xref: /openbmc/qemu/blockdev.c (revision 43709a0c)
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/runstate.h"
60 #include "sysemu/replay.h"
61 #include "qemu/cutils.h"
62 #include "qemu/help_option.h"
63 #include "qemu/main-loop.h"
64 #include "qemu/throttle-options.h"
65 
66 QTAILQ_HEAD(, BlockDriverState) monitor_bdrv_states =
67     QTAILQ_HEAD_INITIALIZER(monitor_bdrv_states);
68 
69 void bdrv_set_monitor_owned(BlockDriverState *bs)
70 {
71     QTAILQ_INSERT_TAIL(&monitor_bdrv_states, bs, monitor_list);
72 }
73 
74 static const char *const if_name[IF_COUNT] = {
75     [IF_NONE] = "none",
76     [IF_IDE] = "ide",
77     [IF_SCSI] = "scsi",
78     [IF_FLOPPY] = "floppy",
79     [IF_PFLASH] = "pflash",
80     [IF_MTD] = "mtd",
81     [IF_SD] = "sd",
82     [IF_VIRTIO] = "virtio",
83     [IF_XEN] = "xen",
84 };
85 
86 static int if_max_devs[IF_COUNT] = {
87     /*
88      * Do not change these numbers!  They govern how drive option
89      * index maps to unit and bus.  That mapping is ABI.
90      *
91      * All controllers used to implement if=T drives need to support
92      * if_max_devs[T] units, for any T with if_max_devs[T] != 0.
93      * Otherwise, some index values map to "impossible" bus, unit
94      * values.
95      *
96      * For instance, if you change [IF_SCSI] to 255, -drive
97      * if=scsi,index=12 no longer means bus=1,unit=5, but
98      * bus=0,unit=12.  With an lsi53c895a controller (7 units max),
99      * the drive can't be set up.  Regression.
100      */
101     [IF_IDE] = 2,
102     [IF_SCSI] = 7,
103 };
104 
105 /**
106  * Boards may call this to offer board-by-board overrides
107  * of the default, global values.
108  */
109 void override_max_devs(BlockInterfaceType type, int max_devs)
110 {
111     BlockBackend *blk;
112     DriveInfo *dinfo;
113 
114     if (max_devs <= 0) {
115         return;
116     }
117 
118     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
119         dinfo = blk_legacy_dinfo(blk);
120         if (dinfo->type == type) {
121             fprintf(stderr, "Cannot override units-per-bus property of"
122                     " the %s interface, because a drive of that type has"
123                     " already been added.\n", if_name[type]);
124             g_assert_not_reached();
125         }
126     }
127 
128     if_max_devs[type] = max_devs;
129 }
130 
131 /*
132  * We automatically delete the drive when a device using it gets
133  * unplugged.  Questionable feature, but we can't just drop it.
134  * Device models call blockdev_mark_auto_del() to schedule the
135  * automatic deletion, and generic qdev code calls blockdev_auto_del()
136  * when deletion is actually safe.
137  */
138 void blockdev_mark_auto_del(BlockBackend *blk)
139 {
140     DriveInfo *dinfo = blk_legacy_dinfo(blk);
141     BlockJob *job;
142 
143     if (!dinfo) {
144         return;
145     }
146 
147     for (job = block_job_next(NULL); job; job = block_job_next(job)) {
148         if (block_job_has_bdrv(job, blk_bs(blk))) {
149             AioContext *aio_context = job->job.aio_context;
150             aio_context_acquire(aio_context);
151 
152             job_cancel(&job->job, false);
153 
154             aio_context_release(aio_context);
155         }
156     }
157 
158     dinfo->auto_del = 1;
159 }
160 
161 void blockdev_auto_del(BlockBackend *blk)
162 {
163     DriveInfo *dinfo = blk_legacy_dinfo(blk);
164 
165     if (dinfo && dinfo->auto_del) {
166         monitor_remove_blk(blk);
167         blk_unref(blk);
168     }
169 }
170 
171 /**
172  * Returns the current mapping of how many units per bus
173  * a particular interface can support.
174  *
175  *  A positive integer indicates n units per bus.
176  *  0 implies the mapping has not been established.
177  * -1 indicates an invalid BlockInterfaceType was given.
178  */
179 int drive_get_max_devs(BlockInterfaceType type)
180 {
181     if (type >= IF_IDE && type < IF_COUNT) {
182         return if_max_devs[type];
183     }
184 
185     return -1;
186 }
187 
188 static int drive_index_to_bus_id(BlockInterfaceType type, int index)
189 {
190     int max_devs = if_max_devs[type];
191     return max_devs ? index / max_devs : 0;
192 }
193 
194 static int drive_index_to_unit_id(BlockInterfaceType type, int index)
195 {
196     int max_devs = if_max_devs[type];
197     return max_devs ? index % max_devs : index;
198 }
199 
200 QemuOpts *drive_def(const char *optstr)
201 {
202     return qemu_opts_parse_noisily(qemu_find_opts("drive"), optstr, false);
203 }
204 
205 QemuOpts *drive_add(BlockInterfaceType type, int index, const char *file,
206                     const char *optstr)
207 {
208     QemuOpts *opts;
209 
210     opts = drive_def(optstr);
211     if (!opts) {
212         return NULL;
213     }
214     if (type != IF_DEFAULT) {
215         qemu_opt_set(opts, "if", if_name[type], &error_abort);
216     }
217     if (index >= 0) {
218         qemu_opt_set_number(opts, "index", index, &error_abort);
219     }
220     if (file)
221         qemu_opt_set(opts, "file", file, &error_abort);
222     return opts;
223 }
224 
225 DriveInfo *drive_get(BlockInterfaceType type, int bus, int unit)
226 {
227     BlockBackend *blk;
228     DriveInfo *dinfo;
229 
230     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
231         dinfo = blk_legacy_dinfo(blk);
232         if (dinfo && dinfo->type == type
233             && dinfo->bus == bus && dinfo->unit == unit) {
234             return dinfo;
235         }
236     }
237 
238     return NULL;
239 }
240 
241 /*
242  * Check board claimed all -drive that are meant to be claimed.
243  * Fatal error if any remain unclaimed.
244  */
245 void drive_check_orphaned(void)
246 {
247     BlockBackend *blk;
248     DriveInfo *dinfo;
249     Location loc;
250     bool orphans = false;
251 
252     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
253         dinfo = blk_legacy_dinfo(blk);
254         /*
255          * Ignore default drives, because we create certain default
256          * drives unconditionally, then leave them unclaimed.  Not the
257          * users fault.
258          * Ignore IF_VIRTIO, because it gets desugared into -device,
259          * so we can leave failing to -device.
260          * Ignore IF_NONE, because leaving unclaimed IF_NONE remains
261          * available for device_add is a feature.
262          */
263         if (dinfo->is_default || dinfo->type == IF_VIRTIO
264             || dinfo->type == IF_NONE) {
265             continue;
266         }
267         if (!blk_get_attached_dev(blk)) {
268             loc_push_none(&loc);
269             qemu_opts_loc_restore(dinfo->opts);
270             error_report("machine type does not support"
271                          " if=%s,bus=%d,unit=%d",
272                          if_name[dinfo->type], dinfo->bus, dinfo->unit);
273             loc_pop(&loc);
274             orphans = true;
275         }
276     }
277 
278     if (orphans) {
279         exit(1);
280     }
281 }
282 
283 DriveInfo *drive_get_by_index(BlockInterfaceType type, int index)
284 {
285     return drive_get(type,
286                      drive_index_to_bus_id(type, index),
287                      drive_index_to_unit_id(type, index));
288 }
289 
290 int drive_get_max_bus(BlockInterfaceType type)
291 {
292     int max_bus;
293     BlockBackend *blk;
294     DriveInfo *dinfo;
295 
296     max_bus = -1;
297     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
298         dinfo = blk_legacy_dinfo(blk);
299         if (dinfo && dinfo->type == type && dinfo->bus > max_bus) {
300             max_bus = dinfo->bus;
301         }
302     }
303     return max_bus;
304 }
305 
306 /* Get a block device.  This should only be used for single-drive devices
307    (e.g. SD/Floppy/MTD).  Multi-disk devices (scsi/ide) should use the
308    appropriate bus.  */
309 DriveInfo *drive_get_next(BlockInterfaceType type)
310 {
311     static int next_block_unit[IF_COUNT];
312 
313     return drive_get(type, 0, next_block_unit[type]++);
314 }
315 
316 static void bdrv_format_print(void *opaque, const char *name)
317 {
318     qemu_printf(" %s", name);
319 }
320 
321 typedef struct {
322     QEMUBH *bh;
323     BlockDriverState *bs;
324 } BDRVPutRefBH;
325 
326 static int parse_block_error_action(const char *buf, bool is_read, Error **errp)
327 {
328     if (!strcmp(buf, "ignore")) {
329         return BLOCKDEV_ON_ERROR_IGNORE;
330     } else if (!is_read && !strcmp(buf, "enospc")) {
331         return BLOCKDEV_ON_ERROR_ENOSPC;
332     } else if (!strcmp(buf, "stop")) {
333         return BLOCKDEV_ON_ERROR_STOP;
334     } else if (!strcmp(buf, "report")) {
335         return BLOCKDEV_ON_ERROR_REPORT;
336     } else {
337         error_setg(errp, "'%s' invalid %s error action",
338                    buf, is_read ? "read" : "write");
339         return -1;
340     }
341 }
342 
343 static bool parse_stats_intervals(BlockAcctStats *stats, QList *intervals,
344                                   Error **errp)
345 {
346     const QListEntry *entry;
347     for (entry = qlist_first(intervals); entry; entry = qlist_next(entry)) {
348         switch (qobject_type(entry->value)) {
349 
350         case QTYPE_QSTRING: {
351             unsigned long long length;
352             const char *str = qstring_get_str(qobject_to(QString,
353                                                          entry->value));
354             if (parse_uint_full(str, &length, 10) == 0 &&
355                 length > 0 && length <= UINT_MAX) {
356                 block_acct_add_interval(stats, (unsigned) length);
357             } else {
358                 error_setg(errp, "Invalid interval length: %s", str);
359                 return false;
360             }
361             break;
362         }
363 
364         case QTYPE_QNUM: {
365             int64_t length = qnum_get_int(qobject_to(QNum, entry->value));
366 
367             if (length > 0 && length <= UINT_MAX) {
368                 block_acct_add_interval(stats, (unsigned) length);
369             } else {
370                 error_setg(errp, "Invalid interval length: %" PRId64, length);
371                 return false;
372             }
373             break;
374         }
375 
376         default:
377             error_setg(errp, "The specification of stats-intervals is invalid");
378             return false;
379         }
380     }
381     return true;
382 }
383 
384 typedef enum { MEDIA_DISK, MEDIA_CDROM } DriveMediaType;
385 
386 /* All parameters but @opts are optional and may be set to NULL. */
387 static void extract_common_blockdev_options(QemuOpts *opts, int *bdrv_flags,
388     const char **throttling_group, ThrottleConfig *throttle_cfg,
389     BlockdevDetectZeroesOptions *detect_zeroes, Error **errp)
390 {
391     Error *local_error = NULL;
392     const char *aio;
393 
394     if (bdrv_flags) {
395         if (qemu_opt_get_bool(opts, "copy-on-read", false)) {
396             *bdrv_flags |= BDRV_O_COPY_ON_READ;
397         }
398 
399         if ((aio = qemu_opt_get(opts, "aio")) != NULL) {
400             if (bdrv_parse_aio(aio, bdrv_flags) < 0) {
401                 error_setg(errp, "invalid aio option");
402                 return;
403             }
404         }
405     }
406 
407     /* disk I/O throttling */
408     if (throttling_group) {
409         *throttling_group = qemu_opt_get(opts, "throttling.group");
410     }
411 
412     if (throttle_cfg) {
413         throttle_config_init(throttle_cfg);
414         throttle_cfg->buckets[THROTTLE_BPS_TOTAL].avg =
415             qemu_opt_get_number(opts, "throttling.bps-total", 0);
416         throttle_cfg->buckets[THROTTLE_BPS_READ].avg  =
417             qemu_opt_get_number(opts, "throttling.bps-read", 0);
418         throttle_cfg->buckets[THROTTLE_BPS_WRITE].avg =
419             qemu_opt_get_number(opts, "throttling.bps-write", 0);
420         throttle_cfg->buckets[THROTTLE_OPS_TOTAL].avg =
421             qemu_opt_get_number(opts, "throttling.iops-total", 0);
422         throttle_cfg->buckets[THROTTLE_OPS_READ].avg =
423             qemu_opt_get_number(opts, "throttling.iops-read", 0);
424         throttle_cfg->buckets[THROTTLE_OPS_WRITE].avg =
425             qemu_opt_get_number(opts, "throttling.iops-write", 0);
426 
427         throttle_cfg->buckets[THROTTLE_BPS_TOTAL].max =
428             qemu_opt_get_number(opts, "throttling.bps-total-max", 0);
429         throttle_cfg->buckets[THROTTLE_BPS_READ].max  =
430             qemu_opt_get_number(opts, "throttling.bps-read-max", 0);
431         throttle_cfg->buckets[THROTTLE_BPS_WRITE].max =
432             qemu_opt_get_number(opts, "throttling.bps-write-max", 0);
433         throttle_cfg->buckets[THROTTLE_OPS_TOTAL].max =
434             qemu_opt_get_number(opts, "throttling.iops-total-max", 0);
435         throttle_cfg->buckets[THROTTLE_OPS_READ].max =
436             qemu_opt_get_number(opts, "throttling.iops-read-max", 0);
437         throttle_cfg->buckets[THROTTLE_OPS_WRITE].max =
438             qemu_opt_get_number(opts, "throttling.iops-write-max", 0);
439 
440         throttle_cfg->buckets[THROTTLE_BPS_TOTAL].burst_length =
441             qemu_opt_get_number(opts, "throttling.bps-total-max-length", 1);
442         throttle_cfg->buckets[THROTTLE_BPS_READ].burst_length  =
443             qemu_opt_get_number(opts, "throttling.bps-read-max-length", 1);
444         throttle_cfg->buckets[THROTTLE_BPS_WRITE].burst_length =
445             qemu_opt_get_number(opts, "throttling.bps-write-max-length", 1);
446         throttle_cfg->buckets[THROTTLE_OPS_TOTAL].burst_length =
447             qemu_opt_get_number(opts, "throttling.iops-total-max-length", 1);
448         throttle_cfg->buckets[THROTTLE_OPS_READ].burst_length =
449             qemu_opt_get_number(opts, "throttling.iops-read-max-length", 1);
450         throttle_cfg->buckets[THROTTLE_OPS_WRITE].burst_length =
451             qemu_opt_get_number(opts, "throttling.iops-write-max-length", 1);
452 
453         throttle_cfg->op_size =
454             qemu_opt_get_number(opts, "throttling.iops-size", 0);
455 
456         if (!throttle_is_valid(throttle_cfg, errp)) {
457             return;
458         }
459     }
460 
461     if (detect_zeroes) {
462         *detect_zeroes =
463             qapi_enum_parse(&BlockdevDetectZeroesOptions_lookup,
464                             qemu_opt_get(opts, "detect-zeroes"),
465                             BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF,
466                             &local_error);
467         if (local_error) {
468             error_propagate(errp, local_error);
469             return;
470         }
471     }
472 }
473 
474 /* Takes the ownership of bs_opts */
475 static BlockBackend *blockdev_init(const char *file, QDict *bs_opts,
476                                    Error **errp)
477 {
478     const char *buf;
479     int bdrv_flags = 0;
480     int on_read_error, on_write_error;
481     bool account_invalid, account_failed;
482     bool writethrough, read_only;
483     BlockBackend *blk;
484     BlockDriverState *bs;
485     ThrottleConfig cfg;
486     int snapshot = 0;
487     Error *error = NULL;
488     QemuOpts *opts;
489     QDict *interval_dict = NULL;
490     QList *interval_list = NULL;
491     const char *id;
492     BlockdevDetectZeroesOptions detect_zeroes =
493         BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF;
494     const char *throttling_group = NULL;
495 
496     /* Check common options by copying from bs_opts to opts, all other options
497      * stay in bs_opts for processing by bdrv_open(). */
498     id = qdict_get_try_str(bs_opts, "id");
499     opts = qemu_opts_create(&qemu_common_drive_opts, id, 1, errp);
500     if (!opts) {
501         goto err_no_opts;
502     }
503 
504     if (!qemu_opts_absorb_qdict(opts, bs_opts, errp)) {
505         goto early_err;
506     }
507 
508     if (id) {
509         qdict_del(bs_opts, "id");
510     }
511 
512     /* extract parameters */
513     snapshot = qemu_opt_get_bool(opts, "snapshot", 0);
514 
515     account_invalid = qemu_opt_get_bool(opts, "stats-account-invalid", true);
516     account_failed = qemu_opt_get_bool(opts, "stats-account-failed", true);
517 
518     writethrough = !qemu_opt_get_bool(opts, BDRV_OPT_CACHE_WB, true);
519 
520     id = qemu_opts_id(opts);
521 
522     qdict_extract_subqdict(bs_opts, &interval_dict, "stats-intervals.");
523     qdict_array_split(interval_dict, &interval_list);
524 
525     if (qdict_size(interval_dict) != 0) {
526         error_setg(errp, "Invalid option stats-intervals.%s",
527                    qdict_first(interval_dict)->key);
528         goto early_err;
529     }
530 
531     extract_common_blockdev_options(opts, &bdrv_flags, &throttling_group, &cfg,
532                                     &detect_zeroes, &error);
533     if (error) {
534         error_propagate(errp, error);
535         goto early_err;
536     }
537 
538     if ((buf = qemu_opt_get(opts, "format")) != NULL) {
539         if (is_help_option(buf)) {
540             qemu_printf("Supported formats:");
541             bdrv_iterate_format(bdrv_format_print, NULL, false);
542             qemu_printf("\nSupported formats (read-only):");
543             bdrv_iterate_format(bdrv_format_print, NULL, true);
544             qemu_printf("\n");
545             goto early_err;
546         }
547 
548         if (qdict_haskey(bs_opts, "driver")) {
549             error_setg(errp, "Cannot specify both 'driver' and 'format'");
550             goto early_err;
551         }
552         qdict_put_str(bs_opts, "driver", buf);
553     }
554 
555     on_write_error = BLOCKDEV_ON_ERROR_ENOSPC;
556     if ((buf = qemu_opt_get(opts, "werror")) != NULL) {
557         on_write_error = parse_block_error_action(buf, 0, &error);
558         if (error) {
559             error_propagate(errp, error);
560             goto early_err;
561         }
562     }
563 
564     on_read_error = BLOCKDEV_ON_ERROR_REPORT;
565     if ((buf = qemu_opt_get(opts, "rerror")) != NULL) {
566         on_read_error = parse_block_error_action(buf, 1, &error);
567         if (error) {
568             error_propagate(errp, error);
569             goto early_err;
570         }
571     }
572 
573     if (snapshot) {
574         bdrv_flags |= BDRV_O_SNAPSHOT;
575     }
576 
577     read_only = qemu_opt_get_bool(opts, BDRV_OPT_READ_ONLY, false);
578 
579     /* init */
580     if ((!file || !*file) && !qdict_size(bs_opts)) {
581         BlockBackendRootState *blk_rs;
582 
583         blk = blk_new(qemu_get_aio_context(), 0, BLK_PERM_ALL);
584         blk_rs = blk_get_root_state(blk);
585         blk_rs->open_flags    = bdrv_flags | (read_only ? 0 : BDRV_O_RDWR);
586         blk_rs->detect_zeroes = detect_zeroes;
587 
588         qobject_unref(bs_opts);
589     } else {
590         if (file && !*file) {
591             file = NULL;
592         }
593 
594         /* bdrv_open() defaults to the values in bdrv_flags (for compatibility
595          * with other callers) rather than what we want as the real defaults.
