xref: /openbmc/qemu/blockdev.c (revision ac1d8878)
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 "sysemu/block-backend.h"
34 #include "sysemu/blockdev.h"
35 #include "hw/block/block.h"
36 #include "block/blockjob.h"
37 #include "block/throttle-groups.h"
38 #include "monitor/monitor.h"
39 #include "qemu/error-report.h"
40 #include "qemu/option.h"
41 #include "qemu/config-file.h"
42 #include "qapi/qmp/types.h"
43 #include "qapi-visit.h"
44 #include "qapi/qmp/qerror.h"
45 #include "qapi/qmp-output-visitor.h"
46 #include "qapi/util.h"
47 #include "sysemu/sysemu.h"
48 #include "block/block_int.h"
49 #include "qmp-commands.h"
50 #include "trace.h"
51 #include "sysemu/arch_init.h"
52 
53 static const char *const if_name[IF_COUNT] = {
54     [IF_NONE] = "none",
55     [IF_IDE] = "ide",
56     [IF_SCSI] = "scsi",
57     [IF_FLOPPY] = "floppy",
58     [IF_PFLASH] = "pflash",
59     [IF_MTD] = "mtd",
60     [IF_SD] = "sd",
61     [IF_VIRTIO] = "virtio",
62     [IF_XEN] = "xen",
63 };
64 
65 static int if_max_devs[IF_COUNT] = {
66     /*
67      * Do not change these numbers!  They govern how drive option
68      * index maps to unit and bus.  That mapping is ABI.
69      *
70      * All controllers used to imlement if=T drives need to support
71      * if_max_devs[T] units, for any T with if_max_devs[T] != 0.
72      * Otherwise, some index values map to "impossible" bus, unit
73      * values.
74      *
75      * For instance, if you change [IF_SCSI] to 255, -drive
76      * if=scsi,index=12 no longer means bus=1,unit=5, but
77      * bus=0,unit=12.  With an lsi53c895a controller (7 units max),
78      * the drive can't be set up.  Regression.
79      */
80     [IF_IDE] = 2,
81     [IF_SCSI] = 7,
82 };
83 
84 /**
85  * Boards may call this to offer board-by-board overrides
86  * of the default, global values.
87  */
88 void override_max_devs(BlockInterfaceType type, int max_devs)
89 {
90     BlockBackend *blk;
91     DriveInfo *dinfo;
92 
93     if (max_devs <= 0) {
94         return;
95     }
96 
97     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
98         dinfo = blk_legacy_dinfo(blk);
99         if (dinfo->type == type) {
100             fprintf(stderr, "Cannot override units-per-bus property of"
101                     " the %s interface, because a drive of that type has"
102                     " already been added.\n", if_name[type]);
103             g_assert_not_reached();
104         }
105     }
106 
107     if_max_devs[type] = max_devs;
108 }
109 
110 /*
111  * We automatically delete the drive when a device using it gets
112  * unplugged.  Questionable feature, but we can't just drop it.
113  * Device models call blockdev_mark_auto_del() to schedule the
114  * automatic deletion, and generic qdev code calls blockdev_auto_del()
115  * when deletion is actually safe.
116  */
117 void blockdev_mark_auto_del(BlockBackend *blk)
118 {
119     DriveInfo *dinfo = blk_legacy_dinfo(blk);
120     BlockDriverState *bs = blk_bs(blk);
121     AioContext *aio_context;
122 
123     if (!dinfo) {
124         return;
125     }
126 
127     if (bs) {
128         aio_context = bdrv_get_aio_context(bs);
129         aio_context_acquire(aio_context);
130 
131         if (bs->job) {
132             block_job_cancel(bs->job);
133         }
134 
135         aio_context_release(aio_context);
136     }
137 
138     dinfo->auto_del = 1;
139 }
140 
141 void blockdev_auto_del(BlockBackend *blk)
142 {
143     DriveInfo *dinfo = blk_legacy_dinfo(blk);
144 
145     if (dinfo && dinfo->auto_del) {
146         blk_unref(blk);
147     }
148 }
149 
150 /**
151  * Returns the current mapping of how many units per bus
152  * a particular interface can support.
153  *
154  *  A positive integer indicates n units per bus.
155  *  0 implies the mapping has not been established.
156  * -1 indicates an invalid BlockInterfaceType was given.
157  */
158 int drive_get_max_devs(BlockInterfaceType type)
159 {
160     if (type >= IF_IDE && type < IF_COUNT) {
161         return if_max_devs[type];
162     }
163 
164     return -1;
165 }
166 
167 static int drive_index_to_bus_id(BlockInterfaceType type, int index)
168 {
169     int max_devs = if_max_devs[type];
170     return max_devs ? index / max_devs : 0;
171 }
172 
173 static int drive_index_to_unit_id(BlockInterfaceType type, int index)
174 {
175     int max_devs = if_max_devs[type];
176     return max_devs ? index % max_devs : index;
177 }
178 
179 QemuOpts *drive_def(const char *optstr)
180 {
181     return qemu_opts_parse_noisily(qemu_find_opts("drive"), optstr, false);
182 }
183 
184 QemuOpts *drive_add(BlockInterfaceType type, int index, const char *file,
185                     const char *optstr)
186 {
187     QemuOpts *opts;
188 
189     opts = drive_def(optstr);
190     if (!opts) {
191         return NULL;
192     }
193     if (type != IF_DEFAULT) {
194         qemu_opt_set(opts, "if", if_name[type], &error_abort);
195     }
196     if (index >= 0) {
197         qemu_opt_set_number(opts, "index", index, &error_abort);
198     }
199     if (file)
200         qemu_opt_set(opts, "file", file, &error_abort);
201     return opts;
202 }
203 
204 DriveInfo *drive_get(BlockInterfaceType type, int bus, int unit)
205 {
206     BlockBackend *blk;
207     DriveInfo *dinfo;
208 
209     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
210         dinfo = blk_legacy_dinfo(blk);
211         if (dinfo && dinfo->type == type
212             && dinfo->bus == bus && dinfo->unit == unit) {
213             return dinfo;
214         }
215     }
216 
217     return NULL;
218 }
219 
220 bool drive_check_orphaned(void)
221 {
222     BlockBackend *blk;
223     DriveInfo *dinfo;
224     bool rs = false;
225 
226     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
227         dinfo = blk_legacy_dinfo(blk);
228         /* If dinfo->bdrv->dev is NULL, it has no device attached. */
229         /* Unless this is a default drive, this may be an oversight. */
230         if (!blk_get_attached_dev(blk) && !dinfo->is_default &&
231             dinfo->type != IF_NONE) {
232             fprintf(stderr, "Warning: Orphaned drive without device: "
233                     "id=%s,file=%s,if=%s,bus=%d,unit=%d\n",
234                     blk_name(blk), blk_bs(blk) ? blk_bs(blk)->filename : "",
235                     if_name[dinfo->type], dinfo->bus, dinfo->unit);
236             rs = true;
237         }
238     }
239 
240     return rs;
241 }
242 
243 DriveInfo *drive_get_by_index(BlockInterfaceType type, int index)
244 {
245     return drive_get(type,
246                      drive_index_to_bus_id(type, index),
247                      drive_index_to_unit_id(type, index));
248 }
249 
250 int drive_get_max_bus(BlockInterfaceType type)
251 {
252     int max_bus;
253     BlockBackend *blk;
254     DriveInfo *dinfo;
255 
256     max_bus = -1;
257     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
258         dinfo = blk_legacy_dinfo(blk);
259         if (dinfo && dinfo->type == type && dinfo->bus > max_bus) {
260             max_bus = dinfo->bus;
261         }
262     }
263     return max_bus;
264 }
265 
266 /* Get a block device.  This should only be used for single-drive devices
267    (e.g. SD/Floppy/MTD).  Multi-disk devices (scsi/ide) should use the
268    appropriate bus.  */
269 DriveInfo *drive_get_next(BlockInterfaceType type)
270 {
271     static int next_block_unit[IF_COUNT];
272 
273     return drive_get(type, 0, next_block_unit[type]++);
274 }
275 
276 static void bdrv_format_print(void *opaque, const char *name)
277 {
278     error_printf(" %s", name);
279 }
280 
281 typedef struct {
282     QEMUBH *bh;
283     BlockDriverState *bs;
284 } BDRVPutRefBH;
285 
286 static int parse_block_error_action(const char *buf, bool is_read, Error **errp)
287 {
288     if (!strcmp(buf, "ignore")) {
289         return BLOCKDEV_ON_ERROR_IGNORE;
290     } else if (!is_read && !strcmp(buf, "enospc")) {
291         return BLOCKDEV_ON_ERROR_ENOSPC;
292     } else if (!strcmp(buf, "stop")) {
293         return BLOCKDEV_ON_ERROR_STOP;
294     } else if (!strcmp(buf, "report")) {
295         return BLOCKDEV_ON_ERROR_REPORT;
296     } else {
297         error_setg(errp, "'%s' invalid %s error action",
298                    buf, is_read ? "read" : "write");
299         return -1;
300     }
301 }
302 
303 static bool parse_stats_intervals(BlockAcctStats *stats, QList *intervals,
304                                   Error **errp)
305 {
306     const QListEntry *entry;
307     for (entry = qlist_first(intervals); entry; entry = qlist_next(entry)) {
308         switch (qobject_type(entry->value)) {
309 
310         case QTYPE_QSTRING: {
311             unsigned long long length;
312             const char *str = qstring_get_str(qobject_to_qstring(entry->value));
313             if (parse_uint_full(str, &length, 10) == 0 &&
314                 length > 0 && length <= UINT_MAX) {
315                 block_acct_add_interval(stats, (unsigned) length);
316             } else {
317                 error_setg(errp, "Invalid interval length: %s", str);
318                 return false;
319             }
320             break;
321         }
322 
323         case QTYPE_QINT: {
324             int64_t length = qint_get_int(qobject_to_qint(entry->value));
325             if (length > 0 && length <= UINT_MAX) {
326                 block_acct_add_interval(stats, (unsigned) length);
327             } else {
328                 error_setg(errp, "Invalid interval length: %" PRId64, length);
329                 return false;
330             }
331             break;
332         }
333 
334         default:
335             error_setg(errp, "The specification of stats-intervals is invalid");
336             return false;
337         }
338     }
339     return true;
340 }
341 
342 static bool check_throttle_config(ThrottleConfig *cfg, Error **errp)
343 {
344     if (throttle_conflicting(cfg)) {
345         error_setg(errp, "bps/iops/max total values and read/write values"
346                          " cannot be used at the same time");
347         return false;
348     }
349 
350     if (!throttle_is_valid(cfg)) {
351         error_setg(errp, "bps/iops/maxs values must be 0 or greater");
352         return false;
353     }
354 
355     if (throttle_max_is_missing_limit(cfg)) {
356         error_setg(errp, "bps_max/iops_max require corresponding"
357                          " bps/iops values");
358         return false;
359     }
360 
361     return true;
362 }
363 
364 typedef enum { MEDIA_DISK, MEDIA_CDROM } DriveMediaType;
365 
366 /* All parameters but @opts are optional and may be set to NULL. */
367 static void extract_common_blockdev_options(QemuOpts *opts, int *bdrv_flags,
368     const char **throttling_group, ThrottleConfig *throttle_cfg,
369     BlockdevDetectZeroesOptions *detect_zeroes, Error **errp)
370 {
371     const char *discard;
372     Error *local_error = NULL;
373     const char *aio;
374 
375     if (bdrv_flags) {
376         if (!qemu_opt_get_bool(opts, "read-only", false)) {
377             *bdrv_flags |= BDRV_O_RDWR;
378         }
379         if (qemu_opt_get_bool(opts, "copy-on-read", false)) {
380             *bdrv_flags |= BDRV_O_COPY_ON_READ;
381         }
382 
383         if ((discard = qemu_opt_get(opts, "discard")) != NULL) {
384             if (bdrv_parse_discard_flags(discard, bdrv_flags) != 0) {
385                 error_setg(errp, "Invalid discard option");
386                 return;
387             }
388         }
389 
390         if (qemu_opt_get_bool(opts, BDRV_OPT_CACHE_WB, true)) {
391             *bdrv_flags |= BDRV_O_CACHE_WB;
392         }
393         if (qemu_opt_get_bool(opts, BDRV_OPT_CACHE_DIRECT, false)) {
394             *bdrv_flags |= BDRV_O_NOCACHE;
395         }
396         if (qemu_opt_get_bool(opts, BDRV_OPT_CACHE_NO_FLUSH, false)) {
397             *bdrv_flags |= BDRV_O_NO_FLUSH;
398         }
399 
400         if ((aio = qemu_opt_get(opts, "aio")) != NULL) {
401             if (!strcmp(aio, "native")) {
402                 *bdrv_flags |= BDRV_O_NATIVE_AIO;
403             } else if (!strcmp(aio, "threads")) {
404                 /* this is the default */
405             } else {
406                error_setg(errp, "invalid aio option");
407                return;
408             }
409         }
410     }
411 
412     /* disk I/O throttling */
413     if (throttling_group) {
414         *throttling_group = qemu_opt_get(opts, "throttling.group");
415     }
416 
417     if (throttle_cfg) {
418         memset(throttle_cfg, 0, sizeof(*throttle_cfg));
419         throttle_cfg->buckets[THROTTLE_BPS_TOTAL].avg =
420             qemu_opt_get_number(opts, "throttling.bps-total", 0);
421         throttle_cfg->buckets[THROTTLE_BPS_READ].avg  =
422             qemu_opt_get_number(opts, "throttling.bps-read", 0);
423         throttle_cfg->buckets[THROTTLE_BPS_WRITE].avg =
424             qemu_opt_get_number(opts, "throttling.bps-write", 0);
425         throttle_cfg->buckets[THROTTLE_OPS_TOTAL].avg =
426             qemu_opt_get_number(opts, "throttling.iops-total", 0);
427         throttle_cfg->buckets[THROTTLE_OPS_READ].avg =
428             qemu_opt_get_number(opts, "throttling.iops-read", 0);
429         throttle_cfg->buckets[THROTTLE_OPS_WRITE].avg =
430             qemu_opt_get_number(opts, "throttling.iops-write", 0);
431 
432         throttle_cfg->buckets[THROTTLE_BPS_TOTAL].max =
433             qemu_opt_get_number(opts, "throttling.bps-total-max", 0);
434         throttle_cfg->buckets[THROTTLE_BPS_READ].max  =
435             qemu_opt_get_number(opts, "throttling.bps-read-max", 0);
436         throttle_cfg->buckets[THROTTLE_BPS_WRITE].max =
437             qemu_opt_get_number(opts, "throttling.bps-write-max", 0);
438         throttle_cfg->buckets[THROTTLE_OPS_TOTAL].max =
439             qemu_opt_get_number(opts, "throttling.iops-total-max", 0);
440         throttle_cfg->buckets[THROTTLE_OPS_READ].max =
441             qemu_opt_get_number(opts, "throttling.iops-read-max", 0);
442         throttle_cfg->buckets[THROTTLE_OPS_WRITE].max =
443             qemu_opt_get_number(opts, "throttling.iops-write-max", 0);
444 
445         throttle_cfg->op_size =
446             qemu_opt_get_number(opts, "throttling.iops-size", 0);
447 
448         if (!check_throttle_config(throttle_cfg, errp)) {
449             return;
450         }
451     }
452 
453     if (detect_zeroes) {
454         *detect_zeroes =
455             qapi_enum_parse(BlockdevDetectZeroesOptions_lookup,
456                             qemu_opt_get(opts, "detect-zeroes"),
457                             BLOCKDEV_DETECT_ZEROES_OPTIONS__MAX,
458                             BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF,
459                             &local_error);
460         if (local_error) {
461             error_propagate(errp, local_error);
462             return;
463         }
464 
465         if (bdrv_flags &&
466             *detect_zeroes == BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP &&
467             !(*bdrv_flags & BDRV_O_UNMAP))
468         {
469             error_setg(errp, "setting detect-zeroes to unmap is not allowed "
470                              "without setting discard operation to unmap");
471             return;
472         }
473     }
474 }
475 
476 /* Takes the ownership of bs_opts */
477 static BlockBackend *blockdev_init(const char *file, QDict *bs_opts,
478                                    Error **errp)
479 {
480     const char *buf;
481     int bdrv_flags = 0;
482     int on_read_error, on_write_error;
483     bool account_invalid, account_failed;
484     BlockBackend *blk;
485     BlockDriverState *bs;
486     ThrottleConfig cfg;
487     int snapshot = 0;
488     Error *error = NULL;
489     QemuOpts *opts;
490     QDict *interval_dict = NULL;
491     QList *interval_list = NULL;
492     const char *id;
493     bool has_driver_specific_opts;
494     BlockdevDetectZeroesOptions detect_zeroes =
495         BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF;
496     const char *throttling_group = NULL;
497 
498     /* Check common options by copying from bs_opts to opts, all other options
499      * stay in bs_opts for processing by bdrv_open(). */
500     id = qdict_get_try_str(bs_opts, "id");
501     opts = qemu_opts_create(&qemu_common_drive_opts, id, 1, &error);
502     if (error) {
503         error_propagate(errp, error);
504         goto err_no_opts;
505     }
506 
507     qemu_opts_absorb_qdict(opts, bs_opts, &error);
508     if (error) {
509         error_propagate(errp, error);
510         goto early_err;
511     }
512 
513     if (id) {
514         qdict_del(bs_opts, "id");
515     }
516 
