xref: /openbmc/qemu/blockdev.c (revision db7a99cd)
1 /*
2  * QEMU host block devices
3  *
4  * Copyright (c) 2003-2008 Fabrice Bellard
5  *
6  * This work is licensed under the terms of the GNU GPL, version 2 or
7  * later.  See the COPYING file in the top-level directory.
8  *
9  * This file incorporates work covered by the following copyright and
10  * permission notice:
11  *
12  * Copyright (c) 2003-2008 Fabrice Bellard
13  *
14  * Permission is hereby granted, free of charge, to any person obtaining a copy
15  * of this software and associated documentation files (the "Software"), to deal
16  * in the Software without restriction, including without limitation the rights
17  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
18  * copies of the Software, and to permit persons to whom the Software is
19  * furnished to do so, subject to the following conditions:
20  *
21  * The above copyright notice and this permission notice shall be included in
22  * all copies or substantial portions of the Software.
23  *
24  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
25  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
26  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
27  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
28  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
29  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
30  * THE SOFTWARE.
31  */
32 
33 #include "qemu/osdep.h"
34 #include "sysemu/block-backend.h"
35 #include "sysemu/blockdev.h"
36 #include "hw/block/block.h"
37 #include "block/blockjob.h"
38 #include "block/throttle-groups.h"
39 #include "monitor/monitor.h"
40 #include "qemu/error-report.h"
41 #include "qemu/option.h"
42 #include "qemu/config-file.h"
43 #include "qapi/qmp/types.h"
44 #include "qapi-visit.h"
45 #include "qapi/qmp/qerror.h"
46 #include "qapi/qobject-output-visitor.h"
47 #include "qapi/util.h"
48 #include "sysemu/sysemu.h"
49 #include "block/block_int.h"
50 #include "qmp-commands.h"
51 #include "block/trace.h"
52 #include "sysemu/arch_init.h"
53 #include "qemu/cutils.h"
54 #include "qemu/help_option.h"
55 #include "qemu/throttle-options.h"
56 
57 static QTAILQ_HEAD(, BlockDriverState) monitor_bdrv_states =
58     QTAILQ_HEAD_INITIALIZER(monitor_bdrv_states);
59 
60 static int do_open_tray(const char *blk_name, const char *qdev_id,
61                         bool force, Error **errp);
62 
63 static const char *const if_name[IF_COUNT] = {
64     [IF_NONE] = "none",
65     [IF_IDE] = "ide",
66     [IF_SCSI] = "scsi",
67     [IF_FLOPPY] = "floppy",
68     [IF_PFLASH] = "pflash",
69     [IF_MTD] = "mtd",
70     [IF_SD] = "sd",
71     [IF_VIRTIO] = "virtio",
72     [IF_XEN] = "xen",
73 };
74 
75 static int if_max_devs[IF_COUNT] = {
76     /*
77      * Do not change these numbers!  They govern how drive option
78      * index maps to unit and bus.  That mapping is ABI.
79      *
80      * All controllers used to implement if=T drives need to support
81      * if_max_devs[T] units, for any T with if_max_devs[T] != 0.
82      * Otherwise, some index values map to "impossible" bus, unit
83      * values.
84      *
85      * For instance, if you change [IF_SCSI] to 255, -drive
86      * if=scsi,index=12 no longer means bus=1,unit=5, but
87      * bus=0,unit=12.  With an lsi53c895a controller (7 units max),
88      * the drive can't be set up.  Regression.
89      */
90     [IF_IDE] = 2,
91     [IF_SCSI] = 7,
92 };
93 
94 /**
95  * Boards may call this to offer board-by-board overrides
96  * of the default, global values.
97  */
98 void override_max_devs(BlockInterfaceType type, int max_devs)
99 {
100     BlockBackend *blk;
101     DriveInfo *dinfo;
102 
103     if (max_devs <= 0) {
104         return;
105     }
106 
107     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
108         dinfo = blk_legacy_dinfo(blk);
109         if (dinfo->type == type) {
110             fprintf(stderr, "Cannot override units-per-bus property of"
111                     " the %s interface, because a drive of that type has"
112                     " already been added.\n", if_name[type]);
113             g_assert_not_reached();
114         }
115     }
116 
117     if_max_devs[type] = max_devs;
118 }
119 
120 /*
121  * We automatically delete the drive when a device using it gets
122  * unplugged.  Questionable feature, but we can't just drop it.
123  * Device models call blockdev_mark_auto_del() to schedule the
124  * automatic deletion, and generic qdev code calls blockdev_auto_del()
125  * when deletion is actually safe.
126  */
127 void blockdev_mark_auto_del(BlockBackend *blk)
128 {
129     DriveInfo *dinfo = blk_legacy_dinfo(blk);
130     BlockDriverState *bs = blk_bs(blk);
131     AioContext *aio_context;
132 
133     if (!dinfo) {
134         return;
135     }
136 
137     if (bs) {
138         aio_context = bdrv_get_aio_context(bs);
139         aio_context_acquire(aio_context);
140 
141         if (bs->job) {
142             block_job_cancel(bs->job);
143         }
144 
145         aio_context_release(aio_context);
146     }
147 
148     dinfo->auto_del = 1;
149 }
150 
151 void blockdev_auto_del(BlockBackend *blk)
152 {
153     DriveInfo *dinfo = blk_legacy_dinfo(blk);
154 
155     if (dinfo && dinfo->auto_del) {
156         monitor_remove_blk(blk);
157         blk_unref(blk);
158     }
159 }
160 
161 /**
162  * Returns the current mapping of how many units per bus
163  * a particular interface can support.
164  *
165  *  A positive integer indicates n units per bus.
166  *  0 implies the mapping has not been established.
167  * -1 indicates an invalid BlockInterfaceType was given.
168  */
169 int drive_get_max_devs(BlockInterfaceType type)
170 {
171     if (type >= IF_IDE && type < IF_COUNT) {
172         return if_max_devs[type];
173     }
174 
175     return -1;
176 }
177 
178 static int drive_index_to_bus_id(BlockInterfaceType type, int index)
179 {
180     int max_devs = if_max_devs[type];
181     return max_devs ? index / max_devs : 0;
182 }
183 
184 static int drive_index_to_unit_id(BlockInterfaceType type, int index)
185 {
186     int max_devs = if_max_devs[type];
187     return max_devs ? index % max_devs : index;
188 }
189 
190 QemuOpts *drive_def(const char *optstr)
191 {
192     return qemu_opts_parse_noisily(qemu_find_opts("drive"), optstr, false);
193 }
194 
195 QemuOpts *drive_add(BlockInterfaceType type, int index, const char *file,
196                     const char *optstr)
197 {
198     QemuOpts *opts;
199 
200     opts = drive_def(optstr);
201     if (!opts) {
202         return NULL;
203     }
204     if (type != IF_DEFAULT) {
205         qemu_opt_set(opts, "if", if_name[type], &error_abort);
206     }
207     if (index >= 0) {
208         qemu_opt_set_number(opts, "index", index, &error_abort);
209     }
210     if (file)
211         qemu_opt_set(opts, "file", file, &error_abort);
212     return opts;
213 }
214 
215 DriveInfo *drive_get(BlockInterfaceType type, int bus, int unit)
216 {
217     BlockBackend *blk;
218     DriveInfo *dinfo;
219 
220     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
221         dinfo = blk_legacy_dinfo(blk);
222         if (dinfo && dinfo->type == type
223             && dinfo->bus == bus && dinfo->unit == unit) {
224             return dinfo;
225         }
226     }
227 
228     return NULL;
229 }
230 
231 void drive_check_orphaned(void)
232 {
233     BlockBackend *blk;
234     DriveInfo *dinfo;
235     Location loc;
236     bool orphans = false;
237 
238     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
239         dinfo = blk_legacy_dinfo(blk);
240         if (!blk_get_attached_dev(blk) && !dinfo->is_default &&
241             dinfo->type != IF_NONE) {
242             loc_push_none(&loc);
243             qemu_opts_loc_restore(dinfo->opts);
244             error_report("machine type does not support"
245                          " if=%s,bus=%d,unit=%d",
246                          if_name[dinfo->type], dinfo->bus, dinfo->unit);
247             loc_pop(&loc);
248             orphans = true;
249         }
250     }
251 
252     if (orphans) {
253         exit(1);
254     }
255 }
256 
257 DriveInfo *drive_get_by_index(BlockInterfaceType type, int index)
258 {
259     return drive_get(type,
260                      drive_index_to_bus_id(type, index),
261                      drive_index_to_unit_id(type, index));
262 }
263 
264 int drive_get_max_bus(BlockInterfaceType type)
265 {
266     int max_bus;
267     BlockBackend *blk;
268     DriveInfo *dinfo;
269 
270     max_bus = -1;
271     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
272         dinfo = blk_legacy_dinfo(blk);
273         if (dinfo && dinfo->type == type && dinfo->bus > max_bus) {
274             max_bus = dinfo->bus;
275         }
276     }
277     return max_bus;
278 }
279 
280 /* Get a block device.  This should only be used for single-drive devices
281    (e.g. SD/Floppy/MTD).  Multi-disk devices (scsi/ide) should use the
282    appropriate bus.  */
283 DriveInfo *drive_get_next(BlockInterfaceType type)
284 {
285     static int next_block_unit[IF_COUNT];
286 
287     return drive_get(type, 0, next_block_unit[type]++);
288 }
289 
290 static void bdrv_format_print(void *opaque, const char *name)
291 {
292     error_printf(" %s", name);
293 }
294 
295 typedef struct {
296     QEMUBH *bh;
297     BlockDriverState *bs;
298 } BDRVPutRefBH;
299 
300 static int parse_block_error_action(const char *buf, bool is_read, Error **errp)
301 {
302     if (!strcmp(buf, "ignore")) {
303         return BLOCKDEV_ON_ERROR_IGNORE;
304     } else if (!is_read && !strcmp(buf, "enospc")) {
305         return BLOCKDEV_ON_ERROR_ENOSPC;
306     } else if (!strcmp(buf, "stop")) {
307         return BLOCKDEV_ON_ERROR_STOP;
308     } else if (!strcmp(buf, "report")) {
309         return BLOCKDEV_ON_ERROR_REPORT;
310     } else {
311         error_setg(errp, "'%s' invalid %s error action",
312                    buf, is_read ? "read" : "write");
313         return -1;
314     }
315 }
316 
317 static bool parse_stats_intervals(BlockAcctStats *stats, QList *intervals,
318                                   Error **errp)
319 {
320     const QListEntry *entry;
321     for (entry = qlist_first(intervals); entry; entry = qlist_next(entry)) {
322         switch (qobject_type(entry->value)) {
323 
324         case QTYPE_QSTRING: {
325             unsigned long long length;
326             const char *str = qstring_get_str(qobject_to_qstring(entry->value));
327             if (parse_uint_full(str, &length, 10) == 0 &&
328                 length > 0 && length <= UINT_MAX) {
329                 block_acct_add_interval(stats, (unsigned) length);
330             } else {
331                 error_setg(errp, "Invalid interval length: %s", str);
332                 return false;
333             }
334             break;
335         }
336 
337         case QTYPE_QINT: {
338             int64_t length = qint_get_int(qobject_to_qint(entry->value));
339             if (length > 0 && length <= UINT_MAX) {
340                 block_acct_add_interval(stats, (unsigned) length);
341             } else {
342                 error_setg(errp, "Invalid interval length: %" PRId64, length);
343                 return false;
344             }
345             break;
346         }
347 
348         default:
349             error_setg(errp, "The specification of stats-intervals is invalid");
350             return false;
351         }
352     }
353     return true;
354 }
355 
356 typedef enum { MEDIA_DISK, MEDIA_CDROM } DriveMediaType;
357 
358 /* All parameters but @opts are optional and may be set to NULL. */
359 static void extract_common_blockdev_options(QemuOpts *opts, int *bdrv_flags,
360     const char **throttling_group, ThrottleConfig *throttle_cfg,
361     BlockdevDetectZeroesOptions *detect_zeroes, Error **errp)
362 {
363     Error *local_error = NULL;
364     const char *aio;
365 
366     if (bdrv_flags) {
367         if (qemu_opt_get_bool(opts, "copy-on-read", false)) {
368             *bdrv_flags |= BDRV_O_COPY_ON_READ;
369         }
370 
371         if ((aio = qemu_opt_get(opts, "aio")) != NULL) {
372             if (!strcmp(aio, "native")) {
373                 *bdrv_flags |= BDRV_O_NATIVE_AIO;
374             } else if (!strcmp(aio, "threads")) {
375                 /* this is the default */
376             } else {
377                error_setg(errp, "invalid aio option");
378                return;
379             }
380         }
381     }
382 
383     /* disk I/O throttling */
384     if (throttling_group) {
385         *throttling_group = qemu_opt_get(opts, "throttling.group");
386     }
387 
388     if (throttle_cfg) {
389         throttle_config_init(throttle_cfg);
390         throttle_cfg->buckets[THROTTLE_BPS_TOTAL].avg =
391             qemu_opt_get_number(opts, "throttling.bps-total", 0);
392         throttle_cfg->buckets[THROTTLE_BPS_READ].avg  =
393             qemu_opt_get_number(opts, "throttling.bps-read", 0);
394         throttle_cfg->buckets[THROTTLE_BPS_WRITE].avg =
395             qemu_opt_get_number(opts, "throttling.bps-write", 0);
396         throttle_cfg->buckets[THROTTLE_OPS_TOTAL].avg =
397             qemu_opt_get_number(opts, "throttling.iops-total", 0);
398         throttle_cfg->buckets[THROTTLE_OPS_READ].avg =
399             qemu_opt_get_number(opts, "throttling.iops-read", 0);
400         throttle_cfg->buckets[THROTTLE_OPS_WRITE].avg =
401             qemu_opt_get_number(opts, "throttling.iops-write", 0);
402 
403         throttle_cfg->buckets[THROTTLE_BPS_TOTAL].max =
404             qemu_opt_get_number(opts, "throttling.bps-total-max", 0);
405         throttle_cfg->buckets[THROTTLE_BPS_READ].max  =
406             qemu_opt_get_number(opts, "throttling.bps-read-max", 0);
407         throttle_cfg->buckets[THROTTLE_BPS_WRITE].max =
408             qemu_opt_get_number(opts, "throttling.bps-write-max", 0);
409         throttle_cfg->buckets[THROTTLE_OPS_TOTAL].max =
410             qemu_opt_get_number(opts, "throttling.iops-total-max", 0);
411         throttle_cfg->buckets[THROTTLE_OPS_READ].max =
412             qemu_opt_get_number(opts, "throttling.iops-read-max", 0);
413         throttle_cfg->buckets[THROTTLE_OPS_WRITE].max =
414             qemu_opt_get_number(opts, "throttling.iops-write-max", 0);
415 
416         throttle_cfg->buckets[THROTTLE_BPS_TOTAL].burst_length =
417             qemu_opt_get_number(opts, "throttling.bps-total-max-length", 1);
418         throttle_cfg->buckets[THROTTLE_BPS_READ].burst_length  =
419             qemu_opt_get_number(opts, "throttling.bps-read-max-length", 1);
420         throttle_cfg->buckets[THROTTLE_BPS_WRITE].burst_length =
421             qemu_opt_get_number(opts, "throttling.bps-write-max-length", 1);
422         throttle_cfg->buckets[THROTTLE_OPS_TOTAL].burst_length =
423             qemu_opt_get_number(opts, "throttling.iops-total-max-length", 1);
424         throttle_cfg->buckets[THROTTLE_OPS_READ].burst_length =
425             qemu_opt_get_number(opts, "throttling.iops-read-max-length", 1);
426         throttle_cfg->buckets[THROTTLE_OPS_WRITE].burst_length =
427             qemu_opt_get_number(opts, "throttling.iops-write-max-length", 1);
428 
429         throttle_cfg->op_size =
430             qemu_opt_get_number(opts, "throttling.iops-size", 0);
431 
432         if (!throttle_is_valid(throttle_cfg, errp)) {
433             return;
434         }
435     }
436 
437     if (detect_zeroes) {
438         *detect_zeroes =
439             qapi_enum_parse(BlockdevDetectZeroesOptions_lookup,
440                             qemu_opt_get(opts, "detect-zeroes"),
441                             BLOCKDEV_DETECT_ZEROES_OPTIONS__MAX,
442                             BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF,
443                             &local_error);
444         if (local_error) {
445             error_propagate(errp, local_error);
446             return;
447         }
448     }
449 }
450 
451 /* Takes the ownership of bs_opts */
452 static BlockBackend *blockdev_init(const char *file, QDict *bs_opts,
453                                    Error **errp)
454 {
455     const char *buf;
456     int bdrv_flags = 0;
457     int on_read_error, on_write_error;
458     bool account_invalid, account_failed;
459     bool writethrough, read_only;
460     BlockBackend *blk;
461     BlockDriverState *bs;
462     ThrottleConfig cfg;
463     int snapshot = 0;
464     Error *error = NULL;
465     QemuOpts *opts;
466     QDict *interval_dict = NULL;
467     QList *interval_list = NULL;
468     const char *id;
469     BlockdevDetectZeroesOptions detect_zeroes =
470         BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF;
471     const char *throttling_group = NULL;
472 
473     /* Check common options by copying from bs_opts to opts, all other options
474      * stay in bs_opts for processing by bdrv_open(). */
475     id = qdict_get_try_str(bs_opts, "id");
476     opts = qemu_opts_create(&qemu_common_drive_opts, id, 1, &error);
477     if (error) {
478         error_propagate(errp, error);
479         goto err_no_opts;
480     }
481 
482     qemu_opts_absorb_qdict(opts, bs_opts, &error);
483     if (error) {
484         error_propagate(errp, error);
485         goto early_err;
486     }
487 
488     if (id) {
489         qdict_del(bs_opts, "id");
490     }
491 
492     /* extract parameters */
493     snapshot = qemu_opt_get_bool(opts, "snapshot", 0);
494 
495     account_invalid = qemu_opt_get_bool(opts, "stats-account-invalid", true);
496     account_failed = qemu_opt_get_bool(opts, "stats-account-failed", true);
497 
498     writethrough = !qemu_opt_get_bool(opts, BDRV_OPT_CACHE_WB, true);
499 
500     id = qemu_opts_id(opts);
501 
502     qdict_extract_subqdict(bs_opts, &interval_dict, "stats-intervals.");
503     qdict_array_split(interval_dict, &interval_list);
504 
505     if (qdict_size(interval_dict) != 0) {
506         error_setg(errp, "Invalid option stats-intervals.%s",
507                    qdict_first(interval_dict)->key);
508         goto early_err;
509     }
510 
511     extract_common_blockdev_options(opts, &bdrv_flags, &throttling_group, &cfg,
512                                     &detect_zeroes, &error);
513     if (error) {
514         error_propagate(errp, error);
515         goto early_err;
516     }
517 
518     if ((buf = qemu_opt_get(opts, "format")) != NULL) {
519         if (is_help_option(buf)) {
520             error_printf("Supported formats:");
521             bdrv_iterate_format(bdrv_format_print, NULL);
522             error_printf("\n");
523             goto early_err;
524         }
525 
526         if (qdict_haskey(bs_opts, "driver")) {
527             error_setg(errp, "Cannot specify both 'driver' and 'format'");
528             goto early_err;
529         }
530         qdict_put_str(bs_opts, "driver", buf);
531     }
532 
533     on_write_error = BLOCKDEV_ON_ERROR_ENOSPC;
534     if ((buf = qemu_opt_get(opts, "werror")) != NULL) {
535         on_write_error = parse_block_error_action(buf, 0, &error);
536         if (error) {
537             error_propagate(errp, error);
538             goto early_err;
539         }
540     }
541 
542     on_read_error = BLOCKDEV_ON_ERROR_REPORT;
543     if ((buf = qemu_opt_get(opts, "rerror")) != NULL) {
544         on_read_error = parse_block_error_action(buf, 1, &error);
545         if (error) {
546             error_propagate(errp, error);
547             goto early_err;
548         }
549     }
550 
551     if (snapshot) {
552         bdrv_flags |= BDRV_O_SNAPSHOT;
553     }
554 
555     read_only = qemu_opt_get_bool(opts, BDRV_OPT_READ_ONLY, false);
556 
557     /* init */
558     if ((!file || !*file) && !qdict_size(bs_opts)) {
559         BlockBackendRootState *blk_rs;
560 
561         blk = blk_new(0, BLK_PERM_ALL);
562         blk_rs = blk_get_root_state(blk);
563         blk_rs->open_flags    = bdrv_flags;
564         blk_rs->read_only     = read_only;
565         blk_rs->detect_zeroes = detect_zeroes;
566 
567         QDECREF(bs_opts);
568     } else {
569         if (file && !*file) {
570             file = NULL;
571         }
572 
573         /* bdrv_open() defaults to the values in bdrv_flags (for compatibility
574          * with other callers) rather than what we want as the real defaults.
