]> git.proxmox.com Git - mirror_qemu.git/blame - block.c
accel/tcg: Always require can_do_io
[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
10e5d707
KW
282static int bdrv_can_set_read_only(BlockDriverState *bs, bool read_only,
283 bool ignore_allow_rdw, Error **errp)
fe5241bf 284{
384a48fb
EGE
285 IO_CODE();
286
e2b8247a
JC
287 /* Do not set read_only if copy_on_read is enabled */
288 if (bs->copy_on_read && read_only) {
289 error_setg(errp, "Can't set node '%s' to r/o with copy-on-read enabled",
290 bdrv_get_device_or_node_name(bs));
291 return -EINVAL;
292 }
293
d6fcdf06 294 /* Do not clear read_only if it is prohibited */
54a32bfe
KW
295 if (!read_only && !(bs->open_flags & BDRV_O_ALLOW_RDWR) &&
296 !ignore_allow_rdw)
297 {
d6fcdf06
JC
298 error_setg(errp, "Node '%s' is read only",
299 bdrv_get_device_or_node_name(bs));
300 return -EPERM;
301 }
302
45803a03
JC
303 return 0;
304}
305
eaa2410f
KW
306/*
307 * Called by a driver that can only provide a read-only image.
308 *
309 * Returns 0 if the node is already read-only or it could switch the node to
310 * read-only because BDRV_O_AUTO_RDONLY is set.
311 *
312 * Returns -EACCES if the node is read-write and BDRV_O_AUTO_RDONLY is not set
313 * or bdrv_can_set_read_only() forbids making the node read-only. If @errmsg
314 * is not NULL, it is used as the error message for the Error object.
315 */
316int bdrv_apply_auto_read_only(BlockDriverState *bs, const char *errmsg,
317 Error **errp)
45803a03
JC
318{
319 int ret = 0;
384a48fb 320 IO_CODE();
45803a03 321
eaa2410f
KW
322 if (!(bs->open_flags & BDRV_O_RDWR)) {
323 return 0;
324 }
325 if (!(bs->open_flags & BDRV_O_AUTO_RDONLY)) {
326 goto fail;
45803a03
JC
327 }
328
eaa2410f
KW
329 ret = bdrv_can_set_read_only(bs, true, false, NULL);
330 if (ret < 0) {
331 goto fail;
eeae6a59
KW
332 }
333
eaa2410f
KW
334 bs->open_flags &= ~BDRV_O_RDWR;
335
e2b8247a 336 return 0;
eaa2410f
KW
337
338fail:
339 error_setg(errp, "%s", errmsg ?: "Image is read-only");
340 return -EACCES;
fe5241bf
JC
341}
342
645ae7d8
HR
343/*
344 * If @backing is empty, this function returns NULL without setting
345 * @errp. In all other cases, NULL will only be returned with @errp
346 * set.
347 *
348 * Therefore, a return value of NULL without @errp set means that
349 * there is no backing file; if @errp is set, there is one but its
350 * absolute filename cannot be generated.
351 */
352char *bdrv_get_full_backing_filename_from_filename(const char *backed,
353 const char *backing,
354 Error **errp)
dc5a1371 355{
645ae7d8
HR
356 if (backing[0] == '\0') {
357 return NULL;
358 } else if (path_has_protocol(backing) || path_is_absolute(backing)) {
359 return g_strdup(backing);
9f07429e
HR
360 } else if (backed[0] == '\0' || strstart(backed, "json:", NULL)) {
361 error_setg(errp, "Cannot use relative backing file names for '%s'",
362 backed);
645ae7d8 363 return NULL;
dc5a1371 364 } else {
645ae7d8 365 return path_combine(backed, backing);
dc5a1371
PB
366 }
367}
368
9f4793d8
HR
369/*
370 * If @filename is empty or NULL, this function returns NULL without
371 * setting @errp. In all other cases, NULL will only be returned with
372 * @errp set.
373 */
374static char *bdrv_make_absolute_filename(BlockDriverState *relative_to,
375 const char *filename, Error **errp)
0a82855a 376{
8df68616 377 char *dir, *full_name;
9f4793d8 378
8df68616
HR
379 if (!filename || filename[0] == '\0') {
380 return NULL;
381 } else if (path_has_protocol(filename) || path_is_absolute(filename)) {
382 return g_strdup(filename);
383 }
9f07429e 384
8df68616
HR
385 dir = bdrv_dirname(relative_to, errp);
386 if (!dir) {
387 return NULL;
388 }
f30c66ba 389
8df68616
HR
390 full_name = g_strconcat(dir, filename, NULL);
391 g_free(dir);
392 return full_name;
9f4793d8
HR
393}
394
395char *bdrv_get_full_backing_filename(BlockDriverState *bs, Error **errp)
396{
f791bf7f 397 GLOBAL_STATE_CODE();
9f4793d8 398 return bdrv_make_absolute_filename(bs, bs->backing_file, errp);
0a82855a
HR
399}
400
0eb7217e
SH
401void bdrv_register(BlockDriver *bdrv)
402{
a15f08dc 403 assert(bdrv->format_name);
f791bf7f 404 GLOBAL_STATE_CODE();
8a22f02a 405 QLIST_INSERT_HEAD(&bdrv_drivers, bdrv, list);
ea2384d3 406}
b338082b 407
e4e9986b
MA
408BlockDriverState *bdrv_new(void)
409{
410 BlockDriverState *bs;
411 int i;
412
f791bf7f
EGE
413 GLOBAL_STATE_CODE();
414
5839e53b 415 bs = g_new0(BlockDriverState, 1);
e4654d2d 416 QLIST_INIT(&bs->dirty_bitmaps);
fbe40ff7
FZ
417 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
418 QLIST_INIT(&bs->op_blockers[i]);
419 }
fa9185fc 420 qemu_mutex_init(&bs->reqs_lock);
2119882c 421 qemu_mutex_init(&bs->dirty_bitmap_mutex);
9fcb0251 422 bs->refcnt = 1;
dcd04228 423 bs->aio_context = qemu_get_aio_context();
d7d512f6 424
3ff2f67a
EY
425 qemu_co_queue_init(&bs->flush_queue);
426
0bc329fb
HR
427 qemu_co_mutex_init(&bs->bsc_modify_lock);
428 bs->block_status_cache = g_new0(BdrvBlockStatusCache, 1);
429
0f12264e
KW
430 for (i = 0; i < bdrv_drain_all_count; i++) {
431 bdrv_drained_begin(bs);
432 }
433
2c1d04e0
HR
434 QTAILQ_INSERT_TAIL(&all_bdrv_states, bs, bs_list);
435
b338082b
FB
436 return bs;
437}
438
88d88798 439static BlockDriver *bdrv_do_find_format(const char *format_name)
ea2384d3
FB
440{
441 BlockDriver *drv1;
bdb73476 442 GLOBAL_STATE_CODE();
88d88798 443
8a22f02a
SH
444 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
445 if (!strcmp(drv1->format_name, format_name)) {
ea2384d3 446 return drv1;
8a22f02a 447 }
ea2384d3 448 }
88d88798 449
ea2384d3
FB
450 return NULL;
451}
452
88d88798
MM
453BlockDriver *bdrv_find_format(const char *format_name)
454{
455 BlockDriver *drv1;
456 int i;
457
f791bf7f
EGE
458 GLOBAL_STATE_CODE();
459
88d88798
MM
460 drv1 = bdrv_do_find_format(format_name);
461 if (drv1) {
462 return drv1;
463 }
464
465 /* The driver isn't registered, maybe we need to load a module */
466 for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); ++i) {
467 if (!strcmp(block_driver_modules[i].format_name, format_name)) {
c551fb0b
CF
468 Error *local_err = NULL;
469 int rv = block_module_load(block_driver_modules[i].library_name,
470 &local_err);
471 if (rv > 0) {
472 return bdrv_do_find_format(format_name);
473 } else if (rv < 0) {
474 error_report_err(local_err);
475 }
88d88798
MM
476 break;
477 }
478 }
c551fb0b 479 return NULL;
88d88798
MM
480}
481
9ac404c5 482static int bdrv_format_is_whitelisted(const char *format_name, bool read_only)
eb852011 483{
b64ec4e4
FZ
484 static const char *whitelist_rw[] = {
485 CONFIG_BDRV_RW_WHITELIST
859aef02 486 NULL
b64ec4e4
FZ
487 };
488 static const char *whitelist_ro[] = {
489 CONFIG_BDRV_RO_WHITELIST
859aef02 490 NULL
eb852011
MA
491 };
492 const char **p;
493
b64ec4e4 494 if (!whitelist_rw[0] && !whitelist_ro[0]) {
eb852011 495 return 1; /* no whitelist, anything goes */
b64ec4e4 496 }
eb852011 497
b64ec4e4 498 for (p = whitelist_rw; *p; p++) {
9ac404c5 499 if (!strcmp(format_name, *p)) {
eb852011
MA
500 return 1;
501 }
502 }
b64ec4e4
FZ
503 if (read_only) {
504 for (p = whitelist_ro; *p; p++) {
9ac404c5 505 if (!strcmp(format_name, *p)) {
b64ec4e4
FZ
506 return 1;
507 }
508 }
509 }
eb852011
MA
510 return 0;
511}
512
9ac404c5
AS
513int bdrv_is_whitelisted(BlockDriver *drv, bool read_only)
514{
f791bf7f 515 GLOBAL_STATE_CODE();
9ac404c5
AS
516 return bdrv_format_is_whitelisted(drv->format_name, read_only);
517}
518
e6ff69bf
DB
519bool bdrv_uses_whitelist(void)
520{
521 return use_bdrv_whitelist;
522}
523
5b7e1542
ZYW
524typedef struct CreateCo {
525 BlockDriver *drv;
526 char *filename;
83d0521a 527 QemuOpts *opts;
5b7e1542 528 int ret;
cc84d90f 529 Error *err;
5b7e1542
ZYW
530} CreateCo;
531
741443eb
EGE
532int coroutine_fn bdrv_co_create(BlockDriver *drv, const char *filename,
533 QemuOpts *opts, Error **errp)
5b7e1542 534{
cc84d90f 535 int ret;
84bdf21f
EGE
536 GLOBAL_STATE_CODE();
537 ERRP_GUARD();
538
539 if (!drv->bdrv_co_create_opts) {
540 error_setg(errp, "Driver '%s' does not support image creation",
541 drv->format_name);
542 return -ENOTSUP;
543 }
544
545 ret = drv->bdrv_co_create_opts(drv, filename, opts, errp);
546 if (ret < 0 && !*errp) {
547 error_setg_errno(errp, -ret, "Could not create image");
548 }
cc84d90f 549
84bdf21f
EGE
550 return ret;
551}
552
fd17146c
HR
553/**
554 * Helper function for bdrv_create_file_fallback(): Resize @blk to at
555 * least the given @minimum_size.
556 *
557 * On success, return @blk's actual length.
558 * Otherwise, return -errno.
559 */
84569a7d
PB
560static int64_t coroutine_fn GRAPH_UNLOCKED
561create_file_fallback_truncate(BlockBackend *blk, int64_t minimum_size,
562 Error **errp)
84a12e66 563{
cc84d90f 564 Error *local_err = NULL;
fd17146c 565 int64_t size;
cc84d90f 566 int ret;
84a12e66 567
bdb73476
EGE
568 GLOBAL_STATE_CODE();
569
84569a7d
PB
570 ret = blk_co_truncate(blk, minimum_size, false, PREALLOC_MODE_OFF, 0,
571 &local_err);
fd17146c
HR
572 if (ret < 0 && ret != -ENOTSUP) {
573 error_propagate(errp, local_err);
574 return ret;
575 }
576
84569a7d 577 size = blk_co_getlength(blk);
fd17146c
HR
578 if (size < 0) {
579 error_free(local_err);
580 error_setg_errno(errp, -size,
581 "Failed to inquire the new image file's length");
582 return size;
583 }
584
585 if (size < minimum_size) {
586 /* Need to grow the image, but we failed to do that */
587 error_propagate(errp, local_err);
588 return -ENOTSUP;
589 }
590
591 error_free(local_err);
592 local_err = NULL;
593
594 return size;
595}
596
597/**
598 * Helper function for bdrv_create_file_fallback(): Zero the first
599 * sector to remove any potentially pre-existing image header.
600 */
881a4c55
PB
601static int coroutine_fn
602create_file_fallback_zero_first_sector(BlockBackend *blk,
603 int64_t current_size,
604 Error **errp)
fd17146c
HR
605{
606 int64_t bytes_to_clear;
607 int ret;
608
bdb73476
EGE
609 GLOBAL_STATE_CODE();
610
fd17146c
HR
611 bytes_to_clear = MIN(current_size, BDRV_SECTOR_SIZE);
612 if (bytes_to_clear) {
ce47ff20 613 ret = blk_co_pwrite_zeroes(blk, 0, bytes_to_clear, BDRV_REQ_MAY_UNMAP);
fd17146c
HR
614 if (ret < 0) {
615 error_setg_errno(errp, -ret,
616 "Failed to clear the new image's first sector");
617 return ret;
618 }
619 }
620
621 return 0;
622}
623
5a5e7f8c
ML
624/**
625 * Simple implementation of bdrv_co_create_opts for protocol drivers
626 * which only support creation via opening a file
627 * (usually existing raw storage device)
628 */
629int coroutine_fn bdrv_co_create_opts_simple(BlockDriver *drv,
630 const char *filename,
631 QemuOpts *opts,
632 Error **errp)
fd17146c
HR
633{
634 BlockBackend *blk;
eeea1faa 635 QDict *options;
fd17146c
HR
636 int64_t size = 0;
637 char *buf = NULL;
638 PreallocMode prealloc;
639 Error *local_err = NULL;
640 int ret;
641
b4ad82aa
EGE
642 GLOBAL_STATE_CODE();
643
fd17146c
HR
644 size = qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0);
645 buf = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC);
646 prealloc = qapi_enum_parse(&PreallocMode_lookup, buf,
647 PREALLOC_MODE_OFF, &local_err);
648 g_free(buf);
649 if (local_err) {
650 error_propagate(errp, local_err);
651 return -EINVAL;
652 }
653
654 if (prealloc != PREALLOC_MODE_OFF) {
655 error_setg(errp, "Unsupported preallocation mode '%s'",
656 PreallocMode_str(prealloc));
657 return -ENOTSUP;
658 }
659
eeea1faa 660 options = qdict_new();
fd17146c
HR
661 qdict_put_str(options, "driver", drv->format_name);
662
be1a732c
KW
663 blk = blk_co_new_open(filename, NULL, options,
664 BDRV_O_RDWR | BDRV_O_RESIZE, errp);
fd17146c 665 if (!blk) {
81624867
HC
666 error_prepend(errp, "Protocol driver '%s' does not support creating "
667 "new images, so an existing image must be selected as "
668 "the target; however, opening the given target as an "
669 "existing image failed: ",
fd17146c
HR
670 drv->format_name);
671 return -EINVAL;
672 }
673
674 size = create_file_fallback_truncate(blk, size, errp);
675 if (size < 0) {
676 ret = size;
677 goto out;
678 }
679
680 ret = create_file_fallback_zero_first_sector(blk, size, errp);
681 if (ret < 0) {
682 goto out;
683 }
684
685 ret = 0;
686out:
b2ab5f54 687 blk_co_unref(blk);
fd17146c
HR
688 return ret;
689}
690
2475a0d0
EGE
691int coroutine_fn bdrv_co_create_file(const char *filename, QemuOpts *opts,
692 Error **errp)
fd17146c 693{
729222af 694 QemuOpts *protocol_opts;
fd17146c 695 BlockDriver *drv;
729222af
SG
696 QDict *qdict;
697 int ret;
fd17146c 698
f791bf7f
EGE
699 GLOBAL_STATE_CODE();
700
b65a5e12 701 drv = bdrv_find_protocol(filename, true, errp);
84a12e66 702 if (drv == NULL) {
16905d71 703 return -ENOENT;
84a12e66
CH
704 }
705
729222af
SG
706 if (!drv->create_opts) {
707 error_setg(errp, "Driver '%s' does not support image creation",
708 drv->format_name);
709 return -ENOTSUP;
710 }
711
712 /*
713 * 'opts' contains a QemuOptsList with a combination of format and protocol
714 * default values.
715 *
716 * The format properly removes its options, but the default values remain
717 * in 'opts->list'. So if the protocol has options with the same name
718 * (e.g. rbd has 'cluster_size' as qcow2), it will see the default values
719 * of the format, since for overlapping options, the format wins.
720 *
721 * To avoid this issue, lets convert QemuOpts to QDict, in this way we take
722 * only the set options, and then convert it back to QemuOpts, using the
723 * create_opts of the protocol. So the new QemuOpts, will contain only the
724 * protocol defaults.
725 */
726 qdict = qemu_opts_to_qdict(opts, NULL);
727 protocol_opts = qemu_opts_from_qdict(drv->create_opts, qdict, errp);
728 if (protocol_opts == NULL) {
729 ret = -EINVAL;
730 goto out;
731 }
732
2475a0d0 733 ret = bdrv_co_create(drv, filename, protocol_opts, errp);
729222af
SG
734out:
735 qemu_opts_del(protocol_opts);
736 qobject_unref(qdict);
737 return ret;
84a12e66
CH
738}
739
e1d7f8bb
DHB
740int coroutine_fn bdrv_co_delete_file(BlockDriverState *bs, Error **errp)
741{
742 Error *local_err = NULL;
743 int ret;
744
384a48fb 745 IO_CODE();
e1d7f8bb 746 assert(bs != NULL);
48aef794 747 assert_bdrv_graph_readable();
e1d7f8bb
DHB
748
749 if (!bs->drv) {
750 error_setg(errp, "Block node '%s' is not opened", bs->filename);
751 return -ENOMEDIUM;
752 }
753
754 if (!bs->drv->bdrv_co_delete_file) {
755 error_setg(errp, "Driver '%s' does not support image deletion",
756 bs->drv->format_name);
757 return -ENOTSUP;
758 }
759
760 ret = bs->drv->bdrv_co_delete_file(bs, &local_err);
761 if (ret < 0) {
762 error_propagate(errp, local_err);
763 }
764
765 return ret;
766}
767
a890f08e
ML
768void coroutine_fn bdrv_co_delete_file_noerr(BlockDriverState *bs)
769{
770 Error *local_err = NULL;
771 int ret;
384a48fb 772 IO_CODE();
a890f08e
ML
773
774 if (!bs) {
775 return;
776 }
777
778 ret = bdrv_co_delete_file(bs, &local_err);
779 /*
780 * ENOTSUP will happen if the block driver doesn't support
781 * the 'bdrv_co_delete_file' interface. This is a predictable
782 * scenario and shouldn't be reported back to the user.
783 */
784 if (ret == -ENOTSUP) {
785 error_free(local_err);
786 } else if (ret < 0) {
787 error_report_err(local_err);
788 }
789}
790
892b7de8
ET
791/**
792 * Try to get @bs's logical and physical block size.
793 * On success, store them in @bsz struct and return 0.
794 * On failure return -errno.
795 * @bs must not be empty.
796 */
797int bdrv_probe_blocksizes(BlockDriverState *bs, BlockSizes *bsz)
798{
799 BlockDriver *drv = bs->drv;
93393e69 800 BlockDriverState *filtered = bdrv_filter_bs(bs);
f791bf7f 801 GLOBAL_STATE_CODE();
892b7de8
ET
802
803 if (drv && drv->bdrv_probe_blocksizes) {
804 return drv->bdrv_probe_blocksizes(bs, bsz);
93393e69
HR
805 } else if (filtered) {
806 return bdrv_probe_blocksizes(filtered, bsz);
892b7de8
ET
807 }
808
809 return -ENOTSUP;
810}
811
812/**
813 * Try to get @bs's geometry (cyls, heads, sectors).
814 * On success, store them in @geo struct and return 0.
815 * On failure return -errno.
816 * @bs must not be empty.
817 */
818int bdrv_probe_geometry(BlockDriverState *bs, HDGeometry *geo)
819{
820 BlockDriver *drv = bs->drv;
93393e69 821 BlockDriverState *filtered = bdrv_filter_bs(bs);
f791bf7f 822 GLOBAL_STATE_CODE();
892b7de8
ET
823
824 if (drv && drv->bdrv_probe_geometry) {
825 return drv->bdrv_probe_geometry(bs, geo);
93393e69
HR
826 } else if (filtered) {
827 return bdrv_probe_geometry(filtered, geo);
892b7de8
ET
828 }
829
830 return -ENOTSUP;
831}
832
eba25057
JM
833/*
834 * Create a uniquely-named empty temporary file.
69fbfff9
BM
835 * Return the actual file name used upon success, otherwise NULL.
836 * This string should be freed with g_free() when not needed any longer.
837 *
838 * Note: creating a temporary file for the caller to (re)open is
839 * inherently racy. Use g_file_open_tmp() instead whenever practical.
eba25057 840 */
69fbfff9 841char *create_tmp_file(Error **errp)
d5249393 842{
67b915a5 843 int fd;
7ccfb2eb 844 const char *tmpdir;
69fbfff9
BM
845 g_autofree char *filename = NULL;
846
847 tmpdir = g_get_tmp_dir();
848#ifndef _WIN32
849 /*
850 * See commit 69bef79 ("block: use /var/tmp instead of /tmp for -snapshot")
851 *
852 * This function is used to create temporary disk images (like -snapshot),
853 * so the files can become very large. /tmp is often a tmpfs where as
854 * /var/tmp is usually on a disk, so more appropriate for disk images.
855 */
856 if (!g_strcmp0(tmpdir, "/tmp")) {
69bef793
AS
857 tmpdir = "/var/tmp";
858 }
69fbfff9
BM
859#endif
860
861 filename = g_strdup_printf("%s/vl.XXXXXX", tmpdir);
862 fd = g_mkstemp(filename);
fe235a06 863 if (fd < 0) {
69fbfff9
BM
864 error_setg_errno(errp, errno, "Could not open temporary file '%s'",
865 filename);
866 return NULL;
eba25057 867 }
6b6471ee 868 close(fd);
69fbfff9
BM
869
870 return g_steal_pointer(&filename);
eba25057 871}
fc01f7e7 872
84a12e66
CH
873/*
874 * Detect host devices. By convention, /dev/cdrom[N] is always
875 * recognized as a host CDROM.
876 */
877static BlockDriver *find_hdev_driver(const char *filename)
878{
879 int score_max = 0, score;
880 BlockDriver *drv = NULL, *d;
bdb73476 881 GLOBAL_STATE_CODE();
84a12e66
CH
882
883 QLIST_FOREACH(d, &bdrv_drivers, list) {
884 if (d->bdrv_probe_device) {
885 score = d->bdrv_probe_device(filename);
886 if (score > score_max) {
887 score_max = score;
888 drv = d;
889 }
890 }
891 }
892
893 return drv;
894}
895
88d88798
MM
896static BlockDriver *bdrv_do_find_protocol(const char *protocol)
897{
898 BlockDriver *drv1;
bdb73476 899 GLOBAL_STATE_CODE();
88d88798
MM
900
901 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
902 if (drv1->protocol_name && !strcmp(drv1->protocol_name, protocol)) {
903 return drv1;
904 }
905 }
906
907 return NULL;
908}
909
98289620 910BlockDriver *bdrv_find_protocol(const char *filename,
b65a5e12
HR
911 bool allow_protocol_prefix,
912 Error **errp)
83f64091
FB
913{
914 BlockDriver *drv1;
915 char protocol[128];
1cec71e3 916 int len;
83f64091 917 const char *p;
88d88798 918 int i;
19cb3738 919
f791bf7f 920 GLOBAL_STATE_CODE();
66f82cee
KW
921 /* TODO Drivers without bdrv_file_open must be specified explicitly */
922
39508e7a
CH
923 /*
924 * XXX(hch): we really should not let host device detection
925 * override an explicit protocol specification, but moving this
926 * later breaks access to device names with colons in them.
927 * Thanks to the brain-dead persistent naming schemes on udev-
928 * based Linux systems those actually are quite common.
929 */
930 drv1 = find_hdev_driver(filename);
931 if (drv1) {
932 return drv1;
933 }
934
98289620 935 if (!path_has_protocol(filename) || !allow_protocol_prefix) {
ef810437 936 return &bdrv_file;
84a12e66 937 }
98289620 938
9e0b22f4
SH
939 p = strchr(filename, ':');
940 assert(p != NULL);
1cec71e3
AL
941 len = p - filename;
942 if (len > sizeof(protocol) - 1)
943 len = sizeof(protocol) - 1;
944 memcpy(protocol, filename, len);
945 protocol[len] = '\0';
88d88798
MM
946
947 drv1 = bdrv_do_find_protocol(protocol);
948 if (drv1) {
949 return drv1;
950 }
951
952 for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); ++i) {
953 if (block_driver_modules[i].protocol_name &&
954 !strcmp(block_driver_modules[i].protocol_name, protocol)) {
c551fb0b
CF
955 int rv = block_module_load(block_driver_modules[i].library_name, errp);
956 if (rv > 0) {
957 drv1 = bdrv_do_find_protocol(protocol);
958 } else if (rv < 0) {
959 return NULL;
960 }
88d88798 961 break;
8a22f02a 962 }
83f64091 963 }
b65a5e12 964
88d88798
MM
965 if (!drv1) {
966 error_setg(errp, "Unknown protocol '%s'", protocol);
967 }
968 return drv1;
83f64091
FB
969}
970
c6684249
MA
971/*
972 * Guess image format by probing its contents.
973 * This is not a good idea when your image is raw (CVE-2008-2004), but
974 * we do it anyway for backward compatibility.
975 *
976 * @buf contains the image's first @buf_size bytes.
7cddd372
KW
977 * @buf_size is the buffer size in bytes (generally BLOCK_PROBE_BUF_SIZE,
978 * but can be smaller if the image file is smaller)
c6684249
MA
979 * @filename is its filename.
980 *
981 * For all block drivers, call the bdrv_probe() method to get its
982 * probing score.
983 * Return the first block driver with the highest probing score.
984 */
38f3ef57
KW
985BlockDriver *bdrv_probe_all(const uint8_t *buf, int buf_size,
986 const char *filename)
c6684249
MA
987{
988 int score_max = 0, score;
989 BlockDriver *drv = NULL, *d;
967d7905 990 IO_CODE();
c6684249
MA
991
992 QLIST_FOREACH(d, &bdrv_drivers, list) {
993 if (d->bdrv_probe) {
994 score = d->bdrv_probe(buf, buf_size, filename);
995 if (score > score_max) {
996 score_max = score;
997 drv = d;
998 }
999 }
1000 }
1001
1002 return drv;
1003}
1004
5696c6e3 1005static int find_image_format(BlockBackend *file, const char *filename,
34b5d2c6 1006 BlockDriver **pdrv, Error **errp)
f3a5d3f8 1007{
c6684249 1008 BlockDriver *drv;
7cddd372 1009 uint8_t buf[BLOCK_PROBE_BUF_SIZE];
f500a6d3 1010 int ret = 0;
f8ea0b00 1011
bdb73476
EGE
1012 GLOBAL_STATE_CODE();
1013
08a00559 1014 /* Return the raw BlockDriver * to scsi-generic devices or empty drives */
5696c6e3 1015 if (blk_is_sg(file) || !blk_is_inserted(file) || blk_getlength(file) == 0) {
ef810437 1016 *pdrv = &bdrv_raw;
c98ac35d 1017 return ret;
1a396859 1018 }
f8ea0b00 1019
a9262f55 1020 ret = blk_pread(file, 0, sizeof(buf), buf, 0);
83f64091 1021 if (ret < 0) {
34b5d2c6
HR
1022 error_setg_errno(errp, -ret, "Could not read image for determining its "
1023 "format");
c98ac35d
SW
1024 *pdrv = NULL;
1025 return ret;
83f64091
FB
1026 }
1027
bf5b16fa 1028 drv = bdrv_probe_all(buf, sizeof(buf), filename);
c98ac35d 1029 if (!drv) {
34b5d2c6
HR
1030 error_setg(errp, "Could not determine image format: No compatible "
1031 "driver found");
bf5b16fa
AF
1032 *pdrv = NULL;
1033 return -ENOENT;
c98ac35d 1034 }
bf5b16fa 1035
c98ac35d 1036 *pdrv = drv;
bf5b16fa 1037 return 0;
ea2384d3
FB
1038}
1039
51762288
SH
1040/**
1041 * Set the current 'total_sectors' value
65a9bb25 1042 * Return 0 on success, -errno on error.
51762288 1043 */
c86422c5
EGE
1044int coroutine_fn bdrv_co_refresh_total_sectors(BlockDriverState *bs,
1045 int64_t hint)
51762288
SH
1046{
1047 BlockDriver *drv = bs->drv;
967d7905 1048 IO_CODE();
8ab8140a 1049 assert_bdrv_graph_readable();
51762288 1050
d470ad42
HR
1051 if (!drv) {
1052 return -ENOMEDIUM;
1053 }
1054
c86422c5 1055 /* Do not attempt drv->bdrv_co_getlength() on scsi-generic devices */
b192af8a 1056 if (bdrv_is_sg(bs))
396759ad
NB
1057 return 0;
1058
51762288 1059 /* query actual device if possible, otherwise just trust the hint */
c86422c5
EGE
1060 if (drv->bdrv_co_getlength) {
1061 int64_t length = drv->bdrv_co_getlength(bs);
51762288
SH
1062 if (length < 0) {
1063 return length;
1064 }
7e382003 1065 hint = DIV_ROUND_UP(length, BDRV_SECTOR_SIZE);
51762288
SH
1066 }
1067
1068 bs->total_sectors = hint;
8b117001
VSO
1069
1070 if (bs->total_sectors * BDRV_SECTOR_SIZE > BDRV_MAX_LENGTH) {
1071 return -EFBIG;
1072 }
1073
51762288
SH
1074 return 0;
1075}
1076
cddff5ba
KW
1077/**
1078 * Combines a QDict of new block driver @options with any missing options taken
1079 * from @old_options, so that leaving out an option defaults to its old value.
1080 */
1081static void bdrv_join_options(BlockDriverState *bs, QDict *options,
1082 QDict *old_options)
1083{
da359909 1084 GLOBAL_STATE_CODE();
cddff5ba
KW
1085 if (bs->drv && bs->drv->bdrv_join_options) {
1086 bs->drv->bdrv_join_options(options, old_options);
1087 } else {
1088 qdict_join(options, old_options, false);
1089 }
1090}
1091
543770bd
AG
1092static BlockdevDetectZeroesOptions bdrv_parse_detect_zeroes(QemuOpts *opts,
1093 int open_flags,
1094 Error **errp)
1095{
1096 Error *local_err = NULL;
1097 char *value = qemu_opt_get_del(opts, "detect-zeroes");
1098 BlockdevDetectZeroesOptions detect_zeroes =
1099 qapi_enum_parse(&BlockdevDetectZeroesOptions_lookup, value,
1100 BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF, &local_err);
bdb73476 1101 GLOBAL_STATE_CODE();
543770bd
AG
1102 g_free(value);
1103 if (local_err) {
1104 error_propagate(errp, local_err);
1105 return detect_zeroes;
1106 }
1107
1108 if (detect_zeroes == BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP &&
1109 !(open_flags & BDRV_O_UNMAP))
1110 {
1111 error_setg(errp, "setting detect-zeroes to unmap is not allowed "
1112 "without setting discard operation to unmap");
1113 }
1114
1115 return detect_zeroes;
1116}
1117
f80f2673
AM
1118/**
1119 * Set open flags for aio engine
1120 *
1121 * Return 0 on success, -1 if the engine specified is invalid
1122 */
1123int bdrv_parse_aio(const char *mode, int *flags)
1124{
1125 if (!strcmp(mode, "threads")) {
1126 /* do nothing, default */
1127 } else if (!strcmp(mode, "native")) {
1128 *flags |= BDRV_O_NATIVE_AIO;
1129#ifdef CONFIG_LINUX_IO_URING
1130 } else if (!strcmp(mode, "io_uring")) {
1131 *flags |= BDRV_O_IO_URING;
1132#endif
1133 } else {
1134 return -1;
1135 }
1136
1137 return 0;
1138}
1139
9e8f1835
PB
1140/**
1141 * Set open flags for a given discard mode
1142 *
1143 * Return 0 on success, -1 if the discard mode was invalid.
1144 */
1145int bdrv_parse_discard_flags(const char *mode, int *flags)
1146{
1147 *flags &= ~BDRV_O_UNMAP;
1148
1149 if (!strcmp(mode, "off") || !strcmp(mode, "ignore")) {
1150 /* do nothing */
1151 } else if (!strcmp(mode, "on") || !strcmp(mode, "unmap")) {
1152 *flags |= BDRV_O_UNMAP;
1153 } else {
1154 return -1;
1155 }
1156
1157 return 0;
1158}
1159
c3993cdc
SH
1160/**
1161 * Set open flags for a given cache mode
1162 *
1163 * Return 0 on success, -1 if the cache mode was invalid.
1164 */
53e8ae01 1165int bdrv_parse_cache_mode(const char *mode, int *flags, bool *writethrough)
c3993cdc
SH
1166{
1167 *flags &= ~BDRV_O_CACHE_MASK;
1168
1169 if (!strcmp(mode, "off") || !strcmp(mode, "none")) {
53e8ae01
KW
1170 *writethrough = false;
1171 *flags |= BDRV_O_NOCACHE;
92196b2f 1172 } else if (!strcmp(mode, "directsync")) {
53e8ae01 1173 *writethrough = true;
92196b2f 1174 *flags |= BDRV_O_NOCACHE;
c3993cdc 1175 } else if (!strcmp(mode, "writeback")) {
53e8ae01 1176 *writethrough = false;
c3993cdc 1177 } else if (!strcmp(mode, "unsafe")) {
53e8ae01 1178 *writethrough = false;
c3993cdc
SH
1179 *flags |= BDRV_O_NO_FLUSH;
1180 } else if (!strcmp(mode, "writethrough")) {
53e8ae01 1181 *writethrough = true;
c3993cdc
SH
1182 } else {
1183 return -1;
1184 }
1185
1186 return 0;
1187}
1188
b5411555
KW
1189static char *bdrv_child_get_parent_desc(BdrvChild *c)
1190{
1191 BlockDriverState *parent = c->opaque;
2c0a3acb 1192 return g_strdup_printf("node '%s'", bdrv_get_node_name(parent));
b5411555
KW
1193}
1194
20018e12
KW
1195static void bdrv_child_cb_drained_begin(BdrvChild *child)
1196{
1197 BlockDriverState *bs = child->opaque;
a82a3bd1 1198 bdrv_do_drained_begin_quiesce(bs, NULL);
20018e12
KW
1199}
1200
89bd0305
KW
1201static bool bdrv_child_cb_drained_poll(BdrvChild *child)
1202{
1203 BlockDriverState *bs = child->opaque;
299403ae 1204 return bdrv_drain_poll(bs, NULL, false);
89bd0305
KW
1205}
1206
2f65df6e 1207static void bdrv_child_cb_drained_end(BdrvChild *child)
20018e12
KW
1208{
1209 BlockDriverState *bs = child->opaque;
2f65df6e 1210 bdrv_drained_end(bs);
20018e12
KW
1211}
1212
38701b6a
KW
1213static int bdrv_child_cb_inactivate(BdrvChild *child)
1214{
1215 BlockDriverState *bs = child->opaque;
bdb73476 1216 GLOBAL_STATE_CODE();
38701b6a
KW
1217 assert(bs->open_flags & BDRV_O_INACTIVE);
1218 return 0;
1219}
1220
27633e74
EGE
1221static bool bdrv_child_cb_change_aio_ctx(BdrvChild *child, AioContext *ctx,
1222 GHashTable *visited, Transaction *tran,
1223 Error **errp)
53a7d041
KW
1224{
1225 BlockDriverState *bs = child->opaque;
27633e74 1226 return bdrv_change_aio_context(bs, ctx, visited, tran, errp);
53a7d041
KW
1227}
1228
b1e6fc08 1229/*
73176bee
KW
1230 * Returns the options and flags that a temporary snapshot should get, based on
1231 * the originally requested flags (the originally requested image will have
1232 * flags like a backing file)
b1e6fc08 1233 */
73176bee
KW
1234static void bdrv_temp_snapshot_options(int *child_flags, QDict *child_options,
1235 int parent_flags, QDict *parent_options)
b1e6fc08 1236{
bdb73476 1237 GLOBAL_STATE_CODE();
73176bee
KW
1238 *child_flags = (parent_flags & ~BDRV_O_SNAPSHOT) | BDRV_O_TEMPORARY;
1239
1240 /* For temporary files, unconditional cache=unsafe is fine */
73176bee
KW
1241 qdict_set_default_str(child_options, BDRV_OPT_CACHE_DIRECT, "off");
1242 qdict_set_default_str(child_options, BDRV_OPT_CACHE_NO_FLUSH, "on");
41869044 1243
3f48686f 1244 /* Copy the read-only and discard options from the parent */
f87a0e29 1245 qdict_copy_default(child_options, parent_options, BDRV_OPT_READ_ONLY);
3f48686f 1246 qdict_copy_default(child_options, parent_options, BDRV_OPT_DISCARD);
f87a0e29 1247
41869044
KW
1248 /* aio=native doesn't work for cache.direct=off, so disable it for the
1249 * temporary snapshot */
1250 *child_flags &= ~BDRV_O_NATIVE_AIO;
b1e6fc08
KW
1251}
1252
db95dbba
KW
1253static void bdrv_backing_attach(BdrvChild *c)
1254{
1255 BlockDriverState *parent = c->opaque;
1256 BlockDriverState *backing_hd = c->bs;
1257
bdb73476 1258 GLOBAL_STATE_CODE();
db95dbba
KW
1259 assert(!parent->backing_blocker);
1260 error_setg(&parent->backing_blocker,
1261 "node is used as backing hd of '%s'",
1262 bdrv_get_device_or_node_name(parent));
1263
f30c66ba
HR
1264 bdrv_refresh_filename(backing_hd);
1265
db95dbba 1266 parent->open_flags &= ~BDRV_O_NO_BACKING;
db95dbba
KW
1267
1268 bdrv_op_block_all(backing_hd, parent->backing_blocker);
1269 /* Otherwise we won't be able to commit or stream */
1270 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_COMMIT_TARGET,
1271 parent->backing_blocker);
1272 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_STREAM,
1273 parent->backing_blocker);
1274 /*
1275 * We do backup in 3 ways:
1276 * 1. drive backup
1277 * The target bs is new opened, and the source is top BDS
1278 * 2. blockdev backup
1279 * Both the source and the target are top BDSes.
1280 * 3. internal backup(used for block replication)
1281 * Both the source and the target are backing file
1282 *
1283 * In case 1 and 2, neither the source nor the target is the backing file.
1284 * In case 3, we will block the top BDS, so there is only one block job
1285 * for the top BDS and its backing chain.
1286 */
1287 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_BACKUP_SOURCE,
1288 parent->backing_blocker);
1289 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_BACKUP_TARGET,
1290 parent->backing_blocker);
ca2f1234 1291}
d736f119 1292
db95dbba
KW
1293static void bdrv_backing_detach(BdrvChild *c)
1294{
1295 BlockDriverState *parent = c->opaque;
1296
bdb73476 1297 GLOBAL_STATE_CODE();
db95dbba
KW
1298 assert(parent->backing_blocker);
1299 bdrv_op_unblock_all(c->bs, parent->backing_blocker);
1300 error_free(parent->backing_blocker);
1301 parent->backing_blocker = NULL;
48e08288 1302}
d736f119 1303
6858eba0
KW
1304static int bdrv_backing_update_filename(BdrvChild *c, BlockDriverState *base,
1305 const char *filename, Error **errp)
1306{
1307 BlockDriverState *parent = c->opaque;
e94d3dba 1308 bool read_only = bdrv_is_read_only(parent);
6858eba0 1309 int ret;
bdb73476 1310 GLOBAL_STATE_CODE();
6858eba0 1311
e94d3dba
AG
1312 if (read_only) {
1313 ret = bdrv_reopen_set_read_only(parent, false, errp);
61f09cea
KW
1314 if (ret < 0) {
1315 return ret;
1316 }
1317 }
1318
6858eba0 1319 ret = bdrv_change_backing_file(parent, filename,
e54ee1b3
EB
1320 base->drv ? base->drv->format_name : "",
1321 false);
6858eba0 1322 if (ret < 0) {
64730694 1323 error_setg_errno(errp, -ret, "Could not update backing file link");
6858eba0
KW
1324 }
1325
e94d3dba
AG
1326 if (read_only) {
1327 bdrv_reopen_set_read_only(parent, true, NULL);
61f09cea
KW
1328 }
1329
6858eba0
KW
1330 return ret;
1331}
1332
fae8bd39
HR
1333/*
1334 * Returns the options and flags that a generic child of a BDS should
1335 * get, based on the given options and flags for the parent BDS.
1336 */
00ff7ffd
HR
1337static void bdrv_inherited_options(BdrvChildRole role, bool parent_is_format,
1338 int *child_flags, QDict *child_options,
1339 int parent_flags, QDict *parent_options)
fae8bd39
HR
1340{
1341 int flags = parent_flags;
bdb73476 1342 GLOBAL_STATE_CODE();
fae8bd39
HR
1343
1344 /*
1345 * First, decide whether to set, clear, or leave BDRV_O_PROTOCOL.
1346 * Generally, the question to answer is: Should this child be
1347 * format-probed by default?
1348 */
1349
1350 /*
1351 * Pure and non-filtered data children of non-format nodes should
1352 * be probed by default (even when the node itself has BDRV_O_PROTOCOL
1353 * set). This only affects a very limited set of drivers (namely
1354 * quorum and blkverify when this comment was written).
1355 * Force-clear BDRV_O_PROTOCOL then.
1356 */
1357 if (!parent_is_format &&
1358 (role & BDRV_CHILD_DATA) &&
1359 !(role & (BDRV_CHILD_METADATA | BDRV_CHILD_FILTERED)))
1360 {
1361 flags &= ~BDRV_O_PROTOCOL;
1362 }
1363
1364 /*
1365 * All children of format nodes (except for COW children) and all
1366 * metadata children in general should never be format-probed.
1367 * Force-set BDRV_O_PROTOCOL then.
1368 */
1369 if ((parent_is_format && !(role & BDRV_CHILD_COW)) ||
1370 (role & BDRV_CHILD_METADATA))
1371 {
1372 flags |= BDRV_O_PROTOCOL;
1373 }
1374
1375 /*
1376 * If the cache mode isn't explicitly set, inherit direct and no-flush from
1377 * the parent.
1378 */
1379 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_DIRECT);
1380 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_NO_FLUSH);
1381 qdict_copy_default(child_options, parent_options, BDRV_OPT_FORCE_SHARE);
1382
1383 if (role & BDRV_CHILD_COW) {
1384 /* backing files are opened read-only by default */
1385 qdict_set_default_str(child_options, BDRV_OPT_READ_ONLY, "on");
1386 qdict_set_default_str(child_options, BDRV_OPT_AUTO_READ_ONLY, "off");
1387 } else {
1388 /* Inherit the read-only option from the parent if it's not set */
1389 qdict_copy_default(child_options, parent_options, BDRV_OPT_READ_ONLY);
1390 qdict_copy_default(child_options, parent_options,
1391 BDRV_OPT_AUTO_READ_ONLY);
1392 }
1393
1394 /*
1395 * bdrv_co_pdiscard() respects unmap policy for the parent, so we
1396 * can default to enable it on lower layers regardless of the
1397 * parent option.