517     has_driver_specific_opts = !!qdict_size(bs_opts);
518 
519     /* extract parameters */
520     snapshot = qemu_opt_get_bool(opts, "snapshot", 0);
521 
522     account_invalid = qemu_opt_get_bool(opts, "stats-account-invalid", true);
523     account_failed = qemu_opt_get_bool(opts, "stats-account-failed", true);
524 
525     qdict_extract_subqdict(bs_opts, &interval_dict, "stats-intervals.");
526     qdict_array_split(interval_dict, &interval_list);
527 
528     if (qdict_size(interval_dict) != 0) {
529         error_setg(errp, "Invalid option stats-intervals.%s",
530                    qdict_first(interval_dict)->key);
531         goto early_err;
532     }
533 
534     extract_common_blockdev_options(opts, &bdrv_flags, &throttling_group, &cfg,
535                                     &detect_zeroes, &error);
536     if (error) {
537         error_propagate(errp, error);
538         goto early_err;
539     }
540 
541     if ((buf = qemu_opt_get(opts, "format")) != NULL) {
542         if (is_help_option(buf)) {
543             error_printf("Supported formats:");
544             bdrv_iterate_format(bdrv_format_print, NULL);
545             error_printf("\n");
546             goto early_err;
547         }
548 
549         if (qdict_haskey(bs_opts, "driver")) {
550             error_setg(errp, "Cannot specify both 'driver' and 'format'");
551             goto early_err;
552         }
553         qdict_put(bs_opts, "driver", qstring_from_str(buf));
554     }
555 
556     on_write_error = BLOCKDEV_ON_ERROR_ENOSPC;
557     if ((buf = qemu_opt_get(opts, "werror")) != NULL) {
558         on_write_error = parse_block_error_action(buf, 0, &error);
559         if (error) {
560             error_propagate(errp, error);
561             goto early_err;
562         }
563     }
564 
565     on_read_error = BLOCKDEV_ON_ERROR_REPORT;
566     if ((buf = qemu_opt_get(opts, "rerror")) != NULL) {
567         on_read_error = parse_block_error_action(buf, 1, &error);
568         if (error) {
569             error_propagate(errp, error);
570             goto early_err;
571         }
572     }
573 
574     if (snapshot) {
575         /* always use cache=unsafe with snapshot */
576         bdrv_flags &= ~BDRV_O_CACHE_MASK;
577         bdrv_flags |= (BDRV_O_SNAPSHOT|BDRV_O_CACHE_WB|BDRV_O_NO_FLUSH);
578     }
579 
580     /* init */
581     if ((!file || !*file) && !has_driver_specific_opts) {
582         BlockBackendRootState *blk_rs;
583 
584         blk = blk_new(qemu_opts_id(opts), errp);
585         if (!blk) {
586             goto early_err;
587         }
588 
589         blk_rs = blk_get_root_state(blk);
590         blk_rs->open_flags    = bdrv_flags;
591         blk_rs->read_only     = !(bdrv_flags & BDRV_O_RDWR);
592         blk_rs->detect_zeroes = detect_zeroes;
593 
594         if (throttle_enabled(&cfg)) {
595             if (!throttling_group) {
596                 throttling_group = blk_name(blk);
597             }
598             blk_rs->throttle_group = g_strdup(throttling_group);
599             blk_rs->throttle_state = throttle_group_incref(throttling_group);
600             blk_rs->throttle_state->cfg = cfg;
601         }
602 
603         QDECREF(bs_opts);
604     } else {
605         if (file && !*file) {
606             file = NULL;
607         }
608 
609         blk = blk_new_open(qemu_opts_id(opts), file, NULL, bs_opts, bdrv_flags,
610                            errp);
611         if (!blk) {
612             goto err_no_bs_opts;
613         }
614         bs = blk_bs(blk);
615 
616         bs->detect_zeroes = detect_zeroes;
617 
618         /* disk I/O throttling */
619         if (throttle_enabled(&cfg)) {
620             if (!throttling_group) {
621                 throttling_group = blk_name(blk);
622             }
623             bdrv_io_limits_enable(bs, throttling_group);
624             bdrv_set_io_limits(bs, &cfg);
625         }
626 
627         if (bdrv_key_required(bs)) {
628             autostart = 0;
629         }
630 
631         block_acct_init(blk_get_stats(blk), account_invalid, account_failed);
632 
633         if (!parse_stats_intervals(blk_get_stats(blk), interval_list, errp)) {
634             blk_unref(blk);
635             blk = NULL;
636             goto err_no_bs_opts;
637         }
638     }
639 
640     blk_set_on_error(blk, on_read_error, on_write_error);
641 
642 err_no_bs_opts:
643     qemu_opts_del(opts);
644     QDECREF(interval_dict);
645     QDECREF(interval_list);
646     return blk;
647 
648 early_err:
649     qemu_opts_del(opts);
650     QDECREF(interval_dict);
651     QDECREF(interval_list);
652 err_no_opts:
653     QDECREF(bs_opts);
654     return NULL;
655 }
656 
657 static QemuOptsList qemu_root_bds_opts;
658 
659 /* Takes the ownership of bs_opts */
660 static BlockDriverState *bds_tree_init(QDict *bs_opts, Error **errp)
661 {
662     BlockDriverState *bs;
663     QemuOpts *opts;
664     Error *local_error = NULL;
665     BlockdevDetectZeroesOptions detect_zeroes;
666     int ret;
667     int bdrv_flags = 0;
668 
669     opts = qemu_opts_create(&qemu_root_bds_opts, NULL, 1, errp);
670     if (!opts) {
671         goto fail;
672     }
673 
674     qemu_opts_absorb_qdict(opts, bs_opts, &local_error);
675     if (local_error) {
676         error_propagate(errp, local_error);
677         goto fail;
678     }
679 
680     extract_common_blockdev_options(opts, &bdrv_flags, NULL, NULL,
681                                     &detect_zeroes, &local_error);
682     if (local_error) {
683         error_propagate(errp, local_error);
684         goto fail;
685     }
686 
687     bs = NULL;
688     ret = bdrv_open(&bs, NULL, NULL, bs_opts, bdrv_flags, errp);
689     if (ret < 0) {
690         goto fail_no_bs_opts;
691     }
692 
693     bs->detect_zeroes = detect_zeroes;
694 
695 fail_no_bs_opts:
696     qemu_opts_del(opts);
697     return bs;
698 
699 fail:
700     qemu_opts_del(opts);
701     QDECREF(bs_opts);
702     return NULL;
703 }
704 
705 static void qemu_opt_rename(QemuOpts *opts, const char *from, const char *to,
706                             Error **errp)
707 {
708     const char *value;
709 
710     value = qemu_opt_get(opts, from);
711     if (value) {
712         if (qemu_opt_find(opts, to)) {
713             error_setg(errp, "'%s' and its alias '%s' can't be used at the "
714                        "same time", to, from);
715             return;
716         }
717     }
718 
719     /* rename all items in opts */
720     while ((value = qemu_opt_get(opts, from))) {
721         qemu_opt_set(opts, to, value, &error_abort);
722         qemu_opt_unset(opts, from);
723     }
724 }
725 
726 QemuOptsList qemu_legacy_drive_opts = {
727     .name = "drive",
728     .head = QTAILQ_HEAD_INITIALIZER(qemu_legacy_drive_opts.head),
729     .desc = {
730         {
731             .name = "bus",
732             .type = QEMU_OPT_NUMBER,
733             .help = "bus number",
734         },{
735             .name = "unit",
736             .type = QEMU_OPT_NUMBER,
737             .help = "unit number (i.e. lun for scsi)",
738         },{
739             .name = "index",
740             .type = QEMU_OPT_NUMBER,
741             .help = "index number",
742         },{
743             .name = "media",
744             .type = QEMU_OPT_STRING,
745             .help = "media type (disk, cdrom)",
746         },{
747             .name = "if",
748             .type = QEMU_OPT_STRING,
749             .help = "interface (ide, scsi, sd, mtd, floppy, pflash, virtio)",
750         },{
751             .name = "cyls",
752             .type = QEMU_OPT_NUMBER,
753             .help = "number of cylinders (ide disk geometry)",
754         },{
755             .name = "heads",
756             .type = QEMU_OPT_NUMBER,
757             .help = "number of heads (ide disk geometry)",
758         },{
759             .name = "secs",
760             .type = QEMU_OPT_NUMBER,
761             .help = "number of sectors (ide disk geometry)",
762         },{
763             .name = "trans",
764             .type = QEMU_OPT_STRING,
765             .help = "chs translation (auto, lba, none)",
766         },{
767             .name = "boot",
768             .type = QEMU_OPT_BOOL,
769             .help = "(deprecated, ignored)",
770         },{
771             .name = "addr",
772             .type = QEMU_OPT_STRING,
773             .help = "pci address (virtio only)",
774         },{
775             .name = "serial",
776             .type = QEMU_OPT_STRING,
777             .help = "disk serial number",
778         },{
779             .name = "file",
780             .type = QEMU_OPT_STRING,
781             .help = "file name",
782         },
783 
784         /* Options that are passed on, but have special semantics with -drive */
785         {
786             .name = "read-only",
787             .type = QEMU_OPT_BOOL,
788             .help = "open drive file as read-only",
789         },{
790             .name = "rerror",
791             .type = QEMU_OPT_STRING,
792             .help = "read error action",
793         },{
794             .name = "werror",
795             .type = QEMU_OPT_STRING,
796             .help = "write error action",
797         },{
798             .name = "copy-on-read",
799             .type = QEMU_OPT_BOOL,
800             .help = "copy read data from backing file into image file",
801         },
802 
803         { /* end of list */ }
804     },
805 };
806 
807 DriveInfo *drive_new(QemuOpts *all_opts, BlockInterfaceType block_default_type)
808 {
809     const char *value;
810     BlockBackend *blk;
811     DriveInfo *dinfo = NULL;
812     QDict *bs_opts;
813     QemuOpts *legacy_opts;
814     DriveMediaType media = MEDIA_DISK;
815     BlockInterfaceType type;
816     int cyls, heads, secs, translation;
817     int max_devs, bus_id, unit_id, index;
818     const char *devaddr;
819     const char *werror, *rerror;
820     bool read_only = false;
821     bool copy_on_read;
822     const char *serial;
823     const char *filename;
824     Error *local_err = NULL;
825     int i;
826 
827     /* Change legacy command line options into QMP ones */
828     static const struct {
829         const char *from;
830         const char *to;
831     } opt_renames[] = {
832         { "iops",           "throttling.iops-total" },
833         { "iops_rd",        "throttling.iops-read" },
834         { "iops_wr",        "throttling.iops-write" },
835 
836         { "bps",            "throttling.bps-total" },
837         { "bps_rd",         "throttling.bps-read" },
838         { "bps_wr",         "throttling.bps-write" },
839 
840         { "iops_max",       "throttling.iops-total-max" },
841         { "iops_rd_max",    "throttling.iops-read-max" },
842         { "iops_wr_max",    "throttling.iops-write-max" },
843 
844         { "bps_max",        "throttling.bps-total-max" },
845         { "bps_rd_max",     "throttling.bps-read-max" },
846         { "bps_wr_max",     "throttling.bps-write-max" },
847 
848         { "iops_size",      "throttling.iops-size" },
849 
850         { "group",          "throttling.group" },
851 
852         { "readonly",       "read-only" },
853     };
854 
855     for (i = 0; i < ARRAY_SIZE(opt_renames); i++) {
856         qemu_opt_rename(all_opts, opt_renames[i].from, opt_renames[i].to,
857                         &local_err);
858         if (local_err) {
859             error_report_err(local_err);
860             return NULL;
861         }
862     }
863 
864     value = qemu_opt_get(all_opts, "cache");
865     if (value) {
866         int flags = 0;
867 
868         if (bdrv_parse_cache_flags(value, &flags) != 0) {
869             error_report("invalid cache option");
870             return NULL;
871         }
872 
873         /* Specific options take precedence */
874         if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_WB)) {
875             qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_WB,
876                               !!(flags & BDRV_O_CACHE_WB), &error_abort);
877         }
878         if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_DIRECT)) {
879             qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_DIRECT,
880                               !!(flags & BDRV_O_NOCACHE), &error_abort);
881         }
882         if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_NO_FLUSH)) {
883             qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_NO_FLUSH,
884                               !!(flags & BDRV_O_NO_FLUSH), &error_abort);
885         }
886         qemu_opt_unset(all_opts, "cache");
887     }
888 
889     /* Get a QDict for processing the options */
890     bs_opts = qdict_new();
891     qemu_opts_to_qdict(all_opts, bs_opts);
892 
893     legacy_opts = qemu_opts_create(&qemu_legacy_drive_opts, NULL, 0,
894                                    &error_abort);
895     qemu_opts_absorb_qdict(legacy_opts, bs_opts, &local_err);
896     if (local_err) {
897         error_report_err(local_err);
898         goto fail;
899     }
900 
901     /* Deprecated option boot=[on|off] */
902     if (qemu_opt_get(legacy_opts, "boot") != NULL) {
903         fprintf(stderr, "qemu-kvm: boot=on|off is deprecated and will be "
904                 "ignored. Future versions will reject this parameter. Please "
905                 "update your scripts.\n");
906     }
907 
908     /* Media type */
909     value = qemu_opt_get(legacy_opts, "media");
910     if (value) {
911         if (!strcmp(value, "disk")) {
912             media = MEDIA_DISK;
913         } else if (!strcmp(value, "cdrom")) {
914             media = MEDIA_CDROM;
915             read_only = true;
916         } else {
917             error_report("'%s' invalid media", value);
918             goto fail;
919         }
920     }
921 
922     /* copy-on-read is disabled with a warning for read-only devices */
923     read_only |= qemu_opt_get_bool(legacy_opts, "read-only", false);
924     copy_on_read = qemu_opt_get_bool(legacy_opts, "copy-on-read", false);
925 
926     if (read_only && copy_on_read) {
927         error_report("warning: disabling copy-on-read on read-only drive");
928         copy_on_read = false;
929     }
930 
931     qdict_put(bs_opts, "read-only",
932               qstring_from_str(read_only ? "on" : "off"));
933     qdict_put(bs_opts, "copy-on-read",
934               qstring_from_str(copy_on_read ? "on" :"off"));
935 
936     /* Controller type */
937     value = qemu_opt_get(legacy_opts, "if");
938     if (value) {
939         for (type = 0;
940              type < IF_COUNT && strcmp(value, if_name[type]);
941              type++) {
942         }
943         if (type == IF_COUNT) {
944             error_report("unsupported bus type '%s'", value);
945             goto fail;
946         }
947     } else {
948         type = block_default_type;
949     }
950 
951     /* Geometry */
952     cyls  = qemu_opt_get_number(legacy_opts, "cyls", 0);
953     heads = qemu_opt_get_number(legacy_opts, "heads", 0);
954     secs  = qemu_opt_get_number(legacy_opts, "secs", 0);
955 
956     if (cyls || heads || secs) {
957         if (cyls < 1) {
958             error_report("invalid physical cyls number");
959             goto fail;
960         }
961         if (heads < 1) {
962             error_report("invalid physical heads number");
963             goto fail;
964         }
965         if (secs < 1) {
966             error_report("invalid physical secs number");
967             goto fail;
968         }
969     }
970 
971     translation = BIOS_ATA_TRANSLATION_AUTO;
972     value = qemu_opt_get(legacy_opts, "trans");
973     if (value != NULL) {
974         if (!cyls) {
975             error_report("'%s' trans must be used with cyls, heads and secs",
976                          value);
977             goto fail;
978         }
979         if (!strcmp(value, "none")) {
980             translation = BIOS_ATA_TRANSLATION_NONE;
981         } else if (!strcmp(value, "lba")) {
982             translation = BIOS_ATA_TRANSLATION_LBA;
983         } else if (!strcmp(value, "large")) {
984             translation = BIOS_ATA_TRANSLATION_LARGE;
985         } else if (!strcmp(value, "rechs")) {
986             translation = BIOS_ATA_TRANSLATION_RECHS;
987         } else if (!strcmp(value, "auto")) {
988             translation = BIOS_ATA_TRANSLATION_AUTO;
989         } else {
990             error_report("'%s' invalid translation type", value);
991             goto fail;
992         }
993     }
994 
995     if (media == MEDIA_CDROM) {
996         if (cyls || secs || heads) {
997             error_report("CHS can't be set with media=cdrom");
998             goto fail;
999         }
1000     }
1001 
1002     /* Device address specified by bus/unit or index.
