?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
tests/avocado: s390x cpu topology core
[qemu/armbru.git] / blockdev.c
bloba01c62596b08f772fdb46bb549452699531eda00
1 /*
2 * QEMU host block devices
3 *
4 * Copyright (c) 2003-2008 Fabrice Bellard
5 *
6 * This work is licensed under the terms of the GNU GPL, version 2 or
7 * later. See the COPYING file in the top-level directory.
8 *
9 * This file incorporates work covered by the following copyright and
10 * permission notice:
11 *
12 * Copyright (c) 2003-2008 Fabrice Bellard
13 *
14 * Permission is hereby granted, free of charge, to any person obtaining a copy
15 * of this software and associated documentation files (the "Software"), to deal
16 * in the Software without restriction, including without limitation the rights
17 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
18 * copies of the Software, and to permit persons to whom the Software is
19 * furnished to do so, subject to the following conditions:
20 *
21 * The above copyright notice and this permission notice shall be included in
22 * all copies or substantial portions of the Software.
23 *
24 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
25 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
26 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
27 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
28 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
29 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
30 * THE SOFTWARE.
31 */
32
33 #include "qemu/osdep.h"
34 #include "sysemu/block-backend.h"
35 #include "sysemu/blockdev.h"
36 #include "hw/block/block.h"
37 #include "block/blockjob.h"
38 #include "block/dirty-bitmap.h"
39 #include "block/qdict.h"
40 #include "block/throttle-groups.h"
41 #include "monitor/monitor.h"
42 #include "qemu/error-report.h"
43 #include "qemu/option.h"
44 #include "qemu/qemu-print.h"
45 #include "qemu/config-file.h"
46 #include "qapi/qapi-commands-block.h"
47 #include "qapi/qapi-commands-transaction.h"
48 #include "qapi/qapi-visit-block-core.h"
49 #include "qapi/qmp/qdict.h"
50 #include "qapi/qmp/qnum.h"
51 #include "qapi/qmp/qstring.h"
52 #include "qapi/error.h"
53 #include "qapi/qmp/qerror.h"
54 #include "qapi/qmp/qlist.h"
55 #include "qapi/qobject-output-visitor.h"
56 #include "sysemu/sysemu.h"
57 #include "sysemu/iothread.h"
58 #include "block/block_int.h"
59 #include "block/trace.h"
60 #include "sysemu/runstate.h"
61 #include "sysemu/replay.h"
62 #include "qemu/cutils.h"
63 #include "qemu/help_option.h"
64 #include "qemu/main-loop.h"
65 #include "qemu/throttle-options.h"
66
67 /* Protected by BQL */
68 QTAILQ_HEAD(, BlockDriverState) monitor_bdrv_states =
69 QTAILQ_HEAD_INITIALIZER(monitor_bdrv_states);
70
71 void bdrv_set_monitor_owned(BlockDriverState *bs)
72 {
73 GLOBAL_STATE_CODE();
74 QTAILQ_INSERT_TAIL(&monitor_bdrv_states, bs, monitor_list);
75 }
76
77 static const char *const if_name[IF_COUNT] = {
78 [IF_NONE] = "none",
79 [IF_IDE] = "ide",
80 [IF_SCSI] = "scsi",
81 [IF_FLOPPY] = "floppy",
82 [IF_PFLASH] = "pflash",
83 [IF_MTD] = "mtd",
84 [IF_SD] = "sd",
85 [IF_VIRTIO] = "virtio",
86 [IF_XEN] = "xen",
87 };
88
89 static int if_max_devs[IF_COUNT] = {
90 /*
91 * Do not change these numbers! They govern how drive option
92 * index maps to unit and bus. That mapping is ABI.
93 *
94 * All controllers used to implement if=T drives need to support
95 * if_max_devs[T] units, for any T with if_max_devs[T] != 0.
96 * Otherwise, some index values map to "impossible" bus, unit
97 * values.
98 *
99 * For instance, if you change [IF_SCSI] to 255, -drive
100 * if=scsi,index=12 no longer means bus=1,unit=5, but
101 * bus=0,unit=12. With an lsi53c895a controller (7 units max),
102 * the drive can't be set up. Regression.
103 */
104 [IF_IDE] = 2,
105 [IF_SCSI] = 7,
106 };
107
108 /**
109 * Boards may call this to offer board-by-board overrides
110 * of the default, global values.
111 */
112 void override_max_devs(BlockInterfaceType type, int max_devs)
113 {
114 BlockBackend *blk;
115 DriveInfo *dinfo;
116
117 GLOBAL_STATE_CODE();
118
119 if (max_devs <= 0) {
120 return;
121 }
122
123 for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
124 dinfo = blk_legacy_dinfo(blk);
125 if (dinfo->type == type) {
126 fprintf(stderr, "Cannot override units-per-bus property of"
127 " the %s interface, because a drive of that type has"
128 " already been added.\n", if_name[type]);
129 g_assert_not_reached();
130 }
131 }
132
133 if_max_devs[type] = max_devs;
134 }
135
136 /*
137 * We automatically delete the drive when a device using it gets
138 * unplugged. Questionable feature, but we can't just drop it.
139 * Device models call blockdev_mark_auto_del() to schedule the
140 * automatic deletion, and generic qdev code calls blockdev_auto_del()
141 * when deletion is actually safe.
142 */
143 void blockdev_mark_auto_del(BlockBackend *blk)
144 {
145 DriveInfo *dinfo = blk_legacy_dinfo(blk);
146 BlockJob *job;
147
148 GLOBAL_STATE_CODE();
149
150 if (!dinfo) {
151 return;
152 }
153
154 JOB_LOCK_GUARD();
155
156 do {
157 job = block_job_next_locked(NULL);
158 while (job && (job->job.cancelled ||
159 job->job.deferred_to_main_loop ||
160 !block_job_has_bdrv(job, blk_bs(blk))))
161 {
162 job = block_job_next_locked(job);
163 }
164 if (job) {
165 /*
166 * This drops the job lock temporarily and polls, so we need to
167 * restart processing the list from the start after this.
168 */
169 job_cancel_locked(&job->job, false);
170 }
171 } while (job);
172
173 dinfo->auto_del = 1;
174 }
175
176 void blockdev_auto_del(BlockBackend *blk)
177 {
178 DriveInfo *dinfo = blk_legacy_dinfo(blk);
179 GLOBAL_STATE_CODE();
180
181 if (dinfo && dinfo->auto_del) {
182 monitor_remove_blk(blk);
183 blk_unref(blk);
184 }
185 }
186
187 static int drive_index_to_bus_id(BlockInterfaceType type, int index)
188 {
189 int max_devs = if_max_devs[type];
190 return max_devs ? index / max_devs : 0;
191 }
192
193 static int drive_index_to_unit_id(BlockInterfaceType type, int index)
194 {
195 int max_devs = if_max_devs[type];
196 return max_devs ? index % max_devs : index;
197 }
198
199 QemuOpts *drive_add(BlockInterfaceType type, int index, const char *file,
200 const char *optstr)
201 {
202 QemuOpts *opts;
203
204 GLOBAL_STATE_CODE();
205
206 opts = qemu_opts_parse_noisily(qemu_find_opts("drive"), optstr, false);
207 if (!opts) {
208 return NULL;
209 }
210 if (type != IF_DEFAULT) {
211 qemu_opt_set(opts, "if", if_name[type], &error_abort);
212 }
213 if (index >= 0) {
214 qemu_opt_set_number(opts, "index", index, &error_abort);
215 }
216 if (file)
217 qemu_opt_set(opts, "file", file, &error_abort);
218 return opts;
219 }
220
221 DriveInfo *drive_get(BlockInterfaceType type, int bus, int unit)
222 {
223 BlockBackend *blk;
224 DriveInfo *dinfo;
225
226 GLOBAL_STATE_CODE();
227
228 for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
229 dinfo = blk_legacy_dinfo(blk);
230 if (dinfo && dinfo->type == type
231 && dinfo->bus == bus && dinfo->unit == unit) {
232 return dinfo;
233 }
234 }
235
236 return NULL;
237 }
238
239 /*
240 * Check board claimed all -drive that are meant to be claimed.
241 * Fatal error if any remain unclaimed.
242 */
243 void drive_check_orphaned(void)
244 {
245 BlockBackend *blk;
246 DriveInfo *dinfo;
247 Location loc;
248 bool orphans = false;
249
250 GLOBAL_STATE_CODE();
251
252 for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
253 dinfo = blk_legacy_dinfo(blk);
254 /*
255 * Ignore default drives, because we create certain default
256 * drives unconditionally, then leave them unclaimed. Not the
257 * users fault.
258 * Ignore IF_VIRTIO, because it gets desugared into -device,
259 * so we can leave failing to -device.
260 * Ignore IF_NONE, because leaving unclaimed IF_NONE remains
261 * available for device_add is a feature.
262 */
263 if (dinfo->is_default || dinfo->type == IF_VIRTIO
264 || dinfo->type == IF_NONE) {
265 continue;
266 }
267 if (!blk_get_attached_dev(blk)) {
268 loc_push_none(&loc);
269 qemu_opts_loc_restore(dinfo->opts);
270 error_report("machine type does not support"
271 " if=%s,bus=%d,unit=%d",
272 if_name[dinfo->type], dinfo->bus, dinfo->unit);
273 loc_pop(&loc);
274 orphans = true;
275 }
276 }
277
278 if (orphans) {
279 exit(1);
280 }
281 }
282
283 DriveInfo *drive_get_by_index(BlockInterfaceType type, int index)
284 {
285 GLOBAL_STATE_CODE();
286 return drive_get(type,
287 drive_index_to_bus_id(type, index),
288 drive_index_to_unit_id(type, index));
289 }
290
291 int drive_get_max_bus(BlockInterfaceType type)
292 {
293 int max_bus;
294 BlockBackend *blk;
295 DriveInfo *dinfo;
296
297 GLOBAL_STATE_CODE();
298
299 max_bus = -1;
300 for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
301 dinfo = blk_legacy_dinfo(blk);
302 if (dinfo && dinfo->type == type && dinfo->bus > max_bus) {
303 max_bus = dinfo->bus;
304 }
305 }
306 return max_bus;
307 }
308
309 static void bdrv_format_print(void *opaque, const char *name)
310 {
311 qemu_printf(" %s", name);
312 }
313
314 typedef struct {
315 QEMUBH *bh;
316 BlockDriverState *bs;
317 } BDRVPutRefBH;
318
319 static int parse_block_error_action(const char *buf, bool is_read, Error **errp)
320 {
321 if (!strcmp(buf, "ignore")) {
322 return BLOCKDEV_ON_ERROR_IGNORE;
323 } else if (!is_read && !strcmp(buf, "enospc")) {
324 return BLOCKDEV_ON_ERROR_ENOSPC;
325 } else if (!strcmp(buf, "stop")) {
326 return BLOCKDEV_ON_ERROR_STOP;
327 } else if (!strcmp(buf, "report")) {
328 return BLOCKDEV_ON_ERROR_REPORT;
329 } else {
330 error_setg(errp, "'%s' invalid %s error action",
331 buf, is_read ? "read" : "write");
332 return -1;
333 }
334 }
335
336 static bool parse_stats_intervals(BlockAcctStats *stats, QList *intervals,
337 Error **errp)
338 {
339 const QListEntry *entry;
340 for (entry = qlist_first(intervals); entry; entry = qlist_next(entry)) {
341 switch (qobject_type(entry->value)) {
342
343 case QTYPE_QSTRING: {
344 uint64_t length;
345 const char *str = qstring_get_str(qobject_to(QString,
346 entry->value));
347 if (parse_uint_full(str, 10, &length) == 0 &&
348 length > 0 && length <= UINT_MAX) {
349 block_acct_add_interval(stats, (unsigned) length);
350 } else {
351 error_setg(errp, "Invalid interval length: %s", str);
352 return false;
353 }
354 break;
355 }
356
357 case QTYPE_QNUM: {
358 int64_t length = qnum_get_int(qobject_to(QNum, entry->value));
359
360 if (length > 0 && length <= UINT_MAX) {
361 block_acct_add_interval(stats, (unsigned) length);
362 } else {
363 error_setg(errp, "Invalid interval length: %" PRId64, length);
364 return false;
365 }
366 break;
367 }
368
369 default:
370 error_setg(errp, "The specification of stats-intervals is invalid");
371 return false;
372 }
373 }
374 return true;
375 }
376
377 typedef enum { MEDIA_DISK, MEDIA_CDROM } DriveMediaType;
378
379 /* All parameters but @opts are optional and may be set to NULL. */
380 static void extract_common_blockdev_options(QemuOpts *opts, int *bdrv_flags,
381 const char **throttling_group, ThrottleConfig *throttle_cfg,
382 BlockdevDetectZeroesOptions *detect_zeroes, Error **errp)
383 {
384 Error *local_error = NULL;
385 const char *aio;
386
387 if (bdrv_flags) {
388 if (qemu_opt_get_bool(opts, "copy-on-read", false)) {
389 *bdrv_flags |= BDRV_O_COPY_ON_READ;
390 }
391
392 if ((aio = qemu_opt_get(opts, "aio")) != NULL) {
393 if (bdrv_parse_aio(aio, bdrv_flags) < 0) {
394 error_setg(errp, "invalid aio option");
395 return;
396 }
397 }
398 }
399
400 /* disk I/O throttling */
401 if (throttling_group) {
402 *throttling_group = qemu_opt_get(opts, "throttling.group");
403 }
404
405 if (throttle_cfg) {
406 throttle_config_init(throttle_cfg);
407 throttle_cfg->buckets[THROTTLE_BPS_TOTAL].avg =
408 qemu_opt_get_number(opts, "throttling.bps-total", 0);
409 throttle_cfg->buckets[THROTTLE_BPS_READ].avg =
410 qemu_opt_get_number(opts, "throttling.bps-read", 0);
411 throttle_cfg->buckets[THROTTLE_BPS_WRITE].avg =
412 qemu_opt_get_number(opts, "throttling.bps-write", 0);
413 throttle_cfg->buckets[THROTTLE_OPS_TOTAL].avg =
414 qemu_opt_get_number(opts, "throttling.iops-total", 0);
415 throttle_cfg->buckets[THROTTLE_OPS_READ].avg =
416 qemu_opt_get_number(opts, "throttling.iops-read", 0);
417 throttle_cfg->buckets[THROTTLE_OPS_WRITE].avg =
418 qemu_opt_get_number(opts, "throttling.iops-write", 0);
419
420 throttle_cfg->buckets[THROTTLE_BPS_TOTAL].max =
421 qemu_opt_get_number(opts, "throttling.bps-total-max", 0);
422 throttle_cfg->buckets[THROTTLE_BPS_READ].max =
423 qemu_opt_get_number(opts, "throttling.bps-read-max", 0);
424 throttle_cfg->buckets[THROTTLE_BPS_WRITE].max =
425 qemu_opt_get_number(opts, "throttling.bps-write-max", 0);
426 throttle_cfg->buckets[THROTTLE_OPS_TOTAL].max =
427 qemu_opt_get_number(opts, "throttling.iops-total-max", 0);
428 throttle_cfg->buckets[THROTTLE_OPS_READ].max =
429 qemu_opt_get_number(opts, "throttling.iops-read-max", 0);
430 throttle_cfg->buckets[THROTTLE_OPS_WRITE].max =
431 qemu_opt_get_number(opts, "throttling.iops-write-max", 0);
432
433 throttle_cfg->buckets[THROTTLE_BPS_TOTAL].burst_length =
434 qemu_opt_get_number(opts, "throttling.bps-total-max-length", 1);
435 throttle_cfg->buckets[THROTTLE_BPS_READ].burst_length =
436 qemu_opt_get_number(opts, "throttling.bps-read-max-length", 1);
437 throttle_cfg->buckets[THROTTLE_BPS_WRITE].burst_length =
438 qemu_opt_get_number(opts, "throttling.bps-write-max-length", 1);
439 throttle_cfg->buckets[THROTTLE_OPS_TOTAL].burst_length =
440 qemu_opt_get_number(opts, "throttling.iops-total-max-length", 1);
441 throttle_cfg->buckets[THROTTLE_OPS_READ].burst_length =
