xref: /openbmc/qemu/blockdev.c (revision 0d5fae3e)
1 /*
2  * QEMU host block devices
3  *
4  * Copyright (c) 2003-2008 Fabrice Bellard
5  *
6  * This work is licensed under the terms of the GNU GPL, version 2 or
7  * later.  See the COPYING file in the top-level directory.
8  *
9  * This file incorporates work covered by the following copyright and
10  * permission notice:
11  *
12  * Copyright (c) 2003-2008 Fabrice Bellard
13  *
14  * Permission is hereby granted, free of charge, to any person obtaining a copy
15  * of this software and associated documentation files (the "Software"), to deal
16  * in the Software without restriction, including without limitation the rights
17  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
18  * copies of the Software, and to permit persons to whom the Software is
19  * furnished to do so, subject to the following conditions:
20  *
21  * The above copyright notice and this permission notice shall be included in
22  * all copies or substantial portions of the Software.
23  *
24  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
25  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
26  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
27  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
28  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
29  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
30  * THE SOFTWARE.
31  */
32 
33 #include "qemu/osdep.h"
34 #include "sysemu/block-backend.h"
35 #include "sysemu/blockdev.h"
36 #include "hw/block/block.h"
37 #include "block/blockjob.h"
38 #include "block/qdict.h"
39 #include "block/throttle-groups.h"
40 #include "monitor/monitor.h"
41 #include "qemu/error-report.h"
42 #include "qemu/option.h"
43 #include "qemu/qemu-print.h"
44 #include "qemu/config-file.h"
45 #include "qapi/qapi-commands-block.h"
46 #include "qapi/qapi-commands-transaction.h"
47 #include "qapi/qapi-visit-block-core.h"
48 #include "qapi/qmp/qdict.h"
49 #include "qapi/qmp/qnum.h"
50 #include "qapi/qmp/qstring.h"
51 #include "qapi/error.h"
52 #include "qapi/qmp/qerror.h"
53 #include "qapi/qmp/qlist.h"
54 #include "qapi/qobject-output-visitor.h"
55 #include "sysemu/sysemu.h"
56 #include "sysemu/iothread.h"
57 #include "block/block_int.h"
58 #include "block/trace.h"
59 #include "sysemu/arch_init.h"
60 #include "sysemu/qtest.h"
61 #include "sysemu/runstate.h"
62 #include "qemu/cutils.h"
63 #include "qemu/help_option.h"
64 #include "qemu/main-loop.h"
65 #include "qemu/throttle-options.h"
66 
67 static QTAILQ_HEAD(, BlockDriverState) monitor_bdrv_states =
68     QTAILQ_HEAD_INITIALIZER(monitor_bdrv_states);
69 
70 static int do_open_tray(const char *blk_name, const char *qdev_id,
71                         bool force, Error **errp);
72 static void blockdev_remove_medium(bool has_device, const char *device,
73                                    bool has_id, const char *id, Error **errp);
74 static void blockdev_insert_medium(bool has_device, const char *device,
75                                    bool has_id, const char *id,
76                                    const char *node_name, Error **errp);
77 
78 static const char *const if_name[IF_COUNT] = {
79     [IF_NONE] = "none",
80     [IF_IDE] = "ide",
81     [IF_SCSI] = "scsi",
82     [IF_FLOPPY] = "floppy",
83     [IF_PFLASH] = "pflash",
84     [IF_MTD] = "mtd",
85     [IF_SD] = "sd",
86     [IF_VIRTIO] = "virtio",
87     [IF_XEN] = "xen",
88 };
89 
90 static int if_max_devs[IF_COUNT] = {
91     /*
92      * Do not change these numbers!  They govern how drive option
93      * index maps to unit and bus.  That mapping is ABI.
94      *
95      * All controllers used to implement if=T drives need to support
96      * if_max_devs[T] units, for any T with if_max_devs[T] != 0.
97      * Otherwise, some index values map to "impossible" bus, unit
98      * values.
99      *
100      * For instance, if you change [IF_SCSI] to 255, -drive
101      * if=scsi,index=12 no longer means bus=1,unit=5, but
102      * bus=0,unit=12.  With an lsi53c895a controller (7 units max),
103      * the drive can't be set up.  Regression.
104      */
105     [IF_IDE] = 2,
106     [IF_SCSI] = 7,
107 };
108 
109 /**
110  * Boards may call this to offer board-by-board overrides
111  * of the default, global values.
112  */
113 void override_max_devs(BlockInterfaceType type, int max_devs)
114 {
115     BlockBackend *blk;
116     DriveInfo *dinfo;
117 
118     if (max_devs <= 0) {
119         return;
120     }
121 
122     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
123         dinfo = blk_legacy_dinfo(blk);
124         if (dinfo->type == type) {
125             fprintf(stderr, "Cannot override units-per-bus property of"
126                     " the %s interface, because a drive of that type has"
127                     " already been added.\n", if_name[type]);
128             g_assert_not_reached();
129         }
130     }
131 
132     if_max_devs[type] = max_devs;
133 }
134 
135 /*
136  * We automatically delete the drive when a device using it gets
137  * unplugged.  Questionable feature, but we can't just drop it.
138  * Device models call blockdev_mark_auto_del() to schedule the
139  * automatic deletion, and generic qdev code calls blockdev_auto_del()
140  * when deletion is actually safe.
141  */
142 void blockdev_mark_auto_del(BlockBackend *blk)
143 {
144     DriveInfo *dinfo = blk_legacy_dinfo(blk);
145     BlockJob *job;
146 
147     if (!dinfo) {
148         return;
149     }
150 
151     for (job = block_job_next(NULL); job; job = block_job_next(job)) {
152         if (block_job_has_bdrv(job, blk_bs(blk))) {
153             AioContext *aio_context = job->job.aio_context;
154             aio_context_acquire(aio_context);
155 
156             job_cancel(&job->job, false);
157 
158             aio_context_release(aio_context);
159         }
160     }
161 
162     dinfo->auto_del = 1;
163 }
164 
165 void blockdev_auto_del(BlockBackend *blk)
166 {
167     DriveInfo *dinfo = blk_legacy_dinfo(blk);
168 
169     if (dinfo && dinfo->auto_del) {
170         monitor_remove_blk(blk);
171         blk_unref(blk);
172     }
173 }
174 
175 /**
176  * Returns the current mapping of how many units per bus
177  * a particular interface can support.
178  *
179  *  A positive integer indicates n units per bus.
180  *  0 implies the mapping has not been established.
181  * -1 indicates an invalid BlockInterfaceType was given.
182  */
183 int drive_get_max_devs(BlockInterfaceType type)
184 {
185     if (type >= IF_IDE && type < IF_COUNT) {
186         return if_max_devs[type];
187     }
188 
189     return -1;
190 }
191 
192 static int drive_index_to_bus_id(BlockInterfaceType type, int index)
193 {
194     int max_devs = if_max_devs[type];
195     return max_devs ? index / max_devs : 0;
196 }
197 
198 static int drive_index_to_unit_id(BlockInterfaceType type, int index)
199 {
200     int max_devs = if_max_devs[type];
201     return max_devs ? index % max_devs : index;
202 }
203 
204 QemuOpts *drive_def(const char *optstr)
205 {
206     return qemu_opts_parse_noisily(qemu_find_opts("drive"), optstr, false);
207 }
208 
209 QemuOpts *drive_add(BlockInterfaceType type, int index, const char *file,
210                     const char *optstr)
211 {
212     QemuOpts *opts;
213 
214     opts = drive_def(optstr);
215     if (!opts) {
216         return NULL;
217     }
218     if (type != IF_DEFAULT) {
219         qemu_opt_set(opts, "if", if_name[type], &error_abort);
220     }
221     if (index >= 0) {
222         qemu_opt_set_number(opts, "index", index, &error_abort);
223     }
224     if (file)
225         qemu_opt_set(opts, "file", file, &error_abort);
226     return opts;
227 }
228 
229 DriveInfo *drive_get(BlockInterfaceType type, int bus, int unit)
230 {
231     BlockBackend *blk;
232     DriveInfo *dinfo;
233 
234     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
235         dinfo = blk_legacy_dinfo(blk);
236         if (dinfo && dinfo->type == type
237             && dinfo->bus == bus && dinfo->unit == unit) {
238             return dinfo;
239         }
240     }
241 
242     return NULL;
243 }
244 
245 void drive_check_orphaned(void)
246 {
247     BlockBackend *blk;
248     DriveInfo *dinfo;
249     Location loc;
250     bool orphans = false;
251 
252     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
253         dinfo = blk_legacy_dinfo(blk);
254         if (!blk_get_attached_dev(blk) && !dinfo->is_default &&
255             dinfo->type != IF_NONE) {
256             loc_push_none(&loc);
257             qemu_opts_loc_restore(dinfo->opts);
258             error_report("machine type does not support"
259                          " if=%s,bus=%d,unit=%d",
260                          if_name[dinfo->type], dinfo->bus, dinfo->unit);
261             loc_pop(&loc);
262             orphans = true;
263         }
264     }
265 
266     if (orphans) {
267         exit(1);
268     }
269 }
270 
271 DriveInfo *drive_get_by_index(BlockInterfaceType type, int index)
272 {
273     return drive_get(type,
274                      drive_index_to_bus_id(type, index),
275                      drive_index_to_unit_id(type, index));
276 }
277 
278 int drive_get_max_bus(BlockInterfaceType type)
279 {
280     int max_bus;
281     BlockBackend *blk;
282     DriveInfo *dinfo;
283 
284     max_bus = -1;
285     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
286         dinfo = blk_legacy_dinfo(blk);
287         if (dinfo && dinfo->type == type && dinfo->bus > max_bus) {
288             max_bus = dinfo->bus;
289         }
290     }
291     return max_bus;
292 }
293 
294 /* Get a block device.  This should only be used for single-drive devices
295    (e.g. SD/Floppy/MTD).  Multi-disk devices (scsi/ide) should use the
296    appropriate bus.  */
297 DriveInfo *drive_get_next(BlockInterfaceType type)
298 {
299     static int next_block_unit[IF_COUNT];
300 
301     return drive_get(type, 0, next_block_unit[type]++);
302 }
303 
304 static void bdrv_format_print(void *opaque, const char *name)
305 {
306     qemu_printf(" %s", name);
307 }
308 
309 typedef struct {
310     QEMUBH *bh;
311     BlockDriverState *bs;
312 } BDRVPutRefBH;
313 
314 static int parse_block_error_action(const char *buf, bool is_read, Error **errp)
315 {
316     if (!strcmp(buf, "ignore")) {
317         return BLOCKDEV_ON_ERROR_IGNORE;
318     } else if (!is_read && !strcmp(buf, "enospc")) {
319         return BLOCKDEV_ON_ERROR_ENOSPC;
320     } else if (!strcmp(buf, "stop")) {
321         return BLOCKDEV_ON_ERROR_STOP;
322     } else if (!strcmp(buf, "report")) {
323         return BLOCKDEV_ON_ERROR_REPORT;
324     } else {
325         error_setg(errp, "'%s' invalid %s error action",
326                    buf, is_read ? "read" : "write");
327         return -1;
328     }
329 }
330 
331 static bool parse_stats_intervals(BlockAcctStats *stats, QList *intervals,
332                                   Error **errp)
333 {
334     const QListEntry *entry;
335     for (entry = qlist_first(intervals); entry; entry = qlist_next(entry)) {
336         switch (qobject_type(entry->value)) {
337 
338         case QTYPE_QSTRING: {
339             unsigned long long length;
340             const char *str = qstring_get_str(qobject_to(QString,
341                                                          entry->value));
342             if (parse_uint_full(str, &length, 10) == 0 &&
343                 length > 0 && length <= UINT_MAX) {
344                 block_acct_add_interval(stats, (unsigned) length);
345             } else {
346                 error_setg(errp, "Invalid interval length: %s", str);
347                 return false;
348             }
349             break;
350         }
351 
352         case QTYPE_QNUM: {
353             int64_t length = qnum_get_int(qobject_to(QNum, entry->value));
354 
355             if (length > 0 && length <= UINT_MAX) {
356                 block_acct_add_interval(stats, (unsigned) length);
357             } else {
358                 error_setg(errp, "Invalid interval length: %" PRId64, length);
359                 return false;
360             }
361             break;
362         }
363 
364         default:
365             error_setg(errp, "The specification of stats-intervals is invalid");
366             return false;
367         }
368     }
369     return true;
370 }
371 
372 typedef enum { MEDIA_DISK, MEDIA_CDROM } DriveMediaType;
373 
374 /* All parameters but @opts are optional and may be set to NULL. */
375 static void extract_common_blockdev_options(QemuOpts *opts, int *bdrv_flags,
376     const char **throttling_group, ThrottleConfig *throttle_cfg,
377     BlockdevDetectZeroesOptions *detect_zeroes, Error **errp)
378 {
379     Error *local_error = NULL;
380     const char *aio;
381 
382     if (bdrv_flags) {
383         if (qemu_opt_get_bool(opts, "copy-on-read", false)) {
384             *bdrv_flags |= BDRV_O_COPY_ON_READ;
385         }
386 
387         if ((aio = qemu_opt_get(opts, "aio")) != NULL) {
388             if (!strcmp(aio, "native")) {
389                 *bdrv_flags |= BDRV_O_NATIVE_AIO;
390             } else if (!strcmp(aio, "threads")) {
391                 /* this is the default */
392             } else {
393                error_setg(errp, "invalid aio option");
394                return;
395             }
396         }
397     }
398 
399     /* disk I/O throttling */
400     if (throttling_group) {
401         *throttling_group = qemu_opt_get(opts, "throttling.group");
402     }
403 
404     if (throttle_cfg) {
405         throttle_config_init(throttle_cfg);
406         throttle_cfg->buckets[THROTTLE_BPS_TOTAL].avg =
407             qemu_opt_get_number(opts, "throttling.bps-total", 0);
408         throttle_cfg->buckets[THROTTLE_BPS_READ].avg  =
409             qemu_opt_get_number(opts, "throttling.bps-read", 0);
410         throttle_cfg->buckets[THROTTLE_BPS_WRITE].avg =
411             qemu_opt_get_number(opts, "throttling.bps-write", 0);
412         throttle_cfg->buckets[THROTTLE_OPS_TOTAL].avg =
413             qemu_opt_get_number(opts, "throttling.iops-total", 0);
414         throttle_cfg->buckets[THROTTLE_OPS_READ].avg =
415             qemu_opt_get_number(opts, "throttling.iops-read", 0);
416         throttle_cfg->buckets[THROTTLE_OPS_WRITE].avg =
417             qemu_opt_get_number(opts, "throttling.iops-write", 0);
418 
419         throttle_cfg->buckets[THROTTLE_BPS_TOTAL].max =
420             qemu_opt_get_number(opts, "throttling.bps-total-max", 0);
421         throttle_cfg->buckets[THROTTLE_BPS_READ].max  =
422             qemu_opt_get_number(opts, "throttling.bps-read-max", 0);
423         throttle_cfg->buckets[THROTTLE_BPS_WRITE].max =
424             qemu_opt_get_number(opts, "throttling.bps-write-max", 0);
425         throttle_cfg->buckets[THROTTLE_OPS_TOTAL].max =
426             qemu_opt_get_number(opts, "throttling.iops-total-max", 0);
427         throttle_cfg->buckets[THROTTLE_OPS_READ].max =
428             qemu_opt_get_number(opts, "throttling.iops-read-max", 0);
429         throttle_cfg->buckets[THROTTLE_OPS_WRITE].max =
430             qemu_opt_get_number(opts, "throttling.iops-write-max", 0);
431 
432         throttle_cfg->buckets[THROTTLE_BPS_TOTAL].burst_length =
433             qemu_opt_get_number(opts, "throttling.bps-total-max-length", 1);
434         throttle_cfg->buckets[THROTTLE_BPS_READ].burst_length  =
435             qemu_opt_get_number(opts, "throttling.bps-read-max-length", 1);
436         throttle_cfg->buckets[THROTTLE_BPS_WRITE].burst_length =
437             qemu_opt_get_number(opts, "throttling.bps-write-max-length", 1);
438         throttle_cfg->buckets[THROTTLE_OPS_TOTAL].burst_length =
439             qemu_opt_get_number(opts, "throttling.iops-total-max-length", 1);
440         throttle_cfg->buckets[THROTTLE_OPS_READ].burst_length =
441             qemu_opt_get_number(opts, "throttling.iops-read-max-length", 1);
442         throttle_cfg->buckets[THROTTLE_OPS_WRITE].burst_length =
443             qemu_opt_get_number(opts, "throttling.iops-write-max-length", 1);
444 
445         throttle_cfg->op_size =
446             qemu_opt_get_number(opts, "throttling.iops-size", 0);
447 
448         if (!throttle_is_valid(throttle_cfg, errp)) {
449             return;
450         }
451     }
452 
453     if (detect_zeroes) {
454         *detect_zeroes =
455             qapi_enum_parse(&BlockdevDetectZeroesOptions_lookup,
456                             qemu_opt_get(opts, "detect-zeroes"),
457                             BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF,
458                             &local_error);
459         if (local_error) {
460             error_propagate(errp, local_error);
461             return;
462         }
463     }
464 }
465 
466 /* Takes the ownership of bs_opts */
467 static BlockBackend *blockdev_init(const char *file, QDict *bs_opts,
468                                    Error **errp)
469 {
470     const char *buf;
471     int bdrv_flags = 0;
472     int on_read_error, on_write_error;
473     bool account_invalid, account_failed;
474     bool writethrough, read_only;
475     BlockBackend *blk;
476     BlockDriverState *bs;
477     ThrottleConfig cfg;
478     int snapshot = 0;
479     Error *error = NULL;
480     QemuOpts *opts;
481     QDict *interval_dict = NULL;
482     QList *interval_list = NULL;
483     const char *id;
484     BlockdevDetectZeroesOptions detect_zeroes =
485         BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF;
486     const char *throttling_group = NULL;
487 
488     /* Check common options by copying from bs_opts to opts, all other options
489      * stay in bs_opts for processing by bdrv_open(). */
490     id = qdict_get_try_str(bs_opts, "id");
491     opts = qemu_opts_create(&qemu_common_drive_opts, id, 1, &error);
492     if (error) {
493         error_propagate(errp, error);
494         goto err_no_opts;
495     }
496 
497     qemu_opts_absorb_qdict(opts, bs_opts, &error);
498     if (error) {
499         error_propagate(errp, error);
500         goto early_err;
501     }
502 
503     if (id) {
504         qdict_del(bs_opts, "id");
505     }
506 
507     /* extract parameters */
508     snapshot = qemu_opt_get_bool(opts, "snapshot", 0);
509 
510     account_invalid = qemu_opt_get_bool(opts, "stats-account-invalid", true);
511     account_failed = qemu_opt_get_bool(opts, "stats-account-failed", true);
512 
513     writethrough = !qemu_opt_get_bool(opts, BDRV_OPT_CACHE_WB, true);
514 
515     id = qemu_opts_id(opts);
516 
517     qdict_extract_subqdict(bs_opts, &interval_dict, "stats-intervals.");
518     qdict_array_split(interval_dict, &interval_list);
519 
520     if (qdict_size(interval_dict) != 0) {
521         error_setg(errp, "Invalid option stats-intervals.%s",
522                    qdict_first(interval_dict)->key);
523         goto early_err;
524     }
525 
526     extract_common_blockdev_options(opts, &bdrv_flags, &throttling_group, &cfg,
527                                     &detect_zeroes, &error);
528     if (error) {
529         error_propagate(errp, error);
530         goto early_err;
531     }
532 
533     if ((buf = qemu_opt_get(opts, "format")) != NULL) {
534         if (is_help_option(buf)) {
535             qemu_printf("Supported formats:");
536             bdrv_iterate_format(bdrv_format_print, NULL, false);
537             qemu_printf("\nSupported formats (read-only):");
538             bdrv_iterate_format(bdrv_format_print, NULL, true);
539             qemu_printf("\n");
540             goto early_err;
541         }
542 
543         if (qdict_haskey(bs_opts, "driver")) {
544             error_setg(errp, "Cannot specify both 'driver' and 'format'");
545             goto early_err;
546         }
547         qdict_put_str(bs_opts, "driver", buf);
548     }
549 
550     on_write_error = BLOCKDEV_ON_ERROR_ENOSPC;
551     if ((buf = qemu_opt_get(opts, "werror")) != NULL) {
552         on_write_error = parse_block_error_action(buf, 0, &error);
553         if (error) {
554             error_propagate(errp, error);
555             goto early_err;
556         }
557     }
558 
559     on_read_error = BLOCKDEV_ON_ERROR_REPORT;
560     if ((buf = qemu_opt_get(opts, "rerror")) != NULL) {
561         on_read_error = parse_block_error_action(buf, 1, &error);
562         if (error) {
563             error_propagate(errp, error);
564             goto early_err;
565         }
566     }
567 
568     if (snapshot) {
569         bdrv_flags |= BDRV_O_SNAPSHOT;
570     }
571 
572     read_only = qemu_opt_get_bool(opts, BDRV_OPT_READ_ONLY, false);
573 
574     /* init */
575     if ((!file || !*file) && !qdict_size(bs_opts)) {
576         BlockBackendRootState *blk_rs;
577 
578         blk = blk_new(qemu_get_aio_context(), 0, BLK_PERM_ALL);
579         blk_rs = blk_get_root_state(blk);
580         blk_rs->open_flags    = bdrv_flags;
581         blk_rs->read_only     = read_only;
582         blk_rs->detect_zeroes = detect_zeroes;
583 
584         qobject_unref(bs_opts);
585     } else {
586         if (file && !*file) {
587             file = NULL;
588         }
589 
590         /* bdrv_open() defaults to the values in bdrv_flags (for compatibility
591          * with other callers) rather than what we want as the real defaults.
