xref: /openbmc/qemu/blockdev.c (revision cae41fda)
1 /*
2  * QEMU host block devices
3  *
4  * Copyright (c) 2003-2008 Fabrice Bellard
5  *
6  * This work is licensed under the terms of the GNU GPL, version 2 or
7  * later.  See the COPYING file in the top-level directory.
8  *
9  * This file incorporates work covered by the following copyright and
10  * permission notice:
11  *
12  * Copyright (c) 2003-2008 Fabrice Bellard
13  *
14  * Permission is hereby granted, free of charge, to any person obtaining a copy
15  * of this software and associated documentation files (the "Software"), to deal
16  * in the Software without restriction, including without limitation the rights
17  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
18  * copies of the Software, and to permit persons to whom the Software is
19  * furnished to do so, subject to the following conditions:
20  *
21  * The above copyright notice and this permission notice shall be included in
22  * all copies or substantial portions of the Software.
23  *
24  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
25  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
26  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
27  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
28  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
29  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
30  * THE SOFTWARE.
31  */
32 
33 #include "qemu/osdep.h"
34 #include "sysemu/block-backend.h"
35 #include "sysemu/blockdev.h"
36 #include "hw/block/block.h"
37 #include "block/blockjob.h"
38 #include "block/throttle-groups.h"
39 #include "monitor/monitor.h"
40 #include "qemu/error-report.h"
41 #include "qemu/option.h"
42 #include "qemu/config-file.h"
43 #include "qapi/qmp/types.h"
44 #include "qapi-visit.h"
45 #include "qapi/qmp/qerror.h"
46 #include "qapi/qobject-output-visitor.h"
47 #include "qapi/util.h"
48 #include "sysemu/sysemu.h"
49 #include "block/block_int.h"
50 #include "qmp-commands.h"
51 #include "trace.h"
52 #include "sysemu/arch_init.h"
53 #include "qemu/cutils.h"
54 #include "qemu/help_option.h"
55 
56 static QTAILQ_HEAD(, BlockDriverState) monitor_bdrv_states =
57     QTAILQ_HEAD_INITIALIZER(monitor_bdrv_states);
58 
59 static int do_open_tray(const char *blk_name, const char *qdev_id,
60                         bool force, Error **errp);
61 
62 static const char *const if_name[IF_COUNT] = {
63     [IF_NONE] = "none",
64     [IF_IDE] = "ide",
65     [IF_SCSI] = "scsi",
66     [IF_FLOPPY] = "floppy",
67     [IF_PFLASH] = "pflash",
68     [IF_MTD] = "mtd",
69     [IF_SD] = "sd",
70     [IF_VIRTIO] = "virtio",
71     [IF_XEN] = "xen",
72 };
73 
74 static int if_max_devs[IF_COUNT] = {
75     /*
76      * Do not change these numbers!  They govern how drive option
77      * index maps to unit and bus.  That mapping is ABI.
78      *
79      * All controllers used to implement if=T drives need to support
80      * if_max_devs[T] units, for any T with if_max_devs[T] != 0.
81      * Otherwise, some index values map to "impossible" bus, unit
82      * values.
83      *
84      * For instance, if you change [IF_SCSI] to 255, -drive
85      * if=scsi,index=12 no longer means bus=1,unit=5, but
86      * bus=0,unit=12.  With an lsi53c895a controller (7 units max),
87      * the drive can't be set up.  Regression.
88      */
89     [IF_IDE] = 2,
90     [IF_SCSI] = 7,
91 };
92 
93 /**
94  * Boards may call this to offer board-by-board overrides
95  * of the default, global values.
96  */
97 void override_max_devs(BlockInterfaceType type, int max_devs)
98 {
99     BlockBackend *blk;
100     DriveInfo *dinfo;
101 
102     if (max_devs <= 0) {
103         return;
104     }
105 
106     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
107         dinfo = blk_legacy_dinfo(blk);
108         if (dinfo->type == type) {
109             fprintf(stderr, "Cannot override units-per-bus property of"
110                     " the %s interface, because a drive of that type has"
111                     " already been added.\n", if_name[type]);
112             g_assert_not_reached();
113         }
114     }
115 
116     if_max_devs[type] = max_devs;
117 }
118 
119 /*
120  * We automatically delete the drive when a device using it gets
121  * unplugged.  Questionable feature, but we can't just drop it.
122  * Device models call blockdev_mark_auto_del() to schedule the
123  * automatic deletion, and generic qdev code calls blockdev_auto_del()
124  * when deletion is actually safe.
125  */
126 void blockdev_mark_auto_del(BlockBackend *blk)
127 {
128     DriveInfo *dinfo = blk_legacy_dinfo(blk);
129     BlockDriverState *bs = blk_bs(blk);
130     AioContext *aio_context;
131 
132     if (!dinfo) {
133         return;
134     }
135 
136     if (bs) {
137         aio_context = bdrv_get_aio_context(bs);
138         aio_context_acquire(aio_context);
139 
140         if (bs->job) {
141             block_job_cancel(bs->job);
142         }
143 
144         aio_context_release(aio_context);
145     }
146 
147     dinfo->auto_del = 1;
148 }
149 
150 void blockdev_auto_del(BlockBackend *blk)
151 {
152     DriveInfo *dinfo = blk_legacy_dinfo(blk);
153 
154     if (dinfo && dinfo->auto_del) {
155         monitor_remove_blk(blk);
156         blk_unref(blk);
157     }
158 }
159 
160 /**
161  * Returns the current mapping of how many units per bus
162  * a particular interface can support.
163  *
164  *  A positive integer indicates n units per bus.
165  *  0 implies the mapping has not been established.
166  * -1 indicates an invalid BlockInterfaceType was given.
167  */
168 int drive_get_max_devs(BlockInterfaceType type)
169 {
170     if (type >= IF_IDE && type < IF_COUNT) {
171         return if_max_devs[type];
172     }
173 
174     return -1;
175 }
176 
177 static int drive_index_to_bus_id(BlockInterfaceType type, int index)
178 {
179     int max_devs = if_max_devs[type];
180     return max_devs ? index / max_devs : 0;
181 }
182 
183 static int drive_index_to_unit_id(BlockInterfaceType type, int index)
184 {
185     int max_devs = if_max_devs[type];
186     return max_devs ? index % max_devs : index;
187 }
188 
189 QemuOpts *drive_def(const char *optstr)
190 {
191     return qemu_opts_parse_noisily(qemu_find_opts("drive"), optstr, false);
192 }
193 
194 QemuOpts *drive_add(BlockInterfaceType type, int index, const char *file,
195                     const char *optstr)
196 {
197     QemuOpts *opts;
198 
199     opts = drive_def(optstr);
200     if (!opts) {
201         return NULL;
202     }
203     if (type != IF_DEFAULT) {
204         qemu_opt_set(opts, "if", if_name[type], &error_abort);
205     }
206     if (index >= 0) {
207         qemu_opt_set_number(opts, "index", index, &error_abort);
208     }
209     if (file)
210         qemu_opt_set(opts, "file", file, &error_abort);
211     return opts;
212 }
213 
214 DriveInfo *drive_get(BlockInterfaceType type, int bus, int unit)
215 {
216     BlockBackend *blk;
217     DriveInfo *dinfo;
218 
219     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
220         dinfo = blk_legacy_dinfo(blk);
221         if (dinfo && dinfo->type == type
222             && dinfo->bus == bus && dinfo->unit == unit) {
223             return dinfo;
224         }
225     }
226 
227     return NULL;
228 }
229 
230 bool drive_check_orphaned(void)
231 {
232     BlockBackend *blk;
233     DriveInfo *dinfo;
234     bool rs = false;
235 
236     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
237         dinfo = blk_legacy_dinfo(blk);
238         /* If dinfo->bdrv->dev is NULL, it has no device attached. */
239         /* Unless this is a default drive, this may be an oversight. */
240         if (!blk_get_attached_dev(blk) && !dinfo->is_default &&
241             dinfo->type != IF_NONE) {
242             fprintf(stderr, "Warning: Orphaned drive without device: "
243                     "id=%s,file=%s,if=%s,bus=%d,unit=%d\n",
244                     blk_name(blk), blk_bs(blk) ? blk_bs(blk)->filename : "",
245                     if_name[dinfo->type], dinfo->bus, dinfo->unit);
246             rs = true;
247         }
248     }
249 
250     return rs;
251 }
252 
253 DriveInfo *drive_get_by_index(BlockInterfaceType type, int index)
254 {
255     return drive_get(type,
256                      drive_index_to_bus_id(type, index),
257                      drive_index_to_unit_id(type, index));
258 }
259 
260 int drive_get_max_bus(BlockInterfaceType type)
261 {
262     int max_bus;
263     BlockBackend *blk;
264     DriveInfo *dinfo;
265 
266     max_bus = -1;
267     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
268         dinfo = blk_legacy_dinfo(blk);
269         if (dinfo && dinfo->type == type && dinfo->bus > max_bus) {
270             max_bus = dinfo->bus;
271         }
272     }
273     return max_bus;
274 }
275 
276 /* Get a block device.  This should only be used for single-drive devices
277    (e.g. SD/Floppy/MTD).  Multi-disk devices (scsi/ide) should use the
278    appropriate bus.  */
279 DriveInfo *drive_get_next(BlockInterfaceType type)
280 {
281     static int next_block_unit[IF_COUNT];
282 
283     return drive_get(type, 0, next_block_unit[type]++);
284 }
285 
286 static void bdrv_format_print(void *opaque, const char *name)
287 {
288     error_printf(" %s", name);
289 }
290 
291 typedef struct {
292     QEMUBH *bh;
293     BlockDriverState *bs;
294 } BDRVPutRefBH;
295 
296 static int parse_block_error_action(const char *buf, bool is_read, Error **errp)
297 {
298     if (!strcmp(buf, "ignore")) {
299         return BLOCKDEV_ON_ERROR_IGNORE;
300     } else if (!is_read && !strcmp(buf, "enospc")) {
301         return BLOCKDEV_ON_ERROR_ENOSPC;
302     } else if (!strcmp(buf, "stop")) {
303         return BLOCKDEV_ON_ERROR_STOP;
304     } else if (!strcmp(buf, "report")) {
305         return BLOCKDEV_ON_ERROR_REPORT;
306     } else {
307         error_setg(errp, "'%s' invalid %s error action",
308                    buf, is_read ? "read" : "write");
309         return -1;
310     }
311 }
312 
313 static bool parse_stats_intervals(BlockAcctStats *stats, QList *intervals,
314                                   Error **errp)
315 {
316     const QListEntry *entry;
317     for (entry = qlist_first(intervals); entry; entry = qlist_next(entry)) {
318         switch (qobject_type(entry->value)) {
319 
320         case QTYPE_QSTRING: {
321             unsigned long long length;
322             const char *str = qstring_get_str(qobject_to_qstring(entry->value));
323             if (parse_uint_full(str, &length, 10) == 0 &&
324                 length > 0 && length <= UINT_MAX) {
325                 block_acct_add_interval(stats, (unsigned) length);
326             } else {
327                 error_setg(errp, "Invalid interval length: %s", str);
328                 return false;
329             }
330             break;
331         }
332 
333         case QTYPE_QINT: {
334             int64_t length = qint_get_int(qobject_to_qint(entry->value));
335             if (length > 0 && length <= UINT_MAX) {
336                 block_acct_add_interval(stats, (unsigned) length);
337             } else {
338                 error_setg(errp, "Invalid interval length: %" PRId64, length);
339                 return false;
340             }
341             break;
342         }
343 
344         default:
345             error_setg(errp, "The specification of stats-intervals is invalid");
346             return false;
347         }
348     }
349     return true;
350 }
351 
352 typedef enum { MEDIA_DISK, MEDIA_CDROM } DriveMediaType;
353 
354 /* All parameters but @opts are optional and may be set to NULL. */
355 static void extract_common_blockdev_options(QemuOpts *opts, int *bdrv_flags,
356     const char **throttling_group, ThrottleConfig *throttle_cfg,
357     BlockdevDetectZeroesOptions *detect_zeroes, Error **errp)
358 {
359     Error *local_error = NULL;
360     const char *aio;
361 
362     if (bdrv_flags) {
363         if (qemu_opt_get_bool(opts, "copy-on-read", false)) {
364             *bdrv_flags |= BDRV_O_COPY_ON_READ;
365         }
366 
367         if ((aio = qemu_opt_get(opts, "aio")) != NULL) {
368             if (!strcmp(aio, "native")) {
369                 *bdrv_flags |= BDRV_O_NATIVE_AIO;
370             } else if (!strcmp(aio, "threads")) {
371                 /* this is the default */
372             } else {
373                error_setg(errp, "invalid aio option");
374                return;
375             }
376         }
377     }
378 
379     /* disk I/O throttling */
380     if (throttling_group) {
381         *throttling_group = qemu_opt_get(opts, "throttling.group");
382     }
383 
384     if (throttle_cfg) {
385         throttle_config_init(throttle_cfg);
386         throttle_cfg->buckets[THROTTLE_BPS_TOTAL].avg =
387             qemu_opt_get_number(opts, "throttling.bps-total", 0);
388         throttle_cfg->buckets[THROTTLE_BPS_READ].avg  =
389             qemu_opt_get_number(opts, "throttling.bps-read", 0);
390         throttle_cfg->buckets[THROTTLE_BPS_WRITE].avg =
391             qemu_opt_get_number(opts, "throttling.bps-write", 0);
392         throttle_cfg->buckets[THROTTLE_OPS_TOTAL].avg =
393             qemu_opt_get_number(opts, "throttling.iops-total", 0);
394         throttle_cfg->buckets[THROTTLE_OPS_READ].avg =
395             qemu_opt_get_number(opts, "throttling.iops-read", 0);
396         throttle_cfg->buckets[THROTTLE_OPS_WRITE].avg =
397             qemu_opt_get_number(opts, "throttling.iops-write", 0);
398 
399         throttle_cfg->buckets[THROTTLE_BPS_TOTAL].max =
400             qemu_opt_get_number(opts, "throttling.bps-total-max", 0);
401         throttle_cfg->buckets[THROTTLE_BPS_READ].max  =
402             qemu_opt_get_number(opts, "throttling.bps-read-max", 0);
403         throttle_cfg->buckets[THROTTLE_BPS_WRITE].max =
404             qemu_opt_get_number(opts, "throttling.bps-write-max", 0);
405         throttle_cfg->buckets[THROTTLE_OPS_TOTAL].max =
406             qemu_opt_get_number(opts, "throttling.iops-total-max", 0);
407         throttle_cfg->buckets[THROTTLE_OPS_READ].max =
408             qemu_opt_get_number(opts, "throttling.iops-read-max", 0);
409         throttle_cfg->buckets[THROTTLE_OPS_WRITE].max =
410             qemu_opt_get_number(opts, "throttling.iops-write-max", 0);
411 
412         throttle_cfg->buckets[THROTTLE_BPS_TOTAL].burst_length =
413             qemu_opt_get_number(opts, "throttling.bps-total-max-length", 1);
414         throttle_cfg->buckets[THROTTLE_BPS_READ].burst_length  =
415             qemu_opt_get_number(opts, "throttling.bps-read-max-length", 1);
416         throttle_cfg->buckets[THROTTLE_BPS_WRITE].burst_length =
417             qemu_opt_get_number(opts, "throttling.bps-write-max-length", 1);
418         throttle_cfg->buckets[THROTTLE_OPS_TOTAL].burst_length =
419             qemu_opt_get_number(opts, "throttling.iops-total-max-length", 1);
420         throttle_cfg->buckets[THROTTLE_OPS_READ].burst_length =
421             qemu_opt_get_number(opts, "throttling.iops-read-max-length", 1);
422         throttle_cfg->buckets[THROTTLE_OPS_WRITE].burst_length =
423             qemu_opt_get_number(opts, "throttling.iops-write-max-length", 1);
424 
425         throttle_cfg->op_size =
426             qemu_opt_get_number(opts, "throttling.iops-size", 0);
427 
428         if (!throttle_is_valid(throttle_cfg, errp)) {
429             return;
430         }
431     }
432 
433     if (detect_zeroes) {
434         *detect_zeroes =
435             qapi_enum_parse(BlockdevDetectZeroesOptions_lookup,
436                             qemu_opt_get(opts, "detect-zeroes"),
437                             BLOCKDEV_DETECT_ZEROES_OPTIONS__MAX,
438                             BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF,
439                             &local_error);
440         if (local_error) {
441             error_propagate(errp, local_error);
442             return;
443         }
444     }
445 }
446 
447 /* Takes the ownership of bs_opts */
448 static BlockBackend *blockdev_init(const char *file, QDict *bs_opts,
449                                    Error **errp)
450 {
451     const char *buf;
452     int bdrv_flags = 0;
453     int on_read_error, on_write_error;
454     bool account_invalid, account_failed;
455     bool writethrough, read_only;
456     BlockBackend *blk;
457     BlockDriverState *bs;
458     ThrottleConfig cfg;
459     int snapshot = 0;
460     Error *error = NULL;
461     QemuOpts *opts;
462     QDict *interval_dict = NULL;
463     QList *interval_list = NULL;
464     const char *id;
465     BlockdevDetectZeroesOptions detect_zeroes =
466         BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF;
467     const char *throttling_group = NULL;
468 
469     /* Check common options by copying from bs_opts to opts, all other options
470      * stay in bs_opts for processing by bdrv_open(). */
471     id = qdict_get_try_str(bs_opts, "id");
472     opts = qemu_opts_create(&qemu_common_drive_opts, id, 1, &error);
473     if (error) {
474         error_propagate(errp, error);
475         goto err_no_opts;
476     }
477 
478     qemu_opts_absorb_qdict(opts, bs_opts, &error);
479     if (error) {
480         error_propagate(errp, error);
481         goto early_err;
482     }
483 
484     if (id) {
485         qdict_del(bs_opts, "id");
486     }
487 
488     /* extract parameters */
489     snapshot = qemu_opt_get_bool(opts, "snapshot", 0);
490 
491     account_invalid = qemu_opt_get_bool(opts, "stats-account-invalid", true);
492     account_failed = qemu_opt_get_bool(opts, "stats-account-failed", true);
493 
494     writethrough = !qemu_opt_get_bool(opts, BDRV_OPT_CACHE_WB, true);
495 
496     id = qemu_opts_id(opts);
497 
498     qdict_extract_subqdict(bs_opts, &interval_dict, "stats-intervals.");
499     qdict_array_split(interval_dict, &interval_list);
500 
501     if (qdict_size(interval_dict) != 0) {
502         error_setg(errp, "Invalid option stats-intervals.%s",
503                    qdict_first(interval_dict)->key);
504         goto early_err;
505     }
506 
507     extract_common_blockdev_options(opts, &bdrv_flags, &throttling_group, &cfg,
508                                     &detect_zeroes, &error);
509     if (error) {
510         error_propagate(errp, error);
511         goto early_err;
512     }
513 
514     if ((buf = qemu_opt_get(opts, "format")) != NULL) {
515         if (is_help_option(buf)) {
516             error_printf("Supported formats:");
517             bdrv_iterate_format(bdrv_format_print, NULL);
518             error_printf("\n");
519             goto early_err;
520         }
521 
522         if (qdict_haskey(bs_opts, "driver")) {
523             error_setg(errp, "Cannot specify both 'driver' and 'format'");
524             goto early_err;
525         }
526         qdict_put(bs_opts, "driver", qstring_from_str(buf));
527     }
528 
529     on_write_error = BLOCKDEV_ON_ERROR_ENOSPC;
530     if ((buf = qemu_opt_get(opts, "werror")) != NULL) {
531         on_write_error = parse_block_error_action(buf, 0, &error);
532         if (error) {
533             error_propagate(errp, error);
534             goto early_err;
535         }
536     }
537 
538     on_read_error = BLOCKDEV_ON_ERROR_REPORT;
539     if ((buf = qemu_opt_get(opts, "rerror")) != NULL) {
540         on_read_error = parse_block_error_action(buf, 1, &error);
541         if (error) {
542             error_propagate(errp, error);
543             goto early_err;
544         }
545     }
546 
547     if (snapshot) {
548         bdrv_flags |= BDRV_O_SNAPSHOT;
549     }
550 
551     read_only = qemu_opt_get_bool(opts, BDRV_OPT_READ_ONLY, false);
552 
553     /* init */
554     if ((!file || !*file) && !qdict_size(bs_opts)) {
555         BlockBackendRootState *blk_rs;
556 
557         blk = blk_new();
558         blk_rs = blk_get_root_state(blk);
559         blk_rs->open_flags    = bdrv_flags;
560         blk_rs->read_only     = read_only;
561         blk_rs->detect_zeroes = detect_zeroes;
562 
563         QDECREF(bs_opts);
564     } else {
565         if (file && !*file) {
566             file = NULL;
567         }
568 
569         /* bdrv_open() defaults to the values in bdrv_flags (for compatibility
570          * with other callers) rather than what we want as the real defaults.
