]> git.proxmox.com Git - mirror_qemu.git/blame - block.c
block: Mark bdrv_filter_or_cow_bs() and callers GRAPH_RDLOCK
[mirror_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"
e2c1c34f 30#include "block/dirty-bitmap.h"
0c9b70d5 31#include "block/fuse.h"
cd7fca95 32#include "block/nbd.h"
609f45ea 33#include "block/qdict.h"
d49b6836 34#include "qemu/error-report.h"
5e5733e5 35#include "block/module_block.h"
db725815 36#include "qemu/main-loop.h"
1de7afc9 37#include "qemu/module.h"
e688df6b 38#include "qapi/error.h"
452fcdbc 39#include "qapi/qmp/qdict.h"
7b1b5d19 40#include "qapi/qmp/qjson.h"
e59a0cf1 41#include "qapi/qmp/qnull.h"
fc81fa1e 42#include "qapi/qmp/qstring.h"
e1d74bc6
KW
43#include "qapi/qobject-output-visitor.h"
44#include "qapi/qapi-visit-block-core.h"
bfb197e0 45#include "sysemu/block-backend.h"
1de7afc9 46#include "qemu/notify.h"
922a01a0 47#include "qemu/option.h"
10817bf0 48#include "qemu/coroutine.h"
c13163fb 49#include "block/qapi.h"
1de7afc9 50#include "qemu/timer.h"
f348b6d1
VB
51#include "qemu/cutils.h"
52#include "qemu/id.h"
0bc329fb
HR
53#include "qemu/range.h"
54#include "qemu/rcu.h"
21c2283e 55#include "block/coroutines.h"
fc01f7e7 56
71e72a19 57#ifdef CONFIG_BSD
7674e7bf 58#include <sys/ioctl.h>
72cf2d4f 59#include <sys/queue.h>
feccdcee 60#if defined(HAVE_SYS_DISK_H)
7674e7bf
FB
61#include <sys/disk.h>
62#endif
c5e97233 63#endif
7674e7bf 64
49dc768d
AL
65#ifdef _WIN32
66#include <windows.h>
67#endif
68
1c9805a3
SH
69#define NOT_DONE 0x7fffffff /* used while emulated sync operation in progress */
70
3b491a90 71/* Protected by BQL */
dc364f4c
BC
72static QTAILQ_HEAD(, BlockDriverState) graph_bdrv_states =
73 QTAILQ_HEAD_INITIALIZER(graph_bdrv_states);
74
3b491a90 75/* Protected by BQL */
2c1d04e0
HR
76static QTAILQ_HEAD(, BlockDriverState) all_bdrv_states =
77 QTAILQ_HEAD_INITIALIZER(all_bdrv_states);
78
3b491a90 79/* Protected by BQL */
8a22f02a
SH
80static QLIST_HEAD(, BlockDriver) bdrv_drivers =
81 QLIST_HEAD_INITIALIZER(bdrv_drivers);
ea2384d3 82
5b363937
HR
83static BlockDriverState *bdrv_open_inherit(const char *filename,
84 const char *reference,
85 QDict *options, int flags,
86 BlockDriverState *parent,
bd86fb99 87 const BdrvChildClass *child_class,
272c02ea 88 BdrvChildRole child_role,
5b363937 89 Error **errp);
f3930ed0 90
bfb8aa6d
KW
91static bool bdrv_recurse_has_child(BlockDriverState *bs,
92 BlockDriverState *child);
93
ad29eb3d
KW
94static void GRAPH_WRLOCK
95bdrv_replace_child_noperm(BdrvChild *child, BlockDriverState *new_bs);
96
2f64e1fc
KW
97static void GRAPH_WRLOCK
98bdrv_remove_child(BdrvChild *child, Transaction *tran);
0978623e 99
72373e40
VSO
100static int bdrv_reopen_prepare(BDRVReopenState *reopen_state,
101 BlockReopenQueue *queue,
ecd30d2d 102 Transaction *change_child_tran, Error **errp);
53e96d1e
VSO
103static void bdrv_reopen_commit(BDRVReopenState *reopen_state);
104static void bdrv_reopen_abort(BDRVReopenState *reopen_state);
105
fa8fc1d0
EGE
106static bool bdrv_backing_overridden(BlockDriverState *bs);
107
7e8c182f 108static bool bdrv_change_aio_context(BlockDriverState *bs, AioContext *ctx,
e08cc001 109 GHashTable *visited, Transaction *tran,
7e8c182f
EGE
110 Error **errp);
111
eb852011
MA
112/* If non-zero, use only whitelisted block drivers */
113static int use_bdrv_whitelist;
114
9e0b22f4
SH
115#ifdef _WIN32
116static int is_windows_drive_prefix(const char *filename)
117{
118 return (((filename[0] >= 'a' && filename[0] <= 'z') ||
119 (filename[0] >= 'A' && filename[0] <= 'Z')) &&
120 filename[1] == ':');
121}
122
123int is_windows_drive(const char *filename)
124{
125 if (is_windows_drive_prefix(filename) &&
126 filename[2] == '\0')
127 return 1;
128 if (strstart(filename, "\\\\.\\", NULL) ||
129 strstart(filename, "//./", NULL))
130 return 1;
131 return 0;
132}
133#endif
134
339064d5
KW
135size_t bdrv_opt_mem_align(BlockDriverState *bs)
136{
137 if (!bs || !bs->drv) {
459b4e66 138 /* page size or 4k (hdd sector size) should be on the safe side */
8e3b0cbb 139 return MAX(4096, qemu_real_host_page_size());
339064d5 140 }
384a48fb 141 IO_CODE();
339064d5
KW
142
143 return bs->bl.opt_mem_alignment;
144}
145
4196d2f0
DL
146size_t bdrv_min_mem_align(BlockDriverState *bs)
147{
148 if (!bs || !bs->drv) {
459b4e66 149 /* page size or 4k (hdd sector size) should be on the safe side */
8e3b0cbb 150 return MAX(4096, qemu_real_host_page_size());
4196d2f0 151 }
384a48fb 152 IO_CODE();
4196d2f0
DL
153
154 return bs->bl.min_mem_alignment;
155}
156
9e0b22f4 157/* check if the path starts with "<protocol>:" */
5c98415b 158int path_has_protocol(const char *path)
9e0b22f4 159{
947995c0
PB
160 const char *p;
161
9e0b22f4
SH
162#ifdef _WIN32
163 if (is_windows_drive(path) ||
164 is_windows_drive_prefix(path)) {
165 return 0;
166 }
947995c0
PB
167 p = path + strcspn(path, ":/\\");
168#else
169 p = path + strcspn(path, ":/");
9e0b22f4
SH
170#endif
171
947995c0 172 return *p == ':';
9e0b22f4
SH
173}
174
83f64091 175int path_is_absolute(const char *path)
3b0d4f61 176{
21664424
FB
177#ifdef _WIN32
178 /* specific case for names like: "\\.\d:" */
f53f4da9 179 if (is_windows_drive(path) || is_windows_drive_prefix(path)) {
21664424 180 return 1;
f53f4da9
PB
181 }
182 return (*path == '/' || *path == '\\');
3b9f94e1 183#else
f53f4da9 184 return (*path == '/');
3b9f94e1 185#endif
3b0d4f61
FB
186}
187
009b03aa 188/* if filename is absolute, just return its duplicate. Otherwise, build a
83f64091
FB
189 path to it by considering it is relative to base_path. URL are
190 supported. */
009b03aa 191char *path_combine(const char *base_path, const char *filename)
3b0d4f61 192{
009b03aa 193 const char *protocol_stripped = NULL;
83f64091 194 const char *p, *p1;
009b03aa 195 char *result;
83f64091
FB
196 int len;
197
83f64091 198 if (path_is_absolute(filename)) {
009b03aa
HR
199 return g_strdup(filename);
200 }
0d54a6fe 201
009b03aa
HR
202 if (path_has_protocol(base_path)) {
203 protocol_stripped = strchr(base_path, ':');
204 if (protocol_stripped) {
205 protocol_stripped++;
0d54a6fe 206 }
009b03aa
HR
207 }
208 p = protocol_stripped ?: base_path;
0d54a6fe 209
009b03aa 210 p1 = strrchr(base_path, '/');
3b9f94e1 211#ifdef _WIN32
009b03aa
HR
212 {
213 const char *p2;
214 p2 = strrchr(base_path, '\\');
215 if (!p1 || p2 > p1) {
216 p1 = p2;
3b9f94e1 217 }
009b03aa 218 }
3b9f94e1 219#endif
009b03aa
HR
220 if (p1) {
221 p1++;
222 } else {
223 p1 = base_path;
224 }
225 if (p1 > p) {
226 p = p1;
3b0d4f61 227 }
009b03aa
HR
228 len = p - base_path;
229
230 result = g_malloc(len + strlen(filename) + 1);
231 memcpy(result, base_path, len);
232 strcpy(result + len, filename);
233
234 return result;
235}
236
03c320d8
HR
237/*
238 * Helper function for bdrv_parse_filename() implementations to remove optional
239 * protocol prefixes (especially "file:") from a filename and for putting the
240 * stripped filename into the options QDict if there is such a prefix.
241 */
242void bdrv_parse_filename_strip_prefix(const char *filename, const char *prefix,
243 QDict *options)
244{
245 if (strstart(filename, prefix, &filename)) {
246 /* Stripping the explicit protocol prefix may result in a protocol
247 * prefix being (wrongly) detected (if the filename contains a colon) */
248 if (path_has_protocol(filename)) {
18cf67c5 249 GString *fat_filename;
03c320d8
HR
250
251 /* This means there is some colon before the first slash; therefore,
252 * this cannot be an absolute path */
253 assert(!path_is_absolute(filename));
254
255 /* And we can thus fix the protocol detection issue by prefixing it
256 * by "./" */
18cf67c5
MA
257 fat_filename = g_string_new("./");
258 g_string_append(fat_filename, filename);
03c320d8 259
18cf67c5 260 assert(!path_has_protocol(fat_filename->str));
03c320d8 261
18cf67c5
MA
262 qdict_put(options, "filename",
263 qstring_from_gstring(fat_filename));
03c320d8
HR
264 } else {
265 /* If no protocol prefix was detected, we can use the shortened
266 * filename as-is */
267 qdict_put_str(options, "filename", filename);
268 }
269 }
270}
271
272
9c5e6594
KW
273/* Returns whether the image file is opened as read-only. Note that this can
274 * return false and writing to the image file is still not possible because the
275 * image is inactivated. */
93ed524e
JC
276bool bdrv_is_read_only(BlockDriverState *bs)
277{
384a48fb 278 IO_CODE();
975da073 279 return !(bs->open_flags & BDRV_O_RDWR);
93ed524e
JC
280}
281
4026f1c4
KW
282static int GRAPH_RDLOCK
283bdrv_can_set_read_only(BlockDriverState *bs, bool read_only,
284 bool ignore_allow_rdw, Error **errp)
fe5241bf 285{
384a48fb
EGE
286 IO_CODE();
287
e2b8247a
JC
288 /* Do not set read_only if copy_on_read is enabled */
289 if (bs->copy_on_read && read_only) {
290 error_setg(errp, "Can't set node '%s' to r/o with copy-on-read enabled",
291 bdrv_get_device_or_node_name(bs));
292 return -EINVAL;
293 }
294
d6fcdf06 295 /* Do not clear read_only if it is prohibited */
54a32bfe
KW
296 if (!read_only && !(bs->open_flags & BDRV_O_ALLOW_RDWR) &&
297 !ignore_allow_rdw)
298 {
d6fcdf06
JC
299 error_setg(errp, "Node '%s' is read only",
300 bdrv_get_device_or_node_name(bs));
301 return -EPERM;
302 }
303
45803a03
JC
304 return 0;
305}
306
eaa2410f
KW
307/*
308 * Called by a driver that can only provide a read-only image.
309 *
310 * Returns 0 if the node is already read-only or it could switch the node to
311 * read-only because BDRV_O_AUTO_RDONLY is set.
312 *
313 * Returns -EACCES if the node is read-write and BDRV_O_AUTO_RDONLY is not set
314 * or bdrv_can_set_read_only() forbids making the node read-only. If @errmsg
315 * is not NULL, it is used as the error message for the Error object.
316 */
317int bdrv_apply_auto_read_only(BlockDriverState *bs, const char *errmsg,
318 Error **errp)
45803a03
JC
319{
320 int ret = 0;
384a48fb 321 IO_CODE();
45803a03 322
eaa2410f
KW
323 if (!(bs->open_flags & BDRV_O_RDWR)) {
324 return 0;
325 }
326 if (!(bs->open_flags & BDRV_O_AUTO_RDONLY)) {
327 goto fail;
45803a03
JC
328 }
329
eaa2410f
KW
330 ret = bdrv_can_set_read_only(bs, true, false, NULL);
331 if (ret < 0) {
332 goto fail;
eeae6a59
KW
333 }
334
eaa2410f
KW
335 bs->open_flags &= ~BDRV_O_RDWR;
336
e2b8247a 337 return 0;
eaa2410f
KW
338
339fail:
340 error_setg(errp, "%s", errmsg ?: "Image is read-only");
341 return -EACCES;
fe5241bf
JC
342}
343
645ae7d8
HR
344/*
345 * If @backing is empty, this function returns NULL without setting
346 * @errp. In all other cases, NULL will only be returned with @errp
347 * set.
348 *
349 * Therefore, a return value of NULL without @errp set means that
350 * there is no backing file; if @errp is set, there is one but its
351 * absolute filename cannot be generated.
352 */
353char *bdrv_get_full_backing_filename_from_filename(const char *backed,
354 const char *backing,
355 Error **errp)
dc5a1371 356{
645ae7d8
HR
357 if (backing[0] == '\0') {
358 return NULL;
359 } else if (path_has_protocol(backing) || path_is_absolute(backing)) {
360 return g_strdup(backing);
9f07429e
HR
361 } else if (backed[0] == '\0' || strstart(backed, "json:", NULL)) {
362 error_setg(errp, "Cannot use relative backing file names for '%s'",
363 backed);
645ae7d8 364 return NULL;
dc5a1371 365 } else {
645ae7d8 366 return path_combine(backed, backing);
dc5a1371
PB
367 }
368}
369
9f4793d8
HR
370/*
371 * If @filename is empty or NULL, this function returns NULL without
372 * setting @errp. In all other cases, NULL will only be returned with
373 * @errp set.
374 */
b7cfc7d5
KW
375static char * GRAPH_RDLOCK
376bdrv_make_absolute_filename(BlockDriverState *relative_to,
377 const char *filename, Error **errp)
0a82855a 378{
8df68616 379 char *dir, *full_name;
9f4793d8 380
8df68616
HR
381 if (!filename || filename[0] == '\0') {
382 return NULL;
383 } else if (path_has_protocol(filename) || path_is_absolute(filename)) {
384 return g_strdup(filename);
385 }
9f07429e 386
8df68616
HR
387 dir = bdrv_dirname(relative_to, errp);
388 if (!dir) {
389 return NULL;
390 }
f30c66ba 391
8df68616
HR
392 full_name = g_strconcat(dir, filename, NULL);
393 g_free(dir);
394 return full_name;
9f4793d8
HR
395}
396
397char *bdrv_get_full_backing_filename(BlockDriverState *bs, Error **errp)
398{
f791bf7f 399 GLOBAL_STATE_CODE();
9f4793d8 400 return bdrv_make_absolute_filename(bs, bs->backing_file, errp);
0a82855a
HR
401}
402
0eb7217e
SH
403void bdrv_register(BlockDriver *bdrv)
404{
a15f08dc 405 assert(bdrv->format_name);
f791bf7f 406 GLOBAL_STATE_CODE();
8a22f02a 407 QLIST_INSERT_HEAD(&bdrv_drivers, bdrv, list);
ea2384d3 408}
b338082b 409
e4e9986b
MA
410BlockDriverState *bdrv_new(void)
411{
412 BlockDriverState *bs;
413 int i;
414
f791bf7f
EGE
415 GLOBAL_STATE_CODE();
416
5839e53b 417 bs = g_new0(BlockDriverState, 1);
e4654d2d 418 QLIST_INIT(&bs->dirty_bitmaps);
fbe40ff7
FZ
419 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
420 QLIST_INIT(&bs->op_blockers[i]);
421 }
fa9185fc 422 qemu_mutex_init(&bs->reqs_lock);
2119882c 423 qemu_mutex_init(&bs->dirty_bitmap_mutex);
9fcb0251 424 bs->refcnt = 1;
dcd04228 425 bs->aio_context = qemu_get_aio_context();
d7d512f6 426
3ff2f67a
EY
427 qemu_co_queue_init(&bs->flush_queue);
428
0bc329fb
HR
429 qemu_co_mutex_init(&bs->bsc_modify_lock);
430 bs->block_status_cache = g_new0(BdrvBlockStatusCache, 1);
431
0f12264e
KW
432 for (i = 0; i < bdrv_drain_all_count; i++) {
433 bdrv_drained_begin(bs);
434 }
435
2c1d04e0
HR
436 QTAILQ_INSERT_TAIL(&all_bdrv_states, bs, bs_list);
437
b338082b
FB
438 return bs;
439}
440
88d88798 441static BlockDriver *bdrv_do_find_format(const char *format_name)
ea2384d3
FB
442{
443 BlockDriver *drv1;
bdb73476 444 GLOBAL_STATE_CODE();
88d88798 445
8a22f02a
SH
446 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
447 if (!strcmp(drv1->format_name, format_name)) {
ea2384d3 448 return drv1;
8a22f02a 449 }
ea2384d3 450 }
88d88798 451
ea2384d3
FB
452 return NULL;
453}
454
88d88798
MM
455BlockDriver *bdrv_find_format(const char *format_name)
456{
457 BlockDriver *drv1;
458 int i;
459
f791bf7f
EGE
460 GLOBAL_STATE_CODE();
461
88d88798
MM
462 drv1 = bdrv_do_find_format(format_name);
463 if (drv1) {
464 return drv1;
465 }
466
467 /* The driver isn't registered, maybe we need to load a module */
468 for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); ++i) {
469 if (!strcmp(block_driver_modules[i].format_name, format_name)) {
c551fb0b
CF
470 Error *local_err = NULL;
471 int rv = block_module_load(block_driver_modules[i].library_name,
472 &local_err);
473 if (rv > 0) {
474 return bdrv_do_find_format(format_name);
475 } else if (rv < 0) {
476 error_report_err(local_err);
477 }
88d88798
MM
478 break;
479 }
480 }
c551fb0b 481 return NULL;
88d88798
MM
482}
483
9ac404c5 484static int bdrv_format_is_whitelisted(const char *format_name, bool read_only)
eb852011 485{
b64ec4e4
FZ
486 static const char *whitelist_rw[] = {
487 CONFIG_BDRV_RW_WHITELIST
859aef02 488 NULL
b64ec4e4
FZ
489 };
490 static const char *whitelist_ro[] = {
491 CONFIG_BDRV_RO_WHITELIST
859aef02 492 NULL
eb852011
MA
493 };
494 const char **p;
495
b64ec4e4 496 if (!whitelist_rw[0] && !whitelist_ro[0]) {
eb852011 497 return 1; /* no whitelist, anything goes */
b64ec4e4 498 }
eb852011 499
b64ec4e4 500 for (p = whitelist_rw; *p; p++) {
9ac404c5 501 if (!strcmp(format_name, *p)) {
eb852011
MA
502 return 1;
503 }
504 }
b64ec4e4
FZ
505 if (read_only) {
506 for (p = whitelist_ro; *p; p++) {
9ac404c5 507 if (!strcmp(format_name, *p)) {
b64ec4e4
FZ
508 return 1;
509 }
510 }
511 }
eb852011
MA
512 return 0;
513}
514
9ac404c5
AS
515int bdrv_is_whitelisted(BlockDriver *drv, bool read_only)
516{
f791bf7f 517 GLOBAL_STATE_CODE();
9ac404c5
AS
518 return bdrv_format_is_whitelisted(drv->format_name, read_only);
519}
520
e6ff69bf
DB
521bool bdrv_uses_whitelist(void)
522{
523 return use_bdrv_whitelist;
524}
525
5b7e1542
ZYW
526typedef struct CreateCo {
527 BlockDriver *drv;
528 char *filename;
83d0521a 529 QemuOpts *opts;
5b7e1542 530 int ret;
cc84d90f 531 Error *err;
5b7e1542
ZYW
532} CreateCo;
533
741443eb
EGE
534int coroutine_fn bdrv_co_create(BlockDriver *drv, const char *filename,
535 QemuOpts *opts, Error **errp)
5b7e1542 536{
cc84d90f 537 int ret;
84bdf21f
EGE
538 GLOBAL_STATE_CODE();
539 ERRP_GUARD();
540
541 if (!drv->bdrv_co_create_opts) {
542 error_setg(errp, "Driver '%s' does not support image creation",
543 drv->format_name);
544 return -ENOTSUP;
545 }
546
547 ret = drv->bdrv_co_create_opts(drv, filename, opts, errp);
548 if (ret < 0 && !*errp) {
549 error_setg_errno(errp, -ret, "Could not create image");
550 }
cc84d90f 551
84bdf21f
EGE
552 return ret;
553}
554
fd17146c
HR
555/**
556 * Helper function for bdrv_create_file_fallback(): Resize @blk to at
557 * least the given @minimum_size.
558 *
559 * On success, return @blk's actual length.
560 * Otherwise, return -errno.
561 */
84569a7d
PB
562static int64_t coroutine_fn GRAPH_UNLOCKED
563create_file_fallback_truncate(BlockBackend *blk, int64_t minimum_size,
564 Error **errp)
84a12e66 565{
cc84d90f 566 Error *local_err = NULL;
fd17146c 567 int64_t size;
cc84d90f 568 int ret;
84a12e66 569
bdb73476
EGE
570 GLOBAL_STATE_CODE();
571
84569a7d
PB
572 ret = blk_co_truncate(blk, minimum_size, false, PREALLOC_MODE_OFF, 0,
573 &local_err);
fd17146c
HR
574 if (ret < 0 && ret != -ENOTSUP) {
575 error_propagate(errp, local_err);
576 return ret;
577 }
578
84569a7d 579 size = blk_co_getlength(blk);
fd17146c
HR
580 if (size < 0) {
581 error_free(local_err);
582 error_setg_errno(errp, -size,
583 "Failed to inquire the new image file's length");
584 return size;
585 }
586
587 if (size < minimum_size) {
588 /* Need to grow the image, but we failed to do that */
589 error_propagate(errp, local_err);
590 return -ENOTSUP;
591 }
592
593 error_free(local_err);
594 local_err = NULL;
595
596 return size;
597}
598
599/**
600 * Helper function for bdrv_create_file_fallback(): Zero the first
601 * sector to remove any potentially pre-existing image header.
602 */
881a4c55
PB
603static int coroutine_fn
604create_file_fallback_zero_first_sector(BlockBackend *blk,
605 int64_t current_size,
606 Error **errp)
fd17146c
HR
607{
608 int64_t bytes_to_clear;
609 int ret;
610
bdb73476
EGE
611 GLOBAL_STATE_CODE();
612
fd17146c
HR
613 bytes_to_clear = MIN(current_size, BDRV_SECTOR_SIZE);
614 if (bytes_to_clear) {
ce47ff20 615 ret = blk_co_pwrite_zeroes(blk, 0, bytes_to_clear, BDRV_REQ_MAY_UNMAP);
fd17146c
HR
616 if (ret < 0) {
617 error_setg_errno(errp, -ret,
618 "Failed to clear the new image's first sector");
619 return ret;
620 }
621 }
622
623 return 0;
624}
625
5a5e7f8c
ML
626/**
627 * Simple implementation of bdrv_co_create_opts for protocol drivers
628 * which only support creation via opening a file
629 * (usually existing raw storage device)
630 */
631int coroutine_fn bdrv_co_create_opts_simple(BlockDriver *drv,
632 const char *filename,
633 QemuOpts *opts,
634 Error **errp)
fd17146c
HR
635{
636 BlockBackend *blk;
eeea1faa 637 QDict *options;
fd17146c
HR
638 int64_t size = 0;
639 char *buf = NULL;
640 PreallocMode prealloc;
641 Error *local_err = NULL;
642 int ret;
643
b4ad82aa
EGE
644 GLOBAL_STATE_CODE();
645
fd17146c
HR
646 size = qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0);
647 buf = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC);
648 prealloc = qapi_enum_parse(&PreallocMode_lookup, buf,
649 PREALLOC_MODE_OFF, &local_err);
650 g_free(buf);
651 if (local_err) {
652 error_propagate(errp, local_err);
653 return -EINVAL;
654 }
655
656 if (prealloc != PREALLOC_MODE_OFF) {
657 error_setg(errp, "Unsupported preallocation mode '%s'",
658 PreallocMode_str(prealloc));
659 return -ENOTSUP;
660 }
661
eeea1faa 662 options = qdict_new();
fd17146c
HR
663 qdict_put_str(options, "driver", drv->format_name);
664
be1a732c
KW
665 blk = blk_co_new_open(filename, NULL, options,
666 BDRV_O_RDWR | BDRV_O_RESIZE, errp);
fd17146c 667 if (!blk) {
81624867
HC
668 error_prepend(errp, "Protocol driver '%s' does not support creating "
669 "new images, so an existing image must be selected as "
670 "the target; however, opening the given target as an "
671 "existing image failed: ",
fd17146c
HR
672 drv->format_name);
673 return -EINVAL;
674 }
675
676 size = create_file_fallback_truncate(blk, size, errp);
677 if (size < 0) {
678 ret = size;
679 goto out;
680 }
681
682 ret = create_file_fallback_zero_first_sector(blk, size, errp);
683 if (ret < 0) {
684 goto out;
685 }
686
687 ret = 0;
688out:
b2ab5f54 689 blk_co_unref(blk);
fd17146c
HR
690 return ret;
691}
692
2475a0d0
EGE
693int coroutine_fn bdrv_co_create_file(const char *filename, QemuOpts *opts,
694 Error **errp)
fd17146c 695{
729222af 696 QemuOpts *protocol_opts;
fd17146c 697 BlockDriver *drv;
729222af
SG
698 QDict *qdict;
699 int ret;
fd17146c 700
f791bf7f
EGE
701 GLOBAL_STATE_CODE();
702
b65a5e12 703 drv = bdrv_find_protocol(filename, true, errp);
84a12e66 704 if (drv == NULL) {
16905d71 705 return -ENOENT;
84a12e66
CH
706 }
707
729222af
SG
708 if (!drv->create_opts) {
709 error_setg(errp, "Driver '%s' does not support image creation",
710 drv->format_name);
711 return -ENOTSUP;
712 }
713
714 /*
715 * 'opts' contains a QemuOptsList with a combination of format and protocol
716 * default values.
717 *
718 * The format properly removes its options, but the default values remain
719 * in 'opts->list'. So if the protocol has options with the same name
720 * (e.g. rbd has 'cluster_size' as qcow2), it will see the default values
721 * of the format, since for overlapping options, the format wins.
722 *
723 * To avoid this issue, lets convert QemuOpts to QDict, in this way we take
724 * only the set options, and then convert it back to QemuOpts, using the
725 * create_opts of the protocol. So the new QemuOpts, will contain only the
726 * protocol defaults.
727 */
728 qdict = qemu_opts_to_qdict(opts, NULL);
729 protocol_opts = qemu_opts_from_qdict(drv->create_opts, qdict, errp);
730 if (protocol_opts == NULL) {
731 ret = -EINVAL;
732 goto out;
733 }
734
2475a0d0 735 ret = bdrv_co_create(drv, filename, protocol_opts, errp);
729222af
SG
736out:
737 qemu_opts_del(protocol_opts);
738 qobject_unref(qdict);
739 return ret;
84a12e66
CH
740}
741
e1d7f8bb
DHB
742int coroutine_fn bdrv_co_delete_file(BlockDriverState *bs, Error **errp)
743{
744 Error *local_err = NULL;
745 int ret;
746
384a48fb 747 IO_CODE();
e1d7f8bb 748 assert(bs != NULL);
48aef794 749 assert_bdrv_graph_readable();
e1d7f8bb
DHB
750
751 if (!bs->drv) {
752 error_setg(errp, "Block node '%s' is not opened", bs->filename);
753 return -ENOMEDIUM;
754 }
755
756 if (!bs->drv->bdrv_co_delete_file) {
757 error_setg(errp, "Driver '%s' does not support image deletion",
758 bs->drv->format_name);
759 return -ENOTSUP;
760 }
761
762 ret = bs->drv->bdrv_co_delete_file(bs, &local_err);
763 if (ret < 0) {
764 error_propagate(errp, local_err);
765 }
766
767 return ret;
768}
769
a890f08e
ML
770void coroutine_fn bdrv_co_delete_file_noerr(BlockDriverState *bs)
771{
772 Error *local_err = NULL;
773 int ret;
384a48fb 774 IO_CODE();
a890f08e
ML
775
776 if (!bs) {
777 return;
778 }
779
780 ret = bdrv_co_delete_file(bs, &local_err);
781 /*
782 * ENOTSUP will happen if the block driver doesn't support
783 * the 'bdrv_co_delete_file' interface. This is a predictable
784 * scenario and shouldn't be reported back to the user.
785 */
786 if (ret == -ENOTSUP) {
787 error_free(local_err);
788 } else if (ret < 0) {
789 error_report_err(local_err);
790 }
791}
792
892b7de8
ET
793/**
794 * Try to get @bs's logical and physical block size.
795 * On success, store them in @bsz struct and return 0.
796 * On failure return -errno.
797 * @bs must not be empty.
798 */
799int bdrv_probe_blocksizes(BlockDriverState *bs, BlockSizes *bsz)
800{
801 BlockDriver *drv = bs->drv;
93393e69 802 BlockDriverState *filtered = bdrv_filter_bs(bs);
f791bf7f 803 GLOBAL_STATE_CODE();
892b7de8
ET
804
805 if (drv && drv->bdrv_probe_blocksizes) {
806 return drv->bdrv_probe_blocksizes(bs, bsz);
93393e69
HR
807 } else if (filtered) {
808 return bdrv_probe_blocksizes(filtered, bsz);
892b7de8
ET
809 }
810
811 return -ENOTSUP;
812}
813
814/**
815 * Try to get @bs's geometry (cyls, heads, sectors).
816 * On success, store them in @geo struct and return 0.
817 * On failure return -errno.
818 * @bs must not be empty.
819 */
820int bdrv_probe_geometry(BlockDriverState *bs, HDGeometry *geo)
821{
822 BlockDriver *drv = bs->drv;
f5a3a270
KW
823 BlockDriverState *filtered;
824
f791bf7f 825 GLOBAL_STATE_CODE();
f5a3a270 826 GRAPH_RDLOCK_GUARD_MAINLOOP();
892b7de8
ET
827
828 if (drv && drv->bdrv_probe_geometry) {
829 return drv->bdrv_probe_geometry(bs, geo);
f5a3a270
KW
830 }
831
832 filtered = bdrv_filter_bs(bs);
833 if (filtered) {
93393e69 834 return bdrv_probe_geometry(filtered, geo);
892b7de8
ET
835 }
836
837 return -ENOTSUP;
838}
839
eba25057
JM
840/*
841 * Create a uniquely-named empty temporary file.
69fbfff9
BM
842 * Return the actual file name used upon success, otherwise NULL.
843 * This string should be freed with g_free() when not needed any longer.
844 *
845 * Note: creating a temporary file for the caller to (re)open is
846 * inherently racy. Use g_file_open_tmp() instead whenever practical.
eba25057 847 */
69fbfff9 848char *create_tmp_file(Error **errp)
d5249393 849{
67b915a5 850 int fd;
7ccfb2eb 851 const char *tmpdir;
69fbfff9
BM
852 g_autofree char *filename = NULL;
853
854 tmpdir = g_get_tmp_dir();
855#ifndef _WIN32
856 /*
857 * See commit 69bef79 ("block: use /var/tmp instead of /tmp for -snapshot")
858 *
859 * This function is used to create temporary disk images (like -snapshot),
860 * so the files can become very large. /tmp is often a tmpfs where as
861 * /var/tmp is usually on a disk, so more appropriate for disk images.
862 */
863 if (!g_strcmp0(tmpdir, "/tmp")) {
69bef793
AS
864 tmpdir = "/var/tmp";
865 }
69fbfff9
BM
866#endif
867
868 filename = g_strdup_printf("%s/vl.XXXXXX", tmpdir);
869 fd = g_mkstemp(filename);
fe235a06 870 if (fd < 0) {
69fbfff9
BM
871 error_setg_errno(errp, errno, "Could not open temporary file '%s'",
872 filename);
873 return NULL;
eba25057 874 }
6b6471ee 875 close(fd);
69fbfff9
BM
876
877 return g_steal_pointer(&filename);
eba25057 878}
fc01f7e7 879
84a12e66
CH
880/*
881 * Detect host devices. By convention, /dev/cdrom[N] is always
882 * recognized as a host CDROM.
883 */
884static BlockDriver *find_hdev_driver(const char *filename)
885{
886 int score_max = 0, score;
887 BlockDriver *drv = NULL, *d;
bdb73476 888 GLOBAL_STATE_CODE();
84a12e66
CH
889
890 QLIST_FOREACH(d, &bdrv_drivers, list) {
891 if (d->bdrv_probe_device) {
892 score = d->bdrv_probe_device(filename);
893 if (score > score_max) {
894 score_max = score;
895 drv = d;
896 }
897 }
898 }
899
900 return drv;
901}
902
88d88798
MM
903static BlockDriver *bdrv_do_find_protocol(const char *protocol)
904{
905 BlockDriver *drv1;
bdb73476 906 GLOBAL_STATE_CODE();
88d88798
MM
907
908 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
909 if (drv1->protocol_name && !strcmp(drv1->protocol_name, protocol)) {
910 return drv1;
911 }
912 }
913
914 return NULL;
915}
916
98289620 917BlockDriver *bdrv_find_protocol(const char *filename,
b65a5e12
HR
918 bool allow_protocol_prefix,
919 Error **errp)
83f64091
FB
920{
921 BlockDriver *drv1;
922 char protocol[128];
1cec71e3 923 int len;
83f64091 924 const char *p;
88d88798 925 int i;
19cb3738 926
f791bf7f 927 GLOBAL_STATE_CODE();
66f82cee
KW
928 /* TODO Drivers without bdrv_file_open must be specified explicitly */
929
39508e7a
CH
930 /*
931 * XXX(hch): we really should not let host device detection
932 * override an explicit protocol specification, but moving this
933 * later breaks access to device names with colons in them.
934 * Thanks to the brain-dead persistent naming schemes on udev-
935 * based Linux systems those actually are quite common.
936 */
937 drv1 = find_hdev_driver(filename);
938 if (drv1) {
939 return drv1;
940 }
941
98289620 942 if (!path_has_protocol(filename) || !allow_protocol_prefix) {
ef810437 943 return &bdrv_file;
84a12e66 944 }
98289620 945
9e0b22f4
SH
946 p = strchr(filename, ':');
947 assert(p != NULL);
1cec71e3
AL
948 len = p - filename;
949 if (len > sizeof(protocol) - 1)
950 len = sizeof(protocol) - 1;
951 memcpy(protocol, filename, len);
952 protocol[len] = '\0';
88d88798
MM
953
954 drv1 = bdrv_do_find_protocol(protocol);
955 if (drv1) {
956 return drv1;
957 }
958
959 for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); ++i) {
960 if (block_driver_modules[i].protocol_name &&
961 !strcmp(block_driver_modules[i].protocol_name, protocol)) {
c551fb0b
CF
962 int rv = block_module_load(block_driver_modules[i].library_name, errp);
963 if (rv > 0) {
964 drv1 = bdrv_do_find_protocol(protocol);
965 } else if (rv < 0) {
966 return NULL;
967 }
88d88798 968 break;
8a22f02a 969 }
83f64091 970 }
b65a5e12 971
88d88798
MM
972 if (!drv1) {
973 error_setg(errp, "Unknown protocol '%s'", protocol);
974 }
975 return drv1;
83f64091
FB
976}
977
c6684249
MA
978/*
979 * Guess image format by probing its contents.
980 * This is not a good idea when your image is raw (CVE-2008-2004), but
981 * we do it anyway for backward compatibility.
982 *
983 * @buf contains the image's first @buf_size bytes.
7cddd372
KW
984 * @buf_size is the buffer size in bytes (generally BLOCK_PROBE_BUF_SIZE,
985 * but can be smaller if the image file is smaller)
c6684249
MA
986 * @filename is its filename.
987 *
988 * For all block drivers, call the bdrv_probe() method to get its
989 * probing score.
990 * Return the first block driver with the highest probing score.
991 */
38f3ef57
KW
992BlockDriver *bdrv_probe_all(const uint8_t *buf, int buf_size,
993 const char *filename)
c6684249
MA
994{
995 int score_max = 0, score;
996 BlockDriver *drv = NULL, *d;
967d7905 997 IO_CODE();
c6684249
MA
998
999 QLIST_FOREACH(d, &bdrv_drivers, list) {
1000 if (d->bdrv_probe) {
1001 score = d->bdrv_probe(buf, buf_size, filename);
1002 if (score > score_max) {
1003 score_max = score;
1004 drv = d;
1005 }
1006 }
1007 }
1008
1009 return drv;
1010}
1011
5696c6e3 1012static int find_image_format(BlockBackend *file, const char *filename,
34b5d2c6 1013 BlockDriver **pdrv, Error **errp)
f3a5d3f8 1014{
c6684249 1015 BlockDriver *drv;
7cddd372 1016 uint8_t buf[BLOCK_PROBE_BUF_SIZE];
f500a6d3 1017 int ret = 0;
f8ea0b00 1018
bdb73476
EGE
1019 GLOBAL_STATE_CODE();
1020
08a00559 1021 /* Return the raw BlockDriver * to scsi-generic devices or empty drives */
5696c6e3 1022 if (blk_is_sg(file) || !blk_is_inserted(file) || blk_getlength(file) == 0) {
ef810437 1023 *pdrv = &bdrv_raw;
c98ac35d 1024 return ret;
1a396859 1025 }
f8ea0b00 1026
a9262f55 1027 ret = blk_pread(file, 0, sizeof(buf), buf, 0);
83f64091 1028 if (ret < 0) {
34b5d2c6
HR
1029 error_setg_errno(errp, -ret, "Could not read image for determining its "
1030 "format");
c98ac35d
SW
1031 *pdrv = NULL;
1032 return ret;
83f64091
FB
1033 }
1034
bf5b16fa 1035 drv = bdrv_probe_all(buf, sizeof(buf), filename);
c98ac35d 1036 if (!drv) {
34b5d2c6
HR
1037 error_setg(errp, "Could not determine image format: No compatible "
1038 "driver found");
bf5b16fa
AF
1039 *pdrv = NULL;
1040 return -ENOENT;
c98ac35d 1041 }
bf5b16fa 1042
c98ac35d 1043 *pdrv = drv;
bf5b16fa 1044 return 0;
ea2384d3
FB
1045}
1046
51762288
SH
1047/**
1048 * Set the current 'total_sectors' value
65a9bb25 1049 * Return 0 on success, -errno on error.
51762288 1050 */
c86422c5
EGE
1051int coroutine_fn bdrv_co_refresh_total_sectors(BlockDriverState *bs,
1052 int64_t hint)
51762288
SH
1053{
1054 BlockDriver *drv = bs->drv;
967d7905 1055 IO_CODE();
8ab8140a 1056 assert_bdrv_graph_readable();
51762288 1057
d470ad42
HR
1058 if (!drv) {
1059 return -ENOMEDIUM;
1060 }
1061
c86422c5 1062 /* Do not attempt drv->bdrv_co_getlength() on scsi-generic devices */
b192af8a 1063 if (bdrv_is_sg(bs))
396759ad
NB
1064 return 0;
1065
51762288 1066 /* query actual device if possible, otherwise just trust the hint */
c86422c5
EGE
1067 if (drv->bdrv_co_getlength) {
1068 int64_t length = drv->bdrv_co_getlength(bs);
51762288
SH
1069 if (length < 0) {
1070 return length;
1071 }
7e382003 1072 hint = DIV_ROUND_UP(length, BDRV_SECTOR_SIZE);
51762288
SH
1073 }
1074
1075 bs->total_sectors = hint;
8b117001
VSO
1076
1077 if (bs->total_sectors * BDRV_SECTOR_SIZE > BDRV_MAX_LENGTH) {
1078 return -EFBIG;
1079 }
1080
51762288
SH
1081 return 0;
1082}
1083
cddff5ba
KW
1084/**
1085 * Combines a QDict of new block driver @options with any missing options taken
1086 * from @old_options, so that leaving out an option defaults to its old value.
1087 */
1088static void bdrv_join_options(BlockDriverState *bs, QDict *options,
1089 QDict *old_options)
1090{
da359909 1091 GLOBAL_STATE_CODE();
cddff5ba
KW
1092 if (bs->drv && bs->drv->bdrv_join_options) {
1093 bs->drv->bdrv_join_options(options, old_options);
1094 } else {
1095 qdict_join(options, old_options, false);
1096 }
1097}
1098
543770bd
AG
1099static BlockdevDetectZeroesOptions bdrv_parse_detect_zeroes(QemuOpts *opts,
1100 int open_flags,
1101 Error **errp)
1102{
1103 Error *local_err = NULL;
1104 char *value = qemu_opt_get_del(opts, "detect-zeroes");
1105 BlockdevDetectZeroesOptions detect_zeroes =
1106 qapi_enum_parse(&BlockdevDetectZeroesOptions_lookup, value,
1107 BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF, &local_err);
bdb73476 1108 GLOBAL_STATE_CODE();
543770bd
AG
1109 g_free(value);
1110 if (local_err) {
1111 error_propagate(errp, local_err);
1112 return detect_zeroes;
1113 }
1114
1115 if (detect_zeroes == BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP &&
1116 !(open_flags & BDRV_O_UNMAP))
1117 {
1118 error_setg(errp, "setting detect-zeroes to unmap is not allowed "
1119 "without setting discard operation to unmap");
1120 }
1121
1122 return detect_zeroes;
1123}
1124
f80f2673
AM
1125/**
1126 * Set open flags for aio engine
1127 *
1128 * Return 0 on success, -1 if the engine specified is invalid
1129 */
1130int bdrv_parse_aio(const char *mode, int *flags)
1131{
1132 if (!strcmp(mode, "threads")) {
1133 /* do nothing, default */
1134 } else if (!strcmp(mode, "native")) {
1135 *flags |= BDRV_O_NATIVE_AIO;
1136#ifdef CONFIG_LINUX_IO_URING
1137 } else if (!strcmp(mode, "io_uring")) {
1138 *flags |= BDRV_O_IO_URING;
1139#endif
1140 } else {
1141 return -1;
1142 }
1143
1144 return 0;
1145}
1146
9e8f1835
PB
1147/**
1148 * Set open flags for a given discard mode
1149 *
1150 * Return 0 on success, -1 if the discard mode was invalid.
1151 */
1152int bdrv_parse_discard_flags(const char *mode, int *flags)
1153{
1154 *flags &= ~BDRV_O_UNMAP;
1155
1156 if (!strcmp(mode, "off") || !strcmp(mode, "ignore")) {
1157 /* do nothing */
1158 } else if (!strcmp(mode, "on") || !strcmp(mode, "unmap")) {
1159 *flags |= BDRV_O_UNMAP;
1160 } else {
1161 return -1;
1162 }
1163
1164 return 0;
1165}
1166
c3993cdc
SH
1167/**
1168 * Set open flags for a given cache mode
1169 *
1170 * Return 0 on success, -1 if the cache mode was invalid.
