tests: add migrate-during-backup
[qemu.git] / block.c
CommitLineData
fc01f7e7
FB
1/*
2 * QEMU System Emulator block driver
5fafdf24 3 *
fc01f7e7 4 * Copyright (c) 2003 Fabrice Bellard
c20555e1 5 * Copyright (c) 2020 Virtuozzo International GmbH.
5fafdf24 6 *
fc01f7e7
FB
7 * Permission is hereby granted, free of charge, to any person obtaining a copy
8 * of this software and associated documentation files (the "Software"), to deal
9 * in the Software without restriction, including without limitation the rights
10 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
11 * copies of the Software, and to permit persons to whom the Software is
12 * furnished to do so, subject to the following conditions:
13 *
14 * The above copyright notice and this permission notice shall be included in
15 * all copies or substantial portions of the Software.
16 *
17 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
18 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
19 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
20 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
21 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
22 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
23 * THE SOFTWARE.
24 */
e688df6b 25
d38ea87a 26#include "qemu/osdep.h"
0ab8ed18 27#include "block/trace.h"
737e150e
PB
28#include "block/block_int.h"
29#include "block/blockjob.h"
0c9b70d5 30#include "block/fuse.h"
cd7fca95 31#include "block/nbd.h"
609f45ea 32#include "block/qdict.h"
d49b6836 33#include "qemu/error-report.h"
5e5733e5 34#include "block/module_block.h"
db725815 35#include "qemu/main-loop.h"
1de7afc9 36#include "qemu/module.h"
e688df6b 37#include "qapi/error.h"
452fcdbc 38#include "qapi/qmp/qdict.h"
7b1b5d19 39#include "qapi/qmp/qjson.h"
e59a0cf1 40#include "qapi/qmp/qnull.h"
fc81fa1e 41#include "qapi/qmp/qstring.h"
e1d74bc6
KW
42#include "qapi/qobject-output-visitor.h"
43#include "qapi/qapi-visit-block-core.h"
bfb197e0 44#include "sysemu/block-backend.h"
1de7afc9 45#include "qemu/notify.h"
922a01a0 46#include "qemu/option.h"
10817bf0 47#include "qemu/coroutine.h"
c13163fb 48#include "block/qapi.h"
1de7afc9 49#include "qemu/timer.h"
f348b6d1
VB
50#include "qemu/cutils.h"
51#include "qemu/id.h"
0bc329fb
HR
52#include "qemu/range.h"
53#include "qemu/rcu.h"
21c2283e 54#include "block/coroutines.h"
fc01f7e7 55
71e72a19 56#ifdef CONFIG_BSD
7674e7bf 57#include <sys/ioctl.h>
72cf2d4f 58#include <sys/queue.h>
feccdcee 59#if defined(HAVE_SYS_DISK_H)
7674e7bf
FB
60#include <sys/disk.h>
61#endif
c5e97233 62#endif
7674e7bf 63
49dc768d
AL
64#ifdef _WIN32
65#include <windows.h>
66#endif
67
1c9805a3
SH
68#define NOT_DONE 0x7fffffff /* used while emulated sync operation in progress */
69
dc364f4c
BC
70static QTAILQ_HEAD(, BlockDriverState) graph_bdrv_states =
71 QTAILQ_HEAD_INITIALIZER(graph_bdrv_states);
72
2c1d04e0
HR
73static QTAILQ_HEAD(, BlockDriverState) all_bdrv_states =
74 QTAILQ_HEAD_INITIALIZER(all_bdrv_states);
75
8a22f02a
SH
76static QLIST_HEAD(, BlockDriver) bdrv_drivers =
77 QLIST_HEAD_INITIALIZER(bdrv_drivers);
ea2384d3 78
5b363937
HR
79static BlockDriverState *bdrv_open_inherit(const char *filename,
80 const char *reference,
81 QDict *options, int flags,
82 BlockDriverState *parent,
bd86fb99 83 const BdrvChildClass *child_class,
272c02ea 84 BdrvChildRole child_role,
5b363937 85 Error **errp);
f3930ed0 86
0978623e
VSO
87static void bdrv_replace_child_noperm(BdrvChild *child,
88 BlockDriverState *new_bs);
e9238278
VSO
89static void bdrv_remove_file_or_backing_child(BlockDriverState *bs,
90 BdrvChild *child,
91 Transaction *tran);
160333e1
VSO
92static void bdrv_remove_filter_or_cow_child(BlockDriverState *bs,
93 Transaction *tran);
0978623e 94
72373e40
VSO
95static int bdrv_reopen_prepare(BDRVReopenState *reopen_state,
96 BlockReopenQueue *queue,
ecd30d2d 97 Transaction *change_child_tran, Error **errp);
53e96d1e
VSO
98static void bdrv_reopen_commit(BDRVReopenState *reopen_state);
99static void bdrv_reopen_abort(BDRVReopenState *reopen_state);
100
eb852011
MA
101/* If non-zero, use only whitelisted block drivers */
102static int use_bdrv_whitelist;
103
9e0b22f4
SH
104#ifdef _WIN32
105static int is_windows_drive_prefix(const char *filename)
106{
107 return (((filename[0] >= 'a' && filename[0] <= 'z') ||
108 (filename[0] >= 'A' && filename[0] <= 'Z')) &&
109 filename[1] == ':');
110}
111
112int is_windows_drive(const char *filename)
113{
114 if (is_windows_drive_prefix(filename) &&
115 filename[2] == '\0')
116 return 1;
117 if (strstart(filename, "\\\\.\\", NULL) ||
118 strstart(filename, "//./", NULL))
119 return 1;
120 return 0;
121}
122#endif
123
339064d5
KW
124size_t bdrv_opt_mem_align(BlockDriverState *bs)
125{
126 if (!bs || !bs->drv) {
459b4e66 127 /* page size or 4k (hdd sector size) should be on the safe side */
038adc2f 128 return MAX(4096, qemu_real_host_page_size);
339064d5
KW
129 }
130
131 return bs->bl.opt_mem_alignment;
132}
133
4196d2f0
DL
134size_t bdrv_min_mem_align(BlockDriverState *bs)
135{
136 if (!bs || !bs->drv) {
459b4e66 137 /* page size or 4k (hdd sector size) should be on the safe side */
038adc2f 138 return MAX(4096, qemu_real_host_page_size);
4196d2f0
DL
139 }
140
141 return bs->bl.min_mem_alignment;
142}
143
9e0b22f4 144/* check if the path starts with "<protocol>:" */
5c98415b 145int path_has_protocol(const char *path)
9e0b22f4 146{
947995c0
PB
147 const char *p;
148
9e0b22f4
SH
149#ifdef _WIN32
150 if (is_windows_drive(path) ||
151 is_windows_drive_prefix(path)) {
152 return 0;
153 }
947995c0
PB
154 p = path + strcspn(path, ":/\\");
155#else
156 p = path + strcspn(path, ":/");
9e0b22f4
SH
157#endif
158
947995c0 159 return *p == ':';
9e0b22f4
SH
160}
161
83f64091 162int path_is_absolute(const char *path)
3b0d4f61 163{
21664424
FB
164#ifdef _WIN32
165 /* specific case for names like: "\\.\d:" */
f53f4da9 166 if (is_windows_drive(path) || is_windows_drive_prefix(path)) {
21664424 167 return 1;
f53f4da9
PB
168 }
169 return (*path == '/' || *path == '\\');
3b9f94e1 170#else
f53f4da9 171 return (*path == '/');
3b9f94e1 172#endif
3b0d4f61
FB
173}
174
009b03aa 175/* if filename is absolute, just return its duplicate. Otherwise, build a
83f64091
FB
176 path to it by considering it is relative to base_path. URL are
177 supported. */
009b03aa 178char *path_combine(const char *base_path, const char *filename)
3b0d4f61 179{
009b03aa 180 const char *protocol_stripped = NULL;
83f64091 181 const char *p, *p1;
009b03aa 182 char *result;
83f64091
FB
183 int len;
184
83f64091 185 if (path_is_absolute(filename)) {
009b03aa
HR
186 return g_strdup(filename);
187 }
0d54a6fe 188
009b03aa
HR
189 if (path_has_protocol(base_path)) {
190 protocol_stripped = strchr(base_path, ':');
191 if (protocol_stripped) {
192 protocol_stripped++;
0d54a6fe 193 }
009b03aa
HR
194 }
195 p = protocol_stripped ?: base_path;
0d54a6fe 196
009b03aa 197 p1 = strrchr(base_path, '/');
3b9f94e1 198#ifdef _WIN32
009b03aa
HR
199 {
200 const char *p2;
201 p2 = strrchr(base_path, '\\');
202 if (!p1 || p2 > p1) {
203 p1 = p2;
3b9f94e1 204 }
009b03aa 205 }
3b9f94e1 206#endif
009b03aa
HR
207 if (p1) {
208 p1++;
209 } else {
210 p1 = base_path;
211 }
212 if (p1 > p) {
213 p = p1;
3b0d4f61 214 }
009b03aa
HR
215 len = p - base_path;
216
217 result = g_malloc(len + strlen(filename) + 1);
218 memcpy(result, base_path, len);
219 strcpy(result + len, filename);
220
221 return result;
222}
223
03c320d8
HR
224/*
225 * Helper function for bdrv_parse_filename() implementations to remove optional
226 * protocol prefixes (especially "file:") from a filename and for putting the
227 * stripped filename into the options QDict if there is such a prefix.
228 */
229void bdrv_parse_filename_strip_prefix(const char *filename, const char *prefix,
230 QDict *options)
231{
232 if (strstart(filename, prefix, &filename)) {
233 /* Stripping the explicit protocol prefix may result in a protocol
234 * prefix being (wrongly) detected (if the filename contains a colon) */
235 if (path_has_protocol(filename)) {
18cf67c5 236 GString *fat_filename;
03c320d8
HR
237
238 /* This means there is some colon before the first slash; therefore,
239 * this cannot be an absolute path */
240 assert(!path_is_absolute(filename));
241
242 /* And we can thus fix the protocol detection issue by prefixing it
243 * by "./" */
18cf67c5
MA
244 fat_filename = g_string_new("./");
245 g_string_append(fat_filename, filename);
03c320d8 246
18cf67c5 247 assert(!path_has_protocol(fat_filename->str));
03c320d8 248
18cf67c5
MA
249 qdict_put(options, "filename",
250 qstring_from_gstring(fat_filename));
03c320d8
HR
251 } else {
252 /* If no protocol prefix was detected, we can use the shortened
253 * filename as-is */
254 qdict_put_str(options, "filename", filename);
255 }
256 }
257}
258
259
9c5e6594
KW
260/* Returns whether the image file is opened as read-only. Note that this can
261 * return false and writing to the image file is still not possible because the
262 * image is inactivated. */
93ed524e
JC
263bool bdrv_is_read_only(BlockDriverState *bs)
264{
975da073 265 return !(bs->open_flags & BDRV_O_RDWR);
93ed524e
JC
266}
267
54a32bfe
KW
268int bdrv_can_set_read_only(BlockDriverState *bs, bool read_only,
269 bool ignore_allow_rdw, Error **errp)
fe5241bf 270{
e2b8247a
JC
271 /* Do not set read_only if copy_on_read is enabled */
272 if (bs->copy_on_read && read_only) {
273 error_setg(errp, "Can't set node '%s' to r/o with copy-on-read enabled",
274 bdrv_get_device_or_node_name(bs));
275 return -EINVAL;
276 }
277
d6fcdf06 278 /* Do not clear read_only if it is prohibited */
54a32bfe
KW
279 if (!read_only && !(bs->open_flags & BDRV_O_ALLOW_RDWR) &&
280 !ignore_allow_rdw)
281 {
d6fcdf06
JC
282 error_setg(errp, "Node '%s' is read only",
283 bdrv_get_device_or_node_name(bs));
284 return -EPERM;
285 }
286
45803a03
JC
287 return 0;
288}
289
eaa2410f
KW
290/*
291 * Called by a driver that can only provide a read-only image.
292 *
293 * Returns 0 if the node is already read-only or it could switch the node to
294 * read-only because BDRV_O_AUTO_RDONLY is set.
295 *
296 * Returns -EACCES if the node is read-write and BDRV_O_AUTO_RDONLY is not set
297 * or bdrv_can_set_read_only() forbids making the node read-only. If @errmsg
298 * is not NULL, it is used as the error message for the Error object.
299 */
300int bdrv_apply_auto_read_only(BlockDriverState *bs, const char *errmsg,
301 Error **errp)
45803a03
JC
302{
303 int ret = 0;
304
eaa2410f
KW
305 if (!(bs->open_flags & BDRV_O_RDWR)) {
306 return 0;
307 }
308 if (!(bs->open_flags & BDRV_O_AUTO_RDONLY)) {
309 goto fail;
45803a03
JC
310 }
311
eaa2410f
KW
312 ret = bdrv_can_set_read_only(bs, true, false, NULL);
313 if (ret < 0) {
314 goto fail;
eeae6a59
KW
315 }
316
eaa2410f
KW
317 bs->open_flags &= ~BDRV_O_RDWR;
318
e2b8247a 319 return 0;
eaa2410f
KW
320
321fail:
322 error_setg(errp, "%s", errmsg ?: "Image is read-only");
323 return -EACCES;
fe5241bf
JC
324}
325
645ae7d8
HR
326/*
327 * If @backing is empty, this function returns NULL without setting
328 * @errp. In all other cases, NULL will only be returned with @errp
329 * set.
330 *
331 * Therefore, a return value of NULL without @errp set means that
332 * there is no backing file; if @errp is set, there is one but its
333 * absolute filename cannot be generated.
334 */
335char *bdrv_get_full_backing_filename_from_filename(const char *backed,
336 const char *backing,
337 Error **errp)
dc5a1371 338{
645ae7d8
HR
339 if (backing[0] == '\0') {
340 return NULL;
341 } else if (path_has_protocol(backing) || path_is_absolute(backing)) {
342 return g_strdup(backing);
9f07429e
HR
343 } else if (backed[0] == '\0' || strstart(backed, "json:", NULL)) {
344 error_setg(errp, "Cannot use relative backing file names for '%s'",
345 backed);
645ae7d8 346 return NULL;
dc5a1371 347 } else {
645ae7d8 348 return path_combine(backed, backing);
dc5a1371
PB
349 }
350}
351
9f4793d8
HR
352/*
353 * If @filename is empty or NULL, this function returns NULL without
354 * setting @errp. In all other cases, NULL will only be returned with
355 * @errp set.
356 */
357static char *bdrv_make_absolute_filename(BlockDriverState *relative_to,
358 const char *filename, Error **errp)
0a82855a 359{
8df68616 360 char *dir, *full_name;
9f4793d8 361
8df68616
HR
362 if (!filename || filename[0] == '\0') {
363 return NULL;
364 } else if (path_has_protocol(filename) || path_is_absolute(filename)) {
365 return g_strdup(filename);
366 }
9f07429e 367
8df68616
HR
368 dir = bdrv_dirname(relative_to, errp);
369 if (!dir) {
370 return NULL;
371 }
f30c66ba 372
8df68616
HR
373 full_name = g_strconcat(dir, filename, NULL);
374 g_free(dir);
375 return full_name;
9f4793d8
HR
376}
377
378char *bdrv_get_full_backing_filename(BlockDriverState *bs, Error **errp)
379{
380 return bdrv_make_absolute_filename(bs, bs->backing_file, errp);
0a82855a
HR
381}
382
0eb7217e
SH
383void bdrv_register(BlockDriver *bdrv)
384{
a15f08dc 385 assert(bdrv->format_name);
8a22f02a 386 QLIST_INSERT_HEAD(&bdrv_drivers, bdrv, list);
ea2384d3 387}
b338082b 388
e4e9986b
MA
389BlockDriverState *bdrv_new(void)
390{
391 BlockDriverState *bs;
392 int i;
393
5839e53b 394 bs = g_new0(BlockDriverState, 1);
e4654d2d 395 QLIST_INIT(&bs->dirty_bitmaps);
fbe40ff7
FZ
396 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
397 QLIST_INIT(&bs->op_blockers[i]);
398 }
3783fa3d 399 qemu_co_mutex_init(&bs->reqs_lock);
2119882c 400 qemu_mutex_init(&bs->dirty_bitmap_mutex);
9fcb0251 401 bs->refcnt = 1;
dcd04228 402 bs->aio_context = qemu_get_aio_context();
d7d512f6 403
3ff2f67a
EY
404 qemu_co_queue_init(&bs->flush_queue);
405
0bc329fb
HR
406 qemu_co_mutex_init(&bs->bsc_modify_lock);
407 bs->block_status_cache = g_new0(BdrvBlockStatusCache, 1);
408
0f12264e
KW
409 for (i = 0; i < bdrv_drain_all_count; i++) {
410 bdrv_drained_begin(bs);
411 }
412
2c1d04e0
HR
413 QTAILQ_INSERT_TAIL(&all_bdrv_states, bs, bs_list);
414
b338082b
FB
415 return bs;
416}
417
88d88798 418static BlockDriver *bdrv_do_find_format(const char *format_name)
ea2384d3
FB
419{
420 BlockDriver *drv1;
88d88798 421
8a22f02a
SH
422 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
423 if (!strcmp(drv1->format_name, format_name)) {
ea2384d3 424 return drv1;
8a22f02a 425 }
ea2384d3 426 }
88d88798 427
ea2384d3
FB
428 return NULL;
429}
430
88d88798
MM
431BlockDriver *bdrv_find_format(const char *format_name)
432{
433 BlockDriver *drv1;
434 int i;
435
436 drv1 = bdrv_do_find_format(format_name);
437 if (drv1) {
438 return drv1;
439 }
440
441 /* The driver isn't registered, maybe we need to load a module */
442 for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); ++i) {
443 if (!strcmp(block_driver_modules[i].format_name, format_name)) {
444 block_module_load_one(block_driver_modules[i].library_name);
445 break;
446 }
447 }
448
449 return bdrv_do_find_format(format_name);
450}
451
9ac404c5 452static int bdrv_format_is_whitelisted(const char *format_name, bool read_only)
eb852011 453{
b64ec4e4
FZ
454 static const char *whitelist_rw[] = {
455 CONFIG_BDRV_RW_WHITELIST
859aef02 456 NULL
b64ec4e4
FZ
457 };
458 static const char *whitelist_ro[] = {
459 CONFIG_BDRV_RO_WHITELIST
859aef02 460 NULL
eb852011
MA
461 };
462 const char **p;
463
b64ec4e4 464 if (!whitelist_rw[0] && !whitelist_ro[0]) {
eb852011 465 return 1; /* no whitelist, anything goes */
b64ec4e4 466 }
eb852011 467
b64ec4e4 468 for (p = whitelist_rw; *p; p++) {
9ac404c5 469 if (!strcmp(format_name, *p)) {
eb852011
MA
470 return 1;
471 }
472 }
b64ec4e4
FZ
473 if (read_only) {
474 for (p = whitelist_ro; *p; p++) {
9ac404c5 475 if (!strcmp(format_name, *p)) {
b64ec4e4
FZ
476 return 1;
477 }
478 }
479 }
eb852011
MA
480 return 0;
481}
482
9ac404c5
AS
483int bdrv_is_whitelisted(BlockDriver *drv, bool read_only)
484{
485 return bdrv_format_is_whitelisted(drv->format_name, read_only);
486}
487
e6ff69bf
DB
488bool bdrv_uses_whitelist(void)
489{
490 return use_bdrv_whitelist;
491}
492
5b7e1542
ZYW
493typedef struct CreateCo {
494 BlockDriver *drv;
495 char *filename;
83d0521a 496 QemuOpts *opts;
5b7e1542 497 int ret;
cc84d90f 498 Error *err;
5b7e1542
ZYW
499} CreateCo;
500
501static void coroutine_fn bdrv_create_co_entry(void *opaque)
502{
cc84d90f
HR
503 Error *local_err = NULL;
504 int ret;
505
5b7e1542
ZYW
506 CreateCo *cco = opaque;
507 assert(cco->drv);
508
b92902df
ML
509 ret = cco->drv->bdrv_co_create_opts(cco->drv,
510 cco->filename, cco->opts, &local_err);
621ff94d 511 error_propagate(&cco->err, local_err);
cc84d90f 512 cco->ret = ret;
5b7e1542
ZYW
513}
514
0e7e1989 515int bdrv_create(BlockDriver *drv, const char* filename,
83d0521a 516 QemuOpts *opts, Error **errp)
ea2384d3 517{
5b7e1542
ZYW
518 int ret;
519
520 Coroutine *co;
521 CreateCo cco = {
522 .drv = drv,
523 .filename = g_strdup(filename),
83d0521a 524 .opts = opts,
5b7e1542 525 .ret = NOT_DONE,
cc84d90f 526 .err = NULL,
5b7e1542
ZYW
527 };
528
efc75e2a 529 if (!drv->bdrv_co_create_opts) {
cc84d90f 530 error_setg(errp, "Driver '%s' does not support image creation", drv->format_name);
80168bff
LC
531 ret = -ENOTSUP;
532 goto out;
5b7e1542
ZYW
533 }
534
535 if (qemu_in_coroutine()) {
536 /* Fast-path if already in coroutine context */
537 bdrv_create_co_entry(&cco);
538 } else {
0b8b8753
PB
539 co = qemu_coroutine_create(bdrv_create_co_entry, &cco);
540 qemu_coroutine_enter(co);
5b7e1542 541 while (cco.ret == NOT_DONE) {
b47ec2c4 542 aio_poll(qemu_get_aio_context(), true);
5b7e1542
ZYW
543 }
544 }
545
546 ret = cco.ret;
cc84d90f 547 if (ret < 0) {
84d18f06 548 if (cco.err) {
cc84d90f
HR
549 error_propagate(errp, cco.err);
550 } else {
551 error_setg_errno(errp, -ret, "Could not create image");
552 }
553 }
0e7e1989 554
80168bff
LC
555out:
556 g_free(cco.filename);
5b7e1542 557 return ret;
ea2384d3
FB
558}
559
fd17146c
HR
560/**
561 * Helper function for bdrv_create_file_fallback(): Resize @blk to at
562 * least the given @minimum_size.
563 *
564 * On success, return @blk's actual length.
565 * Otherwise, return -errno.
566 */
567static int64_t create_file_fallback_truncate(BlockBackend *blk,
568 int64_t minimum_size, Error **errp)
84a12e66 569{
cc84d90f 570 Error *local_err = NULL;
fd17146c 571 int64_t size;
cc84d90f 572 int ret;
84a12e66 573
8c6242b6
KW
574 ret = blk_truncate(blk, minimum_size, false, PREALLOC_MODE_OFF, 0,
575 &local_err);
fd17146c
HR
576 if (ret < 0 && ret != -ENOTSUP) {
577 error_propagate(errp, local_err);
578 return ret;
579 }
580
581 size = blk_getlength(blk);
582 if (size < 0) {
583 error_free(local_err);
584 error_setg_errno(errp, -size,
585 "Failed to inquire the new image file's length");
586 return size;
587 }
588
589 if (size < minimum_size) {
590 /* Need to grow the image, but we failed to do that */
591 error_propagate(errp, local_err);
592 return -ENOTSUP;
593 }
594
595 error_free(local_err);
596 local_err = NULL;
597
598 return size;
599}
600
601/**
602 * Helper function for bdrv_create_file_fallback(): Zero the first
603 * sector to remove any potentially pre-existing image header.
604 */
605static int create_file_fallback_zero_first_sector(BlockBackend *blk,
606 int64_t current_size,
607 Error **errp)
608{
609 int64_t bytes_to_clear;
610 int ret;
611
612 bytes_to_clear = MIN(current_size, BDRV_SECTOR_SIZE);
613 if (bytes_to_clear) {
614 ret = blk_pwrite_zeroes(blk, 0, bytes_to_clear, BDRV_REQ_MAY_UNMAP);
615 if (ret < 0) {
616 error_setg_errno(errp, -ret,
617 "Failed to clear the new image's first sector");
618 return ret;
619 }
620 }
621
622 return 0;
623}
624
5a5e7f8c
ML
625/**
626 * Simple implementation of bdrv_co_create_opts for protocol drivers
627 * which only support creation via opening a file
628 * (usually existing raw storage device)
629 */
630int coroutine_fn bdrv_co_create_opts_simple(BlockDriver *drv,
631 const char *filename,
632 QemuOpts *opts,
633 Error **errp)
fd17146c
HR
634{
635 BlockBackend *blk;
eeea1faa 636 QDict *options;
fd17146c
HR
637 int64_t size = 0;
638 char *buf = NULL;
639 PreallocMode prealloc;
640 Error *local_err = NULL;
641 int ret;
642
643 size = qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0);
644 buf = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC);
645 prealloc = qapi_enum_parse(&PreallocMode_lookup, buf,
646 PREALLOC_MODE_OFF, &local_err);
647 g_free(buf);
648 if (local_err) {
649 error_propagate(errp, local_err);
650 return -EINVAL;
651 }
652
653 if (prealloc != PREALLOC_MODE_OFF) {
654 error_setg(errp, "Unsupported preallocation mode '%s'",
655 PreallocMode_str(prealloc));
656 return -ENOTSUP;
657 }
658
eeea1faa 659 options = qdict_new();
fd17146c
HR
660 qdict_put_str(options, "driver", drv->format_name);
661
662 blk = blk_new_open(filename, NULL, options,
663 BDRV_O_RDWR | BDRV_O_RESIZE, errp);
664 if (!blk) {
665 error_prepend(errp, "Protocol driver '%s' does not support image "
666 "creation, and opening the image failed: ",
667 drv->format_name);
668 return -EINVAL;
669 }
670
671 size = create_file_fallback_truncate(blk, size, errp);
672 if (size < 0) {
673 ret = size;
674 goto out;
675 }
676
677 ret = create_file_fallback_zero_first_sector(blk, size, errp);
678 if (ret < 0) {
679 goto out;
680 }
681
682 ret = 0;
683out:
684 blk_unref(blk);
685 return ret;
686}
687
688int bdrv_create_file(const char *filename, QemuOpts *opts, Error **errp)
689{
729222af 690 QemuOpts *protocol_opts;
fd17146c 691 BlockDriver *drv;
729222af
SG
692 QDict *qdict;
693 int ret;
fd17146c 694
b65a5e12 695 drv = bdrv_find_protocol(filename, true, errp);
84a12e66 696 if (drv == NULL) {
16905d71 697 return -ENOENT;
84a12e66
CH
698 }
699
729222af
SG
700 if (!drv->create_opts) {
701 error_setg(errp, "Driver '%s' does not support image creation",
702 drv->format_name);
703 return -ENOTSUP;
704 }
705
706 /*
707 * 'opts' contains a QemuOptsList with a combination of format and protocol
708 * default values.
709 *
710 * The format properly removes its options, but the default values remain
711 * in 'opts->list'. So if the protocol has options with the same name
712 * (e.g. rbd has 'cluster_size' as qcow2), it will see the default values
713 * of the format, since for overlapping options, the format wins.
714 *
715 * To avoid this issue, lets convert QemuOpts to QDict, in this way we take
716 * only the set options, and then convert it back to QemuOpts, using the
717 * create_opts of the protocol. So the new QemuOpts, will contain only the
718 * protocol defaults.
719 */
720 qdict = qemu_opts_to_qdict(opts, NULL);
721 protocol_opts = qemu_opts_from_qdict(drv->create_opts, qdict, errp);
722 if (protocol_opts == NULL) {
723 ret = -EINVAL;
724 goto out;
725 }
726
727 ret = bdrv_create(drv, filename, protocol_opts, errp);
728out:
729 qemu_opts_del(protocol_opts);
730 qobject_unref(qdict);
731 return ret;
84a12e66
CH
732}
733
e1d7f8bb
DHB
734int coroutine_fn bdrv_co_delete_file(BlockDriverState *bs, Error **errp)
735{
736 Error *local_err = NULL;
737 int ret;
738
739 assert(bs != NULL);
740
741 if (!bs->drv) {
742 error_setg(errp, "Block node '%s' is not opened", bs->filename);
743 return -ENOMEDIUM;
744 }
745
746 if (!bs->drv->bdrv_co_delete_file) {
747 error_setg(errp, "Driver '%s' does not support image deletion",
748 bs->drv->format_name);
749 return -ENOTSUP;
750 }
751
752 ret = bs->drv->bdrv_co_delete_file(bs, &local_err);
753 if (ret < 0) {
754 error_propagate(errp, local_err);
755 }
756
757 return ret;
758}
759
a890f08e
ML
760void coroutine_fn bdrv_co_delete_file_noerr(BlockDriverState *bs)
761{
762 Error *local_err = NULL;
763 int ret;
764
765 if (!bs) {
766 return;
767 }
768
769 ret = bdrv_co_delete_file(bs, &local_err);
770 /*
771 * ENOTSUP will happen if the block driver doesn't support
772 * the 'bdrv_co_delete_file' interface. This is a predictable
773 * scenario and shouldn't be reported back to the user.
774 */
775 if (ret == -ENOTSUP) {
776 error_free(local_err);
777 } else if (ret < 0) {
778 error_report_err(local_err);
779 }
780}
781
892b7de8
ET
782/**
783 * Try to get @bs's logical and physical block size.
784 * On success, store them in @bsz struct and return 0.
785 * On failure return -errno.
786 * @bs must not be empty.
787 */
788int bdrv_probe_blocksizes(BlockDriverState *bs, BlockSizes *bsz)
789{
790 BlockDriver *drv = bs->drv;
93393e69 791 BlockDriverState *filtered = bdrv_filter_bs(bs);
892b7de8
ET
792
793 if (drv && drv->bdrv_probe_blocksizes) {
794 return drv->bdrv_probe_blocksizes(bs, bsz);
93393e69
HR
795 } else if (filtered) {
796 return bdrv_probe_blocksizes(filtered, bsz);
892b7de8
ET
797 }
798
799 return -ENOTSUP;
800}
801
802/**
803 * Try to get @bs's geometry (cyls, heads, sectors).
804 * On success, store them in @geo struct and return 0.
805 * On failure return -errno.
806 * @bs must not be empty.
807 */
808int bdrv_probe_geometry(BlockDriverState *bs, HDGeometry *geo)
809{
810 BlockDriver *drv = bs->drv;
93393e69 811 BlockDriverState *filtered = bdrv_filter_bs(bs);
892b7de8
ET
812
813 if (drv && drv->bdrv_probe_geometry) {
814 return drv->bdrv_probe_geometry(bs, geo);
93393e69
HR
815 } else if (filtered) {
816 return bdrv_probe_geometry(filtered, geo);
892b7de8
ET
817 }
818
819 return -ENOTSUP;
820}
821
eba25057
JM
822/*
823 * Create a uniquely-named empty temporary file.
824 * Return 0 upon success, otherwise a negative errno value.
825 */
826int get_tmp_filename(char *filename, int size)
d5249393 827{
eba25057 828#ifdef _WIN32
3b9f94e1 829 char temp_dir[MAX_PATH];
eba25057
JM
830 /* GetTempFileName requires that its output buffer (4th param)
831 have length MAX_PATH or greater. */
832 assert(size >= MAX_PATH);
833 return (GetTempPath(MAX_PATH, temp_dir)
834 && GetTempFileName(temp_dir, "qem", 0, filename)
835 ? 0 : -GetLastError());
d5249393 836#else
67b915a5 837 int fd;
7ccfb2eb 838 const char *tmpdir;
0badc1ee 839 tmpdir = getenv("TMPDIR");
69bef793
AS
840 if (!tmpdir) {
841 tmpdir = "/var/tmp";
842 }
eba25057
JM
843 if (snprintf(filename, size, "%s/vl.XXXXXX", tmpdir) >= size) {
844 return -EOVERFLOW;
845 }
ea2384d3 846 fd = mkstemp(filename);
fe235a06
DH
847 if (fd < 0) {
848 return -errno;
849 }
850 if (close(fd) != 0) {
851 unlink(filename);
eba25057
JM
852 return -errno;
853 }
854 return 0;
d5249393 855#endif
eba25057 856}
fc01f7e7 857
84a12e66
CH
858/*
859 * Detect host devices. By convention, /dev/cdrom[N] is always
860 * recognized as a host CDROM.
861 */
862static BlockDriver *find_hdev_driver(const char *filename)
863{
864 int score_max = 0, score;
865 BlockDriver *drv = NULL, *d;
866
867 QLIST_FOREACH(d, &bdrv_drivers, list) {
868 if (d->bdrv_probe_device) {
869 score = d->bdrv_probe_device(filename);
870 if (score > score_max) {
871 score_max = score;
872 drv = d;
873 }
874 }
875 }
876
877 return drv;
878}
879
88d88798
MM
880static BlockDriver *bdrv_do_find_protocol(const char *protocol)
881{
882 BlockDriver *drv1;
883
884 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
885 if (drv1->protocol_name && !strcmp(drv1->protocol_name, protocol)) {
886 return drv1;
887 }
888 }
889
890 return NULL;
891}
892
98289620 893BlockDriver *bdrv_find_protocol(const char *filename,
b65a5e12
HR
894 bool allow_protocol_prefix,
895 Error **errp)
83f64091
FB
896{
897 BlockDriver *drv1;
898 char protocol[128];
1cec71e3 899 int len;
83f64091 900 const char *p;
88d88798 901 int i;
19cb3738 902
66f82cee
KW
903 /* TODO Drivers without bdrv_file_open must be specified explicitly */
904
39508e7a
CH
905 /*
906 * XXX(hch): we really should not let host device detection
907 * override an explicit protocol specification, but moving this
908 * later breaks access to device names with colons in them.
909 * Thanks to the brain-dead persistent naming schemes on udev-
910 * based Linux systems those actually are quite common.
911 */
912 drv1 = find_hdev_driver(filename);
913 if (drv1) {
914 return drv1;
915 }
916
98289620 917 if (!path_has_protocol(filename) || !allow_protocol_prefix) {
ef810437 918 return &bdrv_file;
84a12e66 919 }
98289620 920
9e0b22f4
SH
921 p = strchr(filename, ':');
922 assert(p != NULL);
1cec71e3
AL
923 len = p - filename;
924 if (len > sizeof(protocol) - 1)
925 len = sizeof(protocol) - 1;
926 memcpy(protocol, filename, len);
927 protocol[len] = '\0';
88d88798
MM
928
929 drv1 = bdrv_do_find_protocol(protocol);
930 if (drv1) {
931 return drv1;
932 }
933
934 for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); ++i) {
935 if (block_driver_modules[i].protocol_name &&
936 !strcmp(block_driver_modules[i].protocol_name, protocol)) {
937 block_module_load_one(block_driver_modules[i].library_name);
938 break;
8a22f02a 939 }
83f64091 940 }
b65a5e12 941
88d88798
MM
942 drv1 = bdrv_do_find_protocol(protocol);
943 if (!drv1) {
944 error_setg(errp, "Unknown protocol '%s'", protocol);
945 }
946 return drv1;
83f64091
FB
947}
948
c6684249
MA
949/*
950 * Guess image format by probing its contents.
951 * This is not a good idea when your image is raw (CVE-2008-2004), but
952 * we do it anyway for backward compatibility.
953 *
954 * @buf contains the image's first @buf_size bytes.
7cddd372
KW
955 * @buf_size is the buffer size in bytes (generally BLOCK_PROBE_BUF_SIZE,
956 * but can be smaller if the image file is smaller)
c6684249
MA
957 * @filename is its filename.
958 *
959 * For all block drivers, call the bdrv_probe() method to get its
960 * probing score.
961 * Return the first block driver with the highest probing score.
962 */
38f3ef57
KW
963BlockDriver *bdrv_probe_all(const uint8_t *buf, int buf_size,
964 const char *filename)
c6684249
MA
965{
966 int score_max = 0, score;
967 BlockDriver *drv = NULL, *d;
968
969 QLIST_FOREACH(d, &bdrv_drivers, list) {
970 if (d->bdrv_probe) {
971 score = d->bdrv_probe(buf, buf_size, filename);
972 if (score > score_max) {
973 score_max = score;
974 drv = d;
975 }
976 }
977 }
978
979 return drv;
980}
981
5696c6e3 982static int find_image_format(BlockBackend *file, const char *filename,
34b5d2c6 983 BlockDriver **pdrv, Error **errp)
f3a5d3f8 984{
c6684249 985 BlockDriver *drv;
7cddd372 986 uint8_t buf[BLOCK_PROBE_BUF_SIZE];
f500a6d3 987 int ret = 0;
f8ea0b00 988
08a00559 989 /* Return the raw BlockDriver * to scsi-generic devices or empty drives */
5696c6e3 990 if (blk_is_sg(file) || !blk_is_inserted(file) || blk_getlength(file) == 0) {
ef810437 991 *pdrv = &bdrv_raw;
c98ac35d 992 return ret;
1a396859 993 }
f8ea0b00 994
5696c6e3 995 ret = blk_pread(file, 0, buf, sizeof(buf));
83f64091 996 if (ret < 0) {
34b5d2c6
HR
997 error_setg_errno(errp, -ret, "Could not read image for determining its "
998 "format");
c98ac35d
SW
999 *pdrv = NULL;
1000 return ret;
83f64091
FB
1001 }
1002
c6684249 1003 drv = bdrv_probe_all(buf, ret, filename);
c98ac35d 1004 if (!drv) {
34b5d2c6
HR
1005 error_setg(errp, "Could not determine image format: No compatible "
1006 "driver found");
c98ac35d
SW
1007 ret = -ENOENT;
1008 }
1009 *pdrv = drv;
1010 return ret;
ea2384d3
FB
1011}
1012
51762288
SH
1013/**
1014 * Set the current 'total_sectors' value
65a9bb25 1015 * Return 0 on success, -errno on error.
51762288 1016 */
3d9f2d2a 1017int refresh_total_sectors(BlockDriverState *bs, int64_t hint)
51762288
SH
1018{
1019 BlockDriver *drv = bs->drv;
1020
d470ad42
HR
1021 if (!drv) {
1022 return -ENOMEDIUM;
1023 }
1024
396759ad 1025 /* Do not attempt drv->bdrv_getlength() on scsi-generic devices */
b192af8a 1026 if (bdrv_is_sg(bs))
396759ad
NB
1027 return 0;
1028
51762288
SH
1029 /* query actual device if possible, otherwise just trust the hint */
1030 if (drv->bdrv_getlength) {
1031 int64_t length = drv->bdrv_getlength(bs);
1032 if (length < 0) {
1033 return length;
1034 }
7e382003 1035 hint = DIV_ROUND_UP(length, BDRV_SECTOR_SIZE);
51762288
SH
1036 }
1037
1038 bs->total_sectors = hint;
8b117001
VSO
1039
1040 if (bs->total_sectors * BDRV_SECTOR_SIZE > BDRV_MAX_LENGTH) {
1041 return -EFBIG;
1042 }
1043
51762288
SH
1044 return 0;
1045}
1046
cddff5ba
KW
1047/**
1048 * Combines a QDict of new block driver @options with any missing options taken
1049 * from @old_options, so that leaving out an option defaults to its old value.
