block: use transactions as a replacement of ->{can_}set_aio_context()
[qemu.git] / block.c
CommitLineData
fc01f7e7
FB
1/*
2 * QEMU System Emulator block driver
5fafdf24 3 *
fc01f7e7 4 * Copyright (c) 2003 Fabrice Bellard
c20555e1 5 * Copyright (c) 2020 Virtuozzo International GmbH.
5fafdf24 6 *
fc01f7e7
FB
7 * Permission is hereby granted, free of charge, to any person obtaining a copy
8 * of this software and associated documentation files (the "Software"), to deal
9 * in the Software without restriction, including without limitation the rights
10 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
11 * copies of the Software, and to permit persons to whom the Software is
12 * furnished to do so, subject to the following conditions:
13 *
14 * The above copyright notice and this permission notice shall be included in
15 * all copies or substantial portions of the Software.
16 *
17 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
18 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
19 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
20 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
21 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
22 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
23 * THE SOFTWARE.
24 */
e688df6b 25
d38ea87a 26#include "qemu/osdep.h"
0ab8ed18 27#include "block/trace.h"
737e150e
PB
28#include "block/block_int.h"
29#include "block/blockjob.h"
0c9b70d5 30#include "block/fuse.h"
cd7fca95 31#include "block/nbd.h"
609f45ea 32#include "block/qdict.h"
d49b6836 33#include "qemu/error-report.h"
5e5733e5 34#include "block/module_block.h"
db725815 35#include "qemu/main-loop.h"
1de7afc9 36#include "qemu/module.h"
e688df6b 37#include "qapi/error.h"
452fcdbc 38#include "qapi/qmp/qdict.h"
7b1b5d19 39#include "qapi/qmp/qjson.h"
e59a0cf1 40#include "qapi/qmp/qnull.h"
fc81fa1e 41#include "qapi/qmp/qstring.h"
e1d74bc6
KW
42#include "qapi/qobject-output-visitor.h"
43#include "qapi/qapi-visit-block-core.h"
bfb197e0 44#include "sysemu/block-backend.h"
1de7afc9 45#include "qemu/notify.h"
922a01a0 46#include "qemu/option.h"
10817bf0 47#include "qemu/coroutine.h"
c13163fb 48#include "block/qapi.h"
1de7afc9 49#include "qemu/timer.h"
f348b6d1
VB
50#include "qemu/cutils.h"
51#include "qemu/id.h"
0bc329fb
HR
52#include "qemu/range.h"
53#include "qemu/rcu.h"
21c2283e 54#include "block/coroutines.h"
fc01f7e7 55
71e72a19 56#ifdef CONFIG_BSD
7674e7bf 57#include <sys/ioctl.h>
72cf2d4f 58#include <sys/queue.h>
feccdcee 59#if defined(HAVE_SYS_DISK_H)
7674e7bf
FB
60#include <sys/disk.h>
61#endif
c5e97233 62#endif
7674e7bf 63
49dc768d
AL
64#ifdef _WIN32
65#include <windows.h>
66#endif
67
1c9805a3
SH
68#define NOT_DONE 0x7fffffff /* used while emulated sync operation in progress */
69
3b491a90 70/* Protected by BQL */
dc364f4c
BC
71static QTAILQ_HEAD(, BlockDriverState) graph_bdrv_states =
72 QTAILQ_HEAD_INITIALIZER(graph_bdrv_states);
73
3b491a90 74/* Protected by BQL */
2c1d04e0
HR
75static QTAILQ_HEAD(, BlockDriverState) all_bdrv_states =
76 QTAILQ_HEAD_INITIALIZER(all_bdrv_states);
77
3b491a90 78/* Protected by BQL */
8a22f02a
SH
79static QLIST_HEAD(, BlockDriver) bdrv_drivers =
80 QLIST_HEAD_INITIALIZER(bdrv_drivers);
ea2384d3 81
5b363937
HR
82static BlockDriverState *bdrv_open_inherit(const char *filename,
83 const char *reference,
84 QDict *options, int flags,
85 BlockDriverState *parent,
bd86fb99 86 const BdrvChildClass *child_class,
272c02ea 87 BdrvChildRole child_role,
5b363937 88 Error **errp);
f3930ed0 89
bfb8aa6d
KW
90static bool bdrv_recurse_has_child(BlockDriverState *bs,
91 BlockDriverState *child);
92
544acc7d 93static void bdrv_replace_child_noperm(BdrvChild *child,
4eba825a 94 BlockDriverState *new_bs);
57f08941 95static void bdrv_remove_child(BdrvChild *child, Transaction *tran);
160333e1
VSO
96static void bdrv_remove_filter_or_cow_child(BlockDriverState *bs,
97 Transaction *tran);
0978623e 98
72373e40
VSO
99static int bdrv_reopen_prepare(BDRVReopenState *reopen_state,
100 BlockReopenQueue *queue,
ecd30d2d 101 Transaction *change_child_tran, Error **errp);
53e96d1e
VSO
102static void bdrv_reopen_commit(BDRVReopenState *reopen_state);
103static void bdrv_reopen_abort(BDRVReopenState *reopen_state);
104
fa8fc1d0
EGE
105static bool bdrv_backing_overridden(BlockDriverState *bs);
106
7e8c182f
EGE
107static bool bdrv_change_aio_context(BlockDriverState *bs, AioContext *ctx,
108 GSList **visited, Transaction *tran,
109 Error **errp);
110
eb852011
MA
111/* If non-zero, use only whitelisted block drivers */
112static int use_bdrv_whitelist;
113
9e0b22f4
SH
114#ifdef _WIN32
115static int is_windows_drive_prefix(const char *filename)
116{
117 return (((filename[0] >= 'a' && filename[0] <= 'z') ||
118 (filename[0] >= 'A' && filename[0] <= 'Z')) &&
119 filename[1] == ':');
120}
121
122int is_windows_drive(const char *filename)
123{
124 if (is_windows_drive_prefix(filename) &&
125 filename[2] == '\0')
126 return 1;
127 if (strstart(filename, "\\\\.\\", NULL) ||
128 strstart(filename, "//./", NULL))
129 return 1;
130 return 0;
131}
132#endif
133
339064d5
KW
134size_t bdrv_opt_mem_align(BlockDriverState *bs)
135{
136 if (!bs || !bs->drv) {
459b4e66 137 /* page size or 4k (hdd sector size) should be on the safe side */
8e3b0cbb 138 return MAX(4096, qemu_real_host_page_size());
339064d5 139 }
384a48fb 140 IO_CODE();
339064d5
KW
141
142 return bs->bl.opt_mem_alignment;
143}
144
4196d2f0
DL
145size_t bdrv_min_mem_align(BlockDriverState *bs)
146{
147 if (!bs || !bs->drv) {
459b4e66 148 /* page size or 4k (hdd sector size) should be on the safe side */
8e3b0cbb 149 return MAX(4096, qemu_real_host_page_size());
4196d2f0 150 }
384a48fb 151 IO_CODE();
4196d2f0
DL
152
153 return bs->bl.min_mem_alignment;
154}
155
9e0b22f4 156/* check if the path starts with "<protocol>:" */
5c98415b 157int path_has_protocol(const char *path)
9e0b22f4 158{
947995c0
PB
159 const char *p;
160
9e0b22f4
SH
161#ifdef _WIN32
162 if (is_windows_drive(path) ||
163 is_windows_drive_prefix(path)) {
164 return 0;
165 }
947995c0
PB
166 p = path + strcspn(path, ":/\\");
167#else
168 p = path + strcspn(path, ":/");
9e0b22f4
SH
169#endif
170
947995c0 171 return *p == ':';
9e0b22f4
SH
172}
173
83f64091 174int path_is_absolute(const char *path)
3b0d4f61 175{
21664424
FB
176#ifdef _WIN32
177 /* specific case for names like: "\\.\d:" */
f53f4da9 178 if (is_windows_drive(path) || is_windows_drive_prefix(path)) {
21664424 179 return 1;
f53f4da9
PB
180 }
181 return (*path == '/' || *path == '\\');
3b9f94e1 182#else
f53f4da9 183 return (*path == '/');
3b9f94e1 184#endif
3b0d4f61
FB
185}
186
009b03aa 187/* if filename is absolute, just return its duplicate. Otherwise, build a
83f64091
FB
188 path to it by considering it is relative to base_path. URL are
189 supported. */
009b03aa 190char *path_combine(const char *base_path, const char *filename)
3b0d4f61 191{
009b03aa 192 const char *protocol_stripped = NULL;
83f64091 193 const char *p, *p1;
009b03aa 194 char *result;
83f64091
FB
195 int len;
196
83f64091 197 if (path_is_absolute(filename)) {
009b03aa
HR
198 return g_strdup(filename);
199 }
0d54a6fe 200
009b03aa
HR
201 if (path_has_protocol(base_path)) {
202 protocol_stripped = strchr(base_path, ':');
203 if (protocol_stripped) {
204 protocol_stripped++;
0d54a6fe 205 }
009b03aa
HR
206 }
207 p = protocol_stripped ?: base_path;
0d54a6fe 208
009b03aa 209 p1 = strrchr(base_path, '/');
3b9f94e1 210#ifdef _WIN32
009b03aa
HR
211 {
212 const char *p2;
213 p2 = strrchr(base_path, '\\');
214 if (!p1 || p2 > p1) {
215 p1 = p2;
3b9f94e1 216 }
009b03aa 217 }
3b9f94e1 218#endif
009b03aa
HR
219 if (p1) {
220 p1++;
221 } else {
222 p1 = base_path;
223 }
224 if (p1 > p) {
225 p = p1;
3b0d4f61 226 }
009b03aa
HR
227 len = p - base_path;
228
229 result = g_malloc(len + strlen(filename) + 1);
230 memcpy(result, base_path, len);
231 strcpy(result + len, filename);
232
233 return result;
234}
235
03c320d8
HR
236/*
237 * Helper function for bdrv_parse_filename() implementations to remove optional
238 * protocol prefixes (especially "file:") from a filename and for putting the
239 * stripped filename into the options QDict if there is such a prefix.
240 */
241void bdrv_parse_filename_strip_prefix(const char *filename, const char *prefix,
242 QDict *options)
243{
244 if (strstart(filename, prefix, &filename)) {
245 /* Stripping the explicit protocol prefix may result in a protocol
246 * prefix being (wrongly) detected (if the filename contains a colon) */
247 if (path_has_protocol(filename)) {
18cf67c5 248 GString *fat_filename;
03c320d8
HR
249
250 /* This means there is some colon before the first slash; therefore,
251 * this cannot be an absolute path */
252 assert(!path_is_absolute(filename));
253
254 /* And we can thus fix the protocol detection issue by prefixing it
255 * by "./" */
18cf67c5
MA
256 fat_filename = g_string_new("./");
257 g_string_append(fat_filename, filename);
03c320d8 258
18cf67c5 259 assert(!path_has_protocol(fat_filename->str));
03c320d8 260
18cf67c5
MA
261 qdict_put(options, "filename",
262 qstring_from_gstring(fat_filename));
03c320d8
HR
263 } else {
264 /* If no protocol prefix was detected, we can use the shortened
265 * filename as-is */
266 qdict_put_str(options, "filename", filename);
267 }
268 }
269}
270
271
9c5e6594
KW
272/* Returns whether the image file is opened as read-only. Note that this can
273 * return false and writing to the image file is still not possible because the
274 * image is inactivated. */
93ed524e
JC
275bool bdrv_is_read_only(BlockDriverState *bs)
276{
384a48fb 277 IO_CODE();
975da073 278 return !(bs->open_flags & BDRV_O_RDWR);
93ed524e
JC
279}
280
54a32bfe
KW
281int bdrv_can_set_read_only(BlockDriverState *bs, bool read_only,
282 bool ignore_allow_rdw, Error **errp)
fe5241bf 283{
384a48fb
EGE
284 IO_CODE();
285
e2b8247a
JC
286 /* Do not set read_only if copy_on_read is enabled */
287 if (bs->copy_on_read && read_only) {
288 error_setg(errp, "Can't set node '%s' to r/o with copy-on-read enabled",
289 bdrv_get_device_or_node_name(bs));
290 return -EINVAL;
291 }
292
d6fcdf06 293 /* Do not clear read_only if it is prohibited */
54a32bfe
KW
294 if (!read_only && !(bs->open_flags & BDRV_O_ALLOW_RDWR) &&
295 !ignore_allow_rdw)
296 {
d6fcdf06
JC
297 error_setg(errp, "Node '%s' is read only",
298 bdrv_get_device_or_node_name(bs));
299 return -EPERM;
300 }
301
45803a03
JC
302 return 0;
303}
304
eaa2410f
KW
305/*
306 * Called by a driver that can only provide a read-only image.
307 *
308 * Returns 0 if the node is already read-only or it could switch the node to
309 * read-only because BDRV_O_AUTO_RDONLY is set.
310 *
311 * Returns -EACCES if the node is read-write and BDRV_O_AUTO_RDONLY is not set
312 * or bdrv_can_set_read_only() forbids making the node read-only. If @errmsg
313 * is not NULL, it is used as the error message for the Error object.
314 */
315int bdrv_apply_auto_read_only(BlockDriverState *bs, const char *errmsg,
316 Error **errp)
45803a03
JC
317{
318 int ret = 0;
384a48fb 319 IO_CODE();
45803a03 320
eaa2410f
KW
321 if (!(bs->open_flags & BDRV_O_RDWR)) {
322 return 0;
323 }
324 if (!(bs->open_flags & BDRV_O_AUTO_RDONLY)) {
325 goto fail;
45803a03
JC
326 }
327
eaa2410f
KW
328 ret = bdrv_can_set_read_only(bs, true, false, NULL);
329 if (ret < 0) {
330 goto fail;
eeae6a59
KW
331 }
332
eaa2410f
KW
333 bs->open_flags &= ~BDRV_O_RDWR;
334
e2b8247a 335 return 0;
eaa2410f
KW
336
337fail:
338 error_setg(errp, "%s", errmsg ?: "Image is read-only");
339 return -EACCES;
fe5241bf
JC
340}
341
645ae7d8
HR
342/*
343 * If @backing is empty, this function returns NULL without setting
344 * @errp. In all other cases, NULL will only be returned with @errp
345 * set.
346 *
347 * Therefore, a return value of NULL without @errp set means that
348 * there is no backing file; if @errp is set, there is one but its
349 * absolute filename cannot be generated.
350 */
351char *bdrv_get_full_backing_filename_from_filename(const char *backed,
352 const char *backing,
353 Error **errp)
dc5a1371 354{
645ae7d8
HR
355 if (backing[0] == '\0') {
356 return NULL;
357 } else if (path_has_protocol(backing) || path_is_absolute(backing)) {
358 return g_strdup(backing);
9f07429e
HR
359 } else if (backed[0] == '\0' || strstart(backed, "json:", NULL)) {
360 error_setg(errp, "Cannot use relative backing file names for '%s'",
361 backed);
645ae7d8 362 return NULL;
dc5a1371 363 } else {
645ae7d8 364 return path_combine(backed, backing);
dc5a1371
PB
365 }
366}
367
9f4793d8
HR
368/*
369 * If @filename is empty or NULL, this function returns NULL without
370 * setting @errp. In all other cases, NULL will only be returned with
371 * @errp set.
372 */
373static char *bdrv_make_absolute_filename(BlockDriverState *relative_to,
374 const char *filename, Error **errp)
0a82855a 375{
8df68616 376 char *dir, *full_name;
9f4793d8 377
8df68616
HR
378 if (!filename || filename[0] == '\0') {
379 return NULL;
380 } else if (path_has_protocol(filename) || path_is_absolute(filename)) {
381 return g_strdup(filename);
382 }
9f07429e 383
8df68616
HR
384 dir = bdrv_dirname(relative_to, errp);
385 if (!dir) {
386 return NULL;
387 }
f30c66ba 388
8df68616
HR
389 full_name = g_strconcat(dir, filename, NULL);
390 g_free(dir);
391 return full_name;
9f4793d8
HR
392}
393
394char *bdrv_get_full_backing_filename(BlockDriverState *bs, Error **errp)
395{
f791bf7f 396 GLOBAL_STATE_CODE();
9f4793d8 397 return bdrv_make_absolute_filename(bs, bs->backing_file, errp);
0a82855a
HR
398}
399
0eb7217e
SH
400void bdrv_register(BlockDriver *bdrv)
401{
a15f08dc 402 assert(bdrv->format_name);
f791bf7f 403 GLOBAL_STATE_CODE();
8a22f02a 404 QLIST_INSERT_HEAD(&bdrv_drivers, bdrv, list);
ea2384d3 405}
b338082b 406
e4e9986b
MA
407BlockDriverState *bdrv_new(void)
408{
409 BlockDriverState *bs;
410 int i;
411
f791bf7f
EGE
412 GLOBAL_STATE_CODE();
413
5839e53b 414 bs = g_new0(BlockDriverState, 1);
e4654d2d 415 QLIST_INIT(&bs->dirty_bitmaps);
fbe40ff7
FZ
416 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
417 QLIST_INIT(&bs->op_blockers[i]);
418 }
3783fa3d 419 qemu_co_mutex_init(&bs->reqs_lock);
2119882c 420 qemu_mutex_init(&bs->dirty_bitmap_mutex);
9fcb0251 421 bs->refcnt = 1;
dcd04228 422 bs->aio_context = qemu_get_aio_context();
d7d512f6 423
3ff2f67a
EY
424 qemu_co_queue_init(&bs->flush_queue);
425
0bc329fb
HR
426 qemu_co_mutex_init(&bs->bsc_modify_lock);
427 bs->block_status_cache = g_new0(BdrvBlockStatusCache, 1);
428
0f12264e
KW
429 for (i = 0; i < bdrv_drain_all_count; i++) {
430 bdrv_drained_begin(bs);
431 }
432
2c1d04e0
HR
433 QTAILQ_INSERT_TAIL(&all_bdrv_states, bs, bs_list);
434
b338082b
FB
435 return bs;
436}
437
88d88798 438static BlockDriver *bdrv_do_find_format(const char *format_name)
ea2384d3
FB
439{
440 BlockDriver *drv1;
bdb73476 441 GLOBAL_STATE_CODE();
88d88798 442
8a22f02a
SH
443 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
444 if (!strcmp(drv1->format_name, format_name)) {
ea2384d3 445 return drv1;
8a22f02a 446 }
ea2384d3 447 }
88d88798 448
ea2384d3
FB
449 return NULL;
450}
451
88d88798
MM
452BlockDriver *bdrv_find_format(const char *format_name)
453{
454 BlockDriver *drv1;
455 int i;
456
f791bf7f
EGE
457 GLOBAL_STATE_CODE();
458
88d88798
MM
459 drv1 = bdrv_do_find_format(format_name);
460 if (drv1) {
461 return drv1;
462 }
463
464 /* The driver isn't registered, maybe we need to load a module */
465 for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); ++i) {
466 if (!strcmp(block_driver_modules[i].format_name, format_name)) {
467 block_module_load_one(block_driver_modules[i].library_name);
468 break;
469 }
470 }
471
472 return bdrv_do_find_format(format_name);
473}
474
9ac404c5 475static int bdrv_format_is_whitelisted(const char *format_name, bool read_only)
eb852011 476{
b64ec4e4
FZ
477 static const char *whitelist_rw[] = {
478 CONFIG_BDRV_RW_WHITELIST
859aef02 479 NULL
b64ec4e4
FZ
480 };
481 static const char *whitelist_ro[] = {
482 CONFIG_BDRV_RO_WHITELIST
859aef02 483 NULL
eb852011
MA
484 };
485 const char **p;
486
b64ec4e4 487 if (!whitelist_rw[0] && !whitelist_ro[0]) {
eb852011 488 return 1; /* no whitelist, anything goes */
b64ec4e4 489 }
eb852011 490
b64ec4e4 491 for (p = whitelist_rw; *p; p++) {
9ac404c5 492 if (!strcmp(format_name, *p)) {
eb852011
MA
493 return 1;
494 }
495 }
b64ec4e4
FZ
496 if (read_only) {
497 for (p = whitelist_ro; *p; p++) {
9ac404c5 498 if (!strcmp(format_name, *p)) {
b64ec4e4
FZ
499 return 1;
500 }
501 }
502 }
eb852011
MA
503 return 0;
504}
505
9ac404c5
AS
506int bdrv_is_whitelisted(BlockDriver *drv, bool read_only)
507{
f791bf7f 508 GLOBAL_STATE_CODE();
9ac404c5
AS
509 return bdrv_format_is_whitelisted(drv->format_name, read_only);
510}
511
e6ff69bf
DB
512bool bdrv_uses_whitelist(void)
513{
514 return use_bdrv_whitelist;
515}
516
5b7e1542
ZYW
517typedef struct CreateCo {
518 BlockDriver *drv;
519 char *filename;
83d0521a 520 QemuOpts *opts;
5b7e1542 521 int ret;
cc84d90f 522 Error *err;
5b7e1542
ZYW
523} CreateCo;
524
525static void coroutine_fn bdrv_create_co_entry(void *opaque)
526{
cc84d90f
HR
527 Error *local_err = NULL;
528 int ret;
529
5b7e1542
ZYW
530 CreateCo *cco = opaque;
531 assert(cco->drv);
da359909 532 GLOBAL_STATE_CODE();
5b7e1542 533
b92902df
ML
534 ret = cco->drv->bdrv_co_create_opts(cco->drv,
535 cco->filename, cco->opts, &local_err);
621ff94d 536 error_propagate(&cco->err, local_err);
cc84d90f 537 cco->ret = ret;
5b7e1542
ZYW
538}
539
0e7e1989 540int bdrv_create(BlockDriver *drv, const char* filename,
83d0521a 541 QemuOpts *opts, Error **errp)
ea2384d3 542{
5b7e1542
ZYW
543 int ret;
544
f791bf7f
EGE
545 GLOBAL_STATE_CODE();
546
5b7e1542
ZYW
547 Coroutine *co;
548 CreateCo cco = {
549 .drv = drv,
550 .filename = g_strdup(filename),
83d0521a 551 .opts = opts,
5b7e1542 552 .ret = NOT_DONE,
cc84d90f 553 .err = NULL,
5b7e1542
ZYW
554 };
555
efc75e2a 556 if (!drv->bdrv_co_create_opts) {
cc84d90f 557 error_setg(errp, "Driver '%s' does not support image creation", drv->format_name);
80168bff
LC
558 ret = -ENOTSUP;
559 goto out;
5b7e1542
ZYW
560 }
561
562 if (qemu_in_coroutine()) {
563 /* Fast-path if already in coroutine context */
564 bdrv_create_co_entry(&cco);
565 } else {
0b8b8753
PB
566 co = qemu_coroutine_create(bdrv_create_co_entry, &cco);
567 qemu_coroutine_enter(co);
5b7e1542 568 while (cco.ret == NOT_DONE) {
b47ec2c4 569 aio_poll(qemu_get_aio_context(), true);
5b7e1542
ZYW
570 }
571 }
572
573 ret = cco.ret;
cc84d90f 574 if (ret < 0) {
84d18f06 575 if (cco.err) {
cc84d90f
HR
576 error_propagate(errp, cco.err);
577 } else {
578 error_setg_errno(errp, -ret, "Could not create image");
579 }
580 }
0e7e1989 581
80168bff
LC
582out:
583 g_free(cco.filename);
5b7e1542 584 return ret;
ea2384d3
FB
585}
586
fd17146c
HR
587/**
588 * Helper function for bdrv_create_file_fallback(): Resize @blk to at
589 * least the given @minimum_size.
590 *
591 * On success, return @blk's actual length.
592 * Otherwise, return -errno.
593 */
594static int64_t create_file_fallback_truncate(BlockBackend *blk,
595 int64_t minimum_size, Error **errp)
84a12e66 596{
cc84d90f 597 Error *local_err = NULL;
fd17146c 598 int64_t size;
cc84d90f 599 int ret;
84a12e66 600
bdb73476
EGE
601 GLOBAL_STATE_CODE();
602
8c6242b6
KW
603 ret = blk_truncate(blk, minimum_size, false, PREALLOC_MODE_OFF, 0,
604 &local_err);
fd17146c
HR
605 if (ret < 0 && ret != -ENOTSUP) {
606 error_propagate(errp, local_err);
607 return ret;
608 }
609
610 size = blk_getlength(blk);
611 if (size < 0) {
612 error_free(local_err);
613 error_setg_errno(errp, -size,
614 "Failed to inquire the new image file's length");
615 return size;
616 }
617
618 if (size < minimum_size) {
619 /* Need to grow the image, but we failed to do that */
620 error_propagate(errp, local_err);
621 return -ENOTSUP;
622 }
623
624 error_free(local_err);
625 local_err = NULL;
626
627 return size;
628}
629
630/**
631 * Helper function for bdrv_create_file_fallback(): Zero the first
632 * sector to remove any potentially pre-existing image header.
633 */
881a4c55
PB
634static int coroutine_fn
635create_file_fallback_zero_first_sector(BlockBackend *blk,
636 int64_t current_size,
637 Error **errp)
fd17146c
HR
638{
639 int64_t bytes_to_clear;
640 int ret;
641
bdb73476
EGE
642 GLOBAL_STATE_CODE();
643
fd17146c
HR
644 bytes_to_clear = MIN(current_size, BDRV_SECTOR_SIZE);
645 if (bytes_to_clear) {
646 ret = blk_pwrite_zeroes(blk, 0, bytes_to_clear, BDRV_REQ_MAY_UNMAP);
647 if (ret < 0) {
648 error_setg_errno(errp, -ret,
649 "Failed to clear the new image's first sector");
650 return ret;
651 }
652 }
653
654 return 0;
655}
656
5a5e7f8c
ML
657/**
658 * Simple implementation of bdrv_co_create_opts for protocol drivers
659 * which only support creation via opening a file
660 * (usually existing raw storage device)
661 */
662int coroutine_fn bdrv_co_create_opts_simple(BlockDriver *drv,
663 const char *filename,
664 QemuOpts *opts,
665 Error **errp)
fd17146c
HR
666{
667 BlockBackend *blk;
eeea1faa 668 QDict *options;
fd17146c
HR
669 int64_t size = 0;
670 char *buf = NULL;
671 PreallocMode prealloc;
672 Error *local_err = NULL;
673 int ret;
674
b4ad82aa
EGE
675 GLOBAL_STATE_CODE();
676
fd17146c
HR
677 size = qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0);
678 buf = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC);
679 prealloc = qapi_enum_parse(&PreallocMode_lookup, buf,
680 PREALLOC_MODE_OFF, &local_err);
681 g_free(buf);
682 if (local_err) {
683 error_propagate(errp, local_err);
684 return -EINVAL;
685 }
686
687 if (prealloc != PREALLOC_MODE_OFF) {
688 error_setg(errp, "Unsupported preallocation mode '%s'",
689 PreallocMode_str(prealloc));
690 return -ENOTSUP;
691 }
692
eeea1faa 693 options = qdict_new();
fd17146c
HR
694 qdict_put_str(options, "driver", drv->format_name);
695
696 blk = blk_new_open(filename, NULL, options,
697 BDRV_O_RDWR | BDRV_O_RESIZE, errp);
698 if (!blk) {
699 error_prepend(errp, "Protocol driver '%s' does not support image "
700 "creation, and opening the image failed: ",
701 drv->format_name);
702 return -EINVAL;
703 }
704
705 size = create_file_fallback_truncate(blk, size, errp);
706 if (size < 0) {
707 ret = size;
708 goto out;
709 }
710
711 ret = create_file_fallback_zero_first_sector(blk, size, errp);
712 if (ret < 0) {
713 goto out;
714 }
715
716 ret = 0;
717out:
718 blk_unref(blk);
719 return ret;
720}
721
722int bdrv_create_file(const char *filename, QemuOpts *opts, Error **errp)
723{
729222af 724 QemuOpts *protocol_opts;
fd17146c 725 BlockDriver *drv;
729222af
SG
726 QDict *qdict;
727 int ret;
fd17146c 728
f791bf7f
EGE
729 GLOBAL_STATE_CODE();
730
b65a5e12 731 drv = bdrv_find_protocol(filename, true, errp);
84a12e66 732 if (drv == NULL) {
16905d71 733 return -ENOENT;
84a12e66
CH
734 }
735
729222af
SG
736 if (!drv->create_opts) {
737 error_setg(errp, "Driver '%s' does not support image creation",
738 drv->format_name);
739 return -ENOTSUP;
740 }
741
742 /*
743 * 'opts' contains a QemuOptsList with a combination of format and protocol
744 * default values.
745 *
746 * The format properly removes its options, but the default values remain
747 * in 'opts->list'. So if the protocol has options with the same name
748 * (e.g. rbd has 'cluster_size' as qcow2), it will see the default values
749 * of the format, since for overlapping options, the format wins.
750 *
751 * To avoid this issue, lets convert QemuOpts to QDict, in this way we take
752 * only the set options, and then convert it back to QemuOpts, using the
753 * create_opts of the protocol. So the new QemuOpts, will contain only the
754 * protocol defaults.
755 */
756 qdict = qemu_opts_to_qdict(opts, NULL);
757 protocol_opts = qemu_opts_from_qdict(drv->create_opts, qdict, errp);
758 if (protocol_opts == NULL) {
759 ret = -EINVAL;
760 goto out;
761 }
762
763 ret = bdrv_create(drv, filename, protocol_opts, errp);
764out:
765 qemu_opts_del(protocol_opts);
766 qobject_unref(qdict);
767 return ret;
84a12e66
CH
768}
769
e1d7f8bb
DHB
770int coroutine_fn bdrv_co_delete_file(BlockDriverState *bs, Error **errp)
771{
772 Error *local_err = NULL;
773 int ret;
774
384a48fb 775 IO_CODE();
e1d7f8bb
DHB
776 assert(bs != NULL);
777
778 if (!bs->drv) {
779 error_setg(errp, "Block node '%s' is not opened", bs->filename);
780 return -ENOMEDIUM;
781 }
782
783 if (!bs->drv->bdrv_co_delete_file) {
784 error_setg(errp, "Driver '%s' does not support image deletion",
785 bs->drv->format_name);
786 return -ENOTSUP;
787 }
788
789 ret = bs->drv->bdrv_co_delete_file(bs, &local_err);
790 if (ret < 0) {
791 error_propagate(errp, local_err);
792 }
793
794 return ret;
795}
796
a890f08e
ML
797void coroutine_fn bdrv_co_delete_file_noerr(BlockDriverState *bs)
798{
799 Error *local_err = NULL;
800 int ret;
384a48fb 801 IO_CODE();
a890f08e
ML
802
803 if (!bs) {
804 return;
805 }
806
807 ret = bdrv_co_delete_file(bs, &local_err);
808 /*
809 * ENOTSUP will happen if the block driver doesn't support
810 * the 'bdrv_co_delete_file' interface. This is a predictable
811 * scenario and shouldn't be reported back to the user.
812 */
813 if (ret == -ENOTSUP) {
814 error_free(local_err);
815 } else if (ret < 0) {
816 error_report_err(local_err);
817 }
818}
819
892b7de8
ET
820/**
821 * Try to get @bs's logical and physical block size.
822 * On success, store them in @bsz struct and return 0.
823 * On failure return -errno.
824 * @bs must not be empty.
825 */
826int bdrv_probe_blocksizes(BlockDriverState *bs, BlockSizes *bsz)
827{
828 BlockDriver *drv = bs->drv;
93393e69 829 BlockDriverState *filtered = bdrv_filter_bs(bs);
f791bf7f 830 GLOBAL_STATE_CODE();
892b7de8
ET
831
832 if (drv && drv->bdrv_probe_blocksizes) {
833 return drv->bdrv_probe_blocksizes(bs, bsz);
93393e69
HR
834 } else if (filtered) {
835 return bdrv_probe_blocksizes(filtered, bsz);
892b7de8
ET
836 }
837
838 return -ENOTSUP;
839}
840
841/**
842 * Try to get @bs's geometry (cyls, heads, sectors).
843 * On success, store them in @geo struct and return 0.
844 * On failure return -errno.
845 * @bs must not be empty.
846 */
847int bdrv_probe_geometry(BlockDriverState *bs, HDGeometry *geo)
848{
849 BlockDriver *drv = bs->drv;
93393e69 850 BlockDriverState *filtered = bdrv_filter_bs(bs);
f791bf7f 851 GLOBAL_STATE_CODE();
892b7de8
ET
852
853 if (drv && drv->bdrv_probe_geometry) {
854 return drv->bdrv_probe_geometry(bs, geo);
93393e69
HR
855 } else if (filtered) {
856 return bdrv_probe_geometry(filtered, geo);
892b7de8
ET
857 }
858
859 return -ENOTSUP;
860}
861
eba25057
JM
862/*
863 * Create a uniquely-named empty temporary file.
69fbfff9
BM
864 * Return the actual file name used upon success, otherwise NULL.
865 * This string should be freed with g_free() when not needed any longer.
866 *
867 * Note: creating a temporary file for the caller to (re)open is
868 * inherently racy. Use g_file_open_tmp() instead whenever practical.
eba25057 869 */
69fbfff9 870char *create_tmp_file(Error **errp)
d5249393 871{
67b915a5 872 int fd;
7ccfb2eb 873 const char *tmpdir;
69fbfff9
BM
874 g_autofree char *filename = NULL;
875
876 tmpdir = g_get_tmp_dir();
877#ifndef _WIN32
878 /*
879 * See commit 69bef79 ("block: use /var/tmp instead of /tmp for -snapshot")
880 *
881 * This function is used to create temporary disk images (like -snapshot),
882 * so the files can become very large. /tmp is often a tmpfs where as
883 * /var/tmp is usually on a disk, so more appropriate for disk images.
884 */
885 if (!g_strcmp0(tmpdir, "/tmp")) {
69bef793
AS
886 tmpdir = "/var/tmp";
887 }
69fbfff9
BM
888#endif
889
890 filename = g_strdup_printf("%s/vl.XXXXXX", tmpdir);
891 fd = g_mkstemp(filename);
fe235a06 892 if (fd < 0) {
69fbfff9
BM
893 error_setg_errno(errp, errno, "Could not open temporary file '%s'",
894 filename);
895 return NULL;
fe235a06 896 }
6b6471ee 897 close(fd);
69fbfff9
BM
898
899 return g_steal_pointer(&filename);
eba25057 900}
fc01f7e7 901
84a12e66
CH
902/*
903 * Detect host devices. By convention, /dev/cdrom[N] is always
904 * recognized as a host CDROM.
905 */
906static BlockDriver *find_hdev_driver(const char *filename)
907{
908 int score_max = 0, score;
909 BlockDriver *drv = NULL, *d;
bdb73476 910 GLOBAL_STATE_CODE();
84a12e66
CH
911
912 QLIST_FOREACH(d, &bdrv_drivers, list) {
913 if (d->bdrv_probe_device) {
914 score = d->bdrv_probe_device(filename);
915 if (score > score_max) {
916 score_max = score;
917 drv = d;
918 }
919 }
920 }
921
922 return drv;
923}
924
88d88798
MM
925static BlockDriver *bdrv_do_find_protocol(const char *protocol)
926{
927 BlockDriver *drv1;
bdb73476 928 GLOBAL_STATE_CODE();
88d88798
MM
929
930 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
931 if (drv1->protocol_name && !strcmp(drv1->protocol_name, protocol)) {
932 return drv1;
933 }
934 }
935
936 return NULL;
937}
938
98289620 939BlockDriver *bdrv_find_protocol(const char *filename,
b65a5e12
HR
940 bool allow_protocol_prefix,
941 Error **errp)
83f64091
FB
942{
943 BlockDriver *drv1;
944 char protocol[128];
1cec71e3 945 int len;
83f64091 946 const char *p;
88d88798 947 int i;
19cb3738 948
f791bf7f 949 GLOBAL_STATE_CODE();
66f82cee
KW
950 /* TODO Drivers without bdrv_file_open must be specified explicitly */
951
39508e7a
CH
952 /*
953 * XXX(hch): we really should not let host device detection
954 * override an explicit protocol specification, but moving this
955 * later breaks access to device names with colons in them.
956 * Thanks to the brain-dead persistent naming schemes on udev-
957 * based Linux systems those actually are quite common.
958 */
959 drv1 = find_hdev_driver(filename);
960 if (drv1) {
961 return drv1;
962 }
963
98289620 964 if (!path_has_protocol(filename) || !allow_protocol_prefix) {
ef810437 965 return &bdrv_file;
84a12e66 966 }
98289620 967
9e0b22f4
SH
968 p = strchr(filename, ':');
969 assert(p != NULL);
1cec71e3
AL
970 len = p - filename;
971 if (len > sizeof(protocol) - 1)
972 len = sizeof(protocol) - 1;
973 memcpy(protocol, filename, len);
974 protocol[len] = '\0';
88d88798
MM
975
976 drv1 = bdrv_do_find_protocol(protocol);
977 if (drv1) {
978 return drv1;
979 }
980
981 for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); ++i) {
982 if (block_driver_modules[i].protocol_name &&
983 !strcmp(block_driver_modules[i].protocol_name, protocol)) {
984 block_module_load_one(block_driver_modules[i].library_name);
985 break;
8a22f02a 986 }
83f64091 987 }
b65a5e12 988
88d88798
MM
989 drv1 = bdrv_do_find_protocol(protocol);
990 if (!drv1) {
991 error_setg(errp, "Unknown protocol '%s'", protocol);
992 }
993 return drv1;
83f64091
FB
994}
995
c6684249
MA
996/*
997 * Guess image format by probing its contents.
998 * This is not a good idea when your image is raw (CVE-2008-2004), but
999 * we do it anyway for backward compatibility.
1000 *
1001 * @buf contains the image's first @buf_size bytes.
7cddd372
KW
1002 * @buf_size is the buffer size in bytes (generally BLOCK_PROBE_BUF_SIZE,
1003 * but can be smaller if the image file is smaller)
c6684249
MA
1004 * @filename is its filename.
1005 *
1006 * For all block drivers, call the bdrv_probe() method to get its
1007 * probing score.
1008 * Return the first block driver with the highest probing score.
1009 */
38f3ef57
KW
1010BlockDriver *bdrv_probe_all(const uint8_t *buf, int buf_size,
1011 const char *filename)
c6684249
MA
1012{
1013 int score_max = 0, score;
1014 BlockDriver *drv = NULL, *d;
967d7905 1015 IO_CODE();
c6684249
MA
1016
1017 QLIST_FOREACH(d, &bdrv_drivers, list) {
1018 if (d->bdrv_probe) {
1019 score = d->bdrv_probe(buf, buf_size, filename);
1020 if (score > score_max) {
1021 score_max = score;
1022 drv = d;
1023 }
1024 }
1025 }
1026
1027 return drv;
1028}
1029
5696c6e3 1030static int find_image_format(BlockBackend *file, const char *filename,
34b5d2c6 1031 BlockDriver **pdrv, Error **errp)
f3a5d3f8 1032{
c6684249 1033 BlockDriver *drv;
7cddd372 1034 uint8_t buf[BLOCK_PROBE_BUF_SIZE];
f500a6d3 1035 int ret = 0;
f8ea0b00 1036
bdb73476
EGE
1037 GLOBAL_STATE_CODE();
1038
08a00559 1039 /* Return the raw BlockDriver * to scsi-generic devices or empty drives */
5696c6e3 1040 if (blk_is_sg(file) || !blk_is_inserted(file) || blk_getlength(file) == 0) {
ef810437 1041 *pdrv = &bdrv_raw;
c98ac35d 1042 return ret;
1a396859 1043 }
f8ea0b00 1044
a9262f55 1045 ret = blk_pread(file, 0, sizeof(buf), buf, 0);
83f64091 1046 if (ret < 0) {
34b5d2c6
HR
1047 error_setg_errno(errp, -ret, "Could not read image for determining its "
1048 "format");
c98ac35d
SW
1049 *pdrv = NULL;
1050 return ret;
83f64091
FB
1051 }
1052
bf5b16fa 1053 drv = bdrv_probe_all(buf, sizeof(buf), filename);
c98ac35d 1054 if (!drv) {
34b5d2c6
HR
1055 error_setg(errp, "Could not determine image format: No compatible "
1056 "driver found");
bf5b16fa
AF
1057 *pdrv = NULL;
1058 return -ENOENT;
c98ac35d 1059 }
bf5b16fa 1060
c98ac35d 1061 *pdrv = drv;
bf5b16fa 1062 return 0;
ea2384d3
FB
1063}
1064
51762288
SH
1065/**
1066 * Set the current 'total_sectors' value
65a9bb25 1067 * Return 0 on success, -errno on error.
51762288 1068 */
3d9f2d2a 1069int refresh_total_sectors(BlockDriverState *bs, int64_t hint)
51762288
SH
1070{
1071 BlockDriver *drv = bs->drv;
967d7905 1072 IO_CODE();
51762288 1073
d470ad42
HR
1074 if (!drv) {
1075 return -ENOMEDIUM;
1076 }
1077
396759ad 1078 /* Do not attempt drv->bdrv_getlength() on scsi-generic devices */
b192af8a 1079 if (bdrv_is_sg(bs))
396759ad
NB
1080 return 0;
1081
51762288
SH
1082 /* query actual device if possible, otherwise just trust the hint */
1083 if (drv->bdrv_getlength) {
1084 int64_t length = drv->bdrv_getlength(bs);
1085 if (length < 0) {
1086 return length;
1087 }
7e382003 1088 hint = DIV_ROUND_UP(length, BDRV_SECTOR_SIZE);
51762288
SH
1089 }
1090
1091 bs->total_sectors = hint;
8b117001
VSO
1092
1093 if (bs->total_sectors * BDRV_SECTOR_SIZE > BDRV_MAX_LENGTH) {
1094 return -EFBIG;
1095 }
1096
51762288
SH
1097 return 0;
1098}
1099
cddff5ba
KW
1100/**
1101 * Combines a QDict of new block driver @options with any missing options taken
1102 * from @old_options, so that leaving out an option defaults to its old value.
1103 */
1104static void bdrv_join_options(BlockDriverState *bs, QDict *options,
1105 QDict *old_options)
1106{
da359909 1107 GLOBAL_STATE_CODE();
cddff5ba
KW
1108 if (bs->drv && bs->drv->bdrv_join_options) {
1109 bs->drv->bdrv_join_options(options, old_options);
1110 } else {
1111 qdict_join(options, old_options, false);
1112 }
1113}
1114
543770bd
AG
1115static BlockdevDetectZeroesOptions bdrv_parse_detect_zeroes(QemuOpts *opts,
1116 int open_flags,
1117 Error **errp)
1118{
1119 Error *local_err = NULL;
1120 char *value = qemu_opt_get_del(opts, "detect-zeroes");
1121 BlockdevDetectZeroesOptions detect_zeroes =
1122 qapi_enum_parse(&BlockdevDetectZeroesOptions_lookup, value,
1123 BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF, &local_err);
bdb73476 1124 GLOBAL_STATE_CODE();
543770bd
AG
1125 g_free(value);
1126 if (local_err) {
1127 error_propagate(errp, local_err);
1128 return detect_zeroes;
1129 }
1130
1131 if (detect_zeroes == BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP &&
1132 !(open_flags & BDRV_O_UNMAP))
1133 {
1134 error_setg(errp, "setting detect-zeroes to unmap is not allowed "
1135 "without setting discard operation to unmap");
1136 }
1137
1138 return detect_zeroes;
1139}
1140
f80f2673
AM
1141/**
1142 * Set open flags for aio engine
1143 *
1144 * Return 0 on success, -1 if the engine specified is invalid
1145 */
1146int bdrv_parse_aio(const char *mode, int *flags)
1147{
1148 if (!strcmp(mode, "threads")) {
1149 /* do nothing, default */
1150 } else if (!strcmp(mode, "native")) {
1151 *flags |= BDRV_O_NATIVE_AIO;
1152#ifdef CONFIG_LINUX_IO_URING
1153 } else if (!strcmp(mode, "io_uring")) {
1154 *flags |= BDRV_O_IO_URING;
1155#endif
1156 } else {
1157 return -1;
1158 }
1159
1160 return 0;
1161}
1162
9e8f1835
PB
1163/**
1164 * Set open flags for a given discard mode
1165 *
1166 * Return 0 on success, -1 if the discard mode was invalid.