575          * Apply the defaults here instead. */
576         qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_DIRECT, "off");
577         qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_NO_FLUSH, "off");
578         qdict_set_default_str(bs_opts, BDRV_OPT_READ_ONLY,
579                               read_only ? "on" : "off");
580         assert((bdrv_flags & BDRV_O_CACHE_MASK) == 0);
581 
582         if (runstate_check(RUN_STATE_INMIGRATE)) {
583             bdrv_flags |= BDRV_O_INACTIVE;
584         }
585 
586         blk = blk_new_open(file, NULL, bs_opts, bdrv_flags, errp);
587         if (!blk) {
588             goto err_no_bs_opts;
589         }
590         bs = blk_bs(blk);
591 
592         bs->detect_zeroes = detect_zeroes;
593 
594         if (bdrv_key_required(bs)) {
595             autostart = 0;
596         }
597 
598         block_acct_setup(blk_get_stats(blk), account_invalid, account_failed);
599 
600         if (!parse_stats_intervals(blk_get_stats(blk), interval_list, errp)) {
601             blk_unref(blk);
602             blk = NULL;
603             goto err_no_bs_opts;
604         }
605     }
606 
607     /* disk I/O throttling */
608     if (throttle_enabled(&cfg)) {
609         if (!throttling_group) {
610             throttling_group = id;
611         }
612         blk_io_limits_enable(blk, throttling_group);
613         blk_set_io_limits(blk, &cfg);
614     }
615 
616     blk_set_enable_write_cache(blk, !writethrough);
617     blk_set_on_error(blk, on_read_error, on_write_error);
618 
619     if (!monitor_add_blk(blk, id, errp)) {
620         blk_unref(blk);
621         blk = NULL;
622         goto err_no_bs_opts;
623     }
624 
625 err_no_bs_opts:
626     qemu_opts_del(opts);
627     QDECREF(interval_dict);
628     QDECREF(interval_list);
629     return blk;
630 
631 early_err:
632     qemu_opts_del(opts);
633     QDECREF(interval_dict);
634     QDECREF(interval_list);
635 err_no_opts:
636     QDECREF(bs_opts);
637     return NULL;
638 }
639 
640 /* Takes the ownership of bs_opts */
641 static BlockDriverState *bds_tree_init(QDict *bs_opts, Error **errp)
642 {
643     int bdrv_flags = 0;
644 
645     /* bdrv_open() defaults to the values in bdrv_flags (for compatibility
646      * with other callers) rather than what we want as the real defaults.
647      * Apply the defaults here instead. */
648     qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_DIRECT, "off");
649     qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_NO_FLUSH, "off");
650     qdict_set_default_str(bs_opts, BDRV_OPT_READ_ONLY, "off");
651 
652     if (runstate_check(RUN_STATE_INMIGRATE)) {
653         bdrv_flags |= BDRV_O_INACTIVE;
654     }
655 
656     return bdrv_open(NULL, NULL, bs_opts, bdrv_flags, errp);
657 }
658 
659 void blockdev_close_all_bdrv_states(void)
660 {
661     BlockDriverState *bs, *next_bs;
662 
663     QTAILQ_FOREACH_SAFE(bs, &monitor_bdrv_states, monitor_list, next_bs) {
664         AioContext *ctx = bdrv_get_aio_context(bs);
665 
666         aio_context_acquire(ctx);
667         bdrv_unref(bs);
668         aio_context_release(ctx);
669     }
670 }
671 
672 /* Iterates over the list of monitor-owned BlockDriverStates */
673 BlockDriverState *bdrv_next_monitor_owned(BlockDriverState *bs)
674 {
675     return bs ? QTAILQ_NEXT(bs, monitor_list)
676               : QTAILQ_FIRST(&monitor_bdrv_states);
677 }
678 
679 static void qemu_opt_rename(QemuOpts *opts, const char *from, const char *to,
680                             Error **errp)
681 {
682     const char *value;
683 
684     value = qemu_opt_get(opts, from);
685     if (value) {
686         if (qemu_opt_find(opts, to)) {
687             error_setg(errp, "'%s' and its alias '%s' can't be used at the "
688                        "same time", to, from);
689             return;
690         }
691     }
692 
693     /* rename all items in opts */
694     while ((value = qemu_opt_get(opts, from))) {
695         qemu_opt_set(opts, to, value, &error_abort);
696         qemu_opt_unset(opts, from);
697     }
698 }
699 
700 QemuOptsList qemu_legacy_drive_opts = {
701     .name = "drive",
702     .head = QTAILQ_HEAD_INITIALIZER(qemu_legacy_drive_opts.head),
703     .desc = {
704         {
705             .name = "bus",
706             .type = QEMU_OPT_NUMBER,
707             .help = "bus number",
708         },{
709             .name = "unit",
710             .type = QEMU_OPT_NUMBER,
711             .help = "unit number (i.e. lun for scsi)",
712         },{
713             .name = "index",
714             .type = QEMU_OPT_NUMBER,
715             .help = "index number",
716         },{
717             .name = "media",
718             .type = QEMU_OPT_STRING,
719             .help = "media type (disk, cdrom)",
720         },{
721             .name = "if",
722             .type = QEMU_OPT_STRING,
723             .help = "interface (ide, scsi, sd, mtd, floppy, pflash, virtio)",
724         },{
725             .name = "cyls",
726             .type = QEMU_OPT_NUMBER,
727             .help = "number of cylinders (ide disk geometry)",
728         },{
729             .name = "heads",
730             .type = QEMU_OPT_NUMBER,
731             .help = "number of heads (ide disk geometry)",
732         },{
733             .name = "secs",
734             .type = QEMU_OPT_NUMBER,
735             .help = "number of sectors (ide disk geometry)",
736         },{
737             .name = "trans",
738             .type = QEMU_OPT_STRING,
739             .help = "chs translation (auto, lba, none)",
740         },{
741             .name = "boot",
742             .type = QEMU_OPT_BOOL,
743             .help = "(deprecated, ignored)",
744         },{
745             .name = "addr",
746             .type = QEMU_OPT_STRING,
747             .help = "pci address (virtio only)",
748         },{
749             .name = "serial",
750             .type = QEMU_OPT_STRING,
751             .help = "disk serial number",
752         },{
753             .name = "file",
754             .type = QEMU_OPT_STRING,
755             .help = "file name",
756         },
757 
758         /* Options that are passed on, but have special semantics with -drive */
759         {
760             .name = BDRV_OPT_READ_ONLY,
761             .type = QEMU_OPT_BOOL,
762             .help = "open drive file as read-only",
763         },{
764             .name = "rerror",
765             .type = QEMU_OPT_STRING,
766             .help = "read error action",
767         },{
768             .name = "werror",
769             .type = QEMU_OPT_STRING,
770             .help = "write error action",
771         },{
772             .name = "copy-on-read",
773             .type = QEMU_OPT_BOOL,
774             .help = "copy read data from backing file into image file",
775         },
776 
777         { /* end of list */ }
778     },
779 };
780 
781 DriveInfo *drive_new(QemuOpts *all_opts, BlockInterfaceType block_default_type)
782 {
783     const char *value;
784     BlockBackend *blk;
785     DriveInfo *dinfo = NULL;
786     QDict *bs_opts;
787     QemuOpts *legacy_opts;
788     DriveMediaType media = MEDIA_DISK;
789     BlockInterfaceType type;
790     int cyls, heads, secs, translation;
791     int max_devs, bus_id, unit_id, index;
792     const char *devaddr;
793     const char *werror, *rerror;
794     bool read_only = false;
795     bool copy_on_read;
796     const char *serial;
797     const char *filename;
798     Error *local_err = NULL;
799     int i;
800 
801     /* Change legacy command line options into QMP ones */
802     static const struct {
803         const char *from;
804         const char *to;
805     } opt_renames[] = {
806         { "iops",           "throttling.iops-total" },
807         { "iops_rd",        "throttling.iops-read" },
808         { "iops_wr",        "throttling.iops-write" },
809 
810         { "bps",            "throttling.bps-total" },
811         { "bps_rd",         "throttling.bps-read" },
812         { "bps_wr",         "throttling.bps-write" },
813 
814         { "iops_max",       "throttling.iops-total-max" },
815         { "iops_rd_max",    "throttling.iops-read-max" },
816         { "iops_wr_max",    "throttling.iops-write-max" },
817 
818         { "bps_max",        "throttling.bps-total-max" },
819         { "bps_rd_max",     "throttling.bps-read-max" },
820         { "bps_wr_max",     "throttling.bps-write-max" },
821 
822         { "iops_size",      "throttling.iops-size" },
823 
824         { "group",          "throttling.group" },
825 
826         { "readonly",       BDRV_OPT_READ_ONLY },
827     };
828 
829     for (i = 0; i < ARRAY_SIZE(opt_renames); i++) {
830         qemu_opt_rename(all_opts, opt_renames[i].from, opt_renames[i].to,
831                         &local_err);
832         if (local_err) {
833             error_report_err(local_err);
834             return NULL;
835         }
836     }
837 
838     value = qemu_opt_get(all_opts, "cache");
839     if (value) {
840         int flags = 0;
841         bool writethrough;
842 
843         if (bdrv_parse_cache_mode(value, &flags, &writethrough) != 0) {
844             error_report("invalid cache option");
845             return NULL;
846         }
847 
848         /* Specific options take precedence */
849         if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_WB)) {
850             qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_WB,
851                               !writethrough, &error_abort);
852         }
853         if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_DIRECT)) {
854             qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_DIRECT,
855                               !!(flags & BDRV_O_NOCACHE), &error_abort);
856         }
857         if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_NO_FLUSH)) {
858             qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_NO_FLUSH,
859                               !!(flags & BDRV_O_NO_FLUSH), &error_abort);
860         }
861         qemu_opt_unset(all_opts, "cache");
862     }
863 
864     /* Get a QDict for processing the options */
865     bs_opts = qdict_new();
866     qemu_opts_to_qdict(all_opts, bs_opts);
867 
868     legacy_opts = qemu_opts_create(&qemu_legacy_drive_opts, NULL, 0,
869                                    &error_abort);
870     qemu_opts_absorb_qdict(legacy_opts, bs_opts, &local_err);
871     if (local_err) {
872         error_report_err(local_err);
873         goto fail;
874     }
875 
876     /* Deprecated option boot=[on|off] */
877     if (qemu_opt_get(legacy_opts, "boot") != NULL) {
878         fprintf(stderr, "qemu-kvm: boot=on|off is deprecated and will be "
879                 "ignored. Future versions will reject this parameter. Please "
880                 "update your scripts.\n");
881     }
882 
883     /* Media type */
884     value = qemu_opt_get(legacy_opts, "media");
885     if (value) {
886         if (!strcmp(value, "disk")) {
887             media = MEDIA_DISK;
888         } else if (!strcmp(value, "cdrom")) {
889             media = MEDIA_CDROM;
890             read_only = true;
891         } else {
892             error_report("'%s' invalid media", value);
893             goto fail;
894         }
895     }
896 
897     /* copy-on-read is disabled with a warning for read-only devices */
898     read_only |= qemu_opt_get_bool(legacy_opts, BDRV_OPT_READ_ONLY, false);
899     copy_on_read = qemu_opt_get_bool(legacy_opts, "copy-on-read", false);
900 
901     if (read_only && copy_on_read) {
902         error_report("warning: disabling copy-on-read on read-only drive");
903         copy_on_read = false;
904     }
905 
906     qdict_put_str(bs_opts, BDRV_OPT_READ_ONLY, read_only ? "on" : "off");
907     qdict_put_str(bs_opts, "copy-on-read", copy_on_read ? "on" : "off");
908 
909     /* Controller type */
910     value = qemu_opt_get(legacy_opts, "if");
911     if (value) {
912         for (type = 0;
913              type < IF_COUNT && strcmp(value, if_name[type]);
914              type++) {
915         }
916         if (type == IF_COUNT) {
917             error_report("unsupported bus type '%s'", value);
918             goto fail;
919         }
920     } else {
921         type = block_default_type;
922     }
923 
924     /* Geometry */
925     cyls  = qemu_opt_get_number(legacy_opts, "cyls", 0);
926     heads = qemu_opt_get_number(legacy_opts, "heads", 0);
927     secs  = qemu_opt_get_number(legacy_opts, "secs", 0);
928 
929     if (cyls || heads || secs) {
930         if (cyls < 1) {
931             error_report("invalid physical cyls number");
932             goto fail;
933         }
934         if (heads < 1) {
935             error_report("invalid physical heads number");
936             goto fail;
937         }
938         if (secs < 1) {
939             error_report("invalid physical secs number");
940             goto fail;
941         }
942     }
943 
944     translation = BIOS_ATA_TRANSLATION_AUTO;
945     value = qemu_opt_get(legacy_opts, "trans");
946     if (value != NULL) {
947         if (!cyls) {
948             error_report("'%s' trans must be used with cyls, heads and secs",
949                          value);
950             goto fail;
951         }
952         if (!strcmp(value, "none")) {
953             translation = BIOS_ATA_TRANSLATION_NONE;
954         } else if (!strcmp(value, "lba")) {
955             translation = BIOS_ATA_TRANSLATION_LBA;
956         } else if (!strcmp(value, "large")) {
957             translation = BIOS_ATA_TRANSLATION_LARGE;
958         } else if (!strcmp(value, "rechs")) {
959             translation = BIOS_ATA_TRANSLATION_RECHS;
960         } else if (!strcmp(value, "auto")) {
961             translation = BIOS_ATA_TRANSLATION_AUTO;
962         } else {
963             error_report("'%s' invalid translation type", value);
964             goto fail;
965         }
966     }
967 
968     if (media == MEDIA_CDROM) {
969         if (cyls || secs || heads) {
970             error_report("CHS can't be set with media=cdrom");
971             goto fail;
972         }
973     }
974 
975     /* Device address specified by bus/unit or index.