1171 */
53e8ae01 1172int bdrv_parse_cache_mode(const char *mode, int *flags, bool *writethrough)
c3993cdc
SH
1173{
1174 *flags &= ~BDRV_O_CACHE_MASK;
1175
1176 if (!strcmp(mode, "off") || !strcmp(mode, "none")) {
53e8ae01
KW
1177 *writethrough = false;
1178 *flags |= BDRV_O_NOCACHE;
92196b2f 1179 } else if (!strcmp(mode, "directsync")) {
53e8ae01 1180 *writethrough = true;
92196b2f 1181 *flags |= BDRV_O_NOCACHE;
c3993cdc 1182 } else if (!strcmp(mode, "writeback")) {
53e8ae01 1183 *writethrough = false;
c3993cdc 1184 } else if (!strcmp(mode, "unsafe")) {
53e8ae01 1185 *writethrough = false;
c3993cdc
SH
1186 *flags |= BDRV_O_NO_FLUSH;
1187 } else if (!strcmp(mode, "writethrough")) {
53e8ae01 1188 *writethrough = true;
c3993cdc
SH
1189 } else {
1190 return -1;
1191 }
1192
1193 return 0;
1194}
1195
b5411555
KW
1196static char *bdrv_child_get_parent_desc(BdrvChild *c)
1197{
1198 BlockDriverState *parent = c->opaque;
2c0a3acb 1199 return g_strdup_printf("node '%s'", bdrv_get_node_name(parent));
b5411555
KW
1200}
1201
d05ab380 1202static void GRAPH_RDLOCK bdrv_child_cb_drained_begin(BdrvChild *child)
20018e12
KW
1203{
1204 BlockDriverState *bs = child->opaque;
a82a3bd1 1205 bdrv_do_drained_begin_quiesce(bs, NULL);
20018e12
KW
1206}
1207
d05ab380 1208static bool GRAPH_RDLOCK bdrv_child_cb_drained_poll(BdrvChild *child)
89bd0305
KW
1209{
1210 BlockDriverState *bs = child->opaque;
299403ae 1211 return bdrv_drain_poll(bs, NULL, false);
89bd0305
KW
1212}
1213
d05ab380 1214static void GRAPH_RDLOCK bdrv_child_cb_drained_end(BdrvChild *child)
20018e12
KW
1215{
1216 BlockDriverState *bs = child->opaque;
2f65df6e 1217 bdrv_drained_end(bs);
20018e12
KW
1218}
1219
38701b6a
KW
1220static int bdrv_child_cb_inactivate(BdrvChild *child)
1221{
1222 BlockDriverState *bs = child->opaque;
bdb73476 1223 GLOBAL_STATE_CODE();
38701b6a
KW
1224 assert(bs->open_flags & BDRV_O_INACTIVE);
1225 return 0;
1226}
1227
27633e74
EGE
1228static bool bdrv_child_cb_change_aio_ctx(BdrvChild *child, AioContext *ctx,
1229 GHashTable *visited, Transaction *tran,
1230 Error **errp)
53a7d041
KW
1231{
1232 BlockDriverState *bs = child->opaque;
27633e74 1233 return bdrv_change_aio_context(bs, ctx, visited, tran, errp);
53a7d041
KW
1234}
1235
b1e6fc08 1236/*
73176bee
KW
1237 * Returns the options and flags that a temporary snapshot should get, based on
1238 * the originally requested flags (the originally requested image will have
1239 * flags like a backing file)
b1e6fc08 1240 */
73176bee
KW
1241static void bdrv_temp_snapshot_options(int *child_flags, QDict *child_options,
1242 int parent_flags, QDict *parent_options)
b1e6fc08 1243{
bdb73476 1244 GLOBAL_STATE_CODE();
73176bee
KW
1245 *child_flags = (parent_flags & ~BDRV_O_SNAPSHOT) | BDRV_O_TEMPORARY;
1246
1247 /* For temporary files, unconditional cache=unsafe is fine */
73176bee
KW
1248 qdict_set_default_str(child_options, BDRV_OPT_CACHE_DIRECT, "off");
1249 qdict_set_default_str(child_options, BDRV_OPT_CACHE_NO_FLUSH, "on");
41869044 1250
3f48686f 1251 /* Copy the read-only and discard options from the parent */
f87a0e29 1252 qdict_copy_default(child_options, parent_options, BDRV_OPT_READ_ONLY);
3f48686f 1253 qdict_copy_default(child_options, parent_options, BDRV_OPT_DISCARD);
f87a0e29 1254
41869044
KW
1255 /* aio=native doesn't work for cache.direct=off, so disable it for the
1256 * temporary snapshot */
1257 *child_flags &= ~BDRV_O_NATIVE_AIO;
b1e6fc08
KW
1258}
1259
b7cfc7d5 1260static void GRAPH_WRLOCK bdrv_backing_attach(BdrvChild *c)
db95dbba
KW
1261{
1262 BlockDriverState *parent = c->opaque;
1263 BlockDriverState *backing_hd = c->bs;
1264
bdb73476 1265 GLOBAL_STATE_CODE();
db95dbba
KW
1266 assert(!parent->backing_blocker);
1267 error_setg(&parent->backing_blocker,
1268 "node is used as backing hd of '%s'",
1269 bdrv_get_device_or_node_name(parent));
1270
f30c66ba
HR
1271 bdrv_refresh_filename(backing_hd);
1272
db95dbba 1273 parent->open_flags &= ~BDRV_O_NO_BACKING;
db95dbba
KW
1274
1275 bdrv_op_block_all(backing_hd, parent->backing_blocker);
1276 /* Otherwise we won't be able to commit or stream */
1277 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_COMMIT_TARGET,
1278 parent->backing_blocker);
1279 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_STREAM,
1280 parent->backing_blocker);
1281 /*
1282 * We do backup in 3 ways:
1283 * 1. drive backup
1284 * The target bs is new opened, and the source is top BDS
1285 * 2. blockdev backup
1286 * Both the source and the target are top BDSes.
1287 * 3. internal backup(used for block replication)
1288 * Both the source and the target are backing file
1289 *
1290 * In case 1 and 2, neither the source nor the target is the backing file.
1291 * In case 3, we will block the top BDS, so there is only one block job
1292 * for the top BDS and its backing chain.
1293 */
1294 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_BACKUP_SOURCE,
1295 parent->backing_blocker);
1296 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_BACKUP_TARGET,
1297 parent->backing_blocker);
ca2f1234 1298}
d736f119 1299
db95dbba
KW
1300static void bdrv_backing_detach(BdrvChild *c)
1301{
1302 BlockDriverState *parent = c->opaque;
1303
bdb73476 1304 GLOBAL_STATE_CODE();
db95dbba
KW
1305 assert(parent->backing_blocker);
1306 bdrv_op_unblock_all(c->bs, parent->backing_blocker);
1307 error_free(parent->backing_blocker);
1308 parent->backing_blocker = NULL;
48e08288 1309}
d736f119 1310
6858eba0
KW
1311static int bdrv_backing_update_filename(BdrvChild *c, BlockDriverState *base,
1312 const char *filename, Error **errp)
1313{
1314 BlockDriverState *parent = c->opaque;
e94d3dba 1315 bool read_only = bdrv_is_read_only(parent);
6858eba0 1316 int ret;
bdb73476 1317 GLOBAL_STATE_CODE();
6858eba0 1318
e94d3dba
AG
1319 if (read_only) {
1320 ret = bdrv_reopen_set_read_only(parent, false, errp);
61f09cea
KW
1321 if (ret < 0) {
1322 return ret;
1323 }
1324 }
1325
6858eba0 1326 ret = bdrv_change_backing_file(parent, filename,
e54ee1b3
EB
1327 base->drv ? base->drv->format_name : "",
1328 false);
6858eba0 1329 if (ret < 0) {
64730694 1330 error_setg_errno(errp, -ret, "Could not update backing file link");
6858eba0
KW
1331 }
1332
e94d3dba
AG
1333 if (read_only) {
1334 bdrv_reopen_set_read_only(parent, true, NULL);
61f09cea
KW
1335 }
1336
6858eba0
KW
1337 return ret;
1338}
1339
fae8bd39
HR
1340/*
1341 * Returns the options and flags that a generic child of a BDS should
1342 * get, based on the given options and flags for the parent BDS.
1343 */
00ff7ffd
HR
1344static void bdrv_inherited_options(BdrvChildRole role, bool parent_is_format,
1345 int *child_flags, QDict *child_options,
1346 int parent_flags, QDict *parent_options)
fae8bd39
HR
1347{
1348 int flags = parent_flags;
bdb73476 1349 GLOBAL_STATE_CODE();
fae8bd39
HR
1350
1351 /*
1352 * First, decide whether to set, clear, or leave BDRV_O_PROTOCOL.
1353 * Generally, the question to answer is: Should this child be
1354 * format-probed by default?
1355 */
1356
1357 /*
1358 * Pure and non-filtered data children of non-format nodes should
1359 * be probed by default (even when the node itself has BDRV_O_PROTOCOL
1360 * set). This only affects a very limited set of drivers (namely
1361 * quorum and blkverify when this comment was written).
1362 * Force-clear BDRV_O_PROTOCOL then.
1363 */
1364 if (!parent_is_format &&
1365 (role & BDRV_CHILD_DATA) &&
1366 !(role & (BDRV_CHILD_METADATA | BDRV_CHILD_FILTERED)))
1367 {
1368 flags &= ~BDRV_O_PROTOCOL;
1369 }
1370
1371 /*
1372 * All children of format nodes (except for COW children) and all
1373 * metadata children in general should never be format-probed.
1374 * Force-set BDRV_O_PROTOCOL then.
1375 */
1376 if ((parent_is_format && !(role & BDRV_CHILD_COW)) ||
1377 (role & BDRV_CHILD_METADATA))
1378 {
1379 flags |= BDRV_O_PROTOCOL;
1380 }
1381
1382 /*
1383 * If the cache mode isn't explicitly set, inherit direct and no-flush from
1384 * the parent.
1385 */
1386 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_DIRECT);
1387 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_NO_FLUSH);
1388 qdict_copy_default(child_options, parent_options, BDRV_OPT_FORCE_SHARE);
1389
1390 if (role & BDRV_CHILD_COW) {
1391 /* backing files are opened read-only by default */
1392 qdict_set_default_str(child_options, BDRV_OPT_READ_ONLY, "on");
1393 qdict_set_default_str(child_options, BDRV_OPT_AUTO_READ_ONLY, "off");
1394 } else {
1395 /* Inherit the read-only option from the parent if it's not set */
1396 qdict_copy_default(child_options, parent_options, BDRV_OPT_READ_ONLY);
1397 qdict_copy_default(child_options, parent_options,
1398 BDRV_OPT_AUTO_READ_ONLY);
1399 }
1400
1401 /*
1402 * bdrv_co_pdiscard() respects unmap policy for the parent, so we
1403 * can default to enable it on lower layers regardless of the
1404 * parent option.
1405 */
1406 qdict_set_default_str(child_options, BDRV_OPT_DISCARD, "unmap");
1407
1408 /* Clear flags that only apply to the top layer */
1409 flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_COPY_ON_READ);
1410
1411 if (role & BDRV_CHILD_METADATA) {
1412 flags &= ~BDRV_O_NO_IO;
1413 }
1414 if (role & BDRV_CHILD_COW) {
1415 flags &= ~BDRV_O_TEMPORARY;
1416 }
1417
1418 *child_flags = flags;
1419}
1420
303de47b 1421static void GRAPH_WRLOCK bdrv_child_cb_attach(BdrvChild *child)
ca2f1234
HR
1422{
1423 BlockDriverState *bs = child->opaque;
1424
3f35f82e 1425 assert_bdrv_graph_writable();
a225369b 1426 QLIST_INSERT_HEAD(&bs->children, child, next);
5bb04747
VSO
1427 if (bs->drv->is_filter || (child->role & BDRV_CHILD_FILTERED)) {
1428 /*
1429 * Here we handle filters and block/raw-format.c when it behave like
1430 * filter. They generally have a single PRIMARY child, which is also the
1431 * FILTERED child, and that they may have multiple more children, which
1432 * are neither PRIMARY nor FILTERED. And never we have a COW child here.
1433 * So bs->file will be the PRIMARY child, unless the PRIMARY child goes
1434 * into bs->backing on exceptional cases; and bs->backing will be
1435 * nothing else.
1436 */
1437 assert(!(child->role & BDRV_CHILD_COW));
1438 if (child->role & BDRV_CHILD_PRIMARY) {
1439 assert(child->role & BDRV_CHILD_FILTERED);
1440 assert(!bs->backing);
1441 assert(!bs->file);
1442
1443 if (bs->drv->filtered_child_is_backing) {
1444 bs->backing = child;
1445 } else {
1446 bs->file = child;
1447 }
1448 } else {
1449 assert(!(child->role & BDRV_CHILD_FILTERED));
1450 }
1451 } else if (child->role & BDRV_CHILD_COW) {
1452 assert(bs->drv->supports_backing);
1453 assert(!(child->role & BDRV_CHILD_PRIMARY));
1454 assert(!bs->backing);
1455 bs->backing = child;
ca2f1234 1456 bdrv_backing_attach(child);
5bb04747
VSO
1457 } else if (child->role & BDRV_CHILD_PRIMARY) {
1458 assert(!bs->file);
1459 bs->file = child;
ca2f1234 1460 }
ca2f1234
HR
1461}
1462
303de47b 1463static void GRAPH_WRLOCK bdrv_child_cb_detach(BdrvChild *child)
48e08288
HR
1464{
1465 BlockDriverState *bs = child->opaque;
1466
1467 if (child->role & BDRV_CHILD_COW) {
1468 bdrv_backing_detach(child);
1469 }
1470
3f35f82e 1471 assert_bdrv_graph_writable();
a225369b 1472 QLIST_REMOVE(child, next);
5bb04747
VSO
1473 if (child == bs->backing) {
1474 assert(child != bs->file);
1475 bs->backing = NULL;
1476 } else if (child == bs->file) {
1477 bs->file = NULL;
1478 }
48e08288
HR
1479}
1480
43483550
HR
1481static int bdrv_child_cb_update_filename(BdrvChild *c, BlockDriverState *base,
1482 const char *filename, Error **errp)
1483{
1484 if (c->role & BDRV_CHILD_COW) {
1485 return bdrv_backing_update_filename(c, base, filename, errp);
1486 }
1487 return 0;
1488}
1489
fb62b588 1490AioContext *child_of_bds_get_parent_aio_context(BdrvChild *c)
3ca1f322
VSO
1491{
1492 BlockDriverState *bs = c->opaque;
384a48fb 1493 IO_CODE();
3ca1f322
VSO
1494
1495 return bdrv_get_aio_context(bs);
1496}
1497
43483550
HR
1498const BdrvChildClass child_of_bds = {
1499 .parent_is_bds = true,
1500 .get_parent_desc = bdrv_child_get_parent_desc,
1501 .inherit_options = bdrv_inherited_options,
1502 .drained_begin = bdrv_child_cb_drained_begin,
1503 .drained_poll = bdrv_child_cb_drained_poll,
1504 .drained_end = bdrv_child_cb_drained_end,
1505 .attach = bdrv_child_cb_attach,
1506 .detach = bdrv_child_cb_detach,
1507 .inactivate = bdrv_child_cb_inactivate,
27633e74 1508 .change_aio_ctx = bdrv_child_cb_change_aio_ctx,
43483550 1509 .update_filename = bdrv_child_cb_update_filename,
fb62b588 1510 .get_parent_aio_context = child_of_bds_get_parent_aio_context,
43483550
HR
1511};
1512
3ca1f322
VSO
1513AioContext *bdrv_child_get_parent_aio_context(BdrvChild *c)
1514{
d5f8d79c 1515 IO_CODE();
3ca1f322
VSO
1516 return c->klass->get_parent_aio_context(c);
1517}
1518
7b272452
KW
1519static int bdrv_open_flags(BlockDriverState *bs, int flags)
1520{
61de4c68 1521 int open_flags = flags;
bdb73476 1522 GLOBAL_STATE_CODE();
7b272452
KW
1523
1524 /*
1525 * Clear flags that are internal to the block layer before opening the
1526 * image.
1527 */
20cca275 1528 open_flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_PROTOCOL);
7b272452 1529
7b272452
KW
1530 return open_flags;
1531}
1532
91a097e7
KW
1533static void update_flags_from_options(int *flags, QemuOpts *opts)
1534{
bdb73476
EGE
1535 GLOBAL_STATE_CODE();
1536
2a3d4331 1537 *flags &= ~(BDRV_O_CACHE_MASK | BDRV_O_RDWR | BDRV_O_AUTO_RDONLY);
91a097e7 1538
57f9db9a 1539 if (qemu_opt_get_bool_del(opts, BDRV_OPT_CACHE_NO_FLUSH, false)) {
91a097e7
KW
1540 *flags |= BDRV_O_NO_FLUSH;
1541 }
1542
57f9db9a 1543 if (qemu_opt_get_bool_del(opts, BDRV_OPT_CACHE_DIRECT, false)) {
91a097e7
KW
1544 *flags |= BDRV_O_NOCACHE;
1545 }
f87a0e29 1546
57f9db9a 1547 if (!qemu_opt_get_bool_del(opts, BDRV_OPT_READ_ONLY, false)) {
f87a0e29
AG
1548 *flags |= BDRV_O_RDWR;
1549 }
1550
e35bdc12
KW
1551 if (qemu_opt_get_bool_del(opts, BDRV_OPT_AUTO_READ_ONLY, false)) {
1552 *flags |= BDRV_O_AUTO_RDONLY;
1553 }
91a097e7
KW
1554}
1555
1556static void update_options_from_flags(QDict *options, int flags)
1557{
bdb73476 1558 GLOBAL_STATE_CODE();
91a097e7 1559 if (!qdict_haskey(options, BDRV_OPT_CACHE_DIRECT)) {
46f5ac20 1560 qdict_put_bool(options, BDRV_OPT_CACHE_DIRECT, flags & BDRV_O_NOCACHE);
91a097e7
KW
1561 }
1562 if (!qdict_haskey(options, BDRV_OPT_CACHE_NO_FLUSH)) {
46f5ac20
EB
1563 qdict_put_bool(options, BDRV_OPT_CACHE_NO_FLUSH,
1564 flags & BDRV_O_NO_FLUSH);
91a097e7 1565 }
f87a0e29 1566 if (!qdict_haskey(options, BDRV_OPT_READ_ONLY)) {
46f5ac20 1567 qdict_put_bool(options, BDRV_OPT_READ_ONLY, !(flags & BDRV_O_RDWR));
f87a0e29 1568 }
e35bdc12
KW
1569 if (!qdict_haskey(options, BDRV_OPT_AUTO_READ_ONLY)) {
1570 qdict_put_bool(options, BDRV_OPT_AUTO_READ_ONLY,
1571 flags & BDRV_O_AUTO_RDONLY);
1572 }
91a097e7
KW
1573}
1574
636ea370
KW
1575static void bdrv_assign_node_name(BlockDriverState *bs,
1576 const char *node_name,
1577 Error **errp)
6913c0c2 1578{
15489c76 1579 char *gen_node_name = NULL;
bdb73476 1580 GLOBAL_STATE_CODE();
6913c0c2 1581
15489c76
JC
1582 if (!node_name) {
1583 node_name = gen_node_name = id_generate(ID_BLOCK);
1584 } else if (!id_wellformed(node_name)) {
1585 /*
1586 * Check for empty string or invalid characters, but not if it is
1587 * generated (generated names use characters not available to the user)
1588 */
785ec4b1 1589 error_setg(errp, "Invalid node-name: '%s'", node_name);
636ea370 1590 return;
6913c0c2
BC
1591 }
1592
0c5e94ee 1593 /* takes care of avoiding namespaces collisions */
7f06d47e 1594 if (blk_by_name(node_name)) {
0c5e94ee
BC
1595 error_setg(errp, "node-name=%s is conflicting with a device id",
1596 node_name);
15489c76 1597 goto out;
0c5e94ee
BC
1598 }
1599
6913c0c2
BC
1600 /* takes care of avoiding duplicates node names */
1601 if (bdrv_find_node(node_name)) {
785ec4b1 1602 error_setg(errp, "Duplicate nodes with node-name='%s'", node_name);
15489c76 1603 goto out;
6913c0c2
BC
1604 }
1605
824808dd
KW
1606 /* Make sure that the node name isn't truncated */
1607 if (strlen(node_name) >= sizeof(bs->node_name)) {
1608 error_setg(errp, "Node name too long");
1609 goto out;
1610 }
1611
6913c0c2
BC
1612 /* copy node name into the bs and insert it into the graph list */
1613 pstrcpy(bs->node_name, sizeof(bs->node_name), node_name);
1614 QTAILQ_INSERT_TAIL(&graph_bdrv_states, bs, node_list);
15489c76
JC
1615out:
1616 g_free(gen_node_name);
6913c0c2
BC
1617}
1618
c86422c5
EGE
1619/*
1620 * The caller must always hold @bs AioContext lock, because this function calls
1621 * bdrv_refresh_total_sectors() which polls when called from non-coroutine
1622 * context.
1623 */
1a30b0f5
KW
1624static int no_coroutine_fn GRAPH_UNLOCKED
1625bdrv_open_driver(BlockDriverState *bs, BlockDriver *drv, const char *node_name,
1626 QDict *options, int open_flags, Error **errp)
01a56501 1627{
8dc8a60c 1628 AioContext *ctx;
01a56501 1629 Error *local_err = NULL;
0f12264e 1630 int i, ret;
da359909 1631 GLOBAL_STATE_CODE();
01a56501
KW
1632
1633 bdrv_assign_node_name(bs, node_name, &local_err);
1634 if (local_err) {
1635 error_propagate(errp, local_err);
1636 return -EINVAL;
1637 }
1638
1639 bs->drv = drv;
1640 bs->opaque = g_malloc0(drv->instance_size);
1641
1642 if (drv->bdrv_file_open) {
1643 assert(!drv->bdrv_needs_filename || bs->filename[0]);
1644 ret = drv->bdrv_file_open(bs, options, open_flags, &local_err);
680c7f96 1645 } else if (drv->bdrv_open) {
01a56501 1646 ret = drv->bdrv_open(bs, options, open_flags, &local_err);
680c7f96
KW
1647 } else {
1648 ret = 0;
01a56501
KW
1649 }
1650
1651 if (ret < 0) {
1652 if (local_err) {
1653 error_propagate(errp, local_err);
1654 } else if (bs->filename[0]) {
1655 error_setg_errno(errp, -ret, "Could not open '%s'", bs->filename);
1656 } else {
1657 error_setg_errno(errp, -ret, "Could not open image");
1658 }
180ca19a 1659 goto open_failed;
01a56501
KW
1660 }
1661
e8b65355
SH
1662 assert(!(bs->supported_read_flags & ~BDRV_REQ_MASK));
1663 assert(!(bs->supported_write_flags & ~BDRV_REQ_MASK));
1664
1665 /*
1666 * Always allow the BDRV_REQ_REGISTERED_BUF optimization hint. This saves
1667 * drivers that pass read/write requests through to a child the trouble of
1668 * declaring support explicitly.
1669 *
1670 * Drivers must not propagate this flag accidentally when they initiate I/O
1671 * to a bounce buffer. That case should be rare though.
1672 */
1673 bs->supported_read_flags |= BDRV_REQ_REGISTERED_BUF;
1674 bs->supported_write_flags |= BDRV_REQ_REGISTERED_BUF;
1675
8dc8a60c
KW
1676 /* Get the context after .bdrv_open, it can change the context */
1677 ctx = bdrv_get_aio_context(bs);
1678 aio_context_acquire(ctx);
1679
c057960c 1680 ret = bdrv_refresh_total_sectors(bs, bs->total_sectors);
01a56501
KW
1681 if (ret < 0) {
1682 error_setg_errno(errp, -ret, "Could not refresh total sector count");
8dc8a60c 1683 aio_context_release(ctx);
180ca19a 1684 return ret;
01a56501
KW
1685 }
1686
e19b157f 1687 bdrv_graph_rdlock_main_loop();
1e4c797c 1688 bdrv_refresh_limits(bs, NULL, &local_err);
e19b157f 1689 bdrv_graph_rdunlock_main_loop();
8dc8a60c 1690 aio_context_release(ctx);
e19b157f 1691
01a56501
KW
1692 if (local_err) {
1693 error_propagate(errp, local_err);
180ca19a 1694 return -EINVAL;
01a56501
KW
1695 }
1696
1697 assert(bdrv_opt_mem_align(bs) != 0);
1698 assert(bdrv_min_mem_align(bs) != 0);
1699 assert(is_power_of_2(bs->bl.request_alignment));
1700
0f12264e 1701 for (i = 0; i < bs->quiesce_counter; i++) {
5e8ac217
KW
1702 if (drv->bdrv_drain_begin) {
1703 drv->bdrv_drain_begin(bs);
0f12264e
KW
1704 }
1705 }
1706
01a56501 1707 return 0;
180ca19a
MP
1708open_failed:
1709 bs->drv = NULL;
1710 if (bs->file != NULL) {
32a8aba3 1711 bdrv_graph_wrlock(NULL);
180ca19a 1712 bdrv_unref_child(bs, bs->file);
32a8aba3 1713 bdrv_graph_wrunlock();
5bb04747 1714 assert(!bs->file);
180ca19a 1715 }
01a56501
KW
1716 g_free(bs->opaque);
1717 bs->opaque = NULL;
01a56501
KW
1718 return ret;
1719}
1720
621d1737
VSO
1721/*
1722 * Create and open a block node.
1723 *
1724 * @options is a QDict of options to pass to the block drivers, or NULL for an
1725 * empty set of options. The reference to the QDict belongs to the block layer
1726 * after the call (even on failure), so if the caller intends to reuse the
1727 * dictionary, it needs to use qobject_ref() before calling bdrv_open.
1728 */
1729BlockDriverState *bdrv_new_open_driver_opts(BlockDriver *drv,
1730 const char *node_name,
1731 QDict *options, int flags,
1732 Error **errp)
680c7f96
KW
1733{
1734 BlockDriverState *bs;
1735 int ret;
1736
f791bf7f
EGE
1737 GLOBAL_STATE_CODE();
1738
680c7f96
KW
1739 bs = bdrv_new();
1740 bs->open_flags = flags;
621d1737
VSO
1741 bs->options = options ?: qdict_new();
1742 bs->explicit_options = qdict_clone_shallow(bs->options);
680c7f96
KW
1743 bs->opaque = NULL;
1744
1745 update_options_from_flags(bs->options, flags);
1746
1747 ret = bdrv_open_driver(bs, drv, node_name, bs->options, flags, errp);
1748 if (ret < 0) {
cb3e7f08 1749 qobject_unref(bs->explicit_options);
180ca19a 1750 bs->explicit_options = NULL;
cb3e7f08 1751 qobject_unref(bs->options);
180ca19a 1752 bs->options = NULL;
680c7f96
KW
1753 bdrv_unref(bs);
1754 return NULL;
1755 }
1756
1757 return bs;
1758}
1759
621d1737
VSO
1760/* Create and open a block node. */
1761BlockDriverState *bdrv_new_open_driver(BlockDriver *drv, const char *node_name,
1762 int flags, Error **errp)
1763{
f791bf7f 1764 GLOBAL_STATE_CODE();
621d1737
VSO
1765 return bdrv_new_open_driver_opts(drv, node_name, NULL, flags, errp);
1766}
1767
c5f3014b 1768QemuOptsList bdrv_runtime_opts = {
18edf289
KW
1769 .name = "bdrv_common",
1770 .head = QTAILQ_HEAD_INITIALIZER(bdrv_runtime_opts.head),
1771 .desc = {
1772 {
1773 .name = "node-name",
1774 .type = QEMU_OPT_STRING,
1775 .help = "Node name of the block device node",
1776 },
62392ebb
KW
1777 {
1778 .name = "driver",
1779 .type = QEMU_OPT_STRING,
1780 .help = "Block driver to use for the node",
1781 },
91a097e7
KW
1782 {
1783 .name = BDRV_OPT_CACHE_DIRECT,
1784 .type = QEMU_OPT_BOOL,
1785 .help = "Bypass software writeback cache on the host",
1786 },
1787 {
1788 .name = BDRV_OPT_CACHE_NO_FLUSH,
1789 .type = QEMU_OPT_BOOL,
1790 .help = "Ignore flush requests",
1791 },
f87a0e29
AG
1792 {
1793 .name = BDRV_OPT_READ_ONLY,
1794 .type = QEMU_OPT_BOOL,
1795 .help = "Node is opened in read-only mode",
1796 },
e35bdc12
KW
1797 {
1798 .name = BDRV_OPT_AUTO_READ_ONLY,
1799 .type = QEMU_OPT_BOOL,
1800 .help = "Node can become read-only if opening read-write fails",
1801 },
692e01a2
KW
1802 {
1803 .name = "detect-zeroes",
1804 .type = QEMU_OPT_STRING,
1805 .help = "try to optimize zero writes (off, on, unmap)",
1806 },
818584a4 1807 {
415bbca8 1808 .name = BDRV_OPT_DISCARD,
818584a4
KW
1809 .type = QEMU_OPT_STRING,
1810 .help = "discard operation (ignore/off, unmap/on)",
1811 },
5a9347c6
FZ
1812 {
1813 .name = BDRV_OPT_FORCE_SHARE,
1814 .type = QEMU_OPT_BOOL,
1815 .help = "always accept other writers (default: off)",
1816 },
18edf289
KW
1817 { /* end of list */ }
1818 },
1819};
1820
5a5e7f8c
ML
1821QemuOptsList bdrv_create_opts_simple = {
1822 .name = "simple-create-opts",
1823 .head = QTAILQ_HEAD_INITIALIZER(bdrv_create_opts_simple.head),
fd17146c
HR
1824 .desc = {
1825 {
1826 .name = BLOCK_OPT_SIZE,
1827 .type = QEMU_OPT_SIZE,
1828 .help = "Virtual disk size"
1829 },
1830 {
1831 .name = BLOCK_OPT_PREALLOC,
1832 .type = QEMU_OPT_STRING,
1833 .help = "Preallocation mode (allowed values: off)"
1834 },
1835 { /* end of list */ }
1836 }
1837};
1838
57915332
KW
1839/*
1840 * Common part for opening disk images and files
b6ad491a
KW
1841 *
1842 * Removes all processed options from *options.
57915332 1843 */
5696c6e3 1844static int bdrv_open_common(BlockDriverState *bs, BlockBackend *file,
82dc8b41 1845 QDict *options, Error **errp)
57915332
KW
1846{
1847 int ret, open_flags;
035fccdf 1848 const char *filename;
62392ebb 1849 const char *driver_name = NULL;
6913c0c2 1850 const char *node_name = NULL;
818584a4 1851 const char *discard;
18edf289 1852 QemuOpts *opts;
62392ebb 1853 BlockDriver *drv;
34b5d2c6 1854 Error *local_err = NULL;
307261b2 1855 bool ro;
57915332 1856
6405875c 1857 assert(bs->file == NULL);
707ff828 1858 assert(options != NULL && bs->options != options);
bdb73476 1859 GLOBAL_STATE_CODE();
57915332 1860
62392ebb 1861 opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
af175e85 1862 if (!qemu_opts_absorb_qdict(opts, options, errp)) {
62392ebb
KW
1863 ret = -EINVAL;
1864 goto fail_opts;
1865 }
1866
9b7e8691
AG
1867 update_flags_from_options(&bs->open_flags, opts);
1868
62392ebb
KW
1869 driver_name = qemu_opt_get(opts, "driver");
1870 drv = bdrv_find_format(driver_name);
1871 assert(drv != NULL);
1872
5a9347c6
FZ
1873 bs->force_share = qemu_opt_get_bool(opts, BDRV_OPT_FORCE_SHARE, false);
1874
1875 if (bs->force_share && (bs->open_flags & BDRV_O_RDWR)) {
1876 error_setg(errp,
1877 BDRV_OPT_FORCE_SHARE
1878 "=on can only be used with read-only images");
1879 ret = -EINVAL;
1880 goto fail_opts;
1881 }
1882
45673671 1883 if (file != NULL) {
b7cfc7d5 1884 bdrv_graph_rdlock_main_loop();
f30c66ba 1885 bdrv_refresh_filename(blk_bs(file));
b7cfc7d5
KW
1886 bdrv_graph_rdunlock_main_loop();
1887
5696c6e3 1888 filename = blk_bs(file)->filename;
45673671 1889 } else {
129c7d1c
MA
1890 /*
1891 * Caution: while qdict_get_try_str() is fine, getting
1892 * non-string types would require more care. When @options
1893 * come from -blockdev or blockdev_add, its members are typed
1894 * according to the QAPI schema, but when they come from
1895 * -drive, they're all QString.
1896 */
45673671
KW
1897 filename = qdict_get_try_str(options, "filename");
1898 }
1899
4a008240 1900 if (drv->bdrv_needs_filename && (!filename || !filename[0])) {
765003db
KW
1901 error_setg(errp, "The '%s' block driver requires a file name",
1902 drv->format_name);
18edf289
KW
1903 ret = -EINVAL;
1904 goto fail_opts;
6913c0c2 1905 }
6913c0c2 1906
82dc8b41
KW
1907 trace_bdrv_open_common(bs, filename ?: "", bs->open_flags,
1908 drv->format_name);
62392ebb 1909
307261b2
VSO
1910 ro = bdrv_is_read_only(bs);
1911
1912 if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv, ro)) {
1913 if (!ro && bdrv_is_whitelisted(drv, true)) {
018f9dea 1914 bdrv_graph_rdlock_main_loop();
8be25de6 1915 ret = bdrv_apply_auto_read_only(bs, NULL, NULL);
018f9dea 1916 bdrv_graph_rdunlock_main_loop();
8be25de6
KW
1917 } else {
1918 ret = -ENOTSUP;
1919 }
1920 if (ret < 0) {
1921 error_setg(errp,
307261b2 1922 !ro && bdrv_is_whitelisted(drv, true)
8be25de6
KW
1923 ? "Driver '%s' can only be used for read-only devices"
1924 : "Driver '%s' is not whitelisted",
1925 drv->format_name);
1926 goto fail_opts;
1927 }
b64ec4e4 1928 }
57915332 1929
d3faa13e 1930 /* bdrv_new() and bdrv_close() make it so */
d73415a3 1931 assert(qatomic_read(&bs->copy_on_read) == 0);
d3faa13e 1932
82dc8b41 1933 if (bs->open_flags & BDRV_O_COPY_ON_READ) {
307261b2 1934 if (!ro) {
0ebd24e0
KW
1935 bdrv_enable_copy_on_read(bs);
1936 } else {
1937 error_setg(errp, "Can't use copy-on-read on read-only device");
18edf289
KW
1938 ret = -EINVAL;
1939 goto fail_opts;
0ebd24e0 1940 }
53fec9d3
SH
1941 }
1942
415bbca8 1943 discard = qemu_opt_get(opts, BDRV_OPT_DISCARD);
818584a4
KW
1944 if (discard != NULL) {
1945 if (bdrv_parse_discard_flags(discard, &bs->open_flags) != 0) {
1946 error_setg(errp, "Invalid discard option");
1947 ret = -EINVAL;
1948 goto fail_opts;
1949 }
1950 }
1951
543770bd
AG
1952 bs->detect_zeroes =
1953 bdrv_parse_detect_zeroes(opts, bs->open_flags, &local_err);
1954 if (local_err) {
1955 error_propagate(errp, local_err);
1956 ret = -EINVAL;
1957 goto fail_opts;
692e01a2
KW
1958 }
1959
c2ad1b0c
KW
1960 if (filename != NULL) {
1961 pstrcpy(bs->filename, sizeof(bs->filename), filename);
1962 } else {
1963 bs->filename[0] = '\0';
1964 }
91af7014 1965 pstrcpy(bs->exact_filename, sizeof(bs->exact_filename), bs->filename);
57915332 1966
66f82cee 1967 /* Open the image, either directly or using a protocol */
82dc8b41 1968 open_flags = bdrv_open_flags(bs, bs->open_flags);
01a56501 1969 node_name = qemu_opt_get(opts, "node-name");
57915332 1970
01a56501
KW
1971 assert(!drv->bdrv_file_open || file == NULL);
1972 ret = bdrv_open_driver(bs, drv, node_name, options, open_flags, errp);
51762288 1973 if (ret < 0) {
01a56501 1974 goto fail_opts;
3baca891
KW
1975 }
1976
18edf289 1977 qemu_opts_del(opts);
57915332
KW
1978 return 0;
1979
18edf289
KW
1980fail_opts:
1981 qemu_opts_del(opts);
57915332
KW
1982 return ret;
1983}
1984
5e5c4f63
KW
1985static QDict *parse_json_filename(const char *filename, Error **errp)
1986{
1987 QObject *options_obj;
1988 QDict *options;
1989 int ret;
bdb73476 1990 GLOBAL_STATE_CODE();
5e5c4f63
KW
1991
1992 ret = strstart(filename, "json:", &filename);
1993 assert(ret);
1994
5577fff7 1995 options_obj = qobject_from_json(filename, errp);
5e5c4f63 1996 if (!options_obj) {
5577fff7 1997 error_prepend(errp, "Could not parse the JSON options: ");
5e5c4f63
KW
1998 return NULL;
1999 }
2000
7dc847eb 2001 options = qobject_to(QDict, options_obj);
ca6b6e1e 2002 if (!options) {
cb3e7f08 2003 qobject_unref(options_obj);
5e5c4f63
KW
2004 error_setg(errp, "Invalid JSON object given");
2005 return NULL;
2006 }
2007
5e5c4f63
KW
2008 qdict_flatten(options);
2009
2010 return options;
2011}
2012
de3b53f0
KW
2013static void parse_json_protocol(QDict *options, const char **pfilename,
2014 Error **errp)
2015{
2016 QDict *json_options;
2017 Error *local_err = NULL;
bdb73476 2018 GLOBAL_STATE_CODE();
de3b53f0
KW
2019
2020 /* Parse json: pseudo-protocol */
2021 if (!*pfilename || !g_str_has_prefix(*pfilename, "json:")) {
2022 return;
2023 }
2024
2025 json_options = parse_json_filename(*pfilename, &local_err);
2026 if (local_err) {
2027 error_propagate(errp, local_err);
2028 return;
2029 }
2030
2031 /* Options given in the filename have lower priority than options
2032 * specified directly */
2033 qdict_join(options, json_options, false);
cb3e7f08 2034 qobject_unref(json_options);
de3b53f0
KW
2035 *pfilename = NULL;
2036}
2037
b6ce07aa 2038/*
f54120ff
KW
2039 * Fills in default options for opening images and converts the legacy
2040 * filename/flags pair to option QDict entries.
53a29513
HR
2041 * The BDRV_O_PROTOCOL flag in *flags will be set or cleared accordingly if a
2042 * block driver has been specified explicitly.
b6ce07aa 2043 */
de3b53f0 2044static int bdrv_fill_options(QDict **options, const char *filename,
053e1578 2045 int *flags, Error **errp)
ea2384d3 2046{
c2ad1b0c 2047 const char *drvname;
53a29513 2048 bool protocol = *flags & BDRV_O_PROTOCOL;
e3fa4bfa 2049 bool parse_filename = false;
053e1578 2050 BlockDriver *drv = NULL;
34b5d2c6 2051 Error *local_err = NULL;
83f64091 2052
da359909
EGE
2053 GLOBAL_STATE_CODE();
2054
129c7d1c
MA
2055 /*
2056 * Caution: while qdict_get_try_str() is fine, getting non-string
2057 * types would require more care. When @options come from
2058 * -blockdev or blockdev_add, its members are typed according to
2059 * the QAPI schema, but when they come from -drive, they're all
2060 * QString.
2061 */
53a29513 2062 drvname = qdict_get_try_str(*options, "driver");
053e1578
HR
2063 if (drvname) {
2064 drv = bdrv_find_format(drvname);
2065 if (!drv) {
2066 error_setg(errp, "Unknown driver '%s'", drvname);
2067 return -ENOENT;
2068 }
2069 /* If the user has explicitly specified the driver, this choice should
2070 * override the BDRV_O_PROTOCOL flag */
2071 protocol = drv->bdrv_file_open;
53a29513
HR
2072 }
2073
2074 if (protocol) {
2075 *flags |= BDRV_O_PROTOCOL;
2076 } else {
2077 *flags &= ~BDRV_O_PROTOCOL;
2078 }
2079
91a097e7
KW
2080 /* Translate cache options from flags into options */
2081 update_options_from_flags(*options, *flags);
2082
035fccdf 2083 /* Fetch the file name from the options QDict if necessary */
17b005f1 2084 if (protocol && filename) {
f54120ff 2085 if (!qdict_haskey(*options, "filename")) {
46f5ac20 2086 qdict_put_str(*options, "filename", filename);
f54120ff
KW
2087 parse_filename = true;
2088 } else {
2089 error_setg(errp, "Can't specify 'file' and 'filename' options at "
2090 "the same time");
2091 return -EINVAL;
2092 }
035fccdf
KW
2093 }
2094
c2ad1b0c 2095 /* Find the right block driver */
129c7d1c 2096 /* See cautionary note on accessing @options above */
f54120ff 2097 filename = qdict_get_try_str(*options, "filename");
f54120ff 2098
053e1578
HR
2099 if (!drvname && protocol) {
2100 if (filename) {
2101 drv = bdrv_find_protocol(filename, parse_filename, errp);
17b005f1 2102 if (!drv) {
053e1578 2103 return -EINVAL;
17b005f1 2104 }
053e1578
HR
2105
2106 drvname = drv->format_name;
46f5ac20 2107 qdict_put_str(*options, "driver", drvname);
053e1578
HR
2108 } else {
2109 error_setg(errp, "Must specify either driver or file");
2110 return -EINVAL;
98289620 2111 }
c2ad1b0c
KW
2112 }
2113
17b005f1 2114 assert(drv || !protocol);
c2ad1b0c 2115
f54120ff 2116 /* Driver-specific filename parsing */
17b005f1 2117 if (drv && drv->bdrv_parse_filename && parse_filename) {
5acd9d81 2118 drv->bdrv_parse_filename(filename, *options, &local_err);
84d18f06 2119 if (local_err) {
34b5d2c6 2120 error_propagate(errp, local_err);
f54120ff 2121 return -EINVAL;
6963a30d 2122 }
cd5d031e
HR
2123
2124 if (!drv->bdrv_needs_filename) {
2125 qdict_del(*options, "filename");
cd5d031e 2126 }
6963a30d
KW
2127 }
2128
f54120ff
KW
2129 return 0;
2130}
2131
148eb13c
KW
2132typedef struct BlockReopenQueueEntry {
2133 bool prepared;
2134 BDRVReopenState state;
859443b0 2135 QTAILQ_ENTRY(BlockReopenQueueEntry) entry;
148eb13c
KW
2136} BlockReopenQueueEntry;
2137
2138/*
2139 * Return the flags that @bs will have after the reopens in @q have
2140 * successfully completed. If @q is NULL (or @bs is not contained in @q),
2141 * return the current flags.
2142 */
2143static int bdrv_reopen_get_flags(BlockReopenQueue *q, BlockDriverState *bs)
2144{
2145 BlockReopenQueueEntry *entry;
2146
2147 if (q != NULL) {
859443b0 2148 QTAILQ_FOREACH(entry, q, entry) {
148eb13c
KW
2149 if (entry->state.bs == bs) {
2150 return entry->state.flags;
2151 }
2152 }
2153 }
2154
2155 return bs->open_flags;
2156}
2157
2158/* Returns whether the image file can be written to after the reopen queue @q
2159 * has been successfully applied, or right now if @q is NULL. */
cc022140
HR
2160static bool bdrv_is_writable_after_reopen(BlockDriverState *bs,
2161 BlockReopenQueue *q)
148eb13c
KW
2162{
2163 int flags = bdrv_reopen_get_flags(q, bs);
2164
2165 return (flags & (BDRV_O_RDWR | BDRV_O_INACTIVE)) == BDRV_O_RDWR;
2166}
2167
cc022140
HR
2168/*
2169 * Return whether the BDS can be written to. This is not necessarily
2170 * the same as !bdrv_is_read_only(bs), as inactivated images may not
2171 * be written to but do not count as read-only images.
2172 */
2173bool bdrv_is_writable(BlockDriverState *bs)
2174{
384a48fb 2175 IO_CODE();
cc022140
HR
2176 return bdrv_is_writable_after_reopen(bs, NULL);
2177}
2178
3bf416ba
VSO
2179static char *bdrv_child_user_desc(BdrvChild *c)
2180{
f0c28327 2181 GLOBAL_STATE_CODE();
da261b69 2182 return c->klass->get_parent_desc(c);
3bf416ba
VSO
2183}
2184
30ebb9aa
VSO
2185/*
2186 * Check that @a allows everything that @b needs. @a and @b must reference same
2187 * child node.
2188 */
3bf416ba
VSO
2189static bool bdrv_a_allow_b(BdrvChild *a, BdrvChild *b, Error **errp)
2190{
30ebb9aa
VSO
2191 const char *child_bs_name;
2192 g_autofree char *a_user = NULL;
2193 g_autofree char *b_user = NULL;
2194 g_autofree char *perms = NULL;
2195
2196 assert(a->bs);
2197 assert(a->bs == b->bs);
862fded9 2198 GLOBAL_STATE_CODE();
3bf416ba
VSO
2199
2200 if ((b->perm & a->shared_perm) == b->perm) {
2201 return true;
2202 }
2203
30ebb9aa
VSO
2204 child_bs_name = bdrv_get_node_name(b->bs);
2205 a_user = bdrv_child_user_desc(a);
2206 b_user = bdrv_child_user_desc(b);
2207 perms = bdrv_perm_names(b->perm & ~a->shared_perm);
2208
2209 error_setg(errp, "Permission conflict on node '%s': permissions '%s' are "
2210 "both required by %s (uses node '%s' as '%s' child) and "
2211 "unshared by %s (uses node '%s' as '%s' child).",
2212 child_bs_name, perms,
2213 b_user, child_bs_name, b->name,
2214 a_user, child_bs_name, a->name);
3bf416ba
VSO
2215
2216 return false;
2217}
2218
3804e3cf
KW
2219static bool GRAPH_RDLOCK
2220bdrv_parent_perms_conflict(BlockDriverState *bs, Error **errp)
3bf416ba
VSO
2221{
2222 BdrvChild *a, *b;
862fded9 2223 GLOBAL_STATE_CODE();
3bf416ba
VSO
2224
2225 /*
2226 * During the loop we'll look at each pair twice. That's correct because
2227 * bdrv_a_allow_b() is asymmetric and we should check each pair in both
2228 * directions.