442 qemu_opt_get_number(opts, "throttling.iops-read-max-length", 1);
443 throttle_cfg->buckets[THROTTLE_OPS_WRITE].burst_length =
444 qemu_opt_get_number(opts, "throttling.iops-write-max-length", 1);
445
446 throttle_cfg->op_size =
447 qemu_opt_get_number(opts, "throttling.iops-size", 0);
448
449 if (!throttle_is_valid(throttle_cfg, errp)) {
450 return;
451 }
452 }
453
454 if (detect_zeroes) {
455 *detect_zeroes =
456 qapi_enum_parse(&BlockdevDetectZeroesOptions_lookup,
457 qemu_opt_get(opts, "detect-zeroes"),
458 BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF,
459 &local_error);
460 if (local_error) {
461 error_propagate(errp, local_error);
462 return;
463 }
464 }
465 }
466
467 static OnOffAuto account_get_opt(QemuOpts *opts, const char *name)
468 {
469 if (!qemu_opt_find(opts, name)) {
470 return ON_OFF_AUTO_AUTO;
471 }
472 if (qemu_opt_get_bool(opts, name, true)) {
473 return ON_OFF_AUTO_ON;
474 }
475 return ON_OFF_AUTO_OFF;
476 }
477
478 /* Takes the ownership of bs_opts */
479 static BlockBackend *blockdev_init(const char *file, QDict *bs_opts,
480 Error **errp)
481 {
482 const char *buf;
483 int bdrv_flags = 0;
484 int on_read_error, on_write_error;
485 OnOffAuto account_invalid, account_failed;
486 bool writethrough, read_only;
487 BlockBackend *blk;
488 BlockDriverState *bs;
489 ThrottleConfig cfg;
490 int snapshot = 0;
491 Error *error = NULL;
492 QemuOpts *opts;
493 QDict *interval_dict = NULL;
494 QList *interval_list = NULL;
495 const char *id;
496 BlockdevDetectZeroesOptions detect_zeroes =
497 BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF;
498 const char *throttling_group = NULL;
499
500 /* Check common options by copying from bs_opts to opts, all other options
501 * stay in bs_opts for processing by bdrv_open(). */
502 id = qdict_get_try_str(bs_opts, "id");
503 opts = qemu_opts_create(&qemu_common_drive_opts, id, 1, errp);
504 if (!opts) {
505 goto err_no_opts;
506 }
507
508 if (!qemu_opts_absorb_qdict(opts, bs_opts, errp)) {
509 goto early_err;
510 }
511
512 if (id) {
513 qdict_del(bs_opts, "id");
514 }
515
516 /* extract parameters */
517 snapshot = qemu_opt_get_bool(opts, "snapshot", 0);
518
519 account_invalid = account_get_opt(opts, "stats-account-invalid");
520 account_failed = account_get_opt(opts, "stats-account-failed");
521
522 writethrough = !qemu_opt_get_bool(opts, BDRV_OPT_CACHE_WB, true);
523
524 id = qemu_opts_id(opts);
525
526 qdict_extract_subqdict(bs_opts, &interval_dict, "stats-intervals.");
527 qdict_array_split(interval_dict, &interval_list);
528
529 if (qdict_size(interval_dict) != 0) {
530 error_setg(errp, "Invalid option stats-intervals.%s",
531 qdict_first(interval_dict)->key);
532 goto early_err;
533 }
534
535 extract_common_blockdev_options(opts, &bdrv_flags, &throttling_group, &cfg,
536 &detect_zeroes, &error);
537 if (error) {
538 error_propagate(errp, error);
539 goto early_err;
540 }
541
542 if ((buf = qemu_opt_get(opts, "format")) != NULL) {
543 if (is_help_option(buf)) {
544 qemu_printf("Supported formats:");
545 bdrv_iterate_format(bdrv_format_print, NULL, false);
546 qemu_printf("\nSupported formats (read-only):");
547 bdrv_iterate_format(bdrv_format_print, NULL, true);
548 qemu_printf("\n");
549 goto early_err;
550 }
551
552 if (qdict_haskey(bs_opts, "driver")) {
553 error_setg(errp, "Cannot specify both 'driver' and 'format'");
554 goto early_err;
555 }
556 qdict_put_str(bs_opts, "driver", buf);
557 }
558
559 on_write_error = BLOCKDEV_ON_ERROR_ENOSPC;
560 if ((buf = qemu_opt_get(opts, "werror")) != NULL) {
561 on_write_error = parse_block_error_action(buf, 0, &error);
562 if (error) {
563 error_propagate(errp, error);
564 goto early_err;
565 }
566 }
567
568 on_read_error = BLOCKDEV_ON_ERROR_REPORT;
569 if ((buf = qemu_opt_get(opts, "rerror")) != NULL) {
570 on_read_error = parse_block_error_action(buf, 1, &error);
571 if (error) {
572 error_propagate(errp, error);
573 goto early_err;
574 }
575 }
576
577 if (snapshot) {
578 bdrv_flags |= BDRV_O_SNAPSHOT;
579 }
580
581 read_only = qemu_opt_get_bool(opts, BDRV_OPT_READ_ONLY, false);
582
583 /* init */
584 if ((!file || !*file) && !qdict_size(bs_opts)) {
585 BlockBackendRootState *blk_rs;
586
587 blk = blk_new(qemu_get_aio_context(), 0, BLK_PERM_ALL);
588 blk_rs = blk_get_root_state(blk);
589 blk_rs->open_flags = bdrv_flags | (read_only ? 0 : BDRV_O_RDWR);
590 blk_rs->detect_zeroes = detect_zeroes;
591
592 qobject_unref(bs_opts);
593 } else {
594 if (file && !*file) {
595 file = NULL;
596 }
597
598 /* bdrv_open() defaults to the values in bdrv_flags (for compatibility
599 * with other callers) rather than what we want as the real defaults.
600 * Apply the defaults here instead. */
601 qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_DIRECT, "off");
602 qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_NO_FLUSH, "off");
603 qdict_set_default_str(bs_opts, BDRV_OPT_READ_ONLY,
604 read_only ? "on" : "off");
605 qdict_set_default_str(bs_opts, BDRV_OPT_AUTO_READ_ONLY, "on");
606 assert((bdrv_flags & BDRV_O_CACHE_MASK) == 0);
607
608 if (runstate_check(RUN_STATE_INMIGRATE)) {
609 bdrv_flags |= BDRV_O_INACTIVE;
610 }
611
612 blk = blk_new_open(file, NULL, bs_opts, bdrv_flags, errp);
613 if (!blk) {
614 goto err_no_bs_opts;
615 }
616 bs = blk_bs(blk);
617
618 bs->detect_zeroes = detect_zeroes;
619
620 block_acct_setup(blk_get_stats(blk), account_invalid, account_failed);
621
622 if (!parse_stats_intervals(blk_get_stats(blk), interval_list, errp)) {
623 blk_unref(blk);
624 blk = NULL;
625 goto err_no_bs_opts;
626 }
627 }
628
629 /* disk I/O throttling */
630 if (throttle_enabled(&cfg)) {
631 if (!throttling_group) {
632 throttling_group = id;
633 }
634 blk_io_limits_enable(blk, throttling_group);
635 blk_set_io_limits(blk, &cfg);
636 }
637
638 blk_set_enable_write_cache(blk, !writethrough);
639 blk_set_on_error(blk, on_read_error, on_write_error);
640
641 if (!monitor_add_blk(blk, id, errp)) {
642 blk_unref(blk);
643 blk = NULL;
644 goto err_no_bs_opts;
645 }
646
647 err_no_bs_opts:
648 qemu_opts_del(opts);
649 qobject_unref(interval_dict);
650 qobject_unref(interval_list);
651 return blk;
652
653 early_err:
654 qemu_opts_del(opts);
655 qobject_unref(interval_dict);
656 qobject_unref(interval_list);
657 err_no_opts:
658 qobject_unref(bs_opts);
659 return NULL;
660 }
661
662 /* Takes the ownership of bs_opts */
663 BlockDriverState *bds_tree_init(QDict *bs_opts, Error **errp)
664 {
665 BlockDriverState *bs;
666 int bdrv_flags = 0;
667
668 GLOBAL_STATE_CODE();
669 /* bdrv_open() defaults to the values in bdrv_flags (for compatibility
670 * with other callers) rather than what we want as the real defaults.
671 * Apply the defaults here instead. */
672 qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_DIRECT, "off");
673 qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_NO_FLUSH, "off");
674 qdict_set_default_str(bs_opts, BDRV_OPT_READ_ONLY, "off");
675
676 if (runstate_check(RUN_STATE_INMIGRATE)) {
677 bdrv_flags |= BDRV_O_INACTIVE;
678 }
679
680 aio_context_acquire(qemu_get_aio_context());
681 bs = bdrv_open(NULL, NULL, bs_opts, bdrv_flags, errp);
682 aio_context_release(qemu_get_aio_context());
683
684 return bs;
685 }
686
687 void blockdev_close_all_bdrv_states(void)
688 {
689 BlockDriverState *bs, *next_bs;
690
691 GLOBAL_STATE_CODE();
692 QTAILQ_FOREACH_SAFE(bs, &monitor_bdrv_states, monitor_list, next_bs) {
693 AioContext *ctx = bdrv_get_aio_context(bs);
694
695 aio_context_acquire(ctx);
696 bdrv_unref(bs);
697 aio_context_release(ctx);
698 }
699 }
700
701 /* Iterates over the list of monitor-owned BlockDriverStates */
702 BlockDriverState *bdrv_next_monitor_owned(BlockDriverState *bs)
703 {
704 GLOBAL_STATE_CODE();
705 return bs ? QTAILQ_NEXT(bs, monitor_list)
706 : QTAILQ_FIRST(&monitor_bdrv_states);
707 }
708
709 static bool qemu_opt_rename(QemuOpts *opts, const char *from, const char *to,
710 Error **errp)
711 {
712 const char *value;
713
714 value = qemu_opt_get(opts, from);
715 if (value) {
716 if (qemu_opt_find(opts, to)) {
717 error_setg(errp, "'%s' and its alias '%s' can't be used at the "
718 "same time", to, from);
719 return false;
720 }
721 }
722
723 /* rename all items in opts */
724 while ((value = qemu_opt_get(opts, from))) {
725 qemu_opt_set(opts, to, value, &error_abort);
726 qemu_opt_unset(opts, from);
727 }
728 return true;
729 }
730
731 QemuOptsList qemu_legacy_drive_opts = {
732 .name = "drive",
733 .head = QTAILQ_HEAD_INITIALIZER(qemu_legacy_drive_opts.head),
734 .desc = {
735 {
736 .name = "bus",
737 .type = QEMU_OPT_NUMBER,
738 .help = "bus number",
739 },{
740 .name = "unit",
741 .type = QEMU_OPT_NUMBER,
742 .help = "unit number (i.e. lun for scsi)",
743 },{
744 .name = "index",
745 .type = QEMU_OPT_NUMBER,
746 .help = "index number",
747 },{
748 .name = "media",
749 .type = QEMU_OPT_STRING,
750 .help = "media type (disk, cdrom)",
751 },{
752 .name = "if",
753 .type = QEMU_OPT_STRING,
754 .help = "interface (ide, scsi, sd, mtd, floppy, pflash, virtio)",
755 },{
756 .name = "file",
757 .type = QEMU_OPT_STRING,
758 .help = "file name",
759 },
760
761 /* Options that are passed on, but have special semantics with -drive */
762 {
763 .name = BDRV_OPT_READ_ONLY,
764 .type = QEMU_OPT_BOOL,
765 .help = "open drive file as read-only",
766 },{
767 .name = "rerror",
768 .type = QEMU_OPT_STRING,
769 .help = "read error action",
770 },{
771 .name = "werror",
772 .type = QEMU_OPT_STRING,
773 .help = "write error action",
774 },{
775 .name = "copy-on-read",
776 .type = QEMU_OPT_BOOL,
777 .help = "copy read data from backing file into image file",
778 },
779
780 { /* end of list */ }
781 },
782 };
783
784 DriveInfo *drive_new(QemuOpts *all_opts, BlockInterfaceType block_default_type,
785 Error **errp)
786 {
787 const char *value;
788 BlockBackend *blk;
789 DriveInfo *dinfo = NULL;
790 QDict *bs_opts;
791 QemuOpts *legacy_opts;
792 DriveMediaType media = MEDIA_DISK;
793 BlockInterfaceType type;
794 int max_devs, bus_id, unit_id, index;
795 const char *werror, *rerror;
796 bool read_only = false;
797 bool copy_on_read;
798 const char *filename;
799 int i;
800
801 GLOBAL_STATE_CODE();
802
803 /* Change legacy command line options into QMP ones */
804 static const struct {
805 const char *from;
806 const char *to;
807 } opt_renames[] = {
808 { "iops", "throttling.iops-total" },
809 { "iops_rd", "throttling.iops-read" },
810 { "iops_wr", "throttling.iops-write" },
811
812 { "bps", "throttling.bps-total" },
813 { "bps_rd", "throttling.bps-read" },
814 { "bps_wr", "throttling.bps-write" },
815
816 { "iops_max", "throttling.iops-total-max" },
817 { "iops_rd_max", "throttling.iops-read-max" },
818 { "iops_wr_max", "throttling.iops-write-max" },
819
820 { "bps_max", "throttling.bps-total-max" },
821 { "bps_rd_max", "throttling.bps-read-max" },
822 { "bps_wr_max", "throttling.bps-write-max" },
823
824 { "iops_size", "throttling.iops-size" },
825
826 { "group", "throttling.group" },
827
828 { "readonly", BDRV_OPT_READ_ONLY },
829 };
830
831 for (i = 0; i < ARRAY_SIZE(opt_renames); i++) {
832 if (!qemu_opt_rename(all_opts, opt_renames[i].from,
833 opt_renames[i].to, errp)) {
834 return NULL;
835 }
836 }
837
838 value = qemu_opt_get(all_opts, "cache");
839 if (value) {
840 int flags = 0;
841 bool writethrough;
842
843 if (bdrv_parse_cache_mode(value, &flags, &writethrough) != 0) {
844 error_setg(errp, "invalid cache option");
845 return NULL;
846 }
847
848 /* Specific options take precedence */
849 if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_WB)) {
850 qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_WB,
851 !writethrough, &error_abort);
852 }
853 if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_DIRECT)) {
854 qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_DIRECT,
855 !!(flags & BDRV_O_NOCACHE), &error_abort);
856 }
857 if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_NO_FLUSH)) {
858 qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_NO_FLUSH,
859 !!(flags & BDRV_O_NO_FLUSH), &error_abort);
860 }
861 qemu_opt_unset(all_opts, "cache");
862 }
863
864 /* Get a QDict for processing the options */
865 bs_opts = qdict_new();
866 qemu_opts_to_qdict(all_opts, bs_opts);
867
868 legacy_opts = qemu_opts_create(&qemu_legacy_drive_opts, NULL, 0,
869 &error_abort);
870 if (!qemu_opts_absorb_qdict(legacy_opts, bs_opts, errp)) {
871 goto fail;
872 }
873
874 /* Media type */
875 value = qemu_opt_get(legacy_opts, "media");
876 if (value) {
877 if (!strcmp(value, "disk")) {
878 media = MEDIA_DISK;
879 } else if (!strcmp(value, "cdrom")) {
880 media = MEDIA_CDROM;
881 read_only = true;
882 } else {
883 error_setg(errp, "'%s' invalid media", value);
884 goto fail;
885 }
886 }
887
888 /* copy-on-read is disabled with a warning for read-only devices */
889 read_only |= qemu_opt_get_bool(legacy_opts, BDRV_OPT_READ_ONLY, false);
890 copy_on_read = qemu_opt_get_bool(legacy_opts, "copy-on-read", false);
891
892 if (read_only && copy_on_read) {
893 warn_report("disabling copy-on-read on read-only drive");
894 copy_on_read = false;
895 }
896
897 qdict_put_str(bs_opts, BDRV_OPT_READ_ONLY, read_only ? "on" : "off");
898 qdict_put_str(bs_opts, "copy-on-read", copy_on_read ? "on" : "off");
899
900 /* Controller type */
901 value = qemu_opt_get(legacy_opts, "if");
902 if (value) {
903 for (type = 0;
904 type < IF_COUNT && strcmp(value, if_name[type]);
905 type++) {
906 }
907 if (type == IF_COUNT) {
908 error_setg(errp, "unsupported bus type '%s'", value);
909 goto fail;
910 }
911 } else {
912 type = block_default_type;
913 }
914
915 /* Device address specified by bus/unit or index.