1003      * If none was specified, try to find the first free one. */
1004     bus_id  = qemu_opt_get_number(legacy_opts, "bus", 0);
1005     unit_id = qemu_opt_get_number(legacy_opts, "unit", -1);
1006     index   = qemu_opt_get_number(legacy_opts, "index", -1);
1007 
1008     max_devs = if_max_devs[type];
1009 
1010     if (index != -1) {
1011         if (bus_id != 0 || unit_id != -1) {
1012             error_report("index cannot be used with bus and unit");
1013             goto fail;
1014         }
1015         bus_id = drive_index_to_bus_id(type, index);
1016         unit_id = drive_index_to_unit_id(type, index);
1017     }
1018 
1019     if (unit_id == -1) {
1020        unit_id = 0;
1021        while (drive_get(type, bus_id, unit_id) != NULL) {
1022            unit_id++;
1023            if (max_devs && unit_id >= max_devs) {
1024                unit_id -= max_devs;
1025                bus_id++;
1026            }
1027        }
1028     }
1029 
1030     if (max_devs && unit_id >= max_devs) {
1031         error_report("unit %d too big (max is %d)", unit_id, max_devs - 1);
1032         goto fail;
1033     }
1034 
1035     if (drive_get(type, bus_id, unit_id) != NULL) {
1036         error_report("drive with bus=%d, unit=%d (index=%d) exists",
1037                      bus_id, unit_id, index);
1038         goto fail;
1039     }
1040 
1041     /* Serial number */
1042     serial = qemu_opt_get(legacy_opts, "serial");
1043 
1044     /* no id supplied -> create one */
1045     if (qemu_opts_id(all_opts) == NULL) {
1046         char *new_id;
1047         const char *mediastr = "";
1048         if (type == IF_IDE || type == IF_SCSI) {
1049             mediastr = (media == MEDIA_CDROM) ? "-cd" : "-hd";
1050         }
1051         if (max_devs) {
1052             new_id = g_strdup_printf("%s%i%s%i", if_name[type], bus_id,
1053                                      mediastr, unit_id);
1054         } else {
1055             new_id = g_strdup_printf("%s%s%i", if_name[type],
1056                                      mediastr, unit_id);
1057         }
1058         qdict_put(bs_opts, "id", qstring_from_str(new_id));
1059         g_free(new_id);
1060     }
1061 
1062     /* Add virtio block device */
1063     devaddr = qemu_opt_get(legacy_opts, "addr");
1064     if (devaddr && type != IF_VIRTIO) {
1065         error_report("addr is not supported by this bus type");
1066         goto fail;
1067     }
1068 
1069     if (type == IF_VIRTIO) {
1070         QemuOpts *devopts;
1071         devopts = qemu_opts_create(qemu_find_opts("device"), NULL, 0,
1072                                    &error_abort);
1073         if (arch_type == QEMU_ARCH_S390X) {
1074             qemu_opt_set(devopts, "driver", "virtio-blk-ccw", &error_abort);
1075         } else {
1076             qemu_opt_set(devopts, "driver", "virtio-blk-pci", &error_abort);
1077         }
1078         qemu_opt_set(devopts, "drive", qdict_get_str(bs_opts, "id"),
1079                      &error_abort);
1080         if (devaddr) {
1081             qemu_opt_set(devopts, "addr", devaddr, &error_abort);
1082         }
1083     }
1084 
1085     filename = qemu_opt_get(legacy_opts, "file");
1086 
1087     /* Check werror/rerror compatibility with if=... */
1088     werror = qemu_opt_get(legacy_opts, "werror");
1089     if (werror != NULL) {
1090         if (type != IF_IDE && type != IF_SCSI && type != IF_VIRTIO &&
1091             type != IF_NONE) {
1092             error_report("werror is not supported by this bus type");
1093             goto fail;
1094         }
1095         qdict_put(bs_opts, "werror", qstring_from_str(werror));
1096     }
1097 
1098     rerror = qemu_opt_get(legacy_opts, "rerror");
1099     if (rerror != NULL) {
1100         if (type != IF_IDE && type != IF_VIRTIO && type != IF_SCSI &&
1101             type != IF_NONE) {
1102             error_report("rerror is not supported by this bus type");
1103             goto fail;
1104         }
1105         qdict_put(bs_opts, "rerror", qstring_from_str(rerror));
1106     }
1107 
1108     /* Actual block device init: Functionality shared with blockdev-add */
1109     blk = blockdev_init(filename, bs_opts, &local_err);
1110     bs_opts = NULL;
1111     if (!blk) {
1112         if (local_err) {
1113             error_report_err(local_err);
1114         }
1115         goto fail;
1116     } else {
1117         assert(!local_err);
1118     }
1119 
1120     /* Create legacy DriveInfo */
1121     dinfo = g_malloc0(sizeof(*dinfo));
1122     dinfo->opts = all_opts;
1123 
1124     dinfo->cyls = cyls;
1125     dinfo->heads = heads;
1126     dinfo->secs = secs;
1127     dinfo->trans = translation;
1128 
1129     dinfo->type = type;
1130     dinfo->bus = bus_id;
1131     dinfo->unit = unit_id;
1132     dinfo->devaddr = devaddr;
1133     dinfo->serial = g_strdup(serial);
1134 
1135     blk_set_legacy_dinfo(blk, dinfo);
1136 
1137     switch(type) {
1138     case IF_IDE:
1139     case IF_SCSI:
1140     case IF_XEN:
1141     case IF_NONE:
1142         dinfo->media_cd = media == MEDIA_CDROM;
1143         break;
1144     default:
1145         break;
1146     }
1147 
1148 fail:
1149     qemu_opts_del(legacy_opts);
1150     QDECREF(bs_opts);
1151     return dinfo;
1152 }
1153 
1154 void hmp_commit(Monitor *mon, const QDict *qdict)
1155 {
1156     const char *device = qdict_get_str(qdict, "device");
1157     BlockBackend *blk;
1158     int ret;
1159 
1160     if (!strcmp(device, "all")) {
1161         ret = bdrv_commit_all();
1162     } else {
1163         BlockDriverState *bs;
1164         AioContext *aio_context;
1165 
1166         blk = blk_by_name(device);
1167         if (!blk) {
1168             monitor_printf(mon, "Device '%s' not found\n", device);
1169             return;
1170         }
1171         if (!blk_is_available(blk)) {
1172             monitor_printf(mon, "Device '%s' has no medium\n", device);
1173             return;
1174         }
1175 
1176         bs = blk_bs(blk);
1177         aio_context = bdrv_get_aio_context(bs);
1178         aio_context_acquire(aio_context);
1179 
1180         ret = bdrv_commit(bs);
1181 
1182         aio_context_release(aio_context);
1183     }
1184     if (ret < 0) {
1185         monitor_printf(mon, "'commit' error for '%s': %s\n", device,
1186                        strerror(-ret));
1187     }
1188 }
1189 
1190 static void blockdev_do_action(TransactionActionKind type, void *data,
1191                                Error **errp)
1192 {
1193     TransactionAction action;
1194     TransactionActionList list;
1195 
1196     action.type = type;
1197     action.u.data = data;
1198     list.value = &action;
1199     list.next = NULL;
1200     qmp_transaction(&list, false, NULL, errp);
1201 }
1202 
1203 void qmp_blockdev_snapshot_sync(bool has_device, const char *device,
1204                                 bool has_node_name, const char *node_name,
1205                                 const char *snapshot_file,
1206                                 bool has_snapshot_node_name,
1207                                 const char *snapshot_node_name,
1208                                 bool has_format, const char *format,
1209                                 bool has_mode, NewImageMode mode, Error **errp)
1210 {
1211     BlockdevSnapshotSync snapshot = {
1212         .has_device = has_device,
1213         .device = (char *) device,
1214         .has_node_name = has_node_name,
1215         .node_name = (char *) node_name,
1216         .snapshot_file = (char *) snapshot_file,
1217         .has_snapshot_node_name = has_snapshot_node_name,
1218         .snapshot_node_name = (char *) snapshot_node_name,
1219         .has_format = has_format,
1220         .format = (char *) format,
1221         .has_mode = has_mode,
1222         .mode = mode,
1223     };
1224     blockdev_do_action(TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC,
1225                        &snapshot, errp);
1226 }
1227 
1228 void qmp_blockdev_snapshot(const char *node, const char *overlay,
1229                            Error **errp)
1230 {
1231     BlockdevSnapshot snapshot_data = {
1232         .node = (char *) node,
1233         .overlay = (char *) overlay
1234     };
1235 
1236     blockdev_do_action(TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT,
1237                        &snapshot_data, errp);
1238 }
1239 
1240 void qmp_blockdev_snapshot_internal_sync(const char *device,
1241                                          const char *name,
1242                                          Error **errp)
1243 {
1244     BlockdevSnapshotInternal snapshot = {
1245         .device = (char *) device,
1246         .name = (char *) name
1247     };
1248 
1249     blockdev_do_action(TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC,
1250                        &snapshot, errp);
1251 }
1252 
1253 SnapshotInfo *qmp_blockdev_snapshot_delete_internal_sync(const char *device,
1254                                                          bool has_id,
1255                                                          const char *id,
1256                                                          bool has_name,
1257                                                          const char *name,
1258                                                          Error **errp)
1259 {
1260     BlockDriverState *bs;
1261     BlockBackend *blk;
1262     AioContext *aio_context;
1263     QEMUSnapshotInfo sn;
1264     Error *local_err = NULL;
1265     SnapshotInfo *info = NULL;
1266     int ret;
1267 
1268     blk = blk_by_name(device);
1269     if (!blk) {
1270         error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
1271                   "Device '%s' not found", device);
1272         return NULL;
1273     }
1274 
1275     aio_context = blk_get_aio_context(blk);
1276     aio_context_acquire(aio_context);
1277 
1278     if (!has_id) {
1279         id = NULL;
1280     }
1281 
1282     if (!has_name) {
1283         name = NULL;
1284     }
1285 
1286     if (!id && !name) {
1287         error_setg(errp, "Name or id must be provided");
1288         goto out_aio_context;
1289     }
1290 
1291     if (!blk_is_available(blk)) {
1292         error_setg(errp, "Device '%s' has no medium", device);
1293         goto out_aio_context;
1294     }
1295     bs = blk_bs(blk);
1296 
1297     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_INTERNAL_SNAPSHOT_DELETE, errp)) {
1298         goto out_aio_context;
1299     }
1300 
1301     ret = bdrv_snapshot_find_by_id_and_name(bs, id, name, &sn, &local_err);
1302     if (local_err) {
1303         error_propagate(errp, local_err);
1304         goto out_aio_context;
1305     }
1306     if (!ret) {
1307         error_setg(errp,
1308                    "Snapshot with id '%s' and name '%s' does not exist on "
1309                    "device '%s'",
1310                    STR_OR_NULL(id), STR_OR_NULL(name), device);
1311         goto out_aio_context;
1312     }
1313 
1314     bdrv_snapshot_delete(bs, id, name, &local_err);
1315     if (local_err) {
1316         error_propagate(errp, local_err);
1317         goto out_aio_context;
1318     }
1319 
1320     aio_context_release(aio_context);
1321 
1322     info = g_new0(SnapshotInfo, 1);
1323     info->id = g_strdup(sn.id_str);
1324     info->name = g_strdup(sn.name);
1325     info->date_nsec = sn.date_nsec;
1326     info->date_sec = sn.date_sec;
1327     info->vm_state_size = sn.vm_state_size;
1328     info->vm_clock_nsec = sn.vm_clock_nsec % 1000000000;
1329     info->vm_clock_sec = sn.vm_clock_nsec / 1000000000;
1330 
1331     return info;
1332 
1333 out_aio_context:
1334     aio_context_release(aio_context);
1335     return NULL;
1336 }
1337 
1338 /**
1339  * block_dirty_bitmap_lookup:
1340  * Return a dirty bitmap (if present), after validating
1341  * the node reference and bitmap names.
1342  *
1343  * @node: The name of the BDS node to search for bitmaps
1344  * @name: The name of the bitmap to search for
1345  * @pbs: Output pointer for BDS lookup, if desired. Can be NULL.
1346  * @paio: Output pointer for aio_context acquisition, if desired. Can be NULL.
1347  * @errp: Output pointer for error information. Can be NULL.
1348  *
1349  * @return: A bitmap object on success, or NULL on failure.
1350  */
1351 static BdrvDirtyBitmap *block_dirty_bitmap_lookup(const char *node,
1352                                                   const char *name,
1353                                                   BlockDriverState **pbs,
1354                                                   AioContext **paio,
1355                                                   Error **errp)
1356 {
1357     BlockDriverState *bs;
1358     BdrvDirtyBitmap *bitmap;
1359     AioContext *aio_context;
1360 
1361     if (!node) {
1362         error_setg(errp, "Node cannot be NULL");
1363         return NULL;
1364     }
1365     if (!name) {
1366         error_setg(errp, "Bitmap name cannot be NULL");
1367         return NULL;
1368     }
1369     bs = bdrv_lookup_bs(node, node, NULL);
1370     if (!bs) {
1371         error_setg(errp, "Node '%s' not found", node);
1372         return NULL;
1373     }
1374 
1375     aio_context = bdrv_get_aio_context(bs);
1376     aio_context_acquire(aio_context);
1377 
1378     bitmap = bdrv_find_dirty_bitmap(bs, name);
1379     if (!bitmap) {
1380         error_setg(errp, "Dirty bitmap '%s' not found", name);
1381         goto fail;
1382     }
1383 
1384     if (pbs) {
1385         *pbs = bs;
1386     }
1387     if (paio) {
1388         *paio = aio_context;
1389     } else {
1390         aio_context_release(aio_context);
1391     }
1392 
1393     return bitmap;
1394 
1395  fail:
1396     aio_context_release(aio_context);
1397     return NULL;
1398 }
1399 
1400 /* New and old BlockDriverState structs for atomic group operations */
1401 
1402 typedef struct BlkActionState BlkActionState;
1403 
1404 /**
1405  * BlkActionOps:
1406  * Table of operations that define an Action.
1407  *
1408  * @instance_size: Size of state struct, in bytes.
1409  * @prepare: Prepare the work, must NOT be NULL.
1410  * @commit: Commit the changes, can be NULL.
1411  * @abort: Abort the changes on fail, can be NULL.
1412  * @clean: Clean up resources after all transaction actions have called
1413  *         commit() or abort(). Can be NULL.
1414  *
1415  * Only prepare() may fail. In a single transaction, only one of commit() or
1416  * abort() will be called. clean() will always be called if it is present.
1417  */
1418 typedef struct BlkActionOps {
1419     size_t instance_size;
1420     void (*prepare)(BlkActionState *common, Error **errp);
1421     void (*commit)(BlkActionState *common);
1422     void (*abort)(BlkActionState *common);
1423     void (*clean)(BlkActionState *common);
1424 } BlkActionOps;
1425 
1426 /**
1427  * BlkActionState:
1428  * Describes one Action's state within a Transaction.
1429  *
1430  * @action: QAPI-defined enum identifying which Action to perform.
1431  * @ops: Table of ActionOps this Action can perform.
1432  * @block_job_txn: Transaction which this action belongs to.
1433  * @entry: List membership for all Actions in this Transaction.
1434  *
1435  * This structure must be arranged as first member in a subclassed type,
1436  * assuming that the compiler will also arrange it to the same offsets as the
1437  * base class.
