xref: /openbmc/qemu/blockdev.c (revision 2a53cff4)
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/config-file.h"
44 #include "qapi/qapi-commands-block.h"
45 #include "qapi/qapi-commands-transaction.h"
46 #include "qapi/qapi-visit-block-core.h"
47 #include "qapi/qmp/qdict.h"
48 #include "qapi/qmp/qnum.h"
49 #include "qapi/qmp/qstring.h"
50 #include "qapi/error.h"
51 #include "qapi/qmp/qerror.h"
52 #include "qapi/qmp/qlist.h"
53 #include "qapi/qobject-output-visitor.h"
54 #include "sysemu/sysemu.h"
55 #include "sysemu/iothread.h"
56 #include "block/block_int.h"
57 #include "block/trace.h"
58 #include "sysemu/arch_init.h"
59 #include "sysemu/qtest.h"
60 #include "qemu/cutils.h"
61 #include "qemu/help_option.h"
62 #include "qemu/throttle-options.h"
63 
64 static QTAILQ_HEAD(, BlockDriverState) monitor_bdrv_states =
65     QTAILQ_HEAD_INITIALIZER(monitor_bdrv_states);
66 
67 static int do_open_tray(const char *blk_name, const char *qdev_id,
68                         bool force, Error **errp);
69 static void blockdev_remove_medium(bool has_device, const char *device,
70                                    bool has_id, const char *id, Error **errp);
71 static void blockdev_insert_medium(bool has_device, const char *device,
72                                    bool has_id, const char *id,
73                                    const char *node_name, Error **errp);
74 
75 static const char *const if_name[IF_COUNT] = {
76     [IF_NONE] = "none",
77     [IF_IDE] = "ide",
78     [IF_SCSI] = "scsi",
79     [IF_FLOPPY] = "floppy",
80     [IF_PFLASH] = "pflash",
81     [IF_MTD] = "mtd",
82     [IF_SD] = "sd",
83     [IF_VIRTIO] = "virtio",
84     [IF_XEN] = "xen",
85 };
86 
87 static int if_max_devs[IF_COUNT] = {
88     /*
89      * Do not change these numbers!  They govern how drive option
90      * index maps to unit and bus.  That mapping is ABI.
91      *
92      * All controllers used to implement if=T drives need to support
93      * if_max_devs[T] units, for any T with if_max_devs[T] != 0.
94      * Otherwise, some index values map to "impossible" bus, unit
95      * values.
96      *
97      * For instance, if you change [IF_SCSI] to 255, -drive
98      * if=scsi,index=12 no longer means bus=1,unit=5, but
99      * bus=0,unit=12.  With an lsi53c895a controller (7 units max),
100      * the drive can't be set up.  Regression.
101      */
102     [IF_IDE] = 2,
103     [IF_SCSI] = 7,
104 };
105 
106 /**
107  * Boards may call this to offer board-by-board overrides
108  * of the default, global values.
109  */
110 void override_max_devs(BlockInterfaceType type, int max_devs)
111 {
112     BlockBackend *blk;
113     DriveInfo *dinfo;
114 
115     if (max_devs <= 0) {
116         return;
117     }
118 
119     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
120         dinfo = blk_legacy_dinfo(blk);
121         if (dinfo->type == type) {
122             fprintf(stderr, "Cannot override units-per-bus property of"
123                     " the %s interface, because a drive of that type has"
124                     " already been added.\n", if_name[type]);
125             g_assert_not_reached();
126         }
127     }
128 
129     if_max_devs[type] = max_devs;
130 }
131 
132 /*
133  * We automatically delete the drive when a device using it gets
134  * unplugged.  Questionable feature, but we can't just drop it.
135  * Device models call blockdev_mark_auto_del() to schedule the
136  * automatic deletion, and generic qdev code calls blockdev_auto_del()
137  * when deletion is actually safe.
138  */
139 void blockdev_mark_auto_del(BlockBackend *blk)
140 {
141     DriveInfo *dinfo = blk_legacy_dinfo(blk);
142     BlockDriverState *bs = blk_bs(blk);
143     AioContext *aio_context;
144 
145     if (!dinfo) {
146         return;
147     }
148 
149     if (bs) {
150         aio_context = bdrv_get_aio_context(bs);
151         aio_context_acquire(aio_context);
152 
153         if (bs->job) {
154             job_cancel(&bs->job->job, false);
155         }
156 
157         aio_context_release(aio_context);
158     }
159 
160     dinfo->auto_del = 1;
161 }
162 
163 void blockdev_auto_del(BlockBackend *blk)
164 {
165     DriveInfo *dinfo = blk_legacy_dinfo(blk);
166 
167     if (dinfo && dinfo->auto_del) {
168         monitor_remove_blk(blk);
169         blk_unref(blk);
170     }
171 }
172 
173 /**
174  * Returns the current mapping of how many units per bus
175  * a particular interface can support.
176  *
177  *  A positive integer indicates n units per bus.
178  *  0 implies the mapping has not been established.
179  * -1 indicates an invalid BlockInterfaceType was given.
180  */
181 int drive_get_max_devs(BlockInterfaceType type)
182 {
183     if (type >= IF_IDE && type < IF_COUNT) {
184         return if_max_devs[type];
185     }
186 
187     return -1;
188 }
189 
190 static int drive_index_to_bus_id(BlockInterfaceType type, int index)
191 {
192     int max_devs = if_max_devs[type];
193     return max_devs ? index / max_devs : 0;
194 }
195 
196 static int drive_index_to_unit_id(BlockInterfaceType type, int index)
197 {
198     int max_devs = if_max_devs[type];
199     return max_devs ? index % max_devs : index;
200 }
201 
202 QemuOpts *drive_def(const char *optstr)
203 {
204     return qemu_opts_parse_noisily(qemu_find_opts("drive"), optstr, false);
205 }
206 
207 QemuOpts *drive_add(BlockInterfaceType type, int index, const char *file,
208                     const char *optstr)
209 {
210     QemuOpts *opts;
211 
212     opts = drive_def(optstr);
213     if (!opts) {
214         return NULL;
215     }
216     if (type != IF_DEFAULT) {
217         qemu_opt_set(opts, "if", if_name[type], &error_abort);
218     }
219     if (index >= 0) {
220         qemu_opt_set_number(opts, "index", index, &error_abort);
221     }
222     if (file)
223         qemu_opt_set(opts, "file", file, &error_abort);
224     return opts;
225 }
226 
227 DriveInfo *drive_get(BlockInterfaceType type, int bus, int unit)
228 {
229     BlockBackend *blk;
230     DriveInfo *dinfo;
231 
232     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
233         dinfo = blk_legacy_dinfo(blk);
234         if (dinfo && dinfo->type == type
235             && dinfo->bus == bus && dinfo->unit == unit) {
236             return dinfo;
237         }
238     }
239 
240     return NULL;
241 }
242 
243 void drive_check_orphaned(void)
244 {
245     BlockBackend *blk;
246     DriveInfo *dinfo;
247     Location loc;
248     bool orphans = false;
249 
250     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
251         dinfo = blk_legacy_dinfo(blk);
252         if (!blk_get_attached_dev(blk) && !dinfo->is_default &&
253             dinfo->type != IF_NONE) {
254             loc_push_none(&loc);
255             qemu_opts_loc_restore(dinfo->opts);
256             error_report("machine type does not support"
257                          " if=%s,bus=%d,unit=%d",
258                          if_name[dinfo->type], dinfo->bus, dinfo->unit);
259             loc_pop(&loc);
260             orphans = true;
261         }
262     }
263 
264     if (orphans) {
265         exit(1);
266     }
267 }
268 
269 DriveInfo *drive_get_by_index(BlockInterfaceType type, int index)
270 {
271     return drive_get(type,
272                      drive_index_to_bus_id(type, index),
273                      drive_index_to_unit_id(type, index));
274 }
275 
276 int drive_get_max_bus(BlockInterfaceType type)
277 {
278     int max_bus;
279     BlockBackend *blk;
280     DriveInfo *dinfo;
281 
282     max_bus = -1;
283     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
284         dinfo = blk_legacy_dinfo(blk);
285         if (dinfo && dinfo->type == type && dinfo->bus > max_bus) {
286             max_bus = dinfo->bus;
287         }
288     }
289     return max_bus;
290 }
291 
292 /* Get a block device.  This should only be used for single-drive devices
293    (e.g. SD/Floppy/MTD).  Multi-disk devices (scsi/ide) should use the
294    appropriate bus.  */
295 DriveInfo *drive_get_next(BlockInterfaceType type)
296 {
297     static int next_block_unit[IF_COUNT];
298 
299     return drive_get(type, 0, next_block_unit[type]++);
300 }
301 
302 static void bdrv_format_print(void *opaque, const char *name)
303 {
304     error_printf(" %s", name);
305 }
306 
307 typedef struct {
308     QEMUBH *bh;
309     BlockDriverState *bs;
310 } BDRVPutRefBH;
311 
312 static int parse_block_error_action(const char *buf, bool is_read, Error **errp)
313 {
314     if (!strcmp(buf, "ignore")) {
315         return BLOCKDEV_ON_ERROR_IGNORE;
316     } else if (!is_read && !strcmp(buf, "enospc")) {
317         return BLOCKDEV_ON_ERROR_ENOSPC;
318     } else if (!strcmp(buf, "stop")) {
319         return BLOCKDEV_ON_ERROR_STOP;
320     } else if (!strcmp(buf, "report")) {
321         return BLOCKDEV_ON_ERROR_REPORT;
322     } else {
323         error_setg(errp, "'%s' invalid %s error action",
324                    buf, is_read ? "read" : "write");
325         return -1;
326     }
327 }
328 
329 static bool parse_stats_intervals(BlockAcctStats *stats, QList *intervals,
330                                   Error **errp)
331 {
332     const QListEntry *entry;
333     for (entry = qlist_first(intervals); entry; entry = qlist_next(entry)) {
334         switch (qobject_type(entry->value)) {
335 
336         case QTYPE_QSTRING: {
337             unsigned long long length;
338             const char *str = qstring_get_str(qobject_to(QString,
339                                                          entry->value));
340             if (parse_uint_full(str, &length, 10) == 0 &&
341                 length > 0 && length <= UINT_MAX) {
342                 block_acct_add_interval(stats, (unsigned) length);
343             } else {
344                 error_setg(errp, "Invalid interval length: %s", str);
345                 return false;
346             }
347             break;
348         }
349 
350         case QTYPE_QNUM: {
351             int64_t length = qnum_get_int(qobject_to(QNum, entry->value));
352 
353             if (length > 0 && length <= UINT_MAX) {
354                 block_acct_add_interval(stats, (unsigned) length);
355             } else {
356                 error_setg(errp, "Invalid interval length: %" PRId64, length);
357                 return false;
358             }
359             break;
360         }
361 
362         default:
363             error_setg(errp, "The specification of stats-intervals is invalid");
364             return false;
365         }
366     }
367     return true;
368 }
369 
370 typedef enum { MEDIA_DISK, MEDIA_CDROM } DriveMediaType;
371 
372 /* All parameters but @opts are optional and may be set to NULL. */
373 static void extract_common_blockdev_options(QemuOpts *opts, int *bdrv_flags,
374     const char **throttling_group, ThrottleConfig *throttle_cfg,
375     BlockdevDetectZeroesOptions *detect_zeroes, Error **errp)
376 {
377     Error *local_error = NULL;
378     const char *aio;
379 
380     if (bdrv_flags) {
381         if (qemu_opt_get_bool(opts, "copy-on-read", false)) {
382             *bdrv_flags |= BDRV_O_COPY_ON_READ;
383         }
384 
385         if ((aio = qemu_opt_get(opts, "aio")) != NULL) {
386             if (!strcmp(aio, "native")) {
387                 *bdrv_flags |= BDRV_O_NATIVE_AIO;
388             } else if (!strcmp(aio, "threads")) {
389                 /* this is the default */
390             } else {
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, &error);
490     if (error) {
491         error_propagate(errp, error);
492         goto err_no_opts;
493     }
494 
495     qemu_opts_absorb_qdict(opts, bs_opts, &error);
496     if (error) {
497         error_propagate(errp, error);
498         goto early_err;
499     }
500 
501     if (id) {
502         qdict_del(bs_opts, "id");
503     }
504 
505     /* extract parameters */
506     snapshot = qemu_opt_get_bool(opts, "snapshot", 0);
507 
508     account_invalid = qemu_opt_get_bool(opts, "stats-account-invalid", true);
509     account_failed = qemu_opt_get_bool(opts, "stats-account-failed", true);
510 
511     writethrough = !qemu_opt_get_bool(opts, BDRV_OPT_CACHE_WB, true);
512 
513     id = qemu_opts_id(opts);
514 
515     qdict_extract_subqdict(bs_opts, &interval_dict, "stats-intervals.");
516     qdict_array_split(interval_dict, &interval_list);
517 
518     if (qdict_size(interval_dict) != 0) {
519         error_setg(errp, "Invalid option stats-intervals.%s",
520                    qdict_first(interval_dict)->key);
521         goto early_err;
522     }
523 
524     extract_common_blockdev_options(opts, &bdrv_flags, &throttling_group, &cfg,
525                                     &detect_zeroes, &error);
526     if (error) {
527         error_propagate(errp, error);
528         goto early_err;
529     }
530 
531     if ((buf = qemu_opt_get(opts, "format")) != NULL) {
532         if (is_help_option(buf)) {
533             error_printf("Supported formats:");
534             bdrv_iterate_format(bdrv_format_print, NULL, false);
535             error_printf("\nSupported formats (read-only):");
536             bdrv_iterate_format(bdrv_format_print, NULL, true);
537             error_printf("\n");
538             goto early_err;
539         }
540 
541         if (qdict_haskey(bs_opts, "driver")) {
542             error_setg(errp, "Cannot specify both 'driver' and 'format'");
543             goto early_err;
544         }
545         qdict_put_str(bs_opts, "driver", buf);
546     }
547 
548     on_write_error = BLOCKDEV_ON_ERROR_ENOSPC;
549     if ((buf = qemu_opt_get(opts, "werror")) != NULL) {
550         on_write_error = parse_block_error_action(buf, 0, &error);
551         if (error) {
552             error_propagate(errp, error);
553             goto early_err;
554         }
555     }
556 
557     on_read_error = BLOCKDEV_ON_ERROR_REPORT;
558     if ((buf = qemu_opt_get(opts, "rerror")) != NULL) {
559         on_read_error = parse_block_error_action(buf, 1, &error);
560         if (error) {
561             error_propagate(errp, error);
562             goto early_err;
563         }
564     }
565 
566     if (snapshot) {
567         bdrv_flags |= BDRV_O_SNAPSHOT;
568     }
569 
570     read_only = qemu_opt_get_bool(opts, BDRV_OPT_READ_ONLY, false);
571 
572     /* init */
573     if ((!file || !*file) && !qdict_size(bs_opts)) {
574         BlockBackendRootState *blk_rs;
575 
576         blk = blk_new(0, BLK_PERM_ALL);
577         blk_rs = blk_get_root_state(blk);
578         blk_rs->open_flags    = bdrv_flags;
579         blk_rs->read_only     = read_only;
580         blk_rs->detect_zeroes = detect_zeroes;
581 
582         qobject_unref(bs_opts);
583     } else {
584         if (file && !*file) {
585             file = NULL;
586         }
587 
588         /* bdrv_open() defaults to the values in bdrv_flags (for compatibility
589          * with other callers) rather than what we want as the real defaults.
590          * Apply the defaults here instead. */
591         qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_DIRECT, "off");
592         qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_NO_FLUSH, "off");
593         qdict_set_default_str(bs_opts, BDRV_OPT_READ_ONLY,
594                               read_only ? "on" : "off");
595         qdict_set_default_str(bs_opts, BDRV_OPT_AUTO_READ_ONLY, "on");
596         assert((bdrv_flags & BDRV_O_CACHE_MASK) == 0);
597 
598         if (runstate_check(RUN_STATE_INMIGRATE)) {
599             bdrv_flags |= BDRV_O_INACTIVE;
600         }
601 
602         blk = blk_new_open(file, NULL, bs_opts, bdrv_flags, errp);
603         if (!blk) {
604             goto err_no_bs_opts;
605         }
606         bs = blk_bs(blk);
607 
608         bs->detect_zeroes = detect_zeroes;
609 
610         block_acct_setup(blk_get_stats(blk), account_invalid, account_failed);
611 
612         if (!parse_stats_intervals(blk_get_stats(blk), interval_list, errp)) {
613             blk_unref(blk);
614             blk = NULL;
615             goto err_no_bs_opts;
616         }
617     }
618 
619     /* disk I/O throttling */
620     if (throttle_enabled(&cfg)) {
621         if (!throttling_group) {
622             throttling_group = id;
623         }
624         blk_io_limits_enable(blk, throttling_group);
625         blk_set_io_limits(blk, &cfg);
626     }
627 
628     blk_set_enable_write_cache(blk, !writethrough);
629     blk_set_on_error(blk, on_read_error, on_write_error);
630 
631     if (!monitor_add_blk(blk, id, errp)) {
632         blk_unref(blk);
633         blk = NULL;
634         goto err_no_bs_opts;
635     }
636 
637 err_no_bs_opts:
638     qemu_opts_del(opts);
639     qobject_unref(interval_dict);
640     qobject_unref(interval_list);
641     return blk;
642 
643 early_err:
644     qemu_opts_del(opts);
645     qobject_unref(interval_dict);
646     qobject_unref(interval_list);
647 err_no_opts:
648     qobject_unref(bs_opts);
649     return NULL;
650 }
651 
652 /* Takes the ownership of bs_opts */
653 static BlockDriverState *bds_tree_init(QDict *bs_opts, Error **errp)
654 {
655     int bdrv_flags = 0;
656 
657     /* bdrv_open() defaults to the values in bdrv_flags (for compatibility
658      * with other callers) rather than what we want as the real defaults.