916 * If none was specified, try to find the first free one. */
917 bus_id = qemu_opt_get_number(legacy_opts, "bus", 0);
918 unit_id = qemu_opt_get_number(legacy_opts, "unit", -1);
919 index = qemu_opt_get_number(legacy_opts, "index", -1);
920
921 max_devs = if_max_devs[type];
922
923 if (index != -1) {
924 if (bus_id != 0 || unit_id != -1) {
925 error_setg(errp, "index cannot be used with bus and unit");
926 goto fail;
927 }
928 bus_id = drive_index_to_bus_id(type, index);
929 unit_id = drive_index_to_unit_id(type, index);
930 }
931
932 if (unit_id == -1) {
933 unit_id = 0;
934 while (drive_get(type, bus_id, unit_id) != NULL) {
935 unit_id++;
936 if (max_devs && unit_id >= max_devs) {
937 unit_id -= max_devs;
938 bus_id++;
939 }
940 }
941 }
942
943 if (max_devs && unit_id >= max_devs) {
944 error_setg(errp, "unit %d too big (max is %d)", unit_id, max_devs - 1);
945 goto fail;
946 }
947
948 if (drive_get(type, bus_id, unit_id) != NULL) {
949 error_setg(errp, "drive with bus=%d, unit=%d (index=%d) exists",
950 bus_id, unit_id, index);
951 goto fail;
952 }
953
954 /* no id supplied -> create one */
955 if (qemu_opts_id(all_opts) == NULL) {
956 char *new_id;
957 const char *mediastr = "";
958 if (type == IF_IDE || type == IF_SCSI) {
959 mediastr = (media == MEDIA_CDROM) ? "-cd" : "-hd";
960 }
961 if (max_devs) {
962 new_id = g_strdup_printf("%s%i%s%i", if_name[type], bus_id,
963 mediastr, unit_id);
964 } else {
965 new_id = g_strdup_printf("%s%s%i", if_name[type],
966 mediastr, unit_id);
967 }
968 qdict_put_str(bs_opts, "id", new_id);
969 g_free(new_id);
970 }
971
972 /* Add virtio block device */
973 if (type == IF_VIRTIO) {
974 QemuOpts *devopts;
975 devopts = qemu_opts_create(qemu_find_opts("device"), NULL, 0,
976 &error_abort);
977 qemu_opt_set(devopts, "driver", "virtio-blk", &error_abort);
978 qemu_opt_set(devopts, "drive", qdict_get_str(bs_opts, "id"),
979 &error_abort);
980 }
981
982 filename = qemu_opt_get(legacy_opts, "file");
983
984 /* Check werror/rerror compatibility with if=... */
985 werror = qemu_opt_get(legacy_opts, "werror");
986 if (werror != NULL) {
987 if (type != IF_IDE && type != IF_SCSI && type != IF_VIRTIO &&
988 type != IF_NONE) {
989 error_setg(errp, "werror is not supported by this bus type");
990 goto fail;
991 }
992 qdict_put_str(bs_opts, "werror", werror);
993 }
994
995 rerror = qemu_opt_get(legacy_opts, "rerror");
996 if (rerror != NULL) {
997 if (type != IF_IDE && type != IF_VIRTIO && type != IF_SCSI &&
998 type != IF_NONE) {
999 error_setg(errp, "rerror is not supported by this bus type");
1000 goto fail;
1001 }
1002 qdict_put_str(bs_opts, "rerror", rerror);
1003 }
1004
1005 /* Actual block device init: Functionality shared with blockdev-add */
1006 blk = blockdev_init(filename, bs_opts, errp);
1007 bs_opts = NULL;
1008 if (!blk) {
1009 goto fail;
1010 }
1011
1012 /* Create legacy DriveInfo */
1013 dinfo = g_malloc0(sizeof(*dinfo));
1014 dinfo->opts = all_opts;
1015
1016 dinfo->type = type;
1017 dinfo->bus = bus_id;
1018 dinfo->unit = unit_id;
1019
1020 blk_set_legacy_dinfo(blk, dinfo);
1021
1022 switch(type) {
1023 case IF_IDE:
1024 case IF_SCSI:
1025 case IF_XEN:
1026 case IF_NONE:
1027 dinfo->media_cd = media == MEDIA_CDROM;
1028 break;
1029 default:
1030 break;
1031 }
1032
1033 fail:
1034 qemu_opts_del(legacy_opts);
1035 qobject_unref(bs_opts);
1036 return dinfo;
1037 }
1038
1039 static BlockDriverState *qmp_get_root_bs(const char *name, Error **errp)
1040 {
1041 BlockDriverState *bs;
1042 AioContext *aio_context;
1043
1044 GRAPH_RDLOCK_GUARD_MAINLOOP();
1045
1046 bs = bdrv_lookup_bs(name, name, errp);
1047 if (bs == NULL) {
1048 return NULL;
1049 }
1050
1051 if (!bdrv_is_root_node(bs)) {
1052 error_setg(errp, "Need a root block node");
1053 return NULL;
1054 }
1055
1056 aio_context = bdrv_get_aio_context(bs);
1057 aio_context_acquire(aio_context);
1058
1059 if (!bdrv_is_inserted(bs)) {
1060 error_setg(errp, "Device has no medium");
1061 bs = NULL;
1062 }
1063
1064 aio_context_release(aio_context);
1065
1066 return bs;
1067 }
1068
1069 static void blockdev_do_action(TransactionAction *action, Error **errp)
1070 {
1071 TransactionActionList list;
1072
1073 list.value = action;
1074 list.next = NULL;
1075 qmp_transaction(&list, NULL, errp);
1076 }
1077
1078 void qmp_blockdev_snapshot_sync(const char *device, const char *node_name,
1079 const char *snapshot_file,
1080 const char *snapshot_node_name,
1081 const char *format,
1082 bool has_mode, NewImageMode mode, Error **errp)
1083 {
1084 BlockdevSnapshotSync snapshot = {
1085 .device = (char *) device,
1086 .node_name = (char *) node_name,
1087 .snapshot_file = (char *) snapshot_file,
1088 .snapshot_node_name = (char *) snapshot_node_name,
1089 .format = (char *) format,
1090 .has_mode = has_mode,
1091 .mode = mode,
1092 };
1093 TransactionAction action = {
1094 .type = TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC,
1095 .u.blockdev_snapshot_sync.data = &snapshot,
1096 };
1097 blockdev_do_action(&action, errp);
1098 }
1099
1100 void qmp_blockdev_snapshot(const char *node, const char *overlay,
1101 Error **errp)
1102 {
1103 BlockdevSnapshot snapshot_data = {
1104 .node = (char *) node,
1105 .overlay = (char *) overlay
1106 };
1107 TransactionAction action = {
1108 .type = TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT,
1109 .u.blockdev_snapshot.data = &snapshot_data,
1110 };
1111 blockdev_do_action(&action, errp);
1112 }
1113
1114 void qmp_blockdev_snapshot_internal_sync(const char *device,
1115 const char *name,
1116 Error **errp)
1117 {
1118 BlockdevSnapshotInternal snapshot = {
1119 .device = (char *) device,
1120 .name = (char *) name
1121 };
1122 TransactionAction action = {
1123 .type = TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC,
1124 .u.blockdev_snapshot_internal_sync.data = &snapshot,
1125 };
1126 blockdev_do_action(&action, errp);
1127 }
1128
1129 SnapshotInfo *qmp_blockdev_snapshot_delete_internal_sync(const char *device,
1130 const char *id,
1131 const char *name,
1132 Error **errp)
1133 {
1134 BlockDriverState *bs;
1135 AioContext *aio_context;
1136 QEMUSnapshotInfo sn;
1137 Error *local_err = NULL;
1138 SnapshotInfo *info = NULL;
1139 int ret;
1140
1141 GLOBAL_STATE_CODE();
1142 GRAPH_RDLOCK_GUARD_MAINLOOP();
1143
1144 bs = qmp_get_root_bs(device, errp);
1145 if (!bs) {
1146 return NULL;
1147 }
1148 aio_context = bdrv_get_aio_context(bs);
1149 aio_context_acquire(aio_context);
1150
1151 if (!id && !name) {
1152 error_setg(errp, "Name or id must be provided");
1153 goto out_aio_context;
1154 }
1155
1156 if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_INTERNAL_SNAPSHOT_DELETE, errp)) {
1157 goto out_aio_context;
1158 }
1159
1160 ret = bdrv_snapshot_find_by_id_and_name(bs, id, name, &sn, &local_err);
1161 if (local_err) {
1162 error_propagate(errp, local_err);
1163 goto out_aio_context;
1164 }
1165 if (!ret) {
1166 error_setg(errp,
1167 "Snapshot with id '%s' and name '%s' does not exist on "
1168 "device '%s'",
1169 STR_OR_NULL(id), STR_OR_NULL(name), device);
1170 goto out_aio_context;
1171 }
1172
1173 bdrv_snapshot_delete(bs, id, name, &local_err);
1174 if (local_err) {
1175 error_propagate(errp, local_err);
1176 goto out_aio_context;
1177 }
1178
1179 aio_context_release(aio_context);
1180
1181 info = g_new0(SnapshotInfo, 1);
1182 info->id = g_strdup(sn.id_str);
1183 info->name = g_strdup(sn.name);
1184 info->date_nsec = sn.date_nsec;
1185 info->date_sec = sn.date_sec;
1186 info->vm_state_size = sn.vm_state_size;
1187 info->vm_clock_nsec = sn.vm_clock_nsec % 1000000000;
1188 info->vm_clock_sec = sn.vm_clock_nsec / 1000000000;
1189 if (sn.icount != -1ULL) {
1190 info->icount = sn.icount;
1191 info->has_icount = true;
1192 }
1193
1194 return info;
1195
1196 out_aio_context:
1197 aio_context_release(aio_context);
1198 return NULL;
1199 }
1200
1201 /* internal snapshot private data */
1202 typedef struct InternalSnapshotState {
1203 BlockDriverState *bs;
1204 QEMUSnapshotInfo sn;
1205 bool created;
1206 } InternalSnapshotState;
1207
1208 static void internal_snapshot_abort(void *opaque);
1209 static void internal_snapshot_clean(void *opaque);
1210 TransactionActionDrv internal_snapshot_drv = {
1211 .abort = internal_snapshot_abort,
1212 .clean = internal_snapshot_clean,
1213 };
1214
1215 static void internal_snapshot_action(BlockdevSnapshotInternal *internal,
1216 Transaction *tran, Error **errp)
1217 {
1218 Error *local_err = NULL;
1219 const char *device;
1220 const char *name;
1221 BlockDriverState *bs;
1222 QEMUSnapshotInfo old_sn, *sn;
1223 bool ret;
1224 int64_t rt;
1225 InternalSnapshotState *state = g_new0(InternalSnapshotState, 1);
1226 AioContext *aio_context;
1227 int ret1;
1228
1229 GLOBAL_STATE_CODE();
1230 GRAPH_RDLOCK_GUARD_MAINLOOP();
1231
1232 tran_add(tran, &internal_snapshot_drv, state);
1233
1234 device = internal->device;
1235 name = internal->name;
1236
1237 bs = qmp_get_root_bs(device, errp);
1238 if (!bs) {
1239 return;
1240 }
1241
1242 aio_context = bdrv_get_aio_context(bs);
1243 aio_context_acquire(aio_context);
1244
1245 state->bs = bs;
1246
1247 /* Paired with .clean() */
1248 bdrv_drained_begin(bs);
1249
1250 if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_INTERNAL_SNAPSHOT, errp)) {
1251 goto out;
1252 }
1253
1254 if (bdrv_is_read_only(bs)) {
1255 error_setg(errp, "Device '%s' is read only", device);
1256 goto out;
1257 }
1258
1259 if (!bdrv_can_snapshot(bs)) {
1260 error_setg(errp, "Block format '%s' used by device '%s' "
1261 "does not support internal snapshots",
1262 bs->drv->format_name, device);
1263 goto out;
1264 }
1265
1266 if (!strlen(name)) {
1267 error_setg(errp, "Name is empty");
1268 goto out;
1269 }
1270
1271 /* check whether a snapshot with name exist */
1272 ret = bdrv_snapshot_find_by_id_and_name(bs, NULL, name, &old_sn,
1273 &local_err);
1274 if (local_err) {
1275 error_propagate(errp, local_err);
1276 goto out;
1277 } else if (ret) {
1278 error_setg(errp,
1279 "Snapshot with name '%s' already exists on device '%s'",
1280 name, device);
1281 goto out;
1282 }
1283
1284 /* 3. take the snapshot */
1285 sn = &state->sn;
1286 pstrcpy(sn->name, sizeof(sn->name), name);
1287 rt = g_get_real_time();
1288 sn->date_sec = rt / G_USEC_PER_SEC;
1289 sn->date_nsec = (rt % G_USEC_PER_SEC) * 1000;
1290 sn->vm_clock_nsec = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL);
1291 if (replay_mode != REPLAY_MODE_NONE) {
1292 sn->icount = replay_get_current_icount();
1293 } else {
1294 sn->icount = -1ULL;
1295 }
1296
1297 ret1 = bdrv_snapshot_create(bs, sn);
1298 if (ret1 < 0) {
1299 error_setg_errno(errp, -ret1,
1300 "Failed to create snapshot '%s' on device '%s'",
1301 name, device);
1302 goto out;
1303 }
1304
1305 /* 4. succeed, mark a snapshot is created */
1306 state->created = true;
1307
1308 out:
1309 aio_context_release(aio_context);
1310 }
1311
1312 static void internal_snapshot_abort(void *opaque)
1313 {
1314 InternalSnapshotState *state = opaque;
1315 BlockDriverState *bs = state->bs;
1316 QEMUSnapshotInfo *sn = &state->sn;
1317 AioContext *aio_context;
1318 Error *local_error = NULL;
1319
1320 GLOBAL_STATE_CODE();
1321 GRAPH_RDLOCK_GUARD_MAINLOOP();
1322
1323 if (!state->created) {
1324 return;
1325 }
1326
1327 aio_context = bdrv_get_aio_context(state->bs);
1328 aio_context_acquire(aio_context);
1329
1330 if (bdrv_snapshot_delete(bs, sn->id_str, sn->name, &local_error) < 0) {
1331 error_reportf_err(local_error,
1332 "Failed to delete snapshot with id '%s' and "
1333 "name '%s' on device '%s' in abort: ",
1334 sn->id_str, sn->name,
1335 bdrv_get_device_name(bs));
1336 }
1337
1338 aio_context_release(aio_context);
1339 }
1340
1341 static void internal_snapshot_clean(void *opaque)
1342 {
1343 g_autofree InternalSnapshotState *state = opaque;
1344 AioContext *aio_context;
1345
1346 if (!state->bs) {
1347 return;
1348 }
1349
1350 aio_context = bdrv_get_aio_context(state->bs);
1351 aio_context_acquire(aio_context);
1352
1353 bdrv_drained_end(state->bs);
1354
1355 aio_context_release(aio_context);
1356 }
1357
1358 /* external snapshot private data */
1359 typedef struct ExternalSnapshotState {
1360 BlockDriverState *old_bs;
1361 BlockDriverState *new_bs;
1362 bool overlay_appended;
1363 } ExternalSnapshotState;
1364
1365 static void external_snapshot_commit(void *opaque);
1366 static void external_snapshot_abort(void *opaque);
1367 static void external_snapshot_clean(void *opaque);
1368 TransactionActionDrv external_snapshot_drv = {
1369 .commit = external_snapshot_commit,
1370 .abort = external_snapshot_abort,
1371 .clean = external_snapshot_clean,
1372 };
1373
1374 static void external_snapshot_action(TransactionAction *action,
1375 Transaction *tran, Error **errp)
1376 {
1377 int ret;
1378 int flags = 0;
1379 QDict *options = NULL;
1380 Error *local_err = NULL;
1381 /* Device and node name of the image to generate the snapshot from */
1382 const char *device;
1383 const char *node_name;
1384 /* Reference to the new image (for 'blockdev-snapshot') */
1385 const char *snapshot_ref;
1386 /* File name of the new image (for 'blockdev-snapshot-sync') */
1387 const char *new_image_file;
1388 ExternalSnapshotState *state = g_new0(ExternalSnapshotState, 1);
1389 AioContext *aio_context;
1390 uint64_t perm, shared;
1391
1392 /* TODO We'll eventually have to take a writer lock in this function */
1393 GRAPH_RDLOCK_GUARD_MAINLOOP();
1394
1395 tran_add(tran, &external_snapshot_drv, state);
1396
1397 /* 'blockdev-snapshot' and 'blockdev-snapshot-sync' have similar
1398 * purpose but a different set of parameters */
1399 switch (action->type) {
1400 case TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT:
1401 {
1402 BlockdevSnapshot *s = action->u.blockdev_snapshot.data;
1403 device = s->node;
1404 node_name = s->node;
1405 new_image_file = NULL;
1406 snapshot_ref = s->overlay;
1407 }
1408 break;
1409 case TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC:
1410 {
1411 BlockdevSnapshotSync *s = action->u.blockdev_snapshot_sync.data;
1412 device = s->device;
1413 node_name = s->node_name;
1414 new_image_file = s->snapshot_file;
1415 snapshot_ref = NULL;
1416 }
1417 break;
1418 default:
1419 g_assert_not_reached();
1420 }
1421
1422 /* start processing */
1423
1424 state->old_bs = bdrv_lookup_bs(device, node_name, errp);
1425 if (!state->old_bs) {
1426 return;
1427 }
1428
1429 aio_context = bdrv_get_aio_context(state->old_bs);
1430 aio_context_acquire(aio_context);
1431
1432 /* Paired with .clean() */
1433 bdrv_drained_begin(state->old_bs);
1434
1435 if (!bdrv_is_inserted(state->old_bs)) {
1436 error_setg(errp, QERR_DEVICE_HAS_NO_MEDIUM, device);
1437 goto out;
1438 }
1439
1440 if (bdrv_op_is_blocked(state->old_bs,
1441 BLOCK_OP_TYPE_EXTERNAL_SNAPSHOT, errp)) {
1442 goto out;
1443 }
1444
1445 if (!bdrv_is_read_only(state->old_bs)) {
1446 if (bdrv_flush(state->old_bs)) {
1447 error_setg(errp, QERR_IO_ERROR);
1448 goto out;
1449 }
1450 }
1451
1452 if (action->type == TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC) {
1453 BlockdevSnapshotSync *s = action->u.blockdev_snapshot_sync.data;
1454 const char *format = s->format ?: "qcow2";
1455 enum NewImageMode mode;
1456 const char *snapshot_node_name = s->snapshot_node_name;
1457
1458 if (node_name && !snapshot_node_name) {
1459 error_setg(errp, "New overlay node-name missing");
1460 goto out;
1461 }
1462
1463 if (snapshot_node_name &&
1464 bdrv_lookup_bs(snapshot_node_name, snapshot_node_name, NULL)) {
1465 error_setg(errp, "New overlay node-name already in use");
1466 goto out;
1467 }
1468
1469 flags = state->old_bs->open_flags;
1470 flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_COPY_ON_READ);
1471 flags |= BDRV_O_NO_BACKING;
1472
1473 /* create new image w/backing file */
1474 mode = s->has_mode ? s->mode : NEW_IMAGE_MODE_ABSOLUTE_PATHS;
1475 if (mode != NEW_IMAGE_MODE_EXISTING) {
1476 int64_t size = bdrv_getlength(state->old_bs);
1477 if (size < 0) {
1478 error_setg_errno(errp, -size, "bdrv_getlength failed");
1479 goto out;
1480 }
1481 bdrv_refresh_filename(state->old_bs);
1482
1483 aio_context_release(aio_context);
1484 bdrv_img_create(new_image_file, format,
1485 state->old_bs->filename,
1486 state->old_bs->drv->format_name,
1487 NULL, size, flags, false, &local_err);
1488 aio_context_acquire(aio_context);
1489
1490 if (local_err) {
1491 error_propagate(errp, local_err);
1492 goto out;
1493 }
1494 }
1495
1496 options = qdict_new();
1497 if (snapshot_node_name) {
1498 qdict_put_str(options, "node-name", snapshot_node_name);
1499 }
1500 qdict_put_str(options, "driver", format);
1501 }
1502 aio_context_release(aio_context);
1503
1504 aio_context_acquire(qemu_get_aio_context());
1505 state->new_bs = bdrv_open(new_image_file, snapshot_ref, options, flags,
1506 errp);
1507 aio_context_release(qemu_get_aio_context());
1508
1509 /* We will manually add the backing_hd field to the bs later */
1510 if (!state->new_bs) {
1511 return;
1512 }
1513
1514 aio_context_acquire(aio_context);
1515
1516 /*
1517 * Allow attaching a backing file to an overlay that's already in use only
1518 * if the parents don't assume that they are already seeing a valid image.
