlockBackend *blk;
2459     BlockDriverState *bs;
2460     AioContext *old_ctx;
2461 
2462     bs = bdrv_lookup_bs(has_device ? device : NULL,
2463                         has_node_name ? node_name : NULL,
2464                         &local_err);
2465     if (local_err) {
2466         error_propagate(errp, local_err);
2467         return;
2468     }
2469 
2470     if (size < 0) {
2471         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "size", "a >0 size");
2472         return;
2473     }
2474 
2475     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_RESIZE, NULL)) {
2476         error_setg(errp, QERR_DEVICE_IN_USE, device);
2477         return;
2478     }
2479 
2480     blk = blk_new_with_bs(bs, BLK_PERM_RESIZE, BLK_PERM_ALL, errp);
2481     if (!blk) {
2482         return;
2483     }
2484 
2485     bdrv_co_lock(bs);
2486     bdrv_drained_begin(bs);
2487     bdrv_co_unlock(bs);
2488 
2489     old_ctx = bdrv_co_enter(bs);
2490     blk_truncate(blk, size, false, PREALLOC_MODE_OFF, 0, errp);
2491     bdrv_co_leave(bs, old_ctx);
2492 
2493     bdrv_co_lock(bs);
2494     bdrv_drained_end(bs);
2495     blk_unref(blk);
2496     bdrv_co_unlock(bs);
2497 }
2498 
2499 void qmp_block_stream(bool has_job_id, const char *job_id, const char *device,
2500                       bool has_base, const char *base,
2501                       bool has_base_node, const char *base_node,
2502                       bool has_backing_file, const char *backing_file,
2503                       bool has_speed, int64_t speed,
2504                       bool has_on_error, BlockdevOnError on_error,
2505                       bool has_auto_finalize, bool auto_finalize,
2506                       bool has_auto_dismiss, bool auto_dismiss,
2507                       Error **errp)
2508 {
2509     BlockDriverState *bs, *iter;
2510     BlockDriverState *base_bs = NULL;
2511     AioContext *aio_context;
2512     Error *local_err = NULL;
2513     const char *base_name = NULL;
2514     int job_flags = JOB_DEFAULT;
2515 
2516     if (!has_on_error) {
2517         on_error = BLOCKDEV_ON_ERROR_REPORT;
2518     }
2519 
2520     bs = bdrv_lookup_bs(device, device, errp);
2521     if (!bs) {
2522         return;
2523     }
2524 
2525     aio_context = bdrv_get_aio_context(bs);
2526     aio_context_acquire(aio_context);
2527 
2528     if (has_base && has_base_node) {
2529         error_setg(errp, "'base' and 'base-node' cannot be specified "
2530                    "at the same time");
2531         goto out;
2532     }
2533 
2534     if (has_base) {
2535         base_bs = bdrv_find_backing_image(bs, base);
2536         if (base_bs == NULL) {
2537             error_setg(errp, "Can't find '%s' in the backing chain", base);
2538             goto out;
2539         }
2540         assert(bdrv_get_aio_context(base_bs) == aio_context);
2541         base_name = base;
2542     }
2543 
2544     if (has_base_node) {
2545         base_bs = bdrv_lookup_bs(NULL, base_node, errp);
2546         if (!base_bs) {
2547             goto out;
2548         }
2549         if (bs == base_bs || !bdrv_chain_contains(bs, base_bs)) {
2550             error_setg(errp, "Node '%s' is not a backing image of '%s'",
2551                        base_node, device);
2552             goto out;
2553         }
2554         assert(bdrv_get_aio_context(base_bs) == aio_context);
2555         bdrv_refresh_filename(base_bs);
2556         base_name = base_bs->filename;
2557     }
2558 
2559     /* Check for op blockers in the whole chain between bs and base */
2560     for (iter = bs; iter && iter != base_bs;
2561          iter = bdrv_filter_or_cow_bs(iter))
2562     {
2563         if (bdrv_op_is_blocked(iter, BLOCK_OP_TYPE_STREAM, errp)) {
2564             goto out;
2565         }
2566     }
2567 
2568     /* if we are streaming the entire chain, the result will have no backing
2569      * file, and specifying one is therefore an error */
2570     if (base_bs == NULL && has_backing_file) {
2571         error_setg(errp, "backing file specified, but streaming the "
2572                          "entire chain");
2573         goto out;
2574     }
2575 
2576     /* backing_file string overrides base bs filename */
2577     base_name = has_backing_file ? backing_file : base_name;
2578 
2579     if (has_auto_finalize && !auto_finalize) {
2580         job_flags |= JOB_MANUAL_FINALIZE;
2581     }
2582     if (has_auto_dismiss && !auto_dismiss) {
2583         job_flags |= JOB_MANUAL_DISMISS;
2584     }
2585 
2586     stream_start(has_job_id ? job_id : NULL, bs, base_bs, base_name,
2587                  job_flags, has_speed ? speed : 0, on_error, &local_err);
2588     if (local_err) {
2589         error_propagate(errp, local_err);
2590         goto out;
2591     }
2592 
2593     trace_qmp_block_stream(bs);
2594 
2595 out:
2596     aio_context_release(aio_context);
2597 }
2598 
2599 void qmp_block_commit(bool has_job_id, const char *job_id, const char *device,
2600                       bool has_base_node, const char *base_node,
2601                       bool has_base, const char *base,
2602                       bool has_top_node, const char *top_node,
2603                       bool has_top, const char *top,
2604                       bool has_backing_file, const char *backing_file,
2605                       bool has_speed, int64_t speed,