1438  */
1439 struct BlkActionState {
1440     TransactionAction *action;
1441     const BlkActionOps *ops;
1442     BlockJobTxn *block_job_txn;
1443     TransactionProperties *txn_props;
1444     QSIMPLEQ_ENTRY(BlkActionState) entry;
1445 };
1446 
1447 /* internal snapshot private data */
1448 typedef struct InternalSnapshotState {
1449     BlkActionState common;
1450     BlockDriverState *bs;
1451     AioContext *aio_context;
1452     QEMUSnapshotInfo sn;
1453     bool created;
1454 } InternalSnapshotState;
1455 
1456 
1457 static int action_check_completion_mode(BlkActionState *s, Error **errp)
1458 {
1459     if (s->txn_props->completion_mode != ACTION_COMPLETION_MODE_INDIVIDUAL) {
1460         error_setg(errp,
1461                    "Action '%s' does not support Transaction property "
1462                    "completion-mode = %s",
1463                    TransactionActionKind_lookup[s->action->type],
1464                    ActionCompletionMode_lookup[s->txn_props->completion_mode]);
1465         return -1;
1466     }
1467     return 0;
1468 }
1469 
1470 static void internal_snapshot_prepare(BlkActionState *common,
1471                                       Error **errp)
1472 {
1473     Error *local_err = NULL;
1474     const char *device;
1475     const char *name;
1476     BlockBackend *blk;
1477     BlockDriverState *bs;
1478     QEMUSnapshotInfo old_sn, *sn;
1479     bool ret;
1480     qemu_timeval tv;
1481     BlockdevSnapshotInternal *internal;
1482     InternalSnapshotState *state;
1483     int ret1;
1484 
1485     g_assert(common->action->type ==
1486              TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC);
1487     internal = common->action->u.blockdev_snapshot_internal_sync;
1488     state = DO_UPCAST(InternalSnapshotState, common, common);
1489 
1490     /* 1. parse input */
1491     device = internal->device;
1492     name = internal->name;
1493 
1494     /* 2. check for validation */
1495     if (action_check_completion_mode(common, errp) < 0) {
1496         return;
1497     }
1498 
1499     blk = blk_by_name(device);
1500     if (!blk) {
1501         error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
1502                   "Device '%s' not found", device);
1503         return;
1504     }
1505 
1506     /* AioContext is released in .clean() */
1507     state->aio_context = blk_get_aio_context(blk);
1508     aio_context_acquire(state->aio_context);
1509 
1510     if (!blk_is_available(blk)) {
1511         error_setg(errp, QERR_DEVICE_HAS_NO_MEDIUM, device);
1512         return;
1513     }
1514     bs = blk_bs(blk);
1515 
1516     state->bs = bs;
1517     bdrv_drained_begin(bs);
1518 
1519     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_INTERNAL_SNAPSHOT, errp)) {
1520         return;
1521     }
1522 
1523     if (bdrv_is_read_only(bs)) {
1524         error_setg(errp, "Device '%s' is read only", device);
1525         return;
1526     }
1527 
1528     if (!bdrv_can_snapshot(bs)) {
1529         error_setg(errp, "Block format '%s' used by device '%s' "
1530                    "does not support internal snapshots",
1531                    bs->drv->format_name, device);
1532         return;
1533     }
1534 
1535     if (!strlen(name)) {
1536         error_setg(errp, "Name is empty");
1537         return;
1538     }
1539 
1540     /* check whether a snapshot with name exist */
1541     ret = bdrv_snapshot_find_by_id_and_name(bs, NULL, name, &old_sn,
1542                                             &local_err);
1543     if (local_err) {
1544         error_propagate(errp, local_err);
1545         return;
1546     } else if (ret) {
1547         error_setg(errp,
1548                    "Snapshot with name '%s' already exists on device '%s'",
1549                    name, device);
1550         return;
1551     }
1552 
1553     /* 3. take the snapshot */
1554     sn = &state->sn;
1555     pstrcpy(sn->name, sizeof(sn->name), name);
1556     qemu_gettimeofday(&tv);
1557     sn->date_sec = tv.tv_sec;
1558     sn->date_nsec = tv.tv_usec * 1000;
1559     sn->vm_clock_nsec = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL);
1560 
1561     ret1 = bdrv_snapshot_create(bs, sn);
1562     if (ret1 < 0) {
1563         error_setg_errno(errp, -ret1,
1564                          "Failed to create snapshot '%s' on device '%s'",
1565                          name, device);
1566         return;
1567     }
1568 
1569     /* 4. succeed, mark a snapshot is created */
1570     state->created = true;
1571 }
1572 
1573 static void internal_snapshot_abort(BlkActionState *common)
1574 {
1575     InternalSnapshotState *state =
1576                              DO_UPCAST(InternalSnapshotState, common, common);
1577     BlockDriverState *bs = state->bs;
1578     QEMUSnapshotInfo *sn = &state->sn;
1579     Error *local_error = NULL;
1580 
1581     if (!state->created) {
1582         return;
1583     }
1584 
1585     if (bdrv_snapshot_delete(bs, sn->id_str, sn->name, &local_error) < 0) {
1586         error_report("Failed to delete snapshot with id '%s' and name '%s' on "
1587                      "device '%s' in abort: %s",
1588                      sn->id_str,
1589                      sn->name,
1590                      bdrv_get_device_name(bs),
1591                      error_get_pretty(local_error));
1592         error_free(local_error);
1593     }
1594 }
1595 
1596 static void internal_snapshot_clean(BlkActionState *common)
1597 {
1598     InternalSnapshotState *state = DO_UPCAST(InternalSnapshotState,
1599                                              common, common);
1600 
1601     if (state->aio_context) {
1602         if (state->bs) {
1603             bdrv_drained_end(state->bs);
1604         }
1605         aio_context_release(state->aio_context);
1606     }
1607 }
1608 
1609 /* external snapshot private data */
1610 typedef struct ExternalSnapshotState {
1611     BlkActionState common;
1612     BlockDriverState *old_bs;
1613     BlockDriverState *new_bs;
1614     AioContext *aio_context;
1615 } ExternalSnapshotState;
1616 
1617 static void external_snapshot_prepare(BlkActionState *common,
1618                                       Error **errp)
1619 {
1620     int flags = 0, ret;
1621     QDict *options = NULL;
1622     Error *local_err = NULL;
1623     /* Device and node name of the image to generate the snapshot from */
1624     const char *device;
1625     const char *node_name;
1626     /* Reference to the new image (for 'blockdev-snapshot') */
1627     const char *snapshot_ref;
1628     /* File name of the new image (for 'blockdev-snapshot-sync') */
1629     const char *new_image_file;
1630     ExternalSnapshotState *state =
1631                              DO_UPCAST(ExternalSnapshotState, common, common);
1632     TransactionAction *action = common->action;
1633 
1634     /* 'blockdev-snapshot' and 'blockdev-snapshot-sync' have similar
1635      * purpose but a different set of parameters */
1636     switch (action->type) {
1637     case TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT:
1638         {
1639             BlockdevSnapshot *s = action->u.blockdev_snapshot;
1640             device = s->node;
1641             node_name = s->node;
1642             new_image_file = NULL;
1643             snapshot_ref = s->overlay;
1644         }
1645         break;
1646     case TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC:
1647         {
1648             BlockdevSnapshotSync *s = action->u.blockdev_snapshot_sync;
1649             device = s->has_device ? s->device : NULL;
1650             node_name = s->has_node_name ? s->node_name : NULL;
1651             new_image_file = s->snapshot_file;
1652             snapshot_ref = NULL;
1653         }
1654         break;
1655     default:
1656         g_assert_not_reached();
1657     }
1658 
1659     /* start processing */
1660     if (action_check_completion_mode(common, errp) < 0) {
1661         return;
1662     }
1663 
1664     state->old_bs = bdrv_lookup_bs(device, node_name, errp);
1665     if (!state->old_bs) {
1666         return;
1667     }
1668 
1669     /* Acquire AioContext now so any threads operating on old_bs stop */
1670     state->aio_context = bdrv_get_aio_context(state->old_bs);
1671     aio_context_acquire(state->aio_context);
1672     bdrv_drained_begin(state->old_bs);
1673 
1674     if (!bdrv_is_inserted(state->old_bs)) {
1675         error_setg(errp, QERR_DEVICE_HAS_NO_MEDIUM, device);
1676         return;
1677     }
1678 
1679     if (bdrv_op_is_blocked(state->old_bs,
1680                            BLOCK_OP_TYPE_EXTERNAL_SNAPSHOT, errp)) {
1681         return;
1682     }
1683 
1684     if (!bdrv_is_read_only(state->old_bs)) {
1685         if (bdrv_flush(state->old_bs)) {
1686             error_setg(errp, QERR_IO_ERROR);
1687             return;
1688         }
1689     }
1690 
1691     if (!bdrv_is_first_non_filter(state->old_bs)) {
1692         error_setg(errp, QERR_FEATURE_DISABLED, "snapshot");
1693         return;
1694     }
1695 
1696     if (action->type == TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC) {
1697         BlockdevSnapshotSync *s = action->u.blockdev_snapshot_sync;
1698         const char *format = s->has_format ? s->format : "qcow2";
1699         enum NewImageMode mode;
1700         const char *snapshot_node_name =
1701             s->has_snapshot_node_name ? s->snapshot_node_name : NULL;
1702 
1703         if (node_name && !snapshot_node_name) {
1704             error_setg(errp, "New snapshot node name missing");
1705             return;
1706         }
1707 
1708         if (snapshot_node_name &&
1709             bdrv_lookup_bs(snapshot_node_name, snapshot_node_name, NULL)) {
1710             error_setg(errp, "New snapshot node name already in use");
1711             return;
1712         }
1713 
1714         flags = state->old_bs->open_flags;
1715 
1716         /* create new image w/backing file */
1717         mode = s->has_mode ? s->mode : NEW_IMAGE_MODE_ABSOLUTE_PATHS;
1718         if (mode != NEW_IMAGE_MODE_EXISTING) {
1719             bdrv_img_create(new_image_file, format,
1720                             state->old_bs->filename,
1721                             state->old_bs->drv->format_name,
1722                             NULL, -1, flags, &local_err, false);
1723             if (local_err) {
1724                 error_propagate(errp, local_err);
1725                 return;
1726             }
1727         }
1728 
1729         options = qdict_new();
1730         if (s->has_snapshot_node_name) {
1731             qdict_put(options, "node-name",
1732                       qstring_from_str(snapshot_node_name));
1733         }
1734         qdict_put(options, "driver", qstring_from_str(format));
1735 
1736         flags |= BDRV_O_NO_BACKING;
1737     }
1738 
1739     assert(state->new_bs == NULL);
1740     ret = bdrv_open(&state->new_bs, new_image_file, snapshot_ref, options,
1741                     flags, errp);
1742     /* We will manually add the backing_hd field to the bs later */
1743     if (ret != 0) {
1744         return;
1745     }
1746 
1747     if (state->new_bs->blk != NULL) {
1748         error_setg(errp, "The snapshot is already in use by %s",
1749                    blk_name(state->new_bs->blk));
1750         return;
1751     }
1752 
1753     if (bdrv_op_is_blocked(state->new_bs, BLOCK_OP_TYPE_EXTERNAL_SNAPSHOT,
1754                            errp)) {
1755         return;
1756     }
1757 
1758     if (state->new_bs->backing != NULL) {
1759         error_setg(errp, "The snapshot already has a backing image");
1760         return;
1761     }
1762 
1763     if (!state->new_bs->drv->supports_backing) {
1764         error_setg(errp, "The snapshot does not support backing images");
1765     }
1766 }
1767 
1768 static void external_snapshot_commit(BlkActionState *common)
1769 {
1770     ExternalSnapshotState *state =
1771                              DO_UPCAST(ExternalSnapshotState, common, common);
1772 
1773     bdrv_set_aio_context(state->new_bs, state->aio_context);
1774 
1775     /* This removes our old bs and adds the new bs */
1776     bdrv_append(state->new_bs, state->old_bs);
1777     /* We don't need (or want) to use the transactional
1778      * bdrv_reopen_multiple() across all the entries at once, because we
1779      * don't want to abort all of them if one of them fails the reopen */
1780     bdrv_reopen(state->old_bs, state->old_bs->open_flags & ~BDRV_O_RDWR,
1781                 NULL);
1782 }
1783 
1784 static void external_snapshot_abort(BlkActionState *common)
1785 {
1786     ExternalSnapshotState *state =
1787                              DO_UPCAST(ExternalSnapshotState, common, common);
1788     if (state->new_bs) {
1789         bdrv_unref(state->new_bs);
1790     }
1791 }
1792 
1793 static void external_snapshot_clean(BlkActionState *common)
1794 {
1795     ExternalSnapshotState *state =
1796                              DO_UPCAST(ExternalSnapshotState, common, common);
1797     if (state->aio_context) {
1798         bdrv_drained_end(state->old_bs);
1799         aio_context_release(state->aio_context);
1800     }
1801 }
1802 
1803 typedef struct DriveBackupState {
1804     BlkActionState common;
1805     BlockDriverState *bs;
1806     AioContext *aio_context;
1807     BlockJob *job;
1808 } DriveBackupState;
1809 
1810 static void do_drive_backup(const char *device, const char *target,
1811                             bool has_format, const char *format,
1812                             enum MirrorSyncMode sync,
1813                             bool has_mode, enum NewImageMode mode,
1814                             bool has_speed, int64_t speed,
1815                             bool has_bitmap, const char *bitmap,
1816                             bool has_on_source_error,
1817                             BlockdevOnError on_source_error,
1818                             bool has_on_target_error,
1819                             BlockdevOnError on_target_error,
1820                             BlockJobTxn *txn, Error **errp);
1821 
1822 static void drive_backup_prepare(BlkActionState *common, Error **errp)
1823 {
1824     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1825     BlockBackend *blk;
1826     DriveBackup *backup;
1827     Error *local_err = NULL;
1828 
1829     assert(common->action->type == TRANSACTION_ACTION_KIND_DRIVE_BACKUP);
1830     backup = common->action->u.drive_backup;
1831 
1832     blk = blk_by_name(backup->device);
1833     if (!blk) {
1834         error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
1835                   "Device '%s' not found", backup->device);
1836         return;
1837     }
1838 
1839     if (!blk_is_available(blk)) {
1840         error_setg(errp, QERR_DEVICE_HAS_NO_MEDIUM, backup->device);
1841         return;
1842     }
1843 
1844     /* AioContext is released in .clean() */
1845     state->aio_context = blk_get_aio_context(blk);
1846     aio_context_acquire(state->aio_context);
1847     bdrv_drained_begin(blk_bs(blk));
1848     state->bs = blk_bs(blk);
1849 
1850     do_drive_backup(backup->device, backup->target,
1851                     backup->has_format, backup->format,
1852                     backup->sync,
1853                     backup->has_mode, backup->mode,
1854                     backup->has_speed, backup->speed,
1855                     backup->has_bitmap, backup->bitmap,
1856                     backup->has_on_source_error, backup->on_source_error,
1857                     backup->has_on_target_error, backup->on_target_error,
1858                     common->block_job_txn, &local_err);
1859     if (local_err) {
1860         error_propagate(errp, local_err);
1861         return;
1862     }
1863 
1864     state->job = state->bs->job;
1865 }
1866 
1867 static void drive_backup_abort(BlkActionState *common)
1868 {
1869     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1870     BlockDriverState *bs = state->bs;
1871 
1872     /* Only cancel if it's the job we started */
1873     if (bs && bs->job && bs->job == state->job) {
1874         block_job_cancel_sync(bs->job);
1875     }
1876 }
1877 
1878 static void drive_backup_clean(BlkActionState *common)
1879 {
1880     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1881 
1882     if (state->aio_context) {
1883         bdrv_drained_end(state->bs);
1884         aio_context_release(state->aio_context);
1885     }
1886 }
1887 
1888 typedef struct BlockdevBackupState {
1889     BlkActionState common;
1890     BlockDriverState *bs;
1891     BlockJob *job;
1892     AioContext *aio_context;
1893 } BlockdevBackupState;
1894 
1895 static void do_blockdev_backup(const char *device, const char *target,
1896                                enum MirrorSyncMode sync,
1897                                bool has_speed, int64_t speed,
1898                                bool has_on_source_error,
1899                                BlockdevOnError on_source_error,
1900                                bool has_on_target_error,
1901                                BlockdevOnError on_target_error,
1902                                BlockJobTxn *txn, Error **errp);
1903 
1904 static void blockdev_backup_prepare(BlkActionState *common, Error **errp)
1905 {
1906     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1907     BlockdevBackup *backup;
1908     BlockBackend *blk, *target;
1909     Error *local_err = NULL;
1910 
1911     assert(common->action->type == TRANSACTION_ACTION_KIND_BLOCKDEV_BACKUP);
1912     backup = common->action->u.blockdev_backup;
1913 
1914     blk = blk_by_name(backup->device);
1915     if (!blk) {
1916         error_setg(errp, "Device '%s' not found", backup->device);
1917         return;
1918     }
1919 
1920     if (!blk_is_available(blk)) {
1921         error_setg(errp, QERR_DEVICE_HAS_NO_MEDIUM, backup->device);
1922         return;
1923     }
1924 
1925     target = blk_by_name(backup->target);
1926     if (!target) {
1927         error_setg(errp, "Device '%s' not found", backup->target);
1928         return;
1929     }
1930 
1931     /* AioContext is released in .clean() */
1932     state->aio_context = blk_get_aio_context(blk);
1933     if (state->aio_context != blk_get_aio_context(target)) {
1934         state->aio_context = NULL;
1935         error_setg(errp, "Backup between two IO threads is not implemented");
1936         return;
1937     }
1938     aio_context_acquire(state->aio_context);
1939     state->bs = blk_bs(blk);
1940     bdrv_drained_begin(state->bs);
1941 
1942     do_blockdev_backup(backup->device, backup->target,
1943                        backup->sync,
1944                        backup->has_speed, backup->speed,
1945                        backup->has_on_source_error, backup->on_source_error,
1946                        backup->has_on_target_error, backup->on_target_error,
1947                        common->block_job_txn, &local_err);
1948     if (local_err) {
1949         error_propagate(errp, local_err);
1950         return;
1951     }
1952 
1953     state->job = state->bs->job;
1954 }
1955 
1956 static void blockdev_backup_abort(BlkActionState *common)
1957 {
1958     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1959     BlockDriverState *bs = state->bs;
1960 
1961     /* Only cancel if it's the job we started */
1962     if (bs && bs->job && bs->job == state->job) {
1963         block_job_cancel_sync(bs->job);
1964     }
1965 }
1966 
1967 static void blockdev_backup_clean(BlkActionState *common)
1968 {
1969     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1970 
1971     if (state->aio_context) {
1972         bdrv_drained_end(state->bs);
1973         aio_context_release(state->aio_context);
1974     }
1975 }
1976 
1977 typedef struct BlockDirtyBitmapState {
1978     BlkActionState common;
1979     BdrvDirtyBitmap *bitmap;
1980     BlockDriverState *bs;
1981     AioContext *aio_context;
1982     HBitmap *backup;
1983     bool prepared;
1984 } BlockDirtyBitmapState;
1985 
1986 static void block_dirty_bitmap_add_prepare(BlkActionState *common,
1987                                            Error **errp)
1988 {
1989     Error *local_err = NULL;
1990     BlockDirtyBitmapAdd *action;
1991     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
1992                                              common, common);
1993 
1994     if (action_check_completion_mode(common, errp) < 0) {
1995         return;
1996     }
1997 
1998     action = common->action->u.block_dirty_bitmap_add;
1999     /* AIO context taken and released within qmp_block_dirty_bitmap_add */
2000     qmp_block_dirty_bitmap_add(action->node, action->name,
2001                                action->has_granularity, action->granularity,
2002                                &local_err);
2003 
2004     if (!local_err) {
2005         state->prepared = true;
2006     } else {
2007         error_propagate(errp, local_err);
2008     }
2009 }
2010 
2011 static void block_dirty_bitmap_add_abort(BlkActionState *common)
2012 {
2013     BlockDirtyBitmapAdd *action;
2014     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2015                                              common, common);
2016 
2017     action = common->action->u.block_dirty_bitmap_add;
2018     /* Should not be able to fail: IF the bitmap was added via .prepare(),
2019      * then the node reference and bitmap name must have been valid.