2229 */
2230 QLIST_FOREACH(a, &bs->parents, next_parent) {
2231 QLIST_FOREACH(b, &bs->parents, next_parent) {
9397c14f 2232 if (a == b) {
3bf416ba
VSO
2233 continue;
2234 }
2235
2236 if (!bdrv_a_allow_b(a, b, errp)) {
2237 return true;
2238 }
2239 }
2240 }
2241
2242 return false;
2243}
2244
c629b6d2
KW
2245static void GRAPH_RDLOCK
2246bdrv_child_perm(BlockDriverState *bs, BlockDriverState *child_bs,
2247 BdrvChild *c, BdrvChildRole role,
2248 BlockReopenQueue *reopen_queue,
2249 uint64_t parent_perm, uint64_t parent_shared,
2250 uint64_t *nperm, uint64_t *nshared)
ffd1a5a2 2251{
0b3ca76e 2252 assert(bs->drv && bs->drv->bdrv_child_perm);
da359909 2253 GLOBAL_STATE_CODE();
e5d8a406 2254 bs->drv->bdrv_child_perm(bs, c, role, reopen_queue,
0b3ca76e
AG
2255 parent_perm, parent_shared,
2256 nperm, nshared);
e0995dc3 2257 /* TODO Take force_share from reopen_queue */
ffd1a5a2
FZ
2258 if (child_bs && child_bs->force_share) {
2259 *nshared = BLK_PERM_ALL;
2260 }
2261}
2262
bd57f8f7
VSO
2263/*
2264 * Adds the whole subtree of @bs (including @bs itself) to the @list (except for
2265 * nodes that are already in the @list, of course) so that final list is
2266 * topologically sorted. Return the result (GSList @list object is updated, so
2267 * don't use old reference after function call).
2268 *
2269 * On function start @list must be already topologically sorted and for any node
2270 * in the @list the whole subtree of the node must be in the @list as well. The
2271 * simplest way to satisfy this criteria: use only result of
2272 * bdrv_topological_dfs() or NULL as @list parameter.
2273 */
3804e3cf
KW
2274static GSList * GRAPH_RDLOCK
2275bdrv_topological_dfs(GSList *list, GHashTable *found, BlockDriverState *bs)
bd57f8f7
VSO
2276{
2277 BdrvChild *child;
2278 g_autoptr(GHashTable) local_found = NULL;
2279
bdb73476
EGE
2280 GLOBAL_STATE_CODE();
2281
bd57f8f7
VSO
2282 if (!found) {
2283 assert(!list);
2284 found = local_found = g_hash_table_new(NULL, NULL);
2285 }
2286
2287 if (g_hash_table_contains(found, bs)) {
2288 return list;
2289 }
2290 g_hash_table_add(found, bs);
2291
2292 QLIST_FOREACH(child, &bs->children, next) {
2293 list = bdrv_topological_dfs(list, found, child->bs);
2294 }
2295
2296 return g_slist_prepend(list, bs);
2297}
2298
ecb776bd
VSO
2299typedef struct BdrvChildSetPermState {
2300 BdrvChild *child;
2301 uint64_t old_perm;
2302 uint64_t old_shared_perm;
2303} BdrvChildSetPermState;
b0defa83
VSO
2304
2305static void bdrv_child_set_perm_abort(void *opaque)
2306{
ecb776bd
VSO
2307 BdrvChildSetPermState *s = opaque;
2308
862fded9
EGE
2309 GLOBAL_STATE_CODE();
2310
ecb776bd
VSO
2311 s->child->perm = s->old_perm;
2312 s->child->shared_perm = s->old_shared_perm;
b0defa83
VSO
2313}
2314
2315static TransactionActionDrv bdrv_child_set_pem_drv = {
2316 .abort = bdrv_child_set_perm_abort,
ecb776bd 2317 .clean = g_free,
b0defa83
VSO
2318};
2319
ecb776bd
VSO
2320static void bdrv_child_set_perm(BdrvChild *c, uint64_t perm,
2321 uint64_t shared, Transaction *tran)
b0defa83 2322{
ecb776bd 2323 BdrvChildSetPermState *s = g_new(BdrvChildSetPermState, 1);
862fded9 2324 GLOBAL_STATE_CODE();
ecb776bd
VSO
2325
2326 *s = (BdrvChildSetPermState) {
2327 .child = c,
2328 .old_perm = c->perm,
2329 .old_shared_perm = c->shared_perm,
2330 };
b0defa83
VSO
2331
2332 c->perm = perm;
2333 c->shared_perm = shared;
2334
ecb776bd 2335 tran_add(tran, &bdrv_child_set_pem_drv, s);
b0defa83
VSO
2336}
2337
bce73bc2 2338static void GRAPH_RDLOCK bdrv_drv_set_perm_commit(void *opaque)
2513ef59
VSO
2339{
2340 BlockDriverState *bs = opaque;
2341 uint64_t cumulative_perms, cumulative_shared_perms;
da359909 2342 GLOBAL_STATE_CODE();
2513ef59
VSO
2343
2344 if (bs->drv->bdrv_set_perm) {
2345 bdrv_get_cumulative_perm(bs, &cumulative_perms,
2346 &cumulative_shared_perms);
2347 bs->drv->bdrv_set_perm(bs, cumulative_perms, cumulative_shared_perms);
2348 }
2349}
2350
bce73bc2 2351static void GRAPH_RDLOCK bdrv_drv_set_perm_abort(void *opaque)
2513ef59
VSO
2352{
2353 BlockDriverState *bs = opaque;
da359909 2354 GLOBAL_STATE_CODE();
2513ef59
VSO
2355
2356 if (bs->drv->bdrv_abort_perm_update) {
2357 bs->drv->bdrv_abort_perm_update(bs);
2358 }
2359}
2360
2361TransactionActionDrv bdrv_drv_set_perm_drv = {
2362 .abort = bdrv_drv_set_perm_abort,
2363 .commit = bdrv_drv_set_perm_commit,
2364};
2365
bce73bc2
KW
2366/*
2367 * After calling this function, the transaction @tran may only be completed
2368 * while holding a reader lock for the graph.
2369 */
2370static int GRAPH_RDLOCK
2371bdrv_drv_set_perm(BlockDriverState *bs, uint64_t perm, uint64_t shared_perm,
2372 Transaction *tran, Error **errp)
2513ef59 2373{
da359909 2374 GLOBAL_STATE_CODE();
2513ef59
VSO
2375 if (!bs->drv) {
2376 return 0;
2377 }
2378
2379 if (bs->drv->bdrv_check_perm) {
2380 int ret = bs->drv->bdrv_check_perm(bs, perm, shared_perm, errp);
2381 if (ret < 0) {
2382 return ret;
2383 }
2384 }
2385
2386 if (tran) {
2387 tran_add(tran, &bdrv_drv_set_perm_drv, bs);
2388 }
2389
2390 return 0;
2391}
2392
0978623e
VSO
2393typedef struct BdrvReplaceChildState {
2394 BdrvChild *child;
2395 BlockDriverState *old_bs;
2396} BdrvReplaceChildState;
2397
5661a00d 2398static void GRAPH_WRLOCK bdrv_replace_child_commit(void *opaque)
0978623e
VSO
2399{
2400 BdrvReplaceChildState *s = opaque;
bdb73476 2401 GLOBAL_STATE_CODE();
0978623e 2402
5661a00d 2403 bdrv_schedule_unref(s->old_bs);
0978623e
VSO
2404}
2405
5661a00d 2406static void GRAPH_WRLOCK bdrv_replace_child_abort(void *opaque)
0978623e
VSO
2407{
2408 BdrvReplaceChildState *s = opaque;
2409 BlockDriverState *new_bs = s->child->bs;
2410
bdb73476 2411 GLOBAL_STATE_CODE();
5661a00d 2412 assert_bdrv_graph_writable();
ad29eb3d 2413
0f0b1e29 2414 /* old_bs reference is transparently moved from @s to @s->child */
23987471
KW
2415 if (!s->child->bs) {
2416 /*
2417 * The parents were undrained when removing old_bs from the child. New
2418 * requests can't have been made, though, because the child was empty.
2419 *
2420 * TODO Make bdrv_replace_child_noperm() transactionable to avoid
2421 * undraining the parent in the first place. Once this is done, having
2422 * new_bs drained when calling bdrv_replace_child_tran() is not a
2423 * requirement any more.
2424 */
606ed756 2425 bdrv_parent_drained_begin_single(s->child);
23987471
KW
2426 assert(!bdrv_parent_drained_poll_single(s->child));
2427 }
2428 assert(s->child->quiesced_parent);
544acc7d 2429 bdrv_replace_child_noperm(s->child, s->old_bs);
ad29eb3d 2430
0978623e
VSO
2431 bdrv_unref(new_bs);
2432}
2433
2434static TransactionActionDrv bdrv_replace_child_drv = {
2435 .commit = bdrv_replace_child_commit,
2436 .abort = bdrv_replace_child_abort,
2437 .clean = g_free,
2438};
2439
2440/*
4bf021db 2441 * bdrv_replace_child_tran
0978623e
VSO
2442 *
2443 * Note: real unref of old_bs is done only on commit.
4bf021db 2444 *
23987471
KW
2445 * Both @child->bs and @new_bs (if non-NULL) must be drained. @new_bs must be
2446 * kept drained until the transaction is completed.
2447 *
5661a00d
KW
2448 * After calling this function, the transaction @tran may only be completed
2449 * while holding a writer lock for the graph.
2450 *
4bf021db 2451 * The function doesn't update permissions, caller is responsible for this.
0978623e 2452 */
2f64e1fc
KW
2453static void GRAPH_WRLOCK
2454bdrv_replace_child_tran(BdrvChild *child, BlockDriverState *new_bs,
2455 Transaction *tran)
0978623e
VSO
2456{
2457 BdrvReplaceChildState *s = g_new(BdrvReplaceChildState, 1);
23987471
KW
2458
2459 assert(child->quiesced_parent);
2460 assert(!new_bs || new_bs->quiesce_counter);
2461
0978623e 2462 *s = (BdrvReplaceChildState) {
0f0b1e29
VSO
2463 .child = child,
2464 .old_bs = child->bs,
0978623e
VSO
2465 };
2466 tran_add(tran, &bdrv_replace_child_drv, s);
2467
2468 if (new_bs) {
2469 bdrv_ref(new_bs);
2470 }
ad29eb3d 2471
544acc7d 2472 bdrv_replace_child_noperm(child, new_bs);
0f0b1e29 2473 /* old_bs reference is transparently moved from @child to @s */
0978623e
VSO
2474}
2475
33a610c3 2476/*
c20555e1
VSO
2477 * Refresh permissions in @bs subtree. The function is intended to be called
2478 * after some graph modification that was done without permission update.
bce73bc2
KW
2479 *
2480 * After calling this function, the transaction @tran may only be completed
2481 * while holding a reader lock for the graph.
33a610c3 2482 */
bce73bc2
KW
2483static int GRAPH_RDLOCK
2484bdrv_node_refresh_perm(BlockDriverState *bs, BlockReopenQueue *q,
2485 Transaction *tran, Error **errp)
33a610c3
KW
2486{
2487 BlockDriver *drv = bs->drv;
2488 BdrvChild *c;
2489 int ret;
c20555e1 2490 uint64_t cumulative_perms, cumulative_shared_perms;
862fded9 2491 GLOBAL_STATE_CODE();
c20555e1
VSO
2492
2493 bdrv_get_cumulative_perm(bs, &cumulative_perms, &cumulative_shared_perms);
33a610c3
KW
2494
2495 /* Write permissions never work with read-only images */
2496 if ((cumulative_perms & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)) &&
cc022140 2497 !bdrv_is_writable_after_reopen(bs, q))
33a610c3 2498 {
481e0eee
HR
2499 if (!bdrv_is_writable_after_reopen(bs, NULL)) {
2500 error_setg(errp, "Block node is read-only");
2501 } else {
c20555e1
VSO
2502 error_setg(errp, "Read-only block node '%s' cannot support "
2503 "read-write users", bdrv_get_node_name(bs));
481e0eee
HR
2504 }
2505
33a610c3
KW
2506 return -EPERM;
2507 }
2508
9c60a5d1
KW
2509 /*
2510 * Unaligned requests will automatically be aligned to bl.request_alignment
2511 * and without RESIZE we can't extend requests to write to space beyond the
2512 * end of the image, so it's required that the image size is aligned.
2513 */
2514 if ((cumulative_perms & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)) &&
2515 !(cumulative_perms & BLK_PERM_RESIZE))
2516 {
2517 if ((bs->total_sectors * BDRV_SECTOR_SIZE) % bs->bl.request_alignment) {
2518 error_setg(errp, "Cannot get 'write' permission without 'resize': "
2519 "Image size is not a multiple of request "
2520 "alignment");
2521 return -EPERM;
2522 }
2523 }
2524
33a610c3
KW
2525 /* Check this node */
2526 if (!drv) {
2527 return 0;
2528 }
2529
b1d2bbeb 2530 ret = bdrv_drv_set_perm(bs, cumulative_perms, cumulative_shared_perms, tran,
2513ef59
VSO
2531 errp);
2532 if (ret < 0) {
2533 return ret;
33a610c3
KW
2534 }
2535
78e421c9 2536 /* Drivers that never have children can omit .bdrv_child_perm() */
33a610c3 2537 if (!drv->bdrv_child_perm) {
78e421c9 2538 assert(QLIST_EMPTY(&bs->children));
33a610c3
KW
2539 return 0;
2540 }
2541
2542 /* Check all children */
2543 QLIST_FOREACH(c, &bs->children, next) {
2544 uint64_t cur_perm, cur_shared;
9eab1544 2545
e5d8a406 2546 bdrv_child_perm(bs, c->bs, c, c->role, q,
ffd1a5a2
FZ
2547 cumulative_perms, cumulative_shared_perms,
2548 &cur_perm, &cur_shared);
ecb776bd 2549 bdrv_child_set_perm(c, cur_perm, cur_shared, tran);
bd57f8f7
VSO
2550 }
2551
2552 return 0;
2553}
3ef45e02 2554
fb0ff4d1
VSO
2555/*
2556 * @list is a product of bdrv_topological_dfs() (may be called several times) -
2557 * a topologically sorted subgraph.
bce73bc2
KW
2558 *
2559 * After calling this function, the transaction @tran may only be completed
2560 * while holding a reader lock for the graph.
fb0ff4d1 2561 */
3804e3cf
KW
2562static int GRAPH_RDLOCK
2563bdrv_do_refresh_perms(GSList *list, BlockReopenQueue *q, Transaction *tran,
2564 Error **errp)
bd57f8f7
VSO
2565{
2566 int ret;
b1d2bbeb 2567 BlockDriverState *bs;
862fded9 2568 GLOBAL_STATE_CODE();
bd57f8f7 2569
b1d2bbeb
VSO
2570 for ( ; list; list = list->next) {
2571 bs = list->data;
2572
9397c14f 2573 if (bdrv_parent_perms_conflict(bs, errp)) {
b1d2bbeb 2574 return -EINVAL;
bd57f8f7
VSO
2575 }
2576
c20555e1 2577 ret = bdrv_node_refresh_perm(bs, q, tran, errp);
33a610c3
KW
2578 if (ret < 0) {
2579 return ret;
2580 }
2581 }
2582
2583 return 0;
2584}
2585
fb0ff4d1
VSO
2586/*
2587 * @list is any list of nodes. List is completed by all subtrees and
2588 * topologically sorted. It's not a problem if some node occurs in the @list
2589 * several times.
bce73bc2
KW
2590 *
2591 * After calling this function, the transaction @tran may only be completed
2592 * while holding a reader lock for the graph.
fb0ff4d1 2593 */
3804e3cf
KW
2594static int GRAPH_RDLOCK
2595bdrv_list_refresh_perms(GSList *list, BlockReopenQueue *q, Transaction *tran,
2596 Error **errp)
fb0ff4d1
VSO
2597{
2598 g_autoptr(GHashTable) found = g_hash_table_new(NULL, NULL);
2599 g_autoptr(GSList) refresh_list = NULL;
2600
2601 for ( ; list; list = list->next) {
2602 refresh_list = bdrv_topological_dfs(refresh_list, found, list->data);
2603 }
2604
2605 return bdrv_do_refresh_perms(refresh_list, q, tran, errp);
2606}
2607
c7a0f2be
KW
2608void bdrv_get_cumulative_perm(BlockDriverState *bs, uint64_t *perm,
2609 uint64_t *shared_perm)
33a610c3
KW
2610{
2611 BdrvChild *c;
2612 uint64_t cumulative_perms = 0;
2613 uint64_t cumulative_shared_perms = BLK_PERM_ALL;
2614
b4ad82aa
EGE
2615 GLOBAL_STATE_CODE();
2616
33a610c3
KW
2617 QLIST_FOREACH(c, &bs->parents, next_parent) {
2618 cumulative_perms |= c->perm;
2619 cumulative_shared_perms &= c->shared_perm;
2620 }
2621
2622 *perm = cumulative_perms;
2623 *shared_perm = cumulative_shared_perms;
2624}
2625
5176196c 2626char *bdrv_perm_names(uint64_t perm)
d083319f
KW
2627{
2628 struct perm_name {
2629 uint64_t perm;
2630 const char *name;
2631 } permissions[] = {
2632 { BLK_PERM_CONSISTENT_READ, "consistent read" },
2633 { BLK_PERM_WRITE, "write" },
2634 { BLK_PERM_WRITE_UNCHANGED, "write unchanged" },
2635 { BLK_PERM_RESIZE, "resize" },
d083319f
KW
2636 { 0, NULL }
2637 };
2638
e2a7423a 2639 GString *result = g_string_sized_new(30);
d083319f
KW
2640 struct perm_name *p;
2641
2642 for (p = permissions; p->name; p++) {
2643 if (perm & p->perm) {
e2a7423a
AG
2644 if (result->len > 0) {
2645 g_string_append(result, ", ");
2646 }
2647 g_string_append(result, p->name);
d083319f
KW
2648 }
2649 }
2650
e2a7423a 2651 return g_string_free(result, FALSE);
d083319f
KW
2652}
2653
33a610c3 2654
bce73bc2
KW
2655/*
2656 * @tran is allowed to be NULL. In this case no rollback is possible.
2657 *
2658 * After calling this function, the transaction @tran may only be completed
2659 * while holding a reader lock for the graph.
2660 */
3804e3cf
KW
2661static int GRAPH_RDLOCK
2662bdrv_refresh_perms(BlockDriverState *bs, Transaction *tran, Error **errp)
bb87e4d1
VSO
2663{
2664 int ret;
f1316edb 2665 Transaction *local_tran = NULL;
b1d2bbeb 2666 g_autoptr(GSList) list = bdrv_topological_dfs(NULL, NULL, bs);
862fded9 2667 GLOBAL_STATE_CODE();
bb87e4d1 2668
f1316edb
VSO
2669 if (!tran) {
2670 tran = local_tran = tran_new();
2671 }
2672
fb0ff4d1 2673 ret = bdrv_do_refresh_perms(list, NULL, tran, errp);
f1316edb
VSO
2674
2675 if (local_tran) {
2676 tran_finalize(local_tran, ret);
2677 }
bb87e4d1 2678
b1d2bbeb 2679 return ret;
bb87e4d1
VSO
2680}
2681
33a610c3
KW
2682int bdrv_child_try_set_perm(BdrvChild *c, uint64_t perm, uint64_t shared,
2683 Error **errp)
2684{
1046779e 2685 Error *local_err = NULL;
83928dc4 2686 Transaction *tran = tran_new();
33a610c3
KW
2687 int ret;
2688
b4ad82aa
EGE
2689 GLOBAL_STATE_CODE();
2690
ecb776bd 2691 bdrv_child_set_perm(c, perm, shared, tran);
83928dc4 2692
f1316edb 2693 ret = bdrv_refresh_perms(c->bs, tran, &local_err);
83928dc4
VSO
2694
2695 tran_finalize(tran, ret);
2696
33a610c3 2697 if (ret < 0) {
071b474f
VSO
2698 if ((perm & ~c->perm) || (c->shared_perm & ~shared)) {
2699 /* tighten permissions */
1046779e
HR
2700 error_propagate(errp, local_err);
2701 } else {
2702 /*
2703 * Our caller may intend to only loosen restrictions and
2704 * does not expect this function to fail. Errors are not
2705 * fatal in such a case, so we can just hide them from our
2706 * caller.
2707 */
2708 error_free(local_err);
2709 ret = 0;
2710 }
33a610c3
KW
2711 }
2712
83928dc4 2713 return ret;
d5e6f437
KW
2714}
2715
c1087f12
HR
2716int bdrv_child_refresh_perms(BlockDriverState *bs, BdrvChild *c, Error **errp)
2717{
2718 uint64_t parent_perms, parent_shared;
2719 uint64_t perms, shared;
2720
b4ad82aa
EGE
2721 GLOBAL_STATE_CODE();
2722
c1087f12 2723 bdrv_get_cumulative_perm(bs, &parent_perms, &parent_shared);
e5d8a406 2724 bdrv_child_perm(bs, c->bs, c, c->role, NULL,
bf8e925e 2725 parent_perms, parent_shared, &perms, &shared);
c1087f12
HR
2726
2727 return bdrv_child_try_set_perm(c, perms, shared, errp);
2728}
2729
87278af1
HR
2730/*
2731 * Default implementation for .bdrv_child_perm() for block filters:
2732 * Forward CONSISTENT_READ, WRITE, WRITE_UNCHANGED, and RESIZE to the
2733 * filtered child.
2734 */
2735static void bdrv_filter_default_perms(BlockDriverState *bs, BdrvChild *c,
87278af1
HR
2736 BdrvChildRole role,
2737 BlockReopenQueue *reopen_queue,
2738 uint64_t perm, uint64_t shared,
2739 uint64_t *nperm, uint64_t *nshared)
6a1b9ee1 2740{
862fded9 2741 GLOBAL_STATE_CODE();
e444fa83
KW
2742 *nperm = perm & DEFAULT_PERM_PASSTHROUGH;
2743 *nshared = (shared & DEFAULT_PERM_PASSTHROUGH) | DEFAULT_PERM_UNCHANGED;
6a1b9ee1
KW
2744}
2745
70082db4 2746static void bdrv_default_perms_for_cow(BlockDriverState *bs, BdrvChild *c,
70082db4
HR
2747 BdrvChildRole role,
2748 BlockReopenQueue *reopen_queue,
2749 uint64_t perm, uint64_t shared,
2750 uint64_t *nperm, uint64_t *nshared)
2751{
e5d8a406 2752 assert(role & BDRV_CHILD_COW);
862fded9 2753 GLOBAL_STATE_CODE();
70082db4
HR
2754
2755 /*
2756 * We want consistent read from backing files if the parent needs it.
2757 * No other operations are performed on backing files.
2758 */
2759 perm &= BLK_PERM_CONSISTENT_READ;
2760
2761 /*
2762 * If the parent can deal with changing data, we're okay with a
2763 * writable and resizable backing file.
2764 * TODO Require !(perm & BLK_PERM_CONSISTENT_READ), too?
2765 */
2766 if (shared & BLK_PERM_WRITE) {
2767 shared = BLK_PERM_WRITE | BLK_PERM_RESIZE;
2768 } else {
2769 shared = 0;
2770 }
2771
64631f36 2772 shared |= BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE_UNCHANGED;
70082db4
HR
2773
2774 if (bs->open_flags & BDRV_O_INACTIVE) {
2775 shared |= BLK_PERM_WRITE | BLK_PERM_RESIZE;
2776 }
2777
2778 *nperm = perm;
2779 *nshared = shared;
2780}
2781
6f838a4b 2782static void bdrv_default_perms_for_storage(BlockDriverState *bs, BdrvChild *c,
6f838a4b
HR
2783 BdrvChildRole role,
2784 BlockReopenQueue *reopen_queue,
2785 uint64_t perm, uint64_t shared,
2786 uint64_t *nperm, uint64_t *nshared)
2787{
2788 int flags;
2789
862fded9 2790 GLOBAL_STATE_CODE();
e5d8a406 2791 assert(role & (BDRV_CHILD_METADATA | BDRV_CHILD_DATA));
6f838a4b
HR
2792
2793 flags = bdrv_reopen_get_flags(reopen_queue, bs);
2794
2795 /*
2796 * Apart from the modifications below, the same permissions are
2797 * forwarded and left alone as for filters
2798 */
e5d8a406 2799 bdrv_filter_default_perms(bs, c, role, reopen_queue,
6f838a4b
HR
2800 perm, shared, &perm, &shared);
2801
f889054f
HR
2802 if (role & BDRV_CHILD_METADATA) {
2803 /* Format drivers may touch metadata even if the guest doesn't write */
2804 if (bdrv_is_writable_after_reopen(bs, reopen_queue)) {
2805 perm |= BLK_PERM_WRITE | BLK_PERM_RESIZE;
2806 }
2807
2808 /*
2809 * bs->file always needs to be consistent because of the
2810 * metadata. We can never allow other users to resize or write
2811 * to it.
2812 */
2813 if (!(flags & BDRV_O_NO_IO)) {
2814 perm |= BLK_PERM_CONSISTENT_READ;
2815 }
2816 shared &= ~(BLK_PERM_WRITE | BLK_PERM_RESIZE);
6f838a4b
HR
2817 }
2818
f889054f
HR
2819 if (role & BDRV_CHILD_DATA) {
2820 /*
2821 * Technically, everything in this block is a subset of the
2822 * BDRV_CHILD_METADATA path taken above, and so this could
2823 * be an "else if" branch. However, that is not obvious, and
2824 * this function is not performance critical, therefore we let
2825 * this be an independent "if".
2826 */
2827
2828 /*
2829 * We cannot allow other users to resize the file because the
2830 * format driver might have some assumptions about the size
2831 * (e.g. because it is stored in metadata, or because the file
2832 * is split into fixed-size data files).
2833 */
2834 shared &= ~BLK_PERM_RESIZE;
2835
2836 /*
2837 * WRITE_UNCHANGED often cannot be performed as such on the
2838 * data file. For example, the qcow2 driver may still need to
2839 * write copied clusters on copy-on-read.
2840 */
2841 if (perm & BLK_PERM_WRITE_UNCHANGED) {
2842 perm |= BLK_PERM_WRITE;
2843 }
2844
2845 /*
2846 * If the data file is written to, the format driver may
2847 * expect to be able to resize it by writing beyond the EOF.
2848 */
2849 if (perm & BLK_PERM_WRITE) {
2850 perm |= BLK_PERM_RESIZE;
2851 }
6f838a4b 2852 }
6f838a4b
HR
2853
2854 if (bs->open_flags & BDRV_O_INACTIVE) {
2855 shared |= BLK_PERM_WRITE | BLK_PERM_RESIZE;
2856 }
2857
2858 *nperm = perm;
2859 *nshared = shared;
2860}
2861
2519f549 2862void bdrv_default_perms(BlockDriverState *bs, BdrvChild *c,
e5d8a406 2863 BdrvChildRole role, BlockReopenQueue *reopen_queue,
2519f549
HR
2864 uint64_t perm, uint64_t shared,
2865 uint64_t *nperm, uint64_t *nshared)
2866{
b4ad82aa 2867 GLOBAL_STATE_CODE();
2519f549
HR
2868 if (role & BDRV_CHILD_FILTERED) {
2869 assert(!(role & (BDRV_CHILD_DATA | BDRV_CHILD_METADATA |
2870 BDRV_CHILD_COW)));
e5d8a406 2871 bdrv_filter_default_perms(bs, c, role, reopen_queue,
2519f549
HR
2872 perm, shared, nperm, nshared);
2873 } else if (role & BDRV_CHILD_COW) {
2874 assert(!(role & (BDRV_CHILD_DATA | BDRV_CHILD_METADATA)));
e5d8a406 2875 bdrv_default_perms_for_cow(bs, c, role, reopen_queue,
2519f549
HR
2876 perm, shared, nperm, nshared);
2877 } else if (role & (BDRV_CHILD_METADATA | BDRV_CHILD_DATA)) {
e5d8a406 2878 bdrv_default_perms_for_storage(bs, c, role, reopen_queue,
2519f549
HR
2879 perm, shared, nperm, nshared);
2880 } else {
2881 g_assert_not_reached();
2882 }
2883}
2884
7b1d9c4d
HR
2885uint64_t bdrv_qapi_perm_to_blk_perm(BlockPermission qapi_perm)
2886{
2887 static const uint64_t permissions[] = {
2888 [BLOCK_PERMISSION_CONSISTENT_READ] = BLK_PERM_CONSISTENT_READ,
2889 [BLOCK_PERMISSION_WRITE] = BLK_PERM_WRITE,
2890 [BLOCK_PERMISSION_WRITE_UNCHANGED] = BLK_PERM_WRITE_UNCHANGED,
2891 [BLOCK_PERMISSION_RESIZE] = BLK_PERM_RESIZE,
7b1d9c4d
HR
2892 };
2893
2894 QEMU_BUILD_BUG_ON(ARRAY_SIZE(permissions) != BLOCK_PERMISSION__MAX);
2895 QEMU_BUILD_BUG_ON(1UL << ARRAY_SIZE(permissions) != BLK_PERM_ALL + 1);
2896
2897 assert(qapi_perm < BLOCK_PERMISSION__MAX);
2898
2899 return permissions[qapi_perm];
2900}
2901
23987471
KW
2902/*
2903 * Replaces the node that a BdrvChild points to without updating permissions.
2904 *
2905 * If @new_bs is non-NULL, the parent of @child must already be drained through
31b2ddfe 2906 * @child and the caller must hold the AioContext lock for @new_bs.
23987471 2907 */
ad29eb3d
KW
2908static void GRAPH_WRLOCK
2909bdrv_replace_child_noperm(BdrvChild *child, BlockDriverState *new_bs)
e9740bc6
KW
2910{
2911 BlockDriverState *old_bs = child->bs;
debc2927 2912 int new_bs_quiesce_counter;
e9740bc6 2913
2cad1ebe 2914 assert(!child->frozen);
23987471
KW
2915
2916 /*
2917 * If we want to change the BdrvChild to point to a drained node as its new
2918 * child->bs, we need to make sure that its new parent is drained, too. In
2919 * other words, either child->quiesce_parent must already be true or we must
2920 * be able to set it and keep the parent's quiesce_counter consistent with
2921 * that, but without polling or starting new requests (this function
2922 * guarantees that it doesn't poll, and starting new requests would be
2923 * against the invariants of drain sections).
2924 *
2925 * To keep things simple, we pick the first option (child->quiesce_parent
2926 * must already be true). We also generalise the rule a bit to make it
2927 * easier to verify in callers and more likely to be covered in test cases:
2928 * The parent must be quiesced through this child even if new_bs isn't
2929 * currently drained.
2930 *
2931 * The only exception is for callers that always pass new_bs == NULL. In
2932 * this case, we obviously never need to consider the case of a drained
2933 * new_bs, so we can keep the callers simpler by allowing them not to drain
2934 * the parent.
2935 */
2936 assert(!new_bs || child->quiesced_parent);
bfb8aa6d 2937 assert(old_bs != new_bs);
f0c28327 2938 GLOBAL_STATE_CODE();
2cad1ebe 2939
bb2614e9
FZ
2940 if (old_bs && new_bs) {
2941 assert(bdrv_get_aio_context(old_bs) == bdrv_get_aio_context(new_bs));
2942 }
debc2927 2943
e9740bc6 2944 if (old_bs) {
bd86fb99
HR
2945 if (child->klass->detach) {
2946 child->klass->detach(child);
d736f119 2947 }
e9740bc6
KW
2948 QLIST_REMOVE(child, next_parent);
2949 }
36fe1331
KW
2950
2951 child->bs = new_bs;
2952
e9740bc6
KW
2953 if (new_bs) {
2954 QLIST_INSERT_HEAD(&new_bs->parents, child, next_parent);
bd86fb99
HR
2955 if (child->klass->attach) {
2956 child->klass->attach(child);
8ee03995
KW
2957 }
2958 }
debc2927
HR
2959
2960 /*
23987471
KW
2961 * If the parent was drained through this BdrvChild previously, but new_bs
2962 * is not drained, allow requests to come in only after the new node has
2963 * been attached.
debc2927 2964 */
57e05be3
KW
2965 new_bs_quiesce_counter = (new_bs ? new_bs->quiesce_counter : 0);
2966 if (!new_bs_quiesce_counter && child->quiesced_parent) {
debc2927 2967 bdrv_parent_drained_end_single(child);
debc2927 2968 }
8ee03995 2969}
33a610c3 2970
04c9c3a5
HR
2971/**
2972 * Free the given @child.
2973 *
2974 * The child must be empty (i.e. `child->bs == NULL`) and it must be
2975 * unused (i.e. not in a children list).
2976 */
2977static void bdrv_child_free(BdrvChild *child)
df581792 2978{
548a74c0 2979 assert(!child->bs);
bdb73476 2980 GLOBAL_STATE_CODE();
680e0cc4
KW
2981 GRAPH_RDLOCK_GUARD_MAINLOOP();
2982
a225369b 2983 assert(!child->next.le_prev); /* not in children list */
04c9c3a5
HR
2984
2985 g_free(child->name);
2986 g_free(child);
548a74c0 2987}
d5e6f437 2988
548a74c0 2989typedef struct BdrvAttachChildCommonState {
5bb04747 2990 BdrvChild *child;
548a74c0
VSO
2991 AioContext *old_parent_ctx;
2992 AioContext *old_child_ctx;
2993} BdrvAttachChildCommonState;
2994
5661a00d 2995static void GRAPH_WRLOCK bdrv_attach_child_common_abort(void *opaque)
548a74c0
VSO
2996{
2997 BdrvAttachChildCommonState *s = opaque;
5bb04747 2998 BlockDriverState *bs = s->child->bs;
548a74c0 2999
f0c28327 3000 GLOBAL_STATE_CODE();
5661a00d 3001 assert_bdrv_graph_writable();
ad29eb3d 3002
5bb04747 3003 bdrv_replace_child_noperm(s->child, NULL);
548a74c0
VSO
3004
3005 if (bdrv_get_aio_context(bs) != s->old_child_ctx) {
142e6907 3006 bdrv_try_change_aio_context(bs, s->old_child_ctx, NULL, &error_abort);
548a74c0
VSO
3007 }
3008
5bb04747 3009 if (bdrv_child_get_parent_aio_context(s->child) != s->old_parent_ctx) {
f8be48ad
EGE
3010 Transaction *tran;
3011 GHashTable *visited;
3012 bool ret;
548a74c0 3013
f8be48ad 3014 tran = tran_new();
548a74c0 3015
f8be48ad
EGE
3016 /* No need to visit `child`, because it has been detached already */
3017 visited = g_hash_table_new(NULL, NULL);
3018 ret = s->child->klass->change_aio_ctx(s->child, s->old_parent_ctx,
3019 visited, tran, &error_abort);
3020 g_hash_table_destroy(visited);
3021
3022 /* transaction is supposed to always succeed */
3023 assert(ret == true);
3024 tran_commit(tran);
d5e6f437
KW
3025 }
3026
5661a00d 3027 bdrv_schedule_unref(bs);
5bb04747 3028 bdrv_child_free(s->child);
548a74c0
VSO
3029}
3030
3031static TransactionActionDrv bdrv_attach_child_common_drv = {
3032 .abort = bdrv_attach_child_common_abort,
3033 .clean = g_free,
3034};
3035
3036/*
3037 * Common part of attaching bdrv child to bs or to blk or to job
f8d2ad78 3038 *
7ec390d5 3039 * Function doesn't update permissions, caller is responsible for this.
5bb04747 3040 *
5661a00d
KW
3041 * After calling this function, the transaction @tran may only be completed
3042 * while holding a writer lock for the graph.
3043 *
5bb04747 3044 * Returns new created child.
c066e808
KW
3045 *
3046 * The caller must hold the AioContext lock for @child_bs. Both @parent_bs and
3047 * @child_bs can move to a different AioContext in this function. Callers must
3048 * make sure that their AioContext locking is still correct after this.
548a74c0 3049 */
7d4ca9d2
KW
3050static BdrvChild * GRAPH_WRLOCK
3051bdrv_attach_child_common(BlockDriverState *child_bs,
3052 const char *child_name,
3053 const BdrvChildClass *child_class,
3054 BdrvChildRole child_role,
3055 uint64_t perm, uint64_t shared_perm,
3056 void *opaque,
3057 Transaction *tran, Error **errp)
548a74c0
VSO
3058{
3059 BdrvChild *new_child;
c066e808 3060 AioContext *parent_ctx, *new_child_ctx;
548a74c0
VSO
3061 AioContext *child_ctx = bdrv_get_aio_context(child_bs);
3062
da261b69 3063 assert(child_class->get_parent_desc);
bdb73476 3064 GLOBAL_STATE_CODE();
548a74c0
VSO
3065
3066 new_child = g_new(BdrvChild, 1);
3067 *new_child = (BdrvChild) {
d5e6f437
KW
3068 .bs = NULL,
3069 .name = g_strdup(child_name),
bd86fb99 3070 .klass = child_class,
258b7765 3071 .role = child_role,
d5e6f437
KW
3072 .perm = perm,
3073 .shared_perm = shared_perm,
3074 .opaque = opaque,
df581792
KW
3075 };
3076
548a74c0
VSO
3077 /*
3078 * If the AioContexts don't match, first try to move the subtree of
132ada80 3079 * child_bs into the AioContext of the new parent. If this doesn't work,
548a74c0
VSO
3080 * try moving the parent into the AioContext of child_bs instead.
3081 */
3082 parent_ctx = bdrv_child_get_parent_aio_context(new_child);
3083 if (child_ctx != parent_ctx) {
3084 Error *local_err = NULL;
142e6907
EGE
3085 int ret = bdrv_try_change_aio_context(child_bs, parent_ctx, NULL,
3086 &local_err);
548a74c0 3087
f8be48ad 3088 if (ret < 0 && child_class->change_aio_ctx) {
fb2575f9 3089 Transaction *aio_ctx_tran = tran_new();
f8be48ad
EGE
3090 GHashTable *visited = g_hash_table_new(NULL, NULL);
3091 bool ret_child;
3092
3093 g_hash_table_add(visited, new_child);
3094 ret_child = child_class->change_aio_ctx(new_child, child_ctx,
fb2575f9
MA
3095 visited, aio_ctx_tran,
3096 NULL);
f8be48ad 3097 if (ret_child == true) {
132ada80
KW
3098 error_free(local_err);
3099 ret = 0;
132ada80 3100 }
fb2575f9 3101 tran_finalize(aio_ctx_tran, ret_child == true ? 0 : -1);
f8be48ad 3102 g_hash_table_destroy(visited);
132ada80 3103 }
548a74c0 3104
132ada80
KW
3105 if (ret < 0) {
3106 error_propagate(errp, local_err);
04c9c3a5 3107 bdrv_child_free(new_child);
5bb04747 3108 return NULL;
132ada80
KW
3109 }
3110 }
3111
c066e808
KW
3112 new_child_ctx = bdrv_get_aio_context(child_bs);
3113 if (new_child_ctx != child_ctx) {
3114 aio_context_release(child_ctx);
3115 aio_context_acquire(new_child_ctx);
3116 }
3117
548a74c0 3118 bdrv_ref(child_bs);
23987471
KW
3119 /*
3120 * Let every new BdrvChild start with a drained parent. Inserting the child
3121 * in the graph with bdrv_replace_child_noperm() will undrain it if
3122 * @child_bs is not drained.
3123 *
3124 * The child was only just created and is not yet visible in global state
3125 * until bdrv_replace_child_noperm() inserts it into the graph, so nobody
3126 * could have sent requests and polling is not necessary.
3127 *
3128 * Note that this means that the parent isn't fully drained yet, we only
3129 * stop new requests from coming in. This is fine, we don't care about the
3130 * old requests here, they are not for this child. If another place enters a
3131 * drain section for the same parent, but wants it to be fully quiesced, it
3132 * will not run most of the the code in .drained_begin() again (which is not
3133 * a problem, we already did this), but it will still poll until the parent
3134 * is fully quiesced, so it will not be negatively affected either.
3135 */
606ed756 3136 bdrv_parent_drained_begin_single(new_child);
544acc7d 3137 bdrv_replace_child_noperm(new_child, child_bs);
b4b059f6 3138
548a74c0
VSO
3139 BdrvAttachChildCommonState *s = g_new(BdrvAttachChildCommonState, 1);
3140 *s = (BdrvAttachChildCommonState) {
5bb04747 3141 .child = new_child,
548a74c0
VSO
3142 .old_parent_ctx = parent_ctx,
3143 .old_child_ctx = child_ctx,
3144 };
3145 tran_add(tran, &bdrv_attach_child_common_drv, s);
3146
c066e808
KW
3147 if (new_child_ctx != child_ctx) {
3148 aio_context_release(new_child_ctx);
3149 aio_context_acquire(child_ctx);
3150 }
3151
5bb04747 3152 return new_child;
548a74c0
VSO
3153}
3154
f8d2ad78 3155/*
7ec390d5 3156 * Function doesn't update permissions, caller is responsible for this.
c066e808
KW
3157 *
3158 * The caller must hold the AioContext lock for @child_bs. Both @parent_bs and
3159 * @child_bs can move to a different AioContext in this function. Callers must
3160 * make sure that their AioContext locking is still correct after this.
5661a00d
KW
3161 *
3162 * After calling this function, the transaction @tran may only be completed
3163 * while holding a writer lock for the graph.
f8d2ad78 3164 */
7d4ca9d2
KW
3165static BdrvChild * GRAPH_WRLOCK
3166bdrv_attach_child_noperm(BlockDriverState *parent_bs,
3167 BlockDriverState *child_bs,
3168 const char *child_name,
3169 const BdrvChildClass *child_class,
3170 BdrvChildRole child_role,
3171 Transaction *tran,
3172 Error **errp)
aa5a04c7 3173{
aa5a04c7
VSO
3174 uint64_t perm, shared_perm;
3175
3176 assert(parent_bs->drv);
bdb73476 3177 GLOBAL_STATE_CODE();
aa5a04c7 3178
bfb8aa6d
KW
3179 if (bdrv_recurse_has_child(child_bs, parent_bs)) {
3180 error_setg(errp, "Making '%s' a %s child of '%s' would create a cycle",
3181 child_bs->node_name, child_name, parent_bs->node_name);
5bb04747 3182 return NULL;
bfb8aa6d
KW
3183 }
3184
aa5a04c7
VSO
3185 bdrv_get_cumulative_perm(parent_bs, &perm, &shared_perm);
3186 bdrv_child_perm(parent_bs, child_bs, NULL, child_role, NULL,
3187 perm, shared_perm, &perm, &shared_perm);
3188
5bb04747
VSO
3189 return bdrv_attach_child_common(child_bs, child_name, child_class,
3190 child_role, perm, shared_perm, parent_bs,
3191 tran, errp);
aa5a04c7
VSO
3192}
3193
548a74c0
VSO
3194/*
3195 * This function steals the reference to child_bs from the caller.
3196 * That reference is later dropped by bdrv_root_unref_child().
3197 *
3198 * On failure NULL is returned, errp is set and the reference to
3199 * child_bs is also dropped.
3200 *
3201 * The caller must hold the AioContext lock @child_bs, but not that of @ctx
3202 * (unless @child_bs is already in @ctx).
3203 */
3204BdrvChild *bdrv_root_attach_child(BlockDriverState *child_bs,
3205 const char *child_name,
3206 const BdrvChildClass *child_class,
3207 BdrvChildRole child_role,
3208 uint64_t perm, uint64_t shared_perm,
3209 void *opaque, Error **errp)
3210{
3211 int ret;
5bb04747 3212 BdrvChild *child;
548a74c0
VSO
3213 Transaction *tran = tran_new();
3214
b4ad82aa
EGE
3215 GLOBAL_STATE_CODE();
3216
5bb04747 3217 child = bdrv_attach_child_common(child_bs, child_name, child_class,
548a74c0 3218 child_role, perm, shared_perm, opaque,
5bb04747
VSO
3219 tran, errp);
3220 if (!child) {
3221 ret = -EINVAL;
e878bb12 3222 goto out;
548a74c0
VSO
3223 }
3224
f1316edb 3225 ret = bdrv_refresh_perms(child_bs, tran, errp);
548a74c0 3226
e878bb12
KW
3227out:
3228 tran_finalize(tran, ret);
f8d2ad78 3229
03b9eaca 3230 bdrv_schedule_unref(child_bs);
5bb04747
VSO
3231
3232 return ret < 0 ? NULL : child;
df581792
KW
3233}
3234
b441dc71
AG
3235/*
3236 * This function transfers the reference to child_bs from the caller
3237 * to parent_bs. That reference is later dropped by parent_bs on
3238 * bdrv_close() or if someone calls bdrv_unref_child().
3239 *
3240 * On failure NULL is returned, errp is set and the reference to
3241 * child_bs is also dropped.
132ada80
KW
3242 *
3243 * If @parent_bs and @child_bs are in different AioContexts, the caller must
3244 * hold the AioContext lock for @child_bs, but not for @parent_bs.
b441dc71 3245 */
98292c61
WC
3246BdrvChild *bdrv_attach_child(BlockDriverState *parent_bs,
3247 BlockDriverState *child_bs,
3248 const char *child_name,
bd86fb99 3249 const BdrvChildClass *child_class,
258b7765 3250 BdrvChildRole child_role,
8b2ff529 3251 Error **errp)
f21d96d0 3252{
aa5a04c7 3253 int ret;
5bb04747 3254 BdrvChild *child;
aa5a04c7 3255 Transaction *tran = tran_new();
f68c598b 3256
f791bf7f
EGE
3257 GLOBAL_STATE_CODE();
3258
5bb04747
VSO
3259 child = bdrv_attach_child_noperm(parent_bs, child_bs, child_name,
3260 child_class, child_role, tran, errp);
3261 if (!child) {
3262 ret = -EINVAL;
aa5a04c7
VSO
3263 goto out;
3264 }
d5e6f437 3265
f1316edb 3266 ret = bdrv_refresh_perms(parent_bs, tran, errp);
aa5a04c7
VSO
3267 if (ret < 0) {
3268 goto out;
d5e6f437
KW
3269 }
3270
aa5a04c7
VSO
3271out:
3272 tran_finalize(tran, ret);
3273
afdaeb9e 3274 bdrv_schedule_unref(child_bs);
aa5a04c7 3275
5bb04747 3276 return ret < 0 ? NULL : child;
f21d96d0
KW
3277}
3278
7b99a266 3279/* Callers must ensure that child->frozen is false. */
f21d96d0 3280void bdrv_root_unref_child(BdrvChild *child)
33a60407 3281{
00eb93b5 3282 BlockDriverState *child_bs = child->bs;
779020cb 3283
f791bf7f 3284 GLOBAL_STATE_CODE();
00eb93b5
VSO
3285 bdrv_replace_child_noperm(child, NULL);
3286 bdrv_child_free(child);
3287
3288 if (child_bs) {
3289 /*
3290 * Update permissions for old node. We're just taking a parent away, so
3291 * we're loosening restrictions. Errors of permission update are not
3292 * fatal in this case, ignore them.