2606                       bool has_on_error, BlockdevOnError on_error,
2607                       bool has_filter_node_name, const char *filter_node_name,
2608                       bool has_auto_finalize, bool auto_finalize,
2609                       bool has_auto_dismiss, bool auto_dismiss,
2610                       Error **errp)
2611 {
2612     BlockDriverState *bs;
2613     BlockDriverState *iter;
2614     BlockDriverState *base_bs, *top_bs;
2615     AioContext *aio_context;
2616     Error *local_err = NULL;
2617     int job_flags = JOB_DEFAULT;
2618     uint64_t top_perm, top_shared;
2619 
2620     if (!has_speed) {
2621         speed = 0;
2622     }
2623     if (!has_on_error) {
2624         on_error = BLOCKDEV_ON_ERROR_REPORT;
2625     }
2626     if (!has_filter_node_name) {
2627         filter_node_name = NULL;
2628     }
2629     if (has_auto_finalize && !auto_finalize) {
2630         job_flags |= JOB_MANUAL_FINALIZE;
2631     }
2632     if (has_auto_dismiss && !auto_dismiss) {
2633         job_flags |= JOB_MANUAL_DISMISS;
2634     }
2635 
2636     /* Important Note:
2637      *  libvirt relies on the DeviceNotFound error class in order to probe for
2638      *  live commit feature versions; for this to work, we must make sure to
2639      *  perform the device lookup before any generic errors that may occur in a
2640      *  scenario in which all optional arguments are omitted. */
2641     bs = qmp_get_root_bs(device, &local_err);
2642     if (!bs) {
2643         bs = bdrv_lookup_bs(device, device, NULL);
2644         if (!bs) {
2645             error_free(local_err);
2646             error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
2647                       "Device '%s' not found", device);
2648         } else {
2649             error_propagate(errp, local_err);
2650         }
2651         return;
2652     }
2653 
2654     aio_context = bdrv_get_aio_context(bs);
2655     aio_context_acquire(aio_context);
2656 
2657     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_COMMIT_SOURCE, errp)) {
2658         goto out;
2659     }
2660 
2661     /* default top_bs is the active layer */
2662     top_bs = bs;
2663 
2664     if (has_top_node && has_top) {
2665         error_setg(errp, "'top-node' and 'top' are mutually exclusive");
2666         goto out;
2667     } else if (has_top_node) {
2668         top_bs = bdrv_lookup_bs(NULL, top_node, errp);
2669         if (top_bs == NULL) {
2670             goto out;
2671         }
2672         if (!bdrv_chain_contains(bs, top_bs)) {
2673             error_setg(errp, "'%s' is not in this backing file chain",
2674                        top_node);
2675             goto out;
2676         }
2677     } else if (has_top && top) {
2678         /* This strcmp() is just a shortcut, there is no need to
2679          * refresh @bs's filename.  If it mismatches,
2680          * bdrv_find_backing_image() will do the refresh and may still
2681          * return @bs. */
2682         if (strcmp(bs->filename, top) != 0) {
2683             top_bs = bdrv_find_backing_image(bs, top);
2684         }
2685     }
2686 
2687     if (top_bs == NULL) {
2688         error_setg(errp, "Top image file %s not found", top ? top : "NULL");
2689         goto out;
2690     }
2691 
2692     assert(bdrv_get_aio_context(top_bs) == aio_context);
2693 
2694     if (has_base_node && has_base) {
2695         error_setg(errp, "'base-node' and 'base' are mutually exclusive");
2696         goto out;
2697     } else if (has_base_node) {
2698         base_bs = bdrv_lookup_bs(NULL, base_node, errp);
2699         if (base_bs == NULL) {
2700             goto out;
2701         }
2702         if (!bdrv_chain_contains(top_bs, base_bs)) {
2703             error_setg(errp, "'%s' is not in this backing file chain",
2704                        base_node);
2705             goto out;
2706         }
2707     } else if (has_base && base) {
2708         base_bs = bdrv_find_backing_image(top_bs, base);
2709         if (base_bs == NULL) {
2710             error_setg(errp, "Can't find '%s' in the backing chain", base);
2711             goto out;
2712         }
2713     } else {
2714         base_bs = bdrv_find_base(top_bs);
2715         if (base_bs == NULL) {
2716             error_setg(errp, "There is no backimg image");
2717             goto out;
2718         }
2719     }
2720 
2721     assert(bdrv_get_aio_context(base_bs) == aio_context);
2722 
2723     for (iter = top_bs; iter != bdrv_filter_or_cow_bs(base_bs);
2724          iter = bdrv_filter_or_cow_bs(iter))
2725     {
2726         if (bdrv_op_is_blocked(iter, BLOCK_OP_TYPE_COMMIT_TARGET, errp)) {
2727             goto out;
2728         }
2729     }
2730 
2731     /* Do not allow attempts to commit an image into itself */
2732     if (top_bs == base_bs) {
2733         error_setg(errp, "cannot commit an image into itself");
2734         goto out;
2735     }
2736 
2737     /*
2738      * Active commit is required if and only if someone has taken a
2739      * WRITE permission on the top node.  Historically, we have always
2740      * used active commit for top nodes, so continue that practice
2741      * lest we possibly break clients that rely on this behavior, e.g.
2742      * to later attach this node to a writing parent.
2743      * (Active commit is never really wrong.)