1050 */
1051static void bdrv_join_options(BlockDriverState *bs, QDict *options,
1052 QDict *old_options)
1053{
1054 if (bs->drv && bs->drv->bdrv_join_options) {
1055 bs->drv->bdrv_join_options(options, old_options);
1056 } else {
1057 qdict_join(options, old_options, false);
1058 }
1059}
1060
543770bd
AG
1061static BlockdevDetectZeroesOptions bdrv_parse_detect_zeroes(QemuOpts *opts,
1062 int open_flags,
1063 Error **errp)
1064{
1065 Error *local_err = NULL;
1066 char *value = qemu_opt_get_del(opts, "detect-zeroes");
1067 BlockdevDetectZeroesOptions detect_zeroes =
1068 qapi_enum_parse(&BlockdevDetectZeroesOptions_lookup, value,
1069 BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF, &local_err);
1070 g_free(value);
1071 if (local_err) {
1072 error_propagate(errp, local_err);
1073 return detect_zeroes;
1074 }
1075
1076 if (detect_zeroes == BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP &&
1077 !(open_flags & BDRV_O_UNMAP))
1078 {
1079 error_setg(errp, "setting detect-zeroes to unmap is not allowed "
1080 "without setting discard operation to unmap");
1081 }
1082
1083 return detect_zeroes;
1084}
1085
f80f2673
AM
1086/**
1087 * Set open flags for aio engine
1088 *
1089 * Return 0 on success, -1 if the engine specified is invalid
1090 */
1091int bdrv_parse_aio(const char *mode, int *flags)
1092{
1093 if (!strcmp(mode, "threads")) {
1094 /* do nothing, default */
1095 } else if (!strcmp(mode, "native")) {
1096 *flags |= BDRV_O_NATIVE_AIO;
1097#ifdef CONFIG_LINUX_IO_URING
1098 } else if (!strcmp(mode, "io_uring")) {
1099 *flags |= BDRV_O_IO_URING;
1100#endif
1101 } else {
1102 return -1;
1103 }
1104
1105 return 0;
1106}
1107
9e8f1835
PB
1108/**
1109 * Set open flags for a given discard mode
1110 *
1111 * Return 0 on success, -1 if the discard mode was invalid.
1112 */
1113int bdrv_parse_discard_flags(const char *mode, int *flags)
1114{
1115 *flags &= ~BDRV_O_UNMAP;
1116
1117 if (!strcmp(mode, "off") || !strcmp(mode, "ignore")) {
1118 /* do nothing */
1119 } else if (!strcmp(mode, "on") || !strcmp(mode, "unmap")) {
1120 *flags |= BDRV_O_UNMAP;
1121 } else {
1122 return -1;
1123 }
1124
1125 return 0;
1126}
1127
c3993cdc
SH
1128/**
1129 * Set open flags for a given cache mode
1130 *
1131 * Return 0 on success, -1 if the cache mode was invalid.
1132 */
53e8ae01 1133int bdrv_parse_cache_mode(const char *mode, int *flags, bool *writethrough)
c3993cdc
SH
1134{
1135 *flags &= ~BDRV_O_CACHE_MASK;
1136
1137 if (!strcmp(mode, "off") || !strcmp(mode, "none")) {
53e8ae01
KW
1138 *writethrough = false;
1139 *flags |= BDRV_O_NOCACHE;
92196b2f 1140 } else if (!strcmp(mode, "directsync")) {
53e8ae01 1141 *writethrough = true;
92196b2f 1142 *flags |= BDRV_O_NOCACHE;
c3993cdc 1143 } else if (!strcmp(mode, "writeback")) {
53e8ae01 1144 *writethrough = false;
c3993cdc 1145 } else if (!strcmp(mode, "unsafe")) {
53e8ae01 1146 *writethrough = false;
c3993cdc
SH
1147 *flags |= BDRV_O_NO_FLUSH;
1148 } else if (!strcmp(mode, "writethrough")) {
53e8ae01 1149 *writethrough = true;
c3993cdc
SH
1150 } else {
1151 return -1;
1152 }
1153
1154 return 0;
1155}
1156
b5411555
KW
1157static char *bdrv_child_get_parent_desc(BdrvChild *c)
1158{
1159 BlockDriverState *parent = c->opaque;
2c0a3acb 1160 return g_strdup_printf("node '%s'", bdrv_get_node_name(parent));
b5411555
KW
1161}
1162
20018e12
KW
1163static void bdrv_child_cb_drained_begin(BdrvChild *child)
1164{
1165 BlockDriverState *bs = child->opaque;
6cd5c9d7 1166 bdrv_do_drained_begin_quiesce(bs, NULL, false);
20018e12
KW
1167}
1168
89bd0305
KW
1169static bool bdrv_child_cb_drained_poll(BdrvChild *child)
1170{
1171 BlockDriverState *bs = child->opaque;
6cd5c9d7 1172 return bdrv_drain_poll(bs, false, NULL, false);
89bd0305
KW
1173}
1174
e037c09c
HR
1175static void bdrv_child_cb_drained_end(BdrvChild *child,
1176 int *drained_end_counter)
20018e12
KW
1177{
1178 BlockDriverState *bs = child->opaque;
e037c09c 1179 bdrv_drained_end_no_poll(bs, drained_end_counter);
20018e12
KW
1180}
1181
38701b6a
KW
1182static int bdrv_child_cb_inactivate(BdrvChild *child)
1183{
1184 BlockDriverState *bs = child->opaque;
1185 assert(bs->open_flags & BDRV_O_INACTIVE);
1186 return 0;
1187}
1188
5d231849
KW
1189static bool bdrv_child_cb_can_set_aio_ctx(BdrvChild *child, AioContext *ctx,
1190 GSList **ignore, Error **errp)
1191{
1192 BlockDriverState *bs = child->opaque;
1193 return bdrv_can_set_aio_context(bs, ctx, ignore, errp);
1194}
1195
53a7d041
KW
1196static void bdrv_child_cb_set_aio_ctx(BdrvChild *child, AioContext *ctx,
1197 GSList **ignore)
1198{
1199 BlockDriverState *bs = child->opaque;
1200 return bdrv_set_aio_context_ignore(bs, ctx, ignore);
1201}
1202
b1e6fc08 1203/*
73176bee
KW
1204 * Returns the options and flags that a temporary snapshot should get, based on
1205 * the originally requested flags (the originally requested image will have
1206 * flags like a backing file)
b1e6fc08 1207 */
73176bee
KW
1208static void bdrv_temp_snapshot_options(int *child_flags, QDict *child_options,
1209 int parent_flags, QDict *parent_options)
b1e6fc08 1210{
73176bee
KW
1211 *child_flags = (parent_flags & ~BDRV_O_SNAPSHOT) | BDRV_O_TEMPORARY;
1212
1213 /* For temporary files, unconditional cache=unsafe is fine */
73176bee
KW
1214 qdict_set_default_str(child_options, BDRV_OPT_CACHE_DIRECT, "off");
1215 qdict_set_default_str(child_options, BDRV_OPT_CACHE_NO_FLUSH, "on");
41869044 1216
3f48686f 1217 /* Copy the read-only and discard options from the parent */
f87a0e29 1218 qdict_copy_default(child_options, parent_options, BDRV_OPT_READ_ONLY);
3f48686f 1219 qdict_copy_default(child_options, parent_options, BDRV_OPT_DISCARD);
f87a0e29 1220
41869044
KW
1221 /* aio=native doesn't work for cache.direct=off, so disable it for the
1222 * temporary snapshot */
1223 *child_flags &= ~BDRV_O_NATIVE_AIO;
b1e6fc08
KW
1224}
1225
db95dbba
KW
1226static void bdrv_backing_attach(BdrvChild *c)
1227{
1228 BlockDriverState *parent = c->opaque;
1229 BlockDriverState *backing_hd = c->bs;
1230
1231 assert(!parent->backing_blocker);
1232 error_setg(&parent->backing_blocker,
1233 "node is used as backing hd of '%s'",
1234 bdrv_get_device_or_node_name(parent));
1235
f30c66ba
HR
1236 bdrv_refresh_filename(backing_hd);
1237
db95dbba 1238 parent->open_flags &= ~BDRV_O_NO_BACKING;
db95dbba
KW
1239
1240 bdrv_op_block_all(backing_hd, parent->backing_blocker);
1241 /* Otherwise we won't be able to commit or stream */
1242 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_COMMIT_TARGET,
1243 parent->backing_blocker);
1244 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_STREAM,
1245 parent->backing_blocker);
1246 /*
1247 * We do backup in 3 ways:
1248 * 1. drive backup
1249 * The target bs is new opened, and the source is top BDS
1250 * 2. blockdev backup
1251 * Both the source and the target are top BDSes.
1252 * 3. internal backup(used for block replication)
1253 * Both the source and the target are backing file
1254 *
1255 * In case 1 and 2, neither the source nor the target is the backing file.
1256 * In case 3, we will block the top BDS, so there is only one block job
1257 * for the top BDS and its backing chain.
1258 */
1259 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_BACKUP_SOURCE,
1260 parent->backing_blocker);
1261 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_BACKUP_TARGET,
1262 parent->backing_blocker);
ca2f1234 1263}
d736f119 1264
db95dbba
KW
1265static void bdrv_backing_detach(BdrvChild *c)
1266{
1267 BlockDriverState *parent = c->opaque;
1268
1269 assert(parent->backing_blocker);
1270 bdrv_op_unblock_all(c->bs, parent->backing_blocker);
1271 error_free(parent->backing_blocker);
1272 parent->backing_blocker = NULL;
48e08288 1273}
d736f119 1274
6858eba0
KW
1275static int bdrv_backing_update_filename(BdrvChild *c, BlockDriverState *base,
1276 const char *filename, Error **errp)
1277{
1278 BlockDriverState *parent = c->opaque;
e94d3dba 1279 bool read_only = bdrv_is_read_only(parent);
6858eba0
KW
1280 int ret;
1281
e94d3dba
AG
1282 if (read_only) {
1283 ret = bdrv_reopen_set_read_only(parent, false, errp);
61f09cea
KW
1284 if (ret < 0) {
1285 return ret;
1286 }
1287 }
1288
6858eba0 1289 ret = bdrv_change_backing_file(parent, filename,
e54ee1b3
EB
1290 base->drv ? base->drv->format_name : "",
1291 false);
6858eba0 1292 if (ret < 0) {
64730694 1293 error_setg_errno(errp, -ret, "Could not update backing file link");
6858eba0
KW
1294 }
1295
e94d3dba
AG
1296 if (read_only) {
1297 bdrv_reopen_set_read_only(parent, true, NULL);
61f09cea
KW
1298 }
1299
6858eba0
KW
1300 return ret;
1301}
1302
fae8bd39
HR
1303/*
1304 * Returns the options and flags that a generic child of a BDS should
1305 * get, based on the given options and flags for the parent BDS.
1306 */
00ff7ffd
HR
1307static void bdrv_inherited_options(BdrvChildRole role, bool parent_is_format,
1308 int *child_flags, QDict *child_options,
1309 int parent_flags, QDict *parent_options)
fae8bd39
HR
1310{
1311 int flags = parent_flags;
1312
1313 /*
1314 * First, decide whether to set, clear, or leave BDRV_O_PROTOCOL.
1315 * Generally, the question to answer is: Should this child be
1316 * format-probed by default?
1317 */
1318
1319 /*
1320 * Pure and non-filtered data children of non-format nodes should
1321 * be probed by default (even when the node itself has BDRV_O_PROTOCOL
1322 * set). This only affects a very limited set of drivers (namely
1323 * quorum and blkverify when this comment was written).
1324 * Force-clear BDRV_O_PROTOCOL then.
1325 */
1326 if (!parent_is_format &&
1327 (role & BDRV_CHILD_DATA) &&
1328 !(role & (BDRV_CHILD_METADATA | BDRV_CHILD_FILTERED)))
1329 {
1330 flags &= ~BDRV_O_PROTOCOL;
1331 }
1332
1333 /*
1334 * All children of format nodes (except for COW children) and all
1335 * metadata children in general should never be format-probed.
1336 * Force-set BDRV_O_PROTOCOL then.
1337 */
1338 if ((parent_is_format && !(role & BDRV_CHILD_COW)) ||
1339 (role & BDRV_CHILD_METADATA))
1340 {
1341 flags |= BDRV_O_PROTOCOL;
1342 }
1343
1344 /*
1345 * If the cache mode isn't explicitly set, inherit direct and no-flush from
1346 * the parent.
1347 */
1348 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_DIRECT);
1349 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_NO_FLUSH);
1350 qdict_copy_default(child_options, parent_options, BDRV_OPT_FORCE_SHARE);
1351
1352 if (role & BDRV_CHILD_COW) {
1353 /* backing files are opened read-only by default */
1354 qdict_set_default_str(child_options, BDRV_OPT_READ_ONLY, "on");
1355 qdict_set_default_str(child_options, BDRV_OPT_AUTO_READ_ONLY, "off");
1356 } else {
1357 /* Inherit the read-only option from the parent if it's not set */
1358 qdict_copy_default(child_options, parent_options, BDRV_OPT_READ_ONLY);
1359 qdict_copy_default(child_options, parent_options,
1360 BDRV_OPT_AUTO_READ_ONLY);
1361 }
1362
1363 /*
1364 * bdrv_co_pdiscard() respects unmap policy for the parent, so we
1365 * can default to enable it on lower layers regardless of the
1366 * parent option.
1367 */
1368 qdict_set_default_str(child_options, BDRV_OPT_DISCARD, "unmap");
1369
1370 /* Clear flags that only apply to the top layer */
1371 flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_COPY_ON_READ);
1372
1373 if (role & BDRV_CHILD_METADATA) {
1374 flags &= ~BDRV_O_NO_IO;
1375 }
1376 if (role & BDRV_CHILD_COW) {
1377 flags &= ~BDRV_O_TEMPORARY;
1378 }
1379
1380 *child_flags = flags;
1381}
1382
ca2f1234
HR
1383static void bdrv_child_cb_attach(BdrvChild *child)
1384{
1385 BlockDriverState *bs = child->opaque;
1386
1387 if (child->role & BDRV_CHILD_COW) {
1388 bdrv_backing_attach(child);
1389 }
1390
1391 bdrv_apply_subtree_drain(child, bs);
1392}
1393
48e08288
HR
1394static void bdrv_child_cb_detach(BdrvChild *child)
1395{
1396 BlockDriverState *bs = child->opaque;
1397
1398 if (child->role & BDRV_CHILD_COW) {
1399 bdrv_backing_detach(child);
1400 }
1401
1402 bdrv_unapply_subtree_drain(child, bs);
1403}
1404
43483550
HR
1405static int bdrv_child_cb_update_filename(BdrvChild *c, BlockDriverState *base,
1406 const char *filename, Error **errp)
1407{
1408 if (c->role & BDRV_CHILD_COW) {
1409 return bdrv_backing_update_filename(c, base, filename, errp);
1410 }
1411 return 0;
1412}
1413
fb62b588 1414AioContext *child_of_bds_get_parent_aio_context(BdrvChild *c)
3ca1f322
VSO
1415{
1416 BlockDriverState *bs = c->opaque;
1417
1418 return bdrv_get_aio_context(bs);
1419}
1420
43483550
HR
1421const BdrvChildClass child_of_bds = {
1422 .parent_is_bds = true,
1423 .get_parent_desc = bdrv_child_get_parent_desc,
1424 .inherit_options = bdrv_inherited_options,
1425 .drained_begin = bdrv_child_cb_drained_begin,
1426 .drained_poll = bdrv_child_cb_drained_poll,
1427 .drained_end = bdrv_child_cb_drained_end,
1428 .attach = bdrv_child_cb_attach,
1429 .detach = bdrv_child_cb_detach,
1430 .inactivate = bdrv_child_cb_inactivate,
1431 .can_set_aio_ctx = bdrv_child_cb_can_set_aio_ctx,
1432 .set_aio_ctx = bdrv_child_cb_set_aio_ctx,
1433 .update_filename = bdrv_child_cb_update_filename,
fb62b588 1434 .get_parent_aio_context = child_of_bds_get_parent_aio_context,
43483550
HR
1435};
1436
3ca1f322
VSO
1437AioContext *bdrv_child_get_parent_aio_context(BdrvChild *c)
1438{
1439 return c->klass->get_parent_aio_context(c);
1440}
1441
7b272452
KW
1442static int bdrv_open_flags(BlockDriverState *bs, int flags)
1443{
61de4c68 1444 int open_flags = flags;
7b272452
KW
1445
1446 /*
1447 * Clear flags that are internal to the block layer before opening the
1448 * image.
1449 */
20cca275 1450 open_flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_PROTOCOL);
7b272452 1451
7b272452
KW
1452 return open_flags;
1453}
1454
91a097e7
KW
1455static void update_flags_from_options(int *flags, QemuOpts *opts)
1456{
2a3d4331 1457 *flags &= ~(BDRV_O_CACHE_MASK | BDRV_O_RDWR | BDRV_O_AUTO_RDONLY);
91a097e7 1458
57f9db9a 1459 if (qemu_opt_get_bool_del(opts, BDRV_OPT_CACHE_NO_FLUSH, false)) {
91a097e7
KW
1460 *flags |= BDRV_O_NO_FLUSH;
1461 }
1462
57f9db9a 1463 if (qemu_opt_get_bool_del(opts, BDRV_OPT_CACHE_DIRECT, false)) {
91a097e7
KW
1464 *flags |= BDRV_O_NOCACHE;
1465 }
f87a0e29 1466
57f9db9a 1467 if (!qemu_opt_get_bool_del(opts, BDRV_OPT_READ_ONLY, false)) {
f87a0e29
AG
1468 *flags |= BDRV_O_RDWR;
1469 }
1470
e35bdc12
KW
1471 if (qemu_opt_get_bool_del(opts, BDRV_OPT_AUTO_READ_ONLY, false)) {
1472 *flags |= BDRV_O_AUTO_RDONLY;
1473 }
91a097e7
KW
1474}
1475
1476static void update_options_from_flags(QDict *options, int flags)
1477{
91a097e7 1478 if (!qdict_haskey(options, BDRV_OPT_CACHE_DIRECT)) {
46f5ac20 1479 qdict_put_bool(options, BDRV_OPT_CACHE_DIRECT, flags & BDRV_O_NOCACHE);
91a097e7
KW
1480 }
1481 if (!qdict_haskey(options, BDRV_OPT_CACHE_NO_FLUSH)) {
46f5ac20
EB
1482 qdict_put_bool(options, BDRV_OPT_CACHE_NO_FLUSH,
1483 flags & BDRV_O_NO_FLUSH);
91a097e7 1484 }
f87a0e29 1485 if (!qdict_haskey(options, BDRV_OPT_READ_ONLY)) {
46f5ac20 1486 qdict_put_bool(options, BDRV_OPT_READ_ONLY, !(flags & BDRV_O_RDWR));
f87a0e29 1487 }
e35bdc12
KW
1488 if (!qdict_haskey(options, BDRV_OPT_AUTO_READ_ONLY)) {
1489 qdict_put_bool(options, BDRV_OPT_AUTO_READ_ONLY,
1490 flags & BDRV_O_AUTO_RDONLY);
1491 }
91a097e7
KW
1492}
1493
636ea370
KW
1494static void bdrv_assign_node_name(BlockDriverState *bs,
1495 const char *node_name,
1496 Error **errp)
6913c0c2 1497{
15489c76 1498 char *gen_node_name = NULL;
6913c0c2 1499
15489c76
JC
1500 if (!node_name) {
1501 node_name = gen_node_name = id_generate(ID_BLOCK);
1502 } else if (!id_wellformed(node_name)) {
1503 /*
1504 * Check for empty string or invalid characters, but not if it is
1505 * generated (generated names use characters not available to the user)
1506 */
785ec4b1 1507 error_setg(errp, "Invalid node-name: '%s'", node_name);
636ea370 1508 return;
6913c0c2
BC
1509 }
1510
0c5e94ee 1511 /* takes care of avoiding namespaces collisions */
7f06d47e 1512 if (blk_by_name(node_name)) {
0c5e94ee
BC
1513 error_setg(errp, "node-name=%s is conflicting with a device id",
1514 node_name);
15489c76 1515 goto out;
0c5e94ee
BC
1516 }
1517
6913c0c2
BC
1518 /* takes care of avoiding duplicates node names */
1519 if (bdrv_find_node(node_name)) {
785ec4b1 1520 error_setg(errp, "Duplicate nodes with node-name='%s'", node_name);
15489c76 1521 goto out;
6913c0c2
BC
1522 }
1523
824808dd
KW
1524 /* Make sure that the node name isn't truncated */
1525 if (strlen(node_name) >= sizeof(bs->node_name)) {
1526 error_setg(errp, "Node name too long");
1527 goto out;
1528 }
1529
6913c0c2
BC
1530 /* copy node name into the bs and insert it into the graph list */
1531 pstrcpy(bs->node_name, sizeof(bs->node_name), node_name);
1532 QTAILQ_INSERT_TAIL(&graph_bdrv_states, bs, node_list);
15489c76
JC
1533out:
1534 g_free(gen_node_name);
6913c0c2
BC
1535}
1536
01a56501
KW
1537static int bdrv_open_driver(BlockDriverState *bs, BlockDriver *drv,
1538 const char *node_name, QDict *options,
1539 int open_flags, Error **errp)
1540{
1541 Error *local_err = NULL;
0f12264e 1542 int i, ret;
01a56501
KW
1543
1544 bdrv_assign_node_name(bs, node_name, &local_err);
1545 if (local_err) {
1546 error_propagate(errp, local_err);
1547 return -EINVAL;
1548 }
1549
1550 bs->drv = drv;
1551 bs->opaque = g_malloc0(drv->instance_size);
1552
1553 if (drv->bdrv_file_open) {
1554 assert(!drv->bdrv_needs_filename || bs->filename[0]);
1555 ret = drv->bdrv_file_open(bs, options, open_flags, &local_err);
680c7f96 1556 } else if (drv->bdrv_open) {
01a56501 1557 ret = drv->bdrv_open(bs, options, open_flags, &local_err);
680c7f96
KW
1558 } else {
1559 ret = 0;
01a56501
KW
1560 }
1561
1562 if (ret < 0) {
1563 if (local_err) {
1564 error_propagate(errp, local_err);
1565 } else if (bs->filename[0]) {
1566 error_setg_errno(errp, -ret, "Could not open '%s'", bs->filename);
1567 } else {
1568 error_setg_errno(errp, -ret, "Could not open image");
1569 }
180ca19a 1570 goto open_failed;
01a56501
KW
1571 }
1572
1573 ret = refresh_total_sectors(bs, bs->total_sectors);
1574 if (ret < 0) {
1575 error_setg_errno(errp, -ret, "Could not refresh total sector count");
180ca19a 1576 return ret;
01a56501
KW
1577 }
1578
1e4c797c 1579 bdrv_refresh_limits(bs, NULL, &local_err);
01a56501
KW
1580 if (local_err) {
1581 error_propagate(errp, local_err);
180ca19a 1582 return -EINVAL;
01a56501
KW
1583 }
1584
1585 assert(bdrv_opt_mem_align(bs) != 0);
1586 assert(bdrv_min_mem_align(bs) != 0);
1587 assert(is_power_of_2(bs->bl.request_alignment));
1588
0f12264e
KW
1589 for (i = 0; i < bs->quiesce_counter; i++) {
1590 if (drv->bdrv_co_drain_begin) {
1591 drv->bdrv_co_drain_begin(bs);
1592 }
1593 }
1594
01a56501 1595 return 0;
180ca19a
MP
1596open_failed:
1597 bs->drv = NULL;
1598 if (bs->file != NULL) {
1599 bdrv_unref_child(bs, bs->file);
1600 bs->file = NULL;
1601 }
01a56501
KW
1602 g_free(bs->opaque);
1603 bs->opaque = NULL;
01a56501
KW
1604 return ret;
1605}
1606
680c7f96
KW
1607BlockDriverState *bdrv_new_open_driver(BlockDriver *drv, const char *node_name,
1608 int flags, Error **errp)
1609{
1610 BlockDriverState *bs;
1611 int ret;
1612
1613 bs = bdrv_new();
1614 bs->open_flags = flags;
1615 bs->explicit_options = qdict_new();
1616 bs->options = qdict_new();
1617 bs->opaque = NULL;
1618
1619 update_options_from_flags(bs->options, flags);
1620
1621 ret = bdrv_open_driver(bs, drv, node_name, bs->options, flags, errp);
1622 if (ret < 0) {
cb3e7f08 1623 qobject_unref(bs->explicit_options);
180ca19a 1624 bs->explicit_options = NULL;
cb3e7f08 1625 qobject_unref(bs->options);
180ca19a 1626 bs->options = NULL;
680c7f96
KW
1627 bdrv_unref(bs);
1628 return NULL;
1629 }
1630
1631 return bs;
1632}
1633
c5f3014b 1634QemuOptsList bdrv_runtime_opts = {
18edf289
KW
1635 .name = "bdrv_common",
1636 .head = QTAILQ_HEAD_INITIALIZER(bdrv_runtime_opts.head),
1637 .desc = {
1638 {
1639 .name = "node-name",
1640 .type = QEMU_OPT_STRING,
1641 .help = "Node name of the block device node",
1642 },
62392ebb
KW
1643 {
1644 .name = "driver",
1645 .type = QEMU_OPT_STRING,
1646 .help = "Block driver to use for the node",
1647 },
91a097e7
KW
1648 {
1649 .name = BDRV_OPT_CACHE_DIRECT,
1650 .type = QEMU_OPT_BOOL,
1651 .help = "Bypass software writeback cache on the host",
1652 },
1653 {
1654 .name = BDRV_OPT_CACHE_NO_FLUSH,
1655 .type = QEMU_OPT_BOOL,
1656 .help = "Ignore flush requests",
1657 },
f87a0e29
AG
1658 {
1659 .name = BDRV_OPT_READ_ONLY,
1660 .type = QEMU_OPT_BOOL,
1661 .help = "Node is opened in read-only mode",
1662 },
e35bdc12
KW
1663 {
1664 .name = BDRV_OPT_AUTO_READ_ONLY,
1665 .type = QEMU_OPT_BOOL,
1666 .help = "Node can become read-only if opening read-write fails",
1667 },
692e01a2
KW
1668 {
1669 .name = "detect-zeroes",
1670 .type = QEMU_OPT_STRING,
1671 .help = "try to optimize zero writes (off, on, unmap)",
1672 },
818584a4 1673 {
415bbca8 1674 .name = BDRV_OPT_DISCARD,
818584a4
KW
1675 .type = QEMU_OPT_STRING,
1676 .help = "discard operation (ignore/off, unmap/on)",
1677 },
5a9347c6
FZ
1678 {
1679 .name = BDRV_OPT_FORCE_SHARE,
1680 .type = QEMU_OPT_BOOL,
1681 .help = "always accept other writers (default: off)",
1682 },
18edf289
KW
1683 { /* end of list */ }
1684 },
1685};
1686
5a5e7f8c
ML
1687QemuOptsList bdrv_create_opts_simple = {
1688 .name = "simple-create-opts",
1689 .head = QTAILQ_HEAD_INITIALIZER(bdrv_create_opts_simple.head),
fd17146c
HR
1690 .desc = {
1691 {
1692 .name = BLOCK_OPT_SIZE,
1693 .type = QEMU_OPT_SIZE,
1694 .help = "Virtual disk size"
1695 },
1696 {
1697 .name = BLOCK_OPT_PREALLOC,
1698 .type = QEMU_OPT_STRING,
1699 .help = "Preallocation mode (allowed values: off)"
1700 },
1701 { /* end of list */ }
1702 }
1703};
1704
57915332
KW
1705/*
1706 * Common part for opening disk images and files
b6ad491a
KW
1707 *
1708 * Removes all processed options from *options.
57915332 1709 */
5696c6e3 1710static int bdrv_open_common(BlockDriverState *bs, BlockBackend *file,
82dc8b41 1711 QDict *options, Error **errp)
57915332
KW
1712{
1713 int ret, open_flags;
035fccdf 1714 const char *filename;
62392ebb 1715 const char *driver_name = NULL;
6913c0c2 1716 const char *node_name = NULL;
818584a4 1717 const char *discard;
18edf289 1718 QemuOpts *opts;
62392ebb 1719 BlockDriver *drv;
34b5d2c6 1720 Error *local_err = NULL;
307261b2 1721 bool ro;
57915332 1722
6405875c 1723 assert(bs->file == NULL);
707ff828 1724 assert(options != NULL && bs->options != options);
57915332 1725
62392ebb 1726 opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
af175e85 1727 if (!qemu_opts_absorb_qdict(opts, options, errp)) {
62392ebb
KW
1728 ret = -EINVAL;
1729 goto fail_opts;
1730 }
1731
9b7e8691
AG
1732 update_flags_from_options(&bs->open_flags, opts);
1733
62392ebb
KW
1734 driver_name = qemu_opt_get(opts, "driver");
1735 drv = bdrv_find_format(driver_name);
1736 assert(drv != NULL);
1737
5a9347c6
FZ
1738 bs->force_share = qemu_opt_get_bool(opts, BDRV_OPT_FORCE_SHARE, false);
1739
1740 if (bs->force_share && (bs->open_flags & BDRV_O_RDWR)) {
1741 error_setg(errp,
1742 BDRV_OPT_FORCE_SHARE
1743 "=on can only be used with read-only images");
1744 ret = -EINVAL;
1745 goto fail_opts;
1746 }
1747
45673671 1748 if (file != NULL) {
f30c66ba 1749 bdrv_refresh_filename(blk_bs(file));
5696c6e3 1750 filename = blk_bs(file)->filename;
45673671 1751 } else {
129c7d1c
MA
1752 /*
1753 * Caution: while qdict_get_try_str() is fine, getting
1754 * non-string types would require more care. When @options
1755 * come from -blockdev or blockdev_add, its members are typed
1756 * according to the QAPI schema, but when they come from
1757 * -drive, they're all QString.
1758 */
45673671
KW
1759 filename = qdict_get_try_str(options, "filename");
1760 }
1761
4a008240 1762 if (drv->bdrv_needs_filename && (!filename || !filename[0])) {
765003db
KW
1763 error_setg(errp, "The '%s' block driver requires a file name",
1764 drv->format_name);
18edf289
KW
1765 ret = -EINVAL;
1766 goto fail_opts;
6913c0c2 1767 }
6913c0c2 1768
82dc8b41
KW
1769 trace_bdrv_open_common(bs, filename ?: "", bs->open_flags,
1770 drv->format_name);
62392ebb 1771
307261b2
VSO
1772 ro = bdrv_is_read_only(bs);
1773
1774 if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv, ro)) {
1775 if (!ro && bdrv_is_whitelisted(drv, true)) {
8be25de6
KW
1776 ret = bdrv_apply_auto_read_only(bs, NULL, NULL);
1777 } else {
1778 ret = -ENOTSUP;
1779 }
1780 if (ret < 0) {
1781 error_setg(errp,
307261b2 1782 !ro && bdrv_is_whitelisted(drv, true)
8be25de6
KW
1783 ? "Driver '%s' can only be used for read-only devices"
1784 : "Driver '%s' is not whitelisted",
1785 drv->format_name);
1786 goto fail_opts;
1787 }
b64ec4e4 1788 }
57915332 1789
d3faa13e 1790 /* bdrv_new() and bdrv_close() make it so */
d73415a3 1791 assert(qatomic_read(&bs->copy_on_read) == 0);
d3faa13e 1792
82dc8b41 1793 if (bs->open_flags & BDRV_O_COPY_ON_READ) {
307261b2 1794 if (!ro) {
0ebd24e0
KW
1795 bdrv_enable_copy_on_read(bs);
1796 } else {
1797 error_setg(errp, "Can't use copy-on-read on read-only device");
18edf289
KW
1798 ret = -EINVAL;
1799 goto fail_opts;
0ebd24e0 1800 }
53fec9d3
SH
1801 }
1802
415bbca8 1803 discard = qemu_opt_get(opts, BDRV_OPT_DISCARD);
818584a4
KW
1804 if (discard != NULL) {
1805 if (bdrv_parse_discard_flags(discard, &bs->open_flags) != 0) {
1806 error_setg(errp, "Invalid discard option");
1807 ret = -EINVAL;
1808 goto fail_opts;
1809 }
1810 }
1811
543770bd
AG
1812 bs->detect_zeroes =
1813 bdrv_parse_detect_zeroes(opts, bs->open_flags, &local_err);
1814 if (local_err) {
1815 error_propagate(errp, local_err);
1816 ret = -EINVAL;
1817 goto fail_opts;
692e01a2
KW
1818 }
1819
c2ad1b0c
KW
1820 if (filename != NULL) {
1821 pstrcpy(bs->filename, sizeof(bs->filename), filename);
1822 } else {
1823 bs->filename[0] = '\0';
1824 }
91af7014 1825 pstrcpy(bs->exact_filename, sizeof(bs->exact_filename), bs->filename);
57915332 1826
66f82cee 1827 /* Open the image, either directly or using a protocol */
82dc8b41 1828 open_flags = bdrv_open_flags(bs, bs->open_flags);
01a56501 1829 node_name = qemu_opt_get(opts, "node-name");
57915332 1830
01a56501
KW
1831 assert(!drv->bdrv_file_open || file == NULL);
1832 ret = bdrv_open_driver(bs, drv, node_name, options, open_flags, errp);
51762288 1833 if (ret < 0) {
01a56501 1834 goto fail_opts;
3baca891
KW
1835 }
1836
18edf289 1837 qemu_opts_del(opts);
57915332
KW
1838 return 0;
1839
18edf289
KW
1840fail_opts:
1841 qemu_opts_del(opts);
57915332
KW
1842 return ret;
1843}
1844
5e5c4f63
KW
1845static QDict *parse_json_filename(const char *filename, Error **errp)
1846{
1847 QObject *options_obj;
1848 QDict *options;
1849 int ret;
1850
1851 ret = strstart(filename, "json:", &filename);
1852 assert(ret);
1853
5577fff7 1854 options_obj = qobject_from_json(filename, errp);
5e5c4f63 1855 if (!options_obj) {
5577fff7 1856 error_prepend(errp, "Could not parse the JSON options: ");
5e5c4f63
KW
1857 return NULL;
1858 }
1859
7dc847eb 1860 options = qobject_to(QDict, options_obj);
ca6b6e1e 1861 if (!options) {
cb3e7f08 1862 qobject_unref(options_obj);
5e5c4f63
KW
1863 error_setg(errp, "Invalid JSON object given");
1864 return NULL;
1865 }
1866
5e5c4f63
KW
1867 qdict_flatten(options);
1868
1869 return options;
1870}
1871
de3b53f0
KW
1872static void parse_json_protocol(QDict *options, const char **pfilename,
1873 Error **errp)
1874{
1875 QDict *json_options;
1876 Error *local_err = NULL;
1877
1878 /* Parse json: pseudo-protocol */
1879 if (!*pfilename || !g_str_has_prefix(*pfilename, "json:")) {
1880 return;
1881 }
1882
1883 json_options = parse_json_filename(*pfilename, &local_err);
1884 if (local_err) {
1885 error_propagate(errp, local_err);
1886 return;
1887 }
1888
1889 /* Options given in the filename have lower priority than options
1890 * specified directly */
1891 qdict_join(options, json_options, false);
cb3e7f08 1892 qobject_unref(json_options);
de3b53f0
KW
1893 *pfilename = NULL;
1894}
1895
b6ce07aa 1896/*
f54120ff
KW
1897 * Fills in default options for opening images and converts the legacy
1898 * filename/flags pair to option QDict entries.
53a29513
HR
1899 * The BDRV_O_PROTOCOL flag in *flags will be set or cleared accordingly if a
1900 * block driver has been specified explicitly.
b6ce07aa 1901 */
de3b53f0 1902static int bdrv_fill_options(QDict **options, const char *filename,
053e1578 1903 int *flags, Error **errp)
ea2384d3 1904{
c2ad1b0c 1905 const char *drvname;
53a29513 1906 bool protocol = *flags & BDRV_O_PROTOCOL;
e3fa4bfa 1907 bool parse_filename = false;
053e1578 1908 BlockDriver *drv = NULL;
34b5d2c6 1909 Error *local_err = NULL;
83f64091 1910
129c7d1c
MA
1911 /*
1912 * Caution: while qdict_get_try_str() is fine, getting non-string
1913 * types would require more care. When @options come from
1914 * -blockdev or blockdev_add, its members are typed according to
1915 * the QAPI schema, but when they come from -drive, they're all
1916 * QString.
1917 */
53a29513 1918 drvname = qdict_get_try_str(*options, "driver");
053e1578
HR
1919 if (drvname) {
1920 drv = bdrv_find_format(drvname);
1921 if (!drv) {
1922 error_setg(errp, "Unknown driver '%s'", drvname);
1923 return -ENOENT;
1924 }
1925 /* If the user has explicitly specified the driver, this choice should
1926 * override the BDRV_O_PROTOCOL flag */
1927 protocol = drv->bdrv_file_open;
53a29513
HR
1928 }
1929
1930 if (protocol) {
1931 *flags |= BDRV_O_PROTOCOL;
1932 } else {
1933 *flags &= ~BDRV_O_PROTOCOL;
1934 }
1935
91a097e7
KW
1936 /* Translate cache options from flags into options */
1937 update_options_from_flags(*options, *flags);
1938
035fccdf 1939 /* Fetch the file name from the options QDict if necessary */
17b005f1 1940 if (protocol && filename) {
f54120ff 1941 if (!qdict_haskey(*options, "filename")) {
46f5ac20 1942 qdict_put_str(*options, "filename", filename);
f54120ff
KW
1943 parse_filename = true;
1944 } else {
1945 error_setg(errp, "Can't specify 'file' and 'filename' options at "
1946 "the same time");
1947 return -EINVAL;
1948 }
035fccdf
KW
1949 }
1950
c2ad1b0c 1951 /* Find the right block driver */
129c7d1c 1952 /* See cautionary note on accessing @options above */
f54120ff 1953 filename = qdict_get_try_str(*options, "filename");
f54120ff 1954
053e1578
HR
1955 if (!drvname && protocol) {
1956 if (filename) {
1957 drv = bdrv_find_protocol(filename, parse_filename, errp);
17b005f1 1958 if (!drv) {
053e1578 1959 return -EINVAL;
17b005f1 1960 }
053e1578
HR
1961
1962 drvname = drv->format_name;
46f5ac20 1963 qdict_put_str(*options, "driver", drvname);
053e1578
HR
1964 } else {
1965 error_setg(errp, "Must specify either driver or file");
1966 return -EINVAL;
98289620 1967 }
c2ad1b0c
KW
1968 }
1969
17b005f1 1970 assert(drv || !protocol);
c2ad1b0c 1971
f54120ff 1972 /* Driver-specific filename parsing */
17b005f1 1973 if (drv && drv->bdrv_parse_filename && parse_filename) {
5acd9d81 1974 drv->bdrv_parse_filename(filename, *options, &local_err);
84d18f06 1975 if (local_err) {
34b5d2c6 1976 error_propagate(errp, local_err);
f54120ff 1977 return -EINVAL;
6963a30d 1978 }
cd5d031e
HR
1979
1980 if (!drv->bdrv_needs_filename) {
1981 qdict_del(*options, "filename");
cd5d031e 1982 }
6963a30d
KW
1983 }
1984
f54120ff
KW
1985 return 0;
1986}
1987
148eb13c
KW
1988typedef struct BlockReopenQueueEntry {
1989 bool prepared;
69b736e7 1990 bool perms_checked;
148eb13c 1991 BDRVReopenState state;
859443b0 1992 QTAILQ_ENTRY(BlockReopenQueueEntry) entry;
148eb13c
KW
1993} BlockReopenQueueEntry;
1994
1995/*
1996 * Return the flags that @bs will have after the reopens in @q have
1997 * successfully completed. If @q is NULL (or @bs is not contained in @q),
1998 * return the current flags.
