]> git.ipfire.org Git - thirdparty/qemu.git/blame - block.c
block: Avoid memleak on qcow2 image info failure
[thirdparty/qemu.git] / block.c
CommitLineData
fc01f7e7
FB
1/*
2 * QEMU System Emulator block driver
5fafdf24 3 *
fc01f7e7 4 * Copyright (c) 2003 Fabrice Bellard
5fafdf24 5 *
fc01f7e7
FB
6 * Permission is hereby granted, free of charge, to any person obtaining a copy
7 * of this software and associated documentation files (the "Software"), to deal
8 * in the Software without restriction, including without limitation the rights
9 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
10 * copies of the Software, and to permit persons to whom the Software is
11 * furnished to do so, subject to the following conditions:
12 *
13 * The above copyright notice and this permission notice shall be included in
14 * all copies or substantial portions of the Software.
15 *
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
22 * THE SOFTWARE.
23 */
e688df6b 24
d38ea87a 25#include "qemu/osdep.h"
0ab8ed18 26#include "block/trace.h"
737e150e
PB
27#include "block/block_int.h"
28#include "block/blockjob.h"
cd7fca95 29#include "block/nbd.h"
609f45ea 30#include "block/qdict.h"
d49b6836 31#include "qemu/error-report.h"
88d88798 32#include "module_block.h"
db725815 33#include "qemu/main-loop.h"
1de7afc9 34#include "qemu/module.h"
e688df6b 35#include "qapi/error.h"
452fcdbc 36#include "qapi/qmp/qdict.h"
7b1b5d19 37#include "qapi/qmp/qjson.h"
e59a0cf1 38#include "qapi/qmp/qnull.h"
fc81fa1e 39#include "qapi/qmp/qstring.h"
e1d74bc6
KW
40#include "qapi/qobject-output-visitor.h"
41#include "qapi/qapi-visit-block-core.h"
bfb197e0 42#include "sysemu/block-backend.h"
9c17d615 43#include "sysemu/sysemu.h"
1de7afc9 44#include "qemu/notify.h"
922a01a0 45#include "qemu/option.h"
10817bf0 46#include "qemu/coroutine.h"
c13163fb 47#include "block/qapi.h"
1de7afc9 48#include "qemu/timer.h"
f348b6d1
VB
49#include "qemu/cutils.h"
50#include "qemu/id.h"
fc01f7e7 51
71e72a19 52#ifdef CONFIG_BSD
7674e7bf 53#include <sys/ioctl.h>
72cf2d4f 54#include <sys/queue.h>
c5e97233 55#ifndef __DragonFly__
7674e7bf
FB
56#include <sys/disk.h>
57#endif
c5e97233 58#endif
7674e7bf 59
49dc768d
AL
60#ifdef _WIN32
61#include <windows.h>
62#endif
63
1c9805a3
SH
64#define NOT_DONE 0x7fffffff /* used while emulated sync operation in progress */
65
dc364f4c
BC
66static QTAILQ_HEAD(, BlockDriverState) graph_bdrv_states =
67 QTAILQ_HEAD_INITIALIZER(graph_bdrv_states);
68
2c1d04e0
HR
69static QTAILQ_HEAD(, BlockDriverState) all_bdrv_states =
70 QTAILQ_HEAD_INITIALIZER(all_bdrv_states);
71
8a22f02a
SH
72static QLIST_HEAD(, BlockDriver) bdrv_drivers =
73 QLIST_HEAD_INITIALIZER(bdrv_drivers);
ea2384d3 74
5b363937
HR
75static BlockDriverState *bdrv_open_inherit(const char *filename,
76 const char *reference,
77 QDict *options, int flags,
78 BlockDriverState *parent,
79 const BdrvChildRole *child_role,
80 Error **errp);
f3930ed0 81
eb852011
MA
82/* If non-zero, use only whitelisted block drivers */
83static int use_bdrv_whitelist;
84
9e0b22f4
SH
85#ifdef _WIN32
86static int is_windows_drive_prefix(const char *filename)
87{
88 return (((filename[0] >= 'a' && filename[0] <= 'z') ||
89 (filename[0] >= 'A' && filename[0] <= 'Z')) &&
90 filename[1] == ':');
91}
92
93int is_windows_drive(const char *filename)
94{
95 if (is_windows_drive_prefix(filename) &&
96 filename[2] == '\0')
97 return 1;
98 if (strstart(filename, "\\\\.\\", NULL) ||
99 strstart(filename, "//./", NULL))
100 return 1;
101 return 0;
102}
103#endif
104
339064d5
KW
105size_t bdrv_opt_mem_align(BlockDriverState *bs)
106{
107 if (!bs || !bs->drv) {
459b4e66 108 /* page size or 4k (hdd sector size) should be on the safe side */
038adc2f 109 return MAX(4096, qemu_real_host_page_size);
339064d5
KW
110 }
111
112 return bs->bl.opt_mem_alignment;
113}
114
4196d2f0
DL
115size_t bdrv_min_mem_align(BlockDriverState *bs)
116{
117 if (!bs || !bs->drv) {
459b4e66 118 /* page size or 4k (hdd sector size) should be on the safe side */
038adc2f 119 return MAX(4096, qemu_real_host_page_size);
4196d2f0
DL
120 }
121
122 return bs->bl.min_mem_alignment;
123}
124
9e0b22f4 125/* check if the path starts with "<protocol>:" */
5c98415b 126int path_has_protocol(const char *path)
9e0b22f4 127{
947995c0
PB
128 const char *p;
129
9e0b22f4
SH
130#ifdef _WIN32
131 if (is_windows_drive(path) ||
132 is_windows_drive_prefix(path)) {
133 return 0;
134 }
947995c0
PB
135 p = path + strcspn(path, ":/\\");
136#else
137 p = path + strcspn(path, ":/");
9e0b22f4
SH
138#endif
139
947995c0 140 return *p == ':';
9e0b22f4
SH
141}
142
83f64091 143int path_is_absolute(const char *path)
3b0d4f61 144{
21664424
FB
145#ifdef _WIN32
146 /* specific case for names like: "\\.\d:" */
f53f4da9 147 if (is_windows_drive(path) || is_windows_drive_prefix(path)) {
21664424 148 return 1;
f53f4da9
PB
149 }
150 return (*path == '/' || *path == '\\');
3b9f94e1 151#else
f53f4da9 152 return (*path == '/');
3b9f94e1 153#endif
3b0d4f61
FB
154}
155
009b03aa 156/* if filename is absolute, just return its duplicate. Otherwise, build a
83f64091
FB
157 path to it by considering it is relative to base_path. URL are
158 supported. */
009b03aa 159char *path_combine(const char *base_path, const char *filename)
3b0d4f61 160{
009b03aa 161 const char *protocol_stripped = NULL;
83f64091 162 const char *p, *p1;
009b03aa 163 char *result;
83f64091
FB
164 int len;
165
83f64091 166 if (path_is_absolute(filename)) {
009b03aa
HR
167 return g_strdup(filename);
168 }
0d54a6fe 169
009b03aa
HR
170 if (path_has_protocol(base_path)) {
171 protocol_stripped = strchr(base_path, ':');
172 if (protocol_stripped) {
173 protocol_stripped++;
0d54a6fe 174 }
009b03aa
HR
175 }
176 p = protocol_stripped ?: base_path;
0d54a6fe 177
009b03aa 178 p1 = strrchr(base_path, '/');
3b9f94e1 179#ifdef _WIN32
009b03aa
HR
180 {
181 const char *p2;
182 p2 = strrchr(base_path, '\\');
183 if (!p1 || p2 > p1) {
184 p1 = p2;
3b9f94e1 185 }
009b03aa 186 }
3b9f94e1 187#endif
009b03aa
HR
188 if (p1) {
189 p1++;
190 } else {
191 p1 = base_path;
192 }
193 if (p1 > p) {
194 p = p1;
3b0d4f61 195 }
009b03aa
HR
196 len = p - base_path;
197
198 result = g_malloc(len + strlen(filename) + 1);
199 memcpy(result, base_path, len);
200 strcpy(result + len, filename);
201
202 return result;
203}
204
03c320d8
HR
205/*
206 * Helper function for bdrv_parse_filename() implementations to remove optional
207 * protocol prefixes (especially "file:") from a filename and for putting the
208 * stripped filename into the options QDict if there is such a prefix.
209 */
210void bdrv_parse_filename_strip_prefix(const char *filename, const char *prefix,
211 QDict *options)
212{
213 if (strstart(filename, prefix, &filename)) {
214 /* Stripping the explicit protocol prefix may result in a protocol
215 * prefix being (wrongly) detected (if the filename contains a colon) */
216 if (path_has_protocol(filename)) {
217 QString *fat_filename;
218
219 /* This means there is some colon before the first slash; therefore,
220 * this cannot be an absolute path */
221 assert(!path_is_absolute(filename));
222
223 /* And we can thus fix the protocol detection issue by prefixing it
224 * by "./" */
225 fat_filename = qstring_from_str("./");
226 qstring_append(fat_filename, filename);
227
228 assert(!path_has_protocol(qstring_get_str(fat_filename)));
229
230 qdict_put(options, "filename", fat_filename);
231 } else {
232 /* If no protocol prefix was detected, we can use the shortened
233 * filename as-is */
234 qdict_put_str(options, "filename", filename);
235 }
236 }
237}
238
239
9c5e6594
KW
240/* Returns whether the image file is opened as read-only. Note that this can
241 * return false and writing to the image file is still not possible because the
242 * image is inactivated. */
93ed524e
JC
243bool bdrv_is_read_only(BlockDriverState *bs)
244{
245 return bs->read_only;
246}
247
54a32bfe
KW
248int bdrv_can_set_read_only(BlockDriverState *bs, bool read_only,
249 bool ignore_allow_rdw, Error **errp)
fe5241bf 250{
e2b8247a
JC
251 /* Do not set read_only if copy_on_read is enabled */
252 if (bs->copy_on_read && read_only) {
253 error_setg(errp, "Can't set node '%s' to r/o with copy-on-read enabled",
254 bdrv_get_device_or_node_name(bs));
255 return -EINVAL;
256 }
257
d6fcdf06 258 /* Do not clear read_only if it is prohibited */
54a32bfe
KW
259 if (!read_only && !(bs->open_flags & BDRV_O_ALLOW_RDWR) &&
260 !ignore_allow_rdw)
261 {
d6fcdf06
JC
262 error_setg(errp, "Node '%s' is read only",
263 bdrv_get_device_or_node_name(bs));
264 return -EPERM;
265 }
266
45803a03
JC
267 return 0;
268}
269
eaa2410f
KW
270/*
271 * Called by a driver that can only provide a read-only image.
272 *
273 * Returns 0 if the node is already read-only or it could switch the node to
274 * read-only because BDRV_O_AUTO_RDONLY is set.
275 *
276 * Returns -EACCES if the node is read-write and BDRV_O_AUTO_RDONLY is not set
277 * or bdrv_can_set_read_only() forbids making the node read-only. If @errmsg
278 * is not NULL, it is used as the error message for the Error object.
279 */
280int bdrv_apply_auto_read_only(BlockDriverState *bs, const char *errmsg,
281 Error **errp)
45803a03
JC
282{
283 int ret = 0;
284
eaa2410f
KW
285 if (!(bs->open_flags & BDRV_O_RDWR)) {
286 return 0;
287 }
288 if (!(bs->open_flags & BDRV_O_AUTO_RDONLY)) {
289 goto fail;
45803a03
JC
290 }
291
eaa2410f
KW
292 ret = bdrv_can_set_read_only(bs, true, false, NULL);
293 if (ret < 0) {
294 goto fail;
eeae6a59
KW
295 }
296
eaa2410f
KW
297 bs->read_only = true;
298 bs->open_flags &= ~BDRV_O_RDWR;
299
e2b8247a 300 return 0;
eaa2410f
KW
301
302fail:
303 error_setg(errp, "%s", errmsg ?: "Image is read-only");
304 return -EACCES;
fe5241bf
JC
305}
306
645ae7d8
HR
307/*
308 * If @backing is empty, this function returns NULL without setting
309 * @errp. In all other cases, NULL will only be returned with @errp
310 * set.
311 *
312 * Therefore, a return value of NULL without @errp set means that
313 * there is no backing file; if @errp is set, there is one but its
314 * absolute filename cannot be generated.
315 */
316char *bdrv_get_full_backing_filename_from_filename(const char *backed,
317 const char *backing,
318 Error **errp)
dc5a1371 319{
645ae7d8
HR
320 if (backing[0] == '\0') {
321 return NULL;
322 } else if (path_has_protocol(backing) || path_is_absolute(backing)) {
323 return g_strdup(backing);
9f07429e
HR
324 } else if (backed[0] == '\0' || strstart(backed, "json:", NULL)) {
325 error_setg(errp, "Cannot use relative backing file names for '%s'",
326 backed);
645ae7d8 327 return NULL;
dc5a1371 328 } else {
645ae7d8 329 return path_combine(backed, backing);
dc5a1371
PB
330 }
331}
332
9f4793d8
HR
333/*
334 * If @filename is empty or NULL, this function returns NULL without
335 * setting @errp. In all other cases, NULL will only be returned with
336 * @errp set.
337 */
338static char *bdrv_make_absolute_filename(BlockDriverState *relative_to,
339 const char *filename, Error **errp)
0a82855a 340{
8df68616 341 char *dir, *full_name;
9f4793d8 342
8df68616
HR
343 if (!filename || filename[0] == '\0') {
344 return NULL;
345 } else if (path_has_protocol(filename) || path_is_absolute(filename)) {
346 return g_strdup(filename);
347 }
9f07429e 348
8df68616
HR
349 dir = bdrv_dirname(relative_to, errp);
350 if (!dir) {
351 return NULL;
352 }
f30c66ba 353
8df68616
HR
354 full_name = g_strconcat(dir, filename, NULL);
355 g_free(dir);
356 return full_name;
9f4793d8
HR
357}
358
359char *bdrv_get_full_backing_filename(BlockDriverState *bs, Error **errp)
360{
361 return bdrv_make_absolute_filename(bs, bs->backing_file, errp);
0a82855a
HR
362}
363
0eb7217e
SH
364void bdrv_register(BlockDriver *bdrv)
365{
8a22f02a 366 QLIST_INSERT_HEAD(&bdrv_drivers, bdrv, list);
ea2384d3 367}
b338082b 368
e4e9986b
MA
369BlockDriverState *bdrv_new(void)
370{
371 BlockDriverState *bs;
372 int i;
373
5839e53b 374 bs = g_new0(BlockDriverState, 1);
e4654d2d 375 QLIST_INIT(&bs->dirty_bitmaps);
fbe40ff7
FZ
376 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
377 QLIST_INIT(&bs->op_blockers[i]);
378 }
d616b224 379 notifier_with_return_list_init(&bs->before_write_notifiers);
3783fa3d 380 qemu_co_mutex_init(&bs->reqs_lock);
2119882c 381 qemu_mutex_init(&bs->dirty_bitmap_mutex);
9fcb0251 382 bs->refcnt = 1;
dcd04228 383 bs->aio_context = qemu_get_aio_context();
d7d512f6 384
3ff2f67a
EY
385 qemu_co_queue_init(&bs->flush_queue);
386
0f12264e
KW
387 for (i = 0; i < bdrv_drain_all_count; i++) {
388 bdrv_drained_begin(bs);
389 }
390
2c1d04e0
HR
391 QTAILQ_INSERT_TAIL(&all_bdrv_states, bs, bs_list);
392
b338082b
FB
393 return bs;
394}
395
88d88798 396static BlockDriver *bdrv_do_find_format(const char *format_name)
ea2384d3
FB
397{
398 BlockDriver *drv1;
88d88798 399
8a22f02a
SH
400 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
401 if (!strcmp(drv1->format_name, format_name)) {
ea2384d3 402 return drv1;
8a22f02a 403 }
ea2384d3 404 }
88d88798 405
ea2384d3
FB
406 return NULL;
407}
408
88d88798
MM
409BlockDriver *bdrv_find_format(const char *format_name)
410{
411 BlockDriver *drv1;
412 int i;
413
414 drv1 = bdrv_do_find_format(format_name);
415 if (drv1) {
416 return drv1;
417 }
418
419 /* The driver isn't registered, maybe we need to load a module */
420 for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); ++i) {
421 if (!strcmp(block_driver_modules[i].format_name, format_name)) {
422 block_module_load_one(block_driver_modules[i].library_name);
423 break;
424 }
425 }
426
427 return bdrv_do_find_format(format_name);
428}
429
9ac404c5 430static int bdrv_format_is_whitelisted(const char *format_name, bool read_only)
eb852011 431{
b64ec4e4
FZ
432 static const char *whitelist_rw[] = {
433 CONFIG_BDRV_RW_WHITELIST
434 };
435 static const char *whitelist_ro[] = {
436 CONFIG_BDRV_RO_WHITELIST
eb852011
MA
437 };
438 const char **p;
439
b64ec4e4 440 if (!whitelist_rw[0] && !whitelist_ro[0]) {
eb852011 441 return 1; /* no whitelist, anything goes */
b64ec4e4 442 }
eb852011 443
b64ec4e4 444 for (p = whitelist_rw; *p; p++) {
9ac404c5 445 if (!strcmp(format_name, *p)) {
eb852011
MA
446 return 1;
447 }
448 }
b64ec4e4
FZ
449 if (read_only) {
450 for (p = whitelist_ro; *p; p++) {
9ac404c5 451 if (!strcmp(format_name, *p)) {
b64ec4e4
FZ
452 return 1;
453 }
454 }
455 }
eb852011
MA
456 return 0;
457}
458
9ac404c5
AS
459int bdrv_is_whitelisted(BlockDriver *drv, bool read_only)
460{
461 return bdrv_format_is_whitelisted(drv->format_name, read_only);
462}
463
e6ff69bf
DB
464bool bdrv_uses_whitelist(void)
465{
466 return use_bdrv_whitelist;
467}
468
5b7e1542
ZYW
469typedef struct CreateCo {
470 BlockDriver *drv;
471 char *filename;
83d0521a 472 QemuOpts *opts;
5b7e1542 473 int ret;
cc84d90f 474 Error *err;
5b7e1542
ZYW
475} CreateCo;
476
477static void coroutine_fn bdrv_create_co_entry(void *opaque)
478{
cc84d90f
HR
479 Error *local_err = NULL;
480 int ret;
481
5b7e1542
ZYW
482 CreateCo *cco = opaque;
483 assert(cco->drv);
484
efc75e2a 485 ret = cco->drv->bdrv_co_create_opts(cco->filename, cco->opts, &local_err);
621ff94d 486 error_propagate(&cco->err, local_err);
cc84d90f 487 cco->ret = ret;
5b7e1542
ZYW
488}
489
0e7e1989 490int bdrv_create(BlockDriver *drv, const char* filename,
83d0521a 491 QemuOpts *opts, Error **errp)
ea2384d3 492{
5b7e1542
ZYW
493 int ret;
494
495 Coroutine *co;
496 CreateCo cco = {
497 .drv = drv,
498 .filename = g_strdup(filename),
83d0521a 499 .opts = opts,
5b7e1542 500 .ret = NOT_DONE,
cc84d90f 501 .err = NULL,
5b7e1542
ZYW
502 };
503
efc75e2a 504 if (!drv->bdrv_co_create_opts) {
cc84d90f 505 error_setg(errp, "Driver '%s' does not support image creation", drv->format_name);
80168bff
LC
506 ret = -ENOTSUP;
507 goto out;
5b7e1542
ZYW
508 }
509
510 if (qemu_in_coroutine()) {
511 /* Fast-path if already in coroutine context */
512 bdrv_create_co_entry(&cco);
513 } else {
0b8b8753
PB
514 co = qemu_coroutine_create(bdrv_create_co_entry, &cco);
515 qemu_coroutine_enter(co);
5b7e1542 516 while (cco.ret == NOT_DONE) {
b47ec2c4 517 aio_poll(qemu_get_aio_context(), true);
5b7e1542
ZYW
518 }
519 }
520
521 ret = cco.ret;
cc84d90f 522 if (ret < 0) {
84d18f06 523 if (cco.err) {
cc84d90f
HR
524 error_propagate(errp, cco.err);
525 } else {
526 error_setg_errno(errp, -ret, "Could not create image");
527 }
528 }
0e7e1989 529
80168bff
LC
530out:
531 g_free(cco.filename);
5b7e1542 532 return ret;
ea2384d3
FB
533}
534
fd17146c
HR
535/**
536 * Helper function for bdrv_create_file_fallback(): Resize @blk to at
537 * least the given @minimum_size.
538 *
539 * On success, return @blk's actual length.
540 * Otherwise, return -errno.
541 */
542static int64_t create_file_fallback_truncate(BlockBackend *blk,
543 int64_t minimum_size, Error **errp)
84a12e66 544{
cc84d90f 545 Error *local_err = NULL;
fd17146c 546 int64_t size;
cc84d90f 547 int ret;
84a12e66 548
fd17146c
HR
549 ret = blk_truncate(blk, minimum_size, false, PREALLOC_MODE_OFF, &local_err);
550 if (ret < 0 && ret != -ENOTSUP) {
551 error_propagate(errp, local_err);
552 return ret;
553 }
554
555 size = blk_getlength(blk);
556 if (size < 0) {
557 error_free(local_err);
558 error_setg_errno(errp, -size,
559 "Failed to inquire the new image file's length");
560 return size;
561 }
562
563 if (size < minimum_size) {
564 /* Need to grow the image, but we failed to do that */
565 error_propagate(errp, local_err);
566 return -ENOTSUP;
567 }
568
569 error_free(local_err);
570 local_err = NULL;
571
572 return size;
573}
574
575/**
576 * Helper function for bdrv_create_file_fallback(): Zero the first
577 * sector to remove any potentially pre-existing image header.
578 */
579static int create_file_fallback_zero_first_sector(BlockBackend *blk,
580 int64_t current_size,
581 Error **errp)
582{
583 int64_t bytes_to_clear;
584 int ret;
585
586 bytes_to_clear = MIN(current_size, BDRV_SECTOR_SIZE);
587 if (bytes_to_clear) {
588 ret = blk_pwrite_zeroes(blk, 0, bytes_to_clear, BDRV_REQ_MAY_UNMAP);
589 if (ret < 0) {
590 error_setg_errno(errp, -ret,
591 "Failed to clear the new image's first sector");
592 return ret;
593 }
594 }
595
596 return 0;
597}
598
599static int bdrv_create_file_fallback(const char *filename, BlockDriver *drv,
600 QemuOpts *opts, Error **errp)
601{
602 BlockBackend *blk;
eeea1faa 603 QDict *options;
fd17146c
HR
604 int64_t size = 0;
605 char *buf = NULL;
606 PreallocMode prealloc;
607 Error *local_err = NULL;
608 int ret;
609
610 size = qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0);
611 buf = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC);
612 prealloc = qapi_enum_parse(&PreallocMode_lookup, buf,
613 PREALLOC_MODE_OFF, &local_err);
614 g_free(buf);
615 if (local_err) {
616 error_propagate(errp, local_err);
617 return -EINVAL;
618 }
619
620 if (prealloc != PREALLOC_MODE_OFF) {
621 error_setg(errp, "Unsupported preallocation mode '%s'",
622 PreallocMode_str(prealloc));
623 return -ENOTSUP;
624 }
625
eeea1faa 626 options = qdict_new();
fd17146c
HR
627 qdict_put_str(options, "driver", drv->format_name);
628
629 blk = blk_new_open(filename, NULL, options,
630 BDRV_O_RDWR | BDRV_O_RESIZE, errp);
631 if (!blk) {
632 error_prepend(errp, "Protocol driver '%s' does not support image "
633 "creation, and opening the image failed: ",
634 drv->format_name);
635 return -EINVAL;
636 }
637
638 size = create_file_fallback_truncate(blk, size, errp);
639 if (size < 0) {
640 ret = size;
641 goto out;
642 }
643
644 ret = create_file_fallback_zero_first_sector(blk, size, errp);
645 if (ret < 0) {
646 goto out;
647 }
648
649 ret = 0;
650out:
651 blk_unref(blk);
652 return ret;
653}
654
655int bdrv_create_file(const char *filename, QemuOpts *opts, Error **errp)
656{
657 BlockDriver *drv;
658
b65a5e12 659 drv = bdrv_find_protocol(filename, true, errp);
84a12e66 660 if (drv == NULL) {
16905d71 661 return -ENOENT;
84a12e66
CH
662 }
663
fd17146c
HR
664 if (drv->bdrv_co_create_opts) {
665 return bdrv_create(drv, filename, opts, errp);
666 } else {
667 return bdrv_create_file_fallback(filename, drv, opts, errp);
668 }
84a12e66
CH
669}
670
e1d7f8bb
DHB
671int coroutine_fn bdrv_co_delete_file(BlockDriverState *bs, Error **errp)
672{
673 Error *local_err = NULL;
674 int ret;
675
676 assert(bs != NULL);
677
678 if (!bs->drv) {
679 error_setg(errp, "Block node '%s' is not opened", bs->filename);
680 return -ENOMEDIUM;
681 }
682
683 if (!bs->drv->bdrv_co_delete_file) {
684 error_setg(errp, "Driver '%s' does not support image deletion",
685 bs->drv->format_name);
686 return -ENOTSUP;
687 }
688
689 ret = bs->drv->bdrv_co_delete_file(bs, &local_err);
690 if (ret < 0) {
691 error_propagate(errp, local_err);
692 }
693
694 return ret;
695}
696
892b7de8
ET
697/**
698 * Try to get @bs's logical and physical block size.
699 * On success, store them in @bsz struct and return 0.
700 * On failure return -errno.
701 * @bs must not be empty.
702 */
703int bdrv_probe_blocksizes(BlockDriverState *bs, BlockSizes *bsz)
704{
705 BlockDriver *drv = bs->drv;
706
707 if (drv && drv->bdrv_probe_blocksizes) {
708 return drv->bdrv_probe_blocksizes(bs, bsz);
5a612c00
MP
709 } else if (drv && drv->is_filter && bs->file) {
710 return bdrv_probe_blocksizes(bs->file->bs, bsz);
892b7de8
ET
711 }
712
713 return -ENOTSUP;
714}
715
716/**
717 * Try to get @bs's geometry (cyls, heads, sectors).
718 * On success, store them in @geo struct and return 0.
719 * On failure return -errno.
720 * @bs must not be empty.
721 */
722int bdrv_probe_geometry(BlockDriverState *bs, HDGeometry *geo)
723{
724 BlockDriver *drv = bs->drv;
725
726 if (drv && drv->bdrv_probe_geometry) {
727 return drv->bdrv_probe_geometry(bs, geo);
5a612c00
MP
728 } else if (drv && drv->is_filter && bs->file) {
729 return bdrv_probe_geometry(bs->file->bs, geo);
892b7de8
ET
730 }
731
732 return -ENOTSUP;
733}
734
eba25057
JM
735/*
736 * Create a uniquely-named empty temporary file.
737 * Return 0 upon success, otherwise a negative errno value.
738 */
739int get_tmp_filename(char *filename, int size)
d5249393 740{
eba25057 741#ifdef _WIN32
3b9f94e1 742 char temp_dir[MAX_PATH];
eba25057
JM
743 /* GetTempFileName requires that its output buffer (4th param)
744 have length MAX_PATH or greater. */
745 assert(size >= MAX_PATH);
746 return (GetTempPath(MAX_PATH, temp_dir)
747 && GetTempFileName(temp_dir, "qem", 0, filename)
748 ? 0 : -GetLastError());
d5249393 749#else
67b915a5 750 int fd;
7ccfb2eb 751 const char *tmpdir;
0badc1ee 752 tmpdir = getenv("TMPDIR");
69bef793
AS
753 if (!tmpdir) {
754 tmpdir = "/var/tmp";
755 }
eba25057
JM
756 if (snprintf(filename, size, "%s/vl.XXXXXX", tmpdir) >= size) {
757 return -EOVERFLOW;
758 }
ea2384d3 759 fd = mkstemp(filename);
fe235a06
DH
760 if (fd < 0) {
761 return -errno;
762 }
763 if (close(fd) != 0) {
764 unlink(filename);
eba25057
JM
765 return -errno;
766 }
767 return 0;
d5249393 768#endif
eba25057 769}
fc01f7e7 770
84a12e66
CH
771/*
772 * Detect host devices. By convention, /dev/cdrom[N] is always
773 * recognized as a host CDROM.
774 */
775static BlockDriver *find_hdev_driver(const char *filename)
776{
777 int score_max = 0, score;
778 BlockDriver *drv = NULL, *d;
779
780 QLIST_FOREACH(d, &bdrv_drivers, list) {
781 if (d->bdrv_probe_device) {
782 score = d->bdrv_probe_device(filename);
783 if (score > score_max) {
784 score_max = score;
785 drv = d;
786 }
787 }
788 }
789
790 return drv;
791}
792
88d88798
MM
793static BlockDriver *bdrv_do_find_protocol(const char *protocol)
794{
795 BlockDriver *drv1;
796
797 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
798 if (drv1->protocol_name && !strcmp(drv1->protocol_name, protocol)) {
799 return drv1;
800 }
801 }
802
803 return NULL;
804}
805
98289620 806BlockDriver *bdrv_find_protocol(const char *filename,
b65a5e12
HR
807 bool allow_protocol_prefix,
808 Error **errp)
83f64091
FB
809{
810 BlockDriver *drv1;
811 char protocol[128];
1cec71e3 812 int len;
83f64091 813 const char *p;
88d88798 814 int i;
19cb3738 815
66f82cee
KW
816 /* TODO Drivers without bdrv_file_open must be specified explicitly */
817
39508e7a
CH
818 /*
819 * XXX(hch): we really should not let host device detection
820 * override an explicit protocol specification, but moving this
821 * later breaks access to device names with colons in them.
822 * Thanks to the brain-dead persistent naming schemes on udev-
823 * based Linux systems those actually are quite common.
824 */
825 drv1 = find_hdev_driver(filename);
826 if (drv1) {
827 return drv1;
828 }
829
98289620 830 if (!path_has_protocol(filename) || !allow_protocol_prefix) {
ef810437 831 return &bdrv_file;
84a12e66 832 }
98289620 833
9e0b22f4
SH
834 p = strchr(filename, ':');
835 assert(p != NULL);
1cec71e3
AL
836 len = p - filename;
837 if (len > sizeof(protocol) - 1)
838 len = sizeof(protocol) - 1;
839 memcpy(protocol, filename, len);
840 protocol[len] = '\0';
88d88798
MM
841
842 drv1 = bdrv_do_find_protocol(protocol);
843 if (drv1) {
844 return drv1;
845 }
846
847 for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); ++i) {
848 if (block_driver_modules[i].protocol_name &&
849 !strcmp(block_driver_modules[i].protocol_name, protocol)) {
850 block_module_load_one(block_driver_modules[i].library_name);
851 break;
8a22f02a 852 }
83f64091 853 }
b65a5e12 854
88d88798
MM
855 drv1 = bdrv_do_find_protocol(protocol);
856 if (!drv1) {
857 error_setg(errp, "Unknown protocol '%s'", protocol);
858 }
859 return drv1;
83f64091
FB
860}
861
c6684249
MA
862/*
863 * Guess image format by probing its contents.
864 * This is not a good idea when your image is raw (CVE-2008-2004), but
865 * we do it anyway for backward compatibility.
866 *
867 * @buf contains the image's first @buf_size bytes.
7cddd372
KW
868 * @buf_size is the buffer size in bytes (generally BLOCK_PROBE_BUF_SIZE,
869 * but can be smaller if the image file is smaller)
c6684249
MA
870 * @filename is its filename.
871 *
872 * For all block drivers, call the bdrv_probe() method to get its
873 * probing score.
874 * Return the first block driver with the highest probing score.
875 */
38f3ef57
KW
876BlockDriver *bdrv_probe_all(const uint8_t *buf, int buf_size,
877 const char *filename)
c6684249
MA
878{
879 int score_max = 0, score;
880 BlockDriver *drv = NULL, *d;
881
882 QLIST_FOREACH(d, &bdrv_drivers, list) {
883 if (d->bdrv_probe) {
884 score = d->bdrv_probe(buf, buf_size, filename);
885 if (score > score_max) {
886 score_max = score;
887 drv = d;
888 }
889 }
890 }
891
892 return drv;
893}
894
5696c6e3 895static int find_image_format(BlockBackend *file, const char *filename,
34b5d2c6 896 BlockDriver **pdrv, Error **errp)
f3a5d3f8 897{
c6684249 898 BlockDriver *drv;
7cddd372 899 uint8_t buf[BLOCK_PROBE_BUF_SIZE];
f500a6d3 900 int ret = 0;
f8ea0b00 901
08a00559 902 /* Return the raw BlockDriver * to scsi-generic devices or empty drives */
5696c6e3 903 if (blk_is_sg(file) || !blk_is_inserted(file) || blk_getlength(file) == 0) {
ef810437 904 *pdrv = &bdrv_raw;
c98ac35d 905 return ret;
1a396859 906 }
f8ea0b00 907
5696c6e3 908 ret = blk_pread(file, 0, buf, sizeof(buf));
83f64091 909 if (ret < 0) {
34b5d2c6
HR
910 error_setg_errno(errp, -ret, "Could not read image for determining its "
911 "format");
c98ac35d
SW
912 *pdrv = NULL;
913 return ret;
83f64091
FB
914 }
915
c6684249 916 drv = bdrv_probe_all(buf, ret, filename);
c98ac35d 917 if (!drv) {
34b5d2c6
HR
918 error_setg(errp, "Could not determine image format: No compatible "
919 "driver found");
c98ac35d
SW
920 ret = -ENOENT;
921 }
922 *pdrv = drv;
923 return ret;
ea2384d3
FB
924}
925
51762288
SH
926/**
927 * Set the current 'total_sectors' value
65a9bb25 928 * Return 0 on success, -errno on error.
51762288 929 */
3d9f2d2a 930int refresh_total_sectors(BlockDriverState *bs, int64_t hint)
51762288
SH
931{
932 BlockDriver *drv = bs->drv;
933
d470ad42
HR
934 if (!drv) {
935 return -ENOMEDIUM;
936 }
937
396759ad 938 /* Do not attempt drv->bdrv_getlength() on scsi-generic devices */
b192af8a 939 if (bdrv_is_sg(bs))
396759ad
NB
940 return 0;
941
51762288
SH
942 /* query actual device if possible, otherwise just trust the hint */
943 if (drv->bdrv_getlength) {
944 int64_t length = drv->bdrv_getlength(bs);
945 if (length < 0) {
946 return length;
947 }
7e382003 948 hint = DIV_ROUND_UP(length, BDRV_SECTOR_SIZE);
51762288
SH
949 }
950
951 bs->total_sectors = hint;
952 return 0;
953}
954
cddff5ba
KW
955/**
956 * Combines a QDict of new block driver @options with any missing options taken
957 * from @old_options, so that leaving out an option defaults to its old value.