1398 */
1399 qdict_set_default_str(child_options, BDRV_OPT_DISCARD, "unmap");
1400
1401 /* Clear flags that only apply to the top layer */
1402 flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_COPY_ON_READ);
1403
1404 if (role & BDRV_CHILD_METADATA) {
1405 flags &= ~BDRV_O_NO_IO;
1406 }
1407 if (role & BDRV_CHILD_COW) {
1408 flags &= ~BDRV_O_TEMPORARY;
1409 }
1410
1411 *child_flags = flags;
1412}
1413
303de47b 1414static void GRAPH_WRLOCK bdrv_child_cb_attach(BdrvChild *child)
ca2f1234
HR
1415{
1416 BlockDriverState *bs = child->opaque;
1417
3f35f82e 1418 assert_bdrv_graph_writable();
a225369b 1419 QLIST_INSERT_HEAD(&bs->children, child, next);
5bb04747
VSO
1420 if (bs->drv->is_filter || (child->role & BDRV_CHILD_FILTERED)) {
1421 /*
1422 * Here we handle filters and block/raw-format.c when it behave like
1423 * filter. They generally have a single PRIMARY child, which is also the
1424 * FILTERED child, and that they may have multiple more children, which
1425 * are neither PRIMARY nor FILTERED. And never we have a COW child here.
1426 * So bs->file will be the PRIMARY child, unless the PRIMARY child goes
1427 * into bs->backing on exceptional cases; and bs->backing will be
1428 * nothing else.
1429 */
1430 assert(!(child->role & BDRV_CHILD_COW));
1431 if (child->role & BDRV_CHILD_PRIMARY) {
1432 assert(child->role & BDRV_CHILD_FILTERED);
1433 assert(!bs->backing);
1434 assert(!bs->file);
1435
1436 if (bs->drv->filtered_child_is_backing) {
1437 bs->backing = child;
1438 } else {
1439 bs->file = child;
1440 }
1441 } else {
1442 assert(!(child->role & BDRV_CHILD_FILTERED));
1443 }
1444 } else if (child->role & BDRV_CHILD_COW) {
1445 assert(bs->drv->supports_backing);
1446 assert(!(child->role & BDRV_CHILD_PRIMARY));
1447 assert(!bs->backing);
1448 bs->backing = child;
ca2f1234 1449 bdrv_backing_attach(child);
5bb04747
VSO
1450 } else if (child->role & BDRV_CHILD_PRIMARY) {
1451 assert(!bs->file);
1452 bs->file = child;
ca2f1234 1453 }
ca2f1234
HR
1454}
1455
303de47b 1456static void GRAPH_WRLOCK bdrv_child_cb_detach(BdrvChild *child)
48e08288
HR
1457{
1458 BlockDriverState *bs = child->opaque;
1459
1460 if (child->role & BDRV_CHILD_COW) {
1461 bdrv_backing_detach(child);
1462 }
1463
3f35f82e 1464 assert_bdrv_graph_writable();
a225369b 1465 QLIST_REMOVE(child, next);
5bb04747
VSO
1466 if (child == bs->backing) {
1467 assert(child != bs->file);
1468 bs->backing = NULL;
1469 } else if (child == bs->file) {
1470 bs->file = NULL;
1471 }
48e08288
HR
1472}
1473
43483550
HR
1474static int bdrv_child_cb_update_filename(BdrvChild *c, BlockDriverState *base,
1475 const char *filename, Error **errp)
1476{
1477 if (c->role & BDRV_CHILD_COW) {
1478 return bdrv_backing_update_filename(c, base, filename, errp);
1479 }
1480 return 0;
1481}
1482
fb62b588 1483AioContext *child_of_bds_get_parent_aio_context(BdrvChild *c)
3ca1f322
VSO
1484{
1485 BlockDriverState *bs = c->opaque;
384a48fb 1486 IO_CODE();
3ca1f322
VSO
1487
1488 return bdrv_get_aio_context(bs);
1489}
1490
43483550
HR
1491const BdrvChildClass child_of_bds = {
1492 .parent_is_bds = true,
1493 .get_parent_desc = bdrv_child_get_parent_desc,
1494 .inherit_options = bdrv_inherited_options,
1495 .drained_begin = bdrv_child_cb_drained_begin,
1496 .drained_poll = bdrv_child_cb_drained_poll,
1497 .drained_end = bdrv_child_cb_drained_end,
1498 .attach = bdrv_child_cb_attach,
1499 .detach = bdrv_child_cb_detach,
1500 .inactivate = bdrv_child_cb_inactivate,
27633e74 1501 .change_aio_ctx = bdrv_child_cb_change_aio_ctx,
43483550 1502 .update_filename = bdrv_child_cb_update_filename,
fb62b588 1503 .get_parent_aio_context = child_of_bds_get_parent_aio_context,
43483550
HR
1504};
1505
3ca1f322
VSO
1506AioContext *bdrv_child_get_parent_aio_context(BdrvChild *c)
1507{
d5f8d79c 1508 IO_CODE();
3ca1f322
VSO
1509 return c->klass->get_parent_aio_context(c);
1510}
1511
7b272452
KW
1512static int bdrv_open_flags(BlockDriverState *bs, int flags)
1513{
61de4c68 1514 int open_flags = flags;
bdb73476 1515 GLOBAL_STATE_CODE();
7b272452
KW
1516
1517 /*
1518 * Clear flags that are internal to the block layer before opening the
1519 * image.
1520 */
20cca275 1521 open_flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_PROTOCOL);
7b272452 1522
7b272452
KW
1523 return open_flags;
1524}
1525
91a097e7
KW
1526static void update_flags_from_options(int *flags, QemuOpts *opts)
1527{
bdb73476
EGE
1528 GLOBAL_STATE_CODE();
1529
2a3d4331 1530 *flags &= ~(BDRV_O_CACHE_MASK | BDRV_O_RDWR | BDRV_O_AUTO_RDONLY);
91a097e7 1531
57f9db9a 1532 if (qemu_opt_get_bool_del(opts, BDRV_OPT_CACHE_NO_FLUSH, false)) {
91a097e7
KW
1533 *flags |= BDRV_O_NO_FLUSH;
1534 }
1535
57f9db9a 1536 if (qemu_opt_get_bool_del(opts, BDRV_OPT_CACHE_DIRECT, false)) {
91a097e7
KW
1537 *flags |= BDRV_O_NOCACHE;
1538 }
f87a0e29 1539
57f9db9a 1540 if (!qemu_opt_get_bool_del(opts, BDRV_OPT_READ_ONLY, false)) {
f87a0e29
AG
1541 *flags |= BDRV_O_RDWR;
1542 }
1543
e35bdc12
KW
1544 if (qemu_opt_get_bool_del(opts, BDRV_OPT_AUTO_READ_ONLY, false)) {
1545 *flags |= BDRV_O_AUTO_RDONLY;
1546 }
91a097e7
KW
1547}
1548
1549static void update_options_from_flags(QDict *options, int flags)
1550{
bdb73476 1551 GLOBAL_STATE_CODE();
91a097e7 1552 if (!qdict_haskey(options, BDRV_OPT_CACHE_DIRECT)) {
46f5ac20 1553 qdict_put_bool(options, BDRV_OPT_CACHE_DIRECT, flags & BDRV_O_NOCACHE);
91a097e7
KW
1554 }
1555 if (!qdict_haskey(options, BDRV_OPT_CACHE_NO_FLUSH)) {
46f5ac20
EB
1556 qdict_put_bool(options, BDRV_OPT_CACHE_NO_FLUSH,
1557 flags & BDRV_O_NO_FLUSH);
91a097e7 1558 }
f87a0e29 1559 if (!qdict_haskey(options, BDRV_OPT_READ_ONLY)) {
46f5ac20 1560 qdict_put_bool(options, BDRV_OPT_READ_ONLY, !(flags & BDRV_O_RDWR));
f87a0e29 1561 }
e35bdc12
KW
1562 if (!qdict_haskey(options, BDRV_OPT_AUTO_READ_ONLY)) {
1563 qdict_put_bool(options, BDRV_OPT_AUTO_READ_ONLY,
1564 flags & BDRV_O_AUTO_RDONLY);
1565 }
91a097e7
KW
1566}
1567
636ea370
KW
1568static void bdrv_assign_node_name(BlockDriverState *bs,
1569 const char *node_name,
1570 Error **errp)
6913c0c2 1571{
15489c76 1572 char *gen_node_name = NULL;
bdb73476 1573 GLOBAL_STATE_CODE();
6913c0c2 1574
15489c76
JC
1575 if (!node_name) {
1576 node_name = gen_node_name = id_generate(ID_BLOCK);
1577 } else if (!id_wellformed(node_name)) {
1578 /*
1579 * Check for empty string or invalid characters, but not if it is
1580 * generated (generated names use characters not available to the user)
1581 */
785ec4b1 1582 error_setg(errp, "Invalid node-name: '%s'", node_name);
636ea370 1583 return;
6913c0c2
BC
1584 }
1585
0c5e94ee 1586 /* takes care of avoiding namespaces collisions */
7f06d47e 1587 if (blk_by_name(node_name)) {
0c5e94ee
BC
1588 error_setg(errp, "node-name=%s is conflicting with a device id",
1589 node_name);
15489c76 1590 goto out;
0c5e94ee
BC
1591 }
1592
6913c0c2
BC
1593 /* takes care of avoiding duplicates node names */
1594 if (bdrv_find_node(node_name)) {
785ec4b1 1595 error_setg(errp, "Duplicate nodes with node-name='%s'", node_name);
15489c76 1596 goto out;
6913c0c2
BC
1597 }
1598
824808dd
KW
1599 /* Make sure that the node name isn't truncated */
1600 if (strlen(node_name) >= sizeof(bs->node_name)) {
1601 error_setg(errp, "Node name too long");
1602 goto out;
1603 }
1604
6913c0c2
BC
1605 /* copy node name into the bs and insert it into the graph list */
1606 pstrcpy(bs->node_name, sizeof(bs->node_name), node_name);
1607 QTAILQ_INSERT_TAIL(&graph_bdrv_states, bs, node_list);
15489c76
JC
1608out:
1609 g_free(gen_node_name);
6913c0c2
BC
1610}
1611
c86422c5
EGE
1612/*
1613 * The caller must always hold @bs AioContext lock, because this function calls
1614 * bdrv_refresh_total_sectors() which polls when called from non-coroutine
1615 * context.
1616 */
1a30b0f5
KW
1617static int no_coroutine_fn GRAPH_UNLOCKED
1618bdrv_open_driver(BlockDriverState *bs, BlockDriver *drv, const char *node_name,
1619 QDict *options, int open_flags, Error **errp)
01a56501 1620{
8dc8a60c 1621 AioContext *ctx;
01a56501 1622 Error *local_err = NULL;
0f12264e 1623 int i, ret;
da359909 1624 GLOBAL_STATE_CODE();
01a56501
KW
1625
1626 bdrv_assign_node_name(bs, node_name, &local_err);
1627 if (local_err) {
1628 error_propagate(errp, local_err);
1629 return -EINVAL;
1630 }
1631
1632 bs->drv = drv;
1633 bs->opaque = g_malloc0(drv->instance_size);
1634
1635 if (drv->bdrv_file_open) {
1636 assert(!drv->bdrv_needs_filename || bs->filename[0]);
1637 ret = drv->bdrv_file_open(bs, options, open_flags, &local_err);
680c7f96 1638 } else if (drv->bdrv_open) {
01a56501 1639 ret = drv->bdrv_open(bs, options, open_flags, &local_err);
680c7f96
KW
1640 } else {
1641 ret = 0;
01a56501
KW
1642 }
1643
1644 if (ret < 0) {
1645 if (local_err) {
1646 error_propagate(errp, local_err);
1647 } else if (bs->filename[0]) {
1648 error_setg_errno(errp, -ret, "Could not open '%s'", bs->filename);
1649 } else {
1650 error_setg_errno(errp, -ret, "Could not open image");
1651 }
180ca19a 1652 goto open_failed;
01a56501
KW
1653 }
1654
e8b65355
SH
1655 assert(!(bs->supported_read_flags & ~BDRV_REQ_MASK));
1656 assert(!(bs->supported_write_flags & ~BDRV_REQ_MASK));
1657
1658 /*
1659 * Always allow the BDRV_REQ_REGISTERED_BUF optimization hint. This saves
1660 * drivers that pass read/write requests through to a child the trouble of
1661 * declaring support explicitly.
1662 *
1663 * Drivers must not propagate this flag accidentally when they initiate I/O
1664 * to a bounce buffer. That case should be rare though.
1665 */
1666 bs->supported_read_flags |= BDRV_REQ_REGISTERED_BUF;
1667 bs->supported_write_flags |= BDRV_REQ_REGISTERED_BUF;
1668
8dc8a60c
KW
1669 /* Get the context after .bdrv_open, it can change the context */
1670 ctx = bdrv_get_aio_context(bs);
1671 aio_context_acquire(ctx);
1672
c057960c 1673 ret = bdrv_refresh_total_sectors(bs, bs->total_sectors);
01a56501
KW
1674 if (ret < 0) {
1675 error_setg_errno(errp, -ret, "Could not refresh total sector count");
8dc8a60c 1676 aio_context_release(ctx);
180ca19a 1677 return ret;
01a56501
KW
1678 }
1679
e19b157f 1680 bdrv_graph_rdlock_main_loop();
1e4c797c 1681 bdrv_refresh_limits(bs, NULL, &local_err);
e19b157f 1682 bdrv_graph_rdunlock_main_loop();
8dc8a60c 1683 aio_context_release(ctx);
e19b157f 1684
01a56501
KW
1685 if (local_err) {
1686 error_propagate(errp, local_err);
180ca19a 1687 return -EINVAL;
01a56501
KW
1688 }
1689
1690 assert(bdrv_opt_mem_align(bs) != 0);
1691 assert(bdrv_min_mem_align(bs) != 0);
1692 assert(is_power_of_2(bs->bl.request_alignment));
1693
0f12264e 1694 for (i = 0; i < bs->quiesce_counter; i++) {
5e8ac217
KW
1695 if (drv->bdrv_drain_begin) {
1696 drv->bdrv_drain_begin(bs);
0f12264e
KW
1697 }
1698 }
1699
01a56501 1700 return 0;
180ca19a
MP
1701open_failed:
1702 bs->drv = NULL;
1703 if (bs->file != NULL) {
32a8aba3 1704 bdrv_graph_wrlock(NULL);
180ca19a 1705 bdrv_unref_child(bs, bs->file);
32a8aba3 1706 bdrv_graph_wrunlock();
5bb04747 1707 assert(!bs->file);
180ca19a 1708 }
01a56501
KW
1709 g_free(bs->opaque);
1710 bs->opaque = NULL;
01a56501
KW
1711 return ret;
1712}
1713
621d1737
VSO
1714/*
1715 * Create and open a block node.
1716 *
1717 * @options is a QDict of options to pass to the block drivers, or NULL for an
1718 * empty set of options. The reference to the QDict belongs to the block layer
1719 * after the call (even on failure), so if the caller intends to reuse the
1720 * dictionary, it needs to use qobject_ref() before calling bdrv_open.
1721 */
1722BlockDriverState *bdrv_new_open_driver_opts(BlockDriver *drv,
1723 const char *node_name,
1724 QDict *options, int flags,
1725 Error **errp)
680c7f96
KW
1726{
1727 BlockDriverState *bs;
1728 int ret;
1729
f791bf7f
EGE
1730 GLOBAL_STATE_CODE();
1731
680c7f96
KW
1732 bs = bdrv_new();
1733 bs->open_flags = flags;
621d1737
VSO
1734 bs->options = options ?: qdict_new();
1735 bs->explicit_options = qdict_clone_shallow(bs->options);
680c7f96
KW
1736 bs->opaque = NULL;
1737
1738 update_options_from_flags(bs->options, flags);
1739
1740 ret = bdrv_open_driver(bs, drv, node_name, bs->options, flags, errp);
1741 if (ret < 0) {
cb3e7f08 1742 qobject_unref(bs->explicit_options);
180ca19a 1743 bs->explicit_options = NULL;
cb3e7f08 1744 qobject_unref(bs->options);
180ca19a 1745 bs->options = NULL;
680c7f96
KW
1746 bdrv_unref(bs);
1747 return NULL;
1748 }
1749
1750 return bs;
1751}
1752
621d1737
VSO
1753/* Create and open a block node. */
1754BlockDriverState *bdrv_new_open_driver(BlockDriver *drv, const char *node_name,
1755 int flags, Error **errp)
1756{
f791bf7f 1757 GLOBAL_STATE_CODE();
621d1737
VSO
1758 return bdrv_new_open_driver_opts(drv, node_name, NULL, flags, errp);
1759}
1760
c5f3014b 1761QemuOptsList bdrv_runtime_opts = {
18edf289
KW
1762 .name = "bdrv_common",
1763 .head = QTAILQ_HEAD_INITIALIZER(bdrv_runtime_opts.head),
1764 .desc = {
1765 {
1766 .name = "node-name",
1767 .type = QEMU_OPT_STRING,
1768 .help = "Node name of the block device node",
1769 },
62392ebb
KW
1770 {
1771 .name = "driver",
1772 .type = QEMU_OPT_STRING,
1773 .help = "Block driver to use for the node",
1774 },
91a097e7
KW
1775 {
1776 .name = BDRV_OPT_CACHE_DIRECT,
1777 .type = QEMU_OPT_BOOL,
1778 .help = "Bypass software writeback cache on the host",
1779 },
1780 {
1781 .name = BDRV_OPT_CACHE_NO_FLUSH,
1782 .type = QEMU_OPT_BOOL,
1783 .help = "Ignore flush requests",
1784 },
f87a0e29
AG
1785 {
1786 .name = BDRV_OPT_READ_ONLY,
1787 .type = QEMU_OPT_BOOL,
1788 .help = "Node is opened in read-only mode",
1789 },
e35bdc12
KW
1790 {
1791 .name = BDRV_OPT_AUTO_READ_ONLY,
1792 .type = QEMU_OPT_BOOL,
1793 .help = "Node can become read-only if opening read-write fails",
1794 },
692e01a2
KW
1795 {
1796 .name = "detect-zeroes",
1797 .type = QEMU_OPT_STRING,
1798 .help = "try to optimize zero writes (off, on, unmap)",
1799 },
818584a4 1800 {
415bbca8 1801 .name = BDRV_OPT_DISCARD,
818584a4
KW
1802 .type = QEMU_OPT_STRING,
1803 .help = "discard operation (ignore/off, unmap/on)",
1804 },
5a9347c6
FZ
1805 {
1806 .name = BDRV_OPT_FORCE_SHARE,
1807 .type = QEMU_OPT_BOOL,
1808 .help = "always accept other writers (default: off)",
1809 },
18edf289
KW
1810 { /* end of list */ }
1811 },
1812};
1813
5a5e7f8c
ML
1814QemuOptsList bdrv_create_opts_simple = {
1815 .name = "simple-create-opts",
1816 .head = QTAILQ_HEAD_INITIALIZER(bdrv_create_opts_simple.head),
fd17146c
HR
1817 .desc = {
1818 {
1819 .name = BLOCK_OPT_SIZE,
1820 .type = QEMU_OPT_SIZE,
1821 .help = "Virtual disk size"
1822 },
1823 {
1824 .name = BLOCK_OPT_PREALLOC,
1825 .type = QEMU_OPT_STRING,
1826 .help = "Preallocation mode (allowed values: off)"
1827 },
1828 { /* end of list */ }
1829 }
1830};
1831
57915332
KW
1832/*
1833 * Common part for opening disk images and files
b6ad491a
KW
1834 *
1835 * Removes all processed options from *options.
57915332 1836 */
5696c6e3 1837static int bdrv_open_common(BlockDriverState *bs, BlockBackend *file,
82dc8b41 1838 QDict *options, Error **errp)
57915332
KW
1839{
1840 int ret, open_flags;
035fccdf 1841 const char *filename;
62392ebb 1842 const char *driver_name = NULL;
6913c0c2 1843 const char *node_name = NULL;
818584a4 1844 const char *discard;
18edf289 1845 QemuOpts *opts;
62392ebb 1846 BlockDriver *drv;
34b5d2c6 1847 Error *local_err = NULL;
307261b2 1848 bool ro;
57915332 1849
6405875c 1850 assert(bs->file == NULL);
707ff828 1851 assert(options != NULL && bs->options != options);
bdb73476 1852 GLOBAL_STATE_CODE();
57915332 1853
62392ebb 1854 opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
af175e85 1855 if (!qemu_opts_absorb_qdict(opts, options, errp)) {
62392ebb
KW
1856 ret = -EINVAL;
1857 goto fail_opts;
1858 }
1859
9b7e8691
AG
1860 update_flags_from_options(&bs->open_flags, opts);
1861
62392ebb
KW
1862 driver_name = qemu_opt_get(opts, "driver");
1863 drv = bdrv_find_format(driver_name);
1864 assert(drv != NULL);
1865
5a9347c6
FZ
1866 bs->force_share = qemu_opt_get_bool(opts, BDRV_OPT_FORCE_SHARE, false);
1867
1868 if (bs->force_share && (bs->open_flags & BDRV_O_RDWR)) {
1869 error_setg(errp,
1870 BDRV_OPT_FORCE_SHARE
1871 "=on can only be used with read-only images");
1872 ret = -EINVAL;
1873 goto fail_opts;
1874 }
1875
45673671 1876 if (file != NULL) {
f30c66ba 1877 bdrv_refresh_filename(blk_bs(file));
5696c6e3 1878 filename = blk_bs(file)->filename;
45673671 1879 } else {
129c7d1c
MA
1880 /*
1881 * Caution: while qdict_get_try_str() is fine, getting
1882 * non-string types would require more care. When @options
1883 * come from -blockdev or blockdev_add, its members are typed
1884 * according to the QAPI schema, but when they come from
1885 * -drive, they're all QString.
1886 */
45673671
KW
1887 filename = qdict_get_try_str(options, "filename");
1888 }
1889
4a008240 1890 if (drv->bdrv_needs_filename && (!filename || !filename[0])) {
765003db
KW
1891 error_setg(errp, "The '%s' block driver requires a file name",
1892 drv->format_name);
18edf289
KW
1893 ret = -EINVAL;
1894 goto fail_opts;
6913c0c2 1895 }
6913c0c2 1896
82dc8b41
KW
1897 trace_bdrv_open_common(bs, filename ?: "", bs->open_flags,
1898 drv->format_name);
62392ebb 1899
307261b2
VSO
1900 ro = bdrv_is_read_only(bs);
1901
1902 if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv, ro)) {
1903 if (!ro && bdrv_is_whitelisted(drv, true)) {
8be25de6
KW
1904 ret = bdrv_apply_auto_read_only(bs, NULL, NULL);
1905 } else {
1906 ret = -ENOTSUP;
1907 }
1908 if (ret < 0) {
1909 error_setg(errp,
307261b2 1910 !ro && bdrv_is_whitelisted(drv, true)
8be25de6
KW
1911 ? "Driver '%s' can only be used for read-only devices"
1912 : "Driver '%s' is not whitelisted",
1913 drv->format_name);
1914 goto fail_opts;
1915 }
b64ec4e4 1916 }
57915332 1917
d3faa13e 1918 /* bdrv_new() and bdrv_close() make it so */
d73415a3 1919 assert(qatomic_read(&bs->copy_on_read) == 0);
d3faa13e 1920
82dc8b41 1921 if (bs->open_flags & BDRV_O_COPY_ON_READ) {
307261b2 1922 if (!ro) {
0ebd24e0
KW
1923 bdrv_enable_copy_on_read(bs);
1924 } else {
1925 error_setg(errp, "Can't use copy-on-read on read-only device");
18edf289
KW
1926 ret = -EINVAL;
1927 goto fail_opts;
0ebd24e0 1928 }
53fec9d3
SH
1929 }
1930
415bbca8 1931 discard = qemu_opt_get(opts, BDRV_OPT_DISCARD);
818584a4
KW
1932 if (discard != NULL) {
1933 if (bdrv_parse_discard_flags(discard, &bs->open_flags) != 0) {
1934 error_setg(errp, "Invalid discard option");
1935 ret = -EINVAL;
1936 goto fail_opts;
1937 }
1938 }
1939
543770bd
AG
1940 bs->detect_zeroes =
1941 bdrv_parse_detect_zeroes(opts, bs->open_flags, &local_err);
1942 if (local_err) {
1943 error_propagate(errp, local_err);
1944 ret = -EINVAL;
1945 goto fail_opts;
692e01a2
KW
1946 }
1947
c2ad1b0c
KW
1948 if (filename != NULL) {
1949 pstrcpy(bs->filename, sizeof(bs->filename), filename);
1950 } else {
1951 bs->filename[0] = '\0';
1952 }
91af7014 1953 pstrcpy(bs->exact_filename, sizeof(bs->exact_filename), bs->filename);
57915332 1954
66f82cee 1955 /* Open the image, either directly or using a protocol */
82dc8b41 1956 open_flags = bdrv_open_flags(bs, bs->open_flags);
01a56501 1957 node_name = qemu_opt_get(opts, "node-name");
57915332 1958
01a56501
KW
1959 assert(!drv->bdrv_file_open || file == NULL);
1960 ret = bdrv_open_driver(bs, drv, node_name, options, open_flags, errp);
51762288 1961 if (ret < 0) {
01a56501 1962 goto fail_opts;
3baca891
KW
1963 }
1964
18edf289 1965 qemu_opts_del(opts);
57915332
KW
1966 return 0;
1967
18edf289
KW
1968fail_opts:
1969 qemu_opts_del(opts);
57915332
KW
1970 return ret;
1971}
1972
5e5c4f63
KW
1973static QDict *parse_json_filename(const char *filename, Error **errp)
1974{
1975 QObject *options_obj;
1976 QDict *options;
1977 int ret;
bdb73476 1978 GLOBAL_STATE_CODE();
5e5c4f63
KW
1979
1980 ret = strstart(filename, "json:", &filename);
1981 assert(ret);
1982
5577fff7 1983 options_obj = qobject_from_json(filename, errp);
5e5c4f63 1984 if (!options_obj) {
5577fff7 1985 error_prepend(errp, "Could not parse the JSON options: ");
5e5c4f63
KW
1986 return NULL;
1987 }
1988
7dc847eb 1989 options = qobject_to(QDict, options_obj);
ca6b6e1e 1990 if (!options) {
cb3e7f08 1991 qobject_unref(options_obj);
5e5c4f63
KW
1992 error_setg(errp, "Invalid JSON object given");
1993 return NULL;
1994 }
1995
5e5c4f63
KW
1996 qdict_flatten(options);
1997
1998 return options;
1999}
2000
de3b53f0
KW
2001static void parse_json_protocol(QDict *options, const char **pfilename,
2002 Error **errp)
2003{
2004 QDict *json_options;
2005 Error *local_err = NULL;
bdb73476 2006 GLOBAL_STATE_CODE();
de3b53f0
KW
2007
2008 /* Parse json: pseudo-protocol */
2009 if (!*pfilename || !g_str_has_prefix(*pfilename, "json:")) {
2010 return;
2011 }
2012
2013 json_options = parse_json_filename(*pfilename, &local_err);
2014 if (local_err) {
2015 error_propagate(errp, local_err);
2016 return;
2017 }
2018
2019 /* Options given in the filename have lower priority than options
2020 * specified directly */
2021 qdict_join(options, json_options, false);
cb3e7f08 2022 qobject_unref(json_options);
de3b53f0
KW
2023 *pfilename = NULL;
2024}
2025
b6ce07aa 2026/*
f54120ff
KW
2027 * Fills in default options for opening images and converts the legacy
2028 * filename/flags pair to option QDict entries.
53a29513
HR
2029 * The BDRV_O_PROTOCOL flag in *flags will be set or cleared accordingly if a
2030 * block driver has been specified explicitly.
b6ce07aa 2031 */
de3b53f0 2032static int bdrv_fill_options(QDict **options, const char *filename,
053e1578 2033 int *flags, Error **errp)
ea2384d3 2034{
c2ad1b0c 2035 const char *drvname;
53a29513 2036 bool protocol = *flags & BDRV_O_PROTOCOL;
e3fa4bfa 2037 bool parse_filename = false;
053e1578 2038 BlockDriver *drv = NULL;
34b5d2c6 2039 Error *local_err = NULL;
83f64091 2040
da359909
EGE
2041 GLOBAL_STATE_CODE();
2042
129c7d1c
MA
2043 /*
2044 * Caution: while qdict_get_try_str() is fine, getting non-string
2045 * types would require more care. When @options come from
2046 * -blockdev or blockdev_add, its members are typed according to
2047 * the QAPI schema, but when they come from -drive, they're all
2048 * QString.
2049 */
53a29513 2050 drvname = qdict_get_try_str(*options, "driver");
053e1578
HR
2051 if (drvname) {
2052 drv = bdrv_find_format(drvname);
2053 if (!drv) {
2054 error_setg(errp, "Unknown driver '%s'", drvname);
2055 return -ENOENT;
2056 }
2057 /* If the user has explicitly specified the driver, this choice should
2058 * override the BDRV_O_PROTOCOL flag */
2059 protocol = drv->bdrv_file_open;
53a29513
HR
2060 }
2061
2062 if (protocol) {
2063 *flags |= BDRV_O_PROTOCOL;
2064 } else {
2065 *flags &= ~BDRV_O_PROTOCOL;
2066 }
2067
91a097e7
KW
2068 /* Translate cache options from flags into options */
2069 update_options_from_flags(*options, *flags);
2070
035fccdf 2071 /* Fetch the file name from the options QDict if necessary */
17b005f1 2072 if (protocol && filename) {
f54120ff 2073 if (!qdict_haskey(*options, "filename")) {
46f5ac20 2074 qdict_put_str(*options, "filename", filename);
f54120ff
KW
2075 parse_filename = true;
2076 } else {
2077 error_setg(errp, "Can't specify 'file' and 'filename' options at "
2078 "the same time");
2079 return -EINVAL;
2080 }
035fccdf
KW
2081 }
2082
c2ad1b0c 2083 /* Find the right block driver */
129c7d1c 2084 /* See cautionary note on accessing @options above */
f54120ff 2085 filename = qdict_get_try_str(*options, "filename");
f54120ff 2086
053e1578
HR
2087 if (!drvname && protocol) {
2088 if (filename) {
2089 drv = bdrv_find_protocol(filename, parse_filename, errp);
17b005f1 2090 if (!drv) {
053e1578 2091 return -EINVAL;
17b005f1 2092 }
053e1578
HR
2093
2094 drvname = drv->format_name;
46f5ac20 2095 qdict_put_str(*options, "driver", drvname);
053e1578
HR
2096 } else {
2097 error_setg(errp, "Must specify either driver or file");
2098 return -EINVAL;
98289620 2099 }
c2ad1b0c
KW
2100 }
2101
17b005f1 2102 assert(drv || !protocol);
c2ad1b0c 2103
f54120ff 2104 /* Driver-specific filename parsing */
17b005f1 2105 if (drv && drv->bdrv_parse_filename && parse_filename) {
5acd9d81 2106 drv->bdrv_parse_filename(filename, *options, &local_err);
84d18f06 2107 if (local_err) {
34b5d2c6 2108 error_propagate(errp, local_err);
f54120ff 2109 return -EINVAL;
6963a30d 2110 }
cd5d031e
HR
2111
2112 if (!drv->bdrv_needs_filename) {
2113 qdict_del(*options, "filename");
cd5d031e 2114 }
6963a30d
KW
2115 }
2116
f54120ff
KW
2117 return 0;
2118}
2119
148eb13c
KW
2120typedef struct BlockReopenQueueEntry {
2121 bool prepared;
2122 BDRVReopenState state;
859443b0 2123 QTAILQ_ENTRY(BlockReopenQueueEntry) entry;
148eb13c
KW
2124} BlockReopenQueueEntry;
2125
2126/*
2127 * Return the flags that @bs will have after the reopens in @q have
2128 * successfully completed. If @q is NULL (or @bs is not contained in @q),
2129 * return the current flags.
2130 */
2131static int bdrv_reopen_get_flags(BlockReopenQueue *q, BlockDriverState *bs)
2132{
2133 BlockReopenQueueEntry *entry;
2134
2135 if (q != NULL) {
859443b0 2136 QTAILQ_FOREACH(entry, q, entry) {
148eb13c
KW
2137 if (entry->state.bs == bs) {
2138 return entry->state.flags;
2139 }
2140 }
2141 }
2142
2143 return bs->open_flags;
2144}
2145
2146/* Returns whether the image file can be written to after the reopen queue @q
2147 * has been successfully applied, or right now if @q is NULL. */
cc022140
HR
2148static bool bdrv_is_writable_after_reopen(BlockDriverState *bs,
2149 BlockReopenQueue *q)
148eb13c
KW
2150{
2151 int flags = bdrv_reopen_get_flags(q, bs);
2152
2153 return (flags & (BDRV_O_RDWR | BDRV_O_INACTIVE)) == BDRV_O_RDWR;
2154}
2155
cc022140
HR
2156/*
2157 * Return whether the BDS can be written to. This is not necessarily
2158 * the same as !bdrv_is_read_only(bs), as inactivated images may not
2159 * be written to but do not count as read-only images.
2160 */
2161bool bdrv_is_writable(BlockDriverState *bs)
2162{
384a48fb 2163 IO_CODE();
cc022140
HR
2164 return bdrv_is_writable_after_reopen(bs, NULL);
2165}
2166
3bf416ba
VSO
2167static char *bdrv_child_user_desc(BdrvChild *c)
2168{
f0c28327 2169 GLOBAL_STATE_CODE();
da261b69 2170 return c->klass->get_parent_desc(c);
3bf416ba
VSO
2171}
2172
30ebb9aa
VSO
2173/*
2174 * Check that @a allows everything that @b needs. @a and @b must reference same
2175 * child node.
2176 */
3bf416ba
VSO
2177static bool bdrv_a_allow_b(BdrvChild *a, BdrvChild *b, Error **errp)
2178{
30ebb9aa
VSO
2179 const char *child_bs_name;
2180 g_autofree char *a_user = NULL;
2181 g_autofree char *b_user = NULL;
2182 g_autofree char *perms = NULL;
2183
2184 assert(a->bs);
2185 assert(a->bs == b->bs);
862fded9 2186 GLOBAL_STATE_CODE();
3bf416ba
VSO
2187
2188 if ((b->perm & a->shared_perm) == b->perm) {
2189 return true;
2190 }
2191
30ebb9aa
VSO
2192 child_bs_name = bdrv_get_node_name(b->bs);
2193 a_user = bdrv_child_user_desc(a);
2194 b_user = bdrv_child_user_desc(b);
2195 perms = bdrv_perm_names(b->perm & ~a->shared_perm);
2196
2197 error_setg(errp, "Permission conflict on node '%s': permissions '%s' are "
2198 "both required by %s (uses node '%s' as '%s' child) and "
2199 "unshared by %s (uses node '%s' as '%s' child).",
2200 child_bs_name, perms,
2201 b_user, child_bs_name, b->name,
2202 a_user, child_bs_name, a->name);
3bf416ba
VSO
2203
2204 return false;
2205}
2206
3804e3cf
KW
2207static bool GRAPH_RDLOCK
2208bdrv_parent_perms_conflict(BlockDriverState *bs, Error **errp)
3bf416ba
VSO
2209{
2210 BdrvChild *a, *b;
862fded9 2211 GLOBAL_STATE_CODE();
3bf416ba
VSO
2212
2213 /*
2214 * During the loop we'll look at each pair twice. That's correct because
2215 * bdrv_a_allow_b() is asymmetric and we should check each pair in both
2216 * directions.
2217 */
2218 QLIST_FOREACH(a, &bs->parents, next_parent) {
2219 QLIST_FOREACH(b, &bs->parents, next_parent) {
9397c14f 2220 if (a == b) {
3bf416ba
VSO
2221 continue;
2222 }
2223
2224 if (!bdrv_a_allow_b(a, b, errp)) {
2225 return true;
2226 }
2227 }
2228 }
2229
2230 return false;
2231}
2232
c629b6d2
KW
2233static void GRAPH_RDLOCK
2234bdrv_child_perm(BlockDriverState *bs, BlockDriverState *child_bs,
2235 BdrvChild *c, BdrvChildRole role,
2236 BlockReopenQueue *reopen_queue,
2237 uint64_t parent_perm, uint64_t parent_shared,
2238 uint64_t *nperm, uint64_t *nshared)
ffd1a5a2 2239{
0b3ca76e 2240 assert(bs->drv && bs->drv->bdrv_child_perm);
da359909 2241 GLOBAL_STATE_CODE();
e5d8a406 2242 bs->drv->bdrv_child_perm(bs, c, role, reopen_queue,
0b3ca76e
AG
2243 parent_perm, parent_shared,
2244 nperm, nshared);
e0995dc3 2245 /* TODO Take force_share from reopen_queue */
ffd1a5a2
FZ
2246 if (child_bs && child_bs->force_share) {
2247 *nshared = BLK_PERM_ALL;
2248 }
2249}
2250
bd57f8f7
VSO
2251/*
2252 * Adds the whole subtree of @bs (including @bs itself) to the @list (except for
2253 * nodes that are already in the @list, of course) so that final list is
2254 * topologically sorted. Return the result (GSList @list object is updated, so
2255 * don't use old reference after function call).
2256 *
2257 * On function start @list must be already topologically sorted and for any node
2258 * in the @list the whole subtree of the node must be in the @list as well. The
2259 * simplest way to satisfy this criteria: use only result of
2260 * bdrv_topological_dfs() or NULL as @list parameter.
2261 */
3804e3cf
KW
2262static GSList * GRAPH_RDLOCK
2263bdrv_topological_dfs(GSList *list, GHashTable *found, BlockDriverState *bs)
bd57f8f7
VSO
2264{
2265 BdrvChild *child;
2266 g_autoptr(GHashTable) local_found = NULL;
2267
bdb73476
EGE
2268 GLOBAL_STATE_CODE();
2269
bd57f8f7
VSO
2270 if (!found) {
2271 assert(!list);
2272 found = local_found = g_hash_table_new(NULL, NULL);
2273 }
2274
2275 if (g_hash_table_contains(found, bs)) {
2276 return list;
2277 }
2278 g_hash_table_add(found, bs);
2279
2280 QLIST_FOREACH(child, &bs->children, next) {
2281 list = bdrv_topological_dfs(list, found, child->bs);
2282 }
2283
2284 return g_slist_prepend(list, bs);
2285}
2286
ecb776bd
VSO
2287typedef struct BdrvChildSetPermState {
2288 BdrvChild *child;
2289 uint64_t old_perm;
2290 uint64_t old_shared_perm;
2291} BdrvChildSetPermState;
b0defa83
VSO
2292
2293static void bdrv_child_set_perm_abort(void *opaque)
2294{
ecb776bd
VSO
2295 BdrvChildSetPermState *s = opaque;
2296
862fded9
EGE
2297 GLOBAL_STATE_CODE();
2298
ecb776bd
VSO
2299 s->child->perm = s->old_perm;
2300 s->child->shared_perm = s->old_shared_perm;
b0defa83
VSO
2301}
2302
2303static TransactionActionDrv bdrv_child_set_pem_drv = {
2304 .abort = bdrv_child_set_perm_abort,
ecb776bd 2305 .clean = g_free,
b0defa83
VSO
2306};
2307
ecb776bd
VSO
2308static void bdrv_child_set_perm(BdrvChild *c, uint64_t perm,
2309 uint64_t shared, Transaction *tran)
b0defa83 2310{
ecb776bd 2311 BdrvChildSetPermState *s = g_new(BdrvChildSetPermState, 1);
862fded9 2312 GLOBAL_STATE_CODE();
ecb776bd
VSO
2313
2314 *s = (BdrvChildSetPermState) {
2315 .child = c,
2316 .old_perm = c->perm,
2317 .old_shared_perm = c->shared_perm,
2318 };
b0defa83
VSO
2319
2320 c->perm = perm;
2321 c->shared_perm = shared;
2322
ecb776bd 2323 tran_add(tran, &bdrv_child_set_pem_drv, s);
b0defa83
VSO
2324}
2325
bce73bc2 2326static void GRAPH_RDLOCK bdrv_drv_set_perm_commit(void *opaque)
2513ef59
VSO
2327{
2328 BlockDriverState *bs = opaque;
2329 uint64_t cumulative_perms, cumulative_shared_perms;
da359909 2330 GLOBAL_STATE_CODE();
2513ef59
VSO
2331
2332 if (bs->drv->bdrv_set_perm) {
2333 bdrv_get_cumulative_perm(bs, &cumulative_perms,
2334 &cumulative_shared_perms);
2335 bs->drv->bdrv_set_perm(bs, cumulative_perms, cumulative_shared_perms);
2336 }
2337}
2338
bce73bc2 2339static void GRAPH_RDLOCK bdrv_drv_set_perm_abort(void *opaque)
2513ef59
VSO
2340{
2341 BlockDriverState *bs = opaque;
da359909 2342 GLOBAL_STATE_CODE();
2513ef59
VSO
2343
2344 if (bs->drv->bdrv_abort_perm_update) {
2345 bs->drv->bdrv_abort_perm_update(bs);
2346 }
2347}
2348
2349TransactionActionDrv bdrv_drv_set_perm_drv = {
2350 .abort = bdrv_drv_set_perm_abort,
2351 .commit = bdrv_drv_set_perm_commit,
2352};
2353
bce73bc2
KW
2354/*
2355 * After calling this function, the transaction @tran may only be completed
2356 * while holding a reader lock for the graph.
2357 */
2358static int GRAPH_RDLOCK
2359bdrv_drv_set_perm(BlockDriverState *bs, uint64_t perm, uint64_t shared_perm,
2360 Transaction *tran, Error **errp)
2513ef59 2361{
da359909 2362 GLOBAL_STATE_CODE();
2513ef59
VSO
2363 if (!bs->drv) {
2364 return 0;
2365 }
2366
2367 if (bs->drv->bdrv_check_perm) {
2368 int ret = bs->drv->bdrv_check_perm(bs, perm, shared_perm, errp);
2369 if (ret < 0) {
2370 return ret;
2371 }
2372 }
2373
2374 if (tran) {
2375 tran_add(tran, &bdrv_drv_set_perm_drv, bs);
2376 }
2377
2378 return 0;
2379}
2380
0978623e
VSO
2381typedef struct BdrvReplaceChildState {
2382 BdrvChild *child;
2383 BlockDriverState *old_bs;
2384} BdrvReplaceChildState;
2385
5661a00d 2386static void GRAPH_WRLOCK bdrv_replace_child_commit(void *opaque)
0978623e
VSO
2387{
2388 BdrvReplaceChildState *s = opaque;
bdb73476 2389 GLOBAL_STATE_CODE();
0978623e 2390
5661a00d 2391 bdrv_schedule_unref(s->old_bs);
0978623e
VSO
2392}
2393
5661a00d 2394static void GRAPH_WRLOCK bdrv_replace_child_abort(void *opaque)
0978623e
VSO
2395{
2396 BdrvReplaceChildState *s = opaque;
2397 BlockDriverState *new_bs = s->child->bs;
2398
bdb73476 2399 GLOBAL_STATE_CODE();
5661a00d 2400 assert_bdrv_graph_writable();
ad29eb3d 2401
0f0b1e29 2402 /* old_bs reference is transparently moved from @s to @s->child */
23987471
KW
2403 if (!s->child->bs) {
2404 /*
2405 * The parents were undrained when removing old_bs from the child. New
2406 * requests can't have been made, though, because the child was empty.
2407 *
2408 * TODO Make bdrv_replace_child_noperm() transactionable to avoid
2409 * undraining the parent in the first place. Once this is done, having
2410 * new_bs drained when calling bdrv_replace_child_tran() is not a
2411 * requirement any more.