3293 */
f1316edb 3294 bdrv_refresh_perms(child_bs, NULL, NULL);
00eb93b5
VSO
3295
3296 /*
3297 * When the parent requiring a non-default AioContext is removed, the
3298 * node moves back to the main AioContext
3299 */
3300 bdrv_try_change_aio_context(child_bs, qemu_get_aio_context(), NULL,
3301 NULL);
3302 }
f791bf7f 3303
ede01e46 3304 bdrv_schedule_unref(child_bs);
f21d96d0
KW
3305}
3306
332b3a17
VSO
3307typedef struct BdrvSetInheritsFrom {
3308 BlockDriverState *bs;
3309 BlockDriverState *old_inherits_from;
3310} BdrvSetInheritsFrom;
3311
3312static void bdrv_set_inherits_from_abort(void *opaque)
3313{
3314 BdrvSetInheritsFrom *s = opaque;
3315
3316 s->bs->inherits_from = s->old_inherits_from;
3317}
3318
3319static TransactionActionDrv bdrv_set_inherits_from_drv = {
3320 .abort = bdrv_set_inherits_from_abort,
3321 .clean = g_free,
3322};
3323
3324/* @tran is allowed to be NULL. In this case no rollback is possible */
3325static void bdrv_set_inherits_from(BlockDriverState *bs,
3326 BlockDriverState *new_inherits_from,
3327 Transaction *tran)
3328{
3329 if (tran) {
3330 BdrvSetInheritsFrom *s = g_new(BdrvSetInheritsFrom, 1);
3331
3332 *s = (BdrvSetInheritsFrom) {
3333 .bs = bs,
3334 .old_inherits_from = bs->inherits_from,
3335 };
3336
3337 tran_add(tran, &bdrv_set_inherits_from_drv, s);
3338 }
3339
3340 bs->inherits_from = new_inherits_from;
3341}
3342
3cf746b3
HR
3343/**
3344 * Clear all inherits_from pointers from children and grandchildren of
3345 * @root that point to @root, where necessary.
332b3a17 3346 * @tran is allowed to be NULL. In this case no rollback is possible
3cf746b3 3347 */
32a8aba3
KW
3348static void GRAPH_WRLOCK
3349bdrv_unset_inherits_from(BlockDriverState *root, BdrvChild *child,
3350 Transaction *tran)
f21d96d0 3351{
3cf746b3 3352 BdrvChild *c;
4e4bf5c4 3353
3cf746b3
HR
3354 if (child->bs->inherits_from == root) {
3355 /*
3356 * Remove inherits_from only when the last reference between root and
3357 * child->bs goes away.
3358 */
3359 QLIST_FOREACH(c, &root->children, next) {
4e4bf5c4
KW
3360 if (c != child && c->bs == child->bs) {
3361 break;
3362 }
3363 }
3364 if (c == NULL) {
332b3a17 3365 bdrv_set_inherits_from(child->bs, NULL, tran);
4e4bf5c4 3366 }
33a60407
KW
3367 }
3368
3cf746b3 3369 QLIST_FOREACH(c, &child->bs->children, next) {
332b3a17 3370 bdrv_unset_inherits_from(root, c, tran);
3cf746b3
HR
3371 }
3372}
3373
7b99a266 3374/* Callers must ensure that child->frozen is false. */
3cf746b3
HR
3375void bdrv_unref_child(BlockDriverState *parent, BdrvChild *child)
3376{
f791bf7f 3377 GLOBAL_STATE_CODE();
3cf746b3
HR
3378 if (child == NULL) {
3379 return;
3380 }
3381
332b3a17 3382 bdrv_unset_inherits_from(parent, child, NULL);
f21d96d0 3383 bdrv_root_unref_child(child);
33a60407
KW
3384}
3385
5c8cab48 3386
356f4ef6
KW
3387static void GRAPH_RDLOCK
3388bdrv_parent_cb_change_media(BlockDriverState *bs, bool load)
5c8cab48
KW
3389{
3390 BdrvChild *c;
f0c28327 3391 GLOBAL_STATE_CODE();
5c8cab48 3392 QLIST_FOREACH(c, &bs->parents, next_parent) {
bd86fb99
HR
3393 if (c->klass->change_media) {
3394 c->klass->change_media(c, load);
5c8cab48
KW
3395 }
3396 }
3397}
3398
0065c455
AG
3399/* Return true if you can reach parent going through child->inherits_from
3400 * recursively. If parent or child are NULL, return false */
3401static bool bdrv_inherits_from_recursive(BlockDriverState *child,
3402 BlockDriverState *parent)
3403{
3404 while (child && child != parent) {
3405 child = child->inherits_from;
3406 }
3407
3408 return child != NULL;
3409}
3410
25191e5f
HR
3411/*
3412 * Return the BdrvChildRole for @bs's backing child. bs->backing is
3413 * mostly used for COW backing children (role = COW), but also for
3414 * filtered children (role = FILTERED | PRIMARY).
3415 */
3416static BdrvChildRole bdrv_backing_role(BlockDriverState *bs)
3417{
3418 if (bs->drv && bs->drv->is_filter) {
3419 return BDRV_CHILD_FILTERED | BDRV_CHILD_PRIMARY;
3420 } else {
3421 return BDRV_CHILD_COW;
3422 }
3423}
3424
5db15a57 3425/*
e9238278
VSO
3426 * Sets the bs->backing or bs->file link of a BDS. A new reference is created;
3427 * callers which don't need their own reference any more must call bdrv_unref().
7ec390d5 3428 *
7d4ca9d2
KW
3429 * If the respective child is already present (i.e. we're detaching a node),
3430 * that child node must be drained.
3431 *
7ec390d5 3432 * Function doesn't update permissions, caller is responsible for this.
4b408668
KW
3433 *
3434 * The caller must hold the AioContext lock for @child_bs. Both @parent_bs and
3435 * @child_bs can move to a different AioContext in this function. Callers must
3436 * make sure that their AioContext locking is still correct after this.
5661a00d
KW
3437 *
3438 * After calling this function, the transaction @tran may only be completed
3439 * while holding a writer lock for the graph.
5db15a57 3440 */
7d4ca9d2
KW
3441static int GRAPH_WRLOCK
3442bdrv_set_file_or_backing_noperm(BlockDriverState *parent_bs,
3443 BlockDriverState *child_bs,
3444 bool is_backing,
3445 Transaction *tran, Error **errp)
8d24cce1 3446{
e9238278
VSO
3447 bool update_inherits_from =
3448 bdrv_inherits_from_recursive(child_bs, parent_bs);
3449 BdrvChild *child = is_backing ? parent_bs->backing : parent_bs->file;
3450 BdrvChildRole role;
0065c455 3451
bdb73476
EGE
3452 GLOBAL_STATE_CODE();
3453
e9238278
VSO
3454 if (!parent_bs->drv) {
3455 /*
3456 * Node without drv is an object without a class :/. TODO: finally fix
3457 * qcow2 driver to never clear bs->drv and implement format corruption
3458 * handling in other way.
3459 */
3460 error_setg(errp, "Node corrupted");
3461 return -EINVAL;
3462 }
3463
3464 if (child && child->frozen) {
3465 error_setg(errp, "Cannot change frozen '%s' link from '%s' to '%s'",
3466 child->name, parent_bs->node_name, child->bs->node_name);
a1e708fc 3467 return -EPERM;
2cad1ebe
AG
3468 }
3469
25f78d9e
VSO
3470 if (is_backing && !parent_bs->drv->is_filter &&
3471 !parent_bs->drv->supports_backing)
3472 {
3473 error_setg(errp, "Driver '%s' of node '%s' does not support backing "
3474 "files", parent_bs->drv->format_name, parent_bs->node_name);
3475 return -EINVAL;
3476 }
3477
e9238278
VSO
3478 if (parent_bs->drv->is_filter) {
3479 role = BDRV_CHILD_FILTERED | BDRV_CHILD_PRIMARY;
3480 } else if (is_backing) {
3481 role = BDRV_CHILD_COW;
3482 } else {
3483 /*
3484 * We only can use same role as it is in existing child. We don't have
3485 * infrastructure to determine role of file child in generic way
3486 */
3487 if (!child) {
3488 error_setg(errp, "Cannot set file child to format node without "
3489 "file child");
3490 return -EINVAL;
3491 }
3492 role = child->role;
826b6ca0
FZ
3493 }
3494
e9238278 3495 if (child) {
7d4ca9d2 3496 assert(child->bs->quiesce_counter);
e9238278 3497 bdrv_unset_inherits_from(parent_bs, child, tran);
57f08941 3498 bdrv_remove_child(child, tran);
e9238278
VSO
3499 }
3500
3501 if (!child_bs) {
8d24cce1
FZ
3502 goto out;
3503 }
12fa4af6 3504
5bb04747
VSO
3505 child = bdrv_attach_child_noperm(parent_bs, child_bs,
3506 is_backing ? "backing" : "file",
3507 &child_of_bds, role,
3508 tran, errp);
3509 if (!child) {
3510 return -EINVAL;
a1e708fc
VSO
3511 }
3512
160333e1
VSO
3513
3514 /*
e9238278 3515 * If inherits_from pointed recursively to bs then let's update it to
160333e1
VSO
3516 * point directly to bs (else it will become NULL).
3517 */
a1e708fc 3518 if (update_inherits_from) {
e9238278 3519 bdrv_set_inherits_from(child_bs, parent_bs, tran);
0065c455 3520 }
826b6ca0 3521
8d24cce1 3522out:
e9238278 3523 bdrv_refresh_limits(parent_bs, tran, NULL);
160333e1
VSO
3524
3525 return 0;
3526}
3527
4b408668
KW
3528/*
3529 * The caller must hold the AioContext lock for @backing_hd. Both @bs and
3530 * @backing_hd can move to a different AioContext in this function. Callers must
3531 * make sure that their AioContext locking is still correct after this.
7d4ca9d2
KW
3532 *
3533 * If a backing child is already present (i.e. we're detaching a node), that
3534 * child node must be drained.
5661a00d
KW
3535 *
3536 * After calling this function, the transaction @tran may only be completed
3537 * while holding a writer lock for the graph.
4b408668 3538 */
7d4ca9d2
KW
3539static int GRAPH_WRLOCK
3540bdrv_set_backing_noperm(BlockDriverState *bs,
3541 BlockDriverState *backing_hd,
3542 Transaction *tran, Error **errp)
e9238278 3543{
bdb73476 3544 GLOBAL_STATE_CODE();
e9238278
VSO
3545 return bdrv_set_file_or_backing_noperm(bs, backing_hd, true, tran, errp);
3546}
3547
92140b9f
KW
3548int bdrv_set_backing_hd_drained(BlockDriverState *bs,
3549 BlockDriverState *backing_hd,
3550 Error **errp)
160333e1
VSO
3551{
3552 int ret;
3553 Transaction *tran = tran_new();
3554
f791bf7f 3555 GLOBAL_STATE_CODE();
92140b9f 3556 assert(bs->quiesce_counter > 0);
7d4ca9d2
KW
3557 if (bs->backing) {
3558 assert(bs->backing->bs->quiesce_counter > 0);
3559 }
3560 bdrv_graph_wrlock(backing_hd);
c0829cb1 3561
160333e1
VSO
3562 ret = bdrv_set_backing_noperm(bs, backing_hd, tran, errp);
3563 if (ret < 0) {
3564 goto out;
3565 }
3566
f1316edb 3567 ret = bdrv_refresh_perms(bs, tran, errp);
160333e1
VSO
3568out:
3569 tran_finalize(tran, ret);
5661a00d 3570 bdrv_graph_wrunlock();
92140b9f
KW
3571 return ret;
3572}
a1e708fc 3573
92140b9f
KW
3574int bdrv_set_backing_hd(BlockDriverState *bs, BlockDriverState *backing_hd,
3575 Error **errp)
3576{
7d4ca9d2 3577 BlockDriverState *drain_bs = bs->backing ? bs->backing->bs : bs;
92140b9f
KW
3578 int ret;
3579 GLOBAL_STATE_CODE();
3580
7d4ca9d2
KW
3581 bdrv_ref(drain_bs);
3582 bdrv_drained_begin(drain_bs);
92140b9f 3583 ret = bdrv_set_backing_hd_drained(bs, backing_hd, errp);
7d4ca9d2
KW
3584 bdrv_drained_end(drain_bs);
3585 bdrv_unref(drain_bs);
c0829cb1 3586
a1e708fc 3587 return ret;
8d24cce1
FZ
3588}
3589
31ca6d07
KW
3590/*
3591 * Opens the backing file for a BlockDriverState if not yet open
3592 *
d9b7b057
KW
3593 * bdref_key specifies the key for the image's BlockdevRef in the options QDict.
3594 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
3595 * itself, all options starting with "${bdref_key}." are considered part of the
3596 * BlockdevRef.
3597 *
2626d27f
KW
3598 * The caller must hold the main AioContext lock.
3599 *
d9b7b057 3600 * TODO Can this be unified with bdrv_open_image()?
31ca6d07 3601 */
d9b7b057
KW
3602int bdrv_open_backing_file(BlockDriverState *bs, QDict *parent_options,
3603 const char *bdref_key, Error **errp)
9156df12 3604{
6b6833c1 3605 char *backing_filename = NULL;
d9b7b057
KW
3606 char *bdref_key_dot;
3607 const char *reference = NULL;
317fc44e 3608 int ret = 0;
998c2019 3609 bool implicit_backing = false;
8d24cce1 3610 BlockDriverState *backing_hd;
8aa04542 3611 AioContext *backing_hd_ctx;
d9b7b057
KW
3612 QDict *options;
3613 QDict *tmp_parent_options = NULL;
34b5d2c6 3614 Error *local_err = NULL;
9156df12 3615
f791bf7f
EGE
3616 GLOBAL_STATE_CODE();
3617
760e0063 3618 if (bs->backing != NULL) {
1ba4b6a5 3619 goto free_exit;
9156df12
PB
3620 }
3621
31ca6d07 3622 /* NULL means an empty set of options */
d9b7b057
KW
3623 if (parent_options == NULL) {
3624 tmp_parent_options = qdict_new();
3625 parent_options = tmp_parent_options;
31ca6d07
KW
3626 }
3627
9156df12 3628 bs->open_flags &= ~BDRV_O_NO_BACKING;
d9b7b057
KW
3629
3630 bdref_key_dot = g_strdup_printf("%s.", bdref_key);
3631 qdict_extract_subqdict(parent_options, &options, bdref_key_dot);
3632 g_free(bdref_key_dot);
3633
129c7d1c
MA
3634 /*
3635 * Caution: while qdict_get_try_str() is fine, getting non-string
3636 * types would require more care. When @parent_options come from
3637 * -blockdev or blockdev_add, its members are typed according to
3638 * the QAPI schema, but when they come from -drive, they're all
3639 * QString.
3640 */
d9b7b057
KW
3641 reference = qdict_get_try_str(parent_options, bdref_key);
3642 if (reference || qdict_haskey(options, "file.filename")) {
6b6833c1 3643 /* keep backing_filename NULL */
1cb6f506 3644 } else if (bs->backing_file[0] == '\0' && qdict_size(options) == 0) {
cb3e7f08 3645 qobject_unref(options);
1ba4b6a5 3646 goto free_exit;
dbecebdd 3647 } else {
998c2019
HR
3648 if (qdict_size(options) == 0) {
3649 /* If the user specifies options that do not modify the
3650 * backing file's behavior, we might still consider it the
3651 * implicit backing file. But it's easier this way, and
3652 * just specifying some of the backing BDS's options is
3653 * only possible with -drive anyway (otherwise the QAPI
3654 * schema forces the user to specify everything). */
3655 implicit_backing = !strcmp(bs->auto_backing_file, bs->backing_file);
3656 }
3657
b7cfc7d5 3658 bdrv_graph_rdlock_main_loop();
6b6833c1 3659 backing_filename = bdrv_get_full_backing_filename(bs, &local_err);
b7cfc7d5
KW
3660 bdrv_graph_rdunlock_main_loop();
3661
9f07429e
HR
3662 if (local_err) {
3663 ret = -EINVAL;
3664 error_propagate(errp, local_err);
cb3e7f08 3665 qobject_unref(options);
9f07429e
HR
3666 goto free_exit;
3667 }
9156df12
PB
3668 }
3669
8ee79e70
KW
3670 if (!bs->drv || !bs->drv->supports_backing) {
3671 ret = -EINVAL;
3672 error_setg(errp, "Driver doesn't support backing files");
cb3e7f08 3673 qobject_unref(options);
8ee79e70
KW
3674 goto free_exit;
3675 }
3676
6bff597b
PK
3677 if (!reference &&
3678 bs->backing_format[0] != '\0' && !qdict_haskey(options, "driver")) {
46f5ac20 3679 qdict_put_str(options, "driver", bs->backing_format);
9156df12
PB
3680 }
3681
6b6833c1 3682 backing_hd = bdrv_open_inherit(backing_filename, reference, options, 0, bs,
25191e5f 3683 &child_of_bds, bdrv_backing_role(bs), errp);
5b363937 3684 if (!backing_hd) {
9156df12 3685 bs->open_flags |= BDRV_O_NO_BACKING;
e43bfd9c 3686 error_prepend(errp, "Could not open backing file: ");
5b363937 3687 ret = -EINVAL;
1ba4b6a5 3688 goto free_exit;
9156df12 3689 }
df581792 3690
998c2019 3691 if (implicit_backing) {
b7cfc7d5 3692 bdrv_graph_rdlock_main_loop();
998c2019 3693 bdrv_refresh_filename(backing_hd);
b7cfc7d5 3694 bdrv_graph_rdunlock_main_loop();
998c2019
HR
3695 pstrcpy(bs->auto_backing_file, sizeof(bs->auto_backing_file),
3696 backing_hd->filename);
3697 }
3698
5db15a57
KW
3699 /* Hook up the backing file link; drop our reference, bs owns the
3700 * backing_hd reference now */
8aa04542
KW
3701 backing_hd_ctx = bdrv_get_aio_context(backing_hd);
3702 aio_context_acquire(backing_hd_ctx);
dc9c10a1 3703 ret = bdrv_set_backing_hd(bs, backing_hd, errp);
5db15a57 3704 bdrv_unref(backing_hd);
8aa04542
KW
3705 aio_context_release(backing_hd_ctx);
3706
dc9c10a1 3707 if (ret < 0) {
12fa4af6
KW
3708 goto free_exit;
3709 }
d80ac658 3710
d9b7b057
KW
3711 qdict_del(parent_options, bdref_key);
3712
1ba4b6a5
BC
3713free_exit:
3714 g_free(backing_filename);
cb3e7f08 3715 qobject_unref(tmp_parent_options);
1ba4b6a5 3716 return ret;
9156df12
PB
3717}
3718
2d6b86af
KW
3719static BlockDriverState *
3720bdrv_open_child_bs(const char *filename, QDict *options, const char *bdref_key,
bd86fb99 3721 BlockDriverState *parent, const BdrvChildClass *child_class,
272c02ea 3722 BdrvChildRole child_role, bool allow_none, Error **errp)
da557aac 3723{
2d6b86af 3724 BlockDriverState *bs = NULL;
da557aac 3725 QDict *image_options;
da557aac
HR
3726 char *bdref_key_dot;
3727 const char *reference;
3728
bd86fb99 3729 assert(child_class != NULL);
f67503e5 3730
da557aac
HR
3731 bdref_key_dot = g_strdup_printf("%s.", bdref_key);
3732 qdict_extract_subqdict(options, &image_options, bdref_key_dot);
3733 g_free(bdref_key_dot);
3734
129c7d1c
MA
3735 /*
3736 * Caution: while qdict_get_try_str() is fine, getting non-string
3737 * types would require more care. When @options come from
3738 * -blockdev or blockdev_add, its members are typed according to
3739 * the QAPI schema, but when they come from -drive, they're all
3740 * QString.
3741 */
da557aac
HR
3742 reference = qdict_get_try_str(options, bdref_key);
3743 if (!filename && !reference && !qdict_size(image_options)) {
b4b059f6 3744 if (!allow_none) {
da557aac
HR
3745 error_setg(errp, "A block device must be specified for \"%s\"",
3746 bdref_key);
da557aac 3747 }
cb3e7f08 3748 qobject_unref(image_options);
da557aac
HR
3749 goto done;
3750 }
3751
5b363937 3752 bs = bdrv_open_inherit(filename, reference, image_options, 0,
272c02ea 3753 parent, child_class, child_role, errp);
5b363937 3754 if (!bs) {
df581792
KW
3755 goto done;
3756 }
3757
da557aac
HR
3758done:
3759 qdict_del(options, bdref_key);
2d6b86af
KW
3760 return bs;
3761}
3762
3763/*
3764 * Opens a disk image whose options are given as BlockdevRef in another block
3765 * device's options.
3766 *
3767 * If allow_none is true, no image will be opened if filename is false and no
3768 * BlockdevRef is given. NULL will be returned, but errp remains unset.
3769 *
3770 * bdrev_key specifies the key for the image's BlockdevRef in the options QDict.
3771 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
3772 * itself, all options starting with "${bdref_key}." are considered part of the
3773 * BlockdevRef.
3774 *
3775 * The BlockdevRef will be removed from the options QDict.
aa269ff8 3776 *
8394c35e 3777 * The caller must hold the lock of the main AioContext and no other AioContext.
aa269ff8
KW
3778 * @parent can move to a different AioContext in this function. Callers must
3779 * make sure that their AioContext locking is still correct after this.
2d6b86af
KW
3780 */
3781BdrvChild *bdrv_open_child(const char *filename,
3782 QDict *options, const char *bdref_key,
3783 BlockDriverState *parent,
bd86fb99 3784 const BdrvChildClass *child_class,
258b7765 3785 BdrvChildRole child_role,
2d6b86af
KW
3786 bool allow_none, Error **errp)
3787{
3788 BlockDriverState *bs;
8394c35e
KW
3789 BdrvChild *child;
3790 AioContext *ctx;
2d6b86af 3791
f791bf7f
EGE
3792 GLOBAL_STATE_CODE();
3793
bd86fb99 3794 bs = bdrv_open_child_bs(filename, options, bdref_key, parent, child_class,
272c02ea 3795 child_role, allow_none, errp);
2d6b86af
KW
3796 if (bs == NULL) {
3797 return NULL;
3798 }
3799
afdaeb9e 3800 bdrv_graph_wrlock(NULL);
8394c35e
KW
3801 ctx = bdrv_get_aio_context(bs);
3802 aio_context_acquire(ctx);
3803 child = bdrv_attach_child(parent, bs, bdref_key, child_class, child_role,
3804 errp);
3805 aio_context_release(ctx);
afdaeb9e 3806 bdrv_graph_wrunlock();
8394c35e
KW
3807
3808 return child;
b4b059f6
KW
3809}
3810
83930780
VSO
3811/*
3812 * Wrapper on bdrv_open_child() for most popular case: open primary child of bs.
aa269ff8 3813 *
8394c35e 3814 * The caller must hold the lock of the main AioContext and no other AioContext.
aa269ff8
KW
3815 * @parent can move to a different AioContext in this function. Callers must
3816 * make sure that their AioContext locking is still correct after this.
83930780
VSO
3817 */
3818int bdrv_open_file_child(const char *filename,
3819 QDict *options, const char *bdref_key,
3820 BlockDriverState *parent, Error **errp)
3821{
3822 BdrvChildRole role;
3823
3824 /* commit_top and mirror_top don't use this function */
3825 assert(!parent->drv->filtered_child_is_backing);
83930780
VSO
3826 role = parent->drv->is_filter ?
3827 (BDRV_CHILD_FILTERED | BDRV_CHILD_PRIMARY) : BDRV_CHILD_IMAGE;
3828
5bb04747
VSO
3829 if (!bdrv_open_child(filename, options, bdref_key, parent,
3830 &child_of_bds, role, false, errp))
3831 {
3832 return -EINVAL;
3833 }
83930780 3834
5bb04747 3835 return 0;
83930780
VSO
3836}
3837
bd86fb99
HR
3838/*
3839 * TODO Future callers may need to specify parent/child_class in order for
3840 * option inheritance to work. Existing callers use it for the root node.
3841 */
e1d74bc6
KW
3842BlockDriverState *bdrv_open_blockdev_ref(BlockdevRef *ref, Error **errp)
3843{
3844 BlockDriverState *bs = NULL;
e1d74bc6
KW
3845 QObject *obj = NULL;
3846 QDict *qdict = NULL;
3847 const char *reference = NULL;
3848 Visitor *v = NULL;
3849
f791bf7f
EGE
3850 GLOBAL_STATE_CODE();
3851
e1d74bc6
KW
3852 if (ref->type == QTYPE_QSTRING) {
3853 reference = ref->u.reference;
3854 } else {
3855 BlockdevOptions *options = &ref->u.definition;
3856 assert(ref->type == QTYPE_QDICT);
3857
3858 v = qobject_output_visitor_new(&obj);
1f584248 3859 visit_type_BlockdevOptions(v, NULL, &options, &error_abort);
e1d74bc6
KW
3860 visit_complete(v, &obj);
3861
7dc847eb 3862 qdict = qobject_to(QDict, obj);
e1d74bc6
KW
3863 qdict_flatten(qdict);
3864
3865 /* bdrv_open_inherit() defaults to the values in bdrv_flags (for
3866 * compatibility with other callers) rather than what we want as the
3867 * real defaults. Apply the defaults here instead. */
3868 qdict_set_default_str(qdict, BDRV_OPT_CACHE_DIRECT, "off");
3869 qdict_set_default_str(qdict, BDRV_OPT_CACHE_NO_FLUSH, "off");
3870 qdict_set_default_str(qdict, BDRV_OPT_READ_ONLY, "off");
e35bdc12
KW
3871 qdict_set_default_str(qdict, BDRV_OPT_AUTO_READ_ONLY, "off");
3872
e1d74bc6
KW
3873 }
3874
272c02ea 3875 bs = bdrv_open_inherit(NULL, reference, qdict, 0, NULL, NULL, 0, errp);
e1d74bc6 3876 obj = NULL;
cb3e7f08 3877 qobject_unref(obj);
e1d74bc6
KW
3878 visit_free(v);
3879 return bs;
3880}
3881
66836189
HR
3882static BlockDriverState *bdrv_append_temp_snapshot(BlockDriverState *bs,
3883 int flags,
3884 QDict *snapshot_options,
3885 Error **errp)
b998875d 3886{
69fbfff9 3887 g_autofree char *tmp_filename = NULL;
b998875d 3888 int64_t total_size;
83d0521a 3889 QemuOpts *opts = NULL;
ff6ed714 3890 BlockDriverState *bs_snapshot = NULL;
f665f01f 3891 AioContext *ctx = bdrv_get_aio_context(bs);
b998875d
KW
3892 int ret;
3893
bdb73476
EGE
3894 GLOBAL_STATE_CODE();
3895
b998875d
KW
3896 /* if snapshot, we create a temporary backing file and open it
3897 instead of opening 'filename' directly */
3898
3899 /* Get the required size from the image */
f665f01f 3900 aio_context_acquire(ctx);
f187743a 3901 total_size = bdrv_getlength(bs);
f665f01f
KW
3902 aio_context_release(ctx);
3903
f187743a
KW
3904 if (total_size < 0) {
3905 error_setg_errno(errp, -total_size, "Could not get image size");
1ba4b6a5 3906 goto out;
f187743a 3907 }
b998875d
KW
3908
3909 /* Create the temporary image */
69fbfff9
BM
3910 tmp_filename = create_tmp_file(errp);
3911 if (!tmp_filename) {
1ba4b6a5 3912 goto out;
b998875d
KW
3913 }
3914
ef810437 3915 opts = qemu_opts_create(bdrv_qcow2.create_opts, NULL, 0,
c282e1fd 3916 &error_abort);
39101f25 3917 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, total_size, &error_abort);
e43bfd9c 3918 ret = bdrv_create(&bdrv_qcow2, tmp_filename, opts, errp);
83d0521a 3919 qemu_opts_del(opts);
b998875d 3920 if (ret < 0) {
e43bfd9c
MA
3921 error_prepend(errp, "Could not create temporary overlay '%s': ",
3922 tmp_filename);
1ba4b6a5 3923 goto out;
b998875d
KW
3924 }
3925
73176bee 3926 /* Prepare options QDict for the temporary file */
46f5ac20
EB
3927 qdict_put_str(snapshot_options, "file.driver", "file");
3928 qdict_put_str(snapshot_options, "file.filename", tmp_filename);
3929 qdict_put_str(snapshot_options, "driver", "qcow2");
b998875d 3930
5b363937 3931 bs_snapshot = bdrv_open(NULL, NULL, snapshot_options, flags, errp);
73176bee 3932 snapshot_options = NULL;
5b363937 3933 if (!bs_snapshot) {
1ba4b6a5 3934 goto out;
b998875d
KW
3935 }
3936
f665f01f 3937 aio_context_acquire(ctx);
934aee14 3938 ret = bdrv_append(bs_snapshot, bs, errp);
f665f01f
KW
3939 aio_context_release(ctx);
3940
934aee14 3941 if (ret < 0) {
ff6ed714 3942 bs_snapshot = NULL;
b2c2832c
KW
3943 goto out;
3944 }
1ba4b6a5
BC
3945
3946out:
cb3e7f08 3947 qobject_unref(snapshot_options);
ff6ed714 3948 return bs_snapshot;
b998875d
KW
3949}
3950
b6ce07aa
KW
3951/*
3952 * Opens a disk image (raw, qcow2, vmdk, ...)
de9c0cec
KW
3953 *
3954 * options is a QDict of options to pass to the block drivers, or NULL for an
3955 * empty set of options. The reference to the QDict belongs to the block layer
3956 * after the call (even on failure), so if the caller intends to reuse the
cb3e7f08 3957 * dictionary, it needs to use qobject_ref() before calling bdrv_open.
f67503e5
HR
3958 *
3959 * If *pbs is NULL, a new BDS will be created with a pointer to it stored there.
3960 * If it is not NULL, the referenced BDS will be reused.
ddf5636d
HR
3961 *
3962 * The reference parameter may be used to specify an existing block device which
3963 * should be opened. If specified, neither options nor a filename may be given,
3964 * nor can an existing BDS be reused (that is, *pbs has to be NULL).
c86422c5 3965 *
ae400dbb 3966 * The caller must always hold the main AioContext lock.
b6ce07aa 3967 */
32192301
KW
3968static BlockDriverState * no_coroutine_fn
3969bdrv_open_inherit(const char *filename, const char *reference, QDict *options,
3970 int flags, BlockDriverState *parent,
3971 const BdrvChildClass *child_class, BdrvChildRole child_role,
3972 Error **errp)
ea2384d3 3973{
b6ce07aa 3974 int ret;
5696c6e3 3975 BlockBackend *file = NULL;
9a4f4c31 3976 BlockDriverState *bs;
ce343771 3977 BlockDriver *drv = NULL;
2f624b80 3978 BdrvChild *child;
74fe54f2 3979 const char *drvname;
3e8c2e57 3980 const char *backing;
34b5d2c6 3981 Error *local_err = NULL;
73176bee 3982 QDict *snapshot_options = NULL;
b1e6fc08 3983 int snapshot_flags = 0;
f665f01f 3984 AioContext *ctx = qemu_get_aio_context();
712e7874 3985
bd86fb99
HR
3986 assert(!child_class || !flags);
3987 assert(!child_class == !parent);
f0c28327 3988 GLOBAL_STATE_CODE();
32192301 3989 assert(!qemu_in_coroutine());
f67503e5 3990
356f4ef6
KW
3991 /* TODO We'll eventually have to take a writer lock in this function */
3992 GRAPH_RDLOCK_GUARD_MAINLOOP();
3993
ddf5636d
HR
3994 if (reference) {
3995 bool options_non_empty = options ? qdict_size(options) : false;
cb3e7f08 3996 qobject_unref(options);
ddf5636d 3997
ddf5636d
HR
3998 if (filename || options_non_empty) {
3999 error_setg(errp, "Cannot reference an existing block device with "
4000 "additional options or a new filename");
5b363937 4001 return NULL;
ddf5636d
HR
4002 }
4003
4004 bs = bdrv_lookup_bs(reference, reference, errp);
4005 if (!bs) {
5b363937 4006 return NULL;
ddf5636d 4007 }
76b22320 4008
ddf5636d 4009 bdrv_ref(bs);
5b363937 4010 return bs;
ddf5636d
HR
4011 }
4012
5b363937 4013 bs = bdrv_new();
f67503e5 4014
de9c0cec
KW
4015 /* NULL means an empty set of options */
4016 if (options == NULL) {
4017 options = qdict_new();
4018 }
4019
145f598e 4020 /* json: syntax counts as explicit options, as if in the QDict */
de3b53f0
KW
4021 parse_json_protocol(options, &filename, &local_err);
4022 if (local_err) {
de3b53f0
KW
4023 goto fail;
4024 }
4025
145f598e
KW
4026 bs->explicit_options = qdict_clone_shallow(options);
4027
bd86fb99 4028 if (child_class) {
3cdc69d3
HR
4029 bool parent_is_format;
4030
4031 if (parent->drv) {
4032 parent_is_format = parent->drv->is_format;
4033 } else {
4034 /*
4035 * parent->drv is not set yet because this node is opened for
4036 * (potential) format probing. That means that @parent is going
4037 * to be a format node.
4038 */
4039 parent_is_format = true;
4040 }
4041
bddcec37 4042 bs->inherits_from = parent;
3cdc69d3
HR
4043 child_class->inherit_options(child_role, parent_is_format,
4044 &flags, options,
bd86fb99 4045 parent->open_flags, parent->options);
f3930ed0
KW
4046 }
4047
de3b53f0 4048 ret = bdrv_fill_options(&options, filename, &flags, &local_err);
dfde483e 4049 if (ret < 0) {
462f5bcf
KW
4050 goto fail;
4051 }
4052
129c7d1c
MA
4053 /*
4054 * Set the BDRV_O_RDWR and BDRV_O_ALLOW_RDWR flags.
4055 * Caution: getting a boolean member of @options requires care.
4056 * When @options come from -blockdev or blockdev_add, members are
4057 * typed according to the QAPI schema, but when they come from
4058 * -drive, they're all QString.
4059 */
f87a0e29
AG
4060 if (g_strcmp0(qdict_get_try_str(options, BDRV_OPT_READ_ONLY), "on") &&
4061 !qdict_get_try_bool(options, BDRV_OPT_READ_ONLY, false)) {
4062 flags |= (BDRV_O_RDWR | BDRV_O_ALLOW_RDWR);
4063 } else {
4064 flags &= ~BDRV_O_RDWR;
14499ea5
AG
4065 }
4066
4067 if (flags & BDRV_O_SNAPSHOT) {
4068 snapshot_options = qdict_new();
4069 bdrv_temp_snapshot_options(&snapshot_flags, snapshot_options,
4070 flags, options);
f87a0e29
AG
4071 /* Let bdrv_backing_options() override "read-only" */
4072 qdict_del(options, BDRV_OPT_READ_ONLY);
00ff7ffd
HR
4073 bdrv_inherited_options(BDRV_CHILD_COW, true,
4074 &flags, options, flags, options);
14499ea5
AG
4075 }
4076
62392ebb
KW
4077 bs->open_flags = flags;
4078 bs->options = options;
4079 options = qdict_clone_shallow(options);
4080
76c591b0 4081 /* Find the right image format driver */
129c7d1c 4082 /* See cautionary note on accessing @options above */
76c591b0
KW
4083 drvname = qdict_get_try_str(options, "driver");
4084 if (drvname) {
4085 drv = bdrv_find_format(drvname);
76c591b0
KW
4086 if (!drv) {
4087 error_setg(errp, "Unknown driver: '%s'", drvname);
76c591b0
KW
4088 goto fail;
4089 }
4090 }
4091
4092 assert(drvname || !(flags & BDRV_O_PROTOCOL));
76c591b0 4093
129c7d1c 4094 /* See cautionary note on accessing @options above */
3e8c2e57 4095 backing = qdict_get_try_str(options, "backing");
e59a0cf1
HR
4096 if (qobject_to(QNull, qdict_get(options, "backing")) != NULL ||
4097 (backing && *backing == '\0'))
4098 {
4f7be280
HR
4099 if (backing) {
4100 warn_report("Use of \"backing\": \"\" is deprecated; "
4101 "use \"backing\": null instead");
4102 }
3e8c2e57 4103 flags |= BDRV_O_NO_BACKING;
ae0f57f0
KW
4104 qdict_del(bs->explicit_options, "backing");
4105 qdict_del(bs->options, "backing");
3e8c2e57
AG
4106 qdict_del(options, "backing");
4107 }
4108
5696c6e3 4109 /* Open image file without format layer. This BlockBackend is only used for
4e4bf5c4
KW
4110 * probing, the block drivers will do their own bdrv_open_child() for the
4111 * same BDS, which is why we put the node name back into options. */
f4788adc 4112 if ((flags & BDRV_O_PROTOCOL) == 0) {
5696c6e3
KW
4113 BlockDriverState *file_bs;
4114
4115 file_bs = bdrv_open_child_bs(filename, options, "file", bs,
58944401
HR
4116 &child_of_bds, BDRV_CHILD_IMAGE,
4117 true, &local_err);
1fdd6933 4118 if (local_err) {
f4788adc
KW
4119 goto fail;
4120 }
5696c6e3 4121 if (file_bs != NULL) {
dacaa162
KW
4122 /* Not requesting BLK_PERM_CONSISTENT_READ because we're only
4123 * looking at the header to guess the image format. This works even
4124 * in cases where a guest would not see a consistent state. */
f665f01f
KW
4125 ctx = bdrv_get_aio_context(file_bs);
4126 aio_context_acquire(ctx);
4127 file = blk_new(ctx, 0, BLK_PERM_ALL);
d7086422 4128 blk_insert_bs(file, file_bs, &local_err);
5696c6e3 4129 bdrv_unref(file_bs);
f665f01f
KW
4130 aio_context_release(ctx);
4131
d7086422
KW
4132 if (local_err) {
4133 goto fail;
4134 }
5696c6e3 4135
46f5ac20 4136 qdict_put_str(options, "file", bdrv_get_node_name(file_bs));
4e4bf5c4 4137 }
f500a6d3
KW
4138 }
4139
76c591b0 4140 /* Image format probing */
38f3ef57 4141 bs->probed = !drv;
76c591b0 4142 if (!drv && file) {
cf2ab8fc 4143 ret = find_image_format(file, filename, &drv, &local_err);
17b005f1 4144 if (ret < 0) {
8bfea15d 4145 goto fail;
2a05cbe4 4146 }
62392ebb
KW
4147 /*
4148 * This option update would logically belong in bdrv_fill_options(),
4149 * but we first need to open bs->file for the probing to work, while
4150 * opening bs->file already requires the (mostly) final set of options
4151 * so that cache mode etc. can be inherited.
4152 *
4153 * Adding the driver later is somewhat ugly, but it's not an option
4154 * that would ever be inherited, so it's correct. We just need to make
4155 * sure to update both bs->options (which has the full effective
4156 * options for bs) and options (which has file.* already removed).
4157 */
46f5ac20
EB
4158 qdict_put_str(bs->options, "driver", drv->format_name);
4159 qdict_put_str(options, "driver", drv->format_name);
76c591b0 4160 } else if (!drv) {
17b005f1 4161 error_setg(errp, "Must specify either driver or file");
8bfea15d 4162 goto fail;
ea2384d3 4163 }
b6ce07aa 4164
53a29513
HR
4165 /* BDRV_O_PROTOCOL must be set iff a protocol BDS is about to be created */
4166 assert(!!(flags & BDRV_O_PROTOCOL) == !!drv->bdrv_file_open);
4167 /* file must be NULL if a protocol BDS is about to be created
4168 * (the inverse results in an error message from bdrv_open_common()) */
4169 assert(!(flags & BDRV_O_PROTOCOL) || !file);
4170
b6ce07aa 4171 /* Open the image */
82dc8b41 4172 ret = bdrv_open_common(bs, file, options, &local_err);
b6ce07aa 4173 if (ret < 0) {
8bfea15d 4174 goto fail;
6987307c
CH
4175 }
4176
f665f01f
KW
4177 /* The AioContext could have changed during bdrv_open_common() */
4178 ctx = bdrv_get_aio_context(bs);
4179
4e4bf5c4 4180 if (file) {
f665f01f 4181 aio_context_acquire(ctx);
5696c6e3 4182 blk_unref(file);
f665f01f 4183 aio_context_release(ctx);
f500a6d3
KW
4184 file = NULL;
4185 }
4186
b6ce07aa 4187 /* If there is a backing file, use it */
9156df12 4188 if ((flags & BDRV_O_NO_BACKING) == 0) {
d9b7b057 4189 ret = bdrv_open_backing_file(bs, options, "backing", &local_err);
b6ce07aa 4190 if (ret < 0) {
b6ad491a 4191 goto close_and_fail;
b6ce07aa 4192 }
b6ce07aa
KW
4193 }
4194
50196d7a
AG
4195 /* Remove all children options and references
4196 * from bs->options and bs->explicit_options */
2f624b80
AG
4197 QLIST_FOREACH(child, &bs->children, next) {
4198 char *child_key_dot;
4199 child_key_dot = g_strdup_printf("%s.", child->name);
4200 qdict_extract_subqdict(bs->explicit_options, NULL, child_key_dot);
4201 qdict_extract_subqdict(bs->options, NULL, child_key_dot);
50196d7a
AG
4202 qdict_del(bs->explicit_options, child->name);
4203 qdict_del(bs->options, child->name);
2f624b80
AG
4204 g_free(child_key_dot);
4205 }
4206
b6ad491a 4207 /* Check if any unknown options were used */
7ad2757f 4208 if (qdict_size(options) != 0) {
b6ad491a 4209 const QDictEntry *entry = qdict_first(options);
5acd9d81
HR
4210 if (flags & BDRV_O_PROTOCOL) {
4211 error_setg(errp, "Block protocol '%s' doesn't support the option "
4212 "'%s'", drv->format_name, entry->key);
4213 } else {
d0e46a55
HR
4214 error_setg(errp,
4215 "Block format '%s' does not support the option '%s'",
4216 drv->format_name, entry->key);
5acd9d81 4217 }
b6ad491a 4218
b6ad491a
KW
4219 goto close_and_fail;
4220 }
b6ad491a 4221
c01c214b 4222 bdrv_parent_cb_change_media(bs, true);
b6ce07aa 4223
cb3e7f08 4224 qobject_unref(options);
8961be33 4225 options = NULL;
dd62f1ca
KW
4226
4227 /* For snapshot=on, create a temporary qcow2 overlay. bs points to the
4228 * temporary snapshot afterwards. */
4229 if (snapshot_flags) {
66836189
HR
4230 BlockDriverState *snapshot_bs;
4231 snapshot_bs = bdrv_append_temp_snapshot(bs, snapshot_flags,
4232 snapshot_options, &local_err);
73176bee 4233 snapshot_options = NULL;
dd62f1ca
KW
4234 if (local_err) {
4235 goto close_and_fail;
4236 }
5b363937
HR
4237 /* We are not going to return bs but the overlay on top of it
4238 * (snapshot_bs); thus, we have to drop the strong reference to bs
4239 * (which we obtained by calling bdrv_new()). bs will not be deleted,
4240 * though, because the overlay still has a reference to it. */
f665f01f 4241 aio_context_acquire(ctx);
5b363937 4242 bdrv_unref(bs);
f665f01f 4243 aio_context_release(ctx);
5b363937 4244 bs = snapshot_bs;
dd62f1ca
KW
4245 }
4246
5b363937 4247 return bs;
b6ce07aa 4248
8bfea15d 4249fail:
f665f01f 4250 aio_context_acquire(ctx);
5696c6e3 4251 blk_unref(file);
cb3e7f08
MAL
4252 qobject_unref(snapshot_options);
4253 qobject_unref(bs->explicit_options);
4254 qobject_unref(bs->options);
4255 qobject_unref(options);
de9c0cec 4256 bs->options = NULL;
998cbd6a 4257 bs->explicit_options = NULL;
5b363937 4258 bdrv_unref(bs);
f665f01f 4259 aio_context_release(ctx);
621ff94d 4260 error_propagate(errp, local_err);
5b363937 4261 return NULL;
de9c0cec 4262
b6ad491a 4263close_and_fail:
f665f01f 4264 aio_context_acquire(ctx);
5b363937 4265 bdrv_unref(bs);
f665f01f 4266 aio_context_release(ctx);
cb3e7f08
MAL
4267 qobject_unref(snapshot_options);
4268 qobject_unref(options);
621ff94d 4269 error_propagate(errp, local_err);
5b363937 4270 return NULL;
b6ce07aa
KW
4271}
4272
ae400dbb 4273/* The caller must always hold the main AioContext lock. */
5b363937
HR
4274BlockDriverState *bdrv_open(const char *filename, const char *reference,
4275 QDict *options, int flags, Error **errp)
f3930ed0 4276{
f791bf7f
EGE
4277 GLOBAL_STATE_CODE();
4278
5b363937 4279 return bdrv_open_inherit(filename, reference, options, flags, NULL,
272c02ea 4280 NULL, 0, errp);
f3930ed0
KW
4281}
4282
faf116b4
AG
4283/* Return true if the NULL-terminated @list contains @str */
4284static bool is_str_in_list(const char *str, const char *const *list)
4285{
4286 if (str && list) {
4287 int i;
4288 for (i = 0; list[i] != NULL; i++) {
4289 if (!strcmp(str, list[i])) {
4290 return true;
4291 }
4292 }
4293 }
4294 return false;
4295}
4296
4297/*
4298 * Check that every option set in @bs->options is also set in
4299 * @new_opts.
