xref: /openbmc/qemu/blockdev.c (revision fafa4d50)
1 /*
2  * QEMU host block devices
3  *
4  * Copyright (c) 2003-2008 Fabrice Bellard
5  *
6  * This work is licensed under the terms of the GNU GPL, version 2 or
7  * later.  See the COPYING file in the top-level directory.
8  *
9  * This file incorporates work covered by the following copyright and
10  * permission notice:
11  *
12  * Copyright (c) 2003-2008 Fabrice Bellard
13  *
14  * Permission is hereby granted, free of charge, to any person obtaining a copy
15  * of this software and associated documentation files (the "Software"), to deal
16  * in the Software without restriction, including without limitation the rights
17  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
18  * copies of the Software, and to permit persons to whom the Software is
19  * furnished to do so, subject to the following conditions:
20  *
21  * The above copyright notice and this permission notice shall be included in
22  * all copies or substantial portions of the Software.
23  *
24  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
25  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
26  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
27  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
28  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
29  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
30  * THE SOFTWARE.
31  */
32 
33 #include "sysemu/block-backend.h"
34 #include "sysemu/blockdev.h"
35 #include "hw/block/block.h"
36 #include "block/blockjob.h"
37 #include "monitor/monitor.h"
38 #include "qemu/option.h"
39 #include "qemu/config-file.h"
40 #include "qapi/qmp/types.h"
41 #include "qapi-visit.h"
42 #include "qapi/qmp-output-visitor.h"
43 #include "qapi/util.h"
44 #include "sysemu/sysemu.h"
45 #include "block/block_int.h"
46 #include "qmp-commands.h"
47 #include "trace.h"
48 #include "sysemu/arch_init.h"
49 
50 static const char *const if_name[IF_COUNT] = {
51     [IF_NONE] = "none",
52     [IF_IDE] = "ide",
53     [IF_SCSI] = "scsi",
54     [IF_FLOPPY] = "floppy",
55     [IF_PFLASH] = "pflash",
56     [IF_MTD] = "mtd",
57     [IF_SD] = "sd",
58     [IF_VIRTIO] = "virtio",
59     [IF_XEN] = "xen",
60 };
61 
62 static int if_max_devs[IF_COUNT] = {
63     /*
64      * Do not change these numbers!  They govern how drive option
65      * index maps to unit and bus.  That mapping is ABI.
66      *
67      * All controllers used to imlement if=T drives need to support
68      * if_max_devs[T] units, for any T with if_max_devs[T] != 0.
69      * Otherwise, some index values map to "impossible" bus, unit
70      * values.
71      *
72      * For instance, if you change [IF_SCSI] to 255, -drive
73      * if=scsi,index=12 no longer means bus=1,unit=5, but
74      * bus=0,unit=12.  With an lsi53c895a controller (7 units max),
75      * the drive can't be set up.  Regression.
76      */
77     [IF_IDE] = 2,
78     [IF_SCSI] = 7,
79 };
80 
81 /**
82  * Boards may call this to offer board-by-board overrides
83  * of the default, global values.
84  */
85 void override_max_devs(BlockInterfaceType type, int max_devs)
86 {
87     BlockBackend *blk;
88     DriveInfo *dinfo;
89 
90     if (max_devs <= 0) {
91         return;
92     }
93 
94     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
95         dinfo = blk_legacy_dinfo(blk);
96         if (dinfo->type == type) {
97             fprintf(stderr, "Cannot override units-per-bus property of"
98                     " the %s interface, because a drive of that type has"
99                     " already been added.\n", if_name[type]);
100             g_assert_not_reached();
101         }
102     }
103 
104     if_max_devs[type] = max_devs;
105 }
106 
107 /*
108  * We automatically delete the drive when a device using it gets
109  * unplugged.  Questionable feature, but we can't just drop it.
110  * Device models call blockdev_mark_auto_del() to schedule the
111  * automatic deletion, and generic qdev code calls blockdev_auto_del()
112  * when deletion is actually safe.
113  */
114 void blockdev_mark_auto_del(BlockBackend *blk)
115 {
116     DriveInfo *dinfo = blk_legacy_dinfo(blk);
117     BlockDriverState *bs = blk_bs(blk);
118     AioContext *aio_context;
119 
120     if (!dinfo) {
121         return;
122     }
123 
124     aio_context = bdrv_get_aio_context(bs);
125     aio_context_acquire(aio_context);
126 
127     if (bs->job) {
128         block_job_cancel(bs->job);
129     }
130 
131     aio_context_release(aio_context);
132 
133     dinfo->auto_del = 1;
134 }
135 
136 void blockdev_auto_del(BlockBackend *blk)
137 {
138     DriveInfo *dinfo = blk_legacy_dinfo(blk);
139 
140     if (dinfo && dinfo->auto_del) {
141         blk_unref(blk);
142     }
143 }
144 
145 /**
146  * Returns the current mapping of how many units per bus
147  * a particular interface can support.
148  *
149  *  A positive integer indicates n units per bus.
150  *  0 implies the mapping has not been established.
151  * -1 indicates an invalid BlockInterfaceType was given.
152  */
153 int drive_get_max_devs(BlockInterfaceType type)
154 {
155     if (type >= IF_IDE && type < IF_COUNT) {
156         return if_max_devs[type];
157     }
158 
159     return -1;
160 }
161 
162 static int drive_index_to_bus_id(BlockInterfaceType type, int index)
163 {
164     int max_devs = if_max_devs[type];
165     return max_devs ? index / max_devs : 0;
166 }
167 
168 static int drive_index_to_unit_id(BlockInterfaceType type, int index)
169 {
170     int max_devs = if_max_devs[type];
171     return max_devs ? index % max_devs : index;
172 }
173 
174 QemuOpts *drive_def(const char *optstr)
175 {
176     return qemu_opts_parse(qemu_find_opts("drive"), optstr, 0);
177 }
178 
179 QemuOpts *drive_add(BlockInterfaceType type, int index, const char *file,
180                     const char *optstr)
181 {
182     QemuOpts *opts;
183 
184     opts = drive_def(optstr);
185     if (!opts) {
186         return NULL;
187     }
188     if (type != IF_DEFAULT) {
189         qemu_opt_set(opts, "if", if_name[type], &error_abort);
190     }
191     if (index >= 0) {
192         qemu_opt_set_number(opts, "index", index, &error_abort);
193     }
194     if (file)
195         qemu_opt_set(opts, "file", file, &error_abort);
196     return opts;
197 }
198 
199 DriveInfo *drive_get(BlockInterfaceType type, int bus, int unit)
200 {
201     BlockBackend *blk;
202     DriveInfo *dinfo;
203 
204     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
205         dinfo = blk_legacy_dinfo(blk);
206         if (dinfo && dinfo->type == type
207             && dinfo->bus == bus && dinfo->unit == unit) {
208             return dinfo;
209         }
210     }
211 
212     return NULL;
213 }
214 
215 bool drive_check_orphaned(void)
216 {
217     BlockBackend *blk;
218     DriveInfo *dinfo;
219     bool rs = false;
220 
221     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
222         dinfo = blk_legacy_dinfo(blk);
223         /* If dinfo->bdrv->dev is NULL, it has no device attached. */
224         /* Unless this is a default drive, this may be an oversight. */
225         if (!blk_get_attached_dev(blk) && !dinfo->is_default &&
226             dinfo->type != IF_NONE) {
227             fprintf(stderr, "Warning: Orphaned drive without device: "
228                     "id=%s,file=%s,if=%s,bus=%d,unit=%d\n",
229                     blk_name(blk), blk_bs(blk)->filename, if_name[dinfo->type],
230                     dinfo->bus, dinfo->unit);
231             rs = true;
232         }
233     }
234 
235     return rs;
236 }
237 
238 DriveInfo *drive_get_by_index(BlockInterfaceType type, int index)
239 {
240     return drive_get(type,
241                      drive_index_to_bus_id(type, index),
242                      drive_index_to_unit_id(type, index));
243 }
244 
245 int drive_get_max_bus(BlockInterfaceType type)
246 {
247     int max_bus;
248     BlockBackend *blk;
249     DriveInfo *dinfo;
250 
251     max_bus = -1;
252     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
253         dinfo = blk_legacy_dinfo(blk);
254         if (dinfo && dinfo->type == type && dinfo->bus > max_bus) {
255             max_bus = dinfo->bus;
256         }
257     }
258     return max_bus;
259 }
260 
261 /* Get a block device.  This should only be used for single-drive devices
262    (e.g. SD/Floppy/MTD).  Multi-disk devices (scsi/ide) should use the
263    appropriate bus.  */
264 DriveInfo *drive_get_next(BlockInterfaceType type)
265 {
266     static int next_block_unit[IF_COUNT];
267 
268     return drive_get(type, 0, next_block_unit[type]++);
269 }
270 
271 static void bdrv_format_print(void *opaque, const char *name)
272 {
273     error_printf(" %s", name);
274 }
275 
276 typedef struct {
277     QEMUBH *bh;
278     BlockDriverState *bs;
279 } BDRVPutRefBH;
280 
281 static void bdrv_put_ref_bh(void *opaque)
282 {
283     BDRVPutRefBH *s = opaque;
284 
285     bdrv_unref(s->bs);
286     qemu_bh_delete(s->bh);
287     g_free(s);
288 }
289 
290 /*
291  * Release a BDS reference in a BH
292  *
293  * It is not safe to use bdrv_unref() from a callback function when the callers
294  * still need the BlockDriverState.  In such cases we schedule a BH to release
295  * the reference.
296  */
297 static void bdrv_put_ref_bh_schedule(BlockDriverState *bs)
298 {
299     BDRVPutRefBH *s;
300 
301     s = g_new(BDRVPutRefBH, 1);
302     s->bh = qemu_bh_new(bdrv_put_ref_bh, s);
303     s->bs = bs;
304     qemu_bh_schedule(s->bh);
305 }
306 
307 static int parse_block_error_action(const char *buf, bool is_read, Error **errp)
308 {
309     if (!strcmp(buf, "ignore")) {
310         return BLOCKDEV_ON_ERROR_IGNORE;
311     } else if (!is_read && !strcmp(buf, "enospc")) {
312         return BLOCKDEV_ON_ERROR_ENOSPC;
313     } else if (!strcmp(buf, "stop")) {
314         return BLOCKDEV_ON_ERROR_STOP;
315     } else if (!strcmp(buf, "report")) {
316         return BLOCKDEV_ON_ERROR_REPORT;
317     } else {
318         error_setg(errp, "'%s' invalid %s error action",
319                    buf, is_read ? "read" : "write");
320         return -1;
321     }
322 }
323 
324 static bool check_throttle_config(ThrottleConfig *cfg, Error **errp)
325 {
326     if (throttle_conflicting(cfg)) {
327         error_setg(errp, "bps/iops/max total values and read/write values"
328                          " cannot be used at the same time");
329         return false;
330     }
331 
332     if (!throttle_is_valid(cfg)) {
333         error_setg(errp, "bps/iops/maxs values must be 0 or greater");
334         return false;
335     }
336 
337     return true;
338 }
339 
340 typedef enum { MEDIA_DISK, MEDIA_CDROM } DriveMediaType;
341 
342 /* Takes the ownership of bs_opts */
343 static BlockBackend *blockdev_init(const char *file, QDict *bs_opts,
344                                    Error **errp)
345 {
346     const char *buf;
347     int ro = 0;
348     int bdrv_flags = 0;
349     int on_read_error, on_write_error;
350     BlockBackend *blk;
351     BlockDriverState *bs;
352     ThrottleConfig cfg;
353     int snapshot = 0;
354     bool copy_on_read;
355     Error *error = NULL;
356     QemuOpts *opts;
357     const char *id;
358     bool has_driver_specific_opts;
359     BlockdevDetectZeroesOptions detect_zeroes;
360 
361     /* Check common options by copying from bs_opts to opts, all other options
362      * stay in bs_opts for processing by bdrv_open(). */
363     id = qdict_get_try_str(bs_opts, "id");
364     opts = qemu_opts_create(&qemu_common_drive_opts, id, 1, &error);
365     if (error) {
366         error_propagate(errp, error);
367         goto err_no_opts;
368     }
369 
370     qemu_opts_absorb_qdict(opts, bs_opts, &error);
371     if (error) {
372         error_propagate(errp, error);
373         goto early_err;
374     }
375 
376     if (id) {
377         qdict_del(bs_opts, "id");
378     }
379 
380     has_driver_specific_opts = !!qdict_size(bs_opts);
381 
382     /* extract parameters */
383     snapshot = qemu_opt_get_bool(opts, "snapshot", 0);
384     ro = qemu_opt_get_bool(opts, "read-only", 0);
385     copy_on_read = qemu_opt_get_bool(opts, "copy-on-read", false);
386 
387     if ((buf = qemu_opt_get(opts, "discard")) != NULL) {
388         if (bdrv_parse_discard_flags(buf, &bdrv_flags) != 0) {
389             error_setg(errp, "invalid discard option");
390             goto early_err;
391         }
392     }
393 
394     if (qemu_opt_get_bool(opts, "cache.writeback", true)) {
395         bdrv_flags |= BDRV_O_CACHE_WB;
396     }
397     if (qemu_opt_get_bool(opts, "cache.direct", false)) {
398         bdrv_flags |= BDRV_O_NOCACHE;
399     }
400     if (qemu_opt_get_bool(opts, "cache.no-flush", false)) {
401         bdrv_flags |= BDRV_O_NO_FLUSH;
402     }
403 
404 #ifdef CONFIG_LINUX_AIO
405     if ((buf = qemu_opt_get(opts, "aio")) != NULL) {
406         if (!strcmp(buf, "native")) {
407             bdrv_flags |= BDRV_O_NATIVE_AIO;
408         } else if (!strcmp(buf, "threads")) {
409             /* this is the default */
410         } else {
411            error_setg(errp, "invalid aio option");
412            goto early_err;
413         }
414     }
415 #endif
416 
417     if ((buf = qemu_opt_get(opts, "format")) != NULL) {
418         if (is_help_option(buf)) {
419             error_printf("Supported formats:");
420             bdrv_iterate_format(bdrv_format_print, NULL);
421             error_printf("\n");
422             goto early_err;
423         }
424 
425         if (qdict_haskey(bs_opts, "driver")) {
426             error_setg(errp, "Cannot specify both 'driver' and 'format'");
427             goto early_err;
428         }
429         qdict_put(bs_opts, "driver", qstring_from_str(buf));
430     }
431 
432     /* disk I/O throttling */
433     memset(&cfg, 0, sizeof(cfg));
434     cfg.buckets[THROTTLE_BPS_TOTAL].avg =
435         qemu_opt_get_number(opts, "throttling.bps-total", 0);
436     cfg.buckets[THROTTLE_BPS_READ].avg  =
437         qemu_opt_get_number(opts, "throttling.bps-read", 0);
438     cfg.buckets[THROTTLE_BPS_WRITE].avg =
439         qemu_opt_get_number(opts, "throttling.bps-write", 0);
440     cfg.buckets[THROTTLE_OPS_TOTAL].avg =
441         qemu_opt_get_number(opts, "throttling.iops-total", 0);
442     cfg.buckets[THROTTLE_OPS_READ].avg =
443         qemu_opt_get_number(opts, "throttling.iops-read", 0);
444     cfg.buckets[THROTTLE_OPS_WRITE].avg =
445         qemu_opt_get_number(opts, "throttling.iops-write", 0);
446 
447     cfg.buckets[THROTTLE_BPS_TOTAL].max =
448         qemu_opt_get_number(opts, "throttling.bps-total-max", 0);
