]>
Commit | Line | Data |
---|---|---|
fc01f7e7 FB |
1 | /* |
2 | * QEMU System Emulator block driver | |
5fafdf24 | 3 | * |
fc01f7e7 | 4 | * Copyright (c) 2003 Fabrice Bellard |
c20555e1 | 5 | * Copyright (c) 2020 Virtuozzo International GmbH. |
5fafdf24 | 6 | * |
fc01f7e7 FB |
7 | * Permission is hereby granted, free of charge, to any person obtaining a copy |
8 | * of this software and associated documentation files (the "Software"), to deal | |
9 | * in the Software without restriction, including without limitation the rights | |
10 | * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell | |
11 | * copies of the Software, and to permit persons to whom the Software is | |
12 | * furnished to do so, subject to the following conditions: | |
13 | * | |
14 | * The above copyright notice and this permission notice shall be included in | |
15 | * all copies or substantial portions of the Software. | |
16 | * | |
17 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR | |
18 | * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, | |
19 | * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL | |
20 | * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER | |
21 | * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, | |
22 | * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN | |
23 | * THE SOFTWARE. | |
24 | */ | |
e688df6b | 25 | |
d38ea87a | 26 | #include "qemu/osdep.h" |
0ab8ed18 | 27 | #include "block/trace.h" |
737e150e PB |
28 | #include "block/block_int.h" |
29 | #include "block/blockjob.h" | |
e2c1c34f | 30 | #include "block/dirty-bitmap.h" |
0c9b70d5 | 31 | #include "block/fuse.h" |
cd7fca95 | 32 | #include "block/nbd.h" |
609f45ea | 33 | #include "block/qdict.h" |
d49b6836 | 34 | #include "qemu/error-report.h" |
5e5733e5 | 35 | #include "block/module_block.h" |
db725815 | 36 | #include "qemu/main-loop.h" |
1de7afc9 | 37 | #include "qemu/module.h" |
e688df6b | 38 | #include "qapi/error.h" |
452fcdbc | 39 | #include "qapi/qmp/qdict.h" |
7b1b5d19 | 40 | #include "qapi/qmp/qjson.h" |
e59a0cf1 | 41 | #include "qapi/qmp/qnull.h" |
fc81fa1e | 42 | #include "qapi/qmp/qstring.h" |
e1d74bc6 KW |
43 | #include "qapi/qobject-output-visitor.h" |
44 | #include "qapi/qapi-visit-block-core.h" | |
bfb197e0 | 45 | #include "sysemu/block-backend.h" |
1de7afc9 | 46 | #include "qemu/notify.h" |
922a01a0 | 47 | #include "qemu/option.h" |
10817bf0 | 48 | #include "qemu/coroutine.h" |
c13163fb | 49 | #include "block/qapi.h" |
1de7afc9 | 50 | #include "qemu/timer.h" |
f348b6d1 VB |
51 | #include "qemu/cutils.h" |
52 | #include "qemu/id.h" | |
0bc329fb HR |
53 | #include "qemu/range.h" |
54 | #include "qemu/rcu.h" | |
21c2283e | 55 | #include "block/coroutines.h" |
fc01f7e7 | 56 | |
71e72a19 | 57 | #ifdef CONFIG_BSD |
7674e7bf | 58 | #include <sys/ioctl.h> |
72cf2d4f | 59 | #include <sys/queue.h> |
feccdcee | 60 | #if defined(HAVE_SYS_DISK_H) |
7674e7bf FB |
61 | #include <sys/disk.h> |
62 | #endif | |
c5e97233 | 63 | #endif |
7674e7bf | 64 | |
49dc768d AL |
65 | #ifdef _WIN32 |
66 | #include <windows.h> | |
67 | #endif | |
68 | ||
1c9805a3 SH |
69 | #define NOT_DONE 0x7fffffff /* used while emulated sync operation in progress */ |
70 | ||
3b491a90 | 71 | /* Protected by BQL */ |
dc364f4c BC |
72 | static QTAILQ_HEAD(, BlockDriverState) graph_bdrv_states = |
73 | QTAILQ_HEAD_INITIALIZER(graph_bdrv_states); | |
74 | ||
3b491a90 | 75 | /* Protected by BQL */ |
2c1d04e0 HR |
76 | static QTAILQ_HEAD(, BlockDriverState) all_bdrv_states = |
77 | QTAILQ_HEAD_INITIALIZER(all_bdrv_states); | |
78 | ||
3b491a90 | 79 | /* Protected by BQL */ |
8a22f02a SH |
80 | static QLIST_HEAD(, BlockDriver) bdrv_drivers = |
81 | QLIST_HEAD_INITIALIZER(bdrv_drivers); | |
ea2384d3 | 82 | |
5b363937 HR |
83 | static BlockDriverState *bdrv_open_inherit(const char *filename, |
84 | const char *reference, | |
85 | QDict *options, int flags, | |
86 | BlockDriverState *parent, | |
bd86fb99 | 87 | const BdrvChildClass *child_class, |
272c02ea | 88 | BdrvChildRole child_role, |
5b363937 | 89 | Error **errp); |
f3930ed0 | 90 | |
bfb8aa6d KW |
91 | static bool bdrv_recurse_has_child(BlockDriverState *bs, |
92 | BlockDriverState *child); | |
93 | ||
544acc7d | 94 | static void bdrv_replace_child_noperm(BdrvChild *child, |
4eba825a | 95 | BlockDriverState *new_bs); |
57f08941 | 96 | static void bdrv_remove_child(BdrvChild *child, Transaction *tran); |
0978623e | 97 | |
72373e40 VSO |
98 | static int bdrv_reopen_prepare(BDRVReopenState *reopen_state, |
99 | BlockReopenQueue *queue, | |
ecd30d2d | 100 | Transaction *change_child_tran, Error **errp); |
53e96d1e VSO |
101 | static void bdrv_reopen_commit(BDRVReopenState *reopen_state); |
102 | static void bdrv_reopen_abort(BDRVReopenState *reopen_state); | |
103 | ||
fa8fc1d0 EGE |
104 | static bool bdrv_backing_overridden(BlockDriverState *bs); |
105 | ||
7e8c182f | 106 | static bool bdrv_change_aio_context(BlockDriverState *bs, AioContext *ctx, |
e08cc001 | 107 | GHashTable *visited, Transaction *tran, |
7e8c182f EGE |
108 | Error **errp); |
109 | ||
eb852011 MA |
110 | /* If non-zero, use only whitelisted block drivers */ |
111 | static int use_bdrv_whitelist; | |
112 | ||
9e0b22f4 SH |
113 | #ifdef _WIN32 |
114 | static int is_windows_drive_prefix(const char *filename) | |
115 | { | |
116 | return (((filename[0] >= 'a' && filename[0] <= 'z') || | |
117 | (filename[0] >= 'A' && filename[0] <= 'Z')) && | |
118 | filename[1] == ':'); | |
119 | } | |
120 | ||
121 | int is_windows_drive(const char *filename) | |
122 | { | |
123 | if (is_windows_drive_prefix(filename) && | |
124 | filename[2] == '\0') | |
125 | return 1; | |
126 | if (strstart(filename, "\\\\.\\", NULL) || | |
127 | strstart(filename, "//./", NULL)) | |
128 | return 1; | |
129 | return 0; | |
130 | } | |
131 | #endif | |
132 | ||
339064d5 KW |
133 | size_t bdrv_opt_mem_align(BlockDriverState *bs) |
134 | { | |
135 | if (!bs || !bs->drv) { | |
459b4e66 | 136 | /* page size or 4k (hdd sector size) should be on the safe side */ |
8e3b0cbb | 137 | return MAX(4096, qemu_real_host_page_size()); |
339064d5 | 138 | } |
384a48fb | 139 | IO_CODE(); |
339064d5 KW |
140 | |
141 | return bs->bl.opt_mem_alignment; | |
142 | } | |
143 | ||
4196d2f0 DL |
144 | size_t bdrv_min_mem_align(BlockDriverState *bs) |
145 | { | |
146 | if (!bs || !bs->drv) { | |
459b4e66 | 147 | /* page size or 4k (hdd sector size) should be on the safe side */ |
8e3b0cbb | 148 | return MAX(4096, qemu_real_host_page_size()); |
4196d2f0 | 149 | } |
384a48fb | 150 | IO_CODE(); |
4196d2f0 DL |
151 | |
152 | return bs->bl.min_mem_alignment; | |
153 | } | |
154 | ||
9e0b22f4 | 155 | /* check if the path starts with "<protocol>:" */ |
5c98415b | 156 | int path_has_protocol(const char *path) |
9e0b22f4 | 157 | { |
947995c0 PB |
158 | const char *p; |
159 | ||
9e0b22f4 SH |
160 | #ifdef _WIN32 |
161 | if (is_windows_drive(path) || | |
162 | is_windows_drive_prefix(path)) { | |
163 | return 0; | |
164 | } | |
947995c0 PB |
165 | p = path + strcspn(path, ":/\\"); |
166 | #else | |
167 | p = path + strcspn(path, ":/"); | |
9e0b22f4 SH |
168 | #endif |
169 | ||
947995c0 | 170 | return *p == ':'; |
9e0b22f4 SH |
171 | } |
172 | ||
83f64091 | 173 | int path_is_absolute(const char *path) |
3b0d4f61 | 174 | { |
21664424 FB |
175 | #ifdef _WIN32 |
176 | /* specific case for names like: "\\.\d:" */ | |
f53f4da9 | 177 | if (is_windows_drive(path) || is_windows_drive_prefix(path)) { |
21664424 | 178 | return 1; |
f53f4da9 PB |
179 | } |
180 | return (*path == '/' || *path == '\\'); | |
3b9f94e1 | 181 | #else |
f53f4da9 | 182 | return (*path == '/'); |
3b9f94e1 | 183 | #endif |
3b0d4f61 FB |
184 | } |
185 | ||
009b03aa | 186 | /* if filename is absolute, just return its duplicate. Otherwise, build a |
83f64091 FB |
187 | path to it by considering it is relative to base_path. URL are |
188 | supported. */ | |
009b03aa | 189 | char *path_combine(const char *base_path, const char *filename) |
3b0d4f61 | 190 | { |
009b03aa | 191 | const char *protocol_stripped = NULL; |
83f64091 | 192 | const char *p, *p1; |
009b03aa | 193 | char *result; |
83f64091 FB |
194 | int len; |
195 | ||
83f64091 | 196 | if (path_is_absolute(filename)) { |
009b03aa HR |
197 | return g_strdup(filename); |
198 | } | |
0d54a6fe | 199 | |
009b03aa HR |
200 | if (path_has_protocol(base_path)) { |
201 | protocol_stripped = strchr(base_path, ':'); | |
202 | if (protocol_stripped) { | |
203 | protocol_stripped++; | |
0d54a6fe | 204 | } |
009b03aa HR |
205 | } |
206 | p = protocol_stripped ?: base_path; | |
0d54a6fe | 207 | |
009b03aa | 208 | p1 = strrchr(base_path, '/'); |
3b9f94e1 | 209 | #ifdef _WIN32 |
009b03aa HR |
210 | { |
211 | const char *p2; | |
212 | p2 = strrchr(base_path, '\\'); | |
213 | if (!p1 || p2 > p1) { | |
214 | p1 = p2; | |
3b9f94e1 | 215 | } |
009b03aa | 216 | } |
3b9f94e1 | 217 | #endif |
009b03aa HR |
218 | if (p1) { |
219 | p1++; | |
220 | } else { | |
221 | p1 = base_path; | |
222 | } | |
223 | if (p1 > p) { | |
224 | p = p1; | |
3b0d4f61 | 225 | } |
009b03aa HR |
226 | len = p - base_path; |
227 | ||
228 | result = g_malloc(len + strlen(filename) + 1); | |
229 | memcpy(result, base_path, len); | |
230 | strcpy(result + len, filename); | |
231 | ||
232 | return result; | |
233 | } | |
234 | ||
03c320d8 HR |
235 | /* |
236 | * Helper function for bdrv_parse_filename() implementations to remove optional | |
237 | * protocol prefixes (especially "file:") from a filename and for putting the | |
238 | * stripped filename into the options QDict if there is such a prefix. | |
239 | */ | |
240 | void bdrv_parse_filename_strip_prefix(const char *filename, const char *prefix, | |
241 | QDict *options) | |
242 | { | |
243 | if (strstart(filename, prefix, &filename)) { | |
244 | /* Stripping the explicit protocol prefix may result in a protocol | |
245 | * prefix being (wrongly) detected (if the filename contains a colon) */ | |
246 | if (path_has_protocol(filename)) { | |
18cf67c5 | 247 | GString *fat_filename; |
03c320d8 HR |
248 | |
249 | /* This means there is some colon before the first slash; therefore, | |
250 | * this cannot be an absolute path */ | |
251 | assert(!path_is_absolute(filename)); | |
252 | ||
253 | /* And we can thus fix the protocol detection issue by prefixing it | |
254 | * by "./" */ | |
18cf67c5 MA |
255 | fat_filename = g_string_new("./"); |
256 | g_string_append(fat_filename, filename); | |
03c320d8 | 257 | |
18cf67c5 | 258 | assert(!path_has_protocol(fat_filename->str)); |
03c320d8 | 259 | |
18cf67c5 MA |
260 | qdict_put(options, "filename", |
261 | qstring_from_gstring(fat_filename)); | |
03c320d8 HR |
262 | } else { |
263 | /* If no protocol prefix was detected, we can use the shortened | |
264 | * filename as-is */ | |
265 | qdict_put_str(options, "filename", filename); | |
266 | } | |
267 | } | |
268 | } | |
269 | ||
270 | ||
9c5e6594 KW |
271 | /* Returns whether the image file is opened as read-only. Note that this can |
272 | * return false and writing to the image file is still not possible because the | |
273 | * image is inactivated. */ | |
93ed524e JC |
274 | bool bdrv_is_read_only(BlockDriverState *bs) |
275 | { | |
384a48fb | 276 | IO_CODE(); |
975da073 | 277 | return !(bs->open_flags & BDRV_O_RDWR); |
93ed524e JC |
278 | } |
279 | ||
10e5d707 KW |
280 | static int bdrv_can_set_read_only(BlockDriverState *bs, bool read_only, |
281 | bool ignore_allow_rdw, Error **errp) | |
fe5241bf | 282 | { |
384a48fb EGE |
283 | IO_CODE(); |
284 | ||
e2b8247a JC |
285 | /* Do not set read_only if copy_on_read is enabled */ |
286 | if (bs->copy_on_read && read_only) { | |
287 | error_setg(errp, "Can't set node '%s' to r/o with copy-on-read enabled", | |
288 | bdrv_get_device_or_node_name(bs)); | |
289 | return -EINVAL; | |
290 | } | |
291 | ||
d6fcdf06 | 292 | /* Do not clear read_only if it is prohibited */ |
54a32bfe KW |
293 | if (!read_only && !(bs->open_flags & BDRV_O_ALLOW_RDWR) && |
294 | !ignore_allow_rdw) | |
295 | { | |
d6fcdf06 JC |
296 | error_setg(errp, "Node '%s' is read only", |
297 | bdrv_get_device_or_node_name(bs)); | |
298 | return -EPERM; | |
299 | } | |
300 | ||
45803a03 JC |
301 | return 0; |
302 | } | |
303 | ||
eaa2410f KW |
304 | /* |
305 | * Called by a driver that can only provide a read-only image. | |
306 | * | |
307 | * Returns 0 if the node is already read-only or it could switch the node to | |
308 | * read-only because BDRV_O_AUTO_RDONLY is set. | |
309 | * | |
310 | * Returns -EACCES if the node is read-write and BDRV_O_AUTO_RDONLY is not set | |
311 | * or bdrv_can_set_read_only() forbids making the node read-only. If @errmsg | |
312 | * is not NULL, it is used as the error message for the Error object. | |
313 | */ | |
314 | int bdrv_apply_auto_read_only(BlockDriverState *bs, const char *errmsg, | |
315 | Error **errp) | |
45803a03 JC |
316 | { |
317 | int ret = 0; | |
384a48fb | 318 | IO_CODE(); |
45803a03 | 319 | |
eaa2410f KW |
320 | if (!(bs->open_flags & BDRV_O_RDWR)) { |
321 | return 0; | |
322 | } | |
323 | if (!(bs->open_flags & BDRV_O_AUTO_RDONLY)) { | |
324 | goto fail; | |
45803a03 JC |
325 | } |
326 | ||
eaa2410f KW |
327 | ret = bdrv_can_set_read_only(bs, true, false, NULL); |
328 | if (ret < 0) { | |
329 | goto fail; | |
eeae6a59 KW |
330 | } |
331 | ||
eaa2410f KW |
332 | bs->open_flags &= ~BDRV_O_RDWR; |
333 | ||
e2b8247a | 334 | return 0; |
eaa2410f KW |
335 | |
336 | fail: | |
337 | error_setg(errp, "%s", errmsg ?: "Image is read-only"); | |
338 | return -EACCES; | |
fe5241bf JC |
339 | } |
340 | ||
645ae7d8 HR |
341 | /* |
342 | * If @backing is empty, this function returns NULL without setting | |
343 | * @errp. In all other cases, NULL will only be returned with @errp | |
344 | * set. | |
345 | * | |
346 | * Therefore, a return value of NULL without @errp set means that | |
347 | * there is no backing file; if @errp is set, there is one but its | |
348 | * absolute filename cannot be generated. | |
349 | */ | |
350 | char *bdrv_get_full_backing_filename_from_filename(const char *backed, | |
351 | const char *backing, | |
352 | Error **errp) | |
dc5a1371 | 353 | { |
645ae7d8 HR |
354 | if (backing[0] == '\0') { |
355 | return NULL; | |
356 | } else if (path_has_protocol(backing) || path_is_absolute(backing)) { | |
357 | return g_strdup(backing); | |
9f07429e HR |
358 | } else if (backed[0] == '\0' || strstart(backed, "json:", NULL)) { |
359 | error_setg(errp, "Cannot use relative backing file names for '%s'", | |
360 | backed); | |
645ae7d8 | 361 | return NULL; |
dc5a1371 | 362 | } else { |
645ae7d8 | 363 | return path_combine(backed, backing); |
dc5a1371 PB |
364 | } |
365 | } | |
366 | ||
9f4793d8 HR |
367 | /* |
368 | * If @filename is empty or NULL, this function returns NULL without | |
369 | * setting @errp. In all other cases, NULL will only be returned with | |
370 | * @errp set. | |
371 | */ | |
372 | static char *bdrv_make_absolute_filename(BlockDriverState *relative_to, | |
373 | const char *filename, Error **errp) | |