1519 * (Specifically, allow it as a mirror target, which is write-only access.)
1520 */
1521 bdrv_get_cumulative_perm(state->new_bs, &perm, &shared);
1522 if (perm & BLK_PERM_CONSISTENT_READ) {
1523 error_setg(errp, "The overlay is already in use");
1524 goto out;
1525 }
1526
1527 if (state->new_bs->drv->is_filter) {
1528 error_setg(errp, "Filters cannot be used as overlays");
1529 goto out;
1530 }
1531
1532 if (bdrv_cow_child(state->new_bs)) {
1533 error_setg(errp, "The overlay already has a backing image");
1534 goto out;
1535 }
1536
1537 if (!state->new_bs->drv->supports_backing) {
1538 error_setg(errp, "The overlay does not support backing images");
1539 goto out;
1540 }
1541
1542 ret = bdrv_append(state->new_bs, state->old_bs, errp);
1543 if (ret < 0) {
1544 goto out;
1545 }
1546 state->overlay_appended = true;
1547
1548 out:
1549 aio_context_release(aio_context);
1550 }
1551
1552 static void external_snapshot_commit(void *opaque)
1553 {
1554 ExternalSnapshotState *state = opaque;
1555 AioContext *aio_context;
1556
1557 aio_context = bdrv_get_aio_context(state->old_bs);
1558 aio_context_acquire(aio_context);
1559
1560 /* We don't need (or want) to use the transactional
1561 * bdrv_reopen_multiple() across all the entries at once, because we
1562 * don't want to abort all of them if one of them fails the reopen */
1563 if (!qatomic_read(&state->old_bs->copy_on_read)) {
1564 bdrv_reopen_set_read_only(state->old_bs, true, NULL);
1565 }
1566
1567 aio_context_release(aio_context);
1568 }
1569
1570 static void external_snapshot_abort(void *opaque)
1571 {
1572 ExternalSnapshotState *state = opaque;
1573 if (state->new_bs) {
1574 if (state->overlay_appended) {
1575 AioContext *aio_context;
1576 AioContext *tmp_context;
1577 int ret;
1578
1579 aio_context = bdrv_get_aio_context(state->old_bs);
1580 aio_context_acquire(aio_context);
1581
1582 bdrv_ref(state->old_bs); /* we can't let bdrv_set_backind_hd()
1583 close state->old_bs; we need it */
1584 bdrv_set_backing_hd(state->new_bs, NULL, &error_abort);
1585
1586 /*
1587 * The call to bdrv_set_backing_hd() above returns state->old_bs to
1588 * the main AioContext. As we're still going to be using it, return
1589 * it to the AioContext it was before.
1590 */
1591 tmp_context = bdrv_get_aio_context(state->old_bs);
1592 if (aio_context != tmp_context) {
1593 aio_context_release(aio_context);
1594 aio_context_acquire(tmp_context);
1595
1596 ret = bdrv_try_change_aio_context(state->old_bs,
1597 aio_context, NULL, NULL);
1598 assert(ret == 0);
1599
1600 aio_context_release(tmp_context);
1601 aio_context_acquire(aio_context);
1602 }
1603
1604 bdrv_replace_node(state->new_bs, state->old_bs, &error_abort);
1605 bdrv_unref(state->old_bs); /* bdrv_replace_node() ref'ed old_bs */
1606
1607 aio_context_release(aio_context);
1608 }
1609 }
1610 }
1611
1612 static void external_snapshot_clean(void *opaque)
1613 {
1614 g_autofree ExternalSnapshotState *state = opaque;
1615 AioContext *aio_context;
1616
1617 if (!state->old_bs) {
1618 return;
1619 }
1620
1621 aio_context = bdrv_get_aio_context(state->old_bs);
1622 aio_context_acquire(aio_context);
1623
1624 bdrv_drained_end(state->old_bs);
1625 bdrv_unref(state->new_bs);
1626
1627 aio_context_release(aio_context);
1628 }
1629
1630 typedef struct DriveBackupState {
1631 BlockDriverState *bs;
1632 BlockJob *job;
1633 } DriveBackupState;
1634
1635 static BlockJob *do_backup_common(BackupCommon *backup,
1636 BlockDriverState *bs,
1637 BlockDriverState *target_bs,
1638 AioContext *aio_context,
1639 JobTxn *txn, Error **errp);
1640
1641 static void drive_backup_commit(void *opaque);
1642 static void drive_backup_abort(void *opaque);
1643 static void drive_backup_clean(void *opaque);
1644 TransactionActionDrv drive_backup_drv = {
1645 .commit = drive_backup_commit,
1646 .abort = drive_backup_abort,
1647 .clean = drive_backup_clean,
1648 };
1649
1650 static void drive_backup_action(DriveBackup *backup,
1651 JobTxn *block_job_txn,
1652 Transaction *tran, Error **errp)
1653 {
1654 DriveBackupState *state = g_new0(DriveBackupState, 1);
1655 BlockDriverState *bs;
1656 BlockDriverState *target_bs;
1657 BlockDriverState *source = NULL;
1658 AioContext *aio_context;
1659 AioContext *old_context;
1660 const char *format;
1661 QDict *options;
1662 Error *local_err = NULL;
1663 int flags;
1664 int64_t size;
1665 bool set_backing_hd = false;
1666 int ret;
1667
1668 GLOBAL_STATE_CODE();
1669
1670 tran_add(tran, &drive_backup_drv, state);
1671
1672 if (!backup->has_mode) {
1673 backup->mode = NEW_IMAGE_MODE_ABSOLUTE_PATHS;
1674 }
1675
1676 bs = bdrv_lookup_bs(backup->device, backup->device, errp);
1677 if (!bs) {
1678 return;
1679 }
1680
1681 if (!bs->drv) {
1682 error_setg(errp, "Device has no medium");
1683 return;
1684 }
1685
1686 aio_context = bdrv_get_aio_context(bs);
1687 aio_context_acquire(aio_context);
1688
1689 state->bs = bs;
1690 /* Paired with .clean() */
1691 bdrv_drained_begin(bs);
1692
1693 format = backup->format;
1694 if (!format && backup->mode != NEW_IMAGE_MODE_EXISTING) {
1695 format = bs->drv->format_name;
1696 }
1697
1698 /* Early check to avoid creating target */
1699 bdrv_graph_rdlock_main_loop();
1700 if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_BACKUP_SOURCE, errp)) {
1701 bdrv_graph_rdunlock_main_loop();
1702 goto out;
1703 }
1704 bdrv_graph_rdunlock_main_loop();
1705
1706 flags = bs->open_flags | BDRV_O_RDWR;
1707
1708 /*
1709 * See if we have a backing HD we can use to create our new image
1710 * on top of.
1711 */
1712 if (backup->sync == MIRROR_SYNC_MODE_TOP) {
1713 /*
1714 * Backup will not replace the source by the target, so none
1715 * of the filters skipped here will be removed (in contrast to
1716 * mirror). Therefore, we can skip all of them when looking
1717 * for the first COW relationship.
1718 */
1719 source = bdrv_cow_bs(bdrv_skip_filters(bs));
1720 if (!source) {
1721 backup->sync = MIRROR_SYNC_MODE_FULL;
1722 }
1723 }
1724 if (backup->sync == MIRROR_SYNC_MODE_NONE) {
1725 source = bs;
1726 flags |= BDRV_O_NO_BACKING;
1727 set_backing_hd = true;
1728 }
1729
1730 size = bdrv_getlength(bs);
1731 if (size < 0) {
1732 error_setg_errno(errp, -size, "bdrv_getlength failed");
1733 goto out;
1734 }
1735
1736 if (backup->mode != NEW_IMAGE_MODE_EXISTING) {
1737 assert(format);
1738 if (source) {
1739 /* Implicit filters should not appear in the filename */
1740 BlockDriverState *explicit_backing =
1741 bdrv_skip_implicit_filters(source);
1742
1743 bdrv_graph_rdlock_main_loop();
1744 bdrv_refresh_filename(explicit_backing);
1745 bdrv_graph_rdunlock_main_loop();
1746
1747 bdrv_img_create(backup->target, format,
1748 explicit_backing->filename,
1749 explicit_backing->drv->format_name, NULL,
1750 size, flags, false, &local_err);
1751 } else {
1752 bdrv_img_create(backup->target, format, NULL, NULL, NULL,
1753 size, flags, false, &local_err);
1754 }
1755 }
1756
1757 if (local_err) {
1758 error_propagate(errp, local_err);
1759 goto out;
1760 }
1761
1762 options = qdict_new();
1763 qdict_put_str(options, "discard", "unmap");
1764 qdict_put_str(options, "detect-zeroes", "unmap");
1765 if (format) {
1766 qdict_put_str(options, "driver", format);
1767 }
1768 aio_context_release(aio_context);
1769
1770 aio_context_acquire(qemu_get_aio_context());
1771 target_bs = bdrv_open(backup->target, NULL, options, flags, errp);
1772 aio_context_release(qemu_get_aio_context());
1773
1774 if (!target_bs) {
1775 return;
1776 }
1777
1778 /* Honor bdrv_try_change_aio_context() context acquisition requirements. */
1779 old_context = bdrv_get_aio_context(target_bs);
1780 aio_context_acquire(old_context);
1781
1782 ret = bdrv_try_change_aio_context(target_bs, aio_context, NULL, errp);
1783 if (ret < 0) {
1784 bdrv_unref(target_bs);
1785 aio_context_release(old_context);
1786 return;
1787 }
1788
1789 aio_context_release(old_context);
1790 aio_context_acquire(aio_context);
1791
1792 if (set_backing_hd) {
1793 if (bdrv_set_backing_hd(target_bs, source, errp) < 0) {
1794 goto unref;
1795 }
1796 }
1797
1798 state->job = do_backup_common(qapi_DriveBackup_base(backup),
1799 bs, target_bs, aio_context,
1800 block_job_txn, errp);
1801
1802 unref:
1803 bdrv_unref(target_bs);
1804 out:
1805 aio_context_release(aio_context);
1806 }
1807
1808 static void drive_backup_commit(void *opaque)
1809 {
1810 DriveBackupState *state = opaque;
1811 AioContext *aio_context;
1812
1813 aio_context = bdrv_get_aio_context(state->bs);
1814 aio_context_acquire(aio_context);
1815
1816 assert(state->job);
1817 job_start(&state->job->job);
1818
1819 aio_context_release(aio_context);
1820 }
1821
1822 static void drive_backup_abort(void *opaque)
1823 {
1824 DriveBackupState *state = opaque;
1825
1826 if (state->job) {
1827 job_cancel_sync(&state->job->job, true);
1828 }
1829 }
1830
1831 static void drive_backup_clean(void *opaque)
1832 {
1833 g_autofree DriveBackupState *state = opaque;
1834 AioContext *aio_context;
1835
1836 if (!state->bs) {
1837 return;
1838 }
1839
1840 aio_context = bdrv_get_aio_context(state->bs);
1841 aio_context_acquire(aio_context);
1842
1843 bdrv_drained_end(state->bs);
1844
1845 aio_context_release(aio_context);
1846 }
1847
1848 typedef struct BlockdevBackupState {
1849 BlockDriverState *bs;
1850 BlockJob *job;
1851 } BlockdevBackupState;
1852
1853 static void blockdev_backup_commit(void *opaque);
1854 static void blockdev_backup_abort(void *opaque);
1855 static void blockdev_backup_clean(void *opaque);
1856 TransactionActionDrv blockdev_backup_drv = {
1857 .commit = blockdev_backup_commit,
1858 .abort = blockdev_backup_abort,
1859 .clean = blockdev_backup_clean,
1860 };
1861
1862 static void blockdev_backup_action(BlockdevBackup *backup,
1863 JobTxn *block_job_txn,
1864 Transaction *tran, Error **errp)
1865 {
1866 BlockdevBackupState *state = g_new0(BlockdevBackupState, 1);
1867 BlockDriverState *bs;
1868 BlockDriverState *target_bs;
1869 AioContext *aio_context;
1870 AioContext *old_context;
1871 int ret;
1872
1873 tran_add(tran, &blockdev_backup_drv, state);
1874
1875 bs = bdrv_lookup_bs(backup->device, backup->device, errp);
1876 if (!bs) {
1877 return;
1878 }
1879
1880 target_bs = bdrv_lookup_bs(backup->target, backup->target, errp);
1881 if (!target_bs) {
1882 return;
1883 }
1884
1885 /* Honor bdrv_try_change_aio_context() context acquisition requirements. */
1886 aio_context = bdrv_get_aio_context(bs);
1887 old_context = bdrv_get_aio_context(target_bs);
1888 aio_context_acquire(old_context);
1889
1890 ret = bdrv_try_change_aio_context(target_bs, aio_context, NULL, errp);
1891 if (ret < 0) {
1892 aio_context_release(old_context);
1893 return;
1894 }
1895
1896 aio_context_release(old_context);
1897 aio_context_acquire(aio_context);
1898 state->bs = bs;
1899
1900 /* Paired with .clean() */
1901 bdrv_drained_begin(state->bs);
1902
1903 state->job = do_backup_common(qapi_BlockdevBackup_base(backup),
1904 bs, target_bs, aio_context,
1905 block_job_txn, errp);
1906
1907 aio_context_release(aio_context);
1908 }
1909
1910 static void blockdev_backup_commit(void *opaque)
1911 {
1912 BlockdevBackupState *state = opaque;
1913 AioContext *aio_context;
1914