571          * Apply the defaults here instead. */
572         qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_DIRECT, "off");
573         qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_NO_FLUSH, "off");
574         qdict_set_default_str(bs_opts, BDRV_OPT_READ_ONLY,
575                               read_only ? "on" : "off");
576         assert((bdrv_flags & BDRV_O_CACHE_MASK) == 0);
577 
578         if (runstate_check(RUN_STATE_INMIGRATE)) {
579             bdrv_flags |= BDRV_O_INACTIVE;
580         }
581 
582         blk = blk_new_open(file, NULL, bs_opts, bdrv_flags, errp);
583         if (!blk) {
584             goto err_no_bs_opts;
585         }
586         bs = blk_bs(blk);
587 
588         bs->detect_zeroes = detect_zeroes;
589 
590         if (bdrv_key_required(bs)) {
591             autostart = 0;
592         }
593 
594         block_acct_init(blk_get_stats(blk), account_invalid, account_failed);
595 
596         if (!parse_stats_intervals(blk_get_stats(blk), interval_list, errp)) {
597             blk_unref(blk);
598             blk = NULL;
599             goto err_no_bs_opts;
600         }
601     }
602 
603     /* disk I/O throttling */
604     if (throttle_enabled(&cfg)) {
605         if (!throttling_group) {
606             throttling_group = id;
607         }
608         blk_io_limits_enable(blk, throttling_group);
609         blk_set_io_limits(blk, &cfg);
610     }
611 
612     blk_set_enable_write_cache(blk, !writethrough);
613     blk_set_on_error(blk, on_read_error, on_write_error);
614 
615     if (!monitor_add_blk(blk, id, errp)) {
616         blk_unref(blk);
617         blk = NULL;
618         goto err_no_bs_opts;
619     }
620 
621 err_no_bs_opts:
622     qemu_opts_del(opts);
623     QDECREF(interval_dict);
624     QDECREF(interval_list);
625     return blk;
626 
627 early_err:
628     qemu_opts_del(opts);
629     QDECREF(interval_dict);
630     QDECREF(interval_list);
631 err_no_opts:
632     QDECREF(bs_opts);
633     return NULL;
634 }
635 
636 /* Takes the ownership of bs_opts */
637 static BlockDriverState *bds_tree_init(QDict *bs_opts, Error **errp)
638 {
639     int bdrv_flags = 0;
640 
641     /* bdrv_open() defaults to the values in bdrv_flags (for compatibility
642      * with other callers) rather than what we want as the real defaults.
643      * Apply the defaults here instead. */
644     qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_DIRECT, "off");
645     qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_NO_FLUSH, "off");
646     qdict_set_default_str(bs_opts, BDRV_OPT_READ_ONLY, "off");
647 
648     if (runstate_check(RUN_STATE_INMIGRATE)) {
649         bdrv_flags |= BDRV_O_INACTIVE;
650     }
651 
652     return bdrv_open(NULL, NULL, bs_opts, bdrv_flags, errp);
653 }
654 
655 void blockdev_close_all_bdrv_states(void)
656 {
657     BlockDriverState *bs, *next_bs;
658 
659     QTAILQ_FOREACH_SAFE(bs, &monitor_bdrv_states, monitor_list, next_bs) {
660         AioContext *ctx = bdrv_get_aio_context(bs);
661 
662         aio_context_acquire(ctx);
663         bdrv_unref(bs);
664         aio_context_release(ctx);
665     }
666 }
667 
668 /* Iterates over the list of monitor-owned BlockDriverStates */
669 BlockDriverState *bdrv_next_monitor_owned(BlockDriverState *bs)
670 {
671     return bs ? QTAILQ_NEXT(bs, monitor_list)
672               : QTAILQ_FIRST(&monitor_bdrv_states);
673 }
674 
675 static void qemu_opt_rename(QemuOpts *opts, const char *from, const char *to,
676                             Error **errp)
677 {
678     const char *value;
679 
680     value = qemu_opt_get(opts, from);
681     if (value) {
682         if (qemu_opt_find(opts, to)) {
683             error_setg(errp, "'%s' and its alias '%s' can't be used at the "
684                        "same time", to, from);
685             return;
686         }
687     }
688 
689     /* rename all items in opts */
690     while ((value = qemu_opt_get(opts, from))) {
691         qemu_opt_set(opts, to, value, &error_abort);
692         qemu_opt_unset(opts, from);
693     }
694 }
695 
696 QemuOptsList qemu_legacy_drive_opts = {
697     .name = "drive",
698     .head = QTAILQ_HEAD_INITIALIZER(qemu_legacy_drive_opts.head),
699     .desc = {
700         {
701             .name = "bus",
702             .type = QEMU_OPT_NUMBER,
703             .help = "bus number",
704         },{
705             .name = "unit",
706             .type = QEMU_OPT_NUMBER,
707             .help = "unit number (i.e. lun for scsi)",
708         },{
709             .name = "index",
710             .type = QEMU_OPT_NUMBER,
711             .help = "index number",
712         },{
713             .name = "media",
714             .type = QEMU_OPT_STRING,
715             .help = "media type (disk, cdrom)",
716         },{
717             .name = "if",
718             .type = QEMU_OPT_STRING,
719             .help = "interface (ide, scsi, sd, mtd, floppy, pflash, virtio)",
720         },{
721             .name = "cyls",
722             .type = QEMU_OPT_NUMBER,
723             .help = "number of cylinders (ide disk geometry)",
724         },{
725             .name = "heads",
726             .type = QEMU_OPT_NUMBER,
727             .help = "number of heads (ide disk geometry)",
728         },{
729             .name = "secs",
730             .type = QEMU_OPT_NUMBER,
731             .help = "number of sectors (ide disk geometry)",
732         },{
733             .name = "trans",
734             .type = QEMU_OPT_STRING,
735             .help = "chs translation (auto, lba, none)",
736         },{
737             .name = "boot",
738             .type = QEMU_OPT_BOOL,
739             .help = "(deprecated, ignored)",
740         },{
741             .name = "addr",
742             .type = QEMU_OPT_STRING,
743             .help = "pci address (virtio only)",
744         },{
745             .name = "serial",
746             .type = QEMU_OPT_STRING,
747             .help = "disk serial number",
748         },{
749             .name = "file",
750             .type = QEMU_OPT_STRING,
751             .help = "file name",
752         },
753 
754         /* Options that are passed on, but have special semantics with -drive */
755         {
756             .name = BDRV_OPT_READ_ONLY,
757             .type = QEMU_OPT_BOOL,
758             .help = "open drive file as read-only",
759         },{
760             .name = "rerror",
761             .type = QEMU_OPT_STRING,
762             .help = "read error action",
763         },{
764             .name = "werror",
765             .type = QEMU_OPT_STRING,
766             .help = "write error action",
767         },{
768             .name = "copy-on-read",
769             .type = QEMU_OPT_BOOL,
770             .help = "copy read data from backing file into image file",
771         },
772 
773         { /* end of list */ }
774     },
775 };
776 
777 DriveInfo *drive_new(QemuOpts *all_opts, BlockInterfaceType block_default_type)
778 {
779     const char *value;
780     BlockBackend *blk;
781     DriveInfo *dinfo = NULL;
782     QDict *bs_opts;
783     QemuOpts *legacy_opts;
784     DriveMediaType media = MEDIA_DISK;
785     BlockInterfaceType type;
786     int cyls, heads, secs, translation;
787     int max_devs, bus_id, unit_id, index;
788     const char *devaddr;
789     const char *werror, *rerror;
790     bool read_only = false;
791     bool copy_on_read;
792     const char *serial;
793     const char *filename;
794     Error *local_err = NULL;
795     int i;
796 
797     /* Change legacy command line options into QMP ones */
798     static const struct {
799         const char *from;
800         const char *to;
801     } opt_renames[] = {
802         { "iops",           "throttling.iops-total" },
803         { "iops_rd",        "throttling.iops-read" },
804         { "iops_wr",        "throttling.iops-write" },
805 
806         { "bps",            "throttling.bps-total" },
807         { "bps_rd",         "throttling.bps-read" },
808         { "bps_wr",         "throttling.bps-write" },
809 
810         { "iops_max",       "throttling.iops-total-max" },
811         { "iops_rd_max",    "throttling.iops-read-max" },
812         { "iops_wr_max",    "throttling.iops-write-max" },
813 
814         { "bps_max",        "throttling.bps-total-max" },
815         { "bps_rd_max",     "throttling.bps-read-max" },
816         { "bps_wr_max",     "throttling.bps-write-max" },
817 
818         { "iops_size",      "throttling.iops-size" },
819 
820         { "group",          "throttling.group" },
821 
822         { "readonly",       BDRV_OPT_READ_ONLY },
823     };
824 
825     for (i = 0; i < ARRAY_SIZE(opt_renames); i++) {
826         qemu_opt_rename(all_opts, opt_renames[i].from, opt_renames[i].to,
827                         &local_err);
828         if (local_err) {
829             error_report_err(local_err);
830             return NULL;
831         }
832     }
833 
834     value = qemu_opt_get(all_opts, "cache");
835     if (value) {
836         int flags = 0;
837         bool writethrough;
838 
839         if (bdrv_parse_cache_mode(value, &flags, &writethrough) != 0) {
840             error_report("invalid cache option");
841             return NULL;
842         }
843 
844         /* Specific options take precedence */
845         if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_WB)) {
846             qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_WB,
847                               !writethrough, &error_abort);
848         }
849         if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_DIRECT)) {
850             qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_DIRECT,
851                               !!(flags & BDRV_O_NOCACHE), &error_abort);
852         }
853         if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_NO_FLUSH)) {
854             qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_NO_FLUSH,
855                               !!(flags & BDRV_O_NO_FLUSH), &error_abort);
856         }
857         qemu_opt_unset(all_opts, "cache");
858     }
859 
860     /* Get a QDict for processing the options */
861     bs_opts = qdict_new();
862     qemu_opts_to_qdict(all_opts, bs_opts);
863 
864     legacy_opts = qemu_opts_create(&qemu_legacy_drive_opts, NULL, 0,
865                                    &error_abort);
866     qemu_opts_absorb_qdict(legacy_opts, bs_opts, &local_err);
867     if (local_err) {
868         error_report_err(local_err);
869         goto fail;
870     }
871 
872     /* Deprecated option boot=[on|off] */
873     if (qemu_opt_get(legacy_opts, "boot") != NULL) {
874         fprintf(stderr, "qemu-kvm: boot=on|off is deprecated and will be "
875                 "ignored. Future versions will reject this parameter. Please "
876                 "update your scripts.\n");
877     }
878 
879     /* Media type */
880     value = qemu_opt_get(legacy_opts, "media");
881     if (value) {
882         if (!strcmp(value, "disk")) {
883             media = MEDIA_DISK;
884         } else if (!strcmp(value, "cdrom")) {
885             media = MEDIA_CDROM;
886             read_only = true;
887         } else {
888             error_report("'%s' invalid media", value);
889             goto fail;
890         }
891     }
892 
893     /* copy-on-read is disabled with a warning for read-only devices */
894     read_only |= qemu_opt_get_bool(legacy_opts, BDRV_OPT_READ_ONLY, false);
895     copy_on_read = qemu_opt_get_bool(legacy_opts, "copy-on-read", false);
896 
897     if (read_only && copy_on_read) {
898         error_report("warning: disabling copy-on-read on read-only drive");
899         copy_on_read = false;
900     }
901 
902     qdict_put(bs_opts, BDRV_OPT_READ_ONLY,
903               qstring_from_str(read_only ? "on" : "off"));
904     qdict_put(bs_opts, "copy-on-read",
905               qstring_from_str(copy_on_read ? "on" :"off"));
906 
907     /* Controller type */
908     value = qemu_opt_get(legacy_opts, "if");
909     if (value) {
910         for (type = 0;
911              type < IF_COUNT && strcmp(value, if_name[type]);
912              type++) {
913         }
914         if (type == IF_COUNT) {
915             error_report("unsupported bus type '%s'", value);
916             goto fail;
917         }
918     } else {
919         type = block_default_type;
920     }
921 
922     /* Geometry */
923     cyls  = qemu_opt_get_number(legacy_opts, "cyls", 0);
924     heads = qemu_opt_get_number(legacy_opts, "heads", 0);
925     secs  = qemu_opt_get_number(legacy_opts, "secs", 0);
926 
927     if (cyls || heads || secs) {
928         if (cyls < 1) {
929             error_report("invalid physical cyls number");
930             goto fail;
931         }
932         if (heads < 1) {
933             error_report("invalid physical heads number");
934             goto fail;
935         }
936         if (secs < 1) {
937             error_report("invalid physical secs number");
938             goto fail;
939         }
940     }
941 
942     translation = BIOS_ATA_TRANSLATION_AUTO;
943     value = qemu_opt_get(legacy_opts, "trans");
944     if (value != NULL) {
945         if (!cyls) {
946             error_report("'%s' trans must be used with cyls, heads and secs",
947                          value);
948             goto fail;
949         }
950         if (!strcmp(value, "none")) {
951             translation = BIOS_ATA_TRANSLATION_NONE;
952         } else if (!strcmp(value, "lba")) {
953             translation = BIOS_ATA_TRANSLATION_LBA;
954         } else if (!strcmp(value, "large")) {
955             translation = BIOS_ATA_TRANSLATION_LARGE;
956         } else if (!strcmp(value, "rechs")) {
957             translation = BIOS_ATA_TRANSLATION_RECHS;
958         } else if (!strcmp(value, "auto")) {
959             translation = BIOS_ATA_TRANSLATION_AUTO;
960         } else {
961             error_report("'%s' invalid translation type", value);
962             goto fail;
963         }
964     }
965 
966     if (media == MEDIA_CDROM) {
967         if (cyls || secs || heads) {
968             error_report("CHS can't be set with media=cdrom");
969             goto fail;
970         }
971     }
972 
973     /* Device address specified by bus/unit or index.