958 */
959static void bdrv_join_options(BlockDriverState *bs, QDict *options,
960 QDict *old_options)
961{
962 if (bs->drv && bs->drv->bdrv_join_options) {
963 bs->drv->bdrv_join_options(options, old_options);
964 } else {
965 qdict_join(options, old_options, false);
966 }
967}
968
543770bd
AG
969static BlockdevDetectZeroesOptions bdrv_parse_detect_zeroes(QemuOpts *opts,
970 int open_flags,
971 Error **errp)
972{
973 Error *local_err = NULL;
974 char *value = qemu_opt_get_del(opts, "detect-zeroes");
975 BlockdevDetectZeroesOptions detect_zeroes =
976 qapi_enum_parse(&BlockdevDetectZeroesOptions_lookup, value,
977 BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF, &local_err);
978 g_free(value);
979 if (local_err) {
980 error_propagate(errp, local_err);
981 return detect_zeroes;
982 }
983
984 if (detect_zeroes == BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP &&
985 !(open_flags & BDRV_O_UNMAP))
986 {
987 error_setg(errp, "setting detect-zeroes to unmap is not allowed "
988 "without setting discard operation to unmap");
989 }
990
991 return detect_zeroes;
992}
993
f80f2673
AM
994/**
995 * Set open flags for aio engine
996 *
997 * Return 0 on success, -1 if the engine specified is invalid
998 */
999int bdrv_parse_aio(const char *mode, int *flags)
1000{
1001 if (!strcmp(mode, "threads")) {
1002 /* do nothing, default */
1003 } else if (!strcmp(mode, "native")) {
1004 *flags |= BDRV_O_NATIVE_AIO;
1005#ifdef CONFIG_LINUX_IO_URING
1006 } else if (!strcmp(mode, "io_uring")) {
1007 *flags |= BDRV_O_IO_URING;
1008#endif
1009 } else {
1010 return -1;
1011 }
1012
1013 return 0;
1014}
1015
9e8f1835
PB
1016/**
1017 * Set open flags for a given discard mode
1018 *
1019 * Return 0 on success, -1 if the discard mode was invalid.
1020 */
1021int bdrv_parse_discard_flags(const char *mode, int *flags)
1022{
1023 *flags &= ~BDRV_O_UNMAP;
1024
1025 if (!strcmp(mode, "off") || !strcmp(mode, "ignore")) {
1026 /* do nothing */
1027 } else if (!strcmp(mode, "on") || !strcmp(mode, "unmap")) {
1028 *flags |= BDRV_O_UNMAP;
1029 } else {
1030 return -1;
1031 }
1032
1033 return 0;
1034}
1035
c3993cdc
SH
1036/**
1037 * Set open flags for a given cache mode
1038 *
1039 * Return 0 on success, -1 if the cache mode was invalid.
1040 */
53e8ae01 1041int bdrv_parse_cache_mode(const char *mode, int *flags, bool *writethrough)
c3993cdc
SH
1042{
1043 *flags &= ~BDRV_O_CACHE_MASK;
1044
1045 if (!strcmp(mode, "off") || !strcmp(mode, "none")) {
53e8ae01
KW
1046 *writethrough = false;
1047 *flags |= BDRV_O_NOCACHE;
92196b2f 1048 } else if (!strcmp(mode, "directsync")) {
53e8ae01 1049 *writethrough = true;
92196b2f 1050 *flags |= BDRV_O_NOCACHE;
c3993cdc 1051 } else if (!strcmp(mode, "writeback")) {
53e8ae01 1052 *writethrough = false;
c3993cdc 1053 } else if (!strcmp(mode, "unsafe")) {
53e8ae01 1054 *writethrough = false;
c3993cdc
SH
1055 *flags |= BDRV_O_NO_FLUSH;
1056 } else if (!strcmp(mode, "writethrough")) {
53e8ae01 1057 *writethrough = true;
c3993cdc
SH
1058 } else {
1059 return -1;
1060 }
1061
1062 return 0;
1063}
1064
b5411555
KW
1065static char *bdrv_child_get_parent_desc(BdrvChild *c)
1066{
1067 BlockDriverState *parent = c->opaque;
1068 return g_strdup(bdrv_get_device_or_node_name(parent));
1069}
1070
20018e12
KW
1071static void bdrv_child_cb_drained_begin(BdrvChild *child)
1072{
1073 BlockDriverState *bs = child->opaque;
6cd5c9d7 1074 bdrv_do_drained_begin_quiesce(bs, NULL, false);
20018e12
KW
1075}
1076
89bd0305
KW
1077static bool bdrv_child_cb_drained_poll(BdrvChild *child)
1078{
1079 BlockDriverState *bs = child->opaque;
6cd5c9d7 1080 return bdrv_drain_poll(bs, false, NULL, false);
89bd0305
KW
1081}
1082
e037c09c
HR
1083static void bdrv_child_cb_drained_end(BdrvChild *child,
1084 int *drained_end_counter)
20018e12
KW
1085{
1086 BlockDriverState *bs = child->opaque;
e037c09c 1087 bdrv_drained_end_no_poll(bs, drained_end_counter);
20018e12
KW
1088}
1089
d736f119
KW
1090static void bdrv_child_cb_attach(BdrvChild *child)
1091{
1092 BlockDriverState *bs = child->opaque;
1093 bdrv_apply_subtree_drain(child, bs);
1094}
1095
1096static void bdrv_child_cb_detach(BdrvChild *child)
1097{
1098 BlockDriverState *bs = child->opaque;
1099 bdrv_unapply_subtree_drain(child, bs);
1100}
1101
38701b6a
KW
1102static int bdrv_child_cb_inactivate(BdrvChild *child)
1103{
1104 BlockDriverState *bs = child->opaque;
1105 assert(bs->open_flags & BDRV_O_INACTIVE);
1106 return 0;
1107}
1108
5d231849
KW
1109static bool bdrv_child_cb_can_set_aio_ctx(BdrvChild *child, AioContext *ctx,
1110 GSList **ignore, Error **errp)
1111{
1112 BlockDriverState *bs = child->opaque;
1113 return bdrv_can_set_aio_context(bs, ctx, ignore, errp);
1114}
1115
53a7d041
KW
1116static void bdrv_child_cb_set_aio_ctx(BdrvChild *child, AioContext *ctx,
1117 GSList **ignore)
1118{
1119 BlockDriverState *bs = child->opaque;
1120 return bdrv_set_aio_context_ignore(bs, ctx, ignore);
1121}
1122
b1e6fc08 1123/*
73176bee
KW
1124 * Returns the options and flags that a temporary snapshot should get, based on
1125 * the originally requested flags (the originally requested image will have
1126 * flags like a backing file)
b1e6fc08 1127 */
73176bee
KW
1128static void bdrv_temp_snapshot_options(int *child_flags, QDict *child_options,
1129 int parent_flags, QDict *parent_options)
b1e6fc08 1130{
73176bee
KW
1131 *child_flags = (parent_flags & ~BDRV_O_SNAPSHOT) | BDRV_O_TEMPORARY;
1132
1133 /* For temporary files, unconditional cache=unsafe is fine */
73176bee
KW
1134 qdict_set_default_str(child_options, BDRV_OPT_CACHE_DIRECT, "off");
1135 qdict_set_default_str(child_options, BDRV_OPT_CACHE_NO_FLUSH, "on");
41869044 1136
3f48686f 1137 /* Copy the read-only and discard options from the parent */
f87a0e29 1138 qdict_copy_default(child_options, parent_options, BDRV_OPT_READ_ONLY);
3f48686f 1139 qdict_copy_default(child_options, parent_options, BDRV_OPT_DISCARD);
f87a0e29 1140
41869044
KW
1141 /* aio=native doesn't work for cache.direct=off, so disable it for the
1142 * temporary snapshot */
1143 *child_flags &= ~BDRV_O_NATIVE_AIO;
b1e6fc08
KW
1144}
1145
0b50cc88 1146/*
8e2160e2
KW
1147 * Returns the options and flags that bs->file should get if a protocol driver
1148 * is expected, based on the given options and flags for the parent BDS
0b50cc88 1149 */
8e2160e2
KW
1150static void bdrv_inherited_options(int *child_flags, QDict *child_options,
1151 int parent_flags, QDict *parent_options)
0b50cc88 1152{
8e2160e2
KW
1153 int flags = parent_flags;
1154
0b50cc88
KW
1155 /* Enable protocol handling, disable format probing for bs->file */
1156 flags |= BDRV_O_PROTOCOL;
1157
91a097e7
KW
1158 /* If the cache mode isn't explicitly set, inherit direct and no-flush from
1159 * the parent. */
1160 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_DIRECT);
1161 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_NO_FLUSH);
5a9347c6 1162 qdict_copy_default(child_options, parent_options, BDRV_OPT_FORCE_SHARE);
91a097e7 1163
f87a0e29
AG
1164 /* Inherit the read-only option from the parent if it's not set */
1165 qdict_copy_default(child_options, parent_options, BDRV_OPT_READ_ONLY);
e35bdc12 1166 qdict_copy_default(child_options, parent_options, BDRV_OPT_AUTO_READ_ONLY);
f87a0e29 1167
0b50cc88 1168 /* Our block drivers take care to send flushes and respect unmap policy,
91a097e7
KW
1169 * so we can default to enable both on lower layers regardless of the
1170 * corresponding parent options. */
818584a4 1171 qdict_set_default_str(child_options, BDRV_OPT_DISCARD, "unmap");
0b50cc88 1172
0b50cc88 1173 /* Clear flags that only apply to the top layer */
abb06c5a
DB
1174 flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_COPY_ON_READ |
1175 BDRV_O_NO_IO);
0b50cc88 1176
8e2160e2 1177 *child_flags = flags;
0b50cc88
KW
1178}
1179
f3930ed0 1180const BdrvChildRole child_file = {
6cd5c9d7 1181 .parent_is_bds = true,
b5411555 1182 .get_parent_desc = bdrv_child_get_parent_desc,
8e2160e2 1183 .inherit_options = bdrv_inherited_options,
20018e12 1184 .drained_begin = bdrv_child_cb_drained_begin,
89bd0305 1185 .drained_poll = bdrv_child_cb_drained_poll,
20018e12 1186 .drained_end = bdrv_child_cb_drained_end,
d736f119
KW
1187 .attach = bdrv_child_cb_attach,
1188 .detach = bdrv_child_cb_detach,
38701b6a 1189 .inactivate = bdrv_child_cb_inactivate,
5d231849 1190 .can_set_aio_ctx = bdrv_child_cb_can_set_aio_ctx,
53a7d041 1191 .set_aio_ctx = bdrv_child_cb_set_aio_ctx,
f3930ed0
KW
1192};
1193
1194/*
8e2160e2
KW
1195 * Returns the options and flags that bs->file should get if the use of formats
1196 * (and not only protocols) is permitted for it, based on the given options and
1197 * flags for the parent BDS
f3930ed0 1198 */
8e2160e2
KW
1199static void bdrv_inherited_fmt_options(int *child_flags, QDict *child_options,
1200 int parent_flags, QDict *parent_options)
f3930ed0 1201{
8e2160e2
KW
1202 child_file.inherit_options(child_flags, child_options,
1203 parent_flags, parent_options);
1204
abb06c5a 1205 *child_flags &= ~(BDRV_O_PROTOCOL | BDRV_O_NO_IO);
f3930ed0
KW
1206}
1207
1208const BdrvChildRole child_format = {
6cd5c9d7 1209 .parent_is_bds = true,
b5411555 1210 .get_parent_desc = bdrv_child_get_parent_desc,
8e2160e2 1211 .inherit_options = bdrv_inherited_fmt_options,
20018e12 1212 .drained_begin = bdrv_child_cb_drained_begin,
89bd0305 1213 .drained_poll = bdrv_child_cb_drained_poll,
20018e12 1214 .drained_end = bdrv_child_cb_drained_end,
d736f119
KW
1215 .attach = bdrv_child_cb_attach,
1216 .detach = bdrv_child_cb_detach,
38701b6a 1217 .inactivate = bdrv_child_cb_inactivate,
5d231849 1218 .can_set_aio_ctx = bdrv_child_cb_can_set_aio_ctx,
53a7d041 1219 .set_aio_ctx = bdrv_child_cb_set_aio_ctx,
f3930ed0
KW
1220};
1221
db95dbba
KW
1222static void bdrv_backing_attach(BdrvChild *c)
1223{
1224 BlockDriverState *parent = c->opaque;
1225 BlockDriverState *backing_hd = c->bs;
1226
1227 assert(!parent->backing_blocker);
1228 error_setg(&parent->backing_blocker,
1229 "node is used as backing hd of '%s'",
1230 bdrv_get_device_or_node_name(parent));
1231
f30c66ba
HR
1232 bdrv_refresh_filename(backing_hd);
1233
db95dbba
KW
1234 parent->open_flags &= ~BDRV_O_NO_BACKING;
1235 pstrcpy(parent->backing_file, sizeof(parent->backing_file),
1236 backing_hd->filename);
1237 pstrcpy(parent->backing_format, sizeof(parent->backing_format),
1238 backing_hd->drv ? backing_hd->drv->format_name : "");
1239
1240 bdrv_op_block_all(backing_hd, parent->backing_blocker);
1241 /* Otherwise we won't be able to commit or stream */
1242 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_COMMIT_TARGET,
1243 parent->backing_blocker);
1244 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_STREAM,
1245 parent->backing_blocker);
1246 /*
1247 * We do backup in 3 ways:
1248 * 1. drive backup
1249 * The target bs is new opened, and the source is top BDS
1250 * 2. blockdev backup
1251 * Both the source and the target are top BDSes.
1252 * 3. internal backup(used for block replication)
1253 * Both the source and the target are backing file
1254 *
1255 * In case 1 and 2, neither the source nor the target is the backing file.
1256 * In case 3, we will block the top BDS, so there is only one block job
1257 * for the top BDS and its backing chain.
1258 */
1259 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_BACKUP_SOURCE,
1260 parent->backing_blocker);
1261 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_BACKUP_TARGET,
1262 parent->backing_blocker);
d736f119
KW
1263
1264 bdrv_child_cb_attach(c);
db95dbba
KW
1265}
1266
1267static void bdrv_backing_detach(BdrvChild *c)
1268{
1269 BlockDriverState *parent = c->opaque;
1270
1271 assert(parent->backing_blocker);
1272 bdrv_op_unblock_all(c->bs, parent->backing_blocker);
1273 error_free(parent->backing_blocker);
1274 parent->backing_blocker = NULL;
d736f119
KW
1275
1276 bdrv_child_cb_detach(c);
db95dbba
KW
1277}
1278
317fc44e 1279/*
8e2160e2
KW
1280 * Returns the options and flags that bs->backing should get, based on the
1281 * given options and flags for the parent BDS
317fc44e 1282 */
8e2160e2
KW
1283static void bdrv_backing_options(int *child_flags, QDict *child_options,
1284 int parent_flags, QDict *parent_options)
317fc44e 1285{
8e2160e2
KW
1286 int flags = parent_flags;
1287
b8816a43
KW
1288 /* The cache mode is inherited unmodified for backing files; except WCE,
1289 * which is only applied on the top level (BlockBackend) */
91a097e7
KW
1290 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_DIRECT);
1291 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_NO_FLUSH);
5a9347c6 1292 qdict_copy_default(child_options, parent_options, BDRV_OPT_FORCE_SHARE);
91a097e7 1293
317fc44e 1294 /* backing files always opened read-only */
f87a0e29 1295 qdict_set_default_str(child_options, BDRV_OPT_READ_ONLY, "on");
e35bdc12 1296 qdict_set_default_str(child_options, BDRV_OPT_AUTO_READ_ONLY, "off");
f87a0e29 1297 flags &= ~BDRV_O_COPY_ON_READ;
317fc44e
KW
1298
1299 /* snapshot=on is handled on the top layer */
8bfea15d 1300 flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_TEMPORARY);
317fc44e 1301
8e2160e2 1302 *child_flags = flags;
317fc44e
KW
1303}
1304
6858eba0
KW
1305static int bdrv_backing_update_filename(BdrvChild *c, BlockDriverState *base,
1306 const char *filename, Error **errp)
1307{
1308 BlockDriverState *parent = c->opaque;
e94d3dba 1309 bool read_only = bdrv_is_read_only(parent);
6858eba0
KW
1310 int ret;
1311
e94d3dba
AG
1312 if (read_only) {
1313 ret = bdrv_reopen_set_read_only(parent, false, errp);
61f09cea
KW
1314 if (ret < 0) {
1315 return ret;
1316 }
1317 }
1318
6858eba0
KW
1319 ret = bdrv_change_backing_file(parent, filename,
1320 base->drv ? base->drv->format_name : "");
1321 if (ret < 0) {
64730694 1322 error_setg_errno(errp, -ret, "Could not update backing file link");
6858eba0
KW
1323 }
1324
e94d3dba
AG
1325 if (read_only) {
1326 bdrv_reopen_set_read_only(parent, true, NULL);
61f09cea
KW
1327 }
1328
6858eba0
KW
1329 return ret;
1330}
1331
91ef3825 1332const BdrvChildRole child_backing = {
6cd5c9d7 1333 .parent_is_bds = true,
b5411555 1334 .get_parent_desc = bdrv_child_get_parent_desc,
db95dbba
KW
1335 .attach = bdrv_backing_attach,
1336 .detach = bdrv_backing_detach,
8e2160e2 1337 .inherit_options = bdrv_backing_options,
20018e12 1338 .drained_begin = bdrv_child_cb_drained_begin,
89bd0305 1339 .drained_poll = bdrv_child_cb_drained_poll,
20018e12 1340 .drained_end = bdrv_child_cb_drained_end,
38701b6a 1341 .inactivate = bdrv_child_cb_inactivate,
6858eba0 1342 .update_filename = bdrv_backing_update_filename,
5d231849 1343 .can_set_aio_ctx = bdrv_child_cb_can_set_aio_ctx,
53a7d041 1344 .set_aio_ctx = bdrv_child_cb_set_aio_ctx,
f3930ed0
KW
1345};
1346
7b272452
KW
1347static int bdrv_open_flags(BlockDriverState *bs, int flags)
1348{
61de4c68 1349 int open_flags = flags;
7b272452
KW
1350
1351 /*
1352 * Clear flags that are internal to the block layer before opening the
1353 * image.
1354 */
20cca275 1355 open_flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_PROTOCOL);
7b272452 1356
7b272452
KW
1357 return open_flags;
1358}
1359
91a097e7
KW
1360static void update_flags_from_options(int *flags, QemuOpts *opts)
1361{
2a3d4331 1362 *flags &= ~(BDRV_O_CACHE_MASK | BDRV_O_RDWR | BDRV_O_AUTO_RDONLY);
91a097e7 1363
57f9db9a 1364 if (qemu_opt_get_bool_del(opts, BDRV_OPT_CACHE_NO_FLUSH, false)) {
91a097e7
KW
1365 *flags |= BDRV_O_NO_FLUSH;
1366 }
1367
57f9db9a 1368 if (qemu_opt_get_bool_del(opts, BDRV_OPT_CACHE_DIRECT, false)) {
91a097e7
KW
1369 *flags |= BDRV_O_NOCACHE;
1370 }
f87a0e29 1371
57f9db9a 1372 if (!qemu_opt_get_bool_del(opts, BDRV_OPT_READ_ONLY, false)) {
f87a0e29
AG
1373 *flags |= BDRV_O_RDWR;
1374 }
1375
e35bdc12
KW
1376 if (qemu_opt_get_bool_del(opts, BDRV_OPT_AUTO_READ_ONLY, false)) {
1377 *flags |= BDRV_O_AUTO_RDONLY;
1378 }
91a097e7
KW
1379}
1380
1381static void update_options_from_flags(QDict *options, int flags)
1382{
91a097e7 1383 if (!qdict_haskey(options, BDRV_OPT_CACHE_DIRECT)) {
46f5ac20 1384 qdict_put_bool(options, BDRV_OPT_CACHE_DIRECT, flags & BDRV_O_NOCACHE);
91a097e7
KW
1385 }
1386 if (!qdict_haskey(options, BDRV_OPT_CACHE_NO_FLUSH)) {
46f5ac20
EB
1387 qdict_put_bool(options, BDRV_OPT_CACHE_NO_FLUSH,
1388 flags & BDRV_O_NO_FLUSH);
91a097e7 1389 }
f87a0e29 1390 if (!qdict_haskey(options, BDRV_OPT_READ_ONLY)) {
46f5ac20 1391 qdict_put_bool(options, BDRV_OPT_READ_ONLY, !(flags & BDRV_O_RDWR));
f87a0e29 1392 }
e35bdc12
KW
1393 if (!qdict_haskey(options, BDRV_OPT_AUTO_READ_ONLY)) {
1394 qdict_put_bool(options, BDRV_OPT_AUTO_READ_ONLY,
1395 flags & BDRV_O_AUTO_RDONLY);
1396 }
91a097e7
KW
1397}
1398
636ea370
KW
1399static void bdrv_assign_node_name(BlockDriverState *bs,
1400 const char *node_name,
1401 Error **errp)
6913c0c2 1402{
15489c76 1403 char *gen_node_name = NULL;
6913c0c2 1404
15489c76
JC
1405 if (!node_name) {
1406 node_name = gen_node_name = id_generate(ID_BLOCK);
1407 } else if (!id_wellformed(node_name)) {
1408 /*
1409 * Check for empty string or invalid characters, but not if it is
1410 * generated (generated names use characters not available to the user)
1411 */
9aebf3b8 1412 error_setg(errp, "Invalid node name");
636ea370 1413 return;
6913c0c2
BC
1414 }
1415
0c5e94ee 1416 /* takes care of avoiding namespaces collisions */
7f06d47e 1417 if (blk_by_name(node_name)) {
0c5e94ee
BC
1418 error_setg(errp, "node-name=%s is conflicting with a device id",
1419 node_name);
15489c76 1420 goto out;
0c5e94ee
BC
1421 }
1422
6913c0c2
BC
1423 /* takes care of avoiding duplicates node names */
1424 if (bdrv_find_node(node_name)) {
1425 error_setg(errp, "Duplicate node name");
15489c76 1426 goto out;
6913c0c2
BC
1427 }
1428
824808dd
KW
1429 /* Make sure that the node name isn't truncated */
1430 if (strlen(node_name) >= sizeof(bs->node_name)) {
1431 error_setg(errp, "Node name too long");
1432 goto out;
1433 }
1434
6913c0c2
BC
1435 /* copy node name into the bs and insert it into the graph list */
1436 pstrcpy(bs->node_name, sizeof(bs->node_name), node_name);
1437 QTAILQ_INSERT_TAIL(&graph_bdrv_states, bs, node_list);
15489c76
JC
1438out:
1439 g_free(gen_node_name);
6913c0c2
BC
1440}
1441
01a56501
KW
1442static int bdrv_open_driver(BlockDriverState *bs, BlockDriver *drv,
1443 const char *node_name, QDict *options,
1444 int open_flags, Error **errp)
1445{
1446 Error *local_err = NULL;
0f12264e 1447 int i, ret;
01a56501
KW
1448
1449 bdrv_assign_node_name(bs, node_name, &local_err);
1450 if (local_err) {
1451 error_propagate(errp, local_err);
1452 return -EINVAL;
1453 }
1454
1455 bs->drv = drv;
680c7f96 1456 bs->read_only = !(bs->open_flags & BDRV_O_RDWR);
01a56501
KW
1457 bs->opaque = g_malloc0(drv->instance_size);
1458
1459 if (drv->bdrv_file_open) {
1460 assert(!drv->bdrv_needs_filename || bs->filename[0]);
1461 ret = drv->bdrv_file_open(bs, options, open_flags, &local_err);
680c7f96 1462 } else if (drv->bdrv_open) {
01a56501 1463 ret = drv->bdrv_open(bs, options, open_flags, &local_err);
680c7f96
KW
1464 } else {
1465 ret = 0;
01a56501
KW
1466 }
1467
1468 if (ret < 0) {
1469 if (local_err) {
1470 error_propagate(errp, local_err);
1471 } else if (bs->filename[0]) {
1472 error_setg_errno(errp, -ret, "Could not open '%s'", bs->filename);
1473 } else {
1474 error_setg_errno(errp, -ret, "Could not open image");
1475 }
180ca19a 1476 goto open_failed;
01a56501
KW
1477 }
1478
1479 ret = refresh_total_sectors(bs, bs->total_sectors);
1480 if (ret < 0) {
1481 error_setg_errno(errp, -ret, "Could not refresh total sector count");
180ca19a 1482 return ret;
01a56501
KW
1483 }
1484
1485 bdrv_refresh_limits(bs, &local_err);
1486 if (local_err) {
1487 error_propagate(errp, local_err);
180ca19a 1488 return -EINVAL;
01a56501
KW
1489 }
1490
1491 assert(bdrv_opt_mem_align(bs) != 0);
1492 assert(bdrv_min_mem_align(bs) != 0);
1493 assert(is_power_of_2(bs->bl.request_alignment));
1494
0f12264e
KW
1495 for (i = 0; i < bs->quiesce_counter; i++) {
1496 if (drv->bdrv_co_drain_begin) {
1497 drv->bdrv_co_drain_begin(bs);
1498 }
1499 }
1500
01a56501 1501 return 0;
180ca19a
MP
1502open_failed:
1503 bs->drv = NULL;
1504 if (bs->file != NULL) {
1505 bdrv_unref_child(bs, bs->file);
1506 bs->file = NULL;
1507 }
01a56501
KW
1508 g_free(bs->opaque);
1509 bs->opaque = NULL;
01a56501
KW
1510 return ret;
1511}
1512
680c7f96
KW
1513BlockDriverState *bdrv_new_open_driver(BlockDriver *drv, const char *node_name,
1514 int flags, Error **errp)
1515{
1516 BlockDriverState *bs;
1517 int ret;
1518
1519 bs = bdrv_new();
1520 bs->open_flags = flags;
1521 bs->explicit_options = qdict_new();
1522 bs->options = qdict_new();
1523 bs->opaque = NULL;
1524
1525 update_options_from_flags(bs->options, flags);
1526
1527 ret = bdrv_open_driver(bs, drv, node_name, bs->options, flags, errp);
1528 if (ret < 0) {
cb3e7f08 1529 qobject_unref(bs->explicit_options);
180ca19a 1530 bs->explicit_options = NULL;
cb3e7f08 1531 qobject_unref(bs->options);
180ca19a 1532 bs->options = NULL;
680c7f96
KW
1533 bdrv_unref(bs);
1534 return NULL;
1535 }
1536
1537 return bs;
1538}
1539
c5f3014b 1540QemuOptsList bdrv_runtime_opts = {
18edf289
KW
1541 .name = "bdrv_common",
1542 .head = QTAILQ_HEAD_INITIALIZER(bdrv_runtime_opts.head),
1543 .desc = {
1544 {
1545 .name = "node-name",
1546 .type = QEMU_OPT_STRING,
1547 .help = "Node name of the block device node",
1548 },
62392ebb
KW
1549 {
1550 .name = "driver",
1551 .type = QEMU_OPT_STRING,
1552 .help = "Block driver to use for the node",
1553 },
91a097e7
KW
1554 {
1555 .name = BDRV_OPT_CACHE_DIRECT,
1556 .type = QEMU_OPT_BOOL,
1557 .help = "Bypass software writeback cache on the host",
1558 },
1559 {
1560 .name = BDRV_OPT_CACHE_NO_FLUSH,
1561 .type = QEMU_OPT_BOOL,
1562 .help = "Ignore flush requests",
1563 },
f87a0e29
AG
1564 {
1565 .name = BDRV_OPT_READ_ONLY,
1566 .type = QEMU_OPT_BOOL,
1567 .help = "Node is opened in read-only mode",
1568 },
e35bdc12
KW
1569 {
1570 .name = BDRV_OPT_AUTO_READ_ONLY,
1571 .type = QEMU_OPT_BOOL,
1572 .help = "Node can become read-only if opening read-write fails",
1573 },
692e01a2
KW
1574 {
1575 .name = "detect-zeroes",
1576 .type = QEMU_OPT_STRING,
1577 .help = "try to optimize zero writes (off, on, unmap)",
1578 },
818584a4 1579 {
415bbca8 1580 .name = BDRV_OPT_DISCARD,
818584a4
KW
1581 .type = QEMU_OPT_STRING,
1582 .help = "discard operation (ignore/off, unmap/on)",
1583 },
5a9347c6
FZ
1584 {
1585 .name = BDRV_OPT_FORCE_SHARE,
1586 .type = QEMU_OPT_BOOL,
1587 .help = "always accept other writers (default: off)",
1588 },
18edf289
KW
1589 { /* end of list */ }
1590 },
1591};
1592
fd17146c
HR
1593static QemuOptsList fallback_create_opts = {
1594 .name = "fallback-create-opts",
1595 .head = QTAILQ_HEAD_INITIALIZER(fallback_create_opts.head),
1596 .desc = {
1597 {
1598 .name = BLOCK_OPT_SIZE,
1599 .type = QEMU_OPT_SIZE,
1600 .help = "Virtual disk size"
1601 },
1602 {
1603 .name = BLOCK_OPT_PREALLOC,
1604 .type = QEMU_OPT_STRING,
1605 .help = "Preallocation mode (allowed values: off)"
1606 },
1607 { /* end of list */ }
1608 }
1609};
1610
57915332
KW
1611/*
1612 * Common part for opening disk images and files
b6ad491a
KW
1613 *
1614 * Removes all processed options from *options.
57915332 1615 */
5696c6e3 1616static int bdrv_open_common(BlockDriverState *bs, BlockBackend *file,
82dc8b41 1617 QDict *options, Error **errp)
57915332
KW
1618{
1619 int ret, open_flags;
035fccdf 1620 const char *filename;
62392ebb 1621 const char *driver_name = NULL;
6913c0c2 1622 const char *node_name = NULL;
818584a4 1623 const char *discard;
18edf289 1624 QemuOpts *opts;
62392ebb 1625 BlockDriver *drv;
34b5d2c6 1626 Error *local_err = NULL;
57915332 1627
6405875c 1628 assert(bs->file == NULL);
707ff828 1629 assert(options != NULL && bs->options != options);
57915332 1630
62392ebb
KW
1631 opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
1632 qemu_opts_absorb_qdict(opts, options, &local_err);
1633 if (local_err) {
1634 error_propagate(errp, local_err);
1635 ret = -EINVAL;
1636 goto fail_opts;
1637 }
1638
9b7e8691
AG
1639 update_flags_from_options(&bs->open_flags, opts);
1640
62392ebb
KW
1641 driver_name = qemu_opt_get(opts, "driver");
1642 drv = bdrv_find_format(driver_name);
1643 assert(drv != NULL);
1644
5a9347c6
FZ
1645 bs->force_share = qemu_opt_get_bool(opts, BDRV_OPT_FORCE_SHARE, false);
1646
1647 if (bs->force_share && (bs->open_flags & BDRV_O_RDWR)) {
1648 error_setg(errp,
1649 BDRV_OPT_FORCE_SHARE
1650 "=on can only be used with read-only images");
1651 ret = -EINVAL;
1652 goto fail_opts;
1653 }
1654
45673671 1655 if (file != NULL) {
f30c66ba 1656 bdrv_refresh_filename(blk_bs(file));
5696c6e3 1657 filename = blk_bs(file)->filename;
45673671 1658 } else {
129c7d1c
MA
1659 /*
1660 * Caution: while qdict_get_try_str() is fine, getting
1661 * non-string types would require more care. When @options
1662 * come from -blockdev or blockdev_add, its members are typed
1663 * according to the QAPI schema, but when they come from
1664 * -drive, they're all QString.
1665 */
45673671
KW
1666 filename = qdict_get_try_str(options, "filename");
1667 }
1668
4a008240 1669 if (drv->bdrv_needs_filename && (!filename || !filename[0])) {
765003db
KW
1670 error_setg(errp, "The '%s' block driver requires a file name",
1671 drv->format_name);
18edf289
KW
1672 ret = -EINVAL;
1673 goto fail_opts;
6913c0c2 1674 }
6913c0c2 1675
82dc8b41
KW
1676 trace_bdrv_open_common(bs, filename ?: "", bs->open_flags,
1677 drv->format_name);
62392ebb 1678
82dc8b41 1679 bs->read_only = !(bs->open_flags & BDRV_O_RDWR);
b64ec4e4
FZ
1680
1681 if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv, bs->read_only)) {
8be25de6
KW
1682 if (!bs->read_only && bdrv_is_whitelisted(drv, true)) {
1683 ret = bdrv_apply_auto_read_only(bs, NULL, NULL);
1684 } else {
1685 ret = -ENOTSUP;
1686 }
1687 if (ret < 0) {
1688 error_setg(errp,
1689 !bs->read_only && bdrv_is_whitelisted(drv, true)
1690 ? "Driver '%s' can only be used for read-only devices"
1691 : "Driver '%s' is not whitelisted",
1692 drv->format_name);
1693 goto fail_opts;
1694 }
b64ec4e4 1695 }
57915332 1696
d3faa13e
PB
1697 /* bdrv_new() and bdrv_close() make it so */
1698 assert(atomic_read(&bs->copy_on_read) == 0);
1699
82dc8b41 1700 if (bs->open_flags & BDRV_O_COPY_ON_READ) {
0ebd24e0
KW
1701 if (!bs->read_only) {
1702 bdrv_enable_copy_on_read(bs);
1703 } else {
1704 error_setg(errp, "Can't use copy-on-read on read-only device");
18edf289
KW
1705 ret = -EINVAL;
1706 goto fail_opts;
0ebd24e0 1707 }
53fec9d3
SH
1708 }
1709
415bbca8 1710 discard = qemu_opt_get(opts, BDRV_OPT_DISCARD);
818584a4
KW
1711 if (discard != NULL) {
1712 if (bdrv_parse_discard_flags(discard, &bs->open_flags) != 0) {
1713 error_setg(errp, "Invalid discard option");
1714 ret = -EINVAL;
1715 goto fail_opts;
1716 }
1717 }
1718
543770bd
AG
1719 bs->detect_zeroes =
1720 bdrv_parse_detect_zeroes(opts, bs->open_flags, &local_err);
1721 if (local_err) {
1722 error_propagate(errp, local_err);
1723 ret = -EINVAL;
1724 goto fail_opts;
692e01a2
KW
1725 }
1726
c2ad1b0c
KW
1727 if (filename != NULL) {
1728 pstrcpy(bs->filename, sizeof(bs->filename), filename);
1729 } else {
1730 bs->filename[0] = '\0';
1731 }
91af7014 1732 pstrcpy(bs->exact_filename, sizeof(bs->exact_filename), bs->filename);
57915332 1733
66f82cee 1734 /* Open the image, either directly or using a protocol */
82dc8b41 1735 open_flags = bdrv_open_flags(bs, bs->open_flags);
01a56501 1736 node_name = qemu_opt_get(opts, "node-name");
57915332 1737
01a56501
KW
1738 assert(!drv->bdrv_file_open || file == NULL);
1739 ret = bdrv_open_driver(bs, drv, node_name, options, open_flags, errp);
51762288 1740 if (ret < 0) {
01a56501 1741 goto fail_opts;
3baca891
KW
1742 }
1743
18edf289 1744 qemu_opts_del(opts);
57915332
KW
1745 return 0;
1746
18edf289
KW
1747fail_opts:
1748 qemu_opts_del(opts);
57915332
KW
1749 return ret;
1750}
1751
5e5c4f63
KW
1752static QDict *parse_json_filename(const char *filename, Error **errp)
1753{
1754 QObject *options_obj;
1755 QDict *options;
1756 int ret;
1757
1758 ret = strstart(filename, "json:", &filename);
1759 assert(ret);
1760
5577fff7 1761 options_obj = qobject_from_json(filename, errp);
5e5c4f63 1762 if (!options_obj) {
5577fff7 1763 error_prepend(errp, "Could not parse the JSON options: ");
5e5c4f63
KW
1764 return NULL;
1765 }
1766
7dc847eb 1767 options = qobject_to(QDict, options_obj);
ca6b6e1e 1768 if (!options) {
cb3e7f08 1769 qobject_unref(options_obj);
5e5c4f63
KW
1770 error_setg(errp, "Invalid JSON object given");
1771 return NULL;
1772 }
1773
5e5c4f63
KW
1774 qdict_flatten(options);
1775
1776 return options;
1777}
1778
de3b53f0
KW
1779static void parse_json_protocol(QDict *options, const char **pfilename,
1780 Error **errp)
1781{
1782 QDict *json_options;
1783 Error *local_err = NULL;
1784
1785 /* Parse json: pseudo-protocol */
1786 if (!*pfilename || !g_str_has_prefix(*pfilename, "json:")) {
1787 return;
1788 }
1789
1790 json_options = parse_json_filename(*pfilename, &local_err);
1791 if (local_err) {
1792 error_propagate(errp, local_err);
1793 return;
1794 }
1795
1796 /* Options given in the filename have lower priority than options
1797 * specified directly */
1798 qdict_join(options, json_options, false);
cb3e7f08 1799 qobject_unref(json_options);
de3b53f0
KW
1800 *pfilename = NULL;
1801}
1802
b6ce07aa 1803/*
f54120ff
KW
1804 * Fills in default options for opening images and converts the legacy
1805 * filename/flags pair to option QDict entries.
