]> git.proxmox.com Git - mirror_qemu.git/blame - block.c
dump: Set correct vaddr for ELF dump
[mirror_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"
1de7afc9 33#include "qemu/module.h"
e688df6b 34#include "qapi/error.h"
452fcdbc 35#include "qapi/qmp/qdict.h"
7b1b5d19 36#include "qapi/qmp/qjson.h"
e59a0cf1 37#include "qapi/qmp/qnull.h"
fc81fa1e 38#include "qapi/qmp/qstring.h"
e1d74bc6
KW
39#include "qapi/qobject-output-visitor.h"
40#include "qapi/qapi-visit-block-core.h"
bfb197e0 41#include "sysemu/block-backend.h"
9c17d615 42#include "sysemu/sysemu.h"
1de7afc9 43#include "qemu/notify.h"
922a01a0 44#include "qemu/option.h"
10817bf0 45#include "qemu/coroutine.h"
c13163fb 46#include "block/qapi.h"
1de7afc9 47#include "qemu/timer.h"
f348b6d1
VB
48#include "qemu/cutils.h"
49#include "qemu/id.h"
fc01f7e7 50
71e72a19 51#ifdef CONFIG_BSD
7674e7bf 52#include <sys/ioctl.h>
72cf2d4f 53#include <sys/queue.h>
c5e97233 54#ifndef __DragonFly__
7674e7bf
FB
55#include <sys/disk.h>
56#endif
c5e97233 57#endif
7674e7bf 58
49dc768d
AL
59#ifdef _WIN32
60#include <windows.h>
61#endif
62
1c9805a3
SH
63#define NOT_DONE 0x7fffffff /* used while emulated sync operation in progress */
64
dc364f4c
BC
65static QTAILQ_HEAD(, BlockDriverState) graph_bdrv_states =
66 QTAILQ_HEAD_INITIALIZER(graph_bdrv_states);
67
2c1d04e0
HR
68static QTAILQ_HEAD(, BlockDriverState) all_bdrv_states =
69 QTAILQ_HEAD_INITIALIZER(all_bdrv_states);
70
8a22f02a
SH
71static QLIST_HEAD(, BlockDriver) bdrv_drivers =
72 QLIST_HEAD_INITIALIZER(bdrv_drivers);
ea2384d3 73
5b363937
HR
74static BlockDriverState *bdrv_open_inherit(const char *filename,
75 const char *reference,
76 QDict *options, int flags,
77 BlockDriverState *parent,
78 const BdrvChildRole *child_role,
79 Error **errp);
f3930ed0 80
eb852011
MA
81/* If non-zero, use only whitelisted block drivers */
82static int use_bdrv_whitelist;
83
9e0b22f4
SH
84#ifdef _WIN32
85static int is_windows_drive_prefix(const char *filename)
86{
87 return (((filename[0] >= 'a' && filename[0] <= 'z') ||
88 (filename[0] >= 'A' && filename[0] <= 'Z')) &&
89 filename[1] == ':');
90}
91
92int is_windows_drive(const char *filename)
93{
94 if (is_windows_drive_prefix(filename) &&
95 filename[2] == '\0')
96 return 1;
97 if (strstart(filename, "\\\\.\\", NULL) ||
98 strstart(filename, "//./", NULL))
99 return 1;
100 return 0;
101}
102#endif
103
339064d5
KW
104size_t bdrv_opt_mem_align(BlockDriverState *bs)
105{
106 if (!bs || !bs->drv) {
459b4e66
DL
107 /* page size or 4k (hdd sector size) should be on the safe side */
108 return MAX(4096, getpagesize());
339064d5
KW
109 }
110
111 return bs->bl.opt_mem_alignment;
112}
113
4196d2f0
DL
114size_t bdrv_min_mem_align(BlockDriverState *bs)
115{
116 if (!bs || !bs->drv) {
459b4e66
DL
117 /* page size or 4k (hdd sector size) should be on the safe side */
118 return MAX(4096, getpagesize());
4196d2f0
DL
119 }
120
121 return bs->bl.min_mem_alignment;
122}
123
9e0b22f4 124/* check if the path starts with "<protocol>:" */
5c98415b 125int path_has_protocol(const char *path)
9e0b22f4 126{
947995c0
PB
127 const char *p;
128
9e0b22f4
SH
129#ifdef _WIN32
130 if (is_windows_drive(path) ||
131 is_windows_drive_prefix(path)) {
132 return 0;
133 }
947995c0
PB
134 p = path + strcspn(path, ":/\\");
135#else
136 p = path + strcspn(path, ":/");
9e0b22f4
SH
137#endif
138
947995c0 139 return *p == ':';
9e0b22f4
SH
140}
141
83f64091 142int path_is_absolute(const char *path)
3b0d4f61 143{
21664424
FB
144#ifdef _WIN32
145 /* specific case for names like: "\\.\d:" */
f53f4da9 146 if (is_windows_drive(path) || is_windows_drive_prefix(path)) {
21664424 147 return 1;
f53f4da9
PB
148 }
149 return (*path == '/' || *path == '\\');
3b9f94e1 150#else
f53f4da9 151 return (*path == '/');
3b9f94e1 152#endif
3b0d4f61
FB
153}
154
83f64091
FB
155/* if filename is absolute, just copy it to dest. Otherwise, build a
156 path to it by considering it is relative to base_path. URL are
157 supported. */
158void path_combine(char *dest, int dest_size,
159 const char *base_path,
160 const char *filename)
3b0d4f61 161{
83f64091
FB
162 const char *p, *p1;
163 int len;
164
165 if (dest_size <= 0)
166 return;
167 if (path_is_absolute(filename)) {
168 pstrcpy(dest, dest_size, filename);
169 } else {
0d54a6fe
HR
170 const char *protocol_stripped = NULL;
171
172 if (path_has_protocol(base_path)) {
173 protocol_stripped = strchr(base_path, ':');
174 if (protocol_stripped) {
175 protocol_stripped++;
176 }
177 }
178 p = protocol_stripped ?: base_path;
179
3b9f94e1
FB
180 p1 = strrchr(base_path, '/');
181#ifdef _WIN32
182 {
183 const char *p2;
184 p2 = strrchr(base_path, '\\');
185 if (!p1 || p2 > p1)
186 p1 = p2;
187 }
188#endif
83f64091
FB
189 if (p1)
190 p1++;
191 else
192 p1 = base_path;
193 if (p1 > p)
194 p = p1;
195 len = p - base_path;
196 if (len > dest_size - 1)
197 len = dest_size - 1;
198 memcpy(dest, base_path, len);
199 dest[len] = '\0';
200 pstrcat(dest, dest_size, filename);
3b0d4f61 201 }
3b0d4f61
FB
202}
203
03c320d8
HR
204/*
205 * Helper function for bdrv_parse_filename() implementations to remove optional
206 * protocol prefixes (especially "file:") from a filename and for putting the
207 * stripped filename into the options QDict if there is such a prefix.
208 */
209void bdrv_parse_filename_strip_prefix(const char *filename, const char *prefix,
210 QDict *options)
211{
212 if (strstart(filename, prefix, &filename)) {
213 /* Stripping the explicit protocol prefix may result in a protocol
214 * prefix being (wrongly) detected (if the filename contains a colon) */
215 if (path_has_protocol(filename)) {
216 QString *fat_filename;
217
218 /* This means there is some colon before the first slash; therefore,
219 * this cannot be an absolute path */
220 assert(!path_is_absolute(filename));
221
222 /* And we can thus fix the protocol detection issue by prefixing it
223 * by "./" */
224 fat_filename = qstring_from_str("./");
225 qstring_append(fat_filename, filename);
226
227 assert(!path_has_protocol(qstring_get_str(fat_filename)));
228
229 qdict_put(options, "filename", fat_filename);
230 } else {
231 /* If no protocol prefix was detected, we can use the shortened
232 * filename as-is */
233 qdict_put_str(options, "filename", filename);
234 }
235 }
236}
237
238
9c5e6594
KW
239/* Returns whether the image file is opened as read-only. Note that this can
240 * return false and writing to the image file is still not possible because the
241 * image is inactivated. */
93ed524e
JC
242bool bdrv_is_read_only(BlockDriverState *bs)
243{
244 return bs->read_only;
245}
246
54a32bfe
KW
247int bdrv_can_set_read_only(BlockDriverState *bs, bool read_only,
248 bool ignore_allow_rdw, Error **errp)
fe5241bf 249{
e2b8247a
JC
250 /* Do not set read_only if copy_on_read is enabled */
251 if (bs->copy_on_read && read_only) {
252 error_setg(errp, "Can't set node '%s' to r/o with copy-on-read enabled",
253 bdrv_get_device_or_node_name(bs));
254 return -EINVAL;
255 }
256
d6fcdf06 257 /* Do not clear read_only if it is prohibited */
54a32bfe
KW
258 if (!read_only && !(bs->open_flags & BDRV_O_ALLOW_RDWR) &&
259 !ignore_allow_rdw)
260 {
d6fcdf06
JC
261 error_setg(errp, "Node '%s' is read only",
262 bdrv_get_device_or_node_name(bs));
263 return -EPERM;
264 }
265
45803a03
JC
266 return 0;
267}
268
eaa2410f
KW
269/*
270 * Called by a driver that can only provide a read-only image.
271 *
272 * Returns 0 if the node is already read-only or it could switch the node to
273 * read-only because BDRV_O_AUTO_RDONLY is set.
274 *
275 * Returns -EACCES if the node is read-write and BDRV_O_AUTO_RDONLY is not set
276 * or bdrv_can_set_read_only() forbids making the node read-only. If @errmsg
277 * is not NULL, it is used as the error message for the Error object.
278 */
279int bdrv_apply_auto_read_only(BlockDriverState *bs, const char *errmsg,
280 Error **errp)
45803a03
JC
281{
282 int ret = 0;
283
eaa2410f
KW
284 if (!(bs->open_flags & BDRV_O_RDWR)) {
285 return 0;
286 }
287 if (!(bs->open_flags & BDRV_O_AUTO_RDONLY)) {
288 goto fail;
45803a03
JC
289 }
290
eaa2410f
KW
291 ret = bdrv_can_set_read_only(bs, true, false, NULL);
292 if (ret < 0) {
293 goto fail;
eeae6a59
KW
294 }
295
eaa2410f
KW
296 bs->read_only = true;
297 bs->open_flags &= ~BDRV_O_RDWR;
298
e2b8247a 299 return 0;
eaa2410f
KW
300
301fail:
302 error_setg(errp, "%s", errmsg ?: "Image is read-only");
303 return -EACCES;
fe5241bf
JC
304}
305
0a82855a
HR
306void bdrv_get_full_backing_filename_from_filename(const char *backed,
307 const char *backing,
9f07429e
HR
308 char *dest, size_t sz,
309 Error **errp)
dc5a1371 310{
9f07429e
HR
311 if (backing[0] == '\0' || path_has_protocol(backing) ||
312 path_is_absolute(backing))
313 {
0a82855a 314 pstrcpy(dest, sz, backing);
9f07429e
HR
315 } else if (backed[0] == '\0' || strstart(backed, "json:", NULL)) {
316 error_setg(errp, "Cannot use relative backing file names for '%s'",
317 backed);
dc5a1371 318 } else {
0a82855a 319 path_combine(dest, sz, backed, backing);
dc5a1371
PB
320 }
321}
322
9f07429e
HR
323void bdrv_get_full_backing_filename(BlockDriverState *bs, char *dest, size_t sz,
324 Error **errp)
0a82855a 325{
9f07429e
HR
326 char *backed = bs->exact_filename[0] ? bs->exact_filename : bs->filename;
327
328 bdrv_get_full_backing_filename_from_filename(backed, bs->backing_file,
329 dest, sz, errp);
0a82855a
HR
330}
331
0eb7217e
SH
332void bdrv_register(BlockDriver *bdrv)
333{
8a22f02a 334 QLIST_INSERT_HEAD(&bdrv_drivers, bdrv, list);
ea2384d3 335}
b338082b 336
e4e9986b
MA
337BlockDriverState *bdrv_new(void)
338{
339 BlockDriverState *bs;
340 int i;
341
5839e53b 342 bs = g_new0(BlockDriverState, 1);
e4654d2d 343 QLIST_INIT(&bs->dirty_bitmaps);
fbe40ff7
FZ
344 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
345 QLIST_INIT(&bs->op_blockers[i]);
346 }
d616b224 347 notifier_with_return_list_init(&bs->before_write_notifiers);
3783fa3d 348 qemu_co_mutex_init(&bs->reqs_lock);
2119882c 349 qemu_mutex_init(&bs->dirty_bitmap_mutex);
9fcb0251 350 bs->refcnt = 1;
dcd04228 351 bs->aio_context = qemu_get_aio_context();
d7d512f6 352
3ff2f67a
EY
353 qemu_co_queue_init(&bs->flush_queue);
354
0f12264e
KW
355 for (i = 0; i < bdrv_drain_all_count; i++) {
356 bdrv_drained_begin(bs);
357 }
358
2c1d04e0
HR
359 QTAILQ_INSERT_TAIL(&all_bdrv_states, bs, bs_list);
360
b338082b
FB
361 return bs;
362}
363
88d88798 364static BlockDriver *bdrv_do_find_format(const char *format_name)
ea2384d3
FB
365{
366 BlockDriver *drv1;
88d88798 367
8a22f02a
SH
368 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
369 if (!strcmp(drv1->format_name, format_name)) {
ea2384d3 370 return drv1;
8a22f02a 371 }
ea2384d3 372 }
88d88798 373
ea2384d3
FB
374 return NULL;
375}
376
88d88798
MM
377BlockDriver *bdrv_find_format(const char *format_name)
378{
379 BlockDriver *drv1;
380 int i;
381
382 drv1 = bdrv_do_find_format(format_name);
383 if (drv1) {
384 return drv1;
385 }
386
387 /* The driver isn't registered, maybe we need to load a module */
388 for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); ++i) {
389 if (!strcmp(block_driver_modules[i].format_name, format_name)) {
390 block_module_load_one(block_driver_modules[i].library_name);
391 break;
392 }
393 }
394
395 return bdrv_do_find_format(format_name);
396}
397
e8eb8637 398int bdrv_is_whitelisted(BlockDriver *drv, bool read_only)
eb852011 399{
b64ec4e4
FZ
400 static const char *whitelist_rw[] = {
401 CONFIG_BDRV_RW_WHITELIST
402 };
403 static const char *whitelist_ro[] = {
404 CONFIG_BDRV_RO_WHITELIST
eb852011
MA
405 };
406 const char **p;
407
b64ec4e4 408 if (!whitelist_rw[0] && !whitelist_ro[0]) {
eb852011 409 return 1; /* no whitelist, anything goes */
b64ec4e4 410 }
eb852011 411
b64ec4e4 412 for (p = whitelist_rw; *p; p++) {
eb852011
MA
413 if (!strcmp(drv->format_name, *p)) {
414 return 1;
415 }
416 }
b64ec4e4
FZ
417 if (read_only) {
418 for (p = whitelist_ro; *p; p++) {
419 if (!strcmp(drv->format_name, *p)) {
420 return 1;
421 }
422 }
423 }
eb852011
MA
424 return 0;
425}
426
e6ff69bf
DB
427bool bdrv_uses_whitelist(void)
428{
429 return use_bdrv_whitelist;
430}
431
5b7e1542
ZYW
432typedef struct CreateCo {
433 BlockDriver *drv;
434 char *filename;
83d0521a 435 QemuOpts *opts;
5b7e1542 436 int ret;
cc84d90f 437 Error *err;
5b7e1542
ZYW
438} CreateCo;
439
440static void coroutine_fn bdrv_create_co_entry(void *opaque)
441{
cc84d90f
HR
442 Error *local_err = NULL;
443 int ret;
444
5b7e1542
ZYW
445 CreateCo *cco = opaque;
446 assert(cco->drv);
447
efc75e2a 448 ret = cco->drv->bdrv_co_create_opts(cco->filename, cco->opts, &local_err);
621ff94d 449 error_propagate(&cco->err, local_err);
cc84d90f 450 cco->ret = ret;
5b7e1542
ZYW
451}
452
0e7e1989 453int bdrv_create(BlockDriver *drv, const char* filename,
83d0521a 454 QemuOpts *opts, Error **errp)
ea2384d3 455{
5b7e1542
ZYW
456 int ret;
457
458 Coroutine *co;
459 CreateCo cco = {
460 .drv = drv,
461 .filename = g_strdup(filename),
83d0521a 462 .opts = opts,
5b7e1542 463 .ret = NOT_DONE,
cc84d90f 464 .err = NULL,
5b7e1542
ZYW
465 };
466
efc75e2a 467 if (!drv->bdrv_co_create_opts) {
cc84d90f 468 error_setg(errp, "Driver '%s' does not support image creation", drv->format_name);
80168bff
LC
469 ret = -ENOTSUP;
470 goto out;
5b7e1542
ZYW
471 }
472
473 if (qemu_in_coroutine()) {
474 /* Fast-path if already in coroutine context */
475 bdrv_create_co_entry(&cco);
476 } else {
0b8b8753
PB
477 co = qemu_coroutine_create(bdrv_create_co_entry, &cco);
478 qemu_coroutine_enter(co);
5b7e1542 479 while (cco.ret == NOT_DONE) {
b47ec2c4 480 aio_poll(qemu_get_aio_context(), true);
5b7e1542
ZYW
481 }
482 }
483
484 ret = cco.ret;
cc84d90f 485 if (ret < 0) {
84d18f06 486 if (cco.err) {
cc84d90f
HR
487 error_propagate(errp, cco.err);
488 } else {
489 error_setg_errno(errp, -ret, "Could not create image");
490 }
491 }
0e7e1989 492
80168bff
LC
493out:
494 g_free(cco.filename);
5b7e1542 495 return ret;
ea2384d3
FB
496}
497
c282e1fd 498int bdrv_create_file(const char *filename, QemuOpts *opts, Error **errp)
84a12e66
CH
499{
500 BlockDriver *drv;
cc84d90f
HR
501 Error *local_err = NULL;
502 int ret;
84a12e66 503
b65a5e12 504 drv = bdrv_find_protocol(filename, true, errp);
84a12e66 505 if (drv == NULL) {
16905d71 506 return -ENOENT;
84a12e66
CH
507 }
508
c282e1fd 509 ret = bdrv_create(drv, filename, opts, &local_err);
621ff94d 510 error_propagate(errp, local_err);
cc84d90f 511 return ret;
84a12e66
CH
512}
513
892b7de8
ET
514/**
515 * Try to get @bs's logical and physical block size.
516 * On success, store them in @bsz struct and return 0.
517 * On failure return -errno.
518 * @bs must not be empty.
519 */
520int bdrv_probe_blocksizes(BlockDriverState *bs, BlockSizes *bsz)
521{
522 BlockDriver *drv = bs->drv;
523
524 if (drv && drv->bdrv_probe_blocksizes) {
525 return drv->bdrv_probe_blocksizes(bs, bsz);
5a612c00
MP
526 } else if (drv && drv->is_filter && bs->file) {
527 return bdrv_probe_blocksizes(bs->file->bs, bsz);
892b7de8
ET
528 }
529
530 return -ENOTSUP;
531}
532
533/**
534 * Try to get @bs's geometry (cyls, heads, sectors).
535 * On success, store them in @geo struct and return 0.
536 * On failure return -errno.
537 * @bs must not be empty.
538 */
539int bdrv_probe_geometry(BlockDriverState *bs, HDGeometry *geo)
540{
541 BlockDriver *drv = bs->drv;
542
543 if (drv && drv->bdrv_probe_geometry) {
544 return drv->bdrv_probe_geometry(bs, geo);
5a612c00
MP
545 } else if (drv && drv->is_filter && bs->file) {
546 return bdrv_probe_geometry(bs->file->bs, geo);
892b7de8
ET
547 }
548
549 return -ENOTSUP;
550}
551
eba25057
JM
552/*
553 * Create a uniquely-named empty temporary file.
554 * Return 0 upon success, otherwise a negative errno value.
555 */
556int get_tmp_filename(char *filename, int size)
d5249393 557{
eba25057 558#ifdef _WIN32
3b9f94e1 559 char temp_dir[MAX_PATH];
eba25057
JM
560 /* GetTempFileName requires that its output buffer (4th param)
561 have length MAX_PATH or greater. */
562 assert(size >= MAX_PATH);
563 return (GetTempPath(MAX_PATH, temp_dir)
564 && GetTempFileName(temp_dir, "qem", 0, filename)
565 ? 0 : -GetLastError());
d5249393 566#else
67b915a5 567 int fd;
7ccfb2eb 568 const char *tmpdir;
0badc1ee 569 tmpdir = getenv("TMPDIR");
69bef793
AS
570 if (!tmpdir) {
571 tmpdir = "/var/tmp";
572 }
eba25057
JM
573 if (snprintf(filename, size, "%s/vl.XXXXXX", tmpdir) >= size) {
574 return -EOVERFLOW;
575 }
ea2384d3 576 fd = mkstemp(filename);
fe235a06
DH
577 if (fd < 0) {
578 return -errno;
579 }
580 if (close(fd) != 0) {
581 unlink(filename);
eba25057
JM
582 return -errno;
583 }
584 return 0;
d5249393 585#endif
eba25057 586}
fc01f7e7 587
84a12e66
CH
588/*
589 * Detect host devices. By convention, /dev/cdrom[N] is always
590 * recognized as a host CDROM.
591 */
592static BlockDriver *find_hdev_driver(const char *filename)
593{
594 int score_max = 0, score;
595 BlockDriver *drv = NULL, *d;
596
597 QLIST_FOREACH(d, &bdrv_drivers, list) {
598 if (d->bdrv_probe_device) {
599 score = d->bdrv_probe_device(filename);
600 if (score > score_max) {
601 score_max = score;
602 drv = d;
603 }
604 }
605 }
606
607 return drv;
608}
609
88d88798
MM
610static BlockDriver *bdrv_do_find_protocol(const char *protocol)
611{
612 BlockDriver *drv1;
613
614 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
615 if (drv1->protocol_name && !strcmp(drv1->protocol_name, protocol)) {
616 return drv1;
617 }
618 }
619
620 return NULL;
621}
622
98289620 623BlockDriver *bdrv_find_protocol(const char *filename,
b65a5e12
HR
624 bool allow_protocol_prefix,
625 Error **errp)
83f64091
FB
626{
627 BlockDriver *drv1;
628 char protocol[128];
1cec71e3 629 int len;
83f64091 630 const char *p;
88d88798 631 int i;
19cb3738 632
66f82cee
KW
633 /* TODO Drivers without bdrv_file_open must be specified explicitly */
634
39508e7a
CH
635 /*
636 * XXX(hch): we really should not let host device detection
637 * override an explicit protocol specification, but moving this
638 * later breaks access to device names with colons in them.
639 * Thanks to the brain-dead persistent naming schemes on udev-
640 * based Linux systems those actually are quite common.
641 */
642 drv1 = find_hdev_driver(filename);
643 if (drv1) {
644 return drv1;
645 }
646
98289620 647 if (!path_has_protocol(filename) || !allow_protocol_prefix) {
ef810437 648 return &bdrv_file;
84a12e66 649 }
98289620 650
9e0b22f4
SH
651 p = strchr(filename, ':');
652 assert(p != NULL);
1cec71e3
AL
653 len = p - filename;
654 if (len > sizeof(protocol) - 1)
655 len = sizeof(protocol) - 1;
656 memcpy(protocol, filename, len);
657 protocol[len] = '\0';
88d88798
MM
658
659 drv1 = bdrv_do_find_protocol(protocol);
660 if (drv1) {
661 return drv1;
662 }
663
664 for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); ++i) {
665 if (block_driver_modules[i].protocol_name &&
666 !strcmp(block_driver_modules[i].protocol_name, protocol)) {
667 block_module_load_one(block_driver_modules[i].library_name);
668 break;
8a22f02a 669 }
83f64091 670 }
b65a5e12 671
88d88798
MM
672 drv1 = bdrv_do_find_protocol(protocol);
673 if (!drv1) {
674 error_setg(errp, "Unknown protocol '%s'", protocol);
675 }
676 return drv1;
83f64091
FB
677}
678
c6684249
MA
679/*
680 * Guess image format by probing its contents.
681 * This is not a good idea when your image is raw (CVE-2008-2004), but
682 * we do it anyway for backward compatibility.
683 *
684 * @buf contains the image's first @buf_size bytes.
7cddd372
KW
685 * @buf_size is the buffer size in bytes (generally BLOCK_PROBE_BUF_SIZE,
686 * but can be smaller if the image file is smaller)
c6684249
MA
687 * @filename is its filename.
688 *
689 * For all block drivers, call the bdrv_probe() method to get its
690 * probing score.
691 * Return the first block driver with the highest probing score.
692 */
38f3ef57
KW
693BlockDriver *bdrv_probe_all(const uint8_t *buf, int buf_size,
694 const char *filename)
c6684249
MA
695{
696 int score_max = 0, score;
697 BlockDriver *drv = NULL, *d;
698
699 QLIST_FOREACH(d, &bdrv_drivers, list) {
700 if (d->bdrv_probe) {
701 score = d->bdrv_probe(buf, buf_size, filename);
702 if (score > score_max) {
703 score_max = score;
704 drv = d;
705 }
706 }
707 }
708
709 return drv;
710}
711
5696c6e3 712static int find_image_format(BlockBackend *file, const char *filename,
34b5d2c6 713 BlockDriver **pdrv, Error **errp)
f3a5d3f8 714{
c6684249 715 BlockDriver *drv;
7cddd372 716 uint8_t buf[BLOCK_PROBE_BUF_SIZE];
f500a6d3 717 int ret = 0;
f8ea0b00 718
08a00559 719 /* Return the raw BlockDriver * to scsi-generic devices or empty drives */
5696c6e3 720 if (blk_is_sg(file) || !blk_is_inserted(file) || blk_getlength(file) == 0) {
ef810437 721 *pdrv = &bdrv_raw;
c98ac35d 722 return ret;
1a396859 723 }
f8ea0b00 724
5696c6e3 725 ret = blk_pread(file, 0, buf, sizeof(buf));
83f64091 726 if (ret < 0) {
34b5d2c6
HR
727 error_setg_errno(errp, -ret, "Could not read image for determining its "
728 "format");
c98ac35d
SW
729 *pdrv = NULL;
730 return ret;
83f64091
FB
731 }
732
c6684249 733 drv = bdrv_probe_all(buf, ret, filename);
c98ac35d 734 if (!drv) {
34b5d2c6
HR
735 error_setg(errp, "Could not determine image format: No compatible "
736 "driver found");
c98ac35d
SW
737 ret = -ENOENT;
738 }
739 *pdrv = drv;
740 return ret;
ea2384d3
FB
741}
742
51762288
SH
743/**
744 * Set the current 'total_sectors' value
65a9bb25 745 * Return 0 on success, -errno on error.
51762288 746 */
3d9f2d2a 747int refresh_total_sectors(BlockDriverState *bs, int64_t hint)
51762288
SH
748{
749 BlockDriver *drv = bs->drv;
750
d470ad42
HR
751 if (!drv) {
752 return -ENOMEDIUM;
753 }
754
396759ad 755 /* Do not attempt drv->bdrv_getlength() on scsi-generic devices */
b192af8a 756 if (bdrv_is_sg(bs))
396759ad
NB
757 return 0;
758
51762288
SH
759 /* query actual device if possible, otherwise just trust the hint */
760 if (drv->bdrv_getlength) {
761 int64_t length = drv->bdrv_getlength(bs);
762 if (length < 0) {
763 return length;
764 }
7e382003 765 hint = DIV_ROUND_UP(length, BDRV_SECTOR_SIZE);
51762288
SH
766 }
767
768 bs->total_sectors = hint;
769 return 0;
770}
771
cddff5ba
KW
772/**
773 * Combines a QDict of new block driver @options with any missing options taken
774 * from @old_options, so that leaving out an option defaults to its old value.
775 */
776static void bdrv_join_options(BlockDriverState *bs, QDict *options,
777 QDict *old_options)
778{
779 if (bs->drv && bs->drv->bdrv_join_options) {
780 bs->drv->bdrv_join_options(options, old_options);
781 } else {
782 qdict_join(options, old_options, false);
783 }
784}
785
543770bd
AG
786static BlockdevDetectZeroesOptions bdrv_parse_detect_zeroes(QemuOpts *opts,
787 int open_flags,
788 Error **errp)
789{
790 Error *local_err = NULL;
791 char *value = qemu_opt_get_del(opts, "detect-zeroes");
792 BlockdevDetectZeroesOptions detect_zeroes =
793 qapi_enum_parse(&BlockdevDetectZeroesOptions_lookup, value,
794 BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF, &local_err);
795 g_free(value);
796 if (local_err) {
797 error_propagate(errp, local_err);
798 return detect_zeroes;
799 }
800
801 if (detect_zeroes == BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP &&
802 !(open_flags & BDRV_O_UNMAP))
803 {
804 error_setg(errp, "setting detect-zeroes to unmap is not allowed "
805 "without setting discard operation to unmap");
806 }
807
808 return detect_zeroes;
809}
810
9e8f1835
PB
811/**
812 * Set open flags for a given discard mode
813 *
814 * Return 0 on success, -1 if the discard mode was invalid.