449     cfg.buckets[THROTTLE_BPS_READ].max  =
450         qemu_opt_get_number(opts, "throttling.bps-read-max", 0);
451     cfg.buckets[THROTTLE_BPS_WRITE].max =
452         qemu_opt_get_number(opts, "throttling.bps-write-max", 0);
453     cfg.buckets[THROTTLE_OPS_TOTAL].max =
454         qemu_opt_get_number(opts, "throttling.iops-total-max", 0);
455     cfg.buckets[THROTTLE_OPS_READ].max =
456         qemu_opt_get_number(opts, "throttling.iops-read-max", 0);
457     cfg.buckets[THROTTLE_OPS_WRITE].max =
458         qemu_opt_get_number(opts, "throttling.iops-write-max", 0);
459 
460     cfg.op_size = qemu_opt_get_number(opts, "throttling.iops-size", 0);
461 
462     if (!check_throttle_config(&cfg, &error)) {
463         error_propagate(errp, error);
464         goto early_err;
465     }
466 
467     on_write_error = BLOCKDEV_ON_ERROR_ENOSPC;
468     if ((buf = qemu_opt_get(opts, "werror")) != NULL) {
469         on_write_error = parse_block_error_action(buf, 0, &error);
470         if (error) {
471             error_propagate(errp, error);
472             goto early_err;
473         }
474     }
475 
476     on_read_error = BLOCKDEV_ON_ERROR_REPORT;
477     if ((buf = qemu_opt_get(opts, "rerror")) != NULL) {
478         on_read_error = parse_block_error_action(buf, 1, &error);
479         if (error) {
480             error_propagate(errp, error);
481             goto early_err;
482         }
483     }
484 
485     detect_zeroes =
486         qapi_enum_parse(BlockdevDetectZeroesOptions_lookup,
487                         qemu_opt_get(opts, "detect-zeroes"),
488                         BLOCKDEV_DETECT_ZEROES_OPTIONS_MAX,
489                         BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF,
490                         &error);
491     if (error) {
492         error_propagate(errp, error);
493         goto early_err;
494     }
495 
496     if (detect_zeroes == BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP &&
497         !(bdrv_flags & BDRV_O_UNMAP)) {
498         error_setg(errp, "setting detect-zeroes to unmap is not allowed "
499                          "without setting discard operation to unmap");
500         goto early_err;
501     }
502 
503     /* init */
504     if ((!file || !*file) && !has_driver_specific_opts) {
505         blk = blk_new_with_bs(qemu_opts_id(opts), errp);
506         if (!blk) {
507             goto early_err;
508         }
509 
510         bs = blk_bs(blk);
511         bs->open_flags = snapshot ? BDRV_O_SNAPSHOT : 0;
512         bs->read_only = ro;
513 
514         QDECREF(bs_opts);
515     } else {
516         if (file && !*file) {
517             file = NULL;
518         }
519 
520         if (snapshot) {
521             /* always use cache=unsafe with snapshot */
522             bdrv_flags &= ~BDRV_O_CACHE_MASK;
523             bdrv_flags |= (BDRV_O_SNAPSHOT|BDRV_O_CACHE_WB|BDRV_O_NO_FLUSH);
524         }
525 
526         if (copy_on_read) {
527             bdrv_flags |= BDRV_O_COPY_ON_READ;
528         }
529 
530         if (runstate_check(RUN_STATE_INMIGRATE)) {
531             bdrv_flags |= BDRV_O_INCOMING;
532         }
533 
534         bdrv_flags |= ro ? 0 : BDRV_O_RDWR;
535 
536         blk = blk_new_open(qemu_opts_id(opts), file, NULL, bs_opts, bdrv_flags,
537                            errp);
538         if (!blk) {
539             goto err_no_bs_opts;
540         }
541         bs = blk_bs(blk);
542     }
543 
544     bs->detect_zeroes = detect_zeroes;
545 
546     bdrv_set_on_error(bs, on_read_error, on_write_error);
547 
548     /* disk I/O throttling */
549     if (throttle_enabled(&cfg)) {
550         bdrv_io_limits_enable(bs);
551         bdrv_set_io_limits(bs, &cfg);
552     }
553 
554     if (bdrv_key_required(bs)) {
555         autostart = 0;
556     }
557 
558 err_no_bs_opts:
559     qemu_opts_del(opts);
560     return blk;
561 
562 early_err:
563     qemu_opts_del(opts);
564 err_no_opts:
565     QDECREF(bs_opts);
566     return NULL;
567 }
568 
569 static void qemu_opt_rename(QemuOpts *opts, const char *from, const char *to,
570                             Error **errp)
571 {
572     const char *value;
573 
574     value = qemu_opt_get(opts, from);
575     if (value) {
576         if (qemu_opt_find(opts, to)) {
577             error_setg(errp, "'%s' and its alias '%s' can't be used at the "
578                        "same time", to, from);
579             return;
580         }
581     }
582 
583     /* rename all items in opts */
584     while ((value = qemu_opt_get(opts, from))) {
585         qemu_opt_set(opts, to, value, &error_abort);
586         qemu_opt_unset(opts, from);
587     }
588 }
589 
590 QemuOptsList qemu_legacy_drive_opts = {
591     .name = "drive",
592     .head = QTAILQ_HEAD_INITIALIZER(qemu_legacy_drive_opts.head),
593     .desc = {
594         {
595             .name = "bus",
596             .type = QEMU_OPT_NUMBER,
597             .help = "bus number",
598         },{
599             .name = "unit",
600             .type = QEMU_OPT_NUMBER,
601             .help = "unit number (i.e. lun for scsi)",
602         },{
603             .name = "index",
604             .type = QEMU_OPT_NUMBER,
605             .help = "index number",
606         },{
607             .name = "media",
608             .type = QEMU_OPT_STRING,
609             .help = "media type (disk, cdrom)",
610         },{
611             .name = "if",
612             .type = QEMU_OPT_STRING,
613             .help = "interface (ide, scsi, sd, mtd, floppy, pflash, virtio)",
614         },{
615             .name = "cyls",
616             .type = QEMU_OPT_NUMBER,
617             .help = "number of cylinders (ide disk geometry)",
618         },{
619             .name = "heads",
620             .type = QEMU_OPT_NUMBER,
621             .help = "number of heads (ide disk geometry)",
622         },{
623             .name = "secs",
624             .type = QEMU_OPT_NUMBER,
625             .help = "number of sectors (ide disk geometry)",
626         },{
627             .name = "trans",
628             .type = QEMU_OPT_STRING,
629             .help = "chs translation (auto, lba, none)",
630         },{
631             .name = "boot",
632             .type = QEMU_OPT_BOOL,
633             .help = "(deprecated, ignored)",
634         },{
635             .name = "addr",
636             .type = QEMU_OPT_STRING,
637             .help = "pci address (virtio only)",
638         },{
639             .name = "serial",
640             .type = QEMU_OPT_STRING,
641             .help = "disk serial number",
642         },{
643             .name = "file",
644             .type = QEMU_OPT_STRING,
645             .help = "file name",
646         },
647 
648         /* Options that are passed on, but have special semantics with -drive */
649         {
650             .name = "read-only",
651             .type = QEMU_OPT_BOOL,
652             .help = "open drive file as read-only",
653         },{
654             .name = "rerror",
655             .type = QEMU_OPT_STRING,
656             .help = "read error action",
657         },{
658             .name = "werror",
659             .type = QEMU_OPT_STRING,
660             .help = "write error action",
661         },{
662             .name = "copy-on-read",
663             .type = QEMU_OPT_BOOL,
664             .help = "copy read data from backing file into image file",
665         },
666 
667         { /* end of list */ }
668     },
669 };
670 
671 DriveInfo *drive_new(QemuOpts *all_opts, BlockInterfaceType block_default_type)
672 {
673     const char *value;
674     BlockBackend *blk;
675     DriveInfo *dinfo = NULL;
676     QDict *bs_opts;
677     QemuOpts *legacy_opts;
678     DriveMediaType media = MEDIA_DISK;
679     BlockInterfaceType type;
680     int cyls, heads, secs, translation;
681     int max_devs, bus_id, unit_id, index;
682     const char *devaddr;
683     const char *werror, *rerror;
684     bool read_only = false;
685     bool copy_on_read;
686     const char *serial;
687     const char *filename;
688     Error *local_err = NULL;
689     int i;
690 
691     /* Change legacy command line options into QMP ones */
692     static const struct {
693         const char *from;
694         const char *to;
695     } opt_renames[] = {
696         { "iops",           "throttling.iops-total" },
697         { "iops_rd",        "throttling.iops-read" },
698         { "iops_wr",        "throttling.iops-write" },
699 
700         { "bps",            "throttling.bps-total" },
701         { "bps_rd",         "throttling.bps-read" },
702         { "bps_wr",         "throttling.bps-write" },
703 
704         { "iops_max",       "throttling.iops-total-max" },
705         { "iops_rd_max",    "throttling.iops-read-max" },
706         { "iops_wr_max",    "throttling.iops-write-max" },
707 
708         { "bps_max",        "throttling.bps-total-max" },
709         { "bps_rd_max",     "throttling.bps-read-max" },
710         { "bps_wr_max",     "throttling.bps-write-max" },
711 
712         { "iops_size",      "throttling.iops-size" },
713 
714         { "readonly",       "read-only" },
715     };
716 
717     for (i = 0; i < ARRAY_SIZE(opt_renames); i++) {
718         qemu_opt_rename(all_opts, opt_renames[i].from, opt_renames[i].to,
719                         &local_err);
720         if (local_err) {
721             error_report_err(local_err);
722             return NULL;
723         }
724     }
725 
726     value = qemu_opt_get(all_opts, "cache");
727     if (value) {
728         int flags = 0;
729 
730         if (bdrv_parse_cache_flags(value, &flags) != 0) {
731             error_report("invalid cache option");
732             return NULL;
733         }
734 
735         /* Specific options take precedence */
736         if (!qemu_opt_get(all_opts, "cache.writeback")) {
737             qemu_opt_set_bool(all_opts, "cache.writeback",
738                               !!(flags & BDRV_O_CACHE_WB), &error_abort);
739         }
740         if (!qemu_opt_get(all_opts, "cache.direct")) {
741             qemu_opt_set_bool(all_opts, "cache.direct",
742                               !!(flags & BDRV_O_NOCACHE), &error_abort);
743         }
744         if (!qemu_opt_get(all_opts, "cache.no-flush")) {
745             qemu_opt_set_bool(all_opts, "cache.no-flush",
746                               !!(flags & BDRV_O_NO_FLUSH), &error_abort);
747         }
748         qemu_opt_unset(all_opts, "cache");
749     }
750 
751     /* Get a QDict for processing the options */
752     bs_opts = qdict_new();
753     qemu_opts_to_qdict(all_opts, bs_opts);
754 
755     legacy_opts = qemu_opts_create(&qemu_legacy_drive_opts, NULL, 0,
756                                    &error_abort);
757     qemu_opts_absorb_qdict(legacy_opts, bs_opts, &local_err);
758     if (local_err) {
759         error_report_err(local_err);
760         goto fail;
761     }
762 
763     /* Deprecated option boot=[on|off] */
764     if (qemu_opt_get(legacy_opts, "boot") != NULL) {
765         fprintf(stderr, "qemu-kvm: boot=on|off is deprecated and will be "
766                 "ignored. Future versions will reject this parameter. Please "
767                 "update your scripts.\n");
768     }
769 
770     /* Media type */
771     value = qemu_opt_get(legacy_opts, "media");
772     if (value) {
773         if (!strcmp(value, "disk")) {
774             media = MEDIA_DISK;
775         } else if (!strcmp(value, "cdrom")) {
776             media = MEDIA_CDROM;
777             read_only = true;
778         } else {
779             error_report("'%s' invalid media", value);
780             goto fail;
781         }
782     }
783 
784     /* copy-on-read is disabled with a warning for read-only devices */
785     read_only |= qemu_opt_get_bool(legacy_opts, "read-only", false);
786     copy_on_read = qemu_opt_get_bool(legacy_opts, "copy-on-read", false);
787 
788     if (read_only && copy_on_read) {
789         error_report("warning: disabling copy-on-read on read-only drive");
790         copy_on_read = false;
791     }
792 
793     qdict_put(bs_opts, "read-only",
794               qstring_from_str(read_only ? "on" : "off"));
795     qdict_put(bs_opts, "copy-on-read",
796               qstring_from_str(copy_on_read ? "on" :"off"));
797 
798     /* Controller type */
799     value = qemu_opt_get(legacy_opts, "if");
800     if (value) {
801         for (type = 0;
802              type < IF_COUNT && strcmp(value, if_name[type]);
803              type++) {
804         }
805         if (type == IF_COUNT) {
806             error_report("unsupported bus type '%s'", value);
807             goto fail;
808         }
809     } else {
810         type = block_default_type;
811     }
812 
813     /* Geometry */
814     cyls  = qemu_opt_get_number(legacy_opts, "cyls", 0);
815     heads = qemu_opt_get_number(legacy_opts, "heads", 0);
816     secs  = qemu_opt_get_number(legacy_opts, "secs", 0);
817 
818     if (cyls || heads || secs) {
819         if (cyls < 1) {
820             error_report("invalid physical cyls number");
821             goto fail;
822         }
823         if (heads < 1) {
824             error_report("invalid physical heads number");
825             goto fail;
826         }
827         if (secs < 1) {
828             error_report("invalid physical secs number");
829             goto fail;
830         }
831     }
832 
833     translation = BIOS_ATA_TRANSLATION_AUTO;
834     value = qemu_opt_get(legacy_opts, "trans");
835     if (value != NULL) {
836         if (!cyls) {
837             error_report("'%s' trans must be used with cyls, heads and secs",
838                          value);
839             goto fail;
840         }
841         if (!strcmp(value, "none")) {
842             translation = BIOS_ATA_TRANSLATION_NONE;
843         } else if (!strcmp(value, "lba")) {
844             translation = BIOS_ATA_TRANSLATION_LBA;
845         } else if (!strcmp(value, "large")) {
846             translation = BIOS_ATA_TRANSLATION_LARGE;
847         } else if (!strcmp(value, "rechs")) {
848             translation = BIOS_ATA_TRANSLATION_RECHS;
849         } else if (!strcmp(value, "auto")) {
850             translation = BIOS_ATA_TRANSLATION_AUTO;
851         } else {
852             error_report("'%s' invalid translation type", value);
853             goto fail;
854         }
855     }
856 
857     if (media == MEDIA_CDROM) {
858         if (cyls || secs || heads) {
859             error_report("CHS can't be set with media=cdrom");
860             goto fail;
861         }
862     }
863 
864     /* Device address specified by bus/unit or index.
