blob: 30afdcbba60eab8ffdaeffe5b2a0a168cffd040a [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
Kevin Wolfad29eb32023-09-11 11:46:07 +020094static void GRAPH_WRLOCK
95bdrv_replace_child_noperm(BdrvChild *child, BlockDriverState *new_bs);
96
Kevin Wolf2f64e1f2023-09-11 11:46:08 +020097static void GRAPH_WRLOCK
98bdrv_remove_child(BdrvChild *child, Transaction *tran);
Vladimir Sementsov-Ogievskiy09786232021-04-28 18:17:44 +030099
Vladimir Sementsov-Ogievskiy72373e42021-04-28 18:17:58 +0300100static int bdrv_reopen_prepare(BDRVReopenState *reopen_state,
101 BlockReopenQueue *queue,
Alberto Garciaecd30d22021-06-10 15:05:36 +0300102 Transaction *change_child_tran, Error **errp);
Vladimir Sementsov-Ogievskiy53e96d12021-04-28 18:17:35 +0300103static void bdrv_reopen_commit(BDRVReopenState *reopen_state);
104static void bdrv_reopen_abort(BDRVReopenState *reopen_state);
105
Emanuele Giuseppe Espositofa8fc1d2021-12-15 07:11:38 -0500106static bool bdrv_backing_overridden(BlockDriverState *bs);
107
Emanuele Giuseppe Esposito7e8c1822022-10-25 04:49:44 -0400108static bool bdrv_change_aio_context(BlockDriverState *bs, AioContext *ctx,
Emanuele Giuseppe Espositoe08cc002022-10-25 04:49:45 -0400109 GHashTable *visited, Transaction *tran,
Emanuele Giuseppe Esposito7e8c1822022-10-25 04:49:44 -0400110 Error **errp);
111
Markus Armbrustereb852012009-10-27 18:41:44 +0100112/* If non-zero, use only whitelisted block drivers */
113static int use_bdrv_whitelist;
114
Stefan Hajnoczi9e0b22f2010-12-09 11:53:00 +0000115#ifdef _WIN32
116static int is_windows_drive_prefix(const char *filename)
117{
118 return (((filename[0] >= 'a' && filename[0] <= 'z') ||
119 (filename[0] >= 'A' && filename[0] <= 'Z')) &&
120 filename[1] == ':');
121}
122
123int is_windows_drive(const char *filename)
124{
125 if (is_windows_drive_prefix(filename) &&
126 filename[2] == '\0')
127 return 1;
128 if (strstart(filename, "\\\\.\\", NULL) ||
129 strstart(filename, "//./", NULL))
130 return 1;
131 return 0;
132}
133#endif
134
Kevin Wolf339064d2013-11-28 10:23:32 +0100135size_t bdrv_opt_mem_align(BlockDriverState *bs)
136{
137 if (!bs || !bs->drv) {
Denis V. Lunev459b4e62015-05-12 17:30:56 +0300138 /* page size or 4k (hdd sector size) should be on the safe side */
Marc-André Lureau8e3b0cb2022-03-23 19:57:22 +0400139 return MAX(4096, qemu_real_host_page_size());
Kevin Wolf339064d2013-11-28 10:23:32 +0100140 }
Emanuele Giuseppe Esposito384a48f2022-03-03 10:15:50 -0500141 IO_CODE();
Kevin Wolf339064d2013-11-28 10:23:32 +0100142
143 return bs->bl.opt_mem_alignment;
144}
145
Denis V. Lunev4196d2f2015-05-12 17:30:55 +0300146size_t bdrv_min_mem_align(BlockDriverState *bs)
147{
148 if (!bs || !bs->drv) {
Denis V. Lunev459b4e62015-05-12 17:30:56 +0300149 /* page size or 4k (hdd sector size) should be on the safe side */
Marc-André Lureau8e3b0cb2022-03-23 19:57:22 +0400150 return MAX(4096, qemu_real_host_page_size());
Denis V. Lunev4196d2f2015-05-12 17:30:55 +0300151 }
Emanuele Giuseppe Esposito384a48f2022-03-03 10:15:50 -0500152 IO_CODE();
Denis V. Lunev4196d2f2015-05-12 17:30:55 +0300153
154 return bs->bl.min_mem_alignment;
155}
156
Stefan Hajnoczi9e0b22f2010-12-09 11:53:00 +0000157/* check if the path starts with "<protocol>:" */
Max Reitz5c984152014-12-03 14:57:22 +0100158int path_has_protocol(const char *path)
Stefan Hajnoczi9e0b22f2010-12-09 11:53:00 +0000159{
Paolo Bonzini947995c2012-05-08 16:51:48 +0200160 const char *p;
161
Stefan Hajnoczi9e0b22f2010-12-09 11:53:00 +0000162#ifdef _WIN32
163 if (is_windows_drive(path) ||
164 is_windows_drive_prefix(path)) {
165 return 0;
166 }
Paolo Bonzini947995c2012-05-08 16:51:48 +0200167 p = path + strcspn(path, ":/\\");
168#else
169 p = path + strcspn(path, ":/");
Stefan Hajnoczi9e0b22f2010-12-09 11:53:00 +0000170#endif
171
Paolo Bonzini947995c2012-05-08 16:51:48 +0200172 return *p == ':';
Stefan Hajnoczi9e0b22f2010-12-09 11:53:00 +0000173}
174
bellard83f64092006-08-01 16:21:11 +0000175int path_is_absolute(const char *path)
176{
bellard21664422007-01-07 18:22:37 +0000177#ifdef _WIN32
178 /* specific case for names like: "\\.\d:" */
Paolo Bonzinif53f4da2012-05-08 16:51:47 +0200179 if (is_windows_drive(path) || is_windows_drive_prefix(path)) {
bellard21664422007-01-07 18:22:37 +0000180 return 1;
Paolo Bonzinif53f4da2012-05-08 16:51:47 +0200181 }
182 return (*path == '/' || *path == '\\');
bellard3b9f94e2007-01-07 17:27:07 +0000183#else
Paolo Bonzinif53f4da2012-05-08 16:51:47 +0200184 return (*path == '/');
bellard3b9f94e2007-01-07 17:27:07 +0000185#endif
bellard83f64092006-08-01 16:21:11 +0000186}
187
Max Reitz009b03a2019-02-01 20:29:13 +0100188/* if filename is absolute, just return its duplicate. Otherwise, build a
bellard83f64092006-08-01 16:21:11 +0000189 path to it by considering it is relative to base_path. URL are
190 supported. */
Max Reitz009b03a2019-02-01 20:29:13 +0100191char *path_combine(const char *base_path, const char *filename)
bellard83f64092006-08-01 16:21:11 +0000192{
Max Reitz009b03a2019-02-01 20:29:13 +0100193 const char *protocol_stripped = NULL;
bellard83f64092006-08-01 16:21:11 +0000194 const char *p, *p1;
Max Reitz009b03a2019-02-01 20:29:13 +0100195 char *result;
bellard83f64092006-08-01 16:21:11 +0000196 int len;
197
bellard83f64092006-08-01 16:21:11 +0000198 if (path_is_absolute(filename)) {
Max Reitz009b03a2019-02-01 20:29:13 +0100199 return g_strdup(filename);
bellard83f64092006-08-01 16:21:11 +0000200 }
Max Reitz009b03a2019-02-01 20:29:13 +0100201
202 if (path_has_protocol(base_path)) {
203 protocol_stripped = strchr(base_path, ':');
204 if (protocol_stripped) {
205 protocol_stripped++;
206 }
207 }
208 p = protocol_stripped ?: base_path;
209
210 p1 = strrchr(base_path, '/');
211#ifdef _WIN32
212 {
213 const char *p2;
214 p2 = strrchr(base_path, '\\');
215 if (!p1 || p2 > p1) {
216 p1 = p2;
217 }
218 }
219#endif
220 if (p1) {
221 p1++;
222 } else {
223 p1 = base_path;
224 }
225 if (p1 > p) {
226 p = p1;
227 }
228 len = p - base_path;
229
230 result = g_malloc(len + strlen(filename) + 1);
231 memcpy(result, base_path, len);
232 strcpy(result + len, filename);
233
234 return result;
235}
236
Max Reitz03c320d2017-05-22 21:52:16 +0200237/*
238 * Helper function for bdrv_parse_filename() implementations to remove optional
239 * protocol prefixes (especially "file:") from a filename and for putting the
240 * stripped filename into the options QDict if there is such a prefix.
241 */
242void bdrv_parse_filename_strip_prefix(const char *filename, const char *prefix,
243 QDict *options)
244{
245 if (strstart(filename, prefix, &filename)) {
246 /* Stripping the explicit protocol prefix may result in a protocol
247 * prefix being (wrongly) detected (if the filename contains a colon) */
248 if (path_has_protocol(filename)) {
Markus Armbruster18cf67c2020-12-11 18:11:51 +0100249 GString *fat_filename;
Max Reitz03c320d2017-05-22 21:52:16 +0200250
251 /* This means there is some colon before the first slash; therefore,
252 * this cannot be an absolute path */
253 assert(!path_is_absolute(filename));
254
255 /* And we can thus fix the protocol detection issue by prefixing it
256 * by "./" */
Markus Armbruster18cf67c2020-12-11 18:11:51 +0100257 fat_filename = g_string_new("./");
258 g_string_append(fat_filename, filename);
Max Reitz03c320d2017-05-22 21:52:16 +0200259
Markus Armbruster18cf67c2020-12-11 18:11:51 +0100260 assert(!path_has_protocol(fat_filename->str));
Max Reitz03c320d2017-05-22 21:52:16 +0200261
Markus Armbruster18cf67c2020-12-11 18:11:51 +0100262 qdict_put(options, "filename",
263 qstring_from_gstring(fat_filename));
Max Reitz03c320d2017-05-22 21:52:16 +0200264 } else {
265 /* If no protocol prefix was detected, we can use the shortened
266 * filename as-is */
267 qdict_put_str(options, "filename", filename);
268 }
269 }
270}
271
272
Kevin Wolf9c5e6592017-05-04 18:52:40 +0200273/* Returns whether the image file is opened as read-only. Note that this can
274 * return false and writing to the image file is still not possible because the
275 * image is inactivated. */
Jeff Cody93ed5242017-04-07 16:55:28 -0400276bool bdrv_is_read_only(BlockDriverState *bs)
277{
Emanuele Giuseppe Esposito384a48f2022-03-03 10:15:50 -0500278 IO_CODE();
Vladimir Sementsov-Ogievskiy975da072021-05-27 18:40:55 +0300279 return !(bs->open_flags & BDRV_O_RDWR);
Jeff Cody93ed5242017-04-07 16:55:28 -0400280}
281
Kevin Wolf4026f1c2023-09-29 16:51:47 +0200282static int GRAPH_RDLOCK
283bdrv_can_set_read_only(BlockDriverState *bs, bool read_only,
284 bool ignore_allow_rdw, Error **errp)
Jeff Codyfe5241b2017-04-07 16:55:25 -0400285{
Emanuele Giuseppe Esposito384a48f2022-03-03 10:15:50 -0500286 IO_CODE();
287
Jeff Codye2b82472017-04-07 16:55:26 -0400288 /* Do not set read_only if copy_on_read is enabled */
289 if (bs->copy_on_read && read_only) {
290 error_setg(errp, "Can't set node '%s' to r/o with copy-on-read enabled",
291 bdrv_get_device_or_node_name(bs));
292 return -EINVAL;
293 }
294
Jeff Codyd6fcdf02017-04-07 16:55:27 -0400295 /* Do not clear read_only if it is prohibited */
Kevin Wolf54a32bf2017-08-03 17:02:58 +0200296 if (!read_only && !(bs->open_flags & BDRV_O_ALLOW_RDWR) &&
297 !ignore_allow_rdw)
298 {
Jeff Codyd6fcdf02017-04-07 16:55:27 -0400299 error_setg(errp, "Node '%s' is read only",
300 bdrv_get_device_or_node_name(bs));
301 return -EPERM;
302 }
303
Jeff Cody45803a02017-04-07 16:55:29 -0400304 return 0;
305}
306
Kevin Wolfeaa24102018-10-12 11:27:41 +0200307/*
308 * Called by a driver that can only provide a read-only image.
309 *
310 * Returns 0 if the node is already read-only or it could switch the node to
311 * read-only because BDRV_O_AUTO_RDONLY is set.
312 *
313 * Returns -EACCES if the node is read-write and BDRV_O_AUTO_RDONLY is not set
314 * or bdrv_can_set_read_only() forbids making the node read-only. If @errmsg
315 * is not NULL, it is used as the error message for the Error object.
316 */
317int bdrv_apply_auto_read_only(BlockDriverState *bs, const char *errmsg,
318 Error **errp)
Jeff Cody45803a02017-04-07 16:55:29 -0400319{
320 int ret = 0;
Emanuele Giuseppe Esposito384a48f2022-03-03 10:15:50 -0500321 IO_CODE();
Jeff Cody45803a02017-04-07 16:55:29 -0400322
Kevin Wolfeaa24102018-10-12 11:27:41 +0200323 if (!(bs->open_flags & BDRV_O_RDWR)) {
324 return 0;
325 }
326 if (!(bs->open_flags & BDRV_O_AUTO_RDONLY)) {
327 goto fail;
328 }
329
330 ret = bdrv_can_set_read_only(bs, true, false, NULL);
Jeff Cody45803a02017-04-07 16:55:29 -0400331 if (ret < 0) {
Kevin Wolfeaa24102018-10-12 11:27:41 +0200332 goto fail;
Jeff Cody45803a02017-04-07 16:55:29 -0400333 }
334
Kevin Wolfeaa24102018-10-12 11:27:41 +0200335 bs->open_flags &= ~BDRV_O_RDWR;
Kevin Wolfeeae6a52018-10-09 16:57:12 +0200336
Jeff Codye2b82472017-04-07 16:55:26 -0400337 return 0;
Kevin Wolfeaa24102018-10-12 11:27:41 +0200338
339fail:
340 error_setg(errp, "%s", errmsg ?: "Image is read-only");
341 return -EACCES;
Jeff Codyfe5241b2017-04-07 16:55:25 -0400342}
343
Max Reitz645ae7d2019-02-01 20:29:14 +0100344/*
345 * If @backing is empty, this function returns NULL without setting
346 * @errp. In all other cases, NULL will only be returned with @errp
347 * set.
348 *
349 * Therefore, a return value of NULL without @errp set means that
350 * there is no backing file; if @errp is set, there is one but its
351 * absolute filename cannot be generated.
352 */
353char *bdrv_get_full_backing_filename_from_filename(const char *backed,
354 const char *backing,
355 Error **errp)
Max Reitz0a828552014-11-26 17:20:25 +0100356{
Max Reitz645ae7d2019-02-01 20:29:14 +0100357 if (backing[0] == '\0') {
358 return NULL;
359 } else if (path_has_protocol(backing) || path_is_absolute(backing)) {
360 return g_strdup(backing);
Max Reitz9f074292014-11-26 17:20:26 +0100361 } else if (backed[0] == '\0' || strstart(backed, "json:", NULL)) {
362 error_setg(errp, "Cannot use relative backing file names for '%s'",
363 backed);
Max Reitz645ae7d2019-02-01 20:29:14 +0100364 return NULL;
Max Reitz0a828552014-11-26 17:20:25 +0100365 } else {
Max Reitz645ae7d2019-02-01 20:29:14 +0100366 return path_combine(backed, backing);
Max Reitz0a828552014-11-26 17:20:25 +0100367 }
368}
369
Max Reitz9f4793d2019-02-01 20:29:16 +0100370/*
371 * If @filename is empty or NULL, this function returns NULL without
372 * setting @errp. In all other cases, NULL will only be returned with
373 * @errp set.
374 */
Kevin Wolfb7cfc7d2023-09-29 16:51:45 +0200375static char * GRAPH_RDLOCK
376bdrv_make_absolute_filename(BlockDriverState *relative_to,
377 const char *filename, Error **errp)
Max Reitz9f4793d2019-02-01 20:29:16 +0100378{
Max Reitz8df68612019-02-01 20:29:23 +0100379 char *dir, *full_name;
Max Reitz9f4793d2019-02-01 20:29:16 +0100380
Max Reitz8df68612019-02-01 20:29:23 +0100381 if (!filename || filename[0] == '\0') {
382 return NULL;
383 } else if (path_has_protocol(filename) || path_is_absolute(filename)) {
384 return g_strdup(filename);
385 }
Max Reitz9f4793d2019-02-01 20:29:16 +0100386
Max Reitz8df68612019-02-01 20:29:23 +0100387 dir = bdrv_dirname(relative_to, errp);
388 if (!dir) {
389 return NULL;
390 }
Max Reitz9f4793d2019-02-01 20:29:16 +0100391
Max Reitz8df68612019-02-01 20:29:23 +0100392 full_name = g_strconcat(dir, filename, NULL);
393 g_free(dir);
394 return full_name;
Max Reitz9f4793d2019-02-01 20:29:16 +0100395}
396
Max Reitz6b6833c2019-02-01 20:29:15 +0100397char *bdrv_get_full_backing_filename(BlockDriverState *bs, Error **errp)
Paolo Bonzinidc5a1372012-05-08 16:51:50 +0200398{
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -0500399 GLOBAL_STATE_CODE();
Max Reitz9f4793d2019-02-01 20:29:16 +0100400 return bdrv_make_absolute_filename(bs, bs->backing_file, errp);
Paolo Bonzinidc5a1372012-05-08 16:51:50 +0200401}
402
Stefan Hajnoczi0eb72172015-04-28 14:27:51 +0100403void bdrv_register(BlockDriver *bdrv)
404{
Philippe Mathieu-Daudéa15f08d2020-03-18 23:22:35 +0100405 assert(bdrv->format_name);
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -0500406 GLOBAL_STATE_CODE();
Stefan Hajnoczi8a22f022010-04-13 10:29:33 +0100407 QLIST_INSERT_HEAD(&bdrv_drivers, bdrv, list);
bellardea2384d2004-08-01 21:59:26 +0000408}
bellardb3380822004-03-14 21:38:54 +0000409
Markus Armbrustere4e99862014-10-07 13:59:03 +0200410BlockDriverState *bdrv_new(void)
411{
412 BlockDriverState *bs;
413 int i;
414
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -0500415 GLOBAL_STATE_CODE();
416
Markus Armbruster5839e532014-08-19 10:31:08 +0200417 bs = g_new0(BlockDriverState, 1);
Fam Zhenge4654d22013-11-13 18:29:43 +0800418 QLIST_INIT(&bs->dirty_bitmaps);
Fam Zhengfbe40ff2014-05-23 21:29:42 +0800419 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
420 QLIST_INIT(&bs->op_blockers[i]);
421 }
Stefan Hajnoczifa9185f2023-08-08 11:58:52 -0400422 qemu_mutex_init(&bs->reqs_lock);
Paolo Bonzini21198822017-06-05 14:39:03 +0200423 qemu_mutex_init(&bs->dirty_bitmap_mutex);
Fam Zheng9fcb0252013-08-23 09:14:46 +0800424 bs->refcnt = 1;
Stefan Hajnoczidcd04222014-05-08 16:34:37 +0200425 bs->aio_context = qemu_get_aio_context();
Paolo Bonzinid7d512f2012-08-23 11:20:36 +0200426
Evgeny Yakovlev3ff2f672016-07-18 22:39:52 +0300427 qemu_co_queue_init(&bs->flush_queue);
428
Hanna Reitz0bc329f2021-08-12 10:41:44 +0200429 qemu_co_mutex_init(&bs->bsc_modify_lock);
430 bs->block_status_cache = g_new0(BdrvBlockStatusCache, 1);
431
Kevin Wolf0f122642018-03-28 18:29:18 +0200432 for (i = 0; i < bdrv_drain_all_count; i++) {
433 bdrv_drained_begin(bs);
434 }
435
Max Reitz2c1d04e2016-01-29 16:36:11 +0100436 QTAILQ_INSERT_TAIL(&all_bdrv_states, bs, bs_list);
437
bellardb3380822004-03-14 21:38:54 +0000438 return bs;
439}
440
Marc Mari88d88792016-08-12 09:27:03 -0400441static BlockDriver *bdrv_do_find_format(const char *format_name)
bellardea2384d2004-08-01 21:59:26 +0000442{
443 BlockDriver *drv1;
Emanuele Giuseppe Espositobdb73472022-03-03 10:16:02 -0500444 GLOBAL_STATE_CODE();
Marc Mari88d88792016-08-12 09:27:03 -0400445
Stefan Hajnoczi8a22f022010-04-13 10:29:33 +0100446 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
447 if (!strcmp(drv1->format_name, format_name)) {
bellardea2384d2004-08-01 21:59:26 +0000448 return drv1;
Stefan Hajnoczi8a22f022010-04-13 10:29:33 +0100449 }
bellardea2384d2004-08-01 21:59:26 +0000450 }
Marc Mari88d88792016-08-12 09:27:03 -0400451
bellardea2384d2004-08-01 21:59:26 +0000452 return NULL;
453}
454
Marc Mari88d88792016-08-12 09:27:03 -0400455BlockDriver *bdrv_find_format(const char *format_name)
456{
457 BlockDriver *drv1;
458 int i;
459
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -0500460 GLOBAL_STATE_CODE();
461
Marc Mari88d88792016-08-12 09:27:03 -0400462 drv1 = bdrv_do_find_format(format_name);
463 if (drv1) {
464 return drv1;
465 }
466
467 /* The driver isn't registered, maybe we need to load a module */
468 for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); ++i) {
469 if (!strcmp(block_driver_modules[i].format_name, format_name)) {
Claudio Fontanac551fb02022-09-29 11:30:33 +0200470 Error *local_err = NULL;
471 int rv = block_module_load(block_driver_modules[i].library_name,
472 &local_err);
473 if (rv > 0) {
474 return bdrv_do_find_format(format_name);
475 } else if (rv < 0) {
476 error_report_err(local_err);
477 }
Marc Mari88d88792016-08-12 09:27:03 -0400478 break;
479 }
480 }
Claudio Fontanac551fb02022-09-29 11:30:33 +0200481 return NULL;
Marc Mari88d88792016-08-12 09:27:03 -0400482}
483
Andrey Shinkevich9ac404c2019-03-07 16:33:58 +0300484static int bdrv_format_is_whitelisted(const char *format_name, bool read_only)
Markus Armbrustereb852012009-10-27 18:41:44 +0100485{
Fam Zhengb64ec4e2013-05-29 19:35:40 +0800486 static const char *whitelist_rw[] = {
487 CONFIG_BDRV_RW_WHITELIST
Paolo Bonzini859aef02020-08-04 18:14:26 +0200488 NULL
Fam Zhengb64ec4e2013-05-29 19:35:40 +0800489 };
490 static const char *whitelist_ro[] = {
491 CONFIG_BDRV_RO_WHITELIST
Paolo Bonzini859aef02020-08-04 18:14:26 +0200492 NULL
Markus Armbrustereb852012009-10-27 18:41:44 +0100493 };
494 const char **p;
495
Fam Zhengb64ec4e2013-05-29 19:35:40 +0800496 if (!whitelist_rw[0] && !whitelist_ro[0]) {
Markus Armbrustereb852012009-10-27 18:41:44 +0100497 return 1; /* no whitelist, anything goes */
Fam Zhengb64ec4e2013-05-29 19:35:40 +0800498 }
Markus Armbrustereb852012009-10-27 18:41:44 +0100499
Fam Zhengb64ec4e2013-05-29 19:35:40 +0800500 for (p = whitelist_rw; *p; p++) {
Andrey Shinkevich9ac404c2019-03-07 16:33:58 +0300501 if (!strcmp(format_name, *p)) {
Markus Armbrustereb852012009-10-27 18:41:44 +0100502 return 1;
503 }
504 }
Fam Zhengb64ec4e2013-05-29 19:35:40 +0800505 if (read_only) {
506 for (p = whitelist_ro; *p; p++) {
Andrey Shinkevich9ac404c2019-03-07 16:33:58 +0300507 if (!strcmp(format_name, *p)) {
Fam Zhengb64ec4e2013-05-29 19:35:40 +0800508 return 1;
509 }
510 }
511 }
Markus Armbrustereb852012009-10-27 18:41:44 +0100512 return 0;
513}
514
Andrey Shinkevich9ac404c2019-03-07 16:33:58 +0300515int bdrv_is_whitelisted(BlockDriver *drv, bool read_only)
516{
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -0500517 GLOBAL_STATE_CODE();
Andrey Shinkevich9ac404c2019-03-07 16:33:58 +0300518 return bdrv_format_is_whitelisted(drv->format_name, read_only);
519}
520
Daniel P. Berrangee6ff69b2016-03-21 14:11:48 +0000521bool bdrv_uses_whitelist(void)
522{
523 return use_bdrv_whitelist;
524}
525
Zhi Yong Wu5b7e1542012-05-07 16:50:42 +0800526typedef struct CreateCo {
527 BlockDriver *drv;
528 char *filename;
Chunyan Liu83d05212014-06-05 17:20:51 +0800529 QemuOpts *opts;
Zhi Yong Wu5b7e1542012-05-07 16:50:42 +0800530 int ret;
Max Reitzcc84d902013-09-06 17:14:26 +0200531 Error *err;
Zhi Yong Wu5b7e1542012-05-07 16:50:42 +0800532} CreateCo;
533
Emanuele Giuseppe Esposito741443e2022-11-28 09:23:36 -0500534int coroutine_fn bdrv_co_create(BlockDriver *drv, const char *filename,
535 QemuOpts *opts, Error **errp)
Emanuele Giuseppe Esposito84bdf212022-11-28 09:23:30 -0500536{
537 int ret;
538 GLOBAL_STATE_CODE();
539 ERRP_GUARD();
540
541 if (!drv->bdrv_co_create_opts) {
542 error_setg(errp, "Driver '%s' does not support image creation",
543 drv->format_name);
544 return -ENOTSUP;
545 }
546
547 ret = drv->bdrv_co_create_opts(drv, filename, opts, errp);
548 if (ret < 0 && !*errp) {
549 error_setg_errno(errp, -ret, "Could not create image");
550 }
551
552 return ret;
553}
554
Max Reitzfd171462020-01-22 17:45:29 +0100555/**
556 * Helper function for bdrv_create_file_fallback(): Resize @blk to at
557 * least the given @minimum_size.
558 *
559 * On success, return @blk's actual length.
560 * Otherwise, return -errno.
561 */
Paolo Bonzini84569a72023-06-01 13:51:38 +0200562static int64_t coroutine_fn GRAPH_UNLOCKED
563create_file_fallback_truncate(BlockBackend *blk, int64_t minimum_size,
564 Error **errp)
Max Reitzfd171462020-01-22 17:45:29 +0100565{
566 Error *local_err = NULL;
567 int64_t size;
568 int ret;
569
Emanuele Giuseppe Espositobdb73472022-03-03 10:16:02 -0500570 GLOBAL_STATE_CODE();
571
Paolo Bonzini84569a72023-06-01 13:51:38 +0200572 ret = blk_co_truncate(blk, minimum_size, false, PREALLOC_MODE_OFF, 0,
573 &local_err);
Max Reitzfd171462020-01-22 17:45:29 +0100574 if (ret < 0 && ret != -ENOTSUP) {
575 error_propagate(errp, local_err);
576 return ret;
577 }
578
Paolo Bonzini84569a72023-06-01 13:51:38 +0200579 size = blk_co_getlength(blk);
Max Reitzfd171462020-01-22 17:45:29 +0100580 if (size < 0) {
581 error_free(local_err);
582 error_setg_errno(errp, -size,
583 "Failed to inquire the new image file's length");
584 return size;
585 }
586
587 if (size < minimum_size) {
588 /* Need to grow the image, but we failed to do that */
589 error_propagate(errp, local_err);
590 return -ENOTSUP;
591 }
592
593 error_free(local_err);
594 local_err = NULL;
595
596 return size;
597}
598
599/**
600 * Helper function for bdrv_create_file_fallback(): Zero the first
601 * sector to remove any potentially pre-existing image header.
602 */
Paolo Bonzini881a4c52022-09-22 10:49:00 +0200603static int coroutine_fn
604create_file_fallback_zero_first_sector(BlockBackend *blk,
605 int64_t current_size,
606 Error **errp)
Max Reitzfd171462020-01-22 17:45:29 +0100607{
608 int64_t bytes_to_clear;
609 int ret;
610
Emanuele Giuseppe Espositobdb73472022-03-03 10:16:02 -0500611 GLOBAL_STATE_CODE();
612
Max Reitzfd171462020-01-22 17:45:29 +0100613 bytes_to_clear = MIN(current_size, BDRV_SECTOR_SIZE);
614 if (bytes_to_clear) {
Alberto Fariace47ff22022-10-13 14:37:02 +0200615 ret = blk_co_pwrite_zeroes(blk, 0, bytes_to_clear, BDRV_REQ_MAY_UNMAP);
Max Reitzfd171462020-01-22 17:45:29 +0100616 if (ret < 0) {
617 error_setg_errno(errp, -ret,
618 "Failed to clear the new image's first sector");
619 return ret;
620 }
621 }
622
623 return 0;
624}
625
Maxim Levitsky5a5e7f82020-03-26 03:12:18 +0200626/**
627 * Simple implementation of bdrv_co_create_opts for protocol drivers
628 * which only support creation via opening a file
629 * (usually existing raw storage device)
630 */
631int coroutine_fn bdrv_co_create_opts_simple(BlockDriver *drv,
632 const char *filename,
633 QemuOpts *opts,
634 Error **errp)
Max Reitzfd171462020-01-22 17:45:29 +0100635{
636 BlockBackend *blk;
Max Reitzeeea1fa2020-02-25 16:56:18 +0100637 QDict *options;
Max Reitzfd171462020-01-22 17:45:29 +0100638 int64_t size = 0;
639 char *buf = NULL;
640 PreallocMode prealloc;
641 Error *local_err = NULL;
642 int ret;
643
Emanuele Giuseppe Espositob4ad82a2022-03-03 10:15:57 -0500644 GLOBAL_STATE_CODE();
645
Max Reitzfd171462020-01-22 17:45:29 +0100646 size = qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0);
647 buf = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC);
648 prealloc = qapi_enum_parse(&PreallocMode_lookup, buf,
649 PREALLOC_MODE_OFF, &local_err);
650 g_free(buf);
651 if (local_err) {
652 error_propagate(errp, local_err);
653 return -EINVAL;
654 }
655
656 if (prealloc != PREALLOC_MODE_OFF) {
657 error_setg(errp, "Unsupported preallocation mode '%s'",
658 PreallocMode_str(prealloc));
659 return -ENOTSUP;
660 }
661
Max Reitzeeea1fa2020-02-25 16:56:18 +0100662 options = qdict_new();
Max Reitzfd171462020-01-22 17:45:29 +0100663 qdict_put_str(options, "driver", drv->format_name);
664
Kevin Wolfbe1a7322023-01-26 18:24:31 +0100665 blk = blk_co_new_open(filename, NULL, options,
666 BDRV_O_RDWR | BDRV_O_RESIZE, errp);
Max Reitzfd171462020-01-22 17:45:29 +0100667 if (!blk) {
Hanna Czenczek81624862023-07-20 16:00:24 +0200668 error_prepend(errp, "Protocol driver '%s' does not support creating "
669 "new images, so an existing image must be selected as "
670 "the target; however, opening the given target as an "
671 "existing image failed: ",
Max Reitzfd171462020-01-22 17:45:29 +0100672 drv->format_name);
673 return -EINVAL;
674 }
675
676 size = create_file_fallback_truncate(blk, size, errp);
677 if (size < 0) {
678 ret = size;
679 goto out;
680 }
681
682 ret = create_file_fallback_zero_first_sector(blk, size, errp);
683 if (ret < 0) {
684 goto out;
685 }
686
687 ret = 0;
688out:
Kevin Wolfb2ab5f52023-05-04 13:57:33 +0200689 blk_co_unref(blk);
Max Reitzfd171462020-01-22 17:45:29 +0100690 return ret;
691}
692
Emanuele Giuseppe Esposito2475a0d2022-11-28 09:23:31 -0500693int coroutine_fn bdrv_co_create_file(const char *filename, QemuOpts *opts,
694 Error **errp)
Christoph Hellwig84a12e62010-04-07 22:30:24 +0200695{
Stefano Garzarella729222a2021-03-08 17:12:32 +0100696 QemuOpts *protocol_opts;
Christoph Hellwig84a12e62010-04-07 22:30:24 +0200697 BlockDriver *drv;
Stefano Garzarella729222a2021-03-08 17:12:32 +0100698 QDict *qdict;
699 int ret;
Christoph Hellwig84a12e62010-04-07 22:30:24 +0200700
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -0500701 GLOBAL_STATE_CODE();
702
Max Reitzb65a5e12015-02-05 13:58:12 -0500703 drv = bdrv_find_protocol(filename, true, errp);
Christoph Hellwig84a12e62010-04-07 22:30:24 +0200704 if (drv == NULL) {
Stefan Hajnoczi16905d72010-11-30 15:14:14 +0000705 return -ENOENT;
Christoph Hellwig84a12e62010-04-07 22:30:24 +0200706 }
707
Stefano Garzarella729222a2021-03-08 17:12:32 +0100708 if (!drv->create_opts) {
709 error_setg(errp, "Driver '%s' does not support image creation",
710 drv->format_name);
711 return -ENOTSUP;
712 }
713
714 /*
715 * 'opts' contains a QemuOptsList with a combination of format and protocol
716 * default values.
717 *
718 * The format properly removes its options, but the default values remain
719 * in 'opts->list'. So if the protocol has options with the same name
720 * (e.g. rbd has 'cluster_size' as qcow2), it will see the default values
721 * of the format, since for overlapping options, the format wins.
722 *
723 * To avoid this issue, lets convert QemuOpts to QDict, in this way we take
724 * only the set options, and then convert it back to QemuOpts, using the
725 * create_opts of the protocol. So the new QemuOpts, will contain only the
726 * protocol defaults.
727 */
728 qdict = qemu_opts_to_qdict(opts, NULL);
729 protocol_opts = qemu_opts_from_qdict(drv->create_opts, qdict, errp);
730 if (protocol_opts == NULL) {
731 ret = -EINVAL;
732 goto out;
733 }
734
Emanuele Giuseppe Esposito2475a0d2022-11-28 09:23:31 -0500735 ret = bdrv_co_create(drv, filename, protocol_opts, errp);
Stefano Garzarella729222a2021-03-08 17:12:32 +0100736out:
737 qemu_opts_del(protocol_opts);
738 qobject_unref(qdict);
739 return ret;
Christoph Hellwig84a12e62010-04-07 22:30:24 +0200740}
741
Daniel Henrique Barbozae1d7f8b2020-01-30 18:39:05 -0300742int coroutine_fn bdrv_co_delete_file(BlockDriverState *bs, Error **errp)
743{
744 Error *local_err = NULL;
745 int ret;
746
Emanuele Giuseppe Esposito384a48f2022-03-03 10:15:50 -0500747 IO_CODE();
Daniel Henrique Barbozae1d7f8b2020-01-30 18:39:05 -0300748 assert(bs != NULL);
Kevin Wolf48aef792023-02-03 16:22:00 +0100749 assert_bdrv_graph_readable();
Daniel Henrique Barbozae1d7f8b2020-01-30 18:39:05 -0300750
751 if (!bs->drv) {
752 error_setg(errp, "Block node '%s' is not opened", bs->filename);
753 return -ENOMEDIUM;
754 }
755
756 if (!bs->drv->bdrv_co_delete_file) {
757 error_setg(errp, "Driver '%s' does not support image deletion",
758 bs->drv->format_name);
759 return -ENOTSUP;
760 }
761
762 ret = bs->drv->bdrv_co_delete_file(bs, &local_err);
763 if (ret < 0) {
764 error_propagate(errp, local_err);
765 }
766
767 return ret;
768}
769
Maxim Levitskya890f082020-12-17 19:09:03 +0200770void coroutine_fn bdrv_co_delete_file_noerr(BlockDriverState *bs)
771{
772 Error *local_err = NULL;
773 int ret;
Emanuele Giuseppe Esposito384a48f2022-03-03 10:15:50 -0500774 IO_CODE();
Maxim Levitskya890f082020-12-17 19:09:03 +0200775
776 if (!bs) {
777 return;
778 }
779
780 ret = bdrv_co_delete_file(bs, &local_err);
781 /*
782 * ENOTSUP will happen if the block driver doesn't support
783 * the 'bdrv_co_delete_file' interface. This is a predictable
784 * scenario and shouldn't be reported back to the user.
785 */
786 if (ret == -ENOTSUP) {
787 error_free(local_err);
788 } else if (ret < 0) {
789 error_report_err(local_err);
790 }
791}
792
Ekaterina Tumanova892b7de2015-02-16 12:47:54 +0100793/**
794 * Try to get @bs's logical and physical block size.
795 * On success, store them in @bsz struct and return 0.
796 * On failure return -errno.
797 * @bs must not be empty.
798 */
799int bdrv_probe_blocksizes(BlockDriverState *bs, BlockSizes *bsz)
800{
801 BlockDriver *drv = bs->drv;
Max Reitz93393e62019-06-12 17:03:38 +0200802 BlockDriverState *filtered = bdrv_filter_bs(bs);
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -0500803 GLOBAL_STATE_CODE();
Ekaterina Tumanova892b7de2015-02-16 12:47:54 +0100804
805 if (drv && drv->bdrv_probe_blocksizes) {
806 return drv->bdrv_probe_blocksizes(bs, bsz);
Max Reitz93393e62019-06-12 17:03:38 +0200807 } else if (filtered) {
808 return bdrv_probe_blocksizes(filtered, bsz);
Ekaterina Tumanova892b7de2015-02-16 12:47:54 +0100809 }
810
811 return -ENOTSUP;
812}
813
814/**
815 * Try to get @bs's geometry (cyls, heads, sectors).
816 * On success, store them in @geo struct and return 0.
817 * On failure return -errno.
818 * @bs must not be empty.
819 */
820int bdrv_probe_geometry(BlockDriverState *bs, HDGeometry *geo)
821{
822 BlockDriver *drv = bs->drv;
Kevin Wolff5a3a272023-10-27 17:53:12 +0200823 BlockDriverState *filtered;
824
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -0500825 GLOBAL_STATE_CODE();
Kevin Wolff5a3a272023-10-27 17:53:12 +0200826 GRAPH_RDLOCK_GUARD_MAINLOOP();
Ekaterina Tumanova892b7de2015-02-16 12:47:54 +0100827
828 if (drv && drv->bdrv_probe_geometry) {
829 return drv->bdrv_probe_geometry(bs, geo);
Kevin Wolff5a3a272023-10-27 17:53:12 +0200830 }
831
832 filtered = bdrv_filter_bs(bs);
833 if (filtered) {
Max Reitz93393e62019-06-12 17:03:38 +0200834 return bdrv_probe_geometry(filtered, geo);
Ekaterina Tumanova892b7de2015-02-16 12:47:54 +0100835 }
836
837 return -ENOTSUP;
838}
839
Jim Meyeringeba25052012-05-28 09:27:54 +0200840/*
841 * Create a uniquely-named empty temporary file.
Bin Meng69fbfff2022-10-10 12:04:31 +0800842 * Return the actual file name used upon success, otherwise NULL.
843 * This string should be freed with g_free() when not needed any longer.
844 *
845 * Note: creating a temporary file for the caller to (re)open is
846 * inherently racy. Use g_file_open_tmp() instead whenever practical.
Jim Meyeringeba25052012-05-28 09:27:54 +0200847 */
Bin Meng69fbfff2022-10-10 12:04:31 +0800848char *create_tmp_file(Error **errp)
Jim Meyeringeba25052012-05-28 09:27:54 +0200849{
bellardea2384d2004-08-01 21:59:26 +0000850 int fd;
blueswir17ccfb2e2008-09-14 06:45:34 +0000851 const char *tmpdir;
Bin Meng69fbfff2022-10-10 12:04:31 +0800852 g_autofree char *filename = NULL;
853
854 tmpdir = g_get_tmp_dir();
855#ifndef _WIN32
856 /*
857 * See commit 69bef79 ("block: use /var/tmp instead of /tmp for -snapshot")
858 *
859 * This function is used to create temporary disk images (like -snapshot),
860 * so the files can become very large. /tmp is often a tmpfs where as
861 * /var/tmp is usually on a disk, so more appropriate for disk images.
862 */
863 if (!g_strcmp0(tmpdir, "/tmp")) {
Amit Shah69bef792014-02-26 15:12:37 +0530864 tmpdir = "/var/tmp";
865 }
bellardd5249392004-08-03 21:14:23 +0000866#endif
Bin Meng69fbfff2022-10-10 12:04:31 +0800867
868 filename = g_strdup_printf("%s/vl.XXXXXX", tmpdir);
869 fd = g_mkstemp(filename);
bellardea2384d2004-08-01 21:59:26 +0000870 if (fd < 0) {
Bin Meng69fbfff2022-10-10 12:04:31 +0800871 error_setg_errno(errp, errno, "Could not open temporary file '%s'",
872 filename);
873 return NULL;
bellardea2384d2004-08-01 21:59:26 +0000874 }
Bin Meng6b6471e2022-10-10 12:04:30 +0800875 close(fd);
Bin Meng69fbfff2022-10-10 12:04:31 +0800876
877 return g_steal_pointer(&filename);
Jim Meyeringeba25052012-05-28 09:27:54 +0200878}
bellardea2384d2004-08-01 21:59:26 +0000879
Christoph Hellwigf3a5d3f2009-06-15 13:55:19 +0200880/*
881 * Detect host devices. By convention, /dev/cdrom[N] is always
882 * recognized as a host CDROM.
883 */
Christoph Hellwigf3a5d3f2009-06-15 13:55:19 +0200884static BlockDriver *find_hdev_driver(const char *filename)
885{
Christoph Hellwig508c7cb2009-06-15 14:04:22 +0200886 int score_max = 0, score;
887 BlockDriver *drv = NULL, *d;
Emanuele Giuseppe Espositobdb73472022-03-03 10:16:02 -0500888 GLOBAL_STATE_CODE();
Christoph Hellwigf3a5d3f2009-06-15 13:55:19 +0200889
Stefan Hajnoczi8a22f022010-04-13 10:29:33 +0100890 QLIST_FOREACH(d, &bdrv_drivers, list) {
Christoph Hellwig508c7cb2009-06-15 14:04:22 +0200891 if (d->bdrv_probe_device) {
892 score = d->bdrv_probe_device(filename);
893 if (score > score_max) {
894 score_max = score;
895 drv = d;
896 }
897 }
Christoph Hellwigf3a5d3f2009-06-15 13:55:19 +0200898 }
899
Christoph Hellwig508c7cb2009-06-15 14:04:22 +0200900 return drv;
Christoph Hellwigf3a5d3f2009-06-15 13:55:19 +0200901}
Christoph Hellwigf3a5d3f2009-06-15 13:55:19 +0200902
Marc Mari88d88792016-08-12 09:27:03 -0400903static BlockDriver *bdrv_do_find_protocol(const char *protocol)
904{
905 BlockDriver *drv1;
Emanuele Giuseppe Espositobdb73472022-03-03 10:16:02 -0500906 GLOBAL_STATE_CODE();
Marc Mari88d88792016-08-12 09:27:03 -0400907
908 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
909 if (drv1->protocol_name && !strcmp(drv1->protocol_name, protocol)) {
910 return drv1;
911 }
912 }
913
914 return NULL;
915}
916
Kevin Wolf98289622013-07-10 15:47:39 +0200917BlockDriver *bdrv_find_protocol(const char *filename,
Max Reitzb65a5e12015-02-05 13:58:12 -0500918 bool allow_protocol_prefix,
919 Error **errp)
Christoph Hellwig84a12e62010-04-07 22:30:24 +0200920{
921 BlockDriver *drv1;
922 char protocol[128];
923 int len;
924 const char *p;
Marc Mari88d88792016-08-12 09:27:03 -0400925 int i;
Christoph Hellwig84a12e62010-04-07 22:30:24 +0200926
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -0500927 GLOBAL_STATE_CODE();
Kevin Wolf66f82ce2010-04-14 14:17:38 +0200928 /* TODO Drivers without bdrv_file_open must be specified explicitly */
929
Christoph Hellwig39508e72010-06-23 12:25:17 +0200930 /*
931 * XXX(hch): we really should not let host device detection
932 * override an explicit protocol specification, but moving this
933 * later breaks access to device names with colons in them.
934 * Thanks to the brain-dead persistent naming schemes on udev-
935 * based Linux systems those actually are quite common.
936 */
937 drv1 = find_hdev_driver(filename);
938 if (drv1) {
Christoph Hellwig84a12e62010-04-07 22:30:24 +0200939 return drv1;
940 }
Christoph Hellwig39508e72010-06-23 12:25:17 +0200941
Kevin Wolf98289622013-07-10 15:47:39 +0200942 if (!path_has_protocol(filename) || !allow_protocol_prefix) {
Max Reitzef810432014-12-02 18:32:42 +0100943 return &bdrv_file;
Christoph Hellwig39508e72010-06-23 12:25:17 +0200944 }
Kevin Wolf98289622013-07-10 15:47:39 +0200945
Stefan Hajnoczi9e0b22f2010-12-09 11:53:00 +0000946 p = strchr(filename, ':');
947 assert(p != NULL);
Christoph Hellwig84a12e62010-04-07 22:30:24 +0200948 len = p - filename;
949 if (len > sizeof(protocol) - 1)
950 len = sizeof(protocol) - 1;
951 memcpy(protocol, filename, len);
952 protocol[len] = '\0';
Marc Mari88d88792016-08-12 09:27:03 -0400953
954 drv1 = bdrv_do_find_protocol(protocol);
955 if (drv1) {
956 return drv1;
957 }
958
959 for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); ++i) {
960 if (block_driver_modules[i].protocol_name &&
961 !strcmp(block_driver_modules[i].protocol_name, protocol)) {
Claudio Fontanac551fb02022-09-29 11:30:33 +0200962 int rv = block_module_load(block_driver_modules[i].library_name, errp);
963 if (rv > 0) {
964 drv1 = bdrv_do_find_protocol(protocol);
965 } else if (rv < 0) {
966 return NULL;
967 }
Marc Mari88d88792016-08-12 09:27:03 -0400968 break;
Christoph Hellwig84a12e62010-04-07 22:30:24 +0200969 }
970 }
Max Reitzb65a5e12015-02-05 13:58:12 -0500971
Marc Mari88d88792016-08-12 09:27:03 -0400972 if (!drv1) {
973 error_setg(errp, "Unknown protocol '%s'", protocol);
974 }
975 return drv1;
Christoph Hellwig84a12e62010-04-07 22:30:24 +0200976}
977
Markus Armbrusterc6684242014-11-20 16:27:10 +0100978/*
979 * Guess image format by probing its contents.
980 * This is not a good idea when your image is raw (CVE-2008-2004), but
981 * we do it anyway for backward compatibility.
982 *
983 * @buf contains the image's first @buf_size bytes.
Kevin Wolf7cddd372014-11-20 16:27:11 +0100984 * @buf_size is the buffer size in bytes (generally BLOCK_PROBE_BUF_SIZE,
985 * but can be smaller if the image file is smaller)
Markus Armbrusterc6684242014-11-20 16:27:10 +0100986 * @filename is its filename.
987 *
988 * For all block drivers, call the bdrv_probe() method to get its
989 * probing score.
990 * Return the first block driver with the highest probing score.
991 */
Kevin Wolf38f3ef52014-11-20 16:27:12 +0100992BlockDriver *bdrv_probe_all(const uint8_t *buf, int buf_size,
993 const char *filename)
Markus Armbrusterc6684242014-11-20 16:27:10 +0100994{
995 int score_max = 0, score;
996 BlockDriver *drv = NULL, *d;
Emanuele Giuseppe Esposito967d7902022-03-03 10:15:58 -0500997 IO_CODE();
Markus Armbrusterc6684242014-11-20 16:27:10 +0100998
999 QLIST_FOREACH(d, &bdrv_drivers, list) {
1000 if (d->bdrv_probe) {
1001 score = d->bdrv_probe(buf, buf_size, filename);
1002 if (score > score_max) {
1003 score_max = score;
1004 drv = d;
1005 }
1006 }
1007 }
1008
1009 return drv;
1010}
1011
Kevin Wolf5696c6e2017-02-17 18:39:24 +01001012static int find_image_format(BlockBackend *file, const char *filename,
Max Reitz34b5d2c2013-09-05 14:45:29 +02001013 BlockDriver **pdrv, Error **errp)
bellardea2384d2004-08-01 21:59:26 +00001014{
Markus Armbrusterc6684242014-11-20 16:27:10 +01001015 BlockDriver *drv;
Kevin Wolf7cddd372014-11-20 16:27:11 +01001016 uint8_t buf[BLOCK_PROBE_BUF_SIZE];
Kevin Wolff500a6d2012-11-12 17:35:27 +01001017 int ret = 0;
Nicholas Bellingerf8ea0b02010-05-17 09:45:57 -07001018
Emanuele Giuseppe Espositobdb73472022-03-03 10:16:02 -05001019 GLOBAL_STATE_CODE();
1020
Kevin Wolf08a00552010-06-01 18:37:31 +02001021 /* Return the raw BlockDriver * to scsi-generic devices or empty drives */
Kevin Wolf5696c6e2017-02-17 18:39:24 +01001022 if (blk_is_sg(file) || !blk_is_inserted(file) || blk_getlength(file) == 0) {
Max Reitzef810432014-12-02 18:32:42 +01001023 *pdrv = &bdrv_raw;
Stefan Weilc98ac352010-07-21 21:51:51 +02001024 return ret;
Nicholas A. Bellinger1a396852010-05-27 08:56:28 -07001025 }
Nicholas Bellingerf8ea0b02010-05-17 09:45:57 -07001026
Alberto Fariaa9262f52022-07-05 17:15:11 +01001027 ret = blk_pread(file, 0, sizeof(buf), buf, 0);
bellard83f64092006-08-01 16:21:11 +00001028 if (ret < 0) {
Max Reitz34b5d2c2013-09-05 14:45:29 +02001029 error_setg_errno(errp, -ret, "Could not read image for determining its "
1030 "format");
Stefan Weilc98ac352010-07-21 21:51:51 +02001031 *pdrv = NULL;
1032 return ret;
bellard83f64092006-08-01 16:21:11 +00001033 }
1034
Alberto Fariabf5b16f2022-07-05 17:15:09 +01001035 drv = bdrv_probe_all(buf, sizeof(buf), filename);
Stefan Weilc98ac352010-07-21 21:51:51 +02001036 if (!drv) {
Max Reitz34b5d2c2013-09-05 14:45:29 +02001037 error_setg(errp, "Could not determine image format: No compatible "
1038 "driver found");
Alberto Fariabf5b16f2022-07-05 17:15:09 +01001039 *pdrv = NULL;
1040 return -ENOENT;
Stefan Weilc98ac352010-07-21 21:51:51 +02001041 }
Alberto Fariabf5b16f2022-07-05 17:15:09 +01001042
Stefan Weilc98ac352010-07-21 21:51:51 +02001043 *pdrv = drv;
Alberto Fariabf5b16f2022-07-05 17:15:09 +01001044 return 0;
bellardea2384d2004-08-01 21:59:26 +00001045}
1046
Stefan Hajnoczi51762282010-04-19 16:56:41 +01001047/**
1048 * Set the current 'total_sectors' value
Markus Armbruster65a9bb22014-06-26 13:23:17 +02001049 * Return 0 on success, -errno on error.
Stefan Hajnoczi51762282010-04-19 16:56:41 +01001050 */
Emanuele Giuseppe Espositoc86422c2023-01-13 21:42:04 +01001051int coroutine_fn bdrv_co_refresh_total_sectors(BlockDriverState *bs,
1052 int64_t hint)
Stefan Hajnoczi51762282010-04-19 16:56:41 +01001053{
1054 BlockDriver *drv = bs->drv;
Emanuele Giuseppe Esposito967d7902022-03-03 10:15:58 -05001055 IO_CODE();
Kevin Wolf8ab81402023-02-03 16:22:02 +01001056 assert_bdrv_graph_readable();
Stefan Hajnoczi51762282010-04-19 16:56:41 +01001057
Max Reitzd470ad42017-11-10 21:31:09 +01001058 if (!drv) {
1059 return -ENOMEDIUM;
1060 }
1061
Emanuele Giuseppe Espositoc86422c2023-01-13 21:42:04 +01001062 /* Do not attempt drv->bdrv_co_getlength() on scsi-generic devices */
Dimitris Aragiorgisb192af82015-06-23 13:44:56 +03001063 if (bdrv_is_sg(bs))
Nicholas Bellinger396759a2010-05-17 09:46:04 -07001064 return 0;
1065
Stefan Hajnoczi51762282010-04-19 16:56:41 +01001066 /* query actual device if possible, otherwise just trust the hint */
Emanuele Giuseppe Espositoc86422c2023-01-13 21:42:04 +01001067 if (drv->bdrv_co_getlength) {
1068 int64_t length = drv->bdrv_co_getlength(bs);
Stefan Hajnoczi51762282010-04-19 16:56:41 +01001069 if (length < 0) {
1070 return length;
1071 }
Fam Zheng7e382002013-11-06 19:48:06 +08001072 hint = DIV_ROUND_UP(length, BDRV_SECTOR_SIZE);
Stefan Hajnoczi51762282010-04-19 16:56:41 +01001073 }
1074
1075 bs->total_sectors = hint;
Vladimir Sementsov-Ogievskiy8b117002020-12-04 01:27:13 +03001076
1077 if (bs->total_sectors * BDRV_SECTOR_SIZE > BDRV_MAX_LENGTH) {
1078 return -EFBIG;
1079 }
1080
Stefan Hajnoczi51762282010-04-19 16:56:41 +01001081 return 0;
1082}
1083
Stefan Hajnoczic3993cd2011-08-04 12:26:51 +01001084/**
Kevin Wolfcddff5b2015-11-16 16:43:27 +01001085 * Combines a QDict of new block driver @options with any missing options taken
1086 * from @old_options, so that leaving out an option defaults to its old value.
1087 */
1088static void bdrv_join_options(BlockDriverState *bs, QDict *options,
1089 QDict *old_options)
1090{
Emanuele Giuseppe Espositoda359902022-03-03 10:16:11 -05001091 GLOBAL_STATE_CODE();
Kevin Wolfcddff5b2015-11-16 16:43:27 +01001092 if (bs->drv && bs->drv->bdrv_join_options) {
1093 bs->drv->bdrv_join_options(options, old_options);
1094 } else {
1095 qdict_join(options, old_options, false);
1096 }
1097}
1098
Alberto Garcia543770b2018-09-06 12:37:09 +03001099static BlockdevDetectZeroesOptions bdrv_parse_detect_zeroes(QemuOpts *opts,
1100 int open_flags,
1101 Error **errp)
1102{
1103 Error *local_err = NULL;
1104 char *value = qemu_opt_get_del(opts, "detect-zeroes");
1105 BlockdevDetectZeroesOptions detect_zeroes =
1106 qapi_enum_parse(&BlockdevDetectZeroesOptions_lookup, value,
1107 BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF, &local_err);
Emanuele Giuseppe Espositobdb73472022-03-03 10:16:02 -05001108 GLOBAL_STATE_CODE();
Alberto Garcia543770b2018-09-06 12:37:09 +03001109 g_free(value);
1110 if (local_err) {
1111 error_propagate(errp, local_err);
1112 return detect_zeroes;
1113 }
1114
1115 if (detect_zeroes == BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP &&
1116 !(open_flags & BDRV_O_UNMAP))
1117 {
1118 error_setg(errp, "setting detect-zeroes to unmap is not allowed "
1119 "without setting discard operation to unmap");
1120 }
1121
1122 return detect_zeroes;
1123}
1124
Kevin Wolfcddff5b2015-11-16 16:43:27 +01001125/**
Aarushi Mehtaf80f2672020-01-20 14:18:50 +00001126 * Set open flags for aio engine
1127 *
1128 * Return 0 on success, -1 if the engine specified is invalid
1129 */
1130int bdrv_parse_aio(const char *mode, int *flags)
1131{
1132 if (!strcmp(mode, "threads")) {
1133 /* do nothing, default */
1134 } else if (!strcmp(mode, "native")) {
1135 *flags |= BDRV_O_NATIVE_AIO;
1136#ifdef CONFIG_LINUX_IO_URING
1137 } else if (!strcmp(mode, "io_uring")) {
1138 *flags |= BDRV_O_IO_URING;
1139#endif
1140 } else {
1141 return -1;
1142 }
1143
1144 return 0;
1145}
1146
1147/**
Paolo Bonzini9e8f1832013-02-08 14:06:11 +01001148 * Set open flags for a given discard mode
1149 *
1150 * Return 0 on success, -1 if the discard mode was invalid.
1151 */
1152int bdrv_parse_discard_flags(const char *mode, int *flags)
1153{
1154 *flags &= ~BDRV_O_UNMAP;
1155
1156 if (!strcmp(mode, "off") || !strcmp(mode, "ignore")) {
1157 /* do nothing */
1158 } else if (!strcmp(mode, "on") || !strcmp(mode, "unmap")) {
1159 *flags |= BDRV_O_UNMAP;
1160 } else {
1161 return -1;
1162 }
1163
1164 return 0;
1165}
1166
1167/**
Stefan Hajnoczic3993cd2011-08-04 12:26:51 +01001168 * Set open flags for a given cache mode
1169 *
1170 * Return 0 on success, -1 if the cache mode was invalid.
1171 */
Kevin Wolf53e8ae02016-03-18 15:36:58 +01001172int bdrv_parse_cache_mode(const char *mode, int *flags, bool *writethrough)
Stefan Hajnoczic3993cd2011-08-04 12:26:51 +01001173{
1174 *flags &= ~BDRV_O_CACHE_MASK;
1175
1176 if (!strcmp(mode, "off") || !strcmp(mode, "none")) {
Kevin Wolf53e8ae02016-03-18 15:36:58 +01001177 *writethrough = false;
1178 *flags |= BDRV_O_NOCACHE;
Stefan Hajnoczi92196b22011-08-04 12:26:52 +01001179 } else if (!strcmp(mode, "directsync")) {
Kevin Wolf53e8ae02016-03-18 15:36:58 +01001180 *writethrough = true;
Stefan Hajnoczi92196b22011-08-04 12:26:52 +01001181 *flags |= BDRV_O_NOCACHE;
Stefan Hajnoczic3993cd2011-08-04 12:26:51 +01001182 } else if (!strcmp(mode, "writeback")) {
Kevin Wolf53e8ae02016-03-18 15:36:58 +01001183 *writethrough = false;
Stefan Hajnoczic3993cd2011-08-04 12:26:51 +01001184 } else if (!strcmp(mode, "unsafe")) {
Kevin Wolf53e8ae02016-03-18 15:36:58 +01001185 *writethrough = false;
Stefan Hajnoczic3993cd2011-08-04 12:26:51 +01001186 *flags |= BDRV_O_NO_FLUSH;
1187 } else if (!strcmp(mode, "writethrough")) {
Kevin Wolf53e8ae02016-03-18 15:36:58 +01001188 *writethrough = true;
Stefan Hajnoczic3993cd2011-08-04 12:26:51 +01001189 } else {
1190 return -1;
1191 }
1192
1193 return 0;
1194}
1195
Kevin Wolfb5411552017-01-17 15:56:16 +01001196static char *bdrv_child_get_parent_desc(BdrvChild *c)
1197{
1198 BlockDriverState *parent = c->opaque;
Vladimir Sementsov-Ogievskiy2c0a3ac2021-06-01 10:52:15 +03001199 return g_strdup_printf("node '%s'", bdrv_get_node_name(parent));
Kevin Wolfb5411552017-01-17 15:56:16 +01001200}
1201
Emanuele Giuseppe Espositod05ab382023-09-29 16:51:40 +02001202static void GRAPH_RDLOCK bdrv_child_cb_drained_begin(BdrvChild *child)
Kevin Wolf20018e12016-05-23 18:46:59 +02001203{
1204 BlockDriverState *bs = child->opaque;
Kevin Wolfa82a3bd2022-11-18 18:41:07 +01001205 bdrv_do_drained_begin_quiesce(bs, NULL);
Kevin Wolf20018e12016-05-23 18:46:59 +02001206}
1207
Emanuele Giuseppe Espositod05ab382023-09-29 16:51:40 +02001208static bool GRAPH_RDLOCK bdrv_child_cb_drained_poll(BdrvChild *child)
Kevin Wolf89bd0302018-03-22 14:11:20 +01001209{
1210 BlockDriverState *bs = child->opaque;
Kevin Wolf299403a2022-11-18 18:41:05 +01001211 return bdrv_drain_poll(bs, NULL, false);
Kevin Wolf89bd0302018-03-22 14:11:20 +01001212}
1213
Emanuele Giuseppe Espositod05ab382023-09-29 16:51:40 +02001214static void GRAPH_RDLOCK bdrv_child_cb_drained_end(BdrvChild *child)
Kevin Wolf20018e12016-05-23 18:46:59 +02001215{
1216 BlockDriverState *bs = child->opaque;
Kevin Wolf2f65df62022-11-18 18:40:59 +01001217 bdrv_drained_end(bs);
Kevin Wolf20018e12016-05-23 18:46:59 +02001218}
1219
Kevin Wolf38701b62017-05-04 18:52:39 +02001220static int bdrv_child_cb_inactivate(BdrvChild *child)
1221{
1222 BlockDriverState *bs = child->opaque;
Emanuele Giuseppe Espositobdb73472022-03-03 10:16:02 -05001223 GLOBAL_STATE_CODE();
Kevin Wolf38701b62017-05-04 18:52:39 +02001224 assert(bs->open_flags & BDRV_O_INACTIVE);
1225 return 0;
1226}
1227
Emanuele Giuseppe Esposito27633e72022-10-25 04:49:47 -04001228static bool bdrv_child_cb_change_aio_ctx(BdrvChild *child, AioContext *ctx,
1229 GHashTable *visited, Transaction *tran,
1230 Error **errp)
Kevin Wolf5d231842019-05-06 19:17:56 +02001231{
1232 BlockDriverState *bs = child->opaque;
Emanuele Giuseppe Esposito27633e72022-10-25 04:49:47 -04001233 return bdrv_change_aio_context(bs, ctx, visited, tran, errp);
Kevin Wolf53a7d042019-05-06 19:17:59 +02001234}
1235
Kevin Wolf0b50cc82014-04-11 21:29:52 +02001236/*
Kevin Wolf73176be2016-03-07 13:02:15 +01001237 * Returns the options and flags that a temporary snapshot should get, based on
1238 * the originally requested flags (the originally requested image will have
1239 * flags like a backing file)
Kevin Wolfb1e6fc02014-05-06 12:11:42 +02001240 */
Kevin Wolf73176be2016-03-07 13:02:15 +01001241static void bdrv_temp_snapshot_options(int *child_flags, QDict *child_options,
1242 int parent_flags, QDict *parent_options)
Kevin Wolfb1e6fc02014-05-06 12:11:42 +02001243{
Emanuele Giuseppe Espositobdb73472022-03-03 10:16:02 -05001244 GLOBAL_STATE_CODE();
Kevin Wolf73176be2016-03-07 13:02:15 +01001245 *child_flags = (parent_flags & ~BDRV_O_SNAPSHOT) | BDRV_O_TEMPORARY;
1246
1247 /* For temporary files, unconditional cache=unsafe is fine */
Kevin Wolf73176be2016-03-07 13:02:15 +01001248 qdict_set_default_str(child_options, BDRV_OPT_CACHE_DIRECT, "off");
1249 qdict_set_default_str(child_options, BDRV_OPT_CACHE_NO_FLUSH, "on");
Kevin Wolf41869042016-06-16 12:59:30 +02001250
Kevin Wolf3f486862019-04-04 17:04:43 +02001251 /* Copy the read-only and discard options from the parent */
Alberto Garciaf87a0e22016-09-15 17:53:02 +03001252 qdict_copy_default(child_options, parent_options, BDRV_OPT_READ_ONLY);
Kevin Wolf3f486862019-04-04 17:04:43 +02001253 qdict_copy_default(child_options, parent_options, BDRV_OPT_DISCARD);
Alberto Garciaf87a0e22016-09-15 17:53:02 +03001254
Kevin Wolf41869042016-06-16 12:59:30 +02001255 /* aio=native doesn't work for cache.direct=off, so disable it for the
1256 * temporary snapshot */
1257 *child_flags &= ~BDRV_O_NATIVE_AIO;
Kevin Wolfb1e6fc02014-05-06 12:11:42 +02001258}
1259
Kevin Wolfb7cfc7d2023-09-29 16:51:45 +02001260static void GRAPH_WRLOCK bdrv_backing_attach(BdrvChild *c)
Kevin Wolfdb95dbb2017-02-08 11:28:52 +01001261{
1262 BlockDriverState *parent = c->opaque;
1263 BlockDriverState *backing_hd = c->bs;
1264
Emanuele Giuseppe Espositobdb73472022-03-03 10:16:02 -05001265 GLOBAL_STATE_CODE();
Kevin Wolfdb95dbb2017-02-08 11:28:52 +01001266 assert(!parent->backing_blocker);
1267 error_setg(&parent->backing_blocker,
1268 "node is used as backing hd of '%s'",
1269 bdrv_get_device_or_node_name(parent));
1270
Max Reitzf30c66b2019-02-01 20:29:05 +01001271 bdrv_refresh_filename(backing_hd);
1272
Kevin Wolfdb95dbb2017-02-08 11:28:52 +01001273 parent->open_flags &= ~BDRV_O_NO_BACKING;
Kevin Wolfdb95dbb2017-02-08 11:28:52 +01001274
1275 bdrv_op_block_all(backing_hd, parent->backing_blocker);
1276 /* Otherwise we won't be able to commit or stream */
1277 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_COMMIT_TARGET,
1278 parent->backing_blocker);
1279 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_STREAM,
1280 parent->backing_blocker);
1281 /*
1282 * We do backup in 3 ways:
1283 * 1. drive backup
1284 * The target bs is new opened, and the source is top BDS
1285 * 2. blockdev backup
1286 * Both the source and the target are top BDSes.