2744      */
2745     bdrv_get_cumulative_perm(top_bs, &top_perm, &top_shared);
2746     if (top_perm & BLK_PERM_WRITE ||
2747         bdrv_skip_filters(top_bs) == bdrv_skip_filters(bs))
2748     {
2749         if (has_backing_file) {
2750             if (bdrv_skip_filters(top_bs) == bdrv_skip_filters(bs)) {
2751                 error_setg(errp, "'backing-file' specified,"
2752                                  " but 'top' is the active layer");
2753             } else {
2754                 error_setg(errp, "'backing-file' specified, but 'top' has a "
2755                                  "writer on it");
2756             }
2757             goto out;
2758         }
2759         if (!has_job_id) {
2760             /*
2761              * Emulate here what block_job_create() does, because it
2762              * is possible that @bs != @top_bs (the block job should
2763              * be named after @bs, even if @top_bs is the actual
2764              * source)
2765              */
2766             job_id = bdrv_get_device_name(bs);
2767         }
2768         commit_active_start(job_id, top_bs, base_bs, job_flags, speed, on_error,
2769                             filter_node_name, NULL, NULL, false, &local_err);
2770     } else {
2771         BlockDriverState *overlay_bs = bdrv_find_overlay(bs, top_bs);
2772         if (bdrv_op_is_blocked(overlay_bs, BLOCK_OP_TYPE_COMMIT_TARGET, errp)) {
2773             goto out;
2774         }
2775         commit_start(has_job_id ? job_id : NULL, bs, base_bs, top_bs, job_flags,
2776                      speed, on_error, has_backing_file ? backing_file : NULL,
2777                      filter_node_name, &local_err);
2778     }
2779     if (local_err != NULL) {
2780         error_propagate(errp, local_err);
2781         goto out;
2782     }
2783 
2784 out:
2785     aio_context_release(aio_context);
2786 }
2787 
2788 /* Common QMP interface for drive-backup and blockdev-backup */
2789 static BlockJob *do_backup_common(BackupCommon *backup,
2790                                   BlockDriverState *bs,
2791                                   BlockDriverState *target_bs,
2792                                   AioContext *aio_context,
2793                                   JobTxn *txn, Error **errp)
2794 {
2795     BlockJob *job = NULL;
2796     BdrvDirtyBitmap *bmap = NULL;
2797     int job_flags = JOB_DEFAULT;
2798 
2799     if (!backup->has_speed) {
2800         backup->speed = 0;
2801     }
2802     if (!backup->has_on_source_error) {
2803         backup->on_source_error = BLOCKDEV_ON_ERROR_REPORT;
2804     }
2805     if (!backup->has_on_target_error) {
2806         backup->on_target_error = BLOCKDEV_ON_ERROR_REPORT;
2807     }
2808     if (!backup->has_job_id) {
2809         backup->job_id = NULL;
2810     }
2811     if (!backup->has_auto_finalize) {
2812         backup->auto_finalize = true;
2813     }
2814     if (!backup->has_auto_dismiss) {
2815         backup->auto_dismiss = true;
2816     }
2817     if (!backup->has_compress) {
2818         backup->compress = false;
2819     }
2820 
2821     if ((backup->sync == MIRROR_SYNC_MODE_BITMAP) ||
2822         (backup->sync == MIRROR_SYNC_MODE_INCREMENTAL)) {
2823         /* done before desugaring 'incremental' to print the right message */
2824         if (!backup->has_bitmap) {
2825             error_setg(errp, "must provide a valid bitmap name for "
2826                        "'%s' sync mode", MirrorSyncMode_str(backup->sync));
2827             return NULL;
2828         }
2829     }
2830 
2831     if (backup->sync == MIRROR_SYNC_MODE_INCREMENTAL) {
2832         if (backup->has_bitmap_mode &&
2833             backup->bitmap_mode != BITMAP_SYNC_MODE_ON_SUCCESS) {
2834             error_setg(errp, "Bitmap sync mode must be '%s' "
2835                        "when using sync mode '%s'",
2836                        BitmapSyncMode_str(BITMAP_SYNC_MODE_ON_SUCCESS),
2837                        MirrorSyncMode_str(backup->sync));
2838             return NULL;
2839         }
2840         backup->has_bitmap_mode = true;
2841         backup->sync = MIRROR_SYNC_MODE_BITMAP;
2842         backup->bitmap_mode = BITMAP_SYNC_MODE_ON_SUCCESS;
2843     }
2844 
2845     if (backup->has_bitmap) {
2846         bmap = bdrv_find_dirty_bitmap(bs, backup->bitmap);
2847         if (!bmap) {
2848             error_setg(errp, "Bitmap '%s' could not be found", backup->bitmap);
2849             return NULL;
2850         }
2851         if (!backup->has_bitmap_mode) {
2852             error_setg(errp, "Bitmap sync mode must be given "
2853                        "when providing a bitmap");
2854             return NULL;
2855         }
2856         if (bdrv_dirty_bitmap_check(bmap, BDRV_BITMAP_ALLOW_RO, errp)) {
2857             return NULL;
2858         }
2859 
2860         /* This does not produce a useful bitmap artifact: */
2861         if (backup->sync == MIRROR_SYNC_MODE_NONE) {
2862             error_setg(errp, "sync mode '%s' does not produce meaningful bitmap"
2863                        " outputs", MirrorSyncMode_str(backup->sync));
2864             return NULL;
2865         }
2866 
2867         /* If the bitmap isn't used for input or output, this is useless: */
2868         if (backup->bitmap_mode == BITMAP_SYNC_MODE_NEVER &&
2869             backup->sync != MIRROR_SYNC_MODE_BITMAP) {
2870             error_setg(errp, "Bitmap sync mode '%s' has no meaningful effect"
2871                        " when combined with sync mode '%s'",
2872                        BitmapSyncMode_str(backup->bitmap_mode),
2873                        MirrorSyncMode_str(backup->sync));
2874             return NULL;
2875         }
2876     }
2877 
2878     if (!backup->has_bitmap && backup->has_bitmap_mode) {
2879         error_setg(errp, "Cannot specify bitmap sync mode without a bitmap");
2880         return NULL;
2881     }
2882 
2883     if (!backup->auto_finalize) {
2884         job_flags |= JOB_MANUAL_FINALIZE;
2885     }
2886     if (!backup->auto_dismiss) {
2887         job_flags |= JOB_MANUAL_DISMISS;
2888     }
2889 
2890     job = backup_job_create(backup->job_id, bs, target_bs, backup->speed,
2891                             backup->sync, bmap, backup->bitmap_mode,
2892                             backup->compress,
2893                             backup->filter_node_name,
2894                             backup->on_source_error,
2895                             backup->on_target_error,
2896                             job_flags, NULL, NULL, txn, errp);
2897     return job;
2898 }
2899 
2900 void qmp_drive_backup(DriveBackup *backup, Error **errp)
2901 {
2902     TransactionAction action = {
2903         .type = TRANSACTION_ACTION_KIND_DRIVE_BACKUP,
2904         .u.drive_backup.data = backup,
2905     };
2906     blockdev_do_action(&action, errp);
2907 }
2908 
2909 BlockDeviceInfoList *qmp_query_named_block_nodes(bool has_flat,
2910                                                  bool flat,
2911                                                  Error **errp)
2912 {
2913     bool return_flat = has_flat && flat;
2914 
2915     return bdrv_named_nodes_list(return_flat, errp);
2916 }
2917 
2918 XDbgBlockGraph *qmp_x_debug_query_block_graph(Error **errp)
2919 {
2920     return bdrv_get_xdbg_block_graph(errp);
2921 }
2922 
2923 void qmp_blockdev_backup(BlockdevBackup *backup, Error **errp)
2924 {
2925     TransactionAction action = {
2926         .type = TRANSACTION_ACTION_KIND_BLOCKDEV_BACKUP,
2927         .u.blockdev_backup.data = backup,
2928     };
2929     blockdev_do_action(&action, errp);
2930 }
2931 
2932 /* Parameter check and block job starting for drive mirroring.