2020      */
2021     if (state->prepared) {
2022         qmp_block_dirty_bitmap_remove(action->node, action->name, &error_abort);
2023     }
2024 }
2025 
2026 static void block_dirty_bitmap_clear_prepare(BlkActionState *common,
2027                                              Error **errp)
2028 {
2029     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2030                                              common, common);
2031     BlockDirtyBitmap *action;
2032 
2033     if (action_check_completion_mode(common, errp) < 0) {
2034         return;
2035     }
2036 
2037     action = common->action->u.block_dirty_bitmap_clear;
2038     state->bitmap = block_dirty_bitmap_lookup(action->node,
2039                                               action->name,
2040                                               &state->bs,
2041                                               &state->aio_context,
2042                                               errp);
2043     if (!state->bitmap) {
2044         return;
2045     }
2046 
2047     if (bdrv_dirty_bitmap_frozen(state->bitmap)) {
2048         error_setg(errp, "Cannot modify a frozen bitmap");
2049         return;
2050     } else if (!bdrv_dirty_bitmap_enabled(state->bitmap)) {
2051         error_setg(errp, "Cannot clear a disabled bitmap");
2052         return;
2053     }
2054 
2055     bdrv_clear_dirty_bitmap(state->bitmap, &state->backup);
2056     /* AioContext is released in .clean() */
2057 }
2058 
2059 static void block_dirty_bitmap_clear_abort(BlkActionState *common)
2060 {
2061     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2062                                              common, common);
2063 
2064     bdrv_undo_clear_dirty_bitmap(state->bitmap, state->backup);
2065 }
2066 
2067 static void block_dirty_bitmap_clear_commit(BlkActionState *common)
2068 {
2069     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2070                                              common, common);
2071 
2072     hbitmap_free(state->backup);
2073 }
2074 
2075 static void block_dirty_bitmap_clear_clean(BlkActionState *common)
2076 {
2077     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2078                                              common, common);
2079 
2080     if (state->aio_context) {
2081         aio_context_release(state->aio_context);
2082     }
2083 }
2084 
2085 static void abort_prepare(BlkActionState *common, Error **errp)
2086 {
2087     error_setg(errp, "Transaction aborted using Abort action");
2088 }
2089 
2090 static void abort_commit(BlkActionState *common)
2091 {
2092     g_assert_not_reached(); /* this action never succeeds */
2093 }
2094 
2095 static const BlkActionOps actions[] = {
2096     [TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT] = {
2097         .instance_size = sizeof(ExternalSnapshotState),
2098         .prepare  = external_snapshot_prepare,
2099         .commit   = external_snapshot_commit,
2100         .abort = external_snapshot_abort,
2101         .clean = external_snapshot_clean,
2102     },
2103     [TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC] = {
2104         .instance_size = sizeof(ExternalSnapshotState),
2105         .prepare  = external_snapshot_prepare,
2106         .commit   = external_snapshot_commit,
2107         .abort = external_snapshot_abort,
2108         .clean = external_snapshot_clean,
2109     },
2110     [TRANSACTION_ACTION_KIND_DRIVE_BACKUP] = {
2111         .instance_size = sizeof(DriveBackupState),
2112         .prepare = drive_backup_prepare,
2113         .abort = drive_backup_abort,
2114         .clean = drive_backup_clean,
2115     },
2116     [TRANSACTION_ACTION_KIND_BLOCKDEV_BACKUP] = {
2117         .instance_size = sizeof(BlockdevBackupState),
2118         .prepare = blockdev_backup_prepare,
2119         .abort = blockdev_backup_abort,
2120         .clean = blockdev_backup_clean,
2121     },
2122     [TRANSACTION_ACTION_KIND_ABORT] = {
2123         .instance_size = sizeof(BlkActionState),
2124         .prepare = abort_prepare,
2125         .commit = abort_commit,
2126     },
2127     [TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC] = {
2128         .instance_size = sizeof(InternalSnapshotState),
2129         .prepare  = internal_snapshot_prepare,
2130         .abort = internal_snapshot_abort,
2131         .clean = internal_snapshot_clean,
2132     },
2133     [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_ADD] = {
2134         .instance_size = sizeof(BlockDirtyBitmapState),
2135         .prepare = block_dirty_bitmap_add_prepare,
2136         .abort = block_dirty_bitmap_add_abort,
2137     },
2138     [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_CLEAR] = {
2139         .instance_size = sizeof(BlockDirtyBitmapState),
2140         .prepare = block_dirty_bitmap_clear_prepare,
2141         .commit = block_dirty_bitmap_clear_commit,
2142         .abort = block_dirty_bitmap_clear_abort,
2143         .clean = block_dirty_bitmap_clear_clean,
2144     }
2145 };
2146 
2147 /**
2148  * Allocate a TransactionProperties structure if necessary, and fill
2149  * that structure with desired defaults if they are unset.
2150  */
2151 static TransactionProperties *get_transaction_properties(
2152     TransactionProperties *props)
2153 {
2154     if (!props) {
2155         props = g_new0(TransactionProperties, 1);
2156     }
2157 
2158     if (!props->has_completion_mode) {
2159         props->has_completion_mode = true;
2160         props->completion_mode = ACTION_COMPLETION_MODE_INDIVIDUAL;
2161     }
2162 
2163     return props;
2164 }
2165 
2166 /*
2167  * 'Atomic' group operations.  The operations are performed as a set, and if
2168  * any fail then we roll back all operations in the group.
2169  */
2170 void qmp_transaction(TransactionActionList *dev_list,
2171                      bool has_props,
2172                      struct TransactionProperties *props,
2173                      Error **errp)
2174 {
2175     TransactionActionList *dev_entry = dev_list;
2176     BlockJobTxn *block_job_txn = NULL;
2177     BlkActionState *state, *next;
2178     Error *local_err = NULL;
2179 
2180     QSIMPLEQ_HEAD(snap_bdrv_states, BlkActionState) snap_bdrv_states;
2181     QSIMPLEQ_INIT(&snap_bdrv_states);
2182 
2183     /* Does this transaction get canceled as a group on failure?
2184      * If not, we don't really need to make a BlockJobTxn.
2185      */
2186     props = get_transaction_properties(props);
2187     if (props->completion_mode != ACTION_COMPLETION_MODE_INDIVIDUAL) {
2188         block_job_txn = block_job_txn_new();
2189     }
2190 
2191     /* drain all i/o before any operations */
2192     bdrv_drain_all();
2193 
2194     /* We don't do anything in this loop that commits us to the operations */
2195     while (NULL != dev_entry) {
2196         TransactionAction *dev_info = NULL;
2197         const BlkActionOps *ops;
2198 
2199         dev_info = dev_entry->value;
2200         dev_entry = dev_entry->next;
2201 
2202         assert(dev_info->type < ARRAY_SIZE(actions));
2203 
2204         ops = &actions[dev_info->type];
2205         assert(ops->instance_size > 0);
2206 
2207         state = g_malloc0(ops->instance_size);
2208         state->ops = ops;
2209         state->action = dev_info;
2210         state->block_job_txn = block_job_txn;
2211         state->txn_props = props;
2212         QSIMPLEQ_INSERT_TAIL(&snap_bdrv_states, state, entry);
2213 
2214         state->ops->prepare(state, &local_err);
2215         if (local_err) {
2216             error_propagate(errp, local_err);
2217             goto delete_and_fail;
2218         }
2219     }
2220 
2221     QSIMPLEQ_FOREACH(state, &snap_bdrv_states, entry) {
2222         if (state->ops->commit) {
2223             state->ops->commit(state);
2224         }
2225     }
2226 
2227     /* success */
2228     goto exit;
2229 
2230 delete_and_fail:
2231     /* failure, and it is all-or-none; roll back all operations */
2232     QSIMPLEQ_FOREACH(state, &snap_bdrv_states, entry) {
2233         if (state->ops->abort) {
2234             state->ops->abort(state);
2235         }
2236     }
2237 exit:
2238     QSIMPLEQ_FOREACH_SAFE(state, &snap_bdrv_states, entry, next) {
2239         if (state->ops->clean) {
2240             state->ops->clean(state);
2241         }
2242         g_free(state);
2243     }
2244     if (!has_props) {
2245         qapi_free_TransactionProperties(props);
2246     }
2247     block_job_txn_unref(block_job_txn);
2248 }
2249 
2250 void qmp_eject(const char *device, bool has_force, bool force, Error **errp)
2251 {
2252     Error *local_err = NULL;
2253 
2254     qmp_blockdev_open_tray(device, has_force, force, &local_err);
2255     if (local_err) {
2256         error_propagate(errp, local_err);
2257         return;
2258     }
2259 
2260     qmp_x_blockdev_remove_medium(device, errp);
2261 }
2262 
2263 void qmp_block_passwd(bool has_device, const char *device,
2264                       bool has_node_name, const char *node_name,
2265                       const char *password, Error **errp)
2266 {
2267     Error *local_err = NULL;
2268     BlockDriverState *bs;
2269     AioContext *aio_context;
2270 
2271     bs = bdrv_lookup_bs(has_device ? device : NULL,
2272                         has_node_name ? node_name : NULL,
2273                         &local_err);
2274     if (local_err) {
2275         error_propagate(errp, local_err);
2276         return;
2277     }
2278 
2279     aio_context = bdrv_get_aio_context(bs);
2280     aio_context_acquire(aio_context);
2281 
2282     bdrv_add_key(bs, password, errp);
2283 
2284     aio_context_release(aio_context);
2285 }
2286 
2287 void qmp_blockdev_open_tray(const char *device, bool has_force, bool force,
2288                             Error **errp)
2289 {
2290     BlockBackend *blk;
2291     bool locked;
2292 
2293     if (!has_force) {
2294         force = false;
2295     }
2296 
2297     blk = blk_by_name(device);
2298     if (!blk) {
2299         error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
2300                   "Device '%s' not found", device);
2301         return;
2302     }
2303 
2304     if (!blk_dev_has_removable_media(blk)) {
2305         error_setg(errp, "Device '%s' is not removable", device);
2306         return;
2307     }
2308 
2309     if (blk_dev_is_tray_open(blk)) {
2310         return;
2311     }
2312 
2313     locked = blk_dev_is_medium_locked(blk);
2314     if (locked) {
2315         blk_dev_eject_request(blk, force);
2316     }
2317 
2318     if (!locked || force) {
2319         blk_dev_change_media_cb(blk, false);
2320     }
2321 }
2322 
2323 void qmp_blockdev_close_tray(const char *device, Error **errp)
2324 {
2325     BlockBackend *blk;
2326 
2327     blk = blk_by_name(device);
2328     if (!blk) {
2329         error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
2330                   "Device '%s' not found", device);
2331         return;
2332     }
2333 
2334     if (!blk_dev_has_removable_media(blk)) {
2335         error_setg(errp, "Device '%s' is not removable", device);
2336         return;
2337     }
2338 
2339     if (!blk_dev_is_tray_open(blk)) {
2340         return;
2341     }
2342 
2343     blk_dev_change_media_cb(blk, true);
2344 }
2345 
2346 void qmp_x_blockdev_remove_medium(const char *device, Error **errp)
2347 {
2348     BlockBackend *blk;
2349     BlockDriverState *bs;
2350     AioContext *aio_context;
2351     bool has_device;
2352 
2353     blk = blk_by_name(device);
2354     if (!blk) {
2355         error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
2356                   "Device '%s' not found", device);
2357         return;
2358     }
2359 
2360     /* For BBs without a device, we can exchange the BDS tree at will */
2361     has_device = blk_get_attached_dev(blk);
2362 
2363     if (has_device && !blk_dev_has_removable_media(blk)) {
2364         error_setg(errp, "Device '%s' is not removable", device);
2365         return;
2366     }
2367 
2368     if (has_device && !blk_dev_is_tray_open(blk)) {
2369         error_setg(errp, "Tray of device '%s' is not open", device);
2370         return;
2371     }
2372 
2373     bs = blk_bs(blk);
2374     if (!bs) {
2375         return;
2376     }
2377 
2378     aio_context = bdrv_get_aio_context(bs);
2379     aio_context_acquire(aio_context);
2380 
2381     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_EJECT, errp)) {
2382         goto out;
2383     }
2384 
2385     /* This follows the convention established by bdrv_make_anon() */
2386     if (bs->device_list.tqe_prev) {
2387         QTAILQ_REMOVE(&bdrv_states, bs, device_list);
2388         bs->device_list.tqe_prev = NULL;
2389     }
2390 
2391     blk_remove_bs(blk);
2392 
2393 out:
2394     aio_context_release(aio_context);
2395 }
2396 
2397 static void qmp_blockdev_insert_anon_medium(const char *device,
2398                                             BlockDriverState *bs, Error **errp)
2399 {
2400     BlockBackend *blk;
2401     bool has_device;
2402 
2403     blk = blk_by_name(device);
2404     if (!blk) {
2405         error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
2406                   "Device '%s' not found", device);
2407         return;
2408     }
2409 
2410     /* For BBs without a device, we can exchange the BDS tree at will */
2411     has_device = blk_get_attached_dev(blk);
2412 
2413     if (has_device && !blk_dev_has_removable_media(blk)) {
2414         error_setg(errp, "Device '%s' is not removable", device);
2415         return;
2416     }
2417 
2418     if (has_device && !blk_dev_is_tray_open(blk)) {
2419         error_setg(errp, "Tray of device '%s' is not open", device);
2420         return;
2421     }
2422 
2423     if (blk_bs(blk)) {
2424         error_setg(errp, "There already is a medium in device '%s'", device);
2425         return;
2426     }
2427 
2428     blk_insert_bs(blk, bs);
2429 
2430     QTAILQ_INSERT_TAIL(&bdrv_states, bs, device_list);
2431 }
2432 
2433 void qmp_x_blockdev_insert_medium(const char *device, const char *node_name,
2434                                   Error **errp)
2435 {
2436     BlockDriverState *bs;
2437 
2438     bs = bdrv_find_node(node_name);
2439     if (!bs) {
2440         error_setg(errp, "Node '%s' not found", node_name);
2441         return;
2442     }
2443 
2444     if (bs->blk) {
2445         error_setg(errp, "Node '%s' is already in use by '%s'", node_name,
2446                    blk_name(bs->blk));
2447         return;
2448     }
2449 
2450     qmp_blockdev_insert_anon_medium(device, bs, errp);
2451 }
2452 
2453 void qmp_blockdev_change_medium(const char *device, const char *filename,
2454                                 bool has_format, const char *format,
2455                                 bool has_read_only,
2456                                 BlockdevChangeReadOnlyMode read_only,
2457                                 Error **errp)
2458 {
2459     BlockBackend *blk;
2460     BlockDriverState *medium_bs = NULL;
2461     int bdrv_flags, ret;
2462     QDict *options = NULL;
2463     Error *err = NULL;
2464 
2465     blk = blk_by_name(device);
2466     if (!blk) {
2467         error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
2468                   "Device '%s' not found", device);
2469         goto fail;
2470     }
2471 
2472     if (blk_bs(blk)) {
2473         blk_update_root_state(blk);
2474     }
2475 
2476     bdrv_flags = blk_get_open_flags_from_root_state(blk);
2477 
2478     if (!has_read_only) {
2479         read_only = BLOCKDEV_CHANGE_READ_ONLY_MODE_RETAIN;
2480     }
2481 
2482     switch (read_only) {
2483     case BLOCKDEV_CHANGE_READ_ONLY_MODE_RETAIN:
2484         break;
2485 
2486     case BLOCKDEV_CHANGE_READ_ONLY_MODE_READ_ONLY:
2487         bdrv_flags &= ~BDRV_O_RDWR;
2488         break;
2489 
2490     case BLOCKDEV_CHANGE_READ_ONLY_MODE_READ_WRITE:
2491         bdrv_flags |= BDRV_O_RDWR;
2492         break;
2493 
2494     default:
2495         abort();
2496     }
2497 
2498     if (has_format) {
2499         options = qdict_new();
2500         qdict_put(options, "driver", qstring_from_str(format));
2501     }
2502 
2503     assert(!medium_bs);
2504     ret = bdrv_open(&medium_bs, filename, NULL, options, bdrv_flags, errp);
2505     if (ret < 0) {
2506         goto fail;
2507     }
2508 
2509     blk_apply_root_state(blk, medium_bs);
2510 
2511     bdrv_add_key(medium_bs, NULL, &err);