1287 * 3. internal backup(used for block replication)
1288 * Both the source and the target are backing file
1289 *
1290 * In case 1 and 2, neither the source nor the target is the backing file.
1291 * In case 3, we will block the top BDS, so there is only one block job
1292 * for the top BDS and its backing chain.
1293 */
1294 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_BACKUP_SOURCE,
1295 parent->backing_blocker);
1296 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_BACKUP_TARGET,
1297 parent->backing_blocker);
Max Reitzca2f1232020-05-13 13:05:22 +02001298}
Kevin Wolfd736f112017-12-18 16:05:48 +01001299
Kevin Wolfdb95dbb2017-02-08 11:28:52 +01001300static void bdrv_backing_detach(BdrvChild *c)
1301{
1302 BlockDriverState *parent = c->opaque;
1303
Emanuele Giuseppe Espositobdb73472022-03-03 10:16:02 -05001304 GLOBAL_STATE_CODE();
Kevin Wolfdb95dbb2017-02-08 11:28:52 +01001305 assert(parent->backing_blocker);
1306 bdrv_op_unblock_all(c->bs, parent->backing_blocker);
1307 error_free(parent->backing_blocker);
1308 parent->backing_blocker = NULL;
Max Reitz48e08282020-05-13 13:05:23 +02001309}
Kevin Wolfd736f112017-12-18 16:05:48 +01001310
Kevin Wolf6858eba2017-06-29 19:32:21 +02001311static int bdrv_backing_update_filename(BdrvChild *c, BlockDriverState *base,
Peter Krempa4b028cb2023-12-05 18:14:41 +01001312 const char *filename,
1313 bool backing_mask_protocol,
1314 Error **errp)
Kevin Wolf6858eba2017-06-29 19:32:21 +02001315{
1316 BlockDriverState *parent = c->opaque;
Alberto Garciae94d3db2018-11-12 16:00:34 +02001317 bool read_only = bdrv_is_read_only(parent);
Kevin Wolf6858eba2017-06-29 19:32:21 +02001318 int ret;
Peter Krempa4b028cb2023-12-05 18:14:41 +01001319 const char *format_name;
Emanuele Giuseppe Espositobdb73472022-03-03 10:16:02 -05001320 GLOBAL_STATE_CODE();
Kevin Wolf6858eba2017-06-29 19:32:21 +02001321
Alberto Garciae94d3db2018-11-12 16:00:34 +02001322 if (read_only) {
1323 ret = bdrv_reopen_set_read_only(parent, false, errp);
Kevin Wolf61f09ce2017-09-19 16:22:54 +02001324 if (ret < 0) {
1325 return ret;
1326 }
1327 }
1328
Peter Krempa4b028cb2023-12-05 18:14:41 +01001329 if (base->drv) {
1330 /*
1331 * If the new base image doesn't have a format driver layer, which we
1332 * detect by the fact that @base is a protocol driver, we record
1333 * 'raw' as the format instead of putting the protocol name as the
1334 * backing format
1335 */
1336 if (backing_mask_protocol && base->drv->protocol_name) {
1337 format_name = "raw";
1338 } else {
1339 format_name = base->drv->format_name;
1340 }
1341 } else {
1342 format_name = "";
1343 }
1344
1345 ret = bdrv_change_backing_file(parent, filename, format_name, false);
Kevin Wolf6858eba2017-06-29 19:32:21 +02001346 if (ret < 0) {
Kevin Wolf64730692017-11-06 17:52:58 +01001347 error_setg_errno(errp, -ret, "Could not update backing file link");
Kevin Wolf6858eba2017-06-29 19:32:21 +02001348 }
1349
Alberto Garciae94d3db2018-11-12 16:00:34 +02001350 if (read_only) {
1351 bdrv_reopen_set_read_only(parent, true, NULL);
Kevin Wolf61f09ce2017-09-19 16:22:54 +02001352 }
1353
Kevin Wolf6858eba2017-06-29 19:32:21 +02001354 return ret;
1355}
1356
Max Reitzfae8bd32020-05-13 13:05:20 +02001357/*
1358 * Returns the options and flags that a generic child of a BDS should
1359 * get, based on the given options and flags for the parent BDS.
1360 */
Max Reitz00ff7ff2020-05-13 13:05:21 +02001361static void bdrv_inherited_options(BdrvChildRole role, bool parent_is_format,
1362 int *child_flags, QDict *child_options,
1363 int parent_flags, QDict *parent_options)
Max Reitzfae8bd32020-05-13 13:05:20 +02001364{
1365 int flags = parent_flags;
Emanuele Giuseppe Espositobdb73472022-03-03 10:16:02 -05001366 GLOBAL_STATE_CODE();
Max Reitzfae8bd32020-05-13 13:05:20 +02001367
1368 /*
1369 * First, decide whether to set, clear, or leave BDRV_O_PROTOCOL.
1370 * Generally, the question to answer is: Should this child be
1371 * format-probed by default?
1372 */
1373
1374 /*
1375 * Pure and non-filtered data children of non-format nodes should
1376 * be probed by default (even when the node itself has BDRV_O_PROTOCOL
1377 * set). This only affects a very limited set of drivers (namely
1378 * quorum and blkverify when this comment was written).
1379 * Force-clear BDRV_O_PROTOCOL then.
1380 */
1381 if (!parent_is_format &&
1382 (role & BDRV_CHILD_DATA) &&
1383 !(role & (BDRV_CHILD_METADATA | BDRV_CHILD_FILTERED)))
1384 {
1385 flags &= ~BDRV_O_PROTOCOL;
1386 }
1387
1388 /*
1389 * All children of format nodes (except for COW children) and all
1390 * metadata children in general should never be format-probed.
1391 * Force-set BDRV_O_PROTOCOL then.
1392 */
1393 if ((parent_is_format && !(role & BDRV_CHILD_COW)) ||
1394 (role & BDRV_CHILD_METADATA))
1395 {
1396 flags |= BDRV_O_PROTOCOL;
1397 }
1398
1399 /*
1400 * If the cache mode isn't explicitly set, inherit direct and no-flush from
1401 * the parent.
1402 */
1403 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_DIRECT);
1404 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_NO_FLUSH);
1405 qdict_copy_default(child_options, parent_options, BDRV_OPT_FORCE_SHARE);
1406
1407 if (role & BDRV_CHILD_COW) {
1408 /* backing files are opened read-only by default */
1409 qdict_set_default_str(child_options, BDRV_OPT_READ_ONLY, "on");
1410 qdict_set_default_str(child_options, BDRV_OPT_AUTO_READ_ONLY, "off");
1411 } else {
1412 /* Inherit the read-only option from the parent if it's not set */
1413 qdict_copy_default(child_options, parent_options, BDRV_OPT_READ_ONLY);
1414 qdict_copy_default(child_options, parent_options,
1415 BDRV_OPT_AUTO_READ_ONLY);
1416 }
1417
1418 /*
1419 * bdrv_co_pdiscard() respects unmap policy for the parent, so we
1420 * can default to enable it on lower layers regardless of the
1421 * parent option.
1422 */
1423 qdict_set_default_str(child_options, BDRV_OPT_DISCARD, "unmap");
1424
1425 /* Clear flags that only apply to the top layer */
1426 flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_COPY_ON_READ);
1427
1428 if (role & BDRV_CHILD_METADATA) {
1429 flags &= ~BDRV_O_NO_IO;
1430 }
1431 if (role & BDRV_CHILD_COW) {
1432 flags &= ~BDRV_O_TEMPORARY;
1433 }
1434
1435 *child_flags = flags;
1436}
1437
Kevin Wolf303de472022-12-07 14:18:35 +01001438static void GRAPH_WRLOCK bdrv_child_cb_attach(BdrvChild *child)
Max Reitzca2f1232020-05-13 13:05:22 +02001439{
1440 BlockDriverState *bs = child->opaque;
1441
Emanuele Giuseppe Esposito3f35f822022-12-07 14:18:33 +01001442 assert_bdrv_graph_writable();
Hanna Reitza2253692021-11-15 15:53:58 +01001443 QLIST_INSERT_HEAD(&bs->children, child, next);
Vladimir Sementsov-Ogievskiy5bb047472022-07-26 23:11:32 +03001444 if (bs->drv->is_filter || (child->role & BDRV_CHILD_FILTERED)) {
1445 /*
1446 * Here we handle filters and block/raw-format.c when it behave like
1447 * filter. They generally have a single PRIMARY child, which is also the
1448 * FILTERED child, and that they may have multiple more children, which
1449 * are neither PRIMARY nor FILTERED. And never we have a COW child here.
1450 * So bs->file will be the PRIMARY child, unless the PRIMARY child goes
1451 * into bs->backing on exceptional cases; and bs->backing will be
1452 * nothing else.
1453 */
1454 assert(!(child->role & BDRV_CHILD_COW));
1455 if (child->role & BDRV_CHILD_PRIMARY) {
1456 assert(child->role & BDRV_CHILD_FILTERED);
1457 assert(!bs->backing);
1458 assert(!bs->file);
Hanna Reitza2253692021-11-15 15:53:58 +01001459
Vladimir Sementsov-Ogievskiy5bb047472022-07-26 23:11:32 +03001460 if (bs->drv->filtered_child_is_backing) {
1461 bs->backing = child;
1462 } else {
1463 bs->file = child;
1464 }
1465 } else {
1466 assert(!(child->role & BDRV_CHILD_FILTERED));
1467 }
1468 } else if (child->role & BDRV_CHILD_COW) {
1469 assert(bs->drv->supports_backing);
1470 assert(!(child->role & BDRV_CHILD_PRIMARY));
1471 assert(!bs->backing);
1472 bs->backing = child;
Max Reitzca2f1232020-05-13 13:05:22 +02001473 bdrv_backing_attach(child);
Vladimir Sementsov-Ogievskiy5bb047472022-07-26 23:11:32 +03001474 } else if (child->role & BDRV_CHILD_PRIMARY) {
1475 assert(!bs->file);
1476 bs->file = child;
Max Reitzca2f1232020-05-13 13:05:22 +02001477 }
Max Reitzca2f1232020-05-13 13:05:22 +02001478}
1479
Kevin Wolf303de472022-12-07 14:18:35 +01001480static void GRAPH_WRLOCK bdrv_child_cb_detach(BdrvChild *child)
Max Reitz48e08282020-05-13 13:05:23 +02001481{
1482 BlockDriverState *bs = child->opaque;
1483
1484 if (child->role & BDRV_CHILD_COW) {
1485 bdrv_backing_detach(child);
1486 }
1487
Emanuele Giuseppe Esposito3f35f822022-12-07 14:18:33 +01001488 assert_bdrv_graph_writable();
Hanna Reitza2253692021-11-15 15:53:58 +01001489 QLIST_REMOVE(child, next);
Vladimir Sementsov-Ogievskiy5bb047472022-07-26 23:11:32 +03001490 if (child == bs->backing) {
1491 assert(child != bs->file);
1492 bs->backing = NULL;
1493 } else if (child == bs->file) {
1494 bs->file = NULL;
1495 }
Max Reitz48e08282020-05-13 13:05:23 +02001496}
1497
Max Reitz43483552020-05-13 13:05:24 +02001498static int bdrv_child_cb_update_filename(BdrvChild *c, BlockDriverState *base,
Peter Krempa4b028cb2023-12-05 18:14:41 +01001499 const char *filename,
1500 bool backing_mask_protocol,
1501 Error **errp)
Max Reitz43483552020-05-13 13:05:24 +02001502{
1503 if (c->role & BDRV_CHILD_COW) {
Peter Krempa4b028cb2023-12-05 18:14:41 +01001504 return bdrv_backing_update_filename(c, base, filename,
1505 backing_mask_protocol,
1506 errp);
Max Reitz43483552020-05-13 13:05:24 +02001507 }
1508 return 0;
1509}
1510
Vladimir Sementsov-Ogievskiyfb62b582021-05-24 13:12:56 +03001511AioContext *child_of_bds_get_parent_aio_context(BdrvChild *c)
Vladimir Sementsov-Ogievskiy3ca1f322021-04-28 18:17:33 +03001512{
1513 BlockDriverState *bs = c->opaque;
Emanuele Giuseppe Esposito384a48f2022-03-03 10:15:50 -05001514 IO_CODE();
Vladimir Sementsov-Ogievskiy3ca1f322021-04-28 18:17:33 +03001515
1516 return bdrv_get_aio_context(bs);
1517}
1518
Max Reitz43483552020-05-13 13:05:24 +02001519const BdrvChildClass child_of_bds = {
1520 .parent_is_bds = true,
1521 .get_parent_desc = bdrv_child_get_parent_desc,
1522 .inherit_options = bdrv_inherited_options,
1523 .drained_begin = bdrv_child_cb_drained_begin,
1524 .drained_poll = bdrv_child_cb_drained_poll,
1525 .drained_end = bdrv_child_cb_drained_end,
1526 .attach = bdrv_child_cb_attach,
1527 .detach = bdrv_child_cb_detach,
1528 .inactivate = bdrv_child_cb_inactivate,
Emanuele Giuseppe Esposito27633e72022-10-25 04:49:47 -04001529 .change_aio_ctx = bdrv_child_cb_change_aio_ctx,
Max Reitz43483552020-05-13 13:05:24 +02001530 .update_filename = bdrv_child_cb_update_filename,
Vladimir Sementsov-Ogievskiyfb62b582021-05-24 13:12:56 +03001531 .get_parent_aio_context = child_of_bds_get_parent_aio_context,
Max Reitz43483552020-05-13 13:05:24 +02001532};
1533
Vladimir Sementsov-Ogievskiy3ca1f322021-04-28 18:17:33 +03001534AioContext *bdrv_child_get_parent_aio_context(BdrvChild *c)
1535{
Hanna Reitzd5f8d792022-11-07 16:13:19 +01001536 IO_CODE();
Vladimir Sementsov-Ogievskiy3ca1f322021-04-28 18:17:33 +03001537 return c->klass->get_parent_aio_context(c);
1538}
1539
Kevin Wolf7b272452012-11-12 17:05:39 +01001540static int bdrv_open_flags(BlockDriverState *bs, int flags)
1541{
Kevin Wolf61de4c62016-03-18 17:46:45 +01001542 int open_flags = flags;
Emanuele Giuseppe Espositobdb73472022-03-03 10:16:02 -05001543 GLOBAL_STATE_CODE();
Kevin Wolf7b272452012-11-12 17:05:39 +01001544
1545 /*
1546 * Clear flags that are internal to the block layer before opening the
1547 * image.
1548 */
Kevin Wolf20cca272014-06-04 14:33:27 +02001549 open_flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_PROTOCOL);
Kevin Wolf7b272452012-11-12 17:05:39 +01001550
Kevin Wolf7b272452012-11-12 17:05:39 +01001551 return open_flags;
1552}
1553
Kevin Wolf91a097e2015-05-08 17:49:53 +02001554static void update_flags_from_options(int *flags, QemuOpts *opts)
1555{
Emanuele Giuseppe Espositobdb73472022-03-03 10:16:02 -05001556 GLOBAL_STATE_CODE();
1557
Alberto Garcia2a3d4332018-11-12 16:00:48 +02001558 *flags &= ~(BDRV_O_CACHE_MASK | BDRV_O_RDWR | BDRV_O_AUTO_RDONLY);
Kevin Wolf91a097e2015-05-08 17:49:53 +02001559
Alberto Garcia57f9db92018-09-06 12:37:06 +03001560 if (qemu_opt_get_bool_del(opts, BDRV_OPT_CACHE_NO_FLUSH, false)) {
Kevin Wolf91a097e2015-05-08 17:49:53 +02001561 *flags |= BDRV_O_NO_FLUSH;
1562 }
1563
Alberto Garcia57f9db92018-09-06 12:37:06 +03001564 if (qemu_opt_get_bool_del(opts, BDRV_OPT_CACHE_DIRECT, false)) {
Kevin Wolf91a097e2015-05-08 17:49:53 +02001565 *flags |= BDRV_O_NOCACHE;
1566 }
Alberto Garciaf87a0e22016-09-15 17:53:02 +03001567
Alberto Garcia57f9db92018-09-06 12:37:06 +03001568 if (!qemu_opt_get_bool_del(opts, BDRV_OPT_READ_ONLY, false)) {
Alberto Garciaf87a0e22016-09-15 17:53:02 +03001569 *flags |= BDRV_O_RDWR;
1570 }
1571
Kevin Wolfe35bdc12018-10-05 18:57:40 +02001572 if (qemu_opt_get_bool_del(opts, BDRV_OPT_AUTO_READ_ONLY, false)) {
1573 *flags |= BDRV_O_AUTO_RDONLY;
1574 }
Kevin Wolf91a097e2015-05-08 17:49:53 +02001575}
1576
1577static void update_options_from_flags(QDict *options, int flags)
1578{
Emanuele Giuseppe Espositobdb73472022-03-03 10:16:02 -05001579 GLOBAL_STATE_CODE();
Kevin Wolf91a097e2015-05-08 17:49:53 +02001580 if (!qdict_haskey(options, BDRV_OPT_CACHE_DIRECT)) {
Eric Blake46f5ac22017-04-27 16:58:17 -05001581 qdict_put_bool(options, BDRV_OPT_CACHE_DIRECT, flags & BDRV_O_NOCACHE);
Kevin Wolf91a097e2015-05-08 17:49:53 +02001582 }
1583 if (!qdict_haskey(options, BDRV_OPT_CACHE_NO_FLUSH)) {
Eric Blake46f5ac22017-04-27 16:58:17 -05001584 qdict_put_bool(options, BDRV_OPT_CACHE_NO_FLUSH,
1585 flags & BDRV_O_NO_FLUSH);
Kevin Wolf91a097e2015-05-08 17:49:53 +02001586 }
Alberto Garciaf87a0e22016-09-15 17:53:02 +03001587 if (!qdict_haskey(options, BDRV_OPT_READ_ONLY)) {
Eric Blake46f5ac22017-04-27 16:58:17 -05001588 qdict_put_bool(options, BDRV_OPT_READ_ONLY, !(flags & BDRV_O_RDWR));
Alberto Garciaf87a0e22016-09-15 17:53:02 +03001589 }
Kevin Wolfe35bdc12018-10-05 18:57:40 +02001590 if (!qdict_haskey(options, BDRV_OPT_AUTO_READ_ONLY)) {
1591 qdict_put_bool(options, BDRV_OPT_AUTO_READ_ONLY,
1592 flags & BDRV_O_AUTO_RDONLY);
1593 }
Kevin Wolf91a097e2015-05-08 17:49:53 +02001594}
1595
Kevin Wolf636ea372014-01-24 14:11:52 +01001596static void bdrv_assign_node_name(BlockDriverState *bs,
1597 const char *node_name,
1598 Error **errp)
Benoît Canet6913c0c2014-01-23 21:31:33 +01001599{
Jeff Cody15489c72015-10-12 19:36:50 -04001600 char *gen_node_name = NULL;
Emanuele Giuseppe Espositobdb73472022-03-03 10:16:02 -05001601 GLOBAL_STATE_CODE();
Benoît Canet6913c0c2014-01-23 21:31:33 +01001602
Jeff Cody15489c72015-10-12 19:36:50 -04001603 if (!node_name) {
1604 node_name = gen_node_name = id_generate(ID_BLOCK);
1605 } else if (!id_wellformed(node_name)) {
1606 /*
1607 * Check for empty string or invalid characters, but not if it is
1608 * generated (generated names use characters not available to the user)
1609 */
Connor Kuehl785ec4b2021-03-05 09:19:28 -06001610 error_setg(errp, "Invalid node-name: '%s'", node_name);
Kevin Wolf636ea372014-01-24 14:11:52 +01001611 return;
Benoît Canet6913c0c2014-01-23 21:31:33 +01001612 }
1613
Benoît Canet0c5e94e2014-02-12 17:15:07 +01001614 /* takes care of avoiding namespaces collisions */
Markus Armbruster7f06d472014-10-07 13:59:12 +02001615 if (blk_by_name(node_name)) {
Benoît Canet0c5e94e2014-02-12 17:15:07 +01001616 error_setg(errp, "node-name=%s is conflicting with a device id",
1617 node_name);
Jeff Cody15489c72015-10-12 19:36:50 -04001618 goto out;
Benoît Canet0c5e94e2014-02-12 17:15:07 +01001619 }
1620
Benoît Canet6913c0c2014-01-23 21:31:33 +01001621 /* takes care of avoiding duplicates node names */
1622 if (bdrv_find_node(node_name)) {
Connor Kuehl785ec4b2021-03-05 09:19:28 -06001623 error_setg(errp, "Duplicate nodes with node-name='%s'", node_name);
Jeff Cody15489c72015-10-12 19:36:50 -04001624 goto out;
Benoît Canet6913c0c2014-01-23 21:31:33 +01001625 }
1626
Kevin Wolf824808d2018-07-04 13:28:29 +02001627 /* Make sure that the node name isn't truncated */
1628 if (strlen(node_name) >= sizeof(bs->node_name)) {
1629 error_setg(errp, "Node name too long");
1630 goto out;
1631 }
1632
Benoît Canet6913c0c2014-01-23 21:31:33 +01001633 /* copy node name into the bs and insert it into the graph list */
1634 pstrcpy(bs->node_name, sizeof(bs->node_name), node_name);
1635 QTAILQ_INSERT_TAIL(&graph_bdrv_states, bs, node_list);
Jeff Cody15489c72015-10-12 19:36:50 -04001636out:
1637 g_free(gen_node_name);
Benoît Canet6913c0c2014-01-23 21:31:33 +01001638}
1639
Kevin Wolf1a30b0f2023-05-04 13:57:38 +02001640static int no_coroutine_fn GRAPH_UNLOCKED
1641bdrv_open_driver(BlockDriverState *bs, BlockDriver *drv, const char *node_name,
1642 QDict *options, int open_flags, Error **errp)
Kevin Wolf01a56502017-01-18 15:51:56 +01001643{
1644 Error *local_err = NULL;
Kevin Wolf0f122642018-03-28 18:29:18 +02001645 int i, ret;
Emanuele Giuseppe Espositoda359902022-03-03 10:16:11 -05001646 GLOBAL_STATE_CODE();
Kevin Wolf01a56502017-01-18 15:51:56 +01001647
1648 bdrv_assign_node_name(bs, node_name, &local_err);
1649 if (local_err) {
1650 error_propagate(errp, local_err);
1651 return -EINVAL;
1652 }
1653
1654 bs->drv = drv;
1655 bs->opaque = g_malloc0(drv->instance_size);
1656
1657 if (drv->bdrv_file_open) {
1658 assert(!drv->bdrv_needs_filename || bs->filename[0]);
1659 ret = drv->bdrv_file_open(bs, options, open_flags, &local_err);
Kevin Wolf680c7f92017-01-18 17:16:41 +01001660 } else if (drv->bdrv_open) {
Kevin Wolf01a56502017-01-18 15:51:56 +01001661 ret = drv->bdrv_open(bs, options, open_flags, &local_err);
Kevin Wolf680c7f92017-01-18 17:16:41 +01001662 } else {
1663 ret = 0;
Kevin Wolf01a56502017-01-18 15:51:56 +01001664 }
1665
1666 if (ret < 0) {
1667 if (local_err) {
1668 error_propagate(errp, local_err);
1669 } else if (bs->filename[0]) {
1670 error_setg_errno(errp, -ret, "Could not open '%s'", bs->filename);
1671 } else {
1672 error_setg_errno(errp, -ret, "Could not open image");
1673 }
Manos Pitsidianakis180ca192017-07-14 17:35:48 +03001674 goto open_failed;
Kevin Wolf01a56502017-01-18 15:51:56 +01001675 }
1676
Stefan Hajnoczie8b65352022-10-13 14:59:01 -04001677 assert(!(bs->supported_read_flags & ~BDRV_REQ_MASK));
1678 assert(!(bs->supported_write_flags & ~BDRV_REQ_MASK));
1679
1680 /*
1681 * Always allow the BDRV_REQ_REGISTERED_BUF optimization hint. This saves
1682 * drivers that pass read/write requests through to a child the trouble of
1683 * declaring support explicitly.
1684 *
1685 * Drivers must not propagate this flag accidentally when they initiate I/O
1686 * to a bounce buffer. That case should be rare though.
1687 */
1688 bs->supported_read_flags |= BDRV_REQ_REGISTERED_BUF;
1689 bs->supported_write_flags |= BDRV_REQ_REGISTERED_BUF;
1690
Emanuele Giuseppe Espositoc0579602023-01-13 21:42:03 +01001691 ret = bdrv_refresh_total_sectors(bs, bs->total_sectors);
Kevin Wolf01a56502017-01-18 15:51:56 +01001692 if (ret < 0) {
1693 error_setg_errno(errp, -ret, "Could not refresh total sector count");
Manos Pitsidianakis180ca192017-07-14 17:35:48 +03001694 return ret;
Kevin Wolf01a56502017-01-18 15:51:56 +01001695 }
1696
Kevin Wolfe19b1572023-05-04 13:57:50 +02001697 bdrv_graph_rdlock_main_loop();
Vladimir Sementsov-Ogievskiy1e4c7972021-04-28 18:17:55 +03001698 bdrv_refresh_limits(bs, NULL, &local_err);
Kevin Wolfe19b1572023-05-04 13:57:50 +02001699 bdrv_graph_rdunlock_main_loop();
1700
Kevin Wolf01a56502017-01-18 15:51:56 +01001701 if (local_err) {
1702 error_propagate(errp, local_err);
Manos Pitsidianakis180ca192017-07-14 17:35:48 +03001703 return -EINVAL;
Kevin Wolf01a56502017-01-18 15:51:56 +01001704 }
1705
1706 assert(bdrv_opt_mem_align(bs) != 0);
1707 assert(bdrv_min_mem_align(bs) != 0);
1708 assert(is_power_of_2(bs->bl.request_alignment));
1709
Kevin Wolf0f122642018-03-28 18:29:18 +02001710 for (i = 0; i < bs->quiesce_counter; i++) {
Kevin Wolf5e8ac212022-11-18 18:40:58 +01001711 if (drv->bdrv_drain_begin) {
1712 drv->bdrv_drain_begin(bs);
Kevin Wolf0f122642018-03-28 18:29:18 +02001713 }
1714 }
1715
Kevin Wolf01a56502017-01-18 15:51:56 +01001716 return 0;
Manos Pitsidianakis180ca192017-07-14 17:35:48 +03001717open_failed:
1718 bs->drv = NULL;
Kevin Wolf1f051dc2023-10-27 17:53:33 +02001719
Stefan Hajnoczi6bc30f12023-12-05 13:20:02 -05001720 bdrv_graph_wrlock();
Manos Pitsidianakis180ca192017-07-14 17:35:48 +03001721 if (bs->file != NULL) {
1722 bdrv_unref_child(bs, bs->file);
Vladimir Sementsov-Ogievskiy5bb047472022-07-26 23:11:32 +03001723 assert(!bs->file);
Manos Pitsidianakis180ca192017-07-14 17:35:48 +03001724 }
Stefan Hajnoczi6bc30f12023-12-05 13:20:02 -05001725 bdrv_graph_wrunlock();
Kevin Wolf1f051dc2023-10-27 17:53:33 +02001726
Kevin Wolf01a56502017-01-18 15:51:56 +01001727 g_free(bs->opaque);
1728 bs->opaque = NULL;
Kevin Wolf01a56502017-01-18 15:51:56 +01001729 return ret;
1730}
1731
Vladimir Sementsov-Ogievskiy621d1732021-09-20 14:55:34 +03001732/*
1733 * Create and open a block node.
1734 *
1735 * @options is a QDict of options to pass to the block drivers, or NULL for an
1736 * empty set of options. The reference to the QDict belongs to the block layer
1737 * after the call (even on failure), so if the caller intends to reuse the
1738 * dictionary, it needs to use qobject_ref() before calling bdrv_open.
1739 */
1740BlockDriverState *bdrv_new_open_driver_opts(BlockDriver *drv,
1741 const char *node_name,
1742 QDict *options, int flags,
1743 Error **errp)
Kevin Wolf680c7f92017-01-18 17:16:41 +01001744{
1745 BlockDriverState *bs;
1746 int ret;
1747
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05001748 GLOBAL_STATE_CODE();
1749
Kevin Wolf680c7f92017-01-18 17:16:41 +01001750 bs = bdrv_new();
1751 bs->open_flags = flags;
Vladimir Sementsov-Ogievskiy621d1732021-09-20 14:55:34 +03001752 bs->options = options ?: qdict_new();
1753 bs->explicit_options = qdict_clone_shallow(bs->options);
Kevin Wolf680c7f92017-01-18 17:16:41 +01001754 bs->opaque = NULL;
1755
1756 update_options_from_flags(bs->options, flags);
1757
1758 ret = bdrv_open_driver(bs, drv, node_name, bs->options, flags, errp);
1759 if (ret < 0) {
Marc-André Lureaucb3e7f02018-04-19 17:01:43 +02001760 qobject_unref(bs->explicit_options);
Manos Pitsidianakis180ca192017-07-14 17:35:48 +03001761 bs->explicit_options = NULL;
Marc-André Lureaucb3e7f02018-04-19 17:01:43 +02001762 qobject_unref(bs->options);
Manos Pitsidianakis180ca192017-07-14 17:35:48 +03001763 bs->options = NULL;
Kevin Wolf680c7f92017-01-18 17:16:41 +01001764 bdrv_unref(bs);
1765 return NULL;
1766 }
1767
1768 return bs;
1769}
1770
Vladimir Sementsov-Ogievskiy621d1732021-09-20 14:55:34 +03001771/* Create and open a block node. */
1772BlockDriverState *bdrv_new_open_driver(BlockDriver *drv, const char *node_name,
1773 int flags, Error **errp)
1774{
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05001775 GLOBAL_STATE_CODE();
Vladimir Sementsov-Ogievskiy621d1732021-09-20 14:55:34 +03001776 return bdrv_new_open_driver_opts(drv, node_name, NULL, flags, errp);
1777}
1778
Kevin Wolfc5f30142016-10-06 11:33:17 +02001779QemuOptsList bdrv_runtime_opts = {
Kevin Wolf18edf282015-04-07 17:12:56 +02001780 .name = "bdrv_common",
1781 .head = QTAILQ_HEAD_INITIALIZER(bdrv_runtime_opts.head),
1782 .desc = {
1783 {
1784 .name = "node-name",
1785 .type = QEMU_OPT_STRING,
1786 .help = "Node name of the block device node",
1787 },
Kevin Wolf62392eb2015-04-24 16:38:02 +02001788 {
1789 .name = "driver",
1790 .type = QEMU_OPT_STRING,
1791 .help = "Block driver to use for the node",
1792 },
Kevin Wolf91a097e2015-05-08 17:49:53 +02001793 {
Kevin Wolf91a097e2015-05-08 17:49:53 +02001794 .name = BDRV_OPT_CACHE_DIRECT,
1795 .type = QEMU_OPT_BOOL,
1796 .help = "Bypass software writeback cache on the host",
1797 },
1798 {
1799 .name = BDRV_OPT_CACHE_NO_FLUSH,
1800 .type = QEMU_OPT_BOOL,
1801 .help = "Ignore flush requests",
1802 },
Alberto Garciaf87a0e22016-09-15 17:53:02 +03001803 {
1804 .name = BDRV_OPT_READ_ONLY,
1805 .type = QEMU_OPT_BOOL,
1806 .help = "Node is opened in read-only mode",
1807 },
Kevin Wolf692e01a2016-09-12 21:00:41 +02001808 {
Kevin Wolfe35bdc12018-10-05 18:57:40 +02001809 .name = BDRV_OPT_AUTO_READ_ONLY,
1810 .type = QEMU_OPT_BOOL,
1811 .help = "Node can become read-only if opening read-write fails",
1812 },
1813 {
Kevin Wolf692e01a2016-09-12 21:00:41 +02001814 .name = "detect-zeroes",
1815 .type = QEMU_OPT_STRING,
1816 .help = "try to optimize zero writes (off, on, unmap)",
1817 },
Kevin Wolf818584a2016-09-12 18:03:18 +02001818 {
Alberto Garcia415bbca2018-10-03 13:23:13 +03001819 .name = BDRV_OPT_DISCARD,
Kevin Wolf818584a2016-09-12 18:03:18 +02001820 .type = QEMU_OPT_STRING,
1821 .help = "discard operation (ignore/off, unmap/on)",
1822 },
Fam Zheng5a9347c2017-05-03 00:35:37 +08001823 {
1824 .name = BDRV_OPT_FORCE_SHARE,
1825 .type = QEMU_OPT_BOOL,
1826 .help = "always accept other writers (default: off)",
1827 },
Kevin Wolf18edf282015-04-07 17:12:56 +02001828 { /* end of list */ }
1829 },
1830};
1831
Maxim Levitsky5a5e7f82020-03-26 03:12:18 +02001832QemuOptsList bdrv_create_opts_simple = {
1833 .name = "simple-create-opts",
1834 .head = QTAILQ_HEAD_INITIALIZER(bdrv_create_opts_simple.head),
Max Reitzfd171462020-01-22 17:45:29 +01001835 .desc = {
1836 {
1837 .name = BLOCK_OPT_SIZE,
1838 .type = QEMU_OPT_SIZE,
1839 .help = "Virtual disk size"
1840 },
1841 {
1842 .name = BLOCK_OPT_PREALLOC,
1843 .type = QEMU_OPT_STRING,
1844 .help = "Preallocation mode (allowed values: off)"
1845 },
1846 { /* end of list */ }
1847 }
1848};
1849
Kevin Wolfb6ce07a2010-04-12 16:37:13 +02001850/*
Kevin Wolf57915332010-04-14 15:24:50 +02001851 * Common part for opening disk images and files
Kevin Wolfb6ad4912013-03-15 10:35:04 +01001852 *
1853 * Removes all processed options from *options.
Kevin Wolf57915332010-04-14 15:24:50 +02001854 */
Kevin Wolf5696c6e2017-02-17 18:39:24 +01001855static int bdrv_open_common(BlockDriverState *bs, BlockBackend *file,
Kevin Wolf82dc8b42016-01-11 19:07:50 +01001856 QDict *options, Error **errp)
Kevin Wolf57915332010-04-14 15:24:50 +02001857{
1858 int ret, open_flags;
Kevin Wolf035fccd2013-04-09 14:34:19 +02001859 const char *filename;
Kevin Wolf62392eb2015-04-24 16:38:02 +02001860 const char *driver_name = NULL;
Benoît Canet6913c0c2014-01-23 21:31:33 +01001861 const char *node_name = NULL;
Kevin Wolf818584a2016-09-12 18:03:18 +02001862 const char *discard;
Kevin Wolf18edf282015-04-07 17:12:56 +02001863 QemuOpts *opts;
Kevin Wolf62392eb2015-04-24 16:38:02 +02001864 BlockDriver *drv;
Max Reitz34b5d2c2013-09-05 14:45:29 +02001865 Error *local_err = NULL;
Vladimir Sementsov-Ogievskiy307261b2021-05-27 18:40:54 +03001866 bool ro;
Kevin Wolf57915332010-04-14 15:24:50 +02001867
Kevin Wolf1f051dc2023-10-27 17:53:33 +02001868 GLOBAL_STATE_CODE();
1869
1870 bdrv_graph_rdlock_main_loop();
Paolo Bonzini64058752012-05-08 16:51:49 +02001871 assert(bs->file == NULL);
Kevin Wolf707ff822013-03-06 12:20:31 +01001872 assert(options != NULL && bs->options != options);
Kevin Wolf1f051dc2023-10-27 17:53:33 +02001873 bdrv_graph_rdunlock_main_loop();
Kevin Wolf57915332010-04-14 15:24:50 +02001874
Kevin Wolf62392eb2015-04-24 16:38:02 +02001875 opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
Markus Armbrusteraf175e82020-07-07 18:06:03 +02001876 if (!qemu_opts_absorb_qdict(opts, options, errp)) {
Kevin Wolf62392eb2015-04-24 16:38:02 +02001877 ret = -EINVAL;
1878 goto fail_opts;
1879 }
1880
Alberto Garcia9b7e8692016-09-15 17:53:01 +03001881 update_flags_from_options(&bs->open_flags, opts);
1882
Kevin Wolf62392eb2015-04-24 16:38:02 +02001883 driver_name = qemu_opt_get(opts, "driver");
1884 drv = bdrv_find_format(driver_name);
1885 assert(drv != NULL);
1886
Fam Zheng5a9347c2017-05-03 00:35:37 +08001887 bs->force_share = qemu_opt_get_bool(opts, BDRV_OPT_FORCE_SHARE, false);
1888
1889 if (bs->force_share && (bs->open_flags & BDRV_O_RDWR)) {
1890 error_setg(errp,
1891 BDRV_OPT_FORCE_SHARE
1892 "=on can only be used with read-only images");
1893 ret = -EINVAL;
1894 goto fail_opts;
1895 }
1896
Kevin Wolf45673672013-04-22 17:48:40 +02001897 if (file != NULL) {
Kevin Wolfb7cfc7d2023-09-29 16:51:45 +02001898 bdrv_graph_rdlock_main_loop();
Max Reitzf30c66b2019-02-01 20:29:05 +01001899 bdrv_refresh_filename(blk_bs(file));
Kevin Wolfb7cfc7d2023-09-29 16:51:45 +02001900 bdrv_graph_rdunlock_main_loop();
1901
Kevin Wolf5696c6e2017-02-17 18:39:24 +01001902 filename = blk_bs(file)->filename;
Kevin Wolf45673672013-04-22 17:48:40 +02001903 } else {
Markus Armbruster129c7d12017-03-30 19:43:12 +02001904 /*
1905 * Caution: while qdict_get_try_str() is fine, getting
1906 * non-string types would require more care. When @options
1907 * come from -blockdev or blockdev_add, its members are typed
1908 * according to the QAPI schema, but when they come from
1909 * -drive, they're all QString.
1910 */
Kevin Wolf45673672013-04-22 17:48:40 +02001911 filename = qdict_get_try_str(options, "filename");
1912 }
1913
Max Reitz4a008242017-04-13 18:06:24 +02001914 if (drv->bdrv_needs_filename && (!filename || !filename[0])) {
Kevin Wolf765003d2014-02-03 14:49:42 +01001915 error_setg(errp, "The '%s' block driver requires a file name",
1916 drv->format_name);
Kevin Wolf18edf282015-04-07 17:12:56 +02001917 ret = -EINVAL;
1918 goto fail_opts;
1919 }
1920
Kevin Wolf82dc8b42016-01-11 19:07:50 +01001921 trace_bdrv_open_common(bs, filename ?: "", bs->open_flags,
1922 drv->format_name);
Kevin Wolf62392eb2015-04-24 16:38:02 +02001923
Vladimir Sementsov-Ogievskiy307261b2021-05-27 18:40:54 +03001924 ro = bdrv_is_read_only(bs);
1925
1926 if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv, ro)) {
1927 if (!ro && bdrv_is_whitelisted(drv, true)) {
Kevin Wolf018f9de2023-09-29 16:51:53 +02001928 bdrv_graph_rdlock_main_loop();
Kevin Wolf8be25de2019-01-22 13:15:31 +01001929 ret = bdrv_apply_auto_read_only(bs, NULL, NULL);
Kevin Wolf018f9de2023-09-29 16:51:53 +02001930 bdrv_graph_rdunlock_main_loop();
Kevin Wolf8be25de2019-01-22 13:15:31 +01001931 } else {
1932 ret = -ENOTSUP;
1933 }
1934 if (ret < 0) {
1935 error_setg(errp,
Vladimir Sementsov-Ogievskiy307261b2021-05-27 18:40:54 +03001936 !ro && bdrv_is_whitelisted(drv, true)
Kevin Wolf8be25de2019-01-22 13:15:31 +01001937 ? "Driver '%s' can only be used for read-only devices"
1938 : "Driver '%s' is not whitelisted",
1939 drv->format_name);
1940 goto fail_opts;
1941 }
Fam Zhengb64ec4e2013-05-29 19:35:40 +08001942 }
Kevin Wolf57915332010-04-14 15:24:50 +02001943
Paolo Bonzinid3faa132017-06-05 14:38:50 +02001944 /* bdrv_new() and bdrv_close() make it so */
Stefan Hajnoczid73415a2020-09-23 11:56:46 +01001945 assert(qatomic_read(&bs->copy_on_read) == 0);
Paolo Bonzinid3faa132017-06-05 14:38:50 +02001946
Kevin Wolf82dc8b42016-01-11 19:07:50 +01001947 if (bs->open_flags & BDRV_O_COPY_ON_READ) {
Vladimir Sementsov-Ogievskiy307261b2021-05-27 18:40:54 +03001948 if (!ro) {
Kevin Wolf0ebd24e2013-09-19 15:12:18 +02001949 bdrv_enable_copy_on_read(bs);
1950 } else {
1951 error_setg(errp, "Can't use copy-on-read on read-only device");
Kevin Wolf18edf282015-04-07 17:12:56 +02001952 ret = -EINVAL;
1953 goto fail_opts;
Kevin Wolf0ebd24e2013-09-19 15:12:18 +02001954 }
Stefan Hajnoczi53fec9d2011-11-28 16:08:47 +00001955 }
1956
Alberto Garcia415bbca2018-10-03 13:23:13 +03001957 discard = qemu_opt_get(opts, BDRV_OPT_DISCARD);
Kevin Wolf818584a2016-09-12 18:03:18 +02001958 if (discard != NULL) {
1959 if (bdrv_parse_discard_flags(discard, &bs->open_flags) != 0) {
1960 error_setg(errp, "Invalid discard option");
1961 ret = -EINVAL;
1962 goto fail_opts;
1963 }
1964 }
1965
Alberto Garcia543770b2018-09-06 12:37:09 +03001966 bs->detect_zeroes =
1967 bdrv_parse_detect_zeroes(opts, bs->open_flags, &local_err);
1968 if (local_err) {
1969 error_propagate(errp, local_err);
1970 ret = -EINVAL;
1971 goto fail_opts;
Kevin Wolf692e01a2016-09-12 21:00:41 +02001972 }
1973
Kevin Wolfc2ad1b02013-03-18 16:40:51 +01001974 if (filename != NULL) {
1975 pstrcpy(bs->filename, sizeof(bs->filename), filename);
1976 } else {
1977 bs->filename[0] = '\0';
1978 }
Max Reitz91af7012014-07-18 20:24:56 +02001979 pstrcpy(bs->exact_filename, sizeof(bs->exact_filename), bs->filename);
Kevin Wolf57915332010-04-14 15:24:50 +02001980
Kevin Wolf66f82ce2010-04-14 14:17:38 +02001981 /* Open the image, either directly or using a protocol */
Kevin Wolf82dc8b42016-01-11 19:07:50 +01001982 open_flags = bdrv_open_flags(bs, bs->open_flags);
Kevin Wolf01a56502017-01-18 15:51:56 +01001983 node_name = qemu_opt_get(opts, "node-name");
Kevin Wolf66f82ce2010-04-14 14:17:38 +02001984
Kevin Wolf01a56502017-01-18 15:51:56 +01001985 assert(!drv->bdrv_file_open || file == NULL);
1986 ret = bdrv_open_driver(bs, drv, node_name, options, open_flags, errp);
Kevin Wolf57915332010-04-14 15:24:50 +02001987 if (ret < 0) {
Kevin Wolf01a56502017-01-18 15:51:56 +01001988 goto fail_opts;
Kevin Wolf57915332010-04-14 15:24:50 +02001989 }
1990
Kevin Wolf18edf282015-04-07 17:12:56 +02001991 qemu_opts_del(opts);
Kevin Wolf57915332010-04-14 15:24:50 +02001992 return 0;
1993
Kevin Wolf18edf282015-04-07 17:12:56 +02001994fail_opts:
1995 qemu_opts_del(opts);
Kevin Wolf57915332010-04-14 15:24:50 +02001996 return ret;
1997}
1998
Kevin Wolf5e5c4f62014-05-26 11:45:08 +02001999static QDict *parse_json_filename(const char *filename, Error **errp)
2000{
2001 QObject *options_obj;
2002 QDict *options;
2003 int ret;
Emanuele Giuseppe Espositobdb73472022-03-03 10:16:02 -05002004 GLOBAL_STATE_CODE();
Kevin Wolf5e5c4f62014-05-26 11:45:08 +02002005
2006 ret = strstart(filename, "json:", &filename);
2007 assert(ret);
2008
Markus Armbruster5577fff2017-02-28 22:26:59 +01002009 options_obj = qobject_from_json(filename, errp);
Kevin Wolf5e5c4f62014-05-26 11:45:08 +02002010 if (!options_obj) {
Markus Armbruster5577fff2017-02-28 22:26:59 +01002011 error_prepend(errp, "Could not parse the JSON options: ");
Kevin Wolf5e5c4f62014-05-26 11:45:08 +02002012 return NULL;
2013 }
2014
Max Reitz7dc847e2018-02-24 16:40:29 +01002015 options = qobject_to(QDict, options_obj);
Markus Armbrusterca6b6e12017-02-17 21:38:18 +01002016 if (!options) {
Marc-André Lureaucb3e7f02018-04-19 17:01:43 +02002017 qobject_unref(options_obj);
Kevin Wolf5e5c4f62014-05-26 11:45:08 +02002018 error_setg(errp, "Invalid JSON object given");
2019 return NULL;
2020 }
2021
Kevin Wolf5e5c4f62014-05-26 11:45:08 +02002022 qdict_flatten(options);
2023
2024 return options;
2025}
2026
Kevin Wolfde3b53f2015-10-29 15:24:41 +01002027static void parse_json_protocol(QDict *options, const char **pfilename,
2028 Error **errp)
2029{
2030 QDict *json_options;
2031 Error *local_err = NULL;
Emanuele Giuseppe Espositobdb73472022-03-03 10:16:02 -05002032 GLOBAL_STATE_CODE();
Kevin Wolfde3b53f2015-10-29 15:24:41 +01002033
2034 /* Parse json: pseudo-protocol */
2035 if (!*pfilename || !g_str_has_prefix(*pfilename, "json:")) {
2036 return;
2037 }
2038
2039 json_options = parse_json_filename(*pfilename, &local_err);
2040 if (local_err) {
2041 error_propagate(errp, local_err);
2042 return;
2043 }
2044
2045 /* Options given in the filename have lower priority than options
2046 * specified directly */
2047 qdict_join(options, json_options, false);
Marc-André Lureaucb3e7f02018-04-19 17:01:43 +02002048 qobject_unref(json_options);
Kevin Wolfde3b53f2015-10-29 15:24:41 +01002049 *pfilename = NULL;
2050}
2051
Kevin Wolf57915332010-04-14 15:24:50 +02002052/*
Kevin Wolff54120f2014-05-26 11:09:59 +02002053 * Fills in default options for opening images and converts the legacy
2054 * filename/flags pair to option QDict entries.
Max Reitz53a29512015-03-19 14:53:16 -04002055 * The BDRV_O_PROTOCOL flag in *flags will be set or cleared accordingly if a
2056 * block driver has been specified explicitly.
Kevin Wolff54120f2014-05-26 11:09:59 +02002057 */
Kevin Wolfde3b53f2015-10-29 15:24:41 +01002058static int bdrv_fill_options(QDict **options, const char *filename,
Max Reitz053e1572015-08-26 19:47:51 +02002059 int *flags, Error **errp)
Kevin Wolff54120f2014-05-26 11:09:59 +02002060{
2061 const char *drvname;
Max Reitz53a29512015-03-19 14:53:16 -04002062 bool protocol = *flags & BDRV_O_PROTOCOL;
Kevin Wolff54120f2014-05-26 11:09:59 +02002063 bool parse_filename = false;
Max Reitz053e1572015-08-26 19:47:51 +02002064 BlockDriver *drv = NULL;
Kevin Wolff54120f2014-05-26 11:09:59 +02002065 Error *local_err = NULL;
Kevin Wolff54120f2014-05-26 11:09:59 +02002066
Emanuele Giuseppe Espositoda359902022-03-03 10:16:11 -05002067 GLOBAL_STATE_CODE();
2068
Markus Armbruster129c7d12017-03-30 19:43:12 +02002069 /*
2070 * Caution: while qdict_get_try_str() is fine, getting non-string
2071 * types would require more care. When @options come from
2072 * -blockdev or blockdev_add, its members are typed according to
2073 * the QAPI schema, but when they come from -drive, they're all
2074 * QString.
2075 */
Max Reitz53a29512015-03-19 14:53:16 -04002076 drvname = qdict_get_try_str(*options, "driver");
Max Reitz053e1572015-08-26 19:47:51 +02002077 if (drvname) {
2078 drv = bdrv_find_format(drvname);
2079 if (!drv) {
2080 error_setg(errp, "Unknown driver '%s'", drvname);
2081 return -ENOENT;
2082 }
2083 /* If the user has explicitly specified the driver, this choice should
2084 * override the BDRV_O_PROTOCOL flag */
2085 protocol = drv->bdrv_file_open;
Max Reitz53a29512015-03-19 14:53:16 -04002086 }
2087
2088 if (protocol) {
2089 *flags |= BDRV_O_PROTOCOL;
2090 } else {
2091 *flags &= ~BDRV_O_PROTOCOL;
2092 }
2093
Kevin Wolf91a097e2015-05-08 17:49:53 +02002094 /* Translate cache options from flags into options */
2095 update_options_from_flags(*options, *flags);
2096
Kevin Wolff54120f2014-05-26 11:09:59 +02002097 /* Fetch the file name from the options QDict if necessary */
Kevin Wolf17b005f2014-05-27 10:50:29 +02002098 if (protocol && filename) {
Kevin Wolff54120f2014-05-26 11:09:59 +02002099 if (!qdict_haskey(*options, "filename")) {
Eric Blake46f5ac22017-04-27 16:58:17 -05002100 qdict_put_str(*options, "filename", filename);
Kevin Wolff54120f2014-05-26 11:09:59 +02002101 parse_filename = true;
2102 } else {
2103 error_setg(errp, "Can't specify 'file' and 'filename' options at "
2104 "the same time");
2105 return -EINVAL;
2106 }
2107 }
2108
2109 /* Find the right block driver */
Markus Armbruster129c7d12017-03-30 19:43:12 +02002110 /* See cautionary note on accessing @options above */
Kevin Wolff54120f2014-05-26 11:09:59 +02002111 filename = qdict_get_try_str(*options, "filename");
Kevin Wolff54120f2014-05-26 11:09:59 +02002112
Max Reitz053e1572015-08-26 19:47:51 +02002113 if (!drvname && protocol) {
2114 if (filename) {
2115 drv = bdrv_find_protocol(filename, parse_filename, errp);
2116 if (!drv) {
Kevin Wolff54120f2014-05-26 11:09:59 +02002117 return -EINVAL;
2118 }
Max Reitz053e1572015-08-26 19:47:51 +02002119
2120 drvname = drv->format_name;
Eric Blake46f5ac22017-04-27 16:58:17 -05002121 qdict_put_str(*options, "driver", drvname);
Max Reitz053e1572015-08-26 19:47:51 +02002122 } else {
2123 error_setg(errp, "Must specify either driver or file");
2124 return -EINVAL;
Kevin Wolff54120f2014-05-26 11:09:59 +02002125 }
2126 }
2127
Kevin Wolf17b005f2014-05-27 10:50:29 +02002128 assert(drv || !protocol);
Kevin Wolff54120f2014-05-26 11:09:59 +02002129
2130 /* Driver-specific filename parsing */
Kevin Wolf17b005f2014-05-27 10:50:29 +02002131 if (drv && drv->bdrv_parse_filename && parse_filename) {
Kevin Wolff54120f2014-05-26 11:09:59 +02002132 drv->bdrv_parse_filename(filename, *options, &local_err);
2133 if (local_err) {
2134 error_propagate(errp, local_err);
2135 return -EINVAL;
2136 }
2137
2138 if (!drv->bdrv_needs_filename) {
2139 qdict_del(*options, "filename");
2140 }
2141 }
2142
2143 return 0;
2144}
2145
Kevin Wolf148eb132017-09-14 14:32:04 +02002146typedef struct BlockReopenQueueEntry {
2147 bool prepared;
2148 BDRVReopenState state;
Vladimir Sementsov-Ogievskiy859443b2019-09-27 15:23:47 +03002149 QTAILQ_ENTRY(BlockReopenQueueEntry) entry;
Kevin Wolf148eb132017-09-14 14:32:04 +02002150} BlockReopenQueueEntry;
2151
2152/*
2153 * Return the flags that @bs will have after the reopens in @q have
2154 * successfully completed. If @q is NULL (or @bs is not contained in @q),
2155 * return the current flags.
2156 */
2157static int bdrv_reopen_get_flags(BlockReopenQueue *q, BlockDriverState *bs)
2158{
2159 BlockReopenQueueEntry *entry;
2160
2161 if (q != NULL) {
Vladimir Sementsov-Ogievskiy859443b2019-09-27 15:23:47 +03002162 QTAILQ_FOREACH(entry, q, entry) {
Kevin Wolf148eb132017-09-14 14:32:04 +02002163 if (entry->state.bs == bs) {
2164 return entry->state.flags;
2165 }
2166 }
2167 }
2168
2169 return bs->open_flags;
2170}
2171
2172/* Returns whether the image file can be written to after the reopen queue @q
2173 * has been successfully applied, or right now if @q is NULL. */
Max Reitzcc022142018-06-06 21:37:00 +02002174static bool bdrv_is_writable_after_reopen(BlockDriverState *bs,
2175 BlockReopenQueue *q)
Kevin Wolf148eb132017-09-14 14:32:04 +02002176{
2177 int flags = bdrv_reopen_get_flags(q, bs);
2178
2179 return (flags & (BDRV_O_RDWR | BDRV_O_INACTIVE)) == BDRV_O_RDWR;
2180}
2181
Max Reitzcc022142018-06-06 21:37:00 +02002182/*
2183 * Return whether the BDS can be written to. This is not necessarily
2184 * the same as !bdrv_is_read_only(bs), as inactivated images may not
2185 * be written to but do not count as read-only images.
2186 */
2187bool bdrv_is_writable(BlockDriverState *bs)
2188{
Emanuele Giuseppe Esposito384a48f2022-03-03 10:15:50 -05002189 IO_CODE();
Max Reitzcc022142018-06-06 21:37:00 +02002190 return bdrv_is_writable_after_reopen(bs, NULL);
2191}
2192
Vladimir Sementsov-Ogievskiy3bf416b2021-04-28 18:17:37 +03002193static char *bdrv_child_user_desc(BdrvChild *c)
2194{
Emanuele Giuseppe Espositof0c28322022-03-03 10:16:13 -05002195 GLOBAL_STATE_CODE();
Vladimir Sementsov-Ogievskiyda261b62021-06-01 10:52:17 +03002196 return c->klass->get_parent_desc(c);
Vladimir Sementsov-Ogievskiy3bf416b2021-04-28 18:17:37 +03002197}
2198
Vladimir Sementsov-Ogievskiy30ebb9a2021-06-01 10:52:18 +03002199/*
2200 * Check that @a allows everything that @b needs. @a and @b must reference same
2201 * child node.
2202 */
Vladimir Sementsov-Ogievskiy3bf416b2021-04-28 18:17:37 +03002203static bool bdrv_a_allow_b(BdrvChild *a, BdrvChild *b, Error **errp)
2204{
Vladimir Sementsov-Ogievskiy30ebb9a2021-06-01 10:52:18 +03002205 const char *child_bs_name;
2206 g_autofree char *a_user = NULL;
2207 g_autofree char *b_user = NULL;
2208 g_autofree char *perms = NULL;
2209
2210 assert(a->bs);
2211 assert(a->bs == b->bs);
Emanuele Giuseppe Esposito862fded2022-03-03 10:15:55 -05002212 GLOBAL_STATE_CODE();
Vladimir Sementsov-Ogievskiy3bf416b2021-04-28 18:17:37 +03002213
2214 if ((b->perm & a->shared_perm) == b->perm) {
2215 return true;
2216 }
2217
Vladimir Sementsov-Ogievskiy30ebb9a2021-06-01 10:52:18 +03002218 child_bs_name = bdrv_get_node_name(b->bs);
2219 a_user = bdrv_child_user_desc(a);
2220 b_user = bdrv_child_user_desc(b);
2221 perms = bdrv_perm_names(b->perm & ~a->shared_perm);
2222
2223 error_setg(errp, "Permission conflict on node '%s': permissions '%s' are "
2224 "both required by %s (uses node '%s' as '%s' child) and "
2225 "unshared by %s (uses node '%s' as '%s' child).",
2226 child_bs_name, perms,
2227 b_user, child_bs_name, b->name,
2228 a_user, child_bs_name, a->name);
Vladimir Sementsov-Ogievskiy3bf416b2021-04-28 18:17:37 +03002229
2230 return false;
2231}
2232
Kevin Wolf3804e3c2023-09-11 11:46:12 +02002233static bool GRAPH_RDLOCK
2234bdrv_parent_perms_conflict(BlockDriverState *bs, Error **errp)
Vladimir Sementsov-Ogievskiy3bf416b2021-04-28 18:17:37 +03002235{
2236 BdrvChild *a, *b;
Emanuele Giuseppe Esposito862fded2022-03-03 10:15:55 -05002237 GLOBAL_STATE_CODE();
Vladimir Sementsov-Ogievskiy3bf416b2021-04-28 18:17:37 +03002238
2239 /*
2240 * During the loop we'll look at each pair twice. That's correct because
2241 * bdrv_a_allow_b() is asymmetric and we should check each pair in both
2242 * directions.
2243 */
2244 QLIST_FOREACH(a, &bs->parents, next_parent) {
2245 QLIST_FOREACH(b, &bs->parents, next_parent) {
Vladimir Sementsov-Ogievskiy9397c142021-04-28 18:17:53 +03002246 if (a == b) {
Vladimir Sementsov-Ogievskiy3bf416b2021-04-28 18:17:37 +03002247 continue;
2248 }
2249
2250 if (!bdrv_a_allow_b(a, b, errp)) {
2251 return true;
2252 }
2253 }
2254 }
2255
2256 return false;
2257}
2258
Kevin Wolfc629b6d2023-09-11 11:46:14 +02002259static void GRAPH_RDLOCK
2260bdrv_child_perm(BlockDriverState *bs, BlockDriverState *child_bs,
2261 BdrvChild *c, BdrvChildRole role,
2262 BlockReopenQueue *reopen_queue,
2263 uint64_t parent_perm, uint64_t parent_shared,
2264 uint64_t *nperm, uint64_t *nshared)
Fam Zhengffd1a5a2017-05-03 00:35:38 +08002265{
Alberto Garcia0b3ca762019-04-04 14:29:53 +03002266 assert(bs->drv && bs->drv->bdrv_child_perm);
Emanuele Giuseppe Espositoda359902022-03-03 10:16:11 -05002267 GLOBAL_STATE_CODE();
Max Reitze5d8a402020-05-13 13:05:44 +02002268 bs->drv->bdrv_child_perm(bs, c, role, reopen_queue,
Alberto Garcia0b3ca762019-04-04 14:29:53 +03002269 parent_perm, parent_shared,
2270 nperm, nshared);
Kevin Wolfe0995dc2017-09-14 12:47:11 +02002271 /* TODO Take force_share from reopen_queue */
Fam Zhengffd1a5a2017-05-03 00:35:38 +08002272 if (child_bs && child_bs->force_share) {
2273 *nshared = BLK_PERM_ALL;
2274 }
2275}
2276
Vladimir Sementsov-Ogievskiybd57f8f2021-04-28 18:17:41 +03002277/*
2278 * Adds the whole subtree of @bs (including @bs itself) to the @list (except for
2279 * nodes that are already in the @list, of course) so that final list is
2280 * topologically sorted. Return the result (GSList @list object is updated, so
2281 * don't use old reference after function call).