1999 */
2000static int bdrv_reopen_get_flags(BlockReopenQueue *q, BlockDriverState *bs)
2001{
2002 BlockReopenQueueEntry *entry;
2003
2004 if (q != NULL) {
859443b0 2005 QTAILQ_FOREACH(entry, q, entry) {
148eb13c
KW
2006 if (entry->state.bs == bs) {
2007 return entry->state.flags;
2008 }
2009 }
2010 }
2011
2012 return bs->open_flags;
2013}
2014
2015/* Returns whether the image file can be written to after the reopen queue @q
2016 * has been successfully applied, or right now if @q is NULL. */
cc022140
HR
2017static bool bdrv_is_writable_after_reopen(BlockDriverState *bs,
2018 BlockReopenQueue *q)
148eb13c
KW
2019{
2020 int flags = bdrv_reopen_get_flags(q, bs);
2021
2022 return (flags & (BDRV_O_RDWR | BDRV_O_INACTIVE)) == BDRV_O_RDWR;
2023}
2024
cc022140
HR
2025/*
2026 * Return whether the BDS can be written to. This is not necessarily
2027 * the same as !bdrv_is_read_only(bs), as inactivated images may not
2028 * be written to but do not count as read-only images.
2029 */
2030bool bdrv_is_writable(BlockDriverState *bs)
2031{
2032 return bdrv_is_writable_after_reopen(bs, NULL);
2033}
2034
3bf416ba
VSO
2035static char *bdrv_child_user_desc(BdrvChild *c)
2036{
da261b69 2037 return c->klass->get_parent_desc(c);
3bf416ba
VSO
2038}
2039
30ebb9aa
VSO
2040/*
2041 * Check that @a allows everything that @b needs. @a and @b must reference same
2042 * child node.
2043 */
3bf416ba
VSO
2044static bool bdrv_a_allow_b(BdrvChild *a, BdrvChild *b, Error **errp)
2045{
30ebb9aa
VSO
2046 const char *child_bs_name;
2047 g_autofree char *a_user = NULL;
2048 g_autofree char *b_user = NULL;
2049 g_autofree char *perms = NULL;
2050
2051 assert(a->bs);
2052 assert(a->bs == b->bs);
3bf416ba
VSO
2053
2054 if ((b->perm & a->shared_perm) == b->perm) {
2055 return true;
2056 }
2057
30ebb9aa
VSO
2058 child_bs_name = bdrv_get_node_name(b->bs);
2059 a_user = bdrv_child_user_desc(a);
2060 b_user = bdrv_child_user_desc(b);
2061 perms = bdrv_perm_names(b->perm & ~a->shared_perm);
2062
2063 error_setg(errp, "Permission conflict on node '%s': permissions '%s' are "
2064 "both required by %s (uses node '%s' as '%s' child) and "
2065 "unshared by %s (uses node '%s' as '%s' child).",
2066 child_bs_name, perms,
2067 b_user, child_bs_name, b->name,
2068 a_user, child_bs_name, a->name);
3bf416ba
VSO
2069
2070 return false;
2071}
2072
9397c14f 2073static bool bdrv_parent_perms_conflict(BlockDriverState *bs, Error **errp)
3bf416ba
VSO
2074{
2075 BdrvChild *a, *b;
2076
2077 /*
2078 * During the loop we'll look at each pair twice. That's correct because
2079 * bdrv_a_allow_b() is asymmetric and we should check each pair in both
2080 * directions.
2081 */
2082 QLIST_FOREACH(a, &bs->parents, next_parent) {
2083 QLIST_FOREACH(b, &bs->parents, next_parent) {
9397c14f 2084 if (a == b) {
3bf416ba
VSO
2085 continue;
2086 }
2087
2088 if (!bdrv_a_allow_b(a, b, errp)) {
2089 return true;
2090 }
2091 }
2092 }
2093
2094 return false;
2095}
2096
ffd1a5a2 2097static void bdrv_child_perm(BlockDriverState *bs, BlockDriverState *child_bs,
e5d8a406
HR
2098 BdrvChild *c, BdrvChildRole role,
2099 BlockReopenQueue *reopen_queue,
ffd1a5a2
FZ
2100 uint64_t parent_perm, uint64_t parent_shared,
2101 uint64_t *nperm, uint64_t *nshared)
2102{
0b3ca76e 2103 assert(bs->drv && bs->drv->bdrv_child_perm);
e5d8a406 2104 bs->drv->bdrv_child_perm(bs, c, role, reopen_queue,
0b3ca76e
AG
2105 parent_perm, parent_shared,
2106 nperm, nshared);
e0995dc3 2107 /* TODO Take force_share from reopen_queue */
ffd1a5a2
FZ
2108 if (child_bs && child_bs->force_share) {
2109 *nshared = BLK_PERM_ALL;
2110 }
2111}
2112
bd57f8f7
VSO
2113/*
2114 * Adds the whole subtree of @bs (including @bs itself) to the @list (except for
2115 * nodes that are already in the @list, of course) so that final list is
2116 * topologically sorted. Return the result (GSList @list object is updated, so
2117 * don't use old reference after function call).
2118 *
2119 * On function start @list must be already topologically sorted and for any node
2120 * in the @list the whole subtree of the node must be in the @list as well. The
2121 * simplest way to satisfy this criteria: use only result of
2122 * bdrv_topological_dfs() or NULL as @list parameter.
2123 */
2124static GSList *bdrv_topological_dfs(GSList *list, GHashTable *found,
2125 BlockDriverState *bs)
2126{
2127 BdrvChild *child;
2128 g_autoptr(GHashTable) local_found = NULL;
2129
2130 if (!found) {
2131 assert(!list);
2132 found = local_found = g_hash_table_new(NULL, NULL);
2133 }
2134
2135 if (g_hash_table_contains(found, bs)) {
2136 return list;
2137 }
2138 g_hash_table_add(found, bs);
2139
2140 QLIST_FOREACH(child, &bs->children, next) {
2141 list = bdrv_topological_dfs(list, found, child->bs);
2142 }
2143
2144 return g_slist_prepend(list, bs);
2145}
2146
ecb776bd
VSO
2147typedef struct BdrvChildSetPermState {
2148 BdrvChild *child;
2149 uint64_t old_perm;
2150 uint64_t old_shared_perm;
2151} BdrvChildSetPermState;
b0defa83
VSO
2152
2153static void bdrv_child_set_perm_abort(void *opaque)
2154{
ecb776bd
VSO
2155 BdrvChildSetPermState *s = opaque;
2156
2157 s->child->perm = s->old_perm;
2158 s->child->shared_perm = s->old_shared_perm;
b0defa83
VSO
2159}
2160
2161static TransactionActionDrv bdrv_child_set_pem_drv = {
2162 .abort = bdrv_child_set_perm_abort,
ecb776bd 2163 .clean = g_free,
b0defa83
VSO
2164};
2165
ecb776bd
VSO
2166static void bdrv_child_set_perm(BdrvChild *c, uint64_t perm,
2167 uint64_t shared, Transaction *tran)
b0defa83 2168{
ecb776bd
VSO
2169 BdrvChildSetPermState *s = g_new(BdrvChildSetPermState, 1);
2170
2171 *s = (BdrvChildSetPermState) {
2172 .child = c,
2173 .old_perm = c->perm,
2174 .old_shared_perm = c->shared_perm,
2175 };
b0defa83
VSO
2176
2177 c->perm = perm;
2178 c->shared_perm = shared;
2179
ecb776bd 2180 tran_add(tran, &bdrv_child_set_pem_drv, s);
b0defa83
VSO
2181}
2182
2513ef59
VSO
2183static void bdrv_drv_set_perm_commit(void *opaque)
2184{
2185 BlockDriverState *bs = opaque;
2186 uint64_t cumulative_perms, cumulative_shared_perms;
2187
2188 if (bs->drv->bdrv_set_perm) {
2189 bdrv_get_cumulative_perm(bs, &cumulative_perms,
2190 &cumulative_shared_perms);
2191 bs->drv->bdrv_set_perm(bs, cumulative_perms, cumulative_shared_perms);
2192 }
2193}
2194
2195static void bdrv_drv_set_perm_abort(void *opaque)
2196{
2197 BlockDriverState *bs = opaque;
2198
2199 if (bs->drv->bdrv_abort_perm_update) {
2200 bs->drv->bdrv_abort_perm_update(bs);
2201 }
2202}
2203
2204TransactionActionDrv bdrv_drv_set_perm_drv = {
2205 .abort = bdrv_drv_set_perm_abort,
2206 .commit = bdrv_drv_set_perm_commit,
2207};
2208
2209static int bdrv_drv_set_perm(BlockDriverState *bs, uint64_t perm,
2210 uint64_t shared_perm, Transaction *tran,
2211 Error **errp)
2212{
2213 if (!bs->drv) {
2214 return 0;
2215 }
2216
2217 if (bs->drv->bdrv_check_perm) {
2218 int ret = bs->drv->bdrv_check_perm(bs, perm, shared_perm, errp);
2219 if (ret < 0) {
2220 return ret;
2221 }
2222 }
2223
2224 if (tran) {
2225 tran_add(tran, &bdrv_drv_set_perm_drv, bs);
2226 }
2227
2228 return 0;
2229}
2230
0978623e
VSO
2231typedef struct BdrvReplaceChildState {
2232 BdrvChild *child;
2233 BlockDriverState *old_bs;
2234} BdrvReplaceChildState;
2235
2236static void bdrv_replace_child_commit(void *opaque)
2237{
2238 BdrvReplaceChildState *s = opaque;
2239
2240 bdrv_unref(s->old_bs);
2241}
2242
2243static void bdrv_replace_child_abort(void *opaque)
2244{
2245 BdrvReplaceChildState *s = opaque;
2246 BlockDriverState *new_bs = s->child->bs;
2247
2248 /* old_bs reference is transparently moved from @s to @s->child */
2249 bdrv_replace_child_noperm(s->child, s->old_bs);
2250 bdrv_unref(new_bs);
2251}
2252
2253static TransactionActionDrv bdrv_replace_child_drv = {
2254 .commit = bdrv_replace_child_commit,
2255 .abort = bdrv_replace_child_abort,
2256 .clean = g_free,
2257};
2258
2259/*
4bf021db 2260 * bdrv_replace_child_tran
0978623e
VSO
2261 *
2262 * Note: real unref of old_bs is done only on commit.
4bf021db
VSO
2263 *
2264 * The function doesn't update permissions, caller is responsible for this.
0978623e 2265 */
4bf021db
VSO
2266static void bdrv_replace_child_tran(BdrvChild *child, BlockDriverState *new_bs,
2267 Transaction *tran)
0978623e
VSO
2268{
2269 BdrvReplaceChildState *s = g_new(BdrvReplaceChildState, 1);
2270 *s = (BdrvReplaceChildState) {
2271 .child = child,
2272 .old_bs = child->bs,
2273 };
2274 tran_add(tran, &bdrv_replace_child_drv, s);
2275
2276 if (new_bs) {
2277 bdrv_ref(new_bs);
2278 }
2279 bdrv_replace_child_noperm(child, new_bs);
2280 /* old_bs reference is transparently moved from @child to @s */
2281}
2282
33a610c3 2283/*
c20555e1
VSO
2284 * Refresh permissions in @bs subtree. The function is intended to be called
2285 * after some graph modification that was done without permission update.
33a610c3 2286 */
c20555e1
VSO
2287static int bdrv_node_refresh_perm(BlockDriverState *bs, BlockReopenQueue *q,
2288 Transaction *tran, Error **errp)
33a610c3
KW
2289{
2290 BlockDriver *drv = bs->drv;
2291 BdrvChild *c;
2292 int ret;
c20555e1
VSO
2293 uint64_t cumulative_perms, cumulative_shared_perms;
2294
2295 bdrv_get_cumulative_perm(bs, &cumulative_perms, &cumulative_shared_perms);
33a610c3
KW
2296
2297 /* Write permissions never work with read-only images */
2298 if ((cumulative_perms & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)) &&
cc022140 2299 !bdrv_is_writable_after_reopen(bs, q))
33a610c3 2300 {
481e0eee
HR
2301 if (!bdrv_is_writable_after_reopen(bs, NULL)) {
2302 error_setg(errp, "Block node is read-only");
2303 } else {
c20555e1
VSO
2304 error_setg(errp, "Read-only block node '%s' cannot support "
2305 "read-write users", bdrv_get_node_name(bs));
481e0eee
HR
2306 }
2307
33a610c3
KW
2308 return -EPERM;
2309 }
2310
9c60a5d1
KW
2311 /*
2312 * Unaligned requests will automatically be aligned to bl.request_alignment
2313 * and without RESIZE we can't extend requests to write to space beyond the
2314 * end of the image, so it's required that the image size is aligned.
2315 */
2316 if ((cumulative_perms & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)) &&
2317 !(cumulative_perms & BLK_PERM_RESIZE))
2318 {
2319 if ((bs->total_sectors * BDRV_SECTOR_SIZE) % bs->bl.request_alignment) {
2320 error_setg(errp, "Cannot get 'write' permission without 'resize': "
2321 "Image size is not a multiple of request "
2322 "alignment");
2323 return -EPERM;
2324 }
2325 }
2326
33a610c3
KW
2327 /* Check this node */
2328 if (!drv) {
2329 return 0;
2330 }
2331
b1d2bbeb 2332 ret = bdrv_drv_set_perm(bs, cumulative_perms, cumulative_shared_perms, tran,
2513ef59
VSO
2333 errp);
2334 if (ret < 0) {
2335 return ret;
33a610c3
KW
2336 }
2337
78e421c9 2338 /* Drivers that never have children can omit .bdrv_child_perm() */
33a610c3 2339 if (!drv->bdrv_child_perm) {
78e421c9 2340 assert(QLIST_EMPTY(&bs->children));
33a610c3
KW
2341 return 0;
2342 }
2343
2344 /* Check all children */
2345 QLIST_FOREACH(c, &bs->children, next) {
2346 uint64_t cur_perm, cur_shared;
9eab1544 2347
e5d8a406 2348 bdrv_child_perm(bs, c->bs, c, c->role, q,
ffd1a5a2
FZ
2349 cumulative_perms, cumulative_shared_perms,
2350 &cur_perm, &cur_shared);
ecb776bd 2351 bdrv_child_set_perm(c, cur_perm, cur_shared, tran);
bd57f8f7
VSO
2352 }
2353
2354 return 0;
2355}
3ef45e02 2356
25409807
VSO
2357static int bdrv_list_refresh_perms(GSList *list, BlockReopenQueue *q,
2358 Transaction *tran, Error **errp)
bd57f8f7
VSO
2359{
2360 int ret;
b1d2bbeb 2361 BlockDriverState *bs;
bd57f8f7 2362
b1d2bbeb
VSO
2363 for ( ; list; list = list->next) {
2364 bs = list->data;
2365
9397c14f 2366 if (bdrv_parent_perms_conflict(bs, errp)) {
b1d2bbeb 2367 return -EINVAL;
bd57f8f7
VSO
2368 }
2369
c20555e1 2370 ret = bdrv_node_refresh_perm(bs, q, tran, errp);
33a610c3
KW
2371 if (ret < 0) {
2372 return ret;
2373 }
2374 }
2375
2376 return 0;
2377}
2378
c7a0f2be
KW
2379void bdrv_get_cumulative_perm(BlockDriverState *bs, uint64_t *perm,
2380 uint64_t *shared_perm)
33a610c3
KW
2381{
2382 BdrvChild *c;
2383 uint64_t cumulative_perms = 0;
2384 uint64_t cumulative_shared_perms = BLK_PERM_ALL;
2385
2386 QLIST_FOREACH(c, &bs->parents, next_parent) {
2387 cumulative_perms |= c->perm;
2388 cumulative_shared_perms &= c->shared_perm;
2389 }
2390
2391 *perm = cumulative_perms;
2392 *shared_perm = cumulative_shared_perms;
2393}
2394
5176196c 2395char *bdrv_perm_names(uint64_t perm)
d083319f
KW
2396{
2397 struct perm_name {
2398 uint64_t perm;
2399 const char *name;
2400 } permissions[] = {
2401 { BLK_PERM_CONSISTENT_READ, "consistent read" },
2402 { BLK_PERM_WRITE, "write" },
2403 { BLK_PERM_WRITE_UNCHANGED, "write unchanged" },
2404 { BLK_PERM_RESIZE, "resize" },
2405 { BLK_PERM_GRAPH_MOD, "change children" },
2406 { 0, NULL }
2407 };
2408
e2a7423a 2409 GString *result = g_string_sized_new(30);
d083319f
KW
2410 struct perm_name *p;
2411
2412 for (p = permissions; p->name; p++) {
2413 if (perm & p->perm) {
e2a7423a
AG
2414 if (result->len > 0) {
2415 g_string_append(result, ", ");
2416 }
2417 g_string_append(result, p->name);
d083319f
KW
2418 }
2419 }
2420
e2a7423a 2421 return g_string_free(result, FALSE);
d083319f
KW
2422}
2423
33a610c3 2424
071b474f 2425static int bdrv_refresh_perms(BlockDriverState *bs, Error **errp)
bb87e4d1
VSO
2426{
2427 int ret;
b1d2bbeb
VSO
2428 Transaction *tran = tran_new();
2429 g_autoptr(GSList) list = bdrv_topological_dfs(NULL, NULL, bs);
bb87e4d1 2430
b1d2bbeb
VSO
2431 ret = bdrv_list_refresh_perms(list, NULL, tran, errp);
2432 tran_finalize(tran, ret);
bb87e4d1 2433
b1d2bbeb 2434 return ret;
bb87e4d1
VSO
2435}
2436
33a610c3
KW
2437int bdrv_child_try_set_perm(BdrvChild *c, uint64_t perm, uint64_t shared,
2438 Error **errp)
2439{
1046779e 2440 Error *local_err = NULL;
83928dc4 2441 Transaction *tran = tran_new();
33a610c3
KW
2442 int ret;
2443
ecb776bd 2444 bdrv_child_set_perm(c, perm, shared, tran);
83928dc4
VSO
2445
2446 ret = bdrv_refresh_perms(c->bs, &local_err);
2447
2448 tran_finalize(tran, ret);
2449
33a610c3 2450 if (ret < 0) {
071b474f
VSO
2451 if ((perm & ~c->perm) || (c->shared_perm & ~shared)) {
2452 /* tighten permissions */
1046779e
HR
2453 error_propagate(errp, local_err);
2454 } else {
2455 /*
2456 * Our caller may intend to only loosen restrictions and
2457 * does not expect this function to fail. Errors are not
2458 * fatal in such a case, so we can just hide them from our
2459 * caller.
2460 */
2461 error_free(local_err);
2462 ret = 0;
2463 }
33a610c3
KW
2464 }
2465
83928dc4 2466 return ret;
d5e6f437
KW
2467}
2468
c1087f12
HR
2469int bdrv_child_refresh_perms(BlockDriverState *bs, BdrvChild *c, Error **errp)
2470{
2471 uint64_t parent_perms, parent_shared;
2472 uint64_t perms, shared;
2473
2474 bdrv_get_cumulative_perm(bs, &parent_perms, &parent_shared);
e5d8a406 2475 bdrv_child_perm(bs, c->bs, c, c->role, NULL,
bf8e925e 2476 parent_perms, parent_shared, &perms, &shared);
c1087f12
HR
2477
2478 return bdrv_child_try_set_perm(c, perms, shared, errp);
2479}
2480
87278af1
HR
2481/*
2482 * Default implementation for .bdrv_child_perm() for block filters:
2483 * Forward CONSISTENT_READ, WRITE, WRITE_UNCHANGED, and RESIZE to the
2484 * filtered child.
2485 */
2486static void bdrv_filter_default_perms(BlockDriverState *bs, BdrvChild *c,
87278af1
HR
2487 BdrvChildRole role,
2488 BlockReopenQueue *reopen_queue,
2489 uint64_t perm, uint64_t shared,
2490 uint64_t *nperm, uint64_t *nshared)
6a1b9ee1 2491{
e444fa83
KW
2492 *nperm = perm & DEFAULT_PERM_PASSTHROUGH;
2493 *nshared = (shared & DEFAULT_PERM_PASSTHROUGH) | DEFAULT_PERM_UNCHANGED;
6a1b9ee1
KW
2494}
2495
70082db4 2496static void bdrv_default_perms_for_cow(BlockDriverState *bs, BdrvChild *c,
70082db4
HR
2497 BdrvChildRole role,
2498 BlockReopenQueue *reopen_queue,
2499 uint64_t perm, uint64_t shared,
2500 uint64_t *nperm, uint64_t *nshared)
2501{
e5d8a406 2502 assert(role & BDRV_CHILD_COW);
70082db4
HR
2503
2504 /*
2505 * We want consistent read from backing files if the parent needs it.
2506 * No other operations are performed on backing files.
2507 */
2508 perm &= BLK_PERM_CONSISTENT_READ;
2509
2510 /*
2511 * If the parent can deal with changing data, we're okay with a
2512 * writable and resizable backing file.
2513 * TODO Require !(perm & BLK_PERM_CONSISTENT_READ), too?
2514 */
2515 if (shared & BLK_PERM_WRITE) {
2516 shared = BLK_PERM_WRITE | BLK_PERM_RESIZE;
2517 } else {
2518 shared = 0;
2519 }
2520
2521 shared |= BLK_PERM_CONSISTENT_READ | BLK_PERM_GRAPH_MOD |
2522 BLK_PERM_WRITE_UNCHANGED;
2523
2524 if (bs->open_flags & BDRV_O_INACTIVE) {
2525 shared |= BLK_PERM_WRITE | BLK_PERM_RESIZE;
2526 }
2527
2528 *nperm = perm;
2529 *nshared = shared;
2530}
2531
6f838a4b 2532static void bdrv_default_perms_for_storage(BlockDriverState *bs, BdrvChild *c,
6f838a4b
HR
2533 BdrvChildRole role,
2534 BlockReopenQueue *reopen_queue,
2535 uint64_t perm, uint64_t shared,
2536 uint64_t *nperm, uint64_t *nshared)
2537{
2538 int flags;
2539
e5d8a406 2540 assert(role & (BDRV_CHILD_METADATA | BDRV_CHILD_DATA));
6f838a4b
HR
2541
2542 flags = bdrv_reopen_get_flags(reopen_queue, bs);
2543
2544 /*
2545 * Apart from the modifications below, the same permissions are
2546 * forwarded and left alone as for filters
2547 */
e5d8a406 2548 bdrv_filter_default_perms(bs, c, role, reopen_queue,
6f838a4b
HR
2549 perm, shared, &perm, &shared);
2550
f889054f
HR
2551 if (role & BDRV_CHILD_METADATA) {
2552 /* Format drivers may touch metadata even if the guest doesn't write */
2553 if (bdrv_is_writable_after_reopen(bs, reopen_queue)) {
2554 perm |= BLK_PERM_WRITE | BLK_PERM_RESIZE;
2555 }
2556
2557 /*
2558 * bs->file always needs to be consistent because of the
2559 * metadata. We can never allow other users to resize or write
2560 * to it.
2561 */
2562 if (!(flags & BDRV_O_NO_IO)) {
2563 perm |= BLK_PERM_CONSISTENT_READ;
2564 }
2565 shared &= ~(BLK_PERM_WRITE | BLK_PERM_RESIZE);
6f838a4b
HR
2566 }
2567
f889054f
HR
2568 if (role & BDRV_CHILD_DATA) {
2569 /*
2570 * Technically, everything in this block is a subset of the
2571 * BDRV_CHILD_METADATA path taken above, and so this could
2572 * be an "else if" branch. However, that is not obvious, and
2573 * this function is not performance critical, therefore we let
2574 * this be an independent "if".
2575 */
2576
2577 /*
2578 * We cannot allow other users to resize the file because the
2579 * format driver might have some assumptions about the size
2580 * (e.g. because it is stored in metadata, or because the file
2581 * is split into fixed-size data files).
2582 */
2583 shared &= ~BLK_PERM_RESIZE;
2584
2585 /*
2586 * WRITE_UNCHANGED often cannot be performed as such on the
2587 * data file. For example, the qcow2 driver may still need to
2588 * write copied clusters on copy-on-read.
2589 */
2590 if (perm & BLK_PERM_WRITE_UNCHANGED) {
2591 perm |= BLK_PERM_WRITE;
2592 }
2593
2594 /*
2595 * If the data file is written to, the format driver may
2596 * expect to be able to resize it by writing beyond the EOF.
2597 */
2598 if (perm & BLK_PERM_WRITE) {
2599 perm |= BLK_PERM_RESIZE;
2600 }
6f838a4b 2601 }
6f838a4b
HR
2602
2603 if (bs->open_flags & BDRV_O_INACTIVE) {
2604 shared |= BLK_PERM_WRITE | BLK_PERM_RESIZE;
2605 }
2606
2607 *nperm = perm;
2608 *nshared = shared;
2609}
2610
2519f549 2611void bdrv_default_perms(BlockDriverState *bs, BdrvChild *c,
e5d8a406 2612 BdrvChildRole role, BlockReopenQueue *reopen_queue,
2519f549
HR
2613 uint64_t perm, uint64_t shared,
2614 uint64_t *nperm, uint64_t *nshared)
2615{
2519f549
HR
2616 if (role & BDRV_CHILD_FILTERED) {
2617 assert(!(role & (BDRV_CHILD_DATA | BDRV_CHILD_METADATA |
2618 BDRV_CHILD_COW)));
e5d8a406 2619 bdrv_filter_default_perms(bs, c, role, reopen_queue,
2519f549
HR
2620 perm, shared, nperm, nshared);
2621 } else if (role & BDRV_CHILD_COW) {
2622 assert(!(role & (BDRV_CHILD_DATA | BDRV_CHILD_METADATA)));
e5d8a406 2623 bdrv_default_perms_for_cow(bs, c, role, reopen_queue,
2519f549
HR
2624 perm, shared, nperm, nshared);
2625 } else if (role & (BDRV_CHILD_METADATA | BDRV_CHILD_DATA)) {
e5d8a406 2626 bdrv_default_perms_for_storage(bs, c, role, reopen_queue,
2519f549
HR
2627 perm, shared, nperm, nshared);
2628 } else {
2629 g_assert_not_reached();
2630 }
2631}
2632
7b1d9c4d
HR
2633uint64_t bdrv_qapi_perm_to_blk_perm(BlockPermission qapi_perm)
2634{
2635 static const uint64_t permissions[] = {
2636 [BLOCK_PERMISSION_CONSISTENT_READ] = BLK_PERM_CONSISTENT_READ,
2637 [BLOCK_PERMISSION_WRITE] = BLK_PERM_WRITE,
2638 [BLOCK_PERMISSION_WRITE_UNCHANGED] = BLK_PERM_WRITE_UNCHANGED,
2639 [BLOCK_PERMISSION_RESIZE] = BLK_PERM_RESIZE,
2640 [BLOCK_PERMISSION_GRAPH_MOD] = BLK_PERM_GRAPH_MOD,
2641 };
2642
2643 QEMU_BUILD_BUG_ON(ARRAY_SIZE(permissions) != BLOCK_PERMISSION__MAX);
2644 QEMU_BUILD_BUG_ON(1UL << ARRAY_SIZE(permissions) != BLK_PERM_ALL + 1);
2645
2646 assert(qapi_perm < BLOCK_PERMISSION__MAX);
2647
2648 return permissions[qapi_perm];
2649}
2650
8ee03995
KW
2651static void bdrv_replace_child_noperm(BdrvChild *child,
2652 BlockDriverState *new_bs)
e9740bc6
KW
2653{
2654 BlockDriverState *old_bs = child->bs;
debc2927
HR
2655 int new_bs_quiesce_counter;
2656 int drain_saldo;
e9740bc6 2657
2cad1ebe
AG
2658 assert(!child->frozen);
2659
bb2614e9
FZ
2660 if (old_bs && new_bs) {
2661 assert(bdrv_get_aio_context(old_bs) == bdrv_get_aio_context(new_bs));
2662 }
debc2927
HR
2663
2664 new_bs_quiesce_counter = (new_bs ? new_bs->quiesce_counter : 0);
2665 drain_saldo = new_bs_quiesce_counter - child->parent_quiesce_counter;
2666
2667 /*
2668 * If the new child node is drained but the old one was not, flush
2669 * all outstanding requests to the old child node.
2670 */
bd86fb99 2671 while (drain_saldo > 0 && child->klass->drained_begin) {
debc2927
HR
2672 bdrv_parent_drained_begin_single(child, true);
2673 drain_saldo--;
2674 }
2675
e9740bc6 2676 if (old_bs) {
d736f119
KW
2677 /* Detach first so that the recursive drain sections coming from @child
2678 * are already gone and we only end the drain sections that came from
2679 * elsewhere. */
bd86fb99
HR
2680 if (child->klass->detach) {
2681 child->klass->detach(child);
d736f119 2682 }
e9740bc6
KW
2683 QLIST_REMOVE(child, next_parent);
2684 }
36fe1331
KW
2685
2686 child->bs = new_bs;
2687
e9740bc6
KW
2688 if (new_bs) {
2689 QLIST_INSERT_HEAD(&new_bs->parents, child, next_parent);
debc2927
HR
2690
2691 /*
2692 * Detaching the old node may have led to the new node's
2693 * quiesce_counter having been decreased. Not a problem, we
2694 * just need to recognize this here and then invoke
2695 * drained_end appropriately more often.
2696 */
2697 assert(new_bs->quiesce_counter <= new_bs_quiesce_counter);
2698 drain_saldo += new_bs->quiesce_counter - new_bs_quiesce_counter;
33a610c3 2699
d736f119
KW
2700 /* Attach only after starting new drained sections, so that recursive
2701 * drain sections coming from @child don't get an extra .drained_begin
2702 * callback. */
bd86fb99
HR
2703 if (child->klass->attach) {
2704 child->klass->attach(child);
8ee03995
KW
2705 }
2706 }
debc2927
HR
2707
2708 /*
2709 * If the old child node was drained but the new one is not, allow
2710 * requests to come in only after the new node has been attached.
2711 */
bd86fb99 2712 while (drain_saldo < 0 && child->klass->drained_end) {
debc2927
HR
2713 bdrv_parent_drained_end_single(child);
2714 drain_saldo++;
2715 }
8ee03995 2716}
33a610c3 2717
46541ee5
VSO
2718static void bdrv_child_free(void *opaque)
2719{
2720 BdrvChild *c = opaque;
2721
2722 g_free(c->name);
2723 g_free(c);
2724}
2725
548a74c0 2726static void bdrv_remove_empty_child(BdrvChild *child)
df581792 2727{
548a74c0
VSO
2728 assert(!child->bs);
2729 QLIST_SAFE_REMOVE(child, next);
46541ee5 2730 bdrv_child_free(child);
548a74c0 2731}
d5e6f437 2732
548a74c0
VSO
2733typedef struct BdrvAttachChildCommonState {
2734 BdrvChild **child;
2735 AioContext *old_parent_ctx;
2736 AioContext *old_child_ctx;
2737} BdrvAttachChildCommonState;
2738
2739static void bdrv_attach_child_common_abort(void *opaque)
2740{
2741 BdrvAttachChildCommonState *s = opaque;
2742 BdrvChild *child = *s->child;
2743 BlockDriverState *bs = child->bs;
2744
2745 bdrv_replace_child_noperm(child, NULL);
2746
2747 if (bdrv_get_aio_context(bs) != s->old_child_ctx) {
2748 bdrv_try_set_aio_context(bs, s->old_child_ctx, &error_abort);
2749 }
2750
2751 if (bdrv_child_get_parent_aio_context(child) != s->old_parent_ctx) {
2752 GSList *ignore = g_slist_prepend(NULL, child);
2753
2754 child->klass->can_set_aio_ctx(child, s->old_parent_ctx, &ignore,
2755 &error_abort);
2756 g_slist_free(ignore);
2757 ignore = g_slist_prepend(NULL, child);
2758 child->klass->set_aio_ctx(child, s->old_parent_ctx, &ignore);
2759
2760 g_slist_free(ignore);
d5e6f437
KW
2761 }
2762
548a74c0
VSO
2763 bdrv_unref(bs);
2764 bdrv_remove_empty_child(child);
2765 *s->child = NULL;
2766}
2767
2768static TransactionActionDrv bdrv_attach_child_common_drv = {
2769 .abort = bdrv_attach_child_common_abort,
2770 .clean = g_free,
2771};
2772
2773/*
2774 * Common part of attaching bdrv child to bs or to blk or to job
f8d2ad78
VSO
2775 *
2776 * Resulting new child is returned through @child.
2777 * At start *@child must be NULL.
2778 * @child is saved to a new entry of @tran, so that *@child could be reverted to
2779 * NULL on abort(). So referenced variable must live at least until transaction
2780 * end.
7ec390d5
VSO
2781 *
2782 * Function doesn't update permissions, caller is responsible for this.
548a74c0
VSO
2783 */
2784static int bdrv_attach_child_common(BlockDriverState *child_bs,
2785 const char *child_name,
2786 const BdrvChildClass *child_class,
2787 BdrvChildRole child_role,
2788 uint64_t perm, uint64_t shared_perm,
2789 void *opaque, BdrvChild **child,
2790 Transaction *tran, Error **errp)
2791{
2792 BdrvChild *new_child;
2793 AioContext *parent_ctx;
2794 AioContext *child_ctx = bdrv_get_aio_context(child_bs);
2795
2796 assert(child);
2797 assert(*child == NULL);
da261b69 2798 assert(child_class->get_parent_desc);
548a74c0
VSO
2799
2800 new_child = g_new(BdrvChild, 1);
2801 *new_child = (BdrvChild) {
d5e6f437
KW
2802 .bs = NULL,
2803 .name = g_strdup(child_name),
bd86fb99 2804 .klass = child_class,
258b7765 2805 .role = child_role,
d5e6f437
KW
2806 .perm = perm,
2807 .shared_perm = shared_perm,
2808 .opaque = opaque,
df581792
KW
2809 };
2810
548a74c0
VSO
2811 /*
2812 * If the AioContexts don't match, first try to move the subtree of
132ada80 2813 * child_bs into the AioContext of the new parent. If this doesn't work,
548a74c0
VSO
2814 * try moving the parent into the AioContext of child_bs instead.
2815 */
2816 parent_ctx = bdrv_child_get_parent_aio_context(new_child);
2817 if (child_ctx != parent_ctx) {
2818 Error *local_err = NULL;
2819 int ret = bdrv_try_set_aio_context(child_bs, parent_ctx, &local_err);
2820
bd86fb99 2821 if (ret < 0 && child_class->can_set_aio_ctx) {
548a74c0
VSO
2822 GSList *ignore = g_slist_prepend(NULL, new_child);
2823 if (child_class->can_set_aio_ctx(new_child, child_ctx, &ignore,
2824 NULL))
2825 {
132ada80
KW
2826 error_free(local_err);
2827 ret = 0;
2828 g_slist_free(ignore);
548a74c0
VSO
2829 ignore = g_slist_prepend(NULL, new_child);
2830 child_class->set_aio_ctx(new_child, child_ctx, &ignore);
132ada80
KW
2831 }
2832 g_slist_free(ignore);
2833 }
548a74c0 2834
132ada80
KW
2835 if (ret < 0) {
2836 error_propagate(errp, local_err);
548a74c0
VSO
2837 bdrv_remove_empty_child(new_child);
2838 return ret;
132ada80
KW
2839 }
2840 }
2841
548a74c0
VSO
2842 bdrv_ref(child_bs);
2843 bdrv_replace_child_noperm(new_child, child_bs);
2844
2845 *child = new_child;
b4b059f6 2846
548a74c0
VSO
2847 BdrvAttachChildCommonState *s = g_new(BdrvAttachChildCommonState, 1);
2848 *s = (BdrvAttachChildCommonState) {
2849 .child = child,
2850 .old_parent_ctx = parent_ctx,
2851 .old_child_ctx = child_ctx,
2852 };
2853 tran_add(tran, &bdrv_attach_child_common_drv, s);
2854
2855 return 0;
2856}
2857
f8d2ad78
VSO
2858/*
2859 * Variable referenced by @child must live at least until transaction end.
2860 * (see bdrv_attach_child_common() doc for details)
7ec390d5
VSO
2861 *
2862 * Function doesn't update permissions, caller is responsible for this.
f8d2ad78 2863 */
aa5a04c7
VSO
2864static int bdrv_attach_child_noperm(BlockDriverState *parent_bs,
2865 BlockDriverState *child_bs,
2866 const char *child_name,
2867 const BdrvChildClass *child_class,
2868 BdrvChildRole child_role,
2869 BdrvChild **child,
2870 Transaction *tran,
2871 Error **errp)
2872{
2873 int ret;
2874 uint64_t perm, shared_perm;
2875
2876 assert(parent_bs->drv);
2877
2878 bdrv_get_cumulative_perm(parent_bs, &perm, &shared_perm);
2879 bdrv_child_perm(parent_bs, child_bs, NULL, child_role, NULL,
2880 perm, shared_perm, &perm, &shared_perm);
2881
2882 ret = bdrv_attach_child_common(child_bs, child_name, child_class,
2883 child_role, perm, shared_perm, parent_bs,
2884 child, tran, errp);
2885 if (ret < 0) {
2886 return ret;
2887 }
2888
2889 QLIST_INSERT_HEAD(&parent_bs->children, *child, next);
2890 /*
2891 * child is removed in bdrv_attach_child_common_abort(), so don't care to
2892 * abort this change separately.
2893 */
2894
2895 return 0;
2896}
2897
548a74c0
VSO
2898static void bdrv_detach_child(BdrvChild *child)
2899{
4954aace
VSO
2900 BlockDriverState *old_bs = child->bs;
2901
2902 bdrv_replace_child_noperm(child, NULL);
548a74c0 2903 bdrv_remove_empty_child(child);
4954aace
VSO
2904
2905 if (old_bs) {
2906 /*
2907 * Update permissions for old node. We're just taking a parent away, so
2908 * we're loosening restrictions. Errors of permission update are not
2909 * fatal in this case, ignore them.
2910 */
2911 bdrv_refresh_perms(old_bs, NULL);
2912
2913 /*
2914 * When the parent requiring a non-default AioContext is removed, the
2915 * node moves back to the main AioContext
2916 */
2917 bdrv_try_set_aio_context(old_bs, qemu_get_aio_context(), NULL);
2918 }
548a74c0
VSO
2919}
2920
2921/*
2922 * This function steals the reference to child_bs from the caller.
2923 * That reference is later dropped by bdrv_root_unref_child().
2924 *
2925 * On failure NULL is returned, errp is set and the reference to
2926 * child_bs is also dropped.
2927 *
2928 * The caller must hold the AioContext lock @child_bs, but not that of @ctx
2929 * (unless @child_bs is already in @ctx).
2930 */
2931BdrvChild *bdrv_root_attach_child(BlockDriverState *child_bs,
2932 const char *child_name,
2933 const BdrvChildClass *child_class,
2934 BdrvChildRole child_role,
2935 uint64_t perm, uint64_t shared_perm,
2936 void *opaque, Error **errp)
2937{
2938 int ret;
2939 BdrvChild *child = NULL;
2940 Transaction *tran = tran_new();
2941
2942 ret = bdrv_attach_child_common(child_bs, child_name, child_class,
2943 child_role, perm, shared_perm, opaque,
2944 &child, tran, errp);
2945 if (ret < 0) {
e878bb12 2946 goto out;
548a74c0
VSO
2947 }
2948
2949 ret = bdrv_refresh_perms(child_bs, errp);
548a74c0 2950
e878bb12
KW
2951out:
2952 tran_finalize(tran, ret);
f8d2ad78
VSO
2953 /* child is unset on failure by bdrv_attach_child_common_abort() */
2954 assert((ret < 0) == !child);
2955
548a74c0 2956 bdrv_unref(child_bs);
b4b059f6 2957 return child;
df581792
KW
2958}
2959
b441dc71
AG
2960/*
2961 * This function transfers the reference to child_bs from the caller
2962 * to parent_bs. That reference is later dropped by parent_bs on
2963 * bdrv_close() or if someone calls bdrv_unref_child().
