?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
i386: kvm: Add support for MSR filtering
[qemu.git] / blockdev.c
blob392d9476e6acd9c5950b72dd6e696879a9c7e2f5
1 /*
2 * QEMU host block devices
3 *
4 * Copyright (c) 2003-2008 Fabrice Bellard
5 *
6 * This work is licensed under the terms of the GNU GPL, version 2 or
7 * later. See the COPYING file in the top-level directory.
8 *
9 * This file incorporates work covered by the following copyright and
10 * permission notice:
11 *
12 * Copyright (c) 2003-2008 Fabrice Bellard
13 *
14 * Permission is hereby granted, free of charge, to any person obtaining a copy
15 * of this software and associated documentation files (the "Software"), to deal
16 * in the Software without restriction, including without limitation the rights
17 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
18 * copies of the Software, and to permit persons to whom the Software is
19 * furnished to do so, subject to the following conditions:
20 *
21 * The above copyright notice and this permission notice shall be included in
22 * all copies or substantial portions of the Software.
23 *
24 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
25 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
26 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
27 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
28 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
29 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
30 * THE SOFTWARE.
31 */
32
33 #include "qemu/osdep.h"
34 #include "sysemu/block-backend.h"
35 #include "sysemu/blockdev.h"
36 #include "hw/block/block.h"
37 #include "block/blockjob.h"
38 #include "block/qdict.h"
39 #include "block/throttle-groups.h"
40 #include "monitor/monitor.h"
41 #include "qemu/error-report.h"
42 #include "qemu/option.h"
43 #include "qemu/qemu-print.h"
44 #include "qemu/config-file.h"
45 #include "qapi/qapi-commands-block.h"
46 #include "qapi/qapi-commands-transaction.h"
47 #include "qapi/qapi-visit-block-core.h"
48 #include "qapi/qmp/qdict.h"
49 #include "qapi/qmp/qnum.h"
50 #include "qapi/qmp/qstring.h"
51 #include "qapi/error.h"
52 #include "qapi/qmp/qerror.h"
53 #include "qapi/qmp/qlist.h"
54 #include "qapi/qobject-output-visitor.h"
55 #include "sysemu/sysemu.h"
56 #include "sysemu/iothread.h"
57 #include "block/block_int.h"
58 #include "block/trace.h"
59 #include "sysemu/runstate.h"
60 #include "sysemu/replay.h"
61 #include "qemu/cutils.h"
62 #include "qemu/help_option.h"
63 #include "qemu/main-loop.h"
64 #include "qemu/throttle-options.h"
65
66 /* Protected by BQL */
67 QTAILQ_HEAD(, BlockDriverState) monitor_bdrv_states =
68 QTAILQ_HEAD_INITIALIZER(monitor_bdrv_states);
69
70 void bdrv_set_monitor_owned(BlockDriverState *bs)
71 {
72 GLOBAL_STATE_CODE();
73 QTAILQ_INSERT_TAIL(&monitor_bdrv_states, bs, monitor_list);
74 }
75
76 static const char *const if_name[IF_COUNT] = {
77 [IF_NONE] = "none",
78 [IF_IDE] = "ide",
79 [IF_SCSI] = "scsi",
80 [IF_FLOPPY] = "floppy",
81 [IF_PFLASH] = "pflash",
82 [IF_MTD] = "mtd",
83 [IF_SD] = "sd",
84 [IF_VIRTIO] = "virtio",
85 [IF_XEN] = "xen",
86 };
87
88 static int if_max_devs[IF_COUNT] = {
89 /*
90 * Do not change these numbers! They govern how drive option
91 * index maps to unit and bus. That mapping is ABI.
92 *
93 * All controllers used to implement if=T drives need to support
94 * if_max_devs[T] units, for any T with if_max_devs[T] != 0.
95 * Otherwise, some index values map to "impossible" bus, unit
96 * values.
97 *
98 * For instance, if you change [IF_SCSI] to 255, -drive
99 * if=scsi,index=12 no longer means bus=1,unit=5, but
100 * bus=0,unit=12. With an lsi53c895a controller (7 units max),
101 * the drive can't be set up. Regression.
102 */
103 [IF_IDE] = 2,
104 [IF_SCSI] = 7,
105 };
106
107 /**
108 * Boards may call this to offer board-by-board overrides
109 * of the default, global values.
110 */
111 void override_max_devs(BlockInterfaceType type, int max_devs)
112 {
113 BlockBackend *blk;
114 DriveInfo *dinfo;
115
116 GLOBAL_STATE_CODE();
117
118 if (max_devs <= 0) {
119 return;
120 }
121
122 for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
123 dinfo = blk_legacy_dinfo(blk);
124 if (dinfo->type == type) {
125 fprintf(stderr, "Cannot override units-per-bus property of"
126 " the %s interface, because a drive of that type has"
127 " already been added.\n", if_name[type]);
128 g_assert_not_reached();
129 }
130 }
131
132 if_max_devs[type] = max_devs;
133 }
134
135 /*
136 * We automatically delete the drive when a device using it gets
137 * unplugged. Questionable feature, but we can't just drop it.
138 * Device models call blockdev_mark_auto_del() to schedule the
139 * automatic deletion, and generic qdev code calls blockdev_auto_del()
140 * when deletion is actually safe.
141 */
142 void blockdev_mark_auto_del(BlockBackend *blk)
143 {
144 DriveInfo *dinfo = blk_legacy_dinfo(blk);
145 BlockJob *job;
146
147 GLOBAL_STATE_CODE();
148
149 if (!dinfo) {
150 return;
151 }
152
153 for (job = block_job_next(NULL); job; job = block_job_next(job)) {
154 if (block_job_has_bdrv(job, blk_bs(blk))) {
155 AioContext *aio_context = job->job.aio_context;
156 aio_context_acquire(aio_context);
157
158 job_cancel(&job->job, false);
159
160 aio_context_release(aio_context);
161 }
162 }
163
164 dinfo->auto_del = 1;
165 }
166
167 void blockdev_auto_del(BlockBackend *blk)
168 {
169 DriveInfo *dinfo = blk_legacy_dinfo(blk);
170 GLOBAL_STATE_CODE();
171
172 if (dinfo && dinfo->auto_del) {
173 monitor_remove_blk(blk);
174 blk_unref(blk);
175 }
176 }
177
178 static int drive_index_to_bus_id(BlockInterfaceType type, int index)
179 {
180 int max_devs = if_max_devs[type];
181 return max_devs ? index / max_devs : 0;
182 }
183
184 static int drive_index_to_unit_id(BlockInterfaceType type, int index)
185 {
186 int max_devs = if_max_devs[type];
187 return max_devs ? index % max_devs : index;
188 }
189
190 QemuOpts *drive_add(BlockInterfaceType type, int index, const char *file,
191 const char *optstr)
192 {
193 QemuOpts *opts;
194
195 GLOBAL_STATE_CODE();
196
197 opts = qemu_opts_parse_noisily(qemu_find_opts("drive"), optstr, false);
198 if (!opts) {
199 return NULL;
200 }
201 if (type != IF_DEFAULT) {
202 qemu_opt_set(opts, "if", if_name[type], &error_abort);
203 }
204 if (index >= 0) {
205 qemu_opt_set_number(opts, "index", index, &error_abort);
206 }
207 if (file)
208 qemu_opt_set(opts, "file", file, &error_abort);
209 return opts;
210 }
211
212 DriveInfo *drive_get(BlockInterfaceType type, int bus, int unit)
213 {
214 BlockBackend *blk;
215 DriveInfo *dinfo;
216
217 GLOBAL_STATE_CODE();
218
219 for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
220 dinfo = blk_legacy_dinfo(blk);
221 if (dinfo && dinfo->type == type
222 && dinfo->bus == bus && dinfo->unit == unit) {
223 return dinfo;
224 }
225 }
226
227 return NULL;
228 }
229
230 /*
231 * Check board claimed all -drive that are meant to be claimed.
232 * Fatal error if any remain unclaimed.
233 */
234 void drive_check_orphaned(void)
235 {
236 BlockBackend *blk;
237 DriveInfo *dinfo;
238 Location loc;
239 bool orphans = false;
240
241 GLOBAL_STATE_CODE();
242
243 for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
244 dinfo = blk_legacy_dinfo(blk);
245 /*
246 * Ignore default drives, because we create certain default
247 * drives unconditionally, then leave them unclaimed. Not the
248 * users fault.
249 * Ignore IF_VIRTIO, because it gets desugared into -device,
250 * so we can leave failing to -device.
251 * Ignore IF_NONE, because leaving unclaimed IF_NONE remains
252 * available for device_add is a feature.
253 */
254 if (dinfo->is_default || dinfo->type == IF_VIRTIO
255 || dinfo->type == IF_NONE) {
256 continue;
257 }
258 if (!blk_get_attached_dev(blk)) {
259 loc_push_none(&loc);
260 qemu_opts_loc_restore(dinfo->opts);
261 error_report("machine type does not support"
262 " if=%s,bus=%d,unit=%d",
263 if_name[dinfo->type], dinfo->bus, dinfo->unit);
264 loc_pop(&loc);
265 orphans = true;
266 }
267 }
268
269 if (orphans) {
270 exit(1);
271 }
272 }
273
274 DriveInfo *drive_get_by_index(BlockInterfaceType type, int index)
275 {
276 GLOBAL_STATE_CODE();
277 return drive_get(type,
278 drive_index_to_bus_id(type, index),
279 drive_index_to_unit_id(type, index));
280 }
281
282 int drive_get_max_bus(BlockInterfaceType type)
283 {
284 int max_bus;
285 BlockBackend *blk;
286 DriveInfo *dinfo;
287
288 GLOBAL_STATE_CODE();
289
290 max_bus = -1;
291 for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
292 dinfo = blk_legacy_dinfo(blk);
293 if (dinfo && dinfo->type == type && dinfo->bus > max_bus) {
294 max_bus = dinfo->bus;
295 }
296 }
297 return max_bus;
298 }
299
300 static void bdrv_format_print(void *opaque, const char *name)
301 {
302 qemu_printf(" %s", name);
303 }
304
305 typedef struct {
306 QEMUBH *bh;
307 BlockDriverState *bs;
308 } BDRVPutRefBH;
309
310 static int parse_block_error_action(const char *buf, bool is_read, Error **errp)
311 {
312 if (!strcmp(buf, "ignore")) {
313 return BLOCKDEV_ON_ERROR_IGNORE;
314 } else if (!is_read && !strcmp(buf, "enospc")) {
315 return BLOCKDEV_ON_ERROR_ENOSPC;
316 } else if (!strcmp(buf, "stop")) {
317 return BLOCKDEV_ON_ERROR_STOP;
318 } else if (!strcmp(buf, "report")) {
319 return BLOCKDEV_ON_ERROR_REPORT;
320 } else {
321 error_setg(errp, "'%s' invalid %s error action",
322 buf, is_read ? "read" : "write");
323 return -1;
324 }
325 }
326
327 static bool parse_stats_intervals(BlockAcctStats *stats, QList *intervals,
328 Error **errp)
329 {
330 const QListEntry *entry;
331 for (entry = qlist_first(intervals); entry; entry = qlist_next(entry)) {
332 switch (qobject_type(entry->value)) {
333
334 case QTYPE_QSTRING: {
335 unsigned long long length;
336 const char *str = qstring_get_str(qobject_to(QString,
337 entry->value));
338 if (parse_uint_full(str, &length, 10) == 0 &&
339 length > 0 && length <= UINT_MAX) {
340 block_acct_add_interval(stats, (unsigned) length);
341 } else {
342 error_setg(errp, "Invalid interval length: %s", str);
343 return false;
344 }
345 break;
346 }
347
348 case QTYPE_QNUM: {
349 int64_t length = qnum_get_int(qobject_to(QNum, entry->value));
350
351 if (length > 0 && length <= UINT_MAX) {
352 block_acct_add_interval(stats, (unsigned) length);
353 } else {
354 error_setg(errp, "Invalid interval length: %" PRId64, length);
355 return false;
356 }
357 break;
358 }
359
360 default:
361 error_setg(errp, "The specification of stats-intervals is invalid");
362 return false;
363 }
364 }
365 return true;
366 }
367
368 typedef enum { MEDIA_DISK, MEDIA_CDROM } DriveMediaType;
369
370 /* All parameters but @opts are optional and may be set to NULL. */
371 static void extract_common_blockdev_options(QemuOpts *opts, int *bdrv_flags,
372 const char **throttling_group, ThrottleConfig *throttle_cfg,
373 BlockdevDetectZeroesOptions *detect_zeroes, Error **errp)
374 {
375 Error *local_error = NULL;
376 const char *aio;
377
378 if (bdrv_flags) {
379 if (qemu_opt_get_bool(opts, "copy-on-read", false)) {
380 *bdrv_flags |= BDRV_O_COPY_ON_READ;
381 }
382
383 if ((aio = qemu_opt_get(opts, "aio")) != NULL) {
384 if (bdrv_parse_aio(aio, bdrv_flags) < 0) {
385 error_setg(errp, "invalid aio option");
386 return;
387 }
388 }
389 }
390
391 /* disk I/O throttling */
392 if (throttling_group) {
393 *throttling_group = qemu_opt_get(opts, "throttling.group");
394 }
395
396 if (throttle_cfg) {
397 throttle_config_init(throttle_cfg);
398 throttle_cfg->buckets[THROTTLE_BPS_TOTAL].avg =
399 qemu_opt_get_number(opts, "throttling.bps-total", 0);
400 throttle_cfg->buckets[THROTTLE_BPS_READ].avg =
401 qemu_opt_get_number(opts, "throttling.bps-read", 0);
402 throttle_cfg->buckets[THROTTLE_BPS_WRITE].avg =
403 qemu_opt_get_number(opts, "throttling.bps-write", 0);
404 throttle_cfg->buckets[THROTTLE_OPS_TOTAL].avg =
405 qemu_opt_get_number(opts, "throttling.iops-total", 0);
406 throttle_cfg->buckets[THROTTLE_OPS_READ].avg =
407 qemu_opt_get_number(opts, "throttling.iops-read", 0);
408 throttle_cfg->buckets[THROTTLE_OPS_WRITE].avg =
409 qemu_opt_get_number(opts, "throttling.iops-write", 0);
410
411 throttle_cfg->buckets[THROTTLE_BPS_TOTAL].max =
412 qemu_opt_get_number(opts, "throttling.bps-total-max", 0);
413 throttle_cfg->buckets[THROTTLE_BPS_READ].max =
414 qemu_opt_get_number(opts, "throttling.bps-read-max", 0);
415 throttle_cfg->buckets[THROTTLE_BPS_WRITE].max =
416 qemu_opt_get_number(opts, "throttling.bps-write-max", 0);
417 throttle_cfg->buckets[THROTTLE_OPS_TOTAL].max =
418 qemu_opt_get_number(opts, "throttling.iops-total-max", 0);
419 throttle_cfg->buckets[THROTTLE_OPS_READ].max =
420 qemu_opt_get_number(opts, "throttling.iops-read-max", 0);
421 throttle_cfg->buckets[THROTTLE_OPS_WRITE].max =
422 qemu_opt_get_number(opts, "throttling.iops-write-max", 0);
423
424 throttle_cfg->buckets[THROTTLE_BPS_TOTAL].burst_length =
425 qemu_opt_get_number(opts, "throttling.bps-total-max-length", 1);
426 throttle_cfg->buckets[THROTTLE_BPS_READ].burst_length =
427 qemu_opt_get_number(opts, "throttling.bps-read-max-length", 1);
428 throttle_cfg->buckets[THROTTLE_BPS_WRITE].burst_length =
429 qemu_opt_get_number(opts, "throttling.bps-write-max-length", 1);
430 throttle_cfg->buckets[THROTTLE_OPS_TOTAL].burst_length =
431 qemu_opt_get_number(opts, "throttling.iops-total-max-length", 1);
432 throttle_cfg->buckets[THROTTLE_OPS_READ].burst_length =
433 qemu_opt_get_number(opts, "throttling.iops-read-max-length", 1);
434 throttle_cfg->buckets[THROTTLE_OPS_WRITE].burst_length =
435 qemu_opt_get_number(opts, "throttling.iops-write-max-length", 1);
436
437 throttle_cfg->op_size =
438 qemu_opt_get_number(opts, "throttling.iops-size", 0);
439
440 if (!throttle_is_valid(throttle_cfg, errp)) {
441 return;
442 }
443 }
444
445 if (detect_zeroes) {
446 *detect_zeroes =
447 qapi_enum_parse(&BlockdevDetectZeroesOptions_lookup,
448 qemu_opt_get(opts, "detect-zeroes"),
449 BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF,
450 &local_error);
451 if (local_error) {
452 error_propagate(errp, local_error);
453 return;
454 }
455 }
456 }
457
458 static OnOffAuto account_get_opt(QemuOpts *opts, const char *name)
459 {
460 if (!qemu_opt_find(opts, name)) {
461 return ON_OFF_AUTO_AUTO;
462 }
463 if (qemu_opt_get_bool(opts, name, true)) {
464 return ON_OFF_AUTO_ON;
465 }
466 return ON_OFF_AUTO_OFF;
467 }
468
469 /* Takes the ownership of bs_opts */
470 static BlockBackend *blockdev_init(const char *file, QDict *bs_opts,
471 Error **errp)
472 {
473 const char *buf;
474 int bdrv_flags = 0;
475 int on_read_error, on_write_error;
476 OnOffAuto account_invalid, account_failed;
477 bool writethrough, read_only;
478 BlockBackend *blk;
479 BlockDriverState *bs;
480 ThrottleConfig cfg;
481 int snapshot = 0;
482 Error *error = NULL;
483 QemuOpts *opts;
484 QDict *interval_dict = NULL;
485 QList *interval_list = NULL;
486 const char *id;
487 BlockdevDetectZeroesOptions detect_zeroes =
488 BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF;
489 const char *throttling_group = NULL;
490
491 /* Check common options by copying from bs_opts to opts, all other options
492 * stay in bs_opts for processing by bdrv_open(). */
493 id = qdict_get_try_str(bs_opts, "id");
494 opts = qemu_opts_create(&qemu_common_drive_opts, id, 1, errp);
495 if (!opts) {
496 goto err_no_opts;
497 }
498
499 if (!qemu_opts_absorb_qdict(opts, bs_opts, errp)) {
500 goto early_err;
501 }
502
503 if (id) {
504 qdict_del(bs_opts, "id");
505 }
506
507 /* extract parameters */
508 snapshot = qemu_opt_get_bool(opts, "snapshot", 0);
509
510 account_invalid = account_get_opt(opts, "stats-account-invalid");
511 account_failed = account_get_opt(opts, "stats-account-failed");
512
513 writethrough = !qemu_opt_get_bool(opts, BDRV_OPT_CACHE_WB, true);
514
515 id = qemu_opts_id(opts);
516
517 qdict_extract_subqdict(bs_opts, &interval_dict, "stats-intervals.");
518 qdict_array_split(interval_dict, &interval_list);
519
520 if (qdict_size(interval_dict) != 0) {
521 error_setg(errp, "Invalid option stats-intervals.%s",
522 qdict_first(interval_dict)->key);
523 goto early_err;
524 }
525
526 extract_common_blockdev_options(opts, &bdrv_flags, &throttling_group, &cfg,
527 &detect_zeroes, &error);
528 if (error) {
529 error_propagate(errp, error);
530 goto early_err;
531 }
532
533 if ((buf = qemu_opt_get(opts, "format")) != NULL) {
534 if (is_help_option(buf)) {
535 qemu_printf("Supported formats:");
536 bdrv_iterate_format(bdrv_format_print, NULL, false);
537 qemu_printf("\nSupported formats (read-only):");
538 bdrv_iterate_format(bdrv_format_print, NULL, true);
539 qemu_printf("\n");
540 goto early_err;
541 }
542
543 if (qdict_haskey(bs_opts, "driver")) {
544 error_setg(errp, "Cannot specify both 'driver' and 'format'");
545 goto early_err;
546 }
547 qdict_put_str(bs_opts, "driver", buf);
548 }
549
550 on_write_error = BLOCKDEV_ON_ERROR_ENOSPC;
551 if ((buf = qemu_opt_get(opts, "werror")) != NULL) {
552 on_write_error = parse_block_error_action(buf, 0, &error);
553 if (error) {
554 error_propagate(errp, error);
555 goto early_err;
556 }
557 }
558
559 on_read_error = BLOCKDEV_ON_ERROR_REPORT;
560 if ((buf = qemu_opt_get(opts, "rerror")) != NULL) {
561 on_read_error = parse_block_error_action(buf, 1, &error);
562 if (error) {
563 error_propagate(errp, error);
564 goto early_err;
565 }
566 }
567
568 if (snapshot) {
569 bdrv_flags |= BDRV_O_SNAPSHOT;
570 }
571
572 read_only = qemu_opt_get_bool(opts, BDRV_OPT_READ_ONLY, false);
573
574 /* init */
575 if ((!file || !*file) && !qdict_size(bs_opts)) {
576 BlockBackendRootState *blk_rs;
577
578 blk = blk_new(qemu_get_aio_context(), 0, BLK_PERM_ALL);
579 blk_rs = blk_get_root_state(blk);
580 blk_rs->open_flags = bdrv_flags | (read_only ? 0 : BDRV_O_RDWR);
581 blk_rs->detect_zeroes = detect_zeroes;
582
583 qobject_unref(bs_opts);
584 } else {
585 if (file && !*file) {
586 file = NULL;
587 }
588
589 /* bdrv_open() defaults to the values in bdrv_flags (for compatibility
590 * with other callers) rather than what we want as the real defaults.