2412 */
606ed756 2413 bdrv_parent_drained_begin_single(s->child);
23987471
KW
2414 assert(!bdrv_parent_drained_poll_single(s->child));
2415 }
2416 assert(s->child->quiesced_parent);
544acc7d 2417 bdrv_replace_child_noperm(s->child, s->old_bs);
ad29eb3d 2418
0978623e
VSO
2419 bdrv_unref(new_bs);
2420}
2421
2422static TransactionActionDrv bdrv_replace_child_drv = {
2423 .commit = bdrv_replace_child_commit,
2424 .abort = bdrv_replace_child_abort,
2425 .clean = g_free,
2426};
2427
2428/*
4bf021db 2429 * bdrv_replace_child_tran
0978623e
VSO
2430 *
2431 * Note: real unref of old_bs is done only on commit.
4bf021db 2432 *
23987471
KW
2433 * Both @child->bs and @new_bs (if non-NULL) must be drained. @new_bs must be
2434 * kept drained until the transaction is completed.
2435 *
5661a00d
KW
2436 * After calling this function, the transaction @tran may only be completed
2437 * while holding a writer lock for the graph.
2438 *
4bf021db 2439 * The function doesn't update permissions, caller is responsible for this.
0978623e 2440 */
2f64e1fc
KW
2441static void GRAPH_WRLOCK
2442bdrv_replace_child_tran(BdrvChild *child, BlockDriverState *new_bs,
2443 Transaction *tran)
0978623e
VSO
2444{
2445 BdrvReplaceChildState *s = g_new(BdrvReplaceChildState, 1);
23987471
KW
2446
2447 assert(child->quiesced_parent);
2448 assert(!new_bs || new_bs->quiesce_counter);
2449
0978623e 2450 *s = (BdrvReplaceChildState) {
0f0b1e29
VSO
2451 .child = child,
2452 .old_bs = child->bs,
0978623e
VSO
2453 };
2454 tran_add(tran, &bdrv_replace_child_drv, s);
2455
2456 if (new_bs) {
2457 bdrv_ref(new_bs);
2458 }
ad29eb3d 2459
544acc7d 2460 bdrv_replace_child_noperm(child, new_bs);
0f0b1e29 2461 /* old_bs reference is transparently moved from @child to @s */
0978623e
VSO
2462}
2463
33a610c3 2464/*
c20555e1
VSO
2465 * Refresh permissions in @bs subtree. The function is intended to be called
2466 * after some graph modification that was done without permission update.
bce73bc2
KW
2467 *
2468 * After calling this function, the transaction @tran may only be completed
2469 * while holding a reader lock for the graph.
33a610c3 2470 */
bce73bc2
KW
2471static int GRAPH_RDLOCK
2472bdrv_node_refresh_perm(BlockDriverState *bs, BlockReopenQueue *q,
2473 Transaction *tran, Error **errp)
33a610c3
KW
2474{
2475 BlockDriver *drv = bs->drv;
2476 BdrvChild *c;
2477 int ret;
c20555e1 2478 uint64_t cumulative_perms, cumulative_shared_perms;
862fded9 2479 GLOBAL_STATE_CODE();
c20555e1
VSO
2480
2481 bdrv_get_cumulative_perm(bs, &cumulative_perms, &cumulative_shared_perms);
33a610c3
KW
2482
2483 /* Write permissions never work with read-only images */
2484 if ((cumulative_perms & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)) &&
cc022140 2485 !bdrv_is_writable_after_reopen(bs, q))
33a610c3 2486 {
481e0eee
HR
2487 if (!bdrv_is_writable_after_reopen(bs, NULL)) {
2488 error_setg(errp, "Block node is read-only");
2489 } else {
c20555e1
VSO
2490 error_setg(errp, "Read-only block node '%s' cannot support "
2491 "read-write users", bdrv_get_node_name(bs));
481e0eee
HR
2492 }
2493
33a610c3
KW
2494 return -EPERM;
2495 }
2496
9c60a5d1
KW
2497 /*
2498 * Unaligned requests will automatically be aligned to bl.request_alignment
2499 * and without RESIZE we can't extend requests to write to space beyond the
2500 * end of the image, so it's required that the image size is aligned.
2501 */
2502 if ((cumulative_perms & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)) &&
2503 !(cumulative_perms & BLK_PERM_RESIZE))
2504 {
2505 if ((bs->total_sectors * BDRV_SECTOR_SIZE) % bs->bl.request_alignment) {
2506 error_setg(errp, "Cannot get 'write' permission without 'resize': "
2507 "Image size is not a multiple of request "
2508 "alignment");
2509 return -EPERM;
2510 }
2511 }
2512
33a610c3
KW
2513 /* Check this node */
2514 if (!drv) {
2515 return 0;
2516 }
2517
b1d2bbeb 2518 ret = bdrv_drv_set_perm(bs, cumulative_perms, cumulative_shared_perms, tran,
2513ef59
VSO
2519 errp);
2520 if (ret < 0) {
2521 return ret;
33a610c3
KW
2522 }
2523
78e421c9 2524 /* Drivers that never have children can omit .bdrv_child_perm() */
33a610c3 2525 if (!drv->bdrv_child_perm) {
78e421c9 2526 assert(QLIST_EMPTY(&bs->children));
33a610c3
KW
2527 return 0;
2528 }
2529
2530 /* Check all children */
2531 QLIST_FOREACH(c, &bs->children, next) {
2532 uint64_t cur_perm, cur_shared;
9eab1544 2533
e5d8a406 2534 bdrv_child_perm(bs, c->bs, c, c->role, q,
ffd1a5a2
FZ
2535 cumulative_perms, cumulative_shared_perms,
2536 &cur_perm, &cur_shared);
ecb776bd 2537 bdrv_child_set_perm(c, cur_perm, cur_shared, tran);
bd57f8f7
VSO
2538 }
2539
2540 return 0;
2541}
3ef45e02 2542
fb0ff4d1
VSO
2543/*
2544 * @list is a product of bdrv_topological_dfs() (may be called several times) -
2545 * a topologically sorted subgraph.
bce73bc2
KW
2546 *
2547 * After calling this function, the transaction @tran may only be completed
2548 * while holding a reader lock for the graph.
fb0ff4d1 2549 */
3804e3cf
KW
2550static int GRAPH_RDLOCK
2551bdrv_do_refresh_perms(GSList *list, BlockReopenQueue *q, Transaction *tran,
2552 Error **errp)
bd57f8f7
VSO
2553{
2554 int ret;
b1d2bbeb 2555 BlockDriverState *bs;
862fded9 2556 GLOBAL_STATE_CODE();
bd57f8f7 2557
b1d2bbeb
VSO
2558 for ( ; list; list = list->next) {
2559 bs = list->data;
2560
9397c14f 2561 if (bdrv_parent_perms_conflict(bs, errp)) {
b1d2bbeb 2562 return -EINVAL;
bd57f8f7
VSO
2563 }
2564
c20555e1 2565 ret = bdrv_node_refresh_perm(bs, q, tran, errp);
33a610c3
KW
2566 if (ret < 0) {
2567 return ret;
2568 }
2569 }
2570
2571 return 0;
2572}
2573
fb0ff4d1
VSO
2574/*
2575 * @list is any list of nodes. List is completed by all subtrees and
2576 * topologically sorted. It's not a problem if some node occurs in the @list
2577 * several times.
bce73bc2
KW
2578 *
2579 * After calling this function, the transaction @tran may only be completed
2580 * while holding a reader lock for the graph.
fb0ff4d1 2581 */
3804e3cf
KW
2582static int GRAPH_RDLOCK
2583bdrv_list_refresh_perms(GSList *list, BlockReopenQueue *q, Transaction *tran,
2584 Error **errp)
fb0ff4d1
VSO
2585{
2586 g_autoptr(GHashTable) found = g_hash_table_new(NULL, NULL);
2587 g_autoptr(GSList) refresh_list = NULL;
2588
2589 for ( ; list; list = list->next) {
2590 refresh_list = bdrv_topological_dfs(refresh_list, found, list->data);
2591 }
2592
2593 return bdrv_do_refresh_perms(refresh_list, q, tran, errp);
2594}
2595
c7a0f2be
KW
2596void bdrv_get_cumulative_perm(BlockDriverState *bs, uint64_t *perm,
2597 uint64_t *shared_perm)
33a610c3
KW
2598{
2599 BdrvChild *c;
2600 uint64_t cumulative_perms = 0;
2601 uint64_t cumulative_shared_perms = BLK_PERM_ALL;
2602
b4ad82aa
EGE
2603 GLOBAL_STATE_CODE();
2604
33a610c3
KW
2605 QLIST_FOREACH(c, &bs->parents, next_parent) {
2606 cumulative_perms |= c->perm;
2607 cumulative_shared_perms &= c->shared_perm;
2608 }
2609
2610 *perm = cumulative_perms;
2611 *shared_perm = cumulative_shared_perms;
2612}
2613
5176196c 2614char *bdrv_perm_names(uint64_t perm)
d083319f
KW
2615{
2616 struct perm_name {
2617 uint64_t perm;
2618 const char *name;
2619 } permissions[] = {
2620 { BLK_PERM_CONSISTENT_READ, "consistent read" },
2621 { BLK_PERM_WRITE, "write" },
2622 { BLK_PERM_WRITE_UNCHANGED, "write unchanged" },
2623 { BLK_PERM_RESIZE, "resize" },
d083319f
KW
2624 { 0, NULL }
2625 };
2626
e2a7423a 2627 GString *result = g_string_sized_new(30);
d083319f
KW
2628 struct perm_name *p;
2629
2630 for (p = permissions; p->name; p++) {
2631 if (perm & p->perm) {
e2a7423a
AG
2632 if (result->len > 0) {
2633 g_string_append(result, ", ");
2634 }
2635 g_string_append(result, p->name);
d083319f
KW
2636 }
2637 }
2638
e2a7423a 2639 return g_string_free(result, FALSE);
d083319f
KW
2640}
2641
33a610c3 2642
bce73bc2
KW
2643/*
2644 * @tran is allowed to be NULL. In this case no rollback is possible.
2645 *
2646 * After calling this function, the transaction @tran may only be completed
2647 * while holding a reader lock for the graph.
2648 */
3804e3cf
KW
2649static int GRAPH_RDLOCK
2650bdrv_refresh_perms(BlockDriverState *bs, Transaction *tran, Error **errp)
bb87e4d1
VSO
2651{
2652 int ret;
f1316edb 2653 Transaction *local_tran = NULL;
b1d2bbeb 2654 g_autoptr(GSList) list = bdrv_topological_dfs(NULL, NULL, bs);
862fded9 2655 GLOBAL_STATE_CODE();
bb87e4d1 2656
f1316edb
VSO
2657 if (!tran) {
2658 tran = local_tran = tran_new();
2659 }
2660
fb0ff4d1 2661 ret = bdrv_do_refresh_perms(list, NULL, tran, errp);
f1316edb
VSO
2662
2663 if (local_tran) {
2664 tran_finalize(local_tran, ret);
2665 }
bb87e4d1 2666
b1d2bbeb 2667 return ret;
bb87e4d1
VSO
2668}
2669
33a610c3
KW
2670int bdrv_child_try_set_perm(BdrvChild *c, uint64_t perm, uint64_t shared,
2671 Error **errp)
2672{
1046779e 2673 Error *local_err = NULL;
83928dc4 2674 Transaction *tran = tran_new();
33a610c3
KW
2675 int ret;
2676
b4ad82aa
EGE
2677 GLOBAL_STATE_CODE();
2678
ecb776bd 2679 bdrv_child_set_perm(c, perm, shared, tran);
83928dc4 2680
f1316edb 2681 ret = bdrv_refresh_perms(c->bs, tran, &local_err);
83928dc4
VSO
2682
2683 tran_finalize(tran, ret);
2684
33a610c3 2685 if (ret < 0) {
071b474f
VSO
2686 if ((perm & ~c->perm) || (c->shared_perm & ~shared)) {
2687 /* tighten permissions */
1046779e
HR
2688 error_propagate(errp, local_err);
2689 } else {
2690 /*
2691 * Our caller may intend to only loosen restrictions and
2692 * does not expect this function to fail. Errors are not
2693 * fatal in such a case, so we can just hide them from our
2694 * caller.
2695 */
2696 error_free(local_err);
2697 ret = 0;
2698 }
33a610c3
KW
2699 }
2700
83928dc4 2701 return ret;
d5e6f437
KW
2702}
2703
c1087f12
HR
2704int bdrv_child_refresh_perms(BlockDriverState *bs, BdrvChild *c, Error **errp)
2705{
2706 uint64_t parent_perms, parent_shared;
2707 uint64_t perms, shared;
2708
b4ad82aa
EGE
2709 GLOBAL_STATE_CODE();
2710
c1087f12 2711 bdrv_get_cumulative_perm(bs, &parent_perms, &parent_shared);
e5d8a406 2712 bdrv_child_perm(bs, c->bs, c, c->role, NULL,
bf8e925e 2713 parent_perms, parent_shared, &perms, &shared);
c1087f12
HR
2714
2715 return bdrv_child_try_set_perm(c, perms, shared, errp);
2716}
2717
87278af1
HR
2718/*
2719 * Default implementation for .bdrv_child_perm() for block filters:
2720 * Forward CONSISTENT_READ, WRITE, WRITE_UNCHANGED, and RESIZE to the
2721 * filtered child.
2722 */
2723static void bdrv_filter_default_perms(BlockDriverState *bs, BdrvChild *c,
87278af1
HR
2724 BdrvChildRole role,
2725 BlockReopenQueue *reopen_queue,
2726 uint64_t perm, uint64_t shared,
2727 uint64_t *nperm, uint64_t *nshared)
6a1b9ee1 2728{
862fded9 2729 GLOBAL_STATE_CODE();
e444fa83
KW
2730 *nperm = perm & DEFAULT_PERM_PASSTHROUGH;
2731 *nshared = (shared & DEFAULT_PERM_PASSTHROUGH) | DEFAULT_PERM_UNCHANGED;
6a1b9ee1
KW
2732}
2733
70082db4 2734static void bdrv_default_perms_for_cow(BlockDriverState *bs, BdrvChild *c,
70082db4
HR
2735 BdrvChildRole role,
2736 BlockReopenQueue *reopen_queue,
2737 uint64_t perm, uint64_t shared,
2738 uint64_t *nperm, uint64_t *nshared)
2739{
e5d8a406 2740 assert(role & BDRV_CHILD_COW);
862fded9 2741 GLOBAL_STATE_CODE();
70082db4
HR
2742
2743 /*
2744 * We want consistent read from backing files if the parent needs it.
2745 * No other operations are performed on backing files.
2746 */
2747 perm &= BLK_PERM_CONSISTENT_READ;
2748
2749 /*
2750 * If the parent can deal with changing data, we're okay with a
2751 * writable and resizable backing file.
2752 * TODO Require !(perm & BLK_PERM_CONSISTENT_READ), too?
2753 */
2754 if (shared & BLK_PERM_WRITE) {
2755 shared = BLK_PERM_WRITE | BLK_PERM_RESIZE;
2756 } else {
2757 shared = 0;
2758 }
2759
64631f36 2760 shared |= BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE_UNCHANGED;
70082db4
HR
2761
2762 if (bs->open_flags & BDRV_O_INACTIVE) {
2763 shared |= BLK_PERM_WRITE | BLK_PERM_RESIZE;
2764 }
2765
2766 *nperm = perm;
2767 *nshared = shared;
2768}
2769
6f838a4b 2770static void bdrv_default_perms_for_storage(BlockDriverState *bs, BdrvChild *c,
6f838a4b
HR
2771 BdrvChildRole role,
2772 BlockReopenQueue *reopen_queue,
2773 uint64_t perm, uint64_t shared,
2774 uint64_t *nperm, uint64_t *nshared)
2775{
2776 int flags;
2777
862fded9 2778 GLOBAL_STATE_CODE();
e5d8a406 2779 assert(role & (BDRV_CHILD_METADATA | BDRV_CHILD_DATA));
6f838a4b
HR
2780
2781 flags = bdrv_reopen_get_flags(reopen_queue, bs);
2782
2783 /*
2784 * Apart from the modifications below, the same permissions are
2785 * forwarded and left alone as for filters
2786 */
e5d8a406 2787 bdrv_filter_default_perms(bs, c, role, reopen_queue,
6f838a4b
HR
2788 perm, shared, &perm, &shared);
2789
f889054f
HR
2790 if (role & BDRV_CHILD_METADATA) {
2791 /* Format drivers may touch metadata even if the guest doesn't write */
2792 if (bdrv_is_writable_after_reopen(bs, reopen_queue)) {
2793 perm |= BLK_PERM_WRITE | BLK_PERM_RESIZE;
2794 }
2795
2796 /*
2797 * bs->file always needs to be consistent because of the
2798 * metadata. We can never allow other users to resize or write
2799 * to it.
2800 */
2801 if (!(flags & BDRV_O_NO_IO)) {
2802 perm |= BLK_PERM_CONSISTENT_READ;
2803 }
2804 shared &= ~(BLK_PERM_WRITE | BLK_PERM_RESIZE);
6f838a4b
HR
2805 }
2806
f889054f
HR
2807 if (role & BDRV_CHILD_DATA) {
2808 /*
2809 * Technically, everything in this block is a subset of the
2810 * BDRV_CHILD_METADATA path taken above, and so this could
2811 * be an "else if" branch. However, that is not obvious, and
2812 * this function is not performance critical, therefore we let
2813 * this be an independent "if".
2814 */
2815
2816 /*
2817 * We cannot allow other users to resize the file because the
2818 * format driver might have some assumptions about the size
2819 * (e.g. because it is stored in metadata, or because the file
2820 * is split into fixed-size data files).
2821 */
2822 shared &= ~BLK_PERM_RESIZE;
2823
2824 /*
2825 * WRITE_UNCHANGED often cannot be performed as such on the
2826 * data file. For example, the qcow2 driver may still need to
2827 * write copied clusters on copy-on-read.
2828 */
2829 if (perm & BLK_PERM_WRITE_UNCHANGED) {
2830 perm |= BLK_PERM_WRITE;
2831 }
2832
2833 /*
2834 * If the data file is written to, the format driver may
2835 * expect to be able to resize it by writing beyond the EOF.
2836 */
2837 if (perm & BLK_PERM_WRITE) {
2838 perm |= BLK_PERM_RESIZE;
2839 }
6f838a4b 2840 }
6f838a4b
HR
2841
2842 if (bs->open_flags & BDRV_O_INACTIVE) {
2843 shared |= BLK_PERM_WRITE | BLK_PERM_RESIZE;
2844 }
2845
2846 *nperm = perm;
2847 *nshared = shared;
2848}
2849
2519f549 2850void bdrv_default_perms(BlockDriverState *bs, BdrvChild *c,
e5d8a406 2851 BdrvChildRole role, BlockReopenQueue *reopen_queue,
2519f549
HR
2852 uint64_t perm, uint64_t shared,
2853 uint64_t *nperm, uint64_t *nshared)
2854{
b4ad82aa 2855 GLOBAL_STATE_CODE();
2519f549
HR
2856 if (role & BDRV_CHILD_FILTERED) {
2857 assert(!(role & (BDRV_CHILD_DATA | BDRV_CHILD_METADATA |
2858 BDRV_CHILD_COW)));
e5d8a406 2859 bdrv_filter_default_perms(bs, c, role, reopen_queue,
2519f549
HR
2860 perm, shared, nperm, nshared);
2861 } else if (role & BDRV_CHILD_COW) {
2862 assert(!(role & (BDRV_CHILD_DATA | BDRV_CHILD_METADATA)));
e5d8a406 2863 bdrv_default_perms_for_cow(bs, c, role, reopen_queue,
2519f549
HR
2864 perm, shared, nperm, nshared);
2865 } else if (role & (BDRV_CHILD_METADATA | BDRV_CHILD_DATA)) {
e5d8a406 2866 bdrv_default_perms_for_storage(bs, c, role, reopen_queue,
2519f549
HR
2867 perm, shared, nperm, nshared);
2868 } else {
2869 g_assert_not_reached();
2870 }
2871}
2872
7b1d9c4d
HR
2873uint64_t bdrv_qapi_perm_to_blk_perm(BlockPermission qapi_perm)
2874{
2875 static const uint64_t permissions[] = {
2876 [BLOCK_PERMISSION_CONSISTENT_READ] = BLK_PERM_CONSISTENT_READ,
2877 [BLOCK_PERMISSION_WRITE] = BLK_PERM_WRITE,
2878 [BLOCK_PERMISSION_WRITE_UNCHANGED] = BLK_PERM_WRITE_UNCHANGED,
2879 [BLOCK_PERMISSION_RESIZE] = BLK_PERM_RESIZE,
7b1d9c4d
HR
2880 };
2881
2882 QEMU_BUILD_BUG_ON(ARRAY_SIZE(permissions) != BLOCK_PERMISSION__MAX);
2883 QEMU_BUILD_BUG_ON(1UL << ARRAY_SIZE(permissions) != BLK_PERM_ALL + 1);
2884
2885 assert(qapi_perm < BLOCK_PERMISSION__MAX);
2886
2887 return permissions[qapi_perm];
2888}
2889
23987471
KW
2890/*
2891 * Replaces the node that a BdrvChild points to without updating permissions.
2892 *
2893 * If @new_bs is non-NULL, the parent of @child must already be drained through
31b2ddfe 2894 * @child and the caller must hold the AioContext lock for @new_bs.
23987471 2895 */
ad29eb3d
KW
2896static void GRAPH_WRLOCK
2897bdrv_replace_child_noperm(BdrvChild *child, BlockDriverState *new_bs)
e9740bc6
KW
2898{
2899 BlockDriverState *old_bs = child->bs;
debc2927 2900 int new_bs_quiesce_counter;
e9740bc6 2901
2cad1ebe 2902 assert(!child->frozen);
23987471
KW
2903
2904 /*
2905 * If we want to change the BdrvChild to point to a drained node as its new
2906 * child->bs, we need to make sure that its new parent is drained, too. In
2907 * other words, either child->quiesce_parent must already be true or we must
2908 * be able to set it and keep the parent's quiesce_counter consistent with
2909 * that, but without polling or starting new requests (this function
2910 * guarantees that it doesn't poll, and starting new requests would be
2911 * against the invariants of drain sections).
2912 *
2913 * To keep things simple, we pick the first option (child->quiesce_parent
2914 * must already be true). We also generalise the rule a bit to make it
2915 * easier to verify in callers and more likely to be covered in test cases:
2916 * The parent must be quiesced through this child even if new_bs isn't
2917 * currently drained.
2918 *
2919 * The only exception is for callers that always pass new_bs == NULL. In
2920 * this case, we obviously never need to consider the case of a drained
2921 * new_bs, so we can keep the callers simpler by allowing them not to drain
2922 * the parent.
2923 */
2924 assert(!new_bs || child->quiesced_parent);
bfb8aa6d 2925 assert(old_bs != new_bs);
f0c28327 2926 GLOBAL_STATE_CODE();
2cad1ebe 2927
bb2614e9
FZ
2928 if (old_bs && new_bs) {
2929 assert(bdrv_get_aio_context(old_bs) == bdrv_get_aio_context(new_bs));
2930 }
debc2927 2931
e9740bc6 2932 if (old_bs) {
bd86fb99
HR
2933 if (child->klass->detach) {
2934 child->klass->detach(child);
d736f119 2935 }
e9740bc6
KW
2936 QLIST_REMOVE(child, next_parent);
2937 }
36fe1331
KW
2938
2939 child->bs = new_bs;
2940
e9740bc6
KW
2941 if (new_bs) {
2942 QLIST_INSERT_HEAD(&new_bs->parents, child, next_parent);
bd86fb99
HR
2943 if (child->klass->attach) {
2944 child->klass->attach(child);
8ee03995
KW
2945 }
2946 }
debc2927
HR
2947
2948 /*
23987471
KW
2949 * If the parent was drained through this BdrvChild previously, but new_bs
2950 * is not drained, allow requests to come in only after the new node has
2951 * been attached.
debc2927 2952 */
57e05be3
KW
2953 new_bs_quiesce_counter = (new_bs ? new_bs->quiesce_counter : 0);
2954 if (!new_bs_quiesce_counter && child->quiesced_parent) {
debc2927 2955 bdrv_parent_drained_end_single(child);
debc2927 2956 }
8ee03995 2957}
33a610c3 2958
04c9c3a5
HR
2959/**
2960 * Free the given @child.
2961 *
2962 * The child must be empty (i.e. `child->bs == NULL`) and it must be
2963 * unused (i.e. not in a children list).
2964 */
2965static void bdrv_child_free(BdrvChild *child)
df581792 2966{
548a74c0 2967 assert(!child->bs);
bdb73476 2968 GLOBAL_STATE_CODE();
a225369b 2969 assert(!child->next.le_prev); /* not in children list */
04c9c3a5
HR
2970
2971 g_free(child->name);
2972 g_free(child);
548a74c0 2973}
d5e6f437 2974
548a74c0 2975typedef struct BdrvAttachChildCommonState {
5bb04747 2976 BdrvChild *child;
548a74c0
VSO
2977 AioContext *old_parent_ctx;
2978 AioContext *old_child_ctx;
2979} BdrvAttachChildCommonState;
2980
5661a00d 2981static void GRAPH_WRLOCK bdrv_attach_child_common_abort(void *opaque)
548a74c0
VSO
2982{
2983 BdrvAttachChildCommonState *s = opaque;
5bb04747 2984 BlockDriverState *bs = s->child->bs;
548a74c0 2985
f0c28327 2986 GLOBAL_STATE_CODE();
5661a00d 2987 assert_bdrv_graph_writable();
ad29eb3d 2988
5bb04747 2989 bdrv_replace_child_noperm(s->child, NULL);
548a74c0
VSO
2990
2991 if (bdrv_get_aio_context(bs) != s->old_child_ctx) {
142e6907 2992 bdrv_try_change_aio_context(bs, s->old_child_ctx, NULL, &error_abort);
548a74c0
VSO
2993 }
2994
5bb04747 2995 if (bdrv_child_get_parent_aio_context(s->child) != s->old_parent_ctx) {
f8be48ad
EGE
2996 Transaction *tran;
2997 GHashTable *visited;
2998 bool ret;
548a74c0 2999
f8be48ad 3000 tran = tran_new();
548a74c0 3001
f8be48ad
EGE
3002 /* No need to visit `child`, because it has been detached already */
3003 visited = g_hash_table_new(NULL, NULL);
3004 ret = s->child->klass->change_aio_ctx(s->child, s->old_parent_ctx,
3005 visited, tran, &error_abort);
3006 g_hash_table_destroy(visited);
3007
3008 /* transaction is supposed to always succeed */
3009 assert(ret == true);
3010 tran_commit(tran);
d5e6f437
KW
3011 }
3012
5661a00d 3013 bdrv_schedule_unref(bs);
5bb04747 3014 bdrv_child_free(s->child);
548a74c0
VSO
3015}
3016
3017static TransactionActionDrv bdrv_attach_child_common_drv = {
3018 .abort = bdrv_attach_child_common_abort,
3019 .clean = g_free,
3020};
3021
3022/*
3023 * Common part of attaching bdrv child to bs or to blk or to job
f8d2ad78 3024 *
7ec390d5 3025 * Function doesn't update permissions, caller is responsible for this.
5bb04747 3026 *
5661a00d
KW
3027 * After calling this function, the transaction @tran may only be completed
3028 * while holding a writer lock for the graph.
3029 *
5bb04747 3030 * Returns new created child.
c066e808
KW
3031 *
3032 * The caller must hold the AioContext lock for @child_bs. Both @parent_bs and
3033 * @child_bs can move to a different AioContext in this function. Callers must
3034 * make sure that their AioContext locking is still correct after this.
548a74c0 3035 */
7d4ca9d2
KW
3036static BdrvChild * GRAPH_WRLOCK
3037bdrv_attach_child_common(BlockDriverState *child_bs,
3038 const char *child_name,
3039 const BdrvChildClass *child_class,
3040 BdrvChildRole child_role,
3041 uint64_t perm, uint64_t shared_perm,
3042 void *opaque,
3043 Transaction *tran, Error **errp)
548a74c0
VSO
3044{
3045 BdrvChild *new_child;
c066e808 3046 AioContext *parent_ctx, *new_child_ctx;
548a74c0
VSO
3047 AioContext *child_ctx = bdrv_get_aio_context(child_bs);
3048
da261b69 3049 assert(child_class->get_parent_desc);
bdb73476 3050 GLOBAL_STATE_CODE();
548a74c0
VSO
3051
3052 new_child = g_new(BdrvChild, 1);
3053 *new_child = (BdrvChild) {
d5e6f437
KW
3054 .bs = NULL,
3055 .name = g_strdup(child_name),
bd86fb99 3056 .klass = child_class,
258b7765 3057 .role = child_role,
d5e6f437
KW
3058 .perm = perm,
3059 .shared_perm = shared_perm,
3060 .opaque = opaque,
df581792
KW
3061 };
3062
548a74c0
VSO
3063 /*
3064 * If the AioContexts don't match, first try to move the subtree of
132ada80 3065 * child_bs into the AioContext of the new parent. If this doesn't work,
548a74c0
VSO
3066 * try moving the parent into the AioContext of child_bs instead.
3067 */
3068 parent_ctx = bdrv_child_get_parent_aio_context(new_child);
3069 if (child_ctx != parent_ctx) {
3070 Error *local_err = NULL;
142e6907
EGE
3071 int ret = bdrv_try_change_aio_context(child_bs, parent_ctx, NULL,
3072 &local_err);
548a74c0 3073
f8be48ad
EGE
3074 if (ret < 0 && child_class->change_aio_ctx) {
3075 Transaction *tran = tran_new();
3076 GHashTable *visited = g_hash_table_new(NULL, NULL);
3077 bool ret_child;
3078
3079 g_hash_table_add(visited, new_child);
3080 ret_child = child_class->change_aio_ctx(new_child, child_ctx,
3081 visited, tran, NULL);
3082 if (ret_child == true) {
132ada80
KW
3083 error_free(local_err);
3084 ret = 0;
132ada80 3085 }
f8be48ad
EGE
3086 tran_finalize(tran, ret_child == true ? 0 : -1);
3087 g_hash_table_destroy(visited);
132ada80 3088 }
548a74c0 3089
132ada80
KW
3090 if (ret < 0) {
3091 error_propagate(errp, local_err);
04c9c3a5 3092 bdrv_child_free(new_child);
5bb04747 3093 return NULL;
132ada80
KW
3094 }
3095 }
3096
c066e808
KW
3097 new_child_ctx = bdrv_get_aio_context(child_bs);
3098 if (new_child_ctx != child_ctx) {
3099 aio_context_release(child_ctx);
3100 aio_context_acquire(new_child_ctx);
3101 }
3102
548a74c0 3103 bdrv_ref(child_bs);
23987471
KW
3104 /*
3105 * Let every new BdrvChild start with a drained parent. Inserting the child
3106 * in the graph with bdrv_replace_child_noperm() will undrain it if
3107 * @child_bs is not drained.
3108 *
3109 * The child was only just created and is not yet visible in global state
3110 * until bdrv_replace_child_noperm() inserts it into the graph, so nobody
3111 * could have sent requests and polling is not necessary.
3112 *
3113 * Note that this means that the parent isn't fully drained yet, we only
3114 * stop new requests from coming in. This is fine, we don't care about the
3115 * old requests here, they are not for this child. If another place enters a
3116 * drain section for the same parent, but wants it to be fully quiesced, it
3117 * will not run most of the the code in .drained_begin() again (which is not
3118 * a problem, we already did this), but it will still poll until the parent
3119 * is fully quiesced, so it will not be negatively affected either.
3120 */
606ed756 3121 bdrv_parent_drained_begin_single(new_child);
544acc7d 3122 bdrv_replace_child_noperm(new_child, child_bs);
b4b059f6 3123
548a74c0
VSO
3124 BdrvAttachChildCommonState *s = g_new(BdrvAttachChildCommonState, 1);
3125 *s = (BdrvAttachChildCommonState) {
5bb04747 3126 .child = new_child,
548a74c0
VSO
3127 .old_parent_ctx = parent_ctx,
3128 .old_child_ctx = child_ctx,
3129 };
3130 tran_add(tran, &bdrv_attach_child_common_drv, s);
3131
c066e808
KW
3132 if (new_child_ctx != child_ctx) {
3133 aio_context_release(new_child_ctx);
3134 aio_context_acquire(child_ctx);
3135 }
3136
5bb04747 3137 return new_child;
548a74c0
VSO
3138}
3139
f8d2ad78 3140/*
7ec390d5 3141 * Function doesn't update permissions, caller is responsible for this.
c066e808
KW
3142 *
3143 * The caller must hold the AioContext lock for @child_bs. Both @parent_bs and
3144 * @child_bs can move to a different AioContext in this function. Callers must
3145 * make sure that their AioContext locking is still correct after this.
5661a00d
KW
3146 *
3147 * After calling this function, the transaction @tran may only be completed
3148 * while holding a writer lock for the graph.
f8d2ad78 3149 */
7d4ca9d2
KW
3150static BdrvChild * GRAPH_WRLOCK
3151bdrv_attach_child_noperm(BlockDriverState *parent_bs,
3152 BlockDriverState *child_bs,
3153 const char *child_name,
3154 const BdrvChildClass *child_class,
3155 BdrvChildRole child_role,
3156 Transaction *tran,
3157 Error **errp)
aa5a04c7 3158{
aa5a04c7
VSO
3159 uint64_t perm, shared_perm;
3160
3161 assert(parent_bs->drv);
bdb73476 3162 GLOBAL_STATE_CODE();
aa5a04c7 3163
bfb8aa6d
KW
3164 if (bdrv_recurse_has_child(child_bs, parent_bs)) {
3165 error_setg(errp, "Making '%s' a %s child of '%s' would create a cycle",
3166 child_bs->node_name, child_name, parent_bs->node_name);
5bb04747 3167 return NULL;
bfb8aa6d
KW
3168 }
3169
aa5a04c7
VSO
3170 bdrv_get_cumulative_perm(parent_bs, &perm, &shared_perm);
3171 bdrv_child_perm(parent_bs, child_bs, NULL, child_role, NULL,
3172 perm, shared_perm, &perm, &shared_perm);
3173
5bb04747
VSO
3174 return bdrv_attach_child_common(child_bs, child_name, child_class,
3175 child_role, perm, shared_perm, parent_bs,
3176 tran, errp);
aa5a04c7
VSO
3177}
3178
548a74c0
VSO
3179/*
3180 * This function steals the reference to child_bs from the caller.
3181 * That reference is later dropped by bdrv_root_unref_child().
3182 *
3183 * On failure NULL is returned, errp is set and the reference to
3184 * child_bs is also dropped.
3185 *
3186 * The caller must hold the AioContext lock @child_bs, but not that of @ctx
3187 * (unless @child_bs is already in @ctx).
3188 */
3189BdrvChild *bdrv_root_attach_child(BlockDriverState *child_bs,
3190 const char *child_name,
3191 const BdrvChildClass *child_class,
3192 BdrvChildRole child_role,
3193 uint64_t perm, uint64_t shared_perm,
3194 void *opaque, Error **errp)
3195{
3196 int ret;
5bb04747 3197 BdrvChild *child;
548a74c0
VSO
3198 Transaction *tran = tran_new();
3199
b4ad82aa
EGE
3200 GLOBAL_STATE_CODE();
3201
7d4ca9d2
KW
3202 bdrv_graph_wrlock(child_bs);
3203
5bb04747 3204 child = bdrv_attach_child_common(child_bs, child_name, child_class,
548a74c0 3205 child_role, perm, shared_perm, opaque,
5bb04747
VSO
3206 tran, errp);
3207 if (!child) {
3208 ret = -EINVAL;
e878bb12 3209 goto out;
548a74c0
VSO
3210 }
3211
f1316edb 3212 ret = bdrv_refresh_perms(child_bs, tran, errp);
548a74c0 3213
e878bb12
KW
3214out:
3215 tran_finalize(tran, ret);
5661a00d 3216 bdrv_graph_wrunlock();
f8d2ad78 3217
548a74c0 3218 bdrv_unref(child_bs);
5bb04747
VSO
3219
3220 return ret < 0 ? NULL : child;
df581792
KW
3221}
3222
b441dc71
AG
3223/*
3224 * This function transfers the reference to child_bs from the caller
3225 * to parent_bs. That reference is later dropped by parent_bs on
3226 * bdrv_close() or if someone calls bdrv_unref_child().
3227 *
3228 * On failure NULL is returned, errp is set and the reference to
3229 * child_bs is also dropped.
132ada80
KW
3230 *
3231 * If @parent_bs and @child_bs are in different AioContexts, the caller must
3232 * hold the AioContext lock for @child_bs, but not for @parent_bs.
b441dc71 3233 */
98292c61
WC
3234BdrvChild *bdrv_attach_child(BlockDriverState *parent_bs,
3235 BlockDriverState *child_bs,
3236 const char *child_name,
bd86fb99 3237 const BdrvChildClass *child_class,
258b7765 3238 BdrvChildRole child_role,
8b2ff529 3239 Error **errp)
f21d96d0 3240{
aa5a04c7 3241 int ret;
5bb04747 3242 BdrvChild *child;
aa5a04c7 3243 Transaction *tran = tran_new();
f68c598b 3244
f791bf7f
EGE
3245 GLOBAL_STATE_CODE();
3246
5bb04747
VSO
3247 child = bdrv_attach_child_noperm(parent_bs, child_bs, child_name,
3248 child_class, child_role, tran, errp);
3249 if (!child) {
3250 ret = -EINVAL;
aa5a04c7
VSO
3251 goto out;
3252 }
d5e6f437 3253
f1316edb 3254 ret = bdrv_refresh_perms(parent_bs, tran, errp);
aa5a04c7
VSO
3255 if (ret < 0) {
3256 goto out;
d5e6f437
KW
3257 }
3258
aa5a04c7
VSO
3259out:
3260 tran_finalize(tran, ret);
3261
afdaeb9e 3262 bdrv_schedule_unref(child_bs);
aa5a04c7 3263
5bb04747 3264 return ret < 0 ? NULL : child;
f21d96d0
KW
3265}
3266
7b99a266 3267/* Callers must ensure that child->frozen is false. */
f21d96d0 3268void bdrv_root_unref_child(BdrvChild *child)
33a60407 3269{
00eb93b5 3270 BlockDriverState *child_bs = child->bs;
779020cb 3271
f791bf7f 3272 GLOBAL_STATE_CODE();
00eb93b5
VSO
3273 bdrv_replace_child_noperm(child, NULL);
3274 bdrv_child_free(child);
3275
3276 if (child_bs) {
3277 /*
3278 * Update permissions for old node. We're just taking a parent away, so
3279 * we're loosening restrictions. Errors of permission update are not
3280 * fatal in this case, ignore them.
3281 */
f1316edb 3282 bdrv_refresh_perms(child_bs, NULL, NULL);
00eb93b5
VSO
3283
3284 /*
3285 * When the parent requiring a non-default AioContext is removed, the
3286 * node moves back to the main AioContext
3287 */
3288 bdrv_try_change_aio_context(child_bs, qemu_get_aio_context(), NULL,
3289 NULL);
3290 }
f791bf7f 3291
ede01e46 3292 bdrv_schedule_unref(child_bs);
f21d96d0
KW
3293}
3294
332b3a17
VSO
3295typedef struct BdrvSetInheritsFrom {
3296 BlockDriverState *bs;
3297 BlockDriverState *old_inherits_from;
3298} BdrvSetInheritsFrom;
3299
3300static void bdrv_set_inherits_from_abort(void *opaque)
3301{
3302 BdrvSetInheritsFrom *s = opaque;
3303
3304 s->bs->inherits_from = s->old_inherits_from;
3305}
3306
3307static TransactionActionDrv bdrv_set_inherits_from_drv = {
3308 .abort = bdrv_set_inherits_from_abort,
3309 .clean = g_free,
3310};
3311
3312/* @tran is allowed to be NULL. In this case no rollback is possible */
3313static void bdrv_set_inherits_from(BlockDriverState *bs,
3314 BlockDriverState *new_inherits_from,
3315 Transaction *tran)
3316{
3317 if (tran) {
3318 BdrvSetInheritsFrom *s = g_new(BdrvSetInheritsFrom, 1);
3319
3320 *s = (BdrvSetInheritsFrom) {
3321 .bs = bs,
3322 .old_inherits_from = bs->inherits_from,
3323 };
3324
3325 tran_add(tran, &bdrv_set_inherits_from_drv, s);
3326 }
3327
3328 bs->inherits_from = new_inherits_from;
3329}
3330
3cf746b3
HR
3331/**
3332 * Clear all inherits_from pointers from children and grandchildren of
3333 * @root that point to @root, where necessary.
332b3a17 3334 * @tran is allowed to be NULL. In this case no rollback is possible
3cf746b3 3335 */
32a8aba3
KW
3336static void GRAPH_WRLOCK
3337bdrv_unset_inherits_from(BlockDriverState *root, BdrvChild *child,
3338 Transaction *tran)
f21d96d0 3339{
3cf746b3 3340 BdrvChild *c;
4e4bf5c4 3341
3cf746b3
HR
3342 if (child->bs->inherits_from == root) {
3343 /*
3344 * Remove inherits_from only when the last reference between root and
3345 * child->bs goes away.
3346 */
3347 QLIST_FOREACH(c, &root->children, next) {
4e4bf5c4
KW
3348 if (c != child && c->bs == child->bs) {
3349 break;
3350 }
3351 }
3352 if (c == NULL) {
332b3a17 3353 bdrv_set_inherits_from(child->bs, NULL, tran);
4e4bf5c4 3354 }
33a60407
KW
3355 }
3356
3cf746b3 3357 QLIST_FOREACH(c, &child->bs->children, next) {
332b3a17 3358 bdrv_unset_inherits_from(root, c, tran);
3cf746b3
HR
3359 }
3360}
3361
7b99a266 3362/* Callers must ensure that child->frozen is false. */
3cf746b3
HR
3363void bdrv_unref_child(BlockDriverState *parent, BdrvChild *child)
3364{
f791bf7f 3365 GLOBAL_STATE_CODE();
3cf746b3
HR
3366 if (child == NULL) {
3367 return;
3368 }
3369
332b3a17 3370 bdrv_unset_inherits_from(parent, child, NULL);
f21d96d0 3371 bdrv_root_unref_child(child);
33a60407
KW
3372}
3373
5c8cab48 3374
356f4ef6
KW
3375static void GRAPH_RDLOCK
3376bdrv_parent_cb_change_media(BlockDriverState *bs, bool load)
5c8cab48
KW
3377{
3378 BdrvChild *c;
f0c28327 3379 GLOBAL_STATE_CODE();
5c8cab48 3380 QLIST_FOREACH(c, &bs->parents, next_parent) {
bd86fb99
HR
3381 if (c->klass->change_media) {
3382 c->klass->change_media(c, load);
5c8cab48
KW
3383 }
3384 }
3385}
3386
0065c455
AG
3387/* Return true if you can reach parent going through child->inherits_from
3388 * recursively. If parent or child are NULL, return false */
3389static bool bdrv_inherits_from_recursive(BlockDriverState *child,
3390 BlockDriverState *parent)
3391{
3392 while (child && child != parent) {
3393 child = child->inherits_from;
3394 }
3395
3396 return child != NULL;
3397}
3398
25191e5f
HR
3399/*
3400 * Return the BdrvChildRole for @bs's backing child. bs->backing is
3401 * mostly used for COW backing children (role = COW), but also for
3402 * filtered children (role = FILTERED | PRIMARY).