0a82855a | 374 | { |
8df68616 | 375 | char *dir, *full_name; |
9f4793d8 | 376 | |
8df68616 HR |
377 | if (!filename || filename[0] == '\0') { |
378 | return NULL; | |
379 | } else if (path_has_protocol(filename) || path_is_absolute(filename)) { | |
380 | return g_strdup(filename); | |
381 | } | |
9f07429e | 382 | |
8df68616 HR |
383 | dir = bdrv_dirname(relative_to, errp); |
384 | if (!dir) { | |
385 | return NULL; | |
386 | } | |
f30c66ba | 387 | |
8df68616 HR |
388 | full_name = g_strconcat(dir, filename, NULL); |
389 | g_free(dir); | |
390 | return full_name; | |
9f4793d8 HR |
391 | } |
392 | ||
393 | char *bdrv_get_full_backing_filename(BlockDriverState *bs, Error **errp) | |
394 | { | |
f791bf7f | 395 | GLOBAL_STATE_CODE(); |
9f4793d8 | 396 | return bdrv_make_absolute_filename(bs, bs->backing_file, errp); |
0a82855a HR |
397 | } |
398 | ||
0eb7217e SH |
399 | void bdrv_register(BlockDriver *bdrv) |
400 | { | |
a15f08dc | 401 | assert(bdrv->format_name); |
f791bf7f | 402 | GLOBAL_STATE_CODE(); |
8a22f02a | 403 | QLIST_INSERT_HEAD(&bdrv_drivers, bdrv, list); |
ea2384d3 | 404 | } |
b338082b | 405 | |
e4e9986b MA |
406 | BlockDriverState *bdrv_new(void) |
407 | { | |
408 | BlockDriverState *bs; | |
409 | int i; | |
410 | ||
f791bf7f EGE |
411 | GLOBAL_STATE_CODE(); |
412 | ||
5839e53b | 413 | bs = g_new0(BlockDriverState, 1); |
e4654d2d | 414 | QLIST_INIT(&bs->dirty_bitmaps); |
fbe40ff7 FZ |
415 | for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) { |
416 | QLIST_INIT(&bs->op_blockers[i]); | |
417 | } | |
3783fa3d | 418 | qemu_co_mutex_init(&bs->reqs_lock); |
2119882c | 419 | qemu_mutex_init(&bs->dirty_bitmap_mutex); |
9fcb0251 | 420 | bs->refcnt = 1; |
dcd04228 | 421 | bs->aio_context = qemu_get_aio_context(); |
d7d512f6 | 422 | |
3ff2f67a EY |
423 | qemu_co_queue_init(&bs->flush_queue); |
424 | ||
0bc329fb HR |
425 | qemu_co_mutex_init(&bs->bsc_modify_lock); |
426 | bs->block_status_cache = g_new0(BdrvBlockStatusCache, 1); | |
427 | ||
0f12264e KW |
428 | for (i = 0; i < bdrv_drain_all_count; i++) { |
429 | bdrv_drained_begin(bs); | |
430 | } | |
431 | ||
2c1d04e0 HR |
432 | QTAILQ_INSERT_TAIL(&all_bdrv_states, bs, bs_list); |
433 | ||
b338082b FB |
434 | return bs; |
435 | } | |
436 | ||
88d88798 | 437 | static BlockDriver *bdrv_do_find_format(const char *format_name) |
ea2384d3 FB |
438 | { |
439 | BlockDriver *drv1; | |
bdb73476 | 440 | GLOBAL_STATE_CODE(); |
88d88798 | 441 | |
8a22f02a SH |
442 | QLIST_FOREACH(drv1, &bdrv_drivers, list) { |
443 | if (!strcmp(drv1->format_name, format_name)) { | |
ea2384d3 | 444 | return drv1; |
8a22f02a | 445 | } |
ea2384d3 | 446 | } |
88d88798 | 447 | |
ea2384d3 FB |
448 | return NULL; |
449 | } | |
450 | ||
88d88798 MM |
451 | BlockDriver *bdrv_find_format(const char *format_name) |
452 | { | |
453 | BlockDriver *drv1; | |
454 | int i; | |
455 | ||
f791bf7f EGE |
456 | GLOBAL_STATE_CODE(); |
457 | ||
88d88798 MM |
458 | drv1 = bdrv_do_find_format(format_name); |
459 | if (drv1) { | |
460 | return drv1; | |
461 | } | |
462 | ||
463 | /* The driver isn't registered, maybe we need to load a module */ | |
464 | for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); ++i) { | |
465 | if (!strcmp(block_driver_modules[i].format_name, format_name)) { | |
c551fb0b CF |
466 | Error *local_err = NULL; |
467 | int rv = block_module_load(block_driver_modules[i].library_name, | |
468 | &local_err); | |
469 | if (rv > 0) { | |
470 | return bdrv_do_find_format(format_name); | |
471 | } else if (rv < 0) { | |
472 | error_report_err(local_err); | |
473 | } | |
88d88798 MM |
474 | break; |
475 | } | |
476 | } | |
c551fb0b | 477 | return NULL; |
88d88798 MM |
478 | } |
479 | ||
9ac404c5 | 480 | static int bdrv_format_is_whitelisted(const char *format_name, bool read_only) |
eb852011 | 481 | { |
b64ec4e4 FZ |
482 | static const char *whitelist_rw[] = { |
483 | CONFIG_BDRV_RW_WHITELIST | |
859aef02 | 484 | NULL |
b64ec4e4 FZ |
485 | }; |
486 | static const char *whitelist_ro[] = { | |
487 | CONFIG_BDRV_RO_WHITELIST | |
859aef02 | 488 | NULL |
eb852011 MA |
489 | }; |
490 | const char **p; | |
491 | ||
b64ec4e4 | 492 | if (!whitelist_rw[0] && !whitelist_ro[0]) { |
eb852011 | 493 | return 1; /* no whitelist, anything goes */ |
b64ec4e4 | 494 | } |
eb852011 | 495 | |
b64ec4e4 | 496 | for (p = whitelist_rw; *p; p++) { |
9ac404c5 | 497 | if (!strcmp(format_name, *p)) { |
eb852011 MA |
498 | return 1; |
499 | } | |
500 | } | |
b64ec4e4 FZ |
501 | if (read_only) { |
502 | for (p = whitelist_ro; *p; p++) { | |
9ac404c5 | 503 | if (!strcmp(format_name, *p)) { |
b64ec4e4 FZ |
504 | return 1; |
505 | } | |
506 | } | |
507 | } | |
eb852011 MA |
508 | return 0; |
509 | } | |
510 | ||
9ac404c5 AS |
511 | int bdrv_is_whitelisted(BlockDriver *drv, bool read_only) |
512 | { | |
f791bf7f | 513 | GLOBAL_STATE_CODE(); |
9ac404c5 AS |
514 | return bdrv_format_is_whitelisted(drv->format_name, read_only); |
515 | } | |
516 | ||
e6ff69bf DB |
517 | bool bdrv_uses_whitelist(void) |
518 | { | |
519 | return use_bdrv_whitelist; | |
520 | } | |
521 | ||
5b7e1542 ZYW |
522 | typedef struct CreateCo { |
523 | BlockDriver *drv; | |
524 | char *filename; | |
83d0521a | 525 | QemuOpts *opts; |
5b7e1542 | 526 | int ret; |
cc84d90f | 527 | Error *err; |
5b7e1542 ZYW |
528 | } CreateCo; |
529 | ||
741443eb EGE |
530 | int coroutine_fn bdrv_co_create(BlockDriver *drv, const char *filename, |
531 | QemuOpts *opts, Error **errp) | |
5b7e1542 | 532 | { |
cc84d90f | 533 | int ret; |
84bdf21f EGE |
534 | GLOBAL_STATE_CODE(); |
535 | ERRP_GUARD(); | |
536 | ||
537 | if (!drv->bdrv_co_create_opts) { | |
538 | error_setg(errp, "Driver '%s' does not support image creation", | |
539 | drv->format_name); | |
540 | return -ENOTSUP; | |
541 | } | |
542 | ||
543 | ret = drv->bdrv_co_create_opts(drv, filename, opts, errp); | |
544 | if (ret < 0 && !*errp) { | |
545 | error_setg_errno(errp, -ret, "Could not create image"); | |
546 | } | |
cc84d90f | 547 | |
84bdf21f EGE |
548 | return ret; |
549 | } | |
550 | ||
fd17146c HR |
551 | /** |
552 | * Helper function for bdrv_create_file_fallback(): Resize @blk to at | |
553 | * least the given @minimum_size. | |
554 | * | |
555 | * On success, return @blk's actual length. | |
556 | * Otherwise, return -errno. | |
557 | */ | |
558 | static int64_t create_file_fallback_truncate(BlockBackend *blk, | |
559 | int64_t minimum_size, Error **errp) | |
84a12e66 | 560 | { |
cc84d90f | 561 | Error *local_err = NULL; |
fd17146c | 562 | int64_t size; |
cc84d90f | 563 | int ret; |
84a12e66 | 564 | |
bdb73476 EGE |
565 | GLOBAL_STATE_CODE(); |
566 | ||
8c6242b6 KW |
567 | ret = blk_truncate(blk, minimum_size, false, PREALLOC_MODE_OFF, 0, |
568 | &local_err); | |
fd17146c HR |
569 | if (ret < 0 && ret != -ENOTSUP) { |
570 | error_propagate(errp, local_err); | |
571 | return ret; | |
572 | } | |
573 | ||
574 | size = blk_getlength(blk); | |
575 | if (size < 0) { | |
576 | error_free(local_err); | |
577 | error_setg_errno(errp, -size, | |
578 | "Failed to inquire the new image file's length"); | |
579 | return size; | |
580 | } | |
581 | ||
582 | if (size < minimum_size) { | |
583 | /* Need to grow the image, but we failed to do that */ | |
584 | error_propagate(errp, local_err); | |
585 | return -ENOTSUP; | |
586 | } | |
587 | ||
588 | error_free(local_err); | |
589 | local_err = NULL; | |
590 | ||
591 | return size; | |
592 | } | |
593 | ||
594 | /** | |
595 | * Helper function for bdrv_create_file_fallback(): Zero the first | |
596 | * sector to remove any potentially pre-existing image header. | |
597 | */ | |
881a4c55 PB |
598 | static int coroutine_fn |
599 | create_file_fallback_zero_first_sector(BlockBackend *blk, | |
600 | int64_t current_size, | |
601 | Error **errp) | |
fd17146c HR |
602 | { |
603 | int64_t bytes_to_clear; | |
604 | int ret; | |
605 | ||
bdb73476 EGE |
606 | GLOBAL_STATE_CODE(); |
607 | ||
fd17146c HR |
608 | bytes_to_clear = MIN(current_size, BDRV_SECTOR_SIZE); |
609 | if (bytes_to_clear) { | |
ce47ff20 | 610 | ret = blk_co_pwrite_zeroes(blk, 0, bytes_to_clear, BDRV_REQ_MAY_UNMAP); |
fd17146c HR |
611 | if (ret < 0) { |
612 | error_setg_errno(errp, -ret, | |
613 | "Failed to clear the new image's first sector"); | |
614 | return ret; | |
615 | } | |
616 | } | |
617 | ||
618 | return 0; | |
619 | } | |
620 | ||
5a5e7f8c ML |
621 | /** |
622 | * Simple implementation of bdrv_co_create_opts for protocol drivers | |
623 | * which only support creation via opening a file | |
624 | * (usually existing raw storage device) | |
625 | */ | |
626 | int coroutine_fn bdrv_co_create_opts_simple(BlockDriver *drv, | |
627 | const char *filename, | |
628 | QemuOpts *opts, | |
629 | Error **errp) | |
fd17146c HR |
630 | { |
631 | BlockBackend *blk; | |
eeea1faa | 632 | QDict *options; |
fd17146c HR |
633 | int64_t size = 0; |
634 | char *buf = NULL; | |
635 | PreallocMode prealloc; | |
636 | Error *local_err = NULL; | |
637 | int ret; | |
638 | ||
b4ad82aa EGE |
639 | GLOBAL_STATE_CODE(); |
640 | ||
fd17146c HR |
641 | size = qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0); |
642 | buf = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC); | |
643 | prealloc = qapi_enum_parse(&PreallocMode_lookup, buf, | |
644 | PREALLOC_MODE_OFF, &local_err); | |
645 | g_free(buf); | |
646 | if (local_err) { | |
647 | error_propagate(errp, local_err); | |
648 | return -EINVAL; | |
649 | } | |
650 | ||
651 | if (prealloc != PREALLOC_MODE_OFF) { | |
652 | error_setg(errp, "Unsupported preallocation mode '%s'", | |
653 | PreallocMode_str(prealloc)); | |
654 | return -ENOTSUP; | |
655 | } | |
656 | ||
eeea1faa | 657 | options = qdict_new(); |
fd17146c HR |
658 | qdict_put_str(options, "driver", drv->format_name); |
659 | ||
be1a732c KW |
660 | blk = blk_co_new_open(filename, NULL, options, |
661 | BDRV_O_RDWR | BDRV_O_RESIZE, errp); | |
fd17146c HR |
662 | if (!blk) { |
663 | error_prepend(errp, "Protocol driver '%s' does not support image " | |
664 | "creation, and opening the image failed: ", | |
665 | drv->format_name); | |
666 | return -EINVAL; | |
667 | } | |
668 | ||
669 | size = create_file_fallback_truncate(blk, size, errp); | |
670 | if (size < 0) { | |
671 | ret = size; | |
672 | goto out; | |
673 | } | |
674 | ||
675 | ret = create_file_fallback_zero_first_sector(blk, size, errp); | |
676 | if (ret < 0) { | |
677 | goto out; | |
678 | } | |
679 | ||
680 | ret = 0; | |
681 | out: | |
b2ab5f54 | 682 | blk_co_unref(blk); |
fd17146c HR |
683 | return ret; |
684 | } | |
685 | ||
2475a0d0 EGE |
686 | int coroutine_fn bdrv_co_create_file(const char *filename, QemuOpts *opts, |
687 | Error **errp) | |
fd17146c | 688 | { |
729222af | 689 | QemuOpts *protocol_opts; |
fd17146c | 690 | BlockDriver *drv; |
729222af SG |
691 | QDict *qdict; |
692 | int ret; | |
fd17146c | 693 | |
f791bf7f EGE |
694 | GLOBAL_STATE_CODE(); |
695 | ||
b65a5e12 | 696 | drv = bdrv_find_protocol(filename, true, errp); |
84a12e66 | 697 | if (drv == NULL) { |
16905d71 | 698 | return -ENOENT; |
84a12e66 CH |
699 | } |
700 | ||
729222af SG |
701 | if (!drv->create_opts) { |
702 | error_setg(errp, "Driver '%s' does not support image creation", | |
703 | drv->format_name); | |
704 | return -ENOTSUP; | |
705 | } | |
706 | ||
707 | /* | |
708 | * 'opts' contains a QemuOptsList with a combination of format and protocol | |
709 | * default values. | |
710 | * | |
711 | * The format properly removes its options, but the default values remain | |
712 | * in 'opts->list'. So if the protocol has options with the same name | |
713 | * (e.g. rbd has 'cluster_size' as qcow2), it will see the default values | |
714 | * of the format, since for overlapping options, the format wins. | |
715 | * | |
716 | * To avoid this issue, lets convert QemuOpts to QDict, in this way we take | |
717 | * only the set options, and then convert it back to QemuOpts, using the | |
718 | * create_opts of the protocol. So the new QemuOpts, will contain only the | |
719 | * protocol defaults. | |
720 | */ | |
721 | qdict = qemu_opts_to_qdict(opts, NULL); | |
722 | protocol_opts = qemu_opts_from_qdict(drv->create_opts, qdict, errp); | |
723 | if (protocol_opts == NULL) { | |
724 | ret = -EINVAL; | |
725 | goto out; | |
726 | } | |
727 | ||
2475a0d0 | 728 | ret = bdrv_co_create(drv, filename, protocol_opts, errp); |
729222af SG |
729 | out: |
730 | qemu_opts_del(protocol_opts); | |
731 | qobject_unref(qdict); | |
732 | return ret; | |
84a12e66 CH |
733 | } |
734 | ||
e1d7f8bb DHB |
735 | int coroutine_fn bdrv_co_delete_file(BlockDriverState *bs, Error **errp) |
736 | { | |
737 | Error *local_err = NULL; | |
738 | int ret; | |
739 | ||
384a48fb | 740 | IO_CODE(); |
e1d7f8bb | 741 | assert(bs != NULL); |
48aef794 | 742 | assert_bdrv_graph_readable(); |
e1d7f8bb DHB |
743 | |
744 | if (!bs->drv) { | |
745 | error_setg(errp, "Block node '%s' is not opened", bs->filename); | |
746 | return -ENOMEDIUM; | |
747 | } | |
748 | ||
749 | if (!bs->drv->bdrv_co_delete_file) { | |
750 | error_setg(errp, "Driver '%s' does not support image deletion", | |
751 | bs->drv->format_name); | |
752 | return -ENOTSUP; | |
753 | } | |
754 | ||
755 | ret = bs->drv->bdrv_co_delete_file(bs, &local_err); | |
756 | if (ret < 0) { | |
757 | error_propagate(errp, local_err); | |
758 | } | |
759 | ||
760 | return ret; | |
761 | } | |
762 | ||
a890f08e ML |
763 | void coroutine_fn bdrv_co_delete_file_noerr(BlockDriverState *bs) |
764 | { | |
765 | Error *local_err = NULL; | |
766 | int ret; | |
384a48fb | 767 | IO_CODE(); |
a890f08e ML |
768 | |
769 | if (!bs) { | |
770 | return; | |
771 | } | |
772 | ||
773 | ret = bdrv_co_delete_file(bs, &local_err); | |
774 | /* | |
775 | * ENOTSUP will happen if the block driver doesn't support | |
776 | * the 'bdrv_co_delete_file' interface. This is a predictable | |
777 | * scenario and shouldn't be reported back to the user. | |
778 | */ | |
779 | if (ret == -ENOTSUP) { | |
780 | error_free(local_err); | |
781 | } else if (ret < 0) { | |
782 | error_report_err(local_err); | |
783 | } | |
784 | } | |
785 | ||
892b7de8 ET |
786 | /** |
787 | * Try to get @bs's logical and physical block size. | |
788 | * On success, store them in @bsz struct and return 0. | |
789 | * On failure return -errno. | |
790 | * @bs must not be empty. | |
791 | */ | |
792 | int bdrv_probe_blocksizes(BlockDriverState *bs, BlockSizes *bsz) | |
793 | { | |
794 | BlockDriver *drv = bs->drv; | |
93393e69 | 795 | BlockDriverState *filtered = bdrv_filter_bs(bs); |
f791bf7f | 796 | GLOBAL_STATE_CODE(); |
892b7de8 ET |
797 | |
798 | if (drv && drv->bdrv_probe_blocksizes) { | |
799 | return drv->bdrv_probe_blocksizes(bs, bsz); | |
93393e69 HR |
800 | } else if (filtered) { |
801 | return bdrv_probe_blocksizes(filtered, bsz); | |
892b7de8 ET |
802 | } |
803 | ||
804 | return -ENOTSUP; | |
805 | } | |
806 | ||
807 | /** | |
808 | * Try to get @bs's geometry (cyls, heads, sectors). | |
809 | * On success, store them in @geo struct and return 0. | |