815 */
816int bdrv_parse_discard_flags(const char *mode, int *flags)
817{
818 *flags &= ~BDRV_O_UNMAP;
819
820 if (!strcmp(mode, "off") || !strcmp(mode, "ignore")) {
821 /* do nothing */
822 } else if (!strcmp(mode, "on") || !strcmp(mode, "unmap")) {
823 *flags |= BDRV_O_UNMAP;
824 } else {
825 return -1;
826 }
827
828 return 0;
829}
830
c3993cdc
SH
831/**
832 * Set open flags for a given cache mode
833 *
834 * Return 0 on success, -1 if the cache mode was invalid.
835 */
53e8ae01 836int bdrv_parse_cache_mode(const char *mode, int *flags, bool *writethrough)
c3993cdc
SH
837{
838 *flags &= ~BDRV_O_CACHE_MASK;
839
840 if (!strcmp(mode, "off") || !strcmp(mode, "none")) {
53e8ae01
KW
841 *writethrough = false;
842 *flags |= BDRV_O_NOCACHE;
92196b2f 843 } else if (!strcmp(mode, "directsync")) {
53e8ae01 844 *writethrough = true;
92196b2f 845 *flags |= BDRV_O_NOCACHE;
c3993cdc 846 } else if (!strcmp(mode, "writeback")) {
53e8ae01 847 *writethrough = false;
c3993cdc 848 } else if (!strcmp(mode, "unsafe")) {
53e8ae01 849 *writethrough = false;
c3993cdc
SH
850 *flags |= BDRV_O_NO_FLUSH;
851 } else if (!strcmp(mode, "writethrough")) {
53e8ae01 852 *writethrough = true;
c3993cdc
SH
853 } else {
854 return -1;
855 }
856
857 return 0;
858}
859
b5411555
KW
860static char *bdrv_child_get_parent_desc(BdrvChild *c)
861{
862 BlockDriverState *parent = c->opaque;
863 return g_strdup(bdrv_get_device_or_node_name(parent));
864}
865
20018e12
KW
866static void bdrv_child_cb_drained_begin(BdrvChild *child)
867{
868 BlockDriverState *bs = child->opaque;
6cd5c9d7 869 bdrv_do_drained_begin_quiesce(bs, NULL, false);
20018e12
KW
870}
871
89bd0305
KW
872static bool bdrv_child_cb_drained_poll(BdrvChild *child)
873{
874 BlockDriverState *bs = child->opaque;
6cd5c9d7 875 return bdrv_drain_poll(bs, false, NULL, false);
89bd0305
KW
876}
877
20018e12
KW
878static void bdrv_child_cb_drained_end(BdrvChild *child)
879{
880 BlockDriverState *bs = child->opaque;
881 bdrv_drained_end(bs);
882}
883
d736f119
KW
884static void bdrv_child_cb_attach(BdrvChild *child)
885{
886 BlockDriverState *bs = child->opaque;
887 bdrv_apply_subtree_drain(child, bs);
888}
889
890static void bdrv_child_cb_detach(BdrvChild *child)
891{
892 BlockDriverState *bs = child->opaque;
893 bdrv_unapply_subtree_drain(child, bs);
894}
895
38701b6a
KW
896static int bdrv_child_cb_inactivate(BdrvChild *child)
897{
898 BlockDriverState *bs = child->opaque;
899 assert(bs->open_flags & BDRV_O_INACTIVE);
900 return 0;
901}
902
b1e6fc08 903/*
73176bee
KW
904 * Returns the options and flags that a temporary snapshot should get, based on
905 * the originally requested flags (the originally requested image will have
906 * flags like a backing file)
b1e6fc08 907 */
73176bee
KW
908static void bdrv_temp_snapshot_options(int *child_flags, QDict *child_options,
909 int parent_flags, QDict *parent_options)
b1e6fc08 910{
73176bee
KW
911 *child_flags = (parent_flags & ~BDRV_O_SNAPSHOT) | BDRV_O_TEMPORARY;
912
913 /* For temporary files, unconditional cache=unsafe is fine */
73176bee
KW
914 qdict_set_default_str(child_options, BDRV_OPT_CACHE_DIRECT, "off");
915 qdict_set_default_str(child_options, BDRV_OPT_CACHE_NO_FLUSH, "on");
41869044 916
f87a0e29
AG
917 /* Copy the read-only option from the parent */
918 qdict_copy_default(child_options, parent_options, BDRV_OPT_READ_ONLY);
919
41869044
KW
920 /* aio=native doesn't work for cache.direct=off, so disable it for the
921 * temporary snapshot */
922 *child_flags &= ~BDRV_O_NATIVE_AIO;
b1e6fc08
KW
923}
924
0b50cc88 925/*
8e2160e2
KW
926 * Returns the options and flags that bs->file should get if a protocol driver
927 * is expected, based on the given options and flags for the parent BDS
0b50cc88 928 */
8e2160e2
KW
929static void bdrv_inherited_options(int *child_flags, QDict *child_options,
930 int parent_flags, QDict *parent_options)
0b50cc88 931{
8e2160e2
KW
932 int flags = parent_flags;
933
0b50cc88
KW
934 /* Enable protocol handling, disable format probing for bs->file */
935 flags |= BDRV_O_PROTOCOL;
936
91a097e7
KW
937 /* If the cache mode isn't explicitly set, inherit direct and no-flush from
938 * the parent. */
939 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_DIRECT);
940 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_NO_FLUSH);
5a9347c6 941 qdict_copy_default(child_options, parent_options, BDRV_OPT_FORCE_SHARE);
91a097e7 942
f87a0e29
AG
943 /* Inherit the read-only option from the parent if it's not set */
944 qdict_copy_default(child_options, parent_options, BDRV_OPT_READ_ONLY);
e35bdc12 945 qdict_copy_default(child_options, parent_options, BDRV_OPT_AUTO_READ_ONLY);
f87a0e29 946
0b50cc88 947 /* Our block drivers take care to send flushes and respect unmap policy,
91a097e7
KW
948 * so we can default to enable both on lower layers regardless of the
949 * corresponding parent options. */
818584a4 950 qdict_set_default_str(child_options, BDRV_OPT_DISCARD, "unmap");
0b50cc88 951
0b50cc88 952 /* Clear flags that only apply to the top layer */
abb06c5a
DB
953 flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_COPY_ON_READ |
954 BDRV_O_NO_IO);
0b50cc88 955
8e2160e2 956 *child_flags = flags;
0b50cc88
KW
957}
958
f3930ed0 959const BdrvChildRole child_file = {
6cd5c9d7 960 .parent_is_bds = true,
b5411555 961 .get_parent_desc = bdrv_child_get_parent_desc,
8e2160e2 962 .inherit_options = bdrv_inherited_options,
20018e12 963 .drained_begin = bdrv_child_cb_drained_begin,
89bd0305 964 .drained_poll = bdrv_child_cb_drained_poll,
20018e12 965 .drained_end = bdrv_child_cb_drained_end,
d736f119
KW
966 .attach = bdrv_child_cb_attach,
967 .detach = bdrv_child_cb_detach,
38701b6a 968 .inactivate = bdrv_child_cb_inactivate,
f3930ed0
KW
969};
970
971/*
8e2160e2
KW
972 * Returns the options and flags that bs->file should get if the use of formats
973 * (and not only protocols) is permitted for it, based on the given options and
974 * flags for the parent BDS
f3930ed0 975 */
8e2160e2
KW
976static void bdrv_inherited_fmt_options(int *child_flags, QDict *child_options,
977 int parent_flags, QDict *parent_options)
f3930ed0 978{
8e2160e2
KW
979 child_file.inherit_options(child_flags, child_options,
980 parent_flags, parent_options);
981
abb06c5a 982 *child_flags &= ~(BDRV_O_PROTOCOL | BDRV_O_NO_IO);
f3930ed0
KW
983}
984
985const BdrvChildRole child_format = {
6cd5c9d7 986 .parent_is_bds = true,
b5411555 987 .get_parent_desc = bdrv_child_get_parent_desc,
8e2160e2 988 .inherit_options = bdrv_inherited_fmt_options,
20018e12 989 .drained_begin = bdrv_child_cb_drained_begin,
89bd0305 990 .drained_poll = bdrv_child_cb_drained_poll,
20018e12 991 .drained_end = bdrv_child_cb_drained_end,
d736f119
KW
992 .attach = bdrv_child_cb_attach,
993 .detach = bdrv_child_cb_detach,
38701b6a 994 .inactivate = bdrv_child_cb_inactivate,
f3930ed0
KW
995};
996
db95dbba
KW
997static void bdrv_backing_attach(BdrvChild *c)
998{
999 BlockDriverState *parent = c->opaque;
1000 BlockDriverState *backing_hd = c->bs;
1001
1002 assert(!parent->backing_blocker);
1003 error_setg(&parent->backing_blocker,
1004 "node is used as backing hd of '%s'",
1005 bdrv_get_device_or_node_name(parent));
1006
1007 parent->open_flags &= ~BDRV_O_NO_BACKING;
1008 pstrcpy(parent->backing_file, sizeof(parent->backing_file),
1009 backing_hd->filename);
1010 pstrcpy(parent->backing_format, sizeof(parent->backing_format),
1011 backing_hd->drv ? backing_hd->drv->format_name : "");
1012
1013 bdrv_op_block_all(backing_hd, parent->backing_blocker);
1014 /* Otherwise we won't be able to commit or stream */
1015 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_COMMIT_TARGET,
1016 parent->backing_blocker);
1017 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_STREAM,
1018 parent->backing_blocker);
1019 /*
1020 * We do backup in 3 ways:
1021 * 1. drive backup
1022 * The target bs is new opened, and the source is top BDS
1023 * 2. blockdev backup
1024 * Both the source and the target are top BDSes.
1025 * 3. internal backup(used for block replication)
1026 * Both the source and the target are backing file
1027 *
1028 * In case 1 and 2, neither the source nor the target is the backing file.
1029 * In case 3, we will block the top BDS, so there is only one block job
1030 * for the top BDS and its backing chain.
1031 */
1032 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_BACKUP_SOURCE,
1033 parent->backing_blocker);
1034 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_BACKUP_TARGET,
1035 parent->backing_blocker);
d736f119
KW
1036
1037 bdrv_child_cb_attach(c);
db95dbba
KW
1038}
1039
1040static void bdrv_backing_detach(BdrvChild *c)
1041{
1042 BlockDriverState *parent = c->opaque;
1043
1044 assert(parent->backing_blocker);
1045 bdrv_op_unblock_all(c->bs, parent->backing_blocker);
1046 error_free(parent->backing_blocker);
1047 parent->backing_blocker = NULL;
d736f119
KW
1048
1049 bdrv_child_cb_detach(c);
db95dbba
KW
1050}
1051
317fc44e 1052/*
8e2160e2
KW
1053 * Returns the options and flags that bs->backing should get, based on the
1054 * given options and flags for the parent BDS
317fc44e 1055 */
8e2160e2
KW
1056static void bdrv_backing_options(int *child_flags, QDict *child_options,
1057 int parent_flags, QDict *parent_options)
317fc44e 1058{
8e2160e2
KW
1059 int flags = parent_flags;
1060
b8816a43
KW
1061 /* The cache mode is inherited unmodified for backing files; except WCE,
1062 * which is only applied on the top level (BlockBackend) */
91a097e7
KW
1063 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_DIRECT);
1064 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_NO_FLUSH);
5a9347c6 1065 qdict_copy_default(child_options, parent_options, BDRV_OPT_FORCE_SHARE);
91a097e7 1066
317fc44e 1067 /* backing files always opened read-only */
f87a0e29 1068 qdict_set_default_str(child_options, BDRV_OPT_READ_ONLY, "on");
e35bdc12 1069 qdict_set_default_str(child_options, BDRV_OPT_AUTO_READ_ONLY, "off");
f87a0e29 1070 flags &= ~BDRV_O_COPY_ON_READ;
317fc44e
KW
1071
1072 /* snapshot=on is handled on the top layer */
8bfea15d 1073 flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_TEMPORARY);
317fc44e 1074
8e2160e2 1075 *child_flags = flags;
317fc44e
KW
1076}
1077
6858eba0
KW
1078static int bdrv_backing_update_filename(BdrvChild *c, BlockDriverState *base,
1079 const char *filename, Error **errp)
1080{
1081 BlockDriverState *parent = c->opaque;
e94d3dba 1082 bool read_only = bdrv_is_read_only(parent);
6858eba0
KW
1083 int ret;
1084
e94d3dba
AG
1085 if (read_only) {
1086 ret = bdrv_reopen_set_read_only(parent, false, errp);
61f09cea
KW
1087 if (ret < 0) {
1088 return ret;
1089 }
1090 }
1091
6858eba0
KW
1092 ret = bdrv_change_backing_file(parent, filename,
1093 base->drv ? base->drv->format_name : "");
1094 if (ret < 0) {
64730694 1095 error_setg_errno(errp, -ret, "Could not update backing file link");
6858eba0
KW
1096 }
1097
e94d3dba
AG
1098 if (read_only) {
1099 bdrv_reopen_set_read_only(parent, true, NULL);
61f09cea
KW
1100 }
1101
6858eba0
KW
1102 return ret;
1103}
1104
91ef3825 1105const BdrvChildRole child_backing = {
6cd5c9d7 1106 .parent_is_bds = true,
b5411555 1107 .get_parent_desc = bdrv_child_get_parent_desc,
db95dbba
KW
1108 .attach = bdrv_backing_attach,
1109 .detach = bdrv_backing_detach,
8e2160e2 1110 .inherit_options = bdrv_backing_options,
20018e12 1111 .drained_begin = bdrv_child_cb_drained_begin,
89bd0305 1112 .drained_poll = bdrv_child_cb_drained_poll,
20018e12 1113 .drained_end = bdrv_child_cb_drained_end,
38701b6a 1114 .inactivate = bdrv_child_cb_inactivate,
6858eba0 1115 .update_filename = bdrv_backing_update_filename,
f3930ed0
KW
1116};
1117
7b272452
KW
1118static int bdrv_open_flags(BlockDriverState *bs, int flags)
1119{
61de4c68 1120 int open_flags = flags;
7b272452
KW
1121
1122 /*
1123 * Clear flags that are internal to the block layer before opening the
1124 * image.
1125 */
20cca275 1126 open_flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_PROTOCOL);
7b272452
KW
1127
1128 /*
1129 * Snapshots should be writable.
1130 */
8bfea15d 1131 if (flags & BDRV_O_TEMPORARY) {
7b272452
KW
1132 open_flags |= BDRV_O_RDWR;
1133 }
1134
1135 return open_flags;
1136}
1137
91a097e7
KW
1138static void update_flags_from_options(int *flags, QemuOpts *opts)
1139{
2a3d4331 1140 *flags &= ~(BDRV_O_CACHE_MASK | BDRV_O_RDWR | BDRV_O_AUTO_RDONLY);
91a097e7 1141
57f9db9a 1142 if (qemu_opt_get_bool_del(opts, BDRV_OPT_CACHE_NO_FLUSH, false)) {
91a097e7
KW
1143 *flags |= BDRV_O_NO_FLUSH;
1144 }
1145
57f9db9a 1146 if (qemu_opt_get_bool_del(opts, BDRV_OPT_CACHE_DIRECT, false)) {
91a097e7
KW
1147 *flags |= BDRV_O_NOCACHE;
1148 }
f87a0e29 1149
57f9db9a 1150 if (!qemu_opt_get_bool_del(opts, BDRV_OPT_READ_ONLY, false)) {
f87a0e29
AG
1151 *flags |= BDRV_O_RDWR;
1152 }
1153
e35bdc12
KW
1154 if (qemu_opt_get_bool_del(opts, BDRV_OPT_AUTO_READ_ONLY, false)) {
1155 *flags |= BDRV_O_AUTO_RDONLY;
1156 }
91a097e7
KW
1157}
1158
1159static void update_options_from_flags(QDict *options, int flags)
1160{
91a097e7 1161 if (!qdict_haskey(options, BDRV_OPT_CACHE_DIRECT)) {
46f5ac20 1162 qdict_put_bool(options, BDRV_OPT_CACHE_DIRECT, flags & BDRV_O_NOCACHE);
91a097e7
KW
1163 }
1164 if (!qdict_haskey(options, BDRV_OPT_CACHE_NO_FLUSH)) {
46f5ac20
EB
1165 qdict_put_bool(options, BDRV_OPT_CACHE_NO_FLUSH,
1166 flags & BDRV_O_NO_FLUSH);
91a097e7 1167 }
f87a0e29 1168 if (!qdict_haskey(options, BDRV_OPT_READ_ONLY)) {
46f5ac20 1169 qdict_put_bool(options, BDRV_OPT_READ_ONLY, !(flags & BDRV_O_RDWR));
f87a0e29 1170 }
e35bdc12
KW
1171 if (!qdict_haskey(options, BDRV_OPT_AUTO_READ_ONLY)) {
1172 qdict_put_bool(options, BDRV_OPT_AUTO_READ_ONLY,
1173 flags & BDRV_O_AUTO_RDONLY);
1174 }
91a097e7
KW
1175}
1176
636ea370
KW
1177static void bdrv_assign_node_name(BlockDriverState *bs,
1178 const char *node_name,
1179 Error **errp)
6913c0c2 1180{
15489c76 1181 char *gen_node_name = NULL;
6913c0c2 1182
15489c76
JC
1183 if (!node_name) {
1184 node_name = gen_node_name = id_generate(ID_BLOCK);
1185 } else if (!id_wellformed(node_name)) {
1186 /*
1187 * Check for empty string or invalid characters, but not if it is
1188 * generated (generated names use characters not available to the user)
1189 */
9aebf3b8 1190 error_setg(errp, "Invalid node name");
636ea370 1191 return;
6913c0c2
BC
1192 }
1193
0c5e94ee 1194 /* takes care of avoiding namespaces collisions */
7f06d47e 1195 if (blk_by_name(node_name)) {
0c5e94ee
BC
1196 error_setg(errp, "node-name=%s is conflicting with a device id",
1197 node_name);
15489c76 1198 goto out;
0c5e94ee
BC
1199 }
1200
6913c0c2
BC
1201 /* takes care of avoiding duplicates node names */
1202 if (bdrv_find_node(node_name)) {
1203 error_setg(errp, "Duplicate node name");
15489c76 1204 goto out;
6913c0c2
BC
1205 }
1206
824808dd
KW
1207 /* Make sure that the node name isn't truncated */
1208 if (strlen(node_name) >= sizeof(bs->node_name)) {
1209 error_setg(errp, "Node name too long");
1210 goto out;
1211 }
1212
6913c0c2
BC
1213 /* copy node name into the bs and insert it into the graph list */
1214 pstrcpy(bs->node_name, sizeof(bs->node_name), node_name);
1215 QTAILQ_INSERT_TAIL(&graph_bdrv_states, bs, node_list);
15489c76
JC
1216out:
1217 g_free(gen_node_name);
6913c0c2
BC
1218}
1219
01a56501
KW
1220static int bdrv_open_driver(BlockDriverState *bs, BlockDriver *drv,
1221 const char *node_name, QDict *options,
1222 int open_flags, Error **errp)
1223{
1224 Error *local_err = NULL;
0f12264e 1225 int i, ret;
01a56501
KW
1226
1227 bdrv_assign_node_name(bs, node_name, &local_err);
1228 if (local_err) {
1229 error_propagate(errp, local_err);
1230 return -EINVAL;
1231 }
1232
1233 bs->drv = drv;
680c7f96 1234 bs->read_only = !(bs->open_flags & BDRV_O_RDWR);
01a56501
KW
1235 bs->opaque = g_malloc0(drv->instance_size);
1236
1237 if (drv->bdrv_file_open) {
1238 assert(!drv->bdrv_needs_filename || bs->filename[0]);
1239 ret = drv->bdrv_file_open(bs, options, open_flags, &local_err);
680c7f96 1240 } else if (drv->bdrv_open) {
01a56501 1241 ret = drv->bdrv_open(bs, options, open_flags, &local_err);
680c7f96
KW
1242 } else {
1243 ret = 0;
01a56501
KW
1244 }
1245
1246 if (ret < 0) {
1247 if (local_err) {
1248 error_propagate(errp, local_err);
1249 } else if (bs->filename[0]) {
1250 error_setg_errno(errp, -ret, "Could not open '%s'", bs->filename);
1251 } else {
1252 error_setg_errno(errp, -ret, "Could not open image");
1253 }
180ca19a 1254 goto open_failed;
01a56501
KW
1255 }
1256
1257 ret = refresh_total_sectors(bs, bs->total_sectors);
1258 if (ret < 0) {
1259 error_setg_errno(errp, -ret, "Could not refresh total sector count");
180ca19a 1260 return ret;
01a56501
KW
1261 }
1262
1263 bdrv_refresh_limits(bs, &local_err);
1264 if (local_err) {
1265 error_propagate(errp, local_err);
180ca19a 1266 return -EINVAL;
01a56501
KW
1267 }
1268
1269 assert(bdrv_opt_mem_align(bs) != 0);
1270 assert(bdrv_min_mem_align(bs) != 0);
1271 assert(is_power_of_2(bs->bl.request_alignment));
1272
0f12264e
KW
1273 for (i = 0; i < bs->quiesce_counter; i++) {
1274 if (drv->bdrv_co_drain_begin) {
1275 drv->bdrv_co_drain_begin(bs);
1276 }
1277 }
1278
01a56501 1279 return 0;
180ca19a
MP
1280open_failed:
1281 bs->drv = NULL;
1282 if (bs->file != NULL) {
1283 bdrv_unref_child(bs, bs->file);
1284 bs->file = NULL;
1285 }
01a56501
KW
1286 g_free(bs->opaque);
1287 bs->opaque = NULL;
01a56501
KW
1288 return ret;
1289}
1290
680c7f96
KW
1291BlockDriverState *bdrv_new_open_driver(BlockDriver *drv, const char *node_name,
1292 int flags, Error **errp)
1293{
1294 BlockDriverState *bs;
1295 int ret;
1296
1297 bs = bdrv_new();
1298 bs->open_flags = flags;
1299 bs->explicit_options = qdict_new();
1300 bs->options = qdict_new();
1301 bs->opaque = NULL;
1302
1303 update_options_from_flags(bs->options, flags);
1304
1305 ret = bdrv_open_driver(bs, drv, node_name, bs->options, flags, errp);
1306 if (ret < 0) {
cb3e7f08 1307 qobject_unref(bs->explicit_options);
180ca19a 1308 bs->explicit_options = NULL;
cb3e7f08 1309 qobject_unref(bs->options);
180ca19a 1310 bs->options = NULL;
680c7f96
KW
1311 bdrv_unref(bs);
1312 return NULL;
1313 }
1314
1315 return bs;
1316}
1317
c5f3014b 1318QemuOptsList bdrv_runtime_opts = {
18edf289
KW
1319 .name = "bdrv_common",
1320 .head = QTAILQ_HEAD_INITIALIZER(bdrv_runtime_opts.head),
1321 .desc = {
1322 {
1323 .name = "node-name",
1324 .type = QEMU_OPT_STRING,
1325 .help = "Node name of the block device node",
1326 },
62392ebb
KW
1327 {
1328 .name = "driver",
1329 .type = QEMU_OPT_STRING,
1330 .help = "Block driver to use for the node",
1331 },
91a097e7
KW
1332 {
1333 .name = BDRV_OPT_CACHE_DIRECT,
1334 .type = QEMU_OPT_BOOL,
1335 .help = "Bypass software writeback cache on the host",
1336 },
1337 {
1338 .name = BDRV_OPT_CACHE_NO_FLUSH,
1339 .type = QEMU_OPT_BOOL,
1340 .help = "Ignore flush requests",
1341 },
f87a0e29
AG
1342 {
1343 .name = BDRV_OPT_READ_ONLY,
1344 .type = QEMU_OPT_BOOL,
1345 .help = "Node is opened in read-only mode",
1346 },
e35bdc12
KW
1347 {
1348 .name = BDRV_OPT_AUTO_READ_ONLY,
1349 .type = QEMU_OPT_BOOL,
1350 .help = "Node can become read-only if opening read-write fails",
1351 },
692e01a2
KW
1352 {
1353 .name = "detect-zeroes",
1354 .type = QEMU_OPT_STRING,
1355 .help = "try to optimize zero writes (off, on, unmap)",
1356 },
818584a4 1357 {
415bbca8 1358 .name = BDRV_OPT_DISCARD,
818584a4
KW
1359 .type = QEMU_OPT_STRING,
1360 .help = "discard operation (ignore/off, unmap/on)",
1361 },
5a9347c6
FZ
1362 {
1363 .name = BDRV_OPT_FORCE_SHARE,
1364 .type = QEMU_OPT_BOOL,
1365 .help = "always accept other writers (default: off)",
1366 },
18edf289
KW
1367 { /* end of list */ }
1368 },
1369};
1370
57915332
KW
1371/*
1372 * Common part for opening disk images and files
b6ad491a
KW
1373 *
1374 * Removes all processed options from *options.
57915332 1375 */
5696c6e3 1376static int bdrv_open_common(BlockDriverState *bs, BlockBackend *file,
82dc8b41 1377 QDict *options, Error **errp)
57915332
KW
1378{
1379 int ret, open_flags;
035fccdf 1380 const char *filename;
62392ebb 1381 const char *driver_name = NULL;
6913c0c2 1382 const char *node_name = NULL;
818584a4 1383 const char *discard;
18edf289 1384 QemuOpts *opts;
62392ebb 1385 BlockDriver *drv;
34b5d2c6 1386 Error *local_err = NULL;
57915332 1387
6405875c 1388 assert(bs->file == NULL);
707ff828 1389 assert(options != NULL && bs->options != options);
57915332 1390
62392ebb
KW
1391 opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
1392 qemu_opts_absorb_qdict(opts, options, &local_err);
1393 if (local_err) {
1394 error_propagate(errp, local_err);
1395 ret = -EINVAL;
1396 goto fail_opts;
1397 }
1398
9b7e8691
AG
1399 update_flags_from_options(&bs->open_flags, opts);
1400
62392ebb
KW
1401 driver_name = qemu_opt_get(opts, "driver");
1402 drv = bdrv_find_format(driver_name);
1403 assert(drv != NULL);
1404
5a9347c6
FZ
1405 bs->force_share = qemu_opt_get_bool(opts, BDRV_OPT_FORCE_SHARE, false);
1406
1407 if (bs->force_share && (bs->open_flags & BDRV_O_RDWR)) {
1408 error_setg(errp,
1409 BDRV_OPT_FORCE_SHARE
1410 "=on can only be used with read-only images");
1411 ret = -EINVAL;
1412 goto fail_opts;
1413 }
1414
45673671 1415 if (file != NULL) {
5696c6e3 1416 filename = blk_bs(file)->filename;
45673671 1417 } else {
129c7d1c
MA
1418 /*
1419 * Caution: while qdict_get_try_str() is fine, getting
1420 * non-string types would require more care. When @options
1421 * come from -blockdev or blockdev_add, its members are typed
1422 * according to the QAPI schema, but when they come from
1423 * -drive, they're all QString.