3403 */
3404static BdrvChildRole bdrv_backing_role(BlockDriverState *bs)
3405{
3406 if (bs->drv && bs->drv->is_filter) {
3407 return BDRV_CHILD_FILTERED | BDRV_CHILD_PRIMARY;
3408 } else {
3409 return BDRV_CHILD_COW;
3410 }
3411}
3412
5db15a57 3413/*
e9238278
VSO
3414 * Sets the bs->backing or bs->file link of a BDS. A new reference is created;
3415 * callers which don't need their own reference any more must call bdrv_unref().
7ec390d5 3416 *
7d4ca9d2
KW
3417 * If the respective child is already present (i.e. we're detaching a node),
3418 * that child node must be drained.
3419 *
7ec390d5 3420 * Function doesn't update permissions, caller is responsible for this.
4b408668
KW
3421 *
3422 * The caller must hold the AioContext lock for @child_bs. Both @parent_bs and
3423 * @child_bs can move to a different AioContext in this function. Callers must
3424 * make sure that their AioContext locking is still correct after this.
5661a00d
KW
3425 *
3426 * After calling this function, the transaction @tran may only be completed
3427 * while holding a writer lock for the graph.
5db15a57 3428 */
7d4ca9d2
KW
3429static int GRAPH_WRLOCK
3430bdrv_set_file_or_backing_noperm(BlockDriverState *parent_bs,
3431 BlockDriverState *child_bs,
3432 bool is_backing,
3433 Transaction *tran, Error **errp)
8d24cce1 3434{
e9238278
VSO
3435 bool update_inherits_from =
3436 bdrv_inherits_from_recursive(child_bs, parent_bs);
3437 BdrvChild *child = is_backing ? parent_bs->backing : parent_bs->file;
3438 BdrvChildRole role;
0065c455 3439
bdb73476
EGE
3440 GLOBAL_STATE_CODE();
3441
e9238278
VSO
3442 if (!parent_bs->drv) {
3443 /*
3444 * Node without drv is an object without a class :/. TODO: finally fix
3445 * qcow2 driver to never clear bs->drv and implement format corruption
3446 * handling in other way.
3447 */
3448 error_setg(errp, "Node corrupted");
3449 return -EINVAL;
3450 }
3451
3452 if (child && child->frozen) {
3453 error_setg(errp, "Cannot change frozen '%s' link from '%s' to '%s'",
3454 child->name, parent_bs->node_name, child->bs->node_name);
a1e708fc 3455 return -EPERM;
2cad1ebe
AG
3456 }
3457
25f78d9e
VSO
3458 if (is_backing && !parent_bs->drv->is_filter &&
3459 !parent_bs->drv->supports_backing)
3460 {
3461 error_setg(errp, "Driver '%s' of node '%s' does not support backing "
3462 "files", parent_bs->drv->format_name, parent_bs->node_name);
3463 return -EINVAL;
3464 }
3465
e9238278
VSO
3466 if (parent_bs->drv->is_filter) {
3467 role = BDRV_CHILD_FILTERED | BDRV_CHILD_PRIMARY;
3468 } else if (is_backing) {
3469 role = BDRV_CHILD_COW;
3470 } else {
3471 /*
3472 * We only can use same role as it is in existing child. We don't have
3473 * infrastructure to determine role of file child in generic way
3474 */
3475 if (!child) {
3476 error_setg(errp, "Cannot set file child to format node without "
3477 "file child");
3478 return -EINVAL;
3479 }
3480 role = child->role;
826b6ca0
FZ
3481 }
3482
e9238278 3483 if (child) {
7d4ca9d2 3484 assert(child->bs->quiesce_counter);
e9238278 3485 bdrv_unset_inherits_from(parent_bs, child, tran);
57f08941 3486 bdrv_remove_child(child, tran);
e9238278
VSO
3487 }
3488
3489 if (!child_bs) {
8d24cce1
FZ
3490 goto out;
3491 }
12fa4af6 3492
5bb04747
VSO
3493 child = bdrv_attach_child_noperm(parent_bs, child_bs,
3494 is_backing ? "backing" : "file",
3495 &child_of_bds, role,
3496 tran, errp);
3497 if (!child) {
3498 return -EINVAL;
a1e708fc
VSO
3499 }
3500
160333e1
VSO
3501
3502 /*
e9238278 3503 * If inherits_from pointed recursively to bs then let's update it to
160333e1
VSO
3504 * point directly to bs (else it will become NULL).
3505 */
a1e708fc 3506 if (update_inherits_from) {
e9238278 3507 bdrv_set_inherits_from(child_bs, parent_bs, tran);
0065c455 3508 }
826b6ca0 3509
8d24cce1 3510out:
e9238278 3511 bdrv_refresh_limits(parent_bs, tran, NULL);
160333e1
VSO
3512
3513 return 0;
3514}
3515
4b408668
KW
3516/*
3517 * The caller must hold the AioContext lock for @backing_hd. Both @bs and
3518 * @backing_hd can move to a different AioContext in this function. Callers must
3519 * make sure that their AioContext locking is still correct after this.
7d4ca9d2
KW
3520 *
3521 * If a backing child is already present (i.e. we're detaching a node), that
3522 * child node must be drained.
5661a00d
KW
3523 *
3524 * After calling this function, the transaction @tran may only be completed
3525 * while holding a writer lock for the graph.
4b408668 3526 */
7d4ca9d2
KW
3527static int GRAPH_WRLOCK
3528bdrv_set_backing_noperm(BlockDriverState *bs,
3529 BlockDriverState *backing_hd,
3530 Transaction *tran, Error **errp)
e9238278 3531{
bdb73476 3532 GLOBAL_STATE_CODE();
e9238278
VSO
3533 return bdrv_set_file_or_backing_noperm(bs, backing_hd, true, tran, errp);
3534}
3535
92140b9f
KW
3536int bdrv_set_backing_hd_drained(BlockDriverState *bs,
3537 BlockDriverState *backing_hd,
3538 Error **errp)
160333e1
VSO
3539{
3540 int ret;
3541 Transaction *tran = tran_new();
3542
f791bf7f 3543 GLOBAL_STATE_CODE();
92140b9f 3544 assert(bs->quiesce_counter > 0);
7d4ca9d2
KW
3545 if (bs->backing) {
3546 assert(bs->backing->bs->quiesce_counter > 0);
3547 }
3548 bdrv_graph_wrlock(backing_hd);
c0829cb1 3549
160333e1
VSO
3550 ret = bdrv_set_backing_noperm(bs, backing_hd, tran, errp);
3551 if (ret < 0) {
3552 goto out;
3553 }
3554
f1316edb 3555 ret = bdrv_refresh_perms(bs, tran, errp);
160333e1
VSO
3556out:
3557 tran_finalize(tran, ret);
5661a00d 3558 bdrv_graph_wrunlock();
92140b9f
KW
3559 return ret;
3560}
a1e708fc 3561
92140b9f
KW
3562int bdrv_set_backing_hd(BlockDriverState *bs, BlockDriverState *backing_hd,
3563 Error **errp)
3564{
7d4ca9d2 3565 BlockDriverState *drain_bs = bs->backing ? bs->backing->bs : bs;
92140b9f
KW
3566 int ret;
3567 GLOBAL_STATE_CODE();
3568
7d4ca9d2
KW
3569 bdrv_ref(drain_bs);
3570 bdrv_drained_begin(drain_bs);
92140b9f 3571 ret = bdrv_set_backing_hd_drained(bs, backing_hd, errp);
7d4ca9d2
KW
3572 bdrv_drained_end(drain_bs);
3573 bdrv_unref(drain_bs);
c0829cb1 3574
a1e708fc 3575 return ret;
8d24cce1
FZ
3576}
3577
31ca6d07
KW
3578/*
3579 * Opens the backing file for a BlockDriverState if not yet open
3580 *
d9b7b057
KW
3581 * bdref_key specifies the key for the image's BlockdevRef in the options QDict.
3582 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
3583 * itself, all options starting with "${bdref_key}." are considered part of the
3584 * BlockdevRef.
3585 *
2626d27f
KW
3586 * The caller must hold the main AioContext lock.
3587 *
d9b7b057 3588 * TODO Can this be unified with bdrv_open_image()?
31ca6d07 3589 */
d9b7b057
KW
3590int bdrv_open_backing_file(BlockDriverState *bs, QDict *parent_options,
3591 const char *bdref_key, Error **errp)
9156df12 3592{
6b6833c1 3593 char *backing_filename = NULL;
d9b7b057
KW
3594 char *bdref_key_dot;
3595 const char *reference = NULL;
317fc44e 3596 int ret = 0;
998c2019 3597 bool implicit_backing = false;
8d24cce1 3598 BlockDriverState *backing_hd;
8aa04542 3599 AioContext *backing_hd_ctx;
d9b7b057
KW
3600 QDict *options;
3601 QDict *tmp_parent_options = NULL;
34b5d2c6 3602 Error *local_err = NULL;
9156df12 3603
f791bf7f
EGE
3604 GLOBAL_STATE_CODE();
3605
760e0063 3606 if (bs->backing != NULL) {
1ba4b6a5 3607 goto free_exit;
9156df12
PB
3608 }
3609
31ca6d07 3610 /* NULL means an empty set of options */
d9b7b057
KW
3611 if (parent_options == NULL) {
3612 tmp_parent_options = qdict_new();
3613 parent_options = tmp_parent_options;
31ca6d07
KW
3614 }
3615
9156df12 3616 bs->open_flags &= ~BDRV_O_NO_BACKING;
d9b7b057
KW
3617
3618 bdref_key_dot = g_strdup_printf("%s.", bdref_key);
3619 qdict_extract_subqdict(parent_options, &options, bdref_key_dot);
3620 g_free(bdref_key_dot);
3621
129c7d1c
MA
3622 /*
3623 * Caution: while qdict_get_try_str() is fine, getting non-string
3624 * types would require more care. When @parent_options come from
3625 * -blockdev or blockdev_add, its members are typed according to
3626 * the QAPI schema, but when they come from -drive, they're all
3627 * QString.
3628 */
d9b7b057
KW
3629 reference = qdict_get_try_str(parent_options, bdref_key);
3630 if (reference || qdict_haskey(options, "file.filename")) {
6b6833c1 3631 /* keep backing_filename NULL */
1cb6f506 3632 } else if (bs->backing_file[0] == '\0' && qdict_size(options) == 0) {
cb3e7f08 3633 qobject_unref(options);
1ba4b6a5 3634 goto free_exit;
dbecebdd 3635 } else {
998c2019
HR
3636 if (qdict_size(options) == 0) {
3637 /* If the user specifies options that do not modify the
3638 * backing file's behavior, we might still consider it the
3639 * implicit backing file. But it's easier this way, and
3640 * just specifying some of the backing BDS's options is
3641 * only possible with -drive anyway (otherwise the QAPI
3642 * schema forces the user to specify everything). */
3643 implicit_backing = !strcmp(bs->auto_backing_file, bs->backing_file);
3644 }
3645
6b6833c1 3646 backing_filename = bdrv_get_full_backing_filename(bs, &local_err);
9f07429e
HR
3647 if (local_err) {
3648 ret = -EINVAL;
3649 error_propagate(errp, local_err);
cb3e7f08 3650 qobject_unref(options);
9f07429e
HR
3651 goto free_exit;
3652 }
9156df12
PB
3653 }
3654
8ee79e70
KW
3655 if (!bs->drv || !bs->drv->supports_backing) {
3656 ret = -EINVAL;
3657 error_setg(errp, "Driver doesn't support backing files");
cb3e7f08 3658 qobject_unref(options);
8ee79e70
KW
3659 goto free_exit;
3660 }
3661
6bff597b
PK
3662 if (!reference &&
3663 bs->backing_format[0] != '\0' && !qdict_haskey(options, "driver")) {
46f5ac20 3664 qdict_put_str(options, "driver", bs->backing_format);
9156df12
PB
3665 }
3666
6b6833c1 3667 backing_hd = bdrv_open_inherit(backing_filename, reference, options, 0, bs,
25191e5f 3668 &child_of_bds, bdrv_backing_role(bs), errp);
5b363937 3669 if (!backing_hd) {
9156df12 3670 bs->open_flags |= BDRV_O_NO_BACKING;
e43bfd9c 3671 error_prepend(errp, "Could not open backing file: ");
5b363937 3672 ret = -EINVAL;
1ba4b6a5 3673 goto free_exit;
9156df12 3674 }
df581792 3675
998c2019
HR
3676 if (implicit_backing) {
3677 bdrv_refresh_filename(backing_hd);
3678 pstrcpy(bs->auto_backing_file, sizeof(bs->auto_backing_file),
3679 backing_hd->filename);
3680 }
3681
5db15a57
KW
3682 /* Hook up the backing file link; drop our reference, bs owns the
3683 * backing_hd reference now */
8aa04542
KW
3684 backing_hd_ctx = bdrv_get_aio_context(backing_hd);
3685 aio_context_acquire(backing_hd_ctx);
dc9c10a1 3686 ret = bdrv_set_backing_hd(bs, backing_hd, errp);
5db15a57 3687 bdrv_unref(backing_hd);
8aa04542
KW
3688 aio_context_release(backing_hd_ctx);
3689
dc9c10a1 3690 if (ret < 0) {
12fa4af6
KW
3691 goto free_exit;
3692 }
d80ac658 3693
d9b7b057
KW
3694 qdict_del(parent_options, bdref_key);
3695
1ba4b6a5
BC
3696free_exit:
3697 g_free(backing_filename);
cb3e7f08 3698 qobject_unref(tmp_parent_options);
1ba4b6a5 3699 return ret;
9156df12
PB
3700}
3701
2d6b86af
KW
3702static BlockDriverState *
3703bdrv_open_child_bs(const char *filename, QDict *options, const char *bdref_key,
bd86fb99 3704 BlockDriverState *parent, const BdrvChildClass *child_class,
272c02ea 3705 BdrvChildRole child_role, bool allow_none, Error **errp)
da557aac 3706{
2d6b86af 3707 BlockDriverState *bs = NULL;
da557aac 3708 QDict *image_options;
da557aac
HR
3709 char *bdref_key_dot;
3710 const char *reference;
3711
bd86fb99 3712 assert(child_class != NULL);
f67503e5 3713
da557aac
HR
3714 bdref_key_dot = g_strdup_printf("%s.", bdref_key);
3715 qdict_extract_subqdict(options, &image_options, bdref_key_dot);
3716 g_free(bdref_key_dot);
3717
129c7d1c
MA
3718 /*
3719 * Caution: while qdict_get_try_str() is fine, getting non-string
3720 * types would require more care. When @options come from
3721 * -blockdev or blockdev_add, its members are typed according to
3722 * the QAPI schema, but when they come from -drive, they're all
3723 * QString.
3724 */
da557aac
HR
3725 reference = qdict_get_try_str(options, bdref_key);
3726 if (!filename && !reference && !qdict_size(image_options)) {
b4b059f6 3727 if (!allow_none) {
da557aac
HR
3728 error_setg(errp, "A block device must be specified for \"%s\"",
3729 bdref_key);
da557aac 3730 }
cb3e7f08 3731 qobject_unref(image_options);
da557aac
HR
3732 goto done;
3733 }
3734
5b363937 3735 bs = bdrv_open_inherit(filename, reference, image_options, 0,
272c02ea 3736 parent, child_class, child_role, errp);
5b363937 3737 if (!bs) {
df581792
KW
3738 goto done;
3739 }
3740
da557aac
HR
3741done:
3742 qdict_del(options, bdref_key);
2d6b86af
KW
3743 return bs;
3744}
3745
3746/*
3747 * Opens a disk image whose options are given as BlockdevRef in another block
3748 * device's options.
3749 *
3750 * If allow_none is true, no image will be opened if filename is false and no
3751 * BlockdevRef is given. NULL will be returned, but errp remains unset.
3752 *
3753 * bdrev_key specifies the key for the image's BlockdevRef in the options QDict.
3754 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
3755 * itself, all options starting with "${bdref_key}." are considered part of the
3756 * BlockdevRef.
3757 *
3758 * The BlockdevRef will be removed from the options QDict.
aa269ff8 3759 *
8394c35e 3760 * The caller must hold the lock of the main AioContext and no other AioContext.
aa269ff8
KW
3761 * @parent can move to a different AioContext in this function. Callers must
3762 * make sure that their AioContext locking is still correct after this.
2d6b86af
KW
3763 */
3764BdrvChild *bdrv_open_child(const char *filename,
3765 QDict *options, const char *bdref_key,
3766 BlockDriverState *parent,
bd86fb99 3767 const BdrvChildClass *child_class,
258b7765 3768 BdrvChildRole child_role,
2d6b86af
KW
3769 bool allow_none, Error **errp)
3770{
3771 BlockDriverState *bs;
8394c35e
KW
3772 BdrvChild *child;
3773 AioContext *ctx;
2d6b86af 3774
f791bf7f
EGE
3775 GLOBAL_STATE_CODE();
3776
bd86fb99 3777 bs = bdrv_open_child_bs(filename, options, bdref_key, parent, child_class,
272c02ea 3778 child_role, allow_none, errp);
2d6b86af
KW
3779 if (bs == NULL) {
3780 return NULL;
3781 }
3782
afdaeb9e 3783 bdrv_graph_wrlock(NULL);
8394c35e
KW
3784 ctx = bdrv_get_aio_context(bs);
3785 aio_context_acquire(ctx);
3786 child = bdrv_attach_child(parent, bs, bdref_key, child_class, child_role,
3787 errp);
3788 aio_context_release(ctx);
afdaeb9e 3789 bdrv_graph_wrunlock();
8394c35e
KW
3790
3791 return child;
b4b059f6
KW
3792}
3793
83930780
VSO
3794/*
3795 * Wrapper on bdrv_open_child() for most popular case: open primary child of bs.
aa269ff8 3796 *
8394c35e 3797 * The caller must hold the lock of the main AioContext and no other AioContext.
aa269ff8
KW
3798 * @parent can move to a different AioContext in this function. Callers must
3799 * make sure that their AioContext locking is still correct after this.
83930780
VSO
3800 */
3801int bdrv_open_file_child(const char *filename,
3802 QDict *options, const char *bdref_key,
3803 BlockDriverState *parent, Error **errp)
3804{
3805 BdrvChildRole role;
3806
3807 /* commit_top and mirror_top don't use this function */
3808 assert(!parent->drv->filtered_child_is_backing);
83930780
VSO
3809 role = parent->drv->is_filter ?
3810 (BDRV_CHILD_FILTERED | BDRV_CHILD_PRIMARY) : BDRV_CHILD_IMAGE;
3811
5bb04747
VSO
3812 if (!bdrv_open_child(filename, options, bdref_key, parent,
3813 &child_of_bds, role, false, errp))
3814 {
3815 return -EINVAL;
3816 }
83930780 3817
5bb04747 3818 return 0;
83930780
VSO
3819}
3820
bd86fb99
HR
3821/*
3822 * TODO Future callers may need to specify parent/child_class in order for
3823 * option inheritance to work. Existing callers use it for the root node.
3824 */
e1d74bc6
KW
3825BlockDriverState *bdrv_open_blockdev_ref(BlockdevRef *ref, Error **errp)
3826{
3827 BlockDriverState *bs = NULL;
e1d74bc6
KW
3828 QObject *obj = NULL;
3829 QDict *qdict = NULL;
3830 const char *reference = NULL;
3831 Visitor *v = NULL;
3832
f791bf7f
EGE
3833 GLOBAL_STATE_CODE();
3834
e1d74bc6
KW
3835 if (ref->type == QTYPE_QSTRING) {
3836 reference = ref->u.reference;
3837 } else {
3838 BlockdevOptions *options = &ref->u.definition;
3839 assert(ref->type == QTYPE_QDICT);
3840
3841 v = qobject_output_visitor_new(&obj);
1f584248 3842 visit_type_BlockdevOptions(v, NULL, &options, &error_abort);
e1d74bc6
KW
3843 visit_complete(v, &obj);
3844
7dc847eb 3845 qdict = qobject_to(QDict, obj);
e1d74bc6
KW
3846 qdict_flatten(qdict);
3847
3848 /* bdrv_open_inherit() defaults to the values in bdrv_flags (for
3849 * compatibility with other callers) rather than what we want as the
3850 * real defaults. Apply the defaults here instead. */
3851 qdict_set_default_str(qdict, BDRV_OPT_CACHE_DIRECT, "off");
3852 qdict_set_default_str(qdict, BDRV_OPT_CACHE_NO_FLUSH, "off");
3853 qdict_set_default_str(qdict, BDRV_OPT_READ_ONLY, "off");
e35bdc12
KW
3854 qdict_set_default_str(qdict, BDRV_OPT_AUTO_READ_ONLY, "off");
3855
e1d74bc6
KW
3856 }
3857
272c02ea 3858 bs = bdrv_open_inherit(NULL, reference, qdict, 0, NULL, NULL, 0, errp);
e1d74bc6 3859 obj = NULL;
cb3e7f08 3860 qobject_unref(obj);
e1d74bc6
KW
3861 visit_free(v);
3862 return bs;
3863}
3864
66836189
HR
3865static BlockDriverState *bdrv_append_temp_snapshot(BlockDriverState *bs,
3866 int flags,
3867 QDict *snapshot_options,
3868 Error **errp)
b998875d 3869{
69fbfff9 3870 g_autofree char *tmp_filename = NULL;
b998875d 3871 int64_t total_size;
83d0521a 3872 QemuOpts *opts = NULL;
ff6ed714 3873 BlockDriverState *bs_snapshot = NULL;
f665f01f 3874 AioContext *ctx = bdrv_get_aio_context(bs);
b998875d
KW
3875 int ret;
3876
bdb73476
EGE
3877 GLOBAL_STATE_CODE();
3878
b998875d
KW
3879 /* if snapshot, we create a temporary backing file and open it
3880 instead of opening 'filename' directly */
3881
3882 /* Get the required size from the image */
f665f01f 3883 aio_context_acquire(ctx);
f187743a 3884 total_size = bdrv_getlength(bs);
f665f01f
KW
3885 aio_context_release(ctx);
3886
f187743a
KW
3887 if (total_size < 0) {
3888 error_setg_errno(errp, -total_size, "Could not get image size");
1ba4b6a5 3889 goto out;
f187743a 3890 }
b998875d
KW
3891
3892 /* Create the temporary image */
69fbfff9
BM
3893 tmp_filename = create_tmp_file(errp);
3894 if (!tmp_filename) {
1ba4b6a5 3895 goto out;
b998875d
KW
3896 }
3897
ef810437 3898 opts = qemu_opts_create(bdrv_qcow2.create_opts, NULL, 0,
c282e1fd 3899 &error_abort);
39101f25 3900 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, total_size, &error_abort);
e43bfd9c 3901 ret = bdrv_create(&bdrv_qcow2, tmp_filename, opts, errp);
83d0521a 3902 qemu_opts_del(opts);
b998875d 3903 if (ret < 0) {
e43bfd9c
MA
3904 error_prepend(errp, "Could not create temporary overlay '%s': ",
3905 tmp_filename);
1ba4b6a5 3906 goto out;
b998875d
KW
3907 }
3908
73176bee 3909 /* Prepare options QDict for the temporary file */
46f5ac20
EB
3910 qdict_put_str(snapshot_options, "file.driver", "file");
3911 qdict_put_str(snapshot_options, "file.filename", tmp_filename);
3912 qdict_put_str(snapshot_options, "driver", "qcow2");
b998875d 3913
5b363937 3914 bs_snapshot = bdrv_open(NULL, NULL, snapshot_options, flags, errp);
73176bee 3915 snapshot_options = NULL;
5b363937 3916 if (!bs_snapshot) {
1ba4b6a5 3917 goto out;
b998875d
KW
3918 }
3919
f665f01f 3920 aio_context_acquire(ctx);
934aee14 3921 ret = bdrv_append(bs_snapshot, bs, errp);
f665f01f
KW
3922 aio_context_release(ctx);
3923
934aee14 3924 if (ret < 0) {
ff6ed714 3925 bs_snapshot = NULL;
b2c2832c
KW
3926 goto out;
3927 }
1ba4b6a5
BC
3928
3929out:
cb3e7f08 3930 qobject_unref(snapshot_options);
ff6ed714 3931 return bs_snapshot;
b998875d
KW
3932}
3933
b6ce07aa
KW
3934/*
3935 * Opens a disk image (raw, qcow2, vmdk, ...)
de9c0cec
KW
3936 *
3937 * options is a QDict of options to pass to the block drivers, or NULL for an
3938 * empty set of options. The reference to the QDict belongs to the block layer
3939 * after the call (even on failure), so if the caller intends to reuse the
cb3e7f08 3940 * dictionary, it needs to use qobject_ref() before calling bdrv_open.
f67503e5
HR
3941 *
3942 * If *pbs is NULL, a new BDS will be created with a pointer to it stored there.
3943 * If it is not NULL, the referenced BDS will be reused.
ddf5636d
HR
3944 *
3945 * The reference parameter may be used to specify an existing block device which
3946 * should be opened. If specified, neither options nor a filename may be given,
3947 * nor can an existing BDS be reused (that is, *pbs has to be NULL).
c86422c5 3948 *
ae400dbb 3949 * The caller must always hold the main AioContext lock.
b6ce07aa 3950 */
32192301
KW
3951static BlockDriverState * no_coroutine_fn
3952bdrv_open_inherit(const char *filename, const char *reference, QDict *options,
3953 int flags, BlockDriverState *parent,
3954 const BdrvChildClass *child_class, BdrvChildRole child_role,
3955 Error **errp)
ea2384d3 3956{
b6ce07aa 3957 int ret;
5696c6e3 3958 BlockBackend *file = NULL;
9a4f4c31 3959 BlockDriverState *bs;
ce343771 3960 BlockDriver *drv = NULL;
2f624b80 3961 BdrvChild *child;
74fe54f2 3962 const char *drvname;
3e8c2e57 3963 const char *backing;
34b5d2c6 3964 Error *local_err = NULL;
73176bee 3965 QDict *snapshot_options = NULL;
b1e6fc08 3966 int snapshot_flags = 0;
f665f01f 3967 AioContext *ctx = qemu_get_aio_context();
712e7874 3968
bd86fb99
HR
3969 assert(!child_class || !flags);
3970 assert(!child_class == !parent);
f0c28327 3971 GLOBAL_STATE_CODE();
32192301 3972 assert(!qemu_in_coroutine());
f67503e5 3973
356f4ef6
KW
3974 /* TODO We'll eventually have to take a writer lock in this function */
3975 GRAPH_RDLOCK_GUARD_MAINLOOP();
3976
ddf5636d
HR
3977 if (reference) {
3978 bool options_non_empty = options ? qdict_size(options) : false;
cb3e7f08 3979 qobject_unref(options);
ddf5636d 3980
ddf5636d
HR
3981 if (filename || options_non_empty) {
3982 error_setg(errp, "Cannot reference an existing block device with "
3983 "additional options or a new filename");
5b363937 3984 return NULL;
ddf5636d
HR
3985 }
3986
3987 bs = bdrv_lookup_bs(reference, reference, errp);
3988 if (!bs) {
5b363937 3989 return NULL;
ddf5636d 3990 }
76b22320 3991
ddf5636d 3992 bdrv_ref(bs);
5b363937 3993 return bs;
ddf5636d
HR
3994 }
3995
5b363937 3996 bs = bdrv_new();
f67503e5 3997
de9c0cec
KW
3998 /* NULL means an empty set of options */
3999 if (options == NULL) {
4000 options = qdict_new();
4001 }
4002
145f598e 4003 /* json: syntax counts as explicit options, as if in the QDict */
de3b53f0
KW
4004 parse_json_protocol(options, &filename, &local_err);
4005 if (local_err) {
de3b53f0
KW
4006 goto fail;
4007 }
4008
145f598e
KW
4009 bs->explicit_options = qdict_clone_shallow(options);
4010
bd86fb99 4011 if (child_class) {
3cdc69d3
HR
4012 bool parent_is_format;
4013
4014 if (parent->drv) {
4015 parent_is_format = parent->drv->is_format;
4016 } else {
4017 /*
4018 * parent->drv is not set yet because this node is opened for
4019 * (potential) format probing. That means that @parent is going
4020 * to be a format node.
4021 */
4022 parent_is_format = true;
4023 }
4024
bddcec37 4025 bs->inherits_from = parent;
3cdc69d3
HR
4026 child_class->inherit_options(child_role, parent_is_format,
4027 &flags, options,
bd86fb99 4028 parent->open_flags, parent->options);
f3930ed0
KW
4029 }
4030
de3b53f0 4031 ret = bdrv_fill_options(&options, filename, &flags, &local_err);
dfde483e 4032 if (ret < 0) {
462f5bcf
KW
4033 goto fail;
4034 }
4035
129c7d1c
MA
4036 /*
4037 * Set the BDRV_O_RDWR and BDRV_O_ALLOW_RDWR flags.
4038 * Caution: getting a boolean member of @options requires care.
4039 * When @options come from -blockdev or blockdev_add, members are
4040 * typed according to the QAPI schema, but when they come from
4041 * -drive, they're all QString.
4042 */
f87a0e29
AG
4043 if (g_strcmp0(qdict_get_try_str(options, BDRV_OPT_READ_ONLY), "on") &&
4044 !qdict_get_try_bool(options, BDRV_OPT_READ_ONLY, false)) {
4045 flags |= (BDRV_O_RDWR | BDRV_O_ALLOW_RDWR);
4046 } else {
4047 flags &= ~BDRV_O_RDWR;
14499ea5
AG
4048 }
4049
4050 if (flags & BDRV_O_SNAPSHOT) {
4051 snapshot_options = qdict_new();
4052 bdrv_temp_snapshot_options(&snapshot_flags, snapshot_options,
4053 flags, options);
f87a0e29
AG
4054 /* Let bdrv_backing_options() override "read-only" */
4055 qdict_del(options, BDRV_OPT_READ_ONLY);
00ff7ffd
HR
4056 bdrv_inherited_options(BDRV_CHILD_COW, true,
4057 &flags, options, flags, options);
14499ea5
AG
4058 }
4059
62392ebb
KW
4060 bs->open_flags = flags;
4061 bs->options = options;
4062 options = qdict_clone_shallow(options);
4063
76c591b0 4064 /* Find the right image format driver */
129c7d1c 4065 /* See cautionary note on accessing @options above */
76c591b0
KW
4066 drvname = qdict_get_try_str(options, "driver");
4067 if (drvname) {
4068 drv = bdrv_find_format(drvname);
76c591b0
KW
4069 if (!drv) {
4070 error_setg(errp, "Unknown driver: '%s'", drvname);
76c591b0
KW
4071 goto fail;
4072 }
4073 }
4074
4075 assert(drvname || !(flags & BDRV_O_PROTOCOL));
76c591b0 4076
129c7d1c 4077 /* See cautionary note on accessing @options above */
3e8c2e57 4078 backing = qdict_get_try_str(options, "backing");
e59a0cf1
HR
4079 if (qobject_to(QNull, qdict_get(options, "backing")) != NULL ||
4080 (backing && *backing == '\0'))
4081 {
4f7be280
HR
4082 if (backing) {
4083 warn_report("Use of \"backing\": \"\" is deprecated; "
4084 "use \"backing\": null instead");
4085 }
3e8c2e57 4086 flags |= BDRV_O_NO_BACKING;
ae0f57f0
KW
4087 qdict_del(bs->explicit_options, "backing");
4088 qdict_del(bs->options, "backing");
3e8c2e57
AG
4089 qdict_del(options, "backing");
4090 }
4091
5696c6e3 4092 /* Open image file without format layer. This BlockBackend is only used for
4e4bf5c4
KW
4093 * probing, the block drivers will do their own bdrv_open_child() for the
4094 * same BDS, which is why we put the node name back into options. */
f4788adc 4095 if ((flags & BDRV_O_PROTOCOL) == 0) {
5696c6e3
KW
4096 BlockDriverState *file_bs;
4097
4098 file_bs = bdrv_open_child_bs(filename, options, "file", bs,
58944401
HR
4099 &child_of_bds, BDRV_CHILD_IMAGE,
4100 true, &local_err);
1fdd6933 4101 if (local_err) {
f4788adc
KW
4102 goto fail;
4103 }
5696c6e3 4104 if (file_bs != NULL) {
dacaa162
KW
4105 /* Not requesting BLK_PERM_CONSISTENT_READ because we're only
4106 * looking at the header to guess the image format. This works even
4107 * in cases where a guest would not see a consistent state. */
f665f01f
KW
4108 ctx = bdrv_get_aio_context(file_bs);
4109 aio_context_acquire(ctx);
4110 file = blk_new(ctx, 0, BLK_PERM_ALL);
d7086422 4111 blk_insert_bs(file, file_bs, &local_err);
5696c6e3 4112 bdrv_unref(file_bs);
f665f01f
KW
4113 aio_context_release(ctx);
4114
d7086422
KW
4115 if (local_err) {
4116 goto fail;
4117 }
5696c6e3 4118
46f5ac20 4119 qdict_put_str(options, "file", bdrv_get_node_name(file_bs));
4e4bf5c4 4120 }
f500a6d3
KW
4121 }
4122
76c591b0 4123 /* Image format probing */
38f3ef57 4124 bs->probed = !drv;
76c591b0 4125 if (!drv && file) {
cf2ab8fc 4126 ret = find_image_format(file, filename, &drv, &local_err);
17b005f1 4127 if (ret < 0) {
8bfea15d 4128 goto fail;
2a05cbe4 4129 }
62392ebb
KW
4130 /*
4131 * This option update would logically belong in bdrv_fill_options(),
4132 * but we first need to open bs->file for the probing to work, while
4133 * opening bs->file already requires the (mostly) final set of options
4134 * so that cache mode etc. can be inherited.
4135 *
4136 * Adding the driver later is somewhat ugly, but it's not an option
4137 * that would ever be inherited, so it's correct. We just need to make
4138 * sure to update both bs->options (which has the full effective
4139 * options for bs) and options (which has file.* already removed).
4140 */
46f5ac20
EB
4141 qdict_put_str(bs->options, "driver", drv->format_name);
4142 qdict_put_str(options, "driver", drv->format_name);
76c591b0 4143 } else if (!drv) {
17b005f1 4144 error_setg(errp, "Must specify either driver or file");
8bfea15d 4145 goto fail;
ea2384d3 4146 }
b6ce07aa 4147
53a29513
HR
4148 /* BDRV_O_PROTOCOL must be set iff a protocol BDS is about to be created */
4149 assert(!!(flags & BDRV_O_PROTOCOL) == !!drv->bdrv_file_open);
4150 /* file must be NULL if a protocol BDS is about to be created
4151 * (the inverse results in an error message from bdrv_open_common()) */
4152 assert(!(flags & BDRV_O_PROTOCOL) || !file);
4153
b6ce07aa 4154 /* Open the image */
82dc8b41 4155 ret = bdrv_open_common(bs, file, options, &local_err);
b6ce07aa 4156 if (ret < 0) {
8bfea15d 4157 goto fail;
6987307c
CH
4158 }
4159
f665f01f
KW
4160 /* The AioContext could have changed during bdrv_open_common() */
4161 ctx = bdrv_get_aio_context(bs);
4162
4e4bf5c4 4163 if (file) {
f665f01f 4164 aio_context_acquire(ctx);
5696c6e3 4165 blk_unref(file);
f665f01f 4166 aio_context_release(ctx);
f500a6d3
KW
4167 file = NULL;
4168 }
4169
b6ce07aa 4170 /* If there is a backing file, use it */
9156df12 4171 if ((flags & BDRV_O_NO_BACKING) == 0) {
d9b7b057 4172 ret = bdrv_open_backing_file(bs, options, "backing", &local_err);
b6ce07aa 4173 if (ret < 0) {
b6ad491a 4174 goto close_and_fail;
b6ce07aa 4175 }
b6ce07aa
KW
4176 }
4177
50196d7a
AG
4178 /* Remove all children options and references
4179 * from bs->options and bs->explicit_options */
2f624b80
AG
4180 QLIST_FOREACH(child, &bs->children, next) {
4181 char *child_key_dot;
4182 child_key_dot = g_strdup_printf("%s.", child->name);
4183 qdict_extract_subqdict(bs->explicit_options, NULL, child_key_dot);
4184 qdict_extract_subqdict(bs->options, NULL, child_key_dot);
50196d7a
AG
4185 qdict_del(bs->explicit_options, child->name);
4186 qdict_del(bs->options, child->name);
2f624b80
AG
4187 g_free(child_key_dot);
4188 }
4189
b6ad491a 4190 /* Check if any unknown options were used */
7ad2757f 4191 if (qdict_size(options) != 0) {
b6ad491a 4192 const QDictEntry *entry = qdict_first(options);
5acd9d81
HR
4193 if (flags & BDRV_O_PROTOCOL) {
4194 error_setg(errp, "Block protocol '%s' doesn't support the option "
4195 "'%s'", drv->format_name, entry->key);
4196 } else {
d0e46a55
HR
4197 error_setg(errp,
4198 "Block format '%s' does not support the option '%s'",
4199 drv->format_name, entry->key);
5acd9d81 4200 }
b6ad491a 4201
b6ad491a
KW
4202 goto close_and_fail;
4203 }
b6ad491a 4204
c01c214b 4205 bdrv_parent_cb_change_media(bs, true);
b6ce07aa 4206
cb3e7f08 4207 qobject_unref(options);
8961be33 4208 options = NULL;
dd62f1ca
KW
4209
4210 /* For snapshot=on, create a temporary qcow2 overlay. bs points to the
4211 * temporary snapshot afterwards. */
4212 if (snapshot_flags) {
66836189
HR
4213 BlockDriverState *snapshot_bs;
4214 snapshot_bs = bdrv_append_temp_snapshot(bs, snapshot_flags,
4215 snapshot_options, &local_err);
73176bee 4216 snapshot_options = NULL;
dd62f1ca
KW
4217 if (local_err) {
4218 goto close_and_fail;
4219 }
5b363937
HR
4220 /* We are not going to return bs but the overlay on top of it
4221 * (snapshot_bs); thus, we have to drop the strong reference to bs
4222 * (which we obtained by calling bdrv_new()). bs will not be deleted,
4223 * though, because the overlay still has a reference to it. */
f665f01f 4224 aio_context_acquire(ctx);
5b363937 4225 bdrv_unref(bs);
f665f01f 4226 aio_context_release(ctx);
5b363937 4227 bs = snapshot_bs;
dd62f1ca
KW
4228 }
4229
5b363937 4230 return bs;
b6ce07aa 4231
8bfea15d 4232fail:
f665f01f 4233 aio_context_acquire(ctx);
5696c6e3 4234 blk_unref(file);
cb3e7f08
MAL
4235 qobject_unref(snapshot_options);
4236 qobject_unref(bs->explicit_options);
4237 qobject_unref(bs->options);
4238 qobject_unref(options);
de9c0cec 4239 bs->options = NULL;
998cbd6a 4240 bs->explicit_options = NULL;
5b363937 4241 bdrv_unref(bs);
f665f01f 4242 aio_context_release(ctx);
621ff94d 4243 error_propagate(errp, local_err);
5b363937 4244 return NULL;
de9c0cec 4245
b6ad491a 4246close_and_fail:
f665f01f 4247 aio_context_acquire(ctx);
5b363937 4248 bdrv_unref(bs);
f665f01f 4249 aio_context_release(ctx);
cb3e7f08
MAL
4250 qobject_unref(snapshot_options);
4251 qobject_unref(options);
621ff94d 4252 error_propagate(errp, local_err);
5b363937 4253 return NULL;
b6ce07aa
KW
4254}
4255
ae400dbb 4256/* The caller must always hold the main AioContext lock. */
5b363937
HR
4257BlockDriverState *bdrv_open(const char *filename, const char *reference,
4258 QDict *options, int flags, Error **errp)
f3930ed0 4259{
f791bf7f
EGE
4260 GLOBAL_STATE_CODE();
4261
5b363937 4262 return bdrv_open_inherit(filename, reference, options, flags, NULL,
272c02ea 4263 NULL, 0, errp);
f3930ed0
KW
4264}
4265
faf116b4
AG
4266/* Return true if the NULL-terminated @list contains @str */
4267static bool is_str_in_list(const char *str, const char *const *list)
4268{
4269 if (str && list) {
4270 int i;
4271 for (i = 0; list[i] != NULL; i++) {
4272 if (!strcmp(str, list[i])) {
4273 return true;
4274 }
4275 }
4276 }
4277 return false;
4278}
4279
4280/*
4281 * Check that every option set in @bs->options is also set in
4282 * @new_opts.
4283 *
4284 * Options listed in the common_options list and in
4285 * @bs->drv->mutable_opts are skipped.
4286 *
4287 * Return 0 on success, otherwise return -EINVAL and set @errp.
4288 */
4289static int bdrv_reset_options_allowed(BlockDriverState *bs,
4290 const QDict *new_opts, Error **errp)
4291{
4292 const QDictEntry *e;
4293 /* These options are common to all block drivers and are handled
4294 * in bdrv_reopen_prepare() so they can be left out of @new_opts */
4295 const char *const common_options[] = {
4296 "node-name", "discard", "cache.direct", "cache.no-flush",
4297 "read-only", "auto-read-only", "detect-zeroes", NULL
4298 };
4299
4300 for (e = qdict_first(bs->options); e; e = qdict_next(bs->options, e)) {
4301 if (!qdict_haskey(new_opts, e->key) &&
4302 !is_str_in_list(e->key, common_options) &&
4303 !is_str_in_list(e->key, bs->drv->mutable_opts)) {
4304 error_setg(errp, "Option '%s' cannot be reset "
4305 "to its default value", e->key);
4306 return -EINVAL;
4307 }
4308 }
4309
4310 return 0;
4311}
4312
cb828c31
AG
4313/*
4314 * Returns true if @child can be reached recursively from @bs
4315 */
4316static bool bdrv_recurse_has_child(BlockDriverState *bs,
4317 BlockDriverState *child)
4318{
4319 BdrvChild *c;
4320
4321 if (bs == child) {
4322 return true;
4323 }
4324
4325 QLIST_FOREACH(c, &bs->children, next) {
4326 if (bdrv_recurse_has_child(c->bs, child)) {
4327 return true;
4328 }
4329 }
4330
4331 return false;
4332}
4333
e971aa12
JC
4334/*
4335 * Adds a BlockDriverState to a simple queue for an atomic, transactional
4336 * reopen of multiple devices.
4337 *
859443b0 4338 * bs_queue can either be an existing BlockReopenQueue that has had QTAILQ_INIT
e971aa12
JC
4339 * already performed, or alternatively may be NULL a new BlockReopenQueue will
4340 * be created and initialized. This newly created BlockReopenQueue should be
4341 * passed back in for subsequent calls that are intended to be of the same
4342 * atomic 'set'.
4343 *
4344 * bs is the BlockDriverState to add to the reopen queue.
4345 *
4d2cb092
KW
4346 * options contains the changed options for the associated bs
4347 * (the BlockReopenQueue takes ownership)
4348 *
e971aa12
JC
4349 * flags contains the open flags for the associated bs
4350 *
4351 * returns a pointer to bs_queue, which is either the newly allocated
4352 * bs_queue, or the existing bs_queue being used.
4353 *
d22933ac 4354 * bs is drained here and undrained by bdrv_reopen_queue_free().