591 * Apply the defaults here instead. */
592 qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_DIRECT, "off");
593 qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_NO_FLUSH, "off");
594 qdict_set_default_str(bs_opts, BDRV_OPT_READ_ONLY,
595 read_only ? "on" : "off");
596 qdict_set_default_str(bs_opts, BDRV_OPT_AUTO_READ_ONLY, "on");
597 assert((bdrv_flags & BDRV_O_CACHE_MASK) == 0);
598
599 if (runstate_check(RUN_STATE_INMIGRATE)) {
600 bdrv_flags |= BDRV_O_INACTIVE;
601 }
602
603 blk = blk_new_open(file, NULL, bs_opts, bdrv_flags, errp);
604 if (!blk) {
605 goto err_no_bs_opts;
606 }
607 bs = blk_bs(blk);
608
609 bs->detect_zeroes = detect_zeroes;
610
611 block_acct_setup(blk_get_stats(blk), account_invalid, account_failed);
612
613 if (!parse_stats_intervals(blk_get_stats(blk), interval_list, errp)) {
614 blk_unref(blk);
615 blk = NULL;
616 goto err_no_bs_opts;
617 }
618 }
619
620 /* disk I/O throttling */
621 if (throttle_enabled(&cfg)) {
622 if (!throttling_group) {
623 throttling_group = id;
624 }
625 blk_io_limits_enable(blk, throttling_group);
626 blk_set_io_limits(blk, &cfg);
627 }
628
629 blk_set_enable_write_cache(blk, !writethrough);
630 blk_set_on_error(blk, on_read_error, on_write_error);
631
632 if (!monitor_add_blk(blk, id, errp)) {
633 blk_unref(blk);
634 blk = NULL;
635 goto err_no_bs_opts;
636 }
637
638 err_no_bs_opts:
639 qemu_opts_del(opts);
640 qobject_unref(interval_dict);
641 qobject_unref(interval_list);
642 return blk;
643
644 early_err:
645 qemu_opts_del(opts);
646 qobject_unref(interval_dict);
647 qobject_unref(interval_list);
648 err_no_opts:
649 qobject_unref(bs_opts);
650 return NULL;
651 }
652
653 /* Takes the ownership of bs_opts */
654 BlockDriverState *bds_tree_init(QDict *bs_opts, Error **errp)
655 {
656 int bdrv_flags = 0;
657
658 GLOBAL_STATE_CODE();
659 /* bdrv_open() defaults to the values in bdrv_flags (for compatibility
660 * with other callers) rather than what we want as the real defaults.
661 * Apply the defaults here instead. */
662 qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_DIRECT, "off");
663 qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_NO_FLUSH, "off");
664 qdict_set_default_str(bs_opts, BDRV_OPT_READ_ONLY, "off");
665
666 if (runstate_check(RUN_STATE_INMIGRATE)) {
667 bdrv_flags |= BDRV_O_INACTIVE;
668 }
669
670 return bdrv_open(NULL, NULL, bs_opts, bdrv_flags, errp);
671 }
672
673 void blockdev_close_all_bdrv_states(void)
674 {
675 BlockDriverState *bs, *next_bs;
676
677 GLOBAL_STATE_CODE();
678 QTAILQ_FOREACH_SAFE(bs, &monitor_bdrv_states, monitor_list, next_bs) {
679 AioContext *ctx = bdrv_get_aio_context(bs);
680
681 aio_context_acquire(ctx);
682 bdrv_unref(bs);
683 aio_context_release(ctx);
684 }
685 }
686
687 /* Iterates over the list of monitor-owned BlockDriverStates */
688 BlockDriverState *bdrv_next_monitor_owned(BlockDriverState *bs)
689 {
690 GLOBAL_STATE_CODE();
691 return bs ? QTAILQ_NEXT(bs, monitor_list)
692 : QTAILQ_FIRST(&monitor_bdrv_states);
693 }
694
695 static bool qemu_opt_rename(QemuOpts *opts, const char *from, const char *to,
696 Error **errp)
697 {
698 const char *value;
699
700 value = qemu_opt_get(opts, from);
701 if (value) {
702 if (qemu_opt_find(opts, to)) {
703 error_setg(errp, "'%s' and its alias '%s' can't be used at the "
704 "same time", to, from);
705 return false;
706 }
707 }
708
709 /* rename all items in opts */
710 while ((value = qemu_opt_get(opts, from))) {
711 qemu_opt_set(opts, to, value, &error_abort);
712 qemu_opt_unset(opts, from);
713 }
714 return true;
715 }
716
717 QemuOptsList qemu_legacy_drive_opts = {
718 .name = "drive",
719 .head = QTAILQ_HEAD_INITIALIZER(qemu_legacy_drive_opts.head),
720 .desc = {
721 {
722 .name = "bus",
723 .type = QEMU_OPT_NUMBER,
724 .help = "bus number",
725 },{
726 .name = "unit",
727 .type = QEMU_OPT_NUMBER,
728 .help = "unit number (i.e. lun for scsi)",
729 },{
730 .name = "index",
731 .type = QEMU_OPT_NUMBER,
732 .help = "index number",
733 },{
734 .name = "media",
735 .type = QEMU_OPT_STRING,
736 .help = "media type (disk, cdrom)",
737 },{
738 .name = "if",
739 .type = QEMU_OPT_STRING,
740 .help = "interface (ide, scsi, sd, mtd, floppy, pflash, virtio)",
741 },{
742 .name = "file",
743 .type = QEMU_OPT_STRING,
744 .help = "file name",
745 },
746
747 /* Options that are passed on, but have special semantics with -drive */
748 {
749 .name = BDRV_OPT_READ_ONLY,
750 .type = QEMU_OPT_BOOL,
751 .help = "open drive file as read-only",
752 },{
753 .name = "rerror",
754 .type = QEMU_OPT_STRING,
755 .help = "read error action",
756 },{
757 .name = "werror",
758 .type = QEMU_OPT_STRING,
759 .help = "write error action",
760 },{
761 .name = "copy-on-read",
762 .type = QEMU_OPT_BOOL,
763 .help = "copy read data from backing file into image file",
764 },
765
766 { /* end of list */ }
767 },
768 };
769
770 DriveInfo *drive_new(QemuOpts *all_opts, BlockInterfaceType block_default_type,
771 Error **errp)
772 {
773 const char *value;
774 BlockBackend *blk;
775 DriveInfo *dinfo = NULL;
776 QDict *bs_opts;
777 QemuOpts *legacy_opts;
778 DriveMediaType media = MEDIA_DISK;
779 BlockInterfaceType type;
780 int max_devs, bus_id, unit_id, index;
781 const char *werror, *rerror;
782 bool read_only = false;
783 bool copy_on_read;
784 const char *filename;
785 int i;
786
787 GLOBAL_STATE_CODE();
788
789 /* Change legacy command line options into QMP ones */
790 static const struct {
791 const char *from;
792 const char *to;
793 } opt_renames[] = {
794 { "iops", "throttling.iops-total" },
795 { "iops_rd", "throttling.iops-read" },
796 { "iops_wr", "throttling.iops-write" },
797
798 { "bps", "throttling.bps-total" },
799 { "bps_rd", "throttling.bps-read" },
800 { "bps_wr", "throttling.bps-write" },
801
802 { "iops_max", "throttling.iops-total-max" },
803 { "iops_rd_max", "throttling.iops-read-max" },
804 { "iops_wr_max", "throttling.iops-write-max" },
805
806 { "bps_max", "throttling.bps-total-max" },
807 { "bps_rd_max", "throttling.bps-read-max" },
808 { "bps_wr_max", "throttling.bps-write-max" },
809
810 { "iops_size", "throttling.iops-size" },
811
812 { "group", "throttling.group" },
813
814 { "readonly", BDRV_OPT_READ_ONLY },
815 };
816
817 for (i = 0; i < ARRAY_SIZE(opt_renames); i++) {
818 if (!qemu_opt_rename(all_opts, opt_renames[i].from,
819 opt_renames[i].to, errp)) {
820 return NULL;
821 }
822 }
823
824 value = qemu_opt_get(all_opts, "cache");
825 if (value) {
826 int flags = 0;
827 bool writethrough;
828
829 if (bdrv_parse_cache_mode(value, &flags, &writethrough) != 0) {
830 error_setg(errp, "invalid cache option");
831 return NULL;
832 }
833
834 /* Specific options take precedence */
835 if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_WB)) {
836 qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_WB,
837 !writethrough, &error_abort);
838 }
839 if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_DIRECT)) {
840 qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_DIRECT,
841 !!(flags & BDRV_O_NOCACHE), &error_abort);
842 }
843 if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_NO_FLUSH)) {
844 qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_NO_FLUSH,
845 !!(flags & BDRV_O_NO_FLUSH), &error_abort);
846 }
847 qemu_opt_unset(all_opts, "cache");
848 }
849
850 /* Get a QDict for processing the options */
851 bs_opts = qdict_new();
852 qemu_opts_to_qdict(all_opts, bs_opts);
853
854 legacy_opts = qemu_opts_create(&qemu_legacy_drive_opts, NULL, 0,
855 &error_abort);
856 if (!qemu_opts_absorb_qdict(legacy_opts, bs_opts, errp)) {
857 goto fail;
858 }
859
860 /* Media type */
861 value = qemu_opt_get(legacy_opts, "media");
862 if (value) {
863 if (!strcmp(value, "disk")) {
864 media = MEDIA_DISK;
865 } else if (!strcmp(value, "cdrom")) {
866 media = MEDIA_CDROM;
867 read_only = true;
868 } else {
869 error_setg(errp, "'%s' invalid media", value);
870 goto fail;
871 }
872 }
873
874 /* copy-on-read is disabled with a warning for read-only devices */
875 read_only |= qemu_opt_get_bool(legacy_opts, BDRV_OPT_READ_ONLY, false);
876 copy_on_read = qemu_opt_get_bool(legacy_opts, "copy-on-read", false);
877
878 if (read_only && copy_on_read) {
879 warn_report("disabling copy-on-read on read-only drive");
880 copy_on_read = false;
881 }
882
883 qdict_put_str(bs_opts, BDRV_OPT_READ_ONLY, read_only ? "on" : "off");
884 qdict_put_str(bs_opts, "copy-on-read", copy_on_read ? "on" : "off");
885
886 /* Controller type */
887 value = qemu_opt_get(legacy_opts, "if");
888 if (value) {
889 for (type = 0;
890 type < IF_COUNT && strcmp(value, if_name[type]);
891 type++) {
892 }
893 if (type == IF_COUNT) {
894 error_setg(errp, "unsupported bus type '%s'", value);
895 goto fail;
896 }
897 } else {
898 type = block_default_type;
899 }
900
901 /* Device address specified by bus/unit or index.