2964 *
2965 * On failure NULL is returned, errp is set and the reference to
2966 * child_bs is also dropped.
132ada80
KW
2967 *
2968 * If @parent_bs and @child_bs are in different AioContexts, the caller must
2969 * hold the AioContext lock for @child_bs, but not for @parent_bs.
b441dc71 2970 */
98292c61
WC
2971BdrvChild *bdrv_attach_child(BlockDriverState *parent_bs,
2972 BlockDriverState *child_bs,
2973 const char *child_name,
bd86fb99 2974 const BdrvChildClass *child_class,
258b7765 2975 BdrvChildRole child_role,
8b2ff529 2976 Error **errp)
f21d96d0 2977{
aa5a04c7
VSO
2978 int ret;
2979 BdrvChild *child = NULL;
2980 Transaction *tran = tran_new();
f68c598b 2981
aa5a04c7
VSO
2982 ret = bdrv_attach_child_noperm(parent_bs, child_bs, child_name, child_class,
2983 child_role, &child, tran, errp);
2984 if (ret < 0) {
2985 goto out;
2986 }
d5e6f437 2987
aa5a04c7
VSO
2988 ret = bdrv_refresh_perms(parent_bs, errp);
2989 if (ret < 0) {
2990 goto out;
d5e6f437
KW
2991 }
2992
aa5a04c7
VSO
2993out:
2994 tran_finalize(tran, ret);
f8d2ad78
VSO
2995 /* child is unset on failure by bdrv_attach_child_common_abort() */
2996 assert((ret < 0) == !child);
aa5a04c7
VSO
2997
2998 bdrv_unref(child_bs);
2999
f21d96d0
KW
3000 return child;
3001}
3002
7b99a266 3003/* Callers must ensure that child->frozen is false. */
f21d96d0 3004void bdrv_root_unref_child(BdrvChild *child)
33a60407 3005{
779020cb
KW
3006 BlockDriverState *child_bs;
3007
f21d96d0
KW
3008 child_bs = child->bs;
3009 bdrv_detach_child(child);
3010 bdrv_unref(child_bs);
3011}
3012
332b3a17
VSO
3013typedef struct BdrvSetInheritsFrom {
3014 BlockDriverState *bs;
3015 BlockDriverState *old_inherits_from;
3016} BdrvSetInheritsFrom;
3017
3018static void bdrv_set_inherits_from_abort(void *opaque)
3019{
3020 BdrvSetInheritsFrom *s = opaque;
3021
3022 s->bs->inherits_from = s->old_inherits_from;
3023}
3024
3025static TransactionActionDrv bdrv_set_inherits_from_drv = {
3026 .abort = bdrv_set_inherits_from_abort,
3027 .clean = g_free,
3028};
3029
3030/* @tran is allowed to be NULL. In this case no rollback is possible */
3031static void bdrv_set_inherits_from(BlockDriverState *bs,
3032 BlockDriverState *new_inherits_from,
3033 Transaction *tran)
3034{
3035 if (tran) {
3036 BdrvSetInheritsFrom *s = g_new(BdrvSetInheritsFrom, 1);
3037
3038 *s = (BdrvSetInheritsFrom) {
3039 .bs = bs,
3040 .old_inherits_from = bs->inherits_from,
3041 };
3042
3043 tran_add(tran, &bdrv_set_inherits_from_drv, s);
3044 }
3045
3046 bs->inherits_from = new_inherits_from;
3047}
3048
3cf746b3
HR
3049/**
3050 * Clear all inherits_from pointers from children and grandchildren of
3051 * @root that point to @root, where necessary.
332b3a17 3052 * @tran is allowed to be NULL. In this case no rollback is possible
3cf746b3 3053 */
332b3a17
VSO
3054static void bdrv_unset_inherits_from(BlockDriverState *root, BdrvChild *child,
3055 Transaction *tran)
f21d96d0 3056{
3cf746b3 3057 BdrvChild *c;
4e4bf5c4 3058
3cf746b3
HR
3059 if (child->bs->inherits_from == root) {
3060 /*
3061 * Remove inherits_from only when the last reference between root and
3062 * child->bs goes away.
3063 */
3064 QLIST_FOREACH(c, &root->children, next) {
4e4bf5c4
KW
3065 if (c != child && c->bs == child->bs) {
3066 break;
3067 }
3068 }
3069 if (c == NULL) {
332b3a17 3070 bdrv_set_inherits_from(child->bs, NULL, tran);
4e4bf5c4 3071 }
33a60407
KW
3072 }
3073
3cf746b3 3074 QLIST_FOREACH(c, &child->bs->children, next) {
332b3a17 3075 bdrv_unset_inherits_from(root, c, tran);
3cf746b3
HR
3076 }
3077}
3078
7b99a266 3079/* Callers must ensure that child->frozen is false. */
3cf746b3
HR
3080void bdrv_unref_child(BlockDriverState *parent, BdrvChild *child)
3081{
3082 if (child == NULL) {
3083 return;
3084 }
3085
332b3a17 3086 bdrv_unset_inherits_from(parent, child, NULL);
f21d96d0 3087 bdrv_root_unref_child(child);
33a60407
KW
3088}
3089
5c8cab48
KW
3090
3091static void bdrv_parent_cb_change_media(BlockDriverState *bs, bool load)
3092{
3093 BdrvChild *c;
3094 QLIST_FOREACH(c, &bs->parents, next_parent) {
bd86fb99
HR
3095 if (c->klass->change_media) {
3096 c->klass->change_media(c, load);
5c8cab48
KW
3097 }
3098 }
3099}
3100
0065c455
AG
3101/* Return true if you can reach parent going through child->inherits_from
3102 * recursively. If parent or child are NULL, return false */
3103static bool bdrv_inherits_from_recursive(BlockDriverState *child,
3104 BlockDriverState *parent)
3105{
3106 while (child && child != parent) {
3107 child = child->inherits_from;
3108 }
3109
3110 return child != NULL;
3111}
3112
25191e5f
HR
3113/*
3114 * Return the BdrvChildRole for @bs's backing child. bs->backing is
3115 * mostly used for COW backing children (role = COW), but also for
3116 * filtered children (role = FILTERED | PRIMARY).
3117 */
3118static BdrvChildRole bdrv_backing_role(BlockDriverState *bs)
3119{
3120 if (bs->drv && bs->drv->is_filter) {
3121 return BDRV_CHILD_FILTERED | BDRV_CHILD_PRIMARY;
3122 } else {
3123 return BDRV_CHILD_COW;
3124 }
3125}
3126
5db15a57 3127/*
e9238278
VSO
3128 * Sets the bs->backing or bs->file link of a BDS. A new reference is created;
3129 * callers which don't need their own reference any more must call bdrv_unref().
7ec390d5
VSO
3130 *
3131 * Function doesn't update permissions, caller is responsible for this.
5db15a57 3132 */
e9238278
VSO
3133static int bdrv_set_file_or_backing_noperm(BlockDriverState *parent_bs,
3134 BlockDriverState *child_bs,
3135 bool is_backing,
3136 Transaction *tran, Error **errp)
8d24cce1 3137{
a1e708fc 3138 int ret = 0;
e9238278
VSO
3139 bool update_inherits_from =
3140 bdrv_inherits_from_recursive(child_bs, parent_bs);
3141 BdrvChild *child = is_backing ? parent_bs->backing : parent_bs->file;
3142 BdrvChildRole role;
0065c455 3143
e9238278
VSO
3144 if (!parent_bs->drv) {
3145 /*
3146 * Node without drv is an object without a class :/. TODO: finally fix
3147 * qcow2 driver to never clear bs->drv and implement format corruption
3148 * handling in other way.
3149 */
3150 error_setg(errp, "Node corrupted");
3151 return -EINVAL;
3152 }
3153
3154 if (child && child->frozen) {
3155 error_setg(errp, "Cannot change frozen '%s' link from '%s' to '%s'",
3156 child->name, parent_bs->node_name, child->bs->node_name);
a1e708fc 3157 return -EPERM;
2cad1ebe
AG
3158 }
3159
25f78d9e
VSO
3160 if (is_backing && !parent_bs->drv->is_filter &&
3161 !parent_bs->drv->supports_backing)
3162 {
3163 error_setg(errp, "Driver '%s' of node '%s' does not support backing "
3164 "files", parent_bs->drv->format_name, parent_bs->node_name);
3165 return -EINVAL;
3166 }
3167
e9238278
VSO
3168 if (parent_bs->drv->is_filter) {
3169 role = BDRV_CHILD_FILTERED | BDRV_CHILD_PRIMARY;
3170 } else if (is_backing) {
3171 role = BDRV_CHILD_COW;
3172 } else {
3173 /*
3174 * We only can use same role as it is in existing child. We don't have
3175 * infrastructure to determine role of file child in generic way
3176 */
3177 if (!child) {
3178 error_setg(errp, "Cannot set file child to format node without "
3179 "file child");
3180 return -EINVAL;
3181 }
3182 role = child->role;
826b6ca0
FZ
3183 }
3184
e9238278
VSO
3185 if (child) {
3186 bdrv_unset_inherits_from(parent_bs, child, tran);
3187 bdrv_remove_file_or_backing_child(parent_bs, child, tran);
3188 }
3189
3190 if (!child_bs) {
8d24cce1
FZ
3191 goto out;
3192 }
12fa4af6 3193
e9238278
VSO
3194 ret = bdrv_attach_child_noperm(parent_bs, child_bs,
3195 is_backing ? "backing" : "file",
3196 &child_of_bds, role,
3197 is_backing ? &parent_bs->backing :
3198 &parent_bs->file,
3199 tran, errp);
160333e1
VSO
3200 if (ret < 0) {
3201 return ret;
a1e708fc
VSO
3202 }
3203
160333e1
VSO
3204
3205 /*
e9238278 3206 * If inherits_from pointed recursively to bs then let's update it to
160333e1
VSO
3207 * point directly to bs (else it will become NULL).
3208 */
a1e708fc 3209 if (update_inherits_from) {
e9238278 3210 bdrv_set_inherits_from(child_bs, parent_bs, tran);
0065c455 3211 }
826b6ca0 3212
8d24cce1 3213out:
e9238278 3214 bdrv_refresh_limits(parent_bs, tran, NULL);
160333e1
VSO
3215
3216 return 0;
3217}
3218
e9238278
VSO
3219static int bdrv_set_backing_noperm(BlockDriverState *bs,
3220 BlockDriverState *backing_hd,
3221 Transaction *tran, Error **errp)
3222{
3223 return bdrv_set_file_or_backing_noperm(bs, backing_hd, true, tran, errp);
3224}
3225
160333e1
VSO
3226int bdrv_set_backing_hd(BlockDriverState *bs, BlockDriverState *backing_hd,
3227 Error **errp)
3228{
3229 int ret;
3230 Transaction *tran = tran_new();
3231
3232 ret = bdrv_set_backing_noperm(bs, backing_hd, tran, errp);
3233 if (ret < 0) {
3234 goto out;
3235 }
3236
3237 ret = bdrv_refresh_perms(bs, errp);
3238out:
3239 tran_finalize(tran, ret);
a1e708fc
VSO
3240
3241 return ret;
8d24cce1
FZ
3242}
3243
31ca6d07
KW
3244/*
3245 * Opens the backing file for a BlockDriverState if not yet open
3246 *
d9b7b057
KW
3247 * bdref_key specifies the key for the image's BlockdevRef in the options QDict.
3248 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
3249 * itself, all options starting with "${bdref_key}." are considered part of the
3250 * BlockdevRef.
3251 *
3252 * TODO Can this be unified with bdrv_open_image()?
31ca6d07 3253 */
d9b7b057
KW
3254int bdrv_open_backing_file(BlockDriverState *bs, QDict *parent_options,
3255 const char *bdref_key, Error **errp)
9156df12 3256{
6b6833c1 3257 char *backing_filename = NULL;
d9b7b057
KW
3258 char *bdref_key_dot;
3259 const char *reference = NULL;
317fc44e 3260 int ret = 0;
998c2019 3261 bool implicit_backing = false;
8d24cce1 3262 BlockDriverState *backing_hd;
d9b7b057
KW
3263 QDict *options;
3264 QDict *tmp_parent_options = NULL;
34b5d2c6 3265 Error *local_err = NULL;
9156df12 3266
760e0063 3267 if (bs->backing != NULL) {
1ba4b6a5 3268 goto free_exit;
9156df12
PB
3269 }
3270
31ca6d07 3271 /* NULL means an empty set of options */
d9b7b057
KW
3272 if (parent_options == NULL) {
3273 tmp_parent_options = qdict_new();
3274 parent_options = tmp_parent_options;
31ca6d07
KW
3275 }
3276
9156df12 3277 bs->open_flags &= ~BDRV_O_NO_BACKING;
d9b7b057
KW
3278
3279 bdref_key_dot = g_strdup_printf("%s.", bdref_key);
3280 qdict_extract_subqdict(parent_options, &options, bdref_key_dot);
3281 g_free(bdref_key_dot);
3282
129c7d1c
MA
3283 /*
3284 * Caution: while qdict_get_try_str() is fine, getting non-string
3285 * types would require more care. When @parent_options come from
3286 * -blockdev or blockdev_add, its members are typed according to
3287 * the QAPI schema, but when they come from -drive, they're all
3288 * QString.
3289 */
d9b7b057
KW
3290 reference = qdict_get_try_str(parent_options, bdref_key);
3291 if (reference || qdict_haskey(options, "file.filename")) {
6b6833c1 3292 /* keep backing_filename NULL */
1cb6f506 3293 } else if (bs->backing_file[0] == '\0' && qdict_size(options) == 0) {
cb3e7f08 3294 qobject_unref(options);
1ba4b6a5 3295 goto free_exit;
dbecebdd 3296 } else {
998c2019
HR
3297 if (qdict_size(options) == 0) {
3298 /* If the user specifies options that do not modify the
3299 * backing file's behavior, we might still consider it the
3300 * implicit backing file. But it's easier this way, and
3301 * just specifying some of the backing BDS's options is
3302 * only possible with -drive anyway (otherwise the QAPI
3303 * schema forces the user to specify everything). */
3304 implicit_backing = !strcmp(bs->auto_backing_file, bs->backing_file);
3305 }
3306
6b6833c1 3307 backing_filename = bdrv_get_full_backing_filename(bs, &local_err);
9f07429e
HR
3308 if (local_err) {
3309 ret = -EINVAL;
3310 error_propagate(errp, local_err);
cb3e7f08 3311 qobject_unref(options);
9f07429e
HR
3312 goto free_exit;
3313 }
9156df12
PB
3314 }
3315
8ee79e70
KW
3316 if (!bs->drv || !bs->drv->supports_backing) {
3317 ret = -EINVAL;
3318 error_setg(errp, "Driver doesn't support backing files");
cb3e7f08 3319 qobject_unref(options);
8ee79e70
KW
3320 goto free_exit;
3321 }
3322
6bff597b
PK
3323 if (!reference &&
3324 bs->backing_format[0] != '\0' && !qdict_haskey(options, "driver")) {
46f5ac20 3325 qdict_put_str(options, "driver", bs->backing_format);
9156df12
PB
3326 }
3327
6b6833c1 3328 backing_hd = bdrv_open_inherit(backing_filename, reference, options, 0, bs,
25191e5f 3329 &child_of_bds, bdrv_backing_role(bs), errp);
5b363937 3330 if (!backing_hd) {
9156df12 3331 bs->open_flags |= BDRV_O_NO_BACKING;
e43bfd9c 3332 error_prepend(errp, "Could not open backing file: ");
5b363937 3333 ret = -EINVAL;
1ba4b6a5 3334 goto free_exit;
9156df12 3335 }
df581792 3336
998c2019
HR
3337 if (implicit_backing) {
3338 bdrv_refresh_filename(backing_hd);
3339 pstrcpy(bs->auto_backing_file, sizeof(bs->auto_backing_file),
3340 backing_hd->filename);
3341 }
3342
5db15a57
KW
3343 /* Hook up the backing file link; drop our reference, bs owns the
3344 * backing_hd reference now */
dc9c10a1 3345 ret = bdrv_set_backing_hd(bs, backing_hd, errp);
5db15a57 3346 bdrv_unref(backing_hd);
dc9c10a1 3347 if (ret < 0) {
12fa4af6
KW
3348 goto free_exit;
3349 }
d80ac658 3350
d9b7b057
KW
3351 qdict_del(parent_options, bdref_key);
3352
1ba4b6a5
BC
3353free_exit:
3354 g_free(backing_filename);
cb3e7f08 3355 qobject_unref(tmp_parent_options);
1ba4b6a5 3356 return ret;
9156df12
PB
3357}
3358
2d6b86af
KW
3359static BlockDriverState *
3360bdrv_open_child_bs(const char *filename, QDict *options, const char *bdref_key,
bd86fb99 3361 BlockDriverState *parent, const BdrvChildClass *child_class,
272c02ea 3362 BdrvChildRole child_role, bool allow_none, Error **errp)
da557aac 3363{
2d6b86af 3364 BlockDriverState *bs = NULL;
da557aac 3365 QDict *image_options;
da557aac
HR
3366 char *bdref_key_dot;
3367 const char *reference;
3368
bd86fb99 3369 assert(child_class != NULL);
f67503e5 3370
da557aac
HR
3371 bdref_key_dot = g_strdup_printf("%s.", bdref_key);
3372 qdict_extract_subqdict(options, &image_options, bdref_key_dot);
3373 g_free(bdref_key_dot);
3374
129c7d1c
MA
3375 /*
3376 * Caution: while qdict_get_try_str() is fine, getting non-string
3377 * types would require more care. When @options come from
3378 * -blockdev or blockdev_add, its members are typed according to
3379 * the QAPI schema, but when they come from -drive, they're all
3380 * QString.
3381 */
da557aac
HR
3382 reference = qdict_get_try_str(options, bdref_key);
3383 if (!filename && !reference && !qdict_size(image_options)) {
b4b059f6 3384 if (!allow_none) {
da557aac
HR
3385 error_setg(errp, "A block device must be specified for \"%s\"",
3386 bdref_key);
da557aac 3387 }
cb3e7f08 3388 qobject_unref(image_options);
da557aac
HR
3389 goto done;
3390 }
3391
5b363937 3392 bs = bdrv_open_inherit(filename, reference, image_options, 0,
272c02ea 3393 parent, child_class, child_role, errp);
5b363937 3394 if (!bs) {
df581792
KW
3395 goto done;
3396 }
3397
da557aac
HR
3398done:
3399 qdict_del(options, bdref_key);
2d6b86af
KW
3400 return bs;
3401}
3402
3403/*
3404 * Opens a disk image whose options are given as BlockdevRef in another block
3405 * device's options.
3406 *
3407 * If allow_none is true, no image will be opened if filename is false and no
3408 * BlockdevRef is given. NULL will be returned, but errp remains unset.
3409 *
3410 * bdrev_key specifies the key for the image's BlockdevRef in the options QDict.
3411 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
3412 * itself, all options starting with "${bdref_key}." are considered part of the
3413 * BlockdevRef.
3414 *
3415 * The BlockdevRef will be removed from the options QDict.
3416 */
3417BdrvChild *bdrv_open_child(const char *filename,
3418 QDict *options, const char *bdref_key,
3419 BlockDriverState *parent,
bd86fb99 3420 const BdrvChildClass *child_class,
258b7765 3421 BdrvChildRole child_role,
2d6b86af
KW
3422 bool allow_none, Error **errp)
3423{
3424 BlockDriverState *bs;
3425
bd86fb99 3426 bs = bdrv_open_child_bs(filename, options, bdref_key, parent, child_class,
272c02ea 3427 child_role, allow_none, errp);
2d6b86af
KW
3428 if (bs == NULL) {
3429 return NULL;
3430 }
3431
258b7765
HR
3432 return bdrv_attach_child(parent, bs, bdref_key, child_class, child_role,
3433 errp);
b4b059f6
KW
3434}
3435
bd86fb99
HR
3436/*
3437 * TODO Future callers may need to specify parent/child_class in order for
3438 * option inheritance to work. Existing callers use it for the root node.
3439 */
e1d74bc6
KW
3440BlockDriverState *bdrv_open_blockdev_ref(BlockdevRef *ref, Error **errp)
3441{
3442 BlockDriverState *bs = NULL;
e1d74bc6
KW
3443 QObject *obj = NULL;
3444 QDict *qdict = NULL;
3445 const char *reference = NULL;
3446 Visitor *v = NULL;
3447
3448 if (ref->type == QTYPE_QSTRING) {
3449 reference = ref->u.reference;
3450 } else {
3451 BlockdevOptions *options = &ref->u.definition;
3452 assert(ref->type == QTYPE_QDICT);
3453
3454 v = qobject_output_visitor_new(&obj);
1f584248 3455 visit_type_BlockdevOptions(v, NULL, &options, &error_abort);
e1d74bc6
KW
3456 visit_complete(v, &obj);
3457
7dc847eb 3458 qdict = qobject_to(QDict, obj);
e1d74bc6
KW
3459 qdict_flatten(qdict);
3460
3461 /* bdrv_open_inherit() defaults to the values in bdrv_flags (for
3462 * compatibility with other callers) rather than what we want as the
3463 * real defaults. Apply the defaults here instead. */
3464 qdict_set_default_str(qdict, BDRV_OPT_CACHE_DIRECT, "off");
3465 qdict_set_default_str(qdict, BDRV_OPT_CACHE_NO_FLUSH, "off");
3466 qdict_set_default_str(qdict, BDRV_OPT_READ_ONLY, "off");
e35bdc12
KW
3467 qdict_set_default_str(qdict, BDRV_OPT_AUTO_READ_ONLY, "off");
3468
e1d74bc6
KW
3469 }
3470
272c02ea 3471 bs = bdrv_open_inherit(NULL, reference, qdict, 0, NULL, NULL, 0, errp);
e1d74bc6 3472 obj = NULL;
cb3e7f08 3473 qobject_unref(obj);
e1d74bc6
KW
3474 visit_free(v);
3475 return bs;
3476}
3477
66836189
HR
3478static BlockDriverState *bdrv_append_temp_snapshot(BlockDriverState *bs,
3479 int flags,
3480 QDict *snapshot_options,
3481 Error **errp)
b998875d
KW
3482{
3483 /* TODO: extra byte is a hack to ensure MAX_PATH space on Windows. */
1ba4b6a5 3484 char *tmp_filename = g_malloc0(PATH_MAX + 1);
b998875d 3485 int64_t total_size;
83d0521a 3486 QemuOpts *opts = NULL;
ff6ed714 3487 BlockDriverState *bs_snapshot = NULL;
b998875d
KW
3488 int ret;
3489
3490 /* if snapshot, we create a temporary backing file and open it
3491 instead of opening 'filename' directly */
3492
3493 /* Get the required size from the image */
f187743a
KW
3494 total_size = bdrv_getlength(bs);
3495 if (total_size < 0) {
3496 error_setg_errno(errp, -total_size, "Could not get image size");
1ba4b6a5 3497 goto out;
f187743a 3498 }
b998875d
KW
3499
3500 /* Create the temporary image */
1ba4b6a5 3501 ret = get_tmp_filename(tmp_filename, PATH_MAX + 1);
b998875d
KW
3502 if (ret < 0) {
3503 error_setg_errno(errp, -ret, "Could not get temporary filename");
1ba4b6a5 3504 goto out;
b998875d
KW
3505 }
3506
ef810437 3507 opts = qemu_opts_create(bdrv_qcow2.create_opts, NULL, 0,
c282e1fd 3508 &error_abort);
39101f25 3509 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, total_size, &error_abort);
e43bfd9c 3510 ret = bdrv_create(&bdrv_qcow2, tmp_filename, opts, errp);
83d0521a 3511 qemu_opts_del(opts);
b998875d 3512 if (ret < 0) {
e43bfd9c
MA
3513 error_prepend(errp, "Could not create temporary overlay '%s': ",
3514 tmp_filename);
1ba4b6a5 3515 goto out;
b998875d
KW
3516 }
3517
73176bee 3518 /* Prepare options QDict for the temporary file */
46f5ac20
EB
3519 qdict_put_str(snapshot_options, "file.driver", "file");
3520 qdict_put_str(snapshot_options, "file.filename", tmp_filename);
3521 qdict_put_str(snapshot_options, "driver", "qcow2");
b998875d 3522
5b363937 3523 bs_snapshot = bdrv_open(NULL, NULL, snapshot_options, flags, errp);
73176bee 3524 snapshot_options = NULL;
5b363937 3525 if (!bs_snapshot) {
1ba4b6a5 3526 goto out;
b998875d
KW
3527 }
3528
934aee14
VSO
3529 ret = bdrv_append(bs_snapshot, bs, errp);
3530 if (ret < 0) {
ff6ed714 3531 bs_snapshot = NULL;
b2c2832c
KW
3532 goto out;
3533 }
1ba4b6a5
BC
3534
3535out:
cb3e7f08 3536 qobject_unref(snapshot_options);
1ba4b6a5 3537 g_free(tmp_filename);
ff6ed714 3538 return bs_snapshot;
b998875d
KW
3539}
3540
b6ce07aa
KW
3541/*
3542 * Opens a disk image (raw, qcow2, vmdk, ...)
de9c0cec
KW
3543 *
3544 * options is a QDict of options to pass to the block drivers, or NULL for an
3545 * empty set of options. The reference to the QDict belongs to the block layer
3546 * after the call (even on failure), so if the caller intends to reuse the
cb3e7f08 3547 * dictionary, it needs to use qobject_ref() before calling bdrv_open.
f67503e5
HR
3548 *
3549 * If *pbs is NULL, a new BDS will be created with a pointer to it stored there.
3550 * If it is not NULL, the referenced BDS will be reused.
ddf5636d
HR
3551 *
3552 * The reference parameter may be used to specify an existing block device which
3553 * should be opened. If specified, neither options nor a filename may be given,
3554 * nor can an existing BDS be reused (that is, *pbs has to be NULL).
b6ce07aa 3555 */
5b363937
HR
3556static BlockDriverState *bdrv_open_inherit(const char *filename,
3557 const char *reference,
3558 QDict *options, int flags,
3559 BlockDriverState *parent,
bd86fb99 3560 const BdrvChildClass *child_class,
272c02ea 3561 BdrvChildRole child_role,
5b363937 3562 Error **errp)
ea2384d3 3563{
b6ce07aa 3564 int ret;
5696c6e3 3565 BlockBackend *file = NULL;
9a4f4c31 3566 BlockDriverState *bs;
ce343771 3567 BlockDriver *drv = NULL;
2f624b80 3568 BdrvChild *child;
74fe54f2 3569 const char *drvname;
3e8c2e57 3570 const char *backing;
34b5d2c6 3571 Error *local_err = NULL;
73176bee 3572 QDict *snapshot_options = NULL;
b1e6fc08 3573 int snapshot_flags = 0;
712e7874 3574
bd86fb99
HR
3575 assert(!child_class || !flags);
3576 assert(!child_class == !parent);
f67503e5 3577
ddf5636d
HR
3578 if (reference) {
3579 bool options_non_empty = options ? qdict_size(options) : false;
cb3e7f08 3580 qobject_unref(options);
ddf5636d 3581
ddf5636d
HR
3582 if (filename || options_non_empty) {
3583 error_setg(errp, "Cannot reference an existing block device with "
3584 "additional options or a new filename");
5b363937 3585 return NULL;
ddf5636d
HR
3586 }
3587
3588 bs = bdrv_lookup_bs(reference, reference, errp);
3589 if (!bs) {
5b363937 3590 return NULL;
ddf5636d 3591 }
76b22320 3592
ddf5636d 3593 bdrv_ref(bs);
5b363937 3594 return bs;
ddf5636d
HR
3595 }
3596
5b363937 3597 bs = bdrv_new();
f67503e5 3598
de9c0cec
KW
3599 /* NULL means an empty set of options */
3600 if (options == NULL) {
3601 options = qdict_new();
3602 }
3603
145f598e 3604 /* json: syntax counts as explicit options, as if in the QDict */
de3b53f0
KW
3605 parse_json_protocol(options, &filename, &local_err);
3606 if (local_err) {
de3b53f0
KW
3607 goto fail;
3608 }
3609
145f598e
KW
3610 bs->explicit_options = qdict_clone_shallow(options);
3611
bd86fb99 3612 if (child_class) {
3cdc69d3
HR
3613 bool parent_is_format;
3614
3615 if (parent->drv) {
3616 parent_is_format = parent->drv->is_format;
3617 } else {
3618 /*
3619 * parent->drv is not set yet because this node is opened for
3620 * (potential) format probing. That means that @parent is going
3621 * to be a format node.
3622 */
3623 parent_is_format = true;
3624 }
3625
bddcec37 3626 bs->inherits_from = parent;
3cdc69d3
HR
3627 child_class->inherit_options(child_role, parent_is_format,
3628 &flags, options,
bd86fb99 3629 parent->open_flags, parent->options);
f3930ed0
KW
3630 }
3631
de3b53f0 3632 ret = bdrv_fill_options(&options, filename, &flags, &local_err);
dfde483e 3633 if (ret < 0) {
462f5bcf
KW
3634 goto fail;
3635 }
3636
129c7d1c
MA
3637 /*
3638 * Set the BDRV_O_RDWR and BDRV_O_ALLOW_RDWR flags.
3639 * Caution: getting a boolean member of @options requires care.
3640 * When @options come from -blockdev or blockdev_add, members are
3641 * typed according to the QAPI schema, but when they come from
3642 * -drive, they're all QString.
3643 */
f87a0e29
AG
3644 if (g_strcmp0(qdict_get_try_str(options, BDRV_OPT_READ_ONLY), "on") &&
3645 !qdict_get_try_bool(options, BDRV_OPT_READ_ONLY, false)) {
3646 flags |= (BDRV_O_RDWR | BDRV_O_ALLOW_RDWR);
3647 } else {
3648 flags &= ~BDRV_O_RDWR;
14499ea5
AG
3649 }
3650
3651 if (flags & BDRV_O_SNAPSHOT) {
3652 snapshot_options = qdict_new();
3653 bdrv_temp_snapshot_options(&snapshot_flags, snapshot_options,
3654 flags, options);
f87a0e29
AG
3655 /* Let bdrv_backing_options() override "read-only" */
3656 qdict_del(options, BDRV_OPT_READ_ONLY);
00ff7ffd
HR
3657 bdrv_inherited_options(BDRV_CHILD_COW, true,
3658 &flags, options, flags, options);
14499ea5
AG
3659 }
3660
62392ebb
KW
3661 bs->open_flags = flags;
3662 bs->options = options;
3663 options = qdict_clone_shallow(options);
3664
76c591b0 3665 /* Find the right image format driver */
129c7d1c 3666 /* See cautionary note on accessing @options above */
76c591b0
KW
3667 drvname = qdict_get_try_str(options, "driver");
3668 if (drvname) {
3669 drv = bdrv_find_format(drvname);
76c591b0
KW
3670 if (!drv) {
3671 error_setg(errp, "Unknown driver: '%s'", drvname);
76c591b0
KW
3672 goto fail;
3673 }
3674 }
3675
3676 assert(drvname || !(flags & BDRV_O_PROTOCOL));
76c591b0 3677
129c7d1c 3678 /* See cautionary note on accessing @options above */
3e8c2e57 3679 backing = qdict_get_try_str(options, "backing");
e59a0cf1
HR
3680 if (qobject_to(QNull, qdict_get(options, "backing")) != NULL ||
3681 (backing && *backing == '\0'))
3682 {
4f7be280
HR
3683 if (backing) {
3684 warn_report("Use of \"backing\": \"\" is deprecated; "
3685 "use \"backing\": null instead");
3686 }
3e8c2e57 3687 flags |= BDRV_O_NO_BACKING;
ae0f57f0
KW
3688 qdict_del(bs->explicit_options, "backing");
3689 qdict_del(bs->options, "backing");
3e8c2e57
AG
3690 qdict_del(options, "backing");
3691 }
3692
5696c6e3 3693 /* Open image file without format layer. This BlockBackend is only used for
4e4bf5c4
KW
3694 * probing, the block drivers will do their own bdrv_open_child() for the
3695 * same BDS, which is why we put the node name back into options. */
f4788adc 3696 if ((flags & BDRV_O_PROTOCOL) == 0) {
5696c6e3
KW
3697 BlockDriverState *file_bs;
3698
3699 file_bs = bdrv_open_child_bs(filename, options, "file", bs,
58944401
HR
3700 &child_of_bds, BDRV_CHILD_IMAGE,
3701 true, &local_err);
1fdd6933 3702 if (local_err) {
f4788adc
KW
3703 goto fail;
3704 }
5696c6e3 3705 if (file_bs != NULL) {
dacaa162
KW
3706 /* Not requesting BLK_PERM_CONSISTENT_READ because we're only
3707 * looking at the header to guess the image format. This works even
3708 * in cases where a guest would not see a consistent state. */
d861ab3a 3709 file = blk_new(bdrv_get_aio_context(file_bs), 0, BLK_PERM_ALL);
d7086422 3710 blk_insert_bs(file, file_bs, &local_err);
5696c6e3 3711 bdrv_unref(file_bs);
d7086422
KW
3712 if (local_err) {
3713 goto fail;
3714 }
5696c6e3 3715
46f5ac20 3716 qdict_put_str(options, "file", bdrv_get_node_name(file_bs));
4e4bf5c4 3717 }
f500a6d3
KW
3718 }
3719
76c591b0 3720 /* Image format probing */
38f3ef57 3721 bs->probed = !drv;
76c591b0 3722 if (!drv && file) {
cf2ab8fc 3723 ret = find_image_format(file, filename, &drv, &local_err);
17b005f1 3724 if (ret < 0) {
8bfea15d 3725 goto fail;
2a05cbe4 3726 }
62392ebb
KW
3727 /*
3728 * This option update would logically belong in bdrv_fill_options(),
3729 * but we first need to open bs->file for the probing to work, while
3730 * opening bs->file already requires the (mostly) final set of options
3731 * so that cache mode etc. can be inherited.
3732 *
3733 * Adding the driver later is somewhat ugly, but it's not an option
3734 * that would ever be inherited, so it's correct. We just need to make
3735 * sure to update both bs->options (which has the full effective
3736 * options for bs) and options (which has file.* already removed).
3737 */
46f5ac20
EB
3738 qdict_put_str(bs->options, "driver", drv->format_name);
3739 qdict_put_str(options, "driver", drv->format_name);
76c591b0 3740 } else if (!drv) {
17b005f1 3741 error_setg(errp, "Must specify either driver or file");
8bfea15d 3742 goto fail;
ea2384d3 3743 }
b6ce07aa 3744
53a29513
HR
3745 /* BDRV_O_PROTOCOL must be set iff a protocol BDS is about to be created */
3746 assert(!!(flags & BDRV_O_PROTOCOL) == !!drv->bdrv_file_open);
3747 /* file must be NULL if a protocol BDS is about to be created
3748 * (the inverse results in an error message from bdrv_open_common()) */
3749 assert(!(flags & BDRV_O_PROTOCOL) || !file);
3750
b6ce07aa 3751 /* Open the image */
82dc8b41 3752 ret = bdrv_open_common(bs, file, options, &local_err);
b6ce07aa 3753 if (ret < 0) {
8bfea15d 3754 goto fail;
6987307c
CH
3755 }
3756
4e4bf5c4 3757 if (file) {
5696c6e3 3758 blk_unref(file);
f500a6d3
KW
3759 file = NULL;
3760 }
3761
b6ce07aa 3762 /* If there is a backing file, use it */
9156df12 3763 if ((flags & BDRV_O_NO_BACKING) == 0) {
d9b7b057 3764 ret = bdrv_open_backing_file(bs, options, "backing", &local_err);
b6ce07aa 3765 if (ret < 0) {
b6ad491a 3766 goto close_and_fail;
b6ce07aa 3767 }
b6ce07aa
KW
3768 }
3769
50196d7a
AG
3770 /* Remove all children options and references
3771 * from bs->options and bs->explicit_options */
2f624b80
AG
3772 QLIST_FOREACH(child, &bs->children, next) {
3773 char *child_key_dot;
3774 child_key_dot = g_strdup_printf("%s.", child->name);
3775 qdict_extract_subqdict(bs->explicit_options, NULL, child_key_dot);
3776 qdict_extract_subqdict(bs->options, NULL, child_key_dot);
50196d7a
AG
3777 qdict_del(bs->explicit_options, child->name);
3778 qdict_del(bs->options, child->name);
2f624b80
AG
3779 g_free(child_key_dot);
3780 }
3781
b6ad491a 3782 /* Check if any unknown options were used */
7ad2757f 3783 if (qdict_size(options) != 0) {
b6ad491a 3784 const QDictEntry *entry = qdict_first(options);
5acd9d81
HR
3785 if (flags & BDRV_O_PROTOCOL) {
3786 error_setg(errp, "Block protocol '%s' doesn't support the option "
3787 "'%s'", drv->format_name, entry->key);
3788 } else {
d0e46a55
HR
3789 error_setg(errp,
3790 "Block format '%s' does not support the option '%s'",
3791 drv->format_name, entry->key);
5acd9d81 3792 }
b6ad491a 3793
b6ad491a
KW
3794 goto close_and_fail;
3795 }
b6ad491a 3796
c01c214b 3797 bdrv_parent_cb_change_media(bs, true);
b6ce07aa 3798
cb3e7f08 3799 qobject_unref(options);
8961be33 3800 options = NULL;
dd62f1ca
KW
3801
3802 /* For snapshot=on, create a temporary qcow2 overlay. bs points to the
3803 * temporary snapshot afterwards. */
3804 if (snapshot_flags) {
66836189
HR
3805 BlockDriverState *snapshot_bs;
3806 snapshot_bs = bdrv_append_temp_snapshot(bs, snapshot_flags,
3807 snapshot_options, &local_err);
73176bee 3808 snapshot_options = NULL;
dd62f1ca
KW
3809 if (local_err) {
3810 goto close_and_fail;
3811 }
5b363937
HR
3812 /* We are not going to return bs but the overlay on top of it
3813 * (snapshot_bs); thus, we have to drop the strong reference to bs
3814 * (which we obtained by calling bdrv_new()). bs will not be deleted,
3815 * though, because the overlay still has a reference to it. */
3816 bdrv_unref(bs);
3817 bs = snapshot_bs;
dd62f1ca
KW
3818 }
3819
5b363937 3820 return bs;
b6ce07aa 3821
8bfea15d 3822fail:
5696c6e3 3823 blk_unref(file);
cb3e7f08
MAL
3824 qobject_unref(snapshot_options);
3825 qobject_unref(bs->explicit_options);
3826 qobject_unref(bs->options);
3827 qobject_unref(options);
de9c0cec 3828 bs->options = NULL;
998cbd6a 3829 bs->explicit_options = NULL;
5b363937 3830 bdrv_unref(bs);
621ff94d 3831 error_propagate(errp, local_err);
5b363937 3832 return NULL;
de9c0cec 3833
b6ad491a 3834close_and_fail:
5b363937 3835 bdrv_unref(bs);
cb3e7f08
MAL
3836 qobject_unref(snapshot_options);
3837 qobject_unref(options);
621ff94d 3838 error_propagate(errp, local_err);
5b363937 3839 return NULL;
b6ce07aa
KW
3840}
3841
5b363937
HR
3842BlockDriverState *bdrv_open(const char *filename, const char *reference,
3843 QDict *options, int flags, Error **errp)
f3930ed0 3844{
5b363937 3845 return bdrv_open_inherit(filename, reference, options, flags, NULL,
272c02ea 3846 NULL, 0, errp);
f3930ed0
KW
3847}
3848
faf116b4
AG
3849/* Return true if the NULL-terminated @list contains @str */
3850static bool is_str_in_list(const char *str, const char *const *list)
3851{
3852 if (str && list) {
3853 int i;
3854 for (i = 0; list[i] != NULL; i++) {
3855 if (!strcmp(str, list[i])) {
3856 return true;
3857 }
3858 }
3859 }
3860 return false;
3861}
3862
3863/*
3864 * Check that every option set in @bs->options is also set in
3865 * @new_opts.