2e117866
KW
4355 *
4356 * To be called with bs->aio_context locked.
e971aa12 4357 */
28518102
KW
4358static BlockReopenQueue *bdrv_reopen_queue_child(BlockReopenQueue *bs_queue,
4359 BlockDriverState *bs,
4360 QDict *options,
bd86fb99 4361 const BdrvChildClass *klass,
272c02ea 4362 BdrvChildRole role,
3cdc69d3 4363 bool parent_is_format,
28518102 4364 QDict *parent_options,
077e8e20
AG
4365 int parent_flags,
4366 bool keep_old_opts)
e971aa12
JC
4367{
4368 assert(bs != NULL);
4369
4370 BlockReopenQueueEntry *bs_entry;
67251a31 4371 BdrvChild *child;
9aa09ddd
AG
4372 QDict *old_options, *explicit_options, *options_copy;
4373 int flags;
4374 QemuOpts *opts;
67251a31 4375
f0c28327 4376 GLOBAL_STATE_CODE();
1a63a907 4377
d22933ac
KW
4378 bdrv_drained_begin(bs);
4379
e971aa12
JC
4380 if (bs_queue == NULL) {
4381 bs_queue = g_new0(BlockReopenQueue, 1);
859443b0 4382 QTAILQ_INIT(bs_queue);
e971aa12
JC
4383 }
4384
4d2cb092
KW
4385 if (!options) {
4386 options = qdict_new();
4387 }
4388
5b7ba05f 4389 /* Check if this BlockDriverState is already in the queue */
859443b0 4390 QTAILQ_FOREACH(bs_entry, bs_queue, entry) {
5b7ba05f
AG
4391 if (bs == bs_entry->state.bs) {
4392 break;
4393 }
4394 }
4395
28518102
KW
4396 /*
4397 * Precedence of options:
4398 * 1. Explicitly passed in options (highest)
9aa09ddd
AG
4399 * 2. Retained from explicitly set options of bs
4400 * 3. Inherited from parent node
4401 * 4. Retained from effective options of bs
28518102
KW
4402 */
4403
145f598e 4404 /* Old explicitly set values (don't overwrite by inherited value) */
077e8e20
AG
4405 if (bs_entry || keep_old_opts) {
4406 old_options = qdict_clone_shallow(bs_entry ?
4407 bs_entry->state.explicit_options :
4408 bs->explicit_options);
4409 bdrv_join_options(bs, options, old_options);
4410 qobject_unref(old_options);
5b7ba05f 4411 }
145f598e
KW
4412
4413 explicit_options = qdict_clone_shallow(options);
4414
28518102
KW
4415 /* Inherit from parent node */
4416 if (parent_options) {
9aa09ddd 4417 flags = 0;
3cdc69d3 4418 klass->inherit_options(role, parent_is_format, &flags, options,
272c02ea 4419 parent_flags, parent_options);
9aa09ddd
AG
4420 } else {
4421 flags = bdrv_get_flags(bs);
28518102
KW
4422 }
4423
077e8e20
AG
4424 if (keep_old_opts) {
4425 /* Old values are used for options that aren't set yet */
4426 old_options = qdict_clone_shallow(bs->options);
4427 bdrv_join_options(bs, options, old_options);
4428 qobject_unref(old_options);
4429 }
4d2cb092 4430
9aa09ddd
AG
4431 /* We have the final set of options so let's update the flags */
4432 options_copy = qdict_clone_shallow(options);
4433 opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
4434 qemu_opts_absorb_qdict(opts, options_copy, NULL);
4435 update_flags_from_options(&flags, opts);
4436 qemu_opts_del(opts);
4437 qobject_unref(options_copy);
4438
fd452021 4439 /* bdrv_open_inherit() sets and clears some additional flags internally */
f1f25a2e 4440 flags &= ~BDRV_O_PROTOCOL;
fd452021
KW
4441 if (flags & BDRV_O_RDWR) {
4442 flags |= BDRV_O_ALLOW_RDWR;
4443 }
f1f25a2e 4444
1857c97b
KW
4445 if (!bs_entry) {
4446 bs_entry = g_new0(BlockReopenQueueEntry, 1);
859443b0 4447 QTAILQ_INSERT_TAIL(bs_queue, bs_entry, entry);
1857c97b 4448 } else {
cb3e7f08
MAL
4449 qobject_unref(bs_entry->state.options);
4450 qobject_unref(bs_entry->state.explicit_options);
1857c97b
KW
4451 }
4452
4453 bs_entry->state.bs = bs;
4454 bs_entry->state.options = options;
4455 bs_entry->state.explicit_options = explicit_options;
4456 bs_entry->state.flags = flags;
4457
8546632e
AG
4458 /*
4459 * If keep_old_opts is false then it means that unspecified
4460 * options must be reset to their original value. We don't allow
4461 * resetting 'backing' but we need to know if the option is
4462 * missing in order to decide if we have to return an error.
4463 */
4464 if (!keep_old_opts) {
4465 bs_entry->state.backing_missing =
4466 !qdict_haskey(options, "backing") &&
4467 !qdict_haskey(options, "backing.driver");
4468 }
4469
67251a31 4470 QLIST_FOREACH(child, &bs->children, next) {
8546632e
AG
4471 QDict *new_child_options = NULL;
4472 bool child_keep_old = keep_old_opts;
67251a31 4473
4c9dfe5d
KW
4474 /* reopen can only change the options of block devices that were
4475 * implicitly created and inherited options. For other (referenced)
4476 * block devices, a syntax like "backing.foo" results in an error. */
67251a31
KW
4477 if (child->bs->inherits_from != bs) {
4478 continue;
4479 }
4480
8546632e
AG
4481 /* Check if the options contain a child reference */
4482 if (qdict_haskey(options, child->name)) {
4483 const char *childref = qdict_get_try_str(options, child->name);
4484 /*
4485 * The current child must not be reopened if the child
4486 * reference is null or points to a different node.
4487 */
4488 if (g_strcmp0(childref, child->bs->node_name)) {
4489 continue;
4490 }
4491 /*
4492 * If the child reference points to the current child then
4493 * reopen it with its existing set of options (note that
4494 * it can still inherit new options from the parent).
4495 */
4496 child_keep_old = true;
4497 } else {
4498 /* Extract child options ("child-name.*") */
4499 char *child_key_dot = g_strdup_printf("%s.", child->name);
4500 qdict_extract_subqdict(explicit_options, NULL, child_key_dot);
4501 qdict_extract_subqdict(options, &new_child_options, child_key_dot);
4502 g_free(child_key_dot);
4503 }
4c9dfe5d 4504
9aa09ddd 4505 bdrv_reopen_queue_child(bs_queue, child->bs, new_child_options,
3cdc69d3
HR
4506 child->klass, child->role, bs->drv->is_format,
4507 options, flags, child_keep_old);
e971aa12
JC
4508 }
4509
e971aa12
JC
4510 return bs_queue;
4511}
4512
2e117866 4513/* To be called with bs->aio_context locked */
28518102
KW
4514BlockReopenQueue *bdrv_reopen_queue(BlockReopenQueue *bs_queue,
4515 BlockDriverState *bs,
077e8e20 4516 QDict *options, bool keep_old_opts)
28518102 4517{
f791bf7f
EGE
4518 GLOBAL_STATE_CODE();
4519
3cdc69d3
HR
4520 return bdrv_reopen_queue_child(bs_queue, bs, options, NULL, 0, false,
4521 NULL, 0, keep_old_opts);
28518102
KW
4522}
4523
ab5b5228
AG
4524void bdrv_reopen_queue_free(BlockReopenQueue *bs_queue)
4525{
f791bf7f 4526 GLOBAL_STATE_CODE();
ab5b5228
AG
4527 if (bs_queue) {
4528 BlockReopenQueueEntry *bs_entry, *next;
4529 QTAILQ_FOREACH_SAFE(bs_entry, bs_queue, entry, next) {
d22933ac
KW
4530 AioContext *ctx = bdrv_get_aio_context(bs_entry->state.bs);
4531
4532 aio_context_acquire(ctx);
4533 bdrv_drained_end(bs_entry->state.bs);
4534 aio_context_release(ctx);
4535
ab5b5228
AG
4536 qobject_unref(bs_entry->state.explicit_options);
4537 qobject_unref(bs_entry->state.options);
4538 g_free(bs_entry);
4539 }
4540 g_free(bs_queue);
4541 }
4542}
4543
e971aa12
JC
4544/*
4545 * Reopen multiple BlockDriverStates atomically & transactionally.
4546 *
4547 * The queue passed in (bs_queue) must have been built up previous
4548 * via bdrv_reopen_queue().
4549 *
4550 * Reopens all BDS specified in the queue, with the appropriate
4551 * flags. All devices are prepared for reopen, and failure of any
50d6a8a3 4552 * device will cause all device changes to be abandoned, and intermediate
e971aa12
JC
4553 * data cleaned up.
4554 *
4555 * If all devices prepare successfully, then the changes are committed
4556 * to all devices.
4557 *
1a63a907
KW
4558 * All affected nodes must be drained between bdrv_reopen_queue() and
4559 * bdrv_reopen_multiple().
6cf42ca2
KW
4560 *
4561 * To be called from the main thread, with all other AioContexts unlocked.
e971aa12 4562 */
5019aece 4563int bdrv_reopen_multiple(BlockReopenQueue *bs_queue, Error **errp)
e971aa12
JC
4564{
4565 int ret = -1;
4566 BlockReopenQueueEntry *bs_entry, *next;
6cf42ca2 4567 AioContext *ctx;
72373e40 4568 Transaction *tran = tran_new();
72373e40 4569 g_autoptr(GSList) refresh_list = NULL;
e971aa12 4570
6cf42ca2 4571 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
e971aa12 4572 assert(bs_queue != NULL);
da359909 4573 GLOBAL_STATE_CODE();
e971aa12 4574
a2aabf88 4575 QTAILQ_FOREACH(bs_entry, bs_queue, entry) {
6cf42ca2
KW
4576 ctx = bdrv_get_aio_context(bs_entry->state.bs);
4577 aio_context_acquire(ctx);
a2aabf88 4578 ret = bdrv_flush(bs_entry->state.bs);
6cf42ca2 4579 aio_context_release(ctx);
a2aabf88
VSO
4580 if (ret < 0) {
4581 error_setg_errno(errp, -ret, "Error flushing drive");
e3fc91aa 4582 goto abort;
a2aabf88
VSO
4583 }
4584 }
4585
859443b0 4586 QTAILQ_FOREACH(bs_entry, bs_queue, entry) {
1a63a907 4587 assert(bs_entry->state.bs->quiesce_counter > 0);
6cf42ca2
KW
4588 ctx = bdrv_get_aio_context(bs_entry->state.bs);
4589 aio_context_acquire(ctx);
72373e40 4590 ret = bdrv_reopen_prepare(&bs_entry->state, bs_queue, tran, errp);
6cf42ca2 4591 aio_context_release(ctx);
72373e40
VSO
4592 if (ret < 0) {
4593 goto abort;
e971aa12
JC
4594 }
4595 bs_entry->prepared = true;
4596 }
4597
859443b0 4598 QTAILQ_FOREACH(bs_entry, bs_queue, entry) {
69b736e7 4599 BDRVReopenState *state = &bs_entry->state;
72373e40 4600
fb0ff4d1 4601 refresh_list = g_slist_prepend(refresh_list, state->bs);
72373e40 4602 if (state->old_backing_bs) {
fb0ff4d1 4603 refresh_list = g_slist_prepend(refresh_list, state->old_backing_bs);
cb828c31 4604 }
ecd30d2d 4605 if (state->old_file_bs) {
fb0ff4d1 4606 refresh_list = g_slist_prepend(refresh_list, state->old_file_bs);
ecd30d2d 4607 }
72373e40
VSO
4608 }
4609
4610 /*
4611 * Note that file-posix driver rely on permission update done during reopen
4612 * (even if no permission changed), because it wants "new" permissions for
4613 * reconfiguring the fd and that's why it does it in raw_check_perm(), not
4614 * in raw_reopen_prepare() which is called with "old" permissions.
4615 */
3804e3cf 4616 bdrv_graph_rdlock_main_loop();
72373e40 4617 ret = bdrv_list_refresh_perms(refresh_list, bs_queue, tran, errp);
3804e3cf
KW
4618 bdrv_graph_rdunlock_main_loop();
4619
72373e40
VSO
4620 if (ret < 0) {
4621 goto abort;
69b736e7
KW
4622 }
4623
fcd6a4f4
VSO
4624 /*
4625 * If we reach this point, we have success and just need to apply the
4626 * changes.
4627 *
4628 * Reverse order is used to comfort qcow2 driver: on commit it need to write
4629 * IN_USE flag to the image, to mark bitmaps in the image as invalid. But
4630 * children are usually goes after parents in reopen-queue, so go from last
4631 * to first element.
e971aa12 4632 */
fcd6a4f4 4633 QTAILQ_FOREACH_REVERSE(bs_entry, bs_queue, entry) {
6cf42ca2
KW
4634 ctx = bdrv_get_aio_context(bs_entry->state.bs);
4635 aio_context_acquire(ctx);
e971aa12 4636 bdrv_reopen_commit(&bs_entry->state);
6cf42ca2 4637 aio_context_release(ctx);
e971aa12
JC
4638 }
4639
5661a00d 4640 bdrv_graph_wrlock(NULL);
72373e40 4641 tran_commit(tran);
5661a00d 4642 bdrv_graph_wrunlock();
69b736e7 4643
72373e40
VSO
4644 QTAILQ_FOREACH_REVERSE(bs_entry, bs_queue, entry) {
4645 BlockDriverState *bs = bs_entry->state.bs;
74ad9a3b 4646
72373e40 4647 if (bs->drv->bdrv_reopen_commit_post) {
6cf42ca2
KW
4648 ctx = bdrv_get_aio_context(bs);
4649 aio_context_acquire(ctx);
72373e40 4650 bs->drv->bdrv_reopen_commit_post(&bs_entry->state);
6cf42ca2 4651 aio_context_release(ctx);
69b736e7
KW
4652 }
4653 }
17e1e2be 4654
72373e40
VSO
4655 ret = 0;
4656 goto cleanup;
17e1e2be 4657
72373e40 4658abort:
5661a00d 4659 bdrv_graph_wrlock(NULL);
72373e40 4660 tran_abort(tran);
5661a00d 4661 bdrv_graph_wrunlock();
7d4ca9d2 4662
72373e40
VSO
4663 QTAILQ_FOREACH_SAFE(bs_entry, bs_queue, entry, next) {
4664 if (bs_entry->prepared) {
6cf42ca2
KW
4665 ctx = bdrv_get_aio_context(bs_entry->state.bs);
4666 aio_context_acquire(ctx);
72373e40 4667 bdrv_reopen_abort(&bs_entry->state);
6cf42ca2 4668 aio_context_release(ctx);
17e1e2be
PK
4669 }
4670 }
72373e40 4671
e971aa12 4672cleanup:
ab5b5228 4673 bdrv_reopen_queue_free(bs_queue);
40840e41 4674
e971aa12
JC
4675 return ret;
4676}
4677
6cf42ca2
KW
4678int bdrv_reopen(BlockDriverState *bs, QDict *opts, bool keep_old_opts,
4679 Error **errp)
6e1000a8 4680{
6cf42ca2 4681 AioContext *ctx = bdrv_get_aio_context(bs);
6e1000a8 4682 BlockReopenQueue *queue;
6cf42ca2 4683 int ret;
6e1000a8 4684
f791bf7f
EGE
4685 GLOBAL_STATE_CODE();
4686
2e117866
KW
4687 queue = bdrv_reopen_queue(NULL, bs, opts, keep_old_opts);
4688
6cf42ca2
KW
4689 if (ctx != qemu_get_aio_context()) {
4690 aio_context_release(ctx);
4691 }
5019aece 4692 ret = bdrv_reopen_multiple(queue, errp);
6cf42ca2
KW
4693
4694 if (ctx != qemu_get_aio_context()) {
4695 aio_context_acquire(ctx);
4696 }
6e1000a8
AG
4697
4698 return ret;
4699}
4700
6cf42ca2
KW
4701int bdrv_reopen_set_read_only(BlockDriverState *bs, bool read_only,
4702 Error **errp)
4703{
4704 QDict *opts = qdict_new();
4705
f791bf7f
EGE
4706 GLOBAL_STATE_CODE();
4707
6cf42ca2
KW
4708 qdict_put_bool(opts, BDRV_OPT_READ_ONLY, read_only);
4709
4710 return bdrv_reopen(bs, opts, true, errp);
4711}
4712
cb828c31
AG
4713/*
4714 * Take a BDRVReopenState and check if the value of 'backing' in the
4715 * reopen_state->options QDict is valid or not.
4716 *
4717 * If 'backing' is missing from the QDict then return 0.
4718 *
4719 * If 'backing' contains the node name of the backing file of
4720 * reopen_state->bs then return 0.
4721 *
4722 * If 'backing' contains a different node name (or is null) then check
4723 * whether the current backing file can be replaced with the new one.
4724 * If that's the case then reopen_state->replace_backing_bs is set to
4725 * true and reopen_state->new_backing_bs contains a pointer to the new
4726 * backing BlockDriverState (or NULL).
4727 *
5661a00d
KW
4728 * After calling this function, the transaction @tran may only be completed
4729 * while holding a writer lock for the graph.
4730 *
cb828c31 4731 * Return 0 on success, otherwise return < 0 and set @errp.
4b408668
KW
4732 *
4733 * The caller must hold the AioContext lock of @reopen_state->bs.
4734 * @reopen_state->bs can move to a different AioContext in this function.
4735 * Callers must make sure that their AioContext locking is still correct after
4736 * this.
cb828c31 4737 */
ecd30d2d
AG
4738static int bdrv_reopen_parse_file_or_backing(BDRVReopenState *reopen_state,
4739 bool is_backing, Transaction *tran,
4740 Error **errp)
cb828c31
AG
4741{
4742 BlockDriverState *bs = reopen_state->bs;
ecd30d2d
AG
4743 BlockDriverState *new_child_bs;
4744 BlockDriverState *old_child_bs = is_backing ? child_bs(bs->backing) :
4745 child_bs(bs->file);
4746 const char *child_name = is_backing ? "backing" : "file";
cb828c31
AG
4747 QObject *value;
4748 const char *str;
4b408668
KW
4749 AioContext *ctx, *old_ctx;
4750 int ret;
cb828c31 4751
bdb73476
EGE
4752 GLOBAL_STATE_CODE();
4753
ecd30d2d 4754 value = qdict_get(reopen_state->options, child_name);
cb828c31
AG
4755 if (value == NULL) {
4756 return 0;
4757 }
4758
4759 switch (qobject_type(value)) {
4760 case QTYPE_QNULL:
ecd30d2d
AG
4761 assert(is_backing); /* The 'file' option does not allow a null value */
4762 new_child_bs = NULL;
cb828c31
AG
4763 break;
4764 case QTYPE_QSTRING:
410f44f5 4765 str = qstring_get_str(qobject_to(QString, value));
ecd30d2d
AG
4766 new_child_bs = bdrv_lookup_bs(NULL, str, errp);
4767 if (new_child_bs == NULL) {
cb828c31 4768 return -EINVAL;
ecd30d2d
AG
4769 } else if (bdrv_recurse_has_child(new_child_bs, bs)) {
4770 error_setg(errp, "Making '%s' a %s child of '%s' would create a "
4771 "cycle", str, child_name, bs->node_name);
cb828c31
AG
4772 return -EINVAL;
4773 }
4774 break;
4775 default:
ecd30d2d
AG
4776 /*
4777 * The options QDict has been flattened, so 'backing' and 'file'
4778 * do not allow any other data type here.
4779 */
cb828c31
AG
4780 g_assert_not_reached();
4781 }
4782
ecd30d2d
AG
4783 if (old_child_bs == new_child_bs) {
4784 return 0;
4785 }
4786
4787 if (old_child_bs) {
4788 if (bdrv_skip_implicit_filters(old_child_bs) == new_child_bs) {
cbfdb98c
VSO
4789 return 0;
4790 }
4791
ecd30d2d
AG
4792 if (old_child_bs->implicit) {
4793 error_setg(errp, "Cannot replace implicit %s child of %s",
4794 child_name, bs->node_name);
cbfdb98c
VSO
4795 return -EPERM;
4796 }
4797 }
4798
ecd30d2d 4799 if (bs->drv->is_filter && !old_child_bs) {
25f78d9e
VSO
4800 /*
4801 * Filters always have a file or a backing child, so we are trying to
4802 * change wrong child
4803 */
4804 error_setg(errp, "'%s' is a %s filter node that does not support a "
ecd30d2d 4805 "%s child", bs->node_name, bs->drv->format_name, child_name);
1d42f48c
HR
4806 return -EINVAL;
4807 }
4808
ecd30d2d
AG
4809 if (is_backing) {
4810 reopen_state->old_backing_bs = old_child_bs;
4811 } else {
4812 reopen_state->old_file_bs = old_child_bs;
4813 }
4814
7d4ca9d2
KW
4815 if (old_child_bs) {
4816 bdrv_ref(old_child_bs);
4817 bdrv_drained_begin(old_child_bs);
4818 }
4819
4b408668
KW
4820 old_ctx = bdrv_get_aio_context(bs);
4821 ctx = bdrv_get_aio_context(new_child_bs);
4822 if (old_ctx != ctx) {
4823 aio_context_release(old_ctx);
4824 aio_context_acquire(ctx);
4825 }
4826
7d4ca9d2
KW
4827 bdrv_graph_wrlock(new_child_bs);
4828
4b408668
KW
4829 ret = bdrv_set_file_or_backing_noperm(bs, new_child_bs, is_backing,
4830 tran, errp);
4831
7d4ca9d2
KW
4832 bdrv_graph_wrunlock();
4833
4b408668
KW
4834 if (old_ctx != ctx) {
4835 aio_context_release(ctx);
4836 aio_context_acquire(old_ctx);
4837 }
4838
7d4ca9d2
KW
4839 if (old_child_bs) {
4840 bdrv_drained_end(old_child_bs);
4841 bdrv_unref(old_child_bs);
4842 }
4843
4b408668 4844 return ret;
cb828c31
AG
4845}
4846
e971aa12
JC
4847/*
4848 * Prepares a BlockDriverState for reopen. All changes are staged in the
4849 * 'opaque' field of the BDRVReopenState, which is used and allocated by
4850 * the block driver layer .bdrv_reopen_prepare()
4851 *
4852 * bs is the BlockDriverState to reopen
4853 * flags are the new open flags
4854 * queue is the reopen queue
4855 *
4856 * Returns 0 on success, non-zero on error. On error errp will be set
4857 * as well.
4858 *
4859 * On failure, bdrv_reopen_abort() will be called to clean up any data.
4860 * It is the responsibility of the caller to then call the abort() or
4861 * commit() for any other BDS that have been left in a prepare() state
4862 *
4b408668 4863 * The caller must hold the AioContext lock of @reopen_state->bs.
5661a00d
KW
4864 *
4865 * After calling this function, the transaction @change_child_tran may only be
4866 * completed while holding a writer lock for the graph.
e971aa12 4867 */
53e96d1e 4868static int bdrv_reopen_prepare(BDRVReopenState *reopen_state,
72373e40 4869 BlockReopenQueue *queue,
ecd30d2d 4870 Transaction *change_child_tran, Error **errp)
e971aa12
JC
4871{
4872 int ret = -1;
e6d79c41 4873 int old_flags;
e971aa12
JC
4874 Error *local_err = NULL;
4875 BlockDriver *drv;
ccf9dc07 4876 QemuOpts *opts;
4c8350fe 4877 QDict *orig_reopen_opts;
593b3071 4878 char *discard = NULL;
3d8ce171 4879 bool read_only;
9ad08c44 4880 bool drv_prepared = false;
e971aa12
JC
4881
4882 assert(reopen_state != NULL);
4883 assert(reopen_state->bs->drv != NULL);
da359909 4884 GLOBAL_STATE_CODE();
e971aa12
JC
4885 drv = reopen_state->bs->drv;
4886
4c8350fe
AG
4887 /* This function and each driver's bdrv_reopen_prepare() remove
4888 * entries from reopen_state->options as they are processed, so
4889 * we need to make a copy of the original QDict. */
4890 orig_reopen_opts = qdict_clone_shallow(reopen_state->options);
4891
ccf9dc07
KW
4892 /* Process generic block layer options */
4893 opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
af175e85 4894 if (!qemu_opts_absorb_qdict(opts, reopen_state->options, errp)) {
ccf9dc07
KW
4895 ret = -EINVAL;
4896 goto error;
4897 }
4898
e6d79c41
AG
4899 /* This was already called in bdrv_reopen_queue_child() so the flags
4900 * are up-to-date. This time we simply want to remove the options from
4901 * QemuOpts in order to indicate that they have been processed. */
4902 old_flags = reopen_state->flags;
91a097e7 4903 update_flags_from_options(&reopen_state->flags, opts);
e6d79c41 4904 assert(old_flags == reopen_state->flags);
91a097e7 4905
415bbca8 4906 discard = qemu_opt_get_del(opts, BDRV_OPT_DISCARD);
593b3071
AG
4907 if (discard != NULL) {
4908 if (bdrv_parse_discard_flags(discard, &reopen_state->flags) != 0) {
4909 error_setg(errp, "Invalid discard option");
4910 ret = -EINVAL;
4911 goto error;
4912 }
4913 }
4914
543770bd
AG
4915 reopen_state->detect_zeroes =
4916 bdrv_parse_detect_zeroes(opts, reopen_state->flags, &local_err);
4917 if (local_err) {
4918 error_propagate(errp, local_err);
4919 ret = -EINVAL;
4920 goto error;
4921 }
4922
57f9db9a
AG
4923 /* All other options (including node-name and driver) must be unchanged.
4924 * Put them back into the QDict, so that they are checked at the end
4925 * of this function. */
4926 qemu_opts_to_qdict(opts, reopen_state->options);
ccf9dc07 4927
3d8ce171
JC
4928 /* If we are to stay read-only, do not allow permission change
4929 * to r/w. Attempting to set to r/w may fail if either BDRV_O_ALLOW_RDWR is
4930 * not set, or if the BDS still has copy_on_read enabled */
4931 read_only = !(reopen_state->flags & BDRV_O_RDWR);
54a32bfe 4932 ret = bdrv_can_set_read_only(reopen_state->bs, read_only, true, &local_err);
3d8ce171
JC
4933 if (local_err) {
4934 error_propagate(errp, local_err);
e971aa12
JC
4935 goto error;
4936 }
4937
e971aa12 4938 if (drv->bdrv_reopen_prepare) {
faf116b4
AG
4939 /*
4940 * If a driver-specific option is missing, it means that we
4941 * should reset it to its default value.
4942 * But not all options allow that, so we need to check it first.
4943 */
4944 ret = bdrv_reset_options_allowed(reopen_state->bs,
4945 reopen_state->options, errp);
4946 if (ret) {
4947 goto error;
4948 }
4949
e971aa12
JC
4950 ret = drv->bdrv_reopen_prepare(reopen_state, queue, &local_err);
4951 if (ret) {
4952 if (local_err != NULL) {
4953 error_propagate(errp, local_err);
4954 } else {
f30c66ba 4955 bdrv_refresh_filename(reopen_state->bs);
d8b6895f
LC
4956 error_setg(errp, "failed while preparing to reopen image '%s'",
4957 reopen_state->bs->filename);
e971aa12
JC
4958 }
4959 goto error;
4960 }
4961 } else {
4962 /* It is currently mandatory to have a bdrv_reopen_prepare()
4963 * handler for each supported drv. */
81e5f78a
AG
4964 error_setg(errp, "Block format '%s' used by node '%s' "
4965 "does not support reopening files", drv->format_name,
4966 bdrv_get_device_or_node_name(reopen_state->bs));
e971aa12
JC
4967 ret = -1;
4968 goto error;
4969 }
4970
9ad08c44
HR
4971 drv_prepared = true;
4972
bacd9b87
AG
4973 /*
4974 * We must provide the 'backing' option if the BDS has a backing
4975 * file or if the image file has a backing file name as part of
4976 * its metadata. Otherwise the 'backing' option can be omitted.
4977 */
4978 if (drv->supports_backing && reopen_state->backing_missing &&
1d42f48c 4979 (reopen_state->bs->backing || reopen_state->bs->backing_file[0])) {
8546632e
AG
4980 error_setg(errp, "backing is missing for '%s'",
4981 reopen_state->bs->node_name);
4982 ret = -EINVAL;
4983 goto error;
4984 }
4985
cb828c31
AG
4986 /*
4987 * Allow changing the 'backing' option. The new value can be
4988 * either a reference to an existing node (using its node name)
4989 * or NULL to simply detach the current backing file.
4990 */
ecd30d2d
AG
4991 ret = bdrv_reopen_parse_file_or_backing(reopen_state, true,
4992 change_child_tran, errp);
cb828c31
AG
4993 if (ret < 0) {
4994 goto error;
4995 }
4996 qdict_del(reopen_state->options, "backing");
4997
ecd30d2d
AG
4998 /* Allow changing the 'file' option. In this case NULL is not allowed */
4999 ret = bdrv_reopen_parse_file_or_backing(reopen_state, false,
5000 change_child_tran, errp);
5001 if (ret < 0) {
5002 goto error;
5003 }
5004 qdict_del(reopen_state->options, "file");
5005
4d2cb092
KW
5006 /* Options that are not handled are only okay if they are unchanged
5007 * compared to the old state. It is expected that some options are only
5008 * used for the initial open, but not reopen (e.g. filename) */
5009 if (qdict_size(reopen_state->options)) {
5010 const QDictEntry *entry = qdict_first(reopen_state->options);
5011
5012 do {
54fd1b0d
HR
5013 QObject *new = entry->value;
5014 QObject *old = qdict_get(reopen_state->bs->options, entry->key);
5015
db905283
AG
5016 /* Allow child references (child_name=node_name) as long as they
5017 * point to the current child (i.e. everything stays the same). */
5018 if (qobject_type(new) == QTYPE_QSTRING) {
5019 BdrvChild *child;
5020 QLIST_FOREACH(child, &reopen_state->bs->children, next) {
5021 if (!strcmp(child->name, entry->key)) {
5022 break;
5023 }
5024 }
5025
5026 if (child) {
410f44f5
MA
5027 if (!strcmp(child->bs->node_name,
5028 qstring_get_str(qobject_to(QString, new)))) {
db905283
AG
5029 continue; /* Found child with this name, skip option */
5030 }
5031 }
5032 }
5033
129c7d1c 5034 /*
54fd1b0d
HR
5035 * TODO: When using -drive to specify blockdev options, all values
5036 * will be strings; however, when using -blockdev, blockdev-add or
5037 * filenames using the json:{} pseudo-protocol, they will be
5038 * correctly typed.
5039 * In contrast, reopening options are (currently) always strings
5040 * (because you can only specify them through qemu-io; all other
5041 * callers do not specify any options).
5042 * Therefore, when using anything other than -drive to create a BDS,
5043 * this cannot detect non-string options as unchanged, because
5044 * qobject_is_equal() always returns false for objects of different
5045 * type. In the future, this should be remedied by correctly typing
5046 * all options. For now, this is not too big of an issue because
5047 * the user can simply omit options which cannot be changed anyway,
5048 * so they will stay unchanged.
129c7d1c 5049 */
54fd1b0d 5050 if (!qobject_is_equal(new, old)) {
4d2cb092
KW
5051 error_setg(errp, "Cannot change the option '%s'", entry->key);
5052 ret = -EINVAL;
5053 goto error;
5054 }
5055 } while ((entry = qdict_next(reopen_state->options, entry)));
5056 }
5057
e971aa12
JC
5058 ret = 0;
5059
4c8350fe
AG
5060 /* Restore the original reopen_state->options QDict */
5061 qobject_unref(reopen_state->options);
5062 reopen_state->options = qobject_ref(orig_reopen_opts);
5063
e971aa12 5064error:
9ad08c44
HR
5065 if (ret < 0 && drv_prepared) {
5066 /* drv->bdrv_reopen_prepare() has succeeded, so we need to
5067 * call drv->bdrv_reopen_abort() before signaling an error
5068 * (bdrv_reopen_multiple() will not call bdrv_reopen_abort()
5069 * when the respective bdrv_reopen_prepare() has failed) */
5070 if (drv->bdrv_reopen_abort) {
5071 drv->bdrv_reopen_abort(reopen_state);
5072 }
5073 }
ccf9dc07 5074 qemu_opts_del(opts);
4c8350fe 5075 qobject_unref(orig_reopen_opts);
593b3071 5076 g_free(discard);
e971aa12
JC
5077 return ret;
5078}
5079
5080/*
5081 * Takes the staged changes for the reopen from bdrv_reopen_prepare(), and
5082 * makes them final by swapping the staging BlockDriverState contents into
5083 * the active BlockDriverState contents.
5084 */
53e96d1e 5085static void bdrv_reopen_commit(BDRVReopenState *reopen_state)
e971aa12
JC
5086{
5087 BlockDriver *drv;
50bf65ba 5088 BlockDriverState *bs;
50196d7a 5089 BdrvChild *child;
e971aa12
JC
5090
5091 assert(reopen_state != NULL);
50bf65ba
VSO
5092 bs = reopen_state->bs;
5093 drv = bs->drv;
e971aa12 5094 assert(drv != NULL);
da359909 5095 GLOBAL_STATE_CODE();
e971aa12
JC
5096
5097 /* If there are any driver level actions to take */
5098 if (drv->bdrv_reopen_commit) {
5099 drv->bdrv_reopen_commit(reopen_state);
5100 }
5101
5102 /* set BDS specific flags now */
cb3e7f08 5103 qobject_unref(bs->explicit_options);
4c8350fe 5104 qobject_unref(bs->options);
ab5b5228
AG
5105 qobject_ref(reopen_state->explicit_options);
5106 qobject_ref(reopen_state->options);
145f598e 5107
50bf65ba 5108 bs->explicit_options = reopen_state->explicit_options;
4c8350fe 5109 bs->options = reopen_state->options;
50bf65ba 5110 bs->open_flags = reopen_state->flags;
543770bd 5111 bs->detect_zeroes = reopen_state->detect_zeroes;
355ef4ac 5112
50196d7a
AG
5113 /* Remove child references from bs->options and bs->explicit_options.
5114 * Child options were already removed in bdrv_reopen_queue_child() */
5115 QLIST_FOREACH(child, &bs->children, next) {
5116 qdict_del(bs->explicit_options, child->name);
5117 qdict_del(bs->options, child->name);
5118 }
3d0e8743
VSO
5119 /* backing is probably removed, so it's not handled by previous loop */
5120 qdict_del(bs->explicit_options, "backing");
5121 qdict_del(bs->options, "backing");
5122
e19b157f 5123 bdrv_graph_rdlock_main_loop();
1e4c797c 5124 bdrv_refresh_limits(bs, NULL, NULL);
e19b157f 5125 bdrv_graph_rdunlock_main_loop();
439cc330 5126 bdrv_refresh_total_sectors(bs, bs->total_sectors);
e971aa12
JC
5127}
5128
5129/*
5130 * Abort the reopen, and delete and free the staged changes in
5131 * reopen_state
5132 */
53e96d1e 5133static void bdrv_reopen_abort(BDRVReopenState *reopen_state)
e971aa12
JC
5134{
5135 BlockDriver *drv;
5136
5137 assert(reopen_state != NULL);
5138 drv = reopen_state->bs->drv;
5139 assert(drv != NULL);
da359909 5140 GLOBAL_STATE_CODE();
e971aa12
JC
5141
5142 if (drv->bdrv_reopen_abort) {
5143 drv->bdrv_reopen_abort(reopen_state);
5144 }
5145}
5146
5147
64dff520 5148static void bdrv_close(BlockDriverState *bs)
fc01f7e7 5149{
33384421 5150 BdrvAioNotifier *ban, *ban_next;
50a3efb0 5151 BdrvChild *child, *next;
33384421 5152
f791bf7f 5153 GLOBAL_STATE_CODE();
30f55fb8 5154 assert(!bs->refcnt);
99b7e775 5155
fc27291d 5156 bdrv_drained_begin(bs); /* complete I/O */
58fda173 5157 bdrv_flush(bs);
53ec73e2 5158 bdrv_drain(bs); /* in case flush left pending I/O */
fc27291d 5159
3cbc002c 5160 if (bs->drv) {
3c005293 5161 if (bs->drv->bdrv_close) {
7b99a266 5162 /* Must unfreeze all children, so bdrv_unref_child() works */
3c005293
VSO
5163 bs->drv->bdrv_close(bs);
5164 }
9a4f4c31 5165 bs->drv = NULL;
50a3efb0 5166 }
9a7dedbc 5167
32a8aba3 5168 bdrv_graph_wrlock(NULL);
50a3efb0 5169 QLIST_FOREACH_SAFE(child, &bs->children, next, next) {
dd4118c7 5170 bdrv_unref_child(bs, child);
b338082b 5171 }
32a8aba3 5172 bdrv_graph_wrunlock();
98f90dba 5173
5bb04747
VSO
5174 assert(!bs->backing);
5175 assert(!bs->file);
50a3efb0
AG
5176 g_free(bs->opaque);
5177 bs->opaque = NULL;
d73415a3 5178 qatomic_set(&bs->copy_on_read, 0);
50a3efb0
AG
5179 bs->backing_file[0] = '\0';
5180 bs->backing_format[0] = '\0';
5181 bs->total_sectors = 0;
5182 bs->encrypted = false;
5183 bs->sg = false;
cb3e7f08
MAL
5184 qobject_unref(bs->options);
5185 qobject_unref(bs->explicit_options);
50a3efb0
AG
5186 bs->options = NULL;
5187 bs->explicit_options = NULL;
cb3e7f08 5188 qobject_unref(bs->full_open_options);
50a3efb0 5189 bs->full_open_options = NULL;
0bc329fb
HR
5190 g_free(bs->block_status_cache);
5191 bs->block_status_cache = NULL;
50a3efb0 5192
cca43ae1
VSO
5193 bdrv_release_named_dirty_bitmaps(bs);
5194 assert(QLIST_EMPTY(&bs->dirty_bitmaps));
5195
33384421
HR
5196 QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_next) {
5197 g_free(ban);
5198 }
5199 QLIST_INIT(&bs->aio_notifiers);
fc27291d 5200 bdrv_drained_end(bs);
1a6d3bd2
GK
5201
5202 /*
5203 * If we're still inside some bdrv_drain_all_begin()/end() sections, end
5204 * them now since this BDS won't exist anymore when bdrv_drain_all_end()
5205 * gets called.
5206 */
5207 if (bs->quiesce_counter) {
5208 bdrv_drain_all_end_quiesce(bs);
5209 }
b338082b
FB
5210}
5211
2bc93fed
MK
5212void bdrv_close_all(void)
5213{
f791bf7f 5214 GLOBAL_STATE_CODE();
880eeec6 5215 assert(job_next(NULL) == NULL);
ca9bd24c
HR
5216
5217 /* Drop references from requests still in flight, such as canceled block
5218 * jobs whose AIO context has not been polled yet */
5219 bdrv_drain_all();
2bc93fed 5220
ca9bd24c
HR
5221 blk_remove_all_bs();
5222 blockdev_close_all_bdrv_states();
ed78cda3 5223
a1a2af07 5224 assert(QTAILQ_EMPTY(&all_bdrv_states));
2bc93fed
MK
5225}
5226
2f64e1fc 5227static bool GRAPH_RDLOCK should_update_child(BdrvChild *c, BlockDriverState *to)
d0ac0380 5228{
2f30b7c3
VSO
5229 GQueue *queue;
5230 GHashTable *found;
5231 bool ret;
d0ac0380 5232
bd86fb99 5233 if (c->klass->stay_at_node) {
d0ac0380
KW
5234 return false;
5235 }
5236
ec9f10fe
HR
5237 /* If the child @c belongs to the BDS @to, replacing the current
5238 * c->bs by @to would mean to create a loop.
5239 *
5240 * Such a case occurs when appending a BDS to a backing chain.
5241 * For instance, imagine the following chain:
5242 *
5243 * guest device -> node A -> further backing chain...
5244 *
5245 * Now we create a new BDS B which we want to put on top of this
5246 * chain, so we first attach A as its backing node:
5247 *
5248 * node B
5249 * |
5250 * v
5251 * guest device -> node A -> further backing chain...
5252 *
5253 * Finally we want to replace A by B. When doing that, we want to
5254 * replace all pointers to A by pointers to B -- except for the
5255 * pointer from B because (1) that would create a loop, and (2)
5256 * that pointer should simply stay intact:
5257 *
5258 * guest device -> node B
5259 * |
5260 * v
5261 * node A -> further backing chain...
5262 *
5263 * In general, when replacing a node A (c->bs) by a node B (@to),
5264 * if A is a child of B, that means we cannot replace A by B there
5265 * because that would create a loop. Silently detaching A from B
5266 * is also not really an option. So overall just leaving A in
2f30b7c3
VSO
5267 * place there is the most sensible choice.
5268 *
5269 * We would also create a loop in any cases where @c is only
5270 * indirectly referenced by @to. Prevent this by returning false
5271 * if @c is found (by breadth-first search) anywhere in the whole
5272 * subtree of @to.
5273 */
5274
5275 ret = true;
5276 found = g_hash_table_new(NULL, NULL);
5277 g_hash_table_add(found, to);
5278 queue = g_queue_new();
5279 g_queue_push_tail(queue, to);
5280
5281 while (!g_queue_is_empty(queue)) {
5282 BlockDriverState *v = g_queue_pop_head(queue);
5283 BdrvChild *c2;
5284
5285 QLIST_FOREACH(c2, &v->children, next) {
5286 if (c2 == c) {
5287 ret = false;
5288 break;
5289 }
5290
5291 if (g_hash_table_contains(found, c2->bs)) {
5292 continue;
5293 }
5294
5295 g_queue_push_tail(queue, c2->bs);
5296 g_hash_table_add(found, c2->bs);
d0ac0380
KW
5297 }
5298 }
5299
2f30b7c3
VSO
5300 g_queue_free(queue);
5301 g_hash_table_destroy(found);
5302
5303 return ret;
d0ac0380
KW
5304}
5305
57f08941 5306static void bdrv_remove_child_commit(void *opaque)
46541ee5 5307{
bdb73476 5308 GLOBAL_STATE_CODE();
5bb04747 5309 bdrv_child_free(opaque);
82b54cf5
HR
5310}
5311
57f08941
VSO
5312static TransactionActionDrv bdrv_remove_child_drv = {
5313 .commit = bdrv_remove_child_commit,
46541ee5
VSO
5314};
5315
2f64e1fc
KW
5316/*
5317 * Function doesn't update permissions, caller is responsible for this.
5318 *
5319 * @child->bs (if non-NULL) must be drained.
5661a00d
KW
5320 *
5321 * After calling this function, the transaction @tran may only be completed
5322 * while holding a writer lock for the graph.
2f64e1fc
KW
5323 */
5324static void GRAPH_WRLOCK bdrv_remove_child(BdrvChild *child, Transaction *tran)
46541ee5 5325{
46541ee5
VSO
5326 if (!child) {
5327 return;
5328 }
5329
5330 if (child->bs) {
2f64e1fc 5331 assert(child->quiesced_parent);
a2c37a30 5332 bdrv_replace_child_tran(child, NULL, tran);
46541ee5
VSO
5333 }
5334
57f08941 5335 tran_add(tran, &bdrv_remove_child_drv, child);
46541ee5
VSO
5336}
5337
2f64e1fc
KW
5338/*
5339 * Both @from and @to (if non-NULL) must be drained. @to must be kept drained
5340 * until the transaction is completed.
5661a00d
KW
5341 *
5342 * After calling this function, the transaction @tran may only be completed
5343 * while holding a writer lock for the graph.