902 * If none was specified, try to find the first free one. */
903 bus_id = qemu_opt_get_number(legacy_opts, "bus", 0);
904 unit_id = qemu_opt_get_number(legacy_opts, "unit", -1);
905 index = qemu_opt_get_number(legacy_opts, "index", -1);
906
907 max_devs = if_max_devs[type];
908
909 if (index != -1) {
910 if (bus_id != 0 || unit_id != -1) {
911 error_setg(errp, "index cannot be used with bus and unit");
912 goto fail;
913 }
914 bus_id = drive_index_to_bus_id(type, index);
915 unit_id = drive_index_to_unit_id(type, index);
916 }
917
918 if (unit_id == -1) {
919 unit_id = 0;
920 while (drive_get(type, bus_id, unit_id) != NULL) {
921 unit_id++;
922 if (max_devs && unit_id >= max_devs) {
923 unit_id -= max_devs;
924 bus_id++;
925 }
926 }
927 }
928
929 if (max_devs && unit_id >= max_devs) {
930 error_setg(errp, "unit %d too big (max is %d)", unit_id, max_devs - 1);
931 goto fail;
932 }
933
934 if (drive_get(type, bus_id, unit_id) != NULL) {
935 error_setg(errp, "drive with bus=%d, unit=%d (index=%d) exists",
936 bus_id, unit_id, index);
937 goto fail;
938 }
939
940 /* no id supplied -> create one */
941 if (qemu_opts_id(all_opts) == NULL) {
942 char *new_id;
943 const char *mediastr = "";
944 if (type == IF_IDE || type == IF_SCSI) {
945 mediastr = (media == MEDIA_CDROM) ? "-cd" : "-hd";
946 }
947 if (max_devs) {
948 new_id = g_strdup_printf("%s%i%s%i", if_name[type], bus_id,
949 mediastr, unit_id);
950 } else {
951 new_id = g_strdup_printf("%s%s%i", if_name[type],
952 mediastr, unit_id);
953 }
954 qdict_put_str(bs_opts, "id", new_id);
955 g_free(new_id);
956 }
957
958 /* Add virtio block device */
959 if (type == IF_VIRTIO) {
960 QemuOpts *devopts;
961 devopts = qemu_opts_create(qemu_find_opts("device"), NULL, 0,
962 &error_abort);
963 qemu_opt_set(devopts, "driver", "virtio-blk", &error_abort);
964 qemu_opt_set(devopts, "drive", qdict_get_str(bs_opts, "id"),
965 &error_abort);
966 }
967
968 filename = qemu_opt_get(legacy_opts, "file");
969
970 /* Check werror/rerror compatibility with if=... */
971 werror = qemu_opt_get(legacy_opts, "werror");
972 if (werror != NULL) {
973 if (type != IF_IDE && type != IF_SCSI && type != IF_VIRTIO &&
974 type != IF_NONE) {
975 error_setg(errp, "werror is not supported by this bus type");
976 goto fail;
977 }
978 qdict_put_str(bs_opts, "werror", werror);
979 }
980
981 rerror = qemu_opt_get(legacy_opts, "rerror");
982 if (rerror != NULL) {
983 if (type != IF_IDE && type != IF_VIRTIO && type != IF_SCSI &&
984 type != IF_NONE) {
985 error_setg(errp, "rerror is not supported by this bus type");
986 goto fail;
987 }
988 qdict_put_str(bs_opts, "rerror", rerror);
989 }
990
991 /* Actual block device init: Functionality shared with blockdev-add */
992 blk = blockdev_init(filename, bs_opts, errp);
993 bs_opts = NULL;
994 if (!blk) {
995 goto fail;
996 }
997
998 /* Create legacy DriveInfo */
999 dinfo = g_malloc0(sizeof(*dinfo));
1000 dinfo->opts = all_opts;
1001
1002 dinfo->type = type;
1003 dinfo->bus = bus_id;
1004 dinfo->unit = unit_id;
1005
1006 blk_set_legacy_dinfo(blk, dinfo);
1007
1008 switch(type) {
1009 case IF_IDE:
1010 case IF_SCSI:
1011 case IF_XEN:
1012 case IF_NONE:
1013 dinfo->media_cd = media == MEDIA_CDROM;
1014 break;
1015 default:
1016 break;
1017 }
1018
1019 fail:
1020 qemu_opts_del(legacy_opts);
1021 qobject_unref(bs_opts);
1022 return dinfo;
1023 }
1024
1025 static BlockDriverState *qmp_get_root_bs(const char *name, Error **errp)
1026 {
1027 BlockDriverState *bs;
1028
1029 bs = bdrv_lookup_bs(name, name, errp);
1030 if (bs == NULL) {
1031 return NULL;
1032 }
1033
1034 if (!bdrv_is_root_node(bs)) {
1035 error_setg(errp, "Need a root block node");
1036 return NULL;
1037 }
1038
1039 if (!bdrv_is_inserted(bs)) {
1040 error_setg(errp, "Device has no medium");
1041 return NULL;
1042 }
1043
1044 return bs;
1045 }
1046
1047 static void blockdev_do_action(TransactionAction *action, Error **errp)
1048 {
1049 TransactionActionList list;
1050
1051 list.value = action;
1052 list.next = NULL;
1053 qmp_transaction(&list, false, NULL, errp);
1054 }
1055
1056 void qmp_blockdev_snapshot_sync(bool has_device, const char *device,
1057 bool has_node_name, const char *node_name,
1058 const char *snapshot_file,
1059 bool has_snapshot_node_name,
1060 const char *snapshot_node_name,
1061 bool has_format, const char *format,
1062 bool has_mode, NewImageMode mode, Error **errp)
1063 {
1064 BlockdevSnapshotSync snapshot = {
1065 .has_device = has_device,
1066 .device = (char *) device,
1067 .has_node_name = has_node_name,
1068 .node_name = (char *) node_name,
1069 .snapshot_file = (char *) snapshot_file,
1070 .has_snapshot_node_name = has_snapshot_node_name,
1071 .snapshot_node_name = (char *) snapshot_node_name,
1072 .has_format = has_format,
1073 .format = (char *) format,
1074 .has_mode = has_mode,
1075 .mode = mode,
1076 };
1077 TransactionAction action = {
1078 .type = TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC,
1079 .u.blockdev_snapshot_sync.data = &snapshot,
1080 };
1081 blockdev_do_action(&action, errp);
1082 }
1083
1084 void qmp_blockdev_snapshot(const char *node, const char *overlay,
1085 Error **errp)
1086 {
1087 BlockdevSnapshot snapshot_data = {
1088 .node = (char *) node,
1089 .overlay = (char *) overlay
1090 };
1091 TransactionAction action = {
1092 .type = TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT,
1093 .u.blockdev_snapshot.data = &snapshot_data,
1094 };
1095 blockdev_do_action(&action, errp);
1096 }
1097
1098 void qmp_blockdev_snapshot_internal_sync(const char *device,
1099 const char *name,
1100 Error **errp)
1101 {
1102 BlockdevSnapshotInternal snapshot = {
1103 .device = (char *) device,
1104 .name = (char *) name
1105 };
1106 TransactionAction action = {
1107 .type = TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC,
1108 .u.blockdev_snapshot_internal_sync.data = &snapshot,
1109 };
1110 blockdev_do_action(&action, errp);
1111 }
1112
1113 SnapshotInfo *qmp_blockdev_snapshot_delete_internal_sync(const char *device,
1114 bool has_id,
1115 const char *id,
1116 bool has_name,
1117 const char *name,
1118 Error **errp)
1119 {
1120 BlockDriverState *bs;
1121 AioContext *aio_context;
1122 QEMUSnapshotInfo sn;
1123 Error *local_err = NULL;
1124 SnapshotInfo *info = NULL;
1125 int ret;
1126
1127 bs = qmp_get_root_bs(device, errp);
1128 if (!bs) {
1129 return NULL;
1130 }
1131 aio_context = bdrv_get_aio_context(bs);
1132 aio_context_acquire(aio_context);
1133
1134 if (!has_id) {
1135 id = NULL;
1136 }
1137
1138 if (!has_name) {
1139 name = NULL;
1140 }
1141
1142 if (!id && !name) {
1143 error_setg(errp, "Name or id must be provided");
1144 goto out_aio_context;
1145 }
1146
1147 if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_INTERNAL_SNAPSHOT_DELETE, errp)) {
1148 goto out_aio_context;
1149 }
1150
1151 ret = bdrv_snapshot_find_by_id_and_name(bs, id, name, &sn, &local_err);
1152 if (local_err) {
1153 error_propagate(errp, local_err);
1154 goto out_aio_context;
1155 }
1156 if (!ret) {
1157 error_setg(errp,
1158 "Snapshot with id '%s' and name '%s' does not exist on "
1159 "device '%s'",
1160 STR_OR_NULL(id), STR_OR_NULL(name), device);
1161 goto out_aio_context;
1162 }
1163
1164 bdrv_snapshot_delete(bs, id, name, &local_err);
1165 if (local_err) {
1166 error_propagate(errp, local_err);
1167 goto out_aio_context;
1168 }
1169
1170 aio_context_release(aio_context);
1171
1172 info = g_new0(SnapshotInfo, 1);
1173 info->id = g_strdup(sn.id_str);
1174 info->name = g_strdup(sn.name);
1175 info->date_nsec = sn.date_nsec;
1176 info->date_sec = sn.date_sec;
1177 info->vm_state_size = sn.vm_state_size;
1178 info->vm_clock_nsec = sn.vm_clock_nsec % 1000000000;
1179 info->vm_clock_sec = sn.vm_clock_nsec / 1000000000;
1180 if (sn.icount != -1ULL) {
1181 info->icount = sn.icount;
1182 info->has_icount = true;
1183 }
1184
1185 return info;
1186
1187 out_aio_context:
1188 aio_context_release(aio_context);
1189 return NULL;
1190 }
1191
1192 /* New and old BlockDriverState structs for atomic group operations */
1193
1194 typedef struct BlkActionState BlkActionState;
1195
1196 /**
1197 * BlkActionOps:
1198 * Table of operations that define an Action.
1199 *
1200 * @instance_size: Size of state struct, in bytes.
1201 * @prepare: Prepare the work, must NOT be NULL.
1202 * @commit: Commit the changes, can be NULL.
1203 * @abort: Abort the changes on fail, can be NULL.
1204 * @clean: Clean up resources after all transaction actions have called
1205 * commit() or abort(). Can be NULL.
1206 *
1207 * Only prepare() may fail. In a single transaction, only one of commit() or
1208 * abort() will be called. clean() will always be called if it is present.
1209 *
1210 * Always run under BQL.
1211 */
1212 typedef struct BlkActionOps {
1213 size_t instance_size;
1214 void (*prepare)(BlkActionState *common, Error **errp);
1215 void (*commit)(BlkActionState *common);
1216 void (*abort)(BlkActionState *common);
1217 void (*clean)(BlkActionState *common);
1218 } BlkActionOps;
1219
1220 /**
1221 * BlkActionState:
1222 * Describes one Action's state within a Transaction.
1223 *
1224 * @action: QAPI-defined enum identifying which Action to perform.
1225 * @ops: Table of ActionOps this Action can perform.
1226 * @block_job_txn: Transaction which this action belongs to.
1227 * @entry: List membership for all Actions in this Transaction.
1228 *
1229 * This structure must be arranged as first member in a subclassed type,
1230 * assuming that the compiler will also arrange it to the same offsets as the
1231 * base class.
1232 */
1233 struct BlkActionState {
1234 TransactionAction *action;
1235 const BlkActionOps *ops;
1236 JobTxn *block_job_txn;
1237 TransactionProperties *txn_props;
1238 QTAILQ_ENTRY(BlkActionState) entry;
1239 };
1240
1241 /* internal snapshot private data */
1242 typedef struct InternalSnapshotState {
1243 BlkActionState common;
1244 BlockDriverState *bs;
1245 QEMUSnapshotInfo sn;
1246 bool created;
1247 } InternalSnapshotState;
1248
1249
1250 static int action_check_completion_mode(BlkActionState *s, Error **errp)
1251 {
1252 if (s->txn_props->completion_mode != ACTION_COMPLETION_MODE_INDIVIDUAL) {
1253 error_setg(errp,
1254 "Action '%s' does not support Transaction property "
1255 "completion-mode = %s",
1256 TransactionActionKind_str(s->action->type),
1257 ActionCompletionMode_str(s->txn_props->completion_mode));
1258 return -1;
1259 }
1260 return 0;
1261 }
1262
1263 static void internal_snapshot_prepare(BlkActionState *common,
1264 Error **errp)
1265 {
1266 Error *local_err = NULL;
1267 const char *device;
1268 const char *name;
1269 BlockDriverState *bs;
1270 QEMUSnapshotInfo old_sn, *sn;
1271 bool ret;
1272 int64_t rt;
1273 BlockdevSnapshotInternal *internal;
1274 InternalSnapshotState *state;
1275 AioContext *aio_context;
1276 int ret1;
1277
1278 g_assert(common->action->type ==
1279 TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC);
1280 internal = common->action->u.blockdev_snapshot_internal_sync.data;
1281 state = DO_UPCAST(InternalSnapshotState, common, common);
1282
1283 /* 1. parse input */
1284 device = internal->device;
1285 name = internal->name;
1286
1287 /* 2. check for validation */
1288 if (action_check_completion_mode(common, errp) < 0) {
1289 return;
1290 }
1291
1292 bs = qmp_get_root_bs(device, errp);
1293 if (!bs) {
1294 return;
1295 }
1296
1297 aio_context = bdrv_get_aio_context(bs);
1298 aio_context_acquire(aio_context);
1299
1300 state->bs = bs;
1301
1302 /* Paired with .clean() */
1303 bdrv_drained_begin(bs);
1304
1305 if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_INTERNAL_SNAPSHOT, errp)) {
1306 goto out;
1307 }
1308
1309 if (bdrv_is_read_only(bs)) {
1310 error_setg(errp, "Device '%s' is read only", device);
1311 goto out;
1312 }
1313
1314 if (!bdrv_can_snapshot(bs)) {
1315 error_setg(errp, "Block format '%s' used by device '%s' "
1316 "does not support internal snapshots",
1317 bs->drv->format_name, device);
1318 goto out;
1319 }
1320
1321 if (!strlen(name)) {
1322 error_setg(errp, "Name is empty");
1323 goto out;
1324 }
1325
1326 /* check whether a snapshot with name exist */
1327 ret = bdrv_snapshot_find_by_id_and_name(bs, NULL, name, &old_sn,
1328 &local_err);
1329 if (local_err) {
1330 error_propagate(errp, local_err);
1331 goto out;
1332 } else if (ret) {
1333 error_setg(errp,
1334 "Snapshot with name '%s' already exists on device '%s'",
1335 name, device);
1336 goto out;
1337 }
1338
1339 /* 3. take the snapshot */
1340 sn = &state->sn;
1341 pstrcpy(sn->name, sizeof(sn->name), name);
1342 rt = g_get_real_time();
1343 sn->date_sec = rt / G_USEC_PER_SEC;
1344 sn->date_nsec = (rt % G_USEC_PER_SEC) * 1000;
1345 sn->vm_clock_nsec = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL);
1346 if (replay_mode != REPLAY_MODE_NONE) {
1347 sn->icount = replay_get_current_icount();
1348 } else {
1349 sn->icount = -1ULL;
1350 }
1351
1352 ret1 = bdrv_snapshot_create(bs, sn);
1353 if (ret1 < 0) {
1354 error_setg_errno(errp, -ret1,
1355 "Failed to create snapshot '%s' on device '%s'",
1356 name, device);
1357 goto out;
1358 }
1359
1360 /* 4. succeed, mark a snapshot is created */
1361 state->created = true;
1362
1363 out:
1364 aio_context_release(aio_context);
1365 }
1366
1367 static void internal_snapshot_abort(BlkActionState *common)
1368 {
1369 InternalSnapshotState *state =
1370 DO_UPCAST(InternalSnapshotState, common, common);
1371 BlockDriverState *bs = state->bs;
1372 QEMUSnapshotInfo *sn = &state->sn;
1373 AioContext *aio_context;
1374 Error *local_error = NULL;
1375
1376 if (!state->created) {
1377 return;
1378 }
1379
1380 aio_context = bdrv_get_aio_context(state->bs);
1381 aio_context_acquire(aio_context);
1382
1383 if (bdrv_snapshot_delete(bs, sn->id_str, sn->name, &local_error) < 0) {
1384 error_reportf_err(local_error,
1385 "Failed to delete snapshot with id '%s' and "
1386 "name '%s' on device '%s' in abort: ",
1387 sn->id_str, sn->name,
1388 bdrv_get_device_name(bs));
1389 }
1390
1391 aio_context_release(aio_context);
1392 }
1393
1394 static void internal_snapshot_clean(BlkActionState *common)
1395 {
1396 InternalSnapshotState *state = DO_UPCAST(InternalSnapshotState,
1397 common, common);
1398 AioContext *aio_context;
1399
1400 if (!state->bs) {
1401 return;
1402 }
1403
1404 aio_context = bdrv_get_aio_context(state->bs);
1405 aio_context_acquire(aio_context);
1406
1407 bdrv_drained_end(state->bs);
1408
1409 aio_context_release(aio_context);
1410 }
1411
1412 /* external snapshot private data */
1413 typedef struct ExternalSnapshotState {
1414 BlkActionState common;
1415 BlockDriverState *old_bs;
1416 BlockDriverState *new_bs;
1417 bool overlay_appended;
1418 } ExternalSnapshotState;
1419
1420 static void external_snapshot_prepare(BlkActionState *common,
1421 Error **errp)
1422 {
1423 int ret;
1424 int flags = 0;
1425 QDict *options = NULL;
1426 Error *local_err = NULL;
1427 /* Device and node name of the image to generate the snapshot from */
1428 const char *device;
1429 const char *node_name;
1430 /* Reference to the new image (for 'blockdev-snapshot') */
1431 const char *snapshot_ref;
1432 /* File name of the new image (for 'blockdev-snapshot-sync') */
1433 const char *new_image_file;
1434 ExternalSnapshotState *state =
1435 DO_UPCAST(ExternalSnapshotState, common, common);
1436 TransactionAction *action = common->action;
1437 AioContext *aio_context;
1438 uint64_t perm, shared;
1439
1440 /* 'blockdev-snapshot' and 'blockdev-snapshot-sync' have similar
1441 * purpose but a different set of parameters */
1442 switch (action->type) {
1443 case TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT:
1444 {
1445 BlockdevSnapshot *s = action->u.blockdev_snapshot.data;
1446 device = s->node;
1447 node_name = s->node;
1448 new_image_file = NULL;
1449 snapshot_ref = s->overlay;
1450 }
1451 break;
1452 case TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC:
1453 {
1454 BlockdevSnapshotSync *s = action->u.blockdev_snapshot_sync.data;
1455 device = s->has_device ? s->device : NULL;
1456 node_name = s->has_node_name ? s->node_name : NULL;
1457 new_image_file = s->snapshot_file;
1458 snapshot_ref = NULL;
1459 }
1460 break;
1461 default:
1462 g_assert_not_reached();
1463 }
1464
1465 /* start processing */
1466 if (action_check_completion_mode(common, errp) < 0) {
1467 return;
1468 }
1469
1470 state->old_bs = bdrv_lookup_bs(device, node_name, errp);
1471 if (!state->old_bs) {
1472 return;
1473 }
1474
1475 aio_context = bdrv_get_aio_context(state->old_bs);
1476 aio_context_acquire(aio_context);
1477
1478 /* Paired with .clean() */
1479 bdrv_drained_begin(state->old_bs);
1480
1481 if (!bdrv_is_inserted(state->old_bs)) {
1482 error_setg(errp, QERR_DEVICE_HAS_NO_MEDIUM, device);
1483 goto out;
1484 }
1485
1486 if (bdrv_op_is_blocked(state->old_bs,
1487 BLOCK_OP_TYPE_EXTERNAL_SNAPSHOT, errp)) {
1488 goto out;
1489 }
1490
1491 if (!bdrv_is_read_only(state->old_bs)) {
1492 if (bdrv_flush(state->old_bs)) {
1493 error_setg(errp, QERR_IO_ERROR);
1494 goto out;
1495 }
1496 }
1497
1498 if (action->type == TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC) {
1499 BlockdevSnapshotSync *s = action->u.blockdev_snapshot_sync.data;
1500 const char *format = s->has_format ? s->format : "qcow2";
1501 enum NewImageMode mode;
1502 const char *snapshot_node_name =
1503 s->has_snapshot_node_name ? s->snapshot_node_name : NULL;
1504
1505 if (node_name && !snapshot_node_name) {
1506 error_setg(errp, "New overlay node-name missing");
1507 goto out;
1508 }
1509
1510 if (snapshot_node_name &&
1511 bdrv_lookup_bs(snapshot_node_name, snapshot_node_name, NULL)) {
1512 error_setg(errp, "New overlay node-name already in use");
1513 goto out;
1514 }
1515
1516 flags = state->old_bs->open_flags;
1517 flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_COPY_ON_READ);
1518 flags |= BDRV_O_NO_BACKING;
1519
1520 /* create new image w/backing file */
1521 mode = s->has_mode ? s->mode : NEW_IMAGE_MODE_ABSOLUTE_PATHS;
1522 if (mode != NEW_IMAGE_MODE_EXISTING) {
1523 int64_t size = bdrv_getlength(state->old_bs);
1524 if (size < 0) {
1525 error_setg_errno(errp, -size, "bdrv_getlength failed");
1526 goto out;
1527 }
1528 bdrv_refresh_filename(state->old_bs);
1529 bdrv_img_create(new_image_file, format,
1530 state->old_bs->filename,
1531 state->old_bs->drv->format_name,
1532 NULL, size, flags, false, &local_err);
1533 if (local_err) {
1534 error_propagate(errp, local_err);
1535 goto out;
1536 }
1537 }
1538
1539 options = qdict_new();
1540 if (snapshot_node_name) {
1541 qdict_put_str(options, "node-name", snapshot_node_name);
1542 }
1543 qdict_put_str(options, "driver", format);
1544 }
1545
1546 state->new_bs = bdrv_open(new_image_file, snapshot_ref, options, flags,
1547 errp);
1548 /* We will manually add the backing_hd field to the bs later */
1549 if (!state->new_bs) {
1550 goto out;
1551 }
1552
1553 /*
1554 * Allow attaching a backing file to an overlay that's already in use only
1555 * if the parents don't assume that they are already seeing a valid image.