659      * Apply the defaults here instead. */
660     qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_DIRECT, "off");
661     qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_NO_FLUSH, "off");
662     qdict_set_default_str(bs_opts, BDRV_OPT_READ_ONLY, "off");
663 
664     if (runstate_check(RUN_STATE_INMIGRATE)) {
665         bdrv_flags |= BDRV_O_INACTIVE;
666     }
667 
668     return bdrv_open(NULL, NULL, bs_opts, bdrv_flags, errp);
669 }
670 
671 void blockdev_close_all_bdrv_states(void)
672 {
673     BlockDriverState *bs, *next_bs;
674 
675     QTAILQ_FOREACH_SAFE(bs, &monitor_bdrv_states, monitor_list, next_bs) {
676         AioContext *ctx = bdrv_get_aio_context(bs);
677 
678         aio_context_acquire(ctx);
679         bdrv_unref(bs);
680         aio_context_release(ctx);
681     }
682 }
683 
684 /* Iterates over the list of monitor-owned BlockDriverStates */
685 BlockDriverState *bdrv_next_monitor_owned(BlockDriverState *bs)
686 {
687     return bs ? QTAILQ_NEXT(bs, monitor_list)
688               : QTAILQ_FIRST(&monitor_bdrv_states);
689 }
690 
691 static void qemu_opt_rename(QemuOpts *opts, const char *from, const char *to,
692                             Error **errp)
693 {
694     const char *value;
695 
696     value = qemu_opt_get(opts, from);
697     if (value) {
698         if (qemu_opt_find(opts, to)) {
699             error_setg(errp, "'%s' and its alias '%s' can't be used at the "
700                        "same time", to, from);
701             return;
702         }
703     }
704 
705     /* rename all items in opts */
706     while ((value = qemu_opt_get(opts, from))) {
707         qemu_opt_set(opts, to, value, &error_abort);
708         qemu_opt_unset(opts, from);
709     }
710 }
711 
712 QemuOptsList qemu_legacy_drive_opts = {
713     .name = "drive",
714     .head = QTAILQ_HEAD_INITIALIZER(qemu_legacy_drive_opts.head),
715     .desc = {
716         {
717             .name = "bus",
718             .type = QEMU_OPT_NUMBER,
719             .help = "bus number",
720         },{
721             .name = "unit",
722             .type = QEMU_OPT_NUMBER,
723             .help = "unit number (i.e. lun for scsi)",
724         },{
725             .name = "index",
726             .type = QEMU_OPT_NUMBER,
727             .help = "index number",
728         },{
729             .name = "media",
730             .type = QEMU_OPT_STRING,
731             .help = "media type (disk, cdrom)",
732         },{
733             .name = "if",
734             .type = QEMU_OPT_STRING,
735             .help = "interface (ide, scsi, sd, mtd, floppy, pflash, virtio)",
736         },{
737             .name = "file",
738             .type = QEMU_OPT_STRING,
739             .help = "file name",
740         },
741 
742         /* Options that are passed on, but have special semantics with -drive */
743         {
744             .name = BDRV_OPT_READ_ONLY,
745             .type = QEMU_OPT_BOOL,
746             .help = "open drive file as read-only",
747         },{
748             .name = "rerror",
749             .type = QEMU_OPT_STRING,
750             .help = "read error action",
751         },{
752             .name = "werror",
753             .type = QEMU_OPT_STRING,
754             .help = "write error action",
755         },{
756             .name = "copy-on-read",
757             .type = QEMU_OPT_BOOL,
758             .help = "copy read data from backing file into image file",
759         },
760 
761         { /* end of list */ }
762     },
763 };
764 
765 DriveInfo *drive_new(QemuOpts *all_opts, BlockInterfaceType block_default_type,
766                      Error **errp)
767 {
768     const char *value;
769     BlockBackend *blk;
770     DriveInfo *dinfo = NULL;
771     QDict *bs_opts;
772     QemuOpts *legacy_opts;
773     DriveMediaType media = MEDIA_DISK;
774     BlockInterfaceType type;
775     int max_devs, bus_id, unit_id, index;
776     const char *werror, *rerror;
777     bool read_only = false;
778     bool copy_on_read;
779     const char *filename;
780     Error *local_err = NULL;
781     int i;
782 
783     /* Change legacy command line options into QMP ones */
784     static const struct {
785         const char *from;
786         const char *to;
787     } opt_renames[] = {
788         { "iops",           "throttling.iops-total" },
789         { "iops_rd",        "throttling.iops-read" },
790         { "iops_wr",        "throttling.iops-write" },
791 
792         { "bps",            "throttling.bps-total" },
793         { "bps_rd",         "throttling.bps-read" },
794         { "bps_wr",         "throttling.bps-write" },
795 
796         { "iops_max",       "throttling.iops-total-max" },
797         { "iops_rd_max",    "throttling.iops-read-max" },
798         { "iops_wr_max",    "throttling.iops-write-max" },
799 
800         { "bps_max",        "throttling.bps-total-max" },
801         { "bps_rd_max",     "throttling.bps-read-max" },
802         { "bps_wr_max",     "throttling.bps-write-max" },
803 
804         { "iops_size",      "throttling.iops-size" },
805 
806         { "group",          "throttling.group" },
807 
808         { "readonly",       BDRV_OPT_READ_ONLY },
809     };
810 
811     for (i = 0; i < ARRAY_SIZE(opt_renames); i++) {
812         qemu_opt_rename(all_opts, opt_renames[i].from, opt_renames[i].to,
813                         &local_err);
814         if (local_err) {
815             error_propagate(errp, local_err);
816             return NULL;
817         }
818     }
819 
820     value = qemu_opt_get(all_opts, "cache");
821     if (value) {
822         int flags = 0;
823         bool writethrough;
824 
825         if (bdrv_parse_cache_mode(value, &flags, &writethrough) != 0) {
826             error_setg(errp, "invalid cache option");
827             return NULL;
828         }
829 
830         /* Specific options take precedence */
831         if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_WB)) {
832             qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_WB,
833                               !writethrough, &error_abort);
834         }
835         if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_DIRECT)) {
836             qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_DIRECT,
837                               !!(flags & BDRV_O_NOCACHE), &error_abort);
838         }
839         if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_NO_FLUSH)) {
840             qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_NO_FLUSH,
841                               !!(flags & BDRV_O_NO_FLUSH), &error_abort);
842         }
843         qemu_opt_unset(all_opts, "cache");
844     }
845 
846     /* Get a QDict for processing the options */
847     bs_opts = qdict_new();
848     qemu_opts_to_qdict(all_opts, bs_opts);
849 
850     legacy_opts = qemu_opts_create(&qemu_legacy_drive_opts, NULL, 0,
851                                    &error_abort);
852     qemu_opts_absorb_qdict(legacy_opts, bs_opts, &local_err);
853     if (local_err) {
854         error_propagate(errp, local_err);
855         goto fail;
856     }
857 
858     /* Media type */
859     value = qemu_opt_get(legacy_opts, "media");
860     if (value) {
861         if (!strcmp(value, "disk")) {
862             media = MEDIA_DISK;
863         } else if (!strcmp(value, "cdrom")) {
864             media = MEDIA_CDROM;
865             read_only = true;
866         } else {
867             error_setg(errp, "'%s' invalid media", value);
868             goto fail;
869         }
870     }
871 
872     /* copy-on-read is disabled with a warning for read-only devices */
873     read_only |= qemu_opt_get_bool(legacy_opts, BDRV_OPT_READ_ONLY, false);
874     copy_on_read = qemu_opt_get_bool(legacy_opts, "copy-on-read", false);
875 
876     if (read_only && copy_on_read) {
877         warn_report("disabling copy-on-read on read-only drive");
878         copy_on_read = false;
879     }
880 
881     qdict_put_str(bs_opts, BDRV_OPT_READ_ONLY, read_only ? "on" : "off");
882     qdict_put_str(bs_opts, "copy-on-read", copy_on_read ? "on" : "off");
883 
884     /* Controller type */
885     value = qemu_opt_get(legacy_opts, "if");
886     if (value) {
887         for (type = 0;
888              type < IF_COUNT && strcmp(value, if_name[type]);
889              type++) {
890         }
891         if (type == IF_COUNT) {
892             error_setg(errp, "unsupported bus type '%s'", value);
893             goto fail;
894         }
895     } else {
896         type = block_default_type;
897     }
898 
899     /* Device address specified by bus/unit or index.
900      * If none was specified, try to find the first free one. */
901     bus_id  = qemu_opt_get_number(legacy_opts, "bus", 0);
902     unit_id = qemu_opt_get_number(legacy_opts, "unit", -1);
903     index   = qemu_opt_get_number(legacy_opts, "index", -1);
904 
905     max_devs = if_max_devs[type];
906 
907     if (index != -1) {
908         if (bus_id != 0 || unit_id != -1) {
909             error_setg(errp, "index cannot be used with bus and unit");
910             goto fail;
911         }
912         bus_id = drive_index_to_bus_id(type, index);
913         unit_id = drive_index_to_unit_id(type, index);
914     }
915 
916     if (unit_id == -1) {
917        unit_id = 0;
918        while (drive_get(type, bus_id, unit_id) != NULL) {
919            unit_id++;
920            if (max_devs && unit_id >= max_devs) {
921                unit_id -= max_devs;
922                bus_id++;
923            }
924        }
925     }
926 
927     if (max_devs && unit_id >= max_devs) {
928         error_setg(errp, "unit %d too big (max is %d)", unit_id, max_devs - 1);
929         goto fail;
930     }
931 
932     if (drive_get(type, bus_id, unit_id) != NULL) {
933         error_setg(errp, "drive with bus=%d, unit=%d (index=%d) exists",
934                    bus_id, unit_id, index);
935         goto fail;
936     }
937 
938     /* no id supplied -> create one */
939     if (qemu_opts_id(all_opts) == NULL) {
940         char *new_id;
941         const char *mediastr = "";
942         if (type == IF_IDE || type == IF_SCSI) {
943             mediastr = (media == MEDIA_CDROM) ? "-cd" : "-hd";
944         }
945         if (max_devs) {
946             new_id = g_strdup_printf("%s%i%s%i", if_name[type], bus_id,
947                                      mediastr, unit_id);
948         } else {
949             new_id = g_strdup_printf("%s%s%i", if_name[type],
950                                      mediastr, unit_id);
951         }
952         qdict_put_str(bs_opts, "id", new_id);
953         g_free(new_id);
954     }
955 
956     /* Add virtio block device */
957     if (type == IF_VIRTIO) {
958         QemuOpts *devopts;
959         devopts = qemu_opts_create(qemu_find_opts("device"), NULL, 0,
960                                    &error_abort);
961         if (arch_type == QEMU_ARCH_S390X) {
962             qemu_opt_set(devopts, "driver", "virtio-blk-ccw", &error_abort);
963         } else {
964             qemu_opt_set(devopts, "driver", "virtio-blk-pci", &error_abort);
965         }
966         qemu_opt_set(devopts, "drive", qdict_get_str(bs_opts, "id"),
967                      &error_abort);
968     }
969 
970     filename = qemu_opt_get(legacy_opts, "file");
971 
972     /* Check werror/rerror compatibility with if=... */
973     werror = qemu_opt_get(legacy_opts, "werror");
974     if (werror != NULL) {
975         if (type != IF_IDE && type != IF_SCSI && type != IF_VIRTIO &&
976             type != IF_NONE) {
977             error_setg(errp, "werror is not supported by this bus type");
978             goto fail;
979         }
980         qdict_put_str(bs_opts, "werror", werror);
981     }
982 
983     rerror = qemu_opt_get(legacy_opts, "rerror");
984     if (rerror != NULL) {
985         if (type != IF_IDE && type != IF_VIRTIO && type != IF_SCSI &&
986             type != IF_NONE) {
987             error_setg(errp, "rerror is not supported by this bus type");
988             goto fail;
989         }
990         qdict_put_str(bs_opts, "rerror", rerror);
991     }
992 
993     /* Actual block device init: Functionality shared with blockdev-add */
994     blk = blockdev_init(filename, bs_opts, &local_err);
995     bs_opts = NULL;
996     if (!blk) {
997         error_propagate(errp, local_err);
998         goto fail;
999     } else {
1000         assert(!local_err);
1001     }
1002 
1003     /* Create legacy DriveInfo */
1004     dinfo = g_malloc0(sizeof(*dinfo));
1005     dinfo->opts = all_opts;
1006 
1007     dinfo->type = type;
1008     dinfo->bus = bus_id;
1009     dinfo->unit = unit_id;
1010 
1011     blk_set_legacy_dinfo(blk, dinfo);
1012 
1013     switch(type) {
1014     case IF_IDE:
1015     case IF_SCSI:
1016     case IF_XEN:
1017     case IF_NONE:
1018         dinfo->media_cd = media == MEDIA_CDROM;
1019         break;
1020     default:
1021         break;
1022     }
1023 
1024 fail:
1025     qemu_opts_del(legacy_opts);
1026     qobject_unref(bs_opts);
1027     return dinfo;
1028 }
1029 
1030 static BlockDriverState *qmp_get_root_bs(const char *name, Error **errp)
1031 {
1032     BlockDriverState *bs;
1033 
1034     bs = bdrv_lookup_bs(name, name, errp);
1035     if (bs == NULL) {
1036         return NULL;
1037     }
1038 
1039     if (!bdrv_is_root_node(bs)) {
1040         error_setg(errp, "Need a root block node");
1041         return NULL;
1042     }
1043 
1044     if (!bdrv_is_inserted(bs)) {
1045         error_setg(errp, "Device has no medium");
1046         return NULL;
1047     }
1048 
1049     return bs;
1050 }
1051 
1052 static BlockBackend *qmp_get_blk(const char *blk_name, const char *qdev_id,
1053                                  Error **errp)
1054 {
1055     BlockBackend *blk;
1056 
1057     if (!blk_name == !qdev_id) {
1058         error_setg(errp, "Need exactly one of 'device' and 'id'");
1059         return NULL;
1060     }
1061 
1062     if (qdev_id) {
1063         blk = blk_by_qdev_id(qdev_id, errp);
1064     } else {
1065         blk = blk_by_name(blk_name);
1066         if (blk == NULL) {
1067             error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
1068                       "Device '%s' not found", blk_name);
1069         }
1070     }
1071 
1072     return blk;
1073 }
1074 
1075 void hmp_commit(Monitor *mon, const QDict *qdict)
1076 {
1077     const char *device = qdict_get_str(qdict, "device");
1078     BlockBackend *blk;
1079     int ret;
1080 
1081     if (!strcmp(device, "all")) {
1082         ret = blk_commit_all();
1083     } else {
1084         BlockDriverState *bs;
1085         AioContext *aio_context;
1086 
1087         blk = blk_by_name(device);
1088         if (!blk) {
1089             monitor_printf(mon, "Device '%s' not found\n", device);
1090             return;
1091         }
1092         if (!blk_is_available(blk)) {
1093             monitor_printf(mon, "Device '%s' has no medium\n", device);
1094             return;
1095         }
1096 
1097         bs = blk_bs(blk);
1098         aio_context = bdrv_get_aio_context(bs);
1099         aio_context_acquire(aio_context);
1100 
1101         ret = bdrv_commit(bs);
1102 
1103         aio_context_release(aio_context);
1104     }
1105     if (ret < 0) {
1106         monitor_printf(mon, "'commit' error for '%s': %s\n", device,
1107                        strerror(-ret));
1108     }
1109 }
1110 
1111 static void blockdev_do_action(TransactionAction *action, Error **errp)
1112 {
1113     TransactionActionList list;
1114 
1115     list.value = action;
1116     list.next = NULL;
1117     qmp_transaction(&list, false, NULL, errp);
1118 }
1119 
1120 void qmp_blockdev_snapshot_sync(bool has_device, const char *device,
1121                                 bool has_node_name, const char *node_name,
1122                                 const char *snapshot_file,
1123                                 bool has_snapshot_node_name,
1124                                 const char *snapshot_node_name,
1125                                 bool has_format, const char *format,
1126                                 bool has_mode, NewImageMode mode, Error **errp)
1127 {
1128     BlockdevSnapshotSync snapshot = {
1129         .has_device = has_device,
1130         .device = (char *) device,
1131         .has_node_name = has_node_name,
1132         .node_name = (char *) node_name,
1133         .snapshot_file = (char *) snapshot_file,
1134         .has_snapshot_node_name = has_snapshot_node_name,
1135         .snapshot_node_name = (char *) snapshot_node_name,
1136         .has_format = has_format,
1137         .format = (char *) format,
1138         .has_mode = has_mode,
1139         .mode = mode,
1140     };
1141     TransactionAction action = {
1142         .type = TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC,
1143         .u.blockdev_snapshot_sync.data = &snapshot,
1144     };
1145     blockdev_do_action(&action, errp);
1146 }
1147 
1148 void qmp_blockdev_snapshot(const char *node, const char *overlay,
1149                            Error **errp)
1150 {
1151     BlockdevSnapshot snapshot_data = {
1152         .node = (char *) node,
1153         .overlay = (char *) overlay
1154     };
1155     TransactionAction action = {
1156         .type = TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT,
1157         .u.blockdev_snapshot.data = &snapshot_data,
1158     };
1159     blockdev_do_action(&action, errp);
1160 }
1161 
1162 void qmp_blockdev_snapshot_internal_sync(const char *device,
1163                                          const char *name,
1164                                          Error **errp)
1165 {
1166     BlockdevSnapshotInternal snapshot = {
1167         .device = (char *) device,
1168         .name = (char *) name
1169     };
1170     TransactionAction action = {
1171         .type = TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC,
1172         .u.blockdev_snapshot_internal_sync.data = &snapshot,
1173     };
1174     blockdev_do_action(&action, errp);
1175 }
1176 
1177 SnapshotInfo *qmp_blockdev_snapshot_delete_internal_sync(const char *device,
1178                                                          bool has_id,
1179                                                          const char *id,
1180                                                          bool has_name,
1181                                                          const char *name,
1182                                                          Error **errp)
1183 {
1184     BlockDriverState *bs;
1185     AioContext *aio_context;
1186     QEMUSnapshotInfo sn;
1187     Error *local_err = NULL;
1188     SnapshotInfo *info = NULL;
1189     int ret;
1190 
1191     bs = qmp_get_root_bs(device, errp);
1192     if (!bs) {
1193         return NULL;
1194     }
1195     aio_context = bdrv_get_aio_context(bs);
1196     aio_context_acquire(aio_context);
1197 
1198     if (!has_id) {
1199         id = NULL;
1200     }
1201 
1202     if (!has_name) {
1203         name = NULL;
1204     }
1205 
1206     if (!id && !name) {
1207         error_setg(errp, "Name or id must be provided");
1208         goto out_aio_context;
1209     }
1210 
1211     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_INTERNAL_SNAPSHOT_DELETE, errp)) {
1212         goto out_aio_context;
1213     }
1214 
1215     ret = bdrv_snapshot_find_by_id_and_name(bs, id, name, &sn, &local_err);
1216     if (local_err) {
1217         error_propagate(errp, local_err);
1218         goto out_aio_context;
1219     }
1220     if (!ret) {
1221         error_setg(errp,
1222                    "Snapshot with id '%s' and name '%s' does not exist on "
1223                    "device '%s'",
1224                    STR_OR_NULL(id), STR_OR_NULL(name), device);
1225         goto out_aio_context;
1226     }
1227 
1228     bdrv_snapshot_delete(bs, id, name, &local_err);
1229     if (local_err) {
1230         error_propagate(errp, local_err);
1231         goto out_aio_context;
1232     }
1233 
1234     aio_context_release(aio_context);
1235 
1236     info = g_new0(SnapshotInfo, 1);
1237     info->id = g_strdup(sn.id_str);
1238     info->name = g_strdup(sn.name);
1239     info->date_nsec = sn.date_nsec;
1240     info->date_sec = sn.date_sec;
1241     info->vm_state_size = sn.vm_state_size;
1242     info->vm_clock_nsec = sn.vm_clock_nsec % 1000000000;
1243     info->vm_clock_sec = sn.vm_clock_nsec / 1000000000;
1244 
1245     return info;
1246 
1247 out_aio_context:
1248     aio_context_release(aio_context);
1249     return NULL;
1250 }
1251 
1252 /**
1253  * block_dirty_bitmap_lookup:
1254  * Return a dirty bitmap (if present), after validating
1255  * the node reference and bitmap names.
1256  *
1257  * @node: The name of the BDS node to search for bitmaps
1258  * @name: The name of the bitmap to search for
1259  * @pbs: Output pointer for BDS lookup, if desired. Can be NULL.
1260  * @paio: Output pointer for aio_context acquisition, if desired. Can be NULL.
