xref: /openbmc/qemu/blockdev.c (revision 786a4ea8)
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 /* New and old BlockDriverState structs for atomic group operations */
1168 
1169 typedef struct BlkTransactionState BlkTransactionState;
1170 
1171 /* Only prepare() may fail. In a single transaction, only one of commit() or
1172    abort() will be called, clean() will always be called if it present. */
1173 typedef struct BdrvActionOps {
1174     /* Size of state struct, in bytes. */
1175     size_t instance_size;
1176     /* Prepare the work, must NOT be NULL. */
1177     void (*prepare)(BlkTransactionState *common, Error **errp);
1178     /* Commit the changes, can be NULL. */
1179     void (*commit)(BlkTransactionState *common);
1180     /* Abort the changes on fail, can be NULL. */
1181     void (*abort)(BlkTransactionState *common);
1182     /* Clean up resource in the end, can be NULL. */
1183     void (*clean)(BlkTransactionState *common);
1184 } BdrvActionOps;
1185 
1186 /*
1187  * This structure must be arranged as first member in child type, assuming
1188  * that compiler will also arrange it to the same address with parent instance.
1189  * Later it will be used in free().
1190  */
1191 struct BlkTransactionState {
1192     TransactionAction *action;
1193     const BdrvActionOps *ops;
1194     QSIMPLEQ_ENTRY(BlkTransactionState) entry;
1195 };
1196 
1197 /* internal snapshot private data */
1198 typedef struct InternalSnapshotState {
1199     BlkTransactionState common;
1200     BlockDriverState *bs;
1201     AioContext *aio_context;
1202     QEMUSnapshotInfo sn;
1203 } InternalSnapshotState;
1204 
1205 static void internal_snapshot_prepare(BlkTransactionState *common,
1206                                       Error **errp)
1207 {
1208     Error *local_err = NULL;
1209     const char *device;
1210     const char *name;
1211     BlockBackend *blk;
1212     BlockDriverState *bs;
1213     QEMUSnapshotInfo old_sn, *sn;
1214     bool ret;
1215     qemu_timeval tv;
1216     BlockdevSnapshotInternal *internal;
1217     InternalSnapshotState *state;
1218     int ret1;
1219 
1220     g_assert(common->action->kind ==
1221              TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC);
1222     internal = common->action->blockdev_snapshot_internal_sync;
1223     state = DO_UPCAST(InternalSnapshotState, common, common);
1224 
1225     /* 1. parse input */
1226     device = internal->device;
1227     name = internal->name;
1228 
1229     /* 2. check for validation */
1230     blk = blk_by_name(device);
1231     if (!blk) {
1232         error_set(errp, QERR_DEVICE_NOT_FOUND, device);
1233         return;
1234     }
1235     bs = blk_bs(blk);
1236 
1237     /* AioContext is released in .clean() */
1238     state->aio_context = bdrv_get_aio_context(bs);
1239     aio_context_acquire(state->aio_context);
1240 
1241     if (!bdrv_is_inserted(bs)) {
1242         error_set(errp, QERR_DEVICE_HAS_NO_MEDIUM, device);
1243         return;
1244     }
1245 
1246     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_INTERNAL_SNAPSHOT, errp)) {
1247         return;
1248     }
1249 
1250     if (bdrv_is_read_only(bs)) {
1251         error_set(errp, QERR_DEVICE_IS_READ_ONLY, device);
1252         return;
1253     }
1254 
1255     if (!bdrv_can_snapshot(bs)) {
1256         error_set(errp, QERR_BLOCK_FORMAT_FEATURE_NOT_SUPPORTED,
1257                   bs->drv->format_name, device, "internal snapshot");
1258         return;
1259     }
1260 
1261     if (!strlen(name)) {
1262         error_setg(errp, "Name is empty");
1263         return;
1264     }
1265 
1266     /* check whether a snapshot with name exist */
1267     ret = bdrv_snapshot_find_by_id_and_name(bs, NULL, name, &old_sn,
1268                                             &local_err);
1269     if (local_err) {
1270         error_propagate(errp, local_err);
1271         return;
1272     } else if (ret) {
1273         error_setg(errp,
1274                    "Snapshot with name '%s' already exists on device '%s'",
1275                    name, device);
1276         return;
1277     }
1278 
1279     /* 3. take the snapshot */
1280     sn = &state->sn;
1281     pstrcpy(sn->name, sizeof(sn->name), name);
1282     qemu_gettimeofday(&tv);
1283     sn->date_sec = tv.tv_sec;
1284     sn->date_nsec = tv.tv_usec * 1000;
1285     sn->vm_clock_nsec = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL);
1286 
1287     ret1 = bdrv_snapshot_create(bs, sn);
1288     if (ret1 < 0) {
1289         error_setg_errno(errp, -ret1,
1290                          "Failed to create snapshot '%s' on device '%s'",
1291                          name, device);
1292         return;
1293     }
1294 
1295     /* 4. succeed, mark a snapshot is created */
1296     state->bs = bs;
1297 }
1298 
1299 static void internal_snapshot_abort(BlkTransactionState *common)
1300 {
1301     InternalSnapshotState *state =
1302                              DO_UPCAST(InternalSnapshotState, common, common);
1303     BlockDriverState *bs = state->bs;
1304     QEMUSnapshotInfo *sn = &state->sn;
1305     Error *local_error = NULL;
1306 
1307     if (!bs) {
1308         return;
1309     }
1310 
1311     if (bdrv_snapshot_delete(bs, sn->id_str, sn->name, &local_error) < 0) {
1312         error_report("Failed to delete snapshot with id '%s' and name '%s' on "
1313                      "device '%s' in abort: %s",
1314                      sn->id_str,
1315                      sn->name,
1316                      bdrv_get_device_name(bs),
1317                      error_get_pretty(local_error));
1318         error_free(local_error);
1319     }
1320 }
1321 
1322 static void internal_snapshot_clean(BlkTransactionState *common)
1323 {
1324     InternalSnapshotState *state = DO_UPCAST(InternalSnapshotState,
1325                                              common, common);
1326 
1327     if (state->aio_context) {
1328         aio_context_release(state->aio_context);
1329     }
1330 }
1331 
1332 /* external snapshot private data */
1333 typedef struct ExternalSnapshotState {
1334     BlkTransactionState common;
1335     BlockDriverState *old_bs;
1336     BlockDriverState *new_bs;
1337     AioContext *aio_context;
1338 } ExternalSnapshotState;
1339 
1340 static void external_snapshot_prepare(BlkTransactionState *common,
1341                                       Error **errp)
1342 {
1343     BlockDriver *drv;
1344     int flags, ret;
1345     QDict *options = NULL;
1346     Error *local_err = NULL;
1347     bool has_device = false;
1348     const char *device;
1349     bool has_node_name = false;
1350     const char *node_name;
1351     bool has_snapshot_node_name = false;
1352     const char *snapshot_node_name;
1353     const char *new_image_file;
1354     const char *format = "qcow2";
1355     enum NewImageMode mode = NEW_IMAGE_MODE_ABSOLUTE_PATHS;
1356     ExternalSnapshotState *state =
1357                              DO_UPCAST(ExternalSnapshotState, common, common);
1358     TransactionAction *action = common->action;
1359 
1360     /* get parameters */
1361     g_assert(action->kind == TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC);
1362 
1363     has_device = action->blockdev_snapshot_sync->has_device;
1364     device = action->blockdev_snapshot_sync->device;
1365     has_node_name = action->blockdev_snapshot_sync->has_node_name;
1366     node_name = action->blockdev_snapshot_sync->node_name;
1367     has_snapshot_node_name =
1368         action->blockdev_snapshot_sync->has_snapshot_node_name;
1369     snapshot_node_name = action->blockdev_snapshot_sync->snapshot_node_name;
1370 
1371     new_image_file = action->blockdev_snapshot_sync->snapshot_file;
1372     if (action->blockdev_snapshot_sync->has_format) {
1373         format = action->blockdev_snapshot_sync->format;
1374     }
1375     if (action->blockdev_snapshot_sync->has_mode) {
1376         mode = action->blockdev_snapshot_sync->mode;
1377     }
1378 
1379     /* start processing */
1380     drv = bdrv_find_format(format);
1381     if (!drv) {
1382         error_set(errp, QERR_INVALID_BLOCK_FORMAT, format);
1383         return;
1384     }
1385 
1386     state->old_bs = bdrv_lookup_bs(has_device ? device : NULL,
1387                                    has_node_name ? node_name : NULL,
1388                                    &local_err);
1389     if (local_err) {
1390         error_propagate(errp, local_err);
1391         return;
1392     }
1393 
1394     if (has_node_name && !has_snapshot_node_name) {
1395         error_setg(errp, "New snapshot node name missing");
1396         return;
1397     }
1398 
1399     if (has_snapshot_node_name && bdrv_find_node(snapshot_node_name)) {
1400         error_setg(errp, "New snapshot node name already existing");
1401         return;
1402     }
1403 
1404     /* Acquire AioContext now so any threads operating on old_bs stop */
1405     state->aio_context = bdrv_get_aio_context(state->old_bs);
1406     aio_context_acquire(state->aio_context);
1407 
1408     if (!bdrv_is_inserted(state->old_bs)) {
1409         error_set(errp, QERR_DEVICE_HAS_NO_MEDIUM, device);
1410         return;
1411     }
1412 
1413     if (bdrv_op_is_blocked(state->old_bs,
1414                            BLOCK_OP_TYPE_EXTERNAL_SNAPSHOT, errp)) {
1415         return;
1416     }
1417 
1418     if (!bdrv_is_read_only(state->old_bs)) {