976      * If none was specified, try to find the first free one. */
977     bus_id  = qemu_opt_get_number(legacy_opts, "bus", 0);
978     unit_id = qemu_opt_get_number(legacy_opts, "unit", -1);
979     index   = qemu_opt_get_number(legacy_opts, "index", -1);
980 
981     max_devs = if_max_devs[type];
982 
983     if (index != -1) {
984         if (bus_id != 0 || unit_id != -1) {
985             error_report("index cannot be used with bus and unit");
986             goto fail;
987         }
988         bus_id = drive_index_to_bus_id(type, index);
989         unit_id = drive_index_to_unit_id(type, index);
990     }
991 
992     if (unit_id == -1) {
993        unit_id = 0;
994        while (drive_get(type, bus_id, unit_id) != NULL) {
995            unit_id++;
996            if (max_devs && unit_id >= max_devs) {
997                unit_id -= max_devs;
998                bus_id++;
999            }
1000        }
1001     }
1002 
1003     if (max_devs && unit_id >= max_devs) {
1004         error_report("unit %d too big (max is %d)", unit_id, max_devs - 1);
1005         goto fail;
1006     }
1007 
1008     if (drive_get(type, bus_id, unit_id) != NULL) {
1009         error_report("drive with bus=%d, unit=%d (index=%d) exists",
1010                      bus_id, unit_id, index);
1011         goto fail;
1012     }
1013 
1014     /* Serial number */
1015     serial = qemu_opt_get(legacy_opts, "serial");
1016 
1017     /* no id supplied -> create one */
1018     if (qemu_opts_id(all_opts) == NULL) {
1019         char *new_id;
1020         const char *mediastr = "";
1021         if (type == IF_IDE || type == IF_SCSI) {
1022             mediastr = (media == MEDIA_CDROM) ? "-cd" : "-hd";
1023         }
1024         if (max_devs) {
1025             new_id = g_strdup_printf("%s%i%s%i", if_name[type], bus_id,
1026                                      mediastr, unit_id);
1027         } else {
1028             new_id = g_strdup_printf("%s%s%i", if_name[type],
1029                                      mediastr, unit_id);
1030         }
1031         qdict_put_str(bs_opts, "id", new_id);
1032         g_free(new_id);
1033     }
1034 
1035     /* Add virtio block device */
1036     devaddr = qemu_opt_get(legacy_opts, "addr");
1037     if (devaddr && type != IF_VIRTIO) {
1038         error_report("addr is not supported by this bus type");
1039         goto fail;
1040     }
1041 
1042     if (type == IF_VIRTIO) {
1043         QemuOpts *devopts;
1044         devopts = qemu_opts_create(qemu_find_opts("device"), NULL, 0,
1045                                    &error_abort);
1046         if (arch_type == QEMU_ARCH_S390X) {
1047             qemu_opt_set(devopts, "driver", "virtio-blk-ccw", &error_abort);
1048         } else {
1049             qemu_opt_set(devopts, "driver", "virtio-blk-pci", &error_abort);
1050         }
1051         qemu_opt_set(devopts, "drive", qdict_get_str(bs_opts, "id"),
1052                      &error_abort);
1053         if (devaddr) {
1054             qemu_opt_set(devopts, "addr", devaddr, &error_abort);
1055         }
1056     }
1057 
1058     filename = qemu_opt_get(legacy_opts, "file");
1059 
1060     /* Check werror/rerror compatibility with if=... */
1061     werror = qemu_opt_get(legacy_opts, "werror");
1062     if (werror != NULL) {
1063         if (type != IF_IDE && type != IF_SCSI && type != IF_VIRTIO &&
1064             type != IF_NONE) {
1065             error_report("werror is not supported by this bus type");
1066             goto fail;
1067         }
1068         qdict_put_str(bs_opts, "werror", werror);
1069     }
1070 
1071     rerror = qemu_opt_get(legacy_opts, "rerror");
1072     if (rerror != NULL) {
1073         if (type != IF_IDE && type != IF_VIRTIO && type != IF_SCSI &&
1074             type != IF_NONE) {
1075             error_report("rerror is not supported by this bus type");
1076             goto fail;
1077         }
1078         qdict_put_str(bs_opts, "rerror", rerror);
1079     }
1080 
1081     /* Actual block device init: Functionality shared with blockdev-add */
1082     blk = blockdev_init(filename, bs_opts, &local_err);
1083     bs_opts = NULL;
1084     if (!blk) {
1085         if (local_err) {
1086             error_report_err(local_err);
1087         }
1088         goto fail;
1089     } else {
1090         assert(!local_err);
1091     }
1092 
1093     /* Create legacy DriveInfo */
1094     dinfo = g_malloc0(sizeof(*dinfo));
1095     dinfo->opts = all_opts;
1096 
1097     dinfo->cyls = cyls;
1098     dinfo->heads = heads;
1099     dinfo->secs = secs;
1100     dinfo->trans = translation;
1101 
1102     dinfo->type = type;
1103     dinfo->bus = bus_id;
1104     dinfo->unit = unit_id;
1105     dinfo->devaddr = devaddr;
1106     dinfo->serial = g_strdup(serial);
1107 
1108     blk_set_legacy_dinfo(blk, dinfo);
1109 
1110     switch(type) {
1111     case IF_IDE:
1112     case IF_SCSI:
1113     case IF_XEN:
1114     case IF_NONE:
1115         dinfo->media_cd = media == MEDIA_CDROM;
1116         break;
1117     default:
1118         break;
1119     }
1120 
1121 fail:
1122     qemu_opts_del(legacy_opts);
1123     QDECREF(bs_opts);
1124     return dinfo;
1125 }
1126 
1127 static BlockDriverState *qmp_get_root_bs(const char *name, Error **errp)
1128 {
1129     BlockDriverState *bs;
1130 
1131     bs = bdrv_lookup_bs(name, name, errp);
1132     if (bs == NULL) {
1133         return NULL;
1134     }
1135 
1136     if (!bdrv_is_root_node(bs)) {
1137         error_setg(errp, "Need a root block node");
1138         return NULL;
1139     }
1140 
1141     if (!bdrv_is_inserted(bs)) {
1142         error_setg(errp, "Device has no medium");
1143         return NULL;
1144     }
1145 
1146     return bs;
1147 }
1148 
1149 static BlockBackend *qmp_get_blk(const char *blk_name, const char *qdev_id,
1150                                  Error **errp)
1151 {
1152     BlockBackend *blk;
1153 
1154     if (!blk_name == !qdev_id) {
1155         error_setg(errp, "Need exactly one of 'device' and 'id'");
1156         return NULL;
1157     }
1158 
1159     if (qdev_id) {
1160         blk = blk_by_qdev_id(qdev_id, errp);
1161     } else {
1162         blk = blk_by_name(blk_name);
1163         if (blk == NULL) {
1164             error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
1165                       "Device '%s' not found", blk_name);
1166         }
1167     }
1168 
1169     return blk;
1170 }
1171 
1172 void hmp_commit(Monitor *mon, const QDict *qdict)
1173 {
1174     const char *device = qdict_get_str(qdict, "device");
1175     BlockBackend *blk;
1176     int ret;
1177 
1178     if (!strcmp(device, "all")) {
1179         ret = blk_commit_all();
1180     } else {
1181         BlockDriverState *bs;
1182         AioContext *aio_context;
1183 
1184         blk = blk_by_name(device);
1185         if (!blk) {
1186             monitor_printf(mon, "Device '%s' not found\n", device);
1187             return;
1188         }
1189         if (!blk_is_available(blk)) {
1190             monitor_printf(mon, "Device '%s' has no medium\n", device);
1191             return;
1192         }
1193 
1194         bs = blk_bs(blk);
1195         aio_context = bdrv_get_aio_context(bs);
1196         aio_context_acquire(aio_context);
1197 
1198         ret = bdrv_commit(bs);
1199 
1200         aio_context_release(aio_context);
1201     }
1202     if (ret < 0) {
1203         monitor_printf(mon, "'commit' error for '%s': %s\n", device,
1204                        strerror(-ret));
1205     }
1206 }
1207 
1208 static void blockdev_do_action(TransactionAction *action, Error **errp)
1209 {
1210     TransactionActionList list;
1211 
1212     list.value = action;
1213     list.next = NULL;
1214     qmp_transaction(&list, false, NULL, errp);
1215 }
1216 
1217 void qmp_blockdev_snapshot_sync(bool has_device, const char *device,
1218                                 bool has_node_name, const char *node_name,
1219                                 const char *snapshot_file,
1220                                 bool has_snapshot_node_name,
1221                                 const char *snapshot_node_name,
1222                                 bool has_format, const char *format,
1223                                 bool has_mode, NewImageMode mode, Error **errp)
1224 {
1225     BlockdevSnapshotSync snapshot = {
1226         .has_device = has_device,
1227         .device = (char *) device,
1228         .has_node_name = has_node_name,
1229         .node_name = (char *) node_name,
1230         .snapshot_file = (char *) snapshot_file,
1231         .has_snapshot_node_name = has_snapshot_node_name,
1232         .snapshot_node_name = (char *) snapshot_node_name,
1233         .has_format = has_format,
1234         .format = (char *) format,
1235         .has_mode = has_mode,
1236         .mode = mode,
1237     };
1238     TransactionAction action = {
1239         .type = TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC,
1240         .u.blockdev_snapshot_sync.data = &snapshot,
1241     };
1242     blockdev_do_action(&action, errp);
1243 }
1244 
1245 void qmp_blockdev_snapshot(const char *node, const char *overlay,
1246                            Error **errp)
1247 {
1248     BlockdevSnapshot snapshot_data = {
1249         .node = (char *) node,
1250         .overlay = (char *) overlay
1251     };
1252     TransactionAction action = {
1253         .type = TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT,
1254         .u.blockdev_snapshot.data = &snapshot_data,
1255     };
1256     blockdev_do_action(&action, errp);
1257 }
1258 
1259 void qmp_blockdev_snapshot_internal_sync(const char *device,
1260                                          const char *name,
1261                                          Error **errp)
1262 {
1263     BlockdevSnapshotInternal snapshot = {
1264         .device = (char *) device,
1265         .name = (char *) name
1266     };
1267     TransactionAction action = {
1268         .type = TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC,
1269         .u.blockdev_snapshot_internal_sync.data = &snapshot,
1270     };
1271     blockdev_do_action(&action, errp);
1272 }
1273 
1274 SnapshotInfo *qmp_blockdev_snapshot_delete_internal_sync(const char *device,
1275                                                          bool has_id,
1276                                                          const char *id,
1277                                                          bool has_name,
1278                                                          const char *name,
1279                                                          Error **errp)
1280 {
1281     BlockDriverState *bs;
1282     AioContext *aio_context;
1283     QEMUSnapshotInfo sn;
1284     Error *local_err = NULL;
1285     SnapshotInfo *info = NULL;
1286     int ret;
1287 
1288     bs = qmp_get_root_bs(device, errp);
1289     if (!bs) {
1290         return NULL;
1291     }
1292     aio_context = bdrv_get_aio_context(bs);
1293     aio_context_acquire(aio_context);
1294 
1295     if (!has_id) {
1296         id = NULL;
1297     }
1298 
1299     if (!has_name) {
1300         name = NULL;
1301     }
1302 
1303     if (!id && !name) {
1304         error_setg(errp, "Name or id must be provided");
1305         goto out_aio_context;
1306     }
1307 
1308     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_INTERNAL_SNAPSHOT_DELETE, errp)) {
1309         goto out_aio_context;
1310     }
1311 
1312     ret = bdrv_snapshot_find_by_id_and_name(bs, id, name, &sn, &local_err);
1313     if (local_err) {
1314         error_propagate(errp, local_err);
1315         goto out_aio_context;
1316     }
1317     if (!ret) {
1318         error_setg(errp,
1319                    "Snapshot with id '%s' and name '%s' does not exist on "
1320                    "device '%s'",
1321                    STR_OR_NULL(id), STR_OR_NULL(name), device);
1322         goto out_aio_context;
1323     }
1324 
1325     bdrv_snapshot_delete(bs, id, name, &local_err);
1326     if (local_err) {
1327         error_propagate(errp, local_err);
1328         goto out_aio_context;
1329     }
1330 
1331     aio_context_release(aio_context);
1332 
1333     info = g_new0(SnapshotInfo, 1);
1334     info->id = g_strdup(sn.id_str);
1335     info->name = g_strdup(sn.name);
1336     info->date_nsec = sn.date_nsec;
1337     info->date_sec = sn.date_sec;
1338     info->vm_state_size = sn.vm_state_size;
1339     info->vm_clock_nsec = sn.vm_clock_nsec % 1000000000;
1340     info->vm_clock_sec = sn.vm_clock_nsec / 1000000000;
1341 
1342     return info;
1343 
1344 out_aio_context:
1345     aio_context_release(aio_context);
1346     return NULL;
1347 }
1348 
1349 /**
1350  * block_dirty_bitmap_lookup:
1351  * Return a dirty bitmap (if present), after validating
1352  * the node reference and bitmap names.
1353  *
1354  * @node: The name of the BDS node to search for bitmaps
1355  * @name: The name of the bitmap to search for
1356  * @pbs: Output pointer for BDS lookup, if desired. Can be NULL.
1357  * @paio: Output pointer for aio_context acquisition, if desired. Can be NULL.
1358  * @errp: Output pointer for error information. Can be NULL.
1359  *
1360  * @return: A bitmap object on success, or NULL on failure.
1361  */
1362 static BdrvDirtyBitmap *block_dirty_bitmap_lookup(const char *node,
1363                                                   const char *name,
1364                                                   BlockDriverState **pbs,
1365                                                   Error **errp)
1366 {
1367     BlockDriverState *bs;
1368     BdrvDirtyBitmap *bitmap;
1369 
1370     if (!node) {
1371         error_setg(errp, "Node cannot be NULL");
1372         return NULL;
1373     }
1374     if (!name) {
1375         error_setg(errp, "Bitmap name cannot be NULL");
1376         return NULL;
1377     }
1378     bs = bdrv_lookup_bs(node, node, NULL);
1379     if (!bs) {
1380         error_setg(errp, "Node '%s' not found", node);
1381         return NULL;
1382     }
1383 
1384     bitmap = bdrv_find_dirty_bitmap(bs, name);
1385     if (!bitmap) {
1386         error_setg(errp, "Dirty bitmap '%s' not found", name);
1387         return NULL;
1388     }
1389 
1390     if (pbs) {
1391         *pbs = bs;
1392     }
1393 
1394     return bitmap;
1395 }
1396 
1397 /* New and old BlockDriverState structs for atomic group operations */
1398 
1399 typedef struct BlkActionState BlkActionState;
1400 
1401 /**
1402  * BlkActionOps:
1403  * Table of operations that define an Action.
1404  *
1405  * @instance_size: Size of state struct, in bytes.
1406  * @prepare: Prepare the work, must NOT be NULL.
1407  * @commit: Commit the changes, can be NULL.
1408  * @abort: Abort the changes on fail, can be NULL.
1409  * @clean: Clean up resources after all transaction actions have called
1410  *         commit() or abort(). Can be NULL.
1411  *
1412  * Only prepare() may fail. In a single transaction, only one of commit() or
1413  * abort() will be called. clean() will always be called if it is present.
1414  */
1415 typedef struct BlkActionOps {
1416     size_t instance_size;
1417     void (*prepare)(BlkActionState *common, Error **errp);
1418     void (*commit)(BlkActionState *common);
1419     void (*abort)(BlkActionState *common);
1420     void (*clean)(BlkActionState *common);
1421 } BlkActionOps;
1422 
1423 /**
1424  * BlkActionState:
1425  * Describes one Action's state within a Transaction.
1426  *
1427  * @action: QAPI-defined enum identifying which Action to perform.
1428  * @ops: Table of ActionOps this Action can perform.
1429  * @block_job_txn: Transaction which this action belongs to.
1430  * @entry: List membership for all Actions in this Transaction.
1431  *
1432  * This structure must be arranged as first member in a subclassed type,
1433  * assuming that the compiler will also arrange it to the same offsets as the
1434  * base class.