1915 aio_context = bdrv_get_aio_context(state->bs);
1916 aio_context_acquire(aio_context);
1917
1918 assert(state->job);
1919 job_start(&state->job->job);
1920
1921 aio_context_release(aio_context);
1922 }
1923
1924 static void blockdev_backup_abort(void *opaque)
1925 {
1926 BlockdevBackupState *state = opaque;
1927
1928 if (state->job) {
1929 job_cancel_sync(&state->job->job, true);
1930 }
1931 }
1932
1933 static void blockdev_backup_clean(void *opaque)
1934 {
1935 g_autofree BlockdevBackupState *state = opaque;
1936 AioContext *aio_context;
1937
1938 if (!state->bs) {
1939 return;
1940 }
1941
1942 aio_context = bdrv_get_aio_context(state->bs);
1943 aio_context_acquire(aio_context);
1944
1945 bdrv_drained_end(state->bs);
1946
1947 aio_context_release(aio_context);
1948 }
1949
1950 typedef struct BlockDirtyBitmapState {
1951 BdrvDirtyBitmap *bitmap;
1952 BlockDriverState *bs;
1953 HBitmap *backup;
1954 bool was_enabled;
1955 } BlockDirtyBitmapState;
1956
1957 static void block_dirty_bitmap_add_abort(void *opaque);
1958 TransactionActionDrv block_dirty_bitmap_add_drv = {
1959 .abort = block_dirty_bitmap_add_abort,
1960 .clean = g_free,
1961 };
1962
1963 static void block_dirty_bitmap_add_action(BlockDirtyBitmapAdd *action,
1964 Transaction *tran, Error **errp)
1965 {
1966 Error *local_err = NULL;
1967 BlockDirtyBitmapState *state = g_new0(BlockDirtyBitmapState, 1);
1968
1969 tran_add(tran, &block_dirty_bitmap_add_drv, state);
1970
1971 /* AIO context taken and released within qmp_block_dirty_bitmap_add */
1972 qmp_block_dirty_bitmap_add(action->node, action->name,
1973 action->has_granularity, action->granularity,
1974 action->has_persistent, action->persistent,
1975 action->has_disabled, action->disabled,
1976 &local_err);
1977
1978 if (!local_err) {
1979 state->bitmap = block_dirty_bitmap_lookup(action->node, action->name,
1980 NULL, &error_abort);
1981 } else {
1982 error_propagate(errp, local_err);
1983 }
1984 }
1985
1986 static void block_dirty_bitmap_add_abort(void *opaque)
1987 {
1988 BlockDirtyBitmapState *state = opaque;
1989
1990 if (state->bitmap) {
1991 bdrv_release_dirty_bitmap(state->bitmap);
1992 }
1993 }
1994
1995 static void block_dirty_bitmap_restore(void *opaque);
1996 static void block_dirty_bitmap_free_backup(void *opaque);
1997 TransactionActionDrv block_dirty_bitmap_clear_drv = {
1998 .abort = block_dirty_bitmap_restore,
1999 .commit = block_dirty_bitmap_free_backup,
2000 .clean = g_free,
2001 };
2002
2003 static void block_dirty_bitmap_clear_action(BlockDirtyBitmap *action,
2004 Transaction *tran, Error **errp)
2005 {
2006 BlockDirtyBitmapState *state = g_new0(BlockDirtyBitmapState, 1);
2007
2008 tran_add(tran, &block_dirty_bitmap_clear_drv, state);
2009
2010 state->bitmap = block_dirty_bitmap_lookup(action->node,
2011 action->name,
2012 &state->bs,
2013 errp);
2014 if (!state->bitmap) {
2015 return;
2016 }
2017
2018 if (bdrv_dirty_bitmap_check(state->bitmap, BDRV_BITMAP_DEFAULT, errp)) {
2019 return;
2020 }
2021
2022 bdrv_clear_dirty_bitmap(state->bitmap, &state->backup);
2023 }
2024
2025 static void block_dirty_bitmap_restore(void *opaque)
2026 {
2027 BlockDirtyBitmapState *state = opaque;
2028
2029 if (state->backup) {
2030 bdrv_restore_dirty_bitmap(state->bitmap, state->backup);
2031 }
2032 }
2033
2034 static void block_dirty_bitmap_free_backup(void *opaque)
2035 {
2036 BlockDirtyBitmapState *state = opaque;
2037
2038 hbitmap_free(state->backup);
2039 }
2040
2041 static void block_dirty_bitmap_enable_abort(void *opaque);
2042 TransactionActionDrv block_dirty_bitmap_enable_drv = {
2043 .abort = block_dirty_bitmap_enable_abort,
2044 .clean = g_free,
2045 };
2046
2047 static void block_dirty_bitmap_enable_action(BlockDirtyBitmap *action,
2048 Transaction *tran, Error **errp)
2049 {
2050 BlockDirtyBitmapState *state = g_new0(BlockDirtyBitmapState, 1);
2051
2052 tran_add(tran, &block_dirty_bitmap_enable_drv, state);
2053
2054 state->bitmap = block_dirty_bitmap_lookup(action->node,
2055 action->name,
2056 NULL,
2057 errp);
2058 if (!state->bitmap) {
2059 return;
2060 }
2061
2062 if (bdrv_dirty_bitmap_check(state->bitmap, BDRV_BITMAP_ALLOW_RO, errp)) {
2063 return;
2064 }
2065
2066 state->was_enabled = bdrv_dirty_bitmap_enabled(state->bitmap);
2067 bdrv_enable_dirty_bitmap(state->bitmap);
2068 }
2069
2070 static void block_dirty_bitmap_enable_abort(void *opaque)
2071 {
2072 BlockDirtyBitmapState *state = opaque;
2073
2074 if (!state->was_enabled) {
2075 bdrv_disable_dirty_bitmap(state->bitmap);
2076 }
2077 }
2078
2079 static void block_dirty_bitmap_disable_abort(void *opaque);
2080 TransactionActionDrv block_dirty_bitmap_disable_drv = {
2081 .abort = block_dirty_bitmap_disable_abort,
2082 .clean = g_free,
2083 };
2084
2085 static void block_dirty_bitmap_disable_action(BlockDirtyBitmap *action,
2086 Transaction *tran, Error **errp)
2087 {
2088 BlockDirtyBitmapState *state = g_new0(BlockDirtyBitmapState, 1);
2089
2090 tran_add(tran, &block_dirty_bitmap_disable_drv, state);
2091
2092 state->bitmap = block_dirty_bitmap_lookup(action->node,
2093 action->name,
2094 NULL,
2095 errp);
2096 if (!state->bitmap) {
2097 return;
2098 }
2099
2100 if (bdrv_dirty_bitmap_check(state->bitmap, BDRV_BITMAP_ALLOW_RO, errp)) {
2101 return;
2102 }
2103
2104 state->was_enabled = bdrv_dirty_bitmap_enabled(state->bitmap);
2105 bdrv_disable_dirty_bitmap(state->bitmap);
2106 }
2107
2108 static void block_dirty_bitmap_disable_abort(void *opaque)
2109 {
2110 BlockDirtyBitmapState *state = opaque;
2111
2112 if (state->was_enabled) {
2113 bdrv_enable_dirty_bitmap(state->bitmap);
2114 }
2115 }
2116
2117 TransactionActionDrv block_dirty_bitmap_merge_drv = {
2118 .commit = block_dirty_bitmap_free_backup,
2119 .abort = block_dirty_bitmap_restore,
2120 .clean = g_free,
2121 };
2122
2123 static void block_dirty_bitmap_merge_action(BlockDirtyBitmapMerge *action,
2124 Transaction *tran, Error **errp)
2125 {
2126 BlockDirtyBitmapState *state = g_new0(BlockDirtyBitmapState, 1);
2127
2128 tran_add(tran, &block_dirty_bitmap_merge_drv, state);
2129
2130 state->bitmap = block_dirty_bitmap_merge(action->node, action->target,
2131 action->bitmaps, &state->backup,
2132 errp);
2133 }
2134
2135 static void block_dirty_bitmap_remove_commit(void *opaque);
2136 static void block_dirty_bitmap_remove_abort(void *opaque);
2137 TransactionActionDrv block_dirty_bitmap_remove_drv = {
2138 .commit = block_dirty_bitmap_remove_commit,
2139 .abort = block_dirty_bitmap_remove_abort,
2140 .clean = g_free,
2141 };
2142
2143 static void block_dirty_bitmap_remove_action(BlockDirtyBitmap *action,
2144 Transaction *tran, Error **errp)
2145 {
2146 BlockDirtyBitmapState *state = g_new0(BlockDirtyBitmapState, 1);
2147
2148 tran_add(tran, &block_dirty_bitmap_remove_drv, state);
2149
2150
2151 state->bitmap = block_dirty_bitmap_remove(action->node, action->name,
2152 false, &state->bs, errp);
2153 if (state->bitmap) {
2154 bdrv_dirty_bitmap_skip_store(state->bitmap, true);
2155 bdrv_dirty_bitmap_set_busy(state->bitmap, true);
2156 }
2157 }
2158
2159 static void block_dirty_bitmap_remove_abort(void *opaque)
2160 {
2161 BlockDirtyBitmapState *state = opaque;
2162
2163 if (state->bitmap) {
2164 bdrv_dirty_bitmap_skip_store(state->bitmap, false);
2165 bdrv_dirty_bitmap_set_busy(state->bitmap, false);
2166 }
2167 }
2168
2169 static void block_dirty_bitmap_remove_commit(void *opaque)
2170 {
2171 BlockDirtyBitmapState *state = opaque;
2172
2173 bdrv_dirty_bitmap_set_busy(state->bitmap, false);
2174 bdrv_release_dirty_bitmap(state->bitmap);
2175 }
2176
2177 static void abort_commit(void *opaque);
2178 TransactionActionDrv abort_drv = {
2179 .commit = abort_commit,
2180 };
2181
2182 static void abort_action(Transaction *tran, Error **errp)
2183 {
2184 tran_add(tran, &abort_drv, NULL);
2185 error_setg(errp, "Transaction aborted using Abort action");
2186 }
2187
2188 static void abort_commit(void *opaque)
2189 {
2190 g_assert_not_reached(); /* this action never succeeds */
2191 }
2192
2193 static void transaction_action(TransactionAction *act, JobTxn *block_job_txn,
2194 Transaction *tran, Error **errp)
2195 {
2196 switch (act->type) {
2197 case TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT:
2198 case TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC:
2199 external_snapshot_action(act, tran, errp);
2200 return;
2201 case TRANSACTION_ACTION_KIND_DRIVE_BACKUP:
2202 drive_backup_action(act->u.drive_backup.data,
2203 block_job_txn, tran, errp);
2204 return;
2205 case TRANSACTION_ACTION_KIND_BLOCKDEV_BACKUP:
2206 blockdev_backup_action(act->u.blockdev_backup.data,
2207 block_job_txn, tran, errp);
2208 return;
2209 case TRANSACTION_ACTION_KIND_ABORT:
2210 abort_action(tran, errp);
2211 return;
2212 case TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC:
2213 internal_snapshot_action(act->u.blockdev_snapshot_internal_sync.data,
2214 tran, errp);
2215 return;
2216 case TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_ADD:
2217 block_dirty_bitmap_add_action(act->u.block_dirty_bitmap_add.data,
2218 tran, errp);
2219 return;
2220 case TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_CLEAR:
2221 block_dirty_bitmap_clear_action(act->u.block_dirty_bitmap_clear.data,
2222 tran, errp);
2223 return;
2224 case TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_ENABLE:
2225 block_dirty_bitmap_enable_action(act->u.block_dirty_bitmap_enable.data,
2226 tran, errp);
2227 return;
2228 case TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_DISABLE:
2229 block_dirty_bitmap_disable_action(
2230 act->u.block_dirty_bitmap_disable.data, tran, errp);
2231 return;
2232 case TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_MERGE:
2233 block_dirty_bitmap_merge_action(act->u.block_dirty_bitmap_merge.data,
2234 tran, errp);
2235 return;
2236 case TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_REMOVE:
2237 block_dirty_bitmap_remove_action(act->u.block_dirty_bitmap_remove.data,
2238 tran, errp);
2239 return;
2240 /*
2241 * Where are transactions for MIRROR, COMMIT and STREAM?
2242 * Although these blockjobs use transaction callbacks like the backup job,
2243 * these jobs do not necessarily adhere to transaction semantics.
2244 * These jobs may not fully undo all of their actions on abort, nor do they
2245 * necessarily work in transactions with more than one job in them.
2246 */
2247 case TRANSACTION_ACTION_KIND__MAX:
2248 default:
2249 g_assert_not_reached();
2250 };
2251 }
2252
2253
2254 /*
2255 * 'Atomic' group operations. The operations are performed as a set, and if
2256 * any fail then we roll back all operations in the group.
2257 *
2258 * Always run under BQL.
2259 */
2260 void qmp_transaction(TransactionActionList *actions,
2261 struct TransactionProperties *properties,
2262 Error **errp)
2263 {
2264 TransactionActionList *act;
2265 JobTxn *block_job_txn = NULL;
2266 Error *local_err = NULL;
2267 Transaction *tran;
2268 ActionCompletionMode comp_mode =
2269 properties ? properties->completion_mode :
2270 ACTION_COMPLETION_MODE_INDIVIDUAL;
2271
2272 GLOBAL_STATE_CODE();
2273
2274 /* Does this transaction get canceled as a group on failure?
2275 * If not, we don't really need to make a JobTxn.