53a29513
HR
1806 * The BDRV_O_PROTOCOL flag in *flags will be set or cleared accordingly if a
1807 * block driver has been specified explicitly.
b6ce07aa 1808 */
de3b53f0 1809static int bdrv_fill_options(QDict **options, const char *filename,
053e1578 1810 int *flags, Error **errp)
ea2384d3 1811{
c2ad1b0c 1812 const char *drvname;
53a29513 1813 bool protocol = *flags & BDRV_O_PROTOCOL;
e3fa4bfa 1814 bool parse_filename = false;
053e1578 1815 BlockDriver *drv = NULL;
34b5d2c6 1816 Error *local_err = NULL;
83f64091 1817
129c7d1c
MA
1818 /*
1819 * Caution: while qdict_get_try_str() is fine, getting non-string
1820 * types would require more care. When @options come from
1821 * -blockdev or blockdev_add, its members are typed according to
1822 * the QAPI schema, but when they come from -drive, they're all
1823 * QString.
1824 */
53a29513 1825 drvname = qdict_get_try_str(*options, "driver");
053e1578
HR
1826 if (drvname) {
1827 drv = bdrv_find_format(drvname);
1828 if (!drv) {
1829 error_setg(errp, "Unknown driver '%s'", drvname);
1830 return -ENOENT;
1831 }
1832 /* If the user has explicitly specified the driver, this choice should
1833 * override the BDRV_O_PROTOCOL flag */
1834 protocol = drv->bdrv_file_open;
53a29513
HR
1835 }
1836
1837 if (protocol) {
1838 *flags |= BDRV_O_PROTOCOL;
1839 } else {
1840 *flags &= ~BDRV_O_PROTOCOL;
1841 }
1842
91a097e7
KW
1843 /* Translate cache options from flags into options */
1844 update_options_from_flags(*options, *flags);
1845
035fccdf 1846 /* Fetch the file name from the options QDict if necessary */
17b005f1 1847 if (protocol && filename) {
f54120ff 1848 if (!qdict_haskey(*options, "filename")) {
46f5ac20 1849 qdict_put_str(*options, "filename", filename);
f54120ff
KW
1850 parse_filename = true;
1851 } else {
1852 error_setg(errp, "Can't specify 'file' and 'filename' options at "
1853 "the same time");
1854 return -EINVAL;
1855 }
035fccdf
KW
1856 }
1857
c2ad1b0c 1858 /* Find the right block driver */
129c7d1c 1859 /* See cautionary note on accessing @options above */
f54120ff 1860 filename = qdict_get_try_str(*options, "filename");
f54120ff 1861
053e1578
HR
1862 if (!drvname && protocol) {
1863 if (filename) {
1864 drv = bdrv_find_protocol(filename, parse_filename, errp);
17b005f1 1865 if (!drv) {
053e1578 1866 return -EINVAL;
17b005f1 1867 }
053e1578
HR
1868
1869 drvname = drv->format_name;
46f5ac20 1870 qdict_put_str(*options, "driver", drvname);
053e1578
HR
1871 } else {
1872 error_setg(errp, "Must specify either driver or file");
1873 return -EINVAL;
98289620 1874 }
c2ad1b0c
KW
1875 }
1876
17b005f1 1877 assert(drv || !protocol);
c2ad1b0c 1878
f54120ff 1879 /* Driver-specific filename parsing */
17b005f1 1880 if (drv && drv->bdrv_parse_filename && parse_filename) {
5acd9d81 1881 drv->bdrv_parse_filename(filename, *options, &local_err);
84d18f06 1882 if (local_err) {
34b5d2c6 1883 error_propagate(errp, local_err);
f54120ff 1884 return -EINVAL;
6963a30d 1885 }
cd5d031e
HR
1886
1887 if (!drv->bdrv_needs_filename) {
1888 qdict_del(*options, "filename");
cd5d031e 1889 }
6963a30d
KW
1890 }
1891
f54120ff
KW
1892 return 0;
1893}
1894
3121fb45
KW
1895static int bdrv_child_check_perm(BdrvChild *c, BlockReopenQueue *q,
1896 uint64_t perm, uint64_t shared,
9eab1544
HR
1897 GSList *ignore_children,
1898 bool *tighten_restrictions, Error **errp);
c1cef672
FZ
1899static void bdrv_child_abort_perm_update(BdrvChild *c);
1900static void bdrv_child_set_perm(BdrvChild *c, uint64_t perm, uint64_t shared);
1901
148eb13c
KW
1902typedef struct BlockReopenQueueEntry {
1903 bool prepared;
69b736e7 1904 bool perms_checked;
148eb13c 1905 BDRVReopenState state;
859443b0 1906 QTAILQ_ENTRY(BlockReopenQueueEntry) entry;
148eb13c
KW
1907} BlockReopenQueueEntry;
1908
1909/*
1910 * Return the flags that @bs will have after the reopens in @q have
1911 * successfully completed. If @q is NULL (or @bs is not contained in @q),
1912 * return the current flags.
1913 */
1914static int bdrv_reopen_get_flags(BlockReopenQueue *q, BlockDriverState *bs)
1915{
1916 BlockReopenQueueEntry *entry;
1917
1918 if (q != NULL) {
859443b0 1919 QTAILQ_FOREACH(entry, q, entry) {
148eb13c
KW
1920 if (entry->state.bs == bs) {
1921 return entry->state.flags;
1922 }
1923 }
1924 }
1925
1926 return bs->open_flags;
1927}
1928
1929/* Returns whether the image file can be written to after the reopen queue @q
1930 * has been successfully applied, or right now if @q is NULL. */
cc022140
HR
1931static bool bdrv_is_writable_after_reopen(BlockDriverState *bs,
1932 BlockReopenQueue *q)
148eb13c
KW
1933{
1934 int flags = bdrv_reopen_get_flags(q, bs);
1935
1936 return (flags & (BDRV_O_RDWR | BDRV_O_INACTIVE)) == BDRV_O_RDWR;
1937}
1938
cc022140
HR
1939/*
1940 * Return whether the BDS can be written to. This is not necessarily
1941 * the same as !bdrv_is_read_only(bs), as inactivated images may not
1942 * be written to but do not count as read-only images.
1943 */
1944bool bdrv_is_writable(BlockDriverState *bs)
1945{
1946 return bdrv_is_writable_after_reopen(bs, NULL);
1947}
1948
ffd1a5a2 1949static void bdrv_child_perm(BlockDriverState *bs, BlockDriverState *child_bs,
e0995dc3
KW
1950 BdrvChild *c, const BdrvChildRole *role,
1951 BlockReopenQueue *reopen_queue,
ffd1a5a2
FZ
1952 uint64_t parent_perm, uint64_t parent_shared,
1953 uint64_t *nperm, uint64_t *nshared)
1954{
0b3ca76e
AG
1955 assert(bs->drv && bs->drv->bdrv_child_perm);
1956 bs->drv->bdrv_child_perm(bs, c, role, reopen_queue,
1957 parent_perm, parent_shared,
1958 nperm, nshared);
e0995dc3 1959 /* TODO Take force_share from reopen_queue */
ffd1a5a2
FZ
1960 if (child_bs && child_bs->force_share) {
1961 *nshared = BLK_PERM_ALL;
1962 }
1963}
1964
33a610c3
KW
1965/*
1966 * Check whether permissions on this node can be changed in a way that
1967 * @cumulative_perms and @cumulative_shared_perms are the new cumulative
1968 * permissions of all its parents. This involves checking whether all necessary
1969 * permission changes to child nodes can be performed.
1970 *
9eab1544
HR
1971 * Will set *tighten_restrictions to true if and only if new permissions have to
1972 * be taken or currently shared permissions are to be unshared. Otherwise,
1973 * errors are not fatal as long as the caller accepts that the restrictions
1974 * remain tighter than they need to be. The caller still has to abort the
1975 * transaction.
1976 * @tighten_restrictions cannot be used together with @q: When reopening, we may
1977 * encounter fatal errors even though no restrictions are to be tightened. For
1978 * example, changing a node from RW to RO will fail if the WRITE permission is
1979 * to be kept.
1980 *
33a610c3
KW
1981 * A call to this function must always be followed by a call to bdrv_set_perm()
1982 * or bdrv_abort_perm_update().
1983 */
3121fb45
KW
1984static int bdrv_check_perm(BlockDriverState *bs, BlockReopenQueue *q,
1985 uint64_t cumulative_perms,
46181129 1986 uint64_t cumulative_shared_perms,
9eab1544
HR
1987 GSList *ignore_children,
1988 bool *tighten_restrictions, Error **errp)
33a610c3
KW
1989{
1990 BlockDriver *drv = bs->drv;
1991 BdrvChild *c;
1992 int ret;
1993
9eab1544
HR
1994 assert(!q || !tighten_restrictions);
1995
1996 if (tighten_restrictions) {
1997 uint64_t current_perms, current_shared;
1998 uint64_t added_perms, removed_shared_perms;
1999
2000 bdrv_get_cumulative_perm(bs, &current_perms, &current_shared);
2001
2002 added_perms = cumulative_perms & ~current_perms;
2003 removed_shared_perms = current_shared & ~cumulative_shared_perms;
2004
2005 *tighten_restrictions = added_perms || removed_shared_perms;
2006 }
2007
33a610c3
KW
2008 /* Write permissions never work with read-only images */
2009 if ((cumulative_perms & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)) &&
cc022140 2010 !bdrv_is_writable_after_reopen(bs, q))
33a610c3 2011 {
481e0eee
HR
2012 if (!bdrv_is_writable_after_reopen(bs, NULL)) {
2013 error_setg(errp, "Block node is read-only");
2014 } else {
2015 uint64_t current_perms, current_shared;
2016 bdrv_get_cumulative_perm(bs, &current_perms, &current_shared);
2017 if (current_perms & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)) {
2018 error_setg(errp, "Cannot make block node read-only, there is "
2019 "a writer on it");
2020 } else {
2021 error_setg(errp, "Cannot make block node read-only and create "
2022 "a writer on it");
2023 }
2024 }
2025
33a610c3
KW
2026 return -EPERM;
2027 }
2028
2029 /* Check this node */
2030 if (!drv) {
2031 return 0;
2032 }
2033
2034 if (drv->bdrv_check_perm) {
2035 return drv->bdrv_check_perm(bs, cumulative_perms,
2036 cumulative_shared_perms, errp);
2037 }
2038
78e421c9 2039 /* Drivers that never have children can omit .bdrv_child_perm() */
33a610c3 2040 if (!drv->bdrv_child_perm) {
78e421c9 2041 assert(QLIST_EMPTY(&bs->children));
33a610c3
KW
2042 return 0;
2043 }
2044
2045 /* Check all children */
2046 QLIST_FOREACH(c, &bs->children, next) {
2047 uint64_t cur_perm, cur_shared;
9eab1544
HR
2048 bool child_tighten_restr;
2049
3121fb45 2050 bdrv_child_perm(bs, c->bs, c, c->role, q,
ffd1a5a2
FZ
2051 cumulative_perms, cumulative_shared_perms,
2052 &cur_perm, &cur_shared);
9eab1544
HR
2053 ret = bdrv_child_check_perm(c, q, cur_perm, cur_shared, ignore_children,
2054 tighten_restrictions ? &child_tighten_restr
2055 : NULL,
2056 errp);
2057 if (tighten_restrictions) {
2058 *tighten_restrictions |= child_tighten_restr;
2059 }
33a610c3
KW
2060 if (ret < 0) {
2061 return ret;
2062 }
2063 }
2064
2065 return 0;
2066}
2067
2068/*
2069 * Notifies drivers that after a previous bdrv_check_perm() call, the
2070 * permission update is not performed and any preparations made for it (e.g.
2071 * taken file locks) need to be undone.
2072 *
2073 * This function recursively notifies all child nodes.
2074 */
2075static void bdrv_abort_perm_update(BlockDriverState *bs)
2076{
2077 BlockDriver *drv = bs->drv;
2078 BdrvChild *c;
2079
2080 if (!drv) {
2081 return;
2082 }
2083
2084 if (drv->bdrv_abort_perm_update) {
2085 drv->bdrv_abort_perm_update(bs);
2086 }
2087
2088 QLIST_FOREACH(c, &bs->children, next) {
2089 bdrv_child_abort_perm_update(c);
2090 }
2091}
2092
2093static void bdrv_set_perm(BlockDriverState *bs, uint64_t cumulative_perms,
2094 uint64_t cumulative_shared_perms)
2095{
2096 BlockDriver *drv = bs->drv;
2097 BdrvChild *c;
2098
2099 if (!drv) {
2100 return;
2101 }
2102
2103 /* Update this node */
2104 if (drv->bdrv_set_perm) {
2105 drv->bdrv_set_perm(bs, cumulative_perms, cumulative_shared_perms);
2106 }
2107
78e421c9 2108 /* Drivers that never have children can omit .bdrv_child_perm() */
33a610c3 2109 if (!drv->bdrv_child_perm) {
78e421c9 2110 assert(QLIST_EMPTY(&bs->children));
33a610c3
KW
2111 return;
2112 }
2113
2114 /* Update all children */
2115 QLIST_FOREACH(c, &bs->children, next) {
2116 uint64_t cur_perm, cur_shared;
e0995dc3 2117 bdrv_child_perm(bs, c->bs, c, c->role, NULL,
ffd1a5a2
FZ
2118 cumulative_perms, cumulative_shared_perms,
2119 &cur_perm, &cur_shared);
33a610c3
KW
2120 bdrv_child_set_perm(c, cur_perm, cur_shared);
2121 }
2122}
2123
c7a0f2be
KW
2124void bdrv_get_cumulative_perm(BlockDriverState *bs, uint64_t *perm,
2125 uint64_t *shared_perm)
33a610c3
KW
2126{
2127 BdrvChild *c;
2128 uint64_t cumulative_perms = 0;
2129 uint64_t cumulative_shared_perms = BLK_PERM_ALL;
2130
2131 QLIST_FOREACH(c, &bs->parents, next_parent) {
2132 cumulative_perms |= c->perm;
2133 cumulative_shared_perms &= c->shared_perm;
2134 }
2135
2136 *perm = cumulative_perms;
2137 *shared_perm = cumulative_shared_perms;
2138}
2139
d083319f
KW
2140static char *bdrv_child_user_desc(BdrvChild *c)
2141{
2142 if (c->role->get_parent_desc) {
2143 return c->role->get_parent_desc(c);
2144 }
2145
2146 return g_strdup("another user");
2147}
2148
5176196c 2149char *bdrv_perm_names(uint64_t perm)
d083319f
KW
2150{
2151 struct perm_name {
2152 uint64_t perm;
2153 const char *name;
2154 } permissions[] = {
2155 { BLK_PERM_CONSISTENT_READ, "consistent read" },
2156 { BLK_PERM_WRITE, "write" },
2157 { BLK_PERM_WRITE_UNCHANGED, "write unchanged" },
2158 { BLK_PERM_RESIZE, "resize" },
2159 { BLK_PERM_GRAPH_MOD, "change children" },
2160 { 0, NULL }
2161 };
2162
e2a7423a 2163 GString *result = g_string_sized_new(30);
d083319f
KW
2164 struct perm_name *p;
2165
2166 for (p = permissions; p->name; p++) {
2167 if (perm & p->perm) {
e2a7423a
AG
2168 if (result->len > 0) {
2169 g_string_append(result, ", ");
2170 }
2171 g_string_append(result, p->name);
d083319f
KW
2172 }
2173 }
2174
e2a7423a 2175 return g_string_free(result, FALSE);
d083319f
KW
2176}
2177
33a610c3
KW
2178/*
2179 * Checks whether a new reference to @bs can be added if the new user requires
46181129
KW
2180 * @new_used_perm/@new_shared_perm as its permissions. If @ignore_children is
2181 * set, the BdrvChild objects in this list are ignored in the calculations;
2182 * this allows checking permission updates for an existing reference.
33a610c3 2183 *
9eab1544
HR
2184 * See bdrv_check_perm() for the semantics of @tighten_restrictions.
2185 *
33a610c3
KW
2186 * Needs to be followed by a call to either bdrv_set_perm() or
2187 * bdrv_abort_perm_update(). */
3121fb45
KW
2188static int bdrv_check_update_perm(BlockDriverState *bs, BlockReopenQueue *q,
2189 uint64_t new_used_perm,
d5e6f437 2190 uint64_t new_shared_perm,
9eab1544
HR
2191 GSList *ignore_children,
2192 bool *tighten_restrictions,
2193 Error **errp)
d5e6f437
KW
2194{
2195 BdrvChild *c;
33a610c3
KW
2196 uint64_t cumulative_perms = new_used_perm;
2197 uint64_t cumulative_shared_perms = new_shared_perm;
d5e6f437 2198
9eab1544
HR
2199 assert(!q || !tighten_restrictions);
2200
d5e6f437
KW
2201 /* There is no reason why anyone couldn't tolerate write_unchanged */
2202 assert(new_shared_perm & BLK_PERM_WRITE_UNCHANGED);
2203
2204 QLIST_FOREACH(c, &bs->parents, next_parent) {
46181129 2205 if (g_slist_find(ignore_children, c)) {
d5e6f437
KW
2206 continue;
2207 }
2208
d083319f
KW
2209 if ((new_used_perm & c->shared_perm) != new_used_perm) {
2210 char *user = bdrv_child_user_desc(c);
2211 char *perm_names = bdrv_perm_names(new_used_perm & ~c->shared_perm);
9eab1544
HR
2212
2213 if (tighten_restrictions) {
2214 *tighten_restrictions = true;
2215 }
2216
d083319f
KW
2217 error_setg(errp, "Conflicts with use by %s as '%s', which does not "
2218 "allow '%s' on %s",
2219 user, c->name, perm_names, bdrv_get_node_name(c->bs));
2220 g_free(user);
2221 g_free(perm_names);
2222 return -EPERM;
2223 }
2224
2225 if ((c->perm & new_shared_perm) != c->perm) {
2226 char *user = bdrv_child_user_desc(c);
2227 char *perm_names = bdrv_perm_names(c->perm & ~new_shared_perm);
9eab1544
HR
2228
2229 if (tighten_restrictions) {
2230 *tighten_restrictions = true;
2231 }
2232
d083319f
KW
2233 error_setg(errp, "Conflicts with use by %s as '%s', which uses "
2234 "'%s' on %s",
2235 user, c->name, perm_names, bdrv_get_node_name(c->bs));
2236 g_free(user);
2237 g_free(perm_names);
d5e6f437
KW
2238 return -EPERM;
2239 }
33a610c3
KW
2240
2241 cumulative_perms |= c->perm;
2242 cumulative_shared_perms &= c->shared_perm;
d5e6f437
KW
2243 }
2244
3121fb45 2245 return bdrv_check_perm(bs, q, cumulative_perms, cumulative_shared_perms,
9eab1544 2246 ignore_children, tighten_restrictions, errp);
33a610c3
KW
2247}
2248
2249/* Needs to be followed by a call to either bdrv_child_set_perm() or
2250 * bdrv_child_abort_perm_update(). */
3121fb45
KW
2251static int bdrv_child_check_perm(BdrvChild *c, BlockReopenQueue *q,
2252 uint64_t perm, uint64_t shared,
9eab1544
HR
2253 GSList *ignore_children,
2254 bool *tighten_restrictions, Error **errp)
33a610c3 2255{
46181129
KW
2256 int ret;
2257
2258 ignore_children = g_slist_prepend(g_slist_copy(ignore_children), c);
9eab1544
HR
2259 ret = bdrv_check_update_perm(c->bs, q, perm, shared, ignore_children,
2260 tighten_restrictions, errp);
46181129
KW
2261 g_slist_free(ignore_children);
2262
f962e961
VSO
2263 if (ret < 0) {
2264 return ret;
2265 }
2266
2267 if (!c->has_backup_perm) {
2268 c->has_backup_perm = true;
2269 c->backup_perm = c->perm;
2270 c->backup_shared_perm = c->shared_perm;
2271 }
2272 /*
2273 * Note: it's OK if c->has_backup_perm was already set, as we can find the
2274 * same child twice during check_perm procedure
2275 */
2276
2277 c->perm = perm;
2278 c->shared_perm = shared;
2279
2280 return 0;
33a610c3
KW
2281}
2282
c1cef672 2283static void bdrv_child_set_perm(BdrvChild *c, uint64_t perm, uint64_t shared)
33a610c3
KW
2284{
2285 uint64_t cumulative_perms, cumulative_shared_perms;
2286
f962e961
VSO
2287 c->has_backup_perm = false;
2288
33a610c3
KW
2289 c->perm = perm;
2290 c->shared_perm = shared;
2291
2292 bdrv_get_cumulative_perm(c->bs, &cumulative_perms,
2293 &cumulative_shared_perms);
2294 bdrv_set_perm(c->bs, cumulative_perms, cumulative_shared_perms);
2295}
2296
c1cef672 2297static void bdrv_child_abort_perm_update(BdrvChild *c)
33a610c3 2298{
f962e961
VSO
2299 if (c->has_backup_perm) {
2300 c->perm = c->backup_perm;
2301 c->shared_perm = c->backup_shared_perm;
2302 c->has_backup_perm = false;
2303 }
2304
33a610c3
KW
2305 bdrv_abort_perm_update(c->bs);
2306}
2307
2308int bdrv_child_try_set_perm(BdrvChild *c, uint64_t perm, uint64_t shared,
2309 Error **errp)
2310{
1046779e 2311 Error *local_err = NULL;
33a610c3 2312 int ret;
1046779e 2313 bool tighten_restrictions;
33a610c3 2314
1046779e
HR
2315 ret = bdrv_child_check_perm(c, NULL, perm, shared, NULL,
2316 &tighten_restrictions, &local_err);
33a610c3
KW
2317 if (ret < 0) {
2318 bdrv_child_abort_perm_update(c);
1046779e
HR
2319 if (tighten_restrictions) {
2320 error_propagate(errp, local_err);
2321 } else {
2322 /*
2323 * Our caller may intend to only loosen restrictions and
2324 * does not expect this function to fail. Errors are not
2325 * fatal in such a case, so we can just hide them from our
2326 * caller.
2327 */
2328 error_free(local_err);
2329 ret = 0;
2330 }
33a610c3
KW
2331 return ret;
2332 }
2333
2334 bdrv_child_set_perm(c, perm, shared);
2335
d5e6f437
KW
2336 return 0;
2337}
2338
c1087f12
HR
2339int bdrv_child_refresh_perms(BlockDriverState *bs, BdrvChild *c, Error **errp)
2340{
2341 uint64_t parent_perms, parent_shared;
2342 uint64_t perms, shared;
2343
2344 bdrv_get_cumulative_perm(bs, &parent_perms, &parent_shared);
2345 bdrv_child_perm(bs, c->bs, c, c->role, NULL, parent_perms, parent_shared,
2346 &perms, &shared);
2347
2348 return bdrv_child_try_set_perm(c, perms, shared, errp);
2349}
2350
6a1b9ee1
KW
2351void bdrv_filter_default_perms(BlockDriverState *bs, BdrvChild *c,
2352 const BdrvChildRole *role,
e0995dc3 2353 BlockReopenQueue *reopen_queue,
6a1b9ee1
KW
2354 uint64_t perm, uint64_t shared,
2355 uint64_t *nperm, uint64_t *nshared)
2356{
e444fa83
KW
2357 *nperm = perm & DEFAULT_PERM_PASSTHROUGH;
2358 *nshared = (shared & DEFAULT_PERM_PASSTHROUGH) | DEFAULT_PERM_UNCHANGED;
6a1b9ee1
KW
2359}
2360
6b1a044a
KW
2361void bdrv_format_default_perms(BlockDriverState *bs, BdrvChild *c,
2362 const BdrvChildRole *role,
e0995dc3 2363 BlockReopenQueue *reopen_queue,
6b1a044a
KW
2364 uint64_t perm, uint64_t shared,
2365 uint64_t *nperm, uint64_t *nshared)
2366{
2367 bool backing = (role == &child_backing);
2368 assert(role == &child_backing || role == &child_file);
2369
2370 if (!backing) {
5fbfabd3
KW
2371 int flags = bdrv_reopen_get_flags(reopen_queue, bs);
2372
6b1a044a
KW
2373 /* Apart from the modifications below, the same permissions are
2374 * forwarded and left alone as for filters */
e0995dc3
KW
2375 bdrv_filter_default_perms(bs, c, role, reopen_queue, perm, shared,
2376 &perm, &shared);
6b1a044a
KW
2377
2378 /* Format drivers may touch metadata even if the guest doesn't write */
cc022140 2379 if (bdrv_is_writable_after_reopen(bs, reopen_queue)) {
6b1a044a
KW
2380 perm |= BLK_PERM_WRITE | BLK_PERM_RESIZE;
2381 }
2382
2383 /* bs->file always needs to be consistent because of the metadata. We
2384 * can never allow other users to resize or write to it. */
5fbfabd3
KW
2385 if (!(flags & BDRV_O_NO_IO)) {
2386 perm |= BLK_PERM_CONSISTENT_READ;
2387 }
6b1a044a
KW
2388 shared &= ~(BLK_PERM_WRITE | BLK_PERM_RESIZE);
2389 } else {
2390 /* We want consistent read from backing files if the parent needs it.
2391 * No other operations are performed on backing files. */
2392 perm &= BLK_PERM_CONSISTENT_READ;
2393
2394 /* If the parent can deal with changing data, we're okay with a
2395 * writable and resizable backing file. */
2396 /* TODO Require !(perm & BLK_PERM_CONSISTENT_READ), too? */
2397 if (shared & BLK_PERM_WRITE) {
2398 shared = BLK_PERM_WRITE | BLK_PERM_RESIZE;
2399 } else {
2400 shared = 0;
2401 }
2402
2403 shared |= BLK_PERM_CONSISTENT_READ | BLK_PERM_GRAPH_MOD |
2404 BLK_PERM_WRITE_UNCHANGED;
2405 }
2406
9c5e6594
KW
2407 if (bs->open_flags & BDRV_O_INACTIVE) {
2408 shared |= BLK_PERM_WRITE | BLK_PERM_RESIZE;
2409 }
2410
6b1a044a
KW
2411 *nperm = perm;
2412 *nshared = shared;
2413}
2414
7b1d9c4d
HR
2415uint64_t bdrv_qapi_perm_to_blk_perm(BlockPermission qapi_perm)
2416{
2417 static const uint64_t permissions[] = {
2418 [BLOCK_PERMISSION_CONSISTENT_READ] = BLK_PERM_CONSISTENT_READ,
2419 [BLOCK_PERMISSION_WRITE] = BLK_PERM_WRITE,
2420 [BLOCK_PERMISSION_WRITE_UNCHANGED] = BLK_PERM_WRITE_UNCHANGED,
2421 [BLOCK_PERMISSION_RESIZE] = BLK_PERM_RESIZE,
2422 [BLOCK_PERMISSION_GRAPH_MOD] = BLK_PERM_GRAPH_MOD,
2423 };
2424
2425 QEMU_BUILD_BUG_ON(ARRAY_SIZE(permissions) != BLOCK_PERMISSION__MAX);
2426 QEMU_BUILD_BUG_ON(1UL << ARRAY_SIZE(permissions) != BLK_PERM_ALL + 1);
2427
2428 assert(qapi_perm < BLOCK_PERMISSION__MAX);
2429
2430 return permissions[qapi_perm];
2431}
2432
8ee03995
KW
2433static void bdrv_replace_child_noperm(BdrvChild *child,
2434 BlockDriverState *new_bs)
e9740bc6
KW
2435{
2436 BlockDriverState *old_bs = child->bs;
debc2927
HR
2437 int new_bs_quiesce_counter;
2438 int drain_saldo;
e9740bc6 2439
2cad1ebe
AG
2440 assert(!child->frozen);
2441
bb2614e9
FZ
2442 if (old_bs && new_bs) {
2443 assert(bdrv_get_aio_context(old_bs) == bdrv_get_aio_context(new_bs));
2444 }
debc2927
HR
2445
2446 new_bs_quiesce_counter = (new_bs ? new_bs->quiesce_counter : 0);
2447 drain_saldo = new_bs_quiesce_counter - child->parent_quiesce_counter;
2448
2449 /*
2450 * If the new child node is drained but the old one was not, flush
2451 * all outstanding requests to the old child node.
2452 */
2453 while (drain_saldo > 0 && child->role->drained_begin) {
2454 bdrv_parent_drained_begin_single(child, true);
2455 drain_saldo--;
2456 }
2457
e9740bc6 2458 if (old_bs) {
d736f119
KW
2459 /* Detach first so that the recursive drain sections coming from @child
2460 * are already gone and we only end the drain sections that came from
2461 * elsewhere. */
2462 if (child->role->detach) {
2463 child->role->detach(child);
2464 }
e9740bc6
KW
2465 QLIST_REMOVE(child, next_parent);
2466 }
36fe1331
KW
2467
2468 child->bs = new_bs;
2469
e9740bc6
KW
2470 if (new_bs) {
2471 QLIST_INSERT_HEAD(&new_bs->parents, child, next_parent);
debc2927
HR
2472
2473 /*
2474 * Detaching the old node may have led to the new node's
2475 * quiesce_counter having been decreased. Not a problem, we
2476 * just need to recognize this here and then invoke
2477 * drained_end appropriately more often.
2478 */
2479 assert(new_bs->quiesce_counter <= new_bs_quiesce_counter);
2480 drain_saldo += new_bs->quiesce_counter - new_bs_quiesce_counter;
33a610c3 2481
d736f119
KW
2482 /* Attach only after starting new drained sections, so that recursive
2483 * drain sections coming from @child don't get an extra .drained_begin
2484 * callback. */
8ee03995
KW
2485 if (child->role->attach) {
2486 child->role->attach(child);
2487 }
2488 }
debc2927
HR
2489
2490 /*
2491 * If the old child node was drained but the new one is not, allow
2492 * requests to come in only after the new node has been attached.
2493 */
2494 while (drain_saldo < 0 && child->role->drained_end) {
2495 bdrv_parent_drained_end_single(child);
2496 drain_saldo++;
2497 }
8ee03995 2498}
33a610c3 2499
466787fb
KW
2500/*
2501 * Updates @child to change its reference to point to @new_bs, including
2502 * checking and applying the necessary permisson updates both to the old node
2503 * and to @new_bs.
2504 *
2505 * NULL is passed as @new_bs for removing the reference before freeing @child.
2506 *
2507 * If @new_bs is not NULL, bdrv_check_perm() must be called beforehand, as this
2508 * function uses bdrv_set_perm() to update the permissions according to the new
2509 * reference that @new_bs gets.
2510 */
2511static void bdrv_replace_child(BdrvChild *child, BlockDriverState *new_bs)
8ee03995
KW
2512{
2513 BlockDriverState *old_bs = child->bs;
2514 uint64_t perm, shared_perm;
2515
8aecf1d1
KW
2516 bdrv_replace_child_noperm(child, new_bs);
2517
87ace5f8
HR
2518 /*
2519 * Start with the new node's permissions. If @new_bs is a (direct
2520 * or indirect) child of @old_bs, we must complete the permission
2521 * update on @new_bs before we loosen the restrictions on @old_bs.
2522 * Otherwise, bdrv_check_perm() on @old_bs would re-initiate
2523 * updating the permissions of @new_bs, and thus not purely loosen
2524 * restrictions.
2525 */
2526 if (new_bs) {
2527 bdrv_get_cumulative_perm(new_bs, &perm, &shared_perm);
2528 bdrv_set_perm(new_bs, perm, shared_perm);
2529 }
2530
8ee03995 2531 if (old_bs) {
33a610c3
KW
2532 /* Update permissions for old node. This is guaranteed to succeed
2533 * because we're just taking a parent away, so we're loosening
2534 * restrictions. */
1046779e
HR
2535 bool tighten_restrictions;
2536 int ret;
2537
33a610c3 2538 bdrv_get_cumulative_perm(old_bs, &perm, &shared_perm);
1046779e
HR
2539 ret = bdrv_check_perm(old_bs, NULL, perm, shared_perm, NULL,
2540 &tighten_restrictions, NULL);
2541 assert(tighten_restrictions == false);
2542 if (ret < 0) {
2543 /* We only tried to loosen restrictions, so errors are not fatal */
2544 bdrv_abort_perm_update(old_bs);
2545 } else {
2546 bdrv_set_perm(old_bs, perm, shared_perm);
2547 }
ad943dcb
KW
2548
2549 /* When the parent requiring a non-default AioContext is removed, the
2550 * node moves back to the main AioContext */
2551 bdrv_try_set_aio_context(old_bs, qemu_get_aio_context(), NULL);
e9740bc6 2552 }
e9740bc6
KW
2553}
2554
b441dc71
AG
2555/*
2556 * This function steals the reference to child_bs from the caller.
2557 * That reference is later dropped by bdrv_root_unref_child().
2558 *
2559 * On failure NULL is returned, errp is set and the reference to
2560 * child_bs is also dropped.