1556 * (Specifically, allow it as a mirror target, which is write-only access.)
1557 */
1558 bdrv_get_cumulative_perm(state->new_bs, &perm, &shared);
1559 if (perm & BLK_PERM_CONSISTENT_READ) {
1560 error_setg(errp, "The overlay is already in use");
1561 goto out;
1562 }
1563
1564 if (state->new_bs->drv->is_filter) {
1565 error_setg(errp, "Filters cannot be used as overlays");
1566 goto out;
1567 }
1568
1569 if (bdrv_cow_child(state->new_bs)) {
1570 error_setg(errp, "The overlay already has a backing image");
1571 goto out;
1572 }
1573
1574 if (!state->new_bs->drv->supports_backing) {
1575 error_setg(errp, "The overlay does not support backing images");
1576 goto out;
1577 }
1578
1579 ret = bdrv_append(state->new_bs, state->old_bs, errp);
1580 if (ret < 0) {
1581 goto out;
1582 }
1583 state->overlay_appended = true;
1584
1585 out:
1586 aio_context_release(aio_context);
1587 }
1588
1589 static void external_snapshot_commit(BlkActionState *common)
1590 {
1591 ExternalSnapshotState *state =
1592 DO_UPCAST(ExternalSnapshotState, common, common);
1593 AioContext *aio_context;
1594
1595 aio_context = bdrv_get_aio_context(state->old_bs);
1596 aio_context_acquire(aio_context);
1597
1598 /* We don't need (or want) to use the transactional
1599 * bdrv_reopen_multiple() across all the entries at once, because we
1600 * don't want to abort all of them if one of them fails the reopen */
1601 if (!qatomic_read(&state->old_bs->copy_on_read)) {
1602 bdrv_reopen_set_read_only(state->old_bs, true, NULL);
1603 }
1604
1605 aio_context_release(aio_context);
1606 }
1607
1608 static void external_snapshot_abort(BlkActionState *common)
1609 {
1610 ExternalSnapshotState *state =
1611 DO_UPCAST(ExternalSnapshotState, common, common);
1612 if (state->new_bs) {
1613 if (state->overlay_appended) {
1614 AioContext *aio_context;
1615 AioContext *tmp_context;
1616 int ret;
1617
1618 aio_context = bdrv_get_aio_context(state->old_bs);
1619 aio_context_acquire(aio_context);
1620
1621 bdrv_ref(state->old_bs); /* we can't let bdrv_set_backind_hd()
1622 close state->old_bs; we need it */
1623 bdrv_set_backing_hd(state->new_bs, NULL, &error_abort);
1624
1625 /*
1626 * The call to bdrv_set_backing_hd() above returns state->old_bs to
1627 * the main AioContext. As we're still going to be using it, return
1628 * it to the AioContext it was before.
1629 */
1630 tmp_context = bdrv_get_aio_context(state->old_bs);
1631 if (aio_context != tmp_context) {
1632 aio_context_release(aio_context);
1633 aio_context_acquire(tmp_context);
1634
1635 ret = bdrv_try_set_aio_context(state->old_bs,
1636 aio_context, NULL);
1637 assert(ret == 0);
1638
1639 aio_context_release(tmp_context);
1640 aio_context_acquire(aio_context);
1641 }
1642
1643 bdrv_replace_node(state->new_bs, state->old_bs, &error_abort);
1644 bdrv_unref(state->old_bs); /* bdrv_replace_node() ref'ed old_bs */
1645
1646 aio_context_release(aio_context);
1647 }
1648 }
1649 }
1650
1651 static void external_snapshot_clean(BlkActionState *common)
1652 {
1653 ExternalSnapshotState *state =
1654 DO_UPCAST(ExternalSnapshotState, common, common);
1655 AioContext *aio_context;
1656
1657 if (!state->old_bs) {
1658 return;
1659 }
1660
1661 aio_context = bdrv_get_aio_context(state->old_bs);
1662 aio_context_acquire(aio_context);
1663
1664 bdrv_drained_end(state->old_bs);
1665 bdrv_unref(state->new_bs);
1666
1667 aio_context_release(aio_context);
1668 }
1669
1670 typedef struct DriveBackupState {
1671 BlkActionState common;
1672 BlockDriverState *bs;
1673 BlockJob *job;
1674 } DriveBackupState;
1675
1676 static BlockJob *do_backup_common(BackupCommon *backup,
1677 BlockDriverState *bs,
1678 BlockDriverState *target_bs,
1679 AioContext *aio_context,
1680 JobTxn *txn, Error **errp);
1681
1682 static void drive_backup_prepare(BlkActionState *common, Error **errp)
1683 {
1684 DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1685 DriveBackup *backup;
1686 BlockDriverState *bs;
1687 BlockDriverState *target_bs;
1688 BlockDriverState *source = NULL;
1689 AioContext *aio_context;
1690 AioContext *old_context;
1691 QDict *options;
1692 Error *local_err = NULL;
1693 int flags;
1694 int64_t size;
1695 bool set_backing_hd = false;
1696 int ret;
1697
1698 assert(common->action->type == TRANSACTION_ACTION_KIND_DRIVE_BACKUP);
1699 backup = common->action->u.drive_backup.data;
1700
1701 if (!backup->has_mode) {
1702 backup->mode = NEW_IMAGE_MODE_ABSOLUTE_PATHS;
1703 }
1704
1705 bs = bdrv_lookup_bs(backup->device, backup->device, errp);
1706 if (!bs) {
1707 return;
1708 }
1709
1710 if (!bs->drv) {
1711 error_setg(errp, "Device has no medium");
1712 return;
1713 }
1714
1715 aio_context = bdrv_get_aio_context(bs);
1716 aio_context_acquire(aio_context);
1717
1718 state->bs = bs;
1719 /* Paired with .clean() */
1720 bdrv_drained_begin(bs);
1721
1722 if (!backup->has_format) {
1723 backup->format = backup->mode == NEW_IMAGE_MODE_EXISTING ?
1724 NULL : (char *) bs->drv->format_name;
1725 }
1726
1727 /* Early check to avoid creating target */
1728 if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_BACKUP_SOURCE, errp)) {
1729 goto out;
1730 }
1731
1732 flags = bs->open_flags | BDRV_O_RDWR;
1733
1734 /*
1735 * See if we have a backing HD we can use to create our new image
1736 * on top of.
1737 */
1738 if (backup->sync == MIRROR_SYNC_MODE_TOP) {
1739 /*
1740 * Backup will not replace the source by the target, so none
1741 * of the filters skipped here will be removed (in contrast to
1742 * mirror). Therefore, we can skip all of them when looking
1743 * for the first COW relationship.
1744 */
1745 source = bdrv_cow_bs(bdrv_skip_filters(bs));
1746 if (!source) {
1747 backup->sync = MIRROR_SYNC_MODE_FULL;
1748 }
1749 }
1750 if (backup->sync == MIRROR_SYNC_MODE_NONE) {
1751 source = bs;
1752 flags |= BDRV_O_NO_BACKING;
1753 set_backing_hd = true;
1754 }
1755
1756 size = bdrv_getlength(bs);
1757 if (size < 0) {
1758 error_setg_errno(errp, -size, "bdrv_getlength failed");
1759 goto out;
1760 }
1761
1762 if (backup->mode != NEW_IMAGE_MODE_EXISTING) {
1763 assert(backup->format);
1764 if (source) {
1765 /* Implicit filters should not appear in the filename */
1766 BlockDriverState *explicit_backing =
1767 bdrv_skip_implicit_filters(source);
1768
1769 bdrv_refresh_filename(explicit_backing);
1770 bdrv_img_create(backup->target, backup->format,
1771 explicit_backing->filename,
1772 explicit_backing->drv->format_name, NULL,
1773 size, flags, false, &local_err);
1774 } else {
1775 bdrv_img_create(backup->target, backup->format, NULL, NULL, NULL,
1776 size, flags, false, &local_err);
1777 }
1778 }
1779
1780 if (local_err) {
1781 error_propagate(errp, local_err);
1782 goto out;
1783 }
1784
1785 options = qdict_new();
1786 qdict_put_str(options, "discard", "unmap");
1787 qdict_put_str(options, "detect-zeroes", "unmap");
1788 if (backup->format) {
1789 qdict_put_str(options, "driver", backup->format);
1790 }
1791
1792 target_bs = bdrv_open(backup->target, NULL, options, flags, errp);
1793 if (!target_bs) {
1794 goto out;
1795 }
1796
1797 /* Honor bdrv_try_set_aio_context() context acquisition requirements. */
1798 old_context = bdrv_get_aio_context(target_bs);
1799 aio_context_release(aio_context);
1800 aio_context_acquire(old_context);
1801
1802 ret = bdrv_try_set_aio_context(target_bs, aio_context, errp);
1803 if (ret < 0) {
1804 bdrv_unref(target_bs);
1805 aio_context_release(old_context);
1806 return;
1807 }
1808
1809 aio_context_release(old_context);
1810 aio_context_acquire(aio_context);
1811
1812 if (set_backing_hd) {
1813 if (bdrv_set_backing_hd(target_bs, source, errp) < 0) {
1814 goto unref;
1815 }
1816 }
1817
1818 state->job = do_backup_common(qapi_DriveBackup_base(backup),
1819 bs, target_bs, aio_context,
1820 common->block_job_txn, errp);
1821
1822 unref:
1823 bdrv_unref(target_bs);
1824 out:
1825 aio_context_release(aio_context);
1826 }
1827
1828 static void drive_backup_commit(BlkActionState *common)
1829 {
1830 DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1831 AioContext *aio_context;
1832
1833 aio_context = bdrv_get_aio_context(state->bs);
1834 aio_context_acquire(aio_context);
1835
1836 assert(state->job);
1837 job_start(&state->job->job);
1838
1839 aio_context_release(aio_context);
1840 }
1841
1842 static void drive_backup_abort(BlkActionState *common)
1843 {
1844 DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1845
1846 if (state->job) {
1847 AioContext *aio_context;
1848
1849 aio_context = bdrv_get_aio_context(state->bs);
1850 aio_context_acquire(aio_context);
1851
1852 job_cancel_sync(&state->job->job, true);
1853
1854 aio_context_release(aio_context);
1855 }
1856 }
1857
1858 static void drive_backup_clean(BlkActionState *common)
1859 {
1860 DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1861 AioContext *aio_context;
1862
1863 if (!state->bs) {
1864 return;
1865 }
1866
1867 aio_context = bdrv_get_aio_context(state->bs);
1868 aio_context_acquire(aio_context);
1869
1870 bdrv_drained_end(state->bs);
1871
1872 aio_context_release(aio_context);
1873 }
1874
1875 typedef struct BlockdevBackupState {
1876 BlkActionState common;
1877 BlockDriverState *bs;
1878 BlockJob *job;
1879 } BlockdevBackupState;
1880
1881 static void blockdev_backup_prepare(BlkActionState *common, Error **errp)
1882 {
1883 BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1884 BlockdevBackup *backup;
1885 BlockDriverState *bs;
1886 BlockDriverState *target_bs;
1887 AioContext *aio_context;
1888 AioContext *old_context;
1889 int ret;
1890
1891 assert(common->action->type == TRANSACTION_ACTION_KIND_BLOCKDEV_BACKUP);
1892 backup = common->action->u.blockdev_backup.data;
1893
1894 bs = bdrv_lookup_bs(backup->device, backup->device, errp);
1895 if (!bs) {
1896 return;
1897 }
1898
1899 target_bs = bdrv_lookup_bs(backup->target, backup->target, errp);
1900 if (!target_bs) {
1901 return;
1902 }
1903
1904 /* Honor bdrv_try_set_aio_context() context acquisition requirements. */
1905 aio_context = bdrv_get_aio_context(bs);
1906 old_context = bdrv_get_aio_context(target_bs);
1907 aio_context_acquire(old_context);
1908
1909 ret = bdrv_try_set_aio_context(target_bs, aio_context, errp);
1910 if (ret < 0) {
1911 aio_context_release(old_context);
1912 return;
1913 }
1914
1915 aio_context_release(old_context);
1916 aio_context_acquire(aio_context);
1917 state->bs = bs;
1918
1919 /* Paired with .clean() */
1920 bdrv_drained_begin(state->bs);
1921
1922 state->job = do_backup_common(qapi_BlockdevBackup_base(backup),
1923 bs, target_bs, aio_context,
1924 common->block_job_txn, errp);
1925
1926 aio_context_release(aio_context);
1927 }
1928
1929 static void blockdev_backup_commit(BlkActionState *common)
1930 {
1931 BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1932 AioContext *aio_context;
1933
1934 aio_context = bdrv_get_aio_context(state->bs);
1935 aio_context_acquire(aio_context);
1936
1937 assert(state->job);
1938 job_start(&state->job->job);
1939
1940 aio_context_release(aio_context);
1941 }
1942
1943 static void blockdev_backup_abort(BlkActionState *common)
1944 {
1945 BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1946
1947 if (state->job) {
1948 AioContext *aio_context;
1949
1950 aio_context = bdrv_get_aio_context(state->bs);
1951 aio_context_acquire(aio_context);
1952
1953 job_cancel_sync(&state->job->job, true);
1954
1955 aio_context_release(aio_context);
1956 }
1957 }
1958
1959 static void blockdev_backup_clean(BlkActionState *common)
1960 {
1961 BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1962 AioContext *aio_context;
1963
1964 if (!state->bs) {
1965 return;
1966 }
1967
1968 aio_context = bdrv_get_aio_context(state->bs);
1969 aio_context_acquire(aio_context);
1970
1971 bdrv_drained_end(state->bs);
1972
1973 aio_context_release(aio_context);
1974 }
1975
1976 typedef struct BlockDirtyBitmapState {
1977 BlkActionState common;
1978 BdrvDirtyBitmap *bitmap;
1979 BlockDriverState *bs;
1980 HBitmap *backup;
1981 bool prepared;
1982 bool was_enabled;
1983 } BlockDirtyBitmapState;
1984
1985 static void block_dirty_bitmap_add_prepare(BlkActionState *common,
1986 Error **errp)
1987 {
1988 Error *local_err = NULL;
1989 BlockDirtyBitmapAdd *action;
1990 BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
1991 common, common);
1992
1993 if (action_check_completion_mode(common, errp) < 0) {
1994 return;
1995 }
1996
1997 action = common->action->u.block_dirty_bitmap_add.data;
1998 /* AIO context taken and released within qmp_block_dirty_bitmap_add */
1999 qmp_block_dirty_bitmap_add(action->node, action->name,
2000 action->has_granularity, action->granularity,
2001 action->has_persistent, action->persistent,
2002 action->has_disabled, action->disabled,
2003 &local_err);
2004
2005 if (!local_err) {
2006 state->prepared = true;
2007 } else {
2008 error_propagate(errp, local_err);
2009 }
2010 }
2011
2012 static void block_dirty_bitmap_add_abort(BlkActionState *common)
2013 {
2014 BlockDirtyBitmapAdd *action;
2015 BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2016 common, common);
2017
2018 action = common->action->u.block_dirty_bitmap_add.data;
2019 /* Should not be able to fail: IF the bitmap was added via .prepare(),
2020 * then the node reference and bitmap name must have been valid.