1419         if (bdrv_flush(state->old_bs)) {
1420             error_set(errp, QERR_IO_ERROR);
1421             return;
1422         }
1423     }
1424 
1425     if (!bdrv_is_first_non_filter(state->old_bs)) {
1426         error_set(errp, QERR_FEATURE_DISABLED, "snapshot");
1427         return;
1428     }
1429 
1430     flags = state->old_bs->open_flags;
1431 
1432     /* create new image w/backing file */
1433     if (mode != NEW_IMAGE_MODE_EXISTING) {
1434         bdrv_img_create(new_image_file, format,
1435                         state->old_bs->filename,
1436                         state->old_bs->drv->format_name,
1437                         NULL, -1, flags, &local_err, false);
1438         if (local_err) {
1439             error_propagate(errp, local_err);
1440             return;
1441         }
1442     }
1443 
1444     if (has_snapshot_node_name) {
1445         options = qdict_new();
1446         qdict_put(options, "node-name",
1447                   qstring_from_str(snapshot_node_name));
1448     }
1449 
1450     /* TODO Inherit bs->options or only take explicit options with an
1451      * extended QMP command? */
1452     assert(state->new_bs == NULL);
1453     ret = bdrv_open(&state->new_bs, new_image_file, NULL, options,
1454                     flags | BDRV_O_NO_BACKING, drv, &local_err);
1455     /* We will manually add the backing_hd field to the bs later */
1456     if (ret != 0) {
1457         error_propagate(errp, local_err);
1458     }
1459 }
1460 
1461 static void external_snapshot_commit(BlkTransactionState *common)
1462 {
1463     ExternalSnapshotState *state =
1464                              DO_UPCAST(ExternalSnapshotState, common, common);
1465 
1466     bdrv_set_aio_context(state->new_bs, state->aio_context);
1467 
1468     /* This removes our old bs and adds the new bs */
1469     bdrv_append(state->new_bs, state->old_bs);
1470     /* We don't need (or want) to use the transactional
1471      * bdrv_reopen_multiple() across all the entries at once, because we
1472      * don't want to abort all of them if one of them fails the reopen */
1473     bdrv_reopen(state->new_bs, state->new_bs->open_flags & ~BDRV_O_RDWR,
1474                 NULL);
1475 
1476     aio_context_release(state->aio_context);
1477 }
1478 
1479 static void external_snapshot_abort(BlkTransactionState *common)
1480 {
1481     ExternalSnapshotState *state =
1482                              DO_UPCAST(ExternalSnapshotState, common, common);
1483     if (state->new_bs) {
1484         bdrv_unref(state->new_bs);
1485     }
1486     if (state->aio_context) {
1487         aio_context_release(state->aio_context);
1488     }
1489 }
1490 
1491 typedef struct DriveBackupState {
1492     BlkTransactionState common;
1493     BlockDriverState *bs;
1494     AioContext *aio_context;
1495     BlockJob *job;
1496 } DriveBackupState;
1497 
1498 static void drive_backup_prepare(BlkTransactionState *common, Error **errp)
1499 {
1500     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1501     BlockDriverState *bs;
1502     BlockBackend *blk;
1503     DriveBackup *backup;
1504     Error *local_err = NULL;
1505 
1506     assert(common->action->kind == TRANSACTION_ACTION_KIND_DRIVE_BACKUP);
1507     backup = common->action->drive_backup;
1508 
1509     blk = blk_by_name(backup->device);
1510     if (!blk) {
1511         error_set(errp, QERR_DEVICE_NOT_FOUND, backup->device);
1512         return;
1513     }
1514     bs = blk_bs(blk);
1515 
1516     /* AioContext is released in .clean() */
1517     state->aio_context = bdrv_get_aio_context(bs);
1518     aio_context_acquire(state->aio_context);
1519 
1520     qmp_drive_backup(backup->device, backup->target,
1521                      backup->has_format, backup->format,
1522                      backup->sync,
1523                      backup->has_mode, backup->mode,
1524                      backup->has_speed, backup->speed,
1525                      backup->has_on_source_error, backup->on_source_error,
1526                      backup->has_on_target_error, backup->on_target_error,
1527                      &local_err);
1528     if (local_err) {
1529         error_propagate(errp, local_err);
1530         return;
1531     }
1532 
1533     state->bs = bs;
1534     state->job = state->bs->job;
1535 }
1536 
1537 static void drive_backup_abort(BlkTransactionState *common)
1538 {
1539     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1540     BlockDriverState *bs = state->bs;
1541 
1542     /* Only cancel if it's the job we started */
1543     if (bs && bs->job && bs->job == state->job) {
1544         block_job_cancel_sync(bs->job);
1545     }
1546 }
1547 
1548 static void drive_backup_clean(BlkTransactionState *common)
1549 {
1550     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1551 
1552     if (state->aio_context) {
1553         aio_context_release(state->aio_context);
1554     }
1555 }
1556 
1557 typedef struct BlockdevBackupState {
1558     BlkTransactionState common;
1559     BlockDriverState *bs;
1560     BlockJob *job;
1561     AioContext *aio_context;
1562 } BlockdevBackupState;
1563 
1564 static void blockdev_backup_prepare(BlkTransactionState *common, Error **errp)
1565 {
1566     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1567     BlockdevBackup *backup;
1568     BlockDriverState *bs, *target;
1569     BlockBackend *blk;
1570     Error *local_err = NULL;
1571 
1572     assert(common->action->kind == TRANSACTION_ACTION_KIND_BLOCKDEV_BACKUP);
1573     backup = common->action->blockdev_backup;
1574 
1575     blk = blk_by_name(backup->device);
1576     if (!blk) {
1577         error_setg(errp, "Device '%s' not found", backup->device);
1578         return;
1579     }
1580     bs = blk_bs(blk);
1581 
1582     blk = blk_by_name(backup->target);
1583     if (!blk) {
1584         error_setg(errp, "Device '%s' not found", backup->target);
1585         return;
1586     }
1587     target = blk_bs(blk);
1588 
1589     /* AioContext is released in .clean() */
1590     state->aio_context = bdrv_get_aio_context(bs);
1591     if (state->aio_context != bdrv_get_aio_context(target)) {
1592         state->aio_context = NULL;
1593         error_setg(errp, "Backup between two IO threads is not implemented");
1594         return;
1595     }
1596     aio_context_acquire(state->aio_context);
1597 
1598     qmp_blockdev_backup(backup->device, backup->target,
1599                         backup->sync,
1600                         backup->has_speed, backup->speed,
1601                         backup->has_on_source_error, backup->on_source_error,
1602                         backup->has_on_target_error, backup->on_target_error,
1603                         &local_err);
1604     if (local_err) {
1605         error_propagate(errp, local_err);
1606         return;
1607     }
1608 
1609     state->bs = bs;
1610     state->job = state->bs->job;
1611 }
1612 
1613 static void blockdev_backup_abort(BlkTransactionState *common)
1614 {
1615     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1616     BlockDriverState *bs = state->bs;
1617 
1618     /* Only cancel if it's the job we started */
1619     if (bs && bs->job && bs->job == state->job) {
1620         block_job_cancel_sync(bs->job);
1621     }
1622 }
1623 
1624 static void blockdev_backup_clean(BlkTransactionState *common)
1625 {
1626     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1627 
1628     if (state->aio_context) {
1629         aio_context_release(state->aio_context);
1630     }
1631 }
1632 
1633 static void abort_prepare(BlkTransactionState *common, Error **errp)
1634 {
1635     error_setg(errp, "Transaction aborted using Abort action");
1636 }
1637 
1638 static void abort_commit(BlkTransactionState *common)
1639 {
1640     g_assert_not_reached(); /* this action never succeeds */
1641 }
1642 
1643 static const BdrvActionOps actions[] = {
1644     [TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC] = {
1645         .instance_size = sizeof(ExternalSnapshotState),
1646         .prepare  = external_snapshot_prepare,
1647         .commit   = external_snapshot_commit,
1648         .abort = external_snapshot_abort,
1649     },
1650     [TRANSACTION_ACTION_KIND_DRIVE_BACKUP] = {
1651         .instance_size = sizeof(DriveBackupState),
1652         .prepare = drive_backup_prepare,
1653         .abort = drive_backup_abort,
1654         .clean = drive_backup_clean,
1655     },
1656     [TRANSACTION_ACTION_KIND_BLOCKDEV_BACKUP] = {
1657         .instance_size = sizeof(BlockdevBackupState),
1658         .prepare = blockdev_backup_prepare,
1659         .abort = blockdev_backup_abort,
1660         .clean = blockdev_backup_clean,
1661     },
1662     [TRANSACTION_ACTION_KIND_ABORT] = {
1663         .instance_size = sizeof(BlkTransactionState),
1664         .prepare = abort_prepare,
1665         .commit = abort_commit,
1666     },
1667     [TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC] = {
1668         .instance_size = sizeof(InternalSnapshotState),
1669         .prepare  = internal_snapshot_prepare,
1670         .abort = internal_snapshot_abort,
1671         .clean = internal_snapshot_clean,
1672     },
1673 };
1674 
1675 /*
1676  * 'Atomic' group operations.  The operations are performed as a set, and if
1677  * any fail then we roll back all operations in the group.