2f64e1fc
KW
5344 */
5345static int GRAPH_WRLOCK
5346bdrv_replace_node_noperm(BlockDriverState *from,
5347 BlockDriverState *to,
5348 bool auto_skip, Transaction *tran,
5349 Error **errp)
117caba9
VSO
5350{
5351 BdrvChild *c, *next;
5352
bdb73476 5353 GLOBAL_STATE_CODE();
82b54cf5 5354
2f64e1fc
KW
5355 assert(from->quiesce_counter);
5356 assert(to->quiesce_counter);
23987471 5357
117caba9
VSO
5358 QLIST_FOREACH_SAFE(c, &from->parents, next_parent, next) {
5359 assert(c->bs == from);
5360 if (!should_update_child(c, to)) {
5361 if (auto_skip) {
5362 continue;
5363 }
5364 error_setg(errp, "Should not change '%s' link to '%s'",
5365 c->name, from->node_name);
5366 return -EINVAL;
5367 }
5368 if (c->frozen) {
5369 error_setg(errp, "Cannot change '%s' link to '%s'",
5370 c->name, from->node_name);
5371 return -EPERM;
5372 }
0f0b1e29 5373 bdrv_replace_child_tran(c, to, tran);
117caba9
VSO
5374 }
5375
5376 return 0;
5377}
5378
313274bb
VSO
5379/*
5380 * With auto_skip=true bdrv_replace_node_common skips updating from parents
5381 * if it creates a parent-child relation loop or if parent is block-job.
5382 *
5383 * With auto_skip=false the error is returned if from has a parent which should
5384 * not be updated.
3108a15c
VSO
5385 *
5386 * With @detach_subchain=true @to must be in a backing chain of @from. In this
5387 * case backing link of the cow-parent of @to is removed.
313274bb 5388 */
a1e708fc
VSO
5389static int bdrv_replace_node_common(BlockDriverState *from,
5390 BlockDriverState *to,
3108a15c
VSO
5391 bool auto_skip, bool detach_subchain,
5392 Error **errp)
dd62f1ca 5393{
3bb0e298 5394 Transaction *tran = tran_new();
3bb0e298 5395 g_autoptr(GSList) refresh_list = NULL;
2d369d6e 5396 BlockDriverState *to_cow_parent = NULL;
234ac1a9
KW
5397 int ret;
5398
bdb73476 5399 GLOBAL_STATE_CODE();
82b54cf5 5400
3108a15c
VSO
5401 if (detach_subchain) {
5402 assert(bdrv_chain_contains(from, to));
5403 assert(from != to);
5404 for (to_cow_parent = from;
5405 bdrv_filter_or_cow_bs(to_cow_parent) != to;
5406 to_cow_parent = bdrv_filter_or_cow_bs(to_cow_parent))
5407 {
5408 ;
5409 }
5410 }
5411
234ac1a9
KW
5412 /* Make sure that @from doesn't go away until we have successfully attached
5413 * all of its parents to @to. */
5414 bdrv_ref(from);
dd62f1ca 5415
f871abd6 5416 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
30dd65f3 5417 assert(bdrv_get_aio_context(from) == bdrv_get_aio_context(to));
f871abd6 5418 bdrv_drained_begin(from);
2f64e1fc
KW
5419 bdrv_drained_begin(to);
5420
5421 bdrv_graph_wrlock(to);
f871abd6 5422
3bb0e298
VSO
5423 /*
5424 * Do the replacement without permission update.
5425 * Replacement may influence the permissions, we should calculate new
5426 * permissions based on new graph. If we fail, we'll roll-back the
5427 * replacement.
5428 */
117caba9
VSO
5429 ret = bdrv_replace_node_noperm(from, to, auto_skip, tran, errp);
5430 if (ret < 0) {
5431 goto out;
234ac1a9
KW
5432 }
5433
3108a15c 5434 if (detach_subchain) {
2f64e1fc 5435 /* to_cow_parent is already drained because from is drained */
f38eaec4 5436 bdrv_remove_child(bdrv_filter_or_cow_child(to_cow_parent), tran);
3108a15c
VSO
5437 }
5438
fb0ff4d1
VSO
5439 refresh_list = g_slist_prepend(refresh_list, to);
5440 refresh_list = g_slist_prepend(refresh_list, from);
9bd910e2 5441
3bb0e298
VSO
5442 ret = bdrv_list_refresh_perms(refresh_list, NULL, tran, errp);
5443 if (ret < 0) {
5444 goto out;
dd62f1ca 5445 }
234ac1a9 5446
a1e708fc
VSO
5447 ret = 0;
5448
234ac1a9 5449out:
3bb0e298 5450 tran_finalize(tran, ret);
5661a00d 5451 bdrv_graph_wrunlock();
3bb0e298 5452
2f64e1fc 5453 bdrv_drained_end(to);
f871abd6 5454 bdrv_drained_end(from);
234ac1a9 5455 bdrv_unref(from);
a1e708fc
VSO
5456
5457 return ret;
dd62f1ca
KW
5458}
5459
a1e708fc
VSO
5460int bdrv_replace_node(BlockDriverState *from, BlockDriverState *to,
5461 Error **errp)
313274bb 5462{
f791bf7f
EGE
5463 GLOBAL_STATE_CODE();
5464
3108a15c
VSO
5465 return bdrv_replace_node_common(from, to, true, false, errp);
5466}
5467
5468int bdrv_drop_filter(BlockDriverState *bs, Error **errp)
5469{
f791bf7f
EGE
5470 GLOBAL_STATE_CODE();
5471
3108a15c
VSO
5472 return bdrv_replace_node_common(bs, bdrv_filter_or_cow_bs(bs), true, true,
5473 errp);
313274bb
VSO
5474}
5475
4ddc07ca
PB
5476/*
5477 * Add new bs contents at the top of an image chain while the chain is
5478 * live, while keeping required fields on the top layer.
5479 *
5480 * This will modify the BlockDriverState fields, and swap contents
5481 * between bs_new and bs_top. Both bs_new and bs_top are modified.
5482 *
2272edcf
VSO
5483 * bs_new must not be attached to a BlockBackend and must not have backing
5484 * child.
4ddc07ca
PB
5485 *
5486 * This function does not create any image files.
60d90bf4
SG
5487 *
5488 * The caller must hold the AioContext lock for @bs_top.
4ddc07ca 5489 */
a1e708fc
VSO
5490int bdrv_append(BlockDriverState *bs_new, BlockDriverState *bs_top,
5491 Error **errp)
4ddc07ca 5492{
2272edcf 5493 int ret;
5bb04747 5494 BdrvChild *child;
2272edcf 5495 Transaction *tran = tran_new();
60d90bf4 5496 AioContext *old_context, *new_context = NULL;
2272edcf 5497
f791bf7f
EGE
5498 GLOBAL_STATE_CODE();
5499
2272edcf
VSO
5500 assert(!bs_new->backing);
5501
60d90bf4 5502 old_context = bdrv_get_aio_context(bs_top);
7d4ca9d2
KW
5503 bdrv_drained_begin(bs_top);
5504
5505 /*
5506 * bdrv_drained_begin() requires that only the AioContext of the drained
5507 * node is locked, and at this point it can still differ from the AioContext
5508 * of bs_top.
5509 */
5510 new_context = bdrv_get_aio_context(bs_new);
5511 aio_context_release(old_context);
5512 aio_context_acquire(new_context);
5513 bdrv_drained_begin(bs_new);
5514 aio_context_release(new_context);
5515 aio_context_acquire(old_context);
5516 new_context = NULL;
5517
5518 bdrv_graph_wrlock(bs_top);
60d90bf4 5519
5bb04747
VSO
5520 child = bdrv_attach_child_noperm(bs_new, bs_top, "backing",
5521 &child_of_bds, bdrv_backing_role(bs_new),
5522 tran, errp);
5523 if (!child) {
5524 ret = -EINVAL;
2272edcf 5525 goto out;
b2c2832c 5526 }
dd62f1ca 5527
60d90bf4 5528 /*
7d4ca9d2
KW
5529 * bdrv_attach_child_noperm could change the AioContext of bs_top and
5530 * bs_new, but at least they are in the same AioContext now. This is the
5531 * AioContext that we need to lock for the rest of the function.
60d90bf4
SG
5532 */
5533 new_context = bdrv_get_aio_context(bs_top);
5534
5535 if (old_context != new_context) {
5536 aio_context_release(old_context);
5537 aio_context_acquire(new_context);
5538 }
5539
2272edcf 5540 ret = bdrv_replace_node_noperm(bs_top, bs_new, true, tran, errp);
a1e708fc 5541 if (ret < 0) {
2272edcf 5542 goto out;
234ac1a9 5543 }
4ddc07ca 5544
f1316edb 5545 ret = bdrv_refresh_perms(bs_new, tran, errp);
2272edcf
VSO
5546out:
5547 tran_finalize(tran, ret);
5548
1e4c797c 5549 bdrv_refresh_limits(bs_top, NULL, NULL);
5661a00d 5550 bdrv_graph_wrunlock();
2272edcf 5551
7d4ca9d2
KW
5552 bdrv_drained_end(bs_top);
5553 bdrv_drained_end(bs_new);
2f64e1fc 5554
60d90bf4
SG
5555 if (new_context && old_context != new_context) {
5556 aio_context_release(new_context);
5557 aio_context_acquire(old_context);
5558 }
5559
2272edcf 5560 return ret;
8802d1fd
JC
5561}
5562
bd8f4c42
VSO
5563/* Not for empty child */
5564int bdrv_replace_child_bs(BdrvChild *child, BlockDriverState *new_bs,
5565 Error **errp)
5566{
5567 int ret;
5568 Transaction *tran = tran_new();
bd8f4c42
VSO
5569 g_autoptr(GSList) refresh_list = NULL;
5570 BlockDriverState *old_bs = child->bs;
5571
f791bf7f
EGE
5572 GLOBAL_STATE_CODE();
5573
bd8f4c42
VSO
5574 bdrv_ref(old_bs);
5575 bdrv_drained_begin(old_bs);
5576 bdrv_drained_begin(new_bs);
2f64e1fc 5577 bdrv_graph_wrlock(new_bs);
bd8f4c42 5578
0f0b1e29 5579 bdrv_replace_child_tran(child, new_bs, tran);
bd8f4c42 5580
fb0ff4d1
VSO
5581 refresh_list = g_slist_prepend(refresh_list, old_bs);
5582 refresh_list = g_slist_prepend(refresh_list, new_bs);
bd8f4c42
VSO
5583
5584 ret = bdrv_list_refresh_perms(refresh_list, NULL, tran, errp);
5585
5586 tran_finalize(tran, ret);
5587
5661a00d 5588 bdrv_graph_wrunlock();
bd8f4c42
VSO
5589 bdrv_drained_end(old_bs);
5590 bdrv_drained_end(new_bs);
5591 bdrv_unref(old_bs);
5592
5593 return ret;
5594}
5595
4f6fd349 5596static void bdrv_delete(BlockDriverState *bs)
b338082b 5597{
3718d8ab 5598 assert(bdrv_op_blocker_is_empty(bs));
4f6fd349 5599 assert(!bs->refcnt);
f791bf7f 5600 GLOBAL_STATE_CODE();
18846dee 5601
1b7bdbc1 5602 /* remove from list, if necessary */
63eaaae0
KW
5603 if (bs->node_name[0] != '\0') {
5604 QTAILQ_REMOVE(&graph_bdrv_states, bs, node_list);
5605 }
2c1d04e0
HR
5606 QTAILQ_REMOVE(&all_bdrv_states, bs, bs_list);
5607
30c321f9
AK
5608 bdrv_close(bs);
5609
fa9185fc
SH
5610 qemu_mutex_destroy(&bs->reqs_lock);
5611
7267c094 5612 g_free(bs);
fc01f7e7
FB
5613}
5614
96796fae
VSO
5615
5616/*
5617 * Replace @bs by newly created block node.
5618 *
5619 * @options is a QDict of options to pass to the block drivers, or NULL for an
5620 * empty set of options. The reference to the QDict belongs to the block layer
5621 * after the call (even on failure), so if the caller intends to reuse the
5622 * dictionary, it needs to use qobject_ref() before calling bdrv_open.
8823407c
KW
5623 *
5624 * The caller holds the AioContext lock for @bs. It must make sure that @bs
5625 * stays in the same AioContext, i.e. @options must not refer to nodes in a
5626 * different AioContext.
96796fae
VSO
5627 */
5628BlockDriverState *bdrv_insert_node(BlockDriverState *bs, QDict *options,
8872ef78
AS
5629 int flags, Error **errp)
5630{
f053b7e8
VSO
5631 ERRP_GUARD();
5632 int ret;
8823407c 5633 AioContext *ctx = bdrv_get_aio_context(bs);
b11c8739
VSO
5634 BlockDriverState *new_node_bs = NULL;
5635 const char *drvname, *node_name;
5636 BlockDriver *drv;
5637
5638 drvname = qdict_get_try_str(options, "driver");
5639 if (!drvname) {
5640 error_setg(errp, "driver is not specified");
5641 goto fail;
5642 }
5643
5644 drv = bdrv_find_format(drvname);
5645 if (!drv) {
5646 error_setg(errp, "Unknown driver: '%s'", drvname);
5647 goto fail;
5648 }
8872ef78 5649
b11c8739
VSO
5650 node_name = qdict_get_try_str(options, "node-name");
5651
f791bf7f
EGE
5652 GLOBAL_STATE_CODE();
5653
8823407c
KW
5654 aio_context_release(ctx);
5655 aio_context_acquire(qemu_get_aio_context());
b11c8739
VSO
5656 new_node_bs = bdrv_new_open_driver_opts(drv, node_name, options, flags,
5657 errp);
8823407c
KW
5658 aio_context_release(qemu_get_aio_context());
5659 aio_context_acquire(ctx);
5660 assert(bdrv_get_aio_context(bs) == ctx);
5661
b11c8739
VSO
5662 options = NULL; /* bdrv_new_open_driver() eats options */
5663 if (!new_node_bs) {
8872ef78 5664 error_prepend(errp, "Could not create node: ");
b11c8739 5665 goto fail;
8872ef78
AS
5666 }
5667
5668 bdrv_drained_begin(bs);
f053b7e8 5669 ret = bdrv_replace_node(bs, new_node_bs, errp);
8872ef78
AS
5670 bdrv_drained_end(bs);
5671
f053b7e8
VSO
5672 if (ret < 0) {
5673 error_prepend(errp, "Could not replace node: ");
b11c8739 5674 goto fail;
8872ef78
AS
5675 }
5676
5677 return new_node_bs;
b11c8739
VSO
5678
5679fail:
5680 qobject_unref(options);
5681 bdrv_unref(new_node_bs);
5682 return NULL;
8872ef78
AS
5683}
5684
e97fc193
AL
5685/*
5686 * Run consistency checks on an image
5687 *
e076f338 5688 * Returns 0 if the check could be completed (it doesn't mean that the image is
a1c7273b 5689 * free of errors) or -errno when an internal error occurred. The results of the
e076f338 5690 * check are stored in res.
e97fc193 5691 */
21c2283e
VSO
5692int coroutine_fn bdrv_co_check(BlockDriverState *bs,
5693 BdrvCheckResult *res, BdrvCheckMode fix)
e97fc193 5694{
1581a70d 5695 IO_CODE();
1b3ff9fe 5696 assert_bdrv_graph_readable();
908bcd54
HR
5697 if (bs->drv == NULL) {
5698 return -ENOMEDIUM;
5699 }
2fd61638 5700 if (bs->drv->bdrv_co_check == NULL) {
e97fc193
AL
5701 return -ENOTSUP;
5702 }
5703
e076f338 5704 memset(res, 0, sizeof(*res));
2fd61638
PB
5705 return bs->drv->bdrv_co_check(bs, res, fix);
5706}
5707
756e6736
KW
5708/*
5709 * Return values:
5710 * 0 - success
5711 * -EINVAL - backing format specified, but no file
5712 * -ENOSPC - can't update the backing file because no space is left in the
5713 * image file header
5714 * -ENOTSUP - format driver doesn't support changing the backing file
5715 */
e54ee1b3 5716int bdrv_change_backing_file(BlockDriverState *bs, const char *backing_file,
497a30db 5717 const char *backing_fmt, bool require)
756e6736
KW
5718{
5719 BlockDriver *drv = bs->drv;
469ef350 5720 int ret;
756e6736 5721
f791bf7f
EGE
5722 GLOBAL_STATE_CODE();
5723
d470ad42
HR
5724 if (!drv) {
5725 return -ENOMEDIUM;
5726 }
5727
5f377794
PB
5728 /* Backing file format doesn't make sense without a backing file */
5729 if (backing_fmt && !backing_file) {
5730 return -EINVAL;
5731 }
5732
497a30db
EB
5733 if (require && backing_file && !backing_fmt) {
5734 return -EINVAL;
e54ee1b3
EB
5735 }
5736
756e6736 5737 if (drv->bdrv_change_backing_file != NULL) {
469ef350 5738 ret = drv->bdrv_change_backing_file(bs, backing_file, backing_fmt);
756e6736 5739 } else {
469ef350 5740 ret = -ENOTSUP;
756e6736 5741 }
469ef350
PB
5742
5743 if (ret == 0) {
5744 pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_file ?: "");
5745 pstrcpy(bs->backing_format, sizeof(bs->backing_format), backing_fmt ?: "");
998c2019
HR
5746 pstrcpy(bs->auto_backing_file, sizeof(bs->auto_backing_file),
5747 backing_file ?: "");
469ef350
PB
5748 }
5749 return ret;
756e6736
KW
5750}
5751
6ebdcee2 5752/*
dcf3f9b2
HR
5753 * Finds the first non-filter node above bs in the chain between
5754 * active and bs. The returned node is either an immediate parent of
5755 * bs, or there are only filter nodes between the two.
6ebdcee2
JC
5756 *
5757 * Returns NULL if bs is not found in active's image chain,
5758 * or if active == bs.
4caf0fcd
JC
5759 *
5760 * Returns the bottommost base image if bs == NULL.
6ebdcee2
JC
5761 */
5762BlockDriverState *bdrv_find_overlay(BlockDriverState *active,
5763 BlockDriverState *bs)
5764{
f791bf7f
EGE
5765
5766 GLOBAL_STATE_CODE();
5767
dcf3f9b2
HR
5768 bs = bdrv_skip_filters(bs);
5769 active = bdrv_skip_filters(active);
5770
5771 while (active) {
5772 BlockDriverState *next = bdrv_backing_chain_next(active);
5773 if (bs == next) {
5774 return active;
5775 }
5776 active = next;
6ebdcee2
JC
5777 }
5778
dcf3f9b2 5779 return NULL;
4caf0fcd 5780}
6ebdcee2 5781
4caf0fcd
JC
5782/* Given a BDS, searches for the base layer. */
5783BlockDriverState *bdrv_find_base(BlockDriverState *bs)
5784{
f791bf7f
EGE
5785 GLOBAL_STATE_CODE();
5786
4caf0fcd 5787 return bdrv_find_overlay(bs, NULL);
6ebdcee2
JC
5788}
5789
2cad1ebe 5790/*
7b99a266
HR
5791 * Return true if at least one of the COW (backing) and filter links
5792 * between @bs and @base is frozen. @errp is set if that's the case.
0f0998f6 5793 * @base must be reachable from @bs, or NULL.
2cad1ebe
AG
5794 */
5795bool bdrv_is_backing_chain_frozen(BlockDriverState *bs, BlockDriverState *base,
5796 Error **errp)
5797{
5798 BlockDriverState *i;
7b99a266 5799 BdrvChild *child;
2cad1ebe 5800
f791bf7f
EGE
5801 GLOBAL_STATE_CODE();
5802
7b99a266
HR
5803 for (i = bs; i != base; i = child_bs(child)) {
5804 child = bdrv_filter_or_cow_child(i);
5805
5806 if (child && child->frozen) {
2cad1ebe 5807 error_setg(errp, "Cannot change '%s' link from '%s' to '%s'",
7b99a266 5808 child->name, i->node_name, child->bs->node_name);
2cad1ebe
AG
5809 return true;
5810 }
5811 }
5812
5813 return false;
5814}
5815
5816/*
7b99a266 5817 * Freeze all COW (backing) and filter links between @bs and @base.
2cad1ebe
AG
5818 * If any of the links is already frozen the operation is aborted and
5819 * none of the links are modified.
0f0998f6 5820 * @base must be reachable from @bs, or NULL.
2cad1ebe
AG
5821 * Returns 0 on success. On failure returns < 0 and sets @errp.
5822 */
5823int bdrv_freeze_backing_chain(BlockDriverState *bs, BlockDriverState *base,
5824 Error **errp)
5825{
5826 BlockDriverState *i;
7b99a266 5827 BdrvChild *child;
2cad1ebe 5828
f791bf7f
EGE
5829 GLOBAL_STATE_CODE();
5830
2cad1ebe
AG
5831 if (bdrv_is_backing_chain_frozen(bs, base, errp)) {
5832 return -EPERM;
5833 }
5834
7b99a266
HR
5835 for (i = bs; i != base; i = child_bs(child)) {
5836 child = bdrv_filter_or_cow_child(i);
5837 if (child && child->bs->never_freeze) {
e5182c1c 5838 error_setg(errp, "Cannot freeze '%s' link to '%s'",
7b99a266 5839 child->name, child->bs->node_name);
e5182c1c
HR
5840 return -EPERM;
5841 }
5842 }
5843
7b99a266
HR
5844 for (i = bs; i != base; i = child_bs(child)) {
5845 child = bdrv_filter_or_cow_child(i);
5846 if (child) {
5847 child->frozen = true;
0f0998f6 5848 }
2cad1ebe
AG
5849 }
5850
5851 return 0;
5852}
5853
5854/*
7b99a266
HR
5855 * Unfreeze all COW (backing) and filter links between @bs and @base.
5856 * The caller must ensure that all links are frozen before using this
5857 * function.
0f0998f6 5858 * @base must be reachable from @bs, or NULL.
2cad1ebe
AG
5859 */
5860void bdrv_unfreeze_backing_chain(BlockDriverState *bs, BlockDriverState *base)
5861{
5862 BlockDriverState *i;
7b99a266 5863 BdrvChild *child;
2cad1ebe 5864
f791bf7f
EGE
5865 GLOBAL_STATE_CODE();
5866
7b99a266
HR
5867 for (i = bs; i != base; i = child_bs(child)) {
5868 child = bdrv_filter_or_cow_child(i);
5869 if (child) {
5870 assert(child->frozen);
5871 child->frozen = false;
0f0998f6 5872 }
2cad1ebe
AG
5873 }
5874}
5875
6ebdcee2
JC
5876/*
5877 * Drops images above 'base' up to and including 'top', and sets the image
5878 * above 'top' to have base as its backing file.
5879 *
5880 * Requires that the overlay to 'top' is opened r/w, so that the backing file
5881 * information in 'bs' can be properly updated.
5882 *
5883 * E.g., this will convert the following chain:
5884 * bottom <- base <- intermediate <- top <- active
5885 *
5886 * to
5887 *
5888 * bottom <- base <- active
5889 *
5890 * It is allowed for bottom==base, in which case it converts:
5891 *
5892 * base <- intermediate <- top <- active
5893 *
5894 * to
5895 *
5896 * base <- active
5897 *
54e26900
JC
5898 * If backing_file_str is non-NULL, it will be used when modifying top's
5899 * overlay image metadata.
5900 *
6ebdcee2
JC
5901 * Error conditions:
5902 * if active == top, that is considered an error
5903 *
5904 */
bde70715
KW
5905int bdrv_drop_intermediate(BlockDriverState *top, BlockDriverState *base,
5906 const char *backing_file_str)
6ebdcee2 5907{
6bd858b3
AG
5908 BlockDriverState *explicit_top = top;
5909 bool update_inherits_from;
d669ed6a 5910 BdrvChild *c;
12fa4af6 5911 Error *local_err = NULL;
6ebdcee2 5912 int ret = -EIO;
d669ed6a
VSO
5913 g_autoptr(GSList) updated_children = NULL;
5914 GSList *p;
6ebdcee2 5915
f791bf7f
EGE
5916 GLOBAL_STATE_CODE();
5917
6858eba0 5918 bdrv_ref(top);
631086de 5919 bdrv_drained_begin(base);
6858eba0 5920
6ebdcee2
JC
5921 if (!top->drv || !base->drv) {
5922 goto exit;
5923 }
5924
5db15a57
KW
5925 /* Make sure that base is in the backing chain of top */
5926 if (!bdrv_chain_contains(top, base)) {
6ebdcee2
JC
5927 goto exit;
5928 }
5929
6bd858b3
AG
5930 /* If 'base' recursively inherits from 'top' then we should set
5931 * base->inherits_from to top->inherits_from after 'top' and all
5932 * other intermediate nodes have been dropped.
5933 * If 'top' is an implicit node (e.g. "commit_top") we should skip
5934 * it because no one inherits from it. We use explicit_top for that. */
dcf3f9b2 5935 explicit_top = bdrv_skip_implicit_filters(explicit_top);
6bd858b3
AG
5936 update_inherits_from = bdrv_inherits_from_recursive(base, explicit_top);
5937
6ebdcee2 5938 /* success - we can delete the intermediate states, and link top->base */
f30c66ba
HR
5939 if (!backing_file_str) {
5940 bdrv_refresh_filename(base);
5941 backing_file_str = base->filename;
5942 }
61f09cea 5943
733467a1 5944 bdrv_graph_rdlock_main_loop();
d669ed6a
VSO
5945 QLIST_FOREACH(c, &top->parents, next_parent) {
5946 updated_children = g_slist_prepend(updated_children, c);
5947 }
733467a1 5948 bdrv_graph_rdunlock_main_loop();
d669ed6a 5949
3108a15c
VSO
5950 /*
5951 * It seems correct to pass detach_subchain=true here, but it triggers
5952 * one more yet not fixed bug, when due to nested aio_poll loop we switch to
5953 * another drained section, which modify the graph (for example, removing
5954 * the child, which we keep in updated_children list). So, it's a TODO.
5955 *
5956 * Note, bug triggered if pass detach_subchain=true here and run
5957 * test-bdrv-drain. test_drop_intermediate_poll() test-case will crash.
5958 * That's a FIXME.
5959 */
5960 bdrv_replace_node_common(top, base, false, false, &local_err);
d669ed6a
VSO
5961 if (local_err) {
5962 error_report_err(local_err);
5963 goto exit;
5964 }
5965
5966 for (p = updated_children; p; p = p->next) {
5967 c = p->data;
12fa4af6 5968
bd86fb99
HR
5969 if (c->klass->update_filename) {
5970 ret = c->klass->update_filename(c, base, backing_file_str,
5971 &local_err);
61f09cea 5972 if (ret < 0) {
d669ed6a
VSO
5973 /*
5974 * TODO: Actually, we want to rollback all previous iterations
5975 * of this loop, and (which is almost impossible) previous
5976 * bdrv_replace_node()...
5977 *
5978 * Note, that c->klass->update_filename may lead to permission
5979 * update, so it's a bad idea to call it inside permission
5980 * update transaction of bdrv_replace_node.
5981 */
61f09cea
KW
5982 error_report_err(local_err);
5983 goto exit;
5984 }
5985 }
12fa4af6 5986 }
6ebdcee2 5987
6bd858b3
AG
5988 if (update_inherits_from) {
5989 base->inherits_from = explicit_top->inherits_from;
5990 }
5991
6ebdcee2 5992 ret = 0;
6ebdcee2 5993exit:
631086de 5994 bdrv_drained_end(base);
6858eba0 5995 bdrv_unref(top);
6ebdcee2
JC
5996 return ret;
5997}
5998
081e4650 5999/**
82618d7b 6000 * Implementation of BlockDriver.bdrv_co_get_allocated_file_size() that
081e4650
HR
6001 * sums the size of all data-bearing children. (This excludes backing
6002 * children.)
6003 */
de335638
EGE
6004static int64_t coroutine_fn GRAPH_RDLOCK
6005bdrv_sum_allocated_file_size(BlockDriverState *bs)
081e4650
HR
6006{
6007 BdrvChild *child;
6008 int64_t child_size, sum = 0;
6009
6010 QLIST_FOREACH(child, &bs->children, next) {
6011 if (child->role & (BDRV_CHILD_DATA | BDRV_CHILD_METADATA |
6012 BDRV_CHILD_FILTERED))
6013 {
82618d7b 6014 child_size = bdrv_co_get_allocated_file_size(child->bs);
081e4650
HR
6015 if (child_size < 0) {
6016 return child_size;
6017 }
6018 sum += child_size;
6019 }
6020 }
6021
6022 return sum;
6023}
6024
61007b31
SH
6025/**
6026 * Length of a allocated file in bytes. Sparse files are counted by actual
6027 * allocated space. Return < 0 if error or unknown.
6028 */
82618d7b 6029int64_t coroutine_fn bdrv_co_get_allocated_file_size(BlockDriverState *bs)
71d0770c 6030{
61007b31 6031 BlockDriver *drv = bs->drv;
384a48fb 6032 IO_CODE();
de335638 6033 assert_bdrv_graph_readable();
384a48fb 6034
61007b31
SH
6035 if (!drv) {
6036 return -ENOMEDIUM;
8f4754ed 6037 }
82618d7b
EGE
6038 if (drv->bdrv_co_get_allocated_file_size) {
6039 return drv->bdrv_co_get_allocated_file_size(bs);
61007b31 6040 }
081e4650
HR
6041
6042 if (drv->bdrv_file_open) {
6043 /*
6044 * Protocol drivers default to -ENOTSUP (most of their data is
6045 * not stored in any of their children (if they even have any),
6046 * so there is no generic way to figure it out).
6047 */
6048 return -ENOTSUP;
6049 } else if (drv->is_filter) {
6050 /* Filter drivers default to the size of their filtered child */
82618d7b 6051 return bdrv_co_get_allocated_file_size(bdrv_filter_bs(bs));
081e4650
HR
6052 } else {
6053 /* Other drivers default to summing their children's sizes */
6054 return bdrv_sum_allocated_file_size(bs);
1c9805a3
SH
6055 }
6056}
e7a8a783 6057
90880ff1
SH
6058/*
6059 * bdrv_measure:
6060 * @drv: Format driver
6061 * @opts: Creation options for new image
6062 * @in_bs: Existing image containing data for new image (may be NULL)
6063 * @errp: Error object
6064 * Returns: A #BlockMeasureInfo (free using qapi_free_BlockMeasureInfo())
6065 * or NULL on error
6066 *
6067 * Calculate file size required to create a new image.
6068 *
6069 * If @in_bs is given then space for allocated clusters and zero clusters
6070 * from that image are included in the calculation. If @opts contains a
6071 * backing file that is shared by @in_bs then backing clusters may be omitted
6072 * from the calculation.
6073 *
6074 * If @in_bs is NULL then the calculation includes no allocated clusters
6075 * unless a preallocation option is given in @opts.
6076 *
6077 * Note that @in_bs may use a different BlockDriver from @drv.
6078 *
6079 * If an error occurs the @errp pointer is set.
6080 */
6081BlockMeasureInfo *bdrv_measure(BlockDriver *drv, QemuOpts *opts,
6082 BlockDriverState *in_bs, Error **errp)
6083{
384a48fb 6084 IO_CODE();
90880ff1
SH
6085 if (!drv->bdrv_measure) {
6086 error_setg(errp, "Block driver '%s' does not support size measurement",
6087 drv->format_name);
6088 return NULL;
6089 }
6090
6091 return drv->bdrv_measure(opts, in_bs, errp);
6092}
6093
61007b31
SH
6094/**
6095 * Return number of sectors on success, -errno on error.
1c9805a3 6096 */
c86422c5 6097int64_t coroutine_fn bdrv_co_nb_sectors(BlockDriverState *bs)
1c9805a3 6098{
61007b31 6099 BlockDriver *drv = bs->drv;
384a48fb 6100 IO_CODE();
8ab8140a 6101 assert_bdrv_graph_readable();
498e386c 6102
61007b31
SH
6103 if (!drv)
6104 return -ENOMEDIUM;
2572b37a 6105
160a29e2 6106 if (bs->bl.has_variable_length) {
c86422c5 6107 int ret = bdrv_co_refresh_total_sectors(bs, bs->total_sectors);
61007b31
SH
6108 if (ret < 0) {
6109 return ret;
1c9805a3
SH
6110 }
6111 }
61007b31 6112 return bs->total_sectors;
1c9805a3 6113}
b338082b 6114
81f730d4
PB
6115/*
6116 * This wrapper is written by hand because this function is in the hot I/O path,
6117 * via blk_get_geometry.
6118 */
6119int64_t coroutine_mixed_fn bdrv_nb_sectors(BlockDriverState *bs)
6120{
6121 BlockDriver *drv = bs->drv;
6122 IO_CODE();
6123
6124 if (!drv)
6125 return -ENOMEDIUM;
6126
6127 if (bs->bl.has_variable_length) {
6128 int ret = bdrv_refresh_total_sectors(bs, bs->total_sectors);
6129 if (ret < 0) {
6130 return ret;
6131 }
6132 }
6133
6134 return bs->total_sectors;
6135}
6136
61007b31
SH
6137/**
6138 * Return length in bytes on success, -errno on error.
6139 * The length is always a multiple of BDRV_SECTOR_SIZE.
8d3b1a2d 6140 */
c86422c5 6141int64_t coroutine_fn bdrv_co_getlength(BlockDriverState *bs)
8d3b1a2d 6142{
c86422c5 6143 int64_t ret;
384a48fb 6144 IO_CODE();
8ab8140a 6145 assert_bdrv_graph_readable();
8d3b1a2d 6146
c86422c5 6147 ret = bdrv_co_nb_sectors(bs);
122860ba
EB
6148 if (ret < 0) {
6149 return ret;
6150 }
6151 if (ret > INT64_MAX / BDRV_SECTOR_SIZE) {
6152 return -EFBIG;
6153 }
6154 return ret * BDRV_SECTOR_SIZE;
fc01f7e7
FB
6155}
6156
54115412 6157bool bdrv_is_sg(BlockDriverState *bs)
f08145fe 6158{
384a48fb 6159 IO_CODE();
61007b31 6160 return bs->sg;
f08145fe
KW
6161}
6162
ae23f786
HR
6163/**
6164 * Return whether the given node supports compressed writes.
6165 */
6166bool bdrv_supports_compressed_writes(BlockDriverState *bs)
6167{
6168 BlockDriverState *filtered;
384a48fb 6169 IO_CODE();
ae23f786
HR
6170
6171 if (!bs->drv || !block_driver_can_compress(bs->drv)) {
6172 return false;
6173 }
6174
6175 filtered = bdrv_filter_bs(bs);
6176 if (filtered) {
6177 /*
6178 * Filters can only forward compressed writes, so we have to
6179 * check the child.
6180 */
6181 return bdrv_supports_compressed_writes(filtered);
6182 }
6183
6184 return true;
6185}
6186
61007b31 6187const char *bdrv_get_format_name(BlockDriverState *bs)
40b4f539 6188{
384a48fb 6189 IO_CODE();
61007b31 6190 return bs->drv ? bs->drv->format_name : NULL;
40b4f539
KW
6191}
6192
61007b31 6193static int qsort_strcmp(const void *a, const void *b)
40b4f539 6194{
ceff5bd7 6195 return strcmp(*(char *const *)a, *(char *const *)b);
40b4f539
KW
6196}
6197
61007b31 6198void bdrv_iterate_format(void (*it)(void *opaque, const char *name),
9ac404c5 6199 void *opaque, bool read_only)
40b4f539 6200{
61007b31
SH
6201 BlockDriver *drv;
6202 int count = 0;
6203 int i;
6204 const char **formats = NULL;
40b4f539 6205
f791bf7f
EGE
6206 GLOBAL_STATE_CODE();
6207
61007b31
SH
6208 QLIST_FOREACH(drv, &bdrv_drivers, list) {
6209 if (drv->format_name) {
6210 bool found = false;
6211 int i = count;
9ac404c5
AS
6212
6213 if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv, read_only)) {
6214 continue;
6215 }
6216
61007b31
SH
6217 while (formats && i && !found) {
6218 found = !strcmp(formats[--i], drv->format_name);
6219 }
e2a305fb 6220
61007b31
SH
6221 if (!found) {
6222 formats = g_renew(const char *, formats, count + 1);
6223 formats[count++] = drv->format_name;
6224 }
6c5a42ac 6225 }
61007b31 6226 }
6c5a42ac 6227
eb0df69f
HR
6228 for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); i++) {
6229 const char *format_name = block_driver_modules[i].format_name;
6230
6231 if (format_name) {
6232 bool found = false;
6233 int j = count;
6234
9ac404c5
AS
6235 if (use_bdrv_whitelist &&
6236 !bdrv_format_is_whitelisted(format_name, read_only)) {
6237 continue;
6238 }
6239
eb0df69f
HR
6240 while (formats && j && !found) {
6241 found = !strcmp(formats[--j], format_name);
6242 }
6243
6244 if (!found) {
6245 formats = g_renew(const char *, formats, count + 1);
6246 formats[count++] = format_name;
6247 }
6248 }
6249 }
6250
61007b31 6251 qsort(formats, count, sizeof(formats[0]), qsort_strcmp);
40b4f539 6252
61007b31
SH
6253 for (i = 0; i < count; i++) {
6254 it(opaque, formats[i]);
6255 }
40b4f539 6256
61007b31
SH
6257 g_free(formats);
6258}
40b4f539 6259
61007b31
SH
6260/* This function is to find a node in the bs graph */
6261BlockDriverState *bdrv_find_node(const char *node_name)
6262{
6263 BlockDriverState *bs;
391827eb 6264
61007b31 6265 assert(node_name);
f791bf7f 6266 GLOBAL_STATE_CODE();
40b4f539 6267
61007b31
SH
6268 QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
6269 if (!strcmp(node_name, bs->node_name)) {
6270 return bs;
40b4f539
KW
6271 }
6272 }
61007b31 6273 return NULL;
40b4f539
KW
6274}
6275
61007b31 6276/* Put this QMP function here so it can access the static graph_bdrv_states. */
facda544
PK
6277BlockDeviceInfoList *bdrv_named_nodes_list(bool flat,
6278 Error **errp)
40b4f539 6279{
9812e712 6280 BlockDeviceInfoList *list;
61007b31 6281 BlockDriverState *bs;
40b4f539 6282
f791bf7f
EGE
6283 GLOBAL_STATE_CODE();
6284
61007b31
SH
6285 list = NULL;
6286 QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
facda544 6287 BlockDeviceInfo *info = bdrv_block_device_info(NULL, bs, flat, errp);
61007b31
SH
6288 if (!info) {
6289 qapi_free_BlockDeviceInfoList(list);
6290 return NULL;
301db7c2 6291 }
9812e712 6292 QAPI_LIST_PREPEND(list, info);
301db7c2
RH
6293 }
6294
61007b31
SH
6295 return list;
6296}
40b4f539 6297
5d3b4e99
VSO
6298typedef struct XDbgBlockGraphConstructor {
6299 XDbgBlockGraph *graph;
6300 GHashTable *graph_nodes;
6301} XDbgBlockGraphConstructor;
6302
6303static XDbgBlockGraphConstructor *xdbg_graph_new(void)
6304{
6305 XDbgBlockGraphConstructor *gr = g_new(XDbgBlockGraphConstructor, 1);
6306
6307 gr->graph = g_new0(XDbgBlockGraph, 1);
6308 gr->graph_nodes = g_hash_table_new(NULL, NULL);
6309
6310 return gr;
6311}
6312
6313static XDbgBlockGraph *xdbg_graph_finalize(XDbgBlockGraphConstructor *gr)
6314{
6315 XDbgBlockGraph *graph = gr->graph;
6316
6317 g_hash_table_destroy(gr->graph_nodes);
6318 g_free(gr);
6319
6320 return graph;
6321}
6322
6323static uintptr_t xdbg_graph_node_num(XDbgBlockGraphConstructor *gr, void *node)
6324{
6325 uintptr_t ret = (uintptr_t)g_hash_table_lookup(gr->graph_nodes, node);
6326
6327 if (ret != 0) {
6328 return ret;
6329 }
6330
6331 /*
6332 * Start counting from 1, not 0, because 0 interferes with not-found (NULL)
6333 * answer of g_hash_table_lookup.