865      * If none was specified, try to find the first free one. */
866     bus_id  = qemu_opt_get_number(legacy_opts, "bus", 0);
867     unit_id = qemu_opt_get_number(legacy_opts, "unit", -1);
868     index   = qemu_opt_get_number(legacy_opts, "index", -1);
869 
870     max_devs = if_max_devs[type];
871 
872     if (index != -1) {
873         if (bus_id != 0 || unit_id != -1) {
874             error_report("index cannot be used with bus and unit");
875             goto fail;
876         }
877         bus_id = drive_index_to_bus_id(type, index);
878         unit_id = drive_index_to_unit_id(type, index);
879     }
880 
881     if (unit_id == -1) {
882        unit_id = 0;
883        while (drive_get(type, bus_id, unit_id) != NULL) {
884            unit_id++;
885            if (max_devs && unit_id >= max_devs) {
886                unit_id -= max_devs;
887                bus_id++;
888            }
889        }
890     }
891 
892     if (max_devs && unit_id >= max_devs) {
893         error_report("unit %d too big (max is %d)", unit_id, max_devs - 1);
894         goto fail;
895     }
896 
897     if (drive_get(type, bus_id, unit_id) != NULL) {
898         error_report("drive with bus=%d, unit=%d (index=%d) exists",
899                      bus_id, unit_id, index);
900         goto fail;
901     }
902 
903     /* Serial number */
904     serial = qemu_opt_get(legacy_opts, "serial");
905 
906     /* no id supplied -> create one */
907     if (qemu_opts_id(all_opts) == NULL) {
908         char *new_id;
909         const char *mediastr = "";
910         if (type == IF_IDE || type == IF_SCSI) {
911             mediastr = (media == MEDIA_CDROM) ? "-cd" : "-hd";
912         }
913         if (max_devs) {
914             new_id = g_strdup_printf("%s%i%s%i", if_name[type], bus_id,
915                                      mediastr, unit_id);
916         } else {
917             new_id = g_strdup_printf("%s%s%i", if_name[type],
918                                      mediastr, unit_id);
919         }
920         qdict_put(bs_opts, "id", qstring_from_str(new_id));
921         g_free(new_id);
922     }
923 
924     /* Add virtio block device */
925     devaddr = qemu_opt_get(legacy_opts, "addr");
926     if (devaddr && type != IF_VIRTIO) {
927         error_report("addr is not supported by this bus type");
928         goto fail;
929     }
930 
931     if (type == IF_VIRTIO) {
932         QemuOpts *devopts;
933         devopts = qemu_opts_create(qemu_find_opts("device"), NULL, 0,
934                                    &error_abort);
935         if (arch_type == QEMU_ARCH_S390X) {
936             qemu_opt_set(devopts, "driver", "virtio-blk-s390", &error_abort);
937         } else {
938             qemu_opt_set(devopts, "driver", "virtio-blk-pci", &error_abort);
939         }
940         qemu_opt_set(devopts, "drive", qdict_get_str(bs_opts, "id"),
941                      &error_abort);
942         if (devaddr) {
943             qemu_opt_set(devopts, "addr", devaddr, &error_abort);
944         }
945     }
946 
947     filename = qemu_opt_get(legacy_opts, "file");
948 
949     /* Check werror/rerror compatibility with if=... */
950     werror = qemu_opt_get(legacy_opts, "werror");
951     if (werror != NULL) {
952         if (type != IF_IDE && type != IF_SCSI && type != IF_VIRTIO &&
953             type != IF_NONE) {
954             error_report("werror is not supported by this bus type");
955             goto fail;
956         }
957         qdict_put(bs_opts, "werror", qstring_from_str(werror));
958     }
959 
960     rerror = qemu_opt_get(legacy_opts, "rerror");
961     if (rerror != NULL) {
962         if (type != IF_IDE && type != IF_VIRTIO && type != IF_SCSI &&
963             type != IF_NONE) {
964             error_report("rerror is not supported by this bus type");
965             goto fail;
966         }
967         qdict_put(bs_opts, "rerror", qstring_from_str(rerror));
968     }
969 
970     /* Actual block device init: Functionality shared with blockdev-add */
971     blk = blockdev_init(filename, bs_opts, &local_err);
972     bs_opts = NULL;
973     if (!blk) {
974         if (local_err) {
975             error_report_err(local_err);
976         }
977         goto fail;
978     } else {
979         assert(!local_err);
980     }
981 
982     /* Create legacy DriveInfo */
983     dinfo = g_malloc0(sizeof(*dinfo));
984     dinfo->opts = all_opts;
985 
986     dinfo->cyls = cyls;
987     dinfo->heads = heads;
988     dinfo->secs = secs;
989     dinfo->trans = translation;
990 
991     dinfo->type = type;
992     dinfo->bus = bus_id;
993     dinfo->unit = unit_id;
994     dinfo->devaddr = devaddr;
995     dinfo->serial = g_strdup(serial);
996 
997     blk_set_legacy_dinfo(blk, dinfo);
998 
999     switch(type) {
1000     case IF_IDE:
1001     case IF_SCSI:
1002     case IF_XEN:
1003     case IF_NONE:
1004         dinfo->media_cd = media == MEDIA_CDROM;
1005         break;
1006     default:
1007         break;
1008     }
1009 
1010 fail:
1011     qemu_opts_del(legacy_opts);
1012     QDECREF(bs_opts);
1013     return dinfo;
1014 }
1015 
1016 void hmp_commit(Monitor *mon, const QDict *qdict)
1017 {
1018     const char *device = qdict_get_str(qdict, "device");
1019     BlockBackend *blk;
1020     int ret;
1021 
1022     if (!strcmp(device, "all")) {
1023         ret = bdrv_commit_all();
1024     } else {
1025         blk = blk_by_name(device);
1026         if (!blk) {
1027             monitor_printf(mon, "Device '%s' not found\n", device);
1028             return;
1029         }
1030         ret = bdrv_commit(blk_bs(blk));
1031     }
1032     if (ret < 0) {
1033         monitor_printf(mon, "'commit' error for '%s': %s\n", device,
1034                        strerror(-ret));
1035     }
1036 }
1037 
1038 static void blockdev_do_action(int kind, void *data, Error **errp)
1039 {
1040     TransactionAction action;
1041     TransactionActionList list;
1042 
1043     action.kind = kind;
1044     action.data = data;
1045     list.value = &action;
1046     list.next = NULL;
1047     qmp_transaction(&list, errp);
1048 }
1049 
1050 void qmp_blockdev_snapshot_sync(bool has_device, const char *device,
1051                                 bool has_node_name, const char *node_name,
1052                                 const char *snapshot_file,
1053                                 bool has_snapshot_node_name,
1054                                 const char *snapshot_node_name,
1055                                 bool has_format, const char *format,
1056                                 bool has_mode, NewImageMode mode, Error **errp)
1057 {
1058     BlockdevSnapshot snapshot = {
1059         .has_device = has_device,
1060         .device = (char *) device,
1061         .has_node_name = has_node_name,
1062         .node_name = (char *) node_name,
1063         .snapshot_file = (char *) snapshot_file,
1064         .has_snapshot_node_name = has_snapshot_node_name,
1065         .snapshot_node_name = (char *) snapshot_node_name,
1066         .has_format = has_format,
1067         .format = (char *) format,
1068         .has_mode = has_mode,
1069         .mode = mode,
1070     };
1071     blockdev_do_action(TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC,
1072                        &snapshot, errp);
1073 }
1074 
1075 void qmp_blockdev_snapshot_internal_sync(const char *device,
1076                                          const char *name,
1077                                          Error **errp)
1078 {
1079     BlockdevSnapshotInternal snapshot = {
1080         .device = (char *) device,
1081         .name = (char *) name
1082     };
1083 
1084     blockdev_do_action(TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC,
1085                        &snapshot, errp);
1086 }
1087 
1088 SnapshotInfo *qmp_blockdev_snapshot_delete_internal_sync(const char *device,
1089                                                          bool has_id,
1090                                                          const char *id,
1091                                                          bool has_name,
1092                                                          const char *name,
1093                                                          Error **errp)
1094 {
1095     BlockDriverState *bs;
1096     BlockBackend *blk;
1097     AioContext *aio_context;
1098     QEMUSnapshotInfo sn;
1099     Error *local_err = NULL;
1100     SnapshotInfo *info = NULL;
1101     int ret;
1102 
1103     blk = blk_by_name(device);
1104     if (!blk) {
1105         error_set(errp, QERR_DEVICE_NOT_FOUND, device);
1106         return NULL;
1107     }
1108     bs = blk_bs(blk);
1109 
1110     if (!has_id) {
1111         id = NULL;
1112     }
1113 
1114     if (!has_name) {
1115         name = NULL;
1116     }
1117 
1118     if (!id && !name) {
1119         error_setg(errp, "Name or id must be provided");
1120         return NULL;
1121     }
1122 
1123     aio_context = bdrv_get_aio_context(bs);
1124     aio_context_acquire(aio_context);
1125 
1126     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_INTERNAL_SNAPSHOT_DELETE, errp)) {
1127         goto out_aio_context;
1128     }
1129 
1130     ret = bdrv_snapshot_find_by_id_and_name(bs, id, name, &sn, &local_err);
1131     if (local_err) {
1132         error_propagate(errp, local_err);
1133         goto out_aio_context;
1134     }
1135     if (!ret) {
1136         error_setg(errp,
1137                    "Snapshot with id '%s' and name '%s' does not exist on "
1138                    "device '%s'",
1139                    STR_OR_NULL(id), STR_OR_NULL(name), device);
1140         goto out_aio_context;
1141     }
1142 
1143     bdrv_snapshot_delete(bs, id, name, &local_err);
1144     if (local_err) {
1145         error_propagate(errp, local_err);
1146         goto out_aio_context;
1147     }
1148 
1149     aio_context_release(aio_context);
1150 
1151     info = g_new0(SnapshotInfo, 1);
1152     info->id = g_strdup(sn.id_str);
1153     info->name = g_strdup(sn.name);
1154     info->date_nsec = sn.date_nsec;
1155     info->date_sec = sn.date_sec;
1156     info->vm_state_size = sn.vm_state_size;
1157     info->vm_clock_nsec = sn.vm_clock_nsec % 1000000000;
1158     info->vm_clock_sec = sn.vm_clock_nsec / 1000000000;
1159 
1160     return info;
1161 
1162 out_aio_context:
1163     aio_context_release(aio_context);
1164     return NULL;
1165 }
1166 
1167 /**
1168  * block_dirty_bitmap_lookup:
1169  * Return a dirty bitmap (if present), after validating
1170  * the node reference and bitmap names.
1171  *
1172  * @node: The name of the BDS node to search for bitmaps
1173  * @name: The name of the bitmap to search for
1174  * @pbs: Output pointer for BDS lookup, if desired. Can be NULL.
1175  * @paio: Output pointer for aio_context acquisition, if desired. Can be NULL.
1176  * @errp: Output pointer for error information. Can be NULL.
1177  *
1178  * @return: A bitmap object on success, or NULL on failure.