1261  * @errp: Output pointer for error information. Can be NULL.
1262  *
1263  * @return: A bitmap object on success, or NULL on failure.
1264  */
1265 static BdrvDirtyBitmap *block_dirty_bitmap_lookup(const char *node,
1266                                                   const char *name,
1267                                                   BlockDriverState **pbs,
1268                                                   Error **errp)
1269 {
1270     BlockDriverState *bs;
1271     BdrvDirtyBitmap *bitmap;
1272 
1273     if (!node) {
1274         error_setg(errp, "Node cannot be NULL");
1275         return NULL;
1276     }
1277     if (!name) {
1278         error_setg(errp, "Bitmap name cannot be NULL");
1279         return NULL;
1280     }
1281     bs = bdrv_lookup_bs(node, node, NULL);
1282     if (!bs) {
1283         error_setg(errp, "Node '%s' not found", node);
1284         return NULL;
1285     }
1286 
1287     bitmap = bdrv_find_dirty_bitmap(bs, name);
1288     if (!bitmap) {
1289         error_setg(errp, "Dirty bitmap '%s' not found", name);
1290         return NULL;
1291     }
1292 
1293     if (pbs) {
1294         *pbs = bs;
1295     }
1296 
1297     return bitmap;
1298 }
1299 
1300 /* New and old BlockDriverState structs for atomic group operations */
1301 
1302 typedef struct BlkActionState BlkActionState;
1303 
1304 /**
1305  * BlkActionOps:
1306  * Table of operations that define an Action.
1307  *
1308  * @instance_size: Size of state struct, in bytes.
1309  * @prepare: Prepare the work, must NOT be NULL.
1310  * @commit: Commit the changes, can be NULL.
1311  * @abort: Abort the changes on fail, can be NULL.
1312  * @clean: Clean up resources after all transaction actions have called
1313  *         commit() or abort(). Can be NULL.
1314  *
1315  * Only prepare() may fail. In a single transaction, only one of commit() or
1316  * abort() will be called. clean() will always be called if it is present.
1317  */
1318 typedef struct BlkActionOps {
1319     size_t instance_size;
1320     void (*prepare)(BlkActionState *common, Error **errp);
1321     void (*commit)(BlkActionState *common);
1322     void (*abort)(BlkActionState *common);
1323     void (*clean)(BlkActionState *common);
1324 } BlkActionOps;
1325 
1326 /**
1327  * BlkActionState:
1328  * Describes one Action's state within a Transaction.
1329  *
1330  * @action: QAPI-defined enum identifying which Action to perform.
1331  * @ops: Table of ActionOps this Action can perform.
1332  * @block_job_txn: Transaction which this action belongs to.
1333  * @entry: List membership for all Actions in this Transaction.
1334  *
1335  * This structure must be arranged as first member in a subclassed type,
1336  * assuming that the compiler will also arrange it to the same offsets as the
1337  * base class.
1338  */
1339 struct BlkActionState {
1340     TransactionAction *action;
1341     const BlkActionOps *ops;
1342     JobTxn *block_job_txn;
1343     TransactionProperties *txn_props;
1344     QTAILQ_ENTRY(BlkActionState) entry;
1345 };
1346 
1347 /* internal snapshot private data */
1348 typedef struct InternalSnapshotState {
1349     BlkActionState common;
1350     BlockDriverState *bs;
1351     QEMUSnapshotInfo sn;
1352     bool created;
1353 } InternalSnapshotState;
1354 
1355 
1356 static int action_check_completion_mode(BlkActionState *s, Error **errp)
1357 {
1358     if (s->txn_props->completion_mode != ACTION_COMPLETION_MODE_INDIVIDUAL) {
1359         error_setg(errp,
1360                    "Action '%s' does not support Transaction property "
1361                    "completion-mode = %s",
1362                    TransactionActionKind_str(s->action->type),
1363                    ActionCompletionMode_str(s->txn_props->completion_mode));
1364         return -1;
1365     }
1366     return 0;
1367 }
1368 
1369 static void internal_snapshot_prepare(BlkActionState *common,
1370                                       Error **errp)
1371 {
1372     Error *local_err = NULL;
1373     const char *device;
1374     const char *name;
1375     BlockDriverState *bs;
1376     QEMUSnapshotInfo old_sn, *sn;
1377     bool ret;
1378     qemu_timeval tv;
1379     BlockdevSnapshotInternal *internal;
1380     InternalSnapshotState *state;
1381     AioContext *aio_context;
1382     int ret1;
1383 
1384     g_assert(common->action->type ==
1385              TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC);
1386     internal = common->action->u.blockdev_snapshot_internal_sync.data;
1387     state = DO_UPCAST(InternalSnapshotState, common, common);
1388 
1389     /* 1. parse input */
1390     device = internal->device;
1391     name = internal->name;
1392 
1393     /* 2. check for validation */
1394     if (action_check_completion_mode(common, errp) < 0) {
1395         return;
1396     }
1397 
1398     bs = qmp_get_root_bs(device, errp);
1399     if (!bs) {
1400         return;
1401     }
1402 
1403     aio_context = bdrv_get_aio_context(bs);
1404     aio_context_acquire(aio_context);
1405 
1406     state->bs = bs;
1407 
1408     /* Paired with .clean() */
1409     bdrv_drained_begin(bs);
1410 
1411     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_INTERNAL_SNAPSHOT, errp)) {
1412         goto out;
1413     }
1414 
1415     if (bdrv_is_read_only(bs)) {
1416         error_setg(errp, "Device '%s' is read only", device);
1417         goto out;
1418     }
1419 
1420     if (!bdrv_can_snapshot(bs)) {
1421         error_setg(errp, "Block format '%s' used by device '%s' "
1422                    "does not support internal snapshots",
1423                    bs->drv->format_name, device);
1424         goto out;
1425     }
1426 
1427     if (!strlen(name)) {
1428         error_setg(errp, "Name is empty");
1429         goto out;
1430     }
1431 
1432     /* check whether a snapshot with name exist */
1433     ret = bdrv_snapshot_find_by_id_and_name(bs, NULL, name, &old_sn,
1434                                             &local_err);
1435     if (local_err) {
1436         error_propagate(errp, local_err);
1437         goto out;
1438     } else if (ret) {
1439         error_setg(errp,
1440                    "Snapshot with name '%s' already exists on device '%s'",
1441                    name, device);
1442         goto out;
1443     }
1444 
1445     /* 3. take the snapshot */
1446     sn = &state->sn;
1447     pstrcpy(sn->name, sizeof(sn->name), name);
1448     qemu_gettimeofday(&tv);
1449     sn->date_sec = tv.tv_sec;
1450     sn->date_nsec = tv.tv_usec * 1000;
1451     sn->vm_clock_nsec = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL);
1452 
1453     ret1 = bdrv_snapshot_create(bs, sn);
1454     if (ret1 < 0) {
1455         error_setg_errno(errp, -ret1,
1456                          "Failed to create snapshot '%s' on device '%s'",
1457                          name, device);
1458         goto out;
1459     }
1460 
1461     /* 4. succeed, mark a snapshot is created */
1462     state->created = true;
1463 
1464 out:
1465     aio_context_release(aio_context);
1466 }
1467 
1468 static void internal_snapshot_abort(BlkActionState *common)
1469 {
1470     InternalSnapshotState *state =
1471                              DO_UPCAST(InternalSnapshotState, common, common);
1472     BlockDriverState *bs = state->bs;
1473     QEMUSnapshotInfo *sn = &state->sn;
1474     AioContext *aio_context;
1475     Error *local_error = NULL;
1476 
1477     if (!state->created) {
1478         return;
1479     }
1480 
1481     aio_context = bdrv_get_aio_context(state->bs);
1482     aio_context_acquire(aio_context);
1483 
1484     if (bdrv_snapshot_delete(bs, sn->id_str, sn->name, &local_error) < 0) {
1485         error_reportf_err(local_error,
1486                           "Failed to delete snapshot with id '%s' and "
1487                           "name '%s' on device '%s' in abort: ",
1488                           sn->id_str, sn->name,
1489                           bdrv_get_device_name(bs));
1490     }
1491 
1492     aio_context_release(aio_context);
1493 }
1494 
1495 static void internal_snapshot_clean(BlkActionState *common)
1496 {
1497     InternalSnapshotState *state = DO_UPCAST(InternalSnapshotState,
1498                                              common, common);
1499     AioContext *aio_context;
1500 
1501     if (!state->bs) {
1502         return;
1503     }
1504 
1505     aio_context = bdrv_get_aio_context(state->bs);
1506     aio_context_acquire(aio_context);
1507 
1508     bdrv_drained_end(state->bs);
1509 
1510     aio_context_release(aio_context);
1511 }
1512 
1513 /* external snapshot private data */
1514 typedef struct ExternalSnapshotState {
1515     BlkActionState common;
1516     BlockDriverState *old_bs;
1517     BlockDriverState *new_bs;
1518     bool overlay_appended;
1519 } ExternalSnapshotState;
1520 
1521 static void external_snapshot_prepare(BlkActionState *common,
1522                                       Error **errp)
1523 {
1524     int flags = 0;
1525     QDict *options = NULL;
1526     Error *local_err = NULL;
1527     /* Device and node name of the image to generate the snapshot from */
1528     const char *device;
1529     const char *node_name;
1530     /* Reference to the new image (for 'blockdev-snapshot') */
1531     const char *snapshot_ref;
1532     /* File name of the new image (for 'blockdev-snapshot-sync') */
1533     const char *new_image_file;
1534     ExternalSnapshotState *state =
1535                              DO_UPCAST(ExternalSnapshotState, common, common);
1536     TransactionAction *action = common->action;
1537     AioContext *aio_context;
1538 
1539     /* 'blockdev-snapshot' and 'blockdev-snapshot-sync' have similar
1540      * purpose but a different set of parameters */
1541     switch (action->type) {
1542     case TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT:
1543         {
1544             BlockdevSnapshot *s = action->u.blockdev_snapshot.data;
1545             device = s->node;
1546             node_name = s->node;
1547             new_image_file = NULL;
1548             snapshot_ref = s->overlay;
1549         }
1550         break;
1551     case TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC:
1552         {
1553             BlockdevSnapshotSync *s = action->u.blockdev_snapshot_sync.data;
1554             device = s->has_device ? s->device : NULL;
1555             node_name = s->has_node_name ? s->node_name : NULL;
1556             new_image_file = s->snapshot_file;
1557             snapshot_ref = NULL;
1558         }
1559         break;
1560     default:
1561         g_assert_not_reached();
1562     }
1563 
1564     /* start processing */
1565     if (action_check_completion_mode(common, errp) < 0) {
1566         return;
1567     }
1568 
1569     state->old_bs = bdrv_lookup_bs(device, node_name, errp);
1570     if (!state->old_bs) {
1571         return;
1572     }
1573 
1574     aio_context = bdrv_get_aio_context(state->old_bs);
1575     aio_context_acquire(aio_context);
1576 
1577     /* Paired with .clean() */
1578     bdrv_drained_begin(state->old_bs);
1579 
1580     if (!bdrv_is_inserted(state->old_bs)) {
1581         error_setg(errp, QERR_DEVICE_HAS_NO_MEDIUM, device);
1582         goto out;
1583     }
1584 
1585     if (bdrv_op_is_blocked(state->old_bs,
1586                            BLOCK_OP_TYPE_EXTERNAL_SNAPSHOT, errp)) {
1587         goto out;
1588     }
1589 
1590     if (!bdrv_is_read_only(state->old_bs)) {
1591         if (bdrv_flush(state->old_bs)) {
1592             error_setg(errp, QERR_IO_ERROR);
1593             goto out;
1594         }
1595     }
1596 
1597     if (!bdrv_is_first_non_filter(state->old_bs)) {
1598         error_setg(errp, QERR_FEATURE_DISABLED, "snapshot");
1599         goto out;
1600     }
1601 
1602     if (action->type == TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC) {
1603         BlockdevSnapshotSync *s = action->u.blockdev_snapshot_sync.data;
1604         const char *format = s->has_format ? s->format : "qcow2";
1605         enum NewImageMode mode;
1606         const char *snapshot_node_name =
1607             s->has_snapshot_node_name ? s->snapshot_node_name : NULL;
1608 
1609         if (node_name && !snapshot_node_name) {
1610             error_setg(errp, "New snapshot node name missing");
1611             goto out;
1612         }
1613 
1614         if (snapshot_node_name &&
1615             bdrv_lookup_bs(snapshot_node_name, snapshot_node_name, NULL)) {
1616             error_setg(errp, "New snapshot node name already in use");
1617             goto out;
1618         }
1619 
1620         flags = state->old_bs->open_flags;
1621         flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_COPY_ON_READ);
1622         flags |= BDRV_O_NO_BACKING;
1623 
1624         /* create new image w/backing file */
1625         mode = s->has_mode ? s->mode : NEW_IMAGE_MODE_ABSOLUTE_PATHS;
1626         if (mode != NEW_IMAGE_MODE_EXISTING) {
1627             int64_t size = bdrv_getlength(state->old_bs);
1628             if (size < 0) {
1629                 error_setg_errno(errp, -size, "bdrv_getlength failed");
1630                 goto out;
1631             }
1632             bdrv_refresh_filename(state->old_bs);
1633             bdrv_img_create(new_image_file, format,
1634                             state->old_bs->filename,
1635                             state->old_bs->drv->format_name,
1636                             NULL, size, flags, false, &local_err);
1637             if (local_err) {
1638                 error_propagate(errp, local_err);
1639                 goto out;
1640             }
1641         }
1642 
1643         options = qdict_new();
1644         if (snapshot_node_name) {
1645             qdict_put_str(options, "node-name", snapshot_node_name);
1646         }
1647         qdict_put_str(options, "driver", format);
1648     }
1649 
1650     state->new_bs = bdrv_open(new_image_file, snapshot_ref, options, flags,
1651                               errp);
1652     /* We will manually add the backing_hd field to the bs later */
1653     if (!state->new_bs) {
1654         goto out;
1655     }
1656 
1657     if (bdrv_has_blk(state->new_bs)) {
1658         error_setg(errp, "The snapshot is already in use");
1659         goto out;
1660     }
1661 
1662     if (bdrv_op_is_blocked(state->new_bs, BLOCK_OP_TYPE_EXTERNAL_SNAPSHOT,
1663                            errp)) {
1664         goto out;
1665     }
1666 
1667     if (state->new_bs->backing != NULL) {
1668         error_setg(errp, "The snapshot already has a backing image");
1669         goto out;
1670     }
1671 
1672     if (!state->new_bs->drv->supports_backing) {
1673         error_setg(errp, "The snapshot does not support backing images");
1674         goto out;
1675     }
1676 
1677     bdrv_set_aio_context(state->new_bs, aio_context);
1678 
1679     /* This removes our old bs and adds the new bs. This is an operation that
1680      * can fail, so we need to do it in .prepare; undoing it for abort is
1681      * always possible. */
1682     bdrv_ref(state->new_bs);
1683     bdrv_append(state->new_bs, state->old_bs, &local_err);
1684     if (local_err) {
1685         error_propagate(errp, local_err);
1686         goto out;
1687     }
1688     state->overlay_appended = true;
1689 
1690 out:
1691     aio_context_release(aio_context);
1692 }
1693 
1694 static void external_snapshot_commit(BlkActionState *common)
1695 {
1696     ExternalSnapshotState *state =
1697                              DO_UPCAST(ExternalSnapshotState, common, common);
1698     AioContext *aio_context;
1699 
1700     aio_context = bdrv_get_aio_context(state->old_bs);
1701     aio_context_acquire(aio_context);
1702 
1703     /* We don't need (or want) to use the transactional
1704      * bdrv_reopen_multiple() across all the entries at once, because we
1705      * don't want to abort all of them if one of them fails the reopen */
1706     if (!atomic_read(&state->old_bs->copy_on_read)) {
1707         bdrv_reopen_set_read_only(state->old_bs, true, NULL);
1708     }
1709 
1710     aio_context_release(aio_context);
1711 }
1712 
1713 static void external_snapshot_abort(BlkActionState *common)
1714 {
1715     ExternalSnapshotState *state =
1716                              DO_UPCAST(ExternalSnapshotState, common, common);
1717     if (state->new_bs) {
1718         if (state->overlay_appended) {
1719             AioContext *aio_context;
1720 
1721             aio_context = bdrv_get_aio_context(state->old_bs);
1722             aio_context_acquire(aio_context);
1723 
1724             bdrv_ref(state->old_bs);   /* we can't let bdrv_set_backind_hd()
1725                                           close state->old_bs; we need it */
1726             bdrv_set_backing_hd(state->new_bs, NULL, &error_abort);
1727             bdrv_replace_node(state->new_bs, state->old_bs, &error_abort);
1728             bdrv_unref(state->old_bs); /* bdrv_replace_node() ref'ed old_bs */
1729 
1730             aio_context_release(aio_context);
1731         }
1732     }
1733 }
1734 
1735 static void external_snapshot_clean(BlkActionState *common)
1736 {
1737     ExternalSnapshotState *state =
1738                              DO_UPCAST(ExternalSnapshotState, common, common);
1739     AioContext *aio_context;
1740 
1741     if (!state->old_bs) {
1742         return;
1743     }
1744 
1745     aio_context = bdrv_get_aio_context(state->old_bs);
1746     aio_context_acquire(aio_context);
1747 
1748     bdrv_drained_end(state->old_bs);
1749     bdrv_unref(state->new_bs);
1750 
1751     aio_context_release(aio_context);
1752 }
1753 
1754 typedef struct DriveBackupState {
1755     BlkActionState common;
1756     BlockDriverState *bs;
1757     BlockJob *job;
1758 } DriveBackupState;
1759 
1760 static BlockJob *do_drive_backup(DriveBackup *backup, JobTxn *txn,
1761                             Error **errp);
1762 
1763 static void drive_backup_prepare(BlkActionState *common, Error **errp)
1764 {
1765     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1766     BlockDriverState *bs;
1767     DriveBackup *backup;
1768     AioContext *aio_context;
1769     Error *local_err = NULL;
1770 
1771     assert(common->action->type == TRANSACTION_ACTION_KIND_DRIVE_BACKUP);
1772     backup = common->action->u.drive_backup.data;
1773 
1774     bs = qmp_get_root_bs(backup->device, errp);
1775     if (!bs) {
1776         return;
1777     }
1778 
1779     aio_context = bdrv_get_aio_context(bs);
1780     aio_context_acquire(aio_context);
1781 
1782     /* Paired with .clean() */
1783     bdrv_drained_begin(bs);
1784 
1785     state->bs = bs;
1786 
1787     state->job = do_drive_backup(backup, common->block_job_txn, &local_err);
1788     if (local_err) {
1789         error_propagate(errp, local_err);
1790         goto out;
1791     }
1792 
1793 out:
1794     aio_context_release(aio_context);
1795 }
1796 
1797 static void drive_backup_commit(BlkActionState *common)
1798 {
1799     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1800     AioContext *aio_context;
1801 
1802     aio_context = bdrv_get_aio_context(state->bs);
1803     aio_context_acquire(aio_context);
1804 
1805     assert(state->job);
1806     job_start(&state->job->job);
1807 
1808     aio_context_release(aio_context);
1809 }
1810 
1811 static void drive_backup_abort(BlkActionState *common)
1812 {
1813     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1814 
1815     if (state->job) {
1816         AioContext *aio_context;
1817 
1818         aio_context = bdrv_get_aio_context(state->bs);
1819         aio_context_acquire(aio_context);
1820 
1821         job_cancel_sync(&state->job->job);
1822 
1823         aio_context_release(aio_context);
1824     }
1825 }
1826 
1827 static void drive_backup_clean(BlkActionState *common)
1828 {
1829     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1830     AioContext *aio_context;
1831 
1832     if (!state->bs) {
1833         return;
1834     }
1835 
1836     aio_context = bdrv_get_aio_context(state->bs);
1837     aio_context_acquire(aio_context);
1838 
1839     bdrv_drained_end(state->bs);
1840 
1841     aio_context_release(aio_context);
1842 }
1843 
1844 typedef struct BlockdevBackupState {
1845     BlkActionState common;
1846     BlockDriverState *bs;
1847     BlockJob *job;
1848 } BlockdevBackupState;
1849 
1850 static BlockJob *do_blockdev_backup(BlockdevBackup *backup, JobTxn *txn,
1851                                     Error **errp);
1852 
1853 static void blockdev_backup_prepare(BlkActionState *common, Error **errp)
1854 {
1855     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1856     BlockdevBackup *backup;
1857     BlockDriverState *bs, *target;
1858     AioContext *aio_context;
1859     Error *local_err = NULL;
1860 
1861     assert(common->action->type == TRANSACTION_ACTION_KIND_BLOCKDEV_BACKUP);
1862     backup = common->action->u.blockdev_backup.data;
1863 
1864     bs = bdrv_lookup_bs(backup->device, backup->device, errp);
1865     if (!bs) {
1866         return;
1867     }
1868 
1869     target = bdrv_lookup_bs(backup->target, backup->target, errp);
1870     if (!target) {
1871         return;
1872     }
1873 
1874     aio_context = bdrv_get_aio_context(bs);
1875     if (aio_context != bdrv_get_aio_context(target)) {
1876         error_setg(errp, "Backup between two IO threads is not implemented");
1877         return;
1878     }
1879     aio_context_acquire(aio_context);
1880     state->bs = bs;
1881 
1882     /* Paired with .clean() */
1883     bdrv_drained_begin(state->bs);
1884 
1885     state->job = do_blockdev_backup(backup, common->block_job_txn, &local_err);
1886     if (local_err) {
1887         error_propagate(errp, local_err);
1888         goto out;
1889     }
1890 
1891 out:
1892     aio_context_release(aio_context);
1893 }
1894 
1895 static void blockdev_backup_commit(BlkActionState *common)
1896 {
1897     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1898     AioContext *aio_context;
1899 
1900     aio_context = bdrv_get_aio_context(state->bs);
1901     aio_context_acquire(aio_context);
1902 
1903     assert(state->job);
1904     job_start(&state->job->job);
1905 
1906     aio_context_release(aio_context);
1907 }
1908 
1909 static void blockdev_backup_abort(BlkActionState *common)
1910 {
1911     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1912 
1913     if (state->job) {
1914         AioContext *aio_context;
1915 
1916         aio_context = bdrv_get_aio_context(state->bs);
1917         aio_context_acquire(aio_context);
1918 
1919         job_cancel_sync(&state->job->job);
1920 
1921         aio_context_release(aio_context);
1922     }
1923 }
1924 
1925 static void blockdev_backup_clean(BlkActionState *common)
1926 {
1927     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1928     AioContext *aio_context;
1929 
1930     if (!state->bs) {
1931         return;
1932     }
1933 
1934     aio_context = bdrv_get_aio_context(state->bs);
1935     aio_context_acquire(aio_context);
1936 
1937     bdrv_drained_end(state->bs);
1938 
1939     aio_context_release(aio_context);
1940 }
1941 
1942 typedef struct BlockDirtyBitmapState {
1943     BlkActionState common;
1944     BdrvDirtyBitmap *bitmap;
1945     BlockDriverState *bs;
1946     HBitmap *backup;
1947     bool prepared;
1948     bool was_enabled;
1949 } BlockDirtyBitmapState;
1950 
1951 static void block_dirty_bitmap_add_prepare(BlkActionState *common,
1952                                            Error **errp)
1953 {
1954     Error *local_err = NULL;
1955     BlockDirtyBitmapAdd *action;
1956     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
1957                                              common, common);
1958 
1959     if (action_check_completion_mode(common, errp) < 0) {
1960         return;
1961     }
1962 
1963     action = common->action->u.block_dirty_bitmap_add.data;
1964     /* AIO context taken and released within qmp_block_dirty_bitmap_add */
1965     qmp_block_dirty_bitmap_add(action->node, action->name,
1966                                action->has_granularity, action->granularity,
1967                                action->has_persistent, action->persistent,
1968                                action->has_autoload, action->autoload,
1969                                action->has_disabled, action->disabled,
1970                                &local_err);
1971 
1972     if (!local_err) {
1973         state->prepared = true;
1974     } else {
1975         error_propagate(errp, local_err);
1976     }
1977 }
1978 
1979 static void block_dirty_bitmap_add_abort(BlkActionState *common)
1980 {
1981     BlockDirtyBitmapAdd *action;
1982     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
1983                                              common, common);
1984 
1985     action = common->action->u.block_dirty_bitmap_add.data;
1986     /* Should not be able to fail: IF the bitmap was added via .prepare(),
1987      * then the node reference and bitmap name must have been valid.