6334 */
6335 ret = g_hash_table_size(gr->graph_nodes) + 1;
6336 g_hash_table_insert(gr->graph_nodes, node, (void *)ret);
6337
6338 return ret;
6339}
6340
6341static void xdbg_graph_add_node(XDbgBlockGraphConstructor *gr, void *node,
6342 XDbgBlockGraphNodeType type, const char *name)
6343{
6344 XDbgBlockGraphNode *n;
6345
6346 n = g_new0(XDbgBlockGraphNode, 1);
6347
6348 n->id = xdbg_graph_node_num(gr, node);
6349 n->type = type;
6350 n->name = g_strdup(name);
6351
9812e712 6352 QAPI_LIST_PREPEND(gr->graph->nodes, n);
5d3b4e99
VSO
6353}
6354
6355static void xdbg_graph_add_edge(XDbgBlockGraphConstructor *gr, void *parent,
6356 const BdrvChild *child)
6357{
cdb1cec8 6358 BlockPermission qapi_perm;
5d3b4e99 6359 XDbgBlockGraphEdge *edge;
862fded9 6360 GLOBAL_STATE_CODE();
5d3b4e99 6361
5d3b4e99
VSO
6362 edge = g_new0(XDbgBlockGraphEdge, 1);
6363
6364 edge->parent = xdbg_graph_node_num(gr, parent);
6365 edge->child = xdbg_graph_node_num(gr, child->bs);
6366 edge->name = g_strdup(child->name);
6367
cdb1cec8
HR
6368 for (qapi_perm = 0; qapi_perm < BLOCK_PERMISSION__MAX; qapi_perm++) {
6369 uint64_t flag = bdrv_qapi_perm_to_blk_perm(qapi_perm);
6370
6371 if (flag & child->perm) {
9812e712 6372 QAPI_LIST_PREPEND(edge->perm, qapi_perm);
5d3b4e99 6373 }
cdb1cec8 6374 if (flag & child->shared_perm) {
9812e712 6375 QAPI_LIST_PREPEND(edge->shared_perm, qapi_perm);
5d3b4e99
VSO
6376 }
6377 }
6378
9812e712 6379 QAPI_LIST_PREPEND(gr->graph->edges, edge);
5d3b4e99
VSO
6380}
6381
6382
6383XDbgBlockGraph *bdrv_get_xdbg_block_graph(Error **errp)
6384{
6385 BlockBackend *blk;
6386 BlockJob *job;
6387 BlockDriverState *bs;
6388 BdrvChild *child;
6389 XDbgBlockGraphConstructor *gr = xdbg_graph_new();
6390
f791bf7f
EGE
6391 GLOBAL_STATE_CODE();
6392
5d3b4e99
VSO
6393 for (blk = blk_all_next(NULL); blk; blk = blk_all_next(blk)) {
6394 char *allocated_name = NULL;
6395 const char *name = blk_name(blk);
6396
6397 if (!*name) {
6398 name = allocated_name = blk_get_attached_dev_id(blk);
6399 }
6400 xdbg_graph_add_node(gr, blk, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_BACKEND,
6401 name);
6402 g_free(allocated_name);
6403 if (blk_root(blk)) {
6404 xdbg_graph_add_edge(gr, blk, blk_root(blk));
6405 }
6406 }
6407
880eeec6
EGE
6408 WITH_JOB_LOCK_GUARD() {
6409 for (job = block_job_next_locked(NULL); job;
6410 job = block_job_next_locked(job)) {
6411 GSList *el;
5d3b4e99 6412
880eeec6
EGE
6413 xdbg_graph_add_node(gr, job, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_JOB,
6414 job->job.id);
6415 for (el = job->nodes; el; el = el->next) {
6416 xdbg_graph_add_edge(gr, job, (BdrvChild *)el->data);
6417 }
5d3b4e99
VSO
6418 }
6419 }
6420
6421 QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
6422 xdbg_graph_add_node(gr, bs, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_DRIVER,
6423 bs->node_name);
6424 QLIST_FOREACH(child, &bs->children, next) {
6425 xdbg_graph_add_edge(gr, bs, child);
6426 }
6427 }
6428
6429 return xdbg_graph_finalize(gr);
6430}
6431
61007b31
SH
6432BlockDriverState *bdrv_lookup_bs(const char *device,
6433 const char *node_name,
6434 Error **errp)
6435{
6436 BlockBackend *blk;
6437 BlockDriverState *bs;
40b4f539 6438
f791bf7f
EGE
6439 GLOBAL_STATE_CODE();
6440
61007b31
SH
6441 if (device) {
6442 blk = blk_by_name(device);
40b4f539 6443
61007b31 6444 if (blk) {
9f4ed6fb
AG
6445 bs = blk_bs(blk);
6446 if (!bs) {
5433c24f 6447 error_setg(errp, "Device '%s' has no medium", device);
5433c24f
HR
6448 }
6449
9f4ed6fb 6450 return bs;
61007b31
SH
6451 }
6452 }
40b4f539 6453
61007b31
SH
6454 if (node_name) {
6455 bs = bdrv_find_node(node_name);
6d519a5f 6456
61007b31
SH
6457 if (bs) {
6458 return bs;
6459 }
40b4f539
KW
6460 }
6461
785ec4b1 6462 error_setg(errp, "Cannot find device=\'%s\' nor node-name=\'%s\'",
61007b31
SH
6463 device ? device : "",
6464 node_name ? node_name : "");
6465 return NULL;
40b4f539
KW
6466}
6467
61007b31
SH
6468/* If 'base' is in the same chain as 'top', return true. Otherwise,
6469 * return false. If either argument is NULL, return false. */
6470bool bdrv_chain_contains(BlockDriverState *top, BlockDriverState *base)
83f64091 6471{
f791bf7f
EGE
6472
6473 GLOBAL_STATE_CODE();
6474
61007b31 6475 while (top && top != base) {
dcf3f9b2 6476 top = bdrv_filter_or_cow_bs(top);
02c50efe 6477 }
61007b31
SH
6478
6479 return top != NULL;
02c50efe
FZ
6480}
6481
61007b31 6482BlockDriverState *bdrv_next_node(BlockDriverState *bs)
02c50efe 6483{
f791bf7f 6484 GLOBAL_STATE_CODE();
61007b31
SH
6485 if (!bs) {
6486 return QTAILQ_FIRST(&graph_bdrv_states);
02c50efe 6487 }
61007b31 6488 return QTAILQ_NEXT(bs, node_list);
83f64091
FB
6489}
6490
0f12264e
KW
6491BlockDriverState *bdrv_next_all_states(BlockDriverState *bs)
6492{
f791bf7f 6493 GLOBAL_STATE_CODE();
0f12264e
KW
6494 if (!bs) {
6495 return QTAILQ_FIRST(&all_bdrv_states);
6496 }
6497 return QTAILQ_NEXT(bs, bs_list);
6498}
6499
61007b31 6500const char *bdrv_get_node_name(const BlockDriverState *bs)
83f64091 6501{
384a48fb 6502 IO_CODE();
61007b31 6503 return bs->node_name;
beac80cd
FB
6504}
6505
1f0c461b 6506const char *bdrv_get_parent_name(const BlockDriverState *bs)
4c265bf9
KW
6507{
6508 BdrvChild *c;
6509 const char *name;
967d7905 6510 IO_CODE();
4c265bf9
KW
6511
6512 /* If multiple parents have a name, just pick the first one. */
6513 QLIST_FOREACH(c, &bs->parents, next_parent) {
bd86fb99
HR
6514 if (c->klass->get_name) {
6515 name = c->klass->get_name(c);
4c265bf9
KW
6516 if (name && *name) {
6517 return name;
6518 }
6519 }
6520 }
6521
6522 return NULL;
6523}
6524
61007b31
SH
6525/* TODO check what callers really want: bs->node_name or blk_name() */
6526const char *bdrv_get_device_name(const BlockDriverState *bs)
beac80cd 6527{
384a48fb 6528 IO_CODE();
4c265bf9 6529 return bdrv_get_parent_name(bs) ?: "";
f141eafe 6530}
83f64091 6531
61007b31
SH
6532/* This can be used to identify nodes that might not have a device
6533 * name associated. Since node and device names live in the same
6534 * namespace, the result is unambiguous. The exception is if both are
6535 * absent, then this returns an empty (non-null) string. */
6536const char *bdrv_get_device_or_node_name(const BlockDriverState *bs)
f141eafe 6537{
384a48fb 6538 IO_CODE();
4c265bf9 6539 return bdrv_get_parent_name(bs) ?: bs->node_name;
beac80cd 6540}
beac80cd 6541
61007b31 6542int bdrv_get_flags(BlockDriverState *bs)
0b5a2445 6543{
15aee7ac 6544 IO_CODE();
61007b31 6545 return bs->open_flags;
0b5a2445
PB
6546}
6547
61007b31 6548int bdrv_has_zero_init_1(BlockDriverState *bs)
68485420 6549{
f791bf7f 6550 GLOBAL_STATE_CODE();
61007b31 6551 return 1;
0b5a2445
PB
6552}
6553
61007b31 6554int bdrv_has_zero_init(BlockDriverState *bs)
0b5a2445 6555{
93393e69 6556 BlockDriverState *filtered;
f791bf7f 6557 GLOBAL_STATE_CODE();
93393e69 6558
d470ad42
HR
6559 if (!bs->drv) {
6560 return 0;
6561 }
0b5a2445 6562
61007b31
SH
6563 /* If BS is a copy on write image, it is initialized to
6564 the contents of the base image, which may not be zeroes. */
34778172 6565 if (bdrv_cow_child(bs)) {
61007b31
SH
6566 return 0;
6567 }
6568 if (bs->drv->bdrv_has_zero_init) {
6569 return bs->drv->bdrv_has_zero_init(bs);
0b5a2445 6570 }
93393e69
HR
6571
6572 filtered = bdrv_filter_bs(bs);
6573 if (filtered) {
6574 return bdrv_has_zero_init(filtered);
5a612c00 6575 }
61007b31
SH
6576
6577 /* safe default */
6578 return 0;
68485420
KW
6579}
6580
61007b31 6581bool bdrv_can_write_zeroes_with_unmap(BlockDriverState *bs)
68485420 6582{
384a48fb 6583 IO_CODE();
2f0342ef 6584 if (!(bs->open_flags & BDRV_O_UNMAP)) {
61007b31
SH
6585 return false;
6586 }
68485420 6587
e24d813b 6588 return bs->supported_zero_flags & BDRV_REQ_MAY_UNMAP;
68485420
KW
6589}
6590
61007b31
SH
6591void bdrv_get_backing_filename(BlockDriverState *bs,
6592 char *filename, int filename_size)
016f5cf6 6593{
384a48fb 6594 IO_CODE();
61007b31
SH
6595 pstrcpy(filename, filename_size, bs->backing_file);
6596}
d318aea9 6597
3d47eb0a 6598int coroutine_fn bdrv_co_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
61007b31 6599{
8b117001 6600 int ret;
61007b31 6601 BlockDriver *drv = bs->drv;
384a48fb 6602 IO_CODE();
a00e70c0
EGE
6603 assert_bdrv_graph_readable();
6604
5a612c00
MP
6605 /* if bs->drv == NULL, bs is closed, so there's nothing to do here */
6606 if (!drv) {
61007b31 6607 return -ENOMEDIUM;
5a612c00 6608 }
3d47eb0a 6609 if (!drv->bdrv_co_get_info) {
93393e69
HR
6610 BlockDriverState *filtered = bdrv_filter_bs(bs);
6611 if (filtered) {
3d47eb0a 6612 return bdrv_co_get_info(filtered, bdi);
5a612c00 6613 }
61007b31 6614 return -ENOTSUP;
5a612c00 6615 }
61007b31 6616 memset(bdi, 0, sizeof(*bdi));
3d47eb0a 6617 ret = drv->bdrv_co_get_info(bs, bdi);
c54483b6
AD
6618 if (bdi->subcluster_size == 0) {
6619 /*
6620 * If the driver left this unset, subclusters are not supported.
6621 * Then it is safe to treat each cluster as having only one subcluster.
6622 */
6623 bdi->subcluster_size = bdi->cluster_size;
6624 }
8b117001
VSO
6625 if (ret < 0) {
6626 return ret;
6627 }
6628
6629 if (bdi->cluster_size > BDRV_MAX_ALIGNMENT) {
6630 return -EINVAL;
6631 }
6632
6633 return 0;
61007b31 6634}
016f5cf6 6635
1bf6e9ca
AS
6636ImageInfoSpecific *bdrv_get_specific_info(BlockDriverState *bs,
6637 Error **errp)
61007b31
SH
6638{
6639 BlockDriver *drv = bs->drv;
384a48fb 6640 IO_CODE();
61007b31 6641 if (drv && drv->bdrv_get_specific_info) {
1bf6e9ca 6642 return drv->bdrv_get_specific_info(bs, errp);
61007b31
SH
6643 }
6644 return NULL;
016f5cf6
AG
6645}
6646
d9245599
AN
6647BlockStatsSpecific *bdrv_get_specific_stats(BlockDriverState *bs)
6648{
6649 BlockDriver *drv = bs->drv;
384a48fb 6650 IO_CODE();
d9245599
AN
6651 if (!drv || !drv->bdrv_get_specific_stats) {
6652 return NULL;
6653 }
6654 return drv->bdrv_get_specific_stats(bs);
6655}
6656
c834dc05 6657void coroutine_fn bdrv_co_debug_event(BlockDriverState *bs, BlkdebugEvent event)
4265d620 6658{
384a48fb 6659 IO_CODE();
cb2bfaa4
EGE
6660 assert_bdrv_graph_readable();
6661
c834dc05 6662 if (!bs || !bs->drv || !bs->drv->bdrv_co_debug_event) {
61007b31
SH
6663 return;
6664 }
4265d620 6665
c834dc05 6666 bs->drv->bdrv_co_debug_event(bs, event);
4265d620
PB
6667}
6668
d10529a2 6669static BlockDriverState *bdrv_find_debug_node(BlockDriverState *bs)
4265d620 6670{
bdb73476 6671 GLOBAL_STATE_CODE();
61007b31 6672 while (bs && bs->drv && !bs->drv->bdrv_debug_breakpoint) {
f706a92f 6673 bs = bdrv_primary_bs(bs);
61007b31 6674 }
4265d620 6675
61007b31 6676 if (bs && bs->drv && bs->drv->bdrv_debug_breakpoint) {
d10529a2
VSO
6677 assert(bs->drv->bdrv_debug_remove_breakpoint);
6678 return bs;
6679 }
6680
6681 return NULL;
6682}
6683
6684int bdrv_debug_breakpoint(BlockDriverState *bs, const char *event,
6685 const char *tag)
6686{
f791bf7f 6687 GLOBAL_STATE_CODE();
d10529a2
VSO
6688 bs = bdrv_find_debug_node(bs);
6689 if (bs) {
61007b31
SH
6690 return bs->drv->bdrv_debug_breakpoint(bs, event, tag);
6691 }
4265d620 6692
61007b31 6693 return -ENOTSUP;
4265d620
PB
6694}
6695
61007b31 6696int bdrv_debug_remove_breakpoint(BlockDriverState *bs, const char *tag)
ea2384d3 6697{
f791bf7f 6698 GLOBAL_STATE_CODE();
d10529a2
VSO
6699 bs = bdrv_find_debug_node(bs);
6700 if (bs) {
61007b31
SH
6701 return bs->drv->bdrv_debug_remove_breakpoint(bs, tag);
6702 }
6703
6704 return -ENOTSUP;
eb852011
MA
6705}
6706
61007b31 6707int bdrv_debug_resume(BlockDriverState *bs, const char *tag)
ce1a14dc 6708{
f791bf7f 6709 GLOBAL_STATE_CODE();
61007b31 6710 while (bs && (!bs->drv || !bs->drv->bdrv_debug_resume)) {
f706a92f 6711 bs = bdrv_primary_bs(bs);
61007b31 6712 }
ce1a14dc 6713
61007b31
SH
6714 if (bs && bs->drv && bs->drv->bdrv_debug_resume) {
6715 return bs->drv->bdrv_debug_resume(bs, tag);
6716 }
ce1a14dc 6717
61007b31 6718 return -ENOTSUP;
f197fe2b
FZ
6719}
6720
61007b31 6721bool bdrv_debug_is_suspended(BlockDriverState *bs, const char *tag)
ce1a14dc 6722{
f791bf7f 6723 GLOBAL_STATE_CODE();
61007b31 6724 while (bs && bs->drv && !bs->drv->bdrv_debug_is_suspended) {
f706a92f 6725 bs = bdrv_primary_bs(bs);
f197fe2b 6726 }
19cb3738 6727
61007b31
SH
6728 if (bs && bs->drv && bs->drv->bdrv_debug_is_suspended) {
6729 return bs->drv->bdrv_debug_is_suspended(bs, tag);
6730 }
f9f05dc5 6731
61007b31
SH
6732 return false;
6733}
f9f05dc5 6734
61007b31
SH
6735/* backing_file can either be relative, or absolute, or a protocol. If it is
6736 * relative, it must be relative to the chain. So, passing in bs->filename
6737 * from a BDS as backing_file should not be done, as that may be relative to
6738 * the CWD rather than the chain. */
6739BlockDriverState *bdrv_find_backing_image(BlockDriverState *bs,
6740 const char *backing_file)
f9f05dc5 6741{
61007b31
SH
6742 char *filename_full = NULL;
6743 char *backing_file_full = NULL;
6744 char *filename_tmp = NULL;
6745 int is_protocol = 0;
0b877d09 6746 bool filenames_refreshed = false;
61007b31
SH
6747 BlockDriverState *curr_bs = NULL;
6748 BlockDriverState *retval = NULL;
dcf3f9b2 6749 BlockDriverState *bs_below;
f9f05dc5 6750
f791bf7f
EGE
6751 GLOBAL_STATE_CODE();
6752
61007b31
SH
6753 if (!bs || !bs->drv || !backing_file) {
6754 return NULL;
f9f05dc5
KW
6755 }
6756
61007b31
SH
6757 filename_full = g_malloc(PATH_MAX);
6758 backing_file_full = g_malloc(PATH_MAX);
f9f05dc5 6759
61007b31 6760 is_protocol = path_has_protocol(backing_file);
f9f05dc5 6761
dcf3f9b2
HR
6762 /*
6763 * Being largely a legacy function, skip any filters here
6764 * (because filters do not have normal filenames, so they cannot
6765 * match anyway; and allowing json:{} filenames is a bit out of
6766 * scope).
6767 */
6768 for (curr_bs = bdrv_skip_filters(bs);
6769 bdrv_cow_child(curr_bs) != NULL;
6770 curr_bs = bs_below)
6771 {
6772 bs_below = bdrv_backing_chain_next(curr_bs);
f9f05dc5 6773
0b877d09
HR
6774 if (bdrv_backing_overridden(curr_bs)) {
6775 /*
6776 * If the backing file was overridden, we can only compare
6777 * directly against the backing node's filename.
6778 */
6779
6780 if (!filenames_refreshed) {
6781 /*
6782 * This will automatically refresh all of the
6783 * filenames in the rest of the backing chain, so we
6784 * only need to do this once.
6785 */
6786 bdrv_refresh_filename(bs_below);
6787 filenames_refreshed = true;
6788 }
6789
6790 if (strcmp(backing_file, bs_below->filename) == 0) {
6791 retval = bs_below;
6792 break;
6793 }
6794 } else if (is_protocol || path_has_protocol(curr_bs->backing_file)) {
6795 /*
6796 * If either of the filename paths is actually a protocol, then
6797 * compare unmodified paths; otherwise make paths relative.
6798 */
6b6833c1
HR
6799 char *backing_file_full_ret;
6800
61007b31 6801 if (strcmp(backing_file, curr_bs->backing_file) == 0) {
dcf3f9b2 6802 retval = bs_below;
61007b31
SH
6803 break;
6804 }
418661e0 6805 /* Also check against the full backing filename for the image */
6b6833c1
HR
6806 backing_file_full_ret = bdrv_get_full_backing_filename(curr_bs,
6807 NULL);
6808 if (backing_file_full_ret) {
6809 bool equal = strcmp(backing_file, backing_file_full_ret) == 0;
6810 g_free(backing_file_full_ret);
6811 if (equal) {
dcf3f9b2 6812 retval = bs_below;
418661e0
JC
6813 break;
6814 }
418661e0 6815 }
61007b31
SH
6816 } else {
6817 /* If not an absolute filename path, make it relative to the current
6818 * image's filename path */
2d9158ce
HR
6819 filename_tmp = bdrv_make_absolute_filename(curr_bs, backing_file,
6820 NULL);
6821 /* We are going to compare canonicalized absolute pathnames */
6822 if (!filename_tmp || !realpath(filename_tmp, filename_full)) {
6823 g_free(filename_tmp);
61007b31
SH
6824 continue;
6825 }
2d9158ce 6826 g_free(filename_tmp);
07f07615 6827
61007b31
SH
6828 /* We need to make sure the backing filename we are comparing against
6829 * is relative to the current image filename (or absolute) */
2d9158ce
HR
6830 filename_tmp = bdrv_get_full_backing_filename(curr_bs, NULL);
6831 if (!filename_tmp || !realpath(filename_tmp, backing_file_full)) {
6832 g_free(filename_tmp);
61007b31
SH
6833 continue;
6834 }
2d9158ce 6835 g_free(filename_tmp);
eb489bb1 6836
61007b31 6837 if (strcmp(backing_file_full, filename_full) == 0) {
dcf3f9b2 6838 retval = bs_below;
61007b31
SH
6839 break;
6840 }
6841 }
eb489bb1
KW
6842 }
6843
61007b31
SH
6844 g_free(filename_full);
6845 g_free(backing_file_full);
61007b31
SH
6846 return retval;
6847}
6848
61007b31
SH
6849void bdrv_init(void)
6850{
e5f05f8c
KW
6851#ifdef CONFIG_BDRV_WHITELIST_TOOLS
6852 use_bdrv_whitelist = 1;
6853#endif
61007b31
SH
6854 module_call_init(MODULE_INIT_BLOCK);
6855}
29cdb251 6856
61007b31
SH
6857void bdrv_init_with_whitelist(void)
6858{
6859 use_bdrv_whitelist = 1;
6860 bdrv_init();
07f07615
PB
6861}
6862
a94750d9 6863int bdrv_activate(BlockDriverState *bs, Error **errp)
0f15423c 6864{
4417ab7a 6865 BdrvChild *child, *parent;
5a8a30db
KW
6866 Error *local_err = NULL;
6867 int ret;
9c98f145 6868 BdrvDirtyBitmap *bm;
5a8a30db 6869
f791bf7f 6870 GLOBAL_STATE_CODE();
3804e3cf 6871 GRAPH_RDLOCK_GUARD_MAINLOOP();
f791bf7f 6872
3456a8d1 6873 if (!bs->drv) {
5416645f 6874 return -ENOMEDIUM;
3456a8d1
KW
6875 }
6876
16e977d5 6877 QLIST_FOREACH(child, &bs->children, next) {
11d0c9b3 6878 bdrv_activate(child->bs, &local_err);
0d1c5c91 6879 if (local_err) {
0d1c5c91 6880 error_propagate(errp, local_err);
5416645f 6881 return -EINVAL;
0d1c5c91 6882 }
5a8a30db 6883 }
0d1c5c91 6884
dafe0960
KW
6885 /*
6886 * Update permissions, they may differ for inactive nodes.
6887 *
6888 * Note that the required permissions of inactive images are always a
6889 * subset of the permissions required after activating the image. This
6890 * allows us to just get the permissions upfront without restricting
11d0c9b3 6891 * bdrv_co_invalidate_cache().
dafe0960
KW
6892 *
6893 * It also means that in error cases, we don't have to try and revert to
6894 * the old permissions (which is an operation that could fail, too). We can
6895 * just keep the extended permissions for the next time that an activation
6896 * of the image is tried.
6897 */
7bb4941a
KW
6898 if (bs->open_flags & BDRV_O_INACTIVE) {
6899 bs->open_flags &= ~BDRV_O_INACTIVE;
f1316edb 6900 ret = bdrv_refresh_perms(bs, NULL, errp);
7bb4941a 6901 if (ret < 0) {
0d1c5c91 6902 bs->open_flags |= BDRV_O_INACTIVE;
5416645f 6903 return ret;
0d1c5c91 6904 }
3456a8d1 6905
11d0c9b3
EGE
6906 ret = bdrv_invalidate_cache(bs, errp);
6907 if (ret < 0) {
6908 bs->open_flags |= BDRV_O_INACTIVE;
6909 return ret;
7bb4941a 6910 }
9c98f145 6911
7bb4941a
KW
6912 FOR_EACH_DIRTY_BITMAP(bs, bm) {
6913 bdrv_dirty_bitmap_skip_store(bm, false);
6914 }
6915
c057960c 6916 ret = bdrv_refresh_total_sectors(bs, bs->total_sectors);
7bb4941a
KW
6917 if (ret < 0) {
6918 bs->open_flags |= BDRV_O_INACTIVE;
6919 error_setg_errno(errp, -ret, "Could not refresh total sector count");
5416645f 6920 return ret;
7bb4941a 6921 }
5a8a30db 6922 }
4417ab7a
KW
6923
6924 QLIST_FOREACH(parent, &bs->parents, next_parent) {
bd86fb99
HR
6925 if (parent->klass->activate) {
6926 parent->klass->activate(parent, &local_err);
4417ab7a 6927 if (local_err) {
78fc3b3a 6928 bs->open_flags |= BDRV_O_INACTIVE;
4417ab7a 6929 error_propagate(errp, local_err);
5416645f 6930 return -EINVAL;
4417ab7a
KW
6931 }
6932 }
6933 }
5416645f
VSO
6934
6935 return 0;
0f15423c
AL
6936}
6937
11d0c9b3
EGE
6938int coroutine_fn bdrv_co_invalidate_cache(BlockDriverState *bs, Error **errp)
6939{
6940 Error *local_err = NULL;
1581a70d 6941 IO_CODE();
11d0c9b3
EGE
6942
6943 assert(!(bs->open_flags & BDRV_O_INACTIVE));
1b3ff9fe 6944 assert_bdrv_graph_readable();
11d0c9b3
EGE
6945
6946 if (bs->drv->bdrv_co_invalidate_cache) {
6947 bs->drv->bdrv_co_invalidate_cache(bs, &local_err);
6948 if (local_err) {
6949 error_propagate(errp, local_err);
6950 return -EINVAL;
6951 }
6952 }
6953
6954 return 0;
6955}
6956
3b717194 6957void bdrv_activate_all(Error **errp)
0f15423c 6958{
7c8eece4 6959 BlockDriverState *bs;
88be7b4b 6960 BdrvNextIterator it;
0f15423c 6961
f791bf7f
EGE
6962 GLOBAL_STATE_CODE();
6963
88be7b4b 6964 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
ed78cda3 6965 AioContext *aio_context = bdrv_get_aio_context(bs);
5416645f 6966 int ret;
ed78cda3
SH
6967
6968 aio_context_acquire(aio_context);
a94750d9 6969 ret = bdrv_activate(bs, errp);
ed78cda3 6970 aio_context_release(aio_context);
5416645f 6971 if (ret < 0) {
5e003f17 6972 bdrv_next_cleanup(&it);
5a8a30db
KW
6973 return;
6974 }
0f15423c
AL
6975 }
6976}
6977
9e37271f
KW
6978static bool bdrv_has_bds_parent(BlockDriverState *bs, bool only_active)
6979{
6980 BdrvChild *parent;
bdb73476 6981 GLOBAL_STATE_CODE();
9e37271f
KW
6982
6983 QLIST_FOREACH(parent, &bs->parents, next_parent) {
bd86fb99 6984 if (parent->klass->parent_is_bds) {
9e37271f
KW
6985 BlockDriverState *parent_bs = parent->opaque;
6986 if (!only_active || !(parent_bs->open_flags & BDRV_O_INACTIVE)) {
6987 return true;
6988 }
6989 }
6990 }
6991
6992 return false;
6993}
6994
6995static int bdrv_inactivate_recurse(BlockDriverState *bs)
76b1c7fe 6996{
cfa1a572 6997 BdrvChild *child, *parent;
76b1c7fe 6998 int ret;
a13de40a 6999 uint64_t cumulative_perms, cumulative_shared_perms;
76b1c7fe 7000
da359909 7001 GLOBAL_STATE_CODE();
3804e3cf 7002 GRAPH_RDLOCK_GUARD_MAINLOOP();
da359909 7003
d470ad42
HR
7004 if (!bs->drv) {
7005 return -ENOMEDIUM;
7006 }
7007
9e37271f
KW
7008 /* Make sure that we don't inactivate a child before its parent.
7009 * It will be covered by recursion from the yet active parent. */
7010 if (bdrv_has_bds_parent(bs, true)) {
7011 return 0;
7012 }
7013
7014 assert(!(bs->open_flags & BDRV_O_INACTIVE));
7015
7016 /* Inactivate this node */
7017 if (bs->drv->bdrv_inactivate) {
76b1c7fe
KW
7018 ret = bs->drv->bdrv_inactivate(bs);
7019 if (ret < 0) {
7020 return ret;
7021 }
7022 }
7023
9e37271f 7024 QLIST_FOREACH(parent, &bs->parents, next_parent) {
bd86fb99
HR
7025 if (parent->klass->inactivate) {
7026 ret = parent->klass->inactivate(parent);
9e37271f
KW
7027 if (ret < 0) {
7028 return ret;
cfa1a572
KW
7029 }
7030 }
9e37271f 7031 }
9c5e6594 7032
a13de40a
VSO
7033 bdrv_get_cumulative_perm(bs, &cumulative_perms,
7034 &cumulative_shared_perms);
7035 if (cumulative_perms & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)) {
7036 /* Our inactive parents still need write access. Inactivation failed. */
7037 return -EPERM;
7038 }
7039
9e37271f 7040 bs->open_flags |= BDRV_O_INACTIVE;
7d5b5261 7041
bb87e4d1
VSO
7042 /*
7043 * Update permissions, they may differ for inactive nodes.
7044 * We only tried to loosen restrictions, so errors are not fatal, ignore
7045 * them.
7046 */
f1316edb 7047 bdrv_refresh_perms(bs, NULL, NULL);
9e37271f
KW
7048
7049 /* Recursively inactivate children */
38701b6a 7050 QLIST_FOREACH(child, &bs->children, next) {
9e37271f 7051 ret = bdrv_inactivate_recurse(child->bs);
38701b6a
KW
7052 if (ret < 0) {
7053 return ret;
7054 }
7055 }
7056
76b1c7fe
KW
7057 return 0;
7058}
7059
7060int bdrv_inactivate_all(void)
7061{
79720af6 7062 BlockDriverState *bs = NULL;
88be7b4b 7063 BdrvNextIterator it;
aad0b7a0 7064 int ret = 0;
bd6458e4 7065 GSList *aio_ctxs = NULL, *ctx;
76b1c7fe 7066
f791bf7f
EGE
7067 GLOBAL_STATE_CODE();
7068
88be7b4b 7069 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
bd6458e4
PB
7070 AioContext *aio_context = bdrv_get_aio_context(bs);
7071
7072 if (!g_slist_find(aio_ctxs, aio_context)) {
7073 aio_ctxs = g_slist_prepend(aio_ctxs, aio_context);
7074 aio_context_acquire(aio_context);
7075 }
aad0b7a0 7076 }
76b1c7fe 7077
9e37271f
KW
7078 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
7079 /* Nodes with BDS parents are covered by recursion from the last
7080 * parent that gets inactivated. Don't inactivate them a second
7081 * time if that has already happened. */
7082 if (bdrv_has_bds_parent(bs, false)) {
7083 continue;
7084 }
7085 ret = bdrv_inactivate_recurse(bs);
7086 if (ret < 0) {
7087 bdrv_next_cleanup(&it);
7088 goto out;
76b1c7fe
KW
7089 }
7090 }
7091
aad0b7a0 7092out:
bd6458e4
PB
7093 for (ctx = aio_ctxs; ctx != NULL; ctx = ctx->next) {
7094 AioContext *aio_context = ctx->data;
7095 aio_context_release(aio_context);
aad0b7a0 7096 }
bd6458e4 7097 g_slist_free(aio_ctxs);
aad0b7a0
FZ
7098
7099 return ret;
76b1c7fe
KW
7100}
7101
19cb3738
FB
7102/**************************************************************/
7103/* removable device support */
7104
7105/**
7106 * Return TRUE if the media is present
7107 */
1e97be91 7108bool coroutine_fn bdrv_co_is_inserted(BlockDriverState *bs)
19cb3738
FB
7109{
7110 BlockDriver *drv = bs->drv;
28d7a789 7111 BdrvChild *child;
384a48fb 7112 IO_CODE();
c73ff92c 7113 assert_bdrv_graph_readable();
a1aff5bf 7114
e031f750
HR
7115 if (!drv) {
7116 return false;
7117 }
1e97be91
EGE
7118 if (drv->bdrv_co_is_inserted) {
7119 return drv->bdrv_co_is_inserted(bs);
28d7a789
HR
7120 }
7121 QLIST_FOREACH(child, &bs->children, next) {
1e97be91 7122 if (!bdrv_co_is_inserted(child->bs)) {
28d7a789
HR
7123 return false;
7124 }
e031f750 7125 }
28d7a789 7126 return true;
19cb3738
FB
7127}
7128
19cb3738
FB
7129/**
7130 * If eject_flag is TRUE, eject the media. Otherwise, close the tray
7131 */
2531b390 7132void coroutine_fn bdrv_co_eject(BlockDriverState *bs, bool eject_flag)
19cb3738
FB
7133{
7134 BlockDriver *drv = bs->drv;
384a48fb 7135 IO_CODE();
79a292e5 7136 assert_bdrv_graph_readable();
19cb3738 7137
2531b390
EGE
7138 if (drv && drv->bdrv_co_eject) {
7139 drv->bdrv_co_eject(bs, eject_flag);
19cb3738
FB
7140 }
7141}
7142
19cb3738
FB
7143/**
7144 * Lock or unlock the media (if it is locked, the user won't be able
7145 * to eject it manually).
7146 */
2c75261c 7147void coroutine_fn bdrv_co_lock_medium(BlockDriverState *bs, bool locked)
19cb3738
FB
7148{
7149 BlockDriver *drv = bs->drv;
384a48fb 7150 IO_CODE();
79a292e5 7151 assert_bdrv_graph_readable();
025e849a 7152 trace_bdrv_lock_medium(bs, locked);
b8c6d095 7153
2c75261c
EGE
7154 if (drv && drv->bdrv_co_lock_medium) {
7155 drv->bdrv_co_lock_medium(bs, locked);
19cb3738
FB
7156 }
7157}
985a03b0 7158
9fcb0251
FZ
7159/* Get a reference to bs */
7160void bdrv_ref(BlockDriverState *bs)
7161{
f791bf7f 7162 GLOBAL_STATE_CODE();
9fcb0251
FZ
7163 bs->refcnt++;
7164}
7165
7166/* Release a previously grabbed reference to bs.
7167 * If after releasing, reference count is zero, the BlockDriverState is
7168 * deleted. */
7169void bdrv_unref(BlockDriverState *bs)
7170{
f791bf7f 7171 GLOBAL_STATE_CODE();
9a4d5ca6
JC
7172 if (!bs) {
7173 return;
7174 }
9fcb0251
FZ
7175 assert(bs->refcnt > 0);
7176 if (--bs->refcnt == 0) {
7177 bdrv_delete(bs);
7178 }
7179}
7180
ac2ae233
KW
7181/*
7182 * Release a BlockDriverState reference while holding the graph write lock.
7183 *
7184 * Calling bdrv_unref() directly is forbidden while holding the graph lock
7185 * because bdrv_close() both involves polling and taking the graph lock
7186 * internally. bdrv_schedule_unref() instead delays decreasing the refcount and
7187 * possibly closing @bs until the graph lock is released.
7188 */
7189void bdrv_schedule_unref(BlockDriverState *bs)
7190{
7191 if (!bs) {
7192 return;
7193 }
7194 aio_bh_schedule_oneshot(qemu_get_aio_context(),
7195 (QEMUBHFunc *) bdrv_unref, bs);
7196}
7197
fbe40ff7
FZ
7198struct BdrvOpBlocker {
7199 Error *reason;
7200 QLIST_ENTRY(BdrvOpBlocker) list;
7201};
7202
7203bool bdrv_op_is_blocked(BlockDriverState *bs, BlockOpType op, Error **errp)
7204{
7205 BdrvOpBlocker *blocker;
f791bf7f 7206 GLOBAL_STATE_CODE();
fbe40ff7
FZ
7207 assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
7208 if (!QLIST_EMPTY(&bs->op_blockers[op])) {
7209 blocker = QLIST_FIRST(&bs->op_blockers[op]);
4b576648
MA
7210 error_propagate_prepend(errp, error_copy(blocker->reason),
7211 "Node '%s' is busy: ",
7212 bdrv_get_device_or_node_name(bs));
fbe40ff7
FZ
7213 return true;
7214 }
7215 return false;
7216}
7217
7218void bdrv_op_block(BlockDriverState *bs, BlockOpType op, Error *reason)
7219{
7220 BdrvOpBlocker *blocker;
f791bf7f 7221 GLOBAL_STATE_CODE();
fbe40ff7
FZ
7222 assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
7223
5839e53b 7224 blocker = g_new0(BdrvOpBlocker, 1);
fbe40ff7
FZ
7225 blocker->reason = reason;
7226 QLIST_INSERT_HEAD(&bs->op_blockers[op], blocker, list);
7227}
7228
7229void bdrv_op_unblock(BlockDriverState *bs, BlockOpType op, Error *reason)
7230{
7231 BdrvOpBlocker *blocker, *next;
f791bf7f 7232 GLOBAL_STATE_CODE();
fbe40ff7
FZ
7233 assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
7234 QLIST_FOREACH_SAFE(blocker, &bs->op_blockers[op], list, next) {
7235 if (blocker->reason == reason) {
7236 QLIST_REMOVE(blocker, list);
7237 g_free(blocker);
7238 }
7239 }
7240}
7241
7242void bdrv_op_block_all(BlockDriverState *bs, Error *reason)
7243{
7244 int i;
f791bf7f 7245 GLOBAL_STATE_CODE();
fbe40ff7
FZ
7246 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
7247 bdrv_op_block(bs, i, reason);
7248 }
7249}
7250
7251void bdrv_op_unblock_all(BlockDriverState *bs, Error *reason)
7252{
7253 int i;
f791bf7f 7254 GLOBAL_STATE_CODE();
fbe40ff7
FZ
7255 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
7256 bdrv_op_unblock(bs, i, reason);
7257 }
7258}
7259
7260bool bdrv_op_blocker_is_empty(BlockDriverState *bs)
7261{
7262 int i;
f791bf7f 7263 GLOBAL_STATE_CODE();
fbe40ff7
FZ
7264 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
7265 if (!QLIST_EMPTY(&bs->op_blockers[i])) {
7266 return false;
7267 }
7268 }
7269 return true;
7270}
7271
e1355055
KW
7272/*
7273 * Must not be called while holding the lock of an AioContext other than the
7274 * current one.
7275 */
d92ada22
LC
7276void bdrv_img_create(const char *filename, const char *fmt,
7277 const char *base_filename, const char *base_fmt,
9217283d
FZ
7278 char *options, uint64_t img_size, int flags, bool quiet,
7279 Error **errp)
f88e1a42 7280{
83d0521a
CL
7281 QemuOptsList *create_opts = NULL;
7282 QemuOpts *opts = NULL;
7283 const char *backing_fmt, *backing_file;
7284 int64_t size;
f88e1a42 7285 BlockDriver *drv, *proto_drv;
cc84d90f 7286 Error *local_err = NULL;
f88e1a42
JS
7287 int ret = 0;
7288
f791bf7f
EGE
7289 GLOBAL_STATE_CODE();
7290
f88e1a42
JS
7291 /* Find driver and parse its options */
7292 drv = bdrv_find_format(fmt);
7293 if (!drv) {
71c79813 7294 error_setg(errp, "Unknown file format '%s'", fmt);
d92ada22 7295 return;
f88e1a42
JS
7296 }
7297
b65a5e12 7298 proto_drv = bdrv_find_protocol(filename, true, errp);
f88e1a42 7299 if (!proto_drv) {
d92ada22 7300 return;
f88e1a42
JS
7301 }
7302
c6149724
HR
7303 if (!drv->create_opts) {
7304 error_setg(errp, "Format driver '%s' does not support image creation",
7305 drv->format_name);
7306 return;
7307 }
7308
5a5e7f8c
ML
7309 if (!proto_drv->create_opts) {
7310 error_setg(errp, "Protocol driver '%s' does not support image creation",
7311 proto_drv->format_name);
7312 return;
7313 }
7314
c6e0a6de
KW
7315 aio_context_acquire(qemu_get_aio_context());
7316
f6dc1c31 7317 /* Create parameter list */
c282e1fd 7318 create_opts = qemu_opts_append(create_opts, drv->create_opts);
5a5e7f8c 7319 create_opts = qemu_opts_append(create_opts, proto_drv->create_opts);
f88e1a42 7320
83d0521a 7321 opts = qemu_opts_create(create_opts, NULL, 0, &error_abort);
f88e1a42
JS
7322
7323 /* Parse -o options */
7324 if (options) {
a5f9b9df 7325 if (!qemu_opts_do_parse(opts, options, NULL, errp)) {
f88e1a42
JS
7326 goto out;
7327 }
7328 }
7329
f6dc1c31
KW
7330 if (!qemu_opt_get(opts, BLOCK_OPT_SIZE)) {
7331 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, img_size, &error_abort);
7332 } else if (img_size != UINT64_C(-1)) {
7333 error_setg(errp, "The image size must be specified only once");
7334 goto out;
7335 }
7336
f88e1a42 7337 if (base_filename) {
235e59cf 7338 if (!qemu_opt_set(opts, BLOCK_OPT_BACKING_FILE, base_filename,
3882578b 7339 NULL)) {
71c79813
LC
7340 error_setg(errp, "Backing file not supported for file format '%s'",
7341 fmt);
f88e1a42
JS
7342 goto out;
7343 }
7344 }
7345
7346 if (base_fmt) {
3882578b 7347 if (!qemu_opt_set(opts, BLOCK_OPT_BACKING_FMT, base_fmt, NULL)) {
71c79813
LC
7348 error_setg(errp, "Backing file format not supported for file "
7349 "format '%s'", fmt);
f88e1a42
JS
7350 goto out;
7351 }
7352 }
7353
83d0521a
CL
7354 backing_file = qemu_opt_get(opts, BLOCK_OPT_BACKING_FILE);
7355 if (backing_file) {
7356 if (!strcmp(filename, backing_file)) {
71c79813
LC
7357 error_setg(errp, "Error: Trying to create an image with the "
7358 "same filename as the backing file");
792da93a
JS
7359 goto out;
7360 }
975a7bd2
CK
7361 if (backing_file[0] == '\0') {
7362 error_setg(errp, "Expected backing file name, got empty string");
7363 goto out;
7364 }
792da93a
JS
7365 }
7366
83d0521a 7367 backing_fmt = qemu_opt_get(opts, BLOCK_OPT_BACKING_FMT);
f88e1a42 7368
6e6e55f5
JS
7369 /* The size for the image must always be specified, unless we have a backing
7370 * file and we have not been forbidden from opening it. */
a8b42a1c 7371 size = qemu_opt_get_size(opts, BLOCK_OPT_SIZE, img_size);
6e6e55f5
JS
7372 if (backing_file && !(flags & BDRV_O_NO_BACKING)) {
7373 BlockDriverState *bs;
645ae7d8 7374 char *full_backing;
6e6e55f5
JS
7375 int back_flags;
7376 QDict *backing_options = NULL;
7377
645ae7d8
HR
7378 full_backing =
7379 bdrv_get_full_backing_filename_from_filename(filename, backing_file,
7380 &local_err);
6e6e55f5 7381 if (local_err) {
6e6e55f5
JS
7382 goto out;
7383 }
645ae7d8 7384 assert(full_backing);
29168018 7385
d5b23994
HR
7386 /*
7387 * No need to do I/O here, which allows us to open encrypted
7388 * backing images without needing the secret
7389 */
6e6e55f5
JS
7390 back_flags = flags;
7391 back_flags &= ~(BDRV_O_RDWR | BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING);
d5b23994 7392 back_flags |= BDRV_O_NO_IO;
f88e1a42 7393
cc954f01 7394 backing_options = qdict_new();
6e6e55f5 7395 if (backing_fmt) {
6e6e55f5
JS
7396 qdict_put_str(backing_options, "driver", backing_fmt);
7397 }
cc954f01 7398 qdict_put_bool(backing_options, BDRV_OPT_FORCE_SHARE, true);
e6641719 7399
6e6e55f5
JS
7400 bs = bdrv_open(full_backing, NULL, backing_options, back_flags,
7401 &local_err);
7402 g_free(full_backing);
add8200d
EB
7403 if (!bs) {
7404 error_append_hint(&local_err, "Could not open backing image.\n");
6e6e55f5
JS
7405 goto out;
7406 } else {
d9f059aa 7407 if (!backing_fmt) {
497a30db
EB
7408 error_setg(&local_err,
7409 "Backing file specified without backing format");
fbdffb08 7410 error_append_hint(&local_err, "Detected format of %s.\n",
497a30db
EB
7411 bs->drv->format_name);
7412 goto out;
d9f059aa 7413 }
6e6e55f5
JS
7414 if (size == -1) {
7415 /* Opened BS, have no size */
7416 size = bdrv_getlength(bs);
7417 if (size < 0) {
7418 error_setg_errno(errp, -size, "Could not get size of '%s'",
7419 backing_file);
7420 bdrv_unref(bs);
7421 goto out;
7422 }
7423 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, size, &error_abort);
52bf1e72 7424 }
66f6b814 7425 bdrv_unref(bs);
f88e1a42 7426 }
d9f059aa
EB
7427 /* (backing_file && !(flags & BDRV_O_NO_BACKING)) */
7428 } else if (backing_file && !backing_fmt) {
497a30db
EB
7429 error_setg(&local_err,
7430 "Backing file specified without backing format");
7431 goto out;
d9f059aa 7432 }
6e6e55f5
JS
7433
7434 if (size == -1) {
7435 error_setg(errp, "Image creation needs a size parameter");
7436 goto out;
f88e1a42
JS
7437 }
7438
f382d43a 7439 if (!quiet) {
fe646693 7440 printf("Formatting '%s', fmt=%s ", filename, fmt);
43c5d8f8 7441 qemu_opts_print(opts, " ");
f382d43a 7442 puts("");
4e2f4418 7443 fflush(stdout);
f382d43a 7444 }
83d0521a 7445
c282e1fd 7446 ret = bdrv_create(drv, filename, opts, &local_err);
83d0521a 7447
cc84d90f
HR
7448 if (ret == -EFBIG) {
7449 /* This is generally a better message than whatever the driver would
7450 * deliver (especially because of the cluster_size_hint), since that
7451 * is most probably not much different from "image too large". */
7452 const char *cluster_size_hint = "";
83d0521a 7453 if (qemu_opt_get_size(opts, BLOCK_OPT_CLUSTER_SIZE, 0)) {
cc84d90f 7454 cluster_size_hint = " (try using a larger cluster size)";
f88e1a42 7455 }
cc84d90f
HR
7456 error_setg(errp, "The image size is too large for file format '%s'"
7457 "%s", fmt, cluster_size_hint);
7458 error_free(local_err);
7459 local_err = NULL;
f88e1a42
JS
7460 }
7461
7462out:
83d0521a
CL
7463 qemu_opts_del(opts);
7464 qemu_opts_free(create_opts);
621ff94d 7465 error_propagate(errp, local_err);
c6e0a6de 7466 aio_context_release(qemu_get_aio_context());
f88e1a42 7467}
85d126f3
SH
7468
7469AioContext *bdrv_get_aio_context(BlockDriverState *bs)
7470{
384a48fb 7471 IO_CODE();
33f2a757 7472 return bs ? bs->aio_context : qemu_get_aio_context();
dcd04228
SH
7473}
7474
e336fd4c
KW
7475AioContext *coroutine_fn bdrv_co_enter(BlockDriverState *bs)
7476{
7477 Coroutine *self = qemu_coroutine_self();
7478 AioContext *old_ctx = qemu_coroutine_get_aio_context(self);
7479 AioContext *new_ctx;
384a48fb 7480 IO_CODE();
e336fd4c
KW
7481
7482 /*
7483 * Increase bs->in_flight to ensure that this operation is completed before
7484 * moving the node to a different AioContext. Read new_ctx only afterwards.