596          * Apply the defaults here instead. */
597         qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_DIRECT, "off");
598         qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_NO_FLUSH, "off");
599         qdict_set_default_str(bs_opts, BDRV_OPT_READ_ONLY,
600                               read_only ? "on" : "off");
601         qdict_set_default_str(bs_opts, BDRV_OPT_AUTO_READ_ONLY, "on");
602         assert((bdrv_flags & BDRV_O_CACHE_MASK) == 0);
603 
604         if (runstate_check(RUN_STATE_INMIGRATE)) {
605             bdrv_flags |= BDRV_O_INACTIVE;
606         }
607 
608         blk = blk_new_open(file, NULL, bs_opts, bdrv_flags, errp);
609         if (!blk) {
610             goto err_no_bs_opts;
611         }
612         bs = blk_bs(blk);
613 
614         bs->detect_zeroes = detect_zeroes;
615 
616         block_acct_setup(blk_get_stats(blk), account_invalid, account_failed);
617 
618         if (!parse_stats_intervals(blk_get_stats(blk), interval_list, errp)) {
619             blk_unref(blk);
620             blk = NULL;
621             goto err_no_bs_opts;
622         }
623     }
624 
625     /* disk I/O throttling */
626     if (throttle_enabled(&cfg)) {
627         if (!throttling_group) {
628             throttling_group = id;
629         }
630         blk_io_limits_enable(blk, throttling_group);
631         blk_set_io_limits(blk, &cfg);
632     }
633 
634     blk_set_enable_write_cache(blk, !writethrough);
635     blk_set_on_error(blk, on_read_error, on_write_error);
636 
637     if (!monitor_add_blk(blk, id, errp)) {
638         blk_unref(blk);
639         blk = NULL;
640         goto err_no_bs_opts;
641     }
642 
643 err_no_bs_opts:
644     qemu_opts_del(opts);
645     qobject_unref(interval_dict);
646     qobject_unref(interval_list);
647     return blk;
648 
649 early_err:
650     qemu_opts_del(opts);
651     qobject_unref(interval_dict);
652     qobject_unref(interval_list);
653 err_no_opts:
654     qobject_unref(bs_opts);
655     return NULL;
656 }
657 
658 /* Takes the ownership of bs_opts */
659 BlockDriverState *bds_tree_init(QDict *bs_opts, Error **errp)
660 {
661     int bdrv_flags = 0;
662 
663     /* bdrv_open() defaults to the values in bdrv_flags (for compatibility
664      * with other callers) rather than what we want as the real defaults.
665      * Apply the defaults here instead. */
666     qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_DIRECT, "off");
667     qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_NO_FLUSH, "off");
668     qdict_set_default_str(bs_opts, BDRV_OPT_READ_ONLY, "off");
669 
670     if (runstate_check(RUN_STATE_INMIGRATE)) {
671         bdrv_flags |= BDRV_O_INACTIVE;
672     }
673 
674     return bdrv_open(NULL, NULL, bs_opts, bdrv_flags, errp);
675 }
676 
677 void blockdev_close_all_bdrv_states(void)
678 {
679     BlockDriverState *bs, *next_bs;
680 
681     QTAILQ_FOREACH_SAFE(bs, &monitor_bdrv_states, monitor_list, next_bs) {
682         AioContext *ctx = bdrv_get_aio_context(bs);
683 
684         aio_context_acquire(ctx);
685         bdrv_unref(bs);
686         aio_context_release(ctx);
687     }
688 }
689 
690 /* Iterates over the list of monitor-owned BlockDriverStates */
691 BlockDriverState *bdrv_next_monitor_owned(BlockDriverState *bs)
692 {
693     return bs ? QTAILQ_NEXT(bs, monitor_list)
694               : QTAILQ_FIRST(&monitor_bdrv_states);
695 }
696 
697 static bool qemu_opt_rename(QemuOpts *opts, const char *from, const char *to,
698                             Error **errp)
699 {
700     const char *value;
701 
702     value = qemu_opt_get(opts, from);
703     if (value) {
704         if (qemu_opt_find(opts, to)) {
705             error_setg(errp, "'%s' and its alias '%s' can't be used at the "
706                        "same time", to, from);
707             return false;
708         }
709     }
710 
711     /* rename all items in opts */
712     while ((value = qemu_opt_get(opts, from))) {
713         qemu_opt_set(opts, to, value, &error_abort);
714         qemu_opt_unset(opts, from);
715     }
716     return true;
717 }
718 
719 QemuOptsList qemu_legacy_drive_opts = {
720     .name = "drive",
721     .head = QTAILQ_HEAD_INITIALIZER(qemu_legacy_drive_opts.head),
722     .desc = {
723         {
724             .name = "bus",
725             .type = QEMU_OPT_NUMBER,
726             .help = "bus number",
727         },{
728             .name = "unit",
729             .type = QEMU_OPT_NUMBER,
730             .help = "unit number (i.e. lun for scsi)",
731         },{
732             .name = "index",
733             .type = QEMU_OPT_NUMBER,
734             .help = "index number",
735         },{
736             .name = "media",
737             .type = QEMU_OPT_STRING,
738             .help = "media type (disk, cdrom)",
739         },{
740             .name = "if",
741             .type = QEMU_OPT_STRING,
742             .help = "interface (ide, scsi, sd, mtd, floppy, pflash, virtio)",
743         },{
744             .name = "file",
745             .type = QEMU_OPT_STRING,
746             .help = "file name",
747         },
748 
749         /* Options that are passed on, but have special semantics with -drive */
750         {
751             .name = BDRV_OPT_READ_ONLY,
752             .type = QEMU_OPT_BOOL,
753             .help = "open drive file as read-only",
754         },{
755             .name = "rerror",
756             .type = QEMU_OPT_STRING,
757             .help = "read error action",
758         },{
759             .name = "werror",
760             .type = QEMU_OPT_STRING,
761             .help = "write error action",
762         },{
763             .name = "copy-on-read",
764             .type = QEMU_OPT_BOOL,
765             .help = "copy read data from backing file into image file",
766         },
767 
768         { /* end of list */ }
769     },
770 };
771 
772 DriveInfo *drive_new(QemuOpts *all_opts, BlockInterfaceType block_default_type,
773                      Error **errp)
774 {
775     const char *value;
776     BlockBackend *blk;
777     DriveInfo *dinfo = NULL;
778     QDict *bs_opts;
779     QemuOpts *legacy_opts;
780     DriveMediaType media = MEDIA_DISK;
781     BlockInterfaceType type;
782     int max_devs, bus_id, unit_id, index;
783     const char *werror, *rerror;
784     bool read_only = false;
785     bool copy_on_read;
786     const char *filename;
787     int i;
788 
789     /* Change legacy command line options into QMP ones */
790     static const struct {
791         const char *from;
792         const char *to;
793     } opt_renames[] = {
794         { "iops",           "throttling.iops-total" },
795         { "iops_rd",        "throttling.iops-read" },
796         { "iops_wr",        "throttling.iops-write" },
797 
798         { "bps",            "throttling.bps-total" },
799         { "bps_rd",         "throttling.bps-read" },
800         { "bps_wr",         "throttling.bps-write" },
801 
802         { "iops_max",       "throttling.iops-total-max" },
803         { "iops_rd_max",    "throttling.iops-read-max" },
804         { "iops_wr_max",    "throttling.iops-write-max" },
805 
806         { "bps_max",        "throttling.bps-total-max" },
807         { "bps_rd_max",     "throttling.bps-read-max" },
808         { "bps_wr_max",     "throttling.bps-write-max" },
809 
810         { "iops_size",      "throttling.iops-size" },
811 
812         { "group",          "throttling.group" },
813 
814         { "readonly",       BDRV_OPT_READ_ONLY },
815     };
816 
817     for (i = 0; i < ARRAY_SIZE(opt_renames); i++) {
818         if (!qemu_opt_rename(all_opts, opt_renames[i].from,
819                              opt_renames[i].to, errp)) {
820             return NULL;
821         }
822     }
823 
824     value = qemu_opt_get(all_opts, "cache");
825     if (value) {
826         int flags = 0;
827         bool writethrough;
828 
829         if (bdrv_parse_cache_mode(value, &flags, &writethrough) != 0) {
830             error_setg(errp, "invalid cache option");
831             return NULL;
832         }
833 
834         /* Specific options take precedence */
835         if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_WB)) {
836             qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_WB,
837                               !writethrough, &error_abort);
838         }
839         if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_DIRECT)) {
840             qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_DIRECT,
841                               !!(flags & BDRV_O_NOCACHE), &error_abort);
842         }
843         if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_NO_FLUSH)) {
844             qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_NO_FLUSH,
845                               !!(flags & BDRV_O_NO_FLUSH), &error_abort);
846         }
847         qemu_opt_unset(all_opts, "cache");
848     }
849 
850     /* Get a QDict for processing the options */
851     bs_opts = qdict_new();
852     qemu_opts_to_qdict(all_opts, bs_opts);
853 
854     legacy_opts = qemu_opts_create(&qemu_legacy_drive_opts, NULL, 0,
855                                    &error_abort);
856     if (!qemu_opts_absorb_qdict(legacy_opts, bs_opts, errp)) {
857         goto fail;
858     }
859 
860     /* Media type */
861     value = qemu_opt_get(legacy_opts, "media");
862     if (value) {
863         if (!strcmp(value, "disk")) {
864             media = MEDIA_DISK;
865         } else if (!strcmp(value, "cdrom")) {
866             media = MEDIA_CDROM;
867             read_only = true;
868         } else {
869             error_setg(errp, "'%s' invalid media", value);
870             goto fail;
871         }
872     }
873 
874     /* copy-on-read is disabled with a warning for read-only devices */
875     read_only |= qemu_opt_get_bool(legacy_opts, BDRV_OPT_READ_ONLY, false);
876     copy_on_read = qemu_opt_get_bool(legacy_opts, "copy-on-read", false);
877 
878     if (read_only && copy_on_read) {
879         warn_report("disabling copy-on-read on read-only drive");
880         copy_on_read = false;
881     }
882 
883     qdict_put_str(bs_opts, BDRV_OPT_READ_ONLY, read_only ? "on" : "off");
884     qdict_put_str(bs_opts, "copy-on-read", copy_on_read ? "on" : "off");
885 
886     /* Controller type */
887     value = qemu_opt_get(legacy_opts, "if");
888     if (value) {
889         for (type = 0;
890              type < IF_COUNT && strcmp(value, if_name[type]);
891              type++) {
892         }
893         if (type == IF_COUNT) {
894             error_setg(errp, "unsupported bus type '%s'", value);
895             goto fail;
896         }
897     } else {
898         type = block_default_type;
899     }
900 
901     /* Device address specified by bus/unit or index.