592          * Apply the defaults here instead. */
593         qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_DIRECT, "off");
594         qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_NO_FLUSH, "off");
595         qdict_set_default_str(bs_opts, BDRV_OPT_READ_ONLY,
596                               read_only ? "on" : "off");
597         qdict_set_default_str(bs_opts, BDRV_OPT_AUTO_READ_ONLY, "on");
598         assert((bdrv_flags & BDRV_O_CACHE_MASK) == 0);
599 
600         if (runstate_check(RUN_STATE_INMIGRATE)) {
601             bdrv_flags |= BDRV_O_INACTIVE;
602         }
603 
604         blk = blk_new_open(file, NULL, bs_opts, bdrv_flags, errp);
605         if (!blk) {
606             goto err_no_bs_opts;
607         }
608         bs = blk_bs(blk);
609 
610         bs->detect_zeroes = detect_zeroes;
611 
612         block_acct_setup(blk_get_stats(blk), account_invalid, account_failed);
613 
614         if (!parse_stats_intervals(blk_get_stats(blk), interval_list, errp)) {
615             blk_unref(blk);
616             blk = NULL;
617             goto err_no_bs_opts;
618         }
619     }
620 
621     /* disk I/O throttling */
622     if (throttle_enabled(&cfg)) {
623         if (!throttling_group) {
624             throttling_group = id;
625         }
626         blk_io_limits_enable(blk, throttling_group);
627         blk_set_io_limits(blk, &cfg);
628     }
629 
630     blk_set_enable_write_cache(blk, !writethrough);
631     blk_set_on_error(blk, on_read_error, on_write_error);
632 
633     if (!monitor_add_blk(blk, id, errp)) {
634         blk_unref(blk);
635         blk = NULL;
636         goto err_no_bs_opts;
637     }
638 
639 err_no_bs_opts:
640     qemu_opts_del(opts);
641     qobject_unref(interval_dict);
642     qobject_unref(interval_list);
643     return blk;
644 
645 early_err:
646     qemu_opts_del(opts);
647     qobject_unref(interval_dict);
648     qobject_unref(interval_list);
649 err_no_opts:
650     qobject_unref(bs_opts);
651     return NULL;
652 }
653 
654 /* Takes the ownership of bs_opts */
655 static BlockDriverState *bds_tree_init(QDict *bs_opts, Error **errp)
656 {
657     int bdrv_flags = 0;
658 
659     /* bdrv_open() defaults to the values in bdrv_flags (for compatibility
660      * with other callers) rather than what we want as the real defaults.
661      * Apply the defaults here instead. */
662     qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_DIRECT, "off");
663     qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_NO_FLUSH, "off");
664     qdict_set_default_str(bs_opts, BDRV_OPT_READ_ONLY, "off");
665 
666     if (runstate_check(RUN_STATE_INMIGRATE)) {
667         bdrv_flags |= BDRV_O_INACTIVE;
668     }
669 
670     return bdrv_open(NULL, NULL, bs_opts, bdrv_flags, errp);
671 }
672 
673 void blockdev_close_all_bdrv_states(void)
674 {
675     BlockDriverState *bs, *next_bs;
676 
677     QTAILQ_FOREACH_SAFE(bs, &monitor_bdrv_states, monitor_list, next_bs) {
678         AioContext *ctx = bdrv_get_aio_context(bs);
679 
680         aio_context_acquire(ctx);
681         bdrv_unref(bs);
682         aio_context_release(ctx);
683     }
684 }
685 
686 /* Iterates over the list of monitor-owned BlockDriverStates */
687 BlockDriverState *bdrv_next_monitor_owned(BlockDriverState *bs)
688 {
689     return bs ? QTAILQ_NEXT(bs, monitor_list)
690               : QTAILQ_FIRST(&monitor_bdrv_states);
691 }
692 
693 static void qemu_opt_rename(QemuOpts *opts, const char *from, const char *to,
694                             Error **errp)
695 {
696     const char *value;
697 
698     value = qemu_opt_get(opts, from);
699     if (value) {
700         if (qemu_opt_find(opts, to)) {
701             error_setg(errp, "'%s' and its alias '%s' can't be used at the "
702                        "same time", to, from);
703             return;
704         }
705     }
706 
707     /* rename all items in opts */
708     while ((value = qemu_opt_get(opts, from))) {
709         qemu_opt_set(opts, to, value, &error_abort);
710         qemu_opt_unset(opts, from);
711     }
712 }
713 
714 QemuOptsList qemu_legacy_drive_opts = {
715     .name = "drive",
716     .head = QTAILQ_HEAD_INITIALIZER(qemu_legacy_drive_opts.head),
717     .desc = {
718         {
719             .name = "bus",
720             .type = QEMU_OPT_NUMBER,
721             .help = "bus number",
722         },{
723             .name = "unit",
724             .type = QEMU_OPT_NUMBER,
725             .help = "unit number (i.e. lun for scsi)",
726         },{
727             .name = "index",
728             .type = QEMU_OPT_NUMBER,
729             .help = "index number",
730         },{
731             .name = "media",
732             .type = QEMU_OPT_STRING,
733             .help = "media type (disk, cdrom)",
734         },{
735             .name = "if",
736             .type = QEMU_OPT_STRING,
737             .help = "interface (ide, scsi, sd, mtd, floppy, pflash, virtio)",
738         },{
739             .name = "file",
740             .type = QEMU_OPT_STRING,
741             .help = "file name",
742         },
743 
744         /* Options that are passed on, but have special semantics with -drive */
745         {
746             .name = BDRV_OPT_READ_ONLY,
747             .type = QEMU_OPT_BOOL,
748             .help = "open drive file as read-only",
749         },{
750             .name = "rerror",
751             .type = QEMU_OPT_STRING,
752             .help = "read error action",
753         },{
754             .name = "werror",
755             .type = QEMU_OPT_STRING,
756             .help = "write error action",
757         },{
758             .name = "copy-on-read",
759             .type = QEMU_OPT_BOOL,
760             .help = "copy read data from backing file into image file",
761         },
762 
763         { /* end of list */ }
764     },
765 };
766 
767 DriveInfo *drive_new(QemuOpts *all_opts, BlockInterfaceType block_default_type,
768                      Error **errp)
769 {
770     const char *value;
771     BlockBackend *blk;
772     DriveInfo *dinfo = NULL;
773     QDict *bs_opts;
774     QemuOpts *legacy_opts;
775     DriveMediaType media = MEDIA_DISK;
776     BlockInterfaceType type;
777     int max_devs, bus_id, unit_id, index;
778     const char *werror, *rerror;
779     bool read_only = false;
780     bool copy_on_read;
781     const char *filename;
782     Error *local_err = NULL;
783     int i;
784 
785     /* Change legacy command line options into QMP ones */
786     static const struct {
787         const char *from;
788         const char *to;
789     } opt_renames[] = {
790         { "iops",           "throttling.iops-total" },
791         { "iops_rd",        "throttling.iops-read" },
792         { "iops_wr",        "throttling.iops-write" },
793 
794         { "bps",            "throttling.bps-total" },
795         { "bps_rd",         "throttling.bps-read" },
796         { "bps_wr",         "throttling.bps-write" },
797 
798         { "iops_max",       "throttling.iops-total-max" },
799         { "iops_rd_max",    "throttling.iops-read-max" },
800         { "iops_wr_max",    "throttling.iops-write-max" },
801 
802         { "bps_max",        "throttling.bps-total-max" },
803         { "bps_rd_max",     "throttling.bps-read-max" },
804         { "bps_wr_max",     "throttling.bps-write-max" },
805 
806         { "iops_size",      "throttling.iops-size" },
807 
808         { "group",          "throttling.group" },
809 
810         { "readonly",       BDRV_OPT_READ_ONLY },
811     };
812 
813     for (i = 0; i < ARRAY_SIZE(opt_renames); i++) {
814         qemu_opt_rename(all_opts, opt_renames[i].from, opt_renames[i].to,
815                         &local_err);
816         if (local_err) {
817             error_propagate(errp, local_err);
818             return NULL;
819         }
820     }
821 
822     value = qemu_opt_get(all_opts, "cache");
823     if (value) {
824         int flags = 0;
825         bool writethrough;
826 
827         if (bdrv_parse_cache_mode(value, &flags, &writethrough) != 0) {
828             error_setg(errp, "invalid cache option");
829             return NULL;
830         }
831 
832         /* Specific options take precedence */
833         if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_WB)) {
834             qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_WB,
835                               !writethrough, &error_abort);
836         }
837         if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_DIRECT)) {
838             qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_DIRECT,
839                               !!(flags & BDRV_O_NOCACHE), &error_abort);
840         }
841         if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_NO_FLUSH)) {
842             qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_NO_FLUSH,
843                               !!(flags & BDRV_O_NO_FLUSH), &error_abort);
844         }
845         qemu_opt_unset(all_opts, "cache");
846     }
847 
848     /* Get a QDict for processing the options */
849     bs_opts = qdict_new();
850     qemu_opts_to_qdict(all_opts, bs_opts);
851 
852     legacy_opts = qemu_opts_create(&qemu_legacy_drive_opts, NULL, 0,
853                                    &error_abort);
854     qemu_opts_absorb_qdict(legacy_opts, bs_opts, &local_err);
855     if (local_err) {
856         error_propagate(errp, local_err);
857         goto fail;
858     }
859 
860     /* Media type */
861     value = qemu_opt_get(legacy_opts, "media");
862     if (value) {
863         if (!strcmp(value, "disk")) {
864             media = MEDIA_DISK;
865         } else if (!strcmp(value, "cdrom")) {
866             media = MEDIA_CDROM;
867             read_only = true;
868         } else {
869             error_setg(errp, "'%s' invalid media", value);
870             goto fail;
871         }
872     }
873 
874     /* copy-on-read is disabled with a warning for read-only devices */
875     read_only |= qemu_opt_get_bool(legacy_opts, BDRV_OPT_READ_ONLY, false);
876     copy_on_read = qemu_opt_get_bool(legacy_opts, "copy-on-read", false);
877 
878     if (read_only && copy_on_read) {
879         warn_report("disabling copy-on-read on read-only drive");
880         copy_on_read = false;
881     }
882 
883     qdict_put_str(bs_opts, BDRV_OPT_READ_ONLY, read_only ? "on" : "off");
884     qdict_put_str(bs_opts, "copy-on-read", copy_on_read ? "on" : "off");
885 
886     /* Controller type */
887     value = qemu_opt_get(legacy_opts, "if");
888     if (value) {
889         for (type = 0;
890              type < IF_COUNT && strcmp(value, if_name[type]);
891              type++) {
892         }
893         if (type == IF_COUNT) {
894             error_setg(errp, "unsupported bus type '%s'", value);
895             goto fail;
896         }
897     } else {
898         type = block_default_type;
899     }
900 
901     /* Device address specified by bus/unit or index.
902      * If none was specified, try to find the first free one. */
903     bus_id  = qemu_opt_get_number(legacy_opts, "bus", 0);
904     unit_id = qemu_opt_get_number(legacy_opts, "unit", -1);
905     index   = qemu_opt_get_number(legacy_opts, "index", -1);
906 
907     max_devs = if_max_devs[type];
908 
909     if (index != -1) {
910         if (bus_id != 0 || unit_id != -1) {
911             error_setg(errp, "index cannot be used with bus and unit");
912             goto fail;
913         }
914         bus_id = drive_index_to_bus_id(type, index);
915         unit_id = drive_index_to_unit_id(type, index);
916     }
917 
918     if (unit_id == -1) {
919        unit_id = 0;
920        while (drive_get(type, bus_id, unit_id) != NULL) {
921            unit_id++;
922            if (max_devs && unit_id >= max_devs) {
923                unit_id -= max_devs;
924                bus_id++;
925            }
926        }
927     }
928 
929     if (max_devs && unit_id >= max_devs) {
930         error_setg(errp, "unit %d too big (max is %d)", unit_id, max_devs - 1);
931         goto fail;
932     }
933 
934     if (drive_get(type, bus_id, unit_id) != NULL) {
935         error_setg(errp, "drive with bus=%d, unit=%d (index=%d) exists",
936                    bus_id, unit_id, index);
937         goto fail;
938     }
939 
940     /* no id supplied -> create one */
941     if (qemu_opts_id(all_opts) == NULL) {
942         char *new_id;
943         const char *mediastr = "";
944         if (type == IF_IDE || type == IF_SCSI) {
945             mediastr = (media == MEDIA_CDROM) ? "-cd" : "-hd";
946         }
947         if (max_devs) {
948             new_id = g_strdup_printf("%s%i%s%i", if_name[type], bus_id,
949                                      mediastr, unit_id);
950         } else {
951             new_id = g_strdup_printf("%s%s%i", if_name[type],
952                                      mediastr, unit_id);
953         }
954         qdict_put_str(bs_opts, "id", new_id);
955         g_free(new_id);
956     }
957 
958     /* Add virtio block device */
959     if (type == IF_VIRTIO) {
960         QemuOpts *devopts;
961         devopts = qemu_opts_create(qemu_find_opts("device"), NULL, 0,
962                                    &error_abort);
963         if (arch_type == QEMU_ARCH_S390X) {
964             qemu_opt_set(devopts, "driver", "virtio-blk-ccw", &error_abort);
965         } else {
966             qemu_opt_set(devopts, "driver", "virtio-blk-pci", &error_abort);
967         }
968         qemu_opt_set(devopts, "drive", qdict_get_str(bs_opts, "id"),
969                      &error_abort);
970     }
971 
972     filename = qemu_opt_get(legacy_opts, "file");
973 
974     /* Check werror/rerror compatibility with if=... */
975     werror = qemu_opt_get(legacy_opts, "werror");
976     if (werror != NULL) {
977         if (type != IF_IDE && type != IF_SCSI && type != IF_VIRTIO &&
978             type != IF_NONE) {
979             error_setg(errp, "werror is not supported by this bus type");
980             goto fail;
981         }
982         qdict_put_str(bs_opts, "werror", werror);
983     }
984 
985     rerror = qemu_opt_get(legacy_opts, "rerror");
986     if (rerror != NULL) {
987         if (type != IF_IDE && type != IF_VIRTIO && type != IF_SCSI &&
988             type != IF_NONE) {
989             error_setg(errp, "rerror is not supported by this bus type");
990             goto fail;
991         }
992         qdict_put_str(bs_opts, "rerror", rerror);
993     }
994 
995     /* Actual block device init: Functionality shared with blockdev-add */
996     blk = blockdev_init(filename, bs_opts, &local_err);
997     bs_opts = NULL;
998     if (!blk) {
999         error_propagate(errp, local_err);
1000         goto fail;
1001     } else {
1002         assert(!local_err);
1003     }
1004 
1005     /* Create legacy DriveInfo */
1006     dinfo = g_malloc0(sizeof(*dinfo));
1007     dinfo->opts = all_opts;
1008 
1009     dinfo->type = type;
1010     dinfo->bus = bus_id;
1011     dinfo->unit = unit_id;
1012 
1013     blk_set_legacy_dinfo(blk, dinfo);
1014 
1015     switch(type) {
1016     case IF_IDE:
1017     case IF_SCSI:
1018     case IF_XEN:
1019     case IF_NONE:
1020         dinfo->media_cd = media == MEDIA_CDROM;
1021         break;
1022     default:
1023         break;
1024     }
1025 
1026 fail:
1027     qemu_opts_del(legacy_opts);
1028     qobject_unref(bs_opts);
1029     return dinfo;
1030 }
1031 
1032 static BlockDriverState *qmp_get_root_bs(const char *name, Error **errp)
1033 {
1034     BlockDriverState *bs;
1035 
1036     bs = bdrv_lookup_bs(name, name, errp);
1037     if (bs == NULL) {
1038         return NULL;
1039     }
1040 
1041     if (!bdrv_is_root_node(bs)) {
1042         error_setg(errp, "Need a root block node");
1043         return NULL;
1044     }
1045 
1046     if (!bdrv_is_inserted(bs)) {
1047         error_setg(errp, "Device has no medium");
1048         return NULL;
1049     }
1050 
1051     return bs;
1052 }
1053 
1054 static BlockBackend *qmp_get_blk(const char *blk_name, const char *qdev_id,
1055                                  Error **errp)
1056 {
1057     BlockBackend *blk;
1058 
1059     if (!blk_name == !qdev_id) {
1060         error_setg(errp, "Need exactly one of 'device' and 'id'");
1061         return NULL;
1062     }
1063 
1064     if (qdev_id) {
1065         blk = blk_by_qdev_id(qdev_id, errp);
1066     } else {
1067         blk = blk_by_name(blk_name);
1068         if (blk == NULL) {
1069             error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
1070                       "Device '%s' not found", blk_name);
1071         }
1072     }
1073 
1074     return blk;
1075 }
1076 
1077 void hmp_commit(Monitor *mon, const QDict *qdict)
1078 {
1079     const char *device = qdict_get_str(qdict, "device");
1080     BlockBackend *blk;
1081     int ret;
1082 
1083     if (!strcmp(device, "all")) {
1084         ret = blk_commit_all();
1085     } else {
1086         BlockDriverState *bs;
1087         AioContext *aio_context;
1088 
1089         blk = blk_by_name(device);
1090         if (!blk) {
1091             monitor_printf(mon, "Device '%s' not found\n", device);
1092             return;
1093         }
1094         if (!blk_is_available(blk)) {
1095             monitor_printf(mon, "Device '%s' has no medium\n", device);
1096             return;
1097         }
1098 
1099         bs = blk_bs(blk);
1100         aio_context = bdrv_get_aio_context(bs);
1101         aio_context_acquire(aio_context);
1102 
1103         ret = bdrv_commit(bs);
1104 
1105         aio_context_release(aio_context);
1106     }
1107     if (ret < 0) {
1108         monitor_printf(mon, "'commit' error for '%s': %s\n", device,
1109                        strerror(-ret));
1110     }
1111 }
1112 
1113 static void blockdev_do_action(TransactionAction *action, Error **errp)
1114 {
1115     TransactionActionList list;
1116 
1117     list.value = action;
1118     list.next = NULL;
1119     qmp_transaction(&list, false, NULL, errp);
1120 }
1121 
1122 void qmp_blockdev_snapshot_sync(bool has_device, const char *device,
1123                                 bool has_node_name, const char *node_name,
1124                                 const char *snapshot_file,
1125                                 bool has_snapshot_node_name,
1126                                 const char *snapshot_node_name,
1127                                 bool has_format, const char *format,
1128                                 bool has_mode, NewImageMode mode, Error **errp)
1129 {
1130     BlockdevSnapshotSync snapshot = {
1131         .has_device = has_device,
1132         .device = (char *) device,
1133         .has_node_name = has_node_name,
1134         .node_name = (char *) node_name,
1135         .snapshot_file = (char *) snapshot_file,
1136         .has_snapshot_node_name = has_snapshot_node_name,
1137         .snapshot_node_name = (char *) snapshot_node_name,
1138         .has_format = has_format,
1139         .format = (char *) format,
1140         .has_mode = has_mode,
1141         .mode = mode,
1142     };
1143     TransactionAction action = {
1144         .type = TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC,
1145         .u.blockdev_snapshot_sync.data = &snapshot,
1146     };
1147     blockdev_do_action(&action, errp);
1148 }
1149 
1150 void qmp_blockdev_snapshot(const char *node, const char *overlay,
1151                            Error **errp)
1152 {
1153     BlockdevSnapshot snapshot_data = {
1154         .node = (char *) node,
1155         .overlay = (char *) overlay
1156     };
1157     TransactionAction action = {
1158         .type = TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT,
1159         .u.blockdev_snapshot.data = &snapshot_data,
1160     };
1161     blockdev_do_action(&action, errp);
1162 }
1163 
1164 void qmp_blockdev_snapshot_internal_sync(const char *device,
1165                                          const char *name,
1166                                          Error **errp)
1167 {
1168     BlockdevSnapshotInternal snapshot = {
1169         .device = (char *) device,
1170         .name = (char *) name
1171     };
1172     TransactionAction action = {
1173         .type = TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC,
1174         .u.blockdev_snapshot_internal_sync.data = &snapshot,
1175     };
1176     blockdev_do_action(&action, errp);
1177 }
1178 
1179 SnapshotInfo *qmp_blockdev_snapshot_delete_internal_sync(const char *device,
1180                                                          bool has_id,
1181                                                          const char *id,
1182                                                          bool has_name,
1183                                                          const char *name,
1184                                                          Error **errp)
1185 {
1186     BlockDriverState *bs;
1187     AioContext *aio_context;
1188     QEMUSnapshotInfo sn;
1189     Error *local_err = NULL;
1190     SnapshotInfo *info = NULL;
1191     int ret;
1192 
1193     bs = qmp_get_root_bs(device, errp);
1194     if (!bs) {
1195         return NULL;
1196     }
1197     aio_context = bdrv_get_aio_context(bs);
1198     aio_context_acquire(aio_context);
1199 
1200     if (!has_id) {
1201         id = NULL;
1202     }
1203 
1204     if (!has_name) {
1205         name = NULL;
1206     }
1207 
1208     if (!id && !name) {
1209         error_setg(errp, "Name or id must be provided");
1210         goto out_aio_context;
1211     }
1212 
1213     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_INTERNAL_SNAPSHOT_DELETE, errp)) {
1214         goto out_aio_context;
1215     }
1216 
1217     ret = bdrv_snapshot_find_by_id_and_name(bs, id, name, &sn, &local_err);
1218     if (local_err) {
1219         error_propagate(errp, local_err);
1220         goto out_aio_context;
1221     }
1222     if (!ret) {
1223         error_setg(errp,
1224                    "Snapshot with id '%s' and name '%s' does not exist on "
1225                    "device '%s'",
1226                    STR_OR_NULL(id), STR_OR_NULL(name), device);
1227         goto out_aio_context;
1228     }
1229 
1230     bdrv_snapshot_delete(bs, id, name, &local_err);
1231     if (local_err) {
1232         error_propagate(errp, local_err);
1233         goto out_aio_context;
1234     }
1235 
1236     aio_context_release(aio_context);
1237 
1238     info = g_new0(SnapshotInfo, 1);
1239     info->id = g_strdup(sn.id_str);
1240     info->name = g_strdup(sn.name);
1241     info->date_nsec = sn.date_nsec;
1242     info->date_sec = sn.date_sec;
1243     info->vm_state_size = sn.vm_state_size;
1244     info->vm_clock_nsec = sn.vm_clock_nsec % 1000000000;
1245     info->vm_clock_sec = sn.vm_clock_nsec / 1000000000;
1246 
1247     return info;
1248 
1249 out_aio_context:
1250     aio_context_release(aio_context);
1251     return NULL;
1252 }
1253 
1254 /**
1255  * block_dirty_bitmap_lookup:
1256  * Return a dirty bitmap (if present), after validating
1257  * the node reference and bitmap names.