1179  */
1180 static BdrvDirtyBitmap *block_dirty_bitmap_lookup(const char *node,
1181                                                   const char *name,
1182                                                   BlockDriverState **pbs,
1183                                                   AioContext **paio,
1184                                                   Error **errp)
1185 {
1186     BlockDriverState *bs;
1187     BdrvDirtyBitmap *bitmap;
1188     AioContext *aio_context;
1189 
1190     if (!node) {
1191         error_setg(errp, "Node cannot be NULL");
1192         return NULL;
1193     }
1194     if (!name) {
1195         error_setg(errp, "Bitmap name cannot be NULL");
1196         return NULL;
1197     }
1198     bs = bdrv_lookup_bs(node, node, NULL);
1199     if (!bs) {
1200         error_setg(errp, "Node '%s' not found", node);
1201         return NULL;
1202     }
1203 
1204     aio_context = bdrv_get_aio_context(bs);
1205     aio_context_acquire(aio_context);
1206 
1207     bitmap = bdrv_find_dirty_bitmap(bs, name);
1208     if (!bitmap) {
1209         error_setg(errp, "Dirty bitmap '%s' not found", name);
1210         goto fail;
1211     }
1212 
1213     if (pbs) {
1214         *pbs = bs;
1215     }
1216     if (paio) {
1217         *paio = aio_context;
1218     } else {
1219         aio_context_release(aio_context);
1220     }
1221 
1222     return bitmap;
1223 
1224  fail:
1225     aio_context_release(aio_context);
1226     return NULL;
1227 }
1228 
1229 /* New and old BlockDriverState structs for atomic group operations */
1230 
1231 typedef struct BlkTransactionState BlkTransactionState;
1232 
1233 /* Only prepare() may fail. In a single transaction, only one of commit() or
1234    abort() will be called, clean() will always be called if it present. */
1235 typedef struct BdrvActionOps {
1236     /* Size of state struct, in bytes. */
1237     size_t instance_size;
1238     /* Prepare the work, must NOT be NULL. */
1239     void (*prepare)(BlkTransactionState *common, Error **errp);
1240     /* Commit the changes, can be NULL. */
1241     void (*commit)(BlkTransactionState *common);
1242     /* Abort the changes on fail, can be NULL. */
1243     void (*abort)(BlkTransactionState *common);
1244     /* Clean up resource in the end, can be NULL. */
1245     void (*clean)(BlkTransactionState *common);
1246 } BdrvActionOps;
1247 
1248 /*
1249  * This structure must be arranged as first member in child type, assuming
1250  * that compiler will also arrange it to the same address with parent instance.
1251  * Later it will be used in free().
1252  */
1253 struct BlkTransactionState {
1254     TransactionAction *action;
1255     const BdrvActionOps *ops;
1256     QSIMPLEQ_ENTRY(BlkTransactionState) entry;
1257 };
1258 
1259 /* internal snapshot private data */
1260 typedef struct InternalSnapshotState {
1261     BlkTransactionState common;
1262     BlockDriverState *bs;
1263     AioContext *aio_context;
1264     QEMUSnapshotInfo sn;
1265 } InternalSnapshotState;
1266 
1267 static void internal_snapshot_prepare(BlkTransactionState *common,
1268                                       Error **errp)
1269 {
1270     Error *local_err = NULL;
1271     const char *device;
1272     const char *name;
1273     BlockBackend *blk;
1274     BlockDriverState *bs;
1275     QEMUSnapshotInfo old_sn, *sn;
1276     bool ret;
1277     qemu_timeval tv;
1278     BlockdevSnapshotInternal *internal;
1279     InternalSnapshotState *state;
1280     int ret1;
1281 
1282     g_assert(common->action->kind ==
1283              TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC);
1284     internal = common->action->blockdev_snapshot_internal_sync;
1285     state = DO_UPCAST(InternalSnapshotState, common, common);
1286 
1287     /* 1. parse input */
1288     device = internal->device;
1289     name = internal->name;
1290 
1291     /* 2. check for validation */
1292     blk = blk_by_name(device);
1293     if (!blk) {
1294         error_set(errp, QERR_DEVICE_NOT_FOUND, device);
1295         return;
1296     }
1297     bs = blk_bs(blk);
1298 
1299     /* AioContext is released in .clean() */
1300     state->aio_context = bdrv_get_aio_context(bs);
1301     aio_context_acquire(state->aio_context);
1302 
1303     if (!bdrv_is_inserted(bs)) {
1304         error_set(errp, QERR_DEVICE_HAS_NO_MEDIUM, device);
1305         return;
1306     }
1307 
1308     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_INTERNAL_SNAPSHOT, errp)) {
1309         return;
1310     }
1311 
1312     if (bdrv_is_read_only(bs)) {
1313         error_setg(errp, "Device '%s' is read only", device);
1314         return;
1315     }
1316 
1317     if (!bdrv_can_snapshot(bs)) {
1318         error_setg(errp, "Block format '%s' used by device '%s' "
1319                    "does not support internal snapshots",
1320                    bs->drv->format_name, device);
1321         return;
1322     }
1323 
1324     if (!strlen(name)) {
1325         error_setg(errp, "Name is empty");
1326         return;
1327     }
1328 
1329     /* check whether a snapshot with name exist */
1330     ret = bdrv_snapshot_find_by_id_and_name(bs, NULL, name, &old_sn,
1331                                             &local_err);
1332     if (local_err) {
1333         error_propagate(errp, local_err);
1334         return;
1335     } else if (ret) {
1336         error_setg(errp,
1337                    "Snapshot with name '%s' already exists on device '%s'",
1338                    name, device);
1339         return;
1340     }
1341 
1342     /* 3. take the snapshot */
1343     sn = &state->sn;
1344     pstrcpy(sn->name, sizeof(sn->name), name);
1345     qemu_gettimeofday(&tv);
1346     sn->date_sec = tv.tv_sec;
1347     sn->date_nsec = tv.tv_usec * 1000;
1348     sn->vm_clock_nsec = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL);
1349 
1350     ret1 = bdrv_snapshot_create(bs, sn);
1351     if (ret1 < 0) {
1352         error_setg_errno(errp, -ret1,
1353                          "Failed to create snapshot '%s' on device '%s'",
1354                          name, device);
1355         return;
1356     }
1357 
1358     /* 4. succeed, mark a snapshot is created */
1359     state->bs = bs;
1360 }
1361 
1362 static void internal_snapshot_abort(BlkTransactionState *common)
1363 {
1364     InternalSnapshotState *state =
1365                              DO_UPCAST(InternalSnapshotState, common, common);
1366     BlockDriverState *bs = state->bs;
1367     QEMUSnapshotInfo *sn = &state->sn;
1368     Error *local_error = NULL;
1369 
1370     if (!bs) {
1371         return;
1372     }
1373 
1374     if (bdrv_snapshot_delete(bs, sn->id_str, sn->name, &local_error) < 0) {
1375         error_report("Failed to delete snapshot with id '%s' and name '%s' on "
1376                      "device '%s' in abort: %s",
1377                      sn->id_str,
1378                      sn->name,
1379                      bdrv_get_device_name(bs),
1380                      error_get_pretty(local_error));
1381         error_free(local_error);
1382     }
1383 }
1384 
1385 static void internal_snapshot_clean(BlkTransactionState *common)
1386 {
1387     InternalSnapshotState *state = DO_UPCAST(InternalSnapshotState,
1388                                              common, common);
1389 
1390     if (state->aio_context) {
1391         aio_context_release(state->aio_context);
1392     }
1393 }
1394 
1395 /* external snapshot private data */
1396 typedef struct ExternalSnapshotState {
1397     BlkTransactionState common;
1398     BlockDriverState *old_bs;
1399     BlockDriverState *new_bs;
1400     AioContext *aio_context;
1401 } ExternalSnapshotState;
1402 
1403 static void external_snapshot_prepare(BlkTransactionState *common,
1404                                       Error **errp)
1405 {
1406     BlockDriver *drv;
1407     int flags, ret;
1408     QDict *options = NULL;
1409     Error *local_err = NULL;
1410     bool has_device = false;
1411     const char *device;
1412     bool has_node_name = false;
1413     const char *node_name;
1414     bool has_snapshot_node_name = false;
1415     const char *snapshot_node_name;
1416     const char *new_image_file;
1417     const char *format = "qcow2";
1418     enum NewImageMode mode = NEW_IMAGE_MODE_ABSOLUTE_PATHS;
1419     ExternalSnapshotState *state =
1420                              DO_UPCAST(ExternalSnapshotState, common, common);
1421     TransactionAction *action = common->action;
1422 
1423     /* get parameters */
1424     g_assert(action->kind == TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC);
1425 
1426     has_device = action->blockdev_snapshot_sync->has_device;
1427     device = action->blockdev_snapshot_sync->device;
1428     has_node_name = action->blockdev_snapshot_sync->has_node_name;
1429     node_name = action->blockdev_snapshot_sync->node_name;
1430     has_snapshot_node_name =
1431         action->blockdev_snapshot_sync->has_snapshot_node_name;
1432     snapshot_node_name = action->blockdev_snapshot_sync->snapshot_node_name;
1433 
1434     new_image_file = action->blockdev_snapshot_sync->snapshot_file;
1435     if (action->blockdev_snapshot_sync->has_format) {
1436         format = action->blockdev_snapshot_sync->format;
1437     }
1438     if (action->blockdev_snapshot_sync->has_mode) {
1439         mode = action->blockdev_snapshot_sync->mode;
1440     }
1441 
1442     /* start processing */
1443     drv = bdrv_find_format(format);
1444     if (!drv) {
1445         error_set(errp, QERR_INVALID_BLOCK_FORMAT, format);
1446         return;
1447     }
1448 
1449     state->old_bs = bdrv_lookup_bs(has_device ? device : NULL,
1450                                    has_node_name ? node_name : NULL,
1451                                    &local_err);
1452     if (local_err) {
1453         error_propagate(errp, local_err);
1454         return;
1455     }
1456 
1457     if (has_node_name && !has_snapshot_node_name) {
1458         error_setg(errp, "New snapshot node name missing");
1459         return;
1460     }
1461 
1462     if (has_snapshot_node_name && bdrv_find_node(snapshot_node_name)) {
1463         error_setg(errp, "New snapshot node name already existing");
1464         return;
1465     }
1466 
1467     /* Acquire AioContext now so any threads operating on old_bs stop */
1468     state->aio_context = bdrv_get_aio_context(state->old_bs);
1469     aio_context_acquire(state->aio_context);
1470 
1471     if (!bdrv_is_inserted(state->old_bs)) {
1472         error_set(errp, QERR_DEVICE_HAS_NO_MEDIUM, device);
1473         return;
1474     }
1475 
1476     if (bdrv_op_is_blocked(state->old_bs,
1477                            BLOCK_OP_TYPE_EXTERNAL_SNAPSHOT, errp)) {
1478         return;
1479     }
1480 
1481     if (!bdrv_is_read_only(state->old_bs)) {
1482         if (bdrv_flush(state->old_bs)) {
1483             error_set(errp, QERR_IO_ERROR);
1484             return;
1485         }
1486     }
1487 
1488     if (!bdrv_is_first_non_filter(state->old_bs)) {
1489         error_set(errp, QERR_FEATURE_DISABLED, "snapshot");
1490         return;
1491     }
1492 
1493     flags = state->old_bs->open_flags;
1494 
1495     /* create new image w/backing file */
1496     if (mode != NEW_IMAGE_MODE_EXISTING) {
1497         bdrv_img_create(new_image_file, format,
1498                         state->old_bs->filename,
1499                         state->old_bs->drv->format_name,
1500                         NULL, -1, flags, &local_err, false);
1501         if (local_err) {
1502             error_propagate(errp, local_err);
1503             return;
1504         }
1505     }
1506 
1507     if (has_snapshot_node_name) {
1508         options = qdict_new();
1509         qdict_put(options, "node-name",
1510                   qstring_from_str(snapshot_node_name));
1511     }
1512 
1513     /* TODO Inherit bs->options or only take explicit options with an
1514      * extended QMP command? */
1515     assert(state->new_bs == NULL);
1516     ret = bdrv_open(&state->new_bs, new_image_file, NULL, options,
1517                     flags | BDRV_O_NO_BACKING, drv, &local_err);
1518     /* We will manually add the backing_hd field to the bs later */
1519     if (ret != 0) {
1520         error_propagate(errp, local_err);
1521     }
1522 }
1523 
1524 static void external_snapshot_commit(BlkTransactionState *common)
1525 {
1526     ExternalSnapshotState *state =
1527                              DO_UPCAST(ExternalSnapshotState, common, common);
1528 
1529     bdrv_set_aio_context(state->new_bs, state->aio_context);
1530 
1531     /* This removes our old bs and adds the new bs */
1532     bdrv_append(state->new_bs, state->old_bs);
1533     /* We don't need (or want) to use the transactional
1534      * bdrv_reopen_multiple() across all the entries at once, because we
1535      * don't want to abort all of them if one of them fails the reopen */
1536     bdrv_reopen(state->new_bs, state->new_bs->open_flags & ~BDRV_O_RDWR,
1537                 NULL);
1538 
1539     aio_context_release(state->aio_context);
1540 }
1541 
1542 static void external_snapshot_abort(BlkTransactionState *common)
1543 {
1544     ExternalSnapshotState *state =
1545                              DO_UPCAST(ExternalSnapshotState, common, common);
1546     if (state->new_bs) {
1547         bdrv_unref(state->new_bs);
1548     }
1549     if (state->aio_context) {
1550         aio_context_release(state->aio_context);
1551     }
1552 }
1553 
1554 typedef struct DriveBackupState {
1555     BlkTransactionState common;
1556     BlockDriverState *bs;
1557     AioContext *aio_context;
1558     BlockJob *job;
1559 } DriveBackupState;
1560 
1561 static void drive_backup_prepare(BlkTransactionState *common, Error **errp)
1562 {
1563     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1564     BlockDriverState *bs;
1565     BlockBackend *blk;
1566     DriveBackup *backup;
1567     Error *local_err = NULL;
1568 
1569     assert(common->action->kind == TRANSACTION_ACTION_KIND_DRIVE_BACKUP);
1570     backup = common->action->drive_backup;
1571 
1572     blk = blk_by_name(backup->device);
1573     if (!blk) {
1574         error_set(errp, QERR_DEVICE_NOT_FOUND, backup->device);
1575         return;
1576     }
1577     bs = blk_bs(blk);
1578 
1579     /* AioContext is released in .clean() */
1580     state->aio_context = bdrv_get_aio_context(bs);
1581     aio_context_acquire(state->aio_context);
1582 
1583     qmp_drive_backup(backup->device, backup->target,
1584                      backup->has_format, backup->format,
1585                      backup->sync,
1586                      backup->has_mode, backup->mode,
1587                      backup->has_speed, backup->speed,
1588                      backup->has_bitmap, backup->bitmap,
1589                      backup->has_on_source_error, backup->on_source_error,
1590                      backup->has_on_target_error, backup->on_target_error,
1591                      &local_err);
1592     if (local_err) {
1593         error_propagate(errp, local_err);
1594         return;
1595     }
1596 
1597     state->bs = bs;
1598     state->job = state->bs->job;
1599 }
1600 
1601 static void drive_backup_abort(BlkTransactionState *common)
1602 {
1603     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1604     BlockDriverState *bs = state->bs;
1605 
1606     /* Only cancel if it's the job we started */
1607     if (bs && bs->job && bs->job == state->job) {
1608         block_job_cancel_sync(bs->job);
1609     }
1610 }
1611 
1612 static void drive_backup_clean(BlkTransactionState *common)
1613 {
1614     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1615 
1616     if (state->aio_context) {