2282 *
2283 * On function start @list must be already topologically sorted and for any node
2284 * in the @list the whole subtree of the node must be in the @list as well. The
2285 * simplest way to satisfy this criteria: use only result of
2286 * bdrv_topological_dfs() or NULL as @list parameter.
2287 */
Kevin Wolf3804e3c2023-09-11 11:46:12 +02002288static GSList * GRAPH_RDLOCK
2289bdrv_topological_dfs(GSList *list, GHashTable *found, BlockDriverState *bs)
Vladimir Sementsov-Ogievskiybd57f8f2021-04-28 18:17:41 +03002290{
2291 BdrvChild *child;
2292 g_autoptr(GHashTable) local_found = NULL;
2293
Emanuele Giuseppe Espositobdb73472022-03-03 10:16:02 -05002294 GLOBAL_STATE_CODE();
2295
Vladimir Sementsov-Ogievskiybd57f8f2021-04-28 18:17:41 +03002296 if (!found) {
2297 assert(!list);
2298 found = local_found = g_hash_table_new(NULL, NULL);
2299 }
2300
2301 if (g_hash_table_contains(found, bs)) {
2302 return list;
2303 }
2304 g_hash_table_add(found, bs);
2305
2306 QLIST_FOREACH(child, &bs->children, next) {
2307 list = bdrv_topological_dfs(list, found, child->bs);
2308 }
2309
2310 return g_slist_prepend(list, bs);
2311}
2312
Vladimir Sementsov-Ogievskiyecb776b2021-04-28 18:18:02 +03002313typedef struct BdrvChildSetPermState {
2314 BdrvChild *child;
2315 uint64_t old_perm;
2316 uint64_t old_shared_perm;
2317} BdrvChildSetPermState;
Vladimir Sementsov-Ogievskiyb0defa82021-04-28 18:17:38 +03002318
2319static void bdrv_child_set_perm_abort(void *opaque)
2320{
Vladimir Sementsov-Ogievskiyecb776b2021-04-28 18:18:02 +03002321 BdrvChildSetPermState *s = opaque;
2322
Emanuele Giuseppe Esposito862fded2022-03-03 10:15:55 -05002323 GLOBAL_STATE_CODE();
2324
Vladimir Sementsov-Ogievskiyecb776b2021-04-28 18:18:02 +03002325 s->child->perm = s->old_perm;
2326 s->child->shared_perm = s->old_shared_perm;
Vladimir Sementsov-Ogievskiyb0defa82021-04-28 18:17:38 +03002327}
2328
2329static TransactionActionDrv bdrv_child_set_pem_drv = {
2330 .abort = bdrv_child_set_perm_abort,
Vladimir Sementsov-Ogievskiyecb776b2021-04-28 18:18:02 +03002331 .clean = g_free,
Vladimir Sementsov-Ogievskiyb0defa82021-04-28 18:17:38 +03002332};
2333
Vladimir Sementsov-Ogievskiyecb776b2021-04-28 18:18:02 +03002334static void bdrv_child_set_perm(BdrvChild *c, uint64_t perm,
2335 uint64_t shared, Transaction *tran)
Vladimir Sementsov-Ogievskiyb0defa82021-04-28 18:17:38 +03002336{
Vladimir Sementsov-Ogievskiyecb776b2021-04-28 18:18:02 +03002337 BdrvChildSetPermState *s = g_new(BdrvChildSetPermState, 1);
Emanuele Giuseppe Esposito862fded2022-03-03 10:15:55 -05002338 GLOBAL_STATE_CODE();
Vladimir Sementsov-Ogievskiyecb776b2021-04-28 18:18:02 +03002339
2340 *s = (BdrvChildSetPermState) {
2341 .child = c,
2342 .old_perm = c->perm,
2343 .old_shared_perm = c->shared_perm,
2344 };
Vladimir Sementsov-Ogievskiyb0defa82021-04-28 18:17:38 +03002345
2346 c->perm = perm;
2347 c->shared_perm = shared;
2348
Vladimir Sementsov-Ogievskiyecb776b2021-04-28 18:18:02 +03002349 tran_add(tran, &bdrv_child_set_pem_drv, s);
Vladimir Sementsov-Ogievskiyb0defa82021-04-28 18:17:38 +03002350}
2351
Kevin Wolfbce73bc2023-09-11 11:46:13 +02002352static void GRAPH_RDLOCK bdrv_drv_set_perm_commit(void *opaque)
Vladimir Sementsov-Ogievskiy2513ef52021-04-28 18:17:42 +03002353{
2354 BlockDriverState *bs = opaque;
2355 uint64_t cumulative_perms, cumulative_shared_perms;
Emanuele Giuseppe Espositoda359902022-03-03 10:16:11 -05002356 GLOBAL_STATE_CODE();
Vladimir Sementsov-Ogievskiy2513ef52021-04-28 18:17:42 +03002357
2358 if (bs->drv->bdrv_set_perm) {
2359 bdrv_get_cumulative_perm(bs, &cumulative_perms,
2360 &cumulative_shared_perms);
2361 bs->drv->bdrv_set_perm(bs, cumulative_perms, cumulative_shared_perms);
2362 }
2363}
2364
Kevin Wolfbce73bc2023-09-11 11:46:13 +02002365static void GRAPH_RDLOCK bdrv_drv_set_perm_abort(void *opaque)
Vladimir Sementsov-Ogievskiy2513ef52021-04-28 18:17:42 +03002366{
2367 BlockDriverState *bs = opaque;
Emanuele Giuseppe Espositoda359902022-03-03 10:16:11 -05002368 GLOBAL_STATE_CODE();
Vladimir Sementsov-Ogievskiy2513ef52021-04-28 18:17:42 +03002369
2370 if (bs->drv->bdrv_abort_perm_update) {
2371 bs->drv->bdrv_abort_perm_update(bs);
2372 }
2373}
2374
2375TransactionActionDrv bdrv_drv_set_perm_drv = {
2376 .abort = bdrv_drv_set_perm_abort,
2377 .commit = bdrv_drv_set_perm_commit,
2378};
2379
Kevin Wolfbce73bc2023-09-11 11:46:13 +02002380/*
2381 * After calling this function, the transaction @tran may only be completed
2382 * while holding a reader lock for the graph.
2383 */
2384static int GRAPH_RDLOCK
2385bdrv_drv_set_perm(BlockDriverState *bs, uint64_t perm, uint64_t shared_perm,
2386 Transaction *tran, Error **errp)
Vladimir Sementsov-Ogievskiy2513ef52021-04-28 18:17:42 +03002387{
Emanuele Giuseppe Espositoda359902022-03-03 10:16:11 -05002388 GLOBAL_STATE_CODE();
Vladimir Sementsov-Ogievskiy2513ef52021-04-28 18:17:42 +03002389 if (!bs->drv) {
2390 return 0;
2391 }
2392
2393 if (bs->drv->bdrv_check_perm) {
2394 int ret = bs->drv->bdrv_check_perm(bs, perm, shared_perm, errp);
2395 if (ret < 0) {
2396 return ret;
2397 }
2398 }
2399
2400 if (tran) {
2401 tran_add(tran, &bdrv_drv_set_perm_drv, bs);
2402 }
2403
2404 return 0;
2405}
2406
Vladimir Sementsov-Ogievskiy09786232021-04-28 18:17:44 +03002407typedef struct BdrvReplaceChildState {
2408 BdrvChild *child;
2409 BlockDriverState *old_bs;
2410} BdrvReplaceChildState;
2411
Kevin Wolf5661a002023-09-11 11:46:10 +02002412static void GRAPH_WRLOCK bdrv_replace_child_commit(void *opaque)
Vladimir Sementsov-Ogievskiy09786232021-04-28 18:17:44 +03002413{
2414 BdrvReplaceChildState *s = opaque;
Emanuele Giuseppe Espositobdb73472022-03-03 10:16:02 -05002415 GLOBAL_STATE_CODE();
Vladimir Sementsov-Ogievskiy09786232021-04-28 18:17:44 +03002416
Kevin Wolf5661a002023-09-11 11:46:10 +02002417 bdrv_schedule_unref(s->old_bs);
Vladimir Sementsov-Ogievskiy09786232021-04-28 18:17:44 +03002418}
2419
Kevin Wolf5661a002023-09-11 11:46:10 +02002420static void GRAPH_WRLOCK bdrv_replace_child_abort(void *opaque)
Vladimir Sementsov-Ogievskiy09786232021-04-28 18:17:44 +03002421{
2422 BdrvReplaceChildState *s = opaque;
2423 BlockDriverState *new_bs = s->child->bs;
2424
Emanuele Giuseppe Espositobdb73472022-03-03 10:16:02 -05002425 GLOBAL_STATE_CODE();
Kevin Wolf5661a002023-09-11 11:46:10 +02002426 assert_bdrv_graph_writable();
Kevin Wolfad29eb32023-09-11 11:46:07 +02002427
Vladimir Sementsov-Ogievskiy0f0b1e22022-07-26 23:11:29 +03002428 /* old_bs reference is transparently moved from @s to @s->child */
Kevin Wolf23987472022-11-18 18:41:09 +01002429 if (!s->child->bs) {
2430 /*
2431 * The parents were undrained when removing old_bs from the child. New
2432 * requests can't have been made, though, because the child was empty.
2433 *
2434 * TODO Make bdrv_replace_child_noperm() transactionable to avoid
2435 * undraining the parent in the first place. Once this is done, having
2436 * new_bs drained when calling bdrv_replace_child_tran() is not a
2437 * requirement any more.
2438 */
Kevin Wolf606ed752022-11-18 18:41:10 +01002439 bdrv_parent_drained_begin_single(s->child);
Kevin Wolf23987472022-11-18 18:41:09 +01002440 assert(!bdrv_parent_drained_poll_single(s->child));
2441 }
2442 assert(s->child->quiesced_parent);
Vladimir Sementsov-Ogievskiy544acc72022-07-26 23:11:31 +03002443 bdrv_replace_child_noperm(s->child, s->old_bs);
Kevin Wolfad29eb32023-09-11 11:46:07 +02002444
Vladimir Sementsov-Ogievskiy09786232021-04-28 18:17:44 +03002445 bdrv_unref(new_bs);
2446}
2447
2448static TransactionActionDrv bdrv_replace_child_drv = {
2449 .commit = bdrv_replace_child_commit,
2450 .abort = bdrv_replace_child_abort,
2451 .clean = g_free,
2452};
2453
2454/*
Vladimir Sementsov-Ogievskiy4bf021d2021-06-10 14:25:44 +03002455 * bdrv_replace_child_tran
Vladimir Sementsov-Ogievskiy09786232021-04-28 18:17:44 +03002456 *
2457 * Note: real unref of old_bs is done only on commit.
Vladimir Sementsov-Ogievskiy4bf021d2021-06-10 14:25:44 +03002458 *
Kevin Wolf23987472022-11-18 18:41:09 +01002459 * Both @child->bs and @new_bs (if non-NULL) must be drained. @new_bs must be
2460 * kept drained until the transaction is completed.
2461 *
Kevin Wolf5661a002023-09-11 11:46:10 +02002462 * After calling this function, the transaction @tran may only be completed
2463 * while holding a writer lock for the graph.
2464 *
Vladimir Sementsov-Ogievskiy4bf021d2021-06-10 14:25:44 +03002465 * The function doesn't update permissions, caller is responsible for this.
Vladimir Sementsov-Ogievskiy09786232021-04-28 18:17:44 +03002466 */
Kevin Wolf2f64e1f2023-09-11 11:46:08 +02002467static void GRAPH_WRLOCK
2468bdrv_replace_child_tran(BdrvChild *child, BlockDriverState *new_bs,
2469 Transaction *tran)
Vladimir Sementsov-Ogievskiy09786232021-04-28 18:17:44 +03002470{
2471 BdrvReplaceChildState *s = g_new(BdrvReplaceChildState, 1);
Kevin Wolf23987472022-11-18 18:41:09 +01002472
2473 assert(child->quiesced_parent);
2474 assert(!new_bs || new_bs->quiesce_counter);
2475
Vladimir Sementsov-Ogievskiy09786232021-04-28 18:17:44 +03002476 *s = (BdrvReplaceChildState) {
Vladimir Sementsov-Ogievskiy0f0b1e22022-07-26 23:11:29 +03002477 .child = child,
2478 .old_bs = child->bs,
Vladimir Sementsov-Ogievskiy09786232021-04-28 18:17:44 +03002479 };
2480 tran_add(tran, &bdrv_replace_child_drv, s);
2481
2482 if (new_bs) {
2483 bdrv_ref(new_bs);
2484 }
Kevin Wolfad29eb32023-09-11 11:46:07 +02002485
Vladimir Sementsov-Ogievskiy544acc72022-07-26 23:11:31 +03002486 bdrv_replace_child_noperm(child, new_bs);
Vladimir Sementsov-Ogievskiy0f0b1e22022-07-26 23:11:29 +03002487 /* old_bs reference is transparently moved from @child to @s */
Vladimir Sementsov-Ogievskiy09786232021-04-28 18:17:44 +03002488}
2489
Kevin Wolf33a610c2016-12-15 13:04:20 +01002490/*
Vladimir Sementsov-Ogievskiyc20555e2021-04-28 18:18:04 +03002491 * Refresh permissions in @bs subtree. The function is intended to be called
2492 * after some graph modification that was done without permission update.
Kevin Wolfbce73bc2023-09-11 11:46:13 +02002493 *
2494 * After calling this function, the transaction @tran may only be completed
2495 * while holding a reader lock for the graph.
Kevin Wolf33a610c2016-12-15 13:04:20 +01002496 */
Kevin Wolfbce73bc2023-09-11 11:46:13 +02002497static int GRAPH_RDLOCK
2498bdrv_node_refresh_perm(BlockDriverState *bs, BlockReopenQueue *q,
2499 Transaction *tran, Error **errp)
Kevin Wolf33a610c2016-12-15 13:04:20 +01002500{
2501 BlockDriver *drv = bs->drv;
2502 BdrvChild *c;
2503 int ret;
Vladimir Sementsov-Ogievskiyc20555e2021-04-28 18:18:04 +03002504 uint64_t cumulative_perms, cumulative_shared_perms;
Emanuele Giuseppe Esposito862fded2022-03-03 10:15:55 -05002505 GLOBAL_STATE_CODE();
Vladimir Sementsov-Ogievskiyc20555e2021-04-28 18:18:04 +03002506
2507 bdrv_get_cumulative_perm(bs, &cumulative_perms, &cumulative_shared_perms);
Kevin Wolf33a610c2016-12-15 13:04:20 +01002508
2509 /* Write permissions never work with read-only images */
2510 if ((cumulative_perms & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)) &&
Max Reitzcc022142018-06-06 21:37:00 +02002511 !bdrv_is_writable_after_reopen(bs, q))
Kevin Wolf33a610c2016-12-15 13:04:20 +01002512 {
Max Reitz481e0ee2019-05-15 22:15:00 +02002513 if (!bdrv_is_writable_after_reopen(bs, NULL)) {
2514 error_setg(errp, "Block node is read-only");
2515 } else {
Vladimir Sementsov-Ogievskiyc20555e2021-04-28 18:18:04 +03002516 error_setg(errp, "Read-only block node '%s' cannot support "
2517 "read-write users", bdrv_get_node_name(bs));
Max Reitz481e0ee2019-05-15 22:15:00 +02002518 }
2519
Kevin Wolf33a610c2016-12-15 13:04:20 +01002520 return -EPERM;
2521 }
2522
Kevin Wolf9c60a5d2020-07-16 16:26:00 +02002523 /*
2524 * Unaligned requests will automatically be aligned to bl.request_alignment
2525 * and without RESIZE we can't extend requests to write to space beyond the
2526 * end of the image, so it's required that the image size is aligned.
2527 */
2528 if ((cumulative_perms & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)) &&
2529 !(cumulative_perms & BLK_PERM_RESIZE))
2530 {
2531 if ((bs->total_sectors * BDRV_SECTOR_SIZE) % bs->bl.request_alignment) {
2532 error_setg(errp, "Cannot get 'write' permission without 'resize': "
2533 "Image size is not a multiple of request "
2534 "alignment");
2535 return -EPERM;
2536 }
2537 }
2538
Kevin Wolf33a610c2016-12-15 13:04:20 +01002539 /* Check this node */
2540 if (!drv) {
2541 return 0;
2542 }
2543
Vladimir Sementsov-Ogievskiyb1d2bbe2021-04-28 18:17:43 +03002544 ret = bdrv_drv_set_perm(bs, cumulative_perms, cumulative_shared_perms, tran,
Vladimir Sementsov-Ogievskiy2513ef52021-04-28 18:17:42 +03002545 errp);
2546 if (ret < 0) {
2547 return ret;
Kevin Wolf33a610c2016-12-15 13:04:20 +01002548 }
2549
Kevin Wolf78e421c2016-12-20 23:25:12 +01002550 /* Drivers that never have children can omit .bdrv_child_perm() */
Kevin Wolf33a610c2016-12-15 13:04:20 +01002551 if (!drv->bdrv_child_perm) {
Kevin Wolf78e421c2016-12-20 23:25:12 +01002552 assert(QLIST_EMPTY(&bs->children));
Kevin Wolf33a610c2016-12-15 13:04:20 +01002553 return 0;
2554 }
2555
2556 /* Check all children */
2557 QLIST_FOREACH(c, &bs->children, next) {
2558 uint64_t cur_perm, cur_shared;
Max Reitz9eab1542019-05-22 19:03:50 +02002559
Max Reitze5d8a402020-05-13 13:05:44 +02002560 bdrv_child_perm(bs, c->bs, c, c->role, q,
Fam Zhengffd1a5a2017-05-03 00:35:38 +08002561 cumulative_perms, cumulative_shared_perms,
2562 &cur_perm, &cur_shared);
Vladimir Sementsov-Ogievskiyecb776b2021-04-28 18:18:02 +03002563 bdrv_child_set_perm(c, cur_perm, cur_shared, tran);
Vladimir Sementsov-Ogievskiyb1d2bbe2021-04-28 18:17:43 +03002564 }
2565
2566 return 0;
2567}
2568
Vladimir Sementsov-Ogievskiyfb0ff4d2022-11-07 19:35:58 +03002569/*
2570 * @list is a product of bdrv_topological_dfs() (may be called several times) -
2571 * a topologically sorted subgraph.
Kevin Wolfbce73bc2023-09-11 11:46:13 +02002572 *
2573 * After calling this function, the transaction @tran may only be completed
2574 * while holding a reader lock for the graph.
Vladimir Sementsov-Ogievskiyfb0ff4d2022-11-07 19:35:58 +03002575 */
Kevin Wolf3804e3c2023-09-11 11:46:12 +02002576static int GRAPH_RDLOCK
2577bdrv_do_refresh_perms(GSList *list, BlockReopenQueue *q, Transaction *tran,
2578 Error **errp)
Vladimir Sementsov-Ogievskiyb1d2bbe2021-04-28 18:17:43 +03002579{
2580 int ret;
2581 BlockDriverState *bs;
Emanuele Giuseppe Esposito862fded2022-03-03 10:15:55 -05002582 GLOBAL_STATE_CODE();
Vladimir Sementsov-Ogievskiyb1d2bbe2021-04-28 18:17:43 +03002583
Vladimir Sementsov-Ogievskiyb1d2bbe2021-04-28 18:17:43 +03002584 for ( ; list; list = list->next) {
2585 bs = list->data;
2586
Vladimir Sementsov-Ogievskiy9397c142021-04-28 18:17:53 +03002587 if (bdrv_parent_perms_conflict(bs, errp)) {
Vladimir Sementsov-Ogievskiyb1d2bbe2021-04-28 18:17:43 +03002588 return -EINVAL;
2589 }
2590
Vladimir Sementsov-Ogievskiyc20555e2021-04-28 18:18:04 +03002591 ret = bdrv_node_refresh_perm(bs, q, tran, errp);
Vladimir Sementsov-Ogievskiyb1d2bbe2021-04-28 18:17:43 +03002592 if (ret < 0) {
2593 return ret;
2594 }
Vladimir Sementsov-Ogievskiybd57f8f2021-04-28 18:17:41 +03002595 }
Vladimir Sementsov-Ogievskiy3ef45e02021-04-28 18:17:40 +03002596
Vladimir Sementsov-Ogievskiybd57f8f2021-04-28 18:17:41 +03002597 return 0;
2598}
2599
Vladimir Sementsov-Ogievskiyfb0ff4d2022-11-07 19:35:58 +03002600/*
2601 * @list is any list of nodes. List is completed by all subtrees and
2602 * topologically sorted. It's not a problem if some node occurs in the @list
2603 * several times.
Kevin Wolfbce73bc2023-09-11 11:46:13 +02002604 *
2605 * After calling this function, the transaction @tran may only be completed
2606 * while holding a reader lock for the graph.
Vladimir Sementsov-Ogievskiyfb0ff4d2022-11-07 19:35:58 +03002607 */
Kevin Wolf3804e3c2023-09-11 11:46:12 +02002608static int GRAPH_RDLOCK
2609bdrv_list_refresh_perms(GSList *list, BlockReopenQueue *q, Transaction *tran,
2610 Error **errp)
Vladimir Sementsov-Ogievskiyfb0ff4d2022-11-07 19:35:58 +03002611{
2612 g_autoptr(GHashTable) found = g_hash_table_new(NULL, NULL);
2613 g_autoptr(GSList) refresh_list = NULL;
2614
2615 for ( ; list; list = list->next) {
2616 refresh_list = bdrv_topological_dfs(refresh_list, found, list->data);
2617 }
2618
2619 return bdrv_do_refresh_perms(refresh_list, q, tran, errp);
2620}
2621
Kevin Wolfc7a0f2b2020-03-10 12:38:25 +01002622void bdrv_get_cumulative_perm(BlockDriverState *bs, uint64_t *perm,
2623 uint64_t *shared_perm)
Kevin Wolf33a610c2016-12-15 13:04:20 +01002624{
2625 BdrvChild *c;
2626 uint64_t cumulative_perms = 0;
2627 uint64_t cumulative_shared_perms = BLK_PERM_ALL;
2628
Emanuele Giuseppe Espositob4ad82a2022-03-03 10:15:57 -05002629 GLOBAL_STATE_CODE();
2630
Kevin Wolf33a610c2016-12-15 13:04:20 +01002631 QLIST_FOREACH(c, &bs->parents, next_parent) {
2632 cumulative_perms |= c->perm;
2633 cumulative_shared_perms &= c->shared_perm;
2634 }
2635
2636 *perm = cumulative_perms;
2637 *shared_perm = cumulative_shared_perms;
2638}
2639
Fam Zheng51761962017-05-03 00:35:36 +08002640char *bdrv_perm_names(uint64_t perm)
Kevin Wolfd0833192017-01-16 18:26:20 +01002641{
2642 struct perm_name {
2643 uint64_t perm;
2644 const char *name;
2645 } permissions[] = {
2646 { BLK_PERM_CONSISTENT_READ, "consistent read" },
2647 { BLK_PERM_WRITE, "write" },
2648 { BLK_PERM_WRITE_UNCHANGED, "write unchanged" },
2649 { BLK_PERM_RESIZE, "resize" },
Kevin Wolfd0833192017-01-16 18:26:20 +01002650 { 0, NULL }
2651 };
2652
Alberto Garciae2a74232020-01-10 18:15:18 +01002653 GString *result = g_string_sized_new(30);
Kevin Wolfd0833192017-01-16 18:26:20 +01002654 struct perm_name *p;
2655
2656 for (p = permissions; p->name; p++) {
2657 if (perm & p->perm) {
Alberto Garciae2a74232020-01-10 18:15:18 +01002658 if (result->len > 0) {
2659 g_string_append(result, ", ");
2660 }
2661 g_string_append(result, p->name);
Kevin Wolfd0833192017-01-16 18:26:20 +01002662 }
2663 }
2664
Alberto Garciae2a74232020-01-10 18:15:18 +01002665 return g_string_free(result, FALSE);
Kevin Wolfd0833192017-01-16 18:26:20 +01002666}
2667
Kevin Wolf33a610c2016-12-15 13:04:20 +01002668
Kevin Wolfbce73bc2023-09-11 11:46:13 +02002669/*
2670 * @tran is allowed to be NULL. In this case no rollback is possible.
2671 *
2672 * After calling this function, the transaction @tran may only be completed
2673 * while holding a reader lock for the graph.
2674 */
Kevin Wolf3804e3c2023-09-11 11:46:12 +02002675static int GRAPH_RDLOCK
2676bdrv_refresh_perms(BlockDriverState *bs, Transaction *tran, Error **errp)
Vladimir Sementsov-Ogievskiybb87e4d2020-11-06 15:42:38 +03002677{
2678 int ret;
Vladimir Sementsov-Ogievskiyf1316ed2022-11-07 19:35:57 +03002679 Transaction *local_tran = NULL;
Vladimir Sementsov-Ogievskiyb1d2bbe2021-04-28 18:17:43 +03002680 g_autoptr(GSList) list = bdrv_topological_dfs(NULL, NULL, bs);
Emanuele Giuseppe Esposito862fded2022-03-03 10:15:55 -05002681 GLOBAL_STATE_CODE();
Vladimir Sementsov-Ogievskiybb87e4d2020-11-06 15:42:38 +03002682
Vladimir Sementsov-Ogievskiyf1316ed2022-11-07 19:35:57 +03002683 if (!tran) {
2684 tran = local_tran = tran_new();
2685 }
2686
Vladimir Sementsov-Ogievskiyfb0ff4d2022-11-07 19:35:58 +03002687 ret = bdrv_do_refresh_perms(list, NULL, tran, errp);
Vladimir Sementsov-Ogievskiyf1316ed2022-11-07 19:35:57 +03002688
2689 if (local_tran) {
2690 tran_finalize(local_tran, ret);
2691 }
Vladimir Sementsov-Ogievskiybb87e4d2020-11-06 15:42:38 +03002692
Vladimir Sementsov-Ogievskiyb1d2bbe2021-04-28 18:17:43 +03002693 return ret;
Vladimir Sementsov-Ogievskiybb87e4d2020-11-06 15:42:38 +03002694}
2695
Kevin Wolf33a610c2016-12-15 13:04:20 +01002696int bdrv_child_try_set_perm(BdrvChild *c, uint64_t perm, uint64_t shared,
2697 Error **errp)
2698{
Max Reitz10467792019-05-22 19:03:51 +02002699 Error *local_err = NULL;
Vladimir Sementsov-Ogievskiy83928dc2021-04-28 18:17:39 +03002700 Transaction *tran = tran_new();
Kevin Wolf33a610c2016-12-15 13:04:20 +01002701 int ret;
2702
Emanuele Giuseppe Espositob4ad82a2022-03-03 10:15:57 -05002703 GLOBAL_STATE_CODE();
2704
Vladimir Sementsov-Ogievskiyecb776b2021-04-28 18:18:02 +03002705 bdrv_child_set_perm(c, perm, shared, tran);
Vladimir Sementsov-Ogievskiy83928dc2021-04-28 18:17:39 +03002706
Vladimir Sementsov-Ogievskiyf1316ed2022-11-07 19:35:57 +03002707 ret = bdrv_refresh_perms(c->bs, tran, &local_err);
Vladimir Sementsov-Ogievskiy83928dc2021-04-28 18:17:39 +03002708
2709 tran_finalize(tran, ret);
2710
Kevin Wolf33a610c2016-12-15 13:04:20 +01002711 if (ret < 0) {
Vladimir Sementsov-Ogievskiy071b4742020-11-06 15:42:41 +03002712 if ((perm & ~c->perm) || (c->shared_perm & ~shared)) {
2713 /* tighten permissions */
Max Reitz10467792019-05-22 19:03:51 +02002714 error_propagate(errp, local_err);
2715 } else {
2716 /*
2717 * Our caller may intend to only loosen restrictions and
2718 * does not expect this function to fail. Errors are not
2719 * fatal in such a case, so we can just hide them from our
2720 * caller.
2721 */
2722 error_free(local_err);
2723 ret = 0;
2724 }
Kevin Wolf33a610c2016-12-15 13:04:20 +01002725 }
2726
Vladimir Sementsov-Ogievskiy83928dc2021-04-28 18:17:39 +03002727 return ret;
Kevin Wolfd5e6f432016-12-14 17:24:36 +01002728}
2729
Max Reitzc1087f12019-05-22 19:03:46 +02002730int bdrv_child_refresh_perms(BlockDriverState *bs, BdrvChild *c, Error **errp)
2731{
2732 uint64_t parent_perms, parent_shared;
2733 uint64_t perms, shared;
2734
Emanuele Giuseppe Espositob4ad82a2022-03-03 10:15:57 -05002735 GLOBAL_STATE_CODE();
2736
Max Reitzc1087f12019-05-22 19:03:46 +02002737 bdrv_get_cumulative_perm(bs, &parent_perms, &parent_shared);
Max Reitze5d8a402020-05-13 13:05:44 +02002738 bdrv_child_perm(bs, c->bs, c, c->role, NULL,
Max Reitzbf8e9252020-05-13 13:05:16 +02002739 parent_perms, parent_shared, &perms, &shared);
Max Reitzc1087f12019-05-22 19:03:46 +02002740
2741 return bdrv_child_try_set_perm(c, perms, shared, errp);
2742}
2743
Max Reitz87278af2020-05-13 13:05:40 +02002744/*
2745 * Default implementation for .bdrv_child_perm() for block filters:
2746 * Forward CONSISTENT_READ, WRITE, WRITE_UNCHANGED, and RESIZE to the
2747 * filtered child.
2748 */
2749static void bdrv_filter_default_perms(BlockDriverState *bs, BdrvChild *c,
Max Reitz87278af2020-05-13 13:05:40 +02002750 BdrvChildRole role,
2751 BlockReopenQueue *reopen_queue,
2752 uint64_t perm, uint64_t shared,
2753 uint64_t *nperm, uint64_t *nshared)
Kevin Wolf6a1b9ee2016-12-15 11:27:32 +01002754{
Emanuele Giuseppe Esposito862fded2022-03-03 10:15:55 -05002755 GLOBAL_STATE_CODE();
Kevin Wolfe444fa82019-08-02 15:59:41 +02002756 *nperm = perm & DEFAULT_PERM_PASSTHROUGH;
2757 *nshared = (shared & DEFAULT_PERM_PASSTHROUGH) | DEFAULT_PERM_UNCHANGED;
Kevin Wolf6a1b9ee2016-12-15 11:27:32 +01002758}
2759
Max Reitz70082db2020-05-13 13:05:26 +02002760static void bdrv_default_perms_for_cow(BlockDriverState *bs, BdrvChild *c,
Max Reitz70082db2020-05-13 13:05:26 +02002761 BdrvChildRole role,
2762 BlockReopenQueue *reopen_queue,
2763 uint64_t perm, uint64_t shared,
2764 uint64_t *nperm, uint64_t *nshared)
2765{
Max Reitze5d8a402020-05-13 13:05:44 +02002766 assert(role & BDRV_CHILD_COW);
Emanuele Giuseppe Esposito862fded2022-03-03 10:15:55 -05002767 GLOBAL_STATE_CODE();
Max Reitz70082db2020-05-13 13:05:26 +02002768
2769 /*
2770 * We want consistent read from backing files if the parent needs it.
2771 * No other operations are performed on backing files.
2772 */
2773 perm &= BLK_PERM_CONSISTENT_READ;
2774
2775 /*
2776 * If the parent can deal with changing data, we're okay with a
2777 * writable and resizable backing file.
2778 * TODO Require !(perm & BLK_PERM_CONSISTENT_READ), too?
2779 */
2780 if (shared & BLK_PERM_WRITE) {
2781 shared = BLK_PERM_WRITE | BLK_PERM_RESIZE;
2782 } else {
2783 shared = 0;
2784 }
2785
Vladimir Sementsov-Ogievskiy64631f32021-09-02 12:37:54 +03002786 shared |= BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE_UNCHANGED;
Max Reitz70082db2020-05-13 13:05:26 +02002787
2788 if (bs->open_flags & BDRV_O_INACTIVE) {
2789 shared |= BLK_PERM_WRITE | BLK_PERM_RESIZE;
2790 }
2791
2792 *nperm = perm;
2793 *nshared = shared;
2794}
2795
Max Reitz6f838a42020-05-13 13:05:27 +02002796static void bdrv_default_perms_for_storage(BlockDriverState *bs, BdrvChild *c,
Max Reitz6f838a42020-05-13 13:05:27 +02002797 BdrvChildRole role,
2798 BlockReopenQueue *reopen_queue,
2799 uint64_t perm, uint64_t shared,
2800 uint64_t *nperm, uint64_t *nshared)
2801{
2802 int flags;
2803
Emanuele Giuseppe Esposito862fded2022-03-03 10:15:55 -05002804 GLOBAL_STATE_CODE();
Max Reitze5d8a402020-05-13 13:05:44 +02002805 assert(role & (BDRV_CHILD_METADATA | BDRV_CHILD_DATA));
Max Reitz6f838a42020-05-13 13:05:27 +02002806
2807 flags = bdrv_reopen_get_flags(reopen_queue, bs);
2808
2809 /*
2810 * Apart from the modifications below, the same permissions are
2811 * forwarded and left alone as for filters
2812 */
Max Reitze5d8a402020-05-13 13:05:44 +02002813 bdrv_filter_default_perms(bs, c, role, reopen_queue,
Max Reitz6f838a42020-05-13 13:05:27 +02002814 perm, shared, &perm, &shared);
2815
Max Reitzf8890542020-05-13 13:05:28 +02002816 if (role & BDRV_CHILD_METADATA) {
2817 /* Format drivers may touch metadata even if the guest doesn't write */
2818 if (bdrv_is_writable_after_reopen(bs, reopen_queue)) {
2819 perm |= BLK_PERM_WRITE | BLK_PERM_RESIZE;
2820 }
2821
2822 /*
2823 * bs->file always needs to be consistent because of the
2824 * metadata. We can never allow other users to resize or write
2825 * to it.
2826 */
2827 if (!(flags & BDRV_O_NO_IO)) {
2828 perm |= BLK_PERM_CONSISTENT_READ;
2829 }
2830 shared &= ~(BLK_PERM_WRITE | BLK_PERM_RESIZE);
Max Reitz6f838a42020-05-13 13:05:27 +02002831 }
2832
Max Reitzf8890542020-05-13 13:05:28 +02002833 if (role & BDRV_CHILD_DATA) {
2834 /*
2835 * Technically, everything in this block is a subset of the
2836 * BDRV_CHILD_METADATA path taken above, and so this could
2837 * be an "else if" branch. However, that is not obvious, and
2838 * this function is not performance critical, therefore we let
2839 * this be an independent "if".
2840 */
2841
2842 /*
2843 * We cannot allow other users to resize the file because the
2844 * format driver might have some assumptions about the size
2845 * (e.g. because it is stored in metadata, or because the file
2846 * is split into fixed-size data files).
2847 */
2848 shared &= ~BLK_PERM_RESIZE;
2849
2850 /*
2851 * WRITE_UNCHANGED often cannot be performed as such on the
2852 * data file. For example, the qcow2 driver may still need to
2853 * write copied clusters on copy-on-read.
2854 */
2855 if (perm & BLK_PERM_WRITE_UNCHANGED) {
2856 perm |= BLK_PERM_WRITE;
2857 }
2858
2859 /*
2860 * If the data file is written to, the format driver may
2861 * expect to be able to resize it by writing beyond the EOF.
2862 */
2863 if (perm & BLK_PERM_WRITE) {
2864 perm |= BLK_PERM_RESIZE;
2865 }
Max Reitz6f838a42020-05-13 13:05:27 +02002866 }
Max Reitz6f838a42020-05-13 13:05:27 +02002867
2868 if (bs->open_flags & BDRV_O_INACTIVE) {
2869 shared |= BLK_PERM_WRITE | BLK_PERM_RESIZE;
2870 }
2871
2872 *nperm = perm;
2873 *nshared = shared;
2874}
2875
Max Reitz2519f542020-05-13 13:05:29 +02002876void bdrv_default_perms(BlockDriverState *bs, BdrvChild *c,
Max Reitze5d8a402020-05-13 13:05:44 +02002877 BdrvChildRole role, BlockReopenQueue *reopen_queue,
Max Reitz2519f542020-05-13 13:05:29 +02002878 uint64_t perm, uint64_t shared,
2879 uint64_t *nperm, uint64_t *nshared)
2880{
Emanuele Giuseppe Espositob4ad82a2022-03-03 10:15:57 -05002881 GLOBAL_STATE_CODE();
Max Reitz2519f542020-05-13 13:05:29 +02002882 if (role & BDRV_CHILD_FILTERED) {
2883 assert(!(role & (BDRV_CHILD_DATA | BDRV_CHILD_METADATA |
2884 BDRV_CHILD_COW)));
Max Reitze5d8a402020-05-13 13:05:44 +02002885 bdrv_filter_default_perms(bs, c, role, reopen_queue,
Max Reitz2519f542020-05-13 13:05:29 +02002886 perm, shared, nperm, nshared);
2887 } else if (role & BDRV_CHILD_COW) {
2888 assert(!(role & (BDRV_CHILD_DATA | BDRV_CHILD_METADATA)));
Max Reitze5d8a402020-05-13 13:05:44 +02002889 bdrv_default_perms_for_cow(bs, c, role, reopen_queue,
Max Reitz2519f542020-05-13 13:05:29 +02002890 perm, shared, nperm, nshared);
2891 } else if (role & (BDRV_CHILD_METADATA | BDRV_CHILD_DATA)) {
Max Reitze5d8a402020-05-13 13:05:44 +02002892 bdrv_default_perms_for_storage(bs, c, role, reopen_queue,
Max Reitz2519f542020-05-13 13:05:29 +02002893 perm, shared, nperm, nshared);
2894 } else {
2895 g_assert_not_reached();
2896 }
2897}
2898
Max Reitz7b1d9c42019-11-08 13:34:51 +01002899uint64_t bdrv_qapi_perm_to_blk_perm(BlockPermission qapi_perm)
2900{
2901 static const uint64_t permissions[] = {
2902 [BLOCK_PERMISSION_CONSISTENT_READ] = BLK_PERM_CONSISTENT_READ,
2903 [BLOCK_PERMISSION_WRITE] = BLK_PERM_WRITE,
2904 [BLOCK_PERMISSION_WRITE_UNCHANGED] = BLK_PERM_WRITE_UNCHANGED,
2905 [BLOCK_PERMISSION_RESIZE] = BLK_PERM_RESIZE,
Max Reitz7b1d9c42019-11-08 13:34:51 +01002906 };
2907
2908 QEMU_BUILD_BUG_ON(ARRAY_SIZE(permissions) != BLOCK_PERMISSION__MAX);
2909 QEMU_BUILD_BUG_ON(1UL << ARRAY_SIZE(permissions) != BLK_PERM_ALL + 1);
2910
2911 assert(qapi_perm < BLOCK_PERMISSION__MAX);
2912
2913 return permissions[qapi_perm];
2914}
2915
Kevin Wolf23987472022-11-18 18:41:09 +01002916/*
2917 * Replaces the node that a BdrvChild points to without updating permissions.
2918 *
2919 * If @new_bs is non-NULL, the parent of @child must already be drained through
Stefan Hajnoczi23c983c2023-12-05 13:20:11 -05002920 * @child.
Kevin Wolf23987472022-11-18 18:41:09 +01002921 */
Kevin Wolfad29eb32023-09-11 11:46:07 +02002922static void GRAPH_WRLOCK
2923bdrv_replace_child_noperm(BdrvChild *child, BlockDriverState *new_bs)
Kevin Wolfe9740bc2016-05-23 15:52:26 +02002924{
2925 BlockDriverState *old_bs = child->bs;
Max Reitzdebc2922019-07-22 15:33:44 +02002926 int new_bs_quiesce_counter;
Kevin Wolfe9740bc2016-05-23 15:52:26 +02002927
Alberto Garcia2cad1eb2019-03-12 18:48:40 +02002928 assert(!child->frozen);
Kevin Wolf23987472022-11-18 18:41:09 +01002929
2930 /*
2931 * If we want to change the BdrvChild to point to a drained node as its new
2932 * child->bs, we need to make sure that its new parent is drained, too. In
2933 * other words, either child->quiesce_parent must already be true or we must
2934 * be able to set it and keep the parent's quiesce_counter consistent with
2935 * that, but without polling or starting new requests (this function
2936 * guarantees that it doesn't poll, and starting new requests would be
2937 * against the invariants of drain sections).
2938 *
2939 * To keep things simple, we pick the first option (child->quiesce_parent
2940 * must already be true). We also generalise the rule a bit to make it
2941 * easier to verify in callers and more likely to be covered in test cases:
2942 * The parent must be quiesced through this child even if new_bs isn't
2943 * currently drained.
2944 *
2945 * The only exception is for callers that always pass new_bs == NULL. In
2946 * this case, we obviously never need to consider the case of a drained
2947 * new_bs, so we can keep the callers simpler by allowing them not to drain
2948 * the parent.
2949 */
2950 assert(!new_bs || child->quiesced_parent);
Kevin Wolfbfb8aa62021-10-18 15:47:14 +02002951 assert(old_bs != new_bs);
Emanuele Giuseppe Espositof0c28322022-03-03 10:16:13 -05002952 GLOBAL_STATE_CODE();
Alberto Garcia2cad1eb2019-03-12 18:48:40 +02002953
Fam Zhengbb2614e2017-04-07 14:54:10 +08002954 if (old_bs && new_bs) {
2955 assert(bdrv_get_aio_context(old_bs) == bdrv_get_aio_context(new_bs));
2956 }
Max Reitzdebc2922019-07-22 15:33:44 +02002957
Kevin Wolfe9740bc2016-05-23 15:52:26 +02002958 if (old_bs) {
Max Reitzbd86fb92020-05-13 13:05:13 +02002959 if (child->klass->detach) {
2960 child->klass->detach(child);
Kevin Wolfd736f112017-12-18 16:05:48 +01002961 }
Kevin Wolfe9740bc2016-05-23 15:52:26 +02002962 QLIST_REMOVE(child, next_parent);
2963 }
Kevin Wolfe9740bc2016-05-23 15:52:26 +02002964
2965 child->bs = new_bs;
Kevin Wolf36fe1332016-05-17 14:51:55 +02002966
2967 if (new_bs) {
2968 QLIST_INSERT_HEAD(&new_bs->parents, child, next_parent);
Max Reitzbd86fb92020-05-13 13:05:13 +02002969 if (child->klass->attach) {
2970 child->klass->attach(child);
Kevin Wolfdb95dbb2017-02-08 11:28:52 +01002971 }
Kevin Wolf36fe1332016-05-17 14:51:55 +02002972 }
Max Reitzdebc2922019-07-22 15:33:44 +02002973
2974 /*
Kevin Wolf23987472022-11-18 18:41:09 +01002975 * If the parent was drained through this BdrvChild previously, but new_bs
2976 * is not drained, allow requests to come in only after the new node has
2977 * been attached.
Max Reitzdebc2922019-07-22 15:33:44 +02002978 */
Kevin Wolf57e05be2022-11-18 18:41:06 +01002979 new_bs_quiesce_counter = (new_bs ? new_bs->quiesce_counter : 0);
2980 if (!new_bs_quiesce_counter && child->quiesced_parent) {
Max Reitzdebc2922019-07-22 15:33:44 +02002981 bdrv_parent_drained_end_single(child);
Max Reitzdebc2922019-07-22 15:33:44 +02002982 }
Kevin Wolfe9740bc2016-05-23 15:52:26 +02002983}
2984
Hanna Reitz04c9c3a2021-11-15 15:53:59 +01002985/**
2986 * Free the given @child.
2987 *
2988 * The child must be empty (i.e. `child->bs == NULL`) and it must be
2989 * unused (i.e. not in a children list).
2990 */
2991static void bdrv_child_free(BdrvChild *child)
Vladimir Sementsov-Ogievskiy548a74c2021-04-28 18:17:46 +03002992{
2993 assert(!child->bs);
Emanuele Giuseppe Espositobdb73472022-03-03 10:16:02 -05002994 GLOBAL_STATE_CODE();
Kevin Wolf680e0cc2023-09-29 16:51:56 +02002995 GRAPH_RDLOCK_GUARD_MAINLOOP();
2996
Hanna Reitza2253692021-11-15 15:53:58 +01002997 assert(!child->next.le_prev); /* not in children list */
Hanna Reitz04c9c3a2021-11-15 15:53:59 +01002998
2999 g_free(child->name);
3000 g_free(child);
Vladimir Sementsov-Ogievskiy548a74c2021-04-28 18:17:46 +03003001}
3002
3003typedef struct BdrvAttachChildCommonState {
Vladimir Sementsov-Ogievskiy5bb047472022-07-26 23:11:32 +03003004 BdrvChild *child;
Vladimir Sementsov-Ogievskiy548a74c2021-04-28 18:17:46 +03003005 AioContext *old_parent_ctx;
3006 AioContext *old_child_ctx;
3007} BdrvAttachChildCommonState;
3008
Kevin Wolf5661a002023-09-11 11:46:10 +02003009static void GRAPH_WRLOCK bdrv_attach_child_common_abort(void *opaque)
Vladimir Sementsov-Ogievskiy548a74c2021-04-28 18:17:46 +03003010{
3011 BdrvAttachChildCommonState *s = opaque;
Vladimir Sementsov-Ogievskiy5bb047472022-07-26 23:11:32 +03003012 BlockDriverState *bs = s->child->bs;
Vladimir Sementsov-Ogievskiy548a74c2021-04-28 18:17:46 +03003013
Emanuele Giuseppe Espositof0c28322022-03-03 10:16:13 -05003014 GLOBAL_STATE_CODE();
Kevin Wolf5661a002023-09-11 11:46:10 +02003015 assert_bdrv_graph_writable();
Kevin Wolfad29eb32023-09-11 11:46:07 +02003016
Vladimir Sementsov-Ogievskiy5bb047472022-07-26 23:11:32 +03003017 bdrv_replace_child_noperm(s->child, NULL);
Vladimir Sementsov-Ogievskiy548a74c2021-04-28 18:17:46 +03003018
3019 if (bdrv_get_aio_context(bs) != s->old_child_ctx) {
Emanuele Giuseppe Esposito142e6902022-10-25 04:49:52 -04003020 bdrv_try_change_aio_context(bs, s->old_child_ctx, NULL, &error_abort);
Vladimir Sementsov-Ogievskiy548a74c2021-04-28 18:17:46 +03003021 }
3022
Vladimir Sementsov-Ogievskiy5bb047472022-07-26 23:11:32 +03003023 if (bdrv_child_get_parent_aio_context(s->child) != s->old_parent_ctx) {
Emanuele Giuseppe Espositof8be48a2022-10-25 04:49:49 -04003024 Transaction *tran;
3025 GHashTable *visited;
3026 bool ret;
Vladimir Sementsov-Ogievskiy548a74c2021-04-28 18:17:46 +03003027
Emanuele Giuseppe Espositof8be48a2022-10-25 04:49:49 -04003028 tran = tran_new();
Vladimir Sementsov-Ogievskiy548a74c2021-04-28 18:17:46 +03003029
Emanuele Giuseppe Espositof8be48a2022-10-25 04:49:49 -04003030 /* No need to visit `child`, because it has been detached already */
3031 visited = g_hash_table_new(NULL, NULL);
3032 ret = s->child->klass->change_aio_ctx(s->child, s->old_parent_ctx,
3033 visited, tran, &error_abort);
3034 g_hash_table_destroy(visited);
3035
3036 /* transaction is supposed to always succeed */
3037 assert(ret == true);
3038 tran_commit(tran);
Vladimir Sementsov-Ogievskiy548a74c2021-04-28 18:17:46 +03003039 }
3040
Kevin Wolf5661a002023-09-11 11:46:10 +02003041 bdrv_schedule_unref(bs);
Vladimir Sementsov-Ogievskiy5bb047472022-07-26 23:11:32 +03003042 bdrv_child_free(s->child);
Vladimir Sementsov-Ogievskiy548a74c2021-04-28 18:17:46 +03003043}
3044
3045static TransactionActionDrv bdrv_attach_child_common_drv = {
3046 .abort = bdrv_attach_child_common_abort,
3047 .clean = g_free,
3048};
3049
3050/*
3051 * Common part of attaching bdrv child to bs or to blk or to job
Vladimir Sementsov-Ogievskiyf8d2ad72021-06-01 10:52:13 +03003052 *
Vladimir Sementsov-Ogievskiy7ec390d2021-06-10 14:25:45 +03003053 * Function doesn't update permissions, caller is responsible for this.
Vladimir Sementsov-Ogievskiy5bb047472022-07-26 23:11:32 +03003054 *
Kevin Wolf5661a002023-09-11 11:46:10 +02003055 * After calling this function, the transaction @tran may only be completed
3056 * while holding a writer lock for the graph.
3057 *
Vladimir Sementsov-Ogievskiy5bb047472022-07-26 23:11:32 +03003058 * Returns new created child.
Kevin Wolfc066e802023-06-05 10:57:05 +02003059 *
Stefan Hajnoczi23c983c2023-12-05 13:20:11 -05003060 * Both @parent_bs and @child_bs can move to a different AioContext in this
3061 * function.
Vladimir Sementsov-Ogievskiy548a74c2021-04-28 18:17:46 +03003062 */
Kevin Wolf7d4ca9d2023-09-11 11:46:09 +02003063static BdrvChild * GRAPH_WRLOCK
3064bdrv_attach_child_common(BlockDriverState *child_bs,
3065 const char *child_name,
3066 const BdrvChildClass *child_class,
3067 BdrvChildRole child_role,
3068 uint64_t perm, uint64_t shared_perm,
3069 void *opaque,
3070 Transaction *tran, Error **errp)
Vladimir Sementsov-Ogievskiy548a74c2021-04-28 18:17:46 +03003071{
3072 BdrvChild *new_child;
Stefan Hajnoczib49f4752023-12-05 13:20:03 -05003073 AioContext *parent_ctx;
Vladimir Sementsov-Ogievskiy548a74c2021-04-28 18:17:46 +03003074 AioContext *child_ctx = bdrv_get_aio_context(child_bs);
3075
Vladimir Sementsov-Ogievskiyda261b62021-06-01 10:52:17 +03003076 assert(child_class->get_parent_desc);
Emanuele Giuseppe Espositobdb73472022-03-03 10:16:02 -05003077 GLOBAL_STATE_CODE();
Vladimir Sementsov-Ogievskiy548a74c2021-04-28 18:17:46 +03003078
3079 new_child = g_new(BdrvChild, 1);
3080 *new_child = (BdrvChild) {
3081 .bs = NULL,
3082 .name = g_strdup(child_name),
3083 .klass = child_class,
3084 .role = child_role,
3085 .perm = perm,
3086 .shared_perm = shared_perm,
3087 .opaque = opaque,
3088 };
3089
3090 /*
3091 * If the AioContexts don't match, first try to move the subtree of
3092 * child_bs into the AioContext of the new parent. If this doesn't work,
3093 * try moving the parent into the AioContext of child_bs instead.
3094 */
3095 parent_ctx = bdrv_child_get_parent_aio_context(new_child);
3096 if (child_ctx != parent_ctx) {
3097 Error *local_err = NULL;
Emanuele Giuseppe Esposito142e6902022-10-25 04:49:52 -04003098 int ret = bdrv_try_change_aio_context(child_bs, parent_ctx, NULL,
3099 &local_err);
Vladimir Sementsov-Ogievskiy548a74c2021-04-28 18:17:46 +03003100
Emanuele Giuseppe Espositof8be48a2022-10-25 04:49:49 -04003101 if (ret < 0 && child_class->change_aio_ctx) {
Markus Armbrusterfb2575f2023-09-21 14:13:11 +02003102 Transaction *aio_ctx_tran = tran_new();
Emanuele Giuseppe Espositof8be48a2022-10-25 04:49:49 -04003103 GHashTable *visited = g_hash_table_new(NULL, NULL);
3104 bool ret_child;
3105
3106 g_hash_table_add(visited, new_child);
3107 ret_child = child_class->change_aio_ctx(new_child, child_ctx,
Markus Armbrusterfb2575f2023-09-21 14:13:11 +02003108 visited, aio_ctx_tran,
3109 NULL);
Emanuele Giuseppe Espositof8be48a2022-10-25 04:49:49 -04003110 if (ret_child == true) {
Vladimir Sementsov-Ogievskiy548a74c2021-04-28 18:17:46 +03003111 error_free(local_err);
3112 ret = 0;
Vladimir Sementsov-Ogievskiy548a74c2021-04-28 18:17:46 +03003113 }
Markus Armbrusterfb2575f2023-09-21 14:13:11 +02003114 tran_finalize(aio_ctx_tran, ret_child == true ? 0 : -1);
Emanuele Giuseppe Espositof8be48a2022-10-25 04:49:49 -04003115 g_hash_table_destroy(visited);
Vladimir Sementsov-Ogievskiy548a74c2021-04-28 18:17:46 +03003116 }
3117
3118 if (ret < 0) {
3119 error_propagate(errp, local_err);
Hanna Reitz04c9c3a2021-11-15 15:53:59 +01003120 bdrv_child_free(new_child);
Vladimir Sementsov-Ogievskiy5bb047472022-07-26 23:11:32 +03003121 return NULL;
Vladimir Sementsov-Ogievskiy548a74c2021-04-28 18:17:46 +03003122 }
3123 }
3124
3125 bdrv_ref(child_bs);
Kevin Wolf23987472022-11-18 18:41:09 +01003126 /*
3127 * Let every new BdrvChild start with a drained parent. Inserting the child
3128 * in the graph with bdrv_replace_child_noperm() will undrain it if
3129 * @child_bs is not drained.
3130 *
3131 * The child was only just created and is not yet visible in global state
3132 * until bdrv_replace_child_noperm() inserts it into the graph, so nobody
3133 * could have sent requests and polling is not necessary.
3134 *
3135 * Note that this means that the parent isn't fully drained yet, we only
3136 * stop new requests from coming in. This is fine, we don't care about the
3137 * old requests here, they are not for this child. If another place enters a
3138 * drain section for the same parent, but wants it to be fully quiesced, it
3139 * will not run most of the the code in .drained_begin() again (which is not
3140 * a problem, we already did this), but it will still poll until the parent
3141 * is fully quiesced, so it will not be negatively affected either.
3142 */
Kevin Wolf606ed752022-11-18 18:41:10 +01003143 bdrv_parent_drained_begin_single(new_child);
Vladimir Sementsov-Ogievskiy544acc72022-07-26 23:11:31 +03003144 bdrv_replace_child_noperm(new_child, child_bs);
Vladimir Sementsov-Ogievskiy548a74c2021-04-28 18:17:46 +03003145
3146 BdrvAttachChildCommonState *s = g_new(BdrvAttachChildCommonState, 1);
3147 *s = (BdrvAttachChildCommonState) {
Vladimir Sementsov-Ogievskiy5bb047472022-07-26 23:11:32 +03003148 .child = new_child,
Vladimir Sementsov-Ogievskiy548a74c2021-04-28 18:17:46 +03003149 .old_parent_ctx = parent_ctx,
3150 .old_child_ctx = child_ctx,
3151 };
3152 tran_add(tran, &bdrv_attach_child_common_drv, s);
3153
Vladimir Sementsov-Ogievskiy5bb047472022-07-26 23:11:32 +03003154 return new_child;
Vladimir Sementsov-Ogievskiy548a74c2021-04-28 18:17:46 +03003155}
3156
Vladimir Sementsov-Ogievskiyf8d2ad72021-06-01 10:52:13 +03003157/*
Vladimir Sementsov-Ogievskiy7ec390d2021-06-10 14:25:45 +03003158 * Function doesn't update permissions, caller is responsible for this.
Kevin Wolfc066e802023-06-05 10:57:05 +02003159 *
Stefan Hajnoczi23c983c2023-12-05 13:20:11 -05003160 * Both @parent_bs and @child_bs can move to a different AioContext in this
3161 * function.
Kevin Wolf5661a002023-09-11 11:46:10 +02003162 *
3163 * After calling this function, the transaction @tran may only be completed
3164 * while holding a writer lock for the graph.
Vladimir Sementsov-Ogievskiyf8d2ad72021-06-01 10:52:13 +03003165 */
Kevin Wolf7d4ca9d2023-09-11 11:46:09 +02003166static BdrvChild * GRAPH_WRLOCK
3167bdrv_attach_child_noperm(BlockDriverState *parent_bs,
3168 BlockDriverState *child_bs,
3169 const char *child_name,
3170 const BdrvChildClass *child_class,
3171 BdrvChildRole child_role,
3172 Transaction *tran,
3173 Error **errp)
Vladimir Sementsov-Ogievskiyaa5a04c2021-04-28 18:17:47 +03003174{
Vladimir Sementsov-Ogievskiyaa5a04c2021-04-28 18:17:47 +03003175 uint64_t perm, shared_perm;
3176
3177 assert(parent_bs->drv);
Emanuele Giuseppe Espositobdb73472022-03-03 10:16:02 -05003178 GLOBAL_STATE_CODE();
Vladimir Sementsov-Ogievskiyaa5a04c2021-04-28 18:17:47 +03003179
Kevin Wolfbfb8aa62021-10-18 15:47:14 +02003180 if (bdrv_recurse_has_child(child_bs, parent_bs)) {
3181 error_setg(errp, "Making '%s' a %s child of '%s' would create a cycle",
3182 child_bs->node_name, child_name, parent_bs->node_name);
Vladimir Sementsov-Ogievskiy5bb047472022-07-26 23:11:32 +03003183 return NULL;
Kevin Wolfbfb8aa62021-10-18 15:47:14 +02003184 }
3185
Vladimir Sementsov-Ogievskiyaa5a04c2021-04-28 18:17:47 +03003186 bdrv_get_cumulative_perm(parent_bs, &perm, &shared_perm);
3187 bdrv_child_perm(parent_bs, child_bs, NULL, child_role, NULL,
3188 perm, shared_perm, &perm, &shared_perm);
3189
Vladimir Sementsov-Ogievskiy5bb047472022-07-26 23:11:32 +03003190 return bdrv_attach_child_common(child_bs, child_name, child_class,
3191 child_role, perm, shared_perm, parent_bs,
3192 tran, errp);
Vladimir Sementsov-Ogievskiyaa5a04c2021-04-28 18:17:47 +03003193}
3194
Alberto Garciab441dc72019-05-13 16:46:18 +03003195/*
3196 * This function steals the reference to child_bs from the caller.
3197 * That reference is later dropped by bdrv_root_unref_child().
3198 *
3199 * On failure NULL is returned, errp is set and the reference to
3200 * child_bs is also dropped.
3201 */
Kevin Wolff21d96d2016-03-08 13:47:46 +01003202BdrvChild *bdrv_root_attach_child(BlockDriverState *child_bs,
3203 const char *child_name,
Max Reitzbd86fb92020-05-13 13:05:13 +02003204 const BdrvChildClass *child_class,
Max Reitz258b7762020-05-13 13:05:15 +02003205 BdrvChildRole child_role,
Kevin Wolfd5e6f432016-12-14 17:24:36 +01003206 uint64_t perm, uint64_t shared_perm,
3207 void *opaque, Error **errp)
Kevin Wolfdf581792015-06-15 11:53:47 +02003208{
Kevin Wolfd5e6f432016-12-14 17:24:36 +01003209 int ret;
Vladimir Sementsov-Ogievskiy5bb047472022-07-26 23:11:32 +03003210 BdrvChild *child;
Vladimir Sementsov-Ogievskiy548a74c2021-04-28 18:17:46 +03003211 Transaction *tran = tran_new();
Kevin Wolfd5e6f432016-12-14 17:24:36 +01003212
Emanuele Giuseppe Espositob4ad82a2022-03-03 10:15:57 -05003213 GLOBAL_STATE_CODE();
3214
Vladimir Sementsov-Ogievskiy5bb047472022-07-26 23:11:32 +03003215 child = bdrv_attach_child_common(child_bs, child_name, child_class,
Vladimir Sementsov-Ogievskiy548a74c2021-04-28 18:17:46 +03003216 child_role, perm, shared_perm, opaque,
Vladimir Sementsov-Ogievskiy5bb047472022-07-26 23:11:32 +03003217 tran, errp);
3218 if (!child) {
3219 ret = -EINVAL;
Kevin Wolfe878bb12021-05-03 13:05:54 +02003220 goto out;
Kevin Wolfd5e6f432016-12-14 17:24:36 +01003221 }
3222
Vladimir Sementsov-Ogievskiyf1316ed2022-11-07 19:35:57 +03003223 ret = bdrv_refresh_perms(child_bs, tran, errp);
Kevin Wolfdf581792015-06-15 11:53:47 +02003224
Kevin Wolfe878bb12021-05-03 13:05:54 +02003225out:
3226 tran_finalize(tran, ret);
Vladimir Sementsov-Ogievskiyf8d2ad72021-06-01 10:52:13 +03003227
Kevin Wolf03b9eac2023-10-27 17:53:13 +02003228 bdrv_schedule_unref(child_bs);
Vladimir Sementsov-Ogievskiy5bb047472022-07-26 23:11:32 +03003229
3230 return ret < 0 ? NULL : child;
Kevin Wolfdf581792015-06-15 11:53:47 +02003231}
3232
Alberto Garciab441dc72019-05-13 16:46:18 +03003233/*
3234 * This function transfers the reference to child_bs from the caller
3235 * to parent_bs. That reference is later dropped by parent_bs on
3236 * bdrv_close() or if someone calls bdrv_unref_child().