1435  */
1436 struct BlkActionState {
1437     TransactionAction *action;
1438     const BlkActionOps *ops;
1439     BlockJobTxn *block_job_txn;
1440     TransactionProperties *txn_props;
1441     QSIMPLEQ_ENTRY(BlkActionState) entry;
1442 };
1443 
1444 /* internal snapshot private data */
1445 typedef struct InternalSnapshotState {
1446     BlkActionState common;
1447     BlockDriverState *bs;
1448     AioContext *aio_context;
1449     QEMUSnapshotInfo sn;
1450     bool created;
1451 } InternalSnapshotState;
1452 
1453 
1454 static int action_check_completion_mode(BlkActionState *s, Error **errp)
1455 {
1456     if (s->txn_props->completion_mode != ACTION_COMPLETION_MODE_INDIVIDUAL) {
1457         error_setg(errp,
1458                    "Action '%s' does not support Transaction property "
1459                    "completion-mode = %s",
1460                    TransactionActionKind_lookup[s->action->type],
1461                    ActionCompletionMode_lookup[s->txn_props->completion_mode]);
1462         return -1;
1463     }
1464     return 0;
1465 }
1466 
1467 static void internal_snapshot_prepare(BlkActionState *common,
1468                                       Error **errp)
1469 {
1470     Error *local_err = NULL;
1471     const char *device;
1472     const char *name;
1473     BlockDriverState *bs;
1474     QEMUSnapshotInfo old_sn, *sn;
1475     bool ret;
1476     qemu_timeval tv;
1477     BlockdevSnapshotInternal *internal;
1478     InternalSnapshotState *state;
1479     int ret1;
1480 
1481     g_assert(common->action->type ==
1482              TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC);
1483     internal = common->action->u.blockdev_snapshot_internal_sync.data;
1484     state = DO_UPCAST(InternalSnapshotState, common, common);
1485 
1486     /* 1. parse input */
1487     device = internal->device;
1488     name = internal->name;
1489 
1490     /* 2. check for validation */
1491     if (action_check_completion_mode(common, errp) < 0) {
1492         return;
1493     }
1494 
1495     bs = qmp_get_root_bs(device, errp);
1496     if (!bs) {
1497         return;
1498     }
1499 
1500     /* AioContext is released in .clean() */
1501     state->aio_context = bdrv_get_aio_context(bs);
1502     aio_context_acquire(state->aio_context);
1503 
1504     state->bs = bs;
1505     bdrv_drained_begin(bs);
1506 
1507     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_INTERNAL_SNAPSHOT, errp)) {
1508         return;
1509     }
1510 
1511     if (bdrv_is_read_only(bs)) {
1512         error_setg(errp, "Device '%s' is read only", device);
1513         return;
1514     }
1515 
1516     if (!bdrv_can_snapshot(bs)) {
1517         error_setg(errp, "Block format '%s' used by device '%s' "
1518                    "does not support internal snapshots",
1519                    bs->drv->format_name, device);
1520         return;
1521     }
1522 
1523     if (!strlen(name)) {
1524         error_setg(errp, "Name is empty");
1525         return;
1526     }
1527 
1528     /* check whether a snapshot with name exist */
1529     ret = bdrv_snapshot_find_by_id_and_name(bs, NULL, name, &old_sn,
1530                                             &local_err);
1531     if (local_err) {
1532         error_propagate(errp, local_err);
1533         return;
1534     } else if (ret) {
1535         error_setg(errp,
1536                    "Snapshot with name '%s' already exists on device '%s'",
1537                    name, device);
1538         return;
1539     }
1540 
1541     /* 3. take the snapshot */
1542     sn = &state->sn;
1543     pstrcpy(sn->name, sizeof(sn->name), name);
1544     qemu_gettimeofday(&tv);
1545     sn->date_sec = tv.tv_sec;
1546     sn->date_nsec = tv.tv_usec * 1000;
1547     sn->vm_clock_nsec = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL);
1548 
1549     ret1 = bdrv_snapshot_create(bs, sn);
1550     if (ret1 < 0) {
1551         error_setg_errno(errp, -ret1,
1552                          "Failed to create snapshot '%s' on device '%s'",
1553                          name, device);
1554         return;
1555     }
1556 
1557     /* 4. succeed, mark a snapshot is created */
1558     state->created = true;
1559 }
1560 
1561 static void internal_snapshot_abort(BlkActionState *common)
1562 {
1563     InternalSnapshotState *state =
1564                              DO_UPCAST(InternalSnapshotState, common, common);
1565     BlockDriverState *bs = state->bs;
1566     QEMUSnapshotInfo *sn = &state->sn;
1567     Error *local_error = NULL;
1568 
1569     if (!state->created) {
1570         return;
1571     }
1572 
1573     if (bdrv_snapshot_delete(bs, sn->id_str, sn->name, &local_error) < 0) {
1574         error_reportf_err(local_error,
1575                           "Failed to delete snapshot with id '%s' and "
1576                           "name '%s' on device '%s' in abort: ",
1577                           sn->id_str, sn->name,
1578                           bdrv_get_device_name(bs));
1579     }
1580 }
1581 
1582 static void internal_snapshot_clean(BlkActionState *common)
1583 {
1584     InternalSnapshotState *state = DO_UPCAST(InternalSnapshotState,
1585                                              common, common);
1586 
1587     if (state->aio_context) {
1588         if (state->bs) {
1589             bdrv_drained_end(state->bs);
1590         }
1591         aio_context_release(state->aio_context);
1592     }
1593 }
1594 
1595 /* external snapshot private data */
1596 typedef struct ExternalSnapshotState {
1597     BlkActionState common;
1598     BlockDriverState *old_bs;
1599     BlockDriverState *new_bs;
1600     AioContext *aio_context;
1601     bool overlay_appended;
1602 } ExternalSnapshotState;
1603 
1604 static void external_snapshot_prepare(BlkActionState *common,
1605                                       Error **errp)
1606 {
1607     int flags = 0;
1608     QDict *options = NULL;
1609     Error *local_err = NULL;
1610     /* Device and node name of the image to generate the snapshot from */
1611     const char *device;
1612     const char *node_name;
1613     /* Reference to the new image (for 'blockdev-snapshot') */
1614     const char *snapshot_ref;
1615     /* File name of the new image (for 'blockdev-snapshot-sync') */
1616     const char *new_image_file;
1617     ExternalSnapshotState *state =
1618                              DO_UPCAST(ExternalSnapshotState, common, common);
1619     TransactionAction *action = common->action;
1620 
1621     /* 'blockdev-snapshot' and 'blockdev-snapshot-sync' have similar
1622      * purpose but a different set of parameters */
1623     switch (action->type) {
1624     case TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT:
1625         {
1626             BlockdevSnapshot *s = action->u.blockdev_snapshot.data;
1627             device = s->node;
1628             node_name = s->node;
1629             new_image_file = NULL;
1630             snapshot_ref = s->overlay;
1631         }
1632         break;
1633     case TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC:
1634         {
1635             BlockdevSnapshotSync *s = action->u.blockdev_snapshot_sync.data;
1636             device = s->has_device ? s->device : NULL;
1637             node_name = s->has_node_name ? s->node_name : NULL;
1638             new_image_file = s->snapshot_file;
1639             snapshot_ref = NULL;
1640         }
1641         break;
1642     default:
1643         g_assert_not_reached();
1644     }
1645 
1646     /* start processing */
1647     if (action_check_completion_mode(common, errp) < 0) {
1648         return;
1649     }
1650 
1651     state->old_bs = bdrv_lookup_bs(device, node_name, errp);
1652     if (!state->old_bs) {
1653         return;
1654     }
1655 
1656     /* Acquire AioContext now so any threads operating on old_bs stop */
1657     state->aio_context = bdrv_get_aio_context(state->old_bs);
1658     aio_context_acquire(state->aio_context);
1659     bdrv_drained_begin(state->old_bs);
1660 
1661     if (!bdrv_is_inserted(state->old_bs)) {
1662         error_setg(errp, QERR_DEVICE_HAS_NO_MEDIUM, device);
1663         return;
1664     }
1665 
1666     if (bdrv_op_is_blocked(state->old_bs,
1667                            BLOCK_OP_TYPE_EXTERNAL_SNAPSHOT, errp)) {
1668         return;
1669     }
1670 
1671     if (!bdrv_is_read_only(state->old_bs)) {
1672         if (bdrv_flush(state->old_bs)) {
1673             error_setg(errp, QERR_IO_ERROR);
1674             return;
1675         }
1676     }
1677 
1678     if (!bdrv_is_first_non_filter(state->old_bs)) {
1679         error_setg(errp, QERR_FEATURE_DISABLED, "snapshot");
1680         return;
1681     }
1682 
1683     if (action->type == TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC) {
1684         BlockdevSnapshotSync *s = action->u.blockdev_snapshot_sync.data;
1685         const char *format = s->has_format ? s->format : "qcow2";
1686         enum NewImageMode mode;
1687         const char *snapshot_node_name =
1688             s->has_snapshot_node_name ? s->snapshot_node_name : NULL;
1689 
1690         if (node_name && !snapshot_node_name) {
1691             error_setg(errp, "New snapshot node name missing");
1692             return;
1693         }
1694 
1695         if (snapshot_node_name &&
1696             bdrv_lookup_bs(snapshot_node_name, snapshot_node_name, NULL)) {
1697             error_setg(errp, "New snapshot node name already in use");
1698             return;
1699         }
1700 
1701         flags = state->old_bs->open_flags;
1702         flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_COPY_ON_READ);
1703 
1704         /* create new image w/backing file */
1705         mode = s->has_mode ? s->mode : NEW_IMAGE_MODE_ABSOLUTE_PATHS;
1706         if (mode != NEW_IMAGE_MODE_EXISTING) {
1707             int64_t size = bdrv_getlength(state->old_bs);
1708             if (size < 0) {
1709                 error_setg_errno(errp, -size, "bdrv_getlength failed");
1710                 return;
1711             }
1712             bdrv_img_create(new_image_file, format,
1713                             state->old_bs->filename,
1714                             state->old_bs->drv->format_name,
1715                             NULL, size, flags, false, &local_err);
1716             if (local_err) {
1717                 error_propagate(errp, local_err);
1718                 return;
1719             }
1720         }
1721 
1722         options = qdict_new();
1723         if (s->has_snapshot_node_name) {
1724             qdict_put_str(options, "node-name", snapshot_node_name);
1725         }
1726         qdict_put_str(options, "driver", format);
1727 
1728         flags |= BDRV_O_NO_BACKING;
1729     }
1730 
1731     state->new_bs = bdrv_open(new_image_file, snapshot_ref, options, flags,
1732                               errp);
1733     /* We will manually add the backing_hd field to the bs later */
1734     if (!state->new_bs) {
1735         return;
1736     }
1737 
1738     if (bdrv_has_blk(state->new_bs)) {
1739         error_setg(errp, "The snapshot is already in use");
1740         return;
1741     }
1742 
1743     if (bdrv_op_is_blocked(state->new_bs, BLOCK_OP_TYPE_EXTERNAL_SNAPSHOT,
1744                            errp)) {
1745         return;
1746     }
1747 
1748     if (state->new_bs->backing != NULL) {
1749         error_setg(errp, "The snapshot already has a backing image");
1750         return;
1751     }
1752 
1753     if (!state->new_bs->drv->supports_backing) {
1754         error_setg(errp, "The snapshot does not support backing images");
1755         return;
1756     }
1757 
1758     bdrv_set_aio_context(state->new_bs, state->aio_context);
1759 
1760     /* This removes our old bs and adds the new bs. This is an operation that
1761      * can fail, so we need to do it in .prepare; undoing it for abort is
1762      * always possible. */
1763     bdrv_ref(state->new_bs);
1764     bdrv_append(state->new_bs, state->old_bs, &local_err);
1765     if (local_err) {
1766         error_propagate(errp, local_err);
1767         return;
1768     }
1769     state->overlay_appended = true;
1770 }
1771 
1772 static void external_snapshot_commit(BlkActionState *common)
1773 {
1774     ExternalSnapshotState *state =
1775                              DO_UPCAST(ExternalSnapshotState, common, common);
1776 
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     if (!atomic_read(&state->old_bs->copy_on_read)) {
1781         bdrv_reopen(state->old_bs, state->old_bs->open_flags & ~BDRV_O_RDWR,
1782                     NULL);
1783     }
1784 }
1785 
1786 static void external_snapshot_abort(BlkActionState *common)
1787 {
1788     ExternalSnapshotState *state =
1789                              DO_UPCAST(ExternalSnapshotState, common, common);
1790     if (state->new_bs) {
1791         if (state->overlay_appended) {
1792             bdrv_ref(state->old_bs);   /* we can't let bdrv_set_backind_hd()
1793                                           close state->old_bs; we need it */
1794             bdrv_set_backing_hd(state->new_bs, NULL, &error_abort);
1795             bdrv_replace_node(state->new_bs, state->old_bs, &error_abort);
1796             bdrv_unref(state->old_bs); /* bdrv_replace_node() ref'ed old_bs */
1797         }
1798     }
1799 }
1800 
1801 static void external_snapshot_clean(BlkActionState *common)
1802 {
1803     ExternalSnapshotState *state =
1804                              DO_UPCAST(ExternalSnapshotState, common, common);
1805     if (state->aio_context) {
1806         bdrv_drained_end(state->old_bs);
1807         aio_context_release(state->aio_context);
1808         bdrv_unref(state->new_bs);
1809     }
1810 }
1811 
1812 typedef struct DriveBackupState {
1813     BlkActionState common;
1814     BlockDriverState *bs;
1815     AioContext *aio_context;
1816     BlockJob *job;
1817 } DriveBackupState;
1818 
1819 static BlockJob *do_drive_backup(DriveBackup *backup, BlockJobTxn *txn,
1820                             Error **errp);
1821 
1822 static void drive_backup_prepare(BlkActionState *common, Error **errp)
1823 {
1824     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1825     BlockDriverState *bs;
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.data;
1831 
1832     bs = qmp_get_root_bs(backup->device, errp);
1833     if (!bs) {
1834         return;
1835     }
1836 
1837     /* AioContext is released in .clean() */
1838     state->aio_context = bdrv_get_aio_context(bs);
1839     aio_context_acquire(state->aio_context);
1840     bdrv_drained_begin(bs);
1841     state->bs = bs;
1842 
1843     state->job = do_drive_backup(backup, common->block_job_txn, &local_err);
1844     if (local_err) {
1845         error_propagate(errp, local_err);
1846         return;
1847     }
1848 }
1849 
1850 static void drive_backup_commit(BlkActionState *common)
1851 {
1852     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1853     assert(state->job);
1854     block_job_start(state->job);
1855 }
1856 
1857 static void drive_backup_abort(BlkActionState *common)
1858 {
1859     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1860 
1861     if (state->job) {
1862         block_job_cancel_sync(state->job);
1863     }
1864 }
1865 
1866 static void drive_backup_clean(BlkActionState *common)
1867 {
1868     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1869 
1870     if (state->aio_context) {
1871         bdrv_drained_end(state->bs);
1872         aio_context_release(state->aio_context);
1873     }
1874 }
1875 
1876 typedef struct BlockdevBackupState {
1877     BlkActionState common;
1878     BlockDriverState *bs;
1879     BlockJob *job;
1880     AioContext *aio_context;
1881 } BlockdevBackupState;
1882 
1883 static BlockJob *do_blockdev_backup(BlockdevBackup *backup, BlockJobTxn *txn,
1884                                     Error **errp);
1885 
1886 static void blockdev_backup_prepare(BlkActionState *common, Error **errp)
1887 {
1888     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1889     BlockdevBackup *backup;
1890     BlockDriverState *bs, *target;
1891     Error *local_err = NULL;
1892 
1893     assert(common->action->type == TRANSACTION_ACTION_KIND_BLOCKDEV_BACKUP);
1894     backup = common->action->u.blockdev_backup.data;
1895 
1896     bs = qmp_get_root_bs(backup->device, errp);
1897     if (!bs) {
1898         return;
1899     }
1900 
1901     target = bdrv_lookup_bs(backup->target, backup->target, errp);
1902     if (!target) {
1903         return;
1904     }
1905 
1906     /* AioContext is released in .clean() */
1907     state->aio_context = bdrv_get_aio_context(bs);
1908     if (state->aio_context != bdrv_get_aio_context(target)) {
1909         state->aio_context = NULL;
1910         error_setg(errp, "Backup between two IO threads is not implemented");
1911         return;
1912     }
1913     aio_context_acquire(state->aio_context);
1914     state->bs = bs;
1915     bdrv_drained_begin(state->bs);
1916 
1917     state->job = do_blockdev_backup(backup, common->block_job_txn, &local_err);
1918     if (local_err) {
1919         error_propagate(errp, local_err);
1920         return;
1921     }
1922 }
1923 
1924 static void blockdev_backup_commit(BlkActionState *common)
1925 {
1926     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1927     assert(state->job);
1928     block_job_start(state->job);
1929 }
1930 
1931 static void blockdev_backup_abort(BlkActionState *common)
1932 {
1933     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1934 
1935     if (state->job) {
1936         block_job_cancel_sync(state->job);
1937     }
1938 }
1939 
1940 static void blockdev_backup_clean(BlkActionState *common)
1941 {
1942     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1943 
1944     if (state->aio_context) {
1945         bdrv_drained_end(state->bs);
1946         aio_context_release(state->aio_context);
1947     }
1948 }
1949 
1950 typedef struct BlockDirtyBitmapState {
1951     BlkActionState common;
1952     BdrvDirtyBitmap *bitmap;
1953     BlockDriverState *bs;
1954     AioContext *aio_context;
1955     HBitmap *backup;
1956     bool prepared;
1957 } BlockDirtyBitmapState;
1958 
1959 static void block_dirty_bitmap_add_prepare(BlkActionState *common,
1960                                            Error **errp)
1961 {
1962     Error *local_err = NULL;
1963     BlockDirtyBitmapAdd *action;
1964     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
1965                                              common, common);
1966 
1967     if (action_check_completion_mode(common, errp) < 0) {
1968         return;
1969     }
1970 
1971     action = common->action->u.block_dirty_bitmap_add.data;
1972     /* AIO context taken and released within qmp_block_dirty_bitmap_add */
1973     qmp_block_dirty_bitmap_add(action->node, action->name,
1974                                action->has_granularity, action->granularity,
1975                                &local_err);
1976 
1977     if (!local_err) {
1978         state->prepared = true;
1979     } else {
1980         error_propagate(errp, local_err);
1981     }
1982 }
1983 
1984 static void block_dirty_bitmap_add_abort(BlkActionState *common)
1985 {
1986     BlockDirtyBitmapAdd *action;
1987     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
1988                                              common, common);
1989 
1990     action = common->action->u.block_dirty_bitmap_add.data;
1991     /* Should not be able to fail: IF the bitmap was added via .prepare(),
1992      * then the node reference and bitmap name must have been valid.
