xref: /openbmc/qemu/blockdev.c (revision 6016b7b4)
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 static void bdrv_format_print(void *opaque, const char *name)
307 {
308     qemu_printf(" %s", name);
309 }
310 
311 typedef struct {
312     QEMUBH *bh;
313     BlockDriverState *bs;
314 } BDRVPutRefBH;
315 
316 static int parse_block_error_action(const char *buf, bool is_read, Error **errp)
317 {
318     if (!strcmp(buf, "ignore")) {
319         return BLOCKDEV_ON_ERROR_IGNORE;
320     } else if (!is_read && !strcmp(buf, "enospc")) {
321         return BLOCKDEV_ON_ERROR_ENOSPC;
322     } else if (!strcmp(buf, "stop")) {
323         return BLOCKDEV_ON_ERROR_STOP;
324     } else if (!strcmp(buf, "report")) {
325         return BLOCKDEV_ON_ERROR_REPORT;
326     } else {
327         error_setg(errp, "'%s' invalid %s error action",
328                    buf, is_read ? "read" : "write");
329         return -1;
330     }
331 }
332 
333 static bool parse_stats_intervals(BlockAcctStats *stats, QList *intervals,
334                                   Error **errp)
335 {
336     const QListEntry *entry;
337     for (entry = qlist_first(intervals); entry; entry = qlist_next(entry)) {
338         switch (qobject_type(entry->value)) {
339 
340         case QTYPE_QSTRING: {
341             unsigned long long length;
342             const char *str = qstring_get_str(qobject_to(QString,
343                                                          entry->value));
344             if (parse_uint_full(str, &length, 10) == 0 &&
345                 length > 0 && length <= UINT_MAX) {
346                 block_acct_add_interval(stats, (unsigned) length);
347             } else {
348                 error_setg(errp, "Invalid interval length: %s", str);
349                 return false;
350             }
351             break;
352         }
353 
354         case QTYPE_QNUM: {
355             int64_t length = qnum_get_int(qobject_to(QNum, entry->value));
356 
357             if (length > 0 && length <= UINT_MAX) {
358                 block_acct_add_interval(stats, (unsigned) length);
359             } else {
360                 error_setg(errp, "Invalid interval length: %" PRId64, length);
361                 return false;
362             }
363             break;
364         }
365 
366         default:
367             error_setg(errp, "The specification of stats-intervals is invalid");
368             return false;
369         }
370     }
371     return true;
372 }
373 
374 typedef enum { MEDIA_DISK, MEDIA_CDROM } DriveMediaType;
375 
376 /* All parameters but @opts are optional and may be set to NULL. */
377 static void extract_common_blockdev_options(QemuOpts *opts, int *bdrv_flags,
378     const char **throttling_group, ThrottleConfig *throttle_cfg,
379     BlockdevDetectZeroesOptions *detect_zeroes, Error **errp)
380 {
381     Error *local_error = NULL;
382     const char *aio;
383 
384     if (bdrv_flags) {
385         if (qemu_opt_get_bool(opts, "copy-on-read", false)) {
386             *bdrv_flags |= BDRV_O_COPY_ON_READ;
387         }
388 
389         if ((aio = qemu_opt_get(opts, "aio")) != NULL) {
390             if (bdrv_parse_aio(aio, bdrv_flags) < 0) {
391                 error_setg(errp, "invalid aio option");
392                 return;
393             }
394         }
395     }
396 
397     /* disk I/O throttling */
398     if (throttling_group) {
399         *throttling_group = qemu_opt_get(opts, "throttling.group");
400     }
401 
402     if (throttle_cfg) {
403         throttle_config_init(throttle_cfg);
404         throttle_cfg->buckets[THROTTLE_BPS_TOTAL].avg =
405             qemu_opt_get_number(opts, "throttling.bps-total", 0);
406         throttle_cfg->buckets[THROTTLE_BPS_READ].avg  =
407             qemu_opt_get_number(opts, "throttling.bps-read", 0);
408         throttle_cfg->buckets[THROTTLE_BPS_WRITE].avg =
409             qemu_opt_get_number(opts, "throttling.bps-write", 0);
410         throttle_cfg->buckets[THROTTLE_OPS_TOTAL].avg =
411             qemu_opt_get_number(opts, "throttling.iops-total", 0);
412         throttle_cfg->buckets[THROTTLE_OPS_READ].avg =
413             qemu_opt_get_number(opts, "throttling.iops-read", 0);
414         throttle_cfg->buckets[THROTTLE_OPS_WRITE].avg =
415             qemu_opt_get_number(opts, "throttling.iops-write", 0);
416 
417         throttle_cfg->buckets[THROTTLE_BPS_TOTAL].max =
418             qemu_opt_get_number(opts, "throttling.bps-total-max", 0);
419         throttle_cfg->buckets[THROTTLE_BPS_READ].max  =
420             qemu_opt_get_number(opts, "throttling.bps-read-max", 0);
421         throttle_cfg->buckets[THROTTLE_BPS_WRITE].max =
422             qemu_opt_get_number(opts, "throttling.bps-write-max", 0);
423         throttle_cfg->buckets[THROTTLE_OPS_TOTAL].max =
424             qemu_opt_get_number(opts, "throttling.iops-total-max", 0);
425         throttle_cfg->buckets[THROTTLE_OPS_READ].max =
426             qemu_opt_get_number(opts, "throttling.iops-read-max", 0);
427         throttle_cfg->buckets[THROTTLE_OPS_WRITE].max =
428             qemu_opt_get_number(opts, "throttling.iops-write-max", 0);
429 
430         throttle_cfg->buckets[THROTTLE_BPS_TOTAL].burst_length =
431             qemu_opt_get_number(opts, "throttling.bps-total-max-length", 1);
432         throttle_cfg->buckets[THROTTLE_BPS_READ].burst_length  =
433             qemu_opt_get_number(opts, "throttling.bps-read-max-length", 1);
434         throttle_cfg->buckets[THROTTLE_BPS_WRITE].burst_length =
435             qemu_opt_get_number(opts, "throttling.bps-write-max-length", 1);
436         throttle_cfg->buckets[THROTTLE_OPS_TOTAL].burst_length =
437             qemu_opt_get_number(opts, "throttling.iops-total-max-length", 1);
438         throttle_cfg->buckets[THROTTLE_OPS_READ].burst_length =
439             qemu_opt_get_number(opts, "throttling.iops-read-max-length", 1);
440         throttle_cfg->buckets[THROTTLE_OPS_WRITE].burst_length =
441             qemu_opt_get_number(opts, "throttling.iops-write-max-length", 1);
442 
443         throttle_cfg->op_size =
444             qemu_opt_get_number(opts, "throttling.iops-size", 0);
445 
446         if (!throttle_is_valid(throttle_cfg, errp)) {
447             return;
448         }
449     }
450 
451     if (detect_zeroes) {
452         *detect_zeroes =
453             qapi_enum_parse(&BlockdevDetectZeroesOptions_lookup,
454                             qemu_opt_get(opts, "detect-zeroes"),
455                             BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF,
456                             &local_error);
457         if (local_error) {
458             error_propagate(errp, local_error);
459             return;
460         }
461     }
462 }
463 
464 /* Takes the ownership of bs_opts */
465 static BlockBackend *blockdev_init(const char *file, QDict *bs_opts,
466                                    Error **errp)
467 {
468     const char *buf;
469     int bdrv_flags = 0;
470     int on_read_error, on_write_error;
471     bool account_invalid, account_failed;
472     bool writethrough, read_only;
473     BlockBackend *blk;
474     BlockDriverState *bs;
475     ThrottleConfig cfg;
476     int snapshot = 0;
477     Error *error = NULL;
478     QemuOpts *opts;
479     QDict *interval_dict = NULL;
480     QList *interval_list = NULL;
481     const char *id;
482     BlockdevDetectZeroesOptions detect_zeroes =
483         BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF;
484     const char *throttling_group = NULL;
485 
486     /* Check common options by copying from bs_opts to opts, all other options
487      * stay in bs_opts for processing by bdrv_open(). */
488     id = qdict_get_try_str(bs_opts, "id");
489     opts = qemu_opts_create(&qemu_common_drive_opts, id, 1, errp);
490     if (!opts) {
491         goto err_no_opts;
492     }
493 
494     if (!qemu_opts_absorb_qdict(opts, bs_opts, errp)) {
495         goto early_err;
496     }
497 
498     if (id) {
499         qdict_del(bs_opts, "id");
500     }
501 
502     /* extract parameters */
503     snapshot = qemu_opt_get_bool(opts, "snapshot", 0);
504 
505     account_invalid = qemu_opt_get_bool(opts, "stats-account-invalid", true);
506     account_failed = qemu_opt_get_bool(opts, "stats-account-failed", true);
507 
508     writethrough = !qemu_opt_get_bool(opts, BDRV_OPT_CACHE_WB, true);
509 
510     id = qemu_opts_id(opts);
511 
512     qdict_extract_subqdict(bs_opts, &interval_dict, "stats-intervals.");
513     qdict_array_split(interval_dict, &interval_list);
514 
515     if (qdict_size(interval_dict) != 0) {
516         error_setg(errp, "Invalid option stats-intervals.%s",
517                    qdict_first(interval_dict)->key);
518         goto early_err;
519     }
520 
521     extract_common_blockdev_options(opts, &bdrv_flags, &throttling_group, &cfg,
522                                     &detect_zeroes, &error);
523     if (error) {
524         error_propagate(errp, error);
525         goto early_err;
526     }
527 
528     if ((buf = qemu_opt_get(opts, "format")) != NULL) {
529         if (is_help_option(buf)) {
530             qemu_printf("Supported formats:");
531             bdrv_iterate_format(bdrv_format_print, NULL, false);
532             qemu_printf("\nSupported formats (read-only):");
533             bdrv_iterate_format(bdrv_format_print, NULL, true);
534             qemu_printf("\n");
535             goto early_err;
536         }
537 
538         if (qdict_haskey(bs_opts, "driver")) {
539             error_setg(errp, "Cannot specify both 'driver' and 'format'");
540             goto early_err;
541         }
542         qdict_put_str(bs_opts, "driver", buf);
543     }
544 
545     on_write_error = BLOCKDEV_ON_ERROR_ENOSPC;
546     if ((buf = qemu_opt_get(opts, "werror")) != NULL) {
547         on_write_error = parse_block_error_action(buf, 0, &error);
548         if (error) {
549             error_propagate(errp, error);
550             goto early_err;
551         }
552     }
553 
554     on_read_error = BLOCKDEV_ON_ERROR_REPORT;
555     if ((buf = qemu_opt_get(opts, "rerror")) != NULL) {
556         on_read_error = parse_block_error_action(buf, 1, &error);
557         if (error) {
558             error_propagate(errp, error);
559             goto early_err;
560         }
561     }
562 
563     if (snapshot) {
564         bdrv_flags |= BDRV_O_SNAPSHOT;
565     }
566 
567     read_only = qemu_opt_get_bool(opts, BDRV_OPT_READ_ONLY, false);
568 
569     /* init */
570     if ((!file || !*file) && !qdict_size(bs_opts)) {
571         BlockBackendRootState *blk_rs;
572 
573         blk = blk_new(qemu_get_aio_context(), 0, BLK_PERM_ALL);
574         blk_rs = blk_get_root_state(blk);
575         blk_rs->open_flags    = bdrv_flags | (read_only ? 0 : BDRV_O_RDWR);
576         blk_rs->detect_zeroes = detect_zeroes;
577 
578         qobject_unref(bs_opts);
579     } else {
580         if (file && !*file) {
581             file = NULL;
582         }
583 
584         /* bdrv_open() defaults to the values in bdrv_flags (for compatibility
585          * with other callers) rather than what we want as the real defaults.
586          * Apply the defaults here instead. */
587         qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_DIRECT, "off");
588         qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_NO_FLUSH, "off");
589         qdict_set_default_str(bs_opts, BDRV_OPT_READ_ONLY,
590                               read_only ? "on" : "off");
591         qdict_set_default_str(bs_opts, BDRV_OPT_AUTO_READ_ONLY, "on");
592         assert((bdrv_flags & BDRV_O_CACHE_MASK) == 0);
593 
594         if (runstate_check(RUN_STATE_INMIGRATE)) {
595             bdrv_flags |= BDRV_O_INACTIVE;
596         }
597 
598         blk = blk_new_open(file, NULL, bs_opts, bdrv_flags, errp);
599         if (!blk) {
600             goto err_no_bs_opts;
601         }
602         bs = blk_bs(blk);
603 
604         bs->detect_zeroes = detect_zeroes;
605 
606         block_acct_setup(blk_get_stats(blk), account_invalid, account_failed);
607 
608         if (!parse_stats_intervals(blk_get_stats(blk), interval_list, errp)) {
609             blk_unref(blk);
610             blk = NULL;
611             goto err_no_bs_opts;
612         }
613     }
614 
615     /* disk I/O throttling */
616     if (throttle_enabled(&cfg)) {
617         if (!throttling_group) {
618             throttling_group = id;
619         }
620         blk_io_limits_enable(blk, throttling_group);
621         blk_set_io_limits(blk, &cfg);
622     }
623 
624     blk_set_enable_write_cache(blk, !writethrough);
625     blk_set_on_error(blk, on_read_error, on_write_error);
626 
627     if (!monitor_add_blk(blk, id, errp)) {
628         blk_unref(blk);
629         blk = NULL;
630         goto err_no_bs_opts;
631     }
632 
633 err_no_bs_opts:
634     qemu_opts_del(opts);
635     qobject_unref(interval_dict);
636     qobject_unref(interval_list);
637     return blk;
638 
639 early_err:
640     qemu_opts_del(opts);
641     qobject_unref(interval_dict);
642     qobject_unref(interval_list);
643 err_no_opts:
644     qobject_unref(bs_opts);
645     return NULL;
646 }
647 
648 /* Takes the ownership of bs_opts */
649 BlockDriverState *bds_tree_init(QDict *bs_opts, Error **errp)
650 {
651     int bdrv_flags = 0;
652 
653     /* bdrv_open() defaults to the values in bdrv_flags (for compatibility
654      * with other callers) rather than what we want as the real defaults.