1617         aio_context_release(state->aio_context);
1618     }
1619 }
1620 
1621 typedef struct BlockdevBackupState {
1622     BlkTransactionState common;
1623     BlockDriverState *bs;
1624     BlockJob *job;
1625     AioContext *aio_context;
1626 } BlockdevBackupState;
1627 
1628 static void blockdev_backup_prepare(BlkTransactionState *common, Error **errp)
1629 {
1630     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1631     BlockdevBackup *backup;
1632     BlockDriverState *bs, *target;
1633     BlockBackend *blk;
1634     Error *local_err = NULL;
1635 
1636     assert(common->action->kind == TRANSACTION_ACTION_KIND_BLOCKDEV_BACKUP);
1637     backup = common->action->blockdev_backup;
1638 
1639     blk = blk_by_name(backup->device);
1640     if (!blk) {
1641         error_setg(errp, "Device '%s' not found", backup->device);
1642         return;
1643     }
1644     bs = blk_bs(blk);
1645 
1646     blk = blk_by_name(backup->target);
1647     if (!blk) {
1648         error_setg(errp, "Device '%s' not found", backup->target);
1649         return;
1650     }
1651     target = blk_bs(blk);
1652 
1653     /* AioContext is released in .clean() */
1654     state->aio_context = bdrv_get_aio_context(bs);
1655     if (state->aio_context != bdrv_get_aio_context(target)) {
1656         state->aio_context = NULL;
1657         error_setg(errp, "Backup between two IO threads is not implemented");
1658         return;
1659     }
1660     aio_context_acquire(state->aio_context);
1661 
1662     qmp_blockdev_backup(backup->device, backup->target,
1663                         backup->sync,
1664                         backup->has_speed, backup->speed,
1665                         backup->has_on_source_error, backup->on_source_error,
1666                         backup->has_on_target_error, backup->on_target_error,
1667                         &local_err);
1668     if (local_err) {
1669         error_propagate(errp, local_err);
1670         return;
1671     }
1672 
1673     state->bs = bs;
1674     state->job = state->bs->job;
1675 }
1676 
1677 static void blockdev_backup_abort(BlkTransactionState *common)
1678 {
1679     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1680     BlockDriverState *bs = state->bs;
1681 
1682     /* Only cancel if it's the job we started */
1683     if (bs && bs->job && bs->job == state->job) {
1684         block_job_cancel_sync(bs->job);
1685     }
1686 }
1687 
1688 static void blockdev_backup_clean(BlkTransactionState *common)
1689 {
1690     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1691 
1692     if (state->aio_context) {
1693         aio_context_release(state->aio_context);
1694     }
1695 }
1696 
1697 static void abort_prepare(BlkTransactionState *common, Error **errp)
1698 {
1699     error_setg(errp, "Transaction aborted using Abort action");
1700 }
1701 
1702 static void abort_commit(BlkTransactionState *common)
1703 {
1704     g_assert_not_reached(); /* this action never succeeds */
1705 }
1706 
1707 static const BdrvActionOps actions[] = {
1708     [TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC] = {
1709         .instance_size = sizeof(ExternalSnapshotState),
1710         .prepare  = external_snapshot_prepare,
1711         .commit   = external_snapshot_commit,
1712         .abort = external_snapshot_abort,
1713     },
1714     [TRANSACTION_ACTION_KIND_DRIVE_BACKUP] = {
1715         .instance_size = sizeof(DriveBackupState),
1716         .prepare = drive_backup_prepare,
1717         .abort = drive_backup_abort,
1718         .clean = drive_backup_clean,
1719     },
1720     [TRANSACTION_ACTION_KIND_BLOCKDEV_BACKUP] = {
1721         .instance_size = sizeof(BlockdevBackupState),
1722         .prepare = blockdev_backup_prepare,
1723         .abort = blockdev_backup_abort,
1724         .clean = blockdev_backup_clean,
1725     },
1726     [TRANSACTION_ACTION_KIND_ABORT] = {
1727         .instance_size = sizeof(BlkTransactionState),
1728         .prepare = abort_prepare,
1729         .commit = abort_commit,
1730     },
1731     [TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC] = {
1732         .instance_size = sizeof(InternalSnapshotState),
1733         .prepare  = internal_snapshot_prepare,
1734         .abort = internal_snapshot_abort,
1735         .clean = internal_snapshot_clean,
1736     },
1737 };
1738 
1739 /*
1740  * 'Atomic' group operations.  The operations are performed as a set, and if
1741  * any fail then we roll back all operations in the group.
1742  */
1743 void qmp_transaction(TransactionActionList *dev_list, Error **errp)
1744 {
1745     TransactionActionList *dev_entry = dev_list;
1746     BlkTransactionState *state, *next;
1747     Error *local_err = NULL;
1748 
1749     QSIMPLEQ_HEAD(snap_bdrv_states, BlkTransactionState) snap_bdrv_states;
1750     QSIMPLEQ_INIT(&snap_bdrv_states);
1751 
1752     /* drain all i/o before any operations */
1753     bdrv_drain_all();
1754 
1755     /* We don't do anything in this loop that commits us to the operations */
1756     while (NULL != dev_entry) {
1757         TransactionAction *dev_info = NULL;
1758         const BdrvActionOps *ops;
1759 
1760         dev_info = dev_entry->value;
1761         dev_entry = dev_entry->next;
1762 
1763         assert(dev_info->kind < ARRAY_SIZE(actions));
1764 
1765         ops = &actions[dev_info->kind];
1766         assert(ops->instance_size > 0);
1767 
1768         state = g_malloc0(ops->instance_size);
1769         state->ops = ops;
1770         state->action = dev_info;
1771         QSIMPLEQ_INSERT_TAIL(&snap_bdrv_states, state, entry);
1772 
1773         state->ops->prepare(state, &local_err);
1774         if (local_err) {
1775             error_propagate(errp, local_err);
1776             goto delete_and_fail;
1777         }
1778     }
1779 
1780     QSIMPLEQ_FOREACH(state, &snap_bdrv_states, entry) {
1781         if (state->ops->commit) {
1782             state->ops->commit(state);
1783         }
1784     }
1785 
1786     /* success */
1787     goto exit;
1788 
1789 delete_and_fail:
1790     /* failure, and it is all-or-none; roll back all operations */
1791     QSIMPLEQ_FOREACH(state, &snap_bdrv_states, entry) {
1792         if (state->ops->abort) {
1793             state->ops->abort(state);
1794         }
1795     }
1796 exit:
1797     QSIMPLEQ_FOREACH_SAFE(state, &snap_bdrv_states, entry, next) {
1798         if (state->ops->clean) {
1799             state->ops->clean(state);
1800         }
1801         g_free(state);
1802     }
1803 }
1804 
1805 
1806 static void eject_device(BlockBackend *blk, int force, Error **errp)
1807 {
1808     BlockDriverState *bs = blk_bs(blk);
1809     AioContext *aio_context;
1810 
1811     aio_context = bdrv_get_aio_context(bs);
1812     aio_context_acquire(aio_context);
1813 
1814     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_EJECT, errp)) {
1815         goto out;
1816     }
1817     if (!blk_dev_has_removable_media(blk)) {
1818         error_setg(errp, "Device '%s' is not removable",
1819                    bdrv_get_device_name(bs));
1820         goto out;
1821     }
1822 
1823     if (blk_dev_is_medium_locked(blk) && !blk_dev_is_tray_open(blk)) {
1824         blk_dev_eject_request(blk, force);
1825         if (!force) {
1826             error_setg(errp, "Device '%s' is locked",
1827                        bdrv_get_device_name(bs));
1828             goto out;
1829         }
1830     }
1831 
1832     bdrv_close(bs);
1833 
1834 out:
1835     aio_context_release(aio_context);
1836 }
1837 
1838 void qmp_eject(const char *device, bool has_force, bool force, Error **errp)
1839 {
1840     BlockBackend *blk;
1841 
1842     blk = blk_by_name(device);
1843     if (!blk) {
1844         error_set(errp, QERR_DEVICE_NOT_FOUND, device);
1845         return;
1846     }
1847 
1848     eject_device(blk, force, errp);
1849 }
1850 
1851 void qmp_block_passwd(bool has_device, const char *device,
1852                       bool has_node_name, const char *node_name,
1853                       const char *password, Error **errp)
1854 {
1855     Error *local_err = NULL;
1856     BlockDriverState *bs;
1857     AioContext *aio_context;
1858 
1859     bs = bdrv_lookup_bs(has_device ? device : NULL,
1860                         has_node_name ? node_name : NULL,
1861                         &local_err);
1862     if (local_err) {
1863         error_propagate(errp, local_err);
1864         return;
1865     }
1866 
1867     aio_context = bdrv_get_aio_context(bs);
1868     aio_context_acquire(aio_context);
1869 
1870     bdrv_add_key(bs, password, errp);
1871 
1872     aio_context_release(aio_context);
1873 }
1874 
1875 /* Assumes AioContext is held */
1876 static void qmp_bdrv_open_encrypted(BlockDriverState *bs, const char *filename,
1877                                     int bdrv_flags, BlockDriver *drv,
1878                                     const char *password, Error **errp)
1879 {
1880     Error *local_err = NULL;
1881     int ret;
1882 
1883     ret = bdrv_open(&bs, filename, NULL, NULL, bdrv_flags, drv, &local_err);
1884     if (ret < 0) {
1885         error_propagate(errp, local_err);
1886         return;
1887     }
1888 
1889     bdrv_add_key(bs, password, errp);
1890 }
1891 
1892 void qmp_change_blockdev(const char *device, const char *filename,
1893                          const char *format, Error **errp)
1894 {
1895     BlockBackend *blk;
1896     BlockDriverState *bs;
1897     AioContext *aio_context;
1898     BlockDriver *drv = NULL;
1899     int bdrv_flags;
1900     Error *err = NULL;
1901 
1902     blk = blk_by_name(device);
1903     if (!blk) {
1904         error_set(errp, QERR_DEVICE_NOT_FOUND, device);
1905         return;
1906     }
1907     bs = blk_bs(blk);
1908 
1909     aio_context = bdrv_get_aio_context(bs);
1910     aio_context_acquire(aio_context);
1911 
1912     if (format) {
1913         drv = bdrv_find_whitelisted_format(format, bs->read_only);
1914         if (!drv) {
1915             error_set(errp, QERR_INVALID_BLOCK_FORMAT, format);
1916             goto out;
1917         }
1918     }
1919 
1920     eject_device(blk, 0, &err);
1921     if (err) {
1922         error_propagate(errp, err);
1923         goto out;
1924     }
1925 
1926     bdrv_flags = bdrv_is_read_only(bs) ? 0 : BDRV_O_RDWR;
1927     bdrv_flags |= bdrv_is_snapshot(bs) ? BDRV_O_SNAPSHOT : 0;
1928 
1929     qmp_bdrv_open_encrypted(bs, filename, bdrv_flags, drv, NULL, errp);
1930 
1931 out:
1932     aio_context_release(aio_context);
1933 }
1934 
1935 /* throttling disk I/O limits */
1936 void qmp_block_set_io_throttle(const char *device, int64_t bps, int64_t bps_rd,
1937                                int64_t bps_wr,
1938                                int64_t iops,
1939                                int64_t iops_rd,
1940                                int64_t iops_wr,
1941                                bool has_bps_max,
1942                                int64_t bps_max,
1943                                bool has_bps_rd_max,
1944                                int64_t bps_rd_max,
1945                                bool has_bps_wr_max,
1946                                int64_t bps_wr_max,
1947                                bool has_iops_max,
1948                                int64_t iops_max,
1949                                bool has_iops_rd_max,
1950                                int64_t iops_rd_max,
1951                                bool has_iops_wr_max,
1952                                int64_t iops_wr_max,
1953                                bool has_iops_size,
1954                                int64_t iops_size, Error **errp)
1955 {
1956     ThrottleConfig cfg;
1957     BlockDriverState *bs;
1958     BlockBackend *blk;
1959     AioContext *aio_context;
1960 
1961     blk = blk_by_name(device);
1962     if (!blk) {
1963         error_set(errp, QERR_DEVICE_NOT_FOUND, device);
1964         return;
1965     }
1966     bs = blk_bs(blk);
1967 
1968     memset(&cfg, 0, sizeof(cfg));
1969     cfg.buckets[THROTTLE_BPS_TOTAL].avg = bps;
1970     cfg.buckets[THROTTLE_BPS_READ].avg  = bps_rd;
1971     cfg.buckets[THROTTLE_BPS_WRITE].avg = bps_wr;
1972 
1973     cfg.buckets[THROTTLE_OPS_TOTAL].avg = iops;
1974     cfg.buckets[THROTTLE_OPS_READ].avg  = iops_rd;
1975     cfg.buckets[THROTTLE_OPS_WRITE].avg = iops_wr;
1976 
1977     if (has_bps_max) {
1978         cfg.buckets[THROTTLE_BPS_TOTAL].max = bps_max;
1979     }
1980     if (has_bps_rd_max) {
1981         cfg.buckets[THROTTLE_BPS_READ].max = bps_rd_max;
1982     }
1983     if (has_bps_wr_max) {
1984         cfg.buckets[THROTTLE_BPS_WRITE].max = bps_wr_max;
1985     }
1986     if (has_iops_max) {
1987         cfg.buckets[THROTTLE_OPS_TOTAL].max = iops_max;
1988     }
1989     if (has_iops_rd_max) {
1990         cfg.buckets[THROTTLE_OPS_READ].max = iops_rd_max;
1991     }
1992     if (has_iops_wr_max) {
1993         cfg.buckets[THROTTLE_OPS_WRITE].max = iops_wr_max;
1994     }
1995 
1996     if (has_iops_size) {
1997         cfg.op_size = iops_size;
1998     }
1999 
2000     if (!check_throttle_config(&cfg, errp)) {
2001         return;
2002     }
2003 
2004     aio_context = bdrv_get_aio_context(bs);
2005     aio_context_acquire(aio_context);
2006 
2007     if (!bs->io_limits_enabled && throttle_enabled(&cfg)) {
2008         bdrv_io_limits_enable(bs);
2009     } else if (bs->io_limits_enabled && !throttle_enabled(&cfg)) {
2010         bdrv_io_limits_disable(bs);
2011     }
2012 
2013     if (bs->io_limits_enabled) {
2014         bdrv_set_io_limits(bs, &cfg);
2015     }
2016 
2017     aio_context_release(aio_context);
2018 }
2019 
2020 void qmp_block_dirty_bitmap_add(const char *node, const char *name,
2021                                 bool has_granularity, uint32_t granularity,
2022                                 Error **errp)
2023 {
2024     AioContext *aio_context;
2025     BlockDriverState *bs;
2026 
2027     if (!name || name[0] == '\0') {
2028         error_setg(errp, "Bitmap name cannot be empty");
2029         return;
2030     }
2031 
2032     bs = bdrv_lookup_bs(node, node, errp);
2033     if (!bs) {
2034         return;
2035     }
2036 
2037     aio_context = bdrv_get_aio_context(bs);
2038     aio_context_acquire(aio_context);
2039 
2040     if (has_granularity) {
2041         if (granularity < 512 || !is_power_of_2(granularity)) {
2042             error_setg(errp, "Granularity must be power of 2 "
2043                              "and at least 512");
2044             goto out;
2045         }
2046     } else {
2047         /* Default to cluster size, if available: */
2048         granularity = bdrv_get_default_bitmap_granularity(bs);
2049     }
2050 
2051     bdrv_create_dirty_bitmap(bs, granularity, name, errp);
2052 
2053  out:
2054     aio_context_release(aio_context);
2055 }
2056 
2057 void qmp_block_dirty_bitmap_remove(const char *node, const char *name,
2058                                    Error **errp)
2059 {
2060     AioContext *aio_context;
2061     BlockDriverState *bs;
2062     BdrvDirtyBitmap *bitmap;
2063 
2064     bitmap = block_dirty_bitmap_lookup(node, name, &bs, &aio_context, errp);
2065     if (!bitmap || !bs) {
2066         return;
2067     }
2068 
2069     if (bdrv_dirty_bitmap_frozen(bitmap)) {
2070         error_setg(errp,
2071                    "Bitmap '%s' is currently frozen and cannot be removed",
2072                    name);
2073         goto out;
2074     }
2075     bdrv_dirty_bitmap_make_anon(bitmap);
2076     bdrv_release_dirty_bitmap(bs, bitmap);
2077 
2078  out:
2079     aio_context_release(aio_context);
2080 }
2081 
2082 /**
2083  * Completely clear a bitmap, for the purposes of synchronizing a bitmap
2084  * immediately after a full backup operation.