1167 */
1168int bdrv_parse_discard_flags(const char *mode, int *flags)
1169{
1170 *flags &= ~BDRV_O_UNMAP;
1171
1172 if (!strcmp(mode, "off") || !strcmp(mode, "ignore")) {
1173 /* do nothing */
1174 } else if (!strcmp(mode, "on") || !strcmp(mode, "unmap")) {
1175 *flags |= BDRV_O_UNMAP;
1176 } else {
1177 return -1;
1178 }
1179
1180 return 0;
1181}
1182
c3993cdc
SH
1183/**
1184 * Set open flags for a given cache mode
1185 *
1186 * Return 0 on success, -1 if the cache mode was invalid.
1187 */
53e8ae01 1188int bdrv_parse_cache_mode(const char *mode, int *flags, bool *writethrough)
c3993cdc
SH
1189{
1190 *flags &= ~BDRV_O_CACHE_MASK;
1191
1192 if (!strcmp(mode, "off") || !strcmp(mode, "none")) {
53e8ae01
KW
1193 *writethrough = false;
1194 *flags |= BDRV_O_NOCACHE;
92196b2f 1195 } else if (!strcmp(mode, "directsync")) {
53e8ae01 1196 *writethrough = true;
92196b2f 1197 *flags |= BDRV_O_NOCACHE;
c3993cdc 1198 } else if (!strcmp(mode, "writeback")) {
53e8ae01 1199 *writethrough = false;
c3993cdc 1200 } else if (!strcmp(mode, "unsafe")) {
53e8ae01 1201 *writethrough = false;
c3993cdc
SH
1202 *flags |= BDRV_O_NO_FLUSH;
1203 } else if (!strcmp(mode, "writethrough")) {
53e8ae01 1204 *writethrough = true;
c3993cdc
SH
1205 } else {
1206 return -1;
1207 }
1208
1209 return 0;
1210}
1211
b5411555
KW
1212static char *bdrv_child_get_parent_desc(BdrvChild *c)
1213{
1214 BlockDriverState *parent = c->opaque;
2c0a3acb 1215 return g_strdup_printf("node '%s'", bdrv_get_node_name(parent));
b5411555
KW
1216}
1217
20018e12
KW
1218static void bdrv_child_cb_drained_begin(BdrvChild *child)
1219{
1220 BlockDriverState *bs = child->opaque;
6cd5c9d7 1221 bdrv_do_drained_begin_quiesce(bs, NULL, false);
20018e12
KW
1222}
1223
89bd0305
KW
1224static bool bdrv_child_cb_drained_poll(BdrvChild *child)
1225{
1226 BlockDriverState *bs = child->opaque;
6cd5c9d7 1227 return bdrv_drain_poll(bs, false, NULL, false);
89bd0305
KW
1228}
1229
e037c09c
HR
1230static void bdrv_child_cb_drained_end(BdrvChild *child,
1231 int *drained_end_counter)
20018e12
KW
1232{
1233 BlockDriverState *bs = child->opaque;
e037c09c 1234 bdrv_drained_end_no_poll(bs, drained_end_counter);
20018e12
KW
1235}
1236
38701b6a
KW
1237static int bdrv_child_cb_inactivate(BdrvChild *child)
1238{
1239 BlockDriverState *bs = child->opaque;
bdb73476 1240 GLOBAL_STATE_CODE();
38701b6a
KW
1241 assert(bs->open_flags & BDRV_O_INACTIVE);
1242 return 0;
1243}
1244
5d231849
KW
1245static bool bdrv_child_cb_can_set_aio_ctx(BdrvChild *child, AioContext *ctx,
1246 GSList **ignore, Error **errp)
1247{
1248 BlockDriverState *bs = child->opaque;
1249 return bdrv_can_set_aio_context(bs, ctx, ignore, errp);
1250}
1251
53a7d041
KW
1252static void bdrv_child_cb_set_aio_ctx(BdrvChild *child, AioContext *ctx,
1253 GSList **ignore)
1254{
1255 BlockDriverState *bs = child->opaque;
1256 return bdrv_set_aio_context_ignore(bs, ctx, ignore);
1257}
1258
b1e6fc08 1259/*
73176bee
KW
1260 * Returns the options and flags that a temporary snapshot should get, based on
1261 * the originally requested flags (the originally requested image will have
1262 * flags like a backing file)
b1e6fc08 1263 */
73176bee
KW
1264static void bdrv_temp_snapshot_options(int *child_flags, QDict *child_options,
1265 int parent_flags, QDict *parent_options)
b1e6fc08 1266{
bdb73476 1267 GLOBAL_STATE_CODE();
73176bee
KW
1268 *child_flags = (parent_flags & ~BDRV_O_SNAPSHOT) | BDRV_O_TEMPORARY;
1269
1270 /* For temporary files, unconditional cache=unsafe is fine */
73176bee
KW
1271 qdict_set_default_str(child_options, BDRV_OPT_CACHE_DIRECT, "off");
1272 qdict_set_default_str(child_options, BDRV_OPT_CACHE_NO_FLUSH, "on");
41869044 1273
3f48686f 1274 /* Copy the read-only and discard options from the parent */
f87a0e29 1275 qdict_copy_default(child_options, parent_options, BDRV_OPT_READ_ONLY);
3f48686f 1276 qdict_copy_default(child_options, parent_options, BDRV_OPT_DISCARD);
f87a0e29 1277
41869044
KW
1278 /* aio=native doesn't work for cache.direct=off, so disable it for the
1279 * temporary snapshot */
1280 *child_flags &= ~BDRV_O_NATIVE_AIO;
b1e6fc08
KW
1281}
1282
db95dbba
KW
1283static void bdrv_backing_attach(BdrvChild *c)
1284{
1285 BlockDriverState *parent = c->opaque;
1286 BlockDriverState *backing_hd = c->bs;
1287
bdb73476 1288 GLOBAL_STATE_CODE();
db95dbba
KW
1289 assert(!parent->backing_blocker);
1290 error_setg(&parent->backing_blocker,
1291 "node is used as backing hd of '%s'",
1292 bdrv_get_device_or_node_name(parent));
1293
f30c66ba
HR
1294 bdrv_refresh_filename(backing_hd);
1295
db95dbba 1296 parent->open_flags &= ~BDRV_O_NO_BACKING;
db95dbba
KW
1297
1298 bdrv_op_block_all(backing_hd, parent->backing_blocker);
1299 /* Otherwise we won't be able to commit or stream */
1300 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_COMMIT_TARGET,
1301 parent->backing_blocker);
1302 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_STREAM,
1303 parent->backing_blocker);
1304 /*
1305 * We do backup in 3 ways:
1306 * 1. drive backup
1307 * The target bs is new opened, and the source is top BDS
1308 * 2. blockdev backup
1309 * Both the source and the target are top BDSes.
1310 * 3. internal backup(used for block replication)
1311 * Both the source and the target are backing file
1312 *
1313 * In case 1 and 2, neither the source nor the target is the backing file.
1314 * In case 3, we will block the top BDS, so there is only one block job
1315 * for the top BDS and its backing chain.
1316 */
1317 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_BACKUP_SOURCE,
1318 parent->backing_blocker);
1319 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_BACKUP_TARGET,
1320 parent->backing_blocker);
ca2f1234 1321}
d736f119 1322
db95dbba
KW
1323static void bdrv_backing_detach(BdrvChild *c)
1324{
1325 BlockDriverState *parent = c->opaque;
1326
bdb73476 1327 GLOBAL_STATE_CODE();
db95dbba
KW
1328 assert(parent->backing_blocker);
1329 bdrv_op_unblock_all(c->bs, parent->backing_blocker);
1330 error_free(parent->backing_blocker);
1331 parent->backing_blocker = NULL;
48e08288 1332}
d736f119 1333
6858eba0
KW
1334static int bdrv_backing_update_filename(BdrvChild *c, BlockDriverState *base,
1335 const char *filename, Error **errp)
1336{
1337 BlockDriverState *parent = c->opaque;
e94d3dba 1338 bool read_only = bdrv_is_read_only(parent);
6858eba0 1339 int ret;
bdb73476 1340 GLOBAL_STATE_CODE();
6858eba0 1341
e94d3dba
AG
1342 if (read_only) {
1343 ret = bdrv_reopen_set_read_only(parent, false, errp);
61f09cea
KW
1344 if (ret < 0) {
1345 return ret;
1346 }
1347 }
1348
6858eba0 1349 ret = bdrv_change_backing_file(parent, filename,
e54ee1b3
EB
1350 base->drv ? base->drv->format_name : "",
1351 false);
6858eba0 1352 if (ret < 0) {
64730694 1353 error_setg_errno(errp, -ret, "Could not update backing file link");
6858eba0
KW
1354 }
1355
e94d3dba
AG
1356 if (read_only) {
1357 bdrv_reopen_set_read_only(parent, true, NULL);
61f09cea
KW
1358 }
1359
6858eba0
KW
1360 return ret;
1361}
1362
fae8bd39
HR
1363/*
1364 * Returns the options and flags that a generic child of a BDS should
1365 * get, based on the given options and flags for the parent BDS.
1366 */
00ff7ffd
HR
1367static void bdrv_inherited_options(BdrvChildRole role, bool parent_is_format,
1368 int *child_flags, QDict *child_options,
1369 int parent_flags, QDict *parent_options)
fae8bd39
HR
1370{
1371 int flags = parent_flags;
bdb73476 1372 GLOBAL_STATE_CODE();
fae8bd39
HR
1373
1374 /*
1375 * First, decide whether to set, clear, or leave BDRV_O_PROTOCOL.
1376 * Generally, the question to answer is: Should this child be
1377 * format-probed by default?
1378 */
1379
1380 /*
1381 * Pure and non-filtered data children of non-format nodes should
1382 * be probed by default (even when the node itself has BDRV_O_PROTOCOL
1383 * set). This only affects a very limited set of drivers (namely
1384 * quorum and blkverify when this comment was written).
1385 * Force-clear BDRV_O_PROTOCOL then.
1386 */
1387 if (!parent_is_format &&
1388 (role & BDRV_CHILD_DATA) &&
1389 !(role & (BDRV_CHILD_METADATA | BDRV_CHILD_FILTERED)))
1390 {
1391 flags &= ~BDRV_O_PROTOCOL;
1392 }
1393
1394 /*
1395 * All children of format nodes (except for COW children) and all
1396 * metadata children in general should never be format-probed.
1397 * Force-set BDRV_O_PROTOCOL then.
1398 */
1399 if ((parent_is_format && !(role & BDRV_CHILD_COW)) ||
1400 (role & BDRV_CHILD_METADATA))
1401 {
1402 flags |= BDRV_O_PROTOCOL;
1403 }
1404
1405 /*
1406 * If the cache mode isn't explicitly set, inherit direct and no-flush from
1407 * the parent.
1408 */
1409 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_DIRECT);
1410 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_NO_FLUSH);
1411 qdict_copy_default(child_options, parent_options, BDRV_OPT_FORCE_SHARE);
1412
1413 if (role & BDRV_CHILD_COW) {
1414 /* backing files are opened read-only by default */
1415 qdict_set_default_str(child_options, BDRV_OPT_READ_ONLY, "on");
1416 qdict_set_default_str(child_options, BDRV_OPT_AUTO_READ_ONLY, "off");
1417 } else {
1418 /* Inherit the read-only option from the parent if it's not set */
1419 qdict_copy_default(child_options, parent_options, BDRV_OPT_READ_ONLY);
1420 qdict_copy_default(child_options, parent_options,
1421 BDRV_OPT_AUTO_READ_ONLY);
1422 }
1423
1424 /*
1425 * bdrv_co_pdiscard() respects unmap policy for the parent, so we
1426 * can default to enable it on lower layers regardless of the
1427 * parent option.
1428 */
1429 qdict_set_default_str(child_options, BDRV_OPT_DISCARD, "unmap");
1430
1431 /* Clear flags that only apply to the top layer */
1432 flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_COPY_ON_READ);
1433
1434 if (role & BDRV_CHILD_METADATA) {
1435 flags &= ~BDRV_O_NO_IO;
1436 }
1437 if (role & BDRV_CHILD_COW) {
1438 flags &= ~BDRV_O_TEMPORARY;
1439 }
1440
1441 *child_flags = flags;
1442}
1443
ca2f1234
HR
1444static void bdrv_child_cb_attach(BdrvChild *child)
1445{
1446 BlockDriverState *bs = child->opaque;
1447
696bf4c7 1448 assert_bdrv_graph_writable(bs);
a225369b 1449 QLIST_INSERT_HEAD(&bs->children, child, next);
5bb04747
VSO
1450 if (bs->drv->is_filter || (child->role & BDRV_CHILD_FILTERED)) {
1451 /*
1452 * Here we handle filters and block/raw-format.c when it behave like
1453 * filter. They generally have a single PRIMARY child, which is also the
1454 * FILTERED child, and that they may have multiple more children, which
1455 * are neither PRIMARY nor FILTERED. And never we have a COW child here.
1456 * So bs->file will be the PRIMARY child, unless the PRIMARY child goes
1457 * into bs->backing on exceptional cases; and bs->backing will be
1458 * nothing else.
1459 */
1460 assert(!(child->role & BDRV_CHILD_COW));
1461 if (child->role & BDRV_CHILD_PRIMARY) {
1462 assert(child->role & BDRV_CHILD_FILTERED);
1463 assert(!bs->backing);
1464 assert(!bs->file);
1465
1466 if (bs->drv->filtered_child_is_backing) {
1467 bs->backing = child;
1468 } else {
1469 bs->file = child;
1470 }
1471 } else {
1472 assert(!(child->role & BDRV_CHILD_FILTERED));
1473 }
1474 } else if (child->role & BDRV_CHILD_COW) {
1475 assert(bs->drv->supports_backing);
1476 assert(!(child->role & BDRV_CHILD_PRIMARY));
1477 assert(!bs->backing);
1478 bs->backing = child;
ca2f1234 1479 bdrv_backing_attach(child);
5bb04747
VSO
1480 } else if (child->role & BDRV_CHILD_PRIMARY) {
1481 assert(!bs->file);
1482 bs->file = child;
ca2f1234
HR
1483 }
1484
1485 bdrv_apply_subtree_drain(child, bs);
1486}
1487
48e08288
HR
1488static void bdrv_child_cb_detach(BdrvChild *child)
1489{
1490 BlockDriverState *bs = child->opaque;
1491
1492 if (child->role & BDRV_CHILD_COW) {
1493 bdrv_backing_detach(child);
1494 }
1495
1496 bdrv_unapply_subtree_drain(child, bs);
a225369b 1497
696bf4c7 1498 assert_bdrv_graph_writable(bs);
a225369b 1499 QLIST_REMOVE(child, next);
5bb04747
VSO
1500 if (child == bs->backing) {
1501 assert(child != bs->file);
1502 bs->backing = NULL;
1503 } else if (child == bs->file) {
1504 bs->file = NULL;
1505 }
48e08288
HR
1506}
1507
43483550
HR
1508static int bdrv_child_cb_update_filename(BdrvChild *c, BlockDriverState *base,
1509 const char *filename, Error **errp)
1510{
1511 if (c->role & BDRV_CHILD_COW) {
1512 return bdrv_backing_update_filename(c, base, filename, errp);
1513 }
1514 return 0;
1515}
1516
fb62b588 1517AioContext *child_of_bds_get_parent_aio_context(BdrvChild *c)
3ca1f322
VSO
1518{
1519 BlockDriverState *bs = c->opaque;
384a48fb 1520 IO_CODE();
3ca1f322
VSO
1521
1522 return bdrv_get_aio_context(bs);
1523}
1524
43483550
HR
1525const BdrvChildClass child_of_bds = {
1526 .parent_is_bds = true,
1527 .get_parent_desc = bdrv_child_get_parent_desc,
1528 .inherit_options = bdrv_inherited_options,
1529 .drained_begin = bdrv_child_cb_drained_begin,
1530 .drained_poll = bdrv_child_cb_drained_poll,
1531 .drained_end = bdrv_child_cb_drained_end,
1532 .attach = bdrv_child_cb_attach,
1533 .detach = bdrv_child_cb_detach,
1534 .inactivate = bdrv_child_cb_inactivate,
1535 .can_set_aio_ctx = bdrv_child_cb_can_set_aio_ctx,
1536 .set_aio_ctx = bdrv_child_cb_set_aio_ctx,
1537 .update_filename = bdrv_child_cb_update_filename,
fb62b588 1538 .get_parent_aio_context = child_of_bds_get_parent_aio_context,
43483550
HR
1539};
1540
3ca1f322
VSO
1541AioContext *bdrv_child_get_parent_aio_context(BdrvChild *c)
1542{
f0c28327 1543 GLOBAL_STATE_CODE();
3ca1f322
VSO
1544 return c->klass->get_parent_aio_context(c);
1545}
1546
7b272452
KW
1547static int bdrv_open_flags(BlockDriverState *bs, int flags)
1548{
61de4c68 1549 int open_flags = flags;
bdb73476 1550 GLOBAL_STATE_CODE();
7b272452
KW
1551
1552 /*
1553 * Clear flags that are internal to the block layer before opening the
1554 * image.
1555 */
20cca275 1556 open_flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_PROTOCOL);
7b272452 1557
7b272452
KW
1558 return open_flags;
1559}
1560
91a097e7
KW
1561static void update_flags_from_options(int *flags, QemuOpts *opts)
1562{
bdb73476
EGE
1563 GLOBAL_STATE_CODE();
1564
2a3d4331 1565 *flags &= ~(BDRV_O_CACHE_MASK | BDRV_O_RDWR | BDRV_O_AUTO_RDONLY);
91a097e7 1566
57f9db9a 1567 if (qemu_opt_get_bool_del(opts, BDRV_OPT_CACHE_NO_FLUSH, false)) {
91a097e7
KW
1568 *flags |= BDRV_O_NO_FLUSH;
1569 }
1570
57f9db9a 1571 if (qemu_opt_get_bool_del(opts, BDRV_OPT_CACHE_DIRECT, false)) {
91a097e7
KW
1572 *flags |= BDRV_O_NOCACHE;
1573 }
f87a0e29 1574
57f9db9a 1575 if (!qemu_opt_get_bool_del(opts, BDRV_OPT_READ_ONLY, false)) {
f87a0e29
AG
1576 *flags |= BDRV_O_RDWR;
1577 }
1578
e35bdc12
KW
1579 if (qemu_opt_get_bool_del(opts, BDRV_OPT_AUTO_READ_ONLY, false)) {
1580 *flags |= BDRV_O_AUTO_RDONLY;
1581 }
91a097e7
KW
1582}
1583
1584static void update_options_from_flags(QDict *options, int flags)
1585{
bdb73476 1586 GLOBAL_STATE_CODE();
91a097e7 1587 if (!qdict_haskey(options, BDRV_OPT_CACHE_DIRECT)) {
46f5ac20 1588 qdict_put_bool(options, BDRV_OPT_CACHE_DIRECT, flags & BDRV_O_NOCACHE);
91a097e7
KW
1589 }
1590 if (!qdict_haskey(options, BDRV_OPT_CACHE_NO_FLUSH)) {
46f5ac20
EB
1591 qdict_put_bool(options, BDRV_OPT_CACHE_NO_FLUSH,
1592 flags & BDRV_O_NO_FLUSH);
91a097e7 1593 }
f87a0e29 1594 if (!qdict_haskey(options, BDRV_OPT_READ_ONLY)) {
46f5ac20 1595 qdict_put_bool(options, BDRV_OPT_READ_ONLY, !(flags & BDRV_O_RDWR));
f87a0e29 1596 }
e35bdc12
KW
1597 if (!qdict_haskey(options, BDRV_OPT_AUTO_READ_ONLY)) {
1598 qdict_put_bool(options, BDRV_OPT_AUTO_READ_ONLY,
1599 flags & BDRV_O_AUTO_RDONLY);
1600 }
91a097e7
KW
1601}
1602
636ea370
KW
1603static void bdrv_assign_node_name(BlockDriverState *bs,
1604 const char *node_name,
1605 Error **errp)
6913c0c2 1606{
15489c76 1607 char *gen_node_name = NULL;
bdb73476 1608 GLOBAL_STATE_CODE();
6913c0c2 1609
15489c76
JC
1610 if (!node_name) {
1611 node_name = gen_node_name = id_generate(ID_BLOCK);
1612 } else if (!id_wellformed(node_name)) {
1613 /*
1614 * Check for empty string or invalid characters, but not if it is
1615 * generated (generated names use characters not available to the user)
1616 */
785ec4b1 1617 error_setg(errp, "Invalid node-name: '%s'", node_name);
636ea370 1618 return;
6913c0c2
BC
1619 }
1620
0c5e94ee 1621 /* takes care of avoiding namespaces collisions */
7f06d47e 1622 if (blk_by_name(node_name)) {
0c5e94ee
BC
1623 error_setg(errp, "node-name=%s is conflicting with a device id",
1624 node_name);
15489c76 1625 goto out;
0c5e94ee
BC
1626 }
1627
6913c0c2
BC
1628 /* takes care of avoiding duplicates node names */
1629 if (bdrv_find_node(node_name)) {
785ec4b1 1630 error_setg(errp, "Duplicate nodes with node-name='%s'", node_name);
15489c76 1631 goto out;
6913c0c2
BC
1632 }
1633
824808dd
KW
1634 /* Make sure that the node name isn't truncated */
1635 if (strlen(node_name) >= sizeof(bs->node_name)) {
1636 error_setg(errp, "Node name too long");
1637 goto out;
1638 }
1639
6913c0c2
BC
1640 /* copy node name into the bs and insert it into the graph list */
1641 pstrcpy(bs->node_name, sizeof(bs->node_name), node_name);
1642 QTAILQ_INSERT_TAIL(&graph_bdrv_states, bs, node_list);
15489c76
JC
1643out:
1644 g_free(gen_node_name);
6913c0c2
BC
1645}
1646
01a56501
KW
1647static int bdrv_open_driver(BlockDriverState *bs, BlockDriver *drv,
1648 const char *node_name, QDict *options,
1649 int open_flags, Error **errp)
1650{
1651 Error *local_err = NULL;
0f12264e 1652 int i, ret;
da359909 1653 GLOBAL_STATE_CODE();
01a56501
KW
1654
1655 bdrv_assign_node_name(bs, node_name, &local_err);
1656 if (local_err) {
1657 error_propagate(errp, local_err);
1658 return -EINVAL;
1659 }
1660
1661 bs->drv = drv;
1662 bs->opaque = g_malloc0(drv->instance_size);
1663
1664 if (drv->bdrv_file_open) {
1665 assert(!drv->bdrv_needs_filename || bs->filename[0]);
1666 ret = drv->bdrv_file_open(bs, options, open_flags, &local_err);
680c7f96 1667 } else if (drv->bdrv_open) {
01a56501 1668 ret = drv->bdrv_open(bs, options, open_flags, &local_err);
680c7f96
KW
1669 } else {
1670 ret = 0;
01a56501
KW
1671 }
1672
1673 if (ret < 0) {
1674 if (local_err) {
1675 error_propagate(errp, local_err);
1676 } else if (bs->filename[0]) {
1677 error_setg_errno(errp, -ret, "Could not open '%s'", bs->filename);
1678 } else {
1679 error_setg_errno(errp, -ret, "Could not open image");
1680 }
180ca19a 1681 goto open_failed;
01a56501
KW
1682 }
1683
1684 ret = refresh_total_sectors(bs, bs->total_sectors);
1685 if (ret < 0) {
1686 error_setg_errno(errp, -ret, "Could not refresh total sector count");
180ca19a 1687 return ret;
01a56501
KW
1688 }
1689
1e4c797c 1690 bdrv_refresh_limits(bs, NULL, &local_err);
01a56501
KW
1691 if (local_err) {
1692 error_propagate(errp, local_err);
180ca19a 1693 return -EINVAL;
01a56501
KW
1694 }
1695
1696 assert(bdrv_opt_mem_align(bs) != 0);
1697 assert(bdrv_min_mem_align(bs) != 0);
1698 assert(is_power_of_2(bs->bl.request_alignment));
1699
0f12264e
KW
1700 for (i = 0; i < bs->quiesce_counter; i++) {
1701 if (drv->bdrv_co_drain_begin) {
1702 drv->bdrv_co_drain_begin(bs);
1703 }
1704 }
1705
01a56501 1706 return 0;
180ca19a
MP
1707open_failed:
1708 bs->drv = NULL;
1709 if (bs->file != NULL) {
1710 bdrv_unref_child(bs, bs->file);
5bb04747 1711 assert(!bs->file);
180ca19a 1712 }
01a56501
KW
1713 g_free(bs->opaque);
1714 bs->opaque = NULL;
01a56501
KW
1715 return ret;
1716}
1717
621d1737
VSO
1718/*
1719 * Create and open a block node.
1720 *
1721 * @options is a QDict of options to pass to the block drivers, or NULL for an
1722 * empty set of options. The reference to the QDict belongs to the block layer
1723 * after the call (even on failure), so if the caller intends to reuse the
1724 * dictionary, it needs to use qobject_ref() before calling bdrv_open.
1725 */
1726BlockDriverState *bdrv_new_open_driver_opts(BlockDriver *drv,
1727 const char *node_name,
1728 QDict *options, int flags,
1729 Error **errp)
680c7f96
KW
1730{
1731 BlockDriverState *bs;
1732 int ret;
1733
f791bf7f
EGE
1734 GLOBAL_STATE_CODE();
1735
680c7f96
KW
1736 bs = bdrv_new();
1737 bs->open_flags = flags;
621d1737
VSO
1738 bs->options = options ?: qdict_new();
1739 bs->explicit_options = qdict_clone_shallow(bs->options);
680c7f96
KW
1740 bs->opaque = NULL;
1741
1742 update_options_from_flags(bs->options, flags);
1743
1744 ret = bdrv_open_driver(bs, drv, node_name, bs->options, flags, errp);
1745 if (ret < 0) {
cb3e7f08 1746 qobject_unref(bs->explicit_options);
180ca19a 1747 bs->explicit_options = NULL;
cb3e7f08 1748 qobject_unref(bs->options);
180ca19a 1749 bs->options = NULL;
680c7f96
KW
1750 bdrv_unref(bs);
1751 return NULL;
1752 }
1753
1754 return bs;
1755}
1756
621d1737
VSO
1757/* Create and open a block node. */
1758BlockDriverState *bdrv_new_open_driver(BlockDriver *drv, const char *node_name,
1759 int flags, Error **errp)
1760{
f791bf7f 1761 GLOBAL_STATE_CODE();
621d1737
VSO
1762 return bdrv_new_open_driver_opts(drv, node_name, NULL, flags, errp);
1763}
1764
c5f3014b 1765QemuOptsList bdrv_runtime_opts = {
18edf289
KW
1766 .name = "bdrv_common",
1767 .head = QTAILQ_HEAD_INITIALIZER(bdrv_runtime_opts.head),
1768 .desc = {
1769 {
1770 .name = "node-name",
1771 .type = QEMU_OPT_STRING,
1772 .help = "Node name of the block device node",
1773 },
62392ebb
KW
1774 {
1775 .name = "driver",
1776 .type = QEMU_OPT_STRING,
1777 .help = "Block driver to use for the node",
1778 },
91a097e7
KW
1779 {
1780 .name = BDRV_OPT_CACHE_DIRECT,
1781 .type = QEMU_OPT_BOOL,
1782 .help = "Bypass software writeback cache on the host",
1783 },
1784 {
1785 .name = BDRV_OPT_CACHE_NO_FLUSH,
1786 .type = QEMU_OPT_BOOL,
1787 .help = "Ignore flush requests",
1788 },
f87a0e29
AG
1789 {
1790 .name = BDRV_OPT_READ_ONLY,
1791 .type = QEMU_OPT_BOOL,
1792 .help = "Node is opened in read-only mode",
1793 },
e35bdc12
KW
1794 {
1795 .name = BDRV_OPT_AUTO_READ_ONLY,
1796 .type = QEMU_OPT_BOOL,
1797 .help = "Node can become read-only if opening read-write fails",
1798 },
692e01a2
KW
1799 {
1800 .name = "detect-zeroes",
1801 .type = QEMU_OPT_STRING,
1802 .help = "try to optimize zero writes (off, on, unmap)",
1803 },
818584a4 1804 {
415bbca8 1805 .name = BDRV_OPT_DISCARD,
818584a4
KW
1806 .type = QEMU_OPT_STRING,
1807 .help = "discard operation (ignore/off, unmap/on)",
1808 },
5a9347c6
FZ
1809 {
1810 .name = BDRV_OPT_FORCE_SHARE,
1811 .type = QEMU_OPT_BOOL,
1812 .help = "always accept other writers (default: off)",
1813 },
18edf289
KW
1814 { /* end of list */ }
1815 },
1816};
1817
5a5e7f8c
ML
1818QemuOptsList bdrv_create_opts_simple = {
1819 .name = "simple-create-opts",
1820 .head = QTAILQ_HEAD_INITIALIZER(bdrv_create_opts_simple.head),
fd17146c
HR
1821 .desc = {
1822 {
1823 .name = BLOCK_OPT_SIZE,
1824 .type = QEMU_OPT_SIZE,
1825 .help = "Virtual disk size"
1826 },
1827 {
1828 .name = BLOCK_OPT_PREALLOC,
1829 .type = QEMU_OPT_STRING,
1830 .help = "Preallocation mode (allowed values: off)"
1831 },
1832 { /* end of list */ }
1833 }
1834};
1835
57915332
KW
1836/*
1837 * Common part for opening disk images and files
b6ad491a
KW
1838 *
1839 * Removes all processed options from *options.
57915332 1840 */
5696c6e3 1841static int bdrv_open_common(BlockDriverState *bs, BlockBackend *file,
82dc8b41 1842 QDict *options, Error **errp)
57915332
KW
1843{
1844 int ret, open_flags;
035fccdf 1845 const char *filename;
62392ebb 1846 const char *driver_name = NULL;
6913c0c2 1847 const char *node_name = NULL;
818584a4 1848 const char *discard;
18edf289 1849 QemuOpts *opts;
62392ebb 1850 BlockDriver *drv;
34b5d2c6 1851 Error *local_err = NULL;
307261b2 1852 bool ro;
57915332 1853
6405875c 1854 assert(bs->file == NULL);
707ff828 1855 assert(options != NULL && bs->options != options);
bdb73476 1856 GLOBAL_STATE_CODE();
57915332 1857
62392ebb 1858 opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
af175e85 1859 if (!qemu_opts_absorb_qdict(opts, options, errp)) {
62392ebb
KW
1860 ret = -EINVAL;
1861 goto fail_opts;
1862 }
1863
9b7e8691
AG
1864 update_flags_from_options(&bs->open_flags, opts);
1865
62392ebb
KW
1866 driver_name = qemu_opt_get(opts, "driver");
1867 drv = bdrv_find_format(driver_name);
1868 assert(drv != NULL);
1869
5a9347c6
FZ
1870 bs->force_share = qemu_opt_get_bool(opts, BDRV_OPT_FORCE_SHARE, false);
1871
1872 if (bs->force_share && (bs->open_flags & BDRV_O_RDWR)) {
1873 error_setg(errp,
1874 BDRV_OPT_FORCE_SHARE
1875 "=on can only be used with read-only images");
1876 ret = -EINVAL;
1877 goto fail_opts;
1878 }
1879
45673671 1880 if (file != NULL) {
f30c66ba 1881 bdrv_refresh_filename(blk_bs(file));
5696c6e3 1882 filename = blk_bs(file)->filename;
45673671 1883 } else {
129c7d1c
MA
1884 /*
1885 * Caution: while qdict_get_try_str() is fine, getting
1886 * non-string types would require more care. When @options
1887 * come from -blockdev or blockdev_add, its members are typed
1888 * according to the QAPI schema, but when they come from
1889 * -drive, they're all QString.
1890 */
45673671
KW
1891 filename = qdict_get_try_str(options, "filename");
1892 }
1893
4a008240 1894 if (drv->bdrv_needs_filename && (!filename || !filename[0])) {
765003db
KW
1895 error_setg(errp, "The '%s' block driver requires a file name",
1896 drv->format_name);
18edf289
KW
1897 ret = -EINVAL;
1898 goto fail_opts;
6913c0c2 1899 }
6913c0c2 1900
82dc8b41
KW
1901 trace_bdrv_open_common(bs, filename ?: "", bs->open_flags,
1902 drv->format_name);
62392ebb 1903
307261b2
VSO
1904 ro = bdrv_is_read_only(bs);
1905
1906 if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv, ro)) {
1907 if (!ro && bdrv_is_whitelisted(drv, true)) {
8be25de6
KW
1908 ret = bdrv_apply_auto_read_only(bs, NULL, NULL);
1909 } else {
1910 ret = -ENOTSUP;
1911 }
1912 if (ret < 0) {
1913 error_setg(errp,
307261b2 1914 !ro && bdrv_is_whitelisted(drv, true)
8be25de6
KW
1915 ? "Driver '%s' can only be used for read-only devices"
1916 : "Driver '%s' is not whitelisted",
1917 drv->format_name);
1918 goto fail_opts;
1919 }
b64ec4e4 1920 }
57915332 1921
d3faa13e 1922 /* bdrv_new() and bdrv_close() make it so */
d73415a3 1923 assert(qatomic_read(&bs->copy_on_read) == 0);
d3faa13e 1924
82dc8b41 1925 if (bs->open_flags & BDRV_O_COPY_ON_READ) {
307261b2 1926 if (!ro) {
0ebd24e0
KW
1927 bdrv_enable_copy_on_read(bs);
1928 } else {
1929 error_setg(errp, "Can't use copy-on-read on read-only device");
18edf289
KW
1930 ret = -EINVAL;
1931 goto fail_opts;
0ebd24e0 1932 }
53fec9d3
SH
1933 }
1934
415bbca8 1935 discard = qemu_opt_get(opts, BDRV_OPT_DISCARD);
818584a4
KW
1936 if (discard != NULL) {
1937 if (bdrv_parse_discard_flags(discard, &bs->open_flags) != 0) {
1938 error_setg(errp, "Invalid discard option");
1939 ret = -EINVAL;
1940 goto fail_opts;
1941 }
1942 }
1943
543770bd
AG
1944 bs->detect_zeroes =
1945 bdrv_parse_detect_zeroes(opts, bs->open_flags, &local_err);
1946 if (local_err) {
1947 error_propagate(errp, local_err);
1948 ret = -EINVAL;
1949 goto fail_opts;
692e01a2
KW
1950 }
1951
c2ad1b0c
KW
1952 if (filename != NULL) {
1953 pstrcpy(bs->filename, sizeof(bs->filename), filename);
1954 } else {
1955 bs->filename[0] = '\0';
1956 }
91af7014 1957 pstrcpy(bs->exact_filename, sizeof(bs->exact_filename), bs->filename);
57915332 1958
66f82cee 1959 /* Open the image, either directly or using a protocol */
82dc8b41 1960 open_flags = bdrv_open_flags(bs, bs->open_flags);
01a56501 1961 node_name = qemu_opt_get(opts, "node-name");
57915332 1962
01a56501
KW
1963 assert(!drv->bdrv_file_open || file == NULL);
1964 ret = bdrv_open_driver(bs, drv, node_name, options, open_flags, errp);
51762288 1965 if (ret < 0) {
01a56501 1966 goto fail_opts;
3baca891
KW
1967 }
1968
18edf289 1969 qemu_opts_del(opts);
57915332
KW
1970 return 0;
1971
18edf289
KW
1972fail_opts:
1973 qemu_opts_del(opts);
57915332
KW
1974 return ret;
1975}
1976
5e5c4f63
KW
1977static QDict *parse_json_filename(const char *filename, Error **errp)
1978{
1979 QObject *options_obj;
1980 QDict *options;
1981 int ret;
bdb73476 1982 GLOBAL_STATE_CODE();
5e5c4f63
KW
1983
1984 ret = strstart(filename, "json:", &filename);
1985 assert(ret);
1986
5577fff7 1987 options_obj = qobject_from_json(filename, errp);
5e5c4f63 1988 if (!options_obj) {
5577fff7 1989 error_prepend(errp, "Could not parse the JSON options: ");
5e5c4f63
KW
1990 return NULL;
1991 }
1992
7dc847eb 1993 options = qobject_to(QDict, options_obj);
ca6b6e1e 1994 if (!options) {
cb3e7f08 1995 qobject_unref(options_obj);
5e5c4f63
KW
1996 error_setg(errp, "Invalid JSON object given");
1997 return NULL;
1998 }
1999
5e5c4f63
KW
2000 qdict_flatten(options);
2001
2002 return options;
2003}
2004
de3b53f0
KW
2005static void parse_json_protocol(QDict *options, const char **pfilename,
2006 Error **errp)
2007{
2008 QDict *json_options;
2009 Error *local_err = NULL;
bdb73476 2010 GLOBAL_STATE_CODE();
de3b53f0
KW
2011
2012 /* Parse json: pseudo-protocol */
2013 if (!*pfilename || !g_str_has_prefix(*pfilename, "json:")) {
2014 return;
2015 }
2016
2017 json_options = parse_json_filename(*pfilename, &local_err);
2018 if (local_err) {
2019 error_propagate(errp, local_err);
2020 return;
2021 }
2022
2023 /* Options given in the filename have lower priority than options
2024 * specified directly */
2025 qdict_join(options, json_options, false);
cb3e7f08 2026 qobject_unref(json_options);
de3b53f0
KW
2027 *pfilename = NULL;
2028}
2029
b6ce07aa 2030/*
f54120ff
KW
2031 * Fills in default options for opening images and converts the legacy
2032 * filename/flags pair to option QDict entries.
53a29513
HR
2033 * The BDRV_O_PROTOCOL flag in *flags will be set or cleared accordingly if a
2034 * block driver has been specified explicitly.
b6ce07aa 2035 */
de3b53f0 2036static int bdrv_fill_options(QDict **options, const char *filename,
053e1578 2037 int *flags, Error **errp)
ea2384d3 2038{
c2ad1b0c 2039 const char *drvname;
53a29513 2040 bool protocol = *flags & BDRV_O_PROTOCOL;
e3fa4bfa 2041 bool parse_filename = false;
053e1578 2042 BlockDriver *drv = NULL;
34b5d2c6 2043 Error *local_err = NULL;
83f64091 2044
da359909
EGE
2045 GLOBAL_STATE_CODE();
2046
129c7d1c
MA
2047 /*
2048 * Caution: while qdict_get_try_str() is fine, getting non-string
2049 * types would require more care. When @options come from
2050 * -blockdev or blockdev_add, its members are typed according to
2051 * the QAPI schema, but when they come from -drive, they're all
2052 * QString.
2053 */
53a29513 2054 drvname = qdict_get_try_str(*options, "driver");
053e1578
HR
2055 if (drvname) {
2056 drv = bdrv_find_format(drvname);
2057 if (!drv) {
2058 error_setg(errp, "Unknown driver '%s'", drvname);
2059 return -ENOENT;
2060 }
2061 /* If the user has explicitly specified the driver, this choice should
2062 * override the BDRV_O_PROTOCOL flag */
2063 protocol = drv->bdrv_file_open;
53a29513
HR
2064 }
2065
2066 if (protocol) {
2067 *flags |= BDRV_O_PROTOCOL;
2068 } else {
2069 *flags &= ~BDRV_O_PROTOCOL;
2070 }
2071
91a097e7
KW
2072 /* Translate cache options from flags into options */
2073 update_options_from_flags(*options, *flags);
2074
035fccdf 2075 /* Fetch the file name from the options QDict if necessary */
17b005f1 2076 if (protocol && filename) {
f54120ff 2077 if (!qdict_haskey(*options, "filename")) {
46f5ac20 2078 qdict_put_str(*options, "filename", filename);
f54120ff
KW
2079 parse_filename = true;
2080 } else {
2081 error_setg(errp, "Can't specify 'file' and 'filename' options at "
2082 "the same time");
2083 return -EINVAL;
2084 }
035fccdf
KW
2085 }
2086
c2ad1b0c 2087 /* Find the right block driver */
129c7d1c 2088 /* See cautionary note on accessing @options above */
f54120ff 2089 filename = qdict_get_try_str(*options, "filename");
f54120ff 2090
053e1578
HR
2091 if (!drvname && protocol) {
2092 if (filename) {
2093 drv = bdrv_find_protocol(filename, parse_filename, errp);
17b005f1 2094 if (!drv) {
053e1578 2095 return -EINVAL;
17b005f1 2096 }
053e1578
HR
2097
2098 drvname = drv->format_name;
46f5ac20 2099 qdict_put_str(*options, "driver", drvname);
053e1578
HR
2100 } else {
2101 error_setg(errp, "Must specify either driver or file");
2102 return -EINVAL;
98289620 2103 }
c2ad1b0c
KW
2104 }
2105
17b005f1 2106 assert(drv || !protocol);
c2ad1b0c 2107
f54120ff 2108 /* Driver-specific filename parsing */
17b005f1 2109 if (drv && drv->bdrv_parse_filename && parse_filename) {
5acd9d81 2110 drv->bdrv_parse_filename(filename, *options, &local_err);
84d18f06 2111 if (local_err) {
34b5d2c6 2112 error_propagate(errp, local_err);
f54120ff 2113 return -EINVAL;
6963a30d 2114 }
cd5d031e
HR
2115
2116 if (!drv->bdrv_needs_filename) {
2117 qdict_del(*options, "filename");
cd5d031e 2118 }
6963a30d
KW
2119 }
2120
f54120ff
KW
2121 return 0;
2122}
2123
148eb13c
KW
2124typedef struct BlockReopenQueueEntry {
2125 bool prepared;
69b736e7 2126 bool perms_checked;
148eb13c 2127 BDRVReopenState state;
859443b0 2128 QTAILQ_ENTRY(BlockReopenQueueEntry) entry;
148eb13c
KW
2129} BlockReopenQueueEntry;
2130
2131/*
2132 * Return the flags that @bs will have after the reopens in @q have
2133 * successfully completed. If @q is NULL (or @bs is not contained in @q),
2134 * return the current flags.