974      * If none was specified, try to find the first free one. */
975     bus_id  = qemu_opt_get_number(legacy_opts, "bus", 0);
976     unit_id = qemu_opt_get_number(legacy_opts, "unit", -1);
977     index   = qemu_opt_get_number(legacy_opts, "index", -1);
978 
979     max_devs = if_max_devs[type];
980 
981     if (index != -1) {
982         if (bus_id != 0 || unit_id != -1) {
983             error_report("index cannot be used with bus and unit");
984             goto fail;
985         }
986         bus_id = drive_index_to_bus_id(type, index);
987         unit_id = drive_index_to_unit_id(type, index);
988     }
989 
990     if (unit_id == -1) {
991        unit_id = 0;
992        while (drive_get(type, bus_id, unit_id) != NULL) {
993            unit_id++;
994            if (max_devs && unit_id >= max_devs) {
995                unit_id -= max_devs;
996                bus_id++;
997            }
998        }
999     }
1000 
1001     if (max_devs && unit_id >= max_devs) {
1002         error_report("unit %d too big (max is %d)", unit_id, max_devs - 1);
1003         goto fail;
1004     }
1005 
1006     if (drive_get(type, bus_id, unit_id) != NULL) {
1007         error_report("drive with bus=%d, unit=%d (index=%d) exists",
1008                      bus_id, unit_id, index);
1009         goto fail;
1010     }
1011 
1012     /* Serial number */
1013     serial = qemu_opt_get(legacy_opts, "serial");
1014 
1015     /* no id supplied -> create one */
1016     if (qemu_opts_id(all_opts) == NULL) {
1017         char *new_id;
1018         const char *mediastr = "";
1019         if (type == IF_IDE || type == IF_SCSI) {
1020             mediastr = (media == MEDIA_CDROM) ? "-cd" : "-hd";
1021         }
1022         if (max_devs) {
1023             new_id = g_strdup_printf("%s%i%s%i", if_name[type], bus_id,
1024                                      mediastr, unit_id);
1025         } else {
1026             new_id = g_strdup_printf("%s%s%i", if_name[type],
1027                                      mediastr, unit_id);
1028         }
1029         qdict_put(bs_opts, "id", qstring_from_str(new_id));
1030         g_free(new_id);
1031     }
1032 
1033     /* Add virtio block device */
1034     devaddr = qemu_opt_get(legacy_opts, "addr");
1035     if (devaddr && type != IF_VIRTIO) {
1036         error_report("addr is not supported by this bus type");
1037         goto fail;
1038     }
1039 
1040     if (type == IF_VIRTIO) {
1041         QemuOpts *devopts;
1042         devopts = qemu_opts_create(qemu_find_opts("device"), NULL, 0,
1043                                    &error_abort);
1044         if (arch_type == QEMU_ARCH_S390X) {
1045             qemu_opt_set(devopts, "driver", "virtio-blk-ccw", &error_abort);
1046         } else {
1047             qemu_opt_set(devopts, "driver", "virtio-blk-pci", &error_abort);
1048         }
1049         qemu_opt_set(devopts, "drive", qdict_get_str(bs_opts, "id"),
1050                      &error_abort);
1051         if (devaddr) {
1052             qemu_opt_set(devopts, "addr", devaddr, &error_abort);
1053         }
1054     }
1055 
1056     filename = qemu_opt_get(legacy_opts, "file");
1057 
1058     /* Check werror/rerror compatibility with if=... */
1059     werror = qemu_opt_get(legacy_opts, "werror");
1060     if (werror != NULL) {
1061         if (type != IF_IDE && type != IF_SCSI && type != IF_VIRTIO &&
1062             type != IF_NONE) {
1063             error_report("werror is not supported by this bus type");
1064             goto fail;
1065         }
1066         qdict_put(bs_opts, "werror", qstring_from_str(werror));
1067     }
1068 
1069     rerror = qemu_opt_get(legacy_opts, "rerror");
1070     if (rerror != NULL) {
1071         if (type != IF_IDE && type != IF_VIRTIO && type != IF_SCSI &&
1072             type != IF_NONE) {
1073             error_report("rerror is not supported by this bus type");
1074             goto fail;
1075         }
1076         qdict_put(bs_opts, "rerror", qstring_from_str(rerror));
1077     }
1078 
1079     /* Actual block device init: Functionality shared with blockdev-add */
1080     blk = blockdev_init(filename, bs_opts, &local_err);
1081     bs_opts = NULL;
1082     if (!blk) {
1083         if (local_err) {
1084             error_report_err(local_err);
1085         }
1086         goto fail;
1087     } else {
1088         assert(!local_err);
1089     }
1090 
1091     /* Create legacy DriveInfo */
1092     dinfo = g_malloc0(sizeof(*dinfo));
1093     dinfo->opts = all_opts;
1094 
1095     dinfo->cyls = cyls;
1096     dinfo->heads = heads;
1097     dinfo->secs = secs;
1098     dinfo->trans = translation;
1099 
1100     dinfo->type = type;
1101     dinfo->bus = bus_id;
1102     dinfo->unit = unit_id;
1103     dinfo->devaddr = devaddr;
1104     dinfo->serial = g_strdup(serial);
1105 
1106     blk_set_legacy_dinfo(blk, dinfo);
1107 
1108     switch(type) {
1109     case IF_IDE:
1110     case IF_SCSI:
1111     case IF_XEN:
1112     case IF_NONE:
1113         dinfo->media_cd = media == MEDIA_CDROM;
1114         break;
1115     default:
1116         break;
1117     }
1118 
1119 fail:
1120     qemu_opts_del(legacy_opts);
1121     QDECREF(bs_opts);
1122     return dinfo;
1123 }
1124 
1125 static BlockDriverState *qmp_get_root_bs(const char *name, Error **errp)
1126 {
1127     BlockDriverState *bs;
1128 
1129     bs = bdrv_lookup_bs(name, name, errp);
1130     if (bs == NULL) {
1131         return NULL;
1132     }
1133 
1134     if (!bdrv_is_root_node(bs)) {
1135         error_setg(errp, "Need a root block node");
1136         return NULL;
1137     }
1138 
1139     if (!bdrv_is_inserted(bs)) {
1140         error_setg(errp, "Device has no medium");
1141         return NULL;
1142     }
1143 
1144     return bs;
1145 }
1146 
1147 static BlockBackend *qmp_get_blk(const char *blk_name, const char *qdev_id,
1148                                  Error **errp)
1149 {
1150     BlockBackend *blk;
1151 
1152     if (!blk_name == !qdev_id) {
1153         error_setg(errp, "Need exactly one of 'device' and 'id'");
1154         return NULL;
1155     }
1156 
1157     if (qdev_id) {
1158         blk = blk_by_qdev_id(qdev_id, errp);
1159     } else {
1160         blk = blk_by_name(blk_name);
1161         if (blk == NULL) {
1162             error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
1163                       "Device '%s' not found", blk_name);
1164         }
1165     }
1166 
1167     return blk;
1168 }
1169 
1170 void hmp_commit(Monitor *mon, const QDict *qdict)
1171 {
1172     const char *device = qdict_get_str(qdict, "device");
1173     BlockBackend *blk;
1174     int ret;
1175 
1176     if (!strcmp(device, "all")) {
1177         ret = blk_commit_all();
1178     } else {
1179         BlockDriverState *bs;
1180         AioContext *aio_context;
1181 
1182         blk = blk_by_name(device);
1183         if (!blk) {
1184             monitor_printf(mon, "Device '%s' not found\n", device);
1185             return;
1186         }
1187         if (!blk_is_available(blk)) {
1188             monitor_printf(mon, "Device '%s' has no medium\n", device);
1189             return;
1190         }
1191 
1192         bs = blk_bs(blk);
1193         aio_context = bdrv_get_aio_context(bs);
1194         aio_context_acquire(aio_context);
1195 
1196         ret = bdrv_commit(bs);
1197 
1198         aio_context_release(aio_context);
1199     }
1200     if (ret < 0) {
1201         monitor_printf(mon, "'commit' error for '%s': %s\n", device,
1202                        strerror(-ret));
1203     }
1204 }
1205 
1206 static void blockdev_do_action(TransactionAction *action, Error **errp)
1207 {
1208     TransactionActionList list;
1209 
1210     list.value = action;
1211     list.next = NULL;
1212     qmp_transaction(&list, false, NULL, errp);
1213 }
1214 
1215 void qmp_blockdev_snapshot_sync(bool has_device, const char *device,
1216                                 bool has_node_name, const char *node_name,
1217                                 const char *snapshot_file,
1218                                 bool has_snapshot_node_name,
1219                                 const char *snapshot_node_name,
1220                                 bool has_format, const char *format,
1221                                 bool has_mode, NewImageMode mode, Error **errp)
1222 {
1223     BlockdevSnapshotSync snapshot = {
1224         .has_device = has_device,
1225         .device = (char *) device,
1226         .has_node_name = has_node_name,
1227         .node_name = (char *) node_name,
1228         .snapshot_file = (char *) snapshot_file,
1229         .has_snapshot_node_name = has_snapshot_node_name,
1230         .snapshot_node_name = (char *) snapshot_node_name,
1231         .has_format = has_format,
1232         .format = (char *) format,
1233         .has_mode = has_mode,
1234         .mode = mode,
1235     };
1236     TransactionAction action = {
1237         .type = TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC,
1238         .u.blockdev_snapshot_sync.data = &snapshot,
1239     };
1240     blockdev_do_action(&action, errp);
1241 }
1242 
1243 void qmp_blockdev_snapshot(const char *node, const char *overlay,
1244                            Error **errp)
1245 {
1246     BlockdevSnapshot snapshot_data = {
1247         .node = (char *) node,
1248         .overlay = (char *) overlay
1249     };
1250     TransactionAction action = {
1251         .type = TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT,
1252         .u.blockdev_snapshot.data = &snapshot_data,
1253     };
1254     blockdev_do_action(&action, errp);
1255 }
1256 
1257 void qmp_blockdev_snapshot_internal_sync(const char *device,
1258                                          const char *name,
1259                                          Error **errp)
1260 {
1261     BlockdevSnapshotInternal snapshot = {
1262         .device = (char *) device,
1263         .name = (char *) name
1264     };
1265     TransactionAction action = {
1266         .type = TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC,
1267         .u.blockdev_snapshot_internal_sync.data = &snapshot,
1268     };
1269     blockdev_do_action(&action, errp);
1270 }
1271 
1272 SnapshotInfo *qmp_blockdev_snapshot_delete_internal_sync(const char *device,
1273                                                          bool has_id,
1274                                                          const char *id,
1275                                                          bool has_name,
1276                                                          const char *name,
1277                                                          Error **errp)
1278 {
1279     BlockDriverState *bs;
1280     AioContext *aio_context;
1281     QEMUSnapshotInfo sn;
1282     Error *local_err = NULL;
1283     SnapshotInfo *info = NULL;
1284     int ret;
1285 
1286     bs = qmp_get_root_bs(device, errp);
1287     if (!bs) {
1288         return NULL;
1289     }
1290     aio_context = bdrv_get_aio_context(bs);
1291     aio_context_acquire(aio_context);
1292 
1293     if (!has_id) {
1294         id = NULL;
1295     }
1296 
1297     if (!has_name) {
1298         name = NULL;
1299     }
1300 
1301     if (!id && !name) {
1302         error_setg(errp, "Name or id must be provided");
1303         goto out_aio_context;
1304     }
1305 
1306     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_INTERNAL_SNAPSHOT_DELETE, errp)) {
1307         goto out_aio_context;
1308     }
1309 
1310     ret = bdrv_snapshot_find_by_id_and_name(bs, id, name, &sn, &local_err);
1311     if (local_err) {
1312         error_propagate(errp, local_err);
1313         goto out_aio_context;
1314     }
1315     if (!ret) {
1316         error_setg(errp,
1317                    "Snapshot with id '%s' and name '%s' does not exist on "
1318                    "device '%s'",
1319                    STR_OR_NULL(id), STR_OR_NULL(name), device);
1320         goto out_aio_context;
1321     }
1322 
1323     bdrv_snapshot_delete(bs, id, name, &local_err);
1324     if (local_err) {
1325         error_propagate(errp, local_err);
1326         goto out_aio_context;
1327     }
1328 
1329     aio_context_release(aio_context);
1330 
1331     info = g_new0(SnapshotInfo, 1);
1332     info->id = g_strdup(sn.id_str);
1333     info->name = g_strdup(sn.name);
1334     info->date_nsec = sn.date_nsec;
1335     info->date_sec = sn.date_sec;
1336     info->vm_state_size = sn.vm_state_size;
1337     info->vm_clock_nsec = sn.vm_clock_nsec % 1000000000;
1338     info->vm_clock_sec = sn.vm_clock_nsec / 1000000000;
1339 
1340     return info;
1341 
1342 out_aio_context:
1343     aio_context_release(aio_context);
1344     return NULL;
1345 }
1346 
1347 /**
1348  * block_dirty_bitmap_lookup:
1349  * Return a dirty bitmap (if present), after validating
1350  * the node reference and bitmap names.
1351  *
1352  * @node: The name of the BDS node to search for bitmaps
1353  * @name: The name of the bitmap to search for
1354  * @pbs: Output pointer for BDS lookup, if desired. Can be NULL.
1355  * @paio: Output pointer for aio_context acquisition, if desired. Can be NULL.
1356  * @errp: Output pointer for error information. Can be NULL.
1357  *
1358  * @return: A bitmap object on success, or NULL on failure.
1359  */
1360 static BdrvDirtyBitmap *block_dirty_bitmap_lookup(const char *node,
1361                                                   const char *name,
1362                                                   BlockDriverState **pbs,
1363                                                   AioContext **paio,
1364                                                   Error **errp)
1365 {
1366     BlockDriverState *bs;
1367     BdrvDirtyBitmap *bitmap;
1368     AioContext *aio_context;
1369 
1370     if (!node) {
1371         error_setg(errp, "Node cannot be NULL");
1372         return NULL;
1373     }
1374     if (!name) {
1375         error_setg(errp, "Bitmap name cannot be NULL");
1376         return NULL;
1377     }
1378     bs = bdrv_lookup_bs(node, node, NULL);
1379     if (!bs) {
1380         error_setg(errp, "Node '%s' not found", node);
1381         return NULL;
1382     }
1383 
1384     aio_context = bdrv_get_aio_context(bs);
1385     aio_context_acquire(aio_context);
1386 
1387     bitmap = bdrv_find_dirty_bitmap(bs, name);
1388     if (!bitmap) {
1389         error_setg(errp, "Dirty bitmap '%s' not found", name);
1390         goto fail;
1391     }
1392 
1393     if (pbs) {
1394         *pbs = bs;
1395     }
1396     if (paio) {
1397         *paio = aio_context;
1398     } else {
1399         aio_context_release(aio_context);
1400     }
1401 
1402     return bitmap;
1403 
1404  fail:
1405     aio_context_release(aio_context);
1406     return NULL;
1407 }
1408 
1409 /* New and old BlockDriverState structs for atomic group operations */
1410 
1411 typedef struct BlkActionState BlkActionState;
1412 
1413 /**
1414  * BlkActionOps:
1415  * Table of operations that define an Action.
1416  *
1417  * @instance_size: Size of state struct, in bytes.
1418  * @prepare: Prepare the work, must NOT be NULL.
1419  * @commit: Commit the changes, can be NULL.
1420  * @abort: Abort the changes on fail, can be NULL.
1421  * @clean: Clean up resources after all transaction actions have called
1422  *         commit() or abort(). Can be NULL.
1423  *
1424  * Only prepare() may fail. In a single transaction, only one of commit() or
1425  * abort() will be called. clean() will always be called if it is present.
1426  */
1427 typedef struct BlkActionOps {
1428     size_t instance_size;
1429     void (*prepare)(BlkActionState *common, Error **errp);
1430     void (*commit)(BlkActionState *common);
1431     void (*abort)(BlkActionState *common);
1432     void (*clean)(BlkActionState *common);
1433 } BlkActionOps;
1434 
1435 /**
1436  * BlkActionState:
1437  * Describes one Action's state within a Transaction.
1438  *
1439  * @action: QAPI-defined enum identifying which Action to perform.
1440  * @ops: Table of ActionOps this Action can perform.
1441  * @block_job_txn: Transaction which this action belongs to.
1442  * @entry: List membership for all Actions in this Transaction.
1443  *
1444  * This structure must be arranged as first member in a subclassed type,
1445  * assuming that the compiler will also arrange it to the same offsets as the
1446  * base class.