1678  */
1679 void qmp_transaction(TransactionActionList *dev_list, Error **errp)
1680 {
1681     TransactionActionList *dev_entry = dev_list;
1682     BlkTransactionState *state, *next;
1683     Error *local_err = NULL;
1684 
1685     QSIMPLEQ_HEAD(snap_bdrv_states, BlkTransactionState) snap_bdrv_states;
1686     QSIMPLEQ_INIT(&snap_bdrv_states);
1687 
1688     /* drain all i/o before any operations */
1689     bdrv_drain_all();
1690 
1691     /* We don't do anything in this loop that commits us to the operations */
1692     while (NULL != dev_entry) {
1693         TransactionAction *dev_info = NULL;
1694         const BdrvActionOps *ops;
1695 
1696         dev_info = dev_entry->value;
1697         dev_entry = dev_entry->next;
1698 
1699         assert(dev_info->kind < ARRAY_SIZE(actions));
1700 
1701         ops = &actions[dev_info->kind];
1702         assert(ops->instance_size > 0);
1703 
1704         state = g_malloc0(ops->instance_size);
1705         state->ops = ops;
1706         state->action = dev_info;
1707         QSIMPLEQ_INSERT_TAIL(&snap_bdrv_states, state, entry);
1708 
1709         state->ops->prepare(state, &local_err);
1710         if (local_err) {
1711             error_propagate(errp, local_err);
1712             goto delete_and_fail;
1713         }
1714     }
1715 
1716     QSIMPLEQ_FOREACH(state, &snap_bdrv_states, entry) {
1717         if (state->ops->commit) {
1718             state->ops->commit(state);
1719         }
1720     }
1721 
1722     /* success */
1723     goto exit;
1724 
1725 delete_and_fail:
1726     /* failure, and it is all-or-none; roll back all operations */
1727     QSIMPLEQ_FOREACH(state, &snap_bdrv_states, entry) {
1728         if (state->ops->abort) {
1729             state->ops->abort(state);
1730         }
1731     }
1732 exit:
1733     QSIMPLEQ_FOREACH_SAFE(state, &snap_bdrv_states, entry, next) {
1734         if (state->ops->clean) {
1735             state->ops->clean(state);
1736         }
1737         g_free(state);
1738     }
1739 }
1740 
1741 
1742 static void eject_device(BlockBackend *blk, int force, Error **errp)
1743 {
1744     BlockDriverState *bs = blk_bs(blk);
1745     AioContext *aio_context;
1746 
1747     aio_context = bdrv_get_aio_context(bs);
1748     aio_context_acquire(aio_context);
1749 
1750     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_EJECT, errp)) {
1751         goto out;
1752     }
1753     if (!blk_dev_has_removable_media(blk)) {
1754         error_setg(errp, "Device '%s' is not removable",
1755                    bdrv_get_device_name(bs));
1756         goto out;
1757     }
1758 
1759     if (blk_dev_is_medium_locked(blk) && !blk_dev_is_tray_open(blk)) {
1760         blk_dev_eject_request(blk, force);
1761         if (!force) {
1762             error_setg(errp, "Device '%s' is locked",
1763                        bdrv_get_device_name(bs));
1764             goto out;
1765         }
1766     }
1767 
1768     bdrv_close(bs);
1769 
1770 out:
1771     aio_context_release(aio_context);
1772 }
1773 
1774 void qmp_eject(const char *device, bool has_force, bool force, Error **errp)
1775 {
1776     BlockBackend *blk;
1777 
1778     blk = blk_by_name(device);
1779     if (!blk) {
1780         error_set(errp, QERR_DEVICE_NOT_FOUND, device);
1781         return;
1782     }
1783 
1784     eject_device(blk, force, errp);
1785 }
1786 
1787 void qmp_block_passwd(bool has_device, const char *device,
1788                       bool has_node_name, const char *node_name,
1789                       const char *password, Error **errp)
1790 {
1791     Error *local_err = NULL;
1792     BlockDriverState *bs;
1793     AioContext *aio_context;
1794 
1795     bs = bdrv_lookup_bs(has_device ? device : NULL,
1796                         has_node_name ? node_name : NULL,
1797                         &local_err);
1798     if (local_err) {
1799         error_propagate(errp, local_err);
1800         return;
1801     }
1802 
1803     aio_context = bdrv_get_aio_context(bs);
1804     aio_context_acquire(aio_context);
1805 
1806     bdrv_add_key(bs, password, errp);
1807 
1808     aio_context_release(aio_context);
1809 }
1810 
1811 /* Assumes AioContext is held */
1812 static void qmp_bdrv_open_encrypted(BlockDriverState *bs, const char *filename,
1813                                     int bdrv_flags, BlockDriver *drv,
1814                                     const char *password, Error **errp)
1815 {
1816     Error *local_err = NULL;
1817     int ret;
1818 
1819     ret = bdrv_open(&bs, filename, NULL, NULL, bdrv_flags, drv, &local_err);
1820     if (ret < 0) {
1821         error_propagate(errp, local_err);
1822         return;
1823     }
1824 
1825     bdrv_add_key(bs, password, errp);
1826 }
1827 
1828 void qmp_change_blockdev(const char *device, const char *filename,
1829                          const char *format, Error **errp)
1830 {
1831     BlockBackend *blk;
1832     BlockDriverState *bs;
1833     AioContext *aio_context;
1834     BlockDriver *drv = NULL;
1835     int bdrv_flags;
1836     Error *err = NULL;
1837 
1838     blk = blk_by_name(device);
1839     if (!blk) {
1840         error_set(errp, QERR_DEVICE_NOT_FOUND, device);
1841         return;
1842     }
1843     bs = blk_bs(blk);
1844 
1845     aio_context = bdrv_get_aio_context(bs);
1846     aio_context_acquire(aio_context);
1847 
1848     if (format) {
1849         drv = bdrv_find_whitelisted_format(format, bs->read_only);
1850         if (!drv) {
1851             error_set(errp, QERR_INVALID_BLOCK_FORMAT, format);
1852             goto out;
1853         }
1854     }
1855 
1856     eject_device(blk, 0, &err);
1857     if (err) {
1858         error_propagate(errp, err);
1859         goto out;
1860     }
1861 
1862     bdrv_flags = bdrv_is_read_only(bs) ? 0 : BDRV_O_RDWR;
1863     bdrv_flags |= bdrv_is_snapshot(bs) ? BDRV_O_SNAPSHOT : 0;
1864 
1865     qmp_bdrv_open_encrypted(bs, filename, bdrv_flags, drv, NULL, errp);
1866 
1867 out:
1868     aio_context_release(aio_context);
1869 }
1870 
1871 /* throttling disk I/O limits */
1872 void qmp_block_set_io_throttle(const char *device, int64_t bps, int64_t bps_rd,
1873                                int64_t bps_wr,
1874                                int64_t iops,
1875                                int64_t iops_rd,
1876                                int64_t iops_wr,
1877                                bool has_bps_max,
1878                                int64_t bps_max,
1879                                bool has_bps_rd_max,
1880                                int64_t bps_rd_max,