132ada80
KW
2561 *
2562 * The caller must hold the AioContext lock @child_bs, but not that of @ctx
2563 * (unless @child_bs is already in @ctx).
b441dc71 2564 */
f21d96d0
KW
2565BdrvChild *bdrv_root_attach_child(BlockDriverState *child_bs,
2566 const char *child_name,
36fe1331 2567 const BdrvChildRole *child_role,
132ada80 2568 AioContext *ctx,
d5e6f437
KW
2569 uint64_t perm, uint64_t shared_perm,
2570 void *opaque, Error **errp)
df581792 2571{
d5e6f437 2572 BdrvChild *child;
132ada80 2573 Error *local_err = NULL;
d5e6f437
KW
2574 int ret;
2575
9eab1544
HR
2576 ret = bdrv_check_update_perm(child_bs, NULL, perm, shared_perm, NULL, NULL,
2577 errp);
d5e6f437 2578 if (ret < 0) {
33a610c3 2579 bdrv_abort_perm_update(child_bs);
b441dc71 2580 bdrv_unref(child_bs);
d5e6f437
KW
2581 return NULL;
2582 }
2583
2584 child = g_new(BdrvChild, 1);
df581792 2585 *child = (BdrvChild) {
d5e6f437
KW
2586 .bs = NULL,
2587 .name = g_strdup(child_name),
2588 .role = child_role,
2589 .perm = perm,
2590 .shared_perm = shared_perm,
2591 .opaque = opaque,
df581792
KW
2592 };
2593
132ada80
KW
2594 /* If the AioContexts don't match, first try to move the subtree of
2595 * child_bs into the AioContext of the new parent. If this doesn't work,
2596 * try moving the parent into the AioContext of child_bs instead. */
2597 if (bdrv_get_aio_context(child_bs) != ctx) {
2598 ret = bdrv_try_set_aio_context(child_bs, ctx, &local_err);
2599 if (ret < 0 && child_role->can_set_aio_ctx) {
0beab811 2600 GSList *ignore = g_slist_prepend(NULL, child);
132ada80
KW
2601 ctx = bdrv_get_aio_context(child_bs);
2602 if (child_role->can_set_aio_ctx(child, ctx, &ignore, NULL)) {
2603 error_free(local_err);
2604 ret = 0;
2605 g_slist_free(ignore);
0beab811 2606 ignore = g_slist_prepend(NULL, child);
132ada80
KW
2607 child_role->set_aio_ctx(child, ctx, &ignore);
2608 }
2609 g_slist_free(ignore);
2610 }
2611 if (ret < 0) {
2612 error_propagate(errp, local_err);
2613 g_free(child);
2614 bdrv_abort_perm_update(child_bs);
2615 return NULL;
2616 }
2617 }
2618
33a610c3 2619 /* This performs the matching bdrv_set_perm() for the above check. */
466787fb 2620 bdrv_replace_child(child, child_bs);
b4b059f6
KW
2621
2622 return child;
df581792
KW
2623}
2624
b441dc71
AG
2625/*
2626 * This function transfers the reference to child_bs from the caller
2627 * to parent_bs. That reference is later dropped by parent_bs on
2628 * bdrv_close() or if someone calls bdrv_unref_child().
2629 *
2630 * On failure NULL is returned, errp is set and the reference to
2631 * child_bs is also dropped.
132ada80
KW
2632 *
2633 * If @parent_bs and @child_bs are in different AioContexts, the caller must
2634 * hold the AioContext lock for @child_bs, but not for @parent_bs.
b441dc71 2635 */
98292c61
WC
2636BdrvChild *bdrv_attach_child(BlockDriverState *parent_bs,
2637 BlockDriverState *child_bs,
2638 const char *child_name,
8b2ff529
KW
2639 const BdrvChildRole *child_role,
2640 Error **errp)
f21d96d0 2641{
d5e6f437 2642 BdrvChild *child;
f68c598b
KW
2643 uint64_t perm, shared_perm;
2644
2645 bdrv_get_cumulative_perm(parent_bs, &perm, &shared_perm);
2646
2647 assert(parent_bs->drv);
e0995dc3 2648 bdrv_child_perm(parent_bs, child_bs, NULL, child_role, NULL,
ffd1a5a2 2649 perm, shared_perm, &perm, &shared_perm);
d5e6f437 2650
d5e6f437 2651 child = bdrv_root_attach_child(child_bs, child_name, child_role,
132ada80 2652 bdrv_get_aio_context(parent_bs),
f68c598b 2653 perm, shared_perm, parent_bs, errp);
d5e6f437
KW
2654 if (child == NULL) {
2655 return NULL;
2656 }
2657
f21d96d0
KW
2658 QLIST_INSERT_HEAD(&parent_bs->children, child, next);
2659 return child;
2660}
2661
3f09bfbc 2662static void bdrv_detach_child(BdrvChild *child)
33a60407 2663{
195ed8cb 2664 QLIST_SAFE_REMOVE(child, next);
e9740bc6 2665
466787fb 2666 bdrv_replace_child(child, NULL);
e9740bc6 2667
260fecf1 2668 g_free(child->name);
33a60407
KW
2669 g_free(child);
2670}
2671
f21d96d0 2672void bdrv_root_unref_child(BdrvChild *child)
33a60407 2673{
779020cb
KW
2674 BlockDriverState *child_bs;
2675
f21d96d0
KW
2676 child_bs = child->bs;
2677 bdrv_detach_child(child);
2678 bdrv_unref(child_bs);
2679}
2680
3cf746b3
HR
2681/**
2682 * Clear all inherits_from pointers from children and grandchildren of
2683 * @root that point to @root, where necessary.
2684 */
2685static void bdrv_unset_inherits_from(BlockDriverState *root, BdrvChild *child)
f21d96d0 2686{
3cf746b3 2687 BdrvChild *c;
4e4bf5c4 2688
3cf746b3
HR
2689 if (child->bs->inherits_from == root) {
2690 /*
2691 * Remove inherits_from only when the last reference between root and
2692 * child->bs goes away.
2693 */
2694 QLIST_FOREACH(c, &root->children, next) {
4e4bf5c4
KW
2695 if (c != child && c->bs == child->bs) {
2696 break;
2697 }
2698 }
2699 if (c == NULL) {
2700 child->bs->inherits_from = NULL;
2701 }
33a60407
KW
2702 }
2703
3cf746b3
HR
2704 QLIST_FOREACH(c, &child->bs->children, next) {
2705 bdrv_unset_inherits_from(root, c);
2706 }
2707}
2708
2709void bdrv_unref_child(BlockDriverState *parent, BdrvChild *child)
2710{
2711 if (child == NULL) {
2712 return;
2713 }
2714
2715 bdrv_unset_inherits_from(parent, child);
f21d96d0 2716 bdrv_root_unref_child(child);
33a60407
KW
2717}
2718
5c8cab48
KW
2719
2720static void bdrv_parent_cb_change_media(BlockDriverState *bs, bool load)
2721{
2722 BdrvChild *c;
2723 QLIST_FOREACH(c, &bs->parents, next_parent) {
2724 if (c->role->change_media) {
2725 c->role->change_media(c, load);
2726 }
2727 }
2728}
2729
0065c455
AG
2730/* Return true if you can reach parent going through child->inherits_from
2731 * recursively. If parent or child are NULL, return false */
2732static bool bdrv_inherits_from_recursive(BlockDriverState *child,
2733 BlockDriverState *parent)
2734{
2735 while (child && child != parent) {
2736 child = child->inherits_from;
2737 }
2738
2739 return child != NULL;
2740}
2741
5db15a57
KW
2742/*
2743 * Sets the backing file link of a BDS. A new reference is created; callers
2744 * which don't need their own reference any more must call bdrv_unref().
2745 */
12fa4af6
KW
2746void bdrv_set_backing_hd(BlockDriverState *bs, BlockDriverState *backing_hd,
2747 Error **errp)
8d24cce1 2748{
0065c455
AG
2749 bool update_inherits_from = bdrv_chain_contains(bs, backing_hd) &&
2750 bdrv_inherits_from_recursive(backing_hd, bs);
2751
2cad1ebe
AG
2752 if (bdrv_is_backing_chain_frozen(bs, backing_bs(bs), errp)) {
2753 return;
2754 }
2755
5db15a57
KW
2756 if (backing_hd) {
2757 bdrv_ref(backing_hd);
2758 }
8d24cce1 2759
760e0063 2760 if (bs->backing) {
5db15a57 2761 bdrv_unref_child(bs, bs->backing);
826b6ca0
FZ
2762 }
2763
8d24cce1 2764 if (!backing_hd) {
760e0063 2765 bs->backing = NULL;
8d24cce1
FZ
2766 goto out;
2767 }
12fa4af6 2768
8b2ff529 2769 bs->backing = bdrv_attach_child(bs, backing_hd, "backing", &child_backing,
12fa4af6 2770 errp);
0065c455
AG
2771 /* If backing_hd was already part of bs's backing chain, and
2772 * inherits_from pointed recursively to bs then let's update it to
2773 * point directly to bs (else it will become NULL). */
b441dc71 2774 if (bs->backing && update_inherits_from) {
0065c455
AG
2775 backing_hd->inherits_from = bs;
2776 }
826b6ca0 2777
8d24cce1 2778out:
3baca891 2779 bdrv_refresh_limits(bs, NULL);
8d24cce1
FZ
2780}
2781
31ca6d07
KW
2782/*
2783 * Opens the backing file for a BlockDriverState if not yet open
2784 *
d9b7b057
KW
2785 * bdref_key specifies the key for the image's BlockdevRef in the options QDict.
2786 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
2787 * itself, all options starting with "${bdref_key}." are considered part of the
2788 * BlockdevRef.
2789 *
2790 * TODO Can this be unified with bdrv_open_image()?
31ca6d07 2791 */
d9b7b057
KW
2792int bdrv_open_backing_file(BlockDriverState *bs, QDict *parent_options,
2793 const char *bdref_key, Error **errp)
9156df12 2794{
6b6833c1 2795 char *backing_filename = NULL;
d9b7b057
KW
2796 char *bdref_key_dot;
2797 const char *reference = NULL;
317fc44e 2798 int ret = 0;
998c2019 2799 bool implicit_backing = false;
8d24cce1 2800 BlockDriverState *backing_hd;
d9b7b057
KW
2801 QDict *options;
2802 QDict *tmp_parent_options = NULL;
34b5d2c6 2803 Error *local_err = NULL;
9156df12 2804
760e0063 2805 if (bs->backing != NULL) {
1ba4b6a5 2806 goto free_exit;
9156df12
PB
2807 }
2808
31ca6d07 2809 /* NULL means an empty set of options */
d9b7b057
KW
2810 if (parent_options == NULL) {
2811 tmp_parent_options = qdict_new();
2812 parent_options = tmp_parent_options;
31ca6d07
KW
2813 }
2814
9156df12 2815 bs->open_flags &= ~BDRV_O_NO_BACKING;
d9b7b057
KW
2816
2817 bdref_key_dot = g_strdup_printf("%s.", bdref_key);
2818 qdict_extract_subqdict(parent_options, &options, bdref_key_dot);
2819 g_free(bdref_key_dot);
2820
129c7d1c
MA
2821 /*
2822 * Caution: while qdict_get_try_str() is fine, getting non-string
2823 * types would require more care. When @parent_options come from
2824 * -blockdev or blockdev_add, its members are typed according to
2825 * the QAPI schema, but when they come from -drive, they're all
2826 * QString.
2827 */
d9b7b057
KW
2828 reference = qdict_get_try_str(parent_options, bdref_key);
2829 if (reference || qdict_haskey(options, "file.filename")) {
6b6833c1 2830 /* keep backing_filename NULL */
1cb6f506 2831 } else if (bs->backing_file[0] == '\0' && qdict_size(options) == 0) {
cb3e7f08 2832 qobject_unref(options);
1ba4b6a5 2833 goto free_exit;
dbecebdd 2834 } else {
998c2019
HR
2835 if (qdict_size(options) == 0) {
2836 /* If the user specifies options that do not modify the
2837 * backing file's behavior, we might still consider it the
2838 * implicit backing file. But it's easier this way, and
2839 * just specifying some of the backing BDS's options is
2840 * only possible with -drive anyway (otherwise the QAPI
2841 * schema forces the user to specify everything). */
2842 implicit_backing = !strcmp(bs->auto_backing_file, bs->backing_file);
2843 }
2844
6b6833c1 2845 backing_filename = bdrv_get_full_backing_filename(bs, &local_err);
9f07429e
HR
2846 if (local_err) {
2847 ret = -EINVAL;
2848 error_propagate(errp, local_err);
cb3e7f08 2849 qobject_unref(options);
9f07429e
HR
2850 goto free_exit;
2851 }
9156df12
PB
2852 }
2853
8ee79e70
KW
2854 if (!bs->drv || !bs->drv->supports_backing) {
2855 ret = -EINVAL;
2856 error_setg(errp, "Driver doesn't support backing files");
cb3e7f08 2857 qobject_unref(options);
8ee79e70
KW
2858 goto free_exit;
2859 }
2860
6bff597b
PK
2861 if (!reference &&
2862 bs->backing_format[0] != '\0' && !qdict_haskey(options, "driver")) {
46f5ac20 2863 qdict_put_str(options, "driver", bs->backing_format);
9156df12
PB
2864 }
2865
6b6833c1
HR
2866 backing_hd = bdrv_open_inherit(backing_filename, reference, options, 0, bs,
2867 &child_backing, errp);
5b363937 2868 if (!backing_hd) {
9156df12 2869 bs->open_flags |= BDRV_O_NO_BACKING;
e43bfd9c 2870 error_prepend(errp, "Could not open backing file: ");
5b363937 2871 ret = -EINVAL;
1ba4b6a5 2872 goto free_exit;
9156df12 2873 }
df581792 2874
998c2019
HR
2875 if (implicit_backing) {
2876 bdrv_refresh_filename(backing_hd);
2877 pstrcpy(bs->auto_backing_file, sizeof(bs->auto_backing_file),
2878 backing_hd->filename);
2879 }
2880
5db15a57
KW
2881 /* Hook up the backing file link; drop our reference, bs owns the
2882 * backing_hd reference now */
12fa4af6 2883 bdrv_set_backing_hd(bs, backing_hd, &local_err);
5db15a57 2884 bdrv_unref(backing_hd);
12fa4af6 2885 if (local_err) {
8cd1a3e4 2886 error_propagate(errp, local_err);
12fa4af6
KW
2887 ret = -EINVAL;
2888 goto free_exit;
2889 }
d80ac658 2890
d9b7b057
KW
2891 qdict_del(parent_options, bdref_key);
2892
1ba4b6a5
BC
2893free_exit:
2894 g_free(backing_filename);
cb3e7f08 2895 qobject_unref(tmp_parent_options);
1ba4b6a5 2896 return ret;
9156df12
PB
2897}
2898
2d6b86af
KW
2899static BlockDriverState *
2900bdrv_open_child_bs(const char *filename, QDict *options, const char *bdref_key,
2901 BlockDriverState *parent, const BdrvChildRole *child_role,
2902 bool allow_none, Error **errp)
da557aac 2903{
2d6b86af 2904 BlockDriverState *bs = NULL;
da557aac 2905 QDict *image_options;
da557aac
HR
2906 char *bdref_key_dot;
2907 const char *reference;
2908
df581792 2909 assert(child_role != NULL);
f67503e5 2910
da557aac
HR
2911 bdref_key_dot = g_strdup_printf("%s.", bdref_key);
2912 qdict_extract_subqdict(options, &image_options, bdref_key_dot);
2913 g_free(bdref_key_dot);
2914
129c7d1c
MA
2915 /*
2916 * Caution: while qdict_get_try_str() is fine, getting non-string
2917 * types would require more care. When @options come from
2918 * -blockdev or blockdev_add, its members are typed according to
2919 * the QAPI schema, but when they come from -drive, they're all
2920 * QString.
2921 */
da557aac
HR
2922 reference = qdict_get_try_str(options, bdref_key);
2923 if (!filename && !reference && !qdict_size(image_options)) {
b4b059f6 2924 if (!allow_none) {
da557aac
HR
2925 error_setg(errp, "A block device must be specified for \"%s\"",
2926 bdref_key);
da557aac 2927 }
cb3e7f08 2928 qobject_unref(image_options);
da557aac
HR
2929 goto done;
2930 }
2931
5b363937
HR
2932 bs = bdrv_open_inherit(filename, reference, image_options, 0,
2933 parent, child_role, errp);
2934 if (!bs) {
df581792
KW
2935 goto done;
2936 }
2937
da557aac
HR
2938done:
2939 qdict_del(options, bdref_key);
2d6b86af
KW
2940 return bs;
2941}
2942
2943/*
2944 * Opens a disk image whose options are given as BlockdevRef in another block
2945 * device's options.
2946 *
2947 * If allow_none is true, no image will be opened if filename is false and no
2948 * BlockdevRef is given. NULL will be returned, but errp remains unset.
2949 *
2950 * bdrev_key specifies the key for the image's BlockdevRef in the options QDict.
2951 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
2952 * itself, all options starting with "${bdref_key}." are considered part of the
2953 * BlockdevRef.
2954 *
2955 * The BlockdevRef will be removed from the options QDict.
2956 */
2957BdrvChild *bdrv_open_child(const char *filename,
2958 QDict *options, const char *bdref_key,
2959 BlockDriverState *parent,
2960 const BdrvChildRole *child_role,
2961 bool allow_none, Error **errp)
2962{
2963 BlockDriverState *bs;
2964
2965 bs = bdrv_open_child_bs(filename, options, bdref_key, parent, child_role,
2966 allow_none, errp);
2967 if (bs == NULL) {
2968 return NULL;
2969 }
2970
b441dc71 2971 return bdrv_attach_child(parent, bs, bdref_key, child_role, errp);
b4b059f6
KW
2972}
2973
e1d74bc6
KW
2974/* TODO Future callers may need to specify parent/child_role in order for
2975 * option inheritance to work. Existing callers use it for the root node. */
2976BlockDriverState *bdrv_open_blockdev_ref(BlockdevRef *ref, Error **errp)
2977{
2978 BlockDriverState *bs = NULL;
2979 Error *local_err = NULL;
2980 QObject *obj = NULL;
2981 QDict *qdict = NULL;
2982 const char *reference = NULL;
2983 Visitor *v = NULL;
2984
2985 if (ref->type == QTYPE_QSTRING) {
2986 reference = ref->u.reference;
2987 } else {
2988 BlockdevOptions *options = &ref->u.definition;
2989 assert(ref->type == QTYPE_QDICT);
2990
2991 v = qobject_output_visitor_new(&obj);
2992 visit_type_BlockdevOptions(v, NULL, &options, &local_err);
2993 if (local_err) {
2994 error_propagate(errp, local_err);
2995 goto fail;
2996 }
2997 visit_complete(v, &obj);
2998
7dc847eb 2999 qdict = qobject_to(QDict, obj);
e1d74bc6
KW
3000 qdict_flatten(qdict);
3001
3002 /* bdrv_open_inherit() defaults to the values in bdrv_flags (for
3003 * compatibility with other callers) rather than what we want as the
3004 * real defaults. Apply the defaults here instead. */
3005 qdict_set_default_str(qdict, BDRV_OPT_CACHE_DIRECT, "off");
3006 qdict_set_default_str(qdict, BDRV_OPT_CACHE_NO_FLUSH, "off");
3007 qdict_set_default_str(qdict, BDRV_OPT_READ_ONLY, "off");
e35bdc12
KW
3008 qdict_set_default_str(qdict, BDRV_OPT_AUTO_READ_ONLY, "off");
3009
e1d74bc6
KW
3010 }
3011
3012 bs = bdrv_open_inherit(NULL, reference, qdict, 0, NULL, NULL, errp);
3013 obj = NULL;
3014
3015fail:
cb3e7f08 3016 qobject_unref(obj);
e1d74bc6
KW
3017 visit_free(v);
3018 return bs;
3019}
3020
66836189
HR
3021static BlockDriverState *bdrv_append_temp_snapshot(BlockDriverState *bs,
3022 int flags,
3023 QDict *snapshot_options,
3024 Error **errp)
b998875d
KW
3025{
3026 /* TODO: extra byte is a hack to ensure MAX_PATH space on Windows. */
1ba4b6a5 3027 char *tmp_filename = g_malloc0(PATH_MAX + 1);
b998875d 3028 int64_t total_size;
83d0521a 3029 QemuOpts *opts = NULL;
ff6ed714 3030 BlockDriverState *bs_snapshot = NULL;
b2c2832c 3031 Error *local_err = NULL;
b998875d
KW
3032 int ret;
3033
3034 /* if snapshot, we create a temporary backing file and open it
3035 instead of opening 'filename' directly */
3036
3037 /* Get the required size from the image */
f187743a
KW
3038 total_size = bdrv_getlength(bs);
3039 if (total_size < 0) {
3040 error_setg_errno(errp, -total_size, "Could not get image size");
1ba4b6a5 3041 goto out;
f187743a 3042 }
b998875d
KW
3043
3044 /* Create the temporary image */
1ba4b6a5 3045 ret = get_tmp_filename(tmp_filename, PATH_MAX + 1);
b998875d
KW
3046 if (ret < 0) {
3047 error_setg_errno(errp, -ret, "Could not get temporary filename");
1ba4b6a5 3048 goto out;
b998875d
KW
3049 }
3050
ef810437 3051 opts = qemu_opts_create(bdrv_qcow2.create_opts, NULL, 0,
c282e1fd 3052 &error_abort);
39101f25 3053 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, total_size, &error_abort);
e43bfd9c 3054 ret = bdrv_create(&bdrv_qcow2, tmp_filename, opts, errp);
83d0521a 3055 qemu_opts_del(opts);
b998875d 3056 if (ret < 0) {
e43bfd9c
MA
3057 error_prepend(errp, "Could not create temporary overlay '%s': ",
3058 tmp_filename);
1ba4b6a5 3059 goto out;
b998875d
KW
3060 }
3061
73176bee 3062 /* Prepare options QDict for the temporary file */
46f5ac20
EB
3063 qdict_put_str(snapshot_options, "file.driver", "file");
3064 qdict_put_str(snapshot_options, "file.filename", tmp_filename);
3065 qdict_put_str(snapshot_options, "driver", "qcow2");
b998875d 3066
5b363937 3067 bs_snapshot = bdrv_open(NULL, NULL, snapshot_options, flags, errp);
73176bee 3068 snapshot_options = NULL;
5b363937 3069 if (!bs_snapshot) {
1ba4b6a5 3070 goto out;
b998875d
KW
3071 }
3072
ff6ed714
EB
3073 /* bdrv_append() consumes a strong reference to bs_snapshot
3074 * (i.e. it will call bdrv_unref() on it) even on error, so in
3075 * order to be able to return one, we have to increase
3076 * bs_snapshot's refcount here */
66836189 3077 bdrv_ref(bs_snapshot);
b2c2832c
KW
3078 bdrv_append(bs_snapshot, bs, &local_err);
3079 if (local_err) {
3080 error_propagate(errp, local_err);
ff6ed714 3081 bs_snapshot = NULL;
b2c2832c
KW
3082 goto out;
3083 }
1ba4b6a5
BC
3084
3085out:
cb3e7f08 3086 qobject_unref(snapshot_options);
1ba4b6a5 3087 g_free(tmp_filename);
ff6ed714 3088 return bs_snapshot;
b998875d
KW
3089}
3090
b6ce07aa
KW
3091/*
3092 * Opens a disk image (raw, qcow2, vmdk, ...)
de9c0cec
KW
3093 *
3094 * options is a QDict of options to pass to the block drivers, or NULL for an
3095 * empty set of options. The reference to the QDict belongs to the block layer
3096 * after the call (even on failure), so if the caller intends to reuse the
cb3e7f08 3097 * dictionary, it needs to use qobject_ref() before calling bdrv_open.
f67503e5
HR
3098 *
3099 * If *pbs is NULL, a new BDS will be created with a pointer to it stored there.
3100 * If it is not NULL, the referenced BDS will be reused.
ddf5636d
HR
3101 *
3102 * The reference parameter may be used to specify an existing block device which
3103 * should be opened. If specified, neither options nor a filename may be given,
3104 * nor can an existing BDS be reused (that is, *pbs has to be NULL).
b6ce07aa 3105 */
5b363937
HR
3106static BlockDriverState *bdrv_open_inherit(const char *filename,
3107 const char *reference,
3108 QDict *options, int flags,
3109 BlockDriverState *parent,
3110 const BdrvChildRole *child_role,
3111 Error **errp)
ea2384d3 3112{
b6ce07aa 3113 int ret;
5696c6e3 3114 BlockBackend *file = NULL;
9a4f4c31 3115 BlockDriverState *bs;
ce343771 3116 BlockDriver *drv = NULL;
2f624b80 3117 BdrvChild *child;
74fe54f2 3118 const char *drvname;
3e8c2e57 3119 const char *backing;
34b5d2c6 3120 Error *local_err = NULL;
73176bee 3121 QDict *snapshot_options = NULL;
b1e6fc08 3122 int snapshot_flags = 0;
712e7874 3123
f3930ed0
KW
3124 assert(!child_role || !flags);
3125 assert(!child_role == !parent);
f67503e5 3126
ddf5636d
HR
3127 if (reference) {
3128 bool options_non_empty = options ? qdict_size(options) : false;
cb3e7f08 3129 qobject_unref(options);
ddf5636d 3130
ddf5636d
HR
3131 if (filename || options_non_empty) {
3132 error_setg(errp, "Cannot reference an existing block device with "
3133 "additional options or a new filename");
5b363937 3134 return NULL;
ddf5636d
HR
3135 }
3136
3137 bs = bdrv_lookup_bs(reference, reference, errp);
3138 if (!bs) {
5b363937 3139 return NULL;
ddf5636d 3140 }
76b22320 3141
ddf5636d 3142 bdrv_ref(bs);
5b363937 3143 return bs;
ddf5636d
HR
3144 }
3145
5b363937 3146 bs = bdrv_new();
f67503e5 3147
de9c0cec
KW
3148 /* NULL means an empty set of options */
3149 if (options == NULL) {
3150 options = qdict_new();
3151 }
3152
145f598e 3153 /* json: syntax counts as explicit options, as if in the QDict */
de3b53f0
KW
3154 parse_json_protocol(options, &filename, &local_err);
3155 if (local_err) {
de3b53f0
KW
3156 goto fail;
3157 }
3158
145f598e
KW
3159 bs->explicit_options = qdict_clone_shallow(options);
3160
f3930ed0 3161 if (child_role) {
bddcec37 3162 bs->inherits_from = parent;
8e2160e2
KW
3163 child_role->inherit_options(&flags, options,
3164 parent->open_flags, parent->options);
f3930ed0
KW
3165 }
3166
de3b53f0 3167 ret = bdrv_fill_options(&options, filename, &flags, &local_err);
462f5bcf
KW
3168 if (local_err) {
3169 goto fail;
3170 }
3171
129c7d1c
MA
3172 /*
3173 * Set the BDRV_O_RDWR and BDRV_O_ALLOW_RDWR flags.
3174 * Caution: getting a boolean member of @options requires care.
3175 * When @options come from -blockdev or blockdev_add, members are
3176 * typed according to the QAPI schema, but when they come from
3177 * -drive, they're all QString.
3178 */
f87a0e29
AG
3179 if (g_strcmp0(qdict_get_try_str(options, BDRV_OPT_READ_ONLY), "on") &&
3180 !qdict_get_try_bool(options, BDRV_OPT_READ_ONLY, false)) {
3181 flags |= (BDRV_O_RDWR | BDRV_O_ALLOW_RDWR);
3182 } else {
3183 flags &= ~BDRV_O_RDWR;
14499ea5
AG
3184 }
3185
3186 if (flags & BDRV_O_SNAPSHOT) {
3187 snapshot_options = qdict_new();
3188 bdrv_temp_snapshot_options(&snapshot_flags, snapshot_options,
3189 flags, options);
f87a0e29
AG
3190 /* Let bdrv_backing_options() override "read-only" */
3191 qdict_del(options, BDRV_OPT_READ_ONLY);
14499ea5
AG
3192 bdrv_backing_options(&flags, options, flags, options);
3193 }
3194
62392ebb
KW
3195 bs->open_flags = flags;
3196 bs->options = options;
3197 options = qdict_clone_shallow(options);
3198
76c591b0 3199 /* Find the right image format driver */
129c7d1c 3200 /* See cautionary note on accessing @options above */
76c591b0
KW
3201 drvname = qdict_get_try_str(options, "driver");
3202 if (drvname) {
3203 drv = bdrv_find_format(drvname);
76c591b0
KW
3204 if (!drv) {
3205 error_setg(errp, "Unknown driver: '%s'", drvname);
76c591b0
KW
3206 goto fail;
3207 }
3208 }
3209
3210 assert(drvname || !(flags & BDRV_O_PROTOCOL));
76c591b0 3211
129c7d1c 3212 /* See cautionary note on accessing @options above */
3e8c2e57 3213 backing = qdict_get_try_str(options, "backing");
e59a0cf1
HR
3214 if (qobject_to(QNull, qdict_get(options, "backing")) != NULL ||
3215 (backing && *backing == '\0'))
3216 {
4f7be280
HR
3217 if (backing) {
3218 warn_report("Use of \"backing\": \"\" is deprecated; "
3219 "use \"backing\": null instead");
3220 }
3e8c2e57 3221 flags |= BDRV_O_NO_BACKING;
ae0f57f0
KW
3222 qdict_del(bs->explicit_options, "backing");
3223 qdict_del(bs->options, "backing");
3e8c2e57
AG
3224 qdict_del(options, "backing");
3225 }
3226
5696c6e3 3227 /* Open image file without format layer. This BlockBackend is only used for
4e4bf5c4
KW
3228 * probing, the block drivers will do their own bdrv_open_child() for the
3229 * same BDS, which is why we put the node name back into options. */
f4788adc 3230 if ((flags & BDRV_O_PROTOCOL) == 0) {
5696c6e3
KW
3231 BlockDriverState *file_bs;
3232
3233 file_bs = bdrv_open_child_bs(filename, options, "file", bs,
3234 &child_file, true, &local_err);
1fdd6933 3235 if (local_err) {
f4788adc
KW
3236 goto fail;
3237 }
5696c6e3 3238 if (file_bs != NULL) {
dacaa162
KW
3239 /* Not requesting BLK_PERM_CONSISTENT_READ because we're only
3240 * looking at the header to guess the image format. This works even
3241 * in cases where a guest would not see a consistent state. */
d861ab3a 3242 file = blk_new(bdrv_get_aio_context(file_bs), 0, BLK_PERM_ALL);
d7086422 3243 blk_insert_bs(file, file_bs, &local_err);
5696c6e3 3244 bdrv_unref(file_bs);
d7086422
KW
3245 if (local_err) {
3246 goto fail;
3247 }
5696c6e3 3248
46f5ac20 3249 qdict_put_str(options, "file", bdrv_get_node_name(file_bs));
4e4bf5c4 3250 }
f500a6d3
KW
3251 }
3252
76c591b0 3253 /* Image format probing */
38f3ef57 3254 bs->probed = !drv;
76c591b0 3255 if (!drv && file) {
cf2ab8fc 3256 ret = find_image_format(file, filename, &drv, &local_err);
17b005f1 3257 if (ret < 0) {
8bfea15d 3258 goto fail;
2a05cbe4 3259 }
62392ebb
KW
3260 /*
3261 * This option update would logically belong in bdrv_fill_options(),
3262 * but we first need to open bs->file for the probing to work, while
3263 * opening bs->file already requires the (mostly) final set of options
3264 * so that cache mode etc. can be inherited.
3265 *
3266 * Adding the driver later is somewhat ugly, but it's not an option
3267 * that would ever be inherited, so it's correct. We just need to make
3268 * sure to update both bs->options (which has the full effective
3269 * options for bs) and options (which has file.* already removed).
3270 */
46f5ac20
EB
3271 qdict_put_str(bs->options, "driver", drv->format_name);
3272 qdict_put_str(options, "driver", drv->format_name);
76c591b0 3273 } else if (!drv) {
17b005f1 3274 error_setg(errp, "Must specify either driver or file");
8bfea15d 3275 goto fail;
ea2384d3 3276 }
b6ce07aa 3277
53a29513
HR
3278 /* BDRV_O_PROTOCOL must be set iff a protocol BDS is about to be created */
3279 assert(!!(flags & BDRV_O_PROTOCOL) == !!drv->bdrv_file_open);
3280 /* file must be NULL if a protocol BDS is about to be created
3281 * (the inverse results in an error message from bdrv_open_common()) */
3282 assert(!(flags & BDRV_O_PROTOCOL) || !file);
3283
b6ce07aa 3284 /* Open the image */
82dc8b41 3285 ret = bdrv_open_common(bs, file, options, &local_err);
b6ce07aa 3286 if (ret < 0) {
8bfea15d 3287 goto fail;
6987307c
CH
3288 }
3289
4e4bf5c4 3290 if (file) {
5696c6e3 3291 blk_unref(file);
f500a6d3
KW
3292 file = NULL;
3293 }
3294
b6ce07aa 3295 /* If there is a backing file, use it */
9156df12 3296 if ((flags & BDRV_O_NO_BACKING) == 0) {
d9b7b057 3297 ret = bdrv_open_backing_file(bs, options, "backing", &local_err);
b6ce07aa 3298 if (ret < 0) {
b6ad491a 3299 goto close_and_fail;
b6ce07aa 3300 }
b6ce07aa
KW
3301 }
3302
50196d7a
AG
3303 /* Remove all children options and references
3304 * from bs->options and bs->explicit_options */
2f624b80
AG
3305 QLIST_FOREACH(child, &bs->children, next) {
3306 char *child_key_dot;
3307 child_key_dot = g_strdup_printf("%s.", child->name);
3308 qdict_extract_subqdict(bs->explicit_options, NULL, child_key_dot);
3309 qdict_extract_subqdict(bs->options, NULL, child_key_dot);
50196d7a
AG
3310 qdict_del(bs->explicit_options, child->name);
3311 qdict_del(bs->options, child->name);
2f624b80
AG
3312 g_free(child_key_dot);
3313 }
3314
b6ad491a 3315 /* Check if any unknown options were used */
7ad2757f 3316 if (qdict_size(options) != 0) {
b6ad491a 3317 const QDictEntry *entry = qdict_first(options);
5acd9d81
HR
3318 if (flags & BDRV_O_PROTOCOL) {
3319 error_setg(errp, "Block protocol '%s' doesn't support the option "
3320 "'%s'", drv->format_name, entry->key);
3321 } else {
d0e46a55
HR
3322 error_setg(errp,
3323 "Block format '%s' does not support the option '%s'",
3324 drv->format_name, entry->key);
5acd9d81 3325 }
b6ad491a 3326
b6ad491a
KW
3327 goto close_and_fail;
3328 }
b6ad491a 3329
c01c214b 3330 bdrv_parent_cb_change_media(bs, true);
b6ce07aa 3331
cb3e7f08 3332 qobject_unref(options);
8961be33 3333 options = NULL;
dd62f1ca
KW
3334
3335 /* For snapshot=on, create a temporary qcow2 overlay. bs points to the
3336 * temporary snapshot afterwards. */
3337 if (snapshot_flags) {
66836189
HR
3338 BlockDriverState *snapshot_bs;
3339 snapshot_bs = bdrv_append_temp_snapshot(bs, snapshot_flags,
3340 snapshot_options, &local_err);
73176bee 3341 snapshot_options = NULL;
dd62f1ca
KW
3342 if (local_err) {
3343 goto close_and_fail;
3344 }
5b363937
HR
3345 /* We are not going to return bs but the overlay on top of it
3346 * (snapshot_bs); thus, we have to drop the strong reference to bs
3347 * (which we obtained by calling bdrv_new()). bs will not be deleted,
3348 * though, because the overlay still has a reference to it. */
3349 bdrv_unref(bs);
3350 bs = snapshot_bs;
dd62f1ca
KW
3351 }
3352
5b363937 3353 return bs;
b6ce07aa 3354
8bfea15d 3355fail:
5696c6e3 3356 blk_unref(file);
cb3e7f08
MAL
3357 qobject_unref(snapshot_options);
3358 qobject_unref(bs->explicit_options);
3359 qobject_unref(bs->options);
3360 qobject_unref(options);
de9c0cec 3361 bs->options = NULL;
998cbd6a 3362 bs->explicit_options = NULL;
5b363937 3363 bdrv_unref(bs);
621ff94d 3364 error_propagate(errp, local_err);
5b363937 3365 return NULL;
de9c0cec 3366
b6ad491a 3367close_and_fail:
5b363937 3368 bdrv_unref(bs);
cb3e7f08
MAL
3369 qobject_unref(snapshot_options);
3370 qobject_unref(options);
621ff94d 3371 error_propagate(errp, local_err);
5b363937 3372 return NULL;
b6ce07aa
KW
3373}
3374
5b363937
HR
3375BlockDriverState *bdrv_open(const char *filename, const char *reference,
3376 QDict *options, int flags, Error **errp)
f3930ed0 3377{
5b363937 3378 return bdrv_open_inherit(filename, reference, options, flags, NULL,
ce343771 3379 NULL, errp);
f3930ed0
KW
3380}
3381
faf116b4
AG
3382/* Return true if the NULL-terminated @list contains @str */
3383static bool is_str_in_list(const char *str, const char *const *list)
3384{
3385 if (str && list) {
3386 int i;
3387 for (i = 0; list[i] != NULL; i++) {
3388 if (!strcmp(str, list[i])) {
3389 return true;
3390 }
3391 }
3392 }
3393 return false;
3394}
3395
3396/*
3397 * Check that every option set in @bs->options is also set in
3398 * @new_opts.
