blob: a307c151a8dbba4a145393afa56f18d40088c552 [file] [log] [blame]
bellardfc01f7e2003-06-30 10:03:06 +00001/*
2 * QEMU System Emulator block driver
ths5fafdf22007-09-16 21:08:06 +00003 *
bellardfc01f7e2003-06-30 10:03:06 +00004 * Copyright (c) 2003 Fabrice Bellard
Vladimir Sementsov-Ogievskiyc20555e2021-04-28 18:18:04 +03005 * Copyright (c) 2020 Virtuozzo International GmbH.
ths5fafdf22007-09-16 21:08:06 +00006 *
bellardfc01f7e2003-06-30 10:03:06 +00007 * Permission is hereby granted, free of charge, to any person obtaining a copy
8 * of this software and associated documentation files (the "Software"), to deal
9 * in the Software without restriction, including without limitation the rights
10 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
11 * copies of the Software, and to permit persons to whom the Software is
12 * furnished to do so, subject to the following conditions:
13 *
14 * The above copyright notice and this permission notice shall be included in
15 * all copies or substantial portions of the Software.
16 *
17 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
18 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
19 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
20 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
21 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
22 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
23 * THE SOFTWARE.
24 */
Markus Armbrustere688df62018-02-01 12:18:31 +010025
Peter Maydelld38ea872016-01-29 17:50:05 +000026#include "qemu/osdep.h"
Daniel P. Berrange0ab8ed12017-01-25 16:14:15 +000027#include "block/trace.h"
Paolo Bonzini737e1502012-12-17 18:19:44 +010028#include "block/block_int.h"
29#include "block/blockjob.h"
Markus Armbrustere2c1c342022-12-21 14:35:49 +010030#include "block/dirty-bitmap.h"
Max Reitz0c9b70d2020-10-27 20:05:42 +010031#include "block/fuse.h"
Kevin Wolfcd7fca92016-07-06 11:22:39 +020032#include "block/nbd.h"
Max Reitz609f45e2018-06-14 21:14:28 +020033#include "block/qdict.h"
Markus Armbrusterd49b6832015-03-17 18:29:20 +010034#include "qemu/error-report.h"
Marc-André Lureau5e5733e2019-08-29 22:34:43 +040035#include "block/module_block.h"
Markus Armbrusterdb725812019-08-12 07:23:50 +020036#include "qemu/main-loop.h"
Paolo Bonzini1de7afc2012-12-17 18:20:00 +010037#include "qemu/module.h"
Markus Armbrustere688df62018-02-01 12:18:31 +010038#include "qapi/error.h"
Markus Armbruster452fcdb2018-02-01 12:18:39 +010039#include "qapi/qmp/qdict.h"
Paolo Bonzini7b1b5d12012-12-17 18:19:43 +010040#include "qapi/qmp/qjson.h"
Max Reitze59a0cf2018-02-24 16:40:32 +010041#include "qapi/qmp/qnull.h"
Markus Armbrusterfc81fa12018-02-01 12:18:40 +010042#include "qapi/qmp/qstring.h"
Kevin Wolfe1d74bc2018-01-10 15:52:33 +010043#include "qapi/qobject-output-visitor.h"
44#include "qapi/qapi-visit-block-core.h"
Markus Armbrusterbfb197e2014-10-07 13:59:11 +020045#include "sysemu/block-backend.h"
Paolo Bonzini1de7afc2012-12-17 18:20:00 +010046#include "qemu/notify.h"
Markus Armbruster922a01a2018-02-01 12:18:46 +010047#include "qemu/option.h"
Daniel P. Berrange10817bf2015-09-01 14:48:02 +010048#include "qemu/coroutine.h"
Benoît Canetc13163f2014-01-23 21:31:34 +010049#include "block/qapi.h"
Paolo Bonzini1de7afc2012-12-17 18:20:00 +010050#include "qemu/timer.h"
Veronia Bahaaf348b6d2016-03-20 19:16:19 +020051#include "qemu/cutils.h"
52#include "qemu/id.h"
Hanna Reitz0bc329f2021-08-12 10:41:44 +020053#include "qemu/range.h"
54#include "qemu/rcu.h"
Vladimir Sementsov-Ogievskiy21c22832020-09-24 21:54:10 +030055#include "block/coroutines.h"
bellardfc01f7e2003-06-30 10:03:06 +000056
Juan Quintela71e72a12009-07-27 16:12:56 +020057#ifdef CONFIG_BSD
bellard7674e7b2005-04-26 21:59:26 +000058#include <sys/ioctl.h>
Blue Swirl72cf2d42009-09-12 07:36:22 +000059#include <sys/queue.h>
Joelle van Dynefeccdce2021-03-15 11:03:39 -070060#if defined(HAVE_SYS_DISK_H)
bellard7674e7b2005-04-26 21:59:26 +000061#include <sys/disk.h>
62#endif
blueswir1c5e97232009-03-07 20:06:23 +000063#endif
bellard7674e7b2005-04-26 21:59:26 +000064
aliguori49dc7682009-03-08 16:26:59 +000065#ifdef _WIN32
66#include <windows.h>
67#endif
68
Stefan Hajnoczi1c9805a2011-10-13 13:08:22 +010069#define NOT_DONE 0x7fffffff /* used while emulated sync operation in progress */
70
Emanuele Giuseppe Esposito3b491a92022-03-03 10:15:48 -050071/* Protected by BQL */
Benoît Canetdc364f42014-01-23 21:31:32 +010072static QTAILQ_HEAD(, BlockDriverState) graph_bdrv_states =
73 QTAILQ_HEAD_INITIALIZER(graph_bdrv_states);
74
Emanuele Giuseppe Esposito3b491a92022-03-03 10:15:48 -050075/* Protected by BQL */
Max Reitz2c1d04e2016-01-29 16:36:11 +010076static QTAILQ_HEAD(, BlockDriverState) all_bdrv_states =
77 QTAILQ_HEAD_INITIALIZER(all_bdrv_states);
78
Emanuele Giuseppe Esposito3b491a92022-03-03 10:15:48 -050079/* Protected by BQL */
Stefan Hajnoczi8a22f022010-04-13 10:29:33 +010080static QLIST_HEAD(, BlockDriver) bdrv_drivers =
81 QLIST_HEAD_INITIALIZER(bdrv_drivers);
bellardea2384d2004-08-01 21:59:26 +000082
Max Reitz5b363932016-05-17 16:41:31 +020083static BlockDriverState *bdrv_open_inherit(const char *filename,
84 const char *reference,
85 QDict *options, int flags,
86 BlockDriverState *parent,
Max Reitzbd86fb92020-05-13 13:05:13 +020087 const BdrvChildClass *child_class,
Max Reitz272c02e2020-05-13 13:05:17 +020088 BdrvChildRole child_role,
Max Reitz5b363932016-05-17 16:41:31 +020089 Error **errp);
Kevin Wolff3930ed2015-04-08 13:43:47 +020090
Kevin Wolfbfb8aa62021-10-18 15:47:14 +020091static bool bdrv_recurse_has_child(BlockDriverState *bs,
92 BlockDriverState *child);
93
Vladimir Sementsov-Ogievskiy544acc72022-07-26 23:11:31 +030094static void bdrv_replace_child_noperm(BdrvChild *child,
Vladimir Sementsov-Ogievskiy4eba8252022-07-26 23:11:28 +030095 BlockDriverState *new_bs);
Vladimir Sementsov-Ogievskiy57f08942022-07-26 23:11:34 +030096static void bdrv_remove_child(BdrvChild *child, Transaction *tran);
Vladimir Sementsov-Ogievskiy09786232021-04-28 18:17:44 +030097
Vladimir Sementsov-Ogievskiy72373e42021-04-28 18:17:58 +030098static int bdrv_reopen_prepare(BDRVReopenState *reopen_state,
99 BlockReopenQueue *queue,
Alberto Garciaecd30d22021-06-10 15:05:36 +0300100 Transaction *change_child_tran, Error **errp);
Vladimir Sementsov-Ogievskiy53e96d12021-04-28 18:17:35 +0300101static void bdrv_reopen_commit(BDRVReopenState *reopen_state);
102static void bdrv_reopen_abort(BDRVReopenState *reopen_state);
103
Emanuele Giuseppe Espositofa8fc1d2021-12-15 07:11:38 -0500104static bool bdrv_backing_overridden(BlockDriverState *bs);
105
Emanuele Giuseppe Esposito7e8c1822022-10-25 04:49:44 -0400106static bool bdrv_change_aio_context(BlockDriverState *bs, AioContext *ctx,
Emanuele Giuseppe Espositoe08cc002022-10-25 04:49:45 -0400107 GHashTable *visited, Transaction *tran,
Emanuele Giuseppe Esposito7e8c1822022-10-25 04:49:44 -0400108 Error **errp);
109
Markus Armbrustereb852012009-10-27 18:41:44 +0100110/* If non-zero, use only whitelisted block drivers */
111static int use_bdrv_whitelist;
112
Stefan Hajnoczi9e0b22f2010-12-09 11:53:00 +0000113#ifdef _WIN32
114static int is_windows_drive_prefix(const char *filename)
115{
116 return (((filename[0] >= 'a' && filename[0] <= 'z') ||
117 (filename[0] >= 'A' && filename[0] <= 'Z')) &&
118 filename[1] == ':');
119}
120
121int is_windows_drive(const char *filename)
122{
123 if (is_windows_drive_prefix(filename) &&
124 filename[2] == '\0')
125 return 1;
126 if (strstart(filename, "\\\\.\\", NULL) ||
127 strstart(filename, "//./", NULL))
128 return 1;
129 return 0;
130}
131#endif
132
Kevin Wolf339064d2013-11-28 10:23:32 +0100133size_t bdrv_opt_mem_align(BlockDriverState *bs)
134{
135 if (!bs || !bs->drv) {
Denis V. Lunev459b4e62015-05-12 17:30:56 +0300136 /* page size or 4k (hdd sector size) should be on the safe side */
Marc-André Lureau8e3b0cb2022-03-23 19:57:22 +0400137 return MAX(4096, qemu_real_host_page_size());
Kevin Wolf339064d2013-11-28 10:23:32 +0100138 }
Emanuele Giuseppe Esposito384a48f2022-03-03 10:15:50 -0500139 IO_CODE();
Kevin Wolf339064d2013-11-28 10:23:32 +0100140
141 return bs->bl.opt_mem_alignment;
142}
143
Denis V. Lunev4196d2f2015-05-12 17:30:55 +0300144size_t bdrv_min_mem_align(BlockDriverState *bs)
145{
146 if (!bs || !bs->drv) {
Denis V. Lunev459b4e62015-05-12 17:30:56 +0300147 /* page size or 4k (hdd sector size) should be on the safe side */
Marc-André Lureau8e3b0cb2022-03-23 19:57:22 +0400148 return MAX(4096, qemu_real_host_page_size());
Denis V. Lunev4196d2f2015-05-12 17:30:55 +0300149 }
Emanuele Giuseppe Esposito384a48f2022-03-03 10:15:50 -0500150 IO_CODE();
Denis V. Lunev4196d2f2015-05-12 17:30:55 +0300151
152 return bs->bl.min_mem_alignment;
153}
154
Stefan Hajnoczi9e0b22f2010-12-09 11:53:00 +0000155/* check if the path starts with "<protocol>:" */
Max Reitz5c984152014-12-03 14:57:22 +0100156int path_has_protocol(const char *path)
Stefan Hajnoczi9e0b22f2010-12-09 11:53:00 +0000157{
Paolo Bonzini947995c2012-05-08 16:51:48 +0200158 const char *p;
159
Stefan Hajnoczi9e0b22f2010-12-09 11:53:00 +0000160#ifdef _WIN32
161 if (is_windows_drive(path) ||
162 is_windows_drive_prefix(path)) {
163 return 0;
164 }
Paolo Bonzini947995c2012-05-08 16:51:48 +0200165 p = path + strcspn(path, ":/\\");
166#else
167 p = path + strcspn(path, ":/");
Stefan Hajnoczi9e0b22f2010-12-09 11:53:00 +0000168#endif
169
Paolo Bonzini947995c2012-05-08 16:51:48 +0200170 return *p == ':';
Stefan Hajnoczi9e0b22f2010-12-09 11:53:00 +0000171}
172
bellard83f64092006-08-01 16:21:11 +0000173int path_is_absolute(const char *path)
174{
bellard21664422007-01-07 18:22:37 +0000175#ifdef _WIN32
176 /* specific case for names like: "\\.\d:" */
Paolo Bonzinif53f4da2012-05-08 16:51:47 +0200177 if (is_windows_drive(path) || is_windows_drive_prefix(path)) {
bellard21664422007-01-07 18:22:37 +0000178 return 1;
Paolo Bonzinif53f4da2012-05-08 16:51:47 +0200179 }
180 return (*path == '/' || *path == '\\');
bellard3b9f94e2007-01-07 17:27:07 +0000181#else
Paolo Bonzinif53f4da2012-05-08 16:51:47 +0200182 return (*path == '/');
bellard3b9f94e2007-01-07 17:27:07 +0000183#endif
bellard83f64092006-08-01 16:21:11 +0000184}
185
Max Reitz009b03a2019-02-01 20:29:13 +0100186/* if filename is absolute, just return its duplicate. Otherwise, build a
bellard83f64092006-08-01 16:21:11 +0000187 path to it by considering it is relative to base_path. URL are
188 supported. */
Max Reitz009b03a2019-02-01 20:29:13 +0100189char *path_combine(const char *base_path, const char *filename)
bellard83f64092006-08-01 16:21:11 +0000190{
Max Reitz009b03a2019-02-01 20:29:13 +0100191 const char *protocol_stripped = NULL;
bellard83f64092006-08-01 16:21:11 +0000192 const char *p, *p1;
Max Reitz009b03a2019-02-01 20:29:13 +0100193 char *result;
bellard83f64092006-08-01 16:21:11 +0000194 int len;
195
bellard83f64092006-08-01 16:21:11 +0000196 if (path_is_absolute(filename)) {
Max Reitz009b03a2019-02-01 20:29:13 +0100197 return g_strdup(filename);
bellard83f64092006-08-01 16:21:11 +0000198 }
Max Reitz009b03a2019-02-01 20:29:13 +0100199
200 if (path_has_protocol(base_path)) {
201 protocol_stripped = strchr(base_path, ':');
202 if (protocol_stripped) {
203 protocol_stripped++;
204 }
205 }
206 p = protocol_stripped ?: base_path;
207
208 p1 = strrchr(base_path, '/');
209#ifdef _WIN32
210 {
211 const char *p2;
212 p2 = strrchr(base_path, '\\');
213 if (!p1 || p2 > p1) {
214 p1 = p2;
215 }
216 }
217#endif
218 if (p1) {
219 p1++;
220 } else {
221 p1 = base_path;
222 }
223 if (p1 > p) {
224 p = p1;
225 }
226 len = p - base_path;
227
228 result = g_malloc(len + strlen(filename) + 1);
229 memcpy(result, base_path, len);
230 strcpy(result + len, filename);
231
232 return result;
233}
234
Max Reitz03c320d2017-05-22 21:52:16 +0200235/*
236 * Helper function for bdrv_parse_filename() implementations to remove optional
237 * protocol prefixes (especially "file:") from a filename and for putting the
238 * stripped filename into the options QDict if there is such a prefix.
239 */
240void bdrv_parse_filename_strip_prefix(const char *filename, const char *prefix,
241 QDict *options)
242{
243 if (strstart(filename, prefix, &filename)) {
244 /* Stripping the explicit protocol prefix may result in a protocol
245 * prefix being (wrongly) detected (if the filename contains a colon) */
246 if (path_has_protocol(filename)) {
Markus Armbruster18cf67c2020-12-11 18:11:51 +0100247 GString *fat_filename;
Max Reitz03c320d2017-05-22 21:52:16 +0200248
249 /* This means there is some colon before the first slash; therefore,
250 * this cannot be an absolute path */
251 assert(!path_is_absolute(filename));
252
253 /* And we can thus fix the protocol detection issue by prefixing it
254 * by "./" */
Markus Armbruster18cf67c2020-12-11 18:11:51 +0100255 fat_filename = g_string_new("./");
256 g_string_append(fat_filename, filename);
Max Reitz03c320d2017-05-22 21:52:16 +0200257
Markus Armbruster18cf67c2020-12-11 18:11:51 +0100258 assert(!path_has_protocol(fat_filename->str));
Max Reitz03c320d2017-05-22 21:52:16 +0200259
Markus Armbruster18cf67c2020-12-11 18:11:51 +0100260 qdict_put(options, "filename",
261 qstring_from_gstring(fat_filename));
Max Reitz03c320d2017-05-22 21:52:16 +0200262 } else {
263 /* If no protocol prefix was detected, we can use the shortened
264 * filename as-is */
265 qdict_put_str(options, "filename", filename);
266 }
267 }
268}
269
270
Kevin Wolf9c5e6592017-05-04 18:52:40 +0200271/* Returns whether the image file is opened as read-only. Note that this can
272 * return false and writing to the image file is still not possible because the
273 * image is inactivated. */
Jeff Cody93ed5242017-04-07 16:55:28 -0400274bool bdrv_is_read_only(BlockDriverState *bs)
275{
Emanuele Giuseppe Esposito384a48f2022-03-03 10:15:50 -0500276 IO_CODE();
Vladimir Sementsov-Ogievskiy975da072021-05-27 18:40:55 +0300277 return !(bs->open_flags & BDRV_O_RDWR);
Jeff Cody93ed5242017-04-07 16:55:28 -0400278}
279
Kevin Wolf10e5d702023-02-03 16:21:40 +0100280static int bdrv_can_set_read_only(BlockDriverState *bs, bool read_only,
281 bool ignore_allow_rdw, Error **errp)
Jeff Codyfe5241b2017-04-07 16:55:25 -0400282{
Emanuele Giuseppe Esposito384a48f2022-03-03 10:15:50 -0500283 IO_CODE();
284
Jeff Codye2b82472017-04-07 16:55:26 -0400285 /* Do not set read_only if copy_on_read is enabled */
286 if (bs->copy_on_read && read_only) {
287 error_setg(errp, "Can't set node '%s' to r/o with copy-on-read enabled",
288 bdrv_get_device_or_node_name(bs));
289 return -EINVAL;
290 }
291
Jeff Codyd6fcdf02017-04-07 16:55:27 -0400292 /* Do not clear read_only if it is prohibited */
Kevin Wolf54a32bf2017-08-03 17:02:58 +0200293 if (!read_only && !(bs->open_flags & BDRV_O_ALLOW_RDWR) &&
294 !ignore_allow_rdw)
295 {
Jeff Codyd6fcdf02017-04-07 16:55:27 -0400296 error_setg(errp, "Node '%s' is read only",
297 bdrv_get_device_or_node_name(bs));
298 return -EPERM;
299 }
300
Jeff Cody45803a02017-04-07 16:55:29 -0400301 return 0;
302}
303
Kevin Wolfeaa24102018-10-12 11:27:41 +0200304/*
305 * Called by a driver that can only provide a read-only image.
306 *
307 * Returns 0 if the node is already read-only or it could switch the node to
308 * read-only because BDRV_O_AUTO_RDONLY is set.
309 *
310 * Returns -EACCES if the node is read-write and BDRV_O_AUTO_RDONLY is not set
311 * or bdrv_can_set_read_only() forbids making the node read-only. If @errmsg
312 * is not NULL, it is used as the error message for the Error object.
313 */
314int bdrv_apply_auto_read_only(BlockDriverState *bs, const char *errmsg,
315 Error **errp)
Jeff Cody45803a02017-04-07 16:55:29 -0400316{
317 int ret = 0;
Emanuele Giuseppe Esposito384a48f2022-03-03 10:15:50 -0500318 IO_CODE();
Jeff Cody45803a02017-04-07 16:55:29 -0400319
Kevin Wolfeaa24102018-10-12 11:27:41 +0200320 if (!(bs->open_flags & BDRV_O_RDWR)) {
321 return 0;
322 }
323 if (!(bs->open_flags & BDRV_O_AUTO_RDONLY)) {
324 goto fail;
325 }
326
327 ret = bdrv_can_set_read_only(bs, true, false, NULL);
Jeff Cody45803a02017-04-07 16:55:29 -0400328 if (ret < 0) {
Kevin Wolfeaa24102018-10-12 11:27:41 +0200329 goto fail;
Jeff Cody45803a02017-04-07 16:55:29 -0400330 }
331
Kevin Wolfeaa24102018-10-12 11:27:41 +0200332 bs->open_flags &= ~BDRV_O_RDWR;
Kevin Wolfeeae6a52018-10-09 16:57:12 +0200333
Jeff Codye2b82472017-04-07 16:55:26 -0400334 return 0;
Kevin Wolfeaa24102018-10-12 11:27:41 +0200335
336fail:
337 error_setg(errp, "%s", errmsg ?: "Image is read-only");
338 return -EACCES;
Jeff Codyfe5241b2017-04-07 16:55:25 -0400339}
340
Max Reitz645ae7d2019-02-01 20:29:14 +0100341/*
342 * If @backing is empty, this function returns NULL without setting
343 * @errp. In all other cases, NULL will only be returned with @errp
344 * set.
345 *
346 * Therefore, a return value of NULL without @errp set means that
347 * there is no backing file; if @errp is set, there is one but its
348 * absolute filename cannot be generated.
349 */
350char *bdrv_get_full_backing_filename_from_filename(const char *backed,
351 const char *backing,
352 Error **errp)
Max Reitz0a828552014-11-26 17:20:25 +0100353{
Max Reitz645ae7d2019-02-01 20:29:14 +0100354 if (backing[0] == '\0') {
355 return NULL;
356 } else if (path_has_protocol(backing) || path_is_absolute(backing)) {
357 return g_strdup(backing);
Max Reitz9f074292014-11-26 17:20:26 +0100358 } else if (backed[0] == '\0' || strstart(backed, "json:", NULL)) {
359 error_setg(errp, "Cannot use relative backing file names for '%s'",
360 backed);
Max Reitz645ae7d2019-02-01 20:29:14 +0100361 return NULL;
Max Reitz0a828552014-11-26 17:20:25 +0100362 } else {
Max Reitz645ae7d2019-02-01 20:29:14 +0100363 return path_combine(backed, backing);
Max Reitz0a828552014-11-26 17:20:25 +0100364 }
365}
366
Max Reitz9f4793d2019-02-01 20:29:16 +0100367/*
368 * If @filename is empty or NULL, this function returns NULL without
369 * setting @errp. In all other cases, NULL will only be returned with
370 * @errp set.
371 */
372static char *bdrv_make_absolute_filename(BlockDriverState *relative_to,
373 const char *filename, Error **errp)
374{
Max Reitz8df68612019-02-01 20:29:23 +0100375 char *dir, *full_name;
Max Reitz9f4793d2019-02-01 20:29:16 +0100376
Max Reitz8df68612019-02-01 20:29:23 +0100377 if (!filename || filename[0] == '\0') {
378 return NULL;
379 } else if (path_has_protocol(filename) || path_is_absolute(filename)) {
380 return g_strdup(filename);
381 }
Max Reitz9f4793d2019-02-01 20:29:16 +0100382
Max Reitz8df68612019-02-01 20:29:23 +0100383 dir = bdrv_dirname(relative_to, errp);
384 if (!dir) {
385 return NULL;
386 }
Max Reitz9f4793d2019-02-01 20:29:16 +0100387
Max Reitz8df68612019-02-01 20:29:23 +0100388 full_name = g_strconcat(dir, filename, NULL);
389 g_free(dir);
390 return full_name;
Max Reitz9f4793d2019-02-01 20:29:16 +0100391}
392
Max Reitz6b6833c2019-02-01 20:29:15 +0100393char *bdrv_get_full_backing_filename(BlockDriverState *bs, Error **errp)
Paolo Bonzinidc5a1372012-05-08 16:51:50 +0200394{
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -0500395 GLOBAL_STATE_CODE();
Max Reitz9f4793d2019-02-01 20:29:16 +0100396 return bdrv_make_absolute_filename(bs, bs->backing_file, errp);
Paolo Bonzinidc5a1372012-05-08 16:51:50 +0200397}
398
Stefan Hajnoczi0eb72172015-04-28 14:27:51 +0100399void bdrv_register(BlockDriver *bdrv)
400{
Philippe Mathieu-Daudéa15f08d2020-03-18 23:22:35 +0100401 assert(bdrv->format_name);
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -0500402 GLOBAL_STATE_CODE();
Stefan Hajnoczi8a22f022010-04-13 10:29:33 +0100403 QLIST_INSERT_HEAD(&bdrv_drivers, bdrv, list);
bellardea2384d2004-08-01 21:59:26 +0000404}
bellardb3380822004-03-14 21:38:54 +0000405
Markus Armbrustere4e99862014-10-07 13:59:03 +0200406BlockDriverState *bdrv_new(void)
407{
408 BlockDriverState *bs;
409 int i;
410
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -0500411 GLOBAL_STATE_CODE();
412
Markus Armbruster5839e532014-08-19 10:31:08 +0200413 bs = g_new0(BlockDriverState, 1);
Fam Zhenge4654d22013-11-13 18:29:43 +0800414 QLIST_INIT(&bs->dirty_bitmaps);
Fam Zhengfbe40ff2014-05-23 21:29:42 +0800415 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
416 QLIST_INIT(&bs->op_blockers[i]);
417 }
Paolo Bonzini3783fa32017-06-05 14:39:02 +0200418 qemu_co_mutex_init(&bs->reqs_lock);
Paolo Bonzini21198822017-06-05 14:39:03 +0200419 qemu_mutex_init(&bs->dirty_bitmap_mutex);
Fam Zheng9fcb0252013-08-23 09:14:46 +0800420 bs->refcnt = 1;
Stefan Hajnoczidcd04222014-05-08 16:34:37 +0200421 bs->aio_context = qemu_get_aio_context();
Paolo Bonzinid7d512f2012-08-23 11:20:36 +0200422
Evgeny Yakovlev3ff2f672016-07-18 22:39:52 +0300423 qemu_co_queue_init(&bs->flush_queue);
424
Hanna Reitz0bc329f2021-08-12 10:41:44 +0200425 qemu_co_mutex_init(&bs->bsc_modify_lock);
426 bs->block_status_cache = g_new0(BdrvBlockStatusCache, 1);
427
Kevin Wolf0f122642018-03-28 18:29:18 +0200428 for (i = 0; i < bdrv_drain_all_count; i++) {
429 bdrv_drained_begin(bs);
430 }
431
Max Reitz2c1d04e2016-01-29 16:36:11 +0100432 QTAILQ_INSERT_TAIL(&all_bdrv_states, bs, bs_list);
433
bellardb3380822004-03-14 21:38:54 +0000434 return bs;
435}
436
Marc Mari88d88792016-08-12 09:27:03 -0400437static BlockDriver *bdrv_do_find_format(const char *format_name)
bellardea2384d2004-08-01 21:59:26 +0000438{
439 BlockDriver *drv1;
Emanuele Giuseppe Espositobdb73472022-03-03 10:16:02 -0500440 GLOBAL_STATE_CODE();
Marc Mari88d88792016-08-12 09:27:03 -0400441
Stefan Hajnoczi8a22f022010-04-13 10:29:33 +0100442 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
443 if (!strcmp(drv1->format_name, format_name)) {
bellardea2384d2004-08-01 21:59:26 +0000444 return drv1;
Stefan Hajnoczi8a22f022010-04-13 10:29:33 +0100445 }
bellardea2384d2004-08-01 21:59:26 +0000446 }
Marc Mari88d88792016-08-12 09:27:03 -0400447
bellardea2384d2004-08-01 21:59:26 +0000448 return NULL;
449}
450
Marc Mari88d88792016-08-12 09:27:03 -0400451BlockDriver *bdrv_find_format(const char *format_name)
452{
453 BlockDriver *drv1;
454 int i;
455
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -0500456 GLOBAL_STATE_CODE();
457
Marc Mari88d88792016-08-12 09:27:03 -0400458 drv1 = bdrv_do_find_format(format_name);
459 if (drv1) {
460 return drv1;
461 }
462
463 /* The driver isn't registered, maybe we need to load a module */
464 for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); ++i) {
465 if (!strcmp(block_driver_modules[i].format_name, format_name)) {
Claudio Fontanac551fb02022-09-29 11:30:33 +0200466 Error *local_err = NULL;
467 int rv = block_module_load(block_driver_modules[i].library_name,
468 &local_err);
469 if (rv > 0) {
470 return bdrv_do_find_format(format_name);
471 } else if (rv < 0) {
472 error_report_err(local_err);
473 }
Marc Mari88d88792016-08-12 09:27:03 -0400474 break;
475 }
476 }
Claudio Fontanac551fb02022-09-29 11:30:33 +0200477 return NULL;
Marc Mari88d88792016-08-12 09:27:03 -0400478}
479
Andrey Shinkevich9ac404c2019-03-07 16:33:58 +0300480static int bdrv_format_is_whitelisted(const char *format_name, bool read_only)
Markus Armbrustereb852012009-10-27 18:41:44 +0100481{
Fam Zhengb64ec4e2013-05-29 19:35:40 +0800482 static const char *whitelist_rw[] = {
483 CONFIG_BDRV_RW_WHITELIST
Paolo Bonzini859aef02020-08-04 18:14:26 +0200484 NULL
Fam Zhengb64ec4e2013-05-29 19:35:40 +0800485 };
486 static const char *whitelist_ro[] = {
487 CONFIG_BDRV_RO_WHITELIST
Paolo Bonzini859aef02020-08-04 18:14:26 +0200488 NULL
Markus Armbrustereb852012009-10-27 18:41:44 +0100489 };
490 const char **p;
491
Fam Zhengb64ec4e2013-05-29 19:35:40 +0800492 if (!whitelist_rw[0] && !whitelist_ro[0]) {
Markus Armbrustereb852012009-10-27 18:41:44 +0100493 return 1; /* no whitelist, anything goes */
Fam Zhengb64ec4e2013-05-29 19:35:40 +0800494 }
Markus Armbrustereb852012009-10-27 18:41:44 +0100495
Fam Zhengb64ec4e2013-05-29 19:35:40 +0800496 for (p = whitelist_rw; *p; p++) {
Andrey Shinkevich9ac404c2019-03-07 16:33:58 +0300497 if (!strcmp(format_name, *p)) {
Markus Armbrustereb852012009-10-27 18:41:44 +0100498 return 1;
499 }
500 }
Fam Zhengb64ec4e2013-05-29 19:35:40 +0800501 if (read_only) {
502 for (p = whitelist_ro; *p; p++) {
Andrey Shinkevich9ac404c2019-03-07 16:33:58 +0300503 if (!strcmp(format_name, *p)) {
Fam Zhengb64ec4e2013-05-29 19:35:40 +0800504 return 1;
505 }
506 }
507 }
Markus Armbrustereb852012009-10-27 18:41:44 +0100508 return 0;
509}
510
Andrey Shinkevich9ac404c2019-03-07 16:33:58 +0300511int bdrv_is_whitelisted(BlockDriver *drv, bool read_only)
512{
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -0500513 GLOBAL_STATE_CODE();
Andrey Shinkevich9ac404c2019-03-07 16:33:58 +0300514 return bdrv_format_is_whitelisted(drv->format_name, read_only);
515}
516
Daniel P. Berrangee6ff69b2016-03-21 14:11:48 +0000517bool bdrv_uses_whitelist(void)
518{
519 return use_bdrv_whitelist;
520}
521
Zhi Yong Wu5b7e1542012-05-07 16:50:42 +0800522typedef struct CreateCo {
523 BlockDriver *drv;
524 char *filename;
Chunyan Liu83d05212014-06-05 17:20:51 +0800525 QemuOpts *opts;
Zhi Yong Wu5b7e1542012-05-07 16:50:42 +0800526 int ret;
Max Reitzcc84d902013-09-06 17:14:26 +0200527 Error *err;
Zhi Yong Wu5b7e1542012-05-07 16:50:42 +0800528} CreateCo;
529
Emanuele Giuseppe Esposito741443e2022-11-28 09:23:36 -0500530int coroutine_fn bdrv_co_create(BlockDriver *drv, const char *filename,
531 QemuOpts *opts, Error **errp)
Emanuele Giuseppe Esposito84bdf212022-11-28 09:23:30 -0500532{
533 int ret;
534 GLOBAL_STATE_CODE();
535 ERRP_GUARD();
536
537 if (!drv->bdrv_co_create_opts) {
538 error_setg(errp, "Driver '%s' does not support image creation",
539 drv->format_name);
540 return -ENOTSUP;
541 }
542
543 ret = drv->bdrv_co_create_opts(drv, filename, opts, errp);
544 if (ret < 0 && !*errp) {
545 error_setg_errno(errp, -ret, "Could not create image");
546 }
547
548 return ret;
549}
550
Max Reitzfd171462020-01-22 17:45:29 +0100551/**
552 * Helper function for bdrv_create_file_fallback(): Resize @blk to at
553 * least the given @minimum_size.
554 *
555 * On success, return @blk's actual length.
556 * Otherwise, return -errno.
557 */
Paolo Bonzini84569a72023-06-01 13:51:38 +0200558static int64_t coroutine_fn GRAPH_UNLOCKED
559create_file_fallback_truncate(BlockBackend *blk, int64_t minimum_size,
560 Error **errp)
Max Reitzfd171462020-01-22 17:45:29 +0100561{
562 Error *local_err = NULL;
563 int64_t size;
564 int ret;
565
Emanuele Giuseppe Espositobdb73472022-03-03 10:16:02 -0500566 GLOBAL_STATE_CODE();
567
Paolo Bonzini84569a72023-06-01 13:51:38 +0200568 ret = blk_co_truncate(blk, minimum_size, false, PREALLOC_MODE_OFF, 0,
569 &local_err);
Max Reitzfd171462020-01-22 17:45:29 +0100570 if (ret < 0 && ret != -ENOTSUP) {
571 error_propagate(errp, local_err);
572 return ret;
573 }
574
Paolo Bonzini84569a72023-06-01 13:51:38 +0200575 size = blk_co_getlength(blk);
Max Reitzfd171462020-01-22 17:45:29 +0100576 if (size < 0) {
577 error_free(local_err);
578 error_setg_errno(errp, -size,
579 "Failed to inquire the new image file's length");
580 return size;
581 }
582
583 if (size < minimum_size) {
584 /* Need to grow the image, but we failed to do that */
585 error_propagate(errp, local_err);
586 return -ENOTSUP;
587 }
588
589 error_free(local_err);
590 local_err = NULL;
591
592 return size;
593}
594
595/**
596 * Helper function for bdrv_create_file_fallback(): Zero the first
597 * sector to remove any potentially pre-existing image header.
598 */
Paolo Bonzini881a4c52022-09-22 10:49:00 +0200599static int coroutine_fn
600create_file_fallback_zero_first_sector(BlockBackend *blk,
601 int64_t current_size,
602 Error **errp)
Max Reitzfd171462020-01-22 17:45:29 +0100603{
604 int64_t bytes_to_clear;
605 int ret;
606
Emanuele Giuseppe Espositobdb73472022-03-03 10:16:02 -0500607 GLOBAL_STATE_CODE();
608
Max Reitzfd171462020-01-22 17:45:29 +0100609 bytes_to_clear = MIN(current_size, BDRV_SECTOR_SIZE);
610 if (bytes_to_clear) {
Alberto Fariace47ff22022-10-13 14:37:02 +0200611 ret = blk_co_pwrite_zeroes(blk, 0, bytes_to_clear, BDRV_REQ_MAY_UNMAP);
Max Reitzfd171462020-01-22 17:45:29 +0100612 if (ret < 0) {
613 error_setg_errno(errp, -ret,
614 "Failed to clear the new image's first sector");
615 return ret;
616 }
617 }
618
619 return 0;
620}
621
Maxim Levitsky5a5e7f82020-03-26 03:12:18 +0200622/**
623 * Simple implementation of bdrv_co_create_opts for protocol drivers
624 * which only support creation via opening a file
625 * (usually existing raw storage device)
626 */
627int coroutine_fn bdrv_co_create_opts_simple(BlockDriver *drv,
628 const char *filename,
629 QemuOpts *opts,
630 Error **errp)
Max Reitzfd171462020-01-22 17:45:29 +0100631{
632 BlockBackend *blk;
Max Reitzeeea1fa2020-02-25 16:56:18 +0100633 QDict *options;
Max Reitzfd171462020-01-22 17:45:29 +0100634 int64_t size = 0;
635 char *buf = NULL;
636 PreallocMode prealloc;
637 Error *local_err = NULL;
638 int ret;
639
Emanuele Giuseppe Espositob4ad82a2022-03-03 10:15:57 -0500640 GLOBAL_STATE_CODE();
641
Max Reitzfd171462020-01-22 17:45:29 +0100642 size = qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0);
643 buf = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC);
644 prealloc = qapi_enum_parse(&PreallocMode_lookup, buf,
645 PREALLOC_MODE_OFF, &local_err);
646 g_free(buf);
647 if (local_err) {
648 error_propagate(errp, local_err);
649 return -EINVAL;
650 }
651
652 if (prealloc != PREALLOC_MODE_OFF) {
653 error_setg(errp, "Unsupported preallocation mode '%s'",
654 PreallocMode_str(prealloc));
655 return -ENOTSUP;
656 }
657
Max Reitzeeea1fa2020-02-25 16:56:18 +0100658 options = qdict_new();
Max Reitzfd171462020-01-22 17:45:29 +0100659 qdict_put_str(options, "driver", drv->format_name);
660
Kevin Wolfbe1a7322023-01-26 18:24:31 +0100661 blk = blk_co_new_open(filename, NULL, options,
662 BDRV_O_RDWR | BDRV_O_RESIZE, errp);
Max Reitzfd171462020-01-22 17:45:29 +0100663 if (!blk) {
664 error_prepend(errp, "Protocol driver '%s' does not support image "
665 "creation, and opening the image failed: ",
666 drv->format_name);
667 return -EINVAL;
668 }
669
670 size = create_file_fallback_truncate(blk, size, errp);
671 if (size < 0) {
672 ret = size;
673 goto out;
674 }
675
676 ret = create_file_fallback_zero_first_sector(blk, size, errp);
677 if (ret < 0) {
678 goto out;
679 }
680
681 ret = 0;
682out:
Kevin Wolfb2ab5f52023-05-04 13:57:33 +0200683 blk_co_unref(blk);
Max Reitzfd171462020-01-22 17:45:29 +0100684 return ret;
685}
686
Emanuele Giuseppe Esposito2475a0d2022-11-28 09:23:31 -0500687int coroutine_fn bdrv_co_create_file(const char *filename, QemuOpts *opts,
688 Error **errp)
Christoph Hellwig84a12e62010-04-07 22:30:24 +0200689{
Stefano Garzarella729222a2021-03-08 17:12:32 +0100690 QemuOpts *protocol_opts;
Christoph Hellwig84a12e62010-04-07 22:30:24 +0200691 BlockDriver *drv;
Stefano Garzarella729222a2021-03-08 17:12:32 +0100692 QDict *qdict;
693 int ret;
Christoph Hellwig84a12e62010-04-07 22:30:24 +0200694
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -0500695 GLOBAL_STATE_CODE();
696
Max Reitzb65a5e12015-02-05 13:58:12 -0500697 drv = bdrv_find_protocol(filename, true, errp);
Christoph Hellwig84a12e62010-04-07 22:30:24 +0200698 if (drv == NULL) {
Stefan Hajnoczi16905d72010-11-30 15:14:14 +0000699 return -ENOENT;
Christoph Hellwig84a12e62010-04-07 22:30:24 +0200700 }
701
Stefano Garzarella729222a2021-03-08 17:12:32 +0100702 if (!drv->create_opts) {
703 error_setg(errp, "Driver '%s' does not support image creation",
704 drv->format_name);
705 return -ENOTSUP;
706 }
707
708 /*
709 * 'opts' contains a QemuOptsList with a combination of format and protocol
710 * default values.
711 *
712 * The format properly removes its options, but the default values remain
713 * in 'opts->list'. So if the protocol has options with the same name
714 * (e.g. rbd has 'cluster_size' as qcow2), it will see the default values
715 * of the format, since for overlapping options, the format wins.
716 *
717 * To avoid this issue, lets convert QemuOpts to QDict, in this way we take
718 * only the set options, and then convert it back to QemuOpts, using the
719 * create_opts of the protocol. So the new QemuOpts, will contain only the
720 * protocol defaults.
721 */
722 qdict = qemu_opts_to_qdict(opts, NULL);
723 protocol_opts = qemu_opts_from_qdict(drv->create_opts, qdict, errp);
724 if (protocol_opts == NULL) {
725 ret = -EINVAL;
726 goto out;
727 }
728
Emanuele Giuseppe Esposito2475a0d2022-11-28 09:23:31 -0500729 ret = bdrv_co_create(drv, filename, protocol_opts, errp);
Stefano Garzarella729222a2021-03-08 17:12:32 +0100730out:
731 qemu_opts_del(protocol_opts);
732 qobject_unref(qdict);
733 return ret;
Christoph Hellwig84a12e62010-04-07 22:30:24 +0200734}
735
Daniel Henrique Barbozae1d7f8b2020-01-30 18:39:05 -0300736int coroutine_fn bdrv_co_delete_file(BlockDriverState *bs, Error **errp)
737{
738 Error *local_err = NULL;
739 int ret;
740
Emanuele Giuseppe Esposito384a48f2022-03-03 10:15:50 -0500741 IO_CODE();
Daniel Henrique Barbozae1d7f8b2020-01-30 18:39:05 -0300742 assert(bs != NULL);
Kevin Wolf48aef792023-02-03 16:22:00 +0100743 assert_bdrv_graph_readable();
Daniel Henrique Barbozae1d7f8b2020-01-30 18:39:05 -0300744
745 if (!bs->drv) {
746 error_setg(errp, "Block node '%s' is not opened", bs->filename);
747 return -ENOMEDIUM;
748 }
749
750 if (!bs->drv->bdrv_co_delete_file) {
751 error_setg(errp, "Driver '%s' does not support image deletion",
752 bs->drv->format_name);
753 return -ENOTSUP;
754 }
755
756 ret = bs->drv->bdrv_co_delete_file(bs, &local_err);
757 if (ret < 0) {
758 error_propagate(errp, local_err);
759 }
760
761 return ret;
762}
763
Maxim Levitskya890f082020-12-17 19:09:03 +0200764void coroutine_fn bdrv_co_delete_file_noerr(BlockDriverState *bs)
765{
766 Error *local_err = NULL;
767 int ret;
Emanuele Giuseppe Esposito384a48f2022-03-03 10:15:50 -0500768 IO_CODE();
Maxim Levitskya890f082020-12-17 19:09:03 +0200769
770 if (!bs) {
771 return;
772 }
773
774 ret = bdrv_co_delete_file(bs, &local_err);
775 /*
776 * ENOTSUP will happen if the block driver doesn't support
777 * the 'bdrv_co_delete_file' interface. This is a predictable
778 * scenario and shouldn't be reported back to the user.
779 */
780 if (ret == -ENOTSUP) {
781 error_free(local_err);
782 } else if (ret < 0) {
783 error_report_err(local_err);
784 }
785}
786
Ekaterina Tumanova892b7de2015-02-16 12:47:54 +0100787/**
788 * Try to get @bs's logical and physical block size.
789 * On success, store them in @bsz struct and return 0.
790 * On failure return -errno.
791 * @bs must not be empty.
792 */
793int bdrv_probe_blocksizes(BlockDriverState *bs, BlockSizes *bsz)
794{
795 BlockDriver *drv = bs->drv;
Max Reitz93393e62019-06-12 17:03:38 +0200796 BlockDriverState *filtered = bdrv_filter_bs(bs);
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -0500797 GLOBAL_STATE_CODE();
Ekaterina Tumanova892b7de2015-02-16 12:47:54 +0100798
799 if (drv && drv->bdrv_probe_blocksizes) {
800 return drv->bdrv_probe_blocksizes(bs, bsz);
Max Reitz93393e62019-06-12 17:03:38 +0200801 } else if (filtered) {
802 return bdrv_probe_blocksizes(filtered, bsz);
Ekaterina Tumanova892b7de2015-02-16 12:47:54 +0100803 }
804
805 return -ENOTSUP;
806}
807
808/**
809 * Try to get @bs's geometry (cyls, heads, sectors).
810 * On success, store them in @geo struct and return 0.
811 * On failure return -errno.
812 * @bs must not be empty.
813 */
814int bdrv_probe_geometry(BlockDriverState *bs, HDGeometry *geo)
815{
816 BlockDriver *drv = bs->drv;
Max Reitz93393e62019-06-12 17:03:38 +0200817 BlockDriverState *filtered = bdrv_filter_bs(bs);
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -0500818 GLOBAL_STATE_CODE();
Ekaterina Tumanova892b7de2015-02-16 12:47:54 +0100819
820 if (drv && drv->bdrv_probe_geometry) {
821 return drv->bdrv_probe_geometry(bs, geo);
Max Reitz93393e62019-06-12 17:03:38 +0200822 } else if (filtered) {
823 return bdrv_probe_geometry(filtered, geo);
Ekaterina Tumanova892b7de2015-02-16 12:47:54 +0100824 }
825
826 return -ENOTSUP;
827}
828
Jim Meyeringeba25052012-05-28 09:27:54 +0200829/*
830 * Create a uniquely-named empty temporary file.
Bin Meng69fbfff2022-10-10 12:04:31 +0800831 * Return the actual file name used upon success, otherwise NULL.
832 * This string should be freed with g_free() when not needed any longer.
833 *
834 * Note: creating a temporary file for the caller to (re)open is
835 * inherently racy. Use g_file_open_tmp() instead whenever practical.
Jim Meyeringeba25052012-05-28 09:27:54 +0200836 */
Bin Meng69fbfff2022-10-10 12:04:31 +0800837char *create_tmp_file(Error **errp)
Jim Meyeringeba25052012-05-28 09:27:54 +0200838{
bellardea2384d2004-08-01 21:59:26 +0000839 int fd;
blueswir17ccfb2e2008-09-14 06:45:34 +0000840 const char *tmpdir;
Bin Meng69fbfff2022-10-10 12:04:31 +0800841 g_autofree char *filename = NULL;
842
843 tmpdir = g_get_tmp_dir();
844#ifndef _WIN32
845 /*
846 * See commit 69bef79 ("block: use /var/tmp instead of /tmp for -snapshot")
847 *
848 * This function is used to create temporary disk images (like -snapshot),
849 * so the files can become very large. /tmp is often a tmpfs where as
850 * /var/tmp is usually on a disk, so more appropriate for disk images.
851 */
852 if (!g_strcmp0(tmpdir, "/tmp")) {
Amit Shah69bef792014-02-26 15:12:37 +0530853 tmpdir = "/var/tmp";
854 }
bellardd5249392004-08-03 21:14:23 +0000855#endif
Bin Meng69fbfff2022-10-10 12:04:31 +0800856
857 filename = g_strdup_printf("%s/vl.XXXXXX", tmpdir);
858 fd = g_mkstemp(filename);
bellardea2384d2004-08-01 21:59:26 +0000859 if (fd < 0) {
Bin Meng69fbfff2022-10-10 12:04:31 +0800860 error_setg_errno(errp, errno, "Could not open temporary file '%s'",
861 filename);
862 return NULL;
bellardea2384d2004-08-01 21:59:26 +0000863 }
Bin Meng6b6471e2022-10-10 12:04:30 +0800864 close(fd);
Bin Meng69fbfff2022-10-10 12:04:31 +0800865
866 return g_steal_pointer(&filename);
Jim Meyeringeba25052012-05-28 09:27:54 +0200867}
bellardea2384d2004-08-01 21:59:26 +0000868
Christoph Hellwigf3a5d3f2009-06-15 13:55:19 +0200869/*
870 * Detect host devices. By convention, /dev/cdrom[N] is always
871 * recognized as a host CDROM.
872 */
Christoph Hellwigf3a5d3f2009-06-15 13:55:19 +0200873static BlockDriver *find_hdev_driver(const char *filename)
874{
Christoph Hellwig508c7cb2009-06-15 14:04:22 +0200875 int score_max = 0, score;
876 BlockDriver *drv = NULL, *d;
Emanuele Giuseppe Espositobdb73472022-03-03 10:16:02 -0500877 GLOBAL_STATE_CODE();
Christoph Hellwigf3a5d3f2009-06-15 13:55:19 +0200878
Stefan Hajnoczi8a22f022010-04-13 10:29:33 +0100879 QLIST_FOREACH(d, &bdrv_drivers, list) {
Christoph Hellwig508c7cb2009-06-15 14:04:22 +0200880 if (d->bdrv_probe_device) {
881 score = d->bdrv_probe_device(filename);
882 if (score > score_max) {
883 score_max = score;
884 drv = d;
885 }
886 }
Christoph Hellwigf3a5d3f2009-06-15 13:55:19 +0200887 }
888
Christoph Hellwig508c7cb2009-06-15 14:04:22 +0200889 return drv;
Christoph Hellwigf3a5d3f2009-06-15 13:55:19 +0200890}
Christoph Hellwigf3a5d3f2009-06-15 13:55:19 +0200891
Marc Mari88d88792016-08-12 09:27:03 -0400892static BlockDriver *bdrv_do_find_protocol(const char *protocol)
893{
894 BlockDriver *drv1;
Emanuele Giuseppe Espositobdb73472022-03-03 10:16:02 -0500895 GLOBAL_STATE_CODE();
Marc Mari88d88792016-08-12 09:27:03 -0400896
897 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
898 if (drv1->protocol_name && !strcmp(drv1->protocol_name, protocol)) {
899 return drv1;
900 }
901 }
902
903 return NULL;
904}
905
Kevin Wolf98289622013-07-10 15:47:39 +0200906BlockDriver *bdrv_find_protocol(const char *filename,
Max Reitzb65a5e12015-02-05 13:58:12 -0500907 bool allow_protocol_prefix,
908 Error **errp)
Christoph Hellwig84a12e62010-04-07 22:30:24 +0200909{
910 BlockDriver *drv1;
911 char protocol[128];
912 int len;
913 const char *p;
Marc Mari88d88792016-08-12 09:27:03 -0400914 int i;
Christoph Hellwig84a12e62010-04-07 22:30:24 +0200915
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -0500916 GLOBAL_STATE_CODE();
Kevin Wolf66f82ce2010-04-14 14:17:38 +0200917 /* TODO Drivers without bdrv_file_open must be specified explicitly */
918
Christoph Hellwig39508e72010-06-23 12:25:17 +0200919 /*
920 * XXX(hch): we really should not let host device detection
921 * override an explicit protocol specification, but moving this
922 * later breaks access to device names with colons in them.
923 * Thanks to the brain-dead persistent naming schemes on udev-
924 * based Linux systems those actually are quite common.
925 */
926 drv1 = find_hdev_driver(filename);
927 if (drv1) {
Christoph Hellwig84a12e62010-04-07 22:30:24 +0200928 return drv1;
929 }
Christoph Hellwig39508e72010-06-23 12:25:17 +0200930
Kevin Wolf98289622013-07-10 15:47:39 +0200931 if (!path_has_protocol(filename) || !allow_protocol_prefix) {
Max Reitzef810432014-12-02 18:32:42 +0100932 return &bdrv_file;
Christoph Hellwig39508e72010-06-23 12:25:17 +0200933 }
Kevin Wolf98289622013-07-10 15:47:39 +0200934
Stefan Hajnoczi9e0b22f2010-12-09 11:53:00 +0000935 p = strchr(filename, ':');
936 assert(p != NULL);
Christoph Hellwig84a12e62010-04-07 22:30:24 +0200937 len = p - filename;
938 if (len > sizeof(protocol) - 1)
939 len = sizeof(protocol) - 1;
940 memcpy(protocol, filename, len);
941 protocol[len] = '\0';
Marc Mari88d88792016-08-12 09:27:03 -0400942
943 drv1 = bdrv_do_find_protocol(protocol);
944 if (drv1) {
945 return drv1;
946 }
947
948 for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); ++i) {
949 if (block_driver_modules[i].protocol_name &&
950 !strcmp(block_driver_modules[i].protocol_name, protocol)) {
Claudio Fontanac551fb02022-09-29 11:30:33 +0200951 int rv = block_module_load(block_driver_modules[i].library_name, errp);
952 if (rv > 0) {
953 drv1 = bdrv_do_find_protocol(protocol);
954 } else if (rv < 0) {
955 return NULL;
956 }
Marc Mari88d88792016-08-12 09:27:03 -0400957 break;
Christoph Hellwig84a12e62010-04-07 22:30:24 +0200958 }
959 }
Max Reitzb65a5e12015-02-05 13:58:12 -0500960
Marc Mari88d88792016-08-12 09:27:03 -0400961 if (!drv1) {
962 error_setg(errp, "Unknown protocol '%s'", protocol);
963 }
964 return drv1;
Christoph Hellwig84a12e62010-04-07 22:30:24 +0200965}
966
Markus Armbrusterc6684242014-11-20 16:27:10 +0100967/*
968 * Guess image format by probing its contents.
969 * This is not a good idea when your image is raw (CVE-2008-2004), but
970 * we do it anyway for backward compatibility.
971 *
972 * @buf contains the image's first @buf_size bytes.
Kevin Wolf7cddd372014-11-20 16:27:11 +0100973 * @buf_size is the buffer size in bytes (generally BLOCK_PROBE_BUF_SIZE,
974 * but can be smaller if the image file is smaller)
Markus Armbrusterc6684242014-11-20 16:27:10 +0100975 * @filename is its filename.
976 *
977 * For all block drivers, call the bdrv_probe() method to get its
978 * probing score.
979 * Return the first block driver with the highest probing score.
980 */
Kevin Wolf38f3ef52014-11-20 16:27:12 +0100981BlockDriver *bdrv_probe_all(const uint8_t *buf, int buf_size,
982 const char *filename)
Markus Armbrusterc6684242014-11-20 16:27:10 +0100983{
984 int score_max = 0, score;
985 BlockDriver *drv = NULL, *d;
Emanuele Giuseppe Esposito967d7902022-03-03 10:15:58 -0500986 IO_CODE();
Markus Armbrusterc6684242014-11-20 16:27:10 +0100987
988 QLIST_FOREACH(d, &bdrv_drivers, list) {
989 if (d->bdrv_probe) {
990 score = d->bdrv_probe(buf, buf_size, filename);
991 if (score > score_max) {
992 score_max = score;
993 drv = d;
994 }
995 }
996 }
997
998 return drv;
999}
1000
Kevin Wolf5696c6e2017-02-17 18:39:24 +01001001static int find_image_format(BlockBackend *file, const char *filename,
Max Reitz34b5d2c2013-09-05 14:45:29 +02001002 BlockDriver **pdrv, Error **errp)
bellardea2384d2004-08-01 21:59:26 +00001003{
Markus Armbrusterc6684242014-11-20 16:27:10 +01001004 BlockDriver *drv;
Kevin Wolf7cddd372014-11-20 16:27:11 +01001005 uint8_t buf[BLOCK_PROBE_BUF_SIZE];
Kevin Wolff500a6d2012-11-12 17:35:27 +01001006 int ret = 0;
Nicholas Bellingerf8ea0b02010-05-17 09:45:57 -07001007
Emanuele Giuseppe Espositobdb73472022-03-03 10:16:02 -05001008 GLOBAL_STATE_CODE();
1009
Kevin Wolf08a00552010-06-01 18:37:31 +02001010 /* Return the raw BlockDriver * to scsi-generic devices or empty drives */
Kevin Wolf5696c6e2017-02-17 18:39:24 +01001011 if (blk_is_sg(file) || !blk_is_inserted(file) || blk_getlength(file) == 0) {
Max Reitzef810432014-12-02 18:32:42 +01001012 *pdrv = &bdrv_raw;
Stefan Weilc98ac352010-07-21 21:51:51 +02001013 return ret;
Nicholas A. Bellinger1a396852010-05-27 08:56:28 -07001014 }
Nicholas Bellingerf8ea0b02010-05-17 09:45:57 -07001015
Alberto Fariaa9262f52022-07-05 17:15:11 +01001016 ret = blk_pread(file, 0, sizeof(buf), buf, 0);
bellard83f64092006-08-01 16:21:11 +00001017 if (ret < 0) {
Max Reitz34b5d2c2013-09-05 14:45:29 +02001018 error_setg_errno(errp, -ret, "Could not read image for determining its "
1019 "format");
Stefan Weilc98ac352010-07-21 21:51:51 +02001020 *pdrv = NULL;
1021 return ret;
bellard83f64092006-08-01 16:21:11 +00001022 }
1023
Alberto Fariabf5b16f2022-07-05 17:15:09 +01001024 drv = bdrv_probe_all(buf, sizeof(buf), filename);
Stefan Weilc98ac352010-07-21 21:51:51 +02001025 if (!drv) {
Max Reitz34b5d2c2013-09-05 14:45:29 +02001026 error_setg(errp, "Could not determine image format: No compatible "
1027 "driver found");
Alberto Fariabf5b16f2022-07-05 17:15:09 +01001028 *pdrv = NULL;
1029 return -ENOENT;
Stefan Weilc98ac352010-07-21 21:51:51 +02001030 }
Alberto Fariabf5b16f2022-07-05 17:15:09 +01001031
Stefan Weilc98ac352010-07-21 21:51:51 +02001032 *pdrv = drv;
Alberto Fariabf5b16f2022-07-05 17:15:09 +01001033 return 0;
bellardea2384d2004-08-01 21:59:26 +00001034}
1035
Stefan Hajnoczi51762282010-04-19 16:56:41 +01001036/**
1037 * Set the current 'total_sectors' value
Markus Armbruster65a9bb22014-06-26 13:23:17 +02001038 * Return 0 on success, -errno on error.
Stefan Hajnoczi51762282010-04-19 16:56:41 +01001039 */
Emanuele Giuseppe Espositoc86422c2023-01-13 21:42:04 +01001040int coroutine_fn bdrv_co_refresh_total_sectors(BlockDriverState *bs,
1041 int64_t hint)
Stefan Hajnoczi51762282010-04-19 16:56:41 +01001042{
1043 BlockDriver *drv = bs->drv;
Emanuele Giuseppe Esposito967d7902022-03-03 10:15:58 -05001044 IO_CODE();
Kevin Wolf8ab81402023-02-03 16:22:02 +01001045 assert_bdrv_graph_readable();
Stefan Hajnoczi51762282010-04-19 16:56:41 +01001046
Max Reitzd470ad42017-11-10 21:31:09 +01001047 if (!drv) {
1048 return -ENOMEDIUM;
1049 }
1050
Emanuele Giuseppe Espositoc86422c2023-01-13 21:42:04 +01001051 /* Do not attempt drv->bdrv_co_getlength() on scsi-generic devices */
Dimitris Aragiorgisb192af82015-06-23 13:44:56 +03001052 if (bdrv_is_sg(bs))
Nicholas Bellinger396759a2010-05-17 09:46:04 -07001053 return 0;
1054
Stefan Hajnoczi51762282010-04-19 16:56:41 +01001055 /* query actual device if possible, otherwise just trust the hint */
Emanuele Giuseppe Espositoc86422c2023-01-13 21:42:04 +01001056 if (drv->bdrv_co_getlength) {
1057 int64_t length = drv->bdrv_co_getlength(bs);
Stefan Hajnoczi51762282010-04-19 16:56:41 +01001058 if (length < 0) {
1059 return length;
1060 }
Fam Zheng7e382002013-11-06 19:48:06 +08001061 hint = DIV_ROUND_UP(length, BDRV_SECTOR_SIZE);
Stefan Hajnoczi51762282010-04-19 16:56:41 +01001062 }
1063
1064 bs->total_sectors = hint;
Vladimir Sementsov-Ogievskiy8b117002020-12-04 01:27:13 +03001065
1066 if (bs->total_sectors * BDRV_SECTOR_SIZE > BDRV_MAX_LENGTH) {
1067 return -EFBIG;
1068 }
1069
Stefan Hajnoczi51762282010-04-19 16:56:41 +01001070 return 0;
1071}
1072
Stefan Hajnoczic3993cd2011-08-04 12:26:51 +01001073/**
Kevin Wolfcddff5b2015-11-16 16:43:27 +01001074 * Combines a QDict of new block driver @options with any missing options taken
1075 * from @old_options, so that leaving out an option defaults to its old value.
1076 */
1077static void bdrv_join_options(BlockDriverState *bs, QDict *options,
1078 QDict *old_options)
1079{
Emanuele Giuseppe Espositoda359902022-03-03 10:16:11 -05001080 GLOBAL_STATE_CODE();
Kevin Wolfcddff5b2015-11-16 16:43:27 +01001081 if (bs->drv && bs->drv->bdrv_join_options) {
1082 bs->drv->bdrv_join_options(options, old_options);
1083 } else {
1084 qdict_join(options, old_options, false);
1085 }
1086}
1087
Alberto Garcia543770b2018-09-06 12:37:09 +03001088static BlockdevDetectZeroesOptions bdrv_parse_detect_zeroes(QemuOpts *opts,
1089 int open_flags,
1090 Error **errp)
1091{
1092 Error *local_err = NULL;
1093 char *value = qemu_opt_get_del(opts, "detect-zeroes");
1094 BlockdevDetectZeroesOptions detect_zeroes =
1095 qapi_enum_parse(&BlockdevDetectZeroesOptions_lookup, value,
1096 BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF, &local_err);
Emanuele Giuseppe Espositobdb73472022-03-03 10:16:02 -05001097 GLOBAL_STATE_CODE();
Alberto Garcia543770b2018-09-06 12:37:09 +03001098 g_free(value);
1099 if (local_err) {
1100 error_propagate(errp, local_err);
1101 return detect_zeroes;
1102 }
1103
1104 if (detect_zeroes == BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP &&
1105 !(open_flags & BDRV_O_UNMAP))
1106 {
1107 error_setg(errp, "setting detect-zeroes to unmap is not allowed "
1108 "without setting discard operation to unmap");
1109 }
1110
1111 return detect_zeroes;
1112}
1113
Kevin Wolfcddff5b2015-11-16 16:43:27 +01001114/**
Aarushi Mehtaf80f2672020-01-20 14:18:50 +00001115 * Set open flags for aio engine
1116 *
1117 * Return 0 on success, -1 if the engine specified is invalid
1118 */
1119int bdrv_parse_aio(const char *mode, int *flags)
1120{
1121 if (!strcmp(mode, "threads")) {
1122 /* do nothing, default */
1123 } else if (!strcmp(mode, "native")) {
1124 *flags |= BDRV_O_NATIVE_AIO;
1125#ifdef CONFIG_LINUX_IO_URING
1126 } else if (!strcmp(mode, "io_uring")) {
1127 *flags |= BDRV_O_IO_URING;
1128#endif
1129 } else {
1130 return -1;
1131 }
1132
1133 return 0;
1134}
1135
1136/**
Paolo Bonzini9e8f1832013-02-08 14:06:11 +01001137 * Set open flags for a given discard mode
1138 *
1139 * Return 0 on success, -1 if the discard mode was invalid.