2933  * Caller should hold @device and @target's aio context (must be the same).
2934  **/
2935 static void blockdev_mirror_common(const char *job_id, BlockDriverState *bs,
2936                                    BlockDriverState *target,
2937                                    bool has_replaces, const char *replaces,
2938                                    enum MirrorSyncMode sync,
2939                                    BlockMirrorBackingMode backing_mode,
2940                                    bool zero_target,
2941                                    bool has_speed, int64_t speed,
2942                                    bool has_granularity, uint32_t granularity,
2943                                    bool has_buf_size, int64_t buf_size,
2944                                    bool has_on_source_error,
2945                                    BlockdevOnError on_source_error,
2946                                    bool has_on_target_error,
2947                                    BlockdevOnError on_target_error,
2948                                    bool has_unmap, bool unmap,
2949                                    bool has_filter_node_name,
2950                                    const char *filter_node_name,
2951                                    bool has_copy_mode, MirrorCopyMode copy_mode,
2952                                    bool has_auto_finalize, bool auto_finalize,
2953                                    bool has_auto_dismiss, bool auto_dismiss,
2954                                    Error **errp)
2955 {
2956     BlockDriverState *unfiltered_bs;
2957     int job_flags = JOB_DEFAULT;
2958 
2959     if (!has_speed) {
2960         speed = 0;
2961     }
2962     if (!has_on_source_error) {
2963         on_source_error = BLOCKDEV_ON_ERROR_REPORT;
2964     }
2965     if (!has_on_target_error) {
2966         on_target_error = BLOCKDEV_ON_ERROR_REPORT;
2967     }
2968     if (!has_granularity) {
2969         granularity = 0;
2970     }
2971     if (!has_buf_size) {
2972         buf_size = 0;
2973     }
2974     if (!has_unmap) {
2975         unmap = true;
2976     }
2977     if (!has_filter_node_name) {
2978         filter_node_name = NULL;
2979     }
2980     if (!has_copy_mode) {
2981         copy_mode = MIRROR_COPY_MODE_BACKGROUND;
2982     }
2983     if (has_auto_finalize && !auto_finalize) {
2984         job_flags |= JOB_MANUAL_FINALIZE;
2985     }
2986     if (has_auto_dismiss && !auto_dismiss) {
2987         job_flags |= JOB_MANUAL_DISMISS;
2988     }
2989 
2990     if (granularity != 0 && (granularity < 512 || granularity > 1048576 * 64)) {
2991         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "granularity",
2992                    "a value in range [512B, 64MB]");
2993         return;
2994     }
2995     if (granularity & (granularity - 1)) {
2996         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "granularity",
2997                    "a power of 2");
2998         return;
2999     }
3000 
3001     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_MIRROR_SOURCE, errp)) {
3002         return;
3003     }
3004     if (bdrv_op_is_blocked(target, BLOCK_OP_TYPE_MIRROR_TARGET, errp)) {
3005         return;
3006     }
3007 
3008     if (!bdrv_backing_chain_next(bs) && sync == MIRROR_SYNC_MODE_TOP) {
3009         sync = MIRROR_SYNC_MODE_FULL;
3010     }
3011 
3012     if (!has_replaces) {
3013         /* We want to mirror from @bs, but keep implicit filters on top */
3014         unfiltered_bs = bdrv_skip_implicit_filters(bs);
3015         if (unfiltered_bs != bs) {
3016             replaces = unfiltered_bs->node_name;
3017             has_replaces = true;
3018         }
3019     }
3020 
3021     if (has_replaces) {
3022         BlockDriverState *to_replace_bs;
3023         AioContext *replace_aio_context;
3024         int64_t bs_size, replace_size;
3025 
3026         bs_size = bdrv_getlength(bs);
3027         if (bs_size < 0) {
3028             error_setg_errno(errp, -bs_size, "Failed to query device's size");
3029             return;
3030         }
3031 
3032         to_replace_bs = check_to_replace_node(bs, replaces, errp);
3033         if (!to_replace_bs) {
3034             return;
3035         }
3036 
3037         replace_aio_context = bdrv_get_aio_context(to_replace_bs);
3038         aio_context_acquire(replace_aio_context);
3039         replace_size = bdrv_getlength(to_replace_bs);
3040         aio_context_release(replace_aio_context);
3041 
3042         if (replace_size < 0) {
3043             error_setg_errno(errp, -replace_size,
3044                              "Failed to query the replacement node's size");
3045             return;
3046         }
3047         if (bs_size != replace_size) {
3048             error_setg(errp, "cannot replace image with a mirror image of "
3049                              "different size");
3050             return;
3051         }
3052     }
3053 
3054     /* pass the node name to replace to mirror start since it's loose coupling
3055      * and will allow to check whether the node still exist at mirror completion
3056      */
3057     mirror_start(job_id, bs, target,
3058                  has_replaces ? replaces : NULL, job_flags,
3059                  speed, granularity, buf_size, sync, backing_mode, zero_target,
3060                  on_source_error, on_target_error, unmap, filter_node_name,
3061                  copy_mode, errp);
3062 }
3063 
3064 void qmp_drive_mirror(DriveMirror *arg, Error **errp)
3065 {
3066     BlockDriverState *bs;
3067     BlockDriverState *target_backing_bs, *target_bs;
3068     AioContext *aio_context;
3069     AioContext *old_context;
3070     BlockMirrorBackingMode backing_mode;
3071     Error *local_err = NULL;
3072     QDict *options = NULL;
3073     int flags;
3074     int64_t size;
3075     const char *format = arg->format;
3076     bool zero_target;