2085  */
2086 void qmp_block_dirty_bitmap_clear(const char *node, const char *name,
2087                                   Error **errp)
2088 {
2089     AioContext *aio_context;
2090     BdrvDirtyBitmap *bitmap;
2091     BlockDriverState *bs;
2092 
2093     bitmap = block_dirty_bitmap_lookup(node, name, &bs, &aio_context, errp);
2094     if (!bitmap || !bs) {
2095         return;
2096     }
2097 
2098     if (bdrv_dirty_bitmap_frozen(bitmap)) {
2099         error_setg(errp,
2100                    "Bitmap '%s' is currently frozen and cannot be modified",
2101                    name);
2102         goto out;
2103     } else if (!bdrv_dirty_bitmap_enabled(bitmap)) {
2104         error_setg(errp,
2105                    "Bitmap '%s' is currently disabled and cannot be cleared",
2106                    name);
2107         goto out;
2108     }
2109 
2110     bdrv_clear_dirty_bitmap(bitmap);
2111 
2112  out:
2113     aio_context_release(aio_context);
2114 }
2115 
2116 void hmp_drive_del(Monitor *mon, const QDict *qdict)
2117 {
2118     const char *id = qdict_get_str(qdict, "id");
2119     BlockBackend *blk;
2120     BlockDriverState *bs;
2121     AioContext *aio_context;
2122     Error *local_err = NULL;
2123 
2124     blk = blk_by_name(id);
2125     if (!blk) {
2126         error_report("Device '%s' not found", id);
2127         return;
2128     }
2129     bs = blk_bs(blk);
2130 
2131     if (!blk_legacy_dinfo(blk)) {
2132         error_report("Deleting device added with blockdev-add"
2133                      " is not supported");
2134         return;
2135     }
2136 
2137     aio_context = bdrv_get_aio_context(bs);
2138     aio_context_acquire(aio_context);
2139 
2140     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_DRIVE_DEL, &local_err)) {
2141         error_report_err(local_err);
2142         aio_context_release(aio_context);
2143         return;
2144     }
2145 
2146     /* quiesce block driver; prevent further io */
2147     bdrv_drain_all();
2148     bdrv_flush(bs);
2149     bdrv_close(bs);
2150 
2151     /* if we have a device attached to this BlockDriverState
2152      * then we need to make the drive anonymous until the device
2153      * can be removed.  If this is a drive with no device backing
2154      * then we can just get rid of the block driver state right here.
2155      */
2156     if (blk_get_attached_dev(blk)) {
2157         blk_hide_on_behalf_of_hmp_drive_del(blk);
2158         /* Further I/O must not pause the guest */
2159         bdrv_set_on_error(bs, BLOCKDEV_ON_ERROR_REPORT,
2160                           BLOCKDEV_ON_ERROR_REPORT);
2161     } else {
2162         blk_unref(blk);
2163     }
2164 
2165     aio_context_release(aio_context);
2166 }
2167 
2168 void qmp_block_resize(bool has_device, const char *device,
2169                       bool has_node_name, const char *node_name,
2170                       int64_t size, Error **errp)
2171 {
2172     Error *local_err = NULL;
2173     BlockDriverState *bs;
2174     AioContext *aio_context;
2175     int ret;
2176 
2177     bs = bdrv_lookup_bs(has_device ? device : NULL,
2178                         has_node_name ? node_name : NULL,
2179                         &local_err);
2180     if (local_err) {
2181         error_propagate(errp, local_err);
2182         return;
2183     }
2184 
2185     aio_context = bdrv_get_aio_context(bs);
2186     aio_context_acquire(aio_context);
2187 
2188     if (!bdrv_is_first_non_filter(bs)) {
2189         error_set(errp, QERR_FEATURE_DISABLED, "resize");
2190         goto out;
2191     }
2192 
2193     if (size < 0) {
2194         error_set(errp, QERR_INVALID_PARAMETER_VALUE, "size", "a >0 size");
2195         goto out;
2196     }
2197 
2198     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_RESIZE, NULL)) {
2199         error_set(errp, QERR_DEVICE_IN_USE, device);
2200         goto out;
2201     }
2202 
2203     /* complete all in-flight operations before resizing the device */
2204     bdrv_drain_all();
2205 
2206     ret = bdrv_truncate(bs, size);
2207     switch (ret) {
2208     case 0:
2209         break;
2210     case -ENOMEDIUM:
2211         error_set(errp, QERR_DEVICE_HAS_NO_MEDIUM, device);
2212         break;
2213     case -ENOTSUP:
2214         error_set(errp, QERR_UNSUPPORTED);
2215         break;
2216     case -EACCES:
2217         error_setg(errp, "Device '%s' is read only", device);
2218         break;
2219     case -EBUSY:
2220         error_set(errp, QERR_DEVICE_IN_USE, device);
2221         break;
2222     default:
2223         error_setg_errno(errp, -ret, "Could not resize");
2224         break;
2225     }
2226 
2227 out:
2228     aio_context_release(aio_context);
2229 }
2230 
2231 static void block_job_cb(void *opaque, int ret)
2232 {
2233     /* Note that this function may be executed from another AioContext besides
2234      * the QEMU main loop.  If you need to access anything that assumes the
2235      * QEMU global mutex, use a BH or introduce a mutex.
2236      */
2237 
2238     BlockDriverState *bs = opaque;
2239     const char *msg = NULL;
2240 
2241     trace_block_job_cb(bs, bs->job, ret);
2242 
2243     assert(bs->job);
2244 
2245     if (ret < 0) {
2246         msg = strerror(-ret);
2247     }
2248 
2249     if (block_job_is_cancelled(bs->job)) {
2250         block_job_event_cancelled(bs->job);
2251     } else {
2252         block_job_event_completed(bs->job, msg);
2253     }
2254 
2255     bdrv_put_ref_bh_schedule(bs);
2256 }
2257 
2258 void qmp_block_stream(const char *device,
2259                       bool has_base, const char *base,
2260                       bool has_backing_file, const char *backing_file,
2261                       bool has_speed, int64_t speed,
2262                       bool has_on_error, BlockdevOnError on_error,
2263                       Error **errp)
2264 {
2265     BlockBackend *blk;
2266     BlockDriverState *bs;
2267     BlockDriverState *base_bs = NULL;
2268     AioContext *aio_context;
2269     Error *local_err = NULL;
2270     const char *base_name = NULL;
2271 
2272     if (!has_on_error) {
2273         on_error = BLOCKDEV_ON_ERROR_REPORT;
2274     }
2275 
2276     blk = blk_by_name(device);
2277     if (!blk) {
2278         error_set(errp, QERR_DEVICE_NOT_FOUND, device);
2279         return;
2280     }
2281     bs = blk_bs(blk);
2282 
2283     aio_context = bdrv_get_aio_context(bs);
2284     aio_context_acquire(aio_context);
2285 
2286     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_STREAM, errp)) {
2287         goto out;
2288     }
2289 
2290     if (has_base) {
2291         base_bs = bdrv_find_backing_image(bs, base);
2292         if (base_bs == NULL) {
2293             error_set(errp, QERR_BASE_NOT_FOUND, base);
2294             goto out;
2295         }
2296         assert(bdrv_get_aio_context(base_bs) == aio_context);
2297         base_name = base;
2298     }
2299 
2300     /* if we are streaming the entire chain, the result will have no backing
2301      * file, and specifying one is therefore an error */
2302     if (base_bs == NULL && has_backing_file) {
2303         error_setg(errp, "backing file specified, but streaming the "
2304                          "entire chain");
2305         goto out;
2306     }
2307 
2308     /* backing_file string overrides base bs filename */
2309     base_name = has_backing_file ? backing_file : base_name;
2310 
2311     stream_start(bs, base_bs, base_name, has_speed ? speed : 0,
2312                  on_error, block_job_cb, bs, &local_err);
2313     if (local_err) {
2314         error_propagate(errp, local_err);
2315         goto out;
2316     }
2317 
2318     trace_qmp_block_stream(bs, bs->job);
2319 
2320 out:
2321     aio_context_release(aio_context);
2322 }
2323 
2324 void qmp_block_commit(const char *device,
2325                       bool has_base, const char *base,
2326                       bool has_top, const char *top,
2327                       bool has_backing_file, const char *backing_file,
2328                       bool has_speed, int64_t speed,
2329                       Error **errp)
2330 {
2331     BlockBackend *blk;
2332     BlockDriverState *bs;
2333     BlockDriverState *base_bs, *top_bs;
2334     AioContext *aio_context;
2335     Error *local_err = NULL;
2336     /* This will be part of the QMP command, if/when the
2337      * BlockdevOnError change for blkmirror makes it in
2338      */
2339     BlockdevOnError on_error = BLOCKDEV_ON_ERROR_REPORT;
2340 
2341     if (!has_speed) {
2342         speed = 0;
2343     }
2344 
2345     /* Important Note:
2346      *  libvirt relies on the DeviceNotFound error class in order to probe for
2347      *  live commit feature versions; for this to work, we must make sure to
2348      *  perform the device lookup before any generic errors that may occur in a
2349      *  scenario in which all optional arguments are omitted. */
2350     blk = blk_by_name(device);
2351     if (!blk) {
2352         error_set(errp, QERR_DEVICE_NOT_FOUND, device);
2353         return;
2354     }
2355     bs = blk_bs(blk);
2356 
2357     aio_context = bdrv_get_aio_context(bs);
2358     aio_context_acquire(aio_context);
2359 
2360     /* drain all i/o before commits */
2361     bdrv_drain_all();
2362 
2363     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_COMMIT_SOURCE, errp)) {
2364         goto out;
2365     }
2366 
2367     /* default top_bs is the active layer */
2368     top_bs = bs;
2369 
2370     if (has_top && top) {
2371         if (strcmp(bs->filename, top) != 0) {
2372             top_bs = bdrv_find_backing_image(bs, top);
2373         }
2374     }
2375 
2376     if (top_bs == NULL) {
2377         error_setg(errp, "Top image file %s not found", top ? top : "NULL");
2378         goto out;
2379     }
2380 
2381     assert(bdrv_get_aio_context(top_bs) == aio_context);
2382 
2383     if (has_base && base) {
2384         base_bs = bdrv_find_backing_image(top_bs, base);
2385     } else {
2386         base_bs = bdrv_find_base(top_bs);
2387     }
2388 
2389     if (base_bs == NULL) {
2390         error_set(errp, QERR_BASE_NOT_FOUND, base ? base : "NULL");
2391         goto out;
2392     }
2393 
2394     assert(bdrv_get_aio_context(base_bs) == aio_context);
2395 
2396     if (bdrv_op_is_blocked(base_bs, BLOCK_OP_TYPE_COMMIT_TARGET, errp)) {
2397         goto out;
2398     }
2399 
2400     /* Do not allow attempts to commit an image into itself */
2401     if (top_bs == base_bs) {
2402         error_setg(errp, "cannot commit an image into itself");
2403         goto out;
2404     }
2405 
2406     if (top_bs == bs) {
2407         if (has_backing_file) {
2408             error_setg(errp, "'backing-file' specified,"
2409                              " but 'top' is the active layer");
2410             goto out;
2411         }
2412         commit_active_start(bs, base_bs, speed, on_error, block_job_cb,
2413                             bs, &local_err);
2414     } else {
2415         commit_start(bs, base_bs, top_bs, speed, on_error, block_job_cb, bs,
2416                      has_backing_file ? backing_file : NULL, &local_err);
2417     }
2418     if (local_err != NULL) {
2419         error_propagate(errp, local_err);
2420         goto out;
2421     }
2422 
2423 out:
2424     aio_context_release(aio_context);
2425 }
2426 
2427 void qmp_drive_backup(const char *device, const char *target,
2428                       bool has_format, const char *format,
2429                       enum MirrorSyncMode sync,
2430                       bool has_mode, enum NewImageMode mode,
2431                       bool has_speed, int64_t speed,
2432                       bool has_bitmap, const char *bitmap,
2433                       bool has_on_source_error, BlockdevOnError on_source_error,