1258  *
1259  * @node: The name of the BDS node to search for bitmaps
1260  * @name: The name of the bitmap to search for
1261  * @pbs: Output pointer for BDS lookup, if desired. Can be NULL.
1262  * @errp: Output pointer for error information. Can be NULL.
1263  *
1264  * @return: A bitmap object on success, or NULL on failure.
1265  */
1266 static BdrvDirtyBitmap *block_dirty_bitmap_lookup(const char *node,
1267                                                   const char *name,
1268                                                   BlockDriverState **pbs,
1269                                                   Error **errp)
1270 {
1271     BlockDriverState *bs;
1272     BdrvDirtyBitmap *bitmap;
1273 
1274     if (!node) {
1275         error_setg(errp, "Node cannot be NULL");
1276         return NULL;
1277     }
1278     if (!name) {
1279         error_setg(errp, "Bitmap name cannot be NULL");
1280         return NULL;
1281     }
1282     bs = bdrv_lookup_bs(node, node, NULL);
1283     if (!bs) {
1284         error_setg(errp, "Node '%s' not found", node);
1285         return NULL;
1286     }
1287 
1288     bitmap = bdrv_find_dirty_bitmap(bs, name);
1289     if (!bitmap) {
1290         error_setg(errp, "Dirty bitmap '%s' not found", name);
1291         return NULL;
1292     }
1293 
1294     if (pbs) {
1295         *pbs = bs;
1296     }
1297 
1298     return bitmap;
1299 }
1300 
1301 /* New and old BlockDriverState structs for atomic group operations */
1302 
1303 typedef struct BlkActionState BlkActionState;
1304 
1305 /**
1306  * BlkActionOps:
1307  * Table of operations that define an Action.
1308  *
1309  * @instance_size: Size of state struct, in bytes.
1310  * @prepare: Prepare the work, must NOT be NULL.
1311  * @commit: Commit the changes, can be NULL.
1312  * @abort: Abort the changes on fail, can be NULL.
1313  * @clean: Clean up resources after all transaction actions have called
1314  *         commit() or abort(). Can be NULL.
1315  *
1316  * Only prepare() may fail. In a single transaction, only one of commit() or
1317  * abort() will be called. clean() will always be called if it is present.
1318  */
1319 typedef struct BlkActionOps {
1320     size_t instance_size;
1321     void (*prepare)(BlkActionState *common, Error **errp);
1322     void (*commit)(BlkActionState *common);
1323     void (*abort)(BlkActionState *common);
1324     void (*clean)(BlkActionState *common);
1325 } BlkActionOps;
1326 
1327 /**
1328  * BlkActionState:
1329  * Describes one Action's state within a Transaction.
1330  *
1331  * @action: QAPI-defined enum identifying which Action to perform.
1332  * @ops: Table of ActionOps this Action can perform.
1333  * @block_job_txn: Transaction which this action belongs to.
1334  * @entry: List membership for all Actions in this Transaction.
1335  *
1336  * This structure must be arranged as first member in a subclassed type,
1337  * assuming that the compiler will also arrange it to the same offsets as the
1338  * base class.
1339  */
1340 struct BlkActionState {
1341     TransactionAction *action;
1342     const BlkActionOps *ops;
1343     JobTxn *block_job_txn;
1344     TransactionProperties *txn_props;
1345     QTAILQ_ENTRY(BlkActionState) entry;
1346 };
1347 
1348 /* internal snapshot private data */
1349 typedef struct InternalSnapshotState {
1350     BlkActionState common;
1351     BlockDriverState *bs;
1352     QEMUSnapshotInfo sn;
1353     bool created;
1354 } InternalSnapshotState;
1355 
1356 
1357 static int action_check_completion_mode(BlkActionState *s, Error **errp)
1358 {
1359     if (s->txn_props->completion_mode != ACTION_COMPLETION_MODE_INDIVIDUAL) {
1360         error_setg(errp,
1361                    "Action '%s' does not support Transaction property "
1362                    "completion-mode = %s",
1363                    TransactionActionKind_str(s->action->type),
1364                    ActionCompletionMode_str(s->txn_props->completion_mode));
1365         return -1;
1366     }
1367     return 0;
1368 }
1369 
1370 static void internal_snapshot_prepare(BlkActionState *common,
1371                                       Error **errp)
1372 {
1373     Error *local_err = NULL;
1374     const char *device;
1375     const char *name;
1376     BlockDriverState *bs;
1377     QEMUSnapshotInfo old_sn, *sn;
1378     bool ret;
1379     qemu_timeval tv;
1380     BlockdevSnapshotInternal *internal;
1381     InternalSnapshotState *state;
1382     AioContext *aio_context;
1383     int ret1;
1384 
1385     g_assert(common->action->type ==
1386              TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC);
1387     internal = common->action->u.blockdev_snapshot_internal_sync.data;
1388     state = DO_UPCAST(InternalSnapshotState, common, common);
1389 
1390     /* 1. parse input */
1391     device = internal->device;
1392     name = internal->name;
1393 
1394     /* 2. check for validation */
1395     if (action_check_completion_mode(common, errp) < 0) {
1396         return;
1397     }
1398 
1399     bs = qmp_get_root_bs(device, errp);
1400     if (!bs) {
1401         return;
1402     }
1403 
1404     aio_context = bdrv_get_aio_context(bs);
1405     aio_context_acquire(aio_context);
1406 
1407     state->bs = bs;
1408 
1409     /* Paired with .clean() */
1410     bdrv_drained_begin(bs);
1411 
1412     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_INTERNAL_SNAPSHOT, errp)) {
1413         goto out;
1414     }
1415 
1416     if (bdrv_is_read_only(bs)) {
1417         error_setg(errp, "Device '%s' is read only", device);
1418         goto out;
1419     }
1420 
1421     if (!bdrv_can_snapshot(bs)) {
1422         error_setg(errp, "Block format '%s' used by device '%s' "
1423                    "does not support internal snapshots",
1424                    bs->drv->format_name, device);
1425         goto out;
1426     }
1427 
1428     if (!strlen(name)) {
1429         error_setg(errp, "Name is empty");
1430         goto out;
1431     }
1432 
1433     /* check whether a snapshot with name exist */
1434     ret = bdrv_snapshot_find_by_id_and_name(bs, NULL, name, &old_sn,
1435                                             &local_err);
1436     if (local_err) {
1437         error_propagate(errp, local_err);
1438         goto out;
1439     } else if (ret) {
1440         error_setg(errp,
1441                    "Snapshot with name '%s' already exists on device '%s'",
1442                    name, device);
1443         goto out;
1444     }
1445 
1446     /* 3. take the snapshot */
1447     sn = &state->sn;
1448     pstrcpy(sn->name, sizeof(sn->name), name);
1449     qemu_gettimeofday(&tv);
1450     sn->date_sec = tv.tv_sec;
1451     sn->date_nsec = tv.tv_usec * 1000;
1452     sn->vm_clock_nsec = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL);
1453 
1454     ret1 = bdrv_snapshot_create(bs, sn);
1455     if (ret1 < 0) {
1456         error_setg_errno(errp, -ret1,
1457                          "Failed to create snapshot '%s' on device '%s'",
1458                          name, device);
1459         goto out;
1460     }
1461 
1462     /* 4. succeed, mark a snapshot is created */
1463     state->created = true;
1464 
1465 out:
1466     aio_context_release(aio_context);
1467 }
1468 
1469 static void internal_snapshot_abort(BlkActionState *common)
1470 {
1471     InternalSnapshotState *state =
1472                              DO_UPCAST(InternalSnapshotState, common, common);
1473     BlockDriverState *bs = state->bs;
1474     QEMUSnapshotInfo *sn = &state->sn;
1475     AioContext *aio_context;
1476     Error *local_error = NULL;
1477 
1478     if (!state->created) {
1479         return;
1480     }
1481 
1482     aio_context = bdrv_get_aio_context(state->bs);
1483     aio_context_acquire(aio_context);
1484 
1485     if (bdrv_snapshot_delete(bs, sn->id_str, sn->name, &local_error) < 0) {
1486         error_reportf_err(local_error,
1487                           "Failed to delete snapshot with id '%s' and "
1488                           "name '%s' on device '%s' in abort: ",
1489                           sn->id_str, sn->name,
1490                           bdrv_get_device_name(bs));
1491     }
1492 
1493     aio_context_release(aio_context);
1494 }
1495 
1496 static void internal_snapshot_clean(BlkActionState *common)
1497 {
1498     InternalSnapshotState *state = DO_UPCAST(InternalSnapshotState,
1499                                              common, common);
1500     AioContext *aio_context;
1501 
1502     if (!state->bs) {
1503         return;
1504     }
1505 
1506     aio_context = bdrv_get_aio_context(state->bs);
1507     aio_context_acquire(aio_context);
1508 
1509     bdrv_drained_end(state->bs);
1510 
1511     aio_context_release(aio_context);
1512 }
1513 
1514 /* external snapshot private data */
1515 typedef struct ExternalSnapshotState {
1516     BlkActionState common;
1517     BlockDriverState *old_bs;
1518     BlockDriverState *new_bs;
1519     bool overlay_appended;
1520 } ExternalSnapshotState;
1521 
1522 static void external_snapshot_prepare(BlkActionState *common,
1523                                       Error **errp)
1524 {
1525     int flags = 0;
1526     QDict *options = NULL;
1527     Error *local_err = NULL;
1528     /* Device and node name of the image to generate the snapshot from */
1529     const char *device;
1530     const char *node_name;
1531     /* Reference to the new image (for 'blockdev-snapshot') */
1532     const char *snapshot_ref;
1533     /* File name of the new image (for 'blockdev-snapshot-sync') */
1534     const char *new_image_file;
1535     ExternalSnapshotState *state =
1536                              DO_UPCAST(ExternalSnapshotState, common, common);
1537     TransactionAction *action = common->action;
1538     AioContext *aio_context;
1539     int ret;
1540 
1541     /* 'blockdev-snapshot' and 'blockdev-snapshot-sync' have similar
1542      * purpose but a different set of parameters */
1543     switch (action->type) {
1544     case TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT:
1545         {
1546             BlockdevSnapshot *s = action->u.blockdev_snapshot.data;
1547             device = s->node;
1548             node_name = s->node;
1549             new_image_file = NULL;
1550             snapshot_ref = s->overlay;
1551         }
1552         break;
1553     case TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC:
1554         {
1555             BlockdevSnapshotSync *s = action->u.blockdev_snapshot_sync.data;
1556             device = s->has_device ? s->device : NULL;
1557             node_name = s->has_node_name ? s->node_name : NULL;
1558             new_image_file = s->snapshot_file;
1559             snapshot_ref = NULL;
1560         }
1561         break;
1562     default:
1563         g_assert_not_reached();
1564     }
1565 
1566     /* start processing */
1567     if (action_check_completion_mode(common, errp) < 0) {
1568         return;
1569     }
1570 
1571     state->old_bs = bdrv_lookup_bs(device, node_name, errp);
1572     if (!state->old_bs) {
1573         return;
1574     }
1575 
1576     aio_context = bdrv_get_aio_context(state->old_bs);
1577     aio_context_acquire(aio_context);
1578 
1579     /* Paired with .clean() */
1580     bdrv_drained_begin(state->old_bs);
1581 
1582     if (!bdrv_is_inserted(state->old_bs)) {
1583         error_setg(errp, QERR_DEVICE_HAS_NO_MEDIUM, device);
1584         goto out;
1585     }
1586 
1587     if (bdrv_op_is_blocked(state->old_bs,
1588                            BLOCK_OP_TYPE_EXTERNAL_SNAPSHOT, errp)) {
1589         goto out;
1590     }
1591 
1592     if (!bdrv_is_read_only(state->old_bs)) {
1593         if (bdrv_flush(state->old_bs)) {
1594             error_setg(errp, QERR_IO_ERROR);
1595             goto out;
1596         }
1597     }
1598 
1599     if (!bdrv_is_first_non_filter(state->old_bs)) {
1600         error_setg(errp, QERR_FEATURE_DISABLED, "snapshot");
1601         goto out;
1602     }
1603 
1604     if (action->type == TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC) {
1605         BlockdevSnapshotSync *s = action->u.blockdev_snapshot_sync.data;
1606         const char *format = s->has_format ? s->format : "qcow2";
1607         enum NewImageMode mode;
1608         const char *snapshot_node_name =
1609             s->has_snapshot_node_name ? s->snapshot_node_name : NULL;
1610 
1611         if (node_name && !snapshot_node_name) {
1612             error_setg(errp, "New overlay node name missing");
1613             goto out;
1614         }
1615 
1616         if (snapshot_node_name &&
1617             bdrv_lookup_bs(snapshot_node_name, snapshot_node_name, NULL)) {
1618             error_setg(errp, "New overlay node name already in use");
1619             goto out;
1620         }
1621 
1622         flags = state->old_bs->open_flags;
1623         flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_COPY_ON_READ);
1624         flags |= BDRV_O_NO_BACKING;
1625 
1626         /* create new image w/backing file */
1627         mode = s->has_mode ? s->mode : NEW_IMAGE_MODE_ABSOLUTE_PATHS;
1628         if (mode != NEW_IMAGE_MODE_EXISTING) {
1629             int64_t size = bdrv_getlength(state->old_bs);
1630             if (size < 0) {
1631                 error_setg_errno(errp, -size, "bdrv_getlength failed");
1632                 goto out;
1633             }
1634             bdrv_refresh_filename(state->old_bs);
1635             bdrv_img_create(new_image_file, format,
1636                             state->old_bs->filename,
1637                             state->old_bs->drv->format_name,
1638                             NULL, size, flags, false, &local_err);
1639             if (local_err) {
1640                 error_propagate(errp, local_err);
1641                 goto out;
1642             }
1643         }
1644 
1645         options = qdict_new();
1646         if (snapshot_node_name) {
1647             qdict_put_str(options, "node-name", snapshot_node_name);
1648         }
1649         qdict_put_str(options, "driver", format);
1650     }
1651 
1652     state->new_bs = bdrv_open(new_image_file, snapshot_ref, options, flags,
1653                               errp);
1654     /* We will manually add the backing_hd field to the bs later */
1655     if (!state->new_bs) {
1656         goto out;
1657     }
1658 
1659     if (bdrv_has_blk(state->new_bs)) {
1660         error_setg(errp, "The overlay is already in use");
1661         goto out;
1662     }
1663 
1664     if (bdrv_op_is_blocked(state->new_bs, BLOCK_OP_TYPE_EXTERNAL_SNAPSHOT,
1665                            errp)) {
1666         goto out;
1667     }
1668 
1669     if (state->new_bs->backing != NULL) {
1670         error_setg(errp, "The overlay already has a backing image");
1671         goto out;
1672     }
1673 
1674     if (!state->new_bs->drv->supports_backing) {
1675         error_setg(errp, "The overlay does not support backing images");
1676         goto out;
1677     }
1678 
1679     ret = bdrv_try_set_aio_context(state->new_bs, aio_context, errp);
1680     if (ret < 0) {
1681         goto out;
1682     }
1683 
1684     /* This removes our old bs and adds the new bs. This is an operation that
1685      * can fail, so we need to do it in .prepare; undoing it for abort is
1686      * always possible. */
1687     bdrv_ref(state->new_bs);
1688     bdrv_append(state->new_bs, state->old_bs, &local_err);
1689     if (local_err) {
1690         error_propagate(errp, local_err);
1691         goto out;
1692     }
1693     state->overlay_appended = true;
1694 
1695 out:
1696     aio_context_release(aio_context);
1697 }
1698 
1699 static void external_snapshot_commit(BlkActionState *common)
1700 {
1701     ExternalSnapshotState *state =
1702                              DO_UPCAST(ExternalSnapshotState, common, common);
1703     AioContext *aio_context;
1704 
1705     aio_context = bdrv_get_aio_context(state->old_bs);
1706     aio_context_acquire(aio_context);
1707 
1708     /* We don't need (or want) to use the transactional
1709      * bdrv_reopen_multiple() across all the entries at once, because we
1710      * don't want to abort all of them if one of them fails the reopen */
1711     if (!atomic_read(&state->old_bs->copy_on_read)) {
1712         bdrv_reopen_set_read_only(state->old_bs, true, NULL);
1713     }
1714 
1715     aio_context_release(aio_context);
1716 }
1717 
1718 static void external_snapshot_abort(BlkActionState *common)
1719 {
1720     ExternalSnapshotState *state =
1721                              DO_UPCAST(ExternalSnapshotState, common, common);
1722     if (state->new_bs) {
1723         if (state->overlay_appended) {
1724             AioContext *aio_context;
1725 
1726             aio_context = bdrv_get_aio_context(state->old_bs);
1727             aio_context_acquire(aio_context);
1728 
1729             bdrv_ref(state->old_bs);   /* we can't let bdrv_set_backind_hd()
1730                                           close state->old_bs; we need it */
1731             bdrv_set_backing_hd(state->new_bs, NULL, &error_abort);
1732             bdrv_replace_node(state->new_bs, state->old_bs, &error_abort);
1733             bdrv_unref(state->old_bs); /* bdrv_replace_node() ref'ed old_bs */
1734 
1735             aio_context_release(aio_context);
1736         }
1737     }
1738 }
1739 
1740 static void external_snapshot_clean(BlkActionState *common)
1741 {
1742     ExternalSnapshotState *state =
1743                              DO_UPCAST(ExternalSnapshotState, common, common);
1744     AioContext *aio_context;
1745 
1746     if (!state->old_bs) {
1747         return;
1748     }
1749 
1750     aio_context = bdrv_get_aio_context(state->old_bs);
1751     aio_context_acquire(aio_context);
1752 
1753     bdrv_drained_end(state->old_bs);
1754     bdrv_unref(state->new_bs);
1755 
1756     aio_context_release(aio_context);
1757 }
1758 
1759 typedef struct DriveBackupState {
1760     BlkActionState common;
1761     BlockDriverState *bs;
1762     BlockJob *job;
1763 } DriveBackupState;
1764 
1765 static BlockJob *do_drive_backup(DriveBackup *backup, JobTxn *txn,
1766                             Error **errp);
1767 
1768 static void drive_backup_prepare(BlkActionState *common, Error **errp)
1769 {
1770     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1771     BlockDriverState *bs;
1772     DriveBackup *backup;
1773     AioContext *aio_context;
1774     Error *local_err = NULL;
1775 
1776     assert(common->action->type == TRANSACTION_ACTION_KIND_DRIVE_BACKUP);
1777     backup = common->action->u.drive_backup.data;
1778 
1779     bs = bdrv_lookup_bs(backup->device, backup->device, errp);
1780     if (!bs) {
1781         return;
1782     }
1783 
1784     aio_context = bdrv_get_aio_context(bs);
1785     aio_context_acquire(aio_context);
1786 
1787     /* Paired with .clean() */
1788     bdrv_drained_begin(bs);
1789 
1790     state->bs = bs;
1791 