1424 */
45673671
KW
1425 filename = qdict_get_try_str(options, "filename");
1426 }
1427
4a008240 1428 if (drv->bdrv_needs_filename && (!filename || !filename[0])) {
765003db
KW
1429 error_setg(errp, "The '%s' block driver requires a file name",
1430 drv->format_name);
18edf289
KW
1431 ret = -EINVAL;
1432 goto fail_opts;
6913c0c2 1433 }
6913c0c2 1434
82dc8b41
KW
1435 trace_bdrv_open_common(bs, filename ?: "", bs->open_flags,
1436 drv->format_name);
62392ebb 1437
82dc8b41 1438 bs->read_only = !(bs->open_flags & BDRV_O_RDWR);
b64ec4e4
FZ
1439
1440 if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv, bs->read_only)) {
8be25de6
KW
1441 if (!bs->read_only && bdrv_is_whitelisted(drv, true)) {
1442 ret = bdrv_apply_auto_read_only(bs, NULL, NULL);
1443 } else {
1444 ret = -ENOTSUP;
1445 }
1446 if (ret < 0) {
1447 error_setg(errp,
1448 !bs->read_only && bdrv_is_whitelisted(drv, true)
1449 ? "Driver '%s' can only be used for read-only devices"
1450 : "Driver '%s' is not whitelisted",
1451 drv->format_name);
1452 goto fail_opts;
1453 }
b64ec4e4 1454 }
57915332 1455
d3faa13e
PB
1456 /* bdrv_new() and bdrv_close() make it so */
1457 assert(atomic_read(&bs->copy_on_read) == 0);
1458
82dc8b41 1459 if (bs->open_flags & BDRV_O_COPY_ON_READ) {
0ebd24e0
KW
1460 if (!bs->read_only) {
1461 bdrv_enable_copy_on_read(bs);
1462 } else {
1463 error_setg(errp, "Can't use copy-on-read on read-only device");
18edf289
KW
1464 ret = -EINVAL;
1465 goto fail_opts;
0ebd24e0 1466 }
53fec9d3
SH
1467 }
1468
415bbca8 1469 discard = qemu_opt_get(opts, BDRV_OPT_DISCARD);
818584a4
KW
1470 if (discard != NULL) {
1471 if (bdrv_parse_discard_flags(discard, &bs->open_flags) != 0) {
1472 error_setg(errp, "Invalid discard option");
1473 ret = -EINVAL;
1474 goto fail_opts;
1475 }
1476 }
1477
543770bd
AG
1478 bs->detect_zeroes =
1479 bdrv_parse_detect_zeroes(opts, bs->open_flags, &local_err);
1480 if (local_err) {
1481 error_propagate(errp, local_err);
1482 ret = -EINVAL;
1483 goto fail_opts;
692e01a2
KW
1484 }
1485
c2ad1b0c
KW
1486 if (filename != NULL) {
1487 pstrcpy(bs->filename, sizeof(bs->filename), filename);
1488 } else {
1489 bs->filename[0] = '\0';
1490 }
91af7014 1491 pstrcpy(bs->exact_filename, sizeof(bs->exact_filename), bs->filename);
57915332 1492
66f82cee 1493 /* Open the image, either directly or using a protocol */
82dc8b41 1494 open_flags = bdrv_open_flags(bs, bs->open_flags);
01a56501 1495 node_name = qemu_opt_get(opts, "node-name");
57915332 1496
01a56501
KW
1497 assert(!drv->bdrv_file_open || file == NULL);
1498 ret = bdrv_open_driver(bs, drv, node_name, options, open_flags, errp);
51762288 1499 if (ret < 0) {
01a56501 1500 goto fail_opts;
3baca891
KW
1501 }
1502
18edf289 1503 qemu_opts_del(opts);
57915332
KW
1504 return 0;
1505
18edf289
KW
1506fail_opts:
1507 qemu_opts_del(opts);
57915332
KW
1508 return ret;
1509}
1510
5e5c4f63
KW
1511static QDict *parse_json_filename(const char *filename, Error **errp)
1512{
1513 QObject *options_obj;
1514 QDict *options;
1515 int ret;
1516
1517 ret = strstart(filename, "json:", &filename);
1518 assert(ret);
1519
5577fff7 1520 options_obj = qobject_from_json(filename, errp);
5e5c4f63 1521 if (!options_obj) {
5577fff7 1522 error_prepend(errp, "Could not parse the JSON options: ");
5e5c4f63
KW
1523 return NULL;
1524 }
1525
7dc847eb 1526 options = qobject_to(QDict, options_obj);
ca6b6e1e 1527 if (!options) {
cb3e7f08 1528 qobject_unref(options_obj);
5e5c4f63
KW
1529 error_setg(errp, "Invalid JSON object given");
1530 return NULL;
1531 }
1532
5e5c4f63
KW
1533 qdict_flatten(options);
1534
1535 return options;
1536}
1537
de3b53f0
KW
1538static void parse_json_protocol(QDict *options, const char **pfilename,
1539 Error **errp)
1540{
1541 QDict *json_options;
1542 Error *local_err = NULL;
1543
1544 /* Parse json: pseudo-protocol */
1545 if (!*pfilename || !g_str_has_prefix(*pfilename, "json:")) {
1546 return;
1547 }
1548
1549 json_options = parse_json_filename(*pfilename, &local_err);
1550 if (local_err) {
1551 error_propagate(errp, local_err);
1552 return;
1553 }
1554
1555 /* Options given in the filename have lower priority than options
1556 * specified directly */
1557 qdict_join(options, json_options, false);
cb3e7f08 1558 qobject_unref(json_options);
de3b53f0
KW
1559 *pfilename = NULL;
1560}
1561
b6ce07aa 1562/*
f54120ff
KW
1563 * Fills in default options for opening images and converts the legacy
1564 * filename/flags pair to option QDict entries.
53a29513
HR
1565 * The BDRV_O_PROTOCOL flag in *flags will be set or cleared accordingly if a
1566 * block driver has been specified explicitly.
b6ce07aa 1567 */
de3b53f0 1568static int bdrv_fill_options(QDict **options, const char *filename,
053e1578 1569 int *flags, Error **errp)
ea2384d3 1570{
c2ad1b0c 1571 const char *drvname;
53a29513 1572 bool protocol = *flags & BDRV_O_PROTOCOL;
e3fa4bfa 1573 bool parse_filename = false;
053e1578 1574 BlockDriver *drv = NULL;
34b5d2c6 1575 Error *local_err = NULL;
83f64091 1576
129c7d1c
MA
1577 /*
1578 * Caution: while qdict_get_try_str() is fine, getting non-string
1579 * types would require more care. When @options come from
1580 * -blockdev or blockdev_add, its members are typed according to
1581 * the QAPI schema, but when they come from -drive, they're all
1582 * QString.
1583 */
53a29513 1584 drvname = qdict_get_try_str(*options, "driver");
053e1578
HR
1585 if (drvname) {
1586 drv = bdrv_find_format(drvname);
1587 if (!drv) {
1588 error_setg(errp, "Unknown driver '%s'", drvname);
1589 return -ENOENT;
1590 }
1591 /* If the user has explicitly specified the driver, this choice should
1592 * override the BDRV_O_PROTOCOL flag */
1593 protocol = drv->bdrv_file_open;
53a29513
HR
1594 }
1595
1596 if (protocol) {
1597 *flags |= BDRV_O_PROTOCOL;
1598 } else {
1599 *flags &= ~BDRV_O_PROTOCOL;
1600 }
1601
91a097e7
KW
1602 /* Translate cache options from flags into options */
1603 update_options_from_flags(*options, *flags);
1604
035fccdf 1605 /* Fetch the file name from the options QDict if necessary */
17b005f1 1606 if (protocol && filename) {
f54120ff 1607 if (!qdict_haskey(*options, "filename")) {
46f5ac20 1608 qdict_put_str(*options, "filename", filename);
f54120ff
KW
1609 parse_filename = true;
1610 } else {
1611 error_setg(errp, "Can't specify 'file' and 'filename' options at "
1612 "the same time");
1613 return -EINVAL;
1614 }
035fccdf
KW
1615 }
1616
c2ad1b0c 1617 /* Find the right block driver */
129c7d1c 1618 /* See cautionary note on accessing @options above */
f54120ff 1619 filename = qdict_get_try_str(*options, "filename");
f54120ff 1620
053e1578
HR
1621 if (!drvname && protocol) {
1622 if (filename) {
1623 drv = bdrv_find_protocol(filename, parse_filename, errp);
17b005f1 1624 if (!drv) {
053e1578 1625 return -EINVAL;
17b005f1 1626 }
053e1578
HR
1627
1628 drvname = drv->format_name;
46f5ac20 1629 qdict_put_str(*options, "driver", drvname);
053e1578
HR
1630 } else {
1631 error_setg(errp, "Must specify either driver or file");
1632 return -EINVAL;
98289620 1633 }
c2ad1b0c
KW
1634 }
1635
17b005f1 1636 assert(drv || !protocol);
c2ad1b0c 1637
f54120ff 1638 /* Driver-specific filename parsing */
17b005f1 1639 if (drv && drv->bdrv_parse_filename && parse_filename) {
5acd9d81 1640 drv->bdrv_parse_filename(filename, *options, &local_err);
84d18f06 1641 if (local_err) {
34b5d2c6 1642 error_propagate(errp, local_err);
f54120ff 1643 return -EINVAL;
6963a30d 1644 }
cd5d031e
HR
1645
1646 if (!drv->bdrv_needs_filename) {
1647 qdict_del(*options, "filename");
cd5d031e 1648 }
6963a30d
KW
1649 }
1650
f54120ff
KW
1651 return 0;
1652}
1653
3121fb45
KW
1654static int bdrv_child_check_perm(BdrvChild *c, BlockReopenQueue *q,
1655 uint64_t perm, uint64_t shared,
c1cef672
FZ
1656 GSList *ignore_children, Error **errp);
1657static void bdrv_child_abort_perm_update(BdrvChild *c);
1658static void bdrv_child_set_perm(BdrvChild *c, uint64_t perm, uint64_t shared);
1659
148eb13c
KW
1660typedef struct BlockReopenQueueEntry {
1661 bool prepared;
1662 BDRVReopenState state;
1663 QSIMPLEQ_ENTRY(BlockReopenQueueEntry) entry;
1664} BlockReopenQueueEntry;
1665
1666/*
1667 * Return the flags that @bs will have after the reopens in @q have
1668 * successfully completed. If @q is NULL (or @bs is not contained in @q),
1669 * return the current flags.
1670 */
1671static int bdrv_reopen_get_flags(BlockReopenQueue *q, BlockDriverState *bs)
1672{
1673 BlockReopenQueueEntry *entry;
1674
1675 if (q != NULL) {
1676 QSIMPLEQ_FOREACH(entry, q, entry) {
1677 if (entry->state.bs == bs) {
1678 return entry->state.flags;
1679 }
1680 }
1681 }
1682
1683 return bs->open_flags;
1684}
1685
1686/* Returns whether the image file can be written to after the reopen queue @q
1687 * has been successfully applied, or right now if @q is NULL. */
cc022140
HR
1688static bool bdrv_is_writable_after_reopen(BlockDriverState *bs,
1689 BlockReopenQueue *q)
148eb13c
KW
1690{
1691 int flags = bdrv_reopen_get_flags(q, bs);
1692
1693 return (flags & (BDRV_O_RDWR | BDRV_O_INACTIVE)) == BDRV_O_RDWR;
1694}
1695
cc022140
HR
1696/*
1697 * Return whether the BDS can be written to. This is not necessarily
1698 * the same as !bdrv_is_read_only(bs), as inactivated images may not
1699 * be written to but do not count as read-only images.
1700 */
1701bool bdrv_is_writable(BlockDriverState *bs)
1702{
1703 return bdrv_is_writable_after_reopen(bs, NULL);
1704}
1705
ffd1a5a2 1706static void bdrv_child_perm(BlockDriverState *bs, BlockDriverState *child_bs,
e0995dc3
KW
1707 BdrvChild *c, const BdrvChildRole *role,
1708 BlockReopenQueue *reopen_queue,
ffd1a5a2
FZ
1709 uint64_t parent_perm, uint64_t parent_shared,
1710 uint64_t *nperm, uint64_t *nshared)
1711{
1712 if (bs->drv && bs->drv->bdrv_child_perm) {
e0995dc3 1713 bs->drv->bdrv_child_perm(bs, c, role, reopen_queue,
ffd1a5a2
FZ
1714 parent_perm, parent_shared,
1715 nperm, nshared);
1716 }
e0995dc3 1717 /* TODO Take force_share from reopen_queue */
ffd1a5a2
FZ
1718 if (child_bs && child_bs->force_share) {
1719 *nshared = BLK_PERM_ALL;
1720 }
1721}
1722
33a610c3
KW
1723/*
1724 * Check whether permissions on this node can be changed in a way that
1725 * @cumulative_perms and @cumulative_shared_perms are the new cumulative
1726 * permissions of all its parents. This involves checking whether all necessary
1727 * permission changes to child nodes can be performed.
1728 *
1729 * A call to this function must always be followed by a call to bdrv_set_perm()
1730 * or bdrv_abort_perm_update().
1731 */
3121fb45
KW
1732static int bdrv_check_perm(BlockDriverState *bs, BlockReopenQueue *q,
1733 uint64_t cumulative_perms,
46181129
KW
1734 uint64_t cumulative_shared_perms,
1735 GSList *ignore_children, Error **errp)
33a610c3
KW
1736{
1737 BlockDriver *drv = bs->drv;
1738 BdrvChild *c;
1739 int ret;
1740
1741 /* Write permissions never work with read-only images */
1742 if ((cumulative_perms & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)) &&
cc022140 1743 !bdrv_is_writable_after_reopen(bs, q))
33a610c3
KW
1744 {
1745 error_setg(errp, "Block node is read-only");
1746 return -EPERM;
1747 }
1748
1749 /* Check this node */
1750 if (!drv) {
1751 return 0;
1752 }
1753
1754 if (drv->bdrv_check_perm) {
1755 return drv->bdrv_check_perm(bs, cumulative_perms,
1756 cumulative_shared_perms, errp);
1757 }
1758
78e421c9 1759 /* Drivers that never have children can omit .bdrv_child_perm() */
33a610c3 1760 if (!drv->bdrv_child_perm) {
78e421c9 1761 assert(QLIST_EMPTY(&bs->children));
33a610c3
KW
1762 return 0;
1763 }
1764
1765 /* Check all children */
1766 QLIST_FOREACH(c, &bs->children, next) {
1767 uint64_t cur_perm, cur_shared;
3121fb45 1768 bdrv_child_perm(bs, c->bs, c, c->role, q,
ffd1a5a2
FZ
1769 cumulative_perms, cumulative_shared_perms,
1770 &cur_perm, &cur_shared);
3121fb45
KW
1771 ret = bdrv_child_check_perm(c, q, cur_perm, cur_shared,
1772 ignore_children, errp);
33a610c3
KW
1773 if (ret < 0) {
1774 return ret;
1775 }
1776 }
1777
1778 return 0;
1779}
1780
1781/*
1782 * Notifies drivers that after a previous bdrv_check_perm() call, the
1783 * permission update is not performed and any preparations made for it (e.g.
1784 * taken file locks) need to be undone.
1785 *
1786 * This function recursively notifies all child nodes.
1787 */
1788static void bdrv_abort_perm_update(BlockDriverState *bs)
1789{
1790 BlockDriver *drv = bs->drv;
1791 BdrvChild *c;
1792
1793 if (!drv) {
1794 return;
1795 }
1796
1797 if (drv->bdrv_abort_perm_update) {
1798 drv->bdrv_abort_perm_update(bs);
1799 }
1800
1801 QLIST_FOREACH(c, &bs->children, next) {
1802 bdrv_child_abort_perm_update(c);
1803 }
1804}
1805
1806static void bdrv_set_perm(BlockDriverState *bs, uint64_t cumulative_perms,
1807 uint64_t cumulative_shared_perms)
1808{
1809 BlockDriver *drv = bs->drv;
1810 BdrvChild *c;
1811
1812 if (!drv) {
1813 return;
1814 }
1815
1816 /* Update this node */
1817 if (drv->bdrv_set_perm) {
1818 drv->bdrv_set_perm(bs, cumulative_perms, cumulative_shared_perms);
1819 }
1820
78e421c9 1821 /* Drivers that never have children can omit .bdrv_child_perm() */
33a610c3 1822 if (!drv->bdrv_child_perm) {
78e421c9 1823 assert(QLIST_EMPTY(&bs->children));
33a610c3
KW
1824 return;
1825 }
1826
1827 /* Update all children */
1828 QLIST_FOREACH(c, &bs->children, next) {
1829 uint64_t cur_perm, cur_shared;
e0995dc3 1830 bdrv_child_perm(bs, c->bs, c, c->role, NULL,
ffd1a5a2
FZ
1831 cumulative_perms, cumulative_shared_perms,
1832 &cur_perm, &cur_shared);
33a610c3
KW
1833 bdrv_child_set_perm(c, cur_perm, cur_shared);
1834 }
1835}
1836
1837static void bdrv_get_cumulative_perm(BlockDriverState *bs, uint64_t *perm,
1838 uint64_t *shared_perm)
1839{
1840 BdrvChild *c;
1841 uint64_t cumulative_perms = 0;
1842 uint64_t cumulative_shared_perms = BLK_PERM_ALL;
1843
1844 QLIST_FOREACH(c, &bs->parents, next_parent) {
1845 cumulative_perms |= c->perm;
1846 cumulative_shared_perms &= c->shared_perm;
1847 }
1848
1849 *perm = cumulative_perms;
1850 *shared_perm = cumulative_shared_perms;
1851}
1852
d083319f
KW
1853static char *bdrv_child_user_desc(BdrvChild *c)
1854{
1855 if (c->role->get_parent_desc) {
1856 return c->role->get_parent_desc(c);
1857 }
1858
1859 return g_strdup("another user");
1860}
1861
5176196c 1862char *bdrv_perm_names(uint64_t perm)
d083319f
KW
1863{
1864 struct perm_name {
1865 uint64_t perm;
1866 const char *name;
1867 } permissions[] = {
1868 { BLK_PERM_CONSISTENT_READ, "consistent read" },
1869 { BLK_PERM_WRITE, "write" },
1870 { BLK_PERM_WRITE_UNCHANGED, "write unchanged" },
1871 { BLK_PERM_RESIZE, "resize" },
1872 { BLK_PERM_GRAPH_MOD, "change children" },
1873 { 0, NULL }
1874 };
1875
1876 char *result = g_strdup("");
1877 struct perm_name *p;
1878
1879 for (p = permissions; p->name; p++) {
1880 if (perm & p->perm) {
1881 char *old = result;
1882 result = g_strdup_printf("%s%s%s", old, *old ? ", " : "", p->name);
1883 g_free(old);
1884 }
1885 }
1886
1887 return result;
1888}
1889
33a610c3
KW
1890/*
1891 * Checks whether a new reference to @bs can be added if the new user requires
46181129
KW
1892 * @new_used_perm/@new_shared_perm as its permissions. If @ignore_children is
1893 * set, the BdrvChild objects in this list are ignored in the calculations;
1894 * this allows checking permission updates for an existing reference.
33a610c3
KW
1895 *
1896 * Needs to be followed by a call to either bdrv_set_perm() or
1897 * bdrv_abort_perm_update(). */
3121fb45
KW
1898static int bdrv_check_update_perm(BlockDriverState *bs, BlockReopenQueue *q,
1899 uint64_t new_used_perm,
d5e6f437 1900 uint64_t new_shared_perm,
46181129 1901 GSList *ignore_children, Error **errp)
d5e6f437
KW
1902{
1903 BdrvChild *c;
33a610c3
KW
1904 uint64_t cumulative_perms = new_used_perm;
1905 uint64_t cumulative_shared_perms = new_shared_perm;
d5e6f437
KW
1906
1907 /* There is no reason why anyone couldn't tolerate write_unchanged */
1908 assert(new_shared_perm & BLK_PERM_WRITE_UNCHANGED);
1909
1910 QLIST_FOREACH(c, &bs->parents, next_parent) {
46181129 1911 if (g_slist_find(ignore_children, c)) {
d5e6f437
KW
1912 continue;
1913 }
1914
d083319f
KW
1915 if ((new_used_perm & c->shared_perm) != new_used_perm) {
1916 char *user = bdrv_child_user_desc(c);
1917 char *perm_names = bdrv_perm_names(new_used_perm & ~c->shared_perm);
1918 error_setg(errp, "Conflicts with use by %s as '%s', which does not "
1919 "allow '%s' on %s",
1920 user, c->name, perm_names, bdrv_get_node_name(c->bs));
1921 g_free(user);
1922 g_free(perm_names);
1923 return -EPERM;
1924 }
1925
1926 if ((c->perm & new_shared_perm) != c->perm) {
1927 char *user = bdrv_child_user_desc(c);
1928 char *perm_names = bdrv_perm_names(c->perm & ~new_shared_perm);
1929 error_setg(errp, "Conflicts with use by %s as '%s', which uses "
1930 "'%s' on %s",
1931 user, c->name, perm_names, bdrv_get_node_name(c->bs));
1932 g_free(user);
1933 g_free(perm_names);
d5e6f437
KW
1934 return -EPERM;
1935 }
33a610c3
KW
1936
1937 cumulative_perms |= c->perm;
1938 cumulative_shared_perms &= c->shared_perm;
d5e6f437
KW
1939 }
1940
3121fb45 1941 return bdrv_check_perm(bs, q, cumulative_perms, cumulative_shared_perms,
46181129 1942 ignore_children, errp);
33a610c3
KW
1943}
1944
1945/* Needs to be followed by a call to either bdrv_child_set_perm() or
1946 * bdrv_child_abort_perm_update(). */
3121fb45
KW
1947static int bdrv_child_check_perm(BdrvChild *c, BlockReopenQueue *q,
1948 uint64_t perm, uint64_t shared,
c1cef672 1949 GSList *ignore_children, Error **errp)
33a610c3 1950{
46181129
KW
1951 int ret;
1952
1953 ignore_children = g_slist_prepend(g_slist_copy(ignore_children), c);
3121fb45 1954 ret = bdrv_check_update_perm(c->bs, q, perm, shared, ignore_children, errp);
46181129
KW
1955 g_slist_free(ignore_children);
1956
1957 return ret;
33a610c3
KW
1958}
1959
c1cef672 1960static void bdrv_child_set_perm(BdrvChild *c, uint64_t perm, uint64_t shared)
33a610c3
KW
1961{
1962 uint64_t cumulative_perms, cumulative_shared_perms;
1963
1964 c->perm = perm;
1965 c->shared_perm = shared;
1966
1967 bdrv_get_cumulative_perm(c->bs, &cumulative_perms,
1968 &cumulative_shared_perms);
1969 bdrv_set_perm(c->bs, cumulative_perms, cumulative_shared_perms);
1970}
1971
c1cef672 1972static void bdrv_child_abort_perm_update(BdrvChild *c)
33a610c3
KW
1973{
1974 bdrv_abort_perm_update(c->bs);
1975}
1976
1977int bdrv_child_try_set_perm(BdrvChild *c, uint64_t perm, uint64_t shared,
1978 Error **errp)
1979{
1980 int ret;
1981
3121fb45 1982 ret = bdrv_child_check_perm(c, NULL, perm, shared, NULL, errp);
33a610c3
KW
1983 if (ret < 0) {
1984 bdrv_child_abort_perm_update(c);
1985 return ret;
1986 }
1987
1988 bdrv_child_set_perm(c, perm, shared);
1989
d5e6f437
KW
1990 return 0;
1991}
1992
6a1b9ee1
KW
1993void bdrv_filter_default_perms(BlockDriverState *bs, BdrvChild *c,
1994 const BdrvChildRole *role,
e0995dc3 1995 BlockReopenQueue *reopen_queue,
6a1b9ee1
KW
1996 uint64_t perm, uint64_t shared,
1997 uint64_t *nperm, uint64_t *nshared)
1998{
1999 if (c == NULL) {
2000 *nperm = perm & DEFAULT_PERM_PASSTHROUGH;
2001 *nshared = (shared & DEFAULT_PERM_PASSTHROUGH) | DEFAULT_PERM_UNCHANGED;
2002 return;
2003 }
2004
2005 *nperm = (perm & DEFAULT_PERM_PASSTHROUGH) |
2006 (c->perm & DEFAULT_PERM_UNCHANGED);
2007 *nshared = (shared & DEFAULT_PERM_PASSTHROUGH) |
2008 (c->shared_perm & DEFAULT_PERM_UNCHANGED);
2009}
2010
6b1a044a
KW
2011void bdrv_format_default_perms(BlockDriverState *bs, BdrvChild *c,
2012 const BdrvChildRole *role,
e0995dc3 2013 BlockReopenQueue *reopen_queue,
6b1a044a
KW
2014 uint64_t perm, uint64_t shared,
2015 uint64_t *nperm, uint64_t *nshared)
2016{
2017 bool backing = (role == &child_backing);
2018 assert(role == &child_backing || role == &child_file);
2019
2020 if (!backing) {
5fbfabd3
KW
2021 int flags = bdrv_reopen_get_flags(reopen_queue, bs);
2022
6b1a044a
KW
2023 /* Apart from the modifications below, the same permissions are
2024 * forwarded and left alone as for filters */
e0995dc3
KW
2025 bdrv_filter_default_perms(bs, c, role, reopen_queue, perm, shared,
2026 &perm, &shared);
6b1a044a
KW
2027
2028 /* Format drivers may touch metadata even if the guest doesn't write */
cc022140 2029 if (bdrv_is_writable_after_reopen(bs, reopen_queue)) {
6b1a044a
KW
2030 perm |= BLK_PERM_WRITE | BLK_PERM_RESIZE;
2031 }
2032
2033 /* bs->file always needs to be consistent because of the metadata. We
2034 * can never allow other users to resize or write to it. */
5fbfabd3
KW
2035 if (!(flags & BDRV_O_NO_IO)) {
2036 perm |= BLK_PERM_CONSISTENT_READ;
2037 }
6b1a044a
KW
2038 shared &= ~(BLK_PERM_WRITE | BLK_PERM_RESIZE);
2039 } else {
2040 /* We want consistent read from backing files if the parent needs it.
2041 * No other operations are performed on backing files. */
2042 perm &= BLK_PERM_CONSISTENT_READ;
2043
2044 /* If the parent can deal with changing data, we're okay with a
2045 * writable and resizable backing file. */
2046 /* TODO Require !(perm & BLK_PERM_CONSISTENT_READ), too? */
2047 if (shared & BLK_PERM_WRITE) {
2048 shared = BLK_PERM_WRITE | BLK_PERM_RESIZE;
2049 } else {
2050 shared = 0;
2051 }
2052
2053 shared |= BLK_PERM_CONSISTENT_READ | BLK_PERM_GRAPH_MOD |
2054 BLK_PERM_WRITE_UNCHANGED;
2055 }
2056
9c5e6594
KW
2057 if (bs->open_flags & BDRV_O_INACTIVE) {
2058 shared |= BLK_PERM_WRITE | BLK_PERM_RESIZE;
2059 }
2060
6b1a044a
KW
2061 *nperm = perm;
2062 *nshared = shared;
2063}
2064
8ee03995
KW
2065static void bdrv_replace_child_noperm(BdrvChild *child,
2066 BlockDriverState *new_bs)
e9740bc6
KW
2067{
2068 BlockDriverState *old_bs = child->bs;
0152bf40 2069 int i;
e9740bc6 2070
bb2614e9
FZ
2071 if (old_bs && new_bs) {
2072 assert(bdrv_get_aio_context(old_bs) == bdrv_get_aio_context(new_bs));
2073 }
e9740bc6 2074 if (old_bs) {
d736f119
KW
2075 /* Detach first so that the recursive drain sections coming from @child
2076 * are already gone and we only end the drain sections that came from
2077 * elsewhere. */
2078 if (child->role->detach) {
2079 child->role->detach(child);
2080 }
36fe1331 2081 if (old_bs->quiesce_counter && child->role->drained_end) {
0f12264e
KW
2082 int num = old_bs->quiesce_counter;
2083 if (child->role->parent_is_bds) {
2084 num -= bdrv_drain_all_count;
2085 }
2086 assert(num >= 0);
2087 for (i = 0; i < num; i++) {
0152bf40
KW
2088 child->role->drained_end(child);
2089 }
36fe1331 2090 }
e9740bc6
KW
2091 QLIST_REMOVE(child, next_parent);
2092 }
36fe1331
KW
2093
2094 child->bs = new_bs;
2095
e9740bc6
KW
2096 if (new_bs) {
2097 QLIST_INSERT_HEAD(&new_bs->parents, child, next_parent);
36fe1331 2098 if (new_bs->quiesce_counter && child->role->drained_begin) {
0f12264e
KW
2099 int num = new_bs->quiesce_counter;
2100 if (child->role->parent_is_bds) {
2101 num -= bdrv_drain_all_count;
2102 }
2103 assert(num >= 0);
2104 for (i = 0; i < num; i++) {
4be6a6d1 2105 bdrv_parent_drained_begin_single(child, true);
0152bf40 2106 }
36fe1331 2107 }
33a610c3 2108
d736f119
KW
2109 /* Attach only after starting new drained sections, so that recursive
2110 * drain sections coming from @child don't get an extra .drained_begin
2111 * callback. */
8ee03995
KW
2112 if (child->role->attach) {
2113 child->role->attach(child);
2114 }
2115 }
2116}
33a610c3 2117
466787fb
KW
2118/*
2119 * Updates @child to change its reference to point to @new_bs, including
2120 * checking and applying the necessary permisson updates both to the old node
2121 * and to @new_bs.
2122 *
2123 * NULL is passed as @new_bs for removing the reference before freeing @child.
2124 *
2125 * If @new_bs is not NULL, bdrv_check_perm() must be called beforehand, as this
2126 * function uses bdrv_set_perm() to update the permissions according to the new
2127 * reference that @new_bs gets.