655      * Apply the defaults here instead. */
656     qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_DIRECT, "off");
657     qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_NO_FLUSH, "off");
658     qdict_set_default_str(bs_opts, BDRV_OPT_READ_ONLY, "off");
659 
660     if (runstate_check(RUN_STATE_INMIGRATE)) {
661         bdrv_flags |= BDRV_O_INACTIVE;
662     }
663 
664     return bdrv_open(NULL, NULL, bs_opts, bdrv_flags, errp);
665 }
666 
667 void blockdev_close_all_bdrv_states(void)
668 {
669     BlockDriverState *bs, *next_bs;
670 
671     QTAILQ_FOREACH_SAFE(bs, &monitor_bdrv_states, monitor_list, next_bs) {
672         AioContext *ctx = bdrv_get_aio_context(bs);
673 
674         aio_context_acquire(ctx);
675         bdrv_unref(bs);
676         aio_context_release(ctx);
677     }
678 }
679 
680 /* Iterates over the list of monitor-owned BlockDriverStates */
681 BlockDriverState *bdrv_next_monitor_owned(BlockDriverState *bs)
682 {
683     return bs ? QTAILQ_NEXT(bs, monitor_list)
684               : QTAILQ_FIRST(&monitor_bdrv_states);
685 }
686 
687 static bool qemu_opt_rename(QemuOpts *opts, const char *from, const char *to,
688                             Error **errp)
689 {
690     const char *value;
691 
692     value = qemu_opt_get(opts, from);
693     if (value) {
694         if (qemu_opt_find(opts, to)) {
695             error_setg(errp, "'%s' and its alias '%s' can't be used at the "
696                        "same time", to, from);
697             return false;
698         }
699     }
700 
701     /* rename all items in opts */
702     while ((value = qemu_opt_get(opts, from))) {
703         qemu_opt_set(opts, to, value, &error_abort);
704         qemu_opt_unset(opts, from);
705     }
706     return true;
707 }
708 
709 QemuOptsList qemu_legacy_drive_opts = {
710     .name = "drive",
711     .head = QTAILQ_HEAD_INITIALIZER(qemu_legacy_drive_opts.head),
712     .desc = {
713         {
714             .name = "bus",
715             .type = QEMU_OPT_NUMBER,
716             .help = "bus number",
717         },{
718             .name = "unit",
719             .type = QEMU_OPT_NUMBER,
720             .help = "unit number (i.e. lun for scsi)",
721         },{
722             .name = "index",
723             .type = QEMU_OPT_NUMBER,
724             .help = "index number",
725         },{
726             .name = "media",
727             .type = QEMU_OPT_STRING,
728             .help = "media type (disk, cdrom)",
729         },{
730             .name = "if",
731             .type = QEMU_OPT_STRING,
732             .help = "interface (ide, scsi, sd, mtd, floppy, pflash, virtio)",
733         },{
734             .name = "file",
735             .type = QEMU_OPT_STRING,
736             .help = "file name",
737         },
738 
739         /* Options that are passed on, but have special semantics with -drive */
740         {
741             .name = BDRV_OPT_READ_ONLY,
742             .type = QEMU_OPT_BOOL,
743             .help = "open drive file as read-only",
744         },{
745             .name = "rerror",
746             .type = QEMU_OPT_STRING,
747             .help = "read error action",
748         },{
749             .name = "werror",
750             .type = QEMU_OPT_STRING,
751             .help = "write error action",
752         },{
753             .name = "copy-on-read",
754             .type = QEMU_OPT_BOOL,
755             .help = "copy read data from backing file into image file",
756         },
757 
758         { /* end of list */ }
759     },
760 };
761 
762 DriveInfo *drive_new(QemuOpts *all_opts, BlockInterfaceType block_default_type,
763                      Error **errp)
764 {
765     const char *value;
766     BlockBackend *blk;
767     DriveInfo *dinfo = NULL;
768     QDict *bs_opts;
769     QemuOpts *legacy_opts;
770     DriveMediaType media = MEDIA_DISK;
771     BlockInterfaceType type;
772     int max_devs, bus_id, unit_id, index;
773     const char *werror, *rerror;
774     bool read_only = false;
775     bool copy_on_read;
776     const char *filename;
777     int i;
778 
779     /* Change legacy command line options into QMP ones */
780     static const struct {
781         const char *from;
782         const char *to;
783     } opt_renames[] = {
784         { "iops",           "throttling.iops-total" },
785         { "iops_rd",        "throttling.iops-read" },
786         { "iops_wr",        "throttling.iops-write" },
787 
788         { "bps",            "throttling.bps-total" },
789         { "bps_rd",         "throttling.bps-read" },
790         { "bps_wr",         "throttling.bps-write" },
791 
792         { "iops_max",       "throttling.iops-total-max" },
793         { "iops_rd_max",    "throttling.iops-read-max" },
794         { "iops_wr_max",    "throttling.iops-write-max" },
795 
796         { "bps_max",        "throttling.bps-total-max" },
797         { "bps_rd_max",     "throttling.bps-read-max" },
798         { "bps_wr_max",     "throttling.bps-write-max" },
799 
800         { "iops_size",      "throttling.iops-size" },
801 
802         { "group",          "throttling.group" },
803 
804         { "readonly",       BDRV_OPT_READ_ONLY },
805     };
806 
807     for (i = 0; i < ARRAY_SIZE(opt_renames); i++) {
808         if (!qemu_opt_rename(all_opts, opt_renames[i].from,
809                              opt_renames[i].to, errp)) {
810             return NULL;
811         }
812     }
813 
814     value = qemu_opt_get(all_opts, "cache");
815     if (value) {
816         int flags = 0;
817         bool writethrough;
818 
819         if (bdrv_parse_cache_mode(value, &flags, &writethrough) != 0) {
820             error_setg(errp, "invalid cache option");
821             return NULL;
822         }
823 
824         /* Specific options take precedence */
825         if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_WB)) {
826             qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_WB,
827                               !writethrough, &error_abort);
828         }
829         if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_DIRECT)) {
830             qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_DIRECT,
831                               !!(flags & BDRV_O_NOCACHE), &error_abort);
832         }
833         if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_NO_FLUSH)) {
834             qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_NO_FLUSH,
835                               !!(flags & BDRV_O_NO_FLUSH), &error_abort);
836         }
837         qemu_opt_unset(all_opts, "cache");
838     }
839 
840     /* Get a QDict for processing the options */
841     bs_opts = qdict_new();
842     qemu_opts_to_qdict(all_opts, bs_opts);
843 
844     legacy_opts = qemu_opts_create(&qemu_legacy_drive_opts, NULL, 0,
845                                    &error_abort);
846     if (!qemu_opts_absorb_qdict(legacy_opts, bs_opts, errp)) {
847         goto fail;
848     }
849 
850     /* Media type */
851     value = qemu_opt_get(legacy_opts, "media");
852     if (value) {
853         if (!strcmp(value, "disk")) {
854             media = MEDIA_DISK;
855         } else if (!strcmp(value, "cdrom")) {
856             media = MEDIA_CDROM;
857             read_only = true;
858         } else {
859             error_setg(errp, "'%s' invalid media", value);
860             goto fail;
861         }
862     }
863 
864     /* copy-on-read is disabled with a warning for read-only devices */
865     read_only |= qemu_opt_get_bool(legacy_opts, BDRV_OPT_READ_ONLY, false);
866     copy_on_read = qemu_opt_get_bool(legacy_opts, "copy-on-read", false);
867 
868     if (read_only && copy_on_read) {
869         warn_report("disabling copy-on-read on read-only drive");
870         copy_on_read = false;
871     }
872 
873     qdict_put_str(bs_opts, BDRV_OPT_READ_ONLY, read_only ? "on" : "off");
874     qdict_put_str(bs_opts, "copy-on-read", copy_on_read ? "on" : "off");
875 
876     /* Controller type */
877     value = qemu_opt_get(legacy_opts, "if");
878     if (value) {
879         for (type = 0;
880              type < IF_COUNT && strcmp(value, if_name[type]);
881              type++) {
882         }
883         if (type == IF_COUNT) {
884             error_setg(errp, "unsupported bus type '%s'", value);
885             goto fail;
886         }
887     } else {
888         type = block_default_type;
889     }
890 
891     /* Device address specified by bus/unit or index.
892      * If none was specified, try to find the first free one. */
893     bus_id  = qemu_opt_get_number(legacy_opts, "bus", 0);
894     unit_id = qemu_opt_get_number(legacy_opts, "unit", -1);
895     index   = qemu_opt_get_number(legacy_opts, "index", -1);
896 
897     max_devs = if_max_devs[type];
898 
899     if (index != -1) {
900         if (bus_id != 0 || unit_id != -1) {
901             error_setg(errp, "index cannot be used with bus and unit");
902             goto fail;
903         }
904         bus_id = drive_index_to_bus_id(type, index);
905         unit_id = drive_index_to_unit_id(type, index);
906     }
907 
908     if (unit_id == -1) {
909        unit_id = 0;
910        while (drive_get(type, bus_id, unit_id) != NULL) {
911            unit_id++;
912            if (max_devs && unit_id >= max_devs) {
913                unit_id -= max_devs;
914                bus_id++;
915            }
916        }
917     }
918 
919     if (max_devs && unit_id >= max_devs) {
920         error_setg(errp, "unit %d too big (max is %d)", unit_id, max_devs - 1);
921         goto fail;
922     }
923 
924     if (drive_get(type, bus_id, unit_id) != NULL) {
925         error_setg(errp, "drive with bus=%d, unit=%d (index=%d) exists",
926                    bus_id, unit_id, index);
927         goto fail;
928     }
929 
930     /* no id supplied -> create one */
931     if (qemu_opts_id(all_opts) == NULL) {
932         char *new_id;
933         const char *mediastr = "";
934         if (type == IF_IDE || type == IF_SCSI) {
935             mediastr = (media == MEDIA_CDROM) ? "-cd" : "-hd";
936         }
937         if (max_devs) {
938             new_id = g_strdup_printf("%s%i%s%i", if_name[type], bus_id,
939                                      mediastr, unit_id);
940         } else {
941             new_id = g_strdup_printf("%s%s%i", if_name[type],
942                                      mediastr, unit_id);
943         }
944         qdict_put_str(bs_opts, "id", new_id);
945         g_free(new_id);
946     }
947 
948     /* Add virtio block device */
949     if (type == IF_VIRTIO) {
950         QemuOpts *devopts;
951         devopts = qemu_opts_create(qemu_find_opts("device"), NULL, 0,
952                                    &error_abort);
953         qemu_opt_set(devopts, "driver", "virtio-blk", &error_abort);
954         qemu_opt_set(devopts, "drive", qdict_get_str(bs_opts, "id"),
955                      &error_abort);
956     }
957 
958     filename = qemu_opt_get(legacy_opts, "file");
959 
960     /* Check werror/rerror compatibility with if=... */
961     werror = qemu_opt_get(legacy_opts, "werror");
962     if (werror != NULL) {
963         if (type != IF_IDE && type != IF_SCSI && type != IF_VIRTIO &&
964             type != IF_NONE) {
965             error_setg(errp, "werror is not supported by this bus type");
966             goto fail;
967         }
968         qdict_put_str(bs_opts, "werror", werror);
969     }
970 
971     rerror = qemu_opt_get(legacy_opts, "rerror");
972     if (rerror != NULL) {
973         if (type != IF_IDE && type != IF_VIRTIO && type != IF_SCSI &&
974             type != IF_NONE) {
975             error_setg(errp, "rerror is not supported by this bus type");
976             goto fail;
977         }
978         qdict_put_str(bs_opts, "rerror", rerror);
979     }
980 
981     /* Actual block device init: Functionality shared with blockdev-add */
982     blk = blockdev_init(filename, bs_opts, errp);
983     bs_opts = NULL;
984     if (!blk) {
985         goto fail;
986     }
987 
988     /* Create legacy DriveInfo */
989     dinfo = g_malloc0(sizeof(*dinfo));
990     dinfo->opts = all_opts;
991 
992     dinfo->type = type;
993     dinfo->bus = bus_id;
994     dinfo->unit = unit_id;
995 
996     blk_set_legacy_dinfo(blk, dinfo);
997 
998     switch(type) {
999     case IF_IDE:
1000     case IF_SCSI:
1001     case IF_XEN:
1002     case IF_NONE:
1003         dinfo->media_cd = media == MEDIA_CDROM;
1004         break;
1005     default:
1006         break;
1007     }
1008 
1009 fail:
1010     qemu_opts_del(legacy_opts);
1011     qobject_unref(bs_opts);
1012     return dinfo;
1013 }
1014 
1015 static BlockDriverState *qmp_get_root_bs(const char *name, Error **errp)
1016 {
1017     BlockDriverState *bs;
1018 
1019     bs = bdrv_lookup_bs(name, name, errp);
1020     if (bs == NULL) {
1021         return NULL;
1022     }
1023 
1024     if (!bdrv_is_root_node(bs)) {
1025         error_setg(errp, "Need a root block node");
1026         return NULL;
1027     }
1028 
1029     if (!bdrv_is_inserted(bs)) {
1030         error_setg(errp, "Device has no medium");
1031         return NULL;
1032     }
1033 
1034     return bs;
1035 }
1036 
1037 static void blockdev_do_action(TransactionAction *action, Error **errp)
1038 {
1039     TransactionActionList list;
1040 
1041     list.value = action;
1042     list.next = NULL;
1043     qmp_transaction(&list, false, NULL, errp);
1044 }
1045 
1046 void qmp_blockdev_snapshot_sync(bool has_device, const char *device,
1047                                 bool has_node_name, const char *node_name,
1048                                 const char *snapshot_file,
1049                                 bool has_snapshot_node_name,
1050                                 const char *snapshot_node_name,
1051                                 bool has_format, const char *format,
1052                                 bool has_mode, NewImageMode mode, Error **errp)
1053 {
1054     BlockdevSnapshotSync snapshot = {
1055         .has_device = has_device,
1056         .device = (char *) device,
1057         .has_node_name = has_node_name,
1058         .node_name = (char *) node_name,
1059         .snapshot_file = (char *) snapshot_file,
1060         .has_snapshot_node_name = has_snapshot_node_name,
1061         .snapshot_node_name = (char *) snapshot_node_name,
1062         .has_format = has_format,
1063         .format = (char *) format,
1064         .has_mode = has_mode,
1065         .mode = mode,
1066     };
1067     TransactionAction action = {
1068         .type = TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC,
1069         .u.blockdev_snapshot_sync.data = &snapshot,
1070     };
1071     blockdev_do_action(&action, errp);
1072 }
1073 
1074 void qmp_blockdev_snapshot(const char *node, const char *overlay,
1075                            Error **errp)
1076 {
1077     BlockdevSnapshot snapshot_data = {
1078         .node = (char *) node,
1079         .overlay = (char *) overlay
1080     };
1081     TransactionAction action = {
1082         .type = TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT,
1083         .u.blockdev_snapshot.data = &snapshot_data,
1084     };
1085     blockdev_do_action(&action, errp);
1086 }
1087 
1088 void qmp_blockdev_snapshot_internal_sync(const char *device,
1089                                          const char *name,
1090                                          Error **errp)
1091 {
1092     BlockdevSnapshotInternal snapshot = {
1093         .device = (char *) device,
1094         .name = (char *) name
1095     };
1096     TransactionAction action = {
1097         .type = TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC,
1098         .u.blockdev_snapshot_internal_sync.data = &snapshot,
1099     };
1100     blockdev_do_action(&action, errp);
1101 }
1102 
1103 SnapshotInfo *qmp_blockdev_snapshot_delete_internal_sync(const char *device,
1104                                                          bool has_id,
1105                                                          const char *id,
1106                                                          bool has_name,
1107                                                          const char *name,
1108                                                          Error **errp)
1109 {
1110     BlockDriverState *bs;
1111     AioContext *aio_context;
1112     QEMUSnapshotInfo sn;
1113     Error *local_err = NULL;
1114     SnapshotInfo *info = NULL;
1115     int ret;
1116 
1117     bs = qmp_get_root_bs(device, errp);
1118     if (!bs) {
1119         return NULL;
1120     }
1121     aio_context = bdrv_get_aio_context(bs);
1122     aio_context_acquire(aio_context);
1123 
1124     if (!has_id) {
1125         id = NULL;
1126     }
1127 
1128     if (!has_name) {
1129         name = NULL;
1130     }
1131 
1132     if (!id && !name) {
1133         error_setg(errp, "Name or id must be provided");
1134         goto out_aio_context;
1135     }
1136 
1137     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_INTERNAL_SNAPSHOT_DELETE, errp)) {
1138         goto out_aio_context;
1139     }
1140 
1141     ret = bdrv_snapshot_find_by_id_and_name(bs, id, name, &sn, &local_err);
1142     if (local_err) {
1143         error_propagate(errp, local_err);
1144         goto out_aio_context;
1145     }
1146     if (!ret) {
1147         error_setg(errp,
1148                    "Snapshot with id '%s' and name '%s' does not exist on "
1149                    "device '%s'",
1150                    STR_OR_NULL(id), STR_OR_NULL(name), device);
1151         goto out_aio_context;
1152     }
1153 
1154     bdrv_snapshot_delete(bs, id, name, &local_err);
1155     if (local_err) {
1156         error_propagate(errp, local_err);
1157         goto out_aio_context;
1158     }
1159 
1160     aio_context_release(aio_context);
1161 
1162     info = g_new0(SnapshotInfo, 1);
1163     info->id = g_strdup(sn.id_str);
1164     info->name = g_strdup(sn.name);
1165     info->date_nsec = sn.date_nsec;
1166     info->date_sec = sn.date_sec;
1167     info->vm_state_size = sn.vm_state_size;
1168     info->vm_clock_nsec = sn.vm_clock_nsec % 1000000000;
1169     info->vm_clock_sec = sn.vm_clock_nsec / 1000000000;
1170     if (sn.icount != -1ULL) {
1171         info->icount = sn.icount;
1172         info->has_icount = true;
1173     }
1174 
1175     return info;
1176 
1177 out_aio_context:
1178     aio_context_release(aio_context);
1179     return NULL;
1180 }
1181 
1182 /* New and old BlockDriverState structs for atomic group operations */
1183 
1184 typedef struct BlkActionState BlkActionState;
1185 
1186 /**
1187  * BlkActionOps:
1188  * Table of operations that define an Action.