810 | * On failure return -errno. | |
811 | * @bs must not be empty. | |
812 | */ | |
813 | int bdrv_probe_geometry(BlockDriverState *bs, HDGeometry *geo) | |
814 | { | |
815 | BlockDriver *drv = bs->drv; | |
93393e69 | 816 | BlockDriverState *filtered = bdrv_filter_bs(bs); |
f791bf7f | 817 | GLOBAL_STATE_CODE(); |
892b7de8 ET |
818 | |
819 | if (drv && drv->bdrv_probe_geometry) { | |
820 | return drv->bdrv_probe_geometry(bs, geo); | |
93393e69 HR |
821 | } else if (filtered) { |
822 | return bdrv_probe_geometry(filtered, geo); | |
892b7de8 ET |
823 | } |
824 | ||
825 | return -ENOTSUP; | |
826 | } | |
827 | ||
eba25057 JM |
828 | /* |
829 | * Create a uniquely-named empty temporary file. | |
69fbfff9 BM |
830 | * Return the actual file name used upon success, otherwise NULL. |
831 | * This string should be freed with g_free() when not needed any longer. | |
832 | * | |
833 | * Note: creating a temporary file for the caller to (re)open is | |
834 | * inherently racy. Use g_file_open_tmp() instead whenever practical. | |
eba25057 | 835 | */ |
69fbfff9 | 836 | char *create_tmp_file(Error **errp) |
d5249393 | 837 | { |
67b915a5 | 838 | int fd; |
7ccfb2eb | 839 | const char *tmpdir; |
69fbfff9 BM |
840 | g_autofree char *filename = NULL; |
841 | ||
842 | tmpdir = g_get_tmp_dir(); | |
843 | #ifndef _WIN32 | |
844 | /* | |
845 | * See commit 69bef79 ("block: use /var/tmp instead of /tmp for -snapshot") | |
846 | * | |
847 | * This function is used to create temporary disk images (like -snapshot), | |
848 | * so the files can become very large. /tmp is often a tmpfs where as | |
849 | * /var/tmp is usually on a disk, so more appropriate for disk images. | |
850 | */ | |
851 | if (!g_strcmp0(tmpdir, "/tmp")) { | |
69bef793 AS |
852 | tmpdir = "/var/tmp"; |
853 | } | |
69fbfff9 BM |
854 | #endif |
855 | ||
856 | filename = g_strdup_printf("%s/vl.XXXXXX", tmpdir); | |
857 | fd = g_mkstemp(filename); | |
fe235a06 | 858 | if (fd < 0) { |
69fbfff9 BM |
859 | error_setg_errno(errp, errno, "Could not open temporary file '%s'", |
860 | filename); | |
861 | return NULL; | |
eba25057 | 862 | } |
6b6471ee | 863 | close(fd); |
69fbfff9 BM |
864 | |
865 | return g_steal_pointer(&filename); | |
eba25057 | 866 | } |
fc01f7e7 | 867 | |
84a12e66 CH |
868 | /* |
869 | * Detect host devices. By convention, /dev/cdrom[N] is always | |
870 | * recognized as a host CDROM. | |
871 | */ | |
872 | static BlockDriver *find_hdev_driver(const char *filename) | |
873 | { | |
874 | int score_max = 0, score; | |
875 | BlockDriver *drv = NULL, *d; | |
bdb73476 | 876 | GLOBAL_STATE_CODE(); |
84a12e66 CH |
877 | |
878 | QLIST_FOREACH(d, &bdrv_drivers, list) { | |
879 | if (d->bdrv_probe_device) { | |
880 | score = d->bdrv_probe_device(filename); | |
881 | if (score > score_max) { | |
882 | score_max = score; | |
883 | drv = d; | |
884 | } | |
885 | } | |
886 | } | |
887 | ||
888 | return drv; | |
889 | } | |
890 | ||
88d88798 MM |
891 | static BlockDriver *bdrv_do_find_protocol(const char *protocol) |
892 | { | |
893 | BlockDriver *drv1; | |
bdb73476 | 894 | GLOBAL_STATE_CODE(); |
88d88798 MM |
895 | |
896 | QLIST_FOREACH(drv1, &bdrv_drivers, list) { | |
897 | if (drv1->protocol_name && !strcmp(drv1->protocol_name, protocol)) { | |
898 | return drv1; | |
899 | } | |
900 | } | |
901 | ||
902 | return NULL; | |
903 | } | |
904 | ||
98289620 | 905 | BlockDriver *bdrv_find_protocol(const char *filename, |
b65a5e12 HR |
906 | bool allow_protocol_prefix, |
907 | Error **errp) | |
83f64091 FB |
908 | { |
909 | BlockDriver *drv1; | |
910 | char protocol[128]; | |
1cec71e3 | 911 | int len; |
83f64091 | 912 | const char *p; |
88d88798 | 913 | int i; |
19cb3738 | 914 | |
f791bf7f | 915 | GLOBAL_STATE_CODE(); |
66f82cee KW |
916 | /* TODO Drivers without bdrv_file_open must be specified explicitly */ |
917 | ||
39508e7a CH |
918 | /* |
919 | * XXX(hch): we really should not let host device detection | |
920 | * override an explicit protocol specification, but moving this | |
921 | * later breaks access to device names with colons in them. | |
922 | * Thanks to the brain-dead persistent naming schemes on udev- | |
923 | * based Linux systems those actually are quite common. | |
924 | */ | |
925 | drv1 = find_hdev_driver(filename); | |
926 | if (drv1) { | |
927 | return drv1; | |
928 | } | |
929 | ||
98289620 | 930 | if (!path_has_protocol(filename) || !allow_protocol_prefix) { |
ef810437 | 931 | return &bdrv_file; |
84a12e66 | 932 | } |
98289620 | 933 | |
9e0b22f4 SH |
934 | p = strchr(filename, ':'); |
935 | assert(p != NULL); | |
1cec71e3 AL |
936 | len = p - filename; |
937 | if (len > sizeof(protocol) - 1) | |
938 | len = sizeof(protocol) - 1; | |
939 | memcpy(protocol, filename, len); | |
940 | protocol[len] = '\0'; | |
88d88798 MM |
941 | |
942 | drv1 = bdrv_do_find_protocol(protocol); | |
943 | if (drv1) { | |
944 | return drv1; | |
945 | } | |
946 | ||
947 | for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); ++i) { | |
948 | if (block_driver_modules[i].protocol_name && | |
949 | !strcmp(block_driver_modules[i].protocol_name, protocol)) { | |
c551fb0b CF |
950 | int rv = block_module_load(block_driver_modules[i].library_name, errp); |
951 | if (rv > 0) { | |
952 | drv1 = bdrv_do_find_protocol(protocol); | |
953 | } else if (rv < 0) { | |
954 | return NULL; | |
955 | } | |
88d88798 | 956 | break; |
8a22f02a | 957 | } |
83f64091 | 958 | } |
b65a5e12 | 959 | |
88d88798 MM |
960 | if (!drv1) { |
961 | error_setg(errp, "Unknown protocol '%s'", protocol); | |
962 | } | |
963 | return drv1; | |
83f64091 FB |
964 | } |
965 | ||
c6684249 MA |
966 | /* |
967 | * Guess image format by probing its contents. | |
968 | * This is not a good idea when your image is raw (CVE-2008-2004), but | |
969 | * we do it anyway for backward compatibility. | |
970 | * | |
971 | * @buf contains the image's first @buf_size bytes. | |
7cddd372 KW |
972 | * @buf_size is the buffer size in bytes (generally BLOCK_PROBE_BUF_SIZE, |
973 | * but can be smaller if the image file is smaller) | |
c6684249 MA |
974 | * @filename is its filename. |
975 | * | |
976 | * For all block drivers, call the bdrv_probe() method to get its | |
977 | * probing score. | |
978 | * Return the first block driver with the highest probing score. | |
979 | */ | |
38f3ef57 KW |
980 | BlockDriver *bdrv_probe_all(const uint8_t *buf, int buf_size, |
981 | const char *filename) | |
c6684249 MA |
982 | { |
983 | int score_max = 0, score; | |
984 | BlockDriver *drv = NULL, *d; | |
967d7905 | 985 | IO_CODE(); |
c6684249 MA |
986 | |
987 | QLIST_FOREACH(d, &bdrv_drivers, list) { | |
988 | if (d->bdrv_probe) { | |
989 | score = d->bdrv_probe(buf, buf_size, filename); | |
990 | if (score > score_max) { | |
991 | score_max = score; | |
992 | drv = d; | |
993 | } | |
994 | } | |
995 | } | |
996 | ||
997 | return drv; | |
998 | } | |
999 | ||
5696c6e3 | 1000 | static int find_image_format(BlockBackend *file, const char *filename, |
34b5d2c6 | 1001 | BlockDriver **pdrv, Error **errp) |
f3a5d3f8 | 1002 | { |
c6684249 | 1003 | BlockDriver *drv; |
7cddd372 | 1004 | uint8_t buf[BLOCK_PROBE_BUF_SIZE]; |
f500a6d3 | 1005 | int ret = 0; |
f8ea0b00 | 1006 | |
bdb73476 EGE |
1007 | GLOBAL_STATE_CODE(); |
1008 | ||
08a00559 | 1009 | /* Return the raw BlockDriver * to scsi-generic devices or empty drives */ |
5696c6e3 | 1010 | if (blk_is_sg(file) || !blk_is_inserted(file) || blk_getlength(file) == 0) { |
ef810437 | 1011 | *pdrv = &bdrv_raw; |
c98ac35d | 1012 | return ret; |
1a396859 | 1013 | } |
f8ea0b00 | 1014 | |
a9262f55 | 1015 | ret = blk_pread(file, 0, sizeof(buf), buf, 0); |
83f64091 | 1016 | if (ret < 0) { |
34b5d2c6 HR |
1017 | error_setg_errno(errp, -ret, "Could not read image for determining its " |
1018 | "format"); | |
c98ac35d SW |
1019 | *pdrv = NULL; |
1020 | return ret; | |
83f64091 FB |
1021 | } |
1022 | ||
bf5b16fa | 1023 | drv = bdrv_probe_all(buf, sizeof(buf), filename); |
c98ac35d | 1024 | if (!drv) { |
34b5d2c6 HR |
1025 | error_setg(errp, "Could not determine image format: No compatible " |
1026 | "driver found"); | |
bf5b16fa AF |
1027 | *pdrv = NULL; |
1028 | return -ENOENT; | |
c98ac35d | 1029 | } |
bf5b16fa | 1030 | |
c98ac35d | 1031 | *pdrv = drv; |
bf5b16fa | 1032 | return 0; |
ea2384d3 FB |
1033 | } |
1034 | ||
51762288 SH |
1035 | /** |
1036 | * Set the current 'total_sectors' value | |
65a9bb25 | 1037 | * Return 0 on success, -errno on error. |
51762288 | 1038 | */ |
c86422c5 EGE |
1039 | int coroutine_fn bdrv_co_refresh_total_sectors(BlockDriverState *bs, |
1040 | int64_t hint) | |
51762288 SH |
1041 | { |
1042 | BlockDriver *drv = bs->drv; | |
967d7905 | 1043 | IO_CODE(); |
8ab8140a | 1044 | assert_bdrv_graph_readable(); |
51762288 | 1045 | |
d470ad42 HR |
1046 | if (!drv) { |
1047 | return -ENOMEDIUM; | |
1048 | } | |
1049 | ||
c86422c5 | 1050 | /* Do not attempt drv->bdrv_co_getlength() on scsi-generic devices */ |
b192af8a | 1051 | if (bdrv_is_sg(bs)) |
396759ad NB |
1052 | return 0; |
1053 | ||
51762288 | 1054 | /* query actual device if possible, otherwise just trust the hint */ |
c86422c5 EGE |
1055 | if (drv->bdrv_co_getlength) { |
1056 | int64_t length = drv->bdrv_co_getlength(bs); | |
51762288 SH |
1057 | if (length < 0) { |
1058 | return length; | |
1059 | } | |
7e382003 | 1060 | hint = DIV_ROUND_UP(length, BDRV_SECTOR_SIZE); |
51762288 SH |
1061 | } |
1062 | ||
1063 | bs->total_sectors = hint; | |
8b117001 VSO |
1064 | |
1065 | if (bs->total_sectors * BDRV_SECTOR_SIZE > BDRV_MAX_LENGTH) { | |
1066 | return -EFBIG; | |
1067 | } | |
1068 | ||
51762288 SH |
1069 | return 0; |
1070 | } | |
1071 | ||
cddff5ba KW |
1072 | /** |
1073 | * Combines a QDict of new block driver @options with any missing options taken | |
1074 | * from @old_options, so that leaving out an option defaults to its old value. | |
1075 | */ | |
1076 | static void bdrv_join_options(BlockDriverState *bs, QDict *options, | |
1077 | QDict *old_options) | |
1078 | { | |
da359909 | 1079 | GLOBAL_STATE_CODE(); |
cddff5ba KW |
1080 | if (bs->drv && bs->drv->bdrv_join_options) { |
1081 | bs->drv->bdrv_join_options(options, old_options); | |
1082 | } else { | |
1083 | qdict_join(options, old_options, false); | |
1084 | } | |
1085 | } | |
1086 | ||
543770bd AG |
1087 | static BlockdevDetectZeroesOptions bdrv_parse_detect_zeroes(QemuOpts *opts, |
1088 | int open_flags, | |
1089 | Error **errp) | |
1090 | { | |
1091 | Error *local_err = NULL; | |
1092 | char *value = qemu_opt_get_del(opts, "detect-zeroes"); | |
1093 | BlockdevDetectZeroesOptions detect_zeroes = | |
1094 | qapi_enum_parse(&BlockdevDetectZeroesOptions_lookup, value, | |
1095 | BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF, &local_err); | |
bdb73476 | 1096 | GLOBAL_STATE_CODE(); |
543770bd AG |
1097 | g_free(value); |
1098 | if (local_err) { | |
1099 | error_propagate(errp, local_err); | |
1100 | return detect_zeroes; | |
1101 | } | |
1102 | ||
1103 | if (detect_zeroes == BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP && | |
1104 | !(open_flags & BDRV_O_UNMAP)) | |
1105 | { | |
1106 | error_setg(errp, "setting detect-zeroes to unmap is not allowed " | |
1107 | "without setting discard operation to unmap"); | |
1108 | } | |
1109 | ||
1110 | return detect_zeroes; | |
1111 | } | |
1112 | ||
f80f2673 AM |
1113 | /** |
1114 | * Set open flags for aio engine | |
1115 | * | |
1116 | * Return 0 on success, -1 if the engine specified is invalid | |
1117 | */ | |
1118 | int bdrv_parse_aio(const char *mode, int *flags) | |
1119 | { | |
1120 | if (!strcmp(mode, "threads")) { | |
1121 | /* do nothing, default */ | |
1122 | } else if (!strcmp(mode, "native")) { | |
1123 | *flags |= BDRV_O_NATIVE_AIO; | |
1124 | #ifdef CONFIG_LINUX_IO_URING | |
1125 | } else if (!strcmp(mode, "io_uring")) { | |
1126 | *flags |= BDRV_O_IO_URING; | |
1127 | #endif | |
1128 | } else { | |
1129 | return -1; | |
1130 | } | |
1131 | ||
1132 | return 0; | |
1133 | } | |
1134 | ||
9e8f1835 PB |
1135 | /** |
1136 | * Set open flags for a given discard mode | |
1137 | * | |
1138 | * Return 0 on success, -1 if the discard mode was invalid. | |
1139 | */ | |
1140 | int bdrv_parse_discard_flags(const char *mode, int *flags) | |
1141 | { | |
1142 | *flags &= ~BDRV_O_UNMAP; | |
1143 | ||
1144 | if (!strcmp(mode, "off") || !strcmp(mode, "ignore")) { | |
1145 | /* do nothing */ | |
1146 | } else if (!strcmp(mode, "on") || !strcmp(mode, "unmap")) { | |
1147 | *flags |= BDRV_O_UNMAP; | |
1148 | } else { | |
1149 | return -1; | |
1150 | } | |
1151 | ||
1152 | return 0; | |
1153 | } | |
1154 | ||
c3993cdc SH |
1155 | /** |
1156 | * Set open flags for a given cache mode | |
1157 | * | |
1158 | * Return 0 on success, -1 if the cache mode was invalid. | |
1159 | */ | |
53e8ae01 | 1160 | int bdrv_parse_cache_mode(const char *mode, int *flags, bool *writethrough) |
c3993cdc SH |
1161 | { |
1162 | *flags &= ~BDRV_O_CACHE_MASK; | |
1163 | ||
1164 | if (!strcmp(mode, "off") || !strcmp(mode, "none")) { | |
53e8ae01 KW |
1165 | *writethrough = false; |
1166 | *flags |= BDRV_O_NOCACHE; | |
92196b2f | 1167 | } else if (!strcmp(mode, "directsync")) { |
53e8ae01 | 1168 | *writethrough = true; |
92196b2f | 1169 | *flags |= BDRV_O_NOCACHE; |
c3993cdc | 1170 | } else if (!strcmp(mode, "writeback")) { |
53e8ae01 | 1171 | *writethrough = false; |
c3993cdc | 1172 | } else if (!strcmp(mode, "unsafe")) { |
53e8ae01 | 1173 | *writethrough = false; |
c3993cdc SH |
1174 | *flags |= BDRV_O_NO_FLUSH; |
1175 | } else if (!strcmp(mode, "writethrough")) { | |
53e8ae01 | 1176 | *writethrough = true; |
c3993cdc SH |
1177 | } else { |
1178 | return -1; | |
1179 | } | |
1180 | ||
1181 | return 0; | |
1182 | } | |
1183 | ||
b5411555 KW |
1184 | static char *bdrv_child_get_parent_desc(BdrvChild *c) |
1185 | { | |
1186 | BlockDriverState *parent = c->opaque; | |
2c0a3acb | 1187 | return g_strdup_printf("node '%s'", bdrv_get_node_name(parent)); |
b5411555 KW |
1188 | } |
1189 | ||
20018e12 KW |
1190 | static void bdrv_child_cb_drained_begin(BdrvChild *child) |
1191 | { | |
1192 | BlockDriverState *bs = child->opaque; | |
a82a3bd1 | 1193 | bdrv_do_drained_begin_quiesce(bs, NULL); |
20018e12 KW |
1194 | } |
1195 | ||
89bd0305 KW |
1196 | static bool bdrv_child_cb_drained_poll(BdrvChild *child) |
1197 | { | |
1198 | BlockDriverState *bs = child->opaque; | |
299403ae | 1199 | return bdrv_drain_poll(bs, NULL, false); |
89bd0305 KW |
1200 | } |
1201 | ||
2f65df6e | 1202 | static void bdrv_child_cb_drained_end(BdrvChild *child) |
20018e12 KW |
1203 | { |
1204 | BlockDriverState *bs = child->opaque; | |
2f65df6e | 1205 | bdrv_drained_end(bs); |
20018e12 KW |
1206 | } |
1207 | ||
38701b6a KW |
1208 | static int bdrv_child_cb_inactivate(BdrvChild *child) |
1209 | { | |
1210 | BlockDriverState *bs = child->opaque; | |
bdb73476 | 1211 | GLOBAL_STATE_CODE(); |
38701b6a KW |
1212 | assert(bs->open_flags & BDRV_O_INACTIVE); |
1213 | return 0; | |
1214 | } | |
1215 | ||
27633e74 EGE |
1216 | static bool bdrv_child_cb_change_aio_ctx(BdrvChild *child, AioContext *ctx, |