1140 */
1141int bdrv_parse_discard_flags(const char *mode, int *flags)
1142{
1143 *flags &= ~BDRV_O_UNMAP;
1144
1145 if (!strcmp(mode, "off") || !strcmp(mode, "ignore")) {
1146 /* do nothing */
1147 } else if (!strcmp(mode, "on") || !strcmp(mode, "unmap")) {
1148 *flags |= BDRV_O_UNMAP;
1149 } else {
1150 return -1;
1151 }
1152
1153 return 0;
1154}
1155
1156/**
Stefan Hajnoczic3993cd2011-08-04 12:26:51 +01001157 * Set open flags for a given cache mode
1158 *
1159 * Return 0 on success, -1 if the cache mode was invalid.
1160 */
Kevin Wolf53e8ae02016-03-18 15:36:58 +01001161int bdrv_parse_cache_mode(const char *mode, int *flags, bool *writethrough)
Stefan Hajnoczic3993cd2011-08-04 12:26:51 +01001162{
1163 *flags &= ~BDRV_O_CACHE_MASK;
1164
1165 if (!strcmp(mode, "off") || !strcmp(mode, "none")) {
Kevin Wolf53e8ae02016-03-18 15:36:58 +01001166 *writethrough = false;
1167 *flags |= BDRV_O_NOCACHE;
Stefan Hajnoczi92196b22011-08-04 12:26:52 +01001168 } else if (!strcmp(mode, "directsync")) {
Kevin Wolf53e8ae02016-03-18 15:36:58 +01001169 *writethrough = true;
Stefan Hajnoczi92196b22011-08-04 12:26:52 +01001170 *flags |= BDRV_O_NOCACHE;
Stefan Hajnoczic3993cd2011-08-04 12:26:51 +01001171 } else if (!strcmp(mode, "writeback")) {
Kevin Wolf53e8ae02016-03-18 15:36:58 +01001172 *writethrough = false;
Stefan Hajnoczic3993cd2011-08-04 12:26:51 +01001173 } else if (!strcmp(mode, "unsafe")) {
Kevin Wolf53e8ae02016-03-18 15:36:58 +01001174 *writethrough = false;
Stefan Hajnoczic3993cd2011-08-04 12:26:51 +01001175 *flags |= BDRV_O_NO_FLUSH;
1176 } else if (!strcmp(mode, "writethrough")) {
Kevin Wolf53e8ae02016-03-18 15:36:58 +01001177 *writethrough = true;
Stefan Hajnoczic3993cd2011-08-04 12:26:51 +01001178 } else {
1179 return -1;
1180 }
1181
1182 return 0;
1183}
1184
Kevin Wolfb5411552017-01-17 15:56:16 +01001185static char *bdrv_child_get_parent_desc(BdrvChild *c)
1186{
1187 BlockDriverState *parent = c->opaque;
Vladimir Sementsov-Ogievskiy2c0a3ac2021-06-01 10:52:15 +03001188 return g_strdup_printf("node '%s'", bdrv_get_node_name(parent));
Kevin Wolfb5411552017-01-17 15:56:16 +01001189}
1190
Kevin Wolf20018e12016-05-23 18:46:59 +02001191static void bdrv_child_cb_drained_begin(BdrvChild *child)
1192{
1193 BlockDriverState *bs = child->opaque;
Kevin Wolfa82a3bd2022-11-18 18:41:07 +01001194 bdrv_do_drained_begin_quiesce(bs, NULL);
Kevin Wolf20018e12016-05-23 18:46:59 +02001195}
1196
Kevin Wolf89bd0302018-03-22 14:11:20 +01001197static bool bdrv_child_cb_drained_poll(BdrvChild *child)
1198{
1199 BlockDriverState *bs = child->opaque;
Kevin Wolf299403a2022-11-18 18:41:05 +01001200 return bdrv_drain_poll(bs, NULL, false);
Kevin Wolf89bd0302018-03-22 14:11:20 +01001201}
1202
Kevin Wolf2f65df62022-11-18 18:40:59 +01001203static void bdrv_child_cb_drained_end(BdrvChild *child)
Kevin Wolf20018e12016-05-23 18:46:59 +02001204{
1205 BlockDriverState *bs = child->opaque;
Kevin Wolf2f65df62022-11-18 18:40:59 +01001206 bdrv_drained_end(bs);
Kevin Wolf20018e12016-05-23 18:46:59 +02001207}
1208
Kevin Wolf38701b62017-05-04 18:52:39 +02001209static int bdrv_child_cb_inactivate(BdrvChild *child)
1210{
1211 BlockDriverState *bs = child->opaque;
Emanuele Giuseppe Espositobdb73472022-03-03 10:16:02 -05001212 GLOBAL_STATE_CODE();
Kevin Wolf38701b62017-05-04 18:52:39 +02001213 assert(bs->open_flags & BDRV_O_INACTIVE);
1214 return 0;
1215}
1216
Emanuele Giuseppe Esposito27633e72022-10-25 04:49:47 -04001217static bool bdrv_child_cb_change_aio_ctx(BdrvChild *child, AioContext *ctx,
1218 GHashTable *visited, Transaction *tran,
1219 Error **errp)
Kevin Wolf5d231842019-05-06 19:17:56 +02001220{
1221 BlockDriverState *bs = child->opaque;
Emanuele Giuseppe Esposito27633e72022-10-25 04:49:47 -04001222 return bdrv_change_aio_context(bs, ctx, visited, tran, errp);
Kevin Wolf53a7d042019-05-06 19:17:59 +02001223}
1224
Kevin Wolf0b50cc82014-04-11 21:29:52 +02001225/*
Kevin Wolf73176be2016-03-07 13:02:15 +01001226 * Returns the options and flags that a temporary snapshot should get, based on
1227 * the originally requested flags (the originally requested image will have
1228 * flags like a backing file)
Kevin Wolfb1e6fc02014-05-06 12:11:42 +02001229 */
Kevin Wolf73176be2016-03-07 13:02:15 +01001230static void bdrv_temp_snapshot_options(int *child_flags, QDict *child_options,
1231 int parent_flags, QDict *parent_options)
Kevin Wolfb1e6fc02014-05-06 12:11:42 +02001232{
Emanuele Giuseppe Espositobdb73472022-03-03 10:16:02 -05001233 GLOBAL_STATE_CODE();
Kevin Wolf73176be2016-03-07 13:02:15 +01001234 *child_flags = (parent_flags & ~BDRV_O_SNAPSHOT) | BDRV_O_TEMPORARY;
1235
1236 /* For temporary files, unconditional cache=unsafe is fine */
Kevin Wolf73176be2016-03-07 13:02:15 +01001237 qdict_set_default_str(child_options, BDRV_OPT_CACHE_DIRECT, "off");
1238 qdict_set_default_str(child_options, BDRV_OPT_CACHE_NO_FLUSH, "on");
Kevin Wolf41869042016-06-16 12:59:30 +02001239
Kevin Wolf3f486862019-04-04 17:04:43 +02001240 /* Copy the read-only and discard options from the parent */
Alberto Garciaf87a0e22016-09-15 17:53:02 +03001241 qdict_copy_default(child_options, parent_options, BDRV_OPT_READ_ONLY);
Kevin Wolf3f486862019-04-04 17:04:43 +02001242 qdict_copy_default(child_options, parent_options, BDRV_OPT_DISCARD);
Alberto Garciaf87a0e22016-09-15 17:53:02 +03001243
Kevin Wolf41869042016-06-16 12:59:30 +02001244 /* aio=native doesn't work for cache.direct=off, so disable it for the
1245 * temporary snapshot */
1246 *child_flags &= ~BDRV_O_NATIVE_AIO;
Kevin Wolfb1e6fc02014-05-06 12:11:42 +02001247}
1248
Kevin Wolfdb95dbb2017-02-08 11:28:52 +01001249static void bdrv_backing_attach(BdrvChild *c)
1250{
1251 BlockDriverState *parent = c->opaque;
1252 BlockDriverState *backing_hd = c->bs;
1253
Emanuele Giuseppe Espositobdb73472022-03-03 10:16:02 -05001254 GLOBAL_STATE_CODE();
Kevin Wolfdb95dbb2017-02-08 11:28:52 +01001255 assert(!parent->backing_blocker);
1256 error_setg(&parent->backing_blocker,
1257 "node is used as backing hd of '%s'",
1258 bdrv_get_device_or_node_name(parent));
1259
Max Reitzf30c66b2019-02-01 20:29:05 +01001260 bdrv_refresh_filename(backing_hd);
1261
Kevin Wolfdb95dbb2017-02-08 11:28:52 +01001262 parent->open_flags &= ~BDRV_O_NO_BACKING;
Kevin Wolfdb95dbb2017-02-08 11:28:52 +01001263
1264 bdrv_op_block_all(backing_hd, parent->backing_blocker);
1265 /* Otherwise we won't be able to commit or stream */
1266 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_COMMIT_TARGET,
1267 parent->backing_blocker);
1268 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_STREAM,
1269 parent->backing_blocker);
1270 /*
1271 * We do backup in 3 ways:
1272 * 1. drive backup
1273 * The target bs is new opened, and the source is top BDS
1274 * 2. blockdev backup
1275 * Both the source and the target are top BDSes.
1276 * 3. internal backup(used for block replication)
1277 * Both the source and the target are backing file
1278 *
1279 * In case 1 and 2, neither the source nor the target is the backing file.
1280 * In case 3, we will block the top BDS, so there is only one block job
1281 * for the top BDS and its backing chain.
1282 */
1283 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_BACKUP_SOURCE,
1284 parent->backing_blocker);
1285 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_BACKUP_TARGET,
1286 parent->backing_blocker);
Max Reitzca2f1232020-05-13 13:05:22 +02001287}
Kevin Wolfd736f112017-12-18 16:05:48 +01001288
Kevin Wolfdb95dbb2017-02-08 11:28:52 +01001289static void bdrv_backing_detach(BdrvChild *c)
1290{
1291 BlockDriverState *parent = c->opaque;
1292
Emanuele Giuseppe Espositobdb73472022-03-03 10:16:02 -05001293 GLOBAL_STATE_CODE();
Kevin Wolfdb95dbb2017-02-08 11:28:52 +01001294 assert(parent->backing_blocker);
1295 bdrv_op_unblock_all(c->bs, parent->backing_blocker);
1296 error_free(parent->backing_blocker);
1297 parent->backing_blocker = NULL;
Max Reitz48e08282020-05-13 13:05:23 +02001298}
Kevin Wolfd736f112017-12-18 16:05:48 +01001299
Kevin Wolf6858eba2017-06-29 19:32:21 +02001300static int bdrv_backing_update_filename(BdrvChild *c, BlockDriverState *base,
1301 const char *filename, Error **errp)
1302{
1303 BlockDriverState *parent = c->opaque;
Alberto Garciae94d3db2018-11-12 16:00:34 +02001304 bool read_only = bdrv_is_read_only(parent);
Kevin Wolf6858eba2017-06-29 19:32:21 +02001305 int ret;
Emanuele Giuseppe Espositobdb73472022-03-03 10:16:02 -05001306 GLOBAL_STATE_CODE();
Kevin Wolf6858eba2017-06-29 19:32:21 +02001307
Alberto Garciae94d3db2018-11-12 16:00:34 +02001308 if (read_only) {
1309 ret = bdrv_reopen_set_read_only(parent, false, errp);
Kevin Wolf61f09ce2017-09-19 16:22:54 +02001310 if (ret < 0) {
1311 return ret;
1312 }
1313 }
1314
Kevin Wolf6858eba2017-06-29 19:32:21 +02001315 ret = bdrv_change_backing_file(parent, filename,
Eric Blakee54ee1b2020-07-06 15:39:53 -05001316 base->drv ? base->drv->format_name : "",
1317 false);
Kevin Wolf6858eba2017-06-29 19:32:21 +02001318 if (ret < 0) {
Kevin Wolf64730692017-11-06 17:52:58 +01001319 error_setg_errno(errp, -ret, "Could not update backing file link");
Kevin Wolf6858eba2017-06-29 19:32:21 +02001320 }
1321
Alberto Garciae94d3db2018-11-12 16:00:34 +02001322 if (read_only) {
1323 bdrv_reopen_set_read_only(parent, true, NULL);
Kevin Wolf61f09ce2017-09-19 16:22:54 +02001324 }
1325
Kevin Wolf6858eba2017-06-29 19:32:21 +02001326 return ret;
1327}
1328
Max Reitzfae8bd32020-05-13 13:05:20 +02001329/*
1330 * Returns the options and flags that a generic child of a BDS should
1331 * get, based on the given options and flags for the parent BDS.
1332 */
Max Reitz00ff7ff2020-05-13 13:05:21 +02001333static void bdrv_inherited_options(BdrvChildRole role, bool parent_is_format,
1334 int *child_flags, QDict *child_options,
1335 int parent_flags, QDict *parent_options)
Max Reitzfae8bd32020-05-13 13:05:20 +02001336{
1337 int flags = parent_flags;
Emanuele Giuseppe Espositobdb73472022-03-03 10:16:02 -05001338 GLOBAL_STATE_CODE();
Max Reitzfae8bd32020-05-13 13:05:20 +02001339
1340 /*
1341 * First, decide whether to set, clear, or leave BDRV_O_PROTOCOL.
1342 * Generally, the question to answer is: Should this child be
1343 * format-probed by default?
1344 */
1345
1346 /*
1347 * Pure and non-filtered data children of non-format nodes should
1348 * be probed by default (even when the node itself has BDRV_O_PROTOCOL
1349 * set). This only affects a very limited set of drivers (namely
1350 * quorum and blkverify when this comment was written).
1351 * Force-clear BDRV_O_PROTOCOL then.
1352 */
1353 if (!parent_is_format &&
1354 (role & BDRV_CHILD_DATA) &&
1355 !(role & (BDRV_CHILD_METADATA | BDRV_CHILD_FILTERED)))
1356 {
1357 flags &= ~BDRV_O_PROTOCOL;
1358 }
1359
1360 /*
1361 * All children of format nodes (except for COW children) and all
1362 * metadata children in general should never be format-probed.
1363 * Force-set BDRV_O_PROTOCOL then.
1364 */
1365 if ((parent_is_format && !(role & BDRV_CHILD_COW)) ||
1366 (role & BDRV_CHILD_METADATA))
1367 {
1368 flags |= BDRV_O_PROTOCOL;
1369 }
1370
1371 /*
1372 * If the cache mode isn't explicitly set, inherit direct and no-flush from
1373 * the parent.
1374 */
1375 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_DIRECT);
1376 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_NO_FLUSH);
1377 qdict_copy_default(child_options, parent_options, BDRV_OPT_FORCE_SHARE);
1378
1379 if (role & BDRV_CHILD_COW) {
1380 /* backing files are opened read-only by default */
1381 qdict_set_default_str(child_options, BDRV_OPT_READ_ONLY, "on");
1382 qdict_set_default_str(child_options, BDRV_OPT_AUTO_READ_ONLY, "off");
1383 } else {
1384 /* Inherit the read-only option from the parent if it's not set */
1385 qdict_copy_default(child_options, parent_options, BDRV_OPT_READ_ONLY);
1386 qdict_copy_default(child_options, parent_options,
1387 BDRV_OPT_AUTO_READ_ONLY);
1388 }
1389
1390 /*
1391 * bdrv_co_pdiscard() respects unmap policy for the parent, so we
1392 * can default to enable it on lower layers regardless of the
1393 * parent option.
1394 */
1395 qdict_set_default_str(child_options, BDRV_OPT_DISCARD, "unmap");
1396
1397 /* Clear flags that only apply to the top layer */
1398 flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_COPY_ON_READ);
1399
1400 if (role & BDRV_CHILD_METADATA) {
1401 flags &= ~BDRV_O_NO_IO;
1402 }
1403 if (role & BDRV_CHILD_COW) {
1404 flags &= ~BDRV_O_TEMPORARY;
1405 }
1406
1407 *child_flags = flags;
1408}
1409
Kevin Wolf303de472022-12-07 14:18:35 +01001410static void GRAPH_WRLOCK bdrv_child_cb_attach(BdrvChild *child)
Max Reitzca2f1232020-05-13 13:05:22 +02001411{
1412 BlockDriverState *bs = child->opaque;
1413
Emanuele Giuseppe Esposito3f35f822022-12-07 14:18:33 +01001414 assert_bdrv_graph_writable();
Hanna Reitza2253692021-11-15 15:53:58 +01001415 QLIST_INSERT_HEAD(&bs->children, child, next);
Vladimir Sementsov-Ogievskiy5bb047472022-07-26 23:11:32 +03001416 if (bs->drv->is_filter || (child->role & BDRV_CHILD_FILTERED)) {
1417 /*
1418 * Here we handle filters and block/raw-format.c when it behave like
1419 * filter. They generally have a single PRIMARY child, which is also the
1420 * FILTERED child, and that they may have multiple more children, which
1421 * are neither PRIMARY nor FILTERED. And never we have a COW child here.
1422 * So bs->file will be the PRIMARY child, unless the PRIMARY child goes
1423 * into bs->backing on exceptional cases; and bs->backing will be
1424 * nothing else.
1425 */
1426 assert(!(child->role & BDRV_CHILD_COW));
1427 if (child->role & BDRV_CHILD_PRIMARY) {
1428 assert(child->role & BDRV_CHILD_FILTERED);
1429 assert(!bs->backing);
1430 assert(!bs->file);
Hanna Reitza2253692021-11-15 15:53:58 +01001431
Vladimir Sementsov-Ogievskiy5bb047472022-07-26 23:11:32 +03001432 if (bs->drv->filtered_child_is_backing) {
1433 bs->backing = child;
1434 } else {
1435 bs->file = child;
1436 }
1437 } else {
1438 assert(!(child->role & BDRV_CHILD_FILTERED));
1439 }
1440 } else if (child->role & BDRV_CHILD_COW) {
1441 assert(bs->drv->supports_backing);
1442 assert(!(child->role & BDRV_CHILD_PRIMARY));
1443 assert(!bs->backing);
1444 bs->backing = child;
Max Reitzca2f1232020-05-13 13:05:22 +02001445 bdrv_backing_attach(child);
Vladimir Sementsov-Ogievskiy5bb047472022-07-26 23:11:32 +03001446 } else if (child->role & BDRV_CHILD_PRIMARY) {
1447 assert(!bs->file);
1448 bs->file = child;
Max Reitzca2f1232020-05-13 13:05:22 +02001449 }
Max Reitzca2f1232020-05-13 13:05:22 +02001450}
1451
Kevin Wolf303de472022-12-07 14:18:35 +01001452static void GRAPH_WRLOCK bdrv_child_cb_detach(BdrvChild *child)
Max Reitz48e08282020-05-13 13:05:23 +02001453{
1454 BlockDriverState *bs = child->opaque;
1455
1456 if (child->role & BDRV_CHILD_COW) {
1457 bdrv_backing_detach(child);
1458 }
1459
Emanuele Giuseppe Esposito3f35f822022-12-07 14:18:33 +01001460 assert_bdrv_graph_writable();
Hanna Reitza2253692021-11-15 15:53:58 +01001461 QLIST_REMOVE(child, next);
Vladimir Sementsov-Ogievskiy5bb047472022-07-26 23:11:32 +03001462 if (child == bs->backing) {
1463 assert(child != bs->file);
1464 bs->backing = NULL;
1465 } else if (child == bs->file) {
1466 bs->file = NULL;
1467 }
Max Reitz48e08282020-05-13 13:05:23 +02001468}
1469
Max Reitz43483552020-05-13 13:05:24 +02001470static int bdrv_child_cb_update_filename(BdrvChild *c, BlockDriverState *base,
1471 const char *filename, Error **errp)
1472{
1473 if (c->role & BDRV_CHILD_COW) {
1474 return bdrv_backing_update_filename(c, base, filename, errp);
1475 }
1476 return 0;
1477}
1478
Vladimir Sementsov-Ogievskiyfb62b582021-05-24 13:12:56 +03001479AioContext *child_of_bds_get_parent_aio_context(BdrvChild *c)
Vladimir Sementsov-Ogievskiy3ca1f322021-04-28 18:17:33 +03001480{
1481 BlockDriverState *bs = c->opaque;
Emanuele Giuseppe Esposito384a48f2022-03-03 10:15:50 -05001482 IO_CODE();
Vladimir Sementsov-Ogievskiy3ca1f322021-04-28 18:17:33 +03001483
1484 return bdrv_get_aio_context(bs);
1485}
1486
Max Reitz43483552020-05-13 13:05:24 +02001487const BdrvChildClass child_of_bds = {
1488 .parent_is_bds = true,
1489 .get_parent_desc = bdrv_child_get_parent_desc,
1490 .inherit_options = bdrv_inherited_options,
1491 .drained_begin = bdrv_child_cb_drained_begin,
1492 .drained_poll = bdrv_child_cb_drained_poll,
1493 .drained_end = bdrv_child_cb_drained_end,
1494 .attach = bdrv_child_cb_attach,
1495 .detach = bdrv_child_cb_detach,
1496 .inactivate = bdrv_child_cb_inactivate,
Emanuele Giuseppe Esposito27633e72022-10-25 04:49:47 -04001497 .change_aio_ctx = bdrv_child_cb_change_aio_ctx,
Max Reitz43483552020-05-13 13:05:24 +02001498 .update_filename = bdrv_child_cb_update_filename,
Vladimir Sementsov-Ogievskiyfb62b582021-05-24 13:12:56 +03001499 .get_parent_aio_context = child_of_bds_get_parent_aio_context,
Max Reitz43483552020-05-13 13:05:24 +02001500};
1501
Vladimir Sementsov-Ogievskiy3ca1f322021-04-28 18:17:33 +03001502AioContext *bdrv_child_get_parent_aio_context(BdrvChild *c)
1503{
Hanna Reitzd5f8d792022-11-07 16:13:19 +01001504 IO_CODE();
Vladimir Sementsov-Ogievskiy3ca1f322021-04-28 18:17:33 +03001505 return c->klass->get_parent_aio_context(c);
1506}
1507
Kevin Wolf7b272452012-11-12 17:05:39 +01001508static int bdrv_open_flags(BlockDriverState *bs, int flags)
1509{
Kevin Wolf61de4c62016-03-18 17:46:45 +01001510 int open_flags = flags;
Emanuele Giuseppe Espositobdb73472022-03-03 10:16:02 -05001511 GLOBAL_STATE_CODE();
Kevin Wolf7b272452012-11-12 17:05:39 +01001512
1513 /*
1514 * Clear flags that are internal to the block layer before opening the
1515 * image.
1516 */
Kevin Wolf20cca272014-06-04 14:33:27 +02001517 open_flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_PROTOCOL);
Kevin Wolf7b272452012-11-12 17:05:39 +01001518
Kevin Wolf7b272452012-11-12 17:05:39 +01001519 return open_flags;
1520}
1521
Kevin Wolf91a097e2015-05-08 17:49:53 +02001522static void update_flags_from_options(int *flags, QemuOpts *opts)
1523{
Emanuele Giuseppe Espositobdb73472022-03-03 10:16:02 -05001524 GLOBAL_STATE_CODE();
1525
Alberto Garcia2a3d4332018-11-12 16:00:48 +02001526 *flags &= ~(BDRV_O_CACHE_MASK | BDRV_O_RDWR | BDRV_O_AUTO_RDONLY);
Kevin Wolf91a097e2015-05-08 17:49:53 +02001527
Alberto Garcia57f9db92018-09-06 12:37:06 +03001528 if (qemu_opt_get_bool_del(opts, BDRV_OPT_CACHE_NO_FLUSH, false)) {
Kevin Wolf91a097e2015-05-08 17:49:53 +02001529 *flags |= BDRV_O_NO_FLUSH;
1530 }
1531
Alberto Garcia57f9db92018-09-06 12:37:06 +03001532 if (qemu_opt_get_bool_del(opts, BDRV_OPT_CACHE_DIRECT, false)) {
Kevin Wolf91a097e2015-05-08 17:49:53 +02001533 *flags |= BDRV_O_NOCACHE;
1534 }
Alberto Garciaf87a0e22016-09-15 17:53:02 +03001535
Alberto Garcia57f9db92018-09-06 12:37:06 +03001536 if (!qemu_opt_get_bool_del(opts, BDRV_OPT_READ_ONLY, false)) {
Alberto Garciaf87a0e22016-09-15 17:53:02 +03001537 *flags |= BDRV_O_RDWR;
1538 }
1539
Kevin Wolfe35bdc12018-10-05 18:57:40 +02001540 if (qemu_opt_get_bool_del(opts, BDRV_OPT_AUTO_READ_ONLY, false)) {
1541 *flags |= BDRV_O_AUTO_RDONLY;
1542 }
Kevin Wolf91a097e2015-05-08 17:49:53 +02001543}
1544
1545static void update_options_from_flags(QDict *options, int flags)
1546{
Emanuele Giuseppe Espositobdb73472022-03-03 10:16:02 -05001547 GLOBAL_STATE_CODE();
Kevin Wolf91a097e2015-05-08 17:49:53 +02001548 if (!qdict_haskey(options, BDRV_OPT_CACHE_DIRECT)) {
Eric Blake46f5ac22017-04-27 16:58:17 -05001549 qdict_put_bool(options, BDRV_OPT_CACHE_DIRECT, flags & BDRV_O_NOCACHE);
Kevin Wolf91a097e2015-05-08 17:49:53 +02001550 }
1551 if (!qdict_haskey(options, BDRV_OPT_CACHE_NO_FLUSH)) {
Eric Blake46f5ac22017-04-27 16:58:17 -05001552 qdict_put_bool(options, BDRV_OPT_CACHE_NO_FLUSH,
1553 flags & BDRV_O_NO_FLUSH);
Kevin Wolf91a097e2015-05-08 17:49:53 +02001554 }
Alberto Garciaf87a0e22016-09-15 17:53:02 +03001555 if (!qdict_haskey(options, BDRV_OPT_READ_ONLY)) {
Eric Blake46f5ac22017-04-27 16:58:17 -05001556 qdict_put_bool(options, BDRV_OPT_READ_ONLY, !(flags & BDRV_O_RDWR));
Alberto Garciaf87a0e22016-09-15 17:53:02 +03001557 }
Kevin Wolfe35bdc12018-10-05 18:57:40 +02001558 if (!qdict_haskey(options, BDRV_OPT_AUTO_READ_ONLY)) {
1559 qdict_put_bool(options, BDRV_OPT_AUTO_READ_ONLY,
1560 flags & BDRV_O_AUTO_RDONLY);
1561 }
Kevin Wolf91a097e2015-05-08 17:49:53 +02001562}
1563
Kevin Wolf636ea372014-01-24 14:11:52 +01001564static void bdrv_assign_node_name(BlockDriverState *bs,
1565 const char *node_name,
1566 Error **errp)
Benoît Canet6913c0c2014-01-23 21:31:33 +01001567{
Jeff Cody15489c72015-10-12 19:36:50 -04001568 char *gen_node_name = NULL;
Emanuele Giuseppe Espositobdb73472022-03-03 10:16:02 -05001569 GLOBAL_STATE_CODE();
Benoît Canet6913c0c2014-01-23 21:31:33 +01001570
Jeff Cody15489c72015-10-12 19:36:50 -04001571 if (!node_name) {
1572 node_name = gen_node_name = id_generate(ID_BLOCK);
1573 } else if (!id_wellformed(node_name)) {
1574 /*
1575 * Check for empty string or invalid characters, but not if it is
1576 * generated (generated names use characters not available to the user)
1577 */
Connor Kuehl785ec4b2021-03-05 09:19:28 -06001578 error_setg(errp, "Invalid node-name: '%s'", node_name);
Kevin Wolf636ea372014-01-24 14:11:52 +01001579 return;
Benoît Canet6913c0c2014-01-23 21:31:33 +01001580 }
1581
Benoît Canet0c5e94e2014-02-12 17:15:07 +01001582 /* takes care of avoiding namespaces collisions */
Markus Armbruster7f06d472014-10-07 13:59:12 +02001583 if (blk_by_name(node_name)) {
Benoît Canet0c5e94e2014-02-12 17:15:07 +01001584 error_setg(errp, "node-name=%s is conflicting with a device id",
1585 node_name);
Jeff Cody15489c72015-10-12 19:36:50 -04001586 goto out;
Benoît Canet0c5e94e2014-02-12 17:15:07 +01001587 }
1588
Benoît Canet6913c0c2014-01-23 21:31:33 +01001589 /* takes care of avoiding duplicates node names */
1590 if (bdrv_find_node(node_name)) {
Connor Kuehl785ec4b2021-03-05 09:19:28 -06001591 error_setg(errp, "Duplicate nodes with node-name='%s'", node_name);
Jeff Cody15489c72015-10-12 19:36:50 -04001592 goto out;
Benoît Canet6913c0c2014-01-23 21:31:33 +01001593 }
1594
Kevin Wolf824808d2018-07-04 13:28:29 +02001595 /* Make sure that the node name isn't truncated */
1596 if (strlen(node_name) >= sizeof(bs->node_name)) {
1597 error_setg(errp, "Node name too long");
1598 goto out;
1599 }
1600
Benoît Canet6913c0c2014-01-23 21:31:33 +01001601 /* copy node name into the bs and insert it into the graph list */
1602 pstrcpy(bs->node_name, sizeof(bs->node_name), node_name);
1603 QTAILQ_INSERT_TAIL(&graph_bdrv_states, bs, node_list);
Jeff Cody15489c72015-10-12 19:36:50 -04001604out:
1605 g_free(gen_node_name);
Benoît Canet6913c0c2014-01-23 21:31:33 +01001606}
1607
Emanuele Giuseppe Espositoc86422c2023-01-13 21:42:04 +01001608/*
1609 * The caller must always hold @bs AioContext lock, because this function calls
1610 * bdrv_refresh_total_sectors() which polls when called from non-coroutine
1611 * context.
1612 */
Kevin Wolf1a30b0f2023-05-04 13:57:38 +02001613static int no_coroutine_fn GRAPH_UNLOCKED
1614bdrv_open_driver(BlockDriverState *bs, BlockDriver *drv, const char *node_name,
1615 QDict *options, int open_flags, Error **errp)
Kevin Wolf01a56502017-01-18 15:51:56 +01001616{
Kevin Wolf8dc8a602023-05-25 14:47:10 +02001617 AioContext *ctx;
Kevin Wolf01a56502017-01-18 15:51:56 +01001618 Error *local_err = NULL;
Kevin Wolf0f122642018-03-28 18:29:18 +02001619 int i, ret;
Emanuele Giuseppe Espositoda359902022-03-03 10:16:11 -05001620 GLOBAL_STATE_CODE();
Kevin Wolf01a56502017-01-18 15:51:56 +01001621
1622 bdrv_assign_node_name(bs, node_name, &local_err);
1623 if (local_err) {
1624 error_propagate(errp, local_err);
1625 return -EINVAL;
1626 }
1627
1628 bs->drv = drv;
1629 bs->opaque = g_malloc0(drv->instance_size);
1630
1631 if (drv->bdrv_file_open) {
1632 assert(!drv->bdrv_needs_filename || bs->filename[0]);
1633 ret = drv->bdrv_file_open(bs, options, open_flags, &local_err);
Kevin Wolf680c7f92017-01-18 17:16:41 +01001634 } else if (drv->bdrv_open) {
Kevin Wolf01a56502017-01-18 15:51:56 +01001635 ret = drv->bdrv_open(bs, options, open_flags, &local_err);
Kevin Wolf680c7f92017-01-18 17:16:41 +01001636 } else {
1637 ret = 0;
Kevin Wolf01a56502017-01-18 15:51:56 +01001638 }
1639
1640 if (ret < 0) {
1641 if (local_err) {
1642 error_propagate(errp, local_err);
1643 } else if (bs->filename[0]) {
1644 error_setg_errno(errp, -ret, "Could not open '%s'", bs->filename);
1645 } else {
1646 error_setg_errno(errp, -ret, "Could not open image");
1647 }
Manos Pitsidianakis180ca192017-07-14 17:35:48 +03001648 goto open_failed;
Kevin Wolf01a56502017-01-18 15:51:56 +01001649 }
1650
Stefan Hajnoczie8b65352022-10-13 14:59:01 -04001651 assert(!(bs->supported_read_flags & ~BDRV_REQ_MASK));
1652 assert(!(bs->supported_write_flags & ~BDRV_REQ_MASK));
1653
1654 /*
1655 * Always allow the BDRV_REQ_REGISTERED_BUF optimization hint. This saves
1656 * drivers that pass read/write requests through to a child the trouble of
1657 * declaring support explicitly.
1658 *
1659 * Drivers must not propagate this flag accidentally when they initiate I/O
1660 * to a bounce buffer. That case should be rare though.
1661 */
1662 bs->supported_read_flags |= BDRV_REQ_REGISTERED_BUF;
1663 bs->supported_write_flags |= BDRV_REQ_REGISTERED_BUF;
1664
Kevin Wolf8dc8a602023-05-25 14:47:10 +02001665 /* Get the context after .bdrv_open, it can change the context */
1666 ctx = bdrv_get_aio_context(bs);
1667 aio_context_acquire(ctx);
1668
Emanuele Giuseppe Espositoc0579602023-01-13 21:42:03 +01001669 ret = bdrv_refresh_total_sectors(bs, bs->total_sectors);
Kevin Wolf01a56502017-01-18 15:51:56 +01001670 if (ret < 0) {
1671 error_setg_errno(errp, -ret, "Could not refresh total sector count");
Kevin Wolf8dc8a602023-05-25 14:47:10 +02001672 aio_context_release(ctx);
Manos Pitsidianakis180ca192017-07-14 17:35:48 +03001673 return ret;
Kevin Wolf01a56502017-01-18 15:51:56 +01001674 }
1675
Kevin Wolfe19b1572023-05-04 13:57:50 +02001676 bdrv_graph_rdlock_main_loop();
Vladimir Sementsov-Ogievskiy1e4c7972021-04-28 18:17:55 +03001677 bdrv_refresh_limits(bs, NULL, &local_err);
Kevin Wolfe19b1572023-05-04 13:57:50 +02001678 bdrv_graph_rdunlock_main_loop();
Kevin Wolf8dc8a602023-05-25 14:47:10 +02001679 aio_context_release(ctx);
Kevin Wolfe19b1572023-05-04 13:57:50 +02001680
Kevin Wolf01a56502017-01-18 15:51:56 +01001681 if (local_err) {
1682 error_propagate(errp, local_err);
Manos Pitsidianakis180ca192017-07-14 17:35:48 +03001683 return -EINVAL;
Kevin Wolf01a56502017-01-18 15:51:56 +01001684 }
1685
1686 assert(bdrv_opt_mem_align(bs) != 0);
1687 assert(bdrv_min_mem_align(bs) != 0);
1688 assert(is_power_of_2(bs->bl.request_alignment));
1689
Kevin Wolf0f122642018-03-28 18:29:18 +02001690 for (i = 0; i < bs->quiesce_counter; i++) {
Kevin Wolf5e8ac212022-11-18 18:40:58 +01001691 if (drv->bdrv_drain_begin) {
1692 drv->bdrv_drain_begin(bs);
Kevin Wolf0f122642018-03-28 18:29:18 +02001693 }
1694 }
1695
Kevin Wolf01a56502017-01-18 15:51:56 +01001696 return 0;
Manos Pitsidianakis180ca192017-07-14 17:35:48 +03001697open_failed:
1698 bs->drv = NULL;
1699 if (bs->file != NULL) {
1700 bdrv_unref_child(bs, bs->file);
Vladimir Sementsov-Ogievskiy5bb047472022-07-26 23:11:32 +03001701 assert(!bs->file);
Manos Pitsidianakis180ca192017-07-14 17:35:48 +03001702 }
Kevin Wolf01a56502017-01-18 15:51:56 +01001703 g_free(bs->opaque);
1704 bs->opaque = NULL;
Kevin Wolf01a56502017-01-18 15:51:56 +01001705 return ret;
1706}
1707
Vladimir Sementsov-Ogievskiy621d1732021-09-20 14:55:34 +03001708/*
1709 * Create and open a block node.
1710 *
1711 * @options is a QDict of options to pass to the block drivers, or NULL for an
1712 * empty set of options. The reference to the QDict belongs to the block layer
1713 * after the call (even on failure), so if the caller intends to reuse the
1714 * dictionary, it needs to use qobject_ref() before calling bdrv_open.
1715 */
1716BlockDriverState *bdrv_new_open_driver_opts(BlockDriver *drv,
1717 const char *node_name,
1718 QDict *options, int flags,
1719 Error **errp)
Kevin Wolf680c7f92017-01-18 17:16:41 +01001720{
1721 BlockDriverState *bs;
1722 int ret;
1723
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05001724 GLOBAL_STATE_CODE();
1725
Kevin Wolf680c7f92017-01-18 17:16:41 +01001726 bs = bdrv_new();
1727 bs->open_flags = flags;
Vladimir Sementsov-Ogievskiy621d1732021-09-20 14:55:34 +03001728 bs->options = options ?: qdict_new();
1729 bs->explicit_options = qdict_clone_shallow(bs->options);
Kevin Wolf680c7f92017-01-18 17:16:41 +01001730 bs->opaque = NULL;
1731
1732 update_options_from_flags(bs->options, flags);
1733
1734 ret = bdrv_open_driver(bs, drv, node_name, bs->options, flags, errp);
1735 if (ret < 0) {
Marc-André Lureaucb3e7f02018-04-19 17:01:43 +02001736 qobject_unref(bs->explicit_options);
Manos Pitsidianakis180ca192017-07-14 17:35:48 +03001737 bs->explicit_options = NULL;
Marc-André Lureaucb3e7f02018-04-19 17:01:43 +02001738 qobject_unref(bs->options);
Manos Pitsidianakis180ca192017-07-14 17:35:48 +03001739 bs->options = NULL;
Kevin Wolf680c7f92017-01-18 17:16:41 +01001740 bdrv_unref(bs);
1741 return NULL;
1742 }
1743
1744 return bs;
1745}
1746
Vladimir Sementsov-Ogievskiy621d1732021-09-20 14:55:34 +03001747/* Create and open a block node. */
1748BlockDriverState *bdrv_new_open_driver(BlockDriver *drv, const char *node_name,
1749 int flags, Error **errp)
1750{
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05001751 GLOBAL_STATE_CODE();
Vladimir Sementsov-Ogievskiy621d1732021-09-20 14:55:34 +03001752 return bdrv_new_open_driver_opts(drv, node_name, NULL, flags, errp);
1753}
1754
Kevin Wolfc5f30142016-10-06 11:33:17 +02001755QemuOptsList bdrv_runtime_opts = {
Kevin Wolf18edf282015-04-07 17:12:56 +02001756 .name = "bdrv_common",
1757 .head = QTAILQ_HEAD_INITIALIZER(bdrv_runtime_opts.head),
1758 .desc = {
1759 {
1760 .name = "node-name",
1761 .type = QEMU_OPT_STRING,
1762 .help = "Node name of the block device node",
1763 },
Kevin Wolf62392eb2015-04-24 16:38:02 +02001764 {
1765 .name = "driver",
1766 .type = QEMU_OPT_STRING,
1767 .help = "Block driver to use for the node",
1768 },
Kevin Wolf91a097e2015-05-08 17:49:53 +02001769 {
Kevin Wolf91a097e2015-05-08 17:49:53 +02001770 .name = BDRV_OPT_CACHE_DIRECT,
1771 .type = QEMU_OPT_BOOL,
1772 .help = "Bypass software writeback cache on the host",
1773 },
1774 {
1775 .name = BDRV_OPT_CACHE_NO_FLUSH,
1776 .type = QEMU_OPT_BOOL,
1777 .help = "Ignore flush requests",
1778 },
Alberto Garciaf87a0e22016-09-15 17:53:02 +03001779 {
1780 .name = BDRV_OPT_READ_ONLY,
1781 .type = QEMU_OPT_BOOL,
1782 .help = "Node is opened in read-only mode",
1783 },
Kevin Wolf692e01a2016-09-12 21:00:41 +02001784 {
Kevin Wolfe35bdc12018-10-05 18:57:40 +02001785 .name = BDRV_OPT_AUTO_READ_ONLY,
1786 .type = QEMU_OPT_BOOL,
1787 .help = "Node can become read-only if opening read-write fails",
1788 },
1789 {
Kevin Wolf692e01a2016-09-12 21:00:41 +02001790 .name = "detect-zeroes",
1791 .type = QEMU_OPT_STRING,
1792 .help = "try to optimize zero writes (off, on, unmap)",
1793 },
Kevin Wolf818584a2016-09-12 18:03:18 +02001794 {
Alberto Garcia415bbca2018-10-03 13:23:13 +03001795 .name = BDRV_OPT_DISCARD,
Kevin Wolf818584a2016-09-12 18:03:18 +02001796 .type = QEMU_OPT_STRING,
1797 .help = "discard operation (ignore/off, unmap/on)",
1798 },
Fam Zheng5a9347c2017-05-03 00:35:37 +08001799 {
1800 .name = BDRV_OPT_FORCE_SHARE,
1801 .type = QEMU_OPT_BOOL,
1802 .help = "always accept other writers (default: off)",
1803 },
Kevin Wolf18edf282015-04-07 17:12:56 +02001804 { /* end of list */ }
1805 },
1806};
1807
Maxim Levitsky5a5e7f82020-03-26 03:12:18 +02001808QemuOptsList bdrv_create_opts_simple = {
1809 .name = "simple-create-opts",
1810 .head = QTAILQ_HEAD_INITIALIZER(bdrv_create_opts_simple.head),
Max Reitzfd171462020-01-22 17:45:29 +01001811 .desc = {
1812 {
1813 .name = BLOCK_OPT_SIZE,
1814 .type = QEMU_OPT_SIZE,
1815 .help = "Virtual disk size"
1816 },
1817 {
1818 .name = BLOCK_OPT_PREALLOC,
1819 .type = QEMU_OPT_STRING,
1820 .help = "Preallocation mode (allowed values: off)"
1821 },
1822 { /* end of list */ }
1823 }
1824};
1825
Kevin Wolfb6ce07a2010-04-12 16:37:13 +02001826/*
Kevin Wolf57915332010-04-14 15:24:50 +02001827 * Common part for opening disk images and files
Kevin Wolfb6ad4912013-03-15 10:35:04 +01001828 *
1829 * Removes all processed options from *options.
Kevin Wolf57915332010-04-14 15:24:50 +02001830 */
Kevin Wolf5696c6e2017-02-17 18:39:24 +01001831static int bdrv_open_common(BlockDriverState *bs, BlockBackend *file,
Kevin Wolf82dc8b42016-01-11 19:07:50 +01001832 QDict *options, Error **errp)
Kevin Wolf57915332010-04-14 15:24:50 +02001833{
1834 int ret, open_flags;
Kevin Wolf035fccd2013-04-09 14:34:19 +02001835 const char *filename;
Kevin Wolf62392eb2015-04-24 16:38:02 +02001836 const char *driver_name = NULL;
Benoît Canet6913c0c2014-01-23 21:31:33 +01001837 const char *node_name = NULL;
Kevin Wolf818584a2016-09-12 18:03:18 +02001838 const char *discard;
Kevin Wolf18edf282015-04-07 17:12:56 +02001839 QemuOpts *opts;
Kevin Wolf62392eb2015-04-24 16:38:02 +02001840 BlockDriver *drv;
Max Reitz34b5d2c2013-09-05 14:45:29 +02001841 Error *local_err = NULL;
Vladimir Sementsov-Ogievskiy307261b2021-05-27 18:40:54 +03001842 bool ro;
Kevin Wolf57915332010-04-14 15:24:50 +02001843
Paolo Bonzini64058752012-05-08 16:51:49 +02001844 assert(bs->file == NULL);
Kevin Wolf707ff822013-03-06 12:20:31 +01001845 assert(options != NULL && bs->options != options);
Emanuele Giuseppe Espositobdb73472022-03-03 10:16:02 -05001846 GLOBAL_STATE_CODE();
Kevin Wolf57915332010-04-14 15:24:50 +02001847
Kevin Wolf62392eb2015-04-24 16:38:02 +02001848 opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
Markus Armbrusteraf175e82020-07-07 18:06:03 +02001849 if (!qemu_opts_absorb_qdict(opts, options, errp)) {
Kevin Wolf62392eb2015-04-24 16:38:02 +02001850 ret = -EINVAL;
1851 goto fail_opts;
1852 }
1853
Alberto Garcia9b7e8692016-09-15 17:53:01 +03001854 update_flags_from_options(&bs->open_flags, opts);
1855
Kevin Wolf62392eb2015-04-24 16:38:02 +02001856 driver_name = qemu_opt_get(opts, "driver");
1857 drv = bdrv_find_format(driver_name);
1858 assert(drv != NULL);
1859
Fam Zheng5a9347c2017-05-03 00:35:37 +08001860 bs->force_share = qemu_opt_get_bool(opts, BDRV_OPT_FORCE_SHARE, false);
1861
1862 if (bs->force_share && (bs->open_flags & BDRV_O_RDWR)) {
1863 error_setg(errp,
1864 BDRV_OPT_FORCE_SHARE
1865 "=on can only be used with read-only images");
1866 ret = -EINVAL;
1867 goto fail_opts;
1868 }
1869
Kevin Wolf45673672013-04-22 17:48:40 +02001870 if (file != NULL) {
Max Reitzf30c66b2019-02-01 20:29:05 +01001871 bdrv_refresh_filename(blk_bs(file));
Kevin Wolf5696c6e2017-02-17 18:39:24 +01001872 filename = blk_bs(file)->filename;
Kevin Wolf45673672013-04-22 17:48:40 +02001873 } else {
Markus Armbruster129c7d12017-03-30 19:43:12 +02001874 /*
1875 * Caution: while qdict_get_try_str() is fine, getting
1876 * non-string types would require more care. When @options
1877 * come from -blockdev or blockdev_add, its members are typed
1878 * according to the QAPI schema, but when they come from
1879 * -drive, they're all QString.
1880 */
Kevin Wolf45673672013-04-22 17:48:40 +02001881 filename = qdict_get_try_str(options, "filename");
1882 }
1883
Max Reitz4a008242017-04-13 18:06:24 +02001884 if (drv->bdrv_needs_filename && (!filename || !filename[0])) {
Kevin Wolf765003d2014-02-03 14:49:42 +01001885 error_setg(errp, "The '%s' block driver requires a file name",
1886 drv->format_name);
Kevin Wolf18edf282015-04-07 17:12:56 +02001887 ret = -EINVAL;
1888 goto fail_opts;
1889 }
1890
Kevin Wolf82dc8b42016-01-11 19:07:50 +01001891 trace_bdrv_open_common(bs, filename ?: "", bs->open_flags,
1892 drv->format_name);
Kevin Wolf62392eb2015-04-24 16:38:02 +02001893
Vladimir Sementsov-Ogievskiy307261b2021-05-27 18:40:54 +03001894 ro = bdrv_is_read_only(bs);
1895
1896 if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv, ro)) {
1897 if (!ro && bdrv_is_whitelisted(drv, true)) {
Kevin Wolf8be25de2019-01-22 13:15:31 +01001898 ret = bdrv_apply_auto_read_only(bs, NULL, NULL);
1899 } else {
1900 ret = -ENOTSUP;
1901 }
1902 if (ret < 0) {
1903 error_setg(errp,
Vladimir Sementsov-Ogievskiy307261b2021-05-27 18:40:54 +03001904 !ro && bdrv_is_whitelisted(drv, true)
Kevin Wolf8be25de2019-01-22 13:15:31 +01001905 ? "Driver '%s' can only be used for read-only devices"
1906 : "Driver '%s' is not whitelisted",
1907 drv->format_name);
1908 goto fail_opts;
1909 }
Fam Zhengb64ec4e2013-05-29 19:35:40 +08001910 }
Kevin Wolf57915332010-04-14 15:24:50 +02001911
Paolo Bonzinid3faa132017-06-05 14:38:50 +02001912 /* bdrv_new() and bdrv_close() make it so */
Stefan Hajnoczid73415a2020-09-23 11:56:46 +01001913 assert(qatomic_read(&bs->copy_on_read) == 0);
Paolo Bonzinid3faa132017-06-05 14:38:50 +02001914
Kevin Wolf82dc8b42016-01-11 19:07:50 +01001915 if (bs->open_flags & BDRV_O_COPY_ON_READ) {
Vladimir Sementsov-Ogievskiy307261b2021-05-27 18:40:54 +03001916 if (!ro) {
Kevin Wolf0ebd24e2013-09-19 15:12:18 +02001917 bdrv_enable_copy_on_read(bs);
1918 } else {
1919 error_setg(errp, "Can't use copy-on-read on read-only device");
Kevin Wolf18edf282015-04-07 17:12:56 +02001920 ret = -EINVAL;
1921 goto fail_opts;
Kevin Wolf0ebd24e2013-09-19 15:12:18 +02001922 }
Stefan Hajnoczi53fec9d2011-11-28 16:08:47 +00001923 }
1924
Alberto Garcia415bbca2018-10-03 13:23:13 +03001925 discard = qemu_opt_get(opts, BDRV_OPT_DISCARD);
Kevin Wolf818584a2016-09-12 18:03:18 +02001926 if (discard != NULL) {
1927 if (bdrv_parse_discard_flags(discard, &bs->open_flags) != 0) {
1928 error_setg(errp, "Invalid discard option");
1929 ret = -EINVAL;
1930 goto fail_opts;
1931 }
1932 }
1933
Alberto Garcia543770b2018-09-06 12:37:09 +03001934 bs->detect_zeroes =
1935 bdrv_parse_detect_zeroes(opts, bs->open_flags, &local_err);
1936 if (local_err) {
1937 error_propagate(errp, local_err);
1938 ret = -EINVAL;
1939 goto fail_opts;
Kevin Wolf692e01a2016-09-12 21:00:41 +02001940 }
1941
Kevin Wolfc2ad1b02013-03-18 16:40:51 +01001942 if (filename != NULL) {
1943 pstrcpy(bs->filename, sizeof(bs->filename), filename);
1944 } else {
1945 bs->filename[0] = '\0';
1946 }
Max Reitz91af7012014-07-18 20:24:56 +02001947 pstrcpy(bs->exact_filename, sizeof(bs->exact_filename), bs->filename);
Kevin Wolf57915332010-04-14 15:24:50 +02001948
Kevin Wolf66f82ce2010-04-14 14:17:38 +02001949 /* Open the image, either directly or using a protocol */
Kevin Wolf82dc8b42016-01-11 19:07:50 +01001950 open_flags = bdrv_open_flags(bs, bs->open_flags);
Kevin Wolf01a56502017-01-18 15:51:56 +01001951 node_name = qemu_opt_get(opts, "node-name");
Kevin Wolf66f82ce2010-04-14 14:17:38 +02001952
Kevin Wolf01a56502017-01-18 15:51:56 +01001953 assert(!drv->bdrv_file_open || file == NULL);
1954 ret = bdrv_open_driver(bs, drv, node_name, options, open_flags, errp);
Kevin Wolf57915332010-04-14 15:24:50 +02001955 if (ret < 0) {
Kevin Wolf01a56502017-01-18 15:51:56 +01001956 goto fail_opts;
Kevin Wolf57915332010-04-14 15:24:50 +02001957 }
1958
Kevin Wolf18edf282015-04-07 17:12:56 +02001959 qemu_opts_del(opts);
Kevin Wolf57915332010-04-14 15:24:50 +02001960 return 0;
1961
Kevin Wolf18edf282015-04-07 17:12:56 +02001962fail_opts:
1963 qemu_opts_del(opts);
Kevin Wolf57915332010-04-14 15:24:50 +02001964 return ret;
1965}
1966
Kevin Wolf5e5c4f62014-05-26 11:45:08 +02001967static QDict *parse_json_filename(const char *filename, Error **errp)
1968{
1969 QObject *options_obj;
1970 QDict *options;
1971 int ret;
Emanuele Giuseppe Espositobdb73472022-03-03 10:16:02 -05001972 GLOBAL_STATE_CODE();
Kevin Wolf5e5c4f62014-05-26 11:45:08 +02001973
1974 ret = strstart(filename, "json:", &filename);
1975 assert(ret);
1976
Markus Armbruster5577fff2017-02-28 22:26:59 +01001977 options_obj = qobject_from_json(filename, errp);
Kevin Wolf5e5c4f62014-05-26 11:45:08 +02001978 if (!options_obj) {
Markus Armbruster5577fff2017-02-28 22:26:59 +01001979 error_prepend(errp, "Could not parse the JSON options: ");
Kevin Wolf5e5c4f62014-05-26 11:45:08 +02001980 return NULL;
1981 }
1982
Max Reitz7dc847e2018-02-24 16:40:29 +01001983 options = qobject_to(QDict, options_obj);
Markus Armbrusterca6b6e12017-02-17 21:38:18 +01001984 if (!options) {
Marc-André Lureaucb3e7f02018-04-19 17:01:43 +02001985 qobject_unref(options_obj);
Kevin Wolf5e5c4f62014-05-26 11:45:08 +02001986 error_setg(errp, "Invalid JSON object given");
1987 return NULL;
1988 }
1989
Kevin Wolf5e5c4f62014-05-26 11:45:08 +02001990 qdict_flatten(options);
1991
1992 return options;
1993}
1994
Kevin Wolfde3b53f2015-10-29 15:24:41 +01001995static void parse_json_protocol(QDict *options, const char **pfilename,
1996 Error **errp)
1997{
1998 QDict *json_options;
1999 Error *local_err = NULL;
Emanuele Giuseppe Espositobdb73472022-03-03 10:16:02 -05002000 GLOBAL_STATE_CODE();
Kevin Wolfde3b53f2015-10-29 15:24:41 +01002001
2002 /* Parse json: pseudo-protocol */
2003 if (!*pfilename || !g_str_has_prefix(*pfilename, "json:")) {
2004 return;
2005 }
2006
2007 json_options = parse_json_filename(*pfilename, &local_err);
2008 if (local_err) {
2009 error_propagate(errp, local_err);
2010 return;
2011 }
2012
2013 /* Options given in the filename have lower priority than options
2014 * specified directly */
2015 qdict_join(options, json_options, false);
Marc-André Lureaucb3e7f02018-04-19 17:01:43 +02002016 qobject_unref(json_options);
Kevin Wolfde3b53f2015-10-29 15:24:41 +01002017 *pfilename = NULL;
2018}
2019
Kevin Wolf57915332010-04-14 15:24:50 +02002020/*
Kevin Wolff54120f2014-05-26 11:09:59 +02002021 * Fills in default options for opening images and converts the legacy
2022 * filename/flags pair to option QDict entries.
Max Reitz53a29512015-03-19 14:53:16 -04002023 * The BDRV_O_PROTOCOL flag in *flags will be set or cleared accordingly if a
2024 * block driver has been specified explicitly.
Kevin Wolff54120f2014-05-26 11:09:59 +02002025 */
Kevin Wolfde3b53f2015-10-29 15:24:41 +01002026static int bdrv_fill_options(QDict **options, const char *filename,
Max Reitz053e1572015-08-26 19:47:51 +02002027 int *flags, Error **errp)
Kevin Wolff54120f2014-05-26 11:09:59 +02002028{
2029 const char *drvname;
Max Reitz53a29512015-03-19 14:53:16 -04002030 bool protocol = *flags & BDRV_O_PROTOCOL;
Kevin Wolff54120f2014-05-26 11:09:59 +02002031 bool parse_filename = false;
Max Reitz053e1572015-08-26 19:47:51 +02002032 BlockDriver *drv = NULL;
Kevin Wolff54120f2014-05-26 11:09:59 +02002033 Error *local_err = NULL;
Kevin Wolff54120f2014-05-26 11:09:59 +02002034
Emanuele Giuseppe Espositoda359902022-03-03 10:16:11 -05002035 GLOBAL_STATE_CODE();
2036
Markus Armbruster129c7d12017-03-30 19:43:12 +02002037 /*
2038 * Caution: while qdict_get_try_str() is fine, getting non-string
2039 * types would require more care. When @options come from
2040 * -blockdev or blockdev_add, its members are typed according to
2041 * the QAPI schema, but when they come from -drive, they're all
2042 * QString.
2043 */
Max Reitz53a29512015-03-19 14:53:16 -04002044 drvname = qdict_get_try_str(*options, "driver");
Max Reitz053e1572015-08-26 19:47:51 +02002045 if (drvname) {
2046 drv = bdrv_find_format(drvname);
2047 if (!drv) {
2048 error_setg(errp, "Unknown driver '%s'", drvname);
2049 return -ENOENT;
2050 }
2051 /* If the user has explicitly specified the driver, this choice should
2052 * override the BDRV_O_PROTOCOL flag */
2053 protocol = drv->bdrv_file_open;
Max Reitz53a29512015-03-19 14:53:16 -04002054 }
2055
2056 if (protocol) {
2057 *flags |= BDRV_O_PROTOCOL;
2058 } else {
2059 *flags &= ~BDRV_O_PROTOCOL;
2060 }
2061
Kevin Wolf91a097e2015-05-08 17:49:53 +02002062 /* Translate cache options from flags into options */
2063 update_options_from_flags(*options, *flags);
2064
Kevin Wolff54120f2014-05-26 11:09:59 +02002065 /* Fetch the file name from the options QDict if necessary */
Kevin Wolf17b005f2014-05-27 10:50:29 +02002066 if (protocol && filename) {
Kevin Wolff54120f2014-05-26 11:09:59 +02002067 if (!qdict_haskey(*options, "filename")) {
Eric Blake46f5ac22017-04-27 16:58:17 -05002068 qdict_put_str(*options, "filename", filename);
Kevin Wolff54120f2014-05-26 11:09:59 +02002069 parse_filename = true;
2070 } else {
2071 error_setg(errp, "Can't specify 'file' and 'filename' options at "
2072 "the same time");
2073 return -EINVAL;
2074 }
2075 }
2076
2077 /* Find the right block driver */
Markus Armbruster129c7d12017-03-30 19:43:12 +02002078 /* See cautionary note on accessing @options above */
Kevin Wolff54120f2014-05-26 11:09:59 +02002079 filename = qdict_get_try_str(*options, "filename");
Kevin Wolff54120f2014-05-26 11:09:59 +02002080
Max Reitz053e1572015-08-26 19:47:51 +02002081 if (!drvname && protocol) {
2082 if (filename) {
2083 drv = bdrv_find_protocol(filename, parse_filename, errp);
2084 if (!drv) {
Kevin Wolff54120f2014-05-26 11:09:59 +02002085 return -EINVAL;
2086 }
Max Reitz053e1572015-08-26 19:47:51 +02002087
2088 drvname = drv->format_name;
Eric Blake46f5ac22017-04-27 16:58:17 -05002089 qdict_put_str(*options, "driver", drvname);
Max Reitz053e1572015-08-26 19:47:51 +02002090 } else {
2091 error_setg(errp, "Must specify either driver or file");
2092 return -EINVAL;
Kevin Wolff54120f2014-05-26 11:09:59 +02002093 }
2094 }
2095
Kevin Wolf17b005f2014-05-27 10:50:29 +02002096 assert(drv || !protocol);
Kevin Wolff54120f2014-05-26 11:09:59 +02002097
2098 /* Driver-specific filename parsing */
Kevin Wolf17b005f2014-05-27 10:50:29 +02002099 if (drv && drv->bdrv_parse_filename && parse_filename) {
Kevin Wolff54120f2014-05-26 11:09:59 +02002100 drv->bdrv_parse_filename(filename, *options, &local_err);
2101 if (local_err) {
2102 error_propagate(errp, local_err);
2103 return -EINVAL;
2104 }
2105
2106 if (!drv->bdrv_needs_filename) {
2107 qdict_del(*options, "filename");
2108 }
2109 }
2110
2111 return 0;
2112}
2113
Kevin Wolf148eb132017-09-14 14:32:04 +02002114typedef struct BlockReopenQueueEntry {
2115 bool prepared;
Kevin Wolf69b736e2019-03-05 17:18:22 +01002116 bool perms_checked;
Kevin Wolf148eb132017-09-14 14:32:04 +02002117 BDRVReopenState state;
Vladimir Sementsov-Ogievskiy859443b2019-09-27 15:23:47 +03002118 QTAILQ_ENTRY(BlockReopenQueueEntry) entry;
Kevin Wolf148eb132017-09-14 14:32:04 +02002119} BlockReopenQueueEntry;
2120
2121/*
2122 * Return the flags that @bs will have after the reopens in @q have
2123 * successfully completed. If @q is NULL (or @bs is not contained in @q),
2124 * return the current flags.
2125 */
2126static int bdrv_reopen_get_flags(BlockReopenQueue *q, BlockDriverState *bs)
2127{
2128 BlockReopenQueueEntry *entry;
2129
2130 if (q != NULL) {
Vladimir Sementsov-Ogievskiy859443b2019-09-27 15:23:47 +03002131 QTAILQ_FOREACH(entry, q, entry) {
Kevin Wolf148eb132017-09-14 14:32:04 +02002132 if (entry->state.bs == bs) {
2133 return entry->state.flags;
2134 }
2135 }
2136 }
2137
2138 return bs->open_flags;
2139}
2140
2141/* Returns whether the image file can be written to after the reopen queue @q
2142 * has been successfully applied, or right now if @q is NULL. */
Max Reitzcc022142018-06-06 21:37:00 +02002143static bool bdrv_is_writable_after_reopen(BlockDriverState *bs,
2144 BlockReopenQueue *q)
Kevin Wolf148eb132017-09-14 14:32:04 +02002145{
2146 int flags = bdrv_reopen_get_flags(q, bs);
2147
2148 return (flags & (BDRV_O_RDWR | BDRV_O_INACTIVE)) == BDRV_O_RDWR;
2149}
2150
Max Reitzcc022142018-06-06 21:37:00 +02002151/*
2152 * Return whether the BDS can be written to. This is not necessarily
2153 * the same as !bdrv_is_read_only(bs), as inactivated images may not
2154 * be written to but do not count as read-only images.