1792     state->job = do_drive_backup(backup, common->block_job_txn, &local_err);
1793     if (local_err) {
1794         error_propagate(errp, local_err);
1795         goto out;
1796     }
1797 
1798 out:
1799     aio_context_release(aio_context);
1800 }
1801 
1802 static void drive_backup_commit(BlkActionState *common)
1803 {
1804     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1805     AioContext *aio_context;
1806 
1807     aio_context = bdrv_get_aio_context(state->bs);
1808     aio_context_acquire(aio_context);
1809 
1810     assert(state->job);
1811     job_start(&state->job->job);
1812 
1813     aio_context_release(aio_context);
1814 }
1815 
1816 static void drive_backup_abort(BlkActionState *common)
1817 {
1818     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1819 
1820     if (state->job) {
1821         AioContext *aio_context;
1822 
1823         aio_context = bdrv_get_aio_context(state->bs);
1824         aio_context_acquire(aio_context);
1825 
1826         job_cancel_sync(&state->job->job);
1827 
1828         aio_context_release(aio_context);
1829     }
1830 }
1831 
1832 static void drive_backup_clean(BlkActionState *common)
1833 {
1834     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1835     AioContext *aio_context;
1836 
1837     if (!state->bs) {
1838         return;
1839     }
1840 
1841     aio_context = bdrv_get_aio_context(state->bs);
1842     aio_context_acquire(aio_context);
1843 
1844     bdrv_drained_end(state->bs);
1845 
1846     aio_context_release(aio_context);
1847 }
1848 
1849 typedef struct BlockdevBackupState {
1850     BlkActionState common;
1851     BlockDriverState *bs;
1852     BlockJob *job;
1853 } BlockdevBackupState;
1854 
1855 static BlockJob *do_blockdev_backup(BlockdevBackup *backup, JobTxn *txn,
1856                                     Error **errp);
1857 
1858 static void blockdev_backup_prepare(BlkActionState *common, Error **errp)
1859 {
1860     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1861     BlockdevBackup *backup;
1862     BlockDriverState *bs, *target;
1863     AioContext *aio_context;
1864     Error *local_err = NULL;
1865 
1866     assert(common->action->type == TRANSACTION_ACTION_KIND_BLOCKDEV_BACKUP);
1867     backup = common->action->u.blockdev_backup.data;
1868 
1869     bs = bdrv_lookup_bs(backup->device, backup->device, errp);
1870     if (!bs) {
1871         return;
1872     }
1873 
1874     target = bdrv_lookup_bs(backup->target, backup->target, errp);
1875     if (!target) {
1876         return;
1877     }
1878 
1879     aio_context = bdrv_get_aio_context(bs);
1880     aio_context_acquire(aio_context);
1881     state->bs = bs;
1882 
1883     /* Paired with .clean() */
1884     bdrv_drained_begin(state->bs);
1885 
1886     state->job = do_blockdev_backup(backup, common->block_job_txn, &local_err);
1887     if (local_err) {
1888         error_propagate(errp, local_err);
1889         goto out;
1890     }
1891 
1892 out:
1893     aio_context_release(aio_context);
1894 }
1895 
1896 static void blockdev_backup_commit(BlkActionState *common)
1897 {
1898     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1899     AioContext *aio_context;
1900 
1901     aio_context = bdrv_get_aio_context(state->bs);
1902     aio_context_acquire(aio_context);
1903 
1904     assert(state->job);
1905     job_start(&state->job->job);
1906 
1907     aio_context_release(aio_context);
1908 }
1909 
1910 static void blockdev_backup_abort(BlkActionState *common)
1911 {
1912     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1913 
1914     if (state->job) {
1915         AioContext *aio_context;
1916 
1917         aio_context = bdrv_get_aio_context(state->bs);
1918         aio_context_acquire(aio_context);
1919 
1920         job_cancel_sync(&state->job->job);
1921 
1922         aio_context_release(aio_context);
1923     }
1924 }
1925 
1926 static void blockdev_backup_clean(BlkActionState *common)
1927 {
1928     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1929     AioContext *aio_context;
1930 
1931     if (!state->bs) {
1932         return;
1933     }
1934 
1935     aio_context = bdrv_get_aio_context(state->bs);
1936     aio_context_acquire(aio_context);
1937 
1938     bdrv_drained_end(state->bs);
1939 
1940     aio_context_release(aio_context);
1941 }
1942 
1943 typedef struct BlockDirtyBitmapState {
1944     BlkActionState common;
1945     BdrvDirtyBitmap *bitmap;
1946     BlockDriverState *bs;
1947     HBitmap *backup;
1948     bool prepared;
1949     bool was_enabled;
1950 } BlockDirtyBitmapState;
1951 
1952 static void block_dirty_bitmap_add_prepare(BlkActionState *common,
1953                                            Error **errp)
1954 {
1955     Error *local_err = NULL;
1956     BlockDirtyBitmapAdd *action;
1957     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
1958                                              common, common);
1959 
1960     if (action_check_completion_mode(common, errp) < 0) {
1961         return;
1962     }
1963 
1964     action = common->action->u.block_dirty_bitmap_add.data;
1965     /* AIO context taken and released within qmp_block_dirty_bitmap_add */
1966     qmp_block_dirty_bitmap_add(action->node, action->name,
1967                                action->has_granularity, action->granularity,
1968                                action->has_persistent, action->persistent,
1969                                action->has_disabled, action->disabled,
1970                                &local_err);
1971 
1972     if (!local_err) {
1973         state->prepared = true;
1974     } else {
1975         error_propagate(errp, local_err);
1976     }
1977 }
1978 
1979 static void block_dirty_bitmap_add_abort(BlkActionState *common)
1980 {
1981     BlockDirtyBitmapAdd *action;
1982     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
1983                                              common, common);
1984 
1985     action = common->action->u.block_dirty_bitmap_add.data;
1986     /* Should not be able to fail: IF the bitmap was added via .prepare(),
1987      * then the node reference and bitmap name must have been valid.
1988      */
1989     if (state->prepared) {
1990         qmp_block_dirty_bitmap_remove(action->node, action->name, &error_abort);
1991     }
1992 }
1993 
1994 static void block_dirty_bitmap_clear_prepare(BlkActionState *common,
1995                                              Error **errp)
1996 {
1997     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
1998                                              common, common);
1999     BlockDirtyBitmap *action;
2000 
2001     if (action_check_completion_mode(common, errp) < 0) {
2002         return;
2003     }
2004 
2005     action = common->action->u.block_dirty_bitmap_clear.data;
2006     state->bitmap = block_dirty_bitmap_lookup(action->node,
2007                                               action->name,
2008                                               &state->bs,
2009                                               errp);
2010     if (!state->bitmap) {
2011         return;
2012     }
2013 
2014     if (bdrv_dirty_bitmap_check(state->bitmap, BDRV_BITMAP_DEFAULT, errp)) {
2015         return;
2016     }
2017 
2018     bdrv_clear_dirty_bitmap(state->bitmap, &state->backup);
2019 }
2020 
2021 static void block_dirty_bitmap_restore(BlkActionState *common)
2022 {
2023     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2024                                              common, common);
2025 
2026     if (state->backup) {
2027         bdrv_restore_dirty_bitmap(state->bitmap, state->backup);
2028     }
2029 }
2030 
2031 static void block_dirty_bitmap_free_backup(BlkActionState *common)
2032 {
2033     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2034                                              common, common);
2035 
2036     hbitmap_free(state->backup);
2037 }
2038 
2039 static void block_dirty_bitmap_enable_prepare(BlkActionState *common,
2040                                               Error **errp)
2041 {
2042     BlockDirtyBitmap *action;
2043     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2044                                              common, common);
2045 
2046     if (action_check_completion_mode(common, errp) < 0) {
2047         return;
2048     }
2049 
2050     action = common->action->u.block_dirty_bitmap_enable.data;
2051     state->bitmap = block_dirty_bitmap_lookup(action->node,
2052                                               action->name,
2053                                               NULL,
2054                                               errp);
2055     if (!state->bitmap) {
2056         return;
2057     }
2058 
2059     if (bdrv_dirty_bitmap_check(state->bitmap, BDRV_BITMAP_ALLOW_RO, errp)) {
2060         return;
2061     }
2062 
2063     state->was_enabled = bdrv_dirty_bitmap_enabled(state->bitmap);
2064     bdrv_enable_dirty_bitmap(state->bitmap);
2065 }
2066 
2067 static void block_dirty_bitmap_enable_abort(BlkActionState *common)
2068 {
2069     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2070                                              common, common);
2071 
2072     if (!state->was_enabled) {
2073         bdrv_disable_dirty_bitmap(state->bitmap);
2074     }
2075 }
2076 
2077 static void block_dirty_bitmap_disable_prepare(BlkActionState *common,
2078                                                Error **errp)
2079 {
2080     BlockDirtyBitmap *action;
2081     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2082                                              common, common);
2083 
2084     if (action_check_completion_mode(common, errp) < 0) {
2085         return;
2086     }
2087 
2088     action = common->action->u.block_dirty_bitmap_disable.data;
2089     state->bitmap = block_dirty_bitmap_lookup(action->node,
2090                                               action->name,
2091                                               NULL,
2092                                               errp);
2093     if (!state->bitmap) {
2094         return;
2095     }
2096 
2097     if (bdrv_dirty_bitmap_check(state->bitmap, BDRV_BITMAP_ALLOW_RO, errp)) {
2098         return;
2099     }
2100 
2101     state->was_enabled = bdrv_dirty_bitmap_enabled(state->bitmap);
2102     bdrv_disable_dirty_bitmap(state->bitmap);
2103 }
2104 
2105 static void block_dirty_bitmap_disable_abort(BlkActionState *common)
2106 {
2107     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2108                                              common, common);
2109 
2110     if (state->was_enabled) {
2111         bdrv_enable_dirty_bitmap(state->bitmap);
2112     }
2113 }
2114 
2115 static BdrvDirtyBitmap *do_block_dirty_bitmap_merge(
2116         const char *node, const char *target,
2117         BlockDirtyBitmapMergeSourceList *bitmaps,
2118         HBitmap **backup, Error **errp);
2119 
2120 static void block_dirty_bitmap_merge_prepare(BlkActionState *common,
2121                                              Error **errp)
2122 {
2123     BlockDirtyBitmapMerge *action;
2124     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2125                                              common, common);
2126 
2127     if (action_check_completion_mode(common, errp) < 0) {
2128         return;
2129     }
2130 
2131     action = common->action->u.block_dirty_bitmap_merge.data;
2132 
2133     state->bitmap = do_block_dirty_bitmap_merge(action->node, action->target,
2134                                                 action->bitmaps, &state->backup,
2135                                                 errp);
2136 }
2137 
2138 static BdrvDirtyBitmap *do_block_dirty_bitmap_remove(
2139         const char *node, const char *name, bool release,
2140         BlockDriverState **bitmap_bs, Error **errp);
2141 
2142 static void block_dirty_bitmap_remove_prepare(BlkActionState *common,
2143                                               Error **errp)
2144 {
2145     BlockDirtyBitmap *action;
2146     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2147                                              common, common);
2148 
2149     if (action_check_completion_mode(common, errp) < 0) {
2150         return;
2151     }
2152 
2153     action = common->action->u.block_dirty_bitmap_remove.data;
2154 
2155     state->bitmap = do_block_dirty_bitmap_remove(action->node, action->name,
2156                                                  false, &state->bs, errp);
2157     if (state->bitmap) {
2158         bdrv_dirty_bitmap_skip_store(state->bitmap, true);
2159         bdrv_dirty_bitmap_set_busy(state->bitmap, true);
2160     }
2161 }
2162 
2163 static void block_dirty_bitmap_remove_abort(BlkActionState *common)
2164 {
2165     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2166                                              common, common);
2167 
2168     if (state->bitmap) {
2169         bdrv_dirty_bitmap_skip_store(state->bitmap, false);
2170         bdrv_dirty_bitmap_set_busy(state->bitmap, false);
2171     }
2172 }
2173 
2174 static void block_dirty_bitmap_remove_commit(BlkActionState *common)
2175 {
2176     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2177                                              common, common);
2178 
2179     bdrv_dirty_bitmap_set_busy(state->bitmap, false);
2180     bdrv_release_dirty_bitmap(state->bitmap);
2181 }
2182 
2183 static void abort_prepare(BlkActionState *common, Error **errp)
2184 {
2185     error_setg(errp, "Transaction aborted using Abort action");
2186 }
2187 
2188 static void abort_commit(BlkActionState *common)
2189 {
2190     g_assert_not_reached(); /* this action never succeeds */
2191 }
2192 
2193 static const BlkActionOps actions[] = {
2194     [TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT] = {
2195         .instance_size = sizeof(ExternalSnapshotState),
2196         .prepare  = external_snapshot_prepare,
2197         .commit   = external_snapshot_commit,
2198         .abort = external_snapshot_abort,
2199         .clean = external_snapshot_clean,
2200     },
2201     [TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC] = {
2202         .instance_size = sizeof(ExternalSnapshotState),
2203         .prepare  = external_snapshot_prepare,
2204         .commit   = external_snapshot_commit,
2205         .abort = external_snapshot_abort,
2206         .clean = external_snapshot_clean,
2207     },
2208     [TRANSACTION_ACTION_KIND_DRIVE_BACKUP] = {
2209         .instance_size = sizeof(DriveBackupState),
2210         .prepare = drive_backup_prepare,
2211         .commit = drive_backup_commit,
2212         .abort = drive_backup_abort,
2213         .clean = drive_backup_clean,
2214     },
2215     [TRANSACTION_ACTION_KIND_BLOCKDEV_BACKUP] = {
2216         .instance_size = sizeof(BlockdevBackupState),
2217         .prepare = blockdev_backup_prepare,
2218         .commit = blockdev_backup_commit,
2219         .abort = blockdev_backup_abort,
2220         .clean = blockdev_backup_clean,
2221     },
2222     [TRANSACTION_ACTION_KIND_ABORT] = {
2223         .instance_size = sizeof(BlkActionState),
2224         .prepare = abort_prepare,
2225         .commit = abort_commit,
2226     },
2227     [TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC] = {
2228         .instance_size = sizeof(InternalSnapshotState),
2229         .prepare  = internal_snapshot_prepare,
2230         .abort = internal_snapshot_abort,
2231         .clean = internal_snapshot_clean,
2232     },
2233     [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_ADD] = {
2234         .instance_size = sizeof(BlockDirtyBitmapState),
2235         .prepare = block_dirty_bitmap_add_prepare,
2236         .abort = block_dirty_bitmap_add_abort,
2237     },
2238     [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_CLEAR] = {
2239         .instance_size = sizeof(BlockDirtyBitmapState),
2240         .prepare = block_dirty_bitmap_clear_prepare,
2241         .commit = block_dirty_bitmap_free_backup,
2242         .abort = block_dirty_bitmap_restore,
2243     },
2244     [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_ENABLE] = {
2245         .instance_size = sizeof(BlockDirtyBitmapState),
2246         .prepare = block_dirty_bitmap_enable_prepare,
2247         .abort = block_dirty_bitmap_enable_abort,
2248     },
2249     [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_DISABLE] = {
2250         .instance_size = sizeof(BlockDirtyBitmapState),
2251         .prepare = block_dirty_bitmap_disable_prepare,
2252         .abort = block_dirty_bitmap_disable_abort,
2253     },
2254     [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_MERGE] = {
2255         .instance_size = sizeof(BlockDirtyBitmapState),
2256         .prepare = block_dirty_bitmap_merge_prepare,
2257         .commit = block_dirty_bitmap_free_backup,
2258         .abort = block_dirty_bitmap_restore,
2259     },
2260     [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_REMOVE] = {
2261         .instance_size = sizeof(BlockDirtyBitmapState),
2262         .prepare = block_dirty_bitmap_remove_prepare,
2263         .commit = block_dirty_bitmap_remove_commit,
2264         .abort = block_dirty_bitmap_remove_abort,
2265     },
2266     /* Where are transactions for MIRROR, COMMIT and STREAM?
2267      * Although these blockjobs use transaction callbacks like the backup job,
2268      * these jobs do not necessarily adhere to transaction semantics.
2269      * These jobs may not fully undo all of their actions on abort, nor do they
2270      * necessarily work in transactions with more than one job in them.
2271      */
2272 };
2273 
2274 /**
2275  * Allocate a TransactionProperties structure if necessary, and fill
2276  * that structure with desired defaults if they are unset.
2277  */
2278 static TransactionProperties *get_transaction_properties(
2279     TransactionProperties *props)
2280 {
2281     if (!props) {
2282         props = g_new0(TransactionProperties, 1);
2283     }
2284 
2285     if (!props->has_completion_mode) {
2286         props->has_completion_mode = true;
2287         props->completion_mode = ACTION_COMPLETION_MODE_INDIVIDUAL;
2288     }
2289 
2290     return props;
2291 }
2292 
2293 /*
2294  * 'Atomic' group operations.  The operations are performed as a set, and if
2295  * any fail then we roll back all operations in the group.
2296  */
2297 void qmp_transaction(TransactionActionList *dev_list,
2298                      bool has_props,
2299                      struct TransactionProperties *props,
2300                      Error **errp)
2301 {
2302     TransactionActionList *dev_entry = dev_list;
2303     JobTxn *block_job_txn = NULL;
2304     BlkActionState *state, *next;
2305     Error *local_err = NULL;
2306 
2307     QTAILQ_HEAD(, BlkActionState) snap_bdrv_states;
2308     QTAILQ_INIT(&snap_bdrv_states);
2309 
2310     /* Does this transaction get canceled as a group on failure?
2311      * If not, we don't really need to make a JobTxn.