3237 *
3238 * On failure NULL is returned, errp is set and the reference to
3239 * child_bs is also dropped.
3240 */
Wen Congyang98292c62016-05-10 15:36:38 +08003241BdrvChild *bdrv_attach_child(BlockDriverState *parent_bs,
3242 BlockDriverState *child_bs,
3243 const char *child_name,
Max Reitzbd86fb92020-05-13 13:05:13 +02003244 const BdrvChildClass *child_class,
Max Reitz258b7762020-05-13 13:05:15 +02003245 BdrvChildRole child_role,
Kevin Wolf8b2ff522016-12-20 22:21:17 +01003246 Error **errp)
Kevin Wolff21d96d2016-03-08 13:47:46 +01003247{
Vladimir Sementsov-Ogievskiyaa5a04c2021-04-28 18:17:47 +03003248 int ret;
Vladimir Sementsov-Ogievskiy5bb047472022-07-26 23:11:32 +03003249 BdrvChild *child;
Vladimir Sementsov-Ogievskiyaa5a04c2021-04-28 18:17:47 +03003250 Transaction *tran = tran_new();
Kevin Wolfd5e6f432016-12-14 17:24:36 +01003251
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05003252 GLOBAL_STATE_CODE();
3253
Vladimir Sementsov-Ogievskiy5bb047472022-07-26 23:11:32 +03003254 child = bdrv_attach_child_noperm(parent_bs, child_bs, child_name,
3255 child_class, child_role, tran, errp);
3256 if (!child) {
3257 ret = -EINVAL;
Vladimir Sementsov-Ogievskiyaa5a04c2021-04-28 18:17:47 +03003258 goto out;
Kevin Wolfd5e6f432016-12-14 17:24:36 +01003259 }
3260
Vladimir Sementsov-Ogievskiyf1316ed2022-11-07 19:35:57 +03003261 ret = bdrv_refresh_perms(parent_bs, tran, errp);
Vladimir Sementsov-Ogievskiyaa5a04c2021-04-28 18:17:47 +03003262 if (ret < 0) {
3263 goto out;
3264 }
3265
3266out:
3267 tran_finalize(tran, ret);
3268
Kevin Wolfafdaeb92023-09-11 11:46:11 +02003269 bdrv_schedule_unref(child_bs);
Vladimir Sementsov-Ogievskiyaa5a04c2021-04-28 18:17:47 +03003270
Vladimir Sementsov-Ogievskiy5bb047472022-07-26 23:11:32 +03003271 return ret < 0 ? NULL : child;
Kevin Wolff21d96d2016-03-08 13:47:46 +01003272}
3273
Max Reitz7b99a262019-06-12 16:07:11 +02003274/* Callers must ensure that child->frozen is false. */
Kevin Wolff21d96d2016-03-08 13:47:46 +01003275void bdrv_root_unref_child(BdrvChild *child)
Kevin Wolf33a60402015-06-15 13:51:04 +02003276{
Vladimir Sementsov-Ogievskiy00eb93b2022-11-07 19:35:55 +03003277 BlockDriverState *child_bs = child->bs;
Kevin Wolf779020c2015-10-13 14:09:44 +02003278
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05003279 GLOBAL_STATE_CODE();
Vladimir Sementsov-Ogievskiy00eb93b2022-11-07 19:35:55 +03003280 bdrv_replace_child_noperm(child, NULL);
3281 bdrv_child_free(child);
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05003282
Vladimir Sementsov-Ogievskiy00eb93b2022-11-07 19:35:55 +03003283 if (child_bs) {
3284 /*
3285 * Update permissions for old node. We're just taking a parent away, so
3286 * we're loosening restrictions. Errors of permission update are not
3287 * fatal in this case, ignore them.
3288 */
Vladimir Sementsov-Ogievskiyf1316ed2022-11-07 19:35:57 +03003289 bdrv_refresh_perms(child_bs, NULL, NULL);
Vladimir Sementsov-Ogievskiy00eb93b2022-11-07 19:35:55 +03003290
3291 /*
3292 * When the parent requiring a non-default AioContext is removed, the
3293 * node moves back to the main AioContext
3294 */
3295 bdrv_try_change_aio_context(child_bs, qemu_get_aio_context(), NULL,
3296 NULL);
3297 }
3298
Kevin Wolfede01e42023-09-11 11:46:18 +02003299 bdrv_schedule_unref(child_bs);
Kevin Wolff21d96d2016-03-08 13:47:46 +01003300}
3301
Vladimir Sementsov-Ogievskiy332b3a12021-04-28 18:17:54 +03003302typedef struct BdrvSetInheritsFrom {
3303 BlockDriverState *bs;
3304 BlockDriverState *old_inherits_from;
3305} BdrvSetInheritsFrom;
3306
3307static void bdrv_set_inherits_from_abort(void *opaque)
3308{
3309 BdrvSetInheritsFrom *s = opaque;
3310
3311 s->bs->inherits_from = s->old_inherits_from;
3312}
3313
3314static TransactionActionDrv bdrv_set_inherits_from_drv = {
3315 .abort = bdrv_set_inherits_from_abort,
3316 .clean = g_free,
3317};
3318
3319/* @tran is allowed to be NULL. In this case no rollback is possible */
3320static void bdrv_set_inherits_from(BlockDriverState *bs,
3321 BlockDriverState *new_inherits_from,
3322 Transaction *tran)
3323{
3324 if (tran) {
3325 BdrvSetInheritsFrom *s = g_new(BdrvSetInheritsFrom, 1);
3326
3327 *s = (BdrvSetInheritsFrom) {
3328 .bs = bs,
3329 .old_inherits_from = bs->inherits_from,
3330 };
3331
3332 tran_add(tran, &bdrv_set_inherits_from_drv, s);
3333 }
3334
3335 bs->inherits_from = new_inherits_from;
3336}
3337
Max Reitz3cf746b2019-07-03 19:28:07 +02003338/**
3339 * Clear all inherits_from pointers from children and grandchildren of
3340 * @root that point to @root, where necessary.
Vladimir Sementsov-Ogievskiy332b3a12021-04-28 18:17:54 +03003341 * @tran is allowed to be NULL. In this case no rollback is possible
Max Reitz3cf746b2019-07-03 19:28:07 +02003342 */
Kevin Wolf32a8aba2023-09-11 11:46:19 +02003343static void GRAPH_WRLOCK
3344bdrv_unset_inherits_from(BlockDriverState *root, BdrvChild *child,
3345 Transaction *tran)
Kevin Wolff21d96d2016-03-08 13:47:46 +01003346{
Max Reitz3cf746b2019-07-03 19:28:07 +02003347 BdrvChild *c;
Kevin Wolf33a60402015-06-15 13:51:04 +02003348
Max Reitz3cf746b2019-07-03 19:28:07 +02003349 if (child->bs->inherits_from == root) {
3350 /*
3351 * Remove inherits_from only when the last reference between root and
3352 * child->bs goes away.
3353 */
3354 QLIST_FOREACH(c, &root->children, next) {
Kevin Wolf4e4bf5c2016-12-16 18:52:37 +01003355 if (c != child && c->bs == child->bs) {
3356 break;
3357 }
3358 }
3359 if (c == NULL) {
Vladimir Sementsov-Ogievskiy332b3a12021-04-28 18:17:54 +03003360 bdrv_set_inherits_from(child->bs, NULL, tran);
Kevin Wolf4e4bf5c2016-12-16 18:52:37 +01003361 }
Kevin Wolf33a60402015-06-15 13:51:04 +02003362 }
3363
Max Reitz3cf746b2019-07-03 19:28:07 +02003364 QLIST_FOREACH(c, &child->bs->children, next) {
Vladimir Sementsov-Ogievskiy332b3a12021-04-28 18:17:54 +03003365 bdrv_unset_inherits_from(root, c, tran);
Max Reitz3cf746b2019-07-03 19:28:07 +02003366 }
3367}
3368
Max Reitz7b99a262019-06-12 16:07:11 +02003369/* Callers must ensure that child->frozen is false. */
Max Reitz3cf746b2019-07-03 19:28:07 +02003370void bdrv_unref_child(BlockDriverState *parent, BdrvChild *child)
3371{
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05003372 GLOBAL_STATE_CODE();
Max Reitz3cf746b2019-07-03 19:28:07 +02003373 if (child == NULL) {
3374 return;
3375 }
3376
Vladimir Sementsov-Ogievskiy332b3a12021-04-28 18:17:54 +03003377 bdrv_unset_inherits_from(parent, child, NULL);
Kevin Wolff21d96d2016-03-08 13:47:46 +01003378 bdrv_root_unref_child(child);
Kevin Wolf33a60402015-06-15 13:51:04 +02003379}
3380
Kevin Wolf5c8cab42016-02-24 15:13:35 +01003381
Kevin Wolf356f4ef2023-09-11 11:46:15 +02003382static void GRAPH_RDLOCK
3383bdrv_parent_cb_change_media(BlockDriverState *bs, bool load)
Kevin Wolf5c8cab42016-02-24 15:13:35 +01003384{
3385 BdrvChild *c;
Emanuele Giuseppe Espositof0c28322022-03-03 10:16:13 -05003386 GLOBAL_STATE_CODE();
Kevin Wolf5c8cab42016-02-24 15:13:35 +01003387 QLIST_FOREACH(c, &bs->parents, next_parent) {
Max Reitzbd86fb92020-05-13 13:05:13 +02003388 if (c->klass->change_media) {
3389 c->klass->change_media(c, load);
Kevin Wolf5c8cab42016-02-24 15:13:35 +01003390 }
3391 }
3392}
3393
Alberto Garcia0065c452018-10-31 18:16:37 +02003394/* Return true if you can reach parent going through child->inherits_from
3395 * recursively. If parent or child are NULL, return false */
3396static bool bdrv_inherits_from_recursive(BlockDriverState *child,
3397 BlockDriverState *parent)
3398{
3399 while (child && child != parent) {
3400 child = child->inherits_from;
3401 }
3402
3403 return child != NULL;
3404}
3405
Kevin Wolf5db15a52015-09-14 15:33:33 +02003406/*
Max Reitz25191e52020-05-13 13:05:33 +02003407 * Return the BdrvChildRole for @bs's backing child. bs->backing is
3408 * mostly used for COW backing children (role = COW), but also for
3409 * filtered children (role = FILTERED | PRIMARY).
3410 */
3411static BdrvChildRole bdrv_backing_role(BlockDriverState *bs)
3412{
3413 if (bs->drv && bs->drv->is_filter) {
3414 return BDRV_CHILD_FILTERED | BDRV_CHILD_PRIMARY;
3415 } else {
3416 return BDRV_CHILD_COW;
3417 }
3418}
3419
3420/*
Vladimir Sementsov-Ogievskiye9238272021-06-10 15:05:30 +03003421 * Sets the bs->backing or bs->file link of a BDS. A new reference is created;
3422 * callers which don't need their own reference any more must call bdrv_unref().
Vladimir Sementsov-Ogievskiy7ec390d2021-06-10 14:25:45 +03003423 *
Kevin Wolf7d4ca9d2023-09-11 11:46:09 +02003424 * If the respective child is already present (i.e. we're detaching a node),
3425 * that child node must be drained.
3426 *
Vladimir Sementsov-Ogievskiy7ec390d2021-06-10 14:25:45 +03003427 * Function doesn't update permissions, caller is responsible for this.
Kevin Wolf4b408662023-06-05 10:57:06 +02003428 *
Stefan Hajnoczi23c983c2023-12-05 13:20:11 -05003429 * Both @parent_bs and @child_bs can move to a different AioContext in this
3430 * function.
Kevin Wolf5661a002023-09-11 11:46:10 +02003431 *
3432 * After calling this function, the transaction @tran may only be completed
3433 * while holding a writer lock for the graph.
Kevin Wolf5db15a52015-09-14 15:33:33 +02003434 */
Kevin Wolf7d4ca9d2023-09-11 11:46:09 +02003435static int GRAPH_WRLOCK
3436bdrv_set_file_or_backing_noperm(BlockDriverState *parent_bs,
3437 BlockDriverState *child_bs,
3438 bool is_backing,
3439 Transaction *tran, Error **errp)
Fam Zheng8d24cce2014-05-23 21:29:45 +08003440{
Vladimir Sementsov-Ogievskiye9238272021-06-10 15:05:30 +03003441 bool update_inherits_from =
3442 bdrv_inherits_from_recursive(child_bs, parent_bs);
3443 BdrvChild *child = is_backing ? parent_bs->backing : parent_bs->file;
3444 BdrvChildRole role;
Alberto Garcia0065c452018-10-31 18:16:37 +02003445
Emanuele Giuseppe Espositobdb73472022-03-03 10:16:02 -05003446 GLOBAL_STATE_CODE();
3447
Vladimir Sementsov-Ogievskiye9238272021-06-10 15:05:30 +03003448 if (!parent_bs->drv) {
3449 /*
3450 * Node without drv is an object without a class :/. TODO: finally fix
3451 * qcow2 driver to never clear bs->drv and implement format corruption
3452 * handling in other way.
3453 */
3454 error_setg(errp, "Node corrupted");
3455 return -EINVAL;
3456 }
3457
3458 if (child && child->frozen) {
3459 error_setg(errp, "Cannot change frozen '%s' link from '%s' to '%s'",
3460 child->name, parent_bs->node_name, child->bs->node_name);
Vladimir Sementsov-Ogievskiya1e708f2021-02-02 15:49:43 +03003461 return -EPERM;
Alberto Garcia2cad1eb2019-03-12 18:48:40 +02003462 }
3463
Vladimir Sementsov-Ogievskiy25f78d92021-06-10 15:05:34 +03003464 if (is_backing && !parent_bs->drv->is_filter &&
3465 !parent_bs->drv->supports_backing)
3466 {
3467 error_setg(errp, "Driver '%s' of node '%s' does not support backing "
3468 "files", parent_bs->drv->format_name, parent_bs->node_name);
3469 return -EINVAL;
3470 }
3471
Vladimir Sementsov-Ogievskiye9238272021-06-10 15:05:30 +03003472 if (parent_bs->drv->is_filter) {
3473 role = BDRV_CHILD_FILTERED | BDRV_CHILD_PRIMARY;
3474 } else if (is_backing) {
3475 role = BDRV_CHILD_COW;
3476 } else {
3477 /*
3478 * We only can use same role as it is in existing child. We don't have
3479 * infrastructure to determine role of file child in generic way
3480 */
3481 if (!child) {
3482 error_setg(errp, "Cannot set file child to format node without "
3483 "file child");
3484 return -EINVAL;
3485 }
3486 role = child->role;
Fam Zheng826b6ca2014-05-23 21:29:47 +08003487 }
3488
Vladimir Sementsov-Ogievskiye9238272021-06-10 15:05:30 +03003489 if (child) {
Kevin Wolf7d4ca9d2023-09-11 11:46:09 +02003490 assert(child->bs->quiesce_counter);
Vladimir Sementsov-Ogievskiye9238272021-06-10 15:05:30 +03003491 bdrv_unset_inherits_from(parent_bs, child, tran);
Vladimir Sementsov-Ogievskiy57f08942022-07-26 23:11:34 +03003492 bdrv_remove_child(child, tran);
Vladimir Sementsov-Ogievskiye9238272021-06-10 15:05:30 +03003493 }
3494
3495 if (!child_bs) {
Fam Zheng8d24cce2014-05-23 21:29:45 +08003496 goto out;
3497 }
Kevin Wolf12fa4af2017-02-17 20:42:32 +01003498
Vladimir Sementsov-Ogievskiy5bb047472022-07-26 23:11:32 +03003499 child = bdrv_attach_child_noperm(parent_bs, child_bs,
3500 is_backing ? "backing" : "file",
3501 &child_of_bds, role,
3502 tran, errp);
3503 if (!child) {
3504 return -EINVAL;
Vladimir Sementsov-Ogievskiya1e708f2021-02-02 15:49:43 +03003505 }
3506
Vladimir Sementsov-Ogievskiy160333e2021-04-28 18:17:56 +03003507
3508 /*
Vladimir Sementsov-Ogievskiye9238272021-06-10 15:05:30 +03003509 * If inherits_from pointed recursively to bs then let's update it to
Vladimir Sementsov-Ogievskiy160333e2021-04-28 18:17:56 +03003510 * point directly to bs (else it will become NULL).
3511 */
Vladimir Sementsov-Ogievskiya1e708f2021-02-02 15:49:43 +03003512 if (update_inherits_from) {
Vladimir Sementsov-Ogievskiye9238272021-06-10 15:05:30 +03003513 bdrv_set_inherits_from(child_bs, parent_bs, tran);
Alberto Garcia0065c452018-10-31 18:16:37 +02003514 }
Fam Zheng826b6ca2014-05-23 21:29:47 +08003515
Fam Zheng8d24cce2014-05-23 21:29:45 +08003516out:
Vladimir Sementsov-Ogievskiye9238272021-06-10 15:05:30 +03003517 bdrv_refresh_limits(parent_bs, tran, NULL);
Vladimir Sementsov-Ogievskiy160333e2021-04-28 18:17:56 +03003518
3519 return 0;
3520}
3521
Kevin Wolf4b408662023-06-05 10:57:06 +02003522/*
Stefan Hajnoczi23c983c2023-12-05 13:20:11 -05003523 * Both @bs and @backing_hd can move to a different AioContext in this
3524 * function.
Kevin Wolf7d4ca9d2023-09-11 11:46:09 +02003525 *
3526 * If a backing child is already present (i.e. we're detaching a node), that
3527 * child node must be drained.
Kevin Wolf4b408662023-06-05 10:57:06 +02003528 */
Kevin Wolf92140b92022-11-18 18:41:04 +01003529int bdrv_set_backing_hd_drained(BlockDriverState *bs,
3530 BlockDriverState *backing_hd,
3531 Error **errp)
Vladimir Sementsov-Ogievskiy160333e2021-04-28 18:17:56 +03003532{
3533 int ret;
3534 Transaction *tran = tran_new();
3535
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05003536 GLOBAL_STATE_CODE();
Kevin Wolf92140b92022-11-18 18:41:04 +01003537 assert(bs->quiesce_counter > 0);
Kevin Wolf7d4ca9d2023-09-11 11:46:09 +02003538 if (bs->backing) {
3539 assert(bs->backing->bs->quiesce_counter > 0);
3540 }
Vladimir Sementsov-Ogievskiyc0829cb2022-01-24 18:37:41 +01003541
Kevin Wolf3204c2e2023-10-27 17:53:23 +02003542 ret = bdrv_set_file_or_backing_noperm(bs, backing_hd, true, tran, errp);
Vladimir Sementsov-Ogievskiy160333e2021-04-28 18:17:56 +03003543 if (ret < 0) {
3544 goto out;
3545 }
3546
Vladimir Sementsov-Ogievskiyf1316ed2022-11-07 19:35:57 +03003547 ret = bdrv_refresh_perms(bs, tran, errp);
Vladimir Sementsov-Ogievskiy160333e2021-04-28 18:17:56 +03003548out:
3549 tran_finalize(tran, ret);
Kevin Wolf92140b92022-11-18 18:41:04 +01003550 return ret;
3551}
Vladimir Sementsov-Ogievskiya1e708f2021-02-02 15:49:43 +03003552
Kevin Wolf92140b92022-11-18 18:41:04 +01003553int bdrv_set_backing_hd(BlockDriverState *bs, BlockDriverState *backing_hd,
3554 Error **errp)
3555{
Kevin Wolf004915a2023-10-27 17:53:26 +02003556 BlockDriverState *drain_bs;
Kevin Wolf92140b92022-11-18 18:41:04 +01003557 int ret;
3558 GLOBAL_STATE_CODE();
3559
Kevin Wolf004915a2023-10-27 17:53:26 +02003560 bdrv_graph_rdlock_main_loop();
3561 drain_bs = bs->backing ? bs->backing->bs : bs;
3562 bdrv_graph_rdunlock_main_loop();
3563
Kevin Wolf7d4ca9d2023-09-11 11:46:09 +02003564 bdrv_ref(drain_bs);
3565 bdrv_drained_begin(drain_bs);
Stefan Hajnoczi6bc30f12023-12-05 13:20:02 -05003566 bdrv_graph_wrlock();
Kevin Wolf92140b92022-11-18 18:41:04 +01003567 ret = bdrv_set_backing_hd_drained(bs, backing_hd, errp);
Stefan Hajnoczi6bc30f12023-12-05 13:20:02 -05003568 bdrv_graph_wrunlock();
Kevin Wolf7d4ca9d2023-09-11 11:46:09 +02003569 bdrv_drained_end(drain_bs);
3570 bdrv_unref(drain_bs);
Vladimir Sementsov-Ogievskiyc0829cb2022-01-24 18:37:41 +01003571
Vladimir Sementsov-Ogievskiya1e708f2021-02-02 15:49:43 +03003572 return ret;
Fam Zheng8d24cce2014-05-23 21:29:45 +08003573}
3574
Kevin Wolf31ca6d02013-03-28 15:29:24 +01003575/*
3576 * Opens the backing file for a BlockDriverState if not yet open
3577 *
Kevin Wolfd9b7b052015-01-16 18:23:41 +01003578 * bdref_key specifies the key for the image's BlockdevRef in the options QDict.
3579 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
3580 * itself, all options starting with "${bdref_key}." are considered part of the
3581 * BlockdevRef.
3582 *
3583 * TODO Can this be unified with bdrv_open_image()?
Kevin Wolf31ca6d02013-03-28 15:29:24 +01003584 */
Kevin Wolfd9b7b052015-01-16 18:23:41 +01003585int bdrv_open_backing_file(BlockDriverState *bs, QDict *parent_options,
3586 const char *bdref_key, Error **errp)
Paolo Bonzini9156df12012-10-18 16:49:17 +02003587{
Max Reitz6b6833c2019-02-01 20:29:15 +01003588 char *backing_filename = NULL;
Kevin Wolfd9b7b052015-01-16 18:23:41 +01003589 char *bdref_key_dot;
3590 const char *reference = NULL;
Kevin Wolf317fc442014-04-25 13:27:34 +02003591 int ret = 0;
Max Reitz998c2012019-02-01 20:29:08 +01003592 bool implicit_backing = false;
Fam Zheng8d24cce2014-05-23 21:29:45 +08003593 BlockDriverState *backing_hd;
Kevin Wolfd9b7b052015-01-16 18:23:41 +01003594 QDict *options;
3595 QDict *tmp_parent_options = NULL;
Max Reitz34b5d2c2013-09-05 14:45:29 +02003596 Error *local_err = NULL;
Paolo Bonzini9156df12012-10-18 16:49:17 +02003597
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05003598 GLOBAL_STATE_CODE();
Kevin Wolf004915a2023-10-27 17:53:26 +02003599 GRAPH_RDLOCK_GUARD_MAINLOOP();
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05003600
Kevin Wolf760e0062015-06-17 14:55:21 +02003601 if (bs->backing != NULL) {
Benoît Canet1ba4b6a2014-04-22 17:05:27 +02003602 goto free_exit;
Paolo Bonzini9156df12012-10-18 16:49:17 +02003603 }
3604
Kevin Wolf31ca6d02013-03-28 15:29:24 +01003605 /* NULL means an empty set of options */
Kevin Wolfd9b7b052015-01-16 18:23:41 +01003606 if (parent_options == NULL) {
3607 tmp_parent_options = qdict_new();
3608 parent_options = tmp_parent_options;
Kevin Wolf31ca6d02013-03-28 15:29:24 +01003609 }
3610
Paolo Bonzini9156df12012-10-18 16:49:17 +02003611 bs->open_flags &= ~BDRV_O_NO_BACKING;
Kevin Wolfd9b7b052015-01-16 18:23:41 +01003612
3613 bdref_key_dot = g_strdup_printf("%s.", bdref_key);
3614 qdict_extract_subqdict(parent_options, &options, bdref_key_dot);
3615 g_free(bdref_key_dot);
3616
Markus Armbruster129c7d12017-03-30 19:43:12 +02003617 /*
3618 * Caution: while qdict_get_try_str() is fine, getting non-string
3619 * types would require more care. When @parent_options come from
3620 * -blockdev or blockdev_add, its members are typed according to
3621 * the QAPI schema, but when they come from -drive, they're all
3622 * QString.
3623 */
Kevin Wolfd9b7b052015-01-16 18:23:41 +01003624 reference = qdict_get_try_str(parent_options, bdref_key);
3625 if (reference || qdict_haskey(options, "file.filename")) {
Max Reitz6b6833c2019-02-01 20:29:15 +01003626 /* keep backing_filename NULL */
Kevin Wolf1cb6f502013-04-12 20:27:07 +02003627 } else if (bs->backing_file[0] == '\0' && qdict_size(options) == 0) {
Marc-André Lureaucb3e7f02018-04-19 17:01:43 +02003628 qobject_unref(options);
Benoît Canet1ba4b6a2014-04-22 17:05:27 +02003629 goto free_exit;
Fam Zhengdbecebd2013-09-22 20:05:06 +08003630 } else {
Max Reitz998c2012019-02-01 20:29:08 +01003631 if (qdict_size(options) == 0) {
3632 /* If the user specifies options that do not modify the
3633 * backing file's behavior, we might still consider it the
3634 * implicit backing file. But it's easier this way, and
3635 * just specifying some of the backing BDS's options is
3636 * only possible with -drive anyway (otherwise the QAPI
3637 * schema forces the user to specify everything). */
3638 implicit_backing = !strcmp(bs->auto_backing_file, bs->backing_file);
3639 }
3640
Max Reitz6b6833c2019-02-01 20:29:15 +01003641 backing_filename = bdrv_get_full_backing_filename(bs, &local_err);
Max Reitz9f074292014-11-26 17:20:26 +01003642 if (local_err) {
3643 ret = -EINVAL;
3644 error_propagate(errp, local_err);
Marc-André Lureaucb3e7f02018-04-19 17:01:43 +02003645 qobject_unref(options);
Max Reitz9f074292014-11-26 17:20:26 +01003646 goto free_exit;
3647 }
Paolo Bonzini9156df12012-10-18 16:49:17 +02003648 }
3649
Kevin Wolf8ee79e72014-06-04 15:09:35 +02003650 if (!bs->drv || !bs->drv->supports_backing) {
3651 ret = -EINVAL;
3652 error_setg(errp, "Driver doesn't support backing files");
Marc-André Lureaucb3e7f02018-04-19 17:01:43 +02003653 qobject_unref(options);
Kevin Wolf8ee79e72014-06-04 15:09:35 +02003654 goto free_exit;
3655 }
3656
Peter Krempa6bff5972017-10-12 16:14:10 +02003657 if (!reference &&
3658 bs->backing_format[0] != '\0' && !qdict_haskey(options, "driver")) {
Eric Blake46f5ac22017-04-27 16:58:17 -05003659 qdict_put_str(options, "driver", bs->backing_format);
Paolo Bonzini9156df12012-10-18 16:49:17 +02003660 }
3661
Max Reitz6b6833c2019-02-01 20:29:15 +01003662 backing_hd = bdrv_open_inherit(backing_filename, reference, options, 0, bs,
Max Reitz25191e52020-05-13 13:05:33 +02003663 &child_of_bds, bdrv_backing_role(bs), errp);
Max Reitz5b363932016-05-17 16:41:31 +02003664 if (!backing_hd) {
Paolo Bonzini9156df12012-10-18 16:49:17 +02003665 bs->open_flags |= BDRV_O_NO_BACKING;
Markus Armbrustere43bfd92015-12-18 16:35:15 +01003666 error_prepend(errp, "Could not open backing file: ");
Max Reitz5b363932016-05-17 16:41:31 +02003667 ret = -EINVAL;
Benoît Canet1ba4b6a2014-04-22 17:05:27 +02003668 goto free_exit;
Paolo Bonzini9156df12012-10-18 16:49:17 +02003669 }
Kevin Wolfdf581792015-06-15 11:53:47 +02003670
Max Reitz998c2012019-02-01 20:29:08 +01003671 if (implicit_backing) {
3672 bdrv_refresh_filename(backing_hd);
3673 pstrcpy(bs->auto_backing_file, sizeof(bs->auto_backing_file),
3674 backing_hd->filename);
3675 }
3676
Kevin Wolf5db15a52015-09-14 15:33:33 +02003677 /* Hook up the backing file link; drop our reference, bs owns the
3678 * backing_hd reference now */
Vladimir Sementsov-Ogievskiydc9c10a2021-02-02 15:49:47 +03003679 ret = bdrv_set_backing_hd(bs, backing_hd, errp);
Kevin Wolf5db15a52015-09-14 15:33:33 +02003680 bdrv_unref(backing_hd);
Kevin Wolf8aa04542023-06-05 10:57:08 +02003681
Vladimir Sementsov-Ogievskiydc9c10a2021-02-02 15:49:47 +03003682 if (ret < 0) {
Kevin Wolf12fa4af2017-02-17 20:42:32 +01003683 goto free_exit;
3684 }
Peter Feinerd80ac652014-01-08 19:43:25 +00003685
Kevin Wolfd9b7b052015-01-16 18:23:41 +01003686 qdict_del(parent_options, bdref_key);
3687
Benoît Canet1ba4b6a2014-04-22 17:05:27 +02003688free_exit:
3689 g_free(backing_filename);
Marc-André Lureaucb3e7f02018-04-19 17:01:43 +02003690 qobject_unref(tmp_parent_options);
Benoît Canet1ba4b6a2014-04-22 17:05:27 +02003691 return ret;
Paolo Bonzini9156df12012-10-18 16:49:17 +02003692}
3693
Kevin Wolf2d6b86a2017-02-17 17:43:59 +01003694static BlockDriverState *
3695bdrv_open_child_bs(const char *filename, QDict *options, const char *bdref_key,
Max Reitzbd86fb92020-05-13 13:05:13 +02003696 BlockDriverState *parent, const BdrvChildClass *child_class,
Max Reitz272c02e2020-05-13 13:05:17 +02003697 BdrvChildRole child_role, bool allow_none, Error **errp)
Max Reitzda557aa2013-12-20 19:28:11 +01003698{
Kevin Wolf2d6b86a2017-02-17 17:43:59 +01003699 BlockDriverState *bs = NULL;
Max Reitzda557aa2013-12-20 19:28:11 +01003700 QDict *image_options;
Max Reitzda557aa2013-12-20 19:28:11 +01003701 char *bdref_key_dot;
3702 const char *reference;
3703
Max Reitzbd86fb92020-05-13 13:05:13 +02003704 assert(child_class != NULL);
Max Reitzf67503e2014-02-18 18:33:05 +01003705
Max Reitzda557aa2013-12-20 19:28:11 +01003706 bdref_key_dot = g_strdup_printf("%s.", bdref_key);
3707 qdict_extract_subqdict(options, &image_options, bdref_key_dot);
3708 g_free(bdref_key_dot);
3709
Markus Armbruster129c7d12017-03-30 19:43:12 +02003710 /*
3711 * Caution: while qdict_get_try_str() is fine, getting non-string
3712 * types would require more care. When @options come from
3713 * -blockdev or blockdev_add, its members are typed according to
3714 * the QAPI schema, but when they come from -drive, they're all
3715 * QString.
3716 */
Max Reitzda557aa2013-12-20 19:28:11 +01003717 reference = qdict_get_try_str(options, bdref_key);
3718 if (!filename && !reference && !qdict_size(image_options)) {
Kevin Wolfb4b059f2015-06-15 13:24:19 +02003719 if (!allow_none) {
Max Reitzda557aa2013-12-20 19:28:11 +01003720 error_setg(errp, "A block device must be specified for \"%s\"",
3721 bdref_key);
Max Reitzda557aa2013-12-20 19:28:11 +01003722 }
Marc-André Lureaucb3e7f02018-04-19 17:01:43 +02003723 qobject_unref(image_options);
Max Reitzda557aa2013-12-20 19:28:11 +01003724 goto done;
3725 }
3726
Max Reitz5b363932016-05-17 16:41:31 +02003727 bs = bdrv_open_inherit(filename, reference, image_options, 0,
Max Reitz272c02e2020-05-13 13:05:17 +02003728 parent, child_class, child_role, errp);
Max Reitz5b363932016-05-17 16:41:31 +02003729 if (!bs) {
Kevin Wolfdf581792015-06-15 11:53:47 +02003730 goto done;
3731 }
3732
Max Reitzda557aa2013-12-20 19:28:11 +01003733done:
3734 qdict_del(options, bdref_key);
Kevin Wolf2d6b86a2017-02-17 17:43:59 +01003735 return bs;
3736}
3737
3738/*
3739 * Opens a disk image whose options are given as BlockdevRef in another block
3740 * device's options.
3741 *
3742 * If allow_none is true, no image will be opened if filename is false and no
3743 * BlockdevRef is given. NULL will be returned, but errp remains unset.
3744 *
3745 * bdrev_key specifies the key for the image's BlockdevRef in the options QDict.
3746 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
3747 * itself, all options starting with "${bdref_key}." are considered part of the
3748 * BlockdevRef.
3749 *
3750 * The BlockdevRef will be removed from the options QDict.
Kevin Wolfaa269ff2023-05-25 14:47:07 +02003751 *
Stefan Hajnoczi23c983c2023-12-05 13:20:11 -05003752 * @parent can move to a different AioContext in this function.
Kevin Wolf2d6b86a2017-02-17 17:43:59 +01003753 */
3754BdrvChild *bdrv_open_child(const char *filename,
3755 QDict *options, const char *bdref_key,
3756 BlockDriverState *parent,
Max Reitzbd86fb92020-05-13 13:05:13 +02003757 const BdrvChildClass *child_class,
Max Reitz258b7762020-05-13 13:05:15 +02003758 BdrvChildRole child_role,
Kevin Wolf2d6b86a2017-02-17 17:43:59 +01003759 bool allow_none, Error **errp)
3760{
3761 BlockDriverState *bs;
Kevin Wolf8394c352023-06-05 10:57:04 +02003762 BdrvChild *child;
Kevin Wolf2d6b86a2017-02-17 17:43:59 +01003763
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05003764 GLOBAL_STATE_CODE();
3765
Max Reitzbd86fb92020-05-13 13:05:13 +02003766 bs = bdrv_open_child_bs(filename, options, bdref_key, parent, child_class,
Max Reitz272c02e2020-05-13 13:05:17 +02003767 child_role, allow_none, errp);
Kevin Wolf2d6b86a2017-02-17 17:43:59 +01003768 if (bs == NULL) {
3769 return NULL;
3770 }
3771
Stefan Hajnoczi6bc30f12023-12-05 13:20:02 -05003772 bdrv_graph_wrlock();
Kevin Wolf8394c352023-06-05 10:57:04 +02003773 child = bdrv_attach_child(parent, bs, bdref_key, child_class, child_role,
3774 errp);
Stefan Hajnoczi6bc30f12023-12-05 13:20:02 -05003775 bdrv_graph_wrunlock();
Kevin Wolf8394c352023-06-05 10:57:04 +02003776
3777 return child;
Kevin Wolfb4b059f2015-06-15 13:24:19 +02003778}
3779
Max Reitzbd86fb92020-05-13 13:05:13 +02003780/*
Vladimir Sementsov-Ogievskiy83930782022-07-26 23:11:21 +03003781 * Wrapper on bdrv_open_child() for most popular case: open primary child of bs.
Kevin Wolfaa269ff2023-05-25 14:47:07 +02003782 *
Stefan Hajnoczi23c983c2023-12-05 13:20:11 -05003783 * @parent can move to a different AioContext in this function.
Vladimir Sementsov-Ogievskiy83930782022-07-26 23:11:21 +03003784 */
3785int bdrv_open_file_child(const char *filename,
3786 QDict *options, const char *bdref_key,
3787 BlockDriverState *parent, Error **errp)
3788{
3789 BdrvChildRole role;
3790
3791 /* commit_top and mirror_top don't use this function */
3792 assert(!parent->drv->filtered_child_is_backing);
Vladimir Sementsov-Ogievskiy83930782022-07-26 23:11:21 +03003793 role = parent->drv->is_filter ?
3794 (BDRV_CHILD_FILTERED | BDRV_CHILD_PRIMARY) : BDRV_CHILD_IMAGE;
3795
Vladimir Sementsov-Ogievskiy5bb047472022-07-26 23:11:32 +03003796 if (!bdrv_open_child(filename, options, bdref_key, parent,
3797 &child_of_bds, role, false, errp))
3798 {
3799 return -EINVAL;
3800 }
Vladimir Sementsov-Ogievskiy83930782022-07-26 23:11:21 +03003801
Vladimir Sementsov-Ogievskiy5bb047472022-07-26 23:11:32 +03003802 return 0;
Vladimir Sementsov-Ogievskiy83930782022-07-26 23:11:21 +03003803}
3804
3805/*
Max Reitzbd86fb92020-05-13 13:05:13 +02003806 * TODO Future callers may need to specify parent/child_class in order for
3807 * option inheritance to work. Existing callers use it for the root node.
3808 */
Kevin Wolfe1d74bc2018-01-10 15:52:33 +01003809BlockDriverState *bdrv_open_blockdev_ref(BlockdevRef *ref, Error **errp)
3810{
3811 BlockDriverState *bs = NULL;
Kevin Wolfe1d74bc2018-01-10 15:52:33 +01003812 QObject *obj = NULL;
3813 QDict *qdict = NULL;
3814 const char *reference = NULL;
3815 Visitor *v = NULL;
3816
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05003817 GLOBAL_STATE_CODE();
3818
Kevin Wolfe1d74bc2018-01-10 15:52:33 +01003819 if (ref->type == QTYPE_QSTRING) {
3820 reference = ref->u.reference;
3821 } else {
3822 BlockdevOptions *options = &ref->u.definition;
3823 assert(ref->type == QTYPE_QDICT);
3824
3825 v = qobject_output_visitor_new(&obj);
Markus Armbruster1f584242020-04-24 10:43:35 +02003826 visit_type_BlockdevOptions(v, NULL, &options, &error_abort);
Kevin Wolfe1d74bc2018-01-10 15:52:33 +01003827 visit_complete(v, &obj);
3828
Max Reitz7dc847e2018-02-24 16:40:29 +01003829 qdict = qobject_to(QDict, obj);
Kevin Wolfe1d74bc2018-01-10 15:52:33 +01003830 qdict_flatten(qdict);
3831
3832 /* bdrv_open_inherit() defaults to the values in bdrv_flags (for
3833 * compatibility with other callers) rather than what we want as the
3834 * real defaults. Apply the defaults here instead. */
3835 qdict_set_default_str(qdict, BDRV_OPT_CACHE_DIRECT, "off");
3836 qdict_set_default_str(qdict, BDRV_OPT_CACHE_NO_FLUSH, "off");
3837 qdict_set_default_str(qdict, BDRV_OPT_READ_ONLY, "off");
Kevin Wolfe35bdc12018-10-05 18:57:40 +02003838 qdict_set_default_str(qdict, BDRV_OPT_AUTO_READ_ONLY, "off");
3839
Kevin Wolfe1d74bc2018-01-10 15:52:33 +01003840 }
3841
Max Reitz272c02e2020-05-13 13:05:17 +02003842 bs = bdrv_open_inherit(NULL, reference, qdict, 0, NULL, NULL, 0, errp);
Kevin Wolfe1d74bc2018-01-10 15:52:33 +01003843 obj = NULL;
Marc-André Lureaucb3e7f02018-04-19 17:01:43 +02003844 qobject_unref(obj);
Kevin Wolfe1d74bc2018-01-10 15:52:33 +01003845 visit_free(v);
3846 return bs;
3847}
3848
Max Reitz66836182016-05-17 16:41:27 +02003849static BlockDriverState *bdrv_append_temp_snapshot(BlockDriverState *bs,
3850 int flags,
3851 QDict *snapshot_options,
3852 Error **errp)
Kevin Wolfb9988752014-04-03 12:09:34 +02003853{
Bin Meng69fbfff2022-10-10 12:04:31 +08003854 g_autofree char *tmp_filename = NULL;
Kevin Wolfb9988752014-04-03 12:09:34 +02003855 int64_t total_size;
Chunyan Liu83d05212014-06-05 17:20:51 +08003856 QemuOpts *opts = NULL;
Eric Blakeff6ed712017-04-27 16:58:18 -05003857 BlockDriverState *bs_snapshot = NULL;
Kevin Wolfb9988752014-04-03 12:09:34 +02003858 int ret;
3859
Emanuele Giuseppe Espositobdb73472022-03-03 10:16:02 -05003860 GLOBAL_STATE_CODE();
3861
Kevin Wolfb9988752014-04-03 12:09:34 +02003862 /* if snapshot, we create a temporary backing file and open it
3863 instead of opening 'filename' directly */
3864
3865 /* Get the required size from the image */
Kevin Wolff1877432014-04-04 17:07:19 +02003866 total_size = bdrv_getlength(bs);
Kevin Wolff665f012023-06-05 10:57:07 +02003867
Kevin Wolff1877432014-04-04 17:07:19 +02003868 if (total_size < 0) {
3869 error_setg_errno(errp, -total_size, "Could not get image size");
Benoît Canet1ba4b6a2014-04-22 17:05:27 +02003870 goto out;
Kevin Wolff1877432014-04-04 17:07:19 +02003871 }
Kevin Wolfb9988752014-04-03 12:09:34 +02003872
3873 /* Create the temporary image */
Bin Meng69fbfff2022-10-10 12:04:31 +08003874 tmp_filename = create_tmp_file(errp);
3875 if (!tmp_filename) {
Benoît Canet1ba4b6a2014-04-22 17:05:27 +02003876 goto out;
Kevin Wolfb9988752014-04-03 12:09:34 +02003877 }
3878
Max Reitzef810432014-12-02 18:32:42 +01003879 opts = qemu_opts_create(bdrv_qcow2.create_opts, NULL, 0,
Chunyan Liuc282e1f2014-06-05 17:21:11 +08003880 &error_abort);
Markus Armbruster39101f22015-02-12 16:46:36 +01003881 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, total_size, &error_abort);
Markus Armbrustere43bfd92015-12-18 16:35:15 +01003882 ret = bdrv_create(&bdrv_qcow2, tmp_filename, opts, errp);
Chunyan Liu83d05212014-06-05 17:20:51 +08003883 qemu_opts_del(opts);
Kevin Wolfb9988752014-04-03 12:09:34 +02003884 if (ret < 0) {
Markus Armbrustere43bfd92015-12-18 16:35:15 +01003885 error_prepend(errp, "Could not create temporary overlay '%s': ",
3886 tmp_filename);
Benoît Canet1ba4b6a2014-04-22 17:05:27 +02003887 goto out;
Kevin Wolfb9988752014-04-03 12:09:34 +02003888 }
3889
Kevin Wolf73176be2016-03-07 13:02:15 +01003890 /* Prepare options QDict for the temporary file */
Eric Blake46f5ac22017-04-27 16:58:17 -05003891 qdict_put_str(snapshot_options, "file.driver", "file");
3892 qdict_put_str(snapshot_options, "file.filename", tmp_filename);
3893 qdict_put_str(snapshot_options, "driver", "qcow2");
Kevin Wolfb9988752014-04-03 12:09:34 +02003894
Max Reitz5b363932016-05-17 16:41:31 +02003895 bs_snapshot = bdrv_open(NULL, NULL, snapshot_options, flags, errp);
Kevin Wolf73176be2016-03-07 13:02:15 +01003896 snapshot_options = NULL;
Max Reitz5b363932016-05-17 16:41:31 +02003897 if (!bs_snapshot) {
Benoît Canet1ba4b6a2014-04-22 17:05:27 +02003898 goto out;
Kevin Wolfb9988752014-04-03 12:09:34 +02003899 }
3900
Vladimir Sementsov-Ogievskiy934aee12021-02-02 15:49:44 +03003901 ret = bdrv_append(bs_snapshot, bs, errp);
3902 if (ret < 0) {
Eric Blakeff6ed712017-04-27 16:58:18 -05003903 bs_snapshot = NULL;
Kevin Wolfb2c28322017-02-20 12:46:42 +01003904 goto out;
3905 }
Benoît Canet1ba4b6a2014-04-22 17:05:27 +02003906
3907out:
Marc-André Lureaucb3e7f02018-04-19 17:01:43 +02003908 qobject_unref(snapshot_options);
Eric Blakeff6ed712017-04-27 16:58:18 -05003909 return bs_snapshot;
Kevin Wolfb9988752014-04-03 12:09:34 +02003910}
3911
Max Reitzda557aa2013-12-20 19:28:11 +01003912/*
Kevin Wolfb6ce07a2010-04-12 16:37:13 +02003913 * Opens a disk image (raw, qcow2, vmdk, ...)
Kevin Wolfde9c0ce2013-03-15 10:35:02 +01003914 *
3915 * options is a QDict of options to pass to the block drivers, or NULL for an
3916 * empty set of options. The reference to the QDict belongs to the block layer
3917 * after the call (even on failure), so if the caller intends to reuse the
Marc-André Lureaucb3e7f02018-04-19 17:01:43 +02003918 * dictionary, it needs to use qobject_ref() before calling bdrv_open.
Max Reitzf67503e2014-02-18 18:33:05 +01003919 *
3920 * If *pbs is NULL, a new BDS will be created with a pointer to it stored there.
3921 * If it is not NULL, the referenced BDS will be reused.
Max Reitzddf56362014-02-18 18:33:06 +01003922 *
3923 * The reference parameter may be used to specify an existing block device which
3924 * should be opened. If specified, neither options nor a filename may be given,
3925 * nor can an existing BDS be reused (that is, *pbs has to be NULL).
Kevin Wolfb6ce07a2010-04-12 16:37:13 +02003926 */
Kevin Wolf32192302023-01-26 18:24:32 +01003927static BlockDriverState * no_coroutine_fn
3928bdrv_open_inherit(const char *filename, const char *reference, QDict *options,
3929 int flags, BlockDriverState *parent,
3930 const BdrvChildClass *child_class, BdrvChildRole child_role,
3931 Error **errp)
bellardea2384d2004-08-01 21:59:26 +00003932{
Kevin Wolfb6ce07a2010-04-12 16:37:13 +02003933 int ret;
Kevin Wolf5696c6e2017-02-17 18:39:24 +01003934 BlockBackend *file = NULL;
Kevin Wolf9a4f4c32015-06-16 14:19:22 +02003935 BlockDriverState *bs;
Max Reitzce343772015-08-26 19:47:50 +02003936 BlockDriver *drv = NULL;
Alberto Garcia2f624b82018-06-29 14:37:00 +03003937 BdrvChild *child;
Kevin Wolf74fe54f2013-07-09 11:09:02 +02003938 const char *drvname;
Alberto Garcia3e8c2e52015-10-26 14:27:15 +02003939 const char *backing;
Max Reitz34b5d2c2013-09-05 14:45:29 +02003940 Error *local_err = NULL;
Kevin Wolf73176be2016-03-07 13:02:15 +01003941 QDict *snapshot_options = NULL;
Kevin Wolfb1e6fc02014-05-06 12:11:42 +02003942 int snapshot_flags = 0;
bellard712e7872005-04-28 21:09:32 +00003943
Max Reitzbd86fb92020-05-13 13:05:13 +02003944 assert(!child_class || !flags);
3945 assert(!child_class == !parent);
Emanuele Giuseppe Espositof0c28322022-03-03 10:16:13 -05003946 GLOBAL_STATE_CODE();
Kevin Wolf32192302023-01-26 18:24:32 +01003947 assert(!qemu_in_coroutine());
Max Reitzf67503e2014-02-18 18:33:05 +01003948
Kevin Wolf356f4ef2023-09-11 11:46:15 +02003949 /* TODO We'll eventually have to take a writer lock in this function */
3950 GRAPH_RDLOCK_GUARD_MAINLOOP();
3951
Max Reitzddf56362014-02-18 18:33:06 +01003952 if (reference) {
3953 bool options_non_empty = options ? qdict_size(options) : false;
Marc-André Lureaucb3e7f02018-04-19 17:01:43 +02003954 qobject_unref(options);
Max Reitzddf56362014-02-18 18:33:06 +01003955
Max Reitzddf56362014-02-18 18:33:06 +01003956 if (filename || options_non_empty) {
3957 error_setg(errp, "Cannot reference an existing block device with "
3958 "additional options or a new filename");
Max Reitz5b363932016-05-17 16:41:31 +02003959 return NULL;
Max Reitzddf56362014-02-18 18:33:06 +01003960 }
3961
3962 bs = bdrv_lookup_bs(reference, reference, errp);
3963 if (!bs) {
Max Reitz5b363932016-05-17 16:41:31 +02003964 return NULL;
Max Reitzddf56362014-02-18 18:33:06 +01003965 }
Kevin Wolf76b22322016-04-04 17:11:13 +02003966
Max Reitzddf56362014-02-18 18:33:06 +01003967 bdrv_ref(bs);
Max Reitz5b363932016-05-17 16:41:31 +02003968 return bs;
Max Reitzddf56362014-02-18 18:33:06 +01003969 }
3970
Max Reitz5b363932016-05-17 16:41:31 +02003971 bs = bdrv_new();
Max Reitzf67503e2014-02-18 18:33:05 +01003972
Kevin Wolfde9c0ce2013-03-15 10:35:02 +01003973 /* NULL means an empty set of options */
3974 if (options == NULL) {
3975 options = qdict_new();
3976 }
3977
Kevin Wolf145f5982015-05-08 16:15:03 +02003978 /* json: syntax counts as explicit options, as if in the QDict */
Kevin Wolfde3b53f2015-10-29 15:24:41 +01003979 parse_json_protocol(options, &filename, &local_err);
3980 if (local_err) {
Kevin Wolfde3b53f2015-10-29 15:24:41 +01003981 goto fail;
3982 }
3983
Kevin Wolf145f5982015-05-08 16:15:03 +02003984 bs->explicit_options = qdict_clone_shallow(options);
3985
Max Reitzbd86fb92020-05-13 13:05:13 +02003986 if (child_class) {
Max Reitz3cdc69d2020-05-13 13:05:18 +02003987 bool parent_is_format;
3988
3989 if (parent->drv) {
3990 parent_is_format = parent->drv->is_format;
3991 } else {
3992 /*
3993 * parent->drv is not set yet because this node is opened for
3994 * (potential) format probing. That means that @parent is going
3995 * to be a format node.
3996 */
3997 parent_is_format = true;
3998 }
3999
Kevin Wolfbddcec32015-04-09 18:47:50 +02004000 bs->inherits_from = parent;
Max Reitz3cdc69d2020-05-13 13:05:18 +02004001 child_class->inherit_options(child_role, parent_is_format,
4002 &flags, options,
Max Reitzbd86fb92020-05-13 13:05:13 +02004003 parent->open_flags, parent->options);
Kevin Wolff3930ed2015-04-08 13:43:47 +02004004 }
4005
Kevin Wolfde3b53f2015-10-29 15:24:41 +01004006 ret = bdrv_fill_options(&options, filename, &flags, &local_err);
Philippe Mathieu-Daudédfde4832020-04-22 15:31:44 +02004007 if (ret < 0) {
Kevin Wolf462f5bc2014-05-26 11:39:55 +02004008 goto fail;
4009 }
4010
Markus Armbruster129c7d12017-03-30 19:43:12 +02004011 /*
4012 * Set the BDRV_O_RDWR and BDRV_O_ALLOW_RDWR flags.
4013 * Caution: getting a boolean member of @options requires care.
4014 * When @options come from -blockdev or blockdev_add, members are
4015 * typed according to the QAPI schema, but when they come from
4016 * -drive, they're all QString.
4017 */
Alberto Garciaf87a0e22016-09-15 17:53:02 +03004018 if (g_strcmp0(qdict_get_try_str(options, BDRV_OPT_READ_ONLY), "on") &&
4019 !qdict_get_try_bool(options, BDRV_OPT_READ_ONLY, false)) {
4020 flags |= (BDRV_O_RDWR | BDRV_O_ALLOW_RDWR);
4021 } else {
4022 flags &= ~BDRV_O_RDWR;
Alberto Garcia14499ea2016-09-15 17:53:00 +03004023 }
4024
4025 if (flags & BDRV_O_SNAPSHOT) {
4026 snapshot_options = qdict_new();
4027 bdrv_temp_snapshot_options(&snapshot_flags, snapshot_options,
4028 flags, options);
Alberto Garciaf87a0e22016-09-15 17:53:02 +03004029 /* Let bdrv_backing_options() override "read-only" */
4030 qdict_del(options, BDRV_OPT_READ_ONLY);
Max Reitz00ff7ff2020-05-13 13:05:21 +02004031 bdrv_inherited_options(BDRV_CHILD_COW, true,
4032 &flags, options, flags, options);
Alberto Garcia14499ea2016-09-15 17:53:00 +03004033 }
4034
Kevin Wolf62392eb2015-04-24 16:38:02 +02004035 bs->open_flags = flags;
4036 bs->options = options;
4037 options = qdict_clone_shallow(options);
4038
Kevin Wolf76c591b2014-06-04 14:19:44 +02004039 /* Find the right image format driver */
Markus Armbruster129c7d12017-03-30 19:43:12 +02004040 /* See cautionary note on accessing @options above */
Kevin Wolf76c591b2014-06-04 14:19:44 +02004041 drvname = qdict_get_try_str(options, "driver");
4042 if (drvname) {
4043 drv = bdrv_find_format(drvname);
Kevin Wolf76c591b2014-06-04 14:19:44 +02004044 if (!drv) {
4045 error_setg(errp, "Unknown driver: '%s'", drvname);
Kevin Wolf76c591b2014-06-04 14:19:44 +02004046 goto fail;
4047 }
4048 }
4049
4050 assert(drvname || !(flags & BDRV_O_PROTOCOL));
Kevin Wolf76c591b2014-06-04 14:19:44 +02004051
Markus Armbruster129c7d12017-03-30 19:43:12 +02004052 /* See cautionary note on accessing @options above */
Alberto Garcia3e8c2e52015-10-26 14:27:15 +02004053 backing = qdict_get_try_str(options, "backing");
Max Reitze59a0cf2018-02-24 16:40:32 +01004054 if (qobject_to(QNull, qdict_get(options, "backing")) != NULL ||
4055 (backing && *backing == '\0'))
4056 {
Max Reitz4f7be282018-02-24 16:40:33 +01004057 if (backing) {
4058 warn_report("Use of \"backing\": \"\" is deprecated; "
4059 "use \"backing\": null instead");
4060 }
Alberto Garcia3e8c2e52015-10-26 14:27:15 +02004061 flags |= BDRV_O_NO_BACKING;
Kevin Wolfae0f57f2019-11-08 09:36:35 +01004062 qdict_del(bs->explicit_options, "backing");
4063 qdict_del(bs->options, "backing");
Alberto Garcia3e8c2e52015-10-26 14:27:15 +02004064 qdict_del(options, "backing");
4065 }
4066
Kevin Wolf5696c6e2017-02-17 18:39:24 +01004067 /* Open image file without format layer. This BlockBackend is only used for
Kevin Wolf4e4bf5c2016-12-16 18:52:37 +01004068 * probing, the block drivers will do their own bdrv_open_child() for the
4069 * same BDS, which is why we put the node name back into options. */
Kevin Wolff4788ad2014-06-03 16:44:19 +02004070 if ((flags & BDRV_O_PROTOCOL) == 0) {
Kevin Wolf5696c6e2017-02-17 18:39:24 +01004071 BlockDriverState *file_bs;
4072
4073 file_bs = bdrv_open_child_bs(filename, options, "file", bs,
Max Reitz58944402020-05-13 13:05:37 +02004074 &child_of_bds, BDRV_CHILD_IMAGE,
4075 true, &local_err);
Kevin Wolf1fdd6932015-06-15 14:11:51 +02004076 if (local_err) {
Max Reitz5469a2a2014-02-18 18:33:10 +01004077 goto fail;
4078 }
Kevin Wolf5696c6e2017-02-17 18:39:24 +01004079 if (file_bs != NULL) {
Kevin Wolfdacaa162017-11-20 14:59:13 +01004080 /* Not requesting BLK_PERM_CONSISTENT_READ because we're only
4081 * looking at the header to guess the image format. This works even
4082 * in cases where a guest would not see a consistent state. */
Stefan Hajnoczib49f4752023-12-05 13:20:03 -05004083 AioContext *ctx = bdrv_get_aio_context(file_bs);
Kevin Wolff665f012023-06-05 10:57:07 +02004084 file = blk_new(ctx, 0, BLK_PERM_ALL);
Kevin Wolfd7086422017-01-13 19:02:32 +01004085 blk_insert_bs(file, file_bs, &local_err);
Kevin Wolf5696c6e2017-02-17 18:39:24 +01004086 bdrv_unref(file_bs);
Kevin Wolff665f012023-06-05 10:57:07 +02004087
Kevin Wolfd7086422017-01-13 19:02:32 +01004088 if (local_err) {
4089 goto fail;
4090 }
Kevin Wolf5696c6e2017-02-17 18:39:24 +01004091
Eric Blake46f5ac22017-04-27 16:58:17 -05004092 qdict_put_str(options, "file", bdrv_get_node_name(file_bs));
Kevin Wolf4e4bf5c2016-12-16 18:52:37 +01004093 }
Max Reitz5469a2a2014-02-18 18:33:10 +01004094 }
4095
Kevin Wolf76c591b2014-06-04 14:19:44 +02004096 /* Image format probing */
Kevin Wolf38f3ef52014-11-20 16:27:12 +01004097 bs->probed = !drv;
Kevin Wolf76c591b2014-06-04 14:19:44 +02004098 if (!drv && file) {
Kevin Wolfcf2ab8f2016-06-20 18:24:02 +02004099 ret = find_image_format(file, filename, &drv, &local_err);
Kevin Wolf17b005f2014-05-27 10:50:29 +02004100 if (ret < 0) {
Kevin Wolf8bfea152014-04-11 19:16:36 +02004101 goto fail;
Max Reitz2a05cbe2013-12-20 19:28:10 +01004102 }
Kevin Wolf62392eb2015-04-24 16:38:02 +02004103 /*
4104 * This option update would logically belong in bdrv_fill_options(),
4105 * but we first need to open bs->file for the probing to work, while
4106 * opening bs->file already requires the (mostly) final set of options
4107 * so that cache mode etc. can be inherited.
4108 *
4109 * Adding the driver later is somewhat ugly, but it's not an option
4110 * that would ever be inherited, so it's correct. We just need to make
4111 * sure to update both bs->options (which has the full effective
4112 * options for bs) and options (which has file.* already removed).