2155 */
2156bool bdrv_is_writable(BlockDriverState *bs)
2157{
Emanuele Giuseppe Esposito384a48f2022-03-03 10:15:50 -05002158 IO_CODE();
Max Reitzcc022142018-06-06 21:37:00 +02002159 return bdrv_is_writable_after_reopen(bs, NULL);
2160}
2161
Vladimir Sementsov-Ogievskiy3bf416b2021-04-28 18:17:37 +03002162static char *bdrv_child_user_desc(BdrvChild *c)
2163{
Emanuele Giuseppe Espositof0c28322022-03-03 10:16:13 -05002164 GLOBAL_STATE_CODE();
Vladimir Sementsov-Ogievskiyda261b62021-06-01 10:52:17 +03002165 return c->klass->get_parent_desc(c);
Vladimir Sementsov-Ogievskiy3bf416b2021-04-28 18:17:37 +03002166}
2167
Vladimir Sementsov-Ogievskiy30ebb9a2021-06-01 10:52:18 +03002168/*
2169 * Check that @a allows everything that @b needs. @a and @b must reference same
2170 * child node.
2171 */
Vladimir Sementsov-Ogievskiy3bf416b2021-04-28 18:17:37 +03002172static bool bdrv_a_allow_b(BdrvChild *a, BdrvChild *b, Error **errp)
2173{
Vladimir Sementsov-Ogievskiy30ebb9a2021-06-01 10:52:18 +03002174 const char *child_bs_name;
2175 g_autofree char *a_user = NULL;
2176 g_autofree char *b_user = NULL;
2177 g_autofree char *perms = NULL;
2178
2179 assert(a->bs);
2180 assert(a->bs == b->bs);
Emanuele Giuseppe Esposito862fded2022-03-03 10:15:55 -05002181 GLOBAL_STATE_CODE();
Vladimir Sementsov-Ogievskiy3bf416b2021-04-28 18:17:37 +03002182
2183 if ((b->perm & a->shared_perm) == b->perm) {
2184 return true;
2185 }
2186
Vladimir Sementsov-Ogievskiy30ebb9a2021-06-01 10:52:18 +03002187 child_bs_name = bdrv_get_node_name(b->bs);
2188 a_user = bdrv_child_user_desc(a);
2189 b_user = bdrv_child_user_desc(b);
2190 perms = bdrv_perm_names(b->perm & ~a->shared_perm);
2191
2192 error_setg(errp, "Permission conflict on node '%s': permissions '%s' are "
2193 "both required by %s (uses node '%s' as '%s' child) and "
2194 "unshared by %s (uses node '%s' as '%s' child).",
2195 child_bs_name, perms,
2196 b_user, child_bs_name, b->name,
2197 a_user, child_bs_name, a->name);
Vladimir Sementsov-Ogievskiy3bf416b2021-04-28 18:17:37 +03002198
2199 return false;
2200}
2201
Vladimir Sementsov-Ogievskiy9397c142021-04-28 18:17:53 +03002202static bool bdrv_parent_perms_conflict(BlockDriverState *bs, Error **errp)
Vladimir Sementsov-Ogievskiy3bf416b2021-04-28 18:17:37 +03002203{
2204 BdrvChild *a, *b;
Emanuele Giuseppe Esposito862fded2022-03-03 10:15:55 -05002205 GLOBAL_STATE_CODE();
Vladimir Sementsov-Ogievskiy3bf416b2021-04-28 18:17:37 +03002206
2207 /*
2208 * During the loop we'll look at each pair twice. That's correct because
2209 * bdrv_a_allow_b() is asymmetric and we should check each pair in both
2210 * directions.
2211 */
2212 QLIST_FOREACH(a, &bs->parents, next_parent) {
2213 QLIST_FOREACH(b, &bs->parents, next_parent) {
Vladimir Sementsov-Ogievskiy9397c142021-04-28 18:17:53 +03002214 if (a == b) {
Vladimir Sementsov-Ogievskiy3bf416b2021-04-28 18:17:37 +03002215 continue;
2216 }
2217
2218 if (!bdrv_a_allow_b(a, b, errp)) {
2219 return true;
2220 }
2221 }
2222 }
2223
2224 return false;
2225}
2226
Fam Zhengffd1a5a2017-05-03 00:35:38 +08002227static void bdrv_child_perm(BlockDriverState *bs, BlockDriverState *child_bs,
Max Reitze5d8a402020-05-13 13:05:44 +02002228 BdrvChild *c, BdrvChildRole role,
2229 BlockReopenQueue *reopen_queue,
Fam Zhengffd1a5a2017-05-03 00:35:38 +08002230 uint64_t parent_perm, uint64_t parent_shared,
2231 uint64_t *nperm, uint64_t *nshared)
2232{
Alberto Garcia0b3ca762019-04-04 14:29:53 +03002233 assert(bs->drv && bs->drv->bdrv_child_perm);
Emanuele Giuseppe Espositoda359902022-03-03 10:16:11 -05002234 GLOBAL_STATE_CODE();
Max Reitze5d8a402020-05-13 13:05:44 +02002235 bs->drv->bdrv_child_perm(bs, c, role, reopen_queue,
Alberto Garcia0b3ca762019-04-04 14:29:53 +03002236 parent_perm, parent_shared,
2237 nperm, nshared);
Kevin Wolfe0995dc2017-09-14 12:47:11 +02002238 /* TODO Take force_share from reopen_queue */
Fam Zhengffd1a5a2017-05-03 00:35:38 +08002239 if (child_bs && child_bs->force_share) {
2240 *nshared = BLK_PERM_ALL;
2241 }
2242}
2243
Vladimir Sementsov-Ogievskiybd57f8f2021-04-28 18:17:41 +03002244/*
2245 * Adds the whole subtree of @bs (including @bs itself) to the @list (except for
2246 * nodes that are already in the @list, of course) so that final list is
2247 * topologically sorted. Return the result (GSList @list object is updated, so
2248 * don't use old reference after function call).
2249 *
2250 * On function start @list must be already topologically sorted and for any node
2251 * in the @list the whole subtree of the node must be in the @list as well. The
2252 * simplest way to satisfy this criteria: use only result of
2253 * bdrv_topological_dfs() or NULL as @list parameter.
2254 */
2255static GSList *bdrv_topological_dfs(GSList *list, GHashTable *found,
2256 BlockDriverState *bs)
2257{
2258 BdrvChild *child;
2259 g_autoptr(GHashTable) local_found = NULL;
2260
Emanuele Giuseppe Espositobdb73472022-03-03 10:16:02 -05002261 GLOBAL_STATE_CODE();
2262
Vladimir Sementsov-Ogievskiybd57f8f2021-04-28 18:17:41 +03002263 if (!found) {
2264 assert(!list);
2265 found = local_found = g_hash_table_new(NULL, NULL);
2266 }
2267
2268 if (g_hash_table_contains(found, bs)) {
2269 return list;
2270 }
2271 g_hash_table_add(found, bs);
2272
2273 QLIST_FOREACH(child, &bs->children, next) {
2274 list = bdrv_topological_dfs(list, found, child->bs);
2275 }
2276
2277 return g_slist_prepend(list, bs);
2278}
2279
Vladimir Sementsov-Ogievskiyecb776b2021-04-28 18:18:02 +03002280typedef struct BdrvChildSetPermState {
2281 BdrvChild *child;
2282 uint64_t old_perm;
2283 uint64_t old_shared_perm;
2284} BdrvChildSetPermState;
Vladimir Sementsov-Ogievskiyb0defa82021-04-28 18:17:38 +03002285
2286static void bdrv_child_set_perm_abort(void *opaque)
2287{
Vladimir Sementsov-Ogievskiyecb776b2021-04-28 18:18:02 +03002288 BdrvChildSetPermState *s = opaque;
2289
Emanuele Giuseppe Esposito862fded2022-03-03 10:15:55 -05002290 GLOBAL_STATE_CODE();
2291
Vladimir Sementsov-Ogievskiyecb776b2021-04-28 18:18:02 +03002292 s->child->perm = s->old_perm;
2293 s->child->shared_perm = s->old_shared_perm;
Vladimir Sementsov-Ogievskiyb0defa82021-04-28 18:17:38 +03002294}
2295
2296static TransactionActionDrv bdrv_child_set_pem_drv = {
2297 .abort = bdrv_child_set_perm_abort,
Vladimir Sementsov-Ogievskiyecb776b2021-04-28 18:18:02 +03002298 .clean = g_free,
Vladimir Sementsov-Ogievskiyb0defa82021-04-28 18:17:38 +03002299};
2300
Vladimir Sementsov-Ogievskiyecb776b2021-04-28 18:18:02 +03002301static void bdrv_child_set_perm(BdrvChild *c, uint64_t perm,
2302 uint64_t shared, Transaction *tran)
Vladimir Sementsov-Ogievskiyb0defa82021-04-28 18:17:38 +03002303{
Vladimir Sementsov-Ogievskiyecb776b2021-04-28 18:18:02 +03002304 BdrvChildSetPermState *s = g_new(BdrvChildSetPermState, 1);
Emanuele Giuseppe Esposito862fded2022-03-03 10:15:55 -05002305 GLOBAL_STATE_CODE();
Vladimir Sementsov-Ogievskiyecb776b2021-04-28 18:18:02 +03002306
2307 *s = (BdrvChildSetPermState) {
2308 .child = c,
2309 .old_perm = c->perm,
2310 .old_shared_perm = c->shared_perm,
2311 };
Vladimir Sementsov-Ogievskiyb0defa82021-04-28 18:17:38 +03002312
2313 c->perm = perm;
2314 c->shared_perm = shared;
2315
Vladimir Sementsov-Ogievskiyecb776b2021-04-28 18:18:02 +03002316 tran_add(tran, &bdrv_child_set_pem_drv, s);
Vladimir Sementsov-Ogievskiyb0defa82021-04-28 18:17:38 +03002317}
2318
Vladimir Sementsov-Ogievskiy2513ef52021-04-28 18:17:42 +03002319static void bdrv_drv_set_perm_commit(void *opaque)
2320{
2321 BlockDriverState *bs = opaque;
2322 uint64_t cumulative_perms, cumulative_shared_perms;
Emanuele Giuseppe Espositoda359902022-03-03 10:16:11 -05002323 GLOBAL_STATE_CODE();
Vladimir Sementsov-Ogievskiy2513ef52021-04-28 18:17:42 +03002324
2325 if (bs->drv->bdrv_set_perm) {
2326 bdrv_get_cumulative_perm(bs, &cumulative_perms,
2327 &cumulative_shared_perms);
2328 bs->drv->bdrv_set_perm(bs, cumulative_perms, cumulative_shared_perms);
2329 }
2330}
2331
2332static void bdrv_drv_set_perm_abort(void *opaque)
2333{
2334 BlockDriverState *bs = opaque;
Emanuele Giuseppe Espositoda359902022-03-03 10:16:11 -05002335 GLOBAL_STATE_CODE();
Vladimir Sementsov-Ogievskiy2513ef52021-04-28 18:17:42 +03002336
2337 if (bs->drv->bdrv_abort_perm_update) {
2338 bs->drv->bdrv_abort_perm_update(bs);
2339 }
2340}
2341
2342TransactionActionDrv bdrv_drv_set_perm_drv = {
2343 .abort = bdrv_drv_set_perm_abort,
2344 .commit = bdrv_drv_set_perm_commit,
2345};
2346
2347static int bdrv_drv_set_perm(BlockDriverState *bs, uint64_t perm,
2348 uint64_t shared_perm, Transaction *tran,
2349 Error **errp)
2350{
Emanuele Giuseppe Espositoda359902022-03-03 10:16:11 -05002351 GLOBAL_STATE_CODE();
Vladimir Sementsov-Ogievskiy2513ef52021-04-28 18:17:42 +03002352 if (!bs->drv) {
2353 return 0;
2354 }
2355
2356 if (bs->drv->bdrv_check_perm) {
2357 int ret = bs->drv->bdrv_check_perm(bs, perm, shared_perm, errp);
2358 if (ret < 0) {
2359 return ret;
2360 }
2361 }
2362
2363 if (tran) {
2364 tran_add(tran, &bdrv_drv_set_perm_drv, bs);
2365 }
2366
2367 return 0;
2368}
2369
Vladimir Sementsov-Ogievskiy09786232021-04-28 18:17:44 +03002370typedef struct BdrvReplaceChildState {
2371 BdrvChild *child;
2372 BlockDriverState *old_bs;
2373} BdrvReplaceChildState;
2374
2375static void bdrv_replace_child_commit(void *opaque)
2376{
2377 BdrvReplaceChildState *s = opaque;
Emanuele Giuseppe Espositobdb73472022-03-03 10:16:02 -05002378 GLOBAL_STATE_CODE();
Vladimir Sementsov-Ogievskiy09786232021-04-28 18:17:44 +03002379
2380 bdrv_unref(s->old_bs);
2381}
2382
2383static void bdrv_replace_child_abort(void *opaque)
2384{
2385 BdrvReplaceChildState *s = opaque;
2386 BlockDriverState *new_bs = s->child->bs;
2387
Emanuele Giuseppe Espositobdb73472022-03-03 10:16:02 -05002388 GLOBAL_STATE_CODE();
Vladimir Sementsov-Ogievskiy0f0b1e22022-07-26 23:11:29 +03002389 /* old_bs reference is transparently moved from @s to @s->child */
Kevin Wolf23987472022-11-18 18:41:09 +01002390 if (!s->child->bs) {
2391 /*
2392 * The parents were undrained when removing old_bs from the child. New
2393 * requests can't have been made, though, because the child was empty.
2394 *
2395 * TODO Make bdrv_replace_child_noperm() transactionable to avoid
2396 * undraining the parent in the first place. Once this is done, having
2397 * new_bs drained when calling bdrv_replace_child_tran() is not a
2398 * requirement any more.
2399 */
Kevin Wolf606ed752022-11-18 18:41:10 +01002400 bdrv_parent_drained_begin_single(s->child);
Kevin Wolf23987472022-11-18 18:41:09 +01002401 assert(!bdrv_parent_drained_poll_single(s->child));
2402 }
2403 assert(s->child->quiesced_parent);
Vladimir Sementsov-Ogievskiy544acc72022-07-26 23:11:31 +03002404 bdrv_replace_child_noperm(s->child, s->old_bs);
Vladimir Sementsov-Ogievskiy09786232021-04-28 18:17:44 +03002405 bdrv_unref(new_bs);
2406}
2407
2408static TransactionActionDrv bdrv_replace_child_drv = {
2409 .commit = bdrv_replace_child_commit,
2410 .abort = bdrv_replace_child_abort,
2411 .clean = g_free,
2412};
2413
2414/*
Vladimir Sementsov-Ogievskiy4bf021d2021-06-10 14:25:44 +03002415 * bdrv_replace_child_tran
Vladimir Sementsov-Ogievskiy09786232021-04-28 18:17:44 +03002416 *
2417 * Note: real unref of old_bs is done only on commit.
Vladimir Sementsov-Ogievskiy4bf021d2021-06-10 14:25:44 +03002418 *
Kevin Wolf23987472022-11-18 18:41:09 +01002419 * Both @child->bs and @new_bs (if non-NULL) must be drained. @new_bs must be
2420 * kept drained until the transaction is completed.
2421 *
Vladimir Sementsov-Ogievskiy4bf021d2021-06-10 14:25:44 +03002422 * The function doesn't update permissions, caller is responsible for this.
Vladimir Sementsov-Ogievskiy09786232021-04-28 18:17:44 +03002423 */
Vladimir Sementsov-Ogievskiy0f0b1e22022-07-26 23:11:29 +03002424static void bdrv_replace_child_tran(BdrvChild *child, BlockDriverState *new_bs,
Vladimir Sementsov-Ogievskiy4eba8252022-07-26 23:11:28 +03002425 Transaction *tran)
Vladimir Sementsov-Ogievskiy09786232021-04-28 18:17:44 +03002426{
2427 BdrvReplaceChildState *s = g_new(BdrvReplaceChildState, 1);
Kevin Wolf23987472022-11-18 18:41:09 +01002428
2429 assert(child->quiesced_parent);
2430 assert(!new_bs || new_bs->quiesce_counter);
2431
Vladimir Sementsov-Ogievskiy09786232021-04-28 18:17:44 +03002432 *s = (BdrvReplaceChildState) {
Vladimir Sementsov-Ogievskiy0f0b1e22022-07-26 23:11:29 +03002433 .child = child,
2434 .old_bs = child->bs,
Vladimir Sementsov-Ogievskiy09786232021-04-28 18:17:44 +03002435 };
2436 tran_add(tran, &bdrv_replace_child_drv, s);
2437
2438 if (new_bs) {
2439 bdrv_ref(new_bs);
2440 }
Vladimir Sementsov-Ogievskiy544acc72022-07-26 23:11:31 +03002441 bdrv_replace_child_noperm(child, new_bs);
Vladimir Sementsov-Ogievskiy0f0b1e22022-07-26 23:11:29 +03002442 /* old_bs reference is transparently moved from @child to @s */
Vladimir Sementsov-Ogievskiy09786232021-04-28 18:17:44 +03002443}
2444
Kevin Wolf33a610c2016-12-15 13:04:20 +01002445/*
Vladimir Sementsov-Ogievskiyc20555e2021-04-28 18:18:04 +03002446 * Refresh permissions in @bs subtree. The function is intended to be called
2447 * after some graph modification that was done without permission update.
Kevin Wolf33a610c2016-12-15 13:04:20 +01002448 */
Vladimir Sementsov-Ogievskiyc20555e2021-04-28 18:18:04 +03002449static int bdrv_node_refresh_perm(BlockDriverState *bs, BlockReopenQueue *q,
2450 Transaction *tran, Error **errp)
Kevin Wolf33a610c2016-12-15 13:04:20 +01002451{
2452 BlockDriver *drv = bs->drv;
2453 BdrvChild *c;
2454 int ret;
Vladimir Sementsov-Ogievskiyc20555e2021-04-28 18:18:04 +03002455 uint64_t cumulative_perms, cumulative_shared_perms;
Emanuele Giuseppe Esposito862fded2022-03-03 10:15:55 -05002456 GLOBAL_STATE_CODE();
Vladimir Sementsov-Ogievskiyc20555e2021-04-28 18:18:04 +03002457
2458 bdrv_get_cumulative_perm(bs, &cumulative_perms, &cumulative_shared_perms);
Kevin Wolf33a610c2016-12-15 13:04:20 +01002459
2460 /* Write permissions never work with read-only images */
2461 if ((cumulative_perms & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)) &&
Max Reitzcc022142018-06-06 21:37:00 +02002462 !bdrv_is_writable_after_reopen(bs, q))
Kevin Wolf33a610c2016-12-15 13:04:20 +01002463 {
Max Reitz481e0ee2019-05-15 22:15:00 +02002464 if (!bdrv_is_writable_after_reopen(bs, NULL)) {
2465 error_setg(errp, "Block node is read-only");
2466 } else {
Vladimir Sementsov-Ogievskiyc20555e2021-04-28 18:18:04 +03002467 error_setg(errp, "Read-only block node '%s' cannot support "
2468 "read-write users", bdrv_get_node_name(bs));
Max Reitz481e0ee2019-05-15 22:15:00 +02002469 }
2470
Kevin Wolf33a610c2016-12-15 13:04:20 +01002471 return -EPERM;
2472 }
2473
Kevin Wolf9c60a5d2020-07-16 16:26:00 +02002474 /*
2475 * Unaligned requests will automatically be aligned to bl.request_alignment
2476 * and without RESIZE we can't extend requests to write to space beyond the
2477 * end of the image, so it's required that the image size is aligned.
2478 */
2479 if ((cumulative_perms & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)) &&
2480 !(cumulative_perms & BLK_PERM_RESIZE))
2481 {
2482 if ((bs->total_sectors * BDRV_SECTOR_SIZE) % bs->bl.request_alignment) {
2483 error_setg(errp, "Cannot get 'write' permission without 'resize': "
2484 "Image size is not a multiple of request "
2485 "alignment");
2486 return -EPERM;
2487 }
2488 }
2489
Kevin Wolf33a610c2016-12-15 13:04:20 +01002490 /* Check this node */
2491 if (!drv) {
2492 return 0;
2493 }
2494
Vladimir Sementsov-Ogievskiyb1d2bbe2021-04-28 18:17:43 +03002495 ret = bdrv_drv_set_perm(bs, cumulative_perms, cumulative_shared_perms, tran,
Vladimir Sementsov-Ogievskiy2513ef52021-04-28 18:17:42 +03002496 errp);
2497 if (ret < 0) {
2498 return ret;
Kevin Wolf33a610c2016-12-15 13:04:20 +01002499 }
2500
Kevin Wolf78e421c2016-12-20 23:25:12 +01002501 /* Drivers that never have children can omit .bdrv_child_perm() */
Kevin Wolf33a610c2016-12-15 13:04:20 +01002502 if (!drv->bdrv_child_perm) {
Kevin Wolf78e421c2016-12-20 23:25:12 +01002503 assert(QLIST_EMPTY(&bs->children));
Kevin Wolf33a610c2016-12-15 13:04:20 +01002504 return 0;
2505 }
2506
2507 /* Check all children */
2508 QLIST_FOREACH(c, &bs->children, next) {
2509 uint64_t cur_perm, cur_shared;
Max Reitz9eab1542019-05-22 19:03:50 +02002510
Max Reitze5d8a402020-05-13 13:05:44 +02002511 bdrv_child_perm(bs, c->bs, c, c->role, q,
Fam Zhengffd1a5a2017-05-03 00:35:38 +08002512 cumulative_perms, cumulative_shared_perms,
2513 &cur_perm, &cur_shared);
Vladimir Sementsov-Ogievskiyecb776b2021-04-28 18:18:02 +03002514 bdrv_child_set_perm(c, cur_perm, cur_shared, tran);
Vladimir Sementsov-Ogievskiyb1d2bbe2021-04-28 18:17:43 +03002515 }
2516
2517 return 0;
2518}
2519
Vladimir Sementsov-Ogievskiyfb0ff4d2022-11-07 19:35:58 +03002520/*
2521 * @list is a product of bdrv_topological_dfs() (may be called several times) -
2522 * a topologically sorted subgraph.
2523 */
2524static int bdrv_do_refresh_perms(GSList *list, BlockReopenQueue *q,
2525 Transaction *tran, Error **errp)
Vladimir Sementsov-Ogievskiyb1d2bbe2021-04-28 18:17:43 +03002526{
2527 int ret;
2528 BlockDriverState *bs;
Emanuele Giuseppe Esposito862fded2022-03-03 10:15:55 -05002529 GLOBAL_STATE_CODE();
Vladimir Sementsov-Ogievskiyb1d2bbe2021-04-28 18:17:43 +03002530
Vladimir Sementsov-Ogievskiyb1d2bbe2021-04-28 18:17:43 +03002531 for ( ; list; list = list->next) {
2532 bs = list->data;
2533
Vladimir Sementsov-Ogievskiy9397c142021-04-28 18:17:53 +03002534 if (bdrv_parent_perms_conflict(bs, errp)) {
Vladimir Sementsov-Ogievskiyb1d2bbe2021-04-28 18:17:43 +03002535 return -EINVAL;
2536 }
2537
Vladimir Sementsov-Ogievskiyc20555e2021-04-28 18:18:04 +03002538 ret = bdrv_node_refresh_perm(bs, q, tran, errp);
Vladimir Sementsov-Ogievskiyb1d2bbe2021-04-28 18:17:43 +03002539 if (ret < 0) {
2540 return ret;
2541 }
Vladimir Sementsov-Ogievskiybd57f8f2021-04-28 18:17:41 +03002542 }
Vladimir Sementsov-Ogievskiy3ef45e02021-04-28 18:17:40 +03002543
Vladimir Sementsov-Ogievskiybd57f8f2021-04-28 18:17:41 +03002544 return 0;
2545}
2546
Vladimir Sementsov-Ogievskiyfb0ff4d2022-11-07 19:35:58 +03002547/*
2548 * @list is any list of nodes. List is completed by all subtrees and
2549 * topologically sorted. It's not a problem if some node occurs in the @list
2550 * several times.
2551 */
2552static int bdrv_list_refresh_perms(GSList *list, BlockReopenQueue *q,
2553 Transaction *tran, Error **errp)
2554{
2555 g_autoptr(GHashTable) found = g_hash_table_new(NULL, NULL);
2556 g_autoptr(GSList) refresh_list = NULL;
2557
2558 for ( ; list; list = list->next) {
2559 refresh_list = bdrv_topological_dfs(refresh_list, found, list->data);
2560 }
2561
2562 return bdrv_do_refresh_perms(refresh_list, q, tran, errp);
2563}
2564
Kevin Wolfc7a0f2b2020-03-10 12:38:25 +01002565void bdrv_get_cumulative_perm(BlockDriverState *bs, uint64_t *perm,
2566 uint64_t *shared_perm)
Kevin Wolf33a610c2016-12-15 13:04:20 +01002567{
2568 BdrvChild *c;
2569 uint64_t cumulative_perms = 0;
2570 uint64_t cumulative_shared_perms = BLK_PERM_ALL;
2571
Emanuele Giuseppe Espositob4ad82a2022-03-03 10:15:57 -05002572 GLOBAL_STATE_CODE();
2573
Kevin Wolf33a610c2016-12-15 13:04:20 +01002574 QLIST_FOREACH(c, &bs->parents, next_parent) {
2575 cumulative_perms |= c->perm;
2576 cumulative_shared_perms &= c->shared_perm;
2577 }
2578
2579 *perm = cumulative_perms;
2580 *shared_perm = cumulative_shared_perms;
2581}
2582
Fam Zheng51761962017-05-03 00:35:36 +08002583char *bdrv_perm_names(uint64_t perm)
Kevin Wolfd0833192017-01-16 18:26:20 +01002584{
2585 struct perm_name {
2586 uint64_t perm;
2587 const char *name;
2588 } permissions[] = {
2589 { BLK_PERM_CONSISTENT_READ, "consistent read" },
2590 { BLK_PERM_WRITE, "write" },
2591 { BLK_PERM_WRITE_UNCHANGED, "write unchanged" },
2592 { BLK_PERM_RESIZE, "resize" },
Kevin Wolfd0833192017-01-16 18:26:20 +01002593 { 0, NULL }
2594 };
2595
Alberto Garciae2a74232020-01-10 18:15:18 +01002596 GString *result = g_string_sized_new(30);
Kevin Wolfd0833192017-01-16 18:26:20 +01002597 struct perm_name *p;
2598
2599 for (p = permissions; p->name; p++) {
2600 if (perm & p->perm) {
Alberto Garciae2a74232020-01-10 18:15:18 +01002601 if (result->len > 0) {
2602 g_string_append(result, ", ");
2603 }
2604 g_string_append(result, p->name);
Kevin Wolfd0833192017-01-16 18:26:20 +01002605 }
2606 }
2607
Alberto Garciae2a74232020-01-10 18:15:18 +01002608 return g_string_free(result, FALSE);
Kevin Wolfd0833192017-01-16 18:26:20 +01002609}
2610
Kevin Wolf33a610c2016-12-15 13:04:20 +01002611
Vladimir Sementsov-Ogievskiyf1316ed2022-11-07 19:35:57 +03002612/* @tran is allowed to be NULL. In this case no rollback is possible */
2613static int bdrv_refresh_perms(BlockDriverState *bs, Transaction *tran,
2614 Error **errp)
Vladimir Sementsov-Ogievskiybb87e4d2020-11-06 15:42:38 +03002615{
2616 int ret;
Vladimir Sementsov-Ogievskiyf1316ed2022-11-07 19:35:57 +03002617 Transaction *local_tran = NULL;
Vladimir Sementsov-Ogievskiyb1d2bbe2021-04-28 18:17:43 +03002618 g_autoptr(GSList) list = bdrv_topological_dfs(NULL, NULL, bs);
Emanuele Giuseppe Esposito862fded2022-03-03 10:15:55 -05002619 GLOBAL_STATE_CODE();
Vladimir Sementsov-Ogievskiybb87e4d2020-11-06 15:42:38 +03002620
Vladimir Sementsov-Ogievskiyf1316ed2022-11-07 19:35:57 +03002621 if (!tran) {
2622 tran = local_tran = tran_new();
2623 }
2624
Vladimir Sementsov-Ogievskiyfb0ff4d2022-11-07 19:35:58 +03002625 ret = bdrv_do_refresh_perms(list, NULL, tran, errp);
Vladimir Sementsov-Ogievskiyf1316ed2022-11-07 19:35:57 +03002626
2627 if (local_tran) {
2628 tran_finalize(local_tran, ret);
2629 }
Vladimir Sementsov-Ogievskiybb87e4d2020-11-06 15:42:38 +03002630
Vladimir Sementsov-Ogievskiyb1d2bbe2021-04-28 18:17:43 +03002631 return ret;
Vladimir Sementsov-Ogievskiybb87e4d2020-11-06 15:42:38 +03002632}
2633
Kevin Wolf33a610c2016-12-15 13:04:20 +01002634int bdrv_child_try_set_perm(BdrvChild *c, uint64_t perm, uint64_t shared,
2635 Error **errp)
2636{
Max Reitz10467792019-05-22 19:03:51 +02002637 Error *local_err = NULL;
Vladimir Sementsov-Ogievskiy83928dc2021-04-28 18:17:39 +03002638 Transaction *tran = tran_new();
Kevin Wolf33a610c2016-12-15 13:04:20 +01002639 int ret;
2640
Emanuele Giuseppe Espositob4ad82a2022-03-03 10:15:57 -05002641 GLOBAL_STATE_CODE();
2642
Vladimir Sementsov-Ogievskiyecb776b2021-04-28 18:18:02 +03002643 bdrv_child_set_perm(c, perm, shared, tran);
Vladimir Sementsov-Ogievskiy83928dc2021-04-28 18:17:39 +03002644
Vladimir Sementsov-Ogievskiyf1316ed2022-11-07 19:35:57 +03002645 ret = bdrv_refresh_perms(c->bs, tran, &local_err);
Vladimir Sementsov-Ogievskiy83928dc2021-04-28 18:17:39 +03002646
2647 tran_finalize(tran, ret);
2648
Kevin Wolf33a610c2016-12-15 13:04:20 +01002649 if (ret < 0) {
Vladimir Sementsov-Ogievskiy071b4742020-11-06 15:42:41 +03002650 if ((perm & ~c->perm) || (c->shared_perm & ~shared)) {
2651 /* tighten permissions */
Max Reitz10467792019-05-22 19:03:51 +02002652 error_propagate(errp, local_err);
2653 } else {
2654 /*
2655 * Our caller may intend to only loosen restrictions and
2656 * does not expect this function to fail. Errors are not
2657 * fatal in such a case, so we can just hide them from our
2658 * caller.
2659 */
2660 error_free(local_err);
2661 ret = 0;
2662 }
Kevin Wolf33a610c2016-12-15 13:04:20 +01002663 }
2664
Vladimir Sementsov-Ogievskiy83928dc2021-04-28 18:17:39 +03002665 return ret;
Kevin Wolfd5e6f432016-12-14 17:24:36 +01002666}
2667
Max Reitzc1087f12019-05-22 19:03:46 +02002668int bdrv_child_refresh_perms(BlockDriverState *bs, BdrvChild *c, Error **errp)
2669{
2670 uint64_t parent_perms, parent_shared;
2671 uint64_t perms, shared;
2672
Emanuele Giuseppe Espositob4ad82a2022-03-03 10:15:57 -05002673 GLOBAL_STATE_CODE();
2674
Max Reitzc1087f12019-05-22 19:03:46 +02002675 bdrv_get_cumulative_perm(bs, &parent_perms, &parent_shared);
Max Reitze5d8a402020-05-13 13:05:44 +02002676 bdrv_child_perm(bs, c->bs, c, c->role, NULL,
Max Reitzbf8e9252020-05-13 13:05:16 +02002677 parent_perms, parent_shared, &perms, &shared);
Max Reitzc1087f12019-05-22 19:03:46 +02002678
2679 return bdrv_child_try_set_perm(c, perms, shared, errp);
2680}
2681
Max Reitz87278af2020-05-13 13:05:40 +02002682/*
2683 * Default implementation for .bdrv_child_perm() for block filters:
2684 * Forward CONSISTENT_READ, WRITE, WRITE_UNCHANGED, and RESIZE to the
2685 * filtered child.
2686 */
2687static void bdrv_filter_default_perms(BlockDriverState *bs, BdrvChild *c,
Max Reitz87278af2020-05-13 13:05:40 +02002688 BdrvChildRole role,
2689 BlockReopenQueue *reopen_queue,
2690 uint64_t perm, uint64_t shared,
2691 uint64_t *nperm, uint64_t *nshared)
Kevin Wolf6a1b9ee2016-12-15 11:27:32 +01002692{
Emanuele Giuseppe Esposito862fded2022-03-03 10:15:55 -05002693 GLOBAL_STATE_CODE();
Kevin Wolfe444fa82019-08-02 15:59:41 +02002694 *nperm = perm & DEFAULT_PERM_PASSTHROUGH;
2695 *nshared = (shared & DEFAULT_PERM_PASSTHROUGH) | DEFAULT_PERM_UNCHANGED;
Kevin Wolf6a1b9ee2016-12-15 11:27:32 +01002696}
2697
Max Reitz70082db2020-05-13 13:05:26 +02002698static void bdrv_default_perms_for_cow(BlockDriverState *bs, BdrvChild *c,
Max Reitz70082db2020-05-13 13:05:26 +02002699 BdrvChildRole role,
2700 BlockReopenQueue *reopen_queue,
2701 uint64_t perm, uint64_t shared,
2702 uint64_t *nperm, uint64_t *nshared)
2703{
Max Reitze5d8a402020-05-13 13:05:44 +02002704 assert(role & BDRV_CHILD_COW);
Emanuele Giuseppe Esposito862fded2022-03-03 10:15:55 -05002705 GLOBAL_STATE_CODE();
Max Reitz70082db2020-05-13 13:05:26 +02002706
2707 /*
2708 * We want consistent read from backing files if the parent needs it.
2709 * No other operations are performed on backing files.
2710 */
2711 perm &= BLK_PERM_CONSISTENT_READ;
2712
2713 /*
2714 * If the parent can deal with changing data, we're okay with a
2715 * writable and resizable backing file.
2716 * TODO Require !(perm & BLK_PERM_CONSISTENT_READ), too?
2717 */
2718 if (shared & BLK_PERM_WRITE) {
2719 shared = BLK_PERM_WRITE | BLK_PERM_RESIZE;
2720 } else {
2721 shared = 0;
2722 }
2723
Vladimir Sementsov-Ogievskiy64631f32021-09-02 12:37:54 +03002724 shared |= BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE_UNCHANGED;
Max Reitz70082db2020-05-13 13:05:26 +02002725
2726 if (bs->open_flags & BDRV_O_INACTIVE) {
2727 shared |= BLK_PERM_WRITE | BLK_PERM_RESIZE;
2728 }
2729
2730 *nperm = perm;
2731 *nshared = shared;
2732}
2733
Max Reitz6f838a42020-05-13 13:05:27 +02002734static void bdrv_default_perms_for_storage(BlockDriverState *bs, BdrvChild *c,
Max Reitz6f838a42020-05-13 13:05:27 +02002735 BdrvChildRole role,
2736 BlockReopenQueue *reopen_queue,
2737 uint64_t perm, uint64_t shared,
2738 uint64_t *nperm, uint64_t *nshared)
2739{
2740 int flags;
2741
Emanuele Giuseppe Esposito862fded2022-03-03 10:15:55 -05002742 GLOBAL_STATE_CODE();
Max Reitze5d8a402020-05-13 13:05:44 +02002743 assert(role & (BDRV_CHILD_METADATA | BDRV_CHILD_DATA));
Max Reitz6f838a42020-05-13 13:05:27 +02002744
2745 flags = bdrv_reopen_get_flags(reopen_queue, bs);
2746
2747 /*
2748 * Apart from the modifications below, the same permissions are
2749 * forwarded and left alone as for filters
2750 */
Max Reitze5d8a402020-05-13 13:05:44 +02002751 bdrv_filter_default_perms(bs, c, role, reopen_queue,
Max Reitz6f838a42020-05-13 13:05:27 +02002752 perm, shared, &perm, &shared);
2753
Max Reitzf8890542020-05-13 13:05:28 +02002754 if (role & BDRV_CHILD_METADATA) {
2755 /* Format drivers may touch metadata even if the guest doesn't write */
2756 if (bdrv_is_writable_after_reopen(bs, reopen_queue)) {
2757 perm |= BLK_PERM_WRITE | BLK_PERM_RESIZE;
2758 }
2759
2760 /*
2761 * bs->file always needs to be consistent because of the
2762 * metadata. We can never allow other users to resize or write
2763 * to it.
2764 */
2765 if (!(flags & BDRV_O_NO_IO)) {
2766 perm |= BLK_PERM_CONSISTENT_READ;
2767 }
2768 shared &= ~(BLK_PERM_WRITE | BLK_PERM_RESIZE);
Max Reitz6f838a42020-05-13 13:05:27 +02002769 }
2770
Max Reitzf8890542020-05-13 13:05:28 +02002771 if (role & BDRV_CHILD_DATA) {
2772 /*
2773 * Technically, everything in this block is a subset of the
2774 * BDRV_CHILD_METADATA path taken above, and so this could
2775 * be an "else if" branch. However, that is not obvious, and
2776 * this function is not performance critical, therefore we let
2777 * this be an independent "if".
2778 */
2779
2780 /*
2781 * We cannot allow other users to resize the file because the
2782 * format driver might have some assumptions about the size
2783 * (e.g. because it is stored in metadata, or because the file
2784 * is split into fixed-size data files).
2785 */
2786 shared &= ~BLK_PERM_RESIZE;
2787
2788 /*
2789 * WRITE_UNCHANGED often cannot be performed as such on the
2790 * data file. For example, the qcow2 driver may still need to
2791 * write copied clusters on copy-on-read.
2792 */
2793 if (perm & BLK_PERM_WRITE_UNCHANGED) {
2794 perm |= BLK_PERM_WRITE;
2795 }
2796
2797 /*
2798 * If the data file is written to, the format driver may
2799 * expect to be able to resize it by writing beyond the EOF.
2800 */
2801 if (perm & BLK_PERM_WRITE) {
2802 perm |= BLK_PERM_RESIZE;
2803 }
Max Reitz6f838a42020-05-13 13:05:27 +02002804 }
Max Reitz6f838a42020-05-13 13:05:27 +02002805
2806 if (bs->open_flags & BDRV_O_INACTIVE) {
2807 shared |= BLK_PERM_WRITE | BLK_PERM_RESIZE;
2808 }
2809
2810 *nperm = perm;
2811 *nshared = shared;
2812}
2813
Max Reitz2519f542020-05-13 13:05:29 +02002814void bdrv_default_perms(BlockDriverState *bs, BdrvChild *c,
Max Reitze5d8a402020-05-13 13:05:44 +02002815 BdrvChildRole role, BlockReopenQueue *reopen_queue,
Max Reitz2519f542020-05-13 13:05:29 +02002816 uint64_t perm, uint64_t shared,
2817 uint64_t *nperm, uint64_t *nshared)
2818{
Emanuele Giuseppe Espositob4ad82a2022-03-03 10:15:57 -05002819 GLOBAL_STATE_CODE();
Max Reitz2519f542020-05-13 13:05:29 +02002820 if (role & BDRV_CHILD_FILTERED) {
2821 assert(!(role & (BDRV_CHILD_DATA | BDRV_CHILD_METADATA |
2822 BDRV_CHILD_COW)));
Max Reitze5d8a402020-05-13 13:05:44 +02002823 bdrv_filter_default_perms(bs, c, role, reopen_queue,
Max Reitz2519f542020-05-13 13:05:29 +02002824 perm, shared, nperm, nshared);
2825 } else if (role & BDRV_CHILD_COW) {
2826 assert(!(role & (BDRV_CHILD_DATA | BDRV_CHILD_METADATA)));
Max Reitze5d8a402020-05-13 13:05:44 +02002827 bdrv_default_perms_for_cow(bs, c, role, reopen_queue,
Max Reitz2519f542020-05-13 13:05:29 +02002828 perm, shared, nperm, nshared);
2829 } else if (role & (BDRV_CHILD_METADATA | BDRV_CHILD_DATA)) {
Max Reitze5d8a402020-05-13 13:05:44 +02002830 bdrv_default_perms_for_storage(bs, c, role, reopen_queue,
Max Reitz2519f542020-05-13 13:05:29 +02002831 perm, shared, nperm, nshared);
2832 } else {
2833 g_assert_not_reached();
2834 }
2835}
2836
Max Reitz7b1d9c42019-11-08 13:34:51 +01002837uint64_t bdrv_qapi_perm_to_blk_perm(BlockPermission qapi_perm)
2838{
2839 static const uint64_t permissions[] = {
2840 [BLOCK_PERMISSION_CONSISTENT_READ] = BLK_PERM_CONSISTENT_READ,
2841 [BLOCK_PERMISSION_WRITE] = BLK_PERM_WRITE,
2842 [BLOCK_PERMISSION_WRITE_UNCHANGED] = BLK_PERM_WRITE_UNCHANGED,
2843 [BLOCK_PERMISSION_RESIZE] = BLK_PERM_RESIZE,
Max Reitz7b1d9c42019-11-08 13:34:51 +01002844 };
2845
2846 QEMU_BUILD_BUG_ON(ARRAY_SIZE(permissions) != BLOCK_PERMISSION__MAX);
2847 QEMU_BUILD_BUG_ON(1UL << ARRAY_SIZE(permissions) != BLK_PERM_ALL + 1);
2848
2849 assert(qapi_perm < BLOCK_PERMISSION__MAX);
2850
2851 return permissions[qapi_perm];
2852}
2853
Kevin Wolf23987472022-11-18 18:41:09 +01002854/*
2855 * Replaces the node that a BdrvChild points to without updating permissions.
2856 *
2857 * If @new_bs is non-NULL, the parent of @child must already be drained through
Kevin Wolf31b2ddf2023-06-05 10:57:10 +02002858 * @child and the caller must hold the AioContext lock for @new_bs.
Kevin Wolf23987472022-11-18 18:41:09 +01002859 */
Vladimir Sementsov-Ogievskiy544acc72022-07-26 23:11:31 +03002860static void bdrv_replace_child_noperm(BdrvChild *child,
Vladimir Sementsov-Ogievskiy4eba8252022-07-26 23:11:28 +03002861 BlockDriverState *new_bs)
Kevin Wolfe9740bc2016-05-23 15:52:26 +02002862{
2863 BlockDriverState *old_bs = child->bs;
Max Reitzdebc2922019-07-22 15:33:44 +02002864 int new_bs_quiesce_counter;
Kevin Wolfe9740bc2016-05-23 15:52:26 +02002865
Alberto Garcia2cad1eb2019-03-12 18:48:40 +02002866 assert(!child->frozen);
Kevin Wolf23987472022-11-18 18:41:09 +01002867
2868 /*
2869 * If we want to change the BdrvChild to point to a drained node as its new
2870 * child->bs, we need to make sure that its new parent is drained, too. In
2871 * other words, either child->quiesce_parent must already be true or we must
2872 * be able to set it and keep the parent's quiesce_counter consistent with
2873 * that, but without polling or starting new requests (this function
2874 * guarantees that it doesn't poll, and starting new requests would be
2875 * against the invariants of drain sections).
2876 *
2877 * To keep things simple, we pick the first option (child->quiesce_parent
2878 * must already be true). We also generalise the rule a bit to make it
2879 * easier to verify in callers and more likely to be covered in test cases:
2880 * The parent must be quiesced through this child even if new_bs isn't
2881 * currently drained.
2882 *
2883 * The only exception is for callers that always pass new_bs == NULL. In
2884 * this case, we obviously never need to consider the case of a drained
2885 * new_bs, so we can keep the callers simpler by allowing them not to drain
2886 * the parent.
2887 */
2888 assert(!new_bs || child->quiesced_parent);
Kevin Wolfbfb8aa62021-10-18 15:47:14 +02002889 assert(old_bs != new_bs);
Emanuele Giuseppe Espositof0c28322022-03-03 10:16:13 -05002890 GLOBAL_STATE_CODE();
Alberto Garcia2cad1eb2019-03-12 18:48:40 +02002891
Fam Zhengbb2614e2017-04-07 14:54:10 +08002892 if (old_bs && new_bs) {
2893 assert(bdrv_get_aio_context(old_bs) == bdrv_get_aio_context(new_bs));
2894 }
Max Reitzdebc2922019-07-22 15:33:44 +02002895
Emanuele Giuseppe Esposito29312502022-12-07 14:18:31 +01002896 /* TODO Pull this up into the callers to avoid polling here */
Kevin Wolf31b2ddf2023-06-05 10:57:10 +02002897 bdrv_graph_wrlock(new_bs);
Kevin Wolfe9740bc2016-05-23 15:52:26 +02002898 if (old_bs) {
Max Reitzbd86fb92020-05-13 13:05:13 +02002899 if (child->klass->detach) {
2900 child->klass->detach(child);
Kevin Wolfd736f112017-12-18 16:05:48 +01002901 }
Kevin Wolfe9740bc2016-05-23 15:52:26 +02002902 QLIST_REMOVE(child, next_parent);
2903 }
Kevin Wolfe9740bc2016-05-23 15:52:26 +02002904
2905 child->bs = new_bs;
Kevin Wolf36fe1332016-05-17 14:51:55 +02002906
2907 if (new_bs) {
2908 QLIST_INSERT_HEAD(&new_bs->parents, child, next_parent);
Max Reitzbd86fb92020-05-13 13:05:13 +02002909 if (child->klass->attach) {
2910 child->klass->attach(child);
Kevin Wolfdb95dbb2017-02-08 11:28:52 +01002911 }
Kevin Wolf36fe1332016-05-17 14:51:55 +02002912 }
Emanuele Giuseppe Esposito29312502022-12-07 14:18:31 +01002913 bdrv_graph_wrunlock();
Max Reitzdebc2922019-07-22 15:33:44 +02002914
2915 /*
Kevin Wolf23987472022-11-18 18:41:09 +01002916 * If the parent was drained through this BdrvChild previously, but new_bs
2917 * is not drained, allow requests to come in only after the new node has
2918 * been attached.
Max Reitzdebc2922019-07-22 15:33:44 +02002919 */
Kevin Wolf57e05be2022-11-18 18:41:06 +01002920 new_bs_quiesce_counter = (new_bs ? new_bs->quiesce_counter : 0);
2921 if (!new_bs_quiesce_counter && child->quiesced_parent) {
Max Reitzdebc2922019-07-22 15:33:44 +02002922 bdrv_parent_drained_end_single(child);
Max Reitzdebc2922019-07-22 15:33:44 +02002923 }
Kevin Wolfe9740bc2016-05-23 15:52:26 +02002924}
2925
Hanna Reitz04c9c3a2021-11-15 15:53:59 +01002926/**
2927 * Free the given @child.
2928 *
2929 * The child must be empty (i.e. `child->bs == NULL`) and it must be
2930 * unused (i.e. not in a children list).
2931 */
2932static void bdrv_child_free(BdrvChild *child)
Vladimir Sementsov-Ogievskiy548a74c2021-04-28 18:17:46 +03002933{
2934 assert(!child->bs);
Emanuele Giuseppe Espositobdb73472022-03-03 10:16:02 -05002935 GLOBAL_STATE_CODE();
Hanna Reitza2253692021-11-15 15:53:58 +01002936 assert(!child->next.le_prev); /* not in children list */
Hanna Reitz04c9c3a2021-11-15 15:53:59 +01002937
2938 g_free(child->name);
2939 g_free(child);
Vladimir Sementsov-Ogievskiy548a74c2021-04-28 18:17:46 +03002940}
2941
2942typedef struct BdrvAttachChildCommonState {
Vladimir Sementsov-Ogievskiy5bb047472022-07-26 23:11:32 +03002943 BdrvChild *child;
Vladimir Sementsov-Ogievskiy548a74c2021-04-28 18:17:46 +03002944 AioContext *old_parent_ctx;
2945 AioContext *old_child_ctx;
2946} BdrvAttachChildCommonState;
2947
2948static void bdrv_attach_child_common_abort(void *opaque)
2949{
2950 BdrvAttachChildCommonState *s = opaque;
Vladimir Sementsov-Ogievskiy5bb047472022-07-26 23:11:32 +03002951 BlockDriverState *bs = s->child->bs;
Vladimir Sementsov-Ogievskiy548a74c2021-04-28 18:17:46 +03002952
Emanuele Giuseppe Espositof0c28322022-03-03 10:16:13 -05002953 GLOBAL_STATE_CODE();
Vladimir Sementsov-Ogievskiy5bb047472022-07-26 23:11:32 +03002954 bdrv_replace_child_noperm(s->child, NULL);
Vladimir Sementsov-Ogievskiy548a74c2021-04-28 18:17:46 +03002955
2956 if (bdrv_get_aio_context(bs) != s->old_child_ctx) {
Emanuele Giuseppe Esposito142e6902022-10-25 04:49:52 -04002957 bdrv_try_change_aio_context(bs, s->old_child_ctx, NULL, &error_abort);
Vladimir Sementsov-Ogievskiy548a74c2021-04-28 18:17:46 +03002958 }
2959
Vladimir Sementsov-Ogievskiy5bb047472022-07-26 23:11:32 +03002960 if (bdrv_child_get_parent_aio_context(s->child) != s->old_parent_ctx) {
Emanuele Giuseppe Espositof8be48a2022-10-25 04:49:49 -04002961 Transaction *tran;
2962 GHashTable *visited;
2963 bool ret;
Vladimir Sementsov-Ogievskiy548a74c2021-04-28 18:17:46 +03002964
Emanuele Giuseppe Espositof8be48a2022-10-25 04:49:49 -04002965 tran = tran_new();
Vladimir Sementsov-Ogievskiy548a74c2021-04-28 18:17:46 +03002966
Emanuele Giuseppe Espositof8be48a2022-10-25 04:49:49 -04002967 /* No need to visit `child`, because it has been detached already */
2968 visited = g_hash_table_new(NULL, NULL);
2969 ret = s->child->klass->change_aio_ctx(s->child, s->old_parent_ctx,
2970 visited, tran, &error_abort);
2971 g_hash_table_destroy(visited);
2972
2973 /* transaction is supposed to always succeed */
2974 assert(ret == true);
2975 tran_commit(tran);
Vladimir Sementsov-Ogievskiy548a74c2021-04-28 18:17:46 +03002976 }
2977
2978 bdrv_unref(bs);
Vladimir Sementsov-Ogievskiy5bb047472022-07-26 23:11:32 +03002979 bdrv_child_free(s->child);
Vladimir Sementsov-Ogievskiy548a74c2021-04-28 18:17:46 +03002980}
2981
2982static TransactionActionDrv bdrv_attach_child_common_drv = {
2983 .abort = bdrv_attach_child_common_abort,
2984 .clean = g_free,
2985};
2986
2987/*
2988 * Common part of attaching bdrv child to bs or to blk or to job
Vladimir Sementsov-Ogievskiyf8d2ad72021-06-01 10:52:13 +03002989 *
Vladimir Sementsov-Ogievskiy7ec390d2021-06-10 14:25:45 +03002990 * Function doesn't update permissions, caller is responsible for this.
Vladimir Sementsov-Ogievskiy5bb047472022-07-26 23:11:32 +03002991 *
2992 * Returns new created child.
Kevin Wolfc066e802023-06-05 10:57:05 +02002993 *
2994 * The caller must hold the AioContext lock for @child_bs. Both @parent_bs and
2995 * @child_bs can move to a different AioContext in this function. Callers must
2996 * make sure that their AioContext locking is still correct after this.
Vladimir Sementsov-Ogievskiy548a74c2021-04-28 18:17:46 +03002997 */
Vladimir Sementsov-Ogievskiy5bb047472022-07-26 23:11:32 +03002998static BdrvChild *bdrv_attach_child_common(BlockDriverState *child_bs,
2999 const char *child_name,
3000 const BdrvChildClass *child_class,
3001 BdrvChildRole child_role,
3002 uint64_t perm, uint64_t shared_perm,
3003 void *opaque,
3004 Transaction *tran, Error **errp)
Vladimir Sementsov-Ogievskiy548a74c2021-04-28 18:17:46 +03003005{
3006 BdrvChild *new_child;
Kevin Wolfc066e802023-06-05 10:57:05 +02003007 AioContext *parent_ctx, *new_child_ctx;
Vladimir Sementsov-Ogievskiy548a74c2021-04-28 18:17:46 +03003008 AioContext *child_ctx = bdrv_get_aio_context(child_bs);
3009
Vladimir Sementsov-Ogievskiyda261b62021-06-01 10:52:17 +03003010 assert(child_class->get_parent_desc);
Emanuele Giuseppe Espositobdb73472022-03-03 10:16:02 -05003011 GLOBAL_STATE_CODE();
Vladimir Sementsov-Ogievskiy548a74c2021-04-28 18:17:46 +03003012
3013 new_child = g_new(BdrvChild, 1);
3014 *new_child = (BdrvChild) {
3015 .bs = NULL,
3016 .name = g_strdup(child_name),
3017 .klass = child_class,
3018 .role = child_role,
3019 .perm = perm,
3020 .shared_perm = shared_perm,
3021 .opaque = opaque,
3022 };
3023
3024 /*
3025 * If the AioContexts don't match, first try to move the subtree of
3026 * child_bs into the AioContext of the new parent. If this doesn't work,
3027 * try moving the parent into the AioContext of child_bs instead.
3028 */
3029 parent_ctx = bdrv_child_get_parent_aio_context(new_child);
3030 if (child_ctx != parent_ctx) {
3031 Error *local_err = NULL;
Emanuele Giuseppe Esposito142e6902022-10-25 04:49:52 -04003032 int ret = bdrv_try_change_aio_context(child_bs, parent_ctx, NULL,
3033 &local_err);
Vladimir Sementsov-Ogievskiy548a74c2021-04-28 18:17:46 +03003034
Emanuele Giuseppe Espositof8be48a2022-10-25 04:49:49 -04003035 if (ret < 0 && child_class->change_aio_ctx) {
3036 Transaction *tran = tran_new();
3037 GHashTable *visited = g_hash_table_new(NULL, NULL);
3038 bool ret_child;
3039
3040 g_hash_table_add(visited, new_child);
3041 ret_child = child_class->change_aio_ctx(new_child, child_ctx,
3042 visited, tran, NULL);
3043 if (ret_child == true) {
Vladimir Sementsov-Ogievskiy548a74c2021-04-28 18:17:46 +03003044 error_free(local_err);
3045 ret = 0;
Vladimir Sementsov-Ogievskiy548a74c2021-04-28 18:17:46 +03003046 }
Emanuele Giuseppe Espositof8be48a2022-10-25 04:49:49 -04003047 tran_finalize(tran, ret_child == true ? 0 : -1);
3048 g_hash_table_destroy(visited);
Vladimir Sementsov-Ogievskiy548a74c2021-04-28 18:17:46 +03003049 }
3050
3051 if (ret < 0) {
3052 error_propagate(errp, local_err);
Hanna Reitz04c9c3a2021-11-15 15:53:59 +01003053 bdrv_child_free(new_child);
Vladimir Sementsov-Ogievskiy5bb047472022-07-26 23:11:32 +03003054 return NULL;
Vladimir Sementsov-Ogievskiy548a74c2021-04-28 18:17:46 +03003055 }
3056 }
3057
Kevin Wolfc066e802023-06-05 10:57:05 +02003058 new_child_ctx = bdrv_get_aio_context(child_bs);
3059 if (new_child_ctx != child_ctx) {
3060 aio_context_release(child_ctx);
3061 aio_context_acquire(new_child_ctx);
3062 }
3063
Vladimir Sementsov-Ogievskiy548a74c2021-04-28 18:17:46 +03003064 bdrv_ref(child_bs);
Kevin Wolf23987472022-11-18 18:41:09 +01003065 /*
3066 * Let every new BdrvChild start with a drained parent. Inserting the child
3067 * in the graph with bdrv_replace_child_noperm() will undrain it if
3068 * @child_bs is not drained.
3069 *
3070 * The child was only just created and is not yet visible in global state
3071 * until bdrv_replace_child_noperm() inserts it into the graph, so nobody
3072 * could have sent requests and polling is not necessary.
3073 *
3074 * Note that this means that the parent isn't fully drained yet, we only
3075 * stop new requests from coming in. This is fine, we don't care about the
3076 * old requests here, they are not for this child. If another place enters a
3077 * drain section for the same parent, but wants it to be fully quiesced, it
3078 * will not run most of the the code in .drained_begin() again (which is not
3079 * a problem, we already did this), but it will still poll until the parent
3080 * is fully quiesced, so it will not be negatively affected either.
3081 */
Kevin Wolf606ed752022-11-18 18:41:10 +01003082 bdrv_parent_drained_begin_single(new_child);
Vladimir Sementsov-Ogievskiy544acc72022-07-26 23:11:31 +03003083 bdrv_replace_child_noperm(new_child, child_bs);
Vladimir Sementsov-Ogievskiy548a74c2021-04-28 18:17:46 +03003084
3085 BdrvAttachChildCommonState *s = g_new(BdrvAttachChildCommonState, 1);
3086 *s = (BdrvAttachChildCommonState) {
Vladimir Sementsov-Ogievskiy5bb047472022-07-26 23:11:32 +03003087 .child = new_child,
Vladimir Sementsov-Ogievskiy548a74c2021-04-28 18:17:46 +03003088 .old_parent_ctx = parent_ctx,
3089 .old_child_ctx = child_ctx,
3090 };
3091 tran_add(tran, &bdrv_attach_child_common_drv, s);
3092
Kevin Wolfc066e802023-06-05 10:57:05 +02003093 if (new_child_ctx != child_ctx) {
3094 aio_context_release(new_child_ctx);
3095 aio_context_acquire(child_ctx);
3096 }
3097
Vladimir Sementsov-Ogievskiy5bb047472022-07-26 23:11:32 +03003098 return new_child;
Vladimir Sementsov-Ogievskiy548a74c2021-04-28 18:17:46 +03003099}
3100
Vladimir Sementsov-Ogievskiyf8d2ad72021-06-01 10:52:13 +03003101/*
Vladimir Sementsov-Ogievskiy7ec390d2021-06-10 14:25:45 +03003102 * Function doesn't update permissions, caller is responsible for this.
Kevin Wolfc066e802023-06-05 10:57:05 +02003103 *
3104 * The caller must hold the AioContext lock for @child_bs. Both @parent_bs and
3105 * @child_bs can move to a different AioContext in this function. Callers must
3106 * make sure that their AioContext locking is still correct after this.
Vladimir Sementsov-Ogievskiyf8d2ad72021-06-01 10:52:13 +03003107 */
Vladimir Sementsov-Ogievskiy5bb047472022-07-26 23:11:32 +03003108static BdrvChild *bdrv_attach_child_noperm(BlockDriverState *parent_bs,
3109 BlockDriverState *child_bs,
3110 const char *child_name,
3111 const BdrvChildClass *child_class,
3112 BdrvChildRole child_role,
3113 Transaction *tran,
3114 Error **errp)
Vladimir Sementsov-Ogievskiyaa5a04c2021-04-28 18:17:47 +03003115{
Vladimir Sementsov-Ogievskiyaa5a04c2021-04-28 18:17:47 +03003116 uint64_t perm, shared_perm;
3117
3118 assert(parent_bs->drv);
Emanuele Giuseppe Espositobdb73472022-03-03 10:16:02 -05003119 GLOBAL_STATE_CODE();
Vladimir Sementsov-Ogievskiyaa5a04c2021-04-28 18:17:47 +03003120
Kevin Wolfbfb8aa62021-10-18 15:47:14 +02003121 if (bdrv_recurse_has_child(child_bs, parent_bs)) {
3122 error_setg(errp, "Making '%s' a %s child of '%s' would create a cycle",
3123 child_bs->node_name, child_name, parent_bs->node_name);
Vladimir Sementsov-Ogievskiy5bb047472022-07-26 23:11:32 +03003124 return NULL;
Kevin Wolfbfb8aa62021-10-18 15:47:14 +02003125 }
3126
Vladimir Sementsov-Ogievskiyaa5a04c2021-04-28 18:17:47 +03003127 bdrv_get_cumulative_perm(parent_bs, &perm, &shared_perm);
3128 bdrv_child_perm(parent_bs, child_bs, NULL, child_role, NULL,
3129 perm, shared_perm, &perm, &shared_perm);
3130
Vladimir Sementsov-Ogievskiy5bb047472022-07-26 23:11:32 +03003131 return bdrv_attach_child_common(child_bs, child_name, child_class,
3132 child_role, perm, shared_perm, parent_bs,
3133 tran, errp);
Vladimir Sementsov-Ogievskiyaa5a04c2021-04-28 18:17:47 +03003134}
3135
Alberto Garciab441dc72019-05-13 16:46:18 +03003136/*
3137 * This function steals the reference to child_bs from the caller.
3138 * That reference is later dropped by bdrv_root_unref_child().
3139 *
3140 * On failure NULL is returned, errp is set and the reference to
3141 * child_bs is also dropped.
Kevin Wolf132ada82019-04-24 17:41:46 +02003142 *
3143 * The caller must hold the AioContext lock @child_bs, but not that of @ctx
3144 * (unless @child_bs is already in @ctx).