902      * If none was specified, try to find the first free one. */
903     bus_id  = qemu_opt_get_number(legacy_opts, "bus", 0);
904     unit_id = qemu_opt_get_number(legacy_opts, "unit", -1);
905     index   = qemu_opt_get_number(legacy_opts, "index", -1);
906 
907     max_devs = if_max_devs[type];
908 
909     if (index != -1) {
910         if (bus_id != 0 || unit_id != -1) {
911             error_setg(errp, "index cannot be used with bus and unit");
912             goto fail;
913         }
914         bus_id = drive_index_to_bus_id(type, index);
915         unit_id = drive_index_to_unit_id(type, index);
916     }
917 
918     if (unit_id == -1) {
919        unit_id = 0;
920        while (drive_get(type, bus_id, unit_id) != NULL) {
921            unit_id++;
922            if (max_devs && unit_id >= max_devs) {
923                unit_id -= max_devs;
924                bus_id++;
925            }
926        }
927     }
928 
929     if (max_devs && unit_id >= max_devs) {
930         error_setg(errp, "unit %d too big (max is %d)", unit_id, max_devs - 1);
931         goto fail;
932     }
933 
934     if (drive_get(type, bus_id, unit_id) != NULL) {
935         error_setg(errp, "drive with bus=%d, unit=%d (index=%d) exists",
936                    bus_id, unit_id, index);
937         goto fail;
938     }
939 
940     /* no id supplied -> create one */
941     if (qemu_opts_id(all_opts) == NULL) {
942         char *new_id;
943         const char *mediastr = "";
944         if (type == IF_IDE || type == IF_SCSI) {
945             mediastr = (media == MEDIA_CDROM) ? "-cd" : "-hd";
946         }
947         if (max_devs) {
948             new_id = g_strdup_printf("%s%i%s%i", if_name[type], bus_id,
949                                      mediastr, unit_id);
950         } else {
951             new_id = g_strdup_printf("%s%s%i", if_name[type],
952                                      mediastr, unit_id);
953         }
954         qdict_put_str(bs_opts, "id", new_id);
955         g_free(new_id);
956     }
957 
958     /* Add virtio block device */
959     if (type == IF_VIRTIO) {
960         QemuOpts *devopts;
961         devopts = qemu_opts_create(qemu_find_opts("device"), NULL, 0,
962                                    &error_abort);
963         qemu_opt_set(devopts, "driver", "virtio-blk", &error_abort);
964         qemu_opt_set(devopts, "drive", qdict_get_str(bs_opts, "id"),
965                      &error_abort);
966     }
967 
968     filename = qemu_opt_get(legacy_opts, "file");
969 
970     /* Check werror/rerror compatibility with if=... */
971     werror = qemu_opt_get(legacy_opts, "werror");
972     if (werror != NULL) {
973         if (type != IF_IDE && type != IF_SCSI && type != IF_VIRTIO &&
974             type != IF_NONE) {
975             error_setg(errp, "werror is not supported by this bus type");
976             goto fail;
977         }
978         qdict_put_str(bs_opts, "werror", werror);
979     }
980 
981     rerror = qemu_opt_get(legacy_opts, "rerror");
982     if (rerror != NULL) {
983         if (type != IF_IDE && type != IF_VIRTIO && type != IF_SCSI &&
984             type != IF_NONE) {
985             error_setg(errp, "rerror is not supported by this bus type");
986             goto fail;
987         }
988         qdict_put_str(bs_opts, "rerror", rerror);
989     }
990 
991     /* Actual block device init: Functionality shared with blockdev-add */
992     blk = blockdev_init(filename, bs_opts, errp);
993     bs_opts = NULL;
994     if (!blk) {
995         goto fail;
996     }
997 
998     /* Create legacy DriveInfo */
999     dinfo = g_malloc0(sizeof(*dinfo));
1000     dinfo->opts = all_opts;
1001 
1002     dinfo->type = type;
1003     dinfo->bus = bus_id;
1004     dinfo->unit = unit_id;
1005 
1006     blk_set_legacy_dinfo(blk, dinfo);
1007 
1008     switch(type) {
1009     case IF_IDE:
1010     case IF_SCSI:
1011     case IF_XEN:
1012     case IF_NONE:
1013         dinfo->media_cd = media == MEDIA_CDROM;
1014         break;
1015     default:
1016         break;
1017     }
1018 
1019 fail:
1020     qemu_opts_del(legacy_opts);
1021     qobject_unref(bs_opts);
1022     return dinfo;
1023 }
1024 
1025 static BlockDriverState *qmp_get_root_bs(const char *name, Error **errp)
1026 {
1027     BlockDriverState *bs;
1028 
1029     bs = bdrv_lookup_bs(name, name, errp);
1030     if (bs == NULL) {
1031         return NULL;
1032     }
1033 
1034     if (!bdrv_is_root_node(bs)) {
1035         error_setg(errp, "Need a root block node");
1036         return NULL;
1037     }
1038 
1039     if (!bdrv_is_inserted(bs)) {
1040         error_setg(errp, "Device has no medium");
1041         return NULL;
1042     }
1043 
1044     return bs;
1045 }
1046 
1047 static void blockdev_do_action(TransactionAction *action, Error **errp)
1048 {
1049     TransactionActionList list;
1050 
1051     list.value = action;
1052     list.next = NULL;
1053     qmp_transaction(&list, false, NULL, errp);
1054 }
1055 
1056 void qmp_blockdev_snapshot_sync(bool has_device, const char *device,
1057                                 bool has_node_name, const char *node_name,
1058                                 const char *snapshot_file,
1059                                 bool has_snapshot_node_name,
1060                                 const char *snapshot_node_name,
1061                                 bool has_format, const char *format,
1062                                 bool has_mode, NewImageMode mode, Error **errp)
1063 {
1064     BlockdevSnapshotSync snapshot = {
1065         .has_device = has_device,
1066         .device = (char *) device,
1067         .has_node_name = has_node_name,
1068         .node_name = (char *) node_name,
1069         .snapshot_file = (char *) snapshot_file,
1070         .has_snapshot_node_name = has_snapshot_node_name,
1071         .snapshot_node_name = (char *) snapshot_node_name,
1072         .has_format = has_format,
1073         .format = (char *) format,
1074         .has_mode = has_mode,
1075         .mode = mode,
1076     };
1077     TransactionAction action = {
1078         .type = TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC,
1079         .u.blockdev_snapshot_sync.data = &snapshot,
1080     };
1081     blockdev_do_action(&action, errp);
1082 }
1083 
1084 void qmp_blockdev_snapshot(const char *node, const char *overlay,
1085                            Error **errp)
1086 {
1087     BlockdevSnapshot snapshot_data = {
1088         .node = (char *) node,
1089         .overlay = (char *) overlay
1090     };
1091     TransactionAction action = {
1092         .type = TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT,
1093         .u.blockdev_snapshot.data = &snapshot_data,
1094     };
1095     blockdev_do_action(&action, errp);
1096 }
1097 
1098 void qmp_blockdev_snapshot_internal_sync(const char *device,
1099                                          const char *name,
1100                                          Error **errp)
1101 {
1102     BlockdevSnapshotInternal snapshot = {
1103         .device = (char *) device,
1104         .name = (char *) name
1105     };
1106     TransactionAction action = {
1107         .type = TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC,
1108         .u.blockdev_snapshot_internal_sync.data = &snapshot,
1109     };
1110     blockdev_do_action(&action, errp);
1111 }
1112 
1113 SnapshotInfo *qmp_blockdev_snapshot_delete_internal_sync(const char *device,
1114                                                          bool has_id,
1115                                                          const char *id,
1116                                                          bool has_name,
1117                                                          const char *name,
1118                                                          Error **errp)
1119 {
1120     BlockDriverState *bs;
1121     AioContext *aio_context;
1122     QEMUSnapshotInfo sn;
1123     Error *local_err = NULL;
1124     SnapshotInfo *info = NULL;
1125     int ret;
1126 
1127     bs = qmp_get_root_bs(device, errp);
1128     if (!bs) {
1129         return NULL;
1130     }
1131     aio_context = bdrv_get_aio_context(bs);
1132     aio_context_acquire(aio_context);
1133 
1134     if (!has_id) {
1135         id = NULL;
1136     }
1137 
1138     if (!has_name) {
1139         name = NULL;
1140     }
1141 
1142     if (!id && !name) {
1143         error_setg(errp, "Name or id must be provided");
1144         goto out_aio_context;
1145     }
1146 
1147     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_INTERNAL_SNAPSHOT_DELETE, errp)) {
1148         goto out_aio_context;
1149     }
1150 
1151     ret = bdrv_snapshot_find_by_id_and_name(bs, id, name, &sn, &local_err);
1152     if (local_err) {
1153         error_propagate(errp, local_err);
1154         goto out_aio_context;
1155     }
1156     if (!ret) {
1157         error_setg(errp,
1158                    "Snapshot with id '%s' and name '%s' does not exist on "
1159                    "device '%s'",
1160                    STR_OR_NULL(id), STR_OR_NULL(name), device);
1161         goto out_aio_context;
1162     }
1163 
1164     bdrv_snapshot_delete(bs, id, name, &local_err);
1165     if (local_err) {
1166         error_propagate(errp, local_err);
1167         goto out_aio_context;
1168     }
1169 
1170     aio_context_release(aio_context);
1171 
1172     info = g_new0(SnapshotInfo, 1);
1173     info->id = g_strdup(sn.id_str);
1174     info->name = g_strdup(sn.name);
1175     info->date_nsec = sn.date_nsec;
1176     info->date_sec = sn.date_sec;
1177     info->vm_state_size = sn.vm_state_size;
1178     info->vm_clock_nsec = sn.vm_clock_nsec % 1000000000;
1179     info->vm_clock_sec = sn.vm_clock_nsec / 1000000000;
1180     if (sn.icount != -1ULL) {
1181         info->icount = sn.icount;
1182         info->has_icount = true;
1183     }
1184 
1185     return info;
1186 
1187 out_aio_context:
1188     aio_context_release(aio_context);
1189     return NULL;
1190 }
1191 
1192 /* New and old BlockDriverState structs for atomic group operations */
1193 
1194 typedef struct BlkActionState BlkActionState;
1195 
1196 /**
1197  * BlkActionOps:
1198  * Table of operations that define an Action.
1199  *
1200  * @instance_size: Size of state struct, in bytes.
1201  * @prepare: Prepare the work, must NOT be NULL.
1202  * @commit: Commit the changes, can be NULL.
1203  * @abort: Abort the changes on fail, can be NULL.
1204  * @clean: Clean up resources after all transaction actions have called
1205  *         commit() or abort(). Can be NULL.
1206  *
1207  * Only prepare() may fail. In a single transaction, only one of commit() or
1208  * abort() will be called. clean() will always be called if it is present.
1209  */
1210 typedef struct BlkActionOps {
1211     size_t instance_size;
1212     void (*prepare)(BlkActionState *common, Error **errp);
1213     void (*commit)(BlkActionState *common);
1214     void (*abort)(BlkActionState *common);
1215     void (*clean)(BlkActionState *common);
1216 } BlkActionOps;
1217 
1218 /**
1219  * BlkActionState:
1220  * Describes one Action's state within a Transaction.
1221  *
1222  * @action: QAPI-defined enum identifying which Action to perform.
1223  * @ops: Table of ActionOps this Action can perform.
1224  * @block_job_txn: Transaction which this action belongs to.
1225  * @entry: List membership for all Actions in this Transaction.
1226  *
1227  * This structure must be arranged as first member in a subclassed type,
1228  * assuming that the compiler will also arrange it to the same offsets as the
1229  * base class.
1230  */
1231 struct BlkActionState {
1232     TransactionAction *action;
1233     const BlkActionOps *ops;
1234     JobTxn *block_job_txn;
1235     TransactionProperties *txn_props;
1236     QTAILQ_ENTRY(BlkActionState) entry;
1237 };
1238 
1239 /* internal snapshot private data */
1240 typedef struct InternalSnapshotState {
1241     BlkActionState common;
1242     BlockDriverState *bs;
1243     QEMUSnapshotInfo sn;
1244     bool created;
1245 } InternalSnapshotState;
1246 
1247 
1248 static int action_check_completion_mode(BlkActionState *s, Error **errp)
1249 {
1250     if (s->txn_props->completion_mode != ACTION_COMPLETION_MODE_INDIVIDUAL) {
1251         error_setg(errp,
1252                    "Action '%s' does not support Transaction property "
1253                    "completion-mode = %s",
1254                    TransactionActionKind_str(s->action->type),
1255                    ActionCompletionMode_str(s->txn_props->completion_mode));
1256         return -1;
1257     }
1258     return 0;
1259 }
1260 
1261 static void internal_snapshot_prepare(BlkActionState *common,
1262                                       Error **errp)
1263 {
1264     Error *local_err = NULL;
1265     const char *device;
1266     const char *name;
1267     BlockDriverState *bs;
1268     QEMUSnapshotInfo old_sn, *sn;
1269     bool ret;
1270     qemu_timeval tv;
1271     BlockdevSnapshotInternal *internal;
1272     InternalSnapshotState *state;
1273     AioContext *aio_context;
1274     int ret1;
1275 
1276     g_assert(common->action->type ==
1277              TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC);
1278     internal = common->action->u.blockdev_snapshot_internal_sync.data;
1279     state = DO_UPCAST(InternalSnapshotState, common, common);
1280 
1281     /* 1. parse input */
1282     device = internal->device;
1283     name = internal->name;
1284 
1285     /* 2. check for validation */
1286     if (action_check_completion_mode(common, errp) < 0) {
1287         return;
1288     }
1289 
1290     bs = qmp_get_root_bs(device, errp);
1291     if (!bs) {
1292         return;
1293     }
1294 
1295     aio_context = bdrv_get_aio_context(bs);
1296     aio_context_acquire(aio_context);
1297 
1298     state->bs = bs;
1299 
1300     /* Paired with .clean() */
1301     bdrv_drained_begin(bs);
1302 
1303     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_INTERNAL_SNAPSHOT, errp)) {
1304         goto out;
1305     }
1306 
1307     if (bdrv_is_read_only(bs)) {
1308         error_setg(errp, "Device '%s' is read only", device);
1309         goto out;
1310     }
1311 
1312     if (!bdrv_can_snapshot(bs)) {
1313         error_setg(errp, "Block format '%s' used by device '%s' "
1314                    "does not support internal snapshots",
1315                    bs->drv->format_name, device);
1316         goto out;
1317     }
1318 
1319     if (!strlen(name)) {
1320         error_setg(errp, "Name is empty");
1321         goto out;
1322     }
1323 
1324     /* check whether a snapshot with name exist */
1325     ret = bdrv_snapshot_find_by_id_and_name(bs, NULL, name, &old_sn,
1326                                             &local_err);
1327     if (local_err) {
1328         error_propagate(errp, local_err);
1329         goto out;
1330     } else if (ret) {
1331         error_setg(errp,
1332                    "Snapshot with name '%s' already exists on device '%s'",
1333                    name, device);
1334         goto out;
1335     }
1336 
1337     /* 3. take the snapshot */
1338     sn = &state->sn;
1339     pstrcpy(sn->name, sizeof(sn->name), name);
1340     qemu_gettimeofday(&tv);
1341     sn->date_sec = tv.tv_sec;
1342     sn->date_nsec = tv.tv_usec * 1000;
1343     sn->vm_clock_nsec = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL);
1344     if (replay_mode != REPLAY_MODE_NONE) {
1345         sn->icount = replay_get_current_icount();
1346     } else {
1347         sn->icount = -1ULL;
1348     }
1349 
1350     ret1 = bdrv_snapshot_create(bs, sn);
1351     if (ret1 < 0) {
1352         error_setg_errno(errp, -ret1,
1353                          "Failed to create snapshot '%s' on device '%s'",
1354                          name, device);
1355         goto out;
1356     }
1357 
1358     /* 4. succeed, mark a snapshot is created */
1359     state->created = true;
1360 
1361 out:
1362     aio_context_release(aio_context);
1363 }
1364 
1365 static void internal_snapshot_abort(BlkActionState *common)
1366 {
1367     InternalSnapshotState *state =
1368                              DO_UPCAST(InternalSnapshotState, common, common);
1369     BlockDriverState *bs = state->bs;
1370     QEMUSnapshotInfo *sn = &state->sn;
1371     AioContext *aio_context;
1372     Error *local_error = NULL;
1373 
1374     if (!state->created) {
1375         return;
1376     }
1377 
1378     aio_context = bdrv_get_aio_context(state->bs);
1379     aio_context_acquire(aio_context);
1380 
1381     if (bdrv_snapshot_delete(bs, sn->id_str, sn->name, &local_error) < 0) {
1382         error_reportf_err(local_error,
1383                           "Failed to delete snapshot with id '%s' and "
1384                           "name '%s' on device '%s' in abort: ",
1385                           sn->id_str, sn->name,
1386                           bdrv_get_device_name(bs));
1387     }
1388 
1389     aio_context_release(aio_context);
1390 }
1391 
1392 static void internal_snapshot_clean(BlkActionState *common)
1393 {
1394     InternalSnapshotState *state = DO_UPCAST(InternalSnapshotState,
1395                                              common, common);
1396     AioContext *aio_context;
1397 
1398     if (!state->bs) {
1399         return;
1400     }
1401 
1402     aio_context = bdrv_get_aio_context(state->bs);
1403     aio_context_acquire(aio_context);
1404 
1405     bdrv_drained_end(state->bs);
1406 
1407     aio_context_release(aio_context);
1408 }
1409 
1410 /* external snapshot private data */
1411 typedef struct ExternalSnapshotState {
1412     BlkActionState common;
1413     BlockDriverState *old_bs;
1414     BlockDriverState *new_bs;
1415     bool overlay_appended;
1416 } ExternalSnapshotState;
1417 
1418 static void external_snapshot_prepare(BlkActionState *common,
1419                                       Error **errp)
1420 {
1421     int ret;
1422     int flags = 0;
1423     QDict *options = NULL;
1424     Error *local_err = NULL;
1425     /* Device and node name of the image to generate the snapshot from */
1426     const char *device;
1427     const char *node_name;
1428     /* Reference to the new image (for 'blockdev-snapshot') */
1429     const char *snapshot_ref;
1430     /* File name of the new image (for 'blockdev-snapshot-sync') */
1431     const char *new_image_file;
1432     ExternalSnapshotState *state =
1433                              DO_UPCAST(ExternalSnapshotState, common, common);
1434     TransactionAction *action = common->action;
1435     AioContext *aio_context;
1436     uint64_t perm, shared;
1437 
1438     /* 'blockdev-snapshot' and 'blockdev-snapshot-sync' have similar
1439      * purpose but a different set of parameters */
1440     switch (action->type) {
1441     case TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT:
1442         {
1443             BlockdevSnapshot *s = action->u.blockdev_snapshot.data;
1444             device = s->node;
1445             node_name = s->node;
1446             new_image_file = NULL;
1447             snapshot_ref = s->overlay;
1448         }
1449         break;
1450     case TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC:
1451         {
1452             BlockdevSnapshotSync *s = action->u.blockdev_snapshot_sync.data;
1453             device = s->has_device ? s->device : NULL;
1454             node_name = s->has_node_name ? s->node_name : NULL;
1455             new_image_file = s->snapshot_file;
1456             snapshot_ref = NULL;
1457         }
1458         break;
1459     default:
1460         g_assert_not_reached();
1461     }
1462 
1463     /* start processing */
1464     if (action_check_completion_mode(common, errp) < 0) {
1465         return;
1466     }
1467 
1468     state->old_bs = bdrv_lookup_bs(device, node_name, errp);
1469     if (!state->old_bs) {
1470         return;
1471     }
1472 
1473     aio_context = bdrv_get_aio_context(state->old_bs);
1474     aio_context_acquire(aio_context);
1475 
1476     /* Paired with .clean() */
1477     bdrv_drained_begin(state->old_bs);
1478 
1479     if (!bdrv_is_inserted(state->old_bs)) {
1480         error_setg(errp, QERR_DEVICE_HAS_NO_MEDIUM, device);
1481         goto out;
1482     }
1483 
1484     if (bdrv_op_is_blocked(state->old_bs,
1485                            BLOCK_OP_TYPE_EXTERNAL_SNAPSHOT, errp)) {
1486         goto out;
1487     }
1488 
1489     if (!bdrv_is_read_only(state->old_bs)) {
1490         if (bdrv_flush(state->old_bs)) {
1491             error_setg(errp, QERR_IO_ERROR);
1492             goto out;
1493         }
1494     }
1495 
1496     if (action->type == TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC) {
1497         BlockdevSnapshotSync *s = action->u.blockdev_snapshot_sync.data;
1498         const char *format = s->has_format ? s->format : "qcow2";
1499         enum NewImageMode mode;
1500         const char *snapshot_node_name =
1501             s->has_snapshot_node_name ? s->snapshot_node_name : NULL;
1502 
1503         if (node_name && !snapshot_node_name) {
1504             error_setg(errp, "New overlay node-name missing");
1505             goto out;
1506         }
1507 
1508         if (snapshot_node_name &&
1509             bdrv_lookup_bs(snapshot_node_name, snapshot_node_name, NULL)) {
1510             error_setg(errp, "New overlay node-name already in use");
1511             goto out;
1512         }
1513 
1514         flags = state->old_bs->open_flags;
1515         flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_COPY_ON_READ);
1516         flags |= BDRV_O_NO_BACKING;
1517 
1518         /* create new image w/backing file */
1519         mode = s->has_mode ? s->mode : NEW_IMAGE_MODE_ABSOLUTE_PATHS;
1520         if (mode != NEW_IMAGE_MODE_EXISTING) {
1521             int64_t size = bdrv_getlength(state->old_bs);
1522             if (size < 0) {
1523                 error_setg_errno(errp, -size, "bdrv_getlength failed");
1524                 goto out;
1525             }
1526             bdrv_refresh_filename(state->old_bs);
1527             bdrv_img_create(new_image_file, format,
1528                             state->old_bs->filename,
1529                             state->old_bs->drv->format_name,
1530                             NULL, size, flags, false, &local_err);
1531             if (local_err) {
1532                 error_propagate(errp, local_err);
1533                 goto out;
1534             }
1535         }
1536 
1537         options = qdict_new();
1538         if (snapshot_node_name) {
1539             qdict_put_str(options, "node-name", snapshot_node_name);
1540         }
1541         qdict_put_str(options, "driver", format);
1542     }
1543 
1544     state->new_bs = bdrv_open(new_image_file, snapshot_ref, options, flags,
1545                               errp);
1546     /* We will manually add the backing_hd field to the bs later */
1547     if (!state->new_bs) {
1548         goto out;
1549     }
1550 
1551     /*
1552      * Allow attaching a backing file to an overlay that's already in use only
1553      * if the parents don't assume that they are already seeing a valid image.