4113 */
Eric Blake46f5ac22017-04-27 16:58:17 -05004114 qdict_put_str(bs->options, "driver", drv->format_name);
4115 qdict_put_str(options, "driver", drv->format_name);
Kevin Wolf76c591b2014-06-04 14:19:44 +02004116 } else if (!drv) {
Kevin Wolf17b005f2014-05-27 10:50:29 +02004117 error_setg(errp, "Must specify either driver or file");
Kevin Wolf8bfea152014-04-11 19:16:36 +02004118 goto fail;
Kevin Wolff500a6d2012-11-12 17:35:27 +01004119 }
4120
Max Reitz53a29512015-03-19 14:53:16 -04004121 /* BDRV_O_PROTOCOL must be set iff a protocol BDS is about to be created */
4122 assert(!!(flags & BDRV_O_PROTOCOL) == !!drv->bdrv_file_open);
4123 /* file must be NULL if a protocol BDS is about to be created
4124 * (the inverse results in an error message from bdrv_open_common()) */
4125 assert(!(flags & BDRV_O_PROTOCOL) || !file);
4126
Kevin Wolfb6ce07a2010-04-12 16:37:13 +02004127 /* Open the image */
Kevin Wolf82dc8b42016-01-11 19:07:50 +01004128 ret = bdrv_open_common(bs, file, options, &local_err);
Kevin Wolfb6ce07a2010-04-12 16:37:13 +02004129 if (ret < 0) {
Kevin Wolf8bfea152014-04-11 19:16:36 +02004130 goto fail;
Christoph Hellwig69873072010-01-20 18:13:25 +01004131 }
4132
Kevin Wolf4e4bf5c2016-12-16 18:52:37 +01004133 if (file) {
Kevin Wolf5696c6e2017-02-17 18:39:24 +01004134 blk_unref(file);
Kevin Wolff500a6d2012-11-12 17:35:27 +01004135 file = NULL;
4136 }
4137
Kevin Wolfb6ce07a2010-04-12 16:37:13 +02004138 /* If there is a backing file, use it */
Paolo Bonzini9156df12012-10-18 16:49:17 +02004139 if ((flags & BDRV_O_NO_BACKING) == 0) {
Kevin Wolfd9b7b052015-01-16 18:23:41 +01004140 ret = bdrv_open_backing_file(bs, options, "backing", &local_err);
Kevin Wolfb6ce07a2010-04-12 16:37:13 +02004141 if (ret < 0) {
Kevin Wolfb6ad4912013-03-15 10:35:04 +01004142 goto close_and_fail;
Kevin Wolfb6ce07a2010-04-12 16:37:13 +02004143 }
Kevin Wolfb6ce07a2010-04-12 16:37:13 +02004144 }
4145
Alberto Garcia50196d72018-09-06 12:37:03 +03004146 /* Remove all children options and references
4147 * from bs->options and bs->explicit_options */
Alberto Garcia2f624b82018-06-29 14:37:00 +03004148 QLIST_FOREACH(child, &bs->children, next) {
4149 char *child_key_dot;
4150 child_key_dot = g_strdup_printf("%s.", child->name);
4151 qdict_extract_subqdict(bs->explicit_options, NULL, child_key_dot);
4152 qdict_extract_subqdict(bs->options, NULL, child_key_dot);
Alberto Garcia50196d72018-09-06 12:37:03 +03004153 qdict_del(bs->explicit_options, child->name);
4154 qdict_del(bs->options, child->name);
Alberto Garcia2f624b82018-06-29 14:37:00 +03004155 g_free(child_key_dot);
4156 }
4157
Kevin Wolfb6ad4912013-03-15 10:35:04 +01004158 /* Check if any unknown options were used */
Paolo Bonzini7ad27572017-01-04 15:59:14 +01004159 if (qdict_size(options) != 0) {
Kevin Wolfb6ad4912013-03-15 10:35:04 +01004160 const QDictEntry *entry = qdict_first(options);
Max Reitz5acd9d82014-02-18 18:33:11 +01004161 if (flags & BDRV_O_PROTOCOL) {
4162 error_setg(errp, "Block protocol '%s' doesn't support the option "
4163 "'%s'", drv->format_name, entry->key);
4164 } else {
Max Reitzd0e46a52016-03-16 19:54:34 +01004165 error_setg(errp,
4166 "Block format '%s' does not support the option '%s'",
4167 drv->format_name, entry->key);
Max Reitz5acd9d82014-02-18 18:33:11 +01004168 }
Kevin Wolfb6ad4912013-03-15 10:35:04 +01004169
Kevin Wolfb6ad4912013-03-15 10:35:04 +01004170 goto close_and_fail;
4171 }
Kevin Wolfb6ad4912013-03-15 10:35:04 +01004172
Daniel P. Berrangec01c2142017-06-23 17:24:16 +01004173 bdrv_parent_cb_change_media(bs, true);
Kevin Wolfb6ce07a2010-04-12 16:37:13 +02004174
Marc-André Lureaucb3e7f02018-04-19 17:01:43 +02004175 qobject_unref(options);
Alberto Garcia8961be32018-09-06 17:25:41 +03004176 options = NULL;
Kevin Wolfdd62f1c2015-06-18 14:09:57 +02004177
4178 /* For snapshot=on, create a temporary qcow2 overlay. bs points to the
4179 * temporary snapshot afterwards. */
4180 if (snapshot_flags) {
Max Reitz66836182016-05-17 16:41:27 +02004181 BlockDriverState *snapshot_bs;
4182 snapshot_bs = bdrv_append_temp_snapshot(bs, snapshot_flags,
4183 snapshot_options, &local_err);
Kevin Wolf73176be2016-03-07 13:02:15 +01004184 snapshot_options = NULL;
Kevin Wolfdd62f1c2015-06-18 14:09:57 +02004185 if (local_err) {
4186 goto close_and_fail;
4187 }
Max Reitz5b363932016-05-17 16:41:31 +02004188 /* We are not going to return bs but the overlay on top of it
4189 * (snapshot_bs); thus, we have to drop the strong reference to bs
4190 * (which we obtained by calling bdrv_new()). bs will not be deleted,
4191 * though, because the overlay still has a reference to it. */
4192 bdrv_unref(bs);
4193 bs = snapshot_bs;
Max Reitz66836182016-05-17 16:41:27 +02004194 }
4195
Max Reitz5b363932016-05-17 16:41:31 +02004196 return bs;
Kevin Wolfb6ce07a2010-04-12 16:37:13 +02004197
Kevin Wolf8bfea152014-04-11 19:16:36 +02004198fail:
Kevin Wolf5696c6e2017-02-17 18:39:24 +01004199 blk_unref(file);
Marc-André Lureaucb3e7f02018-04-19 17:01:43 +02004200 qobject_unref(snapshot_options);
4201 qobject_unref(bs->explicit_options);
4202 qobject_unref(bs->options);
4203 qobject_unref(options);
Kevin Wolfde9c0ce2013-03-15 10:35:02 +01004204 bs->options = NULL;
Manos Pitsidianakis998cbd62017-07-14 17:35:47 +03004205 bs->explicit_options = NULL;
Max Reitz5b363932016-05-17 16:41:31 +02004206 bdrv_unref(bs);
Eduardo Habkost621ff942016-06-13 18:57:56 -03004207 error_propagate(errp, local_err);
Max Reitz5b363932016-05-17 16:41:31 +02004208 return NULL;
Kevin Wolfde9c0ce2013-03-15 10:35:02 +01004209
Kevin Wolfb6ad4912013-03-15 10:35:04 +01004210close_and_fail:
Max Reitz5b363932016-05-17 16:41:31 +02004211 bdrv_unref(bs);
Marc-André Lureaucb3e7f02018-04-19 17:01:43 +02004212 qobject_unref(snapshot_options);
4213 qobject_unref(options);
Eduardo Habkost621ff942016-06-13 18:57:56 -03004214 error_propagate(errp, local_err);
Max Reitz5b363932016-05-17 16:41:31 +02004215 return NULL;
Kevin Wolfb6ce07a2010-04-12 16:37:13 +02004216}
4217
Max Reitz5b363932016-05-17 16:41:31 +02004218BlockDriverState *bdrv_open(const char *filename, const char *reference,
4219 QDict *options, int flags, Error **errp)
Kevin Wolff3930ed2015-04-08 13:43:47 +02004220{
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05004221 GLOBAL_STATE_CODE();
4222
Max Reitz5b363932016-05-17 16:41:31 +02004223 return bdrv_open_inherit(filename, reference, options, flags, NULL,
Max Reitz272c02e2020-05-13 13:05:17 +02004224 NULL, 0, errp);
Kevin Wolff3930ed2015-04-08 13:43:47 +02004225}
4226
Alberto Garciafaf116b2019-03-12 18:48:49 +02004227/* Return true if the NULL-terminated @list contains @str */
4228static bool is_str_in_list(const char *str, const char *const *list)
4229{
4230 if (str && list) {
4231 int i;
4232 for (i = 0; list[i] != NULL; i++) {
4233 if (!strcmp(str, list[i])) {
4234 return true;
4235 }
4236 }
4237 }
4238 return false;
4239}
4240
4241/*
4242 * Check that every option set in @bs->options is also set in
4243 * @new_opts.
4244 *
4245 * Options listed in the common_options list and in
4246 * @bs->drv->mutable_opts are skipped.
4247 *
4248 * Return 0 on success, otherwise return -EINVAL and set @errp.
4249 */
4250static int bdrv_reset_options_allowed(BlockDriverState *bs,
4251 const QDict *new_opts, Error **errp)
4252{
4253 const QDictEntry *e;
4254 /* These options are common to all block drivers and are handled
4255 * in bdrv_reopen_prepare() so they can be left out of @new_opts */
4256 const char *const common_options[] = {
4257 "node-name", "discard", "cache.direct", "cache.no-flush",
4258 "read-only", "auto-read-only", "detect-zeroes", NULL
4259 };
4260
4261 for (e = qdict_first(bs->options); e; e = qdict_next(bs->options, e)) {
4262 if (!qdict_haskey(new_opts, e->key) &&
4263 !is_str_in_list(e->key, common_options) &&
4264 !is_str_in_list(e->key, bs->drv->mutable_opts)) {
4265 error_setg(errp, "Option '%s' cannot be reset "
4266 "to its default value", e->key);
4267 return -EINVAL;
4268 }
4269 }
4270
4271 return 0;
4272}
4273
Jeff Codye971aa12012-09-20 15:13:19 -04004274/*
Alberto Garciacb828c32019-03-12 18:48:47 +02004275 * Returns true if @child can be reached recursively from @bs
4276 */
Kevin Wolfce433d22023-09-29 16:51:43 +02004277static bool GRAPH_RDLOCK
4278bdrv_recurse_has_child(BlockDriverState *bs, BlockDriverState *child)
Alberto Garciacb828c32019-03-12 18:48:47 +02004279{
4280 BdrvChild *c;
4281
4282 if (bs == child) {
4283 return true;
4284 }
4285
4286 QLIST_FOREACH(c, &bs->children, next) {
4287 if (bdrv_recurse_has_child(c->bs, child)) {
4288 return true;
4289 }
4290 }
4291
4292 return false;
4293}
4294
4295/*
Jeff Codye971aa12012-09-20 15:13:19 -04004296 * Adds a BlockDriverState to a simple queue for an atomic, transactional
4297 * reopen of multiple devices.
4298 *
Vladimir Sementsov-Ogievskiy859443b2019-09-27 15:23:47 +03004299 * bs_queue can either be an existing BlockReopenQueue that has had QTAILQ_INIT
Jeff Codye971aa12012-09-20 15:13:19 -04004300 * already performed, or alternatively may be NULL a new BlockReopenQueue will
4301 * be created and initialized. This newly created BlockReopenQueue should be
4302 * passed back in for subsequent calls that are intended to be of the same
4303 * atomic 'set'.
4304 *
4305 * bs is the BlockDriverState to add to the reopen queue.
4306 *
Kevin Wolf4d2cb092015-04-10 17:50:50 +02004307 * options contains the changed options for the associated bs
4308 * (the BlockReopenQueue takes ownership)
4309 *
Jeff Codye971aa12012-09-20 15:13:19 -04004310 * flags contains the open flags for the associated bs
4311 *
4312 * returns a pointer to bs_queue, which is either the newly allocated
4313 * bs_queue, or the existing bs_queue being used.
4314 *
Kevin Wolfd22933a2022-11-18 18:41:02 +01004315 * bs is drained here and undrained by bdrv_reopen_queue_free().
Kevin Wolf2e117862022-11-18 18:41:01 +01004316 *
4317 * To be called with bs->aio_context locked.
Jeff Codye971aa12012-09-20 15:13:19 -04004318 */
Kevin Wolfce433d22023-09-29 16:51:43 +02004319static BlockReopenQueue * GRAPH_RDLOCK
4320bdrv_reopen_queue_child(BlockReopenQueue *bs_queue, BlockDriverState *bs,
4321 QDict *options, const BdrvChildClass *klass,
4322 BdrvChildRole role, bool parent_is_format,
4323 QDict *parent_options, int parent_flags,
4324 bool keep_old_opts)
Jeff Codye971aa12012-09-20 15:13:19 -04004325{
4326 assert(bs != NULL);
4327
4328 BlockReopenQueueEntry *bs_entry;
Kevin Wolf67251a32015-04-09 18:54:04 +02004329 BdrvChild *child;
Alberto Garcia9aa09dd2018-11-12 16:00:45 +02004330 QDict *old_options, *explicit_options, *options_copy;
4331 int flags;
4332 QemuOpts *opts;
Kevin Wolf67251a32015-04-09 18:54:04 +02004333
Emanuele Giuseppe Espositof0c28322022-03-03 10:16:13 -05004334 GLOBAL_STATE_CODE();
Kevin Wolf1a63a902017-12-06 20:24:44 +01004335
Kevin Wolfce433d22023-09-29 16:51:43 +02004336 /*
4337 * Strictly speaking, draining is illegal under GRAPH_RDLOCK. We know that
4338 * we've been called with bdrv_graph_rdlock_main_loop(), though, so it's ok
4339 * in practice.
4340 */
Kevin Wolfd22933a2022-11-18 18:41:02 +01004341 bdrv_drained_begin(bs);
4342
Jeff Codye971aa12012-09-20 15:13:19 -04004343 if (bs_queue == NULL) {
4344 bs_queue = g_new0(BlockReopenQueue, 1);
Vladimir Sementsov-Ogievskiy859443b2019-09-27 15:23:47 +03004345 QTAILQ_INIT(bs_queue);
Jeff Codye971aa12012-09-20 15:13:19 -04004346 }
4347
Kevin Wolf4d2cb092015-04-10 17:50:50 +02004348 if (!options) {
4349 options = qdict_new();
4350 }
4351
Alberto Garcia5b7ba052016-09-15 17:53:03 +03004352 /* Check if this BlockDriverState is already in the queue */
Vladimir Sementsov-Ogievskiy859443b2019-09-27 15:23:47 +03004353 QTAILQ_FOREACH(bs_entry, bs_queue, entry) {
Alberto Garcia5b7ba052016-09-15 17:53:03 +03004354 if (bs == bs_entry->state.bs) {
4355 break;
4356 }
4357 }
4358
Kevin Wolf28518102015-05-08 17:07:31 +02004359 /*
4360 * Precedence of options:
4361 * 1. Explicitly passed in options (highest)
Alberto Garcia9aa09dd2018-11-12 16:00:45 +02004362 * 2. Retained from explicitly set options of bs
4363 * 3. Inherited from parent node
4364 * 4. Retained from effective options of bs
Kevin Wolf28518102015-05-08 17:07:31 +02004365 */
4366
Kevin Wolf145f5982015-05-08 16:15:03 +02004367 /* Old explicitly set values (don't overwrite by inherited value) */
Alberto Garcia077e8e22019-03-12 18:48:44 +02004368 if (bs_entry || keep_old_opts) {
4369 old_options = qdict_clone_shallow(bs_entry ?
4370 bs_entry->state.explicit_options :
4371 bs->explicit_options);
4372 bdrv_join_options(bs, options, old_options);
4373 qobject_unref(old_options);
Alberto Garcia5b7ba052016-09-15 17:53:03 +03004374 }
Kevin Wolf145f5982015-05-08 16:15:03 +02004375
4376 explicit_options = qdict_clone_shallow(options);
4377
Kevin Wolf28518102015-05-08 17:07:31 +02004378 /* Inherit from parent node */
4379 if (parent_options) {
Alberto Garcia9aa09dd2018-11-12 16:00:45 +02004380 flags = 0;
Max Reitz3cdc69d2020-05-13 13:05:18 +02004381 klass->inherit_options(role, parent_is_format, &flags, options,
Max Reitz272c02e2020-05-13 13:05:17 +02004382 parent_flags, parent_options);
Alberto Garcia9aa09dd2018-11-12 16:00:45 +02004383 } else {
4384 flags = bdrv_get_flags(bs);
Kevin Wolf28518102015-05-08 17:07:31 +02004385 }
4386
Alberto Garcia077e8e22019-03-12 18:48:44 +02004387 if (keep_old_opts) {
4388 /* Old values are used for options that aren't set yet */
4389 old_options = qdict_clone_shallow(bs->options);
4390 bdrv_join_options(bs, options, old_options);
4391 qobject_unref(old_options);
4392 }
Kevin Wolf4d2cb092015-04-10 17:50:50 +02004393
Alberto Garcia9aa09dd2018-11-12 16:00:45 +02004394 /* We have the final set of options so let's update the flags */
4395 options_copy = qdict_clone_shallow(options);
4396 opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
4397 qemu_opts_absorb_qdict(opts, options_copy, NULL);
4398 update_flags_from_options(&flags, opts);
4399 qemu_opts_del(opts);
4400 qobject_unref(options_copy);
4401
Kevin Wolffd452022017-08-03 17:02:59 +02004402 /* bdrv_open_inherit() sets and clears some additional flags internally */
Kevin Wolff1f25a22014-04-25 19:04:55 +02004403 flags &= ~BDRV_O_PROTOCOL;
Kevin Wolffd452022017-08-03 17:02:59 +02004404 if (flags & BDRV_O_RDWR) {
4405 flags |= BDRV_O_ALLOW_RDWR;
4406 }
Kevin Wolff1f25a22014-04-25 19:04:55 +02004407
Kevin Wolf1857c972017-09-14 14:53:46 +02004408 if (!bs_entry) {
4409 bs_entry = g_new0(BlockReopenQueueEntry, 1);
Vladimir Sementsov-Ogievskiy859443b2019-09-27 15:23:47 +03004410 QTAILQ_INSERT_TAIL(bs_queue, bs_entry, entry);
Kevin Wolf1857c972017-09-14 14:53:46 +02004411 } else {
Marc-André Lureaucb3e7f02018-04-19 17:01:43 +02004412 qobject_unref(bs_entry->state.options);
4413 qobject_unref(bs_entry->state.explicit_options);
Kevin Wolf1857c972017-09-14 14:53:46 +02004414 }
4415
4416 bs_entry->state.bs = bs;
4417 bs_entry->state.options = options;
4418 bs_entry->state.explicit_options = explicit_options;
4419 bs_entry->state.flags = flags;
4420
Alberto Garcia85466322019-03-12 18:48:45 +02004421 /*
4422 * If keep_old_opts is false then it means that unspecified
4423 * options must be reset to their original value. We don't allow
4424 * resetting 'backing' but we need to know if the option is
4425 * missing in order to decide if we have to return an error.
4426 */
4427 if (!keep_old_opts) {
4428 bs_entry->state.backing_missing =
4429 !qdict_haskey(options, "backing") &&
4430 !qdict_haskey(options, "backing.driver");
4431 }
4432
Kevin Wolf67251a32015-04-09 18:54:04 +02004433 QLIST_FOREACH(child, &bs->children, next) {
Alberto Garcia85466322019-03-12 18:48:45 +02004434 QDict *new_child_options = NULL;
4435 bool child_keep_old = keep_old_opts;
Kevin Wolf67251a32015-04-09 18:54:04 +02004436
Kevin Wolf4c9dfe52015-05-08 15:14:15 +02004437 /* reopen can only change the options of block devices that were
4438 * implicitly created and inherited options. For other (referenced)
4439 * block devices, a syntax like "backing.foo" results in an error. */
Kevin Wolf67251a32015-04-09 18:54:04 +02004440 if (child->bs->inherits_from != bs) {
4441 continue;
4442 }
4443
Alberto Garcia85466322019-03-12 18:48:45 +02004444 /* Check if the options contain a child reference */
4445 if (qdict_haskey(options, child->name)) {
4446 const char *childref = qdict_get_try_str(options, child->name);
4447 /*
4448 * The current child must not be reopened if the child
4449 * reference is null or points to a different node.
4450 */
4451 if (g_strcmp0(childref, child->bs->node_name)) {
4452 continue;
4453 }
4454 /*
4455 * If the child reference points to the current child then
4456 * reopen it with its existing set of options (note that
4457 * it can still inherit new options from the parent).
4458 */
4459 child_keep_old = true;
4460 } else {
4461 /* Extract child options ("child-name.*") */
4462 char *child_key_dot = g_strdup_printf("%s.", child->name);
4463 qdict_extract_subqdict(explicit_options, NULL, child_key_dot);
4464 qdict_extract_subqdict(options, &new_child_options, child_key_dot);
4465 g_free(child_key_dot);
4466 }
Kevin Wolf4c9dfe52015-05-08 15:14:15 +02004467
Alberto Garcia9aa09dd2018-11-12 16:00:45 +02004468 bdrv_reopen_queue_child(bs_queue, child->bs, new_child_options,
Max Reitz3cdc69d2020-05-13 13:05:18 +02004469 child->klass, child->role, bs->drv->is_format,
4470 options, flags, child_keep_old);
Jeff Codye971aa12012-09-20 15:13:19 -04004471 }
4472
Jeff Codye971aa12012-09-20 15:13:19 -04004473 return bs_queue;
4474}
4475
Kevin Wolf2e117862022-11-18 18:41:01 +01004476/* To be called with bs->aio_context locked */
Kevin Wolf28518102015-05-08 17:07:31 +02004477BlockReopenQueue *bdrv_reopen_queue(BlockReopenQueue *bs_queue,
4478 BlockDriverState *bs,
Alberto Garcia077e8e22019-03-12 18:48:44 +02004479 QDict *options, bool keep_old_opts)
Kevin Wolf28518102015-05-08 17:07:31 +02004480{
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05004481 GLOBAL_STATE_CODE();
Kevin Wolfce433d22023-09-29 16:51:43 +02004482 GRAPH_RDLOCK_GUARD_MAINLOOP();
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05004483
Max Reitz3cdc69d2020-05-13 13:05:18 +02004484 return bdrv_reopen_queue_child(bs_queue, bs, options, NULL, 0, false,
4485 NULL, 0, keep_old_opts);
Kevin Wolf28518102015-05-08 17:07:31 +02004486}
4487
Alberto Garciaab5b52282021-07-08 13:47:05 +02004488void bdrv_reopen_queue_free(BlockReopenQueue *bs_queue)
4489{
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05004490 GLOBAL_STATE_CODE();
Alberto Garciaab5b52282021-07-08 13:47:05 +02004491 if (bs_queue) {
4492 BlockReopenQueueEntry *bs_entry, *next;
4493 QTAILQ_FOREACH_SAFE(bs_entry, bs_queue, entry, next) {
Kevin Wolfd22933a2022-11-18 18:41:02 +01004494 bdrv_drained_end(bs_entry->state.bs);
Alberto Garciaab5b52282021-07-08 13:47:05 +02004495 qobject_unref(bs_entry->state.explicit_options);
4496 qobject_unref(bs_entry->state.options);
4497 g_free(bs_entry);
4498 }
4499 g_free(bs_queue);
4500 }
4501}
4502
Jeff Codye971aa12012-09-20 15:13:19 -04004503/*
4504 * Reopen multiple BlockDriverStates atomically & transactionally.
4505 *
4506 * The queue passed in (bs_queue) must have been built up previous
4507 * via bdrv_reopen_queue().
4508 *
4509 * Reopens all BDS specified in the queue, with the appropriate
4510 * flags. All devices are prepared for reopen, and failure of any
Stefan Weil50d6a8a2018-07-12 21:51:20 +02004511 * device will cause all device changes to be abandoned, and intermediate
Jeff Codye971aa12012-09-20 15:13:19 -04004512 * data cleaned up.
4513 *
4514 * If all devices prepare successfully, then the changes are committed
4515 * to all devices.
4516 *
Kevin Wolf1a63a902017-12-06 20:24:44 +01004517 * All affected nodes must be drained between bdrv_reopen_queue() and
4518 * bdrv_reopen_multiple().
Kevin Wolf6cf42ca2021-07-08 13:47:06 +02004519 *
4520 * To be called from the main thread, with all other AioContexts unlocked.
Jeff Codye971aa12012-09-20 15:13:19 -04004521 */
Alberto Garcia5019aec2019-03-12 18:48:50 +02004522int bdrv_reopen_multiple(BlockReopenQueue *bs_queue, Error **errp)
Jeff Codye971aa12012-09-20 15:13:19 -04004523{
4524 int ret = -1;
4525 BlockReopenQueueEntry *bs_entry, *next;
Vladimir Sementsov-Ogievskiy72373e42021-04-28 18:17:58 +03004526 Transaction *tran = tran_new();
Vladimir Sementsov-Ogievskiy72373e42021-04-28 18:17:58 +03004527 g_autoptr(GSList) refresh_list = NULL;
Jeff Codye971aa12012-09-20 15:13:19 -04004528
Kevin Wolf6cf42ca2021-07-08 13:47:06 +02004529 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
Jeff Codye971aa12012-09-20 15:13:19 -04004530 assert(bs_queue != NULL);
Emanuele Giuseppe Espositoda359902022-03-03 10:16:11 -05004531 GLOBAL_STATE_CODE();
Jeff Codye971aa12012-09-20 15:13:19 -04004532
Vladimir Sementsov-Ogievskiy859443b2019-09-27 15:23:47 +03004533 QTAILQ_FOREACH(bs_entry, bs_queue, entry) {
Vladimir Sementsov-Ogievskiya2aabf82021-04-28 18:17:57 +03004534 ret = bdrv_flush(bs_entry->state.bs);
4535 if (ret < 0) {
4536 error_setg_errno(errp, -ret, "Error flushing drive");
Kevin Wolfe3fc91a2021-05-03 13:05:55 +02004537 goto abort;
Vladimir Sementsov-Ogievskiya2aabf82021-04-28 18:17:57 +03004538 }
4539 }
4540
4541 QTAILQ_FOREACH(bs_entry, bs_queue, entry) {
Kevin Wolf1a63a902017-12-06 20:24:44 +01004542 assert(bs_entry->state.bs->quiesce_counter > 0);
Vladimir Sementsov-Ogievskiy72373e42021-04-28 18:17:58 +03004543 ret = bdrv_reopen_prepare(&bs_entry->state, bs_queue, tran, errp);
4544 if (ret < 0) {
4545 goto abort;
Jeff Codye971aa12012-09-20 15:13:19 -04004546 }
4547 bs_entry->prepared = true;
4548 }
4549
Vladimir Sementsov-Ogievskiy859443b2019-09-27 15:23:47 +03004550 QTAILQ_FOREACH(bs_entry, bs_queue, entry) {
Kevin Wolf69b736e2019-03-05 17:18:22 +01004551 BDRVReopenState *state = &bs_entry->state;
Vladimir Sementsov-Ogievskiy72373e42021-04-28 18:17:58 +03004552
Vladimir Sementsov-Ogievskiyfb0ff4d2022-11-07 19:35:58 +03004553 refresh_list = g_slist_prepend(refresh_list, state->bs);
Vladimir Sementsov-Ogievskiy72373e42021-04-28 18:17:58 +03004554 if (state->old_backing_bs) {
Vladimir Sementsov-Ogievskiyfb0ff4d2022-11-07 19:35:58 +03004555 refresh_list = g_slist_prepend(refresh_list, state->old_backing_bs);
Kevin Wolf69b736e2019-03-05 17:18:22 +01004556 }
Alberto Garciaecd30d22021-06-10 15:05:36 +03004557 if (state->old_file_bs) {
Vladimir Sementsov-Ogievskiyfb0ff4d2022-11-07 19:35:58 +03004558 refresh_list = g_slist_prepend(refresh_list, state->old_file_bs);
Alberto Garciaecd30d22021-06-10 15:05:36 +03004559 }
Vladimir Sementsov-Ogievskiy72373e42021-04-28 18:17:58 +03004560 }
4561
4562 /*
4563 * Note that file-posix driver rely on permission update done during reopen
4564 * (even if no permission changed), because it wants "new" permissions for
4565 * reconfiguring the fd and that's why it does it in raw_check_perm(), not
4566 * in raw_reopen_prepare() which is called with "old" permissions.
4567 */
Kevin Wolf3804e3c2023-09-11 11:46:12 +02004568 bdrv_graph_rdlock_main_loop();
Vladimir Sementsov-Ogievskiy72373e42021-04-28 18:17:58 +03004569 ret = bdrv_list_refresh_perms(refresh_list, bs_queue, tran, errp);
Kevin Wolf3804e3c2023-09-11 11:46:12 +02004570 bdrv_graph_rdunlock_main_loop();
4571
Vladimir Sementsov-Ogievskiy72373e42021-04-28 18:17:58 +03004572 if (ret < 0) {
4573 goto abort;
Kevin Wolf69b736e2019-03-05 17:18:22 +01004574 }
4575
Vladimir Sementsov-Ogievskiyfcd6a4f2019-09-27 15:23:48 +03004576 /*
4577 * If we reach this point, we have success and just need to apply the
4578 * changes.
4579 *
4580 * Reverse order is used to comfort qcow2 driver: on commit it need to write
4581 * IN_USE flag to the image, to mark bitmaps in the image as invalid. But
4582 * children are usually goes after parents in reopen-queue, so go from last
4583 * to first element.
Jeff Codye971aa12012-09-20 15:13:19 -04004584 */
Vladimir Sementsov-Ogievskiyfcd6a4f2019-09-27 15:23:48 +03004585 QTAILQ_FOREACH_REVERSE(bs_entry, bs_queue, entry) {
Jeff Codye971aa12012-09-20 15:13:19 -04004586 bdrv_reopen_commit(&bs_entry->state);
4587 }
4588
Stefan Hajnoczi6bc30f12023-12-05 13:20:02 -05004589 bdrv_graph_wrlock();
Vladimir Sementsov-Ogievskiy72373e42021-04-28 18:17:58 +03004590 tran_commit(tran);
Stefan Hajnoczi6bc30f12023-12-05 13:20:02 -05004591 bdrv_graph_wrunlock();
Vladimir Sementsov-Ogievskiy72373e42021-04-28 18:17:58 +03004592
4593 QTAILQ_FOREACH_REVERSE(bs_entry, bs_queue, entry) {
4594 BlockDriverState *bs = bs_entry->state.bs;
4595
4596 if (bs->drv->bdrv_reopen_commit_post) {
4597 bs->drv->bdrv_reopen_commit_post(&bs_entry->state);
4598 }
4599 }
4600
Jeff Codye971aa12012-09-20 15:13:19 -04004601 ret = 0;
Vladimir Sementsov-Ogievskiy72373e42021-04-28 18:17:58 +03004602 goto cleanup;
4603
4604abort:
Stefan Hajnoczi6bc30f12023-12-05 13:20:02 -05004605 bdrv_graph_wrlock();
Vladimir Sementsov-Ogievskiy72373e42021-04-28 18:17:58 +03004606 tran_abort(tran);
Stefan Hajnoczi6bc30f12023-12-05 13:20:02 -05004607 bdrv_graph_wrunlock();
Kevin Wolf7d4ca9d2023-09-11 11:46:09 +02004608
Vladimir Sementsov-Ogievskiy859443b2019-09-27 15:23:47 +03004609 QTAILQ_FOREACH_SAFE(bs_entry, bs_queue, entry, next) {
Vladimir Sementsov-Ogievskiy72373e42021-04-28 18:17:58 +03004610 if (bs_entry->prepared) {
4611 bdrv_reopen_abort(&bs_entry->state);
Kevin Wolf69b736e2019-03-05 17:18:22 +01004612 }
Kevin Wolf69b736e2019-03-05 17:18:22 +01004613 }
Peter Krempa17e1e2b2020-02-28 13:44:46 +01004614
Jeff Codye971aa12012-09-20 15:13:19 -04004615cleanup:
Alberto Garciaab5b52282021-07-08 13:47:05 +02004616 bdrv_reopen_queue_free(bs_queue);
Alberto Garcia40840e42016-10-28 10:08:03 +03004617
Jeff Codye971aa12012-09-20 15:13:19 -04004618 return ret;
4619}
4620
Kevin Wolf6cf42ca2021-07-08 13:47:06 +02004621int bdrv_reopen(BlockDriverState *bs, QDict *opts, bool keep_old_opts,
4622 Error **errp)
4623{
Kevin Wolf6cf42ca2021-07-08 13:47:06 +02004624 BlockReopenQueue *queue;
Kevin Wolf6cf42ca2021-07-08 13:47:06 +02004625
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05004626 GLOBAL_STATE_CODE();
4627
Kevin Wolf2e117862022-11-18 18:41:01 +01004628 queue = bdrv_reopen_queue(NULL, bs, opts, keep_old_opts);
4629
Stefan Hajnoczib49f4752023-12-05 13:20:03 -05004630 return bdrv_reopen_multiple(queue, errp);
Kevin Wolf6cf42ca2021-07-08 13:47:06 +02004631}
4632
Alberto Garcia6e1000a2018-11-12 16:00:33 +02004633int bdrv_reopen_set_read_only(BlockDriverState *bs, bool read_only,
4634 Error **errp)
4635{
Alberto Garcia6e1000a2018-11-12 16:00:33 +02004636 QDict *opts = qdict_new();
4637
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05004638 GLOBAL_STATE_CODE();
4639
Alberto Garcia6e1000a2018-11-12 16:00:33 +02004640 qdict_put_bool(opts, BDRV_OPT_READ_ONLY, read_only);
4641
Kevin Wolf6cf42ca2021-07-08 13:47:06 +02004642 return bdrv_reopen(bs, opts, true, errp);
Alberto Garcia6e1000a2018-11-12 16:00:33 +02004643}
4644
Jeff Codye971aa12012-09-20 15:13:19 -04004645/*
Alberto Garciacb828c32019-03-12 18:48:47 +02004646 * Take a BDRVReopenState and check if the value of 'backing' in the
4647 * reopen_state->options QDict is valid or not.
4648 *
4649 * If 'backing' is missing from the QDict then return 0.
4650 *
4651 * If 'backing' contains the node name of the backing file of
4652 * reopen_state->bs then return 0.
4653 *
4654 * If 'backing' contains a different node name (or is null) then check
4655 * whether the current backing file can be replaced with the new one.
4656 * If that's the case then reopen_state->replace_backing_bs is set to
4657 * true and reopen_state->new_backing_bs contains a pointer to the new
4658 * backing BlockDriverState (or NULL).
4659 *
Kevin Wolf5661a002023-09-11 11:46:10 +02004660 * After calling this function, the transaction @tran may only be completed
4661 * while holding a writer lock for the graph.
4662 *
Alberto Garciacb828c32019-03-12 18:48:47 +02004663 * Return 0 on success, otherwise return < 0 and set @errp.
Kevin Wolf4b408662023-06-05 10:57:06 +02004664 *
Kevin Wolf4b408662023-06-05 10:57:06 +02004665 * @reopen_state->bs can move to a different AioContext in this function.
Alberto Garciacb828c32019-03-12 18:48:47 +02004666 */
Kevin Wolfce433d22023-09-29 16:51:43 +02004667static int GRAPH_UNLOCKED
4668bdrv_reopen_parse_file_or_backing(BDRVReopenState *reopen_state,
4669 bool is_backing, Transaction *tran,
4670 Error **errp)
Alberto Garciacb828c32019-03-12 18:48:47 +02004671{
4672 BlockDriverState *bs = reopen_state->bs;
Alberto Garciaecd30d22021-06-10 15:05:36 +03004673 BlockDriverState *new_child_bs;
Kevin Wolf004915a2023-10-27 17:53:26 +02004674 BlockDriverState *old_child_bs;
4675
Alberto Garciaecd30d22021-06-10 15:05:36 +03004676 const char *child_name = is_backing ? "backing" : "file";
Alberto Garciacb828c32019-03-12 18:48:47 +02004677 QObject *value;
4678 const char *str;
Kevin Wolfce433d22023-09-29 16:51:43 +02004679 bool has_child;
Kevin Wolf4b408662023-06-05 10:57:06 +02004680 int ret;
Alberto Garciacb828c32019-03-12 18:48:47 +02004681
Emanuele Giuseppe Espositobdb73472022-03-03 10:16:02 -05004682 GLOBAL_STATE_CODE();
4683
Alberto Garciaecd30d22021-06-10 15:05:36 +03004684 value = qdict_get(reopen_state->options, child_name);
Alberto Garciacb828c32019-03-12 18:48:47 +02004685 if (value == NULL) {
4686 return 0;
4687 }
4688
Kevin Wolf430da832023-10-27 17:53:16 +02004689 bdrv_graph_rdlock_main_loop();
4690
Alberto Garciacb828c32019-03-12 18:48:47 +02004691 switch (qobject_type(value)) {
4692 case QTYPE_QNULL:
Alberto Garciaecd30d22021-06-10 15:05:36 +03004693 assert(is_backing); /* The 'file' option does not allow a null value */
4694 new_child_bs = NULL;
Alberto Garciacb828c32019-03-12 18:48:47 +02004695 break;
4696 case QTYPE_QSTRING:
Markus Armbruster410f44f2020-12-11 18:11:42 +01004697 str = qstring_get_str(qobject_to(QString, value));
Alberto Garciaecd30d22021-06-10 15:05:36 +03004698 new_child_bs = bdrv_lookup_bs(NULL, str, errp);
4699 if (new_child_bs == NULL) {
Kevin Wolf430da832023-10-27 17:53:16 +02004700 ret = -EINVAL;
4701 goto out_rdlock;
Kevin Wolfce433d22023-09-29 16:51:43 +02004702 }
4703
Kevin Wolfce433d22023-09-29 16:51:43 +02004704 has_child = bdrv_recurse_has_child(new_child_bs, bs);
Kevin Wolfce433d22023-09-29 16:51:43 +02004705 if (has_child) {
Alberto Garciaecd30d22021-06-10 15:05:36 +03004706 error_setg(errp, "Making '%s' a %s child of '%s' would create a "
4707 "cycle", str, child_name, bs->node_name);
Kevin Wolf430da832023-10-27 17:53:16 +02004708 ret = -EINVAL;
4709 goto out_rdlock;
Alberto Garciacb828c32019-03-12 18:48:47 +02004710 }
4711 break;
4712 default:
Alberto Garciaecd30d22021-06-10 15:05:36 +03004713 /*
4714 * The options QDict has been flattened, so 'backing' and 'file'
4715 * do not allow any other data type here.
4716 */
Alberto Garciacb828c32019-03-12 18:48:47 +02004717 g_assert_not_reached();
4718 }
4719
Kevin Wolf004915a2023-10-27 17:53:26 +02004720 old_child_bs = is_backing ? child_bs(bs->backing) : child_bs(bs->file);
Alberto Garciaecd30d22021-06-10 15:05:36 +03004721 if (old_child_bs == new_child_bs) {
Kevin Wolf430da832023-10-27 17:53:16 +02004722 ret = 0;
4723 goto out_rdlock;
Alberto Garciaecd30d22021-06-10 15:05:36 +03004724 }
4725
4726 if (old_child_bs) {
4727 if (bdrv_skip_implicit_filters(old_child_bs) == new_child_bs) {
Kevin Wolf430da832023-10-27 17:53:16 +02004728 ret = 0;
4729 goto out_rdlock;
Vladimir Sementsov-Ogievskiycbfdb982021-06-10 15:05:33 +03004730 }
4731
Alberto Garciaecd30d22021-06-10 15:05:36 +03004732 if (old_child_bs->implicit) {
4733 error_setg(errp, "Cannot replace implicit %s child of %s",
4734 child_name, bs->node_name);
Kevin Wolf430da832023-10-27 17:53:16 +02004735 ret = -EPERM;
4736 goto out_rdlock;
Vladimir Sementsov-Ogievskiycbfdb982021-06-10 15:05:33 +03004737 }
4738 }
4739
Alberto Garciaecd30d22021-06-10 15:05:36 +03004740 if (bs->drv->is_filter && !old_child_bs) {
Vladimir Sementsov-Ogievskiy25f78d92021-06-10 15:05:34 +03004741 /*
4742 * Filters always have a file or a backing child, so we are trying to
4743 * change wrong child
4744 */
4745 error_setg(errp, "'%s' is a %s filter node that does not support a "
Alberto Garciaecd30d22021-06-10 15:05:36 +03004746 "%s child", bs->node_name, bs->drv->format_name, child_name);
Kevin Wolf430da832023-10-27 17:53:16 +02004747 ret = -EINVAL;
4748 goto out_rdlock;
Max Reitz1d42f482019-06-12 17:24:39 +02004749 }
4750
Alberto Garciaecd30d22021-06-10 15:05:36 +03004751 if (is_backing) {
4752 reopen_state->old_backing_bs = old_child_bs;
4753 } else {
4754 reopen_state->old_file_bs = old_child_bs;
4755 }
4756
Kevin Wolf7d4ca9d2023-09-11 11:46:09 +02004757 if (old_child_bs) {
4758 bdrv_ref(old_child_bs);
4759 bdrv_drained_begin(old_child_bs);
4760 }
4761
Kevin Wolf430da832023-10-27 17:53:16 +02004762 bdrv_graph_rdunlock_main_loop();
Stefan Hajnoczi6bc30f12023-12-05 13:20:02 -05004763 bdrv_graph_wrlock();
Kevin Wolf7d4ca9d2023-09-11 11:46:09 +02004764
Kevin Wolf4b408662023-06-05 10:57:06 +02004765 ret = bdrv_set_file_or_backing_noperm(bs, new_child_bs, is_backing,
4766 tran, errp);
4767
Stefan Hajnoczi6bc30f12023-12-05 13:20:02 -05004768 bdrv_graph_wrunlock();
Kevin Wolf7d4ca9d2023-09-11 11:46:09 +02004769
Kevin Wolf7d4ca9d2023-09-11 11:46:09 +02004770 if (old_child_bs) {
4771 bdrv_drained_end(old_child_bs);
4772 bdrv_unref(old_child_bs);
4773 }
4774
Kevin Wolf4b408662023-06-05 10:57:06 +02004775 return ret;
Kevin Wolf430da832023-10-27 17:53:16 +02004776
4777out_rdlock:
4778 bdrv_graph_rdunlock_main_loop();
4779 return ret;
Alberto Garciacb828c32019-03-12 18:48:47 +02004780}
4781
4782/*
Jeff Codye971aa12012-09-20 15:13:19 -04004783 * Prepares a BlockDriverState for reopen. All changes are staged in the
4784 * 'opaque' field of the BDRVReopenState, which is used and allocated by
4785 * the block driver layer .bdrv_reopen_prepare()
4786 *
4787 * bs is the BlockDriverState to reopen
4788 * flags are the new open flags
4789 * queue is the reopen queue
4790 *
4791 * Returns 0 on success, non-zero on error. On error errp will be set
4792 * as well.
4793 *
4794 * On failure, bdrv_reopen_abort() will be called to clean up any data.
4795 * It is the responsibility of the caller to then call the abort() or
4796 * commit() for any other BDS that have been left in a prepare() state
4797 *
Kevin Wolf5661a002023-09-11 11:46:10 +02004798 * After calling this function, the transaction @change_child_tran may only be
4799 * completed while holding a writer lock for the graph.
Jeff Codye971aa12012-09-20 15:13:19 -04004800 */
Kevin Wolfce433d22023-09-29 16:51:43 +02004801static int GRAPH_UNLOCKED
4802bdrv_reopen_prepare(BDRVReopenState *reopen_state, BlockReopenQueue *queue,
4803 Transaction *change_child_tran, Error **errp)
Jeff Codye971aa12012-09-20 15:13:19 -04004804{
4805 int ret = -1;
Alberto Garciae6d79c42018-11-12 16:00:47 +02004806 int old_flags;
Jeff Codye971aa12012-09-20 15:13:19 -04004807 Error *local_err = NULL;
4808 BlockDriver *drv;
Kevin Wolfccf9dc02015-05-08 17:24:56 +02004809 QemuOpts *opts;
Alberto Garcia4c8350f2018-06-29 14:37:02 +03004810 QDict *orig_reopen_opts;
Alberto Garcia593b3072018-09-06 12:37:08 +03004811 char *discard = NULL;
Jeff Cody3d8ce172017-04-07 16:55:30 -04004812 bool read_only;
Max Reitz9ad08c42018-11-16 17:45:24 +01004813 bool drv_prepared = false;
Jeff Codye971aa12012-09-20 15:13:19 -04004814
4815 assert(reopen_state != NULL);
4816 assert(reopen_state->bs->drv != NULL);
Emanuele Giuseppe Espositoda359902022-03-03 10:16:11 -05004817 GLOBAL_STATE_CODE();
Jeff Codye971aa12012-09-20 15:13:19 -04004818 drv = reopen_state->bs->drv;
4819
Alberto Garcia4c8350f2018-06-29 14:37:02 +03004820 /* This function and each driver's bdrv_reopen_prepare() remove
4821 * entries from reopen_state->options as they are processed, so
4822 * we need to make a copy of the original QDict. */
4823 orig_reopen_opts = qdict_clone_shallow(reopen_state->options);
4824
Kevin Wolfccf9dc02015-05-08 17:24:56 +02004825 /* Process generic block layer options */
4826 opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
Markus Armbrusteraf175e82020-07-07 18:06:03 +02004827 if (!qemu_opts_absorb_qdict(opts, reopen_state->options, errp)) {
Kevin Wolfccf9dc02015-05-08 17:24:56 +02004828 ret = -EINVAL;
4829 goto error;
4830 }
4831
Alberto Garciae6d79c42018-11-12 16:00:47 +02004832 /* This was already called in bdrv_reopen_queue_child() so the flags
4833 * are up-to-date. This time we simply want to remove the options from
4834 * QemuOpts in order to indicate that they have been processed. */
4835 old_flags = reopen_state->flags;
Kevin Wolf91a097e2015-05-08 17:49:53 +02004836 update_flags_from_options(&reopen_state->flags, opts);
Alberto Garciae6d79c42018-11-12 16:00:47 +02004837 assert(old_flags == reopen_state->flags);
Kevin Wolf91a097e2015-05-08 17:49:53 +02004838
Alberto Garcia415bbca2018-10-03 13:23:13 +03004839 discard = qemu_opt_get_del(opts, BDRV_OPT_DISCARD);
Alberto Garcia593b3072018-09-06 12:37:08 +03004840 if (discard != NULL) {
4841 if (bdrv_parse_discard_flags(discard, &reopen_state->flags) != 0) {
4842 error_setg(errp, "Invalid discard option");
4843 ret = -EINVAL;
4844 goto error;
4845 }
4846 }
4847
Alberto Garcia543770b2018-09-06 12:37:09 +03004848 reopen_state->detect_zeroes =
4849 bdrv_parse_detect_zeroes(opts, reopen_state->flags, &local_err);
4850 if (local_err) {
4851 error_propagate(errp, local_err);
4852 ret = -EINVAL;
4853 goto error;
4854 }
4855
Alberto Garcia57f9db92018-09-06 12:37:06 +03004856 /* All other options (including node-name and driver) must be unchanged.
4857 * Put them back into the QDict, so that they are checked at the end
4858 * of this function. */
4859 qemu_opts_to_qdict(opts, reopen_state->options);
Kevin Wolfccf9dc02015-05-08 17:24:56 +02004860
Jeff Cody3d8ce172017-04-07 16:55:30 -04004861 /* If we are to stay read-only, do not allow permission change
4862 * to r/w. Attempting to set to r/w may fail if either BDRV_O_ALLOW_RDWR is
4863 * not set, or if the BDS still has copy_on_read enabled */
4864 read_only = !(reopen_state->flags & BDRV_O_RDWR);
Kevin Wolf4026f1c2023-09-29 16:51:47 +02004865
4866 bdrv_graph_rdlock_main_loop();
Kevin Wolf54a32bf2017-08-03 17:02:58 +02004867 ret = bdrv_can_set_read_only(reopen_state->bs, read_only, true, &local_err);
Kevin Wolf4026f1c2023-09-29 16:51:47 +02004868 bdrv_graph_rdunlock_main_loop();
Jeff Cody3d8ce172017-04-07 16:55:30 -04004869 if (local_err) {
4870 error_propagate(errp, local_err);
Jeff Codye971aa12012-09-20 15:13:19 -04004871 goto error;
4872 }
4873
Jeff Codye971aa12012-09-20 15:13:19 -04004874 if (drv->bdrv_reopen_prepare) {
Alberto Garciafaf116b2019-03-12 18:48:49 +02004875 /*
4876 * If a driver-specific option is missing, it means that we
4877 * should reset it to its default value.
4878 * But not all options allow that, so we need to check it first.
4879 */
4880 ret = bdrv_reset_options_allowed(reopen_state->bs,
4881 reopen_state->options, errp);
4882 if (ret) {
4883 goto error;
4884 }
4885
Jeff Codye971aa12012-09-20 15:13:19 -04004886 ret = drv->bdrv_reopen_prepare(reopen_state, queue, &local_err);
4887 if (ret) {
4888 if (local_err != NULL) {
4889 error_propagate(errp, local_err);
4890 } else {
Kevin Wolfb7cfc7d2023-09-29 16:51:45 +02004891 bdrv_graph_rdlock_main_loop();
Max Reitzf30c66b2019-02-01 20:29:05 +01004892 bdrv_refresh_filename(reopen_state->bs);
Kevin Wolfb7cfc7d2023-09-29 16:51:45 +02004893 bdrv_graph_rdunlock_main_loop();
Luiz Capitulinod8b68952013-06-10 11:29:27 -04004894 error_setg(errp, "failed while preparing to reopen image '%s'",
4895 reopen_state->bs->filename);
Jeff Codye971aa12012-09-20 15:13:19 -04004896 }
4897 goto error;
4898 }
4899 } else {
4900 /* It is currently mandatory to have a bdrv_reopen_prepare()
4901 * handler for each supported drv. */
Kevin Wolf4026f1c2023-09-29 16:51:47 +02004902 bdrv_graph_rdlock_main_loop();
Alberto Garcia81e5f782015-04-08 12:29:19 +03004903 error_setg(errp, "Block format '%s' used by node '%s' "
4904 "does not support reopening files", drv->format_name,
4905 bdrv_get_device_or_node_name(reopen_state->bs));
Kevin Wolf4026f1c2023-09-29 16:51:47 +02004906 bdrv_graph_rdunlock_main_loop();
Jeff Codye971aa12012-09-20 15:13:19 -04004907 ret = -1;
4908 goto error;
4909 }
4910
Max Reitz9ad08c42018-11-16 17:45:24 +01004911 drv_prepared = true;
4912
Alberto Garciabacd9b82019-03-12 18:48:46 +02004913 /*
4914 * We must provide the 'backing' option if the BDS has a backing
4915 * file or if the image file has a backing file name as part of
4916 * its metadata. Otherwise the 'backing' option can be omitted.
4917 */
Kevin Wolf004915a2023-10-27 17:53:26 +02004918 bdrv_graph_rdlock_main_loop();
Alberto Garciabacd9b82019-03-12 18:48:46 +02004919 if (drv->supports_backing && reopen_state->backing_missing &&
Max Reitz1d42f482019-06-12 17:24:39 +02004920 (reopen_state->bs->backing || reopen_state->bs->backing_file[0])) {
Alberto Garcia85466322019-03-12 18:48:45 +02004921 error_setg(errp, "backing is missing for '%s'",
4922 reopen_state->bs->node_name);
Kevin Wolf004915a2023-10-27 17:53:26 +02004923 bdrv_graph_rdunlock_main_loop();
Alberto Garcia85466322019-03-12 18:48:45 +02004924 ret = -EINVAL;
4925 goto error;
4926 }
Kevin Wolf004915a2023-10-27 17:53:26 +02004927 bdrv_graph_rdunlock_main_loop();
Alberto Garcia85466322019-03-12 18:48:45 +02004928
Alberto Garciacb828c32019-03-12 18:48:47 +02004929 /*
4930 * Allow changing the 'backing' option. The new value can be
4931 * either a reference to an existing node (using its node name)
4932 * or NULL to simply detach the current backing file.
4933 */
Alberto Garciaecd30d22021-06-10 15:05:36 +03004934 ret = bdrv_reopen_parse_file_or_backing(reopen_state, true,
4935 change_child_tran, errp);
Alberto Garciacb828c32019-03-12 18:48:47 +02004936 if (ret < 0) {
4937 goto error;
4938 }
4939 qdict_del(reopen_state->options, "backing");
4940
Alberto Garciaecd30d22021-06-10 15:05:36 +03004941 /* Allow changing the 'file' option. In this case NULL is not allowed */
4942 ret = bdrv_reopen_parse_file_or_backing(reopen_state, false,
4943 change_child_tran, errp);
4944 if (ret < 0) {
4945 goto error;
4946 }
4947 qdict_del(reopen_state->options, "file");
4948
Kevin Wolf4d2cb092015-04-10 17:50:50 +02004949 /* Options that are not handled are only okay if they are unchanged
4950 * compared to the old state. It is expected that some options are only
4951 * used for the initial open, but not reopen (e.g. filename) */
4952 if (qdict_size(reopen_state->options)) {
4953 const QDictEntry *entry = qdict_first(reopen_state->options);
4954
Kevin Wolfce433d22023-09-29 16:51:43 +02004955 GRAPH_RDLOCK_GUARD_MAINLOOP();
4956
Kevin Wolf4d2cb092015-04-10 17:50:50 +02004957 do {
Max Reitz54fd1b02017-11-14 19:01:26 +01004958 QObject *new = entry->value;
4959 QObject *old = qdict_get(reopen_state->bs->options, entry->key);
Kevin Wolf4d2cb092015-04-10 17:50:50 +02004960
Alberto Garciadb905282018-09-06 12:37:05 +03004961 /* Allow child references (child_name=node_name) as long as they
4962 * point to the current child (i.e. everything stays the same). */
4963 if (qobject_type(new) == QTYPE_QSTRING) {
4964 BdrvChild *child;
4965 QLIST_FOREACH(child, &reopen_state->bs->children, next) {
4966 if (!strcmp(child->name, entry->key)) {
4967 break;
4968 }
4969 }
4970
4971 if (child) {
Markus Armbruster410f44f2020-12-11 18:11:42 +01004972 if (!strcmp(child->bs->node_name,
4973 qstring_get_str(qobject_to(QString, new)))) {
Alberto Garciadb905282018-09-06 12:37:05 +03004974 continue; /* Found child with this name, skip option */
4975 }
4976 }
4977 }
4978
Max Reitz54fd1b02017-11-14 19:01:26 +01004979 /*
4980 * TODO: When using -drive to specify blockdev options, all values
4981 * will be strings; however, when using -blockdev, blockdev-add or
4982 * filenames using the json:{} pseudo-protocol, they will be
4983 * correctly typed.
4984 * In contrast, reopening options are (currently) always strings
4985 * (because you can only specify them through qemu-io; all other
4986 * callers do not specify any options).
4987 * Therefore, when using anything other than -drive to create a BDS,
4988 * this cannot detect non-string options as unchanged, because
4989 * qobject_is_equal() always returns false for objects of different
4990 * type. In the future, this should be remedied by correctly typing
4991 * all options. For now, this is not too big of an issue because
4992 * the user can simply omit options which cannot be changed anyway,
4993 * so they will stay unchanged.
4994 */
4995 if (!qobject_is_equal(new, old)) {
Kevin Wolf4d2cb092015-04-10 17:50:50 +02004996 error_setg(errp, "Cannot change the option '%s'", entry->key);
4997 ret = -EINVAL;
4998 goto error;
4999 }
5000 } while ((entry = qdict_next(reopen_state->options, entry)));
5001 }
5002
Jeff Codye971aa12012-09-20 15:13:19 -04005003 ret = 0;
5004
Alberto Garcia4c8350f2018-06-29 14:37:02 +03005005 /* Restore the original reopen_state->options QDict */
5006 qobject_unref(reopen_state->options);
5007 reopen_state->options = qobject_ref(orig_reopen_opts);
5008
Jeff Codye971aa12012-09-20 15:13:19 -04005009error:
Max Reitz9ad08c42018-11-16 17:45:24 +01005010 if (ret < 0 && drv_prepared) {
5011 /* drv->bdrv_reopen_prepare() has succeeded, so we need to
5012 * call drv->bdrv_reopen_abort() before signaling an error
5013 * (bdrv_reopen_multiple() will not call bdrv_reopen_abort()
5014 * when the respective bdrv_reopen_prepare() has failed) */
5015 if (drv->bdrv_reopen_abort) {
5016 drv->bdrv_reopen_abort(reopen_state);
5017 }
5018 }
Kevin Wolfccf9dc02015-05-08 17:24:56 +02005019 qemu_opts_del(opts);
Alberto Garcia4c8350f2018-06-29 14:37:02 +03005020 qobject_unref(orig_reopen_opts);
Alberto Garcia593b3072018-09-06 12:37:08 +03005021 g_free(discard);
Jeff Codye971aa12012-09-20 15:13:19 -04005022 return ret;
5023}
5024
5025/*
5026 * Takes the staged changes for the reopen from bdrv_reopen_prepare(), and
5027 * makes them final by swapping the staging BlockDriverState contents into
5028 * the active BlockDriverState contents.
5029 */
Kevin Wolfce433d22023-09-29 16:51:43 +02005030static void GRAPH_UNLOCKED bdrv_reopen_commit(BDRVReopenState *reopen_state)
Jeff Codye971aa12012-09-20 15:13:19 -04005031{
5032 BlockDriver *drv;
Vladimir Sementsov-Ogievskiy50bf65b2017-06-28 15:05:12 +03005033 BlockDriverState *bs;
Alberto Garcia50196d72018-09-06 12:37:03 +03005034 BdrvChild *child;
Jeff Codye971aa12012-09-20 15:13:19 -04005035
5036 assert(reopen_state != NULL);
Vladimir Sementsov-Ogievskiy50bf65b2017-06-28 15:05:12 +03005037 bs = reopen_state->bs;
5038 drv = bs->drv;
Jeff Codye971aa12012-09-20 15:13:19 -04005039 assert(drv != NULL);
Emanuele Giuseppe Espositoda359902022-03-03 10:16:11 -05005040 GLOBAL_STATE_CODE();
Jeff Codye971aa12012-09-20 15:13:19 -04005041
5042 /* If there are any driver level actions to take */
5043 if (drv->bdrv_reopen_commit) {
5044 drv->bdrv_reopen_commit(reopen_state);
5045 }
5046
Kevin Wolfce433d22023-09-29 16:51:43 +02005047 GRAPH_RDLOCK_GUARD_MAINLOOP();
5048
Jeff Codye971aa12012-09-20 15:13:19 -04005049 /* set BDS specific flags now */
Marc-André Lureaucb3e7f02018-04-19 17:01:43 +02005050 qobject_unref(bs->explicit_options);
Alberto Garcia4c8350f2018-06-29 14:37:02 +03005051 qobject_unref(bs->options);
Alberto Garciaab5b52282021-07-08 13:47:05 +02005052 qobject_ref(reopen_state->explicit_options);
5053 qobject_ref(reopen_state->options);
Kevin Wolf145f5982015-05-08 16:15:03 +02005054
Vladimir Sementsov-Ogievskiy50bf65b2017-06-28 15:05:12 +03005055 bs->explicit_options = reopen_state->explicit_options;
Alberto Garcia4c8350f2018-06-29 14:37:02 +03005056 bs->options = reopen_state->options;
Vladimir Sementsov-Ogievskiy50bf65b2017-06-28 15:05:12 +03005057 bs->open_flags = reopen_state->flags;
Alberto Garcia543770b2018-09-06 12:37:09 +03005058 bs->detect_zeroes = reopen_state->detect_zeroes;
Kevin Wolf355ef4a2013-12-11 20:14:09 +01005059
Alberto Garcia50196d72018-09-06 12:37:03 +03005060 /* Remove child references from bs->options and bs->explicit_options.
5061 * Child options were already removed in bdrv_reopen_queue_child() */
5062 QLIST_FOREACH(child, &bs->children, next) {
5063 qdict_del(bs->explicit_options, child->name);
5064 qdict_del(bs->options, child->name);
5065 }
Vladimir Sementsov-Ogievskiy3d0e8742021-06-10 15:05:35 +03005066 /* backing is probably removed, so it's not handled by previous loop */
5067 qdict_del(bs->explicit_options, "backing");
5068 qdict_del(bs->options, "backing");
5069
Vladimir Sementsov-Ogievskiy1e4c7972021-04-28 18:17:55 +03005070 bdrv_refresh_limits(bs, NULL, NULL);
Paolo Bonzini439cc332023-04-07 17:32:58 +02005071 bdrv_refresh_total_sectors(bs, bs->total_sectors);
Jeff Codye971aa12012-09-20 15:13:19 -04005072}
5073
5074/*
5075 * Abort the reopen, and delete and free the staged changes in
5076 * reopen_state
5077 */
Kevin Wolfce433d22023-09-29 16:51:43 +02005078static void GRAPH_UNLOCKED bdrv_reopen_abort(BDRVReopenState *reopen_state)
Jeff Codye971aa12012-09-20 15:13:19 -04005079{
5080 BlockDriver *drv;
5081
5082 assert(reopen_state != NULL);
5083 drv = reopen_state->bs->drv;
5084 assert(drv != NULL);
Emanuele Giuseppe Espositoda359902022-03-03 10:16:11 -05005085 GLOBAL_STATE_CODE();
Jeff Codye971aa12012-09-20 15:13:19 -04005086
5087 if (drv->bdrv_reopen_abort) {
5088 drv->bdrv_reopen_abort(reopen_state);
5089 }
5090}
5091
5092
Max Reitz64dff522016-01-29 16:36:10 +01005093static void bdrv_close(BlockDriverState *bs)
bellardfc01f7e2003-06-30 10:03:06 +00005094{
Max Reitz33384422014-06-20 21:57:33 +02005095 BdrvAioNotifier *ban, *ban_next;
Alberto Garcia50a3efb2017-11-06 16:53:45 +02005096 BdrvChild *child, *next;
Max Reitz33384422014-06-20 21:57:33 +02005097
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05005098 GLOBAL_STATE_CODE();
Max Reitz30f55fb2016-05-17 16:41:32 +02005099 assert(!bs->refcnt);
Alberto Garcia99b7e772015-09-25 16:41:44 +03005100
Paolo Bonzinifc272912015-12-23 11:48:24 +01005101 bdrv_drained_begin(bs); /* complete I/O */
Stefan Hajnoczi58fda172013-07-02 15:36:25 +02005102 bdrv_flush(bs);
Fam Zheng53ec73e2015-05-29 18:53:14 +08005103 bdrv_drain(bs); /* in case flush left pending I/O */
Paolo Bonzinifc272912015-12-23 11:48:24 +01005104
Paolo Bonzini3cbc0022012-10-19 11:36:48 +02005105 if (bs->drv) {
Vladimir Sementsov-Ogievskiy3c005292018-08-14 15:43:19 +03005106 if (bs->drv->bdrv_close) {
Max Reitz7b99a262019-06-12 16:07:11 +02005107 /* Must unfreeze all children, so bdrv_unref_child() works */
Vladimir Sementsov-Ogievskiy3c005292018-08-14 15:43:19 +03005108 bs->drv->bdrv_close(bs);
5109 }
Kevin Wolf9a4f4c32015-06-16 14:19:22 +02005110 bs->drv = NULL;
bellardb3380822004-03-14 21:38:54 +00005111 }
Zhi Yong Wu98f90db2011-11-08 13:00:14 +08005112
Stefan Hajnoczi6bc30f12023-12-05 13:20:02 -05005113 bdrv_graph_wrlock();
Alberto Garcia50a3efb2017-11-06 16:53:45 +02005114 QLIST_FOREACH_SAFE(child, &bs->children, next, next) {
Alberto Garciadd4118c2019-05-13 16:46:17 +03005115 bdrv_unref_child(bs, child);
Alberto Garcia50a3efb2017-11-06 16:53:45 +02005116 }
5117
Vladimir Sementsov-Ogievskiy5bb047472022-07-26 23:11:32 +03005118 assert(!bs->backing);
5119 assert(!bs->file);
Stefan Hajnoczi6bc30f12023-12-05 13:20:02 -05005120 bdrv_graph_wrunlock();
Kevin Wolf004915a2023-10-27 17:53:26 +02005121
Alberto Garcia50a3efb2017-11-06 16:53:45 +02005122 g_free(bs->opaque);
5123 bs->opaque = NULL;
Stefan Hajnoczid73415a2020-09-23 11:56:46 +01005124 qatomic_set(&bs->copy_on_read, 0);
Alberto Garcia50a3efb2017-11-06 16:53:45 +02005125 bs->backing_file[0] = '\0';
5126 bs->backing_format[0] = '\0';
5127 bs->total_sectors = 0;
5128 bs->encrypted = false;
5129 bs->sg = false;
Marc-André Lureaucb3e7f02018-04-19 17:01:43 +02005130 qobject_unref(bs->options);
5131 qobject_unref(bs->explicit_options);
Alberto Garcia50a3efb2017-11-06 16:53:45 +02005132 bs->options = NULL;
5133 bs->explicit_options = NULL;
Marc-André Lureaucb3e7f02018-04-19 17:01:43 +02005134 qobject_unref(bs->full_open_options);
Alberto Garcia50a3efb2017-11-06 16:53:45 +02005135 bs->full_open_options = NULL;
Hanna Reitz0bc329f2021-08-12 10:41:44 +02005136 g_free(bs->block_status_cache);
5137 bs->block_status_cache = NULL;
Alberto Garcia50a3efb2017-11-06 16:53:45 +02005138
Vladimir Sementsov-Ogievskiycca43ae2017-06-28 15:05:16 +03005139 bdrv_release_named_dirty_bitmaps(bs);
5140 assert(QLIST_EMPTY(&bs->dirty_bitmaps));
5141
Max Reitz33384422014-06-20 21:57:33 +02005142 QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_next) {
5143 g_free(ban);
5144 }
5145 QLIST_INIT(&bs->aio_notifiers);
Paolo Bonzinifc272912015-12-23 11:48:24 +01005146 bdrv_drained_end(bs);
Greg Kurz1a6d3bd2020-10-23 17:01:10 +02005147
5148 /*
5149 * If we're still inside some bdrv_drain_all_begin()/end() sections, end
5150 * them now since this BDS won't exist anymore when bdrv_drain_all_end()
5151 * gets called.