1881                                bool has_bps_wr_max,
1882                                int64_t bps_wr_max,
1883                                bool has_iops_max,
1884                                int64_t iops_max,
1885                                bool has_iops_rd_max,
1886                                int64_t iops_rd_max,
1887                                bool has_iops_wr_max,
1888                                int64_t iops_wr_max,
1889                                bool has_iops_size,
1890                                int64_t iops_size, Error **errp)
1891 {
1892     ThrottleConfig cfg;
1893     BlockDriverState *bs;
1894     BlockBackend *blk;
1895     AioContext *aio_context;
1896 
1897     blk = blk_by_name(device);
1898     if (!blk) {
1899         error_set(errp, QERR_DEVICE_NOT_FOUND, device);
1900         return;
1901     }
1902     bs = blk_bs(blk);
1903 
1904     memset(&cfg, 0, sizeof(cfg));
1905     cfg.buckets[THROTTLE_BPS_TOTAL].avg = bps;
1906     cfg.buckets[THROTTLE_BPS_READ].avg  = bps_rd;
1907     cfg.buckets[THROTTLE_BPS_WRITE].avg = bps_wr;
1908 
1909     cfg.buckets[THROTTLE_OPS_TOTAL].avg = iops;
1910     cfg.buckets[THROTTLE_OPS_READ].avg  = iops_rd;
1911     cfg.buckets[THROTTLE_OPS_WRITE].avg = iops_wr;
1912 
1913     if (has_bps_max) {
1914         cfg.buckets[THROTTLE_BPS_TOTAL].max = bps_max;
1915     }
1916     if (has_bps_rd_max) {
1917         cfg.buckets[THROTTLE_BPS_READ].max = bps_rd_max;
1918     }
1919     if (has_bps_wr_max) {
1920         cfg.buckets[THROTTLE_BPS_WRITE].max = bps_wr_max;
1921     }
1922     if (has_iops_max) {
1923         cfg.buckets[THROTTLE_OPS_TOTAL].max = iops_max;
1924     }
1925     if (has_iops_rd_max) {
1926         cfg.buckets[THROTTLE_OPS_READ].max = iops_rd_max;
1927     }
1928     if (has_iops_wr_max) {
1929         cfg.buckets[THROTTLE_OPS_WRITE].max = iops_wr_max;
1930     }
1931 
1932     if (has_iops_size) {
1933         cfg.op_size = iops_size;
1934     }
1935 
1936     if (!check_throttle_config(&cfg, errp)) {
1937         return;
1938     }
1939 
1940     aio_context = bdrv_get_aio_context(bs);
1941     aio_context_acquire(aio_context);
1942 
1943     if (!bs->io_limits_enabled && throttle_enabled(&cfg)) {
1944         bdrv_io_limits_enable(bs);
1945     } else if (bs->io_limits_enabled && !throttle_enabled(&cfg)) {
1946         bdrv_io_limits_disable(bs);
1947     }
1948 
1949     if (bs->io_limits_enabled) {
1950         bdrv_set_io_limits(bs, &cfg);
1951     }
1952 
1953     aio_context_release(aio_context);
1954 }
1955 
1956 int hmp_drive_del(Monitor *mon, const QDict *qdict, QObject **ret_data)
1957 {
1958     const char *id = qdict_get_str(qdict, "id");
1959     BlockBackend *blk;
1960     BlockDriverState *bs;
1961     AioContext *aio_context;
1962     Error *local_err = NULL;
1963 
1964     blk = blk_by_name(id);
1965     if (!blk) {
1966         error_report("Device '%s' not found", id);
1967         return -1;
1968     }
1969     bs = blk_bs(blk);
1970 
1971     if (!blk_legacy_dinfo(blk)) {
1972         error_report("Deleting device added with blockdev-add"
1973                      " is not supported");
1974         return -1;
1975     }
1976 
1977     aio_context = bdrv_get_aio_context(bs);
1978     aio_context_acquire(aio_context);
1979 
1980     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_DRIVE_DEL, &local_err)) {
1981         error_report_err(local_err);
1982         aio_context_release(aio_context);
1983         return -1;
1984     }
1985 
1986     /* quiesce block driver; prevent further io */
1987     bdrv_drain_all();
1988     bdrv_flush(bs);
1989     bdrv_close(bs);
1990 
1991     /* if we have a device attached to this BlockDriverState
1992      * then we need to make the drive anonymous until the device
1993      * can be removed.  If this is a drive with no device backing
1994      * then we can just get rid of the block driver state right here.
1995      */
1996     if (blk_get_attached_dev(blk)) {
1997         blk_hide_on_behalf_of_hmp_drive_del(blk);
1998         /* Further I/O must not pause the guest */
1999         bdrv_set_on_error(bs, BLOCKDEV_ON_ERROR_REPORT,
2000                           BLOCKDEV_ON_ERROR_REPORT);
2001     } else {
2002         blk_unref(blk);
2003     }
2004 
2005     aio_context_release(aio_context);
2006     return 0;
2007 }
2008 
2009 void qmp_block_resize(bool has_device, const char *device,
2010                       bool has_node_name, const char *node_name,
2011                       int64_t size, Error **errp)
2012 {
2013     Error *local_err = NULL;
2014     BlockDriverState *bs;
2015     AioContext *aio_context;
2016     int ret;
2017 
2018     bs = bdrv_lookup_bs(has_device ? device : NULL,
2019                         has_node_name ? node_name : NULL,
2020                         &local_err);
2021     if (local_err) {
2022         error_propagate(errp, local_err);
2023         return;
2024     }
2025 
2026     aio_context = bdrv_get_aio_context(bs);
2027     aio_context_acquire(aio_context);
2028 
2029     if (!bdrv_is_first_non_filter(bs)) {
2030         error_set(errp, QERR_FEATURE_DISABLED, "resize");
2031         goto out;
2032     }
2033 
2034     if (size < 0) {
2035         error_set(errp, QERR_INVALID_PARAMETER_VALUE, "size", "a >0 size");
2036         goto out;
2037     }
2038 
2039     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_RESIZE, NULL)) {
2040         error_set(errp, QERR_DEVICE_IN_USE, device);
2041         goto out;
2042     }
2043 
2044     /* complete all in-flight operations before resizing the device */
2045     bdrv_drain_all();
2046 
2047     ret = bdrv_truncate(bs, size);
2048     switch (ret) {
2049     case 0:
2050         break;
2051     case -ENOMEDIUM:
2052         error_set(errp, QERR_DEVICE_HAS_NO_MEDIUM, device);
2053         break;
2054     case -ENOTSUP:
2055         error_set(errp, QERR_UNSUPPORTED);
2056         break;
2057     case -EACCES:
2058         error_set(errp, QERR_DEVICE_IS_READ_ONLY, device);
2059         break;
2060     case -EBUSY:
2061         error_set(errp, QERR_DEVICE_IN_USE, device);
2062         break;
2063     default:
2064         error_setg_errno(errp, -ret, "Could not resize");
2065         break;
2066     }
2067 
2068 out:
2069     aio_context_release(aio_context);
2070 }
2071 
2072 static void block_job_cb(void *opaque, int ret)
2073 {
2074     /* Note that this function may be executed from another AioContext besides
2075      * the QEMU main loop.  If you need to access anything that assumes the
2076      * QEMU global mutex, use a BH or introduce a mutex.