1189  *
1190  * @instance_size: Size of state struct, in bytes.
1191  * @prepare: Prepare the work, must NOT be NULL.
1192  * @commit: Commit the changes, can be NULL.
1193  * @abort: Abort the changes on fail, can be NULL.
1194  * @clean: Clean up resources after all transaction actions have called
1195  *         commit() or abort(). Can be NULL.
1196  *
1197  * Only prepare() may fail. In a single transaction, only one of commit() or
1198  * abort() will be called. clean() will always be called if it is present.
1199  */
1200 typedef struct BlkActionOps {
1201     size_t instance_size;
1202     void (*prepare)(BlkActionState *common, Error **errp);
1203     void (*commit)(BlkActionState *common);
1204     void (*abort)(BlkActionState *common);
1205     void (*clean)(BlkActionState *common);
1206 } BlkActionOps;
1207 
1208 /**
1209  * BlkActionState:
1210  * Describes one Action's state within a Transaction.
1211  *
1212  * @action: QAPI-defined enum identifying which Action to perform.
1213  * @ops: Table of ActionOps this Action can perform.
1214  * @block_job_txn: Transaction which this action belongs to.
1215  * @entry: List membership for all Actions in this Transaction.
1216  *
1217  * This structure must be arranged as first member in a subclassed type,
1218  * assuming that the compiler will also arrange it to the same offsets as the
1219  * base class.
1220  */
1221 struct BlkActionState {
1222     TransactionAction *action;
1223     const BlkActionOps *ops;
1224     JobTxn *block_job_txn;
1225     TransactionProperties *txn_props;
1226     QTAILQ_ENTRY(BlkActionState) entry;
1227 };
1228 
1229 /* internal snapshot private data */
1230 typedef struct InternalSnapshotState {
1231     BlkActionState common;
1232     BlockDriverState *bs;
1233     QEMUSnapshotInfo sn;
1234     bool created;
1235 } InternalSnapshotState;
1236 
1237 
1238 static int action_check_completion_mode(BlkActionState *s, Error **errp)
1239 {
1240     if (s->txn_props->completion_mode != ACTION_COMPLETION_MODE_INDIVIDUAL) {
1241         error_setg(errp,
1242                    "Action '%s' does not support Transaction property "
1243                    "completion-mode = %s",
1244                    TransactionActionKind_str(s->action->type),
1245                    ActionCompletionMode_str(s->txn_props->completion_mode));
1246         return -1;
1247     }
1248     return 0;
1249 }
1250 
1251 static void internal_snapshot_prepare(BlkActionState *common,
1252                                       Error **errp)
1253 {
1254     Error *local_err = NULL;
1255     const char *device;
1256     const char *name;
1257     BlockDriverState *bs;
1258     QEMUSnapshotInfo old_sn, *sn;
1259     bool ret;
1260     qemu_timeval tv;
1261     BlockdevSnapshotInternal *internal;
1262     InternalSnapshotState *state;
1263     AioContext *aio_context;
1264     int ret1;
1265 
1266     g_assert(common->action->type ==
1267              TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC);
1268     internal = common->action->u.blockdev_snapshot_internal_sync.data;
1269     state = DO_UPCAST(InternalSnapshotState, common, common);
1270 
1271     /* 1. parse input */
1272     device = internal->device;
1273     name = internal->name;
1274 
1275     /* 2. check for validation */
1276     if (action_check_completion_mode(common, errp) < 0) {
1277         return;
1278     }
1279 
1280     bs = qmp_get_root_bs(device, errp);
1281     if (!bs) {
1282         return;
1283     }
1284 
1285     aio_context = bdrv_get_aio_context(bs);
1286     aio_context_acquire(aio_context);
1287 
1288     state->bs = bs;
1289 
1290     /* Paired with .clean() */
1291     bdrv_drained_begin(bs);
1292 
1293     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_INTERNAL_SNAPSHOT, errp)) {
1294         goto out;
1295     }
1296 
1297     if (bdrv_is_read_only(bs)) {
1298         error_setg(errp, "Device '%s' is read only", device);
1299         goto out;
1300     }
1301 
1302     if (!bdrv_can_snapshot(bs)) {
1303         error_setg(errp, "Block format '%s' used by device '%s' "
1304                    "does not support internal snapshots",
1305                    bs->drv->format_name, device);
1306         goto out;
1307     }
1308 
1309     if (!strlen(name)) {
1310         error_setg(errp, "Name is empty");
1311         goto out;
1312     }
1313 
1314     /* check whether a snapshot with name exist */
1315     ret = bdrv_snapshot_find_by_id_and_name(bs, NULL, name, &old_sn,
1316                                             &local_err);
1317     if (local_err) {
1318         error_propagate(errp, local_err);
1319         goto out;
1320     } else if (ret) {
1321         error_setg(errp,
1322                    "Snapshot with name '%s' already exists on device '%s'",
1323                    name, device);
1324         goto out;
1325     }
1326 
1327     /* 3. take the snapshot */
1328     sn = &state->sn;
1329     pstrcpy(sn->name, sizeof(sn->name), name);
1330     qemu_gettimeofday(&tv);
1331     sn->date_sec = tv.tv_sec;
1332     sn->date_nsec = tv.tv_usec * 1000;
1333     sn->vm_clock_nsec = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL);
1334     if (replay_mode != REPLAY_MODE_NONE) {
1335         sn->icount = replay_get_current_icount();
1336     } else {
1337         sn->icount = -1ULL;
1338     }
1339 
1340     ret1 = bdrv_snapshot_create(bs, sn);
1341     if (ret1 < 0) {
1342         error_setg_errno(errp, -ret1,
1343                          "Failed to create snapshot '%s' on device '%s'",
1344                          name, device);
1345         goto out;
1346     }
1347 
1348     /* 4. succeed, mark a snapshot is created */
1349     state->created = true;
1350 
1351 out:
1352     aio_context_release(aio_context);
1353 }
1354 
1355 static void internal_snapshot_abort(BlkActionState *common)
1356 {
1357     InternalSnapshotState *state =
1358                              DO_UPCAST(InternalSnapshotState, common, common);
1359     BlockDriverState *bs = state->bs;
1360     QEMUSnapshotInfo *sn = &state->sn;
1361     AioContext *aio_context;
1362     Error *local_error = NULL;
1363 
1364     if (!state->created) {
1365         return;
1366     }
1367 
1368     aio_context = bdrv_get_aio_context(state->bs);
1369     aio_context_acquire(aio_context);
1370 
1371     if (bdrv_snapshot_delete(bs, sn->id_str, sn->name, &local_error) < 0) {
1372         error_reportf_err(local_error,
1373                           "Failed to delete snapshot with id '%s' and "
1374                           "name '%s' on device '%s' in abort: ",
1375                           sn->id_str, sn->name,
1376                           bdrv_get_device_name(bs));
1377     }
1378 
1379     aio_context_release(aio_context);
1380 }
1381 
1382 static void internal_snapshot_clean(BlkActionState *common)
1383 {
1384     InternalSnapshotState *state = DO_UPCAST(InternalSnapshotState,
1385                                              common, common);
1386     AioContext *aio_context;
1387 
1388     if (!state->bs) {
1389         return;
1390     }
1391 
1392     aio_context = bdrv_get_aio_context(state->bs);
1393     aio_context_acquire(aio_context);
1394 
1395     bdrv_drained_end(state->bs);
1396 
1397     aio_context_release(aio_context);
1398 }
1399 
1400 /* external snapshot private data */
1401 typedef struct ExternalSnapshotState {
1402     BlkActionState common;
1403     BlockDriverState *old_bs;
1404     BlockDriverState *new_bs;
1405     bool overlay_appended;
1406 } ExternalSnapshotState;
1407 
1408 static void external_snapshot_prepare(BlkActionState *common,
1409                                       Error **errp)
1410 {
1411     int ret;
1412     int flags = 0;
1413     QDict *options = NULL;
1414     Error *local_err = NULL;
1415     /* Device and node name of the image to generate the snapshot from */
1416     const char *device;
1417     const char *node_name;
1418     /* Reference to the new image (for 'blockdev-snapshot') */
1419     const char *snapshot_ref;
1420     /* File name of the new image (for 'blockdev-snapshot-sync') */
1421     const char *new_image_file;
1422     ExternalSnapshotState *state =
1423                              DO_UPCAST(ExternalSnapshotState, common, common);
1424     TransactionAction *action = common->action;
1425     AioContext *aio_context;
1426     uint64_t perm, shared;
1427 
1428     /* 'blockdev-snapshot' and 'blockdev-snapshot-sync' have similar
1429      * purpose but a different set of parameters */
1430     switch (action->type) {
1431     case TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT:
1432         {
1433             BlockdevSnapshot *s = action->u.blockdev_snapshot.data;
1434             device = s->node;
1435             node_name = s->node;
1436             new_image_file = NULL;
1437             snapshot_ref = s->overlay;
1438         }
1439         break;
1440     case TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC:
1441         {
1442             BlockdevSnapshotSync *s = action->u.blockdev_snapshot_sync.data;
1443             device = s->has_device ? s->device : NULL;
1444             node_name = s->has_node_name ? s->node_name : NULL;
1445             new_image_file = s->snapshot_file;
1446             snapshot_ref = NULL;
1447         }
1448         break;
1449     default:
1450         g_assert_not_reached();
1451     }
1452 
1453     /* start processing */
1454     if (action_check_completion_mode(common, errp) < 0) {
1455         return;
1456     }
1457 
1458     state->old_bs = bdrv_lookup_bs(device, node_name, errp);
1459     if (!state->old_bs) {
1460         return;
1461     }
1462 
1463     aio_context = bdrv_get_aio_context(state->old_bs);
1464     aio_context_acquire(aio_context);
1465 
1466     /* Paired with .clean() */
1467     bdrv_drained_begin(state->old_bs);
1468 
1469     if (!bdrv_is_inserted(state->old_bs)) {
1470         error_setg(errp, QERR_DEVICE_HAS_NO_MEDIUM, device);
1471         goto out;
1472     }
1473 
1474     if (bdrv_op_is_blocked(state->old_bs,
1475                            BLOCK_OP_TYPE_EXTERNAL_SNAPSHOT, errp)) {
1476         goto out;
1477     }
1478 
1479     if (!bdrv_is_read_only(state->old_bs)) {
1480         if (bdrv_flush(state->old_bs)) {
1481             error_setg(errp, QERR_IO_ERROR);
1482             goto out;
1483         }
1484     }
1485 
1486     if (action->type == TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC) {
1487         BlockdevSnapshotSync *s = action->u.blockdev_snapshot_sync.data;
1488         const char *format = s->has_format ? s->format : "qcow2";
1489         enum NewImageMode mode;
1490         const char *snapshot_node_name =
1491             s->has_snapshot_node_name ? s->snapshot_node_name : NULL;
1492 
1493         if (node_name && !snapshot_node_name) {
1494             error_setg(errp, "New overlay node-name missing");
1495             goto out;
1496         }
1497 
1498         if (snapshot_node_name &&
1499             bdrv_lookup_bs(snapshot_node_name, snapshot_node_name, NULL)) {
1500             error_setg(errp, "New overlay node-name already in use");
1501             goto out;
1502         }
1503 
1504         flags = state->old_bs->open_flags;
1505         flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_COPY_ON_READ);
1506         flags |= BDRV_O_NO_BACKING;
1507 
1508         /* create new image w/backing file */
1509         mode = s->has_mode ? s->mode : NEW_IMAGE_MODE_ABSOLUTE_PATHS;
1510         if (mode != NEW_IMAGE_MODE_EXISTING) {
1511             int64_t size = bdrv_getlength(state->old_bs);
1512             if (size < 0) {
1513                 error_setg_errno(errp, -size, "bdrv_getlength failed");
1514                 goto out;
1515             }
1516             bdrv_refresh_filename(state->old_bs);
1517             bdrv_img_create(new_image_file, format,
1518                             state->old_bs->filename,
1519                             state->old_bs->drv->format_name,
1520                             NULL, size, flags, false, &local_err);
1521             if (local_err) {
1522                 error_propagate(errp, local_err);
1523                 goto out;
1524             }
1525         }
1526 
1527         options = qdict_new();
1528         if (snapshot_node_name) {
1529             qdict_put_str(options, "node-name", snapshot_node_name);
1530         }
1531         qdict_put_str(options, "driver", format);
1532     }
1533 
1534     state->new_bs = bdrv_open(new_image_file, snapshot_ref, options, flags,
1535                               errp);
1536     /* We will manually add the backing_hd field to the bs later */
1537     if (!state->new_bs) {
1538         goto out;
1539     }
1540 
1541     /*
1542      * Allow attaching a backing file to an overlay that's already in use only
1543      * if the parents don't assume that they are already seeing a valid image.
1544      * (Specifically, allow it as a mirror target, which is write-only access.)
1545      */
1546     bdrv_get_cumulative_perm(state->new_bs, &perm, &shared);
1547     if (perm & BLK_PERM_CONSISTENT_READ) {
1548         error_setg(errp, "The overlay is already in use");
1549         goto out;
1550     }
1551 
1552     if (state->new_bs->drv->is_filter) {
1553         error_setg(errp, "Filters cannot be used as overlays");
1554         goto out;
1555     }
1556 
1557     if (bdrv_cow_child(state->new_bs)) {
1558         error_setg(errp, "The overlay already has a backing image");
1559         goto out;
1560     }
1561 
1562     if (!state->new_bs->drv->supports_backing) {
1563         error_setg(errp, "The overlay does not support backing images");
1564         goto out;
1565     }
1566 
1567     ret = bdrv_append(state->new_bs, state->old_bs, errp);
1568     if (ret < 0) {
1569         goto out;
1570     }
1571     state->overlay_appended = true;
1572 
1573 out:
1574     aio_context_release(aio_context);
1575 }
1576 
1577 static void external_snapshot_commit(BlkActionState *common)
1578 {
1579     ExternalSnapshotState *state =
1580                              DO_UPCAST(ExternalSnapshotState, common, common);
1581     AioContext *aio_context;
1582 
1583     aio_context = bdrv_get_aio_context(state->old_bs);
1584     aio_context_acquire(aio_context);
1585 
1586     /* We don't need (or want) to use the transactional
1587      * bdrv_reopen_multiple() across all the entries at once, because we
1588      * don't want to abort all of them if one of them fails the reopen */
1589     if (!qatomic_read(&state->old_bs->copy_on_read)) {
1590         bdrv_reopen_set_read_only(state->old_bs, true, NULL);
1591     }
1592 
1593     aio_context_release(aio_context);
1594 }
1595 
1596 static void external_snapshot_abort(BlkActionState *common)
1597 {
1598     ExternalSnapshotState *state =
1599                              DO_UPCAST(ExternalSnapshotState, common, common);
1600     if (state->new_bs) {
1601         if (state->overlay_appended) {
1602             AioContext *aio_context;
1603             AioContext *tmp_context;
1604             int ret;
1605 
1606             aio_context = bdrv_get_aio_context(state->old_bs);
1607             aio_context_acquire(aio_context);
1608 
1609             bdrv_ref(state->old_bs);   /* we can't let bdrv_set_backind_hd()
1610                                           close state->old_bs; we need it */
1611             bdrv_set_backing_hd(state->new_bs, NULL, &error_abort);
1612 
1613             /*
1614              * The call to bdrv_set_backing_hd() above returns state->old_bs to
1615              * the main AioContext. As we're still going to be using it, return
1616              * it to the AioContext it was before.