2135 */
2136static int bdrv_reopen_get_flags(BlockReopenQueue *q, BlockDriverState *bs)
2137{
2138 BlockReopenQueueEntry *entry;
2139
2140 if (q != NULL) {
859443b0 2141 QTAILQ_FOREACH(entry, q, entry) {
148eb13c
KW
2142 if (entry->state.bs == bs) {
2143 return entry->state.flags;
2144 }
2145 }
2146 }
2147
2148 return bs->open_flags;
2149}
2150
2151/* Returns whether the image file can be written to after the reopen queue @q
2152 * has been successfully applied, or right now if @q is NULL. */
cc022140
HR
2153static bool bdrv_is_writable_after_reopen(BlockDriverState *bs,
2154 BlockReopenQueue *q)
148eb13c
KW
2155{
2156 int flags = bdrv_reopen_get_flags(q, bs);
2157
2158 return (flags & (BDRV_O_RDWR | BDRV_O_INACTIVE)) == BDRV_O_RDWR;
2159}
2160
cc022140
HR
2161/*
2162 * Return whether the BDS can be written to. This is not necessarily
2163 * the same as !bdrv_is_read_only(bs), as inactivated images may not
2164 * be written to but do not count as read-only images.
2165 */
2166bool bdrv_is_writable(BlockDriverState *bs)
2167{
384a48fb 2168 IO_CODE();
cc022140
HR
2169 return bdrv_is_writable_after_reopen(bs, NULL);
2170}
2171
3bf416ba
VSO
2172static char *bdrv_child_user_desc(BdrvChild *c)
2173{
f0c28327 2174 GLOBAL_STATE_CODE();
da261b69 2175 return c->klass->get_parent_desc(c);
3bf416ba
VSO
2176}
2177
30ebb9aa
VSO
2178/*
2179 * Check that @a allows everything that @b needs. @a and @b must reference same
2180 * child node.
2181 */
3bf416ba
VSO
2182static bool bdrv_a_allow_b(BdrvChild *a, BdrvChild *b, Error **errp)
2183{
30ebb9aa
VSO
2184 const char *child_bs_name;
2185 g_autofree char *a_user = NULL;
2186 g_autofree char *b_user = NULL;
2187 g_autofree char *perms = NULL;
2188
2189 assert(a->bs);
2190 assert(a->bs == b->bs);
862fded9 2191 GLOBAL_STATE_CODE();
3bf416ba
VSO
2192
2193 if ((b->perm & a->shared_perm) == b->perm) {
2194 return true;
2195 }
2196
30ebb9aa
VSO
2197 child_bs_name = bdrv_get_node_name(b->bs);
2198 a_user = bdrv_child_user_desc(a);
2199 b_user = bdrv_child_user_desc(b);
2200 perms = bdrv_perm_names(b->perm & ~a->shared_perm);
2201
2202 error_setg(errp, "Permission conflict on node '%s': permissions '%s' are "
2203 "both required by %s (uses node '%s' as '%s' child) and "
2204 "unshared by %s (uses node '%s' as '%s' child).",
2205 child_bs_name, perms,
2206 b_user, child_bs_name, b->name,
2207 a_user, child_bs_name, a->name);
3bf416ba
VSO
2208
2209 return false;
2210}
2211
9397c14f 2212static bool bdrv_parent_perms_conflict(BlockDriverState *bs, Error **errp)
3bf416ba
VSO
2213{
2214 BdrvChild *a, *b;
862fded9 2215 GLOBAL_STATE_CODE();
3bf416ba
VSO
2216
2217 /*
2218 * During the loop we'll look at each pair twice. That's correct because
2219 * bdrv_a_allow_b() is asymmetric and we should check each pair in both
2220 * directions.
2221 */
2222 QLIST_FOREACH(a, &bs->parents, next_parent) {
2223 QLIST_FOREACH(b, &bs->parents, next_parent) {
9397c14f 2224 if (a == b) {
3bf416ba
VSO
2225 continue;
2226 }
2227
2228 if (!bdrv_a_allow_b(a, b, errp)) {
2229 return true;
2230 }
2231 }
2232 }
2233
2234 return false;
2235}
2236
ffd1a5a2 2237static void bdrv_child_perm(BlockDriverState *bs, BlockDriverState *child_bs,
e5d8a406
HR
2238 BdrvChild *c, BdrvChildRole role,
2239 BlockReopenQueue *reopen_queue,
ffd1a5a2
FZ
2240 uint64_t parent_perm, uint64_t parent_shared,
2241 uint64_t *nperm, uint64_t *nshared)
2242{
0b3ca76e 2243 assert(bs->drv && bs->drv->bdrv_child_perm);
da359909 2244 GLOBAL_STATE_CODE();
e5d8a406 2245 bs->drv->bdrv_child_perm(bs, c, role, reopen_queue,
0b3ca76e
AG
2246 parent_perm, parent_shared,
2247 nperm, nshared);
e0995dc3 2248 /* TODO Take force_share from reopen_queue */
ffd1a5a2
FZ
2249 if (child_bs && child_bs->force_share) {
2250 *nshared = BLK_PERM_ALL;
2251 }
2252}
2253
bd57f8f7
VSO
2254/*
2255 * Adds the whole subtree of @bs (including @bs itself) to the @list (except for
2256 * nodes that are already in the @list, of course) so that final list is
2257 * topologically sorted. Return the result (GSList @list object is updated, so
2258 * don't use old reference after function call).
2259 *
2260 * On function start @list must be already topologically sorted and for any node
2261 * in the @list the whole subtree of the node must be in the @list as well. The
2262 * simplest way to satisfy this criteria: use only result of
2263 * bdrv_topological_dfs() or NULL as @list parameter.
2264 */
2265static GSList *bdrv_topological_dfs(GSList *list, GHashTable *found,
2266 BlockDriverState *bs)
2267{
2268 BdrvChild *child;
2269 g_autoptr(GHashTable) local_found = NULL;
2270
bdb73476
EGE
2271 GLOBAL_STATE_CODE();
2272
bd57f8f7
VSO
2273 if (!found) {
2274 assert(!list);
2275 found = local_found = g_hash_table_new(NULL, NULL);
2276 }
2277
2278 if (g_hash_table_contains(found, bs)) {
2279 return list;
2280 }
2281 g_hash_table_add(found, bs);
2282
2283 QLIST_FOREACH(child, &bs->children, next) {
2284 list = bdrv_topological_dfs(list, found, child->bs);
2285 }
2286
2287 return g_slist_prepend(list, bs);
2288}
2289
ecb776bd
VSO
2290typedef struct BdrvChildSetPermState {
2291 BdrvChild *child;
2292 uint64_t old_perm;
2293 uint64_t old_shared_perm;
2294} BdrvChildSetPermState;
b0defa83
VSO
2295
2296static void bdrv_child_set_perm_abort(void *opaque)
2297{
ecb776bd
VSO
2298 BdrvChildSetPermState *s = opaque;
2299
862fded9
EGE
2300 GLOBAL_STATE_CODE();
2301
ecb776bd
VSO
2302 s->child->perm = s->old_perm;
2303 s->child->shared_perm = s->old_shared_perm;
b0defa83
VSO
2304}
2305
2306static TransactionActionDrv bdrv_child_set_pem_drv = {
2307 .abort = bdrv_child_set_perm_abort,
ecb776bd 2308 .clean = g_free,
b0defa83
VSO
2309};
2310
ecb776bd
VSO
2311static void bdrv_child_set_perm(BdrvChild *c, uint64_t perm,
2312 uint64_t shared, Transaction *tran)
b0defa83 2313{
ecb776bd 2314 BdrvChildSetPermState *s = g_new(BdrvChildSetPermState, 1);
862fded9 2315 GLOBAL_STATE_CODE();
ecb776bd
VSO
2316
2317 *s = (BdrvChildSetPermState) {
2318 .child = c,
2319 .old_perm = c->perm,
2320 .old_shared_perm = c->shared_perm,
2321 };
b0defa83
VSO
2322
2323 c->perm = perm;
2324 c->shared_perm = shared;
2325
ecb776bd 2326 tran_add(tran, &bdrv_child_set_pem_drv, s);
b0defa83
VSO
2327}
2328
2513ef59
VSO
2329static void bdrv_drv_set_perm_commit(void *opaque)
2330{
2331 BlockDriverState *bs = opaque;
2332 uint64_t cumulative_perms, cumulative_shared_perms;
da359909 2333 GLOBAL_STATE_CODE();
2513ef59
VSO
2334
2335 if (bs->drv->bdrv_set_perm) {
2336 bdrv_get_cumulative_perm(bs, &cumulative_perms,
2337 &cumulative_shared_perms);
2338 bs->drv->bdrv_set_perm(bs, cumulative_perms, cumulative_shared_perms);
2339 }
2340}
2341
2342static void bdrv_drv_set_perm_abort(void *opaque)
2343{
2344 BlockDriverState *bs = opaque;
da359909 2345 GLOBAL_STATE_CODE();
2513ef59
VSO
2346
2347 if (bs->drv->bdrv_abort_perm_update) {
2348 bs->drv->bdrv_abort_perm_update(bs);
2349 }
2350}
2351
2352TransactionActionDrv bdrv_drv_set_perm_drv = {
2353 .abort = bdrv_drv_set_perm_abort,
2354 .commit = bdrv_drv_set_perm_commit,
2355};
2356
2357static int bdrv_drv_set_perm(BlockDriverState *bs, uint64_t perm,
2358 uint64_t shared_perm, Transaction *tran,
2359 Error **errp)
2360{
da359909 2361 GLOBAL_STATE_CODE();
2513ef59
VSO
2362 if (!bs->drv) {
2363 return 0;
2364 }
2365
2366 if (bs->drv->bdrv_check_perm) {
2367 int ret = bs->drv->bdrv_check_perm(bs, perm, shared_perm, errp);
2368 if (ret < 0) {
2369 return ret;
2370 }
2371 }
2372
2373 if (tran) {
2374 tran_add(tran, &bdrv_drv_set_perm_drv, bs);
2375 }
2376
2377 return 0;
2378}
2379
0978623e
VSO
2380typedef struct BdrvReplaceChildState {
2381 BdrvChild *child;
2382 BlockDriverState *old_bs;
2383} BdrvReplaceChildState;
2384
2385static void bdrv_replace_child_commit(void *opaque)
2386{
2387 BdrvReplaceChildState *s = opaque;
bdb73476 2388 GLOBAL_STATE_CODE();
0978623e
VSO
2389
2390 bdrv_unref(s->old_bs);
2391}
2392
2393static void bdrv_replace_child_abort(void *opaque)
2394{
2395 BdrvReplaceChildState *s = opaque;
2396 BlockDriverState *new_bs = s->child->bs;
2397
bdb73476 2398 GLOBAL_STATE_CODE();
0f0b1e29 2399 /* old_bs reference is transparently moved from @s to @s->child */
544acc7d 2400 bdrv_replace_child_noperm(s->child, s->old_bs);
0978623e
VSO
2401 bdrv_unref(new_bs);
2402}
2403
2404static TransactionActionDrv bdrv_replace_child_drv = {
2405 .commit = bdrv_replace_child_commit,
2406 .abort = bdrv_replace_child_abort,
2407 .clean = g_free,
2408};
2409
2410/*
4bf021db 2411 * bdrv_replace_child_tran
0978623e
VSO
2412 *
2413 * Note: real unref of old_bs is done only on commit.
4bf021db
VSO
2414 *
2415 * The function doesn't update permissions, caller is responsible for this.
0978623e 2416 */
0f0b1e29 2417static void bdrv_replace_child_tran(BdrvChild *child, BlockDriverState *new_bs,
4eba825a 2418 Transaction *tran)
0978623e
VSO
2419{
2420 BdrvReplaceChildState *s = g_new(BdrvReplaceChildState, 1);
2421 *s = (BdrvReplaceChildState) {
0f0b1e29
VSO
2422 .child = child,
2423 .old_bs = child->bs,
0978623e
VSO
2424 };
2425 tran_add(tran, &bdrv_replace_child_drv, s);
2426
2427 if (new_bs) {
2428 bdrv_ref(new_bs);
2429 }
544acc7d 2430 bdrv_replace_child_noperm(child, new_bs);
0f0b1e29 2431 /* old_bs reference is transparently moved from @child to @s */
0978623e
VSO
2432}
2433
33a610c3 2434/*
c20555e1
VSO
2435 * Refresh permissions in @bs subtree. The function is intended to be called
2436 * after some graph modification that was done without permission update.
33a610c3 2437 */
c20555e1
VSO
2438static int bdrv_node_refresh_perm(BlockDriverState *bs, BlockReopenQueue *q,
2439 Transaction *tran, Error **errp)
33a610c3
KW
2440{
2441 BlockDriver *drv = bs->drv;
2442 BdrvChild *c;
2443 int ret;
c20555e1 2444 uint64_t cumulative_perms, cumulative_shared_perms;
862fded9 2445 GLOBAL_STATE_CODE();
c20555e1
VSO
2446
2447 bdrv_get_cumulative_perm(bs, &cumulative_perms, &cumulative_shared_perms);
33a610c3
KW
2448
2449 /* Write permissions never work with read-only images */
2450 if ((cumulative_perms & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)) &&
cc022140 2451 !bdrv_is_writable_after_reopen(bs, q))
33a610c3 2452 {
481e0eee
HR
2453 if (!bdrv_is_writable_after_reopen(bs, NULL)) {
2454 error_setg(errp, "Block node is read-only");
2455 } else {
c20555e1
VSO
2456 error_setg(errp, "Read-only block node '%s' cannot support "
2457 "read-write users", bdrv_get_node_name(bs));
481e0eee
HR
2458 }
2459
33a610c3
KW
2460 return -EPERM;
2461 }
2462
9c60a5d1
KW
2463 /*
2464 * Unaligned requests will automatically be aligned to bl.request_alignment
2465 * and without RESIZE we can't extend requests to write to space beyond the
2466 * end of the image, so it's required that the image size is aligned.
2467 */
2468 if ((cumulative_perms & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)) &&
2469 !(cumulative_perms & BLK_PERM_RESIZE))
2470 {
2471 if ((bs->total_sectors * BDRV_SECTOR_SIZE) % bs->bl.request_alignment) {
2472 error_setg(errp, "Cannot get 'write' permission without 'resize': "
2473 "Image size is not a multiple of request "
2474 "alignment");
2475 return -EPERM;
2476 }
2477 }
2478
33a610c3
KW
2479 /* Check this node */
2480 if (!drv) {
2481 return 0;
2482 }
2483
b1d2bbeb 2484 ret = bdrv_drv_set_perm(bs, cumulative_perms, cumulative_shared_perms, tran,
2513ef59
VSO
2485 errp);
2486 if (ret < 0) {
2487 return ret;
33a610c3
KW
2488 }
2489
78e421c9 2490 /* Drivers that never have children can omit .bdrv_child_perm() */
33a610c3 2491 if (!drv->bdrv_child_perm) {
78e421c9 2492 assert(QLIST_EMPTY(&bs->children));
33a610c3
KW
2493 return 0;
2494 }
2495
2496 /* Check all children */
2497 QLIST_FOREACH(c, &bs->children, next) {
2498 uint64_t cur_perm, cur_shared;
9eab1544 2499
e5d8a406 2500 bdrv_child_perm(bs, c->bs, c, c->role, q,
ffd1a5a2
FZ
2501 cumulative_perms, cumulative_shared_perms,
2502 &cur_perm, &cur_shared);
ecb776bd 2503 bdrv_child_set_perm(c, cur_perm, cur_shared, tran);
bd57f8f7
VSO
2504 }
2505
2506 return 0;
2507}
3ef45e02 2508
25409807
VSO
2509static int bdrv_list_refresh_perms(GSList *list, BlockReopenQueue *q,
2510 Transaction *tran, Error **errp)
bd57f8f7
VSO
2511{
2512 int ret;
b1d2bbeb 2513 BlockDriverState *bs;
862fded9 2514 GLOBAL_STATE_CODE();
bd57f8f7 2515
b1d2bbeb
VSO
2516 for ( ; list; list = list->next) {
2517 bs = list->data;
2518
9397c14f 2519 if (bdrv_parent_perms_conflict(bs, errp)) {
b1d2bbeb 2520 return -EINVAL;
bd57f8f7
VSO
2521 }
2522
c20555e1 2523 ret = bdrv_node_refresh_perm(bs, q, tran, errp);
33a610c3
KW
2524 if (ret < 0) {
2525 return ret;
2526 }
2527 }
2528
2529 return 0;
2530}
2531
c7a0f2be
KW
2532void bdrv_get_cumulative_perm(BlockDriverState *bs, uint64_t *perm,
2533 uint64_t *shared_perm)
33a610c3
KW
2534{
2535 BdrvChild *c;
2536 uint64_t cumulative_perms = 0;
2537 uint64_t cumulative_shared_perms = BLK_PERM_ALL;
2538
b4ad82aa
EGE
2539 GLOBAL_STATE_CODE();
2540
33a610c3
KW
2541 QLIST_FOREACH(c, &bs->parents, next_parent) {
2542 cumulative_perms |= c->perm;
2543 cumulative_shared_perms &= c->shared_perm;
2544 }
2545
2546 *perm = cumulative_perms;
2547 *shared_perm = cumulative_shared_perms;
2548}
2549
5176196c 2550char *bdrv_perm_names(uint64_t perm)
d083319f
KW
2551{
2552 struct perm_name {
2553 uint64_t perm;
2554 const char *name;
2555 } permissions[] = {
2556 { BLK_PERM_CONSISTENT_READ, "consistent read" },
2557 { BLK_PERM_WRITE, "write" },
2558 { BLK_PERM_WRITE_UNCHANGED, "write unchanged" },
2559 { BLK_PERM_RESIZE, "resize" },
d083319f
KW
2560 { 0, NULL }
2561 };
2562
e2a7423a 2563 GString *result = g_string_sized_new(30);
d083319f
KW
2564 struct perm_name *p;
2565
2566 for (p = permissions; p->name; p++) {
2567 if (perm & p->perm) {
e2a7423a
AG
2568 if (result->len > 0) {
2569 g_string_append(result, ", ");
2570 }
2571 g_string_append(result, p->name);
d083319f
KW
2572 }
2573 }
2574
e2a7423a 2575 return g_string_free(result, FALSE);
d083319f
KW
2576}
2577
33a610c3 2578
071b474f 2579static int bdrv_refresh_perms(BlockDriverState *bs, Error **errp)
bb87e4d1
VSO
2580{
2581 int ret;
b1d2bbeb
VSO
2582 Transaction *tran = tran_new();
2583 g_autoptr(GSList) list = bdrv_topological_dfs(NULL, NULL, bs);
862fded9 2584 GLOBAL_STATE_CODE();
bb87e4d1 2585
b1d2bbeb
VSO
2586 ret = bdrv_list_refresh_perms(list, NULL, tran, errp);
2587 tran_finalize(tran, ret);
bb87e4d1 2588
b1d2bbeb 2589 return ret;
bb87e4d1
VSO
2590}
2591
33a610c3
KW
2592int bdrv_child_try_set_perm(BdrvChild *c, uint64_t perm, uint64_t shared,
2593 Error **errp)
2594{
1046779e 2595 Error *local_err = NULL;
83928dc4 2596 Transaction *tran = tran_new();
33a610c3
KW
2597 int ret;
2598
b4ad82aa
EGE
2599 GLOBAL_STATE_CODE();
2600
ecb776bd 2601 bdrv_child_set_perm(c, perm, shared, tran);
83928dc4
VSO
2602
2603 ret = bdrv_refresh_perms(c->bs, &local_err);
2604
2605 tran_finalize(tran, ret);
2606
33a610c3 2607 if (ret < 0) {
071b474f
VSO
2608 if ((perm & ~c->perm) || (c->shared_perm & ~shared)) {
2609 /* tighten permissions */
1046779e
HR
2610 error_propagate(errp, local_err);
2611 } else {
2612 /*
2613 * Our caller may intend to only loosen restrictions and
2614 * does not expect this function to fail. Errors are not
2615 * fatal in such a case, so we can just hide them from our
2616 * caller.
2617 */
2618 error_free(local_err);
2619 ret = 0;
2620 }
33a610c3
KW
2621 }
2622
83928dc4 2623 return ret;
d5e6f437
KW
2624}
2625
c1087f12
HR
2626int bdrv_child_refresh_perms(BlockDriverState *bs, BdrvChild *c, Error **errp)
2627{
2628 uint64_t parent_perms, parent_shared;
2629 uint64_t perms, shared;
2630
b4ad82aa
EGE
2631 GLOBAL_STATE_CODE();
2632
c1087f12 2633 bdrv_get_cumulative_perm(bs, &parent_perms, &parent_shared);
e5d8a406 2634 bdrv_child_perm(bs, c->bs, c, c->role, NULL,
bf8e925e 2635 parent_perms, parent_shared, &perms, &shared);
c1087f12
HR
2636
2637 return bdrv_child_try_set_perm(c, perms, shared, errp);
2638}
2639
87278af1
HR
2640/*
2641 * Default implementation for .bdrv_child_perm() for block filters:
2642 * Forward CONSISTENT_READ, WRITE, WRITE_UNCHANGED, and RESIZE to the
2643 * filtered child.
2644 */
2645static void bdrv_filter_default_perms(BlockDriverState *bs, BdrvChild *c,
87278af1
HR
2646 BdrvChildRole role,
2647 BlockReopenQueue *reopen_queue,
2648 uint64_t perm, uint64_t shared,
2649 uint64_t *nperm, uint64_t *nshared)
6a1b9ee1 2650{
862fded9 2651 GLOBAL_STATE_CODE();
e444fa83
KW
2652 *nperm = perm & DEFAULT_PERM_PASSTHROUGH;
2653 *nshared = (shared & DEFAULT_PERM_PASSTHROUGH) | DEFAULT_PERM_UNCHANGED;
6a1b9ee1
KW
2654}
2655
70082db4 2656static void bdrv_default_perms_for_cow(BlockDriverState *bs, BdrvChild *c,
70082db4
HR
2657 BdrvChildRole role,
2658 BlockReopenQueue *reopen_queue,
2659 uint64_t perm, uint64_t shared,
2660 uint64_t *nperm, uint64_t *nshared)
2661{
e5d8a406 2662 assert(role & BDRV_CHILD_COW);
862fded9 2663 GLOBAL_STATE_CODE();
70082db4
HR
2664
2665 /*
2666 * We want consistent read from backing files if the parent needs it.
2667 * No other operations are performed on backing files.
2668 */
2669 perm &= BLK_PERM_CONSISTENT_READ;
2670
2671 /*
2672 * If the parent can deal with changing data, we're okay with a
2673 * writable and resizable backing file.
2674 * TODO Require !(perm & BLK_PERM_CONSISTENT_READ), too?
2675 */
2676 if (shared & BLK_PERM_WRITE) {
2677 shared = BLK_PERM_WRITE | BLK_PERM_RESIZE;
2678 } else {
2679 shared = 0;
2680 }
2681
64631f36 2682 shared |= BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE_UNCHANGED;
70082db4
HR
2683
2684 if (bs->open_flags & BDRV_O_INACTIVE) {
2685 shared |= BLK_PERM_WRITE | BLK_PERM_RESIZE;
2686 }
2687
2688 *nperm = perm;
2689 *nshared = shared;
2690}
2691
6f838a4b 2692static void bdrv_default_perms_for_storage(BlockDriverState *bs, BdrvChild *c,
6f838a4b
HR
2693 BdrvChildRole role,
2694 BlockReopenQueue *reopen_queue,
2695 uint64_t perm, uint64_t shared,
2696 uint64_t *nperm, uint64_t *nshared)
2697{
2698 int flags;
2699
862fded9 2700 GLOBAL_STATE_CODE();
e5d8a406 2701 assert(role & (BDRV_CHILD_METADATA | BDRV_CHILD_DATA));
6f838a4b
HR
2702
2703 flags = bdrv_reopen_get_flags(reopen_queue, bs);
2704
2705 /*
2706 * Apart from the modifications below, the same permissions are
2707 * forwarded and left alone as for filters
2708 */
e5d8a406 2709 bdrv_filter_default_perms(bs, c, role, reopen_queue,
6f838a4b
HR
2710 perm, shared, &perm, &shared);
2711
f889054f
HR
2712 if (role & BDRV_CHILD_METADATA) {
2713 /* Format drivers may touch metadata even if the guest doesn't write */
2714 if (bdrv_is_writable_after_reopen(bs, reopen_queue)) {
2715 perm |= BLK_PERM_WRITE | BLK_PERM_RESIZE;
2716 }
2717
2718 /*
2719 * bs->file always needs to be consistent because of the
2720 * metadata. We can never allow other users to resize or write
2721 * to it.
2722 */
2723 if (!(flags & BDRV_O_NO_IO)) {
2724 perm |= BLK_PERM_CONSISTENT_READ;
2725 }
2726 shared &= ~(BLK_PERM_WRITE | BLK_PERM_RESIZE);
6f838a4b
HR
2727 }
2728
f889054f
HR
2729 if (role & BDRV_CHILD_DATA) {
2730 /*
2731 * Technically, everything in this block is a subset of the
2732 * BDRV_CHILD_METADATA path taken above, and so this could
2733 * be an "else if" branch. However, that is not obvious, and
2734 * this function is not performance critical, therefore we let
2735 * this be an independent "if".
2736 */
2737
2738 /*
2739 * We cannot allow other users to resize the file because the
2740 * format driver might have some assumptions about the size
2741 * (e.g. because it is stored in metadata, or because the file
2742 * is split into fixed-size data files).
2743 */
2744 shared &= ~BLK_PERM_RESIZE;
2745
2746 /*
2747 * WRITE_UNCHANGED often cannot be performed as such on the
2748 * data file. For example, the qcow2 driver may still need to
2749 * write copied clusters on copy-on-read.
2750 */
2751 if (perm & BLK_PERM_WRITE_UNCHANGED) {
2752 perm |= BLK_PERM_WRITE;
2753 }
2754
2755 /*
2756 * If the data file is written to, the format driver may
2757 * expect to be able to resize it by writing beyond the EOF.
2758 */
2759 if (perm & BLK_PERM_WRITE) {
2760 perm |= BLK_PERM_RESIZE;
2761 }
6f838a4b 2762 }
6f838a4b
HR
2763
2764 if (bs->open_flags & BDRV_O_INACTIVE) {
2765 shared |= BLK_PERM_WRITE | BLK_PERM_RESIZE;
2766 }
2767
2768 *nperm = perm;
2769 *nshared = shared;
2770}
2771
2519f549 2772void bdrv_default_perms(BlockDriverState *bs, BdrvChild *c,
e5d8a406 2773 BdrvChildRole role, BlockReopenQueue *reopen_queue,
2519f549
HR
2774 uint64_t perm, uint64_t shared,
2775 uint64_t *nperm, uint64_t *nshared)
2776{
b4ad82aa 2777 GLOBAL_STATE_CODE();
2519f549
HR
2778 if (role & BDRV_CHILD_FILTERED) {
2779 assert(!(role & (BDRV_CHILD_DATA | BDRV_CHILD_METADATA |
2780 BDRV_CHILD_COW)));
e5d8a406 2781 bdrv_filter_default_perms(bs, c, role, reopen_queue,
2519f549
HR
2782 perm, shared, nperm, nshared);
2783 } else if (role & BDRV_CHILD_COW) {
2784 assert(!(role & (BDRV_CHILD_DATA | BDRV_CHILD_METADATA)));
e5d8a406 2785 bdrv_default_perms_for_cow(bs, c, role, reopen_queue,
2519f549
HR
2786 perm, shared, nperm, nshared);
2787 } else if (role & (BDRV_CHILD_METADATA | BDRV_CHILD_DATA)) {
e5d8a406 2788 bdrv_default_perms_for_storage(bs, c, role, reopen_queue,
2519f549
HR
2789 perm, shared, nperm, nshared);
2790 } else {
2791 g_assert_not_reached();
2792 }
2793}
2794
7b1d9c4d
HR
2795uint64_t bdrv_qapi_perm_to_blk_perm(BlockPermission qapi_perm)
2796{
2797 static const uint64_t permissions[] = {
2798 [BLOCK_PERMISSION_CONSISTENT_READ] = BLK_PERM_CONSISTENT_READ,
2799 [BLOCK_PERMISSION_WRITE] = BLK_PERM_WRITE,
2800 [BLOCK_PERMISSION_WRITE_UNCHANGED] = BLK_PERM_WRITE_UNCHANGED,
2801 [BLOCK_PERMISSION_RESIZE] = BLK_PERM_RESIZE,
7b1d9c4d
HR
2802 };
2803
2804 QEMU_BUILD_BUG_ON(ARRAY_SIZE(permissions) != BLOCK_PERMISSION__MAX);
2805 QEMU_BUILD_BUG_ON(1UL << ARRAY_SIZE(permissions) != BLK_PERM_ALL + 1);
2806
2807 assert(qapi_perm < BLOCK_PERMISSION__MAX);
2808
2809 return permissions[qapi_perm];
2810}
2811
544acc7d 2812static void bdrv_replace_child_noperm(BdrvChild *child,
4eba825a 2813 BlockDriverState *new_bs)
e9740bc6
KW
2814{
2815 BlockDriverState *old_bs = child->bs;
debc2927
HR
2816 int new_bs_quiesce_counter;
2817 int drain_saldo;
e9740bc6 2818
2cad1ebe 2819 assert(!child->frozen);
bfb8aa6d 2820 assert(old_bs != new_bs);
f0c28327 2821 GLOBAL_STATE_CODE();
2cad1ebe 2822
bb2614e9
FZ
2823 if (old_bs && new_bs) {
2824 assert(bdrv_get_aio_context(old_bs) == bdrv_get_aio_context(new_bs));
2825 }
debc2927
HR
2826
2827 new_bs_quiesce_counter = (new_bs ? new_bs->quiesce_counter : 0);
2828 drain_saldo = new_bs_quiesce_counter - child->parent_quiesce_counter;
2829
2830 /*
2831 * If the new child node is drained but the old one was not, flush
2832 * all outstanding requests to the old child node.
2833 */
bd86fb99 2834 while (drain_saldo > 0 && child->klass->drained_begin) {
debc2927
HR
2835 bdrv_parent_drained_begin_single(child, true);
2836 drain_saldo--;
2837 }
2838
e9740bc6 2839 if (old_bs) {
d736f119
KW
2840 /* Detach first so that the recursive drain sections coming from @child
2841 * are already gone and we only end the drain sections that came from
2842 * elsewhere. */
bd86fb99
HR
2843 if (child->klass->detach) {
2844 child->klass->detach(child);
d736f119 2845 }
696bf4c7 2846 assert_bdrv_graph_writable(old_bs);
e9740bc6
KW
2847 QLIST_REMOVE(child, next_parent);
2848 }
36fe1331
KW
2849
2850 child->bs = new_bs;
2851
e9740bc6 2852 if (new_bs) {
696bf4c7 2853 assert_bdrv_graph_writable(new_bs);
e9740bc6 2854 QLIST_INSERT_HEAD(&new_bs->parents, child, next_parent);
debc2927
HR
2855
2856 /*
2857 * Detaching the old node may have led to the new node's
2858 * quiesce_counter having been decreased. Not a problem, we
2859 * just need to recognize this here and then invoke
2860 * drained_end appropriately more often.
2861 */
2862 assert(new_bs->quiesce_counter <= new_bs_quiesce_counter);
2863 drain_saldo += new_bs->quiesce_counter - new_bs_quiesce_counter;
33a610c3 2864
d736f119
KW
2865 /* Attach only after starting new drained sections, so that recursive
2866 * drain sections coming from @child don't get an extra .drained_begin
2867 * callback. */
bd86fb99
HR
2868 if (child->klass->attach) {
2869 child->klass->attach(child);
8ee03995
KW
2870 }
2871 }
debc2927
HR
2872
2873 /*
2874 * If the old child node was drained but the new one is not, allow
2875 * requests to come in only after the new node has been attached.
2876 */
bd86fb99 2877 while (drain_saldo < 0 && child->klass->drained_end) {
debc2927
HR
2878 bdrv_parent_drained_end_single(child);
2879 drain_saldo++;
2880 }
8ee03995 2881}
33a610c3 2882
04c9c3a5
HR
2883/**
2884 * Free the given @child.
2885 *
2886 * The child must be empty (i.e. `child->bs == NULL`) and it must be
2887 * unused (i.e. not in a children list).
2888 */
2889static void bdrv_child_free(BdrvChild *child)
df581792 2890{
548a74c0 2891 assert(!child->bs);
bdb73476 2892 GLOBAL_STATE_CODE();
a225369b 2893 assert(!child->next.le_prev); /* not in children list */
04c9c3a5
HR
2894
2895 g_free(child->name);
2896 g_free(child);
548a74c0 2897}
d5e6f437 2898
548a74c0 2899typedef struct BdrvAttachChildCommonState {
5bb04747 2900 BdrvChild *child;
548a74c0
VSO
2901 AioContext *old_parent_ctx;
2902 AioContext *old_child_ctx;
2903} BdrvAttachChildCommonState;
2904
2905static void bdrv_attach_child_common_abort(void *opaque)
2906{
2907 BdrvAttachChildCommonState *s = opaque;
5bb04747 2908 BlockDriverState *bs = s->child->bs;
548a74c0 2909
f0c28327 2910 GLOBAL_STATE_CODE();
5bb04747 2911 bdrv_replace_child_noperm(s->child, NULL);
548a74c0
VSO
2912
2913 if (bdrv_get_aio_context(bs) != s->old_child_ctx) {
2914 bdrv_try_set_aio_context(bs, s->old_child_ctx, &error_abort);
2915 }
2916
5bb04747 2917 if (bdrv_child_get_parent_aio_context(s->child) != s->old_parent_ctx) {
26518061 2918 GSList *ignore;
548a74c0 2919
26518061
HR
2920 /* No need to ignore `child`, because it has been detached already */
2921 ignore = NULL;
5bb04747
VSO
2922 s->child->klass->can_set_aio_ctx(s->child, s->old_parent_ctx, &ignore,
2923 &error_abort);
548a74c0 2924 g_slist_free(ignore);
548a74c0 2925
26518061 2926 ignore = NULL;
5bb04747 2927 s->child->klass->set_aio_ctx(s->child, s->old_parent_ctx, &ignore);
548a74c0 2928 g_slist_free(ignore);
d5e6f437
KW
2929 }
2930
548a74c0 2931 bdrv_unref(bs);
5bb04747 2932 bdrv_child_free(s->child);
548a74c0
VSO
2933}
2934
2935static TransactionActionDrv bdrv_attach_child_common_drv = {
2936 .abort = bdrv_attach_child_common_abort,
2937 .clean = g_free,
2938};
2939
2940/*
2941 * Common part of attaching bdrv child to bs or to blk or to job
f8d2ad78 2942 *
7ec390d5 2943 * Function doesn't update permissions, caller is responsible for this.
5bb04747
VSO
2944 *
2945 * Returns new created child.
548a74c0 2946 */
5bb04747
VSO
2947static BdrvChild *bdrv_attach_child_common(BlockDriverState *child_bs,
2948 const char *child_name,
2949 const BdrvChildClass *child_class,
2950 BdrvChildRole child_role,
2951 uint64_t perm, uint64_t shared_perm,
2952 void *opaque,
2953 Transaction *tran, Error **errp)
548a74c0
VSO
2954{
2955 BdrvChild *new_child;
2956 AioContext *parent_ctx;
2957 AioContext *child_ctx = bdrv_get_aio_context(child_bs);
2958
da261b69 2959 assert(child_class->get_parent_desc);
bdb73476 2960 GLOBAL_STATE_CODE();
548a74c0
VSO
2961
2962 new_child = g_new(BdrvChild, 1);
2963 *new_child = (BdrvChild) {
d5e6f437
KW
2964 .bs = NULL,
2965 .name = g_strdup(child_name),
bd86fb99 2966 .klass = child_class,
258b7765 2967 .role = child_role,
d5e6f437
KW
2968 .perm = perm,
2969 .shared_perm = shared_perm,
2970 .opaque = opaque,
df581792
KW
2971 };
2972
548a74c0
VSO
2973 /*
2974 * If the AioContexts don't match, first try to move the subtree of
132ada80 2975 * child_bs into the AioContext of the new parent. If this doesn't work,
548a74c0
VSO
2976 * try moving the parent into the AioContext of child_bs instead.
2977 */
2978 parent_ctx = bdrv_child_get_parent_aio_context(new_child);
2979 if (child_ctx != parent_ctx) {
2980 Error *local_err = NULL;
2981 int ret = bdrv_try_set_aio_context(child_bs, parent_ctx, &local_err);
2982
bd86fb99 2983 if (ret < 0 && child_class->can_set_aio_ctx) {
548a74c0
VSO
2984 GSList *ignore = g_slist_prepend(NULL, new_child);
2985 if (child_class->can_set_aio_ctx(new_child, child_ctx, &ignore,
2986 NULL))
2987 {
132ada80
KW
2988 error_free(local_err);
2989 ret = 0;
2990 g_slist_free(ignore);
548a74c0
VSO
2991 ignore = g_slist_prepend(NULL, new_child);
2992 child_class->set_aio_ctx(new_child, child_ctx, &ignore);
132ada80
KW
2993 }
2994 g_slist_free(ignore);
2995 }
548a74c0 2996
132ada80
KW
2997 if (ret < 0) {
2998 error_propagate(errp, local_err);
04c9c3a5 2999 bdrv_child_free(new_child);
5bb04747 3000 return NULL;
132ada80
KW
3001 }
3002 }
3003
548a74c0 3004 bdrv_ref(child_bs);
544acc7d 3005 bdrv_replace_child_noperm(new_child, child_bs);
548a74c0 3006
548a74c0
VSO
3007 BdrvAttachChildCommonState *s = g_new(BdrvAttachChildCommonState, 1);
3008 *s = (BdrvAttachChildCommonState) {
5bb04747 3009 .child = new_child,
548a74c0
VSO
3010 .old_parent_ctx = parent_ctx,
3011 .old_child_ctx = child_ctx,
3012 };
3013 tran_add(tran, &bdrv_attach_child_common_drv, s);
3014
5bb04747 3015 return new_child;
548a74c0
VSO
3016}
3017
f8d2ad78 3018/*
7ec390d5 3019 * Function doesn't update permissions, caller is responsible for this.
f8d2ad78 3020 */
5bb04747
VSO
3021static BdrvChild *bdrv_attach_child_noperm(BlockDriverState *parent_bs,
3022 BlockDriverState *child_bs,
3023 const char *child_name,
3024 const BdrvChildClass *child_class,
3025 BdrvChildRole child_role,
3026 Transaction *tran,
3027 Error **errp)
aa5a04c7 3028{
aa5a04c7
VSO
3029 uint64_t perm, shared_perm;
3030
3031 assert(parent_bs->drv);
bdb73476 3032 GLOBAL_STATE_CODE();
aa5a04c7 3033
bfb8aa6d
KW
3034 if (bdrv_recurse_has_child(child_bs, parent_bs)) {
3035 error_setg(errp, "Making '%s' a %s child of '%s' would create a cycle",
3036 child_bs->node_name, child_name, parent_bs->node_name);
5bb04747 3037 return NULL;
bfb8aa6d
KW
3038 }
3039
aa5a04c7
VSO
3040 bdrv_get_cumulative_perm(parent_bs, &perm, &shared_perm);
3041 bdrv_child_perm(parent_bs, child_bs, NULL, child_role, NULL,
3042 perm, shared_perm, &perm, &shared_perm);
3043
5bb04747
VSO
3044 return bdrv_attach_child_common(child_bs, child_name, child_class,
3045 child_role, perm, shared_perm, parent_bs,
3046 tran, errp);
aa5a04c7
VSO
3047}
3048
544acc7d 3049static void bdrv_detach_child(BdrvChild *child)
548a74c0 3050{
544acc7d 3051 BlockDriverState *old_bs = child->bs;
4954aace 3052
bdb73476 3053 GLOBAL_STATE_CODE();
544acc7d
VSO
3054 bdrv_replace_child_noperm(child, NULL);
3055 bdrv_child_free(child);
4954aace
VSO
3056
3057 if (old_bs) {
3058 /*
3059 * Update permissions for old node. We're just taking a parent away, so
3060 * we're loosening restrictions. Errors of permission update are not
3061 * fatal in this case, ignore them.
3062 */
3063 bdrv_refresh_perms(old_bs, NULL);
3064
3065 /*
3066 * When the parent requiring a non-default AioContext is removed, the
3067 * node moves back to the main AioContext
3068 */
3069 bdrv_try_set_aio_context(old_bs, qemu_get_aio_context(), NULL);
3070 }
548a74c0
VSO
3071}
3072
3073/*
3074 * This function steals the reference to child_bs from the caller.
3075 * That reference is later dropped by bdrv_root_unref_child().
3076 *
3077 * On failure NULL is returned, errp is set and the reference to
3078 * child_bs is also dropped.
3079 *
3080 * The caller must hold the AioContext lock @child_bs, but not that of @ctx
3081 * (unless @child_bs is already in @ctx).
3082 */
3083BdrvChild *bdrv_root_attach_child(BlockDriverState *child_bs,
3084 const char *child_name,
3085 const BdrvChildClass *child_class,
3086 BdrvChildRole child_role,
3087 uint64_t perm, uint64_t shared_perm,
3088 void *opaque, Error **errp)
3089{
3090 int ret;
5bb04747 3091 BdrvChild *child;
548a74c0
VSO
3092 Transaction *tran = tran_new();
3093
b4ad82aa
EGE
3094 GLOBAL_STATE_CODE();
3095
5bb04747 3096 child = bdrv_attach_child_common(child_bs, child_name, child_class,
548a74c0 3097 child_role, perm, shared_perm, opaque,
5bb04747
VSO
3098 tran, errp);
3099 if (!child) {
3100 ret = -EINVAL;
e878bb12 3101 goto out;
548a74c0
VSO
3102 }
3103
3104 ret = bdrv_refresh_perms(child_bs, errp);
548a74c0 3105
e878bb12
KW
3106out:
3107 tran_finalize(tran, ret);
f8d2ad78 3108
548a74c0 3109 bdrv_unref(child_bs);
5bb04747
VSO
3110
3111 return ret < 0 ? NULL : child;
df581792
KW
3112}
3113
b441dc71
AG
3114/*
3115 * This function transfers the reference to child_bs from the caller
3116 * to parent_bs. That reference is later dropped by parent_bs on
3117 * bdrv_close() or if someone calls bdrv_unref_child().
3118 *
3119 * On failure NULL is returned, errp is set and the reference to
3120 * child_bs is also dropped.