1447  */
1448 struct BlkActionState {
1449     TransactionAction *action;
1450     const BlkActionOps *ops;
1451     BlockJobTxn *block_job_txn;
1452     TransactionProperties *txn_props;
1453     QSIMPLEQ_ENTRY(BlkActionState) entry;
1454 };
1455 
1456 /* internal snapshot private data */
1457 typedef struct InternalSnapshotState {
1458     BlkActionState common;
1459     BlockDriverState *bs;
1460     AioContext *aio_context;
1461     QEMUSnapshotInfo sn;
1462     bool created;
1463 } InternalSnapshotState;
1464 
1465 
1466 static int action_check_completion_mode(BlkActionState *s, Error **errp)
1467 {
1468     if (s->txn_props->completion_mode != ACTION_COMPLETION_MODE_INDIVIDUAL) {
1469         error_setg(errp,
1470                    "Action '%s' does not support Transaction property "
1471                    "completion-mode = %s",
1472                    TransactionActionKind_lookup[s->action->type],
1473                    ActionCompletionMode_lookup[s->txn_props->completion_mode]);
1474         return -1;
1475     }
1476     return 0;
1477 }
1478 
1479 static void internal_snapshot_prepare(BlkActionState *common,
1480                                       Error **errp)
1481 {
1482     Error *local_err = NULL;
1483     const char *device;
1484     const char *name;
1485     BlockDriverState *bs;
1486     QEMUSnapshotInfo old_sn, *sn;
1487     bool ret;
1488     qemu_timeval tv;
1489     BlockdevSnapshotInternal *internal;
1490     InternalSnapshotState *state;
1491     int ret1;
1492 
1493     g_assert(common->action->type ==
1494              TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC);
1495     internal = common->action->u.blockdev_snapshot_internal_sync.data;
1496     state = DO_UPCAST(InternalSnapshotState, common, common);
1497 
1498     /* 1. parse input */
1499     device = internal->device;
1500     name = internal->name;
1501 
1502     /* 2. check for validation */
1503     if (action_check_completion_mode(common, errp) < 0) {
1504         return;
1505     }
1506 
1507     bs = qmp_get_root_bs(device, errp);
1508     if (!bs) {
1509         return;
1510     }
1511 
1512     /* AioContext is released in .clean() */
1513     state->aio_context = bdrv_get_aio_context(bs);
1514     aio_context_acquire(state->aio_context);
1515 
1516     state->bs = bs;
1517     bdrv_drained_begin(bs);
1518 
1519     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_INTERNAL_SNAPSHOT, errp)) {
1520         return;
1521     }
1522 
1523     if (bdrv_is_read_only(bs)) {
1524         error_setg(errp, "Device '%s' is read only", device);
1525         return;
1526     }
1527 
1528     if (!bdrv_can_snapshot(bs)) {
1529         error_setg(errp, "Block format '%s' used by device '%s' "
1530                    "does not support internal snapshots",
1531                    bs->drv->format_name, device);
1532         return;
1533     }
1534 
1535     if (!strlen(name)) {
1536         error_setg(errp, "Name is empty");
1537         return;
1538     }
1539 
1540     /* check whether a snapshot with name exist */
1541     ret = bdrv_snapshot_find_by_id_and_name(bs, NULL, name, &old_sn,
1542                                             &local_err);
1543     if (local_err) {
1544         error_propagate(errp, local_err);
1545         return;
1546     } else if (ret) {
1547         error_setg(errp,
1548                    "Snapshot with name '%s' already exists on device '%s'",
1549                    name, device);
1550         return;
1551     }
1552 
1553     /* 3. take the snapshot */
1554     sn = &state->sn;
1555     pstrcpy(sn->name, sizeof(sn->name), name);
1556     qemu_gettimeofday(&tv);
1557     sn->date_sec = tv.tv_sec;
1558     sn->date_nsec = tv.tv_usec * 1000;
1559     sn->vm_clock_nsec = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL);
1560 
1561     ret1 = bdrv_snapshot_create(bs, sn);
1562     if (ret1 < 0) {
1563         error_setg_errno(errp, -ret1,
1564                          "Failed to create snapshot '%s' on device '%s'",
1565                          name, device);
1566         return;
1567     }
1568 
1569     /* 4. succeed, mark a snapshot is created */
1570     state->created = true;
1571 }
1572 
1573 static void internal_snapshot_abort(BlkActionState *common)
1574 {
1575     InternalSnapshotState *state =
1576                              DO_UPCAST(InternalSnapshotState, common, common);
1577     BlockDriverState *bs = state->bs;
1578     QEMUSnapshotInfo *sn = &state->sn;
1579     Error *local_error = NULL;
1580 
1581     if (!state->created) {
1582         return;
1583     }
1584 
1585     if (bdrv_snapshot_delete(bs, sn->id_str, sn->name, &local_error) < 0) {
1586         error_reportf_err(local_error,
1587                           "Failed to delete snapshot with id '%s' and "
1588                           "name '%s' on device '%s' in abort: ",
1589                           sn->id_str, sn->name,
1590                           bdrv_get_device_name(bs));
1591     }
1592 }
1593 
1594 static void internal_snapshot_clean(BlkActionState *common)
1595 {
1596     InternalSnapshotState *state = DO_UPCAST(InternalSnapshotState,
1597                                              common, common);
1598 
1599     if (state->aio_context) {
1600         if (state->bs) {
1601             bdrv_drained_end(state->bs);
1602         }
1603         aio_context_release(state->aio_context);
1604     }
1605 }
1606 
1607 /* external snapshot private data */
1608 typedef struct ExternalSnapshotState {
1609     BlkActionState common;
1610     BlockDriverState *old_bs;
1611     BlockDriverState *new_bs;
1612     AioContext *aio_context;
1613 } ExternalSnapshotState;
1614 
1615 static void external_snapshot_prepare(BlkActionState *common,
1616                                       Error **errp)
1617 {
1618     int flags = 0;
1619     QDict *options = NULL;
1620     Error *local_err = NULL;
1621     /* Device and node name of the image to generate the snapshot from */
1622     const char *device;
1623     const char *node_name;
1624     /* Reference to the new image (for 'blockdev-snapshot') */
1625     const char *snapshot_ref;
1626     /* File name of the new image (for 'blockdev-snapshot-sync') */
1627     const char *new_image_file;
1628     ExternalSnapshotState *state =
1629                              DO_UPCAST(ExternalSnapshotState, common, common);
1630     TransactionAction *action = common->action;
1631 
1632     /* 'blockdev-snapshot' and 'blockdev-snapshot-sync' have similar
1633      * purpose but a different set of parameters */
1634     switch (action->type) {
1635     case TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT:
1636         {
1637             BlockdevSnapshot *s = action->u.blockdev_snapshot.data;
1638             device = s->node;
1639             node_name = s->node;
1640             new_image_file = NULL;
1641             snapshot_ref = s->overlay;
1642         }
1643         break;
1644     case TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC:
1645         {
1646             BlockdevSnapshotSync *s = action->u.blockdev_snapshot_sync.data;
1647             device = s->has_device ? s->device : NULL;
1648             node_name = s->has_node_name ? s->node_name : NULL;
1649             new_image_file = s->snapshot_file;
1650             snapshot_ref = NULL;
1651         }
1652         break;
1653     default:
1654         g_assert_not_reached();
1655     }
1656 
1657     /* start processing */
1658     if (action_check_completion_mode(common, errp) < 0) {
1659         return;
1660     }
1661 
1662     state->old_bs = bdrv_lookup_bs(device, node_name, errp);
1663     if (!state->old_bs) {
1664         return;
1665     }
1666 
1667     /* Acquire AioContext now so any threads operating on old_bs stop */
1668     state->aio_context = bdrv_get_aio_context(state->old_bs);
1669     aio_context_acquire(state->aio_context);
1670     bdrv_drained_begin(state->old_bs);
1671 
1672     if (!bdrv_is_inserted(state->old_bs)) {
1673         error_setg(errp, QERR_DEVICE_HAS_NO_MEDIUM, device);
1674         return;
1675     }
1676 
1677     if (bdrv_op_is_blocked(state->old_bs,
1678                            BLOCK_OP_TYPE_EXTERNAL_SNAPSHOT, errp)) {
1679         return;
1680     }
1681 
1682     if (!bdrv_is_read_only(state->old_bs)) {
1683         if (bdrv_flush(state->old_bs)) {
1684             error_setg(errp, QERR_IO_ERROR);
1685             return;
1686         }
1687     }
1688 
1689     if (!bdrv_is_first_non_filter(state->old_bs)) {
1690         error_setg(errp, QERR_FEATURE_DISABLED, "snapshot");
1691         return;
1692     }
1693 
1694     if (action->type == TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC) {
1695         BlockdevSnapshotSync *s = action->u.blockdev_snapshot_sync.data;
1696         const char *format = s->has_format ? s->format : "qcow2";
1697         enum NewImageMode mode;
1698         const char *snapshot_node_name =
1699             s->has_snapshot_node_name ? s->snapshot_node_name : NULL;
1700 
1701         if (node_name && !snapshot_node_name) {
1702             error_setg(errp, "New snapshot node name missing");
1703             return;
1704         }
1705 
1706         if (snapshot_node_name &&
1707             bdrv_lookup_bs(snapshot_node_name, snapshot_node_name, NULL)) {
1708             error_setg(errp, "New snapshot node name already in use");
1709             return;
1710         }
1711 
1712         flags = state->old_bs->open_flags;
1713         flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_COPY_ON_READ);
1714 
1715         /* create new image w/backing file */
1716         mode = s->has_mode ? s->mode : NEW_IMAGE_MODE_ABSOLUTE_PATHS;
1717         if (mode != NEW_IMAGE_MODE_EXISTING) {
1718             int64_t size = bdrv_getlength(state->old_bs);
1719             if (size < 0) {
1720                 error_setg_errno(errp, -size, "bdrv_getlength failed");
1721                 return;
1722             }
1723             bdrv_img_create(new_image_file, format,
1724                             state->old_bs->filename,
1725                             state->old_bs->drv->format_name,
1726                             NULL, size, flags, &local_err, false);
1727             if (local_err) {
1728                 error_propagate(errp, local_err);
1729                 return;
1730             }
1731         }
1732 
1733         options = qdict_new();
1734         if (s->has_snapshot_node_name) {
1735             qdict_put(options, "node-name",
1736                       qstring_from_str(snapshot_node_name));
1737         }
1738         qdict_put(options, "driver", qstring_from_str(format));
1739 
1740         flags |= BDRV_O_NO_BACKING;
1741     }
1742 
1743     state->new_bs = bdrv_open(new_image_file, snapshot_ref, options, flags,
1744                               errp);
1745     /* We will manually add the backing_hd field to the bs later */
1746     if (!state->new_bs) {
1747         return;
1748     }
1749 
1750     if (bdrv_has_blk(state->new_bs)) {
1751         error_setg(errp, "The snapshot is already in use");
1752         return;
1753     }
1754 
1755     if (bdrv_op_is_blocked(state->new_bs, BLOCK_OP_TYPE_EXTERNAL_SNAPSHOT,
1756                            errp)) {
1757         return;
1758     }
1759 
1760     if (state->new_bs->backing != NULL) {
1761         error_setg(errp, "The snapshot already has a backing image");
1762         return;
1763     }
1764 
1765     if (!state->new_bs->drv->supports_backing) {
1766         error_setg(errp, "The snapshot does not support backing images");
1767     }
1768 }
1769 
1770 static void external_snapshot_commit(BlkActionState *common)
1771 {
1772     ExternalSnapshotState *state =
1773                              DO_UPCAST(ExternalSnapshotState, common, common);
1774 
1775     bdrv_set_aio_context(state->new_bs, state->aio_context);
1776 
1777     /* This removes our old bs and adds the new bs */
1778     bdrv_append(state->new_bs, state->old_bs);
1779     /* We don't need (or want) to use the transactional
1780      * bdrv_reopen_multiple() across all the entries at once, because we
1781      * don't want to abort all of them if one of them fails the reopen */
1782     if (!state->old_bs->copy_on_read) {
1783         bdrv_reopen(state->old_bs, state->old_bs->open_flags & ~BDRV_O_RDWR,
1784                     NULL);
1785     }
1786 }
1787 
1788 static void external_snapshot_abort(BlkActionState *common)
1789 {
1790     ExternalSnapshotState *state =
1791                              DO_UPCAST(ExternalSnapshotState, common, common);
1792     if (state->new_bs) {
1793         bdrv_unref(state->new_bs);
1794     }
1795 }
1796 
1797 static void external_snapshot_clean(BlkActionState *common)
1798 {
1799     ExternalSnapshotState *state =
1800                              DO_UPCAST(ExternalSnapshotState, common, common);
1801     if (state->aio_context) {
1802         bdrv_drained_end(state->old_bs);
1803         aio_context_release(state->aio_context);
1804     }
1805 }
1806 
1807 typedef struct DriveBackupState {
1808     BlkActionState common;
1809     BlockDriverState *bs;
1810     AioContext *aio_context;
1811     BlockJob *job;
1812 } DriveBackupState;
1813 
1814 static BlockJob *do_drive_backup(DriveBackup *backup, BlockJobTxn *txn,
1815                             Error **errp);
1816 
1817 static void drive_backup_prepare(BlkActionState *common, Error **errp)
1818 {
1819     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1820     BlockDriverState *bs;
1821     DriveBackup *backup;
1822     Error *local_err = NULL;
1823 
1824     assert(common->action->type == TRANSACTION_ACTION_KIND_DRIVE_BACKUP);
1825     backup = common->action->u.drive_backup.data;
1826 
1827     bs = qmp_get_root_bs(backup->device, errp);
1828     if (!bs) {
1829         return;
1830     }
1831 
1832     /* AioContext is released in .clean() */
1833     state->aio_context = bdrv_get_aio_context(bs);
1834     aio_context_acquire(state->aio_context);
1835     bdrv_drained_begin(bs);
1836     state->bs = bs;
1837 
1838     state->job = do_drive_backup(backup, common->block_job_txn, &local_err);
1839     if (local_err) {
1840         error_propagate(errp, local_err);
1841         return;
1842     }
1843 }
1844 
1845 static void drive_backup_commit(BlkActionState *common)
1846 {
1847     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1848     assert(state->job);
1849     block_job_start(state->job);
1850 }
1851 
1852 static void drive_backup_abort(BlkActionState *common)
1853 {
1854     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1855 
1856     if (state->job) {
1857         block_job_cancel_sync(state->job);
1858     }
1859 }
1860 
1861 static void drive_backup_clean(BlkActionState *common)
1862 {
1863     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1864 
1865     if (state->aio_context) {
1866         bdrv_drained_end(state->bs);
1867         aio_context_release(state->aio_context);
1868     }
1869 }
1870 
1871 typedef struct BlockdevBackupState {
1872     BlkActionState common;
1873     BlockDriverState *bs;
1874     BlockJob *job;
1875     AioContext *aio_context;
1876 } BlockdevBackupState;
1877 
1878 static BlockJob *do_blockdev_backup(BlockdevBackup *backup, BlockJobTxn *txn,
1879                                     Error **errp);
1880 
1881 static void blockdev_backup_prepare(BlkActionState *common, Error **errp)
1882 {
1883     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1884     BlockdevBackup *backup;
1885     BlockDriverState *bs, *target;
1886     Error *local_err = NULL;
1887 
1888     assert(common->action->type == TRANSACTION_ACTION_KIND_BLOCKDEV_BACKUP);
1889     backup = common->action->u.blockdev_backup.data;
1890 
1891     bs = qmp_get_root_bs(backup->device, errp);
1892     if (!bs) {
1893         return;
1894     }
1895 
1896     target = bdrv_lookup_bs(backup->target, backup->target, errp);
1897     if (!target) {
1898         return;
1899     }
1900 
1901     /* AioContext is released in .clean() */
1902     state->aio_context = bdrv_get_aio_context(bs);
1903     if (state->aio_context != bdrv_get_aio_context(target)) {
1904         state->aio_context = NULL;
1905         error_setg(errp, "Backup between two IO threads is not implemented");
1906         return;
1907     }
1908     aio_context_acquire(state->aio_context);
1909     state->bs = bs;
1910     bdrv_drained_begin(state->bs);
1911 
1912     state->job = do_blockdev_backup(backup, common->block_job_txn, &local_err);
1913     if (local_err) {
1914         error_propagate(errp, local_err);
1915         return;
1916     }
1917 }
1918 
1919 static void blockdev_backup_commit(BlkActionState *common)
1920 {
1921     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1922     assert(state->job);
1923     block_job_start(state->job);
1924 }
1925 
1926 static void blockdev_backup_abort(BlkActionState *common)
1927 {
1928     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1929 
1930     if (state->job) {
1931         block_job_cancel_sync(state->job);
1932     }
1933 }
1934 
1935 static void blockdev_backup_clean(BlkActionState *common)
1936 {
1937     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1938 
1939     if (state->aio_context) {
1940         bdrv_drained_end(state->bs);
1941         aio_context_release(state->aio_context);
1942     }
1943 }
1944 
1945 typedef struct BlockDirtyBitmapState {
1946     BlkActionState common;
1947     BdrvDirtyBitmap *bitmap;
1948     BlockDriverState *bs;
1949     AioContext *aio_context;
1950     HBitmap *backup;
1951     bool prepared;
1952 } BlockDirtyBitmapState;
1953 
1954 static void block_dirty_bitmap_add_prepare(BlkActionState *common,
1955                                            Error **errp)
1956 {
1957     Error *local_err = NULL;
1958     BlockDirtyBitmapAdd *action;
1959     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
1960                                              common, common);
1961 
1962     if (action_check_completion_mode(common, errp) < 0) {
1963         return;
1964     }
1965 
1966     action = common->action->u.block_dirty_bitmap_add.data;
1967     /* AIO context taken and released within qmp_block_dirty_bitmap_add */
1968     qmp_block_dirty_bitmap_add(action->node, action->name,
1969                                action->has_granularity, action->granularity,
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                                               &state->aio_context,
2010                                               errp);
2011     if (!state->bitmap) {
2012         return;
2013     }
2014 
2015     if (bdrv_dirty_bitmap_frozen(state->bitmap)) {
2016         error_setg(errp, "Cannot modify a frozen bitmap");
2017         return;
2018     } else if (!bdrv_dirty_bitmap_enabled(state->bitmap)) {
2019         error_setg(errp, "Cannot clear a disabled bitmap");
2020         return;
2021     }
2022 
2023     bdrv_clear_dirty_bitmap(state->bitmap, &state->backup);
2024     /* AioContext is released in .clean() */
2025 }
2026 
2027 static void block_dirty_bitmap_clear_abort(BlkActionState *common)
2028 {
2029     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2030                                              common, common);
2031 
2032     bdrv_undo_clear_dirty_bitmap(state->bitmap, state->backup);
2033 }
2034 
2035 static void block_dirty_bitmap_clear_commit(BlkActionState *common)
2036 {
2037     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2038                                              common, common);
2039 
2040     hbitmap_free(state->backup);
2041 }
2042 
2043 static void block_dirty_bitmap_clear_clean(BlkActionState *common)
2044 {
2045     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2046                                              common, common);
2047 
2048     if (state->aio_context) {
2049         aio_context_release(state->aio_context);
2050     }
2051 }
2052 
2053 static void abort_prepare(BlkActionState *common, Error **errp)
2054 {
2055     error_setg(errp, "Transaction aborted using Abort action");
2056 }
2057 
2058 static void abort_commit(BlkActionState *common)
2059 {
2060     g_assert_not_reached(); /* this action never succeeds */
2061 }
2062 
2063 static const BlkActionOps actions[] = {
2064     [TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT] = {
2065         .instance_size = sizeof(ExternalSnapshotState),
2066         .prepare  = external_snapshot_prepare,
2067         .commit   = external_snapshot_commit,
2068         .abort = external_snapshot_abort,
2069         .clean = external_snapshot_clean,
2070     },
2071     [TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC] = {
2072         .instance_size = sizeof(ExternalSnapshotState),
2073         .prepare  = external_snapshot_prepare,
2074         .commit   = external_snapshot_commit,
2075         .abort = external_snapshot_abort,
2076         .clean = external_snapshot_clean,
2077     },
2078     [TRANSACTION_ACTION_KIND_DRIVE_BACKUP] = {
2079         .instance_size = sizeof(DriveBackupState),
2080         .prepare = drive_backup_prepare,
2081         .commit = drive_backup_commit,
2082         .abort = drive_backup_abort,
2083         .clean = drive_backup_clean,
2084     },
2085     [TRANSACTION_ACTION_KIND_BLOCKDEV_BACKUP] = {
2086         .instance_size = sizeof(BlockdevBackupState),
2087         .prepare = blockdev_backup_prepare,
2088         .commit = blockdev_backup_commit,
2089         .abort = blockdev_backup_abort,
2090         .clean = blockdev_backup_clean,
2091     },
2092     [TRANSACTION_ACTION_KIND_ABORT] = {
2093         .instance_size = sizeof(BlkActionState),
2094         .prepare = abort_prepare,
2095         .commit = abort_commit,
2096     },
2097     [TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC] = {
2098         .instance_size = sizeof(InternalSnapshotState),
2099         .prepare  = internal_snapshot_prepare,
2100         .abort = internal_snapshot_abort,
2101         .clean = internal_snapshot_clean,
2102     },
2103     [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_ADD] = {
2104         .instance_size = sizeof(BlockDirtyBitmapState),
2105         .prepare = block_dirty_bitmap_add_prepare,
2106         .abort = block_dirty_bitmap_add_abort,
2107     },
2108     [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_CLEAR] = {
2109         .instance_size = sizeof(BlockDirtyBitmapState),
2110         .prepare = block_dirty_bitmap_clear_prepare,
2111         .commit = block_dirty_bitmap_clear_commit,
2112         .abort = block_dirty_bitmap_clear_abort,
2113         .clean = block_dirty_bitmap_clear_clean,
2114     }
2115 };
2116 
2117 /**
2118  * Allocate a TransactionProperties structure if necessary, and fill
2119  * that structure with desired defaults if they are unset.