1617              */
1618             tmp_context = bdrv_get_aio_context(state->old_bs);
1619             if (aio_context != tmp_context) {
1620                 aio_context_release(aio_context);
1621                 aio_context_acquire(tmp_context);
1622 
1623                 ret = bdrv_try_set_aio_context(state->old_bs,
1624                                                aio_context, NULL);
1625                 assert(ret == 0);
1626 
1627                 aio_context_release(tmp_context);
1628                 aio_context_acquire(aio_context);
1629             }
1630 
1631             bdrv_replace_node(state->new_bs, state->old_bs, &error_abort);
1632             bdrv_unref(state->old_bs); /* bdrv_replace_node() ref'ed old_bs */
1633 
1634             aio_context_release(aio_context);
1635         }
1636     }
1637 }
1638 
1639 static void external_snapshot_clean(BlkActionState *common)
1640 {
1641     ExternalSnapshotState *state =
1642                              DO_UPCAST(ExternalSnapshotState, common, common);
1643     AioContext *aio_context;
1644 
1645     if (!state->old_bs) {
1646         return;
1647     }
1648 
1649     aio_context = bdrv_get_aio_context(state->old_bs);
1650     aio_context_acquire(aio_context);
1651 
1652     bdrv_drained_end(state->old_bs);
1653     bdrv_unref(state->new_bs);
1654 
1655     aio_context_release(aio_context);
1656 }
1657 
1658 typedef struct DriveBackupState {
1659     BlkActionState common;
1660     BlockDriverState *bs;
1661     BlockJob *job;
1662 } DriveBackupState;
1663 
1664 static BlockJob *do_backup_common(BackupCommon *backup,
1665                                   BlockDriverState *bs,
1666                                   BlockDriverState *target_bs,
1667                                   AioContext *aio_context,
1668                                   JobTxn *txn, Error **errp);
1669 
1670 static void drive_backup_prepare(BlkActionState *common, Error **errp)
1671 {
1672     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1673     DriveBackup *backup;
1674     BlockDriverState *bs;
1675     BlockDriverState *target_bs;
1676     BlockDriverState *source = NULL;
1677     AioContext *aio_context;
1678     AioContext *old_context;
1679     QDict *options;
1680     Error *local_err = NULL;
1681     int flags;
1682     int64_t size;
1683     bool set_backing_hd = false;
1684     int ret;
1685 
1686     assert(common->action->type == TRANSACTION_ACTION_KIND_DRIVE_BACKUP);
1687     backup = common->action->u.drive_backup.data;
1688 
1689     if (!backup->has_mode) {
1690         backup->mode = NEW_IMAGE_MODE_ABSOLUTE_PATHS;
1691     }
1692 
1693     bs = bdrv_lookup_bs(backup->device, backup->device, errp);
1694     if (!bs) {
1695         return;
1696     }
1697 
1698     if (!bs->drv) {
1699         error_setg(errp, "Device has no medium");
1700         return;
1701     }
1702 
1703     aio_context = bdrv_get_aio_context(bs);
1704     aio_context_acquire(aio_context);
1705 
1706     state->bs = bs;
1707     /* Paired with .clean() */
1708     bdrv_drained_begin(bs);
1709 
1710     if (!backup->has_format) {
1711         backup->format = backup->mode == NEW_IMAGE_MODE_EXISTING ?
1712                          NULL : (char *) bs->drv->format_name;
1713     }
1714 
1715     /* Early check to avoid creating target */
1716     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_BACKUP_SOURCE, errp)) {
1717         goto out;
1718     }
1719 
1720     flags = bs->open_flags | BDRV_O_RDWR;
1721 
1722     /*
1723      * See if we have a backing HD we can use to create our new image
1724      * on top of.
1725      */
1726     if (backup->sync == MIRROR_SYNC_MODE_TOP) {
1727         /*
1728          * Backup will not replace the source by the target, so none
1729          * of the filters skipped here will be removed (in contrast to
1730          * mirror).  Therefore, we can skip all of them when looking
1731          * for the first COW relationship.
1732          */
1733         source = bdrv_cow_bs(bdrv_skip_filters(bs));
1734         if (!source) {
1735             backup->sync = MIRROR_SYNC_MODE_FULL;
1736         }
1737     }
1738     if (backup->sync == MIRROR_SYNC_MODE_NONE) {
1739         source = bs;
1740         flags |= BDRV_O_NO_BACKING;
1741         set_backing_hd = true;
1742     }
1743 
1744     size = bdrv_getlength(bs);
1745     if (size < 0) {
1746         error_setg_errno(errp, -size, "bdrv_getlength failed");
1747         goto out;
1748     }
1749 
1750     if (backup->mode != NEW_IMAGE_MODE_EXISTING) {
1751         assert(backup->format);
1752         if (source) {
1753             /* Implicit filters should not appear in the filename */
1754             BlockDriverState *explicit_backing =
1755                 bdrv_skip_implicit_filters(source);
1756 
1757             bdrv_refresh_filename(explicit_backing);
1758             bdrv_img_create(backup->target, backup->format,
1759                             explicit_backing->filename,
1760                             explicit_backing->drv->format_name, NULL,
1761                             size, flags, false, &local_err);
1762         } else {
1763             bdrv_img_create(backup->target, backup->format, NULL, NULL, NULL,
1764                             size, flags, false, &local_err);
1765         }
1766     }
1767 
1768     if (local_err) {
1769         error_propagate(errp, local_err);
1770         goto out;
1771     }
1772 
1773     options = qdict_new();
1774     qdict_put_str(options, "discard", "unmap");
1775     qdict_put_str(options, "detect-zeroes", "unmap");
1776     if (backup->format) {
1777         qdict_put_str(options, "driver", backup->format);
1778     }
1779 
1780     target_bs = bdrv_open(backup->target, NULL, options, flags, errp);
1781     if (!target_bs) {
1782         goto out;
1783     }
1784 
1785     /* Honor bdrv_try_set_aio_context() context acquisition requirements. */
1786     old_context = bdrv_get_aio_context(target_bs);
1787     aio_context_release(aio_context);
1788     aio_context_acquire(old_context);
1789 
1790     ret = bdrv_try_set_aio_context(target_bs, aio_context, errp);
1791     if (ret < 0) {
1792         bdrv_unref(target_bs);
1793         aio_context_release(old_context);
1794         return;
1795     }
1796 
1797     aio_context_release(old_context);
1798     aio_context_acquire(aio_context);
1799 
1800     if (set_backing_hd) {
1801         if (bdrv_set_backing_hd(target_bs, source, errp) < 0) {
1802             goto unref;
1803         }
1804     }
1805 
1806     state->job = do_backup_common(qapi_DriveBackup_base(backup),
1807                                   bs, target_bs, aio_context,
1808                                   common->block_job_txn, errp);
1809 
1810 unref:
1811     bdrv_unref(target_bs);
1812 out:
1813     aio_context_release(aio_context);
1814 }
1815 
1816 static void drive_backup_commit(BlkActionState *common)
1817 {
1818     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1819     AioContext *aio_context;
1820 
1821     aio_context = bdrv_get_aio_context(state->bs);
1822     aio_context_acquire(aio_context);
1823 
1824     assert(state->job);
1825     job_start(&state->job->job);
1826 
1827     aio_context_release(aio_context);
1828 }
1829 
1830 static void drive_backup_abort(BlkActionState *common)
1831 {
1832     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1833 
1834     if (state->job) {
1835         AioContext *aio_context;
1836 
1837         aio_context = bdrv_get_aio_context(state->bs);
1838         aio_context_acquire(aio_context);
1839 
1840         job_cancel_sync(&state->job->job, true);
1841 
1842         aio_context_release(aio_context);
1843     }
1844 }
1845 
1846 static void drive_backup_clean(BlkActionState *common)
1847 {
1848     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1849     AioContext *aio_context;
1850 
1851     if (!state->bs) {
1852         return;
1853     }
1854 
1855     aio_context = bdrv_get_aio_context(state->bs);
1856     aio_context_acquire(aio_context);
1857 
1858     bdrv_drained_end(state->bs);
1859 
1860     aio_context_release(aio_context);
1861 }
1862 
1863 typedef struct BlockdevBackupState {
1864     BlkActionState common;
1865     BlockDriverState *bs;
1866     BlockJob *job;
1867 } BlockdevBackupState;
1868 
1869 static void blockdev_backup_prepare(BlkActionState *common, Error **errp)
1870 {
1871     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1872     BlockdevBackup *backup;
1873     BlockDriverState *bs;
1874     BlockDriverState *target_bs;
1875     AioContext *aio_context;
1876     AioContext *old_context;
1877     int ret;
1878 
1879     assert(common->action->type == TRANSACTION_ACTION_KIND_BLOCKDEV_BACKUP);
1880     backup = common->action->u.blockdev_backup.data;
1881 
1882     bs = bdrv_lookup_bs(backup->device, backup->device, errp);
1883     if (!bs) {
1884         return;
1885     }
1886 
1887     target_bs = bdrv_lookup_bs(backup->target, backup->target, errp);
1888     if (!target_bs) {
1889         return;
1890     }
1891 
1892     /* Honor bdrv_try_set_aio_context() context acquisition requirements. */
1893     aio_context = bdrv_get_aio_context(bs);
1894     old_context = bdrv_get_aio_context(target_bs);
1895     aio_context_acquire(old_context);
1896 
1897     ret = bdrv_try_set_aio_context(target_bs, aio_context, errp);
1898     if (ret < 0) {
1899         aio_context_release(old_context);
1900         return;
1901     }
1902 
1903     aio_context_release(old_context);
1904     aio_context_acquire(aio_context);
1905     state->bs = bs;
1906 
1907     /* Paired with .clean() */
1908     bdrv_drained_begin(state->bs);
1909 
1910     state->job = do_backup_common(qapi_BlockdevBackup_base(backup),
1911                                   bs, target_bs, aio_context,
1912                                   common->block_job_txn, errp);
1913 
1914     aio_context_release(aio_context);
1915 }
1916 
1917 static void blockdev_backup_commit(BlkActionState *common)
1918 {
1919     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1920     AioContext *aio_context;
1921 
1922     aio_context = bdrv_get_aio_context(state->bs);
1923     aio_context_acquire(aio_context);
1924 
1925     assert(state->job);
1926     job_start(&state->job->job);
1927 
1928     aio_context_release(aio_context);
1929 }
1930 
1931 static void blockdev_backup_abort(BlkActionState *common)
1932 {
1933     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1934 
1935     if (state->job) {
1936         AioContext *aio_context;
1937 
1938         aio_context = bdrv_get_aio_context(state->bs);
1939         aio_context_acquire(aio_context);
1940 
1941         job_cancel_sync(&state->job->job, true);
1942 
1943         aio_context_release(aio_context);
1944     }
1945 }
1946 
1947 static void blockdev_backup_clean(BlkActionState *common)
1948 {
1949     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1950     AioContext *aio_context;
1951 
1952     if (!state->bs) {
1953         return;
1954     }
1955 
1956     aio_context = bdrv_get_aio_context(state->bs);
1957     aio_context_acquire(aio_context);
1958 
1959     bdrv_drained_end(state->bs);
1960 
1961     aio_context_release(aio_context);
1962 }
1963 
1964 typedef struct BlockDirtyBitmapState {
1965     BlkActionState common;
1966     BdrvDirtyBitmap *bitmap;
1967     BlockDriverState *bs;
1968     HBitmap *backup;
1969     bool prepared;
1970     bool was_enabled;
1971 } BlockDirtyBitmapState;
1972 
1973 static void block_dirty_bitmap_add_prepare(BlkActionState *common,
1974                                            Error **errp)
1975 {
1976     Error *local_err = NULL;
1977     BlockDirtyBitmapAdd *action;
1978     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
1979                                              common, common);
1980 
1981     if (action_check_completion_mode(common, errp) < 0) {
1982         return;
1983     }
1984 
1985     action = common->action->u.block_dirty_bitmap_add.data;
1986     /* AIO context taken and released within qmp_block_dirty_bitmap_add */
1987     qmp_block_dirty_bitmap_add(action->node, action->name,
1988                                action->has_granularity, action->granularity,
1989                                action->has_persistent, action->persistent,
1990                                action->has_disabled, action->disabled,
1991                                &local_err);
1992 
1993     if (!local_err) {
1994         state->prepared = true;
1995     } else {
1996         error_propagate(errp, local_err);
1997     }
1998 }
1999 
2000 static void block_dirty_bitmap_add_abort(BlkActionState *common)
2001 {
2002     BlockDirtyBitmapAdd *action;
2003     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2004                                              common, common);
2005 
2006     action = common->action->u.block_dirty_bitmap_add.data;
2007     /* Should not be able to fail: IF the bitmap was added via .prepare(),
2008      * then the node reference and bitmap name must have been valid.