5152 */
5153 if (bs->quiesce_counter) {
5154 bdrv_drain_all_end_quiesce(bs);
5155 }
bellardb3380822004-03-14 21:38:54 +00005156}
5157
MORITA Kazutaka2bc93fe2010-05-28 11:44:57 +09005158void bdrv_close_all(void)
5159{
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05005160 GLOBAL_STATE_CODE();
Emanuele Giuseppe Esposito880eeec2022-09-26 05:32:04 -04005161 assert(job_next(NULL) == NULL);
MORITA Kazutaka2bc93fe2010-05-28 11:44:57 +09005162
Max Reitzca9bd242016-01-29 16:36:14 +01005163 /* Drop references from requests still in flight, such as canceled block
5164 * jobs whose AIO context has not been polled yet */
5165 bdrv_drain_all();
Stefan Hajnoczied78cda2014-05-08 16:34:35 +02005166
Max Reitzca9bd242016-01-29 16:36:14 +01005167 blk_remove_all_bs();
5168 blockdev_close_all_bdrv_states();
5169
Kevin Wolfa1a2af02016-04-08 18:26:37 +02005170 assert(QTAILQ_EMPTY(&all_bdrv_states));
MORITA Kazutaka2bc93fe2010-05-28 11:44:57 +09005171}
5172
Kevin Wolf2f64e1f2023-09-11 11:46:08 +02005173static bool GRAPH_RDLOCK should_update_child(BdrvChild *c, BlockDriverState *to)
Kevin Wolfdd62f1c2015-06-18 14:09:57 +02005174{
Vladimir Sementsov-Ogievskiy2f30b7c2019-02-23 22:20:39 +03005175 GQueue *queue;
5176 GHashTable *found;
5177 bool ret;
Kevin Wolfdd62f1c2015-06-18 14:09:57 +02005178
Max Reitzbd86fb92020-05-13 13:05:13 +02005179 if (c->klass->stay_at_node) {
Kevin Wolfd0ac0382017-03-01 17:30:41 +01005180 return false;
5181 }
5182
Max Reitzec9f10f2018-06-13 20:18:15 +02005183 /* If the child @c belongs to the BDS @to, replacing the current
5184 * c->bs by @to would mean to create a loop.
5185 *
5186 * Such a case occurs when appending a BDS to a backing chain.
5187 * For instance, imagine the following chain:
5188 *
5189 * guest device -> node A -> further backing chain...
5190 *
5191 * Now we create a new BDS B which we want to put on top of this
5192 * chain, so we first attach A as its backing node:
5193 *
5194 * node B
5195 * |
5196 * v
5197 * guest device -> node A -> further backing chain...
5198 *
5199 * Finally we want to replace A by B. When doing that, we want to
5200 * replace all pointers to A by pointers to B -- except for the
5201 * pointer from B because (1) that would create a loop, and (2)
5202 * that pointer should simply stay intact:
5203 *
5204 * guest device -> node B
5205 * |
5206 * v
5207 * node A -> further backing chain...
5208 *
5209 * In general, when replacing a node A (c->bs) by a node B (@to),
5210 * if A is a child of B, that means we cannot replace A by B there
5211 * because that would create a loop. Silently detaching A from B
5212 * is also not really an option. So overall just leaving A in
Vladimir Sementsov-Ogievskiy2f30b7c2019-02-23 22:20:39 +03005213 * place there is the most sensible choice.
5214 *
5215 * We would also create a loop in any cases where @c is only
5216 * indirectly referenced by @to. Prevent this by returning false
5217 * if @c is found (by breadth-first search) anywhere in the whole
5218 * subtree of @to.
5219 */
5220
5221 ret = true;
5222 found = g_hash_table_new(NULL, NULL);
5223 g_hash_table_add(found, to);
5224 queue = g_queue_new();
5225 g_queue_push_tail(queue, to);
5226
5227 while (!g_queue_is_empty(queue)) {
5228 BlockDriverState *v = g_queue_pop_head(queue);
5229 BdrvChild *c2;
5230
5231 QLIST_FOREACH(c2, &v->children, next) {
5232 if (c2 == c) {
5233 ret = false;
5234 break;
5235 }
5236
5237 if (g_hash_table_contains(found, c2->bs)) {
5238 continue;
5239 }
5240
5241 g_queue_push_tail(queue, c2->bs);
5242 g_hash_table_add(found, c2->bs);
Kevin Wolfd0ac0382017-03-01 17:30:41 +01005243 }
5244 }
5245
Vladimir Sementsov-Ogievskiy2f30b7c2019-02-23 22:20:39 +03005246 g_queue_free(queue);
5247 g_hash_table_destroy(found);
5248
5249 return ret;
Kevin Wolfd0ac0382017-03-01 17:30:41 +01005250}
5251
Vladimir Sementsov-Ogievskiy57f08942022-07-26 23:11:34 +03005252static void bdrv_remove_child_commit(void *opaque)
Vladimir Sementsov-Ogievskiy46541ee2021-04-28 18:17:50 +03005253{
Emanuele Giuseppe Espositobdb73472022-03-03 10:16:02 -05005254 GLOBAL_STATE_CODE();
Vladimir Sementsov-Ogievskiy5bb047472022-07-26 23:11:32 +03005255 bdrv_child_free(opaque);
Vladimir Sementsov-Ogievskiy46541ee2021-04-28 18:17:50 +03005256}
5257
Vladimir Sementsov-Ogievskiy57f08942022-07-26 23:11:34 +03005258static TransactionActionDrv bdrv_remove_child_drv = {
5259 .commit = bdrv_remove_child_commit,
Vladimir Sementsov-Ogievskiy46541ee2021-04-28 18:17:50 +03005260};
5261
Kevin Wolf2f64e1f2023-09-11 11:46:08 +02005262/*
5263 * Function doesn't update permissions, caller is responsible for this.
5264 *
5265 * @child->bs (if non-NULL) must be drained.
Kevin Wolf5661a002023-09-11 11:46:10 +02005266 *
5267 * After calling this function, the transaction @tran may only be completed
5268 * while holding a writer lock for the graph.
Kevin Wolf2f64e1f2023-09-11 11:46:08 +02005269 */
5270static void GRAPH_WRLOCK bdrv_remove_child(BdrvChild *child, Transaction *tran)
Vladimir Sementsov-Ogievskiy46541ee2021-04-28 18:17:50 +03005271{
Vladimir Sementsov-Ogievskiy46541ee2021-04-28 18:17:50 +03005272 if (!child) {
5273 return;
5274 }
5275
5276 if (child->bs) {
Kevin Wolf2f64e1f2023-09-11 11:46:08 +02005277 assert(child->quiesced_parent);
Vladimir Sementsov-Ogievskiya2c37a32022-07-26 23:11:30 +03005278 bdrv_replace_child_tran(child, NULL, tran);
Vladimir Sementsov-Ogievskiy46541ee2021-04-28 18:17:50 +03005279 }
5280
Vladimir Sementsov-Ogievskiy57f08942022-07-26 23:11:34 +03005281 tran_add(tran, &bdrv_remove_child_drv, child);
Vladimir Sementsov-Ogievskiy46541ee2021-04-28 18:17:50 +03005282}
5283
Kevin Wolf2f64e1f2023-09-11 11:46:08 +02005284/*
5285 * Both @from and @to (if non-NULL) must be drained. @to must be kept drained
5286 * until the transaction is completed.
Kevin Wolf5661a002023-09-11 11:46:10 +02005287 *
5288 * After calling this function, the transaction @tran may only be completed
5289 * while holding a writer lock for the graph.
Kevin Wolf2f64e1f2023-09-11 11:46:08 +02005290 */
5291static int GRAPH_WRLOCK
5292bdrv_replace_node_noperm(BlockDriverState *from,
5293 BlockDriverState *to,
5294 bool auto_skip, Transaction *tran,
5295 Error **errp)
Vladimir Sementsov-Ogievskiy117caba2021-04-28 18:17:48 +03005296{
5297 BdrvChild *c, *next;
5298
Emanuele Giuseppe Espositobdb73472022-03-03 10:16:02 -05005299 GLOBAL_STATE_CODE();
Hanna Reitz82b54cf2021-11-15 15:54:04 +01005300
Kevin Wolf2f64e1f2023-09-11 11:46:08 +02005301 assert(from->quiesce_counter);
5302 assert(to->quiesce_counter);
Kevin Wolf23987472022-11-18 18:41:09 +01005303
Vladimir Sementsov-Ogievskiy117caba2021-04-28 18:17:48 +03005304 QLIST_FOREACH_SAFE(c, &from->parents, next_parent, next) {
5305 assert(c->bs == from);
5306 if (!should_update_child(c, to)) {
5307 if (auto_skip) {
5308 continue;
5309 }
5310 error_setg(errp, "Should not change '%s' link to '%s'",
5311 c->name, from->node_name);
5312 return -EINVAL;
5313 }
5314 if (c->frozen) {
5315 error_setg(errp, "Cannot change '%s' link to '%s'",
5316 c->name, from->node_name);
5317 return -EPERM;
5318 }
Vladimir Sementsov-Ogievskiy0f0b1e22022-07-26 23:11:29 +03005319 bdrv_replace_child_tran(c, to, tran);
Vladimir Sementsov-Ogievskiy117caba2021-04-28 18:17:48 +03005320 }
5321
5322 return 0;
5323}
5324
Vladimir Sementsov-Ogievskiy313274b2020-11-06 15:42:36 +03005325/*
Kevin Wolf5c0ef492023-10-27 17:53:24 +02005326 * Switch all parents of @from to point to @to instead. @from and @to must be in
5327 * the same AioContext and both must be drained.
5328 *
Vladimir Sementsov-Ogievskiy313274b2020-11-06 15:42:36 +03005329 * With auto_skip=true bdrv_replace_node_common skips updating from parents
5330 * if it creates a parent-child relation loop or if parent is block-job.
5331 *
5332 * With auto_skip=false the error is returned if from has a parent which should
5333 * not be updated.
Vladimir Sementsov-Ogievskiy3108a152021-04-28 18:17:51 +03005334 *
5335 * With @detach_subchain=true @to must be in a backing chain of @from. In this
5336 * case backing link of the cow-parent of @to is removed.
Vladimir Sementsov-Ogievskiy313274b2020-11-06 15:42:36 +03005337 */
Kevin Wolf5c0ef492023-10-27 17:53:24 +02005338static int GRAPH_WRLOCK
5339bdrv_replace_node_common(BlockDriverState *from, BlockDriverState *to,
5340 bool auto_skip, bool detach_subchain, Error **errp)
Kevin Wolfdd62f1c2015-06-18 14:09:57 +02005341{
Vladimir Sementsov-Ogievskiy3bb0e292021-04-28 18:17:45 +03005342 Transaction *tran = tran_new();
Vladimir Sementsov-Ogievskiy3bb0e292021-04-28 18:17:45 +03005343 g_autoptr(GSList) refresh_list = NULL;
Miroslav Rezanina2d369d62021-05-05 03:59:03 -04005344 BlockDriverState *to_cow_parent = NULL;
Kevin Wolf234ac1a2017-03-02 18:43:00 +01005345 int ret;
Kevin Wolfdd62f1c2015-06-18 14:09:57 +02005346
Emanuele Giuseppe Espositobdb73472022-03-03 10:16:02 -05005347 GLOBAL_STATE_CODE();
Hanna Reitz82b54cf2021-11-15 15:54:04 +01005348
Kevin Wolf5c0ef492023-10-27 17:53:24 +02005349 assert(from->quiesce_counter);
5350 assert(to->quiesce_counter);
Kevin Wolf30dd65f2020-03-10 12:38:29 +01005351 assert(bdrv_get_aio_context(from) == bdrv_get_aio_context(to));
Kevin Wolff871abd2019-05-21 19:00:25 +02005352
Kevin Wolf372b69f2023-10-27 17:53:15 +02005353 if (detach_subchain) {
5354 assert(bdrv_chain_contains(from, to));
5355 assert(from != to);
5356 for (to_cow_parent = from;
5357 bdrv_filter_or_cow_bs(to_cow_parent) != to;
5358 to_cow_parent = bdrv_filter_or_cow_bs(to_cow_parent))
5359 {
5360 ;
5361 }
5362 }
5363
Vladimir Sementsov-Ogievskiy3bb0e292021-04-28 18:17:45 +03005364 /*
5365 * Do the replacement without permission update.
5366 * Replacement may influence the permissions, we should calculate new
5367 * permissions based on new graph. If we fail, we'll roll-back the
5368 * replacement.
5369 */
Vladimir Sementsov-Ogievskiy117caba2021-04-28 18:17:48 +03005370 ret = bdrv_replace_node_noperm(from, to, auto_skip, tran, errp);
5371 if (ret < 0) {
5372 goto out;
Kevin Wolf234ac1a2017-03-02 18:43:00 +01005373 }
5374
Vladimir Sementsov-Ogievskiy3108a152021-04-28 18:17:51 +03005375 if (detach_subchain) {
Kevin Wolf2f64e1f2023-09-11 11:46:08 +02005376 /* to_cow_parent is already drained because from is drained */
Vladimir Sementsov-Ogievskiyf38eaec2022-11-07 19:35:56 +03005377 bdrv_remove_child(bdrv_filter_or_cow_child(to_cow_parent), tran);
Vladimir Sementsov-Ogievskiy3108a152021-04-28 18:17:51 +03005378 }
5379
Vladimir Sementsov-Ogievskiyfb0ff4d2022-11-07 19:35:58 +03005380 refresh_list = g_slist_prepend(refresh_list, to);
5381 refresh_list = g_slist_prepend(refresh_list, from);
Vladimir Sementsov-Ogievskiy3bb0e292021-04-28 18:17:45 +03005382
5383 ret = bdrv_list_refresh_perms(refresh_list, NULL, tran, errp);
Kevin Wolf234ac1a2017-03-02 18:43:00 +01005384 if (ret < 0) {
Kevin Wolf234ac1a2017-03-02 18:43:00 +01005385 goto out;
5386 }
5387
Vladimir Sementsov-Ogievskiya1e708f2021-02-02 15:49:43 +03005388 ret = 0;
5389
Kevin Wolf234ac1a2017-03-02 18:43:00 +01005390out:
Vladimir Sementsov-Ogievskiy3bb0e292021-04-28 18:17:45 +03005391 tran_finalize(tran, ret);
Kevin Wolf5c0ef492023-10-27 17:53:24 +02005392 return ret;
5393}
Vladimir Sementsov-Ogievskiy3bb0e292021-04-28 18:17:45 +03005394
Kevin Wolf5c0ef492023-10-27 17:53:24 +02005395int bdrv_replace_node(BlockDriverState *from, BlockDriverState *to,
5396 Error **errp)
5397{
Kevin Wolfccd6a372023-10-27 17:53:25 +02005398 return bdrv_replace_node_common(from, to, true, false, errp);
Kevin Wolfdd62f1c2015-06-18 14:09:57 +02005399}
5400
Vladimir Sementsov-Ogievskiy3108a152021-04-28 18:17:51 +03005401int bdrv_drop_filter(BlockDriverState *bs, Error **errp)
5402{
Kevin Wolf372b69f2023-10-27 17:53:15 +02005403 BlockDriverState *child_bs;
Kevin Wolf5c0ef492023-10-27 17:53:24 +02005404 int ret;
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05005405
Kevin Wolf372b69f2023-10-27 17:53:15 +02005406 GLOBAL_STATE_CODE();
Kevin Wolf5c0ef492023-10-27 17:53:24 +02005407
Kevin Wolf372b69f2023-10-27 17:53:15 +02005408 bdrv_graph_rdlock_main_loop();
5409 child_bs = bdrv_filter_or_cow_bs(bs);
5410 bdrv_graph_rdunlock_main_loop();
5411
Kevin Wolf5c0ef492023-10-27 17:53:24 +02005412 bdrv_drained_begin(child_bs);
Stefan Hajnoczi6bc30f12023-12-05 13:20:02 -05005413 bdrv_graph_wrlock();
Kevin Wolf5c0ef492023-10-27 17:53:24 +02005414 ret = bdrv_replace_node_common(bs, child_bs, true, true, errp);
Stefan Hajnoczi6bc30f12023-12-05 13:20:02 -05005415 bdrv_graph_wrunlock();
Kevin Wolf5c0ef492023-10-27 17:53:24 +02005416 bdrv_drained_end(child_bs);
5417
5418 return ret;
Vladimir Sementsov-Ogievskiy313274b2020-11-06 15:42:36 +03005419}
5420
Jeff Cody8802d1f2012-02-28 15:54:06 -05005421/*
5422 * Add new bs contents at the top of an image chain while the chain is
5423 * live, while keeping required fields on the top layer.
5424 *
5425 * This will modify the BlockDriverState fields, and swap contents
5426 * between bs_new and bs_top. Both bs_new and bs_top are modified.
5427 *
Vladimir Sementsov-Ogievskiy2272edc2021-04-28 18:17:49 +03005428 * bs_new must not be attached to a BlockBackend and must not have backing
5429 * child.
Jeff Codyf6801b82012-03-27 16:30:19 -04005430 *
Jeff Cody8802d1f2012-02-28 15:54:06 -05005431 * This function does not create any image files.
5432 */
Vladimir Sementsov-Ogievskiya1e708f2021-02-02 15:49:43 +03005433int bdrv_append(BlockDriverState *bs_new, BlockDriverState *bs_top,
5434 Error **errp)
Jeff Cody8802d1f2012-02-28 15:54:06 -05005435{
Vladimir Sementsov-Ogievskiy2272edc2021-04-28 18:17:49 +03005436 int ret;
Vladimir Sementsov-Ogievskiy5bb047472022-07-26 23:11:32 +03005437 BdrvChild *child;
Vladimir Sementsov-Ogievskiy2272edc2021-04-28 18:17:49 +03005438 Transaction *tran = tran_new();
5439
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05005440 GLOBAL_STATE_CODE();
5441
Kevin Wolf004915a2023-10-27 17:53:26 +02005442 bdrv_graph_rdlock_main_loop();
Vladimir Sementsov-Ogievskiy2272edc2021-04-28 18:17:49 +03005443 assert(!bs_new->backing);
Kevin Wolf004915a2023-10-27 17:53:26 +02005444 bdrv_graph_rdunlock_main_loop();
Vladimir Sementsov-Ogievskiy2272edc2021-04-28 18:17:49 +03005445
Kevin Wolf7d4ca9d2023-09-11 11:46:09 +02005446 bdrv_drained_begin(bs_top);
Kevin Wolf7d4ca9d2023-09-11 11:46:09 +02005447 bdrv_drained_begin(bs_new);
Kevin Wolf7d4ca9d2023-09-11 11:46:09 +02005448
Stefan Hajnoczi6bc30f12023-12-05 13:20:02 -05005449 bdrv_graph_wrlock();
Stefano Garzarella60d90bf2023-02-14 18:16:21 +01005450
Vladimir Sementsov-Ogievskiy5bb047472022-07-26 23:11:32 +03005451 child = bdrv_attach_child_noperm(bs_new, bs_top, "backing",
5452 &child_of_bds, bdrv_backing_role(bs_new),
5453 tran, errp);
5454 if (!child) {
5455 ret = -EINVAL;
Vladimir Sementsov-Ogievskiy2272edc2021-04-28 18:17:49 +03005456 goto out;
Kevin Wolfb2c28322017-02-20 12:46:42 +01005457 }
Kevin Wolfdd62f1c2015-06-18 14:09:57 +02005458
Vladimir Sementsov-Ogievskiy2272edc2021-04-28 18:17:49 +03005459 ret = bdrv_replace_node_noperm(bs_top, bs_new, true, tran, errp);
Vladimir Sementsov-Ogievskiya1e708f2021-02-02 15:49:43 +03005460 if (ret < 0) {
Vladimir Sementsov-Ogievskiy2272edc2021-04-28 18:17:49 +03005461 goto out;
Kevin Wolf234ac1a2017-03-02 18:43:00 +01005462 }
Kevin Wolfdd62f1c2015-06-18 14:09:57 +02005463
Vladimir Sementsov-Ogievskiyf1316ed2022-11-07 19:35:57 +03005464 ret = bdrv_refresh_perms(bs_new, tran, errp);
Vladimir Sementsov-Ogievskiy2272edc2021-04-28 18:17:49 +03005465out:
5466 tran_finalize(tran, ret);
5467
Vladimir Sementsov-Ogievskiy1e4c7972021-04-28 18:17:55 +03005468 bdrv_refresh_limits(bs_top, NULL, NULL);
Stefan Hajnoczi6bc30f12023-12-05 13:20:02 -05005469 bdrv_graph_wrunlock();
Vladimir Sementsov-Ogievskiy2272edc2021-04-28 18:17:49 +03005470
Kevin Wolf7d4ca9d2023-09-11 11:46:09 +02005471 bdrv_drained_end(bs_top);
5472 bdrv_drained_end(bs_new);
Kevin Wolf2f64e1f2023-09-11 11:46:08 +02005473
Vladimir Sementsov-Ogievskiy2272edc2021-04-28 18:17:49 +03005474 return ret;
Jeff Cody8802d1f2012-02-28 15:54:06 -05005475}
5476
Vladimir Sementsov-Ogievskiybd8f4c42021-08-24 11:38:23 +03005477/* Not for empty child */
5478int bdrv_replace_child_bs(BdrvChild *child, BlockDriverState *new_bs,
5479 Error **errp)
5480{
5481 int ret;
5482 Transaction *tran = tran_new();
Vladimir Sementsov-Ogievskiybd8f4c42021-08-24 11:38:23 +03005483 g_autoptr(GSList) refresh_list = NULL;
5484 BlockDriverState *old_bs = child->bs;
5485
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05005486 GLOBAL_STATE_CODE();
5487
Vladimir Sementsov-Ogievskiybd8f4c42021-08-24 11:38:23 +03005488 bdrv_ref(old_bs);
5489 bdrv_drained_begin(old_bs);
5490 bdrv_drained_begin(new_bs);
Stefan Hajnoczi6bc30f12023-12-05 13:20:02 -05005491 bdrv_graph_wrlock();
Vladimir Sementsov-Ogievskiybd8f4c42021-08-24 11:38:23 +03005492
Vladimir Sementsov-Ogievskiy0f0b1e22022-07-26 23:11:29 +03005493 bdrv_replace_child_tran(child, new_bs, tran);
Vladimir Sementsov-Ogievskiybd8f4c42021-08-24 11:38:23 +03005494
Vladimir Sementsov-Ogievskiyfb0ff4d2022-11-07 19:35:58 +03005495 refresh_list = g_slist_prepend(refresh_list, old_bs);
5496 refresh_list = g_slist_prepend(refresh_list, new_bs);
Vladimir Sementsov-Ogievskiybd8f4c42021-08-24 11:38:23 +03005497
5498 ret = bdrv_list_refresh_perms(refresh_list, NULL, tran, errp);
5499
5500 tran_finalize(tran, ret);
5501
Stefan Hajnoczi6bc30f12023-12-05 13:20:02 -05005502 bdrv_graph_wrunlock();
Vladimir Sementsov-Ogievskiybd8f4c42021-08-24 11:38:23 +03005503 bdrv_drained_end(old_bs);
5504 bdrv_drained_end(new_bs);
5505 bdrv_unref(old_bs);
5506
5507 return ret;
5508}
5509
Fam Zheng4f6fd342013-08-23 09:14:47 +08005510static void bdrv_delete(BlockDriverState *bs)
bellardb3380822004-03-14 21:38:54 +00005511{
Fam Zheng3718d8a2014-05-23 21:29:43 +08005512 assert(bdrv_op_blocker_is_empty(bs));
Fam Zheng4f6fd342013-08-23 09:14:47 +08005513 assert(!bs->refcnt);
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05005514 GLOBAL_STATE_CODE();
Markus Armbruster18846de2010-06-29 16:58:30 +02005515
Stefan Hajnoczi1b7bdbc2010-04-10 07:02:42 +01005516 /* remove from list, if necessary */
Kevin Wolf63eaaae2016-03-18 10:46:57 +01005517 if (bs->node_name[0] != '\0') {
5518 QTAILQ_REMOVE(&graph_bdrv_states, bs, node_list);
5519 }
Max Reitz2c1d04e2016-01-29 16:36:11 +01005520 QTAILQ_REMOVE(&all_bdrv_states, bs, bs_list);
5521
Anton Kuchin30c321f2019-05-07 11:12:56 +03005522 bdrv_close(bs);
5523
Stefan Hajnoczifa9185f2023-08-08 11:58:52 -04005524 qemu_mutex_destroy(&bs->reqs_lock);
5525
Anthony Liguori7267c092011-08-20 22:09:37 -05005526 g_free(bs);
bellardfc01f7e2003-06-30 10:03:06 +00005527}
5528
Vladimir Sementsov-Ogievskiy96796fa2021-09-20 14:55:36 +03005529
5530/*
5531 * Replace @bs by newly created block node.
5532 *
5533 * @options is a QDict of options to pass to the block drivers, or NULL for an
5534 * empty set of options. The reference to the QDict belongs to the block layer
5535 * after the call (even on failure), so if the caller intends to reuse the
5536 * dictionary, it needs to use qobject_ref() before calling bdrv_open.
Kevin Wolf88234072023-05-25 14:47:11 +02005537 *
Stefan Hajnoczi23c983c2023-12-05 13:20:11 -05005538 * The caller must make sure that @bs stays in the same AioContext, i.e.
5539 * @options must not refer to nodes in a different AioContext.
Vladimir Sementsov-Ogievskiy96796fa2021-09-20 14:55:36 +03005540 */
5541BlockDriverState *bdrv_insert_node(BlockDriverState *bs, QDict *options,
Andrey Shinkevich8872ef72020-12-16 09:16:52 +03005542 int flags, Error **errp)
5543{
Vladimir Sementsov-Ogievskiyf053b7e2021-09-20 14:55:35 +03005544 ERRP_GUARD();
5545 int ret;
Kevin Wolf88234072023-05-25 14:47:11 +02005546 AioContext *ctx = bdrv_get_aio_context(bs);
Vladimir Sementsov-Ogievskiyb11c8732021-09-20 14:55:37 +03005547 BlockDriverState *new_node_bs = NULL;
5548 const char *drvname, *node_name;
5549 BlockDriver *drv;
Andrey Shinkevich8872ef72020-12-16 09:16:52 +03005550
Vladimir Sementsov-Ogievskiyb11c8732021-09-20 14:55:37 +03005551 drvname = qdict_get_try_str(options, "driver");
5552 if (!drvname) {
5553 error_setg(errp, "driver is not specified");
5554 goto fail;
5555 }
5556
5557 drv = bdrv_find_format(drvname);
5558 if (!drv) {
5559 error_setg(errp, "Unknown driver: '%s'", drvname);
5560 goto fail;
5561 }
5562
5563 node_name = qdict_get_try_str(options, "node-name");
5564
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05005565 GLOBAL_STATE_CODE();
5566
Vladimir Sementsov-Ogievskiyb11c8732021-09-20 14:55:37 +03005567 new_node_bs = bdrv_new_open_driver_opts(drv, node_name, options, flags,
5568 errp);
Kevin Wolf88234072023-05-25 14:47:11 +02005569 assert(bdrv_get_aio_context(bs) == ctx);
5570
Vladimir Sementsov-Ogievskiyb11c8732021-09-20 14:55:37 +03005571 options = NULL; /* bdrv_new_open_driver() eats options */
5572 if (!new_node_bs) {
Andrey Shinkevich8872ef72020-12-16 09:16:52 +03005573 error_prepend(errp, "Could not create node: ");
Vladimir Sementsov-Ogievskiyb11c8732021-09-20 14:55:37 +03005574 goto fail;
Andrey Shinkevich8872ef72020-12-16 09:16:52 +03005575 }
5576
Kevin Wolfccd6a372023-10-27 17:53:25 +02005577 /*
5578 * Make sure that @bs doesn't go away until we have successfully attached
5579 * all of its parents to @new_node_bs and undrained it again.
5580 */
5581 bdrv_ref(bs);
Andrey Shinkevich8872ef72020-12-16 09:16:52 +03005582 bdrv_drained_begin(bs);
Kevin Wolfccd6a372023-10-27 17:53:25 +02005583 bdrv_drained_begin(new_node_bs);
Stefan Hajnoczi6bc30f12023-12-05 13:20:02 -05005584 bdrv_graph_wrlock();
Vladimir Sementsov-Ogievskiyf053b7e2021-09-20 14:55:35 +03005585 ret = bdrv_replace_node(bs, new_node_bs, errp);
Stefan Hajnoczi6bc30f12023-12-05 13:20:02 -05005586 bdrv_graph_wrunlock();
Kevin Wolfccd6a372023-10-27 17:53:25 +02005587 bdrv_drained_end(new_node_bs);
Andrey Shinkevich8872ef72020-12-16 09:16:52 +03005588 bdrv_drained_end(bs);
Kevin Wolfccd6a372023-10-27 17:53:25 +02005589 bdrv_unref(bs);
Andrey Shinkevich8872ef72020-12-16 09:16:52 +03005590
Vladimir Sementsov-Ogievskiyf053b7e2021-09-20 14:55:35 +03005591 if (ret < 0) {
5592 error_prepend(errp, "Could not replace node: ");
Vladimir Sementsov-Ogievskiyb11c8732021-09-20 14:55:37 +03005593 goto fail;
Andrey Shinkevich8872ef72020-12-16 09:16:52 +03005594 }
5595
5596 return new_node_bs;
Vladimir Sementsov-Ogievskiyb11c8732021-09-20 14:55:37 +03005597
5598fail:
5599 qobject_unref(options);
5600 bdrv_unref(new_node_bs);
5601 return NULL;
Andrey Shinkevich8872ef72020-12-16 09:16:52 +03005602}
5603
aliguorie97fc192009-04-21 23:11:50 +00005604/*
5605 * Run consistency checks on an image
5606 *
Kevin Wolfe076f332010-06-29 11:43:13 +02005607 * Returns 0 if the check could be completed (it doesn't mean that the image is
Stefan Weila1c72732011-04-28 17:20:38 +02005608 * free of errors) or -errno when an internal error occurred. The results of the
Kevin Wolfe076f332010-06-29 11:43:13 +02005609 * check are stored in res.
aliguorie97fc192009-04-21 23:11:50 +00005610 */
Vladimir Sementsov-Ogievskiy21c22832020-09-24 21:54:10 +03005611int coroutine_fn bdrv_co_check(BlockDriverState *bs,
5612 BdrvCheckResult *res, BdrvCheckMode fix)
aliguorie97fc192009-04-21 23:11:50 +00005613{
Emanuele Giuseppe Esposito1581a702022-03-03 10:16:09 -05005614 IO_CODE();
Kevin Wolf1b3ff9f2022-12-07 14:18:38 +01005615 assert_bdrv_graph_readable();
Max Reitz908bcd52014-08-07 22:47:55 +02005616 if (bs->drv == NULL) {
5617 return -ENOMEDIUM;
5618 }
Paolo Bonzini2fd61632018-03-01 17:36:19 +01005619 if (bs->drv->bdrv_co_check == NULL) {
aliguorie97fc192009-04-21 23:11:50 +00005620 return -ENOTSUP;
5621 }
5622
Kevin Wolfe076f332010-06-29 11:43:13 +02005623 memset(res, 0, sizeof(*res));
Paolo Bonzini2fd61632018-03-01 17:36:19 +01005624 return bs->drv->bdrv_co_check(bs, res, fix);
5625}
5626
Kevin Wolf756e6732010-01-12 12:55:17 +01005627/*
5628 * Return values:
5629 * 0 - success
5630 * -EINVAL - backing format specified, but no file
5631 * -ENOSPC - can't update the backing file because no space is left in the
5632 * image file header
5633 * -ENOTSUP - format driver doesn't support changing the backing file
5634 */
Kevin Wolfe2dd2732023-10-27 17:53:28 +02005635int coroutine_fn
5636bdrv_co_change_backing_file(BlockDriverState *bs, const char *backing_file,
5637 const char *backing_fmt, bool require)
Kevin Wolf756e6732010-01-12 12:55:17 +01005638{
5639 BlockDriver *drv = bs->drv;
Paolo Bonzini469ef352012-04-12 14:01:02 +02005640 int ret;
Kevin Wolf756e6732010-01-12 12:55:17 +01005641
Kevin Wolfe2dd2732023-10-27 17:53:28 +02005642 IO_CODE();
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05005643
Max Reitzd470ad42017-11-10 21:31:09 +01005644 if (!drv) {
5645 return -ENOMEDIUM;
5646 }
5647
Paolo Bonzini5f377792012-04-12 14:01:01 +02005648 /* Backing file format doesn't make sense without a backing file */
5649 if (backing_fmt && !backing_file) {
5650 return -EINVAL;
5651 }
5652
Eric Blake497a30d2021-05-03 14:36:00 -07005653 if (require && backing_file && !backing_fmt) {
5654 return -EINVAL;
Eric Blakee54ee1b2020-07-06 15:39:53 -05005655 }
5656
Kevin Wolfe2dd2732023-10-27 17:53:28 +02005657 if (drv->bdrv_co_change_backing_file != NULL) {
5658 ret = drv->bdrv_co_change_backing_file(bs, backing_file, backing_fmt);
Kevin Wolf756e6732010-01-12 12:55:17 +01005659 } else {
Paolo Bonzini469ef352012-04-12 14:01:02 +02005660 ret = -ENOTSUP;
Kevin Wolf756e6732010-01-12 12:55:17 +01005661 }
Paolo Bonzini469ef352012-04-12 14:01:02 +02005662
5663 if (ret == 0) {
5664 pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_file ?: "");
5665 pstrcpy(bs->backing_format, sizeof(bs->backing_format), backing_fmt ?: "");
Max Reitz998c2012019-02-01 20:29:08 +01005666 pstrcpy(bs->auto_backing_file, sizeof(bs->auto_backing_file),
5667 backing_file ?: "");
Paolo Bonzini469ef352012-04-12 14:01:02 +02005668 }
5669 return ret;
Kevin Wolf756e6732010-01-12 12:55:17 +01005670}
5671
Jeff Cody6ebdcee2012-09-27 13:29:12 -04005672/*
Max Reitzdcf3f9b2019-06-12 17:34:45 +02005673 * Finds the first non-filter node above bs in the chain between
5674 * active and bs. The returned node is either an immediate parent of
5675 * bs, or there are only filter nodes between the two.
Jeff Cody6ebdcee2012-09-27 13:29:12 -04005676 *
5677 * Returns NULL if bs is not found in active's image chain,
5678 * or if active == bs.
Jeff Cody4caf0fc2014-06-25 15:35:26 -04005679 *
5680 * Returns the bottommost base image if bs == NULL.
Jeff Cody6ebdcee2012-09-27 13:29:12 -04005681 */
5682BlockDriverState *bdrv_find_overlay(BlockDriverState *active,
5683 BlockDriverState *bs)
5684{
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05005685
5686 GLOBAL_STATE_CODE();
5687
Max Reitzdcf3f9b2019-06-12 17:34:45 +02005688 bs = bdrv_skip_filters(bs);
5689 active = bdrv_skip_filters(active);
5690
5691 while (active) {
5692 BlockDriverState *next = bdrv_backing_chain_next(active);
5693 if (bs == next) {
5694 return active;
5695 }
5696 active = next;
Jeff Cody6ebdcee2012-09-27 13:29:12 -04005697 }
5698
Max Reitzdcf3f9b2019-06-12 17:34:45 +02005699 return NULL;
Jeff Cody4caf0fc2014-06-25 15:35:26 -04005700}
Jeff Cody6ebdcee2012-09-27 13:29:12 -04005701
Jeff Cody4caf0fc2014-06-25 15:35:26 -04005702/* Given a BDS, searches for the base layer. */
5703BlockDriverState *bdrv_find_base(BlockDriverState *bs)
5704{
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05005705 GLOBAL_STATE_CODE();
5706
Jeff Cody4caf0fc2014-06-25 15:35:26 -04005707 return bdrv_find_overlay(bs, NULL);
Jeff Cody6ebdcee2012-09-27 13:29:12 -04005708}
5709
Jeff Cody6ebdcee2012-09-27 13:29:12 -04005710/*
Max Reitz7b99a262019-06-12 16:07:11 +02005711 * Return true if at least one of the COW (backing) and filter links
5712 * between @bs and @base is frozen. @errp is set if that's the case.
Alberto Garcia0f0998f2019-03-28 18:25:09 +02005713 * @base must be reachable from @bs, or NULL.
Alberto Garcia2cad1eb2019-03-12 18:48:40 +02005714 */
Kevin Wolf9275fc72023-10-27 17:53:18 +02005715static bool GRAPH_RDLOCK
5716bdrv_is_backing_chain_frozen(BlockDriverState *bs, BlockDriverState *base,
5717 Error **errp)
Alberto Garcia2cad1eb2019-03-12 18:48:40 +02005718{
5719 BlockDriverState *i;
Max Reitz7b99a262019-06-12 16:07:11 +02005720 BdrvChild *child;
Alberto Garcia2cad1eb2019-03-12 18:48:40 +02005721
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05005722 GLOBAL_STATE_CODE();
5723
Max Reitz7b99a262019-06-12 16:07:11 +02005724 for (i = bs; i != base; i = child_bs(child)) {
5725 child = bdrv_filter_or_cow_child(i);
5726
5727 if (child && child->frozen) {
Alberto Garcia2cad1eb2019-03-12 18:48:40 +02005728 error_setg(errp, "Cannot change '%s' link from '%s' to '%s'",
Max Reitz7b99a262019-06-12 16:07:11 +02005729 child->name, i->node_name, child->bs->node_name);
Alberto Garcia2cad1eb2019-03-12 18:48:40 +02005730 return true;
5731 }
5732 }
5733
5734 return false;
5735}
5736
5737/*
Max Reitz7b99a262019-06-12 16:07:11 +02005738 * Freeze all COW (backing) and filter links between @bs and @base.
Alberto Garcia2cad1eb2019-03-12 18:48:40 +02005739 * If any of the links is already frozen the operation is aborted and
5740 * none of the links are modified.
Alberto Garcia0f0998f2019-03-28 18:25:09 +02005741 * @base must be reachable from @bs, or NULL.
Alberto Garcia2cad1eb2019-03-12 18:48:40 +02005742 * Returns 0 on success. On failure returns < 0 and sets @errp.
5743 */
5744int bdrv_freeze_backing_chain(BlockDriverState *bs, BlockDriverState *base,
5745 Error **errp)
5746{
5747 BlockDriverState *i;
Max Reitz7b99a262019-06-12 16:07:11 +02005748 BdrvChild *child;
Alberto Garcia2cad1eb2019-03-12 18:48:40 +02005749
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05005750 GLOBAL_STATE_CODE();
5751
Alberto Garcia2cad1eb2019-03-12 18:48:40 +02005752 if (bdrv_is_backing_chain_frozen(bs, base, errp)) {
5753 return -EPERM;
5754 }
5755
Max Reitz7b99a262019-06-12 16:07:11 +02005756 for (i = bs; i != base; i = child_bs(child)) {
5757 child = bdrv_filter_or_cow_child(i);
5758 if (child && child->bs->never_freeze) {
Max Reitze5182c12019-07-03 19:28:02 +02005759 error_setg(errp, "Cannot freeze '%s' link to '%s'",
Max Reitz7b99a262019-06-12 16:07:11 +02005760 child->name, child->bs->node_name);
Max Reitze5182c12019-07-03 19:28:02 +02005761 return -EPERM;
5762 }
5763 }
5764
Max Reitz7b99a262019-06-12 16:07:11 +02005765 for (i = bs; i != base; i = child_bs(child)) {
5766 child = bdrv_filter_or_cow_child(i);
5767 if (child) {
5768 child->frozen = true;
Alberto Garcia0f0998f2019-03-28 18:25:09 +02005769 }
Alberto Garcia2cad1eb2019-03-12 18:48:40 +02005770 }
5771
5772 return 0;
5773}
5774
5775/*
Max Reitz7b99a262019-06-12 16:07:11 +02005776 * Unfreeze all COW (backing) and filter links between @bs and @base.
5777 * The caller must ensure that all links are frozen before using this
5778 * function.
Alberto Garcia0f0998f2019-03-28 18:25:09 +02005779 * @base must be reachable from @bs, or NULL.
Alberto Garcia2cad1eb2019-03-12 18:48:40 +02005780 */
5781void bdrv_unfreeze_backing_chain(BlockDriverState *bs, BlockDriverState *base)
5782{
5783 BlockDriverState *i;
Max Reitz7b99a262019-06-12 16:07:11 +02005784 BdrvChild *child;
Alberto Garcia2cad1eb2019-03-12 18:48:40 +02005785
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05005786 GLOBAL_STATE_CODE();
5787
Max Reitz7b99a262019-06-12 16:07:11 +02005788 for (i = bs; i != base; i = child_bs(child)) {
5789 child = bdrv_filter_or_cow_child(i);
5790 if (child) {
5791 assert(child->frozen);
5792 child->frozen = false;
Alberto Garcia0f0998f2019-03-28 18:25:09 +02005793 }
Alberto Garcia2cad1eb2019-03-12 18:48:40 +02005794 }
5795}
5796
5797/*
Jeff Cody6ebdcee2012-09-27 13:29:12 -04005798 * Drops images above 'base' up to and including 'top', and sets the image
5799 * above 'top' to have base as its backing file.
5800 *
5801 * Requires that the overlay to 'top' is opened r/w, so that the backing file
5802 * information in 'bs' can be properly updated.
5803 *
5804 * E.g., this will convert the following chain:
5805 * bottom <- base <- intermediate <- top <- active
5806 *
5807 * to
5808 *
5809 * bottom <- base <- active
5810 *
5811 * It is allowed for bottom==base, in which case it converts:
5812 *
5813 * base <- intermediate <- top <- active
5814 *
5815 * to
5816 *
5817 * base <- active
5818 *
Jeff Cody54e26902014-06-25 15:40:10 -04005819 * If backing_file_str is non-NULL, it will be used when modifying top's
5820 * overlay image metadata.
5821 *
Jeff Cody6ebdcee2012-09-27 13:29:12 -04005822 * Error conditions:
5823 * if active == top, that is considered an error
5824 *
5825 */
Kevin Wolfbde70712017-06-27 20:36:18 +02005826int bdrv_drop_intermediate(BlockDriverState *top, BlockDriverState *base,
Peter Krempa4b028cb2023-12-05 18:14:41 +01005827 const char *backing_file_str,
5828 bool backing_mask_protocol)
Jeff Cody6ebdcee2012-09-27 13:29:12 -04005829{
Alberto Garcia6bd858b2018-10-31 18:16:38 +02005830 BlockDriverState *explicit_top = top;
5831 bool update_inherits_from;
Vladimir Sementsov-Ogievskiyd669ed62020-11-06 15:42:37 +03005832 BdrvChild *c;
Kevin Wolf12fa4af2017-02-17 20:42:32 +01005833 Error *local_err = NULL;
Jeff Cody6ebdcee2012-09-27 13:29:12 -04005834 int ret = -EIO;
Vladimir Sementsov-Ogievskiyd669ed62020-11-06 15:42:37 +03005835 g_autoptr(GSList) updated_children = NULL;
5836 GSList *p;
Jeff Cody6ebdcee2012-09-27 13:29:12 -04005837
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05005838 GLOBAL_STATE_CODE();
5839
Kevin Wolf6858eba2017-06-29 19:32:21 +02005840 bdrv_ref(top);
Kevin Wolf631086d2022-11-18 18:41:03 +01005841 bdrv_drained_begin(base);
Stefan Hajnoczi6bc30f12023-12-05 13:20:02 -05005842 bdrv_graph_wrlock();
Kevin Wolf6858eba2017-06-29 19:32:21 +02005843
Jeff Cody6ebdcee2012-09-27 13:29:12 -04005844 if (!top->drv || !base->drv) {
Kevin Wolf5c0ef492023-10-27 17:53:24 +02005845 goto exit_wrlock;
Jeff Cody6ebdcee2012-09-27 13:29:12 -04005846 }
5847
Kevin Wolf5db15a52015-09-14 15:33:33 +02005848 /* Make sure that base is in the backing chain of top */
5849 if (!bdrv_chain_contains(top, base)) {
Kevin Wolf5c0ef492023-10-27 17:53:24 +02005850 goto exit_wrlock;
Jeff Cody6ebdcee2012-09-27 13:29:12 -04005851 }
5852
Alberto Garcia6bd858b2018-10-31 18:16:38 +02005853 /* If 'base' recursively inherits from 'top' then we should set
5854 * base->inherits_from to top->inherits_from after 'top' and all
5855 * other intermediate nodes have been dropped.
5856 * If 'top' is an implicit node (e.g. "commit_top") we should skip
5857 * it because no one inherits from it. We use explicit_top for that. */
Max Reitzdcf3f9b2019-06-12 17:34:45 +02005858 explicit_top = bdrv_skip_implicit_filters(explicit_top);
Alberto Garcia6bd858b2018-10-31 18:16:38 +02005859 update_inherits_from = bdrv_inherits_from_recursive(base, explicit_top);
5860
Jeff Cody6ebdcee2012-09-27 13:29:12 -04005861 /* success - we can delete the intermediate states, and link top->base */
Max Reitzf30c66b2019-02-01 20:29:05 +01005862 if (!backing_file_str) {
5863 bdrv_refresh_filename(base);
5864 backing_file_str = base->filename;
5865 }
Kevin Wolf61f09ce2017-09-19 16:22:54 +02005866
Vladimir Sementsov-Ogievskiyd669ed62020-11-06 15:42:37 +03005867 QLIST_FOREACH(c, &top->parents, next_parent) {
5868 updated_children = g_slist_prepend(updated_children, c);
5869 }
Kevin Wolf12fa4af2017-02-17 20:42:32 +01005870
Vladimir Sementsov-Ogievskiy3108a152021-04-28 18:17:51 +03005871 /*
5872 * It seems correct to pass detach_subchain=true here, but it triggers
5873 * one more yet not fixed bug, when due to nested aio_poll loop we switch to
5874 * another drained section, which modify the graph (for example, removing
5875 * the child, which we keep in updated_children list). So, it's a TODO.
5876 *
5877 * Note, bug triggered if pass detach_subchain=true here and run
5878 * test-bdrv-drain. test_drop_intermediate_poll() test-case will crash.
5879 * That's a FIXME.
5880 */
5881 bdrv_replace_node_common(top, base, false, false, &local_err);
Stefan Hajnoczi6bc30f12023-12-05 13:20:02 -05005882 bdrv_graph_wrunlock();
Kevin Wolf5c0ef492023-10-27 17:53:24 +02005883
Vladimir Sementsov-Ogievskiyd669ed62020-11-06 15:42:37 +03005884 if (local_err) {
5885 error_report_err(local_err);
5886 goto exit;
5887 }
5888
5889 for (p = updated_children; p; p = p->next) {
5890 c = p->data;
5891
Max Reitzbd86fb92020-05-13 13:05:13 +02005892 if (c->klass->update_filename) {
5893 ret = c->klass->update_filename(c, base, backing_file_str,
Peter Krempa4b028cb2023-12-05 18:14:41 +01005894 backing_mask_protocol,
Max Reitzbd86fb92020-05-13 13:05:13 +02005895 &local_err);
Kevin Wolf61f09ce2017-09-19 16:22:54 +02005896 if (ret < 0) {
Vladimir Sementsov-Ogievskiyd669ed62020-11-06 15:42:37 +03005897 /*
5898 * TODO: Actually, we want to rollback all previous iterations
5899 * of this loop, and (which is almost impossible) previous
5900 * bdrv_replace_node()...
5901 *
5902 * Note, that c->klass->update_filename may lead to permission
5903 * update, so it's a bad idea to call it inside permission
5904 * update transaction of bdrv_replace_node.
5905 */
Kevin Wolf61f09ce2017-09-19 16:22:54 +02005906 error_report_err(local_err);
5907 goto exit;
5908 }
5909 }
Kevin Wolf12fa4af2017-02-17 20:42:32 +01005910 }
Jeff Cody6ebdcee2012-09-27 13:29:12 -04005911
Alberto Garcia6bd858b2018-10-31 18:16:38 +02005912 if (update_inherits_from) {
5913 base->inherits_from = explicit_top->inherits_from;
5914 }
5915
Jeff Cody6ebdcee2012-09-27 13:29:12 -04005916 ret = 0;
Kevin Wolf5c0ef492023-10-27 17:53:24 +02005917 goto exit;
5918
5919exit_wrlock:
Stefan Hajnoczi6bc30f12023-12-05 13:20:02 -05005920 bdrv_graph_wrunlock();
Jeff Cody6ebdcee2012-09-27 13:29:12 -04005921exit:
Kevin Wolf631086d2022-11-18 18:41:03 +01005922 bdrv_drained_end(base);
Kevin Wolf6858eba2017-06-29 19:32:21 +02005923 bdrv_unref(top);
Jeff Cody6ebdcee2012-09-27 13:29:12 -04005924 return ret;
5925}
5926
bellard83f64092006-08-01 16:21:11 +00005927/**
Emanuele Giuseppe Esposito82618d72023-01-13 21:42:07 +01005928 * Implementation of BlockDriver.bdrv_co_get_allocated_file_size() that
Max Reitz081e4652019-06-12 18:14:13 +02005929 * sums the size of all data-bearing children. (This excludes backing
5930 * children.)
5931 */
Emanuele Giuseppe Espositode335632023-05-04 13:57:43 +02005932static int64_t coroutine_fn GRAPH_RDLOCK
5933bdrv_sum_allocated_file_size(BlockDriverState *bs)
Max Reitz081e4652019-06-12 18:14:13 +02005934{
5935 BdrvChild *child;
5936 int64_t child_size, sum = 0;
5937
5938 QLIST_FOREACH(child, &bs->children, next) {
5939 if (child->role & (BDRV_CHILD_DATA | BDRV_CHILD_METADATA |
5940 BDRV_CHILD_FILTERED))
5941 {
Emanuele Giuseppe Esposito82618d72023-01-13 21:42:07 +01005942 child_size = bdrv_co_get_allocated_file_size(child->bs);
Max Reitz081e4652019-06-12 18:14:13 +02005943 if (child_size < 0) {
5944 return child_size;
5945 }
5946 sum += child_size;
5947 }
5948 }
5949
5950 return sum;
5951}
5952
5953/**
Fam Zheng4a1d5e12011-07-12 19:56:39 +08005954 * Length of a allocated file in bytes. Sparse files are counted by actual
5955 * allocated space. Return < 0 if error or unknown.
5956 */
Emanuele Giuseppe Esposito82618d72023-01-13 21:42:07 +01005957int64_t coroutine_fn bdrv_co_get_allocated_file_size(BlockDriverState *bs)
Fam Zheng4a1d5e12011-07-12 19:56:39 +08005958{
5959 BlockDriver *drv = bs->drv;
Emanuele Giuseppe Esposito384a48f2022-03-03 10:15:50 -05005960 IO_CODE();
Emanuele Giuseppe Espositode335632023-05-04 13:57:43 +02005961 assert_bdrv_graph_readable();
Emanuele Giuseppe Esposito384a48f2022-03-03 10:15:50 -05005962
Fam Zheng4a1d5e12011-07-12 19:56:39 +08005963 if (!drv) {
5964 return -ENOMEDIUM;
5965 }
Emanuele Giuseppe Esposito82618d72023-01-13 21:42:07 +01005966 if (drv->bdrv_co_get_allocated_file_size) {
5967 return drv->bdrv_co_get_allocated_file_size(bs);
Fam Zheng4a1d5e12011-07-12 19:56:39 +08005968 }
Max Reitz081e4652019-06-12 18:14:13 +02005969
5970 if (drv->bdrv_file_open) {
5971 /*
5972 * Protocol drivers default to -ENOTSUP (most of their data is
5973 * not stored in any of their children (if they even have any),
5974 * so there is no generic way to figure it out).
5975 */
5976 return -ENOTSUP;
5977 } else if (drv->is_filter) {
5978 /* Filter drivers default to the size of their filtered child */
Emanuele Giuseppe Esposito82618d72023-01-13 21:42:07 +01005979 return bdrv_co_get_allocated_file_size(bdrv_filter_bs(bs));
Max Reitz081e4652019-06-12 18:14:13 +02005980 } else {
5981 /* Other drivers default to summing their children's sizes */
5982 return bdrv_sum_allocated_file_size(bs);
Fam Zheng4a1d5e12011-07-12 19:56:39 +08005983 }
Fam Zheng4a1d5e12011-07-12 19:56:39 +08005984}
5985
Stefan Hajnoczi90880ff2017-07-05 13:57:30 +01005986/*
5987 * bdrv_measure:
5988 * @drv: Format driver
5989 * @opts: Creation options for new image
5990 * @in_bs: Existing image containing data for new image (may be NULL)
5991 * @errp: Error object
5992 * Returns: A #BlockMeasureInfo (free using qapi_free_BlockMeasureInfo())
5993 * or NULL on error
5994 *
5995 * Calculate file size required to create a new image.
5996 *
5997 * If @in_bs is given then space for allocated clusters and zero clusters
5998 * from that image are included in the calculation. If @opts contains a
5999 * backing file that is shared by @in_bs then backing clusters may be omitted
6000 * from the calculation.
6001 *
6002 * If @in_bs is NULL then the calculation includes no allocated clusters
6003 * unless a preallocation option is given in @opts.
6004 *
6005 * Note that @in_bs may use a different BlockDriver from @drv.
6006 *
6007 * If an error occurs the @errp pointer is set.
6008 */
6009BlockMeasureInfo *bdrv_measure(BlockDriver *drv, QemuOpts *opts,
6010 BlockDriverState *in_bs, Error **errp)
6011{
Emanuele Giuseppe Esposito384a48f2022-03-03 10:15:50 -05006012 IO_CODE();
Stefan Hajnoczi90880ff2017-07-05 13:57:30 +01006013 if (!drv->bdrv_measure) {
6014 error_setg(errp, "Block driver '%s' does not support size measurement",
6015 drv->format_name);
6016 return NULL;
6017 }
6018
6019 return drv->bdrv_measure(opts, in_bs, errp);
6020}
6021
Fam Zheng4a1d5e12011-07-12 19:56:39 +08006022/**
Markus Armbruster65a9bb22014-06-26 13:23:17 +02006023 * Return number of sectors on success, -errno on error.
bellard83f64092006-08-01 16:21:11 +00006024 */
Emanuele Giuseppe Espositoc86422c2023-01-13 21:42:04 +01006025int64_t coroutine_fn bdrv_co_nb_sectors(BlockDriverState *bs)
bellard83f64092006-08-01 16:21:11 +00006026{
6027 BlockDriver *drv = bs->drv;
Emanuele Giuseppe Esposito384a48f2022-03-03 10:15:50 -05006028 IO_CODE();
Kevin Wolf8ab81402023-02-03 16:22:02 +01006029 assert_bdrv_graph_readable();
Markus Armbruster65a9bb22014-06-26 13:23:17 +02006030
bellard83f64092006-08-01 16:21:11 +00006031 if (!drv)
bellard19cb3732006-08-19 11:45:59 +00006032 return -ENOMEDIUM;
Stefan Hajnoczi51762282010-04-19 16:56:41 +01006033
Paolo Bonzini160a29e2023-04-07 17:32:56 +02006034 if (bs->bl.has_variable_length) {
Emanuele Giuseppe Espositoc86422c2023-01-13 21:42:04 +01006035 int ret = bdrv_co_refresh_total_sectors(bs, bs->total_sectors);
Kevin Wolfb94a2612013-10-29 12:18:58 +01006036 if (ret < 0) {
6037 return ret;
Stefan Hajnoczi46a4e4e2011-03-29 20:04:41 +01006038 }
bellard83f64092006-08-01 16:21:11 +00006039 }
Markus Armbruster65a9bb22014-06-26 13:23:17 +02006040 return bs->total_sectors;
6041}
6042
Paolo Bonzini81f730d2023-04-07 17:33:03 +02006043/*
6044 * This wrapper is written by hand because this function is in the hot I/O path,
6045 * via blk_get_geometry.
6046 */
6047int64_t coroutine_mixed_fn bdrv_nb_sectors(BlockDriverState *bs)
6048{
6049 BlockDriver *drv = bs->drv;
6050 IO_CODE();
6051
6052 if (!drv)
6053 return -ENOMEDIUM;
6054
6055 if (bs->bl.has_variable_length) {
6056 int ret = bdrv_refresh_total_sectors(bs, bs->total_sectors);
6057 if (ret < 0) {
6058 return ret;
6059 }
6060 }
6061
6062 return bs->total_sectors;
6063}
6064
Markus Armbruster65a9bb22014-06-26 13:23:17 +02006065/**
6066 * Return length in bytes on success, -errno on error.
6067 * The length is always a multiple of BDRV_SECTOR_SIZE.
6068 */
Emanuele Giuseppe Espositoc86422c2023-01-13 21:42:04 +01006069int64_t coroutine_fn bdrv_co_getlength(BlockDriverState *bs)
Markus Armbruster65a9bb22014-06-26 13:23:17 +02006070{
Emanuele Giuseppe Espositoc86422c2023-01-13 21:42:04 +01006071 int64_t ret;
Emanuele Giuseppe Esposito384a48f2022-03-03 10:15:50 -05006072 IO_CODE();
Kevin Wolf8ab81402023-02-03 16:22:02 +01006073 assert_bdrv_graph_readable();
Markus Armbruster65a9bb22014-06-26 13:23:17 +02006074
Emanuele Giuseppe Espositoc86422c2023-01-13 21:42:04 +01006075 ret = bdrv_co_nb_sectors(bs);
Eric Blake122860b2020-11-05 09:51:22 -06006076 if (ret < 0) {
6077 return ret;
6078 }
6079 if (ret > INT64_MAX / BDRV_SECTOR_SIZE) {
6080 return -EFBIG;
6081 }
6082 return ret * BDRV_SECTOR_SIZE;
bellardfc01f7e2003-06-30 10:03:06 +00006083}
6084
Eric Blake54115412016-06-23 16:37:26 -06006085bool bdrv_is_sg(BlockDriverState *bs)
ths985a03b2007-12-24 16:10:43 +00006086{
Emanuele Giuseppe Esposito384a48f2022-03-03 10:15:50 -05006087 IO_CODE();
ths985a03b2007-12-24 16:10:43 +00006088 return bs->sg;
6089}
6090
Max Reitzae23f782019-06-12 22:57:15 +02006091/**
6092 * Return whether the given node supports compressed writes.
6093 */
6094bool bdrv_supports_compressed_writes(BlockDriverState *bs)
6095{
6096 BlockDriverState *filtered;
Emanuele Giuseppe Esposito384a48f2022-03-03 10:15:50 -05006097 IO_CODE();
Max Reitzae23f782019-06-12 22:57:15 +02006098
6099 if (!bs->drv || !block_driver_can_compress(bs->drv)) {
6100 return false;
6101 }
6102
6103 filtered = bdrv_filter_bs(bs);
6104 if (filtered) {
6105 /*
6106 * Filters can only forward compressed writes, so we have to
6107 * check the child.