2077      */
2078 
2079     BlockDriverState *bs = opaque;
2080     const char *msg = NULL;
2081 
2082     trace_block_job_cb(bs, bs->job, ret);
2083 
2084     assert(bs->job);
2085 
2086     if (ret < 0) {
2087         msg = strerror(-ret);
2088     }
2089 
2090     if (block_job_is_cancelled(bs->job)) {
2091         block_job_event_cancelled(bs->job);
2092     } else {
2093         block_job_event_completed(bs->job, msg);
2094     }
2095 
2096     bdrv_put_ref_bh_schedule(bs);
2097 }
2098 
2099 void qmp_block_stream(const char *device,
2100                       bool has_base, const char *base,
2101                       bool has_backing_file, const char *backing_file,
2102                       bool has_speed, int64_t speed,
2103                       bool has_on_error, BlockdevOnError on_error,
2104                       Error **errp)
2105 {
2106     BlockBackend *blk;
2107     BlockDriverState *bs;
2108     BlockDriverState *base_bs = NULL;
2109     AioContext *aio_context;
2110     Error *local_err = NULL;
2111     const char *base_name = NULL;
2112 
2113     if (!has_on_error) {
2114         on_error = BLOCKDEV_ON_ERROR_REPORT;
2115     }
2116 
2117     blk = blk_by_name(device);
2118     if (!blk) {
2119         error_set(errp, QERR_DEVICE_NOT_FOUND, device);
2120         return;
2121     }
2122     bs = blk_bs(blk);
2123 
2124     aio_context = bdrv_get_aio_context(bs);
2125     aio_context_acquire(aio_context);
2126 
2127     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_STREAM, errp)) {
2128         goto out;
2129     }
2130 
2131     if (has_base) {
2132         base_bs = bdrv_find_backing_image(bs, base);
2133         if (base_bs == NULL) {
2134             error_set(errp, QERR_BASE_NOT_FOUND, base);
2135             goto out;
2136         }
2137         assert(bdrv_get_aio_context(base_bs) == aio_context);
2138         base_name = base;
2139     }
2140 
2141     /* if we are streaming the entire chain, the result will have no backing
2142      * file, and specifying one is therefore an error */
2143     if (base_bs == NULL && has_backing_file) {
2144         error_setg(errp, "backing file specified, but streaming the "
2145                          "entire chain");
2146         goto out;
2147     }
2148 
2149     /* backing_file string overrides base bs filename */
2150     base_name = has_backing_file ? backing_file : base_name;
2151 
2152     stream_start(bs, base_bs, base_name, has_speed ? speed : 0,
2153                  on_error, block_job_cb, bs, &local_err);
2154     if (local_err) {
2155         error_propagate(errp, local_err);
2156         goto out;
2157     }
2158 
2159     trace_qmp_block_stream(bs, bs->job);
2160 
2161 out:
2162     aio_context_release(aio_context);
2163 }
2164 
2165 void qmp_block_commit(const char *device,
2166                       bool has_base, const char *base,
2167                       bool has_top, const char *top,
2168                       bool has_backing_file, const char *backing_file,
2169                       bool has_speed, int64_t speed,
2170                       Error **errp)
2171 {
2172     BlockBackend *blk;
2173     BlockDriverState *bs;
2174     BlockDriverState *base_bs, *top_bs;
2175     AioContext *aio_context;
2176     Error *local_err = NULL;
2177     /* This will be part of the QMP command, if/when the
2178      * BlockdevOnError change for blkmirror makes it in
2179      */
2180     BlockdevOnError on_error = BLOCKDEV_ON_ERROR_REPORT;
2181 
2182     if (!has_speed) {
2183         speed = 0;
2184     }
2185 
2186     /* Important Note:
2187      *  libvirt relies on the DeviceNotFound error class in order to probe for
2188      *  live commit feature versions; for this to work, we must make sure to
2189      *  perform the device lookup before any generic errors that may occur in a
2190      *  scenario in which all optional arguments are omitted. */
2191     blk = blk_by_name(device);
2192     if (!blk) {
2193         error_set(errp, QERR_DEVICE_NOT_FOUND, device);
2194         return;
2195     }
2196     bs = blk_bs(blk);
2197 
2198     aio_context = bdrv_get_aio_context(bs);
2199     aio_context_acquire(aio_context);
2200 
2201     /* drain all i/o before commits */
2202     bdrv_drain_all();
2203 
2204     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_COMMIT_SOURCE, errp)) {
2205         goto out;
2206     }
2207 
2208     /* default top_bs is the active layer */
2209     top_bs = bs;
2210 
2211     if (has_top && top) {
2212         if (strcmp(bs->filename, top) != 0) {
2213             top_bs = bdrv_find_backing_image(bs, top);
2214         }
2215     }
2216 
2217     if (top_bs == NULL) {
2218         error_setg(errp, "Top image file %s not found", top ? top : "NULL");
2219         goto out;
2220     }
2221 
2222     assert(bdrv_get_aio_context(top_bs) == aio_context);
2223 
2224     if (has_base && base) {
2225         base_bs = bdrv_find_backing_image(top_bs, base);
2226     } else {
2227         base_bs = bdrv_find_base(top_bs);
2228     }
2229 
2230     if (base_bs == NULL) {
2231         error_set(errp, QERR_BASE_NOT_FOUND, base ? base : "NULL");
2232         goto out;
2233     }
2234 
2235     assert(bdrv_get_aio_context(base_bs) == aio_context);
2236 
2237     if (bdrv_op_is_blocked(base_bs, BLOCK_OP_TYPE_COMMIT_TARGET, errp)) {
2238         goto out;
2239     }
2240 
2241     /* Do not allow attempts to commit an image into itself */
2242     if (top_bs == base_bs) {
2243         error_setg(errp, "cannot commit an image into itself");
2244         goto out;
2245     }
2246 
2247     if (top_bs == bs) {
2248         if (has_backing_file) {
2249             error_setg(errp, "'backing-file' specified,"
2250                              " but 'top' is the active layer");
2251             goto out;
2252         }
2253         commit_active_start(bs, base_bs, speed, on_error, block_job_cb,
2254                             bs, &local_err);
2255     } else {
2256         commit_start(bs, base_bs, top_bs, speed, on_error, block_job_cb, bs,
2257                      has_backing_file ? backing_file : NULL, &local_err);
2258     }
2259     if (local_err != NULL) {
2260         error_propagate(errp, local_err);
2261         goto out;
2262     }
2263 
2264 out:
2265     aio_context_release(aio_context);
2266 }
2267 
2268 void qmp_drive_backup(const char *device, const char *target,
2269                       bool has_format, const char *format,
2270                       enum MirrorSyncMode sync,
2271                       bool has_mode, enum NewImageMode mode,
2272                       bool has_speed, int64_t speed,
2273                       bool has_on_source_error, BlockdevOnError on_source_error,
2274                       bool has_on_target_error, BlockdevOnError on_target_error,
2275                       Error **errp)
2276 {
2277     BlockBackend *blk;
2278     BlockDriverState *bs;
2279     BlockDriverState *target_bs;
2280     BlockDriverState *source = NULL;
2281     AioContext *aio_context;
2282     BlockDriver *drv = NULL;
2283     Error *local_err = NULL;
2284     int flags;
2285     int64_t size;
2286     int ret;
2287 
2288     if (!has_speed) {
2289         speed = 0;
2290     }
2291     if (!has_on_source_error) {
2292         on_source_error = BLOCKDEV_ON_ERROR_REPORT;
2293     }
2294     if (!has_on_target_error) {
2295         on_target_error = BLOCKDEV_ON_ERROR_REPORT;
2296     }
2297     if (!has_mode) {
2298         mode = NEW_IMAGE_MODE_ABSOLUTE_PATHS;
2299     }
2300 
2301     blk = blk_by_name(device);
2302     if (!blk) {
2303         error_set(errp, QERR_DEVICE_NOT_FOUND, device);
2304         return;
2305     }
2306     bs = blk_bs(blk);
2307 
2308     aio_context = bdrv_get_aio_context(bs);
2309     aio_context_acquire(aio_context);
2310 
2311     /* Although backup_run has this check too, we need to use bs->drv below, so
2312      * do an early check redundantly. */
2313     if (!bdrv_is_inserted(bs)) {
2314         error_set(errp, QERR_DEVICE_HAS_NO_MEDIUM, device);
2315         goto out;
2316     }
2317 
2318     if (!has_format) {
2319         format = mode == NEW_IMAGE_MODE_EXISTING ? NULL : bs->drv->format_name;
2320     }
2321     if (format) {
2322         drv = bdrv_find_format(format);
2323         if (!drv) {
2324             error_set(errp, QERR_INVALID_BLOCK_FORMAT, format);
2325             goto out;
2326         }
2327     }
2328 
2329     /* Early check to avoid creating target */
2330     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_BACKUP_SOURCE, errp)) {
2331         goto out;
2332     }
2333 
2334     flags = bs->open_flags | BDRV_O_RDWR;
2335 
2336     /* See if we have a backing HD we can use to create our new image
2337      * on top of. */
2338     if (sync == MIRROR_SYNC_MODE_TOP) {
2339         source = bs->backing_hd;
2340         if (!source) {
2341             sync = MIRROR_SYNC_MODE_FULL;
2342         }
2343     }
2344     if (sync == MIRROR_SYNC_MODE_NONE) {
2345         source = bs;
2346     }
2347 
2348     size = bdrv_getlength(bs);
2349     if (size < 0) {