2512     if (err) {
2513         error_propagate(errp, err);
2514         goto fail;
2515     }
2516 
2517     qmp_blockdev_open_tray(device, false, false, &err);
2518     if (err) {
2519         error_propagate(errp, err);
2520         goto fail;
2521     }
2522 
2523     qmp_x_blockdev_remove_medium(device, &err);
2524     if (err) {
2525         error_propagate(errp, err);
2526         goto fail;
2527     }
2528 
2529     qmp_blockdev_insert_anon_medium(device, medium_bs, &err);
2530     if (err) {
2531         error_propagate(errp, err);
2532         goto fail;
2533     }
2534 
2535     qmp_blockdev_close_tray(device, errp);
2536 
2537 fail:
2538     /* If the medium has been inserted, the device has its own reference, so
2539      * ours must be relinquished; and if it has not been inserted successfully,
2540      * the reference must be relinquished anyway */
2541     bdrv_unref(medium_bs);
2542 }
2543 
2544 /* throttling disk I/O limits */
2545 void qmp_block_set_io_throttle(const char *device, int64_t bps, int64_t bps_rd,
2546                                int64_t bps_wr,
2547                                int64_t iops,
2548                                int64_t iops_rd,
2549                                int64_t iops_wr,
2550                                bool has_bps_max,
2551                                int64_t bps_max,
2552                                bool has_bps_rd_max,
2553                                int64_t bps_rd_max,
2554                                bool has_bps_wr_max,
2555                                int64_t bps_wr_max,
2556                                bool has_iops_max,
2557                                int64_t iops_max,
2558                                bool has_iops_rd_max,
2559                                int64_t iops_rd_max,
2560                                bool has_iops_wr_max,
2561                                int64_t iops_wr_max,
2562                                bool has_iops_size,
2563                                int64_t iops_size,
2564                                bool has_group,
2565                                const char *group, Error **errp)
2566 {
2567     ThrottleConfig cfg;
2568     BlockDriverState *bs;
2569     BlockBackend *blk;
2570     AioContext *aio_context;
2571 
2572     blk = blk_by_name(device);
2573     if (!blk) {
2574         error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
2575                   "Device '%s' not found", device);
2576         return;
2577     }
2578 
2579     aio_context = blk_get_aio_context(blk);
2580     aio_context_acquire(aio_context);
2581 
2582     bs = blk_bs(blk);
2583     if (!bs) {
2584         error_setg(errp, "Device '%s' has no medium", device);
2585         goto out;
2586     }
2587 
2588     memset(&cfg, 0, sizeof(cfg));
2589     cfg.buckets[THROTTLE_BPS_TOTAL].avg = bps;
2590     cfg.buckets[THROTTLE_BPS_READ].avg  = bps_rd;
2591     cfg.buckets[THROTTLE_BPS_WRITE].avg = bps_wr;
2592 
2593     cfg.buckets[THROTTLE_OPS_TOTAL].avg = iops;
2594     cfg.buckets[THROTTLE_OPS_READ].avg  = iops_rd;
2595     cfg.buckets[THROTTLE_OPS_WRITE].avg = iops_wr;
2596 
2597     if (has_bps_max) {
2598         cfg.buckets[THROTTLE_BPS_TOTAL].max = bps_max;
2599     }
2600     if (has_bps_rd_max) {
2601         cfg.buckets[THROTTLE_BPS_READ].max = bps_rd_max;
2602     }
2603     if (has_bps_wr_max) {
2604         cfg.buckets[THROTTLE_BPS_WRITE].max = bps_wr_max;
2605     }
2606     if (has_iops_max) {
2607         cfg.buckets[THROTTLE_OPS_TOTAL].max = iops_max;
2608     }
2609     if (has_iops_rd_max) {
2610         cfg.buckets[THROTTLE_OPS_READ].max = iops_rd_max;
2611     }
2612     if (has_iops_wr_max) {
2613         cfg.buckets[THROTTLE_OPS_WRITE].max = iops_wr_max;
2614     }
2615 
2616     if (has_iops_size) {
2617         cfg.op_size = iops_size;
2618     }
2619 
2620     if (!check_throttle_config(&cfg, errp)) {
2621         goto out;
2622     }
2623 
2624     if (throttle_enabled(&cfg)) {
2625         /* Enable I/O limits if they're not enabled yet, otherwise
2626          * just update the throttling group. */
2627         if (!bs->throttle_state) {
2628             bdrv_io_limits_enable(bs, has_group ? group : device);
2629         } else if (has_group) {
2630             bdrv_io_limits_update_group(bs, group);
2631         }
2632         /* Set the new throttling configuration */
2633         bdrv_set_io_limits(bs, &cfg);
2634     } else if (bs->throttle_state) {
2635         /* If all throttling settings are set to 0, disable I/O limits */
2636         bdrv_io_limits_disable(bs);
2637     }
2638 
2639 out:
2640     aio_context_release(aio_context);
2641 }
2642 
2643 void qmp_block_dirty_bitmap_add(const char *node, const char *name,
2644                                 bool has_granularity, uint32_t granularity,
2645                                 Error **errp)
2646 {
2647     AioContext *aio_context;
2648     BlockDriverState *bs;
2649 
2650     if (!name || name[0] == '\0') {
2651         error_setg(errp, "Bitmap name cannot be empty");
2652         return;
2653     }
2654 
2655     bs = bdrv_lookup_bs(node, node, errp);
2656     if (!bs) {
2657         return;
2658     }
2659 
2660     aio_context = bdrv_get_aio_context(bs);
2661     aio_context_acquire(aio_context);
2662 
2663     if (has_granularity) {
2664         if (granularity < 512 || !is_power_of_2(granularity)) {
2665             error_setg(errp, "Granularity must be power of 2 "
2666                              "and at least 512");
2667             goto out;
2668         }
2669     } else {
2670         /* Default to cluster size, if available: */
2671         granularity = bdrv_get_default_bitmap_granularity(bs);
2672     }
2673 
2674     bdrv_create_dirty_bitmap(bs, granularity, name, errp);
2675 
2676  out:
2677     aio_context_release(aio_context);
2678 }
2679 
2680 void qmp_block_dirty_bitmap_remove(const char *node, const char *name,
2681                                    Error **errp)
2682 {
2683     AioContext *aio_context;
2684     BlockDriverState *bs;
2685     BdrvDirtyBitmap *bitmap;
2686 
2687     bitmap = block_dirty_bitmap_lookup(node, name, &bs, &aio_context, errp);
2688     if (!bitmap || !bs) {
2689         return;
2690     }
2691 
2692     if (bdrv_dirty_bitmap_frozen(bitmap)) {
2693         error_setg(errp,
2694                    "Bitmap '%s' is currently frozen and cannot be removed",
2695                    name);
2696         goto out;
2697     }
2698     bdrv_dirty_bitmap_make_anon(bitmap);
2699     bdrv_release_dirty_bitmap(bs, bitmap);
2700 
2701  out:
2702     aio_context_release(aio_context);
2703 }
2704 
2705 /**
2706  * Completely clear a bitmap, for the purposes of synchronizing a bitmap
2707  * immediately after a full backup operation.
2708  */
2709 void qmp_block_dirty_bitmap_clear(const char *node, const char *name,
2710                                   Error **errp)
2711 {
2712     AioContext *aio_context;
2713     BdrvDirtyBitmap *bitmap;
2714     BlockDriverState *bs;
2715 
2716     bitmap = block_dirty_bitmap_lookup(node, name, &bs, &aio_context, errp);
2717     if (!bitmap || !bs) {
2718         return;
2719     }
2720 
2721     if (bdrv_dirty_bitmap_frozen(bitmap)) {
2722         error_setg(errp,
2723                    "Bitmap '%s' is currently frozen and cannot be modified",
2724                    name);
2725         goto out;
2726     } else if (!bdrv_dirty_bitmap_enabled(bitmap)) {
2727         error_setg(errp,
2728                    "Bitmap '%s' is currently disabled and cannot be cleared",
2729                    name);
2730         goto out;
2731     }
2732 
2733     bdrv_clear_dirty_bitmap(bitmap, NULL);
2734 
2735  out:
2736     aio_context_release(aio_context);
2737 }
2738 
2739 void hmp_drive_del(Monitor *mon, const QDict *qdict)
2740 {
2741     const char *id = qdict_get_str(qdict, "id");
2742     BlockBackend *blk;
2743     BlockDriverState *bs;
2744     AioContext *aio_context;
2745     Error *local_err = NULL;
2746 
2747     blk = blk_by_name(id);
2748     if (!blk) {
2749         error_report("Device '%s' not found", id);
2750         return;
2751     }
2752 
2753     if (!blk_legacy_dinfo(blk)) {
2754         error_report("Deleting device added with blockdev-add"
2755                      " is not supported");
2756         return;
2757     }
2758 
2759     aio_context = blk_get_aio_context(blk);
2760     aio_context_acquire(aio_context);
2761 
2762     bs = blk_bs(blk);
2763     if (bs) {
2764         if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_DRIVE_DEL, &local_err)) {
2765             error_report_err(local_err);
2766             aio_context_release(aio_context);
2767             return;
2768         }
2769 
2770         bdrv_close(bs);
2771     }
2772 
2773     /* if we have a device attached to this BlockDriverState
2774      * then we need to make the drive anonymous until the device
2775      * can be removed.  If this is a drive with no device backing
2776      * then we can just get rid of the block driver state right here.
2777      */
2778     if (blk_get_attached_dev(blk)) {
2779         blk_hide_on_behalf_of_hmp_drive_del(blk);
2780         /* Further I/O must not pause the guest */
2781         blk_set_on_error(blk, BLOCKDEV_ON_ERROR_REPORT,
2782                          BLOCKDEV_ON_ERROR_REPORT);
2783     } else {
2784         blk_unref(blk);
2785     }
2786 
2787     aio_context_release(aio_context);
2788 }
2789 
2790 void qmp_block_resize(bool has_device, const char *device,
2791                       bool has_node_name, const char *node_name,
2792                       int64_t size, Error **errp)
2793 {
2794     Error *local_err = NULL;
2795     BlockDriverState *bs;
2796     AioContext *aio_context;
2797     int ret;
2798 
2799     bs = bdrv_lookup_bs(has_device ? device : NULL,
2800                         has_node_name ? node_name : NULL,
2801                         &local_err);
2802     if (local_err) {
2803         error_propagate(errp, local_err);
2804         return;
2805     }
2806 
2807     aio_context = bdrv_get_aio_context(bs);
2808     aio_context_acquire(aio_context);
2809 
2810     if (!bdrv_is_first_non_filter(bs)) {
2811         error_setg(errp, QERR_FEATURE_DISABLED, "resize");
2812         goto out;
2813     }
2814 
2815     if (size < 0) {
2816         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "size", "a >0 size");
2817         goto out;
2818     }
2819 
2820     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_RESIZE, NULL)) {
2821         error_setg(errp, QERR_DEVICE_IN_USE, device);
2822         goto out;
2823     }
2824 
2825     /* complete all in-flight operations before resizing the device */
2826     bdrv_drain_all();
2827 
2828     ret = bdrv_truncate(bs, size);
2829     switch (ret) {
2830     case 0:
2831         break;
2832     case -ENOMEDIUM:
2833         error_setg(errp, QERR_DEVICE_HAS_NO_MEDIUM, device);
2834         break;
2835     case -ENOTSUP:
2836         error_setg(errp, QERR_UNSUPPORTED);
2837         break;
2838     case -EACCES:
2839         error_setg(errp, "Device '%s' is read only", device);
2840         break;
2841     case -EBUSY:
2842         error_setg(errp, QERR_DEVICE_IN_USE, device);
2843         break;
2844     default:
2845         error_setg_errno(errp, -ret, "Could not resize");
2846         break;
2847     }
2848 
2849 out:
2850     aio_context_release(aio_context);
2851 }
2852 
2853 static void block_job_cb(void *opaque, int ret)
2854 {
2855     /* Note that this function may be executed from another AioContext besides
2856      * the QEMU main loop.  If you need to access anything that assumes the
2857      * QEMU global mutex, use a BH or introduce a mutex.
2858      */
2859 
2860     BlockDriverState *bs = opaque;
2861     const char *msg = NULL;
2862 
2863     trace_block_job_cb(bs, bs->job, ret);
2864 
2865     assert(bs->job);
2866 
2867     if (ret < 0) {
2868         msg = strerror(-ret);
2869     }
2870 
2871     if (block_job_is_cancelled(bs->job)) {
2872         block_job_event_cancelled(bs->job);
2873     } else {
2874         block_job_event_completed(bs->job, msg);
2875     }
2876 }
2877 
2878 void qmp_block_stream(const char *device,
2879                       bool has_base, const char *base,
2880                       bool has_backing_file, const char *backing_file,
2881                       bool has_speed, int64_t speed,
2882                       bool has_on_error, BlockdevOnError on_error,
2883                       Error **errp)
2884 {
2885     BlockBackend *blk;
2886     BlockDriverState *bs;
2887     BlockDriverState *base_bs = NULL;
2888     AioContext *aio_context;
2889     Error *local_err = NULL;
2890     const char *base_name = NULL;
2891 
2892     if (!has_on_error) {
2893         on_error = BLOCKDEV_ON_ERROR_REPORT;
2894     }
2895 
2896     blk = blk_by_name(device);
2897     if (!blk) {
2898         error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
2899                   "Device '%s' not found", device);
2900         return;
2901     }
2902 
2903     aio_context = blk_get_aio_context(blk);
2904     aio_context_acquire(aio_context);
2905 
2906     if (!blk_is_available(blk)) {
2907         error_setg(errp, "Device '%s' has no medium", device);
2908         goto out;
2909     }
2910     bs = blk_bs(blk);
2911 
2912     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_STREAM, errp)) {
2913         goto out;
2914     }
2915 
2916     if (has_base) {
2917         base_bs = bdrv_find_backing_image(bs, base);
2918         if (base_bs == NULL) {
2919             error_setg(errp, QERR_BASE_NOT_FOUND, base);
2920             goto out;
2921         }
2922         assert(bdrv_get_aio_context(base_bs) == aio_context);
2923         base_name = base;
2924     }
2925 
2926     /* if we are streaming the entire chain, the result will have no backing
2927      * file, and specifying one is therefore an error */
2928     if (base_bs == NULL && has_backing_file) {
2929         error_setg(errp, "backing file specified, but streaming the "
2930                          "entire chain");
2931         goto out;
2932     }
2933 
2934     /* backing_file string overrides base bs filename */
2935     base_name = has_backing_file ? backing_file : base_name;
2936 
2937     stream_start(bs, base_bs, base_name, has_speed ? speed : 0,
2938                  on_error, block_job_cb, bs, &local_err);
2939     if (local_err) {
2940         error_propagate(errp, local_err);
2941         goto out;
2942     }
2943 
2944     trace_qmp_block_stream(bs, bs->job);
2945 
2946 out:
2947     aio_context_release(aio_context);
2948 }
2949 
2950 void qmp_block_commit(const char *device,
2951                       bool has_base, const char *base,
2952                       bool has_top, const char *top,
2953                       bool has_backing_file, const char *backing_file,
2954                       bool has_speed, int64_t speed,
2955                       Error **errp)
2956 {
2957     BlockBackend *blk;
2958     BlockDriverState *bs;
2959     BlockDriverState *base_bs, *top_bs;
2960     AioContext *aio_context;
2961     Error *local_err = NULL;
2962     /* This will be part of the QMP command, if/when the
2963      * BlockdevOnError change for blkmirror makes it in
2964      */
2965     BlockdevOnError on_error = BLOCKDEV_ON_ERROR_REPORT;
2966 
2967     if (!has_speed) {
2968         speed = 0;
2969     }
2970 
2971     /* Important Note:
2972      *  libvirt relies on the DeviceNotFound error class in order to probe for
2973      *  live commit feature versions; for this to work, we must make sure to
2974      *  perform the device lookup before any generic errors that may occur in a
2975      *  scenario in which all optional arguments are omitted. */
2976     blk = blk_by_name(device);
2977     if (!blk) {
2978         error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
2979                   "Device '%s' not found", device);
2980         return;
2981     }
2982 
2983     aio_context = blk_get_aio_context(blk);
2984     aio_context_acquire(aio_context);
2985 
2986     if (!blk_is_available(blk)) {
2987         error_setg(errp, "Device '%s' has no medium", device);
2988         goto out;
2989     }
2990     bs = blk_bs(blk);
2991 
2992     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_COMMIT_SOURCE, errp)) {
2993         goto out;
2994     }