2009      */
2010     if (state->prepared) {
2011         qmp_block_dirty_bitmap_remove(action->node, action->name, &error_abort);
2012     }
2013 }
2014 
2015 static void block_dirty_bitmap_clear_prepare(BlkActionState *common,
2016                                              Error **errp)
2017 {
2018     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2019                                              common, common);
2020     BlockDirtyBitmap *action;
2021 
2022     if (action_check_completion_mode(common, errp) < 0) {
2023         return;
2024     }
2025 
2026     action = common->action->u.block_dirty_bitmap_clear.data;
2027     state->bitmap = block_dirty_bitmap_lookup(action->node,
2028                                               action->name,
2029                                               &state->bs,
2030                                               errp);
2031     if (!state->bitmap) {
2032         return;
2033     }
2034 
2035     if (bdrv_dirty_bitmap_check(state->bitmap, BDRV_BITMAP_DEFAULT, errp)) {
2036         return;
2037     }
2038 
2039     bdrv_clear_dirty_bitmap(state->bitmap, &state->backup);
2040 }
2041 
2042 static void block_dirty_bitmap_restore(BlkActionState *common)
2043 {
2044     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2045                                              common, common);
2046 
2047     if (state->backup) {
2048         bdrv_restore_dirty_bitmap(state->bitmap, state->backup);
2049     }
2050 }
2051 
2052 static void block_dirty_bitmap_free_backup(BlkActionState *common)
2053 {
2054     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2055                                              common, common);
2056 
2057     hbitmap_free(state->backup);
2058 }
2059 
2060 static void block_dirty_bitmap_enable_prepare(BlkActionState *common,
2061                                               Error **errp)
2062 {
2063     BlockDirtyBitmap *action;
2064     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2065                                              common, common);
2066 
2067     if (action_check_completion_mode(common, errp) < 0) {
2068         return;
2069     }
2070 
2071     action = common->action->u.block_dirty_bitmap_enable.data;
2072     state->bitmap = block_dirty_bitmap_lookup(action->node,
2073                                               action->name,
2074                                               NULL,
2075                                               errp);
2076     if (!state->bitmap) {
2077         return;
2078     }
2079 
2080     if (bdrv_dirty_bitmap_check(state->bitmap, BDRV_BITMAP_ALLOW_RO, errp)) {
2081         return;
2082     }
2083 
2084     state->was_enabled = bdrv_dirty_bitmap_enabled(state->bitmap);
2085     bdrv_enable_dirty_bitmap(state->bitmap);
2086 }
2087 
2088 static void block_dirty_bitmap_enable_abort(BlkActionState *common)
2089 {
2090     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2091                                              common, common);
2092 
2093     if (!state->was_enabled) {
2094         bdrv_disable_dirty_bitmap(state->bitmap);
2095     }
2096 }
2097 
2098 static void block_dirty_bitmap_disable_prepare(BlkActionState *common,
2099                                                Error **errp)
2100 {
2101     BlockDirtyBitmap *action;
2102     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2103                                              common, common);
2104 
2105     if (action_check_completion_mode(common, errp) < 0) {
2106         return;
2107     }
2108 
2109     action = common->action->u.block_dirty_bitmap_disable.data;
2110     state->bitmap = block_dirty_bitmap_lookup(action->node,
2111                                               action->name,
2112                                               NULL,
2113                                               errp);
2114     if (!state->bitmap) {
2115         return;
2116     }
2117 
2118     if (bdrv_dirty_bitmap_check(state->bitmap, BDRV_BITMAP_ALLOW_RO, errp)) {
2119         return;
2120     }
2121 
2122     state->was_enabled = bdrv_dirty_bitmap_enabled(state->bitmap);
2123     bdrv_disable_dirty_bitmap(state->bitmap);
2124 }
2125 
2126 static void block_dirty_bitmap_disable_abort(BlkActionState *common)
2127 {
2128     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2129                                              common, common);
2130 
2131     if (state->was_enabled) {
2132         bdrv_enable_dirty_bitmap(state->bitmap);
2133     }
2134 }
2135 
2136 static void block_dirty_bitmap_merge_prepare(BlkActionState *common,
2137                                              Error **errp)
2138 {
2139     BlockDirtyBitmapMerge *action;
2140     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2141                                              common, common);
2142 
2143     if (action_check_completion_mode(common, errp) < 0) {
2144         return;
2145     }
2146 
2147     action = common->action->u.block_dirty_bitmap_merge.data;
2148 
2149     state->bitmap = block_dirty_bitmap_merge(action->node, action->target,
2150                                              action->bitmaps, &state->backup,
2151                                              errp);
2152 }
2153 
2154 static void block_dirty_bitmap_remove_prepare(BlkActionState *common,
2155                                               Error **errp)
2156 {
2157     BlockDirtyBitmap *action;
2158     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2159                                              common, common);
2160 
2161     if (action_check_completion_mode(common, errp) < 0) {
2162         return;
2163     }
2164 
2165     action = common->action->u.block_dirty_bitmap_remove.data;
2166 
2167     state->bitmap = block_dirty_bitmap_remove(action->node, action->name,
2168                                               false, &state->bs, errp);
2169     if (state->bitmap) {
2170         bdrv_dirty_bitmap_skip_store(state->bitmap, true);
2171         bdrv_dirty_bitmap_set_busy(state->bitmap, true);
2172     }
2173 }
2174 
2175 static void block_dirty_bitmap_remove_abort(BlkActionState *common)
2176 {
2177     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2178                                              common, common);
2179 
2180     if (state->bitmap) {
2181         bdrv_dirty_bitmap_skip_store(state->bitmap, false);
2182         bdrv_dirty_bitmap_set_busy(state->bitmap, false);
2183     }
2184 }
2185 
2186 static void block_dirty_bitmap_remove_commit(BlkActionState *common)
2187 {
2188     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2189                                              common, common);
2190 
2191     bdrv_dirty_bitmap_set_busy(state->bitmap, false);
2192     bdrv_release_dirty_bitmap(state->bitmap);
2193 }
2194 
2195 static void abort_prepare(BlkActionState *common, Error **errp)
2196 {
2197     error_setg(errp, "Transaction aborted using Abort action");
2198 }
2199 
2200 static void abort_commit(BlkActionState *common)
2201 {
2202     g_assert_not_reached(); /* this action never succeeds */
2203 }
2204 
2205 static const BlkActionOps actions[] = {
2206     [TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT] = {
2207         .instance_size = sizeof(ExternalSnapshotState),
2208         .prepare  = external_snapshot_prepare,
2209         .commit   = external_snapshot_commit,
2210         .abort = external_snapshot_abort,
2211         .clean = external_snapshot_clean,
2212     },
2213     [TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC] = {
2214         .instance_size = sizeof(ExternalSnapshotState),
2215         .prepare  = external_snapshot_prepare,
2216         .commit   = external_snapshot_commit,
2217         .abort = external_snapshot_abort,
2218         .clean = external_snapshot_clean,
2219     },
2220     [TRANSACTION_ACTION_KIND_DRIVE_BACKUP] = {
2221         .instance_size = sizeof(DriveBackupState),
2222         .prepare = drive_backup_prepare,
2223         .commit = drive_backup_commit,
2224         .abort = drive_backup_abort,
2225         .clean = drive_backup_clean,
2226     },
2227     [TRANSACTION_ACTION_KIND_BLOCKDEV_BACKUP] = {
2228         .instance_size = sizeof(BlockdevBackupState),
2229         .prepare = blockdev_backup_prepare,
2230         .commit = blockdev_backup_commit,
2231         .abort = blockdev_backup_abort,
2232         .clean = blockdev_backup_clean,
2233     },
2234     [TRANSACTION_ACTION_KIND_ABORT] = {
2235         .instance_size = sizeof(BlkActionState),
2236         .prepare = abort_prepare,
2237         .commit = abort_commit,
2238     },
2239     [TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC] = {
2240         .instance_size = sizeof(InternalSnapshotState),
2241         .prepare  = internal_snapshot_prepare,
2242         .abort = internal_snapshot_abort,
2243         .clean = internal_snapshot_clean,
2244     },
2245     [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_ADD] = {
2246         .instance_size = sizeof(BlockDirtyBitmapState),
2247         .prepare = block_dirty_bitmap_add_prepare,
2248         .abort = block_dirty_bitmap_add_abort,
2249     },
2250     [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_CLEAR] = {
2251         .instance_size = sizeof(BlockDirtyBitmapState),
2252         .prepare = block_dirty_bitmap_clear_prepare,
2253         .commit = block_dirty_bitmap_free_backup,
2254         .abort = block_dirty_bitmap_restore,
2255     },
2256     [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_ENABLE] = {
2257         .instance_size = sizeof(BlockDirtyBitmapState),
2258         .prepare = block_dirty_bitmap_enable_prepare,
2259         .abort = block_dirty_bitmap_enable_abort,
2260     },
2261     [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_DISABLE] = {
2262         .instance_size = sizeof(BlockDirtyBitmapState),
2263         .prepare = block_dirty_bitmap_disable_prepare,
2264         .abort = block_dirty_bitmap_disable_abort,
2265     },
2266     [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_MERGE] = {
2267         .instance_size = sizeof(BlockDirtyBitmapState),
2268         .prepare = block_dirty_bitmap_merge_prepare,
2269         .commit = block_dirty_bitmap_free_backup,
2270         .abort = block_dirty_bitmap_restore,
2271     },
2272     [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_REMOVE] = {
2273         .instance_size = sizeof(BlockDirtyBitmapState),
2274         .prepare = block_dirty_bitmap_remove_prepare,
2275         .commit = block_dirty_bitmap_remove_commit,
2276         .abort = block_dirty_bitmap_remove_abort,
2277     },
2278     /* Where are transactions for MIRROR, COMMIT and STREAM?
2279      * Although these blockjobs use transaction callbacks like the backup job,
2280      * these jobs do not necessarily adhere to transaction semantics.
2281      * These jobs may not fully undo all of their actions on abort, nor do they
2282      * necessarily work in transactions with more than one job in them.
2283      */
2284 };
2285 
2286 /**
2287  * Allocate a TransactionProperties structure if necessary, and fill
2288  * that structure with desired defaults if they are unset.
2289  */
2290 static TransactionProperties *get_transaction_properties(
2291     TransactionProperties *props)
2292 {
2293     if (!props) {
2294         props = g_new0(TransactionProperties, 1);
2295     }
2296 
2297     if (!props->has_completion_mode) {
2298         props->has_completion_mode = true;
2299         props->completion_mode = ACTION_COMPLETION_MODE_INDIVIDUAL;
2300     }
2301 
2302     return props;
2303 }
2304 
2305 /*
2306  * 'Atomic' group operations.  The operations are performed as a set, and if
2307  * any fail then we roll back all operations in the group.
2308  */
2309 void qmp_transaction(TransactionActionList *dev_list,
2310                      bool has_props,
2311                      struct TransactionProperties *props,
2312                      Error **errp)
2313 {
2314     TransactionActionList *dev_entry = dev_list;
2315     JobTxn *block_job_txn = NULL;
2316     BlkActionState *state, *next;
2317     Error *local_err = NULL;
2318 
2319     QTAILQ_HEAD(, BlkActionState) snap_bdrv_states;
2320     QTAILQ_INIT(&snap_bdrv_states);
2321 
2322     /* Does this transaction get canceled as a group on failure?
2323      * If not, we don't really need to make a JobTxn.
2324      */
2325     props = get_transaction_properties(props);
2326     if (props->completion_mode != ACTION_COMPLETION_MODE_INDIVIDUAL) {
2327         block_job_txn = job_txn_new();
2328     }
2329 
2330     /* drain all i/o before any operations */
2331     bdrv_drain_all();
2332 
2333     /* We don't do anything in this loop that commits us to the operations */
2334     while (NULL != dev_entry) {
2335         TransactionAction *dev_info = NULL;
2336         const BlkActionOps *ops;
2337 
2338         dev_info = dev_entry->value;
2339         dev_entry = dev_entry->next;
2340 
2341         assert(dev_info->type < ARRAY_SIZE(actions));
2342 
2343         ops = &actions[dev_info->type];
2344         assert(ops->instance_size > 0);
2345 
2346         state = g_malloc0(ops->instance_size);
2347         state->ops = ops;
2348         state->action = dev_info;
2349         state->block_job_txn = block_job_txn;
2350         state->txn_props = props;
2351         QTAILQ_INSERT_TAIL(&snap_bdrv_states, state, entry);
2352 
2353         state->ops->prepare(state, &local_err);
2354         if (local_err) {
2355             error_propagate(errp, local_err);
2356             goto delete_and_fail;
2357         }
2358     }
2359 
2360     QTAILQ_FOREACH(state, &snap_bdrv_states, entry) {
2361         if (state->ops->commit) {
2362             state->ops->commit(state);
2363         }
2364     }
2365 
2366     /* success */
2367     goto exit;
2368 
2369 delete_and_fail:
2370     /* failure, and it is all-or-none; roll back all operations */
2371     QTAILQ_FOREACH_REVERSE(state, &snap_bdrv_states, entry) {
2372         if (state->ops->abort) {
2373             state->ops->abort(state);
2374         }
2375     }
2376 exit:
2377     QTAILQ_FOREACH_SAFE(state, &snap_bdrv_states, entry, next) {
2378         if (state->ops->clean) {
2379             state->ops->clean(state);
2380         }
2381         g_free(state);
2382     }
2383     if (!has_props) {
2384         qapi_free_TransactionProperties(props);
2385     }
2386     job_txn_unref(block_job_txn);
2387 }
2388 
2389 BlockDirtyBitmapSha256 *qmp_x_debug_block_dirty_bitmap_sha256(const char *node,
2390                                                               const char *name,
2391                                                               Error **errp)
2392 {
2393     BdrvDirtyBitmap *bitmap;
2394     BlockDriverState *bs;
2395     BlockDirtyBitmapSha256 *ret = NULL;
2396     char *sha256;
2397 
2398     bitmap = block_dirty_bitmap_lookup(node, name, &bs, errp);
2399     if (!bitmap || !bs) {
2400         return NULL;
2401     }
2402 
2403     sha256 = bdrv_dirty_bitmap_sha256(bitmap, errp);
2404     if (sha256 == NULL) {
2405         return NULL;
2406     }
2407 
2408     ret = g_new(BlockDirtyBitmapSha256, 1);
2409     ret->sha256 = sha256;
2410 
2411     return ret;
2412 }
2413 
2414 void coroutine_fn qmp_block_resize(bool has_device, const char *device,
2415                                    bool has_node_name, const char *node_name,
2416                                    int64_t size, Error **errp)
2417 {
2418     Error *local_err = NULL;
2419     BlockBackend *blk;
2420     BlockDriverState *bs;
2421     AioContext *old_ctx;
2422 
2423     bs = bdrv_lookup_bs(has_device ? device : NULL,
2424                         has_node_name ? node_name : NULL,
2425                         &local_err);
2426     if (local_err) {
2427         error_propagate(errp, local_err);
2428         return;
2429     }
2430 
2431     if (size < 0) {
2432         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "size", "a >0 size");
2433         return;
2434     }
2435 
2436     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_RESIZE, NULL)) {
2437         error_setg(errp, QERR_DEVICE_IN_USE, device);
2438         return;
2439     }
2440 
2441     blk = blk_new_with_bs(bs, BLK_PERM_RESIZE, BLK_PERM_ALL, errp);
2442     if (!blk) {
2443         return;
2444     }
2445 
2446     bdrv_co_lock(bs);
2447     bdrv_drained_begin(bs);
2448     bdrv_co_unlock(bs);
2449 
2450     old_ctx = bdrv_co_enter(bs);
2451     blk_truncate(blk, size, false, PREALLOC_MODE_OFF, 0, errp);
2452     bdrv_co_leave(bs, old_ctx);
2453 
2454     bdrv_co_lock(bs);
2455     bdrv_drained_end(bs);
2456     blk_unref(blk);
2457     bdrv_co_unlock(bs);
2458 }
2459 
2460 void qmp_block_stream(bool has_job_id, const char *job_id, const char *device,
2461                       bool has_base, const char *base,
2462                       bool has_base_node, const char *base_node,
2463                       bool has_backing_file, const char *backing_file,
2464                       bool has_bottom, const char *bottom,
2465                       bool has_speed, int64_t speed,
2466                       bool has_on_error, BlockdevOnError on_error,
2467                       bool has_filter_node_name, const char *filter_node_name,
2468                       bool has_auto_finalize, bool auto_finalize,
2469                       bool has_auto_dismiss, bool auto_dismiss,
2470                       Error **errp)
2471 {
2472     BlockDriverState *bs, *iter, *iter_end;
2473     BlockDriverState *base_bs = NULL;
2474     BlockDriverState *bottom_bs = NULL;
2475     AioContext *aio_context;
2476     Error *local_err = NULL;
2477     int job_flags = JOB_DEFAULT;
2478 
2479     if (has_base && has_base_node) {
2480         error_setg(errp, "'base' and 'base-node' cannot be specified "