132ada80
KW
3121 *
3122 * If @parent_bs and @child_bs are in different AioContexts, the caller must
3123 * hold the AioContext lock for @child_bs, but not for @parent_bs.
b441dc71 3124 */
98292c61
WC
3125BdrvChild *bdrv_attach_child(BlockDriverState *parent_bs,
3126 BlockDriverState *child_bs,
3127 const char *child_name,
bd86fb99 3128 const BdrvChildClass *child_class,
258b7765 3129 BdrvChildRole child_role,
8b2ff529 3130 Error **errp)
f21d96d0 3131{
aa5a04c7 3132 int ret;
5bb04747 3133 BdrvChild *child;
aa5a04c7 3134 Transaction *tran = tran_new();
f68c598b 3135
f791bf7f
EGE
3136 GLOBAL_STATE_CODE();
3137
5bb04747
VSO
3138 child = bdrv_attach_child_noperm(parent_bs, child_bs, child_name,
3139 child_class, child_role, tran, errp);
3140 if (!child) {
3141 ret = -EINVAL;
aa5a04c7
VSO
3142 goto out;
3143 }
d5e6f437 3144
aa5a04c7
VSO
3145 ret = bdrv_refresh_perms(parent_bs, errp);
3146 if (ret < 0) {
3147 goto out;
d5e6f437
KW
3148 }
3149
aa5a04c7
VSO
3150out:
3151 tran_finalize(tran, ret);
3152
3153 bdrv_unref(child_bs);
3154
5bb04747 3155 return ret < 0 ? NULL : child;
f21d96d0
KW
3156}
3157
7b99a266 3158/* Callers must ensure that child->frozen is false. */
f21d96d0 3159void bdrv_root_unref_child(BdrvChild *child)
33a60407 3160{
779020cb
KW
3161 BlockDriverState *child_bs;
3162
f791bf7f
EGE
3163 GLOBAL_STATE_CODE();
3164
f21d96d0 3165 child_bs = child->bs;
544acc7d 3166 bdrv_detach_child(child);
f21d96d0
KW
3167 bdrv_unref(child_bs);
3168}
3169
332b3a17
VSO
3170typedef struct BdrvSetInheritsFrom {
3171 BlockDriverState *bs;
3172 BlockDriverState *old_inherits_from;
3173} BdrvSetInheritsFrom;
3174
3175static void bdrv_set_inherits_from_abort(void *opaque)
3176{
3177 BdrvSetInheritsFrom *s = opaque;
3178
3179 s->bs->inherits_from = s->old_inherits_from;
3180}
3181
3182static TransactionActionDrv bdrv_set_inherits_from_drv = {
3183 .abort = bdrv_set_inherits_from_abort,
3184 .clean = g_free,
3185};
3186
3187/* @tran is allowed to be NULL. In this case no rollback is possible */
3188static void bdrv_set_inherits_from(BlockDriverState *bs,
3189 BlockDriverState *new_inherits_from,
3190 Transaction *tran)
3191{
3192 if (tran) {
3193 BdrvSetInheritsFrom *s = g_new(BdrvSetInheritsFrom, 1);
3194
3195 *s = (BdrvSetInheritsFrom) {
3196 .bs = bs,
3197 .old_inherits_from = bs->inherits_from,
3198 };
3199
3200 tran_add(tran, &bdrv_set_inherits_from_drv, s);
3201 }
3202
3203 bs->inherits_from = new_inherits_from;
3204}
3205
3cf746b3
HR
3206/**
3207 * Clear all inherits_from pointers from children and grandchildren of
3208 * @root that point to @root, where necessary.
332b3a17 3209 * @tran is allowed to be NULL. In this case no rollback is possible
3cf746b3 3210 */
332b3a17
VSO
3211static void bdrv_unset_inherits_from(BlockDriverState *root, BdrvChild *child,
3212 Transaction *tran)
f21d96d0 3213{
3cf746b3 3214 BdrvChild *c;
4e4bf5c4 3215
3cf746b3
HR
3216 if (child->bs->inherits_from == root) {
3217 /*
3218 * Remove inherits_from only when the last reference between root and
3219 * child->bs goes away.
3220 */
3221 QLIST_FOREACH(c, &root->children, next) {
4e4bf5c4
KW
3222 if (c != child && c->bs == child->bs) {
3223 break;
3224 }
3225 }
3226 if (c == NULL) {
332b3a17 3227 bdrv_set_inherits_from(child->bs, NULL, tran);
4e4bf5c4 3228 }
33a60407
KW
3229 }
3230
3cf746b3 3231 QLIST_FOREACH(c, &child->bs->children, next) {
332b3a17 3232 bdrv_unset_inherits_from(root, c, tran);
3cf746b3
HR
3233 }
3234}
3235
7b99a266 3236/* Callers must ensure that child->frozen is false. */
3cf746b3
HR
3237void bdrv_unref_child(BlockDriverState *parent, BdrvChild *child)
3238{
f791bf7f 3239 GLOBAL_STATE_CODE();
3cf746b3
HR
3240 if (child == NULL) {
3241 return;
3242 }
3243
332b3a17 3244 bdrv_unset_inherits_from(parent, child, NULL);
f21d96d0 3245 bdrv_root_unref_child(child);
33a60407
KW
3246}
3247
5c8cab48
KW
3248
3249static void bdrv_parent_cb_change_media(BlockDriverState *bs, bool load)
3250{
3251 BdrvChild *c;
f0c28327 3252 GLOBAL_STATE_CODE();
5c8cab48 3253 QLIST_FOREACH(c, &bs->parents, next_parent) {
bd86fb99
HR
3254 if (c->klass->change_media) {
3255 c->klass->change_media(c, load);
5c8cab48
KW
3256 }
3257 }
3258}
3259
0065c455
AG
3260/* Return true if you can reach parent going through child->inherits_from
3261 * recursively. If parent or child are NULL, return false */
3262static bool bdrv_inherits_from_recursive(BlockDriverState *child,
3263 BlockDriverState *parent)
3264{
3265 while (child && child != parent) {
3266 child = child->inherits_from;
3267 }
3268
3269 return child != NULL;
3270}
3271
25191e5f
HR
3272/*
3273 * Return the BdrvChildRole for @bs's backing child. bs->backing is
3274 * mostly used for COW backing children (role = COW), but also for
3275 * filtered children (role = FILTERED | PRIMARY).
3276 */
3277static BdrvChildRole bdrv_backing_role(BlockDriverState *bs)
3278{
3279 if (bs->drv && bs->drv->is_filter) {
3280 return BDRV_CHILD_FILTERED | BDRV_CHILD_PRIMARY;
3281 } else {
3282 return BDRV_CHILD_COW;
3283 }
3284}
3285
5db15a57 3286/*
e9238278
VSO
3287 * Sets the bs->backing or bs->file link of a BDS. A new reference is created;
3288 * callers which don't need their own reference any more must call bdrv_unref().
7ec390d5
VSO
3289 *
3290 * Function doesn't update permissions, caller is responsible for this.
5db15a57 3291 */
e9238278
VSO
3292static int bdrv_set_file_or_backing_noperm(BlockDriverState *parent_bs,
3293 BlockDriverState *child_bs,
3294 bool is_backing,
3295 Transaction *tran, Error **errp)
8d24cce1 3296{
e9238278
VSO
3297 bool update_inherits_from =
3298 bdrv_inherits_from_recursive(child_bs, parent_bs);
3299 BdrvChild *child = is_backing ? parent_bs->backing : parent_bs->file;
3300 BdrvChildRole role;
0065c455 3301
bdb73476
EGE
3302 GLOBAL_STATE_CODE();
3303
e9238278
VSO
3304 if (!parent_bs->drv) {
3305 /*
3306 * Node without drv is an object without a class :/. TODO: finally fix
3307 * qcow2 driver to never clear bs->drv and implement format corruption
3308 * handling in other way.
3309 */
3310 error_setg(errp, "Node corrupted");
3311 return -EINVAL;
3312 }
3313
3314 if (child && child->frozen) {
3315 error_setg(errp, "Cannot change frozen '%s' link from '%s' to '%s'",
3316 child->name, parent_bs->node_name, child->bs->node_name);
a1e708fc 3317 return -EPERM;
2cad1ebe
AG
3318 }
3319
25f78d9e
VSO
3320 if (is_backing && !parent_bs->drv->is_filter &&
3321 !parent_bs->drv->supports_backing)
3322 {
3323 error_setg(errp, "Driver '%s' of node '%s' does not support backing "
3324 "files", parent_bs->drv->format_name, parent_bs->node_name);
3325 return -EINVAL;
3326 }
3327
e9238278
VSO
3328 if (parent_bs->drv->is_filter) {
3329 role = BDRV_CHILD_FILTERED | BDRV_CHILD_PRIMARY;
3330 } else if (is_backing) {
3331 role = BDRV_CHILD_COW;
3332 } else {
3333 /*
3334 * We only can use same role as it is in existing child. We don't have
3335 * infrastructure to determine role of file child in generic way
3336 */
3337 if (!child) {
3338 error_setg(errp, "Cannot set file child to format node without "
3339 "file child");
3340 return -EINVAL;
3341 }
3342 role = child->role;
826b6ca0
FZ
3343 }
3344
e9238278
VSO
3345 if (child) {
3346 bdrv_unset_inherits_from(parent_bs, child, tran);
57f08941 3347 bdrv_remove_child(child, tran);
e9238278
VSO
3348 }
3349
3350 if (!child_bs) {
8d24cce1
FZ
3351 goto out;
3352 }
12fa4af6 3353
5bb04747
VSO
3354 child = bdrv_attach_child_noperm(parent_bs, child_bs,
3355 is_backing ? "backing" : "file",
3356 &child_of_bds, role,
3357 tran, errp);
3358 if (!child) {
3359 return -EINVAL;
a1e708fc
VSO
3360 }
3361
160333e1
VSO
3362
3363 /*
e9238278 3364 * If inherits_from pointed recursively to bs then let's update it to
160333e1
VSO
3365 * point directly to bs (else it will become NULL).
3366 */
a1e708fc 3367 if (update_inherits_from) {
e9238278 3368 bdrv_set_inherits_from(child_bs, parent_bs, tran);
0065c455 3369 }
826b6ca0 3370
8d24cce1 3371out:
e9238278 3372 bdrv_refresh_limits(parent_bs, tran, NULL);
160333e1
VSO
3373
3374 return 0;
3375}
3376
e9238278
VSO
3377static int bdrv_set_backing_noperm(BlockDriverState *bs,
3378 BlockDriverState *backing_hd,
3379 Transaction *tran, Error **errp)
3380{
bdb73476 3381 GLOBAL_STATE_CODE();
e9238278
VSO
3382 return bdrv_set_file_or_backing_noperm(bs, backing_hd, true, tran, errp);
3383}
3384
160333e1
VSO
3385int bdrv_set_backing_hd(BlockDriverState *bs, BlockDriverState *backing_hd,
3386 Error **errp)
3387{
3388 int ret;
3389 Transaction *tran = tran_new();
3390
f791bf7f 3391 GLOBAL_STATE_CODE();
c0829cb1
VSO
3392 bdrv_drained_begin(bs);
3393
160333e1
VSO
3394 ret = bdrv_set_backing_noperm(bs, backing_hd, tran, errp);
3395 if (ret < 0) {
3396 goto out;
3397 }
3398
3399 ret = bdrv_refresh_perms(bs, errp);
3400out:
3401 tran_finalize(tran, ret);
a1e708fc 3402
c0829cb1
VSO
3403 bdrv_drained_end(bs);
3404
a1e708fc 3405 return ret;
8d24cce1
FZ
3406}
3407
31ca6d07
KW
3408/*
3409 * Opens the backing file for a BlockDriverState if not yet open
3410 *
d9b7b057
KW
3411 * bdref_key specifies the key for the image's BlockdevRef in the options QDict.
3412 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
3413 * itself, all options starting with "${bdref_key}." are considered part of the
3414 * BlockdevRef.
3415 *
3416 * TODO Can this be unified with bdrv_open_image()?
31ca6d07 3417 */
d9b7b057
KW
3418int bdrv_open_backing_file(BlockDriverState *bs, QDict *parent_options,
3419 const char *bdref_key, Error **errp)
9156df12 3420{
6b6833c1 3421 char *backing_filename = NULL;
d9b7b057
KW
3422 char *bdref_key_dot;
3423 const char *reference = NULL;
317fc44e 3424 int ret = 0;
998c2019 3425 bool implicit_backing = false;
8d24cce1 3426 BlockDriverState *backing_hd;
d9b7b057
KW
3427 QDict *options;
3428 QDict *tmp_parent_options = NULL;
34b5d2c6 3429 Error *local_err = NULL;
9156df12 3430
f791bf7f
EGE
3431 GLOBAL_STATE_CODE();
3432
760e0063 3433 if (bs->backing != NULL) {
1ba4b6a5 3434 goto free_exit;
9156df12
PB
3435 }
3436
31ca6d07 3437 /* NULL means an empty set of options */
d9b7b057
KW
3438 if (parent_options == NULL) {
3439 tmp_parent_options = qdict_new();
3440 parent_options = tmp_parent_options;
31ca6d07
KW
3441 }
3442
9156df12 3443 bs->open_flags &= ~BDRV_O_NO_BACKING;
d9b7b057
KW
3444
3445 bdref_key_dot = g_strdup_printf("%s.", bdref_key);
3446 qdict_extract_subqdict(parent_options, &options, bdref_key_dot);
3447 g_free(bdref_key_dot);
3448
129c7d1c
MA
3449 /*
3450 * Caution: while qdict_get_try_str() is fine, getting non-string
3451 * types would require more care. When @parent_options come from
3452 * -blockdev or blockdev_add, its members are typed according to
3453 * the QAPI schema, but when they come from -drive, they're all
3454 * QString.
3455 */
d9b7b057
KW
3456 reference = qdict_get_try_str(parent_options, bdref_key);
3457 if (reference || qdict_haskey(options, "file.filename")) {
6b6833c1 3458 /* keep backing_filename NULL */
1cb6f506 3459 } else if (bs->backing_file[0] == '\0' && qdict_size(options) == 0) {
cb3e7f08 3460 qobject_unref(options);
1ba4b6a5 3461 goto free_exit;
dbecebdd 3462 } else {
998c2019
HR
3463 if (qdict_size(options) == 0) {
3464 /* If the user specifies options that do not modify the
3465 * backing file's behavior, we might still consider it the
3466 * implicit backing file. But it's easier this way, and
3467 * just specifying some of the backing BDS's options is
3468 * only possible with -drive anyway (otherwise the QAPI
3469 * schema forces the user to specify everything). */
3470 implicit_backing = !strcmp(bs->auto_backing_file, bs->backing_file);
3471 }
3472
6b6833c1 3473 backing_filename = bdrv_get_full_backing_filename(bs, &local_err);
9f07429e
HR
3474 if (local_err) {
3475 ret = -EINVAL;
3476 error_propagate(errp, local_err);
cb3e7f08 3477 qobject_unref(options);
9f07429e
HR
3478 goto free_exit;
3479 }
9156df12
PB
3480 }
3481
8ee79e70
KW
3482 if (!bs->drv || !bs->drv->supports_backing) {
3483 ret = -EINVAL;
3484 error_setg(errp, "Driver doesn't support backing files");
cb3e7f08 3485 qobject_unref(options);
8ee79e70
KW
3486 goto free_exit;
3487 }
3488
6bff597b
PK
3489 if (!reference &&
3490 bs->backing_format[0] != '\0' && !qdict_haskey(options, "driver")) {
46f5ac20 3491 qdict_put_str(options, "driver", bs->backing_format);
9156df12
PB
3492 }
3493
6b6833c1 3494 backing_hd = bdrv_open_inherit(backing_filename, reference, options, 0, bs,
25191e5f 3495 &child_of_bds, bdrv_backing_role(bs), errp);
5b363937 3496 if (!backing_hd) {
9156df12 3497 bs->open_flags |= BDRV_O_NO_BACKING;
e43bfd9c 3498 error_prepend(errp, "Could not open backing file: ");
5b363937 3499 ret = -EINVAL;
1ba4b6a5 3500 goto free_exit;
9156df12 3501 }
df581792 3502
998c2019
HR
3503 if (implicit_backing) {
3504 bdrv_refresh_filename(backing_hd);
3505 pstrcpy(bs->auto_backing_file, sizeof(bs->auto_backing_file),
3506 backing_hd->filename);
3507 }
3508
5db15a57
KW
3509 /* Hook up the backing file link; drop our reference, bs owns the
3510 * backing_hd reference now */
dc9c10a1 3511 ret = bdrv_set_backing_hd(bs, backing_hd, errp);
5db15a57 3512 bdrv_unref(backing_hd);
dc9c10a1 3513 if (ret < 0) {
12fa4af6
KW
3514 goto free_exit;
3515 }
d80ac658 3516
d9b7b057
KW
3517 qdict_del(parent_options, bdref_key);
3518
1ba4b6a5
BC
3519free_exit:
3520 g_free(backing_filename);
cb3e7f08 3521 qobject_unref(tmp_parent_options);
1ba4b6a5 3522 return ret;
9156df12
PB
3523}
3524
2d6b86af
KW
3525static BlockDriverState *
3526bdrv_open_child_bs(const char *filename, QDict *options, const char *bdref_key,
bd86fb99 3527 BlockDriverState *parent, const BdrvChildClass *child_class,
272c02ea 3528 BdrvChildRole child_role, bool allow_none, Error **errp)
da557aac 3529{
2d6b86af 3530 BlockDriverState *bs = NULL;
da557aac 3531 QDict *image_options;
da557aac
HR
3532 char *bdref_key_dot;
3533 const char *reference;
3534
bd86fb99 3535 assert(child_class != NULL);
f67503e5 3536
da557aac
HR
3537 bdref_key_dot = g_strdup_printf("%s.", bdref_key);
3538 qdict_extract_subqdict(options, &image_options, bdref_key_dot);
3539 g_free(bdref_key_dot);
3540
129c7d1c
MA
3541 /*
3542 * Caution: while qdict_get_try_str() is fine, getting non-string
3543 * types would require more care. When @options come from
3544 * -blockdev or blockdev_add, its members are typed according to
3545 * the QAPI schema, but when they come from -drive, they're all
3546 * QString.
3547 */
da557aac
HR
3548 reference = qdict_get_try_str(options, bdref_key);
3549 if (!filename && !reference && !qdict_size(image_options)) {
b4b059f6 3550 if (!allow_none) {
da557aac
HR
3551 error_setg(errp, "A block device must be specified for \"%s\"",
3552 bdref_key);
da557aac 3553 }
cb3e7f08 3554 qobject_unref(image_options);
da557aac
HR
3555 goto done;
3556 }
3557
5b363937 3558 bs = bdrv_open_inherit(filename, reference, image_options, 0,
272c02ea 3559 parent, child_class, child_role, errp);
5b363937 3560 if (!bs) {
df581792
KW
3561 goto done;
3562 }
3563
da557aac
HR
3564done:
3565 qdict_del(options, bdref_key);
2d6b86af
KW
3566 return bs;
3567}
3568
3569/*
3570 * Opens a disk image whose options are given as BlockdevRef in another block
3571 * device's options.
3572 *
3573 * If allow_none is true, no image will be opened if filename is false and no
3574 * BlockdevRef is given. NULL will be returned, but errp remains unset.
3575 *
3576 * bdrev_key specifies the key for the image's BlockdevRef in the options QDict.
3577 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
3578 * itself, all options starting with "${bdref_key}." are considered part of the
3579 * BlockdevRef.
3580 *
3581 * The BlockdevRef will be removed from the options QDict.
3582 */
3583BdrvChild *bdrv_open_child(const char *filename,
3584 QDict *options, const char *bdref_key,
3585 BlockDriverState *parent,
bd86fb99 3586 const BdrvChildClass *child_class,
258b7765 3587 BdrvChildRole child_role,
2d6b86af
KW
3588 bool allow_none, Error **errp)
3589{
3590 BlockDriverState *bs;
3591
f791bf7f
EGE
3592 GLOBAL_STATE_CODE();
3593
bd86fb99 3594 bs = bdrv_open_child_bs(filename, options, bdref_key, parent, child_class,
272c02ea 3595 child_role, allow_none, errp);
2d6b86af
KW
3596 if (bs == NULL) {
3597 return NULL;
3598 }
3599
258b7765
HR
3600 return bdrv_attach_child(parent, bs, bdref_key, child_class, child_role,
3601 errp);
b4b059f6
KW
3602}
3603
83930780
VSO
3604/*
3605 * Wrapper on bdrv_open_child() for most popular case: open primary child of bs.
3606 */
3607int bdrv_open_file_child(const char *filename,
3608 QDict *options, const char *bdref_key,
3609 BlockDriverState *parent, Error **errp)
3610{
3611 BdrvChildRole role;
3612
3613 /* commit_top and mirror_top don't use this function */
3614 assert(!parent->drv->filtered_child_is_backing);
83930780
VSO
3615 role = parent->drv->is_filter ?
3616 (BDRV_CHILD_FILTERED | BDRV_CHILD_PRIMARY) : BDRV_CHILD_IMAGE;
3617
5bb04747
VSO
3618 if (!bdrv_open_child(filename, options, bdref_key, parent,
3619 &child_of_bds, role, false, errp))
3620 {
3621 return -EINVAL;
3622 }
83930780 3623
5bb04747 3624 return 0;
83930780
VSO
3625}
3626
bd86fb99
HR
3627/*
3628 * TODO Future callers may need to specify parent/child_class in order for
3629 * option inheritance to work. Existing callers use it for the root node.
3630 */
e1d74bc6
KW
3631BlockDriverState *bdrv_open_blockdev_ref(BlockdevRef *ref, Error **errp)
3632{
3633 BlockDriverState *bs = NULL;
e1d74bc6
KW
3634 QObject *obj = NULL;
3635 QDict *qdict = NULL;
3636 const char *reference = NULL;
3637 Visitor *v = NULL;
3638
f791bf7f
EGE
3639 GLOBAL_STATE_CODE();
3640
e1d74bc6
KW
3641 if (ref->type == QTYPE_QSTRING) {
3642 reference = ref->u.reference;
3643 } else {
3644 BlockdevOptions *options = &ref->u.definition;
3645 assert(ref->type == QTYPE_QDICT);
3646
3647 v = qobject_output_visitor_new(&obj);
1f584248 3648 visit_type_BlockdevOptions(v, NULL, &options, &error_abort);
e1d74bc6
KW
3649 visit_complete(v, &obj);
3650
7dc847eb 3651 qdict = qobject_to(QDict, obj);
e1d74bc6
KW
3652 qdict_flatten(qdict);
3653
3654 /* bdrv_open_inherit() defaults to the values in bdrv_flags (for
3655 * compatibility with other callers) rather than what we want as the
3656 * real defaults. Apply the defaults here instead. */
3657 qdict_set_default_str(qdict, BDRV_OPT_CACHE_DIRECT, "off");
3658 qdict_set_default_str(qdict, BDRV_OPT_CACHE_NO_FLUSH, "off");
3659 qdict_set_default_str(qdict, BDRV_OPT_READ_ONLY, "off");
e35bdc12
KW
3660 qdict_set_default_str(qdict, BDRV_OPT_AUTO_READ_ONLY, "off");
3661
e1d74bc6
KW
3662 }
3663
272c02ea 3664 bs = bdrv_open_inherit(NULL, reference, qdict, 0, NULL, NULL, 0, errp);
e1d74bc6 3665 obj = NULL;
cb3e7f08 3666 qobject_unref(obj);
e1d74bc6
KW
3667 visit_free(v);
3668 return bs;
3669}
3670
66836189
HR
3671static BlockDriverState *bdrv_append_temp_snapshot(BlockDriverState *bs,
3672 int flags,
3673 QDict *snapshot_options,
3674 Error **errp)
b998875d 3675{
69fbfff9 3676 g_autofree char *tmp_filename = NULL;
b998875d 3677 int64_t total_size;
83d0521a 3678 QemuOpts *opts = NULL;
ff6ed714 3679 BlockDriverState *bs_snapshot = NULL;
b998875d
KW
3680 int ret;
3681
bdb73476
EGE
3682 GLOBAL_STATE_CODE();
3683
b998875d
KW
3684 /* if snapshot, we create a temporary backing file and open it
3685 instead of opening 'filename' directly */
3686
3687 /* Get the required size from the image */
f187743a
KW
3688 total_size = bdrv_getlength(bs);
3689 if (total_size < 0) {
3690 error_setg_errno(errp, -total_size, "Could not get image size");
1ba4b6a5 3691 goto out;
f187743a 3692 }
b998875d
KW
3693
3694 /* Create the temporary image */
69fbfff9
BM
3695 tmp_filename = create_tmp_file(errp);
3696 if (!tmp_filename) {
1ba4b6a5 3697 goto out;
b998875d
KW
3698 }
3699
ef810437 3700 opts = qemu_opts_create(bdrv_qcow2.create_opts, NULL, 0,
c282e1fd 3701 &error_abort);
39101f25 3702 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, total_size, &error_abort);
e43bfd9c 3703 ret = bdrv_create(&bdrv_qcow2, tmp_filename, opts, errp);
83d0521a 3704 qemu_opts_del(opts);
b998875d 3705 if (ret < 0) {
e43bfd9c
MA
3706 error_prepend(errp, "Could not create temporary overlay '%s': ",
3707 tmp_filename);
1ba4b6a5 3708 goto out;
b998875d
KW
3709 }
3710
73176bee 3711 /* Prepare options QDict for the temporary file */
46f5ac20
EB
3712 qdict_put_str(snapshot_options, "file.driver", "file");
3713 qdict_put_str(snapshot_options, "file.filename", tmp_filename);
3714 qdict_put_str(snapshot_options, "driver", "qcow2");
b998875d 3715
5b363937 3716 bs_snapshot = bdrv_open(NULL, NULL, snapshot_options, flags, errp);
73176bee 3717 snapshot_options = NULL;
5b363937 3718 if (!bs_snapshot) {
1ba4b6a5 3719 goto out;
b998875d
KW
3720 }
3721
934aee14
VSO
3722 ret = bdrv_append(bs_snapshot, bs, errp);
3723 if (ret < 0) {
ff6ed714 3724 bs_snapshot = NULL;
b2c2832c
KW
3725 goto out;
3726 }
1ba4b6a5
BC
3727
3728out:
cb3e7f08 3729 qobject_unref(snapshot_options);
ff6ed714 3730 return bs_snapshot;
b998875d
KW
3731}
3732
b6ce07aa
KW
3733/*
3734 * Opens a disk image (raw, qcow2, vmdk, ...)
de9c0cec
KW
3735 *
3736 * options is a QDict of options to pass to the block drivers, or NULL for an
3737 * empty set of options. The reference to the QDict belongs to the block layer
3738 * after the call (even on failure), so if the caller intends to reuse the
cb3e7f08 3739 * dictionary, it needs to use qobject_ref() before calling bdrv_open.
f67503e5
HR
3740 *
3741 * If *pbs is NULL, a new BDS will be created with a pointer to it stored there.
3742 * If it is not NULL, the referenced BDS will be reused.
ddf5636d
HR
3743 *
3744 * The reference parameter may be used to specify an existing block device which
3745 * should be opened. If specified, neither options nor a filename may be given,
3746 * nor can an existing BDS be reused (that is, *pbs has to be NULL).
b6ce07aa 3747 */
5b363937
HR
3748static BlockDriverState *bdrv_open_inherit(const char *filename,
3749 const char *reference,
3750 QDict *options, int flags,
3751 BlockDriverState *parent,
bd86fb99 3752 const BdrvChildClass *child_class,
272c02ea 3753 BdrvChildRole child_role,
5b363937 3754 Error **errp)
ea2384d3 3755{
b6ce07aa 3756 int ret;
5696c6e3 3757 BlockBackend *file = NULL;
9a4f4c31 3758 BlockDriverState *bs;
ce343771 3759 BlockDriver *drv = NULL;
2f624b80 3760 BdrvChild *child;
74fe54f2 3761 const char *drvname;
3e8c2e57 3762 const char *backing;
34b5d2c6 3763 Error *local_err = NULL;
73176bee 3764 QDict *snapshot_options = NULL;
b1e6fc08 3765 int snapshot_flags = 0;
712e7874 3766
bd86fb99
HR
3767 assert(!child_class || !flags);
3768 assert(!child_class == !parent);
f0c28327 3769 GLOBAL_STATE_CODE();
f67503e5 3770
ddf5636d
HR
3771 if (reference) {
3772 bool options_non_empty = options ? qdict_size(options) : false;
cb3e7f08 3773 qobject_unref(options);
ddf5636d 3774
ddf5636d
HR
3775 if (filename || options_non_empty) {
3776 error_setg(errp, "Cannot reference an existing block device with "
3777 "additional options or a new filename");
5b363937 3778 return NULL;
ddf5636d
HR
3779 }
3780
3781 bs = bdrv_lookup_bs(reference, reference, errp);
3782 if (!bs) {
5b363937 3783 return NULL;
ddf5636d 3784 }
76b22320 3785
ddf5636d 3786 bdrv_ref(bs);
5b363937 3787 return bs;
ddf5636d
HR
3788 }
3789
5b363937 3790 bs = bdrv_new();
f67503e5 3791
de9c0cec
KW
3792 /* NULL means an empty set of options */
3793 if (options == NULL) {
3794 options = qdict_new();
3795 }
3796
145f598e 3797 /* json: syntax counts as explicit options, as if in the QDict */
de3b53f0
KW
3798 parse_json_protocol(options, &filename, &local_err);
3799 if (local_err) {
de3b53f0
KW
3800 goto fail;
3801 }
3802
145f598e
KW
3803 bs->explicit_options = qdict_clone_shallow(options);
3804
bd86fb99 3805 if (child_class) {
3cdc69d3
HR
3806 bool parent_is_format;
3807
3808 if (parent->drv) {
3809 parent_is_format = parent->drv->is_format;
3810 } else {
3811 /*
3812 * parent->drv is not set yet because this node is opened for
3813 * (potential) format probing. That means that @parent is going
3814 * to be a format node.
3815 */
3816 parent_is_format = true;
3817 }
3818
bddcec37 3819 bs->inherits_from = parent;
3cdc69d3
HR
3820 child_class->inherit_options(child_role, parent_is_format,
3821 &flags, options,
bd86fb99 3822 parent->open_flags, parent->options);
f3930ed0
KW
3823 }
3824
de3b53f0 3825 ret = bdrv_fill_options(&options, filename, &flags, &local_err);
dfde483e 3826 if (ret < 0) {
462f5bcf
KW
3827 goto fail;
3828 }
3829
129c7d1c
MA
3830 /*
3831 * Set the BDRV_O_RDWR and BDRV_O_ALLOW_RDWR flags.
3832 * Caution: getting a boolean member of @options requires care.
3833 * When @options come from -blockdev or blockdev_add, members are
3834 * typed according to the QAPI schema, but when they come from
3835 * -drive, they're all QString.
3836 */
f87a0e29
AG
3837 if (g_strcmp0(qdict_get_try_str(options, BDRV_OPT_READ_ONLY), "on") &&
3838 !qdict_get_try_bool(options, BDRV_OPT_READ_ONLY, false)) {
3839 flags |= (BDRV_O_RDWR | BDRV_O_ALLOW_RDWR);
3840 } else {
3841 flags &= ~BDRV_O_RDWR;
14499ea5
AG
3842 }
3843
3844 if (flags & BDRV_O_SNAPSHOT) {
3845 snapshot_options = qdict_new();
3846 bdrv_temp_snapshot_options(&snapshot_flags, snapshot_options,
3847 flags, options);
f87a0e29
AG
3848 /* Let bdrv_backing_options() override "read-only" */
3849 qdict_del(options, BDRV_OPT_READ_ONLY);
00ff7ffd
HR
3850 bdrv_inherited_options(BDRV_CHILD_COW, true,
3851 &flags, options, flags, options);
14499ea5
AG
3852 }
3853
62392ebb
KW
3854 bs->open_flags = flags;
3855 bs->options = options;
3856 options = qdict_clone_shallow(options);
3857
76c591b0 3858 /* Find the right image format driver */
129c7d1c 3859 /* See cautionary note on accessing @options above */
76c591b0
KW
3860 drvname = qdict_get_try_str(options, "driver");
3861 if (drvname) {
3862 drv = bdrv_find_format(drvname);
76c591b0
KW
3863 if (!drv) {
3864 error_setg(errp, "Unknown driver: '%s'", drvname);
76c591b0
KW
3865 goto fail;
3866 }
3867 }
3868
3869 assert(drvname || !(flags & BDRV_O_PROTOCOL));
76c591b0 3870
129c7d1c 3871 /* See cautionary note on accessing @options above */
3e8c2e57 3872 backing = qdict_get_try_str(options, "backing");
e59a0cf1
HR
3873 if (qobject_to(QNull, qdict_get(options, "backing")) != NULL ||
3874 (backing && *backing == '\0'))
3875 {
4f7be280
HR
3876 if (backing) {
3877 warn_report("Use of \"backing\": \"\" is deprecated; "
3878 "use \"backing\": null instead");
3879 }
3e8c2e57 3880 flags |= BDRV_O_NO_BACKING;
ae0f57f0
KW
3881 qdict_del(bs->explicit_options, "backing");
3882 qdict_del(bs->options, "backing");
3e8c2e57
AG
3883 qdict_del(options, "backing");
3884 }
3885
5696c6e3 3886 /* Open image file without format layer. This BlockBackend is only used for
4e4bf5c4
KW
3887 * probing, the block drivers will do their own bdrv_open_child() for the
3888 * same BDS, which is why we put the node name back into options. */
f4788adc 3889 if ((flags & BDRV_O_PROTOCOL) == 0) {
5696c6e3
KW
3890 BlockDriverState *file_bs;
3891
3892 file_bs = bdrv_open_child_bs(filename, options, "file", bs,
58944401
HR
3893 &child_of_bds, BDRV_CHILD_IMAGE,
3894 true, &local_err);
1fdd6933 3895 if (local_err) {
f4788adc
KW
3896 goto fail;
3897 }
5696c6e3 3898 if (file_bs != NULL) {
dacaa162
KW
3899 /* Not requesting BLK_PERM_CONSISTENT_READ because we're only
3900 * looking at the header to guess the image format. This works even
3901 * in cases where a guest would not see a consistent state. */
d861ab3a 3902 file = blk_new(bdrv_get_aio_context(file_bs), 0, BLK_PERM_ALL);
d7086422 3903 blk_insert_bs(file, file_bs, &local_err);
5696c6e3 3904 bdrv_unref(file_bs);
d7086422
KW
3905 if (local_err) {
3906 goto fail;
3907 }
5696c6e3 3908
46f5ac20 3909 qdict_put_str(options, "file", bdrv_get_node_name(file_bs));
4e4bf5c4 3910 }
f500a6d3
KW
3911 }
3912
76c591b0 3913 /* Image format probing */
38f3ef57 3914 bs->probed = !drv;
76c591b0 3915 if (!drv && file) {
cf2ab8fc 3916 ret = find_image_format(file, filename, &drv, &local_err);
17b005f1 3917 if (ret < 0) {
8bfea15d 3918 goto fail;
2a05cbe4 3919 }
62392ebb
KW
3920 /*
3921 * This option update would logically belong in bdrv_fill_options(),
3922 * but we first need to open bs->file for the probing to work, while
3923 * opening bs->file already requires the (mostly) final set of options
3924 * so that cache mode etc. can be inherited.
3925 *
3926 * Adding the driver later is somewhat ugly, but it's not an option
3927 * that would ever be inherited, so it's correct. We just need to make
3928 * sure to update both bs->options (which has the full effective
3929 * options for bs) and options (which has file.* already removed).
3930 */
46f5ac20
EB
3931 qdict_put_str(bs->options, "driver", drv->format_name);
3932 qdict_put_str(options, "driver", drv->format_name);
76c591b0 3933 } else if (!drv) {
17b005f1 3934 error_setg(errp, "Must specify either driver or file");
8bfea15d 3935 goto fail;
ea2384d3 3936 }
b6ce07aa 3937
53a29513
HR
3938 /* BDRV_O_PROTOCOL must be set iff a protocol BDS is about to be created */
3939 assert(!!(flags & BDRV_O_PROTOCOL) == !!drv->bdrv_file_open);
3940 /* file must be NULL if a protocol BDS is about to be created
3941 * (the inverse results in an error message from bdrv_open_common()) */
3942 assert(!(flags & BDRV_O_PROTOCOL) || !file);
3943
b6ce07aa 3944 /* Open the image */
82dc8b41 3945 ret = bdrv_open_common(bs, file, options, &local_err);
b6ce07aa 3946 if (ret < 0) {
8bfea15d 3947 goto fail;
6987307c
CH
3948 }
3949
4e4bf5c4 3950 if (file) {
5696c6e3 3951 blk_unref(file);
f500a6d3
KW
3952 file = NULL;
3953 }
3954
b6ce07aa 3955 /* If there is a backing file, use it */
9156df12 3956 if ((flags & BDRV_O_NO_BACKING) == 0) {
d9b7b057 3957 ret = bdrv_open_backing_file(bs, options, "backing", &local_err);
b6ce07aa 3958 if (ret < 0) {
b6ad491a 3959 goto close_and_fail;
b6ce07aa 3960 }
b6ce07aa
KW
3961 }
3962
50196d7a
AG
3963 /* Remove all children options and references
3964 * from bs->options and bs->explicit_options */
2f624b80
AG
3965 QLIST_FOREACH(child, &bs->children, next) {
3966 char *child_key_dot;
3967 child_key_dot = g_strdup_printf("%s.", child->name);
3968 qdict_extract_subqdict(bs->explicit_options, NULL, child_key_dot);
3969 qdict_extract_subqdict(bs->options, NULL, child_key_dot);
50196d7a
AG
3970 qdict_del(bs->explicit_options, child->name);
3971 qdict_del(bs->options, child->name);
2f624b80
AG
3972 g_free(child_key_dot);
3973 }
3974
b6ad491a 3975 /* Check if any unknown options were used */
7ad2757f 3976 if (qdict_size(options) != 0) {
b6ad491a 3977 const QDictEntry *entry = qdict_first(options);
5acd9d81
HR
3978 if (flags & BDRV_O_PROTOCOL) {
3979 error_setg(errp, "Block protocol '%s' doesn't support the option "
3980 "'%s'", drv->format_name, entry->key);
3981 } else {
d0e46a55
HR
3982 error_setg(errp,
3983 "Block format '%s' does not support the option '%s'",
3984 drv->format_name, entry->key);
5acd9d81 3985 }
b6ad491a 3986
b6ad491a
KW
3987 goto close_and_fail;
3988 }
b6ad491a 3989
c01c214b 3990 bdrv_parent_cb_change_media(bs, true);
b6ce07aa 3991
cb3e7f08 3992 qobject_unref(options);
8961be33 3993 options = NULL;
dd62f1ca
KW
3994
3995 /* For snapshot=on, create a temporary qcow2 overlay. bs points to the
3996 * temporary snapshot afterwards. */
3997 if (snapshot_flags) {
66836189
HR
3998 BlockDriverState *snapshot_bs;
3999 snapshot_bs = bdrv_append_temp_snapshot(bs, snapshot_flags,
4000 snapshot_options, &local_err);
73176bee 4001 snapshot_options = NULL;
dd62f1ca
KW
4002 if (local_err) {
4003 goto close_and_fail;
4004 }
5b363937
HR
4005 /* We are not going to return bs but the overlay on top of it
4006 * (snapshot_bs); thus, we have to drop the strong reference to bs
4007 * (which we obtained by calling bdrv_new()). bs will not be deleted,
4008 * though, because the overlay still has a reference to it. */
4009 bdrv_unref(bs);
4010 bs = snapshot_bs;
dd62f1ca
KW
4011 }
4012
5b363937 4013 return bs;
b6ce07aa 4014
8bfea15d 4015fail:
5696c6e3 4016 blk_unref(file);
cb3e7f08
MAL
4017 qobject_unref(snapshot_options);
4018 qobject_unref(bs->explicit_options);
4019 qobject_unref(bs->options);
4020 qobject_unref(options);
de9c0cec 4021 bs->options = NULL;
998cbd6a 4022 bs->explicit_options = NULL;
5b363937 4023 bdrv_unref(bs);
621ff94d 4024 error_propagate(errp, local_err);
5b363937 4025 return NULL;
de9c0cec 4026
b6ad491a 4027close_and_fail:
5b363937 4028 bdrv_unref(bs);
cb3e7f08
MAL
4029 qobject_unref(snapshot_options);
4030 qobject_unref(options);
621ff94d 4031 error_propagate(errp, local_err);
5b363937 4032 return NULL;
b6ce07aa
KW
4033}
4034
5b363937
HR
4035BlockDriverState *bdrv_open(const char *filename, const char *reference,
4036 QDict *options, int flags, Error **errp)
f3930ed0 4037{
f791bf7f
EGE
4038 GLOBAL_STATE_CODE();
4039
5b363937 4040 return bdrv_open_inherit(filename, reference, options, flags, NULL,
272c02ea 4041 NULL, 0, errp);
f3930ed0
KW
4042}
4043
faf116b4
AG
4044/* Return true if the NULL-terminated @list contains @str */
4045static bool is_str_in_list(const char *str, const char *const *list)
4046{
4047 if (str && list) {
4048 int i;
4049 for (i = 0; list[i] != NULL; i++) {
4050 if (!strcmp(str, list[i])) {
4051 return true;
4052 }
4053 }
4054 }
4055 return false;
4056}
4057
4058/*
4059 * Check that every option set in @bs->options is also set in
4060 * @new_opts.
4061 *
4062 * Options listed in the common_options list and in
4063 * @bs->drv->mutable_opts are skipped.
4064 *
4065 * Return 0 on success, otherwise return -EINVAL and set @errp.
4066 */
4067static int bdrv_reset_options_allowed(BlockDriverState *bs,
4068 const QDict *new_opts, Error **errp)
4069{
4070 const QDictEntry *e;
4071 /* These options are common to all block drivers and are handled
4072 * in bdrv_reopen_prepare() so they can be left out of @new_opts */
4073 const char *const common_options[] = {
4074 "node-name", "discard", "cache.direct", "cache.no-flush",
4075 "read-only", "auto-read-only", "detect-zeroes", NULL
4076 };
4077
4078 for (e = qdict_first(bs->options); e; e = qdict_next(bs->options, e)) {
4079 if (!qdict_haskey(new_opts, e->key) &&
4080 !is_str_in_list(e->key, common_options) &&
4081 !is_str_in_list(e->key, bs->drv->mutable_opts)) {
4082 error_setg(errp, "Option '%s' cannot be reset "
4083 "to its default value", e->key);
4084 return -EINVAL;
4085 }
4086 }
4087
4088 return 0;
4089}
4090
cb828c31
AG
4091/*
4092 * Returns true if @child can be reached recursively from @bs
4093 */
4094static bool bdrv_recurse_has_child(BlockDriverState *bs,
4095 BlockDriverState *child)
4096{
4097 BdrvChild *c;
4098
4099 if (bs == child) {
4100 return true;
4101 }
4102
4103 QLIST_FOREACH(c, &bs->children, next) {
4104 if (bdrv_recurse_has_child(c->bs, child)) {
4105 return true;
4106 }
4107 }
4108
4109 return false;
4110}
4111
e971aa12
JC
4112/*
4113 * Adds a BlockDriverState to a simple queue for an atomic, transactional
4114 * reopen of multiple devices.
4115 *
859443b0 4116 * bs_queue can either be an existing BlockReopenQueue that has had QTAILQ_INIT
e971aa12
JC
4117 * already performed, or alternatively may be NULL a new BlockReopenQueue will
4118 * be created and initialized. This newly created BlockReopenQueue should be
4119 * passed back in for subsequent calls that are intended to be of the same
4120 * atomic 'set'.
4121 *
4122 * bs is the BlockDriverState to add to the reopen queue.