2120  */
2121 static TransactionProperties *get_transaction_properties(
2122     TransactionProperties *props)
2123 {
2124     if (!props) {
2125         props = g_new0(TransactionProperties, 1);
2126     }
2127 
2128     if (!props->has_completion_mode) {
2129         props->has_completion_mode = true;
2130         props->completion_mode = ACTION_COMPLETION_MODE_INDIVIDUAL;
2131     }
2132 
2133     return props;
2134 }
2135 
2136 /*
2137  * 'Atomic' group operations.  The operations are performed as a set, and if
2138  * any fail then we roll back all operations in the group.
2139  */
2140 void qmp_transaction(TransactionActionList *dev_list,
2141                      bool has_props,
2142                      struct TransactionProperties *props,
2143                      Error **errp)
2144 {
2145     TransactionActionList *dev_entry = dev_list;
2146     BlockJobTxn *block_job_txn = NULL;
2147     BlkActionState *state, *next;
2148     Error *local_err = NULL;
2149 
2150     QSIMPLEQ_HEAD(snap_bdrv_states, BlkActionState) snap_bdrv_states;
2151     QSIMPLEQ_INIT(&snap_bdrv_states);
2152 
2153     /* Does this transaction get canceled as a group on failure?
2154      * If not, we don't really need to make a BlockJobTxn.
2155      */
2156     props = get_transaction_properties(props);
2157     if (props->completion_mode != ACTION_COMPLETION_MODE_INDIVIDUAL) {
2158         block_job_txn = block_job_txn_new();
2159     }
2160 
2161     /* drain all i/o before any operations */
2162     bdrv_drain_all();
2163 
2164     /* We don't do anything in this loop that commits us to the operations */
2165     while (NULL != dev_entry) {
2166         TransactionAction *dev_info = NULL;
2167         const BlkActionOps *ops;
2168 
2169         dev_info = dev_entry->value;
2170         dev_entry = dev_entry->next;
2171 
2172         assert(dev_info->type < ARRAY_SIZE(actions));
2173 
2174         ops = &actions[dev_info->type];
2175         assert(ops->instance_size > 0);
2176 
2177         state = g_malloc0(ops->instance_size);
2178         state->ops = ops;
2179         state->action = dev_info;
2180         state->block_job_txn = block_job_txn;
2181         state->txn_props = props;
2182         QSIMPLEQ_INSERT_TAIL(&snap_bdrv_states, state, entry);
2183 
2184         state->ops->prepare(state, &local_err);
2185         if (local_err) {
2186             error_propagate(errp, local_err);
2187             goto delete_and_fail;
2188         }
2189     }
2190 
2191     QSIMPLEQ_FOREACH(state, &snap_bdrv_states, entry) {
2192         if (state->ops->commit) {
2193             state->ops->commit(state);
2194         }
2195     }
2196 
2197     /* success */
2198     goto exit;
2199 
2200 delete_and_fail:
2201     /* failure, and it is all-or-none; roll back all operations */
2202     QSIMPLEQ_FOREACH(state, &snap_bdrv_states, entry) {
2203         if (state->ops->abort) {
2204             state->ops->abort(state);
2205         }
2206     }
2207 exit:
2208     QSIMPLEQ_FOREACH_SAFE(state, &snap_bdrv_states, entry, next) {
2209         if (state->ops->clean) {
2210             state->ops->clean(state);
2211         }
2212         g_free(state);
2213     }
2214     if (!has_props) {
2215         qapi_free_TransactionProperties(props);
2216     }
2217     block_job_txn_unref(block_job_txn);
2218 }
2219 
2220 void qmp_eject(bool has_device, const char *device,
2221                bool has_id, const char *id,
2222                bool has_force, bool force, Error **errp)
2223 {
2224     Error *local_err = NULL;
2225     int rc;
2226 
2227     if (!has_force) {
2228         force = false;
2229     }
2230 
2231     rc = do_open_tray(has_device ? device : NULL,
2232                       has_id ? id : NULL,
2233                       force, &local_err);
2234     if (rc && rc != -ENOSYS) {
2235         error_propagate(errp, local_err);
2236         return;
2237     }
2238     error_free(local_err);
2239 
2240     qmp_x_blockdev_remove_medium(has_device, device, has_id, id, errp);
2241 }
2242 
2243 void qmp_block_passwd(bool has_device, const char *device,
2244                       bool has_node_name, const char *node_name,
2245                       const char *password, Error **errp)
2246 {
2247     Error *local_err = NULL;
2248     BlockDriverState *bs;
2249     AioContext *aio_context;
2250 
2251     bs = bdrv_lookup_bs(has_device ? device : NULL,
2252                         has_node_name ? node_name : NULL,
2253                         &local_err);
2254     if (local_err) {
2255         error_propagate(errp, local_err);
2256         return;
2257     }
2258 
2259     aio_context = bdrv_get_aio_context(bs);
2260     aio_context_acquire(aio_context);
2261 
2262     bdrv_add_key(bs, password, errp);
2263 
2264     aio_context_release(aio_context);
2265 }
2266 
2267 /*
2268  * Attempt to open the tray of @device.
2269  * If @force, ignore its tray lock.
2270  * Else, if the tray is locked, don't open it, but ask the guest to open it.
2271  * On error, store an error through @errp and return -errno.
2272  * If @device does not exist, return -ENODEV.
2273  * If it has no removable media, return -ENOTSUP.
2274  * If it has no tray, return -ENOSYS.
2275  * If the guest was asked to open the tray, return -EINPROGRESS.
2276  * Else, return 0.
2277  */
2278 static int do_open_tray(const char *blk_name, const char *qdev_id,
2279                         bool force, Error **errp)
2280 {
2281     BlockBackend *blk;
2282     const char *device = qdev_id ?: blk_name;
2283     bool locked;
2284 
2285     blk = qmp_get_blk(blk_name, qdev_id, errp);
2286     if (!blk) {
2287         return -ENODEV;
2288     }
2289 
2290     if (!blk_dev_has_removable_media(blk)) {
2291         error_setg(errp, "Device '%s' is not removable", device);
2292         return -ENOTSUP;
2293     }
2294 
2295     if (!blk_dev_has_tray(blk)) {
2296         error_setg(errp, "Device '%s' does not have a tray", device);
2297         return -ENOSYS;
2298     }
2299 
2300     if (blk_dev_is_tray_open(blk)) {
2301         return 0;
2302     }
2303 
2304     locked = blk_dev_is_medium_locked(blk);
2305     if (locked) {
2306         blk_dev_eject_request(blk, force);
2307     }
2308 
2309     if (!locked || force) {
2310         blk_dev_change_media_cb(blk, false);
2311     }
2312 
2313     if (locked && !force) {
2314         error_setg(errp, "Device '%s' is locked and force was not specified, "
2315                    "wait for tray to open and try again", device);
2316         return -EINPROGRESS;
2317     }
2318 
2319     return 0;
2320 }
2321 
2322 void qmp_blockdev_open_tray(bool has_device, const char *device,
2323                             bool has_id, const char *id,
2324                             bool has_force, bool force,
2325                             Error **errp)
2326 {
2327     Error *local_err = NULL;
2328     int rc;
2329 
2330     if (!has_force) {
2331         force = false;
2332     }
2333     rc = do_open_tray(has_device ? device : NULL,
2334                       has_id ? id : NULL,
2335                       force, &local_err);
2336     if (rc && rc != -ENOSYS && rc != -EINPROGRESS) {
2337         error_propagate(errp, local_err);
2338         return;
2339     }
2340     error_free(local_err);
2341 }
2342 
2343 void qmp_blockdev_close_tray(bool has_device, const char *device,
2344                              bool has_id, const char *id,
2345                              Error **errp)
2346 {
2347     BlockBackend *blk;
2348 
2349     device = has_device ? device : NULL;
2350     id = has_id ? id : NULL;
2351 
2352     blk = qmp_get_blk(device, id, errp);
2353     if (!blk) {
2354         return;
2355     }
2356 
2357     if (!blk_dev_has_removable_media(blk)) {
2358         error_setg(errp, "Device '%s' is not removable", device ?: id);
2359         return;
2360     }
2361 
2362     if (!blk_dev_has_tray(blk)) {
2363         /* Ignore this command on tray-less devices */
2364         return;
2365     }
2366 
2367     if (!blk_dev_is_tray_open(blk)) {
2368         return;
2369     }
2370 
2371     blk_dev_change_media_cb(blk, true);
2372 }
2373 
2374 void qmp_x_blockdev_remove_medium(bool has_device, const char *device,
2375                                   bool has_id, const char *id, Error **errp)
2376 {
2377     BlockBackend *blk;
2378     BlockDriverState *bs;
2379     AioContext *aio_context;
2380     bool has_attached_device;
2381 
2382     device = has_device ? device : NULL;
2383     id = has_id ? id : NULL;
2384 
2385     blk = qmp_get_blk(device, id, errp);
2386     if (!blk) {
2387         return;
2388     }
2389 
2390     /* For BBs without a device, we can exchange the BDS tree at will */
2391     has_attached_device = blk_get_attached_dev(blk);
2392 
2393     if (has_attached_device && !blk_dev_has_removable_media(blk)) {
2394         error_setg(errp, "Device '%s' is not removable", device ?: id);
2395         return;
2396     }
2397 
2398     if (has_attached_device && blk_dev_has_tray(blk) &&
2399         !blk_dev_is_tray_open(blk))
2400     {
2401         error_setg(errp, "Tray of device '%s' is not open", device ?: id);
2402         return;
2403     }
2404 
2405     bs = blk_bs(blk);
2406     if (!bs) {
2407         return;
2408     }
2409 
2410     aio_context = bdrv_get_aio_context(bs);
2411     aio_context_acquire(aio_context);
2412 
2413     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_EJECT, errp)) {
2414         goto out;
2415     }
2416 
2417     blk_remove_bs(blk);
2418 
2419     if (!blk_dev_has_tray(blk)) {
2420         /* For tray-less devices, blockdev-open-tray is a no-op (or may not be
2421          * called at all); therefore, the medium needs to be ejected here.
2422          * Do it after blk_remove_bs() so blk_is_inserted(blk) returns the @load
2423          * value passed here (i.e. false). */
2424         blk_dev_change_media_cb(blk, false);
2425     }
2426 
2427 out:
2428     aio_context_release(aio_context);
2429 }
2430 
2431 static void qmp_blockdev_insert_anon_medium(BlockBackend *blk,
2432                                             BlockDriverState *bs, Error **errp)
2433 {
2434     bool has_device;
2435 
2436     /* For BBs without a device, we can exchange the BDS tree at will */
2437     has_device = blk_get_attached_dev(blk);
2438 
2439     if (has_device && !blk_dev_has_removable_media(blk)) {
2440         error_setg(errp, "Device is not removable");
2441         return;
2442     }
2443 
2444     if (has_device && blk_dev_has_tray(blk) && !blk_dev_is_tray_open(blk)) {
2445         error_setg(errp, "Tray of the device is not open");
2446         return;
2447     }
2448 
2449     if (blk_bs(blk)) {
2450         error_setg(errp, "There already is a medium in the device");
2451         return;
2452     }
2453 
2454     blk_insert_bs(blk, bs);
2455 
2456     if (!blk_dev_has_tray(blk)) {
2457         /* For tray-less devices, blockdev-close-tray is a no-op (or may not be
2458          * called at all); therefore, the medium needs to be pushed into the
2459          * slot here.