1988      */
1989     if (state->prepared) {
1990         qmp_block_dirty_bitmap_remove(action->node, action->name, &error_abort);
1991     }
1992 }
1993 
1994 static void block_dirty_bitmap_clear_prepare(BlkActionState *common,
1995                                              Error **errp)
1996 {
1997     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
1998                                              common, common);
1999     BlockDirtyBitmap *action;
2000 
2001     if (action_check_completion_mode(common, errp) < 0) {
2002         return;
2003     }
2004 
2005     action = common->action->u.block_dirty_bitmap_clear.data;
2006     state->bitmap = block_dirty_bitmap_lookup(action->node,
2007                                               action->name,
2008                                               &state->bs,
2009                                               errp);
2010     if (!state->bitmap) {
2011         return;
2012     }
2013 
2014     if (bdrv_dirty_bitmap_user_locked(state->bitmap)) {
2015         error_setg(errp, "Cannot modify a bitmap in use by another operation");
2016         return;
2017     } else if (bdrv_dirty_bitmap_readonly(state->bitmap)) {
2018         error_setg(errp, "Cannot clear a readonly bitmap");
2019         return;
2020     }
2021 
2022     bdrv_clear_dirty_bitmap(state->bitmap, &state->backup);
2023 }
2024 
2025 static void block_dirty_bitmap_restore(BlkActionState *common)
2026 {
2027     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2028                                              common, common);
2029 
2030     if (state->backup) {
2031         bdrv_restore_dirty_bitmap(state->bitmap, state->backup);
2032     }
2033 }
2034 
2035 static void block_dirty_bitmap_free_backup(BlkActionState *common)
2036 {
2037     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2038                                              common, common);
2039 
2040     hbitmap_free(state->backup);
2041 }
2042 
2043 static void block_dirty_bitmap_enable_prepare(BlkActionState *common,
2044                                               Error **errp)
2045 {
2046     BlockDirtyBitmap *action;
2047     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2048                                              common, common);
2049 
2050     if (action_check_completion_mode(common, errp) < 0) {
2051         return;
2052     }
2053 
2054     action = common->action->u.block_dirty_bitmap_enable.data;
2055     state->bitmap = block_dirty_bitmap_lookup(action->node,
2056                                               action->name,
2057                                               NULL,
2058                                               errp);
2059     if (!state->bitmap) {
2060         return;
2061     }
2062 
2063     if (bdrv_dirty_bitmap_user_locked(state->bitmap)) {
2064         error_setg(errp,
2065                    "Bitmap '%s' is currently in use by another operation"
2066                    " and cannot be enabled", action->name);
2067         return;
2068     }
2069 
2070     state->was_enabled = bdrv_dirty_bitmap_enabled(state->bitmap);
2071     bdrv_enable_dirty_bitmap(state->bitmap);
2072 }
2073 
2074 static void block_dirty_bitmap_enable_abort(BlkActionState *common)
2075 {
2076     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2077                                              common, common);
2078 
2079     if (!state->was_enabled) {
2080         bdrv_disable_dirty_bitmap(state->bitmap);
2081     }
2082 }
2083 
2084 static void block_dirty_bitmap_disable_prepare(BlkActionState *common,
2085                                                Error **errp)
2086 {
2087     BlockDirtyBitmap *action;
2088     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2089                                              common, common);
2090 
2091     if (action_check_completion_mode(common, errp) < 0) {
2092         return;
2093     }
2094 
2095     action = common->action->u.block_dirty_bitmap_disable.data;
2096     state->bitmap = block_dirty_bitmap_lookup(action->node,
2097                                               action->name,
2098                                               NULL,
2099                                               errp);
2100     if (!state->bitmap) {
2101         return;
2102     }
2103 
2104     if (bdrv_dirty_bitmap_user_locked(state->bitmap)) {
2105         error_setg(errp,
2106                    "Bitmap '%s' is currently in use by another operation"
2107                    " and cannot be disabled", action->name);
2108         return;
2109     }
2110 
2111     state->was_enabled = bdrv_dirty_bitmap_enabled(state->bitmap);
2112     bdrv_disable_dirty_bitmap(state->bitmap);
2113 }
2114 
2115 static void block_dirty_bitmap_disable_abort(BlkActionState *common)
2116 {
2117     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2118                                              common, common);
2119 
2120     if (state->was_enabled) {
2121         bdrv_enable_dirty_bitmap(state->bitmap);
2122     }
2123 }
2124 
2125 static BdrvDirtyBitmap *do_block_dirty_bitmap_merge(const char *node,
2126                                                     const char *target,
2127                                                     strList *bitmaps,
2128                                                     HBitmap **backup,
2129                                                     Error **errp);
2130 
2131 static void block_dirty_bitmap_merge_prepare(BlkActionState *common,
2132                                              Error **errp)
2133 {
2134     BlockDirtyBitmapMerge *action;
2135     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2136                                              common, common);
2137 
2138     if (action_check_completion_mode(common, errp) < 0) {
2139         return;
2140     }
2141 
2142     action = common->action->u.block_dirty_bitmap_merge.data;
2143 
2144     state->bitmap = do_block_dirty_bitmap_merge(action->node, action->target,
2145                                                 action->bitmaps, &state->backup,
2146                                                 errp);
2147 }
2148 
2149 static void abort_prepare(BlkActionState *common, Error **errp)
2150 {
2151     error_setg(errp, "Transaction aborted using Abort action");
2152 }
2153 
2154 static void abort_commit(BlkActionState *common)
2155 {
2156     g_assert_not_reached(); /* this action never succeeds */
2157 }
2158 
2159 static const BlkActionOps actions[] = {
2160     [TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT] = {
2161         .instance_size = sizeof(ExternalSnapshotState),
2162         .prepare  = external_snapshot_prepare,
2163         .commit   = external_snapshot_commit,
2164         .abort = external_snapshot_abort,
2165         .clean = external_snapshot_clean,
2166     },
2167     [TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC] = {
2168         .instance_size = sizeof(ExternalSnapshotState),
2169         .prepare  = external_snapshot_prepare,
2170         .commit   = external_snapshot_commit,
2171         .abort = external_snapshot_abort,
2172         .clean = external_snapshot_clean,
2173     },
2174     [TRANSACTION_ACTION_KIND_DRIVE_BACKUP] = {
2175         .instance_size = sizeof(DriveBackupState),
2176         .prepare = drive_backup_prepare,
2177         .commit = drive_backup_commit,
2178         .abort = drive_backup_abort,
2179         .clean = drive_backup_clean,
2180     },
2181     [TRANSACTION_ACTION_KIND_BLOCKDEV_BACKUP] = {
2182         .instance_size = sizeof(BlockdevBackupState),
2183         .prepare = blockdev_backup_prepare,
2184         .commit = blockdev_backup_commit,
2185         .abort = blockdev_backup_abort,
2186         .clean = blockdev_backup_clean,
2187     },
2188     [TRANSACTION_ACTION_KIND_ABORT] = {
2189         .instance_size = sizeof(BlkActionState),
2190         .prepare = abort_prepare,
2191         .commit = abort_commit,
2192     },
2193     [TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC] = {
2194         .instance_size = sizeof(InternalSnapshotState),
2195         .prepare  = internal_snapshot_prepare,
2196         .abort = internal_snapshot_abort,
2197         .clean = internal_snapshot_clean,
2198     },
2199     [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_ADD] = {
2200         .instance_size = sizeof(BlockDirtyBitmapState),
2201         .prepare = block_dirty_bitmap_add_prepare,
2202         .abort = block_dirty_bitmap_add_abort,
2203     },
2204     [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_CLEAR] = {
2205         .instance_size = sizeof(BlockDirtyBitmapState),
2206         .prepare = block_dirty_bitmap_clear_prepare,
2207         .commit = block_dirty_bitmap_free_backup,
2208         .abort = block_dirty_bitmap_restore,
2209     },
2210     [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_ENABLE] = {
2211         .instance_size = sizeof(BlockDirtyBitmapState),
2212         .prepare = block_dirty_bitmap_enable_prepare,
2213         .abort = block_dirty_bitmap_enable_abort,
2214     },
2215     [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_DISABLE] = {
2216         .instance_size = sizeof(BlockDirtyBitmapState),
2217         .prepare = block_dirty_bitmap_disable_prepare,
2218         .abort = block_dirty_bitmap_disable_abort,
2219     },
2220     [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_MERGE] = {
2221         .instance_size = sizeof(BlockDirtyBitmapState),
2222         .prepare = block_dirty_bitmap_merge_prepare,
2223         .commit = block_dirty_bitmap_free_backup,
2224         .abort = block_dirty_bitmap_restore,
2225     },
2226     /* Where are transactions for MIRROR, COMMIT and STREAM?
2227      * Although these blockjobs use transaction callbacks like the backup job,
2228      * these jobs do not necessarily adhere to transaction semantics.
2229      * These jobs may not fully undo all of their actions on abort, nor do they
2230      * necessarily work in transactions with more than one job in them.
2231      */
2232 };
2233 
2234 /**
2235  * Allocate a TransactionProperties structure if necessary, and fill
2236  * that structure with desired defaults if they are unset.
2237  */
2238 static TransactionProperties *get_transaction_properties(
2239     TransactionProperties *props)
2240 {
2241     if (!props) {
2242         props = g_new0(TransactionProperties, 1);
2243     }
2244 
2245     if (!props->has_completion_mode) {
2246         props->has_completion_mode = true;
2247         props->completion_mode = ACTION_COMPLETION_MODE_INDIVIDUAL;
2248     }
2249 
2250     return props;
2251 }
2252 
2253 /*
2254  * 'Atomic' group operations.  The operations are performed as a set, and if
2255  * any fail then we roll back all operations in the group.
2256  */
2257 void qmp_transaction(TransactionActionList *dev_list,
2258                      bool has_props,
2259                      struct TransactionProperties *props,
2260                      Error **errp)
2261 {
2262     TransactionActionList *dev_entry = dev_list;
2263     JobTxn *block_job_txn = NULL;
2264     BlkActionState *state, *next;
2265     Error *local_err = NULL;
2266 
2267     QTAILQ_HEAD(, BlkActionState) snap_bdrv_states;
2268     QTAILQ_INIT(&snap_bdrv_states);
2269 
2270     /* Does this transaction get canceled as a group on failure?
2271      * If not, we don't really need to make a JobTxn.
2272      */
2273     props = get_transaction_properties(props);
2274     if (props->completion_mode != ACTION_COMPLETION_MODE_INDIVIDUAL) {
2275         block_job_txn = job_txn_new();
2276     }
2277 
2278     /* drain all i/o before any operations */
2279     bdrv_drain_all();
2280 
2281     /* We don't do anything in this loop that commits us to the operations */
2282     while (NULL != dev_entry) {
2283         TransactionAction *dev_info = NULL;
2284         const BlkActionOps *ops;
2285 
2286         dev_info = dev_entry->value;
2287         dev_entry = dev_entry->next;
2288 
2289         assert(dev_info->type < ARRAY_SIZE(actions));
2290 
2291         ops = &actions[dev_info->type];
2292         assert(ops->instance_size > 0);
2293 
2294         state = g_malloc0(ops->instance_size);
2295         state->ops = ops;
2296         state->action = dev_info;
2297         state->block_job_txn = block_job_txn;
2298         state->txn_props = props;
2299         QTAILQ_INSERT_TAIL(&snap_bdrv_states, state, entry);
2300 
2301         state->ops->prepare(state, &local_err);
2302         if (local_err) {
2303             error_propagate(errp, local_err);
2304             goto delete_and_fail;
2305         }
2306     }
2307 
2308     QTAILQ_FOREACH(state, &snap_bdrv_states, entry) {
2309         if (state->ops->commit) {
2310             state->ops->commit(state);
2311         }
2312     }
2313 
2314     /* success */
2315     goto exit;
2316 
2317 delete_and_fail:
2318     /* failure, and it is all-or-none; roll back all operations */
2319     QTAILQ_FOREACH_REVERSE(state, &snap_bdrv_states, entry) {
2320         if (state->ops->abort) {
2321             state->ops->abort(state);
2322         }
2323     }
2324 exit:
2325     QTAILQ_FOREACH_SAFE(state, &snap_bdrv_states, entry, next) {
2326         if (state->ops->clean) {
2327             state->ops->clean(state);
2328         }
2329         g_free(state);
2330     }
2331     if (!has_props) {
2332         qapi_free_TransactionProperties(props);
2333     }
2334     job_txn_unref(block_job_txn);
2335 }
2336 
2337 void qmp_eject(bool has_device, const char *device,
2338                bool has_id, const char *id,
2339                bool has_force, bool force, Error **errp)
2340 {
2341     Error *local_err = NULL;
2342     int rc;
2343 
2344     if (!has_force) {
2345         force = false;
2346     }
2347 
2348     rc = do_open_tray(has_device ? device : NULL,
2349                       has_id ? id : NULL,
2350                       force, &local_err);
2351     if (rc && rc != -ENOSYS) {
2352         error_propagate(errp, local_err);
2353         return;
2354     }
2355     error_free(local_err);
2356 
2357     blockdev_remove_medium(has_device, device, has_id, id, errp);
2358 }
2359 
2360 void qmp_block_passwd(bool has_device, const char *device,
2361                       bool has_node_name, const char *node_name,
2362                       const char *password, Error **errp)
2363 {
2364     error_setg(errp,
2365                "Setting block passwords directly is no longer supported");
2366 }
2367 
2368 /*
2369  * Attempt to open the tray of @device.
2370  * If @force, ignore its tray lock.
2371  * Else, if the tray is locked, don't open it, but ask the guest to open it.
2372  * On error, store an error through @errp and return -errno.
2373  * If @device does not exist, return -ENODEV.
2374  * If it has no removable media, return -ENOTSUP.
2375  * If it has no tray, return -ENOSYS.
2376  * If the guest was asked to open the tray, return -EINPROGRESS.
2377  * Else, return 0.