3866 *
3867 * Options listed in the common_options list and in
3868 * @bs->drv->mutable_opts are skipped.
3869 *
3870 * Return 0 on success, otherwise return -EINVAL and set @errp.
3871 */
3872static int bdrv_reset_options_allowed(BlockDriverState *bs,
3873 const QDict *new_opts, Error **errp)
3874{
3875 const QDictEntry *e;
3876 /* These options are common to all block drivers and are handled
3877 * in bdrv_reopen_prepare() so they can be left out of @new_opts */
3878 const char *const common_options[] = {
3879 "node-name", "discard", "cache.direct", "cache.no-flush",
3880 "read-only", "auto-read-only", "detect-zeroes", NULL
3881 };
3882
3883 for (e = qdict_first(bs->options); e; e = qdict_next(bs->options, e)) {
3884 if (!qdict_haskey(new_opts, e->key) &&
3885 !is_str_in_list(e->key, common_options) &&
3886 !is_str_in_list(e->key, bs->drv->mutable_opts)) {
3887 error_setg(errp, "Option '%s' cannot be reset "
3888 "to its default value", e->key);
3889 return -EINVAL;
3890 }
3891 }
3892
3893 return 0;
3894}
3895
cb828c31
AG
3896/*
3897 * Returns true if @child can be reached recursively from @bs
3898 */
3899static bool bdrv_recurse_has_child(BlockDriverState *bs,
3900 BlockDriverState *child)
3901{
3902 BdrvChild *c;
3903
3904 if (bs == child) {
3905 return true;
3906 }
3907
3908 QLIST_FOREACH(c, &bs->children, next) {
3909 if (bdrv_recurse_has_child(c->bs, child)) {
3910 return true;
3911 }
3912 }
3913
3914 return false;
3915}
3916
e971aa12
JC
3917/*
3918 * Adds a BlockDriverState to a simple queue for an atomic, transactional
3919 * reopen of multiple devices.
3920 *
859443b0 3921 * bs_queue can either be an existing BlockReopenQueue that has had QTAILQ_INIT
e971aa12
JC
3922 * already performed, or alternatively may be NULL a new BlockReopenQueue will
3923 * be created and initialized. This newly created BlockReopenQueue should be
3924 * passed back in for subsequent calls that are intended to be of the same
3925 * atomic 'set'.
3926 *
3927 * bs is the BlockDriverState to add to the reopen queue.
3928 *
4d2cb092
KW
3929 * options contains the changed options for the associated bs
3930 * (the BlockReopenQueue takes ownership)
3931 *
e971aa12
JC
3932 * flags contains the open flags for the associated bs
3933 *
3934 * returns a pointer to bs_queue, which is either the newly allocated
3935 * bs_queue, or the existing bs_queue being used.
3936 *
1a63a907 3937 * bs must be drained between bdrv_reopen_queue() and bdrv_reopen_multiple().
e971aa12 3938 */
28518102
KW
3939static BlockReopenQueue *bdrv_reopen_queue_child(BlockReopenQueue *bs_queue,
3940 BlockDriverState *bs,
3941 QDict *options,
bd86fb99 3942 const BdrvChildClass *klass,
272c02ea 3943 BdrvChildRole role,
3cdc69d3 3944 bool parent_is_format,
28518102 3945 QDict *parent_options,
077e8e20
AG
3946 int parent_flags,
3947 bool keep_old_opts)
e971aa12
JC
3948{
3949 assert(bs != NULL);
3950
3951 BlockReopenQueueEntry *bs_entry;
67251a31 3952 BdrvChild *child;
9aa09ddd
AG
3953 QDict *old_options, *explicit_options, *options_copy;
3954 int flags;
3955 QemuOpts *opts;
67251a31 3956
1a63a907
KW
3957 /* Make sure that the caller remembered to use a drained section. This is
3958 * important to avoid graph changes between the recursive queuing here and
3959 * bdrv_reopen_multiple(). */
3960 assert(bs->quiesce_counter > 0);
3961
e971aa12
JC
3962 if (bs_queue == NULL) {
3963 bs_queue = g_new0(BlockReopenQueue, 1);
859443b0 3964 QTAILQ_INIT(bs_queue);
e971aa12
JC
3965 }
3966
4d2cb092
KW
3967 if (!options) {
3968 options = qdict_new();
3969 }
3970
5b7ba05f 3971 /* Check if this BlockDriverState is already in the queue */
859443b0 3972 QTAILQ_FOREACH(bs_entry, bs_queue, entry) {
5b7ba05f
AG
3973 if (bs == bs_entry->state.bs) {
3974 break;
3975 }
3976 }
3977
28518102
KW
3978 /*
3979 * Precedence of options:
3980 * 1. Explicitly passed in options (highest)
9aa09ddd
AG
3981 * 2. Retained from explicitly set options of bs
3982 * 3. Inherited from parent node
3983 * 4. Retained from effective options of bs
28518102
KW
3984 */
3985
145f598e 3986 /* Old explicitly set values (don't overwrite by inherited value) */
077e8e20
AG
3987 if (bs_entry || keep_old_opts) {
3988 old_options = qdict_clone_shallow(bs_entry ?
3989 bs_entry->state.explicit_options :
3990 bs->explicit_options);
3991 bdrv_join_options(bs, options, old_options);
3992 qobject_unref(old_options);
5b7ba05f 3993 }
145f598e
KW
3994
3995 explicit_options = qdict_clone_shallow(options);
3996
28518102
KW
3997 /* Inherit from parent node */
3998 if (parent_options) {
9aa09ddd 3999 flags = 0;
3cdc69d3 4000 klass->inherit_options(role, parent_is_format, &flags, options,
272c02ea 4001 parent_flags, parent_options);
9aa09ddd
AG
4002 } else {
4003 flags = bdrv_get_flags(bs);
28518102
KW
4004 }
4005
077e8e20
AG
4006 if (keep_old_opts) {
4007 /* Old values are used for options that aren't set yet */
4008 old_options = qdict_clone_shallow(bs->options);
4009 bdrv_join_options(bs, options, old_options);
4010 qobject_unref(old_options);
4011 }
4d2cb092 4012
9aa09ddd
AG
4013 /* We have the final set of options so let's update the flags */
4014 options_copy = qdict_clone_shallow(options);
4015 opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
4016 qemu_opts_absorb_qdict(opts, options_copy, NULL);
4017 update_flags_from_options(&flags, opts);
4018 qemu_opts_del(opts);
4019 qobject_unref(options_copy);
4020
fd452021 4021 /* bdrv_open_inherit() sets and clears some additional flags internally */
f1f25a2e 4022 flags &= ~BDRV_O_PROTOCOL;
fd452021
KW
4023 if (flags & BDRV_O_RDWR) {
4024 flags |= BDRV_O_ALLOW_RDWR;
4025 }
f1f25a2e 4026
1857c97b
KW
4027 if (!bs_entry) {
4028 bs_entry = g_new0(BlockReopenQueueEntry, 1);
859443b0 4029 QTAILQ_INSERT_TAIL(bs_queue, bs_entry, entry);
1857c97b 4030 } else {
cb3e7f08
MAL
4031 qobject_unref(bs_entry->state.options);
4032 qobject_unref(bs_entry->state.explicit_options);
1857c97b
KW
4033 }
4034
4035 bs_entry->state.bs = bs;
4036 bs_entry->state.options = options;
4037 bs_entry->state.explicit_options = explicit_options;
4038 bs_entry->state.flags = flags;
4039
8546632e
AG
4040 /*
4041 * If keep_old_opts is false then it means that unspecified
4042 * options must be reset to their original value. We don't allow
4043 * resetting 'backing' but we need to know if the option is
4044 * missing in order to decide if we have to return an error.
4045 */
4046 if (!keep_old_opts) {
4047 bs_entry->state.backing_missing =
4048 !qdict_haskey(options, "backing") &&
4049 !qdict_haskey(options, "backing.driver");
4050 }
4051
67251a31 4052 QLIST_FOREACH(child, &bs->children, next) {
8546632e
AG
4053 QDict *new_child_options = NULL;
4054 bool child_keep_old = keep_old_opts;
67251a31 4055
4c9dfe5d
KW
4056 /* reopen can only change the options of block devices that were
4057 * implicitly created and inherited options. For other (referenced)
4058 * block devices, a syntax like "backing.foo" results in an error. */
67251a31
KW
4059 if (child->bs->inherits_from != bs) {
4060 continue;
4061 }
4062
8546632e
AG
4063 /* Check if the options contain a child reference */
4064 if (qdict_haskey(options, child->name)) {
4065 const char *childref = qdict_get_try_str(options, child->name);
4066 /*
4067 * The current child must not be reopened if the child
4068 * reference is null or points to a different node.
4069 */
4070 if (g_strcmp0(childref, child->bs->node_name)) {
4071 continue;
4072 }
4073 /*
4074 * If the child reference points to the current child then
4075 * reopen it with its existing set of options (note that
4076 * it can still inherit new options from the parent).
4077 */
4078 child_keep_old = true;
4079 } else {
4080 /* Extract child options ("child-name.*") */
4081 char *child_key_dot = g_strdup_printf("%s.", child->name);
4082 qdict_extract_subqdict(explicit_options, NULL, child_key_dot);
4083 qdict_extract_subqdict(options, &new_child_options, child_key_dot);
4084 g_free(child_key_dot);
4085 }
4c9dfe5d 4086
9aa09ddd 4087 bdrv_reopen_queue_child(bs_queue, child->bs, new_child_options,
3cdc69d3
HR
4088 child->klass, child->role, bs->drv->is_format,
4089 options, flags, child_keep_old);
e971aa12
JC
4090 }
4091
e971aa12
JC
4092 return bs_queue;
4093}
4094
28518102
KW
4095BlockReopenQueue *bdrv_reopen_queue(BlockReopenQueue *bs_queue,
4096 BlockDriverState *bs,
077e8e20 4097 QDict *options, bool keep_old_opts)
28518102 4098{
3cdc69d3
HR
4099 return bdrv_reopen_queue_child(bs_queue, bs, options, NULL, 0, false,
4100 NULL, 0, keep_old_opts);
28518102
KW
4101}
4102
ab5b5228
AG
4103void bdrv_reopen_queue_free(BlockReopenQueue *bs_queue)
4104{
4105 if (bs_queue) {
4106 BlockReopenQueueEntry *bs_entry, *next;
4107 QTAILQ_FOREACH_SAFE(bs_entry, bs_queue, entry, next) {
4108 qobject_unref(bs_entry->state.explicit_options);
4109 qobject_unref(bs_entry->state.options);
4110 g_free(bs_entry);
4111 }
4112 g_free(bs_queue);
4113 }
4114}
4115
e971aa12
JC
4116/*
4117 * Reopen multiple BlockDriverStates atomically & transactionally.
4118 *
4119 * The queue passed in (bs_queue) must have been built up previous
4120 * via bdrv_reopen_queue().
4121 *
4122 * Reopens all BDS specified in the queue, with the appropriate
4123 * flags. All devices are prepared for reopen, and failure of any
50d6a8a3 4124 * device will cause all device changes to be abandoned, and intermediate
e971aa12
JC
4125 * data cleaned up.
4126 *
4127 * If all devices prepare successfully, then the changes are committed
4128 * to all devices.
4129 *
1a63a907
KW
4130 * All affected nodes must be drained between bdrv_reopen_queue() and
4131 * bdrv_reopen_multiple().
6cf42ca2
KW
4132 *
4133 * To be called from the main thread, with all other AioContexts unlocked.
e971aa12 4134 */
5019aece 4135int bdrv_reopen_multiple(BlockReopenQueue *bs_queue, Error **errp)
e971aa12
JC
4136{
4137 int ret = -1;
4138 BlockReopenQueueEntry *bs_entry, *next;
6cf42ca2 4139 AioContext *ctx;
72373e40
VSO
4140 Transaction *tran = tran_new();
4141 g_autoptr(GHashTable) found = NULL;
4142 g_autoptr(GSList) refresh_list = NULL;
e971aa12 4143
6cf42ca2 4144 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
e971aa12
JC
4145 assert(bs_queue != NULL);
4146
a2aabf88 4147 QTAILQ_FOREACH(bs_entry, bs_queue, entry) {
6cf42ca2
KW
4148 ctx = bdrv_get_aio_context(bs_entry->state.bs);
4149 aio_context_acquire(ctx);
a2aabf88 4150 ret = bdrv_flush(bs_entry->state.bs);
6cf42ca2 4151 aio_context_release(ctx);
a2aabf88
VSO
4152 if (ret < 0) {
4153 error_setg_errno(errp, -ret, "Error flushing drive");
e3fc91aa 4154 goto abort;
a2aabf88
VSO
4155 }
4156 }
4157
859443b0 4158 QTAILQ_FOREACH(bs_entry, bs_queue, entry) {
1a63a907 4159 assert(bs_entry->state.bs->quiesce_counter > 0);
6cf42ca2
KW
4160 ctx = bdrv_get_aio_context(bs_entry->state.bs);
4161 aio_context_acquire(ctx);
72373e40 4162 ret = bdrv_reopen_prepare(&bs_entry->state, bs_queue, tran, errp);
6cf42ca2 4163 aio_context_release(ctx);
72373e40
VSO
4164 if (ret < 0) {
4165 goto abort;
e971aa12
JC
4166 }
4167 bs_entry->prepared = true;
4168 }
4169
72373e40 4170 found = g_hash_table_new(NULL, NULL);
859443b0 4171 QTAILQ_FOREACH(bs_entry, bs_queue, entry) {
69b736e7 4172 BDRVReopenState *state = &bs_entry->state;
72373e40
VSO
4173
4174 refresh_list = bdrv_topological_dfs(refresh_list, found, state->bs);
4175 if (state->old_backing_bs) {
4176 refresh_list = bdrv_topological_dfs(refresh_list, found,
4177 state->old_backing_bs);
cb828c31 4178 }
ecd30d2d
AG
4179 if (state->old_file_bs) {
4180 refresh_list = bdrv_topological_dfs(refresh_list, found,
4181 state->old_file_bs);
4182 }
72373e40
VSO
4183 }
4184
4185 /*
4186 * Note that file-posix driver rely on permission update done during reopen
4187 * (even if no permission changed), because it wants "new" permissions for
4188 * reconfiguring the fd and that's why it does it in raw_check_perm(), not
4189 * in raw_reopen_prepare() which is called with "old" permissions.
4190 */
4191 ret = bdrv_list_refresh_perms(refresh_list, bs_queue, tran, errp);
4192 if (ret < 0) {
4193 goto abort;
69b736e7
KW
4194 }
4195
fcd6a4f4
VSO
4196 /*
4197 * If we reach this point, we have success and just need to apply the
4198 * changes.
4199 *
4200 * Reverse order is used to comfort qcow2 driver: on commit it need to write
4201 * IN_USE flag to the image, to mark bitmaps in the image as invalid. But
4202 * children are usually goes after parents in reopen-queue, so go from last
4203 * to first element.
e971aa12 4204 */
fcd6a4f4 4205 QTAILQ_FOREACH_REVERSE(bs_entry, bs_queue, entry) {
6cf42ca2
KW
4206 ctx = bdrv_get_aio_context(bs_entry->state.bs);
4207 aio_context_acquire(ctx);
e971aa12 4208 bdrv_reopen_commit(&bs_entry->state);
6cf42ca2 4209 aio_context_release(ctx);
e971aa12
JC
4210 }
4211
72373e40 4212 tran_commit(tran);
69b736e7 4213
72373e40
VSO
4214 QTAILQ_FOREACH_REVERSE(bs_entry, bs_queue, entry) {
4215 BlockDriverState *bs = bs_entry->state.bs;
74ad9a3b 4216
72373e40 4217 if (bs->drv->bdrv_reopen_commit_post) {
6cf42ca2
KW
4218 ctx = bdrv_get_aio_context(bs);
4219 aio_context_acquire(ctx);
72373e40 4220 bs->drv->bdrv_reopen_commit_post(&bs_entry->state);
6cf42ca2 4221 aio_context_release(ctx);
69b736e7
KW
4222 }
4223 }
17e1e2be 4224
72373e40
VSO
4225 ret = 0;
4226 goto cleanup;
17e1e2be 4227
72373e40
VSO
4228abort:
4229 tran_abort(tran);
4230 QTAILQ_FOREACH_SAFE(bs_entry, bs_queue, entry, next) {
4231 if (bs_entry->prepared) {
6cf42ca2
KW
4232 ctx = bdrv_get_aio_context(bs_entry->state.bs);
4233 aio_context_acquire(ctx);
72373e40 4234 bdrv_reopen_abort(&bs_entry->state);
6cf42ca2 4235 aio_context_release(ctx);
17e1e2be
PK
4236 }
4237 }
72373e40 4238
e971aa12 4239cleanup:
ab5b5228 4240 bdrv_reopen_queue_free(bs_queue);
40840e41 4241
e971aa12
JC
4242 return ret;
4243}
4244
6cf42ca2
KW
4245int bdrv_reopen(BlockDriverState *bs, QDict *opts, bool keep_old_opts,
4246 Error **errp)
6e1000a8 4247{
6cf42ca2 4248 AioContext *ctx = bdrv_get_aio_context(bs);
6e1000a8 4249 BlockReopenQueue *queue;
6cf42ca2 4250 int ret;
6e1000a8
AG
4251
4252 bdrv_subtree_drained_begin(bs);
6cf42ca2
KW
4253 if (ctx != qemu_get_aio_context()) {
4254 aio_context_release(ctx);
4255 }
4256
4257 queue = bdrv_reopen_queue(NULL, bs, opts, keep_old_opts);
5019aece 4258 ret = bdrv_reopen_multiple(queue, errp);
6cf42ca2
KW
4259
4260 if (ctx != qemu_get_aio_context()) {
4261 aio_context_acquire(ctx);
4262 }
6e1000a8
AG
4263 bdrv_subtree_drained_end(bs);
4264
4265 return ret;
4266}
4267
6cf42ca2
KW
4268int bdrv_reopen_set_read_only(BlockDriverState *bs, bool read_only,
4269 Error **errp)
4270{
4271 QDict *opts = qdict_new();
4272
4273 qdict_put_bool(opts, BDRV_OPT_READ_ONLY, read_only);
4274
4275 return bdrv_reopen(bs, opts, true, errp);
4276}
4277
cb828c31
AG
4278/*
4279 * Take a BDRVReopenState and check if the value of 'backing' in the
4280 * reopen_state->options QDict is valid or not.
4281 *
4282 * If 'backing' is missing from the QDict then return 0.
4283 *
4284 * If 'backing' contains the node name of the backing file of
4285 * reopen_state->bs then return 0.
4286 *
4287 * If 'backing' contains a different node name (or is null) then check
4288 * whether the current backing file can be replaced with the new one.
4289 * If that's the case then reopen_state->replace_backing_bs is set to
4290 * true and reopen_state->new_backing_bs contains a pointer to the new
4291 * backing BlockDriverState (or NULL).
4292 *
4293 * Return 0 on success, otherwise return < 0 and set @errp.
4294 */
ecd30d2d
AG
4295static int bdrv_reopen_parse_file_or_backing(BDRVReopenState *reopen_state,
4296 bool is_backing, Transaction *tran,
4297 Error **errp)
cb828c31
AG
4298{
4299 BlockDriverState *bs = reopen_state->bs;
ecd30d2d
AG
4300 BlockDriverState *new_child_bs;
4301 BlockDriverState *old_child_bs = is_backing ? child_bs(bs->backing) :
4302 child_bs(bs->file);
4303 const char *child_name = is_backing ? "backing" : "file";
cb828c31
AG
4304 QObject *value;
4305 const char *str;
4306
ecd30d2d 4307 value = qdict_get(reopen_state->options, child_name);
cb828c31
AG
4308 if (value == NULL) {
4309 return 0;
4310 }
4311
4312 switch (qobject_type(value)) {
4313 case QTYPE_QNULL:
ecd30d2d
AG
4314 assert(is_backing); /* The 'file' option does not allow a null value */
4315 new_child_bs = NULL;
cb828c31
AG
4316 break;
4317 case QTYPE_QSTRING:
410f44f5 4318 str = qstring_get_str(qobject_to(QString, value));
ecd30d2d
AG
4319 new_child_bs = bdrv_lookup_bs(NULL, str, errp);
4320 if (new_child_bs == NULL) {
cb828c31 4321 return -EINVAL;
ecd30d2d
AG
4322 } else if (bdrv_recurse_has_child(new_child_bs, bs)) {
4323 error_setg(errp, "Making '%s' a %s child of '%s' would create a "
4324 "cycle", str, child_name, bs->node_name);
cb828c31
AG
4325 return -EINVAL;
4326 }
4327 break;
4328 default:
ecd30d2d
AG
4329 /*
4330 * The options QDict has been flattened, so 'backing' and 'file'
4331 * do not allow any other data type here.
4332 */
cb828c31
AG
4333 g_assert_not_reached();
4334 }
4335
ecd30d2d
AG
4336 if (old_child_bs == new_child_bs) {
4337 return 0;
4338 }
4339
4340 if (old_child_bs) {
4341 if (bdrv_skip_implicit_filters(old_child_bs) == new_child_bs) {
cbfdb98c
VSO
4342 return 0;
4343 }
4344
ecd30d2d
AG
4345 if (old_child_bs->implicit) {
4346 error_setg(errp, "Cannot replace implicit %s child of %s",
4347 child_name, bs->node_name);
cbfdb98c
VSO
4348 return -EPERM;
4349 }
4350 }
4351
ecd30d2d 4352 if (bs->drv->is_filter && !old_child_bs) {
25f78d9e
VSO
4353 /*
4354 * Filters always have a file or a backing child, so we are trying to
4355 * change wrong child
4356 */
4357 error_setg(errp, "'%s' is a %s filter node that does not support a "
ecd30d2d 4358 "%s child", bs->node_name, bs->drv->format_name, child_name);
1d42f48c
HR
4359 return -EINVAL;
4360 }
4361
ecd30d2d
AG
4362 if (is_backing) {
4363 reopen_state->old_backing_bs = old_child_bs;
4364 } else {
4365 reopen_state->old_file_bs = old_child_bs;
4366 }
4367
4368 return bdrv_set_file_or_backing_noperm(bs, new_child_bs, is_backing,
4369 tran, errp);
cb828c31
AG
4370}
4371
e971aa12
JC
4372/*
4373 * Prepares a BlockDriverState for reopen. All changes are staged in the
4374 * 'opaque' field of the BDRVReopenState, which is used and allocated by
4375 * the block driver layer .bdrv_reopen_prepare()
4376 *
4377 * bs is the BlockDriverState to reopen
4378 * flags are the new open flags
4379 * queue is the reopen queue
4380 *
4381 * Returns 0 on success, non-zero on error. On error errp will be set
4382 * as well.
4383 *
4384 * On failure, bdrv_reopen_abort() will be called to clean up any data.
4385 * It is the responsibility of the caller to then call the abort() or
4386 * commit() for any other BDS that have been left in a prepare() state
4387 *
4388 */
53e96d1e 4389static int bdrv_reopen_prepare(BDRVReopenState *reopen_state,
72373e40 4390 BlockReopenQueue *queue,
ecd30d2d 4391 Transaction *change_child_tran, Error **errp)
e971aa12
JC
4392{
4393 int ret = -1;
e6d79c41 4394 int old_flags;
e971aa12
JC
4395 Error *local_err = NULL;
4396 BlockDriver *drv;
ccf9dc07 4397 QemuOpts *opts;
4c8350fe 4398 QDict *orig_reopen_opts;
593b3071 4399 char *discard = NULL;
3d8ce171 4400 bool read_only;
9ad08c44 4401 bool drv_prepared = false;
e971aa12
JC
4402
4403 assert(reopen_state != NULL);
4404 assert(reopen_state->bs->drv != NULL);
4405 drv = reopen_state->bs->drv;
4406
4c8350fe
AG
4407 /* This function and each driver's bdrv_reopen_prepare() remove
4408 * entries from reopen_state->options as they are processed, so
4409 * we need to make a copy of the original QDict. */
4410 orig_reopen_opts = qdict_clone_shallow(reopen_state->options);
4411
ccf9dc07
KW
4412 /* Process generic block layer options */
4413 opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
af175e85 4414 if (!qemu_opts_absorb_qdict(opts, reopen_state->options, errp)) {
ccf9dc07
KW
4415 ret = -EINVAL;
4416 goto error;
4417 }
4418
e6d79c41
AG
4419 /* This was already called in bdrv_reopen_queue_child() so the flags
4420 * are up-to-date. This time we simply want to remove the options from
4421 * QemuOpts in order to indicate that they have been processed. */
4422 old_flags = reopen_state->flags;
91a097e7 4423 update_flags_from_options(&reopen_state->flags, opts);
e6d79c41 4424 assert(old_flags == reopen_state->flags);
91a097e7 4425
415bbca8 4426 discard = qemu_opt_get_del(opts, BDRV_OPT_DISCARD);
593b3071
AG
4427 if (discard != NULL) {
4428 if (bdrv_parse_discard_flags(discard, &reopen_state->flags) != 0) {
4429 error_setg(errp, "Invalid discard option");
4430 ret = -EINVAL;
4431 goto error;
4432 }
4433 }
4434
543770bd
AG
4435 reopen_state->detect_zeroes =
4436 bdrv_parse_detect_zeroes(opts, reopen_state->flags, &local_err);
4437 if (local_err) {
4438 error_propagate(errp, local_err);
4439 ret = -EINVAL;
4440 goto error;
4441 }
4442
57f9db9a
AG
4443 /* All other options (including node-name and driver) must be unchanged.
4444 * Put them back into the QDict, so that they are checked at the end
4445 * of this function. */
4446 qemu_opts_to_qdict(opts, reopen_state->options);
ccf9dc07 4447
3d8ce171
JC
4448 /* If we are to stay read-only, do not allow permission change
4449 * to r/w. Attempting to set to r/w may fail if either BDRV_O_ALLOW_RDWR is
4450 * not set, or if the BDS still has copy_on_read enabled */
4451 read_only = !(reopen_state->flags & BDRV_O_RDWR);
54a32bfe 4452 ret = bdrv_can_set_read_only(reopen_state->bs, read_only, true, &local_err);
3d8ce171
JC
4453 if (local_err) {
4454 error_propagate(errp, local_err);
e971aa12
JC
4455 goto error;
4456 }
4457
e971aa12 4458 if (drv->bdrv_reopen_prepare) {
faf116b4
AG
4459 /*
4460 * If a driver-specific option is missing, it means that we
4461 * should reset it to its default value.
4462 * But not all options allow that, so we need to check it first.
4463 */
4464 ret = bdrv_reset_options_allowed(reopen_state->bs,
4465 reopen_state->options, errp);
4466 if (ret) {
4467 goto error;
4468 }
4469
e971aa12
JC
4470 ret = drv->bdrv_reopen_prepare(reopen_state, queue, &local_err);
4471 if (ret) {
4472 if (local_err != NULL) {
4473 error_propagate(errp, local_err);
4474 } else {
f30c66ba 4475 bdrv_refresh_filename(reopen_state->bs);
d8b6895f
LC
4476 error_setg(errp, "failed while preparing to reopen image '%s'",
4477 reopen_state->bs->filename);
e971aa12
JC
4478 }
4479 goto error;
4480 }
4481 } else {
4482 /* It is currently mandatory to have a bdrv_reopen_prepare()
4483 * handler for each supported drv. */
81e5f78a
AG
4484 error_setg(errp, "Block format '%s' used by node '%s' "
4485 "does not support reopening files", drv->format_name,
4486 bdrv_get_device_or_node_name(reopen_state->bs));
e971aa12
JC
4487 ret = -1;
4488 goto error;
4489 }
4490
9ad08c44
HR
4491 drv_prepared = true;
4492
bacd9b87
AG
4493 /*
4494 * We must provide the 'backing' option if the BDS has a backing
4495 * file or if the image file has a backing file name as part of
4496 * its metadata. Otherwise the 'backing' option can be omitted.
4497 */
4498 if (drv->supports_backing && reopen_state->backing_missing &&
1d42f48c 4499 (reopen_state->bs->backing || reopen_state->bs->backing_file[0])) {
8546632e
AG
4500 error_setg(errp, "backing is missing for '%s'",
4501 reopen_state->bs->node_name);
4502 ret = -EINVAL;
4503 goto error;
4504 }
4505
cb828c31
AG
4506 /*
4507 * Allow changing the 'backing' option. The new value can be
4508 * either a reference to an existing node (using its node name)
4509 * or NULL to simply detach the current backing file.
4510 */
ecd30d2d
AG
4511 ret = bdrv_reopen_parse_file_or_backing(reopen_state, true,
4512 change_child_tran, errp);
cb828c31
AG
4513 if (ret < 0) {
4514 goto error;
4515 }
4516 qdict_del(reopen_state->options, "backing");
4517
ecd30d2d
AG
4518 /* Allow changing the 'file' option. In this case NULL is not allowed */
4519 ret = bdrv_reopen_parse_file_or_backing(reopen_state, false,
4520 change_child_tran, errp);
4521 if (ret < 0) {
4522 goto error;
4523 }
4524 qdict_del(reopen_state->options, "file");
4525
4d2cb092
KW
4526 /* Options that are not handled are only okay if they are unchanged
4527 * compared to the old state. It is expected that some options are only
4528 * used for the initial open, but not reopen (e.g. filename) */
4529 if (qdict_size(reopen_state->options)) {
4530 const QDictEntry *entry = qdict_first(reopen_state->options);
4531
4532 do {
54fd1b0d
HR
4533 QObject *new = entry->value;
4534 QObject *old = qdict_get(reopen_state->bs->options, entry->key);
4535
db905283
AG
4536 /* Allow child references (child_name=node_name) as long as they
4537 * point to the current child (i.e. everything stays the same). */
4538 if (qobject_type(new) == QTYPE_QSTRING) {
4539 BdrvChild *child;
4540 QLIST_FOREACH(child, &reopen_state->bs->children, next) {
4541 if (!strcmp(child->name, entry->key)) {
4542 break;
4543 }
4544 }
4545
4546 if (child) {
410f44f5
MA
4547 if (!strcmp(child->bs->node_name,
4548 qstring_get_str(qobject_to(QString, new)))) {
db905283
AG
4549 continue; /* Found child with this name, skip option */
4550 }
4551 }
4552 }
4553
129c7d1c 4554 /*
54fd1b0d
HR
4555 * TODO: When using -drive to specify blockdev options, all values
4556 * will be strings; however, when using -blockdev, blockdev-add or
4557 * filenames using the json:{} pseudo-protocol, they will be
4558 * correctly typed.
4559 * In contrast, reopening options are (currently) always strings
4560 * (because you can only specify them through qemu-io; all other
4561 * callers do not specify any options).
4562 * Therefore, when using anything other than -drive to create a BDS,
4563 * this cannot detect non-string options as unchanged, because
4564 * qobject_is_equal() always returns false for objects of different
4565 * type. In the future, this should be remedied by correctly typing
4566 * all options. For now, this is not too big of an issue because
4567 * the user can simply omit options which cannot be changed anyway,
4568 * so they will stay unchanged.
129c7d1c 4569 */
54fd1b0d 4570 if (!qobject_is_equal(new, old)) {
4d2cb092
KW
4571 error_setg(errp, "Cannot change the option '%s'", entry->key);
4572 ret = -EINVAL;
4573 goto error;
4574 }
4575 } while ((entry = qdict_next(reopen_state->options, entry)));
4576 }
4577
e971aa12
JC
4578 ret = 0;
4579
4c8350fe
AG
4580 /* Restore the original reopen_state->options QDict */
4581 qobject_unref(reopen_state->options);
4582 reopen_state->options = qobject_ref(orig_reopen_opts);
4583
e971aa12 4584error:
9ad08c44
HR
4585 if (ret < 0 && drv_prepared) {
4586 /* drv->bdrv_reopen_prepare() has succeeded, so we need to
4587 * call drv->bdrv_reopen_abort() before signaling an error
4588 * (bdrv_reopen_multiple() will not call bdrv_reopen_abort()
4589 * when the respective bdrv_reopen_prepare() has failed) */
4590 if (drv->bdrv_reopen_abort) {
4591 drv->bdrv_reopen_abort(reopen_state);
4592 }
4593 }
ccf9dc07 4594 qemu_opts_del(opts);
4c8350fe 4595 qobject_unref(orig_reopen_opts);
593b3071 4596 g_free(discard);
e971aa12
JC
4597 return ret;
4598}
4599
4600/*
4601 * Takes the staged changes for the reopen from bdrv_reopen_prepare(), and
4602 * makes them final by swapping the staging BlockDriverState contents into
4603 * the active BlockDriverState contents.
4604 */
53e96d1e 4605static void bdrv_reopen_commit(BDRVReopenState *reopen_state)
e971aa12
JC
4606{
4607 BlockDriver *drv;
50bf65ba 4608 BlockDriverState *bs;
50196d7a 4609 BdrvChild *child;
e971aa12
JC
4610
4611 assert(reopen_state != NULL);
50bf65ba
VSO
4612 bs = reopen_state->bs;
4613 drv = bs->drv;
e971aa12
JC
4614 assert(drv != NULL);
4615
4616 /* If there are any driver level actions to take */
4617 if (drv->bdrv_reopen_commit) {
4618 drv->bdrv_reopen_commit(reopen_state);
4619 }
4620
4621 /* set BDS specific flags now */
cb3e7f08 4622 qobject_unref(bs->explicit_options);
4c8350fe 4623 qobject_unref(bs->options);
ab5b5228
AG
4624 qobject_ref(reopen_state->explicit_options);
4625 qobject_ref(reopen_state->options);
145f598e 4626
50bf65ba 4627 bs->explicit_options = reopen_state->explicit_options;
4c8350fe 4628 bs->options = reopen_state->options;
50bf65ba 4629 bs->open_flags = reopen_state->flags;
543770bd 4630 bs->detect_zeroes = reopen_state->detect_zeroes;
355ef4ac 4631
50196d7a
AG
4632 /* Remove child references from bs->options and bs->explicit_options.
4633 * Child options were already removed in bdrv_reopen_queue_child() */
4634 QLIST_FOREACH(child, &bs->children, next) {
4635 qdict_del(bs->explicit_options, child->name);
4636 qdict_del(bs->options, child->name);
4637 }
3d0e8743
VSO
4638 /* backing is probably removed, so it's not handled by previous loop */
4639 qdict_del(bs->explicit_options, "backing");
4640 qdict_del(bs->options, "backing");
4641
1e4c797c 4642 bdrv_refresh_limits(bs, NULL, NULL);
e971aa12
JC
4643}
4644
4645/*
4646 * Abort the reopen, and delete and free the staged changes in
4647 * reopen_state
4648 */
53e96d1e 4649static void bdrv_reopen_abort(BDRVReopenState *reopen_state)
e971aa12
JC
4650{
4651 BlockDriver *drv;
4652
4653 assert(reopen_state != NULL);
4654 drv = reopen_state->bs->drv;
4655 assert(drv != NULL);
4656
4657 if (drv->bdrv_reopen_abort) {
4658 drv->bdrv_reopen_abort(reopen_state);
4659 }
4660}
4661
4662
64dff520 4663static void bdrv_close(BlockDriverState *bs)
fc01f7e7 4664{
33384421 4665 BdrvAioNotifier *ban, *ban_next;
50a3efb0 4666 BdrvChild *child, *next;
33384421 4667
30f55fb8 4668 assert(!bs->refcnt);
99b7e775 4669
fc27291d 4670 bdrv_drained_begin(bs); /* complete I/O */
58fda173 4671 bdrv_flush(bs);
53ec73e2 4672 bdrv_drain(bs); /* in case flush left pending I/O */
fc27291d 4673
3cbc002c 4674 if (bs->drv) {
3c005293 4675 if (bs->drv->bdrv_close) {
7b99a266 4676 /* Must unfreeze all children, so bdrv_unref_child() works */
3c005293
VSO
4677 bs->drv->bdrv_close(bs);
4678 }
9a4f4c31 4679 bs->drv = NULL;
50a3efb0 4680 }
9a7dedbc 4681
50a3efb0 4682 QLIST_FOREACH_SAFE(child, &bs->children, next, next) {
dd4118c7 4683 bdrv_unref_child(bs, child);
b338082b 4684 }
98f90dba 4685
dd4118c7
AG
4686 bs->backing = NULL;
4687 bs->file = NULL;
50a3efb0
AG
4688 g_free(bs->opaque);
4689 bs->opaque = NULL;
d73415a3 4690 qatomic_set(&bs->copy_on_read, 0);
50a3efb0
AG
4691 bs->backing_file[0] = '\0';
4692 bs->backing_format[0] = '\0';
4693 bs->total_sectors = 0;
4694 bs->encrypted = false;
4695 bs->sg = false;
cb3e7f08
MAL
4696 qobject_unref(bs->options);
4697 qobject_unref(bs->explicit_options);
50a3efb0
AG
4698 bs->options = NULL;
4699 bs->explicit_options = NULL;
cb3e7f08 4700 qobject_unref(bs->full_open_options);
50a3efb0 4701 bs->full_open_options = NULL;
0bc329fb
HR
4702 g_free(bs->block_status_cache);
4703 bs->block_status_cache = NULL;
50a3efb0 4704
cca43ae1
VSO
4705 bdrv_release_named_dirty_bitmaps(bs);
4706 assert(QLIST_EMPTY(&bs->dirty_bitmaps));
4707
33384421
HR
4708 QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_next) {
4709 g_free(ban);
4710 }
4711 QLIST_INIT(&bs->aio_notifiers);
fc27291d 4712 bdrv_drained_end(bs);
1a6d3bd2
GK
4713
4714 /*
4715 * If we're still inside some bdrv_drain_all_begin()/end() sections, end
4716 * them now since this BDS won't exist anymore when bdrv_drain_all_end()
4717 * gets called.
4718 */
4719 if (bs->quiesce_counter) {
4720 bdrv_drain_all_end_quiesce(bs);
4721 }
b338082b
FB
4722}
4723
2bc93fed
MK
4724void bdrv_close_all(void)
4725{
b3b5299d 4726 assert(job_next(NULL) == NULL);
ca9bd24c
HR
4727
4728 /* Drop references from requests still in flight, such as canceled block
4729 * jobs whose AIO context has not been polled yet */
4730 bdrv_drain_all();
2bc93fed 4731
ca9bd24c
HR
4732 blk_remove_all_bs();
4733 blockdev_close_all_bdrv_states();
ed78cda3 4734
a1a2af07 4735 assert(QTAILQ_EMPTY(&all_bdrv_states));
2bc93fed
MK
4736}
4737
d0ac0380
KW
4738static bool should_update_child(BdrvChild *c, BlockDriverState *to)
4739{
2f30b7c3
VSO
4740 GQueue *queue;
4741 GHashTable *found;
4742 bool ret;
d0ac0380 4743
bd86fb99 4744 if (c->klass->stay_at_node) {
d0ac0380
KW
4745 return false;
4746 }
4747
ec9f10fe
HR
4748 /* If the child @c belongs to the BDS @to, replacing the current
4749 * c->bs by @to would mean to create a loop.
4750 *
4751 * Such a case occurs when appending a BDS to a backing chain.