2460          * Do it after blk_insert_bs() so blk_is_inserted(blk) returns the @load
2461          * value passed here (i.e. true). */
2462         blk_dev_change_media_cb(blk, true);
2463     }
2464 }
2465 
2466 void qmp_x_blockdev_insert_medium(bool has_device, const char *device,
2467                                   bool has_id, const char *id,
2468                                   const char *node_name, Error **errp)
2469 {
2470     BlockBackend *blk;
2471     BlockDriverState *bs;
2472 
2473     blk = qmp_get_blk(has_device ? device : NULL,
2474                       has_id ? id : NULL,
2475                       errp);
2476     if (!blk) {
2477         return;
2478     }
2479 
2480     bs = bdrv_find_node(node_name);
2481     if (!bs) {
2482         error_setg(errp, "Node '%s' not found", node_name);
2483         return;
2484     }
2485 
2486     if (bdrv_has_blk(bs)) {
2487         error_setg(errp, "Node '%s' is already in use", node_name);
2488         return;
2489     }
2490 
2491     qmp_blockdev_insert_anon_medium(blk, bs, errp);
2492 }
2493 
2494 void qmp_blockdev_change_medium(bool has_device, const char *device,
2495                                 bool has_id, const char *id,
2496                                 const char *filename,
2497                                 bool has_format, const char *format,
2498                                 bool has_read_only,
2499                                 BlockdevChangeReadOnlyMode read_only,
2500                                 Error **errp)
2501 {
2502     BlockBackend *blk;
2503     BlockDriverState *medium_bs = NULL;
2504     int bdrv_flags;
2505     bool detect_zeroes;
2506     int rc;
2507     QDict *options = NULL;
2508     Error *err = NULL;
2509 
2510     blk = qmp_get_blk(has_device ? device : NULL,
2511                       has_id ? id : NULL,
2512                       errp);
2513     if (!blk) {
2514         goto fail;
2515     }
2516 
2517     if (blk_bs(blk)) {
2518         blk_update_root_state(blk);
2519     }
2520 
2521     bdrv_flags = blk_get_open_flags_from_root_state(blk);
2522     bdrv_flags &= ~(BDRV_O_TEMPORARY | BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING |
2523         BDRV_O_PROTOCOL);
2524 
2525     if (!has_read_only) {
2526         read_only = BLOCKDEV_CHANGE_READ_ONLY_MODE_RETAIN;
2527     }
2528 
2529     switch (read_only) {
2530     case BLOCKDEV_CHANGE_READ_ONLY_MODE_RETAIN:
2531         break;
2532 
2533     case BLOCKDEV_CHANGE_READ_ONLY_MODE_READ_ONLY:
2534         bdrv_flags &= ~BDRV_O_RDWR;
2535         break;
2536 
2537     case BLOCKDEV_CHANGE_READ_ONLY_MODE_READ_WRITE:
2538         bdrv_flags |= BDRV_O_RDWR;
2539         break;
2540 
2541     default:
2542         abort();
2543     }
2544 
2545     options = qdict_new();
2546     detect_zeroes = blk_get_detect_zeroes_from_root_state(blk);
2547     qdict_put(options, "detect-zeroes",
2548               qstring_from_str(detect_zeroes ? "on" : "off"));
2549 
2550     if (has_format) {
2551         qdict_put(options, "driver", qstring_from_str(format));
2552     }
2553 
2554     medium_bs = bdrv_open(filename, NULL, options, bdrv_flags, errp);
2555     if (!medium_bs) {
2556         goto fail;
2557     }
2558 
2559     bdrv_add_key(medium_bs, NULL, &err);
2560     if (err) {
2561         error_propagate(errp, err);
2562         goto fail;
2563     }
2564 
2565     rc = do_open_tray(has_device ? device : NULL,
2566                       has_id ? id : NULL,
2567                       false, &err);
2568     if (rc && rc != -ENOSYS) {
2569         error_propagate(errp, err);
2570         goto fail;
2571     }
2572     error_free(err);
2573     err = NULL;
2574 
2575     qmp_x_blockdev_remove_medium(has_device, device, has_id, id, &err);
2576     if (err) {
2577         error_propagate(errp, err);
2578         goto fail;
2579     }
2580 
2581     qmp_blockdev_insert_anon_medium(blk, medium_bs, &err);
2582     if (err) {
2583         error_propagate(errp, err);
2584         goto fail;
2585     }
2586 
2587     qmp_blockdev_close_tray(has_device, device, has_id, id, errp);
2588 
2589 fail:
2590     /* If the medium has been inserted, the device has its own reference, so
2591      * ours must be relinquished; and if it has not been inserted successfully,
2592      * the reference must be relinquished anyway */
2593     bdrv_unref(medium_bs);
2594 }
2595 
2596 /* throttling disk I/O limits */
2597 void qmp_block_set_io_throttle(BlockIOThrottle *arg, Error **errp)
2598 {
2599     ThrottleConfig cfg;
2600     BlockDriverState *bs;
2601     BlockBackend *blk;
2602     AioContext *aio_context;
2603 
2604     blk = qmp_get_blk(arg->has_device ? arg->device : NULL,
2605                       arg->has_id ? arg->id : NULL,
2606                       errp);
2607     if (!blk) {
2608         return;
2609     }
2610 
2611     aio_context = blk_get_aio_context(blk);
2612     aio_context_acquire(aio_context);
2613 
2614     bs = blk_bs(blk);
2615     if (!bs) {
2616         error_setg(errp, "Device has no medium");
2617         goto out;
2618     }
2619 
2620     throttle_config_init(&cfg);
2621     cfg.buckets[THROTTLE_BPS_TOTAL].avg = arg->bps;
2622     cfg.buckets[THROTTLE_BPS_READ].avg  = arg->bps_rd;
2623     cfg.buckets[THROTTLE_BPS_WRITE].avg = arg->bps_wr;
2624 
2625     cfg.buckets[THROTTLE_OPS_TOTAL].avg = arg->iops;
2626     cfg.buckets[THROTTLE_OPS_READ].avg  = arg->iops_rd;
2627     cfg.buckets[THROTTLE_OPS_WRITE].avg = arg->iops_wr;
2628 
2629     if (arg->has_bps_max) {
2630         cfg.buckets[THROTTLE_BPS_TOTAL].max = arg->bps_max;
2631     }
2632     if (arg->has_bps_rd_max) {
2633         cfg.buckets[THROTTLE_BPS_READ].max = arg->bps_rd_max;
2634     }
2635     if (arg->has_bps_wr_max) {
2636         cfg.buckets[THROTTLE_BPS_WRITE].max = arg->bps_wr_max;
2637     }
2638     if (arg->has_iops_max) {
2639         cfg.buckets[THROTTLE_OPS_TOTAL].max = arg->iops_max;
2640     }
2641     if (arg->has_iops_rd_max) {
2642         cfg.buckets[THROTTLE_OPS_READ].max = arg->iops_rd_max;
2643     }
2644     if (arg->has_iops_wr_max) {
2645         cfg.buckets[THROTTLE_OPS_WRITE].max = arg->iops_wr_max;
2646     }
2647 
2648     if (arg->has_bps_max_length) {
2649         cfg.buckets[THROTTLE_BPS_TOTAL].burst_length = arg->bps_max_length;
2650     }
2651     if (arg->has_bps_rd_max_length) {
2652         cfg.buckets[THROTTLE_BPS_READ].burst_length = arg->bps_rd_max_length;
2653     }
2654     if (arg->has_bps_wr_max_length) {
2655         cfg.buckets[THROTTLE_BPS_WRITE].burst_length = arg->bps_wr_max_length;
2656     }
2657     if (arg->has_iops_max_length) {
2658         cfg.buckets[THROTTLE_OPS_TOTAL].burst_length = arg->iops_max_length;
2659     }
2660     if (arg->has_iops_rd_max_length) {
2661         cfg.buckets[THROTTLE_OPS_READ].burst_length = arg->iops_rd_max_length;
2662     }
2663     if (arg->has_iops_wr_max_length) {
2664         cfg.buckets[THROTTLE_OPS_WRITE].burst_length = arg->iops_wr_max_length;
2665     }
2666 
2667     if (arg->has_iops_size) {
2668         cfg.op_size = arg->iops_size;
2669     }
2670 
2671     if (!throttle_is_valid(&cfg, errp)) {
2672         goto out;
2673     }
2674 
2675     if (throttle_enabled(&cfg)) {
2676         /* Enable I/O limits if they're not enabled yet, otherwise
2677          * just update the throttling group. */
2678         if (!blk_get_public(blk)->throttle_state) {
2679             blk_io_limits_enable(blk,
2680                                  arg->has_group ? arg->group :
2681                                  arg->has_device ? arg->device :
2682                                  arg->id);
2683         } else if (arg->has_group) {
2684             blk_io_limits_update_group(blk, arg->group);
2685         }
2686         /* Set the new throttling configuration */
2687         blk_set_io_limits(blk, &cfg);
2688     } else if (blk_get_public(blk)->throttle_state) {
2689         /* If all throttling settings are set to 0, disable I/O limits */
2690         blk_io_limits_disable(blk);
2691     }
2692 
2693 out:
2694     aio_context_release(aio_context);
2695 }
2696 
2697 void qmp_block_dirty_bitmap_add(const char *node, const char *name,
2698                                 bool has_granularity, uint32_t granularity,
2699                                 Error **errp)
2700 {
2701     AioContext *aio_context;
2702     BlockDriverState *bs;
2703 
2704     if (!name || name[0] == '\0') {
2705         error_setg(errp, "Bitmap name cannot be empty");
2706         return;
2707     }
2708 
2709     bs = bdrv_lookup_bs(node, node, errp);
2710     if (!bs) {
2711         return;
2712     }
2713 
2714     aio_context = bdrv_get_aio_context(bs);
2715     aio_context_acquire(aio_context);
2716 
2717     if (has_granularity) {
2718         if (granularity < 512 || !is_power_of_2(granularity)) {
2719             error_setg(errp, "Granularity must be power of 2 "
2720                              "and at least 512");
2721             goto out;
2722         }
2723     } else {
2724         /* Default to cluster size, if available: */
2725         granularity = bdrv_get_default_bitmap_granularity(bs);
2726     }
2727 
2728     bdrv_create_dirty_bitmap(bs, granularity, name, errp);
2729 
2730  out:
2731     aio_context_release(aio_context);
2732 }
2733 
2734 void qmp_block_dirty_bitmap_remove(const char *node, const char *name,
2735                                    Error **errp)
2736 {
2737     AioContext *aio_context;
2738     BlockDriverState *bs;
2739     BdrvDirtyBitmap *bitmap;
2740 
2741     bitmap = block_dirty_bitmap_lookup(node, name, &bs, &aio_context, errp);
2742     if (!bitmap || !bs) {
2743         return;
2744     }
2745 
2746     if (bdrv_dirty_bitmap_frozen(bitmap)) {
2747         error_setg(errp,
2748                    "Bitmap '%s' is currently frozen and cannot be removed",
2749                    name);
2750         goto out;
2751     }
2752     bdrv_dirty_bitmap_make_anon(bitmap);
2753     bdrv_release_dirty_bitmap(bs, bitmap);
2754 
2755  out:
2756     aio_context_release(aio_context);
2757 }
2758 
2759 /**
2760  * Completely clear a bitmap, for the purposes of synchronizing a bitmap
2761  * immediately after a full backup operation.
2762  */
2763 void qmp_block_dirty_bitmap_clear(const char *node, const char *name,
2764                                   Error **errp)
2765 {
2766     AioContext *aio_context;
2767     BdrvDirtyBitmap *bitmap;
2768     BlockDriverState *bs;
2769 
2770     bitmap = block_dirty_bitmap_lookup(node, name, &bs, &aio_context, errp);
2771     if (!bitmap || !bs) {
2772         return;
2773     }
2774 
2775     if (bdrv_dirty_bitmap_frozen(bitmap)) {
2776         error_setg(errp,
2777                    "Bitmap '%s' is currently frozen and cannot be modified",
2778                    name);
2779         goto out;
2780     } else if (!bdrv_dirty_bitmap_enabled(bitmap)) {
2781         error_setg(errp,
2782                    "Bitmap '%s' is currently disabled and cannot be cleared",
2783                    name);
2784         goto out;
2785     }
2786 
2787     bdrv_clear_dirty_bitmap(bitmap, NULL);
2788 
2789  out:
2790     aio_context_release(aio_context);
2791 }
2792 
2793 void hmp_drive_del(Monitor *mon, const QDict *qdict)
2794 {
2795     const char *id = qdict_get_str(qdict, "id");
2796     BlockBackend *blk;
2797     BlockDriverState *bs;
2798     AioContext *aio_context;
2799     Error *local_err = NULL;
2800 
2801     bs = bdrv_find_node(id);
2802     if (bs) {
2803         qmp_x_blockdev_del(id, &local_err);
2804         if (local_err) {
2805             error_report_err(local_err);
2806         }
2807         return;
2808     }
2809 
2810     blk = blk_by_name(id);
2811     if (!blk) {
2812         error_report("Device '%s' not found", id);
2813         return;
2814     }
2815 
2816     if (!blk_legacy_dinfo(blk)) {
2817         error_report("Deleting device added with blockdev-add"
2818                      " is not supported");
2819         return;
2820     }
2821 
2822     aio_context = blk_get_aio_context(blk);
2823     aio_context_acquire(aio_context);
2824 
2825     bs = blk_bs(blk);
2826     if (bs) {
2827         if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_DRIVE_DEL, &local_err)) {
2828             error_report_err(local_err);
2829             aio_context_release(aio_context);
2830             return;
2831         }
2832 
2833         blk_remove_bs(blk);
2834     }
2835 
2836     /* Make the BlockBackend and the attached BlockDriverState anonymous */
2837     monitor_remove_blk(blk);
2838 
2839     /* If this BlockBackend has a device attached to it, its refcount will be
2840      * decremented when the device is removed; otherwise we have to do so here.
2841      */
2842     if (blk_get_attached_dev(blk)) {
2843         /* Further I/O must not pause the guest */
2844         blk_set_on_error(blk, BLOCKDEV_ON_ERROR_REPORT,
2845                          BLOCKDEV_ON_ERROR_REPORT);
2846     } else {
2847         blk_unref(blk);
2848     }
2849 
2850     aio_context_release(aio_context);
2851 }
2852 
2853 void qmp_block_resize(bool has_device, const char *device,
2854                       bool has_node_name, const char *node_name,
2855                       int64_t size, Error **errp)
2856 {
2857     Error *local_err = NULL;
2858     BlockDriverState *bs;
2859     AioContext *aio_context;
2860     int ret;
2861 
2862     bs = bdrv_lookup_bs(has_device ? device : NULL,
2863                         has_node_name ? node_name : NULL,
2864                         &local_err);
2865     if (local_err) {
2866         error_propagate(errp, local_err);
2867         return;
2868     }
2869 
2870     aio_context = bdrv_get_aio_context(bs);
2871     aio_context_acquire(aio_context);
2872 
2873     if (!bdrv_is_first_non_filter(bs)) {
2874         error_setg(errp, QERR_FEATURE_DISABLED, "resize");
2875         goto out;
2876     }
2877 
2878     if (size < 0) {
2879         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "size", "a >0 size");
2880         goto out;
2881     }
2882 
2883     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_RESIZE, NULL)) {
2884         error_setg(errp, QERR_DEVICE_IN_USE, device);
2885         goto out;
2886     }
2887 
2888     /* complete all in-flight operations before resizing the device */
2889     bdrv_drain_all();
2890 
2891     ret = bdrv_truncate(bs, size);
2892     switch (ret) {
2893     case 0:
2894         break;
2895     case -ENOMEDIUM:
2896         error_setg(errp, QERR_DEVICE_HAS_NO_MEDIUM, device);
2897         break;
2898     case -ENOTSUP:
2899         error_setg(errp, QERR_UNSUPPORTED);
2900         break;
2901     case -EACCES:
2902         error_setg(errp, "Device '%s' is read only", device);
2903         break;
2904     case -EBUSY:
2905         error_setg(errp, QERR_DEVICE_IN_USE, device);
2906         break;
2907     default:
2908         error_setg_errno(errp, -ret, "Could not resize");
2909         break;
2910     }
2911 
2912 out:
2913     aio_context_release(aio_context);
2914 }
2915 
2916 void qmp_block_stream(bool has_job_id, const char *job_id, const char *device,
2917                       bool has_base, const char *base,
2918                       bool has_base_node, const char *base_node,
2919                       bool has_backing_file, const char *backing_file,
2920                       bool has_speed, int64_t speed,
2921                       bool has_on_error, BlockdevOnError on_error,
2922                       Error **errp)
2923 {
2924     BlockDriverState *bs, *iter;
2925     BlockDriverState *base_bs = NULL;
2926     AioContext *aio_context;
2927     Error *local_err = NULL;
2928     const char *base_name = NULL;
2929 
2930     if (!has_on_error) {
2931         on_error = BLOCKDEV_ON_ERROR_REPORT;
2932     }
2933 
2934     bs = bdrv_lookup_bs(device, device, errp);
2935     if (!bs) {
2936         return;
2937     }
2938 
2939     aio_context = bdrv_get_aio_context(bs);
2940     aio_context_acquire(aio_context);
2941 
2942     if (has_base && has_base_node) {
2943         error_setg(errp, "'base' and 'base-node' cannot be specified "
2944                    "at the same time");
2945         goto out;
2946     }
2947 
2948     if (has_base) {
2949         base_bs = bdrv_find_backing_image(bs, base);
2950         if (base_bs == NULL) {
2951             error_setg(errp, QERR_BASE_NOT_FOUND, base);
2952             goto out;
2953         }
2954         assert(bdrv_get_aio_context(base_bs) == aio_context);
2955         base_name = base;
2956     }
2957 
2958     if (has_base_node) {
2959         base_bs = bdrv_lookup_bs(NULL, base_node, errp);
2960         if (!base_bs) {
2961             goto out;
2962         }
2963         if (bs == base_bs || !bdrv_chain_contains(bs, base_bs)) {
2964             error_setg(errp, "Node '%s' is not a backing image of '%s'",
2965                        base_node, device);
2966             goto out;
2967         }
2968         assert(bdrv_get_aio_context(base_bs) == aio_context);
2969         base_name = base_bs->filename;
2970     }
2971 
2972     /* Check for op blockers in the whole chain between bs and base */
2973     for (iter = bs; iter && iter != base_bs; iter = backing_bs(iter)) {
2974         if (bdrv_op_is_blocked(iter, BLOCK_OP_TYPE_STREAM, errp)) {
2975             goto out;
2976         }
2977     }
2978 
2979     /* if we are streaming the entire chain, the result will have no backing
2980      * file, and specifying one is therefore an error */
2981     if (base_bs == NULL && has_backing_file) {
2982         error_setg(errp, "backing file specified, but streaming the "
2983                          "entire chain");
2984         goto out;
2985     }
2986 
2987     /* backing_file string overrides base bs filename */
2988     base_name = has_backing_file ? backing_file : base_name;
2989 
2990     stream_start(has_job_id ? job_id : NULL, bs, base_bs, base_name,
2991                  has_speed ? speed : 0, on_error, &local_err);
2992     if (local_err) {
2993         error_propagate(errp, local_err);
2994         goto out;
2995     }
2996 
2997     trace_qmp_block_stream(bs, bs->job);
2998 
2999 out:
3000     aio_context_release(aio_context);
3001 }
3002 
3003 void qmp_block_commit(bool has_job_id, const char *job_id, const char *device,
3004                       bool has_base, const char *base,
3005                       bool has_top, const char *top,
3006                       bool has_backing_file, const char *backing_file,
3007                       bool has_speed, int64_t speed,
3008                       Error **errp)
3009 {
3010     BlockDriverState *bs;
3011     BlockDriverState *iter;
3012     BlockDriverState *base_bs, *top_bs;
3013     AioContext *aio_context;
3014     Error *local_err = NULL;
3015     /* This will be part of the QMP command, if/when the
3016      * BlockdevOnError change for blkmirror makes it in
3017      */
3018     BlockdevOnError on_error = BLOCKDEV_ON_ERROR_REPORT;
3019 
3020     if (!has_speed) {
3021         speed = 0;
3022     }
3023 
3024     /* Important Note:
3025      *  libvirt relies on the DeviceNotFound error class in order to probe for
3026      *  live commit feature versions; for this to work, we must make sure to
3027      *  perform the device lookup before any generic errors that may occur in a
3028      *  scenario in which all optional arguments are omitted. */
3029     bs = qmp_get_root_bs(device, &local_err);
3030     if (!bs) {
3031         bs = bdrv_lookup_bs(device, device, NULL);
3032         if (!bs) {
3033             error_free(local_err);
3034             error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
3035                       "Device '%s' not found", device);
3036         } else {
3037             error_propagate(errp, local_err);
3038         }
3039         return;
3040     }
3041 
3042     aio_context = bdrv_get_aio_context(bs);
3043     aio_context_acquire(aio_context);
3044 
3045     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_COMMIT_SOURCE, errp)) {
3046         goto out;
3047     }
3048 
3049     /* default top_bs is the active layer */
3050     top_bs = bs;
3051 
3052     if (has_top && top) {
3053         if (strcmp(bs->filename, top) != 0) {
3054             top_bs = bdrv_find_backing_image(bs, top);
3055         }
3056     }
3057 
3058     if (top_bs == NULL) {
3059         error_setg(errp, "Top image file %s not found", top ? top : "NULL");
3060         goto out;
3061     }
3062 
3063     assert(bdrv_get_aio_context(top_bs) == aio_context);
3064 
3065     if (has_base && base) {
3066         base_bs = bdrv_find_backing_image(top_bs, base);
3067     } else {
3068         base_bs = bdrv_find_base(top_bs);
3069     }
3070 
3071     if (base_bs == NULL) {
3072         error_setg(errp, QERR_BASE_NOT_FOUND, base ? base : "NULL");
3073         goto out;
3074     }
3075 
3076     assert(bdrv_get_aio_context(base_bs) == aio_context);
3077 
3078     for (iter = top_bs; iter != backing_bs(base_bs); iter = backing_bs(iter)) {
3079         if (bdrv_op_is_blocked(iter, BLOCK_OP_TYPE_COMMIT_TARGET, errp)) {
3080             goto out;
3081         }
3082     }
3083 
3084     /* Do not allow attempts to commit an image into itself */
3085     if (top_bs == base_bs) {
3086         error_setg(errp, "cannot commit an image into itself");
3087         goto out;
3088     }
3089 
3090     if (top_bs == bs) {
3091         if (has_backing_file) {
3092             error_setg(errp, "'backing-file' specified,"
3093                              " but 'top' is the active layer");
3094             goto out;
3095         }
3096         commit_active_start(has_job_id ? job_id : NULL, bs, base_bs,
3097                             BLOCK_JOB_DEFAULT, speed, on_error, NULL, NULL,
3098                             &local_err, false);
3099     } else {
3100         BlockDriverState *overlay_bs = bdrv_find_overlay(bs, top_bs);
3101         if (bdrv_op_is_blocked(overlay_bs, BLOCK_OP_TYPE_COMMIT_TARGET, errp)) {
3102             goto out;
3103         }
3104         commit_start(has_job_id ? job_id : NULL, bs, base_bs, top_bs, speed,
3105                      on_error, has_backing_file ? backing_file : NULL,
3106                      &local_err);
3107     }
3108     if (local_err != NULL) {
3109         error_propagate(errp, local_err);
3110         goto out;
3111     }
3112 
3113 out:
3114     aio_context_release(aio_context);
3115 }
3116 
3117 static BlockJob *do_drive_backup(DriveBackup *backup, BlockJobTxn *txn,
3118                                  Error **errp)
3119 {
3120     BlockDriverState *bs;
3121     BlockDriverState *target_bs;
3122     BlockDriverState *source = NULL;
3123     BlockJob *job = NULL;
3124     BdrvDirtyBitmap *bmap = NULL;
3125     AioContext *aio_context;
3126     QDict *options = NULL;
3127     Error *local_err = NULL;
3128     int flags;
3129     int64_t size;
3130 
3131     if (!backup->has_speed) {
3132         backup->speed = 0;
3133     }
3134     if (!backup->has_on_source_error) {
3135         backup->on_source_error = BLOCKDEV_ON_ERROR_REPORT;
3136     }
3137     if (!backup->has_on_target_error) {
3138         backup->on_target_error = BLOCKDEV_ON_ERROR_REPORT;
3139     }
3140     if (!backup->has_mode) {
3141         backup->mode = NEW_IMAGE_MODE_ABSOLUTE_PATHS;
3142     }
3143     if (!backup->has_job_id) {
3144         backup->job_id = NULL;
3145     }
3146     if (!backup->has_compress) {
3147         backup->compress = false;
3148     }
3149 
3150     bs = qmp_get_root_bs(backup->device, errp);
3151     if (!bs) {
3152         return NULL;
3153     }
3154 
3155     aio_context = bdrv_get_aio_context(bs);
3156     aio_context_acquire(aio_context);
3157 
3158     if (!backup->has_format) {
3159         backup->format = backup->mode == NEW_IMAGE_MODE_EXISTING ?