3077     int ret;
3078 
3079     bs = qmp_get_root_bs(arg->device, errp);
3080     if (!bs) {
3081         return;
3082     }
3083 
3084     /* Early check to avoid creating target */
3085     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_MIRROR_SOURCE, errp)) {
3086         return;
3087     }
3088 
3089     aio_context = bdrv_get_aio_context(bs);
3090     aio_context_acquire(aio_context);
3091 
3092     if (!arg->has_mode) {
3093         arg->mode = NEW_IMAGE_MODE_ABSOLUTE_PATHS;
3094     }
3095 
3096     if (!arg->has_format) {
3097         format = (arg->mode == NEW_IMAGE_MODE_EXISTING
3098                   ? NULL : bs->drv->format_name);
3099     }
3100 
3101     flags = bs->open_flags | BDRV_O_RDWR;
3102     target_backing_bs = bdrv_cow_bs(bdrv_skip_filters(bs));
3103     if (!target_backing_bs && arg->sync == MIRROR_SYNC_MODE_TOP) {
3104         arg->sync = MIRROR_SYNC_MODE_FULL;
3105     }
3106     if (arg->sync == MIRROR_SYNC_MODE_NONE) {
3107         target_backing_bs = bs;
3108     }
3109 
3110     size = bdrv_getlength(bs);
3111     if (size < 0) {
3112         error_setg_errno(errp, -size, "bdrv_getlength failed");
3113         goto out;
3114     }
3115 
3116     if (arg->has_replaces) {
3117         if (!arg->has_node_name) {
3118             error_setg(errp, "a node-name must be provided when replacing a"
3119                              " named node of the graph");
3120             goto out;
3121         }
3122     }
3123 
3124     if (arg->mode == NEW_IMAGE_MODE_ABSOLUTE_PATHS) {
3125         backing_mode = MIRROR_SOURCE_BACKING_CHAIN;
3126     } else {
3127         backing_mode = MIRROR_OPEN_BACKING_CHAIN;
3128     }
3129 
3130     /* Don't open backing image in create() */
3131     flags |= BDRV_O_NO_BACKING;
3132 
3133     if ((arg->sync == MIRROR_SYNC_MODE_FULL || !target_backing_bs)
3134         && arg->mode != NEW_IMAGE_MODE_EXISTING)
3135     {
3136         /* create new image w/o backing file */
3137         assert(format);
3138         bdrv_img_create(arg->target, format,
3139                         NULL, NULL, NULL, size, flags, false, &local_err);
3140     } else {
3141         /* Implicit filters should not appear in the filename */
3142         BlockDriverState *explicit_backing =
3143             bdrv_skip_implicit_filters(target_backing_bs);
3144 
3145         switch (arg->mode) {
3146         case NEW_IMAGE_MODE_EXISTING:
3147             break;
3148         case NEW_IMAGE_MODE_ABSOLUTE_PATHS:
3149             /* create new image with backing file */
3150             bdrv_refresh_filename(explicit_backing);
3151             bdrv_img_create(arg->target, format,
3152                             explicit_backing->filename,
3153                             explicit_backing->drv->format_name,
3154                             NULL, size, flags, false, &local_err);
3155             break;
3156         default:
3157             abort();
3158         }
3159     }
3160 
3161     if (local_err) {
3162         error_propagate(errp, local_err);
3163         goto out;
3164     }
3165 
3166     options = qdict_new();
3167     if (arg->has_node_name) {
3168         qdict_put_str(options, "node-name", arg->node_name);
3169     }
3170     if (format) {
3171         qdict_put_str(options, "driver", format);
3172     }
3173 
3174     /* Mirroring takes care of copy-on-write using the source's backing
3175      * file.
3176      */
3177     target_bs = bdrv_open(arg->target, NULL, options, flags, errp);
3178     if (!target_bs) {
3179         goto out;
3180     }
3181 
3182     zero_target = (arg->sync == MIRROR_SYNC_MODE_FULL &&
3183                    (arg->mode == NEW_IMAGE_MODE_EXISTING ||
3184                     !bdrv_has_zero_init(target_bs)));
3185 
3186 
3187     /* Honor bdrv_try_set_aio_context() context acquisition requirements. */
3188     old_context = bdrv_get_aio_context(target_bs);
3189     aio_context_release(aio_context);
3190     aio_context_acquire(old_context);
3191 
3192     ret = bdrv_try_set_aio_context(target_bs, aio_context, errp);
3193     if (ret < 0) {
3194         bdrv_unref(target_bs);
3195         aio_context_release(old_context);
3196         return;
3197     }
3198 
3199     aio_context_release(old_context);
3200     aio_context_acquire(aio_context);
3201 
3202     blockdev_mirror_common(arg->has_job_id ? arg->job_id : NULL, bs, target_bs,
3203                            arg->has_replaces, arg->replaces, arg->sync,
3204                            backing_mode, zero_target,
3205                            arg->has_speed, arg->speed,
3206                            arg->has_granularity, arg->granularity,
3207                            arg->has_buf_size, arg->buf_size,
3208                            arg->has_on_source_error, arg->on_source_error,
3209                            arg->has_on_target_error, arg->on_target_error,
3210                            arg->has_unmap, arg->unmap,
3211                            false, NULL,
3212                            arg->has_copy_mode, arg->copy_mode,
3213                            arg->has_auto_finalize, arg->auto_finalize,
3214                            arg->has_auto_dismiss, arg->auto_dismiss,
3215                            errp);
3216     bdrv_unref(target_bs);
3217 out:
3218     aio_context_release(aio_context);
3219 }
3220 
3221 void qmp_blockdev_mirror(bool has_job_id, const char *job_id,
3222                          const char *device, const char *target,
3223                          bool has_replaces, const char *replaces,
3224                          MirrorSyncMode sync,
3225                          bool has_speed, int64_t speed,
3226                          bool has_granularity, uint32_t granularity,
3227                          bool has_buf_size, int64_t buf_size,
3228                          bool has_on_source_error,
3229                          BlockdevOnError on_source_error,