2481                    "at the same time");
2482         return;
2483     }
2484 
2485     if (has_base && has_bottom) {
2486         error_setg(errp, "'base' and 'bottom' cannot be specified "
2487                    "at the same time");
2488         return;
2489     }
2490 
2491     if (has_bottom && has_base_node) {
2492         error_setg(errp, "'bottom' and 'base-node' cannot be specified "
2493                    "at the same time");
2494         return;
2495     }
2496 
2497     if (!has_on_error) {
2498         on_error = BLOCKDEV_ON_ERROR_REPORT;
2499     }
2500 
2501     bs = bdrv_lookup_bs(device, device, errp);
2502     if (!bs) {
2503         return;
2504     }
2505 
2506     aio_context = bdrv_get_aio_context(bs);
2507     aio_context_acquire(aio_context);
2508 
2509     if (has_base) {
2510         base_bs = bdrv_find_backing_image(bs, base);
2511         if (base_bs == NULL) {
2512             error_setg(errp, "Can't find '%s' in the backing chain", base);
2513             goto out;
2514         }
2515         assert(bdrv_get_aio_context(base_bs) == aio_context);
2516     }
2517 
2518     if (has_base_node) {
2519         base_bs = bdrv_lookup_bs(NULL, base_node, errp);
2520         if (!base_bs) {
2521             goto out;
2522         }
2523         if (bs == base_bs || !bdrv_chain_contains(bs, base_bs)) {
2524             error_setg(errp, "Node '%s' is not a backing image of '%s'",
2525                        base_node, device);
2526             goto out;
2527         }
2528         assert(bdrv_get_aio_context(base_bs) == aio_context);
2529         bdrv_refresh_filename(base_bs);
2530     }
2531 
2532     if (has_bottom) {
2533         bottom_bs = bdrv_lookup_bs(NULL, bottom, errp);
2534         if (!bottom_bs) {
2535             goto out;
2536         }
2537         if (!bottom_bs->drv) {
2538             error_setg(errp, "Node '%s' is not open", bottom);
2539             goto out;
2540         }
2541         if (bottom_bs->drv->is_filter) {
2542             error_setg(errp, "Node '%s' is a filter, use a non-filter node "
2543                        "as 'bottom'", bottom);
2544             goto out;
2545         }
2546         if (!bdrv_chain_contains(bs, bottom_bs)) {
2547             error_setg(errp, "Node '%s' is not in a chain starting from '%s'",
2548                        bottom, device);
2549             goto out;
2550         }
2551         assert(bdrv_get_aio_context(bottom_bs) == aio_context);
2552     }
2553 
2554     /*
2555      * Check for op blockers in the whole chain between bs and base (or bottom)
2556      */
2557     iter_end = has_bottom ? bdrv_filter_or_cow_bs(bottom_bs) : base_bs;
2558     for (iter = bs; iter && iter != iter_end;
2559          iter = bdrv_filter_or_cow_bs(iter))
2560     {
2561         if (bdrv_op_is_blocked(iter, BLOCK_OP_TYPE_STREAM, errp)) {
2562             goto out;
2563         }
2564     }
2565 
2566     /* if we are streaming the entire chain, the result will have no backing
2567      * file, and specifying one is therefore an error */
2568     if (base_bs == NULL && has_backing_file) {
2569         error_setg(errp, "backing file specified, but streaming the "
2570                          "entire chain");
2571         goto out;
2572     }
2573 
2574     if (has_auto_finalize && !auto_finalize) {
2575         job_flags |= JOB_MANUAL_FINALIZE;
2576     }
2577     if (has_auto_dismiss && !auto_dismiss) {
2578         job_flags |= JOB_MANUAL_DISMISS;
2579     }
2580 
2581     stream_start(has_job_id ? job_id : NULL, bs, base_bs, backing_file,
2582                  bottom_bs, job_flags, has_speed ? speed : 0, on_error,
2583                  filter_node_name, &local_err);
2584     if (local_err) {
2585         error_propagate(errp, local_err);
2586         goto out;
2587     }
2588 
2589     trace_qmp_block_stream(bs);
2590 
2591 out:
2592     aio_context_release(aio_context);
2593 }
2594 
2595 void qmp_block_commit(bool has_job_id, const char *job_id, const char *device,
2596                       bool has_base_node, const char *base_node,
2597                       bool has_base, const char *base,
2598                       bool has_top_node, const char *top_node,
2599                       bool has_top, const char *top,
2600                       bool has_backing_file, const char *backing_file,
2601                       bool has_speed, int64_t speed,
2602                       bool has_on_error, BlockdevOnError on_error,
2603                       bool has_filter_node_name, const char *filter_node_name,
2604                       bool has_auto_finalize, bool auto_finalize,
2605                       bool has_auto_dismiss, bool auto_dismiss,
2606                       Error **errp)
2607 {
2608     BlockDriverState *bs;
2609     BlockDriverState *iter;
2610     BlockDriverState *base_bs, *top_bs;
2611     AioContext *aio_context;
2612     Error *local_err = NULL;
2613     int job_flags = JOB_DEFAULT;
2614     uint64_t top_perm, top_shared;
2615 
2616     if (!has_speed) {
2617         speed = 0;
2618     }
2619     if (!has_on_error) {
2620         on_error = BLOCKDEV_ON_ERROR_REPORT;
2621     }
2622     if (!has_filter_node_name) {
2623         filter_node_name = NULL;
2624     }
2625     if (has_auto_finalize && !auto_finalize) {
2626         job_flags |= JOB_MANUAL_FINALIZE;
2627     }
2628     if (has_auto_dismiss && !auto_dismiss) {
2629         job_flags |= JOB_MANUAL_DISMISS;
2630     }
2631 
2632     /* Important Note:
2633      *  libvirt relies on the DeviceNotFound error class in order to probe for
2634      *  live commit feature versions; for this to work, we must make sure to
2635      *  perform the device lookup before any generic errors that may occur in a
2636      *  scenario in which all optional arguments are omitted. */
2637     bs = qmp_get_root_bs(device, &local_err);
2638     if (!bs) {
2639         bs = bdrv_lookup_bs(device, device, NULL);
2640         if (!bs) {
2641             error_free(local_err);
2642             error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
2643                       "Device '%s' not found", device);
2644         } else {
2645             error_propagate(errp, local_err);
2646         }
2647         return;
2648     }
2649 
2650     aio_context = bdrv_get_aio_context(bs);
2651     aio_context_acquire(aio_context);
2652 
2653     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_COMMIT_SOURCE, errp)) {
2654         goto out;
2655     }
2656 
2657     /* default top_bs is the active layer */
2658     top_bs = bs;
2659 
2660     if (has_top_node && has_top) {
2661         error_setg(errp, "'top-node' and 'top' are mutually exclusive");
2662         goto out;
2663     } else if (has_top_node) {
2664         top_bs = bdrv_lookup_bs(NULL, top_node, errp);
2665         if (top_bs == NULL) {
2666             goto out;
2667         }
2668         if (!bdrv_chain_contains(bs, top_bs)) {
2669             error_setg(errp, "'%s' is not in this backing file chain",
2670                        top_node);
2671             goto out;
2672         }
2673     } else if (has_top && top) {
2674         /* This strcmp() is just a shortcut, there is no need to
2675          * refresh @bs's filename.  If it mismatches,
2676          * bdrv_find_backing_image() will do the refresh and may still
2677          * return @bs. */
2678         if (strcmp(bs->filename, top) != 0) {
2679             top_bs = bdrv_find_backing_image(bs, top);
2680         }
2681     }
2682 
2683     if (top_bs == NULL) {
2684         error_setg(errp, "Top image file %s not found", top ? top : "NULL");
2685         goto out;
2686     }
2687 
2688     assert(bdrv_get_aio_context(top_bs) == aio_context);
2689 
2690     if (has_base_node && has_base) {
2691         error_setg(errp, "'base-node' and 'base' are mutually exclusive");
2692         goto out;
2693     } else if (has_base_node) {
2694         base_bs = bdrv_lookup_bs(NULL, base_node, errp);
2695         if (base_bs == NULL) {
2696             goto out;
2697         }
2698         if (!bdrv_chain_contains(top_bs, base_bs)) {
2699             error_setg(errp, "'%s' is not in this backing file chain",
2700                        base_node);
2701             goto out;
2702         }
2703     } else if (has_base && base) {
2704         base_bs = bdrv_find_backing_image(top_bs, base);
2705         if (base_bs == NULL) {
2706             error_setg(errp, "Can't find '%s' in the backing chain", base);
2707             goto out;
2708         }
2709     } else {
2710         base_bs = bdrv_find_base(top_bs);
2711         if (base_bs == NULL) {
2712             error_setg(errp, "There is no backimg image");
2713             goto out;
2714         }
2715     }
2716 
2717     assert(bdrv_get_aio_context(base_bs) == aio_context);
2718 
2719     for (iter = top_bs; iter != bdrv_filter_or_cow_bs(base_bs);
2720          iter = bdrv_filter_or_cow_bs(iter))
2721     {
2722         if (bdrv_op_is_blocked(iter, BLOCK_OP_TYPE_COMMIT_TARGET, errp)) {
2723             goto out;
2724         }
2725     }
2726 
2727     /* Do not allow attempts to commit an image into itself */
2728     if (top_bs == base_bs) {
2729         error_setg(errp, "cannot commit an image into itself");
2730         goto out;
2731     }
2732 
2733     /*
2734      * Active commit is required if and only if someone has taken a
2735      * WRITE permission on the top node.  Historically, we have always
2736      * used active commit for top nodes, so continue that practice
2737      * lest we possibly break clients that rely on this behavior, e.g.
2738      * to later attach this node to a writing parent.
2739      * (Active commit is never really wrong.)
2740      */
2741     bdrv_get_cumulative_perm(top_bs, &top_perm, &top_shared);
2742     if (top_perm & BLK_PERM_WRITE ||
2743         bdrv_skip_filters(top_bs) == bdrv_skip_filters(bs))
2744     {
2745         if (has_backing_file) {
2746             if (bdrv_skip_filters(top_bs) == bdrv_skip_filters(bs)) {
2747                 error_setg(errp, "'backing-file' specified,"
2748                                  " but 'top' is the active layer");
2749             } else {
2750                 error_setg(errp, "'backing-file' specified, but 'top' has a "
2751                                  "writer on it");
2752             }
2753             goto out;
2754         }
2755         if (!has_job_id) {
2756             /*
2757              * Emulate here what block_job_create() does, because it
2758              * is possible that @bs != @top_bs (the block job should
2759              * be named after @bs, even if @top_bs is the actual
2760              * source)
2761              */
2762             job_id = bdrv_get_device_name(bs);
2763         }
2764         commit_active_start(job_id, top_bs, base_bs, job_flags, speed, on_error,
2765                             filter_node_name, NULL, NULL, false, &local_err);
2766     } else {
2767         BlockDriverState *overlay_bs = bdrv_find_overlay(bs, top_bs);
2768         if (bdrv_op_is_blocked(overlay_bs, BLOCK_OP_TYPE_COMMIT_TARGET, errp)) {
2769             goto out;
2770         }
2771         commit_start(has_job_id ? job_id : NULL, bs, base_bs, top_bs, job_flags,
2772                      speed, on_error, has_backing_file ? backing_file : NULL,
2773                      filter_node_name, &local_err);
2774     }
2775     if (local_err != NULL) {
2776         error_propagate(errp, local_err);
2777         goto out;
2778     }
2779 
2780 out:
2781     aio_context_release(aio_context);
2782 }
2783 
2784 /* Common QMP interface for drive-backup and blockdev-backup */
2785 static BlockJob *do_backup_common(BackupCommon *backup,
2786                                   BlockDriverState *bs,
2787                                   BlockDriverState *target_bs,
2788                                   AioContext *aio_context,
2789                                   JobTxn *txn, Error **errp)
2790 {
2791     BlockJob *job = NULL;
2792     BdrvDirtyBitmap *bmap = NULL;
2793     BackupPerf perf = { .max_workers = 64 };
2794     int job_flags = JOB_DEFAULT;
2795 
2796     if (!backup->has_speed) {
2797         backup->speed = 0;
2798     }
2799     if (!backup->has_on_source_error) {
2800         backup->on_source_error = BLOCKDEV_ON_ERROR_REPORT;
2801     }
2802     if (!backup->has_on_target_error) {
2803         backup->on_target_error = BLOCKDEV_ON_ERROR_REPORT;
2804     }
2805     if (!backup->has_job_id) {
2806         backup->job_id = NULL;
2807     }
2808     if (!backup->has_auto_finalize) {
2809         backup->auto_finalize = true;
2810     }
2811     if (!backup->has_auto_dismiss) {
2812         backup->auto_dismiss = true;
2813     }
2814     if (!backup->has_compress) {
2815         backup->compress = false;
2816     }
2817 
2818     if (backup->x_perf) {
2819         if (backup->x_perf->has_use_copy_range) {
2820             perf.use_copy_range = backup->x_perf->use_copy_range;
2821         }
2822         if (backup->x_perf->has_max_workers) {
2823             perf.max_workers = backup->x_perf->max_workers;
2824         }
2825         if (backup->x_perf->has_max_chunk) {
2826             perf.max_chunk = backup->x_perf->max_chunk;
2827         }
2828     }
2829 
2830     if ((backup->sync == MIRROR_SYNC_MODE_BITMAP) ||
2831         (backup->sync == MIRROR_SYNC_MODE_INCREMENTAL)) {
2832         /* done before desugaring 'incremental' to print the right message */
2833         if (!backup->has_bitmap) {
2834             error_setg(errp, "must provide a valid bitmap name for "
2835                        "'%s' sync mode", MirrorSyncMode_str(backup->sync));
2836             return NULL;
2837         }
2838     }
2839 
2840     if (backup->sync == MIRROR_SYNC_MODE_INCREMENTAL) {
2841         if (backup->has_bitmap_mode &&
2842             backup->bitmap_mode != BITMAP_SYNC_MODE_ON_SUCCESS) {
2843             error_setg(errp, "Bitmap sync mode must be '%s' "
2844                        "when using sync mode '%s'",
2845                        BitmapSyncMode_str(BITMAP_SYNC_MODE_ON_SUCCESS),
2846                        MirrorSyncMode_str(backup->sync));
2847             return NULL;
2848         }
2849         backup->has_bitmap_mode = true;
2850         backup->sync = MIRROR_SYNC_MODE_BITMAP;
2851         backup->bitmap_mode = BITMAP_SYNC_MODE_ON_SUCCESS;
2852     }
2853 
2854     if (backup->has_bitmap) {
2855         bmap = bdrv_find_dirty_bitmap(bs, backup->bitmap);
2856         if (!bmap) {
2857             error_setg(errp, "Bitmap '%s' could not be found", backup->bitmap);
2858             return NULL;
2859         }
2860         if (!backup->has_bitmap_mode) {
2861             error_setg(errp, "Bitmap sync mode must be given "
2862                        "when providing a bitmap");
2863             return NULL;
2864         }
2865         if (bdrv_dirty_bitmap_check(bmap, BDRV_BITMAP_ALLOW_RO, errp)) {
2866             return NULL;
2867         }
2868 
2869         /* This does not produce a useful bitmap artifact: */
2870         if (backup->sync == MIRROR_SYNC_MODE_NONE) {
2871             error_setg(errp, "sync mode '%s' does not produce meaningful bitmap"
2872                        " outputs", MirrorSyncMode_str(backup->sync));
2873             return NULL;
2874         }
2875 
2876         /* If the bitmap isn't used for input or output, this is useless: */
2877         if (backup->bitmap_mode == BITMAP_SYNC_MODE_NEVER &&
2878             backup->sync != MIRROR_SYNC_MODE_BITMAP) {
2879             error_setg(errp, "Bitmap sync mode '%s' has no meaningful effect"
2880                        " when combined with sync mode '%s'",
2881                        BitmapSyncMode_str(backup->bitmap_mode),
2882                        MirrorSyncMode_str(backup->sync));
2883             return NULL;
2884         }
2885     }
2886 
2887     if (!backup->has_bitmap && backup->has_bitmap_mode) {
2888         error_setg(errp, "Cannot specify bitmap sync mode without a bitmap");
2889         return NULL;
2890     }
2891 
2892     if (!backup->auto_finalize) {
2893         job_flags |= JOB_MANUAL_FINALIZE;
2894     }
2895     if (!backup->auto_dismiss) {
2896         job_flags |= JOB_MANUAL_DISMISS;
2897     }
2898 
2899     job = backup_job_create(backup->job_id, bs, target_bs, backup->speed,
2900                             backup->sync, bmap, backup->bitmap_mode,
2901                             backup->compress,
2902                             backup->filter_node_name,
2903                             &perf,
2904                             backup->on_source_error,
2905                             backup->on_target_error,
2906                             job_flags, NULL, NULL, txn, errp);
2907     return job;
2908 }
2909 
2910 void qmp_drive_backup(DriveBackup *backup, Error **errp)
2911 {
2912     TransactionAction action = {
2913         .type = TRANSACTION_ACTION_KIND_DRIVE_BACKUP,
2914         .u.drive_backup.data = backup,
2915     };
2916     blockdev_do_action(&action, errp);
2917 }
2918 
2919 BlockDeviceInfoList *qmp_query_named_block_nodes(bool has_flat,
2920                                                  bool flat,
2921                                                  Error **errp)
2922 {
2923     bool return_flat = has_flat && flat;
2924 
2925     return bdrv_named_nodes_list(return_flat, errp);
2926 }
2927 
2928 XDbgBlockGraph *qmp_x_debug_query_block_graph(Error **errp)
2929 {
2930     return bdrv_get_xdbg_block_graph(errp);
2931 }
2932 
2933 void qmp_blockdev_backup(BlockdevBackup *backup, Error **errp)
2934 {
2935     TransactionAction action = {
2936         .type = TRANSACTION_ACTION_KIND_BLOCKDEV_BACKUP,
2937         .u.blockdev_backup.data = backup,
2938     };
2939     blockdev_do_action(&action, errp);
2940 }
2941 
2942 /* Parameter check and block job starting for drive mirroring.