2021 */
2022 if (state->prepared) {
2023 qmp_block_dirty_bitmap_remove(action->node, action->name, &error_abort);
2024 }
2025 }
2026
2027 static void block_dirty_bitmap_clear_prepare(BlkActionState *common,
2028 Error **errp)
2029 {
2030 BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2031 common, common);
2032 BlockDirtyBitmap *action;
2033
2034 if (action_check_completion_mode(common, errp) < 0) {
2035 return;
2036 }
2037
2038 action = common->action->u.block_dirty_bitmap_clear.data;
2039 state->bitmap = block_dirty_bitmap_lookup(action->node,
2040 action->name,
2041 &state->bs,
2042 errp);
2043 if (!state->bitmap) {
2044 return;
2045 }
2046
2047 if (bdrv_dirty_bitmap_check(state->bitmap, BDRV_BITMAP_DEFAULT, errp)) {
2048 return;
2049 }
2050
2051 bdrv_clear_dirty_bitmap(state->bitmap, &state->backup);
2052 }
2053
2054 static void block_dirty_bitmap_restore(BlkActionState *common)
2055 {
2056 BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2057 common, common);
2058
2059 if (state->backup) {
2060 bdrv_restore_dirty_bitmap(state->bitmap, state->backup);
2061 }
2062 }
2063
2064 static void block_dirty_bitmap_free_backup(BlkActionState *common)
2065 {
2066 BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2067 common, common);
2068
2069 hbitmap_free(state->backup);
2070 }
2071
2072 static void block_dirty_bitmap_enable_prepare(BlkActionState *common,
2073 Error **errp)
2074 {
2075 BlockDirtyBitmap *action;
2076 BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2077 common, common);
2078
2079 if (action_check_completion_mode(common, errp) < 0) {
2080 return;
2081 }
2082
2083 action = common->action->u.block_dirty_bitmap_enable.data;
2084 state->bitmap = block_dirty_bitmap_lookup(action->node,
2085 action->name,
2086 NULL,
2087 errp);
2088 if (!state->bitmap) {
2089 return;
2090 }
2091
2092 if (bdrv_dirty_bitmap_check(state->bitmap, BDRV_BITMAP_ALLOW_RO, errp)) {
2093 return;
2094 }
2095
2096 state->was_enabled = bdrv_dirty_bitmap_enabled(state->bitmap);
2097 bdrv_enable_dirty_bitmap(state->bitmap);
2098 }
2099
2100 static void block_dirty_bitmap_enable_abort(BlkActionState *common)
2101 {
2102 BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2103 common, common);
2104
2105 if (!state->was_enabled) {
2106 bdrv_disable_dirty_bitmap(state->bitmap);
2107 }
2108 }
2109
2110 static void block_dirty_bitmap_disable_prepare(BlkActionState *common,
2111 Error **errp)
2112 {
2113 BlockDirtyBitmap *action;
2114 BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2115 common, common);
2116
2117 if (action_check_completion_mode(common, errp) < 0) {
2118 return;
2119 }
2120
2121 action = common->action->u.block_dirty_bitmap_disable.data;
2122 state->bitmap = block_dirty_bitmap_lookup(action->node,
2123 action->name,
2124 NULL,
2125 errp);
2126 if (!state->bitmap) {
2127 return;
2128 }
2129
2130 if (bdrv_dirty_bitmap_check(state->bitmap, BDRV_BITMAP_ALLOW_RO, errp)) {
2131 return;
2132 }
2133
2134 state->was_enabled = bdrv_dirty_bitmap_enabled(state->bitmap);
2135 bdrv_disable_dirty_bitmap(state->bitmap);
2136 }
2137
2138 static void block_dirty_bitmap_disable_abort(BlkActionState *common)
2139 {
2140 BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2141 common, common);
2142
2143 if (state->was_enabled) {
2144 bdrv_enable_dirty_bitmap(state->bitmap);
2145 }
2146 }
2147
2148 static void block_dirty_bitmap_merge_prepare(BlkActionState *common,
2149 Error **errp)
2150 {
2151 BlockDirtyBitmapMerge *action;
2152 BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2153 common, common);
2154
2155 if (action_check_completion_mode(common, errp) < 0) {
2156 return;
2157 }
2158
2159 action = common->action->u.block_dirty_bitmap_merge.data;
2160
2161 state->bitmap = block_dirty_bitmap_merge(action->node, action->target,
2162 action->bitmaps, &state->backup,
2163 errp);
2164 }
2165
2166 static void block_dirty_bitmap_remove_prepare(BlkActionState *common,
2167 Error **errp)
2168 {
2169 BlockDirtyBitmap *action;
2170 BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2171 common, common);
2172
2173 if (action_check_completion_mode(common, errp) < 0) {
2174 return;
2175 }
2176
2177 action = common->action->u.block_dirty_bitmap_remove.data;
2178
2179 state->bitmap = block_dirty_bitmap_remove(action->node, action->name,
2180 false, &state->bs, errp);
2181 if (state->bitmap) {
2182 bdrv_dirty_bitmap_skip_store(state->bitmap, true);
2183 bdrv_dirty_bitmap_set_busy(state->bitmap, true);
2184 }
2185 }
2186
2187 static void block_dirty_bitmap_remove_abort(BlkActionState *common)
2188 {
2189 BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2190 common, common);
2191
2192 if (state->bitmap) {
2193 bdrv_dirty_bitmap_skip_store(state->bitmap, false);
2194 bdrv_dirty_bitmap_set_busy(state->bitmap, false);
2195 }
2196 }
2197
2198 static void block_dirty_bitmap_remove_commit(BlkActionState *common)
2199 {
2200 BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2201 common, common);
2202
2203 bdrv_dirty_bitmap_set_busy(state->bitmap, false);
2204 bdrv_release_dirty_bitmap(state->bitmap);
2205 }
2206
2207 static void abort_prepare(BlkActionState *common, Error **errp)
2208 {
2209 error_setg(errp, "Transaction aborted using Abort action");
2210 }
2211
2212 static void abort_commit(BlkActionState *common)
2213 {
2214 g_assert_not_reached(); /* this action never succeeds */
2215 }
2216
2217 static const BlkActionOps actions[] = {
2218 [TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT] = {
2219 .instance_size = sizeof(ExternalSnapshotState),
2220 .prepare = external_snapshot_prepare,
2221 .commit = external_snapshot_commit,
2222 .abort = external_snapshot_abort,
2223 .clean = external_snapshot_clean,
2224 },
2225 [TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC] = {
2226 .instance_size = sizeof(ExternalSnapshotState),
2227 .prepare = external_snapshot_prepare,
2228 .commit = external_snapshot_commit,
2229 .abort = external_snapshot_abort,
2230 .clean = external_snapshot_clean,
2231 },
2232 [TRANSACTION_ACTION_KIND_DRIVE_BACKUP] = {
2233 .instance_size = sizeof(DriveBackupState),
2234 .prepare = drive_backup_prepare,
2235 .commit = drive_backup_commit,
2236 .abort = drive_backup_abort,
2237 .clean = drive_backup_clean,
2238 },
2239 [TRANSACTION_ACTION_KIND_BLOCKDEV_BACKUP] = {
2240 .instance_size = sizeof(BlockdevBackupState),
2241 .prepare = blockdev_backup_prepare,
2242 .commit = blockdev_backup_commit,
2243 .abort = blockdev_backup_abort,
2244 .clean = blockdev_backup_clean,
2245 },
2246 [TRANSACTION_ACTION_KIND_ABORT] = {
2247 .instance_size = sizeof(BlkActionState),
2248 .prepare = abort_prepare,
2249 .commit = abort_commit,
2250 },
2251 [TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC] = {
2252 .instance_size = sizeof(InternalSnapshotState),
2253 .prepare = internal_snapshot_prepare,
2254 .abort = internal_snapshot_abort,
2255 .clean = internal_snapshot_clean,
2256 },
2257 [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_ADD] = {
2258 .instance_size = sizeof(BlockDirtyBitmapState),
2259 .prepare = block_dirty_bitmap_add_prepare,
2260 .abort = block_dirty_bitmap_add_abort,
2261 },
2262 [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_CLEAR] = {
2263 .instance_size = sizeof(BlockDirtyBitmapState),
2264 .prepare = block_dirty_bitmap_clear_prepare,
2265 .commit = block_dirty_bitmap_free_backup,
2266 .abort = block_dirty_bitmap_restore,
2267 },
2268 [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_ENABLE] = {
2269 .instance_size = sizeof(BlockDirtyBitmapState),
2270 .prepare = block_dirty_bitmap_enable_prepare,
2271 .abort = block_dirty_bitmap_enable_abort,
2272 },
2273 [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_DISABLE] = {
2274 .instance_size = sizeof(BlockDirtyBitmapState),
2275 .prepare = block_dirty_bitmap_disable_prepare,
2276 .abort = block_dirty_bitmap_disable_abort,
2277 },
2278 [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_MERGE] = {
2279 .instance_size = sizeof(BlockDirtyBitmapState),
2280 .prepare = block_dirty_bitmap_merge_prepare,
2281 .commit = block_dirty_bitmap_free_backup,
2282 .abort = block_dirty_bitmap_restore,
2283 },
2284 [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_REMOVE] = {
2285 .instance_size = sizeof(BlockDirtyBitmapState),
2286 .prepare = block_dirty_bitmap_remove_prepare,
2287 .commit = block_dirty_bitmap_remove_commit,
2288 .abort = block_dirty_bitmap_remove_abort,
2289 },
2290 /* Where are transactions for MIRROR, COMMIT and STREAM?
2291 * Although these blockjobs use transaction callbacks like the backup job,
2292 * these jobs do not necessarily adhere to transaction semantics.
2293 * These jobs may not fully undo all of their actions on abort, nor do they
2294 * necessarily work in transactions with more than one job in them.
2295 */
2296 };
2297
2298 /**
2299 * Allocate a TransactionProperties structure if necessary, and fill
2300 * that structure with desired defaults if they are unset.
2301 */
2302 static TransactionProperties *get_transaction_properties(
2303 TransactionProperties *props)
2304 {
2305 if (!props) {
2306 props = g_new0(TransactionProperties, 1);
2307 }
2308
2309 if (!props->has_completion_mode) {
2310 props->has_completion_mode = true;
2311 props->completion_mode = ACTION_COMPLETION_MODE_INDIVIDUAL;
2312 }
2313
2314 return props;
2315 }
2316
2317 /*
2318 * 'Atomic' group operations. The operations are performed as a set, and if
2319 * any fail then we roll back all operations in the group.
2320 *
2321 * Always run under BQL.
2322 */
2323 void qmp_transaction(TransactionActionList *dev_list,
2324 bool has_props,
2325 struct TransactionProperties *props,
2326 Error **errp)
2327 {
2328 TransactionActionList *dev_entry = dev_list;
2329 JobTxn *block_job_txn = NULL;
2330 BlkActionState *state, *next;
2331 Error *local_err = NULL;
2332
2333 GLOBAL_STATE_CODE();
2334
2335 QTAILQ_HEAD(, BlkActionState) snap_bdrv_states;
2336 QTAILQ_INIT(&snap_bdrv_states);
2337
2338 /* Does this transaction get canceled as a group on failure?
2339 * If not, we don't really need to make a JobTxn.