3160                          NULL : (char*) bs->drv->format_name;
3161     }
3162 
3163     /* Early check to avoid creating target */
3164     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_BACKUP_SOURCE, errp)) {
3165         goto out;
3166     }
3167 
3168     flags = bs->open_flags | BDRV_O_RDWR;
3169 
3170     /* See if we have a backing HD we can use to create our new image
3171      * on top of. */
3172     if (backup->sync == MIRROR_SYNC_MODE_TOP) {
3173         source = backing_bs(bs);
3174         if (!source) {
3175             backup->sync = MIRROR_SYNC_MODE_FULL;
3176         }
3177     }
3178     if (backup->sync == MIRROR_SYNC_MODE_NONE) {
3179         source = bs;
3180     }
3181 
3182     size = bdrv_getlength(bs);
3183     if (size < 0) {
3184         error_setg_errno(errp, -size, "bdrv_getlength failed");
3185         goto out;
3186     }
3187 
3188     if (backup->mode != NEW_IMAGE_MODE_EXISTING) {
3189         assert(backup->format);
3190         if (source) {
3191             bdrv_img_create(backup->target, backup->format, source->filename,
3192                             source->drv->format_name, NULL,
3193                             size, flags, &local_err, false);
3194         } else {
3195             bdrv_img_create(backup->target, backup->format, NULL, NULL, NULL,
3196                             size, flags, &local_err, false);
3197         }
3198     }
3199 
3200     if (local_err) {
3201         error_propagate(errp, local_err);
3202         goto out;
3203     }
3204 
3205     if (backup->format) {
3206         options = qdict_new();
3207         qdict_put(options, "driver", qstring_from_str(backup->format));
3208     }
3209 
3210     target_bs = bdrv_open(backup->target, NULL, options, flags, errp);
3211     if (!target_bs) {
3212         goto out;
3213     }
3214 
3215     bdrv_set_aio_context(target_bs, aio_context);
3216 
3217     if (backup->has_bitmap) {
3218         bmap = bdrv_find_dirty_bitmap(bs, backup->bitmap);
3219         if (!bmap) {
3220             error_setg(errp, "Bitmap '%s' could not be found", backup->bitmap);
3221             bdrv_unref(target_bs);
3222             goto out;
3223         }
3224     }
3225 
3226     job = backup_job_create(backup->job_id, bs, target_bs, backup->speed,
3227                             backup->sync, bmap, backup->compress,
3228                             backup->on_source_error, backup->on_target_error,
3229                             BLOCK_JOB_DEFAULT, NULL, NULL, txn, &local_err);
3230     bdrv_unref(target_bs);
3231     if (local_err != NULL) {
3232         error_propagate(errp, local_err);
3233         goto out;
3234     }
3235 
3236 out:
3237     aio_context_release(aio_context);
3238     return job;
3239 }
3240 
3241 void qmp_drive_backup(DriveBackup *arg, Error **errp)
3242 {
3243 
3244     BlockJob *job;
3245     job = do_drive_backup(arg, NULL, errp);
3246     if (job) {
3247         block_job_start(job);
3248     }
3249 }
3250 
3251 BlockDeviceInfoList *qmp_query_named_block_nodes(Error **errp)
3252 {
3253     return bdrv_named_nodes_list(errp);
3254 }
3255 
3256 BlockJob *do_blockdev_backup(BlockdevBackup *backup, BlockJobTxn *txn,
3257                              Error **errp)
3258 {
3259     BlockDriverState *bs;
3260     BlockDriverState *target_bs;
3261     Error *local_err = NULL;
3262     AioContext *aio_context;
3263     BlockJob *job = NULL;
3264 
3265     if (!backup->has_speed) {
3266         backup->speed = 0;
3267     }
3268     if (!backup->has_on_source_error) {
3269         backup->on_source_error = BLOCKDEV_ON_ERROR_REPORT;
3270     }
3271     if (!backup->has_on_target_error) {
3272         backup->on_target_error = BLOCKDEV_ON_ERROR_REPORT;
3273     }
3274     if (!backup->has_job_id) {
3275         backup->job_id = NULL;
3276     }
3277     if (!backup->has_compress) {
3278         backup->compress = false;
3279     }
3280 
3281     bs = qmp_get_root_bs(backup->device, errp);
3282     if (!bs) {
3283         return NULL;
3284     }
3285 
3286     aio_context = bdrv_get_aio_context(bs);
3287     aio_context_acquire(aio_context);
3288 
3289     target_bs = bdrv_lookup_bs(backup->target, backup->target, errp);
3290     if (!target_bs) {
3291         goto out;
3292     }
3293 
3294     if (bdrv_get_aio_context(target_bs) != aio_context) {
3295         if (!bdrv_has_blk(target_bs)) {
3296             /* The target BDS is not attached, we can safely move it to another
3297              * AioContext. */
3298             bdrv_set_aio_context(target_bs, aio_context);
3299         } else {
3300             error_setg(errp, "Target is attached to a different thread from "
3301                              "source.");
3302             goto out;
3303         }
3304     }
3305     job = backup_job_create(backup->job_id, bs, target_bs, backup->speed,
3306                             backup->sync, NULL, backup->compress,
3307                             backup->on_source_error, backup->on_target_error,
3308                             BLOCK_JOB_DEFAULT, NULL, NULL, txn, &local_err);
3309     if (local_err != NULL) {
3310         error_propagate(errp, local_err);
3311     }
3312 out:
3313     aio_context_release(aio_context);
3314     return job;
3315 }
3316 
3317 void qmp_blockdev_backup(BlockdevBackup *arg, Error **errp)
3318 {
3319     BlockJob *job;
3320     job = do_blockdev_backup(arg, NULL, errp);
3321     if (job) {
3322         block_job_start(job);
3323     }
3324 }
3325 
3326 /* Parameter check and block job starting for drive mirroring.
3327  * Caller should hold @device and @target's aio context (must be the same).
3328  **/
3329 static void blockdev_mirror_common(const char *job_id, BlockDriverState *bs,
3330                                    BlockDriverState *target,
3331                                    bool has_replaces, const char *replaces,
3332                                    enum MirrorSyncMode sync,
3333                                    BlockMirrorBackingMode backing_mode,
3334                                    bool has_speed, int64_t speed,
3335                                    bool has_granularity, uint32_t granularity,
3336                                    bool has_buf_size, int64_t buf_size,
3337                                    bool has_on_source_error,
3338                                    BlockdevOnError on_source_error,
3339                                    bool has_on_target_error,
3340                                    BlockdevOnError on_target_error,
3341                                    bool has_unmap, bool unmap,
3342                                    Error **errp)
3343 {
3344 
3345     if (!has_speed) {
3346         speed = 0;
3347     }
3348     if (!has_on_source_error) {
3349         on_source_error = BLOCKDEV_ON_ERROR_REPORT;
3350     }
3351     if (!has_on_target_error) {
3352         on_target_error = BLOCKDEV_ON_ERROR_REPORT;
3353     }
3354     if (!has_granularity) {
3355         granularity = 0;
3356     }
3357     if (!has_buf_size) {
3358         buf_size = 0;
3359     }
3360     if (!has_unmap) {
3361         unmap = true;
3362     }
3363 
3364     if (granularity != 0 && (granularity < 512 || granularity > 1048576 * 64)) {
3365         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "granularity",
3366                    "a value in range [512B, 64MB]");
3367         return;
3368     }
3369     if (granularity & (granularity - 1)) {
3370         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "granularity",
3371                    "power of 2");
3372         return;
3373     }
3374 
3375     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_MIRROR_SOURCE, errp)) {
3376         return;
3377     }
3378     if (bdrv_op_is_blocked(target, BLOCK_OP_TYPE_MIRROR_TARGET, errp)) {
3379         return;
3380     }
3381 
3382     if (!bs->backing && sync == MIRROR_SYNC_MODE_TOP) {
3383         sync = MIRROR_SYNC_MODE_FULL;
3384     }
3385 
3386     /* pass the node name to replace to mirror start since it's loose coupling
3387      * and will allow to check whether the node still exist at mirror completion
3388      */
3389     mirror_start(job_id, bs, target,
3390                  has_replaces ? replaces : NULL,
3391                  speed, granularity, buf_size, sync, backing_mode,
3392                  on_source_error, on_target_error, unmap, errp);
3393 }
3394 
3395 void qmp_drive_mirror(DriveMirror *arg, Error **errp)
3396 {
3397     BlockDriverState *bs;
3398     BlockDriverState *source, *target_bs;
3399     AioContext *aio_context;
3400     BlockMirrorBackingMode backing_mode;
3401     Error *local_err = NULL;
3402     QDict *options = NULL;
3403     int flags;
3404     int64_t size;
3405     const char *format = arg->format;
3406 
3407     bs = qmp_get_root_bs(arg->device, errp);
3408     if (!bs) {
3409         return;
3410     }
3411 
3412     aio_context = bdrv_get_aio_context(bs);
3413     aio_context_acquire(aio_context);
3414 
3415     if (!arg->has_mode) {
3416         arg->mode = NEW_IMAGE_MODE_ABSOLUTE_PATHS;
3417     }
3418 
3419     if (!arg->has_format) {
3420         format = (arg->mode == NEW_IMAGE_MODE_EXISTING
3421                   ? NULL : bs->drv->format_name);
3422     }
3423 
3424     flags = bs->open_flags | BDRV_O_RDWR;
3425     source = backing_bs(bs);
3426     if (!source && arg->sync == MIRROR_SYNC_MODE_TOP) {
3427         arg->sync = MIRROR_SYNC_MODE_FULL;
3428     }
3429     if (arg->sync == MIRROR_SYNC_MODE_NONE) {
3430         source = bs;
3431     }
3432 
3433     size = bdrv_getlength(bs);
3434     if (size < 0) {
3435         error_setg_errno(errp, -size, "bdrv_getlength failed");
3436         goto out;
3437     }
3438 
3439     if (arg->has_replaces) {
3440         BlockDriverState *to_replace_bs;
3441         AioContext *replace_aio_context;
3442         int64_t replace_size;
3443 
3444         if (!arg->has_node_name) {
3445             error_setg(errp, "a node-name must be provided when replacing a"
3446                              " named node of the graph");
3447             goto out;
3448         }
3449 
3450         to_replace_bs = check_to_replace_node(bs, arg->replaces, &local_err);
3451 
3452         if (!to_replace_bs) {
3453             error_propagate(errp, local_err);
3454             goto out;
3455         }
3456 
3457         replace_aio_context = bdrv_get_aio_context(to_replace_bs);
3458         aio_context_acquire(replace_aio_context);
3459         replace_size = bdrv_getlength(to_replace_bs);
3460         aio_context_release(replace_aio_context);
3461 
3462         if (size != replace_size) {
3463             error_setg(errp, "cannot replace image with a mirror image of "
3464                              "different size");
3465             goto out;
3466         }
3467     }
3468 
3469     if (arg->mode == NEW_IMAGE_MODE_ABSOLUTE_PATHS) {
3470         backing_mode = MIRROR_SOURCE_BACKING_CHAIN;
3471     } else {
3472         backing_mode = MIRROR_OPEN_BACKING_CHAIN;
3473     }
3474 
3475     if ((arg->sync == MIRROR_SYNC_MODE_FULL || !source)
3476         && arg->mode != NEW_IMAGE_MODE_EXISTING)
3477     {
3478         /* create new image w/o backing file */
3479         assert(format);
3480         bdrv_img_create(arg->target, format,
3481                         NULL, NULL, NULL, size, flags, &local_err, false);
3482     } else {
3483         switch (arg->mode) {
3484         case NEW_IMAGE_MODE_EXISTING:
3485             break;
3486         case NEW_IMAGE_MODE_ABSOLUTE_PATHS:
3487             /* create new image with backing file */
3488             bdrv_img_create(arg->target, format,
3489                             source->filename,
3490                             source->drv->format_name,
3491                             NULL, size, flags, &local_err, false);
3492             break;
3493         default:
3494             abort();
3495         }
3496     }
3497 
3498     if (local_err) {
3499         error_propagate(errp, local_err);
3500         goto out;
3501     }
3502 
3503     options = qdict_new();
3504     if (arg->has_node_name) {
3505         qdict_put(options, "node-name", qstring_from_str(arg->node_name));
3506     }
3507     if (format) {
3508         qdict_put(options, "driver", qstring_from_str(format));
3509     }
3510 
3511     /* Mirroring takes care of copy-on-write using the source's backing
3512      * file.