3230                          bool has_on_target_error,
3231                          BlockdevOnError on_target_error,
3232                          bool has_filter_node_name,
3233                          const char *filter_node_name,
3234                          bool has_copy_mode, MirrorCopyMode copy_mode,
3235                          bool has_auto_finalize, bool auto_finalize,
3236                          bool has_auto_dismiss, bool auto_dismiss,
3237                          Error **errp)
3238 {
3239     BlockDriverState *bs;
3240     BlockDriverState *target_bs;
3241     AioContext *aio_context;
3242     AioContext *old_context;
3243     BlockMirrorBackingMode backing_mode = MIRROR_LEAVE_BACKING_CHAIN;
3244     bool zero_target;
3245     int ret;
3246 
3247     bs = qmp_get_root_bs(device, errp);
3248     if (!bs) {
3249         return;
3250     }
3251 
3252     target_bs = bdrv_lookup_bs(target, target, errp);
3253     if (!target_bs) {
3254         return;
3255     }
3256 
3257     zero_target = (sync == MIRROR_SYNC_MODE_FULL);
3258 
3259     /* Honor bdrv_try_set_aio_context() context acquisition requirements. */
3260     old_context = bdrv_get_aio_context(target_bs);
3261     aio_context = bdrv_get_aio_context(bs);
3262     aio_context_acquire(old_context);
3263 
3264     ret = bdrv_try_set_aio_context(target_bs, aio_context, errp);
3265 
3266     aio_context_release(old_context);
3267     aio_context_acquire(aio_context);
3268 
3269     if (ret < 0) {
3270         goto out;
3271     }
3272 
3273     blockdev_mirror_common(has_job_id ? job_id : NULL, bs, target_bs,
3274                            has_replaces, replaces, sync, backing_mode,
3275                            zero_target, has_speed, speed,
3276                            has_granularity, granularity,
3277                            has_buf_size, buf_size,
3278                            has_on_source_error, on_source_error,
3279                            has_on_target_error, on_target_error,
3280                            true, true,
3281                            has_filter_node_name, filter_node_name,
3282                            has_copy_mode, copy_mode,
3283                            has_auto_finalize, auto_finalize,
3284                            has_auto_dismiss, auto_dismiss,
3285                            errp);
3286 out:
3287     aio_context_release(aio_context);
3288 }
3289 
3290 /* Get a block job using its ID and acquire its AioContext */
3291 static BlockJob *find_block_job(const char *id, AioContext **aio_context,
3292                                 Error **errp)
3293 {
3294     BlockJob *job;
3295 
3296     assert(id != NULL);
3297 
3298     *aio_context = NULL;
3299 
3300     job = block_job_get(id);
3301 
3302     if (!job) {
3303         error_set(errp, ERROR_CLASS_DEVICE_NOT_ACTIVE,
3304                   "Block job '%s' not found", id);
3305         return NULL;
3306     }
3307 
3308     *aio_context = blk_get_aio_context(job->blk);
3309     aio_context_acquire(*aio_context);
3310 
3311     return job;
3312 }
3313 
3314 void qmp_block_job_set_speed(const char *device, int64_t speed, Error **errp)
3315 {
3316     AioContext *aio_context;
3317     BlockJob *job = find_block_job(device, &aio_context, errp);
3318 
3319     if (!job) {
3320         return;
3321     }
3322 
3323     block_job_set_speed(job, speed, errp);
3324     aio_context_release(aio_context);
3325 }
3326 
3327 void qmp_block_job_cancel(const char *device,
3328                           bool has_force, bool force, Error **errp)
3329 {
3330     AioContext *aio_context;
3331     BlockJob *job = find_block_job(device, &aio_context, errp);
3332 
3333     if (!job) {
3334         return;
3335     }
3336 
3337     if (!has_force) {
3338         force = false;
3339     }
3340 
3341     if (job_user_paused(&job->job) && !force) {
3342         error_setg(errp, "The block job for device '%s' is currently paused",
3343                    device);
3344         goto out;
3345     }
3346 
3347     trace_qmp_block_job_cancel(job);
3348     job_user_cancel(&job->job, force, errp);
3349 out:
3350     aio_context_release(aio_context);
3351 }
3352 
3353 void qmp_block_job_pause(const char *device, Error **errp)
3354 {
3355     AioContext *aio_context;
3356     BlockJob *job = find_block_job(device, &aio_context, errp);
3357 
3358     if (!job) {
3359         return;
3360     }
3361 
3362     trace_qmp_block_job_pause(job);
3363     job_user_pause(&job->job, errp);
3364     aio_context_release(aio_context);
3365 }
3366 
3367 void qmp_block_job_resume(const char *device, Error **errp)
3368 {
3369     AioContext *aio_context;
3370     BlockJob *job = find_block_job(device, &aio_context, errp);
3371 
3372     if (!job) {
3373         return;
3374     }
3375 
3376     trace_qmp_block_job_resume(job);
3377     job_user_resume(&job->job, errp);
3378     aio_context_release(aio_context);
3379 }
3380 
3381 void qmp_block_job_complete(const char *device, Error **errp)
3382 {
3383     AioContext *aio_context;
3384     BlockJob *job = find_block_job(device, &aio_context, errp);
3385 
3386     if (!job) {
3387         return;
3388     }
3389 
3390     trace_qmp_block_job_complete(job);
3391     job_complete(&job->job, errp);
3392     aio_context_release(aio_context);
3393 }
3394 
3395 void qmp_block_job_finalize(const char *id, Error **errp)
3396 {
3397     AioContext *aio_context;
3398     BlockJob *job = find_block_job(id, &aio_context, errp);
3399 
3400     if (!job) {
3401         return;
3402     }
3403 
3404     trace_qmp_block_job_finalize(job);
3405     job_ref(&job->job);
3406     job_finalize(&job->job, errp);
3407 
3408     /*
3409      * Job's context might have changed via job_finalize (and job_txn_apply
3410      * automatically acquires the new one), so make sure we release the correct
3411      * one.