2378  */
2379 static int do_open_tray(const char *blk_name, const char *qdev_id,
2380                         bool force, Error **errp)
2381 {
2382     BlockBackend *blk;
2383     const char *device = qdev_id ?: blk_name;
2384     bool locked;
2385 
2386     blk = qmp_get_blk(blk_name, qdev_id, errp);
2387     if (!blk) {
2388         return -ENODEV;
2389     }
2390 
2391     if (!blk_dev_has_removable_media(blk)) {
2392         error_setg(errp, "Device '%s' is not removable", device);
2393         return -ENOTSUP;
2394     }
2395 
2396     if (!blk_dev_has_tray(blk)) {
2397         error_setg(errp, "Device '%s' does not have a tray", device);
2398         return -ENOSYS;
2399     }
2400 
2401     if (blk_dev_is_tray_open(blk)) {
2402         return 0;
2403     }
2404 
2405     locked = blk_dev_is_medium_locked(blk);
2406     if (locked) {
2407         blk_dev_eject_request(blk, force);
2408     }
2409 
2410     if (!locked || force) {
2411         blk_dev_change_media_cb(blk, false, &error_abort);
2412     }
2413 
2414     if (locked && !force) {
2415         error_setg(errp, "Device '%s' is locked and force was not specified, "
2416                    "wait for tray to open and try again", device);
2417         return -EINPROGRESS;
2418     }
2419 
2420     return 0;
2421 }
2422 
2423 void qmp_blockdev_open_tray(bool has_device, const char *device,
2424                             bool has_id, const char *id,
2425                             bool has_force, bool force,
2426                             Error **errp)
2427 {
2428     Error *local_err = NULL;
2429     int rc;
2430 
2431     if (!has_force) {
2432         force = false;
2433     }
2434     rc = do_open_tray(has_device ? device : NULL,
2435                       has_id ? id : NULL,
2436                       force, &local_err);
2437     if (rc && rc != -ENOSYS && rc != -EINPROGRESS) {
2438         error_propagate(errp, local_err);
2439         return;
2440     }
2441     error_free(local_err);
2442 }
2443 
2444 void qmp_blockdev_close_tray(bool has_device, const char *device,
2445                              bool has_id, const char *id,
2446                              Error **errp)
2447 {
2448     BlockBackend *blk;
2449     Error *local_err = NULL;
2450 
2451     device = has_device ? device : NULL;
2452     id = has_id ? id : NULL;
2453 
2454     blk = qmp_get_blk(device, id, errp);
2455     if (!blk) {
2456         return;
2457     }
2458 
2459     if (!blk_dev_has_removable_media(blk)) {
2460         error_setg(errp, "Device '%s' is not removable", device ?: id);
2461         return;
2462     }
2463 
2464     if (!blk_dev_has_tray(blk)) {
2465         /* Ignore this command on tray-less devices */
2466         return;
2467     }
2468 
2469     if (!blk_dev_is_tray_open(blk)) {
2470         return;
2471     }
2472 
2473     blk_dev_change_media_cb(blk, true, &local_err);
2474     if (local_err) {
2475         error_propagate(errp, local_err);
2476         return;
2477     }
2478 }
2479 
2480 static void blockdev_remove_medium(bool has_device, const char *device,
2481                                    bool has_id, const char *id, Error **errp)
2482 {
2483     BlockBackend *blk;
2484     BlockDriverState *bs;
2485     AioContext *aio_context;
2486     bool has_attached_device;
2487 
2488     device = has_device ? device : NULL;
2489     id = has_id ? id : NULL;
2490 
2491     blk = qmp_get_blk(device, id, errp);
2492     if (!blk) {
2493         return;
2494     }
2495 
2496     /* For BBs without a device, we can exchange the BDS tree at will */
2497     has_attached_device = blk_get_attached_dev(blk);
2498 
2499     if (has_attached_device && !blk_dev_has_removable_media(blk)) {
2500         error_setg(errp, "Device '%s' is not removable", device ?: id);
2501         return;
2502     }
2503 
2504     if (has_attached_device && blk_dev_has_tray(blk) &&
2505         !blk_dev_is_tray_open(blk))
2506     {
2507         error_setg(errp, "Tray of device '%s' is not open", device ?: id);
2508         return;
2509     }
2510 
2511     bs = blk_bs(blk);
2512     if (!bs) {
2513         return;
2514     }
2515 
2516     aio_context = bdrv_get_aio_context(bs);
2517     aio_context_acquire(aio_context);
2518 
2519     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_EJECT, errp)) {
2520         goto out;
2521     }
2522 
2523     blk_remove_bs(blk);
2524 
2525     if (!blk_dev_has_tray(blk)) {
2526         /* For tray-less devices, blockdev-open-tray is a no-op (or may not be
2527          * called at all); therefore, the medium needs to be ejected here.
2528          * Do it after blk_remove_bs() so blk_is_inserted(blk) returns the @load
2529          * value passed here (i.e. false). */
2530         blk_dev_change_media_cb(blk, false, &error_abort);
2531     }
2532 
2533 out:
2534     aio_context_release(aio_context);
2535 }
2536 
2537 void qmp_blockdev_remove_medium(const char *id, Error **errp)
2538 {
2539     blockdev_remove_medium(false, NULL, true, id, errp);
2540 }
2541 
2542 static void qmp_blockdev_insert_anon_medium(BlockBackend *blk,
2543                                             BlockDriverState *bs, Error **errp)
2544 {
2545     Error *local_err = NULL;
2546     bool has_device;
2547     int ret;
2548 
2549     /* For BBs without a device, we can exchange the BDS tree at will */
2550     has_device = blk_get_attached_dev(blk);
2551 
2552     if (has_device && !blk_dev_has_removable_media(blk)) {
2553         error_setg(errp, "Device is not removable");
2554         return;
2555     }
2556 
2557     if (has_device && blk_dev_has_tray(blk) && !blk_dev_is_tray_open(blk)) {
2558         error_setg(errp, "Tray of the device is not open");
2559         return;
2560     }
2561 
2562     if (blk_bs(blk)) {
2563         error_setg(errp, "There already is a medium in the device");
2564         return;
2565     }
2566 
2567     ret = blk_insert_bs(blk, bs, errp);
2568     if (ret < 0) {
2569         return;
2570     }
2571 
2572     if (!blk_dev_has_tray(blk)) {
2573         /* For tray-less devices, blockdev-close-tray is a no-op (or may not be
2574          * called at all); therefore, the medium needs to be pushed into the
2575          * slot here.
2576          * Do it after blk_insert_bs() so blk_is_inserted(blk) returns the @load
2577          * value passed here (i.e. true). */
2578         blk_dev_change_media_cb(blk, true, &local_err);
2579         if (local_err) {
2580             error_propagate(errp, local_err);
2581             blk_remove_bs(blk);
2582             return;
2583         }
2584     }
2585 }
2586 
2587 static void blockdev_insert_medium(bool has_device, const char *device,
2588                                    bool has_id, const char *id,
2589                                    const char *node_name, Error **errp)
2590 {
2591     BlockBackend *blk;
2592     BlockDriverState *bs;
2593 
2594     blk = qmp_get_blk(has_device ? device : NULL,
2595                       has_id ? id : NULL,
2596                       errp);
2597     if (!blk) {
2598         return;
2599     }
2600 
2601     bs = bdrv_find_node(node_name);
2602     if (!bs) {
2603         error_setg(errp, "Node '%s' not found", node_name);
2604         return;
2605     }
2606 
2607     if (bdrv_has_blk(bs)) {
2608         error_setg(errp, "Node '%s' is already in use", node_name);
2609         return;
2610     }
2611 
2612     qmp_blockdev_insert_anon_medium(blk, bs, errp);
2613 }
2614 
2615 void qmp_blockdev_insert_medium(const char *id, const char *node_name,
2616                                 Error **errp)
2617 {
2618     blockdev_insert_medium(false, NULL, true, id, node_name, errp);
2619 }
2620 
2621 void qmp_blockdev_change_medium(bool has_device, const char *device,
2622                                 bool has_id, const char *id,
2623                                 const char *filename,
2624                                 bool has_format, const char *format,
2625                                 bool has_read_only,
2626                                 BlockdevChangeReadOnlyMode read_only,
2627                                 Error **errp)
2628 {
2629     BlockBackend *blk;
2630     BlockDriverState *medium_bs = NULL;
2631     int bdrv_flags;
2632     bool detect_zeroes;
2633     int rc;
2634     QDict *options = NULL;
2635     Error *err = NULL;
2636 
2637     blk = qmp_get_blk(has_device ? device : NULL,
2638                       has_id ? id : NULL,
2639                       errp);
2640     if (!blk) {
2641         goto fail;
2642     }
2643 
2644     if (blk_bs(blk)) {
2645         blk_update_root_state(blk);
2646     }
2647 
2648     bdrv_flags = blk_get_open_flags_from_root_state(blk);
2649     bdrv_flags &= ~(BDRV_O_TEMPORARY | BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING |
2650         BDRV_O_PROTOCOL | BDRV_O_AUTO_RDONLY);
2651 
2652     if (!has_read_only) {
2653         read_only = BLOCKDEV_CHANGE_READ_ONLY_MODE_RETAIN;
2654     }
2655 
2656     switch (read_only) {
2657     case BLOCKDEV_CHANGE_READ_ONLY_MODE_RETAIN:
2658         break;
2659 
2660     case BLOCKDEV_CHANGE_READ_ONLY_MODE_READ_ONLY:
2661         bdrv_flags &= ~BDRV_O_RDWR;
2662         break;
2663 
2664     case BLOCKDEV_CHANGE_READ_ONLY_MODE_READ_WRITE:
2665         bdrv_flags |= BDRV_O_RDWR;
2666         break;
2667 
2668     default:
2669         abort();
2670     }
2671 
2672     options = qdict_new();
2673     detect_zeroes = blk_get_detect_zeroes_from_root_state(blk);
2674     qdict_put_str(options, "detect-zeroes", detect_zeroes ? "on" : "off");
2675 
2676     if (has_format) {
2677         qdict_put_str(options, "driver", format);
2678     }
2679 
2680     medium_bs = bdrv_open(filename, NULL, options, bdrv_flags, errp);
2681     if (!medium_bs) {
2682         goto fail;
2683     }
2684 
2685     rc = do_open_tray(has_device ? device : NULL,
2686                       has_id ? id : NULL,
2687                       false, &err);
2688     if (rc && rc != -ENOSYS) {
2689         error_propagate(errp, err);
2690         goto fail;
2691     }
2692     error_free(err);
2693     err = NULL;
2694 
2695     blockdev_remove_medium(has_device, device, has_id, id, &err);
2696     if (err) {
2697         error_propagate(errp, err);
2698         goto fail;
2699     }
2700 
2701     qmp_blockdev_insert_anon_medium(blk, medium_bs, &err);
2702     if (err) {
2703         error_propagate(errp, err);
2704         goto fail;
2705     }
2706 
2707     qmp_blockdev_close_tray(has_device, device, has_id, id, errp);
2708 
2709 fail:
2710     /* If the medium has been inserted, the device has its own reference, so
2711      * ours must be relinquished; and if it has not been inserted successfully,
2712      * the reference must be relinquished anyway */
2713     bdrv_unref(medium_bs);
2714 }
2715 
2716 /* throttling disk I/O limits */
2717 void qmp_block_set_io_throttle(BlockIOThrottle *arg, Error **errp)
2718 {
2719     ThrottleConfig cfg;
2720     BlockDriverState *bs;
2721     BlockBackend *blk;
2722     AioContext *aio_context;
2723 
2724     blk = qmp_get_blk(arg->has_device ? arg->device : NULL,
2725                       arg->has_id ? arg->id : NULL,
2726                       errp);
2727     if (!blk) {
2728         return;
2729     }
2730 
2731     aio_context = blk_get_aio_context(blk);
2732     aio_context_acquire(aio_context);
2733 
2734     bs = blk_bs(blk);
2735     if (!bs) {
2736         error_setg(errp, "Device has no medium");
2737         goto out;
2738     }
2739 
2740     throttle_config_init(&cfg);
2741     cfg.buckets[THROTTLE_BPS_TOTAL].avg = arg->bps;
2742     cfg.buckets[THROTTLE_BPS_READ].avg  = arg->bps_rd;
2743     cfg.buckets[THROTTLE_BPS_WRITE].avg = arg->bps_wr;
2744 
2745     cfg.buckets[THROTTLE_OPS_TOTAL].avg = arg->iops;
2746     cfg.buckets[THROTTLE_OPS_READ].avg  = arg->iops_rd;
2747     cfg.buckets[THROTTLE_OPS_WRITE].avg = arg->iops_wr;
2748 
2749     if (arg->has_bps_max) {
2750         cfg.buckets[THROTTLE_BPS_TOTAL].max = arg->bps_max;
2751     }
2752     if (arg->has_bps_rd_max) {
2753         cfg.buckets[THROTTLE_BPS_READ].max = arg->bps_rd_max;
2754     }
2755     if (arg->has_bps_wr_max) {
2756         cfg.buckets[THROTTLE_BPS_WRITE].max = arg->bps_wr_max;
2757     }
2758     if (arg->has_iops_max) {
2759         cfg.buckets[THROTTLE_OPS_TOTAL].max = arg->iops_max;
2760     }
2761     if (arg->has_iops_rd_max) {
2762         cfg.buckets[THROTTLE_OPS_READ].max = arg->iops_rd_max;
2763     }
2764     if (arg->has_iops_wr_max) {
2765         cfg.buckets[THROTTLE_OPS_WRITE].max = arg->iops_wr_max;
2766     }
2767 
2768     if (arg->has_bps_max_length) {
2769         cfg.buckets[THROTTLE_BPS_TOTAL].burst_length = arg->bps_max_length;
2770     }
2771     if (arg->has_bps_rd_max_length) {
2772         cfg.buckets[THROTTLE_BPS_READ].burst_length = arg->bps_rd_max_length;
2773     }
2774     if (arg->has_bps_wr_max_length) {
2775         cfg.buckets[THROTTLE_BPS_WRITE].burst_length = arg->bps_wr_max_length;
2776     }
2777     if (arg->has_iops_max_length) {
2778         cfg.buckets[THROTTLE_OPS_TOTAL].burst_length = arg->iops_max_length;
2779     }
2780     if (arg->has_iops_rd_max_length) {
2781         cfg.buckets[THROTTLE_OPS_READ].burst_length = arg->iops_rd_max_length;
2782     }
2783     if (arg->has_iops_wr_max_length) {
2784         cfg.buckets[THROTTLE_OPS_WRITE].burst_length = arg->iops_wr_max_length;
2785     }
2786 
2787     if (arg->has_iops_size) {
2788         cfg.op_size = arg->iops_size;
2789     }
2790 
2791     if (!throttle_is_valid(&cfg, errp)) {
2792         goto out;
2793     }
2794 
2795     if (throttle_enabled(&cfg)) {
2796         /* Enable I/O limits if they're not enabled yet, otherwise
2797          * just update the throttling group. */
2798         if (!blk_get_public(blk)->throttle_group_member.throttle_state) {
2799             blk_io_limits_enable(blk,
2800                                  arg->has_group ? arg->group :
2801                                  arg->has_device ? arg->device :
2802                                  arg->id);
2803         } else if (arg->has_group) {
2804             blk_io_limits_update_group(blk, arg->group);
2805         }
2806         /* Set the new throttling configuration */
2807         blk_set_io_limits(blk, &cfg);
2808     } else if (blk_get_public(blk)->throttle_group_member.throttle_state) {
2809         /* If all throttling settings are set to 0, disable I/O limits */
2810         blk_io_limits_disable(blk);
2811     }
2812 
2813 out:
2814     aio_context_release(aio_context);
2815 }
2816 
2817 void qmp_block_dirty_bitmap_add(const char *node, const char *name,
2818                                 bool has_granularity, uint32_t granularity,
2819                                 bool has_persistent, bool persistent,
2820                                 bool has_autoload, bool autoload,
2821                                 bool has_disabled, bool disabled,
2822                                 Error **errp)
2823 {
2824     BlockDriverState *bs;
2825     BdrvDirtyBitmap *bitmap;
2826     AioContext *aio_context = NULL;
2827 
2828     if (!name || name[0] == '\0') {
2829         error_setg(errp, "Bitmap name cannot be empty");
2830         return;
2831     }
2832 
2833     bs = bdrv_lookup_bs(node, node, errp);
2834     if (!bs) {
2835         return;
2836     }
2837 
2838     if (has_granularity) {
2839         if (granularity < 512 || !is_power_of_2(granularity)) {
2840             error_setg(errp, "Granularity must be power of 2 "
2841                              "and at least 512");
2842             return;
2843         }
2844     } else {
2845         /* Default to cluster size, if available: */
2846         granularity = bdrv_get_default_bitmap_granularity(bs);
2847     }
2848 
2849     if (!has_persistent) {
2850         persistent = false;
2851     }
2852 
2853     if (has_autoload) {
2854         warn_report("Autoload option is deprecated and its value is ignored");
2855     }
2856 
2857     if (!has_disabled) {
2858         disabled = false;
2859     }
2860 
2861     if (persistent) {
2862         aio_context = bdrv_get_aio_context(bs);
2863         aio_context_acquire(aio_context);
2864         if (!bdrv_can_store_new_dirty_bitmap(bs, name, granularity, errp)) {
2865             goto out;
2866         }
2867     }
2868 
2869     bitmap = bdrv_create_dirty_bitmap(bs, granularity, name, errp);
2870     if (bitmap == NULL) {
2871         goto out;
2872     }
2873 
2874     if (disabled) {
2875         bdrv_disable_dirty_bitmap(bitmap);
2876     }
2877 
2878     bdrv_dirty_bitmap_set_persistance(bitmap, persistent);
2879  out:
2880     if (aio_context) {
2881         aio_context_release(aio_context);
2882     }
2883 }
2884 
2885 void qmp_block_dirty_bitmap_remove(const char *node, const char *name,
2886                                    Error **errp)
2887 {
2888     BlockDriverState *bs;
2889     BdrvDirtyBitmap *bitmap;
2890     Error *local_err = NULL;
2891     AioContext *aio_context = NULL;
2892 
2893     bitmap = block_dirty_bitmap_lookup(node, name, &bs, errp);
2894     if (!bitmap || !bs) {
2895         return;
2896     }
2897 
2898     if (bdrv_dirty_bitmap_user_locked(bitmap)) {
2899         error_setg(errp,
2900                    "Bitmap '%s' is currently in use by another operation and"
2901                    " cannot be removed", name);
2902         return;
2903     }
2904 
2905     if (bdrv_dirty_bitmap_get_persistance(bitmap)) {
2906         aio_context = bdrv_get_aio_context(bs);
2907         aio_context_acquire(aio_context);
2908         bdrv_remove_persistent_dirty_bitmap(bs, name, &local_err);
2909         if (local_err != NULL) {
2910             error_propagate(errp, local_err);
2911             goto out;
2912         }
2913     }
2914 
2915     bdrv_release_dirty_bitmap(bs, bitmap);
2916  out:
2917     if (aio_context) {
2918         aio_context_release(aio_context);
2919     }
2920 }
2921 
2922 /**
2923  * Completely clear a bitmap, for the purposes of synchronizing a bitmap
2924  * immediately after a full backup operation.