4300 *
4301 * Options listed in the common_options list and in
4302 * @bs->drv->mutable_opts are skipped.
4303 *
4304 * Return 0 on success, otherwise return -EINVAL and set @errp.
4305 */
4306static int bdrv_reset_options_allowed(BlockDriverState *bs,
4307 const QDict *new_opts, Error **errp)
4308{
4309 const QDictEntry *e;
4310 /* These options are common to all block drivers and are handled
4311 * in bdrv_reopen_prepare() so they can be left out of @new_opts */
4312 const char *const common_options[] = {
4313 "node-name", "discard", "cache.direct", "cache.no-flush",
4314 "read-only", "auto-read-only", "detect-zeroes", NULL
4315 };
4316
4317 for (e = qdict_first(bs->options); e; e = qdict_next(bs->options, e)) {
4318 if (!qdict_haskey(new_opts, e->key) &&
4319 !is_str_in_list(e->key, common_options) &&
4320 !is_str_in_list(e->key, bs->drv->mutable_opts)) {
4321 error_setg(errp, "Option '%s' cannot be reset "
4322 "to its default value", e->key);
4323 return -EINVAL;
4324 }
4325 }
4326
4327 return 0;
4328}
4329
cb828c31
AG
4330/*
4331 * Returns true if @child can be reached recursively from @bs
4332 */
ce433d29
KW
4333static bool GRAPH_RDLOCK
4334bdrv_recurse_has_child(BlockDriverState *bs, BlockDriverState *child)
cb828c31
AG
4335{
4336 BdrvChild *c;
4337
4338 if (bs == child) {
4339 return true;
4340 }
4341
4342 QLIST_FOREACH(c, &bs->children, next) {
4343 if (bdrv_recurse_has_child(c->bs, child)) {
4344 return true;
4345 }
4346 }
4347
4348 return false;
4349}
4350
e971aa12
JC
4351/*
4352 * Adds a BlockDriverState to a simple queue for an atomic, transactional
4353 * reopen of multiple devices.
4354 *
859443b0 4355 * bs_queue can either be an existing BlockReopenQueue that has had QTAILQ_INIT
e971aa12
JC
4356 * already performed, or alternatively may be NULL a new BlockReopenQueue will
4357 * be created and initialized. This newly created BlockReopenQueue should be
4358 * passed back in for subsequent calls that are intended to be of the same
4359 * atomic 'set'.
4360 *
4361 * bs is the BlockDriverState to add to the reopen queue.
4362 *
4d2cb092
KW
4363 * options contains the changed options for the associated bs
4364 * (the BlockReopenQueue takes ownership)
4365 *
e971aa12
JC
4366 * flags contains the open flags for the associated bs
4367 *
4368 * returns a pointer to bs_queue, which is either the newly allocated
4369 * bs_queue, or the existing bs_queue being used.
4370 *
d22933ac 4371 * bs is drained here and undrained by bdrv_reopen_queue_free().
2e117866
KW
4372 *
4373 * To be called with bs->aio_context locked.
e971aa12 4374 */
ce433d29
KW
4375static BlockReopenQueue * GRAPH_RDLOCK
4376bdrv_reopen_queue_child(BlockReopenQueue *bs_queue, BlockDriverState *bs,
4377 QDict *options, const BdrvChildClass *klass,
4378 BdrvChildRole role, bool parent_is_format,
4379 QDict *parent_options, int parent_flags,
4380 bool keep_old_opts)
e971aa12
JC
4381{
4382 assert(bs != NULL);
4383
4384 BlockReopenQueueEntry *bs_entry;
67251a31 4385 BdrvChild *child;
9aa09ddd
AG
4386 QDict *old_options, *explicit_options, *options_copy;
4387 int flags;
4388 QemuOpts *opts;
67251a31 4389
f0c28327 4390 GLOBAL_STATE_CODE();
1a63a907 4391
ce433d29
KW
4392 /*
4393 * Strictly speaking, draining is illegal under GRAPH_RDLOCK. We know that
4394 * we've been called with bdrv_graph_rdlock_main_loop(), though, so it's ok
4395 * in practice.
4396 */
d22933ac
KW
4397 bdrv_drained_begin(bs);
4398
e971aa12
JC
4399 if (bs_queue == NULL) {
4400 bs_queue = g_new0(BlockReopenQueue, 1);
859443b0 4401 QTAILQ_INIT(bs_queue);
e971aa12
JC
4402 }
4403
4d2cb092
KW
4404 if (!options) {
4405 options = qdict_new();
4406 }
4407
5b7ba05f 4408 /* Check if this BlockDriverState is already in the queue */
859443b0 4409 QTAILQ_FOREACH(bs_entry, bs_queue, entry) {
5b7ba05f
AG
4410 if (bs == bs_entry->state.bs) {
4411 break;
4412 }
4413 }
4414
28518102
KW
4415 /*
4416 * Precedence of options:
4417 * 1. Explicitly passed in options (highest)
9aa09ddd
AG
4418 * 2. Retained from explicitly set options of bs
4419 * 3. Inherited from parent node
4420 * 4. Retained from effective options of bs
28518102
KW
4421 */
4422
145f598e 4423 /* Old explicitly set values (don't overwrite by inherited value) */
077e8e20
AG
4424 if (bs_entry || keep_old_opts) {
4425 old_options = qdict_clone_shallow(bs_entry ?
4426 bs_entry->state.explicit_options :
4427 bs->explicit_options);
4428 bdrv_join_options(bs, options, old_options);
4429 qobject_unref(old_options);
5b7ba05f 4430 }
145f598e
KW
4431
4432 explicit_options = qdict_clone_shallow(options);
4433
28518102
KW
4434 /* Inherit from parent node */
4435 if (parent_options) {
9aa09ddd 4436 flags = 0;
3cdc69d3 4437 klass->inherit_options(role, parent_is_format, &flags, options,
272c02ea 4438 parent_flags, parent_options);
9aa09ddd
AG
4439 } else {
4440 flags = bdrv_get_flags(bs);
28518102
KW
4441 }
4442
077e8e20
AG
4443 if (keep_old_opts) {
4444 /* Old values are used for options that aren't set yet */
4445 old_options = qdict_clone_shallow(bs->options);
4446 bdrv_join_options(bs, options, old_options);
4447 qobject_unref(old_options);
4448 }
4d2cb092 4449
9aa09ddd
AG
4450 /* We have the final set of options so let's update the flags */
4451 options_copy = qdict_clone_shallow(options);
4452 opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
4453 qemu_opts_absorb_qdict(opts, options_copy, NULL);
4454 update_flags_from_options(&flags, opts);
4455 qemu_opts_del(opts);
4456 qobject_unref(options_copy);
4457
fd452021 4458 /* bdrv_open_inherit() sets and clears some additional flags internally */
f1f25a2e 4459 flags &= ~BDRV_O_PROTOCOL;
fd452021
KW
4460 if (flags & BDRV_O_RDWR) {
4461 flags |= BDRV_O_ALLOW_RDWR;
4462 }
f1f25a2e 4463
1857c97b
KW
4464 if (!bs_entry) {
4465 bs_entry = g_new0(BlockReopenQueueEntry, 1);
859443b0 4466 QTAILQ_INSERT_TAIL(bs_queue, bs_entry, entry);
1857c97b 4467 } else {
cb3e7f08
MAL
4468 qobject_unref(bs_entry->state.options);
4469 qobject_unref(bs_entry->state.explicit_options);
1857c97b
KW
4470 }
4471
4472 bs_entry->state.bs = bs;
4473 bs_entry->state.options = options;
4474 bs_entry->state.explicit_options = explicit_options;
4475 bs_entry->state.flags = flags;
4476
8546632e
AG
4477 /*
4478 * If keep_old_opts is false then it means that unspecified
4479 * options must be reset to their original value. We don't allow
4480 * resetting 'backing' but we need to know if the option is
4481 * missing in order to decide if we have to return an error.
4482 */
4483 if (!keep_old_opts) {
4484 bs_entry->state.backing_missing =
4485 !qdict_haskey(options, "backing") &&
4486 !qdict_haskey(options, "backing.driver");
4487 }
4488
67251a31 4489 QLIST_FOREACH(child, &bs->children, next) {
8546632e
AG
4490 QDict *new_child_options = NULL;
4491 bool child_keep_old = keep_old_opts;
67251a31 4492
4c9dfe5d
KW
4493 /* reopen can only change the options of block devices that were
4494 * implicitly created and inherited options. For other (referenced)
4495 * block devices, a syntax like "backing.foo" results in an error. */
67251a31
KW
4496 if (child->bs->inherits_from != bs) {
4497 continue;
4498 }
4499
8546632e
AG
4500 /* Check if the options contain a child reference */
4501 if (qdict_haskey(options, child->name)) {
4502 const char *childref = qdict_get_try_str(options, child->name);
4503 /*
4504 * The current child must not be reopened if the child
4505 * reference is null or points to a different node.
4506 */
4507 if (g_strcmp0(childref, child->bs->node_name)) {
4508 continue;
4509 }
4510 /*
4511 * If the child reference points to the current child then
4512 * reopen it with its existing set of options (note that
4513 * it can still inherit new options from the parent).
4514 */
4515 child_keep_old = true;
4516 } else {
4517 /* Extract child options ("child-name.*") */
4518 char *child_key_dot = g_strdup_printf("%s.", child->name);
4519 qdict_extract_subqdict(explicit_options, NULL, child_key_dot);
4520 qdict_extract_subqdict(options, &new_child_options, child_key_dot);
4521 g_free(child_key_dot);
4522 }
4c9dfe5d 4523
9aa09ddd 4524 bdrv_reopen_queue_child(bs_queue, child->bs, new_child_options,
3cdc69d3
HR
4525 child->klass, child->role, bs->drv->is_format,
4526 options, flags, child_keep_old);
e971aa12
JC
4527 }
4528
e971aa12
JC
4529 return bs_queue;
4530}
4531
2e117866 4532/* To be called with bs->aio_context locked */
28518102
KW
4533BlockReopenQueue *bdrv_reopen_queue(BlockReopenQueue *bs_queue,
4534 BlockDriverState *bs,
077e8e20 4535 QDict *options, bool keep_old_opts)
28518102 4536{
f791bf7f 4537 GLOBAL_STATE_CODE();
ce433d29 4538 GRAPH_RDLOCK_GUARD_MAINLOOP();
f791bf7f 4539
3cdc69d3
HR
4540 return bdrv_reopen_queue_child(bs_queue, bs, options, NULL, 0, false,
4541 NULL, 0, keep_old_opts);
28518102
KW
4542}
4543
ab5b5228
AG
4544void bdrv_reopen_queue_free(BlockReopenQueue *bs_queue)
4545{
f791bf7f 4546 GLOBAL_STATE_CODE();
ab5b5228
AG
4547 if (bs_queue) {
4548 BlockReopenQueueEntry *bs_entry, *next;
4549 QTAILQ_FOREACH_SAFE(bs_entry, bs_queue, entry, next) {
d22933ac
KW
4550 AioContext *ctx = bdrv_get_aio_context(bs_entry->state.bs);
4551
4552 aio_context_acquire(ctx);
4553 bdrv_drained_end(bs_entry->state.bs);
4554 aio_context_release(ctx);
4555
ab5b5228
AG
4556 qobject_unref(bs_entry->state.explicit_options);
4557 qobject_unref(bs_entry->state.options);
4558 g_free(bs_entry);
4559 }
4560 g_free(bs_queue);
4561 }
4562}
4563
e971aa12
JC
4564/*
4565 * Reopen multiple BlockDriverStates atomically & transactionally.
4566 *
4567 * The queue passed in (bs_queue) must have been built up previous
4568 * via bdrv_reopen_queue().
4569 *
4570 * Reopens all BDS specified in the queue, with the appropriate
4571 * flags. All devices are prepared for reopen, and failure of any
50d6a8a3 4572 * device will cause all device changes to be abandoned, and intermediate
e971aa12
JC
4573 * data cleaned up.
4574 *
4575 * If all devices prepare successfully, then the changes are committed
4576 * to all devices.
4577 *
1a63a907
KW
4578 * All affected nodes must be drained between bdrv_reopen_queue() and
4579 * bdrv_reopen_multiple().
6cf42ca2
KW
4580 *
4581 * To be called from the main thread, with all other AioContexts unlocked.
e971aa12 4582 */
5019aece 4583int bdrv_reopen_multiple(BlockReopenQueue *bs_queue, Error **errp)
e971aa12
JC
4584{
4585 int ret = -1;
4586 BlockReopenQueueEntry *bs_entry, *next;
6cf42ca2 4587 AioContext *ctx;
72373e40 4588 Transaction *tran = tran_new();
72373e40 4589 g_autoptr(GSList) refresh_list = NULL;
e971aa12 4590
6cf42ca2 4591 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
e971aa12 4592 assert(bs_queue != NULL);
da359909 4593 GLOBAL_STATE_CODE();
e971aa12 4594
a2aabf88 4595 QTAILQ_FOREACH(bs_entry, bs_queue, entry) {
6cf42ca2
KW
4596 ctx = bdrv_get_aio_context(bs_entry->state.bs);
4597 aio_context_acquire(ctx);
a2aabf88 4598 ret = bdrv_flush(bs_entry->state.bs);
6cf42ca2 4599 aio_context_release(ctx);
a2aabf88
VSO
4600 if (ret < 0) {
4601 error_setg_errno(errp, -ret, "Error flushing drive");
e3fc91aa 4602 goto abort;
a2aabf88
VSO
4603 }
4604 }
4605
859443b0 4606 QTAILQ_FOREACH(bs_entry, bs_queue, entry) {
1a63a907 4607 assert(bs_entry->state.bs->quiesce_counter > 0);
6cf42ca2
KW
4608 ctx = bdrv_get_aio_context(bs_entry->state.bs);
4609 aio_context_acquire(ctx);
72373e40 4610 ret = bdrv_reopen_prepare(&bs_entry->state, bs_queue, tran, errp);
6cf42ca2 4611 aio_context_release(ctx);
72373e40
VSO
4612 if (ret < 0) {
4613 goto abort;
e971aa12
JC
4614 }
4615 bs_entry->prepared = true;
4616 }
4617
859443b0 4618 QTAILQ_FOREACH(bs_entry, bs_queue, entry) {
69b736e7 4619 BDRVReopenState *state = &bs_entry->state;
72373e40 4620
fb0ff4d1 4621 refresh_list = g_slist_prepend(refresh_list, state->bs);
72373e40 4622 if (state->old_backing_bs) {
fb0ff4d1 4623 refresh_list = g_slist_prepend(refresh_list, state->old_backing_bs);
cb828c31 4624 }
ecd30d2d 4625 if (state->old_file_bs) {
fb0ff4d1 4626 refresh_list = g_slist_prepend(refresh_list, state->old_file_bs);
ecd30d2d 4627 }
72373e40
VSO
4628 }
4629
4630 /*
4631 * Note that file-posix driver rely on permission update done during reopen
4632 * (even if no permission changed), because it wants "new" permissions for
4633 * reconfiguring the fd and that's why it does it in raw_check_perm(), not
4634 * in raw_reopen_prepare() which is called with "old" permissions.
4635 */
3804e3cf 4636 bdrv_graph_rdlock_main_loop();
72373e40 4637 ret = bdrv_list_refresh_perms(refresh_list, bs_queue, tran, errp);
3804e3cf
KW
4638 bdrv_graph_rdunlock_main_loop();
4639
72373e40
VSO
4640 if (ret < 0) {
4641 goto abort;
69b736e7
KW
4642 }
4643
fcd6a4f4
VSO
4644 /*
4645 * If we reach this point, we have success and just need to apply the
4646 * changes.
4647 *
4648 * Reverse order is used to comfort qcow2 driver: on commit it need to write
4649 * IN_USE flag to the image, to mark bitmaps in the image as invalid. But
4650 * children are usually goes after parents in reopen-queue, so go from last
4651 * to first element.
e971aa12 4652 */
fcd6a4f4 4653 QTAILQ_FOREACH_REVERSE(bs_entry, bs_queue, entry) {
6cf42ca2
KW
4654 ctx = bdrv_get_aio_context(bs_entry->state.bs);
4655 aio_context_acquire(ctx);
e971aa12 4656 bdrv_reopen_commit(&bs_entry->state);
6cf42ca2 4657 aio_context_release(ctx);
e971aa12
JC
4658 }
4659
5661a00d 4660 bdrv_graph_wrlock(NULL);
72373e40 4661 tran_commit(tran);
5661a00d 4662 bdrv_graph_wrunlock();
69b736e7 4663
72373e40
VSO
4664 QTAILQ_FOREACH_REVERSE(bs_entry, bs_queue, entry) {
4665 BlockDriverState *bs = bs_entry->state.bs;
74ad9a3b 4666
72373e40 4667 if (bs->drv->bdrv_reopen_commit_post) {
6cf42ca2
KW
4668 ctx = bdrv_get_aio_context(bs);
4669 aio_context_acquire(ctx);
72373e40 4670 bs->drv->bdrv_reopen_commit_post(&bs_entry->state);
6cf42ca2 4671 aio_context_release(ctx);
69b736e7
KW
4672 }
4673 }
17e1e2be 4674
72373e40
VSO
4675 ret = 0;
4676 goto cleanup;
17e1e2be 4677
72373e40 4678abort:
5661a00d 4679 bdrv_graph_wrlock(NULL);
72373e40 4680 tran_abort(tran);
5661a00d 4681 bdrv_graph_wrunlock();
7d4ca9d2 4682
72373e40
VSO
4683 QTAILQ_FOREACH_SAFE(bs_entry, bs_queue, entry, next) {
4684 if (bs_entry->prepared) {
6cf42ca2
KW
4685 ctx = bdrv_get_aio_context(bs_entry->state.bs);
4686 aio_context_acquire(ctx);
72373e40 4687 bdrv_reopen_abort(&bs_entry->state);
6cf42ca2 4688 aio_context_release(ctx);
17e1e2be
PK
4689 }
4690 }
72373e40 4691
e971aa12 4692cleanup:
ab5b5228 4693 bdrv_reopen_queue_free(bs_queue);
40840e41 4694
e971aa12
JC
4695 return ret;
4696}
4697
6cf42ca2
KW
4698int bdrv_reopen(BlockDriverState *bs, QDict *opts, bool keep_old_opts,
4699 Error **errp)
6e1000a8 4700{
6cf42ca2 4701 AioContext *ctx = bdrv_get_aio_context(bs);
6e1000a8 4702 BlockReopenQueue *queue;
6cf42ca2 4703 int ret;
6e1000a8 4704
f791bf7f
EGE
4705 GLOBAL_STATE_CODE();
4706
2e117866
KW
4707 queue = bdrv_reopen_queue(NULL, bs, opts, keep_old_opts);
4708
6cf42ca2
KW
4709 if (ctx != qemu_get_aio_context()) {
4710 aio_context_release(ctx);
4711 }
5019aece 4712 ret = bdrv_reopen_multiple(queue, errp);
6cf42ca2
KW
4713
4714 if (ctx != qemu_get_aio_context()) {
4715 aio_context_acquire(ctx);
4716 }
6e1000a8
AG
4717
4718 return ret;
4719}
4720
6cf42ca2
KW
4721int bdrv_reopen_set_read_only(BlockDriverState *bs, bool read_only,
4722 Error **errp)
4723{
4724 QDict *opts = qdict_new();
4725
f791bf7f
EGE
4726 GLOBAL_STATE_CODE();
4727
6cf42ca2
KW
4728 qdict_put_bool(opts, BDRV_OPT_READ_ONLY, read_only);
4729
4730 return bdrv_reopen(bs, opts, true, errp);
4731}
4732
cb828c31
AG
4733/*
4734 * Take a BDRVReopenState and check if the value of 'backing' in the
4735 * reopen_state->options QDict is valid or not.
4736 *
4737 * If 'backing' is missing from the QDict then return 0.
4738 *
4739 * If 'backing' contains the node name of the backing file of
4740 * reopen_state->bs then return 0.
4741 *
4742 * If 'backing' contains a different node name (or is null) then check
4743 * whether the current backing file can be replaced with the new one.
4744 * If that's the case then reopen_state->replace_backing_bs is set to
4745 * true and reopen_state->new_backing_bs contains a pointer to the new
4746 * backing BlockDriverState (or NULL).
4747 *
5661a00d
KW
4748 * After calling this function, the transaction @tran may only be completed
4749 * while holding a writer lock for the graph.
4750 *
cb828c31 4751 * Return 0 on success, otherwise return < 0 and set @errp.
4b408668
KW
4752 *
4753 * The caller must hold the AioContext lock of @reopen_state->bs.
4754 * @reopen_state->bs can move to a different AioContext in this function.
4755 * Callers must make sure that their AioContext locking is still correct after
4756 * this.
cb828c31 4757 */
ce433d29
KW
4758static int GRAPH_UNLOCKED
4759bdrv_reopen_parse_file_or_backing(BDRVReopenState *reopen_state,
4760 bool is_backing, Transaction *tran,
4761 Error **errp)
cb828c31
AG
4762{
4763 BlockDriverState *bs = reopen_state->bs;
ecd30d2d
AG
4764 BlockDriverState *new_child_bs;
4765 BlockDriverState *old_child_bs = is_backing ? child_bs(bs->backing) :
4766 child_bs(bs->file);
4767 const char *child_name = is_backing ? "backing" : "file";
cb828c31
AG
4768 QObject *value;
4769 const char *str;
4b408668 4770 AioContext *ctx, *old_ctx;
ce433d29 4771 bool has_child;
4b408668 4772 int ret;
cb828c31 4773
bdb73476
EGE
4774 GLOBAL_STATE_CODE();
4775
ecd30d2d 4776 value = qdict_get(reopen_state->options, child_name);
cb828c31
AG
4777 if (value == NULL) {
4778 return 0;
4779 }
4780
4781 switch (qobject_type(value)) {
4782 case QTYPE_QNULL:
ecd30d2d
AG
4783 assert(is_backing); /* The 'file' option does not allow a null value */
4784 new_child_bs = NULL;
cb828c31
AG
4785 break;
4786 case QTYPE_QSTRING:
410f44f5 4787 str = qstring_get_str(qobject_to(QString, value));
ecd30d2d
AG
4788 new_child_bs = bdrv_lookup_bs(NULL, str, errp);
4789 if (new_child_bs == NULL) {
cb828c31 4790 return -EINVAL;
ce433d29
KW
4791 }
4792
4793 bdrv_graph_rdlock_main_loop();
4794 has_child = bdrv_recurse_has_child(new_child_bs, bs);
4795 bdrv_graph_rdunlock_main_loop();
4796
4797 if (has_child) {
ecd30d2d
AG
4798 error_setg(errp, "Making '%s' a %s child of '%s' would create a "
4799 "cycle", str, child_name, bs->node_name);
cb828c31
AG
4800 return -EINVAL;
4801 }
4802 break;
4803 default:
ecd30d2d
AG
4804 /*
4805 * The options QDict has been flattened, so 'backing' and 'file'
4806 * do not allow any other data type here.
4807 */
cb828c31
AG
4808 g_assert_not_reached();
4809 }
4810
ecd30d2d
AG
4811 if (old_child_bs == new_child_bs) {
4812 return 0;
4813 }
4814
4815 if (old_child_bs) {
4816 if (bdrv_skip_implicit_filters(old_child_bs) == new_child_bs) {
cbfdb98c
VSO
4817 return 0;
4818 }
4819
ecd30d2d
AG
4820 if (old_child_bs->implicit) {
4821 error_setg(errp, "Cannot replace implicit %s child of %s",
4822 child_name, bs->node_name);
cbfdb98c
VSO
4823 return -EPERM;
4824 }
4825 }
4826
ecd30d2d 4827 if (bs->drv->is_filter && !old_child_bs) {
25f78d9e
VSO
4828 /*
4829 * Filters always have a file or a backing child, so we are trying to
4830 * change wrong child
4831 */
4832 error_setg(errp, "'%s' is a %s filter node that does not support a "
ecd30d2d 4833 "%s child", bs->node_name, bs->drv->format_name, child_name);
1d42f48c
HR
4834 return -EINVAL;
4835 }
4836
ecd30d2d
AG
4837 if (is_backing) {
4838 reopen_state->old_backing_bs = old_child_bs;
4839 } else {
4840 reopen_state->old_file_bs = old_child_bs;
4841 }
4842
7d4ca9d2
KW
4843 if (old_child_bs) {
4844 bdrv_ref(old_child_bs);
4845 bdrv_drained_begin(old_child_bs);
4846 }
4847
4b408668
KW
4848 old_ctx = bdrv_get_aio_context(bs);
4849 ctx = bdrv_get_aio_context(new_child_bs);
4850 if (old_ctx != ctx) {
4851 aio_context_release(old_ctx);
4852 aio_context_acquire(ctx);
4853 }
4854
7d4ca9d2
KW
4855 bdrv_graph_wrlock(new_child_bs);
4856
4b408668
KW
4857 ret = bdrv_set_file_or_backing_noperm(bs, new_child_bs, is_backing,
4858 tran, errp);
4859
7d4ca9d2
KW
4860 bdrv_graph_wrunlock();
4861
4b408668
KW
4862 if (old_ctx != ctx) {
4863 aio_context_release(ctx);
4864 aio_context_acquire(old_ctx);
4865 }
4866
7d4ca9d2
KW
4867 if (old_child_bs) {
4868 bdrv_drained_end(old_child_bs);
4869 bdrv_unref(old_child_bs);
4870 }
4871
4b408668 4872 return ret;
cb828c31
AG
4873}
4874
e971aa12
JC
4875/*
4876 * Prepares a BlockDriverState for reopen. All changes are staged in the
4877 * 'opaque' field of the BDRVReopenState, which is used and allocated by
4878 * the block driver layer .bdrv_reopen_prepare()
4879 *
4880 * bs is the BlockDriverState to reopen
4881 * flags are the new open flags
4882 * queue is the reopen queue
4883 *
4884 * Returns 0 on success, non-zero on error. On error errp will be set
4885 * as well.
4886 *
4887 * On failure, bdrv_reopen_abort() will be called to clean up any data.
4888 * It is the responsibility of the caller to then call the abort() or
4889 * commit() for any other BDS that have been left in a prepare() state
4890 *
4b408668 4891 * The caller must hold the AioContext lock of @reopen_state->bs.
5661a00d
KW
4892 *
4893 * After calling this function, the transaction @change_child_tran may only be
4894 * completed while holding a writer lock for the graph.
e971aa12 4895 */
ce433d29
KW
4896static int GRAPH_UNLOCKED
4897bdrv_reopen_prepare(BDRVReopenState *reopen_state, BlockReopenQueue *queue,
4898 Transaction *change_child_tran, Error **errp)
e971aa12
JC
4899{
4900 int ret = -1;
e6d79c41 4901 int old_flags;
e971aa12
JC
4902 Error *local_err = NULL;
4903 BlockDriver *drv;
ccf9dc07 4904 QemuOpts *opts;
4c8350fe 4905 QDict *orig_reopen_opts;
593b3071 4906 char *discard = NULL;
3d8ce171 4907 bool read_only;
9ad08c44 4908 bool drv_prepared = false;
e971aa12
JC
4909
4910 assert(reopen_state != NULL);
4911 assert(reopen_state->bs->drv != NULL);
da359909 4912 GLOBAL_STATE_CODE();
e971aa12
JC
4913 drv = reopen_state->bs->drv;
4914
4c8350fe
AG
4915 /* This function and each driver's bdrv_reopen_prepare() remove
4916 * entries from reopen_state->options as they are processed, so
4917 * we need to make a copy of the original QDict. */
4918 orig_reopen_opts = qdict_clone_shallow(reopen_state->options);
4919
ccf9dc07
KW
4920 /* Process generic block layer options */
4921 opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
af175e85 4922 if (!qemu_opts_absorb_qdict(opts, reopen_state->options, errp)) {
ccf9dc07
KW
4923 ret = -EINVAL;
4924 goto error;
4925 }
4926
e6d79c41
AG
4927 /* This was already called in bdrv_reopen_queue_child() so the flags
4928 * are up-to-date. This time we simply want to remove the options from
4929 * QemuOpts in order to indicate that they have been processed. */
4930 old_flags = reopen_state->flags;
91a097e7 4931 update_flags_from_options(&reopen_state->flags, opts);
e6d79c41 4932 assert(old_flags == reopen_state->flags);
91a097e7 4933
415bbca8 4934 discard = qemu_opt_get_del(opts, BDRV_OPT_DISCARD);
593b3071
AG
4935 if (discard != NULL) {
4936 if (bdrv_parse_discard_flags(discard, &reopen_state->flags) != 0) {
4937 error_setg(errp, "Invalid discard option");
4938 ret = -EINVAL;
4939 goto error;
4940 }
4941 }
4942
543770bd
AG
4943 reopen_state->detect_zeroes =
4944 bdrv_parse_detect_zeroes(opts, reopen_state->flags, &local_err);
4945 if (local_err) {
4946 error_propagate(errp, local_err);
4947 ret = -EINVAL;
4948 goto error;
4949 }
4950
57f9db9a
AG
4951 /* All other options (including node-name and driver) must be unchanged.
4952 * Put them back into the QDict, so that they are checked at the end
4953 * of this function. */
4954 qemu_opts_to_qdict(opts, reopen_state->options);
ccf9dc07 4955
3d8ce171
JC
4956 /* If we are to stay read-only, do not allow permission change
4957 * to r/w. Attempting to set to r/w may fail if either BDRV_O_ALLOW_RDWR is
4958 * not set, or if the BDS still has copy_on_read enabled */
4959 read_only = !(reopen_state->flags & BDRV_O_RDWR);
4026f1c4
KW
4960
4961 bdrv_graph_rdlock_main_loop();
54a32bfe 4962 ret = bdrv_can_set_read_only(reopen_state->bs, read_only, true, &local_err);
4026f1c4 4963 bdrv_graph_rdunlock_main_loop();
3d8ce171
JC
4964 if (local_err) {
4965 error_propagate(errp, local_err);
e971aa12
JC
4966 goto error;
4967 }
4968
e971aa12 4969 if (drv->bdrv_reopen_prepare) {
faf116b4
AG
4970 /*
4971 * If a driver-specific option is missing, it means that we
4972 * should reset it to its default value.
4973 * But not all options allow that, so we need to check it first.
4974 */
4975 ret = bdrv_reset_options_allowed(reopen_state->bs,
4976 reopen_state->options, errp);
4977 if (ret) {
4978 goto error;
4979 }
4980
e971aa12
JC
4981 ret = drv->bdrv_reopen_prepare(reopen_state, queue, &local_err);
4982 if (ret) {
4983 if (local_err != NULL) {
4984 error_propagate(errp, local_err);
4985 } else {
b7cfc7d5 4986 bdrv_graph_rdlock_main_loop();
f30c66ba 4987 bdrv_refresh_filename(reopen_state->bs);
b7cfc7d5 4988 bdrv_graph_rdunlock_main_loop();
d8b6895f
LC
4989 error_setg(errp, "failed while preparing to reopen image '%s'",
4990 reopen_state->bs->filename);
e971aa12
JC
4991 }
4992 goto error;
4993 }
4994 } else {
4995 /* It is currently mandatory to have a bdrv_reopen_prepare()
4996 * handler for each supported drv. */
4026f1c4 4997 bdrv_graph_rdlock_main_loop();
81e5f78a
AG
4998 error_setg(errp, "Block format '%s' used by node '%s' "
4999 "does not support reopening files", drv->format_name,
5000 bdrv_get_device_or_node_name(reopen_state->bs));
4026f1c4 5001 bdrv_graph_rdunlock_main_loop();
e971aa12
JC
5002 ret = -1;
5003 goto error;
5004 }
5005
9ad08c44
HR
5006 drv_prepared = true;
5007
bacd9b87
AG
5008 /*
5009 * We must provide the 'backing' option if the BDS has a backing
5010 * file or if the image file has a backing file name as part of
5011 * its metadata. Otherwise the 'backing' option can be omitted.
5012 */
5013 if (drv->supports_backing && reopen_state->backing_missing &&
1d42f48c 5014 (reopen_state->bs->backing || reopen_state->bs->backing_file[0])) {
8546632e
AG
5015 error_setg(errp, "backing is missing for '%s'",
5016 reopen_state->bs->node_name);
5017 ret = -EINVAL;
5018 goto error;
5019 }
5020
cb828c31
AG
5021 /*
5022 * Allow changing the 'backing' option. The new value can be
5023 * either a reference to an existing node (using its node name)
5024 * or NULL to simply detach the current backing file.
5025 */
ecd30d2d
AG
5026 ret = bdrv_reopen_parse_file_or_backing(reopen_state, true,
5027 change_child_tran, errp);
cb828c31
AG
5028 if (ret < 0) {
5029 goto error;
5030 }
5031 qdict_del(reopen_state->options, "backing");
5032
ecd30d2d
AG
5033 /* Allow changing the 'file' option. In this case NULL is not allowed */
5034 ret = bdrv_reopen_parse_file_or_backing(reopen_state, false,
5035 change_child_tran, errp);
5036 if (ret < 0) {
5037 goto error;
5038 }
5039 qdict_del(reopen_state->options, "file");
5040
4d2cb092
KW
5041 /* Options that are not handled are only okay if they are unchanged
5042 * compared to the old state. It is expected that some options are only
5043 * used for the initial open, but not reopen (e.g. filename) */
5044 if (qdict_size(reopen_state->options)) {
5045 const QDictEntry *entry = qdict_first(reopen_state->options);
5046
ce433d29
KW
5047 GRAPH_RDLOCK_GUARD_MAINLOOP();
5048
4d2cb092 5049 do {
54fd1b0d
HR
5050 QObject *new = entry->value;
5051 QObject *old = qdict_get(reopen_state->bs->options, entry->key);
5052
db905283
AG
5053 /* Allow child references (child_name=node_name) as long as they
5054 * point to the current child (i.e. everything stays the same). */
5055 if (qobject_type(new) == QTYPE_QSTRING) {
5056 BdrvChild *child;
5057 QLIST_FOREACH(child, &reopen_state->bs->children, next) {
5058 if (!strcmp(child->name, entry->key)) {
5059 break;
5060 }
5061 }
5062
5063 if (child) {
410f44f5
MA
5064 if (!strcmp(child->bs->node_name,
5065 qstring_get_str(qobject_to(QString, new)))) {
db905283
AG
5066 continue; /* Found child with this name, skip option */
5067 }
5068 }
5069 }
5070
129c7d1c 5071 /*
54fd1b0d
HR
5072 * TODO: When using -drive to specify blockdev options, all values
5073 * will be strings; however, when using -blockdev, blockdev-add or
5074 * filenames using the json:{} pseudo-protocol, they will be
5075 * correctly typed.
5076 * In contrast, reopening options are (currently) always strings
5077 * (because you can only specify them through qemu-io; all other
5078 * callers do not specify any options).
5079 * Therefore, when using anything other than -drive to create a BDS,
5080 * this cannot detect non-string options as unchanged, because
5081 * qobject_is_equal() always returns false for objects of different
5082 * type. In the future, this should be remedied by correctly typing
5083 * all options. For now, this is not too big of an issue because
5084 * the user can simply omit options which cannot be changed anyway,
5085 * so they will stay unchanged.
129c7d1c 5086 */
54fd1b0d 5087 if (!qobject_is_equal(new, old)) {
4d2cb092
KW
5088 error_setg(errp, "Cannot change the option '%s'", entry->key);
5089 ret = -EINVAL;
5090 goto error;
5091 }
5092 } while ((entry = qdict_next(reopen_state->options, entry)));
5093 }
5094
e971aa12
JC
5095 ret = 0;
5096
4c8350fe
AG
5097 /* Restore the original reopen_state->options QDict */
5098 qobject_unref(reopen_state->options);
5099 reopen_state->options = qobject_ref(orig_reopen_opts);
5100
e971aa12 5101error:
9ad08c44
HR
5102 if (ret < 0 && drv_prepared) {
5103 /* drv->bdrv_reopen_prepare() has succeeded, so we need to
5104 * call drv->bdrv_reopen_abort() before signaling an error
5105 * (bdrv_reopen_multiple() will not call bdrv_reopen_abort()
5106 * when the respective bdrv_reopen_prepare() has failed) */
5107 if (drv->bdrv_reopen_abort) {
5108 drv->bdrv_reopen_abort(reopen_state);
5109 }
5110 }
ccf9dc07 5111 qemu_opts_del(opts);
4c8350fe 5112 qobject_unref(orig_reopen_opts);
593b3071 5113 g_free(discard);
e971aa12
JC
5114 return ret;
5115}
5116
5117/*
5118 * Takes the staged changes for the reopen from bdrv_reopen_prepare(), and
5119 * makes them final by swapping the staging BlockDriverState contents into
5120 * the active BlockDriverState contents.
5121 */
ce433d29 5122static void GRAPH_UNLOCKED bdrv_reopen_commit(BDRVReopenState *reopen_state)
e971aa12
JC
5123{
5124 BlockDriver *drv;
50bf65ba 5125 BlockDriverState *bs;
50196d7a 5126 BdrvChild *child;
e971aa12
JC
5127
5128 assert(reopen_state != NULL);
50bf65ba
VSO
5129 bs = reopen_state->bs;
5130 drv = bs->drv;
e971aa12 5131 assert(drv != NULL);
da359909 5132 GLOBAL_STATE_CODE();
e971aa12
JC
5133
5134 /* If there are any driver level actions to take */
5135 if (drv->bdrv_reopen_commit) {
5136 drv->bdrv_reopen_commit(reopen_state);
5137 }
5138
ce433d29
KW
5139 GRAPH_RDLOCK_GUARD_MAINLOOP();
5140
e971aa12 5141 /* set BDS specific flags now */
cb3e7f08 5142 qobject_unref(bs->explicit_options);
4c8350fe 5143 qobject_unref(bs->options);
ab5b5228
AG
5144 qobject_ref(reopen_state->explicit_options);
5145 qobject_ref(reopen_state->options);
145f598e 5146
50bf65ba 5147 bs->explicit_options = reopen_state->explicit_options;
4c8350fe 5148 bs->options = reopen_state->options;
50bf65ba 5149 bs->open_flags = reopen_state->flags;
543770bd 5150 bs->detect_zeroes = reopen_state->detect_zeroes;
355ef4ac 5151
50196d7a
AG
5152 /* Remove child references from bs->options and bs->explicit_options.
5153 * Child options were already removed in bdrv_reopen_queue_child() */
5154 QLIST_FOREACH(child, &bs->children, next) {
5155 qdict_del(bs->explicit_options, child->name);
5156 qdict_del(bs->options, child->name);
5157 }
3d0e8743
VSO
5158 /* backing is probably removed, so it's not handled by previous loop */
5159 qdict_del(bs->explicit_options, "backing");
5160 qdict_del(bs->options, "backing");
5161
1e4c797c 5162 bdrv_refresh_limits(bs, NULL, NULL);
439cc330 5163 bdrv_refresh_total_sectors(bs, bs->total_sectors);
e971aa12
JC
5164}
5165
5166/*
5167 * Abort the reopen, and delete and free the staged changes in
5168 * reopen_state
5169 */
ce433d29 5170static void GRAPH_UNLOCKED bdrv_reopen_abort(BDRVReopenState *reopen_state)
e971aa12
JC
5171{
5172 BlockDriver *drv;
5173
5174 assert(reopen_state != NULL);
5175 drv = reopen_state->bs->drv;
5176 assert(drv != NULL);
da359909 5177 GLOBAL_STATE_CODE();
e971aa12
JC
5178
5179 if (drv->bdrv_reopen_abort) {
5180 drv->bdrv_reopen_abort(reopen_state);
5181 }
5182}
5183
5184
64dff520 5185static void bdrv_close(BlockDriverState *bs)
fc01f7e7 5186{
33384421 5187 BdrvAioNotifier *ban, *ban_next;
50a3efb0 5188 BdrvChild *child, *next;
33384421 5189
f791bf7f 5190 GLOBAL_STATE_CODE();
30f55fb8 5191 assert(!bs->refcnt);
99b7e775 5192
fc27291d 5193 bdrv_drained_begin(bs); /* complete I/O */
58fda173 5194 bdrv_flush(bs);
53ec73e2 5195 bdrv_drain(bs); /* in case flush left pending I/O */
fc27291d 5196
3cbc002c 5197 if (bs->drv) {
3c005293 5198 if (bs->drv->bdrv_close) {
7b99a266 5199 /* Must unfreeze all children, so bdrv_unref_child() works */
3c005293
VSO
5200 bs->drv->bdrv_close(bs);
5201 }
9a4f4c31 5202 bs->drv = NULL;
50a3efb0 5203 }
9a7dedbc 5204
e462c6d2 5205 bdrv_graph_wrlock(bs);
50a3efb0 5206 QLIST_FOREACH_SAFE(child, &bs->children, next, next) {
dd4118c7 5207 bdrv_unref_child(bs, child);
b338082b 5208 }
32a8aba3 5209 bdrv_graph_wrunlock();
98f90dba 5210
5bb04747
VSO
5211 assert(!bs->backing);
5212 assert(!bs->file);
50a3efb0
AG
5213 g_free(bs->opaque);
5214 bs->opaque = NULL;
d73415a3 5215 qatomic_set(&bs->copy_on_read, 0);
50a3efb0
AG
5216 bs->backing_file[0] = '\0';
5217 bs->backing_format[0] = '\0';
5218 bs->total_sectors = 0;
5219 bs->encrypted = false;
5220 bs->sg = false;
cb3e7f08
MAL
5221 qobject_unref(bs->options);
5222 qobject_unref(bs->explicit_options);
50a3efb0
AG
5223 bs->options = NULL;
5224 bs->explicit_options = NULL;
cb3e7f08 5225 qobject_unref(bs->full_open_options);
50a3efb0 5226 bs->full_open_options = NULL;
0bc329fb
HR
5227 g_free(bs->block_status_cache);
5228 bs->block_status_cache = NULL;
50a3efb0 5229
cca43ae1
VSO
5230 bdrv_release_named_dirty_bitmaps(bs);
5231 assert(QLIST_EMPTY(&bs->dirty_bitmaps));
5232
33384421
HR
5233 QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_next) {
5234 g_free(ban);
5235 }
5236 QLIST_INIT(&bs->aio_notifiers);
fc27291d 5237 bdrv_drained_end(bs);
1a6d3bd2
GK
5238
5239 /*
5240 * If we're still inside some bdrv_drain_all_begin()/end() sections, end
5241 * them now since this BDS won't exist anymore when bdrv_drain_all_end()
5242 * gets called.
5243 */
5244 if (bs->quiesce_counter) {
5245 bdrv_drain_all_end_quiesce(bs);
5246 }
b338082b
FB
5247}
5248
2bc93fed
MK
5249void bdrv_close_all(void)
5250{
f791bf7f 5251 GLOBAL_STATE_CODE();
880eeec6 5252 assert(job_next(NULL) == NULL);
ca9bd24c
HR
5253
5254 /* Drop references from requests still in flight, such as canceled block
5255 * jobs whose AIO context has not been polled yet */
5256 bdrv_drain_all();
2bc93fed 5257
ca9bd24c
HR
5258 blk_remove_all_bs();
5259 blockdev_close_all_bdrv_states();
ed78cda3 5260
a1a2af07 5261 assert(QTAILQ_EMPTY(&all_bdrv_states));
2bc93fed
MK
5262}
5263
2f64e1fc 5264static bool GRAPH_RDLOCK should_update_child(BdrvChild *c, BlockDriverState *to)
d0ac0380 5265{
2f30b7c3
VSO
5266 GQueue *queue;
5267 GHashTable *found;
5268 bool ret;
d0ac0380 5269
bd86fb99 5270 if (c->klass->stay_at_node) {
d0ac0380
KW
5271 return false;
5272 }
5273
ec9f10fe
HR
5274 /* If the child @c belongs to the BDS @to, replacing the current
5275 * c->bs by @to would mean to create a loop.
5276 *
5277 * Such a case occurs when appending a BDS to a backing chain.
5278 * For instance, imagine the following chain:
5279 *
5280 * guest device -> node A -> further backing chain...
5281 *
5282 * Now we create a new BDS B which we want to put on top of this
5283 * chain, so we first attach A as its backing node:
5284 *
5285 * node B
5286 * |
5287 * v
5288 * guest device -> node A -> further backing chain...