Alberto Garciab441dc72019-05-13 16:46:18 +03003145 */
Kevin Wolff21d96d2016-03-08 13:47:46 +01003146BdrvChild *bdrv_root_attach_child(BlockDriverState *child_bs,
3147 const char *child_name,
Max Reitzbd86fb92020-05-13 13:05:13 +02003148 const BdrvChildClass *child_class,
Max Reitz258b7762020-05-13 13:05:15 +02003149 BdrvChildRole child_role,
Kevin Wolfd5e6f432016-12-14 17:24:36 +01003150 uint64_t perm, uint64_t shared_perm,
3151 void *opaque, Error **errp)
Kevin Wolfdf581792015-06-15 11:53:47 +02003152{
Kevin Wolfd5e6f432016-12-14 17:24:36 +01003153 int ret;
Vladimir Sementsov-Ogievskiy5bb047472022-07-26 23:11:32 +03003154 BdrvChild *child;
Vladimir Sementsov-Ogievskiy548a74c2021-04-28 18:17:46 +03003155 Transaction *tran = tran_new();
Kevin Wolfd5e6f432016-12-14 17:24:36 +01003156
Emanuele Giuseppe Espositob4ad82a2022-03-03 10:15:57 -05003157 GLOBAL_STATE_CODE();
3158
Vladimir Sementsov-Ogievskiy5bb047472022-07-26 23:11:32 +03003159 child = bdrv_attach_child_common(child_bs, child_name, child_class,
Vladimir Sementsov-Ogievskiy548a74c2021-04-28 18:17:46 +03003160 child_role, perm, shared_perm, opaque,
Vladimir Sementsov-Ogievskiy5bb047472022-07-26 23:11:32 +03003161 tran, errp);
3162 if (!child) {
3163 ret = -EINVAL;
Kevin Wolfe878bb12021-05-03 13:05:54 +02003164 goto out;
Kevin Wolfd5e6f432016-12-14 17:24:36 +01003165 }
3166
Vladimir Sementsov-Ogievskiyf1316ed2022-11-07 19:35:57 +03003167 ret = bdrv_refresh_perms(child_bs, tran, errp);
Kevin Wolfdf581792015-06-15 11:53:47 +02003168
Kevin Wolfe878bb12021-05-03 13:05:54 +02003169out:
3170 tran_finalize(tran, ret);
Vladimir Sementsov-Ogievskiyf8d2ad72021-06-01 10:52:13 +03003171
Vladimir Sementsov-Ogievskiy548a74c2021-04-28 18:17:46 +03003172 bdrv_unref(child_bs);
Vladimir Sementsov-Ogievskiy5bb047472022-07-26 23:11:32 +03003173
3174 return ret < 0 ? NULL : child;
Kevin Wolfdf581792015-06-15 11:53:47 +02003175}
3176
Alberto Garciab441dc72019-05-13 16:46:18 +03003177/*
3178 * This function transfers the reference to child_bs from the caller
3179 * to parent_bs. That reference is later dropped by parent_bs on
3180 * bdrv_close() or if someone calls bdrv_unref_child().
3181 *
3182 * On failure NULL is returned, errp is set and the reference to
3183 * child_bs is also dropped.
Kevin Wolf132ada82019-04-24 17:41:46 +02003184 *
3185 * If @parent_bs and @child_bs are in different AioContexts, the caller must
3186 * hold the AioContext lock for @child_bs, but not for @parent_bs.
Alberto Garciab441dc72019-05-13 16:46:18 +03003187 */
Wen Congyang98292c62016-05-10 15:36:38 +08003188BdrvChild *bdrv_attach_child(BlockDriverState *parent_bs,
3189 BlockDriverState *child_bs,
3190 const char *child_name,
Max Reitzbd86fb92020-05-13 13:05:13 +02003191 const BdrvChildClass *child_class,
Max Reitz258b7762020-05-13 13:05:15 +02003192 BdrvChildRole child_role,
Kevin Wolf8b2ff522016-12-20 22:21:17 +01003193 Error **errp)
Kevin Wolff21d96d2016-03-08 13:47:46 +01003194{
Vladimir Sementsov-Ogievskiyaa5a04c2021-04-28 18:17:47 +03003195 int ret;
Vladimir Sementsov-Ogievskiy5bb047472022-07-26 23:11:32 +03003196 BdrvChild *child;
Vladimir Sementsov-Ogievskiyaa5a04c2021-04-28 18:17:47 +03003197 Transaction *tran = tran_new();
Kevin Wolfd5e6f432016-12-14 17:24:36 +01003198
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05003199 GLOBAL_STATE_CODE();
3200
Vladimir Sementsov-Ogievskiy5bb047472022-07-26 23:11:32 +03003201 child = bdrv_attach_child_noperm(parent_bs, child_bs, child_name,
3202 child_class, child_role, tran, errp);
3203 if (!child) {
3204 ret = -EINVAL;
Vladimir Sementsov-Ogievskiyaa5a04c2021-04-28 18:17:47 +03003205 goto out;
Kevin Wolfd5e6f432016-12-14 17:24:36 +01003206 }
3207
Vladimir Sementsov-Ogievskiyf1316ed2022-11-07 19:35:57 +03003208 ret = bdrv_refresh_perms(parent_bs, tran, errp);
Vladimir Sementsov-Ogievskiyaa5a04c2021-04-28 18:17:47 +03003209 if (ret < 0) {
3210 goto out;
3211 }
3212
3213out:
3214 tran_finalize(tran, ret);
3215
3216 bdrv_unref(child_bs);
3217
Vladimir Sementsov-Ogievskiy5bb047472022-07-26 23:11:32 +03003218 return ret < 0 ? NULL : child;
Kevin Wolff21d96d2016-03-08 13:47:46 +01003219}
3220
Max Reitz7b99a262019-06-12 16:07:11 +02003221/* Callers must ensure that child->frozen is false. */
Kevin Wolff21d96d2016-03-08 13:47:46 +01003222void bdrv_root_unref_child(BdrvChild *child)
Kevin Wolf33a60402015-06-15 13:51:04 +02003223{
Vladimir Sementsov-Ogievskiy00eb93b2022-11-07 19:35:55 +03003224 BlockDriverState *child_bs = child->bs;
Kevin Wolf779020c2015-10-13 14:09:44 +02003225
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05003226 GLOBAL_STATE_CODE();
Vladimir Sementsov-Ogievskiy00eb93b2022-11-07 19:35:55 +03003227 bdrv_replace_child_noperm(child, NULL);
3228 bdrv_child_free(child);
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05003229
Vladimir Sementsov-Ogievskiy00eb93b2022-11-07 19:35:55 +03003230 if (child_bs) {
3231 /*
3232 * Update permissions for old node. We're just taking a parent away, so
3233 * we're loosening restrictions. Errors of permission update are not
3234 * fatal in this case, ignore them.
3235 */
Vladimir Sementsov-Ogievskiyf1316ed2022-11-07 19:35:57 +03003236 bdrv_refresh_perms(child_bs, NULL, NULL);
Vladimir Sementsov-Ogievskiy00eb93b2022-11-07 19:35:55 +03003237
3238 /*
3239 * When the parent requiring a non-default AioContext is removed, the
3240 * node moves back to the main AioContext
3241 */
3242 bdrv_try_change_aio_context(child_bs, qemu_get_aio_context(), NULL,
3243 NULL);
3244 }
3245
Kevin Wolff21d96d2016-03-08 13:47:46 +01003246 bdrv_unref(child_bs);
3247}
3248
Vladimir Sementsov-Ogievskiy332b3a12021-04-28 18:17:54 +03003249typedef struct BdrvSetInheritsFrom {
3250 BlockDriverState *bs;
3251 BlockDriverState *old_inherits_from;
3252} BdrvSetInheritsFrom;
3253
3254static void bdrv_set_inherits_from_abort(void *opaque)
3255{
3256 BdrvSetInheritsFrom *s = opaque;
3257
3258 s->bs->inherits_from = s->old_inherits_from;
3259}
3260
3261static TransactionActionDrv bdrv_set_inherits_from_drv = {
3262 .abort = bdrv_set_inherits_from_abort,
3263 .clean = g_free,
3264};
3265
3266/* @tran is allowed to be NULL. In this case no rollback is possible */
3267static void bdrv_set_inherits_from(BlockDriverState *bs,
3268 BlockDriverState *new_inherits_from,
3269 Transaction *tran)
3270{
3271 if (tran) {
3272 BdrvSetInheritsFrom *s = g_new(BdrvSetInheritsFrom, 1);
3273
3274 *s = (BdrvSetInheritsFrom) {
3275 .bs = bs,
3276 .old_inherits_from = bs->inherits_from,
3277 };
3278
3279 tran_add(tran, &bdrv_set_inherits_from_drv, s);
3280 }
3281
3282 bs->inherits_from = new_inherits_from;
3283}
3284
Max Reitz3cf746b2019-07-03 19:28:07 +02003285/**
3286 * Clear all inherits_from pointers from children and grandchildren of
3287 * @root that point to @root, where necessary.
Vladimir Sementsov-Ogievskiy332b3a12021-04-28 18:17:54 +03003288 * @tran is allowed to be NULL. In this case no rollback is possible
Max Reitz3cf746b2019-07-03 19:28:07 +02003289 */
Vladimir Sementsov-Ogievskiy332b3a12021-04-28 18:17:54 +03003290static void bdrv_unset_inherits_from(BlockDriverState *root, BdrvChild *child,
3291 Transaction *tran)
Kevin Wolff21d96d2016-03-08 13:47:46 +01003292{
Max Reitz3cf746b2019-07-03 19:28:07 +02003293 BdrvChild *c;
Kevin Wolf33a60402015-06-15 13:51:04 +02003294
Max Reitz3cf746b2019-07-03 19:28:07 +02003295 if (child->bs->inherits_from == root) {
3296 /*
3297 * Remove inherits_from only when the last reference between root and
3298 * child->bs goes away.
3299 */
3300 QLIST_FOREACH(c, &root->children, next) {
Kevin Wolf4e4bf5c2016-12-16 18:52:37 +01003301 if (c != child && c->bs == child->bs) {
3302 break;
3303 }
3304 }
3305 if (c == NULL) {
Vladimir Sementsov-Ogievskiy332b3a12021-04-28 18:17:54 +03003306 bdrv_set_inherits_from(child->bs, NULL, tran);
Kevin Wolf4e4bf5c2016-12-16 18:52:37 +01003307 }
Kevin Wolf33a60402015-06-15 13:51:04 +02003308 }
3309
Max Reitz3cf746b2019-07-03 19:28:07 +02003310 QLIST_FOREACH(c, &child->bs->children, next) {
Vladimir Sementsov-Ogievskiy332b3a12021-04-28 18:17:54 +03003311 bdrv_unset_inherits_from(root, c, tran);
Max Reitz3cf746b2019-07-03 19:28:07 +02003312 }
3313}
3314
Max Reitz7b99a262019-06-12 16:07:11 +02003315/* Callers must ensure that child->frozen is false. */
Max Reitz3cf746b2019-07-03 19:28:07 +02003316void bdrv_unref_child(BlockDriverState *parent, BdrvChild *child)
3317{
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05003318 GLOBAL_STATE_CODE();
Max Reitz3cf746b2019-07-03 19:28:07 +02003319 if (child == NULL) {
3320 return;
3321 }
3322
Vladimir Sementsov-Ogievskiy332b3a12021-04-28 18:17:54 +03003323 bdrv_unset_inherits_from(parent, child, NULL);
Kevin Wolff21d96d2016-03-08 13:47:46 +01003324 bdrv_root_unref_child(child);
Kevin Wolf33a60402015-06-15 13:51:04 +02003325}
3326
Kevin Wolf5c8cab42016-02-24 15:13:35 +01003327
3328static void bdrv_parent_cb_change_media(BlockDriverState *bs, bool load)
3329{
3330 BdrvChild *c;
Emanuele Giuseppe Espositof0c28322022-03-03 10:16:13 -05003331 GLOBAL_STATE_CODE();
Kevin Wolf5c8cab42016-02-24 15:13:35 +01003332 QLIST_FOREACH(c, &bs->parents, next_parent) {
Max Reitzbd86fb92020-05-13 13:05:13 +02003333 if (c->klass->change_media) {
3334 c->klass->change_media(c, load);
Kevin Wolf5c8cab42016-02-24 15:13:35 +01003335 }
3336 }
3337}
3338
Alberto Garcia0065c452018-10-31 18:16:37 +02003339/* Return true if you can reach parent going through child->inherits_from
3340 * recursively. If parent or child are NULL, return false */
3341static bool bdrv_inherits_from_recursive(BlockDriverState *child,
3342 BlockDriverState *parent)
3343{
3344 while (child && child != parent) {
3345 child = child->inherits_from;
3346 }
3347
3348 return child != NULL;
3349}
3350
Kevin Wolf5db15a52015-09-14 15:33:33 +02003351/*
Max Reitz25191e52020-05-13 13:05:33 +02003352 * Return the BdrvChildRole for @bs's backing child. bs->backing is
3353 * mostly used for COW backing children (role = COW), but also for
3354 * filtered children (role = FILTERED | PRIMARY).
3355 */
3356static BdrvChildRole bdrv_backing_role(BlockDriverState *bs)
3357{
3358 if (bs->drv && bs->drv->is_filter) {
3359 return BDRV_CHILD_FILTERED | BDRV_CHILD_PRIMARY;
3360 } else {
3361 return BDRV_CHILD_COW;
3362 }
3363}
3364
3365/*
Vladimir Sementsov-Ogievskiye9238272021-06-10 15:05:30 +03003366 * Sets the bs->backing or bs->file link of a BDS. A new reference is created;
3367 * callers which don't need their own reference any more must call bdrv_unref().
Vladimir Sementsov-Ogievskiy7ec390d2021-06-10 14:25:45 +03003368 *
3369 * Function doesn't update permissions, caller is responsible for this.
Kevin Wolf4b408662023-06-05 10:57:06 +02003370 *
3371 * The caller must hold the AioContext lock for @child_bs. Both @parent_bs and
3372 * @child_bs can move to a different AioContext in this function. Callers must
3373 * make sure that their AioContext locking is still correct after this.
Kevin Wolf5db15a52015-09-14 15:33:33 +02003374 */
Vladimir Sementsov-Ogievskiye9238272021-06-10 15:05:30 +03003375static int bdrv_set_file_or_backing_noperm(BlockDriverState *parent_bs,
3376 BlockDriverState *child_bs,
3377 bool is_backing,
3378 Transaction *tran, Error **errp)
Fam Zheng8d24cce2014-05-23 21:29:45 +08003379{
Vladimir Sementsov-Ogievskiye9238272021-06-10 15:05:30 +03003380 bool update_inherits_from =
3381 bdrv_inherits_from_recursive(child_bs, parent_bs);
3382 BdrvChild *child = is_backing ? parent_bs->backing : parent_bs->file;
3383 BdrvChildRole role;
Alberto Garcia0065c452018-10-31 18:16:37 +02003384
Emanuele Giuseppe Espositobdb73472022-03-03 10:16:02 -05003385 GLOBAL_STATE_CODE();
3386
Vladimir Sementsov-Ogievskiye9238272021-06-10 15:05:30 +03003387 if (!parent_bs->drv) {
3388 /*
3389 * Node without drv is an object without a class :/. TODO: finally fix
3390 * qcow2 driver to never clear bs->drv and implement format corruption
3391 * handling in other way.
3392 */
3393 error_setg(errp, "Node corrupted");
3394 return -EINVAL;
3395 }
3396
3397 if (child && child->frozen) {
3398 error_setg(errp, "Cannot change frozen '%s' link from '%s' to '%s'",
3399 child->name, parent_bs->node_name, child->bs->node_name);
Vladimir Sementsov-Ogievskiya1e708f2021-02-02 15:49:43 +03003400 return -EPERM;
Alberto Garcia2cad1eb2019-03-12 18:48:40 +02003401 }
3402
Vladimir Sementsov-Ogievskiy25f78d92021-06-10 15:05:34 +03003403 if (is_backing && !parent_bs->drv->is_filter &&
3404 !parent_bs->drv->supports_backing)
3405 {
3406 error_setg(errp, "Driver '%s' of node '%s' does not support backing "
3407 "files", parent_bs->drv->format_name, parent_bs->node_name);
3408 return -EINVAL;
3409 }
3410
Vladimir Sementsov-Ogievskiye9238272021-06-10 15:05:30 +03003411 if (parent_bs->drv->is_filter) {
3412 role = BDRV_CHILD_FILTERED | BDRV_CHILD_PRIMARY;
3413 } else if (is_backing) {
3414 role = BDRV_CHILD_COW;
3415 } else {
3416 /*
3417 * We only can use same role as it is in existing child. We don't have
3418 * infrastructure to determine role of file child in generic way
3419 */
3420 if (!child) {
3421 error_setg(errp, "Cannot set file child to format node without "
3422 "file child");
3423 return -EINVAL;
3424 }
3425 role = child->role;
Fam Zheng826b6ca2014-05-23 21:29:47 +08003426 }
3427
Vladimir Sementsov-Ogievskiye9238272021-06-10 15:05:30 +03003428 if (child) {
3429 bdrv_unset_inherits_from(parent_bs, child, tran);
Vladimir Sementsov-Ogievskiy57f08942022-07-26 23:11:34 +03003430 bdrv_remove_child(child, tran);
Vladimir Sementsov-Ogievskiye9238272021-06-10 15:05:30 +03003431 }
3432
3433 if (!child_bs) {
Fam Zheng8d24cce2014-05-23 21:29:45 +08003434 goto out;
3435 }
Kevin Wolf12fa4af2017-02-17 20:42:32 +01003436
Vladimir Sementsov-Ogievskiy5bb047472022-07-26 23:11:32 +03003437 child = bdrv_attach_child_noperm(parent_bs, child_bs,
3438 is_backing ? "backing" : "file",
3439 &child_of_bds, role,
3440 tran, errp);
3441 if (!child) {
3442 return -EINVAL;
Vladimir Sementsov-Ogievskiya1e708f2021-02-02 15:49:43 +03003443 }
3444
Vladimir Sementsov-Ogievskiy160333e2021-04-28 18:17:56 +03003445
3446 /*
Vladimir Sementsov-Ogievskiye9238272021-06-10 15:05:30 +03003447 * If inherits_from pointed recursively to bs then let's update it to
Vladimir Sementsov-Ogievskiy160333e2021-04-28 18:17:56 +03003448 * point directly to bs (else it will become NULL).
3449 */
Vladimir Sementsov-Ogievskiya1e708f2021-02-02 15:49:43 +03003450 if (update_inherits_from) {
Vladimir Sementsov-Ogievskiye9238272021-06-10 15:05:30 +03003451 bdrv_set_inherits_from(child_bs, parent_bs, tran);
Alberto Garcia0065c452018-10-31 18:16:37 +02003452 }
Fam Zheng826b6ca2014-05-23 21:29:47 +08003453
Fam Zheng8d24cce2014-05-23 21:29:45 +08003454out:
Kevin Wolfe19b1572023-05-04 13:57:50 +02003455 bdrv_graph_rdlock_main_loop();
Vladimir Sementsov-Ogievskiye9238272021-06-10 15:05:30 +03003456 bdrv_refresh_limits(parent_bs, tran, NULL);
Kevin Wolfe19b1572023-05-04 13:57:50 +02003457 bdrv_graph_rdunlock_main_loop();
Vladimir Sementsov-Ogievskiy160333e2021-04-28 18:17:56 +03003458
3459 return 0;
3460}
3461
Kevin Wolf4b408662023-06-05 10:57:06 +02003462/*
3463 * The caller must hold the AioContext lock for @backing_hd. Both @bs and
3464 * @backing_hd can move to a different AioContext in this function. Callers must
3465 * make sure that their AioContext locking is still correct after this.
3466 */
Vladimir Sementsov-Ogievskiye9238272021-06-10 15:05:30 +03003467static int bdrv_set_backing_noperm(BlockDriverState *bs,
3468 BlockDriverState *backing_hd,
3469 Transaction *tran, Error **errp)
3470{
Emanuele Giuseppe Espositobdb73472022-03-03 10:16:02 -05003471 GLOBAL_STATE_CODE();
Vladimir Sementsov-Ogievskiye9238272021-06-10 15:05:30 +03003472 return bdrv_set_file_or_backing_noperm(bs, backing_hd, true, tran, errp);
3473}
3474
Kevin Wolf92140b92022-11-18 18:41:04 +01003475int bdrv_set_backing_hd_drained(BlockDriverState *bs,
3476 BlockDriverState *backing_hd,
3477 Error **errp)
Vladimir Sementsov-Ogievskiy160333e2021-04-28 18:17:56 +03003478{
3479 int ret;
3480 Transaction *tran = tran_new();
3481
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05003482 GLOBAL_STATE_CODE();
Kevin Wolf92140b92022-11-18 18:41:04 +01003483 assert(bs->quiesce_counter > 0);
Vladimir Sementsov-Ogievskiyc0829cb2022-01-24 18:37:41 +01003484
Vladimir Sementsov-Ogievskiy160333e2021-04-28 18:17:56 +03003485 ret = bdrv_set_backing_noperm(bs, backing_hd, tran, errp);
3486 if (ret < 0) {
3487 goto out;
3488 }
3489
Vladimir Sementsov-Ogievskiyf1316ed2022-11-07 19:35:57 +03003490 ret = bdrv_refresh_perms(bs, tran, errp);
Vladimir Sementsov-Ogievskiy160333e2021-04-28 18:17:56 +03003491out:
3492 tran_finalize(tran, ret);
Kevin Wolf92140b92022-11-18 18:41:04 +01003493 return ret;
3494}
Vladimir Sementsov-Ogievskiya1e708f2021-02-02 15:49:43 +03003495
Kevin Wolf92140b92022-11-18 18:41:04 +01003496int bdrv_set_backing_hd(BlockDriverState *bs, BlockDriverState *backing_hd,
3497 Error **errp)
3498{
3499 int ret;
3500 GLOBAL_STATE_CODE();
3501
3502 bdrv_drained_begin(bs);
3503 ret = bdrv_set_backing_hd_drained(bs, backing_hd, errp);
Vladimir Sementsov-Ogievskiyc0829cb2022-01-24 18:37:41 +01003504 bdrv_drained_end(bs);
3505
Vladimir Sementsov-Ogievskiya1e708f2021-02-02 15:49:43 +03003506 return ret;
Fam Zheng8d24cce2014-05-23 21:29:45 +08003507}
3508
Kevin Wolf31ca6d02013-03-28 15:29:24 +01003509/*
3510 * Opens the backing file for a BlockDriverState if not yet open
3511 *
Kevin Wolfd9b7b052015-01-16 18:23:41 +01003512 * bdref_key specifies the key for the image's BlockdevRef in the options QDict.
3513 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
3514 * itself, all options starting with "${bdref_key}." are considered part of the
3515 * BlockdevRef.
3516 *
Kevin Wolf2626d272023-05-25 14:47:06 +02003517 * The caller must hold the main AioContext lock.
3518 *
Kevin Wolfd9b7b052015-01-16 18:23:41 +01003519 * TODO Can this be unified with bdrv_open_image()?
Kevin Wolf31ca6d02013-03-28 15:29:24 +01003520 */
Kevin Wolfd9b7b052015-01-16 18:23:41 +01003521int bdrv_open_backing_file(BlockDriverState *bs, QDict *parent_options,
3522 const char *bdref_key, Error **errp)
Paolo Bonzini9156df12012-10-18 16:49:17 +02003523{
Max Reitz6b6833c2019-02-01 20:29:15 +01003524 char *backing_filename = NULL;
Kevin Wolfd9b7b052015-01-16 18:23:41 +01003525 char *bdref_key_dot;
3526 const char *reference = NULL;
Kevin Wolf317fc442014-04-25 13:27:34 +02003527 int ret = 0;
Max Reitz998c2012019-02-01 20:29:08 +01003528 bool implicit_backing = false;
Fam Zheng8d24cce2014-05-23 21:29:45 +08003529 BlockDriverState *backing_hd;
Kevin Wolf8aa04542023-06-05 10:57:08 +02003530 AioContext *backing_hd_ctx;
Kevin Wolfd9b7b052015-01-16 18:23:41 +01003531 QDict *options;
3532 QDict *tmp_parent_options = NULL;
Max Reitz34b5d2c2013-09-05 14:45:29 +02003533 Error *local_err = NULL;
Paolo Bonzini9156df12012-10-18 16:49:17 +02003534
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05003535 GLOBAL_STATE_CODE();
3536
Kevin Wolf760e0062015-06-17 14:55:21 +02003537 if (bs->backing != NULL) {
Benoît Canet1ba4b6a2014-04-22 17:05:27 +02003538 goto free_exit;
Paolo Bonzini9156df12012-10-18 16:49:17 +02003539 }
3540
Kevin Wolf31ca6d02013-03-28 15:29:24 +01003541 /* NULL means an empty set of options */
Kevin Wolfd9b7b052015-01-16 18:23:41 +01003542 if (parent_options == NULL) {
3543 tmp_parent_options = qdict_new();
3544 parent_options = tmp_parent_options;
Kevin Wolf31ca6d02013-03-28 15:29:24 +01003545 }
3546
Paolo Bonzini9156df12012-10-18 16:49:17 +02003547 bs->open_flags &= ~BDRV_O_NO_BACKING;
Kevin Wolfd9b7b052015-01-16 18:23:41 +01003548
3549 bdref_key_dot = g_strdup_printf("%s.", bdref_key);
3550 qdict_extract_subqdict(parent_options, &options, bdref_key_dot);
3551 g_free(bdref_key_dot);
3552
Markus Armbruster129c7d12017-03-30 19:43:12 +02003553 /*
3554 * Caution: while qdict_get_try_str() is fine, getting non-string
3555 * types would require more care. When @parent_options come from
3556 * -blockdev or blockdev_add, its members are typed according to
3557 * the QAPI schema, but when they come from -drive, they're all
3558 * QString.
3559 */
Kevin Wolfd9b7b052015-01-16 18:23:41 +01003560 reference = qdict_get_try_str(parent_options, bdref_key);
3561 if (reference || qdict_haskey(options, "file.filename")) {
Max Reitz6b6833c2019-02-01 20:29:15 +01003562 /* keep backing_filename NULL */
Kevin Wolf1cb6f502013-04-12 20:27:07 +02003563 } else if (bs->backing_file[0] == '\0' && qdict_size(options) == 0) {
Marc-André Lureaucb3e7f02018-04-19 17:01:43 +02003564 qobject_unref(options);
Benoît Canet1ba4b6a2014-04-22 17:05:27 +02003565 goto free_exit;
Fam Zhengdbecebd2013-09-22 20:05:06 +08003566 } else {
Max Reitz998c2012019-02-01 20:29:08 +01003567 if (qdict_size(options) == 0) {
3568 /* If the user specifies options that do not modify the
3569 * backing file's behavior, we might still consider it the
3570 * implicit backing file. But it's easier this way, and
3571 * just specifying some of the backing BDS's options is
3572 * only possible with -drive anyway (otherwise the QAPI
3573 * schema forces the user to specify everything). */
3574 implicit_backing = !strcmp(bs->auto_backing_file, bs->backing_file);
3575 }
3576
Max Reitz6b6833c2019-02-01 20:29:15 +01003577 backing_filename = bdrv_get_full_backing_filename(bs, &local_err);
Max Reitz9f074292014-11-26 17:20:26 +01003578 if (local_err) {
3579 ret = -EINVAL;
3580 error_propagate(errp, local_err);
Marc-André Lureaucb3e7f02018-04-19 17:01:43 +02003581 qobject_unref(options);
Max Reitz9f074292014-11-26 17:20:26 +01003582 goto free_exit;
3583 }
Paolo Bonzini9156df12012-10-18 16:49:17 +02003584 }
3585
Kevin Wolf8ee79e72014-06-04 15:09:35 +02003586 if (!bs->drv || !bs->drv->supports_backing) {
3587 ret = -EINVAL;
3588 error_setg(errp, "Driver doesn't support backing files");
Marc-André Lureaucb3e7f02018-04-19 17:01:43 +02003589 qobject_unref(options);
Kevin Wolf8ee79e72014-06-04 15:09:35 +02003590 goto free_exit;
3591 }
3592
Peter Krempa6bff5972017-10-12 16:14:10 +02003593 if (!reference &&
3594 bs->backing_format[0] != '\0' && !qdict_haskey(options, "driver")) {
Eric Blake46f5ac22017-04-27 16:58:17 -05003595 qdict_put_str(options, "driver", bs->backing_format);
Paolo Bonzini9156df12012-10-18 16:49:17 +02003596 }
3597
Max Reitz6b6833c2019-02-01 20:29:15 +01003598 backing_hd = bdrv_open_inherit(backing_filename, reference, options, 0, bs,
Max Reitz25191e52020-05-13 13:05:33 +02003599 &child_of_bds, bdrv_backing_role(bs), errp);
Max Reitz5b363932016-05-17 16:41:31 +02003600 if (!backing_hd) {
Paolo Bonzini9156df12012-10-18 16:49:17 +02003601 bs->open_flags |= BDRV_O_NO_BACKING;
Markus Armbrustere43bfd92015-12-18 16:35:15 +01003602 error_prepend(errp, "Could not open backing file: ");
Max Reitz5b363932016-05-17 16:41:31 +02003603 ret = -EINVAL;
Benoît Canet1ba4b6a2014-04-22 17:05:27 +02003604 goto free_exit;
Paolo Bonzini9156df12012-10-18 16:49:17 +02003605 }
Kevin Wolfdf581792015-06-15 11:53:47 +02003606
Max Reitz998c2012019-02-01 20:29:08 +01003607 if (implicit_backing) {
3608 bdrv_refresh_filename(backing_hd);
3609 pstrcpy(bs->auto_backing_file, sizeof(bs->auto_backing_file),
3610 backing_hd->filename);
3611 }
3612
Kevin Wolf5db15a52015-09-14 15:33:33 +02003613 /* Hook up the backing file link; drop our reference, bs owns the
3614 * backing_hd reference now */
Kevin Wolf8aa04542023-06-05 10:57:08 +02003615 backing_hd_ctx = bdrv_get_aio_context(backing_hd);
3616 aio_context_acquire(backing_hd_ctx);
Vladimir Sementsov-Ogievskiydc9c10a2021-02-02 15:49:47 +03003617 ret = bdrv_set_backing_hd(bs, backing_hd, errp);
Kevin Wolf5db15a52015-09-14 15:33:33 +02003618 bdrv_unref(backing_hd);
Kevin Wolf8aa04542023-06-05 10:57:08 +02003619 aio_context_release(backing_hd_ctx);
3620
Vladimir Sementsov-Ogievskiydc9c10a2021-02-02 15:49:47 +03003621 if (ret < 0) {
Kevin Wolf12fa4af2017-02-17 20:42:32 +01003622 goto free_exit;
3623 }
Peter Feinerd80ac652014-01-08 19:43:25 +00003624
Kevin Wolfd9b7b052015-01-16 18:23:41 +01003625 qdict_del(parent_options, bdref_key);
3626
Benoît Canet1ba4b6a2014-04-22 17:05:27 +02003627free_exit:
3628 g_free(backing_filename);
Marc-André Lureaucb3e7f02018-04-19 17:01:43 +02003629 qobject_unref(tmp_parent_options);
Benoît Canet1ba4b6a2014-04-22 17:05:27 +02003630 return ret;
Paolo Bonzini9156df12012-10-18 16:49:17 +02003631}
3632
Kevin Wolf2d6b86a2017-02-17 17:43:59 +01003633static BlockDriverState *
3634bdrv_open_child_bs(const char *filename, QDict *options, const char *bdref_key,
Max Reitzbd86fb92020-05-13 13:05:13 +02003635 BlockDriverState *parent, const BdrvChildClass *child_class,
Max Reitz272c02e2020-05-13 13:05:17 +02003636 BdrvChildRole child_role, bool allow_none, Error **errp)
Max Reitzda557aa2013-12-20 19:28:11 +01003637{
Kevin Wolf2d6b86a2017-02-17 17:43:59 +01003638 BlockDriverState *bs = NULL;
Max Reitzda557aa2013-12-20 19:28:11 +01003639 QDict *image_options;
Max Reitzda557aa2013-12-20 19:28:11 +01003640 char *bdref_key_dot;
3641 const char *reference;
3642
Max Reitzbd86fb92020-05-13 13:05:13 +02003643 assert(child_class != NULL);
Max Reitzf67503e2014-02-18 18:33:05 +01003644
Max Reitzda557aa2013-12-20 19:28:11 +01003645 bdref_key_dot = g_strdup_printf("%s.", bdref_key);
3646 qdict_extract_subqdict(options, &image_options, bdref_key_dot);
3647 g_free(bdref_key_dot);
3648
Markus Armbruster129c7d12017-03-30 19:43:12 +02003649 /*
3650 * Caution: while qdict_get_try_str() is fine, getting non-string
3651 * types would require more care. When @options come from
3652 * -blockdev or blockdev_add, its members are typed according to
3653 * the QAPI schema, but when they come from -drive, they're all
3654 * QString.
3655 */
Max Reitzda557aa2013-12-20 19:28:11 +01003656 reference = qdict_get_try_str(options, bdref_key);
3657 if (!filename && !reference && !qdict_size(image_options)) {
Kevin Wolfb4b059f2015-06-15 13:24:19 +02003658 if (!allow_none) {
Max Reitzda557aa2013-12-20 19:28:11 +01003659 error_setg(errp, "A block device must be specified for \"%s\"",
3660 bdref_key);
Max Reitzda557aa2013-12-20 19:28:11 +01003661 }
Marc-André Lureaucb3e7f02018-04-19 17:01:43 +02003662 qobject_unref(image_options);
Max Reitzda557aa2013-12-20 19:28:11 +01003663 goto done;
3664 }
3665
Max Reitz5b363932016-05-17 16:41:31 +02003666 bs = bdrv_open_inherit(filename, reference, image_options, 0,
Max Reitz272c02e2020-05-13 13:05:17 +02003667 parent, child_class, child_role, errp);
Max Reitz5b363932016-05-17 16:41:31 +02003668 if (!bs) {
Kevin Wolfdf581792015-06-15 11:53:47 +02003669 goto done;
3670 }
3671
Max Reitzda557aa2013-12-20 19:28:11 +01003672done:
3673 qdict_del(options, bdref_key);
Kevin Wolf2d6b86a2017-02-17 17:43:59 +01003674 return bs;
3675}
3676
3677/*
3678 * Opens a disk image whose options are given as BlockdevRef in another block
3679 * device's options.
3680 *
3681 * If allow_none is true, no image will be opened if filename is false and no
3682 * BlockdevRef is given. NULL will be returned, but errp remains unset.
3683 *
3684 * bdrev_key specifies the key for the image's BlockdevRef in the options QDict.
3685 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
3686 * itself, all options starting with "${bdref_key}." are considered part of the
3687 * BlockdevRef.
3688 *
3689 * The BlockdevRef will be removed from the options QDict.
Kevin Wolfaa269ff2023-05-25 14:47:07 +02003690 *
Kevin Wolf8394c352023-06-05 10:57:04 +02003691 * The caller must hold the lock of the main AioContext and no other AioContext.
Kevin Wolfaa269ff2023-05-25 14:47:07 +02003692 * @parent can move to a different AioContext in this function. Callers must
3693 * make sure that their AioContext locking is still correct after this.
Kevin Wolf2d6b86a2017-02-17 17:43:59 +01003694 */
3695BdrvChild *bdrv_open_child(const char *filename,
3696 QDict *options, const char *bdref_key,
3697 BlockDriverState *parent,
Max Reitzbd86fb92020-05-13 13:05:13 +02003698 const BdrvChildClass *child_class,
Max Reitz258b7762020-05-13 13:05:15 +02003699 BdrvChildRole child_role,
Kevin Wolf2d6b86a2017-02-17 17:43:59 +01003700 bool allow_none, Error **errp)
3701{
3702 BlockDriverState *bs;
Kevin Wolf8394c352023-06-05 10:57:04 +02003703 BdrvChild *child;
3704 AioContext *ctx;
Kevin Wolf2d6b86a2017-02-17 17:43:59 +01003705
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05003706 GLOBAL_STATE_CODE();
3707
Max Reitzbd86fb92020-05-13 13:05:13 +02003708 bs = bdrv_open_child_bs(filename, options, bdref_key, parent, child_class,
Max Reitz272c02e2020-05-13 13:05:17 +02003709 child_role, allow_none, errp);
Kevin Wolf2d6b86a2017-02-17 17:43:59 +01003710 if (bs == NULL) {
3711 return NULL;
3712 }
3713
Kevin Wolf8394c352023-06-05 10:57:04 +02003714 ctx = bdrv_get_aio_context(bs);
3715 aio_context_acquire(ctx);
3716 child = bdrv_attach_child(parent, bs, bdref_key, child_class, child_role,
3717 errp);
3718 aio_context_release(ctx);
3719
3720 return child;
Kevin Wolfb4b059f2015-06-15 13:24:19 +02003721}
3722
Max Reitzbd86fb92020-05-13 13:05:13 +02003723/*
Vladimir Sementsov-Ogievskiy83930782022-07-26 23:11:21 +03003724 * Wrapper on bdrv_open_child() for most popular case: open primary child of bs.
Kevin Wolfaa269ff2023-05-25 14:47:07 +02003725 *
Kevin Wolf8394c352023-06-05 10:57:04 +02003726 * The caller must hold the lock of the main AioContext and no other AioContext.
Kevin Wolfaa269ff2023-05-25 14:47:07 +02003727 * @parent can move to a different AioContext in this function. Callers must
3728 * make sure that their AioContext locking is still correct after this.
Vladimir Sementsov-Ogievskiy83930782022-07-26 23:11:21 +03003729 */
3730int bdrv_open_file_child(const char *filename,
3731 QDict *options, const char *bdref_key,
3732 BlockDriverState *parent, Error **errp)
3733{
3734 BdrvChildRole role;
3735
3736 /* commit_top and mirror_top don't use this function */
3737 assert(!parent->drv->filtered_child_is_backing);
Vladimir Sementsov-Ogievskiy83930782022-07-26 23:11:21 +03003738 role = parent->drv->is_filter ?
3739 (BDRV_CHILD_FILTERED | BDRV_CHILD_PRIMARY) : BDRV_CHILD_IMAGE;
3740
Vladimir Sementsov-Ogievskiy5bb047472022-07-26 23:11:32 +03003741 if (!bdrv_open_child(filename, options, bdref_key, parent,
3742 &child_of_bds, role, false, errp))
3743 {
3744 return -EINVAL;
3745 }
Vladimir Sementsov-Ogievskiy83930782022-07-26 23:11:21 +03003746
Vladimir Sementsov-Ogievskiy5bb047472022-07-26 23:11:32 +03003747 return 0;
Vladimir Sementsov-Ogievskiy83930782022-07-26 23:11:21 +03003748}
3749
3750/*
Max Reitzbd86fb92020-05-13 13:05:13 +02003751 * TODO Future callers may need to specify parent/child_class in order for
3752 * option inheritance to work. Existing callers use it for the root node.
3753 */
Kevin Wolfe1d74bc2018-01-10 15:52:33 +01003754BlockDriverState *bdrv_open_blockdev_ref(BlockdevRef *ref, Error **errp)
3755{
3756 BlockDriverState *bs = NULL;
Kevin Wolfe1d74bc2018-01-10 15:52:33 +01003757 QObject *obj = NULL;
3758 QDict *qdict = NULL;
3759 const char *reference = NULL;
3760 Visitor *v = NULL;
3761
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05003762 GLOBAL_STATE_CODE();
3763
Kevin Wolfe1d74bc2018-01-10 15:52:33 +01003764 if (ref->type == QTYPE_QSTRING) {
3765 reference = ref->u.reference;
3766 } else {
3767 BlockdevOptions *options = &ref->u.definition;
3768 assert(ref->type == QTYPE_QDICT);
3769
3770 v = qobject_output_visitor_new(&obj);
Markus Armbruster1f584242020-04-24 10:43:35 +02003771 visit_type_BlockdevOptions(v, NULL, &options, &error_abort);
Kevin Wolfe1d74bc2018-01-10 15:52:33 +01003772 visit_complete(v, &obj);
3773
Max Reitz7dc847e2018-02-24 16:40:29 +01003774 qdict = qobject_to(QDict, obj);
Kevin Wolfe1d74bc2018-01-10 15:52:33 +01003775 qdict_flatten(qdict);
3776
3777 /* bdrv_open_inherit() defaults to the values in bdrv_flags (for
3778 * compatibility with other callers) rather than what we want as the
3779 * real defaults. Apply the defaults here instead. */
3780 qdict_set_default_str(qdict, BDRV_OPT_CACHE_DIRECT, "off");
3781 qdict_set_default_str(qdict, BDRV_OPT_CACHE_NO_FLUSH, "off");
3782 qdict_set_default_str(qdict, BDRV_OPT_READ_ONLY, "off");
Kevin Wolfe35bdc12018-10-05 18:57:40 +02003783 qdict_set_default_str(qdict, BDRV_OPT_AUTO_READ_ONLY, "off");
3784
Kevin Wolfe1d74bc2018-01-10 15:52:33 +01003785 }
3786
Max Reitz272c02e2020-05-13 13:05:17 +02003787 bs = bdrv_open_inherit(NULL, reference, qdict, 0, NULL, NULL, 0, errp);
Kevin Wolfe1d74bc2018-01-10 15:52:33 +01003788 obj = NULL;
Marc-André Lureaucb3e7f02018-04-19 17:01:43 +02003789 qobject_unref(obj);
Kevin Wolfe1d74bc2018-01-10 15:52:33 +01003790 visit_free(v);
3791 return bs;
3792}
3793
Max Reitz66836182016-05-17 16:41:27 +02003794static BlockDriverState *bdrv_append_temp_snapshot(BlockDriverState *bs,
3795 int flags,
3796 QDict *snapshot_options,
3797 Error **errp)
Kevin Wolfb9988752014-04-03 12:09:34 +02003798{
Bin Meng69fbfff2022-10-10 12:04:31 +08003799 g_autofree char *tmp_filename = NULL;
Kevin Wolfb9988752014-04-03 12:09:34 +02003800 int64_t total_size;
Chunyan Liu83d05212014-06-05 17:20:51 +08003801 QemuOpts *opts = NULL;
Eric Blakeff6ed712017-04-27 16:58:18 -05003802 BlockDriverState *bs_snapshot = NULL;
Kevin Wolff665f012023-06-05 10:57:07 +02003803 AioContext *ctx = bdrv_get_aio_context(bs);
Kevin Wolfb9988752014-04-03 12:09:34 +02003804 int ret;
3805
Emanuele Giuseppe Espositobdb73472022-03-03 10:16:02 -05003806 GLOBAL_STATE_CODE();
3807
Kevin Wolfb9988752014-04-03 12:09:34 +02003808 /* if snapshot, we create a temporary backing file and open it
3809 instead of opening 'filename' directly */
3810
3811 /* Get the required size from the image */
Kevin Wolff665f012023-06-05 10:57:07 +02003812 aio_context_acquire(ctx);
Kevin Wolff1877432014-04-04 17:07:19 +02003813 total_size = bdrv_getlength(bs);
Kevin Wolff665f012023-06-05 10:57:07 +02003814 aio_context_release(ctx);
3815
Kevin Wolff1877432014-04-04 17:07:19 +02003816 if (total_size < 0) {
3817 error_setg_errno(errp, -total_size, "Could not get image size");
Benoît Canet1ba4b6a2014-04-22 17:05:27 +02003818 goto out;
Kevin Wolff1877432014-04-04 17:07:19 +02003819 }
Kevin Wolfb9988752014-04-03 12:09:34 +02003820
3821 /* Create the temporary image */
Bin Meng69fbfff2022-10-10 12:04:31 +08003822 tmp_filename = create_tmp_file(errp);
3823 if (!tmp_filename) {
Benoît Canet1ba4b6a2014-04-22 17:05:27 +02003824 goto out;
Kevin Wolfb9988752014-04-03 12:09:34 +02003825 }
3826
Max Reitzef810432014-12-02 18:32:42 +01003827 opts = qemu_opts_create(bdrv_qcow2.create_opts, NULL, 0,
Chunyan Liuc282e1f2014-06-05 17:21:11 +08003828 &error_abort);
Markus Armbruster39101f22015-02-12 16:46:36 +01003829 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, total_size, &error_abort);
Markus Armbrustere43bfd92015-12-18 16:35:15 +01003830 ret = bdrv_create(&bdrv_qcow2, tmp_filename, opts, errp);
Chunyan Liu83d05212014-06-05 17:20:51 +08003831 qemu_opts_del(opts);
Kevin Wolfb9988752014-04-03 12:09:34 +02003832 if (ret < 0) {
Markus Armbrustere43bfd92015-12-18 16:35:15 +01003833 error_prepend(errp, "Could not create temporary overlay '%s': ",
3834 tmp_filename);
Benoît Canet1ba4b6a2014-04-22 17:05:27 +02003835 goto out;
Kevin Wolfb9988752014-04-03 12:09:34 +02003836 }
3837
Kevin Wolf73176be2016-03-07 13:02:15 +01003838 /* Prepare options QDict for the temporary file */
Eric Blake46f5ac22017-04-27 16:58:17 -05003839 qdict_put_str(snapshot_options, "file.driver", "file");
3840 qdict_put_str(snapshot_options, "file.filename", tmp_filename);
3841 qdict_put_str(snapshot_options, "driver", "qcow2");
Kevin Wolfb9988752014-04-03 12:09:34 +02003842
Max Reitz5b363932016-05-17 16:41:31 +02003843 bs_snapshot = bdrv_open(NULL, NULL, snapshot_options, flags, errp);
Kevin Wolf73176be2016-03-07 13:02:15 +01003844 snapshot_options = NULL;
Max Reitz5b363932016-05-17 16:41:31 +02003845 if (!bs_snapshot) {
Benoît Canet1ba4b6a2014-04-22 17:05:27 +02003846 goto out;
Kevin Wolfb9988752014-04-03 12:09:34 +02003847 }
3848
Kevin Wolff665f012023-06-05 10:57:07 +02003849 aio_context_acquire(ctx);
Vladimir Sementsov-Ogievskiy934aee12021-02-02 15:49:44 +03003850 ret = bdrv_append(bs_snapshot, bs, errp);
Kevin Wolff665f012023-06-05 10:57:07 +02003851 aio_context_release(ctx);
3852
Vladimir Sementsov-Ogievskiy934aee12021-02-02 15:49:44 +03003853 if (ret < 0) {
Eric Blakeff6ed712017-04-27 16:58:18 -05003854 bs_snapshot = NULL;
Kevin Wolfb2c28322017-02-20 12:46:42 +01003855 goto out;
3856 }
Benoît Canet1ba4b6a2014-04-22 17:05:27 +02003857
3858out:
Marc-André Lureaucb3e7f02018-04-19 17:01:43 +02003859 qobject_unref(snapshot_options);
Eric Blakeff6ed712017-04-27 16:58:18 -05003860 return bs_snapshot;
Kevin Wolfb9988752014-04-03 12:09:34 +02003861}
3862
Max Reitzda557aa2013-12-20 19:28:11 +01003863/*
Kevin Wolfb6ce07a2010-04-12 16:37:13 +02003864 * Opens a disk image (raw, qcow2, vmdk, ...)
Kevin Wolfde9c0ce2013-03-15 10:35:02 +01003865 *
3866 * options is a QDict of options to pass to the block drivers, or NULL for an
3867 * empty set of options. The reference to the QDict belongs to the block layer
3868 * after the call (even on failure), so if the caller intends to reuse the
Marc-André Lureaucb3e7f02018-04-19 17:01:43 +02003869 * dictionary, it needs to use qobject_ref() before calling bdrv_open.
Max Reitzf67503e2014-02-18 18:33:05 +01003870 *
3871 * If *pbs is NULL, a new BDS will be created with a pointer to it stored there.
3872 * If it is not NULL, the referenced BDS will be reused.
Max Reitzddf56362014-02-18 18:33:06 +01003873 *
3874 * The reference parameter may be used to specify an existing block device which
3875 * should be opened. If specified, neither options nor a filename may be given,
3876 * nor can an existing BDS be reused (that is, *pbs has to be NULL).
Emanuele Giuseppe Espositoc86422c2023-01-13 21:42:04 +01003877 *
Kevin Wolfae400db2023-05-25 14:47:03 +02003878 * The caller must always hold the main AioContext lock.
Kevin Wolfb6ce07a2010-04-12 16:37:13 +02003879 */
Kevin Wolf32192302023-01-26 18:24:32 +01003880static BlockDriverState * no_coroutine_fn
3881bdrv_open_inherit(const char *filename, const char *reference, QDict *options,
3882 int flags, BlockDriverState *parent,
3883 const BdrvChildClass *child_class, BdrvChildRole child_role,
3884 Error **errp)
bellardea2384d2004-08-01 21:59:26 +00003885{
Kevin Wolfb6ce07a2010-04-12 16:37:13 +02003886 int ret;
Kevin Wolf5696c6e2017-02-17 18:39:24 +01003887 BlockBackend *file = NULL;
Kevin Wolf9a4f4c32015-06-16 14:19:22 +02003888 BlockDriverState *bs;
Max Reitzce343772015-08-26 19:47:50 +02003889 BlockDriver *drv = NULL;
Alberto Garcia2f624b82018-06-29 14:37:00 +03003890 BdrvChild *child;
Kevin Wolf74fe54f2013-07-09 11:09:02 +02003891 const char *drvname;
Alberto Garcia3e8c2e52015-10-26 14:27:15 +02003892 const char *backing;
Max Reitz34b5d2c2013-09-05 14:45:29 +02003893 Error *local_err = NULL;
Kevin Wolf73176be2016-03-07 13:02:15 +01003894 QDict *snapshot_options = NULL;
Kevin Wolfb1e6fc02014-05-06 12:11:42 +02003895 int snapshot_flags = 0;
Kevin Wolff665f012023-06-05 10:57:07 +02003896 AioContext *ctx = qemu_get_aio_context();
bellard712e7872005-04-28 21:09:32 +00003897
Max Reitzbd86fb92020-05-13 13:05:13 +02003898 assert(!child_class || !flags);
3899 assert(!child_class == !parent);
Emanuele Giuseppe Espositof0c28322022-03-03 10:16:13 -05003900 GLOBAL_STATE_CODE();
Kevin Wolf32192302023-01-26 18:24:32 +01003901 assert(!qemu_in_coroutine());
Max Reitzf67503e2014-02-18 18:33:05 +01003902
Max Reitzddf56362014-02-18 18:33:06 +01003903 if (reference) {
3904 bool options_non_empty = options ? qdict_size(options) : false;
Marc-André Lureaucb3e7f02018-04-19 17:01:43 +02003905 qobject_unref(options);
Max Reitzddf56362014-02-18 18:33:06 +01003906
Max Reitzddf56362014-02-18 18:33:06 +01003907 if (filename || options_non_empty) {
3908 error_setg(errp, "Cannot reference an existing block device with "
3909 "additional options or a new filename");
Max Reitz5b363932016-05-17 16:41:31 +02003910 return NULL;
Max Reitzddf56362014-02-18 18:33:06 +01003911 }
3912
3913 bs = bdrv_lookup_bs(reference, reference, errp);
3914 if (!bs) {
Max Reitz5b363932016-05-17 16:41:31 +02003915 return NULL;
Max Reitzddf56362014-02-18 18:33:06 +01003916 }
Kevin Wolf76b22322016-04-04 17:11:13 +02003917
Max Reitzddf56362014-02-18 18:33:06 +01003918 bdrv_ref(bs);
Max Reitz5b363932016-05-17 16:41:31 +02003919 return bs;
Max Reitzddf56362014-02-18 18:33:06 +01003920 }
3921
Max Reitz5b363932016-05-17 16:41:31 +02003922 bs = bdrv_new();
Max Reitzf67503e2014-02-18 18:33:05 +01003923
Kevin Wolfde9c0ce2013-03-15 10:35:02 +01003924 /* NULL means an empty set of options */
3925 if (options == NULL) {
3926 options = qdict_new();
3927 }
3928
Kevin Wolf145f5982015-05-08 16:15:03 +02003929 /* json: syntax counts as explicit options, as if in the QDict */
Kevin Wolfde3b53f2015-10-29 15:24:41 +01003930 parse_json_protocol(options, &filename, &local_err);
3931 if (local_err) {
Kevin Wolfde3b53f2015-10-29 15:24:41 +01003932 goto fail;
3933 }
3934
Kevin Wolf145f5982015-05-08 16:15:03 +02003935 bs->explicit_options = qdict_clone_shallow(options);
3936
Max Reitzbd86fb92020-05-13 13:05:13 +02003937 if (child_class) {
Max Reitz3cdc69d2020-05-13 13:05:18 +02003938 bool parent_is_format;
3939
3940 if (parent->drv) {
3941 parent_is_format = parent->drv->is_format;
3942 } else {
3943 /*
3944 * parent->drv is not set yet because this node is opened for
3945 * (potential) format probing. That means that @parent is going
3946 * to be a format node.
3947 */
3948 parent_is_format = true;
3949 }
3950
Kevin Wolfbddcec32015-04-09 18:47:50 +02003951 bs->inherits_from = parent;
Max Reitz3cdc69d2020-05-13 13:05:18 +02003952 child_class->inherit_options(child_role, parent_is_format,
3953 &flags, options,
Max Reitzbd86fb92020-05-13 13:05:13 +02003954 parent->open_flags, parent->options);
Kevin Wolff3930ed2015-04-08 13:43:47 +02003955 }
3956
Kevin Wolfde3b53f2015-10-29 15:24:41 +01003957 ret = bdrv_fill_options(&options, filename, &flags, &local_err);
Philippe Mathieu-Daudédfde4832020-04-22 15:31:44 +02003958 if (ret < 0) {
Kevin Wolf462f5bc2014-05-26 11:39:55 +02003959 goto fail;
3960 }
3961
Markus Armbruster129c7d12017-03-30 19:43:12 +02003962 /*
3963 * Set the BDRV_O_RDWR and BDRV_O_ALLOW_RDWR flags.
3964 * Caution: getting a boolean member of @options requires care.
3965 * When @options come from -blockdev or blockdev_add, members are
3966 * typed according to the QAPI schema, but when they come from
3967 * -drive, they're all QString.
3968 */
Alberto Garciaf87a0e22016-09-15 17:53:02 +03003969 if (g_strcmp0(qdict_get_try_str(options, BDRV_OPT_READ_ONLY), "on") &&
3970 !qdict_get_try_bool(options, BDRV_OPT_READ_ONLY, false)) {
3971 flags |= (BDRV_O_RDWR | BDRV_O_ALLOW_RDWR);
3972 } else {
3973 flags &= ~BDRV_O_RDWR;
Alberto Garcia14499ea2016-09-15 17:53:00 +03003974 }
3975
3976 if (flags & BDRV_O_SNAPSHOT) {
3977 snapshot_options = qdict_new();
3978 bdrv_temp_snapshot_options(&snapshot_flags, snapshot_options,
3979 flags, options);
Alberto Garciaf87a0e22016-09-15 17:53:02 +03003980 /* Let bdrv_backing_options() override "read-only" */
3981 qdict_del(options, BDRV_OPT_READ_ONLY);
Max Reitz00ff7ff2020-05-13 13:05:21 +02003982 bdrv_inherited_options(BDRV_CHILD_COW, true,
3983 &flags, options, flags, options);
Alberto Garcia14499ea2016-09-15 17:53:00 +03003984 }
3985
Kevin Wolf62392eb2015-04-24 16:38:02 +02003986 bs->open_flags = flags;
3987 bs->options = options;
3988 options = qdict_clone_shallow(options);
3989
Kevin Wolf76c591b2014-06-04 14:19:44 +02003990 /* Find the right image format driver */
Markus Armbruster129c7d12017-03-30 19:43:12 +02003991 /* See cautionary note on accessing @options above */
Kevin Wolf76c591b2014-06-04 14:19:44 +02003992 drvname = qdict_get_try_str(options, "driver");
3993 if (drvname) {
3994 drv = bdrv_find_format(drvname);
Kevin Wolf76c591b2014-06-04 14:19:44 +02003995 if (!drv) {
3996 error_setg(errp, "Unknown driver: '%s'", drvname);
Kevin Wolf76c591b2014-06-04 14:19:44 +02003997 goto fail;
3998 }
3999 }
4000
4001 assert(drvname || !(flags & BDRV_O_PROTOCOL));
Kevin Wolf76c591b2014-06-04 14:19:44 +02004002
Markus Armbruster129c7d12017-03-30 19:43:12 +02004003 /* See cautionary note on accessing @options above */
Alberto Garcia3e8c2e52015-10-26 14:27:15 +02004004 backing = qdict_get_try_str(options, "backing");
Max Reitze59a0cf2018-02-24 16:40:32 +01004005 if (qobject_to(QNull, qdict_get(options, "backing")) != NULL ||
4006 (backing && *backing == '\0'))
4007 {
Max Reitz4f7be282018-02-24 16:40:33 +01004008 if (backing) {
4009 warn_report("Use of \"backing\": \"\" is deprecated; "
4010 "use \"backing\": null instead");
4011 }
Alberto Garcia3e8c2e52015-10-26 14:27:15 +02004012 flags |= BDRV_O_NO_BACKING;
Kevin Wolfae0f57f2019-11-08 09:36:35 +01004013 qdict_del(bs->explicit_options, "backing");
4014 qdict_del(bs->options, "backing");
Alberto Garcia3e8c2e52015-10-26 14:27:15 +02004015 qdict_del(options, "backing");
4016 }
4017
Kevin Wolf5696c6e2017-02-17 18:39:24 +01004018 /* Open image file without format layer. This BlockBackend is only used for
Kevin Wolf4e4bf5c2016-12-16 18:52:37 +01004019 * probing, the block drivers will do their own bdrv_open_child() for the
4020 * same BDS, which is why we put the node name back into options. */
Kevin Wolff4788ad2014-06-03 16:44:19 +02004021 if ((flags & BDRV_O_PROTOCOL) == 0) {
Kevin Wolf5696c6e2017-02-17 18:39:24 +01004022 BlockDriverState *file_bs;
4023
4024 file_bs = bdrv_open_child_bs(filename, options, "file", bs,
Max Reitz58944402020-05-13 13:05:37 +02004025 &child_of_bds, BDRV_CHILD_IMAGE,
4026 true, &local_err);
Kevin Wolf1fdd6932015-06-15 14:11:51 +02004027 if (local_err) {
Max Reitz5469a2a2014-02-18 18:33:10 +01004028 goto fail;
4029 }
Kevin Wolf5696c6e2017-02-17 18:39:24 +01004030 if (file_bs != NULL) {
Kevin Wolfdacaa162017-11-20 14:59:13 +01004031 /* Not requesting BLK_PERM_CONSISTENT_READ because we're only
4032 * looking at the header to guess the image format. This works even
4033 * in cases where a guest would not see a consistent state. */
Kevin Wolff665f012023-06-05 10:57:07 +02004034 ctx = bdrv_get_aio_context(file_bs);
4035 aio_context_acquire(ctx);
4036 file = blk_new(ctx, 0, BLK_PERM_ALL);
Kevin Wolfd7086422017-01-13 19:02:32 +01004037 blk_insert_bs(file, file_bs, &local_err);
Kevin Wolf5696c6e2017-02-17 18:39:24 +01004038 bdrv_unref(file_bs);
Kevin Wolff665f012023-06-05 10:57:07 +02004039 aio_context_release(ctx);
4040
Kevin Wolfd7086422017-01-13 19:02:32 +01004041 if (local_err) {
4042 goto fail;
4043 }
Kevin Wolf5696c6e2017-02-17 18:39:24 +01004044
Eric Blake46f5ac22017-04-27 16:58:17 -05004045 qdict_put_str(options, "file", bdrv_get_node_name(file_bs));
Kevin Wolf4e4bf5c2016-12-16 18:52:37 +01004046 }
Max Reitz5469a2a2014-02-18 18:33:10 +01004047 }
4048
Kevin Wolf76c591b2014-06-04 14:19:44 +02004049 /* Image format probing */
Kevin Wolf38f3ef52014-11-20 16:27:12 +01004050 bs->probed = !drv;
Kevin Wolf76c591b2014-06-04 14:19:44 +02004051 if (!drv && file) {
Kevin Wolfcf2ab8f2016-06-20 18:24:02 +02004052 ret = find_image_format(file, filename, &drv, &local_err);
Kevin Wolf17b005f2014-05-27 10:50:29 +02004053 if (ret < 0) {
Kevin Wolf8bfea152014-04-11 19:16:36 +02004054 goto fail;
Max Reitz2a05cbe2013-12-20 19:28:10 +01004055 }
Kevin Wolf62392eb2015-04-24 16:38:02 +02004056 /*
4057 * This option update would logically belong in bdrv_fill_options(),
4058 * but we first need to open bs->file for the probing to work, while
4059 * opening bs->file already requires the (mostly) final set of options
4060 * so that cache mode etc. can be inherited.
4061 *
4062 * Adding the driver later is somewhat ugly, but it's not an option
4063 * that would ever be inherited, so it's correct. We just need to make
4064 * sure to update both bs->options (which has the full effective
4065 * options for bs) and options (which has file.* already removed).