1554      * (Specifically, allow it as a mirror target, which is write-only access.)
1555      */
1556     bdrv_get_cumulative_perm(state->new_bs, &perm, &shared);
1557     if (perm & BLK_PERM_CONSISTENT_READ) {
1558         error_setg(errp, "The overlay is already in use");
1559         goto out;
1560     }
1561 
1562     if (state->new_bs->drv->is_filter) {
1563         error_setg(errp, "Filters cannot be used as overlays");
1564         goto out;
1565     }
1566 
1567     if (bdrv_cow_child(state->new_bs)) {
1568         error_setg(errp, "The overlay already has a backing image");
1569         goto out;
1570     }
1571 
1572     if (!state->new_bs->drv->supports_backing) {
1573         error_setg(errp, "The overlay does not support backing images");
1574         goto out;
1575     }
1576 
1577     ret = bdrv_append(state->new_bs, state->old_bs, errp);
1578     if (ret < 0) {
1579         goto out;
1580     }
1581     state->overlay_appended = true;
1582 
1583 out:
1584     aio_context_release(aio_context);
1585 }
1586 
1587 static void external_snapshot_commit(BlkActionState *common)
1588 {
1589     ExternalSnapshotState *state =
1590                              DO_UPCAST(ExternalSnapshotState, common, common);
1591     AioContext *aio_context;
1592 
1593     aio_context = bdrv_get_aio_context(state->old_bs);
1594     aio_context_acquire(aio_context);
1595 
1596     /* We don't need (or want) to use the transactional
1597      * bdrv_reopen_multiple() across all the entries at once, because we
1598      * don't want to abort all of them if one of them fails the reopen */
1599     if (!qatomic_read(&state->old_bs->copy_on_read)) {
1600         bdrv_reopen_set_read_only(state->old_bs, true, NULL);
1601     }
1602 
1603     aio_context_release(aio_context);
1604 }
1605 
1606 static void external_snapshot_abort(BlkActionState *common)
1607 {
1608     ExternalSnapshotState *state =
1609                              DO_UPCAST(ExternalSnapshotState, common, common);
1610     if (state->new_bs) {
1611         if (state->overlay_appended) {
1612             AioContext *aio_context;
1613             AioContext *tmp_context;
1614             int ret;
1615 
1616             aio_context = bdrv_get_aio_context(state->old_bs);
1617             aio_context_acquire(aio_context);
1618 
1619             bdrv_ref(state->old_bs);   /* we can't let bdrv_set_backind_hd()
1620                                           close state->old_bs; we need it */
1621             bdrv_set_backing_hd(state->new_bs, NULL, &error_abort);
1622 
1623             /*
1624              * The call to bdrv_set_backing_hd() above returns state->old_bs to
1625              * the main AioContext. As we're still going to be using it, return
1626              * it to the AioContext it was before.
1627              */
1628             tmp_context = bdrv_get_aio_context(state->old_bs);
1629             if (aio_context != tmp_context) {
1630                 aio_context_release(aio_context);
1631                 aio_context_acquire(tmp_context);
1632 
1633                 ret = bdrv_try_set_aio_context(state->old_bs,
1634                                                aio_context, NULL);
1635                 assert(ret == 0);
1636 
1637                 aio_context_release(tmp_context);
1638                 aio_context_acquire(aio_context);
1639             }
1640 
1641             bdrv_replace_node(state->new_bs, state->old_bs, &error_abort);
1642             bdrv_unref(state->old_bs); /* bdrv_replace_node() ref'ed old_bs */
1643 
1644             aio_context_release(aio_context);
1645         }
1646     }
1647 }
1648 
1649 static void external_snapshot_clean(BlkActionState *common)
1650 {
1651     ExternalSnapshotState *state =
1652                              DO_UPCAST(ExternalSnapshotState, common, common);
1653     AioContext *aio_context;
1654 
1655     if (!state->old_bs) {
1656         return;
1657     }
1658 
1659     aio_context = bdrv_get_aio_context(state->old_bs);
1660     aio_context_acquire(aio_context);
1661 
1662     bdrv_drained_end(state->old_bs);
1663     bdrv_unref(state->new_bs);
1664 
1665     aio_context_release(aio_context);
1666 }
1667 
1668 typedef struct DriveBackupState {
1669     BlkActionState common;
1670     BlockDriverState *bs;
1671     BlockJob *job;
1672 } DriveBackupState;
1673 
1674 static BlockJob *do_backup_common(BackupCommon *backup,
1675                                   BlockDriverState *bs,
1676                                   BlockDriverState *target_bs,
1677                                   AioContext *aio_context,
1678                                   JobTxn *txn, Error **errp);
1679 
1680 static void drive_backup_prepare(BlkActionState *common, Error **errp)
1681 {
1682     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1683     DriveBackup *backup;
1684     BlockDriverState *bs;
1685     BlockDriverState *target_bs;
1686     BlockDriverState *source = NULL;
1687     AioContext *aio_context;
1688     AioContext *old_context;
1689     QDict *options;
1690     Error *local_err = NULL;
1691     int flags;
1692     int64_t size;
1693     bool set_backing_hd = false;
1694     int ret;
1695 
1696     assert(common->action->type == TRANSACTION_ACTION_KIND_DRIVE_BACKUP);
1697     backup = common->action->u.drive_backup.data;
1698 
1699     if (!backup->has_mode) {
1700         backup->mode = NEW_IMAGE_MODE_ABSOLUTE_PATHS;
1701     }
1702 
1703     bs = bdrv_lookup_bs(backup->device, backup->device, errp);
1704     if (!bs) {
1705         return;
1706     }
1707 
1708     if (!bs->drv) {
1709         error_setg(errp, "Device has no medium");
1710         return;
1711     }
1712 
1713     aio_context = bdrv_get_aio_context(bs);
1714     aio_context_acquire(aio_context);
1715 
1716     state->bs = bs;
1717     /* Paired with .clean() */
1718     bdrv_drained_begin(bs);
1719 
1720     if (!backup->has_format) {
1721         backup->format = backup->mode == NEW_IMAGE_MODE_EXISTING ?
1722                          NULL : (char *) bs->drv->format_name;
1723     }
1724 
1725     /* Early check to avoid creating target */
1726     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_BACKUP_SOURCE, errp)) {
1727         goto out;
1728     }
1729 
1730     flags = bs->open_flags | BDRV_O_RDWR;
1731 
1732     /*
1733      * See if we have a backing HD we can use to create our new image
1734      * on top of.
1735      */
1736     if (backup->sync == MIRROR_SYNC_MODE_TOP) {
1737         /*
1738          * Backup will not replace the source by the target, so none
1739          * of the filters skipped here will be removed (in contrast to
1740          * mirror).  Therefore, we can skip all of them when looking
1741          * for the first COW relationship.
1742          */
1743         source = bdrv_cow_bs(bdrv_skip_filters(bs));
1744         if (!source) {
1745             backup->sync = MIRROR_SYNC_MODE_FULL;
1746         }
1747     }
1748     if (backup->sync == MIRROR_SYNC_MODE_NONE) {
1749         source = bs;
1750         flags |= BDRV_O_NO_BACKING;
1751         set_backing_hd = true;
1752     }
1753 
1754     size = bdrv_getlength(bs);
1755     if (size < 0) {
1756         error_setg_errno(errp, -size, "bdrv_getlength failed");
1757         goto out;
1758     }
1759 
1760     if (backup->mode != NEW_IMAGE_MODE_EXISTING) {
1761         assert(backup->format);
1762         if (source) {
1763             /* Implicit filters should not appear in the filename */
1764             BlockDriverState *explicit_backing =
1765                 bdrv_skip_implicit_filters(source);
1766 
1767             bdrv_refresh_filename(explicit_backing);
1768             bdrv_img_create(backup->target, backup->format,
1769                             explicit_backing->filename,
1770                             explicit_backing->drv->format_name, NULL,
1771                             size, flags, false, &local_err);
1772         } else {
1773             bdrv_img_create(backup->target, backup->format, NULL, NULL, NULL,
1774                             size, flags, false, &local_err);
1775         }
1776     }
1777 
1778     if (local_err) {
1779         error_propagate(errp, local_err);
1780         goto out;
1781     }
1782 
1783     options = qdict_new();
1784     qdict_put_str(options, "discard", "unmap");
1785     qdict_put_str(options, "detect-zeroes", "unmap");
1786     if (backup->format) {
1787         qdict_put_str(options, "driver", backup->format);
1788     }
1789 
1790     target_bs = bdrv_open(backup->target, NULL, options, flags, errp);
1791     if (!target_bs) {
1792         goto out;
1793     }
1794 
1795     /* Honor bdrv_try_set_aio_context() context acquisition requirements. */
1796     old_context = bdrv_get_aio_context(target_bs);
1797     aio_context_release(aio_context);
1798     aio_context_acquire(old_context);
1799 
1800     ret = bdrv_try_set_aio_context(target_bs, aio_context, errp);
1801     if (ret < 0) {
1802         bdrv_unref(target_bs);
1803         aio_context_release(old_context);
1804         return;
1805     }
1806 
1807     aio_context_release(old_context);
1808     aio_context_acquire(aio_context);
1809 
1810     if (set_backing_hd) {
1811         if (bdrv_set_backing_hd(target_bs, source, errp) < 0) {
1812             goto unref;
1813         }
1814     }
1815 
1816     state->job = do_backup_common(qapi_DriveBackup_base(backup),
1817                                   bs, target_bs, aio_context,
1818                                   common->block_job_txn, errp);
1819 
1820 unref:
1821     bdrv_unref(target_bs);
1822 out:
1823     aio_context_release(aio_context);
1824 }
1825 
1826 static void drive_backup_commit(BlkActionState *common)
1827 {
1828     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1829     AioContext *aio_context;
1830 
1831     aio_context = bdrv_get_aio_context(state->bs);
1832     aio_context_acquire(aio_context);
1833 
1834     assert(state->job);
1835     job_start(&state->job->job);
1836 
1837     aio_context_release(aio_context);
1838 }
1839 
1840 static void drive_backup_abort(BlkActionState *common)
1841 {
1842     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1843 
1844     if (state->job) {
1845         AioContext *aio_context;
1846 
1847         aio_context = bdrv_get_aio_context(state->bs);
1848         aio_context_acquire(aio_context);
1849 
1850         job_cancel_sync(&state->job->job, true);
1851 
1852         aio_context_release(aio_context);
1853     }
1854 }
1855 
1856 static void drive_backup_clean(BlkActionState *common)
1857 {
1858     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1859     AioContext *aio_context;
1860 
1861     if (!state->bs) {
1862         return;
1863     }
1864 
1865     aio_context = bdrv_get_aio_context(state->bs);
1866     aio_context_acquire(aio_context);
1867 
1868     bdrv_drained_end(state->bs);
1869 
1870     aio_context_release(aio_context);
1871 }
1872 
1873 typedef struct BlockdevBackupState {
1874     BlkActionState common;
1875     BlockDriverState *bs;
1876     BlockJob *job;
1877 } BlockdevBackupState;
1878 
1879 static void blockdev_backup_prepare(BlkActionState *common, Error **errp)
1880 {
1881     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1882     BlockdevBackup *backup;
1883     BlockDriverState *bs;
1884     BlockDriverState *target_bs;
1885     AioContext *aio_context;
1886     AioContext *old_context;
1887     int ret;
1888 
1889     assert(common->action->type == TRANSACTION_ACTION_KIND_BLOCKDEV_BACKUP);
1890     backup = common->action->u.blockdev_backup.data;
1891 
1892     bs = bdrv_lookup_bs(backup->device, backup->device, errp);
1893     if (!bs) {
1894         return;
1895     }
1896 
1897     target_bs = bdrv_lookup_bs(backup->target, backup->target, errp);
1898     if (!target_bs) {
1899         return;
1900     }
1901 
1902     /* Honor bdrv_try_set_aio_context() context acquisition requirements. */
1903     aio_context = bdrv_get_aio_context(bs);
1904     old_context = bdrv_get_aio_context(target_bs);
1905     aio_context_acquire(old_context);
1906 
1907     ret = bdrv_try_set_aio_context(target_bs, aio_context, errp);
1908     if (ret < 0) {
1909         aio_context_release(old_context);
1910         return;
1911     }
1912 
1913     aio_context_release(old_context);
1914     aio_context_acquire(aio_context);
1915     state->bs = bs;
1916 
1917     /* Paired with .clean() */
1918     bdrv_drained_begin(state->bs);
1919 
1920     state->job = do_backup_common(qapi_BlockdevBackup_base(backup),
1921                                   bs, target_bs, aio_context,
1922                                   common->block_job_txn, errp);
1923 
1924     aio_context_release(aio_context);
1925 }
1926 
1927 static void blockdev_backup_commit(BlkActionState *common)
1928 {
1929     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1930     AioContext *aio_context;
1931 
1932     aio_context = bdrv_get_aio_context(state->bs);
1933     aio_context_acquire(aio_context);
1934 
1935     assert(state->job);
1936     job_start(&state->job->job);
1937 
1938     aio_context_release(aio_context);
1939 }
1940 
1941 static void blockdev_backup_abort(BlkActionState *common)
1942 {
1943     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1944 
1945     if (state->job) {
1946         AioContext *aio_context;
1947 
1948         aio_context = bdrv_get_aio_context(state->bs);
1949         aio_context_acquire(aio_context);
1950 
1951         job_cancel_sync(&state->job->job, true);
1952 
1953         aio_context_release(aio_context);
1954     }
1955 }
1956 
1957 static void blockdev_backup_clean(BlkActionState *common)
1958 {
1959     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1960     AioContext *aio_context;
1961 
1962     if (!state->bs) {
1963         return;
1964     }
1965 
1966     aio_context = bdrv_get_aio_context(state->bs);
1967     aio_context_acquire(aio_context);
1968 
1969     bdrv_drained_end(state->bs);
1970 
1971     aio_context_release(aio_context);
1972 }
1973 
1974 typedef struct BlockDirtyBitmapState {
1975     BlkActionState common;
1976     BdrvDirtyBitmap *bitmap;
1977     BlockDriverState *bs;
1978     HBitmap *backup;
1979     bool prepared;
1980     bool was_enabled;
1981 } BlockDirtyBitmapState;
1982 
1983 static void block_dirty_bitmap_add_prepare(BlkActionState *common,
1984                                            Error **errp)
1985 {
1986     Error *local_err = NULL;
1987     BlockDirtyBitmapAdd *action;
1988     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
1989                                              common, common);
1990 
1991     if (action_check_completion_mode(common, errp) < 0) {
1992         return;
1993     }
1994 
1995     action = common->action->u.block_dirty_bitmap_add.data;
1996     /* AIO context taken and released within qmp_block_dirty_bitmap_add */
1997     qmp_block_dirty_bitmap_add(action->node, action->name,
1998                                action->has_granularity, action->granularity,
1999                                action->has_persistent, action->persistent,
2000                                action->has_disabled, action->disabled,
2001                                &local_err);
2002 
2003     if (!local_err) {
2004         state->prepared = true;
2005     } else {
2006         error_propagate(errp, local_err);
2007     }
2008 }
2009 
2010 static void block_dirty_bitmap_add_abort(BlkActionState *common)
2011 {
2012     BlockDirtyBitmapAdd *action;
2013     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2014                                              common, common);
2015 
2016     action = common->action->u.block_dirty_bitmap_add.data;
2017     /* Should not be able to fail: IF the bitmap was added via .prepare(),
2018      * then the node reference and bitmap name must have been valid.