2943  * Caller should hold @device and @target's aio context (must be the same).
2944  **/
2945 static void blockdev_mirror_common(const char *job_id, BlockDriverState *bs,
2946                                    BlockDriverState *target,
2947                                    bool has_replaces, const char *replaces,
2948                                    enum MirrorSyncMode sync,
2949                                    BlockMirrorBackingMode backing_mode,
2950                                    bool zero_target,
2951                                    bool has_speed, int64_t speed,
2952                                    bool has_granularity, uint32_t granularity,
2953                                    bool has_buf_size, int64_t buf_size,
2954                                    bool has_on_source_error,
2955                                    BlockdevOnError on_source_error,
2956                                    bool has_on_target_error,
2957                                    BlockdevOnError on_target_error,
2958                                    bool has_unmap, bool unmap,
2959                                    bool has_filter_node_name,
2960                                    const char *filter_node_name,
2961                                    bool has_copy_mode, MirrorCopyMode copy_mode,
2962                                    bool has_auto_finalize, bool auto_finalize,
2963                                    bool has_auto_dismiss, bool auto_dismiss,
2964                                    Error **errp)
2965 {
2966     BlockDriverState *unfiltered_bs;
2967     int job_flags = JOB_DEFAULT;
2968 
2969     if (!has_speed) {
2970         speed = 0;
2971     }
2972     if (!has_on_source_error) {
2973         on_source_error = BLOCKDEV_ON_ERROR_REPORT;
2974     }
2975     if (!has_on_target_error) {
2976         on_target_error = BLOCKDEV_ON_ERROR_REPORT;
2977     }
2978     if (!has_granularity) {
2979         granularity = 0;
2980     }
2981     if (!has_buf_size) {
2982         buf_size = 0;
2983     }
2984     if (!has_unmap) {
2985         unmap = true;
2986     }
2987     if (!has_filter_node_name) {
2988         filter_node_name = NULL;
2989     }
2990     if (!has_copy_mode) {
2991         copy_mode = MIRROR_COPY_MODE_BACKGROUND;
2992     }
2993     if (has_auto_finalize && !auto_finalize) {
2994         job_flags |= JOB_MANUAL_FINALIZE;
2995     }
2996     if (has_auto_dismiss && !auto_dismiss) {
2997         job_flags |= JOB_MANUAL_DISMISS;
2998     }
2999 
3000     if (granularity != 0 && (granularity < 512 || granularity > 1048576 * 64)) {
3001         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "granularity",
3002                    "a value in range [512B, 64MB]");
3003         return;
3004     }
3005     if (granularity & (granularity - 1)) {
3006         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "granularity",
3007                    "a power of 2");
3008         return;
3009     }
3010 
3011     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_MIRROR_SOURCE, errp)) {
3012         return;
3013     }
3014     if (bdrv_op_is_blocked(target, BLOCK_OP_TYPE_MIRROR_TARGET, errp)) {
3015         return;
3016     }
3017 
3018     if (!bdrv_backing_chain_next(bs) && sync == MIRROR_SYNC_MODE_TOP) {
3019         sync = MIRROR_SYNC_MODE_FULL;
3020     }
3021 
3022     if (!has_replaces) {
3023         /* We want to mirror from @bs, but keep implicit filters on top */
3024         unfiltered_bs = bdrv_skip_implicit_filters(bs);
3025         if (unfiltered_bs != bs) {
3026             replaces = unfiltered_bs->node_name;
3027             has_replaces = true;
3028         }
3029     }
3030 
3031     if (has_replaces) {
3032         BlockDriverState *to_replace_bs;
3033         AioContext *replace_aio_context;
3034         int64_t bs_size, replace_size;
3035 
3036         bs_size = bdrv_getlength(bs);
3037         if (bs_size < 0) {
3038             error_setg_errno(errp, -bs_size, "Failed to query device's size");
3039             return;
3040         }
3041 
3042         to_replace_bs = check_to_replace_node(bs, replaces, errp);
3043         if (!to_replace_bs) {
3044             return;
3045         }
3046 
3047         replace_aio_context = bdrv_get_aio_context(to_replace_bs);
3048         aio_context_acquire(replace_aio_context);
3049         replace_size = bdrv_getlength(to_replace_bs);
3050         aio_context_release(replace_aio_context);
3051 
3052         if (replace_size < 0) {
3053             error_setg_errno(errp, -replace_size,
3054                              "Failed to query the replacement node's size");
3055             return;
3056         }
3057         if (bs_size != replace_size) {
3058             error_setg(errp, "cannot replace image with a mirror image of "
3059                              "different size");
3060             return;
3061         }
3062     }
3063 
3064     /* pass the node name to replace to mirror start since it's loose coupling
3065      * and will allow to check whether the node still exist at mirror completion
3066      */
3067     mirror_start(job_id, bs, target,
3068                  has_replaces ? replaces : NULL, job_flags,
3069                  speed, granularity, buf_size, sync, backing_mode, zero_target,
3070                  on_source_error, on_target_error, unmap, filter_node_name,
3071                  copy_mode, errp);
3072 }
3073 
3074 void qmp_drive_mirror(DriveMirror *arg, Error **errp)
3075 {
3076     BlockDriverState *bs;
3077     BlockDriverState *target_backing_bs, *target_bs;
3078     AioContext *aio_context;
3079     AioContext *old_context;
3080     BlockMirrorBackingMode backing_mode;
3081     Error *local_err = NULL;
3082     QDict *options = NULL;
3083     int flags;
3084     int64_t size;
3085     const char *format = arg->format;
3086     bool zero_target;
3087     int ret;
3088 
3089     bs = qmp_get_root_bs(arg->device, errp);
3090     if (!bs) {
3091         return;
3092     }
3093 
3094     /* Early check to avoid creating target */
3095     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_MIRROR_SOURCE, errp)) {
3096         return;
3097     }
3098 
3099     aio_context = bdrv_get_aio_context(bs);
3100     aio_context_acquire(aio_context);
3101 
3102     if (!arg->has_mode) {
3103         arg->mode = NEW_IMAGE_MODE_ABSOLUTE_PATHS;
3104     }
3105 
3106     if (!arg->has_format) {
3107         format = (arg->mode == NEW_IMAGE_MODE_EXISTING
3108                   ? NULL : bs->drv->format_name);
3109     }
3110 
3111     flags = bs->open_flags | BDRV_O_RDWR;
3112     target_backing_bs = bdrv_cow_bs(bdrv_skip_filters(bs));
3113     if (!target_backing_bs && arg->sync == MIRROR_SYNC_MODE_TOP) {
3114         arg->sync = MIRROR_SYNC_MODE_FULL;
3115     }
3116     if (arg->sync == MIRROR_SYNC_MODE_NONE) {
3117         target_backing_bs = bs;
3118     }
3119 
3120     size = bdrv_getlength(bs);
3121     if (size < 0) {
3122         error_setg_errno(errp, -size, "bdrv_getlength failed");
3123         goto out;
3124     }
3125 
3126     if (arg->has_replaces) {
3127         if (!arg->has_node_name) {
3128             error_setg(errp, "a node-name must be provided when replacing a"
3129                              " named node of the graph");
3130             goto out;
3131         }
3132     }
3133 
3134     if (arg->mode == NEW_IMAGE_MODE_ABSOLUTE_PATHS) {
3135         backing_mode = MIRROR_SOURCE_BACKING_CHAIN;
3136     } else {
3137         backing_mode = MIRROR_OPEN_BACKING_CHAIN;
3138     }
3139 
3140     /* Don't open backing image in create() */
3141     flags |= BDRV_O_NO_BACKING;
3142 
3143     if ((arg->sync == MIRROR_SYNC_MODE_FULL || !target_backing_bs)
3144         && arg->mode != NEW_IMAGE_MODE_EXISTING)
3145     {
3146         /* create new image w/o backing file */
3147         assert(format);
3148         bdrv_img_create(arg->target, format,
3149                         NULL, NULL, NULL, size, flags, false, &local_err);
3150     } else {
3151         /* Implicit filters should not appear in the filename */
3152         BlockDriverState *explicit_backing =
3153             bdrv_skip_implicit_filters(target_backing_bs);
3154 
3155         switch (arg->mode) {
3156         case NEW_IMAGE_MODE_EXISTING:
3157             break;
3158         case NEW_IMAGE_MODE_ABSOLUTE_PATHS:
3159             /* create new image with backing file */
3160             bdrv_refresh_filename(explicit_backing);
3161             bdrv_img_create(arg->target, format,
3162                             explicit_backing->filename,
3163                             explicit_backing->drv->format_name,
3164                             NULL, size, flags, false, &local_err);
3165             break;
3166         default:
3167             abort();
3168         }
3169     }
3170 
3171     if (local_err) {
3172         error_propagate(errp, local_err);
3173         goto out;
3174     }
3175 
3176     options = qdict_new();
3177     if (arg->has_node_name) {
3178         qdict_put_str(options, "node-name", arg->node_name);
3179     }
3180     if (format) {
3181         qdict_put_str(options, "driver", format);
3182     }
3183 
3184     /* Mirroring takes care of copy-on-write using the source's backing
3185      * file.
3186      */
3187     target_bs = bdrv_open(arg->target, NULL, options, flags, errp);
3188     if (!target_bs) {
3189         goto out;
3190     }
3191 
3192     zero_target = (arg->sync == MIRROR_SYNC_MODE_FULL &&
3193                    (arg->mode == NEW_IMAGE_MODE_EXISTING ||
3194                     !bdrv_has_zero_init(target_bs)));
3195 
3196 
3197     /* Honor bdrv_try_set_aio_context() context acquisition requirements. */
3198     old_context = bdrv_get_aio_context(target_bs);
3199     aio_context_release(aio_context);
3200     aio_context_acquire(old_context);
3201 
3202     ret = bdrv_try_set_aio_context(target_bs, aio_context, errp);
3203     if (ret < 0) {
3204         bdrv_unref(target_bs);
3205         aio_context_release(old_context);
3206         return;
3207     }
3208 
3209     aio_context_release(old_context);
3210     aio_context_acquire(aio_context);
3211 
3212     blockdev_mirror_common(arg->has_job_id ? arg->job_id : NULL, bs, target_bs,
3213                            arg->has_replaces, arg->replaces, arg->sync,
3214                            backing_mode, zero_target,
3215                            arg->has_speed, arg->speed,
3216                            arg->has_granularity, arg->granularity,
3217                            arg->has_buf_size, arg->buf_size,
3218                            arg->has_on_source_error, arg->on_source_error,
3219                            arg->has_on_target_error, arg->on_target_error,
3220                            arg->has_unmap, arg->unmap,
3221                            false, NULL,
3222                            arg->has_copy_mode, arg->copy_mode,
3223                            arg->has_auto_finalize, arg->auto_finalize,
3224                            arg->has_auto_dismiss, arg->auto_dismiss,
3225                            errp);
3226     bdrv_unref(target_bs);
3227 out:
3228     aio_context_release(aio_context);
3229 }
3230 
3231 void qmp_blockdev_mirror(bool has_job_id, const char *job_id,
3232                          const char *device, const char *target,
3233                          bool has_replaces, const char *replaces,
3234                          MirrorSyncMode sync,
3235                          bool has_speed, int64_t speed,
3236                          bool has_granularity, uint32_t granularity,
3237                          bool has_buf_size, int64_t buf_size,
3238                          bool has_on_source_error,
3239                          BlockdevOnError on_source_error,
3240                          bool has_on_target_error,
3241                          BlockdevOnError on_target_error,
3242                          bool has_filter_node_name,
3243                          const char *filter_node_name,
3244                          bool has_copy_mode, MirrorCopyMode copy_mode,
3245                          bool has_auto_finalize, bool auto_finalize,
3246                          bool has_auto_dismiss, bool auto_dismiss,
3247                          Error **errp)
3248 {
3249     BlockDriverState *bs;
3250     BlockDriverState *target_bs;
3251     AioContext *aio_context;
3252     AioContext *old_context;
3253     BlockMirrorBackingMode backing_mode = MIRROR_LEAVE_BACKING_CHAIN;
3254     bool zero_target;
3255     int ret;
3256 
3257     bs = qmp_get_root_bs(device, errp);
3258     if (!bs) {
3259         return;
3260     }
3261 
3262     target_bs = bdrv_lookup_bs(target, target, errp);
3263     if (!target_bs) {
3264         return;
3265     }
3266 
3267     zero_target = (sync == MIRROR_SYNC_MODE_FULL);
3268 
3269     /* Honor bdrv_try_set_aio_context() context acquisition requirements. */
3270     old_context = bdrv_get_aio_context(target_bs);
3271     aio_context = bdrv_get_aio_context(bs);
3272     aio_context_acquire(old_context);
3273 
3274     ret = bdrv_try_set_aio_context(target_bs, aio_context, errp);
3275 
3276     aio_context_release(old_context);
3277     aio_context_acquire(aio_context);
3278 
3279     if (ret < 0) {
3280         goto out;
3281     }
3282 
3283     blockdev_mirror_common(has_job_id ? job_id : NULL, bs, target_bs,
3284                            has_replaces, replaces, sync, backing_mode,
3285                            zero_target, has_speed, speed,
3286                            has_granularity, granularity,
3287                            has_buf_size, buf_size,
3288                            has_on_source_error, on_source_error,
3289                            has_on_target_error, on_target_error,
3290                            true, true,
3291                            has_filter_node_name, filter_node_name,
3292                            has_copy_mode, copy_mode,
3293                            has_auto_finalize, auto_finalize,
3294                            has_auto_dismiss, auto_dismiss,
3295                            errp);
3296 out:
3297     aio_context_release(aio_context);
3298 }
3299 
3300 /* Get a block job using its ID and acquire its AioContext */
3301 static BlockJob *find_block_job(const char *id, AioContext **aio_context,
3302                                 Error **errp)
3303 {
3304     BlockJob *job;
3305 
3306     assert(id != NULL);
3307 
3308     *aio_context = NULL;
3309 
3310     job = block_job_get(id);
3311 
3312     if (!job) {
3313         error_set(errp, ERROR_CLASS_DEVICE_NOT_ACTIVE,
3314                   "Block job '%s' not found", id);
3315         return NULL;
3316     }
3317 
3318     *aio_context = blk_get_aio_context(job->blk);
3319     aio_context_acquire(*aio_context);
3320 
3321     return job;
3322 }
3323 
3324 void qmp_block_job_set_speed(const char *device, int64_t speed, Error **errp)
3325 {
3326     AioContext *aio_context;
3327     BlockJob *job = find_block_job(device, &aio_context, errp);
3328 
3329     if (!job) {
3330         return;
3331     }
3332 
3333     block_job_set_speed(job, speed, errp);
3334     aio_context_release(aio_context);
3335 }
3336 
3337 void qmp_block_job_cancel(const char *device,
3338                           bool has_force, bool force, Error **errp)
3339 {
3340     AioContext *aio_context;
3341     BlockJob *job = find_block_job(device, &aio_context, errp);
3342 
3343     if (!job) {
3344         return;
3345     }
3346 
3347     if (!has_force) {
3348         force = false;
3349     }
3350 
3351     if (job_user_paused(&job->job) && !force) {
3352         error_setg(errp, "The block job for device '%s' is currently paused",
3353                    device);
3354         goto out;
3355     }
3356 
3357     trace_qmp_block_job_cancel(job);
3358     job_user_cancel(&job->job, force, errp);
3359 out:
3360     aio_context_release(aio_context);
3361 }
3362 
3363 void qmp_block_job_pause(const char *device, Error **errp)
3364 {
3365     AioContext *aio_context;
3366     BlockJob *job = find_block_job(device, &aio_context, errp);
3367 
3368     if (!job) {
3369         return;
3370     }
3371 
3372     trace_qmp_block_job_pause(job);
3373     job_user_pause(&job->job, errp);
3374     aio_context_release(aio_context);
3375 }
3376 
3377 void qmp_block_job_resume(const char *device, Error **errp)
3378 {
3379     AioContext *aio_context;
3380     BlockJob *job = find_block_job(device, &aio_context, errp);
3381 
3382     if (!job) {
3383         return;
3384     }
3385 
3386     trace_qmp_block_job_resume(job);
3387     job_user_resume(&job->job, errp);
3388     aio_context_release(aio_context);
3389 }
3390 
3391 void qmp_block_job_complete(const char *device, Error **errp)
3392 {
3393     AioContext *aio_context;
3394     BlockJob *job = find_block_job(device, &aio_context, errp);
3395 
3396     if (!job) {
3397         return;
3398     }
3399 
3400     trace_qmp_block_job_complete(job);
3401     job_complete(&job->job, errp);
3402     aio_context_release(aio_context);
3403 }
3404 
3405 void qmp_block_job_finalize(const char *id, Error **errp)
3406 {
3407     AioContext *aio_context;
3408     BlockJob *job = find_block_job(id, &aio_context, errp);
3409 
3410     if (!job) {
3411         return;
3412     }
3413 
3414     trace_qmp_block_job_finalize(job);
3415     job_ref(&job->job);
3416     job_finalize(&job->job, errp);
3417 
3418     /*
3419      * Job's context might have changed via job_finalize (and job_txn_apply
3420      * automatically acquires the new one), so make sure we release the correct
3421      * one.