3513      */
3514     target_bs = bdrv_open(arg->target, NULL, options,
3515                           flags | BDRV_O_NO_BACKING, errp);
3516     if (!target_bs) {
3517         goto out;
3518     }
3519 
3520     bdrv_set_aio_context(target_bs, aio_context);
3521 
3522     blockdev_mirror_common(arg->has_job_id ? arg->job_id : NULL, bs, target_bs,
3523                            arg->has_replaces, arg->replaces, arg->sync,
3524                            backing_mode, arg->has_speed, arg->speed,
3525                            arg->has_granularity, arg->granularity,
3526                            arg->has_buf_size, arg->buf_size,
3527                            arg->has_on_source_error, arg->on_source_error,
3528                            arg->has_on_target_error, arg->on_target_error,
3529                            arg->has_unmap, arg->unmap,
3530                            &local_err);
3531     bdrv_unref(target_bs);
3532     error_propagate(errp, local_err);
3533 out:
3534     aio_context_release(aio_context);
3535 }
3536 
3537 void qmp_blockdev_mirror(bool has_job_id, const char *job_id,
3538                          const char *device, const char *target,
3539                          bool has_replaces, const char *replaces,
3540                          MirrorSyncMode sync,
3541                          bool has_speed, int64_t speed,
3542                          bool has_granularity, uint32_t granularity,
3543                          bool has_buf_size, int64_t buf_size,
3544                          bool has_on_source_error,
3545                          BlockdevOnError on_source_error,
3546                          bool has_on_target_error,
3547                          BlockdevOnError on_target_error,
3548                          Error **errp)
3549 {
3550     BlockDriverState *bs;
3551     BlockDriverState *target_bs;
3552     AioContext *aio_context;
3553     BlockMirrorBackingMode backing_mode = MIRROR_LEAVE_BACKING_CHAIN;
3554     Error *local_err = NULL;
3555 
3556     bs = qmp_get_root_bs(device, errp);
3557     if (!bs) {
3558         return;
3559     }
3560 
3561     target_bs = bdrv_lookup_bs(target, target, errp);
3562     if (!target_bs) {
3563         return;
3564     }
3565 
3566     aio_context = bdrv_get_aio_context(bs);
3567     aio_context_acquire(aio_context);
3568 
3569     bdrv_set_aio_context(target_bs, aio_context);
3570 
3571     blockdev_mirror_common(has_job_id ? job_id : NULL, bs, target_bs,
3572                            has_replaces, replaces, sync, backing_mode,
3573                            has_speed, speed,
3574                            has_granularity, granularity,
3575                            has_buf_size, buf_size,
3576                            has_on_source_error, on_source_error,
3577                            has_on_target_error, on_target_error,
3578                            true, true,
3579                            &local_err);
3580     error_propagate(errp, local_err);
3581 
3582     aio_context_release(aio_context);
3583 }
3584 
3585 /* Get a block job using its ID and acquire its AioContext */
3586 static BlockJob *find_block_job(const char *id, AioContext **aio_context,
3587                                 Error **errp)
3588 {
3589     BlockJob *job;
3590 
3591     assert(id != NULL);
3592 
3593     *aio_context = NULL;
3594 
3595     job = block_job_get(id);
3596 
3597     if (!job) {
3598         error_set(errp, ERROR_CLASS_DEVICE_NOT_ACTIVE,
3599                   "Block job '%s' not found", id);
3600         return NULL;
3601     }
3602 
3603     *aio_context = blk_get_aio_context(job->blk);
3604     aio_context_acquire(*aio_context);
3605 
3606     return job;
3607 }
3608 
3609 void qmp_block_job_set_speed(const char *device, int64_t speed, Error **errp)
3610 {
3611     AioContext *aio_context;
3612     BlockJob *job = find_block_job(device, &aio_context, errp);
3613 
3614     if (!job) {
3615         return;
3616     }
3617 
3618     block_job_set_speed(job, speed, errp);
3619     aio_context_release(aio_context);
3620 }
3621 
3622 void qmp_block_job_cancel(const char *device,
3623                           bool has_force, bool force, Error **errp)
3624 {
3625     AioContext *aio_context;
3626     BlockJob *job = find_block_job(device, &aio_context, errp);
3627 
3628     if (!job) {
3629         return;
3630     }
3631 
3632     if (!has_force) {
3633         force = false;
3634     }
3635 
3636     if (block_job_user_paused(job) && !force) {
3637         error_setg(errp, "The block job for device '%s' is currently paused",
3638                    device);
3639         goto out;
3640     }
3641 
3642     trace_qmp_block_job_cancel(job);
3643     block_job_cancel(job);
3644 out:
3645     aio_context_release(aio_context);
3646 }
3647 
3648 void qmp_block_job_pause(const char *device, Error **errp)
3649 {
3650     AioContext *aio_context;
3651     BlockJob *job = find_block_job(device, &aio_context, errp);
3652 
3653     if (!job || block_job_user_paused(job)) {
3654         return;
3655     }
3656 
3657     trace_qmp_block_job_pause(job);
3658     block_job_user_pause(job);
3659     aio_context_release(aio_context);
3660 }
3661 
3662 void qmp_block_job_resume(const char *device, Error **errp)
3663 {
3664     AioContext *aio_context;
3665     BlockJob *job = find_block_job(device, &aio_context, errp);
3666 
3667     if (!job || !block_job_user_paused(job)) {
3668         return;
3669     }
3670 
3671     trace_qmp_block_job_resume(job);
3672     block_job_iostatus_reset(job);
3673     block_job_user_resume(job);
3674     aio_context_release(aio_context);
3675 }
3676 
3677 void qmp_block_job_complete(const char *device, Error **errp)
3678 {
3679     AioContext *aio_context;
3680     BlockJob *job = find_block_job(device, &aio_context, errp);
3681 
3682     if (!job) {
3683         return;
3684     }
3685 
3686     trace_qmp_block_job_complete(job);
3687     block_job_complete(job, errp);
3688     aio_context_release(aio_context);
3689 }
3690 
3691 void qmp_change_backing_file(const char *device,
3692                              const char *image_node_name,
3693                              const char *backing_file,
3694                              Error **errp)
3695 {
3696     BlockDriverState *bs = NULL;
3697     AioContext *aio_context;
3698     BlockDriverState *image_bs = NULL;
3699     Error *local_err = NULL;
3700     bool ro;
3701     int open_flags;
3702     int ret;
3703 
3704     bs = qmp_get_root_bs(device, errp);
3705     if (!bs) {
3706         return;
3707     }
3708 
3709     aio_context = bdrv_get_aio_context(bs);
3710     aio_context_acquire(aio_context);
3711 
3712     image_bs = bdrv_lookup_bs(NULL, image_node_name, &local_err);
3713     if (local_err) {
3714         error_propagate(errp, local_err);
3715         goto out;
3716     }
3717 
3718     if (!image_bs) {
3719         error_setg(errp, "image file not found");
3720         goto out;
3721     }
3722 
3723     if (bdrv_find_base(image_bs) == image_bs) {
3724         error_setg(errp, "not allowing backing file change on an image "
3725                          "without a backing file");
3726         goto out;
3727     }
3728 
3729     /* even though we are not necessarily operating on bs, we need it to
3730      * determine if block ops are currently prohibited on the chain */
3731     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_CHANGE, errp)) {
3732         goto out;
3733     }
3734 
3735     /* final sanity check */
3736     if (!bdrv_chain_contains(bs, image_bs)) {
3737         error_setg(errp, "'%s' and image file are not in the same chain",
3738                    device);
3739         goto out;
3740     }
3741 
3742     /* if not r/w, reopen to make r/w */
3743     open_flags = image_bs->open_flags;
3744     ro = bdrv_is_read_only(image_bs);
3745 
3746     if (ro) {
3747         bdrv_reopen(image_bs, open_flags | BDRV_O_RDWR, &local_err);
3748         if (local_err) {
3749             error_propagate(errp, local_err);
3750             goto out;
3751         }
3752     }
3753 
3754     ret = bdrv_change_backing_file(image_bs, backing_file,
3755                                image_bs->drv ? image_bs->drv->format_name : "");
3756 
3757     if (ret < 0) {
3758         error_setg_errno(errp, -ret, "Could not change backing file to '%s'",
3759                          backing_file);
3760         /* don't exit here, so we can try to restore open flags if
3761          * appropriate */
3762     }
3763 
3764     if (ro) {
3765         bdrv_reopen(image_bs, open_flags, &local_err);
3766         error_propagate(errp, local_err);
3767     }
3768 
3769 out:
3770     aio_context_release(aio_context);
3771 }
3772 
3773 void hmp_drive_add_node(Monitor *mon, const char *optstr)
3774 {
3775     QemuOpts *opts;
3776     QDict *qdict;
3777     Error *local_err = NULL;
3778 
3779     opts = qemu_opts_parse_noisily(&qemu_drive_opts, optstr, false);
3780     if (!opts) {
3781         return;
3782     }
3783 
3784     qdict = qemu_opts_to_qdict(opts, NULL);
3785 
3786     if (!qdict_get_try_str(qdict, "node-name")) {
3787         QDECREF(qdict);
3788         error_report("'node-name' needs to be specified");
3789         goto out;
3790     }
3791 
3792     BlockDriverState *bs = bds_tree_init(qdict, &local_err);
3793     if (!bs) {
3794         error_report_err(local_err);
3795         goto out;
3796     }
3797 
3798     QTAILQ_INSERT_TAIL(&monitor_bdrv_states, bs, monitor_list);
3799 
3800 out:
3801     qemu_opts_del(opts);
3802 }
3803 
3804 void qmp_blockdev_add(BlockdevOptions *options, Error **errp)
3805 {
3806     BlockDriverState *bs;
3807     QObject *obj;
3808     Visitor *v = qobject_output_visitor_new(&obj);
3809     QDict *qdict;
3810     Error *local_err = NULL;
3811 
3812     visit_type_BlockdevOptions(v, NULL, &options, &local_err);
3813     if (local_err) {
3814         error_propagate(errp, local_err);
3815         goto fail;
3816     }
3817 
3818     visit_complete(v, &obj);
3819     qdict = qobject_to_qdict(obj);
3820 
3821     qdict_flatten(qdict);
3822 
3823     if (!qdict_get_try_str(qdict, "node-name")) {
3824         error_setg(errp, "'node-name' must be specified for the root node");
3825         goto fail;
3826     }
3827 
3828     bs = bds_tree_init(qdict, errp);
3829     if (!bs) {
3830         goto fail;
3831     }
3832 
3833     QTAILQ_INSERT_TAIL(&monitor_bdrv_states, bs, monitor_list);
3834 
3835     if (bs && bdrv_key_required(bs)) {
3836         QTAILQ_REMOVE(&monitor_bdrv_states, bs, monitor_list);
3837         bdrv_unref(bs);
3838         error_setg(errp, "blockdev-add doesn't support encrypted devices");
3839         goto fail;
3840     }
3841 
3842 fail:
3843     visit_free(v);
3844 }
3845 
3846 void qmp_x_blockdev_del(const char *node_name, Error **errp)
3847 {
3848     AioContext *aio_context;
3849     BlockDriverState *bs;
3850 
3851     bs = bdrv_find_node(node_name);
3852     if (!bs) {
3853         error_setg(errp, "Cannot find node %s", node_name);
3854         return;
3855     }
3856     if (bdrv_has_blk(bs)) {
3857         error_setg(errp, "Node %s is in use", node_name);
3858         return;
3859     }
3860     aio_context = bdrv_get_aio_context(bs);
3861     aio_context_acquire(aio_context);
3862 
3863     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_DRIVE_DEL, errp)) {
3864         goto out;
3865     }
3866 
3867     if (!bs->monitor_list.tqe_prev) {
3868         error_setg(errp, "Node %s is not owned by the monitor",
3869                    bs->node_name);
3870         goto out;
3871     }
3872 
3873     if (bs->refcnt > 1) {
3874         error_setg(errp, "Block device %s is in use",
3875                    bdrv_get_device_or_node_name(bs));
3876         goto out;
3877     }
3878 
3879     QTAILQ_REMOVE(&monitor_bdrv_states, bs, monitor_list);
3880     bdrv_unref(bs);
3881 
3882 out:
3883     aio_context_release(aio_context);
3884 }
3885 
3886 static BdrvChild *bdrv_find_child(BlockDriverState *parent_bs,
3887                                   const char *child_name)
3888 {
3889     BdrvChild *child;
3890 
3891     QLIST_FOREACH(child, &parent_bs->children, next) {
3892         if (strcmp(child->name, child_name) == 0) {
3893             return child;
3894         }
3895     }
3896 
3897     return NULL;
3898 }
3899 
3900 void qmp_x_blockdev_change(const char *parent, bool has_child,
3901                            const char *child, bool has_node,
3902                            const char *node, Error **errp)
3903 {
3904     BlockDriverState *parent_bs, *new_bs = NULL;
3905     BdrvChild *p_child;
3906 
3907     parent_bs = bdrv_lookup_bs(parent, parent, errp);
3908     if (!parent_bs) {
3909         return;
3910     }
3911 
3912     if (has_child == has_node) {
3913         if (has_child) {
3914             error_setg(errp, "The parameters child and node are in conflict");
3915         } else {
3916             error_setg(errp, "Either child or node must be specified");
3917         }
3918         return;
3919     }
3920 
3921     if (has_child) {
3922         p_child = bdrv_find_child(parent_bs, child);
3923         if (!p_child) {
3924             error_setg(errp, "Node '%s' does not have child '%s'",
3925                        parent, child);
3926             return;
3927         }
3928         bdrv_del_child(parent_bs, p_child, errp);
3929     }
3930 
3931     if (has_node) {
3932         new_bs = bdrv_find_node(node);
3933         if (!new_bs) {
3934             error_setg(errp, "Node '%s' not found", node);
3935             return;
3936         }
3937         bdrv_add_child(parent_bs, new_bs, errp);
3938     }
3939 }
3940 
3941 BlockJobInfoList *qmp_query_block_jobs(Error **errp)
3942 {
3943     BlockJobInfoList *head = NULL, **p_next = &head;
3944     BlockJob *job;
3945 
3946     for (job = block_job_next(NULL); job; job = block_job_next(job)) {
3947         BlockJobInfoList *elem;
3948         AioContext *aio_context;
3949 
3950         if (block_job_is_internal(job)) {
3951             continue;
3952         }
3953         elem = g_new0(BlockJobInfoList, 1);
3954         aio_context = blk_get_aio_context(job->blk);
3955         aio_context_acquire(aio_context);
3956         elem->value = block_job_query(job, errp);
3957         aio_context_release(aio_context);
3958         if (!elem->value) {
3959             g_free(elem);
3960             qapi_free_BlockJobInfoList(head);
3961             return NULL;
3962         }
3963         *p_next = elem;
3964         p_next = &elem->next;
3965     }
3966 
3967     return head;
3968 }
3969 
3970 QemuOptsList qemu_common_drive_opts = {
3971     .name = "drive",
3972     .head = QTAILQ_HEAD_INITIALIZER(qemu_common_drive_opts.head),
3973     .desc = {
3974         {
3975             .name = "snapshot",
3976             .type = QEMU_OPT_BOOL,
3977             .help = "enable/disable snapshot mode",
3978         },{
3979             .name = "aio",
3980             .type = QEMU_OPT_STRING,
3981             .help = "host AIO implementation (threads, native)",
3982         },{
3983             .name = BDRV_OPT_CACHE_WB,
3984             .type = QEMU_OPT_BOOL,
3985             .help = "Enable writeback mode",
3986         },{
3987             .name = "format",
3988             .type = QEMU_OPT_STRING,
3989             .help = "disk format (raw, qcow2, ...)",
3990         },{
3991             .name = "rerror",
3992             .type = QEMU_OPT_STRING,
3993             .help = "read error action",
3994         },{
3995             .name = "werror",
3996             .type = QEMU_OPT_STRING,
3997             .help = "write error action",
3998         },{
3999             .name = BDRV_OPT_READ_ONLY,
4000             .type = QEMU_OPT_BOOL,
4001             .help = "open drive file as read-only",
4002         },{
4003             .name = "throttling.iops-total",
4004             .type = QEMU_OPT_NUMBER,
4005             .help = "limit total I/O operations per second",
4006         },{
4007             .name = "throttling.iops-read",
4008             .type = QEMU_OPT_NUMBER,
4009             .help = "limit read operations per second",
4010         },{
4011             .name = "throttling.iops-write",
4012             .type = QEMU_OPT_NUMBER,
4013             .help = "limit write operations per second",
4014         },{
4015             .name = "throttling.bps-total",
4016             .type = QEMU_OPT_NUMBER,
4017             .help = "limit total bytes per second",
4018         },{
4019             .name = "throttling.bps-read",
4020             .type = QEMU_OPT_NUMBER,
4021             .help = "limit read bytes per second",
4022         },{
4023             .name = "throttling.bps-write",
4024             .type = QEMU_OPT_NUMBER,
4025             .help = "limit write bytes per second",
4026         },{
4027             .name = "throttling.iops-total-max",
4028             .type = QEMU_OPT_NUMBER,
4029             .help = "I/O operations burst",
4030         },{
4031             .name = "throttling.iops-read-max",
4032             .type = QEMU_OPT_NUMBER,
4033             .help = "I/O operations read burst",
4034         },{
4035             .name = "throttling.iops-write-max",
4036             .type = QEMU_OPT_NUMBER,
4037             .help = "I/O operations write burst",
4038         },{
4039             .name = "throttling.bps-total-max",
4040             .type = QEMU_OPT_NUMBER,
4041             .help = "total bytes burst",
4042         },{
4043             .name = "throttling.bps-read-max",
4044             .type = QEMU_OPT_NUMBER,
4045             .help = "total bytes read burst",
4046         },{
4047             .name = "throttling.bps-write-max",
4048             .type = QEMU_OPT_NUMBER,
4049             .help = "total bytes write burst",
4050         },{
4051             .name = "throttling.iops-total-max-length",
4052             .type = QEMU_OPT_NUMBER,
4053             .help = "length of the iops-total-max burst period, in seconds",
4054         },{
4055             .name = "throttling.iops-read-max-length",
4056             .type = QEMU_OPT_NUMBER,
4057             .help = "length of the iops-read-max burst period, in seconds",
4058         },{
4059             .name = "throttling.iops-write-max-length",
4060             .type = QEMU_OPT_NUMBER,
4061             .help = "length of the iops-write-max burst period, in seconds",
4062         },{
4063             .name = "throttling.bps-total-max-length",
4064             .type = QEMU_OPT_NUMBER,
4065             .help = "length of the bps-total-max burst period, in seconds",
4066         },{
4067             .name = "throttling.bps-read-max-length",
4068             .type = QEMU_OPT_NUMBER,
4069             .help = "length of the bps-read-max burst period, in seconds",
4070         },{
4071             .name = "throttling.bps-write-max-length",
4072             .type = QEMU_OPT_NUMBER,
4073             .help = "length of the bps-write-max burst period, in seconds",
4074         },{
4075             .name = "throttling.iops-size",
4076             .type = QEMU_OPT_NUMBER,
4077             .help = "when limiting by iops max size of an I/O in bytes",
4078         },{
4079             .name = "throttling.group",
4080             .type = QEMU_OPT_STRING,
4081             .help = "name of the block throttling group",
4082         },{
4083             .name = "copy-on-read",
4084             .type = QEMU_OPT_BOOL,
4085             .help = "copy read data from backing file into image file",
4086         },{
4087             .name = "detect-zeroes",
4088             .type = QEMU_OPT_STRING,
4089             .help = "try to optimize zero writes (off, on, unmap)",
4090         },{
4091             .name = "stats-account-invalid",
4092             .type = QEMU_OPT_BOOL,
4093             .help = "whether to account for invalid I/O operations "
4094                     "in the statistics",
4095         },{
4096             .name = "stats-account-failed",
4097             .type = QEMU_OPT_BOOL,
4098             .help = "whether to account for failed I/O operations "
4099                     "in the statistics",
4100         },
4101         { /* end of list */ }
4102     },
4103 };
4104 
4105 QemuOptsList qemu_drive_opts = {
4106     .name = "drive",
4107     .head = QTAILQ_HEAD_INITIALIZER(qemu_drive_opts.head),
4108     .desc = {
4109         /*
4110          * no elements => accept any params
4111          * validation will happen later
4112          */
4113         { /* end of list */ }
4114     },
4115 };
4116