3399 *
3400 * Options listed in the common_options list and in
3401 * @bs->drv->mutable_opts are skipped.
3402 *
3403 * Return 0 on success, otherwise return -EINVAL and set @errp.
3404 */
3405static int bdrv_reset_options_allowed(BlockDriverState *bs,
3406 const QDict *new_opts, Error **errp)
3407{
3408 const QDictEntry *e;
3409 /* These options are common to all block drivers and are handled
3410 * in bdrv_reopen_prepare() so they can be left out of @new_opts */
3411 const char *const common_options[] = {
3412 "node-name", "discard", "cache.direct", "cache.no-flush",
3413 "read-only", "auto-read-only", "detect-zeroes", NULL
3414 };
3415
3416 for (e = qdict_first(bs->options); e; e = qdict_next(bs->options, e)) {
3417 if (!qdict_haskey(new_opts, e->key) &&
3418 !is_str_in_list(e->key, common_options) &&
3419 !is_str_in_list(e->key, bs->drv->mutable_opts)) {
3420 error_setg(errp, "Option '%s' cannot be reset "
3421 "to its default value", e->key);
3422 return -EINVAL;
3423 }
3424 }
3425
3426 return 0;
3427}
3428
cb828c31
AG
3429/*
3430 * Returns true if @child can be reached recursively from @bs
3431 */
3432static bool bdrv_recurse_has_child(BlockDriverState *bs,
3433 BlockDriverState *child)
3434{
3435 BdrvChild *c;
3436
3437 if (bs == child) {
3438 return true;
3439 }
3440
3441 QLIST_FOREACH(c, &bs->children, next) {
3442 if (bdrv_recurse_has_child(c->bs, child)) {
3443 return true;
3444 }
3445 }
3446
3447 return false;
3448}
3449
e971aa12
JC
3450/*
3451 * Adds a BlockDriverState to a simple queue for an atomic, transactional
3452 * reopen of multiple devices.
3453 *
859443b0 3454 * bs_queue can either be an existing BlockReopenQueue that has had QTAILQ_INIT
e971aa12
JC
3455 * already performed, or alternatively may be NULL a new BlockReopenQueue will
3456 * be created and initialized. This newly created BlockReopenQueue should be
3457 * passed back in for subsequent calls that are intended to be of the same
3458 * atomic 'set'.
3459 *
3460 * bs is the BlockDriverState to add to the reopen queue.
3461 *
4d2cb092
KW
3462 * options contains the changed options for the associated bs
3463 * (the BlockReopenQueue takes ownership)
3464 *
e971aa12
JC
3465 * flags contains the open flags for the associated bs
3466 *
3467 * returns a pointer to bs_queue, which is either the newly allocated
3468 * bs_queue, or the existing bs_queue being used.
3469 *
1a63a907 3470 * bs must be drained between bdrv_reopen_queue() and bdrv_reopen_multiple().
e971aa12 3471 */
28518102
KW
3472static BlockReopenQueue *bdrv_reopen_queue_child(BlockReopenQueue *bs_queue,
3473 BlockDriverState *bs,
3474 QDict *options,
28518102
KW
3475 const BdrvChildRole *role,
3476 QDict *parent_options,
077e8e20
AG
3477 int parent_flags,
3478 bool keep_old_opts)
e971aa12
JC
3479{
3480 assert(bs != NULL);
3481
3482 BlockReopenQueueEntry *bs_entry;
67251a31 3483 BdrvChild *child;
9aa09ddd
AG
3484 QDict *old_options, *explicit_options, *options_copy;
3485 int flags;
3486 QemuOpts *opts;
67251a31 3487
1a63a907
KW
3488 /* Make sure that the caller remembered to use a drained section. This is
3489 * important to avoid graph changes between the recursive queuing here and
3490 * bdrv_reopen_multiple(). */
3491 assert(bs->quiesce_counter > 0);
3492
e971aa12
JC
3493 if (bs_queue == NULL) {
3494 bs_queue = g_new0(BlockReopenQueue, 1);
859443b0 3495 QTAILQ_INIT(bs_queue);
e971aa12
JC
3496 }
3497
4d2cb092
KW
3498 if (!options) {
3499 options = qdict_new();
3500 }
3501
5b7ba05f 3502 /* Check if this BlockDriverState is already in the queue */
859443b0 3503 QTAILQ_FOREACH(bs_entry, bs_queue, entry) {
5b7ba05f
AG
3504 if (bs == bs_entry->state.bs) {
3505 break;
3506 }
3507 }
3508
28518102
KW
3509 /*
3510 * Precedence of options:
3511 * 1. Explicitly passed in options (highest)
9aa09ddd
AG
3512 * 2. Retained from explicitly set options of bs
3513 * 3. Inherited from parent node
3514 * 4. Retained from effective options of bs
28518102
KW
3515 */
3516
145f598e 3517 /* Old explicitly set values (don't overwrite by inherited value) */
077e8e20
AG
3518 if (bs_entry || keep_old_opts) {
3519 old_options = qdict_clone_shallow(bs_entry ?
3520 bs_entry->state.explicit_options :
3521 bs->explicit_options);
3522 bdrv_join_options(bs, options, old_options);
3523 qobject_unref(old_options);
5b7ba05f 3524 }
145f598e
KW
3525
3526 explicit_options = qdict_clone_shallow(options);
3527
28518102
KW
3528 /* Inherit from parent node */
3529 if (parent_options) {
9aa09ddd 3530 flags = 0;
8e2160e2 3531 role->inherit_options(&flags, options, parent_flags, parent_options);
9aa09ddd
AG
3532 } else {
3533 flags = bdrv_get_flags(bs);
28518102
KW
3534 }
3535
077e8e20
AG
3536 if (keep_old_opts) {
3537 /* Old values are used for options that aren't set yet */
3538 old_options = qdict_clone_shallow(bs->options);
3539 bdrv_join_options(bs, options, old_options);
3540 qobject_unref(old_options);
3541 }
4d2cb092 3542
9aa09ddd
AG
3543 /* We have the final set of options so let's update the flags */
3544 options_copy = qdict_clone_shallow(options);
3545 opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
3546 qemu_opts_absorb_qdict(opts, options_copy, NULL);
3547 update_flags_from_options(&flags, opts);
3548 qemu_opts_del(opts);
3549 qobject_unref(options_copy);
3550
fd452021 3551 /* bdrv_open_inherit() sets and clears some additional flags internally */
f1f25a2e 3552 flags &= ~BDRV_O_PROTOCOL;
fd452021
KW
3553 if (flags & BDRV_O_RDWR) {
3554 flags |= BDRV_O_ALLOW_RDWR;
3555 }
f1f25a2e 3556
1857c97b
KW
3557 if (!bs_entry) {
3558 bs_entry = g_new0(BlockReopenQueueEntry, 1);
859443b0 3559 QTAILQ_INSERT_TAIL(bs_queue, bs_entry, entry);
1857c97b 3560 } else {
cb3e7f08
MAL
3561 qobject_unref(bs_entry->state.options);
3562 qobject_unref(bs_entry->state.explicit_options);
1857c97b
KW
3563 }
3564
3565 bs_entry->state.bs = bs;
3566 bs_entry->state.options = options;
3567 bs_entry->state.explicit_options = explicit_options;
3568 bs_entry->state.flags = flags;
3569
30450259
KW
3570 /* This needs to be overwritten in bdrv_reopen_prepare() */
3571 bs_entry->state.perm = UINT64_MAX;
3572 bs_entry->state.shared_perm = 0;
3573
8546632e
AG
3574 /*
3575 * If keep_old_opts is false then it means that unspecified
3576 * options must be reset to their original value. We don't allow
3577 * resetting 'backing' but we need to know if the option is
3578 * missing in order to decide if we have to return an error.
3579 */
3580 if (!keep_old_opts) {
3581 bs_entry->state.backing_missing =
3582 !qdict_haskey(options, "backing") &&
3583 !qdict_haskey(options, "backing.driver");
3584 }
3585
67251a31 3586 QLIST_FOREACH(child, &bs->children, next) {
8546632e
AG
3587 QDict *new_child_options = NULL;
3588 bool child_keep_old = keep_old_opts;
67251a31 3589
4c9dfe5d
KW
3590 /* reopen can only change the options of block devices that were
3591 * implicitly created and inherited options. For other (referenced)
3592 * block devices, a syntax like "backing.foo" results in an error. */
67251a31
KW
3593 if (child->bs->inherits_from != bs) {
3594 continue;
3595 }
3596
8546632e
AG
3597 /* Check if the options contain a child reference */
3598 if (qdict_haskey(options, child->name)) {
3599 const char *childref = qdict_get_try_str(options, child->name);
3600 /*
3601 * The current child must not be reopened if the child
3602 * reference is null or points to a different node.
3603 */
3604 if (g_strcmp0(childref, child->bs->node_name)) {
3605 continue;
3606 }
3607 /*
3608 * If the child reference points to the current child then
3609 * reopen it with its existing set of options (note that
3610 * it can still inherit new options from the parent).
3611 */
3612 child_keep_old = true;
3613 } else {
3614 /* Extract child options ("child-name.*") */
3615 char *child_key_dot = g_strdup_printf("%s.", child->name);
3616 qdict_extract_subqdict(explicit_options, NULL, child_key_dot);
3617 qdict_extract_subqdict(options, &new_child_options, child_key_dot);
3618 g_free(child_key_dot);
3619 }
4c9dfe5d 3620
9aa09ddd 3621 bdrv_reopen_queue_child(bs_queue, child->bs, new_child_options,
8546632e 3622 child->role, options, flags, child_keep_old);
e971aa12
JC
3623 }
3624
e971aa12
JC
3625 return bs_queue;
3626}
3627
28518102
KW
3628BlockReopenQueue *bdrv_reopen_queue(BlockReopenQueue *bs_queue,
3629 BlockDriverState *bs,
077e8e20 3630 QDict *options, bool keep_old_opts)
28518102 3631{
077e8e20
AG
3632 return bdrv_reopen_queue_child(bs_queue, bs, options, NULL, NULL, 0,
3633 keep_old_opts);
28518102
KW
3634}
3635
e971aa12
JC
3636/*
3637 * Reopen multiple BlockDriverStates atomically & transactionally.
3638 *
3639 * The queue passed in (bs_queue) must have been built up previous
3640 * via bdrv_reopen_queue().
3641 *
3642 * Reopens all BDS specified in the queue, with the appropriate
3643 * flags. All devices are prepared for reopen, and failure of any
50d6a8a3 3644 * device will cause all device changes to be abandoned, and intermediate
e971aa12
JC
3645 * data cleaned up.
3646 *
3647 * If all devices prepare successfully, then the changes are committed
3648 * to all devices.
3649 *
1a63a907
KW
3650 * All affected nodes must be drained between bdrv_reopen_queue() and
3651 * bdrv_reopen_multiple().
e971aa12 3652 */
5019aece 3653int bdrv_reopen_multiple(BlockReopenQueue *bs_queue, Error **errp)
e971aa12
JC
3654{
3655 int ret = -1;
3656 BlockReopenQueueEntry *bs_entry, *next;
e971aa12
JC
3657
3658 assert(bs_queue != NULL);
3659
859443b0 3660 QTAILQ_FOREACH(bs_entry, bs_queue, entry) {
1a63a907 3661 assert(bs_entry->state.bs->quiesce_counter > 0);
a4615ab3 3662 if (bdrv_reopen_prepare(&bs_entry->state, bs_queue, errp)) {
e971aa12
JC
3663 goto cleanup;
3664 }
3665 bs_entry->prepared = true;
3666 }
3667
859443b0 3668 QTAILQ_FOREACH(bs_entry, bs_queue, entry) {
69b736e7
KW
3669 BDRVReopenState *state = &bs_entry->state;
3670 ret = bdrv_check_perm(state->bs, bs_queue, state->perm,
9eab1544 3671 state->shared_perm, NULL, NULL, errp);
69b736e7
KW
3672 if (ret < 0) {
3673 goto cleanup_perm;
3674 }
cb828c31
AG
3675 /* Check if new_backing_bs would accept the new permissions */
3676 if (state->replace_backing_bs && state->new_backing_bs) {
3677 uint64_t nperm, nshared;
3678 bdrv_child_perm(state->bs, state->new_backing_bs,
3679 NULL, &child_backing, bs_queue,
3680 state->perm, state->shared_perm,
3681 &nperm, &nshared);
3682 ret = bdrv_check_update_perm(state->new_backing_bs, NULL,
9eab1544 3683 nperm, nshared, NULL, NULL, errp);
cb828c31
AG
3684 if (ret < 0) {
3685 goto cleanup_perm;
3686 }
3687 }
69b736e7
KW
3688 bs_entry->perms_checked = true;
3689 }
3690
fcd6a4f4
VSO
3691 /*
3692 * If we reach this point, we have success and just need to apply the
3693 * changes.
3694 *
3695 * Reverse order is used to comfort qcow2 driver: on commit it need to write
3696 * IN_USE flag to the image, to mark bitmaps in the image as invalid. But
3697 * children are usually goes after parents in reopen-queue, so go from last
3698 * to first element.
e971aa12 3699 */
fcd6a4f4 3700 QTAILQ_FOREACH_REVERSE(bs_entry, bs_queue, entry) {
e971aa12
JC
3701 bdrv_reopen_commit(&bs_entry->state);
3702 }
3703
3704 ret = 0;
69b736e7 3705cleanup_perm:
859443b0 3706 QTAILQ_FOREACH_SAFE(bs_entry, bs_queue, entry, next) {
69b736e7
KW
3707 BDRVReopenState *state = &bs_entry->state;
3708
3709 if (!bs_entry->perms_checked) {
3710 continue;
3711 }
e971aa12 3712
69b736e7
KW
3713 if (ret == 0) {
3714 bdrv_set_perm(state->bs, state->perm, state->shared_perm);
3715 } else {
3716 bdrv_abort_perm_update(state->bs);
cb828c31
AG
3717 if (state->replace_backing_bs && state->new_backing_bs) {
3718 bdrv_abort_perm_update(state->new_backing_bs);
3719 }
69b736e7
KW
3720 }
3721 }
17e1e2be
PK
3722
3723 if (ret == 0) {
3724 QTAILQ_FOREACH_REVERSE(bs_entry, bs_queue, entry) {
3725 BlockDriverState *bs = bs_entry->state.bs;
3726
3727 if (bs->drv->bdrv_reopen_commit_post)
3728 bs->drv->bdrv_reopen_commit_post(&bs_entry->state);
3729 }
3730 }
e971aa12 3731cleanup:
859443b0 3732 QTAILQ_FOREACH_SAFE(bs_entry, bs_queue, entry, next) {
1bab38e7
AG
3733 if (ret) {
3734 if (bs_entry->prepared) {
3735 bdrv_reopen_abort(&bs_entry->state);
3736 }
cb3e7f08 3737 qobject_unref(bs_entry->state.explicit_options);
4c8350fe 3738 qobject_unref(bs_entry->state.options);
e971aa12 3739 }
cb828c31
AG
3740 if (bs_entry->state.new_backing_bs) {
3741 bdrv_unref(bs_entry->state.new_backing_bs);
3742 }
e971aa12
JC
3743 g_free(bs_entry);
3744 }
3745 g_free(bs_queue);
40840e41 3746
e971aa12
JC
3747 return ret;
3748}
3749
6e1000a8
AG
3750int bdrv_reopen_set_read_only(BlockDriverState *bs, bool read_only,
3751 Error **errp)
3752{
3753 int ret;
3754 BlockReopenQueue *queue;
3755 QDict *opts = qdict_new();
3756
3757 qdict_put_bool(opts, BDRV_OPT_READ_ONLY, read_only);
3758
3759 bdrv_subtree_drained_begin(bs);
077e8e20 3760 queue = bdrv_reopen_queue(NULL, bs, opts, true);
5019aece 3761 ret = bdrv_reopen_multiple(queue, errp);
6e1000a8
AG
3762 bdrv_subtree_drained_end(bs);
3763
3764 return ret;
3765}
3766
30450259
KW
3767static BlockReopenQueueEntry *find_parent_in_reopen_queue(BlockReopenQueue *q,
3768 BdrvChild *c)
3769{
3770 BlockReopenQueueEntry *entry;
3771
859443b0 3772 QTAILQ_FOREACH(entry, q, entry) {
30450259
KW
3773 BlockDriverState *bs = entry->state.bs;
3774 BdrvChild *child;
3775
3776 QLIST_FOREACH(child, &bs->children, next) {
3777 if (child == c) {
3778 return entry;
3779 }
3780 }
3781 }
3782
3783 return NULL;
3784}
3785
3786static void bdrv_reopen_perm(BlockReopenQueue *q, BlockDriverState *bs,
3787 uint64_t *perm, uint64_t *shared)
3788{
3789 BdrvChild *c;
3790 BlockReopenQueueEntry *parent;
3791 uint64_t cumulative_perms = 0;
3792 uint64_t cumulative_shared_perms = BLK_PERM_ALL;
3793
3794 QLIST_FOREACH(c, &bs->parents, next_parent) {
3795 parent = find_parent_in_reopen_queue(q, c);
3796 if (!parent) {
3797 cumulative_perms |= c->perm;
3798 cumulative_shared_perms &= c->shared_perm;
3799 } else {
3800 uint64_t nperm, nshared;
3801
3802 bdrv_child_perm(parent->state.bs, bs, c, c->role, q,
3803 parent->state.perm, parent->state.shared_perm,
3804 &nperm, &nshared);
3805
3806 cumulative_perms |= nperm;
3807 cumulative_shared_perms &= nshared;
3808 }
3809 }
3810 *perm = cumulative_perms;
3811 *shared = cumulative_shared_perms;
3812}
e971aa12 3813
1de6b45f
KW
3814static bool bdrv_reopen_can_attach(BlockDriverState *parent,
3815 BdrvChild *child,
3816 BlockDriverState *new_child,
3817 Error **errp)
3818{
3819 AioContext *parent_ctx = bdrv_get_aio_context(parent);
3820 AioContext *child_ctx = bdrv_get_aio_context(new_child);
3821 GSList *ignore;
3822 bool ret;
3823
3824 ignore = g_slist_prepend(NULL, child);
3825 ret = bdrv_can_set_aio_context(new_child, parent_ctx, &ignore, NULL);
3826 g_slist_free(ignore);
3827 if (ret) {
3828 return ret;
3829 }
3830
3831 ignore = g_slist_prepend(NULL, child);
3832 ret = bdrv_can_set_aio_context(parent, child_ctx, &ignore, errp);
3833 g_slist_free(ignore);
3834 return ret;
3835}
3836
cb828c31
AG
3837/*
3838 * Take a BDRVReopenState and check if the value of 'backing' in the
3839 * reopen_state->options QDict is valid or not.
3840 *
3841 * If 'backing' is missing from the QDict then return 0.
3842 *
3843 * If 'backing' contains the node name of the backing file of
3844 * reopen_state->bs then return 0.
3845 *
3846 * If 'backing' contains a different node name (or is null) then check
3847 * whether the current backing file can be replaced with the new one.
3848 * If that's the case then reopen_state->replace_backing_bs is set to
3849 * true and reopen_state->new_backing_bs contains a pointer to the new
3850 * backing BlockDriverState (or NULL).
3851 *
3852 * Return 0 on success, otherwise return < 0 and set @errp.
3853 */
3854static int bdrv_reopen_parse_backing(BDRVReopenState *reopen_state,
3855 Error **errp)
3856{
3857 BlockDriverState *bs = reopen_state->bs;
3858 BlockDriverState *overlay_bs, *new_backing_bs;
3859 QObject *value;
3860 const char *str;
3861
3862 value = qdict_get(reopen_state->options, "backing");
3863 if (value == NULL) {
3864 return 0;
3865 }
3866
3867 switch (qobject_type(value)) {
3868 case QTYPE_QNULL:
3869 new_backing_bs = NULL;
3870 break;
3871 case QTYPE_QSTRING:
3872 str = qobject_get_try_str(value);
3873 new_backing_bs = bdrv_lookup_bs(NULL, str, errp);
3874 if (new_backing_bs == NULL) {
3875 return -EINVAL;
3876 } else if (bdrv_recurse_has_child(new_backing_bs, bs)) {
3877 error_setg(errp, "Making '%s' a backing file of '%s' "
3878 "would create a cycle", str, bs->node_name);
3879 return -EINVAL;
3880 }
3881 break;
3882 default:
3883 /* 'backing' does not allow any other data type */
3884 g_assert_not_reached();
3885 }
3886
3887 /*
1de6b45f
KW
3888 * Check AioContext compatibility so that the bdrv_set_backing_hd() call in
3889 * bdrv_reopen_commit() won't fail.
cb828c31
AG
3890 */
3891 if (new_backing_bs) {
1de6b45f 3892 if (!bdrv_reopen_can_attach(bs, bs->backing, new_backing_bs, errp)) {
cb828c31
AG
3893 return -EINVAL;
3894 }
3895 }
3896
3897 /*
3898 * Find the "actual" backing file by skipping all links that point
3899 * to an implicit node, if any (e.g. a commit filter node).
3900 */
3901 overlay_bs = bs;
3902 while (backing_bs(overlay_bs) && backing_bs(overlay_bs)->implicit) {
3903 overlay_bs = backing_bs(overlay_bs);
3904 }
3905
3906 /* If we want to replace the backing file we need some extra checks */
3907 if (new_backing_bs != backing_bs(overlay_bs)) {
3908 /* Check for implicit nodes between bs and its backing file */
3909 if (bs != overlay_bs) {
3910 error_setg(errp, "Cannot change backing link if '%s' has "
3911 "an implicit backing file", bs->node_name);
3912 return -EPERM;
3913 }
3914 /* Check if the backing link that we want to replace is frozen */
3915 if (bdrv_is_backing_chain_frozen(overlay_bs, backing_bs(overlay_bs),
3916 errp)) {
3917 return -EPERM;
3918 }
3919 reopen_state->replace_backing_bs = true;
3920 if (new_backing_bs) {
3921 bdrv_ref(new_backing_bs);
3922 reopen_state->new_backing_bs = new_backing_bs;
3923 }
3924 }
3925
3926 return 0;
3927}
3928
e971aa12
JC
3929/*
3930 * Prepares a BlockDriverState for reopen. All changes are staged in the
3931 * 'opaque' field of the BDRVReopenState, which is used and allocated by
3932 * the block driver layer .bdrv_reopen_prepare()
3933 *
3934 * bs is the BlockDriverState to reopen
3935 * flags are the new open flags
3936 * queue is the reopen queue
3937 *
3938 * Returns 0 on success, non-zero on error. On error errp will be set
3939 * as well.
3940 *
3941 * On failure, bdrv_reopen_abort() will be called to clean up any data.
3942 * It is the responsibility of the caller to then call the abort() or
3943 * commit() for any other BDS that have been left in a prepare() state
3944 *
3945 */
3946int bdrv_reopen_prepare(BDRVReopenState *reopen_state, BlockReopenQueue *queue,
3947 Error **errp)
3948{
3949 int ret = -1;
e6d79c41 3950 int old_flags;
e971aa12
JC
3951 Error *local_err = NULL;
3952 BlockDriver *drv;
ccf9dc07 3953 QemuOpts *opts;
4c8350fe 3954 QDict *orig_reopen_opts;
593b3071 3955 char *discard = NULL;
3d8ce171 3956 bool read_only;
9ad08c44 3957 bool drv_prepared = false;
e971aa12
JC
3958
3959 assert(reopen_state != NULL);
3960 assert(reopen_state->bs->drv != NULL);
3961 drv = reopen_state->bs->drv;
3962
4c8350fe
AG
3963 /* This function and each driver's bdrv_reopen_prepare() remove
3964 * entries from reopen_state->options as they are processed, so
3965 * we need to make a copy of the original QDict. */
3966 orig_reopen_opts = qdict_clone_shallow(reopen_state->options);
3967
ccf9dc07
KW
3968 /* Process generic block layer options */
3969 opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
3970 qemu_opts_absorb_qdict(opts, reopen_state->options, &local_err);
3971 if (local_err) {
3972 error_propagate(errp, local_err);
3973 ret = -EINVAL;
3974 goto error;
3975 }
3976
e6d79c41
AG
3977 /* This was already called in bdrv_reopen_queue_child() so the flags
3978 * are up-to-date. This time we simply want to remove the options from
3979 * QemuOpts in order to indicate that they have been processed. */
3980 old_flags = reopen_state->flags;
91a097e7 3981 update_flags_from_options(&reopen_state->flags, opts);
e6d79c41 3982 assert(old_flags == reopen_state->flags);
91a097e7 3983
415bbca8 3984 discard = qemu_opt_get_del(opts, BDRV_OPT_DISCARD);
593b3071
AG
3985 if (discard != NULL) {
3986 if (bdrv_parse_discard_flags(discard, &reopen_state->flags) != 0) {
3987 error_setg(errp, "Invalid discard option");
3988 ret = -EINVAL;
3989 goto error;
3990 }
3991 }
3992
543770bd
AG
3993 reopen_state->detect_zeroes =
3994 bdrv_parse_detect_zeroes(opts, reopen_state->flags, &local_err);
3995 if (local_err) {
3996 error_propagate(errp, local_err);
3997 ret = -EINVAL;
3998 goto error;
3999 }
4000
57f9db9a
AG
4001 /* All other options (including node-name and driver) must be unchanged.
4002 * Put them back into the QDict, so that they are checked at the end
4003 * of this function. */
4004 qemu_opts_to_qdict(opts, reopen_state->options);
ccf9dc07 4005
3d8ce171
JC
4006 /* If we are to stay read-only, do not allow permission change
4007 * to r/w. Attempting to set to r/w may fail if either BDRV_O_ALLOW_RDWR is
4008 * not set, or if the BDS still has copy_on_read enabled */
4009 read_only = !(reopen_state->flags & BDRV_O_RDWR);
54a32bfe 4010 ret = bdrv_can_set_read_only(reopen_state->bs, read_only, true, &local_err);
3d8ce171
JC
4011 if (local_err) {
4012 error_propagate(errp, local_err);
e971aa12
JC
4013 goto error;
4014 }
4015
30450259
KW
4016 /* Calculate required permissions after reopening */
4017 bdrv_reopen_perm(queue, reopen_state->bs,
4018 &reopen_state->perm, &reopen_state->shared_perm);
e971aa12
JC
4019
4020 ret = bdrv_flush(reopen_state->bs);
4021 if (ret) {
455b0fde 4022 error_setg_errno(errp, -ret, "Error flushing drive");
e971aa12
JC
4023 goto error;
4024 }
4025
4026 if (drv->bdrv_reopen_prepare) {
faf116b4
AG
4027 /*
4028 * If a driver-specific option is missing, it means that we
4029 * should reset it to its default value.
4030 * But not all options allow that, so we need to check it first.
4031 */
4032 ret = bdrv_reset_options_allowed(reopen_state->bs,
4033 reopen_state->options, errp);
4034 if (ret) {
4035 goto error;
4036 }
4037
e971aa12
JC
4038 ret = drv->bdrv_reopen_prepare(reopen_state, queue, &local_err);
4039 if (ret) {
4040 if (local_err != NULL) {
4041 error_propagate(errp, local_err);
4042 } else {
f30c66ba 4043 bdrv_refresh_filename(reopen_state->bs);
d8b6895f
LC
4044 error_setg(errp, "failed while preparing to reopen image '%s'",
4045 reopen_state->bs->filename);
e971aa12
JC
4046 }
4047 goto error;
4048 }
4049 } else {
4050 /* It is currently mandatory to have a bdrv_reopen_prepare()
4051 * handler for each supported drv. */
81e5f78a
AG
4052 error_setg(errp, "Block format '%s' used by node '%s' "
4053 "does not support reopening files", drv->format_name,
4054 bdrv_get_device_or_node_name(reopen_state->bs));
e971aa12
JC
4055 ret = -1;
4056 goto error;
4057 }
4058
9ad08c44
HR
4059 drv_prepared = true;
4060
bacd9b87
AG
4061 /*
4062 * We must provide the 'backing' option if the BDS has a backing
4063 * file or if the image file has a backing file name as part of
4064 * its metadata. Otherwise the 'backing' option can be omitted.
4065 */
4066 if (drv->supports_backing && reopen_state->backing_missing &&
4067 (backing_bs(reopen_state->bs) || reopen_state->bs->backing_file[0])) {
8546632e
AG
4068 error_setg(errp, "backing is missing for '%s'",
4069 reopen_state->bs->node_name);
4070 ret = -EINVAL;
4071 goto error;
4072 }
4073
cb828c31
AG
4074 /*
4075 * Allow changing the 'backing' option. The new value can be
4076 * either a reference to an existing node (using its node name)
4077 * or NULL to simply detach the current backing file.
4078 */
4079 ret = bdrv_reopen_parse_backing(reopen_state, errp);
4080 if (ret < 0) {
4081 goto error;
4082 }
4083 qdict_del(reopen_state->options, "backing");
4084
4d2cb092
KW
4085 /* Options that are not handled are only okay if they are unchanged
4086 * compared to the old state. It is expected that some options are only
4087 * used for the initial open, but not reopen (e.g. filename) */
4088 if (qdict_size(reopen_state->options)) {
4089 const QDictEntry *entry = qdict_first(reopen_state->options);
4090
4091 do {
54fd1b0d
HR
4092 QObject *new = entry->value;
4093 QObject *old = qdict_get(reopen_state->bs->options, entry->key);
4094
db905283
AG
4095 /* Allow child references (child_name=node_name) as long as they
4096 * point to the current child (i.e. everything stays the same). */
4097 if (qobject_type(new) == QTYPE_QSTRING) {
4098 BdrvChild *child;
4099 QLIST_FOREACH(child, &reopen_state->bs->children, next) {
4100 if (!strcmp(child->name, entry->key)) {
4101 break;
4102 }
4103 }
4104
4105 if (child) {
4106 const char *str = qobject_get_try_str(new);
4107 if (!strcmp(child->bs->node_name, str)) {
4108 continue; /* Found child with this name, skip option */
4109 }
4110 }
4111 }
4112
129c7d1c 4113 /*
54fd1b0d
HR
4114 * TODO: When using -drive to specify blockdev options, all values
4115 * will be strings; however, when using -blockdev, blockdev-add or
4116 * filenames using the json:{} pseudo-protocol, they will be
4117 * correctly typed.
4118 * In contrast, reopening options are (currently) always strings
4119 * (because you can only specify them through qemu-io; all other
4120 * callers do not specify any options).
4121 * Therefore, when using anything other than -drive to create a BDS,
4122 * this cannot detect non-string options as unchanged, because
4123 * qobject_is_equal() always returns false for objects of different
4124 * type. In the future, this should be remedied by correctly typing
4125 * all options. For now, this is not too big of an issue because
4126 * the user can simply omit options which cannot be changed anyway,
4127 * so they will stay unchanged.
129c7d1c 4128 */
54fd1b0d 4129 if (!qobject_is_equal(new, old)) {
4d2cb092
KW
4130 error_setg(errp, "Cannot change the option '%s'", entry->key);
4131 ret = -EINVAL;
4132 goto error;
4133 }
4134 } while ((entry = qdict_next(reopen_state->options, entry)));
4135 }
4136
e971aa12
JC
4137 ret = 0;
4138
4c8350fe
AG
4139 /* Restore the original reopen_state->options QDict */
4140 qobject_unref(reopen_state->options);
4141 reopen_state->options = qobject_ref(orig_reopen_opts);
4142
e971aa12 4143error:
9ad08c44
HR
4144 if (ret < 0 && drv_prepared) {
4145 /* drv->bdrv_reopen_prepare() has succeeded, so we need to
4146 * call drv->bdrv_reopen_abort() before signaling an error
4147 * (bdrv_reopen_multiple() will not call bdrv_reopen_abort()
4148 * when the respective bdrv_reopen_prepare() has failed) */
4149 if (drv->bdrv_reopen_abort) {
4150 drv->bdrv_reopen_abort(reopen_state);
4151 }
4152 }
ccf9dc07 4153 qemu_opts_del(opts);
4c8350fe 4154 qobject_unref(orig_reopen_opts);
593b3071 4155 g_free(discard);
e971aa12
JC
4156 return ret;
4157}
4158
4159/*
4160 * Takes the staged changes for the reopen from bdrv_reopen_prepare(), and
4161 * makes them final by swapping the staging BlockDriverState contents into
4162 * the active BlockDriverState contents.