1993      */
1994     if (state->prepared) {
1995         qmp_block_dirty_bitmap_remove(action->node, action->name, &error_abort);
1996     }
1997 }
1998 
1999 static void block_dirty_bitmap_clear_prepare(BlkActionState *common,
2000                                              Error **errp)
2001 {
2002     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2003                                              common, common);
2004     BlockDirtyBitmap *action;
2005 
2006     if (action_check_completion_mode(common, errp) < 0) {
2007         return;
2008     }
2009 
2010     action = common->action->u.block_dirty_bitmap_clear.data;
2011     state->bitmap = block_dirty_bitmap_lookup(action->node,
2012                                               action->name,
2013                                               &state->bs,
2014                                               errp);
2015     if (!state->bitmap) {
2016         return;
2017     }
2018 
2019     if (bdrv_dirty_bitmap_frozen(state->bitmap)) {
2020         error_setg(errp, "Cannot modify a frozen bitmap");
2021         return;
2022     } else if (!bdrv_dirty_bitmap_enabled(state->bitmap)) {
2023         error_setg(errp, "Cannot clear a disabled bitmap");
2024         return;
2025     }
2026 
2027     bdrv_clear_dirty_bitmap(state->bitmap, &state->backup);
2028     /* AioContext is released in .clean() */
2029 }
2030 
2031 static void block_dirty_bitmap_clear_abort(BlkActionState *common)
2032 {
2033     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2034                                              common, common);
2035 
2036     if (state->backup) {
2037         bdrv_undo_clear_dirty_bitmap(state->bitmap, state->backup);
2038     }
2039 }
2040 
2041 static void block_dirty_bitmap_clear_commit(BlkActionState *common)
2042 {
2043     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2044                                              common, common);
2045 
2046     hbitmap_free(state->backup);
2047 }
2048 
2049 static void block_dirty_bitmap_clear_clean(BlkActionState *common)
2050 {
2051     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2052                                              common, common);
2053 
2054     if (state->aio_context) {
2055         aio_context_release(state->aio_context);
2056     }
2057 }
2058 
2059 static void abort_prepare(BlkActionState *common, Error **errp)
2060 {
2061     error_setg(errp, "Transaction aborted using Abort action");
2062 }
2063 
2064 static void abort_commit(BlkActionState *common)
2065 {
2066     g_assert_not_reached(); /* this action never succeeds */
2067 }
2068 
2069 static const BlkActionOps actions[] = {
2070     [TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT] = {
2071         .instance_size = sizeof(ExternalSnapshotState),
2072         .prepare  = external_snapshot_prepare,
2073         .commit   = external_snapshot_commit,
2074         .abort = external_snapshot_abort,
2075         .clean = external_snapshot_clean,
2076     },
2077     [TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC] = {
2078         .instance_size = sizeof(ExternalSnapshotState),
2079         .prepare  = external_snapshot_prepare,
2080         .commit   = external_snapshot_commit,
2081         .abort = external_snapshot_abort,
2082         .clean = external_snapshot_clean,
2083     },
2084     [TRANSACTION_ACTION_KIND_DRIVE_BACKUP] = {
2085         .instance_size = sizeof(DriveBackupState),
2086         .prepare = drive_backup_prepare,
2087         .commit = drive_backup_commit,
2088         .abort = drive_backup_abort,
2089         .clean = drive_backup_clean,
2090     },
2091     [TRANSACTION_ACTION_KIND_BLOCKDEV_BACKUP] = {
2092         .instance_size = sizeof(BlockdevBackupState),
2093         .prepare = blockdev_backup_prepare,
2094         .commit = blockdev_backup_commit,
2095         .abort = blockdev_backup_abort,
2096         .clean = blockdev_backup_clean,
2097     },
2098     [TRANSACTION_ACTION_KIND_ABORT] = {
2099         .instance_size = sizeof(BlkActionState),
2100         .prepare = abort_prepare,
2101         .commit = abort_commit,
2102     },
2103     [TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC] = {
2104         .instance_size = sizeof(InternalSnapshotState),
2105         .prepare  = internal_snapshot_prepare,
2106         .abort = internal_snapshot_abort,
2107         .clean = internal_snapshot_clean,
2108     },
2109     [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_ADD] = {
2110         .instance_size = sizeof(BlockDirtyBitmapState),
2111         .prepare = block_dirty_bitmap_add_prepare,
2112         .abort = block_dirty_bitmap_add_abort,
2113     },
2114     [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_CLEAR] = {
2115         .instance_size = sizeof(BlockDirtyBitmapState),
2116         .prepare = block_dirty_bitmap_clear_prepare,
2117         .commit = block_dirty_bitmap_clear_commit,
2118         .abort = block_dirty_bitmap_clear_abort,
2119         .clean = block_dirty_bitmap_clear_clean,
2120     }
2121 };
2122 
2123 /**
2124  * Allocate a TransactionProperties structure if necessary, and fill
2125  * that structure with desired defaults if they are unset.
2126  */
2127 static TransactionProperties *get_transaction_properties(
2128     TransactionProperties *props)
2129 {
2130     if (!props) {
2131         props = g_new0(TransactionProperties, 1);
2132     }
2133 
2134     if (!props->has_completion_mode) {
2135         props->has_completion_mode = true;
2136         props->completion_mode = ACTION_COMPLETION_MODE_INDIVIDUAL;
2137     }
2138 
2139     return props;
2140 }
2141 
2142 /*
2143  * 'Atomic' group operations.  The operations are performed as a set, and if
2144  * any fail then we roll back all operations in the group.
2145  */
2146 void qmp_transaction(TransactionActionList *dev_list,
2147                      bool has_props,
2148                      struct TransactionProperties *props,
2149                      Error **errp)
2150 {
2151     TransactionActionList *dev_entry = dev_list;
2152     BlockJobTxn *block_job_txn = NULL;
2153     BlkActionState *state, *next;
2154     Error *local_err = NULL;
2155 
2156     QSIMPLEQ_HEAD(snap_bdrv_states, BlkActionState) snap_bdrv_states;
2157     QSIMPLEQ_INIT(&snap_bdrv_states);
2158 
2159     /* Does this transaction get canceled as a group on failure?
2160      * If not, we don't really need to make a BlockJobTxn.
2161      */
2162     props = get_transaction_properties(props);
2163     if (props->completion_mode != ACTION_COMPLETION_MODE_INDIVIDUAL) {
2164         block_job_txn = block_job_txn_new();
2165     }
2166 
2167     /* drain all i/o before any operations */
2168     bdrv_drain_all();
2169 
2170     /* We don't do anything in this loop that commits us to the operations */
2171     while (NULL != dev_entry) {
2172         TransactionAction *dev_info = NULL;
2173         const BlkActionOps *ops;
2174 
2175         dev_info = dev_entry->value;
2176         dev_entry = dev_entry->next;
2177 
2178         assert(dev_info->type < ARRAY_SIZE(actions));
2179 
2180         ops = &actions[dev_info->type];
2181         assert(ops->instance_size > 0);
2182 
2183         state = g_malloc0(ops->instance_size);
2184         state->ops = ops;
2185         state->action = dev_info;
2186         state->block_job_txn = block_job_txn;
2187         state->txn_props = props;
2188         QSIMPLEQ_INSERT_TAIL(&snap_bdrv_states, state, entry);
2189 
2190         state->ops->prepare(state, &local_err);
2191         if (local_err) {
2192             error_propagate(errp, local_err);
2193             goto delete_and_fail;
2194         }
2195     }
2196 
2197     QSIMPLEQ_FOREACH(state, &snap_bdrv_states, entry) {
2198         if (state->ops->commit) {
2199             state->ops->commit(state);
2200         }
2201     }
2202 
2203     /* success */
2204     goto exit;
2205 
2206 delete_and_fail:
2207     /* failure, and it is all-or-none; roll back all operations */
2208     QSIMPLEQ_FOREACH(state, &snap_bdrv_states, entry) {
2209         if (state->ops->abort) {
2210             state->ops->abort(state);
2211         }
2212     }
2213 exit:
2214     QSIMPLEQ_FOREACH_SAFE(state, &snap_bdrv_states, entry, next) {
2215         if (state->ops->clean) {
2216             state->ops->clean(state);
2217         }
2218         g_free(state);
2219     }
2220     if (!has_props) {
2221         qapi_free_TransactionProperties(props);
2222     }
2223     block_job_txn_unref(block_job_txn);
2224 }
2225 
2226 void qmp_eject(bool has_device, const char *device,
2227                bool has_id, const char *id,
2228                bool has_force, bool force, Error **errp)
2229 {
2230     Error *local_err = NULL;
2231     int rc;
2232 
2233     if (!has_force) {
2234         force = false;
2235     }
2236 
2237     rc = do_open_tray(has_device ? device : NULL,
2238                       has_id ? id : NULL,
2239                       force, &local_err);
2240     if (rc && rc != -ENOSYS) {
2241         error_propagate(errp, local_err);
2242         return;
2243     }
2244     error_free(local_err);
2245 
2246     qmp_x_blockdev_remove_medium(has_device, device, has_id, id, errp);
2247 }
2248 
2249 void qmp_block_passwd(bool has_device, const char *device,
2250                       bool has_node_name, const char *node_name,
2251                       const char *password, Error **errp)
2252 {
2253     Error *local_err = NULL;
2254     BlockDriverState *bs;
2255     AioContext *aio_context;
2256 
2257     bs = bdrv_lookup_bs(has_device ? device : NULL,
2258                         has_node_name ? node_name : NULL,
2259                         &local_err);
2260     if (local_err) {
2261         error_propagate(errp, local_err);
2262         return;
2263     }
2264 
2265     aio_context = bdrv_get_aio_context(bs);
2266     aio_context_acquire(aio_context);
2267 
2268     bdrv_add_key(bs, password, errp);
2269 
2270     aio_context_release(aio_context);
2271 }
2272 
2273 /*
2274  * Attempt to open the tray of @device.
2275  * If @force, ignore its tray lock.
2276  * Else, if the tray is locked, don't open it, but ask the guest to open it.
2277  * On error, store an error through @errp and return -errno.
2278  * If @device does not exist, return -ENODEV.
2279  * If it has no removable media, return -ENOTSUP.
2280  * If it has no tray, return -ENOSYS.
2281  * If the guest was asked to open the tray, return -EINPROGRESS.
2282  * Else, return 0.
2283  */
2284 static int do_open_tray(const char *blk_name, const char *qdev_id,
2285                         bool force, Error **errp)
2286 {
2287     BlockBackend *blk;
2288     const char *device = qdev_id ?: blk_name;
2289     bool locked;
2290 
2291     blk = qmp_get_blk(blk_name, qdev_id, errp);
2292     if (!blk) {
2293         return -ENODEV;
2294     }
2295 
2296     if (!blk_dev_has_removable_media(blk)) {
2297         error_setg(errp, "Device '%s' is not removable", device);
2298         return -ENOTSUP;
2299     }
2300 
2301     if (!blk_dev_has_tray(blk)) {
2302         error_setg(errp, "Device '%s' does not have a tray", device);
2303         return -ENOSYS;
2304     }
2305 
2306     if (blk_dev_is_tray_open(blk)) {
2307         return 0;
2308     }
2309 
2310     locked = blk_dev_is_medium_locked(blk);
2311     if (locked) {
2312         blk_dev_eject_request(blk, force);
2313     }
2314 
2315     if (!locked || force) {
2316         blk_dev_change_media_cb(blk, false, &error_abort);
2317     }
2318 
2319     if (locked && !force) {
2320         error_setg(errp, "Device '%s' is locked and force was not specified, "
2321                    "wait for tray to open and try again", device);
2322         return -EINPROGRESS;
2323     }
2324 
2325     return 0;
2326 }
2327 
2328 void qmp_blockdev_open_tray(bool has_device, const char *device,
2329                             bool has_id, const char *id,
2330                             bool has_force, bool force,
2331                             Error **errp)
2332 {
2333     Error *local_err = NULL;
2334     int rc;
2335 
2336     if (!has_force) {
2337         force = false;
2338     }
2339     rc = do_open_tray(has_device ? device : NULL,
2340                       has_id ? id : NULL,
2341                       force, &local_err);
2342     if (rc && rc != -ENOSYS && rc != -EINPROGRESS) {
2343         error_propagate(errp, local_err);
2344         return;
2345     }
2346     error_free(local_err);
2347 }
2348 
2349 void qmp_blockdev_close_tray(bool has_device, const char *device,
2350                              bool has_id, const char *id,
2351                              Error **errp)
2352 {
2353     BlockBackend *blk;
2354     Error *local_err = NULL;
2355 
2356     device = has_device ? device : NULL;
2357     id = has_id ? id : NULL;
2358 
2359     blk = qmp_get_blk(device, id, errp);
2360     if (!blk) {
2361         return;
2362     }
2363 
2364     if (!blk_dev_has_removable_media(blk)) {
2365         error_setg(errp, "Device '%s' is not removable", device ?: id);
2366         return;
2367     }
2368 
2369     if (!blk_dev_has_tray(blk)) {
2370         /* Ignore this command on tray-less devices */
2371         return;
2372     }
2373 
2374     if (!blk_dev_is_tray_open(blk)) {
2375         return;
2376     }
2377 
2378     blk_dev_change_media_cb(blk, true, &local_err);
2379     if (local_err) {
2380         error_propagate(errp, local_err);
2381         return;
2382     }
2383 }
2384 
2385 void qmp_x_blockdev_remove_medium(bool has_device, const char *device,
2386                                   bool has_id, const char *id, Error **errp)
2387 {
2388     BlockBackend *blk;
2389     BlockDriverState *bs;
2390     AioContext *aio_context;
2391     bool has_attached_device;
2392 
2393     device = has_device ? device : NULL;
2394     id = has_id ? id : NULL;
2395 
2396     blk = qmp_get_blk(device, id, errp);
2397     if (!blk) {
2398         return;
2399     }
2400 
2401     /* For BBs without a device, we can exchange the BDS tree at will */
2402     has_attached_device = blk_get_attached_dev(blk);
2403 
2404     if (has_attached_device && !blk_dev_has_removable_media(blk)) {
2405         error_setg(errp, "Device '%s' is not removable", device ?: id);
2406         return;
2407     }
2408 
2409     if (has_attached_device && blk_dev_has_tray(blk) &&
2410         !blk_dev_is_tray_open(blk))
2411     {
2412         error_setg(errp, "Tray of device '%s' is not open", device ?: id);
2413         return;
2414     }
2415 
2416     bs = blk_bs(blk);
2417     if (!bs) {
2418         return;
2419     }
2420 
2421     aio_context = bdrv_get_aio_context(bs);
2422     aio_context_acquire(aio_context);
2423 
2424     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_EJECT, errp)) {
2425         goto out;
2426     }
2427 
2428     blk_remove_bs(blk);
2429 
2430     if (!blk_dev_has_tray(blk)) {
2431         /* For tray-less devices, blockdev-open-tray is a no-op (or may not be
2432          * called at all); therefore, the medium needs to be ejected here.
2433          * Do it after blk_remove_bs() so blk_is_inserted(blk) returns the @load
2434          * value passed here (i.e. false). */
2435         blk_dev_change_media_cb(blk, false, &error_abort);
2436     }
2437 
2438 out:
2439     aio_context_release(aio_context);
2440 }
2441 
2442 static void qmp_blockdev_insert_anon_medium(BlockBackend *blk,
2443                                             BlockDriverState *bs, Error **errp)
2444 {
2445     Error *local_err = NULL;
2446     bool has_device;
2447     int ret;
2448 
2449     /* For BBs without a device, we can exchange the BDS tree at will */
2450     has_device = blk_get_attached_dev(blk);
2451 
2452     if (has_device && !blk_dev_has_removable_media(blk)) {
2453         error_setg(errp, "Device is not removable");
2454         return;
2455     }
2456 
2457     if (has_device && blk_dev_has_tray(blk) && !blk_dev_is_tray_open(blk)) {
2458         error_setg(errp, "Tray of the device is not open");
2459         return;
2460     }
2461 
2462     if (blk_bs(blk)) {
2463         error_setg(errp, "There already is a medium in the device");
2464         return;
2465     }
2466 
2467     ret = blk_insert_bs(blk, bs, errp);
2468     if (ret < 0) {
2469         return;
2470     }
2471 
2472     if (!blk_dev_has_tray(blk)) {
2473         /* For tray-less devices, blockdev-close-tray is a no-op (or may not be
2474          * called at all); therefore, the medium needs to be pushed into the
2475          * slot here.