2128 */
2129static void bdrv_replace_child(BdrvChild *child, BlockDriverState *new_bs)
8ee03995
KW
2130{
2131 BlockDriverState *old_bs = child->bs;
2132 uint64_t perm, shared_perm;
2133
8aecf1d1
KW
2134 bdrv_replace_child_noperm(child, new_bs);
2135
8ee03995 2136 if (old_bs) {
33a610c3
KW
2137 /* Update permissions for old node. This is guaranteed to succeed
2138 * because we're just taking a parent away, so we're loosening
2139 * restrictions. */
2140 bdrv_get_cumulative_perm(old_bs, &perm, &shared_perm);
3121fb45 2141 bdrv_check_perm(old_bs, NULL, perm, shared_perm, NULL, &error_abort);
33a610c3 2142 bdrv_set_perm(old_bs, perm, shared_perm);
e9740bc6 2143 }
36fe1331 2144
e9740bc6 2145 if (new_bs) {
33a610c3 2146 bdrv_get_cumulative_perm(new_bs, &perm, &shared_perm);
33a610c3 2147 bdrv_set_perm(new_bs, perm, shared_perm);
e9740bc6 2148 }
e9740bc6
KW
2149}
2150
f21d96d0
KW
2151BdrvChild *bdrv_root_attach_child(BlockDriverState *child_bs,
2152 const char *child_name,
36fe1331 2153 const BdrvChildRole *child_role,
d5e6f437
KW
2154 uint64_t perm, uint64_t shared_perm,
2155 void *opaque, Error **errp)
df581792 2156{
d5e6f437
KW
2157 BdrvChild *child;
2158 int ret;
2159
3121fb45 2160 ret = bdrv_check_update_perm(child_bs, NULL, perm, shared_perm, NULL, errp);
d5e6f437 2161 if (ret < 0) {
33a610c3 2162 bdrv_abort_perm_update(child_bs);
d5e6f437
KW
2163 return NULL;
2164 }
2165
2166 child = g_new(BdrvChild, 1);
df581792 2167 *child = (BdrvChild) {
d5e6f437
KW
2168 .bs = NULL,
2169 .name = g_strdup(child_name),
2170 .role = child_role,
2171 .perm = perm,
2172 .shared_perm = shared_perm,
2173 .opaque = opaque,
df581792
KW
2174 };
2175
33a610c3 2176 /* This performs the matching bdrv_set_perm() for the above check. */
466787fb 2177 bdrv_replace_child(child, child_bs);
b4b059f6
KW
2178
2179 return child;
df581792
KW
2180}
2181
98292c61
WC
2182BdrvChild *bdrv_attach_child(BlockDriverState *parent_bs,
2183 BlockDriverState *child_bs,
2184 const char *child_name,
8b2ff529
KW
2185 const BdrvChildRole *child_role,
2186 Error **errp)
f21d96d0 2187{
d5e6f437 2188 BdrvChild *child;
f68c598b
KW
2189 uint64_t perm, shared_perm;
2190
2191 bdrv_get_cumulative_perm(parent_bs, &perm, &shared_perm);
2192
2193 assert(parent_bs->drv);
bb2614e9 2194 assert(bdrv_get_aio_context(parent_bs) == bdrv_get_aio_context(child_bs));
e0995dc3 2195 bdrv_child_perm(parent_bs, child_bs, NULL, child_role, NULL,
ffd1a5a2 2196 perm, shared_perm, &perm, &shared_perm);
d5e6f437 2197
d5e6f437 2198 child = bdrv_root_attach_child(child_bs, child_name, child_role,
f68c598b 2199 perm, shared_perm, parent_bs, errp);
d5e6f437
KW
2200 if (child == NULL) {
2201 return NULL;
2202 }
2203
f21d96d0
KW
2204 QLIST_INSERT_HEAD(&parent_bs->children, child, next);
2205 return child;
2206}
2207
3f09bfbc 2208static void bdrv_detach_child(BdrvChild *child)
33a60407 2209{
f21d96d0
KW
2210 if (child->next.le_prev) {
2211 QLIST_REMOVE(child, next);
2212 child->next.le_prev = NULL;
2213 }
e9740bc6 2214
466787fb 2215 bdrv_replace_child(child, NULL);
e9740bc6 2216
260fecf1 2217 g_free(child->name);
33a60407
KW
2218 g_free(child);
2219}
2220
f21d96d0 2221void bdrv_root_unref_child(BdrvChild *child)
33a60407 2222{
779020cb
KW
2223 BlockDriverState *child_bs;
2224
f21d96d0
KW
2225 child_bs = child->bs;
2226 bdrv_detach_child(child);
2227 bdrv_unref(child_bs);
2228}
2229
2230void bdrv_unref_child(BlockDriverState *parent, BdrvChild *child)
2231{
779020cb
KW
2232 if (child == NULL) {
2233 return;
2234 }
33a60407
KW
2235
2236 if (child->bs->inherits_from == parent) {
4e4bf5c4
KW
2237 BdrvChild *c;
2238
2239 /* Remove inherits_from only when the last reference between parent and
2240 * child->bs goes away. */
2241 QLIST_FOREACH(c, &parent->children, next) {
2242 if (c != child && c->bs == child->bs) {
2243 break;
2244 }
2245 }
2246 if (c == NULL) {
2247 child->bs->inherits_from = NULL;
2248 }
33a60407
KW
2249 }
2250
f21d96d0 2251 bdrv_root_unref_child(child);
33a60407
KW
2252}
2253
5c8cab48
KW
2254
2255static void bdrv_parent_cb_change_media(BlockDriverState *bs, bool load)
2256{
2257 BdrvChild *c;
2258 QLIST_FOREACH(c, &bs->parents, next_parent) {
2259 if (c->role->change_media) {
2260 c->role->change_media(c, load);
2261 }
2262 }
2263}
2264
0065c455
AG
2265/* Return true if you can reach parent going through child->inherits_from
2266 * recursively. If parent or child are NULL, return false */
2267static bool bdrv_inherits_from_recursive(BlockDriverState *child,
2268 BlockDriverState *parent)
2269{
2270 while (child && child != parent) {
2271 child = child->inherits_from;
2272 }
2273
2274 return child != NULL;
2275}
2276
5db15a57
KW
2277/*
2278 * Sets the backing file link of a BDS. A new reference is created; callers
2279 * which don't need their own reference any more must call bdrv_unref().
2280 */
12fa4af6
KW
2281void bdrv_set_backing_hd(BlockDriverState *bs, BlockDriverState *backing_hd,
2282 Error **errp)
8d24cce1 2283{
0065c455
AG
2284 bool update_inherits_from = bdrv_chain_contains(bs, backing_hd) &&
2285 bdrv_inherits_from_recursive(backing_hd, bs);
2286
5db15a57
KW
2287 if (backing_hd) {
2288 bdrv_ref(backing_hd);
2289 }
8d24cce1 2290
760e0063 2291 if (bs->backing) {
5db15a57 2292 bdrv_unref_child(bs, bs->backing);
826b6ca0
FZ
2293 }
2294
8d24cce1 2295 if (!backing_hd) {
760e0063 2296 bs->backing = NULL;
8d24cce1
FZ
2297 goto out;
2298 }
12fa4af6 2299
8b2ff529 2300 bs->backing = bdrv_attach_child(bs, backing_hd, "backing", &child_backing,
12fa4af6 2301 errp);
0065c455
AG
2302 /* If backing_hd was already part of bs's backing chain, and
2303 * inherits_from pointed recursively to bs then let's update it to
2304 * point directly to bs (else it will become NULL). */
2305 if (update_inherits_from) {
2306 backing_hd->inherits_from = bs;
2307 }
12fa4af6
KW
2308 if (!bs->backing) {
2309 bdrv_unref(backing_hd);
2310 }
826b6ca0 2311
9e7e940c
KW
2312 bdrv_refresh_filename(bs);
2313
8d24cce1 2314out:
3baca891 2315 bdrv_refresh_limits(bs, NULL);
8d24cce1
FZ
2316}
2317
31ca6d07
KW
2318/*
2319 * Opens the backing file for a BlockDriverState if not yet open
2320 *
d9b7b057
KW
2321 * bdref_key specifies the key for the image's BlockdevRef in the options QDict.
2322 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
2323 * itself, all options starting with "${bdref_key}." are considered part of the
2324 * BlockdevRef.
2325 *
2326 * TODO Can this be unified with bdrv_open_image()?
31ca6d07 2327 */
d9b7b057
KW
2328int bdrv_open_backing_file(BlockDriverState *bs, QDict *parent_options,
2329 const char *bdref_key, Error **errp)
9156df12 2330{
1ba4b6a5 2331 char *backing_filename = g_malloc0(PATH_MAX);
d9b7b057
KW
2332 char *bdref_key_dot;
2333 const char *reference = NULL;
317fc44e 2334 int ret = 0;
8d24cce1 2335 BlockDriverState *backing_hd;
d9b7b057
KW
2336 QDict *options;
2337 QDict *tmp_parent_options = NULL;
34b5d2c6 2338 Error *local_err = NULL;
9156df12 2339
760e0063 2340 if (bs->backing != NULL) {
1ba4b6a5 2341 goto free_exit;
9156df12
PB
2342 }
2343
31ca6d07 2344 /* NULL means an empty set of options */
d9b7b057
KW
2345 if (parent_options == NULL) {
2346 tmp_parent_options = qdict_new();
2347 parent_options = tmp_parent_options;
31ca6d07
KW
2348 }
2349
9156df12 2350 bs->open_flags &= ~BDRV_O_NO_BACKING;
d9b7b057
KW
2351
2352 bdref_key_dot = g_strdup_printf("%s.", bdref_key);
2353 qdict_extract_subqdict(parent_options, &options, bdref_key_dot);
2354 g_free(bdref_key_dot);
2355
129c7d1c
MA
2356 /*
2357 * Caution: while qdict_get_try_str() is fine, getting non-string
2358 * types would require more care. When @parent_options come from
2359 * -blockdev or blockdev_add, its members are typed according to
2360 * the QAPI schema, but when they come from -drive, they're all
2361 * QString.
2362 */
d9b7b057
KW
2363 reference = qdict_get_try_str(parent_options, bdref_key);
2364 if (reference || qdict_haskey(options, "file.filename")) {
1cb6f506
KW
2365 backing_filename[0] = '\0';
2366 } else if (bs->backing_file[0] == '\0' && qdict_size(options) == 0) {
cb3e7f08 2367 qobject_unref(options);
1ba4b6a5 2368 goto free_exit;
dbecebdd 2369 } else {
9f07429e
HR
2370 bdrv_get_full_backing_filename(bs, backing_filename, PATH_MAX,
2371 &local_err);
2372 if (local_err) {
2373 ret = -EINVAL;
2374 error_propagate(errp, local_err);
cb3e7f08 2375 qobject_unref(options);
9f07429e
HR
2376 goto free_exit;
2377 }
9156df12
PB
2378 }
2379
8ee79e70
KW
2380 if (!bs->drv || !bs->drv->supports_backing) {
2381 ret = -EINVAL;
2382 error_setg(errp, "Driver doesn't support backing files");
cb3e7f08 2383 qobject_unref(options);
8ee79e70
KW
2384 goto free_exit;
2385 }
2386
6bff597b
PK
2387 if (!reference &&
2388 bs->backing_format[0] != '\0' && !qdict_haskey(options, "driver")) {
46f5ac20 2389 qdict_put_str(options, "driver", bs->backing_format);
9156df12
PB
2390 }
2391
5b363937
HR
2392 backing_hd = bdrv_open_inherit(*backing_filename ? backing_filename : NULL,
2393 reference, options, 0, bs, &child_backing,
2394 errp);
2395 if (!backing_hd) {
9156df12 2396 bs->open_flags |= BDRV_O_NO_BACKING;
e43bfd9c 2397 error_prepend(errp, "Could not open backing file: ");
5b363937 2398 ret = -EINVAL;
1ba4b6a5 2399 goto free_exit;
9156df12 2400 }
5ce6bfe2 2401 bdrv_set_aio_context(backing_hd, bdrv_get_aio_context(bs));
df581792 2402
5db15a57
KW
2403 /* Hook up the backing file link; drop our reference, bs owns the
2404 * backing_hd reference now */
12fa4af6 2405 bdrv_set_backing_hd(bs, backing_hd, &local_err);
5db15a57 2406 bdrv_unref(backing_hd);
12fa4af6 2407 if (local_err) {
8cd1a3e4 2408 error_propagate(errp, local_err);
12fa4af6
KW
2409 ret = -EINVAL;
2410 goto free_exit;
2411 }
d80ac658 2412
d9b7b057
KW
2413 qdict_del(parent_options, bdref_key);
2414
1ba4b6a5
BC
2415free_exit:
2416 g_free(backing_filename);
cb3e7f08 2417 qobject_unref(tmp_parent_options);
1ba4b6a5 2418 return ret;
9156df12
PB
2419}
2420
2d6b86af
KW
2421static BlockDriverState *
2422bdrv_open_child_bs(const char *filename, QDict *options, const char *bdref_key,
2423 BlockDriverState *parent, const BdrvChildRole *child_role,
2424 bool allow_none, Error **errp)
da557aac 2425{
2d6b86af 2426 BlockDriverState *bs = NULL;
da557aac 2427 QDict *image_options;
da557aac
HR
2428 char *bdref_key_dot;
2429 const char *reference;
2430
df581792 2431 assert(child_role != NULL);
f67503e5 2432
da557aac
HR
2433 bdref_key_dot = g_strdup_printf("%s.", bdref_key);
2434 qdict_extract_subqdict(options, &image_options, bdref_key_dot);
2435 g_free(bdref_key_dot);
2436
129c7d1c
MA
2437 /*
2438 * Caution: while qdict_get_try_str() is fine, getting non-string
2439 * types would require more care. When @options come from
2440 * -blockdev or blockdev_add, its members are typed according to
2441 * the QAPI schema, but when they come from -drive, they're all
2442 * QString.
2443 */
da557aac
HR
2444 reference = qdict_get_try_str(options, bdref_key);
2445 if (!filename && !reference && !qdict_size(image_options)) {
b4b059f6 2446 if (!allow_none) {
da557aac
HR
2447 error_setg(errp, "A block device must be specified for \"%s\"",
2448 bdref_key);
da557aac 2449 }
cb3e7f08 2450 qobject_unref(image_options);
da557aac
HR
2451 goto done;
2452 }
2453
5b363937
HR
2454 bs = bdrv_open_inherit(filename, reference, image_options, 0,
2455 parent, child_role, errp);
2456 if (!bs) {
df581792
KW
2457 goto done;
2458 }
2459
da557aac
HR
2460done:
2461 qdict_del(options, bdref_key);
2d6b86af
KW
2462 return bs;
2463}
2464
2465/*
2466 * Opens a disk image whose options are given as BlockdevRef in another block
2467 * device's options.
2468 *
2469 * If allow_none is true, no image will be opened if filename is false and no
2470 * BlockdevRef is given. NULL will be returned, but errp remains unset.
2471 *
2472 * bdrev_key specifies the key for the image's BlockdevRef in the options QDict.
2473 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
2474 * itself, all options starting with "${bdref_key}." are considered part of the
2475 * BlockdevRef.
2476 *
2477 * The BlockdevRef will be removed from the options QDict.
2478 */
2479BdrvChild *bdrv_open_child(const char *filename,
2480 QDict *options, const char *bdref_key,
2481 BlockDriverState *parent,
2482 const BdrvChildRole *child_role,
2483 bool allow_none, Error **errp)
2484{
8b2ff529 2485 BdrvChild *c;
2d6b86af
KW
2486 BlockDriverState *bs;
2487
2488 bs = bdrv_open_child_bs(filename, options, bdref_key, parent, child_role,
2489 allow_none, errp);
2490 if (bs == NULL) {
2491 return NULL;
2492 }
2493
8b2ff529
KW
2494 c = bdrv_attach_child(parent, bs, bdref_key, child_role, errp);
2495 if (!c) {
2496 bdrv_unref(bs);
2497 return NULL;
2498 }
2499
2500 return c;
b4b059f6
KW
2501}
2502
e1d74bc6
KW
2503/* TODO Future callers may need to specify parent/child_role in order for
2504 * option inheritance to work. Existing callers use it for the root node. */
2505BlockDriverState *bdrv_open_blockdev_ref(BlockdevRef *ref, Error **errp)
2506{
2507 BlockDriverState *bs = NULL;
2508 Error *local_err = NULL;
2509 QObject *obj = NULL;
2510 QDict *qdict = NULL;
2511 const char *reference = NULL;
2512 Visitor *v = NULL;
2513
2514 if (ref->type == QTYPE_QSTRING) {
2515 reference = ref->u.reference;
2516 } else {
2517 BlockdevOptions *options = &ref->u.definition;
2518 assert(ref->type == QTYPE_QDICT);
2519
2520 v = qobject_output_visitor_new(&obj);
2521 visit_type_BlockdevOptions(v, NULL, &options, &local_err);
2522 if (local_err) {
2523 error_propagate(errp, local_err);
2524 goto fail;
2525 }
2526 visit_complete(v, &obj);
2527
7dc847eb 2528 qdict = qobject_to(QDict, obj);
e1d74bc6
KW
2529 qdict_flatten(qdict);
2530
2531 /* bdrv_open_inherit() defaults to the values in bdrv_flags (for
2532 * compatibility with other callers) rather than what we want as the
2533 * real defaults. Apply the defaults here instead. */
2534 qdict_set_default_str(qdict, BDRV_OPT_CACHE_DIRECT, "off");
2535 qdict_set_default_str(qdict, BDRV_OPT_CACHE_NO_FLUSH, "off");
2536 qdict_set_default_str(qdict, BDRV_OPT_READ_ONLY, "off");
e35bdc12
KW
2537 qdict_set_default_str(qdict, BDRV_OPT_AUTO_READ_ONLY, "off");
2538
e1d74bc6
KW
2539 }
2540
2541 bs = bdrv_open_inherit(NULL, reference, qdict, 0, NULL, NULL, errp);
2542 obj = NULL;
2543
2544fail:
cb3e7f08 2545 qobject_unref(obj);
e1d74bc6
KW
2546 visit_free(v);
2547 return bs;
2548}
2549
66836189
HR
2550static BlockDriverState *bdrv_append_temp_snapshot(BlockDriverState *bs,
2551 int flags,
2552 QDict *snapshot_options,
2553 Error **errp)
b998875d
KW
2554{
2555 /* TODO: extra byte is a hack to ensure MAX_PATH space on Windows. */
1ba4b6a5 2556 char *tmp_filename = g_malloc0(PATH_MAX + 1);
b998875d 2557 int64_t total_size;
83d0521a 2558 QemuOpts *opts = NULL;
ff6ed714 2559 BlockDriverState *bs_snapshot = NULL;
b2c2832c 2560 Error *local_err = NULL;
b998875d
KW
2561 int ret;
2562
2563 /* if snapshot, we create a temporary backing file and open it
2564 instead of opening 'filename' directly */
2565
2566 /* Get the required size from the image */
f187743a
KW
2567 total_size = bdrv_getlength(bs);
2568 if (total_size < 0) {
2569 error_setg_errno(errp, -total_size, "Could not get image size");
1ba4b6a5 2570 goto out;
f187743a 2571 }
b998875d
KW
2572
2573 /* Create the temporary image */
1ba4b6a5 2574 ret = get_tmp_filename(tmp_filename, PATH_MAX + 1);
b998875d
KW
2575 if (ret < 0) {
2576 error_setg_errno(errp, -ret, "Could not get temporary filename");
1ba4b6a5 2577 goto out;
b998875d
KW
2578 }
2579
ef810437 2580 opts = qemu_opts_create(bdrv_qcow2.create_opts, NULL, 0,
c282e1fd 2581 &error_abort);
39101f25 2582 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, total_size, &error_abort);
e43bfd9c 2583 ret = bdrv_create(&bdrv_qcow2, tmp_filename, opts, errp);
83d0521a 2584 qemu_opts_del(opts);
b998875d 2585 if (ret < 0) {
e43bfd9c
MA
2586 error_prepend(errp, "Could not create temporary overlay '%s': ",
2587 tmp_filename);
1ba4b6a5 2588 goto out;
b998875d
KW
2589 }
2590
73176bee 2591 /* Prepare options QDict for the temporary file */
46f5ac20
EB
2592 qdict_put_str(snapshot_options, "file.driver", "file");
2593 qdict_put_str(snapshot_options, "file.filename", tmp_filename);
2594 qdict_put_str(snapshot_options, "driver", "qcow2");
b998875d 2595
5b363937 2596 bs_snapshot = bdrv_open(NULL, NULL, snapshot_options, flags, errp);
73176bee 2597 snapshot_options = NULL;
5b363937 2598 if (!bs_snapshot) {
1ba4b6a5 2599 goto out;
b998875d
KW
2600 }
2601
ff6ed714
EB
2602 /* bdrv_append() consumes a strong reference to bs_snapshot
2603 * (i.e. it will call bdrv_unref() on it) even on error, so in
2604 * order to be able to return one, we have to increase
2605 * bs_snapshot's refcount here */
66836189 2606 bdrv_ref(bs_snapshot);
b2c2832c
KW
2607 bdrv_append(bs_snapshot, bs, &local_err);
2608 if (local_err) {
2609 error_propagate(errp, local_err);
ff6ed714 2610 bs_snapshot = NULL;
b2c2832c
KW
2611 goto out;
2612 }
1ba4b6a5
BC
2613
2614out:
cb3e7f08 2615 qobject_unref(snapshot_options);
1ba4b6a5 2616 g_free(tmp_filename);
ff6ed714 2617 return bs_snapshot;
b998875d
KW
2618}
2619
b6ce07aa
KW
2620/*
2621 * Opens a disk image (raw, qcow2, vmdk, ...)
de9c0cec
KW
2622 *
2623 * options is a QDict of options to pass to the block drivers, or NULL for an
2624 * empty set of options. The reference to the QDict belongs to the block layer
2625 * after the call (even on failure), so if the caller intends to reuse the
cb3e7f08 2626 * dictionary, it needs to use qobject_ref() before calling bdrv_open.
f67503e5
HR
2627 *
2628 * If *pbs is NULL, a new BDS will be created with a pointer to it stored there.
2629 * If it is not NULL, the referenced BDS will be reused.
ddf5636d
HR
2630 *
2631 * The reference parameter may be used to specify an existing block device which
2632 * should be opened. If specified, neither options nor a filename may be given,
2633 * nor can an existing BDS be reused (that is, *pbs has to be NULL).
b6ce07aa 2634 */
5b363937
HR
2635static BlockDriverState *bdrv_open_inherit(const char *filename,
2636 const char *reference,
2637 QDict *options, int flags,
2638 BlockDriverState *parent,
2639 const BdrvChildRole *child_role,
2640 Error **errp)
ea2384d3 2641{
b6ce07aa 2642 int ret;
5696c6e3 2643 BlockBackend *file = NULL;
9a4f4c31 2644 BlockDriverState *bs;
ce343771 2645 BlockDriver *drv = NULL;
2f624b80 2646 BdrvChild *child;
74fe54f2 2647 const char *drvname;
3e8c2e57 2648 const char *backing;
34b5d2c6 2649 Error *local_err = NULL;
73176bee 2650 QDict *snapshot_options = NULL;
b1e6fc08 2651 int snapshot_flags = 0;
712e7874 2652
f3930ed0
KW
2653 assert(!child_role || !flags);
2654 assert(!child_role == !parent);
f67503e5 2655
ddf5636d
HR
2656 if (reference) {
2657 bool options_non_empty = options ? qdict_size(options) : false;
cb3e7f08 2658 qobject_unref(options);
ddf5636d 2659
ddf5636d
HR
2660 if (filename || options_non_empty) {
2661 error_setg(errp, "Cannot reference an existing block device with "
2662 "additional options or a new filename");
5b363937 2663 return NULL;
ddf5636d
HR
2664 }
2665
2666 bs = bdrv_lookup_bs(reference, reference, errp);
2667 if (!bs) {
5b363937 2668 return NULL;
ddf5636d 2669 }
76b22320 2670
ddf5636d 2671 bdrv_ref(bs);
5b363937 2672 return bs;
ddf5636d
HR
2673 }
2674
5b363937 2675 bs = bdrv_new();
f67503e5 2676
de9c0cec
KW
2677 /* NULL means an empty set of options */
2678 if (options == NULL) {
2679 options = qdict_new();
2680 }
2681
145f598e 2682 /* json: syntax counts as explicit options, as if in the QDict */
de3b53f0
KW
2683 parse_json_protocol(options, &filename, &local_err);
2684 if (local_err) {
de3b53f0
KW
2685 goto fail;
2686 }
2687
145f598e
KW
2688 bs->explicit_options = qdict_clone_shallow(options);
2689
f3930ed0 2690 if (child_role) {
bddcec37 2691 bs->inherits_from = parent;
8e2160e2
KW
2692 child_role->inherit_options(&flags, options,
2693 parent->open_flags, parent->options);
f3930ed0
KW
2694 }
2695
de3b53f0 2696 ret = bdrv_fill_options(&options, filename, &flags, &local_err);
462f5bcf
KW
2697 if (local_err) {
2698 goto fail;
2699 }
2700
129c7d1c
MA
2701 /*
2702 * Set the BDRV_O_RDWR and BDRV_O_ALLOW_RDWR flags.
2703 * Caution: getting a boolean member of @options requires care.
2704 * When @options come from -blockdev or blockdev_add, members are
2705 * typed according to the QAPI schema, but when they come from
2706 * -drive, they're all QString.
2707 */
f87a0e29
AG
2708 if (g_strcmp0(qdict_get_try_str(options, BDRV_OPT_READ_ONLY), "on") &&
2709 !qdict_get_try_bool(options, BDRV_OPT_READ_ONLY, false)) {
2710 flags |= (BDRV_O_RDWR | BDRV_O_ALLOW_RDWR);
2711 } else {
2712 flags &= ~BDRV_O_RDWR;
14499ea5
AG
2713 }
2714
2715 if (flags & BDRV_O_SNAPSHOT) {
2716 snapshot_options = qdict_new();
2717 bdrv_temp_snapshot_options(&snapshot_flags, snapshot_options,
2718 flags, options);
f87a0e29
AG
2719 /* Let bdrv_backing_options() override "read-only" */
2720 qdict_del(options, BDRV_OPT_READ_ONLY);
14499ea5
AG
2721 bdrv_backing_options(&flags, options, flags, options);
2722 }
2723
62392ebb
KW
2724 bs->open_flags = flags;
2725 bs->options = options;
2726 options = qdict_clone_shallow(options);
2727
76c591b0 2728 /* Find the right image format driver */
129c7d1c 2729 /* See cautionary note on accessing @options above */
76c591b0
KW
2730 drvname = qdict_get_try_str(options, "driver");
2731 if (drvname) {
2732 drv = bdrv_find_format(drvname);
76c591b0
KW
2733 if (!drv) {
2734 error_setg(errp, "Unknown driver: '%s'", drvname);
76c591b0
KW
2735 goto fail;
2736 }
2737 }
2738
2739 assert(drvname || !(flags & BDRV_O_PROTOCOL));
76c591b0 2740
129c7d1c 2741 /* See cautionary note on accessing @options above */
3e8c2e57 2742 backing = qdict_get_try_str(options, "backing");
e59a0cf1
HR
2743 if (qobject_to(QNull, qdict_get(options, "backing")) != NULL ||
2744 (backing && *backing == '\0'))
2745 {
4f7be280
HR
2746 if (backing) {
2747 warn_report("Use of \"backing\": \"\" is deprecated; "
2748 "use \"backing\": null instead");
2749 }
3e8c2e57
AG
2750 flags |= BDRV_O_NO_BACKING;
2751 qdict_del(options, "backing");
2752 }
2753
5696c6e3 2754 /* Open image file without format layer. This BlockBackend is only used for
4e4bf5c4
KW
2755 * probing, the block drivers will do their own bdrv_open_child() for the
2756 * same BDS, which is why we put the node name back into options. */
f4788adc 2757 if ((flags & BDRV_O_PROTOCOL) == 0) {
5696c6e3
KW
2758 BlockDriverState *file_bs;
2759
2760 file_bs = bdrv_open_child_bs(filename, options, "file", bs,
2761 &child_file, true, &local_err);
1fdd6933 2762 if (local_err) {
f4788adc
KW
2763 goto fail;
2764 }
5696c6e3 2765 if (file_bs != NULL) {
dacaa162
KW
2766 /* Not requesting BLK_PERM_CONSISTENT_READ because we're only
2767 * looking at the header to guess the image format. This works even
2768 * in cases where a guest would not see a consistent state. */
2769 file = blk_new(0, BLK_PERM_ALL);
d7086422 2770 blk_insert_bs(file, file_bs, &local_err);
5696c6e3 2771 bdrv_unref(file_bs);
d7086422
KW
2772 if (local_err) {
2773 goto fail;
2774 }
5696c6e3 2775
46f5ac20 2776 qdict_put_str(options, "file", bdrv_get_node_name(file_bs));
4e4bf5c4 2777 }
f500a6d3
KW
2778 }
2779
76c591b0 2780 /* Image format probing */
38f3ef57 2781 bs->probed = !drv;
76c591b0 2782 if (!drv && file) {
cf2ab8fc 2783 ret = find_image_format(file, filename, &drv, &local_err);
17b005f1 2784 if (ret < 0) {
8bfea15d 2785 goto fail;
2a05cbe4 2786 }
62392ebb
KW
2787 /*
2788 * This option update would logically belong in bdrv_fill_options(),
2789 * but we first need to open bs->file for the probing to work, while
2790 * opening bs->file already requires the (mostly) final set of options
2791 * so that cache mode etc. can be inherited.
2792 *
2793 * Adding the driver later is somewhat ugly, but it's not an option
2794 * that would ever be inherited, so it's correct. We just need to make
2795 * sure to update both bs->options (which has the full effective
2796 * options for bs) and options (which has file.* already removed).