2312      */
2313     props = get_transaction_properties(props);
2314     if (props->completion_mode != ACTION_COMPLETION_MODE_INDIVIDUAL) {
2315         block_job_txn = job_txn_new();
2316     }
2317 
2318     /* drain all i/o before any operations */
2319     bdrv_drain_all();
2320 
2321     /* We don't do anything in this loop that commits us to the operations */
2322     while (NULL != dev_entry) {
2323         TransactionAction *dev_info = NULL;
2324         const BlkActionOps *ops;
2325 
2326         dev_info = dev_entry->value;
2327         dev_entry = dev_entry->next;
2328 
2329         assert(dev_info->type < ARRAY_SIZE(actions));
2330 
2331         ops = &actions[dev_info->type];
2332         assert(ops->instance_size > 0);
2333 
2334         state = g_malloc0(ops->instance_size);
2335         state->ops = ops;
2336         state->action = dev_info;
2337         state->block_job_txn = block_job_txn;
2338         state->txn_props = props;
2339         QTAILQ_INSERT_TAIL(&snap_bdrv_states, state, entry);
2340 
2341         state->ops->prepare(state, &local_err);
2342         if (local_err) {
2343             error_propagate(errp, local_err);
2344             goto delete_and_fail;
2345         }
2346     }
2347 
2348     QTAILQ_FOREACH(state, &snap_bdrv_states, entry) {
2349         if (state->ops->commit) {
2350             state->ops->commit(state);
2351         }
2352     }
2353 
2354     /* success */
2355     goto exit;
2356 
2357 delete_and_fail:
2358     /* failure, and it is all-or-none; roll back all operations */
2359     QTAILQ_FOREACH_REVERSE(state, &snap_bdrv_states, entry) {
2360         if (state->ops->abort) {
2361             state->ops->abort(state);
2362         }
2363     }
2364 exit:
2365     QTAILQ_FOREACH_SAFE(state, &snap_bdrv_states, entry, next) {
2366         if (state->ops->clean) {
2367             state->ops->clean(state);
2368         }
2369         g_free(state);
2370     }
2371     if (!has_props) {
2372         qapi_free_TransactionProperties(props);
2373     }
2374     job_txn_unref(block_job_txn);
2375 }
2376 
2377 void qmp_eject(bool has_device, const char *device,
2378                bool has_id, const char *id,
2379                bool has_force, bool force, Error **errp)
2380 {
2381     Error *local_err = NULL;
2382     int rc;
2383 
2384     if (!has_force) {
2385         force = false;
2386     }
2387 
2388     rc = do_open_tray(has_device ? device : NULL,
2389                       has_id ? id : NULL,
2390                       force, &local_err);
2391     if (rc && rc != -ENOSYS) {
2392         error_propagate(errp, local_err);
2393         return;
2394     }
2395     error_free(local_err);
2396 
2397     blockdev_remove_medium(has_device, device, has_id, id, errp);
2398 }
2399 
2400 void qmp_block_passwd(bool has_device, const char *device,
2401                       bool has_node_name, const char *node_name,
2402                       const char *password, Error **errp)
2403 {
2404     error_setg(errp,
2405                "Setting block passwords directly is no longer supported");
2406 }
2407 
2408 /*
2409  * Attempt to open the tray of @device.
2410  * If @force, ignore its tray lock.
2411  * Else, if the tray is locked, don't open it, but ask the guest to open it.
2412  * On error, store an error through @errp and return -errno.
2413  * If @device does not exist, return -ENODEV.
2414  * If it has no removable media, return -ENOTSUP.
2415  * If it has no tray, return -ENOSYS.
2416  * If the guest was asked to open the tray, return -EINPROGRESS.
2417  * Else, return 0.
2418  */
2419 static int do_open_tray(const char *blk_name, const char *qdev_id,
2420                         bool force, Error **errp)
2421 {
2422     BlockBackend *blk;
2423     const char *device = qdev_id ?: blk_name;
2424     bool locked;
2425 
2426     blk = qmp_get_blk(blk_name, qdev_id, errp);
2427     if (!blk) {
2428         return -ENODEV;
2429     }
2430 
2431     if (!blk_dev_has_removable_media(blk)) {
2432         error_setg(errp, "Device '%s' is not removable", device);
2433         return -ENOTSUP;
2434     }
2435 
2436     if (!blk_dev_has_tray(blk)) {
2437         error_setg(errp, "Device '%s' does not have a tray", device);
2438         return -ENOSYS;
2439     }
2440 
2441     if (blk_dev_is_tray_open(blk)) {
2442         return 0;
2443     }
2444 
2445     locked = blk_dev_is_medium_locked(blk);
2446     if (locked) {
2447         blk_dev_eject_request(blk, force);
2448     }
2449 
2450     if (!locked || force) {
2451         blk_dev_change_media_cb(blk, false, &error_abort);
2452     }
2453 
2454     if (locked && !force) {
2455         error_setg(errp, "Device '%s' is locked and force was not specified, "
2456                    "wait for tray to open and try again", device);
2457         return -EINPROGRESS;
2458     }
2459 
2460     return 0;
2461 }
2462 
2463 void qmp_blockdev_open_tray(bool has_device, const char *device,
2464                             bool has_id, const char *id,
2465                             bool has_force, bool force,
2466                             Error **errp)
2467 {
2468     Error *local_err = NULL;
2469     int rc;
2470 
2471     if (!has_force) {
2472         force = false;
2473     }
2474     rc = do_open_tray(has_device ? device : NULL,
2475                       has_id ? id : NULL,
2476                       force, &local_err);
2477     if (rc && rc != -ENOSYS && rc != -EINPROGRESS) {
2478         error_propagate(errp, local_err);
2479         return;
2480     }
2481     error_free(local_err);
2482 }
2483 
2484 void qmp_blockdev_close_tray(bool has_device, const char *device,
2485                              bool has_id, const char *id,
2486                              Error **errp)
2487 {
2488     BlockBackend *blk;
2489     Error *local_err = NULL;
2490 
2491     device = has_device ? device : NULL;
2492     id = has_id ? id : NULL;
2493 
2494     blk = qmp_get_blk(device, id, errp);
2495     if (!blk) {
2496         return;
2497     }
2498 
2499     if (!blk_dev_has_removable_media(blk)) {
2500         error_setg(errp, "Device '%s' is not removable", device ?: id);
2501         return;
2502     }
2503 
2504     if (!blk_dev_has_tray(blk)) {
2505         /* Ignore this command on tray-less devices */
2506         return;
2507     }
2508 
2509     if (!blk_dev_is_tray_open(blk)) {
2510         return;
2511     }
2512 
2513     blk_dev_change_media_cb(blk, true, &local_err);
2514     if (local_err) {
2515         error_propagate(errp, local_err);
2516         return;
2517     }
2518 }
2519 
2520 static void blockdev_remove_medium(bool has_device, const char *device,
2521                                    bool has_id, const char *id, Error **errp)
2522 {
2523     BlockBackend *blk;
2524     BlockDriverState *bs;
2525     AioContext *aio_context;
2526     bool has_attached_device;
2527 
2528     device = has_device ? device : NULL;
2529     id = has_id ? id : NULL;
2530 
2531     blk = qmp_get_blk(device, id, errp);
2532     if (!blk) {
2533         return;
2534     }
2535 
2536     /* For BBs without a device, we can exchange the BDS tree at will */
2537     has_attached_device = blk_get_attached_dev(blk);
2538 
2539     if (has_attached_device && !blk_dev_has_removable_media(blk)) {
2540         error_setg(errp, "Device '%s' is not removable", device ?: id);
2541         return;
2542     }
2543 
2544     if (has_attached_device && blk_dev_has_tray(blk) &&
2545         !blk_dev_is_tray_open(blk))
2546     {
2547         error_setg(errp, "Tray of device '%s' is not open", device ?: id);
2548         return;
2549     }
2550 
2551     bs = blk_bs(blk);
2552     if (!bs) {
2553         return;
2554     }
2555 
2556     aio_context = bdrv_get_aio_context(bs);
2557     aio_context_acquire(aio_context);
2558 
2559     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_EJECT, errp)) {
2560         goto out;
2561     }
2562 
2563     blk_remove_bs(blk);
2564 
2565     if (!blk_dev_has_tray(blk)) {
2566         /* For tray-less devices, blockdev-open-tray is a no-op (or may not be
2567          * called at all); therefore, the medium needs to be ejected here.
2568          * Do it after blk_remove_bs() so blk_is_inserted(blk) returns the @load
2569          * value passed here (i.e. false). */
2570         blk_dev_change_media_cb(blk, false, &error_abort);
2571     }
2572 
2573 out:
2574     aio_context_release(aio_context);
2575 }
2576 
2577 void qmp_blockdev_remove_medium(const char *id, Error **errp)
2578 {
2579     blockdev_remove_medium(false, NULL, true, id, errp);
2580 }
2581 
2582 static void qmp_blockdev_insert_anon_medium(BlockBackend *blk,
2583                                             BlockDriverState *bs, Error **errp)
2584 {
2585     Error *local_err = NULL;
2586     bool has_device;
2587     int ret;
2588 
2589     /* For BBs without a device, we can exchange the BDS tree at will */
2590     has_device = blk_get_attached_dev(blk);
2591 
2592     if (has_device && !blk_dev_has_removable_media(blk)) {
2593         error_setg(errp, "Device is not removable");
2594         return;
2595     }
2596 
2597     if (has_device && blk_dev_has_tray(blk) && !blk_dev_is_tray_open(blk)) {
2598         error_setg(errp, "Tray of the device is not open");
2599         return;
2600     }
2601 
2602     if (blk_bs(blk)) {
2603         error_setg(errp, "There already is a medium in the device");
2604         return;
2605     }
2606 
2607     ret = blk_insert_bs(blk, bs, errp);
2608     if (ret < 0) {
2609         return;
2610     }
2611 
2612     if (!blk_dev_has_tray(blk)) {
2613         /* For tray-less devices, blockdev-close-tray is a no-op (or may not be
2614          * called at all); therefore, the medium needs to be pushed into the
2615          * slot here.
2616          * Do it after blk_insert_bs() so blk_is_inserted(blk) returns the @load
2617          * value passed here (i.e. true). */
2618         blk_dev_change_media_cb(blk, true, &local_err);
2619         if (local_err) {
2620             error_propagate(errp, local_err);
2621             blk_remove_bs(blk);
2622             return;
2623         }
2624     }
2625 }
2626 
2627 static void blockdev_insert_medium(bool has_device, const char *device,
2628                                    bool has_id, const char *id,
2629                                    const char *node_name, Error **errp)
2630 {
2631     BlockBackend *blk;
2632     BlockDriverState *bs;
2633 
2634     blk = qmp_get_blk(has_device ? device : NULL,
2635                       has_id ? id : NULL,
2636                       errp);
2637     if (!blk) {
2638         return;
2639     }
2640 
2641     bs = bdrv_find_node(node_name);
2642     if (!bs) {
2643         error_setg(errp, "Node '%s' not found", node_name);
2644         return;
2645     }
2646 
2647     if (bdrv_has_blk(bs)) {
2648         error_setg(errp, "Node '%s' is already in use", node_name);
2649         return;
2650     }
2651 
2652     qmp_blockdev_insert_anon_medium(blk, bs, errp);
2653 }
2654 
2655 void qmp_blockdev_insert_medium(const char *id, const char *node_name,
2656                                 Error **errp)
2657 {
2658     blockdev_insert_medium(false, NULL, true, id, node_name, errp);
2659 }
2660 
2661 void qmp_blockdev_change_medium(bool has_device, const char *device,
2662                                 bool has_id, const char *id,
2663                                 const char *filename,
2664                                 bool has_format, const char *format,
2665                                 bool has_read_only,
2666                                 BlockdevChangeReadOnlyMode read_only,
2667                                 Error **errp)
2668 {
2669     BlockBackend *blk;
2670     BlockDriverState *medium_bs = NULL;
2671     int bdrv_flags;
2672     bool detect_zeroes;
2673     int rc;
2674     QDict *options = NULL;
2675     Error *err = NULL;
2676 
2677     blk = qmp_get_blk(has_device ? device : NULL,
2678                       has_id ? id : NULL,
2679                       errp);
2680     if (!blk) {
2681         goto fail;
2682     }
2683 
2684     if (blk_bs(blk)) {
2685         blk_update_root_state(blk);
2686     }
2687 
2688     bdrv_flags = blk_get_open_flags_from_root_state(blk);
2689     bdrv_flags &= ~(BDRV_O_TEMPORARY | BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING |
2690         BDRV_O_PROTOCOL | BDRV_O_AUTO_RDONLY);
2691 
2692     if (!has_read_only) {
2693         read_only = BLOCKDEV_CHANGE_READ_ONLY_MODE_RETAIN;
2694     }
2695 
2696     switch (read_only) {
2697     case BLOCKDEV_CHANGE_READ_ONLY_MODE_RETAIN:
2698         break;
2699 
2700     case BLOCKDEV_CHANGE_READ_ONLY_MODE_READ_ONLY:
2701         bdrv_flags &= ~BDRV_O_RDWR;
2702         break;
2703 
2704     case BLOCKDEV_CHANGE_READ_ONLY_MODE_READ_WRITE:
2705         bdrv_flags |= BDRV_O_RDWR;
2706         break;
2707 
2708     default:
2709         abort();
2710     }
2711 
2712     options = qdict_new();
2713     detect_zeroes = blk_get_detect_zeroes_from_root_state(blk);
2714     qdict_put_str(options, "detect-zeroes", detect_zeroes ? "on" : "off");
2715 
2716     if (has_format) {
2717         qdict_put_str(options, "driver", format);
2718     }
2719 
2720     medium_bs = bdrv_open(filename, NULL, options, bdrv_flags, errp);
2721     if (!medium_bs) {
2722         goto fail;
2723     }
2724 
2725     rc = do_open_tray(has_device ? device : NULL,
2726                       has_id ? id : NULL,
2727                       false, &err);
2728     if (rc && rc != -ENOSYS) {
2729         error_propagate(errp, err);
2730         goto fail;
2731     }
2732     error_free(err);
2733     err = NULL;
2734 
2735     blockdev_remove_medium(has_device, device, has_id, id, &err);
2736     if (err) {
2737         error_propagate(errp, err);
2738         goto fail;
2739     }
2740 
2741     qmp_blockdev_insert_anon_medium(blk, medium_bs, &err);
2742     if (err) {
2743         error_propagate(errp, err);
2744         goto fail;
2745     }
2746 
2747     qmp_blockdev_close_tray(has_device, device, has_id, id, errp);
2748 
2749 fail:
2750     /* If the medium has been inserted, the device has its own reference, so
2751      * ours must be relinquished; and if it has not been inserted successfully,
2752      * the reference must be relinquished anyway */
2753     bdrv_unref(medium_bs);
2754 }
2755 
2756 /* throttling disk I/O limits */
2757 void qmp_block_set_io_throttle(BlockIOThrottle *arg, Error **errp)
2758 {
2759     ThrottleConfig cfg;
2760     BlockDriverState *bs;
2761     BlockBackend *blk;
2762     AioContext *aio_context;
2763 
2764     blk = qmp_get_blk(arg->has_device ? arg->device : NULL,
2765                       arg->has_id ? arg->id : NULL,
2766                       errp);
2767     if (!blk) {
2768         return;
2769     }
2770 
2771     aio_context = blk_get_aio_context(blk);
2772     aio_context_acquire(aio_context);
2773 
2774     bs = blk_bs(blk);
2775     if (!bs) {
2776         error_setg(errp, "Device has no medium");
2777         goto out;
2778     }
2779 
2780     throttle_config_init(&cfg);
2781     cfg.buckets[THROTTLE_BPS_TOTAL].avg = arg->bps;
2782     cfg.buckets[THROTTLE_BPS_READ].avg  = arg->bps_rd;
2783     cfg.buckets[THROTTLE_BPS_WRITE].avg = arg->bps_wr;
2784 
2785     cfg.buckets[THROTTLE_OPS_TOTAL].avg = arg->iops;
2786     cfg.buckets[THROTTLE_OPS_READ].avg  = arg->iops_rd;
2787     cfg.buckets[THROTTLE_OPS_WRITE].avg = arg->iops_wr;
2788 
2789     if (arg->has_bps_max) {
2790         cfg.buckets[THROTTLE_BPS_TOTAL].max = arg->bps_max;
2791     }
2792     if (arg->has_bps_rd_max) {
2793         cfg.buckets[THROTTLE_BPS_READ].max = arg->bps_rd_max;
2794     }
2795     if (arg->has_bps_wr_max) {
2796         cfg.buckets[THROTTLE_BPS_WRITE].max = arg->bps_wr_max;
2797     }
2798     if (arg->has_iops_max) {
2799         cfg.buckets[THROTTLE_OPS_TOTAL].max = arg->iops_max;
2800     }
2801     if (arg->has_iops_rd_max) {
2802         cfg.buckets[THROTTLE_OPS_READ].max = arg->iops_rd_max;
2803     }
2804     if (arg->has_iops_wr_max) {
2805         cfg.buckets[THROTTLE_OPS_WRITE].max = arg->iops_wr_max;
2806     }
2807 
2808     if (arg->has_bps_max_length) {
2809         cfg.buckets[THROTTLE_BPS_TOTAL].burst_length = arg->bps_max_length;
2810     }
2811     if (arg->has_bps_rd_max_length) {
2812         cfg.buckets[THROTTLE_BPS_READ].burst_length = arg->bps_rd_max_length;
2813     }
2814     if (arg->has_bps_wr_max_length) {
2815         cfg.buckets[THROTTLE_BPS_WRITE].burst_length = arg->bps_wr_max_length;
2816     }
2817     if (arg->has_iops_max_length) {
2818         cfg.buckets[THROTTLE_OPS_TOTAL].burst_length = arg->iops_max_length;
2819     }
2820     if (arg->has_iops_rd_max_length) {
2821         cfg.buckets[THROTTLE_OPS_READ].burst_length = arg->iops_rd_max_length;
2822     }
2823     if (arg->has_iops_wr_max_length) {
2824         cfg.buckets[THROTTLE_OPS_WRITE].burst_length = arg->iops_wr_max_length;
2825     }
2826 
2827     if (arg->has_iops_size) {
2828         cfg.op_size = arg->iops_size;
2829     }
2830 
2831     if (!throttle_is_valid(&cfg, errp)) {
2832         goto out;
2833     }
2834 
2835     if (throttle_enabled(&cfg)) {
2836         /* Enable I/O limits if they're not enabled yet, otherwise
2837          * just update the throttling group. */
2838         if (!blk_get_public(blk)->throttle_group_member.throttle_state) {
2839             blk_io_limits_enable(blk,
2840                                  arg->has_group ? arg->group :
2841                                  arg->has_device ? arg->device :
2842                                  arg->id);
2843         } else if (arg->has_group) {
2844             blk_io_limits_update_group(blk, arg->group);
2845         }
2846         /* Set the new throttling configuration */
2847         blk_set_io_limits(blk, &cfg);
2848     } else if (blk_get_public(blk)->throttle_group_member.throttle_state) {
2849         /* If all throttling settings are set to 0, disable I/O limits */
2850         blk_io_limits_disable(blk);
2851     }
2852 
2853 out:
2854     aio_context_release(aio_context);
2855 }
2856 
2857 void qmp_block_dirty_bitmap_add(const char *node, const char *name,
2858                                 bool has_granularity, uint32_t granularity,
2859                                 bool has_persistent, bool persistent,
2860                                 bool has_disabled, bool disabled,
2861                                 Error **errp)
2862 {
2863     BlockDriverState *bs;
2864     BdrvDirtyBitmap *bitmap;
2865 
2866     if (!name || name[0] == '\0') {
2867         error_setg(errp, "Bitmap name cannot be empty");
2868         return;
2869     }
2870 
2871     bs = bdrv_lookup_bs(node, node, errp);
2872     if (!bs) {
2873         return;
2874     }
2875 
2876     if (has_granularity) {
2877         if (granularity < 512 || !is_power_of_2(granularity)) {
2878             error_setg(errp, "Granularity must be power of 2 "
2879                              "and at least 512");
2880             return;
2881         }
2882     } else {
2883         /* Default to cluster size, if available: */
2884         granularity = bdrv_get_default_bitmap_granularity(bs);
2885     }
2886 
2887     if (!has_persistent) {
2888         persistent = false;
2889     }
2890 
2891     if (!has_disabled) {
2892         disabled = false;
2893     }
2894 
2895     if (persistent &&
2896         !bdrv_can_store_new_dirty_bitmap(bs, name, granularity, errp))
2897     {
2898         return;
2899     }
2900 
2901     bitmap = bdrv_create_dirty_bitmap(bs, granularity, name, errp);
2902     if (bitmap == NULL) {
2903         return;
2904     }
2905 
2906     if (disabled) {
2907         bdrv_disable_dirty_bitmap(bitmap);
2908     }
2909 
2910     bdrv_dirty_bitmap_set_persistence(bitmap, persistent);
2911 }
2912 
2913 static BdrvDirtyBitmap *do_block_dirty_bitmap_remove(
2914         const char *node, const char *name, bool release,
2915         BlockDriverState **bitmap_bs, Error **errp)
2916 {
2917     BlockDriverState *bs;
2918     BdrvDirtyBitmap *bitmap;
2919 
2920     bitmap = block_dirty_bitmap_lookup(node, name, &bs, errp);
2921     if (!bitmap || !bs) {
2922         return NULL;
2923     }
2924 
2925     if (bdrv_dirty_bitmap_check(bitmap, BDRV_BITMAP_BUSY | BDRV_BITMAP_RO,
2926                                 errp)) {
2927         return NULL;
2928     }
2929 
2930     if (bdrv_dirty_bitmap_get_persistence(bitmap) &&
2931         bdrv_remove_persistent_dirty_bitmap(bs, name, errp) < 0)
2932     {
2933             return NULL;
2934     }
2935 
2936     if (release) {
2937         bdrv_release_dirty_bitmap(bitmap);
2938     }
2939 
2940     if (bitmap_bs) {
2941         *bitmap_bs = bs;
2942     }
2943 
2944     return release ? NULL : bitmap;
2945 }
2946 
2947 void qmp_block_dirty_bitmap_remove(const char *node, const char *name,
2948                                    Error **errp)
2949 {
2950     do_block_dirty_bitmap_remove(node, name, true, NULL, errp);
2951 }
2952 
2953 /**
2954  * Completely clear a bitmap, for the purposes of synchronizing a bitmap
2955  * immediately after a full backup operation.