4066 */
Eric Blake46f5ac22017-04-27 16:58:17 -05004067 qdict_put_str(bs->options, "driver", drv->format_name);
4068 qdict_put_str(options, "driver", drv->format_name);
Kevin Wolf76c591b2014-06-04 14:19:44 +02004069 } else if (!drv) {
Kevin Wolf17b005f2014-05-27 10:50:29 +02004070 error_setg(errp, "Must specify either driver or file");
Kevin Wolf8bfea152014-04-11 19:16:36 +02004071 goto fail;
Kevin Wolff500a6d2012-11-12 17:35:27 +01004072 }
4073
Max Reitz53a29512015-03-19 14:53:16 -04004074 /* BDRV_O_PROTOCOL must be set iff a protocol BDS is about to be created */
4075 assert(!!(flags & BDRV_O_PROTOCOL) == !!drv->bdrv_file_open);
4076 /* file must be NULL if a protocol BDS is about to be created
4077 * (the inverse results in an error message from bdrv_open_common()) */
4078 assert(!(flags & BDRV_O_PROTOCOL) || !file);
4079
Kevin Wolfb6ce07a2010-04-12 16:37:13 +02004080 /* Open the image */
Kevin Wolf82dc8b42016-01-11 19:07:50 +01004081 ret = bdrv_open_common(bs, file, options, &local_err);
Kevin Wolfb6ce07a2010-04-12 16:37:13 +02004082 if (ret < 0) {
Kevin Wolf8bfea152014-04-11 19:16:36 +02004083 goto fail;
Christoph Hellwig69873072010-01-20 18:13:25 +01004084 }
4085
Kevin Wolff665f012023-06-05 10:57:07 +02004086 /* The AioContext could have changed during bdrv_open_common() */
4087 ctx = bdrv_get_aio_context(bs);
4088
Kevin Wolf4e4bf5c2016-12-16 18:52:37 +01004089 if (file) {
Kevin Wolff665f012023-06-05 10:57:07 +02004090 aio_context_acquire(ctx);
Kevin Wolf5696c6e2017-02-17 18:39:24 +01004091 blk_unref(file);
Kevin Wolff665f012023-06-05 10:57:07 +02004092 aio_context_release(ctx);
Kevin Wolff500a6d2012-11-12 17:35:27 +01004093 file = NULL;
4094 }
4095
Kevin Wolfb6ce07a2010-04-12 16:37:13 +02004096 /* If there is a backing file, use it */
Paolo Bonzini9156df12012-10-18 16:49:17 +02004097 if ((flags & BDRV_O_NO_BACKING) == 0) {
Kevin Wolfd9b7b052015-01-16 18:23:41 +01004098 ret = bdrv_open_backing_file(bs, options, "backing", &local_err);
Kevin Wolfb6ce07a2010-04-12 16:37:13 +02004099 if (ret < 0) {
Kevin Wolfb6ad4912013-03-15 10:35:04 +01004100 goto close_and_fail;
Kevin Wolfb6ce07a2010-04-12 16:37:13 +02004101 }
Kevin Wolfb6ce07a2010-04-12 16:37:13 +02004102 }
4103
Alberto Garcia50196d72018-09-06 12:37:03 +03004104 /* Remove all children options and references
4105 * from bs->options and bs->explicit_options */
Alberto Garcia2f624b82018-06-29 14:37:00 +03004106 QLIST_FOREACH(child, &bs->children, next) {
4107 char *child_key_dot;
4108 child_key_dot = g_strdup_printf("%s.", child->name);
4109 qdict_extract_subqdict(bs->explicit_options, NULL, child_key_dot);
4110 qdict_extract_subqdict(bs->options, NULL, child_key_dot);
Alberto Garcia50196d72018-09-06 12:37:03 +03004111 qdict_del(bs->explicit_options, child->name);
4112 qdict_del(bs->options, child->name);
Alberto Garcia2f624b82018-06-29 14:37:00 +03004113 g_free(child_key_dot);
4114 }
4115
Kevin Wolfb6ad4912013-03-15 10:35:04 +01004116 /* Check if any unknown options were used */
Paolo Bonzini7ad27572017-01-04 15:59:14 +01004117 if (qdict_size(options) != 0) {
Kevin Wolfb6ad4912013-03-15 10:35:04 +01004118 const QDictEntry *entry = qdict_first(options);
Max Reitz5acd9d82014-02-18 18:33:11 +01004119 if (flags & BDRV_O_PROTOCOL) {
4120 error_setg(errp, "Block protocol '%s' doesn't support the option "
4121 "'%s'", drv->format_name, entry->key);
4122 } else {
Max Reitzd0e46a52016-03-16 19:54:34 +01004123 error_setg(errp,
4124 "Block format '%s' does not support the option '%s'",
4125 drv->format_name, entry->key);
Max Reitz5acd9d82014-02-18 18:33:11 +01004126 }
Kevin Wolfb6ad4912013-03-15 10:35:04 +01004127
Kevin Wolfb6ad4912013-03-15 10:35:04 +01004128 goto close_and_fail;
4129 }
Kevin Wolfb6ad4912013-03-15 10:35:04 +01004130
Daniel P. Berrangec01c2142017-06-23 17:24:16 +01004131 bdrv_parent_cb_change_media(bs, true);
Kevin Wolfb6ce07a2010-04-12 16:37:13 +02004132
Marc-André Lureaucb3e7f02018-04-19 17:01:43 +02004133 qobject_unref(options);
Alberto Garcia8961be32018-09-06 17:25:41 +03004134 options = NULL;
Kevin Wolfdd62f1c2015-06-18 14:09:57 +02004135
4136 /* For snapshot=on, create a temporary qcow2 overlay. bs points to the
4137 * temporary snapshot afterwards. */
4138 if (snapshot_flags) {
Max Reitz66836182016-05-17 16:41:27 +02004139 BlockDriverState *snapshot_bs;
4140 snapshot_bs = bdrv_append_temp_snapshot(bs, snapshot_flags,
4141 snapshot_options, &local_err);
Kevin Wolf73176be2016-03-07 13:02:15 +01004142 snapshot_options = NULL;
Kevin Wolfdd62f1c2015-06-18 14:09:57 +02004143 if (local_err) {
4144 goto close_and_fail;
4145 }
Max Reitz5b363932016-05-17 16:41:31 +02004146 /* We are not going to return bs but the overlay on top of it
4147 * (snapshot_bs); thus, we have to drop the strong reference to bs
4148 * (which we obtained by calling bdrv_new()). bs will not be deleted,
4149 * though, because the overlay still has a reference to it. */
Kevin Wolff665f012023-06-05 10:57:07 +02004150 aio_context_acquire(ctx);
Max Reitz5b363932016-05-17 16:41:31 +02004151 bdrv_unref(bs);
Kevin Wolff665f012023-06-05 10:57:07 +02004152 aio_context_release(ctx);
Max Reitz5b363932016-05-17 16:41:31 +02004153 bs = snapshot_bs;
Max Reitz66836182016-05-17 16:41:27 +02004154 }
4155
Max Reitz5b363932016-05-17 16:41:31 +02004156 return bs;
Kevin Wolfb6ce07a2010-04-12 16:37:13 +02004157
Kevin Wolf8bfea152014-04-11 19:16:36 +02004158fail:
Kevin Wolff665f012023-06-05 10:57:07 +02004159 aio_context_acquire(ctx);
Kevin Wolf5696c6e2017-02-17 18:39:24 +01004160 blk_unref(file);
Marc-André Lureaucb3e7f02018-04-19 17:01:43 +02004161 qobject_unref(snapshot_options);
4162 qobject_unref(bs->explicit_options);
4163 qobject_unref(bs->options);
4164 qobject_unref(options);
Kevin Wolfde9c0ce2013-03-15 10:35:02 +01004165 bs->options = NULL;
Manos Pitsidianakis998cbd62017-07-14 17:35:47 +03004166 bs->explicit_options = NULL;
Max Reitz5b363932016-05-17 16:41:31 +02004167 bdrv_unref(bs);
Kevin Wolff665f012023-06-05 10:57:07 +02004168 aio_context_release(ctx);
Eduardo Habkost621ff942016-06-13 18:57:56 -03004169 error_propagate(errp, local_err);
Max Reitz5b363932016-05-17 16:41:31 +02004170 return NULL;
Kevin Wolfde9c0ce2013-03-15 10:35:02 +01004171
Kevin Wolfb6ad4912013-03-15 10:35:04 +01004172close_and_fail:
Kevin Wolff665f012023-06-05 10:57:07 +02004173 aio_context_acquire(ctx);
Max Reitz5b363932016-05-17 16:41:31 +02004174 bdrv_unref(bs);
Kevin Wolff665f012023-06-05 10:57:07 +02004175 aio_context_release(ctx);
Marc-André Lureaucb3e7f02018-04-19 17:01:43 +02004176 qobject_unref(snapshot_options);
4177 qobject_unref(options);
Eduardo Habkost621ff942016-06-13 18:57:56 -03004178 error_propagate(errp, local_err);
Max Reitz5b363932016-05-17 16:41:31 +02004179 return NULL;
Kevin Wolfb6ce07a2010-04-12 16:37:13 +02004180}
4181
Kevin Wolfae400db2023-05-25 14:47:03 +02004182/* The caller must always hold the main AioContext lock. */
Max Reitz5b363932016-05-17 16:41:31 +02004183BlockDriverState *bdrv_open(const char *filename, const char *reference,
4184 QDict *options, int flags, Error **errp)
Kevin Wolff3930ed2015-04-08 13:43:47 +02004185{
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05004186 GLOBAL_STATE_CODE();
4187
Max Reitz5b363932016-05-17 16:41:31 +02004188 return bdrv_open_inherit(filename, reference, options, flags, NULL,
Max Reitz272c02e2020-05-13 13:05:17 +02004189 NULL, 0, errp);
Kevin Wolff3930ed2015-04-08 13:43:47 +02004190}
4191
Alberto Garciafaf116b2019-03-12 18:48:49 +02004192/* Return true if the NULL-terminated @list contains @str */
4193static bool is_str_in_list(const char *str, const char *const *list)
4194{
4195 if (str && list) {
4196 int i;
4197 for (i = 0; list[i] != NULL; i++) {
4198 if (!strcmp(str, list[i])) {
4199 return true;
4200 }
4201 }
4202 }
4203 return false;
4204}
4205
4206/*
4207 * Check that every option set in @bs->options is also set in
4208 * @new_opts.
4209 *
4210 * Options listed in the common_options list and in
4211 * @bs->drv->mutable_opts are skipped.
4212 *
4213 * Return 0 on success, otherwise return -EINVAL and set @errp.
4214 */
4215static int bdrv_reset_options_allowed(BlockDriverState *bs,
4216 const QDict *new_opts, Error **errp)
4217{
4218 const QDictEntry *e;
4219 /* These options are common to all block drivers and are handled
4220 * in bdrv_reopen_prepare() so they can be left out of @new_opts */
4221 const char *const common_options[] = {
4222 "node-name", "discard", "cache.direct", "cache.no-flush",
4223 "read-only", "auto-read-only", "detect-zeroes", NULL
4224 };
4225
4226 for (e = qdict_first(bs->options); e; e = qdict_next(bs->options, e)) {
4227 if (!qdict_haskey(new_opts, e->key) &&
4228 !is_str_in_list(e->key, common_options) &&
4229 !is_str_in_list(e->key, bs->drv->mutable_opts)) {
4230 error_setg(errp, "Option '%s' cannot be reset "
4231 "to its default value", e->key);
4232 return -EINVAL;
4233 }
4234 }
4235
4236 return 0;
4237}
4238
Jeff Codye971aa12012-09-20 15:13:19 -04004239/*
Alberto Garciacb828c32019-03-12 18:48:47 +02004240 * Returns true if @child can be reached recursively from @bs
4241 */
4242static bool bdrv_recurse_has_child(BlockDriverState *bs,
4243 BlockDriverState *child)
4244{
4245 BdrvChild *c;
4246
4247 if (bs == child) {
4248 return true;
4249 }
4250
4251 QLIST_FOREACH(c, &bs->children, next) {
4252 if (bdrv_recurse_has_child(c->bs, child)) {
4253 return true;
4254 }
4255 }
4256
4257 return false;
4258}
4259
4260/*
Jeff Codye971aa12012-09-20 15:13:19 -04004261 * Adds a BlockDriverState to a simple queue for an atomic, transactional
4262 * reopen of multiple devices.
4263 *
Vladimir Sementsov-Ogievskiy859443b2019-09-27 15:23:47 +03004264 * bs_queue can either be an existing BlockReopenQueue that has had QTAILQ_INIT
Jeff Codye971aa12012-09-20 15:13:19 -04004265 * already performed, or alternatively may be NULL a new BlockReopenQueue will
4266 * be created and initialized. This newly created BlockReopenQueue should be
4267 * passed back in for subsequent calls that are intended to be of the same
4268 * atomic 'set'.
4269 *
4270 * bs is the BlockDriverState to add to the reopen queue.
4271 *
Kevin Wolf4d2cb092015-04-10 17:50:50 +02004272 * options contains the changed options for the associated bs
4273 * (the BlockReopenQueue takes ownership)
4274 *
Jeff Codye971aa12012-09-20 15:13:19 -04004275 * flags contains the open flags for the associated bs
4276 *
4277 * returns a pointer to bs_queue, which is either the newly allocated
4278 * bs_queue, or the existing bs_queue being used.
4279 *
Kevin Wolfd22933a2022-11-18 18:41:02 +01004280 * bs is drained here and undrained by bdrv_reopen_queue_free().
Kevin Wolf2e117862022-11-18 18:41:01 +01004281 *
4282 * To be called with bs->aio_context locked.
Jeff Codye971aa12012-09-20 15:13:19 -04004283 */
Kevin Wolf28518102015-05-08 17:07:31 +02004284static BlockReopenQueue *bdrv_reopen_queue_child(BlockReopenQueue *bs_queue,
4285 BlockDriverState *bs,
4286 QDict *options,
Max Reitzbd86fb92020-05-13 13:05:13 +02004287 const BdrvChildClass *klass,
Max Reitz272c02e2020-05-13 13:05:17 +02004288 BdrvChildRole role,
Max Reitz3cdc69d2020-05-13 13:05:18 +02004289 bool parent_is_format,
Kevin Wolf28518102015-05-08 17:07:31 +02004290 QDict *parent_options,
Alberto Garcia077e8e22019-03-12 18:48:44 +02004291 int parent_flags,
4292 bool keep_old_opts)
Jeff Codye971aa12012-09-20 15:13:19 -04004293{
4294 assert(bs != NULL);
4295
4296 BlockReopenQueueEntry *bs_entry;
Kevin Wolf67251a32015-04-09 18:54:04 +02004297 BdrvChild *child;
Alberto Garcia9aa09dd2018-11-12 16:00:45 +02004298 QDict *old_options, *explicit_options, *options_copy;
4299 int flags;
4300 QemuOpts *opts;
Kevin Wolf67251a32015-04-09 18:54:04 +02004301
Emanuele Giuseppe Espositof0c28322022-03-03 10:16:13 -05004302 GLOBAL_STATE_CODE();
Kevin Wolf1a63a902017-12-06 20:24:44 +01004303
Kevin Wolfd22933a2022-11-18 18:41:02 +01004304 bdrv_drained_begin(bs);
4305
Jeff Codye971aa12012-09-20 15:13:19 -04004306 if (bs_queue == NULL) {
4307 bs_queue = g_new0(BlockReopenQueue, 1);
Vladimir Sementsov-Ogievskiy859443b2019-09-27 15:23:47 +03004308 QTAILQ_INIT(bs_queue);
Jeff Codye971aa12012-09-20 15:13:19 -04004309 }
4310
Kevin Wolf4d2cb092015-04-10 17:50:50 +02004311 if (!options) {
4312 options = qdict_new();
4313 }
4314
Alberto Garcia5b7ba052016-09-15 17:53:03 +03004315 /* Check if this BlockDriverState is already in the queue */
Vladimir Sementsov-Ogievskiy859443b2019-09-27 15:23:47 +03004316 QTAILQ_FOREACH(bs_entry, bs_queue, entry) {
Alberto Garcia5b7ba052016-09-15 17:53:03 +03004317 if (bs == bs_entry->state.bs) {
4318 break;
4319 }
4320 }
4321
Kevin Wolf28518102015-05-08 17:07:31 +02004322 /*
4323 * Precedence of options:
4324 * 1. Explicitly passed in options (highest)
Alberto Garcia9aa09dd2018-11-12 16:00:45 +02004325 * 2. Retained from explicitly set options of bs
4326 * 3. Inherited from parent node
4327 * 4. Retained from effective options of bs
Kevin Wolf28518102015-05-08 17:07:31 +02004328 */
4329
Kevin Wolf145f5982015-05-08 16:15:03 +02004330 /* Old explicitly set values (don't overwrite by inherited value) */
Alberto Garcia077e8e22019-03-12 18:48:44 +02004331 if (bs_entry || keep_old_opts) {
4332 old_options = qdict_clone_shallow(bs_entry ?
4333 bs_entry->state.explicit_options :
4334 bs->explicit_options);
4335 bdrv_join_options(bs, options, old_options);
4336 qobject_unref(old_options);
Alberto Garcia5b7ba052016-09-15 17:53:03 +03004337 }
Kevin Wolf145f5982015-05-08 16:15:03 +02004338
4339 explicit_options = qdict_clone_shallow(options);
4340
Kevin Wolf28518102015-05-08 17:07:31 +02004341 /* Inherit from parent node */
4342 if (parent_options) {
Alberto Garcia9aa09dd2018-11-12 16:00:45 +02004343 flags = 0;
Max Reitz3cdc69d2020-05-13 13:05:18 +02004344 klass->inherit_options(role, parent_is_format, &flags, options,
Max Reitz272c02e2020-05-13 13:05:17 +02004345 parent_flags, parent_options);
Alberto Garcia9aa09dd2018-11-12 16:00:45 +02004346 } else {
4347 flags = bdrv_get_flags(bs);
Kevin Wolf28518102015-05-08 17:07:31 +02004348 }
4349
Alberto Garcia077e8e22019-03-12 18:48:44 +02004350 if (keep_old_opts) {
4351 /* Old values are used for options that aren't set yet */
4352 old_options = qdict_clone_shallow(bs->options);
4353 bdrv_join_options(bs, options, old_options);
4354 qobject_unref(old_options);
4355 }
Kevin Wolf4d2cb092015-04-10 17:50:50 +02004356
Alberto Garcia9aa09dd2018-11-12 16:00:45 +02004357 /* We have the final set of options so let's update the flags */
4358 options_copy = qdict_clone_shallow(options);
4359 opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
4360 qemu_opts_absorb_qdict(opts, options_copy, NULL);
4361 update_flags_from_options(&flags, opts);
4362 qemu_opts_del(opts);
4363 qobject_unref(options_copy);
4364
Kevin Wolffd452022017-08-03 17:02:59 +02004365 /* bdrv_open_inherit() sets and clears some additional flags internally */
Kevin Wolff1f25a22014-04-25 19:04:55 +02004366 flags &= ~BDRV_O_PROTOCOL;
Kevin Wolffd452022017-08-03 17:02:59 +02004367 if (flags & BDRV_O_RDWR) {
4368 flags |= BDRV_O_ALLOW_RDWR;
4369 }
Kevin Wolff1f25a22014-04-25 19:04:55 +02004370
Kevin Wolf1857c972017-09-14 14:53:46 +02004371 if (!bs_entry) {
4372 bs_entry = g_new0(BlockReopenQueueEntry, 1);
Vladimir Sementsov-Ogievskiy859443b2019-09-27 15:23:47 +03004373 QTAILQ_INSERT_TAIL(bs_queue, bs_entry, entry);
Kevin Wolf1857c972017-09-14 14:53:46 +02004374 } else {
Marc-André Lureaucb3e7f02018-04-19 17:01:43 +02004375 qobject_unref(bs_entry->state.options);
4376 qobject_unref(bs_entry->state.explicit_options);
Kevin Wolf1857c972017-09-14 14:53:46 +02004377 }
4378
4379 bs_entry->state.bs = bs;
4380 bs_entry->state.options = options;
4381 bs_entry->state.explicit_options = explicit_options;
4382 bs_entry->state.flags = flags;
4383
Alberto Garcia85466322019-03-12 18:48:45 +02004384 /*
4385 * If keep_old_opts is false then it means that unspecified
4386 * options must be reset to their original value. We don't allow
4387 * resetting 'backing' but we need to know if the option is
4388 * missing in order to decide if we have to return an error.
4389 */
4390 if (!keep_old_opts) {
4391 bs_entry->state.backing_missing =
4392 !qdict_haskey(options, "backing") &&
4393 !qdict_haskey(options, "backing.driver");
4394 }
4395
Kevin Wolf67251a32015-04-09 18:54:04 +02004396 QLIST_FOREACH(child, &bs->children, next) {
Alberto Garcia85466322019-03-12 18:48:45 +02004397 QDict *new_child_options = NULL;
4398 bool child_keep_old = keep_old_opts;
Kevin Wolf67251a32015-04-09 18:54:04 +02004399
Kevin Wolf4c9dfe52015-05-08 15:14:15 +02004400 /* reopen can only change the options of block devices that were
4401 * implicitly created and inherited options. For other (referenced)
4402 * block devices, a syntax like "backing.foo" results in an error. */
Kevin Wolf67251a32015-04-09 18:54:04 +02004403 if (child->bs->inherits_from != bs) {
4404 continue;
4405 }
4406
Alberto Garcia85466322019-03-12 18:48:45 +02004407 /* Check if the options contain a child reference */
4408 if (qdict_haskey(options, child->name)) {
4409 const char *childref = qdict_get_try_str(options, child->name);
4410 /*
4411 * The current child must not be reopened if the child
4412 * reference is null or points to a different node.
4413 */
4414 if (g_strcmp0(childref, child->bs->node_name)) {
4415 continue;
4416 }
4417 /*
4418 * If the child reference points to the current child then
4419 * reopen it with its existing set of options (note that
4420 * it can still inherit new options from the parent).
4421 */
4422 child_keep_old = true;
4423 } else {
4424 /* Extract child options ("child-name.*") */
4425 char *child_key_dot = g_strdup_printf("%s.", child->name);
4426 qdict_extract_subqdict(explicit_options, NULL, child_key_dot);
4427 qdict_extract_subqdict(options, &new_child_options, child_key_dot);
4428 g_free(child_key_dot);
4429 }
Kevin Wolf4c9dfe52015-05-08 15:14:15 +02004430
Alberto Garcia9aa09dd2018-11-12 16:00:45 +02004431 bdrv_reopen_queue_child(bs_queue, child->bs, new_child_options,
Max Reitz3cdc69d2020-05-13 13:05:18 +02004432 child->klass, child->role, bs->drv->is_format,
4433 options, flags, child_keep_old);
Jeff Codye971aa12012-09-20 15:13:19 -04004434 }
4435
Jeff Codye971aa12012-09-20 15:13:19 -04004436 return bs_queue;
4437}
4438
Kevin Wolf2e117862022-11-18 18:41:01 +01004439/* To be called with bs->aio_context locked */
Kevin Wolf28518102015-05-08 17:07:31 +02004440BlockReopenQueue *bdrv_reopen_queue(BlockReopenQueue *bs_queue,
4441 BlockDriverState *bs,
Alberto Garcia077e8e22019-03-12 18:48:44 +02004442 QDict *options, bool keep_old_opts)
Kevin Wolf28518102015-05-08 17:07:31 +02004443{
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05004444 GLOBAL_STATE_CODE();
4445
Max Reitz3cdc69d2020-05-13 13:05:18 +02004446 return bdrv_reopen_queue_child(bs_queue, bs, options, NULL, 0, false,
4447 NULL, 0, keep_old_opts);
Kevin Wolf28518102015-05-08 17:07:31 +02004448}
4449
Alberto Garciaab5b52282021-07-08 13:47:05 +02004450void bdrv_reopen_queue_free(BlockReopenQueue *bs_queue)
4451{
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05004452 GLOBAL_STATE_CODE();
Alberto Garciaab5b52282021-07-08 13:47:05 +02004453 if (bs_queue) {
4454 BlockReopenQueueEntry *bs_entry, *next;
4455 QTAILQ_FOREACH_SAFE(bs_entry, bs_queue, entry, next) {
Kevin Wolfd22933a2022-11-18 18:41:02 +01004456 AioContext *ctx = bdrv_get_aio_context(bs_entry->state.bs);
4457
4458 aio_context_acquire(ctx);
4459 bdrv_drained_end(bs_entry->state.bs);
4460 aio_context_release(ctx);
4461
Alberto Garciaab5b52282021-07-08 13:47:05 +02004462 qobject_unref(bs_entry->state.explicit_options);
4463 qobject_unref(bs_entry->state.options);
4464 g_free(bs_entry);
4465 }
4466 g_free(bs_queue);
4467 }
4468}
4469
Jeff Codye971aa12012-09-20 15:13:19 -04004470/*
4471 * Reopen multiple BlockDriverStates atomically & transactionally.
4472 *
4473 * The queue passed in (bs_queue) must have been built up previous
4474 * via bdrv_reopen_queue().
4475 *
4476 * Reopens all BDS specified in the queue, with the appropriate
4477 * flags. All devices are prepared for reopen, and failure of any
Stefan Weil50d6a8a2018-07-12 21:51:20 +02004478 * device will cause all device changes to be abandoned, and intermediate
Jeff Codye971aa12012-09-20 15:13:19 -04004479 * data cleaned up.
4480 *
4481 * If all devices prepare successfully, then the changes are committed
4482 * to all devices.
4483 *
Kevin Wolf1a63a902017-12-06 20:24:44 +01004484 * All affected nodes must be drained between bdrv_reopen_queue() and
4485 * bdrv_reopen_multiple().
Kevin Wolf6cf42ca2021-07-08 13:47:06 +02004486 *
4487 * To be called from the main thread, with all other AioContexts unlocked.
Jeff Codye971aa12012-09-20 15:13:19 -04004488 */
Alberto Garcia5019aec2019-03-12 18:48:50 +02004489int bdrv_reopen_multiple(BlockReopenQueue *bs_queue, Error **errp)
Jeff Codye971aa12012-09-20 15:13:19 -04004490{
4491 int ret = -1;
4492 BlockReopenQueueEntry *bs_entry, *next;
Kevin Wolf6cf42ca2021-07-08 13:47:06 +02004493 AioContext *ctx;
Vladimir Sementsov-Ogievskiy72373e42021-04-28 18:17:58 +03004494 Transaction *tran = tran_new();
Vladimir Sementsov-Ogievskiy72373e42021-04-28 18:17:58 +03004495 g_autoptr(GSList) refresh_list = NULL;
Jeff Codye971aa12012-09-20 15:13:19 -04004496
Kevin Wolf6cf42ca2021-07-08 13:47:06 +02004497 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
Jeff Codye971aa12012-09-20 15:13:19 -04004498 assert(bs_queue != NULL);
Emanuele Giuseppe Espositoda359902022-03-03 10:16:11 -05004499 GLOBAL_STATE_CODE();
Jeff Codye971aa12012-09-20 15:13:19 -04004500
Vladimir Sementsov-Ogievskiy859443b2019-09-27 15:23:47 +03004501 QTAILQ_FOREACH(bs_entry, bs_queue, entry) {
Kevin Wolf6cf42ca2021-07-08 13:47:06 +02004502 ctx = bdrv_get_aio_context(bs_entry->state.bs);
4503 aio_context_acquire(ctx);
Vladimir Sementsov-Ogievskiya2aabf82021-04-28 18:17:57 +03004504 ret = bdrv_flush(bs_entry->state.bs);
Kevin Wolf6cf42ca2021-07-08 13:47:06 +02004505 aio_context_release(ctx);
Vladimir Sementsov-Ogievskiya2aabf82021-04-28 18:17:57 +03004506 if (ret < 0) {
4507 error_setg_errno(errp, -ret, "Error flushing drive");
Kevin Wolfe3fc91a2021-05-03 13:05:55 +02004508 goto abort;
Vladimir Sementsov-Ogievskiya2aabf82021-04-28 18:17:57 +03004509 }
4510 }
4511
4512 QTAILQ_FOREACH(bs_entry, bs_queue, entry) {
Kevin Wolf1a63a902017-12-06 20:24:44 +01004513 assert(bs_entry->state.bs->quiesce_counter > 0);
Kevin Wolf6cf42ca2021-07-08 13:47:06 +02004514 ctx = bdrv_get_aio_context(bs_entry->state.bs);
4515 aio_context_acquire(ctx);
Vladimir Sementsov-Ogievskiy72373e42021-04-28 18:17:58 +03004516 ret = bdrv_reopen_prepare(&bs_entry->state, bs_queue, tran, errp);
Kevin Wolf6cf42ca2021-07-08 13:47:06 +02004517 aio_context_release(ctx);
Vladimir Sementsov-Ogievskiy72373e42021-04-28 18:17:58 +03004518 if (ret < 0) {
4519 goto abort;
Jeff Codye971aa12012-09-20 15:13:19 -04004520 }
4521 bs_entry->prepared = true;
4522 }
4523
Vladimir Sementsov-Ogievskiy859443b2019-09-27 15:23:47 +03004524 QTAILQ_FOREACH(bs_entry, bs_queue, entry) {
Kevin Wolf69b736e2019-03-05 17:18:22 +01004525 BDRVReopenState *state = &bs_entry->state;
Vladimir Sementsov-Ogievskiy72373e42021-04-28 18:17:58 +03004526
Vladimir Sementsov-Ogievskiyfb0ff4d2022-11-07 19:35:58 +03004527 refresh_list = g_slist_prepend(refresh_list, state->bs);
Vladimir Sementsov-Ogievskiy72373e42021-04-28 18:17:58 +03004528 if (state->old_backing_bs) {
Vladimir Sementsov-Ogievskiyfb0ff4d2022-11-07 19:35:58 +03004529 refresh_list = g_slist_prepend(refresh_list, state->old_backing_bs);
Kevin Wolf69b736e2019-03-05 17:18:22 +01004530 }
Alberto Garciaecd30d22021-06-10 15:05:36 +03004531 if (state->old_file_bs) {
Vladimir Sementsov-Ogievskiyfb0ff4d2022-11-07 19:35:58 +03004532 refresh_list = g_slist_prepend(refresh_list, state->old_file_bs);
Alberto Garciaecd30d22021-06-10 15:05:36 +03004533 }
Vladimir Sementsov-Ogievskiy72373e42021-04-28 18:17:58 +03004534 }
4535
4536 /*
4537 * Note that file-posix driver rely on permission update done during reopen
4538 * (even if no permission changed), because it wants "new" permissions for
4539 * reconfiguring the fd and that's why it does it in raw_check_perm(), not
4540 * in raw_reopen_prepare() which is called with "old" permissions.
4541 */
4542 ret = bdrv_list_refresh_perms(refresh_list, bs_queue, tran, errp);
4543 if (ret < 0) {
4544 goto abort;
Kevin Wolf69b736e2019-03-05 17:18:22 +01004545 }
4546
Vladimir Sementsov-Ogievskiyfcd6a4f2019-09-27 15:23:48 +03004547 /*
4548 * If we reach this point, we have success and just need to apply the
4549 * changes.
4550 *
4551 * Reverse order is used to comfort qcow2 driver: on commit it need to write
4552 * IN_USE flag to the image, to mark bitmaps in the image as invalid. But
4553 * children are usually goes after parents in reopen-queue, so go from last
4554 * to first element.
Jeff Codye971aa12012-09-20 15:13:19 -04004555 */
Vladimir Sementsov-Ogievskiyfcd6a4f2019-09-27 15:23:48 +03004556 QTAILQ_FOREACH_REVERSE(bs_entry, bs_queue, entry) {
Kevin Wolf6cf42ca2021-07-08 13:47:06 +02004557 ctx = bdrv_get_aio_context(bs_entry->state.bs);
4558 aio_context_acquire(ctx);
Jeff Codye971aa12012-09-20 15:13:19 -04004559 bdrv_reopen_commit(&bs_entry->state);
Kevin Wolf6cf42ca2021-07-08 13:47:06 +02004560 aio_context_release(ctx);
Jeff Codye971aa12012-09-20 15:13:19 -04004561 }
4562
Vladimir Sementsov-Ogievskiy72373e42021-04-28 18:17:58 +03004563 tran_commit(tran);
4564
4565 QTAILQ_FOREACH_REVERSE(bs_entry, bs_queue, entry) {
4566 BlockDriverState *bs = bs_entry->state.bs;
4567
4568 if (bs->drv->bdrv_reopen_commit_post) {
Kevin Wolf6cf42ca2021-07-08 13:47:06 +02004569 ctx = bdrv_get_aio_context(bs);
4570 aio_context_acquire(ctx);
Vladimir Sementsov-Ogievskiy72373e42021-04-28 18:17:58 +03004571 bs->drv->bdrv_reopen_commit_post(&bs_entry->state);
Kevin Wolf6cf42ca2021-07-08 13:47:06 +02004572 aio_context_release(ctx);
Vladimir Sementsov-Ogievskiy72373e42021-04-28 18:17:58 +03004573 }
4574 }
4575
Jeff Codye971aa12012-09-20 15:13:19 -04004576 ret = 0;
Vladimir Sementsov-Ogievskiy72373e42021-04-28 18:17:58 +03004577 goto cleanup;
4578
4579abort:
4580 tran_abort(tran);
Vladimir Sementsov-Ogievskiy859443b2019-09-27 15:23:47 +03004581 QTAILQ_FOREACH_SAFE(bs_entry, bs_queue, entry, next) {
Vladimir Sementsov-Ogievskiy72373e42021-04-28 18:17:58 +03004582 if (bs_entry->prepared) {
Kevin Wolf6cf42ca2021-07-08 13:47:06 +02004583 ctx = bdrv_get_aio_context(bs_entry->state.bs);
4584 aio_context_acquire(ctx);
Vladimir Sementsov-Ogievskiy72373e42021-04-28 18:17:58 +03004585 bdrv_reopen_abort(&bs_entry->state);
Kevin Wolf6cf42ca2021-07-08 13:47:06 +02004586 aio_context_release(ctx);
Kevin Wolf69b736e2019-03-05 17:18:22 +01004587 }
Kevin Wolf69b736e2019-03-05 17:18:22 +01004588 }
Peter Krempa17e1e2b2020-02-28 13:44:46 +01004589
Jeff Codye971aa12012-09-20 15:13:19 -04004590cleanup:
Alberto Garciaab5b52282021-07-08 13:47:05 +02004591 bdrv_reopen_queue_free(bs_queue);
Alberto Garcia40840e42016-10-28 10:08:03 +03004592
Jeff Codye971aa12012-09-20 15:13:19 -04004593 return ret;
4594}
4595
Kevin Wolf6cf42ca2021-07-08 13:47:06 +02004596int bdrv_reopen(BlockDriverState *bs, QDict *opts, bool keep_old_opts,
4597 Error **errp)
4598{
4599 AioContext *ctx = bdrv_get_aio_context(bs);
4600 BlockReopenQueue *queue;
4601 int ret;
4602
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05004603 GLOBAL_STATE_CODE();
4604
Kevin Wolf2e117862022-11-18 18:41:01 +01004605 queue = bdrv_reopen_queue(NULL, bs, opts, keep_old_opts);
4606
Kevin Wolf6cf42ca2021-07-08 13:47:06 +02004607 if (ctx != qemu_get_aio_context()) {
4608 aio_context_release(ctx);
4609 }
Kevin Wolf6cf42ca2021-07-08 13:47:06 +02004610 ret = bdrv_reopen_multiple(queue, errp);
4611
4612 if (ctx != qemu_get_aio_context()) {
4613 aio_context_acquire(ctx);
4614 }
Kevin Wolf6cf42ca2021-07-08 13:47:06 +02004615
4616 return ret;
4617}
4618
Alberto Garcia6e1000a2018-11-12 16:00:33 +02004619int bdrv_reopen_set_read_only(BlockDriverState *bs, bool read_only,
4620 Error **errp)
4621{
Alberto Garcia6e1000a2018-11-12 16:00:33 +02004622 QDict *opts = qdict_new();
4623
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05004624 GLOBAL_STATE_CODE();
4625
Alberto Garcia6e1000a2018-11-12 16:00:33 +02004626 qdict_put_bool(opts, BDRV_OPT_READ_ONLY, read_only);
4627
Kevin Wolf6cf42ca2021-07-08 13:47:06 +02004628 return bdrv_reopen(bs, opts, true, errp);
Alberto Garcia6e1000a2018-11-12 16:00:33 +02004629}
4630
Jeff Codye971aa12012-09-20 15:13:19 -04004631/*
Alberto Garciacb828c32019-03-12 18:48:47 +02004632 * Take a BDRVReopenState and check if the value of 'backing' in the
4633 * reopen_state->options QDict is valid or not.
4634 *
4635 * If 'backing' is missing from the QDict then return 0.
4636 *
4637 * If 'backing' contains the node name of the backing file of
4638 * reopen_state->bs then return 0.
4639 *
4640 * If 'backing' contains a different node name (or is null) then check
4641 * whether the current backing file can be replaced with the new one.
4642 * If that's the case then reopen_state->replace_backing_bs is set to
4643 * true and reopen_state->new_backing_bs contains a pointer to the new
4644 * backing BlockDriverState (or NULL).
4645 *
4646 * Return 0 on success, otherwise return < 0 and set @errp.
Kevin Wolf4b408662023-06-05 10:57:06 +02004647 *
4648 * The caller must hold the AioContext lock of @reopen_state->bs.
4649 * @reopen_state->bs can move to a different AioContext in this function.
4650 * Callers must make sure that their AioContext locking is still correct after
4651 * this.
Alberto Garciacb828c32019-03-12 18:48:47 +02004652 */
Alberto Garciaecd30d22021-06-10 15:05:36 +03004653static int bdrv_reopen_parse_file_or_backing(BDRVReopenState *reopen_state,
4654 bool is_backing, Transaction *tran,
4655 Error **errp)
Alberto Garciacb828c32019-03-12 18:48:47 +02004656{
4657 BlockDriverState *bs = reopen_state->bs;
Alberto Garciaecd30d22021-06-10 15:05:36 +03004658 BlockDriverState *new_child_bs;
4659 BlockDriverState *old_child_bs = is_backing ? child_bs(bs->backing) :
4660 child_bs(bs->file);
4661 const char *child_name = is_backing ? "backing" : "file";
Alberto Garciacb828c32019-03-12 18:48:47 +02004662 QObject *value;
4663 const char *str;
Kevin Wolf4b408662023-06-05 10:57:06 +02004664 AioContext *ctx, *old_ctx;
4665 int ret;
Alberto Garciacb828c32019-03-12 18:48:47 +02004666
Emanuele Giuseppe Espositobdb73472022-03-03 10:16:02 -05004667 GLOBAL_STATE_CODE();
4668
Alberto Garciaecd30d22021-06-10 15:05:36 +03004669 value = qdict_get(reopen_state->options, child_name);
Alberto Garciacb828c32019-03-12 18:48:47 +02004670 if (value == NULL) {
4671 return 0;
4672 }
4673
4674 switch (qobject_type(value)) {
4675 case QTYPE_QNULL:
Alberto Garciaecd30d22021-06-10 15:05:36 +03004676 assert(is_backing); /* The 'file' option does not allow a null value */
4677 new_child_bs = NULL;
Alberto Garciacb828c32019-03-12 18:48:47 +02004678 break;
4679 case QTYPE_QSTRING:
Markus Armbruster410f44f2020-12-11 18:11:42 +01004680 str = qstring_get_str(qobject_to(QString, value));
Alberto Garciaecd30d22021-06-10 15:05:36 +03004681 new_child_bs = bdrv_lookup_bs(NULL, str, errp);
4682 if (new_child_bs == NULL) {
Alberto Garciacb828c32019-03-12 18:48:47 +02004683 return -EINVAL;
Alberto Garciaecd30d22021-06-10 15:05:36 +03004684 } else if (bdrv_recurse_has_child(new_child_bs, bs)) {
4685 error_setg(errp, "Making '%s' a %s child of '%s' would create a "
4686 "cycle", str, child_name, bs->node_name);
Alberto Garciacb828c32019-03-12 18:48:47 +02004687 return -EINVAL;
4688 }
4689 break;
4690 default:
Alberto Garciaecd30d22021-06-10 15:05:36 +03004691 /*
4692 * The options QDict has been flattened, so 'backing' and 'file'
4693 * do not allow any other data type here.
4694 */
Alberto Garciacb828c32019-03-12 18:48:47 +02004695 g_assert_not_reached();
4696 }
4697
Alberto Garciaecd30d22021-06-10 15:05:36 +03004698 if (old_child_bs == new_child_bs) {
4699 return 0;
4700 }
4701
4702 if (old_child_bs) {
4703 if (bdrv_skip_implicit_filters(old_child_bs) == new_child_bs) {
Vladimir Sementsov-Ogievskiycbfdb982021-06-10 15:05:33 +03004704 return 0;
4705 }
4706
Alberto Garciaecd30d22021-06-10 15:05:36 +03004707 if (old_child_bs->implicit) {
4708 error_setg(errp, "Cannot replace implicit %s child of %s",
4709 child_name, bs->node_name);
Vladimir Sementsov-Ogievskiycbfdb982021-06-10 15:05:33 +03004710 return -EPERM;
4711 }
4712 }
4713
Alberto Garciaecd30d22021-06-10 15:05:36 +03004714 if (bs->drv->is_filter && !old_child_bs) {
Vladimir Sementsov-Ogievskiy25f78d92021-06-10 15:05:34 +03004715 /*
4716 * Filters always have a file or a backing child, so we are trying to
4717 * change wrong child
4718 */
4719 error_setg(errp, "'%s' is a %s filter node that does not support a "
Alberto Garciaecd30d22021-06-10 15:05:36 +03004720 "%s child", bs->node_name, bs->drv->format_name, child_name);
Max Reitz1d42f482019-06-12 17:24:39 +02004721 return -EINVAL;
4722 }
4723
Alberto Garciaecd30d22021-06-10 15:05:36 +03004724 if (is_backing) {
4725 reopen_state->old_backing_bs = old_child_bs;
4726 } else {
4727 reopen_state->old_file_bs = old_child_bs;
4728 }
4729
Kevin Wolf4b408662023-06-05 10:57:06 +02004730 old_ctx = bdrv_get_aio_context(bs);
4731 ctx = bdrv_get_aio_context(new_child_bs);
4732 if (old_ctx != ctx) {
4733 aio_context_release(old_ctx);
4734 aio_context_acquire(ctx);
4735 }
4736
4737 ret = bdrv_set_file_or_backing_noperm(bs, new_child_bs, is_backing,
4738 tran, errp);
4739
4740 if (old_ctx != ctx) {
4741 aio_context_release(ctx);
4742 aio_context_acquire(old_ctx);
4743 }
4744
4745 return ret;
Alberto Garciacb828c32019-03-12 18:48:47 +02004746}
4747
4748/*
Jeff Codye971aa12012-09-20 15:13:19 -04004749 * Prepares a BlockDriverState for reopen. All changes are staged in the
4750 * 'opaque' field of the BDRVReopenState, which is used and allocated by
4751 * the block driver layer .bdrv_reopen_prepare()
4752 *
4753 * bs is the BlockDriverState to reopen
4754 * flags are the new open flags
4755 * queue is the reopen queue
4756 *
4757 * Returns 0 on success, non-zero on error. On error errp will be set
4758 * as well.
4759 *
4760 * On failure, bdrv_reopen_abort() will be called to clean up any data.
4761 * It is the responsibility of the caller to then call the abort() or
4762 * commit() for any other BDS that have been left in a prepare() state
4763 *
Kevin Wolf4b408662023-06-05 10:57:06 +02004764 * The caller must hold the AioContext lock of @reopen_state->bs.
Jeff Codye971aa12012-09-20 15:13:19 -04004765 */
Vladimir Sementsov-Ogievskiy53e96d12021-04-28 18:17:35 +03004766static int bdrv_reopen_prepare(BDRVReopenState *reopen_state,
Vladimir Sementsov-Ogievskiy72373e42021-04-28 18:17:58 +03004767 BlockReopenQueue *queue,
Alberto Garciaecd30d22021-06-10 15:05:36 +03004768 Transaction *change_child_tran, Error **errp)
Jeff Codye971aa12012-09-20 15:13:19 -04004769{
4770 int ret = -1;
Alberto Garciae6d79c42018-11-12 16:00:47 +02004771 int old_flags;
Jeff Codye971aa12012-09-20 15:13:19 -04004772 Error *local_err = NULL;
4773 BlockDriver *drv;
Kevin Wolfccf9dc02015-05-08 17:24:56 +02004774 QemuOpts *opts;
Alberto Garcia4c8350f2018-06-29 14:37:02 +03004775 QDict *orig_reopen_opts;
Alberto Garcia593b3072018-09-06 12:37:08 +03004776 char *discard = NULL;
Jeff Cody3d8ce172017-04-07 16:55:30 -04004777 bool read_only;
Max Reitz9ad08c42018-11-16 17:45:24 +01004778 bool drv_prepared = false;
Jeff Codye971aa12012-09-20 15:13:19 -04004779
4780 assert(reopen_state != NULL);
4781 assert(reopen_state->bs->drv != NULL);
Emanuele Giuseppe Espositoda359902022-03-03 10:16:11 -05004782 GLOBAL_STATE_CODE();
Jeff Codye971aa12012-09-20 15:13:19 -04004783 drv = reopen_state->bs->drv;
4784
Alberto Garcia4c8350f2018-06-29 14:37:02 +03004785 /* This function and each driver's bdrv_reopen_prepare() remove
4786 * entries from reopen_state->options as they are processed, so
4787 * we need to make a copy of the original QDict. */
4788 orig_reopen_opts = qdict_clone_shallow(reopen_state->options);
4789
Kevin Wolfccf9dc02015-05-08 17:24:56 +02004790 /* Process generic block layer options */
4791 opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
Markus Armbrusteraf175e82020-07-07 18:06:03 +02004792 if (!qemu_opts_absorb_qdict(opts, reopen_state->options, errp)) {
Kevin Wolfccf9dc02015-05-08 17:24:56 +02004793 ret = -EINVAL;
4794 goto error;
4795 }
4796
Alberto Garciae6d79c42018-11-12 16:00:47 +02004797 /* This was already called in bdrv_reopen_queue_child() so the flags
4798 * are up-to-date. This time we simply want to remove the options from
4799 * QemuOpts in order to indicate that they have been processed. */
4800 old_flags = reopen_state->flags;
Kevin Wolf91a097e2015-05-08 17:49:53 +02004801 update_flags_from_options(&reopen_state->flags, opts);
Alberto Garciae6d79c42018-11-12 16:00:47 +02004802 assert(old_flags == reopen_state->flags);
Kevin Wolf91a097e2015-05-08 17:49:53 +02004803
Alberto Garcia415bbca2018-10-03 13:23:13 +03004804 discard = qemu_opt_get_del(opts, BDRV_OPT_DISCARD);
Alberto Garcia593b3072018-09-06 12:37:08 +03004805 if (discard != NULL) {
4806 if (bdrv_parse_discard_flags(discard, &reopen_state->flags) != 0) {
4807 error_setg(errp, "Invalid discard option");
4808 ret = -EINVAL;
4809 goto error;
4810 }
4811 }
4812
Alberto Garcia543770b2018-09-06 12:37:09 +03004813 reopen_state->detect_zeroes =
4814 bdrv_parse_detect_zeroes(opts, reopen_state->flags, &local_err);
4815 if (local_err) {
4816 error_propagate(errp, local_err);
4817 ret = -EINVAL;
4818 goto error;
4819 }
4820
Alberto Garcia57f9db92018-09-06 12:37:06 +03004821 /* All other options (including node-name and driver) must be unchanged.
4822 * Put them back into the QDict, so that they are checked at the end
4823 * of this function. */
4824 qemu_opts_to_qdict(opts, reopen_state->options);
Kevin Wolfccf9dc02015-05-08 17:24:56 +02004825
Jeff Cody3d8ce172017-04-07 16:55:30 -04004826 /* If we are to stay read-only, do not allow permission change
4827 * to r/w. Attempting to set to r/w may fail if either BDRV_O_ALLOW_RDWR is
4828 * not set, or if the BDS still has copy_on_read enabled */
4829 read_only = !(reopen_state->flags & BDRV_O_RDWR);
Kevin Wolf54a32bf2017-08-03 17:02:58 +02004830 ret = bdrv_can_set_read_only(reopen_state->bs, read_only, true, &local_err);
Jeff Cody3d8ce172017-04-07 16:55:30 -04004831 if (local_err) {
4832 error_propagate(errp, local_err);
Jeff Codye971aa12012-09-20 15:13:19 -04004833 goto error;
4834 }
4835
Jeff Codye971aa12012-09-20 15:13:19 -04004836 if (drv->bdrv_reopen_prepare) {
Alberto Garciafaf116b2019-03-12 18:48:49 +02004837 /*
4838 * If a driver-specific option is missing, it means that we
4839 * should reset it to its default value.
4840 * But not all options allow that, so we need to check it first.
4841 */
4842 ret = bdrv_reset_options_allowed(reopen_state->bs,
4843 reopen_state->options, errp);
4844 if (ret) {
4845 goto error;
4846 }
4847
Jeff Codye971aa12012-09-20 15:13:19 -04004848 ret = drv->bdrv_reopen_prepare(reopen_state, queue, &local_err);
4849 if (ret) {
4850 if (local_err != NULL) {
4851 error_propagate(errp, local_err);
4852 } else {
Max Reitzf30c66b2019-02-01 20:29:05 +01004853 bdrv_refresh_filename(reopen_state->bs);
Luiz Capitulinod8b68952013-06-10 11:29:27 -04004854 error_setg(errp, "failed while preparing to reopen image '%s'",
4855 reopen_state->bs->filename);
Jeff Codye971aa12012-09-20 15:13:19 -04004856 }
4857 goto error;
4858 }
4859 } else {
4860 /* It is currently mandatory to have a bdrv_reopen_prepare()
4861 * handler for each supported drv. */
Alberto Garcia81e5f782015-04-08 12:29:19 +03004862 error_setg(errp, "Block format '%s' used by node '%s' "
4863 "does not support reopening files", drv->format_name,
4864 bdrv_get_device_or_node_name(reopen_state->bs));
Jeff Codye971aa12012-09-20 15:13:19 -04004865 ret = -1;
4866 goto error;
4867 }
4868
Max Reitz9ad08c42018-11-16 17:45:24 +01004869 drv_prepared = true;
4870
Alberto Garciabacd9b82019-03-12 18:48:46 +02004871 /*
4872 * We must provide the 'backing' option if the BDS has a backing
4873 * file or if the image file has a backing file name as part of
4874 * its metadata. Otherwise the 'backing' option can be omitted.
4875 */
4876 if (drv->supports_backing && reopen_state->backing_missing &&
Max Reitz1d42f482019-06-12 17:24:39 +02004877 (reopen_state->bs->backing || reopen_state->bs->backing_file[0])) {
Alberto Garcia85466322019-03-12 18:48:45 +02004878 error_setg(errp, "backing is missing for '%s'",
4879 reopen_state->bs->node_name);
4880 ret = -EINVAL;
4881 goto error;
4882 }
4883
Alberto Garciacb828c32019-03-12 18:48:47 +02004884 /*
4885 * Allow changing the 'backing' option. The new value can be
4886 * either a reference to an existing node (using its node name)
4887 * or NULL to simply detach the current backing file.
4888 */
Alberto Garciaecd30d22021-06-10 15:05:36 +03004889 ret = bdrv_reopen_parse_file_or_backing(reopen_state, true,
4890 change_child_tran, errp);
Alberto Garciacb828c32019-03-12 18:48:47 +02004891 if (ret < 0) {
4892 goto error;
4893 }
4894 qdict_del(reopen_state->options, "backing");
4895
Alberto Garciaecd30d22021-06-10 15:05:36 +03004896 /* Allow changing the 'file' option. In this case NULL is not allowed */
4897 ret = bdrv_reopen_parse_file_or_backing(reopen_state, false,
4898 change_child_tran, errp);
4899 if (ret < 0) {
4900 goto error;
4901 }
4902 qdict_del(reopen_state->options, "file");
4903
Kevin Wolf4d2cb092015-04-10 17:50:50 +02004904 /* Options that are not handled are only okay if they are unchanged
4905 * compared to the old state. It is expected that some options are only
4906 * used for the initial open, but not reopen (e.g. filename) */
4907 if (qdict_size(reopen_state->options)) {
4908 const QDictEntry *entry = qdict_first(reopen_state->options);
4909
4910 do {
Max Reitz54fd1b02017-11-14 19:01:26 +01004911 QObject *new = entry->value;
4912 QObject *old = qdict_get(reopen_state->bs->options, entry->key);
Kevin Wolf4d2cb092015-04-10 17:50:50 +02004913
Alberto Garciadb905282018-09-06 12:37:05 +03004914 /* Allow child references (child_name=node_name) as long as they
4915 * point to the current child (i.e. everything stays the same). */
4916 if (qobject_type(new) == QTYPE_QSTRING) {
4917 BdrvChild *child;
4918 QLIST_FOREACH(child, &reopen_state->bs->children, next) {
4919 if (!strcmp(child->name, entry->key)) {
4920 break;
4921 }
4922 }
4923
4924 if (child) {
Markus Armbruster410f44f2020-12-11 18:11:42 +01004925 if (!strcmp(child->bs->node_name,
4926 qstring_get_str(qobject_to(QString, new)))) {
Alberto Garciadb905282018-09-06 12:37:05 +03004927 continue; /* Found child with this name, skip option */
4928 }
4929 }
4930 }
4931
Max Reitz54fd1b02017-11-14 19:01:26 +01004932 /*
4933 * TODO: When using -drive to specify blockdev options, all values
4934 * will be strings; however, when using -blockdev, blockdev-add or
4935 * filenames using the json:{} pseudo-protocol, they will be
4936 * correctly typed.
4937 * In contrast, reopening options are (currently) always strings
4938 * (because you can only specify them through qemu-io; all other
4939 * callers do not specify any options).
4940 * Therefore, when using anything other than -drive to create a BDS,
4941 * this cannot detect non-string options as unchanged, because
4942 * qobject_is_equal() always returns false for objects of different
4943 * type. In the future, this should be remedied by correctly typing
4944 * all options. For now, this is not too big of an issue because
4945 * the user can simply omit options which cannot be changed anyway,
4946 * so they will stay unchanged.
4947 */
4948 if (!qobject_is_equal(new, old)) {
Kevin Wolf4d2cb092015-04-10 17:50:50 +02004949 error_setg(errp, "Cannot change the option '%s'", entry->key);
4950 ret = -EINVAL;
4951 goto error;
4952 }
4953 } while ((entry = qdict_next(reopen_state->options, entry)));
4954 }
4955
Jeff Codye971aa12012-09-20 15:13:19 -04004956 ret = 0;
4957
Alberto Garcia4c8350f2018-06-29 14:37:02 +03004958 /* Restore the original reopen_state->options QDict */
4959 qobject_unref(reopen_state->options);
4960 reopen_state->options = qobject_ref(orig_reopen_opts);
4961
Jeff Codye971aa12012-09-20 15:13:19 -04004962error:
Max Reitz9ad08c42018-11-16 17:45:24 +01004963 if (ret < 0 && drv_prepared) {
4964 /* drv->bdrv_reopen_prepare() has succeeded, so we need to
4965 * call drv->bdrv_reopen_abort() before signaling an error
4966 * (bdrv_reopen_multiple() will not call bdrv_reopen_abort()
4967 * when the respective bdrv_reopen_prepare() has failed) */
4968 if (drv->bdrv_reopen_abort) {
4969 drv->bdrv_reopen_abort(reopen_state);
4970 }
4971 }
Kevin Wolfccf9dc02015-05-08 17:24:56 +02004972 qemu_opts_del(opts);
Alberto Garcia4c8350f2018-06-29 14:37:02 +03004973 qobject_unref(orig_reopen_opts);
Alberto Garcia593b3072018-09-06 12:37:08 +03004974 g_free(discard);
Jeff Codye971aa12012-09-20 15:13:19 -04004975 return ret;
4976}
4977
4978/*
4979 * Takes the staged changes for the reopen from bdrv_reopen_prepare(), and
4980 * makes them final by swapping the staging BlockDriverState contents into
4981 * the active BlockDriverState contents.
4982 */
Vladimir Sementsov-Ogievskiy53e96d12021-04-28 18:17:35 +03004983static void bdrv_reopen_commit(BDRVReopenState *reopen_state)
Jeff Codye971aa12012-09-20 15:13:19 -04004984{
4985 BlockDriver *drv;
Vladimir Sementsov-Ogievskiy50bf65b2017-06-28 15:05:12 +03004986 BlockDriverState *bs;
Alberto Garcia50196d72018-09-06 12:37:03 +03004987 BdrvChild *child;
Jeff Codye971aa12012-09-20 15:13:19 -04004988
4989 assert(reopen_state != NULL);
Vladimir Sementsov-Ogievskiy50bf65b2017-06-28 15:05:12 +03004990 bs = reopen_state->bs;
4991 drv = bs->drv;
Jeff Codye971aa12012-09-20 15:13:19 -04004992 assert(drv != NULL);
Emanuele Giuseppe Espositoda359902022-03-03 10:16:11 -05004993 GLOBAL_STATE_CODE();
Jeff Codye971aa12012-09-20 15:13:19 -04004994
4995 /* If there are any driver level actions to take */
4996 if (drv->bdrv_reopen_commit) {
4997 drv->bdrv_reopen_commit(reopen_state);
4998 }
4999
5000 /* set BDS specific flags now */
Marc-André Lureaucb3e7f02018-04-19 17:01:43 +02005001 qobject_unref(bs->explicit_options);
Alberto Garcia4c8350f2018-06-29 14:37:02 +03005002 qobject_unref(bs->options);
Alberto Garciaab5b52282021-07-08 13:47:05 +02005003 qobject_ref(reopen_state->explicit_options);
5004 qobject_ref(reopen_state->options);
Kevin Wolf145f5982015-05-08 16:15:03 +02005005
Vladimir Sementsov-Ogievskiy50bf65b2017-06-28 15:05:12 +03005006 bs->explicit_options = reopen_state->explicit_options;
Alberto Garcia4c8350f2018-06-29 14:37:02 +03005007 bs->options = reopen_state->options;
Vladimir Sementsov-Ogievskiy50bf65b2017-06-28 15:05:12 +03005008 bs->open_flags = reopen_state->flags;
Alberto Garcia543770b2018-09-06 12:37:09 +03005009 bs->detect_zeroes = reopen_state->detect_zeroes;
Kevin Wolf355ef4a2013-12-11 20:14:09 +01005010
Alberto Garcia50196d72018-09-06 12:37:03 +03005011 /* Remove child references from bs->options and bs->explicit_options.
5012 * Child options were already removed in bdrv_reopen_queue_child() */
5013 QLIST_FOREACH(child, &bs->children, next) {
5014 qdict_del(bs->explicit_options, child->name);
5015 qdict_del(bs->options, child->name);
5016 }
Vladimir Sementsov-Ogievskiy3d0e8742021-06-10 15:05:35 +03005017 /* backing is probably removed, so it's not handled by previous loop */
5018 qdict_del(bs->explicit_options, "backing");
5019 qdict_del(bs->options, "backing");
5020
Kevin Wolfe19b1572023-05-04 13:57:50 +02005021 bdrv_graph_rdlock_main_loop();
Vladimir Sementsov-Ogievskiy1e4c7972021-04-28 18:17:55 +03005022 bdrv_refresh_limits(bs, NULL, NULL);
Kevin Wolfe19b1572023-05-04 13:57:50 +02005023 bdrv_graph_rdunlock_main_loop();
Paolo Bonzini439cc332023-04-07 17:32:58 +02005024 bdrv_refresh_total_sectors(bs, bs->total_sectors);
Jeff Codye971aa12012-09-20 15:13:19 -04005025}
5026
5027/*
5028 * Abort the reopen, and delete and free the staged changes in
5029 * reopen_state
5030 */
Vladimir Sementsov-Ogievskiy53e96d12021-04-28 18:17:35 +03005031static void bdrv_reopen_abort(BDRVReopenState *reopen_state)
Jeff Codye971aa12012-09-20 15:13:19 -04005032{
5033 BlockDriver *drv;
5034
5035 assert(reopen_state != NULL);
5036 drv = reopen_state->bs->drv;
5037 assert(drv != NULL);
Emanuele Giuseppe Espositoda359902022-03-03 10:16:11 -05005038 GLOBAL_STATE_CODE();
Jeff Codye971aa12012-09-20 15:13:19 -04005039
5040 if (drv->bdrv_reopen_abort) {
5041 drv->bdrv_reopen_abort(reopen_state);
5042 }
5043}
5044
5045
Max Reitz64dff522016-01-29 16:36:10 +01005046static void bdrv_close(BlockDriverState *bs)
bellardfc01f7e2003-06-30 10:03:06 +00005047{
Max Reitz33384422014-06-20 21:57:33 +02005048 BdrvAioNotifier *ban, *ban_next;
Alberto Garcia50a3efb2017-11-06 16:53:45 +02005049 BdrvChild *child, *next;
Max Reitz33384422014-06-20 21:57:33 +02005050
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05005051 GLOBAL_STATE_CODE();
Max Reitz30f55fb2016-05-17 16:41:32 +02005052 assert(!bs->refcnt);
Alberto Garcia99b7e772015-09-25 16:41:44 +03005053
Paolo Bonzinifc272912015-12-23 11:48:24 +01005054 bdrv_drained_begin(bs); /* complete I/O */
Stefan Hajnoczi58fda172013-07-02 15:36:25 +02005055 bdrv_flush(bs);
Fam Zheng53ec73e2015-05-29 18:53:14 +08005056 bdrv_drain(bs); /* in case flush left pending I/O */
Paolo Bonzinifc272912015-12-23 11:48:24 +01005057
Paolo Bonzini3cbc0022012-10-19 11:36:48 +02005058 if (bs->drv) {
Vladimir Sementsov-Ogievskiy3c005292018-08-14 15:43:19 +03005059 if (bs->drv->bdrv_close) {
Max Reitz7b99a262019-06-12 16:07:11 +02005060 /* Must unfreeze all children, so bdrv_unref_child() works */
Vladimir Sementsov-Ogievskiy3c005292018-08-14 15:43:19 +03005061 bs->drv->bdrv_close(bs);
5062 }
Kevin Wolf9a4f4c32015-06-16 14:19:22 +02005063 bs->drv = NULL;
bellardb3380822004-03-14 21:38:54 +00005064 }
Zhi Yong Wu98f90db2011-11-08 13:00:14 +08005065
Alberto Garcia50a3efb2017-11-06 16:53:45 +02005066 QLIST_FOREACH_SAFE(child, &bs->children, next, next) {
Alberto Garciadd4118c2019-05-13 16:46:17 +03005067 bdrv_unref_child(bs, child);
Alberto Garcia50a3efb2017-11-06 16:53:45 +02005068 }
5069
Vladimir Sementsov-Ogievskiy5bb047472022-07-26 23:11:32 +03005070 assert(!bs->backing);
5071 assert(!bs->file);
Alberto Garcia50a3efb2017-11-06 16:53:45 +02005072 g_free(bs->opaque);
5073 bs->opaque = NULL;
Stefan Hajnoczid73415a2020-09-23 11:56:46 +01005074 qatomic_set(&bs->copy_on_read, 0);
Alberto Garcia50a3efb2017-11-06 16:53:45 +02005075 bs->backing_file[0] = '\0';
5076 bs->backing_format[0] = '\0';
5077 bs->total_sectors = 0;
5078 bs->encrypted = false;
5079 bs->sg = false;
Marc-André Lureaucb3e7f02018-04-19 17:01:43 +02005080 qobject_unref(bs->options);
5081 qobject_unref(bs->explicit_options);
Alberto Garcia50a3efb2017-11-06 16:53:45 +02005082 bs->options = NULL;
5083 bs->explicit_options = NULL;
Marc-André Lureaucb3e7f02018-04-19 17:01:43 +02005084 qobject_unref(bs->full_open_options);
Alberto Garcia50a3efb2017-11-06 16:53:45 +02005085 bs->full_open_options = NULL;
Hanna Reitz0bc329f2021-08-12 10:41:44 +02005086 g_free(bs->block_status_cache);
5087 bs->block_status_cache = NULL;
Alberto Garcia50a3efb2017-11-06 16:53:45 +02005088
Vladimir Sementsov-Ogievskiycca43ae2017-06-28 15:05:16 +03005089 bdrv_release_named_dirty_bitmaps(bs);
5090 assert(QLIST_EMPTY(&bs->dirty_bitmaps));
5091
Max Reitz33384422014-06-20 21:57:33 +02005092 QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_next) {
5093 g_free(ban);
5094 }
5095 QLIST_INIT(&bs->aio_notifiers);
Paolo Bonzinifc272912015-12-23 11:48:24 +01005096 bdrv_drained_end(bs);
Greg Kurz1a6d3bd2020-10-23 17:01:10 +02005097
5098 /*
5099 * If we're still inside some bdrv_drain_all_begin()/end() sections, end
5100 * them now since this BDS won't exist anymore when bdrv_drain_all_end()
5101 * gets called.