4123 *
4d2cb092
KW
4124 * options contains the changed options for the associated bs
4125 * (the BlockReopenQueue takes ownership)
4126 *
e971aa12
JC
4127 * flags contains the open flags for the associated bs
4128 *
4129 * returns a pointer to bs_queue, which is either the newly allocated
4130 * bs_queue, or the existing bs_queue being used.
4131 *
1a63a907 4132 * bs must be drained between bdrv_reopen_queue() and bdrv_reopen_multiple().
e971aa12 4133 */
28518102
KW
4134static BlockReopenQueue *bdrv_reopen_queue_child(BlockReopenQueue *bs_queue,
4135 BlockDriverState *bs,
4136 QDict *options,
bd86fb99 4137 const BdrvChildClass *klass,
272c02ea 4138 BdrvChildRole role,
3cdc69d3 4139 bool parent_is_format,
28518102 4140 QDict *parent_options,
077e8e20
AG
4141 int parent_flags,
4142 bool keep_old_opts)
e971aa12
JC
4143{
4144 assert(bs != NULL);
4145
4146 BlockReopenQueueEntry *bs_entry;
67251a31 4147 BdrvChild *child;
9aa09ddd
AG
4148 QDict *old_options, *explicit_options, *options_copy;
4149 int flags;
4150 QemuOpts *opts;
67251a31 4151
1a63a907
KW
4152 /* Make sure that the caller remembered to use a drained section. This is
4153 * important to avoid graph changes between the recursive queuing here and
4154 * bdrv_reopen_multiple(). */
4155 assert(bs->quiesce_counter > 0);
f0c28327 4156 GLOBAL_STATE_CODE();
1a63a907 4157
e971aa12
JC
4158 if (bs_queue == NULL) {
4159 bs_queue = g_new0(BlockReopenQueue, 1);
859443b0 4160 QTAILQ_INIT(bs_queue);
e971aa12
JC
4161 }
4162
4d2cb092
KW
4163 if (!options) {
4164 options = qdict_new();
4165 }
4166
5b7ba05f 4167 /* Check if this BlockDriverState is already in the queue */
859443b0 4168 QTAILQ_FOREACH(bs_entry, bs_queue, entry) {
5b7ba05f
AG
4169 if (bs == bs_entry->state.bs) {
4170 break;
4171 }
4172 }
4173
28518102
KW
4174 /*
4175 * Precedence of options:
4176 * 1. Explicitly passed in options (highest)
9aa09ddd
AG
4177 * 2. Retained from explicitly set options of bs
4178 * 3. Inherited from parent node
4179 * 4. Retained from effective options of bs
28518102
KW
4180 */
4181
145f598e 4182 /* Old explicitly set values (don't overwrite by inherited value) */
077e8e20
AG
4183 if (bs_entry || keep_old_opts) {
4184 old_options = qdict_clone_shallow(bs_entry ?
4185 bs_entry->state.explicit_options :
4186 bs->explicit_options);
4187 bdrv_join_options(bs, options, old_options);
4188 qobject_unref(old_options);
5b7ba05f 4189 }
145f598e
KW
4190
4191 explicit_options = qdict_clone_shallow(options);
4192
28518102
KW
4193 /* Inherit from parent node */
4194 if (parent_options) {
9aa09ddd 4195 flags = 0;
3cdc69d3 4196 klass->inherit_options(role, parent_is_format, &flags, options,
272c02ea 4197 parent_flags, parent_options);
9aa09ddd
AG
4198 } else {
4199 flags = bdrv_get_flags(bs);
28518102
KW
4200 }
4201
077e8e20
AG
4202 if (keep_old_opts) {
4203 /* Old values are used for options that aren't set yet */
4204 old_options = qdict_clone_shallow(bs->options);
4205 bdrv_join_options(bs, options, old_options);
4206 qobject_unref(old_options);
4207 }
4d2cb092 4208
9aa09ddd
AG
4209 /* We have the final set of options so let's update the flags */
4210 options_copy = qdict_clone_shallow(options);
4211 opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
4212 qemu_opts_absorb_qdict(opts, options_copy, NULL);
4213 update_flags_from_options(&flags, opts);
4214 qemu_opts_del(opts);
4215 qobject_unref(options_copy);
4216
fd452021 4217 /* bdrv_open_inherit() sets and clears some additional flags internally */
f1f25a2e 4218 flags &= ~BDRV_O_PROTOCOL;
fd452021
KW
4219 if (flags & BDRV_O_RDWR) {
4220 flags |= BDRV_O_ALLOW_RDWR;
4221 }
f1f25a2e 4222
1857c97b
KW
4223 if (!bs_entry) {
4224 bs_entry = g_new0(BlockReopenQueueEntry, 1);
859443b0 4225 QTAILQ_INSERT_TAIL(bs_queue, bs_entry, entry);
1857c97b 4226 } else {
cb3e7f08
MAL
4227 qobject_unref(bs_entry->state.options);
4228 qobject_unref(bs_entry->state.explicit_options);
1857c97b
KW
4229 }
4230
4231 bs_entry->state.bs = bs;
4232 bs_entry->state.options = options;
4233 bs_entry->state.explicit_options = explicit_options;
4234 bs_entry->state.flags = flags;
4235
8546632e
AG
4236 /*
4237 * If keep_old_opts is false then it means that unspecified
4238 * options must be reset to their original value. We don't allow
4239 * resetting 'backing' but we need to know if the option is
4240 * missing in order to decide if we have to return an error.
4241 */
4242 if (!keep_old_opts) {
4243 bs_entry->state.backing_missing =
4244 !qdict_haskey(options, "backing") &&
4245 !qdict_haskey(options, "backing.driver");
4246 }
4247
67251a31 4248 QLIST_FOREACH(child, &bs->children, next) {
8546632e
AG
4249 QDict *new_child_options = NULL;
4250 bool child_keep_old = keep_old_opts;
67251a31 4251
4c9dfe5d
KW
4252 /* reopen can only change the options of block devices that were
4253 * implicitly created and inherited options. For other (referenced)
4254 * block devices, a syntax like "backing.foo" results in an error. */
67251a31
KW
4255 if (child->bs->inherits_from != bs) {
4256 continue;
4257 }
4258
8546632e
AG
4259 /* Check if the options contain a child reference */
4260 if (qdict_haskey(options, child->name)) {
4261 const char *childref = qdict_get_try_str(options, child->name);
4262 /*
4263 * The current child must not be reopened if the child
4264 * reference is null or points to a different node.
4265 */
4266 if (g_strcmp0(childref, child->bs->node_name)) {
4267 continue;
4268 }
4269 /*
4270 * If the child reference points to the current child then
4271 * reopen it with its existing set of options (note that
4272 * it can still inherit new options from the parent).
4273 */
4274 child_keep_old = true;
4275 } else {
4276 /* Extract child options ("child-name.*") */
4277 char *child_key_dot = g_strdup_printf("%s.", child->name);
4278 qdict_extract_subqdict(explicit_options, NULL, child_key_dot);
4279 qdict_extract_subqdict(options, &new_child_options, child_key_dot);
4280 g_free(child_key_dot);
4281 }
4c9dfe5d 4282
9aa09ddd 4283 bdrv_reopen_queue_child(bs_queue, child->bs, new_child_options,
3cdc69d3
HR
4284 child->klass, child->role, bs->drv->is_format,
4285 options, flags, child_keep_old);
e971aa12
JC
4286 }
4287
e971aa12
JC
4288 return bs_queue;
4289}
4290
28518102
KW
4291BlockReopenQueue *bdrv_reopen_queue(BlockReopenQueue *bs_queue,
4292 BlockDriverState *bs,
077e8e20 4293 QDict *options, bool keep_old_opts)
28518102 4294{
f791bf7f
EGE
4295 GLOBAL_STATE_CODE();
4296
3cdc69d3
HR
4297 return bdrv_reopen_queue_child(bs_queue, bs, options, NULL, 0, false,
4298 NULL, 0, keep_old_opts);
28518102
KW
4299}
4300
ab5b5228
AG
4301void bdrv_reopen_queue_free(BlockReopenQueue *bs_queue)
4302{
f791bf7f 4303 GLOBAL_STATE_CODE();
ab5b5228
AG
4304 if (bs_queue) {
4305 BlockReopenQueueEntry *bs_entry, *next;
4306 QTAILQ_FOREACH_SAFE(bs_entry, bs_queue, entry, next) {
4307 qobject_unref(bs_entry->state.explicit_options);
4308 qobject_unref(bs_entry->state.options);
4309 g_free(bs_entry);
4310 }
4311 g_free(bs_queue);
4312 }
4313}
4314
e971aa12
JC
4315/*
4316 * Reopen multiple BlockDriverStates atomically & transactionally.
4317 *
4318 * The queue passed in (bs_queue) must have been built up previous
4319 * via bdrv_reopen_queue().
4320 *
4321 * Reopens all BDS specified in the queue, with the appropriate
4322 * flags. All devices are prepared for reopen, and failure of any
50d6a8a3 4323 * device will cause all device changes to be abandoned, and intermediate
e971aa12
JC
4324 * data cleaned up.
4325 *
4326 * If all devices prepare successfully, then the changes are committed
4327 * to all devices.
4328 *
1a63a907
KW
4329 * All affected nodes must be drained between bdrv_reopen_queue() and
4330 * bdrv_reopen_multiple().
6cf42ca2
KW
4331 *
4332 * To be called from the main thread, with all other AioContexts unlocked.
e971aa12 4333 */
5019aece 4334int bdrv_reopen_multiple(BlockReopenQueue *bs_queue, Error **errp)
e971aa12
JC
4335{
4336 int ret = -1;
4337 BlockReopenQueueEntry *bs_entry, *next;
6cf42ca2 4338 AioContext *ctx;
72373e40
VSO
4339 Transaction *tran = tran_new();
4340 g_autoptr(GHashTable) found = NULL;
4341 g_autoptr(GSList) refresh_list = NULL;
e971aa12 4342
6cf42ca2 4343 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
e971aa12 4344 assert(bs_queue != NULL);
da359909 4345 GLOBAL_STATE_CODE();
e971aa12 4346
a2aabf88 4347 QTAILQ_FOREACH(bs_entry, bs_queue, entry) {
6cf42ca2
KW
4348 ctx = bdrv_get_aio_context(bs_entry->state.bs);
4349 aio_context_acquire(ctx);
a2aabf88 4350 ret = bdrv_flush(bs_entry->state.bs);
6cf42ca2 4351 aio_context_release(ctx);
a2aabf88
VSO
4352 if (ret < 0) {
4353 error_setg_errno(errp, -ret, "Error flushing drive");
e3fc91aa 4354 goto abort;
a2aabf88
VSO
4355 }
4356 }
4357
859443b0 4358 QTAILQ_FOREACH(bs_entry, bs_queue, entry) {
1a63a907 4359 assert(bs_entry->state.bs->quiesce_counter > 0);
6cf42ca2
KW
4360 ctx = bdrv_get_aio_context(bs_entry->state.bs);
4361 aio_context_acquire(ctx);
72373e40 4362 ret = bdrv_reopen_prepare(&bs_entry->state, bs_queue, tran, errp);
6cf42ca2 4363 aio_context_release(ctx);
72373e40
VSO
4364 if (ret < 0) {
4365 goto abort;
e971aa12
JC
4366 }
4367 bs_entry->prepared = true;
4368 }
4369
72373e40 4370 found = g_hash_table_new(NULL, NULL);
859443b0 4371 QTAILQ_FOREACH(bs_entry, bs_queue, entry) {
69b736e7 4372 BDRVReopenState *state = &bs_entry->state;
72373e40
VSO
4373
4374 refresh_list = bdrv_topological_dfs(refresh_list, found, state->bs);
4375 if (state->old_backing_bs) {
4376 refresh_list = bdrv_topological_dfs(refresh_list, found,
4377 state->old_backing_bs);
cb828c31 4378 }
ecd30d2d
AG
4379 if (state->old_file_bs) {
4380 refresh_list = bdrv_topological_dfs(refresh_list, found,
4381 state->old_file_bs);
4382 }
72373e40
VSO
4383 }
4384
4385 /*
4386 * Note that file-posix driver rely on permission update done during reopen
4387 * (even if no permission changed), because it wants "new" permissions for
4388 * reconfiguring the fd and that's why it does it in raw_check_perm(), not
4389 * in raw_reopen_prepare() which is called with "old" permissions.
4390 */
4391 ret = bdrv_list_refresh_perms(refresh_list, bs_queue, tran, errp);
4392 if (ret < 0) {
4393 goto abort;
69b736e7
KW
4394 }
4395
fcd6a4f4
VSO
4396 /*
4397 * If we reach this point, we have success and just need to apply the
4398 * changes.
4399 *
4400 * Reverse order is used to comfort qcow2 driver: on commit it need to write
4401 * IN_USE flag to the image, to mark bitmaps in the image as invalid. But
4402 * children are usually goes after parents in reopen-queue, so go from last
4403 * to first element.
e971aa12 4404 */
fcd6a4f4 4405 QTAILQ_FOREACH_REVERSE(bs_entry, bs_queue, entry) {
6cf42ca2
KW
4406 ctx = bdrv_get_aio_context(bs_entry->state.bs);
4407 aio_context_acquire(ctx);
e971aa12 4408 bdrv_reopen_commit(&bs_entry->state);
6cf42ca2 4409 aio_context_release(ctx);
e971aa12
JC
4410 }
4411
72373e40 4412 tran_commit(tran);
69b736e7 4413
72373e40
VSO
4414 QTAILQ_FOREACH_REVERSE(bs_entry, bs_queue, entry) {
4415 BlockDriverState *bs = bs_entry->state.bs;
74ad9a3b 4416
72373e40 4417 if (bs->drv->bdrv_reopen_commit_post) {
6cf42ca2
KW
4418 ctx = bdrv_get_aio_context(bs);
4419 aio_context_acquire(ctx);
72373e40 4420 bs->drv->bdrv_reopen_commit_post(&bs_entry->state);
6cf42ca2 4421 aio_context_release(ctx);
69b736e7
KW
4422 }
4423 }
17e1e2be 4424
72373e40
VSO
4425 ret = 0;
4426 goto cleanup;
17e1e2be 4427
72373e40
VSO
4428abort:
4429 tran_abort(tran);
4430 QTAILQ_FOREACH_SAFE(bs_entry, bs_queue, entry, next) {
4431 if (bs_entry->prepared) {
6cf42ca2
KW
4432 ctx = bdrv_get_aio_context(bs_entry->state.bs);
4433 aio_context_acquire(ctx);
72373e40 4434 bdrv_reopen_abort(&bs_entry->state);
6cf42ca2 4435 aio_context_release(ctx);
17e1e2be
PK
4436 }
4437 }
72373e40 4438
e971aa12 4439cleanup:
ab5b5228 4440 bdrv_reopen_queue_free(bs_queue);
40840e41 4441
e971aa12
JC
4442 return ret;
4443}
4444
6cf42ca2
KW
4445int bdrv_reopen(BlockDriverState *bs, QDict *opts, bool keep_old_opts,
4446 Error **errp)
6e1000a8 4447{
6cf42ca2 4448 AioContext *ctx = bdrv_get_aio_context(bs);
6e1000a8 4449 BlockReopenQueue *queue;
6cf42ca2 4450 int ret;
6e1000a8 4451
f791bf7f
EGE
4452 GLOBAL_STATE_CODE();
4453
6e1000a8 4454 bdrv_subtree_drained_begin(bs);
6cf42ca2
KW
4455 if (ctx != qemu_get_aio_context()) {
4456 aio_context_release(ctx);
4457 }
4458
4459 queue = bdrv_reopen_queue(NULL, bs, opts, keep_old_opts);
5019aece 4460 ret = bdrv_reopen_multiple(queue, errp);
6cf42ca2
KW
4461
4462 if (ctx != qemu_get_aio_context()) {
4463 aio_context_acquire(ctx);
4464 }
6e1000a8
AG
4465 bdrv_subtree_drained_end(bs);
4466
4467 return ret;
4468}
4469
6cf42ca2
KW
4470int bdrv_reopen_set_read_only(BlockDriverState *bs, bool read_only,
4471 Error **errp)
4472{
4473 QDict *opts = qdict_new();
4474
f791bf7f
EGE
4475 GLOBAL_STATE_CODE();
4476
6cf42ca2
KW
4477 qdict_put_bool(opts, BDRV_OPT_READ_ONLY, read_only);
4478
4479 return bdrv_reopen(bs, opts, true, errp);
4480}
4481
cb828c31
AG
4482/*
4483 * Take a BDRVReopenState and check if the value of 'backing' in the
4484 * reopen_state->options QDict is valid or not.
4485 *
4486 * If 'backing' is missing from the QDict then return 0.
4487 *
4488 * If 'backing' contains the node name of the backing file of
4489 * reopen_state->bs then return 0.
4490 *
4491 * If 'backing' contains a different node name (or is null) then check
4492 * whether the current backing file can be replaced with the new one.
4493 * If that's the case then reopen_state->replace_backing_bs is set to
4494 * true and reopen_state->new_backing_bs contains a pointer to the new
4495 * backing BlockDriverState (or NULL).
4496 *
4497 * Return 0 on success, otherwise return < 0 and set @errp.
4498 */
ecd30d2d
AG
4499static int bdrv_reopen_parse_file_or_backing(BDRVReopenState *reopen_state,
4500 bool is_backing, Transaction *tran,
4501 Error **errp)
cb828c31
AG
4502{
4503 BlockDriverState *bs = reopen_state->bs;
ecd30d2d
AG
4504 BlockDriverState *new_child_bs;
4505 BlockDriverState *old_child_bs = is_backing ? child_bs(bs->backing) :
4506 child_bs(bs->file);
4507 const char *child_name = is_backing ? "backing" : "file";
cb828c31
AG
4508 QObject *value;
4509 const char *str;
4510
bdb73476
EGE
4511 GLOBAL_STATE_CODE();
4512
ecd30d2d 4513 value = qdict_get(reopen_state->options, child_name);
cb828c31
AG
4514 if (value == NULL) {
4515 return 0;
4516 }
4517
4518 switch (qobject_type(value)) {
4519 case QTYPE_QNULL:
ecd30d2d
AG
4520 assert(is_backing); /* The 'file' option does not allow a null value */
4521 new_child_bs = NULL;
cb828c31
AG
4522 break;
4523 case QTYPE_QSTRING:
410f44f5 4524 str = qstring_get_str(qobject_to(QString, value));
ecd30d2d
AG
4525 new_child_bs = bdrv_lookup_bs(NULL, str, errp);
4526 if (new_child_bs == NULL) {
cb828c31 4527 return -EINVAL;
ecd30d2d
AG
4528 } else if (bdrv_recurse_has_child(new_child_bs, bs)) {
4529 error_setg(errp, "Making '%s' a %s child of '%s' would create a "
4530 "cycle", str, child_name, bs->node_name);
cb828c31
AG
4531 return -EINVAL;
4532 }
4533 break;
4534 default:
ecd30d2d
AG
4535 /*
4536 * The options QDict has been flattened, so 'backing' and 'file'
4537 * do not allow any other data type here.
4538 */
cb828c31
AG
4539 g_assert_not_reached();
4540 }
4541
ecd30d2d
AG
4542 if (old_child_bs == new_child_bs) {
4543 return 0;
4544 }
4545
4546 if (old_child_bs) {
4547 if (bdrv_skip_implicit_filters(old_child_bs) == new_child_bs) {
cbfdb98c
VSO
4548 return 0;
4549 }
4550
ecd30d2d
AG
4551 if (old_child_bs->implicit) {
4552 error_setg(errp, "Cannot replace implicit %s child of %s",
4553 child_name, bs->node_name);
cbfdb98c
VSO
4554 return -EPERM;
4555 }
4556 }
4557
ecd30d2d 4558 if (bs->drv->is_filter && !old_child_bs) {
25f78d9e
VSO
4559 /*
4560 * Filters always have a file or a backing child, so we are trying to
4561 * change wrong child
4562 */
4563 error_setg(errp, "'%s' is a %s filter node that does not support a "
ecd30d2d 4564 "%s child", bs->node_name, bs->drv->format_name, child_name);
1d42f48c
HR
4565 return -EINVAL;
4566 }
4567
ecd30d2d
AG
4568 if (is_backing) {
4569 reopen_state->old_backing_bs = old_child_bs;
4570 } else {
4571 reopen_state->old_file_bs = old_child_bs;
4572 }
4573
4574 return bdrv_set_file_or_backing_noperm(bs, new_child_bs, is_backing,
4575 tran, errp);
cb828c31
AG
4576}
4577
e971aa12
JC
4578/*
4579 * Prepares a BlockDriverState for reopen. All changes are staged in the
4580 * 'opaque' field of the BDRVReopenState, which is used and allocated by
4581 * the block driver layer .bdrv_reopen_prepare()
4582 *
4583 * bs is the BlockDriverState to reopen
4584 * flags are the new open flags
4585 * queue is the reopen queue
4586 *
4587 * Returns 0 on success, non-zero on error. On error errp will be set
4588 * as well.
4589 *
4590 * On failure, bdrv_reopen_abort() will be called to clean up any data.
4591 * It is the responsibility of the caller to then call the abort() or
4592 * commit() for any other BDS that have been left in a prepare() state
4593 *
4594 */
53e96d1e 4595static int bdrv_reopen_prepare(BDRVReopenState *reopen_state,
72373e40 4596 BlockReopenQueue *queue,
ecd30d2d 4597 Transaction *change_child_tran, Error **errp)
e971aa12
JC
4598{
4599 int ret = -1;
e6d79c41 4600 int old_flags;
e971aa12
JC
4601 Error *local_err = NULL;
4602 BlockDriver *drv;
ccf9dc07 4603 QemuOpts *opts;
4c8350fe 4604 QDict *orig_reopen_opts;
593b3071 4605 char *discard = NULL;
3d8ce171 4606 bool read_only;
9ad08c44 4607 bool drv_prepared = false;
e971aa12
JC
4608
4609 assert(reopen_state != NULL);
4610 assert(reopen_state->bs->drv != NULL);
da359909 4611 GLOBAL_STATE_CODE();
e971aa12
JC
4612 drv = reopen_state->bs->drv;
4613
4c8350fe
AG
4614 /* This function and each driver's bdrv_reopen_prepare() remove
4615 * entries from reopen_state->options as they are processed, so
4616 * we need to make a copy of the original QDict. */
4617 orig_reopen_opts = qdict_clone_shallow(reopen_state->options);
4618
ccf9dc07
KW
4619 /* Process generic block layer options */
4620 opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
af175e85 4621 if (!qemu_opts_absorb_qdict(opts, reopen_state->options, errp)) {
ccf9dc07
KW
4622 ret = -EINVAL;
4623 goto error;
4624 }
4625
e6d79c41
AG
4626 /* This was already called in bdrv_reopen_queue_child() so the flags
4627 * are up-to-date. This time we simply want to remove the options from
4628 * QemuOpts in order to indicate that they have been processed. */
4629 old_flags = reopen_state->flags;
91a097e7 4630 update_flags_from_options(&reopen_state->flags, opts);
e6d79c41 4631 assert(old_flags == reopen_state->flags);
91a097e7 4632
415bbca8 4633 discard = qemu_opt_get_del(opts, BDRV_OPT_DISCARD);
593b3071
AG
4634 if (discard != NULL) {
4635 if (bdrv_parse_discard_flags(discard, &reopen_state->flags) != 0) {
4636 error_setg(errp, "Invalid discard option");
4637 ret = -EINVAL;
4638 goto error;
4639 }
4640 }
4641
543770bd
AG
4642 reopen_state->detect_zeroes =
4643 bdrv_parse_detect_zeroes(opts, reopen_state->flags, &local_err);
4644 if (local_err) {
4645 error_propagate(errp, local_err);
4646 ret = -EINVAL;
4647 goto error;
4648 }
4649
57f9db9a
AG
4650 /* All other options (including node-name and driver) must be unchanged.
4651 * Put them back into the QDict, so that they are checked at the end
4652 * of this function. */
4653 qemu_opts_to_qdict(opts, reopen_state->options);
ccf9dc07 4654
3d8ce171
JC
4655 /* If we are to stay read-only, do not allow permission change
4656 * to r/w. Attempting to set to r/w may fail if either BDRV_O_ALLOW_RDWR is
4657 * not set, or if the BDS still has copy_on_read enabled */
4658 read_only = !(reopen_state->flags & BDRV_O_RDWR);
54a32bfe 4659 ret = bdrv_can_set_read_only(reopen_state->bs, read_only, true, &local_err);
3d8ce171
JC
4660 if (local_err) {
4661 error_propagate(errp, local_err);
e971aa12
JC
4662 goto error;
4663 }
4664
e971aa12 4665 if (drv->bdrv_reopen_prepare) {
faf116b4
AG
4666 /*
4667 * If a driver-specific option is missing, it means that we
4668 * should reset it to its default value.
4669 * But not all options allow that, so we need to check it first.
4670 */
4671 ret = bdrv_reset_options_allowed(reopen_state->bs,
4672 reopen_state->options, errp);
4673 if (ret) {
4674 goto error;
4675 }
4676
e971aa12
JC
4677 ret = drv->bdrv_reopen_prepare(reopen_state, queue, &local_err);
4678 if (ret) {
4679 if (local_err != NULL) {
4680 error_propagate(errp, local_err);
4681 } else {
f30c66ba 4682 bdrv_refresh_filename(reopen_state->bs);
d8b6895f
LC
4683 error_setg(errp, "failed while preparing to reopen image '%s'",
4684 reopen_state->bs->filename);
e971aa12
JC
4685 }
4686 goto error;
4687 }
4688 } else {
4689 /* It is currently mandatory to have a bdrv_reopen_prepare()
4690 * handler for each supported drv. */
81e5f78a
AG
4691 error_setg(errp, "Block format '%s' used by node '%s' "
4692 "does not support reopening files", drv->format_name,
4693 bdrv_get_device_or_node_name(reopen_state->bs));
e971aa12
JC
4694 ret = -1;
4695 goto error;
4696 }
4697
9ad08c44
HR
4698 drv_prepared = true;
4699
bacd9b87
AG
4700 /*
4701 * We must provide the 'backing' option if the BDS has a backing
4702 * file or if the image file has a backing file name as part of
4703 * its metadata. Otherwise the 'backing' option can be omitted.
4704 */
4705 if (drv->supports_backing && reopen_state->backing_missing &&
1d42f48c 4706 (reopen_state->bs->backing || reopen_state->bs->backing_file[0])) {
8546632e
AG
4707 error_setg(errp, "backing is missing for '%s'",
4708 reopen_state->bs->node_name);
4709 ret = -EINVAL;
4710 goto error;
4711 }
4712
cb828c31
AG
4713 /*
4714 * Allow changing the 'backing' option. The new value can be
4715 * either a reference to an existing node (using its node name)
4716 * or NULL to simply detach the current backing file.
4717 */
ecd30d2d
AG
4718 ret = bdrv_reopen_parse_file_or_backing(reopen_state, true,
4719 change_child_tran, errp);
cb828c31
AG
4720 if (ret < 0) {
4721 goto error;
4722 }
4723 qdict_del(reopen_state->options, "backing");
4724
ecd30d2d
AG
4725 /* Allow changing the 'file' option. In this case NULL is not allowed */
4726 ret = bdrv_reopen_parse_file_or_backing(reopen_state, false,
4727 change_child_tran, errp);
4728 if (ret < 0) {
4729 goto error;
4730 }
4731 qdict_del(reopen_state->options, "file");
4732
4d2cb092
KW
4733 /* Options that are not handled are only okay if they are unchanged
4734 * compared to the old state. It is expected that some options are only
4735 * used for the initial open, but not reopen (e.g. filename) */
4736 if (qdict_size(reopen_state->options)) {
4737 const QDictEntry *entry = qdict_first(reopen_state->options);
4738
4739 do {
54fd1b0d
HR
4740 QObject *new = entry->value;
4741 QObject *old = qdict_get(reopen_state->bs->options, entry->key);
4742
db905283
AG
4743 /* Allow child references (child_name=node_name) as long as they
4744 * point to the current child (i.e. everything stays the same). */
4745 if (qobject_type(new) == QTYPE_QSTRING) {
4746 BdrvChild *child;
4747 QLIST_FOREACH(child, &reopen_state->bs->children, next) {
4748 if (!strcmp(child->name, entry->key)) {
4749 break;
4750 }
4751 }
4752
4753 if (child) {
410f44f5
MA
4754 if (!strcmp(child->bs->node_name,
4755 qstring_get_str(qobject_to(QString, new)))) {
db905283
AG
4756 continue; /* Found child with this name, skip option */
4757 }
4758 }
4759 }
4760
129c7d1c 4761 /*
54fd1b0d
HR
4762 * TODO: When using -drive to specify blockdev options, all values
4763 * will be strings; however, when using -blockdev, blockdev-add or
4764 * filenames using the json:{} pseudo-protocol, they will be
4765 * correctly typed.
4766 * In contrast, reopening options are (currently) always strings
4767 * (because you can only specify them through qemu-io; all other
4768 * callers do not specify any options).
4769 * Therefore, when using anything other than -drive to create a BDS,
4770 * this cannot detect non-string options as unchanged, because
4771 * qobject_is_equal() always returns false for objects of different
4772 * type. In the future, this should be remedied by correctly typing
4773 * all options. For now, this is not too big of an issue because
4774 * the user can simply omit options which cannot be changed anyway,
4775 * so they will stay unchanged.
129c7d1c 4776 */
54fd1b0d 4777 if (!qobject_is_equal(new, old)) {
4d2cb092
KW
4778 error_setg(errp, "Cannot change the option '%s'", entry->key);
4779 ret = -EINVAL;
4780 goto error;
4781 }
4782 } while ((entry = qdict_next(reopen_state->options, entry)));
4783 }
4784
e971aa12
JC
4785 ret = 0;
4786
4c8350fe
AG
4787 /* Restore the original reopen_state->options QDict */
4788 qobject_unref(reopen_state->options);
4789 reopen_state->options = qobject_ref(orig_reopen_opts);
4790
e971aa12 4791error:
9ad08c44
HR
4792 if (ret < 0 && drv_prepared) {
4793 /* drv->bdrv_reopen_prepare() has succeeded, so we need to
4794 * call drv->bdrv_reopen_abort() before signaling an error
4795 * (bdrv_reopen_multiple() will not call bdrv_reopen_abort()
4796 * when the respective bdrv_reopen_prepare() has failed) */
4797 if (drv->bdrv_reopen_abort) {
4798 drv->bdrv_reopen_abort(reopen_state);
4799 }
4800 }
ccf9dc07 4801 qemu_opts_del(opts);
4c8350fe 4802 qobject_unref(orig_reopen_opts);
593b3071 4803 g_free(discard);
e971aa12
JC
4804 return ret;
4805}
4806
4807/*
4808 * Takes the staged changes for the reopen from bdrv_reopen_prepare(), and
4809 * makes them final by swapping the staging BlockDriverState contents into
4810 * the active BlockDriverState contents.
4811 */
53e96d1e 4812static void bdrv_reopen_commit(BDRVReopenState *reopen_state)
e971aa12
JC
4813{
4814 BlockDriver *drv;
50bf65ba 4815 BlockDriverState *bs;
50196d7a 4816 BdrvChild *child;
e971aa12
JC
4817
4818 assert(reopen_state != NULL);
50bf65ba
VSO
4819 bs = reopen_state->bs;
4820 drv = bs->drv;
e971aa12 4821 assert(drv != NULL);
da359909 4822 GLOBAL_STATE_CODE();
e971aa12
JC
4823
4824 /* If there are any driver level actions to take */
4825 if (drv->bdrv_reopen_commit) {
4826 drv->bdrv_reopen_commit(reopen_state);
4827 }
4828
4829 /* set BDS specific flags now */
cb3e7f08 4830 qobject_unref(bs->explicit_options);
4c8350fe 4831 qobject_unref(bs->options);
ab5b5228
AG
4832 qobject_ref(reopen_state->explicit_options);
4833 qobject_ref(reopen_state->options);
145f598e 4834
50bf65ba 4835 bs->explicit_options = reopen_state->explicit_options;
4c8350fe 4836 bs->options = reopen_state->options;
50bf65ba 4837 bs->open_flags = reopen_state->flags;
543770bd 4838 bs->detect_zeroes = reopen_state->detect_zeroes;
355ef4ac 4839
50196d7a
AG
4840 /* Remove child references from bs->options and bs->explicit_options.
4841 * Child options were already removed in bdrv_reopen_queue_child() */
4842 QLIST_FOREACH(child, &bs->children, next) {
4843 qdict_del(bs->explicit_options, child->name);
4844 qdict_del(bs->options, child->name);
4845 }
3d0e8743
VSO
4846 /* backing is probably removed, so it's not handled by previous loop */
4847 qdict_del(bs->explicit_options, "backing");
4848 qdict_del(bs->options, "backing");
4849
1e4c797c 4850 bdrv_refresh_limits(bs, NULL, NULL);
e971aa12
JC
4851}
4852
4853/*
4854 * Abort the reopen, and delete and free the staged changes in
4855 * reopen_state
4856 */
53e96d1e 4857static void bdrv_reopen_abort(BDRVReopenState *reopen_state)
e971aa12
JC
4858{
4859 BlockDriver *drv;
4860
4861 assert(reopen_state != NULL);
4862 drv = reopen_state->bs->drv;
4863 assert(drv != NULL);
da359909 4864 GLOBAL_STATE_CODE();
e971aa12
JC
4865
4866 if (drv->bdrv_reopen_abort) {
4867 drv->bdrv_reopen_abort(reopen_state);
4868 }
4869}
4870
4871
64dff520 4872static void bdrv_close(BlockDriverState *bs)
fc01f7e7 4873{
33384421 4874 BdrvAioNotifier *ban, *ban_next;
50a3efb0 4875 BdrvChild *child, *next;
33384421 4876
f791bf7f 4877 GLOBAL_STATE_CODE();
30f55fb8 4878 assert(!bs->refcnt);
99b7e775 4879
fc27291d 4880 bdrv_drained_begin(bs); /* complete I/O */
58fda173 4881 bdrv_flush(bs);
53ec73e2 4882 bdrv_drain(bs); /* in case flush left pending I/O */
fc27291d 4883
3cbc002c 4884 if (bs->drv) {
3c005293 4885 if (bs->drv->bdrv_close) {
7b99a266 4886 /* Must unfreeze all children, so bdrv_unref_child() works */
3c005293
VSO
4887 bs->drv->bdrv_close(bs);
4888 }
9a4f4c31 4889 bs->drv = NULL;
50a3efb0 4890 }
9a7dedbc 4891
50a3efb0 4892 QLIST_FOREACH_SAFE(child, &bs->children, next, next) {
dd4118c7 4893 bdrv_unref_child(bs, child);
b338082b 4894 }
98f90dba 4895
5bb04747
VSO
4896 assert(!bs->backing);
4897 assert(!bs->file);
50a3efb0
AG
4898 g_free(bs->opaque);
4899 bs->opaque = NULL;
d73415a3 4900 qatomic_set(&bs->copy_on_read, 0);
50a3efb0
AG
4901 bs->backing_file[0] = '\0';
4902 bs->backing_format[0] = '\0';
4903 bs->total_sectors = 0;
4904 bs->encrypted = false;
4905 bs->sg = false;
cb3e7f08
MAL
4906 qobject_unref(bs->options);
4907 qobject_unref(bs->explicit_options);
50a3efb0
AG
4908 bs->options = NULL;
4909 bs->explicit_options = NULL;
cb3e7f08 4910 qobject_unref(bs->full_open_options);
50a3efb0 4911 bs->full_open_options = NULL;
0bc329fb
HR
4912 g_free(bs->block_status_cache);
4913 bs->block_status_cache = NULL;
50a3efb0 4914
cca43ae1
VSO
4915 bdrv_release_named_dirty_bitmaps(bs);
4916 assert(QLIST_EMPTY(&bs->dirty_bitmaps));
4917
33384421
HR
4918 QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_next) {
4919 g_free(ban);
4920 }
4921 QLIST_INIT(&bs->aio_notifiers);
fc27291d 4922 bdrv_drained_end(bs);
1a6d3bd2
GK
4923
4924 /*
4925 * If we're still inside some bdrv_drain_all_begin()/end() sections, end
4926 * them now since this BDS won't exist anymore when bdrv_drain_all_end()
4927 * gets called.
4928 */
4929 if (bs->quiesce_counter) {
4930 bdrv_drain_all_end_quiesce(bs);
4931 }
b338082b
FB
4932}
4933
2bc93fed
MK
4934void bdrv_close_all(void)
4935{
f791bf7f 4936 GLOBAL_STATE_CODE();
880eeec6 4937 assert(job_next(NULL) == NULL);
ca9bd24c
HR
4938
4939 /* Drop references from requests still in flight, such as canceled block
4940 * jobs whose AIO context has not been polled yet */
4941 bdrv_drain_all();
2bc93fed 4942
ca9bd24c
HR
4943 blk_remove_all_bs();
4944 blockdev_close_all_bdrv_states();
ed78cda3 4945
a1a2af07 4946 assert(QTAILQ_EMPTY(&all_bdrv_states));
2bc93fed
MK
4947}
4948
d0ac0380
KW
4949static bool should_update_child(BdrvChild *c, BlockDriverState *to)
4950{
2f30b7c3
VSO
4951 GQueue *queue;
4952 GHashTable *found;
4953 bool ret;
d0ac0380 4954
bd86fb99 4955 if (c->klass->stay_at_node) {
d0ac0380
KW
4956 return false;
4957 }
4958
ec9f10fe
HR
4959 /* If the child @c belongs to the BDS @to, replacing the current
4960 * c->bs by @to would mean to create a loop.
4961 *
4962 * Such a case occurs when appending a BDS to a backing chain.
4963 * For instance, imagine the following chain:
4964 *
4965 * guest device -> node A -> further backing chain...
4966 *
4967 * Now we create a new BDS B which we want to put on top of this
4968 * chain, so we first attach A as its backing node:
4969 *
4970 * node B
4971 * |
4972 * v
4973 * guest device -> node A -> further backing chain...
4974 *
4975 * Finally we want to replace A by B. When doing that, we want to
4976 * replace all pointers to A by pointers to B -- except for the
4977 * pointer from B because (1) that would create a loop, and (2)
4978 * that pointer should simply stay intact:
4979 *
4980 * guest device -> node B
4981 * |
4982 * v
4983 * node A -> further backing chain...
4984 *
4985 * In general, when replacing a node A (c->bs) by a node B (@to),
4986 * if A is a child of B, that means we cannot replace A by B there
4987 * because that would create a loop. Silently detaching A from B
4988 * is also not really an option. So overall just leaving A in
2f30b7c3
VSO
4989 * place there is the most sensible choice.
4990 *
4991 * We would also create a loop in any cases where @c is only
4992 * indirectly referenced by @to. Prevent this by returning false
4993 * if @c is found (by breadth-first search) anywhere in the whole
4994 * subtree of @to.
4995 */
4996
4997 ret = true;
4998 found = g_hash_table_new(NULL, NULL);
4999 g_hash_table_add(found, to);
5000 queue = g_queue_new();
5001 g_queue_push_tail(queue, to);
5002
5003 while (!g_queue_is_empty(queue)) {
5004 BlockDriverState *v = g_queue_pop_head(queue);
5005 BdrvChild *c2;
5006
5007 QLIST_FOREACH(c2, &v->children, next) {
5008 if (c2 == c) {
5009 ret = false;
5010 break;
5011 }
5012
5013 if (g_hash_table_contains(found, c2->bs)) {
5014 continue;
5015 }
5016
5017 g_queue_push_tail(queue, c2->bs);
5018 g_hash_table_add(found, c2->bs);
d0ac0380
KW
5019 }
5020 }
5021
2f30b7c3
VSO
5022 g_queue_free(queue);
5023 g_hash_table_destroy(found);
5024
5025 return ret;
d0ac0380
KW
5026}
5027
57f08941 5028static void bdrv_remove_child_commit(void *opaque)
46541ee5 5029{
bdb73476 5030 GLOBAL_STATE_CODE();
5bb04747 5031 bdrv_child_free(opaque);
46541ee5
VSO
5032}
5033
57f08941
VSO
5034static TransactionActionDrv bdrv_remove_child_drv = {
5035 .commit = bdrv_remove_child_commit,
46541ee5
VSO
5036};
5037
57f08941
VSO
5038/* Function doesn't update permissions, caller is responsible for this. */
5039static void bdrv_remove_child(BdrvChild *child, Transaction *tran)
46541ee5 5040{
46541ee5
VSO
5041 if (!child) {
5042 return;
5043 }
5044
5045 if (child->bs) {
a2c37a30 5046 bdrv_replace_child_tran(child, NULL, tran);
46541ee5
VSO
5047 }
5048
57f08941 5049 tran_add(tran, &bdrv_remove_child_drv, child);
46541ee5
VSO
5050}
5051
5b995019
VSO
5052/*
5053 * A function to remove backing-chain child of @bs if exists: cow child for
5054 * format nodes (always .backing) and filter child for filters (may be .file or
5055 * .backing)
5056 */
5057static void bdrv_remove_filter_or_cow_child(BlockDriverState *bs,
5058 Transaction *tran)
5059{
57f08941 5060 bdrv_remove_child(bdrv_filter_or_cow_child(bs), tran);
5b995019
VSO
5061}
5062
117caba9
VSO
5063static int bdrv_replace_node_noperm(BlockDriverState *from,
5064 BlockDriverState *to,
5065 bool auto_skip, Transaction *tran,
5066 Error **errp)
5067{
5068 BdrvChild *c, *next;
5069
bdb73476 5070 GLOBAL_STATE_CODE();
82b54cf5 5071
117caba9
VSO
5072 QLIST_FOREACH_SAFE(c, &from->parents, next_parent, next) {
5073 assert(c->bs == from);
5074 if (!should_update_child(c, to)) {
5075 if (auto_skip) {
5076 continue;
5077 }
5078 error_setg(errp, "Should not change '%s' link to '%s'",
5079 c->name, from->node_name);
5080 return -EINVAL;
5081 }
5082 if (c->frozen) {
5083 error_setg(errp, "Cannot change '%s' link to '%s'",
5084 c->name, from->node_name);
5085 return -EPERM;
5086 }
0f0b1e29 5087 bdrv_replace_child_tran(c, to, tran);
117caba9
VSO
5088 }
5089
5090 return 0;
5091}
5092
313274bb
VSO
5093/*
5094 * With auto_skip=true bdrv_replace_node_common skips updating from parents
5095 * if it creates a parent-child relation loop or if parent is block-job.
5096 *
5097 * With auto_skip=false the error is returned if from has a parent which should
5098 * not be updated.
3108a15c
VSO
5099 *
5100 * With @detach_subchain=true @to must be in a backing chain of @from. In this
5101 * case backing link of the cow-parent of @to is removed.