2476          * Do it after blk_insert_bs() so blk_is_inserted(blk) returns the @load
2477          * value passed here (i.e. true). */
2478         blk_dev_change_media_cb(blk, true, &local_err);
2479         if (local_err) {
2480             error_propagate(errp, local_err);
2481             blk_remove_bs(blk);
2482             return;
2483         }
2484     }
2485 }
2486 
2487 void qmp_x_blockdev_insert_medium(bool has_device, const char *device,
2488                                   bool has_id, const char *id,
2489                                   const char *node_name, Error **errp)
2490 {
2491     BlockBackend *blk;
2492     BlockDriverState *bs;
2493 
2494     blk = qmp_get_blk(has_device ? device : NULL,
2495                       has_id ? id : NULL,
2496                       errp);
2497     if (!blk) {
2498         return;
2499     }
2500 
2501     bs = bdrv_find_node(node_name);
2502     if (!bs) {
2503         error_setg(errp, "Node '%s' not found", node_name);
2504         return;
2505     }
2506 
2507     if (bdrv_has_blk(bs)) {
2508         error_setg(errp, "Node '%s' is already in use", node_name);
2509         return;
2510     }
2511 
2512     qmp_blockdev_insert_anon_medium(blk, bs, errp);
2513 }
2514 
2515 void qmp_blockdev_change_medium(bool has_device, const char *device,
2516                                 bool has_id, const char *id,
2517                                 const char *filename,
2518                                 bool has_format, const char *format,
2519                                 bool has_read_only,
2520                                 BlockdevChangeReadOnlyMode read_only,
2521                                 Error **errp)
2522 {
2523     BlockBackend *blk;
2524     BlockDriverState *medium_bs = NULL;
2525     int bdrv_flags;
2526     bool detect_zeroes;
2527     int rc;
2528     QDict *options = NULL;
2529     Error *err = NULL;
2530 
2531     blk = qmp_get_blk(has_device ? device : NULL,
2532                       has_id ? id : NULL,
2533                       errp);
2534     if (!blk) {
2535         goto fail;
2536     }
2537 
2538     if (blk_bs(blk)) {
2539         blk_update_root_state(blk);
2540     }
2541 
2542     bdrv_flags = blk_get_open_flags_from_root_state(blk);
2543     bdrv_flags &= ~(BDRV_O_TEMPORARY | BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING |
2544         BDRV_O_PROTOCOL);
2545 
2546     if (!has_read_only) {
2547         read_only = BLOCKDEV_CHANGE_READ_ONLY_MODE_RETAIN;
2548     }
2549 
2550     switch (read_only) {
2551     case BLOCKDEV_CHANGE_READ_ONLY_MODE_RETAIN:
2552         break;
2553 
2554     case BLOCKDEV_CHANGE_READ_ONLY_MODE_READ_ONLY:
2555         bdrv_flags &= ~BDRV_O_RDWR;
2556         break;
2557 
2558     case BLOCKDEV_CHANGE_READ_ONLY_MODE_READ_WRITE:
2559         bdrv_flags |= BDRV_O_RDWR;
2560         break;
2561 
2562     default:
2563         abort();
2564     }
2565 
2566     options = qdict_new();
2567     detect_zeroes = blk_get_detect_zeroes_from_root_state(blk);
2568     qdict_put_str(options, "detect-zeroes", detect_zeroes ? "on" : "off");
2569 
2570     if (has_format) {
2571         qdict_put_str(options, "driver", format);
2572     }
2573 
2574     medium_bs = bdrv_open(filename, NULL, options, bdrv_flags, errp);
2575     if (!medium_bs) {
2576         goto fail;
2577     }
2578 
2579     bdrv_add_key(medium_bs, NULL, &err);
2580     if (err) {
2581         error_propagate(errp, err);
2582         goto fail;
2583     }
2584 
2585     rc = do_open_tray(has_device ? device : NULL,
2586                       has_id ? id : NULL,
2587                       false, &err);
2588     if (rc && rc != -ENOSYS) {
2589         error_propagate(errp, err);
2590         goto fail;
2591     }
2592     error_free(err);
2593     err = NULL;
2594 
2595     qmp_x_blockdev_remove_medium(has_device, device, has_id, id, &err);
2596     if (err) {
2597         error_propagate(errp, err);
2598         goto fail;
2599     }
2600 
2601     qmp_blockdev_insert_anon_medium(blk, medium_bs, &err);
2602     if (err) {
2603         error_propagate(errp, err);
2604         goto fail;
2605     }
2606 
2607     qmp_blockdev_close_tray(has_device, device, has_id, id, errp);
2608 
2609 fail:
2610     /* If the medium has been inserted, the device has its own reference, so
2611      * ours must be relinquished; and if it has not been inserted successfully,
2612      * the reference must be relinquished anyway */
2613     bdrv_unref(medium_bs);
2614 }
2615 
2616 /* throttling disk I/O limits */
2617 void qmp_block_set_io_throttle(BlockIOThrottle *arg, Error **errp)
2618 {
2619     ThrottleConfig cfg;
2620     BlockDriverState *bs;
2621     BlockBackend *blk;
2622     AioContext *aio_context;
2623 
2624     blk = qmp_get_blk(arg->has_device ? arg->device : NULL,
2625                       arg->has_id ? arg->id : NULL,
2626                       errp);
2627     if (!blk) {
2628         return;
2629     }
2630 
2631     aio_context = blk_get_aio_context(blk);
2632     aio_context_acquire(aio_context);
2633 
2634     bs = blk_bs(blk);
2635     if (!bs) {
2636         error_setg(errp, "Device has no medium");
2637         goto out;
2638     }
2639 
2640     throttle_config_init(&cfg);
2641     cfg.buckets[THROTTLE_BPS_TOTAL].avg = arg->bps;
2642     cfg.buckets[THROTTLE_BPS_READ].avg  = arg->bps_rd;
2643     cfg.buckets[THROTTLE_BPS_WRITE].avg = arg->bps_wr;
2644 
2645     cfg.buckets[THROTTLE_OPS_TOTAL].avg = arg->iops;
2646     cfg.buckets[THROTTLE_OPS_READ].avg  = arg->iops_rd;
2647     cfg.buckets[THROTTLE_OPS_WRITE].avg = arg->iops_wr;
2648 
2649     if (arg->has_bps_max) {
2650         cfg.buckets[THROTTLE_BPS_TOTAL].max = arg->bps_max;
2651     }
2652     if (arg->has_bps_rd_max) {
2653         cfg.buckets[THROTTLE_BPS_READ].max = arg->bps_rd_max;
2654     }
2655     if (arg->has_bps_wr_max) {
2656         cfg.buckets[THROTTLE_BPS_WRITE].max = arg->bps_wr_max;
2657     }
2658     if (arg->has_iops_max) {
2659         cfg.buckets[THROTTLE_OPS_TOTAL].max = arg->iops_max;
2660     }
2661     if (arg->has_iops_rd_max) {
2662         cfg.buckets[THROTTLE_OPS_READ].max = arg->iops_rd_max;
2663     }
2664     if (arg->has_iops_wr_max) {
2665         cfg.buckets[THROTTLE_OPS_WRITE].max = arg->iops_wr_max;
2666     }
2667 
2668     if (arg->has_bps_max_length) {
2669         cfg.buckets[THROTTLE_BPS_TOTAL].burst_length = arg->bps_max_length;
2670     }
2671     if (arg->has_bps_rd_max_length) {
2672         cfg.buckets[THROTTLE_BPS_READ].burst_length = arg->bps_rd_max_length;
2673     }
2674     if (arg->has_bps_wr_max_length) {
2675         cfg.buckets[THROTTLE_BPS_WRITE].burst_length = arg->bps_wr_max_length;
2676     }
2677     if (arg->has_iops_max_length) {
2678         cfg.buckets[THROTTLE_OPS_TOTAL].burst_length = arg->iops_max_length;
2679     }
2680     if (arg->has_iops_rd_max_length) {
2681         cfg.buckets[THROTTLE_OPS_READ].burst_length = arg->iops_rd_max_length;
2682     }
2683     if (arg->has_iops_wr_max_length) {
2684         cfg.buckets[THROTTLE_OPS_WRITE].burst_length = arg->iops_wr_max_length;
2685     }
2686 
2687     if (arg->has_iops_size) {
2688         cfg.op_size = arg->iops_size;
2689     }
2690 
2691     if (!throttle_is_valid(&cfg, errp)) {
2692         goto out;
2693     }
2694 
2695     if (throttle_enabled(&cfg)) {
2696         /* Enable I/O limits if they're not enabled yet, otherwise
2697          * just update the throttling group. */
2698         if (!blk_get_public(blk)->throttle_state) {
2699             blk_io_limits_enable(blk,
2700                                  arg->has_group ? arg->group :
2701                                  arg->has_device ? arg->device :
2702                                  arg->id);
2703         } else if (arg->has_group) {
2704             blk_io_limits_update_group(blk, arg->group);
2705         }
2706         /* Set the new throttling configuration */
2707         blk_set_io_limits(blk, &cfg);
2708     } else if (blk_get_public(blk)->throttle_state) {
2709         /* If all throttling settings are set to 0, disable I/O limits */
2710         blk_io_limits_disable(blk);
2711     }
2712 
2713 out:
2714     aio_context_release(aio_context);
2715 }
2716 
2717 void qmp_block_dirty_bitmap_add(const char *node, const char *name,
2718                                 bool has_granularity, uint32_t granularity,
2719                                 Error **errp)
2720 {
2721     BlockDriverState *bs;
2722 
2723     if (!name || name[0] == '\0') {
2724         error_setg(errp, "Bitmap name cannot be empty");
2725         return;
2726     }
2727 
2728     bs = bdrv_lookup_bs(node, node, errp);
2729     if (!bs) {
2730         return;
2731     }
2732 
2733     if (has_granularity) {
2734         if (granularity < 512 || !is_power_of_2(granularity)) {
2735             error_setg(errp, "Granularity must be power of 2 "
2736                              "and at least 512");
2737             return;
2738         }
2739     } else {
2740         /* Default to cluster size, if available: */
2741         granularity = bdrv_get_default_bitmap_granularity(bs);
2742     }
2743 
2744     bdrv_create_dirty_bitmap(bs, granularity, name, errp);
2745 }
2746 
2747 void qmp_block_dirty_bitmap_remove(const char *node, const char *name,
2748                                    Error **errp)
2749 {
2750     BlockDriverState *bs;
2751     BdrvDirtyBitmap *bitmap;
2752 
2753     bitmap = block_dirty_bitmap_lookup(node, name, &bs, errp);
2754     if (!bitmap || !bs) {
2755         return;
2756     }
2757 
2758     if (bdrv_dirty_bitmap_frozen(bitmap)) {
2759         error_setg(errp,
2760                    "Bitmap '%s' is currently frozen and cannot be removed",
2761                    name);
2762         return;
2763     }
2764     bdrv_dirty_bitmap_make_anon(bitmap);
2765     bdrv_release_dirty_bitmap(bs, bitmap);
2766 }
2767 
2768 /**
2769  * Completely clear a bitmap, for the purposes of synchronizing a bitmap
2770  * immediately after a full backup operation.
2771  */
2772 void qmp_block_dirty_bitmap_clear(const char *node, const char *name,
2773                                   Error **errp)
2774 {
2775     BdrvDirtyBitmap *bitmap;
2776     BlockDriverState *bs;
2777 
2778     bitmap = block_dirty_bitmap_lookup(node, name, &bs, errp);
2779     if (!bitmap || !bs) {
2780         return;
2781     }
2782 
2783     if (bdrv_dirty_bitmap_frozen(bitmap)) {
2784         error_setg(errp,
2785                    "Bitmap '%s' is currently frozen and cannot be modified",
2786                    name);
2787         return;
2788     } else if (!bdrv_dirty_bitmap_enabled(bitmap)) {
2789         error_setg(errp,
2790                    "Bitmap '%s' is currently disabled and cannot be cleared",
2791                    name);
2792         return;
2793     }
2794 
2795     bdrv_clear_dirty_bitmap(bitmap, NULL);
2796 }
2797 
2798 void hmp_drive_del(Monitor *mon, const QDict *qdict)
2799 {
2800     const char *id = qdict_get_str(qdict, "id");
2801     BlockBackend *blk;
2802     BlockDriverState *bs;
2803     AioContext *aio_context;
2804     Error *local_err = NULL;
2805 
2806     bs = bdrv_find_node(id);
2807     if (bs) {
2808         qmp_blockdev_del(id, &local_err);
2809         if (local_err) {
2810             error_report_err(local_err);
2811         }
2812         return;
2813     }
2814 
2815     blk = blk_by_name(id);
2816     if (!blk) {
2817         error_report("Device '%s' not found", id);
2818         return;
2819     }
2820 
2821     if (!blk_legacy_dinfo(blk)) {
2822         error_report("Deleting device added with blockdev-add"
2823                      " is not supported");
2824         return;
2825     }
2826 
2827     aio_context = blk_get_aio_context(blk);
2828     aio_context_acquire(aio_context);
2829 
2830     bs = blk_bs(blk);
2831     if (bs) {
2832         if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_DRIVE_DEL, &local_err)) {
2833             error_report_err(local_err);
2834             aio_context_release(aio_context);
2835             return;
2836         }
2837 
2838         blk_remove_bs(blk);
2839     }
2840 
2841     /* Make the BlockBackend and the attached BlockDriverState anonymous */
2842     monitor_remove_blk(blk);
2843 
2844     /* If this BlockBackend has a device attached to it, its refcount will be
2845      * decremented when the device is removed; otherwise we have to do so here.
2846      */
2847     if (blk_get_attached_dev(blk)) {
2848         /* Further I/O must not pause the guest */
2849         blk_set_on_error(blk, BLOCKDEV_ON_ERROR_REPORT,
2850                          BLOCKDEV_ON_ERROR_REPORT);
2851     } else {
2852         blk_unref(blk);
2853     }
2854 
2855     aio_context_release(aio_context);
2856 }
2857 
2858 void qmp_block_resize(bool has_device, const char *device,
2859                       bool has_node_name, const char *node_name,
2860                       int64_t size, Error **errp)
2861 {
2862     Error *local_err = NULL;
2863     BlockBackend *blk = NULL;
2864     BlockDriverState *bs;
2865     AioContext *aio_context;
2866     int ret;
2867 
2868     bs = bdrv_lookup_bs(has_device ? device : NULL,
2869                         has_node_name ? node_name : NULL,
2870                         &local_err);
2871     if (local_err) {
2872         error_propagate(errp, local_err);
2873         return;
2874     }
2875 
2876     aio_context = bdrv_get_aio_context(bs);
2877     aio_context_acquire(aio_context);
2878 
2879     if (!bdrv_is_first_non_filter(bs)) {
2880         error_setg(errp, QERR_FEATURE_DISABLED, "resize");
2881         goto out;
2882     }
2883 
2884     if (size < 0) {
2885         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "size", "a >0 size");
2886         goto out;
2887     }
2888 
2889     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_RESIZE, NULL)) {
2890         error_setg(errp, QERR_DEVICE_IN_USE, device);
2891         goto out;
2892     }
2893 
2894     blk = blk_new(BLK_PERM_RESIZE, BLK_PERM_ALL);
2895     ret = blk_insert_bs(blk, bs, errp);
2896     if (ret < 0) {
2897         goto out;
2898     }
2899 
2900     bdrv_drained_begin(bs);
2901     ret = blk_truncate(blk, size, errp);
2902     bdrv_drained_end(bs);
2903 
2904 out:
2905     blk_unref(blk);
2906     aio_context_release(aio_context);
2907 }
2908 
2909 void qmp_block_stream(bool has_job_id, const char *job_id, const char *device,
2910                       bool has_base, const char *base,
2911                       bool has_base_node, const char *base_node,
2912                       bool has_backing_file, const char *backing_file,
2913                       bool has_speed, int64_t speed,
2914                       bool has_on_error, BlockdevOnError on_error,
2915                       Error **errp)
2916 {
2917     BlockDriverState *bs, *iter;
2918     BlockDriverState *base_bs = NULL;
2919     AioContext *aio_context;
2920     Error *local_err = NULL;
2921     const char *base_name = NULL;
2922 
2923     if (!has_on_error) {
2924         on_error = BLOCKDEV_ON_ERROR_REPORT;
2925     }
2926 
2927     bs = bdrv_lookup_bs(device, device, errp);
2928     if (!bs) {
2929         return;
2930     }
2931 
2932     aio_context = bdrv_get_aio_context(bs);
2933     aio_context_acquire(aio_context);
2934 
2935     if (has_base && has_base_node) {
2936         error_setg(errp, "'base' and 'base-node' cannot be specified "
2937                    "at the same time");
2938         goto out;
2939     }
2940 
2941     if (has_base) {
2942         base_bs = bdrv_find_backing_image(bs, base);
2943         if (base_bs == NULL) {
2944             error_setg(errp, QERR_BASE_NOT_FOUND, base);
2945             goto out;
2946         }
2947         assert(bdrv_get_aio_context(base_bs) == aio_context);
2948         base_name = base;
2949     }
2950 
2951     if (has_base_node) {
2952         base_bs = bdrv_lookup_bs(NULL, base_node, errp);
2953         if (!base_bs) {
2954             goto out;
2955         }
2956         if (bs == base_bs || !bdrv_chain_contains(bs, base_bs)) {
2957             error_setg(errp, "Node '%s' is not a backing image of '%s'",
2958                        base_node, device);
2959             goto out;
2960         }
2961         assert(bdrv_get_aio_context(base_bs) == aio_context);
2962         base_name = base_bs->filename;
2963     }
2964 
2965     /* Check for op blockers in the whole chain between bs and base */
2966     for (iter = bs; iter && iter != base_bs; iter = backing_bs(iter)) {
2967         if (bdrv_op_is_blocked(iter, BLOCK_OP_TYPE_STREAM, errp)) {
2968             goto out;
2969         }
2970     }
2971 
2972     /* if we are streaming the entire chain, the result will have no backing
2973      * file, and specifying one is therefore an error */
2974     if (base_bs == NULL && has_backing_file) {
2975         error_setg(errp, "backing file specified, but streaming the "
2976                          "entire chain");
2977         goto out;
2978     }
2979 
2980     /* backing_file string overrides base bs filename */
2981     base_name = has_backing_file ? backing_file : base_name;
2982 
2983     stream_start(has_job_id ? job_id : NULL, bs, base_bs, base_name,
2984                  has_speed ? speed : 0, on_error, &local_err);
2985     if (local_err) {
2986         error_propagate(errp, local_err);
2987         goto out;
2988     }
2989 
2990     trace_qmp_block_stream(bs, bs->job);
2991 
2992 out:
2993     aio_context_release(aio_context);
2994 }
2995 
2996 void qmp_block_commit(bool has_job_id, const char *job_id, const char *device,
2997                       bool has_base, const char *base,
2998                       bool has_top, const char *top,
2999                       bool has_backing_file, const char *backing_file,
3000                       bool has_speed, int64_t speed,
3001                       bool has_filter_node_name, const char *filter_node_name,
3002                       Error **errp)
3003 {
3004     BlockDriverState *bs;
3005     BlockDriverState *iter;
3006     BlockDriverState *base_bs, *top_bs;
3007     AioContext *aio_context;
3008     Error *local_err = NULL;
3009     /* This will be part of the QMP command, if/when the
3010      * BlockdevOnError change for blkmirror makes it in
3011      */
3012     BlockdevOnError on_error = BLOCKDEV_ON_ERROR_REPORT;
3013 
3014     if (!has_speed) {
3015         speed = 0;
3016     }
3017     if (!has_filter_node_name) {
3018         filter_node_name = NULL;
3019     }
3020 
3021     /* Important Note:
3022      *  libvirt relies on the DeviceNotFound error class in order to probe for
3023      *  live commit feature versions; for this to work, we must make sure to
3024      *  perform the device lookup before any generic errors that may occur in a
3025      *  scenario in which all optional arguments are omitted. */
3026     bs = qmp_get_root_bs(device, &local_err);
3027     if (!bs) {
3028         bs = bdrv_lookup_bs(device, device, NULL);
3029         if (!bs) {
3030             error_free(local_err);
3031             error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
3032                       "Device '%s' not found", device);
3033         } else {
3034             error_propagate(errp, local_err);
3035         }
3036         return;
3037     }
3038 
3039     aio_context = bdrv_get_aio_context(bs);
3040     aio_context_acquire(aio_context);
3041 
3042     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_COMMIT_SOURCE, errp)) {
3043         goto out;
3044     }
3045 
3046     /* default top_bs is the active layer */
3047     top_bs = bs;
3048 
3049     if (has_top && top) {
3050         if (strcmp(bs->filename, top) != 0) {
3051             top_bs = bdrv_find_backing_image(bs, top);
3052         }
3053     }
3054 
3055     if (top_bs == NULL) {
3056         error_setg(errp, "Top image file %s not found", top ? top : "NULL");
3057         goto out;
3058     }
3059 
3060     assert(bdrv_get_aio_context(top_bs) == aio_context);
3061 
3062     if (has_base && base) {
3063         base_bs = bdrv_find_backing_image(top_bs, base);
3064     } else {
3065         base_bs = bdrv_find_base(top_bs);
3066     }
3067 
3068     if (base_bs == NULL) {
3069         error_setg(errp, QERR_BASE_NOT_FOUND, base ? base : "NULL");
3070         goto out;
3071     }
3072 
3073     assert(bdrv_get_aio_context(base_bs) == aio_context);
3074 
3075     for (iter = top_bs; iter != backing_bs(base_bs); iter = backing_bs(iter)) {
3076         if (bdrv_op_is_blocked(iter, BLOCK_OP_TYPE_COMMIT_TARGET, errp)) {
3077             goto out;
3078         }
3079     }
3080 
3081     /* Do not allow attempts to commit an image into itself */
3082     if (top_bs == base_bs) {
3083         error_setg(errp, "cannot commit an image into itself");
3084         goto out;
3085     }
3086 
3087     if (top_bs == bs) {
3088         if (has_backing_file) {
3089             error_setg(errp, "'backing-file' specified,"
3090                              " but 'top' is the active layer");
3091             goto out;
3092         }
3093         commit_active_start(has_job_id ? job_id : NULL, bs, base_bs,
3094                             BLOCK_JOB_DEFAULT, speed, on_error,
3095                             filter_node_name, NULL, NULL, false, &local_err);
3096     } else {
3097         BlockDriverState *overlay_bs = bdrv_find_overlay(bs, top_bs);
3098         if (bdrv_op_is_blocked(overlay_bs, BLOCK_OP_TYPE_COMMIT_TARGET, errp)) {
3099             goto out;
3100         }
3101         commit_start(has_job_id ? job_id : NULL, bs, base_bs, top_bs, speed,
3102                      on_error, has_backing_file ? backing_file : NULL,
3103                      filter_node_name, &local_err);
3104     }
3105     if (local_err != NULL) {
3106         error_propagate(errp, local_err);
3107         goto out;
3108     }
3109 
3110 out:
3111     aio_context_release(aio_context);
3112 }
3113 
3114 static BlockJob *do_drive_backup(DriveBackup *backup, BlockJobTxn *txn,
3115                                  Error **errp)
3116 {
3117     BlockDriverState *bs;
3118     BlockDriverState *target_bs;
3119     BlockDriverState *source = NULL;
3120     BlockJob *job = NULL;
3121     BdrvDirtyBitmap *bmap = NULL;
3122     AioContext *aio_context;
3123     QDict *options = NULL;
3124     Error *local_err = NULL;
3125     int flags;
3126     int64_t size;
3127     bool set_backing_hd = false;
3128 
3129     if (!backup->has_speed) {
3130         backup->speed = 0;
3131     }
3132     if (!backup->has_on_source_error) {
3133         backup->on_source_error = BLOCKDEV_ON_ERROR_REPORT;
3134     }
3135     if (!backup->has_on_target_error) {
3136         backup->on_target_error = BLOCKDEV_ON_ERROR_REPORT;
3137     }
3138     if (!backup->has_mode) {
3139         backup->mode = NEW_IMAGE_MODE_ABSOLUTE_PATHS;
3140     }
3141     if (!backup->has_job_id) {
3142         backup->job_id = NULL;
3143     }
3144     if (!backup->has_compress) {
3145         backup->compress = false;
3146     }
3147 
3148     bs = qmp_get_root_bs(backup->device, errp);
3149     if (!bs) {
3150         return NULL;
3151     }
3152 
3153     aio_context = bdrv_get_aio_context(bs);
3154     aio_context_acquire(aio_context);
3155 
3156     if (!backup->has_format) {
3157         backup->format = backup->mode == NEW_IMAGE_MODE_EXISTING ?