4752 * For instance, imagine the following chain:
4753 *
4754 * guest device -> node A -> further backing chain...
4755 *
4756 * Now we create a new BDS B which we want to put on top of this
4757 * chain, so we first attach A as its backing node:
4758 *
4759 * node B
4760 * |
4761 * v
4762 * guest device -> node A -> further backing chain...
4763 *
4764 * Finally we want to replace A by B. When doing that, we want to
4765 * replace all pointers to A by pointers to B -- except for the
4766 * pointer from B because (1) that would create a loop, and (2)
4767 * that pointer should simply stay intact:
4768 *
4769 * guest device -> node B
4770 * |
4771 * v
4772 * node A -> further backing chain...
4773 *
4774 * In general, when replacing a node A (c->bs) by a node B (@to),
4775 * if A is a child of B, that means we cannot replace A by B there
4776 * because that would create a loop. Silently detaching A from B
4777 * is also not really an option. So overall just leaving A in
2f30b7c3
VSO
4778 * place there is the most sensible choice.
4779 *
4780 * We would also create a loop in any cases where @c is only
4781 * indirectly referenced by @to. Prevent this by returning false
4782 * if @c is found (by breadth-first search) anywhere in the whole
4783 * subtree of @to.
4784 */
4785
4786 ret = true;
4787 found = g_hash_table_new(NULL, NULL);
4788 g_hash_table_add(found, to);
4789 queue = g_queue_new();
4790 g_queue_push_tail(queue, to);
4791
4792 while (!g_queue_is_empty(queue)) {
4793 BlockDriverState *v = g_queue_pop_head(queue);
4794 BdrvChild *c2;
4795
4796 QLIST_FOREACH(c2, &v->children, next) {
4797 if (c2 == c) {
4798 ret = false;
4799 break;
4800 }
4801
4802 if (g_hash_table_contains(found, c2->bs)) {
4803 continue;
4804 }
4805
4806 g_queue_push_tail(queue, c2->bs);
4807 g_hash_table_add(found, c2->bs);
d0ac0380
KW
4808 }
4809 }
4810
2f30b7c3
VSO
4811 g_queue_free(queue);
4812 g_hash_table_destroy(found);
4813
4814 return ret;
d0ac0380
KW
4815}
4816
46541ee5
VSO
4817typedef struct BdrvRemoveFilterOrCowChild {
4818 BdrvChild *child;
4819 bool is_backing;
4820} BdrvRemoveFilterOrCowChild;
4821
4822static void bdrv_remove_filter_or_cow_child_abort(void *opaque)
4823{
4824 BdrvRemoveFilterOrCowChild *s = opaque;
4825 BlockDriverState *parent_bs = s->child->opaque;
4826
4827 QLIST_INSERT_HEAD(&parent_bs->children, s->child, next);
4828 if (s->is_backing) {
4829 parent_bs->backing = s->child;
4830 } else {
4831 parent_bs->file = s->child;
4832 }
4833
4834 /*
4bf021db 4835 * We don't have to restore child->bs here to undo bdrv_replace_child_tran()
46541ee5
VSO
4836 * because that function is transactionable and it registered own completion
4837 * entries in @tran, so .abort() for bdrv_replace_child_safe() will be
4838 * called automatically.
4839 */
4840}
4841
4842static void bdrv_remove_filter_or_cow_child_commit(void *opaque)
4843{
4844 BdrvRemoveFilterOrCowChild *s = opaque;
4845
4846 bdrv_child_free(s->child);
4847}
4848
4849static TransactionActionDrv bdrv_remove_filter_or_cow_child_drv = {
4850 .abort = bdrv_remove_filter_or_cow_child_abort,
4851 .commit = bdrv_remove_filter_or_cow_child_commit,
4852 .clean = g_free,
4853};
4854
4855/*
5b995019 4856 * A function to remove backing or file child of @bs.
7ec390d5 4857 * Function doesn't update permissions, caller is responsible for this.
46541ee5 4858 */
5b995019
VSO
4859static void bdrv_remove_file_or_backing_child(BlockDriverState *bs,
4860 BdrvChild *child,
4861 Transaction *tran)
46541ee5
VSO
4862{
4863 BdrvRemoveFilterOrCowChild *s;
5b995019
VSO
4864
4865 assert(child == bs->backing || child == bs->file);
46541ee5
VSO
4866
4867 if (!child) {
4868 return;
4869 }
4870
4871 if (child->bs) {
4bf021db 4872 bdrv_replace_child_tran(child, NULL, tran);
46541ee5
VSO
4873 }
4874
4875 s = g_new(BdrvRemoveFilterOrCowChild, 1);
4876 *s = (BdrvRemoveFilterOrCowChild) {
4877 .child = child,
4878 .is_backing = (child == bs->backing),
4879 };
4880 tran_add(tran, &bdrv_remove_filter_or_cow_child_drv, s);
4881
4882 QLIST_SAFE_REMOVE(child, next);
4883 if (s->is_backing) {
4884 bs->backing = NULL;
4885 } else {
4886 bs->file = NULL;
4887 }
4888}
4889
5b995019
VSO
4890/*
4891 * A function to remove backing-chain child of @bs if exists: cow child for
4892 * format nodes (always .backing) and filter child for filters (may be .file or
4893 * .backing)
4894 */
4895static void bdrv_remove_filter_or_cow_child(BlockDriverState *bs,
4896 Transaction *tran)
4897{
4898 bdrv_remove_file_or_backing_child(bs, bdrv_filter_or_cow_child(bs), tran);
4899}
4900
117caba9
VSO
4901static int bdrv_replace_node_noperm(BlockDriverState *from,
4902 BlockDriverState *to,
4903 bool auto_skip, Transaction *tran,
4904 Error **errp)
4905{
4906 BdrvChild *c, *next;
4907
4908 QLIST_FOREACH_SAFE(c, &from->parents, next_parent, next) {
4909 assert(c->bs == from);
4910 if (!should_update_child(c, to)) {
4911 if (auto_skip) {
4912 continue;
4913 }
4914 error_setg(errp, "Should not change '%s' link to '%s'",
4915 c->name, from->node_name);
4916 return -EINVAL;
4917 }
4918 if (c->frozen) {
4919 error_setg(errp, "Cannot change '%s' link to '%s'",
4920 c->name, from->node_name);
4921 return -EPERM;
4922 }
4bf021db 4923 bdrv_replace_child_tran(c, to, tran);
117caba9
VSO
4924 }
4925
4926 return 0;
4927}
4928
313274bb
VSO
4929/*
4930 * With auto_skip=true bdrv_replace_node_common skips updating from parents
4931 * if it creates a parent-child relation loop or if parent is block-job.
4932 *
4933 * With auto_skip=false the error is returned if from has a parent which should
4934 * not be updated.
3108a15c
VSO
4935 *
4936 * With @detach_subchain=true @to must be in a backing chain of @from. In this
4937 * case backing link of the cow-parent of @to is removed.
313274bb 4938 */
a1e708fc
VSO
4939static int bdrv_replace_node_common(BlockDriverState *from,
4940 BlockDriverState *to,
3108a15c
VSO
4941 bool auto_skip, bool detach_subchain,
4942 Error **errp)
dd62f1ca 4943{
3bb0e298
VSO
4944 Transaction *tran = tran_new();
4945 g_autoptr(GHashTable) found = NULL;
4946 g_autoptr(GSList) refresh_list = NULL;
2d369d6e 4947 BlockDriverState *to_cow_parent = NULL;
234ac1a9
KW
4948 int ret;
4949
3108a15c
VSO
4950 if (detach_subchain) {
4951 assert(bdrv_chain_contains(from, to));
4952 assert(from != to);
4953 for (to_cow_parent = from;
4954 bdrv_filter_or_cow_bs(to_cow_parent) != to;
4955 to_cow_parent = bdrv_filter_or_cow_bs(to_cow_parent))
4956 {
4957 ;
4958 }
4959 }
4960
234ac1a9
KW
4961 /* Make sure that @from doesn't go away until we have successfully attached
4962 * all of its parents to @to. */
4963 bdrv_ref(from);
dd62f1ca 4964
f871abd6 4965 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
30dd65f3 4966 assert(bdrv_get_aio_context(from) == bdrv_get_aio_context(to));
f871abd6
KW
4967 bdrv_drained_begin(from);
4968
3bb0e298
VSO
4969 /*
4970 * Do the replacement without permission update.
4971 * Replacement may influence the permissions, we should calculate new
4972 * permissions based on new graph. If we fail, we'll roll-back the
4973 * replacement.
4974 */
117caba9
VSO
4975 ret = bdrv_replace_node_noperm(from, to, auto_skip, tran, errp);
4976 if (ret < 0) {
4977 goto out;
234ac1a9
KW
4978 }
4979
3108a15c
VSO
4980 if (detach_subchain) {
4981 bdrv_remove_filter_or_cow_child(to_cow_parent, tran);
4982 }
4983
3bb0e298 4984 found = g_hash_table_new(NULL, NULL);
234ac1a9 4985
3bb0e298
VSO
4986 refresh_list = bdrv_topological_dfs(refresh_list, found, to);
4987 refresh_list = bdrv_topological_dfs(refresh_list, found, from);
9bd910e2 4988
3bb0e298
VSO
4989 ret = bdrv_list_refresh_perms(refresh_list, NULL, tran, errp);
4990 if (ret < 0) {
4991 goto out;
dd62f1ca 4992 }
234ac1a9 4993
a1e708fc
VSO
4994 ret = 0;
4995
234ac1a9 4996out:
3bb0e298
VSO
4997 tran_finalize(tran, ret);
4998
f871abd6 4999 bdrv_drained_end(from);
234ac1a9 5000 bdrv_unref(from);
a1e708fc
VSO
5001
5002 return ret;
dd62f1ca
KW
5003}
5004
a1e708fc
VSO
5005int bdrv_replace_node(BlockDriverState *from, BlockDriverState *to,
5006 Error **errp)
313274bb 5007{
3108a15c
VSO
5008 return bdrv_replace_node_common(from, to, true, false, errp);
5009}
5010
5011int bdrv_drop_filter(BlockDriverState *bs, Error **errp)
5012{
5013 return bdrv_replace_node_common(bs, bdrv_filter_or_cow_bs(bs), true, true,
5014 errp);
313274bb
VSO
5015}
5016
4ddc07ca
PB
5017/*
5018 * Add new bs contents at the top of an image chain while the chain is
5019 * live, while keeping required fields on the top layer.
5020 *
5021 * This will modify the BlockDriverState fields, and swap contents
5022 * between bs_new and bs_top. Both bs_new and bs_top are modified.
5023 *
2272edcf
VSO
5024 * bs_new must not be attached to a BlockBackend and must not have backing
5025 * child.
4ddc07ca
PB
5026 *
5027 * This function does not create any image files.
5028 */
a1e708fc
VSO
5029int bdrv_append(BlockDriverState *bs_new, BlockDriverState *bs_top,
5030 Error **errp)
4ddc07ca 5031{
2272edcf
VSO
5032 int ret;
5033 Transaction *tran = tran_new();
5034
5035 assert(!bs_new->backing);
5036
5037 ret = bdrv_attach_child_noperm(bs_new, bs_top, "backing",
5038 &child_of_bds, bdrv_backing_role(bs_new),
5039 &bs_new->backing, tran, errp);
a1e708fc 5040 if (ret < 0) {
2272edcf 5041 goto out;
b2c2832c 5042 }
dd62f1ca 5043
2272edcf 5044 ret = bdrv_replace_node_noperm(bs_top, bs_new, true, tran, errp);
a1e708fc 5045 if (ret < 0) {
2272edcf 5046 goto out;
234ac1a9 5047 }
4ddc07ca 5048
2272edcf
VSO
5049 ret = bdrv_refresh_perms(bs_new, errp);
5050out:
5051 tran_finalize(tran, ret);
5052
1e4c797c 5053 bdrv_refresh_limits(bs_top, NULL, NULL);
2272edcf
VSO
5054
5055 return ret;
8802d1fd
JC
5056}
5057
bd8f4c42
VSO
5058/* Not for empty child */
5059int bdrv_replace_child_bs(BdrvChild *child, BlockDriverState *new_bs,
5060 Error **errp)
5061{
5062 int ret;
5063 Transaction *tran = tran_new();
5064 g_autoptr(GHashTable) found = NULL;
5065 g_autoptr(GSList) refresh_list = NULL;
5066 BlockDriverState *old_bs = child->bs;
5067
5068 bdrv_ref(old_bs);
5069 bdrv_drained_begin(old_bs);
5070 bdrv_drained_begin(new_bs);
5071
5072 bdrv_replace_child_tran(child, new_bs, tran);
5073
5074 found = g_hash_table_new(NULL, NULL);
5075 refresh_list = bdrv_topological_dfs(refresh_list, found, old_bs);
5076 refresh_list = bdrv_topological_dfs(refresh_list, found, new_bs);
5077
5078 ret = bdrv_list_refresh_perms(refresh_list, NULL, tran, errp);
5079
5080 tran_finalize(tran, ret);
5081
5082 bdrv_drained_end(old_bs);
5083 bdrv_drained_end(new_bs);
5084 bdrv_unref(old_bs);
5085
5086 return ret;
5087}
5088
4f6fd349 5089static void bdrv_delete(BlockDriverState *bs)
b338082b 5090{
3718d8ab 5091 assert(bdrv_op_blocker_is_empty(bs));
4f6fd349 5092 assert(!bs->refcnt);
18846dee 5093
1b7bdbc1 5094 /* remove from list, if necessary */
63eaaae0
KW
5095 if (bs->node_name[0] != '\0') {
5096 QTAILQ_REMOVE(&graph_bdrv_states, bs, node_list);
5097 }
2c1d04e0
HR
5098 QTAILQ_REMOVE(&all_bdrv_states, bs, bs_list);
5099
30c321f9
AK
5100 bdrv_close(bs);
5101
7267c094 5102 g_free(bs);
fc01f7e7
FB
5103}
5104
8872ef78
AS
5105BlockDriverState *bdrv_insert_node(BlockDriverState *bs, QDict *node_options,
5106 int flags, Error **errp)
5107{
5108 BlockDriverState *new_node_bs;
5109 Error *local_err = NULL;
5110
5111 new_node_bs = bdrv_open(NULL, NULL, node_options, flags, errp);
5112 if (new_node_bs == NULL) {
5113 error_prepend(errp, "Could not create node: ");
5114 return NULL;
5115 }
5116
5117 bdrv_drained_begin(bs);
5118 bdrv_replace_node(bs, new_node_bs, &local_err);
5119 bdrv_drained_end(bs);
5120
5121 if (local_err) {
5122 bdrv_unref(new_node_bs);
5123 error_propagate(errp, local_err);
5124 return NULL;
5125 }
5126
5127 return new_node_bs;
5128}
5129
e97fc193
AL
5130/*
5131 * Run consistency checks on an image
5132 *
e076f338 5133 * Returns 0 if the check could be completed (it doesn't mean that the image is
a1c7273b 5134 * free of errors) or -errno when an internal error occurred. The results of the
e076f338 5135 * check are stored in res.
e97fc193 5136 */
21c2283e
VSO
5137int coroutine_fn bdrv_co_check(BlockDriverState *bs,
5138 BdrvCheckResult *res, BdrvCheckMode fix)
e97fc193 5139{
908bcd54
HR
5140 if (bs->drv == NULL) {
5141 return -ENOMEDIUM;
5142 }
2fd61638 5143 if (bs->drv->bdrv_co_check == NULL) {
e97fc193
AL
5144 return -ENOTSUP;
5145 }
5146
e076f338 5147 memset(res, 0, sizeof(*res));
2fd61638
PB
5148 return bs->drv->bdrv_co_check(bs, res, fix);
5149}
5150
756e6736
KW
5151/*
5152 * Return values:
5153 * 0 - success
5154 * -EINVAL - backing format specified, but no file
5155 * -ENOSPC - can't update the backing file because no space is left in the
5156 * image file header
5157 * -ENOTSUP - format driver doesn't support changing the backing file
5158 */
e54ee1b3 5159int bdrv_change_backing_file(BlockDriverState *bs, const char *backing_file,
497a30db 5160 const char *backing_fmt, bool require)
756e6736
KW
5161{
5162 BlockDriver *drv = bs->drv;
469ef350 5163 int ret;
756e6736 5164
d470ad42
HR
5165 if (!drv) {
5166 return -ENOMEDIUM;
5167 }
5168
5f377794
PB
5169 /* Backing file format doesn't make sense without a backing file */
5170 if (backing_fmt && !backing_file) {
5171 return -EINVAL;
5172 }
5173
497a30db
EB
5174 if (require && backing_file && !backing_fmt) {
5175 return -EINVAL;
e54ee1b3
EB
5176 }
5177
756e6736 5178 if (drv->bdrv_change_backing_file != NULL) {
469ef350 5179 ret = drv->bdrv_change_backing_file(bs, backing_file, backing_fmt);
756e6736 5180 } else {
469ef350 5181 ret = -ENOTSUP;
756e6736 5182 }
469ef350
PB
5183
5184 if (ret == 0) {
5185 pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_file ?: "");
5186 pstrcpy(bs->backing_format, sizeof(bs->backing_format), backing_fmt ?: "");
998c2019
HR
5187 pstrcpy(bs->auto_backing_file, sizeof(bs->auto_backing_file),
5188 backing_file ?: "");
469ef350
PB
5189 }
5190 return ret;
756e6736
KW
5191}
5192
6ebdcee2 5193/*
dcf3f9b2
HR
5194 * Finds the first non-filter node above bs in the chain between
5195 * active and bs. The returned node is either an immediate parent of
5196 * bs, or there are only filter nodes between the two.
6ebdcee2
JC
5197 *
5198 * Returns NULL if bs is not found in active's image chain,
5199 * or if active == bs.
4caf0fcd
JC
5200 *
5201 * Returns the bottommost base image if bs == NULL.
6ebdcee2
JC
5202 */
5203BlockDriverState *bdrv_find_overlay(BlockDriverState *active,
5204 BlockDriverState *bs)
5205{
dcf3f9b2
HR
5206 bs = bdrv_skip_filters(bs);
5207 active = bdrv_skip_filters(active);
5208
5209 while (active) {
5210 BlockDriverState *next = bdrv_backing_chain_next(active);
5211 if (bs == next) {
5212 return active;
5213 }
5214 active = next;
6ebdcee2
JC
5215 }
5216
dcf3f9b2 5217 return NULL;
4caf0fcd 5218}
6ebdcee2 5219
4caf0fcd
JC
5220/* Given a BDS, searches for the base layer. */
5221BlockDriverState *bdrv_find_base(BlockDriverState *bs)
5222{
5223 return bdrv_find_overlay(bs, NULL);
6ebdcee2
JC
5224}
5225
2cad1ebe 5226/*
7b99a266
HR
5227 * Return true if at least one of the COW (backing) and filter links
5228 * between @bs and @base is frozen. @errp is set if that's the case.
0f0998f6 5229 * @base must be reachable from @bs, or NULL.
2cad1ebe
AG
5230 */
5231bool bdrv_is_backing_chain_frozen(BlockDriverState *bs, BlockDriverState *base,
5232 Error **errp)
5233{
5234 BlockDriverState *i;
7b99a266 5235 BdrvChild *child;
2cad1ebe 5236
7b99a266
HR
5237 for (i = bs; i != base; i = child_bs(child)) {
5238 child = bdrv_filter_or_cow_child(i);
5239
5240 if (child && child->frozen) {
2cad1ebe 5241 error_setg(errp, "Cannot change '%s' link from '%s' to '%s'",
7b99a266 5242 child->name, i->node_name, child->bs->node_name);
2cad1ebe
AG
5243 return true;
5244 }
5245 }
5246
5247 return false;
5248}
5249
5250/*
7b99a266 5251 * Freeze all COW (backing) and filter links between @bs and @base.
2cad1ebe
AG
5252 * If any of the links is already frozen the operation is aborted and
5253 * none of the links are modified.
0f0998f6 5254 * @base must be reachable from @bs, or NULL.
2cad1ebe
AG
5255 * Returns 0 on success. On failure returns < 0 and sets @errp.
5256 */
5257int bdrv_freeze_backing_chain(BlockDriverState *bs, BlockDriverState *base,
5258 Error **errp)
5259{
5260 BlockDriverState *i;
7b99a266 5261 BdrvChild *child;
2cad1ebe
AG
5262
5263 if (bdrv_is_backing_chain_frozen(bs, base, errp)) {
5264 return -EPERM;
5265 }
5266
7b99a266
HR
5267 for (i = bs; i != base; i = child_bs(child)) {
5268 child = bdrv_filter_or_cow_child(i);
5269 if (child && child->bs->never_freeze) {
e5182c1c 5270 error_setg(errp, "Cannot freeze '%s' link to '%s'",
7b99a266 5271 child->name, child->bs->node_name);
e5182c1c
HR
5272 return -EPERM;
5273 }
5274 }
5275
7b99a266
HR
5276 for (i = bs; i != base; i = child_bs(child)) {
5277 child = bdrv_filter_or_cow_child(i);
5278 if (child) {
5279 child->frozen = true;
0f0998f6 5280 }
2cad1ebe
AG
5281 }
5282
5283 return 0;
5284}
5285
5286/*
7b99a266
HR
5287 * Unfreeze all COW (backing) and filter links between @bs and @base.
5288 * The caller must ensure that all links are frozen before using this
5289 * function.
0f0998f6 5290 * @base must be reachable from @bs, or NULL.
2cad1ebe
AG
5291 */
5292void bdrv_unfreeze_backing_chain(BlockDriverState *bs, BlockDriverState *base)
5293{
5294 BlockDriverState *i;
7b99a266 5295 BdrvChild *child;
2cad1ebe 5296
7b99a266
HR
5297 for (i = bs; i != base; i = child_bs(child)) {
5298 child = bdrv_filter_or_cow_child(i);
5299 if (child) {
5300 assert(child->frozen);
5301 child->frozen = false;
0f0998f6 5302 }
2cad1ebe
AG
5303 }
5304}
5305
6ebdcee2
JC
5306/*
5307 * Drops images above 'base' up to and including 'top', and sets the image
5308 * above 'top' to have base as its backing file.
5309 *
5310 * Requires that the overlay to 'top' is opened r/w, so that the backing file
5311 * information in 'bs' can be properly updated.
5312 *
5313 * E.g., this will convert the following chain:
5314 * bottom <- base <- intermediate <- top <- active
5315 *
5316 * to
5317 *
5318 * bottom <- base <- active
5319 *
5320 * It is allowed for bottom==base, in which case it converts:
5321 *
5322 * base <- intermediate <- top <- active
5323 *
5324 * to
5325 *
5326 * base <- active
5327 *
54e26900
JC
5328 * If backing_file_str is non-NULL, it will be used when modifying top's
5329 * overlay image metadata.
5330 *
6ebdcee2
JC
5331 * Error conditions:
5332 * if active == top, that is considered an error
5333 *
5334 */
bde70715
KW
5335int bdrv_drop_intermediate(BlockDriverState *top, BlockDriverState *base,
5336 const char *backing_file_str)
6ebdcee2 5337{
6bd858b3
AG
5338 BlockDriverState *explicit_top = top;
5339 bool update_inherits_from;
d669ed6a 5340 BdrvChild *c;
12fa4af6 5341 Error *local_err = NULL;
6ebdcee2 5342 int ret = -EIO;
d669ed6a
VSO
5343 g_autoptr(GSList) updated_children = NULL;
5344 GSList *p;
6ebdcee2 5345
6858eba0 5346 bdrv_ref(top);
637d54a5 5347 bdrv_subtree_drained_begin(top);
6858eba0 5348
6ebdcee2
JC
5349 if (!top->drv || !base->drv) {
5350 goto exit;
5351 }
5352
5db15a57
KW
5353 /* Make sure that base is in the backing chain of top */
5354 if (!bdrv_chain_contains(top, base)) {
6ebdcee2
JC
5355 goto exit;
5356 }
5357
6bd858b3
AG
5358 /* If 'base' recursively inherits from 'top' then we should set
5359 * base->inherits_from to top->inherits_from after 'top' and all
5360 * other intermediate nodes have been dropped.
5361 * If 'top' is an implicit node (e.g. "commit_top") we should skip
5362 * it because no one inherits from it. We use explicit_top for that. */
dcf3f9b2 5363 explicit_top = bdrv_skip_implicit_filters(explicit_top);
6bd858b3
AG
5364 update_inherits_from = bdrv_inherits_from_recursive(base, explicit_top);
5365
6ebdcee2 5366 /* success - we can delete the intermediate states, and link top->base */
bde70715
KW
5367 /* TODO Check graph modification op blockers (BLK_PERM_GRAPH_MOD) once
5368 * we've figured out how they should work. */
f30c66ba
HR
5369 if (!backing_file_str) {
5370 bdrv_refresh_filename(base);
5371 backing_file_str = base->filename;
5372 }
61f09cea 5373
d669ed6a
VSO
5374 QLIST_FOREACH(c, &top->parents, next_parent) {
5375 updated_children = g_slist_prepend(updated_children, c);
5376 }
5377
3108a15c
VSO
5378 /*
5379 * It seems correct to pass detach_subchain=true here, but it triggers
5380 * one more yet not fixed bug, when due to nested aio_poll loop we switch to
5381 * another drained section, which modify the graph (for example, removing
5382 * the child, which we keep in updated_children list). So, it's a TODO.
5383 *
5384 * Note, bug triggered if pass detach_subchain=true here and run
5385 * test-bdrv-drain. test_drop_intermediate_poll() test-case will crash.
5386 * That's a FIXME.
5387 */
5388 bdrv_replace_node_common(top, base, false, false, &local_err);
d669ed6a
VSO
5389 if (local_err) {
5390 error_report_err(local_err);
5391 goto exit;
5392 }
5393
5394 for (p = updated_children; p; p = p->next) {
5395 c = p->data;
12fa4af6 5396
bd86fb99
HR
5397 if (c->klass->update_filename) {
5398 ret = c->klass->update_filename(c, base, backing_file_str,
5399 &local_err);
61f09cea 5400 if (ret < 0) {
d669ed6a
VSO
5401 /*
5402 * TODO: Actually, we want to rollback all previous iterations
5403 * of this loop, and (which is almost impossible) previous
5404 * bdrv_replace_node()...
5405 *
5406 * Note, that c->klass->update_filename may lead to permission
5407 * update, so it's a bad idea to call it inside permission
5408 * update transaction of bdrv_replace_node.
5409 */
61f09cea
KW
5410 error_report_err(local_err);
5411 goto exit;
5412 }
5413 }
12fa4af6 5414 }
6ebdcee2 5415
6bd858b3
AG
5416 if (update_inherits_from) {
5417 base->inherits_from = explicit_top->inherits_from;
5418 }
5419
6ebdcee2 5420 ret = 0;
6ebdcee2 5421exit:
637d54a5 5422 bdrv_subtree_drained_end(top);
6858eba0 5423 bdrv_unref(top);
6ebdcee2
JC
5424 return ret;
5425}
5426
081e4650
HR
5427/**
5428 * Implementation of BlockDriver.bdrv_get_allocated_file_size() that
5429 * sums the size of all data-bearing children. (This excludes backing
5430 * children.)
5431 */
5432static int64_t bdrv_sum_allocated_file_size(BlockDriverState *bs)
5433{
5434 BdrvChild *child;
5435 int64_t child_size, sum = 0;
5436
5437 QLIST_FOREACH(child, &bs->children, next) {
5438 if (child->role & (BDRV_CHILD_DATA | BDRV_CHILD_METADATA |
5439 BDRV_CHILD_FILTERED))
5440 {
5441 child_size = bdrv_get_allocated_file_size(child->bs);
5442 if (child_size < 0) {
5443 return child_size;
5444 }
5445 sum += child_size;
5446 }
5447 }
5448
5449 return sum;
5450}
5451
61007b31
SH
5452/**
5453 * Length of a allocated file in bytes. Sparse files are counted by actual
5454 * allocated space. Return < 0 if error or unknown.
5455 */
5456int64_t bdrv_get_allocated_file_size(BlockDriverState *bs)
71d0770c 5457{
61007b31
SH
5458 BlockDriver *drv = bs->drv;
5459 if (!drv) {
5460 return -ENOMEDIUM;
8f4754ed 5461 }
61007b31
SH
5462 if (drv->bdrv_get_allocated_file_size) {
5463 return drv->bdrv_get_allocated_file_size(bs);
5464 }
081e4650
HR
5465
5466 if (drv->bdrv_file_open) {
5467 /*
5468 * Protocol drivers default to -ENOTSUP (most of their data is
5469 * not stored in any of their children (if they even have any),
5470 * so there is no generic way to figure it out).
5471 */
5472 return -ENOTSUP;
5473 } else if (drv->is_filter) {
5474 /* Filter drivers default to the size of their filtered child */
5475 return bdrv_get_allocated_file_size(bdrv_filter_bs(bs));
5476 } else {
5477 /* Other drivers default to summing their children's sizes */
5478 return bdrv_sum_allocated_file_size(bs);
1c9805a3
SH
5479 }
5480}
e7a8a783 5481
90880ff1
SH
5482/*
5483 * bdrv_measure:
5484 * @drv: Format driver
5485 * @opts: Creation options for new image
5486 * @in_bs: Existing image containing data for new image (may be NULL)
5487 * @errp: Error object
5488 * Returns: A #BlockMeasureInfo (free using qapi_free_BlockMeasureInfo())
5489 * or NULL on error
5490 *
5491 * Calculate file size required to create a new image.
5492 *
5493 * If @in_bs is given then space for allocated clusters and zero clusters
5494 * from that image are included in the calculation. If @opts contains a
5495 * backing file that is shared by @in_bs then backing clusters may be omitted
5496 * from the calculation.
5497 *
5498 * If @in_bs is NULL then the calculation includes no allocated clusters
5499 * unless a preallocation option is given in @opts.
5500 *
5501 * Note that @in_bs may use a different BlockDriver from @drv.
5502 *
5503 * If an error occurs the @errp pointer is set.
5504 */
5505BlockMeasureInfo *bdrv_measure(BlockDriver *drv, QemuOpts *opts,
5506 BlockDriverState *in_bs, Error **errp)
5507{
5508 if (!drv->bdrv_measure) {
5509 error_setg(errp, "Block driver '%s' does not support size measurement",
5510 drv->format_name);
5511 return NULL;
5512 }
5513
5514 return drv->bdrv_measure(opts, in_bs, errp);
5515}
5516
61007b31
SH
5517/**
5518 * Return number of sectors on success, -errno on error.
1c9805a3 5519 */
61007b31 5520int64_t bdrv_nb_sectors(BlockDriverState *bs)
1c9805a3 5521{
61007b31 5522 BlockDriver *drv = bs->drv;
498e386c 5523
61007b31
SH
5524 if (!drv)
5525 return -ENOMEDIUM;
2572b37a 5526
61007b31
SH
5527 if (drv->has_variable_length) {
5528 int ret = refresh_total_sectors(bs, bs->total_sectors);
5529 if (ret < 0) {
5530 return ret;
1c9805a3
SH
5531 }
5532 }
61007b31 5533 return bs->total_sectors;
1c9805a3 5534}
b338082b 5535
61007b31
SH
5536/**
5537 * Return length in bytes on success, -errno on error.
5538 * The length is always a multiple of BDRV_SECTOR_SIZE.
8d3b1a2d 5539 */
61007b31 5540int64_t bdrv_getlength(BlockDriverState *bs)
8d3b1a2d 5541{
61007b31 5542 int64_t ret = bdrv_nb_sectors(bs);
8d3b1a2d 5543
122860ba
EB
5544 if (ret < 0) {
5545 return ret;
5546 }
5547 if (ret > INT64_MAX / BDRV_SECTOR_SIZE) {
5548 return -EFBIG;
5549 }
5550 return ret * BDRV_SECTOR_SIZE;
fc01f7e7
FB
5551}
5552
61007b31
SH
5553/* return 0 as number of sectors if no device present or error */
5554void bdrv_get_geometry(BlockDriverState *bs, uint64_t *nb_sectors_ptr)
07d27a44 5555{
61007b31 5556 int64_t nb_sectors = bdrv_nb_sectors(bs);
07d27a44 5557
61007b31 5558 *nb_sectors_ptr = nb_sectors < 0 ? 0 : nb_sectors;
07d27a44
MA
5559}
5560
54115412 5561bool bdrv_is_sg(BlockDriverState *bs)
f08145fe 5562{
61007b31 5563 return bs->sg;
f08145fe
KW
5564}
5565
ae23f786
HR
5566/**
5567 * Return whether the given node supports compressed writes.
5568 */
5569bool bdrv_supports_compressed_writes(BlockDriverState *bs)
5570{
5571 BlockDriverState *filtered;
5572
5573 if (!bs->drv || !block_driver_can_compress(bs->drv)) {
5574 return false;
5575 }
5576
5577 filtered = bdrv_filter_bs(bs);
5578 if (filtered) {
5579 /*
5580 * Filters can only forward compressed writes, so we have to
5581 * check the child.
5582 */
5583 return bdrv_supports_compressed_writes(filtered);
5584 }
5585
5586 return true;
5587}
5588
61007b31 5589const char *bdrv_get_format_name(BlockDriverState *bs)
40b4f539 5590{
61007b31 5591 return bs->drv ? bs->drv->format_name : NULL;
40b4f539
KW
5592}
5593
61007b31 5594static int qsort_strcmp(const void *a, const void *b)
40b4f539 5595{
ceff5bd7 5596 return strcmp(*(char *const *)a, *(char *const *)b);
40b4f539
KW
5597}
5598
61007b31 5599void bdrv_iterate_format(void (*it)(void *opaque, const char *name),
9ac404c5 5600 void *opaque, bool read_only)
40b4f539 5601{
61007b31
SH
5602 BlockDriver *drv;
5603 int count = 0;
5604 int i;
5605 const char **formats = NULL;
40b4f539 5606
61007b31
SH
5607 QLIST_FOREACH(drv, &bdrv_drivers, list) {
5608 if (drv->format_name) {
5609 bool found = false;
5610 int i = count;
9ac404c5
AS
5611
5612 if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv, read_only)) {
5613 continue;
5614 }
5615
61007b31
SH
5616 while (formats && i && !found) {
5617 found = !strcmp(formats[--i], drv->format_name);
5618 }
e2a305fb 5619
61007b31
SH
5620 if (!found) {
5621 formats = g_renew(const char *, formats, count + 1);
5622 formats[count++] = drv->format_name;
5623 }
6c5a42ac 5624 }
61007b31 5625 }
6c5a42ac 5626
eb0df69f
HR
5627 for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); i++) {
5628 const char *format_name = block_driver_modules[i].format_name;
5629
5630 if (format_name) {
5631 bool found = false;
5632 int j = count;
5633
9ac404c5
AS
5634 if (use_bdrv_whitelist &&
5635 !bdrv_format_is_whitelisted(format_name, read_only)) {
5636 continue;
5637 }
5638
eb0df69f
HR
5639 while (formats && j && !found) {
5640 found = !strcmp(formats[--j], format_name);
5641 }
5642
5643 if (!found) {
5644 formats = g_renew(const char *, formats, count + 1);
5645 formats[count++] = format_name;
5646 }
5647 }
5648 }
5649
61007b31 5650 qsort(formats, count, sizeof(formats[0]), qsort_strcmp);
40b4f539 5651
61007b31
SH
5652 for (i = 0; i < count; i++) {
5653 it(opaque, formats[i]);
5654 }
40b4f539 5655
61007b31
SH
5656 g_free(formats);
5657}
40b4f539 5658
61007b31
SH
5659/* This function is to find a node in the bs graph */
5660BlockDriverState *bdrv_find_node(const char *node_name)
5661{
5662 BlockDriverState *bs;
391827eb 5663
61007b31 5664 assert(node_name);
40b4f539 5665
61007b31
SH
5666 QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
5667 if (!strcmp(node_name, bs->node_name)) {
5668 return bs;
40b4f539
KW
5669 }
5670 }
61007b31 5671 return NULL;
40b4f539
KW
5672}
5673
61007b31 5674/* Put this QMP function here so it can access the static graph_bdrv_states. */
facda544
PK
5675BlockDeviceInfoList *bdrv_named_nodes_list(bool flat,
5676 Error **errp)
40b4f539 5677{
9812e712 5678 BlockDeviceInfoList *list;
61007b31 5679 BlockDriverState *bs;
40b4f539 5680
61007b31
SH
5681 list = NULL;
5682 QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
facda544 5683 BlockDeviceInfo *info = bdrv_block_device_info(NULL, bs, flat, errp);
61007b31
SH
5684 if (!info) {
5685 qapi_free_BlockDeviceInfoList(list);
5686 return NULL;
301db7c2 5687 }
9812e712 5688 QAPI_LIST_PREPEND(list, info);
301db7c2
RH
5689 }
5690
61007b31
SH
5691 return list;
5692}
40b4f539 5693
5d3b4e99
VSO
5694typedef struct XDbgBlockGraphConstructor {
5695 XDbgBlockGraph *graph;
5696 GHashTable *graph_nodes;
5697} XDbgBlockGraphConstructor;
5698
5699static XDbgBlockGraphConstructor *xdbg_graph_new(void)
5700{
5701 XDbgBlockGraphConstructor *gr = g_new(XDbgBlockGraphConstructor, 1);
5702
5703 gr->graph = g_new0(XDbgBlockGraph, 1);
5704 gr->graph_nodes = g_hash_table_new(NULL, NULL);
5705
5706 return gr;
5707}
5708
5709static XDbgBlockGraph *xdbg_graph_finalize(XDbgBlockGraphConstructor *gr)
5710{
5711 XDbgBlockGraph *graph = gr->graph;
5712
5713 g_hash_table_destroy(gr->graph_nodes);
5714 g_free(gr);
5715
5716 return graph;
5717}
5718
5719static uintptr_t xdbg_graph_node_num(XDbgBlockGraphConstructor *gr, void *node)
5720{
5721 uintptr_t ret = (uintptr_t)g_hash_table_lookup(gr->graph_nodes, node);
5722
5723 if (ret != 0) {
5724 return ret;
5725 }
5726
5727 /*
5728 * Start counting from 1, not 0, because 0 interferes with not-found (NULL)
5729 * answer of g_hash_table_lookup.