3412      */
3413     aio_context = blk_get_aio_context(job->blk);
3414     job_unref(&job->job);
3415     aio_context_release(aio_context);
3416 }
3417 
3418 void qmp_block_job_dismiss(const char *id, Error **errp)
3419 {
3420     AioContext *aio_context;
3421     BlockJob *bjob = find_block_job(id, &aio_context, errp);
3422     Job *job;
3423 
3424     if (!bjob) {
3425         return;
3426     }
3427 
3428     trace_qmp_block_job_dismiss(bjob);
3429     job = &bjob->job;
3430     job_dismiss(&job, errp);
3431     aio_context_release(aio_context);
3432 }
3433 
3434 void qmp_change_backing_file(const char *device,
3435                              const char *image_node_name,
3436                              const char *backing_file,
3437                              Error **errp)
3438 {
3439     BlockDriverState *bs = NULL;
3440     AioContext *aio_context;
3441     BlockDriverState *image_bs = NULL;
3442     Error *local_err = NULL;
3443     bool ro;
3444     int ret;
3445 
3446     bs = qmp_get_root_bs(device, errp);
3447     if (!bs) {
3448         return;
3449     }
3450 
3451     aio_context = bdrv_get_aio_context(bs);
3452     aio_context_acquire(aio_context);
3453 
3454     image_bs = bdrv_lookup_bs(NULL, image_node_name, &local_err);
3455     if (local_err) {
3456         error_propagate(errp, local_err);
3457         goto out;
3458     }
3459 
3460     if (!image_bs) {
3461         error_setg(errp, "image file not found");
3462         goto out;
3463     }
3464 
3465     if (bdrv_find_base(image_bs) == image_bs) {
3466         error_setg(errp, "not allowing backing file change on an image "
3467                          "without a backing file");
3468         goto out;
3469     }
3470 
3471     /* even though we are not necessarily operating on bs, we need it to
3472      * determine if block ops are currently prohibited on the chain */
3473     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_CHANGE, errp)) {
3474         goto out;
3475     }
3476 
3477     /* final sanity check */
3478     if (!bdrv_chain_contains(bs, image_bs)) {
3479         error_setg(errp, "'%s' and image file are not in the same chain",
3480                    device);
3481         goto out;
3482     }
3483 
3484     /* if not r/w, reopen to make r/w */
3485     ro = bdrv_is_read_only(image_bs);
3486 
3487     if (ro) {
3488         if (bdrv_reopen_set_read_only(image_bs, false, errp) != 0) {
3489             goto out;
3490         }
3491     }
3492 
3493     ret = bdrv_change_backing_file(image_bs, backing_file,
3494                                    image_bs->drv ? image_bs->drv->format_name : "",
3495                                    false);
3496 
3497     if (ret < 0) {
3498         error_setg_errno(errp, -ret, "Could not change backing file to '%s'",
3499                          backing_file);
3500         /* don't exit here, so we can try to restore open flags if
3501          * appropriate */
3502     }
3503 
3504     if (ro) {
3505         bdrv_reopen_set_read_only(image_bs, true, errp);
3506     }
3507 
3508 out:
3509     aio_context_release(aio_context);
3510 }
3511 
3512 void qmp_blockdev_add(BlockdevOptions *options, Error **errp)
3513 {
3514     BlockDriverState *bs;
3515     QObject *obj;
3516     Visitor *v = qobject_output_visitor_new(&obj);
3517     QDict *qdict;
3518 
3519     visit_type_BlockdevOptions(v, NULL, &options, &error_abort);
3520     visit_complete(v, &obj);
3521     qdict = qobject_to(QDict, obj);
3522 
3523     qdict_flatten(qdict);
3524 
3525     if (!qdict_get_try_str(qdict, "node-name")) {
3526         error_setg(errp, "'node-name' must be specified for the root node");
3527         goto fail;
3528     }
3529 
3530     bs = bds_tree_init(qdict, errp);
3531     if (!bs) {
3532         goto fail;
3533     }
3534 
3535     bdrv_set_monitor_owned(bs);
3536 
3537 fail:
3538     visit_free(v);
3539 }
3540 
3541 void qmp_x_blockdev_reopen(BlockdevOptions *options, Error **errp)
3542 {
3543     BlockDriverState *bs;
3544     AioContext *ctx;
3545     QObject *obj;
3546     Visitor *v = qobject_output_visitor_new(&obj);
3547     BlockReopenQueue *queue;
3548     QDict *qdict;
3549 
3550     /* Check for the selected node name */
3551     if (!options->has_node_name) {
3552         error_setg(errp, "Node name not specified");
3553         goto fail;
3554     }
3555 
3556     bs = bdrv_find_node(options->node_name);
3557     if (!bs) {
3558         error_setg(errp, "Cannot find node named '%s'", options->node_name);
3559         goto fail;
3560     }
3561 
3562     /* Put all options in a QDict and flatten it */
3563     visit_type_BlockdevOptions(v, NULL, &options, &error_abort);
3564     visit_complete(v, &obj);
3565     qdict = qobject_to(QDict, obj);
3566 
3567     qdict_flatten(qdict);
3568 
3569     /* Perform the reopen operation */
3570     ctx = bdrv_get_aio_context(bs);
3571     aio_context_acquire(ctx);
3572     bdrv_subtree_drained_begin(bs);
3573     queue = bdrv_reopen_queue(NULL, bs, qdict, false);
3574     bdrv_reopen_multiple(queue, errp);
3575     bdrv_subtree_drained_end(bs);
3576     aio_context_release(ctx);
3577 
3578 fail:
3579     visit_free(v);
3580 }
3581 
3582 void qmp_blockdev_del(const char *node_name, Error **errp)
3583 {
3584     AioContext *aio_context;
3585     BlockDriverState *bs;
3586 
3587     bs = bdrv_find_node(node_name);
3588     if (!bs) {
3589         error_setg(errp, "Cannot find node %s", node_name);
3590         return;
3591     }
3592     if (bdrv_has_blk(bs)) {
3593         error_setg(errp, "Node %s is in use", node_name);
3594         return;
3595     }
3596     aio_context = bdrv_get_aio_context(bs);
3597     aio_context_acquire(aio_context);
3598 
3599     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_DRIVE_DEL, errp)) {
3600         goto out;
3601     }
3602 
3603     if (!QTAILQ_IN_USE(bs, monitor_list)) {
3604         error_setg(errp, "Node %s is not owned by the monitor",
3605                    bs->node_name);
3606         goto out;
3607     }
3608 
3609     if (bs->refcnt > 1) {
3610         error_setg(errp, "Block device %s is in use",
3611                    bdrv_get_device_or_node_name(bs));
3612         goto out;
3613     }
3614 
3615     QTAILQ_REMOVE(&monitor_bdrv_states, bs, monitor_list);
3616     bdrv_unref(bs);
3617 
3618 out:
3619     aio_context_release(aio_context);
3620 }
3621 
3622 static BdrvChild *bdrv_find_child(BlockDriverState *parent_bs,
3623                                   const char *child_name)
3624 {
3625     BdrvChild *child;