2956  */
2957 void qmp_block_dirty_bitmap_clear(const char *node, const char *name,
2958                                   Error **errp)
2959 {
2960     BdrvDirtyBitmap *bitmap;
2961     BlockDriverState *bs;
2962 
2963     bitmap = block_dirty_bitmap_lookup(node, name, &bs, errp);
2964     if (!bitmap || !bs) {
2965         return;
2966     }
2967 
2968     if (bdrv_dirty_bitmap_check(bitmap, BDRV_BITMAP_DEFAULT, errp)) {
2969         return;
2970     }
2971 
2972     bdrv_clear_dirty_bitmap(bitmap, NULL);
2973 }
2974 
2975 void qmp_block_dirty_bitmap_enable(const char *node, const char *name,
2976                                    Error **errp)
2977 {
2978     BlockDriverState *bs;
2979     BdrvDirtyBitmap *bitmap;
2980 
2981     bitmap = block_dirty_bitmap_lookup(node, name, &bs, errp);
2982     if (!bitmap) {
2983         return;
2984     }
2985 
2986     if (bdrv_dirty_bitmap_check(bitmap, BDRV_BITMAP_ALLOW_RO, errp)) {
2987         return;
2988     }
2989 
2990     bdrv_enable_dirty_bitmap(bitmap);
2991 }
2992 
2993 void qmp_block_dirty_bitmap_disable(const char *node, const char *name,
2994                                     Error **errp)
2995 {
2996     BlockDriverState *bs;
2997     BdrvDirtyBitmap *bitmap;
2998 
2999     bitmap = block_dirty_bitmap_lookup(node, name, &bs, errp);
3000     if (!bitmap) {
3001         return;
3002     }
3003 
3004     if (bdrv_dirty_bitmap_check(bitmap, BDRV_BITMAP_ALLOW_RO, errp)) {
3005         return;
3006     }
3007 
3008     bdrv_disable_dirty_bitmap(bitmap);
3009 }
3010 
3011 static BdrvDirtyBitmap *do_block_dirty_bitmap_merge(
3012         const char *node, const char *target,
3013         BlockDirtyBitmapMergeSourceList *bitmaps,
3014         HBitmap **backup, Error **errp)
3015 {
3016     BlockDriverState *bs;
3017     BdrvDirtyBitmap *dst, *src, *anon;
3018     BlockDirtyBitmapMergeSourceList *lst;
3019     Error *local_err = NULL;
3020 
3021     dst = block_dirty_bitmap_lookup(node, target, &bs, errp);
3022     if (!dst) {
3023         return NULL;
3024     }
3025 
3026     anon = bdrv_create_dirty_bitmap(bs, bdrv_dirty_bitmap_granularity(dst),
3027                                     NULL, errp);
3028     if (!anon) {
3029         return NULL;
3030     }
3031 
3032     for (lst = bitmaps; lst; lst = lst->next) {
3033         switch (lst->value->type) {
3034             const char *name, *node;
3035         case QTYPE_QSTRING:
3036             name = lst->value->u.local;
3037             src = bdrv_find_dirty_bitmap(bs, name);
3038             if (!src) {
3039                 error_setg(errp, "Dirty bitmap '%s' not found", name);
3040                 dst = NULL;
3041                 goto out;
3042             }
3043             break;
3044         case QTYPE_QDICT:
3045             node = lst->value->u.external.node;
3046             name = lst->value->u.external.name;
3047             src = block_dirty_bitmap_lookup(node, name, NULL, errp);
3048             if (!src) {
3049                 dst = NULL;
3050                 goto out;
3051             }
3052             break;
3053         default:
3054             abort();
3055         }
3056 
3057         bdrv_merge_dirty_bitmap(anon, src, NULL, &local_err);
3058         if (local_err) {
3059             error_propagate(errp, local_err);
3060             dst = NULL;
3061             goto out;
3062         }
3063     }
3064 
3065     /* Merge into dst; dst is unchanged on failure. */
3066     bdrv_merge_dirty_bitmap(dst, anon, backup, errp);
3067 
3068  out:
3069     bdrv_release_dirty_bitmap(anon);
3070     return dst;
3071 }
3072 
3073 void qmp_block_dirty_bitmap_merge(const char *node, const char *target,
3074                                   BlockDirtyBitmapMergeSourceList *bitmaps,
3075                                   Error **errp)
3076 {
3077     do_block_dirty_bitmap_merge(node, target, bitmaps, NULL, errp);
3078 }
3079 
3080 BlockDirtyBitmapSha256 *qmp_x_debug_block_dirty_bitmap_sha256(const char *node,
3081                                                               const char *name,
3082                                                               Error **errp)
3083 {
3084     BdrvDirtyBitmap *bitmap;
3085     BlockDriverState *bs;
3086     BlockDirtyBitmapSha256 *ret = NULL;
3087     char *sha256;
3088 
3089     bitmap = block_dirty_bitmap_lookup(node, name, &bs, errp);
3090     if (!bitmap || !bs) {
3091         return NULL;
3092     }
3093 
3094     sha256 = bdrv_dirty_bitmap_sha256(bitmap, errp);
3095     if (sha256 == NULL) {
3096         return NULL;
3097     }
3098 
3099     ret = g_new(BlockDirtyBitmapSha256, 1);
3100     ret->sha256 = sha256;
3101 
3102     return ret;
3103 }
3104 
3105 void hmp_drive_del(Monitor *mon, const QDict *qdict)
3106 {
3107     const char *id = qdict_get_str(qdict, "id");
3108     BlockBackend *blk;
3109     BlockDriverState *bs;
3110     AioContext *aio_context;
3111     Error *local_err = NULL;
3112 
3113     bs = bdrv_find_node(id);
3114     if (bs) {
3115         qmp_blockdev_del(id, &local_err);
3116         if (local_err) {
3117             error_report_err(local_err);
3118         }
3119         return;
3120     }
3121 
3122     blk = blk_by_name(id);
3123     if (!blk) {
3124         error_report("Device '%s' not found", id);
3125         return;
3126     }
3127 
3128     if (!blk_legacy_dinfo(blk)) {
3129         error_report("Deleting device added with blockdev-add"
3130                      " is not supported");
3131         return;
3132     }
3133 
3134     aio_context = blk_get_aio_context(blk);
3135     aio_context_acquire(aio_context);
3136 
3137     bs = blk_bs(blk);
3138     if (bs) {
3139         if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_DRIVE_DEL, &local_err)) {
3140             error_report_err(local_err);
3141             aio_context_release(aio_context);
3142             return;
3143         }
3144 
3145         blk_remove_bs(blk);
3146     }
3147 
3148     /* Make the BlockBackend and the attached BlockDriverState anonymous */
3149     monitor_remove_blk(blk);
3150 
3151     /* If this BlockBackend has a device attached to it, its refcount will be
3152      * decremented when the device is removed; otherwise we have to do so here.
3153      */
3154     if (blk_get_attached_dev(blk)) {
3155         /* Further I/O must not pause the guest */
3156         blk_set_on_error(blk, BLOCKDEV_ON_ERROR_REPORT,
3157                          BLOCKDEV_ON_ERROR_REPORT);
3158     } else {
3159         blk_unref(blk);
3160     }
3161 
3162     aio_context_release(aio_context);
3163 }
3164 
3165 void qmp_block_resize(bool has_device, const char *device,
3166                       bool has_node_name, const char *node_name,
3167                       int64_t size, Error **errp)
3168 {
3169     Error *local_err = NULL;
3170     BlockBackend *blk = NULL;
3171     BlockDriverState *bs;
3172     AioContext *aio_context;
3173     int ret;
3174 
3175     bs = bdrv_lookup_bs(has_device ? device : NULL,
3176                         has_node_name ? node_name : NULL,
3177                         &local_err);
3178     if (local_err) {
3179         error_propagate(errp, local_err);
3180         return;
3181     }
3182 
3183     aio_context = bdrv_get_aio_context(bs);
3184     aio_context_acquire(aio_context);
3185 
3186     if (!bdrv_is_first_non_filter(bs)) {
3187         error_setg(errp, QERR_FEATURE_DISABLED, "resize");
3188         goto out;
3189     }
3190 
3191     if (size < 0) {
3192         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "size", "a >0 size");
3193         goto out;
3194     }
3195 
3196     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_RESIZE, NULL)) {
3197         error_setg(errp, QERR_DEVICE_IN_USE, device);
3198         goto out;
3199     }
3200 
3201     blk = blk_new(bdrv_get_aio_context(bs), BLK_PERM_RESIZE, BLK_PERM_ALL);
3202     ret = blk_insert_bs(blk, bs, errp);
3203     if (ret < 0) {
3204         goto out;
3205     }
3206 
3207     bdrv_drained_begin(bs);
3208     ret = blk_truncate(blk, size, PREALLOC_MODE_OFF, errp);
3209     bdrv_drained_end(bs);
3210 
3211 out:
3212     blk_unref(blk);
3213     aio_context_release(aio_context);
3214 }
3215 
3216 void qmp_block_stream(bool has_job_id, const char *job_id, const char *device,
3217                       bool has_base, const char *base,
3218                       bool has_base_node, const char *base_node,
3219                       bool has_backing_file, const char *backing_file,
3220                       bool has_speed, int64_t speed,
3221                       bool has_on_error, BlockdevOnError on_error,
3222                       bool has_auto_finalize, bool auto_finalize,
3223                       bool has_auto_dismiss, bool auto_dismiss,
3224                       Error **errp)
3225 {
3226     BlockDriverState *bs, *iter;
3227     BlockDriverState *base_bs = NULL;
3228     AioContext *aio_context;
3229     Error *local_err = NULL;
3230     const char *base_name = NULL;
3231     int job_flags = JOB_DEFAULT;
3232 
3233     if (!has_on_error) {
3234         on_error = BLOCKDEV_ON_ERROR_REPORT;
3235     }
3236 
3237     bs = bdrv_lookup_bs(device, device, errp);
3238     if (!bs) {
3239         return;
3240     }
3241 
3242     aio_context = bdrv_get_aio_context(bs);
3243     aio_context_acquire(aio_context);
3244 
3245     if (has_base && has_base_node) {
3246         error_setg(errp, "'base' and 'base-node' cannot be specified "
3247                    "at the same time");
3248         goto out;
3249     }
3250 
3251     if (has_base) {
3252         base_bs = bdrv_find_backing_image(bs, base);
3253         if (base_bs == NULL) {
3254             error_setg(errp, QERR_BASE_NOT_FOUND, base);
3255             goto out;
3256         }
3257         assert(bdrv_get_aio_context(base_bs) == aio_context);
3258         base_name = base;
3259     }
3260 
3261     if (has_base_node) {
3262         base_bs = bdrv_lookup_bs(NULL, base_node, errp);
3263         if (!base_bs) {
3264             goto out;
3265         }
3266         if (bs == base_bs || !bdrv_chain_contains(bs, base_bs)) {
3267             error_setg(errp, "Node '%s' is not a backing image of '%s'",
3268                        base_node, device);
3269             goto out;
3270         }
3271         assert(bdrv_get_aio_context(base_bs) == aio_context);
3272         bdrv_refresh_filename(base_bs);
3273         base_name = base_bs->filename;
3274     }
3275 
3276     /* Check for op blockers in the whole chain between bs and base */
3277     for (iter = bs; iter && iter != base_bs; iter = backing_bs(iter)) {
3278         if (bdrv_op_is_blocked(iter, BLOCK_OP_TYPE_STREAM, errp)) {
3279             goto out;
3280         }
3281     }
3282 
3283     /* if we are streaming the entire chain, the result will have no backing
3284      * file, and specifying one is therefore an error */
3285     if (base_bs == NULL && has_backing_file) {
3286         error_setg(errp, "backing file specified, but streaming the "
3287                          "entire chain");
3288         goto out;
3289     }
3290 
3291     /* backing_file string overrides base bs filename */
3292     base_name = has_backing_file ? backing_file : base_name;
3293 
3294     if (has_auto_finalize && !auto_finalize) {
3295         job_flags |= JOB_MANUAL_FINALIZE;
3296     }
3297     if (has_auto_dismiss && !auto_dismiss) {
3298         job_flags |= JOB_MANUAL_DISMISS;
3299     }
3300 
3301     stream_start(has_job_id ? job_id : NULL, bs, base_bs, base_name,
3302                  job_flags, has_speed ? speed : 0, on_error, &local_err);
3303     if (local_err) {
3304         error_propagate(errp, local_err);
3305         goto out;
3306     }
3307 
3308     trace_qmp_block_stream(bs);
3309 
3310 out:
3311     aio_context_release(aio_context);
3312 }
3313 
3314 void qmp_block_commit(bool has_job_id, const char *job_id, const char *device,
3315                       bool has_base_node, const char *base_node,
3316                       bool has_base, const char *base,
3317                       bool has_top_node, const char *top_node,
3318                       bool has_top, const char *top,
3319                       bool has_backing_file, const char *backing_file,
3320                       bool has_speed, int64_t speed,
3321                       bool has_filter_node_name, const char *filter_node_name,
3322                       bool has_auto_finalize, bool auto_finalize,
3323                       bool has_auto_dismiss, bool auto_dismiss,
3324                       Error **errp)
3325 {
3326     BlockDriverState *bs;
3327     BlockDriverState *iter;
3328     BlockDriverState *base_bs, *top_bs;
3329     AioContext *aio_context;
3330     Error *local_err = NULL;
3331     /* This will be part of the QMP command, if/when the
3332      * BlockdevOnError change for blkmirror makes it in
3333      */
3334     BlockdevOnError on_error = BLOCKDEV_ON_ERROR_REPORT;
3335     int job_flags = JOB_DEFAULT;
3336 
3337     if (!has_speed) {
3338         speed = 0;
3339     }
3340     if (!has_filter_node_name) {
3341         filter_node_name = NULL;
3342     }
3343     if (has_auto_finalize && !auto_finalize) {
3344         job_flags |= JOB_MANUAL_FINALIZE;
3345     }
3346     if (has_auto_dismiss && !auto_dismiss) {
3347         job_flags |= JOB_MANUAL_DISMISS;
3348     }
3349 
3350     /* Important Note:
3351      *  libvirt relies on the DeviceNotFound error class in order to probe for
3352      *  live commit feature versions; for this to work, we must make sure to
3353      *  perform the device lookup before any generic errors that may occur in a
3354      *  scenario in which all optional arguments are omitted. */
3355     bs = qmp_get_root_bs(device, &local_err);
3356     if (!bs) {
3357         bs = bdrv_lookup_bs(device, device, NULL);
3358         if (!bs) {
3359             error_free(local_err);
3360             error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
3361                       "Device '%s' not found", device);
3362         } else {
3363             error_propagate(errp, local_err);
3364         }
3365         return;
3366     }
3367 
3368     aio_context = bdrv_get_aio_context(bs);
3369     aio_context_acquire(aio_context);
3370 
3371     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_COMMIT_SOURCE, errp)) {
3372         goto out;
3373     }
3374 
3375     /* default top_bs is the active layer */
3376     top_bs = bs;
3377 
3378     if (has_top_node && has_top) {
3379         error_setg(errp, "'top-node' and 'top' are mutually exclusive");
3380         goto out;
3381     } else if (has_top_node) {
3382         top_bs = bdrv_lookup_bs(NULL, top_node, errp);
3383         if (top_bs == NULL) {
3384             goto out;
3385         }
3386         if (!bdrv_chain_contains(bs, top_bs)) {
3387             error_setg(errp, "'%s' is not in this backing file chain",
3388                        top_node);
3389             goto out;
3390         }
3391     } else if (has_top && top) {
3392         /* This strcmp() is just a shortcut, there is no need to
3393          * refresh @bs's filename.  If it mismatches,
3394          * bdrv_find_backing_image() will do the refresh and may still
3395          * return @bs. */
3396         if (strcmp(bs->filename, top) != 0) {
3397             top_bs = bdrv_find_backing_image(bs, top);
3398         }
3399     }
3400 
3401     if (top_bs == NULL) {
3402         error_setg(errp, "Top image file %s not found", top ? top : "NULL");
3403         goto out;
3404     }
3405 
3406     assert(bdrv_get_aio_context(top_bs) == aio_context);
3407 
3408     if (has_base_node && has_base) {
3409         error_setg(errp, "'base-node' and 'base' are mutually exclusive");
3410         goto out;
3411     } else if (has_base_node) {
3412         base_bs = bdrv_lookup_bs(NULL, base_node, errp);
3413         if (base_bs == NULL) {
3414             goto out;
3415         }
3416         if (!bdrv_chain_contains(top_bs, base_bs)) {
3417             error_setg(errp, "'%s' is not in this backing file chain",
3418                        base_node);
3419             goto out;
3420         }
3421     } else if (has_base && base) {
3422         base_bs = bdrv_find_backing_image(top_bs, base);
3423     } else {
3424         base_bs = bdrv_find_base(top_bs);
3425     }
3426 
3427     if (base_bs == NULL) {
3428         error_setg(errp, QERR_BASE_NOT_FOUND, base ? base : "NULL");
3429         goto out;
3430     }
3431 
3432     assert(bdrv_get_aio_context(base_bs) == aio_context);
3433 
3434     for (iter = top_bs; iter != backing_bs(base_bs); iter = backing_bs(iter)) {
3435         if (bdrv_op_is_blocked(iter, BLOCK_OP_TYPE_COMMIT_TARGET, errp)) {
3436             goto out;
3437         }
3438     }
3439 
3440     /* Do not allow attempts to commit an image into itself */
3441     if (top_bs == base_bs) {
3442         error_setg(errp, "cannot commit an image into itself");
3443         goto out;
3444     }
3445 
3446     if (top_bs == bs) {
3447         if (has_backing_file) {
3448             error_setg(errp, "'backing-file' specified,"
3449                              " but 'top' is the active layer");
3450             goto out;
3451         }
3452         commit_active_start(has_job_id ? job_id : NULL, bs, base_bs,
3453                             job_flags, speed, on_error,
3454                             filter_node_name, NULL, NULL, false, &local_err);
3455     } else {
3456         BlockDriverState *overlay_bs = bdrv_find_overlay(bs, top_bs);
3457         if (bdrv_op_is_blocked(overlay_bs, BLOCK_OP_TYPE_COMMIT_TARGET, errp)) {
3458             goto out;
3459         }
3460         commit_start(has_job_id ? job_id : NULL, bs, base_bs, top_bs, job_flags,
3461                      speed, on_error, has_backing_file ? backing_file : NULL,
3462                      filter_node_name, &local_err);
3463     }
3464     if (local_err != NULL) {
3465         error_propagate(errp, local_err);
3466         goto out;
3467     }
3468 
3469 out:
3470     aio_context_release(aio_context);
3471 }
3472 
3473 /* Common QMP interface for drive-backup and blockdev-backup */
3474 static BlockJob *do_backup_common(BackupCommon *backup,
3475                                   BlockDriverState *bs,
3476                                   BlockDriverState *target_bs,
3477                                   AioContext *aio_context,
3478                                   JobTxn *txn, Error **errp)
3479 {
3480     BlockJob *job = NULL;
3481     BdrvDirtyBitmap *bmap = NULL;
3482     int job_flags = JOB_DEFAULT;
3483     int ret;
3484 
3485     if (!backup->has_speed) {
3486         backup->speed = 0;
3487     }
3488     if (!backup->has_on_source_error) {
3489         backup->on_source_error = BLOCKDEV_ON_ERROR_REPORT;
3490     }
3491     if (!backup->has_on_target_error) {
3492         backup->on_target_error = BLOCKDEV_ON_ERROR_REPORT;
3493     }
3494     if (!backup->has_job_id) {
3495         backup->job_id = NULL;
3496     }
3497     if (!backup->has_auto_finalize) {
3498         backup->auto_finalize = true;
3499     }
3500     if (!backup->has_auto_dismiss) {
3501         backup->auto_dismiss = true;
3502     }
3503     if (!backup->has_compress) {
3504         backup->compress = false;
3505     }
3506 
3507     ret = bdrv_try_set_aio_context(target_bs, aio_context, errp);
3508     if (ret < 0) {
3509         return NULL;
3510     }
3511 
3512     if ((backup->sync == MIRROR_SYNC_MODE_BITMAP) ||
3513         (backup->sync == MIRROR_SYNC_MODE_INCREMENTAL)) {
3514         /* done before desugaring 'incremental' to print the right message */
3515         if (!backup->has_bitmap) {
3516             error_setg(errp, "must provide a valid bitmap name for "
3517                        "'%s' sync mode", MirrorSyncMode_str(backup->sync));
3518             return NULL;
3519         }
3520     }
3521 
3522     if (backup->sync == MIRROR_SYNC_MODE_INCREMENTAL) {
3523         if (backup->has_bitmap_mode &&
3524             backup->bitmap_mode != BITMAP_SYNC_MODE_ON_SUCCESS) {
3525             error_setg(errp, "Bitmap sync mode must be '%s' "
3526                        "when using sync mode '%s'",
3527                        BitmapSyncMode_str(BITMAP_SYNC_MODE_ON_SUCCESS),
3528                        MirrorSyncMode_str(backup->sync));
3529             return NULL;
3530         }
3531         backup->has_bitmap_mode = true;
3532         backup->sync = MIRROR_SYNC_MODE_BITMAP;
3533         backup->bitmap_mode = BITMAP_SYNC_MODE_ON_SUCCESS;
3534     }
3535 
3536     if (backup->has_bitmap) {
3537         bmap = bdrv_find_dirty_bitmap(bs, backup->bitmap);
3538         if (!bmap) {
3539             error_setg(errp, "Bitmap '%s' could not be found", backup->bitmap);
3540             return NULL;
3541         }
3542         if (!backup->has_bitmap_mode) {
3543             error_setg(errp, "Bitmap sync mode must be given "
3544                        "when providing a bitmap");
3545             return NULL;
3546         }
3547         if (bdrv_dirty_bitmap_check(bmap, BDRV_BITMAP_ALLOW_RO, errp)) {
3548             return NULL;
3549         }
3550 
3551         /* This does not produce a useful bitmap artifact: */
3552         if (backup->sync == MIRROR_SYNC_MODE_NONE) {
3553             error_setg(errp, "sync mode '%s' does not produce meaningful bitmap"
3554                        " outputs", MirrorSyncMode_str(backup->sync));
3555             return NULL;
3556         }
3557 
3558         /* If the bitmap isn't used for input or output, this is useless: */
3559         if (backup->bitmap_mode == BITMAP_SYNC_MODE_NEVER &&
3560             backup->sync != MIRROR_SYNC_MODE_BITMAP) {
3561             error_setg(errp, "Bitmap sync mode '%s' has no meaningful effect"
3562                        " when combined with sync mode '%s'",
3563                        BitmapSyncMode_str(backup->bitmap_mode),
3564                        MirrorSyncMode_str(backup->sync));
3565             return NULL;
3566         }
3567     }
3568 
3569     if (!backup->has_bitmap && backup->has_bitmap_mode) {
3570         error_setg(errp, "Cannot specify bitmap sync mode without a bitmap");
3571         return NULL;
3572     }
3573 
3574     if (!backup->auto_finalize) {
3575         job_flags |= JOB_MANUAL_FINALIZE;
3576     }
3577     if (!backup->auto_dismiss) {
3578         job_flags |= JOB_MANUAL_DISMISS;
3579     }
3580 
3581     job = backup_job_create(backup->job_id, bs, target_bs, backup->speed,
3582                             backup->sync, bmap, backup->bitmap_mode,
3583                             backup->compress,
3584                             backup->filter_node_name,
3585                             backup->on_source_error,
3586                             backup->on_target_error,
3587                             job_flags, NULL, NULL, txn, errp);
3588     return job;
3589 }
3590 
3591 static BlockJob *do_drive_backup(DriveBackup *backup, JobTxn *txn,
3592                                  Error **errp)
3593 {
3594     BlockDriverState *bs;
3595     BlockDriverState *target_bs;
3596     BlockDriverState *source = NULL;
3597     BlockJob *job = NULL;
3598     AioContext *aio_context;
3599     QDict *options;
3600     Error *local_err = NULL;
3601     int flags;
3602     int64_t size;
3603     bool set_backing_hd = false;
3604 
3605     if (!backup->has_mode) {
3606         backup->mode = NEW_IMAGE_MODE_ABSOLUTE_PATHS;
3607     }
3608 
3609     bs = bdrv_lookup_bs(backup->device, backup->device, errp);
3610     if (!bs) {
3611         return NULL;
3612     }
3613 
3614     if (!bs->drv) {
3615         error_setg(errp, "Device has no medium");
3616         return NULL;
3617     }
3618 
3619     aio_context = bdrv_get_aio_context(bs);
3620     aio_context_acquire(aio_context);
3621 
3622     if (!backup->has_format) {
3623         backup->format = backup->mode == NEW_IMAGE_MODE_EXISTING ?