2797 */
46f5ac20
EB
2798 qdict_put_str(bs->options, "driver", drv->format_name);
2799 qdict_put_str(options, "driver", drv->format_name);
76c591b0 2800 } else if (!drv) {
17b005f1 2801 error_setg(errp, "Must specify either driver or file");
8bfea15d 2802 goto fail;
ea2384d3 2803 }
b6ce07aa 2804
53a29513
HR
2805 /* BDRV_O_PROTOCOL must be set iff a protocol BDS is about to be created */
2806 assert(!!(flags & BDRV_O_PROTOCOL) == !!drv->bdrv_file_open);
2807 /* file must be NULL if a protocol BDS is about to be created
2808 * (the inverse results in an error message from bdrv_open_common()) */
2809 assert(!(flags & BDRV_O_PROTOCOL) || !file);
2810
b6ce07aa 2811 /* Open the image */
82dc8b41 2812 ret = bdrv_open_common(bs, file, options, &local_err);
b6ce07aa 2813 if (ret < 0) {
8bfea15d 2814 goto fail;
6987307c
CH
2815 }
2816
4e4bf5c4 2817 if (file) {
5696c6e3 2818 blk_unref(file);
f500a6d3
KW
2819 file = NULL;
2820 }
2821
b6ce07aa 2822 /* If there is a backing file, use it */
9156df12 2823 if ((flags & BDRV_O_NO_BACKING) == 0) {
d9b7b057 2824 ret = bdrv_open_backing_file(bs, options, "backing", &local_err);
b6ce07aa 2825 if (ret < 0) {
b6ad491a 2826 goto close_and_fail;
b6ce07aa 2827 }
b6ce07aa
KW
2828 }
2829
50196d7a
AG
2830 /* Remove all children options and references
2831 * from bs->options and bs->explicit_options */
2f624b80
AG
2832 QLIST_FOREACH(child, &bs->children, next) {
2833 char *child_key_dot;
2834 child_key_dot = g_strdup_printf("%s.", child->name);
2835 qdict_extract_subqdict(bs->explicit_options, NULL, child_key_dot);
2836 qdict_extract_subqdict(bs->options, NULL, child_key_dot);
50196d7a
AG
2837 qdict_del(bs->explicit_options, child->name);
2838 qdict_del(bs->options, child->name);
2f624b80
AG
2839 g_free(child_key_dot);
2840 }
2841
91af7014
HR
2842 bdrv_refresh_filename(bs);
2843
b6ad491a 2844 /* Check if any unknown options were used */
7ad2757f 2845 if (qdict_size(options) != 0) {
b6ad491a 2846 const QDictEntry *entry = qdict_first(options);
5acd9d81
HR
2847 if (flags & BDRV_O_PROTOCOL) {
2848 error_setg(errp, "Block protocol '%s' doesn't support the option "
2849 "'%s'", drv->format_name, entry->key);
2850 } else {
d0e46a55
HR
2851 error_setg(errp,
2852 "Block format '%s' does not support the option '%s'",
2853 drv->format_name, entry->key);
5acd9d81 2854 }
b6ad491a 2855
b6ad491a
KW
2856 goto close_and_fail;
2857 }
b6ad491a 2858
c01c214b 2859 bdrv_parent_cb_change_media(bs, true);
b6ce07aa 2860
cb3e7f08 2861 qobject_unref(options);
8961be33 2862 options = NULL;
dd62f1ca
KW
2863
2864 /* For snapshot=on, create a temporary qcow2 overlay. bs points to the
2865 * temporary snapshot afterwards. */
2866 if (snapshot_flags) {
66836189
HR
2867 BlockDriverState *snapshot_bs;
2868 snapshot_bs = bdrv_append_temp_snapshot(bs, snapshot_flags,
2869 snapshot_options, &local_err);
73176bee 2870 snapshot_options = NULL;
dd62f1ca
KW
2871 if (local_err) {
2872 goto close_and_fail;
2873 }
5b363937
HR
2874 /* We are not going to return bs but the overlay on top of it
2875 * (snapshot_bs); thus, we have to drop the strong reference to bs
2876 * (which we obtained by calling bdrv_new()). bs will not be deleted,
2877 * though, because the overlay still has a reference to it. */
2878 bdrv_unref(bs);
2879 bs = snapshot_bs;
dd62f1ca
KW
2880 }
2881
5b363937 2882 return bs;
b6ce07aa 2883
8bfea15d 2884fail:
5696c6e3 2885 blk_unref(file);
cb3e7f08
MAL
2886 qobject_unref(snapshot_options);
2887 qobject_unref(bs->explicit_options);
2888 qobject_unref(bs->options);
2889 qobject_unref(options);
de9c0cec 2890 bs->options = NULL;
998cbd6a 2891 bs->explicit_options = NULL;
5b363937 2892 bdrv_unref(bs);
621ff94d 2893 error_propagate(errp, local_err);
5b363937 2894 return NULL;
de9c0cec 2895
b6ad491a 2896close_and_fail:
5b363937 2897 bdrv_unref(bs);
cb3e7f08
MAL
2898 qobject_unref(snapshot_options);
2899 qobject_unref(options);
621ff94d 2900 error_propagate(errp, local_err);
5b363937 2901 return NULL;
b6ce07aa
KW
2902}
2903
5b363937
HR
2904BlockDriverState *bdrv_open(const char *filename, const char *reference,
2905 QDict *options, int flags, Error **errp)
f3930ed0 2906{
5b363937 2907 return bdrv_open_inherit(filename, reference, options, flags, NULL,
ce343771 2908 NULL, errp);
f3930ed0
KW
2909}
2910
e971aa12
JC
2911/*
2912 * Adds a BlockDriverState to a simple queue for an atomic, transactional
2913 * reopen of multiple devices.
2914 *
2915 * bs_queue can either be an existing BlockReopenQueue that has had QSIMPLE_INIT
2916 * already performed, or alternatively may be NULL a new BlockReopenQueue will
2917 * be created and initialized. This newly created BlockReopenQueue should be
2918 * passed back in for subsequent calls that are intended to be of the same
2919 * atomic 'set'.
2920 *
2921 * bs is the BlockDriverState to add to the reopen queue.
2922 *
4d2cb092
KW
2923 * options contains the changed options for the associated bs
2924 * (the BlockReopenQueue takes ownership)
2925 *
e971aa12
JC
2926 * flags contains the open flags for the associated bs
2927 *
2928 * returns a pointer to bs_queue, which is either the newly allocated
2929 * bs_queue, or the existing bs_queue being used.
2930 *
1a63a907 2931 * bs must be drained between bdrv_reopen_queue() and bdrv_reopen_multiple().
e971aa12 2932 */
28518102
KW
2933static BlockReopenQueue *bdrv_reopen_queue_child(BlockReopenQueue *bs_queue,
2934 BlockDriverState *bs,
2935 QDict *options,
28518102
KW
2936 const BdrvChildRole *role,
2937 QDict *parent_options,
2938 int parent_flags)
e971aa12
JC
2939{
2940 assert(bs != NULL);
2941
2942 BlockReopenQueueEntry *bs_entry;
67251a31 2943 BdrvChild *child;
9aa09ddd
AG
2944 QDict *old_options, *explicit_options, *options_copy;
2945 int flags;
2946 QemuOpts *opts;
67251a31 2947
1a63a907
KW
2948 /* Make sure that the caller remembered to use a drained section. This is
2949 * important to avoid graph changes between the recursive queuing here and
2950 * bdrv_reopen_multiple(). */
2951 assert(bs->quiesce_counter > 0);
2952
e971aa12
JC
2953 if (bs_queue == NULL) {
2954 bs_queue = g_new0(BlockReopenQueue, 1);
2955 QSIMPLEQ_INIT(bs_queue);
2956 }
2957
4d2cb092
KW
2958 if (!options) {
2959 options = qdict_new();
2960 }
2961
5b7ba05f
AG
2962 /* Check if this BlockDriverState is already in the queue */
2963 QSIMPLEQ_FOREACH(bs_entry, bs_queue, entry) {
2964 if (bs == bs_entry->state.bs) {
2965 break;
2966 }
2967 }
2968
28518102
KW
2969 /*
2970 * Precedence of options:
2971 * 1. Explicitly passed in options (highest)
9aa09ddd
AG
2972 * 2. Retained from explicitly set options of bs
2973 * 3. Inherited from parent node
2974 * 4. Retained from effective options of bs
28518102
KW
2975 */
2976
145f598e 2977 /* Old explicitly set values (don't overwrite by inherited value) */
5b7ba05f
AG
2978 if (bs_entry) {
2979 old_options = qdict_clone_shallow(bs_entry->state.explicit_options);
2980 } else {
2981 old_options = qdict_clone_shallow(bs->explicit_options);
2982 }
145f598e 2983 bdrv_join_options(bs, options, old_options);
cb3e7f08 2984 qobject_unref(old_options);
145f598e
KW
2985
2986 explicit_options = qdict_clone_shallow(options);
2987
28518102
KW
2988 /* Inherit from parent node */
2989 if (parent_options) {
9aa09ddd 2990 flags = 0;
8e2160e2 2991 role->inherit_options(&flags, options, parent_flags, parent_options);
9aa09ddd
AG
2992 } else {
2993 flags = bdrv_get_flags(bs);
28518102
KW
2994 }
2995
2996 /* Old values are used for options that aren't set yet */
4d2cb092 2997 old_options = qdict_clone_shallow(bs->options);
cddff5ba 2998 bdrv_join_options(bs, options, old_options);
cb3e7f08 2999 qobject_unref(old_options);
4d2cb092 3000
9aa09ddd
AG
3001 /* We have the final set of options so let's update the flags */
3002 options_copy = qdict_clone_shallow(options);
3003 opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
3004 qemu_opts_absorb_qdict(opts, options_copy, NULL);
3005 update_flags_from_options(&flags, opts);
3006 qemu_opts_del(opts);
3007 qobject_unref(options_copy);
3008
fd452021 3009 /* bdrv_open_inherit() sets and clears some additional flags internally */
f1f25a2e 3010 flags &= ~BDRV_O_PROTOCOL;
fd452021
KW
3011 if (flags & BDRV_O_RDWR) {
3012 flags |= BDRV_O_ALLOW_RDWR;
3013 }
f1f25a2e 3014
1857c97b
KW
3015 if (!bs_entry) {
3016 bs_entry = g_new0(BlockReopenQueueEntry, 1);
3017 QSIMPLEQ_INSERT_TAIL(bs_queue, bs_entry, entry);
3018 } else {
cb3e7f08
MAL
3019 qobject_unref(bs_entry->state.options);
3020 qobject_unref(bs_entry->state.explicit_options);
1857c97b
KW
3021 }
3022
3023 bs_entry->state.bs = bs;
3024 bs_entry->state.options = options;
3025 bs_entry->state.explicit_options = explicit_options;
3026 bs_entry->state.flags = flags;
3027
30450259
KW
3028 /* This needs to be overwritten in bdrv_reopen_prepare() */
3029 bs_entry->state.perm = UINT64_MAX;
3030 bs_entry->state.shared_perm = 0;
3031
67251a31 3032 QLIST_FOREACH(child, &bs->children, next) {
4c9dfe5d
KW
3033 QDict *new_child_options;
3034 char *child_key_dot;
67251a31 3035
4c9dfe5d
KW
3036 /* reopen can only change the options of block devices that were
3037 * implicitly created and inherited options. For other (referenced)
3038 * block devices, a syntax like "backing.foo" results in an error. */
67251a31
KW
3039 if (child->bs->inherits_from != bs) {
3040 continue;
3041 }
3042
4c9dfe5d 3043 child_key_dot = g_strdup_printf("%s.", child->name);
2f624b80 3044 qdict_extract_subqdict(explicit_options, NULL, child_key_dot);
4c9dfe5d
KW
3045 qdict_extract_subqdict(options, &new_child_options, child_key_dot);
3046 g_free(child_key_dot);
3047
9aa09ddd 3048 bdrv_reopen_queue_child(bs_queue, child->bs, new_child_options,
28518102 3049 child->role, options, flags);
e971aa12
JC
3050 }
3051
e971aa12
JC
3052 return bs_queue;
3053}
3054
28518102
KW
3055BlockReopenQueue *bdrv_reopen_queue(BlockReopenQueue *bs_queue,
3056 BlockDriverState *bs,
2e891722 3057 QDict *options)
28518102 3058{
9aa09ddd 3059 return bdrv_reopen_queue_child(bs_queue, bs, options, NULL, NULL, 0);
28518102
KW
3060}
3061
e971aa12
JC
3062/*
3063 * Reopen multiple BlockDriverStates atomically & transactionally.
3064 *
3065 * The queue passed in (bs_queue) must have been built up previous
3066 * via bdrv_reopen_queue().
3067 *
3068 * Reopens all BDS specified in the queue, with the appropriate
3069 * flags. All devices are prepared for reopen, and failure of any
50d6a8a3 3070 * device will cause all device changes to be abandoned, and intermediate
e971aa12
JC
3071 * data cleaned up.
3072 *
3073 * If all devices prepare successfully, then the changes are committed
3074 * to all devices.
3075 *
1a63a907
KW
3076 * All affected nodes must be drained between bdrv_reopen_queue() and
3077 * bdrv_reopen_multiple().
e971aa12 3078 */
720150f3 3079int bdrv_reopen_multiple(AioContext *ctx, BlockReopenQueue *bs_queue, Error **errp)
e971aa12
JC
3080{
3081 int ret = -1;
3082 BlockReopenQueueEntry *bs_entry, *next;
3083 Error *local_err = NULL;
3084
3085 assert(bs_queue != NULL);
3086
e971aa12 3087 QSIMPLEQ_FOREACH(bs_entry, bs_queue, entry) {
1a63a907 3088 assert(bs_entry->state.bs->quiesce_counter > 0);
e971aa12
JC
3089 if (bdrv_reopen_prepare(&bs_entry->state, bs_queue, &local_err)) {
3090 error_propagate(errp, local_err);
3091 goto cleanup;
3092 }
3093 bs_entry->prepared = true;
3094 }
3095
3096 /* If we reach this point, we have success and just need to apply the
3097 * changes
3098 */
3099 QSIMPLEQ_FOREACH(bs_entry, bs_queue, entry) {
3100 bdrv_reopen_commit(&bs_entry->state);
3101 }
3102
3103 ret = 0;
3104
3105cleanup:
3106 QSIMPLEQ_FOREACH_SAFE(bs_entry, bs_queue, entry, next) {
1bab38e7
AG
3107 if (ret) {
3108 if (bs_entry->prepared) {
3109 bdrv_reopen_abort(&bs_entry->state);
3110 }
cb3e7f08 3111 qobject_unref(bs_entry->state.explicit_options);
4c8350fe 3112 qobject_unref(bs_entry->state.options);
e971aa12
JC
3113 }
3114 g_free(bs_entry);
3115 }
3116 g_free(bs_queue);
40840e41 3117
e971aa12
JC
3118 return ret;
3119}
3120
6e1000a8
AG
3121int bdrv_reopen_set_read_only(BlockDriverState *bs, bool read_only,
3122 Error **errp)
3123{
3124 int ret;
3125 BlockReopenQueue *queue;
3126 QDict *opts = qdict_new();
3127
3128 qdict_put_bool(opts, BDRV_OPT_READ_ONLY, read_only);
3129
3130 bdrv_subtree_drained_begin(bs);
2e891722 3131 queue = bdrv_reopen_queue(NULL, bs, opts);
6e1000a8
AG
3132 ret = bdrv_reopen_multiple(bdrv_get_aio_context(bs), queue, errp);
3133 bdrv_subtree_drained_end(bs);
3134
3135 return ret;
3136}
3137
30450259
KW
3138static BlockReopenQueueEntry *find_parent_in_reopen_queue(BlockReopenQueue *q,
3139 BdrvChild *c)
3140{
3141 BlockReopenQueueEntry *entry;
3142
3143 QSIMPLEQ_FOREACH(entry, q, entry) {
3144 BlockDriverState *bs = entry->state.bs;
3145 BdrvChild *child;
3146
3147 QLIST_FOREACH(child, &bs->children, next) {
3148 if (child == c) {
3149 return entry;
3150 }
3151 }
3152 }
3153
3154 return NULL;
3155}
3156
3157static void bdrv_reopen_perm(BlockReopenQueue *q, BlockDriverState *bs,
3158 uint64_t *perm, uint64_t *shared)
3159{
3160 BdrvChild *c;
3161 BlockReopenQueueEntry *parent;
3162 uint64_t cumulative_perms = 0;
3163 uint64_t cumulative_shared_perms = BLK_PERM_ALL;
3164
3165 QLIST_FOREACH(c, &bs->parents, next_parent) {
3166 parent = find_parent_in_reopen_queue(q, c);
3167 if (!parent) {
3168 cumulative_perms |= c->perm;
3169 cumulative_shared_perms &= c->shared_perm;
3170 } else {
3171 uint64_t nperm, nshared;
3172
3173 bdrv_child_perm(parent->state.bs, bs, c, c->role, q,
3174 parent->state.perm, parent->state.shared_perm,
3175 &nperm, &nshared);
3176
3177 cumulative_perms |= nperm;
3178 cumulative_shared_perms &= nshared;
3179 }
3180 }
3181 *perm = cumulative_perms;
3182 *shared = cumulative_shared_perms;
3183}
e971aa12
JC
3184
3185/*
3186 * Prepares a BlockDriverState for reopen. All changes are staged in the
3187 * 'opaque' field of the BDRVReopenState, which is used and allocated by
3188 * the block driver layer .bdrv_reopen_prepare()
3189 *
3190 * bs is the BlockDriverState to reopen
3191 * flags are the new open flags
3192 * queue is the reopen queue
3193 *
3194 * Returns 0 on success, non-zero on error. On error errp will be set
3195 * as well.
3196 *
3197 * On failure, bdrv_reopen_abort() will be called to clean up any data.
3198 * It is the responsibility of the caller to then call the abort() or
3199 * commit() for any other BDS that have been left in a prepare() state
3200 *
3201 */
3202int bdrv_reopen_prepare(BDRVReopenState *reopen_state, BlockReopenQueue *queue,
3203 Error **errp)
3204{
3205 int ret = -1;
e6d79c41 3206 int old_flags;
e971aa12
JC
3207 Error *local_err = NULL;
3208 BlockDriver *drv;
ccf9dc07 3209 QemuOpts *opts;
4c8350fe 3210 QDict *orig_reopen_opts;
593b3071 3211 char *discard = NULL;
3d8ce171 3212 bool read_only;
9ad08c44 3213 bool drv_prepared = false;
e971aa12
JC
3214
3215 assert(reopen_state != NULL);
3216 assert(reopen_state->bs->drv != NULL);
3217 drv = reopen_state->bs->drv;
3218
4c8350fe
AG
3219 /* This function and each driver's bdrv_reopen_prepare() remove
3220 * entries from reopen_state->options as they are processed, so
3221 * we need to make a copy of the original QDict. */
3222 orig_reopen_opts = qdict_clone_shallow(reopen_state->options);
3223
ccf9dc07
KW
3224 /* Process generic block layer options */
3225 opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
3226 qemu_opts_absorb_qdict(opts, reopen_state->options, &local_err);
3227 if (local_err) {
3228 error_propagate(errp, local_err);
3229 ret = -EINVAL;
3230 goto error;
3231 }
3232
e6d79c41
AG
3233 /* This was already called in bdrv_reopen_queue_child() so the flags
3234 * are up-to-date. This time we simply want to remove the options from
3235 * QemuOpts in order to indicate that they have been processed. */
3236 old_flags = reopen_state->flags;
91a097e7 3237 update_flags_from_options(&reopen_state->flags, opts);
e6d79c41 3238 assert(old_flags == reopen_state->flags);
91a097e7 3239
415bbca8 3240 discard = qemu_opt_get_del(opts, BDRV_OPT_DISCARD);
593b3071
AG
3241 if (discard != NULL) {
3242 if (bdrv_parse_discard_flags(discard, &reopen_state->flags) != 0) {
3243 error_setg(errp, "Invalid discard option");
3244 ret = -EINVAL;
3245 goto error;
3246 }
3247 }
3248
543770bd
AG
3249 reopen_state->detect_zeroes =
3250 bdrv_parse_detect_zeroes(opts, reopen_state->flags, &local_err);
3251 if (local_err) {
3252 error_propagate(errp, local_err);
3253 ret = -EINVAL;
3254 goto error;
3255 }
3256
57f9db9a
AG
3257 /* All other options (including node-name and driver) must be unchanged.
3258 * Put them back into the QDict, so that they are checked at the end
3259 * of this function. */
3260 qemu_opts_to_qdict(opts, reopen_state->options);
ccf9dc07 3261
3d8ce171
JC
3262 /* If we are to stay read-only, do not allow permission change
3263 * to r/w. Attempting to set to r/w may fail if either BDRV_O_ALLOW_RDWR is
3264 * not set, or if the BDS still has copy_on_read enabled */
3265 read_only = !(reopen_state->flags & BDRV_O_RDWR);
54a32bfe 3266 ret = bdrv_can_set_read_only(reopen_state->bs, read_only, true, &local_err);
3d8ce171
JC
3267 if (local_err) {
3268 error_propagate(errp, local_err);
e971aa12
JC
3269 goto error;
3270 }
3271
30450259
KW
3272 /* Calculate required permissions after reopening */
3273 bdrv_reopen_perm(queue, reopen_state->bs,
3274 &reopen_state->perm, &reopen_state->shared_perm);
e971aa12
JC
3275
3276 ret = bdrv_flush(reopen_state->bs);
3277 if (ret) {
455b0fde 3278 error_setg_errno(errp, -ret, "Error flushing drive");
e971aa12
JC
3279 goto error;
3280 }
3281
3282 if (drv->bdrv_reopen_prepare) {
3283 ret = drv->bdrv_reopen_prepare(reopen_state, queue, &local_err);
3284 if (ret) {
3285 if (local_err != NULL) {
3286 error_propagate(errp, local_err);
3287 } else {
d8b6895f
LC
3288 error_setg(errp, "failed while preparing to reopen image '%s'",
3289 reopen_state->bs->filename);
e971aa12
JC
3290 }
3291 goto error;
3292 }
3293 } else {
3294 /* It is currently mandatory to have a bdrv_reopen_prepare()
3295 * handler for each supported drv. */
81e5f78a
AG
3296 error_setg(errp, "Block format '%s' used by node '%s' "
3297 "does not support reopening files", drv->format_name,
3298 bdrv_get_device_or_node_name(reopen_state->bs));
e971aa12
JC
3299 ret = -1;
3300 goto error;
3301 }
3302
9ad08c44
HR
3303 drv_prepared = true;
3304
4d2cb092
KW
3305 /* Options that are not handled are only okay if they are unchanged
3306 * compared to the old state. It is expected that some options are only
3307 * used for the initial open, but not reopen (e.g. filename) */
3308 if (qdict_size(reopen_state->options)) {
3309 const QDictEntry *entry = qdict_first(reopen_state->options);
3310
3311 do {
54fd1b0d
HR
3312 QObject *new = entry->value;
3313 QObject *old = qdict_get(reopen_state->bs->options, entry->key);
3314
db905283
AG
3315 /* Allow child references (child_name=node_name) as long as they
3316 * point to the current child (i.e. everything stays the same). */
3317 if (qobject_type(new) == QTYPE_QSTRING) {
3318 BdrvChild *child;
3319 QLIST_FOREACH(child, &reopen_state->bs->children, next) {
3320 if (!strcmp(child->name, entry->key)) {
3321 break;
3322 }
3323 }
3324
3325 if (child) {
3326 const char *str = qobject_get_try_str(new);
3327 if (!strcmp(child->bs->node_name, str)) {
3328 continue; /* Found child with this name, skip option */
3329 }
3330 }
3331 }
3332
129c7d1c 3333 /*
54fd1b0d
HR
3334 * TODO: When using -drive to specify blockdev options, all values
3335 * will be strings; however, when using -blockdev, blockdev-add or
3336 * filenames using the json:{} pseudo-protocol, they will be
3337 * correctly typed.
3338 * In contrast, reopening options are (currently) always strings
3339 * (because you can only specify them through qemu-io; all other
3340 * callers do not specify any options).
3341 * Therefore, when using anything other than -drive to create a BDS,
3342 * this cannot detect non-string options as unchanged, because
3343 * qobject_is_equal() always returns false for objects of different
3344 * type. In the future, this should be remedied by correctly typing
3345 * all options. For now, this is not too big of an issue because
3346 * the user can simply omit options which cannot be changed anyway,
3347 * so they will stay unchanged.
129c7d1c 3348 */
54fd1b0d 3349 if (!qobject_is_equal(new, old)) {
4d2cb092
KW
3350 error_setg(errp, "Cannot change the option '%s'", entry->key);
3351 ret = -EINVAL;
3352 goto error;
3353 }
3354 } while ((entry = qdict_next(reopen_state->options, entry)));
3355 }
3356
30450259
KW
3357 ret = bdrv_check_perm(reopen_state->bs, queue, reopen_state->perm,
3358 reopen_state->shared_perm, NULL, errp);
3359 if (ret < 0) {
3360 goto error;
3361 }
3362
e971aa12
JC
3363 ret = 0;
3364
4c8350fe
AG
3365 /* Restore the original reopen_state->options QDict */
3366 qobject_unref(reopen_state->options);
3367 reopen_state->options = qobject_ref(orig_reopen_opts);
3368
e971aa12 3369error:
9ad08c44
HR
3370 if (ret < 0 && drv_prepared) {
3371 /* drv->bdrv_reopen_prepare() has succeeded, so we need to
3372 * call drv->bdrv_reopen_abort() before signaling an error
3373 * (bdrv_reopen_multiple() will not call bdrv_reopen_abort()
3374 * when the respective bdrv_reopen_prepare() has failed) */
3375 if (drv->bdrv_reopen_abort) {
3376 drv->bdrv_reopen_abort(reopen_state);
3377 }
3378 }
ccf9dc07 3379 qemu_opts_del(opts);
4c8350fe 3380 qobject_unref(orig_reopen_opts);
593b3071 3381 g_free(discard);
e971aa12
JC
3382 return ret;
3383}
3384
3385/*
3386 * Takes the staged changes for the reopen from bdrv_reopen_prepare(), and
3387 * makes them final by swapping the staging BlockDriverState contents into
3388 * the active BlockDriverState contents.
3389 */
3390void bdrv_reopen_commit(BDRVReopenState *reopen_state)
3391{
3392 BlockDriver *drv;
50bf65ba 3393 BlockDriverState *bs;
50196d7a 3394 BdrvChild *child;
cb9ff6c2 3395 bool old_can_write, new_can_write;
e971aa12
JC
3396
3397 assert(reopen_state != NULL);
50bf65ba
VSO
3398 bs = reopen_state->bs;
3399 drv = bs->drv;
e971aa12
JC
3400 assert(drv != NULL);
3401
cb9ff6c2
VSO
3402 old_can_write =
3403 !bdrv_is_read_only(bs) && !(bdrv_get_flags(bs) & BDRV_O_INACTIVE);
3404
e971aa12
JC
3405 /* If there are any driver level actions to take */
3406 if (drv->bdrv_reopen_commit) {
3407 drv->bdrv_reopen_commit(reopen_state);
3408 }
3409
3410 /* set BDS specific flags now */
cb3e7f08 3411 qobject_unref(bs->explicit_options);
4c8350fe 3412 qobject_unref(bs->options);
145f598e 3413
50bf65ba 3414 bs->explicit_options = reopen_state->explicit_options;
4c8350fe 3415 bs->options = reopen_state->options;
50bf65ba
VSO
3416 bs->open_flags = reopen_state->flags;
3417 bs->read_only = !(reopen_state->flags & BDRV_O_RDWR);
543770bd 3418 bs->detect_zeroes = reopen_state->detect_zeroes;
355ef4ac 3419
50196d7a
AG
3420 /* Remove child references from bs->options and bs->explicit_options.
3421 * Child options were already removed in bdrv_reopen_queue_child() */
3422 QLIST_FOREACH(child, &bs->children, next) {
3423 qdict_del(bs->explicit_options, child->name);
3424 qdict_del(bs->options, child->name);
3425 }
3426
50bf65ba 3427 bdrv_refresh_limits(bs, NULL);
cb9ff6c2 3428
30450259
KW
3429 bdrv_set_perm(reopen_state->bs, reopen_state->perm,
3430 reopen_state->shared_perm);
3431
cb9ff6c2
VSO
3432 new_can_write =
3433 !bdrv_is_read_only(bs) && !(bdrv_get_flags(bs) & BDRV_O_INACTIVE);
3434 if (!old_can_write && new_can_write && drv->bdrv_reopen_bitmaps_rw) {
3435 Error *local_err = NULL;
3436 if (drv->bdrv_reopen_bitmaps_rw(bs, &local_err) < 0) {
3437 /* This is not fatal, bitmaps just left read-only, so all following
3438 * writes will fail. User can remove read-only bitmaps to unblock
3439 * writes.