5289 *
5290 * Finally we want to replace A by B. When doing that, we want to
5291 * replace all pointers to A by pointers to B -- except for the
5292 * pointer from B because (1) that would create a loop, and (2)
5293 * that pointer should simply stay intact:
5294 *
5295 * guest device -> node B
5296 * |
5297 * v
5298 * node A -> further backing chain...
5299 *
5300 * In general, when replacing a node A (c->bs) by a node B (@to),
5301 * if A is a child of B, that means we cannot replace A by B there
5302 * because that would create a loop. Silently detaching A from B
5303 * is also not really an option. So overall just leaving A in
2f30b7c3
VSO
5304 * place there is the most sensible choice.
5305 *
5306 * We would also create a loop in any cases where @c is only
5307 * indirectly referenced by @to. Prevent this by returning false
5308 * if @c is found (by breadth-first search) anywhere in the whole
5309 * subtree of @to.
5310 */
5311
5312 ret = true;
5313 found = g_hash_table_new(NULL, NULL);
5314 g_hash_table_add(found, to);
5315 queue = g_queue_new();
5316 g_queue_push_tail(queue, to);
5317
5318 while (!g_queue_is_empty(queue)) {
5319 BlockDriverState *v = g_queue_pop_head(queue);
5320 BdrvChild *c2;
5321
5322 QLIST_FOREACH(c2, &v->children, next) {
5323 if (c2 == c) {
5324 ret = false;
5325 break;
5326 }
5327
5328 if (g_hash_table_contains(found, c2->bs)) {
5329 continue;
5330 }
5331
5332 g_queue_push_tail(queue, c2->bs);
5333 g_hash_table_add(found, c2->bs);
d0ac0380
KW
5334 }
5335 }
5336
2f30b7c3
VSO
5337 g_queue_free(queue);
5338 g_hash_table_destroy(found);
5339
5340 return ret;
d0ac0380
KW
5341}
5342
57f08941 5343static void bdrv_remove_child_commit(void *opaque)
46541ee5 5344{
bdb73476 5345 GLOBAL_STATE_CODE();
5bb04747 5346 bdrv_child_free(opaque);
82b54cf5
HR
5347}
5348
57f08941
VSO
5349static TransactionActionDrv bdrv_remove_child_drv = {
5350 .commit = bdrv_remove_child_commit,
46541ee5
VSO
5351};
5352
2f64e1fc
KW
5353/*
5354 * Function doesn't update permissions, caller is responsible for this.
5355 *
5356 * @child->bs (if non-NULL) must be drained.
5661a00d
KW
5357 *
5358 * After calling this function, the transaction @tran may only be completed
5359 * while holding a writer lock for the graph.
2f64e1fc
KW
5360 */
5361static void GRAPH_WRLOCK bdrv_remove_child(BdrvChild *child, Transaction *tran)
46541ee5 5362{
46541ee5
VSO
5363 if (!child) {
5364 return;
5365 }
5366
5367 if (child->bs) {
2f64e1fc 5368 assert(child->quiesced_parent);
a2c37a30 5369 bdrv_replace_child_tran(child, NULL, tran);
46541ee5
VSO
5370 }
5371
57f08941 5372 tran_add(tran, &bdrv_remove_child_drv, child);
46541ee5
VSO
5373}
5374
2f64e1fc
KW
5375/*
5376 * Both @from and @to (if non-NULL) must be drained. @to must be kept drained
5377 * until the transaction is completed.
5661a00d
KW
5378 *
5379 * After calling this function, the transaction @tran may only be completed
5380 * while holding a writer lock for the graph.
2f64e1fc
KW
5381 */
5382static int GRAPH_WRLOCK
5383bdrv_replace_node_noperm(BlockDriverState *from,
5384 BlockDriverState *to,
5385 bool auto_skip, Transaction *tran,
5386 Error **errp)
117caba9
VSO
5387{
5388 BdrvChild *c, *next;
5389
bdb73476 5390 GLOBAL_STATE_CODE();
82b54cf5 5391
2f64e1fc
KW
5392 assert(from->quiesce_counter);
5393 assert(to->quiesce_counter);
23987471 5394
117caba9
VSO
5395 QLIST_FOREACH_SAFE(c, &from->parents, next_parent, next) {
5396 assert(c->bs == from);
5397 if (!should_update_child(c, to)) {
5398 if (auto_skip) {
5399 continue;
5400 }
5401 error_setg(errp, "Should not change '%s' link to '%s'",
5402 c->name, from->node_name);
5403 return -EINVAL;
5404 }
5405 if (c->frozen) {
5406 error_setg(errp, "Cannot change '%s' link to '%s'",
5407 c->name, from->node_name);
5408 return -EPERM;
5409 }
0f0b1e29 5410 bdrv_replace_child_tran(c, to, tran);
117caba9
VSO
5411 }
5412
5413 return 0;
5414}
5415
313274bb
VSO
5416/*
5417 * With auto_skip=true bdrv_replace_node_common skips updating from parents
5418 * if it creates a parent-child relation loop or if parent is block-job.
5419 *
5420 * With auto_skip=false the error is returned if from has a parent which should
5421 * not be updated.
3108a15c
VSO
5422 *
5423 * With @detach_subchain=true @to must be in a backing chain of @from. In this
5424 * case backing link of the cow-parent of @to is removed.
313274bb 5425 */
a1e708fc
VSO
5426static int bdrv_replace_node_common(BlockDriverState *from,
5427 BlockDriverState *to,
3108a15c
VSO
5428 bool auto_skip, bool detach_subchain,
5429 Error **errp)
dd62f1ca 5430{
3bb0e298 5431 Transaction *tran = tran_new();
3bb0e298 5432 g_autoptr(GSList) refresh_list = NULL;
2d369d6e 5433 BlockDriverState *to_cow_parent = NULL;
234ac1a9
KW
5434 int ret;
5435
bdb73476 5436 GLOBAL_STATE_CODE();
82b54cf5 5437
234ac1a9
KW
5438 /* Make sure that @from doesn't go away until we have successfully attached
5439 * all of its parents to @to. */
5440 bdrv_ref(from);
dd62f1ca 5441
f871abd6 5442 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
30dd65f3 5443 assert(bdrv_get_aio_context(from) == bdrv_get_aio_context(to));
f871abd6 5444 bdrv_drained_begin(from);
2f64e1fc
KW
5445 bdrv_drained_begin(to);
5446
5447 bdrv_graph_wrlock(to);
f871abd6 5448
372b69f5
KW
5449 if (detach_subchain) {
5450 assert(bdrv_chain_contains(from, to));
5451 assert(from != to);
5452 for (to_cow_parent = from;
5453 bdrv_filter_or_cow_bs(to_cow_parent) != to;
5454 to_cow_parent = bdrv_filter_or_cow_bs(to_cow_parent))
5455 {
5456 ;
5457 }
5458 }
5459
3bb0e298
VSO
5460 /*
5461 * Do the replacement without permission update.
5462 * Replacement may influence the permissions, we should calculate new
5463 * permissions based on new graph. If we fail, we'll roll-back the
5464 * replacement.
5465 */
117caba9
VSO
5466 ret = bdrv_replace_node_noperm(from, to, auto_skip, tran, errp);
5467 if (ret < 0) {
5468 goto out;
234ac1a9
KW
5469 }
5470
3108a15c 5471 if (detach_subchain) {
2f64e1fc 5472 /* to_cow_parent is already drained because from is drained */
f38eaec4 5473 bdrv_remove_child(bdrv_filter_or_cow_child(to_cow_parent), tran);
3108a15c
VSO
5474 }
5475
fb0ff4d1
VSO
5476 refresh_list = g_slist_prepend(refresh_list, to);
5477 refresh_list = g_slist_prepend(refresh_list, from);
9bd910e2 5478
3bb0e298
VSO
5479 ret = bdrv_list_refresh_perms(refresh_list, NULL, tran, errp);
5480 if (ret < 0) {
5481 goto out;
dd62f1ca 5482 }
234ac1a9 5483
a1e708fc
VSO
5484 ret = 0;
5485
234ac1a9 5486out:
3bb0e298 5487 tran_finalize(tran, ret);
5661a00d 5488 bdrv_graph_wrunlock();
3bb0e298 5489
2f64e1fc 5490 bdrv_drained_end(to);
f871abd6 5491 bdrv_drained_end(from);
234ac1a9 5492 bdrv_unref(from);
a1e708fc
VSO
5493
5494 return ret;
dd62f1ca
KW
5495}
5496
a1e708fc
VSO
5497int bdrv_replace_node(BlockDriverState *from, BlockDriverState *to,
5498 Error **errp)
313274bb 5499{
f791bf7f
EGE
5500 GLOBAL_STATE_CODE();
5501
3108a15c
VSO
5502 return bdrv_replace_node_common(from, to, true, false, errp);
5503}
5504
5505int bdrv_drop_filter(BlockDriverState *bs, Error **errp)
5506{
372b69f5
KW
5507 BlockDriverState *child_bs;
5508
f791bf7f 5509 GLOBAL_STATE_CODE();
372b69f5
KW
5510 bdrv_graph_rdlock_main_loop();
5511 child_bs = bdrv_filter_or_cow_bs(bs);
5512 bdrv_graph_rdunlock_main_loop();
f791bf7f 5513
372b69f5 5514 return bdrv_replace_node_common(bs, child_bs, true, true, errp);
313274bb
VSO
5515}
5516
4ddc07ca
PB
5517/*
5518 * Add new bs contents at the top of an image chain while the chain is
5519 * live, while keeping required fields on the top layer.
5520 *
5521 * This will modify the BlockDriverState fields, and swap contents
5522 * between bs_new and bs_top. Both bs_new and bs_top are modified.
5523 *
2272edcf
VSO
5524 * bs_new must not be attached to a BlockBackend and must not have backing
5525 * child.
4ddc07ca
PB
5526 *
5527 * This function does not create any image files.
60d90bf4
SG
5528 *
5529 * The caller must hold the AioContext lock for @bs_top.
4ddc07ca 5530 */
a1e708fc
VSO
5531int bdrv_append(BlockDriverState *bs_new, BlockDriverState *bs_top,
5532 Error **errp)
4ddc07ca 5533{
2272edcf 5534 int ret;
5bb04747 5535 BdrvChild *child;
2272edcf 5536 Transaction *tran = tran_new();
60d90bf4 5537 AioContext *old_context, *new_context = NULL;
2272edcf 5538
f791bf7f
EGE
5539 GLOBAL_STATE_CODE();
5540
2272edcf
VSO
5541 assert(!bs_new->backing);
5542
60d90bf4 5543 old_context = bdrv_get_aio_context(bs_top);
7d4ca9d2
KW
5544 bdrv_drained_begin(bs_top);
5545
5546 /*
5547 * bdrv_drained_begin() requires that only the AioContext of the drained
5548 * node is locked, and at this point it can still differ from the AioContext
5549 * of bs_top.
5550 */
5551 new_context = bdrv_get_aio_context(bs_new);
5552 aio_context_release(old_context);
5553 aio_context_acquire(new_context);
5554 bdrv_drained_begin(bs_new);
5555 aio_context_release(new_context);
5556 aio_context_acquire(old_context);
5557 new_context = NULL;
5558
5559 bdrv_graph_wrlock(bs_top);
60d90bf4 5560
5bb04747
VSO
5561 child = bdrv_attach_child_noperm(bs_new, bs_top, "backing",
5562 &child_of_bds, bdrv_backing_role(bs_new),
5563 tran, errp);
5564 if (!child) {
5565 ret = -EINVAL;
2272edcf 5566 goto out;
b2c2832c 5567 }
dd62f1ca 5568
60d90bf4 5569 /*
7d4ca9d2
KW
5570 * bdrv_attach_child_noperm could change the AioContext of bs_top and
5571 * bs_new, but at least they are in the same AioContext now. This is the
5572 * AioContext that we need to lock for the rest of the function.
60d90bf4
SG
5573 */
5574 new_context = bdrv_get_aio_context(bs_top);
5575
5576 if (old_context != new_context) {
5577 aio_context_release(old_context);
5578 aio_context_acquire(new_context);
5579 }
5580
2272edcf 5581 ret = bdrv_replace_node_noperm(bs_top, bs_new, true, tran, errp);
a1e708fc 5582 if (ret < 0) {
2272edcf 5583 goto out;
234ac1a9 5584 }
4ddc07ca 5585
f1316edb 5586 ret = bdrv_refresh_perms(bs_new, tran, errp);
2272edcf
VSO
5587out:
5588 tran_finalize(tran, ret);
5589
1e4c797c 5590 bdrv_refresh_limits(bs_top, NULL, NULL);
5661a00d 5591 bdrv_graph_wrunlock();
2272edcf 5592
7d4ca9d2
KW
5593 bdrv_drained_end(bs_top);
5594 bdrv_drained_end(bs_new);
2f64e1fc 5595
60d90bf4
SG
5596 if (new_context && old_context != new_context) {
5597 aio_context_release(new_context);
5598 aio_context_acquire(old_context);
5599 }
5600
2272edcf 5601 return ret;
8802d1fd
JC
5602}
5603
bd8f4c42
VSO
5604/* Not for empty child */
5605int bdrv_replace_child_bs(BdrvChild *child, BlockDriverState *new_bs,
5606 Error **errp)
5607{
5608 int ret;
5609 Transaction *tran = tran_new();
bd8f4c42
VSO
5610 g_autoptr(GSList) refresh_list = NULL;
5611 BlockDriverState *old_bs = child->bs;
5612
f791bf7f
EGE
5613 GLOBAL_STATE_CODE();
5614
bd8f4c42
VSO
5615 bdrv_ref(old_bs);
5616 bdrv_drained_begin(old_bs);
5617 bdrv_drained_begin(new_bs);
2f64e1fc 5618 bdrv_graph_wrlock(new_bs);
bd8f4c42 5619
0f0b1e29 5620 bdrv_replace_child_tran(child, new_bs, tran);
bd8f4c42 5621
fb0ff4d1
VSO
5622 refresh_list = g_slist_prepend(refresh_list, old_bs);
5623 refresh_list = g_slist_prepend(refresh_list, new_bs);
bd8f4c42
VSO
5624
5625 ret = bdrv_list_refresh_perms(refresh_list, NULL, tran, errp);
5626
5627 tran_finalize(tran, ret);
5628
5661a00d 5629 bdrv_graph_wrunlock();
bd8f4c42
VSO
5630 bdrv_drained_end(old_bs);
5631 bdrv_drained_end(new_bs);
5632 bdrv_unref(old_bs);
5633
5634 return ret;
5635}
5636
4f6fd349 5637static void bdrv_delete(BlockDriverState *bs)
b338082b 5638{
3718d8ab 5639 assert(bdrv_op_blocker_is_empty(bs));
4f6fd349 5640 assert(!bs->refcnt);
f791bf7f 5641 GLOBAL_STATE_CODE();
18846dee 5642
1b7bdbc1 5643 /* remove from list, if necessary */
63eaaae0
KW
5644 if (bs->node_name[0] != '\0') {
5645 QTAILQ_REMOVE(&graph_bdrv_states, bs, node_list);
5646 }
2c1d04e0
HR
5647 QTAILQ_REMOVE(&all_bdrv_states, bs, bs_list);
5648
30c321f9
AK
5649 bdrv_close(bs);
5650
fa9185fc
SH
5651 qemu_mutex_destroy(&bs->reqs_lock);
5652
7267c094 5653 g_free(bs);
fc01f7e7
FB
5654}
5655
96796fae
VSO
5656
5657/*
5658 * Replace @bs by newly created block node.
5659 *
5660 * @options is a QDict of options to pass to the block drivers, or NULL for an
5661 * empty set of options. The reference to the QDict belongs to the block layer
5662 * after the call (even on failure), so if the caller intends to reuse the
5663 * dictionary, it needs to use qobject_ref() before calling bdrv_open.
8823407c
KW
5664 *
5665 * The caller holds the AioContext lock for @bs. It must make sure that @bs
5666 * stays in the same AioContext, i.e. @options must not refer to nodes in a
5667 * different AioContext.
96796fae
VSO
5668 */
5669BlockDriverState *bdrv_insert_node(BlockDriverState *bs, QDict *options,
8872ef78
AS
5670 int flags, Error **errp)
5671{
f053b7e8
VSO
5672 ERRP_GUARD();
5673 int ret;
8823407c 5674 AioContext *ctx = bdrv_get_aio_context(bs);
b11c8739
VSO
5675 BlockDriverState *new_node_bs = NULL;
5676 const char *drvname, *node_name;
5677 BlockDriver *drv;
5678
5679 drvname = qdict_get_try_str(options, "driver");
5680 if (!drvname) {
5681 error_setg(errp, "driver is not specified");
5682 goto fail;
5683 }
5684
5685 drv = bdrv_find_format(drvname);
5686 if (!drv) {
5687 error_setg(errp, "Unknown driver: '%s'", drvname);
5688 goto fail;
5689 }
8872ef78 5690
b11c8739
VSO
5691 node_name = qdict_get_try_str(options, "node-name");
5692
f791bf7f
EGE
5693 GLOBAL_STATE_CODE();
5694
8823407c
KW
5695 aio_context_release(ctx);
5696 aio_context_acquire(qemu_get_aio_context());
b11c8739
VSO
5697 new_node_bs = bdrv_new_open_driver_opts(drv, node_name, options, flags,
5698 errp);
8823407c
KW
5699 aio_context_release(qemu_get_aio_context());
5700 aio_context_acquire(ctx);
5701 assert(bdrv_get_aio_context(bs) == ctx);
5702
b11c8739
VSO
5703 options = NULL; /* bdrv_new_open_driver() eats options */
5704 if (!new_node_bs) {
8872ef78 5705 error_prepend(errp, "Could not create node: ");
b11c8739 5706 goto fail;
8872ef78
AS
5707 }
5708
5709 bdrv_drained_begin(bs);
f053b7e8 5710 ret = bdrv_replace_node(bs, new_node_bs, errp);
8872ef78
AS
5711 bdrv_drained_end(bs);
5712
f053b7e8
VSO
5713 if (ret < 0) {
5714 error_prepend(errp, "Could not replace node: ");
b11c8739 5715 goto fail;
8872ef78
AS
5716 }
5717
5718 return new_node_bs;
b11c8739
VSO
5719
5720fail:
5721 qobject_unref(options);
5722 bdrv_unref(new_node_bs);
5723 return NULL;
8872ef78
AS
5724}
5725
e97fc193
AL
5726/*
5727 * Run consistency checks on an image
5728 *
e076f338 5729 * Returns 0 if the check could be completed (it doesn't mean that the image is
a1c7273b 5730 * free of errors) or -errno when an internal error occurred. The results of the
e076f338 5731 * check are stored in res.
e97fc193 5732 */
21c2283e
VSO
5733int coroutine_fn bdrv_co_check(BlockDriverState *bs,
5734 BdrvCheckResult *res, BdrvCheckMode fix)
e97fc193 5735{
1581a70d 5736 IO_CODE();
1b3ff9fe 5737 assert_bdrv_graph_readable();
908bcd54
HR
5738 if (bs->drv == NULL) {
5739 return -ENOMEDIUM;
5740 }
2fd61638 5741 if (bs->drv->bdrv_co_check == NULL) {
e97fc193
AL
5742 return -ENOTSUP;
5743 }
5744
e076f338 5745 memset(res, 0, sizeof(*res));
2fd61638
PB
5746 return bs->drv->bdrv_co_check(bs, res, fix);
5747}
5748
756e6736
KW
5749/*
5750 * Return values:
5751 * 0 - success
5752 * -EINVAL - backing format specified, but no file
5753 * -ENOSPC - can't update the backing file because no space is left in the
5754 * image file header
5755 * -ENOTSUP - format driver doesn't support changing the backing file
5756 */
e54ee1b3 5757int bdrv_change_backing_file(BlockDriverState *bs, const char *backing_file,
497a30db 5758 const char *backing_fmt, bool require)
756e6736
KW
5759{
5760 BlockDriver *drv = bs->drv;
469ef350 5761 int ret;
756e6736 5762
f791bf7f
EGE
5763 GLOBAL_STATE_CODE();
5764
d470ad42
HR
5765 if (!drv) {
5766 return -ENOMEDIUM;
5767 }
5768
5f377794
PB
5769 /* Backing file format doesn't make sense without a backing file */
5770 if (backing_fmt && !backing_file) {
5771 return -EINVAL;
5772 }
5773
497a30db
EB
5774 if (require && backing_file && !backing_fmt) {
5775 return -EINVAL;
e54ee1b3
EB
5776 }
5777
756e6736 5778 if (drv->bdrv_change_backing_file != NULL) {
469ef350 5779 ret = drv->bdrv_change_backing_file(bs, backing_file, backing_fmt);
756e6736 5780 } else {
469ef350 5781 ret = -ENOTSUP;
756e6736 5782 }
469ef350
PB
5783
5784 if (ret == 0) {
5785 pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_file ?: "");
5786 pstrcpy(bs->backing_format, sizeof(bs->backing_format), backing_fmt ?: "");
998c2019
HR
5787 pstrcpy(bs->auto_backing_file, sizeof(bs->auto_backing_file),
5788 backing_file ?: "");
469ef350
PB
5789 }
5790 return ret;
756e6736
KW
5791}
5792
6ebdcee2 5793/*
dcf3f9b2
HR
5794 * Finds the first non-filter node above bs in the chain between
5795 * active and bs. The returned node is either an immediate parent of
5796 * bs, or there are only filter nodes between the two.
6ebdcee2
JC
5797 *
5798 * Returns NULL if bs is not found in active's image chain,
5799 * or if active == bs.
4caf0fcd
JC
5800 *
5801 * Returns the bottommost base image if bs == NULL.
6ebdcee2
JC
5802 */
5803BlockDriverState *bdrv_find_overlay(BlockDriverState *active,
5804 BlockDriverState *bs)
5805{
f791bf7f
EGE
5806
5807 GLOBAL_STATE_CODE();
5808
dcf3f9b2
HR
5809 bs = bdrv_skip_filters(bs);
5810 active = bdrv_skip_filters(active);
5811
5812 while (active) {
5813 BlockDriverState *next = bdrv_backing_chain_next(active);
5814 if (bs == next) {
5815 return active;
5816 }
5817 active = next;
6ebdcee2
JC
5818 }
5819
dcf3f9b2 5820 return NULL;
4caf0fcd 5821}
6ebdcee2 5822
4caf0fcd
JC
5823/* Given a BDS, searches for the base layer. */
5824BlockDriverState *bdrv_find_base(BlockDriverState *bs)
5825{
f791bf7f
EGE
5826 GLOBAL_STATE_CODE();
5827
4caf0fcd 5828 return bdrv_find_overlay(bs, NULL);
6ebdcee2
JC
5829}
5830
2cad1ebe 5831/*
7b99a266
HR
5832 * Return true if at least one of the COW (backing) and filter links
5833 * between @bs and @base is frozen. @errp is set if that's the case.
0f0998f6 5834 * @base must be reachable from @bs, or NULL.
2cad1ebe
AG
5835 */
5836bool bdrv_is_backing_chain_frozen(BlockDriverState *bs, BlockDriverState *base,
5837 Error **errp)
5838{
5839 BlockDriverState *i;
7b99a266 5840 BdrvChild *child;
2cad1ebe 5841
f791bf7f
EGE
5842 GLOBAL_STATE_CODE();
5843
7b99a266
HR
5844 for (i = bs; i != base; i = child_bs(child)) {
5845 child = bdrv_filter_or_cow_child(i);
5846
5847 if (child && child->frozen) {
2cad1ebe 5848 error_setg(errp, "Cannot change '%s' link from '%s' to '%s'",
7b99a266 5849 child->name, i->node_name, child->bs->node_name);
2cad1ebe
AG
5850 return true;
5851 }
5852 }
5853
5854 return false;
5855}
5856
5857/*
7b99a266 5858 * Freeze all COW (backing) and filter links between @bs and @base.
2cad1ebe
AG
5859 * If any of the links is already frozen the operation is aborted and
5860 * none of the links are modified.
0f0998f6 5861 * @base must be reachable from @bs, or NULL.
2cad1ebe
AG
5862 * Returns 0 on success. On failure returns < 0 and sets @errp.
5863 */
5864int bdrv_freeze_backing_chain(BlockDriverState *bs, BlockDriverState *base,
5865 Error **errp)
5866{
5867 BlockDriverState *i;
7b99a266 5868 BdrvChild *child;
2cad1ebe 5869
f791bf7f
EGE
5870 GLOBAL_STATE_CODE();
5871
2cad1ebe
AG
5872 if (bdrv_is_backing_chain_frozen(bs, base, errp)) {
5873 return -EPERM;
5874 }
5875
7b99a266
HR
5876 for (i = bs; i != base; i = child_bs(child)) {
5877 child = bdrv_filter_or_cow_child(i);
5878 if (child && child->bs->never_freeze) {
e5182c1c 5879 error_setg(errp, "Cannot freeze '%s' link to '%s'",
7b99a266 5880 child->name, child->bs->node_name);
e5182c1c
HR
5881 return -EPERM;
5882 }
5883 }
5884
7b99a266
HR
5885 for (i = bs; i != base; i = child_bs(child)) {
5886 child = bdrv_filter_or_cow_child(i);
5887 if (child) {
5888 child->frozen = true;
0f0998f6 5889 }
2cad1ebe
AG
5890 }
5891
5892 return 0;
5893}
5894
5895/*
7b99a266
HR
5896 * Unfreeze all COW (backing) and filter links between @bs and @base.
5897 * The caller must ensure that all links are frozen before using this
5898 * function.
0f0998f6 5899 * @base must be reachable from @bs, or NULL.
2cad1ebe
AG
5900 */
5901void bdrv_unfreeze_backing_chain(BlockDriverState *bs, BlockDriverState *base)
5902{
5903 BlockDriverState *i;
7b99a266 5904 BdrvChild *child;
2cad1ebe 5905
f791bf7f
EGE
5906 GLOBAL_STATE_CODE();
5907
7b99a266
HR
5908 for (i = bs; i != base; i = child_bs(child)) {
5909 child = bdrv_filter_or_cow_child(i);
5910 if (child) {
5911 assert(child->frozen);
5912 child->frozen = false;
0f0998f6 5913 }
2cad1ebe
AG
5914 }
5915}
5916
6ebdcee2
JC
5917/*
5918 * Drops images above 'base' up to and including 'top', and sets the image
5919 * above 'top' to have base as its backing file.
5920 *
5921 * Requires that the overlay to 'top' is opened r/w, so that the backing file
5922 * information in 'bs' can be properly updated.
5923 *
5924 * E.g., this will convert the following chain:
5925 * bottom <- base <- intermediate <- top <- active
5926 *
5927 * to
5928 *
5929 * bottom <- base <- active
5930 *
5931 * It is allowed for bottom==base, in which case it converts:
5932 *
5933 * base <- intermediate <- top <- active
5934 *
5935 * to
5936 *
5937 * base <- active
5938 *
54e26900
JC
5939 * If backing_file_str is non-NULL, it will be used when modifying top's
5940 * overlay image metadata.
5941 *
6ebdcee2
JC
5942 * Error conditions:
5943 * if active == top, that is considered an error
5944 *
5945 */
bde70715
KW
5946int bdrv_drop_intermediate(BlockDriverState *top, BlockDriverState *base,
5947 const char *backing_file_str)
6ebdcee2 5948{
6bd858b3
AG
5949 BlockDriverState *explicit_top = top;
5950 bool update_inherits_from;
d669ed6a 5951 BdrvChild *c;
12fa4af6 5952 Error *local_err = NULL;
6ebdcee2 5953 int ret = -EIO;
d669ed6a
VSO
5954 g_autoptr(GSList) updated_children = NULL;
5955 GSList *p;
6ebdcee2 5956
f791bf7f
EGE
5957 GLOBAL_STATE_CODE();
5958
6858eba0 5959 bdrv_ref(top);
631086de 5960 bdrv_drained_begin(base);
b7cfc7d5 5961 bdrv_graph_rdlock_main_loop();
6858eba0 5962
6ebdcee2
JC
5963 if (!top->drv || !base->drv) {
5964 goto exit;
5965 }
5966
5db15a57
KW
5967 /* Make sure that base is in the backing chain of top */
5968 if (!bdrv_chain_contains(top, base)) {
6ebdcee2
JC
5969 goto exit;
5970 }
5971
6bd858b3
AG
5972 /* If 'base' recursively inherits from 'top' then we should set
5973 * base->inherits_from to top->inherits_from after 'top' and all
5974 * other intermediate nodes have been dropped.
5975 * If 'top' is an implicit node (e.g. "commit_top") we should skip
5976 * it because no one inherits from it. We use explicit_top for that. */
dcf3f9b2 5977 explicit_top = bdrv_skip_implicit_filters(explicit_top);
6bd858b3
AG
5978 update_inherits_from = bdrv_inherits_from_recursive(base, explicit_top);
5979
6ebdcee2 5980 /* success - we can delete the intermediate states, and link top->base */
f30c66ba
HR
5981 if (!backing_file_str) {
5982 bdrv_refresh_filename(base);
5983 backing_file_str = base->filename;
5984 }
61f09cea 5985
d669ed6a
VSO
5986 QLIST_FOREACH(c, &top->parents, next_parent) {
5987 updated_children = g_slist_prepend(updated_children, c);
5988 }
5989
3108a15c
VSO
5990 /*
5991 * It seems correct to pass detach_subchain=true here, but it triggers
5992 * one more yet not fixed bug, when due to nested aio_poll loop we switch to
5993 * another drained section, which modify the graph (for example, removing
5994 * the child, which we keep in updated_children list). So, it's a TODO.
5995 *
5996 * Note, bug triggered if pass detach_subchain=true here and run
5997 * test-bdrv-drain. test_drop_intermediate_poll() test-case will crash.
5998 * That's a FIXME.
5999 */
6000 bdrv_replace_node_common(top, base, false, false, &local_err);
d669ed6a
VSO
6001 if (local_err) {
6002 error_report_err(local_err);
6003 goto exit;
6004 }
6005
6006 for (p = updated_children; p; p = p->next) {
6007 c = p->data;
12fa4af6 6008
bd86fb99
HR
6009 if (c->klass->update_filename) {
6010 ret = c->klass->update_filename(c, base, backing_file_str,
6011 &local_err);
61f09cea 6012 if (ret < 0) {
d669ed6a
VSO
6013 /*
6014 * TODO: Actually, we want to rollback all previous iterations
6015 * of this loop, and (which is almost impossible) previous
6016 * bdrv_replace_node()...
6017 *
6018 * Note, that c->klass->update_filename may lead to permission
6019 * update, so it's a bad idea to call it inside permission
6020 * update transaction of bdrv_replace_node.
6021 */
61f09cea
KW
6022 error_report_err(local_err);
6023 goto exit;
6024 }
6025 }
12fa4af6 6026 }
6ebdcee2 6027
6bd858b3
AG
6028 if (update_inherits_from) {
6029 base->inherits_from = explicit_top->inherits_from;
6030 }
6031
6ebdcee2 6032 ret = 0;
6ebdcee2 6033exit:
b7cfc7d5 6034 bdrv_graph_rdunlock_main_loop();
631086de 6035 bdrv_drained_end(base);
6858eba0 6036 bdrv_unref(top);
6ebdcee2
JC
6037 return ret;
6038}
6039
081e4650 6040/**
82618d7b 6041 * Implementation of BlockDriver.bdrv_co_get_allocated_file_size() that
081e4650
HR
6042 * sums the size of all data-bearing children. (This excludes backing
6043 * children.)
6044 */
de335638
EGE
6045static int64_t coroutine_fn GRAPH_RDLOCK
6046bdrv_sum_allocated_file_size(BlockDriverState *bs)
081e4650
HR
6047{
6048 BdrvChild *child;
6049 int64_t child_size, sum = 0;
6050
6051 QLIST_FOREACH(child, &bs->children, next) {
6052 if (child->role & (BDRV_CHILD_DATA | BDRV_CHILD_METADATA |
6053 BDRV_CHILD_FILTERED))
6054 {
82618d7b 6055 child_size = bdrv_co_get_allocated_file_size(child->bs);
081e4650
HR
6056 if (child_size < 0) {
6057 return child_size;
6058 }
6059 sum += child_size;
6060 }
6061 }
6062
6063 return sum;
6064}
6065
61007b31
SH
6066/**
6067 * Length of a allocated file in bytes. Sparse files are counted by actual
6068 * allocated space. Return < 0 if error or unknown.
6069 */
82618d7b 6070int64_t coroutine_fn bdrv_co_get_allocated_file_size(BlockDriverState *bs)
71d0770c 6071{
61007b31 6072 BlockDriver *drv = bs->drv;
384a48fb 6073 IO_CODE();
de335638 6074 assert_bdrv_graph_readable();
384a48fb 6075
61007b31
SH
6076 if (!drv) {
6077 return -ENOMEDIUM;
8f4754ed 6078 }
82618d7b
EGE
6079 if (drv->bdrv_co_get_allocated_file_size) {
6080 return drv->bdrv_co_get_allocated_file_size(bs);
61007b31 6081 }
081e4650
HR
6082
6083 if (drv->bdrv_file_open) {
6084 /*
6085 * Protocol drivers default to -ENOTSUP (most of their data is
6086 * not stored in any of their children (if they even have any),
6087 * so there is no generic way to figure it out).
6088 */
6089 return -ENOTSUP;
6090 } else if (drv->is_filter) {
6091 /* Filter drivers default to the size of their filtered child */
82618d7b 6092 return bdrv_co_get_allocated_file_size(bdrv_filter_bs(bs));
081e4650
HR
6093 } else {
6094 /* Other drivers default to summing their children's sizes */
6095 return bdrv_sum_allocated_file_size(bs);
1c9805a3
SH
6096 }
6097}
e7a8a783 6098
90880ff1
SH
6099/*
6100 * bdrv_measure:
6101 * @drv: Format driver
6102 * @opts: Creation options for new image
6103 * @in_bs: Existing image containing data for new image (may be NULL)
6104 * @errp: Error object
6105 * Returns: A #BlockMeasureInfo (free using qapi_free_BlockMeasureInfo())
6106 * or NULL on error
6107 *
6108 * Calculate file size required to create a new image.
6109 *
6110 * If @in_bs is given then space for allocated clusters and zero clusters
6111 * from that image are included in the calculation. If @opts contains a
6112 * backing file that is shared by @in_bs then backing clusters may be omitted
6113 * from the calculation.
6114 *
6115 * If @in_bs is NULL then the calculation includes no allocated clusters
6116 * unless a preallocation option is given in @opts.
6117 *
6118 * Note that @in_bs may use a different BlockDriver from @drv.
6119 *
6120 * If an error occurs the @errp pointer is set.
6121 */
6122BlockMeasureInfo *bdrv_measure(BlockDriver *drv, QemuOpts *opts,
6123 BlockDriverState *in_bs, Error **errp)
6124{
384a48fb 6125 IO_CODE();
90880ff1
SH
6126 if (!drv->bdrv_measure) {
6127 error_setg(errp, "Block driver '%s' does not support size measurement",
6128 drv->format_name);
6129 return NULL;
6130 }
6131
6132 return drv->bdrv_measure(opts, in_bs, errp);
6133}
6134
61007b31
SH
6135/**
6136 * Return number of sectors on success, -errno on error.
1c9805a3 6137 */
c86422c5 6138int64_t coroutine_fn bdrv_co_nb_sectors(BlockDriverState *bs)
1c9805a3 6139{
61007b31 6140 BlockDriver *drv = bs->drv;
384a48fb 6141 IO_CODE();
8ab8140a 6142 assert_bdrv_graph_readable();
498e386c 6143
61007b31
SH
6144 if (!drv)
6145 return -ENOMEDIUM;
2572b37a 6146
160a29e2 6147 if (bs->bl.has_variable_length) {
c86422c5 6148 int ret = bdrv_co_refresh_total_sectors(bs, bs->total_sectors);
61007b31
SH
6149 if (ret < 0) {
6150 return ret;
1c9805a3
SH
6151 }
6152 }
61007b31 6153 return bs->total_sectors;
1c9805a3 6154}
b338082b 6155
81f730d4
PB
6156/*
6157 * This wrapper is written by hand because this function is in the hot I/O path,
6158 * via blk_get_geometry.
6159 */
6160int64_t coroutine_mixed_fn bdrv_nb_sectors(BlockDriverState *bs)
6161{
6162 BlockDriver *drv = bs->drv;
6163 IO_CODE();
6164
6165 if (!drv)
6166 return -ENOMEDIUM;
6167
6168 if (bs->bl.has_variable_length) {
6169 int ret = bdrv_refresh_total_sectors(bs, bs->total_sectors);
6170 if (ret < 0) {
6171 return ret;
6172 }
6173 }
6174
6175 return bs->total_sectors;
6176}
6177
61007b31
SH
6178/**
6179 * Return length in bytes on success, -errno on error.
6180 * The length is always a multiple of BDRV_SECTOR_SIZE.
8d3b1a2d 6181 */
c86422c5 6182int64_t coroutine_fn bdrv_co_getlength(BlockDriverState *bs)
8d3b1a2d 6183{
c86422c5 6184 int64_t ret;
384a48fb 6185 IO_CODE();
8ab8140a 6186 assert_bdrv_graph_readable();
8d3b1a2d 6187
c86422c5 6188 ret = bdrv_co_nb_sectors(bs);
122860ba
EB
6189 if (ret < 0) {
6190 return ret;
6191 }
6192 if (ret > INT64_MAX / BDRV_SECTOR_SIZE) {
6193 return -EFBIG;
6194 }
6195 return ret * BDRV_SECTOR_SIZE;
fc01f7e7
FB
6196}
6197
54115412 6198bool bdrv_is_sg(BlockDriverState *bs)
f08145fe 6199{
384a48fb 6200 IO_CODE();
61007b31 6201 return bs->sg;
f08145fe
KW
6202}
6203
ae23f786
HR
6204/**
6205 * Return whether the given node supports compressed writes.
6206 */
6207bool bdrv_supports_compressed_writes(BlockDriverState *bs)
6208{
6209 BlockDriverState *filtered;
384a48fb 6210 IO_CODE();
ae23f786
HR
6211
6212 if (!bs->drv || !block_driver_can_compress(bs->drv)) {
6213 return false;
6214 }
6215
6216 filtered = bdrv_filter_bs(bs);
6217 if (filtered) {
6218 /*
6219 * Filters can only forward compressed writes, so we have to
6220 * check the child.
6221 */
6222 return bdrv_supports_compressed_writes(filtered);
6223 }
6224
6225 return true;
6226}
6227
61007b31 6228const char *bdrv_get_format_name(BlockDriverState *bs)
40b4f539 6229{
384a48fb 6230 IO_CODE();
61007b31 6231 return bs->drv ? bs->drv->format_name : NULL;
40b4f539
KW
6232}
6233
61007b31 6234static int qsort_strcmp(const void *a, const void *b)
40b4f539 6235{
ceff5bd7 6236 return strcmp(*(char *const *)a, *(char *const *)b);
40b4f539
KW
6237}
6238
61007b31 6239void bdrv_iterate_format(void (*it)(void *opaque, const char *name),
9ac404c5 6240 void *opaque, bool read_only)
40b4f539 6241{
61007b31
SH
6242 BlockDriver *drv;
6243 int count = 0;
6244 int i;
6245 const char **formats = NULL;
40b4f539 6246
f791bf7f
EGE
6247 GLOBAL_STATE_CODE();
6248
61007b31
SH
6249 QLIST_FOREACH(drv, &bdrv_drivers, list) {
6250 if (drv->format_name) {
6251 bool found = false;
9ac404c5
AS
6252
6253 if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv, read_only)) {
6254 continue;
6255 }
6256
fb2575f9 6257 i = count;
61007b31
SH
6258 while (formats && i && !found) {
6259 found = !strcmp(formats[--i], drv->format_name);
6260 }
e2a305fb 6261
61007b31
SH
6262 if (!found) {
6263 formats = g_renew(const char *, formats, count + 1);
6264 formats[count++] = drv->format_name;
6265 }
6c5a42ac 6266 }
61007b31 6267 }
6c5a42ac 6268
eb0df69f
HR
6269 for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); i++) {
6270 const char *format_name = block_driver_modules[i].format_name;
6271
6272 if (format_name) {
6273 bool found = false;
6274 int j = count;
6275
9ac404c5
AS
6276 if (use_bdrv_whitelist &&
6277 !bdrv_format_is_whitelisted(format_name, read_only)) {
6278 continue;
6279 }
6280
eb0df69f
HR
6281 while (formats && j && !found) {
6282 found = !strcmp(formats[--j], format_name);
6283 }
6284
6285 if (!found) {
6286 formats = g_renew(const char *, formats, count + 1);
6287 formats[count++] = format_name;
6288 }
6289 }
6290 }
6291
61007b31 6292 qsort(formats, count, sizeof(formats[0]), qsort_strcmp);
40b4f539 6293
61007b31
SH
6294 for (i = 0; i < count; i++) {
6295 it(opaque, formats[i]);
6296 }
40b4f539 6297
61007b31
SH
6298 g_free(formats);
6299}
40b4f539 6300
61007b31
SH
6301/* This function is to find a node in the bs graph */
6302BlockDriverState *bdrv_find_node(const char *node_name)
6303{
6304 BlockDriverState *bs;
391827eb 6305
61007b31 6306 assert(node_name);
f791bf7f 6307 GLOBAL_STATE_CODE();
40b4f539 6308
61007b31
SH
6309 QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
6310 if (!strcmp(node_name, bs->node_name)) {
6311 return bs;
40b4f539
KW
6312 }
6313 }
61007b31 6314 return NULL;
40b4f539
KW
6315}
6316
61007b31 6317/* Put this QMP function here so it can access the static graph_bdrv_states. */
facda544
PK
6318BlockDeviceInfoList *bdrv_named_nodes_list(bool flat,
6319 Error **errp)
40b4f539 6320{
9812e712 6321 BlockDeviceInfoList *list;
61007b31 6322 BlockDriverState *bs;
40b4f539 6323
f791bf7f 6324 GLOBAL_STATE_CODE();
b7cfc7d5 6325 GRAPH_RDLOCK_GUARD_MAINLOOP();
f791bf7f 6326
61007b31
SH
6327 list = NULL;
6328 QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
facda544 6329 BlockDeviceInfo *info = bdrv_block_device_info(NULL, bs, flat, errp);
61007b31
SH
6330 if (!info) {
6331 qapi_free_BlockDeviceInfoList(list);
6332 return NULL;
301db7c2 6333 }
9812e712 6334 QAPI_LIST_PREPEND(list, info);
301db7c2
RH
6335 }
6336
61007b31
SH
6337 return list;
6338}
40b4f539 6339
5d3b4e99
VSO
6340typedef struct XDbgBlockGraphConstructor {
6341 XDbgBlockGraph *graph;
6342 GHashTable *graph_nodes;
6343} XDbgBlockGraphConstructor;
6344
6345static XDbgBlockGraphConstructor *xdbg_graph_new(void)
6346{
6347 XDbgBlockGraphConstructor *gr = g_new(XDbgBlockGraphConstructor, 1);
6348
6349 gr->graph = g_new0(XDbgBlockGraph, 1);
6350 gr->graph_nodes = g_hash_table_new(NULL, NULL);
6351
6352 return gr;
6353}
6354
6355static XDbgBlockGraph *xdbg_graph_finalize(XDbgBlockGraphConstructor *gr)
6356{
6357 XDbgBlockGraph *graph = gr->graph;
6358
6359 g_hash_table_destroy(gr->graph_nodes);
6360 g_free(gr);
6361
6362 return graph;
6363}
6364
6365static uintptr_t xdbg_graph_node_num(XDbgBlockGraphConstructor *gr, void *node)
6366{
6367 uintptr_t ret = (uintptr_t)g_hash_table_lookup(gr->graph_nodes, node);
6368
6369 if (ret != 0) {
6370 return ret;
6371 }
6372
6373 /*
6374 * Start counting from 1, not 0, because 0 interferes with not-found (NULL)
6375 * answer of g_hash_table_lookup.