2019      */
2020     if (state->prepared) {
2021         qmp_block_dirty_bitmap_remove(action->node, action->name, &error_abort);
2022     }
2023 }
2024 
2025 static void block_dirty_bitmap_clear_prepare(BlkActionState *common,
2026                                              Error **errp)
2027 {
2028     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2029                                              common, common);
2030     BlockDirtyBitmap *action;
2031 
2032     if (action_check_completion_mode(common, errp) < 0) {
2033         return;
2034     }
2035 
2036     action = common->action->u.block_dirty_bitmap_clear.data;
2037     state->bitmap = block_dirty_bitmap_lookup(action->node,
2038                                               action->name,
2039                                               &state->bs,
2040                                               errp);
2041     if (!state->bitmap) {
2042         return;
2043     }
2044 
2045     if (bdrv_dirty_bitmap_check(state->bitmap, BDRV_BITMAP_DEFAULT, errp)) {
2046         return;
2047     }
2048 
2049     bdrv_clear_dirty_bitmap(state->bitmap, &state->backup);
2050 }
2051 
2052 static void block_dirty_bitmap_restore(BlkActionState *common)
2053 {
2054     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2055                                              common, common);
2056 
2057     if (state->backup) {
2058         bdrv_restore_dirty_bitmap(state->bitmap, state->backup);
2059     }
2060 }
2061 
2062 static void block_dirty_bitmap_free_backup(BlkActionState *common)
2063 {
2064     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2065                                              common, common);
2066 
2067     hbitmap_free(state->backup);
2068 }
2069 
2070 static void block_dirty_bitmap_enable_prepare(BlkActionState *common,
2071                                               Error **errp)
2072 {
2073     BlockDirtyBitmap *action;
2074     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2075                                              common, common);
2076 
2077     if (action_check_completion_mode(common, errp) < 0) {
2078         return;
2079     }
2080 
2081     action = common->action->u.block_dirty_bitmap_enable.data;
2082     state->bitmap = block_dirty_bitmap_lookup(action->node,
2083                                               action->name,
2084                                               NULL,
2085                                               errp);
2086     if (!state->bitmap) {
2087         return;
2088     }
2089 
2090     if (bdrv_dirty_bitmap_check(state->bitmap, BDRV_BITMAP_ALLOW_RO, errp)) {
2091         return;
2092     }
2093 
2094     state->was_enabled = bdrv_dirty_bitmap_enabled(state->bitmap);
2095     bdrv_enable_dirty_bitmap(state->bitmap);
2096 }
2097 
2098 static void block_dirty_bitmap_enable_abort(BlkActionState *common)
2099 {
2100     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2101                                              common, common);
2102 
2103     if (!state->was_enabled) {
2104         bdrv_disable_dirty_bitmap(state->bitmap);
2105     }
2106 }
2107 
2108 static void block_dirty_bitmap_disable_prepare(BlkActionState *common,
2109                                                Error **errp)
2110 {
2111     BlockDirtyBitmap *action;
2112     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2113                                              common, common);
2114 
2115     if (action_check_completion_mode(common, errp) < 0) {
2116         return;
2117     }
2118 
2119     action = common->action->u.block_dirty_bitmap_disable.data;
2120     state->bitmap = block_dirty_bitmap_lookup(action->node,
2121                                               action->name,
2122                                               NULL,
2123                                               errp);
2124     if (!state->bitmap) {
2125         return;
2126     }
2127 
2128     if (bdrv_dirty_bitmap_check(state->bitmap, BDRV_BITMAP_ALLOW_RO, errp)) {
2129         return;
2130     }
2131 
2132     state->was_enabled = bdrv_dirty_bitmap_enabled(state->bitmap);
2133     bdrv_disable_dirty_bitmap(state->bitmap);
2134 }
2135 
2136 static void block_dirty_bitmap_disable_abort(BlkActionState *common)
2137 {
2138     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2139                                              common, common);
2140 
2141     if (state->was_enabled) {
2142         bdrv_enable_dirty_bitmap(state->bitmap);
2143     }
2144 }
2145 
2146 static void block_dirty_bitmap_merge_prepare(BlkActionState *common,
2147                                              Error **errp)
2148 {
2149     BlockDirtyBitmapMerge *action;
2150     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2151                                              common, common);
2152 
2153     if (action_check_completion_mode(common, errp) < 0) {
2154         return;
2155     }
2156 
2157     action = common->action->u.block_dirty_bitmap_merge.data;
2158 
2159     state->bitmap = block_dirty_bitmap_merge(action->node, action->target,
2160                                              action->bitmaps, &state->backup,
2161                                              errp);
2162 }
2163 
2164 static void block_dirty_bitmap_remove_prepare(BlkActionState *common,
2165                                               Error **errp)
2166 {
2167     BlockDirtyBitmap *action;
2168     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2169                                              common, common);
2170 
2171     if (action_check_completion_mode(common, errp) < 0) {
2172         return;
2173     }
2174 
2175     action = common->action->u.block_dirty_bitmap_remove.data;
2176 
2177     state->bitmap = block_dirty_bitmap_remove(action->node, action->name,
2178                                               false, &state->bs, errp);
2179     if (state->bitmap) {
2180         bdrv_dirty_bitmap_skip_store(state->bitmap, true);
2181         bdrv_dirty_bitmap_set_busy(state->bitmap, true);
2182     }
2183 }
2184 
2185 static void block_dirty_bitmap_remove_abort(BlkActionState *common)
2186 {
2187     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2188                                              common, common);
2189 
2190     if (state->bitmap) {
2191         bdrv_dirty_bitmap_skip_store(state->bitmap, false);
2192         bdrv_dirty_bitmap_set_busy(state->bitmap, false);
2193     }
2194 }
2195 
2196 static void block_dirty_bitmap_remove_commit(BlkActionState *common)
2197 {
2198     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2199                                              common, common);
2200 
2201     bdrv_dirty_bitmap_set_busy(state->bitmap, false);
2202     bdrv_release_dirty_bitmap(state->bitmap);
2203 }
2204 
2205 static void abort_prepare(BlkActionState *common, Error **errp)
2206 {
2207     error_setg(errp, "Transaction aborted using Abort action");
2208 }
2209 
2210 static void abort_commit(BlkActionState *common)
2211 {
2212     g_assert_not_reached(); /* this action never succeeds */
2213 }
2214 
2215 static const BlkActionOps actions[] = {
2216     [TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT] = {
2217         .instance_size = sizeof(ExternalSnapshotState),
2218         .prepare  = external_snapshot_prepare,
2219         .commit   = external_snapshot_commit,
2220         .abort = external_snapshot_abort,
2221         .clean = external_snapshot_clean,
2222     },
2223     [TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC] = {
2224         .instance_size = sizeof(ExternalSnapshotState),
2225         .prepare  = external_snapshot_prepare,
2226         .commit   = external_snapshot_commit,
2227         .abort = external_snapshot_abort,
2228         .clean = external_snapshot_clean,
2229     },
2230     [TRANSACTION_ACTION_KIND_DRIVE_BACKUP] = {
2231         .instance_size = sizeof(DriveBackupState),
2232         .prepare = drive_backup_prepare,
2233         .commit = drive_backup_commit,
2234         .abort = drive_backup_abort,
2235         .clean = drive_backup_clean,
2236     },
2237     [TRANSACTION_ACTION_KIND_BLOCKDEV_BACKUP] = {
2238         .instance_size = sizeof(BlockdevBackupState),
2239         .prepare = blockdev_backup_prepare,
2240         .commit = blockdev_backup_commit,
2241         .abort = blockdev_backup_abort,
2242         .clean = blockdev_backup_clean,
2243     },
2244     [TRANSACTION_ACTION_KIND_ABORT] = {
2245         .instance_size = sizeof(BlkActionState),
2246         .prepare = abort_prepare,
2247         .commit = abort_commit,
2248     },
2249     [TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC] = {
2250         .instance_size = sizeof(InternalSnapshotState),
2251         .prepare  = internal_snapshot_prepare,
2252         .abort = internal_snapshot_abort,
2253         .clean = internal_snapshot_clean,
2254     },
2255     [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_ADD] = {
2256         .instance_size = sizeof(BlockDirtyBitmapState),
2257         .prepare = block_dirty_bitmap_add_prepare,
2258         .abort = block_dirty_bitmap_add_abort,
2259     },
2260     [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_CLEAR] = {
2261         .instance_size = sizeof(BlockDirtyBitmapState),
2262         .prepare = block_dirty_bitmap_clear_prepare,
2263         .commit = block_dirty_bitmap_free_backup,
2264         .abort = block_dirty_bitmap_restore,
2265     },
2266     [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_ENABLE] = {
2267         .instance_size = sizeof(BlockDirtyBitmapState),
2268         .prepare = block_dirty_bitmap_enable_prepare,
2269         .abort = block_dirty_bitmap_enable_abort,
2270     },
2271     [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_DISABLE] = {
2272         .instance_size = sizeof(BlockDirtyBitmapState),
2273         .prepare = block_dirty_bitmap_disable_prepare,
2274         .abort = block_dirty_bitmap_disable_abort,
2275     },
2276     [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_MERGE] = {
2277         .instance_size = sizeof(BlockDirtyBitmapState),
2278         .prepare = block_dirty_bitmap_merge_prepare,
2279         .commit = block_dirty_bitmap_free_backup,
2280         .abort = block_dirty_bitmap_restore,
2281     },
2282     [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_REMOVE] = {
2283         .instance_size = sizeof(BlockDirtyBitmapState),
2284         .prepare = block_dirty_bitmap_remove_prepare,
2285         .commit = block_dirty_bitmap_remove_commit,
2286         .abort = block_dirty_bitmap_remove_abort,
2287     },
2288     /* Where are transactions for MIRROR, COMMIT and STREAM?
2289      * Although these blockjobs use transaction callbacks like the backup job,
2290      * these jobs do not necessarily adhere to transaction semantics.
2291      * These jobs may not fully undo all of their actions on abort, nor do they
2292      * necessarily work in transactions with more than one job in them.
2293      */
2294 };
2295 
2296 /**
2297  * Allocate a TransactionProperties structure if necessary, and fill
2298  * that structure with desired defaults if they are unset.
2299  */
2300 static TransactionProperties *get_transaction_properties(
2301     TransactionProperties *props)
2302 {
2303     if (!props) {
2304         props = g_new0(TransactionProperties, 1);
2305     }
2306 
2307     if (!props->has_completion_mode) {
2308         props->has_completion_mode = true;
2309         props->completion_mode = ACTION_COMPLETION_MODE_INDIVIDUAL;
2310     }
2311 
2312     return props;
2313 }
2314 
2315 /*
2316  * 'Atomic' group operations.  The operations are performed as a set, and if
2317  * any fail then we roll back all operations in the group.
2318  */
2319 void qmp_transaction(TransactionActionList *dev_list,
2320                      bool has_props,
2321                      struct TransactionProperties *props,
2322                      Error **errp)
2323 {
2324     TransactionActionList *dev_entry = dev_list;
2325     JobTxn *block_job_txn = NULL;
2326     BlkActionState *state, *next;
2327     Error *local_err = NULL;
2328 
2329     QTAILQ_HEAD(, BlkActionState) snap_bdrv_states;
2330     QTAILQ_INIT(&snap_bdrv_states);
2331 
2332     /* Does this transaction get canceled as a group on failure?
2333      * If not, we don't really need to make a JobTxn.