6108 */
6109 return bdrv_supports_compressed_writes(filtered);
6110 }
6111
6112 return true;
6113}
6114
Markus Armbrusterf8d6bba2012-06-13 10:11:48 +02006115const char *bdrv_get_format_name(BlockDriverState *bs)
bellardea2384d2004-08-01 21:59:26 +00006116{
Emanuele Giuseppe Esposito384a48f2022-03-03 10:15:50 -05006117 IO_CODE();
Markus Armbrusterf8d6bba2012-06-13 10:11:48 +02006118 return bs->drv ? bs->drv->format_name : NULL;
bellardea2384d2004-08-01 21:59:26 +00006119}
6120
Stefan Hajnocziada42402014-08-27 12:08:55 +01006121static int qsort_strcmp(const void *a, const void *b)
6122{
Max Reitzceff5bd2016-10-12 22:49:05 +02006123 return strcmp(*(char *const *)a, *(char *const *)b);
Stefan Hajnocziada42402014-08-27 12:08:55 +01006124}
6125
ths5fafdf22007-09-16 21:08:06 +00006126void bdrv_iterate_format(void (*it)(void *opaque, const char *name),
Andrey Shinkevich9ac404c2019-03-07 16:33:58 +03006127 void *opaque, bool read_only)
bellardea2384d2004-08-01 21:59:26 +00006128{
6129 BlockDriver *drv;
Jeff Codye855e4f2014-04-28 18:29:54 -04006130 int count = 0;
Stefan Hajnocziada42402014-08-27 12:08:55 +01006131 int i;
Jeff Codye855e4f2014-04-28 18:29:54 -04006132 const char **formats = NULL;
bellardea2384d2004-08-01 21:59:26 +00006133
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05006134 GLOBAL_STATE_CODE();
6135
Stefan Hajnoczi8a22f022010-04-13 10:29:33 +01006136 QLIST_FOREACH(drv, &bdrv_drivers, list) {
Jeff Codye855e4f2014-04-28 18:29:54 -04006137 if (drv->format_name) {
6138 bool found = false;
Andrey Shinkevich9ac404c2019-03-07 16:33:58 +03006139
6140 if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv, read_only)) {
6141 continue;
6142 }
6143
Markus Armbrusterfb2575f2023-09-21 14:13:11 +02006144 i = count;
Jeff Codye855e4f2014-04-28 18:29:54 -04006145 while (formats && i && !found) {
6146 found = !strcmp(formats[--i], drv->format_name);
6147 }
6148
6149 if (!found) {
Markus Armbruster5839e532014-08-19 10:31:08 +02006150 formats = g_renew(const char *, formats, count + 1);
Jeff Codye855e4f2014-04-28 18:29:54 -04006151 formats[count++] = drv->format_name;
Jeff Codye855e4f2014-04-28 18:29:54 -04006152 }
6153 }
bellardea2384d2004-08-01 21:59:26 +00006154 }
Stefan Hajnocziada42402014-08-27 12:08:55 +01006155
Max Reitzeb0df692016-10-12 22:49:06 +02006156 for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); i++) {
6157 const char *format_name = block_driver_modules[i].format_name;
6158
6159 if (format_name) {
6160 bool found = false;
6161 int j = count;
6162
Andrey Shinkevich9ac404c2019-03-07 16:33:58 +03006163 if (use_bdrv_whitelist &&
6164 !bdrv_format_is_whitelisted(format_name, read_only)) {
6165 continue;
6166 }
6167
Max Reitzeb0df692016-10-12 22:49:06 +02006168 while (formats && j && !found) {
6169 found = !strcmp(formats[--j], format_name);
6170 }
6171
6172 if (!found) {
6173 formats = g_renew(const char *, formats, count + 1);
6174 formats[count++] = format_name;
6175 }
6176 }
6177 }
6178
Stefan Hajnocziada42402014-08-27 12:08:55 +01006179 qsort(formats, count, sizeof(formats[0]), qsort_strcmp);
6180
6181 for (i = 0; i < count; i++) {
6182 it(opaque, formats[i]);
6183 }
6184
Jeff Codye855e4f2014-04-28 18:29:54 -04006185 g_free(formats);
bellardea2384d2004-08-01 21:59:26 +00006186}
6187
Benoît Canetdc364f42014-01-23 21:31:32 +01006188/* This function is to find a node in the bs graph */
6189BlockDriverState *bdrv_find_node(const char *node_name)
6190{
6191 BlockDriverState *bs;
6192
6193 assert(node_name);
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05006194 GLOBAL_STATE_CODE();
Benoît Canetdc364f42014-01-23 21:31:32 +01006195
6196 QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
6197 if (!strcmp(node_name, bs->node_name)) {
6198 return bs;
6199 }
6200 }
6201 return NULL;
6202}
6203
Benoît Canetc13163f2014-01-23 21:31:34 +01006204/* Put this QMP function here so it can access the static graph_bdrv_states. */
Peter Krempafacda542020-01-20 09:50:49 +01006205BlockDeviceInfoList *bdrv_named_nodes_list(bool flat,
6206 Error **errp)
Benoît Canetc13163f2014-01-23 21:31:34 +01006207{
Eric Blake9812e712020-10-27 00:05:47 -05006208 BlockDeviceInfoList *list;
Benoît Canetc13163f2014-01-23 21:31:34 +01006209 BlockDriverState *bs;
6210
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05006211 GLOBAL_STATE_CODE();
Kevin Wolfb7cfc7d2023-09-29 16:51:45 +02006212 GRAPH_RDLOCK_GUARD_MAINLOOP();
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05006213
Benoît Canetc13163f2014-01-23 21:31:34 +01006214 list = NULL;
6215 QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
Peter Krempafacda542020-01-20 09:50:49 +01006216 BlockDeviceInfo *info = bdrv_block_device_info(NULL, bs, flat, errp);
Alberto Garciad5a8ee62015-04-17 14:52:43 +03006217 if (!info) {
6218 qapi_free_BlockDeviceInfoList(list);
6219 return NULL;
6220 }
Eric Blake9812e712020-10-27 00:05:47 -05006221 QAPI_LIST_PREPEND(list, info);
Benoît Canetc13163f2014-01-23 21:31:34 +01006222 }
6223
6224 return list;
6225}
6226
Vladimir Sementsov-Ogievskiy5d3b4e92018-12-21 20:09:07 +03006227typedef struct XDbgBlockGraphConstructor {
6228 XDbgBlockGraph *graph;
6229 GHashTable *graph_nodes;
6230} XDbgBlockGraphConstructor;
6231
6232static XDbgBlockGraphConstructor *xdbg_graph_new(void)
6233{
6234 XDbgBlockGraphConstructor *gr = g_new(XDbgBlockGraphConstructor, 1);
6235
6236 gr->graph = g_new0(XDbgBlockGraph, 1);
6237 gr->graph_nodes = g_hash_table_new(NULL, NULL);
6238
6239 return gr;
6240}
6241
6242static XDbgBlockGraph *xdbg_graph_finalize(XDbgBlockGraphConstructor *gr)
6243{
6244 XDbgBlockGraph *graph = gr->graph;
6245
6246 g_hash_table_destroy(gr->graph_nodes);
6247 g_free(gr);
6248
6249 return graph;
6250}
6251
6252static uintptr_t xdbg_graph_node_num(XDbgBlockGraphConstructor *gr, void *node)
6253{
6254 uintptr_t ret = (uintptr_t)g_hash_table_lookup(gr->graph_nodes, node);
6255
6256 if (ret != 0) {
6257 return ret;
6258 }
6259
6260 /*
6261 * Start counting from 1, not 0, because 0 interferes with not-found (NULL)
6262 * answer of g_hash_table_lookup.
6263 */
6264 ret = g_hash_table_size(gr->graph_nodes) + 1;
6265 g_hash_table_insert(gr->graph_nodes, node, (void *)ret);
6266
6267 return ret;
6268}
6269
6270static void xdbg_graph_add_node(XDbgBlockGraphConstructor *gr, void *node,
6271 XDbgBlockGraphNodeType type, const char *name)
6272{
6273 XDbgBlockGraphNode *n;
6274
6275 n = g_new0(XDbgBlockGraphNode, 1);
6276
6277 n->id = xdbg_graph_node_num(gr, node);
6278 n->type = type;
6279 n->name = g_strdup(name);
6280
Eric Blake9812e712020-10-27 00:05:47 -05006281 QAPI_LIST_PREPEND(gr->graph->nodes, n);
Vladimir Sementsov-Ogievskiy5d3b4e92018-12-21 20:09:07 +03006282}
6283
6284static void xdbg_graph_add_edge(XDbgBlockGraphConstructor *gr, void *parent,
6285 const BdrvChild *child)
6286{
Max Reitzcdb1cec2019-11-08 13:34:52 +01006287 BlockPermission qapi_perm;
Vladimir Sementsov-Ogievskiy5d3b4e92018-12-21 20:09:07 +03006288 XDbgBlockGraphEdge *edge;
Emanuele Giuseppe Esposito862fded2022-03-03 10:15:55 -05006289 GLOBAL_STATE_CODE();
Vladimir Sementsov-Ogievskiy5d3b4e92018-12-21 20:09:07 +03006290
Vladimir Sementsov-Ogievskiy5d3b4e92018-12-21 20:09:07 +03006291 edge = g_new0(XDbgBlockGraphEdge, 1);
6292
6293 edge->parent = xdbg_graph_node_num(gr, parent);
6294 edge->child = xdbg_graph_node_num(gr, child->bs);
6295 edge->name = g_strdup(child->name);
6296
Max Reitzcdb1cec2019-11-08 13:34:52 +01006297 for (qapi_perm = 0; qapi_perm < BLOCK_PERMISSION__MAX; qapi_perm++) {
6298 uint64_t flag = bdrv_qapi_perm_to_blk_perm(qapi_perm);
6299
6300 if (flag & child->perm) {
Eric Blake9812e712020-10-27 00:05:47 -05006301 QAPI_LIST_PREPEND(edge->perm, qapi_perm);
Vladimir Sementsov-Ogievskiy5d3b4e92018-12-21 20:09:07 +03006302 }
Max Reitzcdb1cec2019-11-08 13:34:52 +01006303 if (flag & child->shared_perm) {
Eric Blake9812e712020-10-27 00:05:47 -05006304 QAPI_LIST_PREPEND(edge->shared_perm, qapi_perm);
Vladimir Sementsov-Ogievskiy5d3b4e92018-12-21 20:09:07 +03006305 }
6306 }
6307
Eric Blake9812e712020-10-27 00:05:47 -05006308 QAPI_LIST_PREPEND(gr->graph->edges, edge);
Vladimir Sementsov-Ogievskiy5d3b4e92018-12-21 20:09:07 +03006309}
6310
6311
6312XDbgBlockGraph *bdrv_get_xdbg_block_graph(Error **errp)
6313{
6314 BlockBackend *blk;
6315 BlockJob *job;
6316 BlockDriverState *bs;
6317 BdrvChild *child;
6318 XDbgBlockGraphConstructor *gr = xdbg_graph_new();
6319
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05006320 GLOBAL_STATE_CODE();
6321
Vladimir Sementsov-Ogievskiy5d3b4e92018-12-21 20:09:07 +03006322 for (blk = blk_all_next(NULL); blk; blk = blk_all_next(blk)) {
6323 char *allocated_name = NULL;
6324 const char *name = blk_name(blk);
6325
6326 if (!*name) {
6327 name = allocated_name = blk_get_attached_dev_id(blk);
6328 }
6329 xdbg_graph_add_node(gr, blk, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_BACKEND,
6330 name);
6331 g_free(allocated_name);
6332 if (blk_root(blk)) {
6333 xdbg_graph_add_edge(gr, blk, blk_root(blk));
6334 }
6335 }
6336
Emanuele Giuseppe Esposito880eeec2022-09-26 05:32:04 -04006337 WITH_JOB_LOCK_GUARD() {
6338 for (job = block_job_next_locked(NULL); job;
6339 job = block_job_next_locked(job)) {
6340 GSList *el;
Vladimir Sementsov-Ogievskiy5d3b4e92018-12-21 20:09:07 +03006341
Emanuele Giuseppe Esposito880eeec2022-09-26 05:32:04 -04006342 xdbg_graph_add_node(gr, job, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_JOB,
6343 job->job.id);
6344 for (el = job->nodes; el; el = el->next) {
6345 xdbg_graph_add_edge(gr, job, (BdrvChild *)el->data);
6346 }
Vladimir Sementsov-Ogievskiy5d3b4e92018-12-21 20:09:07 +03006347 }
6348 }
6349
6350 QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
6351 xdbg_graph_add_node(gr, bs, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_DRIVER,
6352 bs->node_name);
6353 QLIST_FOREACH(child, &bs->children, next) {
6354 xdbg_graph_add_edge(gr, bs, child);
6355 }
6356 }
6357
6358 return xdbg_graph_finalize(gr);
6359}
6360
Benoît Canet12d3ba82014-01-23 21:31:35 +01006361BlockDriverState *bdrv_lookup_bs(const char *device,
6362 const char *node_name,
6363 Error **errp)
6364{
Markus Armbruster7f06d472014-10-07 13:59:12 +02006365 BlockBackend *blk;
6366 BlockDriverState *bs;
Benoît Canet12d3ba82014-01-23 21:31:35 +01006367
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05006368 GLOBAL_STATE_CODE();
6369
Benoît Canet12d3ba82014-01-23 21:31:35 +01006370 if (device) {
Markus Armbruster7f06d472014-10-07 13:59:12 +02006371 blk = blk_by_name(device);
Benoît Canet12d3ba82014-01-23 21:31:35 +01006372
Markus Armbruster7f06d472014-10-07 13:59:12 +02006373 if (blk) {
Alberto Garcia9f4ed6f2015-10-26 16:46:49 +02006374 bs = blk_bs(blk);
6375 if (!bs) {
Max Reitz5433c242015-10-19 17:53:29 +02006376 error_setg(errp, "Device '%s' has no medium", device);
Max Reitz5433c242015-10-19 17:53:29 +02006377 }
6378
Alberto Garcia9f4ed6f2015-10-26 16:46:49 +02006379 return bs;
Benoît Canet12d3ba82014-01-23 21:31:35 +01006380 }
Benoît Canet12d3ba82014-01-23 21:31:35 +01006381 }
6382
Benoît Canetdd67fa52014-02-12 17:15:06 +01006383 if (node_name) {
6384 bs = bdrv_find_node(node_name);
Benoît Canet12d3ba82014-01-23 21:31:35 +01006385
Benoît Canetdd67fa52014-02-12 17:15:06 +01006386 if (bs) {
6387 return bs;
6388 }
Benoît Canet12d3ba82014-01-23 21:31:35 +01006389 }
6390
Connor Kuehl785ec4b2021-03-05 09:19:28 -06006391 error_setg(errp, "Cannot find device=\'%s\' nor node-name=\'%s\'",
Benoît Canetdd67fa52014-02-12 17:15:06 +01006392 device ? device : "",
6393 node_name ? node_name : "");
6394 return NULL;
Benoît Canet12d3ba82014-01-23 21:31:35 +01006395}
6396
Jeff Cody5a6684d2014-06-25 15:40:09 -04006397/* If 'base' is in the same chain as 'top', return true. Otherwise,
6398 * return false. If either argument is NULL, return false. */
6399bool bdrv_chain_contains(BlockDriverState *top, BlockDriverState *base)
6400{
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05006401
6402 GLOBAL_STATE_CODE();
6403
Jeff Cody5a6684d2014-06-25 15:40:09 -04006404 while (top && top != base) {
Max Reitzdcf3f9b2019-06-12 17:34:45 +02006405 top = bdrv_filter_or_cow_bs(top);
Jeff Cody5a6684d2014-06-25 15:40:09 -04006406 }
6407
6408 return top != NULL;
6409}
6410
Fam Zheng04df7652014-10-31 11:32:54 +08006411BlockDriverState *bdrv_next_node(BlockDriverState *bs)
6412{
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05006413 GLOBAL_STATE_CODE();
Fam Zheng04df7652014-10-31 11:32:54 +08006414 if (!bs) {
6415 return QTAILQ_FIRST(&graph_bdrv_states);
6416 }
6417 return QTAILQ_NEXT(bs, node_list);
6418}
6419
Kevin Wolf0f122642018-03-28 18:29:18 +02006420BlockDriverState *bdrv_next_all_states(BlockDriverState *bs)
6421{
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05006422 GLOBAL_STATE_CODE();
Kevin Wolf0f122642018-03-28 18:29:18 +02006423 if (!bs) {
6424 return QTAILQ_FIRST(&all_bdrv_states);
6425 }
6426 return QTAILQ_NEXT(bs, bs_list);
6427}
6428
Fam Zheng20a9e772014-10-31 11:32:55 +08006429const char *bdrv_get_node_name(const BlockDriverState *bs)
6430{
Emanuele Giuseppe Esposito384a48f2022-03-03 10:15:50 -05006431 IO_CODE();
Fam Zheng20a9e772014-10-31 11:32:55 +08006432 return bs->node_name;
6433}
6434
Kevin Wolf1f0c4612016-03-22 18:38:44 +01006435const char *bdrv_get_parent_name(const BlockDriverState *bs)
Kevin Wolf4c265bf2016-02-26 10:22:16 +01006436{
6437 BdrvChild *c;
6438 const char *name;
Emanuele Giuseppe Esposito967d7902022-03-03 10:15:58 -05006439 IO_CODE();
Kevin Wolf4c265bf2016-02-26 10:22:16 +01006440
6441 /* If multiple parents have a name, just pick the first one. */
6442 QLIST_FOREACH(c, &bs->parents, next_parent) {
Max Reitzbd86fb92020-05-13 13:05:13 +02006443 if (c->klass->get_name) {
6444 name = c->klass->get_name(c);
Kevin Wolf4c265bf2016-02-26 10:22:16 +01006445 if (name && *name) {
6446 return name;
6447 }
6448 }
6449 }
6450
6451 return NULL;
6452}
6453
Markus Armbruster7f06d472014-10-07 13:59:12 +02006454/* TODO check what callers really want: bs->node_name or blk_name() */
Markus Armbrusterbfb197e2014-10-07 13:59:11 +02006455const char *bdrv_get_device_name(const BlockDriverState *bs)
bellardea2384d2004-08-01 21:59:26 +00006456{
Emanuele Giuseppe Esposito384a48f2022-03-03 10:15:50 -05006457 IO_CODE();
Kevin Wolf4c265bf2016-02-26 10:22:16 +01006458 return bdrv_get_parent_name(bs) ?: "";
bellardea2384d2004-08-01 21:59:26 +00006459}
6460
Alberto Garcia9b2aa842015-04-08 12:29:18 +03006461/* This can be used to identify nodes that might not have a device
6462 * name associated. Since node and device names live in the same
6463 * namespace, the result is unambiguous. The exception is if both are
6464 * absent, then this returns an empty (non-null) string. */
6465const char *bdrv_get_device_or_node_name(const BlockDriverState *bs)
6466{
Emanuele Giuseppe Esposito384a48f2022-03-03 10:15:50 -05006467 IO_CODE();
Kevin Wolf4c265bf2016-02-26 10:22:16 +01006468 return bdrv_get_parent_name(bs) ?: bs->node_name;
Alberto Garcia9b2aa842015-04-08 12:29:18 +03006469}
6470
Markus Armbrusterc8433282012-06-05 16:49:24 +02006471int bdrv_get_flags(BlockDriverState *bs)
6472{
Hanna Reitz15aee7a2022-04-27 13:40:54 +02006473 IO_CODE();
Markus Armbrusterc8433282012-06-05 16:49:24 +02006474 return bs->open_flags;
6475}
6476
Peter Lieven3ac21622013-06-28 12:47:42 +02006477int bdrv_has_zero_init_1(BlockDriverState *bs)
6478{
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05006479 GLOBAL_STATE_CODE();
Peter Lieven3ac21622013-06-28 12:47:42 +02006480 return 1;
6481}
6482
Kevin Wolf06717982023-10-27 17:53:11 +02006483int coroutine_mixed_fn bdrv_has_zero_init(BlockDriverState *bs)
Kevin Wolff2feebb2010-04-14 17:30:35 +02006484{
Max Reitz93393e62019-06-12 17:03:38 +02006485 BlockDriverState *filtered;
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05006486 GLOBAL_STATE_CODE();
Max Reitz93393e62019-06-12 17:03:38 +02006487
Max Reitzd470ad42017-11-10 21:31:09 +01006488 if (!bs->drv) {
6489 return 0;
6490 }
Kevin Wolff2feebb2010-04-14 17:30:35 +02006491
Paolo Bonzini11212d82013-09-04 19:00:27 +02006492 /* If BS is a copy on write image, it is initialized to
6493 the contents of the base image, which may not be zeroes. */
Max Reitz34778172019-06-12 17:10:46 +02006494 if (bdrv_cow_child(bs)) {
Paolo Bonzini11212d82013-09-04 19:00:27 +02006495 return 0;
6496 }
Kevin Wolf336c1c12010-07-28 11:26:29 +02006497 if (bs->drv->bdrv_has_zero_init) {
6498 return bs->drv->bdrv_has_zero_init(bs);
Kevin Wolff2feebb2010-04-14 17:30:35 +02006499 }
Max Reitz93393e62019-06-12 17:03:38 +02006500
6501 filtered = bdrv_filter_bs(bs);
6502 if (filtered) {
6503 return bdrv_has_zero_init(filtered);
Manos Pitsidianakis5a612c02017-07-13 18:30:25 +03006504 }
Kevin Wolff2feebb2010-04-14 17:30:35 +02006505
Peter Lieven3ac21622013-06-28 12:47:42 +02006506 /* safe default */
6507 return 0;
Kevin Wolff2feebb2010-04-14 17:30:35 +02006508}
6509
Peter Lieven4ce78692013-10-24 12:06:54 +02006510bool bdrv_can_write_zeroes_with_unmap(BlockDriverState *bs)
6511{
Emanuele Giuseppe Esposito384a48f2022-03-03 10:15:50 -05006512 IO_CODE();
Denis V. Lunev2f0342e2016-07-14 16:33:26 +03006513 if (!(bs->open_flags & BDRV_O_UNMAP)) {
Peter Lieven4ce78692013-10-24 12:06:54 +02006514 return false;
6515 }
6516
Eric Blakee24d8132018-01-26 13:34:39 -06006517 return bs->supported_zero_flags & BDRV_REQ_MAY_UNMAP;
Peter Lieven4ce78692013-10-24 12:06:54 +02006518}
6519
ths5fafdf22007-09-16 21:08:06 +00006520void bdrv_get_backing_filename(BlockDriverState *bs,
bellard83f64092006-08-01 16:21:11 +00006521 char *filename, int filename_size)
bellardea2384d2004-08-01 21:59:26 +00006522{
Emanuele Giuseppe Esposito384a48f2022-03-03 10:15:50 -05006523 IO_CODE();
Kevin Wolf3574c602011-10-26 11:02:11 +02006524 pstrcpy(filename, filename_size, bs->backing_file);
bellardea2384d2004-08-01 21:59:26 +00006525}
6526
Emanuele Giuseppe Esposito3d47eb02023-01-13 21:42:08 +01006527int coroutine_fn bdrv_co_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
bellardfaea38e2006-08-05 21:31:00 +00006528{
Vladimir Sementsov-Ogievskiy8b117002020-12-04 01:27:13 +03006529 int ret;
bellardfaea38e2006-08-05 21:31:00 +00006530 BlockDriver *drv = bs->drv;
Emanuele Giuseppe Esposito384a48f2022-03-03 10:15:50 -05006531 IO_CODE();
Emanuele Giuseppe Espositoa00e70c2023-05-04 13:57:44 +02006532 assert_bdrv_graph_readable();
6533
Manos Pitsidianakis5a612c02017-07-13 18:30:25 +03006534 /* if bs->drv == NULL, bs is closed, so there's nothing to do here */
6535 if (!drv) {
bellard19cb3732006-08-19 11:45:59 +00006536 return -ENOMEDIUM;
Manos Pitsidianakis5a612c02017-07-13 18:30:25 +03006537 }
Emanuele Giuseppe Esposito3d47eb02023-01-13 21:42:08 +01006538 if (!drv->bdrv_co_get_info) {
Max Reitz93393e62019-06-12 17:03:38 +02006539 BlockDriverState *filtered = bdrv_filter_bs(bs);
6540 if (filtered) {
Emanuele Giuseppe Esposito3d47eb02023-01-13 21:42:08 +01006541 return bdrv_co_get_info(filtered, bdi);
Manos Pitsidianakis5a612c02017-07-13 18:30:25 +03006542 }
bellardfaea38e2006-08-05 21:31:00 +00006543 return -ENOTSUP;
Manos Pitsidianakis5a612c02017-07-13 18:30:25 +03006544 }
bellardfaea38e2006-08-05 21:31:00 +00006545 memset(bdi, 0, sizeof(*bdi));
Emanuele Giuseppe Esposito3d47eb02023-01-13 21:42:08 +01006546 ret = drv->bdrv_co_get_info(bs, bdi);
Andrey Drobyshevc54483b2023-07-11 20:25:51 +03006547 if (bdi->subcluster_size == 0) {
6548 /*
6549 * If the driver left this unset, subclusters are not supported.
6550 * Then it is safe to treat each cluster as having only one subcluster.
6551 */
6552 bdi->subcluster_size = bdi->cluster_size;
6553 }
Vladimir Sementsov-Ogievskiy8b117002020-12-04 01:27:13 +03006554 if (ret < 0) {
6555 return ret;
6556 }
6557
6558 if (bdi->cluster_size > BDRV_MAX_ALIGNMENT) {
6559 return -EINVAL;
6560 }
6561
6562 return 0;
bellardfaea38e2006-08-05 21:31:00 +00006563}
6564
Andrey Shinkevich1bf6e9c2019-02-08 18:06:06 +03006565ImageInfoSpecific *bdrv_get_specific_info(BlockDriverState *bs,
6566 Error **errp)
Max Reitzeae041f2013-10-09 10:46:16 +02006567{
6568 BlockDriver *drv = bs->drv;
Emanuele Giuseppe Esposito384a48f2022-03-03 10:15:50 -05006569 IO_CODE();
Max Reitzeae041f2013-10-09 10:46:16 +02006570 if (drv && drv->bdrv_get_specific_info) {
Andrey Shinkevich1bf6e9c2019-02-08 18:06:06 +03006571 return drv->bdrv_get_specific_info(bs, errp);
Max Reitzeae041f2013-10-09 10:46:16 +02006572 }
6573 return NULL;
6574}
6575
Anton Nefedovd9245592019-09-23 15:17:37 +03006576BlockStatsSpecific *bdrv_get_specific_stats(BlockDriverState *bs)
6577{
6578 BlockDriver *drv = bs->drv;
Emanuele Giuseppe Esposito384a48f2022-03-03 10:15:50 -05006579 IO_CODE();
Anton Nefedovd9245592019-09-23 15:17:37 +03006580 if (!drv || !drv->bdrv_get_specific_stats) {
6581 return NULL;
6582 }
6583 return drv->bdrv_get_specific_stats(bs);
6584}
6585
Emanuele Giuseppe Espositoc834dc02023-01-13 21:42:11 +01006586void coroutine_fn bdrv_co_debug_event(BlockDriverState *bs, BlkdebugEvent event)
Kevin Wolf8b9b0cc2010-03-15 17:27:00 +01006587{
Emanuele Giuseppe Esposito384a48f2022-03-03 10:15:50 -05006588 IO_CODE();
Emanuele Giuseppe Espositocb2bfaa2023-05-04 13:57:45 +02006589 assert_bdrv_graph_readable();
6590
Emanuele Giuseppe Espositoc834dc02023-01-13 21:42:11 +01006591 if (!bs || !bs->drv || !bs->drv->bdrv_co_debug_event) {
Kevin Wolf8b9b0cc2010-03-15 17:27:00 +01006592 return;
6593 }
6594
Emanuele Giuseppe Espositoc834dc02023-01-13 21:42:11 +01006595 bs->drv->bdrv_co_debug_event(bs, event);
Kevin Wolf41c695c2012-12-06 14:32:58 +01006596}
Kevin Wolf8b9b0cc2010-03-15 17:27:00 +01006597
Kevin Wolfc0fc5122023-09-29 16:51:46 +02006598static BlockDriverState * GRAPH_RDLOCK
6599bdrv_find_debug_node(BlockDriverState *bs)
Kevin Wolf41c695c2012-12-06 14:32:58 +01006600{
Emanuele Giuseppe Espositobdb73472022-03-03 10:16:02 -05006601 GLOBAL_STATE_CODE();
Kevin Wolf41c695c2012-12-06 14:32:58 +01006602 while (bs && bs->drv && !bs->drv->bdrv_debug_breakpoint) {
Max Reitzf706a922019-06-12 17:42:13 +02006603 bs = bdrv_primary_bs(bs);
Kevin Wolf41c695c2012-12-06 14:32:58 +01006604 }
6605
6606 if (bs && bs->drv && bs->drv->bdrv_debug_breakpoint) {
Vladimir Sementsov-Ogievskiyd10529a2019-09-20 17:20:49 +03006607 assert(bs->drv->bdrv_debug_remove_breakpoint);
6608 return bs;
6609 }
6610
6611 return NULL;
6612}
6613
6614int bdrv_debug_breakpoint(BlockDriverState *bs, const char *event,
6615 const char *tag)
6616{
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05006617 GLOBAL_STATE_CODE();
Kevin Wolfc0fc5122023-09-29 16:51:46 +02006618 GRAPH_RDLOCK_GUARD_MAINLOOP();
6619
Vladimir Sementsov-Ogievskiyd10529a2019-09-20 17:20:49 +03006620 bs = bdrv_find_debug_node(bs);
6621 if (bs) {
Kevin Wolf41c695c2012-12-06 14:32:58 +01006622 return bs->drv->bdrv_debug_breakpoint(bs, event, tag);
6623 }
6624
6625 return -ENOTSUP;
6626}
6627
Fam Zheng4cc70e92013-11-20 10:01:54 +08006628int bdrv_debug_remove_breakpoint(BlockDriverState *bs, const char *tag)
6629{
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05006630 GLOBAL_STATE_CODE();
Kevin Wolfc0fc5122023-09-29 16:51:46 +02006631 GRAPH_RDLOCK_GUARD_MAINLOOP();
6632
Vladimir Sementsov-Ogievskiyd10529a2019-09-20 17:20:49 +03006633 bs = bdrv_find_debug_node(bs);
6634 if (bs) {
Fam Zheng4cc70e92013-11-20 10:01:54 +08006635 return bs->drv->bdrv_debug_remove_breakpoint(bs, tag);
6636 }
6637
6638 return -ENOTSUP;
6639}
6640
Kevin Wolf41c695c2012-12-06 14:32:58 +01006641int bdrv_debug_resume(BlockDriverState *bs, const char *tag)
6642{
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05006643 GLOBAL_STATE_CODE();
Kevin Wolfc0fc5122023-09-29 16:51:46 +02006644 GRAPH_RDLOCK_GUARD_MAINLOOP();
6645
Max Reitz938789e2014-03-10 23:44:08 +01006646 while (bs && (!bs->drv || !bs->drv->bdrv_debug_resume)) {
Max Reitzf706a922019-06-12 17:42:13 +02006647 bs = bdrv_primary_bs(bs);
Kevin Wolf41c695c2012-12-06 14:32:58 +01006648 }
6649
6650 if (bs && bs->drv && bs->drv->bdrv_debug_resume) {
6651 return bs->drv->bdrv_debug_resume(bs, tag);
6652 }
6653
6654 return -ENOTSUP;
6655}
6656
6657bool bdrv_debug_is_suspended(BlockDriverState *bs, const char *tag)
6658{
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05006659 GLOBAL_STATE_CODE();
Kevin Wolfc0fc5122023-09-29 16:51:46 +02006660 GRAPH_RDLOCK_GUARD_MAINLOOP();
6661
Kevin Wolf41c695c2012-12-06 14:32:58 +01006662 while (bs && bs->drv && !bs->drv->bdrv_debug_is_suspended) {
Max Reitzf706a922019-06-12 17:42:13 +02006663 bs = bdrv_primary_bs(bs);
Kevin Wolf41c695c2012-12-06 14:32:58 +01006664 }
6665
6666 if (bs && bs->drv && bs->drv->bdrv_debug_is_suspended) {
6667 return bs->drv->bdrv_debug_is_suspended(bs, tag);
6668 }
6669
6670 return false;
Kevin Wolf8b9b0cc2010-03-15 17:27:00 +01006671}
6672
Jeff Codyb1b1d782012-10-16 15:49:09 -04006673/* backing_file can either be relative, or absolute, or a protocol. If it is
6674 * relative, it must be relative to the chain. So, passing in bs->filename
6675 * from a BDS as backing_file should not be done, as that may be relative to
6676 * the CWD rather than the chain. */
Marcelo Tosattie8a6bb92012-01-18 14:40:51 +00006677BlockDriverState *bdrv_find_backing_image(BlockDriverState *bs,
6678 const char *backing_file)
6679{
Jeff Codyb1b1d782012-10-16 15:49:09 -04006680 char *filename_full = NULL;
6681 char *backing_file_full = NULL;
6682 char *filename_tmp = NULL;
6683 int is_protocol = 0;
Max Reitz0b877d02018-08-01 20:34:11 +02006684 bool filenames_refreshed = false;
Jeff Codyb1b1d782012-10-16 15:49:09 -04006685 BlockDriverState *curr_bs = NULL;
6686 BlockDriverState *retval = NULL;
Max Reitzdcf3f9b2019-06-12 17:34:45 +02006687 BlockDriverState *bs_below;
Jeff Codyb1b1d782012-10-16 15:49:09 -04006688
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05006689 GLOBAL_STATE_CODE();
Kevin Wolfb7cfc7d2023-09-29 16:51:45 +02006690 GRAPH_RDLOCK_GUARD_MAINLOOP();
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05006691
Jeff Codyb1b1d782012-10-16 15:49:09 -04006692 if (!bs || !bs->drv || !backing_file) {
Marcelo Tosattie8a6bb92012-01-18 14:40:51 +00006693 return NULL;
6694 }
6695
Jeff Codyb1b1d782012-10-16 15:49:09 -04006696 filename_full = g_malloc(PATH_MAX);
6697 backing_file_full = g_malloc(PATH_MAX);
Jeff Codyb1b1d782012-10-16 15:49:09 -04006698
6699 is_protocol = path_has_protocol(backing_file);
6700
Max Reitzdcf3f9b2019-06-12 17:34:45 +02006701 /*
6702 * Being largely a legacy function, skip any filters here
6703 * (because filters do not have normal filenames, so they cannot
6704 * match anyway; and allowing json:{} filenames is a bit out of
6705 * scope).
6706 */
6707 for (curr_bs = bdrv_skip_filters(bs);
6708 bdrv_cow_child(curr_bs) != NULL;
6709 curr_bs = bs_below)
6710 {
6711 bs_below = bdrv_backing_chain_next(curr_bs);
Jeff Codyb1b1d782012-10-16 15:49:09 -04006712
Max Reitz0b877d02018-08-01 20:34:11 +02006713 if (bdrv_backing_overridden(curr_bs)) {
6714 /*
6715 * If the backing file was overridden, we can only compare
6716 * directly against the backing node's filename.
6717 */
6718
6719 if (!filenames_refreshed) {
6720 /*
6721 * This will automatically refresh all of the
6722 * filenames in the rest of the backing chain, so we
6723 * only need to do this once.
6724 */
6725 bdrv_refresh_filename(bs_below);
6726 filenames_refreshed = true;
6727 }
6728
6729 if (strcmp(backing_file, bs_below->filename) == 0) {
6730 retval = bs_below;
6731 break;
6732 }
6733 } else if (is_protocol || path_has_protocol(curr_bs->backing_file)) {
6734 /*
6735 * If either of the filename paths is actually a protocol, then
6736 * compare unmodified paths; otherwise make paths relative.
6737 */
Max Reitz6b6833c2019-02-01 20:29:15 +01006738 char *backing_file_full_ret;
6739
Jeff Codyb1b1d782012-10-16 15:49:09 -04006740 if (strcmp(backing_file, curr_bs->backing_file) == 0) {
Max Reitzdcf3f9b2019-06-12 17:34:45 +02006741 retval = bs_below;
Jeff Codyb1b1d782012-10-16 15:49:09 -04006742 break;
6743 }
Jeff Cody418661e2017-01-25 20:08:20 -05006744 /* Also check against the full backing filename for the image */
Max Reitz6b6833c2019-02-01 20:29:15 +01006745 backing_file_full_ret = bdrv_get_full_backing_filename(curr_bs,
6746 NULL);
6747 if (backing_file_full_ret) {
6748 bool equal = strcmp(backing_file, backing_file_full_ret) == 0;
6749 g_free(backing_file_full_ret);
6750 if (equal) {
Max Reitzdcf3f9b2019-06-12 17:34:45 +02006751 retval = bs_below;
Jeff Cody418661e2017-01-25 20:08:20 -05006752 break;
6753 }
Jeff Cody418661e2017-01-25 20:08:20 -05006754 }
Marcelo Tosattie8a6bb92012-01-18 14:40:51 +00006755 } else {
Jeff Codyb1b1d782012-10-16 15:49:09 -04006756 /* If not an absolute filename path, make it relative to the current
6757 * image's filename path */
Max Reitz2d9158c2019-02-01 20:29:17 +01006758 filename_tmp = bdrv_make_absolute_filename(curr_bs, backing_file,
6759 NULL);
6760 /* We are going to compare canonicalized absolute pathnames */
6761 if (!filename_tmp || !realpath(filename_tmp, filename_full)) {
6762 g_free(filename_tmp);
Jeff Codyb1b1d782012-10-16 15:49:09 -04006763 continue;
6764 }
Max Reitz2d9158c2019-02-01 20:29:17 +01006765 g_free(filename_tmp);
Jeff Codyb1b1d782012-10-16 15:49:09 -04006766
6767 /* We need to make sure the backing filename we are comparing against
6768 * is relative to the current image filename (or absolute) */
Max Reitz2d9158c2019-02-01 20:29:17 +01006769 filename_tmp = bdrv_get_full_backing_filename(curr_bs, NULL);
6770 if (!filename_tmp || !realpath(filename_tmp, backing_file_full)) {
6771 g_free(filename_tmp);
Jeff Codyb1b1d782012-10-16 15:49:09 -04006772 continue;
6773 }
Max Reitz2d9158c2019-02-01 20:29:17 +01006774 g_free(filename_tmp);
Jeff Codyb1b1d782012-10-16 15:49:09 -04006775
6776 if (strcmp(backing_file_full, filename_full) == 0) {
Max Reitzdcf3f9b2019-06-12 17:34:45 +02006777 retval = bs_below;
Jeff Codyb1b1d782012-10-16 15:49:09 -04006778 break;
6779 }
Marcelo Tosattie8a6bb92012-01-18 14:40:51 +00006780 }
6781 }
6782
Jeff Codyb1b1d782012-10-16 15:49:09 -04006783 g_free(filename_full);
6784 g_free(backing_file_full);
Jeff Codyb1b1d782012-10-16 15:49:09 -04006785 return retval;
Marcelo Tosattie8a6bb92012-01-18 14:40:51 +00006786}
6787
bellardea2384d2004-08-01 21:59:26 +00006788void bdrv_init(void)
6789{
Kevin Wolfe5f05f82021-07-09 18:41:41 +02006790#ifdef CONFIG_BDRV_WHITELIST_TOOLS
6791 use_bdrv_whitelist = 1;
6792#endif
Anthony Liguori5efa9d52009-05-09 17:03:42 -05006793 module_call_init(MODULE_INIT_BLOCK);
bellardea2384d2004-08-01 21:59:26 +00006794}
pbrookce1a14d2006-08-07 02:38:06 +00006795
Markus Armbrustereb852012009-10-27 18:41:44 +01006796void bdrv_init_with_whitelist(void)
6797{
6798 use_bdrv_whitelist = 1;
6799 bdrv_init();
6800}
6801
Emanuele Giuseppe Espositoa94750d2022-02-09 05:54:50 -05006802int bdrv_activate(BlockDriverState *bs, Error **errp)
6803{
Kevin Wolf4417ab72017-05-04 18:52:37 +02006804 BdrvChild *child, *parent;
Kevin Wolf5a8a30d2014-03-12 15:59:16 +01006805 Error *local_err = NULL;
6806 int ret;
Vladimir Sementsov-Ogievskiy9c98f142018-10-29 16:23:17 -04006807 BdrvDirtyBitmap *bm;
Kevin Wolf5a8a30d2014-03-12 15:59:16 +01006808
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05006809 GLOBAL_STATE_CODE();
Kevin Wolf3804e3c2023-09-11 11:46:12 +02006810 GRAPH_RDLOCK_GUARD_MAINLOOP();
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05006811
Kevin Wolf3456a8d2014-03-11 10:58:39 +01006812 if (!bs->drv) {
Vladimir Sementsov-Ogievskiy54166452020-09-24 21:54:08 +03006813 return -ENOMEDIUM;
Anthony Liguori0f154232011-11-14 15:09:45 -06006814 }
Kevin Wolf3456a8d2014-03-11 10:58:39 +01006815
Vladimir Sementsov-Ogievskiy16e977d2017-01-31 14:23:08 +03006816 QLIST_FOREACH(child, &bs->children, next) {
Emanuele Giuseppe Esposito11d0c9b2022-02-09 05:54:52 -05006817 bdrv_activate(child->bs, &local_err);
Fam Zheng0d1c5c92016-05-11 10:45:33 +08006818 if (local_err) {
Fam Zheng0d1c5c92016-05-11 10:45:33 +08006819 error_propagate(errp, local_err);
Vladimir Sementsov-Ogievskiy54166452020-09-24 21:54:08 +03006820 return -EINVAL;
Fam Zheng0d1c5c92016-05-11 10:45:33 +08006821 }
Kevin Wolf5a8a30d2014-03-12 15:59:16 +01006822 }
Fam Zheng0d1c5c92016-05-11 10:45:33 +08006823
Kevin Wolfdafe0962017-11-16 13:00:01 +01006824 /*
6825 * Update permissions, they may differ for inactive nodes.
6826 *
6827 * Note that the required permissions of inactive images are always a
6828 * subset of the permissions required after activating the image. This
6829 * allows us to just get the permissions upfront without restricting
Emanuele Giuseppe Esposito11d0c9b2022-02-09 05:54:52 -05006830 * bdrv_co_invalidate_cache().
Kevin Wolfdafe0962017-11-16 13:00:01 +01006831 *
6832 * It also means that in error cases, we don't have to try and revert to
6833 * the old permissions (which is an operation that could fail, too). We can
6834 * just keep the extended permissions for the next time that an activation
6835 * of the image is tried.
6836 */
Kevin Wolf7bb49412019-12-17 15:06:38 +01006837 if (bs->open_flags & BDRV_O_INACTIVE) {
6838 bs->open_flags &= ~BDRV_O_INACTIVE;
Vladimir Sementsov-Ogievskiyf1316ed2022-11-07 19:35:57 +03006839 ret = bdrv_refresh_perms(bs, NULL, errp);
Kevin Wolf7bb49412019-12-17 15:06:38 +01006840 if (ret < 0) {
Fam Zheng0d1c5c92016-05-11 10:45:33 +08006841 bs->open_flags |= BDRV_O_INACTIVE;
Vladimir Sementsov-Ogievskiy54166452020-09-24 21:54:08 +03006842 return ret;
Fam Zheng0d1c5c92016-05-11 10:45:33 +08006843 }
Kevin Wolf3456a8d2014-03-11 10:58:39 +01006844
Emanuele Giuseppe Esposito11d0c9b2022-02-09 05:54:52 -05006845 ret = bdrv_invalidate_cache(bs, errp);
6846 if (ret < 0) {
6847 bs->open_flags |= BDRV_O_INACTIVE;
6848 return ret;
Kevin Wolf7bb49412019-12-17 15:06:38 +01006849 }
Vladimir Sementsov-Ogievskiy9c98f142018-10-29 16:23:17 -04006850
Kevin Wolf7bb49412019-12-17 15:06:38 +01006851 FOR_EACH_DIRTY_BITMAP(bs, bm) {
6852 bdrv_dirty_bitmap_skip_store(bm, false);
6853 }
6854
Emanuele Giuseppe Espositoc0579602023-01-13 21:42:03 +01006855 ret = bdrv_refresh_total_sectors(bs, bs->total_sectors);
Kevin Wolf7bb49412019-12-17 15:06:38 +01006856 if (ret < 0) {
6857 bs->open_flags |= BDRV_O_INACTIVE;
6858 error_setg_errno(errp, -ret, "Could not refresh total sector count");
Vladimir Sementsov-Ogievskiy54166452020-09-24 21:54:08 +03006859 return ret;
Kevin Wolf7bb49412019-12-17 15:06:38 +01006860 }
Kevin Wolf5a8a30d2014-03-12 15:59:16 +01006861 }
Kevin Wolf4417ab72017-05-04 18:52:37 +02006862
6863 QLIST_FOREACH(parent, &bs->parents, next_parent) {
Max Reitzbd86fb92020-05-13 13:05:13 +02006864 if (parent->klass->activate) {
6865 parent->klass->activate(parent, &local_err);
Kevin Wolf4417ab72017-05-04 18:52:37 +02006866 if (local_err) {
Kevin Wolf78fc3b32019-01-31 15:16:10 +01006867 bs->open_flags |= BDRV_O_INACTIVE;
Kevin Wolf4417ab72017-05-04 18:52:37 +02006868 error_propagate(errp, local_err);
Vladimir Sementsov-Ogievskiy54166452020-09-24 21:54:08 +03006869 return -EINVAL;
Kevin Wolf4417ab72017-05-04 18:52:37 +02006870 }
6871 }
6872 }
Vladimir Sementsov-Ogievskiy54166452020-09-24 21:54:08 +03006873
6874 return 0;
Anthony Liguori0f154232011-11-14 15:09:45 -06006875}
6876
Emanuele Giuseppe Esposito11d0c9b2022-02-09 05:54:52 -05006877int coroutine_fn bdrv_co_invalidate_cache(BlockDriverState *bs, Error **errp)
6878{
6879 Error *local_err = NULL;
Emanuele Giuseppe Esposito1581a702022-03-03 10:16:09 -05006880 IO_CODE();
Emanuele Giuseppe Esposito11d0c9b2022-02-09 05:54:52 -05006881
6882 assert(!(bs->open_flags & BDRV_O_INACTIVE));
Kevin Wolf1b3ff9f2022-12-07 14:18:38 +01006883 assert_bdrv_graph_readable();
Emanuele Giuseppe Esposito11d0c9b2022-02-09 05:54:52 -05006884
6885 if (bs->drv->bdrv_co_invalidate_cache) {
6886 bs->drv->bdrv_co_invalidate_cache(bs, &local_err);
6887 if (local_err) {
6888 error_propagate(errp, local_err);
6889 return -EINVAL;
6890 }
6891 }
6892
6893 return 0;
6894}
6895
Emanuele Giuseppe Esposito3b717192022-02-09 05:54:51 -05006896void bdrv_activate_all(Error **errp)
Anthony Liguori0f154232011-11-14 15:09:45 -06006897{
Kevin Wolf7c8eece2016-03-22 18:58:50 +01006898 BlockDriverState *bs;
Kevin Wolf88be7b42016-05-20 18:49:07 +02006899 BdrvNextIterator it;
Anthony Liguori0f154232011-11-14 15:09:45 -06006900
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05006901 GLOBAL_STATE_CODE();
Kevin Wolf2b3912f2023-09-29 16:51:39 +02006902 GRAPH_RDLOCK_GUARD_MAINLOOP();
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05006903
Kevin Wolf88be7b42016-05-20 18:49:07 +02006904 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
Vladimir Sementsov-Ogievskiy54166452020-09-24 21:54:08 +03006905 int ret;
Stefan Hajnoczied78cda2014-05-08 16:34:35 +02006906
Emanuele Giuseppe Espositoa94750d2022-02-09 05:54:50 -05006907 ret = bdrv_activate(bs, errp);
Vladimir Sementsov-Ogievskiy54166452020-09-24 21:54:08 +03006908 if (ret < 0) {
Max Reitz5e003f12017-11-10 18:25:45 +01006909 bdrv_next_cleanup(&it);
Kevin Wolf5a8a30d2014-03-12 15:59:16 +01006910 return;
6911 }
Anthony Liguori0f154232011-11-14 15:09:45 -06006912 }
6913}
6914
Kevin Wolf0e6bad12023-09-29 16:51:38 +02006915static bool GRAPH_RDLOCK
6916bdrv_has_bds_parent(BlockDriverState *bs, bool only_active)
Kevin Wolf9e372712018-11-23 15:11:14 +01006917{
6918 BdrvChild *parent;
Emanuele Giuseppe Espositobdb73472022-03-03 10:16:02 -05006919 GLOBAL_STATE_CODE();
Kevin Wolf9e372712018-11-23 15:11:14 +01006920
6921 QLIST_FOREACH(parent, &bs->parents, next_parent) {
Max Reitzbd86fb92020-05-13 13:05:13 +02006922 if (parent->klass->parent_is_bds) {
Kevin Wolf9e372712018-11-23 15:11:14 +01006923 BlockDriverState *parent_bs = parent->opaque;
6924 if (!only_active || !(parent_bs->open_flags & BDRV_O_INACTIVE)) {
6925 return true;
6926 }
6927 }
6928 }
6929
6930 return false;
6931}
6932
Kevin Wolf0e6bad12023-09-29 16:51:38 +02006933static int GRAPH_RDLOCK bdrv_inactivate_recurse(BlockDriverState *bs)
Kevin Wolf76b1c7f2015-12-22 14:07:08 +01006934{
Kevin Wolfcfa1a572017-05-04 18:52:38 +02006935 BdrvChild *child, *parent;
Kevin Wolf76b1c7f2015-12-22 14:07:08 +01006936 int ret;
Vladimir Sementsov-Ogievskiya13de402021-09-11 15:00:27 +03006937 uint64_t cumulative_perms, cumulative_shared_perms;
Kevin Wolf76b1c7f2015-12-22 14:07:08 +01006938
Emanuele Giuseppe Espositoda359902022-03-03 10:16:11 -05006939 GLOBAL_STATE_CODE();
6940
Max Reitzd470ad42017-11-10 21:31:09 +01006941 if (!bs->drv) {
6942 return -ENOMEDIUM;
6943 }
6944
Kevin Wolf9e372712018-11-23 15:11:14 +01006945 /* Make sure that we don't inactivate a child before its parent.
6946 * It will be covered by recursion from the yet active parent. */
6947 if (bdrv_has_bds_parent(bs, true)) {
6948 return 0;
6949 }
6950
6951 assert(!(bs->open_flags & BDRV_O_INACTIVE));
6952
6953 /* Inactivate this node */
6954 if (bs->drv->bdrv_inactivate) {
Kevin Wolf76b1c7f2015-12-22 14:07:08 +01006955 ret = bs->drv->bdrv_inactivate(bs);
6956 if (ret < 0) {
6957 return ret;
6958 }
6959 }
6960
Kevin Wolf9e372712018-11-23 15:11:14 +01006961 QLIST_FOREACH(parent, &bs->parents, next_parent) {
Max Reitzbd86fb92020-05-13 13:05:13 +02006962 if (parent->klass->inactivate) {
6963 ret = parent->klass->inactivate(parent);
Kevin Wolf9e372712018-11-23 15:11:14 +01006964 if (ret < 0) {
6965 return ret;
Kevin Wolfcfa1a572017-05-04 18:52:38 +02006966 }
6967 }
Fam Zhengaad0b7a2016-05-11 10:45:35 +08006968 }
Kevin Wolf38701b62017-05-04 18:52:39 +02006969
Vladimir Sementsov-Ogievskiya13de402021-09-11 15:00:27 +03006970 bdrv_get_cumulative_perm(bs, &cumulative_perms,
6971 &cumulative_shared_perms);
6972 if (cumulative_perms & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)) {
6973 /* Our inactive parents still need write access. Inactivation failed. */
6974 return -EPERM;
6975 }
6976
Kevin Wolf9e372712018-11-23 15:11:14 +01006977 bs->open_flags |= BDRV_O_INACTIVE;
6978
Vladimir Sementsov-Ogievskiybb87e4d2020-11-06 15:42:38 +03006979 /*
6980 * Update permissions, they may differ for inactive nodes.
6981 * We only tried to loosen restrictions, so errors are not fatal, ignore
6982 * them.
6983 */
Vladimir Sementsov-Ogievskiyf1316ed2022-11-07 19:35:57 +03006984 bdrv_refresh_perms(bs, NULL, NULL);
Kevin Wolf9e372712018-11-23 15:11:14 +01006985
6986 /* Recursively inactivate children */
Kevin Wolf38701b62017-05-04 18:52:39 +02006987 QLIST_FOREACH(child, &bs->children, next) {
Kevin Wolf9e372712018-11-23 15:11:14 +01006988 ret = bdrv_inactivate_recurse(child->bs);
Kevin Wolf38701b62017-05-04 18:52:39 +02006989 if (ret < 0) {
6990 return ret;
6991 }
6992 }
6993
Kevin Wolf76b1c7f2015-12-22 14:07:08 +01006994 return 0;
6995}
6996
6997int bdrv_inactivate_all(void)
6998{
Max Reitz79720af2016-03-16 19:54:44 +01006999 BlockDriverState *bs = NULL;
Kevin Wolf88be7b42016-05-20 18:49:07 +02007000 BdrvNextIterator it;
Fam Zhengaad0b7a2016-05-11 10:45:35 +08007001 int ret = 0;
Kevin Wolf76b1c7f2015-12-22 14:07:08 +01007002
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05007003 GLOBAL_STATE_CODE();
Kevin Wolf0e6bad12023-09-29 16:51:38 +02007004 GRAPH_RDLOCK_GUARD_MAINLOOP();
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05007005
Kevin Wolf88be7b42016-05-20 18:49:07 +02007006 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
Kevin Wolf9e372712018-11-23 15:11:14 +01007007 /* Nodes with BDS parents are covered by recursion from the last
7008 * parent that gets inactivated. Don't inactivate them a second
7009 * time if that has already happened. */
7010 if (bdrv_has_bds_parent(bs, false)) {
7011 continue;
7012 }
7013 ret = bdrv_inactivate_recurse(bs);
7014 if (ret < 0) {
7015 bdrv_next_cleanup(&it);
Stefan Hajnoczib49f4752023-12-05 13:20:03 -05007016 break;
Kevin Wolf76b1c7f2015-12-22 14:07:08 +01007017 }
7018 }
7019
Fam Zhengaad0b7a2016-05-11 10:45:35 +08007020 return ret;
Kevin Wolf76b1c7f2015-12-22 14:07:08 +01007021}
7022
Kevin Wolff9f05dc2011-07-15 13:50:26 +02007023/**************************************************************/
bellard19cb3732006-08-19 11:45:59 +00007024/* removable device support */
7025
7026/**
7027 * Return TRUE if the media is present
7028 */
Emanuele Giuseppe Esposito1e97be92023-01-13 21:42:02 +01007029bool coroutine_fn bdrv_co_is_inserted(BlockDriverState *bs)
bellard19cb3732006-08-19 11:45:59 +00007030{
7031 BlockDriver *drv = bs->drv;
Max Reitz28d7a782015-10-19 17:53:13 +02007032 BdrvChild *child;
Emanuele Giuseppe Esposito384a48f2022-03-03 10:15:50 -05007033 IO_CODE();
Emanuele Giuseppe Espositoc73ff922023-02-03 16:21:57 +01007034 assert_bdrv_graph_readable();
Markus Armbrustera1aff5b2011-09-06 18:58:41 +02007035
Max Reitze031f752015-10-19 17:53:11 +02007036 if (!drv) {
7037 return false;
7038 }
Emanuele Giuseppe Esposito1e97be92023-01-13 21:42:02 +01007039 if (drv->bdrv_co_is_inserted) {
7040 return drv->bdrv_co_is_inserted(bs);
Max Reitze031f752015-10-19 17:53:11 +02007041 }
Max Reitz28d7a782015-10-19 17:53:13 +02007042 QLIST_FOREACH(child, &bs->children, next) {
Emanuele Giuseppe Esposito1e97be92023-01-13 21:42:02 +01007043 if (!bdrv_co_is_inserted(child->bs)) {
Max Reitz28d7a782015-10-19 17:53:13 +02007044 return false;
7045 }
7046 }
7047 return true;
bellard19cb3732006-08-19 11:45:59 +00007048}
7049
7050/**
bellard19cb3732006-08-19 11:45:59 +00007051 * If eject_flag is TRUE, eject the media. Otherwise, close the tray
7052 */
Emanuele Giuseppe Esposito2531b392023-01-13 21:42:09 +01007053void coroutine_fn bdrv_co_eject(BlockDriverState *bs, bool eject_flag)
bellard19cb3732006-08-19 11:45:59 +00007054{
7055 BlockDriver *drv = bs->drv;
Emanuele Giuseppe Esposito384a48f2022-03-03 10:15:50 -05007056 IO_CODE();
Kevin Wolf79a292e2023-02-03 16:21:58 +01007057 assert_bdrv_graph_readable();
bellard19cb3732006-08-19 11:45:59 +00007058
Emanuele Giuseppe Esposito2531b392023-01-13 21:42:09 +01007059 if (drv && drv->bdrv_co_eject) {
7060 drv->bdrv_co_eject(bs, eject_flag);
bellard19cb3732006-08-19 11:45:59 +00007061 }
bellard19cb3732006-08-19 11:45:59 +00007062}
7063
bellard19cb3732006-08-19 11:45:59 +00007064/**
7065 * Lock or unlock the media (if it is locked, the user won't be able
7066 * to eject it manually).
7067 */
Emanuele Giuseppe Esposito2c752612023-01-13 21:42:10 +01007068void coroutine_fn bdrv_co_lock_medium(BlockDriverState *bs, bool locked)
bellard19cb3732006-08-19 11:45:59 +00007069{
7070 BlockDriver *drv = bs->drv;
Emanuele Giuseppe Esposito384a48f2022-03-03 10:15:50 -05007071 IO_CODE();
Kevin Wolf79a292e2023-02-03 16:21:58 +01007072 assert_bdrv_graph_readable();
Markus Armbruster025e8492011-09-06 18:58:47 +02007073 trace_bdrv_lock_medium(bs, locked);
Stefan Hajnoczib8c6d092011-03-29 20:04:40 +01007074
Emanuele Giuseppe Esposito2c752612023-01-13 21:42:10 +01007075 if (drv && drv->bdrv_co_lock_medium) {
7076 drv->bdrv_co_lock_medium(bs, locked);
bellard19cb3732006-08-19 11:45:59 +00007077 }
7078}
ths985a03b2007-12-24 16:10:43 +00007079
Fam Zheng9fcb0252013-08-23 09:14:46 +08007080/* Get a reference to bs */
7081void bdrv_ref(BlockDriverState *bs)
7082{
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05007083 GLOBAL_STATE_CODE();
Fam Zheng9fcb0252013-08-23 09:14:46 +08007084 bs->refcnt++;
7085}
7086
7087/* Release a previously grabbed reference to bs.
7088 * If after releasing, reference count is zero, the BlockDriverState is
7089 * deleted. */
7090void bdrv_unref(BlockDriverState *bs)
7091{
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05007092 GLOBAL_STATE_CODE();
Jeff Cody9a4d5ca2014-07-23 17:22:57 -04007093 if (!bs) {
7094 return;
7095 }
Fam Zheng9fcb0252013-08-23 09:14:46 +08007096 assert(bs->refcnt > 0);
7097 if (--bs->refcnt == 0) {
7098 bdrv_delete(bs);
7099 }
7100}
7101
Kevin Wolf6bc0bcc2023-11-15 18:20:10 +01007102static void bdrv_schedule_unref_bh(void *opaque)
7103{
7104 BlockDriverState *bs = opaque;
Kevin Wolf6bc0bcc2023-11-15 18:20:10 +01007105
Kevin Wolf6bc0bcc2023-11-15 18:20:10 +01007106 bdrv_unref(bs);
Kevin Wolf6bc0bcc2023-11-15 18:20:10 +01007107}
7108
Kevin Wolfac2ae232023-09-11 11:46:04 +02007109/*
7110 * Release a BlockDriverState reference while holding the graph write lock.
7111 *
7112 * Calling bdrv_unref() directly is forbidden while holding the graph lock
7113 * because bdrv_close() both involves polling and taking the graph lock
7114 * internally. bdrv_schedule_unref() instead delays decreasing the refcount and
7115 * possibly closing @bs until the graph lock is released.
7116 */
7117void bdrv_schedule_unref(BlockDriverState *bs)
7118{
7119 if (!bs) {
7120 return;
7121 }
Kevin Wolf6bc0bcc2023-11-15 18:20:10 +01007122 aio_bh_schedule_oneshot(qemu_get_aio_context(), bdrv_schedule_unref_bh, bs);
Kevin Wolfac2ae232023-09-11 11:46:04 +02007123}
7124
Fam Zhengfbe40ff2014-05-23 21:29:42 +08007125struct BdrvOpBlocker {
7126 Error *reason;
7127 QLIST_ENTRY(BdrvOpBlocker) list;
7128};
7129
7130bool bdrv_op_is_blocked(BlockDriverState *bs, BlockOpType op, Error **errp)
7131{
7132 BdrvOpBlocker *blocker;
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05007133 GLOBAL_STATE_CODE();
Kevin Wolf0bb79c92023-09-29 16:51:49 +02007134
Fam Zhengfbe40ff2014-05-23 21:29:42 +08007135 assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
7136 if (!QLIST_EMPTY(&bs->op_blockers[op])) {
7137 blocker = QLIST_FIRST(&bs->op_blockers[op]);
Markus Armbruster4b576642018-10-17 10:26:25 +02007138 error_propagate_prepend(errp, error_copy(blocker->reason),
7139 "Node '%s' is busy: ",
7140 bdrv_get_device_or_node_name(bs));
Fam Zhengfbe40ff2014-05-23 21:29:42 +08007141 return true;
7142 }
7143 return false;
7144}
7145
7146void bdrv_op_block(BlockDriverState *bs, BlockOpType op, Error *reason)
7147{
7148 BdrvOpBlocker *blocker;
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05007149 GLOBAL_STATE_CODE();
Fam Zhengfbe40ff2014-05-23 21:29:42 +08007150 assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
7151
Markus Armbruster5839e532014-08-19 10:31:08 +02007152 blocker = g_new0(BdrvOpBlocker, 1);
Fam Zhengfbe40ff2014-05-23 21:29:42 +08007153 blocker->reason = reason;
7154 QLIST_INSERT_HEAD(&bs->op_blockers[op], blocker, list);
7155}
7156
7157void bdrv_op_unblock(BlockDriverState *bs, BlockOpType op, Error *reason)
7158{
7159 BdrvOpBlocker *blocker, *next;
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05007160 GLOBAL_STATE_CODE();
Fam Zhengfbe40ff2014-05-23 21:29:42 +08007161 assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
7162 QLIST_FOREACH_SAFE(blocker, &bs->op_blockers[op], list, next) {
7163 if (blocker->reason == reason) {
7164 QLIST_REMOVE(blocker, list);
7165 g_free(blocker);
7166 }
7167 }
7168}
7169
7170void bdrv_op_block_all(BlockDriverState *bs, Error *reason)
7171{
7172 int i;
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05007173 GLOBAL_STATE_CODE();
Fam Zhengfbe40ff2014-05-23 21:29:42 +08007174 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
7175 bdrv_op_block(bs, i, reason);
7176 }
7177}
7178
7179void bdrv_op_unblock_all(BlockDriverState *bs, Error *reason)
7180{
7181 int i;
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05007182 GLOBAL_STATE_CODE();
Fam Zhengfbe40ff2014-05-23 21:29:42 +08007183 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
7184 bdrv_op_unblock(bs, i, reason);
7185 }
7186}
7187
7188bool bdrv_op_blocker_is_empty(BlockDriverState *bs)
7189{
7190 int i;
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05007191 GLOBAL_STATE_CODE();
Fam Zhengfbe40ff2014-05-23 21:29:42 +08007192 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
7193 if (!QLIST_EMPTY(&bs->op_blockers[i])) {
7194 return false;
7195 }
7196 }
7197 return true;
7198}
7199
Kevin Wolfe1355052022-12-07 14:18:30 +01007200/*
7201 * Must not be called while holding the lock of an AioContext other than the
7202 * current one.