2434                       bool has_on_target_error, BlockdevOnError on_target_error,
2435                       Error **errp)
2436 {
2437     BlockBackend *blk;
2438     BlockDriverState *bs;
2439     BlockDriverState *target_bs;
2440     BlockDriverState *source = NULL;
2441     BdrvDirtyBitmap *bmap = NULL;
2442     AioContext *aio_context;
2443     BlockDriver *drv = NULL;
2444     Error *local_err = NULL;
2445     int flags;
2446     int64_t size;
2447     int ret;
2448 
2449     if (!has_speed) {
2450         speed = 0;
2451     }
2452     if (!has_on_source_error) {
2453         on_source_error = BLOCKDEV_ON_ERROR_REPORT;
2454     }
2455     if (!has_on_target_error) {
2456         on_target_error = BLOCKDEV_ON_ERROR_REPORT;
2457     }
2458     if (!has_mode) {
2459         mode = NEW_IMAGE_MODE_ABSOLUTE_PATHS;
2460     }
2461 
2462     blk = blk_by_name(device);
2463     if (!blk) {
2464         error_set(errp, QERR_DEVICE_NOT_FOUND, device);
2465         return;
2466     }
2467     bs = blk_bs(blk);
2468 
2469     aio_context = bdrv_get_aio_context(bs);
2470     aio_context_acquire(aio_context);
2471 
2472     /* Although backup_run has this check too, we need to use bs->drv below, so
2473      * do an early check redundantly. */
2474     if (!bdrv_is_inserted(bs)) {
2475         error_set(errp, QERR_DEVICE_HAS_NO_MEDIUM, device);
2476         goto out;
2477     }
2478 
2479     if (!has_format) {
2480         format = mode == NEW_IMAGE_MODE_EXISTING ? NULL : bs->drv->format_name;
2481     }
2482     if (format) {
2483         drv = bdrv_find_format(format);
2484         if (!drv) {
2485             error_set(errp, QERR_INVALID_BLOCK_FORMAT, format);
2486             goto out;
2487         }
2488     }
2489 
2490     /* Early check to avoid creating target */
2491     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_BACKUP_SOURCE, errp)) {
2492         goto out;
2493     }
2494 
2495     flags = bs->open_flags | BDRV_O_RDWR;
2496 
2497     /* See if we have a backing HD we can use to create our new image
2498      * on top of. */
2499     if (sync == MIRROR_SYNC_MODE_TOP) {
2500         source = bs->backing_hd;
2501         if (!source) {
2502             sync = MIRROR_SYNC_MODE_FULL;
2503         }
2504     }
2505     if (sync == MIRROR_SYNC_MODE_NONE) {
2506         source = bs;
2507     }
2508 
2509     size = bdrv_getlength(bs);
2510     if (size < 0) {
2511         error_setg_errno(errp, -size, "bdrv_getlength failed");
2512         goto out;
2513     }
2514 
2515     if (mode != NEW_IMAGE_MODE_EXISTING) {
2516         assert(format && drv);
2517         if (source) {
2518             bdrv_img_create(target, format, source->filename,
2519                             source->drv->format_name, NULL,
2520                             size, flags, &local_err, false);
2521         } else {
2522             bdrv_img_create(target, format, NULL, NULL, NULL,
2523                             size, flags, &local_err, false);
2524         }
2525     }
2526 
2527     if (local_err) {
2528         error_propagate(errp, local_err);
2529         goto out;
2530     }
2531 
2532     target_bs = NULL;
2533     ret = bdrv_open(&target_bs, target, NULL, NULL, flags, drv, &local_err);
2534     if (ret < 0) {
2535         error_propagate(errp, local_err);
2536         goto out;
2537     }
2538 
2539     bdrv_set_aio_context(target_bs, aio_context);
2540 
2541     if (has_bitmap) {
2542         bmap = bdrv_find_dirty_bitmap(bs, bitmap);
2543         if (!bmap) {
2544             error_setg(errp, "Bitmap '%s' could not be found", bitmap);
2545             goto out;
2546         }
2547     }
2548 
2549     backup_start(bs, target_bs, speed, sync, bmap,
2550                  on_source_error, on_target_error,
2551                  block_job_cb, bs, &local_err);
2552     if (local_err != NULL) {
2553         bdrv_unref(target_bs);
2554         error_propagate(errp, local_err);
2555         goto out;
2556     }
2557 
2558 out:
2559     aio_context_release(aio_context);
2560 }
2561 
2562 BlockDeviceInfoList *qmp_query_named_block_nodes(Error **errp)
2563 {
2564     return bdrv_named_nodes_list(errp);
2565 }
2566 
2567 void qmp_blockdev_backup(const char *device, const char *target,
2568                          enum MirrorSyncMode sync,
2569                          bool has_speed, int64_t speed,
2570                          bool has_on_source_error,
2571                          BlockdevOnError on_source_error,
2572                          bool has_on_target_error,
2573                          BlockdevOnError on_target_error,
2574                          Error **errp)
2575 {
2576     BlockBackend *blk;
2577     BlockDriverState *bs;
2578     BlockDriverState *target_bs;
2579     Error *local_err = NULL;
2580     AioContext *aio_context;
2581 
2582     if (!has_speed) {
2583         speed = 0;
2584     }
2585     if (!has_on_source_error) {
2586         on_source_error = BLOCKDEV_ON_ERROR_REPORT;
2587     }
2588     if (!has_on_target_error) {
2589         on_target_error = BLOCKDEV_ON_ERROR_REPORT;
2590     }
2591 
2592     blk = blk_by_name(device);
2593     if (!blk) {
2594         error_setg(errp, "Device '%s' not found", device);
2595         return;
2596     }
2597     bs = blk_bs(blk);
2598 
2599     aio_context = bdrv_get_aio_context(bs);
2600     aio_context_acquire(aio_context);
2601 
2602     blk = blk_by_name(target);
2603     if (!blk) {
2604         error_setg(errp, "Device '%s' not found", target);
2605         goto out;
2606     }
2607     target_bs = blk_bs(blk);
2608 
2609     bdrv_ref(target_bs);
2610     bdrv_set_aio_context(target_bs, aio_context);
2611     backup_start(bs, target_bs, speed, sync, NULL, on_source_error,
2612                  on_target_error, block_job_cb, bs, &local_err);
2613     if (local_err != NULL) {
2614         bdrv_unref(target_bs);
2615         error_propagate(errp, local_err);
2616     }
2617 out:
2618     aio_context_release(aio_context);
2619 }
2620 
2621 #define DEFAULT_MIRROR_BUF_SIZE   (10 << 20)
2622 
2623 void qmp_drive_mirror(const char *device, const char *target,
2624                       bool has_format, const char *format,
2625                       bool has_node_name, const char *node_name,
2626                       bool has_replaces, const char *replaces,
2627                       enum MirrorSyncMode sync,
2628                       bool has_mode, enum NewImageMode mode,
2629                       bool has_speed, int64_t speed,
2630                       bool has_granularity, uint32_t granularity,
2631                       bool has_buf_size, int64_t buf_size,
2632                       bool has_on_source_error, BlockdevOnError on_source_error,
2633                       bool has_on_target_error, BlockdevOnError on_target_error,
2634                       Error **errp)
2635 {
2636     BlockBackend *blk;
2637     BlockDriverState *bs;
2638     BlockDriverState *source, *target_bs;
2639     AioContext *aio_context;
2640     BlockDriver *drv = NULL;
2641     Error *local_err = NULL;
2642     QDict *options = NULL;
2643     int flags;
2644     int64_t size;
2645     int ret;
2646 
2647     if (!has_speed) {
2648         speed = 0;
2649     }
2650     if (!has_on_source_error) {
2651         on_source_error = BLOCKDEV_ON_ERROR_REPORT;
2652     }
2653     if (!has_on_target_error) {
2654         on_target_error = BLOCKDEV_ON_ERROR_REPORT;
2655     }
2656     if (!has_mode) {
2657         mode = NEW_IMAGE_MODE_ABSOLUTE_PATHS;
2658     }
2659     if (!has_granularity) {
2660         granularity = 0;
2661     }
2662     if (!has_buf_size) {
2663         buf_size = DEFAULT_MIRROR_BUF_SIZE;
2664     }
2665 
2666     if (granularity != 0 && (granularity < 512 || granularity > 1048576 * 64)) {
2667         error_set(errp, QERR_INVALID_PARAMETER_VALUE, "granularity",
2668                   "a value in range [512B, 64MB]");
2669         return;
2670     }
2671     if (granularity & (granularity - 1)) {
2672         error_set(errp, QERR_INVALID_PARAMETER_VALUE, "granularity", "power of 2");
2673         return;
2674     }
2675 
2676     blk = blk_by_name(device);
2677     if (!blk) {
2678         error_set(errp, QERR_DEVICE_NOT_FOUND, device);
2679         return;
2680     }
2681     bs = blk_bs(blk);
2682 
2683     aio_context = bdrv_get_aio_context(bs);
2684     aio_context_acquire(aio_context);
2685 
2686     if (!bdrv_is_inserted(bs)) {
2687         error_set(errp, QERR_DEVICE_HAS_NO_MEDIUM, device);
2688         goto out;
2689     }
2690 
2691     if (!has_format) {
2692         format = mode == NEW_IMAGE_MODE_EXISTING ? NULL : bs->drv->format_name;
2693     }
2694     if (format) {
2695         drv = bdrv_find_format(format);
2696         if (!drv) {
2697             error_set(errp, QERR_INVALID_BLOCK_FORMAT, format);
2698             goto out;
2699         }
2700     }
2701 
2702     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_MIRROR, errp)) {
2703         goto out;
2704     }
2705 
2706     flags = bs->open_flags | BDRV_O_RDWR;
2707     source = bs->backing_hd;
2708     if (!source && sync == MIRROR_SYNC_MODE_TOP) {
2709         sync = MIRROR_SYNC_MODE_FULL;
2710     }
2711     if (sync == MIRROR_SYNC_MODE_NONE) {
2712         source = bs;
2713     }
2714 
2715     size = bdrv_getlength(bs);
2716     if (size < 0) {
2717         error_setg_errno(errp, -size, "bdrv_getlength failed");
2718         goto out;
2719     }
2720 
2721     if (has_replaces) {
2722         BlockDriverState *to_replace_bs;
2723         AioContext *replace_aio_context;
2724         int64_t replace_size;
2725 
2726         if (!has_node_name) {
2727             error_setg(errp, "a node-name must be provided when replacing a"
2728                              " named node of the graph");
2729             goto out;
2730         }
2731 
2732         to_replace_bs = check_to_replace_node(replaces, &local_err);
2733 
2734         if (!to_replace_bs) {
2735             error_propagate(errp, local_err);
2736             goto out;
2737         }
2738 
2739         replace_aio_context = bdrv_get_aio_context(to_replace_bs);
2740         aio_context_acquire(replace_aio_context);
2741         replace_size = bdrv_getlength(to_replace_bs);
2742         aio_context_release(replace_aio_context);
2743 
2744         if (size != replace_size) {
2745             error_setg(errp, "cannot replace image with a mirror image of "
2746                              "different size");
2747             goto out;
2748         }
2749     }
2750 
2751     if ((sync == MIRROR_SYNC_MODE_FULL || !source)
2752         && mode != NEW_IMAGE_MODE_EXISTING)
2753     {
2754         /* create new image w/o backing file */
2755         assert(format && drv);
2756         bdrv_img_create(target, format,
2757                         NULL, NULL, NULL, size, flags, &local_err, false);
2758     } else {
2759         switch (mode) {
2760         case NEW_IMAGE_MODE_EXISTING:
2761             break;
2762         case NEW_IMAGE_MODE_ABSOLUTE_PATHS:
2763             /* create new image with backing file */
2764             bdrv_img_create(target, format,
2765                             source->filename,
2766                             source->drv->format_name,
2767                             NULL, size, flags, &local_err, false);
2768             break;
2769         default:
2770             abort();
2771         }
2772     }
2773 
2774     if (local_err) {
2775         error_propagate(errp, local_err);
2776         goto out;
2777     }
2778 
2779     if (has_node_name) {
2780         options = qdict_new();
2781         qdict_put(options, "node-name", qstring_from_str(node_name));
2782     }
2783 
2784     /* Mirroring takes care of copy-on-write using the source's backing
2785      * file.