2925  */
2926 void qmp_block_dirty_bitmap_clear(const char *node, const char *name,
2927                                   Error **errp)
2928 {
2929     BdrvDirtyBitmap *bitmap;
2930     BlockDriverState *bs;
2931 
2932     bitmap = block_dirty_bitmap_lookup(node, name, &bs, errp);
2933     if (!bitmap || !bs) {
2934         return;
2935     }
2936 
2937     if (bdrv_dirty_bitmap_user_locked(bitmap)) {
2938         error_setg(errp,
2939                    "Bitmap '%s' is currently in use by another operation"
2940                    " and cannot be cleared", name);
2941         return;
2942     } else if (bdrv_dirty_bitmap_readonly(bitmap)) {
2943         error_setg(errp, "Bitmap '%s' is readonly and cannot be cleared", name);
2944         return;
2945     }
2946 
2947     bdrv_clear_dirty_bitmap(bitmap, NULL);
2948 }
2949 
2950 void qmp_block_dirty_bitmap_enable(const char *node, const char *name,
2951                                    Error **errp)
2952 {
2953     BlockDriverState *bs;
2954     BdrvDirtyBitmap *bitmap;
2955 
2956     bitmap = block_dirty_bitmap_lookup(node, name, &bs, errp);
2957     if (!bitmap) {
2958         return;
2959     }
2960 
2961     if (bdrv_dirty_bitmap_user_locked(bitmap)) {
2962         error_setg(errp,
2963                    "Bitmap '%s' is currently in use by another operation"
2964                    " and cannot be enabled", name);
2965         return;
2966     }
2967 
2968     bdrv_enable_dirty_bitmap(bitmap);
2969 }
2970 
2971 void qmp_block_dirty_bitmap_disable(const char *node, const char *name,
2972                                     Error **errp)
2973 {
2974     BlockDriverState *bs;
2975     BdrvDirtyBitmap *bitmap;
2976 
2977     bitmap = block_dirty_bitmap_lookup(node, name, &bs, errp);
2978     if (!bitmap) {
2979         return;
2980     }
2981 
2982     if (bdrv_dirty_bitmap_user_locked(bitmap)) {
2983         error_setg(errp,
2984                    "Bitmap '%s' is currently in use by another operation"
2985                    " and cannot be disabled", name);
2986         return;
2987     }
2988 
2989     bdrv_disable_dirty_bitmap(bitmap);
2990 }
2991 
2992 static BdrvDirtyBitmap *do_block_dirty_bitmap_merge(const char *node,
2993                                                     const char *target,
2994                                                     strList *bitmaps,
2995                                                     HBitmap **backup,
2996                                                     Error **errp)
2997 {
2998     BlockDriverState *bs;
2999     BdrvDirtyBitmap *dst, *src, *anon;
3000     strList *lst;
3001     Error *local_err = NULL;
3002 
3003     dst = block_dirty_bitmap_lookup(node, target, &bs, errp);
3004     if (!dst) {
3005         return NULL;
3006     }
3007 
3008     anon = bdrv_create_dirty_bitmap(bs, bdrv_dirty_bitmap_granularity(dst),
3009                                     NULL, errp);
3010     if (!anon) {
3011         return NULL;
3012     }
3013 
3014     for (lst = bitmaps; lst; lst = lst->next) {
3015         src = bdrv_find_dirty_bitmap(bs, lst->value);
3016         if (!src) {
3017             error_setg(errp, "Dirty bitmap '%s' not found", lst->value);
3018             dst = NULL;
3019             goto out;
3020         }
3021 
3022         bdrv_merge_dirty_bitmap(anon, src, NULL, &local_err);
3023         if (local_err) {
3024             error_propagate(errp, local_err);
3025             dst = NULL;
3026             goto out;
3027         }
3028     }
3029 
3030     /* Merge into dst; dst is unchanged on failure. */
3031     bdrv_merge_dirty_bitmap(dst, anon, backup, errp);
3032 
3033  out:
3034     bdrv_release_dirty_bitmap(bs, anon);
3035     return dst;
3036 }
3037 
3038 void qmp_block_dirty_bitmap_merge(const char *node, const char *target,
3039                                   strList *bitmaps, Error **errp)
3040 {
3041     do_block_dirty_bitmap_merge(node, target, bitmaps, NULL, errp);
3042 }
3043 
3044 BlockDirtyBitmapSha256 *qmp_x_debug_block_dirty_bitmap_sha256(const char *node,
3045                                                               const char *name,
3046                                                               Error **errp)
3047 {
3048     BdrvDirtyBitmap *bitmap;
3049     BlockDriverState *bs;
3050     BlockDirtyBitmapSha256 *ret = NULL;
3051     char *sha256;
3052 
3053     bitmap = block_dirty_bitmap_lookup(node, name, &bs, errp);
3054     if (!bitmap || !bs) {
3055         return NULL;
3056     }
3057 
3058     sha256 = bdrv_dirty_bitmap_sha256(bitmap, errp);
3059     if (sha256 == NULL) {
3060         return NULL;
3061     }
3062 
3063     ret = g_new(BlockDirtyBitmapSha256, 1);
3064     ret->sha256 = sha256;
3065 
3066     return ret;
3067 }
3068 
3069 void hmp_drive_del(Monitor *mon, const QDict *qdict)
3070 {
3071     const char *id = qdict_get_str(qdict, "id");
3072     BlockBackend *blk;
3073     BlockDriverState *bs;
3074     AioContext *aio_context;
3075     Error *local_err = NULL;
3076 
3077     bs = bdrv_find_node(id);
3078     if (bs) {
3079         qmp_blockdev_del(id, &local_err);
3080         if (local_err) {
3081             error_report_err(local_err);
3082         }
3083         return;
3084     }
3085 
3086     blk = blk_by_name(id);
3087     if (!blk) {
3088         error_report("Device '%s' not found", id);
3089         return;
3090     }
3091 
3092     if (!blk_legacy_dinfo(blk)) {
3093         error_report("Deleting device added with blockdev-add"
3094                      " is not supported");
3095         return;
3096     }
3097 
3098     aio_context = blk_get_aio_context(blk);
3099     aio_context_acquire(aio_context);
3100 
3101     bs = blk_bs(blk);
3102     if (bs) {
3103         if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_DRIVE_DEL, &local_err)) {
3104             error_report_err(local_err);
3105             aio_context_release(aio_context);
3106             return;
3107         }
3108 
3109         blk_remove_bs(blk);
3110     }
3111 
3112     /* Make the BlockBackend and the attached BlockDriverState anonymous */
3113     monitor_remove_blk(blk);
3114 
3115     /* If this BlockBackend has a device attached to it, its refcount will be
3116      * decremented when the device is removed; otherwise we have to do so here.
3117      */
3118     if (blk_get_attached_dev(blk)) {
3119         /* Further I/O must not pause the guest */
3120         blk_set_on_error(blk, BLOCKDEV_ON_ERROR_REPORT,
3121                          BLOCKDEV_ON_ERROR_REPORT);
3122     } else {
3123         blk_unref(blk);
3124     }
3125 
3126     aio_context_release(aio_context);
3127 }
3128 
3129 void qmp_block_resize(bool has_device, const char *device,
3130                       bool has_node_name, const char *node_name,
3131                       int64_t size, Error **errp)
3132 {
3133     Error *local_err = NULL;
3134     BlockBackend *blk = NULL;
3135     BlockDriverState *bs;
3136     AioContext *aio_context;
3137     int ret;
3138 
3139     bs = bdrv_lookup_bs(has_device ? device : NULL,
3140                         has_node_name ? node_name : NULL,
3141                         &local_err);
3142     if (local_err) {
3143         error_propagate(errp, local_err);
3144         return;
3145     }
3146 
3147     aio_context = bdrv_get_aio_context(bs);
3148     aio_context_acquire(aio_context);
3149 
3150     if (!bdrv_is_first_non_filter(bs)) {
3151         error_setg(errp, QERR_FEATURE_DISABLED, "resize");
3152         goto out;
3153     }
3154 
3155     if (size < 0) {
3156         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "size", "a >0 size");
3157         goto out;
3158     }
3159 
3160     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_RESIZE, NULL)) {
3161         error_setg(errp, QERR_DEVICE_IN_USE, device);
3162         goto out;
3163     }
3164 
3165     blk = blk_new(BLK_PERM_RESIZE, BLK_PERM_ALL);
3166     ret = blk_insert_bs(blk, bs, errp);
3167     if (ret < 0) {
3168         goto out;
3169     }
3170 
3171     bdrv_drained_begin(bs);
3172     ret = blk_truncate(blk, size, PREALLOC_MODE_OFF, errp);
3173     bdrv_drained_end(bs);
3174 
3175 out:
3176     blk_unref(blk);
3177     aio_context_release(aio_context);
3178 }
3179 
3180 void qmp_block_stream(bool has_job_id, const char *job_id, const char *device,
3181                       bool has_base, const char *base,
3182                       bool has_base_node, const char *base_node,
3183                       bool has_backing_file, const char *backing_file,
3184                       bool has_speed, int64_t speed,
3185                       bool has_on_error, BlockdevOnError on_error,
3186                       bool has_auto_finalize, bool auto_finalize,
3187                       bool has_auto_dismiss, bool auto_dismiss,
3188                       Error **errp)
3189 {
3190     BlockDriverState *bs, *iter;
3191     BlockDriverState *base_bs = NULL;
3192     AioContext *aio_context;
3193     Error *local_err = NULL;
3194     const char *base_name = NULL;
3195     int job_flags = JOB_DEFAULT;
3196 
3197     if (!has_on_error) {
3198         on_error = BLOCKDEV_ON_ERROR_REPORT;
3199     }
3200 
3201     bs = bdrv_lookup_bs(device, device, errp);
3202     if (!bs) {
3203         return;
3204     }
3205 
3206     aio_context = bdrv_get_aio_context(bs);
3207     aio_context_acquire(aio_context);
3208 
3209     if (has_base && has_base_node) {
3210         error_setg(errp, "'base' and 'base-node' cannot be specified "
3211                    "at the same time");
3212         goto out;
3213     }
3214 
3215     if (has_base) {
3216         base_bs = bdrv_find_backing_image(bs, base);
3217         if (base_bs == NULL) {
3218             error_setg(errp, QERR_BASE_NOT_FOUND, base);
3219             goto out;
3220         }
3221         assert(bdrv_get_aio_context(base_bs) == aio_context);
3222         base_name = base;
3223     }
3224 
3225     if (has_base_node) {
3226         base_bs = bdrv_lookup_bs(NULL, base_node, errp);
3227         if (!base_bs) {
3228             goto out;
3229         }
3230         if (bs == base_bs || !bdrv_chain_contains(bs, base_bs)) {
3231             error_setg(errp, "Node '%s' is not a backing image of '%s'",
3232                        base_node, device);
3233             goto out;
3234         }
3235         assert(bdrv_get_aio_context(base_bs) == aio_context);
3236         bdrv_refresh_filename(base_bs);
3237         base_name = base_bs->filename;
3238     }
3239 
3240     /* Check for op blockers in the whole chain between bs and base */
3241     for (iter = bs; iter && iter != base_bs; iter = backing_bs(iter)) {
3242         if (bdrv_op_is_blocked(iter, BLOCK_OP_TYPE_STREAM, errp)) {
3243             goto out;
3244         }
3245     }
3246 
3247     /* if we are streaming the entire chain, the result will have no backing
3248      * file, and specifying one is therefore an error */
3249     if (base_bs == NULL && has_backing_file) {
3250         error_setg(errp, "backing file specified, but streaming the "
3251                          "entire chain");
3252         goto out;
3253     }
3254 
3255     /* backing_file string overrides base bs filename */
3256     base_name = has_backing_file ? backing_file : base_name;
3257 
3258     if (has_auto_finalize && !auto_finalize) {
3259         job_flags |= JOB_MANUAL_FINALIZE;
3260     }
3261     if (has_auto_dismiss && !auto_dismiss) {
3262         job_flags |= JOB_MANUAL_DISMISS;
3263     }
3264 
3265     stream_start(has_job_id ? job_id : NULL, bs, base_bs, base_name,
3266                  job_flags, has_speed ? speed : 0, on_error, &local_err);
3267     if (local_err) {
3268         error_propagate(errp, local_err);
3269         goto out;
3270     }
3271 
3272     trace_qmp_block_stream(bs, bs->job);
3273 
3274 out:
3275     aio_context_release(aio_context);
3276 }
3277 
3278 void qmp_block_commit(bool has_job_id, const char *job_id, const char *device,
3279                       bool has_base_node, const char *base_node,
3280                       bool has_base, const char *base,
3281                       bool has_top_node, const char *top_node,
3282                       bool has_top, const char *top,
3283                       bool has_backing_file, const char *backing_file,
3284                       bool has_speed, int64_t speed,
3285                       bool has_filter_node_name, const char *filter_node_name,
3286                       bool has_auto_finalize, bool auto_finalize,
3287                       bool has_auto_dismiss, bool auto_dismiss,
3288                       Error **errp)
3289 {
3290     BlockDriverState *bs;
3291     BlockDriverState *iter;
3292     BlockDriverState *base_bs, *top_bs;
3293     AioContext *aio_context;
3294     Error *local_err = NULL;
3295     /* This will be part of the QMP command, if/when the
3296      * BlockdevOnError change for blkmirror makes it in
3297      */
3298     BlockdevOnError on_error = BLOCKDEV_ON_ERROR_REPORT;
3299     int job_flags = JOB_DEFAULT;
3300 
3301     if (!has_speed) {
3302         speed = 0;
3303     }
3304     if (!has_filter_node_name) {
3305         filter_node_name = NULL;
3306     }
3307     if (has_auto_finalize && !auto_finalize) {
3308         job_flags |= JOB_MANUAL_FINALIZE;
3309     }
3310     if (has_auto_dismiss && !auto_dismiss) {
3311         job_flags |= JOB_MANUAL_DISMISS;
3312     }
3313 
3314     /* Important Note:
3315      *  libvirt relies on the DeviceNotFound error class in order to probe for
3316      *  live commit feature versions; for this to work, we must make sure to
3317      *  perform the device lookup before any generic errors that may occur in a
3318      *  scenario in which all optional arguments are omitted. */
3319     bs = qmp_get_root_bs(device, &local_err);
3320     if (!bs) {
3321         bs = bdrv_lookup_bs(device, device, NULL);
3322         if (!bs) {
3323             error_free(local_err);
3324             error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
3325                       "Device '%s' not found", device);
3326         } else {
3327             error_propagate(errp, local_err);
3328         }
3329         return;
3330     }
3331 
3332     aio_context = bdrv_get_aio_context(bs);
3333     aio_context_acquire(aio_context);
3334 
3335     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_COMMIT_SOURCE, errp)) {
3336         goto out;
3337     }
3338 
3339     /* default top_bs is the active layer */
3340     top_bs = bs;
3341 
3342     if (has_top_node && has_top) {
3343         error_setg(errp, "'top-node' and 'top' are mutually exclusive");
3344         goto out;
3345     } else if (has_top_node) {
3346         top_bs = bdrv_lookup_bs(NULL, top_node, errp);
3347         if (top_bs == NULL) {
3348             goto out;
3349         }
3350         if (!bdrv_chain_contains(bs, top_bs)) {
3351             error_setg(errp, "'%s' is not in this backing file chain",
3352                        top_node);
3353             goto out;
3354         }
3355     } else if (has_top && top) {
3356         /* This strcmp() is just a shortcut, there is no need to
3357          * refresh @bs's filename.  If it mismatches,
3358          * bdrv_find_backing_image() will do the refresh and may still
3359          * return @bs. */
3360         if (strcmp(bs->filename, top) != 0) {
3361             top_bs = bdrv_find_backing_image(bs, top);
3362         }
3363     }
3364 
3365     if (top_bs == NULL) {
3366         error_setg(errp, "Top image file %s not found", top ? top : "NULL");
3367         goto out;
3368     }
3369 
3370     assert(bdrv_get_aio_context(top_bs) == aio_context);
3371 
3372     if (has_base_node && has_base) {
3373         error_setg(errp, "'base-node' and 'base' are mutually exclusive");
3374         goto out;
3375     } else if (has_base_node) {
3376         base_bs = bdrv_lookup_bs(NULL, base_node, errp);
3377         if (base_bs == NULL) {
3378             goto out;
3379         }
3380         if (!bdrv_chain_contains(top_bs, base_bs)) {
3381             error_setg(errp, "'%s' is not in this backing file chain",
3382                        base_node);
3383             goto out;
3384         }
3385     } else if (has_base && base) {
3386         base_bs = bdrv_find_backing_image(top_bs, base);
3387     } else {
3388         base_bs = bdrv_find_base(top_bs);
3389     }
3390 
3391     if (base_bs == NULL) {
3392         error_setg(errp, QERR_BASE_NOT_FOUND, base ? base : "NULL");
3393         goto out;
3394     }
3395 
3396     assert(bdrv_get_aio_context(base_bs) == aio_context);
3397 
3398     for (iter = top_bs; iter != backing_bs(base_bs); iter = backing_bs(iter)) {
3399         if (bdrv_op_is_blocked(iter, BLOCK_OP_TYPE_COMMIT_TARGET, errp)) {
3400             goto out;
3401         }
3402     }
3403 
3404     /* Do not allow attempts to commit an image into itself */
3405     if (top_bs == base_bs) {
3406         error_setg(errp, "cannot commit an image into itself");
3407         goto out;
3408     }
3409 
3410     if (top_bs == bs) {
3411         if (has_backing_file) {
3412             error_setg(errp, "'backing-file' specified,"
3413                              " but 'top' is the active layer");
3414             goto out;
3415         }
3416         commit_active_start(has_job_id ? job_id : NULL, bs, base_bs,
3417                             job_flags, speed, on_error,
3418                             filter_node_name, NULL, NULL, false, &local_err);
3419     } else {
3420         BlockDriverState *overlay_bs = bdrv_find_overlay(bs, top_bs);
3421         if (bdrv_op_is_blocked(overlay_bs, BLOCK_OP_TYPE_COMMIT_TARGET, errp)) {
3422             goto out;
3423         }
3424         commit_start(has_job_id ? job_id : NULL, bs, base_bs, top_bs, job_flags,
3425                      speed, on_error, has_backing_file ? backing_file : NULL,
3426                      filter_node_name, &local_err);
3427     }
3428     if (local_err != NULL) {
3429         error_propagate(errp, local_err);
3430         goto out;
3431     }
3432 
3433 out:
3434     aio_context_release(aio_context);
3435 }
3436 
3437 static BlockJob *do_drive_backup(DriveBackup *backup, JobTxn *txn,
3438                                  Error **errp)
3439 {
3440     BlockDriverState *bs;
3441     BlockDriverState *target_bs;
3442     BlockDriverState *source = NULL;
3443     BlockJob *job = NULL;
3444     BdrvDirtyBitmap *bmap = NULL;
3445     AioContext *aio_context;
3446     QDict *options = NULL;
3447     Error *local_err = NULL;
3448     int flags, job_flags = JOB_DEFAULT;
3449     int64_t size;
3450     bool set_backing_hd = false;
3451 
3452     if (!backup->has_speed) {
3453         backup->speed = 0;
3454     }
3455     if (!backup->has_on_source_error) {
3456         backup->on_source_error = BLOCKDEV_ON_ERROR_REPORT;
3457     }
3458     if (!backup->has_on_target_error) {
3459         backup->on_target_error = BLOCKDEV_ON_ERROR_REPORT;
3460     }
3461     if (!backup->has_mode) {
3462         backup->mode = NEW_IMAGE_MODE_ABSOLUTE_PATHS;
3463     }
3464     if (!backup->has_job_id) {
3465         backup->job_id = NULL;
3466     }
3467     if (!backup->has_auto_finalize) {
3468         backup->auto_finalize = true;
3469     }
3470     if (!backup->has_auto_dismiss) {
3471         backup->auto_dismiss = true;
3472     }
3473     if (!backup->has_compress) {
3474         backup->compress = false;
3475     }
3476 
3477     bs = qmp_get_root_bs(backup->device, errp);
3478     if (!bs) {
3479         return NULL;
3480     }
3481 
3482     aio_context = bdrv_get_aio_context(bs);
3483     aio_context_acquire(aio_context);
3484 
3485     if (!backup->has_format) {
3486         backup->format = backup->mode == NEW_IMAGE_MODE_EXISTING ?