3158                          NULL : (char*) bs->drv->format_name;
3159     }
3160 
3161     /* Early check to avoid creating target */
3162     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_BACKUP_SOURCE, errp)) {
3163         goto out;
3164     }
3165 
3166     flags = bs->open_flags | BDRV_O_RDWR;
3167 
3168     /* See if we have a backing HD we can use to create our new image
3169      * on top of. */
3170     if (backup->sync == MIRROR_SYNC_MODE_TOP) {
3171         source = backing_bs(bs);
3172         if (!source) {
3173             backup->sync = MIRROR_SYNC_MODE_FULL;
3174         }
3175     }
3176     if (backup->sync == MIRROR_SYNC_MODE_NONE) {
3177         source = bs;
3178         flags |= BDRV_O_NO_BACKING;
3179         set_backing_hd = true;
3180     }
3181 
3182     size = bdrv_getlength(bs);
3183     if (size < 0) {
3184         error_setg_errno(errp, -size, "bdrv_getlength failed");
3185         goto out;
3186     }
3187 
3188     if (backup->mode != NEW_IMAGE_MODE_EXISTING) {
3189         assert(backup->format);
3190         if (source) {
3191             bdrv_img_create(backup->target, backup->format, source->filename,
3192                             source->drv->format_name, NULL,
3193                             size, flags, false, &local_err);
3194         } else {
3195             bdrv_img_create(backup->target, backup->format, NULL, NULL, NULL,
3196                             size, flags, false, &local_err);
3197         }
3198     }
3199 
3200     if (local_err) {
3201         error_propagate(errp, local_err);
3202         goto out;
3203     }
3204 
3205     if (backup->format) {
3206         if (!options) {
3207             options = qdict_new();
3208         }
3209         qdict_put_str(options, "driver", backup->format);
3210     }
3211 
3212     target_bs = bdrv_open(backup->target, NULL, options, flags, errp);
3213     if (!target_bs) {
3214         goto out;
3215     }
3216 
3217     bdrv_set_aio_context(target_bs, aio_context);
3218 
3219     if (set_backing_hd) {
3220         bdrv_set_backing_hd(target_bs, source, &local_err);
3221         if (local_err) {
3222             bdrv_unref(target_bs);
3223             goto out;
3224         }
3225     }
3226 
3227     if (backup->has_bitmap) {
3228         bmap = bdrv_find_dirty_bitmap(bs, backup->bitmap);
3229         if (!bmap) {
3230             error_setg(errp, "Bitmap '%s' could not be found", backup->bitmap);
3231             bdrv_unref(target_bs);
3232             goto out;
3233         }
3234     }
3235 
3236     job = backup_job_create(backup->job_id, bs, target_bs, backup->speed,
3237                             backup->sync, bmap, backup->compress,
3238                             backup->on_source_error, backup->on_target_error,
3239                             BLOCK_JOB_DEFAULT, NULL, NULL, txn, &local_err);
3240     bdrv_unref(target_bs);
3241     if (local_err != NULL) {
3242         error_propagate(errp, local_err);
3243         goto out;
3244     }
3245 
3246 out:
3247     aio_context_release(aio_context);
3248     return job;
3249 }
3250 
3251 void qmp_drive_backup(DriveBackup *arg, Error **errp)
3252 {
3253 
3254     BlockJob *job;
3255     job = do_drive_backup(arg, NULL, errp);
3256     if (job) {
3257         block_job_start(job);
3258     }
3259 }
3260 
3261 BlockDeviceInfoList *qmp_query_named_block_nodes(Error **errp)
3262 {
3263     return bdrv_named_nodes_list(errp);
3264 }
3265 
3266 BlockJob *do_blockdev_backup(BlockdevBackup *backup, BlockJobTxn *txn,
3267                              Error **errp)
3268 {
3269     BlockDriverState *bs;
3270     BlockDriverState *target_bs;
3271     Error *local_err = NULL;
3272     AioContext *aio_context;
3273     BlockJob *job = NULL;
3274 
3275     if (!backup->has_speed) {
3276         backup->speed = 0;
3277     }
3278     if (!backup->has_on_source_error) {
3279         backup->on_source_error = BLOCKDEV_ON_ERROR_REPORT;
3280     }
3281     if (!backup->has_on_target_error) {
3282         backup->on_target_error = BLOCKDEV_ON_ERROR_REPORT;
3283     }
3284     if (!backup->has_job_id) {
3285         backup->job_id = NULL;
3286     }
3287     if (!backup->has_compress) {
3288         backup->compress = false;
3289     }
3290 
3291     bs = qmp_get_root_bs(backup->device, errp);
3292     if (!bs) {
3293         return NULL;
3294     }
3295 
3296     aio_context = bdrv_get_aio_context(bs);
3297     aio_context_acquire(aio_context);
3298 
3299     target_bs = bdrv_lookup_bs(backup->target, backup->target, errp);
3300     if (!target_bs) {
3301         goto out;
3302     }
3303 
3304     if (bdrv_get_aio_context(target_bs) != aio_context) {
3305         if (!bdrv_has_blk(target_bs)) {
3306             /* The target BDS is not attached, we can safely move it to another
3307              * AioContext. */
3308             bdrv_set_aio_context(target_bs, aio_context);
3309         } else {
3310             error_setg(errp, "Target is attached to a different thread from "
3311                              "source.");
3312             goto out;
3313         }
3314     }
3315     job = backup_job_create(backup->job_id, bs, target_bs, backup->speed,
3316                             backup->sync, NULL, backup->compress,
3317                             backup->on_source_error, backup->on_target_error,
3318                             BLOCK_JOB_DEFAULT, NULL, NULL, txn, &local_err);
3319     if (local_err != NULL) {
3320         error_propagate(errp, local_err);
3321     }
3322 out:
3323     aio_context_release(aio_context);
3324     return job;
3325 }
3326 
3327 void qmp_blockdev_backup(BlockdevBackup *arg, Error **errp)
3328 {
3329     BlockJob *job;
3330     job = do_blockdev_backup(arg, NULL, errp);
3331     if (job) {
3332         block_job_start(job);
3333     }
3334 }
3335 
3336 /* Parameter check and block job starting for drive mirroring.
3337  * Caller should hold @device and @target's aio context (must be the same).
3338  **/
3339 static void blockdev_mirror_common(const char *job_id, BlockDriverState *bs,
3340                                    BlockDriverState *target,
3341                                    bool has_replaces, const char *replaces,
3342                                    enum MirrorSyncMode sync,
3343                                    BlockMirrorBackingMode backing_mode,
3344                                    bool has_speed, int64_t speed,
3345                                    bool has_granularity, uint32_t granularity,
3346                                    bool has_buf_size, int64_t buf_size,
3347                                    bool has_on_source_error,
3348                                    BlockdevOnError on_source_error,
3349                                    bool has_on_target_error,
3350                                    BlockdevOnError on_target_error,
3351                                    bool has_unmap, bool unmap,
3352                                    bool has_filter_node_name,
3353                                    const char *filter_node_name,
3354                                    Error **errp)
3355 {
3356 
3357     if (!has_speed) {
3358         speed = 0;
3359     }
3360     if (!has_on_source_error) {
3361         on_source_error = BLOCKDEV_ON_ERROR_REPORT;
3362     }
3363     if (!has_on_target_error) {
3364         on_target_error = BLOCKDEV_ON_ERROR_REPORT;
3365     }
3366     if (!has_granularity) {
3367         granularity = 0;
3368     }
3369     if (!has_buf_size) {
3370         buf_size = 0;
3371     }
3372     if (!has_unmap) {
3373         unmap = true;
3374     }
3375     if (!has_filter_node_name) {
3376         filter_node_name = NULL;
3377     }
3378 
3379     if (granularity != 0 && (granularity < 512 || granularity > 1048576 * 64)) {
3380         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "granularity",
3381                    "a value in range [512B, 64MB]");
3382         return;
3383     }
3384     if (granularity & (granularity - 1)) {
3385         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "granularity",
3386                    "power of 2");
3387         return;
3388     }
3389 
3390     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_MIRROR_SOURCE, errp)) {
3391         return;
3392     }
3393     if (bdrv_op_is_blocked(target, BLOCK_OP_TYPE_MIRROR_TARGET, errp)) {
3394         return;
3395     }
3396 
3397     if (!bs->backing && sync == MIRROR_SYNC_MODE_TOP) {
3398         sync = MIRROR_SYNC_MODE_FULL;
3399     }
3400 
3401     /* pass the node name to replace to mirror start since it's loose coupling
3402      * and will allow to check whether the node still exist at mirror completion
3403      */
3404     mirror_start(job_id, bs, target,
3405                  has_replaces ? replaces : NULL,
3406                  speed, granularity, buf_size, sync, backing_mode,
3407                  on_source_error, on_target_error, unmap, filter_node_name,
3408                  errp);
3409 }
3410 
3411 void qmp_drive_mirror(DriveMirror *arg, Error **errp)
3412 {
3413     BlockDriverState *bs;
3414     BlockDriverState *source, *target_bs;
3415     AioContext *aio_context;
3416     BlockMirrorBackingMode backing_mode;
3417     Error *local_err = NULL;
3418     QDict *options = NULL;
3419     int flags;
3420     int64_t size;
3421     const char *format = arg->format;
3422 
3423     bs = qmp_get_root_bs(arg->device, errp);
3424     if (!bs) {
3425         return;
3426     }
3427 
3428     aio_context = bdrv_get_aio_context(bs);
3429     aio_context_acquire(aio_context);
3430 
3431     if (!arg->has_mode) {
3432         arg->mode = NEW_IMAGE_MODE_ABSOLUTE_PATHS;
3433     }
3434 
3435     if (!arg->has_format) {
3436         format = (arg->mode == NEW_IMAGE_MODE_EXISTING
3437                   ? NULL : bs->drv->format_name);
3438     }
3439 
3440     flags = bs->open_flags | BDRV_O_RDWR;
3441     source = backing_bs(bs);
3442     if (!source && arg->sync == MIRROR_SYNC_MODE_TOP) {
3443         arg->sync = MIRROR_SYNC_MODE_FULL;
3444     }
3445     if (arg->sync == MIRROR_SYNC_MODE_NONE) {
3446         source = bs;
3447     }
3448 
3449     size = bdrv_getlength(bs);
3450     if (size < 0) {
3451         error_setg_errno(errp, -size, "bdrv_getlength failed");
3452         goto out;
3453     }
3454 
3455     if (arg->has_replaces) {
3456         BlockDriverState *to_replace_bs;
3457         AioContext *replace_aio_context;
3458         int64_t replace_size;
3459 
3460         if (!arg->has_node_name) {
3461             error_setg(errp, "a node-name must be provided when replacing a"
3462                              " named node of the graph");
3463             goto out;
3464         }
3465 
3466         to_replace_bs = check_to_replace_node(bs, arg->replaces, &local_err);
3467 
3468         if (!to_replace_bs) {
3469             error_propagate(errp, local_err);
3470             goto out;
3471         }
3472 
3473         replace_aio_context = bdrv_get_aio_context(to_replace_bs);
3474         aio_context_acquire(replace_aio_context);
3475         replace_size = bdrv_getlength(to_replace_bs);
3476         aio_context_release(replace_aio_context);
3477 
3478         if (size != replace_size) {
3479             error_setg(errp, "cannot replace image with a mirror image of "
3480                              "different size");
3481             goto out;
3482         }
3483     }
3484 
3485     if (arg->mode == NEW_IMAGE_MODE_ABSOLUTE_PATHS) {
3486         backing_mode = MIRROR_SOURCE_BACKING_CHAIN;
3487     } else {
3488         backing_mode = MIRROR_OPEN_BACKING_CHAIN;
3489     }
3490 
3491     if ((arg->sync == MIRROR_SYNC_MODE_FULL || !source)
3492         && arg->mode != NEW_IMAGE_MODE_EXISTING)
3493     {
3494         /* create new image w/o backing file */
3495         assert(format);
3496         bdrv_img_create(arg->target, format,
3497                         NULL, NULL, NULL, size, flags, false, &local_err);
3498     } else {
3499         switch (arg->mode) {
3500         case NEW_IMAGE_MODE_EXISTING:
3501             break;
3502         case NEW_IMAGE_MODE_ABSOLUTE_PATHS:
3503             /* create new image with backing file */
3504             bdrv_img_create(arg->target, format,
3505                             source->filename,
3506                             source->drv->format_name,
3507                             NULL, size, flags, false, &local_err);
3508             break;
3509         default:
3510             abort();
3511         }
3512     }
3513 
3514     if (local_err) {
3515         error_propagate(errp, local_err);
3516         goto out;
3517     }
3518 
3519     options = qdict_new();
3520     if (arg->has_node_name) {
3521         qdict_put_str(options, "node-name", arg->node_name);
3522     }
3523     if (format) {
3524         qdict_put_str(options, "driver", format);
3525     }
3526 
3527     /* Mirroring takes care of copy-on-write using the source's backing
3528      * file.