3440 */
3441 error_reportf_err(local_err,
3442 "%s: Failed to make dirty bitmaps writable: ",
3443 bdrv_get_node_name(bs));
3444 }
3445 }
e971aa12
JC
3446}
3447
3448/*
3449 * Abort the reopen, and delete and free the staged changes in
3450 * reopen_state
3451 */
3452void bdrv_reopen_abort(BDRVReopenState *reopen_state)
3453{
3454 BlockDriver *drv;
3455
3456 assert(reopen_state != NULL);
3457 drv = reopen_state->bs->drv;
3458 assert(drv != NULL);
3459
3460 if (drv->bdrv_reopen_abort) {
3461 drv->bdrv_reopen_abort(reopen_state);
3462 }
145f598e 3463
30450259 3464 bdrv_abort_perm_update(reopen_state->bs);
e971aa12
JC
3465}
3466
3467
64dff520 3468static void bdrv_close(BlockDriverState *bs)
fc01f7e7 3469{
33384421 3470 BdrvAioNotifier *ban, *ban_next;
50a3efb0 3471 BdrvChild *child, *next;
33384421 3472
ca9bd24c 3473 assert(!bs->job);
30f55fb8 3474 assert(!bs->refcnt);
99b7e775 3475
fc27291d 3476 bdrv_drained_begin(bs); /* complete I/O */
58fda173 3477 bdrv_flush(bs);
53ec73e2 3478 bdrv_drain(bs); /* in case flush left pending I/O */
fc27291d 3479
3cbc002c 3480 if (bs->drv) {
3c005293
VSO
3481 if (bs->drv->bdrv_close) {
3482 bs->drv->bdrv_close(bs);
3483 }
9a4f4c31 3484 bs->drv = NULL;
50a3efb0 3485 }
9a7dedbc 3486
50a3efb0 3487 bdrv_set_backing_hd(bs, NULL, &error_abort);
9a7dedbc 3488
50a3efb0
AG
3489 if (bs->file != NULL) {
3490 bdrv_unref_child(bs, bs->file);
3491 bs->file = NULL;
3492 }
9a4f4c31 3493
50a3efb0
AG
3494 QLIST_FOREACH_SAFE(child, &bs->children, next, next) {
3495 /* TODO Remove bdrv_unref() from drivers' close function and use
3496 * bdrv_unref_child() here */
3497 if (child->bs->inherits_from == bs) {
3498 child->bs->inherits_from = NULL;
6e93e7c4 3499 }
50a3efb0 3500 bdrv_detach_child(child);
b338082b 3501 }
98f90dba 3502
50a3efb0
AG
3503 g_free(bs->opaque);
3504 bs->opaque = NULL;
3505 atomic_set(&bs->copy_on_read, 0);
3506 bs->backing_file[0] = '\0';
3507 bs->backing_format[0] = '\0';
3508 bs->total_sectors = 0;
3509 bs->encrypted = false;
3510 bs->sg = false;
cb3e7f08
MAL
3511 qobject_unref(bs->options);
3512 qobject_unref(bs->explicit_options);
50a3efb0
AG
3513 bs->options = NULL;
3514 bs->explicit_options = NULL;
cb3e7f08 3515 qobject_unref(bs->full_open_options);
50a3efb0
AG
3516 bs->full_open_options = NULL;
3517
cca43ae1
VSO
3518 bdrv_release_named_dirty_bitmaps(bs);
3519 assert(QLIST_EMPTY(&bs->dirty_bitmaps));
3520
33384421
HR
3521 QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_next) {
3522 g_free(ban);
3523 }
3524 QLIST_INIT(&bs->aio_notifiers);
fc27291d 3525 bdrv_drained_end(bs);
b338082b
FB
3526}
3527
2bc93fed
MK
3528void bdrv_close_all(void)
3529{
b3b5299d 3530 assert(job_next(NULL) == NULL);
cd7fca95 3531 nbd_export_close_all();
ca9bd24c
HR
3532
3533 /* Drop references from requests still in flight, such as canceled block
3534 * jobs whose AIO context has not been polled yet */
3535 bdrv_drain_all();
2bc93fed 3536
ca9bd24c
HR
3537 blk_remove_all_bs();
3538 blockdev_close_all_bdrv_states();
ed78cda3 3539
a1a2af07 3540 assert(QTAILQ_EMPTY(&all_bdrv_states));
2bc93fed
MK
3541}
3542
d0ac0380
KW
3543static bool should_update_child(BdrvChild *c, BlockDriverState *to)
3544{
3545 BdrvChild *to_c;
3546
3547 if (c->role->stay_at_node) {
3548 return false;
3549 }
3550
ec9f10fe
HR
3551 /* If the child @c belongs to the BDS @to, replacing the current
3552 * c->bs by @to would mean to create a loop.
3553 *
3554 * Such a case occurs when appending a BDS to a backing chain.
3555 * For instance, imagine the following chain:
3556 *
3557 * guest device -> node A -> further backing chain...
3558 *
3559 * Now we create a new BDS B which we want to put on top of this
3560 * chain, so we first attach A as its backing node:
3561 *
3562 * node B
3563 * |
3564 * v
3565 * guest device -> node A -> further backing chain...
3566 *
3567 * Finally we want to replace A by B. When doing that, we want to
3568 * replace all pointers to A by pointers to B -- except for the
3569 * pointer from B because (1) that would create a loop, and (2)
3570 * that pointer should simply stay intact:
3571 *
3572 * guest device -> node B
3573 * |
3574 * v
3575 * node A -> further backing chain...
3576 *
3577 * In general, when replacing a node A (c->bs) by a node B (@to),
3578 * if A is a child of B, that means we cannot replace A by B there
3579 * because that would create a loop. Silently detaching A from B
3580 * is also not really an option. So overall just leaving A in
3581 * place there is the most sensible choice. */
3582 QLIST_FOREACH(to_c, &to->children, next) {
3583 if (to_c == c) {
d0ac0380
KW
3584 return false;
3585 }
3586 }
3587
3588 return true;
3589}
3590
5fe31c25
KW
3591void bdrv_replace_node(BlockDriverState *from, BlockDriverState *to,
3592 Error **errp)
dd62f1ca 3593{
d0ac0380 3594 BdrvChild *c, *next;
234ac1a9
KW
3595 GSList *list = NULL, *p;
3596 uint64_t old_perm, old_shared;
3597 uint64_t perm = 0, shared = BLK_PERM_ALL;
3598 int ret;
3599
5fe31c25
KW
3600 assert(!atomic_read(&from->in_flight));
3601 assert(!atomic_read(&to->in_flight));
dd62f1ca 3602
234ac1a9
KW
3603 /* Make sure that @from doesn't go away until we have successfully attached
3604 * all of its parents to @to. */
3605 bdrv_ref(from);
dd62f1ca 3606
234ac1a9 3607 /* Put all parents into @list and calculate their cumulative permissions */
dd62f1ca 3608 QLIST_FOREACH_SAFE(c, &from->parents, next_parent, next) {
ec9f10fe 3609 assert(c->bs == from);
d0ac0380 3610 if (!should_update_child(c, to)) {
26de9438
KW
3611 continue;
3612 }
234ac1a9
KW
3613 list = g_slist_prepend(list, c);
3614 perm |= c->perm;
3615 shared &= c->shared_perm;
3616 }
3617
3618 /* Check whether the required permissions can be granted on @to, ignoring
3619 * all BdrvChild in @list so that they can't block themselves. */
3121fb45 3620 ret = bdrv_check_update_perm(to, NULL, perm, shared, list, errp);
234ac1a9
KW
3621 if (ret < 0) {
3622 bdrv_abort_perm_update(to);
3623 goto out;
3624 }
3625
3626 /* Now actually perform the change. We performed the permission check for
3627 * all elements of @list at once, so set the permissions all at once at the
3628 * very end. */
3629 for (p = list; p != NULL; p = p->next) {
3630 c = p->data;
9bd910e2 3631
dd62f1ca 3632 bdrv_ref(to);
234ac1a9 3633 bdrv_replace_child_noperm(c, to);
dd62f1ca
KW
3634 bdrv_unref(from);
3635 }
234ac1a9
KW
3636
3637 bdrv_get_cumulative_perm(to, &old_perm, &old_shared);
3638 bdrv_set_perm(to, old_perm | perm, old_shared | shared);
3639
3640out:
3641 g_slist_free(list);
3642 bdrv_unref(from);
dd62f1ca
KW
3643}
3644
4ddc07ca
PB
3645/*
3646 * Add new bs contents at the top of an image chain while the chain is
3647 * live, while keeping required fields on the top layer.
3648 *
3649 * This will modify the BlockDriverState fields, and swap contents
3650 * between bs_new and bs_top. Both bs_new and bs_top are modified.
3651 *
bfb197e0 3652 * bs_new must not be attached to a BlockBackend.
4ddc07ca
PB
3653 *
3654 * This function does not create any image files.
dd62f1ca
KW
3655 *
3656 * bdrv_append() takes ownership of a bs_new reference and unrefs it because
3657 * that's what the callers commonly need. bs_new will be referenced by the old
3658 * parents of bs_top after bdrv_append() returns. If the caller needs to keep a
3659 * reference of its own, it must call bdrv_ref().
4ddc07ca 3660 */
b2c2832c
KW
3661void bdrv_append(BlockDriverState *bs_new, BlockDriverState *bs_top,
3662 Error **errp)
4ddc07ca 3663{
b2c2832c
KW
3664 Error *local_err = NULL;
3665
b2c2832c
KW
3666 bdrv_set_backing_hd(bs_new, bs_top, &local_err);
3667 if (local_err) {
3668 error_propagate(errp, local_err);
3669 goto out;
3670 }
dd62f1ca 3671
5fe31c25 3672 bdrv_replace_node(bs_top, bs_new, &local_err);
234ac1a9
KW
3673 if (local_err) {
3674 error_propagate(errp, local_err);
3675 bdrv_set_backing_hd(bs_new, NULL, &error_abort);
3676 goto out;
3677 }
4ddc07ca 3678
dd62f1ca
KW
3679 /* bs_new is now referenced by its new parents, we don't need the
3680 * additional reference any more. */
b2c2832c 3681out:
dd62f1ca 3682 bdrv_unref(bs_new);
8802d1fd
JC
3683}
3684
4f6fd349 3685static void bdrv_delete(BlockDriverState *bs)
b338082b 3686{
3e914655 3687 assert(!bs->job);
3718d8ab 3688 assert(bdrv_op_blocker_is_empty(bs));
4f6fd349 3689 assert(!bs->refcnt);
18846dee 3690
e1b5c52e
SH
3691 bdrv_close(bs);
3692
1b7bdbc1 3693 /* remove from list, if necessary */
63eaaae0
KW
3694 if (bs->node_name[0] != '\0') {
3695 QTAILQ_REMOVE(&graph_bdrv_states, bs, node_list);
3696 }
2c1d04e0
HR
3697 QTAILQ_REMOVE(&all_bdrv_states, bs, bs_list);
3698
7267c094 3699 g_free(bs);
fc01f7e7
FB
3700}
3701
e97fc193
AL
3702/*
3703 * Run consistency checks on an image
3704 *
e076f338 3705 * Returns 0 if the check could be completed (it doesn't mean that the image is
a1c7273b 3706 * free of errors) or -errno when an internal error occurred. The results of the
e076f338 3707 * check are stored in res.
e97fc193 3708 */
2fd61638
PB
3709static int coroutine_fn bdrv_co_check(BlockDriverState *bs,
3710 BdrvCheckResult *res, BdrvCheckMode fix)
e97fc193 3711{
908bcd54
HR
3712 if (bs->drv == NULL) {
3713 return -ENOMEDIUM;
3714 }
2fd61638 3715 if (bs->drv->bdrv_co_check == NULL) {
e97fc193
AL
3716 return -ENOTSUP;
3717 }
3718
e076f338 3719 memset(res, 0, sizeof(*res));
2fd61638
PB
3720 return bs->drv->bdrv_co_check(bs, res, fix);
3721}
3722
3723typedef struct CheckCo {
3724 BlockDriverState *bs;
3725 BdrvCheckResult *res;
3726 BdrvCheckMode fix;
3727 int ret;
3728} CheckCo;
3729
3730static void bdrv_check_co_entry(void *opaque)
3731{
3732 CheckCo *cco = opaque;
3733 cco->ret = bdrv_co_check(cco->bs, cco->res, cco->fix);
4720cbee 3734 aio_wait_kick();
2fd61638
PB
3735}
3736
3737int bdrv_check(BlockDriverState *bs,
3738 BdrvCheckResult *res, BdrvCheckMode fix)
3739{
3740 Coroutine *co;
3741 CheckCo cco = {
3742 .bs = bs,
3743 .res = res,
3744 .ret = -EINPROGRESS,
3745 .fix = fix,
3746 };
3747
3748 if (qemu_in_coroutine()) {
3749 /* Fast-path if already in coroutine context */
3750 bdrv_check_co_entry(&cco);
3751 } else {
3752 co = qemu_coroutine_create(bdrv_check_co_entry, &cco);
4720cbee 3753 bdrv_coroutine_enter(bs, co);
2fd61638
PB
3754 BDRV_POLL_WHILE(bs, cco.ret == -EINPROGRESS);
3755 }
3756
3757 return cco.ret;
e97fc193
AL
3758}
3759
756e6736
KW
3760/*
3761 * Return values:
3762 * 0 - success
3763 * -EINVAL - backing format specified, but no file
3764 * -ENOSPC - can't update the backing file because no space is left in the
3765 * image file header
3766 * -ENOTSUP - format driver doesn't support changing the backing file
3767 */
3768int bdrv_change_backing_file(BlockDriverState *bs,
3769 const char *backing_file, const char *backing_fmt)
3770{
3771 BlockDriver *drv = bs->drv;
469ef350 3772 int ret;
756e6736 3773
d470ad42
HR
3774 if (!drv) {
3775 return -ENOMEDIUM;
3776 }
3777
5f377794
PB
3778 /* Backing file format doesn't make sense without a backing file */
3779 if (backing_fmt && !backing_file) {
3780 return -EINVAL;
3781 }
3782
756e6736 3783 if (drv->bdrv_change_backing_file != NULL) {
469ef350 3784 ret = drv->bdrv_change_backing_file(bs, backing_file, backing_fmt);
756e6736 3785 } else {
469ef350 3786 ret = -ENOTSUP;
756e6736 3787 }
469ef350
PB
3788
3789 if (ret == 0) {
3790 pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_file ?: "");
3791 pstrcpy(bs->backing_format, sizeof(bs->backing_format), backing_fmt ?: "");
3792 }
3793 return ret;
756e6736
KW
3794}
3795
6ebdcee2
JC
3796/*
3797 * Finds the image layer in the chain that has 'bs' as its backing file.
3798 *
3799 * active is the current topmost image.
3800 *
3801 * Returns NULL if bs is not found in active's image chain,
3802 * or if active == bs.
4caf0fcd
JC
3803 *
3804 * Returns the bottommost base image if bs == NULL.
6ebdcee2
JC
3805 */
3806BlockDriverState *bdrv_find_overlay(BlockDriverState *active,
3807 BlockDriverState *bs)
3808{
760e0063
KW
3809 while (active && bs != backing_bs(active)) {
3810 active = backing_bs(active);
6ebdcee2
JC
3811 }
3812
4caf0fcd
JC
3813 return active;
3814}
6ebdcee2 3815
4caf0fcd
JC
3816/* Given a BDS, searches for the base layer. */
3817BlockDriverState *bdrv_find_base(BlockDriverState *bs)
3818{
3819 return bdrv_find_overlay(bs, NULL);
6ebdcee2
JC
3820}
3821
6ebdcee2
JC
3822/*
3823 * Drops images above 'base' up to and including 'top', and sets the image
3824 * above 'top' to have base as its backing file.
3825 *
3826 * Requires that the overlay to 'top' is opened r/w, so that the backing file
3827 * information in 'bs' can be properly updated.
3828 *
3829 * E.g., this will convert the following chain:
3830 * bottom <- base <- intermediate <- top <- active
3831 *
3832 * to
3833 *
3834 * bottom <- base <- active
3835 *
3836 * It is allowed for bottom==base, in which case it converts:
3837 *
3838 * base <- intermediate <- top <- active
3839 *
3840 * to
3841 *
3842 * base <- active
3843 *
54e26900
JC
3844 * If backing_file_str is non-NULL, it will be used when modifying top's
3845 * overlay image metadata.
3846 *
6ebdcee2
JC
3847 * Error conditions:
3848 * if active == top, that is considered an error
3849 *
3850 */
bde70715
KW
3851int bdrv_drop_intermediate(BlockDriverState *top, BlockDriverState *base,
3852 const char *backing_file_str)
6ebdcee2 3853{
6bd858b3
AG
3854 BlockDriverState *explicit_top = top;
3855 bool update_inherits_from;
61f09cea 3856 BdrvChild *c, *next;
12fa4af6 3857 Error *local_err = NULL;
6ebdcee2
JC
3858 int ret = -EIO;
3859
6858eba0
KW
3860 bdrv_ref(top);
3861
6ebdcee2
JC
3862 if (!top->drv || !base->drv) {
3863 goto exit;
3864 }
3865
5db15a57
KW
3866 /* Make sure that base is in the backing chain of top */
3867 if (!bdrv_chain_contains(top, base)) {
6ebdcee2
JC
3868 goto exit;
3869 }
3870
6bd858b3
AG
3871 /* If 'base' recursively inherits from 'top' then we should set
3872 * base->inherits_from to top->inherits_from after 'top' and all
3873 * other intermediate nodes have been dropped.
3874 * If 'top' is an implicit node (e.g. "commit_top") we should skip
3875 * it because no one inherits from it. We use explicit_top for that. */
3876 while (explicit_top && explicit_top->implicit) {
3877 explicit_top = backing_bs(explicit_top);
3878 }
3879 update_inherits_from = bdrv_inherits_from_recursive(base, explicit_top);
3880
6ebdcee2 3881 /* success - we can delete the intermediate states, and link top->base */
bde70715
KW
3882 /* TODO Check graph modification op blockers (BLK_PERM_GRAPH_MOD) once
3883 * we've figured out how they should work. */
61f09cea
KW
3884 backing_file_str = backing_file_str ? backing_file_str : base->filename;
3885
3886 QLIST_FOREACH_SAFE(c, &top->parents, next_parent, next) {
3887 /* Check whether we are allowed to switch c from top to base */
3888 GSList *ignore_children = g_slist_prepend(NULL, c);
3889 bdrv_check_update_perm(base, NULL, c->perm, c->shared_perm,
3890 ignore_children, &local_err);
2c860e79 3891 g_slist_free(ignore_children);
61f09cea
KW
3892 if (local_err) {
3893 ret = -EPERM;
3894 error_report_err(local_err);
6858eba0
KW
3895 goto exit;
3896 }
12fa4af6 3897
61f09cea
KW
3898 /* If so, update the backing file path in the image file */
3899 if (c->role->update_filename) {
3900 ret = c->role->update_filename(c, base, backing_file_str,
3901 &local_err);
3902 if (ret < 0) {
3903 bdrv_abort_perm_update(base);
3904 error_report_err(local_err);
3905 goto exit;
3906 }
3907 }
3908
3909 /* Do the actual switch in the in-memory graph.
3910 * Completes bdrv_check_update_perm() transaction internally. */
3911 bdrv_ref(base);
3912 bdrv_replace_child(c, base);
3913 bdrv_unref(top);
12fa4af6 3914 }
6ebdcee2 3915
6bd858b3
AG
3916 if (update_inherits_from) {
3917 base->inherits_from = explicit_top->inherits_from;
3918 }
3919
6ebdcee2 3920 ret = 0;
6ebdcee2 3921exit:
6858eba0 3922 bdrv_unref(top);
6ebdcee2
JC
3923 return ret;
3924}
3925
61007b31
SH
3926/**
3927 * Length of a allocated file in bytes. Sparse files are counted by actual
3928 * allocated space. Return < 0 if error or unknown.
3929 */
3930int64_t bdrv_get_allocated_file_size(BlockDriverState *bs)
71d0770c 3931{
61007b31
SH
3932 BlockDriver *drv = bs->drv;
3933 if (!drv) {
3934 return -ENOMEDIUM;
8f4754ed 3935 }
61007b31
SH
3936 if (drv->bdrv_get_allocated_file_size) {
3937 return drv->bdrv_get_allocated_file_size(bs);
3938 }
3939 if (bs->file) {
9a4f4c31 3940 return bdrv_get_allocated_file_size(bs->file->bs);
1c9805a3 3941 }
61007b31 3942 return -ENOTSUP;
1c9805a3 3943}
e7a8a783 3944
90880ff1
SH
3945/*
3946 * bdrv_measure:
3947 * @drv: Format driver
3948 * @opts: Creation options for new image
3949 * @in_bs: Existing image containing data for new image (may be NULL)
3950 * @errp: Error object
3951 * Returns: A #BlockMeasureInfo (free using qapi_free_BlockMeasureInfo())
3952 * or NULL on error
3953 *
3954 * Calculate file size required to create a new image.
3955 *
3956 * If @in_bs is given then space for allocated clusters and zero clusters
3957 * from that image are included in the calculation. If @opts contains a
3958 * backing file that is shared by @in_bs then backing clusters may be omitted
3959 * from the calculation.
3960 *
3961 * If @in_bs is NULL then the calculation includes no allocated clusters
3962 * unless a preallocation option is given in @opts.
3963 *
3964 * Note that @in_bs may use a different BlockDriver from @drv.
3965 *
3966 * If an error occurs the @errp pointer is set.
3967 */
3968BlockMeasureInfo *bdrv_measure(BlockDriver *drv, QemuOpts *opts,
3969 BlockDriverState *in_bs, Error **errp)
3970{
3971 if (!drv->bdrv_measure) {
3972 error_setg(errp, "Block driver '%s' does not support size measurement",
3973 drv->format_name);
3974 return NULL;
3975 }
3976
3977 return drv->bdrv_measure(opts, in_bs, errp);
3978}
3979
61007b31
SH
3980/**
3981 * Return number of sectors on success, -errno on error.
1c9805a3 3982 */
61007b31 3983int64_t bdrv_nb_sectors(BlockDriverState *bs)
1c9805a3 3984{
61007b31 3985 BlockDriver *drv = bs->drv;
498e386c 3986
61007b31
SH
3987 if (!drv)
3988 return -ENOMEDIUM;
2572b37a 3989
61007b31
SH
3990 if (drv->has_variable_length) {
3991 int ret = refresh_total_sectors(bs, bs->total_sectors);
3992 if (ret < 0) {
3993 return ret;
1c9805a3
SH
3994 }
3995 }
61007b31 3996 return bs->total_sectors;
1c9805a3 3997}
b338082b 3998
61007b31
SH
3999/**
4000 * Return length in bytes on success, -errno on error.
4001 * The length is always a multiple of BDRV_SECTOR_SIZE.
8d3b1a2d 4002 */
61007b31 4003int64_t bdrv_getlength(BlockDriverState *bs)
8d3b1a2d 4004{
61007b31 4005 int64_t ret = bdrv_nb_sectors(bs);
8d3b1a2d 4006
4a9c9ea0 4007 ret = ret > INT64_MAX / BDRV_SECTOR_SIZE ? -EFBIG : ret;
61007b31 4008 return ret < 0 ? ret : ret * BDRV_SECTOR_SIZE;
fc01f7e7
FB
4009}
4010
61007b31
SH
4011/* return 0 as number of sectors if no device present or error */
4012void bdrv_get_geometry(BlockDriverState *bs, uint64_t *nb_sectors_ptr)
07d27a44 4013{
61007b31 4014 int64_t nb_sectors = bdrv_nb_sectors(bs);
07d27a44 4015
61007b31 4016 *nb_sectors_ptr = nb_sectors < 0 ? 0 : nb_sectors;
07d27a44
MA
4017}
4018
54115412 4019bool bdrv_is_sg(BlockDriverState *bs)
f08145fe 4020{
61007b31 4021 return bs->sg;
f08145fe
KW
4022}
4023
54115412 4024bool bdrv_is_encrypted(BlockDriverState *bs)
fc3959e4 4025{
760e0063 4026 if (bs->backing && bs->backing->bs->encrypted) {
54115412 4027 return true;
760e0063 4028 }
61007b31 4029 return bs->encrypted;
fc3959e4
FZ
4030}
4031
61007b31 4032const char *bdrv_get_format_name(BlockDriverState *bs)
40b4f539 4033{
61007b31 4034 return bs->drv ? bs->drv->format_name : NULL;
40b4f539
KW
4035}
4036
61007b31 4037static int qsort_strcmp(const void *a, const void *b)
40b4f539 4038{
ceff5bd7 4039 return strcmp(*(char *const *)a, *(char *const *)b);
40b4f539
KW
4040}
4041
61007b31
SH
4042void bdrv_iterate_format(void (*it)(void *opaque, const char *name),
4043 void *opaque)
40b4f539 4044{
61007b31
SH
4045 BlockDriver *drv;
4046 int count = 0;
4047 int i;
4048 const char **formats = NULL;
40b4f539 4049
61007b31
SH
4050 QLIST_FOREACH(drv, &bdrv_drivers, list) {
4051 if (drv->format_name) {
4052 bool found = false;
4053 int i = count;
4054 while (formats && i && !found) {
4055 found = !strcmp(formats[--i], drv->format_name);
4056 }
e2a305fb 4057
61007b31
SH
4058 if (!found) {
4059 formats = g_renew(const char *, formats, count + 1);
4060 formats[count++] = drv->format_name;
4061 }
6c5a42ac 4062 }
61007b31 4063 }
6c5a42ac 4064
eb0df69f
HR
4065 for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); i++) {
4066 const char *format_name = block_driver_modules[i].format_name;
4067
4068 if (format_name) {
4069 bool found = false;
4070 int j = count;
4071
4072 while (formats && j && !found) {
4073 found = !strcmp(formats[--j], format_name);
4074 }
4075
4076 if (!found) {
4077 formats = g_renew(const char *, formats, count + 1);
4078 formats[count++] = format_name;
4079 }
4080 }
4081 }
4082
61007b31 4083 qsort(formats, count, sizeof(formats[0]), qsort_strcmp);
40b4f539 4084
61007b31
SH
4085 for (i = 0; i < count; i++) {
4086 it(opaque, formats[i]);
4087 }
40b4f539 4088
61007b31
SH
4089 g_free(formats);
4090}
40b4f539 4091
61007b31
SH
4092/* This function is to find a node in the bs graph */
4093BlockDriverState *bdrv_find_node(const char *node_name)
4094{
4095 BlockDriverState *bs;
391827eb 4096
61007b31 4097 assert(node_name);
40b4f539 4098
61007b31
SH
4099 QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
4100 if (!strcmp(node_name, bs->node_name)) {
4101 return bs;
40b4f539
KW
4102 }
4103 }
61007b31 4104 return NULL;
40b4f539
KW
4105}
4106
61007b31
SH
4107/* Put this QMP function here so it can access the static graph_bdrv_states. */
4108BlockDeviceInfoList *bdrv_named_nodes_list(Error **errp)
40b4f539 4109{
61007b31
SH
4110 BlockDeviceInfoList *list, *entry;
4111 BlockDriverState *bs;
40b4f539 4112
61007b31
SH
4113 list = NULL;
4114 QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
c83f9fba 4115 BlockDeviceInfo *info = bdrv_block_device_info(NULL, bs, errp);
61007b31
SH
4116 if (!info) {
4117 qapi_free_BlockDeviceInfoList(list);
4118 return NULL;
301db7c2 4119 }
61007b31
SH
4120 entry = g_malloc0(sizeof(*entry));
4121 entry->value = info;
4122 entry->next = list;
4123 list = entry;
301db7c2
RH
4124 }
4125
61007b31
SH
4126 return list;
4127}
40b4f539 4128
5d3b4e99
VSO
4129#define QAPI_LIST_ADD(list, element) do { \
4130 typeof(list) _tmp = g_new(typeof(*(list)), 1); \
4131 _tmp->value = (element); \
4132 _tmp->next = (list); \
4133 (list) = _tmp; \
4134} while (0)
4135
4136typedef struct XDbgBlockGraphConstructor {
4137 XDbgBlockGraph *graph;
4138 GHashTable *graph_nodes;
4139} XDbgBlockGraphConstructor;
4140
4141static XDbgBlockGraphConstructor *xdbg_graph_new(void)
4142{
4143 XDbgBlockGraphConstructor *gr = g_new(XDbgBlockGraphConstructor, 1);
4144
4145 gr->graph = g_new0(XDbgBlockGraph, 1);
4146 gr->graph_nodes = g_hash_table_new(NULL, NULL);
4147
4148 return gr;
4149}
4150
4151static XDbgBlockGraph *xdbg_graph_finalize(XDbgBlockGraphConstructor *gr)
4152{
4153 XDbgBlockGraph *graph = gr->graph;
4154
4155 g_hash_table_destroy(gr->graph_nodes);
4156 g_free(gr);
4157
4158 return graph;
4159}
4160
4161static uintptr_t xdbg_graph_node_num(XDbgBlockGraphConstructor *gr, void *node)
4162{
4163 uintptr_t ret = (uintptr_t)g_hash_table_lookup(gr->graph_nodes, node);
4164
4165 if (ret != 0) {
4166 return ret;
4167 }
4168
4169 /*
4170 * Start counting from 1, not 0, because 0 interferes with not-found (NULL)
4171 * answer of g_hash_table_lookup.