4163 */
4164void bdrv_reopen_commit(BDRVReopenState *reopen_state)
4165{
4166 BlockDriver *drv;
50bf65ba 4167 BlockDriverState *bs;
50196d7a 4168 BdrvChild *child;
e971aa12
JC
4169
4170 assert(reopen_state != NULL);
50bf65ba
VSO
4171 bs = reopen_state->bs;
4172 drv = bs->drv;
e971aa12
JC
4173 assert(drv != NULL);
4174
4175 /* If there are any driver level actions to take */
4176 if (drv->bdrv_reopen_commit) {
4177 drv->bdrv_reopen_commit(reopen_state);
4178 }
4179
4180 /* set BDS specific flags now */
cb3e7f08 4181 qobject_unref(bs->explicit_options);
4c8350fe 4182 qobject_unref(bs->options);
145f598e 4183
50bf65ba 4184 bs->explicit_options = reopen_state->explicit_options;
4c8350fe 4185 bs->options = reopen_state->options;
50bf65ba
VSO
4186 bs->open_flags = reopen_state->flags;
4187 bs->read_only = !(reopen_state->flags & BDRV_O_RDWR);
543770bd 4188 bs->detect_zeroes = reopen_state->detect_zeroes;
355ef4ac 4189
cb828c31
AG
4190 if (reopen_state->replace_backing_bs) {
4191 qdict_del(bs->explicit_options, "backing");
4192 qdict_del(bs->options, "backing");
4193 }
4194
50196d7a
AG
4195 /* Remove child references from bs->options and bs->explicit_options.
4196 * Child options were already removed in bdrv_reopen_queue_child() */
4197 QLIST_FOREACH(child, &bs->children, next) {
4198 qdict_del(bs->explicit_options, child->name);
4199 qdict_del(bs->options, child->name);
4200 }
4201
cb828c31
AG
4202 /*
4203 * Change the backing file if a new one was specified. We do this
4204 * after updating bs->options, so bdrv_refresh_filename() (called
4205 * from bdrv_set_backing_hd()) has the new values.
4206 */
4207 if (reopen_state->replace_backing_bs) {
4208 BlockDriverState *old_backing_bs = backing_bs(bs);
4209 assert(!old_backing_bs || !old_backing_bs->implicit);
4210 /* Abort the permission update on the backing bs we're detaching */
4211 if (old_backing_bs) {
4212 bdrv_abort_perm_update(old_backing_bs);
4213 }
4214 bdrv_set_backing_hd(bs, reopen_state->new_backing_bs, &error_abort);
4215 }
4216
50bf65ba 4217 bdrv_refresh_limits(bs, NULL);
e971aa12
JC
4218}
4219
4220/*
4221 * Abort the reopen, and delete and free the staged changes in
4222 * reopen_state
4223 */
4224void bdrv_reopen_abort(BDRVReopenState *reopen_state)
4225{
4226 BlockDriver *drv;
4227
4228 assert(reopen_state != NULL);
4229 drv = reopen_state->bs->drv;
4230 assert(drv != NULL);
4231
4232 if (drv->bdrv_reopen_abort) {
4233 drv->bdrv_reopen_abort(reopen_state);
4234 }
4235}
4236
4237
64dff520 4238static void bdrv_close(BlockDriverState *bs)
fc01f7e7 4239{
33384421 4240 BdrvAioNotifier *ban, *ban_next;
50a3efb0 4241 BdrvChild *child, *next;
33384421 4242
30f55fb8 4243 assert(!bs->refcnt);
99b7e775 4244
fc27291d 4245 bdrv_drained_begin(bs); /* complete I/O */
58fda173 4246 bdrv_flush(bs);
53ec73e2 4247 bdrv_drain(bs); /* in case flush left pending I/O */
fc27291d 4248
3cbc002c 4249 if (bs->drv) {
3c005293
VSO
4250 if (bs->drv->bdrv_close) {
4251 bs->drv->bdrv_close(bs);
4252 }
9a4f4c31 4253 bs->drv = NULL;
50a3efb0 4254 }
9a7dedbc 4255
50a3efb0 4256 QLIST_FOREACH_SAFE(child, &bs->children, next, next) {
dd4118c7 4257 bdrv_unref_child(bs, child);
b338082b 4258 }
98f90dba 4259
dd4118c7
AG
4260 bs->backing = NULL;
4261 bs->file = NULL;
50a3efb0
AG
4262 g_free(bs->opaque);
4263 bs->opaque = NULL;
4264 atomic_set(&bs->copy_on_read, 0);
4265 bs->backing_file[0] = '\0';
4266 bs->backing_format[0] = '\0';
4267 bs->total_sectors = 0;
4268 bs->encrypted = false;
4269 bs->sg = false;
cb3e7f08
MAL
4270 qobject_unref(bs->options);
4271 qobject_unref(bs->explicit_options);
50a3efb0
AG
4272 bs->options = NULL;
4273 bs->explicit_options = NULL;
cb3e7f08 4274 qobject_unref(bs->full_open_options);
50a3efb0
AG
4275 bs->full_open_options = NULL;
4276
cca43ae1
VSO
4277 bdrv_release_named_dirty_bitmaps(bs);
4278 assert(QLIST_EMPTY(&bs->dirty_bitmaps));
4279
33384421
HR
4280 QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_next) {
4281 g_free(ban);
4282 }
4283 QLIST_INIT(&bs->aio_notifiers);
fc27291d 4284 bdrv_drained_end(bs);
b338082b
FB
4285}
4286
2bc93fed
MK
4287void bdrv_close_all(void)
4288{
b3b5299d 4289 assert(job_next(NULL) == NULL);
cd7fca95 4290 nbd_export_close_all();
ca9bd24c
HR
4291
4292 /* Drop references from requests still in flight, such as canceled block
4293 * jobs whose AIO context has not been polled yet */
4294 bdrv_drain_all();
2bc93fed 4295
ca9bd24c
HR
4296 blk_remove_all_bs();
4297 blockdev_close_all_bdrv_states();
ed78cda3 4298
a1a2af07 4299 assert(QTAILQ_EMPTY(&all_bdrv_states));
2bc93fed
MK
4300}
4301
d0ac0380
KW
4302static bool should_update_child(BdrvChild *c, BlockDriverState *to)
4303{
2f30b7c3
VSO
4304 GQueue *queue;
4305 GHashTable *found;
4306 bool ret;
d0ac0380
KW
4307
4308 if (c->role->stay_at_node) {
4309 return false;
4310 }
4311
ec9f10fe
HR
4312 /* If the child @c belongs to the BDS @to, replacing the current
4313 * c->bs by @to would mean to create a loop.
4314 *
4315 * Such a case occurs when appending a BDS to a backing chain.
4316 * For instance, imagine the following chain:
4317 *
4318 * guest device -> node A -> further backing chain...
4319 *
4320 * Now we create a new BDS B which we want to put on top of this
4321 * chain, so we first attach A as its backing node:
4322 *
4323 * node B
4324 * |
4325 * v
4326 * guest device -> node A -> further backing chain...
4327 *
4328 * Finally we want to replace A by B. When doing that, we want to
4329 * replace all pointers to A by pointers to B -- except for the
4330 * pointer from B because (1) that would create a loop, and (2)
4331 * that pointer should simply stay intact:
4332 *
4333 * guest device -> node B
4334 * |
4335 * v
4336 * node A -> further backing chain...
4337 *
4338 * In general, when replacing a node A (c->bs) by a node B (@to),
4339 * if A is a child of B, that means we cannot replace A by B there
4340 * because that would create a loop. Silently detaching A from B
4341 * is also not really an option. So overall just leaving A in
2f30b7c3
VSO
4342 * place there is the most sensible choice.
4343 *
4344 * We would also create a loop in any cases where @c is only
4345 * indirectly referenced by @to. Prevent this by returning false
4346 * if @c is found (by breadth-first search) anywhere in the whole
4347 * subtree of @to.
4348 */
4349
4350 ret = true;
4351 found = g_hash_table_new(NULL, NULL);
4352 g_hash_table_add(found, to);
4353 queue = g_queue_new();
4354 g_queue_push_tail(queue, to);
4355
4356 while (!g_queue_is_empty(queue)) {
4357 BlockDriverState *v = g_queue_pop_head(queue);
4358 BdrvChild *c2;
4359
4360 QLIST_FOREACH(c2, &v->children, next) {
4361 if (c2 == c) {
4362 ret = false;
4363 break;
4364 }
4365
4366 if (g_hash_table_contains(found, c2->bs)) {
4367 continue;
4368 }
4369
4370 g_queue_push_tail(queue, c2->bs);
4371 g_hash_table_add(found, c2->bs);
d0ac0380
KW
4372 }
4373 }
4374
2f30b7c3
VSO
4375 g_queue_free(queue);
4376 g_hash_table_destroy(found);
4377
4378 return ret;
d0ac0380
KW
4379}
4380
5fe31c25
KW
4381void bdrv_replace_node(BlockDriverState *from, BlockDriverState *to,
4382 Error **errp)
dd62f1ca 4383{
d0ac0380 4384 BdrvChild *c, *next;
234ac1a9 4385 GSList *list = NULL, *p;
234ac1a9
KW
4386 uint64_t perm = 0, shared = BLK_PERM_ALL;
4387 int ret;
4388
4389 /* Make sure that @from doesn't go away until we have successfully attached
4390 * all of its parents to @to. */
4391 bdrv_ref(from);
dd62f1ca 4392
f871abd6 4393 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
30dd65f3 4394 assert(bdrv_get_aio_context(from) == bdrv_get_aio_context(to));
f871abd6
KW
4395 bdrv_drained_begin(from);
4396
234ac1a9 4397 /* Put all parents into @list and calculate their cumulative permissions */
dd62f1ca 4398 QLIST_FOREACH_SAFE(c, &from->parents, next_parent, next) {
ec9f10fe 4399 assert(c->bs == from);
d0ac0380 4400 if (!should_update_child(c, to)) {
26de9438
KW
4401 continue;
4402 }
2cad1ebe
AG
4403 if (c->frozen) {
4404 error_setg(errp, "Cannot change '%s' link to '%s'",
4405 c->name, from->node_name);
4406 goto out;
4407 }
234ac1a9
KW
4408 list = g_slist_prepend(list, c);
4409 perm |= c->perm;
4410 shared &= c->shared_perm;
4411 }
4412
4413 /* Check whether the required permissions can be granted on @to, ignoring
4414 * all BdrvChild in @list so that they can't block themselves. */
9eab1544 4415 ret = bdrv_check_update_perm(to, NULL, perm, shared, list, NULL, errp);
234ac1a9
KW
4416 if (ret < 0) {
4417 bdrv_abort_perm_update(to);
4418 goto out;
4419 }
4420
4421 /* Now actually perform the change. We performed the permission check for
4422 * all elements of @list at once, so set the permissions all at once at the
4423 * very end. */
4424 for (p = list; p != NULL; p = p->next) {
4425 c = p->data;
9bd910e2 4426
dd62f1ca 4427 bdrv_ref(to);
234ac1a9 4428 bdrv_replace_child_noperm(c, to);
dd62f1ca
KW
4429 bdrv_unref(from);
4430 }
234ac1a9 4431
b503de61
VSO
4432 bdrv_get_cumulative_perm(to, &perm, &shared);
4433 bdrv_set_perm(to, perm, shared);
234ac1a9
KW
4434
4435out:
4436 g_slist_free(list);
f871abd6 4437 bdrv_drained_end(from);
234ac1a9 4438 bdrv_unref(from);
dd62f1ca
KW
4439}
4440
4ddc07ca
PB
4441/*
4442 * Add new bs contents at the top of an image chain while the chain is
4443 * live, while keeping required fields on the top layer.
4444 *
4445 * This will modify the BlockDriverState fields, and swap contents
4446 * between bs_new and bs_top. Both bs_new and bs_top are modified.
4447 *
bfb197e0 4448 * bs_new must not be attached to a BlockBackend.
4ddc07ca
PB
4449 *
4450 * This function does not create any image files.
dd62f1ca
KW
4451 *
4452 * bdrv_append() takes ownership of a bs_new reference and unrefs it because
4453 * that's what the callers commonly need. bs_new will be referenced by the old
4454 * parents of bs_top after bdrv_append() returns. If the caller needs to keep a
4455 * reference of its own, it must call bdrv_ref().
4ddc07ca 4456 */
b2c2832c
KW
4457void bdrv_append(BlockDriverState *bs_new, BlockDriverState *bs_top,
4458 Error **errp)
4ddc07ca 4459{
b2c2832c
KW
4460 Error *local_err = NULL;
4461
b2c2832c
KW
4462 bdrv_set_backing_hd(bs_new, bs_top, &local_err);
4463 if (local_err) {
4464 error_propagate(errp, local_err);
4465 goto out;
4466 }
dd62f1ca 4467
5fe31c25 4468 bdrv_replace_node(bs_top, bs_new, &local_err);
234ac1a9
KW
4469 if (local_err) {
4470 error_propagate(errp, local_err);
4471 bdrv_set_backing_hd(bs_new, NULL, &error_abort);
4472 goto out;
4473 }
4ddc07ca 4474
dd62f1ca
KW
4475 /* bs_new is now referenced by its new parents, we don't need the
4476 * additional reference any more. */
b2c2832c 4477out:
dd62f1ca 4478 bdrv_unref(bs_new);
8802d1fd
JC
4479}
4480
4f6fd349 4481static void bdrv_delete(BlockDriverState *bs)
b338082b 4482{
3718d8ab 4483 assert(bdrv_op_blocker_is_empty(bs));
4f6fd349 4484 assert(!bs->refcnt);
18846dee 4485
1b7bdbc1 4486 /* remove from list, if necessary */
63eaaae0
KW
4487 if (bs->node_name[0] != '\0') {
4488 QTAILQ_REMOVE(&graph_bdrv_states, bs, node_list);
4489 }
2c1d04e0
HR
4490 QTAILQ_REMOVE(&all_bdrv_states, bs, bs_list);
4491
30c321f9
AK
4492 bdrv_close(bs);
4493
7267c094 4494 g_free(bs);
fc01f7e7
FB
4495}
4496
e97fc193
AL
4497/*
4498 * Run consistency checks on an image
4499 *
e076f338 4500 * Returns 0 if the check could be completed (it doesn't mean that the image is
a1c7273b 4501 * free of errors) or -errno when an internal error occurred. The results of the
e076f338 4502 * check are stored in res.
e97fc193 4503 */
2fd61638
PB
4504static int coroutine_fn bdrv_co_check(BlockDriverState *bs,
4505 BdrvCheckResult *res, BdrvCheckMode fix)
e97fc193 4506{
908bcd54
HR
4507 if (bs->drv == NULL) {
4508 return -ENOMEDIUM;
4509 }
2fd61638 4510 if (bs->drv->bdrv_co_check == NULL) {
e97fc193
AL
4511 return -ENOTSUP;
4512 }
4513
e076f338 4514 memset(res, 0, sizeof(*res));
2fd61638
PB
4515 return bs->drv->bdrv_co_check(bs, res, fix);
4516}
4517
4518typedef struct CheckCo {
4519 BlockDriverState *bs;
4520 BdrvCheckResult *res;
4521 BdrvCheckMode fix;
4522 int ret;
4523} CheckCo;
4524
66a5bdf3 4525static void coroutine_fn bdrv_check_co_entry(void *opaque)
2fd61638
PB
4526{
4527 CheckCo *cco = opaque;
4528 cco->ret = bdrv_co_check(cco->bs, cco->res, cco->fix);
4720cbee 4529 aio_wait_kick();
2fd61638
PB
4530}
4531
4532int bdrv_check(BlockDriverState *bs,
4533 BdrvCheckResult *res, BdrvCheckMode fix)
4534{
4535 Coroutine *co;
4536 CheckCo cco = {
4537 .bs = bs,
4538 .res = res,
4539 .ret = -EINPROGRESS,
4540 .fix = fix,
4541 };
4542
4543 if (qemu_in_coroutine()) {
4544 /* Fast-path if already in coroutine context */
4545 bdrv_check_co_entry(&cco);
4546 } else {
4547 co = qemu_coroutine_create(bdrv_check_co_entry, &cco);
4720cbee 4548 bdrv_coroutine_enter(bs, co);
2fd61638
PB
4549 BDRV_POLL_WHILE(bs, cco.ret == -EINPROGRESS);
4550 }
4551
4552 return cco.ret;
e97fc193
AL
4553}
4554
756e6736
KW
4555/*
4556 * Return values:
4557 * 0 - success
4558 * -EINVAL - backing format specified, but no file
4559 * -ENOSPC - can't update the backing file because no space is left in the
4560 * image file header
4561 * -ENOTSUP - format driver doesn't support changing the backing file
4562 */
4563int bdrv_change_backing_file(BlockDriverState *bs,
4564 const char *backing_file, const char *backing_fmt)
4565{
4566 BlockDriver *drv = bs->drv;
469ef350 4567 int ret;
756e6736 4568
d470ad42
HR
4569 if (!drv) {
4570 return -ENOMEDIUM;
4571 }
4572
5f377794
PB
4573 /* Backing file format doesn't make sense without a backing file */
4574 if (backing_fmt && !backing_file) {
4575 return -EINVAL;
4576 }
4577
756e6736 4578 if (drv->bdrv_change_backing_file != NULL) {
469ef350 4579 ret = drv->bdrv_change_backing_file(bs, backing_file, backing_fmt);
756e6736 4580 } else {
469ef350 4581 ret = -ENOTSUP;
756e6736 4582 }
469ef350
PB
4583
4584 if (ret == 0) {
4585 pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_file ?: "");
4586 pstrcpy(bs->backing_format, sizeof(bs->backing_format), backing_fmt ?: "");
998c2019
HR
4587 pstrcpy(bs->auto_backing_file, sizeof(bs->auto_backing_file),
4588 backing_file ?: "");
469ef350
PB
4589 }
4590 return ret;
756e6736
KW
4591}
4592
6ebdcee2
JC
4593/*
4594 * Finds the image layer in the chain that has 'bs' as its backing file.
4595 *
4596 * active is the current topmost image.
4597 *
4598 * Returns NULL if bs is not found in active's image chain,
4599 * or if active == bs.
4caf0fcd
JC
4600 *
4601 * Returns the bottommost base image if bs == NULL.
6ebdcee2
JC
4602 */
4603BlockDriverState *bdrv_find_overlay(BlockDriverState *active,
4604 BlockDriverState *bs)
4605{
760e0063
KW
4606 while (active && bs != backing_bs(active)) {
4607 active = backing_bs(active);
6ebdcee2
JC
4608 }
4609
4caf0fcd
JC
4610 return active;
4611}
6ebdcee2 4612
4caf0fcd
JC
4613/* Given a BDS, searches for the base layer. */
4614BlockDriverState *bdrv_find_base(BlockDriverState *bs)
4615{
4616 return bdrv_find_overlay(bs, NULL);
6ebdcee2
JC
4617}
4618
2cad1ebe
AG
4619/*
4620 * Return true if at least one of the backing links between @bs and
4621 * @base is frozen. @errp is set if that's the case.
0f0998f6 4622 * @base must be reachable from @bs, or NULL.
2cad1ebe
AG
4623 */
4624bool bdrv_is_backing_chain_frozen(BlockDriverState *bs, BlockDriverState *base,
4625 Error **errp)
4626{
4627 BlockDriverState *i;
4628
0f0998f6
AG
4629 for (i = bs; i != base; i = backing_bs(i)) {
4630 if (i->backing && i->backing->frozen) {
2cad1ebe
AG
4631 error_setg(errp, "Cannot change '%s' link from '%s' to '%s'",
4632 i->backing->name, i->node_name,
4633 backing_bs(i)->node_name);
4634 return true;
4635 }
4636 }
4637
4638 return false;
4639}
4640
4641/*
4642 * Freeze all backing links between @bs and @base.
4643 * If any of the links is already frozen the operation is aborted and
4644 * none of the links are modified.
0f0998f6 4645 * @base must be reachable from @bs, or NULL.
2cad1ebe
AG
4646 * Returns 0 on success. On failure returns < 0 and sets @errp.
4647 */
4648int bdrv_freeze_backing_chain(BlockDriverState *bs, BlockDriverState *base,
4649 Error **errp)
4650{
4651 BlockDriverState *i;
4652
4653 if (bdrv_is_backing_chain_frozen(bs, base, errp)) {
4654 return -EPERM;
4655 }
4656
e5182c1c
HR
4657 for (i = bs; i != base; i = backing_bs(i)) {
4658 if (i->backing && backing_bs(i)->never_freeze) {
4659 error_setg(errp, "Cannot freeze '%s' link to '%s'",
4660 i->backing->name, backing_bs(i)->node_name);
4661 return -EPERM;
4662 }
4663 }
4664
0f0998f6
AG
4665 for (i = bs; i != base; i = backing_bs(i)) {
4666 if (i->backing) {
4667 i->backing->frozen = true;
4668 }
2cad1ebe
AG
4669 }
4670
4671 return 0;
4672}
4673
4674/*
4675 * Unfreeze all backing links between @bs and @base. The caller must
4676 * ensure that all links are frozen before using this function.
0f0998f6 4677 * @base must be reachable from @bs, or NULL.
2cad1ebe
AG
4678 */
4679void bdrv_unfreeze_backing_chain(BlockDriverState *bs, BlockDriverState *base)
4680{
4681 BlockDriverState *i;
4682
0f0998f6
AG
4683 for (i = bs; i != base; i = backing_bs(i)) {
4684 if (i->backing) {
4685 assert(i->backing->frozen);
4686 i->backing->frozen = false;
4687 }
2cad1ebe
AG
4688 }
4689}
4690
6ebdcee2
JC
4691/*
4692 * Drops images above 'base' up to and including 'top', and sets the image
4693 * above 'top' to have base as its backing file.
4694 *
4695 * Requires that the overlay to 'top' is opened r/w, so that the backing file
4696 * information in 'bs' can be properly updated.
4697 *
4698 * E.g., this will convert the following chain:
4699 * bottom <- base <- intermediate <- top <- active
4700 *
4701 * to
4702 *
4703 * bottom <- base <- active
4704 *
4705 * It is allowed for bottom==base, in which case it converts:
4706 *
4707 * base <- intermediate <- top <- active
4708 *
4709 * to
4710 *
4711 * base <- active
4712 *
54e26900
JC
4713 * If backing_file_str is non-NULL, it will be used when modifying top's
4714 * overlay image metadata.
4715 *
6ebdcee2
JC
4716 * Error conditions:
4717 * if active == top, that is considered an error
4718 *
4719 */
bde70715
KW
4720int bdrv_drop_intermediate(BlockDriverState *top, BlockDriverState *base,
4721 const char *backing_file_str)
6ebdcee2 4722{
6bd858b3
AG
4723 BlockDriverState *explicit_top = top;
4724 bool update_inherits_from;
61f09cea 4725 BdrvChild *c, *next;
12fa4af6 4726 Error *local_err = NULL;
6ebdcee2
JC
4727 int ret = -EIO;
4728
6858eba0 4729 bdrv_ref(top);
637d54a5 4730 bdrv_subtree_drained_begin(top);
6858eba0 4731
6ebdcee2
JC
4732 if (!top->drv || !base->drv) {
4733 goto exit;
4734 }
4735
5db15a57
KW
4736 /* Make sure that base is in the backing chain of top */
4737 if (!bdrv_chain_contains(top, base)) {
6ebdcee2
JC
4738 goto exit;
4739 }
4740
2cad1ebe
AG
4741 /* This function changes all links that point to top and makes
4742 * them point to base. Check that none of them is frozen. */
4743 QLIST_FOREACH(c, &top->parents, next_parent) {
4744 if (c->frozen) {
4745 goto exit;
4746 }
4747 }
4748
6bd858b3
AG
4749 /* If 'base' recursively inherits from 'top' then we should set
4750 * base->inherits_from to top->inherits_from after 'top' and all
4751 * other intermediate nodes have been dropped.
4752 * If 'top' is an implicit node (e.g. "commit_top") we should skip
4753 * it because no one inherits from it. We use explicit_top for that. */
4754 while (explicit_top && explicit_top->implicit) {
4755 explicit_top = backing_bs(explicit_top);
4756 }
4757 update_inherits_from = bdrv_inherits_from_recursive(base, explicit_top);
4758
6ebdcee2 4759 /* success - we can delete the intermediate states, and link top->base */
bde70715
KW
4760 /* TODO Check graph modification op blockers (BLK_PERM_GRAPH_MOD) once
4761 * we've figured out how they should work. */
f30c66ba
HR
4762 if (!backing_file_str) {
4763 bdrv_refresh_filename(base);
4764 backing_file_str = base->filename;
4765 }
61f09cea
KW
4766
4767 QLIST_FOREACH_SAFE(c, &top->parents, next_parent, next) {
4768 /* Check whether we are allowed to switch c from top to base */
4769 GSList *ignore_children = g_slist_prepend(NULL, c);
2345bde6 4770 ret = bdrv_check_update_perm(base, NULL, c->perm, c->shared_perm,
9eab1544 4771 ignore_children, NULL, &local_err);
2c860e79 4772 g_slist_free(ignore_children);
2345bde6 4773 if (ret < 0) {
61f09cea 4774 error_report_err(local_err);
6858eba0
KW
4775 goto exit;
4776 }
12fa4af6 4777
61f09cea
KW
4778 /* If so, update the backing file path in the image file */
4779 if (c->role->update_filename) {
4780 ret = c->role->update_filename(c, base, backing_file_str,
4781 &local_err);
4782 if (ret < 0) {
4783 bdrv_abort_perm_update(base);
4784 error_report_err(local_err);
4785 goto exit;
4786 }
4787 }
4788
4789 /* Do the actual switch in the in-memory graph.
4790 * Completes bdrv_check_update_perm() transaction internally. */
4791 bdrv_ref(base);
4792 bdrv_replace_child(c, base);
4793 bdrv_unref(top);
12fa4af6 4794 }
6ebdcee2 4795
6bd858b3
AG
4796 if (update_inherits_from) {
4797 base->inherits_from = explicit_top->inherits_from;
4798 }
4799
6ebdcee2 4800 ret = 0;
6ebdcee2 4801exit:
637d54a5 4802 bdrv_subtree_drained_end(top);
6858eba0 4803 bdrv_unref(top);
6ebdcee2
JC
4804 return ret;
4805}
4806
61007b31
SH
4807/**
4808 * Length of a allocated file in bytes. Sparse files are counted by actual
4809 * allocated space. Return < 0 if error or unknown.
4810 */
4811int64_t bdrv_get_allocated_file_size(BlockDriverState *bs)
71d0770c 4812{
61007b31
SH
4813 BlockDriver *drv = bs->drv;
4814 if (!drv) {
4815 return -ENOMEDIUM;
8f4754ed 4816 }
61007b31
SH
4817 if (drv->bdrv_get_allocated_file_size) {
4818 return drv->bdrv_get_allocated_file_size(bs);
4819 }
4820 if (bs->file) {
9a4f4c31 4821 return bdrv_get_allocated_file_size(bs->file->bs);
1c9805a3 4822 }
61007b31 4823 return -ENOTSUP;
1c9805a3 4824}
e7a8a783 4825
90880ff1
SH
4826/*
4827 * bdrv_measure:
4828 * @drv: Format driver
4829 * @opts: Creation options for new image
4830 * @in_bs: Existing image containing data for new image (may be NULL)
4831 * @errp: Error object
4832 * Returns: A #BlockMeasureInfo (free using qapi_free_BlockMeasureInfo())
4833 * or NULL on error
4834 *
4835 * Calculate file size required to create a new image.
4836 *
4837 * If @in_bs is given then space for allocated clusters and zero clusters
4838 * from that image are included in the calculation. If @opts contains a
4839 * backing file that is shared by @in_bs then backing clusters may be omitted
4840 * from the calculation.
4841 *
4842 * If @in_bs is NULL then the calculation includes no allocated clusters
4843 * unless a preallocation option is given in @opts.
4844 *
4845 * Note that @in_bs may use a different BlockDriver from @drv.
4846 *
4847 * If an error occurs the @errp pointer is set.
4848 */
4849BlockMeasureInfo *bdrv_measure(BlockDriver *drv, QemuOpts *opts,
4850 BlockDriverState *in_bs, Error **errp)
4851{
4852 if (!drv->bdrv_measure) {
4853 error_setg(errp, "Block driver '%s' does not support size measurement",
4854 drv->format_name);
4855 return NULL;
4856 }
4857
4858 return drv->bdrv_measure(opts, in_bs, errp);
4859}
4860
61007b31
SH
4861/**
4862 * Return number of sectors on success, -errno on error.
1c9805a3 4863 */
61007b31 4864int64_t bdrv_nb_sectors(BlockDriverState *bs)
1c9805a3 4865{
61007b31 4866 BlockDriver *drv = bs->drv;
498e386c 4867
61007b31
SH
4868 if (!drv)
4869 return -ENOMEDIUM;
2572b37a 4870
61007b31
SH
4871 if (drv->has_variable_length) {
4872 int ret = refresh_total_sectors(bs, bs->total_sectors);
4873 if (ret < 0) {
4874 return ret;
1c9805a3
SH
4875 }
4876 }
61007b31 4877 return bs->total_sectors;
1c9805a3 4878}
b338082b 4879
61007b31
SH
4880/**
4881 * Return length in bytes on success, -errno on error.
4882 * The length is always a multiple of BDRV_SECTOR_SIZE.
8d3b1a2d 4883 */
61007b31 4884int64_t bdrv_getlength(BlockDriverState *bs)
8d3b1a2d 4885{
61007b31 4886 int64_t ret = bdrv_nb_sectors(bs);
8d3b1a2d 4887
4a9c9ea0 4888 ret = ret > INT64_MAX / BDRV_SECTOR_SIZE ? -EFBIG : ret;
61007b31 4889 return ret < 0 ? ret : ret * BDRV_SECTOR_SIZE;
fc01f7e7
FB
4890}
4891
61007b31
SH
4892/* return 0 as number of sectors if no device present or error */
4893void bdrv_get_geometry(BlockDriverState *bs, uint64_t *nb_sectors_ptr)
07d27a44 4894{
61007b31 4895 int64_t nb_sectors = bdrv_nb_sectors(bs);
07d27a44 4896
61007b31 4897 *nb_sectors_ptr = nb_sectors < 0 ? 0 : nb_sectors;
07d27a44
MA
4898}
4899
54115412 4900bool bdrv_is_sg(BlockDriverState *bs)
f08145fe 4901{
61007b31 4902 return bs->sg;
f08145fe
KW
4903}
4904
54115412 4905bool bdrv_is_encrypted(BlockDriverState *bs)
fc3959e4 4906{
760e0063 4907 if (bs->backing && bs->backing->bs->encrypted) {
54115412 4908 return true;
760e0063 4909 }
61007b31 4910 return bs->encrypted;
fc3959e4
FZ
4911}
4912
61007b31 4913const char *bdrv_get_format_name(BlockDriverState *bs)
40b4f539 4914{
61007b31 4915 return bs->drv ? bs->drv->format_name : NULL;
40b4f539
KW
4916}
4917
61007b31 4918static int qsort_strcmp(const void *a, const void *b)
40b4f539 4919{
ceff5bd7 4920 return strcmp(*(char *const *)a, *(char *const *)b);
40b4f539
KW
4921}
4922
61007b31 4923void bdrv_iterate_format(void (*it)(void *opaque, const char *name),
9ac404c5 4924 void *opaque, bool read_only)
40b4f539 4925{
61007b31
SH
4926 BlockDriver *drv;
4927 int count = 0;
4928 int i;
4929 const char **formats = NULL;
40b4f539 4930
61007b31
SH
4931 QLIST_FOREACH(drv, &bdrv_drivers, list) {
4932 if (drv->format_name) {
4933 bool found = false;
4934 int i = count;
9ac404c5
AS
4935
4936 if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv, read_only)) {
4937 continue;
4938 }
4939
61007b31
SH
4940 while (formats && i && !found) {
4941 found = !strcmp(formats[--i], drv->format_name);
4942 }
e2a305fb 4943
61007b31
SH
4944 if (!found) {
4945 formats = g_renew(const char *, formats, count + 1);
4946 formats[count++] = drv->format_name;
4947 }
6c5a42ac 4948 }
61007b31 4949 }
6c5a42ac 4950
eb0df69f
HR
4951 for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); i++) {
4952 const char *format_name = block_driver_modules[i].format_name;
4953
4954 if (format_name) {
4955 bool found = false;
4956 int j = count;
4957
9ac404c5
AS
4958 if (use_bdrv_whitelist &&
4959 !bdrv_format_is_whitelisted(format_name, read_only)) {
4960 continue;
4961 }
4962
eb0df69f
HR
4963 while (formats && j && !found) {
4964 found = !strcmp(formats[--j], format_name);
4965 }
4966
4967 if (!found) {
4968 formats = g_renew(const char *, formats, count + 1);
4969 formats[count++] = format_name;
4970 }
4971 }
4972 }
4973
61007b31 4974 qsort(formats, count, sizeof(formats[0]), qsort_strcmp);
40b4f539 4975
61007b31
SH
4976 for (i = 0; i < count; i++) {
4977 it(opaque, formats[i]);
4978 }
40b4f539 4979
61007b31
SH
4980 g_free(formats);
4981}
40b4f539 4982
61007b31
SH
4983/* This function is to find a node in the bs graph */
4984BlockDriverState *bdrv_find_node(const char *node_name)
4985{
4986 BlockDriverState *bs;
391827eb 4987
61007b31 4988 assert(node_name);
40b4f539 4989
61007b31
SH
4990 QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
4991 if (!strcmp(node_name, bs->node_name)) {
4992 return bs;
40b4f539
KW
4993 }
4994 }
61007b31 4995 return NULL;
40b4f539
KW
4996}
4997
61007b31 4998/* Put this QMP function here so it can access the static graph_bdrv_states. */
facda544
PK
4999BlockDeviceInfoList *bdrv_named_nodes_list(bool flat,
5000 Error **errp)
40b4f539 5001{
61007b31
SH
5002 BlockDeviceInfoList *list, *entry;
5003 BlockDriverState *bs;
40b4f539 5004
61007b31
SH
5005 list = NULL;
5006 QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
facda544 5007 BlockDeviceInfo *info = bdrv_block_device_info(NULL, bs, flat, errp);
61007b31
SH
5008 if (!info) {
5009 qapi_free_BlockDeviceInfoList(list);
5010 return NULL;
301db7c2 5011 }
61007b31
SH
5012 entry = g_malloc0(sizeof(*entry));
5013 entry->value = info;
5014 entry->next = list;
5015 list = entry;
301db7c2
RH
5016 }
5017
61007b31
SH
5018 return list;
5019}
40b4f539 5020
5d3b4e99
VSO
5021#define QAPI_LIST_ADD(list, element) do { \
5022 typeof(list) _tmp = g_new(typeof(*(list)), 1); \
5023 _tmp->value = (element); \
5024 _tmp->next = (list); \
5025 (list) = _tmp; \
5026} while (0)
5027
5028typedef struct XDbgBlockGraphConstructor {
5029 XDbgBlockGraph *graph;
5030 GHashTable *graph_nodes;
5031} XDbgBlockGraphConstructor;
5032
5033static XDbgBlockGraphConstructor *xdbg_graph_new(void)
5034{
5035 XDbgBlockGraphConstructor *gr = g_new(XDbgBlockGraphConstructor, 1);
5036
5037 gr->graph = g_new0(XDbgBlockGraph, 1);
5038 gr->graph_nodes = g_hash_table_new(NULL, NULL);
5039
5040 return gr;
5041}
5042
5043static XDbgBlockGraph *xdbg_graph_finalize(XDbgBlockGraphConstructor *gr)
5044{
5045 XDbgBlockGraph *graph = gr->graph;
5046
5047 g_hash_table_destroy(gr->graph_nodes);
5048 g_free(gr);
5049
5050 return graph;
5051}
5052
5053static uintptr_t xdbg_graph_node_num(XDbgBlockGraphConstructor *gr, void *node)
5054{
5055 uintptr_t ret = (uintptr_t)g_hash_table_lookup(gr->graph_nodes, node);
5056
5057 if (ret != 0) {
5058 return ret;
5059 }
5060
5061 /*
5062 * Start counting from 1, not 0, because 0 interferes with not-found (NULL)
5063 * answer of g_hash_table_lookup.