2276 */
2277 if (comp_mode != ACTION_COMPLETION_MODE_INDIVIDUAL) {
2278 for (act = actions; act; act = act->next) {
2279 TransactionActionKind type = act->value->type;
2280
2281 if (type != TRANSACTION_ACTION_KIND_BLOCKDEV_BACKUP &&
2282 type != TRANSACTION_ACTION_KIND_DRIVE_BACKUP)
2283 {
2284 error_setg(errp,
2285 "Action '%s' does not support transaction property "
2286 "completion-mode = %s",
2287 TransactionActionKind_str(type),
2288 ActionCompletionMode_str(comp_mode));
2289 return;
2290 }
2291 }
2292
2293 block_job_txn = job_txn_new();
2294 }
2295
2296 /* drain all i/o before any operations */
2297 bdrv_drain_all();
2298
2299 tran = tran_new();
2300
2301 /* We don't do anything in this loop that commits us to the operations */
2302 for (act = actions; act; act = act->next) {
2303 transaction_action(act->value, block_job_txn, tran, &local_err);
2304 if (local_err) {
2305 error_propagate(errp, local_err);
2306 goto delete_and_fail;
2307 }
2308 }
2309
2310 tran_commit(tran);
2311
2312 /* success */
2313 goto exit;
2314
2315 delete_and_fail:
2316 /* failure, and it is all-or-none; roll back all operations */
2317 tran_abort(tran);
2318 exit:
2319 job_txn_unref(block_job_txn);
2320 }
2321
2322 BlockDirtyBitmapSha256 *qmp_x_debug_block_dirty_bitmap_sha256(const char *node,
2323 const char *name,
2324 Error **errp)
2325 {
2326 BdrvDirtyBitmap *bitmap;
2327 BlockDriverState *bs;
2328 BlockDirtyBitmapSha256 *ret = NULL;
2329 char *sha256;
2330
2331 bitmap = block_dirty_bitmap_lookup(node, name, &bs, errp);
2332 if (!bitmap || !bs) {
2333 return NULL;
2334 }
2335
2336 sha256 = bdrv_dirty_bitmap_sha256(bitmap, errp);
2337 if (sha256 == NULL) {
2338 return NULL;
2339 }
2340
2341 ret = g_new(BlockDirtyBitmapSha256, 1);
2342 ret->sha256 = sha256;
2343
2344 return ret;
2345 }
2346
2347 void coroutine_fn qmp_block_resize(const char *device, const char *node_name,
2348 int64_t size, Error **errp)
2349 {
2350 Error *local_err = NULL;
2351 BlockBackend *blk;
2352 BlockDriverState *bs;
2353 AioContext *old_ctx;
2354
2355 bs = bdrv_lookup_bs(device, node_name, &local_err);
2356 if (local_err) {
2357 error_propagate(errp, local_err);
2358 return;
2359 }
2360
2361 if (size < 0) {
2362 error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "size", "a >0 size");
2363 return;
2364 }
2365
2366 bdrv_graph_co_rdlock();
2367 if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_RESIZE, NULL)) {
2368 error_setg(errp, QERR_DEVICE_IN_USE, device);
2369 bdrv_graph_co_rdunlock();
2370 return;
2371 }
2372 bdrv_graph_co_rdunlock();
2373
2374 blk = blk_co_new_with_bs(bs, BLK_PERM_RESIZE, BLK_PERM_ALL, errp);
2375 if (!blk) {
2376 return;
2377 }
2378
2379 bdrv_co_lock(bs);
2380 bdrv_drained_begin(bs);
2381 bdrv_co_unlock(bs);
2382
2383 old_ctx = bdrv_co_enter(bs);
2384 blk_co_truncate(blk, size, false, PREALLOC_MODE_OFF, 0, errp);
2385 bdrv_co_leave(bs, old_ctx);
2386
2387 bdrv_co_lock(bs);
2388 bdrv_drained_end(bs);
2389 blk_co_unref(blk);
2390 bdrv_co_unlock(bs);
2391 }
2392
2393 void qmp_block_stream(const char *job_id, const char *device,
2394 const char *base,
2395 const char *base_node,
2396 const char *backing_file,
2397 const char *bottom,
2398 bool has_speed, int64_t speed,
2399 bool has_on_error, BlockdevOnError on_error,
2400 const char *filter_node_name,
2401 bool has_auto_finalize, bool auto_finalize,
2402 bool has_auto_dismiss, bool auto_dismiss,
2403 Error **errp)
2404 {
2405 BlockDriverState *bs, *iter, *iter_end;
2406 BlockDriverState *base_bs = NULL;
2407 BlockDriverState *bottom_bs = NULL;
2408 AioContext *aio_context;
2409 Error *local_err = NULL;
2410 int job_flags = JOB_DEFAULT;
2411
2412 GLOBAL_STATE_CODE();
2413
2414 if (base && base_node) {
2415 error_setg(errp, "'base' and 'base-node' cannot be specified "
2416 "at the same time");
2417 return;
2418 }
2419
2420 if (base && bottom) {
2421 error_setg(errp, "'base' and 'bottom' cannot be specified "
2422 "at the same time");
2423 return;
2424 }
2425
2426 if (bottom && base_node) {
2427 error_setg(errp, "'bottom' and 'base-node' cannot be specified "
2428 "at the same time");
2429 return;
2430 }
2431
2432 if (!has_on_error) {
2433 on_error = BLOCKDEV_ON_ERROR_REPORT;
2434 }
2435
2436 bs = bdrv_lookup_bs(device, device, errp);
2437 if (!bs) {
2438 return;
2439 }
2440
2441 aio_context = bdrv_get_aio_context(bs);
2442 aio_context_acquire(aio_context);
2443
2444 if (base) {
2445 base_bs = bdrv_find_backing_image(bs, base);
2446 if (base_bs == NULL) {
2447 error_setg(errp, "Can't find '%s' in the backing chain", base);
2448 goto out;
2449 }
2450 assert(bdrv_get_aio_context(base_bs) == aio_context);
2451 }
2452
2453 if (base_node) {
2454 base_bs = bdrv_lookup_bs(NULL, base_node, errp);
2455 if (!base_bs) {
2456 goto out;
2457 }
2458 if (bs == base_bs || !bdrv_chain_contains(bs, base_bs)) {
2459 error_setg(errp, "Node '%s' is not a backing image of '%s'",
2460 base_node, device);
2461 goto out;
2462 }
2463 assert(bdrv_get_aio_context(base_bs) == aio_context);
2464
2465 bdrv_graph_rdlock_main_loop();
2466 bdrv_refresh_filename(base_bs);
2467 bdrv_graph_rdunlock_main_loop();
2468 }
2469
2470 if (bottom) {
2471 bottom_bs = bdrv_lookup_bs(NULL, bottom, errp);
2472 if (!bottom_bs) {
2473 goto out;
2474 }
2475 if (!bottom_bs->drv) {
2476 error_setg(errp, "Node '%s' is not open", bottom);
2477 goto out;
2478 }
2479 if (bottom_bs->drv->is_filter) {
2480 error_setg(errp, "Node '%s' is a filter, use a non-filter node "
2481 "as 'bottom'", bottom);
2482 goto out;
2483 }
2484 if (!bdrv_chain_contains(bs, bottom_bs)) {
2485 error_setg(errp, "Node '%s' is not in a chain starting from '%s'",
2486 bottom, device);
2487 goto out;
2488 }
2489 assert(bdrv_get_aio_context(bottom_bs) == aio_context);
2490 }
2491
2492 /*
2493 * Check for op blockers in the whole chain between bs and base (or bottom)
2494 */
2495 iter_end = bottom ? bdrv_filter_or_cow_bs(bottom_bs) : base_bs;
2496 bdrv_graph_rdlock_main_loop();
2497 for (iter = bs; iter && iter != iter_end;
2498 iter = bdrv_filter_or_cow_bs(iter))
2499 {
2500 if (bdrv_op_is_blocked(iter, BLOCK_OP_TYPE_STREAM, errp)) {
2501 bdrv_graph_rdunlock_main_loop();
2502 goto out;
2503 }
2504 }
2505 bdrv_graph_rdunlock_main_loop();
2506
2507 /* if we are streaming the entire chain, the result will have no backing
2508 * file, and specifying one is therefore an error */
2509 if (!base_bs && backing_file) {
2510 error_setg(errp, "backing file specified, but streaming the "
2511 "entire chain");
2512 goto out;
2513 }
2514
2515 if (has_auto_finalize && !auto_finalize) {
2516 job_flags |= JOB_MANUAL_FINALIZE;
2517 }
2518 if (has_auto_dismiss && !auto_dismiss) {
2519 job_flags |= JOB_MANUAL_DISMISS;
2520 }
2521
2522 stream_start(job_id, bs, base_bs, backing_file,
2523 bottom_bs, job_flags, has_speed ? speed : 0, on_error,
2524 filter_node_name, &local_err);
2525 if (local_err) {
2526 error_propagate(errp, local_err);
2527 goto out;
2528 }
2529
2530 trace_qmp_block_stream(bs);
2531
2532 out:
2533 aio_context_release(aio_context);
2534 }
2535
2536 void qmp_block_commit(const char *job_id, const char *device,
2537 const char *base_node,
2538 const char *base,
2539 const char *top_node,
2540 const char *top,
2541 const char *backing_file,
2542 bool has_speed, int64_t speed,
2543 bool has_on_error, BlockdevOnError on_error,
2544 const char *filter_node_name,
2545 bool has_auto_finalize, bool auto_finalize,
2546 bool has_auto_dismiss, bool auto_dismiss,
2547 Error **errp)
2548 {
2549 BlockDriverState *bs;
2550 BlockDriverState *iter;
2551 BlockDriverState *base_bs, *top_bs;
2552 AioContext *aio_context;
2553 Error *local_err = NULL;
2554 int job_flags = JOB_DEFAULT;
2555 uint64_t top_perm, top_shared;
2556
2557 /* TODO We'll eventually have to take a writer lock in this function */
2558 GRAPH_RDLOCK_GUARD_MAINLOOP();
2559
2560 if (!has_speed) {
2561 speed = 0;
2562 }
2563 if (!has_on_error) {
2564 on_error = BLOCKDEV_ON_ERROR_REPORT;
2565 }
2566 if (has_auto_finalize && !auto_finalize) {
2567 job_flags |= JOB_MANUAL_FINALIZE;
2568 }
2569 if (has_auto_dismiss && !auto_dismiss) {
2570 job_flags |= JOB_MANUAL_DISMISS;
2571 }
2572
2573 /* Important Note:
2574 * libvirt relies on the DeviceNotFound error class in order to probe for
2575 * live commit feature versions; for this to work, we must make sure to
2576 * perform the device lookup before any generic errors that may occur in a
2577 * scenario in which all optional arguments are omitted. */
2578 bs = qmp_get_root_bs(device, &local_err);
2579 if (!bs) {
2580 bs = bdrv_lookup_bs(device, device, NULL);
2581 if (!bs) {
2582 error_free(local_err);
2583 error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
2584 "Device '%s' not found", device);
2585 } else {
2586 error_propagate(errp, local_err);
2587 }
2588 return;
2589 }
2590
2591 aio_context = bdrv_get_aio_context(bs);
2592 aio_context_acquire(aio_context);
2593
2594 if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_COMMIT_SOURCE, errp)) {
2595 goto out;
2596 }
2597
2598 /* default top_bs is the active layer */
2599 top_bs = bs;
2600
2601 if (top_node && top) {
2602 error_setg(errp, "'top-node' and 'top' are mutually exclusive");
2603 goto out;
2604 } else if (top_node) {
2605 top_bs = bdrv_lookup_bs(NULL, top_node, errp);
2606 if (top_bs == NULL) {
2607 goto out;
2608 }
2609 if (!bdrv_chain_contains(bs, top_bs)) {
2610 error_setg(errp, "'%s' is not in this backing file chain",
2611 top_node);
2612 goto out;
2613 }
2614 } else if (top) {
2615 /* This strcmp() is just a shortcut, there is no need to
2616 * refresh @bs's filename. If it mismatches,
2617 * bdrv_find_backing_image() will do the refresh and may still
2618 * return @bs. */
2619 if (strcmp(bs->filename, top) != 0) {
2620 top_bs = bdrv_find_backing_image(bs, top);
2621 }
2622 }
2623
2624 if (top_bs == NULL) {
2625 error_setg(errp, "Top image file %s not found", top ? top : "NULL");
2626 goto out;
2627 }
2628
2629 assert(bdrv_get_aio_context(top_bs) == aio_context);
2630
2631 if (base_node && base) {
2632 error_setg(errp, "'base-node' and 'base' are mutually exclusive");
2633 goto out;
2634 } else if (base_node) {
2635 base_bs = bdrv_lookup_bs(NULL, base_node, errp);
2636 if (base_bs == NULL) {
2637 goto out;
2638 }
2639 if (!bdrv_chain_contains(top_bs, base_bs)) {
2640 error_setg(errp, "'%s' is not in this backing file chain",
2641 base_node);
2642 goto out;
2643 }
2644 } else if (base) {
2645 base_bs = bdrv_find_backing_image(top_bs, base);
2646 if (base_bs == NULL) {
2647 error_setg(errp, "Can't find '%s' in the backing chain", base);
2648 goto out;
2649 }
2650 } else {
2651 base_bs = bdrv_find_base(top_bs);
2652 if (base_bs == NULL) {
2653 error_setg(errp, "There is no backimg image");
2654 goto out;
2655 }
2656 }
2657
2658 assert(bdrv_get_aio_context(base_bs) == aio_context);
2659
2660 for (iter = top_bs; iter != bdrv_filter_or_cow_bs(base_bs);
2661 iter = bdrv_filter_or_cow_bs(iter))
2662 {
2663 if (bdrv_op_is_blocked(iter, BLOCK_OP_TYPE_COMMIT_TARGET, errp)) {
2664 goto out;
2665 }
2666 }
2667
2668 /* Do not allow attempts to commit an image into itself */
2669 if (top_bs == base_bs) {
2670 error_setg(errp, "cannot commit an image into itself");
2671 goto out;
2672 }
2673
2674 /*
2675 * Active commit is required if and only if someone has taken a
2676 * WRITE permission on the top node. Historically, we have always
2677 * used active commit for top nodes, so continue that practice
2678 * lest we possibly break clients that rely on this behavior, e.g.
2679 * to later attach this node to a writing parent.
2680 * (Active commit is never really wrong.)
2681 */
2682 bdrv_get_cumulative_perm(top_bs, &top_perm, &top_shared);
2683 if (top_perm & BLK_PERM_WRITE ||
2684 bdrv_skip_filters(top_bs) == bdrv_skip_filters(bs))
2685 {
2686 if (backing_file) {
2687 if (bdrv_skip_filters(top_bs) == bdrv_skip_filters(bs)) {
2688 error_setg(errp, "'backing-file' specified,"
2689 " but 'top' is the active layer");
2690 } else {
2691 error_setg(errp, "'backing-file' specified, but 'top' has a "
2692 "writer on it");
2693 }
2694 goto out;
2695 }
2696 if (!job_id) {
2697 /*
2698 * Emulate here what block_job_create() does, because it
2699 * is possible that @bs != @top_bs (the block job should
2700 * be named after @bs, even if @top_bs is the actual
2701 * source)
2702 */
2703 job_id = bdrv_get_device_name(bs);
2704 }
2705 commit_active_start(job_id, top_bs, base_bs, job_flags, speed, on_error,
2706 filter_node_name, NULL, NULL, false, &local_err);
2707 } else {
2708 BlockDriverState *overlay_bs = bdrv_find_overlay(bs, top_bs);
2709 if (bdrv_op_is_blocked(overlay_bs, BLOCK_OP_TYPE_COMMIT_TARGET, errp)) {
2710 goto out;
2711 }
2712 commit_start(job_id, bs, base_bs, top_bs, job_flags,
2713 speed, on_error, backing_file,
2714 filter_node_name, &local_err);
2715 }
2716 if (local_err != NULL) {
2717 error_propagate(errp, local_err);
2718 goto out;
2719 }
2720
2721 out:
2722 aio_context_release(aio_context);
2723 }
2724
2725 /* Common QMP interface for drive-backup and blockdev-backup */
2726 static BlockJob *do_backup_common(BackupCommon *backup,
2727 BlockDriverState *bs,
2728 BlockDriverState *target_bs,
2729 AioContext *aio_context,
2730 JobTxn *txn, Error **errp)
2731 {
2732 BlockJob *job = NULL;
2733 BdrvDirtyBitmap *bmap = NULL;
2734 BackupPerf perf = { .max_workers = 64 };
2735 int job_flags = JOB_DEFAULT;
2736
2737 if (!backup->has_speed) {
2738 backup->speed = 0;
2739 }
2740 if (!backup->has_on_source_error) {
2741 backup->on_source_error = BLOCKDEV_ON_ERROR_REPORT;
2742 }
2743 if (!backup->has_on_target_error) {
2744 backup->on_target_error = BLOCKDEV_ON_ERROR_REPORT;
2745 }
2746 if (!backup->has_auto_finalize) {
2747 backup->auto_finalize = true;
2748 }
2749 if (!backup->has_auto_dismiss) {
2750 backup->auto_dismiss = true;
2751 }
2752 if (!backup->has_compress) {
2753 backup->compress = false;
2754 }
2755
2756 if (backup->x_perf) {
2757 if (backup->x_perf->has_use_copy_range) {
2758 perf.use_copy_range = backup->x_perf->use_copy_range;
2759 }
2760 if (backup->x_perf->has_max_workers) {
2761 perf.max_workers = backup->x_perf->max_workers;
2762 }
2763 if (backup->x_perf->has_max_chunk) {
2764 perf.max_chunk = backup->x_perf->max_chunk;
2765 }
2766 }
2767
2768 if ((backup->sync == MIRROR_SYNC_MODE_BITMAP) ||
2769 (backup->sync == MIRROR_SYNC_MODE_INCREMENTAL)) {
2770 /* done before desugaring 'incremental' to print the right message */
2771 if (!backup->bitmap) {
2772 error_setg(errp, "must provide a valid bitmap name for "
2773 "'%s' sync mode", MirrorSyncMode_str(backup->sync));
2774 return NULL;
2775 }
2776 }
2777
2778 if (backup->sync == MIRROR_SYNC_MODE_INCREMENTAL) {
2779 if (backup->has_bitmap_mode &&
2780 backup->bitmap_mode != BITMAP_SYNC_MODE_ON_SUCCESS) {
2781 error_setg(errp, "Bitmap sync mode must be '%s' "
2782 "when using sync mode '%s'",
2783 BitmapSyncMode_str(BITMAP_SYNC_MODE_ON_SUCCESS),
2784 MirrorSyncMode_str(backup->sync));
2785 return NULL;
2786 }
2787 backup->has_bitmap_mode = true;
2788 backup->sync = MIRROR_SYNC_MODE_BITMAP;
2789 backup->bitmap_mode = BITMAP_SYNC_MODE_ON_SUCCESS;
2790 }
2791
2792 if (backup->bitmap) {
2793 bmap = bdrv_find_dirty_bitmap(bs, backup->bitmap);
2794 if (!bmap) {
2795 error_setg(errp, "Bitmap '%s' could not be found", backup->bitmap);
2796 return NULL;
2797 }
2798 if (!backup->has_bitmap_mode) {
2799 error_setg(errp, "Bitmap sync mode must be given "
2800 "when providing a bitmap");
2801 return NULL;
2802 }
2803 if (bdrv_dirty_bitmap_check(bmap, BDRV_BITMAP_ALLOW_RO, errp)) {
2804 return NULL;
2805 }
2806
2807 /* This does not produce a useful bitmap artifact: */
2808 if (backup->sync == MIRROR_SYNC_MODE_NONE) {
2809 error_setg(errp, "sync mode '%s' does not produce meaningful bitmap"
2810 " outputs", MirrorSyncMode_str(backup->sync));
2811 return NULL;
2812 }
2813
2814 /* If the bitmap isn't used for input or output, this is useless: */
2815 if (backup->bitmap_mode == BITMAP_SYNC_MODE_NEVER &&
2816 backup->sync != MIRROR_SYNC_MODE_BITMAP) {
2817 error_setg(errp, "Bitmap sync mode '%s' has no meaningful effect"
2818 " when combined with sync mode '%s'",
2819 BitmapSyncMode_str(backup->bitmap_mode),
2820 MirrorSyncMode_str(backup->sync));
2821 return NULL;
2822 }
2823 }
2824
2825 if (!backup->bitmap && backup->has_bitmap_mode) {
2826 error_setg(errp, "Cannot specify bitmap sync mode without a bitmap");
2827 return NULL;
2828 }
2829
2830 if (!backup->auto_finalize) {
2831 job_flags |= JOB_MANUAL_FINALIZE;
2832 }
2833 if (!backup->auto_dismiss) {
2834 job_flags |= JOB_MANUAL_DISMISS;
2835 }
2836
2837 job = backup_job_create(backup->job_id, bs, target_bs, backup->speed,
2838 backup->sync, bmap, backup->bitmap_mode,
2839 backup->compress,
2840 backup->filter_node_name,
2841 &perf,
2842 backup->on_source_error,
2843 backup->on_target_error,
2844 job_flags, NULL, NULL, txn, errp);
2845 return job;
2846 }
2847
2848 void qmp_drive_backup(DriveBackup *backup, Error **errp)
2849 {
2850 TransactionAction action = {
2851 .type = TRANSACTION_ACTION_KIND_DRIVE_BACKUP,
2852 .u.drive_backup.data = backup,
2853 };
2854 blockdev_do_action(&action, errp);
2855 }
2856
2857 BlockDeviceInfoList *qmp_query_named_block_nodes(bool has_flat,
2858 bool flat,
2859 Error **errp)
2860 {
2861 bool return_flat = has_flat && flat;
2862
2863 return bdrv_named_nodes_list(return_flat, errp);
2864 }
2865
2866 XDbgBlockGraph *qmp_x_debug_query_block_graph(Error **errp)
2867 {
2868 GRAPH_RDLOCK_GUARD_MAINLOOP();
2869
2870 return bdrv_get_xdbg_block_graph(errp);
2871 }
2872
2873 void qmp_blockdev_backup(BlockdevBackup *backup, Error **errp)
2874 {
2875 TransactionAction action = {
2876 .type = TRANSACTION_ACTION_KIND_BLOCKDEV_BACKUP,
2877 .u.blockdev_backup.data = backup,
2878 };
2879 blockdev_do_action(&action, errp);
2880 }
2881
2882 /* Parameter check and block job starting for drive mirroring.