5730 */
5731 ret = g_hash_table_size(gr->graph_nodes) + 1;
5732 g_hash_table_insert(gr->graph_nodes, node, (void *)ret);
5733
5734 return ret;
5735}
5736
5737static void xdbg_graph_add_node(XDbgBlockGraphConstructor *gr, void *node,
5738 XDbgBlockGraphNodeType type, const char *name)
5739{
5740 XDbgBlockGraphNode *n;
5741
5742 n = g_new0(XDbgBlockGraphNode, 1);
5743
5744 n->id = xdbg_graph_node_num(gr, node);
5745 n->type = type;
5746 n->name = g_strdup(name);
5747
9812e712 5748 QAPI_LIST_PREPEND(gr->graph->nodes, n);
5d3b4e99
VSO
5749}
5750
5751static void xdbg_graph_add_edge(XDbgBlockGraphConstructor *gr, void *parent,
5752 const BdrvChild *child)
5753{
cdb1cec8 5754 BlockPermission qapi_perm;
5d3b4e99
VSO
5755 XDbgBlockGraphEdge *edge;
5756
5d3b4e99
VSO
5757 edge = g_new0(XDbgBlockGraphEdge, 1);
5758
5759 edge->parent = xdbg_graph_node_num(gr, parent);
5760 edge->child = xdbg_graph_node_num(gr, child->bs);
5761 edge->name = g_strdup(child->name);
5762
cdb1cec8
HR
5763 for (qapi_perm = 0; qapi_perm < BLOCK_PERMISSION__MAX; qapi_perm++) {
5764 uint64_t flag = bdrv_qapi_perm_to_blk_perm(qapi_perm);
5765
5766 if (flag & child->perm) {
9812e712 5767 QAPI_LIST_PREPEND(edge->perm, qapi_perm);
5d3b4e99 5768 }
cdb1cec8 5769 if (flag & child->shared_perm) {
9812e712 5770 QAPI_LIST_PREPEND(edge->shared_perm, qapi_perm);
5d3b4e99
VSO
5771 }
5772 }
5773
9812e712 5774 QAPI_LIST_PREPEND(gr->graph->edges, edge);
5d3b4e99
VSO
5775}
5776
5777
5778XDbgBlockGraph *bdrv_get_xdbg_block_graph(Error **errp)
5779{
5780 BlockBackend *blk;
5781 BlockJob *job;
5782 BlockDriverState *bs;
5783 BdrvChild *child;
5784 XDbgBlockGraphConstructor *gr = xdbg_graph_new();
5785
5786 for (blk = blk_all_next(NULL); blk; blk = blk_all_next(blk)) {
5787 char *allocated_name = NULL;
5788 const char *name = blk_name(blk);
5789
5790 if (!*name) {
5791 name = allocated_name = blk_get_attached_dev_id(blk);
5792 }
5793 xdbg_graph_add_node(gr, blk, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_BACKEND,
5794 name);
5795 g_free(allocated_name);
5796 if (blk_root(blk)) {
5797 xdbg_graph_add_edge(gr, blk, blk_root(blk));
5798 }
5799 }
5800
5801 for (job = block_job_next(NULL); job; job = block_job_next(job)) {
5802 GSList *el;
5803
5804 xdbg_graph_add_node(gr, job, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_JOB,
5805 job->job.id);
5806 for (el = job->nodes; el; el = el->next) {
5807 xdbg_graph_add_edge(gr, job, (BdrvChild *)el->data);
5808 }
5809 }
5810
5811 QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
5812 xdbg_graph_add_node(gr, bs, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_DRIVER,
5813 bs->node_name);
5814 QLIST_FOREACH(child, &bs->children, next) {
5815 xdbg_graph_add_edge(gr, bs, child);
5816 }
5817 }
5818
5819 return xdbg_graph_finalize(gr);
5820}
5821
61007b31
SH
5822BlockDriverState *bdrv_lookup_bs(const char *device,
5823 const char *node_name,
5824 Error **errp)
5825{
5826 BlockBackend *blk;
5827 BlockDriverState *bs;
40b4f539 5828
61007b31
SH
5829 if (device) {
5830 blk = blk_by_name(device);
40b4f539 5831
61007b31 5832 if (blk) {
9f4ed6fb
AG
5833 bs = blk_bs(blk);
5834 if (!bs) {
5433c24f 5835 error_setg(errp, "Device '%s' has no medium", device);
5433c24f
HR
5836 }
5837
9f4ed6fb 5838 return bs;
61007b31
SH
5839 }
5840 }
40b4f539 5841
61007b31
SH
5842 if (node_name) {
5843 bs = bdrv_find_node(node_name);
6d519a5f 5844
61007b31
SH
5845 if (bs) {
5846 return bs;
5847 }
40b4f539
KW
5848 }
5849
785ec4b1 5850 error_setg(errp, "Cannot find device=\'%s\' nor node-name=\'%s\'",
61007b31
SH
5851 device ? device : "",
5852 node_name ? node_name : "");
5853 return NULL;
40b4f539
KW
5854}
5855
61007b31
SH
5856/* If 'base' is in the same chain as 'top', return true. Otherwise,
5857 * return false. If either argument is NULL, return false. */
5858bool bdrv_chain_contains(BlockDriverState *top, BlockDriverState *base)
83f64091 5859{
61007b31 5860 while (top && top != base) {
dcf3f9b2 5861 top = bdrv_filter_or_cow_bs(top);
02c50efe 5862 }
61007b31
SH
5863
5864 return top != NULL;
02c50efe
FZ
5865}
5866
61007b31 5867BlockDriverState *bdrv_next_node(BlockDriverState *bs)
02c50efe 5868{
61007b31
SH
5869 if (!bs) {
5870 return QTAILQ_FIRST(&graph_bdrv_states);
02c50efe 5871 }
61007b31 5872 return QTAILQ_NEXT(bs, node_list);
83f64091
FB
5873}
5874
0f12264e
KW
5875BlockDriverState *bdrv_next_all_states(BlockDriverState *bs)
5876{
5877 if (!bs) {
5878 return QTAILQ_FIRST(&all_bdrv_states);
5879 }
5880 return QTAILQ_NEXT(bs, bs_list);
5881}
5882
61007b31 5883const char *bdrv_get_node_name(const BlockDriverState *bs)
83f64091 5884{
61007b31 5885 return bs->node_name;
beac80cd
FB
5886}
5887
1f0c461b 5888const char *bdrv_get_parent_name(const BlockDriverState *bs)
4c265bf9
KW
5889{
5890 BdrvChild *c;
5891 const char *name;
5892
5893 /* If multiple parents have a name, just pick the first one. */
5894 QLIST_FOREACH(c, &bs->parents, next_parent) {
bd86fb99
HR
5895 if (c->klass->get_name) {
5896 name = c->klass->get_name(c);
4c265bf9
KW
5897 if (name && *name) {
5898 return name;
5899 }
5900 }
5901 }
5902
5903 return NULL;
5904}
5905
61007b31
SH
5906/* TODO check what callers really want: bs->node_name or blk_name() */
5907const char *bdrv_get_device_name(const BlockDriverState *bs)
beac80cd 5908{
4c265bf9 5909 return bdrv_get_parent_name(bs) ?: "";
f141eafe 5910}
83f64091 5911
61007b31
SH
5912/* This can be used to identify nodes that might not have a device
5913 * name associated. Since node and device names live in the same
5914 * namespace, the result is unambiguous. The exception is if both are
5915 * absent, then this returns an empty (non-null) string. */
5916const char *bdrv_get_device_or_node_name(const BlockDriverState *bs)
f141eafe 5917{
4c265bf9 5918 return bdrv_get_parent_name(bs) ?: bs->node_name;
beac80cd 5919}
beac80cd 5920
61007b31 5921int bdrv_get_flags(BlockDriverState *bs)
0b5a2445 5922{
61007b31 5923 return bs->open_flags;
0b5a2445
PB
5924}
5925
61007b31 5926int bdrv_has_zero_init_1(BlockDriverState *bs)
68485420 5927{
61007b31 5928 return 1;
0b5a2445
PB
5929}
5930
61007b31 5931int bdrv_has_zero_init(BlockDriverState *bs)
0b5a2445 5932{
93393e69
HR
5933 BlockDriverState *filtered;
5934
d470ad42
HR
5935 if (!bs->drv) {
5936 return 0;
5937 }
0b5a2445 5938
61007b31
SH
5939 /* If BS is a copy on write image, it is initialized to
5940 the contents of the base image, which may not be zeroes. */
34778172 5941 if (bdrv_cow_child(bs)) {
61007b31
SH
5942 return 0;
5943 }
5944 if (bs->drv->bdrv_has_zero_init) {
5945 return bs->drv->bdrv_has_zero_init(bs);
0b5a2445 5946 }
93393e69
HR
5947
5948 filtered = bdrv_filter_bs(bs);
5949 if (filtered) {
5950 return bdrv_has_zero_init(filtered);
5a612c00 5951 }
61007b31
SH
5952
5953 /* safe default */
5954 return 0;
68485420
KW
5955}
5956
61007b31 5957bool bdrv_can_write_zeroes_with_unmap(BlockDriverState *bs)
68485420 5958{
2f0342ef 5959 if (!(bs->open_flags & BDRV_O_UNMAP)) {
61007b31
SH
5960 return false;
5961 }
68485420 5962
e24d813b 5963 return bs->supported_zero_flags & BDRV_REQ_MAY_UNMAP;
68485420
KW
5964}
5965
61007b31
SH
5966void bdrv_get_backing_filename(BlockDriverState *bs,
5967 char *filename, int filename_size)
016f5cf6 5968{
61007b31
SH
5969 pstrcpy(filename, filename_size, bs->backing_file);
5970}
d318aea9 5971
61007b31
SH
5972int bdrv_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
5973{
8b117001 5974 int ret;
61007b31 5975 BlockDriver *drv = bs->drv;
5a612c00
MP
5976 /* if bs->drv == NULL, bs is closed, so there's nothing to do here */
5977 if (!drv) {
61007b31 5978 return -ENOMEDIUM;
5a612c00
MP
5979 }
5980 if (!drv->bdrv_get_info) {
93393e69
HR
5981 BlockDriverState *filtered = bdrv_filter_bs(bs);
5982 if (filtered) {
5983 return bdrv_get_info(filtered, bdi);
5a612c00 5984 }
61007b31 5985 return -ENOTSUP;
5a612c00 5986 }
61007b31 5987 memset(bdi, 0, sizeof(*bdi));
8b117001
VSO
5988 ret = drv->bdrv_get_info(bs, bdi);
5989 if (ret < 0) {
5990 return ret;
5991 }
5992
5993 if (bdi->cluster_size > BDRV_MAX_ALIGNMENT) {
5994 return -EINVAL;
5995 }
5996
5997 return 0;
61007b31 5998}
016f5cf6 5999
1bf6e9ca
AS
6000ImageInfoSpecific *bdrv_get_specific_info(BlockDriverState *bs,
6001 Error **errp)
61007b31
SH
6002{
6003 BlockDriver *drv = bs->drv;
6004 if (drv && drv->bdrv_get_specific_info) {
1bf6e9ca 6005 return drv->bdrv_get_specific_info(bs, errp);
61007b31
SH
6006 }
6007 return NULL;
016f5cf6
AG
6008}
6009
d9245599
AN
6010BlockStatsSpecific *bdrv_get_specific_stats(BlockDriverState *bs)
6011{
6012 BlockDriver *drv = bs->drv;
6013 if (!drv || !drv->bdrv_get_specific_stats) {
6014 return NULL;
6015 }
6016 return drv->bdrv_get_specific_stats(bs);
6017}
6018
a31939e6 6019void bdrv_debug_event(BlockDriverState *bs, BlkdebugEvent event)
4265d620 6020{
61007b31
SH
6021 if (!bs || !bs->drv || !bs->drv->bdrv_debug_event) {
6022 return;
6023 }
4265d620 6024
61007b31 6025 bs->drv->bdrv_debug_event(bs, event);
4265d620
PB
6026}
6027
d10529a2 6028static BlockDriverState *bdrv_find_debug_node(BlockDriverState *bs)
4265d620 6029{
61007b31 6030 while (bs && bs->drv && !bs->drv->bdrv_debug_breakpoint) {
f706a92f 6031 bs = bdrv_primary_bs(bs);
61007b31 6032 }
4265d620 6033
61007b31 6034 if (bs && bs->drv && bs->drv->bdrv_debug_breakpoint) {
d10529a2
VSO
6035 assert(bs->drv->bdrv_debug_remove_breakpoint);
6036 return bs;
6037 }
6038
6039 return NULL;
6040}
6041
6042int bdrv_debug_breakpoint(BlockDriverState *bs, const char *event,
6043 const char *tag)
6044{
6045 bs = bdrv_find_debug_node(bs);
6046 if (bs) {
61007b31
SH
6047 return bs->drv->bdrv_debug_breakpoint(bs, event, tag);
6048 }
4265d620 6049
61007b31 6050 return -ENOTSUP;
4265d620
PB
6051}
6052
61007b31 6053int bdrv_debug_remove_breakpoint(BlockDriverState *bs, const char *tag)
ea2384d3 6054{
d10529a2
VSO
6055 bs = bdrv_find_debug_node(bs);
6056 if (bs) {
61007b31
SH
6057 return bs->drv->bdrv_debug_remove_breakpoint(bs, tag);
6058 }
6059
6060 return -ENOTSUP;
eb852011
MA
6061}
6062
61007b31 6063int bdrv_debug_resume(BlockDriverState *bs, const char *tag)
ce1a14dc 6064{
61007b31 6065 while (bs && (!bs->drv || !bs->drv->bdrv_debug_resume)) {
f706a92f 6066 bs = bdrv_primary_bs(bs);
61007b31 6067 }
ce1a14dc 6068
61007b31
SH
6069 if (bs && bs->drv && bs->drv->bdrv_debug_resume) {
6070 return bs->drv->bdrv_debug_resume(bs, tag);
6071 }
ce1a14dc 6072
61007b31 6073 return -ENOTSUP;
f197fe2b
FZ
6074}
6075
61007b31 6076bool bdrv_debug_is_suspended(BlockDriverState *bs, const char *tag)
ce1a14dc 6077{
61007b31 6078 while (bs && bs->drv && !bs->drv->bdrv_debug_is_suspended) {
f706a92f 6079 bs = bdrv_primary_bs(bs);
f197fe2b 6080 }
19cb3738 6081
61007b31
SH
6082 if (bs && bs->drv && bs->drv->bdrv_debug_is_suspended) {
6083 return bs->drv->bdrv_debug_is_suspended(bs, tag);
6084 }
f9f05dc5 6085
61007b31
SH
6086 return false;
6087}
f9f05dc5 6088
61007b31
SH
6089/* backing_file can either be relative, or absolute, or a protocol. If it is
6090 * relative, it must be relative to the chain. So, passing in bs->filename
6091 * from a BDS as backing_file should not be done, as that may be relative to
6092 * the CWD rather than the chain. */
6093BlockDriverState *bdrv_find_backing_image(BlockDriverState *bs,
6094 const char *backing_file)
f9f05dc5 6095{
61007b31
SH
6096 char *filename_full = NULL;
6097 char *backing_file_full = NULL;
6098 char *filename_tmp = NULL;
6099 int is_protocol = 0;
0b877d09 6100 bool filenames_refreshed = false;
61007b31
SH
6101 BlockDriverState *curr_bs = NULL;
6102 BlockDriverState *retval = NULL;
dcf3f9b2 6103 BlockDriverState *bs_below;
f9f05dc5 6104
61007b31
SH
6105 if (!bs || !bs->drv || !backing_file) {
6106 return NULL;
f9f05dc5
KW
6107 }
6108
61007b31
SH
6109 filename_full = g_malloc(PATH_MAX);
6110 backing_file_full = g_malloc(PATH_MAX);
f9f05dc5 6111
61007b31 6112 is_protocol = path_has_protocol(backing_file);
f9f05dc5 6113
dcf3f9b2
HR
6114 /*
6115 * Being largely a legacy function, skip any filters here
6116 * (because filters do not have normal filenames, so they cannot
6117 * match anyway; and allowing json:{} filenames is a bit out of
6118 * scope).
6119 */
6120 for (curr_bs = bdrv_skip_filters(bs);
6121 bdrv_cow_child(curr_bs) != NULL;
6122 curr_bs = bs_below)
6123 {
6124 bs_below = bdrv_backing_chain_next(curr_bs);
f9f05dc5 6125
0b877d09
HR
6126 if (bdrv_backing_overridden(curr_bs)) {
6127 /*
6128 * If the backing file was overridden, we can only compare
6129 * directly against the backing node's filename.
6130 */
6131
6132 if (!filenames_refreshed) {
6133 /*
6134 * This will automatically refresh all of the
6135 * filenames in the rest of the backing chain, so we
6136 * only need to do this once.
6137 */
6138 bdrv_refresh_filename(bs_below);
6139 filenames_refreshed = true;
6140 }
6141
6142 if (strcmp(backing_file, bs_below->filename) == 0) {
6143 retval = bs_below;
6144 break;
6145 }
6146 } else if (is_protocol || path_has_protocol(curr_bs->backing_file)) {
6147 /*
6148 * If either of the filename paths is actually a protocol, then
6149 * compare unmodified paths; otherwise make paths relative.
6150 */
6b6833c1
HR
6151 char *backing_file_full_ret;
6152
61007b31 6153 if (strcmp(backing_file, curr_bs->backing_file) == 0) {
dcf3f9b2 6154 retval = bs_below;
61007b31
SH
6155 break;
6156 }
418661e0 6157 /* Also check against the full backing filename for the image */
6b6833c1
HR
6158 backing_file_full_ret = bdrv_get_full_backing_filename(curr_bs,
6159 NULL);
6160 if (backing_file_full_ret) {
6161 bool equal = strcmp(backing_file, backing_file_full_ret) == 0;
6162 g_free(backing_file_full_ret);
6163 if (equal) {
dcf3f9b2 6164 retval = bs_below;
418661e0
JC
6165 break;
6166 }
418661e0 6167 }
61007b31
SH
6168 } else {
6169 /* If not an absolute filename path, make it relative to the current
6170 * image's filename path */
2d9158ce
HR
6171 filename_tmp = bdrv_make_absolute_filename(curr_bs, backing_file,
6172 NULL);
6173 /* We are going to compare canonicalized absolute pathnames */
6174 if (!filename_tmp || !realpath(filename_tmp, filename_full)) {
6175 g_free(filename_tmp);
61007b31
SH
6176 continue;
6177 }
2d9158ce 6178 g_free(filename_tmp);
07f07615 6179
61007b31
SH
6180 /* We need to make sure the backing filename we are comparing against
6181 * is relative to the current image filename (or absolute) */
2d9158ce
HR
6182 filename_tmp = bdrv_get_full_backing_filename(curr_bs, NULL);
6183 if (!filename_tmp || !realpath(filename_tmp, backing_file_full)) {
6184 g_free(filename_tmp);
61007b31
SH
6185 continue;
6186 }
2d9158ce 6187 g_free(filename_tmp);
eb489bb1 6188
61007b31 6189 if (strcmp(backing_file_full, filename_full) == 0) {
dcf3f9b2 6190 retval = bs_below;
61007b31
SH
6191 break;
6192 }
6193 }
eb489bb1
KW
6194 }
6195
61007b31
SH
6196 g_free(filename_full);
6197 g_free(backing_file_full);
61007b31
SH
6198 return retval;
6199}
6200
61007b31
SH
6201void bdrv_init(void)
6202{
e5f05f8c
KW
6203#ifdef CONFIG_BDRV_WHITELIST_TOOLS
6204 use_bdrv_whitelist = 1;
6205#endif
61007b31
SH
6206 module_call_init(MODULE_INIT_BLOCK);
6207}
29cdb251 6208
61007b31
SH
6209void bdrv_init_with_whitelist(void)
6210{
6211 use_bdrv_whitelist = 1;
6212 bdrv_init();
07f07615
PB
6213}
6214
21c2283e 6215int coroutine_fn bdrv_co_invalidate_cache(BlockDriverState *bs, Error **errp)
0f15423c 6216{
4417ab7a 6217 BdrvChild *child, *parent;
5a8a30db
KW
6218 Error *local_err = NULL;
6219 int ret;
9c98f145 6220 BdrvDirtyBitmap *bm;
5a8a30db 6221
3456a8d1 6222 if (!bs->drv) {
5416645f 6223 return -ENOMEDIUM;
3456a8d1
KW
6224 }
6225
16e977d5 6226 QLIST_FOREACH(child, &bs->children, next) {
2b148f39 6227 bdrv_co_invalidate_cache(child->bs, &local_err);
0d1c5c91 6228 if (local_err) {
0d1c5c91 6229 error_propagate(errp, local_err);
5416645f 6230 return -EINVAL;
0d1c5c91 6231 }
5a8a30db 6232 }
0d1c5c91 6233
dafe0960
KW
6234 /*
6235 * Update permissions, they may differ for inactive nodes.
6236 *
6237 * Note that the required permissions of inactive images are always a
6238 * subset of the permissions required after activating the image. This
6239 * allows us to just get the permissions upfront without restricting
6240 * drv->bdrv_invalidate_cache().
6241 *
6242 * It also means that in error cases, we don't have to try and revert to
6243 * the old permissions (which is an operation that could fail, too). We can
6244 * just keep the extended permissions for the next time that an activation
6245 * of the image is tried.
6246 */
7bb4941a
KW
6247 if (bs->open_flags & BDRV_O_INACTIVE) {
6248 bs->open_flags &= ~BDRV_O_INACTIVE;
071b474f 6249 ret = bdrv_refresh_perms(bs, errp);
7bb4941a 6250 if (ret < 0) {
0d1c5c91 6251 bs->open_flags |= BDRV_O_INACTIVE;
5416645f 6252 return ret;
0d1c5c91 6253 }
3456a8d1 6254
7bb4941a
KW
6255 if (bs->drv->bdrv_co_invalidate_cache) {
6256 bs->drv->bdrv_co_invalidate_cache(bs, &local_err);
6257 if (local_err) {
6258 bs->open_flags |= BDRV_O_INACTIVE;
6259 error_propagate(errp, local_err);
5416645f 6260 return -EINVAL;
7bb4941a
KW
6261 }
6262 }
9c98f145 6263
7bb4941a
KW
6264 FOR_EACH_DIRTY_BITMAP(bs, bm) {
6265 bdrv_dirty_bitmap_skip_store(bm, false);
6266 }
6267
6268 ret = refresh_total_sectors(bs, bs->total_sectors);
6269 if (ret < 0) {
6270 bs->open_flags |= BDRV_O_INACTIVE;
6271 error_setg_errno(errp, -ret, "Could not refresh total sector count");
5416645f 6272 return ret;
7bb4941a 6273 }
5a8a30db 6274 }
4417ab7a
KW
6275
6276 QLIST_FOREACH(parent, &bs->parents, next_parent) {
bd86fb99
HR
6277 if (parent->klass->activate) {
6278 parent->klass->activate(parent, &local_err);
4417ab7a 6279 if (local_err) {
78fc3b3a 6280 bs->open_flags |= BDRV_O_INACTIVE;
4417ab7a 6281 error_propagate(errp, local_err);
5416645f 6282 return -EINVAL;
4417ab7a
KW
6283 }
6284 }
6285 }
5416645f
VSO
6286
6287 return 0;
0f15423c
AL
6288}
6289
5a8a30db 6290void bdrv_invalidate_cache_all(Error **errp)
0f15423c 6291{
7c8eece4 6292 BlockDriverState *bs;
88be7b4b 6293 BdrvNextIterator it;
0f15423c 6294
88be7b4b 6295 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
ed78cda3 6296 AioContext *aio_context = bdrv_get_aio_context(bs);
5416645f 6297 int ret;
ed78cda3
SH
6298
6299 aio_context_acquire(aio_context);
5416645f 6300 ret = bdrv_invalidate_cache(bs, errp);
ed78cda3 6301 aio_context_release(aio_context);
5416645f 6302 if (ret < 0) {
5e003f17 6303 bdrv_next_cleanup(&it);
5a8a30db
KW
6304 return;
6305 }
0f15423c
AL
6306 }
6307}
6308
9e37271f
KW
6309static bool bdrv_has_bds_parent(BlockDriverState *bs, bool only_active)
6310{
6311 BdrvChild *parent;
6312
6313 QLIST_FOREACH(parent, &bs->parents, next_parent) {
bd86fb99 6314 if (parent->klass->parent_is_bds) {
9e37271f
KW
6315 BlockDriverState *parent_bs = parent->opaque;
6316 if (!only_active || !(parent_bs->open_flags & BDRV_O_INACTIVE)) {
6317 return true;
6318 }
6319 }
6320 }
6321
6322 return false;
6323}
6324
6325static int bdrv_inactivate_recurse(BlockDriverState *bs)
76b1c7fe 6326{
cfa1a572 6327 BdrvChild *child, *parent;
76b1c7fe
KW
6328 int ret;
6329
d470ad42
HR
6330 if (!bs->drv) {
6331 return -ENOMEDIUM;
6332 }
6333
9e37271f
KW
6334 /* Make sure that we don't inactivate a child before its parent.
6335 * It will be covered by recursion from the yet active parent. */
6336 if (bdrv_has_bds_parent(bs, true)) {
6337 return 0;
6338 }
6339
6340 assert(!(bs->open_flags & BDRV_O_INACTIVE));
6341
6342 /* Inactivate this node */
6343 if (bs->drv->bdrv_inactivate) {
76b1c7fe
KW
6344 ret = bs->drv->bdrv_inactivate(bs);
6345 if (ret < 0) {
6346 return ret;
6347 }
6348 }
6349
9e37271f 6350 QLIST_FOREACH(parent, &bs->parents, next_parent) {
bd86fb99
HR
6351 if (parent->klass->inactivate) {
6352 ret = parent->klass->inactivate(parent);
9e37271f
KW
6353 if (ret < 0) {
6354 return ret;
cfa1a572
KW
6355 }
6356 }
9e37271f 6357 }
9c5e6594 6358
9e37271f 6359 bs->open_flags |= BDRV_O_INACTIVE;
7d5b5261 6360
bb87e4d1
VSO
6361 /*
6362 * Update permissions, they may differ for inactive nodes.
6363 * We only tried to loosen restrictions, so errors are not fatal, ignore
6364 * them.
6365 */
071b474f 6366 bdrv_refresh_perms(bs, NULL);
9e37271f
KW
6367
6368 /* Recursively inactivate children */
38701b6a 6369 QLIST_FOREACH(child, &bs->children, next) {
9e37271f 6370 ret = bdrv_inactivate_recurse(child->bs);
38701b6a
KW
6371 if (ret < 0) {
6372 return ret;
6373 }
6374 }
6375
76b1c7fe
KW
6376 return 0;
6377}
6378
6379int bdrv_inactivate_all(void)
6380{
79720af6 6381 BlockDriverState *bs = NULL;
88be7b4b 6382 BdrvNextIterator it;
aad0b7a0 6383 int ret = 0;
bd6458e4 6384 GSList *aio_ctxs = NULL, *ctx;
76b1c7fe 6385
88be7b4b 6386 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
bd6458e4
PB
6387 AioContext *aio_context = bdrv_get_aio_context(bs);
6388
6389 if (!g_slist_find(aio_ctxs, aio_context)) {
6390 aio_ctxs = g_slist_prepend(aio_ctxs, aio_context);
6391 aio_context_acquire(aio_context);
6392 }
aad0b7a0 6393 }
76b1c7fe 6394
9e37271f
KW
6395 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
6396 /* Nodes with BDS parents are covered by recursion from the last
6397 * parent that gets inactivated. Don't inactivate them a second
6398 * time if that has already happened. */
6399 if (bdrv_has_bds_parent(bs, false)) {
6400 continue;
6401 }
6402 ret = bdrv_inactivate_recurse(bs);
6403 if (ret < 0) {
6404 bdrv_next_cleanup(&it);
6405 goto out;
76b1c7fe
KW
6406 }
6407 }
6408
aad0b7a0 6409out:
bd6458e4
PB
6410 for (ctx = aio_ctxs; ctx != NULL; ctx = ctx->next) {
6411 AioContext *aio_context = ctx->data;
6412 aio_context_release(aio_context);
aad0b7a0 6413 }
bd6458e4 6414 g_slist_free(aio_ctxs);
aad0b7a0
FZ
6415
6416 return ret;
76b1c7fe
KW
6417}
6418
19cb3738
FB
6419/**************************************************************/
6420/* removable device support */
6421
6422/**
6423 * Return TRUE if the media is present
6424 */
e031f750 6425bool bdrv_is_inserted(BlockDriverState *bs)
19cb3738
FB
6426{
6427 BlockDriver *drv = bs->drv;
28d7a789 6428 BdrvChild *child;
a1aff5bf 6429
e031f750
HR
6430 if (!drv) {
6431 return false;
6432 }
28d7a789
HR
6433 if (drv->bdrv_is_inserted) {
6434 return drv->bdrv_is_inserted(bs);
6435 }
6436 QLIST_FOREACH(child, &bs->children, next) {
6437 if (!bdrv_is_inserted(child->bs)) {
6438 return false;
6439 }
e031f750 6440 }
28d7a789 6441 return true;
19cb3738
FB
6442}
6443
19cb3738
FB
6444/**
6445 * If eject_flag is TRUE, eject the media. Otherwise, close the tray
6446 */
f36f3949 6447void bdrv_eject(BlockDriverState *bs, bool eject_flag)
19cb3738
FB
6448{
6449 BlockDriver *drv = bs->drv;
19cb3738 6450
822e1cd1
MA
6451 if (drv && drv->bdrv_eject) {
6452 drv->bdrv_eject(bs, eject_flag);
19cb3738
FB
6453 }
6454}
6455
19cb3738
FB
6456/**
6457 * Lock or unlock the media (if it is locked, the user won't be able
6458 * to eject it manually).
6459 */
025e849a 6460void bdrv_lock_medium(BlockDriverState *bs, bool locked)
19cb3738
FB
6461{
6462 BlockDriver *drv = bs->drv;
6463
025e849a 6464 trace_bdrv_lock_medium(bs, locked);
b8c6d095 6465
025e849a
MA
6466 if (drv && drv->bdrv_lock_medium) {
6467 drv->bdrv_lock_medium(bs, locked);
19cb3738
FB
6468 }
6469}
985a03b0 6470
9fcb0251
FZ
6471/* Get a reference to bs */
6472void bdrv_ref(BlockDriverState *bs)
6473{
6474 bs->refcnt++;
6475}
6476
6477/* Release a previously grabbed reference to bs.
6478 * If after releasing, reference count is zero, the BlockDriverState is
6479 * deleted. */
6480void bdrv_unref(BlockDriverState *bs)
6481{
9a4d5ca6
JC
6482 if (!bs) {
6483 return;
6484 }
9fcb0251
FZ
6485 assert(bs->refcnt > 0);
6486 if (--bs->refcnt == 0) {
6487 bdrv_delete(bs);
6488 }
6489}
6490
fbe40ff7
FZ
6491struct BdrvOpBlocker {
6492 Error *reason;
6493 QLIST_ENTRY(BdrvOpBlocker) list;
6494};
6495
6496bool bdrv_op_is_blocked(BlockDriverState *bs, BlockOpType op, Error **errp)
6497{
6498 BdrvOpBlocker *blocker;
6499 assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
6500 if (!QLIST_EMPTY(&bs->op_blockers[op])) {
6501 blocker = QLIST_FIRST(&bs->op_blockers[op]);
4b576648
MA
6502 error_propagate_prepend(errp, error_copy(blocker->reason),
6503 "Node '%s' is busy: ",
6504 bdrv_get_device_or_node_name(bs));
fbe40ff7
FZ
6505 return true;
6506 }
6507 return false;
6508}
6509
6510void bdrv_op_block(BlockDriverState *bs, BlockOpType op, Error *reason)
6511{
6512 BdrvOpBlocker *blocker;
6513 assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
6514
5839e53b 6515 blocker = g_new0(BdrvOpBlocker, 1);
fbe40ff7
FZ
6516 blocker->reason = reason;
6517 QLIST_INSERT_HEAD(&bs->op_blockers[op], blocker, list);
6518}
6519
6520void bdrv_op_unblock(BlockDriverState *bs, BlockOpType op, Error *reason)
6521{
6522 BdrvOpBlocker *blocker, *next;
6523 assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
6524 QLIST_FOREACH_SAFE(blocker, &bs->op_blockers[op], list, next) {
6525 if (blocker->reason == reason) {
6526 QLIST_REMOVE(blocker, list);
6527 g_free(blocker);
6528 }
6529 }
6530}
6531
6532void bdrv_op_block_all(BlockDriverState *bs, Error *reason)
6533{
6534 int i;
6535 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
6536 bdrv_op_block(bs, i, reason);
6537 }
6538}
6539
6540void bdrv_op_unblock_all(BlockDriverState *bs, Error *reason)
6541{
6542 int i;
6543 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
6544 bdrv_op_unblock(bs, i, reason);
6545 }
6546}
6547
6548bool bdrv_op_blocker_is_empty(BlockDriverState *bs)
6549{
6550 int i;
6551
6552 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
6553 if (!QLIST_EMPTY(&bs->op_blockers[i])) {
6554 return false;
6555 }
6556 }
6557 return true;
6558}
6559
d92ada22
LC
6560void bdrv_img_create(const char *filename, const char *fmt,
6561 const char *base_filename, const char *base_fmt,
9217283d
FZ
6562 char *options, uint64_t img_size, int flags, bool quiet,
6563 Error **errp)
f88e1a42 6564{
83d0521a
CL
6565 QemuOptsList *create_opts = NULL;
6566 QemuOpts *opts = NULL;
6567 const char *backing_fmt, *backing_file;
6568 int64_t size;
f88e1a42 6569 BlockDriver *drv, *proto_drv;
cc84d90f 6570 Error *local_err = NULL;
f88e1a42
JS
6571 int ret = 0;
6572
6573 /* Find driver and parse its options */
6574 drv = bdrv_find_format(fmt);
6575 if (!drv) {
71c79813 6576 error_setg(errp, "Unknown file format '%s'", fmt);
d92ada22 6577 return;
f88e1a42
JS
6578 }
6579
b65a5e12 6580 proto_drv = bdrv_find_protocol(filename, true, errp);
f88e1a42 6581 if (!proto_drv) {
d92ada22 6582 return;
f88e1a42
JS
6583 }
6584
c6149724
HR
6585 if (!drv->create_opts) {
6586 error_setg(errp, "Format driver '%s' does not support image creation",
6587 drv->format_name);
6588 return;
6589 }
6590
5a5e7f8c
ML
6591 if (!proto_drv->create_opts) {
6592 error_setg(errp, "Protocol driver '%s' does not support image creation",
6593 proto_drv->format_name);
6594 return;
6595 }
6596
f6dc1c31 6597 /* Create parameter list */
c282e1fd 6598 create_opts = qemu_opts_append(create_opts, drv->create_opts);
5a5e7f8c 6599 create_opts = qemu_opts_append(create_opts, proto_drv->create_opts);
f88e1a42 6600
83d0521a 6601 opts = qemu_opts_create(create_opts, NULL, 0, &error_abort);
f88e1a42
JS
6602
6603 /* Parse -o options */
6604 if (options) {
a5f9b9df 6605 if (!qemu_opts_do_parse(opts, options, NULL, errp)) {
f88e1a42
JS
6606 goto out;
6607 }
6608 }
6609
f6dc1c31
KW
6610 if (!qemu_opt_get(opts, BLOCK_OPT_SIZE)) {
6611 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, img_size, &error_abort);
6612 } else if (img_size != UINT64_C(-1)) {
6613 error_setg(errp, "The image size must be specified only once");
6614 goto out;
6615 }
6616
f88e1a42 6617 if (base_filename) {
235e59cf 6618 if (!qemu_opt_set(opts, BLOCK_OPT_BACKING_FILE, base_filename,
3882578b 6619 NULL)) {
71c79813
LC
6620 error_setg(errp, "Backing file not supported for file format '%s'",
6621 fmt);
f88e1a42
JS
6622 goto out;
6623 }
6624 }
6625
6626 if (base_fmt) {
3882578b 6627 if (!qemu_opt_set(opts, BLOCK_OPT_BACKING_FMT, base_fmt, NULL)) {
71c79813
LC
6628 error_setg(errp, "Backing file format not supported for file "
6629 "format '%s'", fmt);
f88e1a42
JS
6630 goto out;
6631 }
6632 }
6633
83d0521a
CL
6634 backing_file = qemu_opt_get(opts, BLOCK_OPT_BACKING_FILE);
6635 if (backing_file) {
6636 if (!strcmp(filename, backing_file)) {
71c79813
LC
6637 error_setg(errp, "Error: Trying to create an image with the "
6638 "same filename as the backing file");
792da93a
JS
6639 goto out;
6640 }
975a7bd2
CK
6641 if (backing_file[0] == '\0') {
6642 error_setg(errp, "Expected backing file name, got empty string");
6643 goto out;
6644 }
792da93a
JS
6645 }
6646
83d0521a 6647 backing_fmt = qemu_opt_get(opts, BLOCK_OPT_BACKING_FMT);
f88e1a42 6648
6e6e55f5
JS
6649 /* The size for the image must always be specified, unless we have a backing
6650 * file and we have not been forbidden from opening it. */
a8b42a1c 6651 size = qemu_opt_get_size(opts, BLOCK_OPT_SIZE, img_size);
6e6e55f5
JS
6652 if (backing_file && !(flags & BDRV_O_NO_BACKING)) {
6653 BlockDriverState *bs;
645ae7d8 6654 char *full_backing;
6e6e55f5
JS
6655 int back_flags;
6656 QDict *backing_options = NULL;
6657
645ae7d8
HR
6658 full_backing =
6659 bdrv_get_full_backing_filename_from_filename(filename, backing_file,
6660 &local_err);
6e6e55f5 6661 if (local_err) {
6e6e55f5
JS
6662 goto out;
6663 }
645ae7d8 6664 assert(full_backing);
29168018 6665
d5b23994
HR
6666 /*
6667 * No need to do I/O here, which allows us to open encrypted
6668 * backing images without needing the secret
6669 */
6e6e55f5
JS
6670 back_flags = flags;
6671 back_flags &= ~(BDRV_O_RDWR | BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING);
d5b23994 6672 back_flags |= BDRV_O_NO_IO;
f88e1a42 6673
cc954f01 6674 backing_options = qdict_new();
6e6e55f5 6675 if (backing_fmt) {
6e6e55f5
JS
6676 qdict_put_str(backing_options, "driver", backing_fmt);
6677 }
cc954f01 6678 qdict_put_bool(backing_options, BDRV_OPT_FORCE_SHARE, true);
e6641719 6679
6e6e55f5
JS
6680 bs = bdrv_open(full_backing, NULL, backing_options, back_flags,
6681 &local_err);
6682 g_free(full_backing);
add8200d
EB
6683 if (!bs) {
6684 error_append_hint(&local_err, "Could not open backing image.\n");
6e6e55f5
JS
6685 goto out;
6686 } else {
d9f059aa 6687 if (!backing_fmt) {
497a30db
EB
6688 error_setg(&local_err,
6689 "Backing file specified without backing format");
6690 error_append_hint(&local_err, "Detected format of %s.",
6691 bs->drv->format_name);
6692 goto out;
d9f059aa 6693 }
6e6e55f5
JS
6694 if (size == -1) {
6695 /* Opened BS, have no size */
6696 size = bdrv_getlength(bs);
6697 if (size < 0) {
6698 error_setg_errno(errp, -size, "Could not get size of '%s'",
6699 backing_file);
6700 bdrv_unref(bs);
6701 goto out;
6702 }
6703 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, size, &error_abort);
52bf1e72 6704 }
66f6b814 6705 bdrv_unref(bs);
f88e1a42 6706 }
d9f059aa
EB
6707 /* (backing_file && !(flags & BDRV_O_NO_BACKING)) */
6708 } else if (backing_file && !backing_fmt) {
497a30db
EB
6709 error_setg(&local_err,
6710 "Backing file specified without backing format");
6711 goto out;
d9f059aa 6712 }
6e6e55f5
JS
6713
6714 if (size == -1) {
6715 error_setg(errp, "Image creation needs a size parameter");
6716 goto out;
f88e1a42
JS
6717 }
6718
f382d43a 6719 if (!quiet) {
fe646693 6720 printf("Formatting '%s', fmt=%s ", filename, fmt);
43c5d8f8 6721 qemu_opts_print(opts, " ");
f382d43a 6722 puts("");
4e2f4418 6723 fflush(stdout);
f382d43a 6724 }
83d0521a 6725
c282e1fd 6726 ret = bdrv_create(drv, filename, opts, &local_err);
83d0521a 6727
cc84d90f
HR
6728 if (ret == -EFBIG) {
6729 /* This is generally a better message than whatever the driver would
6730 * deliver (especially because of the cluster_size_hint), since that
6731 * is most probably not much different from "image too large". */
6732 const char *cluster_size_hint = "";
83d0521a 6733 if (qemu_opt_get_size(opts, BLOCK_OPT_CLUSTER_SIZE, 0)) {
cc84d90f 6734 cluster_size_hint = " (try using a larger cluster size)";
f88e1a42 6735 }
cc84d90f
HR
6736 error_setg(errp, "The image size is too large for file format '%s'"
6737 "%s", fmt, cluster_size_hint);
6738 error_free(local_err);
6739 local_err = NULL;
f88e1a42
JS
6740 }
6741
6742out:
83d0521a
CL
6743 qemu_opts_del(opts);
6744 qemu_opts_free(create_opts);
621ff94d 6745 error_propagate(errp, local_err);
f88e1a42 6746}
85d126f3
SH
6747
6748AioContext *bdrv_get_aio_context(BlockDriverState *bs)
6749{
33f2a757 6750 return bs ? bs->aio_context : qemu_get_aio_context();
dcd04228
SH
6751}
6752
e336fd4c
KW
6753AioContext *coroutine_fn bdrv_co_enter(BlockDriverState *bs)
6754{
6755 Coroutine *self = qemu_coroutine_self();
6756 AioContext *old_ctx = qemu_coroutine_get_aio_context(self);
6757 AioContext *new_ctx;
6758
6759 /*
6760 * Increase bs->in_flight to ensure that this operation is completed before
6761 * moving the node to a different AioContext. Read new_ctx only afterwards.