7203 */
Luiz Capitulinod92ada22012-11-30 10:52:09 -02007204void bdrv_img_create(const char *filename, const char *fmt,
7205 const char *base_filename, const char *base_fmt,
Fam Zheng92172832017-04-21 20:27:01 +08007206 char *options, uint64_t img_size, int flags, bool quiet,
7207 Error **errp)
Jes Sorensenf88e1a42010-12-16 13:52:15 +01007208{
Chunyan Liu83d05212014-06-05 17:20:51 +08007209 QemuOptsList *create_opts = NULL;
7210 QemuOpts *opts = NULL;
7211 const char *backing_fmt, *backing_file;
7212 int64_t size;
Jes Sorensenf88e1a42010-12-16 13:52:15 +01007213 BlockDriver *drv, *proto_drv;
Max Reitzcc84d902013-09-06 17:14:26 +02007214 Error *local_err = NULL;
Jes Sorensenf88e1a42010-12-16 13:52:15 +01007215 int ret = 0;
7216
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05007217 GLOBAL_STATE_CODE();
7218
Jes Sorensenf88e1a42010-12-16 13:52:15 +01007219 /* Find driver and parse its options */
7220 drv = bdrv_find_format(fmt);
7221 if (!drv) {
Luiz Capitulino71c79812012-11-30 10:52:04 -02007222 error_setg(errp, "Unknown file format '%s'", fmt);
Luiz Capitulinod92ada22012-11-30 10:52:09 -02007223 return;
Jes Sorensenf88e1a42010-12-16 13:52:15 +01007224 }
7225
Max Reitzb65a5e12015-02-05 13:58:12 -05007226 proto_drv = bdrv_find_protocol(filename, true, errp);
Jes Sorensenf88e1a42010-12-16 13:52:15 +01007227 if (!proto_drv) {
Luiz Capitulinod92ada22012-11-30 10:52:09 -02007228 return;
Jes Sorensenf88e1a42010-12-16 13:52:15 +01007229 }
7230
Max Reitzc6149722014-12-02 18:32:45 +01007231 if (!drv->create_opts) {
7232 error_setg(errp, "Format driver '%s' does not support image creation",
7233 drv->format_name);
7234 return;
7235 }
7236
Maxim Levitsky5a5e7f82020-03-26 03:12:18 +02007237 if (!proto_drv->create_opts) {
7238 error_setg(errp, "Protocol driver '%s' does not support image creation",
7239 proto_drv->format_name);
7240 return;
7241 }
7242
Kevin Wolff6dc1c32019-11-26 16:45:49 +01007243 /* Create parameter list */
Chunyan Liuc282e1f2014-06-05 17:21:11 +08007244 create_opts = qemu_opts_append(create_opts, drv->create_opts);
Maxim Levitsky5a5e7f82020-03-26 03:12:18 +02007245 create_opts = qemu_opts_append(create_opts, proto_drv->create_opts);
Jes Sorensenf88e1a42010-12-16 13:52:15 +01007246
Chunyan Liu83d05212014-06-05 17:20:51 +08007247 opts = qemu_opts_create(create_opts, NULL, 0, &error_abort);
Jes Sorensenf88e1a42010-12-16 13:52:15 +01007248
7249 /* Parse -o options */
7250 if (options) {
Markus Armbrustera5f9b9d2020-07-07 18:06:05 +02007251 if (!qemu_opts_do_parse(opts, options, NULL, errp)) {
Jes Sorensenf88e1a42010-12-16 13:52:15 +01007252 goto out;
7253 }
7254 }
7255
Kevin Wolff6dc1c32019-11-26 16:45:49 +01007256 if (!qemu_opt_get(opts, BLOCK_OPT_SIZE)) {
7257 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, img_size, &error_abort);
7258 } else if (img_size != UINT64_C(-1)) {
7259 error_setg(errp, "The image size must be specified only once");
7260 goto out;
7261 }
7262
Jes Sorensenf88e1a42010-12-16 13:52:15 +01007263 if (base_filename) {
Markus Armbruster235e59c2020-07-07 18:05:42 +02007264 if (!qemu_opt_set(opts, BLOCK_OPT_BACKING_FILE, base_filename,
Markus Armbruster38825782020-07-07 18:05:43 +02007265 NULL)) {
Luiz Capitulino71c79812012-11-30 10:52:04 -02007266 error_setg(errp, "Backing file not supported for file format '%s'",
7267 fmt);
Jes Sorensenf88e1a42010-12-16 13:52:15 +01007268 goto out;
7269 }
7270 }
7271
7272 if (base_fmt) {
Markus Armbruster38825782020-07-07 18:05:43 +02007273 if (!qemu_opt_set(opts, BLOCK_OPT_BACKING_FMT, base_fmt, NULL)) {
Luiz Capitulino71c79812012-11-30 10:52:04 -02007274 error_setg(errp, "Backing file format not supported for file "
7275 "format '%s'", fmt);
Jes Sorensenf88e1a42010-12-16 13:52:15 +01007276 goto out;
7277 }
7278 }
7279
Chunyan Liu83d05212014-06-05 17:20:51 +08007280 backing_file = qemu_opt_get(opts, BLOCK_OPT_BACKING_FILE);
7281 if (backing_file) {
7282 if (!strcmp(filename, backing_file)) {
Luiz Capitulino71c79812012-11-30 10:52:04 -02007283 error_setg(errp, "Error: Trying to create an image with the "
7284 "same filename as the backing file");
Jes Sorensen792da932010-12-16 13:52:17 +01007285 goto out;
7286 }
Connor Kuehl975a7bd2020-08-13 08:47:22 -05007287 if (backing_file[0] == '\0') {
7288 error_setg(errp, "Expected backing file name, got empty string");
7289 goto out;
7290 }
Jes Sorensen792da932010-12-16 13:52:17 +01007291 }
7292
Chunyan Liu83d05212014-06-05 17:20:51 +08007293 backing_fmt = qemu_opt_get(opts, BLOCK_OPT_BACKING_FMT);
Jes Sorensenf88e1a42010-12-16 13:52:15 +01007294
John Snow6e6e55f2017-07-17 20:34:22 -04007295 /* The size for the image must always be specified, unless we have a backing
7296 * file and we have not been forbidden from opening it. */
Eric Blakea8b42a12017-09-25 09:55:07 -05007297 size = qemu_opt_get_size(opts, BLOCK_OPT_SIZE, img_size);
John Snow6e6e55f2017-07-17 20:34:22 -04007298 if (backing_file && !(flags & BDRV_O_NO_BACKING)) {
7299 BlockDriverState *bs;
Max Reitz645ae7d2019-02-01 20:29:14 +01007300 char *full_backing;
John Snow6e6e55f2017-07-17 20:34:22 -04007301 int back_flags;
7302 QDict *backing_options = NULL;
Paolo Bonzini63090da2012-04-12 14:01:03 +02007303
Max Reitz645ae7d2019-02-01 20:29:14 +01007304 full_backing =
7305 bdrv_get_full_backing_filename_from_filename(filename, backing_file,
7306 &local_err);
John Snow6e6e55f2017-07-17 20:34:22 -04007307 if (local_err) {
Jes Sorensenf88e1a42010-12-16 13:52:15 +01007308 goto out;
7309 }
Max Reitz645ae7d2019-02-01 20:29:14 +01007310 assert(full_backing);
John Snow6e6e55f2017-07-17 20:34:22 -04007311
Max Reitzd5b23992021-06-22 16:00:30 +02007312 /*
7313 * No need to do I/O here, which allows us to open encrypted
7314 * backing images without needing the secret
7315 */
John Snow6e6e55f2017-07-17 20:34:22 -04007316 back_flags = flags;
7317 back_flags &= ~(BDRV_O_RDWR | BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING);
Max Reitzd5b23992021-06-22 16:00:30 +02007318 back_flags |= BDRV_O_NO_IO;
John Snow6e6e55f2017-07-17 20:34:22 -04007319
Fam Zhengcc954f02017-12-15 16:04:45 +08007320 backing_options = qdict_new();
John Snow6e6e55f2017-07-17 20:34:22 -04007321 if (backing_fmt) {
John Snow6e6e55f2017-07-17 20:34:22 -04007322 qdict_put_str(backing_options, "driver", backing_fmt);
7323 }
Fam Zhengcc954f02017-12-15 16:04:45 +08007324 qdict_put_bool(backing_options, BDRV_OPT_FORCE_SHARE, true);
John Snow6e6e55f2017-07-17 20:34:22 -04007325
7326 bs = bdrv_open(full_backing, NULL, backing_options, back_flags,
7327 &local_err);
7328 g_free(full_backing);
Eric Blakeadd82002020-07-06 15:39:50 -05007329 if (!bs) {
7330 error_append_hint(&local_err, "Could not open backing image.\n");
John Snow6e6e55f2017-07-17 20:34:22 -04007331 goto out;
7332 } else {
Eric Blaked9f059a2020-07-06 15:39:54 -05007333 if (!backing_fmt) {
Eric Blake497a30d2021-05-03 14:36:00 -07007334 error_setg(&local_err,
7335 "Backing file specified without backing format");
Michael Tokarevfbdffb02023-04-05 16:34:04 +03007336 error_append_hint(&local_err, "Detected format of %s.\n",
Eric Blake497a30d2021-05-03 14:36:00 -07007337 bs->drv->format_name);
7338 goto out;
Eric Blaked9f059a2020-07-06 15:39:54 -05007339 }
John Snow6e6e55f2017-07-17 20:34:22 -04007340 if (size == -1) {
7341 /* Opened BS, have no size */
7342 size = bdrv_getlength(bs);
7343 if (size < 0) {
7344 error_setg_errno(errp, -size, "Could not get size of '%s'",
7345 backing_file);
7346 bdrv_unref(bs);
7347 goto out;
7348 }
7349 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, size, &error_abort);
7350 }
7351 bdrv_unref(bs);
7352 }
Eric Blaked9f059a2020-07-06 15:39:54 -05007353 /* (backing_file && !(flags & BDRV_O_NO_BACKING)) */
7354 } else if (backing_file && !backing_fmt) {
Eric Blake497a30d2021-05-03 14:36:00 -07007355 error_setg(&local_err,
7356 "Backing file specified without backing format");
7357 goto out;
Eric Blaked9f059a2020-07-06 15:39:54 -05007358 }
John Snow6e6e55f2017-07-17 20:34:22 -04007359
7360 if (size == -1) {
7361 error_setg(errp, "Image creation needs a size parameter");
7362 goto out;
Jes Sorensenf88e1a42010-12-16 13:52:15 +01007363 }
7364
Miroslav Rezaninaf382d432013-02-13 09:09:40 +01007365 if (!quiet) {
Kővágó, Zoltánfe646692015-07-07 16:42:10 +02007366 printf("Formatting '%s', fmt=%s ", filename, fmt);
Fam Zheng43c5d8f2014-12-09 15:38:04 +08007367 qemu_opts_print(opts, " ");
Miroslav Rezaninaf382d432013-02-13 09:09:40 +01007368 puts("");
Eric Blake4e2f4412020-07-06 15:39:45 -05007369 fflush(stdout);
Miroslav Rezaninaf382d432013-02-13 09:09:40 +01007370 }
Chunyan Liu83d05212014-06-05 17:20:51 +08007371
Chunyan Liuc282e1f2014-06-05 17:21:11 +08007372 ret = bdrv_create(drv, filename, opts, &local_err);
Chunyan Liu83d05212014-06-05 17:20:51 +08007373
Max Reitzcc84d902013-09-06 17:14:26 +02007374 if (ret == -EFBIG) {
7375 /* This is generally a better message than whatever the driver would
7376 * deliver (especially because of the cluster_size_hint), since that
7377 * is most probably not much different from "image too large". */
7378 const char *cluster_size_hint = "";
Chunyan Liu83d05212014-06-05 17:20:51 +08007379 if (qemu_opt_get_size(opts, BLOCK_OPT_CLUSTER_SIZE, 0)) {
Max Reitzcc84d902013-09-06 17:14:26 +02007380 cluster_size_hint = " (try using a larger cluster size)";
Jes Sorensenf88e1a42010-12-16 13:52:15 +01007381 }
Max Reitzcc84d902013-09-06 17:14:26 +02007382 error_setg(errp, "The image size is too large for file format '%s'"
7383 "%s", fmt, cluster_size_hint);
7384 error_free(local_err);
7385 local_err = NULL;
Jes Sorensenf88e1a42010-12-16 13:52:15 +01007386 }
7387
7388out:
Chunyan Liu83d05212014-06-05 17:20:51 +08007389 qemu_opts_del(opts);
7390 qemu_opts_free(create_opts);
Eduardo Habkost621ff942016-06-13 18:57:56 -03007391 error_propagate(errp, local_err);
Jes Sorensenf88e1a42010-12-16 13:52:15 +01007392}
Stefan Hajnoczi85d126f2013-03-07 13:41:48 +01007393
7394AioContext *bdrv_get_aio_context(BlockDriverState *bs)
7395{
Emanuele Giuseppe Esposito384a48f2022-03-03 10:15:50 -05007396 IO_CODE();
Stefan Hajnoczi33f2a752018-02-16 16:50:13 +00007397 return bs ? bs->aio_context : qemu_get_aio_context();
Stefan Hajnoczidcd04222014-05-08 16:34:37 +02007398}
7399
Kevin Wolfe336fd42020-10-05 17:58:53 +02007400AioContext *coroutine_fn bdrv_co_enter(BlockDriverState *bs)
7401{
7402 Coroutine *self = qemu_coroutine_self();
7403 AioContext *old_ctx = qemu_coroutine_get_aio_context(self);
7404 AioContext *new_ctx;
Emanuele Giuseppe Esposito384a48f2022-03-03 10:15:50 -05007405 IO_CODE();
Kevin Wolfe336fd42020-10-05 17:58:53 +02007406
7407 /*
7408 * Increase bs->in_flight to ensure that this operation is completed before
7409 * moving the node to a different AioContext. Read new_ctx only afterwards.
7410 */
7411 bdrv_inc_in_flight(bs);
7412
7413 new_ctx = bdrv_get_aio_context(bs);
7414 aio_co_reschedule_self(new_ctx);
7415 return old_ctx;
7416}
7417
7418void coroutine_fn bdrv_co_leave(BlockDriverState *bs, AioContext *old_ctx)
7419{
Emanuele Giuseppe Esposito384a48f2022-03-03 10:15:50 -05007420 IO_CODE();
Kevin Wolfe336fd42020-10-05 17:58:53 +02007421 aio_co_reschedule_self(old_ctx);
7422 bdrv_dec_in_flight(bs);
7423}
7424
Stefan Hajnoczie8a095d2016-06-16 17:56:26 +01007425static void bdrv_do_remove_aio_context_notifier(BdrvAioNotifier *ban)
7426{
Emanuele Giuseppe Espositobdb73472022-03-03 10:16:02 -05007427 GLOBAL_STATE_CODE();
Stefan Hajnoczie8a095d2016-06-16 17:56:26 +01007428 QLIST_REMOVE(ban, list);
7429 g_free(ban);
7430}
7431
Kevin Wolfa3a683c2019-05-06 19:17:57 +02007432static void bdrv_detach_aio_context(BlockDriverState *bs)
Stefan Hajnoczidcd04222014-05-08 16:34:37 +02007433{
Stefan Hajnoczie8a095d2016-06-16 17:56:26 +01007434 BdrvAioNotifier *baf, *baf_tmp;
Max Reitz33384422014-06-20 21:57:33 +02007435
Stefan Hajnoczie8a095d2016-06-16 17:56:26 +01007436 assert(!bs->walking_aio_notifiers);
Emanuele Giuseppe Espositoda359902022-03-03 10:16:11 -05007437 GLOBAL_STATE_CODE();
Stefan Hajnoczie8a095d2016-06-16 17:56:26 +01007438 bs->walking_aio_notifiers = true;
7439 QLIST_FOREACH_SAFE(baf, &bs->aio_notifiers, list, baf_tmp) {
7440 if (baf->deleted) {
7441 bdrv_do_remove_aio_context_notifier(baf);
7442 } else {
7443 baf->detach_aio_context(baf->opaque);
7444 }
Max Reitz33384422014-06-20 21:57:33 +02007445 }
Stefan Hajnoczie8a095d2016-06-16 17:56:26 +01007446 /* Never mind iterating again to check for ->deleted. bdrv_close() will
7447 * remove remaining aio notifiers if we aren't called again.
7448 */
7449 bs->walking_aio_notifiers = false;
Max Reitz33384422014-06-20 21:57:33 +02007450
Kevin Wolf1bffe1a2019-04-17 17:15:25 +02007451 if (bs->drv && bs->drv->bdrv_detach_aio_context) {
Stefan Hajnoczidcd04222014-05-08 16:34:37 +02007452 bs->drv->bdrv_detach_aio_context(bs);
7453 }
Stefan Hajnoczidcd04222014-05-08 16:34:37 +02007454
7455 bs->aio_context = NULL;
7456}
7457
Kevin Wolfa3a683c2019-05-06 19:17:57 +02007458static void bdrv_attach_aio_context(BlockDriverState *bs,
7459 AioContext *new_context)
Stefan Hajnoczidcd04222014-05-08 16:34:37 +02007460{
Stefan Hajnoczie8a095d2016-06-16 17:56:26 +01007461 BdrvAioNotifier *ban, *ban_tmp;
Emanuele Giuseppe Espositoda359902022-03-03 10:16:11 -05007462 GLOBAL_STATE_CODE();
Max Reitz33384422014-06-20 21:57:33 +02007463
Stefan Hajnoczidcd04222014-05-08 16:34:37 +02007464 bs->aio_context = new_context;
7465
Kevin Wolf1bffe1a2019-04-17 17:15:25 +02007466 if (bs->drv && bs->drv->bdrv_attach_aio_context) {
Stefan Hajnoczidcd04222014-05-08 16:34:37 +02007467 bs->drv->bdrv_attach_aio_context(bs, new_context);
7468 }
Max Reitz33384422014-06-20 21:57:33 +02007469
Stefan Hajnoczie8a095d2016-06-16 17:56:26 +01007470 assert(!bs->walking_aio_notifiers);
7471 bs->walking_aio_notifiers = true;
7472 QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_tmp) {
7473 if (ban->deleted) {
7474 bdrv_do_remove_aio_context_notifier(ban);
7475 } else {
7476 ban->attached_aio_context(new_context, ban->opaque);
7477 }
Max Reitz33384422014-06-20 21:57:33 +02007478 }
Stefan Hajnoczie8a095d2016-06-16 17:56:26 +01007479 bs->walking_aio_notifiers = false;
Stefan Hajnoczidcd04222014-05-08 16:34:37 +02007480}
7481
Emanuele Giuseppe Esposito7e8c1822022-10-25 04:49:44 -04007482typedef struct BdrvStateSetAioContext {
7483 AioContext *new_ctx;
7484 BlockDriverState *bs;
7485} BdrvStateSetAioContext;
7486
7487static bool bdrv_parent_change_aio_context(BdrvChild *c, AioContext *ctx,
Emanuele Giuseppe Espositoe08cc002022-10-25 04:49:45 -04007488 GHashTable *visited,
7489 Transaction *tran,
Emanuele Giuseppe Esposito7e8c1822022-10-25 04:49:44 -04007490 Error **errp)
7491{
7492 GLOBAL_STATE_CODE();
Emanuele Giuseppe Espositoe08cc002022-10-25 04:49:45 -04007493 if (g_hash_table_contains(visited, c)) {
Emanuele Giuseppe Esposito7e8c1822022-10-25 04:49:44 -04007494 return true;
7495 }
Emanuele Giuseppe Espositoe08cc002022-10-25 04:49:45 -04007496 g_hash_table_add(visited, c);
Emanuele Giuseppe Esposito7e8c1822022-10-25 04:49:44 -04007497
7498 /*
7499 * A BdrvChildClass that doesn't handle AioContext changes cannot
7500 * tolerate any AioContext changes
7501 */
7502 if (!c->klass->change_aio_ctx) {
7503 char *user = bdrv_child_user_desc(c);
7504 error_setg(errp, "Changing iothreads is not supported by %s", user);
7505 g_free(user);
7506 return false;
7507 }
7508 if (!c->klass->change_aio_ctx(c, ctx, visited, tran, errp)) {
7509 assert(!errp || *errp);
7510 return false;
7511 }
7512 return true;
7513}
7514
Emanuele Giuseppe Esposito7e8c1822022-10-25 04:49:44 -04007515bool bdrv_child_change_aio_context(BdrvChild *c, AioContext *ctx,
Emanuele Giuseppe Espositoe08cc002022-10-25 04:49:45 -04007516 GHashTable *visited, Transaction *tran,
Emanuele Giuseppe Esposito7e8c1822022-10-25 04:49:44 -04007517 Error **errp)
7518{
7519 GLOBAL_STATE_CODE();
Emanuele Giuseppe Espositoe08cc002022-10-25 04:49:45 -04007520 if (g_hash_table_contains(visited, c)) {
Emanuele Giuseppe Esposito7e8c1822022-10-25 04:49:44 -04007521 return true;
7522 }
Emanuele Giuseppe Espositoe08cc002022-10-25 04:49:45 -04007523 g_hash_table_add(visited, c);
Emanuele Giuseppe Esposito7e8c1822022-10-25 04:49:44 -04007524 return bdrv_change_aio_context(c->bs, ctx, visited, tran, errp);
7525}
7526
Emanuele Giuseppe Esposito7e8c1822022-10-25 04:49:44 -04007527static void bdrv_set_aio_context_clean(void *opaque)
7528{
7529 BdrvStateSetAioContext *state = (BdrvStateSetAioContext *) opaque;
7530 BlockDriverState *bs = (BlockDriverState *) state->bs;
7531
7532 /* Paired with bdrv_drained_begin in bdrv_change_aio_context() */
7533 bdrv_drained_end(bs);
7534
7535 g_free(state);
7536}
7537
7538static void bdrv_set_aio_context_commit(void *opaque)
7539{
7540 BdrvStateSetAioContext *state = (BdrvStateSetAioContext *) opaque;
7541 BlockDriverState *bs = (BlockDriverState *) state->bs;
7542 AioContext *new_context = state->new_ctx;
Emanuele Giuseppe Esposito7e8c1822022-10-25 04:49:44 -04007543
Emanuele Giuseppe Esposito7e8c1822022-10-25 04:49:44 -04007544 bdrv_detach_aio_context(bs);
Emanuele Giuseppe Esposito7e8c1822022-10-25 04:49:44 -04007545 bdrv_attach_aio_context(bs, new_context);
7546}
7547
7548static TransactionActionDrv set_aio_context = {
7549 .commit = bdrv_set_aio_context_commit,
7550 .clean = bdrv_set_aio_context_clean,
7551};
7552
Kevin Wolf42a65f02019-05-07 18:31:38 +02007553/*
7554 * Changes the AioContext used for fd handlers, timers, and BHs by this
7555 * BlockDriverState and all its children and parents.
7556 *
Max Reitz43eaaae2019-07-22 15:30:54 +02007557 * Must be called from the main AioContext.
7558 *
Emanuele Giuseppe Esposito7e8c1822022-10-25 04:49:44 -04007559 * @visited will accumulate all visited BdrvChild objects. The caller is
Kevin Wolf42a65f02019-05-07 18:31:38 +02007560 * responsible for freeing the list afterwards.
7561 */
Emanuele Giuseppe Esposito7e8c1822022-10-25 04:49:44 -04007562static bool bdrv_change_aio_context(BlockDriverState *bs, AioContext *ctx,
Emanuele Giuseppe Espositoe08cc002022-10-25 04:49:45 -04007563 GHashTable *visited, Transaction *tran,
Emanuele Giuseppe Esposito7e8c1822022-10-25 04:49:44 -04007564 Error **errp)
Kevin Wolf5d231842019-05-06 19:17:56 +02007565{
7566 BdrvChild *c;
Emanuele Giuseppe Esposito7e8c1822022-10-25 04:49:44 -04007567 BdrvStateSetAioContext *state;
7568
7569 GLOBAL_STATE_CODE();
Kevin Wolf5d231842019-05-06 19:17:56 +02007570
7571 if (bdrv_get_aio_context(bs) == ctx) {
7572 return true;
7573 }
7574
Kevin Wolf7f831d22023-09-11 11:46:17 +02007575 bdrv_graph_rdlock_main_loop();
Kevin Wolf5d231842019-05-06 19:17:56 +02007576 QLIST_FOREACH(c, &bs->parents, next_parent) {
Emanuele Giuseppe Esposito7e8c1822022-10-25 04:49:44 -04007577 if (!bdrv_parent_change_aio_context(c, ctx, visited, tran, errp)) {
Kevin Wolf7f831d22023-09-11 11:46:17 +02007578 bdrv_graph_rdunlock_main_loop();
Kevin Wolf5d231842019-05-06 19:17:56 +02007579 return false;
7580 }
7581 }
Emanuele Giuseppe Esposito7e8c1822022-10-25 04:49:44 -04007582
Kevin Wolf5d231842019-05-06 19:17:56 +02007583 QLIST_FOREACH(c, &bs->children, next) {
Emanuele Giuseppe Esposito7e8c1822022-10-25 04:49:44 -04007584 if (!bdrv_child_change_aio_context(c, ctx, visited, tran, errp)) {
Kevin Wolf7f831d22023-09-11 11:46:17 +02007585 bdrv_graph_rdunlock_main_loop();
Kevin Wolf5d231842019-05-06 19:17:56 +02007586 return false;
7587 }
7588 }
Kevin Wolf7f831d22023-09-11 11:46:17 +02007589 bdrv_graph_rdunlock_main_loop();
Kevin Wolf5d231842019-05-06 19:17:56 +02007590
Emanuele Giuseppe Esposito7e8c1822022-10-25 04:49:44 -04007591 state = g_new(BdrvStateSetAioContext, 1);
7592 *state = (BdrvStateSetAioContext) {
7593 .new_ctx = ctx,
7594 .bs = bs,
7595 };
7596
7597 /* Paired with bdrv_drained_end in bdrv_set_aio_context_clean() */
7598 bdrv_drained_begin(bs);
7599
7600 tran_add(tran, &set_aio_context, state);
7601
Kevin Wolf5d231842019-05-06 19:17:56 +02007602 return true;
7603}
7604
Emanuele Giuseppe Esposito7e8c1822022-10-25 04:49:44 -04007605/*
7606 * Change bs's and recursively all of its parents' and children's AioContext
7607 * to the given new context, returning an error if that isn't possible.
7608 *
7609 * If ignore_child is not NULL, that child (and its subgraph) will not
7610 * be touched.
Emanuele Giuseppe Esposito7e8c1822022-10-25 04:49:44 -04007611 */
Emanuele Giuseppe Espositoa41cfda2022-10-25 04:49:51 -04007612int bdrv_try_change_aio_context(BlockDriverState *bs, AioContext *ctx,
7613 BdrvChild *ignore_child, Error **errp)
Kevin Wolf5d231842019-05-06 19:17:56 +02007614{
Emanuele Giuseppe Esposito7e8c1822022-10-25 04:49:44 -04007615 Transaction *tran;
Emanuele Giuseppe Espositoe08cc002022-10-25 04:49:45 -04007616 GHashTable *visited;
Emanuele Giuseppe Esposito7e8c1822022-10-25 04:49:44 -04007617 int ret;
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05007618 GLOBAL_STATE_CODE();
7619
Emanuele Giuseppe Esposito7e8c1822022-10-25 04:49:44 -04007620 /*
7621 * Recursion phase: go through all nodes of the graph.
7622 * Take care of checking that all nodes support changing AioContext
Michael Tokarev3202d8e2023-07-14 14:06:05 +03007623 * and drain them, building a linear list of callbacks to run if everything
Emanuele Giuseppe Esposito7e8c1822022-10-25 04:49:44 -04007624 * is successful (the transaction itself).
7625 */
7626 tran = tran_new();
Emanuele Giuseppe Espositoe08cc002022-10-25 04:49:45 -04007627 visited = g_hash_table_new(NULL, NULL);
7628 if (ignore_child) {
7629 g_hash_table_add(visited, ignore_child);
7630 }
7631 ret = bdrv_change_aio_context(bs, ctx, visited, tran, errp);
7632 g_hash_table_destroy(visited);
Emanuele Giuseppe Esposito7e8c1822022-10-25 04:49:44 -04007633
7634 /*
7635 * Linear phase: go through all callbacks collected in the transaction.
Stefan Hajnoczi23c983c2023-12-05 13:20:11 -05007636 * Run all callbacks collected in the recursion to switch every node's
7637 * AioContext (transaction commit), or undo all changes done in the
Emanuele Giuseppe Esposito7e8c1822022-10-25 04:49:44 -04007638 * recursion (transaction abort).
7639 */
Kevin Wolf5d231842019-05-06 19:17:56 +02007640
7641 if (!ret) {
Emanuele Giuseppe Esposito7e8c1822022-10-25 04:49:44 -04007642 /* Just run clean() callbacks. No AioContext changed. */
7643 tran_abort(tran);
Kevin Wolf5d231842019-05-06 19:17:56 +02007644 return -EPERM;
7645 }
7646
Emanuele Giuseppe Esposito7e8c1822022-10-25 04:49:44 -04007647 tran_commit(tran);
Kevin Wolf5d231842019-05-06 19:17:56 +02007648 return 0;
7649}
7650
Max Reitz33384422014-06-20 21:57:33 +02007651void bdrv_add_aio_context_notifier(BlockDriverState *bs,
7652 void (*attached_aio_context)(AioContext *new_context, void *opaque),
7653 void (*detach_aio_context)(void *opaque), void *opaque)
7654{
7655 BdrvAioNotifier *ban = g_new(BdrvAioNotifier, 1);
7656 *ban = (BdrvAioNotifier){
7657 .attached_aio_context = attached_aio_context,
7658 .detach_aio_context = detach_aio_context,
7659 .opaque = opaque
7660 };
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05007661 GLOBAL_STATE_CODE();
Max Reitz33384422014-06-20 21:57:33 +02007662
7663 QLIST_INSERT_HEAD(&bs->aio_notifiers, ban, list);
7664}
7665
7666void bdrv_remove_aio_context_notifier(BlockDriverState *bs,
7667 void (*attached_aio_context)(AioContext *,
7668 void *),
7669 void (*detach_aio_context)(void *),
7670 void *opaque)
7671{
7672 BdrvAioNotifier *ban, *ban_next;
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05007673 GLOBAL_STATE_CODE();
Max Reitz33384422014-06-20 21:57:33 +02007674
7675 QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_next) {
7676 if (ban->attached_aio_context == attached_aio_context &&
7677 ban->detach_aio_context == detach_aio_context &&
Stefan Hajnoczie8a095d2016-06-16 17:56:26 +01007678 ban->opaque == opaque &&
7679 ban->deleted == false)
Max Reitz33384422014-06-20 21:57:33 +02007680 {
Stefan Hajnoczie8a095d2016-06-16 17:56:26 +01007681 if (bs->walking_aio_notifiers) {
7682 ban->deleted = true;
7683 } else {
7684 bdrv_do_remove_aio_context_notifier(ban);
7685 }
Max Reitz33384422014-06-20 21:57:33 +02007686 return;
7687 }
7688 }
7689
7690 abort();
7691}
7692
Max Reitz77485432014-10-27 11:12:50 +01007693int bdrv_amend_options(BlockDriverState *bs, QemuOpts *opts,
Max Reitzd1402b52018-05-09 23:00:18 +02007694 BlockDriverAmendStatusCB *status_cb, void *cb_opaque,
Maxim Levitskya3579bf2020-06-25 14:55:38 +02007695 bool force,
Max Reitzd1402b52018-05-09 23:00:18 +02007696 Error **errp)
Max Reitz6f176b42013-09-03 10:09:50 +02007697{
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05007698 GLOBAL_STATE_CODE();
Max Reitzd470ad42017-11-10 21:31:09 +01007699 if (!bs->drv) {
Max Reitzd1402b52018-05-09 23:00:18 +02007700 error_setg(errp, "Node is ejected");
Max Reitzd470ad42017-11-10 21:31:09 +01007701 return -ENOMEDIUM;
7702 }
Chunyan Liuc282e1f2014-06-05 17:21:11 +08007703 if (!bs->drv->bdrv_amend_options) {
Max Reitzd1402b52018-05-09 23:00:18 +02007704 error_setg(errp, "Block driver '%s' does not support option amendment",
7705 bs->drv->format_name);
Max Reitz6f176b42013-09-03 10:09:50 +02007706 return -ENOTSUP;
7707 }
Maxim Levitskya3579bf2020-06-25 14:55:38 +02007708 return bs->drv->bdrv_amend_options(bs, opts, status_cb,
7709 cb_opaque, force, errp);
Max Reitz6f176b42013-09-03 10:09:50 +02007710}
Benoît Canetf6186f42013-10-02 14:33:48 +02007711
Max Reitz5d69b5a2020-02-18 11:34:41 +01007712/*
7713 * This function checks whether the given @to_replace is allowed to be
7714 * replaced by a node that always shows the same data as @bs. This is
7715 * used for example to verify whether the mirror job can replace
7716 * @to_replace by the target mirrored from @bs.
7717 * To be replaceable, @bs and @to_replace may either be guaranteed to
7718 * always show the same data (because they are only connected through
7719 * filters), or some driver may allow replacing one of its children
7720 * because it can guarantee that this child's data is not visible at
7721 * all (for example, for dissenting quorum children that have no other
7722 * parents).
7723 */
7724bool bdrv_recurse_can_replace(BlockDriverState *bs,
7725 BlockDriverState *to_replace)
7726{
Max Reitz93393e62019-06-12 17:03:38 +02007727 BlockDriverState *filtered;
7728
Emanuele Giuseppe Espositob4ad82a2022-03-03 10:15:57 -05007729 GLOBAL_STATE_CODE();
7730
Max Reitz5d69b5a2020-02-18 11:34:41 +01007731 if (!bs || !bs->drv) {
7732 return false;
7733 }
7734
7735 if (bs == to_replace) {
7736 return true;
7737 }
7738
7739 /* See what the driver can do */
7740 if (bs->drv->bdrv_recurse_can_replace) {
7741 return bs->drv->bdrv_recurse_can_replace(bs, to_replace);
7742 }
7743
7744 /* For filters without an own implementation, we can recurse on our own */
Max Reitz93393e62019-06-12 17:03:38 +02007745 filtered = bdrv_filter_bs(bs);
7746 if (filtered) {
7747 return bdrv_recurse_can_replace(filtered, to_replace);
Max Reitz5d69b5a2020-02-18 11:34:41 +01007748 }
7749
7750 /* Safe default */
7751 return false;
7752}
7753
Max Reitz810803a2020-02-18 11:34:44 +01007754/*
7755 * Check whether the given @node_name can be replaced by a node that
7756 * has the same data as @parent_bs. If so, return @node_name's BDS;
7757 * NULL otherwise.
7758 *
7759 * @node_name must be a (recursive) *child of @parent_bs (or this
7760 * function will return NULL).
7761 *
7762 * The result (whether the node can be replaced or not) is only valid
7763 * for as long as no graph or permission changes occur.
7764 */
Wen Congyange12f3782015-07-17 10:12:22 +08007765BlockDriverState *check_to_replace_node(BlockDriverState *parent_bs,
7766 const char *node_name, Error **errp)
Benoît Canet09158f02014-06-27 18:25:25 +02007767{
7768 BlockDriverState *to_replace_bs = bdrv_find_node(node_name);
Stefan Hajnoczi5a7e7a02014-10-21 12:03:58 +01007769
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05007770 GLOBAL_STATE_CODE();
7771
Benoît Canet09158f02014-06-27 18:25:25 +02007772 if (!to_replace_bs) {
Connor Kuehl785ec4b2021-03-05 09:19:28 -06007773 error_setg(errp, "Failed to find node with node-name='%s'", node_name);
Benoît Canet09158f02014-06-27 18:25:25 +02007774 return NULL;
7775 }
7776
7777 if (bdrv_op_is_blocked(to_replace_bs, BLOCK_OP_TYPE_REPLACE, errp)) {
Stefan Hajnoczib49f4752023-12-05 13:20:03 -05007778 return NULL;
Benoît Canet09158f02014-06-27 18:25:25 +02007779 }
7780
7781 /* We don't want arbitrary node of the BDS chain to be replaced only the top
7782 * most non filter in order to prevent data corruption.
7783 * Another benefit is that this tests exclude backing files which are
7784 * blocked by the backing blockers.
7785 */
Max Reitz810803a2020-02-18 11:34:44 +01007786 if (!bdrv_recurse_can_replace(parent_bs, to_replace_bs)) {
7787 error_setg(errp, "Cannot replace '%s' by a node mirrored from '%s', "
7788 "because it cannot be guaranteed that doing so would not "
7789 "lead to an abrupt change of visible data",
7790 node_name, parent_bs->node_name);
Stefan Hajnoczib49f4752023-12-05 13:20:03 -05007791 return NULL;
Benoît Canet09158f02014-06-27 18:25:25 +02007792 }
7793
7794 return to_replace_bs;
7795}
Ming Lei448ad912014-07-04 18:04:33 +08007796
Max Reitz97e2f022019-02-01 20:29:27 +01007797/**
7798 * Iterates through the list of runtime option keys that are said to
7799 * be "strong" for a BDS. An option is called "strong" if it changes
7800 * a BDS's data. For example, the null block driver's "size" and
7801 * "read-zeroes" options are strong, but its "latency-ns" option is
7802 * not.
7803 *
7804 * If a key returned by this function ends with a dot, all options
7805 * starting with that prefix are strong.
7806 */
7807static const char *const *strong_options(BlockDriverState *bs,
7808 const char *const *curopt)
7809{
7810 static const char *const global_options[] = {
7811 "driver", "filename", NULL
7812 };
7813
7814 if (!curopt) {
7815 return &global_options[0];
7816 }
7817
7818 curopt++;
7819 if (curopt == &global_options[ARRAY_SIZE(global_options) - 1] && bs->drv) {
7820 curopt = bs->drv->strong_runtime_opts;
7821 }
7822
7823 return (curopt && *curopt) ? curopt : NULL;
7824}
7825
7826/**
7827 * Copies all strong runtime options from bs->options to the given
7828 * QDict. The set of strong option keys is determined by invoking
7829 * strong_options().
7830 *
7831 * Returns true iff any strong option was present in bs->options (and
7832 * thus copied to the target QDict) with the exception of "filename"
7833 * and "driver". The caller is expected to use this value to decide
7834 * whether the existence of strong options prevents the generation of
7835 * a plain filename.
7836 */
7837static bool append_strong_runtime_options(QDict *d, BlockDriverState *bs)
7838{
7839 bool found_any = false;
7840 const char *const *option_name = NULL;
7841
7842 if (!bs->drv) {
7843 return false;
7844 }
7845
7846 while ((option_name = strong_options(bs, option_name))) {
7847 bool option_given = false;
7848
7849 assert(strlen(*option_name) > 0);
7850 if ((*option_name)[strlen(*option_name) - 1] != '.') {
7851 QObject *entry = qdict_get(bs->options, *option_name);
7852 if (!entry) {
7853 continue;
7854 }
7855
7856 qdict_put_obj(d, *option_name, qobject_ref(entry));
7857 option_given = true;
7858 } else {
7859 const QDictEntry *entry;
7860 for (entry = qdict_first(bs->options); entry;
7861 entry = qdict_next(bs->options, entry))
7862 {
7863 if (strstart(qdict_entry_key(entry), *option_name, NULL)) {
7864 qdict_put_obj(d, qdict_entry_key(entry),
7865 qobject_ref(qdict_entry_value(entry)));
7866 option_given = true;
7867 }
7868 }
7869 }
7870
7871 /* While "driver" and "filename" need to be included in a JSON filename,
7872 * their existence does not prohibit generation of a plain filename. */
7873 if (!found_any && option_given &&
7874 strcmp(*option_name, "driver") && strcmp(*option_name, "filename"))
7875 {
7876 found_any = true;
7877 }
7878 }
7879
Max Reitz62a01a272019-02-01 20:29:34 +01007880 if (!qdict_haskey(d, "driver")) {
7881 /* Drivers created with bdrv_new_open_driver() may not have a
7882 * @driver option. Add it here. */
7883 qdict_put_str(d, "driver", bs->drv->format_name);
7884 }
7885
Max Reitz97e2f022019-02-01 20:29:27 +01007886 return found_any;
7887}
7888
Max Reitz90993622019-02-01 20:29:09 +01007889/* Note: This function may return false positives; it may return true
7890 * even if opening the backing file specified by bs's image header
7891 * would result in exactly bs->backing. */
Kevin Wolf004915a2023-10-27 17:53:26 +02007892static bool GRAPH_RDLOCK bdrv_backing_overridden(BlockDriverState *bs)
Max Reitz90993622019-02-01 20:29:09 +01007893{
Emanuele Giuseppe Espositob4ad82a2022-03-03 10:15:57 -05007894 GLOBAL_STATE_CODE();
Max Reitz90993622019-02-01 20:29:09 +01007895 if (bs->backing) {
7896 return strcmp(bs->auto_backing_file,
7897 bs->backing->bs->filename);
7898 } else {
7899 /* No backing BDS, so if the image header reports any backing
7900 * file, it must have been suppressed */
7901 return bs->auto_backing_file[0] != '\0';
7902 }
7903}
7904
Max Reitz91af7012014-07-18 20:24:56 +02007905/* Updates the following BDS fields:
7906 * - exact_filename: A filename which may be used for opening a block device
7907 * which (mostly) equals the given BDS (even without any
7908 * other options; so reading and writing must return the same
7909 * results, but caching etc. may be different)
7910 * - full_open_options: Options which, when given when opening a block device
7911 * (without a filename), result in a BDS (mostly)
7912 * equalling the given one
7913 * - filename: If exact_filename is set, it is copied here. Otherwise,
7914 * full_open_options is converted to a JSON object, prefixed with
7915 * "json:" (for use through the JSON pseudo protocol) and put here.
7916 */
7917void bdrv_refresh_filename(BlockDriverState *bs)
7918{
7919 BlockDriver *drv = bs->drv;
Max Reitze24518e2019-02-01 20:29:06 +01007920 BdrvChild *child;
Max Reitz52f72d62019-06-12 17:43:03 +02007921 BlockDriverState *primary_child_bs;
Max Reitz91af7012014-07-18 20:24:56 +02007922 QDict *opts;
Max Reitz90993622019-02-01 20:29:09 +01007923 bool backing_overridden;
Max Reitz998b3a12019-02-01 20:29:28 +01007924 bool generate_json_filename; /* Whether our default implementation should
7925 fill exact_filename (false) or not (true) */
Max Reitz91af7012014-07-18 20:24:56 +02007926
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05007927 GLOBAL_STATE_CODE();
7928
Max Reitz91af7012014-07-18 20:24:56 +02007929 if (!drv) {
7930 return;
7931 }
7932
Max Reitze24518e2019-02-01 20:29:06 +01007933 /* This BDS's file name may depend on any of its children's file names, so
7934 * refresh those first */
7935 QLIST_FOREACH(child, &bs->children, next) {
7936 bdrv_refresh_filename(child->bs);
Max Reitz91af7012014-07-18 20:24:56 +02007937 }
7938
Max Reitzbb808d52019-02-01 20:29:07 +01007939 if (bs->implicit) {
7940 /* For implicit nodes, just copy everything from the single child */
7941 child = QLIST_FIRST(&bs->children);
7942 assert(QLIST_NEXT(child, next) == NULL);
7943
7944 pstrcpy(bs->exact_filename, sizeof(bs->exact_filename),
7945 child->bs->exact_filename);
7946 pstrcpy(bs->filename, sizeof(bs->filename), child->bs->filename);
7947
Pan Nengyuancb895612020-01-16 16:56:00 +08007948 qobject_unref(bs->full_open_options);
Max Reitzbb808d52019-02-01 20:29:07 +01007949 bs->full_open_options = qobject_ref(child->bs->full_open_options);
7950
7951 return;
7952 }
7953
Max Reitz90993622019-02-01 20:29:09 +01007954 backing_overridden = bdrv_backing_overridden(bs);
7955
7956 if (bs->open_flags & BDRV_O_NO_IO) {
7957 /* Without I/O, the backing file does not change anything.
7958 * Therefore, in such a case (primarily qemu-img), we can
7959 * pretend the backing file has not been overridden even if
7960 * it technically has been. */
7961 backing_overridden = false;
7962 }
7963
Max Reitz97e2f022019-02-01 20:29:27 +01007964 /* Gather the options QDict */
7965 opts = qdict_new();
Max Reitz998b3a12019-02-01 20:29:28 +01007966 generate_json_filename = append_strong_runtime_options(opts, bs);
7967 generate_json_filename |= backing_overridden;
Max Reitz97e2f022019-02-01 20:29:27 +01007968
7969 if (drv->bdrv_gather_child_options) {
7970 /* Some block drivers may not want to present all of their children's
7971 * options, or name them differently from BdrvChild.name */
7972 drv->bdrv_gather_child_options(bs, opts, backing_overridden);
7973 } else {
7974 QLIST_FOREACH(child, &bs->children, next) {
Max Reitz25191e52020-05-13 13:05:33 +02007975 if (child == bs->backing && !backing_overridden) {
Max Reitz97e2f022019-02-01 20:29:27 +01007976 /* We can skip the backing BDS if it has not been overridden */
7977 continue;
7978 }
7979
7980 qdict_put(opts, child->name,
7981 qobject_ref(child->bs->full_open_options));
7982 }
7983
7984 if (backing_overridden && !bs->backing) {
7985 /* Force no backing file */
7986 qdict_put_null(opts, "backing");
7987 }
7988 }
7989
7990 qobject_unref(bs->full_open_options);
7991 bs->full_open_options = opts;
7992
Max Reitz52f72d62019-06-12 17:43:03 +02007993 primary_child_bs = bdrv_primary_bs(bs);
7994
Max Reitz998b3a12019-02-01 20:29:28 +01007995 if (drv->bdrv_refresh_filename) {
7996 /* Obsolete information is of no use here, so drop the old file name
7997 * information before refreshing it */
7998 bs->exact_filename[0] = '\0';
7999
8000 drv->bdrv_refresh_filename(bs);
Max Reitz52f72d62019-06-12 17:43:03 +02008001 } else if (primary_child_bs) {
8002 /*
8003 * Try to reconstruct valid information from the underlying
8004 * file -- this only works for format nodes (filter nodes
8005 * cannot be probed and as such must be selected by the user
8006 * either through an options dict, or through a special
8007 * filename which the filter driver must construct in its
8008 * .bdrv_refresh_filename() implementation).
8009 */
Max Reitz998b3a12019-02-01 20:29:28 +01008010
8011 bs->exact_filename[0] = '\0';
8012
Max Reitzfb695c72019-02-01 20:29:29 +01008013 /*
8014 * We can use the underlying file's filename if:
8015 * - it has a filename,
Max Reitz52f72d62019-06-12 17:43:03 +02008016 * - the current BDS is not a filter,
Max Reitzfb695c72019-02-01 20:29:29 +01008017 * - the file is a protocol BDS, and
8018 * - opening that file (as this BDS's format) will automatically create
8019 * the BDS tree we have right now, that is:
8020 * - the user did not significantly change this BDS's behavior with
8021 * some explicit (strong) options
8022 * - no non-file child of this BDS has been overridden by the user
8023 * Both of these conditions are represented by generate_json_filename.
8024 */
Max Reitz52f72d62019-06-12 17:43:03 +02008025 if (primary_child_bs->exact_filename[0] &&
8026 primary_child_bs->drv->bdrv_file_open &&
8027 !drv->is_filter && !generate_json_filename)
Max Reitzfb695c72019-02-01 20:29:29 +01008028 {
Max Reitz52f72d62019-06-12 17:43:03 +02008029 strcpy(bs->exact_filename, primary_child_bs->exact_filename);
Max Reitz998b3a12019-02-01 20:29:28 +01008030 }
8031 }
8032
Max Reitz91af7012014-07-18 20:24:56 +02008033 if (bs->exact_filename[0]) {
8034 pstrcpy(bs->filename, sizeof(bs->filename), bs->exact_filename);
Max Reitz97e2f022019-02-01 20:29:27 +01008035 } else {
Markus Armbrustereab3a462020-12-11 18:11:37 +01008036 GString *json = qobject_to_json(QOBJECT(bs->full_open_options));
Eric Blake5c86bdf2020-06-08 13:26:38 -05008037 if (snprintf(bs->filename, sizeof(bs->filename), "json:%s",
Markus Armbrustereab3a462020-12-11 18:11:37 +01008038 json->str) >= sizeof(bs->filename)) {
Eric Blake5c86bdf2020-06-08 13:26:38 -05008039 /* Give user a hint if we truncated things. */
8040 strcpy(bs->filename + sizeof(bs->filename) - 4, "...");
8041 }
Markus Armbrustereab3a462020-12-11 18:11:37 +01008042 g_string_free(json, true);
Max Reitz91af7012014-07-18 20:24:56 +02008043 }
8044}
Wen Congyange06018a2016-05-10 15:36:37 +08008045
Max Reitz1e89d0f2019-02-01 20:29:18 +01008046char *bdrv_dirname(BlockDriverState *bs, Error **errp)
8047{
8048 BlockDriver *drv = bs->drv;
Max Reitz52f72d62019-06-12 17:43:03 +02008049 BlockDriverState *child_bs;
Max Reitz1e89d0f2019-02-01 20:29:18 +01008050
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05008051 GLOBAL_STATE_CODE();
8052
Max Reitz1e89d0f2019-02-01 20:29:18 +01008053 if (!drv) {
8054 error_setg(errp, "Node '%s' is ejected", bs->node_name);
8055 return NULL;
8056 }
8057
8058 if (drv->bdrv_dirname) {
8059 return drv->bdrv_dirname(bs, errp);
8060 }
8061
Max Reitz52f72d62019-06-12 17:43:03 +02008062 child_bs = bdrv_primary_bs(bs);
8063 if (child_bs) {
8064 return bdrv_dirname(child_bs, errp);
Max Reitz1e89d0f2019-02-01 20:29:18 +01008065 }
8066
8067 bdrv_refresh_filename(bs);
8068 if (bs->exact_filename[0] != '\0') {
8069 return path_combine(bs->exact_filename, "");
8070 }
8071
8072 error_setg(errp, "Cannot generate a base directory for %s nodes",
8073 drv->format_name);
8074 return NULL;
8075}
8076
Wen Congyange06018a2016-05-10 15:36:37 +08008077/*
8078 * Hot add/remove a BDS's child. So the user can take a child offline when
8079 * it is broken and take a new child online
8080 */
8081void bdrv_add_child(BlockDriverState *parent_bs, BlockDriverState *child_bs,
8082 Error **errp)
8083{
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05008084 GLOBAL_STATE_CODE();
Wen Congyange06018a2016-05-10 15:36:37 +08008085 if (!parent_bs->drv || !parent_bs->drv->bdrv_add_child) {
8086 error_setg(errp, "The node %s does not support adding a child",
8087 bdrv_get_device_or_node_name(parent_bs));
8088 return;
8089 }
8090
Sam Li774c7262023-05-08 12:55:30 +08008091 /*
8092 * Non-zoned block drivers do not follow zoned storage constraints
8093 * (i.e. sequential writes to zones). Refuse mixing zoned and non-zoned
8094 * drivers in a graph.
8095 */
8096 if (!parent_bs->drv->supports_zoned_children &&
8097 child_bs->bl.zoned == BLK_Z_HM) {
8098 /*
8099 * The host-aware model allows zoned storage constraints and random
8100 * write. Allow mixing host-aware and non-zoned drivers. Using
8101 * host-aware device as a regular device.
8102 */
8103 error_setg(errp, "Cannot add a %s child to a %s parent",
8104 child_bs->bl.zoned == BLK_Z_HM ? "zoned" : "non-zoned",
8105 parent_bs->drv->supports_zoned_children ?
8106 "support zoned children" : "not support zoned children");
8107 return;
8108 }
8109
Wen Congyange06018a2016-05-10 15:36:37 +08008110 if (!QLIST_EMPTY(&child_bs->parents)) {
8111 error_setg(errp, "The node %s already has a parent",
8112 child_bs->node_name);
8113 return;
8114 }
8115
8116 parent_bs->drv->bdrv_add_child(parent_bs, child_bs, errp);
8117}
8118
8119void bdrv_del_child(BlockDriverState *parent_bs, BdrvChild *child, Error **errp)
8120{
8121 BdrvChild *tmp;
8122
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05008123 GLOBAL_STATE_CODE();
Wen Congyange06018a2016-05-10 15:36:37 +08008124 if (!parent_bs->drv || !parent_bs->drv->bdrv_del_child) {
8125 error_setg(errp, "The node %s does not support removing a child",
8126 bdrv_get_device_or_node_name(parent_bs));
8127 return;
8128 }
8129
8130 QLIST_FOREACH(tmp, &parent_bs->children, next) {
8131 if (tmp == child) {
8132 break;
8133 }
8134 }
8135
8136 if (!tmp) {
8137 error_setg(errp, "The node %s does not have a child named %s",
8138 bdrv_get_device_or_node_name(parent_bs),
8139 bdrv_get_device_or_node_name(child->bs));
8140 return;
8141 }
8142
8143 parent_bs->drv->bdrv_del_child(parent_bs, child, errp);
8144}
Max Reitz6f7a3b52020-04-29 16:11:23 +02008145
8146int bdrv_make_empty(BdrvChild *c, Error **errp)
8147{
8148 BlockDriver *drv = c->bs->drv;
8149 int ret;
8150
Emanuele Giuseppe Espositof791bf72022-03-03 10:15:49 -05008151 GLOBAL_STATE_CODE();
Max Reitz6f7a3b52020-04-29 16:11:23 +02008152 assert(c->perm & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED));
8153
8154 if (!drv->bdrv_make_empty) {
8155 error_setg(errp, "%s does not support emptying nodes",
8156 drv->format_name);
8157 return -ENOTSUP;
8158 }
8159
8160 ret = drv->bdrv_make_empty(c->bs);
8161 if (ret < 0) {
8162 error_setg_errno(errp, -ret, "Failed to empty %s",
8163 c->bs->filename);
8164 return ret;
8165 }
8166
8167 return 0;
8168}
Max Reitz9a6fc882019-05-31 15:23:11 +02008169
8170/*
8171 * Return the child that @bs acts as an overlay for, and from which data may be
8172 * copied in COW or COR operations. Usually this is the backing file.
8173 */
8174BdrvChild *bdrv_cow_child(BlockDriverState *bs)
8175{
Emanuele Giuseppe Esposito967d7902022-03-03 10:15:58 -05008176 IO_CODE();
8177
Max Reitz9a6fc882019-05-31 15:23:11 +02008178 if (!bs || !bs->drv) {
8179 return NULL;
8180 }
8181
8182 if (bs->drv->is_filter) {
8183 return NULL;
8184 }
8185
8186 if (!bs->backing) {
8187 return NULL;
8188 }
8189
8190 assert(bs->backing->role & BDRV_CHILD_COW);
8191 return bs->backing;
8192}
8193
8194/*
8195 * If @bs acts as a filter for exactly one of its children, return
8196 * that child.
8197 */
8198BdrvChild *bdrv_filter_child(BlockDriverState *bs)
8199{
8200 BdrvChild *c;
Emanuele Giuseppe Esposito967d7902022-03-03 10:15:58 -05008201 IO_CODE();
Max Reitz9a6fc882019-05-31 15:23:11 +02008202
8203 if (!bs || !bs->drv) {
8204 return NULL;
8205 }
8206
8207 if (!bs->drv->is_filter) {
8208 return NULL;
8209 }
8210
8211 /* Only one of @backing or @file may be used */
8212 assert(!(bs->backing && bs->file));
8213
8214 c = bs->backing ?: bs->file;
8215 if (!c) {
8216 return NULL;
8217 }
8218
8219 assert(c->role & BDRV_CHILD_FILTERED);
8220 return c;
8221}
8222
8223/*
8224 * Return either the result of bdrv_cow_child() or bdrv_filter_child(),
8225 * whichever is non-NULL.
8226 *
8227 * Return NULL if both are NULL.
8228 */
8229BdrvChild *bdrv_filter_or_cow_child(BlockDriverState *bs)
8230{
8231 BdrvChild *cow_child = bdrv_cow_child(bs);
8232 BdrvChild *filter_child = bdrv_filter_child(bs);
Emanuele Giuseppe Esposito967d7902022-03-03 10:15:58 -05008233 IO_CODE();
Max Reitz9a6fc882019-05-31 15:23:11 +02008234
8235 /* Filter nodes cannot have COW backing files */
8236 assert(!(cow_child && filter_child));
8237
8238 return cow_child ?: filter_child;
8239}
8240
8241/*
8242 * Return the primary child of this node: For filters, that is the
8243 * filtered child. For other nodes, that is usually the child storing
8244 * metadata.
8245 * (A generally more helpful description is that this is (usually) the
8246 * child that has the same filename as @bs.)
8247 *
8248 * Drivers do not necessarily have a primary child; for example quorum
8249 * does not.
8250 */
8251BdrvChild *bdrv_primary_child(BlockDriverState *bs)
8252{
8253 BdrvChild *c, *found = NULL;
Emanuele Giuseppe Esposito967d7902022-03-03 10:15:58 -05008254 IO_CODE();
Max Reitz9a6fc882019-05-31 15:23:11 +02008255
8256 QLIST_FOREACH(c, &bs->children, next) {
8257 if (c->role & BDRV_CHILD_PRIMARY) {
8258 assert(!found);
8259 found = c;
8260 }
8261 }
8262
8263 return found;
8264}
Max Reitzd38d7eb2019-06-12 15:06:37 +02008265
Kevin Wolfec82cc42023-10-27 17:53:20 +02008266static BlockDriverState * GRAPH_RDLOCK
8267bdrv_do_skip_filters(BlockDriverState *bs, bool stop_on_explicit_filter)
Max Reitzd38d7eb2019-06-12 15:06:37 +02008268{
8269 BdrvChild *c;
8270
8271 if (!bs) {
8272 return NULL;
8273 }
8274
8275 while (!(stop_on_explicit_filter && !bs->implicit)) {
8276 c = bdrv_filter_child(bs);
8277 if (!c) {
8278 /*
8279 * A filter that is embedded in a working block graph must
8280 * have a child. Assert this here so this function does
8281 * not return a filter node that is not expected by the
8282 * caller.
8283 */
8284 assert(!bs->drv || !bs->drv->is_filter);
8285 break;
8286 }
8287 bs = c->bs;
8288 }
8289 /*
8290 * Note that this treats nodes with bs->drv == NULL as not being
8291 * filters (bs->drv == NULL should be replaced by something else
8292 * anyway).
8293 * The advantage of this behavior is that this function will thus
8294 * always return a non-NULL value (given a non-NULL @bs).
8295 */
8296
8297 return bs;
8298}
8299
8300/*
8301 * Return the first BDS that has not been added implicitly or that
8302 * does not have a filtered child down the chain starting from @bs
8303 * (including @bs itself).
8304 */
8305BlockDriverState *bdrv_skip_implicit_filters(BlockDriverState *bs)
8306{
Emanuele Giuseppe Espositob4ad82a2022-03-03 10:15:57 -05008307 GLOBAL_STATE_CODE();
Max Reitzd38d7eb2019-06-12 15:06:37 +02008308 return bdrv_do_skip_filters(bs, true);
8309}
8310
8311/*
8312 * Return the first BDS that does not have a filtered child down the
8313 * chain starting from @bs (including @bs itself).
8314 */
8315BlockDriverState *bdrv_skip_filters(BlockDriverState *bs)
8316{
Emanuele Giuseppe Esposito967d7902022-03-03 10:15:58 -05008317 IO_CODE();
Max Reitzd38d7eb2019-06-12 15:06:37 +02008318 return bdrv_do_skip_filters(bs, false);
8319}
8320
8321/*
8322 * For a backing chain, return the first non-filter backing image of
8323 * the first non-filter image.
8324 */
8325BlockDriverState *bdrv_backing_chain_next(BlockDriverState *bs)
8326{
Emanuele Giuseppe Esposito967d7902022-03-03 10:15:58 -05008327 IO_CODE();
Max Reitzd38d7eb2019-06-12 15:06:37 +02008328 return bdrv_skip_filters(bdrv_cow_bs(bdrv_skip_filters(bs)));
8329}
Hanna Reitz0bc329f2021-08-12 10:41:44 +02008330
8331/**
8332 * Check whether [offset, offset + bytes) overlaps with the cached
8333 * block-status data region.
8334 *
8335 * If so, and @pnum is not NULL, set *pnum to `bsc.data_end - offset`,
8336 * which is what bdrv_bsc_is_data()'s interface needs.
8337 * Otherwise, *pnum is not touched.
8338 */
8339static bool bdrv_bsc_range_overlaps_locked(BlockDriverState *bs,
8340 int64_t offset, int64_t bytes,
8341 int64_t *pnum)
8342{
8343 BdrvBlockStatusCache *bsc = qatomic_rcu_read(&bs->block_status_cache);
8344 bool overlaps;
8345
8346 overlaps =
8347 qatomic_read(&bsc->valid) &&
8348 ranges_overlap(offset, bytes, bsc->data_start,
8349 bsc->data_end - bsc->data_start);
8350
8351 if (overlaps && pnum) {
8352 *pnum = bsc->data_end - offset;
8353 }
8354
8355 return overlaps;
8356}
8357
8358/**
8359 * See block_int.h for this function's documentation.
8360 */
8361bool bdrv_bsc_is_data(BlockDriverState *bs, int64_t offset, int64_t *pnum)
8362{
Emanuele Giuseppe Esposito967d7902022-03-03 10:15:58 -05008363 IO_CODE();
Hanna Reitz0bc329f2021-08-12 10:41:44 +02008364 RCU_READ_LOCK_GUARD();
Hanna Reitz0bc329f2021-08-12 10:41:44 +02008365 return bdrv_bsc_range_overlaps_locked(bs, offset, 1, pnum);
8366}
8367
8368/**
8369 * See block_int.h for this function's documentation.
8370 */
8371void bdrv_bsc_invalidate_range(BlockDriverState *bs,
8372 int64_t offset, int64_t bytes)
8373{
Emanuele Giuseppe Esposito967d7902022-03-03 10:15:58 -05008374 IO_CODE();
Hanna Reitz0bc329f2021-08-12 10:41:44 +02008375 RCU_READ_LOCK_GUARD();
8376
8377 if (bdrv_bsc_range_overlaps_locked(bs, offset, bytes, NULL)) {
8378 qatomic_set(&bs->block_status_cache->valid, false);
8379 }
8380}
8381
8382/**
8383 * See block_int.h for this function's documentation.
8384 */
8385void bdrv_bsc_fill(BlockDriverState *bs, int64_t offset, int64_t bytes)
8386{
8387 BdrvBlockStatusCache *new_bsc = g_new(BdrvBlockStatusCache, 1);
8388 BdrvBlockStatusCache *old_bsc;
Emanuele Giuseppe Esposito967d7902022-03-03 10:15:58 -05008389 IO_CODE();
Hanna Reitz0bc329f2021-08-12 10:41:44 +02008390
8391 *new_bsc = (BdrvBlockStatusCache) {
8392 .valid = true,
8393 .data_start = offset,
8394 .data_end = offset + bytes,
8395 };
8396
8397 QEMU_LOCK_GUARD(&bs->bsc_modify_lock);
8398
8399 old_bsc = qatomic_rcu_read(&bs->block_status_cache);
8400 qatomic_rcu_set(&bs->block_status_cache, new_bsc);
8401 if (old_bsc) {
8402 g_free_rcu(old_bsc, rcu);
8403 }
8404}