4172 */
4173 ret = g_hash_table_size(gr->graph_nodes) + 1;
4174 g_hash_table_insert(gr->graph_nodes, node, (void *)ret);
4175
4176 return ret;
4177}
4178
4179static void xdbg_graph_add_node(XDbgBlockGraphConstructor *gr, void *node,
4180 XDbgBlockGraphNodeType type, const char *name)
4181{
4182 XDbgBlockGraphNode *n;
4183
4184 n = g_new0(XDbgBlockGraphNode, 1);
4185
4186 n->id = xdbg_graph_node_num(gr, node);
4187 n->type = type;
4188 n->name = g_strdup(name);
4189
4190 QAPI_LIST_ADD(gr->graph->nodes, n);
4191}
4192
4193static void xdbg_graph_add_edge(XDbgBlockGraphConstructor *gr, void *parent,
4194 const BdrvChild *child)
4195{
4196 typedef struct {
4197 unsigned int flag;
4198 BlockPermission num;
4199 } PermissionMap;
4200
4201 static const PermissionMap permissions[] = {
4202 { BLK_PERM_CONSISTENT_READ, BLOCK_PERMISSION_CONSISTENT_READ },
4203 { BLK_PERM_WRITE, BLOCK_PERMISSION_WRITE },
4204 { BLK_PERM_WRITE_UNCHANGED, BLOCK_PERMISSION_WRITE_UNCHANGED },
4205 { BLK_PERM_RESIZE, BLOCK_PERMISSION_RESIZE },
4206 { BLK_PERM_GRAPH_MOD, BLOCK_PERMISSION_GRAPH_MOD },
4207 { 0, 0 }
4208 };
4209 const PermissionMap *p;
4210 XDbgBlockGraphEdge *edge;
4211
4212 QEMU_BUILD_BUG_ON(1UL << (ARRAY_SIZE(permissions) - 1) != BLK_PERM_ALL + 1);
4213
4214 edge = g_new0(XDbgBlockGraphEdge, 1);
4215
4216 edge->parent = xdbg_graph_node_num(gr, parent);
4217 edge->child = xdbg_graph_node_num(gr, child->bs);
4218 edge->name = g_strdup(child->name);
4219
4220 for (p = permissions; p->flag; p++) {
4221 if (p->flag & child->perm) {
4222 QAPI_LIST_ADD(edge->perm, p->num);
4223 }
4224 if (p->flag & child->shared_perm) {
4225 QAPI_LIST_ADD(edge->shared_perm, p->num);
4226 }
4227 }
4228
4229 QAPI_LIST_ADD(gr->graph->edges, edge);
4230}
4231
4232
4233XDbgBlockGraph *bdrv_get_xdbg_block_graph(Error **errp)
4234{
4235 BlockBackend *blk;
4236 BlockJob *job;
4237 BlockDriverState *bs;
4238 BdrvChild *child;
4239 XDbgBlockGraphConstructor *gr = xdbg_graph_new();
4240
4241 for (blk = blk_all_next(NULL); blk; blk = blk_all_next(blk)) {
4242 char *allocated_name = NULL;
4243 const char *name = blk_name(blk);
4244
4245 if (!*name) {
4246 name = allocated_name = blk_get_attached_dev_id(blk);
4247 }
4248 xdbg_graph_add_node(gr, blk, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_BACKEND,
4249 name);
4250 g_free(allocated_name);
4251 if (blk_root(blk)) {
4252 xdbg_graph_add_edge(gr, blk, blk_root(blk));
4253 }
4254 }
4255
4256 for (job = block_job_next(NULL); job; job = block_job_next(job)) {
4257 GSList *el;
4258
4259 xdbg_graph_add_node(gr, job, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_JOB,
4260 job->job.id);
4261 for (el = job->nodes; el; el = el->next) {
4262 xdbg_graph_add_edge(gr, job, (BdrvChild *)el->data);
4263 }
4264 }
4265
4266 QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
4267 xdbg_graph_add_node(gr, bs, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_DRIVER,
4268 bs->node_name);
4269 QLIST_FOREACH(child, &bs->children, next) {
4270 xdbg_graph_add_edge(gr, bs, child);
4271 }
4272 }
4273
4274 return xdbg_graph_finalize(gr);
4275}
4276
61007b31
SH
4277BlockDriverState *bdrv_lookup_bs(const char *device,
4278 const char *node_name,
4279 Error **errp)
4280{
4281 BlockBackend *blk;
4282 BlockDriverState *bs;
40b4f539 4283
61007b31
SH
4284 if (device) {
4285 blk = blk_by_name(device);
40b4f539 4286
61007b31 4287 if (blk) {
9f4ed6fb
AG
4288 bs = blk_bs(blk);
4289 if (!bs) {
5433c24f 4290 error_setg(errp, "Device '%s' has no medium", device);
5433c24f
HR
4291 }
4292
9f4ed6fb 4293 return bs;
61007b31
SH
4294 }
4295 }
40b4f539 4296
61007b31
SH
4297 if (node_name) {
4298 bs = bdrv_find_node(node_name);
6d519a5f 4299
61007b31
SH
4300 if (bs) {
4301 return bs;
4302 }
40b4f539
KW
4303 }
4304
61007b31
SH
4305 error_setg(errp, "Cannot find device=%s nor node_name=%s",
4306 device ? device : "",
4307 node_name ? node_name : "");
4308 return NULL;
40b4f539
KW
4309}
4310
61007b31
SH
4311/* If 'base' is in the same chain as 'top', return true. Otherwise,
4312 * return false. If either argument is NULL, return false. */
4313bool bdrv_chain_contains(BlockDriverState *top, BlockDriverState *base)
83f64091 4314{
61007b31 4315 while (top && top != base) {
760e0063 4316 top = backing_bs(top);
02c50efe 4317 }
61007b31
SH
4318
4319 return top != NULL;
02c50efe
FZ
4320}
4321
61007b31 4322BlockDriverState *bdrv_next_node(BlockDriverState *bs)
02c50efe 4323{
61007b31
SH
4324 if (!bs) {
4325 return QTAILQ_FIRST(&graph_bdrv_states);
02c50efe 4326 }
61007b31 4327 return QTAILQ_NEXT(bs, node_list);
83f64091
FB
4328}
4329
0f12264e
KW
4330BlockDriverState *bdrv_next_all_states(BlockDriverState *bs)
4331{
4332 if (!bs) {
4333 return QTAILQ_FIRST(&all_bdrv_states);
4334 }
4335 return QTAILQ_NEXT(bs, bs_list);
4336}
4337
61007b31 4338const char *bdrv_get_node_name(const BlockDriverState *bs)
83f64091 4339{
61007b31 4340 return bs->node_name;
beac80cd
FB
4341}
4342
1f0c461b 4343const char *bdrv_get_parent_name(const BlockDriverState *bs)
4c265bf9
KW
4344{
4345 BdrvChild *c;
4346 const char *name;
4347
4348 /* If multiple parents have a name, just pick the first one. */
4349 QLIST_FOREACH(c, &bs->parents, next_parent) {
4350 if (c->role->get_name) {
4351 name = c->role->get_name(c);
4352 if (name && *name) {
4353 return name;
4354 }
4355 }
4356 }
4357
4358 return NULL;
4359}
4360
61007b31
SH
4361/* TODO check what callers really want: bs->node_name or blk_name() */
4362const char *bdrv_get_device_name(const BlockDriverState *bs)
beac80cd 4363{
4c265bf9 4364 return bdrv_get_parent_name(bs) ?: "";
f141eafe 4365}
83f64091 4366
61007b31
SH
4367/* This can be used to identify nodes that might not have a device
4368 * name associated. Since node and device names live in the same
4369 * namespace, the result is unambiguous. The exception is if both are
4370 * absent, then this returns an empty (non-null) string. */
4371const char *bdrv_get_device_or_node_name(const BlockDriverState *bs)
f141eafe 4372{
4c265bf9 4373 return bdrv_get_parent_name(bs) ?: bs->node_name;
beac80cd 4374}
beac80cd 4375
61007b31 4376int bdrv_get_flags(BlockDriverState *bs)
0b5a2445 4377{
61007b31 4378 return bs->open_flags;
0b5a2445
PB
4379}
4380
61007b31 4381int bdrv_has_zero_init_1(BlockDriverState *bs)
68485420 4382{
61007b31 4383 return 1;
0b5a2445
PB
4384}
4385
61007b31 4386int bdrv_has_zero_init(BlockDriverState *bs)
0b5a2445 4387{
d470ad42
HR
4388 if (!bs->drv) {
4389 return 0;
4390 }
0b5a2445 4391
61007b31
SH
4392 /* If BS is a copy on write image, it is initialized to
4393 the contents of the base image, which may not be zeroes. */
760e0063 4394 if (bs->backing) {
61007b31
SH
4395 return 0;
4396 }
4397 if (bs->drv->bdrv_has_zero_init) {
4398 return bs->drv->bdrv_has_zero_init(bs);
0b5a2445 4399 }
5a612c00
MP
4400 if (bs->file && bs->drv->is_filter) {
4401 return bdrv_has_zero_init(bs->file->bs);
4402 }
61007b31
SH
4403
4404 /* safe default */
4405 return 0;
68485420
KW
4406}
4407
61007b31 4408bool bdrv_unallocated_blocks_are_zero(BlockDriverState *bs)
b2a61371 4409{
61007b31 4410 BlockDriverInfo bdi;
b2a61371 4411
760e0063 4412 if (bs->backing) {
61007b31
SH
4413 return false;
4414 }
4415
4416 if (bdrv_get_info(bs, &bdi) == 0) {
4417 return bdi.unallocated_blocks_are_zero;
b2a61371
SH
4418 }
4419
61007b31 4420 return false;
b2a61371
SH
4421}
4422
61007b31 4423bool bdrv_can_write_zeroes_with_unmap(BlockDriverState *bs)
68485420 4424{
2f0342ef 4425 if (!(bs->open_flags & BDRV_O_UNMAP)) {
61007b31
SH
4426 return false;
4427 }
68485420 4428
e24d813b 4429 return bs->supported_zero_flags & BDRV_REQ_MAY_UNMAP;
68485420
KW
4430}
4431
61007b31 4432const char *bdrv_get_encrypted_filename(BlockDriverState *bs)
b2e12bc6 4433{
760e0063 4434 if (bs->backing && bs->backing->bs->encrypted)
61007b31
SH
4435 return bs->backing_file;
4436 else if (bs->encrypted)
4437 return bs->filename;
4438 else
4439 return NULL;
b2e12bc6
CH
4440}
4441
61007b31
SH
4442void bdrv_get_backing_filename(BlockDriverState *bs,
4443 char *filename, int filename_size)
016f5cf6 4444{
61007b31
SH
4445 pstrcpy(filename, filename_size, bs->backing_file);
4446}
d318aea9 4447
61007b31
SH
4448int bdrv_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
4449{
4450 BlockDriver *drv = bs->drv;
5a612c00
MP
4451 /* if bs->drv == NULL, bs is closed, so there's nothing to do here */
4452 if (!drv) {
61007b31 4453 return -ENOMEDIUM;
5a612c00
MP
4454 }
4455 if (!drv->bdrv_get_info) {
4456 if (bs->file && drv->is_filter) {
4457 return bdrv_get_info(bs->file->bs, bdi);
4458 }
61007b31 4459 return -ENOTSUP;
5a612c00 4460 }
61007b31
SH
4461 memset(bdi, 0, sizeof(*bdi));
4462 return drv->bdrv_get_info(bs, bdi);
4463}
016f5cf6 4464
61007b31
SH
4465ImageInfoSpecific *bdrv_get_specific_info(BlockDriverState *bs)
4466{
4467 BlockDriver *drv = bs->drv;
4468 if (drv && drv->bdrv_get_specific_info) {
4469 return drv->bdrv_get_specific_info(bs);
4470 }
4471 return NULL;
016f5cf6
AG
4472}
4473
a31939e6 4474void bdrv_debug_event(BlockDriverState *bs, BlkdebugEvent event)
4265d620 4475{
61007b31
SH
4476 if (!bs || !bs->drv || !bs->drv->bdrv_debug_event) {
4477 return;
4478 }
4265d620 4479
61007b31 4480 bs->drv->bdrv_debug_event(bs, event);
4265d620
PB
4481}
4482
61007b31
SH
4483int bdrv_debug_breakpoint(BlockDriverState *bs, const char *event,
4484 const char *tag)
4265d620 4485{
61007b31 4486 while (bs && bs->drv && !bs->drv->bdrv_debug_breakpoint) {
9a4f4c31 4487 bs = bs->file ? bs->file->bs : NULL;
61007b31 4488 }
4265d620 4489
61007b31
SH
4490 if (bs && bs->drv && bs->drv->bdrv_debug_breakpoint) {
4491 return bs->drv->bdrv_debug_breakpoint(bs, event, tag);
4492 }
4265d620 4493
61007b31 4494 return -ENOTSUP;
4265d620
PB
4495}
4496
61007b31 4497int bdrv_debug_remove_breakpoint(BlockDriverState *bs, const char *tag)
ea2384d3 4498{
61007b31 4499 while (bs && bs->drv && !bs->drv->bdrv_debug_remove_breakpoint) {
9a4f4c31 4500 bs = bs->file ? bs->file->bs : NULL;
61007b31 4501 }
ce1a14dc 4502
61007b31
SH
4503 if (bs && bs->drv && bs->drv->bdrv_debug_remove_breakpoint) {
4504 return bs->drv->bdrv_debug_remove_breakpoint(bs, tag);
4505 }
4506
4507 return -ENOTSUP;
eb852011
MA
4508}
4509
61007b31 4510int bdrv_debug_resume(BlockDriverState *bs, const char *tag)
ce1a14dc 4511{
61007b31 4512 while (bs && (!bs->drv || !bs->drv->bdrv_debug_resume)) {
9a4f4c31 4513 bs = bs->file ? bs->file->bs : NULL;
61007b31 4514 }
ce1a14dc 4515
61007b31
SH
4516 if (bs && bs->drv && bs->drv->bdrv_debug_resume) {
4517 return bs->drv->bdrv_debug_resume(bs, tag);
4518 }
ce1a14dc 4519
61007b31 4520 return -ENOTSUP;
f197fe2b
FZ
4521}
4522
61007b31 4523bool bdrv_debug_is_suspended(BlockDriverState *bs, const char *tag)
ce1a14dc 4524{
61007b31 4525 while (bs && bs->drv && !bs->drv->bdrv_debug_is_suspended) {
9a4f4c31 4526 bs = bs->file ? bs->file->bs : NULL;
f197fe2b 4527 }
19cb3738 4528
61007b31
SH
4529 if (bs && bs->drv && bs->drv->bdrv_debug_is_suspended) {
4530 return bs->drv->bdrv_debug_is_suspended(bs, tag);
4531 }
f9f05dc5 4532
61007b31
SH
4533 return false;
4534}
f9f05dc5 4535
61007b31
SH
4536/* backing_file can either be relative, or absolute, or a protocol. If it is
4537 * relative, it must be relative to the chain. So, passing in bs->filename
4538 * from a BDS as backing_file should not be done, as that may be relative to
4539 * the CWD rather than the chain. */
4540BlockDriverState *bdrv_find_backing_image(BlockDriverState *bs,
4541 const char *backing_file)
f9f05dc5 4542{
61007b31
SH
4543 char *filename_full = NULL;
4544 char *backing_file_full = NULL;
4545 char *filename_tmp = NULL;
4546 int is_protocol = 0;
4547 BlockDriverState *curr_bs = NULL;
4548 BlockDriverState *retval = NULL;
418661e0 4549 Error *local_error = NULL;
f9f05dc5 4550
61007b31
SH
4551 if (!bs || !bs->drv || !backing_file) {
4552 return NULL;
f9f05dc5
KW
4553 }
4554
61007b31
SH
4555 filename_full = g_malloc(PATH_MAX);
4556 backing_file_full = g_malloc(PATH_MAX);
4557 filename_tmp = g_malloc(PATH_MAX);
f9f05dc5 4558
61007b31 4559 is_protocol = path_has_protocol(backing_file);
f9f05dc5 4560
760e0063 4561 for (curr_bs = bs; curr_bs->backing; curr_bs = curr_bs->backing->bs) {
f9f05dc5 4562
61007b31
SH
4563 /* If either of the filename paths is actually a protocol, then
4564 * compare unmodified paths; otherwise make paths relative */
4565 if (is_protocol || path_has_protocol(curr_bs->backing_file)) {
4566 if (strcmp(backing_file, curr_bs->backing_file) == 0) {
760e0063 4567 retval = curr_bs->backing->bs;
61007b31
SH
4568 break;
4569 }
418661e0
JC
4570 /* Also check against the full backing filename for the image */
4571 bdrv_get_full_backing_filename(curr_bs, backing_file_full, PATH_MAX,
4572 &local_error);
4573 if (local_error == NULL) {
4574 if (strcmp(backing_file, backing_file_full) == 0) {
4575 retval = curr_bs->backing->bs;
4576 break;
4577 }
4578 } else {
4579 error_free(local_error);
4580 local_error = NULL;
4581 }
61007b31
SH
4582 } else {
4583 /* If not an absolute filename path, make it relative to the current
4584 * image's filename path */
4585 path_combine(filename_tmp, PATH_MAX, curr_bs->filename,
4586 backing_file);
f9f05dc5 4587
61007b31
SH
4588 /* We are going to compare absolute pathnames */
4589 if (!realpath(filename_tmp, filename_full)) {
4590 continue;
4591 }
07f07615 4592
61007b31
SH
4593 /* We need to make sure the backing filename we are comparing against
4594 * is relative to the current image filename (or absolute) */
4595 path_combine(filename_tmp, PATH_MAX, curr_bs->filename,
4596 curr_bs->backing_file);
07f07615 4597
61007b31
SH
4598 if (!realpath(filename_tmp, backing_file_full)) {
4599 continue;
4600 }
eb489bb1 4601
61007b31 4602 if (strcmp(backing_file_full, filename_full) == 0) {
760e0063 4603 retval = curr_bs->backing->bs;
61007b31
SH
4604 break;
4605 }
4606 }
eb489bb1
KW
4607 }
4608
61007b31
SH
4609 g_free(filename_full);
4610 g_free(backing_file_full);
4611 g_free(filename_tmp);
4612 return retval;
4613}
4614
61007b31
SH
4615void bdrv_init(void)
4616{
4617 module_call_init(MODULE_INIT_BLOCK);
4618}
29cdb251 4619
61007b31
SH
4620void bdrv_init_with_whitelist(void)
4621{
4622 use_bdrv_whitelist = 1;
4623 bdrv_init();
07f07615
PB
4624}
4625
2b148f39
PB
4626static void coroutine_fn bdrv_co_invalidate_cache(BlockDriverState *bs,
4627 Error **errp)
0f15423c 4628{
4417ab7a 4629 BdrvChild *child, *parent;
9c5e6594 4630 uint64_t perm, shared_perm;
5a8a30db
KW
4631 Error *local_err = NULL;
4632 int ret;
9c98f145 4633 BdrvDirtyBitmap *bm;
5a8a30db 4634
3456a8d1
KW
4635 if (!bs->drv) {
4636 return;
4637 }
4638
04c01a5c 4639 if (!(bs->open_flags & BDRV_O_INACTIVE)) {
7ea2d269
AK
4640 return;
4641 }
7ea2d269 4642
16e977d5 4643 QLIST_FOREACH(child, &bs->children, next) {
2b148f39 4644 bdrv_co_invalidate_cache(child->bs, &local_err);
0d1c5c91 4645 if (local_err) {
0d1c5c91
FZ
4646 error_propagate(errp, local_err);
4647 return;
4648 }
5a8a30db 4649 }
0d1c5c91 4650
dafe0960
KW
4651 /*
4652 * Update permissions, they may differ for inactive nodes.
4653 *
4654 * Note that the required permissions of inactive images are always a
4655 * subset of the permissions required after activating the image. This
4656 * allows us to just get the permissions upfront without restricting
4657 * drv->bdrv_invalidate_cache().
4658 *
4659 * It also means that in error cases, we don't have to try and revert to
4660 * the old permissions (which is an operation that could fail, too). We can
4661 * just keep the extended permissions for the next time that an activation
4662 * of the image is tried.
4663 */
16e977d5 4664 bs->open_flags &= ~BDRV_O_INACTIVE;
dafe0960
KW
4665 bdrv_get_cumulative_perm(bs, &perm, &shared_perm);
4666 ret = bdrv_check_perm(bs, NULL, perm, shared_perm, NULL, &local_err);
4667 if (ret < 0) {
4668 bs->open_flags |= BDRV_O_INACTIVE;
4669 error_propagate(errp, local_err);
4670 return;
4671 }
4672 bdrv_set_perm(bs, perm, shared_perm);
4673
2b148f39
PB
4674 if (bs->drv->bdrv_co_invalidate_cache) {
4675 bs->drv->bdrv_co_invalidate_cache(bs, &local_err);
0d1c5c91
FZ
4676 if (local_err) {
4677 bs->open_flags |= BDRV_O_INACTIVE;
4678 error_propagate(errp, local_err);
4679 return;
4680 }
0f15423c 4681 }
3456a8d1 4682
9c98f145
VSO
4683 for (bm = bdrv_dirty_bitmap_next(bs, NULL); bm;
4684 bm = bdrv_dirty_bitmap_next(bs, bm))
4685 {
4686 bdrv_dirty_bitmap_set_migration(bm, false);
4687 }
4688
5a8a30db
KW
4689 ret = refresh_total_sectors(bs, bs->total_sectors);
4690 if (ret < 0) {
04c01a5c 4691 bs->open_flags |= BDRV_O_INACTIVE;
5a8a30db
KW
4692 error_setg_errno(errp, -ret, "Could not refresh total sector count");
4693 return;
4694 }
4417ab7a
KW
4695
4696 QLIST_FOREACH(parent, &bs->parents, next_parent) {
4697 if (parent->role->activate) {
4698 parent->role->activate(parent, &local_err);
4699 if (local_err) {
78fc3b3a 4700 bs->open_flags |= BDRV_O_INACTIVE;
4417ab7a
KW
4701 error_propagate(errp, local_err);
4702 return;
4703 }
4704 }
4705 }
0f15423c
AL
4706}
4707
2b148f39
PB
4708typedef struct InvalidateCacheCo {
4709 BlockDriverState *bs;
4710 Error **errp;
4711 bool done;
4712} InvalidateCacheCo;
4713
4714static void coroutine_fn bdrv_invalidate_cache_co_entry(void *opaque)
4715{
4716 InvalidateCacheCo *ico = opaque;
4717 bdrv_co_invalidate_cache(ico->bs, ico->errp);
4718 ico->done = true;
4720cbee 4719 aio_wait_kick();
2b148f39
PB
4720}
4721
4722void bdrv_invalidate_cache(BlockDriverState *bs, Error **errp)
4723{
4724 Coroutine *co;
4725 InvalidateCacheCo ico = {
4726 .bs = bs,
4727 .done = false,
4728 .errp = errp
4729 };
4730
4731 if (qemu_in_coroutine()) {
4732 /* Fast-path if already in coroutine context */
4733 bdrv_invalidate_cache_co_entry(&ico);
4734 } else {
4735 co = qemu_coroutine_create(bdrv_invalidate_cache_co_entry, &ico);
4720cbee 4736 bdrv_coroutine_enter(bs, co);
2b148f39
PB
4737 BDRV_POLL_WHILE(bs, !ico.done);
4738 }
4739}
4740
5a8a30db 4741void bdrv_invalidate_cache_all(Error **errp)
0f15423c 4742{
7c8eece4 4743 BlockDriverState *bs;
5a8a30db 4744 Error *local_err = NULL;
88be7b4b 4745 BdrvNextIterator it;
0f15423c 4746
88be7b4b 4747 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
ed78cda3
SH
4748 AioContext *aio_context = bdrv_get_aio_context(bs);
4749
4750 aio_context_acquire(aio_context);
5a8a30db 4751 bdrv_invalidate_cache(bs, &local_err);
ed78cda3 4752 aio_context_release(aio_context);
5a8a30db
KW
4753 if (local_err) {
4754 error_propagate(errp, local_err);
5e003f17 4755 bdrv_next_cleanup(&it);
5a8a30db
KW
4756 return;
4757 }
0f15423c
AL
4758 }
4759}
4760
9e37271f
KW
4761static bool bdrv_has_bds_parent(BlockDriverState *bs, bool only_active)
4762{
4763 BdrvChild *parent;
4764
4765 QLIST_FOREACH(parent, &bs->parents, next_parent) {
4766 if (parent->role->parent_is_bds) {
4767 BlockDriverState *parent_bs = parent->opaque;
4768 if (!only_active || !(parent_bs->open_flags & BDRV_O_INACTIVE)) {
4769 return true;
4770 }
4771 }
4772 }
4773
4774 return false;
4775}
4776
4777static int bdrv_inactivate_recurse(BlockDriverState *bs)
76b1c7fe 4778{
cfa1a572 4779 BdrvChild *child, *parent;
9e37271f 4780 uint64_t perm, shared_perm;
76b1c7fe
KW
4781 int ret;
4782
d470ad42
HR
4783 if (!bs->drv) {
4784 return -ENOMEDIUM;
4785 }
4786
9e37271f
KW
4787 /* Make sure that we don't inactivate a child before its parent.
4788 * It will be covered by recursion from the yet active parent. */
4789 if (bdrv_has_bds_parent(bs, true)) {
4790 return 0;
4791 }
4792
4793 assert(!(bs->open_flags & BDRV_O_INACTIVE));
4794
4795 /* Inactivate this node */
4796 if (bs->drv->bdrv_inactivate) {
76b1c7fe
KW
4797 ret = bs->drv->bdrv_inactivate(bs);
4798 if (ret < 0) {
4799 return ret;
4800 }
4801 }
4802
9e37271f
KW
4803 QLIST_FOREACH(parent, &bs->parents, next_parent) {
4804 if (parent->role->inactivate) {
4805 ret = parent->role->inactivate(parent);
4806 if (ret < 0) {
4807 return ret;
cfa1a572
KW
4808 }
4809 }
9e37271f 4810 }
9c5e6594 4811
9e37271f 4812 bs->open_flags |= BDRV_O_INACTIVE;
7d5b5261 4813
9e37271f
KW
4814 /* Update permissions, they may differ for inactive nodes */
4815 bdrv_get_cumulative_perm(bs, &perm, &shared_perm);
4816 bdrv_check_perm(bs, NULL, perm, shared_perm, NULL, &error_abort);
4817 bdrv_set_perm(bs, perm, shared_perm);
38701b6a 4818
9e37271f
KW
4819
4820 /* Recursively inactivate children */
38701b6a 4821 QLIST_FOREACH(child, &bs->children, next) {
9e37271f 4822 ret = bdrv_inactivate_recurse(child->bs);
38701b6a
KW
4823 if (ret < 0) {
4824 return ret;
4825 }
4826 }
4827
76b1c7fe
KW
4828 return 0;
4829}
4830
4831int bdrv_inactivate_all(void)
4832{
79720af6 4833 BlockDriverState *bs = NULL;
88be7b4b 4834 BdrvNextIterator it;
aad0b7a0 4835 int ret = 0;
bd6458e4 4836 GSList *aio_ctxs = NULL, *ctx;
76b1c7fe 4837
88be7b4b 4838 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
bd6458e4
PB
4839 AioContext *aio_context = bdrv_get_aio_context(bs);
4840
4841 if (!g_slist_find(aio_ctxs, aio_context)) {
4842 aio_ctxs = g_slist_prepend(aio_ctxs, aio_context);
4843 aio_context_acquire(aio_context);
4844 }
aad0b7a0 4845 }
76b1c7fe 4846
9e37271f
KW
4847 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
4848 /* Nodes with BDS parents are covered by recursion from the last
4849 * parent that gets inactivated. Don't inactivate them a second
4850 * time if that has already happened. */
4851 if (bdrv_has_bds_parent(bs, false)) {
4852 continue;
4853 }
4854 ret = bdrv_inactivate_recurse(bs);
4855 if (ret < 0) {
4856 bdrv_next_cleanup(&it);
4857 goto out;
76b1c7fe
KW
4858 }
4859 }
4860
aad0b7a0 4861out:
bd6458e4
PB
4862 for (ctx = aio_ctxs; ctx != NULL; ctx = ctx->next) {
4863 AioContext *aio_context = ctx->data;
4864 aio_context_release(aio_context);
aad0b7a0 4865 }
bd6458e4 4866 g_slist_free(aio_ctxs);
aad0b7a0
FZ
4867
4868 return ret;
76b1c7fe
KW
4869}
4870
19cb3738
FB
4871/**************************************************************/
4872/* removable device support */
4873
4874/**
4875 * Return TRUE if the media is present
4876 */
e031f750 4877bool bdrv_is_inserted(BlockDriverState *bs)
19cb3738
FB
4878{
4879 BlockDriver *drv = bs->drv;
28d7a789 4880 BdrvChild *child;
a1aff5bf 4881
e031f750
HR
4882 if (!drv) {
4883 return false;
4884 }
28d7a789
HR
4885 if (drv->bdrv_is_inserted) {
4886 return drv->bdrv_is_inserted(bs);
4887 }
4888 QLIST_FOREACH(child, &bs->children, next) {
4889 if (!bdrv_is_inserted(child->bs)) {
4890 return false;
4891 }
e031f750 4892 }
28d7a789 4893 return true;
19cb3738
FB
4894}
4895
19cb3738
FB
4896/**
4897 * If eject_flag is TRUE, eject the media. Otherwise, close the tray
4898 */
f36f3949 4899void bdrv_eject(BlockDriverState *bs, bool eject_flag)
19cb3738
FB
4900{
4901 BlockDriver *drv = bs->drv;
19cb3738 4902
822e1cd1
MA
4903 if (drv && drv->bdrv_eject) {
4904 drv->bdrv_eject(bs, eject_flag);
19cb3738
FB
4905 }
4906}
4907
19cb3738
FB
4908/**
4909 * Lock or unlock the media (if it is locked, the user won't be able
4910 * to eject it manually).
4911 */
025e849a 4912void bdrv_lock_medium(BlockDriverState *bs, bool locked)
19cb3738
FB
4913{
4914 BlockDriver *drv = bs->drv;
4915
025e849a 4916 trace_bdrv_lock_medium(bs, locked);
b8c6d095 4917
025e849a
MA
4918 if (drv && drv->bdrv_lock_medium) {
4919 drv->bdrv_lock_medium(bs, locked);
19cb3738
FB
4920 }
4921}
985a03b0 4922
9fcb0251
FZ
4923/* Get a reference to bs */
4924void bdrv_ref(BlockDriverState *bs)
4925{
4926 bs->refcnt++;
4927}
4928
4929/* Release a previously grabbed reference to bs.