2883 * Caller should hold @device and @target's aio context (must be the same).
2884 **/
2885 static void blockdev_mirror_common(const char *job_id, BlockDriverState *bs,
2886 BlockDriverState *target,
2887 const char *replaces,
2888 enum MirrorSyncMode sync,
2889 BlockMirrorBackingMode backing_mode,
2890 bool zero_target,
2891 bool has_speed, int64_t speed,
2892 bool has_granularity, uint32_t granularity,
2893 bool has_buf_size, int64_t buf_size,
2894 bool has_on_source_error,
2895 BlockdevOnError on_source_error,
2896 bool has_on_target_error,
2897 BlockdevOnError on_target_error,
2898 bool has_unmap, bool unmap,
2899 const char *filter_node_name,
2900 bool has_copy_mode, MirrorCopyMode copy_mode,
2901 bool has_auto_finalize, bool auto_finalize,
2902 bool has_auto_dismiss, bool auto_dismiss,
2903 Error **errp)
2904 {
2905 BlockDriverState *unfiltered_bs;
2906 int job_flags = JOB_DEFAULT;
2907
2908 GLOBAL_STATE_CODE();
2909 GRAPH_RDLOCK_GUARD_MAINLOOP();
2910
2911 if (!has_speed) {
2912 speed = 0;
2913 }
2914 if (!has_on_source_error) {
2915 on_source_error = BLOCKDEV_ON_ERROR_REPORT;
2916 }
2917 if (!has_on_target_error) {
2918 on_target_error = BLOCKDEV_ON_ERROR_REPORT;
2919 }
2920 if (!has_granularity) {
2921 granularity = 0;
2922 }
2923 if (!has_buf_size) {
2924 buf_size = 0;
2925 }
2926 if (!has_unmap) {
2927 unmap = true;
2928 }
2929 if (!has_copy_mode) {
2930 copy_mode = MIRROR_COPY_MODE_BACKGROUND;
2931 }
2932 if (has_auto_finalize && !auto_finalize) {
2933 job_flags |= JOB_MANUAL_FINALIZE;
2934 }
2935 if (has_auto_dismiss && !auto_dismiss) {
2936 job_flags |= JOB_MANUAL_DISMISS;
2937 }
2938
2939 if (granularity != 0 && (granularity < 512 || granularity > 1048576 * 64)) {
2940 error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "granularity",
2941 "a value in range [512B, 64MB]");
2942 return;
2943 }
2944 if (granularity & (granularity - 1)) {
2945 error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "granularity",
2946 "a power of 2");
2947 return;
2948 }
2949
2950 if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_MIRROR_SOURCE, errp)) {
2951 return;
2952 }
2953 if (bdrv_op_is_blocked(target, BLOCK_OP_TYPE_MIRROR_TARGET, errp)) {
2954 return;
2955 }
2956
2957 if (!bdrv_backing_chain_next(bs) && sync == MIRROR_SYNC_MODE_TOP) {
2958 sync = MIRROR_SYNC_MODE_FULL;
2959 }
2960
2961 if (!replaces) {
2962 /* We want to mirror from @bs, but keep implicit filters on top */
2963 unfiltered_bs = bdrv_skip_implicit_filters(bs);
2964 if (unfiltered_bs != bs) {
2965 replaces = unfiltered_bs->node_name;
2966 }
2967 }
2968
2969 if (replaces) {
2970 BlockDriverState *to_replace_bs;
2971 AioContext *replace_aio_context;
2972 int64_t bs_size, replace_size;
2973
2974 bs_size = bdrv_getlength(bs);
2975 if (bs_size < 0) {
2976 error_setg_errno(errp, -bs_size, "Failed to query device's size");
2977 return;
2978 }
2979
2980 to_replace_bs = check_to_replace_node(bs, replaces, errp);
2981 if (!to_replace_bs) {
2982 return;
2983 }
2984
2985 replace_aio_context = bdrv_get_aio_context(to_replace_bs);
2986 aio_context_acquire(replace_aio_context);
2987 replace_size = bdrv_getlength(to_replace_bs);
2988 aio_context_release(replace_aio_context);
2989
2990 if (replace_size < 0) {
2991 error_setg_errno(errp, -replace_size,
2992 "Failed to query the replacement node's size");
2993 return;
2994 }
2995 if (bs_size != replace_size) {
2996 error_setg(errp, "cannot replace image with a mirror image of "
2997 "different size");
2998 return;
2999 }
3000 }
3001
3002 /* pass the node name to replace to mirror start since it's loose coupling
3003 * and will allow to check whether the node still exist at mirror completion
3004 */
3005 mirror_start(job_id, bs, target,
3006 replaces, job_flags,
3007 speed, granularity, buf_size, sync, backing_mode, zero_target,
3008 on_source_error, on_target_error, unmap, filter_node_name,
3009 copy_mode, errp);
3010 }
3011
3012 void qmp_drive_mirror(DriveMirror *arg, Error **errp)
3013 {
3014 BlockDriverState *bs;
3015 BlockDriverState *target_backing_bs, *target_bs;
3016 AioContext *aio_context;
3017 AioContext *old_context;
3018 BlockMirrorBackingMode backing_mode;
3019 Error *local_err = NULL;
3020 QDict *options = NULL;
3021 int flags;
3022 int64_t size;
3023 const char *format = arg->format;
3024 bool zero_target;
3025 int ret;
3026
3027 bs = qmp_get_root_bs(arg->device, errp);
3028 if (!bs) {
3029 return;
3030 }
3031
3032 /* Early check to avoid creating target */
3033 bdrv_graph_rdlock_main_loop();
3034 if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_MIRROR_SOURCE, errp)) {
3035 bdrv_graph_rdunlock_main_loop();
3036 return;
3037 }
3038 bdrv_graph_rdunlock_main_loop();
3039
3040 aio_context = bdrv_get_aio_context(bs);
3041 aio_context_acquire(aio_context);
3042
3043 if (!arg->has_mode) {
3044 arg->mode = NEW_IMAGE_MODE_ABSOLUTE_PATHS;
3045 }
3046
3047 if (!arg->format) {
3048 format = (arg->mode == NEW_IMAGE_MODE_EXISTING
3049 ? NULL : bs->drv->format_name);
3050 }
3051
3052 flags = bs->open_flags | BDRV_O_RDWR;
3053 target_backing_bs = bdrv_cow_bs(bdrv_skip_filters(bs));
3054 if (!target_backing_bs && arg->sync == MIRROR_SYNC_MODE_TOP) {
3055 arg->sync = MIRROR_SYNC_MODE_FULL;
3056 }
3057 if (arg->sync == MIRROR_SYNC_MODE_NONE) {
3058 target_backing_bs = bs;
3059 }
3060
3061 size = bdrv_getlength(bs);
3062 if (size < 0) {
3063 error_setg_errno(errp, -size, "bdrv_getlength failed");
3064 goto out;
3065 }
3066
3067 if (arg->replaces) {
3068 if (!arg->node_name) {
3069 error_setg(errp, "a node-name must be provided when replacing a"
3070 " named node of the graph");
3071 goto out;
3072 }
3073 }
3074
3075 if (arg->mode == NEW_IMAGE_MODE_ABSOLUTE_PATHS) {
3076 backing_mode = MIRROR_SOURCE_BACKING_CHAIN;
3077 } else {
3078 backing_mode = MIRROR_OPEN_BACKING_CHAIN;
3079 }
3080
3081 /* Don't open backing image in create() */
3082 flags |= BDRV_O_NO_BACKING;
3083
3084 if ((arg->sync == MIRROR_SYNC_MODE_FULL || !target_backing_bs)
3085 && arg->mode != NEW_IMAGE_MODE_EXISTING)
3086 {
3087 /* create new image w/o backing file */
3088 assert(format);
3089 bdrv_img_create(arg->target, format,
3090 NULL, NULL, NULL, size, flags, false, &local_err);
3091 } else {
3092 /* Implicit filters should not appear in the filename */
3093 BlockDriverState *explicit_backing =
3094 bdrv_skip_implicit_filters(target_backing_bs);
3095
3096 switch (arg->mode) {
3097 case NEW_IMAGE_MODE_EXISTING:
3098 break;
3099 case NEW_IMAGE_MODE_ABSOLUTE_PATHS:
3100 /* create new image with backing file */
3101 bdrv_graph_rdlock_main_loop();
3102 bdrv_refresh_filename(explicit_backing);
3103 bdrv_graph_rdunlock_main_loop();
3104
3105 bdrv_img_create(arg->target, format,
3106 explicit_backing->filename,
3107 explicit_backing->drv->format_name,
3108 NULL, size, flags, false, &local_err);
3109 break;
3110 default:
3111 abort();
3112 }
3113 }
3114
3115 if (local_err) {
3116 error_propagate(errp, local_err);
3117 goto out;
3118 }
3119
3120 options = qdict_new();
3121 if (arg->node_name) {
3122 qdict_put_str(options, "node-name", arg->node_name);
3123 }
3124 if (format) {
3125 qdict_put_str(options, "driver", format);
3126 }
3127 aio_context_release(aio_context);
3128
3129 /* Mirroring takes care of copy-on-write using the source's backing
3130 * file.
3131 */
3132 aio_context_acquire(qemu_get_aio_context());
3133 target_bs = bdrv_open(arg->target, NULL, options, flags, errp);
3134 aio_context_release(qemu_get_aio_context());
3135
3136 if (!target_bs) {
3137 return;
3138 }
3139
3140 zero_target = (arg->sync == MIRROR_SYNC_MODE_FULL &&
3141 (arg->mode == NEW_IMAGE_MODE_EXISTING ||
3142 !bdrv_has_zero_init(target_bs)));
3143
3144
3145 /* Honor bdrv_try_change_aio_context() context acquisition requirements. */
3146 old_context = bdrv_get_aio_context(target_bs);
3147 aio_context_acquire(old_context);
3148
3149 ret = bdrv_try_change_aio_context(target_bs, aio_context, NULL, errp);
3150 if (ret < 0) {
3151 bdrv_unref(target_bs);
3152 aio_context_release(old_context);
3153 return;
3154 }
3155
3156 aio_context_release(old_context);
3157 aio_context_acquire(aio_context);
3158
3159 blockdev_mirror_common(arg->job_id, bs, target_bs,
3160 arg->replaces, arg->sync,
3161 backing_mode, zero_target,
3162 arg->has_speed, arg->speed,
3163 arg->has_granularity, arg->granularity,
3164 arg->has_buf_size, arg->buf_size,
3165 arg->has_on_source_error, arg->on_source_error,
3166 arg->has_on_target_error, arg->on_target_error,
3167 arg->has_unmap, arg->unmap,
3168 NULL,
3169 arg->has_copy_mode, arg->copy_mode,
3170 arg->has_auto_finalize, arg->auto_finalize,
3171 arg->has_auto_dismiss, arg->auto_dismiss,
3172 errp);
3173 bdrv_unref(target_bs);
3174 out:
3175 aio_context_release(aio_context);
3176 }
3177
3178 void qmp_blockdev_mirror(const char *job_id,
3179 const char *device, const char *target,
3180 const char *replaces,
3181 MirrorSyncMode sync,
3182 bool has_speed, int64_t speed,
3183 bool has_granularity, uint32_t granularity,
3184 bool has_buf_size, int64_t buf_size,
3185 bool has_on_source_error,
3186 BlockdevOnError on_source_error,
3187 bool has_on_target_error,
3188 BlockdevOnError on_target_error,
3189 const char *filter_node_name,
3190 bool has_copy_mode, MirrorCopyMode copy_mode,
3191 bool has_auto_finalize, bool auto_finalize,
3192 bool has_auto_dismiss, bool auto_dismiss,
3193 Error **errp)
3194 {
3195 BlockDriverState *bs;
3196 BlockDriverState *target_bs;
3197 AioContext *aio_context;
3198 AioContext *old_context;
3199 BlockMirrorBackingMode backing_mode = MIRROR_LEAVE_BACKING_CHAIN;
3200 bool zero_target;
3201 int ret;
3202
3203 bs = qmp_get_root_bs(device, errp);
3204 if (!bs) {
3205 return;
3206 }
3207
3208 target_bs = bdrv_lookup_bs(target, target, errp);
3209 if (!target_bs) {
3210 return;
3211 }
3212
3213 zero_target = (sync == MIRROR_SYNC_MODE_FULL);
3214
3215 /* Honor bdrv_try_change_aio_context() context acquisition requirements. */
3216 old_context = bdrv_get_aio_context(target_bs);
3217 aio_context = bdrv_get_aio_context(bs);
3218 aio_context_acquire(old_context);
3219
3220 ret = bdrv_try_change_aio_context(target_bs, aio_context, NULL, errp);
3221
3222 aio_context_release(old_context);
3223 aio_context_acquire(aio_context);
3224
3225 if (ret < 0) {
3226 goto out;
3227 }
3228
3229 blockdev_mirror_common(job_id, bs, target_bs,
3230 replaces, sync, backing_mode,
3231 zero_target, has_speed, speed,
3232 has_granularity, granularity,
3233 has_buf_size, buf_size,
3234 has_on_source_error, on_source_error,
3235 has_on_target_error, on_target_error,
3236 true, true, filter_node_name,
3237 has_copy_mode, copy_mode,
3238 has_auto_finalize, auto_finalize,
3239 has_auto_dismiss, auto_dismiss,
3240 errp);
3241 out:
3242 aio_context_release(aio_context);
3243 }
3244
3245 /*
3246 * Get a block job using its ID. Called with job_mutex held.