5064 */
5065 ret = g_hash_table_size(gr->graph_nodes) + 1;
5066 g_hash_table_insert(gr->graph_nodes, node, (void *)ret);
5067
5068 return ret;
5069}
5070
5071static void xdbg_graph_add_node(XDbgBlockGraphConstructor *gr, void *node,
5072 XDbgBlockGraphNodeType type, const char *name)
5073{
5074 XDbgBlockGraphNode *n;
5075
5076 n = g_new0(XDbgBlockGraphNode, 1);
5077
5078 n->id = xdbg_graph_node_num(gr, node);
5079 n->type = type;
5080 n->name = g_strdup(name);
5081
5082 QAPI_LIST_ADD(gr->graph->nodes, n);
5083}
5084
5085static void xdbg_graph_add_edge(XDbgBlockGraphConstructor *gr, void *parent,
5086 const BdrvChild *child)
5087{
cdb1cec8 5088 BlockPermission qapi_perm;
5d3b4e99
VSO
5089 XDbgBlockGraphEdge *edge;
5090
5d3b4e99
VSO
5091 edge = g_new0(XDbgBlockGraphEdge, 1);
5092
5093 edge->parent = xdbg_graph_node_num(gr, parent);
5094 edge->child = xdbg_graph_node_num(gr, child->bs);
5095 edge->name = g_strdup(child->name);
5096
cdb1cec8
HR
5097 for (qapi_perm = 0; qapi_perm < BLOCK_PERMISSION__MAX; qapi_perm++) {
5098 uint64_t flag = bdrv_qapi_perm_to_blk_perm(qapi_perm);
5099
5100 if (flag & child->perm) {
5101 QAPI_LIST_ADD(edge->perm, qapi_perm);
5d3b4e99 5102 }
cdb1cec8
HR
5103 if (flag & child->shared_perm) {
5104 QAPI_LIST_ADD(edge->shared_perm, qapi_perm);
5d3b4e99
VSO
5105 }
5106 }
5107
5108 QAPI_LIST_ADD(gr->graph->edges, edge);
5109}
5110
5111
5112XDbgBlockGraph *bdrv_get_xdbg_block_graph(Error **errp)
5113{
5114 BlockBackend *blk;
5115 BlockJob *job;
5116 BlockDriverState *bs;
5117 BdrvChild *child;
5118 XDbgBlockGraphConstructor *gr = xdbg_graph_new();
5119
5120 for (blk = blk_all_next(NULL); blk; blk = blk_all_next(blk)) {
5121 char *allocated_name = NULL;
5122 const char *name = blk_name(blk);
5123
5124 if (!*name) {
5125 name = allocated_name = blk_get_attached_dev_id(blk);
5126 }
5127 xdbg_graph_add_node(gr, blk, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_BACKEND,
5128 name);
5129 g_free(allocated_name);
5130 if (blk_root(blk)) {
5131 xdbg_graph_add_edge(gr, blk, blk_root(blk));
5132 }
5133 }
5134
5135 for (job = block_job_next(NULL); job; job = block_job_next(job)) {
5136 GSList *el;
5137
5138 xdbg_graph_add_node(gr, job, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_JOB,
5139 job->job.id);
5140 for (el = job->nodes; el; el = el->next) {
5141 xdbg_graph_add_edge(gr, job, (BdrvChild *)el->data);
5142 }
5143 }
5144
5145 QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
5146 xdbg_graph_add_node(gr, bs, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_DRIVER,
5147 bs->node_name);
5148 QLIST_FOREACH(child, &bs->children, next) {
5149 xdbg_graph_add_edge(gr, bs, child);
5150 }
5151 }
5152
5153 return xdbg_graph_finalize(gr);
5154}
5155
61007b31
SH
5156BlockDriverState *bdrv_lookup_bs(const char *device,
5157 const char *node_name,
5158 Error **errp)
5159{
5160 BlockBackend *blk;
5161 BlockDriverState *bs;
40b4f539 5162
61007b31
SH
5163 if (device) {
5164 blk = blk_by_name(device);
40b4f539 5165
61007b31 5166 if (blk) {
9f4ed6fb
AG
5167 bs = blk_bs(blk);
5168 if (!bs) {
5433c24f 5169 error_setg(errp, "Device '%s' has no medium", device);
5433c24f
HR
5170 }
5171
9f4ed6fb 5172 return bs;
61007b31
SH
5173 }
5174 }
40b4f539 5175
61007b31
SH
5176 if (node_name) {
5177 bs = bdrv_find_node(node_name);
6d519a5f 5178
61007b31
SH
5179 if (bs) {
5180 return bs;
5181 }
40b4f539
KW
5182 }
5183
61007b31
SH
5184 error_setg(errp, "Cannot find device=%s nor node_name=%s",
5185 device ? device : "",
5186 node_name ? node_name : "");
5187 return NULL;
40b4f539
KW
5188}
5189
61007b31
SH
5190/* If 'base' is in the same chain as 'top', return true. Otherwise,
5191 * return false. If either argument is NULL, return false. */
5192bool bdrv_chain_contains(BlockDriverState *top, BlockDriverState *base)
83f64091 5193{
61007b31 5194 while (top && top != base) {
760e0063 5195 top = backing_bs(top);
02c50efe 5196 }
61007b31
SH
5197
5198 return top != NULL;
02c50efe
FZ
5199}
5200
61007b31 5201BlockDriverState *bdrv_next_node(BlockDriverState *bs)
02c50efe 5202{
61007b31
SH
5203 if (!bs) {
5204 return QTAILQ_FIRST(&graph_bdrv_states);
02c50efe 5205 }
61007b31 5206 return QTAILQ_NEXT(bs, node_list);
83f64091
FB
5207}
5208
0f12264e
KW
5209BlockDriverState *bdrv_next_all_states(BlockDriverState *bs)
5210{
5211 if (!bs) {
5212 return QTAILQ_FIRST(&all_bdrv_states);
5213 }
5214 return QTAILQ_NEXT(bs, bs_list);
5215}
5216
61007b31 5217const char *bdrv_get_node_name(const BlockDriverState *bs)
83f64091 5218{
61007b31 5219 return bs->node_name;
beac80cd
FB
5220}
5221
1f0c461b 5222const char *bdrv_get_parent_name(const BlockDriverState *bs)
4c265bf9
KW
5223{
5224 BdrvChild *c;
5225 const char *name;
5226
5227 /* If multiple parents have a name, just pick the first one. */
5228 QLIST_FOREACH(c, &bs->parents, next_parent) {
5229 if (c->role->get_name) {
5230 name = c->role->get_name(c);
5231 if (name && *name) {
5232 return name;
5233 }
5234 }
5235 }
5236
5237 return NULL;
5238}
5239
61007b31
SH
5240/* TODO check what callers really want: bs->node_name or blk_name() */
5241const char *bdrv_get_device_name(const BlockDriverState *bs)
beac80cd 5242{
4c265bf9 5243 return bdrv_get_parent_name(bs) ?: "";
f141eafe 5244}
83f64091 5245
61007b31
SH
5246/* This can be used to identify nodes that might not have a device
5247 * name associated. Since node and device names live in the same
5248 * namespace, the result is unambiguous. The exception is if both are
5249 * absent, then this returns an empty (non-null) string. */
5250const char *bdrv_get_device_or_node_name(const BlockDriverState *bs)
f141eafe 5251{
4c265bf9 5252 return bdrv_get_parent_name(bs) ?: bs->node_name;
beac80cd 5253}
beac80cd 5254
61007b31 5255int bdrv_get_flags(BlockDriverState *bs)
0b5a2445 5256{
61007b31 5257 return bs->open_flags;
0b5a2445
PB
5258}
5259
61007b31 5260int bdrv_has_zero_init_1(BlockDriverState *bs)
68485420 5261{
61007b31 5262 return 1;
0b5a2445
PB
5263}
5264
61007b31 5265int bdrv_has_zero_init(BlockDriverState *bs)
0b5a2445 5266{
d470ad42
HR
5267 if (!bs->drv) {
5268 return 0;
5269 }
0b5a2445 5270
61007b31
SH
5271 /* If BS is a copy on write image, it is initialized to
5272 the contents of the base image, which may not be zeroes. */
760e0063 5273 if (bs->backing) {
61007b31
SH
5274 return 0;
5275 }
5276 if (bs->drv->bdrv_has_zero_init) {
5277 return bs->drv->bdrv_has_zero_init(bs);
0b5a2445 5278 }
5a612c00
MP
5279 if (bs->file && bs->drv->is_filter) {
5280 return bdrv_has_zero_init(bs->file->bs);
5281 }
61007b31
SH
5282
5283 /* safe default */
5284 return 0;
68485420
KW
5285}
5286
ceaca56f
HR
5287int bdrv_has_zero_init_truncate(BlockDriverState *bs)
5288{
5289 if (!bs->drv) {
5290 return 0;
5291 }
5292
5293 if (bs->backing) {
5294 /* Depends on the backing image length, but better safe than sorry */
5295 return 0;
5296 }
5297 if (bs->drv->bdrv_has_zero_init_truncate) {
5298 return bs->drv->bdrv_has_zero_init_truncate(bs);
5299 }
5300 if (bs->file && bs->drv->is_filter) {
5301 return bdrv_has_zero_init_truncate(bs->file->bs);
5302 }
5303
5304 /* safe default */
5305 return 0;
5306}
5307
61007b31 5308bool bdrv_unallocated_blocks_are_zero(BlockDriverState *bs)
b2a61371 5309{
61007b31 5310 BlockDriverInfo bdi;
b2a61371 5311
760e0063 5312 if (bs->backing) {
61007b31
SH
5313 return false;
5314 }
5315
5316 if (bdrv_get_info(bs, &bdi) == 0) {
5317 return bdi.unallocated_blocks_are_zero;
b2a61371
SH
5318 }
5319
61007b31 5320 return false;
b2a61371
SH
5321}
5322
61007b31 5323bool bdrv_can_write_zeroes_with_unmap(BlockDriverState *bs)
68485420 5324{
2f0342ef 5325 if (!(bs->open_flags & BDRV_O_UNMAP)) {
61007b31
SH
5326 return false;
5327 }
68485420 5328
e24d813b 5329 return bs->supported_zero_flags & BDRV_REQ_MAY_UNMAP;
68485420
KW
5330}
5331
61007b31
SH
5332void bdrv_get_backing_filename(BlockDriverState *bs,
5333 char *filename, int filename_size)
016f5cf6 5334{
61007b31
SH
5335 pstrcpy(filename, filename_size, bs->backing_file);
5336}
d318aea9 5337
61007b31
SH
5338int bdrv_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
5339{
5340 BlockDriver *drv = bs->drv;
5a612c00
MP
5341 /* if bs->drv == NULL, bs is closed, so there's nothing to do here */
5342 if (!drv) {
61007b31 5343 return -ENOMEDIUM;
5a612c00
MP
5344 }
5345 if (!drv->bdrv_get_info) {
5346 if (bs->file && drv->is_filter) {
5347 return bdrv_get_info(bs->file->bs, bdi);
5348 }
61007b31 5349 return -ENOTSUP;
5a612c00 5350 }
61007b31
SH
5351 memset(bdi, 0, sizeof(*bdi));
5352 return drv->bdrv_get_info(bs, bdi);
5353}
016f5cf6 5354
1bf6e9ca
AS
5355ImageInfoSpecific *bdrv_get_specific_info(BlockDriverState *bs,
5356 Error **errp)
61007b31
SH
5357{
5358 BlockDriver *drv = bs->drv;
5359 if (drv && drv->bdrv_get_specific_info) {
1bf6e9ca 5360 return drv->bdrv_get_specific_info(bs, errp);
61007b31
SH
5361 }
5362 return NULL;
016f5cf6
AG
5363}
5364
d9245599
AN
5365BlockStatsSpecific *bdrv_get_specific_stats(BlockDriverState *bs)
5366{
5367 BlockDriver *drv = bs->drv;
5368 if (!drv || !drv->bdrv_get_specific_stats) {
5369 return NULL;
5370 }
5371 return drv->bdrv_get_specific_stats(bs);
5372}
5373
a31939e6 5374void bdrv_debug_event(BlockDriverState *bs, BlkdebugEvent event)
4265d620 5375{
61007b31
SH
5376 if (!bs || !bs->drv || !bs->drv->bdrv_debug_event) {
5377 return;
5378 }
4265d620 5379
61007b31 5380 bs->drv->bdrv_debug_event(bs, event);
4265d620
PB
5381}
5382
d10529a2 5383static BlockDriverState *bdrv_find_debug_node(BlockDriverState *bs)
4265d620 5384{
61007b31 5385 while (bs && bs->drv && !bs->drv->bdrv_debug_breakpoint) {
d10529a2
VSO
5386 if (bs->file) {
5387 bs = bs->file->bs;
5388 continue;
5389 }
5390
5391 if (bs->drv->is_filter && bs->backing) {
5392 bs = bs->backing->bs;
5393 continue;
5394 }
5395
5396 break;
61007b31 5397 }
4265d620 5398
61007b31 5399 if (bs && bs->drv && bs->drv->bdrv_debug_breakpoint) {
d10529a2
VSO
5400 assert(bs->drv->bdrv_debug_remove_breakpoint);
5401 return bs;
5402 }
5403
5404 return NULL;
5405}
5406
5407int bdrv_debug_breakpoint(BlockDriverState *bs, const char *event,
5408 const char *tag)
5409{
5410 bs = bdrv_find_debug_node(bs);
5411 if (bs) {
61007b31
SH
5412 return bs->drv->bdrv_debug_breakpoint(bs, event, tag);
5413 }
4265d620 5414
61007b31 5415 return -ENOTSUP;
4265d620
PB
5416}
5417
61007b31 5418int bdrv_debug_remove_breakpoint(BlockDriverState *bs, const char *tag)
ea2384d3 5419{
d10529a2
VSO
5420 bs = bdrv_find_debug_node(bs);
5421 if (bs) {
61007b31
SH
5422 return bs->drv->bdrv_debug_remove_breakpoint(bs, tag);
5423 }
5424
5425 return -ENOTSUP;
eb852011
MA
5426}
5427
61007b31 5428int bdrv_debug_resume(BlockDriverState *bs, const char *tag)
ce1a14dc 5429{
61007b31 5430 while (bs && (!bs->drv || !bs->drv->bdrv_debug_resume)) {
9a4f4c31 5431 bs = bs->file ? bs->file->bs : NULL;
61007b31 5432 }
ce1a14dc 5433
61007b31
SH
5434 if (bs && bs->drv && bs->drv->bdrv_debug_resume) {
5435 return bs->drv->bdrv_debug_resume(bs, tag);
5436 }
ce1a14dc 5437
61007b31 5438 return -ENOTSUP;
f197fe2b
FZ
5439}
5440
61007b31 5441bool bdrv_debug_is_suspended(BlockDriverState *bs, const char *tag)
ce1a14dc 5442{
61007b31 5443 while (bs && bs->drv && !bs->drv->bdrv_debug_is_suspended) {
9a4f4c31 5444 bs = bs->file ? bs->file->bs : NULL;
f197fe2b 5445 }
19cb3738 5446
61007b31
SH
5447 if (bs && bs->drv && bs->drv->bdrv_debug_is_suspended) {
5448 return bs->drv->bdrv_debug_is_suspended(bs, tag);
5449 }
f9f05dc5 5450
61007b31
SH
5451 return false;
5452}
f9f05dc5 5453
61007b31
SH
5454/* backing_file can either be relative, or absolute, or a protocol. If it is
5455 * relative, it must be relative to the chain. So, passing in bs->filename
5456 * from a BDS as backing_file should not be done, as that may be relative to
5457 * the CWD rather than the chain. */
5458BlockDriverState *bdrv_find_backing_image(BlockDriverState *bs,
5459 const char *backing_file)
f9f05dc5 5460{
61007b31
SH
5461 char *filename_full = NULL;
5462 char *backing_file_full = NULL;
5463 char *filename_tmp = NULL;
5464 int is_protocol = 0;
5465 BlockDriverState *curr_bs = NULL;
5466 BlockDriverState *retval = NULL;
f9f05dc5 5467
61007b31
SH
5468 if (!bs || !bs->drv || !backing_file) {
5469 return NULL;
f9f05dc5
KW
5470 }
5471
61007b31
SH
5472 filename_full = g_malloc(PATH_MAX);
5473 backing_file_full = g_malloc(PATH_MAX);
f9f05dc5 5474
61007b31 5475 is_protocol = path_has_protocol(backing_file);
f9f05dc5 5476
760e0063 5477 for (curr_bs = bs; curr_bs->backing; curr_bs = curr_bs->backing->bs) {
f9f05dc5 5478
61007b31
SH
5479 /* If either of the filename paths is actually a protocol, then
5480 * compare unmodified paths; otherwise make paths relative */
5481 if (is_protocol || path_has_protocol(curr_bs->backing_file)) {
6b6833c1
HR
5482 char *backing_file_full_ret;
5483
61007b31 5484 if (strcmp(backing_file, curr_bs->backing_file) == 0) {
760e0063 5485 retval = curr_bs->backing->bs;
61007b31
SH
5486 break;
5487 }
418661e0 5488 /* Also check against the full backing filename for the image */
6b6833c1
HR
5489 backing_file_full_ret = bdrv_get_full_backing_filename(curr_bs,
5490 NULL);
5491 if (backing_file_full_ret) {
5492 bool equal = strcmp(backing_file, backing_file_full_ret) == 0;
5493 g_free(backing_file_full_ret);
5494 if (equal) {
418661e0
JC
5495 retval = curr_bs->backing->bs;
5496 break;
5497 }
418661e0 5498 }
61007b31
SH
5499 } else {
5500 /* If not an absolute filename path, make it relative to the current
5501 * image's filename path */
2d9158ce
HR
5502 filename_tmp = bdrv_make_absolute_filename(curr_bs, backing_file,
5503 NULL);
5504 /* We are going to compare canonicalized absolute pathnames */
5505 if (!filename_tmp || !realpath(filename_tmp, filename_full)) {
5506 g_free(filename_tmp);
61007b31
SH
5507 continue;
5508 }
2d9158ce 5509 g_free(filename_tmp);
07f07615 5510
61007b31
SH
5511 /* We need to make sure the backing filename we are comparing against
5512 * is relative to the current image filename (or absolute) */
2d9158ce
HR
5513 filename_tmp = bdrv_get_full_backing_filename(curr_bs, NULL);
5514 if (!filename_tmp || !realpath(filename_tmp, backing_file_full)) {
5515 g_free(filename_tmp);
61007b31
SH
5516 continue;
5517 }
2d9158ce 5518 g_free(filename_tmp);
eb489bb1 5519
61007b31 5520 if (strcmp(backing_file_full, filename_full) == 0) {
760e0063 5521 retval = curr_bs->backing->bs;
61007b31
SH
5522 break;
5523 }
5524 }
eb489bb1
KW
5525 }
5526
61007b31
SH
5527 g_free(filename_full);
5528 g_free(backing_file_full);
61007b31
SH
5529 return retval;
5530}
5531
61007b31
SH
5532void bdrv_init(void)
5533{
5534 module_call_init(MODULE_INIT_BLOCK);
5535}
29cdb251 5536
61007b31
SH
5537void bdrv_init_with_whitelist(void)
5538{
5539 use_bdrv_whitelist = 1;
5540 bdrv_init();
07f07615
PB
5541}
5542
2b148f39
PB
5543static void coroutine_fn bdrv_co_invalidate_cache(BlockDriverState *bs,
5544 Error **errp)
0f15423c 5545{
4417ab7a 5546 BdrvChild *child, *parent;
9c5e6594 5547 uint64_t perm, shared_perm;
5a8a30db
KW
5548 Error *local_err = NULL;
5549 int ret;
9c98f145 5550 BdrvDirtyBitmap *bm;
5a8a30db 5551
3456a8d1
KW
5552 if (!bs->drv) {
5553 return;
5554 }
5555
16e977d5 5556 QLIST_FOREACH(child, &bs->children, next) {
2b148f39 5557 bdrv_co_invalidate_cache(child->bs, &local_err);
0d1c5c91 5558 if (local_err) {
0d1c5c91
FZ
5559 error_propagate(errp, local_err);
5560 return;
5561 }
5a8a30db 5562 }
0d1c5c91 5563
dafe0960
KW
5564 /*
5565 * Update permissions, they may differ for inactive nodes.
5566 *
5567 * Note that the required permissions of inactive images are always a
5568 * subset of the permissions required after activating the image. This
5569 * allows us to just get the permissions upfront without restricting
5570 * drv->bdrv_invalidate_cache().
5571 *
5572 * It also means that in error cases, we don't have to try and revert to
5573 * the old permissions (which is an operation that could fail, too). We can
5574 * just keep the extended permissions for the next time that an activation
5575 * of the image is tried.
5576 */
7bb4941a
KW
5577 if (bs->open_flags & BDRV_O_INACTIVE) {
5578 bs->open_flags &= ~BDRV_O_INACTIVE;
5579 bdrv_get_cumulative_perm(bs, &perm, &shared_perm);
5580 ret = bdrv_check_perm(bs, NULL, perm, shared_perm, NULL, NULL, &local_err);
5581 if (ret < 0) {
0d1c5c91
FZ
5582 bs->open_flags |= BDRV_O_INACTIVE;
5583 error_propagate(errp, local_err);
5584 return;
5585 }
7bb4941a 5586 bdrv_set_perm(bs, perm, shared_perm);
3456a8d1 5587
7bb4941a
KW
5588 if (bs->drv->bdrv_co_invalidate_cache) {
5589 bs->drv->bdrv_co_invalidate_cache(bs, &local_err);
5590 if (local_err) {
5591 bs->open_flags |= BDRV_O_INACTIVE;
5592 error_propagate(errp, local_err);
5593 return;
5594 }
5595 }
9c98f145 5596
7bb4941a
KW
5597 FOR_EACH_DIRTY_BITMAP(bs, bm) {
5598 bdrv_dirty_bitmap_skip_store(bm, false);
5599 }
5600
5601 ret = refresh_total_sectors(bs, bs->total_sectors);
5602 if (ret < 0) {
5603 bs->open_flags |= BDRV_O_INACTIVE;
5604 error_setg_errno(errp, -ret, "Could not refresh total sector count");
5605 return;
5606 }
5a8a30db 5607 }
4417ab7a
KW
5608
5609 QLIST_FOREACH(parent, &bs->parents, next_parent) {
5610 if (parent->role->activate) {
5611 parent->role->activate(parent, &local_err);
5612 if (local_err) {
78fc3b3a 5613 bs->open_flags |= BDRV_O_INACTIVE;
4417ab7a
KW
5614 error_propagate(errp, local_err);
5615 return;
5616 }
5617 }
5618 }
0f15423c
AL
5619}
5620
2b148f39
PB
5621typedef struct InvalidateCacheCo {
5622 BlockDriverState *bs;
5623 Error **errp;
5624 bool done;
5625} InvalidateCacheCo;
5626
5627static void coroutine_fn bdrv_invalidate_cache_co_entry(void *opaque)
5628{
5629 InvalidateCacheCo *ico = opaque;
5630 bdrv_co_invalidate_cache(ico->bs, ico->errp);
5631 ico->done = true;
4720cbee 5632 aio_wait_kick();
2b148f39
PB
5633}
5634
5635void bdrv_invalidate_cache(BlockDriverState *bs, Error **errp)
5636{
5637 Coroutine *co;
5638 InvalidateCacheCo ico = {
5639 .bs = bs,
5640 .done = false,
5641 .errp = errp
5642 };
5643
5644 if (qemu_in_coroutine()) {
5645 /* Fast-path if already in coroutine context */
5646 bdrv_invalidate_cache_co_entry(&ico);
5647 } else {
5648 co = qemu_coroutine_create(bdrv_invalidate_cache_co_entry, &ico);
4720cbee 5649 bdrv_coroutine_enter(bs, co);
2b148f39
PB
5650 BDRV_POLL_WHILE(bs, !ico.done);
5651 }
5652}
5653
5a8a30db 5654void bdrv_invalidate_cache_all(Error **errp)
0f15423c 5655{
7c8eece4 5656 BlockDriverState *bs;
5a8a30db 5657 Error *local_err = NULL;
88be7b4b 5658 BdrvNextIterator it;
0f15423c 5659
88be7b4b 5660 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
ed78cda3
SH
5661 AioContext *aio_context = bdrv_get_aio_context(bs);
5662
5663 aio_context_acquire(aio_context);
5a8a30db 5664 bdrv_invalidate_cache(bs, &local_err);
ed78cda3 5665 aio_context_release(aio_context);
5a8a30db
KW
5666 if (local_err) {
5667 error_propagate(errp, local_err);
5e003f17 5668 bdrv_next_cleanup(&it);
5a8a30db
KW
5669 return;
5670 }
0f15423c
AL
5671 }
5672}
5673
9e37271f
KW
5674static bool bdrv_has_bds_parent(BlockDriverState *bs, bool only_active)
5675{
5676 BdrvChild *parent;
5677
5678 QLIST_FOREACH(parent, &bs->parents, next_parent) {
5679 if (parent->role->parent_is_bds) {
5680 BlockDriverState *parent_bs = parent->opaque;
5681 if (!only_active || !(parent_bs->open_flags & BDRV_O_INACTIVE)) {
5682 return true;
5683 }
5684 }
5685 }
5686
5687 return false;
5688}
5689
5690static int bdrv_inactivate_recurse(BlockDriverState *bs)
76b1c7fe 5691{
cfa1a572 5692 BdrvChild *child, *parent;
1046779e 5693 bool tighten_restrictions;
9e37271f 5694 uint64_t perm, shared_perm;
76b1c7fe
KW
5695 int ret;
5696
d470ad42
HR
5697 if (!bs->drv) {
5698 return -ENOMEDIUM;
5699 }
5700
9e37271f
KW
5701 /* Make sure that we don't inactivate a child before its parent.
5702 * It will be covered by recursion from the yet active parent. */
5703 if (bdrv_has_bds_parent(bs, true)) {
5704 return 0;
5705 }
5706
5707 assert(!(bs->open_flags & BDRV_O_INACTIVE));
5708
5709 /* Inactivate this node */
5710 if (bs->drv->bdrv_inactivate) {
76b1c7fe
KW
5711 ret = bs->drv->bdrv_inactivate(bs);
5712 if (ret < 0) {
5713 return ret;
5714 }
5715 }
5716
9e37271f
KW
5717 QLIST_FOREACH(parent, &bs->parents, next_parent) {
5718 if (parent->role->inactivate) {
5719 ret = parent->role->inactivate(parent);
5720 if (ret < 0) {
5721 return ret;
cfa1a572
KW
5722 }
5723 }
9e37271f 5724 }
9c5e6594 5725
9e37271f 5726 bs->open_flags |= BDRV_O_INACTIVE;
7d5b5261 5727
9e37271f
KW
5728 /* Update permissions, they may differ for inactive nodes */
5729 bdrv_get_cumulative_perm(bs, &perm, &shared_perm);
1046779e
HR
5730 ret = bdrv_check_perm(bs, NULL, perm, shared_perm, NULL,
5731 &tighten_restrictions, NULL);
5732 assert(tighten_restrictions == false);
5733 if (ret < 0) {
5734 /* We only tried to loosen restrictions, so errors are not fatal */
5735 bdrv_abort_perm_update(bs);
5736 } else {
5737 bdrv_set_perm(bs, perm, shared_perm);
5738 }
38701b6a 5739
9e37271f
KW
5740
5741 /* Recursively inactivate children */
38701b6a 5742 QLIST_FOREACH(child, &bs->children, next) {
9e37271f 5743 ret = bdrv_inactivate_recurse(child->bs);
38701b6a
KW
5744 if (ret < 0) {
5745 return ret;
5746 }
5747 }
5748
76b1c7fe
KW
5749 return 0;
5750}
5751
5752int bdrv_inactivate_all(void)
5753{
79720af6 5754 BlockDriverState *bs = NULL;
88be7b4b 5755 BdrvNextIterator it;
aad0b7a0 5756 int ret = 0;
bd6458e4 5757 GSList *aio_ctxs = NULL, *ctx;
76b1c7fe 5758
88be7b4b 5759 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
bd6458e4
PB
5760 AioContext *aio_context = bdrv_get_aio_context(bs);
5761
5762 if (!g_slist_find(aio_ctxs, aio_context)) {
5763 aio_ctxs = g_slist_prepend(aio_ctxs, aio_context);
5764 aio_context_acquire(aio_context);
5765 }
aad0b7a0 5766 }
76b1c7fe 5767
9e37271f
KW
5768 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
5769 /* Nodes with BDS parents are covered by recursion from the last
5770 * parent that gets inactivated. Don't inactivate them a second
5771 * time if that has already happened. */
5772 if (bdrv_has_bds_parent(bs, false)) {
5773 continue;
5774 }
5775 ret = bdrv_inactivate_recurse(bs);
5776 if (ret < 0) {
5777 bdrv_next_cleanup(&it);
5778 goto out;
76b1c7fe
KW
5779 }
5780 }
5781
aad0b7a0 5782out:
bd6458e4
PB
5783 for (ctx = aio_ctxs; ctx != NULL; ctx = ctx->next) {
5784 AioContext *aio_context = ctx->data;
5785 aio_context_release(aio_context);
aad0b7a0 5786 }
bd6458e4 5787 g_slist_free(aio_ctxs);
aad0b7a0
FZ
5788
5789 return ret;
76b1c7fe
KW
5790}
5791
19cb3738
FB
5792/**************************************************************/
5793/* removable device support */
5794
5795/**
5796 * Return TRUE if the media is present
5797 */
e031f750 5798bool bdrv_is_inserted(BlockDriverState *bs)
19cb3738
FB
5799{
5800 BlockDriver *drv = bs->drv;
28d7a789 5801 BdrvChild *child;
a1aff5bf 5802
e031f750
HR
5803 if (!drv) {
5804 return false;
5805 }
28d7a789
HR
5806 if (drv->bdrv_is_inserted) {
5807 return drv->bdrv_is_inserted(bs);
5808 }
5809 QLIST_FOREACH(child, &bs->children, next) {
5810 if (!bdrv_is_inserted(child->bs)) {
5811 return false;
5812 }
e031f750 5813 }
28d7a789 5814 return true;
19cb3738
FB
5815}
5816
19cb3738
FB
5817/**
5818 * If eject_flag is TRUE, eject the media. Otherwise, close the tray
5819 */
f36f3949 5820void bdrv_eject(BlockDriverState *bs, bool eject_flag)
19cb3738
FB
5821{
5822 BlockDriver *drv = bs->drv;
19cb3738 5823
822e1cd1
MA
5824 if (drv && drv->bdrv_eject) {
5825 drv->bdrv_eject(bs, eject_flag);
19cb3738
FB
5826 }
5827}
5828
19cb3738
FB
5829/**
5830 * Lock or unlock the media (if it is locked, the user won't be able
5831 * to eject it manually).
5832 */
025e849a 5833void bdrv_lock_medium(BlockDriverState *bs, bool locked)
19cb3738
FB
5834{
5835 BlockDriver *drv = bs->drv;
5836
025e849a 5837 trace_bdrv_lock_medium(bs, locked);
b8c6d095 5838
025e849a
MA
5839 if (drv && drv->bdrv_lock_medium) {
5840 drv->bdrv_lock_medium(bs, locked);
19cb3738
FB
5841 }
5842}
985a03b0 5843
9fcb0251
FZ
5844/* Get a reference to bs */
5845void bdrv_ref(BlockDriverState *bs)
5846{
5847 bs->refcnt++;
5848}
5849
5850/* Release a previously grabbed reference to bs.