2334      */
2335     props = get_transaction_properties(props);
2336     if (props->completion_mode != ACTION_COMPLETION_MODE_INDIVIDUAL) {
2337         block_job_txn = job_txn_new();
2338     }
2339 
2340     /* drain all i/o before any operations */
2341     bdrv_drain_all();
2342 
2343     /* We don't do anything in this loop that commits us to the operations */
2344     while (NULL != dev_entry) {
2345         TransactionAction *dev_info = NULL;
2346         const BlkActionOps *ops;
2347 
2348         dev_info = dev_entry->value;
2349         dev_entry = dev_entry->next;
2350 
2351         assert(dev_info->type < ARRAY_SIZE(actions));
2352 
2353         ops = &actions[dev_info->type];
2354         assert(ops->instance_size > 0);
2355 
2356         state = g_malloc0(ops->instance_size);
2357         state->ops = ops;
2358         state->action = dev_info;
2359         state->block_job_txn = block_job_txn;
2360         state->txn_props = props;
2361         QTAILQ_INSERT_TAIL(&snap_bdrv_states, state, entry);
2362 
2363         state->ops->prepare(state, &local_err);
2364         if (local_err) {
2365             error_propagate(errp, local_err);
2366             goto delete_and_fail;
2367         }
2368     }
2369 
2370     QTAILQ_FOREACH(state, &snap_bdrv_states, entry) {
2371         if (state->ops->commit) {
2372             state->ops->commit(state);
2373         }
2374     }
2375 
2376     /* success */
2377     goto exit;
2378 
2379 delete_and_fail:
2380     /* failure, and it is all-or-none; roll back all operations */
2381     QTAILQ_FOREACH_REVERSE(state, &snap_bdrv_states, entry) {
2382         if (state->ops->abort) {
2383             state->ops->abort(state);
2384         }
2385     }
2386 exit:
2387     QTAILQ_FOREACH_SAFE(state, &snap_bdrv_states, entry, next) {
2388         if (state->ops->clean) {
2389             state->ops->clean(state);
2390         }
2391         g_free(state);
2392     }
2393     if (!has_props) {
2394         qapi_free_TransactionProperties(props);
2395     }
2396     job_txn_unref(block_job_txn);
2397 }
2398 
2399 BlockDirtyBitmapSha256 *qmp_x_debug_block_dirty_bitmap_sha256(const char *node,
2400                                                               const char *name,
2401                                                               Error **errp)
2402 {
2403     BdrvDirtyBitmap *bitmap;
2404     BlockDriverState *bs;
2405     BlockDirtyBitmapSha256 *ret = NULL;
2406     char *sha256;
2407 
2408     bitmap = block_dirty_bitmap_lookup(node, name, &bs, errp);
2409     if (!bitmap || !bs) {
2410         return NULL;
2411     }
2412 
2413     sha256 = bdrv_dirty_bitmap_sha256(bitmap, errp);
2414     if (sha256 == NULL) {
2415         return NULL;
2416     }
2417 
2418     ret = g_new(BlockDirtyBitmapSha256, 1);
2419     ret->sha256 = sha256;
2420 
2421     return ret;
2422 }
2423 
2424 void coroutine_fn qmp_block_resize(bool has_device, const char *device,
2425                                    bool has_node_name, const char *node_name,
2426                                    int64_t size, Error **errp)
2427 {
2428     Error *local_err = NULL;
2429     BlockBackend *blk;
2430     BlockDriverState *bs;
2431     AioContext *old_ctx;
2432 
2433     bs = bdrv_lookup_bs(has_device ? device : NULL,
2434                         has_node_name ? node_name : NULL,
2435                         &local_err);
2436     if (local_err) {
2437         error_propagate(errp, local_err);
2438         return;
2439     }
2440 
2441     if (size < 0) {
2442         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "size", "a >0 size");
2443         return;
2444     }
2445 
2446     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_RESIZE, NULL)) {
2447         error_setg(errp, QERR_DEVICE_IN_USE, device);
2448         return;
2449     }
2450 
2451     blk = blk_new_with_bs(bs, BLK_PERM_RESIZE, BLK_PERM_ALL, errp);
2452     if (!blk) {
2453         return;
2454     }
2455 
2456     bdrv_co_lock(bs);
2457     bdrv_drained_begin(bs);
2458     bdrv_co_unlock(bs);
2459 
2460     old_ctx = bdrv_co_enter(bs);
2461     blk_truncate(blk, size, false, PREALLOC_MODE_OFF, 0, errp);
2462     bdrv_co_leave(bs, old_ctx);
2463 
2464     bdrv_co_lock(bs);
2465     bdrv_drained_end(bs);
2466     blk_unref(blk);
2467     bdrv_co_unlock(bs);
2468 }
2469 
2470 void qmp_block_stream(bool has_job_id, const char *job_id, const char *device,
2471                       bool has_base, const char *base,
2472                       bool has_base_node, const char *base_node,
2473                       bool has_backing_file, const char *backing_file,
2474                       bool has_bottom, const char *bottom,
2475                       bool has_speed, int64_t speed,
2476                       bool has_on_error, BlockdevOnError on_error,
2477                       bool has_filter_node_name, const char *filter_node_name,
2478                       bool has_auto_finalize, bool auto_finalize,
2479                       bool has_auto_dismiss, bool auto_dismiss,
2480                       Error **errp)
2481 {
2482     BlockDriverState *bs, *iter, *iter_end;
2483     BlockDriverState *base_bs = NULL;
2484     BlockDriverState *bottom_bs = NULL;
2485     AioContext *aio_context;
2486     Error *local_err = NULL;
2487     int job_flags = JOB_DEFAULT;
2488 
2489     if (has_base && has_base_node) {
2490         error_setg(errp, "'base' and 'base-node' cannot be specified "
2491                    "at the same time");
2492         return;
2493     }
2494 
2495     if (has_base && has_bottom) {
2496         error_setg(errp, "'base' and 'bottom' cannot be specified "
2497                    "at the same time");
2498         return;
2499     }
2500 
2501     if (has_bottom && has_base_node) {
2502         error_setg(errp, "'bottom' and 'base-node' cannot be specified "
2503                    "at the same time");
2504         return;
2505     }
2506 
2507     if (!has_on_error) {
2508         on_error = BLOCKDEV_ON_ERROR_REPORT;
2509     }
2510 
2511     bs = bdrv_lookup_bs(device, device, errp);
2512     if (!bs) {
2513         return;
2514     }
2515 
2516     aio_context = bdrv_get_aio_context(bs);
2517     aio_context_acquire(aio_context);
2518 
2519     if (has_base) {
2520         base_bs = bdrv_find_backing_image(bs, base);
2521         if (base_bs == NULL) {
2522             error_setg(errp, "Can't find '%s' in the backing chain", base);
2523             goto out;
2524         }
2525         assert(bdrv_get_aio_context(base_bs) == aio_context);
2526     }
2527 
2528     if (has_base_node) {
2529         base_bs = bdrv_lookup_bs(NULL, base_node, errp);
2530         if (!base_bs) {
2531             goto out;
2532         }
2533         if (bs == base_bs || !bdrv_chain_contains(bs, base_bs)) {
2534             error_setg(errp, "Node '%s' is not a backing image of '%s'",
2535                        base_node, device);
2536             goto out;
2537         }
2538         assert(bdrv_get_aio_context(base_bs) == aio_context);
2539         bdrv_refresh_filename(base_bs);
2540     }
2541 
2542     if (has_bottom) {
2543         bottom_bs = bdrv_lookup_bs(NULL, bottom, errp);
2544         if (!bottom_bs) {
2545             goto out;
2546         }
2547         if (!bottom_bs->drv) {
2548             error_setg(errp, "Node '%s' is not open", bottom);
2549             goto out;
2550         }
2551         if (bottom_bs->drv->is_filter) {
2552             error_setg(errp, "Node '%s' is a filter, use a non-filter node "
2553                        "as 'bottom'", bottom);
2554             goto out;
2555         }
2556         if (!bdrv_chain_contains(bs, bottom_bs)) {
2557             error_setg(errp, "Node '%s' is not in a chain starting from '%s'",
2558                        bottom, device);
2559             goto out;
2560         }
2561         assert(bdrv_get_aio_context(bottom_bs) == aio_context);
2562     }
2563 
2564     /*
2565      * Check for op blockers in the whole chain between bs and base (or bottom)
2566      */
2567     iter_end = has_bottom ? bdrv_filter_or_cow_bs(bottom_bs) : base_bs;
2568     for (iter = bs; iter && iter != iter_end;
2569          iter = bdrv_filter_or_cow_bs(iter))
2570     {
2571         if (bdrv_op_is_blocked(iter, BLOCK_OP_TYPE_STREAM, errp)) {
2572             goto out;
2573         }
2574     }
2575 
2576     /* if we are streaming the entire chain, the result will have no backing
2577      * file, and specifying one is therefore an error */
2578     if (base_bs == NULL && has_backing_file) {
2579         error_setg(errp, "backing file specified, but streaming the "
2580                          "entire chain");
2581         goto out;
2582     }
2583 
2584     if (has_auto_finalize && !auto_finalize) {
2585         job_flags |= JOB_MANUAL_FINALIZE;
2586     }
2587     if (has_auto_dismiss && !auto_dismiss) {
2588         job_flags |= JOB_MANUAL_DISMISS;
2589     }
2590 
2591     stream_start(has_job_id ? job_id : NULL, bs, base_bs, backing_file,
2592                  bottom_bs, job_flags, has_speed ? speed : 0, on_error,
2593                  filter_node_name, &local_err);
2594     if (local_err) {
2595         error_propagate(errp, local_err);
2596         goto out;
2597     }
2598 
2599     trace_qmp_block_stream(bs);
2600 
2601 out:
2602     aio_context_release(aio_context);
2603 }
2604 
2605 void qmp_block_commit(bool has_job_id, const char *job_id, const char *device,
2606                       bool has_base_node, const char *base_node,
2607                       bool has_base, const char *base,
2608                       bool has_top_node, const char *top_node,
2609                       bool has_top, const char *top,
2610                       bool has_backing_file, const char *backing_file,
2611                       bool has_speed, int64_t speed,
2612                       bool has_on_error, BlockdevOnError on_error,
2613                       bool has_filter_node_name, const char *filter_node_name,
2614                       bool has_auto_finalize, bool auto_finalize,
2615                       bool has_auto_dismiss, bool auto_dismiss,
2616                       Error **errp)
2617 {
2618     BlockDriverState *bs;
2619     BlockDriverState *iter;
2620     BlockDriverState *base_bs, *top_bs;
2621     AioContext *aio_context;
2622     Error *local_err = NULL;
2623     int job_flags = JOB_DEFAULT;
2624     uint64_t top_perm, top_shared;
2625 
2626     if (!has_speed) {
2627         speed = 0;
2628     }
2629     if (!has_on_error) {
2630         on_error = BLOCKDEV_ON_ERROR_REPORT;
2631     }
2632     if (!has_filter_node_name) {
2633         filter_node_name = NULL;
2634     }
2635     if (has_auto_finalize && !auto_finalize) {
2636         job_flags |= JOB_MANUAL_FINALIZE;
2637     }
2638     if (has_auto_dismiss && !auto_dismiss) {
2639         job_flags |= JOB_MANUAL_DISMISS;
2640     }
2641 
2642     /* Important Note:
2643      *  libvirt relies on the DeviceNotFound error class in order to probe for
2644      *  live commit feature versions; for this to work, we must make sure to
2645      *  perform the device lookup before any generic errors that may occur in a
2646      *  scenario in which all optional arguments are omitted. */
2647     bs = qmp_get_root_bs(device, &local_err);
2648     if (!bs) {
2649         bs = bdrv_lookup_bs(device, device, NULL);
2650         if (!bs) {
2651             error_free(local_err);
2652             error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
2653                       "Device '%s' not found", device);
2654         } else {
2655             error_propagate(errp, local_err);
2656         }
2657         return;
2658     }
2659 
2660     aio_context = bdrv_get_aio_context(bs);
2661     aio_context_acquire(aio_context);
2662 
2663     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_COMMIT_SOURCE, errp)) {
2664         goto out;
2665     }
2666 
2667     /* default top_bs is the active layer */
2668     top_bs = bs;
2669 
2670     if (has_top_node && has_top) {
2671         error_setg(errp, "'top-node' and 'top' are mutually exclusive");
2672         goto out;
2673     } else if (has_top_node) {
2674         top_bs = bdrv_lookup_bs(NULL, top_node, errp);
2675         if (top_bs == NULL) {
2676             goto out;
2677         }
2678         if (!bdrv_chain_contains(bs, top_bs)) {
2679             error_setg(errp, "'%s' is not in this backing file chain",
2680                        top_node);
2681             goto out;
2682         }
2683     } else if (has_top && top) {
2684         /* This strcmp() is just a shortcut, there is no need to
2685          * refresh @bs's filename.  If it mismatches,
2686          * bdrv_find_backing_image() will do the refresh and may still
2687          * return @bs. */
2688         if (strcmp(bs->filename, top) != 0) {
2689             top_bs = bdrv_find_backing_image(bs, top);
2690         }
2691     }
2692 
2693     if (top_bs == NULL) {
2694         error_setg(errp, "Top image file %s not found", top ? top : "NULL");
2695         goto out;
2696     }
2697 
2698     assert(bdrv_get_aio_context(top_bs) == aio_context);
2699 
2700     if (has_base_node && has_base) {
2701         error_setg(errp, "'base-node' and 'base' are mutually exclusive");
2702         goto out;
2703     } else if (has_base_node) {
2704         base_bs = bdrv_lookup_bs(NULL, base_node, errp);
2705         if (base_bs == NULL) {
2706             goto out;
2707         }
2708         if (!bdrv_chain_contains(top_bs, base_bs)) {
2709             error_setg(errp, "'%s' is not in this backing file chain",
2710                        base_node);
2711             goto out;
2712         }
2713     } else if (has_base && base) {
2714         base_bs = bdrv_find_backing_image(top_bs, base);
2715         if (base_bs == NULL) {
2716             error_setg(errp, "Can't find '%s' in the backing chain", base);
2717             goto out;
2718         }
2719     } else {
2720         base_bs = bdrv_find_base(top_bs);
2721         if (base_bs == NULL) {
2722             error_setg(errp, "There is no backimg image");
2723             goto out;
2724         }
2725     }
2726 
2727     assert(bdrv_get_aio_context(base_bs) == aio_context);
2728 
2729     for (iter = top_bs; iter != bdrv_filter_or_cow_bs(base_bs);
2730          iter = bdrv_filter_or_cow_bs(iter))
2731     {
2732         if (bdrv_op_is_blocked(iter, BLOCK_OP_TYPE_COMMIT_TARGET, errp)) {
2733             goto out;
2734         }
2735     }
2736 
2737     /* Do not allow attempts to commit an image into itself */
2738     if (top_bs == base_bs) {
2739         error_setg(errp, "cannot commit an image into itself");
2740         goto out;
2741     }
2742 
2743     /*
2744      * Active commit is required if and only if someone has taken a
2745      * WRITE permission on the top node.  Historically, we have always
2746      * used active commit for top nodes, so continue that practice
2747      * lest we possibly break clients that rely on this behavior, e.g.
2748      * to later attach this node to a writing parent.
2749      * (Active commit is never really wrong.)
2750      */
2751     bdrv_get_cumulative_perm(top_bs, &top_perm, &top_shared);
2752     if (top_perm & BLK_PERM_WRITE ||
2753         bdrv_skip_filters(top_bs) == bdrv_skip_filters(bs))
2754     {
2755         if (has_backing_file) {
2756             if (bdrv_skip_filters(top_bs) == bdrv_skip_filters(bs)) {
2757                 error_setg(errp, "'backing-file' specified,"
2758                                  " but 'top' is the active layer");
2759             } else {
2760                 error_setg(errp, "'backing-file' specified, but 'top' has a "
2761                                  "writer on it");
2762             }
2763             goto out;
2764         }
2765         if (!has_job_id) {
2766             /*
2767              * Emulate here what block_job_create() does, because it
2768              * is possible that @bs != @top_bs (the block job should
2769              * be named after @bs, even if @top_bs is the actual
2770              * source)
2771              */
2772             job_id = bdrv_get_device_name(bs);
2773         }
2774         commit_active_start(job_id, top_bs, base_bs, job_flags, speed, on_error,
2775                             filter_node_name, NULL, NULL, false, &local_err);
2776     } else {
2777         BlockDriverState *overlay_bs = bdrv_find_overlay(bs, top_bs);
2778         if (bdrv_op_is_blocked(overlay_bs, BLOCK_OP_TYPE_COMMIT_TARGET, errp)) {
2779             goto out;
2780         }
2781         commit_start(has_job_id ? job_id : NULL, bs, base_bs, top_bs, job_flags,
2782                      speed, on_error, has_backing_file ? backing_file : NULL,
2783                      filter_node_name, &local_err);
2784     }
2785     if (local_err != NULL) {
2786         error_propagate(errp, local_err);
2787         goto out;
2788     }
2789 
2790 out:
2791     aio_context_release(aio_context);
2792 }
2793 
2794 /* Common QMP interface for drive-backup and blockdev-backup */
2795 static BlockJob *do_backup_common(BackupCommon *backup,
2796                                   BlockDriverState *bs,
2797                                   BlockDriverState *target_bs,
2798                                   AioContext *aio_context,
2799                                   JobTxn *txn, Error **errp)
2800 {
2801     BlockJob *job = NULL;
2802     BdrvDirtyBitmap *bmap = NULL;
2803     BackupPerf perf = { .max_workers = 64 };
2804     int job_flags = JOB_DEFAULT;
2805 
2806     if (!backup->has_speed) {
2807         backup->speed = 0;
2808     }
2809     if (!backup->has_on_source_error) {
2810         backup->on_source_error = BLOCKDEV_ON_ERROR_REPORT;
2811     }
2812     if (!backup->has_on_target_error) {
2813         backup->on_target_error = BLOCKDEV_ON_ERROR_REPORT;
2814     }
2815     if (!backup->has_job_id) {
2816         backup->job_id = NULL;
2817     }
2818     if (!backup->has_auto_finalize) {
2819         backup->auto_finalize = true;
2820     }
2821     if (!backup->has_auto_dismiss) {
2822         backup->auto_dismiss = true;
2823     }
2824     if (!backup->has_compress) {
2825         backup->compress = false;
2826     }
2827 
2828     if (backup->x_perf) {
2829         if (backup->x_perf->has_use_copy_range) {
2830             perf.use_copy_range = backup->x_perf->use_copy_range;
2831         }
2832         if (backup->x_perf->has_max_workers) {
2833             perf.max_workers = backup->x_perf->max_workers;
2834         }
2835         if (backup->x_perf->has_max_chunk) {
2836             perf.max_chunk = backup->x_perf->max_chunk;
2837         }
2838     }
2839 
2840     if ((backup->sync == MIRROR_SYNC_MODE_BITMAP) ||
2841         (backup->sync == MIRROR_SYNC_MODE_INCREMENTAL)) {
2842         /* done before desugaring 'incremental' to print the right message */
2843         if (!backup->has_bitmap) {
2844             error_setg(errp, "must provide a valid bitmap name for "
2845                        "'%s' sync mode", MirrorSyncMode_str(backup->sync));
2846             return NULL;
2847         }
2848     }
2849 
2850     if (backup->sync == MIRROR_SYNC_MODE_INCREMENTAL) {
2851         if (backup->has_bitmap_mode &&
2852             backup->bitmap_mode != BITMAP_SYNC_MODE_ON_SUCCESS) {
2853             error_setg(errp, "Bitmap sync mode must be '%s' "
2854                        "when using sync mode '%s'",
2855                        BitmapSyncMode_str(BITMAP_SYNC_MODE_ON_SUCCESS),
2856                        MirrorSyncMode_str(backup->sync));
2857             return NULL;
2858         }
2859         backup->has_bitmap_mode = true;
2860         backup->sync = MIRROR_SYNC_MODE_BITMAP;
2861         backup->bitmap_mode = BITMAP_SYNC_MODE_ON_SUCCESS;
2862     }
2863 
2864     if (backup->has_bitmap) {
2865         bmap = bdrv_find_dirty_bitmap(bs, backup->bitmap);
2866         if (!bmap) {
2867             error_setg(errp, "Bitmap '%s' could not be found", backup->bitmap);
2868             return NULL;
2869         }
2870         if (!backup->has_bitmap_mode) {
2871             error_setg(errp, "Bitmap sync mode must be given "
2872                        "when providing a bitmap");
2873             return NULL;
2874         }
2875         if (bdrv_dirty_bitmap_check(bmap, BDRV_BITMAP_ALLOW_RO, errp)) {
2876             return NULL;
2877         }
2878 
2879         /* This does not produce a useful bitmap artifact: */
2880         if (backup->sync == MIRROR_SYNC_MODE_NONE) {
2881             error_setg(errp, "sync mode '%s' does not produce meaningful bitmap"
2882                        " outputs", MirrorSyncMode_str(backup->sync));
2883             return NULL;
2884         }
2885 
2886         /* If the bitmap isn't used for input or output, this is useless: */
2887         if (backup->bitmap_mode == BITMAP_SYNC_MODE_NEVER &&
2888             backup->sync != MIRROR_SYNC_MODE_BITMAP) {
2889             error_setg(errp, "Bitmap sync mode '%s' has no meaningful effect"
2890                        " when combined with sync mode '%s'",
2891                        BitmapSyncMode_str(backup->bitmap_mode),
2892                        MirrorSyncMode_str(backup->sync));
2893             return NULL;
2894         }
2895     }
2896 
2897     if (!backup->has_bitmap && backup->has_bitmap_mode) {
2898         error_setg(errp, "Cannot specify bitmap sync mode without a bitmap");
2899         return NULL;
2900     }
2901 
2902     if (!backup->auto_finalize) {
2903         job_flags |= JOB_MANUAL_FINALIZE;
2904     }
2905     if (!backup->auto_dismiss) {
2906         job_flags |= JOB_MANUAL_DISMISS;
2907     }
2908 
2909     job = backup_job_create(backup->job_id, bs, target_bs, backup->speed,
2910                             backup->sync, bmap, backup->bitmap_mode,
2911                             backup->compress,
2912                             backup->filter_node_name,
2913                             &perf,
2914                             backup->on_source_error,
2915                             backup->on_target_error,
2916                             job_flags, NULL, NULL, txn, errp);
2917     return job;
2918 }
2919 
2920 void qmp_drive_backup(DriveBackup *backup, Error **errp)
2921 {
2922     TransactionAction action = {
2923         .type = TRANSACTION_ACTION_KIND_DRIVE_BACKUP,
2924         .u.drive_backup.data = backup,
2925     };
2926     blockdev_do_action(&action, errp);
2927 }
2928 
2929 BlockDeviceInfoList *qmp_query_named_block_nodes(bool has_flat,
2930                                                  bool flat,
2931                                                  Error **errp)
2932 {
2933     bool return_flat = has_flat && flat;
2934 
2935     return bdrv_named_nodes_list(return_flat, errp);
2936 }
2937 
2938 XDbgBlockGraph *qmp_x_debug_query_block_graph(Error **errp)
2939 {
2940     return bdrv_get_xdbg_block_graph(errp);
2941 }
2942 
2943 void qmp_blockdev_backup(BlockdevBackup *backup, Error **errp)
2944 {
2945     TransactionAction action = {
2946         .type = TRANSACTION_ACTION_KIND_BLOCKDEV_BACKUP,
2947         .u.blockdev_backup.data = backup,
2948     };
2949     blockdev_do_action(&action, errp);
2950 }
2951 
2952 /* Parameter check and block job starting for drive mirroring.