2995 
2996     /* default top_bs is the active layer */
2997     top_bs = bs;
2998 
2999     if (has_top && top) {
3000         if (strcmp(bs->filename, top) != 0) {
3001             top_bs = bdrv_find_backing_image(bs, top);
3002         }
3003     }
3004 
3005     if (top_bs == NULL) {
3006         error_setg(errp, "Top image file %s not found", top ? top : "NULL");
3007         goto out;
3008     }
3009 
3010     assert(bdrv_get_aio_context(top_bs) == aio_context);
3011 
3012     if (has_base && base) {
3013         base_bs = bdrv_find_backing_image(top_bs, base);
3014     } else {
3015         base_bs = bdrv_find_base(top_bs);
3016     }
3017 
3018     if (base_bs == NULL) {
3019         error_setg(errp, QERR_BASE_NOT_FOUND, base ? base : "NULL");
3020         goto out;
3021     }
3022 
3023     assert(bdrv_get_aio_context(base_bs) == aio_context);
3024 
3025     if (bdrv_op_is_blocked(base_bs, BLOCK_OP_TYPE_COMMIT_TARGET, errp)) {
3026         goto out;
3027     }
3028 
3029     /* Do not allow attempts to commit an image into itself */
3030     if (top_bs == base_bs) {
3031         error_setg(errp, "cannot commit an image into itself");
3032         goto out;
3033     }
3034 
3035     if (top_bs == bs) {
3036         if (has_backing_file) {
3037             error_setg(errp, "'backing-file' specified,"
3038                              " but 'top' is the active layer");
3039             goto out;
3040         }
3041         commit_active_start(bs, base_bs, speed, on_error, block_job_cb,
3042                             bs, &local_err);
3043     } else {
3044         commit_start(bs, base_bs, top_bs, speed, on_error, block_job_cb, bs,
3045                      has_backing_file ? backing_file : NULL, &local_err);
3046     }
3047     if (local_err != NULL) {
3048         error_propagate(errp, local_err);
3049         goto out;
3050     }
3051 
3052 out:
3053     aio_context_release(aio_context);
3054 }
3055 
3056 static void do_drive_backup(const char *device, const char *target,
3057                             bool has_format, const char *format,
3058                             enum MirrorSyncMode sync,
3059                             bool has_mode, enum NewImageMode mode,
3060                             bool has_speed, int64_t speed,
3061                             bool has_bitmap, const char *bitmap,
3062                             bool has_on_source_error,
3063                             BlockdevOnError on_source_error,
3064                             bool has_on_target_error,
3065                             BlockdevOnError on_target_error,
3066                             BlockJobTxn *txn, Error **errp)
3067 {
3068     BlockBackend *blk;
3069     BlockDriverState *bs;
3070     BlockDriverState *target_bs;
3071     BlockDriverState *source = NULL;
3072     BdrvDirtyBitmap *bmap = NULL;
3073     AioContext *aio_context;
3074     QDict *options = NULL;
3075     Error *local_err = NULL;
3076     int flags;
3077     int64_t size;
3078     int ret;
3079 
3080     if (!has_speed) {
3081         speed = 0;
3082     }
3083     if (!has_on_source_error) {
3084         on_source_error = BLOCKDEV_ON_ERROR_REPORT;
3085     }
3086     if (!has_on_target_error) {
3087         on_target_error = BLOCKDEV_ON_ERROR_REPORT;
3088     }
3089     if (!has_mode) {
3090         mode = NEW_IMAGE_MODE_ABSOLUTE_PATHS;
3091     }
3092 
3093     blk = blk_by_name(device);
3094     if (!blk) {
3095         error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
3096                   "Device '%s' not found", device);
3097         return;
3098     }
3099 
3100     aio_context = blk_get_aio_context(blk);
3101     aio_context_acquire(aio_context);
3102 
3103     /* Although backup_run has this check too, we need to use bs->drv below, so
3104      * do an early check redundantly. */
3105     if (!blk_is_available(blk)) {
3106         error_setg(errp, QERR_DEVICE_HAS_NO_MEDIUM, device);
3107         goto out;
3108     }
3109     bs = blk_bs(blk);
3110 
3111     if (!has_format) {
3112         format = mode == NEW_IMAGE_MODE_EXISTING ? NULL : bs->drv->format_name;
3113     }
3114 
3115     /* Early check to avoid creating target */
3116     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_BACKUP_SOURCE, errp)) {
3117         goto out;
3118     }
3119 
3120     flags = bs->open_flags | BDRV_O_RDWR;
3121 
3122     /* See if we have a backing HD we can use to create our new image
3123      * on top of. */
3124     if (sync == MIRROR_SYNC_MODE_TOP) {
3125         source = backing_bs(bs);
3126         if (!source) {
3127             sync = MIRROR_SYNC_MODE_FULL;
3128         }
3129     }
3130     if (sync == MIRROR_SYNC_MODE_NONE) {
3131         source = bs;
3132     }
3133 
3134     size = bdrv_getlength(bs);
3135     if (size < 0) {
3136         error_setg_errno(errp, -size, "bdrv_getlength failed");
3137         goto out;
3138     }
3139 
3140     if (mode != NEW_IMAGE_MODE_EXISTING) {
3141         assert(format);
3142         if (source) {
3143             bdrv_img_create(target, format, source->filename,
3144                             source->drv->format_name, NULL,
3145                             size, flags, &local_err, false);
3146         } else {
3147             bdrv_img_create(target, format, NULL, NULL, NULL,
3148                             size, flags, &local_err, false);
3149         }
3150     }
3151 
3152     if (local_err) {
3153         error_propagate(errp, local_err);
3154         goto out;
3155     }
3156 
3157     if (format) {
3158         options = qdict_new();
3159         qdict_put(options, "driver", qstring_from_str(format));
3160     }
3161 
3162     target_bs = NULL;
3163     ret = bdrv_open(&target_bs, target, NULL, options, flags, &local_err);
3164     if (ret < 0) {
3165         error_propagate(errp, local_err);
3166         goto out;
3167     }
3168 
3169     bdrv_set_aio_context(target_bs, aio_context);
3170 
3171     if (has_bitmap) {
3172         bmap = bdrv_find_dirty_bitmap(bs, bitmap);
3173         if (!bmap) {
3174             error_setg(errp, "Bitmap '%s' could not be found", bitmap);
3175             bdrv_unref(target_bs);
3176             goto out;
3177         }
3178     }
3179 
3180     backup_start(bs, target_bs, speed, sync, bmap,
3181                  on_source_error, on_target_error,
3182                  block_job_cb, bs, txn, &local_err);
3183     if (local_err != NULL) {
3184         bdrv_unref(target_bs);
3185         error_propagate(errp, local_err);
3186         goto out;
3187     }
3188 
3189 out:
3190     aio_context_release(aio_context);
3191 }
3192 
3193 void qmp_drive_backup(const char *device, const char *target,
3194                       bool has_format, const char *format,
3195                       enum MirrorSyncMode sync,
3196                       bool has_mode, enum NewImageMode mode,
3197                       bool has_speed, int64_t speed,
3198                       bool has_bitmap, const char *bitmap,
3199                       bool has_on_source_error, BlockdevOnError on_source_error,
3200                       bool has_on_target_error, BlockdevOnError on_target_error,
3201                       Error **errp)
3202 {
3203     return do_drive_backup(device, target, has_format, format, sync,
3204                            has_mode, mode, has_speed, speed,
3205                            has_bitmap, bitmap,
3206                            has_on_source_error, on_source_error,
3207                            has_on_target_error, on_target_error,
3208                            NULL, errp);
3209 }
3210 
3211 BlockDeviceInfoList *qmp_query_named_block_nodes(Error **errp)
3212 {
3213     return bdrv_named_nodes_list(errp);
3214 }
3215 
3216 void do_blockdev_backup(const char *device, const char *target,
3217                          enum MirrorSyncMode sync,
3218                          bool has_speed, int64_t speed,
3219                          bool has_on_source_error,
3220                          BlockdevOnError on_source_error,
3221                          bool has_on_target_error,
3222                          BlockdevOnError on_target_error,
3223                          BlockJobTxn *txn, Error **errp)
3224 {
3225     BlockBackend *blk, *target_blk;
3226     BlockDriverState *bs;
3227     BlockDriverState *target_bs;
3228     Error *local_err = NULL;
3229     AioContext *aio_context;
3230 
3231     if (!has_speed) {
3232         speed = 0;
3233     }
3234     if (!has_on_source_error) {
3235         on_source_error = BLOCKDEV_ON_ERROR_REPORT;
3236     }
3237     if (!has_on_target_error) {
3238         on_target_error = BLOCKDEV_ON_ERROR_REPORT;
3239     }
3240 
3241     blk = blk_by_name(device);
3242     if (!blk) {
3243         error_setg(errp, "Device '%s' not found", device);
3244         return;
3245     }
3246 
3247     aio_context = blk_get_aio_context(blk);
3248     aio_context_acquire(aio_context);
3249 
3250     if (!blk_is_available(blk)) {
3251         error_setg(errp, "Device '%s' has no medium", device);
3252         goto out;
3253     }
3254     bs = blk_bs(blk);
3255 
3256     target_blk = blk_by_name(target);
3257     if (!target_blk) {
3258         error_setg(errp, "Device '%s' not found", target);
3259         goto out;
3260     }
3261 
3262     if (!blk_is_available(target_blk)) {
3263         error_setg(errp, "Device '%s' has no medium", target);
3264         goto out;
3265     }
3266     target_bs = blk_bs(target_blk);
3267 
3268     bdrv_ref(target_bs);
3269     bdrv_set_aio_context(target_bs, aio_context);
3270     backup_start(bs, target_bs, speed, sync, NULL, on_source_error,
3271                  on_target_error, block_job_cb, bs, txn, &local_err);
3272     if (local_err != NULL) {
3273         bdrv_unref(target_bs);
3274         error_propagate(errp, local_err);
3275     }
3276 out:
3277     aio_context_release(aio_context);
3278 }
3279 
3280 void qmp_blockdev_backup(const char *device, const char *target,
3281                          enum MirrorSyncMode sync,
3282                          bool has_speed, int64_t speed,
3283                          bool has_on_source_error,
3284                          BlockdevOnError on_source_error,
3285                          bool has_on_target_error,
3286                          BlockdevOnError on_target_error,
3287                          Error **errp)
3288 {
3289     do_blockdev_backup(device, target, sync, has_speed, speed,
3290                        has_on_source_error, on_source_error,
3291                        has_on_target_error, on_target_error,
3292                        NULL, errp);
3293 }
3294 
3295 void qmp_drive_mirror(const char *device, const char *target,
3296                       bool has_format, const char *format,
3297                       bool has_node_name, const char *node_name,
3298                       bool has_replaces, const char *replaces,
3299                       enum MirrorSyncMode sync,
3300                       bool has_mode, enum NewImageMode mode,
3301                       bool has_speed, int64_t speed,
3302                       bool has_granularity, uint32_t granularity,
3303                       bool has_buf_size, int64_t buf_size,
3304                       bool has_on_source_error, BlockdevOnError on_source_error,
3305                       bool has_on_target_error, BlockdevOnError on_target_error,
3306                       bool has_unmap, bool unmap,
3307                       Error **errp)
3308 {
3309     BlockBackend *blk;
3310     BlockDriverState *bs;
3311     BlockDriverState *source, *target_bs;
3312     AioContext *aio_context;
3313     Error *local_err = NULL;
3314     QDict *options;
3315     int flags;
3316     int64_t size;
3317     int ret;
3318 
3319     if (!has_speed) {
3320         speed = 0;
3321     }
3322     if (!has_on_source_error) {
3323         on_source_error = BLOCKDEV_ON_ERROR_REPORT;
3324     }
3325     if (!has_on_target_error) {
3326         on_target_error = BLOCKDEV_ON_ERROR_REPORT;
3327     }
3328     if (!has_mode) {
3329         mode = NEW_IMAGE_MODE_ABSOLUTE_PATHS;
3330     }
3331     if (!has_granularity) {
3332         granularity = 0;
3333     }
3334     if (!has_buf_size) {
3335         buf_size = 0;
3336     }
3337     if (!has_unmap) {
3338         unmap = true;
3339     }
3340 
3341     if (granularity != 0 && (granularity < 512 || granularity > 1048576 * 64)) {
3342         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "granularity",
3343                    "a value in range [512B, 64MB]");
3344         return;
3345     }
3346     if (granularity & (granularity - 1)) {
3347         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "granularity",
3348                    "power of 2");
3349         return;
3350     }
3351 
3352     blk = blk_by_name(device);
3353     if (!blk) {
3354         error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
3355                   "Device '%s' not found", device);
3356         return;
3357     }
3358 
3359     aio_context = blk_get_aio_context(blk);
3360     aio_context_acquire(aio_context);
3361 
3362     if (!blk_is_available(blk)) {
3363         error_setg(errp, QERR_DEVICE_HAS_NO_MEDIUM, device);
3364         goto out;
3365     }
3366     bs = blk_bs(blk);
3367 
3368     if (!has_format) {
3369         format = mode == NEW_IMAGE_MODE_EXISTING ? NULL : bs->drv->format_name;
3370     }
3371 
3372     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_MIRROR, errp)) {
3373         goto out;
3374     }
3375 
3376     flags = bs->open_flags | BDRV_O_RDWR;
3377     source = backing_bs(bs);
3378     if (!source && sync == MIRROR_SYNC_MODE_TOP) {
3379         sync = MIRROR_SYNC_MODE_FULL;
3380     }
3381     if (sync == MIRROR_SYNC_MODE_NONE) {
3382         source = bs;
3383     }
3384 
3385     size = bdrv_getlength(bs);
3386     if (size < 0) {
3387         error_setg_errno(errp, -size, "bdrv_getlength failed");
3388         goto out;
3389     }
3390 
3391     if (has_replaces) {
3392         BlockDriverState *to_replace_bs;
3393         AioContext *replace_aio_context;
3394         int64_t replace_size;
3395 
3396         if (!has_node_name) {
3397             error_setg(errp, "a node-name must be provided when replacing a"
3398                              " named node of the graph");
3399             goto out;
3400         }
3401 
3402         to_replace_bs = check_to_replace_node(bs, replaces, &local_err);
3403 
3404         if (!to_replace_bs) {
3405             error_propagate(errp, local_err);
3406             goto out;
3407         }
3408 
3409         replace_aio_context = bdrv_get_aio_context(to_replace_bs);
3410         aio_context_acquire(replace_aio_context);
3411         replace_size = bdrv_getlength(to_replace_bs);
3412         aio_context_release(replace_aio_context);
3413 
3414         if (size != replace_size) {
3415             error_setg(errp, "cannot replace image with a mirror image of "
3416                              "different size");
3417             goto out;
3418         }
3419     }
3420 
3421     if ((sync == MIRROR_SYNC_MODE_FULL || !source)
3422         && mode != NEW_IMAGE_MODE_EXISTING)
3423     {
3424         /* create new image w/o backing file */
3425         assert(format);
3426         bdrv_img_create(target, format,
3427                         NULL, NULL, NULL, size, flags, &local_err, false);
3428     } else {
3429         switch (mode) {
3430         case NEW_IMAGE_MODE_EXISTING:
3431             break;
3432         case NEW_IMAGE_MODE_ABSOLUTE_PATHS:
3433             /* create new image with backing file */
3434             bdrv_img_create(target, format,
3435                             source->filename,
3436                             source->drv->format_name,
3437                             NULL, size, flags, &local_err, false);
3438             break;
3439         default:
3440             abort();
3441         }
3442     }
3443 
3444     if (local_err) {
3445         error_propagate(errp, local_err);
3446         goto out;
3447     }
3448 
3449     options = qdict_new();
3450     if (has_node_name) {
3451         qdict_put(options, "node-name", qstring_from_str(node_name));
3452     }
3453     if (format) {
3454         qdict_put(options, "driver", qstring_from_str(format));
3455     }
3456 
3457     /* Mirroring takes care of copy-on-write using the source's backing
3458      * file.