313274bb 5102 */
a1e708fc
VSO
5103static int bdrv_replace_node_common(BlockDriverState *from,
5104 BlockDriverState *to,
3108a15c
VSO
5105 bool auto_skip, bool detach_subchain,
5106 Error **errp)
dd62f1ca 5107{
3bb0e298
VSO
5108 Transaction *tran = tran_new();
5109 g_autoptr(GHashTable) found = NULL;
5110 g_autoptr(GSList) refresh_list = NULL;
2d369d6e 5111 BlockDriverState *to_cow_parent = NULL;
234ac1a9
KW
5112 int ret;
5113
bdb73476 5114 GLOBAL_STATE_CODE();
82b54cf5 5115
3108a15c
VSO
5116 if (detach_subchain) {
5117 assert(bdrv_chain_contains(from, to));
5118 assert(from != to);
5119 for (to_cow_parent = from;
5120 bdrv_filter_or_cow_bs(to_cow_parent) != to;
5121 to_cow_parent = bdrv_filter_or_cow_bs(to_cow_parent))
5122 {
5123 ;
5124 }
5125 }
5126
234ac1a9
KW
5127 /* Make sure that @from doesn't go away until we have successfully attached
5128 * all of its parents to @to. */
5129 bdrv_ref(from);
dd62f1ca 5130
f871abd6 5131 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
30dd65f3 5132 assert(bdrv_get_aio_context(from) == bdrv_get_aio_context(to));
f871abd6
KW
5133 bdrv_drained_begin(from);
5134
3bb0e298
VSO
5135 /*
5136 * Do the replacement without permission update.
5137 * Replacement may influence the permissions, we should calculate new
5138 * permissions based on new graph. If we fail, we'll roll-back the
5139 * replacement.
5140 */
117caba9
VSO
5141 ret = bdrv_replace_node_noperm(from, to, auto_skip, tran, errp);
5142 if (ret < 0) {
5143 goto out;
234ac1a9
KW
5144 }
5145
3108a15c
VSO
5146 if (detach_subchain) {
5147 bdrv_remove_filter_or_cow_child(to_cow_parent, tran);
5148 }
5149
3bb0e298 5150 found = g_hash_table_new(NULL, NULL);
234ac1a9 5151
3bb0e298
VSO
5152 refresh_list = bdrv_topological_dfs(refresh_list, found, to);
5153 refresh_list = bdrv_topological_dfs(refresh_list, found, from);
9bd910e2 5154
3bb0e298
VSO
5155 ret = bdrv_list_refresh_perms(refresh_list, NULL, tran, errp);
5156 if (ret < 0) {
5157 goto out;
dd62f1ca 5158 }
234ac1a9 5159
a1e708fc
VSO
5160 ret = 0;
5161
234ac1a9 5162out:
3bb0e298
VSO
5163 tran_finalize(tran, ret);
5164
f871abd6 5165 bdrv_drained_end(from);
234ac1a9 5166 bdrv_unref(from);
a1e708fc
VSO
5167
5168 return ret;
dd62f1ca
KW
5169}
5170
a1e708fc
VSO
5171int bdrv_replace_node(BlockDriverState *from, BlockDriverState *to,
5172 Error **errp)
313274bb 5173{
f791bf7f
EGE
5174 GLOBAL_STATE_CODE();
5175
3108a15c
VSO
5176 return bdrv_replace_node_common(from, to, true, false, errp);
5177}
5178
5179int bdrv_drop_filter(BlockDriverState *bs, Error **errp)
5180{
f791bf7f
EGE
5181 GLOBAL_STATE_CODE();
5182
3108a15c
VSO
5183 return bdrv_replace_node_common(bs, bdrv_filter_or_cow_bs(bs), true, true,
5184 errp);
313274bb
VSO
5185}
5186
4ddc07ca
PB
5187/*
5188 * Add new bs contents at the top of an image chain while the chain is
5189 * live, while keeping required fields on the top layer.
5190 *
5191 * This will modify the BlockDriverState fields, and swap contents
5192 * between bs_new and bs_top. Both bs_new and bs_top are modified.
5193 *
2272edcf
VSO
5194 * bs_new must not be attached to a BlockBackend and must not have backing
5195 * child.
4ddc07ca
PB
5196 *
5197 * This function does not create any image files.
5198 */
a1e708fc
VSO
5199int bdrv_append(BlockDriverState *bs_new, BlockDriverState *bs_top,
5200 Error **errp)
4ddc07ca 5201{
2272edcf 5202 int ret;
5bb04747 5203 BdrvChild *child;
2272edcf
VSO
5204 Transaction *tran = tran_new();
5205
f791bf7f
EGE
5206 GLOBAL_STATE_CODE();
5207
2272edcf
VSO
5208 assert(!bs_new->backing);
5209
5bb04747
VSO
5210 child = bdrv_attach_child_noperm(bs_new, bs_top, "backing",
5211 &child_of_bds, bdrv_backing_role(bs_new),
5212 tran, errp);
5213 if (!child) {
5214 ret = -EINVAL;
2272edcf 5215 goto out;
b2c2832c 5216 }
dd62f1ca 5217
2272edcf 5218 ret = bdrv_replace_node_noperm(bs_top, bs_new, true, tran, errp);
a1e708fc 5219 if (ret < 0) {
2272edcf 5220 goto out;
234ac1a9 5221 }
4ddc07ca 5222
2272edcf
VSO
5223 ret = bdrv_refresh_perms(bs_new, errp);
5224out:
5225 tran_finalize(tran, ret);
5226
1e4c797c 5227 bdrv_refresh_limits(bs_top, NULL, NULL);
2272edcf
VSO
5228
5229 return ret;
8802d1fd
JC
5230}
5231
bd8f4c42
VSO
5232/* Not for empty child */
5233int bdrv_replace_child_bs(BdrvChild *child, BlockDriverState *new_bs,
5234 Error **errp)
5235{
5236 int ret;
5237 Transaction *tran = tran_new();
5238 g_autoptr(GHashTable) found = NULL;
5239 g_autoptr(GSList) refresh_list = NULL;
5240 BlockDriverState *old_bs = child->bs;
5241
f791bf7f
EGE
5242 GLOBAL_STATE_CODE();
5243
bd8f4c42
VSO
5244 bdrv_ref(old_bs);
5245 bdrv_drained_begin(old_bs);
5246 bdrv_drained_begin(new_bs);
5247
0f0b1e29 5248 bdrv_replace_child_tran(child, new_bs, tran);
bd8f4c42
VSO
5249
5250 found = g_hash_table_new(NULL, NULL);
5251 refresh_list = bdrv_topological_dfs(refresh_list, found, old_bs);
5252 refresh_list = bdrv_topological_dfs(refresh_list, found, new_bs);
5253
5254 ret = bdrv_list_refresh_perms(refresh_list, NULL, tran, errp);
5255
5256 tran_finalize(tran, ret);
5257
5258 bdrv_drained_end(old_bs);
5259 bdrv_drained_end(new_bs);
5260 bdrv_unref(old_bs);
5261
5262 return ret;
5263}
5264
4f6fd349 5265static void bdrv_delete(BlockDriverState *bs)
b338082b 5266{
3718d8ab 5267 assert(bdrv_op_blocker_is_empty(bs));
4f6fd349 5268 assert(!bs->refcnt);
f791bf7f 5269 GLOBAL_STATE_CODE();
18846dee 5270
1b7bdbc1 5271 /* remove from list, if necessary */
63eaaae0
KW
5272 if (bs->node_name[0] != '\0') {
5273 QTAILQ_REMOVE(&graph_bdrv_states, bs, node_list);
5274 }
2c1d04e0
HR
5275 QTAILQ_REMOVE(&all_bdrv_states, bs, bs_list);
5276
30c321f9
AK
5277 bdrv_close(bs);
5278
7267c094 5279 g_free(bs);
fc01f7e7
FB
5280}
5281
96796fae
VSO
5282
5283/*
5284 * Replace @bs by newly created block node.
5285 *
5286 * @options is a QDict of options to pass to the block drivers, or NULL for an
5287 * empty set of options. The reference to the QDict belongs to the block layer
5288 * after the call (even on failure), so if the caller intends to reuse the
5289 * dictionary, it needs to use qobject_ref() before calling bdrv_open.
5290 */
5291BlockDriverState *bdrv_insert_node(BlockDriverState *bs, QDict *options,
8872ef78
AS
5292 int flags, Error **errp)
5293{
f053b7e8
VSO
5294 ERRP_GUARD();
5295 int ret;
b11c8739
VSO
5296 BlockDriverState *new_node_bs = NULL;
5297 const char *drvname, *node_name;
5298 BlockDriver *drv;
5299
5300 drvname = qdict_get_try_str(options, "driver");
5301 if (!drvname) {
5302 error_setg(errp, "driver is not specified");
5303 goto fail;
5304 }
5305
5306 drv = bdrv_find_format(drvname);
5307 if (!drv) {
5308 error_setg(errp, "Unknown driver: '%s'", drvname);
5309 goto fail;
5310 }
8872ef78 5311
b11c8739
VSO
5312 node_name = qdict_get_try_str(options, "node-name");
5313
f791bf7f
EGE
5314 GLOBAL_STATE_CODE();
5315
b11c8739
VSO
5316 new_node_bs = bdrv_new_open_driver_opts(drv, node_name, options, flags,
5317 errp);
5318 options = NULL; /* bdrv_new_open_driver() eats options */
5319 if (!new_node_bs) {
8872ef78 5320 error_prepend(errp, "Could not create node: ");
b11c8739 5321 goto fail;
8872ef78
AS
5322 }
5323
5324 bdrv_drained_begin(bs);
f053b7e8 5325 ret = bdrv_replace_node(bs, new_node_bs, errp);
8872ef78
AS
5326 bdrv_drained_end(bs);
5327
f053b7e8
VSO
5328 if (ret < 0) {
5329 error_prepend(errp, "Could not replace node: ");
b11c8739 5330 goto fail;
8872ef78
AS
5331 }
5332
5333 return new_node_bs;
b11c8739
VSO
5334
5335fail:
5336 qobject_unref(options);
5337 bdrv_unref(new_node_bs);
5338 return NULL;
8872ef78
AS
5339}
5340
e97fc193
AL
5341/*
5342 * Run consistency checks on an image
5343 *
e076f338 5344 * Returns 0 if the check could be completed (it doesn't mean that the image is
a1c7273b 5345 * free of errors) or -errno when an internal error occurred. The results of the
e076f338 5346 * check are stored in res.
e97fc193 5347 */
21c2283e
VSO
5348int coroutine_fn bdrv_co_check(BlockDriverState *bs,
5349 BdrvCheckResult *res, BdrvCheckMode fix)
e97fc193 5350{
1581a70d 5351 IO_CODE();
908bcd54
HR
5352 if (bs->drv == NULL) {
5353 return -ENOMEDIUM;
5354 }
2fd61638 5355 if (bs->drv->bdrv_co_check == NULL) {
e97fc193
AL
5356 return -ENOTSUP;
5357 }
5358
e076f338 5359 memset(res, 0, sizeof(*res));
2fd61638
PB
5360 return bs->drv->bdrv_co_check(bs, res, fix);
5361}
5362
756e6736
KW
5363/*
5364 * Return values:
5365 * 0 - success
5366 * -EINVAL - backing format specified, but no file
5367 * -ENOSPC - can't update the backing file because no space is left in the
5368 * image file header
5369 * -ENOTSUP - format driver doesn't support changing the backing file
5370 */
e54ee1b3 5371int bdrv_change_backing_file(BlockDriverState *bs, const char *backing_file,
497a30db 5372 const char *backing_fmt, bool require)
756e6736
KW
5373{
5374 BlockDriver *drv = bs->drv;
469ef350 5375 int ret;
756e6736 5376
f791bf7f
EGE
5377 GLOBAL_STATE_CODE();
5378
d470ad42
HR
5379 if (!drv) {
5380 return -ENOMEDIUM;
5381 }
5382
5f377794
PB
5383 /* Backing file format doesn't make sense without a backing file */
5384 if (backing_fmt && !backing_file) {
5385 return -EINVAL;
5386 }
5387
497a30db
EB
5388 if (require && backing_file && !backing_fmt) {
5389 return -EINVAL;
e54ee1b3
EB
5390 }
5391
756e6736 5392 if (drv->bdrv_change_backing_file != NULL) {
469ef350 5393 ret = drv->bdrv_change_backing_file(bs, backing_file, backing_fmt);
756e6736 5394 } else {
469ef350 5395 ret = -ENOTSUP;
756e6736 5396 }
469ef350
PB
5397
5398 if (ret == 0) {
5399 pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_file ?: "");
5400 pstrcpy(bs->backing_format, sizeof(bs->backing_format), backing_fmt ?: "");
998c2019
HR
5401 pstrcpy(bs->auto_backing_file, sizeof(bs->auto_backing_file),
5402 backing_file ?: "");
469ef350
PB
5403 }
5404 return ret;
756e6736
KW
5405}
5406
6ebdcee2 5407/*
dcf3f9b2
HR
5408 * Finds the first non-filter node above bs in the chain between
5409 * active and bs. The returned node is either an immediate parent of
5410 * bs, or there are only filter nodes between the two.
6ebdcee2
JC
5411 *
5412 * Returns NULL if bs is not found in active's image chain,
5413 * or if active == bs.
4caf0fcd
JC
5414 *
5415 * Returns the bottommost base image if bs == NULL.
6ebdcee2
JC
5416 */
5417BlockDriverState *bdrv_find_overlay(BlockDriverState *active,
5418 BlockDriverState *bs)
5419{
f791bf7f
EGE
5420
5421 GLOBAL_STATE_CODE();
5422
dcf3f9b2
HR
5423 bs = bdrv_skip_filters(bs);
5424 active = bdrv_skip_filters(active);
5425
5426 while (active) {
5427 BlockDriverState *next = bdrv_backing_chain_next(active);
5428 if (bs == next) {
5429 return active;
5430 }
5431 active = next;
6ebdcee2
JC
5432 }
5433
dcf3f9b2 5434 return NULL;
4caf0fcd 5435}
6ebdcee2 5436
4caf0fcd
JC
5437/* Given a BDS, searches for the base layer. */
5438BlockDriverState *bdrv_find_base(BlockDriverState *bs)
5439{
f791bf7f
EGE
5440 GLOBAL_STATE_CODE();
5441
4caf0fcd 5442 return bdrv_find_overlay(bs, NULL);
6ebdcee2
JC
5443}
5444
2cad1ebe 5445/*
7b99a266
HR
5446 * Return true if at least one of the COW (backing) and filter links
5447 * between @bs and @base is frozen. @errp is set if that's the case.
0f0998f6 5448 * @base must be reachable from @bs, or NULL.
2cad1ebe
AG
5449 */
5450bool bdrv_is_backing_chain_frozen(BlockDriverState *bs, BlockDriverState *base,
5451 Error **errp)
5452{
5453 BlockDriverState *i;
7b99a266 5454 BdrvChild *child;
2cad1ebe 5455
f791bf7f
EGE
5456 GLOBAL_STATE_CODE();
5457
7b99a266
HR
5458 for (i = bs; i != base; i = child_bs(child)) {
5459 child = bdrv_filter_or_cow_child(i);
5460
5461 if (child && child->frozen) {
2cad1ebe 5462 error_setg(errp, "Cannot change '%s' link from '%s' to '%s'",
7b99a266 5463 child->name, i->node_name, child->bs->node_name);
2cad1ebe
AG
5464 return true;
5465 }
5466 }
5467
5468 return false;
5469}
5470
5471/*
7b99a266 5472 * Freeze all COW (backing) and filter links between @bs and @base.
2cad1ebe
AG
5473 * If any of the links is already frozen the operation is aborted and
5474 * none of the links are modified.
0f0998f6 5475 * @base must be reachable from @bs, or NULL.
2cad1ebe
AG
5476 * Returns 0 on success. On failure returns < 0 and sets @errp.
5477 */
5478int bdrv_freeze_backing_chain(BlockDriverState *bs, BlockDriverState *base,
5479 Error **errp)
5480{
5481 BlockDriverState *i;
7b99a266 5482 BdrvChild *child;
2cad1ebe 5483
f791bf7f
EGE
5484 GLOBAL_STATE_CODE();
5485
2cad1ebe
AG
5486 if (bdrv_is_backing_chain_frozen(bs, base, errp)) {
5487 return -EPERM;
5488 }
5489
7b99a266
HR
5490 for (i = bs; i != base; i = child_bs(child)) {
5491 child = bdrv_filter_or_cow_child(i);
5492 if (child && child->bs->never_freeze) {
e5182c1c 5493 error_setg(errp, "Cannot freeze '%s' link to '%s'",
7b99a266 5494 child->name, child->bs->node_name);
e5182c1c
HR
5495 return -EPERM;
5496 }
5497 }
5498
7b99a266
HR
5499 for (i = bs; i != base; i = child_bs(child)) {
5500 child = bdrv_filter_or_cow_child(i);
5501 if (child) {
5502 child->frozen = true;
0f0998f6 5503 }
2cad1ebe
AG
5504 }
5505
5506 return 0;
5507}
5508
5509/*
7b99a266
HR
5510 * Unfreeze all COW (backing) and filter links between @bs and @base.
5511 * The caller must ensure that all links are frozen before using this
5512 * function.
0f0998f6 5513 * @base must be reachable from @bs, or NULL.
2cad1ebe
AG
5514 */
5515void bdrv_unfreeze_backing_chain(BlockDriverState *bs, BlockDriverState *base)
5516{
5517 BlockDriverState *i;
7b99a266 5518 BdrvChild *child;
2cad1ebe 5519
f791bf7f
EGE
5520 GLOBAL_STATE_CODE();
5521
7b99a266
HR
5522 for (i = bs; i != base; i = child_bs(child)) {
5523 child = bdrv_filter_or_cow_child(i);
5524 if (child) {
5525 assert(child->frozen);
5526 child->frozen = false;
0f0998f6 5527 }
2cad1ebe
AG
5528 }
5529}
5530
6ebdcee2
JC
5531/*
5532 * Drops images above 'base' up to and including 'top', and sets the image
5533 * above 'top' to have base as its backing file.
5534 *
5535 * Requires that the overlay to 'top' is opened r/w, so that the backing file
5536 * information in 'bs' can be properly updated.
5537 *
5538 * E.g., this will convert the following chain:
5539 * bottom <- base <- intermediate <- top <- active
5540 *
5541 * to
5542 *
5543 * bottom <- base <- active
5544 *
5545 * It is allowed for bottom==base, in which case it converts:
5546 *
5547 * base <- intermediate <- top <- active
5548 *
5549 * to
5550 *
5551 * base <- active
5552 *
54e26900
JC
5553 * If backing_file_str is non-NULL, it will be used when modifying top's
5554 * overlay image metadata.
5555 *
6ebdcee2
JC
5556 * Error conditions:
5557 * if active == top, that is considered an error
5558 *
5559 */
bde70715
KW
5560int bdrv_drop_intermediate(BlockDriverState *top, BlockDriverState *base,
5561 const char *backing_file_str)
6ebdcee2 5562{
6bd858b3
AG
5563 BlockDriverState *explicit_top = top;
5564 bool update_inherits_from;
d669ed6a 5565 BdrvChild *c;
12fa4af6 5566 Error *local_err = NULL;
6ebdcee2 5567 int ret = -EIO;
d669ed6a
VSO
5568 g_autoptr(GSList) updated_children = NULL;
5569 GSList *p;
6ebdcee2 5570
f791bf7f
EGE
5571 GLOBAL_STATE_CODE();
5572
6858eba0 5573 bdrv_ref(top);
637d54a5 5574 bdrv_subtree_drained_begin(top);
6858eba0 5575
6ebdcee2
JC
5576 if (!top->drv || !base->drv) {
5577 goto exit;
5578 }
5579
5db15a57
KW
5580 /* Make sure that base is in the backing chain of top */
5581 if (!bdrv_chain_contains(top, base)) {
6ebdcee2
JC
5582 goto exit;
5583 }
5584
6bd858b3
AG
5585 /* If 'base' recursively inherits from 'top' then we should set
5586 * base->inherits_from to top->inherits_from after 'top' and all
5587 * other intermediate nodes have been dropped.
5588 * If 'top' is an implicit node (e.g. "commit_top") we should skip
5589 * it because no one inherits from it. We use explicit_top for that. */
dcf3f9b2 5590 explicit_top = bdrv_skip_implicit_filters(explicit_top);
6bd858b3
AG
5591 update_inherits_from = bdrv_inherits_from_recursive(base, explicit_top);
5592
6ebdcee2 5593 /* success - we can delete the intermediate states, and link top->base */
f30c66ba
HR
5594 if (!backing_file_str) {
5595 bdrv_refresh_filename(base);
5596 backing_file_str = base->filename;
5597 }
61f09cea 5598
d669ed6a
VSO
5599 QLIST_FOREACH(c, &top->parents, next_parent) {
5600 updated_children = g_slist_prepend(updated_children, c);
5601 }
5602
3108a15c
VSO
5603 /*
5604 * It seems correct to pass detach_subchain=true here, but it triggers
5605 * one more yet not fixed bug, when due to nested aio_poll loop we switch to
5606 * another drained section, which modify the graph (for example, removing
5607 * the child, which we keep in updated_children list). So, it's a TODO.
5608 *
5609 * Note, bug triggered if pass detach_subchain=true here and run
5610 * test-bdrv-drain. test_drop_intermediate_poll() test-case will crash.
5611 * That's a FIXME.
5612 */
5613 bdrv_replace_node_common(top, base, false, false, &local_err);
d669ed6a
VSO
5614 if (local_err) {
5615 error_report_err(local_err);
5616 goto exit;
5617 }
5618
5619 for (p = updated_children; p; p = p->next) {
5620 c = p->data;
12fa4af6 5621
bd86fb99
HR
5622 if (c->klass->update_filename) {
5623 ret = c->klass->update_filename(c, base, backing_file_str,
5624 &local_err);
61f09cea 5625 if (ret < 0) {
d669ed6a
VSO
5626 /*
5627 * TODO: Actually, we want to rollback all previous iterations
5628 * of this loop, and (which is almost impossible) previous
5629 * bdrv_replace_node()...
5630 *
5631 * Note, that c->klass->update_filename may lead to permission
5632 * update, so it's a bad idea to call it inside permission
5633 * update transaction of bdrv_replace_node.
5634 */
61f09cea
KW
5635 error_report_err(local_err);
5636 goto exit;
5637 }
5638 }
12fa4af6 5639 }
6ebdcee2 5640
6bd858b3
AG
5641 if (update_inherits_from) {
5642 base->inherits_from = explicit_top->inherits_from;
5643 }
5644
6ebdcee2 5645 ret = 0;
6ebdcee2 5646exit:
637d54a5 5647 bdrv_subtree_drained_end(top);
6858eba0 5648 bdrv_unref(top);
6ebdcee2
JC
5649 return ret;
5650}
5651
081e4650
HR
5652/**
5653 * Implementation of BlockDriver.bdrv_get_allocated_file_size() that
5654 * sums the size of all data-bearing children. (This excludes backing
5655 * children.)
5656 */
5657static int64_t bdrv_sum_allocated_file_size(BlockDriverState *bs)
5658{
5659 BdrvChild *child;
5660 int64_t child_size, sum = 0;
5661
5662 QLIST_FOREACH(child, &bs->children, next) {
5663 if (child->role & (BDRV_CHILD_DATA | BDRV_CHILD_METADATA |
5664 BDRV_CHILD_FILTERED))
5665 {
5666 child_size = bdrv_get_allocated_file_size(child->bs);
5667 if (child_size < 0) {
5668 return child_size;
5669 }
5670 sum += child_size;
5671 }
5672 }
5673
5674 return sum;
5675}
5676
61007b31
SH
5677/**
5678 * Length of a allocated file in bytes. Sparse files are counted by actual
5679 * allocated space. Return < 0 if error or unknown.
5680 */
5681int64_t bdrv_get_allocated_file_size(BlockDriverState *bs)
71d0770c 5682{
61007b31 5683 BlockDriver *drv = bs->drv;
384a48fb
EGE
5684 IO_CODE();
5685
61007b31
SH
5686 if (!drv) {
5687 return -ENOMEDIUM;
8f4754ed 5688 }
61007b31
SH
5689 if (drv->bdrv_get_allocated_file_size) {
5690 return drv->bdrv_get_allocated_file_size(bs);
5691 }
081e4650
HR
5692
5693 if (drv->bdrv_file_open) {
5694 /*
5695 * Protocol drivers default to -ENOTSUP (most of their data is
5696 * not stored in any of their children (if they even have any),
5697 * so there is no generic way to figure it out).
5698 */
5699 return -ENOTSUP;
5700 } else if (drv->is_filter) {
5701 /* Filter drivers default to the size of their filtered child */
5702 return bdrv_get_allocated_file_size(bdrv_filter_bs(bs));
5703 } else {
5704 /* Other drivers default to summing their children's sizes */
5705 return bdrv_sum_allocated_file_size(bs);
1c9805a3
SH
5706 }
5707}
e7a8a783 5708
90880ff1
SH
5709/*
5710 * bdrv_measure:
5711 * @drv: Format driver
5712 * @opts: Creation options for new image
5713 * @in_bs: Existing image containing data for new image (may be NULL)
5714 * @errp: Error object
5715 * Returns: A #BlockMeasureInfo (free using qapi_free_BlockMeasureInfo())
5716 * or NULL on error
5717 *
5718 * Calculate file size required to create a new image.
5719 *
5720 * If @in_bs is given then space for allocated clusters and zero clusters
5721 * from that image are included in the calculation. If @opts contains a
5722 * backing file that is shared by @in_bs then backing clusters may be omitted
5723 * from the calculation.
5724 *
5725 * If @in_bs is NULL then the calculation includes no allocated clusters
5726 * unless a preallocation option is given in @opts.
5727 *
5728 * Note that @in_bs may use a different BlockDriver from @drv.
5729 *
5730 * If an error occurs the @errp pointer is set.
5731 */
5732BlockMeasureInfo *bdrv_measure(BlockDriver *drv, QemuOpts *opts,
5733 BlockDriverState *in_bs, Error **errp)
5734{
384a48fb 5735 IO_CODE();
90880ff1
SH
5736 if (!drv->bdrv_measure) {
5737 error_setg(errp, "Block driver '%s' does not support size measurement",
5738 drv->format_name);
5739 return NULL;
5740 }
5741
5742 return drv->bdrv_measure(opts, in_bs, errp);
5743}
5744
61007b31
SH
5745/**
5746 * Return number of sectors on success, -errno on error.
1c9805a3 5747 */
61007b31 5748int64_t bdrv_nb_sectors(BlockDriverState *bs)
1c9805a3 5749{
61007b31 5750 BlockDriver *drv = bs->drv;
384a48fb 5751 IO_CODE();
498e386c 5752
61007b31
SH
5753 if (!drv)
5754 return -ENOMEDIUM;
2572b37a 5755
61007b31
SH
5756 if (drv->has_variable_length) {
5757 int ret = refresh_total_sectors(bs, bs->total_sectors);
5758 if (ret < 0) {
5759 return ret;
1c9805a3
SH
5760 }
5761 }
61007b31 5762 return bs->total_sectors;
1c9805a3 5763}
b338082b 5764
61007b31
SH
5765/**
5766 * Return length in bytes on success, -errno on error.
5767 * The length is always a multiple of BDRV_SECTOR_SIZE.
8d3b1a2d 5768 */
61007b31 5769int64_t bdrv_getlength(BlockDriverState *bs)
8d3b1a2d 5770{
61007b31 5771 int64_t ret = bdrv_nb_sectors(bs);
384a48fb 5772 IO_CODE();
8d3b1a2d 5773
122860ba
EB
5774 if (ret < 0) {
5775 return ret;
5776 }
5777 if (ret > INT64_MAX / BDRV_SECTOR_SIZE) {
5778 return -EFBIG;
5779 }
5780 return ret * BDRV_SECTOR_SIZE;
fc01f7e7
FB
5781}
5782
61007b31
SH
5783/* return 0 as number of sectors if no device present or error */
5784void bdrv_get_geometry(BlockDriverState *bs, uint64_t *nb_sectors_ptr)
07d27a44 5785{
61007b31 5786 int64_t nb_sectors = bdrv_nb_sectors(bs);
384a48fb 5787 IO_CODE();
07d27a44 5788
61007b31 5789 *nb_sectors_ptr = nb_sectors < 0 ? 0 : nb_sectors;
07d27a44
MA
5790}
5791
54115412 5792bool bdrv_is_sg(BlockDriverState *bs)
f08145fe 5793{
384a48fb 5794 IO_CODE();
61007b31 5795 return bs->sg;
f08145fe
KW
5796}
5797
ae23f786
HR
5798/**
5799 * Return whether the given node supports compressed writes.
5800 */
5801bool bdrv_supports_compressed_writes(BlockDriverState *bs)
5802{
5803 BlockDriverState *filtered;
384a48fb 5804 IO_CODE();
ae23f786
HR
5805
5806 if (!bs->drv || !block_driver_can_compress(bs->drv)) {
5807 return false;
5808 }
5809
5810 filtered = bdrv_filter_bs(bs);
5811 if (filtered) {
5812 /*
5813 * Filters can only forward compressed writes, so we have to
5814 * check the child.
5815 */
5816 return bdrv_supports_compressed_writes(filtered);
5817 }
5818
5819 return true;
5820}
5821
61007b31 5822const char *bdrv_get_format_name(BlockDriverState *bs)
40b4f539 5823{
384a48fb 5824 IO_CODE();
61007b31 5825 return bs->drv ? bs->drv->format_name : NULL;
40b4f539
KW
5826}
5827
61007b31 5828static int qsort_strcmp(const void *a, const void *b)
40b4f539 5829{
ceff5bd7 5830 return strcmp(*(char *const *)a, *(char *const *)b);
40b4f539
KW
5831}
5832
61007b31 5833void bdrv_iterate_format(void (*it)(void *opaque, const char *name),
9ac404c5 5834 void *opaque, bool read_only)
40b4f539 5835{
61007b31
SH
5836 BlockDriver *drv;
5837 int count = 0;
5838 int i;
5839 const char **formats = NULL;
40b4f539 5840
f791bf7f
EGE
5841 GLOBAL_STATE_CODE();
5842
61007b31
SH
5843 QLIST_FOREACH(drv, &bdrv_drivers, list) {
5844 if (drv->format_name) {
5845 bool found = false;
5846 int i = count;
9ac404c5
AS
5847
5848 if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv, read_only)) {
5849 continue;
5850 }
5851
61007b31
SH
5852 while (formats && i && !found) {
5853 found = !strcmp(formats[--i], drv->format_name);
5854 }
e2a305fb 5855
61007b31
SH
5856 if (!found) {
5857 formats = g_renew(const char *, formats, count + 1);
5858 formats[count++] = drv->format_name;
5859 }
6c5a42ac 5860 }
61007b31 5861 }
6c5a42ac 5862
eb0df69f
HR
5863 for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); i++) {
5864 const char *format_name = block_driver_modules[i].format_name;
5865
5866 if (format_name) {
5867 bool found = false;
5868 int j = count;
5869
9ac404c5
AS
5870 if (use_bdrv_whitelist &&
5871 !bdrv_format_is_whitelisted(format_name, read_only)) {
5872 continue;
5873 }
5874
eb0df69f
HR
5875 while (formats && j && !found) {
5876 found = !strcmp(formats[--j], format_name);
5877 }
5878
5879 if (!found) {
5880 formats = g_renew(const char *, formats, count + 1);
5881 formats[count++] = format_name;
5882 }
5883 }
5884 }
5885
61007b31 5886 qsort(formats, count, sizeof(formats[0]), qsort_strcmp);
40b4f539 5887
61007b31
SH
5888 for (i = 0; i < count; i++) {
5889 it(opaque, formats[i]);
5890 }
40b4f539 5891
61007b31
SH
5892 g_free(formats);
5893}
40b4f539 5894
61007b31
SH
5895/* This function is to find a node in the bs graph */
5896BlockDriverState *bdrv_find_node(const char *node_name)
5897{
5898 BlockDriverState *bs;
391827eb 5899
61007b31 5900 assert(node_name);
f791bf7f 5901 GLOBAL_STATE_CODE();
40b4f539 5902
61007b31
SH
5903 QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
5904 if (!strcmp(node_name, bs->node_name)) {
5905 return bs;
40b4f539
KW
5906 }
5907 }
61007b31 5908 return NULL;
40b4f539
KW
5909}
5910
61007b31 5911/* Put this QMP function here so it can access the static graph_bdrv_states. */
facda544
PK
5912BlockDeviceInfoList *bdrv_named_nodes_list(bool flat,
5913 Error **errp)
40b4f539 5914{
9812e712 5915 BlockDeviceInfoList *list;
61007b31 5916 BlockDriverState *bs;
40b4f539 5917
f791bf7f
EGE
5918 GLOBAL_STATE_CODE();
5919
61007b31
SH
5920 list = NULL;
5921 QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
facda544 5922 BlockDeviceInfo *info = bdrv_block_device_info(NULL, bs, flat, errp);
61007b31
SH
5923 if (!info) {
5924 qapi_free_BlockDeviceInfoList(list);
5925 return NULL;
301db7c2 5926 }
9812e712 5927 QAPI_LIST_PREPEND(list, info);
301db7c2
RH
5928 }
5929
61007b31
SH
5930 return list;
5931}
40b4f539 5932
5d3b4e99
VSO
5933typedef struct XDbgBlockGraphConstructor {
5934 XDbgBlockGraph *graph;
5935 GHashTable *graph_nodes;
5936} XDbgBlockGraphConstructor;
5937
5938static XDbgBlockGraphConstructor *xdbg_graph_new(void)
5939{
5940 XDbgBlockGraphConstructor *gr = g_new(XDbgBlockGraphConstructor, 1);
5941
5942 gr->graph = g_new0(XDbgBlockGraph, 1);
5943 gr->graph_nodes = g_hash_table_new(NULL, NULL);
5944
5945 return gr;
5946}
5947
5948static XDbgBlockGraph *xdbg_graph_finalize(XDbgBlockGraphConstructor *gr)
5949{
5950 XDbgBlockGraph *graph = gr->graph;
5951
5952 g_hash_table_destroy(gr->graph_nodes);
5953 g_free(gr);
5954
5955 return graph;
5956}
5957
5958static uintptr_t xdbg_graph_node_num(XDbgBlockGraphConstructor *gr, void *node)
5959{
5960 uintptr_t ret = (uintptr_t)g_hash_table_lookup(gr->graph_nodes, node);
5961
5962 if (ret != 0) {
5963 return ret;
5964 }
5965
5966 /*
5967 * Start counting from 1, not 0, because 0 interferes with not-found (NULL)
5968 * answer of g_hash_table_lookup.
5969 */
5970 ret = g_hash_table_size(gr->graph_nodes) + 1;
5971 g_hash_table_insert(gr->graph_nodes, node, (void *)ret);
5972
5973 return ret;
5974}
5975
5976static void xdbg_graph_add_node(XDbgBlockGraphConstructor *gr, void *node,
5977 XDbgBlockGraphNodeType type, const char *name)
5978{
5979 XDbgBlockGraphNode *n;
5980
5981 n = g_new0(XDbgBlockGraphNode, 1);
5982
5983 n->id = xdbg_graph_node_num(gr, node);
5984 n->type = type;
5985 n->name = g_strdup(name);
5986
9812e712 5987 QAPI_LIST_PREPEND(gr->graph->nodes, n);
5d3b4e99
VSO
5988}
5989
5990static void xdbg_graph_add_edge(XDbgBlockGraphConstructor *gr, void *parent,
5991 const BdrvChild *child)
5992{
cdb1cec8 5993 BlockPermission qapi_perm;
5d3b4e99 5994 XDbgBlockGraphEdge *edge;
862fded9 5995 GLOBAL_STATE_CODE();
5d3b4e99 5996
5d3b4e99
VSO
5997 edge = g_new0(XDbgBlockGraphEdge, 1);
5998
5999 edge->parent = xdbg_graph_node_num(gr, parent);
6000 edge->child = xdbg_graph_node_num(gr, child->bs);
6001 edge->name = g_strdup(child->name);
6002
cdb1cec8
HR
6003 for (qapi_perm = 0; qapi_perm < BLOCK_PERMISSION__MAX; qapi_perm++) {
6004 uint64_t flag = bdrv_qapi_perm_to_blk_perm(qapi_perm);
6005
6006 if (flag & child->perm) {
9812e712 6007 QAPI_LIST_PREPEND(edge->perm, qapi_perm);
5d3b4e99 6008 }
cdb1cec8 6009 if (flag & child->shared_perm) {
9812e712 6010 QAPI_LIST_PREPEND(edge->shared_perm, qapi_perm);
5d3b4e99
VSO
6011 }
6012 }
6013
9812e712 6014 QAPI_LIST_PREPEND(gr->graph->edges, edge);
5d3b4e99
VSO
6015}
6016
6017
6018XDbgBlockGraph *bdrv_get_xdbg_block_graph(Error **errp)
6019{
6020 BlockBackend *blk;
6021 BlockJob *job;
6022 BlockDriverState *bs;
6023 BdrvChild *child;
6024 XDbgBlockGraphConstructor *gr = xdbg_graph_new();
6025
f791bf7f
EGE
6026 GLOBAL_STATE_CODE();
6027
5d3b4e99
VSO
6028 for (blk = blk_all_next(NULL); blk; blk = blk_all_next(blk)) {
6029 char *allocated_name = NULL;
6030 const char *name = blk_name(blk);
6031
6032 if (!*name) {
6033 name = allocated_name = blk_get_attached_dev_id(blk);
6034 }
6035 xdbg_graph_add_node(gr, blk, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_BACKEND,
6036 name);
6037 g_free(allocated_name);
6038 if (blk_root(blk)) {
6039 xdbg_graph_add_edge(gr, blk, blk_root(blk));
6040 }
6041 }
6042
880eeec6
EGE
6043 WITH_JOB_LOCK_GUARD() {
6044 for (job = block_job_next_locked(NULL); job;
6045 job = block_job_next_locked(job)) {
6046 GSList *el;
5d3b4e99 6047
880eeec6
EGE
6048 xdbg_graph_add_node(gr, job, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_JOB,
6049 job->job.id);
6050 for (el = job->nodes; el; el = el->next) {
6051 xdbg_graph_add_edge(gr, job, (BdrvChild *)el->data);
6052 }
5d3b4e99
VSO
6053 }
6054 }
6055
6056 QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
6057 xdbg_graph_add_node(gr, bs, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_DRIVER,
6058 bs->node_name);
6059 QLIST_FOREACH(child, &bs->children, next) {
6060 xdbg_graph_add_edge(gr, bs, child);
6061 }
6062 }
6063
6064 return xdbg_graph_finalize(gr);
6065}
6066
61007b31
SH
6067BlockDriverState *bdrv_lookup_bs(const char *device,
6068 const char *node_name,
6069 Error **errp)
6070{
6071 BlockBackend *blk;
6072 BlockDriverState *bs;
40b4f539 6073
f791bf7f
EGE
6074 GLOBAL_STATE_CODE();
6075
61007b31
SH
6076 if (device) {
6077 blk = blk_by_name(device);
40b4f539 6078
61007b31 6079 if (blk) {
9f4ed6fb
AG
6080 bs = blk_bs(blk);
6081 if (!bs) {
5433c24f 6082 error_setg(errp, "Device '%s' has no medium", device);
5433c24f
HR
6083 }
6084
9f4ed6fb 6085 return bs;
61007b31
SH
6086 }
6087 }
40b4f539 6088
61007b31
SH
6089 if (node_name) {
6090 bs = bdrv_find_node(node_name);
6d519a5f 6091
61007b31
SH
6092 if (bs) {
6093 return bs;
6094 }
40b4f539
KW
6095 }
6096
785ec4b1 6097 error_setg(errp, "Cannot find device=\'%s\' nor node-name=\'%s\'",
61007b31
SH
6098 device ? device : "",
6099 node_name ? node_name : "");
6100 return NULL;
40b4f539
KW
6101}
6102
61007b31
SH
6103/* If 'base' is in the same chain as 'top', return true. Otherwise,
6104 * return false. If either argument is NULL, return false. */
6105bool bdrv_chain_contains(BlockDriverState *top, BlockDriverState *base)
83f64091 6106{
f791bf7f
EGE
6107
6108 GLOBAL_STATE_CODE();
6109
61007b31 6110 while (top && top != base) {
dcf3f9b2 6111 top = bdrv_filter_or_cow_bs(top);
02c50efe 6112 }
61007b31
SH
6113
6114 return top != NULL;
02c50efe
FZ
6115}
6116
61007b31 6117BlockDriverState *bdrv_next_node(BlockDriverState *bs)
02c50efe 6118{
f791bf7f 6119 GLOBAL_STATE_CODE();
61007b31
SH
6120 if (!bs) {
6121 return QTAILQ_FIRST(&graph_bdrv_states);
02c50efe 6122 }
61007b31 6123 return QTAILQ_NEXT(bs, node_list);
83f64091
FB
6124}
6125
0f12264e
KW
6126BlockDriverState *bdrv_next_all_states(BlockDriverState *bs)
6127{
f791bf7f 6128 GLOBAL_STATE_CODE();
0f12264e
KW
6129 if (!bs) {
6130 return QTAILQ_FIRST(&all_bdrv_states);
6131 }
6132 return QTAILQ_NEXT(bs, bs_list);
6133}
6134
61007b31 6135const char *bdrv_get_node_name(const BlockDriverState *bs)
83f64091 6136{
384a48fb 6137 IO_CODE();
61007b31 6138 return bs->node_name;
beac80cd
FB
6139}
6140
1f0c461b 6141const char *bdrv_get_parent_name(const BlockDriverState *bs)
4c265bf9
KW
6142{
6143 BdrvChild *c;
6144 const char *name;
967d7905 6145 IO_CODE();
4c265bf9
KW
6146
6147 /* If multiple parents have a name, just pick the first one. */
6148 QLIST_FOREACH(c, &bs->parents, next_parent) {
bd86fb99
HR
6149 if (c->klass->get_name) {
6150 name = c->klass->get_name(c);
4c265bf9
KW
6151 if (name && *name) {
6152 return name;
6153 }
6154 }
6155 }
6156
6157 return NULL;
6158}
6159
61007b31
SH
6160/* TODO check what callers really want: bs->node_name or blk_name() */
6161const char *bdrv_get_device_name(const BlockDriverState *bs)
beac80cd 6162{
384a48fb 6163 IO_CODE();
4c265bf9 6164 return bdrv_get_parent_name(bs) ?: "";
f141eafe 6165}
83f64091 6166
61007b31
SH
6167/* This can be used to identify nodes that might not have a device
6168 * name associated. Since node and device names live in the same
6169 * namespace, the result is unambiguous. The exception is if both are
6170 * absent, then this returns an empty (non-null) string. */
6171const char *bdrv_get_device_or_node_name(const BlockDriverState *bs)
f141eafe 6172{
384a48fb 6173 IO_CODE();
4c265bf9 6174 return bdrv_get_parent_name(bs) ?: bs->node_name;
beac80cd 6175}
beac80cd 6176
61007b31 6177int bdrv_get_flags(BlockDriverState *bs)
0b5a2445 6178{
15aee7ac 6179 IO_CODE();
61007b31 6180 return bs->open_flags;
0b5a2445
PB
6181}
6182
61007b31 6183int bdrv_has_zero_init_1(BlockDriverState *bs)
68485420 6184{
f791bf7f 6185 GLOBAL_STATE_CODE();
61007b31 6186 return 1;
0b5a2445
PB
6187}
6188
61007b31 6189int bdrv_has_zero_init(BlockDriverState *bs)
0b5a2445 6190{
93393e69 6191 BlockDriverState *filtered;
f791bf7f 6192 GLOBAL_STATE_CODE();
93393e69 6193
d470ad42
HR
6194 if (!bs->drv) {
6195 return 0;
6196 }
0b5a2445 6197
61007b31
SH
6198 /* If BS is a copy on write image, it is initialized to
6199 the contents of the base image, which may not be zeroes. */
34778172 6200 if (bdrv_cow_child(bs)) {
61007b31
SH
6201 return 0;
6202 }
6203 if (bs->drv->bdrv_has_zero_init) {
6204 return bs->drv->bdrv_has_zero_init(bs);
0b5a2445 6205 }
93393e69
HR
6206
6207 filtered = bdrv_filter_bs(bs);
6208 if (filtered) {
6209 return bdrv_has_zero_init(filtered);
5a612c00 6210 }
61007b31
SH
6211
6212 /* safe default */
6213 return 0;
68485420
KW
6214}
6215
61007b31 6216bool bdrv_can_write_zeroes_with_unmap(BlockDriverState *bs)
68485420 6217{
384a48fb 6218 IO_CODE();
2f0342ef 6219 if (!(bs->open_flags & BDRV_O_UNMAP)) {
61007b31
SH
6220 return false;
6221 }
68485420 6222
e24d813b 6223 return bs->supported_zero_flags & BDRV_REQ_MAY_UNMAP;
68485420
KW
6224}
6225
61007b31
SH
6226void bdrv_get_backing_filename(BlockDriverState *bs,
6227 char *filename, int filename_size)
016f5cf6 6228{
384a48fb 6229 IO_CODE();
61007b31
SH
6230 pstrcpy(filename, filename_size, bs->backing_file);
6231}
d318aea9 6232
61007b31
SH
6233int bdrv_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
6234{
8b117001 6235 int ret;
61007b31 6236 BlockDriver *drv = bs->drv;
384a48fb 6237 IO_CODE();
5a612c00
MP
6238 /* if bs->drv == NULL, bs is closed, so there's nothing to do here */
6239 if (!drv) {
61007b31 6240 return -ENOMEDIUM;
5a612c00
MP
6241 }
6242 if (!drv->bdrv_get_info) {
93393e69
HR
6243 BlockDriverState *filtered = bdrv_filter_bs(bs);
6244 if (filtered) {
6245 return bdrv_get_info(filtered, bdi);
5a612c00 6246 }
61007b31 6247 return -ENOTSUP;
5a612c00 6248 }
61007b31 6249 memset(bdi, 0, sizeof(*bdi));
8b117001
VSO
6250 ret = drv->bdrv_get_info(bs, bdi);
6251 if (ret < 0) {
6252 return ret;
6253 }
6254
6255 if (bdi->cluster_size > BDRV_MAX_ALIGNMENT) {
6256 return -EINVAL;
6257 }
6258
6259 return 0;
61007b31 6260}
016f5cf6 6261
1bf6e9ca
AS
6262ImageInfoSpecific *bdrv_get_specific_info(BlockDriverState *bs,
6263 Error **errp)
61007b31
SH
6264{
6265 BlockDriver *drv = bs->drv;
384a48fb 6266 IO_CODE();
61007b31 6267 if (drv && drv->bdrv_get_specific_info) {
1bf6e9ca 6268 return drv->bdrv_get_specific_info(bs, errp);
61007b31
SH
6269 }
6270 return NULL;
016f5cf6
AG
6271}
6272
d9245599
AN
6273BlockStatsSpecific *bdrv_get_specific_stats(BlockDriverState *bs)
6274{
6275 BlockDriver *drv = bs->drv;
384a48fb 6276 IO_CODE();
d9245599
AN
6277 if (!drv || !drv->bdrv_get_specific_stats) {
6278 return NULL;
6279 }
6280 return drv->bdrv_get_specific_stats(bs);
6281}
6282
a31939e6 6283void bdrv_debug_event(BlockDriverState *bs, BlkdebugEvent event)
4265d620 6284{
384a48fb 6285 IO_CODE();
61007b31
SH
6286 if (!bs || !bs->drv || !bs->drv->bdrv_debug_event) {
6287 return;
6288 }
4265d620 6289
61007b31 6290 bs->drv->bdrv_debug_event(bs, event);
4265d620
PB
6291}
6292
d10529a2 6293static BlockDriverState *bdrv_find_debug_node(BlockDriverState *bs)
4265d620 6294{
bdb73476 6295 GLOBAL_STATE_CODE();
61007b31 6296 while (bs && bs->drv && !bs->drv->bdrv_debug_breakpoint) {
f706a92f 6297 bs = bdrv_primary_bs(bs);
61007b31 6298 }
4265d620 6299
61007b31 6300 if (bs && bs->drv && bs->drv->bdrv_debug_breakpoint) {
d10529a2
VSO
6301 assert(bs->drv->bdrv_debug_remove_breakpoint);
6302 return bs;
6303 }
6304
6305 return NULL;
6306}
6307
6308int bdrv_debug_breakpoint(BlockDriverState *bs, const char *event,
6309 const char *tag)
6310{
f791bf7f 6311 GLOBAL_STATE_CODE();
d10529a2
VSO
6312 bs = bdrv_find_debug_node(bs);
6313 if (bs) {
61007b31
SH
6314 return bs->drv->bdrv_debug_breakpoint(bs, event, tag);
6315 }
4265d620 6316
61007b31 6317 return -ENOTSUP;
4265d620
PB
6318}
6319
61007b31 6320int bdrv_debug_remove_breakpoint(BlockDriverState *bs, const char *tag)
ea2384d3 6321{
f791bf7f 6322 GLOBAL_STATE_CODE();
d10529a2
VSO
6323 bs = bdrv_find_debug_node(bs);
6324 if (bs) {
61007b31
SH
6325 return bs->drv->bdrv_debug_remove_breakpoint(bs, tag);
6326 }
6327
6328 return -ENOTSUP;
eb852011
MA
6329}
6330
61007b31 6331int bdrv_debug_resume(BlockDriverState *bs, const char *tag)
ce1a14dc 6332{
f791bf7f 6333 GLOBAL_STATE_CODE();
61007b31 6334 while (bs && (!bs->drv || !bs->drv->bdrv_debug_resume)) {
f706a92f 6335 bs = bdrv_primary_bs(bs);
61007b31 6336 }
ce1a14dc 6337
61007b31
SH
6338 if (bs && bs->drv && bs->drv->bdrv_debug_resume) {
6339 return bs->drv->bdrv_debug_resume(bs, tag);
6340 }
ce1a14dc 6341
61007b31 6342 return -ENOTSUP;
f197fe2b
FZ
6343}
6344
61007b31 6345bool bdrv_debug_is_suspended(BlockDriverState *bs, const char *tag)
ce1a14dc 6346{
f791bf7f 6347 GLOBAL_STATE_CODE();
61007b31 6348 while (bs && bs->drv && !bs->drv->bdrv_debug_is_suspended) {
f706a92f 6349 bs = bdrv_primary_bs(bs);
f197fe2b 6350 }
19cb3738 6351
61007b31
SH
6352 if (bs && bs->drv && bs->drv->bdrv_debug_is_suspended) {
6353 return bs->drv->bdrv_debug_is_suspended(bs, tag);
6354 }
f9f05dc5 6355
61007b31
SH
6356 return false;
6357}
f9f05dc5 6358
61007b31
SH
6359/* backing_file can either be relative, or absolute, or a protocol. If it is
6360 * relative, it must be relative to the chain. So, passing in bs->filename
6361 * from a BDS as backing_file should not be done, as that may be relative to
6362 * the CWD rather than the chain. */
6363BlockDriverState *bdrv_find_backing_image(BlockDriverState *bs,
6364 const char *backing_file)
f9f05dc5 6365{
61007b31
SH
6366 char *filename_full = NULL;
6367 char *backing_file_full = NULL;
6368 char *filename_tmp = NULL;
6369 int is_protocol = 0;
0b877d09 6370 bool filenames_refreshed = false;
61007b31
SH
6371 BlockDriverState *curr_bs = NULL;
6372 BlockDriverState *retval = NULL;
dcf3f9b2 6373 BlockDriverState *bs_below;
f9f05dc5 6374
f791bf7f
EGE
6375 GLOBAL_STATE_CODE();
6376
61007b31
SH
6377 if (!bs || !bs->drv || !backing_file) {
6378 return NULL;
f9f05dc5
KW
6379 }
6380
61007b31
SH
6381 filename_full = g_malloc(PATH_MAX);
6382 backing_file_full = g_malloc(PATH_MAX);
f9f05dc5 6383
61007b31 6384 is_protocol = path_has_protocol(backing_file);
f9f05dc5 6385
dcf3f9b2
HR
6386 /*
6387 * Being largely a legacy function, skip any filters here
6388 * (because filters do not have normal filenames, so they cannot
6389 * match anyway; and allowing json:{} filenames is a bit out of
6390 * scope).