2953  * Caller should hold @device and @target's aio context (must be the same).
2954  **/
2955 static void blockdev_mirror_common(const char *job_id, BlockDriverState *bs,
2956                                    BlockDriverState *target,
2957                                    bool has_replaces, const char *replaces,
2958                                    enum MirrorSyncMode sync,
2959                                    BlockMirrorBackingMode backing_mode,
2960                                    bool zero_target,
2961                                    bool has_speed, int64_t speed,
2962                                    bool has_granularity, uint32_t granularity,
2963                                    bool has_buf_size, int64_t buf_size,
2964                                    bool has_on_source_error,
2965                                    BlockdevOnError on_source_error,
2966                                    bool has_on_target_error,
2967                                    BlockdevOnError on_target_error,
2968                                    bool has_unmap, bool unmap,
2969                                    bool has_filter_node_name,
2970                                    const char *filter_node_name,
2971                                    bool has_copy_mode, MirrorCopyMode copy_mode,
2972                                    bool has_auto_finalize, bool auto_finalize,
2973                                    bool has_auto_dismiss, bool auto_dismiss,
2974                                    Error **errp)
2975 {
2976     BlockDriverState *unfiltered_bs;
2977     int job_flags = JOB_DEFAULT;
2978 
2979     if (!has_speed) {
2980         speed = 0;
2981     }
2982     if (!has_on_source_error) {
2983         on_source_error = BLOCKDEV_ON_ERROR_REPORT;
2984     }
2985     if (!has_on_target_error) {
2986         on_target_error = BLOCKDEV_ON_ERROR_REPORT;
2987     }
2988     if (!has_granularity) {
2989         granularity = 0;
2990     }
2991     if (!has_buf_size) {
2992         buf_size = 0;
2993     }
2994     if (!has_unmap) {
2995         unmap = true;
2996     }
2997     if (!has_filter_node_name) {
2998         filter_node_name = NULL;
2999     }
3000     if (!has_copy_mode) {
3001         copy_mode = MIRROR_COPY_MODE_BACKGROUND;
3002     }
3003     if (has_auto_finalize && !auto_finalize) {
3004         job_flags |= JOB_MANUAL_FINALIZE;
3005     }
3006     if (has_auto_dismiss && !auto_dismiss) {
3007         job_flags |= JOB_MANUAL_DISMISS;
3008     }
3009 
3010     if (granularity != 0 && (granularity < 512 || granularity > 1048576 * 64)) {
3011         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "granularity",
3012                    "a value in range [512B, 64MB]");
3013         return;
3014     }
3015     if (granularity & (granularity - 1)) {
3016         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "granularity",
3017                    "a power of 2");
3018         return;
3019     }
3020 
3021     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_MIRROR_SOURCE, errp)) {
3022         return;
3023     }
3024     if (bdrv_op_is_blocked(target, BLOCK_OP_TYPE_MIRROR_TARGET, errp)) {
3025         return;
3026     }
3027 
3028     if (!bdrv_backing_chain_next(bs) && sync == MIRROR_SYNC_MODE_TOP) {
3029         sync = MIRROR_SYNC_MODE_FULL;
3030     }
3031 
3032     if (!has_replaces) {
3033         /* We want to mirror from @bs, but keep implicit filters on top */
3034         unfiltered_bs = bdrv_skip_implicit_filters(bs);
3035         if (unfiltered_bs != bs) {
3036             replaces = unfiltered_bs->node_name;
3037             has_replaces = true;
3038         }
3039     }
3040 
3041     if (has_replaces) {
3042         BlockDriverState *to_replace_bs;
3043         AioContext *replace_aio_context;
3044         int64_t bs_size, replace_size;
3045 
3046         bs_size = bdrv_getlength(bs);
3047         if (bs_size < 0) {
3048             error_setg_errno(errp, -bs_size, "Failed to query device's size");
3049             return;
3050         }
3051 
3052         to_replace_bs = check_to_replace_node(bs, replaces, errp);
3053         if (!to_replace_bs) {
3054             return;
3055         }
3056 
3057         replace_aio_context = bdrv_get_aio_context(to_replace_bs);
3058         aio_context_acquire(replace_aio_context);
3059         replace_size = bdrv_getlength(to_replace_bs);
3060         aio_context_release(replace_aio_context);
3061 
3062         if (replace_size < 0) {
3063             error_setg_errno(errp, -replace_size,
3064                              "Failed to query the replacement node's size");
3065             return;
3066         }
3067         if (bs_size != replace_size) {
3068             error_setg(errp, "cannot replace image with a mirror image of "
3069                              "different size");
3070             return;
3071         }
3072     }
3073 
3074     /* pass the node name to replace to mirror start since it's loose coupling
3075      * and will allow to check whether the node still exist at mirror completion
3076      */
3077     mirror_start(job_id, bs, target,
3078                  has_replaces ? replaces : NULL, job_flags,
3079                  speed, granularity, buf_size, sync, backing_mode, zero_target,
3080                  on_source_error, on_target_error, unmap, filter_node_name,
3081                  copy_mode, errp);
3082 }
3083 
3084 void qmp_drive_mirror(DriveMirror *arg, Error **errp)
3085 {
3086     BlockDriverState *bs;
3087     BlockDriverState *target_backing_bs, *target_bs;
3088     AioContext *aio_context;
3089     AioContext *old_context;
3090     BlockMirrorBackingMode backing_mode;
3091     Error *local_err = NULL;
3092     QDict *options = NULL;
3093     int flags;
3094     int64_t size;
3095     const char *format = arg->format;
3096     bool zero_target;
3097     int ret;
3098 
3099     bs = qmp_get_root_bs(arg->device, errp);
3100     if (!bs) {
3101         return;
3102     }
3103 
3104     /* Early check to avoid creating target */
3105     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_MIRROR_SOURCE, errp)) {
3106         return;
3107     }
3108 
3109     aio_context = bdrv_get_aio_context(bs);
3110     aio_context_acquire(aio_context);
3111 
3112     if (!arg->has_mode) {
3113         arg->mode = NEW_IMAGE_MODE_ABSOLUTE_PATHS;
3114     }
3115 
3116     if (!arg->has_format) {
3117         format = (arg->mode == NEW_IMAGE_MODE_EXISTING
3118                   ? NULL : bs->drv->format_name);
3119     }
3120 
3121     flags = bs->open_flags | BDRV_O_RDWR;
3122     target_backing_bs = bdrv_cow_bs(bdrv_skip_filters(bs));
3123     if (!target_backing_bs && arg->sync == MIRROR_SYNC_MODE_TOP) {
3124         arg->sync = MIRROR_SYNC_MODE_FULL;
3125     }
3126     if (arg->sync == MIRROR_SYNC_MODE_NONE) {
3127         target_backing_bs = bs;
3128     }
3129 
3130     size = bdrv_getlength(bs);
3131     if (size < 0) {
3132         error_setg_errno(errp, -size, "bdrv_getlength failed");
3133         goto out;
3134     }
3135 
3136     if (arg->has_replaces) {
3137         if (!arg->has_node_name) {
3138             error_setg(errp, "a node-name must be provided when replacing a"
3139                              " named node of the graph");
3140             goto out;
3141         }
3142     }
3143 
3144     if (arg->mode == NEW_IMAGE_MODE_ABSOLUTE_PATHS) {
3145         backing_mode = MIRROR_SOURCE_BACKING_CHAIN;
3146     } else {
3147         backing_mode = MIRROR_OPEN_BACKING_CHAIN;
3148     }
3149 
3150     /* Don't open backing image in create() */
3151     flags |= BDRV_O_NO_BACKING;
3152 
3153     if ((arg->sync == MIRROR_SYNC_MODE_FULL || !target_backing_bs)
3154         && arg->mode != NEW_IMAGE_MODE_EXISTING)
3155     {
3156         /* create new image w/o backing file */
3157         assert(format);
3158         bdrv_img_create(arg->target, format,
3159                         NULL, NULL, NULL, size, flags, false, &local_err);
3160     } else {
3161         /* Implicit filters should not appear in the filename */
3162         BlockDriverState *explicit_backing =
3163             bdrv_skip_implicit_filters(target_backing_bs);
3164 
3165         switch (arg->mode) {
3166         case NEW_IMAGE_MODE_EXISTING:
3167             break;
3168         case NEW_IMAGE_MODE_ABSOLUTE_PATHS:
3169             /* create new image with backing file */
3170             bdrv_refresh_filename(explicit_backing);
3171             bdrv_img_create(arg->target, format,
3172                             explicit_backing->filename,
3173                             explicit_backing->drv->format_name,
3174                             NULL, size, flags, false, &local_err);
3175             break;
3176         default:
3177             abort();
3178         }
3179     }
3180 
3181     if (local_err) {
3182         error_propagate(errp, local_err);
3183         goto out;
3184     }
3185 
3186     options = qdict_new();
3187     if (arg->has_node_name) {
3188         qdict_put_str(options, "node-name", arg->node_name);
3189     }
3190     if (format) {
3191         qdict_put_str(options, "driver", format);
3192     }
3193 
3194     /* Mirroring takes care of copy-on-write using the source's backing
3195      * file.
3196      */
3197     target_bs = bdrv_open(arg->target, NULL, options, flags, errp);
3198     if (!target_bs) {
3199         goto out;
3200     }
3201 
3202     zero_target = (arg->sync == MIRROR_SYNC_MODE_FULL &&
3203                    (arg->mode == NEW_IMAGE_MODE_EXISTING ||
3204                     !bdrv_has_zero_init(target_bs)));
3205 
3206 
3207     /* Honor bdrv_try_set_aio_context() context acquisition requirements. */
3208     old_context = bdrv_get_aio_context(target_bs);
3209     aio_context_release(aio_context);
3210     aio_context_acquire(old_context);
3211 
3212     ret = bdrv_try_set_aio_context(target_bs, aio_context, errp);
3213     if (ret < 0) {
3214         bdrv_unref(target_bs);
3215         aio_context_release(old_context);
3216         return;
3217     }
3218 
3219     aio_context_release(old_context);
3220     aio_context_acquire(aio_context);
3221 
3222     blockdev_mirror_common(arg->has_job_id ? arg->job_id : NULL, bs, target_bs,
3223                            arg->has_replaces, arg->replaces, arg->sync,
3224                            backing_mode, zero_target,
3225                            arg->has_speed, arg->speed,
3226                            arg->has_granularity, arg->granularity,
3227                            arg->has_buf_size, arg->buf_size,
3228                            arg->has_on_source_error, arg->on_source_error,
3229                            arg->has_on_target_error, arg->on_target_error,
3230                            arg->has_unmap, arg->unmap,
3231                            false, NULL,
3232                            arg->has_copy_mode, arg->copy_mode,
3233                            arg->has_auto_finalize, arg->auto_finalize,
3234                            arg->has_auto_dismiss, arg->auto_dismiss,
3235                            errp);
3236     bdrv_unref(target_bs);
3237 out:
3238     aio_context_release(aio_context);
3239 }
3240 
3241 void qmp_blockdev_mirror(bool has_job_id, const char *job_id,
3242                          const char *device, const char *target,
3243                          bool has_replaces, const char *replaces,
3244                          MirrorSyncMode sync,
3245                          bool has_speed, int64_t speed,
3246                          bool has_granularity, uint32_t granularity,
3247                          bool has_buf_size, int64_t buf_size,
3248                          bool has_on_source_error,
3249                          BlockdevOnError on_source_error,
3250                          bool has_on_target_error,
3251                          BlockdevOnError on_target_error,
3252                          bool has_filter_node_name,
3253                          const char *filter_node_name,
3254                          bool has_copy_mode, MirrorCopyMode copy_mode,
3255                          bool has_auto_finalize, bool auto_finalize,
3256                          bool has_auto_dismiss, bool auto_dismiss,
3257                          Error **errp)
3258 {
3259     BlockDriverState *bs;
3260     BlockDriverState *target_bs;
3261     AioContext *aio_context;
3262     AioContext *old_context;
3263     BlockMirrorBackingMode backing_mode = MIRROR_LEAVE_BACKING_CHAIN;
3264     bool zero_target;
3265     int ret;
3266 
3267     bs = qmp_get_root_bs(device, errp);
3268     if (!bs) {
3269         return;
3270     }
3271 
3272     target_bs = bdrv_lookup_bs(target, target, errp);
3273     if (!target_bs) {
3274         return;
3275     }
3276 
3277     zero_target = (sync == MIRROR_SYNC_MODE_FULL);
3278 
3279     /* Honor bdrv_try_set_aio_context() context acquisition requirements. */
3280     old_context = bdrv_get_aio_context(target_bs);
3281     aio_context = bdrv_get_aio_context(bs);
3282     aio_context_acquire(old_context);
3283 
3284     ret = bdrv_try_set_aio_context(target_bs, aio_context, errp);
3285 
3286     aio_context_release(old_context);
3287     aio_context_acquire(aio_context);
3288 
3289     if (ret < 0) {
3290         goto out;
3291     }
3292 
3293     blockdev_mirror_common(has_job_id ? job_id : NULL, bs, target_bs,
3294                            has_replaces, replaces, sync, backing_mode,
3295                            zero_target, has_speed, speed,
3296                            has_granularity, granularity,
3297                            has_buf_size, buf_size,
3298                            has_on_source_error, on_source_error,
3299                            has_on_target_error, on_target_error,
3300                            true, true,
3301                            has_filter_node_name, filter_node_name,
3302                            has_copy_mode, copy_mode,
3303                            has_auto_finalize, auto_finalize,
3304                            has_auto_dismiss, auto_dismiss,
3305                            errp);
3306 out:
3307     aio_context_release(aio_context);
3308 }
3309 
3310 /* Get a block job using its ID and acquire its AioContext */
3311 static BlockJob *find_block_job(const char *id, AioContext **aio_context,
3312                                 Error **errp)
3313 {
3314     BlockJob *job;
3315 
3316     assert(id != NULL);
3317 
3318     *aio_context = NULL;
3319 
3320     job = block_job_get(id);
3321 
3322     if (!job) {
3323         error_set(errp, ERROR_CLASS_DEVICE_NOT_ACTIVE,
3324                   "Block job '%s' not found", id);
3325         return NULL;
3326     }
3327 
3328     *aio_context = blk_get_aio_context(job->blk);
3329     aio_context_acquire(*aio_context);
3330 
3331     return job;
3332 }
3333 
3334 void qmp_block_job_set_speed(const char *device, int64_t speed, Error **errp)
3335 {
3336     AioContext *aio_context;
3337     BlockJob *job = find_block_job(device, &aio_context, errp);
3338 
3339     if (!job) {
3340         return;
3341     }
3342 
3343     block_job_set_speed(job, speed, errp);
3344     aio_context_release(aio_context);
3345 }
3346 
3347 void qmp_block_job_cancel(const char *device,
3348                           bool has_force, bool force, Error **errp)
3349 {
3350     AioContext *aio_context;
3351     BlockJob *job = find_block_job(device, &aio_context, errp);
3352 
3353     if (!job) {
3354         return;
3355     }
3356 
3357     if (!has_force) {
3358         force = false;
3359     }
3360 
3361     if (job_user_paused(&job->job) && !force) {
3362         error_setg(errp, "The block job for device '%s' is currently paused",
3363                    device);
3364         goto out;
3365     }
3366 
3367     trace_qmp_block_job_cancel(job);
3368     job_user_cancel(&job->job, force, errp);
3369 out:
3370     aio_context_release(aio_context);
3371 }
3372 
3373 void qmp_block_job_pause(const char *device, Error **errp)
3374 {
3375     AioContext *aio_context;
3376     BlockJob *job = find_block_job(device, &aio_context, errp);
3377 
3378     if (!job) {
3379         return;
3380     }
3381 
3382     trace_qmp_block_job_pause(job);
3383     job_user_pause(&job->job, errp);
3384     aio_context_release(aio_context);
3385 }
3386 
3387 void qmp_block_job_resume(const char *device, Error **errp)
3388 {
3389     AioContext *aio_context;
3390     BlockJob *job = find_block_job(device, &aio_context, errp);
3391 
3392     if (!job) {
3393         return;
3394     }
3395 
3396     trace_qmp_block_job_resume(job);
3397     job_user_resume(&job->job, errp);
3398     aio_context_release(aio_context);
3399 }
3400 
3401 void qmp_block_job_complete(const char *device, Error **errp)
3402 {
3403     AioContext *aio_context;
3404     BlockJob *job = find_block_job(device, &aio_context, errp);
3405 
3406     if (!job) {
3407         return;
3408     }
3409 
3410     trace_qmp_block_job_complete(job);
3411     job_complete(&job->job, errp);
3412     aio_context_release(aio_context);
3413 }
3414 
3415 void qmp_block_job_finalize(const char *id, Error **errp)
3416 {
3417     AioContext *aio_context;
3418     BlockJob *job = find_block_job(id, &aio_context, errp);
3419 
3420     if (!job) {
3421         return;
3422     }
3423 
3424     trace_qmp_block_job_finalize(job);
3425     job_ref(&job->job);
3426     job_finalize(&job->job, errp);
3427 
3428     /*
3429      * Job's context might have changed via job_finalize (and job_txn_apply
3430      * automatically acquires the new one), so make sure we release the correct
3431      * one.