3626 
3627     QLIST_FOREACH(child, &parent_bs->children, next) {
3628         if (strcmp(child->name, child_name) == 0) {
3629             return child;
3630         }
3631     }
3632 
3633     return NULL;
3634 }
3635 
3636 void qmp_x_blockdev_change(const char *parent, bool has_child,
3637                            const char *child, bool has_node,
3638                            const char *node, Error **errp)
3639 {
3640     BlockDriverState *parent_bs, *new_bs = NULL;
3641     BdrvChild *p_child;
3642 
3643     parent_bs = bdrv_lookup_bs(parent, parent, errp);
3644     if (!parent_bs) {
3645         return;
3646     }
3647 
3648     if (has_child == has_node) {
3649         if (has_child) {
3650             error_setg(errp, "The parameters child and node are in conflict");
3651         } else {
3652             error_setg(errp, "Either child or node must be specified");
3653         }
3654         return;
3655     }
3656 
3657     if (has_child) {
3658         p_child = bdrv_find_child(parent_bs, child);
3659         if (!p_child) {
3660             error_setg(errp, "Node '%s' does not have child '%s'",
3661                        parent, child);
3662             return;
3663         }
3664         bdrv_del_child(parent_bs, p_child, errp);
3665     }
3666 
3667     if (has_node) {
3668         new_bs = bdrv_find_node(node);
3669         if (!new_bs) {
3670             error_setg(errp, "Node '%s' not found", node);
3671             return;
3672         }
3673         bdrv_add_child(parent_bs, new_bs, errp);
3674     }
3675 }
3676 
3677 BlockJobInfoList *qmp_query_block_jobs(Error **errp)
3678 {
3679     BlockJobInfoList *head = NULL, **p_next = &head;
3680     BlockJob *job;
3681 
3682     for (job = block_job_next(NULL); job; job = block_job_next(job)) {
3683         BlockJobInfoList *elem;
3684         AioContext *aio_context;
3685 
3686         if (block_job_is_internal(job)) {
3687             continue;
3688         }
3689         elem = g_new0(BlockJobInfoList, 1);
3690         aio_context = blk_get_aio_context(job->blk);
3691         aio_context_acquire(aio_context);
3692         elem->value = block_job_query(job, errp);
3693         aio_context_release(aio_context);
3694         if (!elem->value) {
3695             g_free(elem);
3696             qapi_free_BlockJobInfoList(head);
3697             return NULL;
3698         }
3699         *p_next = elem;
3700         p_next = &elem->next;
3701     }
3702 
3703     return head;
3704 }
3705 
3706 void qmp_x_blockdev_set_iothread(const char *node_name, StrOrNull *iothread,
3707                                  bool has_force, bool force, Error **errp)
3708 {
3709     AioContext *old_context;
3710     AioContext *new_context;
3711     BlockDriverState *bs;
3712 
3713     bs = bdrv_find_node(node_name);
3714     if (!bs) {
3715         error_setg(errp, "Cannot find node %s", node_name);
3716         return;
3717     }
3718 
3719     /* Protects against accidents. */
3720     if (!(has_force && force) && bdrv_has_blk(bs)) {
3721         error_setg(errp, "Node %s is associated with a BlockBackend and could "
3722                          "be in use (use force=true to override this check)",
3723                          node_name);
3724         return;
3725     }
3726 
3727     if (iothread->type == QTYPE_QSTRING) {
3728         IOThread *obj = iothread_by_id(iothread->u.s);
3729         if (!obj) {
3730             error_setg(errp, "Cannot find iothread %s", iothread->u.s);
3731             return;
3732         }
3733 
3734         new_context = iothread_get_aio_context(obj);
3735     } else {
3736         new_context = qemu_get_aio_context();
3737     }
3738 
3739     old_context = bdrv_get_aio_context(bs);
3740     aio_context_acquire(old_context);
3741 
3742     bdrv_try_set_aio_context(bs, new_context, errp);
3743 
3744     aio_context_release(old_context);
3745 }
3746 
3747 QemuOptsList qemu_common_drive_opts = {
3748     .name = "drive",
3749     .head = QTAILQ_HEAD_INITIALIZER(qemu_common_drive_opts.head),
3750     .desc = {
3751         {
3752             .name = "snapshot",
3753             .type = QEMU_OPT_BOOL,
3754             .help = "enable/disable snapshot mode",
3755         },{
3756             .name = "aio",
3757             .type = QEMU_OPT_STRING,
3758             .help = "host AIO implementation (threads, native, io_uring)",
3759         },{
3760             .name = BDRV_OPT_CACHE_WB,
3761             .type = QEMU_OPT_BOOL,
3762             .help = "Enable writeback mode",
3763         },{
3764             .name = "format",
3765             .type = QEMU_OPT_STRING,
3766             .help = "disk format (raw, qcow2, ...)",
3767         },{
3768             .name = "rerror",
3769             .type = QEMU_OPT_STRING,
3770             .help = "read error action",
3771         },{
3772             .name = "werror",
3773             .type = QEMU_OPT_STRING,
3774             .help = "write error action",
3775         },{
3776             .name = BDRV_OPT_READ_ONLY,
3777             .type = QEMU_OPT_BOOL,
3778             .help = "open drive file as read-only",
3779         },
3780 
3781         THROTTLE_OPTS,
3782 
3783         {
3784             .name = "throttling.group",
3785             .type = QEMU_OPT_STRING,
3786             .help = "name of the block throttling group",
3787         },{
3788             .name = "copy-on-read",
3789             .type = QEMU_OPT_BOOL,
3790             .help = "copy read data from backing file into image file",
3791         },{
3792             .name = "detect-zeroes",
3793             .type = QEMU_OPT_STRING,
3794             .help = "try to optimize zero writes (off, on, unmap)",
3795         },{
3796             .name = "stats-account-invalid",
3797             .type = QEMU_OPT_BOOL,
3798             .help = "whether to account for invalid I/O operations "
3799                     "in the statistics",
3800         },{
3801             .name = "stats-account-failed",
3802             .type = QEMU_OPT_BOOL,
3803             .help = "whether to account for failed I/O operations "
3804                     "in the statistics",
3805         },
3806         { /* end of list */ }
3807     },
3808 };
3809 
3810 QemuOptsList qemu_drive_opts = {
3811     .name = "drive",
3812     .head = QTAILQ_HEAD_INITIALIZER(qemu_drive_opts.head),
3813     .desc = {
3814         /*
3815          * no elements => accept any params
3816          * validation will happen later
3817          */
3818         { /* end of list */ }
3819     },
3820 };
3821