3487                          NULL : (char*) bs->drv->format_name;
3488     }
3489 
3490     /* Early check to avoid creating target */
3491     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_BACKUP_SOURCE, errp)) {
3492         goto out;
3493     }
3494 
3495     flags = bs->open_flags | BDRV_O_RDWR;
3496 
3497     /* See if we have a backing HD we can use to create our new image
3498      * on top of. */
3499     if (backup->sync == MIRROR_SYNC_MODE_TOP) {
3500         source = backing_bs(bs);
3501         if (!source) {
3502             backup->sync = MIRROR_SYNC_MODE_FULL;
3503         }
3504     }
3505     if (backup->sync == MIRROR_SYNC_MODE_NONE) {
3506         source = bs;
3507         flags |= BDRV_O_NO_BACKING;
3508         set_backing_hd = true;
3509     }
3510 
3511     size = bdrv_getlength(bs);
3512     if (size < 0) {
3513         error_setg_errno(errp, -size, "bdrv_getlength failed");
3514         goto out;
3515     }
3516 
3517     if (backup->mode != NEW_IMAGE_MODE_EXISTING) {
3518         assert(backup->format);
3519         if (source) {
3520             bdrv_refresh_filename(source);
3521             bdrv_img_create(backup->target, backup->format, source->filename,
3522                             source->drv->format_name, NULL,
3523                             size, flags, false, &local_err);
3524         } else {
3525             bdrv_img_create(backup->target, backup->format, NULL, NULL, NULL,
3526                             size, flags, false, &local_err);
3527         }
3528     }
3529 
3530     if (local_err) {
3531         error_propagate(errp, local_err);
3532         goto out;
3533     }
3534 
3535     if (backup->format) {
3536         if (!options) {
3537             options = qdict_new();
3538         }
3539         qdict_put_str(options, "driver", backup->format);
3540     }
3541 
3542     target_bs = bdrv_open(backup->target, NULL, options, flags, errp);
3543     if (!target_bs) {
3544         goto out;
3545     }
3546 
3547     bdrv_set_aio_context(target_bs, aio_context);
3548 
3549     if (set_backing_hd) {
3550         bdrv_set_backing_hd(target_bs, source, &local_err);
3551         if (local_err) {
3552             bdrv_unref(target_bs);
3553             goto out;
3554         }
3555     }
3556 
3557     if (backup->has_bitmap) {
3558         bmap = bdrv_find_dirty_bitmap(bs, backup->bitmap);
3559         if (!bmap) {
3560             error_setg(errp, "Bitmap '%s' could not be found", backup->bitmap);
3561             bdrv_unref(target_bs);
3562             goto out;
3563         }
3564         if (bdrv_dirty_bitmap_user_locked(bmap)) {
3565             error_setg(errp,
3566                        "Bitmap '%s' is currently in use by another operation"
3567                        " and cannot be used for backup", backup->bitmap);
3568             goto out;
3569         }
3570     }
3571     if (!backup->auto_finalize) {
3572         job_flags |= JOB_MANUAL_FINALIZE;
3573     }
3574     if (!backup->auto_dismiss) {
3575         job_flags |= JOB_MANUAL_DISMISS;
3576     }
3577 
3578     job = backup_job_create(backup->job_id, bs, target_bs, backup->speed,
3579                             backup->sync, bmap, backup->compress,
3580                             backup->on_source_error, backup->on_target_error,
3581                             job_flags, NULL, NULL, txn, &local_err);
3582     bdrv_unref(target_bs);
3583     if (local_err != NULL) {
3584         error_propagate(errp, local_err);
3585         goto out;
3586     }
3587 
3588 out:
3589     aio_context_release(aio_context);
3590     return job;
3591 }
3592 
3593 void qmp_drive_backup(DriveBackup *arg, Error **errp)
3594 {
3595 
3596     BlockJob *job;
3597     job = do_drive_backup(arg, NULL, errp);
3598     if (job) {
3599         job_start(&job->job);
3600     }
3601 }
3602 
3603 BlockDeviceInfoList *qmp_query_named_block_nodes(Error **errp)
3604 {
3605     return bdrv_named_nodes_list(errp);
3606 }
3607 
3608 XDbgBlockGraph *qmp_x_debug_query_block_graph(Error **errp)
3609 {
3610     return bdrv_get_xdbg_block_graph(errp);
3611 }
3612 
3613 BlockJob *do_blockdev_backup(BlockdevBackup *backup, JobTxn *txn,
3614                              Error **errp)
3615 {
3616     BlockDriverState *bs;
3617     BlockDriverState *target_bs;
3618     Error *local_err = NULL;
3619     BdrvDirtyBitmap *bmap = NULL;
3620     AioContext *aio_context;
3621     BlockJob *job = NULL;
3622     int job_flags = JOB_DEFAULT;
3623 
3624     if (!backup->has_speed) {
3625         backup->speed = 0;
3626     }
3627     if (!backup->has_on_source_error) {
3628         backup->on_source_error = BLOCKDEV_ON_ERROR_REPORT;
3629     }
3630     if (!backup->has_on_target_error) {
3631         backup->on_target_error = BLOCKDEV_ON_ERROR_REPORT;
3632     }
3633     if (!backup->has_job_id) {
3634         backup->job_id = NULL;
3635     }
3636     if (!backup->has_auto_finalize) {
3637         backup->auto_finalize = true;
3638     }
3639     if (!backup->has_auto_dismiss) {
3640         backup->auto_dismiss = true;
3641     }
3642     if (!backup->has_compress) {
3643         backup->compress = false;
3644     }
3645 
3646     bs = bdrv_lookup_bs(backup->device, backup->device, errp);
3647     if (!bs) {
3648         return NULL;
3649     }
3650 
3651     aio_context = bdrv_get_aio_context(bs);
3652     aio_context_acquire(aio_context);
3653 
3654     target_bs = bdrv_lookup_bs(backup->target, backup->target, errp);
3655     if (!target_bs) {
3656         goto out;
3657     }
3658 
3659     if (bdrv_get_aio_context(target_bs) != aio_context) {
3660         if (!bdrv_has_blk(target_bs)) {
3661             /* The target BDS is not attached, we can safely move it to another
3662              * AioContext. */
3663             bdrv_set_aio_context(target_bs, aio_context);
3664         } else {
3665             error_setg(errp, "Target is attached to a different thread from "
3666                              "source.");
3667             goto out;
3668         }
3669     }
3670 
3671     if (backup->has_bitmap) {
3672         bmap = bdrv_find_dirty_bitmap(bs, backup->bitmap);
3673         if (!bmap) {
3674             error_setg(errp, "Bitmap '%s' could not be found", backup->bitmap);
3675             goto out;
3676         }
3677         if (bdrv_dirty_bitmap_user_locked(bmap)) {
3678             error_setg(errp,
3679                        "Bitmap '%s' is currently in use by another operation"
3680                        " and cannot be used for backup", backup->bitmap);
3681             goto out;
3682         }
3683     }
3684 
3685     if (!backup->auto_finalize) {
3686         job_flags |= JOB_MANUAL_FINALIZE;
3687     }
3688     if (!backup->auto_dismiss) {
3689         job_flags |= JOB_MANUAL_DISMISS;
3690     }
3691     job = backup_job_create(backup->job_id, bs, target_bs, backup->speed,
3692                             backup->sync, bmap, backup->compress,
3693                             backup->on_source_error, backup->on_target_error,
3694                             job_flags, NULL, NULL, txn, &local_err);
3695     if (local_err != NULL) {
3696         error_propagate(errp, local_err);
3697     }
3698 out:
3699     aio_context_release(aio_context);
3700     return job;
3701 }
3702 
3703 void qmp_blockdev_backup(BlockdevBackup *arg, Error **errp)
3704 {
3705     BlockJob *job;
3706     job = do_blockdev_backup(arg, NULL, errp);
3707     if (job) {
3708         job_start(&job->job);
3709     }
3710 }
3711 
3712 /* Parameter check and block job starting for drive mirroring.
3713  * Caller should hold @device and @target's aio context (must be the same).
3714  **/
3715 static void blockdev_mirror_common(const char *job_id, BlockDriverState *bs,
3716                                    BlockDriverState *target,
3717                                    bool has_replaces, const char *replaces,
3718                                    enum MirrorSyncMode sync,
3719                                    BlockMirrorBackingMode backing_mode,
3720                                    bool has_speed, int64_t speed,
3721                                    bool has_granularity, uint32_t granularity,
3722                                    bool has_buf_size, int64_t buf_size,
3723                                    bool has_on_source_error,
3724                                    BlockdevOnError on_source_error,
3725                                    bool has_on_target_error,
3726                                    BlockdevOnError on_target_error,
3727                                    bool has_unmap, bool unmap,
3728                                    bool has_filter_node_name,
3729                                    const char *filter_node_name,
3730                                    bool has_copy_mode, MirrorCopyMode copy_mode,
3731                                    bool has_auto_finalize, bool auto_finalize,
3732                                    bool has_auto_dismiss, bool auto_dismiss,
3733                                    Error **errp)
3734 {
3735     int job_flags = JOB_DEFAULT;
3736 
3737     if (!has_speed) {
3738         speed = 0;
3739     }
3740     if (!has_on_source_error) {
3741         on_source_error = BLOCKDEV_ON_ERROR_REPORT;
3742     }
3743     if (!has_on_target_error) {
3744         on_target_error = BLOCKDEV_ON_ERROR_REPORT;
3745     }
3746     if (!has_granularity) {
3747         granularity = 0;
3748     }
3749     if (!has_buf_size) {
3750         buf_size = 0;
3751     }
3752     if (!has_unmap) {
3753         unmap = true;
3754     }
3755     if (!has_filter_node_name) {
3756         filter_node_name = NULL;
3757     }
3758     if (!has_copy_mode) {
3759         copy_mode = MIRROR_COPY_MODE_BACKGROUND;
3760     }
3761     if (has_auto_finalize && !auto_finalize) {
3762         job_flags |= JOB_MANUAL_FINALIZE;
3763     }
3764     if (has_auto_dismiss && !auto_dismiss) {
3765         job_flags |= JOB_MANUAL_DISMISS;
3766     }
3767 
3768     if (granularity != 0 && (granularity < 512 || granularity > 1048576 * 64)) {
3769         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "granularity",
3770                    "a value in range [512B, 64MB]");
3771         return;
3772     }
3773     if (granularity & (granularity - 1)) {
3774         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "granularity",
3775                    "power of 2");
3776         return;
3777     }
3778 
3779     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_MIRROR_SOURCE, errp)) {
3780         return;
3781     }
3782     if (bdrv_op_is_blocked(target, BLOCK_OP_TYPE_MIRROR_TARGET, errp)) {
3783         return;
3784     }
3785 
3786     if (!bs->backing && sync == MIRROR_SYNC_MODE_TOP) {
3787         sync = MIRROR_SYNC_MODE_FULL;
3788     }
3789 
3790     /* pass the node name to replace to mirror start since it's loose coupling
3791      * and will allow to check whether the node still exist at mirror completion
3792      */
3793     mirror_start(job_id, bs, target,
3794                  has_replaces ? replaces : NULL, job_flags,
3795                  speed, granularity, buf_size, sync, backing_mode,
3796                  on_source_error, on_target_error, unmap, filter_node_name,
3797                  copy_mode, errp);
3798 }
3799 
3800 void qmp_drive_mirror(DriveMirror *arg, Error **errp)
3801 {
3802     BlockDriverState *bs;
3803     BlockDriverState *source, *target_bs;
3804     AioContext *aio_context;
3805     BlockMirrorBackingMode backing_mode;
3806     Error *local_err = NULL;
3807     QDict *options = NULL;
3808     int flags;
3809     int64_t size;
3810     const char *format = arg->format;
3811 
3812     bs = qmp_get_root_bs(arg->device, errp);
3813     if (!bs) {
3814         return;
3815     }
3816 
3817     /* Early check to avoid creating target */
3818     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_MIRROR_SOURCE, errp)) {
3819         return;
3820     }
3821 
3822     aio_context = bdrv_get_aio_context(bs);
3823     aio_context_acquire(aio_context);
3824 
3825     if (!arg->has_mode) {
3826         arg->mode = NEW_IMAGE_MODE_ABSOLUTE_PATHS;
3827     }
3828 
3829     if (!arg->has_format) {
3830         format = (arg->mode == NEW_IMAGE_MODE_EXISTING
3831                   ? NULL : bs->drv->format_name);
3832     }
3833 
3834     flags = bs->open_flags | BDRV_O_RDWR;
3835     source = backing_bs(bs);
3836     if (!source && arg->sync == MIRROR_SYNC_MODE_TOP) {
3837         arg->sync = MIRROR_SYNC_MODE_FULL;
3838     }
3839     if (arg->sync == MIRROR_SYNC_MODE_NONE) {
3840         source = bs;
3841     }
3842 
3843     size = bdrv_getlength(bs);
3844     if (size < 0) {
3845         error_setg_errno(errp, -size, "bdrv_getlength failed");
3846         goto out;
3847     }
3848 
3849     if (arg->has_replaces) {
3850         BlockDriverState *to_replace_bs;
3851         AioContext *replace_aio_context;
3852         int64_t replace_size;
3853 
3854         if (!arg->has_node_name) {
3855             error_setg(errp, "a node-name must be provided when replacing a"
3856                              " named node of the graph");
3857             goto out;
3858         }
3859 
3860         to_replace_bs = check_to_replace_node(bs, arg->replaces, &local_err);
3861 
3862         if (!to_replace_bs) {
3863             error_propagate(errp, local_err);
3864             goto out;
3865         }
3866 
3867         replace_aio_context = bdrv_get_aio_context(to_replace_bs);
3868         aio_context_acquire(replace_aio_context);
3869         replace_size = bdrv_getlength(to_replace_bs);
3870         aio_context_release(replace_aio_context);
3871 
3872         if (size != replace_size) {
3873             error_setg(errp, "cannot replace image with a mirror image of "
3874                              "different size");
3875             goto out;
3876         }
3877     }
3878 
3879     if (arg->mode == NEW_IMAGE_MODE_ABSOLUTE_PATHS) {
3880         backing_mode = MIRROR_SOURCE_BACKING_CHAIN;
3881     } else {
3882         backing_mode = MIRROR_OPEN_BACKING_CHAIN;
3883     }
3884 
3885     /* Don't open backing image in create() */
3886     flags |= BDRV_O_NO_BACKING;
3887 
3888     if ((arg->sync == MIRROR_SYNC_MODE_FULL || !source)
3889         && arg->mode != NEW_IMAGE_MODE_EXISTING)
3890     {
3891         /* create new image w/o backing file */
3892         assert(format);
3893         bdrv_img_create(arg->target, format,
3894                         NULL, NULL, NULL, size, flags, false, &local_err);
3895     } else {
3896         switch (arg->mode) {
3897         case NEW_IMAGE_MODE_EXISTING:
3898             break;
3899         case NEW_IMAGE_MODE_ABSOLUTE_PATHS:
3900             /* create new image with backing file */
3901             bdrv_refresh_filename(source);
3902             bdrv_img_create(arg->target, format,
3903                             source->filename,
3904                             source->drv->format_name,
3905                             NULL, size, flags, false, &local_err);
3906             break;
3907         default:
3908             abort();
3909         }
3910     }
3911 
3912     if (local_err) {
3913         error_propagate(errp, local_err);
3914         goto out;
3915     }
3916 
3917     options = qdict_new();
3918     if (arg->has_node_name) {
3919         qdict_put_str(options, "node-name", arg->node_name);
3920     }
3921     if (format) {
3922         qdict_put_str(options, "driver", format);
3923     }
3924 
3925     /* Mirroring takes care of copy-on-write using the source's backing
3926      * file.