3432      */
3433     aio_context = blk_get_aio_context(job->blk);
3434     job_unref(&job->job);
3435     aio_context_release(aio_context);
3436 }
3437 
3438 void qmp_block_job_dismiss(const char *id, Error **errp)
3439 {
3440     AioContext *aio_context;
3441     BlockJob *bjob = find_block_job(id, &aio_context, errp);
3442     Job *job;
3443 
3444     if (!bjob) {
3445         return;
3446     }
3447 
3448     trace_qmp_block_job_dismiss(bjob);
3449     job = &bjob->job;
3450     job_dismiss(&job, errp);
3451     aio_context_release(aio_context);
3452 }
3453 
3454 void qmp_change_backing_file(const char *device,
3455                              const char *image_node_name,
3456                              const char *backing_file,
3457                              Error **errp)
3458 {
3459     BlockDriverState *bs = NULL;
3460     AioContext *aio_context;
3461     BlockDriverState *image_bs = NULL;
3462     Error *local_err = NULL;
3463     bool ro;
3464     int ret;
3465 
3466     bs = qmp_get_root_bs(device, errp);
3467     if (!bs) {
3468         return;
3469     }
3470 
3471     aio_context = bdrv_get_aio_context(bs);
3472     aio_context_acquire(aio_context);
3473 
3474     image_bs = bdrv_lookup_bs(NULL, image_node_name, &local_err);
3475     if (local_err) {
3476         error_propagate(errp, local_err);
3477         goto out;
3478     }
3479 
3480     if (!image_bs) {
3481         error_setg(errp, "image file not found");
3482         goto out;
3483     }
3484 
3485     if (bdrv_find_base(image_bs) == image_bs) {
3486         error_setg(errp, "not allowing backing file change on an image "
3487                          "without a backing file");
3488         goto out;
3489     }
3490 
3491     /* even though we are not necessarily operating on bs, we need it to
3492      * determine if block ops are currently prohibited on the chain */
3493     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_CHANGE, errp)) {
3494         goto out;
3495     }
3496 
3497     /* final sanity check */
3498     if (!bdrv_chain_contains(bs, image_bs)) {
3499         error_setg(errp, "'%s' and image file are not in the same chain",
3500                    device);
3501         goto out;
3502     }
3503 
3504     /* if not r/w, reopen to make r/w */
3505     ro = bdrv_is_read_only(image_bs);
3506 
3507     if (ro) {
3508         if (bdrv_reopen_set_read_only(image_bs, false, errp) != 0) {
3509             goto out;
3510         }
3511     }
3512 
3513     ret = bdrv_change_backing_file(image_bs, backing_file,
3514                                    image_bs->drv ? image_bs->drv->format_name : "",
3515                                    false);
3516 
3517     if (ret < 0) {
3518         error_setg_errno(errp, -ret, "Could not change backing file to '%s'",
3519                          backing_file);
3520         /* don't exit here, so we can try to restore open flags if
3521          * appropriate */
3522     }
3523 
3524     if (ro) {
3525         bdrv_reopen_set_read_only(image_bs, true, errp);
3526     }
3527 
3528 out:
3529     aio_context_release(aio_context);
3530 }
3531 
3532 void qmp_blockdev_add(BlockdevOptions *options, Error **errp)
3533 {
3534     BlockDriverState *bs;
3535     QObject *obj;
3536     Visitor *v = qobject_output_visitor_new(&obj);
3537     QDict *qdict;
3538 
3539     visit_type_BlockdevOptions(v, NULL, &options, &error_abort);
3540     visit_complete(v, &obj);
3541     qdict = qobject_to(QDict, obj);
3542 
3543     qdict_flatten(qdict);
3544 
3545     if (!qdict_get_try_str(qdict, "node-name")) {
3546         error_setg(errp, "'node-name' must be specified for the root node");
3547         goto fail;
3548     }
3549 
3550     bs = bds_tree_init(qdict, errp);
3551     if (!bs) {
3552         goto fail;
3553     }
3554 
3555     bdrv_set_monitor_owned(bs);
3556 
3557 fail:
3558     visit_free(v);
3559 }
3560 
3561 void qmp_blockdev_reopen(BlockdevOptionsList *reopen_list, Error **errp)
3562 {
3563     BlockReopenQueue *queue = NULL;
3564     GSList *drained = NULL;
3565 
3566     /* Add each one of the BDS that we want to reopen to the queue */
3567     for (; reopen_list != NULL; reopen_list = reopen_list->next) {
3568         BlockdevOptions *options = reopen_list->value;
3569         BlockDriverState *bs;
3570         AioContext *ctx;
3571         QObject *obj;
3572         Visitor *v;
3573         QDict *qdict;
3574 
3575         /* Check for the selected node name */
3576         if (!options->has_node_name) {
3577             error_setg(errp, "node-name not specified");
3578             goto fail;
3579         }
3580 
3581         bs = bdrv_find_node(options->node_name);
3582         if (!bs) {
3583             error_setg(errp, "Failed to find node with node-name='%s'",
3584                        options->node_name);
3585             goto fail;
3586         }
3587 
3588         /* Put all options in a QDict and flatten it */
3589         v = qobject_output_visitor_new(&obj);
3590         visit_type_BlockdevOptions(v, NULL, &options, &error_abort);
3591         visit_complete(v, &obj);
3592         visit_free(v);
3593 
3594         qdict = qobject_to(QDict, obj);
3595 
3596         qdict_flatten(qdict);
3597 
3598         ctx = bdrv_get_aio_context(bs);
3599         aio_context_acquire(ctx);
3600 
3601         bdrv_subtree_drained_begin(bs);
3602         queue = bdrv_reopen_queue(queue, bs, qdict, false);
3603         drained = g_slist_prepend(drained, bs);
3604 
3605         aio_context_release(ctx);
3606     }
3607 
3608     /* Perform the reopen operation */
3609     bdrv_reopen_multiple(queue, errp);
3610     queue = NULL;
3611 
3612 fail:
3613     bdrv_reopen_queue_free(queue);
3614     g_slist_free_full(drained, (GDestroyNotify) bdrv_subtree_drained_end);
3615 }
3616 
3617 void qmp_blockdev_del(const char *node_name, Error **errp)
3618 {
3619     AioContext *aio_context;
3620     BlockDriverState *bs;
3621 
3622     bs = bdrv_find_node(node_name);
3623     if (!bs) {
3624         error_setg(errp, "Failed to find node with node-name='%s'", node_name);
3625         return;
3626     }
3627     if (bdrv_has_blk(bs)) {
3628         error_setg(errp, "Node %s is in use", node_name);
3629         return;
3630     }
3631     aio_context = bdrv_get_aio_context(bs);
3632     aio_context_acquire(aio_context);
3633 
3634     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_DRIVE_DEL, errp)) {
3635         goto out;
3636     }
3637 
3638     if (!QTAILQ_IN_USE(bs, monitor_list)) {
3639         error_setg(errp, "Node %s is not owned by the monitor",
3640                    bs->node_name);
3641         goto out;
3642     }
3643 
3644     if (bs->refcnt > 1) {
3645         error_setg(errp, "Block device %s is in use",
3646                    bdrv_get_device_or_node_name(bs));
3647         goto out;
3648     }
3649 
3650     QTAILQ_REMOVE(&monitor_bdrv_states, bs, monitor_list);
3651     bdrv_unref(bs);
3652 
3653 out:
3654     aio_context_release(aio_context);
3655 }
3656 
3657 static BdrvChild *bdrv_find_child(BlockDriverState *parent_bs,
3658                                   const char *child_name)
3659 {
3660     BdrvChild *child;
3661 
3662     QLIST_FOREACH(child, &parent_bs->children, next) {
3663         if (strcmp(child->name, child_name) == 0) {
3664             return child;
3665         }
3666     }
3667 
3668     return NULL;
3669 }
3670 
3671 void qmp_x_blockdev_change(const char *parent, bool has_child,
3672                            const char *child, bool has_node,
3673                            const char *node, Error **errp)
3674 {
3675     BlockDriverState *parent_bs, *new_bs = NULL;
3676     BdrvChild *p_child;
3677 
3678     parent_bs = bdrv_lookup_bs(parent, parent, errp);
3679     if (!parent_bs) {
3680         return;
3681     }
3682 
3683     if (has_child == has_node) {
3684         if (has_child) {
3685             error_setg(errp, "The parameters child and node are in conflict");
3686         } else {
3687             error_setg(errp, "Either child or node must be specified");
3688         }
3689         return;
3690     }
3691 
3692     if (has_child) {
3693         p_child = bdrv_find_child(parent_bs, child);
3694         if (!p_child) {
3695             error_setg(errp, "Node '%s' does not have child '%s'",
3696                        parent, child);
3697             return;
3698         }
3699         bdrv_del_child(parent_bs, p_child, errp);
3700     }
3701 
3702     if (has_node) {
3703         new_bs = bdrv_find_node(node);
3704         if (!new_bs) {
3705             error_setg(errp, "Node '%s' not found", node);
3706             return;
3707         }
3708         bdrv_add_child(parent_bs, new_bs, errp);
3709     }
3710 }
3711 
3712 BlockJobInfoList *qmp_query_block_jobs(Error **errp)
3713 {
3714     BlockJobInfoList *head = NULL, **tail = &head;
3715     BlockJob *job;
3716 
3717     for (job = block_job_next(NULL); job; job = block_job_next(job)) {
3718         BlockJobInfo *value;
3719         AioContext *aio_context;
3720 
3721         if (block_job_is_internal(job)) {
3722             continue;
3723         }
3724         aio_context = blk_get_aio_context(job->blk);
3725         aio_context_acquire(aio_context);
3726         value = block_job_query(job, errp);
3727         aio_context_release(aio_context);
3728         if (!value) {
3729             qapi_free_BlockJobInfoList(head);
3730             return NULL;
3731         }
3732         QAPI_LIST_APPEND(tail, value);
3733     }
3734 
3735     return head;
3736 }
3737 
3738 void qmp_x_blockdev_set_iothread(const char *node_name, StrOrNull *iothread,
3739                                  bool has_force, bool force, Error **errp)
3740 {
3741     AioContext *old_context;
3742     AioContext *new_context;
3743     BlockDriverState *bs;
3744 
3745     bs = bdrv_find_node(node_name);
3746     if (!bs) {
3747         error_setg(errp, "Failed to find node with node-name='%s'", node_name);
3748         return;
3749     }
3750 
3751     /* Protects against accidents. */
3752     if (!(has_force && force) && bdrv_has_blk(bs)) {
3753         error_setg(errp, "Node %s is associated with a BlockBackend and could "
3754                          "be in use (use force=true to override this check)",
3755                          node_name);
3756         return;
3757     }
3758 
3759     if (iothread->type == QTYPE_QSTRING) {
3760         IOThread *obj = iothread_by_id(iothread->u.s);
3761         if (!obj) {
3762             error_setg(errp, "Cannot find iothread %s", iothread->u.s);
3763             return;
3764         }
3765 
3766         new_context = iothread_get_aio_context(obj);
3767     } else {
3768         new_context = qemu_get_aio_context();
3769     }
3770 
3771     old_context = bdrv_get_aio_context(bs);
3772     aio_context_acquire(old_context);
3773 
3774     bdrv_try_set_aio_context(bs, new_context, errp);
3775 
3776     aio_context_release(old_context);
3777 }
3778 
3779 QemuOptsList qemu_common_drive_opts = {
3780     .name = "drive",
3781     .head = QTAILQ_HEAD_INITIALIZER(qemu_common_drive_opts.head),
3782     .desc = {
3783         {
3784             .name = "snapshot",
3785             .type = QEMU_OPT_BOOL,
3786             .help = "enable/disable snapshot mode",
3787         },{
3788             .name = "aio",
3789             .type = QEMU_OPT_STRING,
3790             .help = "host AIO implementation (threads, native, io_uring)",
3791         },{
3792             .name = BDRV_OPT_CACHE_WB,
3793             .type = QEMU_OPT_BOOL,
3794             .help = "Enable writeback mode",
3795         },{
3796             .name = "format",
3797             .type = QEMU_OPT_STRING,
3798             .help = "disk format (raw, qcow2, ...)",
3799         },{
3800             .name = "rerror",
3801             .type = QEMU_OPT_STRING,
3802             .help = "read error action",
3803         },{
3804             .name = "werror",
3805             .type = QEMU_OPT_STRING,
3806             .help = "write error action",
3807         },{
3808             .name = BDRV_OPT_READ_ONLY,
3809             .type = QEMU_OPT_BOOL,
3810             .help = "open drive file as read-only",
3811         },
3812 
3813         THROTTLE_OPTS,
3814 
3815         {
3816             .name = "throttling.group",
3817             .type = QEMU_OPT_STRING,
3818             .help = "name of the block throttling group",
3819         },{
3820             .name = "copy-on-read",
3821             .type = QEMU_OPT_BOOL,
3822             .help = "copy read data from backing file into image file",
3823         },{
3824             .name = "detect-zeroes",
3825             .type = QEMU_OPT_STRING,
3826             .help = "try to optimize zero writes (off, on, unmap)",
3827         },{
3828             .name = "stats-account-invalid",
3829             .type = QEMU_OPT_BOOL,
3830             .help = "whether to account for invalid I/O operations "
3831                     "in the statistics",
3832         },{
3833             .name = "stats-account-failed",
3834             .type = QEMU_OPT_BOOL,
3835             .help = "whether to account for failed I/O operations "
3836                     "in the statistics",
3837         },
3838         { /* end of list */ }
3839     },
3840 };
3841 
3842 QemuOptsList qemu_drive_opts = {
3843     .name = "drive",
3844     .head = QTAILQ_HEAD_INITIALIZER(qemu_drive_opts.head),
3845     .desc = {
3846         /*
3847          * no elements => accept any params
3848          * validation will happen later
3849          */
3850         { /* end of list */ }
3851     },
3852 };
3853