2786      */
2787     target_bs = NULL;
2788     ret = bdrv_open(&target_bs, target, NULL, options,
2789                     flags | BDRV_O_NO_BACKING, drv, &local_err);
2790     if (ret < 0) {
2791         error_propagate(errp, local_err);
2792         goto out;
2793     }
2794 
2795     bdrv_set_aio_context(target_bs, aio_context);
2796 
2797     /* pass the node name to replace to mirror start since it's loose coupling
2798      * and will allow to check whether the node still exist at mirror completion
2799      */
2800     mirror_start(bs, target_bs,
2801                  has_replaces ? replaces : NULL,
2802                  speed, granularity, buf_size, sync,
2803                  on_source_error, on_target_error,
2804                  block_job_cb, bs, &local_err);
2805     if (local_err != NULL) {
2806         bdrv_unref(target_bs);
2807         error_propagate(errp, local_err);
2808         goto out;
2809     }
2810 
2811 out:
2812     aio_context_release(aio_context);
2813 }
2814 
2815 /* Get the block job for a given device name and acquire its AioContext */
2816 static BlockJob *find_block_job(const char *device, AioContext **aio_context,
2817                                 Error **errp)
2818 {
2819     BlockBackend *blk;
2820     BlockDriverState *bs;
2821 
2822     blk = blk_by_name(device);
2823     if (!blk) {
2824         goto notfound;
2825     }
2826     bs = blk_bs(blk);
2827 
2828     *aio_context = bdrv_get_aio_context(bs);
2829     aio_context_acquire(*aio_context);
2830 
2831     if (!bs->job) {
2832         aio_context_release(*aio_context);
2833         goto notfound;
2834     }
2835 
2836     return bs->job;
2837 
2838 notfound:
2839     error_set(errp, ERROR_CLASS_DEVICE_NOT_ACTIVE,
2840               "No active block job on device '%s'", device);
2841     *aio_context = NULL;
2842     return NULL;
2843 }
2844 
2845 void qmp_block_job_set_speed(const char *device, int64_t speed, Error **errp)
2846 {
2847     AioContext *aio_context;
2848     BlockJob *job = find_block_job(device, &aio_context, errp);
2849 
2850     if (!job) {
2851         return;
2852     }
2853 
2854     block_job_set_speed(job, speed, errp);
2855     aio_context_release(aio_context);
2856 }
2857 
2858 void qmp_block_job_cancel(const char *device,
2859                           bool has_force, bool force, Error **errp)
2860 {
2861     AioContext *aio_context;
2862     BlockJob *job = find_block_job(device, &aio_context, errp);
2863 
2864     if (!job) {
2865         return;
2866     }
2867 
2868     if (!has_force) {
2869         force = false;
2870     }
2871 
2872     if (job->user_paused && !force) {
2873         error_setg(errp, "The block job for device '%s' is currently paused",
2874                    device);
2875         goto out;
2876     }
2877 
2878     trace_qmp_block_job_cancel(job);
2879     block_job_cancel(job);
2880 out:
2881     aio_context_release(aio_context);
2882 }
2883 
2884 void qmp_block_job_pause(const char *device, Error **errp)
2885 {
2886     AioContext *aio_context;
2887     BlockJob *job = find_block_job(device, &aio_context, errp);
2888 
2889     if (!job || job->user_paused) {
2890         return;
2891     }
2892 
2893     job->user_paused = true;
2894     trace_qmp_block_job_pause(job);
2895     block_job_pause(job);
2896     aio_context_release(aio_context);
2897 }
2898 
2899 void qmp_block_job_resume(const char *device, Error **errp)
2900 {
2901     AioContext *aio_context;
2902     BlockJob *job = find_block_job(device, &aio_context, errp);
2903 
2904     if (!job || !job->user_paused) {
2905         return;
2906     }
2907 
2908     job->user_paused = false;
2909     trace_qmp_block_job_resume(job);
2910     block_job_resume(job);
2911     aio_context_release(aio_context);
2912 }
2913 
2914 void qmp_block_job_complete(const char *device, Error **errp)
2915 {
2916     AioContext *aio_context;
2917     BlockJob *job = find_block_job(device, &aio_context, errp);
2918 
2919     if (!job) {
2920         return;
2921     }
2922 
2923     trace_qmp_block_job_complete(job);
2924     block_job_complete(job, errp);
2925     aio_context_release(aio_context);
2926 }
2927 
2928 void qmp_change_backing_file(const char *device,
2929                              const char *image_node_name,
2930                              const char *backing_file,
2931                              Error **errp)
2932 {
2933     BlockBackend *blk;
2934     BlockDriverState *bs = NULL;
2935     AioContext *aio_context;
2936     BlockDriverState *image_bs = NULL;
2937     Error *local_err = NULL;
2938     bool ro;
2939     int open_flags;
2940     int ret;
2941 
2942     blk = blk_by_name(device);
2943     if (!blk) {
2944         error_set(errp, QERR_DEVICE_NOT_FOUND, device);
2945         return;
2946     }
2947     bs = blk_bs(blk);
2948 
2949     aio_context = bdrv_get_aio_context(bs);
2950     aio_context_acquire(aio_context);
2951 
2952     image_bs = bdrv_lookup_bs(NULL, image_node_name, &local_err);
2953     if (local_err) {
2954         error_propagate(errp, local_err);
2955         goto out;
2956     }
2957 
2958     if (!image_bs) {
2959         error_setg(errp, "image file not found");
2960         goto out;
2961     }
2962 
2963     if (bdrv_find_base(image_bs) == image_bs) {
2964         error_setg(errp, "not allowing backing file change on an image "
2965                          "without a backing file");
2966         goto out;
2967     }
2968 
2969     /* even though we are not necessarily operating on bs, we need it to
2970      * determine if block ops are currently prohibited on the chain */
2971     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_CHANGE, errp)) {
2972         goto out;
2973     }
2974 
2975     /* final sanity check */
2976     if (!bdrv_chain_contains(bs, image_bs)) {
2977         error_setg(errp, "'%s' and image file are not in the same chain",
2978                    device);
2979         goto out;
2980     }
2981 
2982     /* if not r/w, reopen to make r/w */
2983     open_flags = image_bs->open_flags;
2984     ro = bdrv_is_read_only(image_bs);
2985 
2986     if (ro) {
2987         bdrv_reopen(image_bs, open_flags | BDRV_O_RDWR, &local_err);
2988         if (local_err) {
2989             error_propagate(errp, local_err);
2990             goto out;
2991         }
2992     }
2993 
2994     ret = bdrv_change_backing_file(image_bs, backing_file,
2995                                image_bs->drv ? image_bs->drv->format_name : "");
2996 
2997     if (ret < 0) {
2998         error_setg_errno(errp, -ret, "Could not change backing file to '%s'",
2999                          backing_file);
3000         /* don't exit here, so we can try to restore open flags if
3001          * appropriate */
3002     }
3003 
3004     if (ro) {
3005         bdrv_reopen(image_bs, open_flags, &local_err);
3006         if (local_err) {
3007             error_propagate(errp, local_err); /* will preserve prior errp */
3008         }
3009     }
3010 
3011 out:
3012     aio_context_release(aio_context);
3013 }
3014 
3015 void qmp_blockdev_add(BlockdevOptions *options, Error **errp)
3016 {
3017     QmpOutputVisitor *ov = qmp_output_visitor_new();
3018     BlockBackend *blk;
3019     QObject *obj;
3020     QDict *qdict;
3021     Error *local_err = NULL;
3022 
3023     /* Require an ID in the top level */
3024     if (!options->has_id) {
3025         error_setg(errp, "Block device needs an ID");
3026         goto fail;
3027     }
3028 
3029     /* TODO Sort it out in raw-posix and drive_new(): Reject aio=native with
3030      * cache.direct=false instead of silently switching to aio=threads, except
3031      * when called from drive_new().
3032      *
3033      * For now, simply forbidding the combination for all drivers will do. */
3034     if (options->has_aio && options->aio == BLOCKDEV_AIO_OPTIONS_NATIVE) {
3035         bool direct = options->has_cache &&
3036                       options->cache->has_direct &&
3037                       options->cache->direct;
3038         if (!direct) {
3039             error_setg(errp, "aio=native requires cache.direct=true");
3040             goto fail;
3041         }
3042     }
3043 
3044     visit_type_BlockdevOptions(qmp_output_get_visitor(ov),
3045                                &options, NULL, &local_err);
3046     if (local_err) {
3047         error_propagate(errp, local_err);
3048         goto fail;
3049     }
3050 
3051     obj = qmp_output_get_qobject(ov);
3052     qdict = qobject_to_qdict(obj);
3053 
3054     qdict_flatten(qdict);
3055 
3056     blk = blockdev_init(NULL, qdict, &local_err);
3057     if (local_err) {
3058         error_propagate(errp, local_err);
3059         goto fail;
3060     }
3061 
3062     if (bdrv_key_required(blk_bs(blk))) {
3063         blk_unref(blk);
3064         error_setg(errp, "blockdev-add doesn't support encrypted devices");
3065         goto fail;
3066     }
3067 
3068 fail:
3069     qmp_output_visitor_cleanup(ov);
3070 }
3071 
3072 BlockJobInfoList *qmp_query_block_jobs(Error **errp)
3073 {
3074     BlockJobInfoList *head = NULL, **p_next = &head;
3075     BlockDriverState *bs;
3076 
3077     for (bs = bdrv_next(NULL); bs; bs = bdrv_next(bs)) {
3078         AioContext *aio_context = bdrv_get_aio_context(bs);
3079 
3080         aio_context_acquire(aio_context);
3081 
3082         if (bs->job) {
3083             BlockJobInfoList *elem = g_new0(BlockJobInfoList, 1);
3084             elem->value = block_job_query(bs->job);
3085             *p_next = elem;
3086             p_next = &elem->next;
3087         }
3088 
3089         aio_context_release(aio_context);
3090     }
3091 
3092     return head;
3093 }
3094 
3095 QemuOptsList qemu_common_drive_opts = {
3096     .name = "drive",
3097     .head = QTAILQ_HEAD_INITIALIZER(qemu_common_drive_opts.head),
3098     .desc = {
3099         {
3100             .name = "snapshot",
3101             .type = QEMU_OPT_BOOL,
3102             .help = "enable/disable snapshot mode",
3103         },{
3104             .name = "discard",
3105             .type = QEMU_OPT_STRING,
3106             .help = "discard operation (ignore/off, unmap/on)",
3107         },{
3108             .name = "cache.writeback",
3109             .type = QEMU_OPT_BOOL,
3110             .help = "enables writeback mode for any caches",
3111         },{
3112             .name = "cache.direct",
3113             .type = QEMU_OPT_BOOL,
3114             .help = "enables use of O_DIRECT (bypass the host page cache)",
3115         },{
3116             .name = "cache.no-flush",
3117             .type = QEMU_OPT_BOOL,
3118             .help = "ignore any flush requests for the device",
3119         },{
3120             .name = "aio",
3121             .type = QEMU_OPT_STRING,
3122             .help = "host AIO implementation (threads, native)",
3123         },{
3124             .name = "format",
3125             .type = QEMU_OPT_STRING,
3126             .help = "disk format (raw, qcow2, ...)",
3127         },{
3128             .name = "rerror",
3129             .type = QEMU_OPT_STRING,
3130             .help = "read error action",
3131         },{
3132             .name = "werror",
3133             .type = QEMU_OPT_STRING,
3134             .help = "write error action",
3135         },{
3136             .name = "read-only",
3137             .type = QEMU_OPT_BOOL,
3138             .help = "open drive file as read-only",
3139         },{
3140             .name = "throttling.iops-total",
3141             .type = QEMU_OPT_NUMBER,
3142             .help = "limit total I/O operations per second",
3143         },{
3144             .name = "throttling.iops-read",
3145             .type = QEMU_OPT_NUMBER,
3146             .help = "limit read operations per second",
3147         },{
3148             .name = "throttling.iops-write",
3149             .type = QEMU_OPT_NUMBER,
3150             .help = "limit write operations per second",
3151         },{
3152             .name = "throttling.bps-total",
3153             .type = QEMU_OPT_NUMBER,
3154             .help = "limit total bytes per second",
3155         },{
3156             .name = "throttling.bps-read",
3157             .type = QEMU_OPT_NUMBER,
3158             .help = "limit read bytes per second",
3159         },{
3160             .name = "throttling.bps-write",
3161             .type = QEMU_OPT_NUMBER,
3162             .help = "limit write bytes per second",
3163         },{
3164             .name = "throttling.iops-total-max",
3165             .type = QEMU_OPT_NUMBER,
3166             .help = "I/O operations burst",
3167         },{
3168             .name = "throttling.iops-read-max",
3169             .type = QEMU_OPT_NUMBER,
3170             .help = "I/O operations read burst",
3171         },{
3172             .name = "throttling.iops-write-max",
3173             .type = QEMU_OPT_NUMBER,
3174             .help = "I/O operations write burst",
3175         },{
3176             .name = "throttling.bps-total-max",
3177             .type = QEMU_OPT_NUMBER,
3178             .help = "total bytes burst",
3179         },{
3180             .name = "throttling.bps-read-max",
3181             .type = QEMU_OPT_NUMBER,
3182             .help = "total bytes read burst",
3183         },{
3184             .name = "throttling.bps-write-max",
3185             .type = QEMU_OPT_NUMBER,
3186             .help = "total bytes write burst",
3187         },{
3188             .name = "throttling.iops-size",
3189             .type = QEMU_OPT_NUMBER,
3190             .help = "when limiting by iops max size of an I/O in bytes",
3191         },{
3192             .name = "copy-on-read",
3193             .type = QEMU_OPT_BOOL,
3194             .help = "copy read data from backing file into image file",
3195         },{
3196             .name = "detect-zeroes",
3197             .type = QEMU_OPT_STRING,
3198             .help = "try to optimize zero writes (off, on, unmap)",
3199         },
3200         { /* end of list */ }
3201     },
3202 };
3203 
3204 QemuOptsList qemu_drive_opts = {
3205     .name = "drive",
3206     .head = QTAILQ_HEAD_INITIALIZER(qemu_drive_opts.head),
3207     .desc = {
3208         /*
3209          * no elements => accept any params
3210          * validation will happen later
3211          */
3212         { /* end of list */ }
3213     },
3214 };
3215