3247 */
3248 static BlockJob *find_block_job_locked(const char *id, Error **errp)
3249 {
3250 BlockJob *job;
3251
3252 assert(id != NULL);
3253
3254 job = block_job_get_locked(id);
3255
3256 if (!job) {
3257 error_set(errp, ERROR_CLASS_DEVICE_NOT_ACTIVE,
3258 "Block job '%s' not found", id);
3259 return NULL;
3260 }
3261
3262 return job;
3263 }
3264
3265 void qmp_block_job_set_speed(const char *device, int64_t speed, Error **errp)
3266 {
3267 BlockJob *job;
3268
3269 JOB_LOCK_GUARD();
3270 job = find_block_job_locked(device, errp);
3271
3272 if (!job) {
3273 return;
3274 }
3275
3276 block_job_set_speed_locked(job, speed, errp);
3277 }
3278
3279 void qmp_block_job_cancel(const char *device,
3280 bool has_force, bool force, Error **errp)
3281 {
3282 BlockJob *job;
3283
3284 JOB_LOCK_GUARD();
3285 job = find_block_job_locked(device, errp);
3286
3287 if (!job) {
3288 return;
3289 }
3290
3291 if (!has_force) {
3292 force = false;
3293 }
3294
3295 if (job_user_paused_locked(&job->job) && !force) {
3296 error_setg(errp, "The block job for device '%s' is currently paused",
3297 device);
3298 return;
3299 }
3300
3301 trace_qmp_block_job_cancel(job);
3302 job_user_cancel_locked(&job->job, force, errp);
3303 }
3304
3305 void qmp_block_job_pause(const char *device, Error **errp)
3306 {
3307 BlockJob *job;
3308
3309 JOB_LOCK_GUARD();
3310 job = find_block_job_locked(device, errp);
3311
3312 if (!job) {
3313 return;
3314 }
3315
3316 trace_qmp_block_job_pause(job);
3317 job_user_pause_locked(&job->job, errp);
3318 }
3319
3320 void qmp_block_job_resume(const char *device, Error **errp)
3321 {
3322 BlockJob *job;
3323
3324 JOB_LOCK_GUARD();
3325 job = find_block_job_locked(device, errp);
3326
3327 if (!job) {
3328 return;
3329 }
3330
3331 trace_qmp_block_job_resume(job);
3332 job_user_resume_locked(&job->job, errp);
3333 }
3334
3335 void qmp_block_job_complete(const char *device, Error **errp)
3336 {
3337 BlockJob *job;
3338
3339 JOB_LOCK_GUARD();
3340 job = find_block_job_locked(device, errp);
3341
3342 if (!job) {
3343 return;
3344 }
3345
3346 trace_qmp_block_job_complete(job);
3347 job_complete_locked(&job->job, errp);
3348 }
3349
3350 void qmp_block_job_finalize(const char *id, Error **errp)
3351 {
3352 BlockJob *job;
3353
3354 JOB_LOCK_GUARD();
3355 job = find_block_job_locked(id, errp);
3356
3357 if (!job) {
3358 return;
3359 }
3360
3361 trace_qmp_block_job_finalize(job);
3362 job_ref_locked(&job->job);
3363 job_finalize_locked(&job->job, errp);
3364
3365 job_unref_locked(&job->job);
3366 }
3367
3368 void qmp_block_job_dismiss(const char *id, Error **errp)
3369 {
3370 BlockJob *bjob;
3371 Job *job;
3372
3373 JOB_LOCK_GUARD();
3374 bjob = find_block_job_locked(id, errp);
3375
3376 if (!bjob) {
3377 return;
3378 }
3379
3380 trace_qmp_block_job_dismiss(bjob);
3381 job = &bjob->job;
3382 job_dismiss_locked(&job, errp);
3383 }
3384
3385 void qmp_change_backing_file(const char *device,
3386 const char *image_node_name,
3387 const char *backing_file,
3388 Error **errp)
3389 {
3390 BlockDriverState *bs = NULL;
3391 AioContext *aio_context;
3392 BlockDriverState *image_bs = NULL;
3393 Error *local_err = NULL;
3394 bool ro;
3395 int ret;
3396
3397 bs = qmp_get_root_bs(device, errp);
3398 if (!bs) {
3399 return;
3400 }
3401
3402 aio_context = bdrv_get_aio_context(bs);
3403 aio_context_acquire(aio_context);
3404
3405 image_bs = bdrv_lookup_bs(NULL, image_node_name, &local_err);
3406 if (local_err) {
3407 error_propagate(errp, local_err);
3408 goto out;
3409 }
3410
3411 if (!image_bs) {
3412 error_setg(errp, "image file not found");
3413 goto out;
3414 }
3415
3416 if (bdrv_find_base(image_bs) == image_bs) {
3417 error_setg(errp, "not allowing backing file change on an image "
3418 "without a backing file");
3419 goto out;
3420 }
3421
3422 /* even though we are not necessarily operating on bs, we need it to
3423 * determine if block ops are currently prohibited on the chain */
3424 bdrv_graph_rdlock_main_loop();
3425 if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_CHANGE, errp)) {
3426 bdrv_graph_rdunlock_main_loop();
3427 goto out;
3428 }
3429 bdrv_graph_rdunlock_main_loop();
3430
3431 /* final sanity check */
3432 if (!bdrv_chain_contains(bs, image_bs)) {
3433 error_setg(errp, "'%s' and image file are not in the same chain",
3434 device);
3435 goto out;
3436 }
3437
3438 /* if not r/w, reopen to make r/w */
3439 ro = bdrv_is_read_only(image_bs);
3440
3441 if (ro) {
3442 if (bdrv_reopen_set_read_only(image_bs, false, errp) != 0) {
3443 goto out;
3444 }
3445 }
3446
3447 ret = bdrv_change_backing_file(image_bs, backing_file,
3448 image_bs->drv ? image_bs->drv->format_name : "",
3449 false);
3450
3451 if (ret < 0) {
3452 error_setg_errno(errp, -ret, "Could not change backing file to '%s'",
3453 backing_file);
3454 /* don't exit here, so we can try to restore open flags if
3455 * appropriate */
3456 }
3457
3458 if (ro) {
3459 bdrv_reopen_set_read_only(image_bs, true, errp);
3460 }
3461
3462 out:
3463 aio_context_release(aio_context);
3464 }
3465
3466 void qmp_blockdev_add(BlockdevOptions *options, Error **errp)
3467 {
3468 BlockDriverState *bs;
3469 QObject *obj;
3470 Visitor *v = qobject_output_visitor_new(&obj);
3471 QDict *qdict;
3472
3473 visit_type_BlockdevOptions(v, NULL, &options, &error_abort);
3474 visit_complete(v, &obj);
3475 qdict = qobject_to(QDict, obj);
3476
3477 qdict_flatten(qdict);
3478
3479 if (!qdict_get_try_str(qdict, "node-name")) {
3480 error_setg(errp, "'node-name' must be specified for the root node");
3481 goto fail;
3482 }
3483
3484 bs = bds_tree_init(qdict, errp);
3485 if (!bs) {
3486 goto fail;
3487 }
3488
3489 bdrv_set_monitor_owned(bs);
3490
3491 fail:
3492 visit_free(v);
3493 }
3494
3495 void qmp_blockdev_reopen(BlockdevOptionsList *reopen_list, Error **errp)
3496 {
3497 BlockReopenQueue *queue = NULL;
3498
3499 /* Add each one of the BDS that we want to reopen to the queue */
3500 for (; reopen_list != NULL; reopen_list = reopen_list->next) {
3501 BlockdevOptions *options = reopen_list->value;
3502 BlockDriverState *bs;
3503 AioContext *ctx;
3504 QObject *obj;
3505 Visitor *v;
3506 QDict *qdict;
3507
3508 /* Check for the selected node name */
3509 if (!options->node_name) {
3510 error_setg(errp, "node-name not specified");
3511 goto fail;
3512 }
3513
3514 bs = bdrv_find_node(options->node_name);
3515 if (!bs) {
3516 error_setg(errp, "Failed to find node with node-name='%s'",
3517 options->node_name);
3518 goto fail;
3519 }
3520
3521 /* Put all options in a QDict and flatten it */
3522 v = qobject_output_visitor_new(&obj);
3523 visit_type_BlockdevOptions(v, NULL, &options, &error_abort);
3524 visit_complete(v, &obj);
3525 visit_free(v);
3526
3527 qdict = qobject_to(QDict, obj);
3528
3529 qdict_flatten(qdict);
3530
3531 ctx = bdrv_get_aio_context(bs);
3532 aio_context_acquire(ctx);
3533
3534 queue = bdrv_reopen_queue(queue, bs, qdict, false);
3535
3536 aio_context_release(ctx);
3537 }
3538
3539 /* Perform the reopen operation */
3540 bdrv_reopen_multiple(queue, errp);
3541 queue = NULL;
3542
3543 fail:
3544 bdrv_reopen_queue_free(queue);
3545 }
3546
3547 void qmp_blockdev_del(const char *node_name, Error **errp)
3548 {
3549 AioContext *aio_context;
3550 BlockDriverState *bs;
3551
3552 GLOBAL_STATE_CODE();
3553 GRAPH_RDLOCK_GUARD_MAINLOOP();
3554
3555 bs = bdrv_find_node(node_name);
3556 if (!bs) {
3557 error_setg(errp, "Failed to find node with node-name='%s'", node_name);
3558 return;
3559 }
3560 if (bdrv_has_blk(bs)) {
3561 error_setg(errp, "Node %s is in use", node_name);
3562 return;
3563 }
3564 aio_context = bdrv_get_aio_context(bs);
3565 aio_context_acquire(aio_context);
3566
3567 if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_DRIVE_DEL, errp)) {
3568 goto out;
3569 }
3570
3571 if (!QTAILQ_IN_USE(bs, monitor_list)) {
3572 error_setg(errp, "Node %s is not owned by the monitor",
3573 bs->node_name);
3574 goto out;
3575 }
3576
3577 if (bs->refcnt > 1) {
3578 error_setg(errp, "Block device %s is in use",
3579 bdrv_get_device_or_node_name(bs));
3580 goto out;
3581 }
3582
3583 QTAILQ_REMOVE(&monitor_bdrv_states, bs, monitor_list);
3584 bdrv_unref(bs);
3585
3586 out:
3587 aio_context_release(aio_context);
3588 }
3589
3590 static BdrvChild * GRAPH_RDLOCK
3591 bdrv_find_child(BlockDriverState *parent_bs, const char *child_name)
3592 {
3593 BdrvChild *child;
3594
3595 QLIST_FOREACH(child, &parent_bs->children, next) {
3596 if (strcmp(child->name, child_name) == 0) {
3597 return child;
3598 }
3599 }
3600
3601 return NULL;
3602 }
3603
3604 void qmp_x_blockdev_change(const char *parent, const char *child,
3605 const char *node, Error **errp)
3606 {
3607 BlockDriverState *parent_bs, *new_bs = NULL;
3608 BdrvChild *p_child;
3609
3610 bdrv_graph_wrlock(NULL);
3611
3612 parent_bs = bdrv_lookup_bs(parent, parent, errp);
3613 if (!parent_bs) {
3614 goto out;
3615 }
3616
3617 if (!child == !node) {
3618 if (child) {
3619 error_setg(errp, "The parameters child and node are in conflict");
3620 } else {
3621 error_setg(errp, "Either child or node must be specified");
3622 }
3623 goto out;
3624 }
3625
3626 if (child) {
3627 p_child = bdrv_find_child(parent_bs, child);
3628 if (!p_child) {
3629 error_setg(errp, "Node '%s' does not have child '%s'",
3630 parent, child);
3631 goto out;
3632 }
3633 bdrv_del_child(parent_bs, p_child, errp);
3634 }
3635
3636 if (node) {
3637 new_bs = bdrv_find_node(node);
3638 if (!new_bs) {
3639 error_setg(errp, "Node '%s' not found", node);
3640 goto out;
3641 }
3642 bdrv_add_child(parent_bs, new_bs, errp);
3643 }
3644
3645 out:
3646 bdrv_graph_wrunlock();
3647 }
3648
3649 BlockJobInfoList *qmp_query_block_jobs(Error **errp)
3650 {
3651 BlockJobInfoList *head = NULL, **tail = &head;
3652 BlockJob *job;
3653
3654 JOB_LOCK_GUARD();
3655
3656 for (job = block_job_next_locked(NULL); job;
3657 job = block_job_next_locked(job)) {
3658 BlockJobInfo *value;
3659
3660 if (block_job_is_internal(job)) {
3661 continue;
3662 }
3663 value = block_job_query_locked(job, errp);
3664 if (!value) {
3665 qapi_free_BlockJobInfoList(head);
3666 return NULL;
3667 }
3668 QAPI_LIST_APPEND(tail, value);
3669 }
3670
3671 return head;
3672 }
3673
3674 void qmp_x_blockdev_set_iothread(const char *node_name, StrOrNull *iothread,
3675 bool has_force, bool force, Error **errp)
3676 {
3677 AioContext *old_context;
3678 AioContext *new_context;
3679 BlockDriverState *bs;
3680
3681 GRAPH_RDLOCK_GUARD_MAINLOOP();
3682
3683 bs = bdrv_find_node(node_name);
3684 if (!bs) {
3685 error_setg(errp, "Failed to find node with node-name='%s'", node_name);
3686 return;
3687 }
3688
3689 /* Protects against accidents. */
3690 if (!(has_force && force) && bdrv_has_blk(bs)) {
3691 error_setg(errp, "Node %s is associated with a BlockBackend and could "
3692 "be in use (use force=true to override this check)",
3693 node_name);
3694 return;
3695 }
3696
3697 if (iothread->type == QTYPE_QSTRING) {
3698 IOThread *obj = iothread_by_id(iothread->u.s);
3699 if (!obj) {
3700 error_setg(errp, "Cannot find iothread %s", iothread->u.s);
3701 return;
3702 }
3703
3704 new_context = iothread_get_aio_context(obj);
3705 } else {
3706 new_context = qemu_get_aio_context();
3707 }
3708
3709 old_context = bdrv_get_aio_context(bs);
3710 aio_context_acquire(old_context);
3711
3712 bdrv_try_change_aio_context(bs, new_context, NULL, errp);
3713
3714 aio_context_release(old_context);
3715 }
3716
3717 QemuOptsList qemu_common_drive_opts = {
3718 .name = "drive",
3719 .head = QTAILQ_HEAD_INITIALIZER(qemu_common_drive_opts.head),
3720 .desc = {
3721 {
3722 .name = "snapshot",
3723 .type = QEMU_OPT_BOOL,
3724 .help = "enable/disable snapshot mode",
3725 },{
3726 .name = "aio",
3727 .type = QEMU_OPT_STRING,
3728 .help = "host AIO implementation (threads, native, io_uring)",
3729 },{
3730 .name = BDRV_OPT_CACHE_WB,
3731 .type = QEMU_OPT_BOOL,
3732 .help = "Enable writeback mode",
3733 },{
3734 .name = "format",
3735 .type = QEMU_OPT_STRING,
3736 .help = "disk format (raw, qcow2, ...)",
3737 },{
3738 .name = "rerror",
3739 .type = QEMU_OPT_STRING,
3740 .help = "read error action",
3741 },{
3742 .name = "werror",
3743 .type = QEMU_OPT_STRING,
3744 .help = "write error action",
3745 },{
3746 .name = BDRV_OPT_READ_ONLY,
3747 .type = QEMU_OPT_BOOL,
3748 .help = "open drive file as read-only",
3749 },
3750
3751 THROTTLE_OPTS,
3752
3753 {
3754 .name = "throttling.group",
3755 .type = QEMU_OPT_STRING,
3756 .help = "name of the block throttling group",
3757 },{
3758 .name = "copy-on-read",
3759 .type = QEMU_OPT_BOOL,
3760 .help = "copy read data from backing file into image file",
3761 },{
3762 .name = "detect-zeroes",
3763 .type = QEMU_OPT_STRING,
3764 .help = "try to optimize zero writes (off, on, unmap)",
3765 },{
3766 .name = "stats-account-invalid",
3767 .type = QEMU_OPT_BOOL,
3768 .help = "whether to account for invalid I/O operations "
3769 "in the statistics",
3770 },{
3771 .name = "stats-account-failed",
3772 .type = QEMU_OPT_BOOL,
3773 .help = "whether to account for failed I/O operations "
3774 "in the statistics",
3775 },
3776 { /* end of list */ }
3777 },
3778 };
3779
3780 QemuOptsList qemu_drive_opts = {
3781 .name = "drive",
3782 .head = QTAILQ_HEAD_INITIALIZER(qemu_drive_opts.head),
3783 .desc = {
3784 /*
3785 * no elements => accept any params
3786 * validation will happen later
3787 */
3788 { /* end of list */ }
3789 },
3790 };
armbru's QEMU hackery