3422      */
3423     aio_context = blk_get_aio_context(job->blk);
3424     job_unref(&job->job);
3425     aio_context_release(aio_context);
3426 }
3427 
3428 void qmp_block_job_dismiss(const char *id, Error **errp)
3429 {
3430     AioContext *aio_context;
3431     BlockJob *bjob = find_block_job(id, &aio_context, errp);
3432     Job *job;
3433 
3434     if (!bjob) {
3435         return;
3436     }
3437 
3438     trace_qmp_block_job_dismiss(bjob);
3439     job = &bjob->job;
3440     job_dismiss(&job, errp);
3441     aio_context_release(aio_context);
3442 }
3443 
3444 void qmp_change_backing_file(const char *device,
3445                              const char *image_node_name,
3446                              const char *backing_file,
3447                              Error **errp)
3448 {
3449     BlockDriverState *bs = NULL;
3450     AioContext *aio_context;
3451     BlockDriverState *image_bs = NULL;
3452     Error *local_err = NULL;
3453     bool ro;
3454     int ret;
3455 
3456     bs = qmp_get_root_bs(device, errp);
3457     if (!bs) {
3458         return;
3459     }
3460 
3461     aio_context = bdrv_get_aio_context(bs);
3462     aio_context_acquire(aio_context);
3463 
3464     image_bs = bdrv_lookup_bs(NULL, image_node_name, &local_err);
3465     if (local_err) {
3466         error_propagate(errp, local_err);
3467         goto out;
3468     }
3469 
3470     if (!image_bs) {
3471         error_setg(errp, "image file not found");
3472         goto out;
3473     }
3474 
3475     if (bdrv_find_base(image_bs) == image_bs) {
3476         error_setg(errp, "not allowing backing file change on an image "
3477                          "without a backing file");
3478         goto out;
3479     }
3480 
3481     /* even though we are not necessarily operating on bs, we need it to
3482      * determine if block ops are currently prohibited on the chain */
3483     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_CHANGE, errp)) {
3484         goto out;
3485     }
3486 
3487     /* final sanity check */
3488     if (!bdrv_chain_contains(bs, image_bs)) {
3489         error_setg(errp, "'%s' and image file are not in the same chain",
3490                    device);
3491         goto out;
3492     }
3493 
3494     /* if not r/w, reopen to make r/w */
3495     ro = bdrv_is_read_only(image_bs);
3496 
3497     if (ro) {
3498         if (bdrv_reopen_set_read_only(image_bs, false, errp) != 0) {
3499             goto out;
3500         }
3501     }
3502 
3503     ret = bdrv_change_backing_file(image_bs, backing_file,
3504                                    image_bs->drv ? image_bs->drv->format_name : "",
3505                                    false);
3506 
3507     if (ret < 0) {
3508         error_setg_errno(errp, -ret, "Could not change backing file to '%s'",
3509                          backing_file);
3510         /* don't exit here, so we can try to restore open flags if
3511          * appropriate */
3512     }
3513 
3514     if (ro) {
3515         bdrv_reopen_set_read_only(image_bs, true, errp);
3516     }
3517 
3518 out:
3519     aio_context_release(aio_context);
3520 }
3521 
3522 void qmp_blockdev_add(BlockdevOptions *options, Error **errp)
3523 {
3524     BlockDriverState *bs;
3525     QObject *obj;
3526     Visitor *v = qobject_output_visitor_new(&obj);
3527     QDict *qdict;
3528 
3529     visit_type_BlockdevOptions(v, NULL, &options, &error_abort);
3530     visit_complete(v, &obj);
3531     qdict = qobject_to(QDict, obj);
3532 
3533     qdict_flatten(qdict);
3534 
3535     if (!qdict_get_try_str(qdict, "node-name")) {
3536         error_setg(errp, "'node-name' must be specified for the root node");
3537         goto fail;
3538     }
3539 
3540     bs = bds_tree_init(qdict, errp);
3541     if (!bs) {
3542         goto fail;
3543     }
3544 
3545     bdrv_set_monitor_owned(bs);
3546 
3547 fail:
3548     visit_free(v);
3549 }
3550 
3551 void qmp_blockdev_reopen(BlockdevOptionsList *reopen_list, Error **errp)
3552 {
3553     BlockReopenQueue *queue = NULL;
3554     GSList *drained = NULL;
3555 
3556     /* Add each one of the BDS that we want to reopen to the queue */
3557     for (; reopen_list != NULL; reopen_list = reopen_list->next) {
3558         BlockdevOptions *options = reopen_list->value;
3559         BlockDriverState *bs;
3560         AioContext *ctx;
3561         QObject *obj;
3562         Visitor *v;
3563         QDict *qdict;
3564 
3565         /* Check for the selected node name */
3566         if (!options->has_node_name) {
3567             error_setg(errp, "node-name not specified");
3568             goto fail;
3569         }
3570 
3571         bs = bdrv_find_node(options->node_name);
3572         if (!bs) {
3573             error_setg(errp, "Failed to find node with node-name='%s'",
3574                        options->node_name);
3575             goto fail;
3576         }
3577 
3578         /* Put all options in a QDict and flatten it */
3579         v = qobject_output_visitor_new(&obj);
3580         visit_type_BlockdevOptions(v, NULL, &options, &error_abort);
3581         visit_complete(v, &obj);
3582         visit_free(v);
3583 
3584         qdict = qobject_to(QDict, obj);
3585 
3586         qdict_flatten(qdict);
3587 
3588         ctx = bdrv_get_aio_context(bs);
3589         aio_context_acquire(ctx);
3590 
3591         bdrv_subtree_drained_begin(bs);
3592         queue = bdrv_reopen_queue(queue, bs, qdict, false);
3593         drained = g_slist_prepend(drained, bs);
3594 
3595         aio_context_release(ctx);
3596     }
3597 
3598     /* Perform the reopen operation */
3599     bdrv_reopen_multiple(queue, errp);
3600     queue = NULL;
3601 
3602 fail:
3603     bdrv_reopen_queue_free(queue);
3604     g_slist_free_full(drained, (GDestroyNotify) bdrv_subtree_drained_end);
3605 }
3606 
3607 void qmp_blockdev_del(const char *node_name, Error **errp)
3608 {
3609     AioContext *aio_context;
3610     BlockDriverState *bs;
3611 
3612     bs = bdrv_find_node(node_name);
3613     if (!bs) {
3614         error_setg(errp, "Failed to find node with node-name='%s'", node_name);
3615         return;
3616     }
3617     if (bdrv_has_blk(bs)) {
3618         error_setg(errp, "Node %s is in use", node_name);
3619         return;
3620     }
3621     aio_context = bdrv_get_aio_context(bs);
3622     aio_context_acquire(aio_context);
3623 
3624     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_DRIVE_DEL, errp)) {
3625         goto out;
3626     }
3627 
3628     if (!QTAILQ_IN_USE(bs, monitor_list)) {
3629         error_setg(errp, "Node %s is not owned by the monitor",
3630                    bs->node_name);
3631         goto out;
3632     }
3633 
3634     if (bs->refcnt > 1) {
3635         error_setg(errp, "Block device %s is in use",
3636                    bdrv_get_device_or_node_name(bs));
3637         goto out;
3638     }
3639 
3640     QTAILQ_REMOVE(&monitor_bdrv_states, bs, monitor_list);
3641     bdrv_unref(bs);
3642 
3643 out:
3644     aio_context_release(aio_context);
3645 }
3646 
3647 static BdrvChild *bdrv_find_child(BlockDriverState *parent_bs,
3648                                   const char *child_name)
3649 {
3650     BdrvChild *child;
3651 
3652     QLIST_FOREACH(child, &parent_bs->children, next) {
3653         if (strcmp(child->name, child_name) == 0) {
3654             return child;
3655         }
3656     }
3657 
3658     return NULL;
3659 }
3660 
3661 void qmp_x_blockdev_change(const char *parent, bool has_child,
3662                            const char *child, bool has_node,
3663                            const char *node, Error **errp)
3664 {
3665     BlockDriverState *parent_bs, *new_bs = NULL;
3666     BdrvChild *p_child;
3667 
3668     parent_bs = bdrv_lookup_bs(parent, parent, errp);
3669     if (!parent_bs) {
3670         return;
3671     }
3672 
3673     if (has_child == has_node) {
3674         if (has_child) {
3675             error_setg(errp, "The parameters child and node are in conflict");
3676         } else {
3677             error_setg(errp, "Either child or node must be specified");
3678         }
3679         return;
3680     }
3681 
3682     if (has_child) {
3683         p_child = bdrv_find_child(parent_bs, child);
3684         if (!p_child) {
3685             error_setg(errp, "Node '%s' does not have child '%s'",
3686                        parent, child);
3687             return;
3688         }
3689         bdrv_del_child(parent_bs, p_child, errp);
3690     }
3691 
3692     if (has_node) {
3693         new_bs = bdrv_find_node(node);
3694         if (!new_bs) {
3695             error_setg(errp, "Node '%s' not found", node);
3696             return;
3697         }
3698         bdrv_add_child(parent_bs, new_bs, errp);
3699     }
3700 }
3701 
3702 BlockJobInfoList *qmp_query_block_jobs(Error **errp)
3703 {
3704     BlockJobInfoList *head = NULL, **tail = &head;
3705     BlockJob *job;
3706 
3707     for (job = block_job_next(NULL); job; job = block_job_next(job)) {
3708         BlockJobInfo *value;
3709         AioContext *aio_context;
3710 
3711         if (block_job_is_internal(job)) {
3712             continue;
3713         }
3714         aio_context = blk_get_aio_context(job->blk);
3715         aio_context_acquire(aio_context);
3716         value = block_job_query(job, errp);
3717         aio_context_release(aio_context);
3718         if (!value) {
3719             qapi_free_BlockJobInfoList(head);
3720             return NULL;
3721         }
3722         QAPI_LIST_APPEND(tail, value);
3723     }
3724 
3725     return head;
3726 }
3727 
3728 void qmp_x_blockdev_set_iothread(const char *node_name, StrOrNull *iothread,
3729                                  bool has_force, bool force, Error **errp)
3730 {
3731     AioContext *old_context;
3732     AioContext *new_context;
3733     BlockDriverState *bs;
3734 
3735     bs = bdrv_find_node(node_name);
3736     if (!bs) {
3737         error_setg(errp, "Failed to find node with node-name='%s'", node_name);
3738         return;
3739     }
3740 
3741     /* Protects against accidents. */
3742     if (!(has_force && force) && bdrv_has_blk(bs)) {
3743         error_setg(errp, "Node %s is associated with a BlockBackend and could "
3744                          "be in use (use force=true to override this check)",
3745                          node_name);
3746         return;
3747     }
3748 
3749     if (iothread->type == QTYPE_QSTRING) {
3750         IOThread *obj = iothread_by_id(iothread->u.s);
3751         if (!obj) {
3752             error_setg(errp, "Cannot find iothread %s", iothread->u.s);
3753             return;
3754         }
3755 
3756         new_context = iothread_get_aio_context(obj);
3757     } else {
3758         new_context = qemu_get_aio_context();
3759     }
3760 
3761     old_context = bdrv_get_aio_context(bs);
3762     aio_context_acquire(old_context);
3763 
3764     bdrv_try_set_aio_context(bs, new_context, errp);
3765 
3766     aio_context_release(old_context);
3767 }
3768 
3769 QemuOptsList qemu_common_drive_opts = {
3770     .name = "drive",
3771     .head = QTAILQ_HEAD_INITIALIZER(qemu_common_drive_opts.head),
3772     .desc = {
3773         {
3774             .name = "snapshot",
3775             .type = QEMU_OPT_BOOL,
3776             .help = "enable/disable snapshot mode",
3777         },{
3778             .name = "aio",
3779             .type = QEMU_OPT_STRING,
3780             .help = "host AIO implementation (threads, native, io_uring)",
3781         },{
3782             .name = BDRV_OPT_CACHE_WB,
3783             .type = QEMU_OPT_BOOL,
3784             .help = "Enable writeback mode",
3785         },{
3786             .name = "format",
3787             .type = QEMU_OPT_STRING,
3788             .help = "disk format (raw, qcow2, ...)",
3789         },{
3790             .name = "rerror",
3791             .type = QEMU_OPT_STRING,
3792             .help = "read error action",
3793         },{
3794             .name = "werror",
3795             .type = QEMU_OPT_STRING,
3796             .help = "write error action",
3797         },{
3798             .name = BDRV_OPT_READ_ONLY,
3799             .type = QEMU_OPT_BOOL,
3800             .help = "open drive file as read-only",
3801         },
3802 
3803         THROTTLE_OPTS,
3804 
3805         {
3806             .name = "throttling.group",
3807             .type = QEMU_OPT_STRING,
3808             .help = "name of the block throttling group",
3809         },{
3810             .name = "copy-on-read",
3811             .type = QEMU_OPT_BOOL,
3812             .help = "copy read data from backing file into image file",
3813         },{
3814             .name = "detect-zeroes",
3815             .type = QEMU_OPT_STRING,
3816             .help = "try to optimize zero writes (off, on, unmap)",
3817         },{
3818             .name = "stats-account-invalid",
3819             .type = QEMU_OPT_BOOL,
3820             .help = "whether to account for invalid I/O operations "
3821                     "in the statistics",
3822         },{
3823             .name = "stats-account-failed",
3824             .type = QEMU_OPT_BOOL,
3825             .help = "whether to account for failed I/O operations "
3826                     "in the statistics",
3827         },
3828         { /* end of list */ }
3829     },
3830 };
3831 
3832 QemuOptsList qemu_drive_opts = {
3833     .name = "drive",
3834     .head = QTAILQ_HEAD_INITIALIZER(qemu_drive_opts.head),
3835     .desc = {
3836         /*
3837          * no elements => accept any params
3838          * validation will happen later
3839          */
3840         { /* end of list */ }
3841     },
3842 };
3843