2350         error_setg_errno(errp, -size, "bdrv_getlength failed");
2351         goto out;
2352     }
2353 
2354     if (mode != NEW_IMAGE_MODE_EXISTING) {
2355         assert(format && drv);
2356         if (source) {
2357             bdrv_img_create(target, format, source->filename,
2358                             source->drv->format_name, NULL,
2359                             size, flags, &local_err, false);
2360         } else {
2361             bdrv_img_create(target, format, NULL, NULL, NULL,
2362                             size, flags, &local_err, false);
2363         }
2364     }
2365 
2366     if (local_err) {
2367         error_propagate(errp, local_err);
2368         goto out;
2369     }
2370 
2371     target_bs = NULL;
2372     ret = bdrv_open(&target_bs, target, NULL, NULL, flags, drv, &local_err);
2373     if (ret < 0) {
2374         error_propagate(errp, local_err);
2375         goto out;
2376     }
2377 
2378     bdrv_set_aio_context(target_bs, aio_context);
2379 
2380     backup_start(bs, target_bs, speed, sync, on_source_error, on_target_error,
2381                  block_job_cb, bs, &local_err);
2382     if (local_err != NULL) {
2383         bdrv_unref(target_bs);
2384         error_propagate(errp, local_err);
2385         goto out;
2386     }
2387 
2388 out:
2389     aio_context_release(aio_context);
2390 }
2391 
2392 BlockDeviceInfoList *qmp_query_named_block_nodes(Error **errp)
2393 {
2394     return bdrv_named_nodes_list();
2395 }
2396 
2397 void qmp_blockdev_backup(const char *device, const char *target,
2398                          enum MirrorSyncMode sync,
2399                          bool has_speed, int64_t speed,
2400                          bool has_on_source_error,
2401                          BlockdevOnError on_source_error,
2402                          bool has_on_target_error,
2403                          BlockdevOnError on_target_error,
2404                          Error **errp)
2405 {
2406     BlockBackend *blk;
2407     BlockDriverState *bs;
2408     BlockDriverState *target_bs;
2409     Error *local_err = NULL;
2410     AioContext *aio_context;
2411 
2412     if (!has_speed) {
2413         speed = 0;
2414     }
2415     if (!has_on_source_error) {
2416         on_source_error = BLOCKDEV_ON_ERROR_REPORT;
2417     }
2418     if (!has_on_target_error) {
2419         on_target_error = BLOCKDEV_ON_ERROR_REPORT;
2420     }
2421 
2422     blk = blk_by_name(device);
2423     if (!blk) {
2424         error_setg(errp, "Device '%s' not found", device);
2425         return;
2426     }
2427     bs = blk_bs(blk);
2428 
2429     aio_context = bdrv_get_aio_context(bs);
2430     aio_context_acquire(aio_context);
2431 
2432     blk = blk_by_name(target);
2433     if (!blk) {
2434         error_setg(errp, "Device '%s' not found", target);
2435         goto out;
2436     }
2437     target_bs = blk_bs(blk);
2438 
2439     bdrv_ref(target_bs);
2440     bdrv_set_aio_context(target_bs, aio_context);
2441     backup_start(bs, target_bs, speed, sync, on_source_error, on_target_error,
2442                  block_job_cb, bs, &local_err);
2443     if (local_err != NULL) {
2444         bdrv_unref(target_bs);
2445         error_propagate(errp, local_err);
2446     }
2447 out:
2448     aio_context_release(aio_context);
2449 }
2450 
2451 #define DEFAULT_MIRROR_BUF_SIZE   (10 << 20)
2452 
2453 void qmp_drive_mirror(const char *device, const char *target,
2454                       bool has_format, const char *format,
2455                       bool has_node_name, const char *node_name,
2456                       bool has_replaces, const char *replaces,
2457                       enum MirrorSyncMode sync,
2458                       bool has_mode, enum NewImageMode mode,
2459                       bool has_speed, int64_t speed,
2460                       bool has_granularity, uint32_t granularity,
2461                       bool has_buf_size, int64_t buf_size,
2462                       bool has_on_source_error, BlockdevOnError on_source_error,
2463                       bool has_on_target_error, BlockdevOnError on_target_error,
2464                       Error **errp)
2465 {
2466     BlockBackend *blk;
2467     BlockDriverState *bs;
2468     BlockDriverState *source, *target_bs;
2469     AioContext *aio_context;
2470     BlockDriver *drv = NULL;
2471     Error *local_err = NULL;
2472     QDict *options = NULL;
2473     int flags;
2474     int64_t size;
2475     int ret;
2476 
2477     if (!has_speed) {
2478         speed = 0;
2479     }
2480     if (!has_on_source_error) {
2481         on_source_error = BLOCKDEV_ON_ERROR_REPORT;
2482     }
2483     if (!has_on_target_error) {
2484         on_target_error = BLOCKDEV_ON_ERROR_REPORT;
2485     }
2486     if (!has_mode) {
2487         mode = NEW_IMAGE_MODE_ABSOLUTE_PATHS;
2488     }
2489     if (!has_granularity) {
2490         granularity = 0;
2491     }
2492     if (!has_buf_size) {
2493         buf_size = DEFAULT_MIRROR_BUF_SIZE;
2494     }
2495 
2496     if (granularity != 0 && (granularity < 512 || granularity > 1048576 * 64)) {
2497         error_set(errp, QERR_INVALID_PARAMETER_VALUE, "granularity",
2498                   "a value in range [512B, 64MB]");
2499         return;
2500     }
2501     if (granularity & (granularity - 1)) {
2502         error_set(errp, QERR_INVALID_PARAMETER_VALUE, "granularity", "power of 2");
2503         return;
2504     }
2505 
2506     blk = blk_by_name(device);
2507     if (!blk) {
2508         error_set(errp, QERR_DEVICE_NOT_FOUND, device);
2509         return;
2510     }
2511     bs = blk_bs(blk);
2512 
2513     aio_context = bdrv_get_aio_context(bs);
2514     aio_context_acquire(aio_context);
2515 
2516     if (!bdrv_is_inserted(bs)) {
2517         error_set(errp, QERR_DEVICE_HAS_NO_MEDIUM, device);
2518         goto out;
2519     }
2520 
2521     if (!has_format) {
2522         format = mode == NEW_IMAGE_MODE_EXISTING ? NULL : bs->drv->format_name;
2523     }
2524     if (format) {
2525         drv = bdrv_find_format(format);
2526         if (!drv) {
2527             error_set(errp, QERR_INVALID_BLOCK_FORMAT, format);
2528             goto out;
2529         }
2530     }
2531 
2532     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_MIRROR, errp)) {
2533         goto out;
2534     }
2535 
2536     flags = bs->open_flags | BDRV_O_RDWR;
2537     source = bs->backing_hd;
2538     if (!source && sync == MIRROR_SYNC_MODE_TOP) {
2539         sync = MIRROR_SYNC_MODE_FULL;
2540     }
2541     if (sync == MIRROR_SYNC_MODE_NONE) {
2542         source = bs;
2543     }
2544 
2545     size = bdrv_getlength(bs);
2546     if (size < 0) {
2547         error_setg_errno(errp, -size, "bdrv_getlength failed");
2548         goto out;
2549     }
2550 
2551     if (has_replaces) {
2552         BlockDriverState *to_replace_bs;
2553         AioContext *replace_aio_context;
2554         int64_t replace_size;
2555 
2556         if (!has_node_name) {
2557             error_setg(errp, "a node-name must be provided when replacing a"
2558                              " named node of the graph");
2559             goto out;
2560         }
2561 
2562         to_replace_bs = check_to_replace_node(replaces, &local_err);
2563 
2564         if (!to_replace_bs) {
2565             error_propagate(errp, local_err);
2566             goto out;
2567         }
2568 
2569         replace_aio_context = bdrv_get_aio_context(to_replace_bs);
2570         aio_context_acquire(replace_aio_context);
2571         replace_size = bdrv_getlength(to_replace_bs);
2572         aio_context_release(replace_aio_context);
2573 
2574         if (size != replace_size) {
2575             error_setg(errp, "cannot replace image with a mirror image of "
2576                              "different size");
2577             goto out;
2578         }
2579     }
2580 
2581     if ((sync == MIRROR_SYNC_MODE_FULL || !source)
2582         && mode != NEW_IMAGE_MODE_EXISTING)
2583     {
2584         /* create new image w/o backing file */
2585         assert(format && drv);
2586         bdrv_img_create(target, format,
2587                         NULL, NULL, NULL, size, flags, &local_err, false);
2588     } else {
2589         switch (mode) {
2590         case NEW_IMAGE_MODE_EXISTING:
2591             break;
2592         case NEW_IMAGE_MODE_ABSOLUTE_PATHS:
2593             /* create new image with backing file */
2594             bdrv_img_create(target, format,
2595                             source->filename,
2596                             source->drv->format_name,
2597                             NULL, size, flags, &local_err, false);
2598             break;
2599         default:
2600             abort();
2601         }
2602     }
2603 
2604     if (local_err) {
2605         error_propagate(errp, local_err);
2606         goto out;
2607     }
2608 
2609     if (has_node_name) {
2610         options = qdict_new();
2611         qdict_put(options, "node-name", qstring_from_str(node_name));
2612     }
2613 
2614     /* Mirroring takes care of copy-on-write using the source's backing
2615      * file.