4930 * If after releasing, reference count is zero, the BlockDriverState is
4931 * deleted. */
4932void bdrv_unref(BlockDriverState *bs)
4933{
9a4d5ca6
JC
4934 if (!bs) {
4935 return;
4936 }
9fcb0251
FZ
4937 assert(bs->refcnt > 0);
4938 if (--bs->refcnt == 0) {
4939 bdrv_delete(bs);
4940 }
4941}
4942
fbe40ff7
FZ
4943struct BdrvOpBlocker {
4944 Error *reason;
4945 QLIST_ENTRY(BdrvOpBlocker) list;
4946};
4947
4948bool bdrv_op_is_blocked(BlockDriverState *bs, BlockOpType op, Error **errp)
4949{
4950 BdrvOpBlocker *blocker;
4951 assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
4952 if (!QLIST_EMPTY(&bs->op_blockers[op])) {
4953 blocker = QLIST_FIRST(&bs->op_blockers[op]);
4b576648
MA
4954 error_propagate_prepend(errp, error_copy(blocker->reason),
4955 "Node '%s' is busy: ",
4956 bdrv_get_device_or_node_name(bs));
fbe40ff7
FZ
4957 return true;
4958 }
4959 return false;
4960}
4961
4962void bdrv_op_block(BlockDriverState *bs, BlockOpType op, Error *reason)
4963{
4964 BdrvOpBlocker *blocker;
4965 assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
4966
5839e53b 4967 blocker = g_new0(BdrvOpBlocker, 1);
fbe40ff7
FZ
4968 blocker->reason = reason;
4969 QLIST_INSERT_HEAD(&bs->op_blockers[op], blocker, list);
4970}
4971
4972void bdrv_op_unblock(BlockDriverState *bs, BlockOpType op, Error *reason)
4973{
4974 BdrvOpBlocker *blocker, *next;
4975 assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
4976 QLIST_FOREACH_SAFE(blocker, &bs->op_blockers[op], list, next) {
4977 if (blocker->reason == reason) {
4978 QLIST_REMOVE(blocker, list);
4979 g_free(blocker);
4980 }
4981 }
4982}
4983
4984void bdrv_op_block_all(BlockDriverState *bs, Error *reason)
4985{
4986 int i;
4987 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
4988 bdrv_op_block(bs, i, reason);
4989 }
4990}
4991
4992void bdrv_op_unblock_all(BlockDriverState *bs, Error *reason)
4993{
4994 int i;
4995 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
4996 bdrv_op_unblock(bs, i, reason);
4997 }
4998}
4999
5000bool bdrv_op_blocker_is_empty(BlockDriverState *bs)
5001{
5002 int i;
5003
5004 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
5005 if (!QLIST_EMPTY(&bs->op_blockers[i])) {
5006 return false;
5007 }
5008 }
5009 return true;
5010}
5011
d92ada22
LC
5012void bdrv_img_create(const char *filename, const char *fmt,
5013 const char *base_filename, const char *base_fmt,
9217283d
FZ
5014 char *options, uint64_t img_size, int flags, bool quiet,
5015 Error **errp)
f88e1a42 5016{
83d0521a
CL
5017 QemuOptsList *create_opts = NULL;
5018 QemuOpts *opts = NULL;
5019 const char *backing_fmt, *backing_file;
5020 int64_t size;
f88e1a42 5021 BlockDriver *drv, *proto_drv;
cc84d90f 5022 Error *local_err = NULL;
f88e1a42
JS
5023 int ret = 0;
5024
5025 /* Find driver and parse its options */
5026 drv = bdrv_find_format(fmt);
5027 if (!drv) {
71c79813 5028 error_setg(errp, "Unknown file format '%s'", fmt);
d92ada22 5029 return;
f88e1a42
JS
5030 }
5031
b65a5e12 5032 proto_drv = bdrv_find_protocol(filename, true, errp);
f88e1a42 5033 if (!proto_drv) {
d92ada22 5034 return;
f88e1a42
JS
5035 }
5036
c6149724
HR
5037 if (!drv->create_opts) {
5038 error_setg(errp, "Format driver '%s' does not support image creation",
5039 drv->format_name);
5040 return;
5041 }
5042
5043 if (!proto_drv->create_opts) {
5044 error_setg(errp, "Protocol driver '%s' does not support image creation",
5045 proto_drv->format_name);
5046 return;
5047 }
5048
c282e1fd
CL
5049 create_opts = qemu_opts_append(create_opts, drv->create_opts);
5050 create_opts = qemu_opts_append(create_opts, proto_drv->create_opts);
f88e1a42
JS
5051
5052 /* Create parameter list with default values */
83d0521a 5053 opts = qemu_opts_create(create_opts, NULL, 0, &error_abort);
39101f25 5054 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, img_size, &error_abort);
f88e1a42
JS
5055
5056 /* Parse -o options */
5057 if (options) {
dc523cd3
MA
5058 qemu_opts_do_parse(opts, options, NULL, &local_err);
5059 if (local_err) {
f88e1a42
JS
5060 goto out;
5061 }
5062 }
5063
5064 if (base_filename) {
f43e47db 5065 qemu_opt_set(opts, BLOCK_OPT_BACKING_FILE, base_filename, &local_err);
6be4194b 5066 if (local_err) {
71c79813
LC
5067 error_setg(errp, "Backing file not supported for file format '%s'",
5068 fmt);
f88e1a42
JS
5069 goto out;
5070 }
5071 }
5072
5073 if (base_fmt) {
f43e47db 5074 qemu_opt_set(opts, BLOCK_OPT_BACKING_FMT, base_fmt, &local_err);
6be4194b 5075 if (local_err) {
71c79813
LC
5076 error_setg(errp, "Backing file format not supported for file "
5077 "format '%s'", fmt);
f88e1a42
JS
5078 goto out;
5079 }
5080 }
5081
83d0521a
CL
5082 backing_file = qemu_opt_get(opts, BLOCK_OPT_BACKING_FILE);
5083 if (backing_file) {
5084 if (!strcmp(filename, backing_file)) {
71c79813
LC
5085 error_setg(errp, "Error: Trying to create an image with the "
5086 "same filename as the backing file");
792da93a
JS
5087 goto out;
5088 }
5089 }
5090
83d0521a 5091 backing_fmt = qemu_opt_get(opts, BLOCK_OPT_BACKING_FMT);
f88e1a42 5092
6e6e55f5
JS
5093 /* The size for the image must always be specified, unless we have a backing
5094 * file and we have not been forbidden from opening it. */
a8b42a1c 5095 size = qemu_opt_get_size(opts, BLOCK_OPT_SIZE, img_size);
6e6e55f5
JS
5096 if (backing_file && !(flags & BDRV_O_NO_BACKING)) {
5097 BlockDriverState *bs;
5098 char *full_backing = g_new0(char, PATH_MAX);
5099 int back_flags;
5100 QDict *backing_options = NULL;
5101
5102 bdrv_get_full_backing_filename_from_filename(filename, backing_file,
5103 full_backing, PATH_MAX,
5104 &local_err);
5105 if (local_err) {
5106 g_free(full_backing);
5107 goto out;
5108 }
29168018 5109
6e6e55f5
JS
5110 /* backing files always opened read-only */
5111 back_flags = flags;
5112 back_flags &= ~(BDRV_O_RDWR | BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING);
f88e1a42 5113
cc954f01 5114 backing_options = qdict_new();
6e6e55f5 5115 if (backing_fmt) {
6e6e55f5
JS
5116 qdict_put_str(backing_options, "driver", backing_fmt);
5117 }
cc954f01 5118 qdict_put_bool(backing_options, BDRV_OPT_FORCE_SHARE, true);
e6641719 5119
6e6e55f5
JS
5120 bs = bdrv_open(full_backing, NULL, backing_options, back_flags,
5121 &local_err);
5122 g_free(full_backing);
5123 if (!bs && size != -1) {
5124 /* Couldn't open BS, but we have a size, so it's nonfatal */
5125 warn_reportf_err(local_err,
5126 "Could not verify backing image. "
5127 "This may become an error in future versions.\n");
5128 local_err = NULL;
5129 } else if (!bs) {
5130 /* Couldn't open bs, do not have size */
5131 error_append_hint(&local_err,
5132 "Could not open backing image to determine size.\n");
5133 goto out;
5134 } else {
5135 if (size == -1) {
5136 /* Opened BS, have no size */
5137 size = bdrv_getlength(bs);
5138 if (size < 0) {
5139 error_setg_errno(errp, -size, "Could not get size of '%s'",
5140 backing_file);
5141 bdrv_unref(bs);
5142 goto out;
5143 }
5144 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, size, &error_abort);
52bf1e72 5145 }
66f6b814 5146 bdrv_unref(bs);
f88e1a42 5147 }
6e6e55f5
JS
5148 } /* (backing_file && !(flags & BDRV_O_NO_BACKING)) */
5149
5150 if (size == -1) {
5151 error_setg(errp, "Image creation needs a size parameter");
5152 goto out;
f88e1a42
JS
5153 }
5154
f382d43a 5155 if (!quiet) {
fe646693 5156 printf("Formatting '%s', fmt=%s ", filename, fmt);
43c5d8f8 5157 qemu_opts_print(opts, " ");
f382d43a
MR
5158 puts("");
5159 }
83d0521a 5160
c282e1fd 5161 ret = bdrv_create(drv, filename, opts, &local_err);
83d0521a 5162
cc84d90f
HR
5163 if (ret == -EFBIG) {
5164 /* This is generally a better message than whatever the driver would
5165 * deliver (especially because of the cluster_size_hint), since that
5166 * is most probably not much different from "image too large". */
5167 const char *cluster_size_hint = "";
83d0521a 5168 if (qemu_opt_get_size(opts, BLOCK_OPT_CLUSTER_SIZE, 0)) {
cc84d90f 5169 cluster_size_hint = " (try using a larger cluster size)";
f88e1a42 5170 }
cc84d90f
HR
5171 error_setg(errp, "The image size is too large for file format '%s'"
5172 "%s", fmt, cluster_size_hint);
5173 error_free(local_err);
5174 local_err = NULL;
f88e1a42
JS
5175 }
5176
5177out:
83d0521a
CL
5178 qemu_opts_del(opts);
5179 qemu_opts_free(create_opts);
621ff94d 5180 error_propagate(errp, local_err);
f88e1a42 5181}
85d126f3
SH
5182
5183AioContext *bdrv_get_aio_context(BlockDriverState *bs)
5184{
33f2a757 5185 return bs ? bs->aio_context : qemu_get_aio_context();
dcd04228
SH
5186}
5187
052a7572
FZ
5188void bdrv_coroutine_enter(BlockDriverState *bs, Coroutine *co)
5189{
5190 aio_co_enter(bdrv_get_aio_context(bs), co);
5191}
5192
e8a095da
SH
5193static void bdrv_do_remove_aio_context_notifier(BdrvAioNotifier *ban)
5194{
5195 QLIST_REMOVE(ban, list);
5196 g_free(ban);
5197}
5198
dcd04228
SH
5199void bdrv_detach_aio_context(BlockDriverState *bs)
5200{
e8a095da 5201 BdrvAioNotifier *baf, *baf_tmp;
b97511c7 5202 BdrvChild *child;
33384421 5203
dcd04228
SH
5204 if (!bs->drv) {
5205 return;
5206 }
5207
e8a095da
SH
5208 assert(!bs->walking_aio_notifiers);
5209 bs->walking_aio_notifiers = true;
5210 QLIST_FOREACH_SAFE(baf, &bs->aio_notifiers, list, baf_tmp) {
5211 if (baf->deleted) {
5212 bdrv_do_remove_aio_context_notifier(baf);
5213 } else {
5214 baf->detach_aio_context(baf->opaque);
5215 }
33384421 5216 }
e8a095da
SH
5217 /* Never mind iterating again to check for ->deleted. bdrv_close() will
5218 * remove remaining aio notifiers if we aren't called again.
5219 */
5220 bs->walking_aio_notifiers = false;
33384421 5221
dcd04228
SH
5222 if (bs->drv->bdrv_detach_aio_context) {
5223 bs->drv->bdrv_detach_aio_context(bs);
5224 }
b97511c7
HR
5225 QLIST_FOREACH(child, &bs->children, next) {
5226 bdrv_detach_aio_context(child->bs);
dcd04228
SH
5227 }
5228
5229 bs->aio_context = NULL;
5230}
5231
5232void bdrv_attach_aio_context(BlockDriverState *bs,
5233 AioContext *new_context)
5234{
e8a095da 5235 BdrvAioNotifier *ban, *ban_tmp;
b97511c7 5236 BdrvChild *child;
33384421 5237
dcd04228
SH
5238 if (!bs->drv) {
5239 return;
5240 }
5241
5242 bs->aio_context = new_context;
5243
b97511c7
HR
5244 QLIST_FOREACH(child, &bs->children, next) {
5245 bdrv_attach_aio_context(child->bs, new_context);
dcd04228
SH
5246 }
5247 if (bs->drv->bdrv_attach_aio_context) {
5248 bs->drv->bdrv_attach_aio_context(bs, new_context);
5249 }
33384421 5250
e8a095da
SH
5251 assert(!bs->walking_aio_notifiers);
5252 bs->walking_aio_notifiers = true;
5253 QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_tmp) {
5254 if (ban->deleted) {
5255 bdrv_do_remove_aio_context_notifier(ban);
5256 } else {
5257 ban->attached_aio_context(new_context, ban->opaque);
5258 }
33384421 5259 }
e8a095da 5260 bs->walking_aio_notifiers = false;
dcd04228
SH
5261}
5262
5263void bdrv_set_aio_context(BlockDriverState *bs, AioContext *new_context)
5264{
aabf5910 5265 AioContext *ctx = bdrv_get_aio_context(bs);
c2b6428d 5266
aabf5910 5267 aio_disable_external(ctx);
6cd5c9d7 5268 bdrv_parent_drained_begin(bs, NULL, false);
53ec73e2 5269 bdrv_drain(bs); /* ensure there are no in-flight requests */
dcd04228 5270
c2b6428d
PB
5271 while (aio_poll(ctx, false)) {
5272 /* wait for all bottom halves to execute */
5273 }
5274
dcd04228
SH
5275 bdrv_detach_aio_context(bs);
5276
5277 /* This function executes in the old AioContext so acquire the new one in
5278 * case it runs in a different thread.
5279 */
5280 aio_context_acquire(new_context);
5281 bdrv_attach_aio_context(bs, new_context);
6cd5c9d7 5282 bdrv_parent_drained_end(bs, NULL, false);
aabf5910 5283 aio_enable_external(ctx);
dcd04228 5284 aio_context_release(new_context);
85d126f3 5285}
d616b224 5286
33384421
HR
5287void bdrv_add_aio_context_notifier(BlockDriverState *bs,
5288 void (*attached_aio_context)(AioContext *new_context, void *opaque),
5289 void (*detach_aio_context)(void *opaque), void *opaque)
5290{
5291 BdrvAioNotifier *ban = g_new(BdrvAioNotifier, 1);
5292 *ban = (BdrvAioNotifier){
5293 .attached_aio_context = attached_aio_context,
5294 .detach_aio_context = detach_aio_context,
5295 .opaque = opaque
5296 };
5297
5298 QLIST_INSERT_HEAD(&bs->aio_notifiers, ban, list);
5299}
5300
5301void bdrv_remove_aio_context_notifier(BlockDriverState *bs,
5302 void (*attached_aio_context)(AioContext *,
5303 void *),
5304 void (*detach_aio_context)(void *),
5305 void *opaque)
5306{
5307 BdrvAioNotifier *ban, *ban_next;
5308
5309 QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_next) {
5310 if (ban->attached_aio_context == attached_aio_context &&
5311 ban->detach_aio_context == detach_aio_context &&
e8a095da
SH
5312 ban->opaque == opaque &&
5313 ban->deleted == false)
33384421 5314 {
e8a095da
SH
5315 if (bs->walking_aio_notifiers) {
5316 ban->deleted = true;
5317 } else {
5318 bdrv_do_remove_aio_context_notifier(ban);
5319 }
33384421
HR
5320 return;
5321 }
5322 }
5323
5324 abort();
5325}
5326
77485434 5327int bdrv_amend_options(BlockDriverState *bs, QemuOpts *opts,
d1402b50
HR
5328 BlockDriverAmendStatusCB *status_cb, void *cb_opaque,
5329 Error **errp)
6f176b48 5330{
d470ad42 5331 if (!bs->drv) {
d1402b50 5332 error_setg(errp, "Node is ejected");
d470ad42
HR
5333 return -ENOMEDIUM;
5334 }
c282e1fd 5335 if (!bs->drv->bdrv_amend_options) {
d1402b50
HR
5336 error_setg(errp, "Block driver '%s' does not support option amendment",
5337 bs->drv->format_name);
6f176b48
HR
5338 return -ENOTSUP;
5339 }
d1402b50 5340 return bs->drv->bdrv_amend_options(bs, opts, status_cb, cb_opaque, errp);
6f176b48 5341}
f6186f49 5342
b5042a36
BC
5343/* This function will be called by the bdrv_recurse_is_first_non_filter method
5344 * of block filter and by bdrv_is_first_non_filter.
5345 * It is used to test if the given bs is the candidate or recurse more in the
5346 * node graph.
212a5a8f 5347 */
b5042a36 5348bool bdrv_recurse_is_first_non_filter(BlockDriverState *bs,
212a5a8f 5349 BlockDriverState *candidate)
f6186f49 5350{
b5042a36
BC
5351 /* return false if basic checks fails */
5352 if (!bs || !bs->drv) {
212a5a8f 5353 return false;
f6186f49
BC
5354 }
5355
b5042a36
BC
5356 /* the code reached a non block filter driver -> check if the bs is
5357 * the same as the candidate. It's the recursion termination condition.
5358 */
5359 if (!bs->drv->is_filter) {
5360 return bs == candidate;
212a5a8f 5361 }
b5042a36 5362 /* Down this path the driver is a block filter driver */
212a5a8f 5363
b5042a36
BC
5364 /* If the block filter recursion method is defined use it to recurse down
5365 * the node graph.
5366 */
5367 if (bs->drv->bdrv_recurse_is_first_non_filter) {
212a5a8f 5368 return bs->drv->bdrv_recurse_is_first_non_filter(bs, candidate);
f6186f49
BC
5369 }
5370
b5042a36
BC
5371 /* the driver is a block filter but don't allow to recurse -> return false
5372 */
5373 return false;
f6186f49
BC
5374}
5375
212a5a8f
BC
5376/* This function checks if the candidate is the first non filter bs down it's
5377 * bs chain. Since we don't have pointers to parents it explore all bs chains
5378 * from the top. Some filters can choose not to pass down the recursion.
5379 */
5380bool bdrv_is_first_non_filter(BlockDriverState *candidate)
f6186f49 5381{
7c8eece4 5382 BlockDriverState *bs;
88be7b4b 5383 BdrvNextIterator it;
212a5a8f
BC
5384
5385 /* walk down the bs forest recursively */
88be7b4b 5386 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
212a5a8f
BC
5387 bool perm;
5388
b5042a36 5389 /* try to recurse in this top level bs */
e6dc8a1f 5390 perm = bdrv_recurse_is_first_non_filter(bs, candidate);
212a5a8f
BC
5391
5392 /* candidate is the first non filter */
5393 if (perm) {
5e003f17 5394 bdrv_next_cleanup(&it);
212a5a8f
BC
5395 return true;
5396 }
5397 }
5398
5399 return false;
f6186f49 5400}
09158f00 5401
e12f3784
WC
5402BlockDriverState *check_to_replace_node(BlockDriverState *parent_bs,
5403 const char *node_name, Error **errp)
09158f00
BC
5404{
5405 BlockDriverState *to_replace_bs = bdrv_find_node(node_name);
5a7e7a0b
SH
5406 AioContext *aio_context;
5407
09158f00
BC
5408 if (!to_replace_bs) {
5409 error_setg(errp, "Node name '%s' not found", node_name);
5410 return NULL;
5411 }
5412
5a7e7a0b
SH
5413 aio_context = bdrv_get_aio_context(to_replace_bs);
5414 aio_context_acquire(aio_context);
5415
09158f00 5416 if (bdrv_op_is_blocked(to_replace_bs, BLOCK_OP_TYPE_REPLACE, errp)) {
5a7e7a0b
SH
5417 to_replace_bs = NULL;
5418 goto out;
09158f00
BC
5419 }
5420
5421 /* We don't want arbitrary node of the BDS chain to be replaced only the top
5422 * most non filter in order to prevent data corruption.
5423 * Another benefit is that this tests exclude backing files which are
5424 * blocked by the backing blockers.
5425 */
e12f3784 5426 if (!bdrv_recurse_is_first_non_filter(parent_bs, to_replace_bs)) {
09158f00 5427 error_setg(errp, "Only top most non filter can be replaced");
5a7e7a0b
SH
5428 to_replace_bs = NULL;
5429 goto out;
09158f00
BC
5430 }
5431
5a7e7a0b
SH
5432out:
5433 aio_context_release(aio_context);
09158f00
BC
5434 return to_replace_bs;
5435}
448ad91d 5436
91af7014
HR
5437static bool append_open_options(QDict *d, BlockDriverState *bs)
5438{
5439 const QDictEntry *entry;
9e700c1a 5440 QemuOptDesc *desc;
91af7014
HR
5441 bool found_any = false;
5442
5443 for (entry = qdict_first(bs->options); entry;
5444 entry = qdict_next(bs->options, entry))
5445 {
a600aadd 5446 /* Exclude all non-driver-specific options */
9e700c1a
KW
5447 for (desc = bdrv_runtime_opts.desc; desc->name; desc++) {
5448 if (!strcmp(qdict_entry_key(entry), desc->name)) {
5449 break;
5450 }
5451 }
5452 if (desc->name) {
5453 continue;
5454 }
5455
f5a74a5a
MAL
5456 qdict_put_obj(d, qdict_entry_key(entry),
5457 qobject_ref(qdict_entry_value(entry)));
9e700c1a 5458 found_any = true;
91af7014
HR
5459 }
5460
5461 return found_any;
5462}
5463
5464/* Updates the following BDS fields:
5465 * - exact_filename: A filename which may be used for opening a block device
5466 * which (mostly) equals the given BDS (even without any
5467 * other options; so reading and writing must return the same
5468 * results, but caching etc. may be different)
5469 * - full_open_options: Options which, when given when opening a block device
5470 * (without a filename), result in a BDS (mostly)
5471 * equalling the given one
5472 * - filename: If exact_filename is set, it is copied here. Otherwise,
5473 * full_open_options is converted to a JSON object, prefixed with
5474 * "json:" (for use through the JSON pseudo protocol) and put here.
5475 */
5476void bdrv_refresh_filename(BlockDriverState *bs)
5477{
5478 BlockDriver *drv = bs->drv;
5479 QDict *opts;
5480
5481 if (!drv) {
5482 return;
5483 }
5484
5485 /* This BDS's file name will most probably depend on its file's name, so
5486 * refresh that first */
5487 if (bs->file) {
9a4f4c31 5488 bdrv_refresh_filename(bs->file->bs);
91af7014
HR
5489 }
5490
5491 if (drv->bdrv_refresh_filename) {
5492 /* Obsolete information is of no use here, so drop the old file name
5493 * information before refreshing it */
5494 bs->exact_filename[0] = '\0';
5495 if (bs->full_open_options) {
cb3e7f08 5496 qobject_unref(bs->full_open_options);
91af7014
HR
5497 bs->full_open_options = NULL;
5498 }
5499
4cdd01d3
KW
5500 opts = qdict_new();
5501 append_open_options(opts, bs);
5502 drv->bdrv_refresh_filename(bs, opts);
cb3e7f08 5503 qobject_unref(opts);
91af7014
HR
5504 } else if (bs->file) {
5505 /* Try to reconstruct valid information from the underlying file */
5506 bool has_open_options;
5507
5508 bs->exact_filename[0] = '\0';
5509 if (bs->full_open_options) {
cb3e7f08 5510 qobject_unref(bs->full_open_options);
91af7014
HR
5511 bs->full_open_options = NULL;
5512 }
5513
5514 opts = qdict_new();
5515 has_open_options = append_open_options(opts, bs);
5516
5517 /* If no specific options have been given for this BDS, the filename of
5518 * the underlying file should suffice for this one as well */
9a4f4c31
KW
5519 if (bs->file->bs->exact_filename[0] && !has_open_options) {
5520 strcpy(bs->exact_filename, bs->file->bs->exact_filename);
91af7014
HR
5521 }
5522 /* Reconstructing the full options QDict is simple for most format block
5523 * drivers, as long as the full options are known for the underlying
5524 * file BDS. The full options QDict of that file BDS should somehow
5525 * contain a representation of the filename, therefore the following
5526 * suffices without querying the (exact_)filename of this BDS. */
9a4f4c31 5527 if (bs->file->bs->full_open_options) {
46f5ac20 5528 qdict_put_str(opts, "driver", drv->format_name);
f5a74a5a
MAL
5529 qdict_put(opts, "file",
5530 qobject_ref(bs->file->bs->full_open_options));
91af7014
HR
5531
5532 bs->full_open_options = opts;
5533 } else {
cb3e7f08 5534 qobject_unref(opts);
91af7014
HR
5535 }
5536 } else if (!bs->full_open_options && qdict_size(bs->options)) {
5537 /* There is no underlying file BDS (at least referenced by BDS.file),
5538 * so the full options QDict should be equal to the options given
5539 * specifically for this block device when it was opened (plus the
5540 * driver specification).
5541 * Because those options don't change, there is no need to update
5542 * full_open_options when it's already set. */
5543
5544 opts = qdict_new();
5545 append_open_options(opts, bs);
46f5ac20 5546 qdict_put_str(opts, "driver", drv->format_name);
91af7014
HR
5547
5548 if (bs->exact_filename[0]) {
5549 /* This may not work for all block protocol drivers (some may
5550 * require this filename to be parsed), but we have to find some
5551 * default solution here, so just include it. If some block driver
5552 * does not support pure options without any filename at all or
5553 * needs some special format of the options QDict, it needs to
5554 * implement the driver-specific bdrv_refresh_filename() function.
5555 */
46f5ac20 5556 qdict_put_str(opts, "filename", bs->exact_filename);
91af7014
HR
5557 }
5558
5559 bs->full_open_options = opts;
5560 }
5561
5562 if (bs->exact_filename[0]) {
5563 pstrcpy(bs->filename, sizeof(bs->filename), bs->exact_filename);
5564 } else if (bs->full_open_options) {
5565 QString *json = qobject_to_json(QOBJECT(bs->full_open_options));
5566 snprintf(bs->filename, sizeof(bs->filename), "json:%s",
5567 qstring_get_str(json));
cb3e7f08 5568 qobject_unref(json);
91af7014
HR
5569 }
5570}
e06018ad
WC
5571
5572/*
5573 * Hot add/remove a BDS's child. So the user can take a child offline when
5574 * it is broken and take a new child online
5575 */
5576void bdrv_add_child(BlockDriverState *parent_bs, BlockDriverState *child_bs,
5577 Error **errp)
5578{
5579
5580 if (!parent_bs->drv || !parent_bs->drv->bdrv_add_child) {
5581 error_setg(errp, "The node %s does not support adding a child",
5582 bdrv_get_device_or_node_name(parent_bs));
5583 return;
5584 }
5585
5586 if (!QLIST_EMPTY(&child_bs->parents)) {
5587 error_setg(errp, "The node %s already has a parent",
5588 child_bs->node_name);
5589 return;
5590 }
5591
5592 parent_bs->drv->bdrv_add_child(parent_bs, child_bs, errp);
5593}
5594
5595void bdrv_del_child(BlockDriverState *parent_bs, BdrvChild *child, Error **errp)
5596{
5597 BdrvChild *tmp;
5598
5599 if (!parent_bs->drv || !parent_bs->drv->bdrv_del_child) {
5600 error_setg(errp, "The node %s does not support removing a child",
5601 bdrv_get_device_or_node_name(parent_bs));
5602 return;
5603 }
5604
5605 QLIST_FOREACH(tmp, &parent_bs->children, next) {
5606 if (tmp == child) {
5607 break;
5608 }
5609 }
5610
5611 if (!tmp) {
5612 error_setg(errp, "The node %s does not have a child named %s",
5613 bdrv_get_device_or_node_name(parent_bs),
5614 bdrv_get_device_or_node_name(child->bs));
5615 return;
5616 }
5617
5618 parent_bs->drv->bdrv_del_child(parent_bs, child, errp);
5619}
67b792f5
VSO
5620
5621bool bdrv_can_store_new_dirty_bitmap(BlockDriverState *bs, const char *name,
5622 uint32_t granularity, Error **errp)
5623{
5624 BlockDriver *drv = bs->drv;
5625
5626 if (!drv) {
5627 error_setg_errno(errp, ENOMEDIUM,
5628 "Can't store persistent bitmaps to %s",
5629 bdrv_get_device_or_node_name(bs));
5630 return false;
5631 }
5632
5633 if (!drv->bdrv_can_store_new_dirty_bitmap) {
5634 error_setg_errno(errp, ENOTSUP,
5635 "Can't store persistent bitmaps to %s",
5636 bdrv_get_device_or_node_name(bs));
5637 return false;
5638 }
5639
5640 return drv->bdrv_can_store_new_dirty_bitmap(bs, name, granularity, errp);
5641}