3459      */
3460     target_bs = NULL;
3461     ret = bdrv_open(&target_bs, target, NULL, options,
3462                     flags | BDRV_O_NO_BACKING, &local_err);
3463     if (ret < 0) {
3464         error_propagate(errp, local_err);
3465         goto out;
3466     }
3467 
3468     bdrv_set_aio_context(target_bs, aio_context);
3469 
3470     /* pass the node name to replace to mirror start since it's loose coupling
3471      * and will allow to check whether the node still exist at mirror completion
3472      */
3473     mirror_start(bs, target_bs,
3474                  has_replaces ? replaces : NULL,
3475                  speed, granularity, buf_size, sync,
3476                  on_source_error, on_target_error,
3477                  unmap,
3478                  block_job_cb, bs, &local_err);
3479     if (local_err != NULL) {
3480         bdrv_unref(target_bs);
3481         error_propagate(errp, local_err);
3482         goto out;
3483     }
3484 
3485 out:
3486     aio_context_release(aio_context);
3487 }
3488 
3489 /* Get the block job for a given device name and acquire its AioContext */
3490 static BlockJob *find_block_job(const char *device, AioContext **aio_context,
3491                                 Error **errp)
3492 {
3493     BlockBackend *blk;
3494     BlockDriverState *bs;
3495 
3496     *aio_context = NULL;
3497 
3498     blk = blk_by_name(device);
3499     if (!blk) {
3500         goto notfound;
3501     }
3502 
3503     *aio_context = blk_get_aio_context(blk);
3504     aio_context_acquire(*aio_context);
3505 
3506     if (!blk_is_available(blk)) {
3507         goto notfound;
3508     }
3509     bs = blk_bs(blk);
3510 
3511     if (!bs->job) {
3512         goto notfound;
3513     }
3514 
3515     return bs->job;
3516 
3517 notfound:
3518     error_set(errp, ERROR_CLASS_DEVICE_NOT_ACTIVE,
3519               "No active block job on device '%s'", device);
3520     if (*aio_context) {
3521         aio_context_release(*aio_context);
3522         *aio_context = NULL;
3523     }
3524     return NULL;
3525 }
3526 
3527 void qmp_block_job_set_speed(const char *device, int64_t speed, Error **errp)
3528 {
3529     AioContext *aio_context;
3530     BlockJob *job = find_block_job(device, &aio_context, errp);
3531 
3532     if (!job) {
3533         return;
3534     }
3535 
3536     block_job_set_speed(job, speed, errp);
3537     aio_context_release(aio_context);
3538 }
3539 
3540 void qmp_block_job_cancel(const char *device,
3541                           bool has_force, bool force, Error **errp)
3542 {
3543     AioContext *aio_context;
3544     BlockJob *job = find_block_job(device, &aio_context, errp);
3545 
3546     if (!job) {
3547         return;
3548     }
3549 
3550     if (!has_force) {
3551         force = false;
3552     }
3553 
3554     if (job->user_paused && !force) {
3555         error_setg(errp, "The block job for device '%s' is currently paused",
3556                    device);
3557         goto out;
3558     }
3559 
3560     trace_qmp_block_job_cancel(job);
3561     block_job_cancel(job);
3562 out:
3563     aio_context_release(aio_context);
3564 }
3565 
3566 void qmp_block_job_pause(const char *device, Error **errp)
3567 {
3568     AioContext *aio_context;
3569     BlockJob *job = find_block_job(device, &aio_context, errp);
3570 
3571     if (!job || job->user_paused) {
3572         return;
3573     }
3574 
3575     job->user_paused = true;
3576     trace_qmp_block_job_pause(job);
3577     block_job_pause(job);
3578     aio_context_release(aio_context);
3579 }
3580 
3581 void qmp_block_job_resume(const char *device, Error **errp)
3582 {
3583     AioContext *aio_context;
3584     BlockJob *job = find_block_job(device, &aio_context, errp);
3585 
3586     if (!job || !job->user_paused) {
3587         return;
3588     }
3589 
3590     job->user_paused = false;
3591     trace_qmp_block_job_resume(job);
3592     block_job_resume(job);
3593     aio_context_release(aio_context);
3594 }
3595 
3596 void qmp_block_job_complete(const char *device, Error **errp)
3597 {
3598     AioContext *aio_context;
3599     BlockJob *job = find_block_job(device, &aio_context, errp);
3600 
3601     if (!job) {
3602         return;
3603     }
3604 
3605     trace_qmp_block_job_complete(job);
3606     block_job_complete(job, errp);
3607     aio_context_release(aio_context);
3608 }
3609 
3610 void qmp_change_backing_file(const char *device,
3611                              const char *image_node_name,
3612                              const char *backing_file,
3613                              Error **errp)
3614 {
3615     BlockBackend *blk;
3616     BlockDriverState *bs = NULL;
3617     AioContext *aio_context;
3618     BlockDriverState *image_bs = NULL;
3619     Error *local_err = NULL;
3620     bool ro;
3621     int open_flags;
3622     int ret;
3623 
3624     blk = blk_by_name(device);
3625     if (!blk) {
3626         error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
3627                   "Device '%s' not found", device);
3628         return;
3629     }
3630 
3631     aio_context = blk_get_aio_context(blk);
3632     aio_context_acquire(aio_context);
3633 
3634     if (!blk_is_available(blk)) {
3635         error_setg(errp, "Device '%s' has no medium", device);
3636         goto out;
3637     }
3638     bs = blk_bs(blk);
3639 
3640     image_bs = bdrv_lookup_bs(NULL, image_node_name, &local_err);
3641     if (local_err) {
3642         error_propagate(errp, local_err);
3643         goto out;
3644     }
3645 
3646     if (!image_bs) {
3647         error_setg(errp, "image file not found");
3648         goto out;
3649     }
3650 
3651     if (bdrv_find_base(image_bs) == image_bs) {
3652         error_setg(errp, "not allowing backing file change on an image "
3653                          "without a backing file");
3654         goto out;
3655     }
3656 
3657     /* even though we are not necessarily operating on bs, we need it to
3658      * determine if block ops are currently prohibited on the chain */
3659     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_CHANGE, errp)) {
3660         goto out;
3661     }
3662 
3663     /* final sanity check */
3664     if (!bdrv_chain_contains(bs, image_bs)) {
3665         error_setg(errp, "'%s' and image file are not in the same chain",
3666                    device);
3667         goto out;
3668     }
3669 
3670     /* if not r/w, reopen to make r/w */
3671     open_flags = image_bs->open_flags;
3672     ro = bdrv_is_read_only(image_bs);
3673 
3674     if (ro) {
3675         bdrv_reopen(image_bs, open_flags | BDRV_O_RDWR, &local_err);
3676         if (local_err) {
3677             error_propagate(errp, local_err);
3678             goto out;
3679         }
3680     }
3681 
3682     ret = bdrv_change_backing_file(image_bs, backing_file,
3683                                image_bs->drv ? image_bs->drv->format_name : "");
3684 
3685     if (ret < 0) {
3686         error_setg_errno(errp, -ret, "Could not change backing file to '%s'",
3687                          backing_file);
3688         /* don't exit here, so we can try to restore open flags if
3689          * appropriate */
3690     }
3691 
3692     if (ro) {
3693         bdrv_reopen(image_bs, open_flags, &local_err);
3694         if (local_err) {
3695             error_propagate(errp, local_err); /* will preserve prior errp */
3696         }
3697     }
3698 
3699 out:
3700     aio_context_release(aio_context);
3701 }
3702 
3703 void qmp_blockdev_add(BlockdevOptions *options, Error **errp)
3704 {
3705     QmpOutputVisitor *ov = qmp_output_visitor_new();
3706     BlockDriverState *bs;
3707     BlockBackend *blk = NULL;
3708     QObject *obj;
3709     QDict *qdict;
3710     Error *local_err = NULL;
3711 
3712     /* TODO Sort it out in raw-posix and drive_new(): Reject aio=native with
3713      * cache.direct=false instead of silently switching to aio=threads, except
3714      * when called from drive_new().
3715      *
3716      * For now, simply forbidding the combination for all drivers will do. */
3717     if (options->has_aio && options->aio == BLOCKDEV_AIO_OPTIONS_NATIVE) {
3718         bool direct = options->has_cache &&
3719                       options->cache->has_direct &&
3720                       options->cache->direct;
3721         if (!direct) {
3722             error_setg(errp, "aio=native requires cache.direct=true");
3723             goto fail;
3724         }
3725     }
3726 
3727     visit_type_BlockdevOptions(qmp_output_get_visitor(ov),
3728                                &options, NULL, &local_err);
3729     if (local_err) {
3730         error_propagate(errp, local_err);
3731         goto fail;
3732     }
3733 
3734     obj = qmp_output_get_qobject(ov);
3735     qdict = qobject_to_qdict(obj);
3736 
3737     qdict_flatten(qdict);
3738 
3739     if (options->has_id) {
3740         blk = blockdev_init(NULL, qdict, &local_err);
3741         if (local_err) {
3742             error_propagate(errp, local_err);
3743             goto fail;
3744         }
3745 
3746         bs = blk_bs(blk);
3747     } else {
3748         if (!qdict_get_try_str(qdict, "node-name")) {
3749             error_setg(errp, "'id' and/or 'node-name' need to be specified for "
3750                        "the root node");
3751             goto fail;
3752         }
3753 
3754         bs = bds_tree_init(qdict, errp);
3755         if (!bs) {
3756             goto fail;
3757         }
3758     }
3759 
3760     if (bs && bdrv_key_required(bs)) {
3761         if (blk) {
3762             blk_unref(blk);
3763         } else {
3764             bdrv_unref(bs);
3765         }
3766         error_setg(errp, "blockdev-add doesn't support encrypted devices");
3767         goto fail;
3768     }
3769 
3770 fail:
3771     qmp_output_visitor_cleanup(ov);
3772 }
3773 
3774 void qmp_x_blockdev_del(bool has_id, const char *id,
3775                         bool has_node_name, const char *node_name, Error **errp)
3776 {
3777     AioContext *aio_context;
3778     BlockBackend *blk;
3779     BlockDriverState *bs;
3780 
3781     if (has_id && has_node_name) {
3782         error_setg(errp, "Only one of id and node-name must be specified");
3783         return;
3784     } else if (!has_id && !has_node_name) {
3785         error_setg(errp, "No block device specified");
3786         return;
3787     }
3788 
3789     if (has_id) {
3790         blk = blk_by_name(id);
3791         if (!blk) {
3792             error_setg(errp, "Cannot find block backend %s", id);
3793             return;
3794         }
3795         if (blk_get_refcnt(blk) > 1) {
3796             error_setg(errp, "Block backend %s is in use", id);
3797             return;
3798         }
3799         bs = blk_bs(blk);
3800         aio_context = blk_get_aio_context(blk);
3801     } else {
3802         bs = bdrv_find_node(node_name);
3803         if (!bs) {
3804             error_setg(errp, "Cannot find node %s", node_name);
3805             return;
3806         }
3807         blk = bs->blk;
3808         if (blk) {
3809             error_setg(errp, "Node %s is in use by %s",
3810                        node_name, blk_name(blk));
3811             return;
3812         }
3813         aio_context = bdrv_get_aio_context(bs);
3814     }
3815 
3816     aio_context_acquire(aio_context);
3817 
3818     if (bs) {
3819         if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_DRIVE_DEL, errp)) {
3820             goto out;
3821         }
3822 
3823         if (bs->refcnt > 1 || !QLIST_EMPTY(&bs->parents)) {
3824             error_setg(errp, "Block device %s is in use",
3825                        bdrv_get_device_or_node_name(bs));
3826             goto out;
3827         }
3828     }
3829 
3830     if (blk) {
3831         blk_unref(blk);
3832     } else {
3833         bdrv_unref(bs);
3834     }
3835 
3836 out:
3837     aio_context_release(aio_context);
3838 }
3839 
3840 BlockJobInfoList *qmp_query_block_jobs(Error **errp)
3841 {
3842     BlockJobInfoList *head = NULL, **p_next = &head;
3843     BlockDriverState *bs;
3844 
3845     for (bs = bdrv_next(NULL); bs; bs = bdrv_next(bs)) {
3846         AioContext *aio_context = bdrv_get_aio_context(bs);
3847 
3848         aio_context_acquire(aio_context);
3849 
3850         if (bs->job) {
3851             BlockJobInfoList *elem = g_new0(BlockJobInfoList, 1);
3852             elem->value = block_job_query(bs->job);
3853             *p_next = elem;
3854             p_next = &elem->next;
3855         }
3856 
3857         aio_context_release(aio_context);
3858     }
3859 
3860     return head;
3861 }
3862 
3863 QemuOptsList qemu_common_drive_opts = {
3864     .name = "drive",
3865     .head = QTAILQ_HEAD_INITIALIZER(qemu_common_drive_opts.head),
3866     .desc = {
3867         {
3868             .name = "snapshot",
3869             .type = QEMU_OPT_BOOL,
3870             .help = "enable/disable snapshot mode",
3871         },{
3872             .name = "discard",
3873             .type = QEMU_OPT_STRING,
3874             .help = "discard operation (ignore/off, unmap/on)",
3875         },{
3876             .name = BDRV_OPT_CACHE_WB,
3877             .type = QEMU_OPT_BOOL,
3878             .help = "enables writeback mode for any caches",
3879         },{
3880             .name = BDRV_OPT_CACHE_DIRECT,
3881             .type = QEMU_OPT_BOOL,
3882             .help = "enables use of O_DIRECT (bypass the host page cache)",
3883         },{
3884             .name = BDRV_OPT_CACHE_NO_FLUSH,
3885             .type = QEMU_OPT_BOOL,
3886             .help = "ignore any flush requests for the device",
3887         },{
3888             .name = "aio",
3889             .type = QEMU_OPT_STRING,
3890             .help = "host AIO implementation (threads, native)",
3891         },{
3892             .name = "format",
3893             .type = QEMU_OPT_STRING,
3894             .help = "disk format (raw, qcow2, ...)",
3895         },{
3896             .name = "rerror",
3897             .type = QEMU_OPT_STRING,
3898             .help = "read error action",
3899         },{
3900             .name = "werror",
3901             .type = QEMU_OPT_STRING,
3902             .help = "write error action",
3903         },{
3904             .name = "read-only",
3905             .type = QEMU_OPT_BOOL,
3906             .help = "open drive file as read-only",
3907         },{
3908             .name = "throttling.iops-total",
3909             .type = QEMU_OPT_NUMBER,
3910             .help = "limit total I/O operations per second",
3911         },{
3912             .name = "throttling.iops-read",
3913             .type = QEMU_OPT_NUMBER,
3914             .help = "limit read operations per second",
3915         },{
3916             .name = "throttling.iops-write",
3917             .type = QEMU_OPT_NUMBER,
3918             .help = "limit write operations per second",
3919         },{
3920             .name = "throttling.bps-total",
3921             .type = QEMU_OPT_NUMBER,
3922             .help = "limit total bytes per second",
3923         },{
3924             .name = "throttling.bps-read",
3925             .type = QEMU_OPT_NUMBER,
3926             .help = "limit read bytes per second",
3927         },{
3928             .name = "throttling.bps-write",
3929             .type = QEMU_OPT_NUMBER,
3930             .help = "limit write bytes per second",
3931         },{
3932             .name = "throttling.iops-total-max",
3933             .type = QEMU_OPT_NUMBER,
3934             .help = "I/O operations burst",
3935         },{
3936             .name = "throttling.iops-read-max",
3937             .type = QEMU_OPT_NUMBER,
3938             .help = "I/O operations read burst",
3939         },{
3940             .name = "throttling.iops-write-max",
3941             .type = QEMU_OPT_NUMBER,
3942             .help = "I/O operations write burst",
3943         },{
3944             .name = "throttling.bps-total-max",
3945             .type = QEMU_OPT_NUMBER,
3946             .help = "total bytes burst",
3947         },{
3948             .name = "throttling.bps-read-max",
3949             .type = QEMU_OPT_NUMBER,
3950             .help = "total bytes read burst",
3951         },{
3952             .name = "throttling.bps-write-max",
3953             .type = QEMU_OPT_NUMBER,
3954             .help = "total bytes write burst",
3955         },{
3956             .name = "throttling.iops-size",
3957             .type = QEMU_OPT_NUMBER,
3958             .help = "when limiting by iops max size of an I/O in bytes",
3959         },{
3960             .name = "throttling.group",
3961             .type = QEMU_OPT_STRING,
3962             .help = "name of the block throttling group",
3963         },{
3964             .name = "copy-on-read",
3965             .type = QEMU_OPT_BOOL,
3966             .help = "copy read data from backing file into image file",
3967         },{
3968             .name = "detect-zeroes",
3969             .type = QEMU_OPT_STRING,
3970             .help = "try to optimize zero writes (off, on, unmap)",
3971         },{
3972             .name = "stats-account-invalid",
3973             .type = QEMU_OPT_BOOL,
3974             .help = "whether to account for invalid I/O operations "
3975                     "in the statistics",
3976         },{
3977             .name = "stats-account-failed",
3978             .type = QEMU_OPT_BOOL,
3979             .help = "whether to account for failed I/O operations "
3980                     "in the statistics",
3981         },
3982         { /* end of list */ }
3983     },
3984 };
3985 
3986 static QemuOptsList qemu_root_bds_opts = {
3987     .name = "root-bds",
3988     .head = QTAILQ_HEAD_INITIALIZER(qemu_common_drive_opts.head),
3989     .desc = {
3990         {
3991             .name = "discard",
3992             .type = QEMU_OPT_STRING,
3993             .help = "discard operation (ignore/off, unmap/on)",
3994         },{
3995             .name = "cache.writeback",
3996             .type = QEMU_OPT_BOOL,
3997             .help = "enables writeback mode for any caches",
3998         },{
3999             .name = "cache.direct",
4000             .type = QEMU_OPT_BOOL,
4001             .help = "enables use of O_DIRECT (bypass the host page cache)",
4002         },{
4003             .name = "cache.no-flush",
4004             .type = QEMU_OPT_BOOL,
4005             .help = "ignore any flush requests for the device",
4006         },{
4007             .name = "aio",
4008             .type = QEMU_OPT_STRING,
4009             .help = "host AIO implementation (threads, native)",
4010         },{
4011             .name = "read-only",
4012             .type = QEMU_OPT_BOOL,
4013             .help = "open drive file as read-only",
4014         },{
4015             .name = "copy-on-read",
4016             .type = QEMU_OPT_BOOL,
4017             .help = "copy read data from backing file into image file",
4018         },{
4019             .name = "detect-zeroes",
4020             .type = QEMU_OPT_STRING,
4021             .help = "try to optimize zero writes (off, on, unmap)",
4022         },
4023         { /* end of list */ }
4024     },
4025 };
4026 
4027 QemuOptsList qemu_drive_opts = {
4028     .name = "drive",
4029     .head = QTAILQ_HEAD_INITIALIZER(qemu_drive_opts.head),
4030     .desc = {
4031         /*
4032          * no elements => accept any params
4033          * validation will happen later
4034          */
4035         { /* end of list */ }
4036     },
4037 };
4038