1217 | GHashTable *visited, Transaction *tran, | |
1218 | Error **errp) | |
53a7d041 KW |
1219 | { |
1220 | BlockDriverState *bs = child->opaque; | |
27633e74 | 1221 | return bdrv_change_aio_context(bs, ctx, visited, tran, errp); |
53a7d041 KW |
1222 | } |
1223 | ||
b1e6fc08 | 1224 | /* |
73176bee KW |
1225 | * Returns the options and flags that a temporary snapshot should get, based on |
1226 | * the originally requested flags (the originally requested image will have | |
1227 | * flags like a backing file) | |
b1e6fc08 | 1228 | */ |
73176bee KW |
1229 | static void bdrv_temp_snapshot_options(int *child_flags, QDict *child_options, |
1230 | int parent_flags, QDict *parent_options) | |
b1e6fc08 | 1231 | { |
bdb73476 | 1232 | GLOBAL_STATE_CODE(); |
73176bee KW |
1233 | *child_flags = (parent_flags & ~BDRV_O_SNAPSHOT) | BDRV_O_TEMPORARY; |
1234 | ||
1235 | /* For temporary files, unconditional cache=unsafe is fine */ | |
73176bee KW |
1236 | qdict_set_default_str(child_options, BDRV_OPT_CACHE_DIRECT, "off"); |
1237 | qdict_set_default_str(child_options, BDRV_OPT_CACHE_NO_FLUSH, "on"); | |
41869044 | 1238 | |
3f48686f | 1239 | /* Copy the read-only and discard options from the parent */ |
f87a0e29 | 1240 | qdict_copy_default(child_options, parent_options, BDRV_OPT_READ_ONLY); |
3f48686f | 1241 | qdict_copy_default(child_options, parent_options, BDRV_OPT_DISCARD); |
f87a0e29 | 1242 | |
41869044 KW |
1243 | /* aio=native doesn't work for cache.direct=off, so disable it for the |
1244 | * temporary snapshot */ | |
1245 | *child_flags &= ~BDRV_O_NATIVE_AIO; | |
b1e6fc08 KW |
1246 | } |
1247 | ||
db95dbba KW |
1248 | static void bdrv_backing_attach(BdrvChild *c) |
1249 | { | |
1250 | BlockDriverState *parent = c->opaque; | |
1251 | BlockDriverState *backing_hd = c->bs; | |
1252 | ||
bdb73476 | 1253 | GLOBAL_STATE_CODE(); |
db95dbba KW |
1254 | assert(!parent->backing_blocker); |
1255 | error_setg(&parent->backing_blocker, | |
1256 | "node is used as backing hd of '%s'", | |
1257 | bdrv_get_device_or_node_name(parent)); | |
1258 | ||
f30c66ba HR |
1259 | bdrv_refresh_filename(backing_hd); |
1260 | ||
db95dbba | 1261 | parent->open_flags &= ~BDRV_O_NO_BACKING; |
db95dbba KW |
1262 | |
1263 | bdrv_op_block_all(backing_hd, parent->backing_blocker); | |
1264 | /* Otherwise we won't be able to commit or stream */ | |
1265 | bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_COMMIT_TARGET, | |
1266 | parent->backing_blocker); | |
1267 | bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_STREAM, | |
1268 | parent->backing_blocker); | |
1269 | /* | |
1270 | * We do backup in 3 ways: | |
1271 | * 1. drive backup | |
1272 | * The target bs is new opened, and the source is top BDS | |
1273 | * 2. blockdev backup | |
1274 | * Both the source and the target are top BDSes. | |
1275 | * 3. internal backup(used for block replication) | |
1276 | * Both the source and the target are backing file | |
1277 | * | |
1278 | * In case 1 and 2, neither the source nor the target is the backing file. | |
1279 | * In case 3, we will block the top BDS, so there is only one block job | |
1280 | * for the top BDS and its backing chain. | |
1281 | */ | |
1282 | bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_BACKUP_SOURCE, | |
1283 | parent->backing_blocker); | |
1284 | bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_BACKUP_TARGET, | |
1285 | parent->backing_blocker); | |
ca2f1234 | 1286 | } |
d736f119 | 1287 | |
db95dbba KW |
1288 | static void bdrv_backing_detach(BdrvChild *c) |
1289 | { | |
1290 | BlockDriverState *parent = c->opaque; | |
1291 | ||
bdb73476 | 1292 | GLOBAL_STATE_CODE(); |
db95dbba KW |
1293 | assert(parent->backing_blocker); |
1294 | bdrv_op_unblock_all(c->bs, parent->backing_blocker); | |
1295 | error_free(parent->backing_blocker); | |
1296 | parent->backing_blocker = NULL; | |
48e08288 | 1297 | } |
d736f119 | 1298 | |
6858eba0 KW |
1299 | static int bdrv_backing_update_filename(BdrvChild *c, BlockDriverState *base, |
1300 | const char *filename, Error **errp) | |
1301 | { | |
1302 | BlockDriverState *parent = c->opaque; | |
e94d3dba | 1303 | bool read_only = bdrv_is_read_only(parent); |
6858eba0 | 1304 | int ret; |
bdb73476 | 1305 | GLOBAL_STATE_CODE(); |
6858eba0 | 1306 | |
e94d3dba AG |
1307 | if (read_only) { |
1308 | ret = bdrv_reopen_set_read_only(parent, false, errp); | |
61f09cea KW |
1309 | if (ret < 0) { |
1310 | return ret; | |
1311 | } | |
1312 | } | |
1313 | ||
6858eba0 | 1314 | ret = bdrv_change_backing_file(parent, filename, |
e54ee1b3 EB |
1315 | base->drv ? base->drv->format_name : "", |
1316 | false); | |
6858eba0 | 1317 | if (ret < 0) { |
64730694 | 1318 | error_setg_errno(errp, -ret, "Could not update backing file link"); |
6858eba0 KW |
1319 | } |
1320 | ||
e94d3dba AG |
1321 | if (read_only) { |
1322 | bdrv_reopen_set_read_only(parent, true, NULL); | |
61f09cea KW |
1323 | } |
1324 | ||
6858eba0 KW |
1325 | return ret; |
1326 | } | |
1327 | ||
fae8bd39 HR |
1328 | /* |
1329 | * Returns the options and flags that a generic child of a BDS should | |
1330 | * get, based on the given options and flags for the parent BDS. | |
1331 | */ | |
00ff7ffd HR |
1332 | static void bdrv_inherited_options(BdrvChildRole role, bool parent_is_format, |
1333 | int *child_flags, QDict *child_options, | |
1334 | int parent_flags, QDict *parent_options) | |
fae8bd39 HR |
1335 | { |
1336 | int flags = parent_flags; | |
bdb73476 | 1337 | GLOBAL_STATE_CODE(); |
fae8bd39 HR |
1338 | |
1339 | /* | |
1340 | * First, decide whether to set, clear, or leave BDRV_O_PROTOCOL. | |
1341 | * Generally, the question to answer is: Should this child be | |
1342 | * format-probed by default? | |
1343 | */ | |
1344 | ||
1345 | /* | |
1346 | * Pure and non-filtered data children of non-format nodes should | |
1347 | * be probed by default (even when the node itself has BDRV_O_PROTOCOL | |
1348 | * set). This only affects a very limited set of drivers (namely | |
1349 | * quorum and blkverify when this comment was written). | |
1350 | * Force-clear BDRV_O_PROTOCOL then. | |
1351 | */ | |
1352 | if (!parent_is_format && | |
1353 | (role & BDRV_CHILD_DATA) && | |
1354 | !(role & (BDRV_CHILD_METADATA | BDRV_CHILD_FILTERED))) | |
1355 | { | |
1356 | flags &= ~BDRV_O_PROTOCOL; | |
1357 | } | |
1358 | ||
1359 | /* | |
1360 | * All children of format nodes (except for COW children) and all | |
1361 | * metadata children in general should never be format-probed. | |
1362 | * Force-set BDRV_O_PROTOCOL then. | |
1363 | */ | |
1364 | if ((parent_is_format && !(role & BDRV_CHILD_COW)) || | |
1365 | (role & BDRV_CHILD_METADATA)) | |
1366 | { | |
1367 | flags |= BDRV_O_PROTOCOL; | |
1368 | } | |
1369 | ||
1370 | /* | |
1371 | * If the cache mode isn't explicitly set, inherit direct and no-flush from | |
1372 | * the parent. | |
1373 | */ | |
1374 | qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_DIRECT); | |
1375 | qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_NO_FLUSH); | |
1376 | qdict_copy_default(child_options, parent_options, BDRV_OPT_FORCE_SHARE); | |
1377 | ||
1378 | if (role & BDRV_CHILD_COW) { | |
1379 | /* backing files are opened read-only by default */ | |
1380 | qdict_set_default_str(child_options, BDRV_OPT_READ_ONLY, "on"); | |
1381 | qdict_set_default_str(child_options, BDRV_OPT_AUTO_READ_ONLY, "off"); | |
1382 | } else { | |
1383 | /* Inherit the read-only option from the parent if it's not set */ | |
1384 | qdict_copy_default(child_options, parent_options, BDRV_OPT_READ_ONLY); | |
1385 | qdict_copy_default(child_options, parent_options, | |
1386 | BDRV_OPT_AUTO_READ_ONLY); | |
1387 | } | |
1388 | ||
1389 | /* | |
1390 | * bdrv_co_pdiscard() respects unmap policy for the parent, so we | |
1391 | * can default to enable it on lower layers regardless of the | |
1392 | * parent option. | |
1393 | */ | |
1394 | qdict_set_default_str(child_options, BDRV_OPT_DISCARD, "unmap"); | |
1395 | ||
1396 | /* Clear flags that only apply to the top layer */ | |
1397 | flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_COPY_ON_READ); | |
1398 | ||
1399 | if (role & BDRV_CHILD_METADATA) { | |
1400 | flags &= ~BDRV_O_NO_IO; | |
1401 | } | |
1402 | if (role & BDRV_CHILD_COW) { | |
1403 | flags &= ~BDRV_O_TEMPORARY; | |
1404 | } | |
1405 | ||
1406 | *child_flags = flags; | |
1407 | } | |
1408 | ||
303de47b | 1409 | static void GRAPH_WRLOCK bdrv_child_cb_attach(BdrvChild *child) |
ca2f1234 HR |
1410 | { |
1411 | BlockDriverState *bs = child->opaque; | |
1412 | ||
3f35f82e | 1413 | assert_bdrv_graph_writable(); |
a225369b | 1414 | QLIST_INSERT_HEAD(&bs->children, child, next); |
5bb04747 VSO |
1415 | if (bs->drv->is_filter || (child->role & BDRV_CHILD_FILTERED)) { |
1416 | /* | |
1417 | * Here we handle filters and block/raw-format.c when it behave like | |
1418 | * filter. They generally have a single PRIMARY child, which is also the | |
1419 | * FILTERED child, and that they may have multiple more children, which | |
1420 | * are neither PRIMARY nor FILTERED. And never we have a COW child here. | |
1421 | * So bs->file will be the PRIMARY child, unless the PRIMARY child goes | |
1422 | * into bs->backing on exceptional cases; and bs->backing will be | |
1423 | * nothing else. | |
1424 | */ | |
1425 | assert(!(child->role & BDRV_CHILD_COW)); | |
1426 | if (child->role & BDRV_CHILD_PRIMARY) { | |
1427 | assert(child->role & BDRV_CHILD_FILTERED); | |
1428 | assert(!bs->backing); | |
1429 | assert(!bs->file); | |
1430 | ||
1431 | if (bs->drv->filtered_child_is_backing) { | |
1432 | bs->backing = child; | |
1433 | } else { | |
1434 | bs->file = child; | |
1435 | } | |
1436 | } else { | |
1437 | assert(!(child->role & BDRV_CHILD_FILTERED)); | |
1438 | } | |
1439 | } else if (child->role & BDRV_CHILD_COW) { | |
1440 | assert(bs->drv->supports_backing); | |
1441 | assert(!(child->role & BDRV_CHILD_PRIMARY)); | |
1442 | assert(!bs->backing); | |
1443 | bs->backing = child; | |
ca2f1234 | 1444 | bdrv_backing_attach(child); |
5bb04747 VSO |
1445 | } else if (child->role & BDRV_CHILD_PRIMARY) { |
1446 | assert(!bs->file); | |
1447 | bs->file = child; | |
ca2f1234 | 1448 | } |
ca2f1234 HR |
1449 | } |
1450 | ||
303de47b | 1451 | static void GRAPH_WRLOCK bdrv_child_cb_detach(BdrvChild *child) |
48e08288 HR |
1452 | { |
1453 | BlockDriverState *bs = child->opaque; | |
1454 | ||
1455 | if (child->role & BDRV_CHILD_COW) { | |
1456 | bdrv_backing_detach(child); | |
1457 | } | |
1458 | ||
3f35f82e | 1459 | assert_bdrv_graph_writable(); |
a225369b | 1460 | QLIST_REMOVE(child, next); |
5bb04747 VSO |
1461 | if (child == bs->backing) { |
1462 | assert(child != bs->file); | |
1463 | bs->backing = NULL; | |
1464 | } else if (child == bs->file) { | |
1465 | bs->file = NULL; | |
1466 | } | |
48e08288 HR |
1467 | } |
1468 | ||
43483550 HR |
1469 | static int bdrv_child_cb_update_filename(BdrvChild *c, BlockDriverState *base, |
1470 | const char *filename, Error **errp) | |
1471 | { | |
1472 | if (c->role & BDRV_CHILD_COW) { | |
1473 | return bdrv_backing_update_filename(c, base, filename, errp); | |
1474 | } | |
1475 | return 0; | |
1476 | } | |
1477 | ||
fb62b588 | 1478 | AioContext *child_of_bds_get_parent_aio_context(BdrvChild *c) |
3ca1f322 VSO |
1479 | { |
1480 | BlockDriverState *bs = c->opaque; | |
384a48fb | 1481 | IO_CODE(); |
3ca1f322 VSO |
1482 | |
1483 | return bdrv_get_aio_context(bs); | |
1484 | } | |
1485 | ||
43483550 HR |
1486 | const BdrvChildClass child_of_bds = { |
1487 | .parent_is_bds = true, | |
1488 | .get_parent_desc = bdrv_child_get_parent_desc, | |
1489 | .inherit_options = bdrv_inherited_options, | |
1490 | .drained_begin = bdrv_child_cb_drained_begin, | |
1491 | .drained_poll = bdrv_child_cb_drained_poll, | |
1492 | .drained_end = bdrv_child_cb_drained_end, | |
1493 | .attach = bdrv_child_cb_attach, | |
1494 | .detach = bdrv_child_cb_detach, | |
1495 | .inactivate = bdrv_child_cb_inactivate, | |
27633e74 | 1496 | .change_aio_ctx = bdrv_child_cb_change_aio_ctx, |
43483550 | 1497 | .update_filename = bdrv_child_cb_update_filename, |
fb62b588 | 1498 | .get_parent_aio_context = child_of_bds_get_parent_aio_context, |
43483550 HR |
1499 | }; |
1500 | ||
3ca1f322 VSO |
1501 | AioContext *bdrv_child_get_parent_aio_context(BdrvChild *c) |
1502 | { | |
d5f8d79c | 1503 | IO_CODE(); |
3ca1f322 VSO |
1504 | return c->klass->get_parent_aio_context(c); |
1505 | } | |
1506 | ||
7b272452 KW |
1507 | static int bdrv_open_flags(BlockDriverState *bs, int flags) |
1508 | { | |
61de4c68 | 1509 | int open_flags = flags; |
bdb73476 | 1510 | GLOBAL_STATE_CODE(); |
7b272452 KW |
1511 | |
1512 | /* | |
1513 | * Clear flags that are internal to the block layer before opening the | |
1514 | * image. | |
1515 | */ | |
20cca275 | 1516 | open_flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_PROTOCOL); |
7b272452 | 1517 | |
7b272452 KW |
1518 | return open_flags; |
1519 | } | |
1520 | ||
91a097e7 KW |
1521 | static void update_flags_from_options(int *flags, QemuOpts *opts) |
1522 | { | |
bdb73476 EGE |
1523 | GLOBAL_STATE_CODE(); |
1524 | ||
2a3d4331 | 1525 | *flags &= ~(BDRV_O_CACHE_MASK | BDRV_O_RDWR | BDRV_O_AUTO_RDONLY); |
91a097e7 | 1526 | |
57f9db9a | 1527 | if (qemu_opt_get_bool_del(opts, BDRV_OPT_CACHE_NO_FLUSH, false)) { |
91a097e7 KW |
1528 | *flags |= BDRV_O_NO_FLUSH; |
1529 | } | |
1530 | ||
57f9db9a | 1531 | if (qemu_opt_get_bool_del(opts, BDRV_OPT_CACHE_DIRECT, false)) { |
91a097e7 KW |
1532 | *flags |= BDRV_O_NOCACHE; |
1533 | } | |
f87a0e29 | 1534 | |
57f9db9a | 1535 | if (!qemu_opt_get_bool_del(opts, BDRV_OPT_READ_ONLY, false)) { |
f87a0e29 AG |
1536 | *flags |= BDRV_O_RDWR; |
1537 | } | |
1538 | ||
e35bdc12 KW |
1539 | if (qemu_opt_get_bool_del(opts, BDRV_OPT_AUTO_READ_ONLY, false)) { |
1540 | *flags |= BDRV_O_AUTO_RDONLY; | |
1541 | } | |
91a097e7 KW |
1542 | } |
1543 | ||
1544 | static void update_options_from_flags(QDict *options, int flags) | |
1545 | { | |
bdb73476 | 1546 | GLOBAL_STATE_CODE(); |
91a097e7 | 1547 | if (!qdict_haskey(options, BDRV_OPT_CACHE_DIRECT)) { |
46f5ac20 | 1548 | qdict_put_bool(options, BDRV_OPT_CACHE_DIRECT, flags & BDRV_O_NOCACHE); |
91a097e7 KW |
1549 | } |
1550 | if (!qdict_haskey(options, BDRV_OPT_CACHE_NO_FLUSH)) { | |
46f5ac20 EB |
1551 | qdict_put_bool(options, BDRV_OPT_CACHE_NO_FLUSH, |
1552 | flags & BDRV_O_NO_FLUSH); | |
91a097e7 | 1553 | } |
f87a0e29 | 1554 | if (!qdict_haskey(options, BDRV_OPT_READ_ONLY)) { |
46f5ac20 | 1555 | qdict_put_bool(options, BDRV_OPT_READ_ONLY, !(flags & BDRV_O_RDWR)); |
f87a0e29 | 1556 | } |
e35bdc12 KW |
1557 | if (!qdict_haskey(options, BDRV_OPT_AUTO_READ_ONLY)) { |
1558 | qdict_put_bool(options, BDRV_OPT_AUTO_READ_ONLY, | |
1559 | flags & BDRV_O_AUTO_RDONLY); | |
1560 | } | |
91a097e7 KW |
1561 | } |
1562 | ||
636ea370 KW |
1563 | static void bdrv_assign_node_name(BlockDriverState *bs, |
1564 | const char *node_name, | |
1565 | Error **errp) | |
6913c0c2 | 1566 | { |
15489c76 | 1567 | char *gen_node_name = NULL; |