2340 */
2341 props = get_transaction_properties(props);
2342 if (props->completion_mode != ACTION_COMPLETION_MODE_INDIVIDUAL) {
2343 block_job_txn = job_txn_new();
2344 }
2345
2346 /* drain all i/o before any operations */
2347 bdrv_drain_all();
2348
2349 /* We don't do anything in this loop that commits us to the operations */
2350 while (NULL != dev_entry) {
2351 TransactionAction *dev_info = NULL;
2352 const BlkActionOps *ops;
2353
2354 dev_info = dev_entry->value;
2355 dev_entry = dev_entry->next;
2356
2357 assert(dev_info->type < ARRAY_SIZE(actions));
2358
2359 ops = &actions[dev_info->type];
2360 assert(ops->instance_size > 0);
2361
2362 state = g_malloc0(ops->instance_size);
2363 state->ops = ops;
2364 state->action = dev_info;
2365 state->block_job_txn = block_job_txn;
2366 state->txn_props = props;
2367 QTAILQ_INSERT_TAIL(&snap_bdrv_states, state, entry);
2368
2369 state->ops->prepare(state, &local_err);
2370 if (local_err) {
2371 error_propagate(errp, local_err);
2372 goto delete_and_fail;
2373 }
2374 }
2375
2376 QTAILQ_FOREACH(state, &snap_bdrv_states, entry) {
2377 if (state->ops->commit) {
2378 state->ops->commit(state);
2379 }
2380 }
2381
2382 /* success */
2383 goto exit;
2384
2385 delete_and_fail:
2386 /* failure, and it is all-or-none; roll back all operations */
2387 QTAILQ_FOREACH_REVERSE(state, &snap_bdrv_states, entry) {
2388 if (state->ops->abort) {
2389 state->ops->abort(state);
2390 }
2391 }
2392 exit:
2393 QTAILQ_FOREACH_SAFE(state, &snap_bdrv_states, entry, next) {
2394 if (state->ops->clean) {
2395 state->ops->clean(state);
2396 }
2397 g_free(state);
2398 }
2399 if (!has_props) {
2400 qapi_free_TransactionProperties(props);
2401 }
2402 job_txn_unref(block_job_txn);
2403 }
2404
2405 BlockDirtyBitmapSha256 *qmp_x_debug_block_dirty_bitmap_sha256(const char *node,
2406 const char *name,
2407 Error **errp)
2408 {
2409 BdrvDirtyBitmap *bitmap;
2410 BlockDriverState *bs;
2411 BlockDirtyBitmapSha256 *ret = NULL;
2412 char *sha256;
2413
2414 bitmap = block_dirty_bitmap_lookup(node, name, &bs, errp);
2415 if (!bitmap || !bs) {
2416 return NULL;
2417 }
2418
2419 sha256 = bdrv_dirty_bitmap_sha256(bitmap, errp);
2420 if (sha256 == NULL) {
2421 return NULL;
2422 }
2423
2424 ret = g_new(BlockDirtyBitmapSha256, 1);
2425 ret->sha256 = sha256;
2426
2427 return ret;
2428 }
2429
2430 void coroutine_fn qmp_block_resize(bool has_device, const char *device,
2431 bool has_node_name, const char *node_name,
2432 int64_t size, Error **errp)
2433 {
2434 Error *local_err = NULL;
2435 BlockBackend *blk;
2436 BlockDriverState *bs;
2437 AioContext *old_ctx;
2438
2439 bs = bdrv_lookup_bs(has_device ? device : NULL,
2440 has_node_name ? node_name : NULL,
2441 &local_err);
2442 if (local_err) {
2443 error_propagate(errp, local_err);
2444 return;
2445 }
2446
2447 if (size < 0) {
2448 error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "size", "a >0 size");
2449 return;
2450 }
2451
2452 if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_RESIZE, NULL)) {
2453 error_setg(errp, QERR_DEVICE_IN_USE, device);
2454 return;
2455 }
2456
2457 blk = blk_new_with_bs(bs, BLK_PERM_RESIZE, BLK_PERM_ALL, errp);
2458 if (!blk) {
2459 return;
2460 }
2461
2462 bdrv_co_lock(bs);
2463 bdrv_drained_begin(bs);
2464 bdrv_co_unlock(bs);
2465
2466 old_ctx = bdrv_co_enter(bs);
2467 blk_truncate(blk, size, false, PREALLOC_MODE_OFF, 0, errp);
2468 bdrv_co_leave(bs, old_ctx);
2469
2470 bdrv_co_lock(bs);
2471 bdrv_drained_end(bs);
2472 blk_unref(blk);
2473 bdrv_co_unlock(bs);
2474 }
2475
2476 void qmp_block_stream(bool has_job_id, const char *job_id, const char *device,
2477 bool has_base, const char *base,
2478 bool has_base_node, const char *base_node,
2479 bool has_backing_file, const char *backing_file,
2480 bool has_bottom, const char *bottom,
2481 bool has_speed, int64_t speed,
2482 bool has_on_error, BlockdevOnError on_error,
2483 bool has_filter_node_name, const char *filter_node_name,
2484 bool has_auto_finalize, bool auto_finalize,
2485 bool has_auto_dismiss, bool auto_dismiss,
2486 Error **errp)
2487 {
2488 BlockDriverState *bs, *iter, *iter_end;
2489 BlockDriverState *base_bs = NULL;
2490 BlockDriverState *bottom_bs = NULL;
2491 AioContext *aio_context;
2492 Error *local_err = NULL;
2493 int job_flags = JOB_DEFAULT;
2494
2495 if (has_base && has_base_node) {
2496 error_setg(errp, "'base' and 'base-node' cannot be specified "
2497 "at the same time");
2498 return;
2499 }
2500
2501 if (has_base && has_bottom) {
2502 error_setg(errp, "'base' and 'bottom' cannot be specified "
2503 "at the same time");
2504 return;
2505 }
2506
2507 if (has_bottom && has_base_node) {
2508 error_setg(errp, "'bottom' and 'base-node' cannot be specified "
2509 "at the same time");
2510 return;
2511 }
2512
2513 if (!has_on_error) {
2514 on_error = BLOCKDEV_ON_ERROR_REPORT;
2515 }
2516
2517 bs = bdrv_lookup_bs(device, device, errp);
2518 if (!bs) {
2519 return;
2520 }
2521
2522 aio_context = bdrv_get_aio_context(bs);
2523 aio_context_acquire(aio_context);
2524
2525 if (has_base) {
2526 base_bs = bdrv_find_backing_image(bs, base);
2527 if (base_bs == NULL) {
2528 error_setg(errp, "Can't find '%s' in the backing chain", base);
2529 goto out;
2530 }
2531 assert(bdrv_get_aio_context(base_bs) == aio_context);
2532 }
2533
2534 if (has_base_node) {
2535 base_bs = bdrv_lookup_bs(NULL, base_node, errp);
2536 if (!base_bs) {
2537 goto out;
2538 }
2539 if (bs == base_bs || !bdrv_chain_contains(bs, base_bs)) {
2540 error_setg(errp, "Node '%s' is not a backing image of '%s'",
2541 base_node, device);
2542 goto out;
2543 }
2544 assert(bdrv_get_aio_context(base_bs) == aio_context);
2545 bdrv_refresh_filename(base_bs);
2546 }
2547
2548 if (has_bottom) {
2549 bottom_bs = bdrv_lookup_bs(NULL, bottom, errp);
2550 if (!bottom_bs) {
2551 goto out;
2552 }
2553 if (!bottom_bs->drv) {
2554 error_setg(errp, "Node '%s' is not open", bottom);
2555 goto out;
2556 }
2557 if (bottom_bs->drv->is_filter) {
2558 error_setg(errp, "Node '%s' is a filter, use a non-filter node "
2559 "as 'bottom'", bottom);
2560 goto out;
2561 }
2562 if (!bdrv_chain_contains(bs, bottom_bs)) {
2563 error_setg(errp, "Node '%s' is not in a chain starting from '%s'",
2564 bottom, device);
2565 goto out;
2566 }
2567 assert(bdrv_get_aio_context(bottom_bs) == aio_context);
2568 }
2569
2570 /*
2571 * Check for op blockers in the whole chain between bs and base (or bottom)
2572 */
2573 iter_end = has_bottom ? bdrv_filter_or_cow_bs(bottom_bs) : base_bs;
2574 for (iter = bs; iter && iter != iter_end;
2575 iter = bdrv_filter_or_cow_bs(iter))
2576 {
2577 if (bdrv_op_is_blocked(iter, BLOCK_OP_TYPE_STREAM, errp)) {
2578 goto out;
2579 }
2580 }
2581
2582 /* if we are streaming the entire chain, the result will have no backing
2583 * file, and specifying one is therefore an error */
2584 if (base_bs == NULL && has_backing_file) {
2585 error_setg(errp, "backing file specified, but streaming the "
2586 "entire chain");
2587 goto out;
2588 }
2589
2590 if (has_auto_finalize && !auto_finalize) {
2591 job_flags |= JOB_MANUAL_FINALIZE;
2592 }
2593 if (has_auto_dismiss && !auto_dismiss) {
2594 job_flags |= JOB_MANUAL_DISMISS;
2595 }
2596
2597 stream_start(has_job_id ? job_id : NULL, bs, base_bs, backing_file,
2598 bottom_bs, job_flags, has_speed ? speed : 0, on_error,
2599 filter_node_name, &local_err);
2600 if (local_err) {
2601 error_propagate(errp, local_err);
2602 goto out;
2603 }
2604
2605 trace_qmp_block_stream(bs);
2606
2607 out:
2608 aio_context_release(aio_context);
2609 }
2610
2611 void qmp_block_commit(bool has_job_id, const char *job_id, const char *device,
2612 bool has_base_node, const char *base_node,
2613 bool has_base, const char *base,
2614 bool has_top_node, const char *top_node,
2615 bool has_top, const char *top,
2616 bool has_backing_file, const char *backing_file,
2617 bool has_speed, int64_t speed,
2618 bool has_on_error, BlockdevOnError on_error,
2619 bool has_filter_node_name, const char *filter_node_name,
2620 bool has_auto_finalize, bool auto_finalize,
2621 bool has_auto_dismiss, bool auto_dismiss,
2622 Error **errp)
2623 {
2624 BlockDriverState *bs;
2625 BlockDriverState *iter;
2626 BlockDriverState *base_bs, *top_bs;
2627 AioContext *aio_context;
2628 Error *local_err = NULL;
2629 int job_flags = JOB_DEFAULT;
2630 uint64_t top_perm, top_shared;
2631
2632 if (!has_speed) {
2633 speed = 0;
2634 }
2635 if (!has_on_error) {
2636 on_error = BLOCKDEV_ON_ERROR_REPORT;
2637 }
2638 if (!has_filter_node_name) {
2639 filter_node_name = NULL;
2640 }
2641 if (has_auto_finalize && !auto_finalize) {
2642 job_flags |= JOB_MANUAL_FINALIZE;
2643 }
2644 if (has_auto_dismiss && !auto_dismiss) {
2645 job_flags |= JOB_MANUAL_DISMISS;
2646 }
2647
2648 /* Important Note:
2649 * libvirt relies on the DeviceNotFound error class in order to probe for
2650 * live commit feature versions; for this to work, we must make sure to
2651 * perform the device lookup before any generic errors that may occur in a
2652 * scenario in which all optional arguments are omitted. */
2653 bs = qmp_get_root_bs(device, &local_err);
2654 if (!bs) {
2655 bs = bdrv_lookup_bs(device, device, NULL);
2656 if (!bs) {
2657 error_free(local_err);
2658 error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
2659 "Device '%s' not found", device);
2660 } else {
2661 error_propagate(errp, local_err);
2662 }
2663 return;
2664 }
2665
2666 aio_context = bdrv_get_aio_context(bs);
2667 aio_context_acquire(aio_context);
2668
2669 if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_COMMIT_SOURCE, errp)) {
2670 goto out;
2671 }
2672
2673 /* default top_bs is the active layer */
2674 top_bs = bs;
2675
2676 if (has_top_node && has_top) {
2677 error_setg(errp, "'top-node' and 'top' are mutually exclusive");
2678 goto out;
2679 } else if (has_top_node) {
2680 top_bs = bdrv_lookup_bs(NULL, top_node, errp);
2681 if (top_bs == NULL) {
2682 goto out;
2683 }
2684 if (!bdrv_chain_contains(bs, top_bs)) {
2685 error_setg(errp, "'%s' is not in this backing file chain",
2686 top_node);
2687 goto out;
2688 }
2689 } else if (has_top && top) {
2690 /* This strcmp() is just a shortcut, there is no need to
2691 * refresh @bs's filename. If it mismatches,
2692 * bdrv_find_backing_image() will do the refresh and may still
2693 * return @bs. */
2694 if (strcmp(bs->filename, top) != 0) {
2695 top_bs = bdrv_find_backing_image(bs, top);
2696 }
2697 }
2698
2699 if (top_bs == NULL) {
2700 error_setg(errp, "Top image file %s not found", top ? top : "NULL");
2701 goto out;
2702 }
2703
2704 assert(bdrv_get_aio_context(top_bs) == aio_context);
2705
2706 if (has_base_node && has_base) {
2707 error_setg(errp, "'base-node' and 'base' are mutually exclusive");
2708 goto out;
2709 } else if (has_base_node) {
2710 base_bs = bdrv_lookup_bs(NULL, base_node, errp);
2711 if (base_bs == NULL) {
2712 goto out;
2713 }
2714 if (!bdrv_chain_contains(top_bs, base_bs)) {
2715 error_setg(errp, "'%s' is not in this backing file chain",
2716 base_node);
2717 goto out;
2718 }
2719 } else if (has_base && base) {
2720 base_bs = bdrv_find_backing_image(top_bs, base);
2721 if (base_bs == NULL) {
2722 error_setg(errp, "Can't find '%s' in the backing chain", base);
2723 goto out;
2724 }
2725 } else {
2726 base_bs = bdrv_find_base(top_bs);
2727 if (base_bs == NULL) {
2728 error_setg(errp, "There is no backimg image");
2729 goto out;
2730 }
2731 }
2732
2733 assert(bdrv_get_aio_context(base_bs) == aio_context);
2734
2735 for (iter = top_bs; iter != bdrv_filter_or_cow_bs(base_bs);
2736 iter = bdrv_filter_or_cow_bs(iter))
2737 {
2738 if (bdrv_op_is_blocked(iter, BLOCK_OP_TYPE_COMMIT_TARGET, errp)) {
2739 goto out;
2740 }
2741 }
2742
2743 /* Do not allow attempts to commit an image into itself */
2744 if (top_bs == base_bs) {
2745 error_setg(errp, "cannot commit an image into itself");
2746 goto out;
2747 }
2748
2749 /*
2750 * Active commit is required if and only if someone has taken a
2751 * WRITE permission on the top node. Historically, we have always
2752 * used active commit for top nodes, so continue that practice
2753 * lest we possibly break clients that rely on this behavior, e.g.
2754 * to later attach this node to a writing parent.
2755 * (Active commit is never really wrong.)
2756 */
2757 bdrv_get_cumulative_perm(top_bs, &top_perm, &top_shared);
2758 if (top_perm & BLK_PERM_WRITE ||
2759 bdrv_skip_filters(top_bs) == bdrv_skip_filters(bs))
2760 {
2761 if (has_backing_file) {
2762 if (bdrv_skip_filters(top_bs) == bdrv_skip_filters(bs)) {
2763 error_setg(errp, "'backing-file' specified,"
2764 " but 'top' is the active layer");
2765 } else {
2766 error_setg(errp, "'backing-file' specified, but 'top' has a "
2767 "writer on it");
2768 }
2769 goto out;
2770 }
2771 if (!has_job_id) {
2772 /*
2773 * Emulate here what block_job_create() does, because it
2774 * is possible that @bs != @top_bs (the block job should
2775 * be named after @bs, even if @top_bs is the actual
2776 * source)
2777 */
2778 job_id = bdrv_get_device_name(bs);
2779 }
2780 commit_active_start(job_id, top_bs, base_bs, job_flags, speed, on_error,
2781 filter_node_name, NULL, NULL, false, &local_err);
2782 } else {
2783 BlockDriverState *overlay_bs = bdrv_find_overlay(bs, top_bs);
2784 if (bdrv_op_is_blocked(overlay_bs, BLOCK_OP_TYPE_COMMIT_TARGET, errp)) {
2785 goto out;
2786 }
2787 commit_start(has_job_id ? job_id : NULL, bs, base_bs, top_bs, job_flags,
2788 speed, on_error, has_backing_file ? backing_file : NULL,
2789 filter_node_name, &local_err);
2790 }
2791 if (local_err != NULL) {
2792 error_propagate(errp, local_err);
2793 goto out;
2794 }
2795
2796 out:
2797 aio_context_release(aio_context);
2798 }
2799
2800 /* Common QMP interface for drive-backup and blockdev-backup */
2801 static BlockJob *do_backup_common(BackupCommon *backup,
2802 BlockDriverState *bs,
2803 BlockDriverState *target_bs,
2804 AioContext *aio_context,
2805 JobTxn *txn, Error **errp)
2806 {
2807 BlockJob *job = NULL;
2808 BdrvDirtyBitmap *bmap = NULL;
2809 BackupPerf perf = { .max_workers = 64 };
2810 int job_flags = JOB_DEFAULT;
2811
2812 if (!backup->has_speed) {
2813 backup->speed = 0;
2814 }
2815 if (!backup->has_on_source_error) {
2816 backup->on_source_error = BLOCKDEV_ON_ERROR_REPORT;
2817 }
2818 if (!backup->has_on_target_error) {
2819 backup->on_target_error = BLOCKDEV_ON_ERROR_REPORT;
2820 }
2821 if (!backup->has_job_id) {
2822 backup->job_id = NULL;
2823 }
2824 if (!backup->has_auto_finalize) {
2825 backup->auto_finalize = true;
2826 }
2827 if (!backup->has_auto_dismiss) {
2828 backup->auto_dismiss = true;
2829 }
2830 if (!backup->has_compress) {
2831 backup->compress = false;
2832 }
2833
2834 if (backup->x_perf) {
2835 if (backup->x_perf->has_use_copy_range) {
2836 perf.use_copy_range = backup->x_perf->use_copy_range;
2837 }
2838 if (backup->x_perf->has_max_workers) {
2839 perf.max_workers = backup->x_perf->max_workers;
2840 }
2841 if (backup->x_perf->has_max_chunk) {
2842 perf.max_chunk = backup->x_perf->max_chunk;
2843 }
2844 }
2845
2846 if ((backup->sync == MIRROR_SYNC_MODE_BITMAP) ||
2847 (backup->sync == MIRROR_SYNC_MODE_INCREMENTAL)) {
2848 /* done before desugaring 'incremental' to print the right message */
2849 if (!backup->has_bitmap) {
2850 error_setg(errp, "must provide a valid bitmap name for "
2851 "'%s' sync mode", MirrorSyncMode_str(backup->sync));
2852 return NULL;
2853 }
2854 }
2855
2856 if (backup->sync == MIRROR_SYNC_MODE_INCREMENTAL) {
2857 if (backup->has_bitmap_mode &&
2858 backup->bitmap_mode != BITMAP_SYNC_MODE_ON_SUCCESS) {
2859 error_setg(errp, "Bitmap sync mode must be '%s' "
2860 "when using sync mode '%s'",
2861 BitmapSyncMode_str(BITMAP_SYNC_MODE_ON_SUCCESS),
2862 MirrorSyncMode_str(backup->sync));
2863 return NULL;
2864 }
2865 backup->has_bitmap_mode = true;
2866 backup->sync = MIRROR_SYNC_MODE_BITMAP;
2867 backup->bitmap_mode = BITMAP_SYNC_MODE_ON_SUCCESS;
2868 }
2869
2870 if (backup->has_bitmap) {
2871 bmap = bdrv_find_dirty_bitmap(bs, backup->bitmap);
2872 if (!bmap) {
2873 error_setg(errp, "Bitmap '%s' could not be found", backup->bitmap);
2874 return NULL;
2875 }
2876 if (!backup->has_bitmap_mode) {
2877 error_setg(errp, "Bitmap sync mode must be given "
2878 "when providing a bitmap");
2879 return NULL;
2880 }
2881 if (bdrv_dirty_bitmap_check(bmap, BDRV_BITMAP_ALLOW_RO, errp)) {
2882 return NULL;
2883 }
2884
2885 /* This does not produce a useful bitmap artifact: */
2886 if (backup->sync == MIRROR_SYNC_MODE_NONE) {
2887 error_setg(errp, "sync mode '%s' does not produce meaningful bitmap"
2888 " outputs", MirrorSyncMode_str(backup->sync));
2889 return NULL;
2890 }
2891
2892 /* If the bitmap isn't used for input or output, this is useless: */
2893 if (backup->bitmap_mode == BITMAP_SYNC_MODE_NEVER &&
2894 backup->sync != MIRROR_SYNC_MODE_BITMAP) {
2895 error_setg(errp, "Bitmap sync mode '%s' has no meaningful effect"
2896 " when combined with sync mode '%s'",
2897 BitmapSyncMode_str(backup->bitmap_mode),
2898 MirrorSyncMode_str(backup->sync));
2899 return NULL;
2900 }
2901 }
2902
2903 if (!backup->has_bitmap && backup->has_bitmap_mode) {
2904 error_setg(errp, "Cannot specify bitmap sync mode without a bitmap");
2905 return NULL;
2906 }
2907
2908 if (!backup->auto_finalize) {
2909 job_flags |= JOB_MANUAL_FINALIZE;
2910 }
2911 if (!backup->auto_dismiss) {
2912 job_flags |= JOB_MANUAL_DISMISS;
2913 }
2914
2915 job = backup_job_create(backup->job_id, bs, target_bs, backup->speed,
2916 backup->sync, bmap, backup->bitmap_mode,
2917 backup->compress,
2918 backup->filter_node_name,
2919 &perf,
2920 backup->on_source_error,
2921 backup->on_target_error,
2922 job_flags, NULL, NULL, txn, errp);
2923 return job;
2924 }
2925
2926 void qmp_drive_backup(DriveBackup *backup, Error **errp)
2927 {
2928 TransactionAction action = {
2929 .type = TRANSACTION_ACTION_KIND_DRIVE_BACKUP,
2930 .u.drive_backup.data = backup,
2931 };
2932 blockdev_do_action(&action, errp);
2933 }
2934
2935 BlockDeviceInfoList *qmp_query_named_block_nodes(bool has_flat,
2936 bool flat,
2937 Error **errp)
2938 {
2939 bool return_flat = has_flat && flat;
2940
2941 return bdrv_named_nodes_list(return_flat, errp);
2942 }
2943
2944 XDbgBlockGraph *qmp_x_debug_query_block_graph(Error **errp)
2945 {
2946 return bdrv_get_xdbg_block_graph(errp);
2947 }
2948
2949 void qmp_blockdev_backup(BlockdevBackup *backup, Error **errp)
2950 {
2951 TransactionAction action = {
2952 .type = TRANSACTION_ACTION_KIND_BLOCKDEV_BACKUP,
2953 .u.blockdev_backup.data = backup,
2954 };
2955 blockdev_do_action(&action, errp);
2956 }
2957
2958 /* Parameter check and block job starting for drive mirroring.