5851 * If after releasing, reference count is zero, the BlockDriverState is
5852 * deleted. */
5853void bdrv_unref(BlockDriverState *bs)
5854{
9a4d5ca6
JC
5855 if (!bs) {
5856 return;
5857 }
9fcb0251
FZ
5858 assert(bs->refcnt > 0);
5859 if (--bs->refcnt == 0) {
5860 bdrv_delete(bs);
5861 }
5862}
5863
fbe40ff7
FZ
5864struct BdrvOpBlocker {
5865 Error *reason;
5866 QLIST_ENTRY(BdrvOpBlocker) list;
5867};
5868
5869bool bdrv_op_is_blocked(BlockDriverState *bs, BlockOpType op, Error **errp)
5870{
5871 BdrvOpBlocker *blocker;
5872 assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
5873 if (!QLIST_EMPTY(&bs->op_blockers[op])) {
5874 blocker = QLIST_FIRST(&bs->op_blockers[op]);
4b576648
MA
5875 error_propagate_prepend(errp, error_copy(blocker->reason),
5876 "Node '%s' is busy: ",
5877 bdrv_get_device_or_node_name(bs));
fbe40ff7
FZ
5878 return true;
5879 }
5880 return false;
5881}
5882
5883void bdrv_op_block(BlockDriverState *bs, BlockOpType op, Error *reason)
5884{
5885 BdrvOpBlocker *blocker;
5886 assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
5887
5839e53b 5888 blocker = g_new0(BdrvOpBlocker, 1);
fbe40ff7
FZ
5889 blocker->reason = reason;
5890 QLIST_INSERT_HEAD(&bs->op_blockers[op], blocker, list);
5891}
5892
5893void bdrv_op_unblock(BlockDriverState *bs, BlockOpType op, Error *reason)
5894{
5895 BdrvOpBlocker *blocker, *next;
5896 assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
5897 QLIST_FOREACH_SAFE(blocker, &bs->op_blockers[op], list, next) {
5898 if (blocker->reason == reason) {
5899 QLIST_REMOVE(blocker, list);
5900 g_free(blocker);
5901 }
5902 }
5903}
5904
5905void bdrv_op_block_all(BlockDriverState *bs, Error *reason)
5906{
5907 int i;
5908 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
5909 bdrv_op_block(bs, i, reason);
5910 }
5911}
5912
5913void bdrv_op_unblock_all(BlockDriverState *bs, Error *reason)
5914{
5915 int i;
5916 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
5917 bdrv_op_unblock(bs, i, reason);
5918 }
5919}
5920
5921bool bdrv_op_blocker_is_empty(BlockDriverState *bs)
5922{
5923 int i;
5924
5925 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
5926 if (!QLIST_EMPTY(&bs->op_blockers[i])) {
5927 return false;
5928 }
5929 }
5930 return true;
5931}
5932
d92ada22
LC
5933void bdrv_img_create(const char *filename, const char *fmt,
5934 const char *base_filename, const char *base_fmt,
9217283d
FZ
5935 char *options, uint64_t img_size, int flags, bool quiet,
5936 Error **errp)
f88e1a42 5937{
83d0521a
CL
5938 QemuOptsList *create_opts = NULL;
5939 QemuOpts *opts = NULL;
5940 const char *backing_fmt, *backing_file;
5941 int64_t size;
f88e1a42 5942 BlockDriver *drv, *proto_drv;
cc84d90f 5943 Error *local_err = NULL;
f88e1a42
JS
5944 int ret = 0;
5945
5946 /* Find driver and parse its options */
5947 drv = bdrv_find_format(fmt);
5948 if (!drv) {
71c79813 5949 error_setg(errp, "Unknown file format '%s'", fmt);
d92ada22 5950 return;
f88e1a42
JS
5951 }
5952
b65a5e12 5953 proto_drv = bdrv_find_protocol(filename, true, errp);
f88e1a42 5954 if (!proto_drv) {
d92ada22 5955 return;
f88e1a42
JS
5956 }
5957
c6149724
HR
5958 if (!drv->create_opts) {
5959 error_setg(errp, "Format driver '%s' does not support image creation",
5960 drv->format_name);
5961 return;
5962 }
5963
f6dc1c31 5964 /* Create parameter list */
c282e1fd 5965 create_opts = qemu_opts_append(create_opts, drv->create_opts);
fd17146c
HR
5966 if (proto_drv->create_opts) {
5967 create_opts = qemu_opts_append(create_opts, proto_drv->create_opts);
5968 } else {
5969 create_opts = qemu_opts_append(create_opts, &fallback_create_opts);
5970 }
f88e1a42 5971
83d0521a 5972 opts = qemu_opts_create(create_opts, NULL, 0, &error_abort);
f88e1a42
JS
5973
5974 /* Parse -o options */
5975 if (options) {
dc523cd3
MA
5976 qemu_opts_do_parse(opts, options, NULL, &local_err);
5977 if (local_err) {
f88e1a42
JS
5978 goto out;
5979 }
5980 }
5981
f6dc1c31
KW
5982 if (!qemu_opt_get(opts, BLOCK_OPT_SIZE)) {
5983 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, img_size, &error_abort);
5984 } else if (img_size != UINT64_C(-1)) {
5985 error_setg(errp, "The image size must be specified only once");
5986 goto out;
5987 }
5988
f88e1a42 5989 if (base_filename) {
f43e47db 5990 qemu_opt_set(opts, BLOCK_OPT_BACKING_FILE, base_filename, &local_err);
6be4194b 5991 if (local_err) {
71c79813
LC
5992 error_setg(errp, "Backing file not supported for file format '%s'",
5993 fmt);
f88e1a42
JS
5994 goto out;
5995 }
5996 }
5997
5998 if (base_fmt) {
f43e47db 5999 qemu_opt_set(opts, BLOCK_OPT_BACKING_FMT, base_fmt, &local_err);
6be4194b 6000 if (local_err) {
71c79813
LC
6001 error_setg(errp, "Backing file format not supported for file "
6002 "format '%s'", fmt);
f88e1a42
JS
6003 goto out;
6004 }
6005 }
6006
83d0521a
CL
6007 backing_file = qemu_opt_get(opts, BLOCK_OPT_BACKING_FILE);
6008 if (backing_file) {
6009 if (!strcmp(filename, backing_file)) {
71c79813
LC
6010 error_setg(errp, "Error: Trying to create an image with the "
6011 "same filename as the backing file");
792da93a
JS
6012 goto out;
6013 }
6014 }
6015
83d0521a 6016 backing_fmt = qemu_opt_get(opts, BLOCK_OPT_BACKING_FMT);
f88e1a42 6017
6e6e55f5
JS
6018 /* The size for the image must always be specified, unless we have a backing
6019 * file and we have not been forbidden from opening it. */
a8b42a1c 6020 size = qemu_opt_get_size(opts, BLOCK_OPT_SIZE, img_size);
6e6e55f5
JS
6021 if (backing_file && !(flags & BDRV_O_NO_BACKING)) {
6022 BlockDriverState *bs;
645ae7d8 6023 char *full_backing;
6e6e55f5
JS
6024 int back_flags;
6025 QDict *backing_options = NULL;
6026
645ae7d8
HR
6027 full_backing =
6028 bdrv_get_full_backing_filename_from_filename(filename, backing_file,
6029 &local_err);
6e6e55f5 6030 if (local_err) {
6e6e55f5
JS
6031 goto out;
6032 }
645ae7d8 6033 assert(full_backing);
29168018 6034
6e6e55f5
JS
6035 /* backing files always opened read-only */
6036 back_flags = flags;
6037 back_flags &= ~(BDRV_O_RDWR | BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING);
f88e1a42 6038
cc954f01 6039 backing_options = qdict_new();
6e6e55f5 6040 if (backing_fmt) {
6e6e55f5
JS
6041 qdict_put_str(backing_options, "driver", backing_fmt);
6042 }
cc954f01 6043 qdict_put_bool(backing_options, BDRV_OPT_FORCE_SHARE, true);
e6641719 6044
6e6e55f5
JS
6045 bs = bdrv_open(full_backing, NULL, backing_options, back_flags,
6046 &local_err);
6047 g_free(full_backing);
6048 if (!bs && size != -1) {
6049 /* Couldn't open BS, but we have a size, so it's nonfatal */
6050 warn_reportf_err(local_err,
6051 "Could not verify backing image. "
6052 "This may become an error in future versions.\n");
6053 local_err = NULL;
6054 } else if (!bs) {
6055 /* Couldn't open bs, do not have size */
6056 error_append_hint(&local_err,
6057 "Could not open backing image to determine size.\n");
6058 goto out;
6059 } else {
6060 if (size == -1) {
6061 /* Opened BS, have no size */
6062 size = bdrv_getlength(bs);
6063 if (size < 0) {
6064 error_setg_errno(errp, -size, "Could not get size of '%s'",
6065 backing_file);
6066 bdrv_unref(bs);
6067 goto out;
6068 }
6069 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, size, &error_abort);
52bf1e72 6070 }
66f6b814 6071 bdrv_unref(bs);
f88e1a42 6072 }
6e6e55f5
JS
6073 } /* (backing_file && !(flags & BDRV_O_NO_BACKING)) */
6074
6075 if (size == -1) {
6076 error_setg(errp, "Image creation needs a size parameter");
6077 goto out;
f88e1a42
JS
6078 }
6079
f382d43a 6080 if (!quiet) {
fe646693 6081 printf("Formatting '%s', fmt=%s ", filename, fmt);
43c5d8f8 6082 qemu_opts_print(opts, " ");
f382d43a
MR
6083 puts("");
6084 }
83d0521a 6085
c282e1fd 6086 ret = bdrv_create(drv, filename, opts, &local_err);
83d0521a 6087
cc84d90f
HR
6088 if (ret == -EFBIG) {
6089 /* This is generally a better message than whatever the driver would
6090 * deliver (especially because of the cluster_size_hint), since that
6091 * is most probably not much different from "image too large". */
6092 const char *cluster_size_hint = "";
83d0521a 6093 if (qemu_opt_get_size(opts, BLOCK_OPT_CLUSTER_SIZE, 0)) {
cc84d90f 6094 cluster_size_hint = " (try using a larger cluster size)";
f88e1a42 6095 }
cc84d90f
HR
6096 error_setg(errp, "The image size is too large for file format '%s'"
6097 "%s", fmt, cluster_size_hint);
6098 error_free(local_err);
6099 local_err = NULL;
f88e1a42
JS
6100 }
6101
6102out:
83d0521a
CL
6103 qemu_opts_del(opts);
6104 qemu_opts_free(create_opts);
621ff94d 6105 error_propagate(errp, local_err);
f88e1a42 6106}
85d126f3
SH
6107
6108AioContext *bdrv_get_aio_context(BlockDriverState *bs)
6109{
33f2a757 6110 return bs ? bs->aio_context : qemu_get_aio_context();
dcd04228
SH
6111}
6112
052a7572
FZ
6113void bdrv_coroutine_enter(BlockDriverState *bs, Coroutine *co)
6114{
6115 aio_co_enter(bdrv_get_aio_context(bs), co);
6116}
6117
e8a095da
SH
6118static void bdrv_do_remove_aio_context_notifier(BdrvAioNotifier *ban)
6119{
6120 QLIST_REMOVE(ban, list);
6121 g_free(ban);
6122}
6123
a3a683c3 6124static void bdrv_detach_aio_context(BlockDriverState *bs)
dcd04228 6125{
e8a095da 6126 BdrvAioNotifier *baf, *baf_tmp;
33384421 6127
e8a095da
SH
6128 assert(!bs->walking_aio_notifiers);
6129 bs->walking_aio_notifiers = true;
6130 QLIST_FOREACH_SAFE(baf, &bs->aio_notifiers, list, baf_tmp) {
6131 if (baf->deleted) {
6132 bdrv_do_remove_aio_context_notifier(baf);
6133 } else {
6134 baf->detach_aio_context(baf->opaque);
6135 }
33384421 6136 }
e8a095da
SH
6137 /* Never mind iterating again to check for ->deleted. bdrv_close() will
6138 * remove remaining aio notifiers if we aren't called again.
6139 */
6140 bs->walking_aio_notifiers = false;
33384421 6141
1bffe1ae 6142 if (bs->drv && bs->drv->bdrv_detach_aio_context) {
dcd04228
SH
6143 bs->drv->bdrv_detach_aio_context(bs);
6144 }
dcd04228 6145
e64f25f3
KW
6146 if (bs->quiesce_counter) {
6147 aio_enable_external(bs->aio_context);
6148 }
dcd04228
SH
6149 bs->aio_context = NULL;
6150}
6151
a3a683c3
KW
6152static void bdrv_attach_aio_context(BlockDriverState *bs,
6153 AioContext *new_context)
dcd04228 6154{
e8a095da 6155 BdrvAioNotifier *ban, *ban_tmp;
33384421 6156
e64f25f3
KW
6157 if (bs->quiesce_counter) {
6158 aio_disable_external(new_context);
6159 }
6160
dcd04228
SH
6161 bs->aio_context = new_context;
6162
1bffe1ae 6163 if (bs->drv && bs->drv->bdrv_attach_aio_context) {
dcd04228
SH
6164 bs->drv->bdrv_attach_aio_context(bs, new_context);
6165 }
33384421 6166
e8a095da
SH
6167 assert(!bs->walking_aio_notifiers);
6168 bs->walking_aio_notifiers = true;
6169 QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_tmp) {
6170 if (ban->deleted) {
6171 bdrv_do_remove_aio_context_notifier(ban);
6172 } else {
6173 ban->attached_aio_context(new_context, ban->opaque);
6174 }
33384421 6175 }
e8a095da 6176 bs->walking_aio_notifiers = false;
dcd04228
SH
6177}
6178
42a65f02
KW
6179/*
6180 * Changes the AioContext used for fd handlers, timers, and BHs by this
6181 * BlockDriverState and all its children and parents.
6182 *
43eaaaef
HR
6183 * Must be called from the main AioContext.
6184 *
42a65f02
KW
6185 * The caller must own the AioContext lock for the old AioContext of bs, but it
6186 * must not own the AioContext lock for new_context (unless new_context is the
6187 * same as the current context of bs).
6188 *
6189 * @ignore will accumulate all visited BdrvChild object. The caller is
6190 * responsible for freeing the list afterwards.
6191 */
53a7d041
KW
6192void bdrv_set_aio_context_ignore(BlockDriverState *bs,
6193 AioContext *new_context, GSList **ignore)
dcd04228 6194{
e037c09c 6195 AioContext *old_context = bdrv_get_aio_context(bs);
0d83708a
KW
6196 BdrvChild *child;
6197
43eaaaef
HR
6198 g_assert(qemu_get_current_aio_context() == qemu_get_aio_context());
6199
e037c09c 6200 if (old_context == new_context) {
57830a49
DP
6201 return;
6202 }
6203
d70d5954 6204 bdrv_drained_begin(bs);
0d83708a
KW
6205
6206 QLIST_FOREACH(child, &bs->children, next) {
53a7d041
KW
6207 if (g_slist_find(*ignore, child)) {
6208 continue;
6209 }
6210 *ignore = g_slist_prepend(*ignore, child);
6211 bdrv_set_aio_context_ignore(child->bs, new_context, ignore);
6212 }
6213 QLIST_FOREACH(child, &bs->parents, next_parent) {
6214 if (g_slist_find(*ignore, child)) {
6215 continue;
6216 }
42a65f02
KW
6217 assert(child->role->set_aio_ctx);
6218 *ignore = g_slist_prepend(*ignore, child);
6219 child->role->set_aio_ctx(child, new_context, ignore);
0d83708a
KW
6220 }
6221
dcd04228
SH
6222 bdrv_detach_aio_context(bs);
6223
e037c09c 6224 /* Acquire the new context, if necessary */
43eaaaef 6225 if (qemu_get_aio_context() != new_context) {
e037c09c
HR
6226 aio_context_acquire(new_context);
6227 }
6228
dcd04228 6229 bdrv_attach_aio_context(bs, new_context);
e037c09c
HR
6230
6231 /*
6232 * If this function was recursively called from
6233 * bdrv_set_aio_context_ignore(), there may be nodes in the
6234 * subtree that have not yet been moved to the new AioContext.
6235 * Release the old one so bdrv_drained_end() can poll them.
6236 */
43eaaaef 6237 if (qemu_get_aio_context() != old_context) {
e037c09c
HR
6238 aio_context_release(old_context);
6239 }
6240
d70d5954 6241 bdrv_drained_end(bs);
e037c09c 6242
43eaaaef 6243 if (qemu_get_aio_context() != old_context) {
e037c09c
HR
6244 aio_context_acquire(old_context);
6245 }
43eaaaef 6246 if (qemu_get_aio_context() != new_context) {
e037c09c
HR
6247 aio_context_release(new_context);
6248 }
85d126f3 6249}
d616b224 6250
5d231849
KW
6251static bool bdrv_parent_can_set_aio_context(BdrvChild *c, AioContext *ctx,
6252 GSList **ignore, Error **errp)
6253{
6254 if (g_slist_find(*ignore, c)) {
6255 return true;
6256 }
6257 *ignore = g_slist_prepend(*ignore, c);
6258
6259 /* A BdrvChildRole that doesn't handle AioContext changes cannot
6260 * tolerate any AioContext changes */
6261 if (!c->role->can_set_aio_ctx) {
6262 char *user = bdrv_child_user_desc(c);
6263 error_setg(errp, "Changing iothreads is not supported by %s", user);
6264 g_free(user);
6265 return false;
6266 }
6267 if (!c->role->can_set_aio_ctx(c, ctx, ignore, errp)) {
6268 assert(!errp || *errp);
6269 return false;
6270 }
6271 return true;
6272}
6273
6274bool bdrv_child_can_set_aio_context(BdrvChild *c, AioContext *ctx,
6275 GSList **ignore, Error **errp)
6276{
6277 if (g_slist_find(*ignore, c)) {
6278 return true;
6279 }
6280 *ignore = g_slist_prepend(*ignore, c);
6281 return bdrv_can_set_aio_context(c->bs, ctx, ignore, errp);
6282}
6283
6284/* @ignore will accumulate all visited BdrvChild object. The caller is
6285 * responsible for freeing the list afterwards. */
6286bool bdrv_can_set_aio_context(BlockDriverState *bs, AioContext *ctx,
6287 GSList **ignore, Error **errp)
6288{
6289 BdrvChild *c;
6290
6291 if (bdrv_get_aio_context(bs) == ctx) {
6292 return true;
6293 }
6294
6295 QLIST_FOREACH(c, &bs->parents, next_parent) {
6296 if (!bdrv_parent_can_set_aio_context(c, ctx, ignore, errp)) {
6297 return false;
6298 }
6299 }
6300 QLIST_FOREACH(c, &bs->children, next) {
6301 if (!bdrv_child_can_set_aio_context(c, ctx, ignore, errp)) {
6302 return false;
6303 }
6304 }
6305
6306 return true;
6307}
6308
6309int bdrv_child_try_set_aio_context(BlockDriverState *bs, AioContext *ctx,
6310 BdrvChild *ignore_child, Error **errp)
6311{
6312 GSList *ignore;
6313 bool ret;
6314
6315 ignore = ignore_child ? g_slist_prepend(NULL, ignore_child) : NULL;
6316 ret = bdrv_can_set_aio_context(bs, ctx, &ignore, errp);
6317 g_slist_free(ignore);
6318
6319 if (!ret) {
6320 return -EPERM;
6321 }
6322
53a7d041
KW
6323 ignore = ignore_child ? g_slist_prepend(NULL, ignore_child) : NULL;
6324 bdrv_set_aio_context_ignore(bs, ctx, &ignore);
6325 g_slist_free(ignore);
6326
5d231849
KW
6327 return 0;
6328}
6329
6330int bdrv_try_set_aio_context(BlockDriverState *bs, AioContext *ctx,
6331 Error **errp)
6332{
6333 return bdrv_child_try_set_aio_context(bs, ctx, NULL, errp);
6334}
6335
33384421
HR
6336void bdrv_add_aio_context_notifier(BlockDriverState *bs,
6337 void (*attached_aio_context)(AioContext *new_context, void *opaque),
6338 void (*detach_aio_context)(void *opaque), void *opaque)
6339{
6340 BdrvAioNotifier *ban = g_new(BdrvAioNotifier, 1);
6341 *ban = (BdrvAioNotifier){
6342 .attached_aio_context = attached_aio_context,
6343 .detach_aio_context = detach_aio_context,
6344 .opaque = opaque
6345 };
6346
6347 QLIST_INSERT_HEAD(&bs->aio_notifiers, ban, list);
6348}
6349
6350void bdrv_remove_aio_context_notifier(BlockDriverState *bs,
6351 void (*attached_aio_context)(AioContext *,
6352 void *),
6353 void (*detach_aio_context)(void *),
6354 void *opaque)
6355{
6356 BdrvAioNotifier *ban, *ban_next;
6357
6358 QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_next) {
6359 if (ban->attached_aio_context == attached_aio_context &&
6360 ban->detach_aio_context == detach_aio_context &&
e8a095da
SH
6361 ban->opaque == opaque &&
6362 ban->deleted == false)
33384421 6363 {
e8a095da
SH
6364 if (bs->walking_aio_notifiers) {
6365 ban->deleted = true;
6366 } else {
6367 bdrv_do_remove_aio_context_notifier(ban);
6368 }
33384421
HR
6369 return;
6370 }
6371 }
6372
6373 abort();
6374}
6375
77485434 6376int bdrv_amend_options(BlockDriverState *bs, QemuOpts *opts,
d1402b50
HR
6377 BlockDriverAmendStatusCB *status_cb, void *cb_opaque,
6378 Error **errp)
6f176b48 6379{
d470ad42 6380 if (!bs->drv) {
d1402b50 6381 error_setg(errp, "Node is ejected");
d470ad42
HR
6382 return -ENOMEDIUM;
6383 }
c282e1fd 6384 if (!bs->drv->bdrv_amend_options) {
d1402b50
HR
6385 error_setg(errp, "Block driver '%s' does not support option amendment",
6386 bs->drv->format_name);
6f176b48
HR
6387 return -ENOTSUP;
6388 }
d1402b50 6389 return bs->drv->bdrv_amend_options(bs, opts, status_cb, cb_opaque, errp);
6f176b48 6390}
f6186f49 6391
5d69b5ab
HR
6392/*
6393 * This function checks whether the given @to_replace is allowed to be
6394 * replaced by a node that always shows the same data as @bs. This is
6395 * used for example to verify whether the mirror job can replace
6396 * @to_replace by the target mirrored from @bs.
6397 * To be replaceable, @bs and @to_replace may either be guaranteed to
6398 * always show the same data (because they are only connected through
6399 * filters), or some driver may allow replacing one of its children
6400 * because it can guarantee that this child's data is not visible at
6401 * all (for example, for dissenting quorum children that have no other
6402 * parents).
6403 */
6404bool bdrv_recurse_can_replace(BlockDriverState *bs,
6405 BlockDriverState *to_replace)
6406{
6407 if (!bs || !bs->drv) {
6408 return false;
6409 }
6410
6411 if (bs == to_replace) {
6412 return true;
6413 }
6414
6415 /* See what the driver can do */
6416 if (bs->drv->bdrv_recurse_can_replace) {
6417 return bs->drv->bdrv_recurse_can_replace(bs, to_replace);
6418 }
6419
6420 /* For filters without an own implementation, we can recurse on our own */
6421 if (bs->drv->is_filter) {
6422 BdrvChild *child = bs->file ?: bs->backing;
6423 return bdrv_recurse_can_replace(child->bs, to_replace);
6424 }
6425
6426 /* Safe default */
6427 return false;
6428}
6429
810803a8
HR
6430/*
6431 * Check whether the given @node_name can be replaced by a node that
6432 * has the same data as @parent_bs. If so, return @node_name's BDS;
6433 * NULL otherwise.
6434 *
6435 * @node_name must be a (recursive) *child of @parent_bs (or this
6436 * function will return NULL).
6437 *
6438 * The result (whether the node can be replaced or not) is only valid
6439 * for as long as no graph or permission changes occur.
6440 */
e12f3784
WC
6441BlockDriverState *check_to_replace_node(BlockDriverState *parent_bs,
6442 const char *node_name, Error **errp)
09158f00
BC
6443{
6444 BlockDriverState *to_replace_bs = bdrv_find_node(node_name);
5a7e7a0b
SH
6445 AioContext *aio_context;
6446
09158f00
BC
6447 if (!to_replace_bs) {
6448 error_setg(errp, "Node name '%s' not found", node_name);
6449 return NULL;
6450 }
6451
5a7e7a0b
SH
6452 aio_context = bdrv_get_aio_context(to_replace_bs);
6453 aio_context_acquire(aio_context);
6454
09158f00 6455 if (bdrv_op_is_blocked(to_replace_bs, BLOCK_OP_TYPE_REPLACE, errp)) {
5a7e7a0b
SH
6456 to_replace_bs = NULL;
6457 goto out;
09158f00
BC
6458 }
6459
6460 /* We don't want arbitrary node of the BDS chain to be replaced only the top
6461 * most non filter in order to prevent data corruption.
6462 * Another benefit is that this tests exclude backing files which are
6463 * blocked by the backing blockers.
6464 */
810803a8
HR
6465 if (!bdrv_recurse_can_replace(parent_bs, to_replace_bs)) {
6466 error_setg(errp, "Cannot replace '%s' by a node mirrored from '%s', "
6467 "because it cannot be guaranteed that doing so would not "
6468 "lead to an abrupt change of visible data",
6469 node_name, parent_bs->node_name);
5a7e7a0b
SH
6470 to_replace_bs = NULL;
6471 goto out;
09158f00
BC
6472 }
6473
5a7e7a0b
SH
6474out:
6475 aio_context_release(aio_context);
09158f00
BC
6476 return to_replace_bs;
6477}
448ad91d 6478
97e2f021
HR
6479/**
6480 * Iterates through the list of runtime option keys that are said to
6481 * be "strong" for a BDS. An option is called "strong" if it changes
6482 * a BDS's data. For example, the null block driver's "size" and
6483 * "read-zeroes" options are strong, but its "latency-ns" option is
6484 * not.
6485 *
6486 * If a key returned by this function ends with a dot, all options
6487 * starting with that prefix are strong.
6488 */
6489static const char *const *strong_options(BlockDriverState *bs,
6490 const char *const *curopt)
6491{
6492 static const char *const global_options[] = {
6493 "driver", "filename", NULL
6494 };
6495
6496 if (!curopt) {
6497 return &global_options[0];
6498 }
6499
6500 curopt++;
6501 if (curopt == &global_options[ARRAY_SIZE(global_options) - 1] && bs->drv) {
6502 curopt = bs->drv->strong_runtime_opts;
6503 }
6504
6505 return (curopt && *curopt) ? curopt : NULL;
6506}
6507
6508/**
6509 * Copies all strong runtime options from bs->options to the given
6510 * QDict. The set of strong option keys is determined by invoking
6511 * strong_options().
6512 *
6513 * Returns true iff any strong option was present in bs->options (and
6514 * thus copied to the target QDict) with the exception of "filename"
6515 * and "driver". The caller is expected to use this value to decide
6516 * whether the existence of strong options prevents the generation of
6517 * a plain filename.
6518 */
6519static bool append_strong_runtime_options(QDict *d, BlockDriverState *bs)
6520{
6521 bool found_any = false;
6522 const char *const *option_name = NULL;
6523
6524 if (!bs->drv) {
6525 return false;
6526 }
6527
6528 while ((option_name = strong_options(bs, option_name))) {
6529 bool option_given = false;
6530
6531 assert(strlen(*option_name) > 0);
6532 if ((*option_name)[strlen(*option_name) - 1] != '.') {
6533 QObject *entry = qdict_get(bs->options, *option_name);
6534 if (!entry) {
6535 continue;
6536 }
6537
6538 qdict_put_obj(d, *option_name, qobject_ref(entry));
6539 option_given = true;
6540 } else {
6541 const QDictEntry *entry;
6542 for (entry = qdict_first(bs->options); entry;
6543 entry = qdict_next(bs->options, entry))
6544 {
6545 if (strstart(qdict_entry_key(entry), *option_name, NULL)) {
6546 qdict_put_obj(d, qdict_entry_key(entry),
6547 qobject_ref(qdict_entry_value(entry)));
6548 option_given = true;
6549 }
6550 }
6551 }
6552
6553 /* While "driver" and "filename" need to be included in a JSON filename,
6554 * their existence does not prohibit generation of a plain filename. */
6555 if (!found_any && option_given &&
6556 strcmp(*option_name, "driver") && strcmp(*option_name, "filename"))
6557 {
6558 found_any = true;
6559 }
6560 }
6561
62a01a27
HR
6562 if (!qdict_haskey(d, "driver")) {
6563 /* Drivers created with bdrv_new_open_driver() may not have a
6564 * @driver option. Add it here. */
6565 qdict_put_str(d, "driver", bs->drv->format_name);
6566 }
6567
97e2f021
HR
6568 return found_any;
6569}
6570
90993623
HR
6571/* Note: This function may return false positives; it may return true
6572 * even if opening the backing file specified by bs's image header
6573 * would result in exactly bs->backing. */
6574static bool bdrv_backing_overridden(BlockDriverState *bs)
6575{
6576 if (bs->backing) {
6577 return strcmp(bs->auto_backing_file,
6578 bs->backing->bs->filename);
6579 } else {
6580 /* No backing BDS, so if the image header reports any backing
6581 * file, it must have been suppressed */
6582 return bs->auto_backing_file[0] != '\0';
6583 }
6584}
6585
91af7014
HR
6586/* Updates the following BDS fields:
6587 * - exact_filename: A filename which may be used for opening a block device
6588 * which (mostly) equals the given BDS (even without any
6589 * other options; so reading and writing must return the same
6590 * results, but caching etc. may be different)
6591 * - full_open_options: Options which, when given when opening a block device
6592 * (without a filename), result in a BDS (mostly)
6593 * equalling the given one
6594 * - filename: If exact_filename is set, it is copied here. Otherwise,
6595 * full_open_options is converted to a JSON object, prefixed with
6596 * "json:" (for use through the JSON pseudo protocol) and put here.
6597 */
6598void bdrv_refresh_filename(BlockDriverState *bs)
6599{
6600 BlockDriver *drv = bs->drv;
e24518e3 6601 BdrvChild *child;
91af7014 6602 QDict *opts;
90993623 6603 bool backing_overridden;
998b3a1e
HR
6604 bool generate_json_filename; /* Whether our default implementation should
6605 fill exact_filename (false) or not (true) */
91af7014
HR
6606
6607 if (!drv) {
6608 return;
6609 }
6610
e24518e3
HR
6611 /* This BDS's file name may depend on any of its children's file names, so
6612 * refresh those first */
6613 QLIST_FOREACH(child, &bs->children, next) {
6614 bdrv_refresh_filename(child->bs);
91af7014
HR
6615 }
6616
bb808d5f
HR
6617 if (bs->implicit) {
6618 /* For implicit nodes, just copy everything from the single child */
6619 child = QLIST_FIRST(&bs->children);
6620 assert(QLIST_NEXT(child, next) == NULL);
6621
6622 pstrcpy(bs->exact_filename, sizeof(bs->exact_filename),
6623 child->bs->exact_filename);
6624 pstrcpy(bs->filename, sizeof(bs->filename), child->bs->filename);
6625
cb895614 6626 qobject_unref(bs->full_open_options);
bb808d5f
HR
6627 bs->full_open_options = qobject_ref(child->bs->full_open_options);
6628
6629 return;
6630 }
6631
90993623
HR
6632 backing_overridden = bdrv_backing_overridden(bs);
6633
6634 if (bs->open_flags & BDRV_O_NO_IO) {
6635 /* Without I/O, the backing file does not change anything.
6636 * Therefore, in such a case (primarily qemu-img), we can
6637 * pretend the backing file has not been overridden even if
6638 * it technically has been. */
6639 backing_overridden = false;
6640 }
6641
97e2f021
HR
6642 /* Gather the options QDict */
6643 opts = qdict_new();
998b3a1e
HR
6644 generate_json_filename = append_strong_runtime_options(opts, bs);
6645 generate_json_filename |= backing_overridden;
97e2f021
HR
6646
6647 if (drv->bdrv_gather_child_options) {
6648 /* Some block drivers may not want to present all of their children's
6649 * options, or name them differently from BdrvChild.name */
6650 drv->bdrv_gather_child_options(bs, opts, backing_overridden);
6651 } else {
6652 QLIST_FOREACH(child, &bs->children, next) {
6653 if (child->role == &child_backing && !backing_overridden) {
6654 /* We can skip the backing BDS if it has not been overridden */
6655 continue;
6656 }
6657
6658 qdict_put(opts, child->name,
6659 qobject_ref(child->bs->full_open_options));
6660 }
6661
6662 if (backing_overridden && !bs->backing) {
6663 /* Force no backing file */
6664 qdict_put_null(opts, "backing");
6665 }
6666 }
6667
6668 qobject_unref(bs->full_open_options);
6669 bs->full_open_options = opts;
6670
998b3a1e
HR
6671 if (drv->bdrv_refresh_filename) {
6672 /* Obsolete information is of no use here, so drop the old file name
6673 * information before refreshing it */
6674 bs->exact_filename[0] = '\0';
6675
6676 drv->bdrv_refresh_filename(bs);
6677 } else if (bs->file) {
6678 /* Try to reconstruct valid information from the underlying file */
6679
6680 bs->exact_filename[0] = '\0';
6681
fb695c74
HR
6682 /*
6683 * We can use the underlying file's filename if:
6684 * - it has a filename,
6685 * - the file is a protocol BDS, and
6686 * - opening that file (as this BDS's format) will automatically create
6687 * the BDS tree we have right now, that is:
6688 * - the user did not significantly change this BDS's behavior with
6689 * some explicit (strong) options
6690 * - no non-file child of this BDS has been overridden by the user
6691 * Both of these conditions are represented by generate_json_filename.
6692 */
6693 if (bs->file->bs->exact_filename[0] &&
6694 bs->file->bs->drv->bdrv_file_open &&
6695 !generate_json_filename)
6696 {
998b3a1e
HR
6697 strcpy(bs->exact_filename, bs->file->bs->exact_filename);
6698 }
6699 }
6700
91af7014
HR
6701 if (bs->exact_filename[0]) {
6702 pstrcpy(bs->filename, sizeof(bs->filename), bs->exact_filename);
97e2f021 6703 } else {
91af7014
HR
6704 QString *json = qobject_to_json(QOBJECT(bs->full_open_options));
6705 snprintf(bs->filename, sizeof(bs->filename), "json:%s",
6706 qstring_get_str(json));
cb3e7f08 6707 qobject_unref(json);
91af7014
HR
6708 }
6709}
e06018ad 6710
1e89d0f9
HR
6711char *bdrv_dirname(BlockDriverState *bs, Error **errp)
6712{
6713 BlockDriver *drv = bs->drv;
6714
6715 if (!drv) {
6716 error_setg(errp, "Node '%s' is ejected", bs->node_name);
6717 return NULL;
6718 }
6719
6720 if (drv->bdrv_dirname) {
6721 return drv->bdrv_dirname(bs, errp);
6722 }
6723
6724 if (bs->file) {
6725 return bdrv_dirname(bs->file->bs, errp);
6726 }
6727
6728 bdrv_refresh_filename(bs);
6729 if (bs->exact_filename[0] != '\0') {
6730 return path_combine(bs->exact_filename, "");
6731 }
6732
6733 error_setg(errp, "Cannot generate a base directory for %s nodes",
6734 drv->format_name);
6735 return NULL;
6736}
6737
e06018ad
WC
6738/*
6739 * Hot add/remove a BDS's child. So the user can take a child offline when
6740 * it is broken and take a new child online
6741 */
6742void bdrv_add_child(BlockDriverState *parent_bs, BlockDriverState *child_bs,
6743 Error **errp)
6744{
6745
6746 if (!parent_bs->drv || !parent_bs->drv->bdrv_add_child) {
6747 error_setg(errp, "The node %s does not support adding a child",
6748 bdrv_get_device_or_node_name(parent_bs));
6749 return;
6750 }
6751
6752 if (!QLIST_EMPTY(&child_bs->parents)) {
6753 error_setg(errp, "The node %s already has a parent",
6754 child_bs->node_name);
6755 return;
6756 }
6757
6758 parent_bs->drv->bdrv_add_child(parent_bs, child_bs, errp);
6759}
6760
6761void bdrv_del_child(BlockDriverState *parent_bs, BdrvChild *child, Error **errp)
6762{
6763 BdrvChild *tmp;
6764
6765 if (!parent_bs->drv || !parent_bs->drv->bdrv_del_child) {
6766 error_setg(errp, "The node %s does not support removing a child",
6767 bdrv_get_device_or_node_name(parent_bs));
6768 return;
6769 }
6770
6771 QLIST_FOREACH(tmp, &parent_bs->children, next) {
6772 if (tmp == child) {
6773 break;
6774 }
6775 }
6776
6777 if (!tmp) {
6778 error_setg(errp, "The node %s does not have a child named %s",
6779 bdrv_get_device_or_node_name(parent_bs),
6780 bdrv_get_device_or_node_name(child->bs));
6781 return;
6782 }
6783
6784 parent_bs->drv->bdrv_del_child(parent_bs, child, errp);
6785}