bdb73476 | 1568 | GLOBAL_STATE_CODE(); |
6913c0c2 | 1569 | |
15489c76 JC |
1570 | if (!node_name) { |
1571 | node_name = gen_node_name = id_generate(ID_BLOCK); | |
1572 | } else if (!id_wellformed(node_name)) { | |
1573 | /* | |
1574 | * Check for empty string or invalid characters, but not if it is | |
1575 | * generated (generated names use characters not available to the user) | |
1576 | */ | |
785ec4b1 | 1577 | error_setg(errp, "Invalid node-name: '%s'", node_name); |
636ea370 | 1578 | return; |
6913c0c2 BC |
1579 | } |
1580 | ||
0c5e94ee | 1581 | /* takes care of avoiding namespaces collisions */ |
7f06d47e | 1582 | if (blk_by_name(node_name)) { |
0c5e94ee BC |
1583 | error_setg(errp, "node-name=%s is conflicting with a device id", |
1584 | node_name); | |
15489c76 | 1585 | goto out; |
0c5e94ee BC |
1586 | } |
1587 | ||
6913c0c2 BC |
1588 | /* takes care of avoiding duplicates node names */ |
1589 | if (bdrv_find_node(node_name)) { | |
785ec4b1 | 1590 | error_setg(errp, "Duplicate nodes with node-name='%s'", node_name); |
15489c76 | 1591 | goto out; |
6913c0c2 BC |
1592 | } |
1593 | ||
824808dd KW |
1594 | /* Make sure that the node name isn't truncated */ |
1595 | if (strlen(node_name) >= sizeof(bs->node_name)) { | |
1596 | error_setg(errp, "Node name too long"); | |
1597 | goto out; | |
1598 | } | |
1599 | ||
6913c0c2 BC |
1600 | /* copy node name into the bs and insert it into the graph list */ |
1601 | pstrcpy(bs->node_name, sizeof(bs->node_name), node_name); | |
1602 | QTAILQ_INSERT_TAIL(&graph_bdrv_states, bs, node_list); | |
15489c76 JC |
1603 | out: |
1604 | g_free(gen_node_name); | |
6913c0c2 BC |
1605 | } |
1606 | ||
c86422c5 EGE |
1607 | /* |
1608 | * The caller must always hold @bs AioContext lock, because this function calls | |
1609 | * bdrv_refresh_total_sectors() which polls when called from non-coroutine | |
1610 | * context. | |
1611 | */ | |
1a30b0f5 KW |
1612 | static int no_coroutine_fn GRAPH_UNLOCKED |
1613 | bdrv_open_driver(BlockDriverState *bs, BlockDriver *drv, const char *node_name, | |
1614 | QDict *options, int open_flags, Error **errp) | |
01a56501 | 1615 | { |
8dc8a60c | 1616 | AioContext *ctx; |
01a56501 | 1617 | Error *local_err = NULL; |
0f12264e | 1618 | int i, ret; |
da359909 | 1619 | GLOBAL_STATE_CODE(); |
01a56501 KW |
1620 | |
1621 | bdrv_assign_node_name(bs, node_name, &local_err); | |
1622 | if (local_err) { | |
1623 | error_propagate(errp, local_err); | |
1624 | return -EINVAL; | |
1625 | } | |
1626 | ||
1627 | bs->drv = drv; | |
1628 | bs->opaque = g_malloc0(drv->instance_size); | |
1629 | ||
1630 | if (drv->bdrv_file_open) { | |
1631 | assert(!drv->bdrv_needs_filename || bs->filename[0]); | |
1632 | ret = drv->bdrv_file_open(bs, options, open_flags, &local_err); | |
680c7f96 | 1633 | } else if (drv->bdrv_open) { |
01a56501 | 1634 | ret = drv->bdrv_open(bs, options, open_flags, &local_err); |
680c7f96 KW |
1635 | } else { |
1636 | ret = 0; | |
01a56501 KW |
1637 | } |
1638 | ||
1639 | if (ret < 0) { | |
1640 | if (local_err) { | |
1641 | error_propagate(errp, local_err); | |
1642 | } else if (bs->filename[0]) { | |
1643 | error_setg_errno(errp, -ret, "Could not open '%s'", bs->filename); | |
1644 | } else { | |
1645 | error_setg_errno(errp, -ret, "Could not open image"); | |
1646 | } | |
180ca19a | 1647 | goto open_failed; |
01a56501 KW |
1648 | } |
1649 | ||
e8b65355 SH |
1650 | assert(!(bs->supported_read_flags & ~BDRV_REQ_MASK)); |
1651 | assert(!(bs->supported_write_flags & ~BDRV_REQ_MASK)); | |
1652 | ||
1653 | /* | |
1654 | * Always allow the BDRV_REQ_REGISTERED_BUF optimization hint. This saves | |
1655 | * drivers that pass read/write requests through to a child the trouble of | |
1656 | * declaring support explicitly. | |
1657 | * | |
1658 | * Drivers must not propagate this flag accidentally when they initiate I/O | |
1659 | * to a bounce buffer. That case should be rare though. | |
1660 | */ | |
1661 | bs->supported_read_flags |= BDRV_REQ_REGISTERED_BUF; | |
1662 | bs->supported_write_flags |= BDRV_REQ_REGISTERED_BUF; | |
1663 | ||
8dc8a60c KW |
1664 | /* Get the context after .bdrv_open, it can change the context */ |
1665 | ctx = bdrv_get_aio_context(bs); | |
1666 | aio_context_acquire(ctx); | |
1667 | ||
c057960c | 1668 | ret = bdrv_refresh_total_sectors(bs, bs->total_sectors); |
01a56501 KW |
1669 | if (ret < 0) { |
1670 | error_setg_errno(errp, -ret, "Could not refresh total sector count"); | |
8dc8a60c | 1671 | aio_context_release(ctx); |
180ca19a | 1672 | return ret; |
01a56501 KW |
1673 | } |
1674 | ||
e19b157f | 1675 | bdrv_graph_rdlock_main_loop(); |
1e4c797c | 1676 | bdrv_refresh_limits(bs, NULL, &local_err); |
e19b157f | 1677 | bdrv_graph_rdunlock_main_loop(); |
8dc8a60c | 1678 | aio_context_release(ctx); |
e19b157f | 1679 | |
01a56501 KW |
1680 | if (local_err) { |
1681 | error_propagate(errp, local_err); | |
180ca19a | 1682 | return -EINVAL; |
01a56501 KW |
1683 | } |
1684 | ||
1685 | assert(bdrv_opt_mem_align(bs) != 0); | |
1686 | assert(bdrv_min_mem_align(bs) != 0); | |
1687 | assert(is_power_of_2(bs->bl.request_alignment)); | |
1688 | ||
0f12264e | 1689 | for (i = 0; i < bs->quiesce_counter; i++) { |
5e8ac217 KW |
1690 | if (drv->bdrv_drain_begin) { |
1691 | drv->bdrv_drain_begin(bs); | |
0f12264e KW |
1692 | } |
1693 | } | |
1694 | ||
01a56501 | 1695 | return 0; |
180ca19a MP |
1696 | open_failed: |
1697 | bs->drv = NULL; | |
1698 | if (bs->file != NULL) { | |
1699 | bdrv_unref_child(bs, bs->file); | |
5bb04747 | 1700 | assert(!bs->file); |
180ca19a | 1701 | } |
01a56501 KW |
1702 | g_free(bs->opaque); |
1703 | bs->opaque = NULL; | |
01a56501 KW |
1704 | return ret; |
1705 | } | |
1706 | ||
621d1737 VSO |
1707 | /* |
1708 | * Create and open a block node. | |
1709 | * | |
1710 | * @options is a QDict of options to pass to the block drivers, or NULL for an | |
1711 | * empty set of options. The reference to the QDict belongs to the block layer | |
1712 | * after the call (even on failure), so if the caller intends to reuse the | |
1713 | * dictionary, it needs to use qobject_ref() before calling bdrv_open. | |
1714 | */ | |
1715 | BlockDriverState *bdrv_new_open_driver_opts(BlockDriver *drv, | |
1716 | const char *node_name, | |
1717 | QDict *options, int flags, | |
1718 | Error **errp) | |
680c7f96 KW |
1719 | { |
1720 | BlockDriverState *bs; | |
1721 | int ret; | |
1722 | ||
f791bf7f EGE |
1723 | GLOBAL_STATE_CODE(); |
1724 | ||
680c7f96 KW |
1725 | bs = bdrv_new(); |
1726 | bs->open_flags = flags; | |
621d1737 VSO |
1727 | bs->options = options ?: qdict_new(); |
1728 | bs->explicit_options = qdict_clone_shallow(bs->options); | |
680c7f96 KW |
1729 | bs->opaque = NULL; |
1730 | ||
1731 | update_options_from_flags(bs->options, flags); | |
1732 | ||
1733 | ret = bdrv_open_driver(bs, drv, node_name, bs->options, flags, errp); | |
1734 | if (ret < 0) { | |
cb3e7f08 | 1735 | qobject_unref(bs->explicit_options); |
180ca19a | 1736 | bs->explicit_options = NULL; |
cb3e7f08 | 1737 | qobject_unref(bs->options); |
180ca19a | 1738 | bs->options = NULL; |
680c7f96 KW |
1739 | bdrv_unref(bs); |
1740 | return NULL; | |
1741 | } | |
1742 | ||
1743 | return bs; | |
1744 | } | |
1745 | ||
621d1737 VSO |
1746 | /* Create and open a block node. */ |
1747 | BlockDriverState *bdrv_new_open_driver(BlockDriver *drv, const char *node_name, | |
1748 | int flags, Error **errp) | |
1749 | { | |
f791bf7f | 1750 | GLOBAL_STATE_CODE(); |
621d1737 VSO |
1751 | return bdrv_new_open_driver_opts(drv, node_name, NULL, flags, errp); |
1752 | } | |
1753 | ||
c5f3014b | 1754 | QemuOptsList bdrv_runtime_opts = { |
18edf289 KW |
1755 | .name = "bdrv_common", |
1756 | .head = QTAILQ_HEAD_INITIALIZER(bdrv_runtime_opts.head), | |
1757 | .desc = { | |
1758 | { | |
1759 | .name = "node-name", | |
1760 | .type = QEMU_OPT_STRING, | |
1761 | .help = "Node name of the block device node", | |
1762 | }, | |
62392ebb KW |
1763 | { |
1764 | .name = "driver", | |
1765 | .type = QEMU_OPT_STRING, | |
1766 | .help = "Block driver to use for the node", | |
1767 | }, | |
91a097e7 KW |
1768 | { |
1769 | .name = BDRV_OPT_CACHE_DIRECT, | |
1770 | .type = QEMU_OPT_BOOL, | |
1771 | .help = "Bypass software writeback cache on the host", | |
1772 | }, | |
1773 | { | |
1774 | .name = BDRV_OPT_CACHE_NO_FLUSH, | |
1775 | .type = QEMU_OPT_BOOL, | |
1776 | .help = "Ignore flush requests", | |
1777 | }, | |
f87a0e29 AG |
1778 | { |
1779 | .name = BDRV_OPT_READ_ONLY, | |
1780 | .type = QEMU_OPT_BOOL, | |
1781 | .help = "Node is opened in read-only mode", | |
1782 | }, | |
e35bdc12 KW |
1783 | { |
1784 | .name = BDRV_OPT_AUTO_READ_ONLY, | |
1785 | .type = QEMU_OPT_BOOL, | |
1786 | .help = "Node can become read-only if opening read-write fails", | |
1787 | }, | |
692e01a2 KW |
1788 | { |
1789 | .name = "detect-zeroes", | |
1790 | .type = QEMU_OPT_STRING, | |
1791 | .help = "try to optimize zero writes (off, on, unmap)", | |
1792 | }, | |
818584a4 | 1793 | { |
415bbca8 | 1794 | .name = BDRV_OPT_DISCARD, |
818584a4 KW |
1795 | .type = QEMU_OPT_STRING, |
1796 | .help = "discard operation (ignore/off, unmap/on)", | |
1797 | }, | |
5a9347c6 FZ |
1798 | { |
1799 | .name = BDRV_OPT_FORCE_SHARE, | |
1800 | .type = QEMU_OPT_BOOL, | |
1801 | .help = "always accept other writers (default: off)", | |
1802 | }, | |
18edf289 KW |
1803 | { /* end of list */ } |
1804 | }, | |
1805 | }; | |
1806 | ||
5a5e7f8c ML |
1807 | QemuOptsList bdrv_create_opts_simple = { |
1808 | .name = "simple-create-opts", | |
1809 | .head = QTAILQ_HEAD_INITIALIZER(bdrv_create_opts_simple.head), | |
fd17146c HR |
1810 | .desc = { |
1811 | { | |
1812 | .name = BLOCK_OPT_SIZE, | |
1813 | .type = QEMU_OPT_SIZE, | |
1814 | .help = "Virtual disk size" | |
1815 | }, | |
1816 | { | |
1817 | .name = BLOCK_OPT_PREALLOC, | |
1818 | .type = QEMU_OPT_STRING, | |
1819 | .help = "Preallocation mode (allowed values: off)" | |
1820 | }, | |
1821 | { /* end of list */ } | |
1822 | } | |
1823 | }; | |
1824 | ||
57915332 KW |
1825 | /* |
1826 | * Common part for opening disk images and files | |
b6ad491a KW |
1827 | * |
1828 | * Removes all processed options from *options. | |
57915332 | 1829 | */ |
5696c6e3 | 1830 | static int bdrv_open_common(BlockDriverState *bs, BlockBackend *file, |
82dc8b41 | 1831 | QDict *options, Error **errp) |
57915332 KW |
1832 | { |
1833 | int ret, open_flags; | |
035fccdf | 1834 | const char *filename; |
62392ebb | 1835 | const char *driver_name = NULL; |
6913c0c2 | 1836 | const char *node_name = NULL; |
818584a4 | 1837 | const char *discard; |
18edf289 | 1838 | QemuOpts *opts; |
62392ebb | 1839 | BlockDriver *drv; |
34b5d2c6 | 1840 | Error *local_err = NULL; |
307261b2 | 1841 | bool ro; |
57915332 | 1842 | |
6405875c | 1843 | assert(bs->file == NULL); |
707ff828 | 1844 | assert(options != NULL && bs->options != options); |
bdb73476 | 1845 | GLOBAL_STATE_CODE(); |
57915332 | 1846 | |
62392ebb | 1847 | opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort); |
af175e85 | 1848 | if (!qemu_opts_absorb_qdict(opts, options, errp)) { |
62392ebb KW |
1849 | ret = -EINVAL; |
1850 | goto fail_opts; | |
1851 | } | |
1852 | ||
9b7e8691 AG |
1853 | update_flags_from_options(&bs->open_flags, opts); |
1854 | ||
62392ebb KW |
1855 | driver_name = qemu_opt_get(opts, "driver"); |
1856 | drv = bdrv_find_format(driver_name); | |
1857 | assert(drv != NULL); | |
1858 | ||
5a9347c6 FZ |
1859 | bs->force_share = qemu_opt_get_bool(opts, BDRV_OPT_FORCE_SHARE, false); |
1860 | ||
1861 | if (bs->force_share && (bs->open_flags & BDRV_O_RDWR)) { | |
1862 | error_setg(errp, | |
1863 | BDRV_OPT_FORCE_SHARE | |
1864 | "=on can only be used with read-only images"); | |
1865 | ret = -EINVAL; | |
1866 | goto fail_opts; | |
1867 | } | |
1868 | ||
45673671 | 1869 | if (file != NULL) { |
f30c66ba | 1870 | bdrv_refresh_filename(blk_bs(file)); |
5696c6e3 | 1871 | filename = blk_bs(file)->filename; |
45673671 | 1872 | } else { |
129c7d1c MA |
1873 | /* |
1874 | * Caution: while qdict_get_try_str() is fine, getting | |
1875 | * non-string types would require more care. When @options | |
1876 | * come from -blockdev or blockdev_add, its members are typed | |
1877 | * according to the QAPI schema, but when they come from | |
1878 | * -drive, they're all QString. | |
1879 | */ | |
45673671 KW |
1880 | filename = qdict_get_try_str(options, "filename"); |
1881 | } | |
1882 | ||
4a008240 | 1883 | if (drv->bdrv_needs_filename && (!filename || !filename[0])) { |
765003db KW |
1884 | error_setg(errp, "The '%s' block driver requires a file name", |
1885 | drv->format_name); | |
18edf289 KW |
1886 | ret = -EINVAL; |
1887 | goto fail_opts; | |
6913c0c2 | 1888 | } |
6913c0c2 | 1889 | |
82dc8b41 KW |
1890 | trace_bdrv_open_common(bs, filename ?: "", bs->open_flags, |
1891 | drv->format_name); | |
62392ebb | 1892 | |
307261b2 VSO |
1893 | ro = bdrv_is_read_only(bs); |
1894 | ||
1895 | if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv, ro)) { | |
1896 | if (!ro && bdrv_is_whitelisted(drv, true)) { | |
8be25de6 KW |
1897 | ret = bdrv_apply_auto_read_only(bs, NULL, NULL); |
1898 | } else { | |
1899 | ret = -ENOTSUP; | |
1900 | } | |
1901 | if (ret < 0) { | |
1902 | error_setg(errp, | |
307261b2 | 1903 | !ro && bdrv_is_whitelisted(drv, true) |
8be25de6 KW |
1904 | ? "Driver '%s' can only be used for read-only devices" |
1905 | : "Driver '%s' is not whitelisted", | |
1906 | drv->format_name); | |
1907 | goto fail_opts; | |
1908 | } | |
b64ec4e4 | 1909 | } |
57915332 | 1910 | |
d3faa13e | 1911 | /* bdrv_new() and bdrv_close() make it so */ |
d73415a3 | 1912 | assert(qatomic_read(&bs->copy_on_read) == 0); |
d3faa13e | 1913 | |
82dc8b41 | 1914 | if (bs->open_flags & BDRV_O_COPY_ON_READ) { |
307261b2 | 1915 | if (!ro) { |
0ebd24e0 KW |
1916 | bdrv_enable_copy_on_read(bs); |
1917 | } else { | |
1918 | error_setg(errp, "Can't use copy-on-read on read-only device"); | |
18edf289 KW |
1919 | ret = -EINVAL; |
1920 | goto fail_opts; | |
0ebd24e0 | 1921 | } |
53fec9d3 SH |
1922 | } |
1923 | ||
415bbca8 | 1924 | discard = qemu_opt_get(opts, BDRV_OPT_DISCARD); |
818584a4 KW |
1925 | if (discard != NULL) { |
1926 | if (bdrv_parse_discard_flags(discard, &bs->open_flags) != 0) { | |
1927 | error_setg(errp, "Invalid discard option"); | |
1928 | ret = -EINVAL; | |
1929 | goto fail_opts; | |
1930 | } | |
1931 | } | |
1932 | ||
543770bd AG |
1933 | bs->detect_zeroes = |
1934 | bdrv_parse_detect_zeroes(opts, bs->open_flags, &local_err); | |
1935 | if (local_err) { | |
1936 | error_propagate(errp, local_err); | |
1937 | ret = -EINVAL; | |
1938 | goto fail_opts; | |
692e01a2 KW |
1939 | } |
1940 | ||
c2ad1b0c KW |
1941 | if (filename != NULL) { |
1942 | pstrcpy(bs->filename, sizeof(bs->filename), filename); | |
1943 | } else { | |
1944 | bs->filename[0] = '\0'; | |
1945 | } | |
91af7014 | 1946 | pstrcpy(bs->exact_filename, sizeof(bs->exact_filename), bs->filename); |
57915332 | 1947 | |
66f82cee | 1948 | /* Open the image, either directly or using a protocol */ |