2959 * Caller should hold @device and @target's aio context (must be the same).
2960 **/
2961 static void blockdev_mirror_common(const char *job_id, BlockDriverState *bs,
2962 BlockDriverState *target,
2963 bool has_replaces, const char *replaces,
2964 enum MirrorSyncMode sync,
2965 BlockMirrorBackingMode backing_mode,
2966 bool zero_target,
2967 bool has_speed, int64_t speed,
2968 bool has_granularity, uint32_t granularity,
2969 bool has_buf_size, int64_t buf_size,
2970 bool has_on_source_error,
2971 BlockdevOnError on_source_error,
2972 bool has_on_target_error,
2973 BlockdevOnError on_target_error,
2974 bool has_unmap, bool unmap,
2975 bool has_filter_node_name,
2976 const char *filter_node_name,
2977 bool has_copy_mode, MirrorCopyMode copy_mode,
2978 bool has_auto_finalize, bool auto_finalize,
2979 bool has_auto_dismiss, bool auto_dismiss,
2980 Error **errp)
2981 {
2982 BlockDriverState *unfiltered_bs;
2983 int job_flags = JOB_DEFAULT;
2984
2985 if (!has_speed) {
2986 speed = 0;
2987 }
2988 if (!has_on_source_error) {
2989 on_source_error = BLOCKDEV_ON_ERROR_REPORT;
2990 }
2991 if (!has_on_target_error) {
2992 on_target_error = BLOCKDEV_ON_ERROR_REPORT;
2993 }
2994 if (!has_granularity) {
2995 granularity = 0;
2996 }
2997 if (!has_buf_size) {
2998 buf_size = 0;
2999 }
3000 if (!has_unmap) {
3001 unmap = true;
3002 }
3003 if (!has_filter_node_name) {
3004 filter_node_name = NULL;
3005 }
3006 if (!has_copy_mode) {
3007 copy_mode = MIRROR_COPY_MODE_BACKGROUND;
3008 }
3009 if (has_auto_finalize && !auto_finalize) {
3010 job_flags |= JOB_MANUAL_FINALIZE;
3011 }
3012 if (has_auto_dismiss && !auto_dismiss) {
3013 job_flags |= JOB_MANUAL_DISMISS;
3014 }
3015
3016 if (granularity != 0 && (granularity < 512 || granularity > 1048576 * 64)) {
3017 error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "granularity",
3018 "a value in range [512B, 64MB]");
3019 return;
3020 }
3021 if (granularity & (granularity - 1)) {
3022 error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "granularity",
3023 "a power of 2");
3024 return;
3025 }
3026
3027 if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_MIRROR_SOURCE, errp)) {
3028 return;
3029 }
3030 if (bdrv_op_is_blocked(target, BLOCK_OP_TYPE_MIRROR_TARGET, errp)) {
3031 return;
3032 }
3033
3034 if (!bdrv_backing_chain_next(bs) && sync == MIRROR_SYNC_MODE_TOP) {
3035 sync = MIRROR_SYNC_MODE_FULL;
3036 }
3037
3038 if (!has_replaces) {
3039 /* We want to mirror from @bs, but keep implicit filters on top */
3040 unfiltered_bs = bdrv_skip_implicit_filters(bs);
3041 if (unfiltered_bs != bs) {
3042 replaces = unfiltered_bs->node_name;
3043 has_replaces = true;
3044 }
3045 }
3046
3047 if (has_replaces) {
3048 BlockDriverState *to_replace_bs;
3049 AioContext *replace_aio_context;
3050 int64_t bs_size, replace_size;
3051
3052 bs_size = bdrv_getlength(bs);
3053 if (bs_size < 0) {
3054 error_setg_errno(errp, -bs_size, "Failed to query device's size");
3055 return;
3056 }
3057
3058 to_replace_bs = check_to_replace_node(bs, replaces, errp);
3059 if (!to_replace_bs) {
3060 return;
3061 }
3062
3063 replace_aio_context = bdrv_get_aio_context(to_replace_bs);
3064 aio_context_acquire(replace_aio_context);
3065 replace_size = bdrv_getlength(to_replace_bs);
3066 aio_context_release(replace_aio_context);
3067
3068 if (replace_size < 0) {
3069 error_setg_errno(errp, -replace_size,
3070 "Failed to query the replacement node's size");
3071 return;
3072 }
3073 if (bs_size != replace_size) {
3074 error_setg(errp, "cannot replace image with a mirror image of "
3075 "different size");
3076 return;
3077 }
3078 }
3079
3080 /* pass the node name to replace to mirror start since it's loose coupling
3081 * and will allow to check whether the node still exist at mirror completion
3082 */
3083 mirror_start(job_id, bs, target,
3084 has_replaces ? replaces : NULL, job_flags,
3085 speed, granularity, buf_size, sync, backing_mode, zero_target,
3086 on_source_error, on_target_error, unmap, filter_node_name,
3087 copy_mode, errp);
3088 }
3089
3090 void qmp_drive_mirror(DriveMirror *arg, Error **errp)
3091 {
3092 BlockDriverState *bs;
3093 BlockDriverState *target_backing_bs, *target_bs;
3094 AioContext *aio_context;
3095 AioContext *old_context;
3096 BlockMirrorBackingMode backing_mode;
3097 Error *local_err = NULL;
3098 QDict *options = NULL;
3099 int flags;
3100 int64_t size;
3101 const char *format = arg->format;
3102 bool zero_target;
3103 int ret;
3104
3105 bs = qmp_get_root_bs(arg->device, errp);
3106 if (!bs) {
3107 return;
3108 }
3109
3110 /* Early check to avoid creating target */
3111 if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_MIRROR_SOURCE, errp)) {
3112 return;
3113 }
3114
3115 aio_context = bdrv_get_aio_context(bs);
3116 aio_context_acquire(aio_context);
3117
3118 if (!arg->has_mode) {
3119 arg->mode = NEW_IMAGE_MODE_ABSOLUTE_PATHS;
3120 }
3121
3122 if (!arg->has_format) {
3123 format = (arg->mode == NEW_IMAGE_MODE_EXISTING
3124 ? NULL : bs->drv->format_name);
3125 }
3126
3127 flags = bs->open_flags | BDRV_O_RDWR;
3128 target_backing_bs = bdrv_cow_bs(bdrv_skip_filters(bs));
3129 if (!target_backing_bs && arg->sync == MIRROR_SYNC_MODE_TOP) {
3130 arg->sync = MIRROR_SYNC_MODE_FULL;
3131 }
3132 if (arg->sync == MIRROR_SYNC_MODE_NONE) {
3133 target_backing_bs = bs;
3134 }
3135
3136 size = bdrv_getlength(bs);
3137 if (size < 0) {
3138 error_setg_errno(errp, -size, "bdrv_getlength failed");
3139 goto out;
3140 }
3141
3142 if (arg->has_replaces) {
3143 if (!arg->has_node_name) {
3144 error_setg(errp, "a node-name must be provided when replacing a"
3145 " named node of the graph");
3146 goto out;
3147 }
3148 }
3149
3150 if (arg->mode == NEW_IMAGE_MODE_ABSOLUTE_PATHS) {
3151 backing_mode = MIRROR_SOURCE_BACKING_CHAIN;
3152 } else {
3153 backing_mode = MIRROR_OPEN_BACKING_CHAIN;
3154 }
3155
3156 /* Don't open backing image in create() */
3157 flags |= BDRV_O_NO_BACKING;
3158
3159 if ((arg->sync == MIRROR_SYNC_MODE_FULL || !target_backing_bs)
3160 && arg->mode != NEW_IMAGE_MODE_EXISTING)
3161 {
3162 /* create new image w/o backing file */
3163 assert(format);
3164 bdrv_img_create(arg->target, format,
3165 NULL, NULL, NULL, size, flags, false, &local_err);
3166 } else {
3167 /* Implicit filters should not appear in the filename */
3168 BlockDriverState *explicit_backing =
3169 bdrv_skip_implicit_filters(target_backing_bs);
3170
3171 switch (arg->mode) {
3172 case NEW_IMAGE_MODE_EXISTING:
3173 break;
3174 case NEW_IMAGE_MODE_ABSOLUTE_PATHS:
3175 /* create new image with backing file */
3176 bdrv_refresh_filename(explicit_backing);
3177 bdrv_img_create(arg->target, format,
3178 explicit_backing->filename,
3179 explicit_backing->drv->format_name,
3180 NULL, size, flags, false, &local_err);
3181 break;
3182 default:
3183 abort();
3184 }
3185 }
3186
3187 if (local_err) {
3188 error_propagate(errp, local_err);
3189 goto out;
3190 }
3191
3192 options = qdict_new();
3193 if (arg->has_node_name) {
3194 qdict_put_str(options, "node-name", arg->node_name);
3195 }
3196 if (format) {
3197 qdict_put_str(options, "driver", format);
3198 }
3199
3200 /* Mirroring takes care of copy-on-write using the source's backing
3201 * file.
3202 */
3203 target_bs = bdrv_open(arg->target, NULL, options, flags, errp);
3204 if (!target_bs) {
3205 goto out;
3206 }
3207
3208 zero_target = (arg->sync == MIRROR_SYNC_MODE_FULL &&
3209 (arg->mode == NEW_IMAGE_MODE_EXISTING ||
3210 !bdrv_has_zero_init(target_bs)));
3211
3212
3213 /* Honor bdrv_try_set_aio_context() context acquisition requirements. */
3214 old_context = bdrv_get_aio_context(target_bs);
3215 aio_context_release(aio_context);
3216 aio_context_acquire(old_context);
3217
3218 ret = bdrv_try_set_aio_context(target_bs, aio_context, errp);
3219 if (ret < 0) {
3220 bdrv_unref(target_bs);
3221 aio_context_release(old_context);
3222 return;
3223 }
3224
3225 aio_context_release(old_context);
3226 aio_context_acquire(aio_context);
3227
3228 blockdev_mirror_common(arg->has_job_id ? arg->job_id : NULL, bs, target_bs,
3229 arg->has_replaces, arg->replaces, arg->sync,
3230 backing_mode, zero_target,
3231 arg->has_speed, arg->speed,
3232 arg->has_granularity, arg->granularity,
3233 arg->has_buf_size, arg->buf_size,
3234 arg->has_on_source_error, arg->on_source_error,
3235 arg->has_on_target_error, arg->on_target_error,
3236 arg->has_unmap, arg->unmap,
3237 false, NULL,
3238 arg->has_copy_mode, arg->copy_mode,
3239 arg->has_auto_finalize, arg->auto_finalize,
3240 arg->has_auto_dismiss, arg->auto_dismiss,
3241 errp);
3242 bdrv_unref(target_bs);
3243 out:
3244 aio_context_release(aio_context);
3245 }
3246
3247 void qmp_blockdev_mirror(bool has_job_id, const char *job_id,
3248 const char *device, const char *target,
3249 bool has_replaces, const char *replaces,
3250 MirrorSyncMode sync,
3251 bool has_speed, int64_t speed,
3252 bool has_granularity, uint32_t granularity,
3253 bool has_buf_size, int64_t buf_size,
3254 bool has_on_source_error,
3255 BlockdevOnError on_source_error,
3256 bool has_on_target_error,
3257 BlockdevOnError on_target_error,
3258 bool has_filter_node_name,
3259 const char *filter_node_name,
3260 bool has_copy_mode, MirrorCopyMode copy_mode,
3261 bool has_auto_finalize, bool auto_finalize,
3262 bool has_auto_dismiss, bool auto_dismiss,
3263 Error **errp)
3264 {
3265 BlockDriverState *bs;
3266 BlockDriverState *target_bs;
3267 AioContext *aio_context;
3268 AioContext *old_context;
3269 BlockMirrorBackingMode backing_mode = MIRROR_LEAVE_BACKING_CHAIN;
3270 bool zero_target;
3271 int ret;
3272
3273 bs = qmp_get_root_bs(device, errp);
3274 if (!bs) {
3275 return;
3276 }
3277
3278 target_bs = bdrv_lookup_bs(target, target, errp);
3279 if (!target_bs) {
3280 return;
3281 }
3282
3283 zero_target = (sync == MIRROR_SYNC_MODE_FULL);
3284
3285 /* Honor bdrv_try_set_aio_context() context acquisition requirements. */
3286 old_context = bdrv_get_aio_context(target_bs);
3287 aio_context = bdrv_get_aio_context(bs);
3288 aio_context_acquire(old_context);
3289
3290 ret = bdrv_try_set_aio_context(target_bs, aio_context, errp);
3291
3292 aio_context_release(old_context);
3293 aio_context_acquire(aio_context);
3294
3295 if (ret < 0) {
3296 goto out;
3297 }
3298
3299 blockdev_mirror_common(has_job_id ? job_id : NULL, bs, target_bs,
3300 has_replaces, replaces, sync, backing_mode,
3301 zero_target, has_speed, speed,
3302 has_granularity, granularity,
3303 has_buf_size, buf_size,
3304 has_on_source_error, on_source_error,
3305 has_on_target_error, on_target_error,
3306 true, true,
3307 has_filter_node_name, filter_node_name,
3308 has_copy_mode, copy_mode,
3309 has_auto_finalize, auto_finalize,
3310 has_auto_dismiss, auto_dismiss,
3311 errp);
3312 out:
3313 aio_context_release(aio_context);
3314 }
3315
3316 /* Get a block job using its ID and acquire its AioContext */
3317 static BlockJob *find_block_job(const char *id, AioContext **aio_context,
3318 Error **errp)
3319 {
3320 BlockJob *job;
3321
3322 assert(id != NULL);
3323
3324 *aio_context = NULL;
3325
3326 job = block_job_get(id);
3327
3328 if (!job) {
3329 error_set(errp, ERROR_CLASS_DEVICE_NOT_ACTIVE,
3330 "Block job '%s' not found", id);
3331 return NULL;
3332 }
3333
3334 *aio_context = block_job_get_aio_context(job);
3335 aio_context_acquire(*aio_context);
3336
3337 return job;
3338 }
3339
3340 void qmp_block_job_set_speed(const char *device, int64_t speed, Error **errp)
3341 {
3342 AioContext *aio_context;
3343 BlockJob *job = find_block_job(device, &aio_context, errp);
3344
3345 if (!job) {
3346 return;
3347 }
3348
3349 block_job_set_speed(job, speed, errp);
3350 aio_context_release(aio_context);
3351 }
3352
3353 void qmp_block_job_cancel(const char *device,
3354 bool has_force, bool force, Error **errp)
3355 {
3356 AioContext *aio_context;
3357 BlockJob *job = find_block_job(device, &aio_context, errp);
3358
3359 if (!job) {
3360 return;
3361 }
3362
3363 if (!has_force) {
3364 force = false;
3365 }
3366
3367 if (job_user_paused(&job->job) && !force) {
3368 error_setg(errp, "The block job for device '%s' is currently paused",
3369 device);
3370 goto out;
3371 }
3372
3373 trace_qmp_block_job_cancel(job);
3374 job_user_cancel(&job->job, force, errp);
3375 out:
3376 aio_context_release(aio_context);
3377 }
3378
3379 void qmp_block_job_pause(const char *device, Error **errp)
3380 {
3381 AioContext *aio_context;
3382 BlockJob *job = find_block_job(device, &aio_context, errp);
3383
3384 if (!job) {
3385 return;
3386 }
3387
3388 trace_qmp_block_job_pause(job);
3389 job_user_pause(&job->job, errp);
3390 aio_context_release(aio_context);
3391 }
3392
3393 void qmp_block_job_resume(const char *device, Error **errp)
3394 {
3395 AioContext *aio_context;
3396 BlockJob *job = find_block_job(device, &aio_context, errp);
3397
3398 if (!job) {
3399 return;
3400 }
3401
3402 trace_qmp_block_job_resume(job);
3403 job_user_resume(&job->job, errp);
3404 aio_context_release(aio_context);
3405 }
3406
3407 void qmp_block_job_complete(const char *device, Error **errp)
3408 {
3409 AioContext *aio_context;
3410 BlockJob *job = find_block_job(device, &aio_context, errp);
3411
3412 if (!job) {
3413 return;
3414 }
3415
3416 trace_qmp_block_job_complete(job);
3417 job_complete(&job->job, errp);
3418 aio_context_release(aio_context);
3419 }
3420
3421 void qmp_block_job_finalize(const char *id, Error **errp)
3422 {
3423 AioContext *aio_context;
3424 BlockJob *job = find_block_job(id, &aio_context, errp);
3425
3426 if (!job) {
3427 return;
3428 }
3429
3430 trace_qmp_block_job_finalize(job);
3431 job_ref(&job->job);
3432 job_finalize(&job->job, errp);
3433
3434 /*
3435 * Job's context might have changed via job_finalize (and job_txn_apply
3436 * automatically acquires the new one), so make sure we release the correct
3437 * one.