6762 */
6763 bdrv_inc_in_flight(bs);
6764
6765 new_ctx = bdrv_get_aio_context(bs);
6766 aio_co_reschedule_self(new_ctx);
6767 return old_ctx;
6768}
6769
6770void coroutine_fn bdrv_co_leave(BlockDriverState *bs, AioContext *old_ctx)
6771{
6772 aio_co_reschedule_self(old_ctx);
6773 bdrv_dec_in_flight(bs);
6774}
6775
18c6ac1c
KW
6776void coroutine_fn bdrv_co_lock(BlockDriverState *bs)
6777{
6778 AioContext *ctx = bdrv_get_aio_context(bs);
6779
6780 /* In the main thread, bs->aio_context won't change concurrently */
6781 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
6782
6783 /*
6784 * We're in coroutine context, so we already hold the lock of the main
6785 * loop AioContext. Don't lock it twice to avoid deadlocks.
6786 */
6787 assert(qemu_in_coroutine());
6788 if (ctx != qemu_get_aio_context()) {
6789 aio_context_acquire(ctx);
6790 }
6791}
6792
6793void coroutine_fn bdrv_co_unlock(BlockDriverState *bs)
6794{
6795 AioContext *ctx = bdrv_get_aio_context(bs);
6796
6797 assert(qemu_in_coroutine());
6798 if (ctx != qemu_get_aio_context()) {
6799 aio_context_release(ctx);
6800 }
6801}
6802
052a7572
FZ
6803void bdrv_coroutine_enter(BlockDriverState *bs, Coroutine *co)
6804{
6805 aio_co_enter(bdrv_get_aio_context(bs), co);
6806}
6807
e8a095da
SH
6808static void bdrv_do_remove_aio_context_notifier(BdrvAioNotifier *ban)
6809{
6810 QLIST_REMOVE(ban, list);
6811 g_free(ban);
6812}
6813
a3a683c3 6814static void bdrv_detach_aio_context(BlockDriverState *bs)
dcd04228 6815{
e8a095da 6816 BdrvAioNotifier *baf, *baf_tmp;
33384421 6817
e8a095da
SH
6818 assert(!bs->walking_aio_notifiers);
6819 bs->walking_aio_notifiers = true;
6820 QLIST_FOREACH_SAFE(baf, &bs->aio_notifiers, list, baf_tmp) {
6821 if (baf->deleted) {
6822 bdrv_do_remove_aio_context_notifier(baf);
6823 } else {
6824 baf->detach_aio_context(baf->opaque);
6825 }
33384421 6826 }
e8a095da
SH
6827 /* Never mind iterating again to check for ->deleted. bdrv_close() will
6828 * remove remaining aio notifiers if we aren't called again.
6829 */
6830 bs->walking_aio_notifiers = false;
33384421 6831
1bffe1ae 6832 if (bs->drv && bs->drv->bdrv_detach_aio_context) {
dcd04228
SH
6833 bs->drv->bdrv_detach_aio_context(bs);
6834 }
dcd04228 6835
e64f25f3
KW
6836 if (bs->quiesce_counter) {
6837 aio_enable_external(bs->aio_context);
6838 }
dcd04228
SH
6839 bs->aio_context = NULL;
6840}
6841
a3a683c3
KW
6842static void bdrv_attach_aio_context(BlockDriverState *bs,
6843 AioContext *new_context)
dcd04228 6844{
e8a095da 6845 BdrvAioNotifier *ban, *ban_tmp;
33384421 6846
e64f25f3
KW
6847 if (bs->quiesce_counter) {
6848 aio_disable_external(new_context);
6849 }
6850
dcd04228
SH
6851 bs->aio_context = new_context;
6852
1bffe1ae 6853 if (bs->drv && bs->drv->bdrv_attach_aio_context) {
dcd04228
SH
6854 bs->drv->bdrv_attach_aio_context(bs, new_context);
6855 }
33384421 6856
e8a095da
SH
6857 assert(!bs->walking_aio_notifiers);
6858 bs->walking_aio_notifiers = true;
6859 QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_tmp) {
6860 if (ban->deleted) {
6861 bdrv_do_remove_aio_context_notifier(ban);
6862 } else {
6863 ban->attached_aio_context(new_context, ban->opaque);
6864 }
33384421 6865 }
e8a095da 6866 bs->walking_aio_notifiers = false;
dcd04228
SH
6867}
6868
42a65f02
KW
6869/*
6870 * Changes the AioContext used for fd handlers, timers, and BHs by this
6871 * BlockDriverState and all its children and parents.
6872 *
43eaaaef
HR
6873 * Must be called from the main AioContext.
6874 *
42a65f02
KW
6875 * The caller must own the AioContext lock for the old AioContext of bs, but it
6876 * must not own the AioContext lock for new_context (unless new_context is the
6877 * same as the current context of bs).
6878 *
6879 * @ignore will accumulate all visited BdrvChild object. The caller is
6880 * responsible for freeing the list afterwards.
6881 */
53a7d041
KW
6882void bdrv_set_aio_context_ignore(BlockDriverState *bs,
6883 AioContext *new_context, GSList **ignore)
dcd04228 6884{
e037c09c 6885 AioContext *old_context = bdrv_get_aio_context(bs);
722d8e73
SL
6886 GSList *children_to_process = NULL;
6887 GSList *parents_to_process = NULL;
6888 GSList *entry;
6889 BdrvChild *child, *parent;
0d83708a 6890
43eaaaef
HR
6891 g_assert(qemu_get_current_aio_context() == qemu_get_aio_context());
6892
e037c09c 6893 if (old_context == new_context) {
57830a49
DP
6894 return;
6895 }
6896
d70d5954 6897 bdrv_drained_begin(bs);
0d83708a
KW
6898
6899 QLIST_FOREACH(child, &bs->children, next) {
53a7d041
KW
6900 if (g_slist_find(*ignore, child)) {
6901 continue;
6902 }
6903 *ignore = g_slist_prepend(*ignore, child);
722d8e73 6904 children_to_process = g_slist_prepend(children_to_process, child);
53a7d041 6905 }
722d8e73
SL
6906
6907 QLIST_FOREACH(parent, &bs->parents, next_parent) {
6908 if (g_slist_find(*ignore, parent)) {
53a7d041
KW
6909 continue;
6910 }
722d8e73
SL
6911 *ignore = g_slist_prepend(*ignore, parent);
6912 parents_to_process = g_slist_prepend(parents_to_process, parent);
6913 }
6914
6915 for (entry = children_to_process;
6916 entry != NULL;
6917 entry = g_slist_next(entry)) {
6918 child = entry->data;
6919 bdrv_set_aio_context_ignore(child->bs, new_context, ignore);
6920 }
6921 g_slist_free(children_to_process);
6922
6923 for (entry = parents_to_process;
6924 entry != NULL;
6925 entry = g_slist_next(entry)) {
6926 parent = entry->data;
6927 assert(parent->klass->set_aio_ctx);
6928 parent->klass->set_aio_ctx(parent, new_context, ignore);
0d83708a 6929 }
722d8e73 6930 g_slist_free(parents_to_process);
0d83708a 6931
dcd04228
SH
6932 bdrv_detach_aio_context(bs);
6933
e037c09c 6934 /* Acquire the new context, if necessary */
43eaaaef 6935 if (qemu_get_aio_context() != new_context) {
e037c09c
HR
6936 aio_context_acquire(new_context);
6937 }
6938
dcd04228 6939 bdrv_attach_aio_context(bs, new_context);
e037c09c
HR
6940
6941 /*
6942 * If this function was recursively called from
6943 * bdrv_set_aio_context_ignore(), there may be nodes in the
6944 * subtree that have not yet been moved to the new AioContext.
6945 * Release the old one so bdrv_drained_end() can poll them.
6946 */
43eaaaef 6947 if (qemu_get_aio_context() != old_context) {
e037c09c
HR
6948 aio_context_release(old_context);
6949 }
6950
d70d5954 6951 bdrv_drained_end(bs);
e037c09c 6952
43eaaaef 6953 if (qemu_get_aio_context() != old_context) {
e037c09c
HR
6954 aio_context_acquire(old_context);
6955 }
43eaaaef 6956 if (qemu_get_aio_context() != new_context) {
e037c09c
HR
6957 aio_context_release(new_context);
6958 }
85d126f3 6959}
d616b224 6960
5d231849
KW
6961static bool bdrv_parent_can_set_aio_context(BdrvChild *c, AioContext *ctx,
6962 GSList **ignore, Error **errp)
6963{
6964 if (g_slist_find(*ignore, c)) {
6965 return true;
6966 }
6967 *ignore = g_slist_prepend(*ignore, c);
6968
bd86fb99
HR
6969 /*
6970 * A BdrvChildClass that doesn't handle AioContext changes cannot
6971 * tolerate any AioContext changes
6972 */
6973 if (!c->klass->can_set_aio_ctx) {
5d231849
KW
6974 char *user = bdrv_child_user_desc(c);
6975 error_setg(errp, "Changing iothreads is not supported by %s", user);
6976 g_free(user);
6977 return false;
6978 }
bd86fb99 6979 if (!c->klass->can_set_aio_ctx(c, ctx, ignore, errp)) {
5d231849
KW
6980 assert(!errp || *errp);
6981 return false;
6982 }
6983 return true;
6984}
6985
6986bool bdrv_child_can_set_aio_context(BdrvChild *c, AioContext *ctx,
6987 GSList **ignore, Error **errp)
6988{
6989 if (g_slist_find(*ignore, c)) {
6990 return true;
6991 }
6992 *ignore = g_slist_prepend(*ignore, c);
6993 return bdrv_can_set_aio_context(c->bs, ctx, ignore, errp);
6994}
6995
6996/* @ignore will accumulate all visited BdrvChild object. The caller is
6997 * responsible for freeing the list afterwards. */
6998bool bdrv_can_set_aio_context(BlockDriverState *bs, AioContext *ctx,
6999 GSList **ignore, Error **errp)
7000{
7001 BdrvChild *c;
7002
7003 if (bdrv_get_aio_context(bs) == ctx) {
7004 return true;
7005 }
7006
7007 QLIST_FOREACH(c, &bs->parents, next_parent) {
7008 if (!bdrv_parent_can_set_aio_context(c, ctx, ignore, errp)) {
7009 return false;
7010 }
7011 }
7012 QLIST_FOREACH(c, &bs->children, next) {
7013 if (!bdrv_child_can_set_aio_context(c, ctx, ignore, errp)) {
7014 return false;
7015 }
7016 }
7017
7018 return true;
7019}
7020
7021int bdrv_child_try_set_aio_context(BlockDriverState *bs, AioContext *ctx,
7022 BdrvChild *ignore_child, Error **errp)
7023{
7024 GSList *ignore;
7025 bool ret;
7026
7027 ignore = ignore_child ? g_slist_prepend(NULL, ignore_child) : NULL;
7028 ret = bdrv_can_set_aio_context(bs, ctx, &ignore, errp);
7029 g_slist_free(ignore);
7030
7031 if (!ret) {
7032 return -EPERM;
7033 }
7034
53a7d041
KW
7035 ignore = ignore_child ? g_slist_prepend(NULL, ignore_child) : NULL;
7036 bdrv_set_aio_context_ignore(bs, ctx, &ignore);
7037 g_slist_free(ignore);
7038
5d231849
KW
7039 return 0;
7040}
7041
7042int bdrv_try_set_aio_context(BlockDriverState *bs, AioContext *ctx,
7043 Error **errp)
7044{
7045 return bdrv_child_try_set_aio_context(bs, ctx, NULL, errp);
7046}
7047
33384421
HR
7048void bdrv_add_aio_context_notifier(BlockDriverState *bs,
7049 void (*attached_aio_context)(AioContext *new_context, void *opaque),
7050 void (*detach_aio_context)(void *opaque), void *opaque)
7051{
7052 BdrvAioNotifier *ban = g_new(BdrvAioNotifier, 1);
7053 *ban = (BdrvAioNotifier){
7054 .attached_aio_context = attached_aio_context,
7055 .detach_aio_context = detach_aio_context,
7056 .opaque = opaque
7057 };
7058
7059 QLIST_INSERT_HEAD(&bs->aio_notifiers, ban, list);
7060}
7061
7062void bdrv_remove_aio_context_notifier(BlockDriverState *bs,
7063 void (*attached_aio_context)(AioContext *,
7064 void *),
7065 void (*detach_aio_context)(void *),
7066 void *opaque)
7067{
7068 BdrvAioNotifier *ban, *ban_next;
7069
7070 QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_next) {
7071 if (ban->attached_aio_context == attached_aio_context &&
7072 ban->detach_aio_context == detach_aio_context &&
e8a095da
SH
7073 ban->opaque == opaque &&
7074 ban->deleted == false)
33384421 7075 {
e8a095da
SH
7076 if (bs->walking_aio_notifiers) {
7077 ban->deleted = true;
7078 } else {
7079 bdrv_do_remove_aio_context_notifier(ban);
7080 }
33384421
HR
7081 return;
7082 }
7083 }
7084
7085 abort();
7086}
7087
77485434 7088int bdrv_amend_options(BlockDriverState *bs, QemuOpts *opts,
d1402b50 7089 BlockDriverAmendStatusCB *status_cb, void *cb_opaque,
a3579bfa 7090 bool force,
d1402b50 7091 Error **errp)
6f176b48 7092{
d470ad42 7093 if (!bs->drv) {
d1402b50 7094 error_setg(errp, "Node is ejected");
d470ad42
HR
7095 return -ENOMEDIUM;
7096 }
c282e1fd 7097 if (!bs->drv->bdrv_amend_options) {
d1402b50
HR
7098 error_setg(errp, "Block driver '%s' does not support option amendment",
7099 bs->drv->format_name);
6f176b48
HR
7100 return -ENOTSUP;
7101 }
a3579bfa
ML
7102 return bs->drv->bdrv_amend_options(bs, opts, status_cb,
7103 cb_opaque, force, errp);
6f176b48 7104}
f6186f49 7105
5d69b5ab
HR
7106/*
7107 * This function checks whether the given @to_replace is allowed to be
7108 * replaced by a node that always shows the same data as @bs. This is
7109 * used for example to verify whether the mirror job can replace
7110 * @to_replace by the target mirrored from @bs.
7111 * To be replaceable, @bs and @to_replace may either be guaranteed to
7112 * always show the same data (because they are only connected through
7113 * filters), or some driver may allow replacing one of its children
7114 * because it can guarantee that this child's data is not visible at
7115 * all (for example, for dissenting quorum children that have no other
7116 * parents).
7117 */
7118bool bdrv_recurse_can_replace(BlockDriverState *bs,
7119 BlockDriverState *to_replace)
7120{
93393e69
HR
7121 BlockDriverState *filtered;
7122
5d69b5ab
HR
7123 if (!bs || !bs->drv) {
7124 return false;
7125 }
7126
7127 if (bs == to_replace) {
7128 return true;
7129 }
7130
7131 /* See what the driver can do */
7132 if (bs->drv->bdrv_recurse_can_replace) {
7133 return bs->drv->bdrv_recurse_can_replace(bs, to_replace);
7134 }
7135
7136 /* For filters without an own implementation, we can recurse on our own */
93393e69
HR
7137 filtered = bdrv_filter_bs(bs);
7138 if (filtered) {
7139 return bdrv_recurse_can_replace(filtered, to_replace);
5d69b5ab
HR
7140 }
7141
7142 /* Safe default */
7143 return false;
7144}
7145
810803a8
HR
7146/*
7147 * Check whether the given @node_name can be replaced by a node that
7148 * has the same data as @parent_bs. If so, return @node_name's BDS;
7149 * NULL otherwise.
7150 *
7151 * @node_name must be a (recursive) *child of @parent_bs (or this
7152 * function will return NULL).
7153 *
7154 * The result (whether the node can be replaced or not) is only valid
7155 * for as long as no graph or permission changes occur.
7156 */
e12f3784
WC
7157BlockDriverState *check_to_replace_node(BlockDriverState *parent_bs,
7158 const char *node_name, Error **errp)
09158f00
BC
7159{
7160 BlockDriverState *to_replace_bs = bdrv_find_node(node_name);
5a7e7a0b
SH
7161 AioContext *aio_context;
7162
09158f00 7163 if (!to_replace_bs) {
785ec4b1 7164 error_setg(errp, "Failed to find node with node-name='%s'", node_name);
09158f00
BC
7165 return NULL;
7166 }
7167
5a7e7a0b
SH
7168 aio_context = bdrv_get_aio_context(to_replace_bs);
7169 aio_context_acquire(aio_context);
7170
09158f00 7171 if (bdrv_op_is_blocked(to_replace_bs, BLOCK_OP_TYPE_REPLACE, errp)) {
5a7e7a0b
SH
7172 to_replace_bs = NULL;
7173 goto out;
09158f00
BC
7174 }
7175
7176 /* We don't want arbitrary node of the BDS chain to be replaced only the top
7177 * most non filter in order to prevent data corruption.
7178 * Another benefit is that this tests exclude backing files which are
7179 * blocked by the backing blockers.
7180 */
810803a8
HR
7181 if (!bdrv_recurse_can_replace(parent_bs, to_replace_bs)) {
7182 error_setg(errp, "Cannot replace '%s' by a node mirrored from '%s', "
7183 "because it cannot be guaranteed that doing so would not "
7184 "lead to an abrupt change of visible data",
7185 node_name, parent_bs->node_name);
5a7e7a0b
SH
7186 to_replace_bs = NULL;
7187 goto out;
09158f00
BC
7188 }
7189
5a7e7a0b
SH
7190out:
7191 aio_context_release(aio_context);
09158f00
BC
7192 return to_replace_bs;
7193}
448ad91d 7194
97e2f021
HR
7195/**
7196 * Iterates through the list of runtime option keys that are said to
7197 * be "strong" for a BDS. An option is called "strong" if it changes
7198 * a BDS's data. For example, the null block driver's "size" and
7199 * "read-zeroes" options are strong, but its "latency-ns" option is
7200 * not.
7201 *
7202 * If a key returned by this function ends with a dot, all options
7203 * starting with that prefix are strong.
7204 */
7205static const char *const *strong_options(BlockDriverState *bs,
7206 const char *const *curopt)
7207{
7208 static const char *const global_options[] = {
7209 "driver", "filename", NULL
7210 };
7211
7212 if (!curopt) {
7213 return &global_options[0];
7214 }
7215
7216 curopt++;
7217 if (curopt == &global_options[ARRAY_SIZE(global_options) - 1] && bs->drv) {
7218 curopt = bs->drv->strong_runtime_opts;
7219 }
7220
7221 return (curopt && *curopt) ? curopt : NULL;
7222}
7223
7224/**
7225 * Copies all strong runtime options from bs->options to the given
7226 * QDict. The set of strong option keys is determined by invoking
7227 * strong_options().
7228 *
7229 * Returns true iff any strong option was present in bs->options (and
7230 * thus copied to the target QDict) with the exception of "filename"
7231 * and "driver". The caller is expected to use this value to decide
7232 * whether the existence of strong options prevents the generation of
7233 * a plain filename.
7234 */
7235static bool append_strong_runtime_options(QDict *d, BlockDriverState *bs)
7236{
7237 bool found_any = false;
7238 const char *const *option_name = NULL;
7239
7240 if (!bs->drv) {
7241 return false;
7242 }
7243
7244 while ((option_name = strong_options(bs, option_name))) {
7245 bool option_given = false;
7246
7247 assert(strlen(*option_name) > 0);
7248 if ((*option_name)[strlen(*option_name) - 1] != '.') {
7249 QObject *entry = qdict_get(bs->options, *option_name);
7250 if (!entry) {
7251 continue;
7252 }
7253
7254 qdict_put_obj(d, *option_name, qobject_ref(entry));
7255 option_given = true;
7256 } else {
7257 const QDictEntry *entry;
7258 for (entry = qdict_first(bs->options); entry;
7259 entry = qdict_next(bs->options, entry))
7260 {
7261 if (strstart(qdict_entry_key(entry), *option_name, NULL)) {
7262 qdict_put_obj(d, qdict_entry_key(entry),
7263 qobject_ref(qdict_entry_value(entry)));
7264 option_given = true;
7265 }
7266 }
7267 }
7268
7269 /* While "driver" and "filename" need to be included in a JSON filename,
7270 * their existence does not prohibit generation of a plain filename. */
7271 if (!found_any && option_given &&
7272 strcmp(*option_name, "driver") && strcmp(*option_name, "filename"))
7273 {
7274 found_any = true;
7275 }
7276 }
7277
62a01a27
HR
7278 if (!qdict_haskey(d, "driver")) {
7279 /* Drivers created with bdrv_new_open_driver() may not have a
7280 * @driver option. Add it here. */
7281 qdict_put_str(d, "driver", bs->drv->format_name);
7282 }
7283
97e2f021
HR
7284 return found_any;
7285}
7286
90993623
HR
7287/* Note: This function may return false positives; it may return true
7288 * even if opening the backing file specified by bs's image header
7289 * would result in exactly bs->backing. */
0b877d09 7290bool bdrv_backing_overridden(BlockDriverState *bs)
90993623
HR
7291{
7292 if (bs->backing) {
7293 return strcmp(bs->auto_backing_file,
7294 bs->backing->bs->filename);
7295 } else {
7296 /* No backing BDS, so if the image header reports any backing
7297 * file, it must have been suppressed */
7298 return bs->auto_backing_file[0] != '\0';
7299 }
7300}
7301
91af7014
HR
7302/* Updates the following BDS fields:
7303 * - exact_filename: A filename which may be used for opening a block device
7304 * which (mostly) equals the given BDS (even without any
7305 * other options; so reading and writing must return the same
7306 * results, but caching etc. may be different)
7307 * - full_open_options: Options which, when given when opening a block device
7308 * (without a filename), result in a BDS (mostly)
7309 * equalling the given one
7310 * - filename: If exact_filename is set, it is copied here. Otherwise,
7311 * full_open_options is converted to a JSON object, prefixed with
7312 * "json:" (for use through the JSON pseudo protocol) and put here.
7313 */
7314void bdrv_refresh_filename(BlockDriverState *bs)
7315{
7316 BlockDriver *drv = bs->drv;
e24518e3 7317 BdrvChild *child;
52f72d6f 7318 BlockDriverState *primary_child_bs;
91af7014 7319 QDict *opts;
90993623 7320 bool backing_overridden;
998b3a1e
HR
7321 bool generate_json_filename; /* Whether our default implementation should
7322 fill exact_filename (false) or not (true) */
91af7014
HR
7323
7324 if (!drv) {
7325 return;
7326 }
7327
e24518e3
HR
7328 /* This BDS's file name may depend on any of its children's file names, so
7329 * refresh those first */
7330 QLIST_FOREACH(child, &bs->children, next) {
7331 bdrv_refresh_filename(child->bs);
91af7014
HR
7332 }
7333
bb808d5f
HR
7334 if (bs->implicit) {
7335 /* For implicit nodes, just copy everything from the single child */
7336 child = QLIST_FIRST(&bs->children);
7337 assert(QLIST_NEXT(child, next) == NULL);
7338
7339 pstrcpy(bs->exact_filename, sizeof(bs->exact_filename),
7340 child->bs->exact_filename);
7341 pstrcpy(bs->filename, sizeof(bs->filename), child->bs->filename);
7342
cb895614 7343 qobject_unref(bs->full_open_options);
bb808d5f
HR
7344 bs->full_open_options = qobject_ref(child->bs->full_open_options);
7345
7346 return;
7347 }
7348
90993623
HR
7349 backing_overridden = bdrv_backing_overridden(bs);
7350
7351 if (bs->open_flags & BDRV_O_NO_IO) {
7352 /* Without I/O, the backing file does not change anything.
7353 * Therefore, in such a case (primarily qemu-img), we can
7354 * pretend the backing file has not been overridden even if
7355 * it technically has been. */
7356 backing_overridden = false;
7357 }
7358
97e2f021
HR
7359 /* Gather the options QDict */
7360 opts = qdict_new();
998b3a1e
HR
7361 generate_json_filename = append_strong_runtime_options(opts, bs);
7362 generate_json_filename |= backing_overridden;
97e2f021
HR
7363
7364 if (drv->bdrv_gather_child_options) {
7365 /* Some block drivers may not want to present all of their children's
7366 * options, or name them differently from BdrvChild.name */
7367 drv->bdrv_gather_child_options(bs, opts, backing_overridden);
7368 } else {
7369 QLIST_FOREACH(child, &bs->children, next) {
25191e5f 7370 if (child == bs->backing && !backing_overridden) {
97e2f021
HR
7371 /* We can skip the backing BDS if it has not been overridden */
7372 continue;
7373 }
7374
7375 qdict_put(opts, child->name,
7376 qobject_ref(child->bs->full_open_options));
7377 }
7378
7379 if (backing_overridden && !bs->backing) {
7380 /* Force no backing file */
7381 qdict_put_null(opts, "backing");
7382 }
7383 }
7384
7385 qobject_unref(bs->full_open_options);
7386 bs->full_open_options = opts;
7387
52f72d6f
HR
7388 primary_child_bs = bdrv_primary_bs(bs);
7389
998b3a1e
HR
7390 if (drv->bdrv_refresh_filename) {
7391 /* Obsolete information is of no use here, so drop the old file name
7392 * information before refreshing it */
7393 bs->exact_filename[0] = '\0';
7394
7395 drv->bdrv_refresh_filename(bs);
52f72d6f
HR
7396 } else if (primary_child_bs) {
7397 /*
7398 * Try to reconstruct valid information from the underlying
7399 * file -- this only works for format nodes (filter nodes
7400 * cannot be probed and as such must be selected by the user
7401 * either through an options dict, or through a special
7402 * filename which the filter driver must construct in its
7403 * .bdrv_refresh_filename() implementation).
7404 */
998b3a1e
HR
7405
7406 bs->exact_filename[0] = '\0';
7407
fb695c74
HR
7408 /*
7409 * We can use the underlying file's filename if:
7410 * - it has a filename,
52f72d6f 7411 * - the current BDS is not a filter,
fb695c74
HR
7412 * - the file is a protocol BDS, and
7413 * - opening that file (as this BDS's format) will automatically create
7414 * the BDS tree we have right now, that is:
7415 * - the user did not significantly change this BDS's behavior with
7416 * some explicit (strong) options
7417 * - no non-file child of this BDS has been overridden by the user
7418 * Both of these conditions are represented by generate_json_filename.
7419 */
52f72d6f
HR
7420 if (primary_child_bs->exact_filename[0] &&
7421 primary_child_bs->drv->bdrv_file_open &&
7422 !drv->is_filter && !generate_json_filename)
fb695c74 7423 {
52f72d6f 7424 strcpy(bs->exact_filename, primary_child_bs->exact_filename);
998b3a1e
HR
7425 }
7426 }
7427
91af7014
HR
7428 if (bs->exact_filename[0]) {
7429 pstrcpy(bs->filename, sizeof(bs->filename), bs->exact_filename);
97e2f021 7430 } else {
eab3a467 7431 GString *json = qobject_to_json(QOBJECT(bs->full_open_options));
5c86bdf1 7432 if (snprintf(bs->filename, sizeof(bs->filename), "json:%s",
eab3a467 7433 json->str) >= sizeof(bs->filename)) {
5c86bdf1
EB
7434 /* Give user a hint if we truncated things. */
7435 strcpy(bs->filename + sizeof(bs->filename) - 4, "...");
7436 }
eab3a467 7437 g_string_free(json, true);
91af7014
HR
7438 }
7439}
e06018ad 7440
1e89d0f9
HR
7441char *bdrv_dirname(BlockDriverState *bs, Error **errp)
7442{
7443 BlockDriver *drv = bs->drv;
52f72d6f 7444 BlockDriverState *child_bs;
1e89d0f9
HR
7445
7446 if (!drv) {
7447 error_setg(errp, "Node '%s' is ejected", bs->node_name);
7448 return NULL;
7449 }
7450
7451 if (drv->bdrv_dirname) {
7452 return drv->bdrv_dirname(bs, errp);
7453 }
7454
52f72d6f
HR
7455 child_bs = bdrv_primary_bs(bs);
7456 if (child_bs) {
7457 return bdrv_dirname(child_bs, errp);
1e89d0f9
HR
7458 }
7459
7460 bdrv_refresh_filename(bs);
7461 if (bs->exact_filename[0] != '\0') {
7462 return path_combine(bs->exact_filename, "");
7463 }
7464
7465 error_setg(errp, "Cannot generate a base directory for %s nodes",
7466 drv->format_name);
7467 return NULL;
7468}
7469
e06018ad
WC
7470/*
7471 * Hot add/remove a BDS's child. So the user can take a child offline when
7472 * it is broken and take a new child online
7473 */
7474void bdrv_add_child(BlockDriverState *parent_bs, BlockDriverState *child_bs,
7475 Error **errp)
7476{
7477
7478 if (!parent_bs->drv || !parent_bs->drv->bdrv_add_child) {
7479 error_setg(errp, "The node %s does not support adding a child",
7480 bdrv_get_device_or_node_name(parent_bs));
7481 return;
7482 }
7483
7484 if (!QLIST_EMPTY(&child_bs->parents)) {
7485 error_setg(errp, "The node %s already has a parent",
7486 child_bs->node_name);
7487 return;
7488 }
7489
7490 parent_bs->drv->bdrv_add_child(parent_bs, child_bs, errp);
7491}
7492
7493void bdrv_del_child(BlockDriverState *parent_bs, BdrvChild *child, Error **errp)
7494{
7495 BdrvChild *tmp;
7496
7497 if (!parent_bs->drv || !parent_bs->drv->bdrv_del_child) {
7498 error_setg(errp, "The node %s does not support removing a child",
7499 bdrv_get_device_or_node_name(parent_bs));
7500 return;
7501 }
7502
7503 QLIST_FOREACH(tmp, &parent_bs->children, next) {
7504 if (tmp == child) {
7505 break;
7506 }
7507 }
7508
7509 if (!tmp) {
7510 error_setg(errp, "The node %s does not have a child named %s",
7511 bdrv_get_device_or_node_name(parent_bs),
7512 bdrv_get_device_or_node_name(child->bs));
7513 return;
7514 }
7515
7516 parent_bs->drv->bdrv_del_child(parent_bs, child, errp);
7517}
6f7a3b53
HR
7518
7519int bdrv_make_empty(BdrvChild *c, Error **errp)
7520{
7521 BlockDriver *drv = c->bs->drv;
7522 int ret;
7523
7524 assert(c->perm & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED));
7525
7526 if (!drv->bdrv_make_empty) {
7527 error_setg(errp, "%s does not support emptying nodes",
7528 drv->format_name);
7529 return -ENOTSUP;
7530 }
7531
7532 ret = drv->bdrv_make_empty(c->bs);
7533 if (ret < 0) {
7534 error_setg_errno(errp, -ret, "Failed to empty %s",
7535 c->bs->filename);
7536 return ret;
7537 }
7538
7539 return 0;
7540}
9a6fc887
HR
7541
7542/*
7543 * Return the child that @bs acts as an overlay for, and from which data may be
7544 * copied in COW or COR operations. Usually this is the backing file.
7545 */
7546BdrvChild *bdrv_cow_child(BlockDriverState *bs)
7547{
7548 if (!bs || !bs->drv) {
7549 return NULL;
7550 }
7551
7552 if (bs->drv->is_filter) {
7553 return NULL;
7554 }
7555
7556 if (!bs->backing) {
7557 return NULL;
7558 }
7559
7560 assert(bs->backing->role & BDRV_CHILD_COW);
7561 return bs->backing;
7562}
7563
7564/*
7565 * If @bs acts as a filter for exactly one of its children, return
7566 * that child.
7567 */
7568BdrvChild *bdrv_filter_child(BlockDriverState *bs)
7569{
7570 BdrvChild *c;
7571
7572 if (!bs || !bs->drv) {
7573 return NULL;
7574 }
7575
7576 if (!bs->drv->is_filter) {
7577 return NULL;
7578 }
7579
7580 /* Only one of @backing or @file may be used */
7581 assert(!(bs->backing && bs->file));
7582
7583 c = bs->backing ?: bs->file;
7584 if (!c) {
7585 return NULL;
7586 }
7587
7588 assert(c->role & BDRV_CHILD_FILTERED);
7589 return c;
7590}
7591
7592/*
7593 * Return either the result of bdrv_cow_child() or bdrv_filter_child(),
7594 * whichever is non-NULL.
7595 *
7596 * Return NULL if both are NULL.
7597 */
7598BdrvChild *bdrv_filter_or_cow_child(BlockDriverState *bs)
7599{
7600 BdrvChild *cow_child = bdrv_cow_child(bs);
7601 BdrvChild *filter_child = bdrv_filter_child(bs);
7602
7603 /* Filter nodes cannot have COW backing files */
7604 assert(!(cow_child && filter_child));
7605
7606 return cow_child ?: filter_child;
7607}
7608
7609/*
7610 * Return the primary child of this node: For filters, that is the
7611 * filtered child. For other nodes, that is usually the child storing
7612 * metadata.
7613 * (A generally more helpful description is that this is (usually) the
7614 * child that has the same filename as @bs.)
7615 *
7616 * Drivers do not necessarily have a primary child; for example quorum
7617 * does not.
7618 */
7619BdrvChild *bdrv_primary_child(BlockDriverState *bs)
7620{
7621 BdrvChild *c, *found = NULL;
7622
7623 QLIST_FOREACH(c, &bs->children, next) {
7624 if (c->role & BDRV_CHILD_PRIMARY) {
7625 assert(!found);
7626 found = c;
7627 }
7628 }
7629
7630 return found;
7631}
d38d7eb8
HR
7632
7633static BlockDriverState *bdrv_do_skip_filters(BlockDriverState *bs,
7634 bool stop_on_explicit_filter)
7635{
7636 BdrvChild *c;
7637
7638 if (!bs) {
7639 return NULL;
7640 }
7641
7642 while (!(stop_on_explicit_filter && !bs->implicit)) {
7643 c = bdrv_filter_child(bs);
7644 if (!c) {
7645 /*
7646 * A filter that is embedded in a working block graph must
7647 * have a child. Assert this here so this function does
7648 * not return a filter node that is not expected by the
7649 * caller.
7650 */
7651 assert(!bs->drv || !bs->drv->is_filter);
7652 break;
7653 }
7654 bs = c->bs;
7655 }
7656 /*
7657 * Note that this treats nodes with bs->drv == NULL as not being
7658 * filters (bs->drv == NULL should be replaced by something else
7659 * anyway).
7660 * The advantage of this behavior is that this function will thus
7661 * always return a non-NULL value (given a non-NULL @bs).
7662 */
7663
7664 return bs;
7665}
7666
7667/*
7668 * Return the first BDS that has not been added implicitly or that
7669 * does not have a filtered child down the chain starting from @bs
7670 * (including @bs itself).
7671 */
7672BlockDriverState *bdrv_skip_implicit_filters(BlockDriverState *bs)
7673{
7674 return bdrv_do_skip_filters(bs, true);
7675}
7676
7677/*
7678 * Return the first BDS that does not have a filtered child down the
7679 * chain starting from @bs (including @bs itself).
7680 */
7681BlockDriverState *bdrv_skip_filters(BlockDriverState *bs)
7682{
7683 return bdrv_do_skip_filters(bs, false);
7684}
7685
7686/*
7687 * For a backing chain, return the first non-filter backing image of
7688 * the first non-filter image.
7689 */
7690BlockDriverState *bdrv_backing_chain_next(BlockDriverState *bs)
7691{
7692 return bdrv_skip_filters(bdrv_cow_bs(bdrv_skip_filters(bs)));
7693}
0bc329fb
HR
7694
7695/**
7696 * Check whether [offset, offset + bytes) overlaps with the cached
7697 * block-status data region.
7698 *
7699 * If so, and @pnum is not NULL, set *pnum to `bsc.data_end - offset`,
7700 * which is what bdrv_bsc_is_data()'s interface needs.
7701 * Otherwise, *pnum is not touched.
7702 */
7703static bool bdrv_bsc_range_overlaps_locked(BlockDriverState *bs,
7704 int64_t offset, int64_t bytes,
7705 int64_t *pnum)
7706{
7707 BdrvBlockStatusCache *bsc = qatomic_rcu_read(&bs->block_status_cache);
7708 bool overlaps;
7709
7710 overlaps =
7711 qatomic_read(&bsc->valid) &&
7712 ranges_overlap(offset, bytes, bsc->data_start,
7713 bsc->data_end - bsc->data_start);
7714
7715 if (overlaps && pnum) {
7716 *pnum = bsc->data_end - offset;
7717 }
7718
7719 return overlaps;
7720}
7721
7722/**
7723 * See block_int.h for this function's documentation.
7724 */
7725bool bdrv_bsc_is_data(BlockDriverState *bs, int64_t offset, int64_t *pnum)
7726{
7727 RCU_READ_LOCK_GUARD();
7728
7729 return bdrv_bsc_range_overlaps_locked(bs, offset, 1, pnum);
7730}
7731
7732/**
7733 * See block_int.h for this function's documentation.
7734 */
7735void bdrv_bsc_invalidate_range(BlockDriverState *bs,
7736 int64_t offset, int64_t bytes)
7737{
7738 RCU_READ_LOCK_GUARD();
7739
7740 if (bdrv_bsc_range_overlaps_locked(bs, offset, bytes, NULL)) {
7741 qatomic_set(&bs->block_status_cache->valid, false);
7742 }
7743}
7744
7745/**
7746 * See block_int.h for this function's documentation.
7747 */
7748void bdrv_bsc_fill(BlockDriverState *bs, int64_t offset, int64_t bytes)
7749{
7750 BdrvBlockStatusCache *new_bsc = g_new(BdrvBlockStatusCache, 1);
7751 BdrvBlockStatusCache *old_bsc;
7752
7753 *new_bsc = (BdrvBlockStatusCache) {
7754 .valid = true,
7755 .data_start = offset,
7756 .data_end = offset + bytes,
7757 };
7758
7759 QEMU_LOCK_GUARD(&bs->bsc_modify_lock);
7760
7761 old_bsc = qatomic_rcu_read(&bs->block_status_cache);
7762 qatomic_rcu_set(&bs->block_status_cache, new_bsc);
7763 if (old_bsc) {
7764 g_free_rcu(old_bsc, rcu);
7765 }
7766}
This page took 2.719191 seconds and 4 git commands to generate.