2616      */
2617     target_bs = NULL;
2618     ret = bdrv_open(&target_bs, target, NULL, options,
2619                     flags | BDRV_O_NO_BACKING, drv, &local_err);
2620     if (ret < 0) {
2621         error_propagate(errp, local_err);
2622         goto out;
2623     }
2624 
2625     bdrv_set_aio_context(target_bs, aio_context);
2626 
2627     /* pass the node name to replace to mirror start since it's loose coupling
2628      * and will allow to check whether the node still exist at mirror completion
2629      */
2630     mirror_start(bs, target_bs,
2631                  has_replaces ? replaces : NULL,
2632                  speed, granularity, buf_size, sync,
2633                  on_source_error, on_target_error,
2634                  block_job_cb, bs, &local_err);
2635     if (local_err != NULL) {
2636         bdrv_unref(target_bs);
2637         error_propagate(errp, local_err);
2638         goto out;
2639     }
2640 
2641 out:
2642     aio_context_release(aio_context);
2643 }
2644 
2645 /* Get the block job for a given device name and acquire its AioContext */
2646 static BlockJob *find_block_job(const char *device, AioContext **aio_context,
2647                                 Error **errp)
2648 {
2649     BlockBackend *blk;
2650     BlockDriverState *bs;
2651 
2652     blk = blk_by_name(device);
2653     if (!blk) {
2654         goto notfound;
2655     }
2656     bs = blk_bs(blk);
2657 
2658     *aio_context = bdrv_get_aio_context(bs);
2659     aio_context_acquire(*aio_context);
2660 
2661     if (!bs->job) {
2662         aio_context_release(*aio_context);
2663         goto notfound;
2664     }
2665 
2666     return bs->job;
2667 
2668 notfound:
2669     error_set(errp, ERROR_CLASS_DEVICE_NOT_ACTIVE,
2670               "No active block job on device '%s'", device);
2671     *aio_context = NULL;
2672     return NULL;
2673 }
2674 
2675 void qmp_block_job_set_speed(const char *device, int64_t speed, Error **errp)
2676 {
2677     AioContext *aio_context;
2678     BlockJob *job = find_block_job(device, &aio_context, errp);
2679 
2680     if (!job) {
2681         return;
2682     }
2683 
2684     block_job_set_speed(job, speed, errp);
2685     aio_context_release(aio_context);
2686 }
2687 
2688 void qmp_block_job_cancel(const char *device,
2689                           bool has_force, bool force, Error **errp)
2690 {
2691     AioContext *aio_context;
2692     BlockJob *job = find_block_job(device, &aio_context, errp);
2693 
2694     if (!job) {
2695         return;
2696     }
2697 
2698     if (!has_force) {
2699         force = false;
2700     }
2701 
2702     if (job->paused && !force) {
2703         error_setg(errp, "The block job for device '%s' is currently paused",
2704                    device);
2705         goto out;
2706     }
2707 
2708     trace_qmp_block_job_cancel(job);
2709     block_job_cancel(job);
2710 out:
2711     aio_context_release(aio_context);
2712 }
2713 
2714 void qmp_block_job_pause(const char *device, Error **errp)
2715 {
2716     AioContext *aio_context;
2717     BlockJob *job = find_block_job(device, &aio_context, errp);
2718 
2719     if (!job) {
2720         return;
2721     }
2722 
2723     trace_qmp_block_job_pause(job);
2724     block_job_pause(job);
2725     aio_context_release(aio_context);
2726 }
2727 
2728 void qmp_block_job_resume(const char *device, Error **errp)
2729 {
2730     AioContext *aio_context;
2731     BlockJob *job = find_block_job(device, &aio_context, errp);
2732 
2733     if (!job) {
2734         return;
2735     }
2736 
2737     trace_qmp_block_job_resume(job);
2738     block_job_resume(job);
2739     aio_context_release(aio_context);
2740 }
2741 
2742 void qmp_block_job_complete(const char *device, Error **errp)
2743 {
2744     AioContext *aio_context;
2745     BlockJob *job = find_block_job(device, &aio_context, errp);
2746 
2747     if (!job) {
2748         return;
2749     }
2750 
2751     trace_qmp_block_job_complete(job);
2752     block_job_complete(job, errp);
2753     aio_context_release(aio_context);
2754 }
2755 
2756 void qmp_change_backing_file(const char *device,
2757                              const char *image_node_name,
2758                              const char *backing_file,
2759                              Error **errp)
2760 {
2761     BlockBackend *blk;
2762     BlockDriverState *bs = NULL;
2763     AioContext *aio_context;
2764     BlockDriverState *image_bs = NULL;
2765     Error *local_err = NULL;
2766     bool ro;
2767     int open_flags;
2768     int ret;
2769 
2770     blk = blk_by_name(device);
2771     if (!blk) {
2772         error_set(errp, QERR_DEVICE_NOT_FOUND, device);
2773         return;
2774     }
2775     bs = blk_bs(blk);
2776 
2777     aio_context = bdrv_get_aio_context(bs);
2778     aio_context_acquire(aio_context);
2779 
2780     image_bs = bdrv_lookup_bs(NULL, image_node_name, &local_err);
2781     if (local_err) {
2782         error_propagate(errp, local_err);
2783         goto out;
2784     }
2785 
2786     if (!image_bs) {
2787         error_setg(errp, "image file not found");
2788         goto out;
2789     }
2790 
2791     if (bdrv_find_base(image_bs) == image_bs) {
2792         error_setg(errp, "not allowing backing file change on an image "
2793                          "without a backing file");
2794         goto out;
2795     }
2796 
2797     /* even though we are not necessarily operating on bs, we need it to
2798      * determine if block ops are currently prohibited on the chain */
2799     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_CHANGE, errp)) {
2800         goto out;
2801     }
2802 
2803     /* final sanity check */
2804     if (!bdrv_chain_contains(bs, image_bs)) {
2805         error_setg(errp, "'%s' and image file are not in the same chain",
2806                    device);
2807         goto out;
2808     }
2809 
2810     /* if not r/w, reopen to make r/w */
2811     open_flags = image_bs->open_flags;
2812     ro = bdrv_is_read_only(image_bs);
2813 
2814     if (ro) {
2815         bdrv_reopen(image_bs, open_flags | BDRV_O_RDWR, &local_err);
2816         if (local_err) {
2817             error_propagate(errp, local_err);
2818             goto out;
2819         }
2820     }
2821 
2822     ret = bdrv_change_backing_file(image_bs, backing_file,
2823                                image_bs->drv ? image_bs->drv->format_name : "");
2824 
2825     if (ret < 0) {
2826         error_setg_errno(errp, -ret, "Could not change backing file to '%s'",
2827                          backing_file);
2828         /* don't exit here, so we can try to restore open flags if
2829          * appropriate */
2830     }
2831 
2832     if (ro) {
2833         bdrv_reopen(image_bs, open_flags, &local_err);
2834         if (local_err) {
2835             error_propagate(errp, local_err); /* will preserve prior errp */
2836         }
2837     }
2838 
2839 out:
2840     aio_context_release(aio_context);
2841 }
2842 
2843 void qmp_blockdev_add(BlockdevOptions *options, Error **errp)
2844 {
2845     QmpOutputVisitor *ov = qmp_output_visitor_new();
2846     BlockBackend *blk;
2847     QObject *obj;
2848     QDict *qdict;
2849     Error *local_err = NULL;
2850 
2851     /* Require an ID in the top level */
2852     if (!options->has_id) {
2853         error_setg(errp, "Block device needs an ID");
2854         goto fail;
2855     }
2856 
2857     /* TODO Sort it out in raw-posix and drive_new(): Reject aio=native with
2858      * cache.direct=false instead of silently switching to aio=threads, except
2859      * when called from drive_new().