5102 */
5103 if (bs->quiesce_counter) {
5104 bdrv_drain_all_end_quiesce(bs);
5105 }
bellardb3380822004-03-14 21:38:54 +00005106}
5107
MORITA Kazutaka2bc93fe2010-05-28 11:44:57 +09005108void bdrv_close_all(void)
5109{
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05005110 GLOBAL_STATE_CODE();
Emanuele Giuseppe Esposito880eeec2022-09-26 05:32:04 -04005111 assert(job_next(NULL) == NULL);
MORITA Kazutaka2bc93fe2010-05-28 11:44:57 +09005112
Max Reitzca9bd242016-01-29 16:36:14 +01005113 /* Drop references from requests still in flight, such as canceled block
5114 * jobs whose AIO context has not been polled yet */
5115 bdrv_drain_all();
Stefan Hajnoczied78cda2014-05-08 16:34:35 +02005116
Max Reitzca9bd242016-01-29 16:36:14 +01005117 blk_remove_all_bs();
5118 blockdev_close_all_bdrv_states();
5119
Kevin Wolfa1a2af02016-04-08 18:26:37 +02005120 assert(QTAILQ_EMPTY(&all_bdrv_states));
MORITA Kazutaka2bc93fe2010-05-28 11:44:57 +09005121}
5122
Kevin Wolfd0ac0382017-03-01 17:30:41 +01005123static bool should_update_child(BdrvChild *c, BlockDriverState *to)
Kevin Wolfdd62f1c2015-06-18 14:09:57 +02005124{
Vladimir Sementsov-Ogievskiy2f30b7c2019-02-23 22:20:39 +03005125 GQueue *queue;
5126 GHashTable *found;
5127 bool ret;
Kevin Wolfdd62f1c2015-06-18 14:09:57 +02005128
Max Reitzbd86fb92020-05-13 13:05:13 +02005129 if (c->klass->stay_at_node) {
Kevin Wolfd0ac0382017-03-01 17:30:41 +01005130 return false;
5131 }
5132
Max Reitzec9f10f2018-06-13 20:18:15 +02005133 /* If the child @c belongs to the BDS @to, replacing the current
5134 * c->bs by @to would mean to create a loop.
5135 *
5136 * Such a case occurs when appending a BDS to a backing chain.
5137 * For instance, imagine the following chain:
5138 *
5139 * guest device -> node A -> further backing chain...
5140 *
5141 * Now we create a new BDS B which we want to put on top of this
5142 * chain, so we first attach A as its backing node:
5143 *
5144 * node B
5145 * |
5146 * v
5147 * guest device -> node A -> further backing chain...
5148 *
5149 * Finally we want to replace A by B. When doing that, we want to
5150 * replace all pointers to A by pointers to B -- except for the
5151 * pointer from B because (1) that would create a loop, and (2)
5152 * that pointer should simply stay intact:
5153 *
5154 * guest device -> node B
5155 * |
5156 * v
5157 * node A -> further backing chain...
5158 *
5159 * In general, when replacing a node A (c->bs) by a node B (@to),
5160 * if A is a child of B, that means we cannot replace A by B there
5161 * because that would create a loop. Silently detaching A from B
5162 * is also not really an option. So overall just leaving A in
Vladimir Sementsov-Ogievskiy2f30b7c2019-02-23 22:20:39 +03005163 * place there is the most sensible choice.
5164 *
5165 * We would also create a loop in any cases where @c is only
5166 * indirectly referenced by @to. Prevent this by returning false
5167 * if @c is found (by breadth-first search) anywhere in the whole
5168 * subtree of @to.
5169 */
5170
5171 ret = true;
5172 found = g_hash_table_new(NULL, NULL);
5173 g_hash_table_add(found, to);
5174 queue = g_queue_new();
5175 g_queue_push_tail(queue, to);
5176
5177 while (!g_queue_is_empty(queue)) {
5178 BlockDriverState *v = g_queue_pop_head(queue);
5179 BdrvChild *c2;
5180
5181 QLIST_FOREACH(c2, &v->children, next) {
5182 if (c2 == c) {
5183 ret = false;
5184 break;
5185 }
5186
5187 if (g_hash_table_contains(found, c2->bs)) {
5188 continue;
5189 }
5190
5191 g_queue_push_tail(queue, c2->bs);
5192 g_hash_table_add(found, c2->bs);
Kevin Wolfd0ac0382017-03-01 17:30:41 +01005193 }
5194 }
5195
Vladimir Sementsov-Ogievskiy2f30b7c2019-02-23 22:20:39 +03005196 g_queue_free(queue);
5197 g_hash_table_destroy(found);
5198
5199 return ret;
Kevin Wolfd0ac0382017-03-01 17:30:41 +01005200}
5201
Vladimir Sementsov-Ogievskiy57f08942022-07-26 23:11:34 +03005202static void bdrv_remove_child_commit(void *opaque)
Vladimir Sementsov-Ogievskiy46541ee2021-04-28 18:17:50 +03005203{
Emanuele Giuseppe Espositobdb73472022-03-03 10:16:02 -05005204 GLOBAL_STATE_CODE();
Vladimir Sementsov-Ogievskiy5bb047472022-07-26 23:11:32 +03005205 bdrv_child_free(opaque);
Vladimir Sementsov-Ogievskiy46541ee2021-04-28 18:17:50 +03005206}
5207
Vladimir Sementsov-Ogievskiy57f08942022-07-26 23:11:34 +03005208static TransactionActionDrv bdrv_remove_child_drv = {
5209 .commit = bdrv_remove_child_commit,
Vladimir Sementsov-Ogievskiy46541ee2021-04-28 18:17:50 +03005210};
5211
Vladimir Sementsov-Ogievskiy57f08942022-07-26 23:11:34 +03005212/* Function doesn't update permissions, caller is responsible for this. */
5213static void bdrv_remove_child(BdrvChild *child, Transaction *tran)
Vladimir Sementsov-Ogievskiy46541ee2021-04-28 18:17:50 +03005214{
Vladimir Sementsov-Ogievskiy46541ee2021-04-28 18:17:50 +03005215 if (!child) {
5216 return;
5217 }
5218
5219 if (child->bs) {
Kevin Wolf23987472022-11-18 18:41:09 +01005220 BlockDriverState *bs = child->bs;
5221 bdrv_drained_begin(bs);
Vladimir Sementsov-Ogievskiya2c37a32022-07-26 23:11:30 +03005222 bdrv_replace_child_tran(child, NULL, tran);
Kevin Wolf23987472022-11-18 18:41:09 +01005223 bdrv_drained_end(bs);
Vladimir Sementsov-Ogievskiy46541ee2021-04-28 18:17:50 +03005224 }
5225
Vladimir Sementsov-Ogievskiy57f08942022-07-26 23:11:34 +03005226 tran_add(tran, &bdrv_remove_child_drv, child);
Vladimir Sementsov-Ogievskiy46541ee2021-04-28 18:17:50 +03005227}
5228
Kevin Wolf23987472022-11-18 18:41:09 +01005229static void undrain_on_clean_cb(void *opaque)
5230{
5231 bdrv_drained_end(opaque);
5232}
5233
5234static TransactionActionDrv undrain_on_clean = {
5235 .clean = undrain_on_clean_cb,
5236};
5237
Vladimir Sementsov-Ogievskiy117caba2021-04-28 18:17:48 +03005238static int bdrv_replace_node_noperm(BlockDriverState *from,
5239 BlockDriverState *to,
5240 bool auto_skip, Transaction *tran,
5241 Error **errp)
5242{
5243 BdrvChild *c, *next;
5244
Emanuele Giuseppe Espositobdb73472022-03-03 10:16:02 -05005245 GLOBAL_STATE_CODE();
Hanna Reitz82b54cf2021-11-15 15:54:04 +01005246
Kevin Wolf23987472022-11-18 18:41:09 +01005247 bdrv_drained_begin(from);
5248 bdrv_drained_begin(to);
5249 tran_add(tran, &undrain_on_clean, from);
5250 tran_add(tran, &undrain_on_clean, to);
5251
Vladimir Sementsov-Ogievskiy117caba2021-04-28 18:17:48 +03005252 QLIST_FOREACH_SAFE(c, &from->parents, next_parent, next) {
5253 assert(c->bs == from);
5254 if (!should_update_child(c, to)) {
5255 if (auto_skip) {
5256 continue;
5257 }
5258 error_setg(errp, "Should not change '%s' link to '%s'",
5259 c->name, from->node_name);
5260 return -EINVAL;
5261 }
5262 if (c->frozen) {
5263 error_setg(errp, "Cannot change '%s' link to '%s'",
5264 c->name, from->node_name);
5265 return -EPERM;
5266 }
Vladimir Sementsov-Ogievskiy0f0b1e22022-07-26 23:11:29 +03005267 bdrv_replace_child_tran(c, to, tran);
Vladimir Sementsov-Ogievskiy117caba2021-04-28 18:17:48 +03005268 }
5269
5270 return 0;
5271}
5272
Vladimir Sementsov-Ogievskiy313274b2020-11-06 15:42:36 +03005273/*
5274 * With auto_skip=true bdrv_replace_node_common skips updating from parents
5275 * if it creates a parent-child relation loop or if parent is block-job.
5276 *
5277 * With auto_skip=false the error is returned if from has a parent which should
5278 * not be updated.
Vladimir Sementsov-Ogievskiy3108a152021-04-28 18:17:51 +03005279 *
5280 * With @detach_subchain=true @to must be in a backing chain of @from. In this
5281 * case backing link of the cow-parent of @to is removed.
Vladimir Sementsov-Ogievskiy313274b2020-11-06 15:42:36 +03005282 */
Vladimir Sementsov-Ogievskiya1e708f2021-02-02 15:49:43 +03005283static int bdrv_replace_node_common(BlockDriverState *from,
5284 BlockDriverState *to,
Vladimir Sementsov-Ogievskiy3108a152021-04-28 18:17:51 +03005285 bool auto_skip, bool detach_subchain,
5286 Error **errp)
Kevin Wolfdd62f1c2015-06-18 14:09:57 +02005287{
Vladimir Sementsov-Ogievskiy3bb0e292021-04-28 18:17:45 +03005288 Transaction *tran = tran_new();
Vladimir Sementsov-Ogievskiy3bb0e292021-04-28 18:17:45 +03005289 g_autoptr(GSList) refresh_list = NULL;
Miroslav Rezanina2d369d62021-05-05 03:59:03 -04005290 BlockDriverState *to_cow_parent = NULL;
Kevin Wolf234ac1a2017-03-02 18:43:00 +01005291 int ret;
Kevin Wolfdd62f1c2015-06-18 14:09:57 +02005292
Emanuele Giuseppe Espositobdb73472022-03-03 10:16:02 -05005293 GLOBAL_STATE_CODE();
Hanna Reitz82b54cf2021-11-15 15:54:04 +01005294
Vladimir Sementsov-Ogievskiy3108a152021-04-28 18:17:51 +03005295 if (detach_subchain) {
5296 assert(bdrv_chain_contains(from, to));
5297 assert(from != to);
5298 for (to_cow_parent = from;
5299 bdrv_filter_or_cow_bs(to_cow_parent) != to;
5300 to_cow_parent = bdrv_filter_or_cow_bs(to_cow_parent))
5301 {
5302 ;
5303 }
5304 }
5305
Kevin Wolf234ac1a2017-03-02 18:43:00 +01005306 /* Make sure that @from doesn't go away until we have successfully attached
5307 * all of its parents to @to. */
5308 bdrv_ref(from);
5309
Kevin Wolff871abd2019-05-21 19:00:25 +02005310 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
Kevin Wolf30dd65f2020-03-10 12:38:29 +01005311 assert(bdrv_get_aio_context(from) == bdrv_get_aio_context(to));
Kevin Wolff871abd2019-05-21 19:00:25 +02005312 bdrv_drained_begin(from);
5313
Vladimir Sementsov-Ogievskiy3bb0e292021-04-28 18:17:45 +03005314 /*
5315 * Do the replacement without permission update.
5316 * Replacement may influence the permissions, we should calculate new
5317 * permissions based on new graph. If we fail, we'll roll-back the
5318 * replacement.
5319 */
Vladimir Sementsov-Ogievskiy117caba2021-04-28 18:17:48 +03005320 ret = bdrv_replace_node_noperm(from, to, auto_skip, tran, errp);
5321 if (ret < 0) {
5322 goto out;
Kevin Wolf234ac1a2017-03-02 18:43:00 +01005323 }
5324
Vladimir Sementsov-Ogievskiy3108a152021-04-28 18:17:51 +03005325 if (detach_subchain) {
Vladimir Sementsov-Ogievskiyf38eaec2022-11-07 19:35:56 +03005326 bdrv_remove_child(bdrv_filter_or_cow_child(to_cow_parent), tran);
Vladimir Sementsov-Ogievskiy3108a152021-04-28 18:17:51 +03005327 }
5328
Vladimir Sementsov-Ogievskiyfb0ff4d2022-11-07 19:35:58 +03005329 refresh_list = g_slist_prepend(refresh_list, to);
5330 refresh_list = g_slist_prepend(refresh_list, from);
Vladimir Sementsov-Ogievskiy3bb0e292021-04-28 18:17:45 +03005331
5332 ret = bdrv_list_refresh_perms(refresh_list, NULL, tran, errp);
Kevin Wolf234ac1a2017-03-02 18:43:00 +01005333 if (ret < 0) {
Kevin Wolf234ac1a2017-03-02 18:43:00 +01005334 goto out;
5335 }
5336
Vladimir Sementsov-Ogievskiya1e708f2021-02-02 15:49:43 +03005337 ret = 0;
5338
Kevin Wolf234ac1a2017-03-02 18:43:00 +01005339out:
Vladimir Sementsov-Ogievskiy3bb0e292021-04-28 18:17:45 +03005340 tran_finalize(tran, ret);
5341
Kevin Wolff871abd2019-05-21 19:00:25 +02005342 bdrv_drained_end(from);
Kevin Wolf234ac1a2017-03-02 18:43:00 +01005343 bdrv_unref(from);
Vladimir Sementsov-Ogievskiya1e708f2021-02-02 15:49:43 +03005344
5345 return ret;
Kevin Wolfdd62f1c2015-06-18 14:09:57 +02005346}
5347
Vladimir Sementsov-Ogievskiya1e708f2021-02-02 15:49:43 +03005348int bdrv_replace_node(BlockDriverState *from, BlockDriverState *to,
5349 Error **errp)
Vladimir Sementsov-Ogievskiy313274b2020-11-06 15:42:36 +03005350{
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05005351 GLOBAL_STATE_CODE();
5352
Vladimir Sementsov-Ogievskiy3108a152021-04-28 18:17:51 +03005353 return bdrv_replace_node_common(from, to, true, false, errp);
5354}
5355
5356int bdrv_drop_filter(BlockDriverState *bs, Error **errp)
5357{
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05005358 GLOBAL_STATE_CODE();
5359
Vladimir Sementsov-Ogievskiy3108a152021-04-28 18:17:51 +03005360 return bdrv_replace_node_common(bs, bdrv_filter_or_cow_bs(bs), true, true,
5361 errp);
Vladimir Sementsov-Ogievskiy313274b2020-11-06 15:42:36 +03005362}
5363
Jeff Cody8802d1f2012-02-28 15:54:06 -05005364/*
5365 * Add new bs contents at the top of an image chain while the chain is
5366 * live, while keeping required fields on the top layer.
5367 *
5368 * This will modify the BlockDriverState fields, and swap contents
5369 * between bs_new and bs_top. Both bs_new and bs_top are modified.
5370 *
Vladimir Sementsov-Ogievskiy2272edc2021-04-28 18:17:49 +03005371 * bs_new must not be attached to a BlockBackend and must not have backing
5372 * child.
Jeff Codyf6801b82012-03-27 16:30:19 -04005373 *
Jeff Cody8802d1f2012-02-28 15:54:06 -05005374 * This function does not create any image files.
Stefano Garzarella60d90bf2023-02-14 18:16:21 +01005375 *
5376 * The caller must hold the AioContext lock for @bs_top.
Jeff Cody8802d1f2012-02-28 15:54:06 -05005377 */
Vladimir Sementsov-Ogievskiya1e708f2021-02-02 15:49:43 +03005378int bdrv_append(BlockDriverState *bs_new, BlockDriverState *bs_top,
5379 Error **errp)
Jeff Cody8802d1f2012-02-28 15:54:06 -05005380{
Vladimir Sementsov-Ogievskiy2272edc2021-04-28 18:17:49 +03005381 int ret;
Vladimir Sementsov-Ogievskiy5bb047472022-07-26 23:11:32 +03005382 BdrvChild *child;
Vladimir Sementsov-Ogievskiy2272edc2021-04-28 18:17:49 +03005383 Transaction *tran = tran_new();
Stefano Garzarella60d90bf2023-02-14 18:16:21 +01005384 AioContext *old_context, *new_context = NULL;
Vladimir Sementsov-Ogievskiy2272edc2021-04-28 18:17:49 +03005385
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05005386 GLOBAL_STATE_CODE();
5387
Vladimir Sementsov-Ogievskiy2272edc2021-04-28 18:17:49 +03005388 assert(!bs_new->backing);
5389
Stefano Garzarella60d90bf2023-02-14 18:16:21 +01005390 old_context = bdrv_get_aio_context(bs_top);
5391
Vladimir Sementsov-Ogievskiy5bb047472022-07-26 23:11:32 +03005392 child = bdrv_attach_child_noperm(bs_new, bs_top, "backing",
5393 &child_of_bds, bdrv_backing_role(bs_new),
5394 tran, errp);
5395 if (!child) {
5396 ret = -EINVAL;
Vladimir Sementsov-Ogievskiy2272edc2021-04-28 18:17:49 +03005397 goto out;
Kevin Wolfb2c28322017-02-20 12:46:42 +01005398 }
Kevin Wolfdd62f1c2015-06-18 14:09:57 +02005399
Stefano Garzarella60d90bf2023-02-14 18:16:21 +01005400 /*
5401 * bdrv_attach_child_noperm could change the AioContext of bs_top.
5402 * bdrv_replace_node_noperm calls bdrv_drained_begin, so let's temporarily
5403 * hold the new AioContext, since bdrv_drained_begin calls BDRV_POLL_WHILE
5404 * that assumes the new lock is taken.
5405 */
5406 new_context = bdrv_get_aio_context(bs_top);
5407
5408 if (old_context != new_context) {
5409 aio_context_release(old_context);
5410 aio_context_acquire(new_context);
5411 }
5412
Vladimir Sementsov-Ogievskiy2272edc2021-04-28 18:17:49 +03005413 ret = bdrv_replace_node_noperm(bs_top, bs_new, true, tran, errp);
Vladimir Sementsov-Ogievskiya1e708f2021-02-02 15:49:43 +03005414 if (ret < 0) {
Vladimir Sementsov-Ogievskiy2272edc2021-04-28 18:17:49 +03005415 goto out;
Kevin Wolf234ac1a2017-03-02 18:43:00 +01005416 }
Kevin Wolfdd62f1c2015-06-18 14:09:57 +02005417
Vladimir Sementsov-Ogievskiyf1316ed2022-11-07 19:35:57 +03005418 ret = bdrv_refresh_perms(bs_new, tran, errp);
Vladimir Sementsov-Ogievskiy2272edc2021-04-28 18:17:49 +03005419out:
5420 tran_finalize(tran, ret);
5421
Kevin Wolfe19b1572023-05-04 13:57:50 +02005422 bdrv_graph_rdlock_main_loop();
Vladimir Sementsov-Ogievskiy1e4c7972021-04-28 18:17:55 +03005423 bdrv_refresh_limits(bs_top, NULL, NULL);
Kevin Wolfe19b1572023-05-04 13:57:50 +02005424 bdrv_graph_rdunlock_main_loop();
Vladimir Sementsov-Ogievskiy2272edc2021-04-28 18:17:49 +03005425
Stefano Garzarella60d90bf2023-02-14 18:16:21 +01005426 if (new_context && old_context != new_context) {
5427 aio_context_release(new_context);
5428 aio_context_acquire(old_context);
5429 }
5430
Vladimir Sementsov-Ogievskiy2272edc2021-04-28 18:17:49 +03005431 return ret;
Jeff Cody8802d1f2012-02-28 15:54:06 -05005432}
5433
Vladimir Sementsov-Ogievskiybd8f4c42021-08-24 11:38:23 +03005434/* Not for empty child */
5435int bdrv_replace_child_bs(BdrvChild *child, BlockDriverState *new_bs,
5436 Error **errp)
5437{
5438 int ret;
5439 Transaction *tran = tran_new();
Vladimir Sementsov-Ogievskiybd8f4c42021-08-24 11:38:23 +03005440 g_autoptr(GSList) refresh_list = NULL;
5441 BlockDriverState *old_bs = child->bs;
5442
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05005443 GLOBAL_STATE_CODE();
5444
Vladimir Sementsov-Ogievskiybd8f4c42021-08-24 11:38:23 +03005445 bdrv_ref(old_bs);
5446 bdrv_drained_begin(old_bs);
5447 bdrv_drained_begin(new_bs);
5448
Vladimir Sementsov-Ogievskiy0f0b1e22022-07-26 23:11:29 +03005449 bdrv_replace_child_tran(child, new_bs, tran);
Vladimir Sementsov-Ogievskiybd8f4c42021-08-24 11:38:23 +03005450
Vladimir Sementsov-Ogievskiyfb0ff4d2022-11-07 19:35:58 +03005451 refresh_list = g_slist_prepend(refresh_list, old_bs);
5452 refresh_list = g_slist_prepend(refresh_list, new_bs);
Vladimir Sementsov-Ogievskiybd8f4c42021-08-24 11:38:23 +03005453
5454 ret = bdrv_list_refresh_perms(refresh_list, NULL, tran, errp);
5455
5456 tran_finalize(tran, ret);
5457
5458 bdrv_drained_end(old_bs);
5459 bdrv_drained_end(new_bs);
5460 bdrv_unref(old_bs);
5461
5462 return ret;
5463}
5464
Fam Zheng4f6fd342013-08-23 09:14:47 +08005465static void bdrv_delete(BlockDriverState *bs)
bellardb3380822004-03-14 21:38:54 +00005466{
Fam Zheng3718d8a2014-05-23 21:29:43 +08005467 assert(bdrv_op_blocker_is_empty(bs));
Fam Zheng4f6fd342013-08-23 09:14:47 +08005468 assert(!bs->refcnt);
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05005469 GLOBAL_STATE_CODE();
Markus Armbruster18846de2010-06-29 16:58:30 +02005470
Stefan Hajnoczi1b7bdbc2010-04-10 07:02:42 +01005471 /* remove from list, if necessary */
Kevin Wolf63eaaae2016-03-18 10:46:57 +01005472 if (bs->node_name[0] != '\0') {
5473 QTAILQ_REMOVE(&graph_bdrv_states, bs, node_list);
5474 }
Max Reitz2c1d04e2016-01-29 16:36:11 +01005475 QTAILQ_REMOVE(&all_bdrv_states, bs, bs_list);
5476
Anton Kuchin30c321f2019-05-07 11:12:56 +03005477 bdrv_close(bs);
5478
Anthony Liguori7267c092011-08-20 22:09:37 -05005479 g_free(bs);
bellardfc01f7e2003-06-30 10:03:06 +00005480}
5481
Vladimir Sementsov-Ogievskiy96796fa2021-09-20 14:55:36 +03005482
5483/*
5484 * Replace @bs by newly created block node.
5485 *
5486 * @options is a QDict of options to pass to the block drivers, or NULL for an
5487 * empty set of options. The reference to the QDict belongs to the block layer
5488 * after the call (even on failure), so if the caller intends to reuse the
5489 * dictionary, it needs to use qobject_ref() before calling bdrv_open.
Kevin Wolf88234072023-05-25 14:47:11 +02005490 *
5491 * The caller holds the AioContext lock for @bs. It must make sure that @bs
5492 * stays in the same AioContext, i.e. @options must not refer to nodes in a
5493 * different AioContext.
Vladimir Sementsov-Ogievskiy96796fa2021-09-20 14:55:36 +03005494 */
5495BlockDriverState *bdrv_insert_node(BlockDriverState *bs, QDict *options,
Andrey Shinkevich8872ef72020-12-16 09:16:52 +03005496 int flags, Error **errp)
5497{
Vladimir Sementsov-Ogievskiyf053b7e2021-09-20 14:55:35 +03005498 ERRP_GUARD();
5499 int ret;
Kevin Wolf88234072023-05-25 14:47:11 +02005500 AioContext *ctx = bdrv_get_aio_context(bs);
Vladimir Sementsov-Ogievskiyb11c8732021-09-20 14:55:37 +03005501 BlockDriverState *new_node_bs = NULL;
5502 const char *drvname, *node_name;
5503 BlockDriver *drv;
Andrey Shinkevich8872ef72020-12-16 09:16:52 +03005504
Vladimir Sementsov-Ogievskiyb11c8732021-09-20 14:55:37 +03005505 drvname = qdict_get_try_str(options, "driver");
5506 if (!drvname) {
5507 error_setg(errp, "driver is not specified");
5508 goto fail;
5509 }
5510
5511 drv = bdrv_find_format(drvname);
5512 if (!drv) {
5513 error_setg(errp, "Unknown driver: '%s'", drvname);
5514 goto fail;
5515 }
5516
5517 node_name = qdict_get_try_str(options, "node-name");
5518
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05005519 GLOBAL_STATE_CODE();
5520
Kevin Wolf88234072023-05-25 14:47:11 +02005521 aio_context_release(ctx);
5522 aio_context_acquire(qemu_get_aio_context());
Vladimir Sementsov-Ogievskiyb11c8732021-09-20 14:55:37 +03005523 new_node_bs = bdrv_new_open_driver_opts(drv, node_name, options, flags,
5524 errp);
Kevin Wolf88234072023-05-25 14:47:11 +02005525 aio_context_release(qemu_get_aio_context());
5526 aio_context_acquire(ctx);
5527 assert(bdrv_get_aio_context(bs) == ctx);
5528
Vladimir Sementsov-Ogievskiyb11c8732021-09-20 14:55:37 +03005529 options = NULL; /* bdrv_new_open_driver() eats options */
5530 if (!new_node_bs) {
Andrey Shinkevich8872ef72020-12-16 09:16:52 +03005531 error_prepend(errp, "Could not create node: ");
Vladimir Sementsov-Ogievskiyb11c8732021-09-20 14:55:37 +03005532 goto fail;
Andrey Shinkevich8872ef72020-12-16 09:16:52 +03005533 }
5534
5535 bdrv_drained_begin(bs);
Vladimir Sementsov-Ogievskiyf053b7e2021-09-20 14:55:35 +03005536 ret = bdrv_replace_node(bs, new_node_bs, errp);
Andrey Shinkevich8872ef72020-12-16 09:16:52 +03005537 bdrv_drained_end(bs);
5538
Vladimir Sementsov-Ogievskiyf053b7e2021-09-20 14:55:35 +03005539 if (ret < 0) {
5540 error_prepend(errp, "Could not replace node: ");
Vladimir Sementsov-Ogievskiyb11c8732021-09-20 14:55:37 +03005541 goto fail;
Andrey Shinkevich8872ef72020-12-16 09:16:52 +03005542 }
5543
5544 return new_node_bs;
Vladimir Sementsov-Ogievskiyb11c8732021-09-20 14:55:37 +03005545
5546fail:
5547 qobject_unref(options);
5548 bdrv_unref(new_node_bs);
5549 return NULL;
Andrey Shinkevich8872ef72020-12-16 09:16:52 +03005550}
5551
aliguorie97fc192009-04-21 23:11:50 +00005552/*
5553 * Run consistency checks on an image
5554 *
Kevin Wolfe076f332010-06-29 11:43:13 +02005555 * Returns 0 if the check could be completed (it doesn't mean that the image is
Stefan Weila1c72732011-04-28 17:20:38 +02005556 * free of errors) or -errno when an internal error occurred. The results of the
Kevin Wolfe076f332010-06-29 11:43:13 +02005557 * check are stored in res.
aliguorie97fc192009-04-21 23:11:50 +00005558 */
Vladimir Sementsov-Ogievskiy21c22832020-09-24 21:54:10 +03005559int coroutine_fn bdrv_co_check(BlockDriverState *bs,
5560 BdrvCheckResult *res, BdrvCheckMode fix)
aliguorie97fc192009-04-21 23:11:50 +00005561{
Emanuele Giuseppe Esposito1581a702022-03-03 10:16:09 -05005562 IO_CODE();
Kevin Wolf1b3ff9f2022-12-07 14:18:38 +01005563 assert_bdrv_graph_readable();
Max Reitz908bcd52014-08-07 22:47:55 +02005564 if (bs->drv == NULL) {
5565 return -ENOMEDIUM;
5566 }
Paolo Bonzini2fd61632018-03-01 17:36:19 +01005567 if (bs->drv->bdrv_co_check == NULL) {
aliguorie97fc192009-04-21 23:11:50 +00005568 return -ENOTSUP;
5569 }
5570
Kevin Wolfe076f332010-06-29 11:43:13 +02005571 memset(res, 0, sizeof(*res));
Paolo Bonzini2fd61632018-03-01 17:36:19 +01005572 return bs->drv->bdrv_co_check(bs, res, fix);
5573}
5574
Kevin Wolf756e6732010-01-12 12:55:17 +01005575/*
5576 * Return values:
5577 * 0 - success
5578 * -EINVAL - backing format specified, but no file
5579 * -ENOSPC - can't update the backing file because no space is left in the
5580 * image file header
5581 * -ENOTSUP - format driver doesn't support changing the backing file
5582 */
Eric Blakee54ee1b2020-07-06 15:39:53 -05005583int bdrv_change_backing_file(BlockDriverState *bs, const char *backing_file,
Eric Blake497a30d2021-05-03 14:36:00 -07005584 const char *backing_fmt, bool require)
Kevin Wolf756e6732010-01-12 12:55:17 +01005585{
5586 BlockDriver *drv = bs->drv;
Paolo Bonzini469ef352012-04-12 14:01:02 +02005587 int ret;
Kevin Wolf756e6732010-01-12 12:55:17 +01005588
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05005589 GLOBAL_STATE_CODE();
5590
Max Reitzd470ad42017-11-10 21:31:09 +01005591 if (!drv) {
5592 return -ENOMEDIUM;
5593 }
5594
Paolo Bonzini5f377792012-04-12 14:01:01 +02005595 /* Backing file format doesn't make sense without a backing file */
5596 if (backing_fmt && !backing_file) {
5597 return -EINVAL;
5598 }
5599
Eric Blake497a30d2021-05-03 14:36:00 -07005600 if (require && backing_file && !backing_fmt) {
5601 return -EINVAL;
Eric Blakee54ee1b2020-07-06 15:39:53 -05005602 }
5603
Kevin Wolf756e6732010-01-12 12:55:17 +01005604 if (drv->bdrv_change_backing_file != NULL) {
Paolo Bonzini469ef352012-04-12 14:01:02 +02005605 ret = drv->bdrv_change_backing_file(bs, backing_file, backing_fmt);
Kevin Wolf756e6732010-01-12 12:55:17 +01005606 } else {
Paolo Bonzini469ef352012-04-12 14:01:02 +02005607 ret = -ENOTSUP;
Kevin Wolf756e6732010-01-12 12:55:17 +01005608 }
Paolo Bonzini469ef352012-04-12 14:01:02 +02005609
5610 if (ret == 0) {
5611 pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_file ?: "");
5612 pstrcpy(bs->backing_format, sizeof(bs->backing_format), backing_fmt ?: "");
Max Reitz998c2012019-02-01 20:29:08 +01005613 pstrcpy(bs->auto_backing_file, sizeof(bs->auto_backing_file),
5614 backing_file ?: "");
Paolo Bonzini469ef352012-04-12 14:01:02 +02005615 }
5616 return ret;
Kevin Wolf756e6732010-01-12 12:55:17 +01005617}
5618
Jeff Cody6ebdcee2012-09-27 13:29:12 -04005619/*
Max Reitzdcf3f9b2019-06-12 17:34:45 +02005620 * Finds the first non-filter node above bs in the chain between
5621 * active and bs. The returned node is either an immediate parent of
5622 * bs, or there are only filter nodes between the two.
Jeff Cody6ebdcee2012-09-27 13:29:12 -04005623 *
5624 * Returns NULL if bs is not found in active's image chain,
5625 * or if active == bs.
Jeff Cody4caf0fc2014-06-25 15:35:26 -04005626 *
5627 * Returns the bottommost base image if bs == NULL.
Jeff Cody6ebdcee2012-09-27 13:29:12 -04005628 */
5629BlockDriverState *bdrv_find_overlay(BlockDriverState *active,
5630 BlockDriverState *bs)
5631{
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05005632
5633 GLOBAL_STATE_CODE();
5634
Max Reitzdcf3f9b2019-06-12 17:34:45 +02005635 bs = bdrv_skip_filters(bs);
5636 active = bdrv_skip_filters(active);
5637
5638 while (active) {
5639 BlockDriverState *next = bdrv_backing_chain_next(active);
5640 if (bs == next) {
5641 return active;
5642 }
5643 active = next;
Jeff Cody6ebdcee2012-09-27 13:29:12 -04005644 }
5645
Max Reitzdcf3f9b2019-06-12 17:34:45 +02005646 return NULL;
Jeff Cody4caf0fc2014-06-25 15:35:26 -04005647}
Jeff Cody6ebdcee2012-09-27 13:29:12 -04005648
Jeff Cody4caf0fc2014-06-25 15:35:26 -04005649/* Given a BDS, searches for the base layer. */
5650BlockDriverState *bdrv_find_base(BlockDriverState *bs)
5651{
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05005652 GLOBAL_STATE_CODE();
5653
Jeff Cody4caf0fc2014-06-25 15:35:26 -04005654 return bdrv_find_overlay(bs, NULL);
Jeff Cody6ebdcee2012-09-27 13:29:12 -04005655}
5656
Jeff Cody6ebdcee2012-09-27 13:29:12 -04005657/*
Max Reitz7b99a262019-06-12 16:07:11 +02005658 * Return true if at least one of the COW (backing) and filter links
5659 * between @bs and @base is frozen. @errp is set if that's the case.
Alberto Garcia0f0998f2019-03-28 18:25:09 +02005660 * @base must be reachable from @bs, or NULL.
Alberto Garcia2cad1eb2019-03-12 18:48:40 +02005661 */
5662bool bdrv_is_backing_chain_frozen(BlockDriverState *bs, BlockDriverState *base,
5663 Error **errp)
5664{
5665 BlockDriverState *i;
Max Reitz7b99a262019-06-12 16:07:11 +02005666 BdrvChild *child;
Alberto Garcia2cad1eb2019-03-12 18:48:40 +02005667
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05005668 GLOBAL_STATE_CODE();
5669
Max Reitz7b99a262019-06-12 16:07:11 +02005670 for (i = bs; i != base; i = child_bs(child)) {
5671 child = bdrv_filter_or_cow_child(i);
5672
5673 if (child && child->frozen) {
Alberto Garcia2cad1eb2019-03-12 18:48:40 +02005674 error_setg(errp, "Cannot change '%s' link from '%s' to '%s'",
Max Reitz7b99a262019-06-12 16:07:11 +02005675 child->name, i->node_name, child->bs->node_name);
Alberto Garcia2cad1eb2019-03-12 18:48:40 +02005676 return true;
5677 }
5678 }
5679
5680 return false;
5681}
5682
5683/*
Max Reitz7b99a262019-06-12 16:07:11 +02005684 * Freeze all COW (backing) and filter links between @bs and @base.
Alberto Garcia2cad1eb2019-03-12 18:48:40 +02005685 * If any of the links is already frozen the operation is aborted and
5686 * none of the links are modified.
Alberto Garcia0f0998f2019-03-28 18:25:09 +02005687 * @base must be reachable from @bs, or NULL.
Alberto Garcia2cad1eb2019-03-12 18:48:40 +02005688 * Returns 0 on success. On failure returns < 0 and sets @errp.
5689 */
5690int bdrv_freeze_backing_chain(BlockDriverState *bs, BlockDriverState *base,
5691 Error **errp)
5692{
5693 BlockDriverState *i;
Max Reitz7b99a262019-06-12 16:07:11 +02005694 BdrvChild *child;
Alberto Garcia2cad1eb2019-03-12 18:48:40 +02005695
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05005696 GLOBAL_STATE_CODE();
5697
Alberto Garcia2cad1eb2019-03-12 18:48:40 +02005698 if (bdrv_is_backing_chain_frozen(bs, base, errp)) {
5699 return -EPERM;
5700 }
5701
Max Reitz7b99a262019-06-12 16:07:11 +02005702 for (i = bs; i != base; i = child_bs(child)) {
5703 child = bdrv_filter_or_cow_child(i);
5704 if (child && child->bs->never_freeze) {
Max Reitze5182c12019-07-03 19:28:02 +02005705 error_setg(errp, "Cannot freeze '%s' link to '%s'",
Max Reitz7b99a262019-06-12 16:07:11 +02005706 child->name, child->bs->node_name);
Max Reitze5182c12019-07-03 19:28:02 +02005707 return -EPERM;
5708 }
5709 }
5710
Max Reitz7b99a262019-06-12 16:07:11 +02005711 for (i = bs; i != base; i = child_bs(child)) {
5712 child = bdrv_filter_or_cow_child(i);
5713 if (child) {
5714 child->frozen = true;
Alberto Garcia0f0998f2019-03-28 18:25:09 +02005715 }
Alberto Garcia2cad1eb2019-03-12 18:48:40 +02005716 }
5717
5718 return 0;
5719}
5720
5721/*
Max Reitz7b99a262019-06-12 16:07:11 +02005722 * Unfreeze all COW (backing) and filter links between @bs and @base.
5723 * The caller must ensure that all links are frozen before using this
5724 * function.
Alberto Garcia0f0998f2019-03-28 18:25:09 +02005725 * @base must be reachable from @bs, or NULL.
Alberto Garcia2cad1eb2019-03-12 18:48:40 +02005726 */
5727void bdrv_unfreeze_backing_chain(BlockDriverState *bs, BlockDriverState *base)
5728{
5729 BlockDriverState *i;
Max Reitz7b99a262019-06-12 16:07:11 +02005730 BdrvChild *child;
Alberto Garcia2cad1eb2019-03-12 18:48:40 +02005731
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05005732 GLOBAL_STATE_CODE();
5733
Max Reitz7b99a262019-06-12 16:07:11 +02005734 for (i = bs; i != base; i = child_bs(child)) {
5735 child = bdrv_filter_or_cow_child(i);
5736 if (child) {
5737 assert(child->frozen);
5738 child->frozen = false;
Alberto Garcia0f0998f2019-03-28 18:25:09 +02005739 }
Alberto Garcia2cad1eb2019-03-12 18:48:40 +02005740 }
5741}
5742
5743/*
Jeff Cody6ebdcee2012-09-27 13:29:12 -04005744 * Drops images above 'base' up to and including 'top', and sets the image
5745 * above 'top' to have base as its backing file.
5746 *
5747 * Requires that the overlay to 'top' is opened r/w, so that the backing file
5748 * information in 'bs' can be properly updated.
5749 *
5750 * E.g., this will convert the following chain:
5751 * bottom <- base <- intermediate <- top <- active
5752 *
5753 * to
5754 *
5755 * bottom <- base <- active
5756 *
5757 * It is allowed for bottom==base, in which case it converts:
5758 *
5759 * base <- intermediate <- top <- active
5760 *
5761 * to
5762 *
5763 * base <- active
5764 *
Jeff Cody54e26902014-06-25 15:40:10 -04005765 * If backing_file_str is non-NULL, it will be used when modifying top's
5766 * overlay image metadata.
5767 *
Jeff Cody6ebdcee2012-09-27 13:29:12 -04005768 * Error conditions:
5769 * if active == top, that is considered an error
5770 *
5771 */
Kevin Wolfbde70712017-06-27 20:36:18 +02005772int bdrv_drop_intermediate(BlockDriverState *top, BlockDriverState *base,
5773 const char *backing_file_str)
Jeff Cody6ebdcee2012-09-27 13:29:12 -04005774{
Alberto Garcia6bd858b2018-10-31 18:16:38 +02005775 BlockDriverState *explicit_top = top;
5776 bool update_inherits_from;
Vladimir Sementsov-Ogievskiyd669ed62020-11-06 15:42:37 +03005777 BdrvChild *c;
Kevin Wolf12fa4af2017-02-17 20:42:32 +01005778 Error *local_err = NULL;
Jeff Cody6ebdcee2012-09-27 13:29:12 -04005779 int ret = -EIO;
Vladimir Sementsov-Ogievskiyd669ed62020-11-06 15:42:37 +03005780 g_autoptr(GSList) updated_children = NULL;
5781 GSList *p;
Jeff Cody6ebdcee2012-09-27 13:29:12 -04005782
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05005783 GLOBAL_STATE_CODE();
5784
Kevin Wolf6858eba2017-06-29 19:32:21 +02005785 bdrv_ref(top);
Kevin Wolf631086d2022-11-18 18:41:03 +01005786 bdrv_drained_begin(base);
Kevin Wolf6858eba2017-06-29 19:32:21 +02005787
Jeff Cody6ebdcee2012-09-27 13:29:12 -04005788 if (!top->drv || !base->drv) {
5789 goto exit;
5790 }
5791
Kevin Wolf5db15a52015-09-14 15:33:33 +02005792 /* Make sure that base is in the backing chain of top */
5793 if (!bdrv_chain_contains(top, base)) {
Jeff Cody6ebdcee2012-09-27 13:29:12 -04005794 goto exit;
5795 }
5796
Alberto Garcia6bd858b2018-10-31 18:16:38 +02005797 /* If 'base' recursively inherits from 'top' then we should set
5798 * base->inherits_from to top->inherits_from after 'top' and all
5799 * other intermediate nodes have been dropped.
5800 * If 'top' is an implicit node (e.g. "commit_top") we should skip
5801 * it because no one inherits from it. We use explicit_top for that. */
Max Reitzdcf3f9b2019-06-12 17:34:45 +02005802 explicit_top = bdrv_skip_implicit_filters(explicit_top);
Alberto Garcia6bd858b2018-10-31 18:16:38 +02005803 update_inherits_from = bdrv_inherits_from_recursive(base, explicit_top);
5804
Jeff Cody6ebdcee2012-09-27 13:29:12 -04005805 /* success - we can delete the intermediate states, and link top->base */
Max Reitzf30c66b2019-02-01 20:29:05 +01005806 if (!backing_file_str) {
5807 bdrv_refresh_filename(base);
5808 backing_file_str = base->filename;
5809 }
Kevin Wolf61f09ce2017-09-19 16:22:54 +02005810
Vladimir Sementsov-Ogievskiyd669ed62020-11-06 15:42:37 +03005811 QLIST_FOREACH(c, &top->parents, next_parent) {
5812 updated_children = g_slist_prepend(updated_children, c);
5813 }
Kevin Wolf12fa4af2017-02-17 20:42:32 +01005814
Vladimir Sementsov-Ogievskiy3108a152021-04-28 18:17:51 +03005815 /*
5816 * It seems correct to pass detach_subchain=true here, but it triggers
5817 * one more yet not fixed bug, when due to nested aio_poll loop we switch to
5818 * another drained section, which modify the graph (for example, removing
5819 * the child, which we keep in updated_children list). So, it's a TODO.
5820 *
5821 * Note, bug triggered if pass detach_subchain=true here and run
5822 * test-bdrv-drain. test_drop_intermediate_poll() test-case will crash.
5823 * That's a FIXME.
5824 */
5825 bdrv_replace_node_common(top, base, false, false, &local_err);
Vladimir Sementsov-Ogievskiyd669ed62020-11-06 15:42:37 +03005826 if (local_err) {
5827 error_report_err(local_err);
5828 goto exit;
5829 }
5830
5831 for (p = updated_children; p; p = p->next) {
5832 c = p->data;
5833
Max Reitzbd86fb92020-05-13 13:05:13 +02005834 if (c->klass->update_filename) {
5835 ret = c->klass->update_filename(c, base, backing_file_str,
5836 &local_err);
Kevin Wolf61f09ce2017-09-19 16:22:54 +02005837 if (ret < 0) {
Vladimir Sementsov-Ogievskiyd669ed62020-11-06 15:42:37 +03005838 /*
5839 * TODO: Actually, we want to rollback all previous iterations
5840 * of this loop, and (which is almost impossible) previous
5841 * bdrv_replace_node()...
5842 *
5843 * Note, that c->klass->update_filename may lead to permission
5844 * update, so it's a bad idea to call it inside permission
5845 * update transaction of bdrv_replace_node.
5846 */
Kevin Wolf61f09ce2017-09-19 16:22:54 +02005847 error_report_err(local_err);
5848 goto exit;
5849 }
5850 }
Kevin Wolf12fa4af2017-02-17 20:42:32 +01005851 }
Jeff Cody6ebdcee2012-09-27 13:29:12 -04005852
Alberto Garcia6bd858b2018-10-31 18:16:38 +02005853 if (update_inherits_from) {
5854 base->inherits_from = explicit_top->inherits_from;
5855 }
5856
Jeff Cody6ebdcee2012-09-27 13:29:12 -04005857 ret = 0;
Jeff Cody6ebdcee2012-09-27 13:29:12 -04005858exit:
Kevin Wolf631086d2022-11-18 18:41:03 +01005859 bdrv_drained_end(base);
Kevin Wolf6858eba2017-06-29 19:32:21 +02005860 bdrv_unref(top);
Jeff Cody6ebdcee2012-09-27 13:29:12 -04005861 return ret;
5862}
5863
bellard83f64092006-08-01 16:21:11 +00005864/**
Emanuele Giuseppe Esposito82618d72023-01-13 21:42:07 +01005865 * Implementation of BlockDriver.bdrv_co_get_allocated_file_size() that
Max Reitz081e4652019-06-12 18:14:13 +02005866 * sums the size of all data-bearing children. (This excludes backing
5867 * children.)
5868 */
Emanuele Giuseppe Espositode335632023-05-04 13:57:43 +02005869static int64_t coroutine_fn GRAPH_RDLOCK
5870bdrv_sum_allocated_file_size(BlockDriverState *bs)
Max Reitz081e4652019-06-12 18:14:13 +02005871{
5872 BdrvChild *child;
5873 int64_t child_size, sum = 0;
5874
5875 QLIST_FOREACH(child, &bs->children, next) {
5876 if (child->role & (BDRV_CHILD_DATA | BDRV_CHILD_METADATA |
5877 BDRV_CHILD_FILTERED))
5878 {
Emanuele Giuseppe Esposito82618d72023-01-13 21:42:07 +01005879 child_size = bdrv_co_get_allocated_file_size(child->bs);
Max Reitz081e4652019-06-12 18:14:13 +02005880 if (child_size < 0) {
5881 return child_size;
5882 }
5883 sum += child_size;
5884 }
5885 }
5886
5887 return sum;
5888}
5889
5890/**
Fam Zheng4a1d5e12011-07-12 19:56:39 +08005891 * Length of a allocated file in bytes. Sparse files are counted by actual
5892 * allocated space. Return < 0 if error or unknown.
5893 */
Emanuele Giuseppe Esposito82618d72023-01-13 21:42:07 +01005894int64_t coroutine_fn bdrv_co_get_allocated_file_size(BlockDriverState *bs)
Fam Zheng4a1d5e12011-07-12 19:56:39 +08005895{
5896 BlockDriver *drv = bs->drv;
Emanuele Giuseppe Esposito384a48f2022-03-03 10:15:50 -05005897 IO_CODE();
Emanuele Giuseppe Espositode335632023-05-04 13:57:43 +02005898 assert_bdrv_graph_readable();
Emanuele Giuseppe Esposito384a48f2022-03-03 10:15:50 -05005899
Fam Zheng4a1d5e12011-07-12 19:56:39 +08005900 if (!drv) {
5901 return -ENOMEDIUM;
5902 }
Emanuele Giuseppe Esposito82618d72023-01-13 21:42:07 +01005903 if (drv->bdrv_co_get_allocated_file_size) {
5904 return drv->bdrv_co_get_allocated_file_size(bs);
Fam Zheng4a1d5e12011-07-12 19:56:39 +08005905 }
Max Reitz081e4652019-06-12 18:14:13 +02005906
5907 if (drv->bdrv_file_open) {
5908 /*
5909 * Protocol drivers default to -ENOTSUP (most of their data is
5910 * not stored in any of their children (if they even have any),
5911 * so there is no generic way to figure it out).
5912 */
5913 return -ENOTSUP;
5914 } else if (drv->is_filter) {
5915 /* Filter drivers default to the size of their filtered child */
Emanuele Giuseppe Esposito82618d72023-01-13 21:42:07 +01005916 return bdrv_co_get_allocated_file_size(bdrv_filter_bs(bs));
Max Reitz081e4652019-06-12 18:14:13 +02005917 } else {
5918 /* Other drivers default to summing their children's sizes */
5919 return bdrv_sum_allocated_file_size(bs);
Fam Zheng4a1d5e12011-07-12 19:56:39 +08005920 }
Fam Zheng4a1d5e12011-07-12 19:56:39 +08005921}
5922
Stefan Hajnoczi90880ff2017-07-05 13:57:30 +01005923/*
5924 * bdrv_measure:
5925 * @drv: Format driver
5926 * @opts: Creation options for new image
5927 * @in_bs: Existing image containing data for new image (may be NULL)
5928 * @errp: Error object
5929 * Returns: A #BlockMeasureInfo (free using qapi_free_BlockMeasureInfo())
5930 * or NULL on error
5931 *
5932 * Calculate file size required to create a new image.
5933 *
5934 * If @in_bs is given then space for allocated clusters and zero clusters
5935 * from that image are included in the calculation. If @opts contains a
5936 * backing file that is shared by @in_bs then backing clusters may be omitted
5937 * from the calculation.
5938 *
5939 * If @in_bs is NULL then the calculation includes no allocated clusters
5940 * unless a preallocation option is given in @opts.
5941 *
5942 * Note that @in_bs may use a different BlockDriver from @drv.
5943 *
5944 * If an error occurs the @errp pointer is set.
5945 */
5946BlockMeasureInfo *bdrv_measure(BlockDriver *drv, QemuOpts *opts,
5947 BlockDriverState *in_bs, Error **errp)
5948{
Emanuele Giuseppe Esposito384a48f2022-03-03 10:15:50 -05005949 IO_CODE();
Stefan Hajnoczi90880ff2017-07-05 13:57:30 +01005950 if (!drv->bdrv_measure) {
5951 error_setg(errp, "Block driver '%s' does not support size measurement",
5952 drv->format_name);
5953 return NULL;
5954 }
5955
5956 return drv->bdrv_measure(opts, in_bs, errp);
5957}
5958
Fam Zheng4a1d5e12011-07-12 19:56:39 +08005959/**
Markus Armbruster65a9bb22014-06-26 13:23:17 +02005960 * Return number of sectors on success, -errno on error.
bellard83f64092006-08-01 16:21:11 +00005961 */
Emanuele Giuseppe Espositoc86422c2023-01-13 21:42:04 +01005962int64_t coroutine_fn bdrv_co_nb_sectors(BlockDriverState *bs)
bellard83f64092006-08-01 16:21:11 +00005963{
5964 BlockDriver *drv = bs->drv;
Emanuele Giuseppe Esposito384a48f2022-03-03 10:15:50 -05005965 IO_CODE();
Kevin Wolf8ab81402023-02-03 16:22:02 +01005966 assert_bdrv_graph_readable();
Markus Armbruster65a9bb22014-06-26 13:23:17 +02005967
bellard83f64092006-08-01 16:21:11 +00005968 if (!drv)
bellard19cb3732006-08-19 11:45:59 +00005969 return -ENOMEDIUM;
Stefan Hajnoczi51762282010-04-19 16:56:41 +01005970
Paolo Bonzini160a29e2023-04-07 17:32:56 +02005971 if (bs->bl.has_variable_length) {
Emanuele Giuseppe Espositoc86422c2023-01-13 21:42:04 +01005972 int ret = bdrv_co_refresh_total_sectors(bs, bs->total_sectors);
Kevin Wolfb94a2612013-10-29 12:18:58 +01005973 if (ret < 0) {
5974 return ret;
Stefan Hajnoczi46a4e4e2011-03-29 20:04:41 +01005975 }
bellard83f64092006-08-01 16:21:11 +00005976 }
Markus Armbruster65a9bb22014-06-26 13:23:17 +02005977 return bs->total_sectors;
5978}
5979
Paolo Bonzini81f730d2023-04-07 17:33:03 +02005980/*
5981 * This wrapper is written by hand because this function is in the hot I/O path,
5982 * via blk_get_geometry.
5983 */
5984int64_t coroutine_mixed_fn bdrv_nb_sectors(BlockDriverState *bs)
5985{
5986 BlockDriver *drv = bs->drv;
5987 IO_CODE();
5988
5989 if (!drv)
5990 return -ENOMEDIUM;
5991
5992 if (bs->bl.has_variable_length) {
5993 int ret = bdrv_refresh_total_sectors(bs, bs->total_sectors);
5994 if (ret < 0) {
5995 return ret;
5996 }
5997 }
5998
5999 return bs->total_sectors;
6000}
6001
Markus Armbruster65a9bb22014-06-26 13:23:17 +02006002/**
6003 * Return length in bytes on success, -errno on error.
6004 * The length is always a multiple of BDRV_SECTOR_SIZE.
6005 */
Emanuele Giuseppe Espositoc86422c2023-01-13 21:42:04 +01006006int64_t coroutine_fn bdrv_co_getlength(BlockDriverState *bs)
Markus Armbruster65a9bb22014-06-26 13:23:17 +02006007{
Emanuele Giuseppe Espositoc86422c2023-01-13 21:42:04 +01006008 int64_t ret;
Emanuele Giuseppe Esposito384a48f2022-03-03 10:15:50 -05006009 IO_CODE();
Kevin Wolf8ab81402023-02-03 16:22:02 +01006010 assert_bdrv_graph_readable();
Markus Armbruster65a9bb22014-06-26 13:23:17 +02006011
Emanuele Giuseppe Espositoc86422c2023-01-13 21:42:04 +01006012 ret = bdrv_co_nb_sectors(bs);
Eric Blake122860b2020-11-05 09:51:22 -06006013 if (ret < 0) {
6014 return ret;
6015 }
6016 if (ret > INT64_MAX / BDRV_SECTOR_SIZE) {
6017 return -EFBIG;
6018 }
6019 return ret * BDRV_SECTOR_SIZE;
bellardfc01f7e2003-06-30 10:03:06 +00006020}
6021
Eric Blake54115412016-06-23 16:37:26 -06006022bool bdrv_is_sg(BlockDriverState *bs)
ths985a03b2007-12-24 16:10:43 +00006023{
Emanuele Giuseppe Esposito384a48f2022-03-03 10:15:50 -05006024 IO_CODE();
ths985a03b2007-12-24 16:10:43 +00006025 return bs->sg;
6026}
6027
Max Reitzae23f782019-06-12 22:57:15 +02006028/**
6029 * Return whether the given node supports compressed writes.
6030 */
6031bool bdrv_supports_compressed_writes(BlockDriverState *bs)
6032{
6033 BlockDriverState *filtered;
Emanuele Giuseppe Esposito384a48f2022-03-03 10:15:50 -05006034 IO_CODE();
Max Reitzae23f782019-06-12 22:57:15 +02006035
6036 if (!bs->drv || !block_driver_can_compress(bs->drv)) {
6037 return false;
6038 }
6039
6040 filtered = bdrv_filter_bs(bs);
6041 if (filtered) {
6042 /*
6043 * Filters can only forward compressed writes, so we have to
6044 * check the child.
6045 */
6046 return bdrv_supports_compressed_writes(filtered);
6047 }
6048
6049 return true;
6050}
6051
Markus Armbrusterf8d6bba2012-06-13 10:11:48 +02006052const char *bdrv_get_format_name(BlockDriverState *bs)
bellardea2384d2004-08-01 21:59:26 +00006053{
Emanuele Giuseppe Esposito384a48f2022-03-03 10:15:50 -05006054 IO_CODE();
Markus Armbrusterf8d6bba2012-06-13 10:11:48 +02006055 return bs->drv ? bs->drv->format_name : NULL;
bellardea2384d2004-08-01 21:59:26 +00006056}
6057
Stefan Hajnocziada42402014-08-27 12:08:55 +01006058static int qsort_strcmp(const void *a, const void *b)
6059{
Max Reitzceff5bd2016-10-12 22:49:05 +02006060 return strcmp(*(char *const *)a, *(char *const *)b);
Stefan Hajnocziada42402014-08-27 12:08:55 +01006061}
6062
ths5fafdf22007-09-16 21:08:06 +00006063void bdrv_iterate_format(void (*it)(void *opaque, const char *name),
Andrey Shinkevich9ac404c2019-03-07 16:33:58 +03006064 void *opaque, bool read_only)
bellardea2384d2004-08-01 21:59:26 +00006065{
6066 BlockDriver *drv;
Jeff Codye855e4f2014-04-28 18:29:54 -04006067 int count = 0;
Stefan Hajnocziada42402014-08-27 12:08:55 +01006068 int i;
Jeff Codye855e4f2014-04-28 18:29:54 -04006069 const char **formats = NULL;
bellardea2384d2004-08-01 21:59:26 +00006070
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05006071 GLOBAL_STATE_CODE();
6072
Stefan Hajnoczi8a22f022010-04-13 10:29:33 +01006073 QLIST_FOREACH(drv, &bdrv_drivers, list) {
Jeff Codye855e4f2014-04-28 18:29:54 -04006074 if (drv->format_name) {
6075 bool found = false;
6076 int i = count;
Andrey Shinkevich9ac404c2019-03-07 16:33:58 +03006077
6078 if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv, read_only)) {
6079 continue;
6080 }
6081
Jeff Codye855e4f2014-04-28 18:29:54 -04006082 while (formats && i && !found) {
6083 found = !strcmp(formats[--i], drv->format_name);
6084 }
6085
6086 if (!found) {
Markus Armbruster5839e532014-08-19 10:31:08 +02006087 formats = g_renew(const char *, formats, count + 1);
Jeff Codye855e4f2014-04-28 18:29:54 -04006088 formats[count++] = drv->format_name;
Jeff Codye855e4f2014-04-28 18:29:54 -04006089 }
6090 }
bellardea2384d2004-08-01 21:59:26 +00006091 }
Stefan Hajnocziada42402014-08-27 12:08:55 +01006092
Max Reitzeb0df692016-10-12 22:49:06 +02006093 for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); i++) {
6094 const char *format_name = block_driver_modules[i].format_name;
6095
6096 if (format_name) {
6097 bool found = false;
6098 int j = count;
6099
Andrey Shinkevich9ac404c2019-03-07 16:33:58 +03006100 if (use_bdrv_whitelist &&
6101 !bdrv_format_is_whitelisted(format_name, read_only)) {
6102 continue;
6103 }
6104
Max Reitzeb0df692016-10-12 22:49:06 +02006105 while (formats && j && !found) {
6106 found = !strcmp(formats[--j], format_name);
6107 }
6108
6109 if (!found) {
6110 formats = g_renew(const char *, formats, count + 1);
6111 formats[count++] = format_name;
6112 }
6113 }
6114 }
6115
Stefan Hajnocziada42402014-08-27 12:08:55 +01006116 qsort(formats, count, sizeof(formats[0]), qsort_strcmp);
6117
6118 for (i = 0; i < count; i++) {
6119 it(opaque, formats[i]);
6120 }
6121
Jeff Codye855e4f2014-04-28 18:29:54 -04006122 g_free(formats);
bellardea2384d2004-08-01 21:59:26 +00006123}
6124
Benoît Canetdc364f42014-01-23 21:31:32 +01006125/* This function is to find a node in the bs graph */
6126BlockDriverState *bdrv_find_node(const char *node_name)
6127{
6128 BlockDriverState *bs;
6129
6130 assert(node_name);
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05006131 GLOBAL_STATE_CODE();
Benoît Canetdc364f42014-01-23 21:31:32 +01006132
6133 QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
6134 if (!strcmp(node_name, bs->node_name)) {
6135 return bs;
6136 }
6137 }
6138 return NULL;
6139}
6140
Benoît Canetc13163f2014-01-23 21:31:34 +01006141/* Put this QMP function here so it can access the static graph_bdrv_states. */
Peter Krempafacda542020-01-20 09:50:49 +01006142BlockDeviceInfoList *bdrv_named_nodes_list(bool flat,
6143 Error **errp)
Benoît Canetc13163f2014-01-23 21:31:34 +01006144{
Eric Blake9812e712020-10-27 00:05:47 -05006145 BlockDeviceInfoList *list;
Benoît Canetc13163f2014-01-23 21:31:34 +01006146 BlockDriverState *bs;
6147
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05006148 GLOBAL_STATE_CODE();
6149
Benoît Canetc13163f2014-01-23 21:31:34 +01006150 list = NULL;
6151 QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
Peter Krempafacda542020-01-20 09:50:49 +01006152 BlockDeviceInfo *info = bdrv_block_device_info(NULL, bs, flat, errp);
Alberto Garciad5a8ee62015-04-17 14:52:43 +03006153 if (!info) {
6154 qapi_free_BlockDeviceInfoList(list);
6155 return NULL;
6156 }
Eric Blake9812e712020-10-27 00:05:47 -05006157 QAPI_LIST_PREPEND(list, info);
Benoît Canetc13163f2014-01-23 21:31:34 +01006158 }
6159
6160 return list;
6161}
6162
Vladimir Sementsov-Ogievskiy5d3b4e92018-12-21 20:09:07 +03006163typedef struct XDbgBlockGraphConstructor {
6164 XDbgBlockGraph *graph;
6165 GHashTable *graph_nodes;
6166} XDbgBlockGraphConstructor;
6167
6168static XDbgBlockGraphConstructor *xdbg_graph_new(void)
6169{
6170 XDbgBlockGraphConstructor *gr = g_new(XDbgBlockGraphConstructor, 1);
6171
6172 gr->graph = g_new0(XDbgBlockGraph, 1);
6173 gr->graph_nodes = g_hash_table_new(NULL, NULL);
6174
6175 return gr;
6176}
6177
6178static XDbgBlockGraph *xdbg_graph_finalize(XDbgBlockGraphConstructor *gr)
6179{
6180 XDbgBlockGraph *graph = gr->graph;
6181
6182 g_hash_table_destroy(gr->graph_nodes);
6183 g_free(gr);
6184
6185 return graph;
6186}
6187
6188static uintptr_t xdbg_graph_node_num(XDbgBlockGraphConstructor *gr, void *node)
6189{
6190 uintptr_t ret = (uintptr_t)g_hash_table_lookup(gr->graph_nodes, node);
6191
6192 if (ret != 0) {
6193 return ret;
6194 }
6195
6196 /*
6197 * Start counting from 1, not 0, because 0 interferes with not-found (NULL)
6198 * answer of g_hash_table_lookup.