82dc8b41 | 1949 | open_flags = bdrv_open_flags(bs, bs->open_flags); |
01a56501 | 1950 | node_name = qemu_opt_get(opts, "node-name"); |
57915332 | 1951 | |
01a56501 KW |
1952 | assert(!drv->bdrv_file_open || file == NULL); |
1953 | ret = bdrv_open_driver(bs, drv, node_name, options, open_flags, errp); | |
51762288 | 1954 | if (ret < 0) { |
01a56501 | 1955 | goto fail_opts; |
3baca891 KW |
1956 | } |
1957 | ||
18edf289 | 1958 | qemu_opts_del(opts); |
57915332 KW |
1959 | return 0; |
1960 | ||
18edf289 KW |
1961 | fail_opts: |
1962 | qemu_opts_del(opts); | |
57915332 KW |
1963 | return ret; |
1964 | } | |
1965 | ||
5e5c4f63 KW |
1966 | static QDict *parse_json_filename(const char *filename, Error **errp) |
1967 | { | |
1968 | QObject *options_obj; | |
1969 | QDict *options; | |
1970 | int ret; | |
bdb73476 | 1971 | GLOBAL_STATE_CODE(); |
5e5c4f63 KW |
1972 | |
1973 | ret = strstart(filename, "json:", &filename); | |
1974 | assert(ret); | |
1975 | ||
5577fff7 | 1976 | options_obj = qobject_from_json(filename, errp); |
5e5c4f63 | 1977 | if (!options_obj) { |
5577fff7 | 1978 | error_prepend(errp, "Could not parse the JSON options: "); |
5e5c4f63 KW |
1979 | return NULL; |
1980 | } | |
1981 | ||
7dc847eb | 1982 | options = qobject_to(QDict, options_obj); |
ca6b6e1e | 1983 | if (!options) { |
cb3e7f08 | 1984 | qobject_unref(options_obj); |
5e5c4f63 KW |
1985 | error_setg(errp, "Invalid JSON object given"); |
1986 | return NULL; | |
1987 | } | |
1988 | ||
5e5c4f63 KW |
1989 | qdict_flatten(options); |
1990 | ||
1991 | return options; | |
1992 | } | |
1993 | ||
de3b53f0 KW |
1994 | static void parse_json_protocol(QDict *options, const char **pfilename, |
1995 | Error **errp) | |
1996 | { | |
1997 | QDict *json_options; | |
1998 | Error *local_err = NULL; | |
bdb73476 | 1999 | GLOBAL_STATE_CODE(); |
de3b53f0 KW |
2000 | |
2001 | /* Parse json: pseudo-protocol */ | |
2002 | if (!*pfilename || !g_str_has_prefix(*pfilename, "json:")) { | |
2003 | return; | |
2004 | } | |
2005 | ||
2006 | json_options = parse_json_filename(*pfilename, &local_err); | |
2007 | if (local_err) { | |
2008 | error_propagate(errp, local_err); | |
2009 | return; | |
2010 | } | |
2011 | ||
2012 | /* Options given in the filename have lower priority than options | |
2013 | * specified directly */ | |
2014 | qdict_join(options, json_options, false); | |
cb3e7f08 | 2015 | qobject_unref(json_options); |
de3b53f0 KW |
2016 | *pfilename = NULL; |
2017 | } | |
2018 | ||
b6ce07aa | 2019 | /* |
f54120ff KW |
2020 | * Fills in default options for opening images and converts the legacy |
2021 | * filename/flags pair to option QDict entries. | |
53a29513 HR |
2022 | * The BDRV_O_PROTOCOL flag in *flags will be set or cleared accordingly if a |
2023 | * block driver has been specified explicitly. | |
b6ce07aa | 2024 | */ |
de3b53f0 | 2025 | static int bdrv_fill_options(QDict **options, const char *filename, |
053e1578 | 2026 | int *flags, Error **errp) |
ea2384d3 | 2027 | { |
c2ad1b0c | 2028 | const char *drvname; |
53a29513 | 2029 | bool protocol = *flags & BDRV_O_PROTOCOL; |
e3fa4bfa | 2030 | bool parse_filename = false; |
053e1578 | 2031 | BlockDriver *drv = NULL; |
34b5d2c6 | 2032 | Error *local_err = NULL; |
83f64091 | 2033 | |
da359909 EGE |
2034 | GLOBAL_STATE_CODE(); |
2035 | ||
129c7d1c MA |
2036 | /* |
2037 | * Caution: while qdict_get_try_str() is fine, getting non-string | |
2038 | * types would require more care. When @options come from | |
2039 | * -blockdev or blockdev_add, its members are typed according to | |
2040 | * the QAPI schema, but when they come from -drive, they're all | |
2041 | * QString. | |
2042 | */ | |
53a29513 | 2043 | drvname = qdict_get_try_str(*options, "driver"); |
053e1578 HR |
2044 | if (drvname) { |
2045 | drv = bdrv_find_format(drvname); | |
2046 | if (!drv) { | |
2047 | error_setg(errp, "Unknown driver '%s'", drvname); | |
2048 | return -ENOENT; | |
2049 | } | |
2050 | /* If the user has explicitly specified the driver, this choice should | |
2051 | * override the BDRV_O_PROTOCOL flag */ | |
2052 | protocol = drv->bdrv_file_open; | |
53a29513 HR |
2053 | } |
2054 | ||
2055 | if (protocol) { | |
2056 | *flags |= BDRV_O_PROTOCOL; | |
2057 | } else { | |
2058 | *flags &= ~BDRV_O_PROTOCOL; | |
2059 | } | |
2060 | ||
91a097e7 KW |
2061 | /* Translate cache options from flags into options */ |
2062 | update_options_from_flags(*options, *flags); | |
2063 | ||
035fccdf | 2064 | /* Fetch the file name from the options QDict if necessary */ |
17b005f1 | 2065 | if (protocol && filename) { |
f54120ff | 2066 | if (!qdict_haskey(*options, "filename")) { |
46f5ac20 | 2067 | qdict_put_str(*options, "filename", filename); |
f54120ff KW |
2068 | parse_filename = true; |
2069 | } else { | |
2070 | error_setg(errp, "Can't specify 'file' and 'filename' options at " | |
2071 | "the same time"); | |
2072 | return -EINVAL; | |
2073 | } | |
035fccdf KW |
2074 | } |
2075 | ||
c2ad1b0c | 2076 | /* Find the right block driver */ |
129c7d1c | 2077 | /* See cautionary note on accessing @options above */ |
f54120ff | 2078 | filename = qdict_get_try_str(*options, "filename"); |
f54120ff | 2079 | |
053e1578 HR |
2080 | if (!drvname && protocol) { |
2081 | if (filename) { | |
2082 | drv = bdrv_find_protocol(filename, parse_filename, errp); | |
17b005f1 | 2083 | if (!drv) { |
053e1578 | 2084 | return -EINVAL; |
17b005f1 | 2085 | } |
053e1578 HR |
2086 | |
2087 | drvname = drv->format_name; | |
46f5ac20 | 2088 | qdict_put_str(*options, "driver", drvname); |
053e1578 HR |
2089 | } else { |
2090 | error_setg(errp, "Must specify either driver or file"); | |
2091 | return -EINVAL; | |
98289620 | 2092 | } |
c2ad1b0c KW |
2093 | } |
2094 | ||
17b005f1 | 2095 | assert(drv || !protocol); |
c2ad1b0c | 2096 | |
f54120ff | 2097 | /* Driver-specific filename parsing */ |
17b005f1 | 2098 | if (drv && drv->bdrv_parse_filename && parse_filename) { |
5acd9d81 | 2099 | drv->bdrv_parse_filename(filename, *options, &local_err); |
84d18f06 | 2100 | if (local_err) { |
34b5d2c6 | 2101 | error_propagate(errp, local_err); |
f54120ff | 2102 | return -EINVAL; |
6963a30d | 2103 | } |
cd5d031e HR |
2104 | |
2105 | if (!drv->bdrv_needs_filename) { | |
2106 | qdict_del(*options, "filename"); | |
cd5d031e | 2107 | } |
6963a30d KW |
2108 | } |
2109 | ||
f54120ff KW |
2110 | return 0; |
2111 | } | |
2112 | ||
148eb13c KW |
2113 | typedef struct BlockReopenQueueEntry { |
2114 | bool prepared; | |
69b736e7 | 2115 | bool perms_checked; |
148eb13c | 2116 | BDRVReopenState state; |
859443b0 | 2117 | QTAILQ_ENTRY(BlockReopenQueueEntry) entry; |
148eb13c KW |
2118 | } BlockReopenQueueEntry; |
2119 | ||
2120 | /* | |
2121 | * Return the flags that @bs will have after the reopens in @q have | |
2122 | * successfully completed. If @q is NULL (or @bs is not contained in @q), | |
2123 | * return the current flags. | |
2124 | */ | |
2125 | static int bdrv_reopen_get_flags(BlockReopenQueue *q, BlockDriverState *bs) | |
2126 | { | |
2127 | BlockReopenQueueEntry *entry; | |
2128 | ||
2129 | if (q != NULL) { | |
859443b0 | 2130 | QTAILQ_FOREACH(entry, q, entry) { |
148eb13c KW |
2131 | if (entry->state.bs == bs) { |
2132 | return entry->state.flags; | |
2133 | } | |
2134 | } | |
2135 | } | |
2136 | ||
2137 | return bs->open_flags; | |
2138 | } | |
2139 | ||
2140 | /* Returns whether the image file can be written to after the reopen queue @q | |
2141 | * has been successfully applied, or right now if @q is NULL. */ | |
cc022140 HR |
2142 | static bool bdrv_is_writable_after_reopen(BlockDriverState *bs, |
2143 | BlockReopenQueue *q) | |
148eb13c KW |
2144 | { |
2145 | int flags = bdrv_reopen_get_flags(q, bs); | |
2146 | ||
2147 | return (flags & (BDRV_O_RDWR | BDRV_O_INACTIVE)) == BDRV_O_RDWR; | |
2148 | } | |
2149 | ||
cc022140 HR |
2150 | /* |
2151 | * Return whether the BDS can be written to. This is not necessarily | |
2152 | * the same as !bdrv_is_read_only(bs), as inactivated images may not | |
2153 | * be written to but do not count as read-only images. | |
2154 | */ | |
2155 | bool bdrv_is_writable(BlockDriverState *bs) | |
2156 | { | |
384a48fb | 2157 | IO_CODE(); |
cc022140 HR |
2158 | return bdrv_is_writable_after_reopen(bs, NULL); |
2159 | } | |
2160 | ||
3bf416ba VSO |
2161 | static char *bdrv_child_user_desc(BdrvChild *c) |
2162 | { | |
f0c28327 | 2163 | GLOBAL_STATE_CODE(); |
da261b69 | 2164 | return c->klass->get_parent_desc(c); |
3bf416ba VSO |
2165 | } |
2166 | ||
30ebb9aa VSO |
2167 | /* |
2168 | * Check that @a allows everything that @b needs. @a and @b must reference same | |
2169 | * child node. | |
2170 | */ | |
3bf416ba VSO |
2171 | static bool bdrv_a_allow_b(BdrvChild *a, BdrvChild *b, Error **errp) |
2172 | { | |
30ebb9aa VSO |
2173 | const char *child_bs_name; |
2174 | g_autofree char *a_user = NULL; | |
2175 | g_autofree char *b_user = NULL; | |
2176 | g_autofree char *perms = NULL; | |
2177 | ||
2178 | assert(a->bs); | |
2179 | assert(a->bs == b->bs); | |
862fded9 | 2180 | GLOBAL_STATE_CODE(); |
3bf416ba VSO |
2181 | |
2182 | if ((b->perm & a->shared_perm) == b->perm) { | |
2183 | return true; | |
2184 | } | |
2185 | ||
30ebb9aa VSO |
2186 | child_bs_name = bdrv_get_node_name(b->bs); |
2187 | a_user = bdrv_child_user_desc(a); | |
2188 | b_user = bdrv_child_user_desc(b); | |
2189 | perms = bdrv_perm_names(b->perm & ~a->shared_perm); | |
2190 | ||
2191 | error_setg(errp, "Permission conflict on node '%s': permissions '%s' are " | |
2192 | "both required by %s (uses node '%s' as '%s' child) and " | |
2193 | "unshared by %s (uses node '%s' as '%s' child).", | |
2194 | child_bs_name, perms, | |
2195 | b_user, child_bs_name, b->name, | |
2196 | a_user, child_bs_name, a->name); | |
3bf416ba VSO |
2197 | |
2198 | return false; | |
2199 | } | |
2200 | ||
9397c14f | 2201 | static bool bdrv_parent_perms_conflict(BlockDriverState *bs, Error **errp) |
3bf416ba VSO |
2202 | { |
2203 | BdrvChild *a, *b; | |
862fded9 | 2204 | GLOBAL_STATE_CODE(); |
3bf416ba VSO |
2205 | |
2206 | /* | |
2207 | * During the loop we'll look at each pair twice. That's correct because | |
2208 | * bdrv_a_allow_b() is asymmetric and we should check each pair in both | |
2209 | * directions. | |
2210 | */ | |
2211 | QLIST_FOREACH(a, &bs->parents, next_parent) { | |
2212 | QLIST_FOREACH(b, &bs->parents, next_parent) { | |
9397c14f | 2213 | if (a == b) { |
3bf416ba VSO |
2214 | continue; |
2215 | } | |
2216 | ||
2217 | if (!bdrv_a_allow_b(a, b, errp)) { | |
2218 | return true; | |
2219 | } | |
2220 | } | |
2221 | } | |
2222 | ||
2223 | return false; | |
2224 | } | |
2225 | ||
ffd1a5a2 | 2226 | static void bdrv_child_perm(BlockDriverState *bs, BlockDriverState *child_bs, |
e5d8a406 HR |
2227 | BdrvChild *c, BdrvChildRole role, |
2228 | BlockReopenQueue *reopen_queue, | |
ffd1a5a2 FZ |
2229 | uint64_t parent_perm, uint64_t parent_shared, |
2230 | uint64_t *nperm, uint64_t *nshared) | |
2231 | { | |
0b3ca76e | 2232 | assert(bs->drv && bs->drv->bdrv_child_perm); |
da359909 | 2233 | GLOBAL_STATE_CODE(); |
e5d8a406 | 2234 | bs->drv->bdrv_child_perm(bs, c, role, reopen_queue, |
0b3ca76e AG |
2235 | parent_perm, parent_shared, |
2236 | nperm, nshared); | |
e0995dc3 | 2237 | /* TODO Take force_share from reopen_queue */ |
ffd1a5a2 FZ |
2238 | if (child_bs && child_bs->force_share) { |
2239 | *nshared = BLK_PERM_ALL; | |
2240 | } | |
2241 | } | |
2242 | ||
bd57f8f7 VSO |
2243 | /* |
2244 | * Adds the whole subtree of @bs (including @bs itself) to the @list (except for | |
2245 | * nodes that are already in the @list, of course) so that final list is | |
2246 | * topologically sorted. Return the result (GSList @list object is updated, so | |
2247 | * don't use old reference after function call). | |
2248 | * | |
2249 | * On function start @list must be already topologically sorted and for any node | |
2250 | * in the @list the whole subtree of the node must be in the @list as well. The | |
2251 | * simplest way to satisfy this criteria: use only result of | |
2252 | * bdrv_topological_dfs() or NULL as @list parameter. | |
2253 | */ | |
2254 | static GSList *bdrv_topological_dfs(GSList *list, GHashTable *found, | |
2255 | BlockDriverState *bs) | |
2256 | { | |
2257 | BdrvChild *child; | |
2258 | g_autoptr(GHashTable) local_found = NULL; | |
2259 | ||
bdb73476 EGE |
2260 | GLOBAL_STATE_CODE(); |
2261 | ||
bd57f8f7 VSO |
2262 | if (!found) { |
2263 | assert(!list); | |
2264 | found = local_found = g_hash_table_new(NULL, NULL); | |
2265 | } | |
2266 | ||
2267 | if (g_hash_table_contains(found, bs)) { | |
2268 | return list; | |
2269 | } | |
2270 | g_hash_table_add(found, bs); | |
2271 | ||
2272 | QLIST_FOREACH(child, &bs->children, next) { | |
2273 | list = bdrv_topological_dfs(list, found, child->bs); | |
2274 | } | |
2275 | ||
2276 | return g_slist_prepend(list, bs); | |
2277 | } | |
2278 | ||
ecb776bd VSO |
2279 | typedef struct BdrvChildSetPermState { |
2280 | BdrvChild *child; | |
2281 | uint64_t old_perm; | |
2282 | uint64_t old_shared_perm; | |
2283 | } BdrvChildSetPermState; | |
b0defa83 VSO |
2284 | |
2285 | static void bdrv_child_set_perm_abort(void *opaque) | |
2286 | { | |
ecb776bd VSO |
2287 | BdrvChildSetPermState *s = opaque; |
2288 | ||
862fded9 EGE |
2289 | GLOBAL_STATE_CODE(); |
2290 | ||
ecb776bd VSO |
2291 | s->child->perm = s->old_perm; |
2292 | s->child->shared_perm = s->old_shared_perm; | |
b0defa83 VSO |
2293 | } |
2294 | ||
2295 | static TransactionActionDrv bdrv_child_set_pem_drv = { | |
2296 | .abort = bdrv_child_set_perm_abort, | |
ecb776bd | 2297 | .clean = g_free, |
b0defa83 VSO |
2298 | }; |
2299 | ||
ecb776bd VSO |
2300 | static void bdrv_child_set_perm(BdrvChild *c, uint64_t perm, |
2301 | uint64_t shared, Transaction *tran) | |
b0defa83 | 2302 | { |
ecb776bd | 2303 | BdrvChildSetPermState *s = g_new(BdrvChildSetPermState, 1); |
862fded9 | 2304 | GLOBAL_STATE_CODE(); |
ecb776bd VSO |
2305 | |
2306 | *s = (BdrvChildSetPermState) { | |
2307 | .child = c, | |
2308 | .old_perm = c->perm, | |
2309 | .old_shared_perm = c->shared_perm, | |
2310 | }; | |
b0defa83 VSO |
2311 | |
2312 | c->perm = perm; | |
2313 | c->shared_perm = shared; | |
2314 | ||
ecb776bd | 2315 | tran_add(tran, &bdrv_child_set_pem_drv, s); |
b0defa83 VSO |
2316 | } |
2317 | ||
2513ef59 VSO |
2318 | static void bdrv_drv_set_perm_commit(void *opaque) |
2319 | { | |
2320 | BlockDriverState *bs = opaque; | |
2321 | uint64_t cumulative_perms, cumulative_shared_perms; | |
da359909 | 2322 | GLOBAL_STATE_CODE(); |
2513ef59 VSO |
2323 | |
2324 | if (bs->drv->bdrv_set_perm) { | |
2325 | bdrv_get_cumulative_perm(bs, &cumulative_perms, | |
2326 | &cumulative_shared_perms); | |
2327 | bs->drv->bdrv_set_perm(bs, cumulative_perms, cumulative_shared_perms); | |
2328 | } | |
2329 | } | |
2330 | ||
2331 | static void bdrv_drv_set_perm_abort(void *opaque) | |
2332 | { | |
2333 | BlockDriverState *bs = opaque; | |
da359909 | 2334 | GLOBAL_STATE_CODE(); |
2513ef59 VSO |
2335 | |
2336 | if (bs->drv->bdrv_abort_perm_update) { | |