6391 */
6392 for (curr_bs = bdrv_skip_filters(bs);
6393 bdrv_cow_child(curr_bs) != NULL;
6394 curr_bs = bs_below)
6395 {
6396 bs_below = bdrv_backing_chain_next(curr_bs);
f9f05dc5 6397
0b877d09
HR
6398 if (bdrv_backing_overridden(curr_bs)) {
6399 /*
6400 * If the backing file was overridden, we can only compare
6401 * directly against the backing node's filename.
6402 */
6403
6404 if (!filenames_refreshed) {
6405 /*
6406 * This will automatically refresh all of the
6407 * filenames in the rest of the backing chain, so we
6408 * only need to do this once.
6409 */
6410 bdrv_refresh_filename(bs_below);
6411 filenames_refreshed = true;
6412 }
6413
6414 if (strcmp(backing_file, bs_below->filename) == 0) {
6415 retval = bs_below;
6416 break;
6417 }
6418 } else if (is_protocol || path_has_protocol(curr_bs->backing_file)) {
6419 /*
6420 * If either of the filename paths is actually a protocol, then
6421 * compare unmodified paths; otherwise make paths relative.
6422 */
6b6833c1
HR
6423 char *backing_file_full_ret;
6424
61007b31 6425 if (strcmp(backing_file, curr_bs->backing_file) == 0) {
dcf3f9b2 6426 retval = bs_below;
61007b31
SH
6427 break;
6428 }
418661e0 6429 /* Also check against the full backing filename for the image */
6b6833c1
HR
6430 backing_file_full_ret = bdrv_get_full_backing_filename(curr_bs,
6431 NULL);
6432 if (backing_file_full_ret) {
6433 bool equal = strcmp(backing_file, backing_file_full_ret) == 0;
6434 g_free(backing_file_full_ret);
6435 if (equal) {
dcf3f9b2 6436 retval = bs_below;
418661e0
JC
6437 break;
6438 }
418661e0 6439 }
61007b31
SH
6440 } else {
6441 /* If not an absolute filename path, make it relative to the current
6442 * image's filename path */
2d9158ce
HR
6443 filename_tmp = bdrv_make_absolute_filename(curr_bs, backing_file,
6444 NULL);
6445 /* We are going to compare canonicalized absolute pathnames */
6446 if (!filename_tmp || !realpath(filename_tmp, filename_full)) {
6447 g_free(filename_tmp);
61007b31
SH
6448 continue;
6449 }
2d9158ce 6450 g_free(filename_tmp);
07f07615 6451
61007b31
SH
6452 /* We need to make sure the backing filename we are comparing against
6453 * is relative to the current image filename (or absolute) */
2d9158ce
HR
6454 filename_tmp = bdrv_get_full_backing_filename(curr_bs, NULL);
6455 if (!filename_tmp || !realpath(filename_tmp, backing_file_full)) {
6456 g_free(filename_tmp);
61007b31
SH
6457 continue;
6458 }
2d9158ce 6459 g_free(filename_tmp);
eb489bb1 6460
61007b31 6461 if (strcmp(backing_file_full, filename_full) == 0) {
dcf3f9b2 6462 retval = bs_below;
61007b31
SH
6463 break;
6464 }
6465 }
eb489bb1
KW
6466 }
6467
61007b31
SH
6468 g_free(filename_full);
6469 g_free(backing_file_full);
61007b31
SH
6470 return retval;
6471}
6472
61007b31
SH
6473void bdrv_init(void)
6474{
e5f05f8c
KW
6475#ifdef CONFIG_BDRV_WHITELIST_TOOLS
6476 use_bdrv_whitelist = 1;
6477#endif
61007b31
SH
6478 module_call_init(MODULE_INIT_BLOCK);
6479}
29cdb251 6480
61007b31
SH
6481void bdrv_init_with_whitelist(void)
6482{
6483 use_bdrv_whitelist = 1;
6484 bdrv_init();
07f07615
PB
6485}
6486
a94750d9 6487int bdrv_activate(BlockDriverState *bs, Error **errp)
0f15423c 6488{
4417ab7a 6489 BdrvChild *child, *parent;
5a8a30db
KW
6490 Error *local_err = NULL;
6491 int ret;
9c98f145 6492 BdrvDirtyBitmap *bm;
5a8a30db 6493
f791bf7f
EGE
6494 GLOBAL_STATE_CODE();
6495
3456a8d1 6496 if (!bs->drv) {
5416645f 6497 return -ENOMEDIUM;
3456a8d1
KW
6498 }
6499
16e977d5 6500 QLIST_FOREACH(child, &bs->children, next) {
11d0c9b3 6501 bdrv_activate(child->bs, &local_err);
0d1c5c91 6502 if (local_err) {
0d1c5c91 6503 error_propagate(errp, local_err);
5416645f 6504 return -EINVAL;
0d1c5c91 6505 }
5a8a30db 6506 }
0d1c5c91 6507
dafe0960
KW
6508 /*
6509 * Update permissions, they may differ for inactive nodes.
6510 *
6511 * Note that the required permissions of inactive images are always a
6512 * subset of the permissions required after activating the image. This
6513 * allows us to just get the permissions upfront without restricting
11d0c9b3 6514 * bdrv_co_invalidate_cache().
dafe0960
KW
6515 *
6516 * It also means that in error cases, we don't have to try and revert to
6517 * the old permissions (which is an operation that could fail, too). We can
6518 * just keep the extended permissions for the next time that an activation
6519 * of the image is tried.
6520 */
7bb4941a
KW
6521 if (bs->open_flags & BDRV_O_INACTIVE) {
6522 bs->open_flags &= ~BDRV_O_INACTIVE;
071b474f 6523 ret = bdrv_refresh_perms(bs, errp);
7bb4941a 6524 if (ret < 0) {
0d1c5c91 6525 bs->open_flags |= BDRV_O_INACTIVE;
5416645f 6526 return ret;
0d1c5c91 6527 }
3456a8d1 6528
11d0c9b3
EGE
6529 ret = bdrv_invalidate_cache(bs, errp);
6530 if (ret < 0) {
6531 bs->open_flags |= BDRV_O_INACTIVE;
6532 return ret;
7bb4941a 6533 }
9c98f145 6534
7bb4941a
KW
6535 FOR_EACH_DIRTY_BITMAP(bs, bm) {
6536 bdrv_dirty_bitmap_skip_store(bm, false);
6537 }
6538
6539 ret = refresh_total_sectors(bs, bs->total_sectors);
6540 if (ret < 0) {
6541 bs->open_flags |= BDRV_O_INACTIVE;
6542 error_setg_errno(errp, -ret, "Could not refresh total sector count");
5416645f 6543 return ret;
7bb4941a 6544 }
5a8a30db 6545 }
4417ab7a
KW
6546
6547 QLIST_FOREACH(parent, &bs->parents, next_parent) {
bd86fb99
HR
6548 if (parent->klass->activate) {
6549 parent->klass->activate(parent, &local_err);
4417ab7a 6550 if (local_err) {
78fc3b3a 6551 bs->open_flags |= BDRV_O_INACTIVE;
4417ab7a 6552 error_propagate(errp, local_err);
5416645f 6553 return -EINVAL;
4417ab7a
KW
6554 }
6555 }
6556 }
5416645f
VSO
6557
6558 return 0;
0f15423c
AL
6559}
6560
11d0c9b3
EGE
6561int coroutine_fn bdrv_co_invalidate_cache(BlockDriverState *bs, Error **errp)
6562{
6563 Error *local_err = NULL;
1581a70d 6564 IO_CODE();
11d0c9b3
EGE
6565
6566 assert(!(bs->open_flags & BDRV_O_INACTIVE));
6567
6568 if (bs->drv->bdrv_co_invalidate_cache) {
6569 bs->drv->bdrv_co_invalidate_cache(bs, &local_err);
6570 if (local_err) {
6571 error_propagate(errp, local_err);
6572 return -EINVAL;
6573 }
6574 }
6575
6576 return 0;
6577}
6578
3b717194 6579void bdrv_activate_all(Error **errp)
0f15423c 6580{
7c8eece4 6581 BlockDriverState *bs;
88be7b4b 6582 BdrvNextIterator it;
0f15423c 6583
f791bf7f
EGE
6584 GLOBAL_STATE_CODE();
6585
88be7b4b 6586 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
ed78cda3 6587 AioContext *aio_context = bdrv_get_aio_context(bs);
5416645f 6588 int ret;
ed78cda3
SH
6589
6590 aio_context_acquire(aio_context);
a94750d9 6591 ret = bdrv_activate(bs, errp);
ed78cda3 6592 aio_context_release(aio_context);
5416645f 6593 if (ret < 0) {
5e003f17 6594 bdrv_next_cleanup(&it);
5a8a30db
KW
6595 return;
6596 }
0f15423c
AL
6597 }
6598}
6599
9e37271f
KW
6600static bool bdrv_has_bds_parent(BlockDriverState *bs, bool only_active)
6601{
6602 BdrvChild *parent;
bdb73476 6603 GLOBAL_STATE_CODE();
9e37271f
KW
6604
6605 QLIST_FOREACH(parent, &bs->parents, next_parent) {
bd86fb99 6606 if (parent->klass->parent_is_bds) {
9e37271f
KW
6607 BlockDriverState *parent_bs = parent->opaque;
6608 if (!only_active || !(parent_bs->open_flags & BDRV_O_INACTIVE)) {
6609 return true;
6610 }
6611 }
6612 }
6613
6614 return false;
6615}
6616
6617static int bdrv_inactivate_recurse(BlockDriverState *bs)
76b1c7fe 6618{
cfa1a572 6619 BdrvChild *child, *parent;
76b1c7fe 6620 int ret;
a13de40a 6621 uint64_t cumulative_perms, cumulative_shared_perms;
76b1c7fe 6622
da359909
EGE
6623 GLOBAL_STATE_CODE();
6624
d470ad42
HR
6625 if (!bs->drv) {
6626 return -ENOMEDIUM;
6627 }
6628
9e37271f
KW
6629 /* Make sure that we don't inactivate a child before its parent.
6630 * It will be covered by recursion from the yet active parent. */
6631 if (bdrv_has_bds_parent(bs, true)) {
6632 return 0;
6633 }
6634
6635 assert(!(bs->open_flags & BDRV_O_INACTIVE));
6636
6637 /* Inactivate this node */
6638 if (bs->drv->bdrv_inactivate) {
76b1c7fe
KW
6639 ret = bs->drv->bdrv_inactivate(bs);
6640 if (ret < 0) {
6641 return ret;
6642 }
6643 }
6644
9e37271f 6645 QLIST_FOREACH(parent, &bs->parents, next_parent) {
bd86fb99
HR
6646 if (parent->klass->inactivate) {
6647 ret = parent->klass->inactivate(parent);
9e37271f
KW
6648 if (ret < 0) {
6649 return ret;
cfa1a572
KW
6650 }
6651 }
9e37271f 6652 }
9c5e6594 6653
a13de40a
VSO
6654 bdrv_get_cumulative_perm(bs, &cumulative_perms,
6655 &cumulative_shared_perms);
6656 if (cumulative_perms & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)) {
6657 /* Our inactive parents still need write access. Inactivation failed. */
6658 return -EPERM;
6659 }
6660
9e37271f 6661 bs->open_flags |= BDRV_O_INACTIVE;
7d5b5261 6662
bb87e4d1
VSO
6663 /*
6664 * Update permissions, they may differ for inactive nodes.
6665 * We only tried to loosen restrictions, so errors are not fatal, ignore
6666 * them.
6667 */
071b474f 6668 bdrv_refresh_perms(bs, NULL);
9e37271f
KW
6669
6670 /* Recursively inactivate children */
38701b6a 6671 QLIST_FOREACH(child, &bs->children, next) {
9e37271f 6672 ret = bdrv_inactivate_recurse(child->bs);
38701b6a
KW
6673 if (ret < 0) {
6674 return ret;
6675 }
6676 }
6677
76b1c7fe
KW
6678 return 0;
6679}
6680
6681int bdrv_inactivate_all(void)
6682{
79720af6 6683 BlockDriverState *bs = NULL;
88be7b4b 6684 BdrvNextIterator it;
aad0b7a0 6685 int ret = 0;
bd6458e4 6686 GSList *aio_ctxs = NULL, *ctx;
76b1c7fe 6687
f791bf7f
EGE
6688 GLOBAL_STATE_CODE();
6689
88be7b4b 6690 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
bd6458e4
PB
6691 AioContext *aio_context = bdrv_get_aio_context(bs);
6692
6693 if (!g_slist_find(aio_ctxs, aio_context)) {
6694 aio_ctxs = g_slist_prepend(aio_ctxs, aio_context);
6695 aio_context_acquire(aio_context);
6696 }
aad0b7a0 6697 }
76b1c7fe 6698
9e37271f
KW
6699 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
6700 /* Nodes with BDS parents are covered by recursion from the last
6701 * parent that gets inactivated. Don't inactivate them a second
6702 * time if that has already happened. */
6703 if (bdrv_has_bds_parent(bs, false)) {
6704 continue;
6705 }
6706 ret = bdrv_inactivate_recurse(bs);
6707 if (ret < 0) {
6708 bdrv_next_cleanup(&it);
6709 goto out;
76b1c7fe
KW
6710 }
6711 }
6712
aad0b7a0 6713out:
bd6458e4
PB
6714 for (ctx = aio_ctxs; ctx != NULL; ctx = ctx->next) {
6715 AioContext *aio_context = ctx->data;
6716 aio_context_release(aio_context);
aad0b7a0 6717 }
bd6458e4 6718 g_slist_free(aio_ctxs);
aad0b7a0
FZ
6719
6720 return ret;
76b1c7fe
KW
6721}
6722
19cb3738
FB
6723/**************************************************************/
6724/* removable device support */
6725
6726/**
6727 * Return TRUE if the media is present
6728 */
e031f750 6729bool bdrv_is_inserted(BlockDriverState *bs)
19cb3738
FB
6730{
6731 BlockDriver *drv = bs->drv;
28d7a789 6732 BdrvChild *child;
384a48fb 6733 IO_CODE();
a1aff5bf 6734
e031f750
HR
6735 if (!drv) {
6736 return false;
6737 }
28d7a789
HR
6738 if (drv->bdrv_is_inserted) {
6739 return drv->bdrv_is_inserted(bs);
6740 }
6741 QLIST_FOREACH(child, &bs->children, next) {
6742 if (!bdrv_is_inserted(child->bs)) {
6743 return false;
6744 }
e031f750 6745 }
28d7a789 6746 return true;
19cb3738
FB
6747}
6748
19cb3738
FB
6749/**
6750 * If eject_flag is TRUE, eject the media. Otherwise, close the tray
6751 */
f36f3949 6752void bdrv_eject(BlockDriverState *bs, bool eject_flag)
19cb3738
FB
6753{
6754 BlockDriver *drv = bs->drv;
384a48fb 6755 IO_CODE();
19cb3738 6756
822e1cd1
MA
6757 if (drv && drv->bdrv_eject) {
6758 drv->bdrv_eject(bs, eject_flag);
19cb3738
FB
6759 }
6760}
6761
19cb3738
FB
6762/**
6763 * Lock or unlock the media (if it is locked, the user won't be able
6764 * to eject it manually).
6765 */
025e849a 6766void bdrv_lock_medium(BlockDriverState *bs, bool locked)
19cb3738
FB
6767{
6768 BlockDriver *drv = bs->drv;
384a48fb 6769 IO_CODE();
025e849a 6770 trace_bdrv_lock_medium(bs, locked);
b8c6d095 6771
025e849a
MA
6772 if (drv && drv->bdrv_lock_medium) {
6773 drv->bdrv_lock_medium(bs, locked);
19cb3738
FB
6774 }
6775}
985a03b0 6776
9fcb0251
FZ
6777/* Get a reference to bs */
6778void bdrv_ref(BlockDriverState *bs)
6779{
f791bf7f 6780 GLOBAL_STATE_CODE();
9fcb0251
FZ
6781 bs->refcnt++;
6782}
6783
6784/* Release a previously grabbed reference to bs.
6785 * If after releasing, reference count is zero, the BlockDriverState is
6786 * deleted. */
6787void bdrv_unref(BlockDriverState *bs)
6788{
f791bf7f 6789 GLOBAL_STATE_CODE();
9a4d5ca6
JC
6790 if (!bs) {
6791 return;
6792 }
9fcb0251
FZ
6793 assert(bs->refcnt > 0);
6794 if (--bs->refcnt == 0) {
6795 bdrv_delete(bs);
6796 }
6797}
6798
fbe40ff7
FZ
6799struct BdrvOpBlocker {
6800 Error *reason;
6801 QLIST_ENTRY(BdrvOpBlocker) list;
6802};
6803
6804bool bdrv_op_is_blocked(BlockDriverState *bs, BlockOpType op, Error **errp)
6805{
6806 BdrvOpBlocker *blocker;
f791bf7f 6807 GLOBAL_STATE_CODE();
fbe40ff7
FZ
6808 assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
6809 if (!QLIST_EMPTY(&bs->op_blockers[op])) {
6810 blocker = QLIST_FIRST(&bs->op_blockers[op]);
4b576648
MA
6811 error_propagate_prepend(errp, error_copy(blocker->reason),
6812 "Node '%s' is busy: ",
6813 bdrv_get_device_or_node_name(bs));
fbe40ff7
FZ
6814 return true;
6815 }
6816 return false;
6817}
6818
6819void bdrv_op_block(BlockDriverState *bs, BlockOpType op, Error *reason)
6820{
6821 BdrvOpBlocker *blocker;
f791bf7f 6822 GLOBAL_STATE_CODE();
fbe40ff7
FZ
6823 assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
6824
5839e53b 6825 blocker = g_new0(BdrvOpBlocker, 1);
fbe40ff7
FZ
6826 blocker->reason = reason;
6827 QLIST_INSERT_HEAD(&bs->op_blockers[op], blocker, list);
6828}
6829
6830void bdrv_op_unblock(BlockDriverState *bs, BlockOpType op, Error *reason)
6831{
6832 BdrvOpBlocker *blocker, *next;
f791bf7f 6833 GLOBAL_STATE_CODE();
fbe40ff7
FZ
6834 assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
6835 QLIST_FOREACH_SAFE(blocker, &bs->op_blockers[op], list, next) {
6836 if (blocker->reason == reason) {
6837 QLIST_REMOVE(blocker, list);
6838 g_free(blocker);
6839 }
6840 }
6841}
6842
6843void bdrv_op_block_all(BlockDriverState *bs, Error *reason)
6844{
6845 int i;
f791bf7f 6846 GLOBAL_STATE_CODE();
fbe40ff7
FZ
6847 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
6848 bdrv_op_block(bs, i, reason);
6849 }
6850}
6851
6852void bdrv_op_unblock_all(BlockDriverState *bs, Error *reason)
6853{
6854 int i;
f791bf7f 6855 GLOBAL_STATE_CODE();
fbe40ff7
FZ
6856 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
6857 bdrv_op_unblock(bs, i, reason);
6858 }
6859}
6860
6861bool bdrv_op_blocker_is_empty(BlockDriverState *bs)
6862{
6863 int i;
f791bf7f 6864 GLOBAL_STATE_CODE();
fbe40ff7
FZ
6865 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
6866 if (!QLIST_EMPTY(&bs->op_blockers[i])) {
6867 return false;
6868 }
6869 }
6870 return true;
6871}
6872
d92ada22
LC
6873void bdrv_img_create(const char *filename, const char *fmt,
6874 const char *base_filename, const char *base_fmt,
9217283d
FZ
6875 char *options, uint64_t img_size, int flags, bool quiet,
6876 Error **errp)
f88e1a42 6877{
83d0521a
CL
6878 QemuOptsList *create_opts = NULL;
6879 QemuOpts *opts = NULL;
6880 const char *backing_fmt, *backing_file;
6881 int64_t size;
f88e1a42 6882 BlockDriver *drv, *proto_drv;
cc84d90f 6883 Error *local_err = NULL;
f88e1a42
JS
6884 int ret = 0;
6885
f791bf7f
EGE
6886 GLOBAL_STATE_CODE();
6887
f88e1a42
JS
6888 /* Find driver and parse its options */
6889 drv = bdrv_find_format(fmt);
6890 if (!drv) {
71c79813 6891 error_setg(errp, "Unknown file format '%s'", fmt);
d92ada22 6892 return;
f88e1a42
JS
6893 }
6894
b65a5e12 6895 proto_drv = bdrv_find_protocol(filename, true, errp);
f88e1a42 6896 if (!proto_drv) {
d92ada22 6897 return;
f88e1a42
JS
6898 }
6899
c6149724
HR
6900 if (!drv->create_opts) {
6901 error_setg(errp, "Format driver '%s' does not support image creation",
6902 drv->format_name);
6903 return;
6904 }
6905
5a5e7f8c
ML
6906 if (!proto_drv->create_opts) {
6907 error_setg(errp, "Protocol driver '%s' does not support image creation",
6908 proto_drv->format_name);
6909 return;
6910 }
6911
f6dc1c31 6912 /* Create parameter list */
c282e1fd 6913 create_opts = qemu_opts_append(create_opts, drv->create_opts);
5a5e7f8c 6914 create_opts = qemu_opts_append(create_opts, proto_drv->create_opts);
f88e1a42 6915
83d0521a 6916 opts = qemu_opts_create(create_opts, NULL, 0, &error_abort);
f88e1a42
JS
6917
6918 /* Parse -o options */
6919 if (options) {
a5f9b9df 6920 if (!qemu_opts_do_parse(opts, options, NULL, errp)) {
f88e1a42
JS
6921 goto out;
6922 }
6923 }
6924
f6dc1c31
KW
6925 if (!qemu_opt_get(opts, BLOCK_OPT_SIZE)) {
6926 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, img_size, &error_abort);
6927 } else if (img_size != UINT64_C(-1)) {
6928 error_setg(errp, "The image size must be specified only once");
6929 goto out;
6930 }
6931
f88e1a42 6932 if (base_filename) {
235e59cf 6933 if (!qemu_opt_set(opts, BLOCK_OPT_BACKING_FILE, base_filename,
3882578b 6934 NULL)) {
71c79813
LC
6935 error_setg(errp, "Backing file not supported for file format '%s'",
6936 fmt);
f88e1a42
JS
6937 goto out;
6938 }
6939 }
6940
6941 if (base_fmt) {
3882578b 6942 if (!qemu_opt_set(opts, BLOCK_OPT_BACKING_FMT, base_fmt, NULL)) {
71c79813
LC
6943 error_setg(errp, "Backing file format not supported for file "
6944 "format '%s'", fmt);
f88e1a42
JS
6945 goto out;
6946 }
6947 }
6948
83d0521a
CL
6949 backing_file = qemu_opt_get(opts, BLOCK_OPT_BACKING_FILE);
6950 if (backing_file) {
6951 if (!strcmp(filename, backing_file)) {
71c79813
LC
6952 error_setg(errp, "Error: Trying to create an image with the "
6953 "same filename as the backing file");
792da93a
JS
6954 goto out;
6955 }
975a7bd2
CK
6956 if (backing_file[0] == '\0') {
6957 error_setg(errp, "Expected backing file name, got empty string");
6958 goto out;
6959 }
792da93a
JS
6960 }
6961
83d0521a 6962 backing_fmt = qemu_opt_get(opts, BLOCK_OPT_BACKING_FMT);
f88e1a42 6963
6e6e55f5
JS
6964 /* The size for the image must always be specified, unless we have a backing
6965 * file and we have not been forbidden from opening it. */
a8b42a1c 6966 size = qemu_opt_get_size(opts, BLOCK_OPT_SIZE, img_size);
6e6e55f5
JS
6967 if (backing_file && !(flags & BDRV_O_NO_BACKING)) {
6968 BlockDriverState *bs;
645ae7d8 6969 char *full_backing;
6e6e55f5
JS
6970 int back_flags;
6971 QDict *backing_options = NULL;
6972
645ae7d8
HR
6973 full_backing =
6974 bdrv_get_full_backing_filename_from_filename(filename, backing_file,
6975 &local_err);
6e6e55f5 6976 if (local_err) {
6e6e55f5
JS
6977 goto out;
6978 }
645ae7d8 6979 assert(full_backing);
29168018 6980
d5b23994
HR
6981 /*
6982 * No need to do I/O here, which allows us to open encrypted
6983 * backing images without needing the secret
6984 */
6e6e55f5
JS
6985 back_flags = flags;
6986 back_flags &= ~(BDRV_O_RDWR | BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING);
d5b23994 6987 back_flags |= BDRV_O_NO_IO;
f88e1a42 6988
cc954f01 6989 backing_options = qdict_new();
6e6e55f5 6990 if (backing_fmt) {
6e6e55f5
JS
6991 qdict_put_str(backing_options, "driver", backing_fmt);
6992 }
cc954f01 6993 qdict_put_bool(backing_options, BDRV_OPT_FORCE_SHARE, true);
e6641719 6994
6e6e55f5
JS
6995 bs = bdrv_open(full_backing, NULL, backing_options, back_flags,
6996 &local_err);
6997 g_free(full_backing);
add8200d
EB
6998 if (!bs) {
6999 error_append_hint(&local_err, "Could not open backing image.\n");
6e6e55f5
JS
7000 goto out;
7001 } else {
d9f059aa 7002 if (!backing_fmt) {
497a30db
EB
7003 error_setg(&local_err,
7004 "Backing file specified without backing format");
7005 error_append_hint(&local_err, "Detected format of %s.",
7006 bs->drv->format_name);
7007 goto out;
d9f059aa 7008 }
6e6e55f5
JS
7009 if (size == -1) {
7010 /* Opened BS, have no size */
7011 size = bdrv_getlength(bs);
7012 if (size < 0) {
7013 error_setg_errno(errp, -size, "Could not get size of '%s'",
7014 backing_file);
7015 bdrv_unref(bs);
7016 goto out;
7017 }
7018 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, size, &error_abort);
52bf1e72 7019 }
66f6b814 7020 bdrv_unref(bs);
f88e1a42 7021 }
d9f059aa
EB
7022 /* (backing_file && !(flags & BDRV_O_NO_BACKING)) */
7023 } else if (backing_file && !backing_fmt) {
497a30db
EB
7024 error_setg(&local_err,
7025 "Backing file specified without backing format");
7026 goto out;
d9f059aa 7027 }
6e6e55f5
JS
7028
7029 if (size == -1) {
7030 error_setg(errp, "Image creation needs a size parameter");
7031 goto out;
f88e1a42
JS
7032 }
7033
f382d43a 7034 if (!quiet) {
fe646693 7035 printf("Formatting '%s', fmt=%s ", filename, fmt);
43c5d8f8 7036 qemu_opts_print(opts, " ");
f382d43a 7037 puts("");
4e2f4418 7038 fflush(stdout);
f382d43a 7039 }
83d0521a 7040
c282e1fd 7041 ret = bdrv_create(drv, filename, opts, &local_err);
83d0521a 7042
cc84d90f
HR
7043 if (ret == -EFBIG) {
7044 /* This is generally a better message than whatever the driver would
7045 * deliver (especially because of the cluster_size_hint), since that
7046 * is most probably not much different from "image too large". */
7047 const char *cluster_size_hint = "";
83d0521a 7048 if (qemu_opt_get_size(opts, BLOCK_OPT_CLUSTER_SIZE, 0)) {
cc84d90f 7049 cluster_size_hint = " (try using a larger cluster size)";
f88e1a42 7050 }
cc84d90f
HR
7051 error_setg(errp, "The image size is too large for file format '%s'"
7052 "%s", fmt, cluster_size_hint);
7053 error_free(local_err);
7054 local_err = NULL;
f88e1a42
JS
7055 }
7056
7057out:
83d0521a
CL
7058 qemu_opts_del(opts);
7059 qemu_opts_free(create_opts);
621ff94d 7060 error_propagate(errp, local_err);
f88e1a42 7061}
85d126f3
SH
7062
7063AioContext *bdrv_get_aio_context(BlockDriverState *bs)
7064{
384a48fb 7065 IO_CODE();
33f2a757 7066 return bs ? bs->aio_context : qemu_get_aio_context();
dcd04228
SH
7067}
7068
e336fd4c
KW
7069AioContext *coroutine_fn bdrv_co_enter(BlockDriverState *bs)
7070{
7071 Coroutine *self = qemu_coroutine_self();
7072 AioContext *old_ctx = qemu_coroutine_get_aio_context(self);
7073 AioContext *new_ctx;
384a48fb 7074 IO_CODE();
e336fd4c
KW
7075
7076 /*
7077 * Increase bs->in_flight to ensure that this operation is completed before
7078 * moving the node to a different AioContext. Read new_ctx only afterwards.
7079 */
7080 bdrv_inc_in_flight(bs);
7081
7082 new_ctx = bdrv_get_aio_context(bs);
7083 aio_co_reschedule_self(new_ctx);
7084 return old_ctx;
7085}
7086
7087void coroutine_fn bdrv_co_leave(BlockDriverState *bs, AioContext *old_ctx)
7088{
384a48fb 7089 IO_CODE();
e336fd4c
KW
7090 aio_co_reschedule_self(old_ctx);
7091 bdrv_dec_in_flight(bs);
7092}
7093
18c6ac1c
KW
7094void coroutine_fn bdrv_co_lock(BlockDriverState *bs)
7095{
7096 AioContext *ctx = bdrv_get_aio_context(bs);
7097
7098 /* In the main thread, bs->aio_context won't change concurrently */
7099 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
7100
7101 /*
7102 * We're in coroutine context, so we already hold the lock of the main
7103 * loop AioContext. Don't lock it twice to avoid deadlocks.
7104 */
7105 assert(qemu_in_coroutine());
7106 if (ctx != qemu_get_aio_context()) {
7107 aio_context_acquire(ctx);
7108 }
7109}
7110
7111void coroutine_fn bdrv_co_unlock(BlockDriverState *bs)
7112{
7113 AioContext *ctx = bdrv_get_aio_context(bs);
7114
7115 assert(qemu_in_coroutine());
7116 if (ctx != qemu_get_aio_context()) {
7117 aio_context_release(ctx);
7118 }
7119}
7120
052a7572
FZ
7121void bdrv_coroutine_enter(BlockDriverState *bs, Coroutine *co)
7122{
384a48fb 7123 IO_CODE();
052a7572
FZ
7124 aio_co_enter(bdrv_get_aio_context(bs), co);
7125}
7126
e8a095da
SH
7127static void bdrv_do_remove_aio_context_notifier(BdrvAioNotifier *ban)
7128{
bdb73476 7129 GLOBAL_STATE_CODE();
e8a095da
SH
7130 QLIST_REMOVE(ban, list);
7131 g_free(ban);
7132}
7133
a3a683c3 7134static void bdrv_detach_aio_context(BlockDriverState *bs)
dcd04228 7135{
e8a095da 7136 BdrvAioNotifier *baf, *baf_tmp;
33384421 7137
e8a095da 7138 assert(!bs->walking_aio_notifiers);
da359909 7139 GLOBAL_STATE_CODE();
e8a095da
SH
7140 bs->walking_aio_notifiers = true;
7141 QLIST_FOREACH_SAFE(baf, &bs->aio_notifiers, list, baf_tmp) {
7142 if (baf->deleted) {
7143 bdrv_do_remove_aio_context_notifier(baf);
7144 } else {
7145 baf->detach_aio_context(baf->opaque);
7146 }
33384421 7147 }
e8a095da
SH
7148 /* Never mind iterating again to check for ->deleted. bdrv_close() will
7149 * remove remaining aio notifiers if we aren't called again.
7150 */
7151 bs->walking_aio_notifiers = false;
33384421 7152
1bffe1ae 7153 if (bs->drv && bs->drv->bdrv_detach_aio_context) {
dcd04228
SH
7154 bs->drv->bdrv_detach_aio_context(bs);
7155 }
dcd04228 7156
e64f25f3
KW
7157 if (bs->quiesce_counter) {
7158 aio_enable_external(bs->aio_context);
7159 }
7f898610 7160 assert_bdrv_graph_writable(bs);
dcd04228
SH
7161 bs->aio_context = NULL;
7162}
7163
a3a683c3
KW
7164static void bdrv_attach_aio_context(BlockDriverState *bs,
7165 AioContext *new_context)
dcd04228 7166{
e8a095da 7167 BdrvAioNotifier *ban, *ban_tmp;
da359909 7168 GLOBAL_STATE_CODE();
33384421 7169
e64f25f3
KW
7170 if (bs->quiesce_counter) {
7171 aio_disable_external(new_context);
7172 }
7173
7f898610 7174 assert_bdrv_graph_writable(bs);
dcd04228
SH
7175 bs->aio_context = new_context;
7176
1bffe1ae 7177 if (bs->drv && bs->drv->bdrv_attach_aio_context) {
dcd04228
SH
7178 bs->drv->bdrv_attach_aio_context(bs, new_context);
7179 }
33384421 7180
e8a095da
SH
7181 assert(!bs->walking_aio_notifiers);
7182 bs->walking_aio_notifiers = true;
7183 QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_tmp) {
7184 if (ban->deleted) {
7185 bdrv_do_remove_aio_context_notifier(ban);
7186 } else {
7187 ban->attached_aio_context(new_context, ban->opaque);
7188 }
33384421 7189 }
e8a095da 7190 bs->walking_aio_notifiers = false;
dcd04228
SH
7191}
7192
42a65f02
KW
7193/*
7194 * Changes the AioContext used for fd handlers, timers, and BHs by this
7195 * BlockDriverState and all its children and parents.
7196 *
43eaaaef
HR
7197 * Must be called from the main AioContext.
7198 *
42a65f02
KW
7199 * The caller must own the AioContext lock for the old AioContext of bs, but it
7200 * must not own the AioContext lock for new_context (unless new_context is the
7201 * same as the current context of bs).
7202 *
7e8c182f 7203 * @ignore will accumulate all visited BdrvChild objects. The caller is
42a65f02
KW
7204 * responsible for freeing the list afterwards.