6199 */
6200 ret = g_hash_table_size(gr->graph_nodes) + 1;
6201 g_hash_table_insert(gr->graph_nodes, node, (void *)ret);
6202
6203 return ret;
6204}
6205
6206static void xdbg_graph_add_node(XDbgBlockGraphConstructor *gr, void *node,
6207 XDbgBlockGraphNodeType type, const char *name)
6208{
6209 XDbgBlockGraphNode *n;
6210
6211 n = g_new0(XDbgBlockGraphNode, 1);
6212
6213 n->id = xdbg_graph_node_num(gr, node);
6214 n->type = type;
6215 n->name = g_strdup(name);
6216
Eric Blake9812e712020-10-27 00:05:47 -05006217 QAPI_LIST_PREPEND(gr->graph->nodes, n);
Vladimir Sementsov-Ogievskiy5d3b4e92018-12-21 20:09:07 +03006218}
6219
6220static void xdbg_graph_add_edge(XDbgBlockGraphConstructor *gr, void *parent,
6221 const BdrvChild *child)
6222{
Max Reitzcdb1cec2019-11-08 13:34:52 +01006223 BlockPermission qapi_perm;
Vladimir Sementsov-Ogievskiy5d3b4e92018-12-21 20:09:07 +03006224 XDbgBlockGraphEdge *edge;
Emanuele Giuseppe Esposito862fded2022-03-03 10:15:55 -05006225 GLOBAL_STATE_CODE();
Vladimir Sementsov-Ogievskiy5d3b4e92018-12-21 20:09:07 +03006226
Vladimir Sementsov-Ogievskiy5d3b4e92018-12-21 20:09:07 +03006227 edge = g_new0(XDbgBlockGraphEdge, 1);
6228
6229 edge->parent = xdbg_graph_node_num(gr, parent);
6230 edge->child = xdbg_graph_node_num(gr, child->bs);
6231 edge->name = g_strdup(child->name);
6232
Max Reitzcdb1cec2019-11-08 13:34:52 +01006233 for (qapi_perm = 0; qapi_perm < BLOCK_PERMISSION__MAX; qapi_perm++) {
6234 uint64_t flag = bdrv_qapi_perm_to_blk_perm(qapi_perm);
6235
6236 if (flag & child->perm) {
Eric Blake9812e712020-10-27 00:05:47 -05006237 QAPI_LIST_PREPEND(edge->perm, qapi_perm);
Vladimir Sementsov-Ogievskiy5d3b4e92018-12-21 20:09:07 +03006238 }
Max Reitzcdb1cec2019-11-08 13:34:52 +01006239 if (flag & child->shared_perm) {
Eric Blake9812e712020-10-27 00:05:47 -05006240 QAPI_LIST_PREPEND(edge->shared_perm, qapi_perm);
Vladimir Sementsov-Ogievskiy5d3b4e92018-12-21 20:09:07 +03006241 }
6242 }
6243
Eric Blake9812e712020-10-27 00:05:47 -05006244 QAPI_LIST_PREPEND(gr->graph->edges, edge);
Vladimir Sementsov-Ogievskiy5d3b4e92018-12-21 20:09:07 +03006245}
6246
6247
6248XDbgBlockGraph *bdrv_get_xdbg_block_graph(Error **errp)
6249{
6250 BlockBackend *blk;
6251 BlockJob *job;
6252 BlockDriverState *bs;
6253 BdrvChild *child;
6254 XDbgBlockGraphConstructor *gr = xdbg_graph_new();
6255
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05006256 GLOBAL_STATE_CODE();
6257
Vladimir Sementsov-Ogievskiy5d3b4e92018-12-21 20:09:07 +03006258 for (blk = blk_all_next(NULL); blk; blk = blk_all_next(blk)) {
6259 char *allocated_name = NULL;
6260 const char *name = blk_name(blk);
6261
6262 if (!*name) {
6263 name = allocated_name = blk_get_attached_dev_id(blk);
6264 }
6265 xdbg_graph_add_node(gr, blk, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_BACKEND,
6266 name);
6267 g_free(allocated_name);
6268 if (blk_root(blk)) {
6269 xdbg_graph_add_edge(gr, blk, blk_root(blk));
6270 }
6271 }
6272
Emanuele Giuseppe Esposito880eeec2022-09-26 05:32:04 -04006273 WITH_JOB_LOCK_GUARD() {
6274 for (job = block_job_next_locked(NULL); job;
6275 job = block_job_next_locked(job)) {
6276 GSList *el;
Vladimir Sementsov-Ogievskiy5d3b4e92018-12-21 20:09:07 +03006277
Emanuele Giuseppe Esposito880eeec2022-09-26 05:32:04 -04006278 xdbg_graph_add_node(gr, job, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_JOB,
6279 job->job.id);
6280 for (el = job->nodes; el; el = el->next) {
6281 xdbg_graph_add_edge(gr, job, (BdrvChild *)el->data);
6282 }
Vladimir Sementsov-Ogievskiy5d3b4e92018-12-21 20:09:07 +03006283 }
6284 }
6285
6286 QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
6287 xdbg_graph_add_node(gr, bs, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_DRIVER,
6288 bs->node_name);
6289 QLIST_FOREACH(child, &bs->children, next) {
6290 xdbg_graph_add_edge(gr, bs, child);
6291 }
6292 }
6293
6294 return xdbg_graph_finalize(gr);
6295}
6296
Benoît Canet12d3ba82014-01-23 21:31:35 +01006297BlockDriverState *bdrv_lookup_bs(const char *device,
6298 const char *node_name,
6299 Error **errp)
6300{
Markus Armbruster7f06d472014-10-07 13:59:12 +02006301 BlockBackend *blk;
6302 BlockDriverState *bs;
Benoît Canet12d3ba82014-01-23 21:31:35 +01006303
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05006304 GLOBAL_STATE_CODE();
6305
Benoît Canet12d3ba82014-01-23 21:31:35 +01006306 if (device) {
Markus Armbruster7f06d472014-10-07 13:59:12 +02006307 blk = blk_by_name(device);
Benoît Canet12d3ba82014-01-23 21:31:35 +01006308
Markus Armbruster7f06d472014-10-07 13:59:12 +02006309 if (blk) {
Alberto Garcia9f4ed6f2015-10-26 16:46:49 +02006310 bs = blk_bs(blk);
6311 if (!bs) {
Max Reitz5433c242015-10-19 17:53:29 +02006312 error_setg(errp, "Device '%s' has no medium", device);
Max Reitz5433c242015-10-19 17:53:29 +02006313 }
6314
Alberto Garcia9f4ed6f2015-10-26 16:46:49 +02006315 return bs;
Benoît Canet12d3ba82014-01-23 21:31:35 +01006316 }
Benoît Canet12d3ba82014-01-23 21:31:35 +01006317 }
6318
Benoît Canetdd67fa52014-02-12 17:15:06 +01006319 if (node_name) {
6320 bs = bdrv_find_node(node_name);
Benoît Canet12d3ba82014-01-23 21:31:35 +01006321
Benoît Canetdd67fa52014-02-12 17:15:06 +01006322 if (bs) {
6323 return bs;
6324 }
Benoît Canet12d3ba82014-01-23 21:31:35 +01006325 }
6326
Connor Kuehl785ec4b2021-03-05 09:19:28 -06006327 error_setg(errp, "Cannot find device=\'%s\' nor node-name=\'%s\'",
Benoît Canetdd67fa52014-02-12 17:15:06 +01006328 device ? device : "",
6329 node_name ? node_name : "");
6330 return NULL;
Benoît Canet12d3ba82014-01-23 21:31:35 +01006331}
6332
Jeff Cody5a6684d2014-06-25 15:40:09 -04006333/* If 'base' is in the same chain as 'top', return true. Otherwise,
6334 * return false. If either argument is NULL, return false. */
6335bool bdrv_chain_contains(BlockDriverState *top, BlockDriverState *base)
6336{
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05006337
6338 GLOBAL_STATE_CODE();
6339
Jeff Cody5a6684d2014-06-25 15:40:09 -04006340 while (top && top != base) {
Max Reitzdcf3f9b2019-06-12 17:34:45 +02006341 top = bdrv_filter_or_cow_bs(top);
Jeff Cody5a6684d2014-06-25 15:40:09 -04006342 }
6343
6344 return top != NULL;
6345}
6346
Fam Zheng04df7652014-10-31 11:32:54 +08006347BlockDriverState *bdrv_next_node(BlockDriverState *bs)
6348{
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05006349 GLOBAL_STATE_CODE();
Fam Zheng04df7652014-10-31 11:32:54 +08006350 if (!bs) {
6351 return QTAILQ_FIRST(&graph_bdrv_states);
6352 }
6353 return QTAILQ_NEXT(bs, node_list);
6354}
6355
Kevin Wolf0f122642018-03-28 18:29:18 +02006356BlockDriverState *bdrv_next_all_states(BlockDriverState *bs)
6357{
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05006358 GLOBAL_STATE_CODE();
Kevin Wolf0f122642018-03-28 18:29:18 +02006359 if (!bs) {
6360 return QTAILQ_FIRST(&all_bdrv_states);
6361 }
6362 return QTAILQ_NEXT(bs, bs_list);
6363}
6364
Fam Zheng20a9e772014-10-31 11:32:55 +08006365const char *bdrv_get_node_name(const BlockDriverState *bs)
6366{
Emanuele Giuseppe Esposito384a48f2022-03-03 10:15:50 -05006367 IO_CODE();
Fam Zheng20a9e772014-10-31 11:32:55 +08006368 return bs->node_name;
6369}
6370
Kevin Wolf1f0c4612016-03-22 18:38:44 +01006371const char *bdrv_get_parent_name(const BlockDriverState *bs)
Kevin Wolf4c265bf2016-02-26 10:22:16 +01006372{
6373 BdrvChild *c;
6374 const char *name;
Emanuele Giuseppe Esposito967d7902022-03-03 10:15:58 -05006375 IO_CODE();
Kevin Wolf4c265bf2016-02-26 10:22:16 +01006376
6377 /* If multiple parents have a name, just pick the first one. */
6378 QLIST_FOREACH(c, &bs->parents, next_parent) {
Max Reitzbd86fb92020-05-13 13:05:13 +02006379 if (c->klass->get_name) {
6380 name = c->klass->get_name(c);
Kevin Wolf4c265bf2016-02-26 10:22:16 +01006381 if (name && *name) {
6382 return name;
6383 }
6384 }
6385 }
6386
6387 return NULL;
6388}
6389
Markus Armbruster7f06d472014-10-07 13:59:12 +02006390/* TODO check what callers really want: bs->node_name or blk_name() */
Markus Armbrusterbfb197e2014-10-07 13:59:11 +02006391const char *bdrv_get_device_name(const BlockDriverState *bs)
bellardea2384d2004-08-01 21:59:26 +00006392{
Emanuele Giuseppe Esposito384a48f2022-03-03 10:15:50 -05006393 IO_CODE();
Kevin Wolf4c265bf2016-02-26 10:22:16 +01006394 return bdrv_get_parent_name(bs) ?: "";
bellardea2384d2004-08-01 21:59:26 +00006395}
6396
Alberto Garcia9b2aa842015-04-08 12:29:18 +03006397/* This can be used to identify nodes that might not have a device
6398 * name associated. Since node and device names live in the same
6399 * namespace, the result is unambiguous. The exception is if both are
6400 * absent, then this returns an empty (non-null) string. */
6401const char *bdrv_get_device_or_node_name(const BlockDriverState *bs)
6402{
Emanuele Giuseppe Esposito384a48f2022-03-03 10:15:50 -05006403 IO_CODE();
Kevin Wolf4c265bf2016-02-26 10:22:16 +01006404 return bdrv_get_parent_name(bs) ?: bs->node_name;
Alberto Garcia9b2aa842015-04-08 12:29:18 +03006405}
6406
Markus Armbrusterc8433282012-06-05 16:49:24 +02006407int bdrv_get_flags(BlockDriverState *bs)
6408{
Hanna Reitz15aee7a2022-04-27 13:40:54 +02006409 IO_CODE();
Markus Armbrusterc8433282012-06-05 16:49:24 +02006410 return bs->open_flags;
6411}
6412
Peter Lieven3ac21622013-06-28 12:47:42 +02006413int bdrv_has_zero_init_1(BlockDriverState *bs)
6414{
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05006415 GLOBAL_STATE_CODE();
Peter Lieven3ac21622013-06-28 12:47:42 +02006416 return 1;
6417}
6418
Kevin Wolff2feebb2010-04-14 17:30:35 +02006419int bdrv_has_zero_init(BlockDriverState *bs)
6420{
Max Reitz93393e62019-06-12 17:03:38 +02006421 BlockDriverState *filtered;
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05006422 GLOBAL_STATE_CODE();
Max Reitz93393e62019-06-12 17:03:38 +02006423
Max Reitzd470ad42017-11-10 21:31:09 +01006424 if (!bs->drv) {
6425 return 0;
6426 }
Kevin Wolff2feebb2010-04-14 17:30:35 +02006427
Paolo Bonzini11212d82013-09-04 19:00:27 +02006428 /* If BS is a copy on write image, it is initialized to
6429 the contents of the base image, which may not be zeroes. */
Max Reitz34778172019-06-12 17:10:46 +02006430 if (bdrv_cow_child(bs)) {
Paolo Bonzini11212d82013-09-04 19:00:27 +02006431 return 0;
6432 }
Kevin Wolf336c1c12010-07-28 11:26:29 +02006433 if (bs->drv->bdrv_has_zero_init) {
6434 return bs->drv->bdrv_has_zero_init(bs);
Kevin Wolff2feebb2010-04-14 17:30:35 +02006435 }
Max Reitz93393e62019-06-12 17:03:38 +02006436
6437 filtered = bdrv_filter_bs(bs);
6438 if (filtered) {
6439 return bdrv_has_zero_init(filtered);
Manos Pitsidianakis5a612c02017-07-13 18:30:25 +03006440 }
Kevin Wolff2feebb2010-04-14 17:30:35 +02006441
Peter Lieven3ac21622013-06-28 12:47:42 +02006442 /* safe default */
6443 return 0;
Kevin Wolff2feebb2010-04-14 17:30:35 +02006444}
6445
Peter Lieven4ce78692013-10-24 12:06:54 +02006446bool bdrv_can_write_zeroes_with_unmap(BlockDriverState *bs)
6447{
Emanuele Giuseppe Esposito384a48f2022-03-03 10:15:50 -05006448 IO_CODE();
Denis V. Lunev2f0342e2016-07-14 16:33:26 +03006449 if (!(bs->open_flags & BDRV_O_UNMAP)) {
Peter Lieven4ce78692013-10-24 12:06:54 +02006450 return false;
6451 }
6452
Eric Blakee24d8132018-01-26 13:34:39 -06006453 return bs->supported_zero_flags & BDRV_REQ_MAY_UNMAP;
Peter Lieven4ce78692013-10-24 12:06:54 +02006454}
6455
ths5fafdf22007-09-16 21:08:06 +00006456void bdrv_get_backing_filename(BlockDriverState *bs,
bellard83f64092006-08-01 16:21:11 +00006457 char *filename, int filename_size)
bellardea2384d2004-08-01 21:59:26 +00006458{
Emanuele Giuseppe Esposito384a48f2022-03-03 10:15:50 -05006459 IO_CODE();
Kevin Wolf3574c602011-10-26 11:02:11 +02006460 pstrcpy(filename, filename_size, bs->backing_file);
bellardea2384d2004-08-01 21:59:26 +00006461}
6462
Emanuele Giuseppe Esposito3d47eb02023-01-13 21:42:08 +01006463int coroutine_fn bdrv_co_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
bellardfaea38e2006-08-05 21:31:00 +00006464{
Vladimir Sementsov-Ogievskiy8b117002020-12-04 01:27:13 +03006465 int ret;
bellardfaea38e2006-08-05 21:31:00 +00006466 BlockDriver *drv = bs->drv;
Emanuele Giuseppe Esposito384a48f2022-03-03 10:15:50 -05006467 IO_CODE();
Emanuele Giuseppe Espositoa00e70c2023-05-04 13:57:44 +02006468 assert_bdrv_graph_readable();
6469
Manos Pitsidianakis5a612c02017-07-13 18:30:25 +03006470 /* if bs->drv == NULL, bs is closed, so there's nothing to do here */
6471 if (!drv) {
bellard19cb3732006-08-19 11:45:59 +00006472 return -ENOMEDIUM;
Manos Pitsidianakis5a612c02017-07-13 18:30:25 +03006473 }
Emanuele Giuseppe Esposito3d47eb02023-01-13 21:42:08 +01006474 if (!drv->bdrv_co_get_info) {
Max Reitz93393e62019-06-12 17:03:38 +02006475 BlockDriverState *filtered = bdrv_filter_bs(bs);
6476 if (filtered) {
Emanuele Giuseppe Esposito3d47eb02023-01-13 21:42:08 +01006477 return bdrv_co_get_info(filtered, bdi);
Manos Pitsidianakis5a612c02017-07-13 18:30:25 +03006478 }
bellardfaea38e2006-08-05 21:31:00 +00006479 return -ENOTSUP;
Manos Pitsidianakis5a612c02017-07-13 18:30:25 +03006480 }
bellardfaea38e2006-08-05 21:31:00 +00006481 memset(bdi, 0, sizeof(*bdi));
Emanuele Giuseppe Esposito3d47eb02023-01-13 21:42:08 +01006482 ret = drv->bdrv_co_get_info(bs, bdi);
Vladimir Sementsov-Ogievskiy8b117002020-12-04 01:27:13 +03006483 if (ret < 0) {
6484 return ret;
6485 }
6486
6487 if (bdi->cluster_size > BDRV_MAX_ALIGNMENT) {
6488 return -EINVAL;
6489 }
6490
6491 return 0;
bellardfaea38e2006-08-05 21:31:00 +00006492}
6493
Andrey Shinkevich1bf6e9c2019-02-08 18:06:06 +03006494ImageInfoSpecific *bdrv_get_specific_info(BlockDriverState *bs,
6495 Error **errp)
Max Reitzeae041f2013-10-09 10:46:16 +02006496{
6497 BlockDriver *drv = bs->drv;
Emanuele Giuseppe Esposito384a48f2022-03-03 10:15:50 -05006498 IO_CODE();
Max Reitzeae041f2013-10-09 10:46:16 +02006499 if (drv && drv->bdrv_get_specific_info) {
Andrey Shinkevich1bf6e9c2019-02-08 18:06:06 +03006500 return drv->bdrv_get_specific_info(bs, errp);
Max Reitzeae041f2013-10-09 10:46:16 +02006501 }
6502 return NULL;
6503}
6504
Anton Nefedovd9245592019-09-23 15:17:37 +03006505BlockStatsSpecific *bdrv_get_specific_stats(BlockDriverState *bs)
6506{
6507 BlockDriver *drv = bs->drv;
Emanuele Giuseppe Esposito384a48f2022-03-03 10:15:50 -05006508 IO_CODE();
Anton Nefedovd9245592019-09-23 15:17:37 +03006509 if (!drv || !drv->bdrv_get_specific_stats) {
6510 return NULL;
6511 }
6512 return drv->bdrv_get_specific_stats(bs);
6513}
6514
Emanuele Giuseppe Espositoc834dc02023-01-13 21:42:11 +01006515void coroutine_fn bdrv_co_debug_event(BlockDriverState *bs, BlkdebugEvent event)
Kevin Wolf8b9b0cc2010-03-15 17:27:00 +01006516{
Emanuele Giuseppe Esposito384a48f2022-03-03 10:15:50 -05006517 IO_CODE();
Emanuele Giuseppe Espositocb2bfaa2023-05-04 13:57:45 +02006518 assert_bdrv_graph_readable();
6519
Emanuele Giuseppe Espositoc834dc02023-01-13 21:42:11 +01006520 if (!bs || !bs->drv || !bs->drv->bdrv_co_debug_event) {
Kevin Wolf8b9b0cc2010-03-15 17:27:00 +01006521 return;
6522 }
6523
Emanuele Giuseppe Espositoc834dc02023-01-13 21:42:11 +01006524 bs->drv->bdrv_co_debug_event(bs, event);
Kevin Wolf41c695c2012-12-06 14:32:58 +01006525}
Kevin Wolf8b9b0cc2010-03-15 17:27:00 +01006526
Vladimir Sementsov-Ogievskiyd10529a2019-09-20 17:20:49 +03006527static BlockDriverState *bdrv_find_debug_node(BlockDriverState *bs)
Kevin Wolf41c695c2012-12-06 14:32:58 +01006528{
Emanuele Giuseppe Espositobdb73472022-03-03 10:16:02 -05006529 GLOBAL_STATE_CODE();
Kevin Wolf41c695c2012-12-06 14:32:58 +01006530 while (bs && bs->drv && !bs->drv->bdrv_debug_breakpoint) {
Max Reitzf706a922019-06-12 17:42:13 +02006531 bs = bdrv_primary_bs(bs);
Kevin Wolf41c695c2012-12-06 14:32:58 +01006532 }
6533
6534 if (bs && bs->drv && bs->drv->bdrv_debug_breakpoint) {
Vladimir Sementsov-Ogievskiyd10529a2019-09-20 17:20:49 +03006535 assert(bs->drv->bdrv_debug_remove_breakpoint);
6536 return bs;
6537 }
6538
6539 return NULL;
6540}
6541
6542int bdrv_debug_breakpoint(BlockDriverState *bs, const char *event,
6543 const char *tag)
6544{
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05006545 GLOBAL_STATE_CODE();
Vladimir Sementsov-Ogievskiyd10529a2019-09-20 17:20:49 +03006546 bs = bdrv_find_debug_node(bs);
6547 if (bs) {
Kevin Wolf41c695c2012-12-06 14:32:58 +01006548 return bs->drv->bdrv_debug_breakpoint(bs, event, tag);
6549 }
6550
6551 return -ENOTSUP;
6552}
6553
Fam Zheng4cc70e92013-11-20 10:01:54 +08006554int bdrv_debug_remove_breakpoint(BlockDriverState *bs, const char *tag)
6555{
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05006556 GLOBAL_STATE_CODE();
Vladimir Sementsov-Ogievskiyd10529a2019-09-20 17:20:49 +03006557 bs = bdrv_find_debug_node(bs);
6558 if (bs) {
Fam Zheng4cc70e92013-11-20 10:01:54 +08006559 return bs->drv->bdrv_debug_remove_breakpoint(bs, tag);
6560 }
6561
6562 return -ENOTSUP;
6563}
6564
Kevin Wolf41c695c2012-12-06 14:32:58 +01006565int bdrv_debug_resume(BlockDriverState *bs, const char *tag)
6566{
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05006567 GLOBAL_STATE_CODE();
Max Reitz938789e2014-03-10 23:44:08 +01006568 while (bs && (!bs->drv || !bs->drv->bdrv_debug_resume)) {
Max Reitzf706a922019-06-12 17:42:13 +02006569 bs = bdrv_primary_bs(bs);
Kevin Wolf41c695c2012-12-06 14:32:58 +01006570 }
6571
6572 if (bs && bs->drv && bs->drv->bdrv_debug_resume) {
6573 return bs->drv->bdrv_debug_resume(bs, tag);
6574 }
6575
6576 return -ENOTSUP;
6577}
6578
6579bool bdrv_debug_is_suspended(BlockDriverState *bs, const char *tag)
6580{
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05006581 GLOBAL_STATE_CODE();
Kevin Wolf41c695c2012-12-06 14:32:58 +01006582 while (bs && bs->drv && !bs->drv->bdrv_debug_is_suspended) {
Max Reitzf706a922019-06-12 17:42:13 +02006583 bs = bdrv_primary_bs(bs);
Kevin Wolf41c695c2012-12-06 14:32:58 +01006584 }
6585
6586 if (bs && bs->drv && bs->drv->bdrv_debug_is_suspended) {
6587 return bs->drv->bdrv_debug_is_suspended(bs, tag);
6588 }
6589
6590 return false;
Kevin Wolf8b9b0cc2010-03-15 17:27:00 +01006591}
6592
Jeff Codyb1b1d782012-10-16 15:49:09 -04006593/* backing_file can either be relative, or absolute, or a protocol. If it is
6594 * relative, it must be relative to the chain. So, passing in bs->filename
6595 * from a BDS as backing_file should not be done, as that may be relative to
6596 * the CWD rather than the chain. */
Marcelo Tosattie8a6bb92012-01-18 14:40:51 +00006597BlockDriverState *bdrv_find_backing_image(BlockDriverState *bs,
6598 const char *backing_file)
6599{
Jeff Codyb1b1d782012-10-16 15:49:09 -04006600 char *filename_full = NULL;
6601 char *backing_file_full = NULL;
6602 char *filename_tmp = NULL;
6603 int is_protocol = 0;
Max Reitz0b877d02018-08-01 20:34:11 +02006604 bool filenames_refreshed = false;
Jeff Codyb1b1d782012-10-16 15:49:09 -04006605 BlockDriverState *curr_bs = NULL;
6606 BlockDriverState *retval = NULL;
Max Reitzdcf3f9b2019-06-12 17:34:45 +02006607 BlockDriverState *bs_below;
Jeff Codyb1b1d782012-10-16 15:49:09 -04006608
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05006609 GLOBAL_STATE_CODE();
6610
Jeff Codyb1b1d782012-10-16 15:49:09 -04006611 if (!bs || !bs->drv || !backing_file) {
Marcelo Tosattie8a6bb92012-01-18 14:40:51 +00006612 return NULL;
6613 }
6614
Jeff Codyb1b1d782012-10-16 15:49:09 -04006615 filename_full = g_malloc(PATH_MAX);
6616 backing_file_full = g_malloc(PATH_MAX);
Jeff Codyb1b1d782012-10-16 15:49:09 -04006617
6618 is_protocol = path_has_protocol(backing_file);
6619
Max Reitzdcf3f9b2019-06-12 17:34:45 +02006620 /*
6621 * Being largely a legacy function, skip any filters here
6622 * (because filters do not have normal filenames, so they cannot
6623 * match anyway; and allowing json:{} filenames is a bit out of
6624 * scope).
6625 */
6626 for (curr_bs = bdrv_skip_filters(bs);
6627 bdrv_cow_child(curr_bs) != NULL;
6628 curr_bs = bs_below)
6629 {
6630 bs_below = bdrv_backing_chain_next(curr_bs);
Jeff Codyb1b1d782012-10-16 15:49:09 -04006631
Max Reitz0b877d02018-08-01 20:34:11 +02006632 if (bdrv_backing_overridden(curr_bs)) {
6633 /*
6634 * If the backing file was overridden, we can only compare
6635 * directly against the backing node's filename.
6636 */
6637
6638 if (!filenames_refreshed) {
6639 /*
6640 * This will automatically refresh all of the
6641 * filenames in the rest of the backing chain, so we
6642 * only need to do this once.
6643 */
6644 bdrv_refresh_filename(bs_below);
6645 filenames_refreshed = true;
6646 }
6647
6648 if (strcmp(backing_file, bs_below->filename) == 0) {
6649 retval = bs_below;
6650 break;
6651 }
6652 } else if (is_protocol || path_has_protocol(curr_bs->backing_file)) {
6653 /*
6654 * If either of the filename paths is actually a protocol, then
6655 * compare unmodified paths; otherwise make paths relative.
6656 */
Max Reitz6b6833c2019-02-01 20:29:15 +01006657 char *backing_file_full_ret;
6658
Jeff Codyb1b1d782012-10-16 15:49:09 -04006659 if (strcmp(backing_file, curr_bs->backing_file) == 0) {
Max Reitzdcf3f9b2019-06-12 17:34:45 +02006660 retval = bs_below;
Jeff Codyb1b1d782012-10-16 15:49:09 -04006661 break;
6662 }
Jeff Cody418661e2017-01-25 20:08:20 -05006663 /* Also check against the full backing filename for the image */
Max Reitz6b6833c2019-02-01 20:29:15 +01006664 backing_file_full_ret = bdrv_get_full_backing_filename(curr_bs,
6665 NULL);
6666 if (backing_file_full_ret) {
6667 bool equal = strcmp(backing_file, backing_file_full_ret) == 0;
6668 g_free(backing_file_full_ret);
6669 if (equal) {
Max Reitzdcf3f9b2019-06-12 17:34:45 +02006670 retval = bs_below;
Jeff Cody418661e2017-01-25 20:08:20 -05006671 break;
6672 }
Jeff Cody418661e2017-01-25 20:08:20 -05006673 }
Marcelo Tosattie8a6bb92012-01-18 14:40:51 +00006674 } else {
Jeff Codyb1b1d782012-10-16 15:49:09 -04006675 /* If not an absolute filename path, make it relative to the current
6676 * image's filename path */
Max Reitz2d9158c2019-02-01 20:29:17 +01006677 filename_tmp = bdrv_make_absolute_filename(curr_bs, backing_file,
6678 NULL);
6679 /* We are going to compare canonicalized absolute pathnames */
6680 if (!filename_tmp || !realpath(filename_tmp, filename_full)) {
6681 g_free(filename_tmp);
Jeff Codyb1b1d782012-10-16 15:49:09 -04006682 continue;
6683 }
Max Reitz2d9158c2019-02-01 20:29:17 +01006684 g_free(filename_tmp);
Jeff Codyb1b1d782012-10-16 15:49:09 -04006685
6686 /* We need to make sure the backing filename we are comparing against
6687 * is relative to the current image filename (or absolute) */
Max Reitz2d9158c2019-02-01 20:29:17 +01006688 filename_tmp = bdrv_get_full_backing_filename(curr_bs, NULL);
6689 if (!filename_tmp || !realpath(filename_tmp, backing_file_full)) {
6690 g_free(filename_tmp);
Jeff Codyb1b1d782012-10-16 15:49:09 -04006691 continue;
6692 }
Max Reitz2d9158c2019-02-01 20:29:17 +01006693 g_free(filename_tmp);
Jeff Codyb1b1d782012-10-16 15:49:09 -04006694
6695 if (strcmp(backing_file_full, filename_full) == 0) {
Max Reitzdcf3f9b2019-06-12 17:34:45 +02006696 retval = bs_below;
Jeff Codyb1b1d782012-10-16 15:49:09 -04006697 break;
6698 }
Marcelo Tosattie8a6bb92012-01-18 14:40:51 +00006699 }
6700 }
6701
Jeff Codyb1b1d782012-10-16 15:49:09 -04006702 g_free(filename_full);
6703 g_free(backing_file_full);
Jeff Codyb1b1d782012-10-16 15:49:09 -04006704 return retval;
Marcelo Tosattie8a6bb92012-01-18 14:40:51 +00006705}
6706
bellardea2384d2004-08-01 21:59:26 +00006707void bdrv_init(void)
6708{
Kevin Wolfe5f05f82021-07-09 18:41:41 +02006709#ifdef CONFIG_BDRV_WHITELIST_TOOLS
6710 use_bdrv_whitelist = 1;
6711#endif
Anthony Liguori5efa9d52009-05-09 17:03:42 -05006712 module_call_init(MODULE_INIT_BLOCK);
bellardea2384d2004-08-01 21:59:26 +00006713}
pbrookce1a14d2006-08-07 02:38:06 +00006714
Markus Armbrustereb852012009-10-27 18:41:44 +01006715void bdrv_init_with_whitelist(void)
6716{
6717 use_bdrv_whitelist = 1;
6718 bdrv_init();
6719}
6720
Emanuele Giuseppe Espositoa94750d2022-02-09 05:54:50 -05006721int bdrv_activate(BlockDriverState *bs, Error **errp)
6722{
Kevin Wolf4417ab72017-05-04 18:52:37 +02006723 BdrvChild *child, *parent;
Kevin Wolf5a8a30d2014-03-12 15:59:16 +01006724 Error *local_err = NULL;
6725 int ret;
Vladimir Sementsov-Ogievskiy9c98f142018-10-29 16:23:17 -04006726 BdrvDirtyBitmap *bm;
Kevin Wolf5a8a30d2014-03-12 15:59:16 +01006727
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05006728 GLOBAL_STATE_CODE();
6729
Kevin Wolf3456a8d2014-03-11 10:58:39 +01006730 if (!bs->drv) {
Vladimir Sementsov-Ogievskiy54166452020-09-24 21:54:08 +03006731 return -ENOMEDIUM;
Anthony Liguori0f154232011-11-14 15:09:45 -06006732 }
Kevin Wolf3456a8d2014-03-11 10:58:39 +01006733
Vladimir Sementsov-Ogievskiy16e977d2017-01-31 14:23:08 +03006734 QLIST_FOREACH(child, &bs->children, next) {
Emanuele Giuseppe Esposito11d0c9b2022-02-09 05:54:52 -05006735 bdrv_activate(child->bs, &local_err);
Fam Zheng0d1c5c92016-05-11 10:45:33 +08006736 if (local_err) {
Fam Zheng0d1c5c92016-05-11 10:45:33 +08006737 error_propagate(errp, local_err);
Vladimir Sementsov-Ogievskiy54166452020-09-24 21:54:08 +03006738 return -EINVAL;
Fam Zheng0d1c5c92016-05-11 10:45:33 +08006739 }
Kevin Wolf5a8a30d2014-03-12 15:59:16 +01006740 }
Fam Zheng0d1c5c92016-05-11 10:45:33 +08006741
Kevin Wolfdafe0962017-11-16 13:00:01 +01006742 /*
6743 * Update permissions, they may differ for inactive nodes.
6744 *
6745 * Note that the required permissions of inactive images are always a
6746 * subset of the permissions required after activating the image. This
6747 * allows us to just get the permissions upfront without restricting
Emanuele Giuseppe Esposito11d0c9b2022-02-09 05:54:52 -05006748 * bdrv_co_invalidate_cache().
Kevin Wolfdafe0962017-11-16 13:00:01 +01006749 *
6750 * It also means that in error cases, we don't have to try and revert to
6751 * the old permissions (which is an operation that could fail, too). We can
6752 * just keep the extended permissions for the next time that an activation
6753 * of the image is tried.
6754 */
Kevin Wolf7bb49412019-12-17 15:06:38 +01006755 if (bs->open_flags & BDRV_O_INACTIVE) {
6756 bs->open_flags &= ~BDRV_O_INACTIVE;
Vladimir Sementsov-Ogievskiyf1316ed2022-11-07 19:35:57 +03006757 ret = bdrv_refresh_perms(bs, NULL, errp);
Kevin Wolf7bb49412019-12-17 15:06:38 +01006758 if (ret < 0) {
Fam Zheng0d1c5c92016-05-11 10:45:33 +08006759 bs->open_flags |= BDRV_O_INACTIVE;
Vladimir Sementsov-Ogievskiy54166452020-09-24 21:54:08 +03006760 return ret;
Fam Zheng0d1c5c92016-05-11 10:45:33 +08006761 }
Kevin Wolf3456a8d2014-03-11 10:58:39 +01006762
Emanuele Giuseppe Esposito11d0c9b2022-02-09 05:54:52 -05006763 ret = bdrv_invalidate_cache(bs, errp);
6764 if (ret < 0) {
6765 bs->open_flags |= BDRV_O_INACTIVE;
6766 return ret;
Kevin Wolf7bb49412019-12-17 15:06:38 +01006767 }
Vladimir Sementsov-Ogievskiy9c98f142018-10-29 16:23:17 -04006768
Kevin Wolf7bb49412019-12-17 15:06:38 +01006769 FOR_EACH_DIRTY_BITMAP(bs, bm) {
6770 bdrv_dirty_bitmap_skip_store(bm, false);
6771 }
6772
Emanuele Giuseppe Espositoc0579602023-01-13 21:42:03 +01006773 ret = bdrv_refresh_total_sectors(bs, bs->total_sectors);
Kevin Wolf7bb49412019-12-17 15:06:38 +01006774 if (ret < 0) {
6775 bs->open_flags |= BDRV_O_INACTIVE;
6776 error_setg_errno(errp, -ret, "Could not refresh total sector count");
Vladimir Sementsov-Ogievskiy54166452020-09-24 21:54:08 +03006777 return ret;
Kevin Wolf7bb49412019-12-17 15:06:38 +01006778 }
Kevin Wolf5a8a30d2014-03-12 15:59:16 +01006779 }
Kevin Wolf4417ab72017-05-04 18:52:37 +02006780
6781 QLIST_FOREACH(parent, &bs->parents, next_parent) {
Max Reitzbd86fb92020-05-13 13:05:13 +02006782 if (parent->klass->activate) {
6783 parent->klass->activate(parent, &local_err);
Kevin Wolf4417ab72017-05-04 18:52:37 +02006784 if (local_err) {
Kevin Wolf78fc3b32019-01-31 15:16:10 +01006785 bs->open_flags |= BDRV_O_INACTIVE;
Kevin Wolf4417ab72017-05-04 18:52:37 +02006786 error_propagate(errp, local_err);
Vladimir Sementsov-Ogievskiy54166452020-09-24 21:54:08 +03006787 return -EINVAL;
Kevin Wolf4417ab72017-05-04 18:52:37 +02006788 }
6789 }
6790 }
Vladimir Sementsov-Ogievskiy54166452020-09-24 21:54:08 +03006791
6792 return 0;
Anthony Liguori0f154232011-11-14 15:09:45 -06006793}
6794
Emanuele Giuseppe Esposito11d0c9b2022-02-09 05:54:52 -05006795int coroutine_fn bdrv_co_invalidate_cache(BlockDriverState *bs, Error **errp)
6796{
6797 Error *local_err = NULL;
Emanuele Giuseppe Esposito1581a702022-03-03 10:16:09 -05006798 IO_CODE();
Emanuele Giuseppe Esposito11d0c9b2022-02-09 05:54:52 -05006799
6800 assert(!(bs->open_flags & BDRV_O_INACTIVE));
Kevin Wolf1b3ff9f2022-12-07 14:18:38 +01006801 assert_bdrv_graph_readable();
Emanuele Giuseppe Esposito11d0c9b2022-02-09 05:54:52 -05006802
6803 if (bs->drv->bdrv_co_invalidate_cache) {
6804 bs->drv->bdrv_co_invalidate_cache(bs, &local_err);
6805 if (local_err) {
6806 error_propagate(errp, local_err);
6807 return -EINVAL;
6808 }
6809 }
6810
6811 return 0;
6812}
6813
Emanuele Giuseppe Esposito3b717192022-02-09 05:54:51 -05006814void bdrv_activate_all(Error **errp)
Anthony Liguori0f154232011-11-14 15:09:45 -06006815{
Kevin Wolf7c8eece2016-03-22 18:58:50 +01006816 BlockDriverState *bs;
Kevin Wolf88be7b42016-05-20 18:49:07 +02006817 BdrvNextIterator it;
Anthony Liguori0f154232011-11-14 15:09:45 -06006818
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05006819 GLOBAL_STATE_CODE();
6820
Kevin Wolf88be7b42016-05-20 18:49:07 +02006821 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
Stefan Hajnoczied78cda2014-05-08 16:34:35 +02006822 AioContext *aio_context = bdrv_get_aio_context(bs);
Vladimir Sementsov-Ogievskiy54166452020-09-24 21:54:08 +03006823 int ret;
Stefan Hajnoczied78cda2014-05-08 16:34:35 +02006824
6825 aio_context_acquire(aio_context);
Emanuele Giuseppe Espositoa94750d2022-02-09 05:54:50 -05006826 ret = bdrv_activate(bs, errp);
Stefan Hajnoczied78cda2014-05-08 16:34:35 +02006827 aio_context_release(aio_context);
Vladimir Sementsov-Ogievskiy54166452020-09-24 21:54:08 +03006828 if (ret < 0) {
Max Reitz5e003f12017-11-10 18:25:45 +01006829 bdrv_next_cleanup(&it);
Kevin Wolf5a8a30d2014-03-12 15:59:16 +01006830 return;
6831 }
Anthony Liguori0f154232011-11-14 15:09:45 -06006832 }
6833}
6834
Kevin Wolf9e372712018-11-23 15:11:14 +01006835static bool bdrv_has_bds_parent(BlockDriverState *bs, bool only_active)
6836{
6837 BdrvChild *parent;
Emanuele Giuseppe Espositobdb73472022-03-03 10:16:02 -05006838 GLOBAL_STATE_CODE();
Kevin Wolf9e372712018-11-23 15:11:14 +01006839
6840 QLIST_FOREACH(parent, &bs->parents, next_parent) {
Max Reitzbd86fb92020-05-13 13:05:13 +02006841 if (parent->klass->parent_is_bds) {
Kevin Wolf9e372712018-11-23 15:11:14 +01006842 BlockDriverState *parent_bs = parent->opaque;
6843 if (!only_active || !(parent_bs->open_flags & BDRV_O_INACTIVE)) {
6844 return true;
6845 }
6846 }
6847 }
6848
6849 return false;
6850}
6851
6852static int bdrv_inactivate_recurse(BlockDriverState *bs)
Kevin Wolf76b1c7f2015-12-22 14:07:08 +01006853{
Kevin Wolfcfa1a572017-05-04 18:52:38 +02006854 BdrvChild *child, *parent;
Kevin Wolf76b1c7f2015-12-22 14:07:08 +01006855 int ret;
Vladimir Sementsov-Ogievskiya13de402021-09-11 15:00:27 +03006856 uint64_t cumulative_perms, cumulative_shared_perms;
Kevin Wolf76b1c7f2015-12-22 14:07:08 +01006857
Emanuele Giuseppe Espositoda359902022-03-03 10:16:11 -05006858 GLOBAL_STATE_CODE();
6859
Max Reitzd470ad42017-11-10 21:31:09 +01006860 if (!bs->drv) {
6861 return -ENOMEDIUM;
6862 }
6863
Kevin Wolf9e372712018-11-23 15:11:14 +01006864 /* Make sure that we don't inactivate a child before its parent.
6865 * It will be covered by recursion from the yet active parent. */
6866 if (bdrv_has_bds_parent(bs, true)) {
6867 return 0;
6868 }
6869
6870 assert(!(bs->open_flags & BDRV_O_INACTIVE));
6871
6872 /* Inactivate this node */
6873 if (bs->drv->bdrv_inactivate) {
Kevin Wolf76b1c7f2015-12-22 14:07:08 +01006874 ret = bs->drv->bdrv_inactivate(bs);
6875 if (ret < 0) {
6876 return ret;
6877 }
6878 }
6879
Kevin Wolf9e372712018-11-23 15:11:14 +01006880 QLIST_FOREACH(parent, &bs->parents, next_parent) {
Max Reitzbd86fb92020-05-13 13:05:13 +02006881 if (parent->klass->inactivate) {
6882 ret = parent->klass->inactivate(parent);
Kevin Wolf9e372712018-11-23 15:11:14 +01006883 if (ret < 0) {
6884 return ret;
Kevin Wolfcfa1a572017-05-04 18:52:38 +02006885 }
6886 }
Fam Zhengaad0b7a2016-05-11 10:45:35 +08006887 }
Kevin Wolf38701b62017-05-04 18:52:39 +02006888
Vladimir Sementsov-Ogievskiya13de402021-09-11 15:00:27 +03006889 bdrv_get_cumulative_perm(bs, &cumulative_perms,
6890 &cumulative_shared_perms);
6891 if (cumulative_perms & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)) {
6892 /* Our inactive parents still need write access. Inactivation failed. */
6893 return -EPERM;
6894 }
6895
Kevin Wolf9e372712018-11-23 15:11:14 +01006896 bs->open_flags |= BDRV_O_INACTIVE;
6897
Vladimir Sementsov-Ogievskiybb87e4d2020-11-06 15:42:38 +03006898 /*
6899 * Update permissions, they may differ for inactive nodes.
6900 * We only tried to loosen restrictions, so errors are not fatal, ignore
6901 * them.
6902 */
Vladimir Sementsov-Ogievskiyf1316ed2022-11-07 19:35:57 +03006903 bdrv_refresh_perms(bs, NULL, NULL);
Kevin Wolf9e372712018-11-23 15:11:14 +01006904
6905 /* Recursively inactivate children */
Kevin Wolf38701b62017-05-04 18:52:39 +02006906 QLIST_FOREACH(child, &bs->children, next) {
Kevin Wolf9e372712018-11-23 15:11:14 +01006907 ret = bdrv_inactivate_recurse(child->bs);
Kevin Wolf38701b62017-05-04 18:52:39 +02006908 if (ret < 0) {
6909 return ret;
6910 }
6911 }
6912
Kevin Wolf76b1c7f2015-12-22 14:07:08 +01006913 return 0;
6914}
6915
6916int bdrv_inactivate_all(void)
6917{
Max Reitz79720af2016-03-16 19:54:44 +01006918 BlockDriverState *bs = NULL;
Kevin Wolf88be7b42016-05-20 18:49:07 +02006919 BdrvNextIterator it;
Fam Zhengaad0b7a2016-05-11 10:45:35 +08006920 int ret = 0;
Paolo Bonzinibd6458e2017-12-07 20:13:15 +00006921 GSList *aio_ctxs = NULL, *ctx;
Kevin Wolf76b1c7f2015-12-22 14:07:08 +01006922
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05006923 GLOBAL_STATE_CODE();
6924
Kevin Wolf88be7b42016-05-20 18:49:07 +02006925 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
Paolo Bonzinibd6458e2017-12-07 20:13:15 +00006926 AioContext *aio_context = bdrv_get_aio_context(bs);
6927
6928 if (!g_slist_find(aio_ctxs, aio_context)) {
6929 aio_ctxs = g_slist_prepend(aio_ctxs, aio_context);
6930 aio_context_acquire(aio_context);
6931 }
Fam Zhengaad0b7a2016-05-11 10:45:35 +08006932 }
Kevin Wolf76b1c7f2015-12-22 14:07:08 +01006933
Kevin Wolf9e372712018-11-23 15:11:14 +01006934 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
6935 /* Nodes with BDS parents are covered by recursion from the last
6936 * parent that gets inactivated. Don't inactivate them a second
6937 * time if that has already happened. */
6938 if (bdrv_has_bds_parent(bs, false)) {
6939 continue;
6940 }
6941 ret = bdrv_inactivate_recurse(bs);
6942 if (ret < 0) {
6943 bdrv_next_cleanup(&it);
6944 goto out;
Kevin Wolf76b1c7f2015-12-22 14:07:08 +01006945 }
6946 }
6947
Fam Zhengaad0b7a2016-05-11 10:45:35 +08006948out:
Paolo Bonzinibd6458e2017-12-07 20:13:15 +00006949 for (ctx = aio_ctxs; ctx != NULL; ctx = ctx->next) {
6950 AioContext *aio_context = ctx->data;
6951 aio_context_release(aio_context);
Fam Zhengaad0b7a2016-05-11 10:45:35 +08006952 }
Paolo Bonzinibd6458e2017-12-07 20:13:15 +00006953 g_slist_free(aio_ctxs);
Fam Zhengaad0b7a2016-05-11 10:45:35 +08006954
6955 return ret;
Kevin Wolf76b1c7f2015-12-22 14:07:08 +01006956}
6957
Kevin Wolff9f05dc2011-07-15 13:50:26 +02006958/**************************************************************/
bellard19cb3732006-08-19 11:45:59 +00006959/* removable device support */
6960
6961/**
6962 * Return TRUE if the media is present
6963 */
Emanuele Giuseppe Esposito1e97be92023-01-13 21:42:02 +01006964bool coroutine_fn bdrv_co_is_inserted(BlockDriverState *bs)
bellard19cb3732006-08-19 11:45:59 +00006965{
6966 BlockDriver *drv = bs->drv;
Max Reitz28d7a782015-10-19 17:53:13 +02006967 BdrvChild *child;
Emanuele Giuseppe Esposito384a48f2022-03-03 10:15:50 -05006968 IO_CODE();
Emanuele Giuseppe Espositoc73ff922023-02-03 16:21:57 +01006969 assert_bdrv_graph_readable();
Markus Armbrustera1aff5b2011-09-06 18:58:41 +02006970
Max Reitze031f752015-10-19 17:53:11 +02006971 if (!drv) {
6972 return false;
6973 }
Emanuele Giuseppe Esposito1e97be92023-01-13 21:42:02 +01006974 if (drv->bdrv_co_is_inserted) {
6975 return drv->bdrv_co_is_inserted(bs);
Max Reitze031f752015-10-19 17:53:11 +02006976 }
Max Reitz28d7a782015-10-19 17:53:13 +02006977 QLIST_FOREACH(child, &bs->children, next) {
Emanuele Giuseppe Esposito1e97be92023-01-13 21:42:02 +01006978 if (!bdrv_co_is_inserted(child->bs)) {
Max Reitz28d7a782015-10-19 17:53:13 +02006979 return false;
6980 }
6981 }
6982 return true;
bellard19cb3732006-08-19 11:45:59 +00006983}
6984
6985/**
bellard19cb3732006-08-19 11:45:59 +00006986 * If eject_flag is TRUE, eject the media. Otherwise, close the tray
6987 */
Emanuele Giuseppe Esposito2531b392023-01-13 21:42:09 +01006988void coroutine_fn bdrv_co_eject(BlockDriverState *bs, bool eject_flag)
bellard19cb3732006-08-19 11:45:59 +00006989{
6990 BlockDriver *drv = bs->drv;
Emanuele Giuseppe Esposito384a48f2022-03-03 10:15:50 -05006991 IO_CODE();
Kevin Wolf79a292e2023-02-03 16:21:58 +01006992 assert_bdrv_graph_readable();
bellard19cb3732006-08-19 11:45:59 +00006993
Emanuele Giuseppe Esposito2531b392023-01-13 21:42:09 +01006994 if (drv && drv->bdrv_co_eject) {
6995 drv->bdrv_co_eject(bs, eject_flag);
bellard19cb3732006-08-19 11:45:59 +00006996 }
bellard19cb3732006-08-19 11:45:59 +00006997}
6998
bellard19cb3732006-08-19 11:45:59 +00006999/**
7000 * Lock or unlock the media (if it is locked, the user won't be able
7001 * to eject it manually).
7002 */
Emanuele Giuseppe Esposito2c752612023-01-13 21:42:10 +01007003void coroutine_fn bdrv_co_lock_medium(BlockDriverState *bs, bool locked)
bellard19cb3732006-08-19 11:45:59 +00007004{
7005 BlockDriver *drv = bs->drv;
Emanuele Giuseppe Esposito384a48f2022-03-03 10:15:50 -05007006 IO_CODE();
Kevin Wolf79a292e2023-02-03 16:21:58 +01007007 assert_bdrv_graph_readable();
Markus Armbruster025e8492011-09-06 18:58:47 +02007008 trace_bdrv_lock_medium(bs, locked);
Stefan Hajnoczib8c6d092011-03-29 20:04:40 +01007009
Emanuele Giuseppe Esposito2c752612023-01-13 21:42:10 +01007010 if (drv && drv->bdrv_co_lock_medium) {
7011 drv->bdrv_co_lock_medium(bs, locked);
bellard19cb3732006-08-19 11:45:59 +00007012 }
7013}
ths985a03b2007-12-24 16:10:43 +00007014
Fam Zheng9fcb0252013-08-23 09:14:46 +08007015/* Get a reference to bs */
7016void bdrv_ref(BlockDriverState *bs)
7017{
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05007018 GLOBAL_STATE_CODE();
Fam Zheng9fcb0252013-08-23 09:14:46 +08007019 bs->refcnt++;
7020}
7021
7022/* Release a previously grabbed reference to bs.
7023 * If after releasing, reference count is zero, the BlockDriverState is
7024 * deleted. */
7025void bdrv_unref(BlockDriverState *bs)
7026{
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05007027 GLOBAL_STATE_CODE();
Jeff Cody9a4d5ca2014-07-23 17:22:57 -04007028 if (!bs) {
7029 return;
7030 }
Fam Zheng9fcb0252013-08-23 09:14:46 +08007031 assert(bs->refcnt > 0);
7032 if (--bs->refcnt == 0) {
7033 bdrv_delete(bs);
7034 }
7035}
7036
Fam Zhengfbe40ff2014-05-23 21:29:42 +08007037struct BdrvOpBlocker {
7038 Error *reason;
7039 QLIST_ENTRY(BdrvOpBlocker) list;
7040};
7041
7042bool bdrv_op_is_blocked(BlockDriverState *bs, BlockOpType op, Error **errp)
7043{
7044 BdrvOpBlocker *blocker;
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05007045 GLOBAL_STATE_CODE();
Fam Zhengfbe40ff2014-05-23 21:29:42 +08007046 assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
7047 if (!QLIST_EMPTY(&bs->op_blockers[op])) {
7048 blocker = QLIST_FIRST(&bs->op_blockers[op]);
Markus Armbruster4b576642018-10-17 10:26:25 +02007049 error_propagate_prepend(errp, error_copy(blocker->reason),
7050 "Node '%s' is busy: ",
7051 bdrv_get_device_or_node_name(bs));
Fam Zhengfbe40ff2014-05-23 21:29:42 +08007052 return true;
7053 }
7054 return false;
7055}
7056
7057void bdrv_op_block(BlockDriverState *bs, BlockOpType op, Error *reason)
7058{
7059 BdrvOpBlocker *blocker;
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05007060 GLOBAL_STATE_CODE();
Fam Zhengfbe40ff2014-05-23 21:29:42 +08007061 assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
7062
Markus Armbruster5839e532014-08-19 10:31:08 +02007063 blocker = g_new0(BdrvOpBlocker, 1);
Fam Zhengfbe40ff2014-05-23 21:29:42 +08007064 blocker->reason = reason;
7065 QLIST_INSERT_HEAD(&bs->op_blockers[op], blocker, list);
7066}
7067
7068void bdrv_op_unblock(BlockDriverState *bs, BlockOpType op, Error *reason)
7069{
7070 BdrvOpBlocker *blocker, *next;
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05007071 GLOBAL_STATE_CODE();
Fam Zhengfbe40ff2014-05-23 21:29:42 +08007072 assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
7073 QLIST_FOREACH_SAFE(blocker, &bs->op_blockers[op], list, next) {
7074 if (blocker->reason == reason) {
7075 QLIST_REMOVE(blocker, list);
7076 g_free(blocker);
7077 }
7078 }
7079}
7080
7081void bdrv_op_block_all(BlockDriverState *bs, Error *reason)
7082{
7083 int i;
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05007084 GLOBAL_STATE_CODE();
Fam Zhengfbe40ff2014-05-23 21:29:42 +08007085 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
7086 bdrv_op_block(bs, i, reason);
7087 }
7088}
7089
7090void bdrv_op_unblock_all(BlockDriverState *bs, Error *reason)
7091{
7092 int i;
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05007093 GLOBAL_STATE_CODE();
Fam Zhengfbe40ff2014-05-23 21:29:42 +08007094 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
7095 bdrv_op_unblock(bs, i, reason);
7096 }
7097}
7098
7099bool bdrv_op_blocker_is_empty(BlockDriverState *bs)
7100{
7101 int i;
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05007102 GLOBAL_STATE_CODE();
Fam Zhengfbe40ff2014-05-23 21:29:42 +08007103 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
7104 if (!QLIST_EMPTY(&bs->op_blockers[i])) {
7105 return false;
7106 }
7107 }
7108 return true;
7109}
7110
Kevin Wolfe1355052022-12-07 14:18:30 +01007111/*
7112 * Must not be called while holding the lock of an AioContext other than the
7113 * current one.