2337 | bs->drv->bdrv_abort_perm_update(bs); | |
2338 | } | |
2339 | } | |
2340 | ||
2341 | TransactionActionDrv bdrv_drv_set_perm_drv = { | |
2342 | .abort = bdrv_drv_set_perm_abort, | |
2343 | .commit = bdrv_drv_set_perm_commit, | |
2344 | }; | |
2345 | ||
2346 | static int bdrv_drv_set_perm(BlockDriverState *bs, uint64_t perm, | |
2347 | uint64_t shared_perm, Transaction *tran, | |
2348 | Error **errp) | |
2349 | { | |
da359909 | 2350 | GLOBAL_STATE_CODE(); |
2513ef59 VSO |
2351 | if (!bs->drv) { |
2352 | return 0; | |
2353 | } | |
2354 | ||
2355 | if (bs->drv->bdrv_check_perm) { | |
2356 | int ret = bs->drv->bdrv_check_perm(bs, perm, shared_perm, errp); | |
2357 | if (ret < 0) { | |
2358 | return ret; | |
2359 | } | |
2360 | } | |
2361 | ||
2362 | if (tran) { | |
2363 | tran_add(tran, &bdrv_drv_set_perm_drv, bs); | |
2364 | } | |
2365 | ||
2366 | return 0; | |
2367 | } | |
2368 | ||
0978623e VSO |
2369 | typedef struct BdrvReplaceChildState { |
2370 | BdrvChild *child; | |
2371 | BlockDriverState *old_bs; | |
2372 | } BdrvReplaceChildState; | |
2373 | ||
2374 | static void bdrv_replace_child_commit(void *opaque) | |
2375 | { | |
2376 | BdrvReplaceChildState *s = opaque; | |
bdb73476 | 2377 | GLOBAL_STATE_CODE(); |
0978623e VSO |
2378 | |
2379 | bdrv_unref(s->old_bs); | |
2380 | } | |
2381 | ||
2382 | static void bdrv_replace_child_abort(void *opaque) | |
2383 | { | |
2384 | BdrvReplaceChildState *s = opaque; | |
2385 | BlockDriverState *new_bs = s->child->bs; | |
2386 | ||
bdb73476 | 2387 | GLOBAL_STATE_CODE(); |
0f0b1e29 | 2388 | /* old_bs reference is transparently moved from @s to @s->child */ |
23987471 KW |
2389 | if (!s->child->bs) { |
2390 | /* | |
2391 | * The parents were undrained when removing old_bs from the child. New | |
2392 | * requests can't have been made, though, because the child was empty. | |
2393 | * | |
2394 | * TODO Make bdrv_replace_child_noperm() transactionable to avoid | |
2395 | * undraining the parent in the first place. Once this is done, having | |
2396 | * new_bs drained when calling bdrv_replace_child_tran() is not a | |
2397 | * requirement any more. | |
2398 | */ | |
606ed756 | 2399 | bdrv_parent_drained_begin_single(s->child); |
23987471 KW |
2400 | assert(!bdrv_parent_drained_poll_single(s->child)); |
2401 | } | |
2402 | assert(s->child->quiesced_parent); | |
544acc7d | 2403 | bdrv_replace_child_noperm(s->child, s->old_bs); |
0978623e VSO |
2404 | bdrv_unref(new_bs); |
2405 | } | |
2406 | ||
2407 | static TransactionActionDrv bdrv_replace_child_drv = { | |
2408 | .commit = bdrv_replace_child_commit, | |
2409 | .abort = bdrv_replace_child_abort, | |
2410 | .clean = g_free, | |
2411 | }; | |
2412 | ||
2413 | /* | |
4bf021db | 2414 | * bdrv_replace_child_tran |
0978623e VSO |
2415 | * |
2416 | * Note: real unref of old_bs is done only on commit. | |
4bf021db | 2417 | * |
23987471 KW |
2418 | * Both @child->bs and @new_bs (if non-NULL) must be drained. @new_bs must be |
2419 | * kept drained until the transaction is completed. | |
2420 | * | |
4bf021db | 2421 | * The function doesn't update permissions, caller is responsible for this. |
0978623e | 2422 | */ |
0f0b1e29 | 2423 | static void bdrv_replace_child_tran(BdrvChild *child, BlockDriverState *new_bs, |
4eba825a | 2424 | Transaction *tran) |
0978623e VSO |
2425 | { |
2426 | BdrvReplaceChildState *s = g_new(BdrvReplaceChildState, 1); | |
23987471 KW |
2427 | |
2428 | assert(child->quiesced_parent); | |
2429 | assert(!new_bs || new_bs->quiesce_counter); | |
2430 | ||
0978623e | 2431 | *s = (BdrvReplaceChildState) { |
0f0b1e29 VSO |
2432 | .child = child, |
2433 | .old_bs = child->bs, | |
0978623e VSO |
2434 | }; |
2435 | tran_add(tran, &bdrv_replace_child_drv, s); | |
2436 | ||
2437 | if (new_bs) { | |
2438 | bdrv_ref(new_bs); | |
2439 | } | |
544acc7d | 2440 | bdrv_replace_child_noperm(child, new_bs); |
0f0b1e29 | 2441 | /* old_bs reference is transparently moved from @child to @s */ |
0978623e VSO |
2442 | } |
2443 | ||
33a610c3 | 2444 | /* |
c20555e1 VSO |
2445 | * Refresh permissions in @bs subtree. The function is intended to be called |
2446 | * after some graph modification that was done without permission update. | |
33a610c3 | 2447 | */ |
c20555e1 VSO |
2448 | static int bdrv_node_refresh_perm(BlockDriverState *bs, BlockReopenQueue *q, |
2449 | Transaction *tran, Error **errp) | |
33a610c3 KW |
2450 | { |
2451 | BlockDriver *drv = bs->drv; | |
2452 | BdrvChild *c; | |
2453 | int ret; | |
c20555e1 | 2454 | uint64_t cumulative_perms, cumulative_shared_perms; |
862fded9 | 2455 | GLOBAL_STATE_CODE(); |
c20555e1 VSO |
2456 | |
2457 | bdrv_get_cumulative_perm(bs, &cumulative_perms, &cumulative_shared_perms); | |
33a610c3 KW |
2458 | |
2459 | /* Write permissions never work with read-only images */ | |
2460 | if ((cumulative_perms & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)) && | |
cc022140 | 2461 | !bdrv_is_writable_after_reopen(bs, q)) |
33a610c3 | 2462 | { |
481e0eee HR |
2463 | if (!bdrv_is_writable_after_reopen(bs, NULL)) { |
2464 | error_setg(errp, "Block node is read-only"); | |
2465 | } else { | |
c20555e1 VSO |
2466 | error_setg(errp, "Read-only block node '%s' cannot support " |
2467 | "read-write users", bdrv_get_node_name(bs)); | |
481e0eee HR |
2468 | } |
2469 | ||
33a610c3 KW |
2470 | return -EPERM; |
2471 | } | |
2472 | ||
9c60a5d1 KW |
2473 | /* |
2474 | * Unaligned requests will automatically be aligned to bl.request_alignment | |
2475 | * and without RESIZE we can't extend requests to write to space beyond the | |
2476 | * end of the image, so it's required that the image size is aligned. | |
2477 | */ | |
2478 | if ((cumulative_perms & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)) && | |
2479 | !(cumulative_perms & BLK_PERM_RESIZE)) | |
2480 | { | |
2481 | if ((bs->total_sectors * BDRV_SECTOR_SIZE) % bs->bl.request_alignment) { | |
2482 | error_setg(errp, "Cannot get 'write' permission without 'resize': " | |
2483 | "Image size is not a multiple of request " | |
2484 | "alignment"); | |
2485 | return -EPERM; | |
2486 | } | |
2487 | } | |
2488 | ||
33a610c3 KW |
2489 | /* Check this node */ |
2490 | if (!drv) { | |
2491 | return 0; | |
2492 | } | |
2493 | ||
b1d2bbeb | 2494 | ret = bdrv_drv_set_perm(bs, cumulative_perms, cumulative_shared_perms, tran, |
2513ef59 VSO |
2495 | errp); |
2496 | if (ret < 0) { | |
2497 | return ret; | |
33a610c3 KW |
2498 | } |
2499 | ||
78e421c9 | 2500 | /* Drivers that never have children can omit .bdrv_child_perm() */ |
33a610c3 | 2501 | if (!drv->bdrv_child_perm) { |
78e421c9 | 2502 | assert(QLIST_EMPTY(&bs->children)); |
33a610c3 KW |
2503 | return 0; |
2504 | } | |
2505 | ||
2506 | /* Check all children */ | |
2507 | QLIST_FOREACH(c, &bs->children, next) { | |
2508 | uint64_t cur_perm, cur_shared; | |
9eab1544 | 2509 | |
e5d8a406 | 2510 | bdrv_child_perm(bs, c->bs, c, c->role, q, |
ffd1a5a2 FZ |
2511 | cumulative_perms, cumulative_shared_perms, |
2512 | &cur_perm, &cur_shared); | |
ecb776bd | 2513 | bdrv_child_set_perm(c, cur_perm, cur_shared, tran); |
bd57f8f7 VSO |
2514 | } |
2515 | ||
2516 | return 0; | |
2517 | } | |
3ef45e02 | 2518 | |
fb0ff4d1 VSO |
2519 | /* |
2520 | * @list is a product of bdrv_topological_dfs() (may be called several times) - | |
2521 | * a topologically sorted subgraph. | |
2522 | */ | |
2523 | static int bdrv_do_refresh_perms(GSList *list, BlockReopenQueue *q, | |
2524 | Transaction *tran, Error **errp) | |
bd57f8f7 VSO |
2525 | { |
2526 | int ret; | |
b1d2bbeb | 2527 | BlockDriverState *bs; |
862fded9 | 2528 | GLOBAL_STATE_CODE(); |
bd57f8f7 | 2529 | |
b1d2bbeb VSO |
2530 | for ( ; list; list = list->next) { |
2531 | bs = list->data; | |
2532 | ||
9397c14f | 2533 | if (bdrv_parent_perms_conflict(bs, errp)) { |
b1d2bbeb | 2534 | return -EINVAL; |
bd57f8f7 VSO |
2535 | } |
2536 | ||
c20555e1 | 2537 | ret = bdrv_node_refresh_perm(bs, q, tran, errp); |
33a610c3 KW |
2538 | if (ret < 0) { |
2539 | return ret; | |
2540 | } | |
2541 | } | |
2542 | ||
2543 | return 0; | |
2544 | } | |
2545 | ||
fb0ff4d1 VSO |
2546 | /* |
2547 | * @list is any list of nodes. List is completed by all subtrees and | |
2548 | * topologically sorted. It's not a problem if some node occurs in the @list | |
2549 | * several times. | |
2550 | */ | |
2551 | static int bdrv_list_refresh_perms(GSList *list, BlockReopenQueue *q, | |
2552 | Transaction *tran, Error **errp) | |
2553 | { | |
2554 | g_autoptr(GHashTable) found = g_hash_table_new(NULL, NULL); | |
2555 | g_autoptr(GSList) refresh_list = NULL; | |
2556 | ||
2557 | for ( ; list; list = list->next) { | |
2558 | refresh_list = bdrv_topological_dfs(refresh_list, found, list->data); | |
2559 | } | |
2560 | ||
2561 | return bdrv_do_refresh_perms(refresh_list, q, tran, errp); | |
2562 | } | |
2563 | ||
c7a0f2be KW |
2564 | void bdrv_get_cumulative_perm(BlockDriverState *bs, uint64_t *perm, |
2565 | uint64_t *shared_perm) | |
33a610c3 KW |
2566 | { |
2567 | BdrvChild *c; | |
2568 | uint64_t cumulative_perms = 0; | |
2569 | uint64_t cumulative_shared_perms = BLK_PERM_ALL; | |
2570 | ||
b4ad82aa EGE |
2571 | GLOBAL_STATE_CODE(); |
2572 | ||
33a610c3 KW |
2573 | QLIST_FOREACH(c, &bs->parents, next_parent) { |
2574 | cumulative_perms |= c->perm; | |
2575 | cumulative_shared_perms &= c->shared_perm; | |
2576 | } | |
2577 | ||
2578 | *perm = cumulative_perms; | |
2579 | *shared_perm = cumulative_shared_perms; | |
2580 | } | |
2581 | ||
5176196c | 2582 | char *bdrv_perm_names(uint64_t perm) |
d083319f KW |
2583 | { |
2584 | struct perm_name { | |
2585 | uint64_t perm; | |
2586 | const char *name; | |
2587 | } permissions[] = { | |
2588 | { BLK_PERM_CONSISTENT_READ, "consistent read" }, | |
2589 | { BLK_PERM_WRITE, "write" }, | |
2590 | { BLK_PERM_WRITE_UNCHANGED, "write unchanged" }, | |
2591 | { BLK_PERM_RESIZE, "resize" }, | |
d083319f KW |
2592 | { 0, NULL } |
2593 | }; | |
2594 | ||
e2a7423a | 2595 | GString *result = g_string_sized_new(30); |
d083319f KW |
2596 | struct perm_name *p; |
2597 | ||
2598 | for (p = permissions; p->name; p++) { | |
2599 | if (perm & p->perm) { | |
e2a7423a AG |
2600 | if (result->len > 0) { |
2601 | g_string_append(result, ", "); | |
2602 | } | |
2603 | g_string_append(result, p->name); | |
d083319f KW |
2604 | } |
2605 | } | |
2606 | ||
e2a7423a | 2607 | return g_string_free(result, FALSE); |
d083319f KW |
2608 | } |
2609 | ||
33a610c3 | 2610 | |
f1316edb VSO |
2611 | /* @tran is allowed to be NULL. In this case no rollback is possible */ |
2612 | static int bdrv_refresh_perms(BlockDriverState *bs, Transaction *tran, | |
2613 | Error **errp) | |
bb87e4d1 VSO |
2614 | { |
2615 | int ret; | |
f1316edb | 2616 | Transaction *local_tran = NULL; |
b1d2bbeb | 2617 | g_autoptr(GSList) list = bdrv_topological_dfs(NULL, NULL, bs); |
862fded9 | 2618 | GLOBAL_STATE_CODE(); |
bb87e4d1 | 2619 | |
f1316edb VSO |
2620 | if (!tran) { |
2621 | tran = local_tran = tran_new(); | |
2622 | } | |
2623 | ||
fb0ff4d1 | 2624 | ret = bdrv_do_refresh_perms(list, NULL, tran, errp); |
f1316edb VSO |
2625 | |
2626 | if (local_tran) { | |
2627 | tran_finalize(local_tran, ret); | |
2628 | } | |
bb87e4d1 | 2629 | |
b1d2bbeb | 2630 | return ret; |
bb87e4d1 VSO |
2631 | } |
2632 | ||
33a610c3 KW |
2633 | int bdrv_child_try_set_perm(BdrvChild *c, uint64_t perm, uint64_t shared, |
2634 | Error **errp) | |
2635 | { | |
1046779e | 2636 | Error *local_err = NULL; |
83928dc4 | 2637 | Transaction *tran = tran_new(); |
33a610c3 KW |
2638 | int ret; |
2639 | ||
b4ad82aa EGE |
2640 | GLOBAL_STATE_CODE(); |
2641 | ||
ecb776bd | 2642 | bdrv_child_set_perm(c, perm, shared, tran); |
83928dc4 | 2643 | |
f1316edb | 2644 | ret = bdrv_refresh_perms(c->bs, tran, &local_err); |
83928dc4 VSO |
2645 | |
2646 | tran_finalize(tran, ret); | |
2647 | ||
33a610c3 | 2648 | if (ret < 0) { |
071b474f VSO |
2649 | if ((perm & ~c->perm) || (c->shared_perm & ~shared)) { |
2650 | /* tighten permissions */ | |
1046779e HR |
2651 | error_propagate(errp, local_err); |
2652 | } else { | |
2653 | /* | |
2654 | * Our caller may intend to only loosen restrictions and | |
2655 | * does not expect this function to fail. Errors are not | |
2656 | * fatal in such a case, so we can just hide them from our | |
2657 | * caller. | |
2658 | */ | |
2659 | error_free(local_err); | |
2660 | ret = 0; | |
2661 | } | |
33a610c3 KW |
2662 | } |
2663 | ||
83928dc4 | 2664 | return ret; |
d5e6f437 KW |
2665 | } |
2666 | ||
c1087f12 HR |
2667 | int bdrv_child_refresh_perms(BlockDriverState *bs, BdrvChild *c, Error **errp) |
2668 | { | |
2669 | uint64_t parent_perms, parent_shared; | |
2670 | uint64_t perms, shared; | |
2671 | ||
b4ad82aa EGE |
2672 | GLOBAL_STATE_CODE(); |
2673 | ||
c1087f12 | 2674 | bdrv_get_cumulative_perm(bs, &parent_perms, &parent_shared); |
e5d8a406 | 2675 | bdrv_child_perm(bs, c->bs, c, c->role, NULL, |
bf8e925e | 2676 | parent_perms, parent_shared, &perms, &shared); |
c1087f12 HR |
2677 | |
2678 | return bdrv_child_try_set_perm(c, perms, shared, errp); | |
2679 | } | |
2680 | ||
87278af1 HR |
2681 | /* |
2682 | * Default implementation for .bdrv_child_perm() for block filters: | |
2683 | * Forward CONSISTENT_READ, WRITE, WRITE_UNCHANGED, and RESIZE to the | |
2684 | * filtered child. | |
2685 | */ | |
2686 | static void bdrv_filter_default_perms(BlockDriverState *bs, BdrvChild *c, | |
87278af1 HR |
2687 | BdrvChildRole role, |
2688 | BlockReopenQueue *reopen_queue, | |
2689 | uint64_t perm, uint64_t shared, | |
2690 | uint64_t *nperm, uint64_t *nshared) | |
6a1b9ee1 | 2691 | { |
862fded9 | 2692 | GLOBAL_STATE_CODE(); |
e444fa83 KW |
2693 | *nperm = perm & DEFAULT_PERM_PASSTHROUGH; |
2694 | *nshared = (shared & DEFAULT_PERM_PASSTHROUGH) | DEFAULT_PERM_UNCHANGED; | |
6a1b9ee1 KW |
2695 | } |
2696 | ||
70082db4 | 2697 | static void bdrv_default_perms_for_cow(BlockDriverState *bs, BdrvChild *c, |
70082db4 HR |
2698 | BdrvChildRole role, |
2699 | BlockReopenQueue *reopen_queue, | |
2700 | uint64_t perm, uint64_t shared, | |
2701 | uint64_t *nperm, uint64_t *nshared) | |
2702 | { | |
e5d8a406 | 2703 | assert(role & BDRV_CHILD_COW); |
862fded9 | 2704 | GLOBAL_STATE_CODE(); |
70082db4 HR |
2705 | |
2706 | /* | |
2707 | * We want consistent read from backing files if the parent needs it. | |
2708 | * No other operations are performed on backing files. | |
2709 | */ | |
2710 | perm &= BLK_PERM_CONSISTENT_READ; | |
2711 | ||
2712 | /* | |
2713 | * If the parent can deal with changing data, we're okay with a | |
2714 | * writable and resizable backing file. | |
2715 | * TODO Require !(perm & BLK_PERM_CONSISTENT_READ), too? | |
2716 | */ | |
2717 | if (shared & BLK_PERM_WRITE) { | |
2718 | shared = BLK_PERM_WRITE | BLK_PERM_RESIZE; | |
2719 | } else { | |
2720 | shared = 0; | |
2721 | } | |
2722 | ||
64631f36 | 2723 | shared |= BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE_UNCHANGED; |
70082db4 HR |
2724 | |
2725 | if (bs->open_flags & BDRV_O_INACTIVE) { | |
2726 | shared |= BLK_PERM_WRITE | BLK_PERM_RESIZE; | |
2727 | } | |