7205 */
53a7d041
KW
7206void bdrv_set_aio_context_ignore(BlockDriverState *bs,
7207 AioContext *new_context, GSList **ignore)
dcd04228 7208{
e037c09c 7209 AioContext *old_context = bdrv_get_aio_context(bs);
722d8e73
SL
7210 GSList *children_to_process = NULL;
7211 GSList *parents_to_process = NULL;
7212 GSList *entry;
7213 BdrvChild *child, *parent;
0d83708a 7214
43eaaaef 7215 g_assert(qemu_get_current_aio_context() == qemu_get_aio_context());
f0c28327 7216 GLOBAL_STATE_CODE();
43eaaaef 7217
e037c09c 7218 if (old_context == new_context) {
57830a49
DP
7219 return;
7220 }
7221
d70d5954 7222 bdrv_drained_begin(bs);
0d83708a
KW
7223
7224 QLIST_FOREACH(child, &bs->children, next) {
53a7d041
KW
7225 if (g_slist_find(*ignore, child)) {
7226 continue;
7227 }
7228 *ignore = g_slist_prepend(*ignore, child);
722d8e73 7229 children_to_process = g_slist_prepend(children_to_process, child);
53a7d041 7230 }
722d8e73
SL
7231
7232 QLIST_FOREACH(parent, &bs->parents, next_parent) {
7233 if (g_slist_find(*ignore, parent)) {
53a7d041
KW
7234 continue;
7235 }
722d8e73
SL
7236 *ignore = g_slist_prepend(*ignore, parent);
7237 parents_to_process = g_slist_prepend(parents_to_process, parent);
7238 }
7239
7240 for (entry = children_to_process;
7241 entry != NULL;
7242 entry = g_slist_next(entry)) {
7243 child = entry->data;
7244 bdrv_set_aio_context_ignore(child->bs, new_context, ignore);
7245 }
7246 g_slist_free(children_to_process);
7247
7248 for (entry = parents_to_process;
7249 entry != NULL;
7250 entry = g_slist_next(entry)) {
7251 parent = entry->data;
7252 assert(parent->klass->set_aio_ctx);
7253 parent->klass->set_aio_ctx(parent, new_context, ignore);
0d83708a 7254 }
722d8e73 7255 g_slist_free(parents_to_process);
0d83708a 7256
dcd04228
SH
7257 bdrv_detach_aio_context(bs);
7258
e037c09c 7259 /* Acquire the new context, if necessary */
43eaaaef 7260 if (qemu_get_aio_context() != new_context) {
e037c09c
HR
7261 aio_context_acquire(new_context);
7262 }
7263
dcd04228 7264 bdrv_attach_aio_context(bs, new_context);
e037c09c
HR
7265
7266 /*
7267 * If this function was recursively called from
7268 * bdrv_set_aio_context_ignore(), there may be nodes in the
7269 * subtree that have not yet been moved to the new AioContext.
7270 * Release the old one so bdrv_drained_end() can poll them.
7271 */
43eaaaef 7272 if (qemu_get_aio_context() != old_context) {
e037c09c
HR
7273 aio_context_release(old_context);
7274 }
7275
d70d5954 7276 bdrv_drained_end(bs);
e037c09c 7277
43eaaaef 7278 if (qemu_get_aio_context() != old_context) {
e037c09c
HR
7279 aio_context_acquire(old_context);
7280 }
43eaaaef 7281 if (qemu_get_aio_context() != new_context) {
e037c09c
HR
7282 aio_context_release(new_context);
7283 }
85d126f3 7284}
d616b224 7285
5d231849
KW
7286static bool bdrv_parent_can_set_aio_context(BdrvChild *c, AioContext *ctx,
7287 GSList **ignore, Error **errp)
7288{
f0c28327 7289 GLOBAL_STATE_CODE();
5d231849
KW
7290 if (g_slist_find(*ignore, c)) {
7291 return true;
7292 }
7293 *ignore = g_slist_prepend(*ignore, c);
7294
bd86fb99
HR
7295 /*
7296 * A BdrvChildClass that doesn't handle AioContext changes cannot
7297 * tolerate any AioContext changes
7298 */
7299 if (!c->klass->can_set_aio_ctx) {
5d231849
KW
7300 char *user = bdrv_child_user_desc(c);
7301 error_setg(errp, "Changing iothreads is not supported by %s", user);
7302 g_free(user);
7303 return false;
7304 }
bd86fb99 7305 if (!c->klass->can_set_aio_ctx(c, ctx, ignore, errp)) {
5d231849
KW
7306 assert(!errp || *errp);
7307 return false;
7308 }
7309 return true;
7310}
7311
7e8c182f
EGE
7312typedef struct BdrvStateSetAioContext {
7313 AioContext *new_ctx;
7314 BlockDriverState *bs;
7315} BdrvStateSetAioContext;
7316
7317static bool bdrv_parent_change_aio_context(BdrvChild *c, AioContext *ctx,
7318 GSList **visited, Transaction *tran,
7319 Error **errp)
7320{
7321 GLOBAL_STATE_CODE();
7322 if (g_slist_find(*visited, c)) {
7323 return true;
7324 }
7325 *visited = g_slist_prepend(*visited, c);
7326
7327 /*
7328 * A BdrvChildClass that doesn't handle AioContext changes cannot
7329 * tolerate any AioContext changes
7330 */
7331 if (!c->klass->change_aio_ctx) {
7332 char *user = bdrv_child_user_desc(c);
7333 error_setg(errp, "Changing iothreads is not supported by %s", user);
7334 g_free(user);
7335 return false;
7336 }
7337 if (!c->klass->change_aio_ctx(c, ctx, visited, tran, errp)) {
7338 assert(!errp || *errp);
7339 return false;
7340 }
7341 return true;
7342}
7343
5d231849
KW
7344bool bdrv_child_can_set_aio_context(BdrvChild *c, AioContext *ctx,
7345 GSList **ignore, Error **errp)
7346{
f791bf7f 7347 GLOBAL_STATE_CODE();
5d231849
KW
7348 if (g_slist_find(*ignore, c)) {
7349 return true;
7350 }
7351 *ignore = g_slist_prepend(*ignore, c);
7352 return bdrv_can_set_aio_context(c->bs, ctx, ignore, errp);
7353}
7354
7e8c182f
EGE
7355bool bdrv_child_change_aio_context(BdrvChild *c, AioContext *ctx,
7356 GSList **visited, Transaction *tran,
7357 Error **errp)
7358{
7359 GLOBAL_STATE_CODE();
7360 if (g_slist_find(*visited, c)) {
7361 return true;
7362 }
7363 *visited = g_slist_prepend(*visited, c);
7364 return bdrv_change_aio_context(c->bs, ctx, visited, tran, errp);
7365}
7366
5d231849
KW
7367/* @ignore will accumulate all visited BdrvChild object. The caller is
7368 * responsible for freeing the list afterwards. */
7369bool bdrv_can_set_aio_context(BlockDriverState *bs, AioContext *ctx,
7370 GSList **ignore, Error **errp)
7371{
7372 BdrvChild *c;
7373
7374 if (bdrv_get_aio_context(bs) == ctx) {
7375 return true;
7376 }
7377
f791bf7f
EGE
7378 GLOBAL_STATE_CODE();
7379
5d231849
KW
7380 QLIST_FOREACH(c, &bs->parents, next_parent) {
7381 if (!bdrv_parent_can_set_aio_context(c, ctx, ignore, errp)) {
7382 return false;
7383 }
7384 }
7385 QLIST_FOREACH(c, &bs->children, next) {
7386 if (!bdrv_child_can_set_aio_context(c, ctx, ignore, errp)) {
7387 return false;
7388 }
7389 }
7390
7391 return true;
7392}
7393
7e8c182f
EGE
7394static void bdrv_set_aio_context_clean(void *opaque)
7395{
7396 BdrvStateSetAioContext *state = (BdrvStateSetAioContext *) opaque;
7397 BlockDriverState *bs = (BlockDriverState *) state->bs;
7398
7399 /* Paired with bdrv_drained_begin in bdrv_change_aio_context() */
7400 bdrv_drained_end(bs);
7401
7402 g_free(state);
7403}
7404
7405static void bdrv_set_aio_context_commit(void *opaque)
7406{
7407 BdrvStateSetAioContext *state = (BdrvStateSetAioContext *) opaque;
7408 BlockDriverState *bs = (BlockDriverState *) state->bs;
7409 AioContext *new_context = state->new_ctx;
7410 AioContext *old_context = bdrv_get_aio_context(bs);
7411 assert_bdrv_graph_writable(bs);
7412
7413 /*
7414 * Take the old AioContex when detaching it from bs.
7415 * At this point, new_context lock is already acquired, and we are now
7416 * also taking old_context. This is safe as long as bdrv_detach_aio_context
7417 * does not call AIO_POLL_WHILE().
7418 */
7419 if (old_context != qemu_get_aio_context()) {
7420 aio_context_acquire(old_context);
7421 }
7422 bdrv_detach_aio_context(bs);
7423 if (old_context != qemu_get_aio_context()) {
7424 aio_context_release(old_context);
7425 }
7426 bdrv_attach_aio_context(bs, new_context);
7427}
7428
7429static TransactionActionDrv set_aio_context = {
7430 .commit = bdrv_set_aio_context_commit,
7431 .clean = bdrv_set_aio_context_clean,
7432};
7433
7434/*
7435 * Changes the AioContext used for fd handlers, timers, and BHs by this
7436 * BlockDriverState and all its children and parents.
7437 *
7438 * Must be called from the main AioContext.
7439 *
7440 * The caller must own the AioContext lock for the old AioContext of bs, but it
7441 * must not own the AioContext lock for new_context (unless new_context is the
7442 * same as the current context of bs).
7443 *
7444 * @visited will accumulate all visited BdrvChild objects. The caller is
7445 * responsible for freeing the list afterwards.
7446 */
7447static bool bdrv_change_aio_context(BlockDriverState *bs, AioContext *ctx,
7448 GSList **visited, Transaction *tran,
7449 Error **errp)
7450{
7451 BdrvChild *c;
7452 BdrvStateSetAioContext *state;
7453
7454 GLOBAL_STATE_CODE();
7455
7456 if (bdrv_get_aio_context(bs) == ctx) {
7457 return true;
7458 }
7459
7460 QLIST_FOREACH(c, &bs->parents, next_parent) {
7461 if (!bdrv_parent_change_aio_context(c, ctx, visited, tran, errp)) {
7462 return false;
7463 }
7464 }
7465
7466 QLIST_FOREACH(c, &bs->children, next) {
7467 if (!bdrv_child_change_aio_context(c, ctx, visited, tran, errp)) {
7468 return false;
7469 }
7470 }
7471
7472 state = g_new(BdrvStateSetAioContext, 1);
7473 *state = (BdrvStateSetAioContext) {
7474 .new_ctx = ctx,
7475 .bs = bs,
7476 };
7477
7478 /* Paired with bdrv_drained_end in bdrv_set_aio_context_clean() */
7479 bdrv_drained_begin(bs);
7480
7481 tran_add(tran, &set_aio_context, state);
7482
7483 return true;
7484}
7485
5d231849
KW
7486int bdrv_child_try_set_aio_context(BlockDriverState *bs, AioContext *ctx,
7487 BdrvChild *ignore_child, Error **errp)
7488{
7489 GSList *ignore;
7490 bool ret;
7491
f791bf7f
EGE
7492 GLOBAL_STATE_CODE();
7493
5d231849
KW
7494 ignore = ignore_child ? g_slist_prepend(NULL, ignore_child) : NULL;
7495 ret = bdrv_can_set_aio_context(bs, ctx, &ignore, errp);
7496 g_slist_free(ignore);
7497
7498 if (!ret) {
7499 return -EPERM;
7500 }
7501
53a7d041
KW
7502 ignore = ignore_child ? g_slist_prepend(NULL, ignore_child) : NULL;
7503 bdrv_set_aio_context_ignore(bs, ctx, &ignore);
7504 g_slist_free(ignore);
7505
5d231849
KW
7506 return 0;
7507}
7508
7e8c182f
EGE
7509/*
7510 * Change bs's and recursively all of its parents' and children's AioContext
7511 * to the given new context, returning an error if that isn't possible.
7512 *
7513 * If ignore_child is not NULL, that child (and its subgraph) will not
7514 * be touched.
7515 *
7516 * This function still requires the caller to take the bs current
7517 * AioContext lock, otherwise draining will fail since AIO_WAIT_WHILE
7518 * assumes the lock is always held if bs is in another AioContext.
7519 * For the same reason, it temporarily also holds the new AioContext, since
7520 * bdrv_drained_end calls BDRV_POLL_WHILE that assumes the lock is taken too.
7521 * Therefore the new AioContext lock must not be taken by the caller.
7522 */
7523int bdrv_child_try_change_aio_context(BlockDriverState *bs, AioContext *ctx,
7524 BdrvChild *ignore_child, Error **errp)
7525{
7526 Transaction *tran;
7527 GSList *visited;
7528 int ret;
7529 AioContext *old_context = bdrv_get_aio_context(bs);
7530 GLOBAL_STATE_CODE();
7531
7532 /*
7533 * Recursion phase: go through all nodes of the graph.
7534 * Take care of checking that all nodes support changing AioContext
7535 * and drain them, builing a linear list of callbacks to run if everything
7536 * is successful (the transaction itself).
7537 */
7538 tran = tran_new();
7539 visited = ignore_child ? g_slist_prepend(NULL, ignore_child) : NULL;
7540 ret = bdrv_change_aio_context(bs, ctx, &visited, tran, errp);
7541 g_slist_free(visited);
7542
7543 /*
7544 * Linear phase: go through all callbacks collected in the transaction.
7545 * Run all callbacks collected in the recursion to switch all nodes
7546 * AioContext lock (transaction commit), or undo all changes done in the
7547 * recursion (transaction abort).
7548 */
7549
7550 if (!ret) {
7551 /* Just run clean() callbacks. No AioContext changed. */
7552 tran_abort(tran);
7553 return -EPERM;
7554 }
7555
7556 /*
7557 * Release old AioContext, it won't be needed anymore, as all
7558 * bdrv_drained_begin() have been called already.
7559 */
7560 if (qemu_get_aio_context() != old_context) {
7561 aio_context_release(old_context);
7562 }
7563
7564 /*
7565 * Acquire new AioContext since bdrv_drained_end() is going to be called
7566 * after we switched all nodes in the new AioContext, and the function
7567 * assumes that the lock of the bs is always taken.
7568 */
7569 if (qemu_get_aio_context() != ctx) {
7570 aio_context_acquire(ctx);
7571 }
7572
7573 tran_commit(tran);
7574
7575 if (qemu_get_aio_context() != ctx) {
7576 aio_context_release(ctx);
7577 }
7578
7579 /* Re-acquire the old AioContext, since the caller takes and releases it. */
7580 if (qemu_get_aio_context() != old_context) {
7581 aio_context_acquire(old_context);
7582 }
7583
7584 return 0;
7585}
7586
5d231849
KW
7587int bdrv_try_set_aio_context(BlockDriverState *bs, AioContext *ctx,
7588 Error **errp)
7589{
f791bf7f 7590 GLOBAL_STATE_CODE();
5d231849
KW
7591 return bdrv_child_try_set_aio_context(bs, ctx, NULL, errp);
7592}
7593
33384421
HR
7594void bdrv_add_aio_context_notifier(BlockDriverState *bs,
7595 void (*attached_aio_context)(AioContext *new_context, void *opaque),
7596 void (*detach_aio_context)(void *opaque), void *opaque)
7597{
7598 BdrvAioNotifier *ban = g_new(BdrvAioNotifier, 1);
7599 *ban = (BdrvAioNotifier){
7600 .attached_aio_context = attached_aio_context,
7601 .detach_aio_context = detach_aio_context,
7602 .opaque = opaque
7603 };
f791bf7f 7604 GLOBAL_STATE_CODE();
33384421
HR
7605
7606 QLIST_INSERT_HEAD(&bs->aio_notifiers, ban, list);
7607}
7608
7609void bdrv_remove_aio_context_notifier(BlockDriverState *bs,
7610 void (*attached_aio_context)(AioContext *,
7611 void *),
7612 void (*detach_aio_context)(void *),
7613 void *opaque)
7614{
7615 BdrvAioNotifier *ban, *ban_next;
f791bf7f 7616 GLOBAL_STATE_CODE();
33384421
HR
7617
7618 QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_next) {
7619 if (ban->attached_aio_context == attached_aio_context &&
7620 ban->detach_aio_context == detach_aio_context &&
e8a095da
SH
7621 ban->opaque == opaque &&
7622 ban->deleted == false)
33384421 7623 {
e8a095da
SH
7624 if (bs->walking_aio_notifiers) {
7625 ban->deleted = true;
7626 } else {
7627 bdrv_do_remove_aio_context_notifier(ban);
7628 }
33384421
HR
7629 return;
7630 }
7631 }
7632
7633 abort();
7634}
7635
77485434 7636int bdrv_amend_options(BlockDriverState *bs, QemuOpts *opts,
d1402b50 7637 BlockDriverAmendStatusCB *status_cb, void *cb_opaque,
a3579bfa 7638 bool force,
d1402b50 7639 Error **errp)
6f176b48 7640{
f791bf7f 7641 GLOBAL_STATE_CODE();
d470ad42 7642 if (!bs->drv) {
d1402b50 7643 error_setg(errp, "Node is ejected");
d470ad42
HR
7644 return -ENOMEDIUM;
7645 }
c282e1fd 7646 if (!bs->drv->bdrv_amend_options) {
d1402b50
HR
7647 error_setg(errp, "Block driver '%s' does not support option amendment",
7648 bs->drv->format_name);
6f176b48
HR
7649 return -ENOTSUP;
7650 }
a3579bfa
ML
7651 return bs->drv->bdrv_amend_options(bs, opts, status_cb,
7652 cb_opaque, force, errp);
6f176b48 7653}
f6186f49 7654
5d69b5ab
HR
7655/*
7656 * This function checks whether the given @to_replace is allowed to be
7657 * replaced by a node that always shows the same data as @bs. This is
7658 * used for example to verify whether the mirror job can replace
7659 * @to_replace by the target mirrored from @bs.
7660 * To be replaceable, @bs and @to_replace may either be guaranteed to
7661 * always show the same data (because they are only connected through
7662 * filters), or some driver may allow replacing one of its children
7663 * because it can guarantee that this child's data is not visible at
7664 * all (for example, for dissenting quorum children that have no other
7665 * parents).
7666 */
7667bool bdrv_recurse_can_replace(BlockDriverState *bs,
7668 BlockDriverState *to_replace)
7669{
93393e69
HR
7670 BlockDriverState *filtered;
7671
b4ad82aa
EGE
7672 GLOBAL_STATE_CODE();
7673
5d69b5ab
HR
7674 if (!bs || !bs->drv) {
7675 return false;
7676 }
7677
7678 if (bs == to_replace) {
7679 return true;
7680 }
7681
7682 /* See what the driver can do */
7683 if (bs->drv->bdrv_recurse_can_replace) {
7684 return bs->drv->bdrv_recurse_can_replace(bs, to_replace);
7685 }
7686
7687 /* For filters without an own implementation, we can recurse on our own */
93393e69
HR
7688 filtered = bdrv_filter_bs(bs);
7689 if (filtered) {
7690 return bdrv_recurse_can_replace(filtered, to_replace);
5d69b5ab
HR
7691 }
7692
7693 /* Safe default */
7694 return false;
7695}
7696
810803a8
HR
7697/*
7698 * Check whether the given @node_name can be replaced by a node that
7699 * has the same data as @parent_bs. If so, return @node_name's BDS;
7700 * NULL otherwise.
7701 *
7702 * @node_name must be a (recursive) *child of @parent_bs (or this
7703 * function will return NULL).
7704 *
7705 * The result (whether the node can be replaced or not) is only valid
7706 * for as long as no graph or permission changes occur.
7707 */
e12f3784
WC
7708BlockDriverState *check_to_replace_node(BlockDriverState *parent_bs,
7709 const char *node_name, Error **errp)
09158f00
BC
7710{
7711 BlockDriverState *to_replace_bs = bdrv_find_node(node_name);
5a7e7a0b
SH
7712 AioContext *aio_context;
7713
f791bf7f
EGE
7714 GLOBAL_STATE_CODE();
7715
09158f00 7716 if (!to_replace_bs) {
785ec4b1 7717 error_setg(errp, "Failed to find node with node-name='%s'", node_name);
09158f00
BC
7718 return NULL;
7719 }
7720
5a7e7a0b
SH
7721 aio_context = bdrv_get_aio_context(to_replace_bs);
7722 aio_context_acquire(aio_context);
7723
09158f00 7724 if (bdrv_op_is_blocked(to_replace_bs, BLOCK_OP_TYPE_REPLACE, errp)) {
5a7e7a0b
SH
7725 to_replace_bs = NULL;
7726 goto out;
09158f00
BC
7727 }
7728
7729 /* We don't want arbitrary node of the BDS chain to be replaced only the top
7730 * most non filter in order to prevent data corruption.
7731 * Another benefit is that this tests exclude backing files which are
7732 * blocked by the backing blockers.
7733 */
810803a8
HR
7734 if (!bdrv_recurse_can_replace(parent_bs, to_replace_bs)) {
7735 error_setg(errp, "Cannot replace '%s' by a node mirrored from '%s', "
7736 "because it cannot be guaranteed that doing so would not "
7737 "lead to an abrupt change of visible data",
7738 node_name, parent_bs->node_name);
5a7e7a0b
SH
7739 to_replace_bs = NULL;
7740 goto out;
09158f00
BC
7741 }
7742
5a7e7a0b
SH
7743out:
7744 aio_context_release(aio_context);
09158f00
BC
7745 return to_replace_bs;
7746}
448ad91d 7747
97e2f021
HR
7748/**
7749 * Iterates through the list of runtime option keys that are said to
7750 * be "strong" for a BDS. An option is called "strong" if it changes
7751 * a BDS's data. For example, the null block driver's "size" and
7752 * "read-zeroes" options are strong, but its "latency-ns" option is
7753 * not.
7754 *
7755 * If a key returned by this function ends with a dot, all options
7756 * starting with that prefix are strong.
7757 */
7758static const char *const *strong_options(BlockDriverState *bs,
7759 const char *const *curopt)
7760{
7761 static const char *const global_options[] = {
7762 "driver", "filename", NULL
7763 };
7764
7765 if (!curopt) {
7766 return &global_options[0];
7767 }
7768
7769 curopt++;
7770 if (curopt == &global_options[ARRAY_SIZE(global_options) - 1] && bs->drv) {
7771 curopt = bs->drv->strong_runtime_opts;
7772 }
7773
7774 return (curopt && *curopt) ? curopt : NULL;
7775}
7776
7777/**
7778 * Copies all strong runtime options from bs->options to the given
7779 * QDict. The set of strong option keys is determined by invoking
7780 * strong_options().
7781 *
7782 * Returns true iff any strong option was present in bs->options (and
7783 * thus copied to the target QDict) with the exception of "filename"
7784 * and "driver". The caller is expected to use this value to decide
7785 * whether the existence of strong options prevents the generation of
7786 * a plain filename.
7787 */
7788static bool append_strong_runtime_options(QDict *d, BlockDriverState *bs)
7789{
7790 bool found_any = false;
7791 const char *const *option_name = NULL;
7792
7793 if (!bs->drv) {
7794 return false;
7795 }
7796
7797 while ((option_name = strong_options(bs, option_name))) {
7798 bool option_given = false;
7799
7800 assert(strlen(*option_name) > 0);
7801 if ((*option_name)[strlen(*option_name) - 1] != '.') {
7802 QObject *entry = qdict_get(bs->options, *option_name);
7803 if (!entry) {
7804 continue;
7805 }
7806
7807 qdict_put_obj(d, *option_name, qobject_ref(entry));
7808 option_given = true;
7809 } else {
7810 const QDictEntry *entry;
7811 for (entry = qdict_first(bs->options); entry;
7812 entry = qdict_next(bs->options, entry))
7813 {
7814 if (strstart(qdict_entry_key(entry), *option_name, NULL)) {
7815 qdict_put_obj(d, qdict_entry_key(entry),
7816 qobject_ref(qdict_entry_value(entry)));
7817 option_given = true;
7818 }
7819 }
7820 }
7821
7822 /* While "driver" and "filename" need to be included in a JSON filename,
7823 * their existence does not prohibit generation of a plain filename. */
7824 if (!found_any && option_given &&
7825 strcmp(*option_name, "driver") && strcmp(*option_name, "filename"))
7826 {
7827 found_any = true;
7828 }
7829 }
7830
62a01a27
HR
7831 if (!qdict_haskey(d, "driver")) {
7832 /* Drivers created with bdrv_new_open_driver() may not have a
7833 * @driver option. Add it here. */
7834 qdict_put_str(d, "driver", bs->drv->format_name);
7835 }
7836
97e2f021
HR
7837 return found_any;
7838}
7839
90993623
HR
7840/* Note: This function may return false positives; it may return true
7841 * even if opening the backing file specified by bs's image header
7842 * would result in exactly bs->backing. */
fa8fc1d0 7843static bool bdrv_backing_overridden(BlockDriverState *bs)
90993623 7844{
b4ad82aa 7845 GLOBAL_STATE_CODE();
90993623
HR
7846 if (bs->backing) {
7847 return strcmp(bs->auto_backing_file,
7848 bs->backing->bs->filename);
7849 } else {
7850 /* No backing BDS, so if the image header reports any backing
7851 * file, it must have been suppressed */
7852 return bs->auto_backing_file[0] != '\0';
7853 }
7854}
7855
91af7014
HR
7856/* Updates the following BDS fields:
7857 * - exact_filename: A filename which may be used for opening a block device
7858 * which (mostly) equals the given BDS (even without any
7859 * other options; so reading and writing must return the same
7860 * results, but caching etc. may be different)
7861 * - full_open_options: Options which, when given when opening a block device
7862 * (without a filename), result in a BDS (mostly)
7863 * equalling the given one
7864 * - filename: If exact_filename is set, it is copied here. Otherwise,
7865 * full_open_options is converted to a JSON object, prefixed with
7866 * "json:" (for use through the JSON pseudo protocol) and put here.
7867 */
7868void bdrv_refresh_filename(BlockDriverState *bs)
7869{
7870 BlockDriver *drv = bs->drv;
e24518e3 7871 BdrvChild *child;
52f72d6f 7872 BlockDriverState *primary_child_bs;
91af7014 7873 QDict *opts;
90993623 7874 bool backing_overridden;
998b3a1e
HR
7875 bool generate_json_filename; /* Whether our default implementation should
7876 fill exact_filename (false) or not (true) */
91af7014 7877
f791bf7f
EGE
7878 GLOBAL_STATE_CODE();
7879
91af7014
HR
7880 if (!drv) {
7881 return;
7882 }
7883
e24518e3
HR
7884 /* This BDS's file name may depend on any of its children's file names, so
7885 * refresh those first */
7886 QLIST_FOREACH(child, &bs->children, next) {
7887 bdrv_refresh_filename(child->bs);
91af7014
HR
7888 }
7889
bb808d5f
HR
7890 if (bs->implicit) {
7891 /* For implicit nodes, just copy everything from the single child */
7892 child = QLIST_FIRST(&bs->children);
7893 assert(QLIST_NEXT(child, next) == NULL);
7894
7895 pstrcpy(bs->exact_filename, sizeof(bs->exact_filename),
7896 child->bs->exact_filename);
7897 pstrcpy(bs->filename, sizeof(bs->filename), child->bs->filename);
7898
cb895614 7899 qobject_unref(bs->full_open_options);
bb808d5f
HR
7900 bs->full_open_options = qobject_ref(child->bs->full_open_options);
7901
7902 return;
7903 }
7904
90993623
HR
7905 backing_overridden = bdrv_backing_overridden(bs);
7906
7907 if (bs->open_flags & BDRV_O_NO_IO) {
7908 /* Without I/O, the backing file does not change anything.
7909 * Therefore, in such a case (primarily qemu-img), we can
7910 * pretend the backing file has not been overridden even if
7911 * it technically has been. */
7912 backing_overridden = false;
7913 }
7914
97e2f021
HR
7915 /* Gather the options QDict */
7916 opts = qdict_new();
998b3a1e
HR
7917 generate_json_filename = append_strong_runtime_options(opts, bs);
7918 generate_json_filename |= backing_overridden;
97e2f021
HR
7919
7920 if (drv->bdrv_gather_child_options) {
7921 /* Some block drivers may not want to present all of their children's
7922 * options, or name them differently from BdrvChild.name */
7923 drv->bdrv_gather_child_options(bs, opts, backing_overridden);
7924 } else {
7925 QLIST_FOREACH(child, &bs->children, next) {
25191e5f 7926 if (child == bs->backing && !backing_overridden) {
97e2f021
HR
7927 /* We can skip the backing BDS if it has not been overridden */
7928 continue;
7929 }
7930
7931 qdict_put(opts, child->name,
7932 qobject_ref(child->bs->full_open_options));
7933 }
7934
7935 if (backing_overridden && !bs->backing) {
7936 /* Force no backing file */
7937 qdict_put_null(opts, "backing");
7938 }
7939 }
7940
7941 qobject_unref(bs->full_open_options);
7942 bs->full_open_options = opts;
7943
52f72d6f
HR
7944 primary_child_bs = bdrv_primary_bs(bs);
7945
998b3a1e
HR
7946 if (drv->bdrv_refresh_filename) {
7947 /* Obsolete information is of no use here, so drop the old file name
7948 * information before refreshing it */
7949 bs->exact_filename[0] = '\0';
7950
7951 drv->bdrv_refresh_filename(bs);
52f72d6f
HR
7952 } else if (primary_child_bs) {
7953 /*
7954 * Try to reconstruct valid information from the underlying
7955 * file -- this only works for format nodes (filter nodes
7956 * cannot be probed and as such must be selected by the user
7957 * either through an options dict, or through a special
7958 * filename which the filter driver must construct in its
7959 * .bdrv_refresh_filename() implementation).
7960 */
998b3a1e
HR
7961
7962 bs->exact_filename[0] = '\0';
7963
fb695c74
HR
7964 /*
7965 * We can use the underlying file's filename if:
7966 * - it has a filename,
52f72d6f 7967 * - the current BDS is not a filter,
fb695c74
HR
7968 * - the file is a protocol BDS, and
7969 * - opening that file (as this BDS's format) will automatically create
7970 * the BDS tree we have right now, that is:
7971 * - the user did not significantly change this BDS's behavior with
7972 * some explicit (strong) options
7973 * - no non-file child of this BDS has been overridden by the user
7974 * Both of these conditions are represented by generate_json_filename.
7975 */
52f72d6f
HR
7976 if (primary_child_bs->exact_filename[0] &&
7977 primary_child_bs->drv->bdrv_file_open &&
7978 !drv->is_filter && !generate_json_filename)
fb695c74 7979 {
52f72d6f 7980 strcpy(bs->exact_filename, primary_child_bs->exact_filename);
998b3a1e
HR
7981 }
7982 }
7983
91af7014
HR
7984 if (bs->exact_filename[0]) {
7985 pstrcpy(bs->filename, sizeof(bs->filename), bs->exact_filename);
97e2f021 7986 } else {
eab3a467 7987 GString *json = qobject_to_json(QOBJECT(bs->full_open_options));
5c86bdf1 7988 if (snprintf(bs->filename, sizeof(bs->filename), "json:%s",
eab3a467 7989 json->str) >= sizeof(bs->filename)) {
5c86bdf1
EB
7990 /* Give user a hint if we truncated things. */
7991 strcpy(bs->filename + sizeof(bs->filename) - 4, "...");
7992 }
eab3a467 7993 g_string_free(json, true);
91af7014
HR
7994 }
7995}
e06018ad 7996
1e89d0f9
HR
7997char *bdrv_dirname(BlockDriverState *bs, Error **errp)
7998{
7999 BlockDriver *drv = bs->drv;
52f72d6f 8000 BlockDriverState *child_bs;
1e89d0f9 8001
f791bf7f
EGE
8002 GLOBAL_STATE_CODE();
8003
1e89d0f9
HR
8004 if (!drv) {
8005 error_setg(errp, "Node '%s' is ejected", bs->node_name);
8006 return NULL;
8007 }
8008
8009 if (drv->bdrv_dirname) {
8010 return drv->bdrv_dirname(bs, errp);
8011 }
8012
52f72d6f
HR
8013 child_bs = bdrv_primary_bs(bs);
8014 if (child_bs) {
8015 return bdrv_dirname(child_bs, errp);
1e89d0f9
HR
8016 }
8017
8018 bdrv_refresh_filename(bs);
8019 if (bs->exact_filename[0] != '\0') {
8020 return path_combine(bs->exact_filename, "");
8021 }
8022
8023 error_setg(errp, "Cannot generate a base directory for %s nodes",
8024 drv->format_name);
8025 return NULL;
8026}
8027
e06018ad
WC
8028/*
8029 * Hot add/remove a BDS's child. So the user can take a child offline when
8030 * it is broken and take a new child online
8031 */
8032void bdrv_add_child(BlockDriverState *parent_bs, BlockDriverState *child_bs,
8033 Error **errp)
8034{
f791bf7f 8035 GLOBAL_STATE_CODE();
e06018ad
WC
8036 if (!parent_bs->drv || !parent_bs->drv->bdrv_add_child) {
8037 error_setg(errp, "The node %s does not support adding a child",
8038 bdrv_get_device_or_node_name(parent_bs));
8039 return;
8040 }
8041
8042 if (!QLIST_EMPTY(&child_bs->parents)) {
8043 error_setg(errp, "The node %s already has a parent",
8044 child_bs->node_name);
8045 return;
8046 }
8047
8048 parent_bs->drv->bdrv_add_child(parent_bs, child_bs, errp);
8049}
8050
8051void bdrv_del_child(BlockDriverState *parent_bs, BdrvChild *child, Error **errp)
8052{
8053 BdrvChild *tmp;
8054
f791bf7f 8055 GLOBAL_STATE_CODE();
e06018ad
WC
8056 if (!parent_bs->drv || !parent_bs->drv->bdrv_del_child) {
8057 error_setg(errp, "The node %s does not support removing a child",
8058 bdrv_get_device_or_node_name(parent_bs));
8059 return;
8060 }
8061
8062 QLIST_FOREACH(tmp, &parent_bs->children, next) {
8063 if (tmp == child) {
8064 break;
8065 }
8066 }
8067
8068 if (!tmp) {
8069 error_setg(errp, "The node %s does not have a child named %s",
8070 bdrv_get_device_or_node_name(parent_bs),
8071 bdrv_get_device_or_node_name(child->bs));
8072 return;
8073 }
8074
8075 parent_bs->drv->bdrv_del_child(parent_bs, child, errp);
8076}
6f7a3b53
HR
8077
8078int bdrv_make_empty(BdrvChild *c, Error **errp)
8079{
8080 BlockDriver *drv = c->bs->drv;
8081 int ret;
8082
f791bf7f 8083 GLOBAL_STATE_CODE();
6f7a3b53
HR
8084 assert(c->perm & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED));
8085
8086 if (!drv->bdrv_make_empty) {
8087 error_setg(errp, "%s does not support emptying nodes",
8088 drv->format_name);
8089 return -ENOTSUP;
8090 }
8091
8092 ret = drv->bdrv_make_empty(c->bs);
8093 if (ret < 0) {
8094 error_setg_errno(errp, -ret, "Failed to empty %s",
8095 c->bs->filename);
8096 return ret;
8097 }
8098
8099 return 0;
8100}
9a6fc887
HR
8101
8102/*
8103 * Return the child that @bs acts as an overlay for, and from which data may be
8104 * copied in COW or COR operations. Usually this is the backing file.
8105 */
8106BdrvChild *bdrv_cow_child(BlockDriverState *bs)
8107{
967d7905
EGE
8108 IO_CODE();
8109
9a6fc887
HR
8110 if (!bs || !bs->drv) {
8111 return NULL;
8112 }
8113
8114 if (bs->drv->is_filter) {
8115 return NULL;
8116 }
8117
8118 if (!bs->backing) {
8119 return NULL;
8120 }
8121
8122 assert(bs->backing->role & BDRV_CHILD_COW);
8123 return bs->backing;
8124}
8125
8126/*
8127 * If @bs acts as a filter for exactly one of its children, return
8128 * that child.
8129 */
8130BdrvChild *bdrv_filter_child(BlockDriverState *bs)
8131{
8132 BdrvChild *c;
967d7905 8133 IO_CODE();
9a6fc887
HR
8134
8135 if (!bs || !bs->drv) {
8136 return NULL;
8137 }
8138
8139 if (!bs->drv->is_filter) {
8140 return NULL;
8141 }
8142
8143 /* Only one of @backing or @file may be used */
8144 assert(!(bs->backing && bs->file));
8145
8146 c = bs->backing ?: bs->file;
8147 if (!c) {
8148 return NULL;
8149 }
8150
8151 assert(c->role & BDRV_CHILD_FILTERED);
8152 return c;
8153}
8154
8155/*
8156 * Return either the result of bdrv_cow_child() or bdrv_filter_child(),
8157 * whichever is non-NULL.
8158 *
8159 * Return NULL if both are NULL.
8160 */
8161BdrvChild *bdrv_filter_or_cow_child(BlockDriverState *bs)
8162{
8163 BdrvChild *cow_child = bdrv_cow_child(bs);
8164 BdrvChild *filter_child = bdrv_filter_child(bs);
967d7905 8165 IO_CODE();
9a6fc887
HR
8166
8167 /* Filter nodes cannot have COW backing files */
8168 assert(!(cow_child && filter_child));
8169
8170 return cow_child ?: filter_child;
8171}
8172
8173/*
8174 * Return the primary child of this node: For filters, that is the
8175 * filtered child. For other nodes, that is usually the child storing
8176 * metadata.
8177 * (A generally more helpful description is that this is (usually) the
8178 * child that has the same filename as @bs.)
8179 *
8180 * Drivers do not necessarily have a primary child; for example quorum
8181 * does not.
8182 */
8183BdrvChild *bdrv_primary_child(BlockDriverState *bs)
8184{
8185 BdrvChild *c, *found = NULL;
967d7905 8186 IO_CODE();
9a6fc887
HR
8187
8188 QLIST_FOREACH(c, &bs->children, next) {
8189 if (c->role & BDRV_CHILD_PRIMARY) {
8190 assert(!found);
8191 found = c;
8192 }
8193 }
8194
8195 return found;
8196}
d38d7eb8
HR
8197
8198static BlockDriverState *bdrv_do_skip_filters(BlockDriverState *bs,
8199 bool stop_on_explicit_filter)
8200{
8201 BdrvChild *c;
8202
8203 if (!bs) {
8204 return NULL;
8205 }
8206
8207 while (!(stop_on_explicit_filter && !bs->implicit)) {
8208 c = bdrv_filter_child(bs);
8209 if (!c) {
8210 /*
8211 * A filter that is embedded in a working block graph must
8212 * have a child. Assert this here so this function does
8213 * not return a filter node that is not expected by the
8214 * caller.
8215 */
8216 assert(!bs->drv || !bs->drv->is_filter);
8217 break;
8218 }
8219 bs = c->bs;
8220 }
8221 /*
8222 * Note that this treats nodes with bs->drv == NULL as not being
8223 * filters (bs->drv == NULL should be replaced by something else
8224 * anyway).
8225 * The advantage of this behavior is that this function will thus
8226 * always return a non-NULL value (given a non-NULL @bs).
8227 */
8228
8229 return bs;
8230}
8231
8232/*
8233 * Return the first BDS that has not been added implicitly or that
8234 * does not have a filtered child down the chain starting from @bs
8235 * (including @bs itself).
8236 */
8237BlockDriverState *bdrv_skip_implicit_filters(BlockDriverState *bs)
8238{
b4ad82aa 8239 GLOBAL_STATE_CODE();
d38d7eb8
HR
8240 return bdrv_do_skip_filters(bs, true);
8241}
8242
8243/*
8244 * Return the first BDS that does not have a filtered child down the
8245 * chain starting from @bs (including @bs itself).
8246 */
8247BlockDriverState *bdrv_skip_filters(BlockDriverState *bs)
8248{
967d7905 8249 IO_CODE();
d38d7eb8
HR
8250 return bdrv_do_skip_filters(bs, false);
8251}
8252
8253/*
8254 * For a backing chain, return the first non-filter backing image of
8255 * the first non-filter image.
8256 */
8257BlockDriverState *bdrv_backing_chain_next(BlockDriverState *bs)
8258{
967d7905 8259 IO_CODE();
d38d7eb8
HR
8260 return bdrv_skip_filters(bdrv_cow_bs(bdrv_skip_filters(bs)));
8261}
0bc329fb
HR
8262
8263/**
8264 * Check whether [offset, offset + bytes) overlaps with the cached
8265 * block-status data region.
8266 *
8267 * If so, and @pnum is not NULL, set *pnum to `bsc.data_end - offset`,
8268 * which is what bdrv_bsc_is_data()'s interface needs.
8269 * Otherwise, *pnum is not touched.
8270 */
8271static bool bdrv_bsc_range_overlaps_locked(BlockDriverState *bs,
8272 int64_t offset, int64_t bytes,
8273 int64_t *pnum)
8274{
8275 BdrvBlockStatusCache *bsc = qatomic_rcu_read(&bs->block_status_cache);
8276 bool overlaps;
8277
8278 overlaps =
8279 qatomic_read(&bsc->valid) &&
8280 ranges_overlap(offset, bytes, bsc->data_start,
8281 bsc->data_end - bsc->data_start);
8282
8283 if (overlaps && pnum) {
8284 *pnum = bsc->data_end - offset;
8285 }
8286
8287 return overlaps;
8288}
8289
8290/**
8291 * See block_int.h for this function's documentation.
8292 */
8293bool bdrv_bsc_is_data(BlockDriverState *bs, int64_t offset, int64_t *pnum)
8294{
967d7905 8295 IO_CODE();
0bc329fb 8296 RCU_READ_LOCK_GUARD();
0bc329fb
HR
8297 return bdrv_bsc_range_overlaps_locked(bs, offset, 1, pnum);
8298}
8299
8300/**
8301 * See block_int.h for this function's documentation.
8302 */
8303void bdrv_bsc_invalidate_range(BlockDriverState *bs,
8304 int64_t offset, int64_t bytes)
8305{
967d7905 8306 IO_CODE();
0bc329fb
HR
8307 RCU_READ_LOCK_GUARD();
8308
8309 if (bdrv_bsc_range_overlaps_locked(bs, offset, bytes, NULL)) {
8310 qatomic_set(&bs->block_status_cache->valid, false);
8311 }
8312}
8313
8314/**
8315 * See block_int.h for this function's documentation.
8316 */
8317void bdrv_bsc_fill(BlockDriverState *bs, int64_t offset, int64_t bytes)
8318{
8319 BdrvBlockStatusCache *new_bsc = g_new(BdrvBlockStatusCache, 1);
8320 BdrvBlockStatusCache *old_bsc;
967d7905 8321 IO_CODE();
0bc329fb
HR
8322
8323 *new_bsc = (BdrvBlockStatusCache) {
8324 .valid = true,
8325 .data_start = offset,
8326 .data_end = offset + bytes,
8327 };
8328
8329 QEMU_LOCK_GUARD(&bs->bsc_modify_lock);
8330
8331 old_bsc = qatomic_rcu_read(&bs->block_status_cache);
8332 qatomic_rcu_set(&bs->block_status_cache, new_bsc);
8333 if (old_bsc) {
8334 g_free_rcu(old_bsc, rcu);
8335 }
8336}
This page took 3.407551 seconds and 4 git commands to generate.