7114 */
Luiz Capitulinod92ada22012-11-30 10:52:09 -02007115void bdrv_img_create(const char *filename, const char *fmt,
7116 const char *base_filename, const char *base_fmt,
Fam Zheng92172832017-04-21 20:27:01 +08007117 char *options, uint64_t img_size, int flags, bool quiet,
7118 Error **errp)
Jes Sorensenf88e1a42010-12-16 13:52:15 +01007119{
Chunyan Liu83d05212014-06-05 17:20:51 +08007120 QemuOptsList *create_opts = NULL;
7121 QemuOpts *opts = NULL;
7122 const char *backing_fmt, *backing_file;
7123 int64_t size;
Jes Sorensenf88e1a42010-12-16 13:52:15 +01007124 BlockDriver *drv, *proto_drv;
Max Reitzcc84d902013-09-06 17:14:26 +02007125 Error *local_err = NULL;
Jes Sorensenf88e1a42010-12-16 13:52:15 +01007126 int ret = 0;
7127
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05007128 GLOBAL_STATE_CODE();
7129
Jes Sorensenf88e1a42010-12-16 13:52:15 +01007130 /* Find driver and parse its options */
7131 drv = bdrv_find_format(fmt);
7132 if (!drv) {
Luiz Capitulino71c79812012-11-30 10:52:04 -02007133 error_setg(errp, "Unknown file format '%s'", fmt);
Luiz Capitulinod92ada22012-11-30 10:52:09 -02007134 return;
Jes Sorensenf88e1a42010-12-16 13:52:15 +01007135 }
7136
Max Reitzb65a5e12015-02-05 13:58:12 -05007137 proto_drv = bdrv_find_protocol(filename, true, errp);
Jes Sorensenf88e1a42010-12-16 13:52:15 +01007138 if (!proto_drv) {
Luiz Capitulinod92ada22012-11-30 10:52:09 -02007139 return;
Jes Sorensenf88e1a42010-12-16 13:52:15 +01007140 }
7141
Max Reitzc6149722014-12-02 18:32:45 +01007142 if (!drv->create_opts) {
7143 error_setg(errp, "Format driver '%s' does not support image creation",
7144 drv->format_name);
7145 return;
7146 }
7147
Maxim Levitsky5a5e7f82020-03-26 03:12:18 +02007148 if (!proto_drv->create_opts) {
7149 error_setg(errp, "Protocol driver '%s' does not support image creation",
7150 proto_drv->format_name);
7151 return;
7152 }
7153
Kevin Wolfc6e0a6d2023-05-25 14:47:04 +02007154 aio_context_acquire(qemu_get_aio_context());
7155
Kevin Wolff6dc1c32019-11-26 16:45:49 +01007156 /* Create parameter list */
Chunyan Liuc282e1f2014-06-05 17:21:11 +08007157 create_opts = qemu_opts_append(create_opts, drv->create_opts);
Maxim Levitsky5a5e7f82020-03-26 03:12:18 +02007158 create_opts = qemu_opts_append(create_opts, proto_drv->create_opts);
Jes Sorensenf88e1a42010-12-16 13:52:15 +01007159
Chunyan Liu83d05212014-06-05 17:20:51 +08007160 opts = qemu_opts_create(create_opts, NULL, 0, &error_abort);
Jes Sorensenf88e1a42010-12-16 13:52:15 +01007161
7162 /* Parse -o options */
7163 if (options) {
Markus Armbrustera5f9b9d2020-07-07 18:06:05 +02007164 if (!qemu_opts_do_parse(opts, options, NULL, errp)) {
Jes Sorensenf88e1a42010-12-16 13:52:15 +01007165 goto out;
7166 }
7167 }
7168
Kevin Wolff6dc1c32019-11-26 16:45:49 +01007169 if (!qemu_opt_get(opts, BLOCK_OPT_SIZE)) {
7170 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, img_size, &error_abort);
7171 } else if (img_size != UINT64_C(-1)) {
7172 error_setg(errp, "The image size must be specified only once");
7173 goto out;
7174 }
7175
Jes Sorensenf88e1a42010-12-16 13:52:15 +01007176 if (base_filename) {
Markus Armbruster235e59c2020-07-07 18:05:42 +02007177 if (!qemu_opt_set(opts, BLOCK_OPT_BACKING_FILE, base_filename,
Markus Armbruster38825782020-07-07 18:05:43 +02007178 NULL)) {
Luiz Capitulino71c79812012-11-30 10:52:04 -02007179 error_setg(errp, "Backing file not supported for file format '%s'",
7180 fmt);
Jes Sorensenf88e1a42010-12-16 13:52:15 +01007181 goto out;
7182 }
7183 }
7184
7185 if (base_fmt) {
Markus Armbruster38825782020-07-07 18:05:43 +02007186 if (!qemu_opt_set(opts, BLOCK_OPT_BACKING_FMT, base_fmt, NULL)) {
Luiz Capitulino71c79812012-11-30 10:52:04 -02007187 error_setg(errp, "Backing file format not supported for file "
7188 "format '%s'", fmt);
Jes Sorensenf88e1a42010-12-16 13:52:15 +01007189 goto out;
7190 }
7191 }
7192
Chunyan Liu83d05212014-06-05 17:20:51 +08007193 backing_file = qemu_opt_get(opts, BLOCK_OPT_BACKING_FILE);
7194 if (backing_file) {
7195 if (!strcmp(filename, backing_file)) {
Luiz Capitulino71c79812012-11-30 10:52:04 -02007196 error_setg(errp, "Error: Trying to create an image with the "
7197 "same filename as the backing file");
Jes Sorensen792da932010-12-16 13:52:17 +01007198 goto out;
7199 }
Connor Kuehl975a7bd2020-08-13 08:47:22 -05007200 if (backing_file[0] == '\0') {
7201 error_setg(errp, "Expected backing file name, got empty string");
7202 goto out;
7203 }
Jes Sorensen792da932010-12-16 13:52:17 +01007204 }
7205
Chunyan Liu83d05212014-06-05 17:20:51 +08007206 backing_fmt = qemu_opt_get(opts, BLOCK_OPT_BACKING_FMT);
Jes Sorensenf88e1a42010-12-16 13:52:15 +01007207
John Snow6e6e55f2017-07-17 20:34:22 -04007208 /* The size for the image must always be specified, unless we have a backing
7209 * file and we have not been forbidden from opening it. */
Eric Blakea8b42a12017-09-25 09:55:07 -05007210 size = qemu_opt_get_size(opts, BLOCK_OPT_SIZE, img_size);
John Snow6e6e55f2017-07-17 20:34:22 -04007211 if (backing_file && !(flags & BDRV_O_NO_BACKING)) {
7212 BlockDriverState *bs;
Max Reitz645ae7d2019-02-01 20:29:14 +01007213 char *full_backing;
John Snow6e6e55f2017-07-17 20:34:22 -04007214 int back_flags;
7215 QDict *backing_options = NULL;
Paolo Bonzini63090da2012-04-12 14:01:03 +02007216
Max Reitz645ae7d2019-02-01 20:29:14 +01007217 full_backing =
7218 bdrv_get_full_backing_filename_from_filename(filename, backing_file,
7219 &local_err);
John Snow6e6e55f2017-07-17 20:34:22 -04007220 if (local_err) {
Jes Sorensenf88e1a42010-12-16 13:52:15 +01007221 goto out;
7222 }
Max Reitz645ae7d2019-02-01 20:29:14 +01007223 assert(full_backing);
John Snow6e6e55f2017-07-17 20:34:22 -04007224
Max Reitzd5b23992021-06-22 16:00:30 +02007225 /*
7226 * No need to do I/O here, which allows us to open encrypted
7227 * backing images without needing the secret
7228 */
John Snow6e6e55f2017-07-17 20:34:22 -04007229 back_flags = flags;
7230 back_flags &= ~(BDRV_O_RDWR | BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING);
Max Reitzd5b23992021-06-22 16:00:30 +02007231 back_flags |= BDRV_O_NO_IO;
John Snow6e6e55f2017-07-17 20:34:22 -04007232
Fam Zhengcc954f02017-12-15 16:04:45 +08007233 backing_options = qdict_new();
John Snow6e6e55f2017-07-17 20:34:22 -04007234 if (backing_fmt) {
John Snow6e6e55f2017-07-17 20:34:22 -04007235 qdict_put_str(backing_options, "driver", backing_fmt);
7236 }
Fam Zhengcc954f02017-12-15 16:04:45 +08007237 qdict_put_bool(backing_options, BDRV_OPT_FORCE_SHARE, true);
John Snow6e6e55f2017-07-17 20:34:22 -04007238
7239 bs = bdrv_open(full_backing, NULL, backing_options, back_flags,
7240 &local_err);
7241 g_free(full_backing);
Eric Blakeadd82002020-07-06 15:39:50 -05007242 if (!bs) {
7243 error_append_hint(&local_err, "Could not open backing image.\n");
John Snow6e6e55f2017-07-17 20:34:22 -04007244 goto out;
7245 } else {
Eric Blaked9f059a2020-07-06 15:39:54 -05007246 if (!backing_fmt) {
Eric Blake497a30d2021-05-03 14:36:00 -07007247 error_setg(&local_err,
7248 "Backing file specified without backing format");
Michael Tokarevfbdffb02023-04-05 16:34:04 +03007249 error_append_hint(&local_err, "Detected format of %s.\n",
Eric Blake497a30d2021-05-03 14:36:00 -07007250 bs->drv->format_name);
7251 goto out;
Eric Blaked9f059a2020-07-06 15:39:54 -05007252 }
John Snow6e6e55f2017-07-17 20:34:22 -04007253 if (size == -1) {
7254 /* Opened BS, have no size */
7255 size = bdrv_getlength(bs);
7256 if (size < 0) {
7257 error_setg_errno(errp, -size, "Could not get size of '%s'",
7258 backing_file);
7259 bdrv_unref(bs);
7260 goto out;
7261 }
7262 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, size, &error_abort);
7263 }
7264 bdrv_unref(bs);
7265 }
Eric Blaked9f059a2020-07-06 15:39:54 -05007266 /* (backing_file && !(flags & BDRV_O_NO_BACKING)) */
7267 } else if (backing_file && !backing_fmt) {
Eric Blake497a30d2021-05-03 14:36:00 -07007268 error_setg(&local_err,
7269 "Backing file specified without backing format");
7270 goto out;
Eric Blaked9f059a2020-07-06 15:39:54 -05007271 }
John Snow6e6e55f2017-07-17 20:34:22 -04007272
7273 if (size == -1) {
7274 error_setg(errp, "Image creation needs a size parameter");
7275 goto out;
Jes Sorensenf88e1a42010-12-16 13:52:15 +01007276 }
7277
Miroslav Rezaninaf382d432013-02-13 09:09:40 +01007278 if (!quiet) {
Kővágó, Zoltánfe646692015-07-07 16:42:10 +02007279 printf("Formatting '%s', fmt=%s ", filename, fmt);
Fam Zheng43c5d8f2014-12-09 15:38:04 +08007280 qemu_opts_print(opts, " ");
Miroslav Rezaninaf382d432013-02-13 09:09:40 +01007281 puts("");
Eric Blake4e2f4412020-07-06 15:39:45 -05007282 fflush(stdout);
Miroslav Rezaninaf382d432013-02-13 09:09:40 +01007283 }
Chunyan Liu83d05212014-06-05 17:20:51 +08007284
Chunyan Liuc282e1f2014-06-05 17:21:11 +08007285 ret = bdrv_create(drv, filename, opts, &local_err);
Chunyan Liu83d05212014-06-05 17:20:51 +08007286
Max Reitzcc84d902013-09-06 17:14:26 +02007287 if (ret == -EFBIG) {
7288 /* This is generally a better message than whatever the driver would
7289 * deliver (especially because of the cluster_size_hint), since that
7290 * is most probably not much different from "image too large". */
7291 const char *cluster_size_hint = "";
Chunyan Liu83d05212014-06-05 17:20:51 +08007292 if (qemu_opt_get_size(opts, BLOCK_OPT_CLUSTER_SIZE, 0)) {
Max Reitzcc84d902013-09-06 17:14:26 +02007293 cluster_size_hint = " (try using a larger cluster size)";
Jes Sorensenf88e1a42010-12-16 13:52:15 +01007294 }
Max Reitzcc84d902013-09-06 17:14:26 +02007295 error_setg(errp, "The image size is too large for file format '%s'"
7296 "%s", fmt, cluster_size_hint);
7297 error_free(local_err);
7298 local_err = NULL;
Jes Sorensenf88e1a42010-12-16 13:52:15 +01007299 }
7300
7301out:
Chunyan Liu83d05212014-06-05 17:20:51 +08007302 qemu_opts_del(opts);
7303 qemu_opts_free(create_opts);
Eduardo Habkost621ff942016-06-13 18:57:56 -03007304 error_propagate(errp, local_err);
Kevin Wolfc6e0a6d2023-05-25 14:47:04 +02007305 aio_context_release(qemu_get_aio_context());
Jes Sorensenf88e1a42010-12-16 13:52:15 +01007306}
Stefan Hajnoczi85d126f2013-03-07 13:41:48 +01007307
7308AioContext *bdrv_get_aio_context(BlockDriverState *bs)
7309{
Emanuele Giuseppe Esposito384a48f2022-03-03 10:15:50 -05007310 IO_CODE();
Stefan Hajnoczi33f2a752018-02-16 16:50:13 +00007311 return bs ? bs->aio_context : qemu_get_aio_context();
Stefan Hajnoczidcd04222014-05-08 16:34:37 +02007312}
7313
Kevin Wolfe336fd42020-10-05 17:58:53 +02007314AioContext *coroutine_fn bdrv_co_enter(BlockDriverState *bs)
7315{
7316 Coroutine *self = qemu_coroutine_self();
7317 AioContext *old_ctx = qemu_coroutine_get_aio_context(self);
7318 AioContext *new_ctx;
Emanuele Giuseppe Esposito384a48f2022-03-03 10:15:50 -05007319 IO_CODE();
Kevin Wolfe336fd42020-10-05 17:58:53 +02007320
7321 /*
7322 * Increase bs->in_flight to ensure that this operation is completed before
7323 * moving the node to a different AioContext. Read new_ctx only afterwards.
7324 */
7325 bdrv_inc_in_flight(bs);
7326
7327 new_ctx = bdrv_get_aio_context(bs);
7328 aio_co_reschedule_self(new_ctx);
7329 return old_ctx;
7330}
7331
7332void coroutine_fn bdrv_co_leave(BlockDriverState *bs, AioContext *old_ctx)
7333{
Emanuele Giuseppe Esposito384a48f2022-03-03 10:15:50 -05007334 IO_CODE();
Kevin Wolfe336fd42020-10-05 17:58:53 +02007335 aio_co_reschedule_self(old_ctx);
7336 bdrv_dec_in_flight(bs);
7337}
7338
Kevin Wolf18c6ac12020-10-05 17:58:54 +02007339void coroutine_fn bdrv_co_lock(BlockDriverState *bs)
7340{
7341 AioContext *ctx = bdrv_get_aio_context(bs);
7342
7343 /* In the main thread, bs->aio_context won't change concurrently */
7344 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
7345
7346 /*
7347 * We're in coroutine context, so we already hold the lock of the main
7348 * loop AioContext. Don't lock it twice to avoid deadlocks.
7349 */
7350 assert(qemu_in_coroutine());
7351 if (ctx != qemu_get_aio_context()) {
7352 aio_context_acquire(ctx);
7353 }
7354}
7355
7356void coroutine_fn bdrv_co_unlock(BlockDriverState *bs)
7357{
7358 AioContext *ctx = bdrv_get_aio_context(bs);
7359
7360 assert(qemu_in_coroutine());
7361 if (ctx != qemu_get_aio_context()) {
7362 aio_context_release(ctx);
7363 }
7364}
7365
Stefan Hajnoczie8a095d2016-06-16 17:56:26 +01007366static void bdrv_do_remove_aio_context_notifier(BdrvAioNotifier *ban)
7367{
Emanuele Giuseppe Espositobdb73472022-03-03 10:16:02 -05007368 GLOBAL_STATE_CODE();
Stefan Hajnoczie8a095d2016-06-16 17:56:26 +01007369 QLIST_REMOVE(ban, list);
7370 g_free(ban);
7371}
7372
Kevin Wolfa3a683c2019-05-06 19:17:57 +02007373static void bdrv_detach_aio_context(BlockDriverState *bs)
Stefan Hajnoczidcd04222014-05-08 16:34:37 +02007374{
Stefan Hajnoczie8a095d2016-06-16 17:56:26 +01007375 BdrvAioNotifier *baf, *baf_tmp;
Max Reitz33384422014-06-20 21:57:33 +02007376
Stefan Hajnoczie8a095d2016-06-16 17:56:26 +01007377 assert(!bs->walking_aio_notifiers);
Emanuele Giuseppe Espositoda359902022-03-03 10:16:11 -05007378 GLOBAL_STATE_CODE();
Stefan Hajnoczie8a095d2016-06-16 17:56:26 +01007379 bs->walking_aio_notifiers = true;
7380 QLIST_FOREACH_SAFE(baf, &bs->aio_notifiers, list, baf_tmp) {
7381 if (baf->deleted) {
7382 bdrv_do_remove_aio_context_notifier(baf);
7383 } else {
7384 baf->detach_aio_context(baf->opaque);
7385 }
Max Reitz33384422014-06-20 21:57:33 +02007386 }
Stefan Hajnoczie8a095d2016-06-16 17:56:26 +01007387 /* Never mind iterating again to check for ->deleted. bdrv_close() will
7388 * remove remaining aio notifiers if we aren't called again.
7389 */
7390 bs->walking_aio_notifiers = false;
Max Reitz33384422014-06-20 21:57:33 +02007391
Kevin Wolf1bffe1a2019-04-17 17:15:25 +02007392 if (bs->drv && bs->drv->bdrv_detach_aio_context) {
Stefan Hajnoczidcd04222014-05-08 16:34:37 +02007393 bs->drv->bdrv_detach_aio_context(bs);
7394 }
Stefan Hajnoczidcd04222014-05-08 16:34:37 +02007395
7396 bs->aio_context = NULL;
7397}
7398
Kevin Wolfa3a683c2019-05-06 19:17:57 +02007399static void bdrv_attach_aio_context(BlockDriverState *bs,
7400 AioContext *new_context)
Stefan Hajnoczidcd04222014-05-08 16:34:37 +02007401{
Stefan Hajnoczie8a095d2016-06-16 17:56:26 +01007402 BdrvAioNotifier *ban, *ban_tmp;
Emanuele Giuseppe Espositoda359902022-03-03 10:16:11 -05007403 GLOBAL_STATE_CODE();
Max Reitz33384422014-06-20 21:57:33 +02007404
Stefan Hajnoczidcd04222014-05-08 16:34:37 +02007405 bs->aio_context = new_context;
7406
Kevin Wolf1bffe1a2019-04-17 17:15:25 +02007407 if (bs->drv && bs->drv->bdrv_attach_aio_context) {
Stefan Hajnoczidcd04222014-05-08 16:34:37 +02007408 bs->drv->bdrv_attach_aio_context(bs, new_context);
7409 }
Max Reitz33384422014-06-20 21:57:33 +02007410
Stefan Hajnoczie8a095d2016-06-16 17:56:26 +01007411 assert(!bs->walking_aio_notifiers);
7412 bs->walking_aio_notifiers = true;
7413 QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_tmp) {
7414 if (ban->deleted) {
7415 bdrv_do_remove_aio_context_notifier(ban);
7416 } else {
7417 ban->attached_aio_context(new_context, ban->opaque);
7418 }
Max Reitz33384422014-06-20 21:57:33 +02007419 }
Stefan Hajnoczie8a095d2016-06-16 17:56:26 +01007420 bs->walking_aio_notifiers = false;
Stefan Hajnoczidcd04222014-05-08 16:34:37 +02007421}
7422
Emanuele Giuseppe Esposito7e8c1822022-10-25 04:49:44 -04007423typedef struct BdrvStateSetAioContext {
7424 AioContext *new_ctx;
7425 BlockDriverState *bs;
7426} BdrvStateSetAioContext;
7427
7428static bool bdrv_parent_change_aio_context(BdrvChild *c, AioContext *ctx,
Emanuele Giuseppe Espositoe08cc002022-10-25 04:49:45 -04007429 GHashTable *visited,
7430 Transaction *tran,
Emanuele Giuseppe Esposito7e8c1822022-10-25 04:49:44 -04007431 Error **errp)
7432{
7433 GLOBAL_STATE_CODE();
Emanuele Giuseppe Espositoe08cc002022-10-25 04:49:45 -04007434 if (g_hash_table_contains(visited, c)) {
Emanuele Giuseppe Esposito7e8c1822022-10-25 04:49:44 -04007435 return true;
7436 }
Emanuele Giuseppe Espositoe08cc002022-10-25 04:49:45 -04007437 g_hash_table_add(visited, c);
Emanuele Giuseppe Esposito7e8c1822022-10-25 04:49:44 -04007438
7439 /*
7440 * A BdrvChildClass that doesn't handle AioContext changes cannot
7441 * tolerate any AioContext changes
7442 */
7443 if (!c->klass->change_aio_ctx) {
7444 char *user = bdrv_child_user_desc(c);
7445 error_setg(errp, "Changing iothreads is not supported by %s", user);
7446 g_free(user);
7447 return false;
7448 }
7449 if (!c->klass->change_aio_ctx(c, ctx, visited, tran, errp)) {
7450 assert(!errp || *errp);
7451 return false;
7452 }
7453 return true;
7454}
7455
Emanuele Giuseppe Esposito7e8c1822022-10-25 04:49:44 -04007456bool bdrv_child_change_aio_context(BdrvChild *c, AioContext *ctx,
Emanuele Giuseppe Espositoe08cc002022-10-25 04:49:45 -04007457 GHashTable *visited, Transaction *tran,
Emanuele Giuseppe Esposito7e8c1822022-10-25 04:49:44 -04007458 Error **errp)
7459{
7460 GLOBAL_STATE_CODE();
Emanuele Giuseppe Espositoe08cc002022-10-25 04:49:45 -04007461 if (g_hash_table_contains(visited, c)) {
Emanuele Giuseppe Esposito7e8c1822022-10-25 04:49:44 -04007462 return true;
7463 }
Emanuele Giuseppe Espositoe08cc002022-10-25 04:49:45 -04007464 g_hash_table_add(visited, c);
Emanuele Giuseppe Esposito7e8c1822022-10-25 04:49:44 -04007465 return bdrv_change_aio_context(c->bs, ctx, visited, tran, errp);
7466}
7467
Emanuele Giuseppe Esposito7e8c1822022-10-25 04:49:44 -04007468static void bdrv_set_aio_context_clean(void *opaque)
7469{
7470 BdrvStateSetAioContext *state = (BdrvStateSetAioContext *) opaque;
7471 BlockDriverState *bs = (BlockDriverState *) state->bs;
7472
7473 /* Paired with bdrv_drained_begin in bdrv_change_aio_context() */
7474 bdrv_drained_end(bs);
7475
7476 g_free(state);
7477}
7478
7479static void bdrv_set_aio_context_commit(void *opaque)
7480{
7481 BdrvStateSetAioContext *state = (BdrvStateSetAioContext *) opaque;
7482 BlockDriverState *bs = (BlockDriverState *) state->bs;
7483 AioContext *new_context = state->new_ctx;
7484 AioContext *old_context = bdrv_get_aio_context(bs);
Emanuele Giuseppe Esposito7e8c1822022-10-25 04:49:44 -04007485
7486 /*
7487 * Take the old AioContex when detaching it from bs.
7488 * At this point, new_context lock is already acquired, and we are now
7489 * also taking old_context. This is safe as long as bdrv_detach_aio_context
7490 * does not call AIO_POLL_WHILE().
7491 */
7492 if (old_context != qemu_get_aio_context()) {
7493 aio_context_acquire(old_context);
7494 }
7495 bdrv_detach_aio_context(bs);
7496 if (old_context != qemu_get_aio_context()) {
7497 aio_context_release(old_context);
7498 }
7499 bdrv_attach_aio_context(bs, new_context);
7500}
7501
7502static TransactionActionDrv set_aio_context = {
7503 .commit = bdrv_set_aio_context_commit,
7504 .clean = bdrv_set_aio_context_clean,
7505};
7506
Kevin Wolf42a65f02019-05-07 18:31:38 +02007507/*
7508 * Changes the AioContext used for fd handlers, timers, and BHs by this
7509 * BlockDriverState and all its children and parents.
7510 *
Max Reitz43eaaae2019-07-22 15:30:54 +02007511 * Must be called from the main AioContext.
7512 *
Kevin Wolf42a65f02019-05-07 18:31:38 +02007513 * The caller must own the AioContext lock for the old AioContext of bs, but it
7514 * must not own the AioContext lock for new_context (unless new_context is the
7515 * same as the current context of bs).
7516 *
Emanuele Giuseppe Esposito7e8c1822022-10-25 04:49:44 -04007517 * @visited will accumulate all visited BdrvChild objects. The caller is
Kevin Wolf42a65f02019-05-07 18:31:38 +02007518 * responsible for freeing the list afterwards.
7519 */
Emanuele Giuseppe Esposito7e8c1822022-10-25 04:49:44 -04007520static bool bdrv_change_aio_context(BlockDriverState *bs, AioContext *ctx,
Emanuele Giuseppe Espositoe08cc002022-10-25 04:49:45 -04007521 GHashTable *visited, Transaction *tran,
Emanuele Giuseppe Esposito7e8c1822022-10-25 04:49:44 -04007522 Error **errp)
Kevin Wolf5d231842019-05-06 19:17:56 +02007523{
7524 BdrvChild *c;
Emanuele Giuseppe Esposito7e8c1822022-10-25 04:49:44 -04007525 BdrvStateSetAioContext *state;
7526
7527 GLOBAL_STATE_CODE();
Kevin Wolf5d231842019-05-06 19:17:56 +02007528
7529 if (bdrv_get_aio_context(bs) == ctx) {
7530 return true;
7531 }
7532
7533 QLIST_FOREACH(c, &bs->parents, next_parent) {
Emanuele Giuseppe Esposito7e8c1822022-10-25 04:49:44 -04007534 if (!bdrv_parent_change_aio_context(c, ctx, visited, tran, errp)) {
Kevin Wolf5d231842019-05-06 19:17:56 +02007535 return false;
7536 }
7537 }
Emanuele Giuseppe Esposito7e8c1822022-10-25 04:49:44 -04007538
Kevin Wolf5d231842019-05-06 19:17:56 +02007539 QLIST_FOREACH(c, &bs->children, next) {
Emanuele Giuseppe Esposito7e8c1822022-10-25 04:49:44 -04007540 if (!bdrv_child_change_aio_context(c, ctx, visited, tran, errp)) {
Kevin Wolf5d231842019-05-06 19:17:56 +02007541 return false;
7542 }
7543 }
7544
Emanuele Giuseppe Esposito7e8c1822022-10-25 04:49:44 -04007545 state = g_new(BdrvStateSetAioContext, 1);
7546 *state = (BdrvStateSetAioContext) {
7547 .new_ctx = ctx,
7548 .bs = bs,
7549 };
7550
7551 /* Paired with bdrv_drained_end in bdrv_set_aio_context_clean() */
7552 bdrv_drained_begin(bs);
7553
7554 tran_add(tran, &set_aio_context, state);
7555
Kevin Wolf5d231842019-05-06 19:17:56 +02007556 return true;
7557}
7558
Emanuele Giuseppe Esposito7e8c1822022-10-25 04:49:44 -04007559/*
7560 * Change bs's and recursively all of its parents' and children's AioContext
7561 * to the given new context, returning an error if that isn't possible.
7562 *
7563 * If ignore_child is not NULL, that child (and its subgraph) will not
7564 * be touched.
7565 *
7566 * This function still requires the caller to take the bs current
7567 * AioContext lock, otherwise draining will fail since AIO_WAIT_WHILE
7568 * assumes the lock is always held if bs is in another AioContext.
7569 * For the same reason, it temporarily also holds the new AioContext, since
7570 * bdrv_drained_end calls BDRV_POLL_WHILE that assumes the lock is taken too.
7571 * Therefore the new AioContext lock must not be taken by the caller.
7572 */
Emanuele Giuseppe Espositoa41cfda2022-10-25 04:49:51 -04007573int bdrv_try_change_aio_context(BlockDriverState *bs, AioContext *ctx,
7574 BdrvChild *ignore_child, Error **errp)
Kevin Wolf5d231842019-05-06 19:17:56 +02007575{
Emanuele Giuseppe Esposito7e8c1822022-10-25 04:49:44 -04007576 Transaction *tran;
Emanuele Giuseppe Espositoe08cc002022-10-25 04:49:45 -04007577 GHashTable *visited;
Emanuele Giuseppe Esposito7e8c1822022-10-25 04:49:44 -04007578 int ret;
7579 AioContext *old_context = bdrv_get_aio_context(bs);
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05007580 GLOBAL_STATE_CODE();
7581
Emanuele Giuseppe Esposito7e8c1822022-10-25 04:49:44 -04007582 /*
7583 * Recursion phase: go through all nodes of the graph.
7584 * Take care of checking that all nodes support changing AioContext
7585 * and drain them, builing a linear list of callbacks to run if everything
7586 * is successful (the transaction itself).
7587 */
7588 tran = tran_new();
Emanuele Giuseppe Espositoe08cc002022-10-25 04:49:45 -04007589 visited = g_hash_table_new(NULL, NULL);
7590 if (ignore_child) {
7591 g_hash_table_add(visited, ignore_child);
7592 }
7593 ret = bdrv_change_aio_context(bs, ctx, visited, tran, errp);
7594 g_hash_table_destroy(visited);
Emanuele Giuseppe Esposito7e8c1822022-10-25 04:49:44 -04007595
7596 /*
7597 * Linear phase: go through all callbacks collected in the transaction.
7598 * Run all callbacks collected in the recursion to switch all nodes
7599 * AioContext lock (transaction commit), or undo all changes done in the
7600 * recursion (transaction abort).
7601 */
Kevin Wolf5d231842019-05-06 19:17:56 +02007602
7603 if (!ret) {
Emanuele Giuseppe Esposito7e8c1822022-10-25 04:49:44 -04007604 /* Just run clean() callbacks. No AioContext changed. */
7605 tran_abort(tran);
Kevin Wolf5d231842019-05-06 19:17:56 +02007606 return -EPERM;
7607 }
7608
Emanuele Giuseppe Esposito7e8c1822022-10-25 04:49:44 -04007609 /*
7610 * Release old AioContext, it won't be needed anymore, as all
7611 * bdrv_drained_begin() have been called already.
7612 */
7613 if (qemu_get_aio_context() != old_context) {
7614 aio_context_release(old_context);
7615 }
7616
7617 /*
7618 * Acquire new AioContext since bdrv_drained_end() is going to be called
7619 * after we switched all nodes in the new AioContext, and the function
7620 * assumes that the lock of the bs is always taken.
7621 */
7622 if (qemu_get_aio_context() != ctx) {
7623 aio_context_acquire(ctx);
7624 }
7625
7626 tran_commit(tran);
7627
7628 if (qemu_get_aio_context() != ctx) {
7629 aio_context_release(ctx);
7630 }
7631
7632 /* Re-acquire the old AioContext, since the caller takes and releases it. */
7633 if (qemu_get_aio_context() != old_context) {
7634 aio_context_acquire(old_context);
7635 }
Kevin Wolf53a7d042019-05-06 19:17:59 +02007636
Kevin Wolf5d231842019-05-06 19:17:56 +02007637 return 0;
7638}
7639
Max Reitz33384422014-06-20 21:57:33 +02007640void bdrv_add_aio_context_notifier(BlockDriverState *bs,
7641 void (*attached_aio_context)(AioContext *new_context, void *opaque),
7642 void (*detach_aio_context)(void *opaque), void *opaque)
7643{
7644 BdrvAioNotifier *ban = g_new(BdrvAioNotifier, 1);
7645 *ban = (BdrvAioNotifier){
7646 .attached_aio_context = attached_aio_context,
7647 .detach_aio_context = detach_aio_context,
7648 .opaque = opaque
7649 };
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05007650 GLOBAL_STATE_CODE();
Max Reitz33384422014-06-20 21:57:33 +02007651
7652 QLIST_INSERT_HEAD(&bs->aio_notifiers, ban, list);
7653}
7654
7655void bdrv_remove_aio_context_notifier(BlockDriverState *bs,
7656 void (*attached_aio_context)(AioContext *,
7657 void *),
7658 void (*detach_aio_context)(void *),
7659 void *opaque)
7660{
7661 BdrvAioNotifier *ban, *ban_next;
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05007662 GLOBAL_STATE_CODE();
Max Reitz33384422014-06-20 21:57:33 +02007663
7664 QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_next) {
7665 if (ban->attached_aio_context == attached_aio_context &&
7666 ban->detach_aio_context == detach_aio_context &&
Stefan Hajnoczie8a095d2016-06-16 17:56:26 +01007667 ban->opaque == opaque &&
7668 ban->deleted == false)
Max Reitz33384422014-06-20 21:57:33 +02007669 {
Stefan Hajnoczie8a095d2016-06-16 17:56:26 +01007670 if (bs->walking_aio_notifiers) {
7671 ban->deleted = true;
7672 } else {
7673 bdrv_do_remove_aio_context_notifier(ban);
7674 }
Max Reitz33384422014-06-20 21:57:33 +02007675 return;
7676 }
7677 }
7678
7679 abort();
7680}
7681
Max Reitz77485432014-10-27 11:12:50 +01007682int bdrv_amend_options(BlockDriverState *bs, QemuOpts *opts,
Max Reitzd1402b52018-05-09 23:00:18 +02007683 BlockDriverAmendStatusCB *status_cb, void *cb_opaque,
Maxim Levitskya3579bf2020-06-25 14:55:38 +02007684 bool force,
Max Reitzd1402b52018-05-09 23:00:18 +02007685 Error **errp)
Max Reitz6f176b42013-09-03 10:09:50 +02007686{
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05007687 GLOBAL_STATE_CODE();
Max Reitzd470ad42017-11-10 21:31:09 +01007688 if (!bs->drv) {
Max Reitzd1402b52018-05-09 23:00:18 +02007689 error_setg(errp, "Node is ejected");
Max Reitzd470ad42017-11-10 21:31:09 +01007690 return -ENOMEDIUM;
7691 }
Chunyan Liuc282e1f2014-06-05 17:21:11 +08007692 if (!bs->drv->bdrv_amend_options) {
Max Reitzd1402b52018-05-09 23:00:18 +02007693 error_setg(errp, "Block driver '%s' does not support option amendment",
7694 bs->drv->format_name);
Max Reitz6f176b42013-09-03 10:09:50 +02007695 return -ENOTSUP;
7696 }
Maxim Levitskya3579bf2020-06-25 14:55:38 +02007697 return bs->drv->bdrv_amend_options(bs, opts, status_cb,
7698 cb_opaque, force, errp);
Max Reitz6f176b42013-09-03 10:09:50 +02007699}
Benoît Canetf6186f42013-10-02 14:33:48 +02007700
Max Reitz5d69b5a2020-02-18 11:34:41 +01007701/*
7702 * This function checks whether the given @to_replace is allowed to be
7703 * replaced by a node that always shows the same data as @bs. This is
7704 * used for example to verify whether the mirror job can replace
7705 * @to_replace by the target mirrored from @bs.
7706 * To be replaceable, @bs and @to_replace may either be guaranteed to
7707 * always show the same data (because they are only connected through
7708 * filters), or some driver may allow replacing one of its children
7709 * because it can guarantee that this child's data is not visible at
7710 * all (for example, for dissenting quorum children that have no other
7711 * parents).
7712 */
7713bool bdrv_recurse_can_replace(BlockDriverState *bs,
7714 BlockDriverState *to_replace)
7715{
Max Reitz93393e62019-06-12 17:03:38 +02007716 BlockDriverState *filtered;
7717
Emanuele Giuseppe Espositob4ad82a2022-03-03 10:15:57 -05007718 GLOBAL_STATE_CODE();
7719
Max Reitz5d69b5a2020-02-18 11:34:41 +01007720 if (!bs || !bs->drv) {
7721 return false;
7722 }
7723
7724 if (bs == to_replace) {
7725 return true;
7726 }
7727
7728 /* See what the driver can do */
7729 if (bs->drv->bdrv_recurse_can_replace) {
7730 return bs->drv->bdrv_recurse_can_replace(bs, to_replace);
7731 }
7732
7733 /* For filters without an own implementation, we can recurse on our own */
Max Reitz93393e62019-06-12 17:03:38 +02007734 filtered = bdrv_filter_bs(bs);
7735 if (filtered) {
7736 return bdrv_recurse_can_replace(filtered, to_replace);
Max Reitz5d69b5a2020-02-18 11:34:41 +01007737 }
7738
7739 /* Safe default */
7740 return false;
7741}
7742
Max Reitz810803a2020-02-18 11:34:44 +01007743/*
7744 * Check whether the given @node_name can be replaced by a node that
7745 * has the same data as @parent_bs. If so, return @node_name's BDS;
7746 * NULL otherwise.
7747 *
7748 * @node_name must be a (recursive) *child of @parent_bs (or this
7749 * function will return NULL).
7750 *
7751 * The result (whether the node can be replaced or not) is only valid
7752 * for as long as no graph or permission changes occur.
7753 */
Wen Congyange12f3782015-07-17 10:12:22 +08007754BlockDriverState *check_to_replace_node(BlockDriverState *parent_bs,
7755 const char *node_name, Error **errp)
Benoît Canet09158f02014-06-27 18:25:25 +02007756{
7757 BlockDriverState *to_replace_bs = bdrv_find_node(node_name);
Stefan Hajnoczi5a7e7a02014-10-21 12:03:58 +01007758 AioContext *aio_context;
7759
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05007760 GLOBAL_STATE_CODE();
7761
Benoît Canet09158f02014-06-27 18:25:25 +02007762 if (!to_replace_bs) {
Connor Kuehl785ec4b2021-03-05 09:19:28 -06007763 error_setg(errp, "Failed to find node with node-name='%s'", node_name);
Benoît Canet09158f02014-06-27 18:25:25 +02007764 return NULL;
7765 }
7766
Stefan Hajnoczi5a7e7a02014-10-21 12:03:58 +01007767 aio_context = bdrv_get_aio_context(to_replace_bs);
7768 aio_context_acquire(aio_context);
7769
Benoît Canet09158f02014-06-27 18:25:25 +02007770 if (bdrv_op_is_blocked(to_replace_bs, BLOCK_OP_TYPE_REPLACE, errp)) {
Stefan Hajnoczi5a7e7a02014-10-21 12:03:58 +01007771 to_replace_bs = NULL;
7772 goto out;
Benoît Canet09158f02014-06-27 18:25:25 +02007773 }
7774
7775 /* We don't want arbitrary node of the BDS chain to be replaced only the top
7776 * most non filter in order to prevent data corruption.
7777 * Another benefit is that this tests exclude backing files which are
7778 * blocked by the backing blockers.
7779 */
Max Reitz810803a2020-02-18 11:34:44 +01007780 if (!bdrv_recurse_can_replace(parent_bs, to_replace_bs)) {
7781 error_setg(errp, "Cannot replace '%s' by a node mirrored from '%s', "
7782 "because it cannot be guaranteed that doing so would not "
7783 "lead to an abrupt change of visible data",
7784 node_name, parent_bs->node_name);
Stefan Hajnoczi5a7e7a02014-10-21 12:03:58 +01007785 to_replace_bs = NULL;
7786 goto out;
Benoît Canet09158f02014-06-27 18:25:25 +02007787 }
7788
Stefan Hajnoczi5a7e7a02014-10-21 12:03:58 +01007789out:
7790 aio_context_release(aio_context);
Benoît Canet09158f02014-06-27 18:25:25 +02007791 return to_replace_bs;
7792}
Ming Lei448ad912014-07-04 18:04:33 +08007793
Max Reitz97e2f022019-02-01 20:29:27 +01007794/**
7795 * Iterates through the list of runtime option keys that are said to
7796 * be "strong" for a BDS. An option is called "strong" if it changes
7797 * a BDS's data. For example, the null block driver's "size" and
7798 * "read-zeroes" options are strong, but its "latency-ns" option is
7799 * not.
7800 *
7801 * If a key returned by this function ends with a dot, all options
7802 * starting with that prefix are strong.
7803 */
7804static const char *const *strong_options(BlockDriverState *bs,
7805 const char *const *curopt)
7806{
7807 static const char *const global_options[] = {
7808 "driver", "filename", NULL
7809 };
7810
7811 if (!curopt) {
7812 return &global_options[0];
7813 }
7814
7815 curopt++;
7816 if (curopt == &global_options[ARRAY_SIZE(global_options) - 1] && bs->drv) {
7817 curopt = bs->drv->strong_runtime_opts;
7818 }
7819
7820 return (curopt && *curopt) ? curopt : NULL;
7821}
7822
7823/**
7824 * Copies all strong runtime options from bs->options to the given
7825 * QDict. The set of strong option keys is determined by invoking
7826 * strong_options().
7827 *
7828 * Returns true iff any strong option was present in bs->options (and
7829 * thus copied to the target QDict) with the exception of "filename"
7830 * and "driver". The caller is expected to use this value to decide
7831 * whether the existence of strong options prevents the generation of
7832 * a plain filename.
7833 */
7834static bool append_strong_runtime_options(QDict *d, BlockDriverState *bs)
7835{
7836 bool found_any = false;
7837 const char *const *option_name = NULL;
7838
7839 if (!bs->drv) {
7840 return false;
7841 }
7842
7843 while ((option_name = strong_options(bs, option_name))) {
7844 bool option_given = false;
7845
7846 assert(strlen(*option_name) > 0);
7847 if ((*option_name)[strlen(*option_name) - 1] != '.') {
7848 QObject *entry = qdict_get(bs->options, *option_name);
7849 if (!entry) {
7850 continue;
7851 }
7852
7853 qdict_put_obj(d, *option_name, qobject_ref(entry));
7854 option_given = true;
7855 } else {
7856 const QDictEntry *entry;
7857 for (entry = qdict_first(bs->options); entry;
7858 entry = qdict_next(bs->options, entry))
7859 {
7860 if (strstart(qdict_entry_key(entry), *option_name, NULL)) {
7861 qdict_put_obj(d, qdict_entry_key(entry),
7862 qobject_ref(qdict_entry_value(entry)));
7863 option_given = true;
7864 }
7865 }
7866 }
7867
7868 /* While "driver" and "filename" need to be included in a JSON filename,
7869 * their existence does not prohibit generation of a plain filename. */
7870 if (!found_any && option_given &&
7871 strcmp(*option_name, "driver") && strcmp(*option_name, "filename"))
7872 {
7873 found_any = true;
7874 }
7875 }
7876
Max Reitz62a01a272019-02-01 20:29:34 +01007877 if (!qdict_haskey(d, "driver")) {
7878 /* Drivers created with bdrv_new_open_driver() may not have a
7879 * @driver option. Add it here. */
7880 qdict_put_str(d, "driver", bs->drv->format_name);
7881 }
7882
Max Reitz97e2f022019-02-01 20:29:27 +01007883 return found_any;
7884}
7885
Max Reitz90993622019-02-01 20:29:09 +01007886/* Note: This function may return false positives; it may return true
7887 * even if opening the backing file specified by bs's image header
7888 * would result in exactly bs->backing. */
Emanuele Giuseppe Espositofa8fc1d2021-12-15 07:11:38 -05007889static bool bdrv_backing_overridden(BlockDriverState *bs)
Max Reitz90993622019-02-01 20:29:09 +01007890{
Emanuele Giuseppe Espositob4ad82a2022-03-03 10:15:57 -05007891 GLOBAL_STATE_CODE();
Max Reitz90993622019-02-01 20:29:09 +01007892 if (bs->backing) {
7893 return strcmp(bs->auto_backing_file,
7894 bs->backing->bs->filename);
7895 } else {
7896 /* No backing BDS, so if the image header reports any backing
7897 * file, it must have been suppressed */
7898 return bs->auto_backing_file[0] != '\0';
7899 }
7900}
7901
Max Reitz91af7012014-07-18 20:24:56 +02007902/* Updates the following BDS fields:
7903 * - exact_filename: A filename which may be used for opening a block device
7904 * which (mostly) equals the given BDS (even without any
7905 * other options; so reading and writing must return the same
7906 * results, but caching etc. may be different)
7907 * - full_open_options: Options which, when given when opening a block device
7908 * (without a filename), result in a BDS (mostly)
7909 * equalling the given one
7910 * - filename: If exact_filename is set, it is copied here. Otherwise,
7911 * full_open_options is converted to a JSON object, prefixed with
7912 * "json:" (for use through the JSON pseudo protocol) and put here.
7913 */
7914void bdrv_refresh_filename(BlockDriverState *bs)
7915{
7916 BlockDriver *drv = bs->drv;
Max Reitze24518e2019-02-01 20:29:06 +01007917 BdrvChild *child;
Max Reitz52f72d62019-06-12 17:43:03 +02007918 BlockDriverState *primary_child_bs;
Max Reitz91af7012014-07-18 20:24:56 +02007919 QDict *opts;
Max Reitz90993622019-02-01 20:29:09 +01007920 bool backing_overridden;
Max Reitz998b3a12019-02-01 20:29:28 +01007921 bool generate_json_filename; /* Whether our default implementation should
7922 fill exact_filename (false) or not (true) */
Max Reitz91af7012014-07-18 20:24:56 +02007923
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05007924 GLOBAL_STATE_CODE();
7925
Max Reitz91af7012014-07-18 20:24:56 +02007926 if (!drv) {
7927 return;
7928 }
7929
Max Reitze24518e2019-02-01 20:29:06 +01007930 /* This BDS's file name may depend on any of its children's file names, so
7931 * refresh those first */
7932 QLIST_FOREACH(child, &bs->children, next) {
7933 bdrv_refresh_filename(child->bs);
Max Reitz91af7012014-07-18 20:24:56 +02007934 }
7935
Max Reitzbb808d52019-02-01 20:29:07 +01007936 if (bs->implicit) {
7937 /* For implicit nodes, just copy everything from the single child */
7938 child = QLIST_FIRST(&bs->children);
7939 assert(QLIST_NEXT(child, next) == NULL);
7940
7941 pstrcpy(bs->exact_filename, sizeof(bs->exact_filename),
7942 child->bs->exact_filename);
7943 pstrcpy(bs->filename, sizeof(bs->filename), child->bs->filename);
7944
Pan Nengyuancb895612020-01-16 16:56:00 +08007945 qobject_unref(bs->full_open_options);
Max Reitzbb808d52019-02-01 20:29:07 +01007946 bs->full_open_options = qobject_ref(child->bs->full_open_options);
7947
7948 return;
7949 }
7950
Max Reitz90993622019-02-01 20:29:09 +01007951 backing_overridden = bdrv_backing_overridden(bs);
7952
7953 if (bs->open_flags & BDRV_O_NO_IO) {
7954 /* Without I/O, the backing file does not change anything.
7955 * Therefore, in such a case (primarily qemu-img), we can
7956 * pretend the backing file has not been overridden even if
7957 * it technically has been. */
7958 backing_overridden = false;
7959 }
7960
Max Reitz97e2f022019-02-01 20:29:27 +01007961 /* Gather the options QDict */
7962 opts = qdict_new();
Max Reitz998b3a12019-02-01 20:29:28 +01007963 generate_json_filename = append_strong_runtime_options(opts, bs);
7964 generate_json_filename |= backing_overridden;
Max Reitz97e2f022019-02-01 20:29:27 +01007965
7966 if (drv->bdrv_gather_child_options) {
7967 /* Some block drivers may not want to present all of their children's
7968 * options, or name them differently from BdrvChild.name */
7969 drv->bdrv_gather_child_options(bs, opts, backing_overridden);
7970 } else {
7971 QLIST_FOREACH(child, &bs->children, next) {
Max Reitz25191e52020-05-13 13:05:33 +02007972 if (child == bs->backing && !backing_overridden) {
Max Reitz97e2f022019-02-01 20:29:27 +01007973 /* We can skip the backing BDS if it has not been overridden */
7974 continue;
7975 }
7976
7977 qdict_put(opts, child->name,
7978 qobject_ref(child->bs->full_open_options));
7979 }
7980
7981 if (backing_overridden && !bs->backing) {
7982 /* Force no backing file */
7983 qdict_put_null(opts, "backing");
7984 }
7985 }
7986
7987 qobject_unref(bs->full_open_options);
7988 bs->full_open_options = opts;
7989
Max Reitz52f72d62019-06-12 17:43:03 +02007990 primary_child_bs = bdrv_primary_bs(bs);
7991
Max Reitz998b3a12019-02-01 20:29:28 +01007992 if (drv->bdrv_refresh_filename) {
7993 /* Obsolete information is of no use here, so drop the old file name
7994 * information before refreshing it */
7995 bs->exact_filename[0] = '\0';
7996
7997 drv->bdrv_refresh_filename(bs);
Max Reitz52f72d62019-06-12 17:43:03 +02007998 } else if (primary_child_bs) {
7999 /*
8000 * Try to reconstruct valid information from the underlying
8001 * file -- this only works for format nodes (filter nodes
8002 * cannot be probed and as such must be selected by the user
8003 * either through an options dict, or through a special
8004 * filename which the filter driver must construct in its
8005 * .bdrv_refresh_filename() implementation).
8006 */
Max Reitz998b3a12019-02-01 20:29:28 +01008007
8008 bs->exact_filename[0] = '\0';
8009
Max Reitzfb695c72019-02-01 20:29:29 +01008010 /*
8011 * We can use the underlying file's filename if:
8012 * - it has a filename,
Max Reitz52f72d62019-06-12 17:43:03 +02008013 * - the current BDS is not a filter,
Max Reitzfb695c72019-02-01 20:29:29 +01008014 * - the file is a protocol BDS, and
8015 * - opening that file (as this BDS's format) will automatically create
8016 * the BDS tree we have right now, that is:
8017 * - the user did not significantly change this BDS's behavior with
8018 * some explicit (strong) options
8019 * - no non-file child of this BDS has been overridden by the user
8020 * Both of these conditions are represented by generate_json_filename.
8021 */
Max Reitz52f72d62019-06-12 17:43:03 +02008022 if (primary_child_bs->exact_filename[0] &&
8023 primary_child_bs->drv->bdrv_file_open &&
8024 !drv->is_filter && !generate_json_filename)
Max Reitzfb695c72019-02-01 20:29:29 +01008025 {
Max Reitz52f72d62019-06-12 17:43:03 +02008026 strcpy(bs->exact_filename, primary_child_bs->exact_filename);
Max Reitz998b3a12019-02-01 20:29:28 +01008027 }
8028 }
8029
Max Reitz91af7012014-07-18 20:24:56 +02008030 if (bs->exact_filename[0]) {
8031 pstrcpy(bs->filename, sizeof(bs->filename), bs->exact_filename);
Max Reitz97e2f022019-02-01 20:29:27 +01008032 } else {
Markus Armbrustereab3a462020-12-11 18:11:37 +01008033 GString *json = qobject_to_json(QOBJECT(bs->full_open_options));
Eric Blake5c86bdf2020-06-08 13:26:38 -05008034 if (snprintf(bs->filename, sizeof(bs->filename), "json:%s",
Markus Armbrustereab3a462020-12-11 18:11:37 +01008035 json->str) >= sizeof(bs->filename)) {
Eric Blake5c86bdf2020-06-08 13:26:38 -05008036 /* Give user a hint if we truncated things. */
8037 strcpy(bs->filename + sizeof(bs->filename) - 4, "...");
8038 }
Markus Armbrustereab3a462020-12-11 18:11:37 +01008039 g_string_free(json, true);
Max Reitz91af7012014-07-18 20:24:56 +02008040 }
8041}
Wen Congyange06018a2016-05-10 15:36:37 +08008042
Max Reitz1e89d0f2019-02-01 20:29:18 +01008043char *bdrv_dirname(BlockDriverState *bs, Error **errp)
8044{
8045 BlockDriver *drv = bs->drv;
Max Reitz52f72d62019-06-12 17:43:03 +02008046 BlockDriverState *child_bs;
Max Reitz1e89d0f2019-02-01 20:29:18 +01008047
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05008048 GLOBAL_STATE_CODE();
8049
Max Reitz1e89d0f2019-02-01 20:29:18 +01008050 if (!drv) {
8051 error_setg(errp, "Node '%s' is ejected", bs->node_name);
8052 return NULL;
8053 }
8054
8055 if (drv->bdrv_dirname) {
8056 return drv->bdrv_dirname(bs, errp);
8057 }
8058
Max Reitz52f72d62019-06-12 17:43:03 +02008059 child_bs = bdrv_primary_bs(bs);
8060 if (child_bs) {
8061 return bdrv_dirname(child_bs, errp);
Max Reitz1e89d0f2019-02-01 20:29:18 +01008062 }
8063
8064 bdrv_refresh_filename(bs);
8065 if (bs->exact_filename[0] != '\0') {
8066 return path_combine(bs->exact_filename, "");
8067 }
8068
8069 error_setg(errp, "Cannot generate a base directory for %s nodes",
8070 drv->format_name);
8071 return NULL;
8072}
8073
Wen Congyange06018a2016-05-10 15:36:37 +08008074/*
8075 * Hot add/remove a BDS's child. So the user can take a child offline when
8076 * it is broken and take a new child online
8077 */
8078void bdrv_add_child(BlockDriverState *parent_bs, BlockDriverState *child_bs,
8079 Error **errp)
8080{
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05008081 GLOBAL_STATE_CODE();
Wen Congyange06018a2016-05-10 15:36:37 +08008082 if (!parent_bs->drv || !parent_bs->drv->bdrv_add_child) {
8083 error_setg(errp, "The node %s does not support adding a child",
8084 bdrv_get_device_or_node_name(parent_bs));
8085 return;
8086 }
8087
Sam Li774c7262023-05-08 12:55:30 +08008088 /*
8089 * Non-zoned block drivers do not follow zoned storage constraints
8090 * (i.e. sequential writes to zones). Refuse mixing zoned and non-zoned
8091 * drivers in a graph.
8092 */
8093 if (!parent_bs->drv->supports_zoned_children &&
8094 child_bs->bl.zoned == BLK_Z_HM) {
8095 /*
8096 * The host-aware model allows zoned storage constraints and random
8097 * write. Allow mixing host-aware and non-zoned drivers. Using
8098 * host-aware device as a regular device.
8099 */
8100 error_setg(errp, "Cannot add a %s child to a %s parent",
8101 child_bs->bl.zoned == BLK_Z_HM ? "zoned" : "non-zoned",
8102 parent_bs->drv->supports_zoned_children ?
8103 "support zoned children" : "not support zoned children");
8104 return;
8105 }
8106
Wen Congyange06018a2016-05-10 15:36:37 +08008107 if (!QLIST_EMPTY(&child_bs->parents)) {
8108 error_setg(errp, "The node %s already has a parent",
8109 child_bs->node_name);
8110 return;
8111 }
8112
8113 parent_bs->drv->bdrv_add_child(parent_bs, child_bs, errp);
8114}
8115
8116void bdrv_del_child(BlockDriverState *parent_bs, BdrvChild *child, Error **errp)
8117{
8118 BdrvChild *tmp;
8119
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05008120 GLOBAL_STATE_CODE();
Wen Congyange06018a2016-05-10 15:36:37 +08008121 if (!parent_bs->drv || !parent_bs->drv->bdrv_del_child) {
8122 error_setg(errp, "The node %s does not support removing a child",
8123 bdrv_get_device_or_node_name(parent_bs));
8124 return;
8125 }
8126
8127 QLIST_FOREACH(tmp, &parent_bs->children, next) {
8128 if (tmp == child) {
8129 break;
8130 }
8131 }
8132
8133 if (!tmp) {
8134 error_setg(errp, "The node %s does not have a child named %s",
8135 bdrv_get_device_or_node_name(parent_bs),
8136 bdrv_get_device_or_node_name(child->bs));
8137 return;
8138 }
8139
8140 parent_bs->drv->bdrv_del_child(parent_bs, child, errp);
8141}
Max Reitz6f7a3b52020-04-29 16:11:23 +02008142
8143int bdrv_make_empty(BdrvChild *c, Error **errp)
8144{
8145 BlockDriver *drv = c->bs->drv;
8146 int ret;
8147
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05008148 GLOBAL_STATE_CODE();
Max Reitz6f7a3b52020-04-29 16:11:23 +02008149 assert(c->perm & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED));
8150
8151 if (!drv->bdrv_make_empty) {
8152 error_setg(errp, "%s does not support emptying nodes",
8153 drv->format_name);
8154 return -ENOTSUP;
8155 }
8156
8157 ret = drv->bdrv_make_empty(c->bs);
8158 if (ret < 0) {
8159 error_setg_errno(errp, -ret, "Failed to empty %s",
8160 c->bs->filename);
8161 return ret;
8162 }
8163
8164 return 0;
8165}
Max Reitz9a6fc882019-05-31 15:23:11 +02008166
8167/*
8168 * Return the child that @bs acts as an overlay for, and from which data may be
8169 * copied in COW or COR operations. Usually this is the backing file.
8170 */
8171BdrvChild *bdrv_cow_child(BlockDriverState *bs)
8172{
Emanuele Giuseppe Esposito967d7902022-03-03 10:15:58 -05008173 IO_CODE();
8174
Max Reitz9a6fc882019-05-31 15:23:11 +02008175 if (!bs || !bs->drv) {
8176 return NULL;
8177 }
8178
8179 if (bs->drv->is_filter) {
8180 return NULL;
8181 }
8182
8183 if (!bs->backing) {
8184 return NULL;
8185 }
8186
8187 assert(bs->backing->role & BDRV_CHILD_COW);
8188 return bs->backing;
8189}
8190
8191/*
8192 * If @bs acts as a filter for exactly one of its children, return
8193 * that child.
8194 */
8195BdrvChild *bdrv_filter_child(BlockDriverState *bs)
8196{
8197 BdrvChild *c;
Emanuele Giuseppe Esposito967d7902022-03-03 10:15:58 -05008198 IO_CODE();
Max Reitz9a6fc882019-05-31 15:23:11 +02008199
8200 if (!bs || !bs->drv) {
8201 return NULL;
8202 }
8203
8204 if (!bs->drv->is_filter) {
8205 return NULL;
8206 }
8207
8208 /* Only one of @backing or @file may be used */
8209 assert(!(bs->backing && bs->file));
8210
8211 c = bs->backing ?: bs->file;
8212 if (!c) {
8213 return NULL;
8214 }
8215
8216 assert(c->role & BDRV_CHILD_FILTERED);
8217 return c;
8218}
8219
8220/*
8221 * Return either the result of bdrv_cow_child() or bdrv_filter_child(),
8222 * whichever is non-NULL.
8223 *
8224 * Return NULL if both are NULL.
8225 */
8226BdrvChild *bdrv_filter_or_cow_child(BlockDriverState *bs)
8227{
8228 BdrvChild *cow_child = bdrv_cow_child(bs);
8229 BdrvChild *filter_child = bdrv_filter_child(bs);
Emanuele Giuseppe Esposito967d7902022-03-03 10:15:58 -05008230 IO_CODE();
Max Reitz9a6fc882019-05-31 15:23:11 +02008231
8232 /* Filter nodes cannot have COW backing files */
8233 assert(!(cow_child && filter_child));
8234
8235 return cow_child ?: filter_child;
8236}
8237
8238/*
8239 * Return the primary child of this node: For filters, that is the
8240 * filtered child. For other nodes, that is usually the child storing
8241 * metadata.
8242 * (A generally more helpful description is that this is (usually) the
8243 * child that has the same filename as @bs.)
8244 *
8245 * Drivers do not necessarily have a primary child; for example quorum
8246 * does not.
8247 */
8248BdrvChild *bdrv_primary_child(BlockDriverState *bs)
8249{
8250 BdrvChild *c, *found = NULL;
Emanuele Giuseppe Esposito967d7902022-03-03 10:15:58 -05008251 IO_CODE();
Max Reitz9a6fc882019-05-31 15:23:11 +02008252
8253 QLIST_FOREACH(c, &bs->children, next) {
8254 if (c->role & BDRV_CHILD_PRIMARY) {
8255 assert(!found);
8256 found = c;
8257 }
8258 }
8259
8260 return found;
8261}
Max Reitzd38d7eb2019-06-12 15:06:37 +02008262
8263static BlockDriverState *bdrv_do_skip_filters(BlockDriverState *bs,
8264 bool stop_on_explicit_filter)
8265{
8266 BdrvChild *c;
8267
8268 if (!bs) {
8269 return NULL;
8270 }
8271
8272 while (!(stop_on_explicit_filter && !bs->implicit)) {
8273 c = bdrv_filter_child(bs);
8274 if (!c) {
8275 /*
8276 * A filter that is embedded in a working block graph must
8277 * have a child. Assert this here so this function does
8278 * not return a filter node that is not expected by the
8279 * caller.
8280 */
8281 assert(!bs->drv || !bs->drv->is_filter);
8282 break;
8283 }
8284 bs = c->bs;
8285 }
8286 /*
8287 * Note that this treats nodes with bs->drv == NULL as not being
8288 * filters (bs->drv == NULL should be replaced by something else
8289 * anyway).
8290 * The advantage of this behavior is that this function will thus
8291 * always return a non-NULL value (given a non-NULL @bs).
8292 */
8293
8294 return bs;
8295}
8296
8297/*
8298 * Return the first BDS that has not been added implicitly or that
8299 * does not have a filtered child down the chain starting from @bs
8300 * (including @bs itself).
8301 */
8302BlockDriverState *bdrv_skip_implicit_filters(BlockDriverState *bs)
8303{
Emanuele Giuseppe Espositob4ad82a2022-03-03 10:15:57 -05008304 GLOBAL_STATE_CODE();
Max Reitzd38d7eb2019-06-12 15:06:37 +02008305 return bdrv_do_skip_filters(bs, true);
8306}
8307
8308/*
8309 * Return the first BDS that does not have a filtered child down the
8310 * chain starting from @bs (including @bs itself).
8311 */
8312BlockDriverState *bdrv_skip_filters(BlockDriverState *bs)
8313{
Emanuele Giuseppe Esposito967d7902022-03-03 10:15:58 -05008314 IO_CODE();
Max Reitzd38d7eb2019-06-12 15:06:37 +02008315 return bdrv_do_skip_filters(bs, false);
8316}
8317
8318/*
8319 * For a backing chain, return the first non-filter backing image of
8320 * the first non-filter image.
8321 */
8322BlockDriverState *bdrv_backing_chain_next(BlockDriverState *bs)
8323{
Emanuele Giuseppe Esposito967d7902022-03-03 10:15:58 -05008324 IO_CODE();
Max Reitzd38d7eb2019-06-12 15:06:37 +02008325 return bdrv_skip_filters(bdrv_cow_bs(bdrv_skip_filters(bs)));
8326}
Hanna Reitz0bc329f2021-08-12 10:41:44 +02008327
8328/**
8329 * Check whether [offset, offset + bytes) overlaps with the cached
8330 * block-status data region.
8331 *
8332 * If so, and @pnum is not NULL, set *pnum to `bsc.data_end - offset`,
8333 * which is what bdrv_bsc_is_data()'s interface needs.
8334 * Otherwise, *pnum is not touched.
8335 */
8336static bool bdrv_bsc_range_overlaps_locked(BlockDriverState *bs,
8337 int64_t offset, int64_t bytes,
8338 int64_t *pnum)
8339{
8340 BdrvBlockStatusCache *bsc = qatomic_rcu_read(&bs->block_status_cache);
8341 bool overlaps;
8342
8343 overlaps =
8344 qatomic_read(&bsc->valid) &&
8345 ranges_overlap(offset, bytes, bsc->data_start,
8346 bsc->data_end - bsc->data_start);
8347
8348 if (overlaps && pnum) {
8349 *pnum = bsc->data_end - offset;
8350 }
8351
8352 return overlaps;
8353}
8354
8355/**
8356 * See block_int.h for this function's documentation.
8357 */
8358bool bdrv_bsc_is_data(BlockDriverState *bs, int64_t offset, int64_t *pnum)
8359{
Emanuele Giuseppe Esposito967d7902022-03-03 10:15:58 -05008360 IO_CODE();
Hanna Reitz0bc329f2021-08-12 10:41:44 +02008361 RCU_READ_LOCK_GUARD();
Hanna Reitz0bc329f2021-08-12 10:41:44 +02008362 return bdrv_bsc_range_overlaps_locked(bs, offset, 1, pnum);
8363}
8364
8365/**
8366 * See block_int.h for this function's documentation.
8367 */
8368void bdrv_bsc_invalidate_range(BlockDriverState *bs,
8369 int64_t offset, int64_t bytes)
8370{
Emanuele Giuseppe Esposito967d7902022-03-03 10:15:58 -05008371 IO_CODE();
Hanna Reitz0bc329f2021-08-12 10:41:44 +02008372 RCU_READ_LOCK_GUARD();
8373
8374 if (bdrv_bsc_range_overlaps_locked(bs, offset, bytes, NULL)) {
8375 qatomic_set(&bs->block_status_cache->valid, false);
8376 }
8377}
8378
8379/**
8380 * See block_int.h for this function's documentation.
8381 */
8382void bdrv_bsc_fill(BlockDriverState *bs, int64_t offset, int64_t bytes)
8383{
8384 BdrvBlockStatusCache *new_bsc = g_new(BdrvBlockStatusCache, 1);
8385 BdrvBlockStatusCache *old_bsc;
Emanuele Giuseppe Esposito967d7902022-03-03 10:15:58 -05008386 IO_CODE();
Hanna Reitz0bc329f2021-08-12 10:41:44 +02008387
8388 *new_bsc = (BdrvBlockStatusCache) {
8389 .valid = true,
8390 .data_start = offset,
8391 .data_end = offset + bytes,
8392 };
8393
8394 QEMU_LOCK_GUARD(&bs->bsc_modify_lock);
8395
8396 old_bsc = qatomic_rcu_read(&bs->block_status_cache);
8397 qatomic_rcu_set(&bs->block_status_cache, new_bsc);
8398 if (old_bsc) {
8399 g_free_rcu(old_bsc, rcu);
8400 }
8401}