3624                          NULL : (char*) bs->drv->format_name;
3625     }
3626 
3627     /* Early check to avoid creating target */
3628     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_BACKUP_SOURCE, errp)) {
3629         goto out;
3630     }
3631 
3632     flags = bs->open_flags | BDRV_O_RDWR;
3633 
3634     /* See if we have a backing HD we can use to create our new image
3635      * on top of. */
3636     if (backup->sync == MIRROR_SYNC_MODE_TOP) {
3637         source = backing_bs(bs);
3638         if (!source) {
3639             backup->sync = MIRROR_SYNC_MODE_FULL;
3640         }
3641     }
3642     if (backup->sync == MIRROR_SYNC_MODE_NONE) {
3643         source = bs;
3644         flags |= BDRV_O_NO_BACKING;
3645         set_backing_hd = true;
3646     }
3647 
3648     size = bdrv_getlength(bs);
3649     if (size < 0) {
3650         error_setg_errno(errp, -size, "bdrv_getlength failed");
3651         goto out;
3652     }
3653 
3654     if (backup->mode != NEW_IMAGE_MODE_EXISTING) {
3655         assert(backup->format);
3656         if (source) {
3657             bdrv_refresh_filename(source);
3658             bdrv_img_create(backup->target, backup->format, source->filename,
3659                             source->drv->format_name, NULL,
3660                             size, flags, false, &local_err);
3661         } else {
3662             bdrv_img_create(backup->target, backup->format, NULL, NULL, NULL,
3663                             size, flags, false, &local_err);
3664         }
3665     }
3666 
3667     if (local_err) {
3668         error_propagate(errp, local_err);
3669         goto out;
3670     }
3671 
3672     options = qdict_new();
3673     qdict_put_str(options, "discard", "unmap");
3674     qdict_put_str(options, "detect-zeroes", "unmap");
3675     if (backup->format) {
3676         qdict_put_str(options, "driver", backup->format);
3677     }
3678 
3679     target_bs = bdrv_open(backup->target, NULL, options, flags, errp);
3680     if (!target_bs) {
3681         goto out;
3682     }
3683 
3684     if (set_backing_hd) {
3685         bdrv_set_backing_hd(target_bs, source, &local_err);
3686         if (local_err) {
3687             goto unref;
3688         }
3689     }
3690 
3691     job = do_backup_common(qapi_DriveBackup_base(backup),
3692                            bs, target_bs, aio_context, txn, errp);
3693 
3694 unref:
3695     bdrv_unref(target_bs);
3696 out:
3697     aio_context_release(aio_context);
3698     return job;
3699 }
3700 
3701 void qmp_drive_backup(DriveBackup *arg, Error **errp)
3702 {
3703 
3704     BlockJob *job;
3705     job = do_drive_backup(arg, NULL, errp);
3706     if (job) {
3707         job_start(&job->job);
3708     }
3709 }
3710 
3711 BlockDeviceInfoList *qmp_query_named_block_nodes(Error **errp)
3712 {
3713     return bdrv_named_nodes_list(errp);
3714 }
3715 
3716 XDbgBlockGraph *qmp_x_debug_query_block_graph(Error **errp)
3717 {
3718     return bdrv_get_xdbg_block_graph(errp);
3719 }
3720 
3721 BlockJob *do_blockdev_backup(BlockdevBackup *backup, JobTxn *txn,
3722                              Error **errp)
3723 {
3724     BlockDriverState *bs;
3725     BlockDriverState *target_bs;
3726     AioContext *aio_context;
3727     BlockJob *job;
3728 
3729     bs = bdrv_lookup_bs(backup->device, backup->device, errp);
3730     if (!bs) {
3731         return NULL;
3732     }
3733 
3734     target_bs = bdrv_lookup_bs(backup->target, backup->target, errp);
3735     if (!target_bs) {
3736         return NULL;
3737     }
3738 
3739     aio_context = bdrv_get_aio_context(bs);
3740     aio_context_acquire(aio_context);
3741 
3742     job = do_backup_common(qapi_BlockdevBackup_base(backup),
3743                            bs, target_bs, aio_context, txn, errp);
3744 
3745     aio_context_release(aio_context);
3746     return job;
3747 }
3748 
3749 void qmp_blockdev_backup(BlockdevBackup *arg, Error **errp)
3750 {
3751     BlockJob *job;
3752     job = do_blockdev_backup(arg, NULL, errp);
3753     if (job) {
3754         job_start(&job->job);
3755     }
3756 }
3757 
3758 /* Parameter check and block job starting for drive mirroring.
3759  * Caller should hold @device and @target's aio context (must be the same).
3760  **/
3761 static void blockdev_mirror_common(const char *job_id, BlockDriverState *bs,
3762                                    BlockDriverState *target,
3763                                    bool has_replaces, const char *replaces,
3764                                    enum MirrorSyncMode sync,
3765                                    BlockMirrorBackingMode backing_mode,
3766                                    bool zero_target,
3767                                    bool has_speed, int64_t speed,
3768                                    bool has_granularity, uint32_t granularity,
3769                                    bool has_buf_size, int64_t buf_size,
3770                                    bool has_on_source_error,
3771                                    BlockdevOnError on_source_error,
3772                                    bool has_on_target_error,
3773                                    BlockdevOnError on_target_error,
3774                                    bool has_unmap, bool unmap,
3775                                    bool has_filter_node_name,
3776                                    const char *filter_node_name,
3777                                    bool has_copy_mode, MirrorCopyMode copy_mode,
3778                                    bool has_auto_finalize, bool auto_finalize,
3779                                    bool has_auto_dismiss, bool auto_dismiss,
3780                                    Error **errp)
3781 {
3782     int job_flags = JOB_DEFAULT;
3783 
3784     if (!has_speed) {
3785         speed = 0;
3786     }
3787     if (!has_on_source_error) {
3788         on_source_error = BLOCKDEV_ON_ERROR_REPORT;
3789     }
3790     if (!has_on_target_error) {
3791         on_target_error = BLOCKDEV_ON_ERROR_REPORT;
3792     }
3793     if (!has_granularity) {
3794         granularity = 0;
3795     }
3796     if (!has_buf_size) {
3797         buf_size = 0;
3798     }
3799     if (!has_unmap) {
3800         unmap = true;
3801     }
3802     if (!has_filter_node_name) {
3803         filter_node_name = NULL;
3804     }
3805     if (!has_copy_mode) {
3806         copy_mode = MIRROR_COPY_MODE_BACKGROUND;
3807     }
3808     if (has_auto_finalize && !auto_finalize) {
3809         job_flags |= JOB_MANUAL_FINALIZE;
3810     }
3811     if (has_auto_dismiss && !auto_dismiss) {
3812         job_flags |= JOB_MANUAL_DISMISS;
3813     }
3814 
3815     if (granularity != 0 && (granularity < 512 || granularity > 1048576 * 64)) {
3816         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "granularity",
3817                    "a value in range [512B, 64MB]");
3818         return;
3819     }
3820     if (granularity & (granularity - 1)) {
3821         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "granularity",
3822                    "power of 2");
3823         return;
3824     }
3825 
3826     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_MIRROR_SOURCE, errp)) {
3827         return;
3828     }
3829     if (bdrv_op_is_blocked(target, BLOCK_OP_TYPE_MIRROR_TARGET, errp)) {
3830         return;
3831     }
3832 
3833     if (!bs->backing && sync == MIRROR_SYNC_MODE_TOP) {
3834         sync = MIRROR_SYNC_MODE_FULL;
3835     }
3836 
3837     if (has_replaces) {
3838         BlockDriverState *to_replace_bs;
3839         AioContext *replace_aio_context;
3840         int64_t bs_size, replace_size;
3841 
3842         bs_size = bdrv_getlength(bs);
3843         if (bs_size < 0) {
3844             error_setg_errno(errp, -bs_size, "Failed to query device's size");
3845             return;
3846         }
3847 
3848         to_replace_bs = check_to_replace_node(bs, replaces, errp);
3849         if (!to_replace_bs) {
3850             return;
3851         }
3852 
3853         replace_aio_context = bdrv_get_aio_context(to_replace_bs);
3854         aio_context_acquire(replace_aio_context);
3855         replace_size = bdrv_getlength(to_replace_bs);
3856         aio_context_release(replace_aio_context);
3857 
3858         if (replace_size < 0) {
3859             error_setg_errno(errp, -replace_size,
3860                              "Failed to query the replacement node's size");
3861             return;
3862         }
3863         if (bs_size != replace_size) {
3864             error_setg(errp, "cannot replace image with a mirror image of "
3865                              "different size");
3866             return;
3867         }
3868     }
3869 
3870     /* pass the node name to replace to mirror start since it's loose coupling
3871      * and will allow to check whether the node still exist at mirror completion
3872      */
3873     mirror_start(job_id, bs, target,
3874                  has_replaces ? replaces : NULL, job_flags,
3875                  speed, granularity, buf_size, sync, backing_mode, zero_target,
3876                  on_source_error, on_target_error, unmap, filter_node_name,
3877                  copy_mode, errp);
3878 }
3879 
3880 void qmp_drive_mirror(DriveMirror *arg, Error **errp)
3881 {
3882     BlockDriverState *bs;
3883     BlockDriverState *source, *target_bs;
3884     AioContext *aio_context;
3885     BlockMirrorBackingMode backing_mode;
3886     Error *local_err = NULL;
3887     QDict *options = NULL;
3888     int flags;
3889     int64_t size;
3890     const char *format = arg->format;
3891     bool zero_target;
3892     int ret;
3893 
3894     bs = qmp_get_root_bs(arg->device, errp);
3895     if (!bs) {
3896         return;
3897     }
3898 
3899     /* Early check to avoid creating target */
3900     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_MIRROR_SOURCE, errp)) {
3901         return;
3902     }
3903 
3904     aio_context = bdrv_get_aio_context(bs);
3905     aio_context_acquire(aio_context);
3906 
3907     if (!arg->has_mode) {
3908         arg->mode = NEW_IMAGE_MODE_ABSOLUTE_PATHS;
3909     }
3910 
3911     if (!arg->has_format) {
3912         format = (arg->mode == NEW_IMAGE_MODE_EXISTING
3913                   ? NULL : bs->drv->format_name);
3914     }
3915 
3916     flags = bs->open_flags | BDRV_O_RDWR;
3917     source = backing_bs(bs);
3918     if (!source && arg->sync == MIRROR_SYNC_MODE_TOP) {
3919         arg->sync = MIRROR_SYNC_MODE_FULL;
3920     }
3921     if (arg->sync == MIRROR_SYNC_MODE_NONE) {
3922         source = bs;
3923     }
3924 
3925     size = bdrv_getlength(bs);
3926     if (size < 0) {
3927         error_setg_errno(errp, -size, "bdrv_getlength failed");
3928         goto out;
3929     }
3930 
3931     if (arg->has_replaces) {
3932         if (!arg->has_node_name) {
3933             error_setg(errp, "a node-name must be provided when replacing a"
3934                              " named node of the graph");
3935             goto out;
3936         }
3937     }
3938 
3939     if (arg->mode == NEW_IMAGE_MODE_ABSOLUTE_PATHS) {
3940         backing_mode = MIRROR_SOURCE_BACKING_CHAIN;
3941     } else {
3942         backing_mode = MIRROR_OPEN_BACKING_CHAIN;
3943     }
3944 
3945     /* Don't open backing image in create() */
3946     flags |= BDRV_O_NO_BACKING;
3947 
3948     if ((arg->sync == MIRROR_SYNC_MODE_FULL || !source)
3949         && arg->mode != NEW_IMAGE_MODE_EXISTING)
3950     {
3951         /* create new image w/o backing file */
3952         assert(format);
3953         bdrv_img_create(arg->target, format,
3954                         NULL, NULL, NULL, size, flags, false, &local_err);
3955     } else {
3956         switch (arg->mode) {
3957         case NEW_IMAGE_MODE_EXISTING:
3958             break;
3959         case NEW_IMAGE_MODE_ABSOLUTE_PATHS:
3960             /* create new image with backing file */
3961             bdrv_refresh_filename(source);
3962             bdrv_img_create(arg->target, format,
3963                             source->filename,
3964                             source->drv->format_name,
3965                             NULL, size, flags, false, &local_err);
3966             break;
3967         default:
3968             abort();
3969         }
3970     }
3971 
3972     if (local_err) {
3973         error_propagate(errp, local_err);
3974         goto out;
3975     }
3976 
3977     options = qdict_new();
3978     if (arg->has_node_name) {
3979         qdict_put_str(options, "node-name", arg->node_name);
3980     }
3981     if (format) {
3982         qdict_put_str(options, "driver", format);
3983     }
3984 
3985     /* Mirroring takes care of copy-on-write using the source's backing
3986      * file.
