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