6376 */
6377 ret = g_hash_table_size(gr->graph_nodes) + 1;
6378 g_hash_table_insert(gr->graph_nodes, node, (void *)ret);
6379
6380 return ret;
6381}
6382
6383static void xdbg_graph_add_node(XDbgBlockGraphConstructor *gr, void *node,
6384 XDbgBlockGraphNodeType type, const char *name)
6385{
6386 XDbgBlockGraphNode *n;
6387
6388 n = g_new0(XDbgBlockGraphNode, 1);
6389
6390 n->id = xdbg_graph_node_num(gr, node);
6391 n->type = type;
6392 n->name = g_strdup(name);
6393
9812e712 6394 QAPI_LIST_PREPEND(gr->graph->nodes, n);
5d3b4e99
VSO
6395}
6396
6397static void xdbg_graph_add_edge(XDbgBlockGraphConstructor *gr, void *parent,
6398 const BdrvChild *child)
6399{
cdb1cec8 6400 BlockPermission qapi_perm;
5d3b4e99 6401 XDbgBlockGraphEdge *edge;
862fded9 6402 GLOBAL_STATE_CODE();
5d3b4e99 6403
5d3b4e99
VSO
6404 edge = g_new0(XDbgBlockGraphEdge, 1);
6405
6406 edge->parent = xdbg_graph_node_num(gr, parent);
6407 edge->child = xdbg_graph_node_num(gr, child->bs);
6408 edge->name = g_strdup(child->name);
6409
cdb1cec8
HR
6410 for (qapi_perm = 0; qapi_perm < BLOCK_PERMISSION__MAX; qapi_perm++) {
6411 uint64_t flag = bdrv_qapi_perm_to_blk_perm(qapi_perm);
6412
6413 if (flag & child->perm) {
9812e712 6414 QAPI_LIST_PREPEND(edge->perm, qapi_perm);
5d3b4e99 6415 }
cdb1cec8 6416 if (flag & child->shared_perm) {
9812e712 6417 QAPI_LIST_PREPEND(edge->shared_perm, qapi_perm);
5d3b4e99
VSO
6418 }
6419 }
6420
9812e712 6421 QAPI_LIST_PREPEND(gr->graph->edges, edge);
5d3b4e99
VSO
6422}
6423
6424
6425XDbgBlockGraph *bdrv_get_xdbg_block_graph(Error **errp)
6426{
6427 BlockBackend *blk;
6428 BlockJob *job;
6429 BlockDriverState *bs;
6430 BdrvChild *child;
6431 XDbgBlockGraphConstructor *gr = xdbg_graph_new();
6432
f791bf7f
EGE
6433 GLOBAL_STATE_CODE();
6434
5d3b4e99
VSO
6435 for (blk = blk_all_next(NULL); blk; blk = blk_all_next(blk)) {
6436 char *allocated_name = NULL;
6437 const char *name = blk_name(blk);
6438
6439 if (!*name) {
6440 name = allocated_name = blk_get_attached_dev_id(blk);
6441 }
6442 xdbg_graph_add_node(gr, blk, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_BACKEND,
6443 name);
6444 g_free(allocated_name);
6445 if (blk_root(blk)) {
6446 xdbg_graph_add_edge(gr, blk, blk_root(blk));
6447 }
6448 }
6449
880eeec6
EGE
6450 WITH_JOB_LOCK_GUARD() {
6451 for (job = block_job_next_locked(NULL); job;
6452 job = block_job_next_locked(job)) {
6453 GSList *el;
5d3b4e99 6454
880eeec6
EGE
6455 xdbg_graph_add_node(gr, job, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_JOB,
6456 job->job.id);
6457 for (el = job->nodes; el; el = el->next) {
6458 xdbg_graph_add_edge(gr, job, (BdrvChild *)el->data);
6459 }
5d3b4e99
VSO
6460 }
6461 }
6462
6463 QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
6464 xdbg_graph_add_node(gr, bs, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_DRIVER,
6465 bs->node_name);
6466 QLIST_FOREACH(child, &bs->children, next) {
6467 xdbg_graph_add_edge(gr, bs, child);
6468 }
6469 }
6470
6471 return xdbg_graph_finalize(gr);
6472}
6473
61007b31
SH
6474BlockDriverState *bdrv_lookup_bs(const char *device,
6475 const char *node_name,
6476 Error **errp)
6477{
6478 BlockBackend *blk;
6479 BlockDriverState *bs;
40b4f539 6480
f791bf7f
EGE
6481 GLOBAL_STATE_CODE();
6482
61007b31
SH
6483 if (device) {
6484 blk = blk_by_name(device);
40b4f539 6485
61007b31 6486 if (blk) {
9f4ed6fb
AG
6487 bs = blk_bs(blk);
6488 if (!bs) {
5433c24f 6489 error_setg(errp, "Device '%s' has no medium", device);
5433c24f
HR
6490 }
6491
9f4ed6fb 6492 return bs;
61007b31
SH
6493 }
6494 }
40b4f539 6495
61007b31
SH
6496 if (node_name) {
6497 bs = bdrv_find_node(node_name);
6d519a5f 6498
61007b31
SH
6499 if (bs) {
6500 return bs;
6501 }
40b4f539
KW
6502 }
6503
785ec4b1 6504 error_setg(errp, "Cannot find device=\'%s\' nor node-name=\'%s\'",
61007b31
SH
6505 device ? device : "",
6506 node_name ? node_name : "");
6507 return NULL;
40b4f539
KW
6508}
6509
61007b31
SH
6510/* If 'base' is in the same chain as 'top', return true. Otherwise,
6511 * return false. If either argument is NULL, return false. */
6512bool bdrv_chain_contains(BlockDriverState *top, BlockDriverState *base)
83f64091 6513{
f791bf7f
EGE
6514
6515 GLOBAL_STATE_CODE();
372b69f5 6516 GRAPH_RDLOCK_GUARD_MAINLOOP();
f791bf7f 6517
61007b31 6518 while (top && top != base) {
dcf3f9b2 6519 top = bdrv_filter_or_cow_bs(top);
02c50efe 6520 }
61007b31
SH
6521
6522 return top != NULL;
02c50efe
FZ
6523}
6524
61007b31 6525BlockDriverState *bdrv_next_node(BlockDriverState *bs)
02c50efe 6526{
f791bf7f 6527 GLOBAL_STATE_CODE();
61007b31
SH
6528 if (!bs) {
6529 return QTAILQ_FIRST(&graph_bdrv_states);
02c50efe 6530 }
61007b31 6531 return QTAILQ_NEXT(bs, node_list);
83f64091
FB
6532}
6533
0f12264e
KW
6534BlockDriverState *bdrv_next_all_states(BlockDriverState *bs)
6535{
f791bf7f 6536 GLOBAL_STATE_CODE();
0f12264e
KW
6537 if (!bs) {
6538 return QTAILQ_FIRST(&all_bdrv_states);
6539 }
6540 return QTAILQ_NEXT(bs, bs_list);
6541}
6542
61007b31 6543const char *bdrv_get_node_name(const BlockDriverState *bs)
83f64091 6544{
384a48fb 6545 IO_CODE();
61007b31 6546 return bs->node_name;
beac80cd
FB
6547}
6548
1f0c461b 6549const char *bdrv_get_parent_name(const BlockDriverState *bs)
4c265bf9
KW
6550{
6551 BdrvChild *c;
6552 const char *name;
967d7905 6553 IO_CODE();
4c265bf9
KW
6554
6555 /* If multiple parents have a name, just pick the first one. */
6556 QLIST_FOREACH(c, &bs->parents, next_parent) {
bd86fb99
HR
6557 if (c->klass->get_name) {
6558 name = c->klass->get_name(c);
4c265bf9
KW
6559 if (name && *name) {
6560 return name;
6561 }
6562 }
6563 }
6564
6565 return NULL;
6566}
6567
61007b31
SH
6568/* TODO check what callers really want: bs->node_name or blk_name() */
6569const char *bdrv_get_device_name(const BlockDriverState *bs)
beac80cd 6570{
384a48fb 6571 IO_CODE();
4c265bf9 6572 return bdrv_get_parent_name(bs) ?: "";
f141eafe 6573}
83f64091 6574
61007b31
SH
6575/* This can be used to identify nodes that might not have a device
6576 * name associated. Since node and device names live in the same
6577 * namespace, the result is unambiguous. The exception is if both are
6578 * absent, then this returns an empty (non-null) string. */
6579const char *bdrv_get_device_or_node_name(const BlockDriverState *bs)
f141eafe 6580{
384a48fb 6581 IO_CODE();
4c265bf9 6582 return bdrv_get_parent_name(bs) ?: bs->node_name;
beac80cd 6583}
beac80cd 6584
61007b31 6585int bdrv_get_flags(BlockDriverState *bs)
0b5a2445 6586{
15aee7ac 6587 IO_CODE();
61007b31 6588 return bs->open_flags;
0b5a2445
PB
6589}
6590
61007b31 6591int bdrv_has_zero_init_1(BlockDriverState *bs)
68485420 6592{
f791bf7f 6593 GLOBAL_STATE_CODE();
61007b31 6594 return 1;
0b5a2445
PB
6595}
6596
06717986 6597int coroutine_mixed_fn bdrv_has_zero_init(BlockDriverState *bs)
0b5a2445 6598{
93393e69 6599 BlockDriverState *filtered;
f791bf7f 6600 GLOBAL_STATE_CODE();
93393e69 6601
d470ad42
HR
6602 if (!bs->drv) {
6603 return 0;
6604 }
0b5a2445 6605
61007b31
SH
6606 /* If BS is a copy on write image, it is initialized to
6607 the contents of the base image, which may not be zeroes. */
34778172 6608 if (bdrv_cow_child(bs)) {
61007b31
SH
6609 return 0;
6610 }
6611 if (bs->drv->bdrv_has_zero_init) {
6612 return bs->drv->bdrv_has_zero_init(bs);
0b5a2445 6613 }
93393e69
HR
6614
6615 filtered = bdrv_filter_bs(bs);
6616 if (filtered) {
6617 return bdrv_has_zero_init(filtered);
5a612c00 6618 }
61007b31
SH
6619
6620 /* safe default */
6621 return 0;
68485420
KW
6622}
6623
61007b31 6624bool bdrv_can_write_zeroes_with_unmap(BlockDriverState *bs)
68485420 6625{
384a48fb 6626 IO_CODE();
2f0342ef 6627 if (!(bs->open_flags & BDRV_O_UNMAP)) {
61007b31
SH
6628 return false;
6629 }
68485420 6630
e24d813b 6631 return bs->supported_zero_flags & BDRV_REQ_MAY_UNMAP;
68485420
KW
6632}
6633
61007b31
SH
6634void bdrv_get_backing_filename(BlockDriverState *bs,
6635 char *filename, int filename_size)
016f5cf6 6636{
384a48fb 6637 IO_CODE();
61007b31
SH
6638 pstrcpy(filename, filename_size, bs->backing_file);
6639}
d318aea9 6640
3d47eb0a 6641int coroutine_fn bdrv_co_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
61007b31 6642{
8b117001 6643 int ret;
61007b31 6644 BlockDriver *drv = bs->drv;
384a48fb 6645 IO_CODE();
a00e70c0
EGE
6646 assert_bdrv_graph_readable();
6647
5a612c00
MP
6648 /* if bs->drv == NULL, bs is closed, so there's nothing to do here */
6649 if (!drv) {
61007b31 6650 return -ENOMEDIUM;
5a612c00 6651 }
3d47eb0a 6652 if (!drv->bdrv_co_get_info) {
93393e69
HR
6653 BlockDriverState *filtered = bdrv_filter_bs(bs);
6654 if (filtered) {
3d47eb0a 6655 return bdrv_co_get_info(filtered, bdi);
5a612c00 6656 }
61007b31 6657 return -ENOTSUP;
5a612c00 6658 }
61007b31 6659 memset(bdi, 0, sizeof(*bdi));
3d47eb0a 6660 ret = drv->bdrv_co_get_info(bs, bdi);
c54483b6
AD
6661 if (bdi->subcluster_size == 0) {
6662 /*
6663 * If the driver left this unset, subclusters are not supported.
6664 * Then it is safe to treat each cluster as having only one subcluster.
6665 */
6666 bdi->subcluster_size = bdi->cluster_size;
6667 }
8b117001
VSO
6668 if (ret < 0) {
6669 return ret;
6670 }
6671
6672 if (bdi->cluster_size > BDRV_MAX_ALIGNMENT) {
6673 return -EINVAL;
6674 }
6675
6676 return 0;
61007b31 6677}
016f5cf6 6678
1bf6e9ca
AS
6679ImageInfoSpecific *bdrv_get_specific_info(BlockDriverState *bs,
6680 Error **errp)
61007b31
SH
6681{
6682 BlockDriver *drv = bs->drv;
384a48fb 6683 IO_CODE();
61007b31 6684 if (drv && drv->bdrv_get_specific_info) {
1bf6e9ca 6685 return drv->bdrv_get_specific_info(bs, errp);
61007b31
SH
6686 }
6687 return NULL;
016f5cf6
AG
6688}
6689
d9245599
AN
6690BlockStatsSpecific *bdrv_get_specific_stats(BlockDriverState *bs)
6691{
6692 BlockDriver *drv = bs->drv;
384a48fb 6693 IO_CODE();
d9245599
AN
6694 if (!drv || !drv->bdrv_get_specific_stats) {
6695 return NULL;
6696 }
6697 return drv->bdrv_get_specific_stats(bs);
6698}
6699
c834dc05 6700void coroutine_fn bdrv_co_debug_event(BlockDriverState *bs, BlkdebugEvent event)
4265d620 6701{
384a48fb 6702 IO_CODE();
cb2bfaa4
EGE
6703 assert_bdrv_graph_readable();
6704
c834dc05 6705 if (!bs || !bs->drv || !bs->drv->bdrv_co_debug_event) {
61007b31
SH
6706 return;
6707 }
4265d620 6708
c834dc05 6709 bs->drv->bdrv_co_debug_event(bs, event);
4265d620
PB
6710}
6711
c0fc5123
KW
6712static BlockDriverState * GRAPH_RDLOCK
6713bdrv_find_debug_node(BlockDriverState *bs)
4265d620 6714{
bdb73476 6715 GLOBAL_STATE_CODE();
61007b31 6716 while (bs && bs->drv && !bs->drv->bdrv_debug_breakpoint) {
f706a92f 6717 bs = bdrv_primary_bs(bs);
61007b31 6718 }
4265d620 6719
61007b31 6720 if (bs && bs->drv && bs->drv->bdrv_debug_breakpoint) {
d10529a2
VSO
6721 assert(bs->drv->bdrv_debug_remove_breakpoint);
6722 return bs;
6723 }
6724
6725 return NULL;
6726}
6727
6728int bdrv_debug_breakpoint(BlockDriverState *bs, const char *event,
6729 const char *tag)
6730{
f791bf7f 6731 GLOBAL_STATE_CODE();
c0fc5123
KW
6732 GRAPH_RDLOCK_GUARD_MAINLOOP();
6733
d10529a2
VSO
6734 bs = bdrv_find_debug_node(bs);
6735 if (bs) {
61007b31
SH
6736 return bs->drv->bdrv_debug_breakpoint(bs, event, tag);
6737 }
4265d620 6738
61007b31 6739 return -ENOTSUP;
4265d620
PB
6740}
6741
61007b31 6742int bdrv_debug_remove_breakpoint(BlockDriverState *bs, const char *tag)
ea2384d3 6743{
f791bf7f 6744 GLOBAL_STATE_CODE();
c0fc5123
KW
6745 GRAPH_RDLOCK_GUARD_MAINLOOP();
6746
d10529a2
VSO
6747 bs = bdrv_find_debug_node(bs);
6748 if (bs) {
61007b31
SH
6749 return bs->drv->bdrv_debug_remove_breakpoint(bs, tag);
6750 }
6751
6752 return -ENOTSUP;
eb852011
MA
6753}
6754
61007b31 6755int bdrv_debug_resume(BlockDriverState *bs, const char *tag)
ce1a14dc 6756{
f791bf7f 6757 GLOBAL_STATE_CODE();
c0fc5123
KW
6758 GRAPH_RDLOCK_GUARD_MAINLOOP();
6759
61007b31 6760 while (bs && (!bs->drv || !bs->drv->bdrv_debug_resume)) {
f706a92f 6761 bs = bdrv_primary_bs(bs);
61007b31 6762 }
ce1a14dc 6763
61007b31
SH
6764 if (bs && bs->drv && bs->drv->bdrv_debug_resume) {
6765 return bs->drv->bdrv_debug_resume(bs, tag);
6766 }
ce1a14dc 6767
61007b31 6768 return -ENOTSUP;
f197fe2b
FZ
6769}
6770
61007b31 6771bool bdrv_debug_is_suspended(BlockDriverState *bs, const char *tag)
ce1a14dc 6772{
f791bf7f 6773 GLOBAL_STATE_CODE();
c0fc5123
KW
6774 GRAPH_RDLOCK_GUARD_MAINLOOP();
6775
61007b31 6776 while (bs && bs->drv && !bs->drv->bdrv_debug_is_suspended) {
f706a92f 6777 bs = bdrv_primary_bs(bs);
f197fe2b 6778 }
19cb3738 6779
61007b31
SH
6780 if (bs && bs->drv && bs->drv->bdrv_debug_is_suspended) {
6781 return bs->drv->bdrv_debug_is_suspended(bs, tag);
6782 }
f9f05dc5 6783
61007b31
SH
6784 return false;
6785}
f9f05dc5 6786
61007b31
SH
6787/* backing_file can either be relative, or absolute, or a protocol. If it is
6788 * relative, it must be relative to the chain. So, passing in bs->filename
6789 * from a BDS as backing_file should not be done, as that may be relative to
6790 * the CWD rather than the chain. */
6791BlockDriverState *bdrv_find_backing_image(BlockDriverState *bs,
6792 const char *backing_file)
f9f05dc5 6793{
61007b31
SH
6794 char *filename_full = NULL;
6795 char *backing_file_full = NULL;
6796 char *filename_tmp = NULL;
6797 int is_protocol = 0;
0b877d09 6798 bool filenames_refreshed = false;
61007b31
SH
6799 BlockDriverState *curr_bs = NULL;
6800 BlockDriverState *retval = NULL;
dcf3f9b2 6801 BlockDriverState *bs_below;
f9f05dc5 6802
f791bf7f 6803 GLOBAL_STATE_CODE();
b7cfc7d5 6804 GRAPH_RDLOCK_GUARD_MAINLOOP();
f791bf7f 6805
61007b31
SH
6806 if (!bs || !bs->drv || !backing_file) {
6807 return NULL;
f9f05dc5
KW
6808 }
6809
61007b31
SH
6810 filename_full = g_malloc(PATH_MAX);
6811 backing_file_full = g_malloc(PATH_MAX);
f9f05dc5 6812
61007b31 6813 is_protocol = path_has_protocol(backing_file);
f9f05dc5 6814
dcf3f9b2
HR
6815 /*
6816 * Being largely a legacy function, skip any filters here
6817 * (because filters do not have normal filenames, so they cannot
6818 * match anyway; and allowing json:{} filenames is a bit out of
6819 * scope).
6820 */
6821 for (curr_bs = bdrv_skip_filters(bs);
6822 bdrv_cow_child(curr_bs) != NULL;
6823 curr_bs = bs_below)
6824 {
6825 bs_below = bdrv_backing_chain_next(curr_bs);
f9f05dc5 6826
0b877d09
HR
6827 if (bdrv_backing_overridden(curr_bs)) {
6828 /*
6829 * If the backing file was overridden, we can only compare
6830 * directly against the backing node's filename.
6831 */
6832
6833 if (!filenames_refreshed) {
6834 /*
6835 * This will automatically refresh all of the
6836 * filenames in the rest of the backing chain, so we
6837 * only need to do this once.
6838 */
6839 bdrv_refresh_filename(bs_below);
6840 filenames_refreshed = true;
6841 }
6842
6843 if (strcmp(backing_file, bs_below->filename) == 0) {
6844 retval = bs_below;
6845 break;
6846 }
6847 } else if (is_protocol || path_has_protocol(curr_bs->backing_file)) {
6848 /*
6849 * If either of the filename paths is actually a protocol, then
6850 * compare unmodified paths; otherwise make paths relative.
6851 */
6b6833c1
HR
6852 char *backing_file_full_ret;
6853
61007b31 6854 if (strcmp(backing_file, curr_bs->backing_file) == 0) {
dcf3f9b2 6855 retval = bs_below;
61007b31
SH
6856 break;
6857 }
418661e0 6858 /* Also check against the full backing filename for the image */
6b6833c1
HR
6859 backing_file_full_ret = bdrv_get_full_backing_filename(curr_bs,
6860 NULL);
6861 if (backing_file_full_ret) {
6862 bool equal = strcmp(backing_file, backing_file_full_ret) == 0;
6863 g_free(backing_file_full_ret);
6864 if (equal) {
dcf3f9b2 6865 retval = bs_below;
418661e0
JC
6866 break;
6867 }
418661e0 6868 }
61007b31
SH
6869 } else {
6870 /* If not an absolute filename path, make it relative to the current
6871 * image's filename path */
2d9158ce
HR
6872 filename_tmp = bdrv_make_absolute_filename(curr_bs, backing_file,
6873 NULL);
6874 /* We are going to compare canonicalized absolute pathnames */
6875 if (!filename_tmp || !realpath(filename_tmp, filename_full)) {
6876 g_free(filename_tmp);
61007b31
SH
6877 continue;
6878 }
2d9158ce 6879 g_free(filename_tmp);
07f07615 6880
61007b31
SH
6881 /* We need to make sure the backing filename we are comparing against
6882 * is relative to the current image filename (or absolute) */
2d9158ce
HR
6883 filename_tmp = bdrv_get_full_backing_filename(curr_bs, NULL);
6884 if (!filename_tmp || !realpath(filename_tmp, backing_file_full)) {
6885 g_free(filename_tmp);
61007b31
SH
6886 continue;
6887 }
2d9158ce 6888 g_free(filename_tmp);
eb489bb1 6889
61007b31 6890 if (strcmp(backing_file_full, filename_full) == 0) {
dcf3f9b2 6891 retval = bs_below;
61007b31
SH
6892 break;
6893 }
6894 }
eb489bb1
KW
6895 }
6896
61007b31
SH
6897 g_free(filename_full);
6898 g_free(backing_file_full);
61007b31
SH
6899 return retval;
6900}
6901
61007b31
SH
6902void bdrv_init(void)
6903{
e5f05f8c
KW
6904#ifdef CONFIG_BDRV_WHITELIST_TOOLS
6905 use_bdrv_whitelist = 1;
6906#endif
61007b31
SH
6907 module_call_init(MODULE_INIT_BLOCK);
6908}
29cdb251 6909
61007b31
SH
6910void bdrv_init_with_whitelist(void)
6911{
6912 use_bdrv_whitelist = 1;
6913 bdrv_init();
07f07615
PB
6914}
6915
a94750d9 6916int bdrv_activate(BlockDriverState *bs, Error **errp)
0f15423c 6917{
4417ab7a 6918 BdrvChild *child, *parent;
5a8a30db
KW
6919 Error *local_err = NULL;
6920 int ret;
9c98f145 6921 BdrvDirtyBitmap *bm;
5a8a30db 6922
f791bf7f 6923 GLOBAL_STATE_CODE();
3804e3cf 6924 GRAPH_RDLOCK_GUARD_MAINLOOP();
f791bf7f 6925
3456a8d1 6926 if (!bs->drv) {
5416645f 6927 return -ENOMEDIUM;
3456a8d1
KW
6928 }
6929
16e977d5 6930 QLIST_FOREACH(child, &bs->children, next) {
11d0c9b3 6931 bdrv_activate(child->bs, &local_err);
0d1c5c91 6932 if (local_err) {
0d1c5c91 6933 error_propagate(errp, local_err);
5416645f 6934 return -EINVAL;
0d1c5c91 6935 }
5a8a30db 6936 }
0d1c5c91 6937
dafe0960
KW
6938 /*
6939 * Update permissions, they may differ for inactive nodes.
6940 *
6941 * Note that the required permissions of inactive images are always a
6942 * subset of the permissions required after activating the image. This
6943 * allows us to just get the permissions upfront without restricting
11d0c9b3 6944 * bdrv_co_invalidate_cache().
dafe0960
KW
6945 *
6946 * It also means that in error cases, we don't have to try and revert to
6947 * the old permissions (which is an operation that could fail, too). We can
6948 * just keep the extended permissions for the next time that an activation
6949 * of the image is tried.
6950 */
7bb4941a
KW
6951 if (bs->open_flags & BDRV_O_INACTIVE) {
6952 bs->open_flags &= ~BDRV_O_INACTIVE;
f1316edb 6953 ret = bdrv_refresh_perms(bs, NULL, errp);
7bb4941a 6954 if (ret < 0) {
0d1c5c91 6955 bs->open_flags |= BDRV_O_INACTIVE;
5416645f 6956 return ret;
0d1c5c91 6957 }
3456a8d1 6958
11d0c9b3
EGE
6959 ret = bdrv_invalidate_cache(bs, errp);
6960 if (ret < 0) {
6961 bs->open_flags |= BDRV_O_INACTIVE;
6962 return ret;
7bb4941a 6963 }
9c98f145 6964
7bb4941a
KW
6965 FOR_EACH_DIRTY_BITMAP(bs, bm) {
6966 bdrv_dirty_bitmap_skip_store(bm, false);
6967 }
6968
c057960c 6969 ret = bdrv_refresh_total_sectors(bs, bs->total_sectors);
7bb4941a
KW
6970 if (ret < 0) {
6971 bs->open_flags |= BDRV_O_INACTIVE;
6972 error_setg_errno(errp, -ret, "Could not refresh total sector count");
5416645f 6973 return ret;
7bb4941a 6974 }
5a8a30db 6975 }
4417ab7a
KW
6976
6977 QLIST_FOREACH(parent, &bs->parents, next_parent) {
bd86fb99
HR
6978 if (parent->klass->activate) {
6979 parent->klass->activate(parent, &local_err);
4417ab7a 6980 if (local_err) {
78fc3b3a 6981 bs->open_flags |= BDRV_O_INACTIVE;
4417ab7a 6982 error_propagate(errp, local_err);
5416645f 6983 return -EINVAL;
4417ab7a
KW
6984 }
6985 }
6986 }
5416645f
VSO
6987
6988 return 0;
0f15423c
AL
6989}
6990
11d0c9b3
EGE
6991int coroutine_fn bdrv_co_invalidate_cache(BlockDriverState *bs, Error **errp)
6992{
6993 Error *local_err = NULL;
1581a70d 6994 IO_CODE();
11d0c9b3
EGE
6995
6996 assert(!(bs->open_flags & BDRV_O_INACTIVE));
1b3ff9fe 6997 assert_bdrv_graph_readable();
11d0c9b3
EGE
6998
6999 if (bs->drv->bdrv_co_invalidate_cache) {
7000 bs->drv->bdrv_co_invalidate_cache(bs, &local_err);
7001 if (local_err) {
7002 error_propagate(errp, local_err);
7003 return -EINVAL;
7004 }
7005 }
7006
7007 return 0;
7008}
7009
3b717194 7010void bdrv_activate_all(Error **errp)
0f15423c 7011{
7c8eece4 7012 BlockDriverState *bs;
88be7b4b 7013 BdrvNextIterator it;
0f15423c 7014
f791bf7f 7015 GLOBAL_STATE_CODE();
2b3912f1 7016 GRAPH_RDLOCK_GUARD_MAINLOOP();
f791bf7f 7017
88be7b4b 7018 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
ed78cda3 7019 AioContext *aio_context = bdrv_get_aio_context(bs);
5416645f 7020 int ret;
ed78cda3
SH
7021
7022 aio_context_acquire(aio_context);
a94750d9 7023 ret = bdrv_activate(bs, errp);
ed78cda3 7024 aio_context_release(aio_context);
5416645f 7025 if (ret < 0) {
5e003f17 7026 bdrv_next_cleanup(&it);
5a8a30db
KW
7027 return;
7028 }
0f15423c
AL
7029 }
7030}
7031
0e6bad1f
KW
7032static bool GRAPH_RDLOCK
7033bdrv_has_bds_parent(BlockDriverState *bs, bool only_active)
9e37271f
KW
7034{
7035 BdrvChild *parent;
bdb73476 7036 GLOBAL_STATE_CODE();
9e37271f
KW
7037
7038 QLIST_FOREACH(parent, &bs->parents, next_parent) {
bd86fb99 7039 if (parent->klass->parent_is_bds) {
9e37271f
KW
7040 BlockDriverState *parent_bs = parent->opaque;
7041 if (!only_active || !(parent_bs->open_flags & BDRV_O_INACTIVE)) {
7042 return true;
7043 }
7044 }
7045 }
7046
7047 return false;
7048}
7049
0e6bad1f 7050static int GRAPH_RDLOCK bdrv_inactivate_recurse(BlockDriverState *bs)
76b1c7fe 7051{
cfa1a572 7052 BdrvChild *child, *parent;
76b1c7fe 7053 int ret;
a13de40a 7054 uint64_t cumulative_perms, cumulative_shared_perms;
76b1c7fe 7055
da359909
EGE
7056 GLOBAL_STATE_CODE();
7057
d470ad42
HR
7058 if (!bs->drv) {
7059 return -ENOMEDIUM;
7060 }
7061
9e37271f
KW
7062 /* Make sure that we don't inactivate a child before its parent.
7063 * It will be covered by recursion from the yet active parent. */
7064 if (bdrv_has_bds_parent(bs, true)) {
7065 return 0;
7066 }
7067
7068 assert(!(bs->open_flags & BDRV_O_INACTIVE));
7069
7070 /* Inactivate this node */
7071 if (bs->drv->bdrv_inactivate) {
76b1c7fe
KW
7072 ret = bs->drv->bdrv_inactivate(bs);
7073 if (ret < 0) {
7074 return ret;
7075 }
7076 }
7077
9e37271f 7078 QLIST_FOREACH(parent, &bs->parents, next_parent) {
bd86fb99
HR
7079 if (parent->klass->inactivate) {
7080 ret = parent->klass->inactivate(parent);
9e37271f
KW
7081 if (ret < 0) {
7082 return ret;
cfa1a572
KW
7083 }
7084 }
9e37271f 7085 }
9c5e6594 7086
a13de40a
VSO
7087 bdrv_get_cumulative_perm(bs, &cumulative_perms,
7088 &cumulative_shared_perms);
7089 if (cumulative_perms & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)) {
7090 /* Our inactive parents still need write access. Inactivation failed. */
7091 return -EPERM;
7092 }
7093
9e37271f 7094 bs->open_flags |= BDRV_O_INACTIVE;
7d5b5261 7095
bb87e4d1
VSO
7096 /*
7097 * Update permissions, they may differ for inactive nodes.
7098 * We only tried to loosen restrictions, so errors are not fatal, ignore
7099 * them.
7100 */
f1316edb 7101 bdrv_refresh_perms(bs, NULL, NULL);
9e37271f
KW
7102
7103 /* Recursively inactivate children */
38701b6a 7104 QLIST_FOREACH(child, &bs->children, next) {
9e37271f 7105 ret = bdrv_inactivate_recurse(child->bs);
38701b6a
KW
7106 if (ret < 0) {
7107 return ret;
7108 }
7109 }
7110
76b1c7fe
KW
7111 return 0;
7112}
7113
7114int bdrv_inactivate_all(void)
7115{
79720af6 7116 BlockDriverState *bs = NULL;
88be7b4b 7117 BdrvNextIterator it;
aad0b7a0 7118 int ret = 0;
bd6458e4 7119 GSList *aio_ctxs = NULL, *ctx;
76b1c7fe 7120
f791bf7f 7121 GLOBAL_STATE_CODE();
0e6bad1f 7122 GRAPH_RDLOCK_GUARD_MAINLOOP();
f791bf7f 7123
88be7b4b 7124 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
bd6458e4
PB
7125 AioContext *aio_context = bdrv_get_aio_context(bs);
7126
7127 if (!g_slist_find(aio_ctxs, aio_context)) {
7128 aio_ctxs = g_slist_prepend(aio_ctxs, aio_context);
7129 aio_context_acquire(aio_context);
7130 }
aad0b7a0 7131 }
76b1c7fe 7132
9e37271f
KW
7133 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
7134 /* Nodes with BDS parents are covered by recursion from the last
7135 * parent that gets inactivated. Don't inactivate them a second
7136 * time if that has already happened. */
7137 if (bdrv_has_bds_parent(bs, false)) {
7138 continue;
7139 }
7140 ret = bdrv_inactivate_recurse(bs);
7141 if (ret < 0) {
7142 bdrv_next_cleanup(&it);
7143 goto out;
76b1c7fe
KW
7144 }
7145 }
7146
aad0b7a0 7147out:
bd6458e4
PB
7148 for (ctx = aio_ctxs; ctx != NULL; ctx = ctx->next) {
7149 AioContext *aio_context = ctx->data;
7150 aio_context_release(aio_context);
aad0b7a0 7151 }
bd6458e4 7152 g_slist_free(aio_ctxs);
aad0b7a0
FZ
7153
7154 return ret;
76b1c7fe
KW
7155}
7156
19cb3738
FB
7157/**************************************************************/
7158/* removable device support */
7159
7160/**
7161 * Return TRUE if the media is present
7162 */
1e97be91 7163bool coroutine_fn bdrv_co_is_inserted(BlockDriverState *bs)
19cb3738
FB
7164{
7165 BlockDriver *drv = bs->drv;
28d7a789 7166 BdrvChild *child;
384a48fb 7167 IO_CODE();
c73ff92c 7168 assert_bdrv_graph_readable();
a1aff5bf 7169
e031f750
HR
7170 if (!drv) {
7171 return false;
7172 }
1e97be91
EGE
7173 if (drv->bdrv_co_is_inserted) {
7174 return drv->bdrv_co_is_inserted(bs);
28d7a789
HR
7175 }
7176 QLIST_FOREACH(child, &bs->children, next) {
1e97be91 7177 if (!bdrv_co_is_inserted(child->bs)) {
28d7a789
HR
7178 return false;
7179 }
e031f750 7180 }
28d7a789 7181 return true;
19cb3738
FB
7182}
7183
19cb3738
FB
7184/**
7185 * If eject_flag is TRUE, eject the media. Otherwise, close the tray
7186 */
2531b390 7187void coroutine_fn bdrv_co_eject(BlockDriverState *bs, bool eject_flag)
19cb3738
FB
7188{
7189 BlockDriver *drv = bs->drv;
384a48fb 7190 IO_CODE();
79a292e5 7191 assert_bdrv_graph_readable();
19cb3738 7192
2531b390
EGE
7193 if (drv && drv->bdrv_co_eject) {
7194 drv->bdrv_co_eject(bs, eject_flag);
19cb3738
FB
7195 }
7196}
7197
19cb3738
FB
7198/**
7199 * Lock or unlock the media (if it is locked, the user won't be able
7200 * to eject it manually).
7201 */
2c75261c 7202void coroutine_fn bdrv_co_lock_medium(BlockDriverState *bs, bool locked)
19cb3738
FB
7203{
7204 BlockDriver *drv = bs->drv;
384a48fb 7205 IO_CODE();
79a292e5 7206 assert_bdrv_graph_readable();
025e849a 7207 trace_bdrv_lock_medium(bs, locked);
b8c6d095 7208
2c75261c
EGE
7209 if (drv && drv->bdrv_co_lock_medium) {
7210 drv->bdrv_co_lock_medium(bs, locked);
19cb3738
FB
7211 }
7212}
985a03b0 7213
9fcb0251
FZ
7214/* Get a reference to bs */
7215void bdrv_ref(BlockDriverState *bs)
7216{
f791bf7f 7217 GLOBAL_STATE_CODE();
9fcb0251
FZ
7218 bs->refcnt++;
7219}
7220
7221/* Release a previously grabbed reference to bs.
7222 * If after releasing, reference count is zero, the BlockDriverState is
7223 * deleted. */
7224void bdrv_unref(BlockDriverState *bs)
7225{
f791bf7f 7226 GLOBAL_STATE_CODE();
9a4d5ca6
JC
7227 if (!bs) {
7228 return;
7229 }
9fcb0251
FZ
7230 assert(bs->refcnt > 0);
7231 if (--bs->refcnt == 0) {
7232 bdrv_delete(bs);
7233 }
7234}
7235
ac2ae233
KW
7236/*
7237 * Release a BlockDriverState reference while holding the graph write lock.
7238 *
7239 * Calling bdrv_unref() directly is forbidden while holding the graph lock
7240 * because bdrv_close() both involves polling and taking the graph lock
7241 * internally. bdrv_schedule_unref() instead delays decreasing the refcount and
7242 * possibly closing @bs until the graph lock is released.
7243 */
7244void bdrv_schedule_unref(BlockDriverState *bs)
7245{
7246 if (!bs) {
7247 return;
7248 }
7249 aio_bh_schedule_oneshot(qemu_get_aio_context(),
7250 (QEMUBHFunc *) bdrv_unref, bs);
7251}
7252
fbe40ff7
FZ
7253struct BdrvOpBlocker {
7254 Error *reason;
7255 QLIST_ENTRY(BdrvOpBlocker) list;
7256};
7257
7258bool bdrv_op_is_blocked(BlockDriverState *bs, BlockOpType op, Error **errp)
7259{
7260 BdrvOpBlocker *blocker;
f791bf7f 7261 GLOBAL_STATE_CODE();
0bb79c97 7262
fbe40ff7
FZ
7263 assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
7264 if (!QLIST_EMPTY(&bs->op_blockers[op])) {
7265 blocker = QLIST_FIRST(&bs->op_blockers[op]);
4b576648
MA
7266 error_propagate_prepend(errp, error_copy(blocker->reason),
7267 "Node '%s' is busy: ",
7268 bdrv_get_device_or_node_name(bs));
fbe40ff7
FZ
7269 return true;
7270 }
7271 return false;
7272}
7273
7274void bdrv_op_block(BlockDriverState *bs, BlockOpType op, Error *reason)
7275{
7276 BdrvOpBlocker *blocker;
f791bf7f 7277 GLOBAL_STATE_CODE();
fbe40ff7
FZ
7278 assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
7279
5839e53b 7280 blocker = g_new0(BdrvOpBlocker, 1);
fbe40ff7
FZ
7281 blocker->reason = reason;
7282 QLIST_INSERT_HEAD(&bs->op_blockers[op], blocker, list);
7283}
7284
7285void bdrv_op_unblock(BlockDriverState *bs, BlockOpType op, Error *reason)
7286{
7287 BdrvOpBlocker *blocker, *next;
f791bf7f 7288 GLOBAL_STATE_CODE();
fbe40ff7
FZ
7289 assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
7290 QLIST_FOREACH_SAFE(blocker, &bs->op_blockers[op], list, next) {
7291 if (blocker->reason == reason) {
7292 QLIST_REMOVE(blocker, list);
7293 g_free(blocker);
7294 }
7295 }
7296}
7297
7298void bdrv_op_block_all(BlockDriverState *bs, Error *reason)
7299{
7300 int i;
f791bf7f 7301 GLOBAL_STATE_CODE();
fbe40ff7
FZ
7302 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
7303 bdrv_op_block(bs, i, reason);
7304 }
7305}
7306
7307void bdrv_op_unblock_all(BlockDriverState *bs, Error *reason)
7308{
7309 int i;
f791bf7f 7310 GLOBAL_STATE_CODE();
fbe40ff7
FZ
7311 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
7312 bdrv_op_unblock(bs, i, reason);
7313 }
7314}
7315
7316bool bdrv_op_blocker_is_empty(BlockDriverState *bs)
7317{
7318 int i;
f791bf7f 7319 GLOBAL_STATE_CODE();
fbe40ff7
FZ
7320 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
7321 if (!QLIST_EMPTY(&bs->op_blockers[i])) {
7322 return false;
7323 }
7324 }
7325 return true;
7326}
7327
e1355055
KW
7328/*
7329 * Must not be called while holding the lock of an AioContext other than the
7330 * current one.
7331 */
d92ada22
LC
7332void bdrv_img_create(const char *filename, const char *fmt,
7333 const char *base_filename, const char *base_fmt,
9217283d
FZ
7334 char *options, uint64_t img_size, int flags, bool quiet,
7335 Error **errp)
f88e1a42 7336{
83d0521a
CL
7337 QemuOptsList *create_opts = NULL;
7338 QemuOpts *opts = NULL;
7339 const char *backing_fmt, *backing_file;
7340 int64_t size;
f88e1a42 7341 BlockDriver *drv, *proto_drv;
cc84d90f 7342 Error *local_err = NULL;
f88e1a42
JS
7343 int ret = 0;
7344
f791bf7f
EGE
7345 GLOBAL_STATE_CODE();
7346
f88e1a42
JS
7347 /* Find driver and parse its options */
7348 drv = bdrv_find_format(fmt);
7349 if (!drv) {
71c79813 7350 error_setg(errp, "Unknown file format '%s'", fmt);
d92ada22 7351 return;
f88e1a42
JS
7352 }
7353
b65a5e12 7354 proto_drv = bdrv_find_protocol(filename, true, errp);
f88e1a42 7355 if (!proto_drv) {
d92ada22 7356 return;
f88e1a42
JS
7357 }
7358
c6149724
HR
7359 if (!drv->create_opts) {
7360 error_setg(errp, "Format driver '%s' does not support image creation",
7361 drv->format_name);
7362 return;
7363 }
7364
5a5e7f8c
ML
7365 if (!proto_drv->create_opts) {
7366 error_setg(errp, "Protocol driver '%s' does not support image creation",
7367 proto_drv->format_name);
7368 return;
7369 }
7370
c6e0a6de
KW
7371 aio_context_acquire(qemu_get_aio_context());
7372
f6dc1c31 7373 /* Create parameter list */
c282e1fd 7374 create_opts = qemu_opts_append(create_opts, drv->create_opts);
5a5e7f8c 7375 create_opts = qemu_opts_append(create_opts, proto_drv->create_opts);
f88e1a42 7376
83d0521a 7377 opts = qemu_opts_create(create_opts, NULL, 0, &error_abort);
f88e1a42
JS
7378
7379 /* Parse -o options */
7380 if (options) {
a5f9b9df 7381 if (!qemu_opts_do_parse(opts, options, NULL, errp)) {
f88e1a42
JS
7382 goto out;
7383 }
7384 }
7385
f6dc1c31
KW
7386 if (!qemu_opt_get(opts, BLOCK_OPT_SIZE)) {
7387 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, img_size, &error_abort);
7388 } else if (img_size != UINT64_C(-1)) {
7389 error_setg(errp, "The image size must be specified only once");
7390 goto out;
7391 }
7392
f88e1a42 7393 if (base_filename) {
235e59cf 7394 if (!qemu_opt_set(opts, BLOCK_OPT_BACKING_FILE, base_filename,
3882578b 7395 NULL)) {
71c79813
LC
7396 error_setg(errp, "Backing file not supported for file format '%s'",
7397 fmt);
f88e1a42
JS
7398 goto out;
7399 }
7400 }
7401
7402 if (base_fmt) {
3882578b 7403 if (!qemu_opt_set(opts, BLOCK_OPT_BACKING_FMT, base_fmt, NULL)) {
71c79813
LC
7404 error_setg(errp, "Backing file format not supported for file "
7405 "format '%s'", fmt);
f88e1a42
JS
7406 goto out;
7407 }
7408 }
7409
83d0521a
CL
7410 backing_file = qemu_opt_get(opts, BLOCK_OPT_BACKING_FILE);
7411 if (backing_file) {
7412 if (!strcmp(filename, backing_file)) {
71c79813
LC
7413 error_setg(errp, "Error: Trying to create an image with the "
7414 "same filename as the backing file");
792da93a
JS
7415 goto out;
7416 }
975a7bd2
CK
7417 if (backing_file[0] == '\0') {
7418 error_setg(errp, "Expected backing file name, got empty string");
7419 goto out;
7420 }
792da93a
JS
7421 }
7422
83d0521a 7423 backing_fmt = qemu_opt_get(opts, BLOCK_OPT_BACKING_FMT);
f88e1a42 7424
6e6e55f5
JS
7425 /* The size for the image must always be specified, unless we have a backing
7426 * file and we have not been forbidden from opening it. */
a8b42a1c 7427 size = qemu_opt_get_size(opts, BLOCK_OPT_SIZE, img_size);
6e6e55f5
JS
7428 if (backing_file && !(flags & BDRV_O_NO_BACKING)) {
7429 BlockDriverState *bs;
645ae7d8 7430 char *full_backing;
6e6e55f5
JS
7431 int back_flags;
7432 QDict *backing_options = NULL;
7433
645ae7d8
HR
7434 full_backing =
7435 bdrv_get_full_backing_filename_from_filename(filename, backing_file,
7436 &local_err);
6e6e55f5 7437 if (local_err) {
6e6e55f5
JS
7438 goto out;
7439 }
645ae7d8 7440 assert(full_backing);
29168018 7441
d5b23994
HR
7442 /*
7443 * No need to do I/O here, which allows us to open encrypted
7444 * backing images without needing the secret
7445 */
6e6e55f5
JS
7446 back_flags = flags;
7447 back_flags &= ~(BDRV_O_RDWR | BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING);
d5b23994 7448 back_flags |= BDRV_O_NO_IO;
f88e1a42 7449
cc954f01 7450 backing_options = qdict_new();
6e6e55f5 7451 if (backing_fmt) {
6e6e55f5
JS
7452 qdict_put_str(backing_options, "driver", backing_fmt);
7453 }
cc954f01 7454 qdict_put_bool(backing_options, BDRV_OPT_FORCE_SHARE, true);
e6641719 7455
6e6e55f5
JS
7456 bs = bdrv_open(full_backing, NULL, backing_options, back_flags,
7457 &local_err);
7458 g_free(full_backing);
add8200d
EB
7459 if (!bs) {
7460 error_append_hint(&local_err, "Could not open backing image.\n");
6e6e55f5
JS
7461 goto out;
7462 } else {
d9f059aa 7463 if (!backing_fmt) {
497a30db
EB
7464 error_setg(&local_err,
7465 "Backing file specified without backing format");
fbdffb08 7466 error_append_hint(&local_err, "Detected format of %s.\n",
497a30db
EB
7467 bs->drv->format_name);
7468 goto out;
d9f059aa 7469 }
6e6e55f5
JS
7470 if (size == -1) {
7471 /* Opened BS, have no size */
7472 size = bdrv_getlength(bs);
7473 if (size < 0) {
7474 error_setg_errno(errp, -size, "Could not get size of '%s'",
7475 backing_file);
7476 bdrv_unref(bs);
7477 goto out;
7478 }
7479 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, size, &error_abort);
52bf1e72 7480 }
66f6b814 7481 bdrv_unref(bs);
f88e1a42 7482 }
d9f059aa
EB
7483 /* (backing_file && !(flags & BDRV_O_NO_BACKING)) */
7484 } else if (backing_file && !backing_fmt) {
497a30db
EB
7485 error_setg(&local_err,
7486 "Backing file specified without backing format");
7487 goto out;
d9f059aa 7488 }
6e6e55f5
JS
7489
7490 if (size == -1) {
7491 error_setg(errp, "Image creation needs a size parameter");
7492 goto out;
f88e1a42
JS
7493 }
7494
f382d43a 7495 if (!quiet) {
fe646693 7496 printf("Formatting '%s', fmt=%s ", filename, fmt);
43c5d8f8 7497 qemu_opts_print(opts, " ");
f382d43a 7498 puts("");
4e2f4418 7499 fflush(stdout);
f382d43a 7500 }
83d0521a 7501
c282e1fd 7502 ret = bdrv_create(drv, filename, opts, &local_err);
83d0521a 7503
cc84d90f
HR
7504 if (ret == -EFBIG) {
7505 /* This is generally a better message than whatever the driver would
7506 * deliver (especially because of the cluster_size_hint), since that
7507 * is most probably not much different from "image too large". */
7508 const char *cluster_size_hint = "";
83d0521a 7509 if (qemu_opt_get_size(opts, BLOCK_OPT_CLUSTER_SIZE, 0)) {
cc84d90f 7510 cluster_size_hint = " (try using a larger cluster size)";
f88e1a42 7511 }
cc84d90f
HR
7512 error_setg(errp, "The image size is too large for file format '%s'"
7513 "%s", fmt, cluster_size_hint);
7514 error_free(local_err);
7515 local_err = NULL;
f88e1a42
JS
7516 }
7517
7518out:
83d0521a
CL
7519 qemu_opts_del(opts);
7520 qemu_opts_free(create_opts);
621ff94d 7521 error_propagate(errp, local_err);
c6e0a6de 7522 aio_context_release(qemu_get_aio_context());
f88e1a42 7523}
85d126f3
SH
7524
7525AioContext *bdrv_get_aio_context(BlockDriverState *bs)
7526{
384a48fb 7527 IO_CODE();
33f2a757 7528 return bs ? bs->aio_context : qemu_get_aio_context();
dcd04228
SH
7529}
7530
e336fd4c
KW
7531AioContext *coroutine_fn bdrv_co_enter(BlockDriverState *bs)
7532{
7533 Coroutine *self = qemu_coroutine_self();
7534 AioContext *old_ctx = qemu_coroutine_get_aio_context(self);
7535 AioContext *new_ctx;
384a48fb 7536 IO_CODE();
e336fd4c
KW
7537
7538 /*
7539 * Increase bs->in_flight to ensure that this operation is completed before
7540 * moving the node to a different AioContext. Read new_ctx only afterwards.