3927      */
3928     target_bs = bdrv_open(arg->target, NULL, options, flags, errp);
3929     if (!target_bs) {
3930         goto out;
3931     }
3932 
3933     bdrv_set_aio_context(target_bs, aio_context);
3934 
3935     blockdev_mirror_common(arg->has_job_id ? arg->job_id : NULL, bs, target_bs,
3936                            arg->has_replaces, arg->replaces, arg->sync,
3937                            backing_mode, arg->has_speed, arg->speed,
3938                            arg->has_granularity, arg->granularity,
3939                            arg->has_buf_size, arg->buf_size,
3940                            arg->has_on_source_error, arg->on_source_error,
3941                            arg->has_on_target_error, arg->on_target_error,
3942                            arg->has_unmap, arg->unmap,
3943                            false, NULL,
3944                            arg->has_copy_mode, arg->copy_mode,
3945                            arg->has_auto_finalize, arg->auto_finalize,
3946                            arg->has_auto_dismiss, arg->auto_dismiss,
3947                            &local_err);
3948     bdrv_unref(target_bs);
3949     error_propagate(errp, local_err);
3950 out:
3951     aio_context_release(aio_context);
3952 }
3953 
3954 void qmp_blockdev_mirror(bool has_job_id, const char *job_id,
3955                          const char *device, const char *target,
3956                          bool has_replaces, const char *replaces,
3957                          MirrorSyncMode sync,
3958                          bool has_speed, int64_t speed,
3959                          bool has_granularity, uint32_t granularity,
3960                          bool has_buf_size, int64_t buf_size,
3961                          bool has_on_source_error,
3962                          BlockdevOnError on_source_error,
3963                          bool has_on_target_error,
3964                          BlockdevOnError on_target_error,
3965                          bool has_filter_node_name,
3966                          const char *filter_node_name,
3967                          bool has_copy_mode, MirrorCopyMode copy_mode,
3968                          bool has_auto_finalize, bool auto_finalize,
3969                          bool has_auto_dismiss, bool auto_dismiss,
3970                          Error **errp)
3971 {
3972     BlockDriverState *bs;
3973     BlockDriverState *target_bs;
3974     AioContext *aio_context;
3975     BlockMirrorBackingMode backing_mode = MIRROR_LEAVE_BACKING_CHAIN;
3976     Error *local_err = NULL;
3977 
3978     bs = qmp_get_root_bs(device, errp);
3979     if (!bs) {
3980         return;
3981     }
3982 
3983     target_bs = bdrv_lookup_bs(target, target, errp);
3984     if (!target_bs) {
3985         return;
3986     }
3987 
3988     aio_context = bdrv_get_aio_context(bs);
3989     aio_context_acquire(aio_context);
3990 
3991     bdrv_set_aio_context(target_bs, aio_context);
3992 
3993     blockdev_mirror_common(has_job_id ? job_id : NULL, bs, target_bs,
3994                            has_replaces, replaces, sync, backing_mode,
3995                            has_speed, speed,
3996                            has_granularity, granularity,
3997                            has_buf_size, buf_size,
3998                            has_on_source_error, on_source_error,
3999                            has_on_target_error, on_target_error,
4000                            true, true,
4001                            has_filter_node_name, filter_node_name,
4002                            has_copy_mode, copy_mode,
4003                            has_auto_finalize, auto_finalize,
4004                            has_auto_dismiss, auto_dismiss,
4005                            &local_err);
4006     error_propagate(errp, local_err);
4007 
4008     aio_context_release(aio_context);
4009 }
4010 
4011 /* Get a block job using its ID and acquire its AioContext */
4012 static BlockJob *find_block_job(const char *id, AioContext **aio_context,
4013                                 Error **errp)
4014 {
4015     BlockJob *job;
4016 
4017     assert(id != NULL);
4018 
4019     *aio_context = NULL;
4020 
4021     job = block_job_get(id);
4022 
4023     if (!job) {
4024         error_set(errp, ERROR_CLASS_DEVICE_NOT_ACTIVE,
4025                   "Block job '%s' not found", id);
4026         return NULL;
4027     }
4028 
4029     *aio_context = blk_get_aio_context(job->blk);
4030     aio_context_acquire(*aio_context);
4031 
4032     return job;
4033 }
4034 
4035 void qmp_block_job_set_speed(const char *device, int64_t speed, Error **errp)
4036 {
4037     AioContext *aio_context;
4038     BlockJob *job = find_block_job(device, &aio_context, errp);
4039 
4040     if (!job) {
4041         return;
4042     }
4043 
4044     block_job_set_speed(job, speed, errp);
4045     aio_context_release(aio_context);
4046 }
4047 
4048 void qmp_block_job_cancel(const char *device,
4049                           bool has_force, bool force, Error **errp)
4050 {
4051     AioContext *aio_context;
4052     BlockJob *job = find_block_job(device, &aio_context, errp);
4053 
4054     if (!job) {
4055         return;
4056     }
4057 
4058     if (!has_force) {
4059         force = false;
4060     }
4061 
4062     if (job_user_paused(&job->job) && !force) {
4063         error_setg(errp, "The block job for device '%s' is currently paused",
4064                    device);
4065         goto out;
4066     }
4067 
4068     trace_qmp_block_job_cancel(job);
4069     job_user_cancel(&job->job, force, errp);
4070 out:
4071     aio_context_release(aio_context);
4072 }
4073 
4074 void qmp_block_job_pause(const char *device, Error **errp)
4075 {
4076     AioContext *aio_context;
4077     BlockJob *job = find_block_job(device, &aio_context, errp);
4078 
4079     if (!job) {
4080         return;
4081     }
4082 
4083     trace_qmp_block_job_pause(job);
4084     job_user_pause(&job->job, errp);
4085     aio_context_release(aio_context);
4086 }
4087 
4088 void qmp_block_job_resume(const char *device, Error **errp)
4089 {
4090     AioContext *aio_context;
4091     BlockJob *job = find_block_job(device, &aio_context, errp);
4092 
4093     if (!job) {
4094         return;
4095     }
4096 
4097     trace_qmp_block_job_resume(job);
4098     job_user_resume(&job->job, errp);
4099     aio_context_release(aio_context);
4100 }
4101 
4102 void qmp_block_job_complete(const char *device, Error **errp)
4103 {
4104     AioContext *aio_context;
4105     BlockJob *job = find_block_job(device, &aio_context, errp);
4106 
4107     if (!job) {
4108         return;
4109     }
4110 
4111     trace_qmp_block_job_complete(job);
4112     job_complete(&job->job, errp);
4113     aio_context_release(aio_context);
4114 }
4115 
4116 void qmp_block_job_finalize(const char *id, Error **errp)
4117 {
4118     AioContext *aio_context;
4119     BlockJob *job = find_block_job(id, &aio_context, errp);
4120 
4121     if (!job) {
4122         return;
4123     }
4124 
4125     trace_qmp_block_job_finalize(job);
4126     job_finalize(&job->job, errp);
4127     aio_context_release(aio_context);
4128 }
4129 
4130 void qmp_block_job_dismiss(const char *id, Error **errp)
4131 {
4132     AioContext *aio_context;
4133     BlockJob *bjob = find_block_job(id, &aio_context, errp);
4134     Job *job;
4135 
4136     if (!bjob) {
4137         return;
4138     }
4139 
4140     trace_qmp_block_job_dismiss(bjob);
4141     job = &bjob->job;
4142     job_dismiss(&job, errp);
4143     aio_context_release(aio_context);
4144 }
4145 
4146 void qmp_change_backing_file(const char *device,
4147                              const char *image_node_name,
4148                              const char *backing_file,
4149                              Error **errp)
4150 {
4151     BlockDriverState *bs = NULL;
4152     AioContext *aio_context;
4153     BlockDriverState *image_bs = NULL;
4154     Error *local_err = NULL;
4155     bool ro;
4156     int ret;
4157 
4158     bs = qmp_get_root_bs(device, errp);
4159     if (!bs) {
4160         return;
4161     }
4162 
4163     aio_context = bdrv_get_aio_context(bs);
4164     aio_context_acquire(aio_context);
4165 
4166     image_bs = bdrv_lookup_bs(NULL, image_node_name, &local_err);
4167     if (local_err) {
4168         error_propagate(errp, local_err);
4169         goto out;
4170     }
4171 
4172     if (!image_bs) {
4173         error_setg(errp, "image file not found");
4174         goto out;
4175     }
4176 
4177     if (bdrv_find_base(image_bs) == image_bs) {
4178         error_setg(errp, "not allowing backing file change on an image "
4179                          "without a backing file");
4180         goto out;
4181     }
4182 
4183     /* even though we are not necessarily operating on bs, we need it to
4184      * determine if block ops are currently prohibited on the chain */
4185     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_CHANGE, errp)) {
4186         goto out;
4187     }
4188 
4189     /* final sanity check */
4190     if (!bdrv_chain_contains(bs, image_bs)) {
4191         error_setg(errp, "'%s' and image file are not in the same chain",
4192                    device);
4193         goto out;
4194     }
4195 
4196     /* if not r/w, reopen to make r/w */
4197     ro = bdrv_is_read_only(image_bs);
4198 
4199     if (ro) {
4200         if (bdrv_reopen_set_read_only(image_bs, false, errp) != 0) {
4201             goto out;
4202         }
4203     }
4204 
4205     ret = bdrv_change_backing_file(image_bs, backing_file,
4206                                image_bs->drv ? image_bs->drv->format_name : "");
4207 
4208     if (ret < 0) {
4209         error_setg_errno(errp, -ret, "Could not change backing file to '%s'",
4210                          backing_file);
4211         /* don't exit here, so we can try to restore open flags if
4212          * appropriate */
4213     }
4214 
4215     if (ro) {
4216         bdrv_reopen_set_read_only(image_bs, true, &local_err);
4217         error_propagate(errp, local_err);
4218     }
4219 
4220 out:
4221     aio_context_release(aio_context);
4222 }
4223 
4224 void hmp_drive_add_node(Monitor *mon, const char *optstr)
4225 {
4226     QemuOpts *opts;
4227     QDict *qdict;
4228     Error *local_err = NULL;
4229 
4230     opts = qemu_opts_parse_noisily(&qemu_drive_opts, optstr, false);
4231     if (!opts) {
4232         return;
4233     }
4234 
4235     qdict = qemu_opts_to_qdict(opts, NULL);
4236 
4237     if (!qdict_get_try_str(qdict, "node-name")) {
4238         qobject_unref(qdict);
4239         error_report("'node-name' needs to be specified");
4240         goto out;
4241     }
4242 
4243     BlockDriverState *bs = bds_tree_init(qdict, &local_err);
4244     if (!bs) {
4245         error_report_err(local_err);
4246         goto out;
4247     }
4248 
4249     QTAILQ_INSERT_TAIL(&monitor_bdrv_states, bs, monitor_list);
4250 
4251 out:
4252     qemu_opts_del(opts);
4253 }
4254 
4255 void qmp_blockdev_add(BlockdevOptions *options, Error **errp)
4256 {
4257     BlockDriverState *bs;
4258     QObject *obj;
4259     Visitor *v = qobject_output_visitor_new(&obj);
4260     QDict *qdict;
4261     Error *local_err = NULL;
4262 
4263     visit_type_BlockdevOptions(v, NULL, &options, &local_err);
4264     if (local_err) {
4265         error_propagate(errp, local_err);
4266         goto fail;
4267     }
4268 
4269     visit_complete(v, &obj);
4270     qdict = qobject_to(QDict, obj);
4271 
4272     qdict_flatten(qdict);
4273 
4274     if (!qdict_get_try_str(qdict, "node-name")) {
4275         error_setg(errp, "'node-name' must be specified for the root node");
4276         goto fail;
4277     }
4278 
4279     bs = bds_tree_init(qdict, errp);
4280     if (!bs) {
4281         goto fail;
4282     }
4283 
4284     QTAILQ_INSERT_TAIL(&monitor_bdrv_states, bs, monitor_list);
4285 
4286 fail:
4287     visit_free(v);
4288 }
4289 
4290 void qmp_blockdev_del(const char *node_name, Error **errp)
4291 {
4292     AioContext *aio_context;
4293     BlockDriverState *bs;
4294 
4295     bs = bdrv_find_node(node_name);
4296     if (!bs) {
4297         error_setg(errp, "Cannot find node %s", node_name);
4298         return;
4299     }
4300     if (bdrv_has_blk(bs)) {
4301         error_setg(errp, "Node %s is in use", node_name);
4302         return;
4303     }
4304     aio_context = bdrv_get_aio_context(bs);
4305     aio_context_acquire(aio_context);
4306 
4307     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_DRIVE_DEL, errp)) {
4308         goto out;
4309     }
4310 
4311     if (!QTAILQ_IN_USE(bs, monitor_list)) {
4312         error_setg(errp, "Node %s is not owned by the monitor",
4313                    bs->node_name);
4314         goto out;
4315     }
4316 
4317     if (bs->refcnt > 1) {
4318         error_setg(errp, "Block device %s is in use",
4319                    bdrv_get_device_or_node_name(bs));
4320         goto out;
4321     }
4322 
4323     QTAILQ_REMOVE(&monitor_bdrv_states, bs, monitor_list);
4324     bdrv_unref(bs);
4325 
4326 out:
4327     aio_context_release(aio_context);
4328 }
4329 
4330 static BdrvChild *bdrv_find_child(BlockDriverState *parent_bs,
4331                                   const char *child_name)
4332 {
4333     BdrvChild *child;
4334 
4335     QLIST_FOREACH(child, &parent_bs->children, next) {
4336         if (strcmp(child->name, child_name) == 0) {
4337             return child;
4338         }
4339     }
4340 
4341     return NULL;
4342 }
4343 
4344 void qmp_x_blockdev_change(const char *parent, bool has_child,
4345                            const char *child, bool has_node,
4346                            const char *node, Error **errp)
4347 {
4348     BlockDriverState *parent_bs, *new_bs = NULL;
4349     BdrvChild *p_child;
4350 
4351     parent_bs = bdrv_lookup_bs(parent, parent, errp);
4352     if (!parent_bs) {
4353         return;
4354     }
4355 
4356     if (has_child == has_node) {
4357         if (has_child) {
4358             error_setg(errp, "The parameters child and node are in conflict");
4359         } else {
4360             error_setg(errp, "Either child or node must be specified");
4361         }
4362         return;
4363     }
4364 
4365     if (has_child) {
4366         p_child = bdrv_find_child(parent_bs, child);
4367         if (!p_child) {
4368             error_setg(errp, "Node '%s' does not have child '%s'",
4369                        parent, child);
4370             return;
4371         }
4372         bdrv_del_child(parent_bs, p_child, errp);
4373     }
4374 
4375     if (has_node) {
4376         new_bs = bdrv_find_node(node);
4377         if (!new_bs) {
4378             error_setg(errp, "Node '%s' not found", node);
4379             return;
4380         }
4381         bdrv_add_child(parent_bs, new_bs, errp);
4382     }
4383 }
4384 
4385 BlockJobInfoList *qmp_query_block_jobs(Error **errp)
4386 {
4387     BlockJobInfoList *head = NULL, **p_next = &head;
4388     BlockJob *job;
4389 
4390     for (job = block_job_next(NULL); job; job = block_job_next(job)) {
4391         BlockJobInfoList *elem;
4392         AioContext *aio_context;
4393 
4394         if (block_job_is_internal(job)) {
4395             continue;
4396         }
4397         elem = g_new0(BlockJobInfoList, 1);
4398         aio_context = blk_get_aio_context(job->blk);
4399         aio_context_acquire(aio_context);
4400         elem->value = block_job_query(job, errp);
4401         aio_context_release(aio_context);
4402         if (!elem->value) {
4403             g_free(elem);
4404             qapi_free_BlockJobInfoList(head);
4405             return NULL;
4406         }
4407         *p_next = elem;
4408         p_next = &elem->next;
4409     }
4410 
4411     return head;
4412 }
4413 
4414 void qmp_x_blockdev_set_iothread(const char *node_name, StrOrNull *iothread,
4415                                  bool has_force, bool force, Error **errp)
4416 {
4417     AioContext *old_context;
4418     AioContext *new_context;
4419     BlockDriverState *bs;
4420 
4421     bs = bdrv_find_node(node_name);
4422     if (!bs) {
4423         error_setg(errp, "Cannot find node %s", node_name);
4424         return;
4425     }
4426 
4427     /* Protects against accidents. */
4428     if (!(has_force && force) && bdrv_has_blk(bs)) {
4429         error_setg(errp, "Node %s is associated with a BlockBackend and could "
4430                          "be in use (use force=true to override this check)",
4431                          node_name);
4432         return;
4433     }
4434 
4435     if (iothread->type == QTYPE_QSTRING) {
4436         IOThread *obj = iothread_by_id(iothread->u.s);
4437         if (!obj) {
4438             error_setg(errp, "Cannot find iothread %s", iothread->u.s);
4439             return;
4440         }
4441 
4442         new_context = iothread_get_aio_context(obj);
4443     } else {
4444         new_context = qemu_get_aio_context();
4445     }
4446 
4447     old_context = bdrv_get_aio_context(bs);
4448     aio_context_acquire(old_context);
4449 
4450     bdrv_set_aio_context(bs, new_context);
4451 
4452     aio_context_release(old_context);
4453 }
4454 
4455 void qmp_x_block_latency_histogram_set(
4456     const char *device,
4457     bool has_boundaries, uint64List *boundaries,
4458     bool has_boundaries_read, uint64List *boundaries_read,
4459     bool has_boundaries_write, uint64List *boundaries_write,
4460     bool has_boundaries_flush, uint64List *boundaries_flush,
4461     Error **errp)
4462 {
4463     BlockBackend *blk = blk_by_name(device);
4464     BlockAcctStats *stats;
4465     int ret;
4466 
4467     if (!blk) {
4468         error_setg(errp, "Device '%s' not found", device);
4469         return;
4470     }
4471     stats = blk_get_stats(blk);
4472 
4473     if (!has_boundaries && !has_boundaries_read && !has_boundaries_write &&
4474         !has_boundaries_flush)
4475     {
4476         block_latency_histograms_clear(stats);
4477         return;
4478     }
4479 
4480     if (has_boundaries || has_boundaries_read) {
4481         ret = block_latency_histogram_set(
4482             stats, BLOCK_ACCT_READ,
4483             has_boundaries_read ? boundaries_read : boundaries);
4484         if (ret) {
4485             error_setg(errp, "Device '%s' set read boundaries fail", device);
4486             return;
4487         }
4488     }
4489 
4490     if (has_boundaries || has_boundaries_write) {
4491         ret = block_latency_histogram_set(
4492             stats, BLOCK_ACCT_WRITE,
4493             has_boundaries_write ? boundaries_write : boundaries);
4494         if (ret) {
4495             error_setg(errp, "Device '%s' set write boundaries fail", device);
4496             return;
4497         }
4498     }
4499 
4500     if (has_boundaries || has_boundaries_flush) {
4501         ret = block_latency_histogram_set(
4502             stats, BLOCK_ACCT_FLUSH,
4503             has_boundaries_flush ? boundaries_flush : boundaries);
4504         if (ret) {
4505             error_setg(errp, "Device '%s' set flush boundaries fail", device);
4506             return;
4507         }
4508     }
4509 }
4510 
4511 QemuOptsList qemu_common_drive_opts = {
4512     .name = "drive",
4513     .head = QTAILQ_HEAD_INITIALIZER(qemu_common_drive_opts.head),
4514     .desc = {
4515         {
4516             .name = "snapshot",
4517             .type = QEMU_OPT_BOOL,
4518             .help = "enable/disable snapshot mode",
4519         },{
4520             .name = "aio",
4521             .type = QEMU_OPT_STRING,
4522             .help = "host AIO implementation (threads, native)",
4523         },{
4524             .name = BDRV_OPT_CACHE_WB,
4525             .type = QEMU_OPT_BOOL,
4526             .help = "Enable writeback mode",
4527         },{
4528             .name = "format",
4529             .type = QEMU_OPT_STRING,
4530             .help = "disk format (raw, qcow2, ...)",
4531         },{
4532             .name = "rerror",
4533             .type = QEMU_OPT_STRING,
4534             .help = "read error action",
4535         },{
4536             .name = "werror",
4537             .type = QEMU_OPT_STRING,
4538             .help = "write error action",
4539         },{
4540             .name = BDRV_OPT_READ_ONLY,
4541             .type = QEMU_OPT_BOOL,
4542             .help = "open drive file as read-only",
4543         },
4544 
4545         THROTTLE_OPTS,
4546 
4547         {
4548             .name = "throttling.group",
4549             .type = QEMU_OPT_STRING,
4550             .help = "name of the block throttling group",
4551         },{
4552             .name = "copy-on-read",
4553             .type = QEMU_OPT_BOOL,
4554             .help = "copy read data from backing file into image file",
4555         },{
4556             .name = "detect-zeroes",
4557             .type = QEMU_OPT_STRING,
4558             .help = "try to optimize zero writes (off, on, unmap)",
4559         },{
4560             .name = "stats-account-invalid",
4561             .type = QEMU_OPT_BOOL,
4562             .help = "whether to account for invalid I/O operations "
4563                     "in the statistics",
4564         },{
4565             .name = "stats-account-failed",
4566             .type = QEMU_OPT_BOOL,
4567             .help = "whether to account for failed I/O operations "
4568                     "in the statistics",
4569         },
4570         { /* end of list */ }
4571     },
4572 };
4573 
4574 QemuOptsList qemu_drive_opts = {
4575     .name = "drive",
4576     .head = QTAILQ_HEAD_INITIALIZER(qemu_drive_opts.head),
4577     .desc = {
4578         /*
4579          * no elements => accept any params
4580          * validation will happen later
4581          */
4582         { /* end of list */ }
4583     },
4584 };
4585