7485 */
7486 bdrv_inc_in_flight(bs);
7487
7488 new_ctx = bdrv_get_aio_context(bs);
7489 aio_co_reschedule_self(new_ctx);
7490 return old_ctx;
7491}
7492
7493void coroutine_fn bdrv_co_leave(BlockDriverState *bs, AioContext *old_ctx)
7494{
384a48fb 7495 IO_CODE();
e336fd4c
KW
7496 aio_co_reschedule_self(old_ctx);
7497 bdrv_dec_in_flight(bs);
7498}
7499
18c6ac1c
KW
7500void coroutine_fn bdrv_co_lock(BlockDriverState *bs)
7501{
7502 AioContext *ctx = bdrv_get_aio_context(bs);
7503
7504 /* In the main thread, bs->aio_context won't change concurrently */
7505 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
7506
7507 /*
7508 * We're in coroutine context, so we already hold the lock of the main
7509 * loop AioContext. Don't lock it twice to avoid deadlocks.
7510 */
7511 assert(qemu_in_coroutine());
7512 if (ctx != qemu_get_aio_context()) {
7513 aio_context_acquire(ctx);
7514 }
7515}
7516
7517void coroutine_fn bdrv_co_unlock(BlockDriverState *bs)
7518{
7519 AioContext *ctx = bdrv_get_aio_context(bs);
7520
7521 assert(qemu_in_coroutine());
7522 if (ctx != qemu_get_aio_context()) {
7523 aio_context_release(ctx);
7524 }
7525}
7526
e8a095da
SH
7527static void bdrv_do_remove_aio_context_notifier(BdrvAioNotifier *ban)
7528{
bdb73476 7529 GLOBAL_STATE_CODE();
e8a095da
SH
7530 QLIST_REMOVE(ban, list);
7531 g_free(ban);
7532}
7533
a3a683c3 7534static void bdrv_detach_aio_context(BlockDriverState *bs)
dcd04228 7535{
e8a095da 7536 BdrvAioNotifier *baf, *baf_tmp;
33384421 7537
e8a095da 7538 assert(!bs->walking_aio_notifiers);
da359909 7539 GLOBAL_STATE_CODE();
e8a095da
SH
7540 bs->walking_aio_notifiers = true;
7541 QLIST_FOREACH_SAFE(baf, &bs->aio_notifiers, list, baf_tmp) {
7542 if (baf->deleted) {
7543 bdrv_do_remove_aio_context_notifier(baf);
7544 } else {
7545 baf->detach_aio_context(baf->opaque);
7546 }
33384421 7547 }
e8a095da
SH
7548 /* Never mind iterating again to check for ->deleted. bdrv_close() will
7549 * remove remaining aio notifiers if we aren't called again.
7550 */
7551 bs->walking_aio_notifiers = false;
33384421 7552
1bffe1ae 7553 if (bs->drv && bs->drv->bdrv_detach_aio_context) {
dcd04228
SH
7554 bs->drv->bdrv_detach_aio_context(bs);
7555 }
dcd04228
SH
7556
7557 bs->aio_context = NULL;
7558}
7559
a3a683c3
KW
7560static void bdrv_attach_aio_context(BlockDriverState *bs,
7561 AioContext *new_context)
dcd04228 7562{
e8a095da 7563 BdrvAioNotifier *ban, *ban_tmp;
da359909 7564 GLOBAL_STATE_CODE();
33384421 7565
dcd04228
SH
7566 bs->aio_context = new_context;
7567
1bffe1ae 7568 if (bs->drv && bs->drv->bdrv_attach_aio_context) {
dcd04228
SH
7569 bs->drv->bdrv_attach_aio_context(bs, new_context);
7570 }
33384421 7571
e8a095da
SH
7572 assert(!bs->walking_aio_notifiers);
7573 bs->walking_aio_notifiers = true;
7574 QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_tmp) {
7575 if (ban->deleted) {
7576 bdrv_do_remove_aio_context_notifier(ban);
7577 } else {
7578 ban->attached_aio_context(new_context, ban->opaque);
7579 }
33384421 7580 }
e8a095da 7581 bs->walking_aio_notifiers = false;
dcd04228
SH
7582}
7583
7e8c182f
EGE
7584typedef struct BdrvStateSetAioContext {
7585 AioContext *new_ctx;
7586 BlockDriverState *bs;
7587} BdrvStateSetAioContext;
d616b224 7588
7e8c182f 7589static bool bdrv_parent_change_aio_context(BdrvChild *c, AioContext *ctx,
e08cc001
EGE
7590 GHashTable *visited,
7591 Transaction *tran,
7e8c182f 7592 Error **errp)
5d231849 7593{
f0c28327 7594 GLOBAL_STATE_CODE();
e08cc001 7595 if (g_hash_table_contains(visited, c)) {
5d231849
KW
7596 return true;
7597 }
e08cc001 7598 g_hash_table_add(visited, c);
5d231849 7599
bd86fb99
HR
7600 /*
7601 * A BdrvChildClass that doesn't handle AioContext changes cannot
7602 * tolerate any AioContext changes
7603 */
7e8c182f 7604 if (!c->klass->change_aio_ctx) {
5d231849
KW
7605 char *user = bdrv_child_user_desc(c);
7606 error_setg(errp, "Changing iothreads is not supported by %s", user);
7607 g_free(user);
7608 return false;
7609 }
7e8c182f 7610 if (!c->klass->change_aio_ctx(c, ctx, visited, tran, errp)) {
5d231849
KW
7611 assert(!errp || *errp);
7612 return false;
7613 }
7614 return true;
7615}
7616
7e8c182f 7617bool bdrv_child_change_aio_context(BdrvChild *c, AioContext *ctx,
e08cc001 7618 GHashTable *visited, Transaction *tran,
7e8c182f 7619 Error **errp)
5d231849 7620{
f791bf7f 7621 GLOBAL_STATE_CODE();
e08cc001 7622 if (g_hash_table_contains(visited, c)) {
5d231849
KW
7623 return true;
7624 }
e08cc001 7625 g_hash_table_add(visited, c);
7e8c182f 7626 return bdrv_change_aio_context(c->bs, ctx, visited, tran, errp);
5d231849
KW
7627}
7628
7e8c182f
EGE
7629static void bdrv_set_aio_context_clean(void *opaque)
7630{
7631 BdrvStateSetAioContext *state = (BdrvStateSetAioContext *) opaque;
7632 BlockDriverState *bs = (BlockDriverState *) state->bs;
7633
7634 /* Paired with bdrv_drained_begin in bdrv_change_aio_context() */
7635 bdrv_drained_end(bs);
7636
7637 g_free(state);
7638}
7639
7640static void bdrv_set_aio_context_commit(void *opaque)
7641{
7642 BdrvStateSetAioContext *state = (BdrvStateSetAioContext *) opaque;
7643 BlockDriverState *bs = (BlockDriverState *) state->bs;
7644 AioContext *new_context = state->new_ctx;
7645 AioContext *old_context = bdrv_get_aio_context(bs);
7e8c182f
EGE
7646
7647 /*
7648 * Take the old AioContex when detaching it from bs.
7649 * At this point, new_context lock is already acquired, and we are now
7650 * also taking old_context. This is safe as long as bdrv_detach_aio_context
7651 * does not call AIO_POLL_WHILE().
7652 */
7653 if (old_context != qemu_get_aio_context()) {
7654 aio_context_acquire(old_context);
7655 }
7656 bdrv_detach_aio_context(bs);
7657 if (old_context != qemu_get_aio_context()) {
7658 aio_context_release(old_context);
7659 }
7660 bdrv_attach_aio_context(bs, new_context);
7661}
7662
7663static TransactionActionDrv set_aio_context = {
7664 .commit = bdrv_set_aio_context_commit,
7665 .clean = bdrv_set_aio_context_clean,
7666};
7667
7668/*
7669 * Changes the AioContext used for fd handlers, timers, and BHs by this
7670 * BlockDriverState and all its children and parents.
7671 *
7672 * Must be called from the main AioContext.
7673 *
7674 * The caller must own the AioContext lock for the old AioContext of bs, but it
7675 * must not own the AioContext lock for new_context (unless new_context is the
7676 * same as the current context of bs).
7677 *
7678 * @visited will accumulate all visited BdrvChild objects. The caller is
7679 * responsible for freeing the list afterwards.
7680 */
7681static bool bdrv_change_aio_context(BlockDriverState *bs, AioContext *ctx,
e08cc001 7682 GHashTable *visited, Transaction *tran,
7e8c182f 7683 Error **errp)
5d231849
KW
7684{
7685 BdrvChild *c;
7e8c182f
EGE
7686 BdrvStateSetAioContext *state;
7687
7688 GLOBAL_STATE_CODE();
5d231849
KW
7689
7690 if (bdrv_get_aio_context(bs) == ctx) {
7691 return true;
7692 }
7693
7f831d29 7694 bdrv_graph_rdlock_main_loop();
5d231849 7695 QLIST_FOREACH(c, &bs->parents, next_parent) {
7e8c182f 7696 if (!bdrv_parent_change_aio_context(c, ctx, visited, tran, errp)) {
7f831d29 7697 bdrv_graph_rdunlock_main_loop();
5d231849
KW
7698 return false;
7699 }
7700 }
7e8c182f 7701
5d231849 7702 QLIST_FOREACH(c, &bs->children, next) {
7e8c182f 7703 if (!bdrv_child_change_aio_context(c, ctx, visited, tran, errp)) {
7f831d29 7704 bdrv_graph_rdunlock_main_loop();
5d231849
KW
7705 return false;
7706 }
7707 }
7f831d29 7708 bdrv_graph_rdunlock_main_loop();
5d231849 7709
7e8c182f
EGE
7710 state = g_new(BdrvStateSetAioContext, 1);
7711 *state = (BdrvStateSetAioContext) {
7712 .new_ctx = ctx,
7713 .bs = bs,
7714 };
7715
7716 /* Paired with bdrv_drained_end in bdrv_set_aio_context_clean() */
7717 bdrv_drained_begin(bs);
7718
7719 tran_add(tran, &set_aio_context, state);
7720
5d231849
KW
7721 return true;
7722}
7723
7e8c182f
EGE
7724/*
7725 * Change bs's and recursively all of its parents' and children's AioContext
7726 * to the given new context, returning an error if that isn't possible.
7727 *
7728 * If ignore_child is not NULL, that child (and its subgraph) will not
7729 * be touched.
7730 *
7731 * This function still requires the caller to take the bs current
7732 * AioContext lock, otherwise draining will fail since AIO_WAIT_WHILE
7733 * assumes the lock is always held if bs is in another AioContext.
7734 * For the same reason, it temporarily also holds the new AioContext, since
7735 * bdrv_drained_end calls BDRV_POLL_WHILE that assumes the lock is taken too.
7736 * Therefore the new AioContext lock must not be taken by the caller.
7737 */
a41cfda1
EGE
7738int bdrv_try_change_aio_context(BlockDriverState *bs, AioContext *ctx,
7739 BdrvChild *ignore_child, Error **errp)
5d231849 7740{
7e8c182f 7741 Transaction *tran;
e08cc001 7742 GHashTable *visited;
7e8c182f
EGE
7743 int ret;
7744 AioContext *old_context = bdrv_get_aio_context(bs);
f791bf7f
EGE
7745 GLOBAL_STATE_CODE();
7746
7e8c182f
EGE
7747 /*
7748 * Recursion phase: go through all nodes of the graph.
7749 * Take care of checking that all nodes support changing AioContext
3202d8e4 7750 * and drain them, building a linear list of callbacks to run if everything
7e8c182f
EGE
7751 * is successful (the transaction itself).
7752 */
7753 tran = tran_new();
e08cc001
EGE
7754 visited = g_hash_table_new(NULL, NULL);
7755 if (ignore_child) {
7756 g_hash_table_add(visited, ignore_child);
7757 }
7758 ret = bdrv_change_aio_context(bs, ctx, visited, tran, errp);
7759 g_hash_table_destroy(visited);
7e8c182f
EGE
7760
7761 /*
7762 * Linear phase: go through all callbacks collected in the transaction.
7763 * Run all callbacks collected in the recursion to switch all nodes
7764 * AioContext lock (transaction commit), or undo all changes done in the
7765 * recursion (transaction abort).
7766 */
5d231849
KW
7767
7768 if (!ret) {
7e8c182f
EGE
7769 /* Just run clean() callbacks. No AioContext changed. */
7770 tran_abort(tran);
5d231849
KW
7771 return -EPERM;
7772 }
7773
7e8c182f
EGE
7774 /*
7775 * Release old AioContext, it won't be needed anymore, as all
7776 * bdrv_drained_begin() have been called already.
7777 */
7778 if (qemu_get_aio_context() != old_context) {
7779 aio_context_release(old_context);
7780 }
7781
7782 /*
7783 * Acquire new AioContext since bdrv_drained_end() is going to be called
7784 * after we switched all nodes in the new AioContext, and the function
7785 * assumes that the lock of the bs is always taken.
7786 */
7787 if (qemu_get_aio_context() != ctx) {
7788 aio_context_acquire(ctx);
7789 }
53a7d041 7790
7e8c182f 7791 tran_commit(tran);
5d231849 7792
7e8c182f
EGE
7793 if (qemu_get_aio_context() != ctx) {
7794 aio_context_release(ctx);
7795 }
7796
7797 /* Re-acquire the old AioContext, since the caller takes and releases it. */
7798 if (qemu_get_aio_context() != old_context) {
7799 aio_context_acquire(old_context);
7800 }
7801
7802 return 0;
5d231849
KW
7803}
7804
33384421
HR
7805void bdrv_add_aio_context_notifier(BlockDriverState *bs,
7806 void (*attached_aio_context)(AioContext *new_context, void *opaque),
7807 void (*detach_aio_context)(void *opaque), void *opaque)
7808{
7809 BdrvAioNotifier *ban = g_new(BdrvAioNotifier, 1);
7810 *ban = (BdrvAioNotifier){
7811 .attached_aio_context = attached_aio_context,
7812 .detach_aio_context = detach_aio_context,
7813 .opaque = opaque
7814 };
f791bf7f 7815 GLOBAL_STATE_CODE();
33384421
HR
7816
7817 QLIST_INSERT_HEAD(&bs->aio_notifiers, ban, list);
7818}
7819
7820void bdrv_remove_aio_context_notifier(BlockDriverState *bs,
7821 void (*attached_aio_context)(AioContext *,
7822 void *),
7823 void (*detach_aio_context)(void *),
7824 void *opaque)
7825{
7826 BdrvAioNotifier *ban, *ban_next;
f791bf7f 7827 GLOBAL_STATE_CODE();
33384421
HR
7828
7829 QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_next) {
7830 if (ban->attached_aio_context == attached_aio_context &&
7831 ban->detach_aio_context == detach_aio_context &&
e8a095da
SH
7832 ban->opaque == opaque &&
7833 ban->deleted == false)
33384421 7834 {
e8a095da
SH
7835 if (bs->walking_aio_notifiers) {
7836 ban->deleted = true;
7837 } else {
7838 bdrv_do_remove_aio_context_notifier(ban);
7839 }
33384421
HR
7840 return;
7841 }
7842 }
7843
7844 abort();
7845}
7846
77485434 7847int bdrv_amend_options(BlockDriverState *bs, QemuOpts *opts,
d1402b50 7848 BlockDriverAmendStatusCB *status_cb, void *cb_opaque,
a3579bfa 7849 bool force,
d1402b50 7850 Error **errp)
6f176b48 7851{
f791bf7f 7852 GLOBAL_STATE_CODE();
d470ad42 7853 if (!bs->drv) {
d1402b50 7854 error_setg(errp, "Node is ejected");
d470ad42
HR
7855 return -ENOMEDIUM;
7856 }
c282e1fd 7857 if (!bs->drv->bdrv_amend_options) {
d1402b50
HR
7858 error_setg(errp, "Block driver '%s' does not support option amendment",
7859 bs->drv->format_name);
6f176b48
HR
7860 return -ENOTSUP;
7861 }
a3579bfa
ML
7862 return bs->drv->bdrv_amend_options(bs, opts, status_cb,
7863 cb_opaque, force, errp);
6f176b48 7864}
f6186f49 7865
5d69b5ab
HR
7866/*
7867 * This function checks whether the given @to_replace is allowed to be
7868 * replaced by a node that always shows the same data as @bs. This is
7869 * used for example to verify whether the mirror job can replace
7870 * @to_replace by the target mirrored from @bs.
7871 * To be replaceable, @bs and @to_replace may either be guaranteed to
7872 * always show the same data (because they are only connected through
7873 * filters), or some driver may allow replacing one of its children
7874 * because it can guarantee that this child's data is not visible at
7875 * all (for example, for dissenting quorum children that have no other
7876 * parents).
7877 */
7878bool bdrv_recurse_can_replace(BlockDriverState *bs,
7879 BlockDriverState *to_replace)
7880{
93393e69
HR
7881 BlockDriverState *filtered;
7882
b4ad82aa
EGE
7883 GLOBAL_STATE_CODE();
7884
5d69b5ab
HR
7885 if (!bs || !bs->drv) {
7886 return false;
7887 }
7888
7889 if (bs == to_replace) {
7890 return true;
7891 }
7892
7893 /* See what the driver can do */
7894 if (bs->drv->bdrv_recurse_can_replace) {
7895 return bs->drv->bdrv_recurse_can_replace(bs, to_replace);
7896 }
7897
7898 /* For filters without an own implementation, we can recurse on our own */
93393e69
HR
7899 filtered = bdrv_filter_bs(bs);
7900 if (filtered) {
7901 return bdrv_recurse_can_replace(filtered, to_replace);
5d69b5ab
HR
7902 }
7903
7904 /* Safe default */
7905 return false;
7906}
7907
810803a8
HR
7908/*
7909 * Check whether the given @node_name can be replaced by a node that
7910 * has the same data as @parent_bs. If so, return @node_name's BDS;
7911 * NULL otherwise.
7912 *
7913 * @node_name must be a (recursive) *child of @parent_bs (or this
7914 * function will return NULL).
7915 *
7916 * The result (whether the node can be replaced or not) is only valid
7917 * for as long as no graph or permission changes occur.
7918 */
e12f3784
WC
7919BlockDriverState *check_to_replace_node(BlockDriverState *parent_bs,
7920 const char *node_name, Error **errp)
09158f00
BC
7921{
7922 BlockDriverState *to_replace_bs = bdrv_find_node(node_name);
5a7e7a0b
SH
7923 AioContext *aio_context;
7924
f791bf7f
EGE
7925 GLOBAL_STATE_CODE();
7926
09158f00 7927 if (!to_replace_bs) {
785ec4b1 7928 error_setg(errp, "Failed to find node with node-name='%s'", node_name);
09158f00
BC
7929 return NULL;
7930 }
7931
5a7e7a0b
SH
7932 aio_context = bdrv_get_aio_context(to_replace_bs);
7933 aio_context_acquire(aio_context);
7934
09158f00 7935 if (bdrv_op_is_blocked(to_replace_bs, BLOCK_OP_TYPE_REPLACE, errp)) {
5a7e7a0b
SH
7936 to_replace_bs = NULL;
7937 goto out;
09158f00
BC
7938 }
7939
7940 /* We don't want arbitrary node of the BDS chain to be replaced only the top
7941 * most non filter in order to prevent data corruption.
7942 * Another benefit is that this tests exclude backing files which are
7943 * blocked by the backing blockers.
7944 */
810803a8
HR
7945 if (!bdrv_recurse_can_replace(parent_bs, to_replace_bs)) {
7946 error_setg(errp, "Cannot replace '%s' by a node mirrored from '%s', "
7947 "because it cannot be guaranteed that doing so would not "
7948 "lead to an abrupt change of visible data",
7949 node_name, parent_bs->node_name);
5a7e7a0b
SH
7950 to_replace_bs = NULL;
7951 goto out;
09158f00
BC
7952 }
7953
5a7e7a0b
SH
7954out:
7955 aio_context_release(aio_context);
09158f00
BC
7956 return to_replace_bs;
7957}
448ad91d 7958
97e2f021
HR
7959/**
7960 * Iterates through the list of runtime option keys that are said to
7961 * be "strong" for a BDS. An option is called "strong" if it changes
7962 * a BDS's data. For example, the null block driver's "size" and
7963 * "read-zeroes" options are strong, but its "latency-ns" option is
7964 * not.
7965 *
7966 * If a key returned by this function ends with a dot, all options
7967 * starting with that prefix are strong.
7968 */
7969static const char *const *strong_options(BlockDriverState *bs,
7970 const char *const *curopt)
7971{
7972 static const char *const global_options[] = {
7973 "driver", "filename", NULL
7974 };
7975
7976 if (!curopt) {
7977 return &global_options[0];
7978 }
7979
7980 curopt++;
7981 if (curopt == &global_options[ARRAY_SIZE(global_options) - 1] && bs->drv) {
7982 curopt = bs->drv->strong_runtime_opts;
7983 }
7984
7985 return (curopt && *curopt) ? curopt : NULL;
7986}
7987
7988/**
7989 * Copies all strong runtime options from bs->options to the given
7990 * QDict. The set of strong option keys is determined by invoking
7991 * strong_options().
7992 *
7993 * Returns true iff any strong option was present in bs->options (and
7994 * thus copied to the target QDict) with the exception of "filename"
7995 * and "driver". The caller is expected to use this value to decide
7996 * whether the existence of strong options prevents the generation of
7997 * a plain filename.
7998 */
7999static bool append_strong_runtime_options(QDict *d, BlockDriverState *bs)
8000{
8001 bool found_any = false;
8002 const char *const *option_name = NULL;
8003
8004 if (!bs->drv) {
8005 return false;
8006 }
8007
8008 while ((option_name = strong_options(bs, option_name))) {
8009 bool option_given = false;
8010
8011 assert(strlen(*option_name) > 0);
8012 if ((*option_name)[strlen(*option_name) - 1] != '.') {
8013 QObject *entry = qdict_get(bs->options, *option_name);
8014 if (!entry) {
8015 continue;
8016 }
8017
8018 qdict_put_obj(d, *option_name, qobject_ref(entry));
8019 option_given = true;
8020 } else {
8021 const QDictEntry *entry;
8022 for (entry = qdict_first(bs->options); entry;
8023 entry = qdict_next(bs->options, entry))
8024 {
8025 if (strstart(qdict_entry_key(entry), *option_name, NULL)) {
8026 qdict_put_obj(d, qdict_entry_key(entry),
8027 qobject_ref(qdict_entry_value(entry)));
8028 option_given = true;
8029 }
8030 }
8031 }
8032
8033 /* While "driver" and "filename" need to be included in a JSON filename,
8034 * their existence does not prohibit generation of a plain filename. */
8035 if (!found_any && option_given &&
8036 strcmp(*option_name, "driver") && strcmp(*option_name, "filename"))
8037 {
8038 found_any = true;
8039 }
8040 }
8041
62a01a27
HR
8042 if (!qdict_haskey(d, "driver")) {
8043 /* Drivers created with bdrv_new_open_driver() may not have a
8044 * @driver option. Add it here. */
8045 qdict_put_str(d, "driver", bs->drv->format_name);
8046 }
8047
97e2f021
HR
8048 return found_any;
8049}
8050
90993623
HR
8051/* Note: This function may return false positives; it may return true
8052 * even if opening the backing file specified by bs's image header
8053 * would result in exactly bs->backing. */
fa8fc1d0 8054static bool bdrv_backing_overridden(BlockDriverState *bs)
90993623 8055{
b4ad82aa 8056 GLOBAL_STATE_CODE();
90993623
HR
8057 if (bs->backing) {
8058 return strcmp(bs->auto_backing_file,
8059 bs->backing->bs->filename);
8060 } else {
8061 /* No backing BDS, so if the image header reports any backing
8062 * file, it must have been suppressed */
8063 return bs->auto_backing_file[0] != '\0';
8064 }
8065}
8066
91af7014
HR
8067/* Updates the following BDS fields:
8068 * - exact_filename: A filename which may be used for opening a block device
8069 * which (mostly) equals the given BDS (even without any
8070 * other options; so reading and writing must return the same
8071 * results, but caching etc. may be different)
8072 * - full_open_options: Options which, when given when opening a block device
8073 * (without a filename), result in a BDS (mostly)
8074 * equalling the given one
8075 * - filename: If exact_filename is set, it is copied here. Otherwise,
8076 * full_open_options is converted to a JSON object, prefixed with
8077 * "json:" (for use through the JSON pseudo protocol) and put here.
8078 */
8079void bdrv_refresh_filename(BlockDriverState *bs)
8080{
8081 BlockDriver *drv = bs->drv;
e24518e3 8082 BdrvChild *child;
52f72d6f 8083 BlockDriverState *primary_child_bs;
91af7014 8084 QDict *opts;
90993623 8085 bool backing_overridden;
998b3a1e
HR
8086 bool generate_json_filename; /* Whether our default implementation should
8087 fill exact_filename (false) or not (true) */
91af7014 8088
f791bf7f
EGE
8089 GLOBAL_STATE_CODE();
8090
91af7014
HR
8091 if (!drv) {
8092 return;
8093 }
8094
e24518e3
HR
8095 /* This BDS's file name may depend on any of its children's file names, so
8096 * refresh those first */
8097 QLIST_FOREACH(child, &bs->children, next) {
8098 bdrv_refresh_filename(child->bs);
91af7014
HR
8099 }
8100
bb808d5f
HR
8101 if (bs->implicit) {
8102 /* For implicit nodes, just copy everything from the single child */
8103 child = QLIST_FIRST(&bs->children);
8104 assert(QLIST_NEXT(child, next) == NULL);
8105
8106 pstrcpy(bs->exact_filename, sizeof(bs->exact_filename),
8107 child->bs->exact_filename);
8108 pstrcpy(bs->filename, sizeof(bs->filename), child->bs->filename);
8109
cb895614 8110 qobject_unref(bs->full_open_options);
bb808d5f
HR
8111 bs->full_open_options = qobject_ref(child->bs->full_open_options);
8112
8113 return;
8114 }
8115
90993623
HR
8116 backing_overridden = bdrv_backing_overridden(bs);
8117
8118 if (bs->open_flags & BDRV_O_NO_IO) {
8119 /* Without I/O, the backing file does not change anything.
8120 * Therefore, in such a case (primarily qemu-img), we can
8121 * pretend the backing file has not been overridden even if
8122 * it technically has been. */
8123 backing_overridden = false;
8124 }
8125
97e2f021
HR
8126 /* Gather the options QDict */
8127 opts = qdict_new();
998b3a1e
HR
8128 generate_json_filename = append_strong_runtime_options(opts, bs);
8129 generate_json_filename |= backing_overridden;
97e2f021
HR
8130
8131 if (drv->bdrv_gather_child_options) {
8132 /* Some block drivers may not want to present all of their children's
8133 * options, or name them differently from BdrvChild.name */
8134 drv->bdrv_gather_child_options(bs, opts, backing_overridden);
8135 } else {
8136 QLIST_FOREACH(child, &bs->children, next) {
25191e5f 8137 if (child == bs->backing && !backing_overridden) {
97e2f021
HR
8138 /* We can skip the backing BDS if it has not been overridden */
8139 continue;
8140 }
8141
8142 qdict_put(opts, child->name,
8143 qobject_ref(child->bs->full_open_options));
8144 }
8145
8146 if (backing_overridden && !bs->backing) {
8147 /* Force no backing file */
8148 qdict_put_null(opts, "backing");
8149 }
8150 }
8151
8152 qobject_unref(bs->full_open_options);
8153 bs->full_open_options = opts;
8154
52f72d6f
HR
8155 primary_child_bs = bdrv_primary_bs(bs);
8156
998b3a1e
HR
8157 if (drv->bdrv_refresh_filename) {
8158 /* Obsolete information is of no use here, so drop the old file name
8159 * information before refreshing it */
8160 bs->exact_filename[0] = '\0';
8161
8162 drv->bdrv_refresh_filename(bs);
52f72d6f
HR
8163 } else if (primary_child_bs) {
8164 /*
8165 * Try to reconstruct valid information from the underlying
8166 * file -- this only works for format nodes (filter nodes
8167 * cannot be probed and as such must be selected by the user
8168 * either through an options dict, or through a special
8169 * filename which the filter driver must construct in its
8170 * .bdrv_refresh_filename() implementation).
8171 */
998b3a1e
HR
8172
8173 bs->exact_filename[0] = '\0';
8174
fb695c74
HR
8175 /*
8176 * We can use the underlying file's filename if:
8177 * - it has a filename,
52f72d6f 8178 * - the current BDS is not a filter,
fb695c74
HR
8179 * - the file is a protocol BDS, and
8180 * - opening that file (as this BDS's format) will automatically create
8181 * the BDS tree we have right now, that is:
8182 * - the user did not significantly change this BDS's behavior with
8183 * some explicit (strong) options
8184 * - no non-file child of this BDS has been overridden by the user
8185 * Both of these conditions are represented by generate_json_filename.
8186 */
52f72d6f
HR
8187 if (primary_child_bs->exact_filename[0] &&
8188 primary_child_bs->drv->bdrv_file_open &&
8189 !drv->is_filter && !generate_json_filename)
fb695c74 8190 {
52f72d6f 8191 strcpy(bs->exact_filename, primary_child_bs->exact_filename);
998b3a1e
HR
8192 }
8193 }
8194
91af7014
HR
8195 if (bs->exact_filename[0]) {
8196 pstrcpy(bs->filename, sizeof(bs->filename), bs->exact_filename);
97e2f021 8197 } else {
eab3a467 8198 GString *json = qobject_to_json(QOBJECT(bs->full_open_options));
5c86bdf1 8199 if (snprintf(bs->filename, sizeof(bs->filename), "json:%s",
eab3a467 8200 json->str) >= sizeof(bs->filename)) {
5c86bdf1
EB
8201 /* Give user a hint if we truncated things. */
8202 strcpy(bs->filename + sizeof(bs->filename) - 4, "...");
8203 }
eab3a467 8204 g_string_free(json, true);
91af7014
HR
8205 }
8206}
e06018ad 8207
1e89d0f9
HR
8208char *bdrv_dirname(BlockDriverState *bs, Error **errp)
8209{
8210 BlockDriver *drv = bs->drv;
52f72d6f 8211 BlockDriverState *child_bs;
1e89d0f9 8212
f791bf7f
EGE
8213 GLOBAL_STATE_CODE();
8214
1e89d0f9
HR
8215 if (!drv) {
8216 error_setg(errp, "Node '%s' is ejected", bs->node_name);
8217 return NULL;
8218 }
8219
8220 if (drv->bdrv_dirname) {
8221 return drv->bdrv_dirname(bs, errp);
8222 }
8223
52f72d6f
HR
8224 child_bs = bdrv_primary_bs(bs);
8225 if (child_bs) {
8226 return bdrv_dirname(child_bs, errp);
1e89d0f9
HR
8227 }
8228
8229 bdrv_refresh_filename(bs);
8230 if (bs->exact_filename[0] != '\0') {
8231 return path_combine(bs->exact_filename, "");
8232 }
8233
8234 error_setg(errp, "Cannot generate a base directory for %s nodes",
8235 drv->format_name);
8236 return NULL;
8237}
8238
e06018ad
WC
8239/*
8240 * Hot add/remove a BDS's child. So the user can take a child offline when
8241 * it is broken and take a new child online
8242 */
8243void bdrv_add_child(BlockDriverState *parent_bs, BlockDriverState *child_bs,
8244 Error **errp)
8245{
f791bf7f 8246 GLOBAL_STATE_CODE();
e06018ad
WC
8247 if (!parent_bs->drv || !parent_bs->drv->bdrv_add_child) {
8248 error_setg(errp, "The node %s does not support adding a child",
8249 bdrv_get_device_or_node_name(parent_bs));
8250 return;
8251 }
8252
774c726c
SL
8253 /*
8254 * Non-zoned block drivers do not follow zoned storage constraints
8255 * (i.e. sequential writes to zones). Refuse mixing zoned and non-zoned
8256 * drivers in a graph.
8257 */
8258 if (!parent_bs->drv->supports_zoned_children &&
8259 child_bs->bl.zoned == BLK_Z_HM) {
8260 /*
8261 * The host-aware model allows zoned storage constraints and random
8262 * write. Allow mixing host-aware and non-zoned drivers. Using
8263 * host-aware device as a regular device.
8264 */
8265 error_setg(errp, "Cannot add a %s child to a %s parent",
8266 child_bs->bl.zoned == BLK_Z_HM ? "zoned" : "non-zoned",
8267 parent_bs->drv->supports_zoned_children ?
8268 "support zoned children" : "not support zoned children");
8269 return;
8270 }
8271
e06018ad
WC
8272 if (!QLIST_EMPTY(&child_bs->parents)) {
8273 error_setg(errp, "The node %s already has a parent",
8274 child_bs->node_name);
8275 return;
8276 }
8277
8278 parent_bs->drv->bdrv_add_child(parent_bs, child_bs, errp);
8279}
8280
8281void bdrv_del_child(BlockDriverState *parent_bs, BdrvChild *child, Error **errp)
8282{
8283 BdrvChild *tmp;
8284
f791bf7f 8285 GLOBAL_STATE_CODE();
e06018ad
WC
8286 if (!parent_bs->drv || !parent_bs->drv->bdrv_del_child) {
8287 error_setg(errp, "The node %s does not support removing a child",
8288 bdrv_get_device_or_node_name(parent_bs));
8289 return;
8290 }
8291
8292 QLIST_FOREACH(tmp, &parent_bs->children, next) {
8293 if (tmp == child) {
8294 break;
8295 }
8296 }
8297
8298 if (!tmp) {
8299 error_setg(errp, "The node %s does not have a child named %s",
8300 bdrv_get_device_or_node_name(parent_bs),
8301 bdrv_get_device_or_node_name(child->bs));
8302 return;
8303 }
8304
8305 parent_bs->drv->bdrv_del_child(parent_bs, child, errp);
8306}
6f7a3b53
HR
8307
8308int bdrv_make_empty(BdrvChild *c, Error **errp)
8309{
8310 BlockDriver *drv = c->bs->drv;
8311 int ret;
8312
f791bf7f 8313 GLOBAL_STATE_CODE();
6f7a3b53
HR
8314 assert(c->perm & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED));
8315
8316 if (!drv->bdrv_make_empty) {
8317 error_setg(errp, "%s does not support emptying nodes",
8318 drv->format_name);
8319 return -ENOTSUP;
8320 }
8321
8322 ret = drv->bdrv_make_empty(c->bs);
8323 if (ret < 0) {
8324 error_setg_errno(errp, -ret, "Failed to empty %s",
8325 c->bs->filename);
8326 return ret;
8327 }
8328
8329 return 0;
8330}
9a6fc887
HR
8331
8332/*
8333 * Return the child that @bs acts as an overlay for, and from which data may be
8334 * copied in COW or COR operations. Usually this is the backing file.
8335 */
8336BdrvChild *bdrv_cow_child(BlockDriverState *bs)
8337{
967d7905
EGE
8338 IO_CODE();
8339
9a6fc887
HR
8340 if (!bs || !bs->drv) {
8341 return NULL;
8342 }
8343
8344 if (bs->drv->is_filter) {
8345 return NULL;
8346 }
8347
8348 if (!bs->backing) {
8349 return NULL;
8350 }
8351
8352 assert(bs->backing->role & BDRV_CHILD_COW);
8353 return bs->backing;
8354}
8355
8356/*
8357 * If @bs acts as a filter for exactly one of its children, return
8358 * that child.
8359 */
8360BdrvChild *bdrv_filter_child(BlockDriverState *bs)
8361{
8362 BdrvChild *c;
967d7905 8363 IO_CODE();
9a6fc887
HR
8364
8365 if (!bs || !bs->drv) {
8366 return NULL;
8367 }
8368
8369 if (!bs->drv->is_filter) {
8370 return NULL;
8371 }
8372
8373 /* Only one of @backing or @file may be used */
8374 assert(!(bs->backing && bs->file));
8375
8376 c = bs->backing ?: bs->file;
8377 if (!c) {
8378 return NULL;
8379 }
8380
8381 assert(c->role & BDRV_CHILD_FILTERED);
8382 return c;
8383}
8384
8385/*
8386 * Return either the result of bdrv_cow_child() or bdrv_filter_child(),
8387 * whichever is non-NULL.
8388 *
8389 * Return NULL if both are NULL.
8390 */
8391BdrvChild *bdrv_filter_or_cow_child(BlockDriverState *bs)
8392{
8393 BdrvChild *cow_child = bdrv_cow_child(bs);
8394 BdrvChild *filter_child = bdrv_filter_child(bs);
967d7905 8395 IO_CODE();
9a6fc887
HR
8396
8397 /* Filter nodes cannot have COW backing files */
8398 assert(!(cow_child && filter_child));
8399
8400 return cow_child ?: filter_child;
8401}
8402
8403/*
8404 * Return the primary child of this node: For filters, that is the
8405 * filtered child. For other nodes, that is usually the child storing
8406 * metadata.
8407 * (A generally more helpful description is that this is (usually) the
8408 * child that has the same filename as @bs.)
8409 *
8410 * Drivers do not necessarily have a primary child; for example quorum
8411 * does not.
8412 */
8413BdrvChild *bdrv_primary_child(BlockDriverState *bs)
8414{
8415 BdrvChild *c, *found = NULL;
967d7905 8416 IO_CODE();
9a6fc887
HR
8417
8418 QLIST_FOREACH(c, &bs->children, next) {
8419 if (c->role & BDRV_CHILD_PRIMARY) {
8420 assert(!found);
8421 found = c;
8422 }
8423 }
8424
8425 return found;
8426}
d38d7eb8
HR
8427
8428static BlockDriverState *bdrv_do_skip_filters(BlockDriverState *bs,
8429 bool stop_on_explicit_filter)
8430{
8431 BdrvChild *c;
8432
8433 if (!bs) {
8434 return NULL;
8435 }
8436
8437 while (!(stop_on_explicit_filter && !bs->implicit)) {
8438 c = bdrv_filter_child(bs);
8439 if (!c) {
8440 /*
8441 * A filter that is embedded in a working block graph must
8442 * have a child. Assert this here so this function does
8443 * not return a filter node that is not expected by the
8444 * caller.
8445 */
8446 assert(!bs->drv || !bs->drv->is_filter);
8447 break;
8448 }
8449 bs = c->bs;
8450 }
8451 /*
8452 * Note that this treats nodes with bs->drv == NULL as not being
8453 * filters (bs->drv == NULL should be replaced by something else
8454 * anyway).
8455 * The advantage of this behavior is that this function will thus
8456 * always return a non-NULL value (given a non-NULL @bs).
8457 */
8458
8459 return bs;
8460}
8461
8462/*
8463 * Return the first BDS that has not been added implicitly or that
8464 * does not have a filtered child down the chain starting from @bs
8465 * (including @bs itself).
8466 */
8467BlockDriverState *bdrv_skip_implicit_filters(BlockDriverState *bs)
8468{
b4ad82aa 8469 GLOBAL_STATE_CODE();
d38d7eb8
HR
8470 return bdrv_do_skip_filters(bs, true);
8471}
8472
8473/*
8474 * Return the first BDS that does not have a filtered child down the
8475 * chain starting from @bs (including @bs itself).
8476 */
8477BlockDriverState *bdrv_skip_filters(BlockDriverState *bs)
8478{
967d7905 8479 IO_CODE();
d38d7eb8
HR
8480 return bdrv_do_skip_filters(bs, false);
8481}
8482
8483/*
8484 * For a backing chain, return the first non-filter backing image of
8485 * the first non-filter image.
8486 */
8487BlockDriverState *bdrv_backing_chain_next(BlockDriverState *bs)
8488{
967d7905 8489 IO_CODE();
d38d7eb8
HR
8490 return bdrv_skip_filters(bdrv_cow_bs(bdrv_skip_filters(bs)));
8491}
0bc329fb
HR
8492
8493/**
8494 * Check whether [offset, offset + bytes) overlaps with the cached
8495 * block-status data region.
8496 *
8497 * If so, and @pnum is not NULL, set *pnum to `bsc.data_end - offset`,
8498 * which is what bdrv_bsc_is_data()'s interface needs.
8499 * Otherwise, *pnum is not touched.
8500 */
8501static bool bdrv_bsc_range_overlaps_locked(BlockDriverState *bs,
8502 int64_t offset, int64_t bytes,
8503 int64_t *pnum)
8504{
8505 BdrvBlockStatusCache *bsc = qatomic_rcu_read(&bs->block_status_cache);
8506 bool overlaps;
8507
8508 overlaps =
8509 qatomic_read(&bsc->valid) &&
8510 ranges_overlap(offset, bytes, bsc->data_start,
8511 bsc->data_end - bsc->data_start);
8512
8513 if (overlaps && pnum) {
8514 *pnum = bsc->data_end - offset;
8515 }
8516
8517 return overlaps;
8518}
8519
8520/**
8521 * See block_int.h for this function's documentation.
8522 */
8523bool bdrv_bsc_is_data(BlockDriverState *bs, int64_t offset, int64_t *pnum)
8524{
967d7905 8525 IO_CODE();
0bc329fb 8526 RCU_READ_LOCK_GUARD();
0bc329fb
HR
8527 return bdrv_bsc_range_overlaps_locked(bs, offset, 1, pnum);
8528}
8529
8530/**
8531 * See block_int.h for this function's documentation.
8532 */
8533void bdrv_bsc_invalidate_range(BlockDriverState *bs,
8534 int64_t offset, int64_t bytes)
8535{
967d7905 8536 IO_CODE();
0bc329fb
HR
8537 RCU_READ_LOCK_GUARD();
8538
8539 if (bdrv_bsc_range_overlaps_locked(bs, offset, bytes, NULL)) {
8540 qatomic_set(&bs->block_status_cache->valid, false);
8541 }
8542}
8543
8544/**
8545 * See block_int.h for this function's documentation.
8546 */
8547void bdrv_bsc_fill(BlockDriverState *bs, int64_t offset, int64_t bytes)
8548{
8549 BdrvBlockStatusCache *new_bsc = g_new(BdrvBlockStatusCache, 1);
8550 BdrvBlockStatusCache *old_bsc;
967d7905 8551 IO_CODE();
0bc329fb
HR
8552
8553 *new_bsc = (BdrvBlockStatusCache) {
8554 .valid = true,
8555 .data_start = offset,
8556 .data_end = offset + bytes,
8557 };
8558
8559 QEMU_LOCK_GUARD(&bs->bsc_modify_lock);
8560
8561 old_bsc = qatomic_rcu_read(&bs->block_status_cache);
8562 qatomic_rcu_set(&bs->block_status_cache, new_bsc);
8563 if (old_bsc) {
8564 g_free_rcu(old_bsc, rcu);
8565 }
8566}