2860      *
2861      * For now, simply forbidding the combination for all drivers will do. */
2862     if (options->has_aio && options->aio == BLOCKDEV_AIO_OPTIONS_NATIVE) {
2863         bool direct = options->has_cache &&
2864                       options->cache->has_direct &&
2865                       options->cache->direct;
2866         if (!direct) {
2867             error_setg(errp, "aio=native requires cache.direct=true");
2868             goto fail;
2869         }
2870     }
2871 
2872     visit_type_BlockdevOptions(qmp_output_get_visitor(ov),
2873                                &options, NULL, &local_err);
2874     if (local_err) {
2875         error_propagate(errp, local_err);
2876         goto fail;
2877     }
2878 
2879     obj = qmp_output_get_qobject(ov);
2880     qdict = qobject_to_qdict(obj);
2881 
2882     qdict_flatten(qdict);
2883 
2884     blk = blockdev_init(NULL, qdict, &local_err);
2885     if (local_err) {
2886         error_propagate(errp, local_err);
2887         goto fail;
2888     }
2889 
2890     if (bdrv_key_required(blk_bs(blk))) {
2891         blk_unref(blk);
2892         error_setg(errp, "blockdev-add doesn't support encrypted devices");
2893         goto fail;
2894     }
2895 
2896 fail:
2897     qmp_output_visitor_cleanup(ov);
2898 }
2899 
2900 BlockJobInfoList *qmp_query_block_jobs(Error **errp)
2901 {
2902     BlockJobInfoList *head = NULL, **p_next = &head;
2903     BlockDriverState *bs;
2904 
2905     for (bs = bdrv_next(NULL); bs; bs = bdrv_next(bs)) {
2906         AioContext *aio_context = bdrv_get_aio_context(bs);
2907 
2908         aio_context_acquire(aio_context);
2909 
2910         if (bs->job) {
2911             BlockJobInfoList *elem = g_new0(BlockJobInfoList, 1);
2912             elem->value = block_job_query(bs->job);
2913             *p_next = elem;
2914             p_next = &elem->next;
2915         }
2916 
2917         aio_context_release(aio_context);
2918     }
2919 
2920     return head;
2921 }
2922 
2923 QemuOptsList qemu_common_drive_opts = {
2924     .name = "drive",
2925     .head = QTAILQ_HEAD_INITIALIZER(qemu_common_drive_opts.head),
2926     .desc = {
2927         {
2928             .name = "snapshot",
2929             .type = QEMU_OPT_BOOL,
2930             .help = "enable/disable snapshot mode",
2931         },{
2932             .name = "discard",
2933             .type = QEMU_OPT_STRING,
2934             .help = "discard operation (ignore/off, unmap/on)",
2935         },{
2936             .name = "cache.writeback",
2937             .type = QEMU_OPT_BOOL,
2938             .help = "enables writeback mode for any caches",
2939         },{
2940             .name = "cache.direct",
2941             .type = QEMU_OPT_BOOL,
2942             .help = "enables use of O_DIRECT (bypass the host page cache)",
2943         },{
2944             .name = "cache.no-flush",
2945             .type = QEMU_OPT_BOOL,
2946             .help = "ignore any flush requests for the device",
2947         },{
2948             .name = "aio",
2949             .type = QEMU_OPT_STRING,
2950             .help = "host AIO implementation (threads, native)",
2951         },{
2952             .name = "format",
2953             .type = QEMU_OPT_STRING,
2954             .help = "disk format (raw, qcow2, ...)",
2955         },{
2956             .name = "rerror",
2957             .type = QEMU_OPT_STRING,
2958             .help = "read error action",
2959         },{
2960             .name = "werror",
2961             .type = QEMU_OPT_STRING,
2962             .help = "write error action",
2963         },{
2964             .name = "read-only",
2965             .type = QEMU_OPT_BOOL,
2966             .help = "open drive file as read-only",
2967         },{
2968             .name = "throttling.iops-total",
2969             .type = QEMU_OPT_NUMBER,
2970             .help = "limit total I/O operations per second",
2971         },{
2972             .name = "throttling.iops-read",
2973             .type = QEMU_OPT_NUMBER,
2974             .help = "limit read operations per second",
2975         },{
2976             .name = "throttling.iops-write",
2977             .type = QEMU_OPT_NUMBER,
2978             .help = "limit write operations per second",
2979         },{
2980             .name = "throttling.bps-total",
2981             .type = QEMU_OPT_NUMBER,
2982             .help = "limit total bytes per second",
2983         },{
2984             .name = "throttling.bps-read",
2985             .type = QEMU_OPT_NUMBER,
2986             .help = "limit read bytes per second",
2987         },{
2988             .name = "throttling.bps-write",
2989             .type = QEMU_OPT_NUMBER,
2990             .help = "limit write bytes per second",
2991         },{
2992             .name = "throttling.iops-total-max",
2993             .type = QEMU_OPT_NUMBER,
2994             .help = "I/O operations burst",
2995         },{
2996             .name = "throttling.iops-read-max",
2997             .type = QEMU_OPT_NUMBER,
2998             .help = "I/O operations read burst",
2999         },{
3000             .name = "throttling.iops-write-max",
3001             .type = QEMU_OPT_NUMBER,
3002             .help = "I/O operations write burst",
3003         },{
3004             .name = "throttling.bps-total-max",
3005             .type = QEMU_OPT_NUMBER,
3006             .help = "total bytes burst",
3007         },{
3008             .name = "throttling.bps-read-max",
3009             .type = QEMU_OPT_NUMBER,
3010             .help = "total bytes read burst",
3011         },{
3012             .name = "throttling.bps-write-max",
3013             .type = QEMU_OPT_NUMBER,
3014             .help = "total bytes write burst",
3015         },{
3016             .name = "throttling.iops-size",
3017             .type = QEMU_OPT_NUMBER,
3018             .help = "when limiting by iops max size of an I/O in bytes",
3019         },{
3020             .name = "copy-on-read",
3021             .type = QEMU_OPT_BOOL,
3022             .help = "copy read data from backing file into image file",
3023         },{
3024             .name = "detect-zeroes",
3025             .type = QEMU_OPT_STRING,
3026             .help = "try to optimize zero writes (off, on, unmap)",
3027         },
3028         { /* end of list */ }
3029     },
3030 };
3031 
3032 QemuOptsList qemu_drive_opts = {
3033     .name = "drive",
3034     .head = QTAILQ_HEAD_INITIALIZER(qemu_drive_opts.head),
3035     .desc = {
3036         /*
3037          * no elements => accept any params
3038          * validation will happen later
3039          */
3040         { /* end of list */ }
3041     },
3042 };
3043