3438 */
3439 aio_context = block_job_get_aio_context(job);
3440 job_unref(&job->job);
3441 aio_context_release(aio_context);
3442 }
3443
3444 void qmp_block_job_dismiss(const char *id, Error **errp)
3445 {
3446 AioContext *aio_context;
3447 BlockJob *bjob = find_block_job(id, &aio_context, errp);
3448 Job *job;
3449
3450 if (!bjob) {
3451 return;
3452 }
3453
3454 trace_qmp_block_job_dismiss(bjob);
3455 job = &bjob->job;
3456 job_dismiss(&job, errp);
3457 aio_context_release(aio_context);
3458 }
3459
3460 void qmp_change_backing_file(const char *device,
3461 const char *image_node_name,
3462 const char *backing_file,
3463 Error **errp)
3464 {
3465 BlockDriverState *bs = NULL;
3466 AioContext *aio_context;
3467 BlockDriverState *image_bs = NULL;
3468 Error *local_err = NULL;
3469 bool ro;
3470 int ret;
3471
3472 bs = qmp_get_root_bs(device, errp);
3473 if (!bs) {
3474 return;
3475 }
3476
3477 aio_context = bdrv_get_aio_context(bs);
3478 aio_context_acquire(aio_context);
3479
3480 image_bs = bdrv_lookup_bs(NULL, image_node_name, &local_err);
3481 if (local_err) {
3482 error_propagate(errp, local_err);
3483 goto out;
3484 }
3485
3486 if (!image_bs) {
3487 error_setg(errp, "image file not found");
3488 goto out;
3489 }
3490
3491 if (bdrv_find_base(image_bs) == image_bs) {
3492 error_setg(errp, "not allowing backing file change on an image "
3493 "without a backing file");
3494 goto out;
3495 }
3496
3497 /* even though we are not necessarily operating on bs, we need it to
3498 * determine if block ops are currently prohibited on the chain */
3499 if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_CHANGE, errp)) {
3500 goto out;
3501 }
3502
3503 /* final sanity check */
3504 if (!bdrv_chain_contains(bs, image_bs)) {
3505 error_setg(errp, "'%s' and image file are not in the same chain",
3506 device);
3507 goto out;
3508 }
3509
3510 /* if not r/w, reopen to make r/w */
3511 ro = bdrv_is_read_only(image_bs);
3512
3513 if (ro) {
3514 if (bdrv_reopen_set_read_only(image_bs, false, errp) != 0) {
3515 goto out;
3516 }
3517 }
3518
3519 ret = bdrv_change_backing_file(image_bs, backing_file,
3520 image_bs->drv ? image_bs->drv->format_name : "",
3521 false);
3522
3523 if (ret < 0) {
3524 error_setg_errno(errp, -ret, "Could not change backing file to '%s'",
3525 backing_file);
3526 /* don't exit here, so we can try to restore open flags if
3527 * appropriate */
3528 }
3529
3530 if (ro) {
3531 bdrv_reopen_set_read_only(image_bs, true, errp);
3532 }
3533
3534 out:
3535 aio_context_release(aio_context);
3536 }
3537
3538 void qmp_blockdev_add(BlockdevOptions *options, Error **errp)
3539 {
3540 BlockDriverState *bs;
3541 QObject *obj;
3542 Visitor *v = qobject_output_visitor_new(&obj);
3543 QDict *qdict;
3544
3545 visit_type_BlockdevOptions(v, NULL, &options, &error_abort);
3546 visit_complete(v, &obj);
3547 qdict = qobject_to(QDict, obj);
3548
3549 qdict_flatten(qdict);
3550
3551 if (!qdict_get_try_str(qdict, "node-name")) {
3552 error_setg(errp, "'node-name' must be specified for the root node");
3553 goto fail;
3554 }
3555
3556 bs = bds_tree_init(qdict, errp);
3557 if (!bs) {
3558 goto fail;
3559 }
3560
3561 bdrv_set_monitor_owned(bs);
3562
3563 fail:
3564 visit_free(v);
3565 }
3566
3567 void qmp_blockdev_reopen(BlockdevOptionsList *reopen_list, Error **errp)
3568 {
3569 BlockReopenQueue *queue = NULL;
3570 GSList *drained = NULL;
3571 GSList *p;
3572
3573 /* Add each one of the BDS that we want to reopen to the queue */
3574 for (; reopen_list != NULL; reopen_list = reopen_list->next) {
3575 BlockdevOptions *options = reopen_list->value;
3576 BlockDriverState *bs;
3577 AioContext *ctx;
3578 QObject *obj;
3579 Visitor *v;
3580 QDict *qdict;
3581
3582 /* Check for the selected node name */
3583 if (!options->has_node_name) {
3584 error_setg(errp, "node-name not specified");
3585 goto fail;
3586 }
3587
3588 bs = bdrv_find_node(options->node_name);
3589 if (!bs) {
3590 error_setg(errp, "Failed to find node with node-name='%s'",
3591 options->node_name);
3592 goto fail;
3593 }
3594
3595 /* Put all options in a QDict and flatten it */
3596 v = qobject_output_visitor_new(&obj);
3597 visit_type_BlockdevOptions(v, NULL, &options, &error_abort);
3598 visit_complete(v, &obj);
3599 visit_free(v);
3600
3601 qdict = qobject_to(QDict, obj);
3602
3603 qdict_flatten(qdict);
3604
3605 ctx = bdrv_get_aio_context(bs);
3606 aio_context_acquire(ctx);
3607
3608 bdrv_subtree_drained_begin(bs);
3609 queue = bdrv_reopen_queue(queue, bs, qdict, false);
3610 drained = g_slist_prepend(drained, bs);
3611
3612 aio_context_release(ctx);
3613 }
3614
3615 /* Perform the reopen operation */
3616 bdrv_reopen_multiple(queue, errp);
3617 queue = NULL;
3618
3619 fail:
3620 bdrv_reopen_queue_free(queue);
3621 for (p = drained; p; p = p->next) {
3622 BlockDriverState *bs = p->data;
3623 AioContext *ctx = bdrv_get_aio_context(bs);
3624
3625 aio_context_acquire(ctx);
3626 bdrv_subtree_drained_end(bs);
3627 aio_context_release(ctx);
3628 }
3629 g_slist_free(drained);
3630 }
3631
3632 void qmp_blockdev_del(const char *node_name, Error **errp)
3633 {
3634 AioContext *aio_context;
3635 BlockDriverState *bs;
3636
3637 GLOBAL_STATE_CODE();
3638
3639 bs = bdrv_find_node(node_name);
3640 if (!bs) {
3641 error_setg(errp, "Failed to find node with node-name='%s'", node_name);
3642 return;
3643 }
3644 if (bdrv_has_blk(bs)) {
3645 error_setg(errp, "Node %s is in use", node_name);
3646 return;
3647 }
3648 aio_context = bdrv_get_aio_context(bs);
3649 aio_context_acquire(aio_context);
3650
3651 if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_DRIVE_DEL, errp)) {
3652 goto out;
3653 }
3654
3655 if (!QTAILQ_IN_USE(bs, monitor_list)) {
3656 error_setg(errp, "Node %s is not owned by the monitor",
3657 bs->node_name);
3658 goto out;
3659 }
3660
3661 if (bs->refcnt > 1) {
3662 error_setg(errp, "Block device %s is in use",
3663 bdrv_get_device_or_node_name(bs));
3664 goto out;
3665 }
3666
3667 QTAILQ_REMOVE(&monitor_bdrv_states, bs, monitor_list);
3668 bdrv_unref(bs);
3669
3670 out:
3671 aio_context_release(aio_context);
3672 }
3673
3674 static BdrvChild *bdrv_find_child(BlockDriverState *parent_bs,
3675 const char *child_name)
3676 {
3677 BdrvChild *child;
3678
3679 QLIST_FOREACH(child, &parent_bs->children, next) {
3680 if (strcmp(child->name, child_name) == 0) {
3681 return child;
3682 }
3683 }
3684
3685 return NULL;
3686 }
3687
3688 void qmp_x_blockdev_change(const char *parent, bool has_child,
3689 const char *child, bool has_node,
3690 const char *node, Error **errp)
3691 {
3692 BlockDriverState *parent_bs, *new_bs = NULL;
3693 BdrvChild *p_child;
3694
3695 parent_bs = bdrv_lookup_bs(parent, parent, errp);
3696 if (!parent_bs) {
3697 return;
3698 }
3699
3700 if (has_child == has_node) {
3701 if (has_child) {
3702 error_setg(errp, "The parameters child and node are in conflict");
3703 } else {
3704 error_setg(errp, "Either child or node must be specified");
3705 }
3706 return;
3707 }
3708
3709 if (has_child) {
3710 p_child = bdrv_find_child(parent_bs, child);
3711 if (!p_child) {
3712 error_setg(errp, "Node '%s' does not have child '%s'",
3713 parent, child);
3714 return;
3715 }
3716 bdrv_del_child(parent_bs, p_child, errp);
3717 }
3718
3719 if (has_node) {
3720 new_bs = bdrv_find_node(node);
3721 if (!new_bs) {
3722 error_setg(errp, "Node '%s' not found", node);
3723 return;
3724 }
3725 bdrv_add_child(parent_bs, new_bs, errp);
3726 }
3727 }
3728
3729 BlockJobInfoList *qmp_query_block_jobs(Error **errp)
3730 {
3731 BlockJobInfoList *head = NULL, **tail = &head;
3732 BlockJob *job;
3733
3734 for (job = block_job_next(NULL); job; job = block_job_next(job)) {
3735 BlockJobInfo *value;
3736 AioContext *aio_context;
3737
3738 if (block_job_is_internal(job)) {
3739 continue;
3740 }
3741 aio_context = block_job_get_aio_context(job);
3742 aio_context_acquire(aio_context);
3743 value = block_job_query(job, errp);
3744 aio_context_release(aio_context);
3745 if (!value) {
3746 qapi_free_BlockJobInfoList(head);
3747 return NULL;
3748 }
3749 QAPI_LIST_APPEND(tail, value);
3750 }
3751
3752 return head;
3753 }
3754
3755 void qmp_x_blockdev_set_iothread(const char *node_name, StrOrNull *iothread,
3756 bool has_force, bool force, Error **errp)
3757 {
3758 AioContext *old_context;
3759 AioContext *new_context;
3760 BlockDriverState *bs;
3761
3762 bs = bdrv_find_node(node_name);
3763 if (!bs) {
3764 error_setg(errp, "Failed to find node with node-name='%s'", node_name);
3765 return;
3766 }
3767
3768 /* Protects against accidents. */
3769 if (!(has_force && force) && bdrv_has_blk(bs)) {
3770 error_setg(errp, "Node %s is associated with a BlockBackend and could "
3771 "be in use (use force=true to override this check)",
3772 node_name);
3773 return;
3774 }
3775
3776 if (iothread->type == QTYPE_QSTRING) {
3777 IOThread *obj = iothread_by_id(iothread->u.s);
3778 if (!obj) {
3779 error_setg(errp, "Cannot find iothread %s", iothread->u.s);
3780 return;
3781 }
3782
3783 new_context = iothread_get_aio_context(obj);
3784 } else {
3785 new_context = qemu_get_aio_context();
3786 }
3787
3788 old_context = bdrv_get_aio_context(bs);
3789 aio_context_acquire(old_context);
3790
3791 bdrv_try_set_aio_context(bs, new_context, errp);
3792
3793 aio_context_release(old_context);
3794 }
3795
3796 QemuOptsList qemu_common_drive_opts = {
3797 .name = "drive",
3798 .head = QTAILQ_HEAD_INITIALIZER(qemu_common_drive_opts.head),
3799 .desc = {
3800 {
3801 .name = "snapshot",
3802 .type = QEMU_OPT_BOOL,
3803 .help = "enable/disable snapshot mode",
3804 },{
3805 .name = "aio",
3806 .type = QEMU_OPT_STRING,
3807 .help = "host AIO implementation (threads, native, io_uring)",
3808 },{
3809 .name = BDRV_OPT_CACHE_WB,
3810 .type = QEMU_OPT_BOOL,
3811 .help = "Enable writeback mode",
3812 },{
3813 .name = "format",
3814 .type = QEMU_OPT_STRING,
3815 .help = "disk format (raw, qcow2, ...)",
3816 },{
3817 .name = "rerror",
3818 .type = QEMU_OPT_STRING,
3819 .help = "read error action",
3820 },{
3821 .name = "werror",
3822 .type = QEMU_OPT_STRING,
3823 .help = "write error action",
3824 },{
3825 .name = BDRV_OPT_READ_ONLY,
3826 .type = QEMU_OPT_BOOL,
3827 .help = "open drive file as read-only",
3828 },
3829
3830 THROTTLE_OPTS,
3831
3832 {
3833 .name = "throttling.group",
3834 .type = QEMU_OPT_STRING,
3835 .help = "name of the block throttling group",
3836 },{
3837 .name = "copy-on-read",
3838 .type = QEMU_OPT_BOOL,
3839 .help = "copy read data from backing file into image file",
3840 },{
3841 .name = "detect-zeroes",
3842 .type = QEMU_OPT_STRING,
3843 .help = "try to optimize zero writes (off, on, unmap)",
3844 },{
3845 .name = "stats-account-invalid",
3846 .type = QEMU_OPT_BOOL,
3847 .help = "whether to account for invalid I/O operations "
3848 "in the statistics",
3849 },{
3850 .name = "stats-account-failed",
3851 .type = QEMU_OPT_BOOL,
3852 .help = "whether to account for failed I/O operations "
3853 "in the statistics",
3854 },
3855 { /* end of list */ }
3856 },
3857 };
3858
3859 QemuOptsList qemu_drive_opts = {
3860 .name = "drive",
3861 .head = QTAILQ_HEAD_INITIALIZER(qemu_drive_opts.head),
3862 .desc = {
3863 /*
3864 * no elements => accept any params
3865 * validation will happen later
3866 */
3867 { /* end of list */ }
3868 },
3869 };
QEmu