7541 */
7542 bdrv_inc_in_flight(bs);
7543
7544 new_ctx = bdrv_get_aio_context(bs);
7545 aio_co_reschedule_self(new_ctx);
7546 return old_ctx;
7547}
7548
7549void coroutine_fn bdrv_co_leave(BlockDriverState *bs, AioContext *old_ctx)
7550{
384a48fb 7551 IO_CODE();
e336fd4c
KW
7552 aio_co_reschedule_self(old_ctx);
7553 bdrv_dec_in_flight(bs);
7554}
7555
18c6ac1c
KW
7556void coroutine_fn bdrv_co_lock(BlockDriverState *bs)
7557{
7558 AioContext *ctx = bdrv_get_aio_context(bs);
7559
7560 /* In the main thread, bs->aio_context won't change concurrently */
7561 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
7562
7563 /*
7564 * We're in coroutine context, so we already hold the lock of the main
7565 * loop AioContext. Don't lock it twice to avoid deadlocks.
7566 */
7567 assert(qemu_in_coroutine());
7568 if (ctx != qemu_get_aio_context()) {
7569 aio_context_acquire(ctx);
7570 }
7571}
7572
7573void coroutine_fn bdrv_co_unlock(BlockDriverState *bs)
7574{
7575 AioContext *ctx = bdrv_get_aio_context(bs);
7576
7577 assert(qemu_in_coroutine());
7578 if (ctx != qemu_get_aio_context()) {
7579 aio_context_release(ctx);
7580 }
7581}
7582
e8a095da
SH
7583static void bdrv_do_remove_aio_context_notifier(BdrvAioNotifier *ban)
7584{
bdb73476 7585 GLOBAL_STATE_CODE();
e8a095da
SH
7586 QLIST_REMOVE(ban, list);
7587 g_free(ban);
7588}
7589
a3a683c3 7590static void bdrv_detach_aio_context(BlockDriverState *bs)
dcd04228 7591{
e8a095da 7592 BdrvAioNotifier *baf, *baf_tmp;
33384421 7593
e8a095da 7594 assert(!bs->walking_aio_notifiers);
da359909 7595 GLOBAL_STATE_CODE();
e8a095da
SH
7596 bs->walking_aio_notifiers = true;
7597 QLIST_FOREACH_SAFE(baf, &bs->aio_notifiers, list, baf_tmp) {
7598 if (baf->deleted) {
7599 bdrv_do_remove_aio_context_notifier(baf);
7600 } else {
7601 baf->detach_aio_context(baf->opaque);
7602 }
33384421 7603 }
e8a095da
SH
7604 /* Never mind iterating again to check for ->deleted. bdrv_close() will
7605 * remove remaining aio notifiers if we aren't called again.
7606 */
7607 bs->walking_aio_notifiers = false;
33384421 7608
1bffe1ae 7609 if (bs->drv && bs->drv->bdrv_detach_aio_context) {
dcd04228
SH
7610 bs->drv->bdrv_detach_aio_context(bs);
7611 }
dcd04228
SH
7612
7613 bs->aio_context = NULL;
7614}
7615
a3a683c3
KW
7616static void bdrv_attach_aio_context(BlockDriverState *bs,
7617 AioContext *new_context)
dcd04228 7618{
e8a095da 7619 BdrvAioNotifier *ban, *ban_tmp;
da359909 7620 GLOBAL_STATE_CODE();
33384421 7621
dcd04228
SH
7622 bs->aio_context = new_context;
7623
1bffe1ae 7624 if (bs->drv && bs->drv->bdrv_attach_aio_context) {
dcd04228
SH
7625 bs->drv->bdrv_attach_aio_context(bs, new_context);
7626 }
33384421 7627
e8a095da
SH
7628 assert(!bs->walking_aio_notifiers);
7629 bs->walking_aio_notifiers = true;
7630 QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_tmp) {
7631 if (ban->deleted) {
7632 bdrv_do_remove_aio_context_notifier(ban);
7633 } else {
7634 ban->attached_aio_context(new_context, ban->opaque);
7635 }
33384421 7636 }
e8a095da 7637 bs->walking_aio_notifiers = false;
dcd04228
SH
7638}
7639
7e8c182f
EGE
7640typedef struct BdrvStateSetAioContext {
7641 AioContext *new_ctx;
7642 BlockDriverState *bs;
7643} BdrvStateSetAioContext;
d616b224 7644
7e8c182f 7645static bool bdrv_parent_change_aio_context(BdrvChild *c, AioContext *ctx,
e08cc001
EGE
7646 GHashTable *visited,
7647 Transaction *tran,
7e8c182f 7648 Error **errp)
5d231849 7649{
f0c28327 7650 GLOBAL_STATE_CODE();
e08cc001 7651 if (g_hash_table_contains(visited, c)) {
5d231849
KW
7652 return true;
7653 }
e08cc001 7654 g_hash_table_add(visited, c);
5d231849 7655
bd86fb99
HR
7656 /*
7657 * A BdrvChildClass that doesn't handle AioContext changes cannot
7658 * tolerate any AioContext changes
7659 */
7e8c182f 7660 if (!c->klass->change_aio_ctx) {
5d231849
KW
7661 char *user = bdrv_child_user_desc(c);
7662 error_setg(errp, "Changing iothreads is not supported by %s", user);
7663 g_free(user);
7664 return false;
7665 }
7e8c182f 7666 if (!c->klass->change_aio_ctx(c, ctx, visited, tran, errp)) {
5d231849
KW
7667 assert(!errp || *errp);
7668 return false;
7669 }
7670 return true;
7671}
7672
7e8c182f 7673bool bdrv_child_change_aio_context(BdrvChild *c, AioContext *ctx,
e08cc001 7674 GHashTable *visited, Transaction *tran,
7e8c182f 7675 Error **errp)
5d231849 7676{
f791bf7f 7677 GLOBAL_STATE_CODE();
e08cc001 7678 if (g_hash_table_contains(visited, c)) {
5d231849
KW
7679 return true;
7680 }
e08cc001 7681 g_hash_table_add(visited, c);
7e8c182f 7682 return bdrv_change_aio_context(c->bs, ctx, visited, tran, errp);
5d231849
KW
7683}
7684
7e8c182f
EGE
7685static void bdrv_set_aio_context_clean(void *opaque)
7686{
7687 BdrvStateSetAioContext *state = (BdrvStateSetAioContext *) opaque;
7688 BlockDriverState *bs = (BlockDriverState *) state->bs;
7689
7690 /* Paired with bdrv_drained_begin in bdrv_change_aio_context() */
7691 bdrv_drained_end(bs);
7692
7693 g_free(state);
7694}
7695
7696static void bdrv_set_aio_context_commit(void *opaque)
7697{
7698 BdrvStateSetAioContext *state = (BdrvStateSetAioContext *) opaque;
7699 BlockDriverState *bs = (BlockDriverState *) state->bs;
7700 AioContext *new_context = state->new_ctx;
7701 AioContext *old_context = bdrv_get_aio_context(bs);
7e8c182f
EGE
7702
7703 /*
7704 * Take the old AioContex when detaching it from bs.
7705 * At this point, new_context lock is already acquired, and we are now
7706 * also taking old_context. This is safe as long as bdrv_detach_aio_context
7707 * does not call AIO_POLL_WHILE().
7708 */
7709 if (old_context != qemu_get_aio_context()) {
7710 aio_context_acquire(old_context);
7711 }
7712 bdrv_detach_aio_context(bs);
7713 if (old_context != qemu_get_aio_context()) {
7714 aio_context_release(old_context);
7715 }
7716 bdrv_attach_aio_context(bs, new_context);
7717}
7718
7719static TransactionActionDrv set_aio_context = {
7720 .commit = bdrv_set_aio_context_commit,
7721 .clean = bdrv_set_aio_context_clean,
7722};
7723
7724/*
7725 * Changes the AioContext used for fd handlers, timers, and BHs by this
7726 * BlockDriverState and all its children and parents.
7727 *
7728 * Must be called from the main AioContext.
7729 *
7730 * The caller must own the AioContext lock for the old AioContext of bs, but it
7731 * must not own the AioContext lock for new_context (unless new_context is the
7732 * same as the current context of bs).
7733 *
7734 * @visited will accumulate all visited BdrvChild objects. The caller is
7735 * responsible for freeing the list afterwards.
7736 */
7737static bool bdrv_change_aio_context(BlockDriverState *bs, AioContext *ctx,
e08cc001 7738 GHashTable *visited, Transaction *tran,
7e8c182f 7739 Error **errp)
5d231849
KW
7740{
7741 BdrvChild *c;
7e8c182f
EGE
7742 BdrvStateSetAioContext *state;
7743
7744 GLOBAL_STATE_CODE();
5d231849
KW
7745
7746 if (bdrv_get_aio_context(bs) == ctx) {
7747 return true;
7748 }
7749
7f831d29 7750 bdrv_graph_rdlock_main_loop();
5d231849 7751 QLIST_FOREACH(c, &bs->parents, next_parent) {
7e8c182f 7752 if (!bdrv_parent_change_aio_context(c, ctx, visited, tran, errp)) {
7f831d29 7753 bdrv_graph_rdunlock_main_loop();
5d231849
KW
7754 return false;
7755 }
7756 }
7e8c182f 7757
5d231849 7758 QLIST_FOREACH(c, &bs->children, next) {
7e8c182f 7759 if (!bdrv_child_change_aio_context(c, ctx, visited, tran, errp)) {
7f831d29 7760 bdrv_graph_rdunlock_main_loop();
5d231849
KW
7761 return false;
7762 }
7763 }
7f831d29 7764 bdrv_graph_rdunlock_main_loop();
5d231849 7765
7e8c182f
EGE
7766 state = g_new(BdrvStateSetAioContext, 1);
7767 *state = (BdrvStateSetAioContext) {
7768 .new_ctx = ctx,
7769 .bs = bs,
7770 };
7771
7772 /* Paired with bdrv_drained_end in bdrv_set_aio_context_clean() */
7773 bdrv_drained_begin(bs);
7774
7775 tran_add(tran, &set_aio_context, state);
7776
5d231849
KW
7777 return true;
7778}
7779
7e8c182f
EGE
7780/*
7781 * Change bs's and recursively all of its parents' and children's AioContext
7782 * to the given new context, returning an error if that isn't possible.
7783 *
7784 * If ignore_child is not NULL, that child (and its subgraph) will not
7785 * be touched.
7786 *
7787 * This function still requires the caller to take the bs current
7788 * AioContext lock, otherwise draining will fail since AIO_WAIT_WHILE
7789 * assumes the lock is always held if bs is in another AioContext.
7790 * For the same reason, it temporarily also holds the new AioContext, since
7791 * bdrv_drained_end calls BDRV_POLL_WHILE that assumes the lock is taken too.
7792 * Therefore the new AioContext lock must not be taken by the caller.
7793 */
a41cfda1
EGE
7794int bdrv_try_change_aio_context(BlockDriverState *bs, AioContext *ctx,
7795 BdrvChild *ignore_child, Error **errp)
5d231849 7796{
7e8c182f 7797 Transaction *tran;
e08cc001 7798 GHashTable *visited;
7e8c182f
EGE
7799 int ret;
7800 AioContext *old_context = bdrv_get_aio_context(bs);
f791bf7f
EGE
7801 GLOBAL_STATE_CODE();
7802
7e8c182f
EGE
7803 /*
7804 * Recursion phase: go through all nodes of the graph.
7805 * Take care of checking that all nodes support changing AioContext
3202d8e4 7806 * and drain them, building a linear list of callbacks to run if everything
7e8c182f
EGE
7807 * is successful (the transaction itself).
7808 */
7809 tran = tran_new();
e08cc001
EGE
7810 visited = g_hash_table_new(NULL, NULL);
7811 if (ignore_child) {
7812 g_hash_table_add(visited, ignore_child);
7813 }
7814 ret = bdrv_change_aio_context(bs, ctx, visited, tran, errp);
7815 g_hash_table_destroy(visited);
7e8c182f
EGE
7816
7817 /*
7818 * Linear phase: go through all callbacks collected in the transaction.
7819 * Run all callbacks collected in the recursion to switch all nodes
7820 * AioContext lock (transaction commit), or undo all changes done in the
7821 * recursion (transaction abort).
7822 */
5d231849
KW
7823
7824 if (!ret) {
7e8c182f
EGE
7825 /* Just run clean() callbacks. No AioContext changed. */
7826 tran_abort(tran);
5d231849
KW
7827 return -EPERM;
7828 }
7829
7e8c182f
EGE
7830 /*
7831 * Release old AioContext, it won't be needed anymore, as all
7832 * bdrv_drained_begin() have been called already.
7833 */
7834 if (qemu_get_aio_context() != old_context) {
7835 aio_context_release(old_context);
7836 }
7837
7838 /*
7839 * Acquire new AioContext since bdrv_drained_end() is going to be called
7840 * after we switched all nodes in the new AioContext, and the function
7841 * assumes that the lock of the bs is always taken.
7842 */
7843 if (qemu_get_aio_context() != ctx) {
7844 aio_context_acquire(ctx);
7845 }
53a7d041 7846
7e8c182f 7847 tran_commit(tran);
5d231849 7848
7e8c182f
EGE
7849 if (qemu_get_aio_context() != ctx) {
7850 aio_context_release(ctx);
7851 }
7852
7853 /* Re-acquire the old AioContext, since the caller takes and releases it. */
7854 if (qemu_get_aio_context() != old_context) {
7855 aio_context_acquire(old_context);
7856 }
7857
7858 return 0;
5d231849
KW
7859}
7860
33384421
HR
7861void bdrv_add_aio_context_notifier(BlockDriverState *bs,
7862 void (*attached_aio_context)(AioContext *new_context, void *opaque),
7863 void (*detach_aio_context)(void *opaque), void *opaque)
7864{
7865 BdrvAioNotifier *ban = g_new(BdrvAioNotifier, 1);
7866 *ban = (BdrvAioNotifier){
7867 .attached_aio_context = attached_aio_context,
7868 .detach_aio_context = detach_aio_context,
7869 .opaque = opaque
7870 };
f791bf7f 7871 GLOBAL_STATE_CODE();
33384421
HR
7872
7873 QLIST_INSERT_HEAD(&bs->aio_notifiers, ban, list);
7874}
7875
7876void bdrv_remove_aio_context_notifier(BlockDriverState *bs,
7877 void (*attached_aio_context)(AioContext *,
7878 void *),
7879 void (*detach_aio_context)(void *),
7880 void *opaque)
7881{
7882 BdrvAioNotifier *ban, *ban_next;
f791bf7f 7883 GLOBAL_STATE_CODE();
33384421
HR
7884
7885 QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_next) {
7886 if (ban->attached_aio_context == attached_aio_context &&
7887 ban->detach_aio_context == detach_aio_context &&
e8a095da
SH
7888 ban->opaque == opaque &&
7889 ban->deleted == false)
33384421 7890 {
e8a095da
SH
7891 if (bs->walking_aio_notifiers) {
7892 ban->deleted = true;
7893 } else {
7894 bdrv_do_remove_aio_context_notifier(ban);
7895 }
33384421
HR
7896 return;
7897 }
7898 }
7899
7900 abort();
7901}
7902
77485434 7903int bdrv_amend_options(BlockDriverState *bs, QemuOpts *opts,
d1402b50 7904 BlockDriverAmendStatusCB *status_cb, void *cb_opaque,
a3579bfa 7905 bool force,
d1402b50 7906 Error **errp)
6f176b48 7907{
f791bf7f 7908 GLOBAL_STATE_CODE();
d470ad42 7909 if (!bs->drv) {
d1402b50 7910 error_setg(errp, "Node is ejected");
d470ad42
HR
7911 return -ENOMEDIUM;
7912 }
c282e1fd 7913 if (!bs->drv->bdrv_amend_options) {
d1402b50
HR
7914 error_setg(errp, "Block driver '%s' does not support option amendment",
7915 bs->drv->format_name);
6f176b48
HR
7916 return -ENOTSUP;
7917 }
a3579bfa
ML
7918 return bs->drv->bdrv_amend_options(bs, opts, status_cb,
7919 cb_opaque, force, errp);
6f176b48 7920}
f6186f49 7921
5d69b5ab
HR
7922/*
7923 * This function checks whether the given @to_replace is allowed to be
7924 * replaced by a node that always shows the same data as @bs. This is
7925 * used for example to verify whether the mirror job can replace
7926 * @to_replace by the target mirrored from @bs.
7927 * To be replaceable, @bs and @to_replace may either be guaranteed to
7928 * always show the same data (because they are only connected through
7929 * filters), or some driver may allow replacing one of its children
7930 * because it can guarantee that this child's data is not visible at
7931 * all (for example, for dissenting quorum children that have no other
7932 * parents).
7933 */
7934bool bdrv_recurse_can_replace(BlockDriverState *bs,
7935 BlockDriverState *to_replace)
7936{
93393e69
HR
7937 BlockDriverState *filtered;
7938
b4ad82aa
EGE
7939 GLOBAL_STATE_CODE();
7940
5d69b5ab
HR
7941 if (!bs || !bs->drv) {
7942 return false;
7943 }
7944
7945 if (bs == to_replace) {
7946 return true;
7947 }
7948
7949 /* See what the driver can do */
7950 if (bs->drv->bdrv_recurse_can_replace) {
7951 return bs->drv->bdrv_recurse_can_replace(bs, to_replace);
7952 }
7953
7954 /* For filters without an own implementation, we can recurse on our own */
93393e69
HR
7955 filtered = bdrv_filter_bs(bs);
7956 if (filtered) {
7957 return bdrv_recurse_can_replace(filtered, to_replace);
5d69b5ab
HR
7958 }
7959
7960 /* Safe default */
7961 return false;
7962}
7963
810803a8
HR
7964/*
7965 * Check whether the given @node_name can be replaced by a node that
7966 * has the same data as @parent_bs. If so, return @node_name's BDS;
7967 * NULL otherwise.
7968 *
7969 * @node_name must be a (recursive) *child of @parent_bs (or this
7970 * function will return NULL).
7971 *
7972 * The result (whether the node can be replaced or not) is only valid
7973 * for as long as no graph or permission changes occur.
7974 */
e12f3784
WC
7975BlockDriverState *check_to_replace_node(BlockDriverState *parent_bs,
7976 const char *node_name, Error **errp)
09158f00
BC
7977{
7978 BlockDriverState *to_replace_bs = bdrv_find_node(node_name);
5a7e7a0b
SH
7979 AioContext *aio_context;
7980
f791bf7f
EGE
7981 GLOBAL_STATE_CODE();
7982
09158f00 7983 if (!to_replace_bs) {
785ec4b1 7984 error_setg(errp, "Failed to find node with node-name='%s'", node_name);
09158f00
BC
7985 return NULL;
7986 }
7987
5a7e7a0b
SH
7988 aio_context = bdrv_get_aio_context(to_replace_bs);
7989 aio_context_acquire(aio_context);
7990
09158f00 7991 if (bdrv_op_is_blocked(to_replace_bs, BLOCK_OP_TYPE_REPLACE, errp)) {
5a7e7a0b
SH
7992 to_replace_bs = NULL;
7993 goto out;
09158f00
BC
7994 }
7995
7996 /* We don't want arbitrary node of the BDS chain to be replaced only the top
7997 * most non filter in order to prevent data corruption.
7998 * Another benefit is that this tests exclude backing files which are
7999 * blocked by the backing blockers.
8000 */
810803a8
HR
8001 if (!bdrv_recurse_can_replace(parent_bs, to_replace_bs)) {
8002 error_setg(errp, "Cannot replace '%s' by a node mirrored from '%s', "
8003 "because it cannot be guaranteed that doing so would not "
8004 "lead to an abrupt change of visible data",
8005 node_name, parent_bs->node_name);
5a7e7a0b
SH
8006 to_replace_bs = NULL;
8007 goto out;
09158f00
BC
8008 }
8009
5a7e7a0b
SH
8010out:
8011 aio_context_release(aio_context);
09158f00
BC
8012 return to_replace_bs;
8013}
448ad91d 8014
97e2f021
HR
8015/**
8016 * Iterates through the list of runtime option keys that are said to
8017 * be "strong" for a BDS. An option is called "strong" if it changes
8018 * a BDS's data. For example, the null block driver's "size" and
8019 * "read-zeroes" options are strong, but its "latency-ns" option is
8020 * not.
8021 *
8022 * If a key returned by this function ends with a dot, all options
8023 * starting with that prefix are strong.
8024 */
8025static const char *const *strong_options(BlockDriverState *bs,
8026 const char *const *curopt)
8027{
8028 static const char *const global_options[] = {
8029 "driver", "filename", NULL
8030 };
8031
8032 if (!curopt) {
8033 return &global_options[0];
8034 }
8035
8036 curopt++;
8037 if (curopt == &global_options[ARRAY_SIZE(global_options) - 1] && bs->drv) {
8038 curopt = bs->drv->strong_runtime_opts;
8039 }
8040
8041 return (curopt && *curopt) ? curopt : NULL;
8042}
8043
8044/**
8045 * Copies all strong runtime options from bs->options to the given
8046 * QDict. The set of strong option keys is determined by invoking
8047 * strong_options().
8048 *
8049 * Returns true iff any strong option was present in bs->options (and
8050 * thus copied to the target QDict) with the exception of "filename"
8051 * and "driver". The caller is expected to use this value to decide
8052 * whether the existence of strong options prevents the generation of
8053 * a plain filename.
8054 */
8055static bool append_strong_runtime_options(QDict *d, BlockDriverState *bs)
8056{
8057 bool found_any = false;
8058 const char *const *option_name = NULL;
8059
8060 if (!bs->drv) {
8061 return false;
8062 }
8063
8064 while ((option_name = strong_options(bs, option_name))) {
8065 bool option_given = false;
8066
8067 assert(strlen(*option_name) > 0);
8068 if ((*option_name)[strlen(*option_name) - 1] != '.') {
8069 QObject *entry = qdict_get(bs->options, *option_name);
8070 if (!entry) {
8071 continue;
8072 }
8073
8074 qdict_put_obj(d, *option_name, qobject_ref(entry));
8075 option_given = true;
8076 } else {
8077 const QDictEntry *entry;
8078 for (entry = qdict_first(bs->options); entry;
8079 entry = qdict_next(bs->options, entry))
8080 {
8081 if (strstart(qdict_entry_key(entry), *option_name, NULL)) {
8082 qdict_put_obj(d, qdict_entry_key(entry),
8083 qobject_ref(qdict_entry_value(entry)));
8084 option_given = true;
8085 }
8086 }
8087 }
8088
8089 /* While "driver" and "filename" need to be included in a JSON filename,
8090 * their existence does not prohibit generation of a plain filename. */
8091 if (!found_any && option_given &&
8092 strcmp(*option_name, "driver") && strcmp(*option_name, "filename"))
8093 {
8094 found_any = true;
8095 }
8096 }
8097
62a01a27
HR
8098 if (!qdict_haskey(d, "driver")) {
8099 /* Drivers created with bdrv_new_open_driver() may not have a
8100 * @driver option. Add it here. */
8101 qdict_put_str(d, "driver", bs->drv->format_name);
8102 }
8103
97e2f021
HR
8104 return found_any;
8105}
8106
90993623
HR
8107/* Note: This function may return false positives; it may return true
8108 * even if opening the backing file specified by bs's image header
8109 * would result in exactly bs->backing. */
fa8fc1d0 8110static bool bdrv_backing_overridden(BlockDriverState *bs)
90993623 8111{
b4ad82aa 8112 GLOBAL_STATE_CODE();
90993623
HR
8113 if (bs->backing) {
8114 return strcmp(bs->auto_backing_file,
8115 bs->backing->bs->filename);
8116 } else {
8117 /* No backing BDS, so if the image header reports any backing
8118 * file, it must have been suppressed */
8119 return bs->auto_backing_file[0] != '\0';
8120 }
8121}
8122
91af7014
HR
8123/* Updates the following BDS fields:
8124 * - exact_filename: A filename which may be used for opening a block device
8125 * which (mostly) equals the given BDS (even without any
8126 * other options; so reading and writing must return the same
8127 * results, but caching etc. may be different)
8128 * - full_open_options: Options which, when given when opening a block device
8129 * (without a filename), result in a BDS (mostly)
8130 * equalling the given one
8131 * - filename: If exact_filename is set, it is copied here. Otherwise,
8132 * full_open_options is converted to a JSON object, prefixed with
8133 * "json:" (for use through the JSON pseudo protocol) and put here.
8134 */
8135void bdrv_refresh_filename(BlockDriverState *bs)
8136{
8137 BlockDriver *drv = bs->drv;
e24518e3 8138 BdrvChild *child;
52f72d6f 8139 BlockDriverState *primary_child_bs;
91af7014 8140 QDict *opts;
90993623 8141 bool backing_overridden;
998b3a1e
HR
8142 bool generate_json_filename; /* Whether our default implementation should
8143 fill exact_filename (false) or not (true) */
91af7014 8144
f791bf7f
EGE
8145 GLOBAL_STATE_CODE();
8146
91af7014
HR
8147 if (!drv) {
8148 return;
8149 }
8150
e24518e3
HR
8151 /* This BDS's file name may depend on any of its children's file names, so
8152 * refresh those first */
8153 QLIST_FOREACH(child, &bs->children, next) {
8154 bdrv_refresh_filename(child->bs);
91af7014
HR
8155 }
8156
bb808d5f
HR
8157 if (bs->implicit) {
8158 /* For implicit nodes, just copy everything from the single child */
8159 child = QLIST_FIRST(&bs->children);
8160 assert(QLIST_NEXT(child, next) == NULL);
8161
8162 pstrcpy(bs->exact_filename, sizeof(bs->exact_filename),
8163 child->bs->exact_filename);
8164 pstrcpy(bs->filename, sizeof(bs->filename), child->bs->filename);
8165
cb895614 8166 qobject_unref(bs->full_open_options);
bb808d5f
HR
8167 bs->full_open_options = qobject_ref(child->bs->full_open_options);
8168
8169 return;
8170 }
8171
90993623
HR
8172 backing_overridden = bdrv_backing_overridden(bs);
8173
8174 if (bs->open_flags & BDRV_O_NO_IO) {
8175 /* Without I/O, the backing file does not change anything.
8176 * Therefore, in such a case (primarily qemu-img), we can
8177 * pretend the backing file has not been overridden even if
8178 * it technically has been. */
8179 backing_overridden = false;
8180 }
8181
97e2f021
HR
8182 /* Gather the options QDict */
8183 opts = qdict_new();
998b3a1e
HR
8184 generate_json_filename = append_strong_runtime_options(opts, bs);
8185 generate_json_filename |= backing_overridden;
97e2f021
HR
8186
8187 if (drv->bdrv_gather_child_options) {
8188 /* Some block drivers may not want to present all of their children's
8189 * options, or name them differently from BdrvChild.name */
8190 drv->bdrv_gather_child_options(bs, opts, backing_overridden);
8191 } else {
8192 QLIST_FOREACH(child, &bs->children, next) {
25191e5f 8193 if (child == bs->backing && !backing_overridden) {
97e2f021
HR
8194 /* We can skip the backing BDS if it has not been overridden */
8195 continue;
8196 }
8197
8198 qdict_put(opts, child->name,
8199 qobject_ref(child->bs->full_open_options));
8200 }
8201
8202 if (backing_overridden && !bs->backing) {
8203 /* Force no backing file */
8204 qdict_put_null(opts, "backing");
8205 }
8206 }
8207
8208 qobject_unref(bs->full_open_options);
8209 bs->full_open_options = opts;
8210
52f72d6f
HR
8211 primary_child_bs = bdrv_primary_bs(bs);
8212
998b3a1e
HR
8213 if (drv->bdrv_refresh_filename) {
8214 /* Obsolete information is of no use here, so drop the old file name
8215 * information before refreshing it */
8216 bs->exact_filename[0] = '\0';
8217
8218 drv->bdrv_refresh_filename(bs);
52f72d6f
HR
8219 } else if (primary_child_bs) {
8220 /*
8221 * Try to reconstruct valid information from the underlying
8222 * file -- this only works for format nodes (filter nodes
8223 * cannot be probed and as such must be selected by the user
8224 * either through an options dict, or through a special
8225 * filename which the filter driver must construct in its
8226 * .bdrv_refresh_filename() implementation).
8227 */
998b3a1e
HR
8228
8229 bs->exact_filename[0] = '\0';
8230
fb695c74
HR
8231 /*
8232 * We can use the underlying file's filename if:
8233 * - it has a filename,
52f72d6f 8234 * - the current BDS is not a filter,
fb695c74
HR
8235 * - the file is a protocol BDS, and
8236 * - opening that file (as this BDS's format) will automatically create
8237 * the BDS tree we have right now, that is:
8238 * - the user did not significantly change this BDS's behavior with
8239 * some explicit (strong) options
8240 * - no non-file child of this BDS has been overridden by the user
8241 * Both of these conditions are represented by generate_json_filename.
8242 */
52f72d6f
HR
8243 if (primary_child_bs->exact_filename[0] &&
8244 primary_child_bs->drv->bdrv_file_open &&
8245 !drv->is_filter && !generate_json_filename)
fb695c74 8246 {
52f72d6f 8247 strcpy(bs->exact_filename, primary_child_bs->exact_filename);
998b3a1e
HR
8248 }
8249 }
8250
91af7014
HR
8251 if (bs->exact_filename[0]) {
8252 pstrcpy(bs->filename, sizeof(bs->filename), bs->exact_filename);
97e2f021 8253 } else {
eab3a467 8254 GString *json = qobject_to_json(QOBJECT(bs->full_open_options));
5c86bdf1 8255 if (snprintf(bs->filename, sizeof(bs->filename), "json:%s",
eab3a467 8256 json->str) >= sizeof(bs->filename)) {
5c86bdf1
EB
8257 /* Give user a hint if we truncated things. */
8258 strcpy(bs->filename + sizeof(bs->filename) - 4, "...");
8259 }
eab3a467 8260 g_string_free(json, true);
91af7014
HR
8261 }
8262}
e06018ad 8263
1e89d0f9
HR
8264char *bdrv_dirname(BlockDriverState *bs, Error **errp)
8265{
8266 BlockDriver *drv = bs->drv;
52f72d6f 8267 BlockDriverState *child_bs;
1e89d0f9 8268
f791bf7f
EGE
8269 GLOBAL_STATE_CODE();
8270
1e89d0f9
HR
8271 if (!drv) {
8272 error_setg(errp, "Node '%s' is ejected", bs->node_name);
8273 return NULL;
8274 }
8275
8276 if (drv->bdrv_dirname) {
8277 return drv->bdrv_dirname(bs, errp);
8278 }
8279
52f72d6f
HR
8280 child_bs = bdrv_primary_bs(bs);
8281 if (child_bs) {
8282 return bdrv_dirname(child_bs, errp);
1e89d0f9
HR
8283 }
8284
8285 bdrv_refresh_filename(bs);
8286 if (bs->exact_filename[0] != '\0') {
8287 return path_combine(bs->exact_filename, "");
8288 }
8289
8290 error_setg(errp, "Cannot generate a base directory for %s nodes",
8291 drv->format_name);
8292 return NULL;
8293}
8294
e06018ad
WC
8295/*
8296 * Hot add/remove a BDS's child. So the user can take a child offline when
8297 * it is broken and take a new child online
8298 */
8299void bdrv_add_child(BlockDriverState *parent_bs, BlockDriverState *child_bs,
8300 Error **errp)
8301{
f791bf7f 8302 GLOBAL_STATE_CODE();
e06018ad
WC
8303 if (!parent_bs->drv || !parent_bs->drv->bdrv_add_child) {
8304 error_setg(errp, "The node %s does not support adding a child",
8305 bdrv_get_device_or_node_name(parent_bs));
8306 return;
8307 }
8308
774c726c
SL
8309 /*
8310 * Non-zoned block drivers do not follow zoned storage constraints
8311 * (i.e. sequential writes to zones). Refuse mixing zoned and non-zoned
8312 * drivers in a graph.
8313 */
8314 if (!parent_bs->drv->supports_zoned_children &&
8315 child_bs->bl.zoned == BLK_Z_HM) {
8316 /*
8317 * The host-aware model allows zoned storage constraints and random
8318 * write. Allow mixing host-aware and non-zoned drivers. Using
8319 * host-aware device as a regular device.
8320 */
8321 error_setg(errp, "Cannot add a %s child to a %s parent",
8322 child_bs->bl.zoned == BLK_Z_HM ? "zoned" : "non-zoned",
8323 parent_bs->drv->supports_zoned_children ?
8324 "support zoned children" : "not support zoned children");
8325 return;
8326 }
8327
e06018ad
WC
8328 if (!QLIST_EMPTY(&child_bs->parents)) {
8329 error_setg(errp, "The node %s already has a parent",
8330 child_bs->node_name);
8331 return;
8332 }
8333
8334 parent_bs->drv->bdrv_add_child(parent_bs, child_bs, errp);
8335}
8336
8337void bdrv_del_child(BlockDriverState *parent_bs, BdrvChild *child, Error **errp)
8338{
8339 BdrvChild *tmp;
8340
f791bf7f 8341 GLOBAL_STATE_CODE();
e06018ad
WC
8342 if (!parent_bs->drv || !parent_bs->drv->bdrv_del_child) {
8343 error_setg(errp, "The node %s does not support removing a child",
8344 bdrv_get_device_or_node_name(parent_bs));
8345 return;
8346 }
8347
8348 QLIST_FOREACH(tmp, &parent_bs->children, next) {
8349 if (tmp == child) {
8350 break;
8351 }
8352 }
8353
8354 if (!tmp) {
8355 error_setg(errp, "The node %s does not have a child named %s",
8356 bdrv_get_device_or_node_name(parent_bs),
8357 bdrv_get_device_or_node_name(child->bs));
8358 return;
8359 }
8360
8361 parent_bs->drv->bdrv_del_child(parent_bs, child, errp);
8362}
6f7a3b53
HR
8363
8364int bdrv_make_empty(BdrvChild *c, Error **errp)
8365{
8366 BlockDriver *drv = c->bs->drv;
8367 int ret;
8368
f791bf7f 8369 GLOBAL_STATE_CODE();
6f7a3b53
HR
8370 assert(c->perm & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED));
8371
8372 if (!drv->bdrv_make_empty) {
8373 error_setg(errp, "%s does not support emptying nodes",
8374 drv->format_name);
8375 return -ENOTSUP;
8376 }
8377
8378 ret = drv->bdrv_make_empty(c->bs);
8379 if (ret < 0) {
8380 error_setg_errno(errp, -ret, "Failed to empty %s",
8381 c->bs->filename);
8382 return ret;
8383 }
8384
8385 return 0;
8386}
9a6fc887
HR
8387
8388/*
8389 * Return the child that @bs acts as an overlay for, and from which data may be
8390 * copied in COW or COR operations. Usually this is the backing file.
8391 */
8392BdrvChild *bdrv_cow_child(BlockDriverState *bs)
8393{
967d7905
EGE
8394 IO_CODE();
8395
9a6fc887
HR
8396 if (!bs || !bs->drv) {
8397 return NULL;
8398 }
8399
8400 if (bs->drv->is_filter) {
8401 return NULL;
8402 }
8403
8404 if (!bs->backing) {
8405 return NULL;
8406 }
8407
8408 assert(bs->backing->role & BDRV_CHILD_COW);
8409 return bs->backing;
8410}
8411
8412/*
8413 * If @bs acts as a filter for exactly one of its children, return
8414 * that child.
8415 */
8416BdrvChild *bdrv_filter_child(BlockDriverState *bs)
8417{
8418 BdrvChild *c;
967d7905 8419 IO_CODE();
9a6fc887
HR
8420
8421 if (!bs || !bs->drv) {
8422 return NULL;
8423 }
8424
8425 if (!bs->drv->is_filter) {
8426 return NULL;
8427 }
8428
8429 /* Only one of @backing or @file may be used */
8430 assert(!(bs->backing && bs->file));
8431
8432 c = bs->backing ?: bs->file;
8433 if (!c) {
8434 return NULL;
8435 }
8436
8437 assert(c->role & BDRV_CHILD_FILTERED);
8438 return c;
8439}
8440
8441/*
8442 * Return either the result of bdrv_cow_child() or bdrv_filter_child(),
8443 * whichever is non-NULL.
8444 *
8445 * Return NULL if both are NULL.
8446 */
8447BdrvChild *bdrv_filter_or_cow_child(BlockDriverState *bs)
8448{
8449 BdrvChild *cow_child = bdrv_cow_child(bs);
8450 BdrvChild *filter_child = bdrv_filter_child(bs);
967d7905 8451 IO_CODE();
9a6fc887
HR
8452
8453 /* Filter nodes cannot have COW backing files */
8454 assert(!(cow_child && filter_child));
8455
8456 return cow_child ?: filter_child;
8457}
8458
8459/*
8460 * Return the primary child of this node: For filters, that is the
8461 * filtered child. For other nodes, that is usually the child storing
8462 * metadata.
8463 * (A generally more helpful description is that this is (usually) the
8464 * child that has the same filename as @bs.)
8465 *
8466 * Drivers do not necessarily have a primary child; for example quorum
8467 * does not.
8468 */
8469BdrvChild *bdrv_primary_child(BlockDriverState *bs)
8470{
8471 BdrvChild *c, *found = NULL;
967d7905 8472 IO_CODE();
9a6fc887
HR
8473
8474 QLIST_FOREACH(c, &bs->children, next) {
8475 if (c->role & BDRV_CHILD_PRIMARY) {
8476 assert(!found);
8477 found = c;
8478 }
8479 }
8480
8481 return found;
8482}
d38d7eb8
HR
8483
8484static BlockDriverState *bdrv_do_skip_filters(BlockDriverState *bs,
8485 bool stop_on_explicit_filter)
8486{
8487 BdrvChild *c;
8488
8489 if (!bs) {
8490 return NULL;
8491 }
8492
8493 while (!(stop_on_explicit_filter && !bs->implicit)) {
8494 c = bdrv_filter_child(bs);
8495 if (!c) {
8496 /*
8497 * A filter that is embedded in a working block graph must
8498 * have a child. Assert this here so this function does
8499 * not return a filter node that is not expected by the
8500 * caller.
8501 */
8502 assert(!bs->drv || !bs->drv->is_filter);
8503 break;
8504 }
8505 bs = c->bs;
8506 }
8507 /*
8508 * Note that this treats nodes with bs->drv == NULL as not being
8509 * filters (bs->drv == NULL should be replaced by something else
8510 * anyway).
8511 * The advantage of this behavior is that this function will thus
8512 * always return a non-NULL value (given a non-NULL @bs).
8513 */
8514
8515 return bs;
8516}
8517
8518/*
8519 * Return the first BDS that has not been added implicitly or that
8520 * does not have a filtered child down the chain starting from @bs
8521 * (including @bs itself).
8522 */
8523BlockDriverState *bdrv_skip_implicit_filters(BlockDriverState *bs)
8524{
b4ad82aa 8525 GLOBAL_STATE_CODE();
d38d7eb8
HR
8526 return bdrv_do_skip_filters(bs, true);
8527}
8528
8529/*
8530 * Return the first BDS that does not have a filtered child down the
8531 * chain starting from @bs (including @bs itself).
8532 */
8533BlockDriverState *bdrv_skip_filters(BlockDriverState *bs)
8534{
967d7905 8535 IO_CODE();
d38d7eb8
HR
8536 return bdrv_do_skip_filters(bs, false);
8537}
8538
8539/*
8540 * For a backing chain, return the first non-filter backing image of
8541 * the first non-filter image.
8542 */
8543BlockDriverState *bdrv_backing_chain_next(BlockDriverState *bs)
8544{
967d7905 8545 IO_CODE();
d38d7eb8
HR
8546 return bdrv_skip_filters(bdrv_cow_bs(bdrv_skip_filters(bs)));
8547}
0bc329fb
HR
8548
8549/**
8550 * Check whether [offset, offset + bytes) overlaps with the cached
8551 * block-status data region.
8552 *
8553 * If so, and @pnum is not NULL, set *pnum to `bsc.data_end - offset`,
8554 * which is what bdrv_bsc_is_data()'s interface needs.
8555 * Otherwise, *pnum is not touched.
8556 */
8557static bool bdrv_bsc_range_overlaps_locked(BlockDriverState *bs,
8558 int64_t offset, int64_t bytes,
8559 int64_t *pnum)
8560{
8561 BdrvBlockStatusCache *bsc = qatomic_rcu_read(&bs->block_status_cache);
8562 bool overlaps;
8563
8564 overlaps =
8565 qatomic_read(&bsc->valid) &&
8566 ranges_overlap(offset, bytes, bsc->data_start,
8567 bsc->data_end - bsc->data_start);
8568
8569 if (overlaps && pnum) {
8570 *pnum = bsc->data_end - offset;
8571 }
8572
8573 return overlaps;
8574}
8575
8576/**
8577 * See block_int.h for this function's documentation.
8578 */
8579bool bdrv_bsc_is_data(BlockDriverState *bs, int64_t offset, int64_t *pnum)
8580{
967d7905 8581 IO_CODE();
0bc329fb 8582 RCU_READ_LOCK_GUARD();
0bc329fb
HR
8583 return bdrv_bsc_range_overlaps_locked(bs, offset, 1, pnum);
8584}
8585
8586/**
8587 * See block_int.h for this function's documentation.
8588 */
8589void bdrv_bsc_invalidate_range(BlockDriverState *bs,
8590 int64_t offset, int64_t bytes)
8591{
967d7905 8592 IO_CODE();
0bc329fb
HR
8593 RCU_READ_LOCK_GUARD();
8594
8595 if (bdrv_bsc_range_overlaps_locked(bs, offset, bytes, NULL)) {
8596 qatomic_set(&bs->block_status_cache->valid, false);
8597 }
8598}
8599
8600/**
8601 * See block_int.h for this function's documentation.
8602 */
8603void bdrv_bsc_fill(BlockDriverState *bs, int64_t offset, int64_t bytes)
8604{
8605 BdrvBlockStatusCache *new_bsc = g_new(BdrvBlockStatusCache, 1);
8606 BdrvBlockStatusCache *old_bsc;
967d7905 8607 IO_CODE();
0bc329fb
HR
8608
8609 *new_bsc = (BdrvBlockStatusCache) {
8610 .valid = true,
8611 .data_start = offset,
8612 .data_end = offset + bytes,
8613 };
8614
8615 QEMU_LOCK_GUARD(&bs->bsc_modify_lock);
8616
8617 old_bsc = qatomic_rcu_read(&bs->block_status_cache);
8618 qatomic_rcu_set(&bs->block_status_cache, new_bsc);
8619 if (old_bsc) {
8620 g_free_rcu(old_bsc, rcu);
8621 }
8622}