3529      */
3530     target_bs = bdrv_open(arg->target, NULL, options,
3531                           flags | BDRV_O_NO_BACKING, errp);
3532     if (!target_bs) {
3533         goto out;
3534     }
3535 
3536     bdrv_set_aio_context(target_bs, aio_context);
3537 
3538     blockdev_mirror_common(arg->has_job_id ? arg->job_id : NULL, bs, target_bs,
3539                            arg->has_replaces, arg->replaces, arg->sync,
3540                            backing_mode, arg->has_speed, arg->speed,
3541                            arg->has_granularity, arg->granularity,
3542                            arg->has_buf_size, arg->buf_size,
3543                            arg->has_on_source_error, arg->on_source_error,
3544                            arg->has_on_target_error, arg->on_target_error,
3545                            arg->has_unmap, arg->unmap,
3546                            false, NULL,
3547                            &local_err);
3548     bdrv_unref(target_bs);
3549     error_propagate(errp, local_err);
3550 out:
3551     aio_context_release(aio_context);
3552 }
3553 
3554 void qmp_blockdev_mirror(bool has_job_id, const char *job_id,
3555                          const char *device, const char *target,
3556                          bool has_replaces, const char *replaces,
3557                          MirrorSyncMode sync,
3558                          bool has_speed, int64_t speed,
3559                          bool has_granularity, uint32_t granularity,
3560                          bool has_buf_size, int64_t buf_size,
3561                          bool has_on_source_error,
3562                          BlockdevOnError on_source_error,
3563                          bool has_on_target_error,
3564                          BlockdevOnError on_target_error,
3565                          bool has_filter_node_name,
3566                          const char *filter_node_name,
3567                          Error **errp)
3568 {
3569     BlockDriverState *bs;
3570     BlockDriverState *target_bs;
3571     AioContext *aio_context;
3572     BlockMirrorBackingMode backing_mode = MIRROR_LEAVE_BACKING_CHAIN;
3573     Error *local_err = NULL;
3574 
3575     bs = qmp_get_root_bs(device, errp);
3576     if (!bs) {
3577         return;
3578     }
3579 
3580     target_bs = bdrv_lookup_bs(target, target, errp);
3581     if (!target_bs) {
3582         return;
3583     }
3584 
3585     aio_context = bdrv_get_aio_context(bs);
3586     aio_context_acquire(aio_context);
3587 
3588     bdrv_set_aio_context(target_bs, aio_context);
3589 
3590     blockdev_mirror_common(has_job_id ? job_id : NULL, bs, target_bs,
3591                            has_replaces, replaces, sync, backing_mode,
3592                            has_speed, speed,
3593                            has_granularity, granularity,
3594                            has_buf_size, buf_size,
3595                            has_on_source_error, on_source_error,
3596                            has_on_target_error, on_target_error,
3597                            true, true,
3598                            has_filter_node_name, filter_node_name,
3599                            &local_err);
3600     error_propagate(errp, local_err);
3601 
3602     aio_context_release(aio_context);
3603 }
3604 
3605 /* Get a block job using its ID and acquire its AioContext */
3606 static BlockJob *find_block_job(const char *id, AioContext **aio_context,
3607                                 Error **errp)
3608 {
3609     BlockJob *job;
3610 
3611     assert(id != NULL);
3612 
3613     *aio_context = NULL;
3614 
3615     job = block_job_get(id);
3616 
3617     if (!job) {
3618         error_set(errp, ERROR_CLASS_DEVICE_NOT_ACTIVE,
3619                   "Block job '%s' not found", id);
3620         return NULL;
3621     }
3622 
3623     *aio_context = blk_get_aio_context(job->blk);
3624     aio_context_acquire(*aio_context);
3625 
3626     return job;
3627 }
3628 
3629 void qmp_block_job_set_speed(const char *device, int64_t speed, Error **errp)
3630 {
3631     AioContext *aio_context;
3632     BlockJob *job = find_block_job(device, &aio_context, errp);
3633 
3634     if (!job) {
3635         return;
3636     }
3637 
3638     block_job_set_speed(job, speed, errp);
3639     aio_context_release(aio_context);
3640 }
3641 
3642 void qmp_block_job_cancel(const char *device,
3643                           bool has_force, bool force, Error **errp)
3644 {
3645     AioContext *aio_context;
3646     BlockJob *job = find_block_job(device, &aio_context, errp);
3647 
3648     if (!job) {
3649         return;
3650     }
3651 
3652     if (!has_force) {
3653         force = false;
3654     }
3655 
3656     if (block_job_user_paused(job) && !force) {
3657         error_setg(errp, "The block job for device '%s' is currently paused",
3658                    device);
3659         goto out;
3660     }
3661 
3662     trace_qmp_block_job_cancel(job);
3663     block_job_cancel(job);
3664 out:
3665     aio_context_release(aio_context);
3666 }
3667 
3668 void qmp_block_job_pause(const char *device, Error **errp)
3669 {
3670     AioContext *aio_context;
3671     BlockJob *job = find_block_job(device, &aio_context, errp);
3672 
3673     if (!job || block_job_user_paused(job)) {
3674         return;
3675     }
3676 
3677     trace_qmp_block_job_pause(job);
3678     block_job_user_pause(job);
3679     aio_context_release(aio_context);
3680 }
3681 
3682 void qmp_block_job_resume(const char *device, Error **errp)
3683 {
3684     AioContext *aio_context;
3685     BlockJob *job = find_block_job(device, &aio_context, errp);
3686 
3687     if (!job || !block_job_user_paused(job)) {
3688         return;
3689     }
3690 
3691     trace_qmp_block_job_resume(job);
3692     block_job_user_resume(job);
3693     aio_context_release(aio_context);
3694 }
3695 
3696 void qmp_block_job_complete(const char *device, Error **errp)
3697 {
3698     AioContext *aio_context;
3699     BlockJob *job = find_block_job(device, &aio_context, errp);
3700 
3701     if (!job) {
3702         return;
3703     }
3704 
3705     trace_qmp_block_job_complete(job);
3706     block_job_complete(job, errp);
3707     aio_context_release(aio_context);
3708 }
3709 
3710 void qmp_change_backing_file(const char *device,
3711                              const char *image_node_name,
3712                              const char *backing_file,
3713                              Error **errp)
3714 {
3715     BlockDriverState *bs = NULL;
3716     AioContext *aio_context;
3717     BlockDriverState *image_bs = NULL;
3718     Error *local_err = NULL;
3719     bool ro;
3720     int open_flags;
3721     int ret;
3722 
3723     bs = qmp_get_root_bs(device, errp);
3724     if (!bs) {
3725         return;
3726     }
3727 
3728     aio_context = bdrv_get_aio_context(bs);
3729     aio_context_acquire(aio_context);
3730 
3731     image_bs = bdrv_lookup_bs(NULL, image_node_name, &local_err);
3732     if (local_err) {
3733         error_propagate(errp, local_err);
3734         goto out;
3735     }
3736 
3737     if (!image_bs) {
3738         error_setg(errp, "image file not found");
3739         goto out;
3740     }
3741 
3742     if (bdrv_find_base(image_bs) == image_bs) {
3743         error_setg(errp, "not allowing backing file change on an image "
3744                          "without a backing file");
3745         goto out;
3746     }
3747 
3748     /* even though we are not necessarily operating on bs, we need it to
3749      * determine if block ops are currently prohibited on the chain */
3750     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_CHANGE, errp)) {
3751         goto out;
3752     }
3753 
3754     /* final sanity check */
3755     if (!bdrv_chain_contains(bs, image_bs)) {
3756         error_setg(errp, "'%s' and image file are not in the same chain",
3757                    device);
3758         goto out;
3759     }
3760 
3761     /* if not r/w, reopen to make r/w */
3762     open_flags = image_bs->open_flags;
3763     ro = bdrv_is_read_only(image_bs);
3764 
3765     if (ro) {
3766         bdrv_reopen(image_bs, open_flags | BDRV_O_RDWR, &local_err);
3767         if (local_err) {
3768             error_propagate(errp, local_err);
3769             goto out;
3770         }
3771     }
3772 
3773     ret = bdrv_change_backing_file(image_bs, backing_file,
3774                                image_bs->drv ? image_bs->drv->format_name : "");
3775 
3776     if (ret < 0) {
3777         error_setg_errno(errp, -ret, "Could not change backing file to '%s'",
3778                          backing_file);
3779         /* don't exit here, so we can try to restore open flags if
3780          * appropriate */
3781     }
3782 
3783     if (ro) {
3784         bdrv_reopen(image_bs, open_flags, &local_err);
3785         error_propagate(errp, local_err);
3786     }
3787 
3788 out:
3789     aio_context_release(aio_context);
3790 }
3791 
3792 void hmp_drive_add_node(Monitor *mon, const char *optstr)
3793 {
3794     QemuOpts *opts;
3795     QDict *qdict;
3796     Error *local_err = NULL;
3797 
3798     opts = qemu_opts_parse_noisily(&qemu_drive_opts, optstr, false);
3799     if (!opts) {
3800         return;
3801     }
3802 
3803     qdict = qemu_opts_to_qdict(opts, NULL);
3804 
3805     if (!qdict_get_try_str(qdict, "node-name")) {
3806         QDECREF(qdict);
3807         error_report("'node-name' needs to be specified");
3808         goto out;
3809     }
3810 
3811     BlockDriverState *bs = bds_tree_init(qdict, &local_err);
3812     if (!bs) {
3813         error_report_err(local_err);
3814         goto out;
3815     }
3816 
3817     QTAILQ_INSERT_TAIL(&monitor_bdrv_states, bs, monitor_list);
3818 
3819 out:
3820     qemu_opts_del(opts);
3821 }
3822 
3823 void qmp_blockdev_add(BlockdevOptions *options, Error **errp)
3824 {
3825     BlockDriverState *bs;
3826     QObject *obj;
3827     Visitor *v = qobject_output_visitor_new(&obj);
3828     QDict *qdict;
3829     Error *local_err = NULL;
3830 
3831     visit_type_BlockdevOptions(v, NULL, &options, &local_err);
3832     if (local_err) {
3833         error_propagate(errp, local_err);
3834         goto fail;
3835     }
3836 
3837     visit_complete(v, &obj);
3838     qdict = qobject_to_qdict(obj);
3839 
3840     qdict_flatten(qdict);
3841 
3842     if (!qdict_get_try_str(qdict, "node-name")) {
3843         error_setg(errp, "'node-name' must be specified for the root node");
3844         goto fail;
3845     }
3846 
3847     bs = bds_tree_init(qdict, errp);
3848     if (!bs) {
3849         goto fail;
3850     }
3851 
3852     QTAILQ_INSERT_TAIL(&monitor_bdrv_states, bs, monitor_list);
3853 
3854     if (bs && bdrv_key_required(bs)) {
3855         QTAILQ_REMOVE(&monitor_bdrv_states, bs, monitor_list);
3856         bdrv_unref(bs);
3857         error_setg(errp, "blockdev-add doesn't support encrypted devices");
3858         goto fail;
3859     }
3860 
3861 fail:
3862     visit_free(v);
3863 }
3864 
3865 void qmp_blockdev_del(const char *node_name, Error **errp)
3866 {
3867     AioContext *aio_context;
3868     BlockDriverState *bs;
3869 
3870     bs = bdrv_find_node(node_name);
3871     if (!bs) {
3872         error_setg(errp, "Cannot find node %s", node_name);
3873         return;
3874     }
3875     if (bdrv_has_blk(bs)) {
3876         error_setg(errp, "Node %s is in use", node_name);
3877         return;
3878     }
3879     aio_context = bdrv_get_aio_context(bs);
3880     aio_context_acquire(aio_context);
3881 
3882     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_DRIVE_DEL, errp)) {
3883         goto out;
3884     }
3885 
3886     if (!bs->monitor_list.tqe_prev) {
3887         error_setg(errp, "Node %s is not owned by the monitor",
3888                    bs->node_name);
3889         goto out;
3890     }
3891 
3892     if (bs->refcnt > 1) {
3893         error_setg(errp, "Block device %s is in use",
3894                    bdrv_get_device_or_node_name(bs));
3895         goto out;
3896     }
3897 
3898     QTAILQ_REMOVE(&monitor_bdrv_states, bs, monitor_list);
3899     bdrv_unref(bs);
3900 
3901 out:
3902     aio_context_release(aio_context);
3903 }
3904 
3905 static BdrvChild *bdrv_find_child(BlockDriverState *parent_bs,
3906                                   const char *child_name)
3907 {
3908     BdrvChild *child;
3909 
3910     QLIST_FOREACH(child, &parent_bs->children, next) {
3911         if (strcmp(child->name, child_name) == 0) {
3912             return child;
3913         }
3914     }
3915 
3916     return NULL;
3917 }
3918 
3919 void qmp_x_blockdev_change(const char *parent, bool has_child,
3920                            const char *child, bool has_node,
3921                            const char *node, Error **errp)
3922 {
3923     BlockDriverState *parent_bs, *new_bs = NULL;
3924     BdrvChild *p_child;
3925 
3926     parent_bs = bdrv_lookup_bs(parent, parent, errp);
3927     if (!parent_bs) {
3928         return;
3929     }
3930 
3931     if (has_child == has_node) {
3932         if (has_child) {
3933             error_setg(errp, "The parameters child and node are in conflict");
3934         } else {
3935             error_setg(errp, "Either child or node must be specified");
3936         }
3937         return;
3938     }
3939 
3940     if (has_child) {
3941         p_child = bdrv_find_child(parent_bs, child);
3942         if (!p_child) {
3943             error_setg(errp, "Node '%s' does not have child '%s'",
3944                        parent, child);
3945             return;
3946         }
3947         bdrv_del_child(parent_bs, p_child, errp);
3948     }
3949 
3950     if (has_node) {
3951         new_bs = bdrv_find_node(node);
3952         if (!new_bs) {
3953             error_setg(errp, "Node '%s' not found", node);
3954             return;
3955         }
3956         bdrv_add_child(parent_bs, new_bs, errp);
3957     }
3958 }
3959 
3960 BlockJobInfoList *qmp_query_block_jobs(Error **errp)
3961 {
3962     BlockJobInfoList *head = NULL, **p_next = &head;
3963     BlockJob *job;
3964 
3965     for (job = block_job_next(NULL); job; job = block_job_next(job)) {
3966         BlockJobInfoList *elem;
3967         AioContext *aio_context;
3968 
3969         if (block_job_is_internal(job)) {
3970             continue;
3971         }
3972         elem = g_new0(BlockJobInfoList, 1);
3973         aio_context = blk_get_aio_context(job->blk);
3974         aio_context_acquire(aio_context);
3975         elem->value = block_job_query(job, errp);
3976         aio_context_release(aio_context);
3977         if (!elem->value) {
3978             g_free(elem);
3979             qapi_free_BlockJobInfoList(head);
3980             return NULL;
3981         }
3982         *p_next = elem;
3983         p_next = &elem->next;
3984     }
3985 
3986     return head;
3987 }
3988 
3989 QemuOptsList qemu_common_drive_opts = {
3990     .name = "drive",
3991     .head = QTAILQ_HEAD_INITIALIZER(qemu_common_drive_opts.head),
3992     .desc = {
3993         {
3994             .name = "snapshot",
3995             .type = QEMU_OPT_BOOL,
3996             .help = "enable/disable snapshot mode",
3997         },{
3998             .name = "aio",
3999             .type = QEMU_OPT_STRING,
4000             .help = "host AIO implementation (threads, native)",
4001         },{
4002             .name = BDRV_OPT_CACHE_WB,
4003             .type = QEMU_OPT_BOOL,
4004             .help = "Enable writeback mode",
4005         },{
4006             .name = "format",
4007             .type = QEMU_OPT_STRING,
4008             .help = "disk format (raw, qcow2, ...)",
4009         },{
4010             .name = "rerror",
4011             .type = QEMU_OPT_STRING,
4012             .help = "read error action",
4013         },{
4014             .name = "werror",
4015             .type = QEMU_OPT_STRING,
4016             .help = "write error action",
4017         },{
4018             .name = BDRV_OPT_READ_ONLY,
4019             .type = QEMU_OPT_BOOL,
4020             .help = "open drive file as read-only",
4021         },
4022 
4023         THROTTLE_OPTS,
4024 
4025         {
4026             .name = "throttling.group",
4027             .type = QEMU_OPT_STRING,
4028             .help = "name of the block throttling group",
4029         },{
4030             .name = "copy-on-read",
4031             .type = QEMU_OPT_BOOL,
4032             .help = "copy read data from backing file into image file",
4033         },{
4034             .name = "detect-zeroes",
4035             .type = QEMU_OPT_STRING,
4036             .help = "try to optimize zero writes (off, on, unmap)",
4037         },{
4038             .name = "stats-account-invalid",
4039             .type = QEMU_OPT_BOOL,
4040             .help = "whether to account for invalid I/O operations "
4041                     "in the statistics",
4042         },{
4043             .name = "stats-account-failed",
4044             .type = QEMU_OPT_BOOL,
4045             .help = "whether to account for failed I/O operations "
4046                     "in the statistics",
4047         },
4048         { /* end of list */ }
4049     },
4050 };
4051 
4052 QemuOptsList qemu_drive_opts = {
4053     .name = "drive",
4054     .head = QTAILQ_HEAD_INITIALIZER(qemu_drive_opts.head),
4055     .desc = {
4056         /*
4057          * no elements => accept any params
4058          * validation will happen later
4059          */
4060         { /* end of list */ }
4061     },
4062 };
4063