3987      */
3988     target_bs = bdrv_open(arg->target, NULL, options, flags, errp);
3989     if (!target_bs) {
3990         goto out;
3991     }
3992 
3993     zero_target = (arg->sync == MIRROR_SYNC_MODE_FULL &&
3994                    (arg->mode == NEW_IMAGE_MODE_EXISTING ||
3995                     !bdrv_has_zero_init(target_bs)));
3996 
3997     ret = bdrv_try_set_aio_context(target_bs, aio_context, errp);
3998     if (ret < 0) {
3999         bdrv_unref(target_bs);
4000         goto out;
4001     }
4002 
4003     blockdev_mirror_common(arg->has_job_id ? arg->job_id : NULL, bs, target_bs,
4004                            arg->has_replaces, arg->replaces, arg->sync,
4005                            backing_mode, zero_target,
4006                            arg->has_speed, arg->speed,
4007                            arg->has_granularity, arg->granularity,
4008                            arg->has_buf_size, arg->buf_size,
4009                            arg->has_on_source_error, arg->on_source_error,
4010                            arg->has_on_target_error, arg->on_target_error,
4011                            arg->has_unmap, arg->unmap,
4012                            false, NULL,
4013                            arg->has_copy_mode, arg->copy_mode,
4014                            arg->has_auto_finalize, arg->auto_finalize,
4015                            arg->has_auto_dismiss, arg->auto_dismiss,
4016                            &local_err);
4017     bdrv_unref(target_bs);
4018     error_propagate(errp, local_err);
4019 out:
4020     aio_context_release(aio_context);
4021 }
4022 
4023 void qmp_blockdev_mirror(bool has_job_id, const char *job_id,
4024                          const char *device, const char *target,
4025                          bool has_replaces, const char *replaces,
4026                          MirrorSyncMode sync,
4027                          bool has_speed, int64_t speed,
4028                          bool has_granularity, uint32_t granularity,
4029                          bool has_buf_size, int64_t buf_size,
4030                          bool has_on_source_error,
4031                          BlockdevOnError on_source_error,
4032                          bool has_on_target_error,
4033                          BlockdevOnError on_target_error,
4034                          bool has_filter_node_name,
4035                          const char *filter_node_name,
4036                          bool has_copy_mode, MirrorCopyMode copy_mode,
4037                          bool has_auto_finalize, bool auto_finalize,
4038                          bool has_auto_dismiss, bool auto_dismiss,
4039                          Error **errp)
4040 {
4041     BlockDriverState *bs;
4042     BlockDriverState *target_bs;
4043     AioContext *aio_context;
4044     BlockMirrorBackingMode backing_mode = MIRROR_LEAVE_BACKING_CHAIN;
4045     Error *local_err = NULL;
4046     bool zero_target;
4047     int ret;
4048 
4049     bs = qmp_get_root_bs(device, errp);
4050     if (!bs) {
4051         return;
4052     }
4053 
4054     target_bs = bdrv_lookup_bs(target, target, errp);
4055     if (!target_bs) {
4056         return;
4057     }
4058 
4059     zero_target = (sync == MIRROR_SYNC_MODE_FULL);
4060 
4061     aio_context = bdrv_get_aio_context(bs);
4062     aio_context_acquire(aio_context);
4063 
4064     ret = bdrv_try_set_aio_context(target_bs, aio_context, errp);
4065     if (ret < 0) {
4066         goto out;
4067     }
4068 
4069     blockdev_mirror_common(has_job_id ? job_id : NULL, bs, target_bs,
4070                            has_replaces, replaces, sync, backing_mode,
4071                            zero_target, has_speed, speed,
4072                            has_granularity, granularity,
4073                            has_buf_size, buf_size,
4074                            has_on_source_error, on_source_error,
4075                            has_on_target_error, on_target_error,
4076                            true, true,
4077                            has_filter_node_name, filter_node_name,
4078                            has_copy_mode, copy_mode,
4079                            has_auto_finalize, auto_finalize,
4080                            has_auto_dismiss, auto_dismiss,
4081                            &local_err);
4082     error_propagate(errp, local_err);
4083 out:
4084     aio_context_release(aio_context);
4085 }
4086 
4087 /* Get a block job using its ID and acquire its AioContext */
4088 static BlockJob *find_block_job(const char *id, AioContext **aio_context,
4089                                 Error **errp)
4090 {
4091     BlockJob *job;
4092 
4093     assert(id != NULL);
4094 
4095     *aio_context = NULL;
4096 
4097     job = block_job_get(id);
4098 
4099     if (!job) {
4100         error_set(errp, ERROR_CLASS_DEVICE_NOT_ACTIVE,
4101                   "Block job '%s' not found", id);
4102         return NULL;
4103     }
4104 
4105     *aio_context = blk_get_aio_context(job->blk);
4106     aio_context_acquire(*aio_context);
4107 
4108     return job;
4109 }
4110 
4111 void qmp_block_job_set_speed(const char *device, int64_t speed, Error **errp)
4112 {
4113     AioContext *aio_context;
4114     BlockJob *job = find_block_job(device, &aio_context, errp);
4115 
4116     if (!job) {
4117         return;
4118     }
4119 
4120     block_job_set_speed(job, speed, errp);
4121     aio_context_release(aio_context);
4122 }
4123 
4124 void qmp_block_job_cancel(const char *device,
4125                           bool has_force, bool force, Error **errp)
4126 {
4127     AioContext *aio_context;
4128     BlockJob *job = find_block_job(device, &aio_context, errp);
4129 
4130     if (!job) {
4131         return;
4132     }
4133 
4134     if (!has_force) {
4135         force = false;
4136     }
4137 
4138     if (job_user_paused(&job->job) && !force) {
4139         error_setg(errp, "The block job for device '%s' is currently paused",
4140                    device);
4141         goto out;
4142     }
4143 
4144     trace_qmp_block_job_cancel(job);
4145     job_user_cancel(&job->job, force, errp);
4146 out:
4147     aio_context_release(aio_context);
4148 }
4149 
4150 void qmp_block_job_pause(const char *device, Error **errp)
4151 {
4152     AioContext *aio_context;
4153     BlockJob *job = find_block_job(device, &aio_context, errp);
4154 
4155     if (!job) {
4156         return;
4157     }
4158 
4159     trace_qmp_block_job_pause(job);
4160     job_user_pause(&job->job, errp);
4161     aio_context_release(aio_context);
4162 }
4163 
4164 void qmp_block_job_resume(const char *device, Error **errp)
4165 {
4166     AioContext *aio_context;
4167     BlockJob *job = find_block_job(device, &aio_context, errp);
4168 
4169     if (!job) {
4170         return;
4171     }
4172 
4173     trace_qmp_block_job_resume(job);
4174     job_user_resume(&job->job, errp);
4175     aio_context_release(aio_context);
4176 }
4177 
4178 void qmp_block_job_complete(const char *device, Error **errp)
4179 {
4180     AioContext *aio_context;
4181     BlockJob *job = find_block_job(device, &aio_context, errp);
4182 
4183     if (!job) {
4184         return;
4185     }
4186 
4187     trace_qmp_block_job_complete(job);
4188     job_complete(&job->job, errp);
4189     aio_context_release(aio_context);
4190 }
4191 
4192 void qmp_block_job_finalize(const char *id, Error **errp)
4193 {
4194     AioContext *aio_context;
4195     BlockJob *job = find_block_job(id, &aio_context, errp);
4196 
4197     if (!job) {
4198         return;
4199     }
4200 
4201     trace_qmp_block_job_finalize(job);
4202     job_finalize(&job->job, errp);
4203     aio_context_release(aio_context);
4204 }
4205 
4206 void qmp_block_job_dismiss(const char *id, Error **errp)
4207 {
4208     AioContext *aio_context;
4209     BlockJob *bjob = find_block_job(id, &aio_context, errp);
4210     Job *job;
4211 
4212     if (!bjob) {
4213         return;
4214     }
4215 
4216     trace_qmp_block_job_dismiss(bjob);
4217     job = &bjob->job;
4218     job_dismiss(&job, errp);
4219     aio_context_release(aio_context);
4220 }
4221 
4222 void qmp_change_backing_file(const char *device,
4223                              const char *image_node_name,
4224                              const char *backing_file,
4225                              Error **errp)
4226 {
4227     BlockDriverState *bs = NULL;
4228     AioContext *aio_context;
4229     BlockDriverState *image_bs = NULL;
4230     Error *local_err = NULL;
4231     bool ro;
4232     int ret;
4233 
4234     bs = qmp_get_root_bs(device, errp);
4235     if (!bs) {
4236         return;
4237     }
4238 
4239     aio_context = bdrv_get_aio_context(bs);
4240     aio_context_acquire(aio_context);
4241 
4242     image_bs = bdrv_lookup_bs(NULL, image_node_name, &local_err);
4243     if (local_err) {
4244         error_propagate(errp, local_err);
4245         goto out;
4246     }
4247 
4248     if (!image_bs) {
4249         error_setg(errp, "image file not found");
4250         goto out;
4251     }
4252 
4253     if (bdrv_find_base(image_bs) == image_bs) {
4254         error_setg(errp, "not allowing backing file change on an image "
4255                          "without a backing file");
4256         goto out;
4257     }
4258 
4259     /* even though we are not necessarily operating on bs, we need it to
4260      * determine if block ops are currently prohibited on the chain */
4261     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_CHANGE, errp)) {
4262         goto out;
4263     }
4264 
4265     /* final sanity check */
4266     if (!bdrv_chain_contains(bs, image_bs)) {
4267         error_setg(errp, "'%s' and image file are not in the same chain",
4268                    device);
4269         goto out;
4270     }
4271 
4272     /* if not r/w, reopen to make r/w */
4273     ro = bdrv_is_read_only(image_bs);
4274 
4275     if (ro) {
4276         if (bdrv_reopen_set_read_only(image_bs, false, errp) != 0) {
4277             goto out;
4278         }
4279     }
4280 
4281     ret = bdrv_change_backing_file(image_bs, backing_file,
4282                                image_bs->drv ? image_bs->drv->format_name : "");
4283 
4284     if (ret < 0) {
4285         error_setg_errno(errp, -ret, "Could not change backing file to '%s'",
4286                          backing_file);
4287         /* don't exit here, so we can try to restore open flags if
4288          * appropriate */
4289     }
4290 
4291     if (ro) {
4292         bdrv_reopen_set_read_only(image_bs, true, &local_err);
4293         error_propagate(errp, local_err);
4294     }
4295 
4296 out:
4297     aio_context_release(aio_context);
4298 }
4299 
4300 void hmp_drive_add_node(Monitor *mon, const char *optstr)
4301 {
4302     QemuOpts *opts;
4303     QDict *qdict;
4304     Error *local_err = NULL;
4305 
4306     opts = qemu_opts_parse_noisily(&qemu_drive_opts, optstr, false);
4307     if (!opts) {
4308         return;
4309     }
4310 
4311     qdict = qemu_opts_to_qdict(opts, NULL);
4312 
4313     if (!qdict_get_try_str(qdict, "node-name")) {
4314         qobject_unref(qdict);
4315         error_report("'node-name' needs to be specified");
4316         goto out;
4317     }
4318 
4319     BlockDriverState *bs = bds_tree_init(qdict, &local_err);
4320     if (!bs) {
4321         error_report_err(local_err);
4322         goto out;
4323     }
4324 
4325     QTAILQ_INSERT_TAIL(&monitor_bdrv_states, bs, monitor_list);
4326 
4327 out:
4328     qemu_opts_del(opts);
4329 }
4330 
4331 void qmp_blockdev_add(BlockdevOptions *options, Error **errp)
4332 {
4333     BlockDriverState *bs;
4334     QObject *obj;
4335     Visitor *v = qobject_output_visitor_new(&obj);
4336     QDict *qdict;
4337     Error *local_err = NULL;
4338 
4339     visit_type_BlockdevOptions(v, NULL, &options, &local_err);
4340     if (local_err) {
4341         error_propagate(errp, local_err);
4342         goto fail;
4343     }
4344 
4345     visit_complete(v, &obj);
4346     qdict = qobject_to(QDict, obj);
4347 
4348     qdict_flatten(qdict);
4349 
4350     if (!qdict_get_try_str(qdict, "node-name")) {
4351         error_setg(errp, "'node-name' must be specified for the root node");
4352         goto fail;
4353     }
4354 
4355     bs = bds_tree_init(qdict, errp);
4356     if (!bs) {
4357         goto fail;
4358     }
4359 
4360     QTAILQ_INSERT_TAIL(&monitor_bdrv_states, bs, monitor_list);
4361 
4362 fail:
4363     visit_free(v);
4364 }
4365 
4366 void qmp_x_blockdev_reopen(BlockdevOptions *options, Error **errp)
4367 {
4368     BlockDriverState *bs;
4369     AioContext *ctx;
4370     QObject *obj;
4371     Visitor *v = qobject_output_visitor_new(&obj);
4372     Error *local_err = NULL;
4373     BlockReopenQueue *queue;
4374     QDict *qdict;
4375 
4376     /* Check for the selected node name */
4377     if (!options->has_node_name) {
4378         error_setg(errp, "Node name not specified");
4379         goto fail;
4380     }
4381 
4382     bs = bdrv_find_node(options->node_name);
4383     if (!bs) {
4384         error_setg(errp, "Cannot find node named '%s'", options->node_name);
4385         goto fail;
4386     }
4387 
4388     /* Put all options in a QDict and flatten it */
4389     visit_type_BlockdevOptions(v, NULL, &options, &local_err);
4390     if (local_err) {
4391         error_propagate(errp, local_err);
4392         goto fail;
4393     }
4394 
4395     visit_complete(v, &obj);
4396     qdict = qobject_to(QDict, obj);
4397 
4398     qdict_flatten(qdict);
4399 
4400     /* Perform the reopen operation */
4401     ctx = bdrv_get_aio_context(bs);
4402     aio_context_acquire(ctx);
4403     bdrv_subtree_drained_begin(bs);
4404     queue = bdrv_reopen_queue(NULL, bs, qdict, false);
4405     bdrv_reopen_multiple(queue, errp);
4406     bdrv_subtree_drained_end(bs);
4407     aio_context_release(ctx);
4408 
4409 fail:
4410     visit_free(v);
4411 }
4412 
4413 void qmp_blockdev_del(const char *node_name, Error **errp)
4414 {
4415     AioContext *aio_context;
4416     BlockDriverState *bs;
4417 
4418     bs = bdrv_find_node(node_name);
4419     if (!bs) {
4420         error_setg(errp, "Cannot find node %s", node_name);
4421         return;
4422     }
4423     if (bdrv_has_blk(bs)) {
4424         error_setg(errp, "Node %s is in use", node_name);
4425         return;
4426     }
4427     aio_context = bdrv_get_aio_context(bs);
4428     aio_context_acquire(aio_context);
4429 
4430     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_DRIVE_DEL, errp)) {
4431         goto out;
4432     }
4433 
4434     if (!QTAILQ_IN_USE(bs, monitor_list)) {
4435         error_setg(errp, "Node %s is not owned by the monitor",
4436                    bs->node_name);
4437         goto out;
4438     }
4439 
4440     if (bs->refcnt > 1) {
4441         error_setg(errp, "Block device %s is in use",
4442                    bdrv_get_device_or_node_name(bs));
4443         goto out;
4444     }
4445 
4446     QTAILQ_REMOVE(&monitor_bdrv_states, bs, monitor_list);
4447     bdrv_unref(bs);
4448 
4449 out:
4450     aio_context_release(aio_context);
4451 }
4452 
4453 static BdrvChild *bdrv_find_child(BlockDriverState *parent_bs,
4454                                   const char *child_name)
4455 {
4456     BdrvChild *child;
4457 
4458     QLIST_FOREACH(child, &parent_bs->children, next) {
4459         if (strcmp(child->name, child_name) == 0) {
4460             return child;
4461         }
4462     }
4463 
4464     return NULL;
4465 }
4466 
4467 void qmp_x_blockdev_change(const char *parent, bool has_child,
4468                            const char *child, bool has_node,
4469                            const char *node, Error **errp)
4470 {
4471     BlockDriverState *parent_bs, *new_bs = NULL;
4472     BdrvChild *p_child;
4473 
4474     parent_bs = bdrv_lookup_bs(parent, parent, errp);
4475     if (!parent_bs) {
4476         return;
4477     }
4478 
4479     if (has_child == has_node) {
4480         if (has_child) {
4481             error_setg(errp, "The parameters child and node are in conflict");
4482         } else {
4483             error_setg(errp, "Either child or node must be specified");
4484         }
4485         return;
4486     }
4487 
4488     if (has_child) {
4489         p_child = bdrv_find_child(parent_bs, child);
4490         if (!p_child) {
4491             error_setg(errp, "Node '%s' does not have child '%s'",
4492                        parent, child);
4493             return;
4494         }
4495         bdrv_del_child(parent_bs, p_child, errp);
4496     }
4497 
4498     if (has_node) {
4499         new_bs = bdrv_find_node(node);
4500         if (!new_bs) {
4501             error_setg(errp, "Node '%s' not found", node);
4502             return;
4503         }
4504         bdrv_add_child(parent_bs, new_bs, errp);
4505     }
4506 }
4507 
4508 BlockJobInfoList *qmp_query_block_jobs(Error **errp)
4509 {
4510     BlockJobInfoList *head = NULL, **p_next = &head;
4511     BlockJob *job;
4512 
4513     for (job = block_job_next(NULL); job; job = block_job_next(job)) {
4514         BlockJobInfoList *elem;
4515         AioContext *aio_context;
4516 
4517         if (block_job_is_internal(job)) {
4518             continue;
4519         }
4520         elem = g_new0(BlockJobInfoList, 1);
4521         aio_context = blk_get_aio_context(job->blk);
4522         aio_context_acquire(aio_context);
4523         elem->value = block_job_query(job, errp);
4524         aio_context_release(aio_context);
4525         if (!elem->value) {
4526             g_free(elem);
4527             qapi_free_BlockJobInfoList(head);
4528             return NULL;
4529         }
4530         *p_next = elem;
4531         p_next = &elem->next;
4532     }
4533 
4534     return head;
4535 }
4536 
4537 void qmp_x_blockdev_set_iothread(const char *node_name, StrOrNull *iothread,
4538                                  bool has_force, bool force, Error **errp)
4539 {
4540     AioContext *old_context;
4541     AioContext *new_context;
4542     BlockDriverState *bs;
4543 
4544     bs = bdrv_find_node(node_name);
4545     if (!bs) {
4546         error_setg(errp, "Cannot find node %s", node_name);
4547         return;
4548     }
4549 
4550     /* Protects against accidents. */
4551     if (!(has_force && force) && bdrv_has_blk(bs)) {
4552         error_setg(errp, "Node %s is associated with a BlockBackend and could "
4553                          "be in use (use force=true to override this check)",
4554                          node_name);
4555         return;
4556     }
4557 
4558     if (iothread->type == QTYPE_QSTRING) {
4559         IOThread *obj = iothread_by_id(iothread->u.s);
4560         if (!obj) {
4561             error_setg(errp, "Cannot find iothread %s", iothread->u.s);
4562             return;
4563         }
4564 
4565         new_context = iothread_get_aio_context(obj);
4566     } else {
4567         new_context = qemu_get_aio_context();
4568     }
4569 
4570     old_context = bdrv_get_aio_context(bs);
4571     aio_context_acquire(old_context);
4572 
4573     bdrv_try_set_aio_context(bs, new_context, errp);
4574 
4575     aio_context_release(old_context);
4576 }
4577 
4578 void qmp_block_latency_histogram_set(
4579     const char *id,
4580     bool has_boundaries, uint64List *boundaries,
4581     bool has_boundaries_read, uint64List *boundaries_read,
4582     bool has_boundaries_write, uint64List *boundaries_write,
4583     bool has_boundaries_flush, uint64List *boundaries_flush,
4584     Error **errp)
4585 {
4586     BlockBackend *blk = qmp_get_blk(NULL, id, errp);
4587     BlockAcctStats *stats;
4588     int ret;
4589 
4590     if (!blk) {
4591         return;
4592     }
4593 
4594     stats = blk_get_stats(blk);
4595 
4596     if (!has_boundaries && !has_boundaries_read && !has_boundaries_write &&
4597         !has_boundaries_flush)
4598     {
4599         block_latency_histograms_clear(stats);
4600         return;
4601     }
4602 
4603     if (has_boundaries || has_boundaries_read) {
4604         ret = block_latency_histogram_set(
4605             stats, BLOCK_ACCT_READ,
4606             has_boundaries_read ? boundaries_read : boundaries);
4607         if (ret) {
4608             error_setg(errp, "Device '%s' set read boundaries fail", id);
4609             return;
4610         }
4611     }
4612 
4613     if (has_boundaries || has_boundaries_write) {
4614         ret = block_latency_histogram_set(
4615             stats, BLOCK_ACCT_WRITE,
4616             has_boundaries_write ? boundaries_write : boundaries);
4617         if (ret) {
4618             error_setg(errp, "Device '%s' set write boundaries fail", id);
4619             return;
4620         }
4621     }
4622 
4623     if (has_boundaries || has_boundaries_flush) {
4624         ret = block_latency_histogram_set(
4625             stats, BLOCK_ACCT_FLUSH,
4626             has_boundaries_flush ? boundaries_flush : boundaries);
4627         if (ret) {
4628             error_setg(errp, "Device '%s' set flush boundaries fail", id);
4629             return;
4630         }
4631     }
4632 }
4633 
4634 QemuOptsList qemu_common_drive_opts = {
4635     .name = "drive",
4636     .head = QTAILQ_HEAD_INITIALIZER(qemu_common_drive_opts.head),
4637     .desc = {
4638         {
4639             .name = "snapshot",
4640             .type = QEMU_OPT_BOOL,
4641             .help = "enable/disable snapshot mode",
4642         },{
4643             .name = "aio",
4644             .type = QEMU_OPT_STRING,
4645             .help = "host AIO implementation (threads, native)",
4646         },{
4647             .name = BDRV_OPT_CACHE_WB,
4648             .type = QEMU_OPT_BOOL,
4649             .help = "Enable writeback mode",
4650         },{
4651             .name = "format",
4652             .type = QEMU_OPT_STRING,
4653             .help = "disk format (raw, qcow2, ...)",
4654         },{
4655             .name = "rerror",
4656             .type = QEMU_OPT_STRING,
4657             .help = "read error action",
4658         },{
4659             .name = "werror",
4660             .type = QEMU_OPT_STRING,
4661             .help = "write error action",
4662         },{
4663             .name = BDRV_OPT_READ_ONLY,
4664             .type = QEMU_OPT_BOOL,
4665             .help = "open drive file as read-only",
4666         },
4667 
4668         THROTTLE_OPTS,
4669 
4670         {
4671             .name = "throttling.group",
4672             .type = QEMU_OPT_STRING,
4673             .help = "name of the block throttling group",
4674         },{
4675             .name = "copy-on-read",
4676             .type = QEMU_OPT_BOOL,
4677             .help = "copy read data from backing file into image file",
4678         },{
4679             .name = "detect-zeroes",
4680             .type = QEMU_OPT_STRING,
4681             .help = "try to optimize zero writes (off, on, unmap)",
4682         },{
4683             .name = "stats-account-invalid",
4684             .type = QEMU_OPT_BOOL,
4685             .help = "whether to account for invalid I/O operations "
4686                     "in the statistics",
4687         },{
4688             .name = "stats-account-failed",
4689             .type = QEMU_OPT_BOOL,
4690             .help = "whether to account for failed I/O operations "
4691                     "in the statistics",
4692         },
4693         { /* end of list */ }
4694     },
4695 };
4696 
4697 QemuOptsList qemu_drive_opts = {
4698     .name = "drive",
4699     .head = QTAILQ_HEAD_INITIALIZER(qemu_drive_opts.head),
4700     .desc = {
4701         /*
4702          * no elements => accept any params
4703          * validation will happen later
4704          */
4705         { /* end of list */ }
4706     },
4707 };
4708