2728 | ||
2729 | *nperm = perm; | |
2730 | *nshared = shared; | |
2731 | } | |
2732 | ||
6f838a4b | 2733 | static void bdrv_default_perms_for_storage(BlockDriverState *bs, BdrvChild *c, |
6f838a4b HR |
2734 | BdrvChildRole role, |
2735 | BlockReopenQueue *reopen_queue, | |
2736 | uint64_t perm, uint64_t shared, | |
2737 | uint64_t *nperm, uint64_t *nshared) | |
2738 | { | |
2739 | int flags; | |
2740 | ||
862fded9 | 2741 | GLOBAL_STATE_CODE(); |
e5d8a406 | 2742 | assert(role & (BDRV_CHILD_METADATA | BDRV_CHILD_DATA)); |
6f838a4b HR |
2743 | |
2744 | flags = bdrv_reopen_get_flags(reopen_queue, bs); | |
2745 | ||
2746 | /* | |
2747 | * Apart from the modifications below, the same permissions are | |
2748 | * forwarded and left alone as for filters | |
2749 | */ | |
e5d8a406 | 2750 | bdrv_filter_default_perms(bs, c, role, reopen_queue, |
6f838a4b HR |
2751 | perm, shared, &perm, &shared); |
2752 | ||
f889054f HR |
2753 | if (role & BDRV_CHILD_METADATA) { |
2754 | /* Format drivers may touch metadata even if the guest doesn't write */ | |
2755 | if (bdrv_is_writable_after_reopen(bs, reopen_queue)) { | |
2756 | perm |= BLK_PERM_WRITE | BLK_PERM_RESIZE; | |
2757 | } | |
2758 | ||
2759 | /* | |
2760 | * bs->file always needs to be consistent because of the | |
2761 | * metadata. We can never allow other users to resize or write | |
2762 | * to it. | |
2763 | */ | |
2764 | if (!(flags & BDRV_O_NO_IO)) { | |
2765 | perm |= BLK_PERM_CONSISTENT_READ; | |
2766 | } | |
2767 | shared &= ~(BLK_PERM_WRITE | BLK_PERM_RESIZE); | |
6f838a4b HR |
2768 | } |
2769 | ||
f889054f HR |
2770 | if (role & BDRV_CHILD_DATA) { |
2771 | /* | |
2772 | * Technically, everything in this block is a subset of the | |
2773 | * BDRV_CHILD_METADATA path taken above, and so this could | |
2774 | * be an "else if" branch. However, that is not obvious, and | |
2775 | * this function is not performance critical, therefore we let | |
2776 | * this be an independent "if". | |
2777 | */ | |
2778 | ||
2779 | /* | |
2780 | * We cannot allow other users to resize the file because the | |
2781 | * format driver might have some assumptions about the size | |
2782 | * (e.g. because it is stored in metadata, or because the file | |
2783 | * is split into fixed-size data files). | |
2784 | */ | |
2785 | shared &= ~BLK_PERM_RESIZE; | |
2786 | ||
2787 | /* | |
2788 | * WRITE_UNCHANGED often cannot be performed as such on the | |
2789 | * data file. For example, the qcow2 driver may still need to | |
2790 | * write copied clusters on copy-on-read. | |
2791 | */ | |
2792 | if (perm & BLK_PERM_WRITE_UNCHANGED) { | |
2793 | perm |= BLK_PERM_WRITE; | |
2794 | } | |
2795 | ||
2796 | /* | |
2797 | * If the data file is written to, the format driver may | |
2798 | * expect to be able to resize it by writing beyond the EOF. | |
2799 | */ | |
2800 | if (perm & BLK_PERM_WRITE) { | |
2801 | perm |= BLK_PERM_RESIZE; | |
2802 | } | |
6f838a4b | 2803 | } |
6f838a4b HR |
2804 | |
2805 | if (bs->open_flags & BDRV_O_INACTIVE) { | |
2806 | shared |= BLK_PERM_WRITE | BLK_PERM_RESIZE; | |
2807 | } | |
2808 | ||
2809 | *nperm = perm; | |
2810 | *nshared = shared; | |
2811 | } | |
2812 | ||
2519f549 | 2813 | void bdrv_default_perms(BlockDriverState *bs, BdrvChild *c, |
e5d8a406 | 2814 | BdrvChildRole role, BlockReopenQueue *reopen_queue, |
2519f549 HR |
2815 | uint64_t perm, uint64_t shared, |
2816 | uint64_t *nperm, uint64_t *nshared) | |
2817 | { | |
b4ad82aa | 2818 | GLOBAL_STATE_CODE(); |
2519f549 HR |
2819 | if (role & BDRV_CHILD_FILTERED) { |
2820 | assert(!(role & (BDRV_CHILD_DATA | BDRV_CHILD_METADATA | | |
2821 | BDRV_CHILD_COW))); | |
e5d8a406 | 2822 | bdrv_filter_default_perms(bs, c, role, reopen_queue, |
2519f549 HR |
2823 | perm, shared, nperm, nshared); |
2824 | } else if (role & BDRV_CHILD_COW) { | |
2825 | assert(!(role & (BDRV_CHILD_DATA | BDRV_CHILD_METADATA))); | |
e5d8a406 | 2826 | bdrv_default_perms_for_cow(bs, c, role, reopen_queue, |
2519f549 HR |
2827 | perm, shared, nperm, nshared); |
2828 | } else if (role & (BDRV_CHILD_METADATA | BDRV_CHILD_DATA)) { | |
e5d8a406 | 2829 | bdrv_default_perms_for_storage(bs, c, role, reopen_queue, |
2519f549 HR |
2830 | perm, shared, nperm, nshared); |
2831 | } else { | |
2832 | g_assert_not_reached(); | |
2833 | } | |
2834 | } | |
2835 | ||
7b1d9c4d HR |
2836 | uint64_t bdrv_qapi_perm_to_blk_perm(BlockPermission qapi_perm) |
2837 | { | |
2838 | static const uint64_t permissions[] = { | |
2839 | [BLOCK_PERMISSION_CONSISTENT_READ] = BLK_PERM_CONSISTENT_READ, | |
2840 | [BLOCK_PERMISSION_WRITE] = BLK_PERM_WRITE, | |
2841 | [BLOCK_PERMISSION_WRITE_UNCHANGED] = BLK_PERM_WRITE_UNCHANGED, | |
2842 | [BLOCK_PERMISSION_RESIZE] = BLK_PERM_RESIZE, | |
7b1d9c4d HR |
2843 | }; |
2844 | ||
2845 | QEMU_BUILD_BUG_ON(ARRAY_SIZE(permissions) != BLOCK_PERMISSION__MAX); | |
2846 | QEMU_BUILD_BUG_ON(1UL << ARRAY_SIZE(permissions) != BLK_PERM_ALL + 1); | |
2847 | ||
2848 | assert(qapi_perm < BLOCK_PERMISSION__MAX); | |
2849 | ||
2850 | return permissions[qapi_perm]; | |
2851 | } | |
2852 | ||
23987471 KW |
2853 | /* |
2854 | * Replaces the node that a BdrvChild points to without updating permissions. | |
2855 | * | |
2856 | * If @new_bs is non-NULL, the parent of @child must already be drained through | |
2857 | * @child. | |
23987471 | 2858 | */ |
544acc7d | 2859 | static void bdrv_replace_child_noperm(BdrvChild *child, |
4eba825a | 2860 | BlockDriverState *new_bs) |
e9740bc6 KW |
2861 | { |
2862 | BlockDriverState *old_bs = child->bs; | |
debc2927 | 2863 | int new_bs_quiesce_counter; |
e9740bc6 | 2864 | |
2cad1ebe | 2865 | assert(!child->frozen); |
23987471 KW |
2866 | |
2867 | /* | |
2868 | * If we want to change the BdrvChild to point to a drained node as its new | |
2869 | * child->bs, we need to make sure that its new parent is drained, too. In | |
2870 | * other words, either child->quiesce_parent must already be true or we must | |
2871 | * be able to set it and keep the parent's quiesce_counter consistent with | |
2872 | * that, but without polling or starting new requests (this function | |
2873 | * guarantees that it doesn't poll, and starting new requests would be | |
2874 | * against the invariants of drain sections). | |
2875 | * | |
2876 | * To keep things simple, we pick the first option (child->quiesce_parent | |
2877 | * must already be true). We also generalise the rule a bit to make it | |
2878 | * easier to verify in callers and more likely to be covered in test cases: | |
2879 | * The parent must be quiesced through this child even if new_bs isn't | |
2880 | * currently drained. | |
2881 | * | |
2882 | * The only exception is for callers that always pass new_bs == NULL. In | |
2883 | * this case, we obviously never need to consider the case of a drained | |
2884 | * new_bs, so we can keep the callers simpler by allowing them not to drain | |
2885 | * the parent. | |
2886 | */ | |
2887 | assert(!new_bs || child->quiesced_parent); | |
bfb8aa6d | 2888 | assert(old_bs != new_bs); |
f0c28327 | 2889 | GLOBAL_STATE_CODE(); |
2cad1ebe | 2890 | |
bb2614e9 FZ |
2891 | if (old_bs && new_bs) { |
2892 | assert(bdrv_get_aio_context(old_bs) == bdrv_get_aio_context(new_bs)); | |
2893 | } | |
debc2927 | 2894 | |
29312507 EGE |
2895 | /* TODO Pull this up into the callers to avoid polling here */ |
2896 | bdrv_graph_wrlock(); | |
e9740bc6 | 2897 | if (old_bs) { |
bd86fb99 HR |
2898 | if (child->klass->detach) { |
2899 | child->klass->detach(child); | |
d736f119 | 2900 | } |
e9740bc6 KW |
2901 | QLIST_REMOVE(child, next_parent); |
2902 | } | |
36fe1331 KW |
2903 | |
2904 | child->bs = new_bs; | |
2905 | ||
e9740bc6 KW |
2906 | if (new_bs) { |
2907 | QLIST_INSERT_HEAD(&new_bs->parents, child, next_parent); | |
bd86fb99 HR |
2908 | if (child->klass->attach) { |
2909 | child->klass->attach(child); | |
8ee03995 KW |
2910 | } |
2911 | } | |
29312507 | 2912 | bdrv_graph_wrunlock(); |
debc2927 HR |
2913 | |
2914 | /* | |
23987471 KW |
2915 | * If the parent was drained through this BdrvChild previously, but new_bs |
2916 | * is not drained, allow requests to come in only after the new node has | |
2917 | * been attached. | |
debc2927 | 2918 | */ |
57e05be3 KW |
2919 | new_bs_quiesce_counter = (new_bs ? new_bs->quiesce_counter : 0); |
2920 | if (!new_bs_quiesce_counter && child->quiesced_parent) { | |
debc2927 | 2921 | bdrv_parent_drained_end_single(child); |
debc2927 | 2922 | } |
8ee03995 | 2923 | } |
33a610c3 | 2924 | |
04c9c3a5 HR |
2925 | /** |
2926 | * Free the given @child. | |
2927 | * | |
2928 | * The child must be empty (i.e. `child->bs == NULL`) and it must be | |
2929 | * unused (i.e. not in a children list). | |
2930 | */ | |
2931 | static void bdrv_child_free(BdrvChild *child) | |
df581792 | 2932 | { |
548a74c0 | 2933 | assert(!child->bs); |
bdb73476 | 2934 | GLOBAL_STATE_CODE(); |
a225369b | 2935 | assert(!child->next.le_prev); /* not in children list */ |
04c9c3a5 HR |
2936 | |
2937 | g_free(child->name); | |
2938 | g_free(child); | |
548a74c0 | 2939 | } |
d5e6f437 | 2940 | |
548a74c0 | 2941 | typedef struct BdrvAttachChildCommonState { |
5bb04747 | 2942 | BdrvChild *child; |
548a74c0 VSO |
2943 | AioContext *old_parent_ctx; |
2944 | AioContext *old_child_ctx; | |
2945 | } BdrvAttachChildCommonState; | |
2946 | ||
2947 | static void bdrv_attach_child_common_abort(void *opaque) | |
2948 | { | |
2949 | BdrvAttachChildCommonState *s = opaque; | |
5bb04747 | 2950 | BlockDriverState *bs = s->child->bs; |
548a74c0 | 2951 | |
f0c28327 | 2952 | GLOBAL_STATE_CODE(); |
5bb04747 | 2953 | bdrv_replace_child_noperm(s->child, NULL); |
548a74c0 VSO |
2954 | |
2955 | if (bdrv_get_aio_context(bs) != s->old_child_ctx) { | |
142e6907 | 2956 | bdrv_try_change_aio_context(bs, s->old_child_ctx, NULL, &error_abort); |
548a74c0 VSO |
2957 | } |
2958 | ||
5bb04747 | 2959 | if (bdrv_child_get_parent_aio_context(s->child) != s->old_parent_ctx) { |
f8be48ad EGE |
2960 | Transaction *tran; |
2961 | GHashTable *visited; | |
2962 | bool ret; | |
548a74c0 | 2963 | |
f8be48ad | 2964 | tran = tran_new(); |
548a74c0 | 2965 | |
f8be48ad EGE |
2966 | /* No need to visit `child`, because it has been detached already */ |
2967 | visited = g_hash_table_new(NULL, NULL); | |
2968 | ret = s->child->klass->change_aio_ctx(s->child, s->old_parent_ctx, | |
2969 | visited, tran, &error_abort); | |
2970 | g_hash_table_destroy(visited); | |
2971 | ||
2972 | /* transaction is supposed to always succeed */ | |
2973 | assert(ret == true); | |
2974 | tran_commit(tran); | |
d5e6f437 KW |
2975 | } |
2976 | ||
548a74c0 | 2977 | bdrv_unref(bs); |
5bb04747 | 2978 | bdrv_child_free(s->child); |
548a74c0 VSO |
2979 | } |
2980 | ||
2981 | static TransactionActionDrv bdrv_attach_child_common_drv = { | |
2982 | .abort = bdrv_attach_child_common_abort, | |
2983 | .clean = g_free, | |
2984 | }; | |
2985 | ||
2986 | /* | |
2987 | * Common part of attaching bdrv child to bs or to blk or to job | |
f8d2ad78 | 2988 | * |
7ec390d5 | 2989 | * Function doesn't update permissions, caller is responsible for this. |
5bb04747 VSO |
2990 | * |
2991 | * Returns new created child. | |
548a74c0 | 2992 | */ |
5bb04747 VSO |
2993 | static BdrvChild *bdrv_attach_child_common(BlockDriverState *child_bs, |
2994 | const char *child_name, | |
2995 | const BdrvChildClass *child_class, | |
2996 | BdrvChildRole child_role, | |
2997 | uint64_t perm, uint64_t shared_perm, | |
2998 | void *opaque, | |
2999 | Transaction *tran, Error **errp) | |
548a74c0 VSO |
3000 | { |
3001 | BdrvChild *new_child; | |
3002 | AioContext *parent_ctx; | |
3003 | AioContext *child_ctx = bdrv_get_aio_context(child_bs); | |
3004 | ||
da261b69 | 3005 | assert(child_class->get_parent_desc); |
bdb73476 | 3006 | GLOBAL_STATE_CODE(); |
548a74c0 VSO |
3007 | |
3008 | new_child = g_new(BdrvChild, 1); | |
3009 | *new_child = (BdrvChild) { | |
d5e6f437 KW |
3010 | .bs = NULL, |
3011 | .name = g_strdup(child_name), | |
bd86fb99 | 3012 | .klass = child_class, |
258b7765 | 3013 | .role = child_role, |
d5e6f437 KW |
3014 | .perm = perm, |
3015 | .shared_perm = shared_perm, | |
3016 | .opaque = opaque, | |
df581792 KW |
3017 | }; |
3018 | ||
548a74c0 VSO |
3019 | /* |
3020 | * If the AioContexts don't match, first try to move the subtree of | |
132ada80 | 3021 | * child_bs into the AioContext of the new parent. If this doesn't work, |
548a74c0 VSO |
3022 | * try moving the parent into the AioContext of child_bs instead. |
3023 | */ | |
3024 | parent_ctx = bdrv_child_get_parent_aio_context(new_child); | |
3025 | if (child_ctx != parent_ctx) { | |
3026 | Error *local_err = NULL; | |
142e6907 EGE |
3027 | int ret = bdrv_try_change_aio_context(child_bs, parent_ctx, NULL, |
3028 | &local_err); | |
548a74c0 | 3029 | |
f8be48ad EGE |
3030 | if (ret < 0 && child_class->change_aio_ctx) { |
3031 | Transaction *tran = tran_new(); | |
3032 | GHashTable *visited = g_hash_table_new(NULL, NULL); | |
3033 | bool ret_child; | |
3034 | ||
3035 | g_hash_table_add(visited, new_child); | |
3036 | ret_child = child_class->change_aio_ctx(new_child, child_ctx, | |
3037 | visited, tran, NULL); | |
3038 | if (ret_child == true) { | |
132ada80 KW |
3039 | error_free(local_err); |
3040 | ret = 0; | |
132ada80 | 3041 | } |
f8be48ad EGE |
3042 | tran_finalize(tran, ret_child == true ? 0 : -1); |
3043 | g_hash_table_destroy(visited); | |
132ada80 | 3044 | } |
548a74c0 | 3045 | |
132ada80 KW |
3046 | if (ret < 0) { |
3047 | error_propagate(errp, local_err); | |
04c9c3a5 | 3048 | bdrv_child_free(new_child); |
5bb04747 | 3049 | return NULL; |
132ada80 KW |
3050 | } |
3051 | } | |
3052 | ||
548a74c0 | 3053 | bdrv_ref(child_bs); |
23987471 KW |
3054 | /* |
3055 | * Let every new BdrvChild start with a drained parent. Inserting the child | |
3056 | * in the graph with bdrv_replace_child_noperm() will undrain it if | |
3057 | * @child_bs is not drained. | |
3058 | * | |
3059 | * The child was only just created and is not yet visible in global state | |
3060 | * until bdrv_replace_child_noperm() inserts it into the graph, so nobody | |
3061 | * could have sent requests and polling is not necessary. | |
3062 | * | |
3063 | * Note that this means that the parent isn't fully drained yet, we only | |
3064 | * stop new requests from coming in. This is fine, we don't care about the | |
3065 | * old requests here, they are not for this child. If another place enters a | |
3066 | * drain section for the same parent, but wants it to be fully quiesced, it | |
3067 | * will not run most of the the code in .drained_begin() again (which is not | |
3068 | * a problem, we already did this), but it will still poll until the parent | |
3069 | * is fully quiesced, so it will not be negatively affected either. | |
3070 | */ | |
606ed756 | 3071 | bdrv_parent_drained_begin_single(new_child); |
544acc7d | 3072 | bdrv_replace_child_noperm(new_child, child_bs); |
b4b059f6 | 3073 | |
548a74c0 VSO |
3074 | BdrvAttachChildCommonState *s = g_new(BdrvAttachChildCommonState, 1); |
3075 | *s = (BdrvAttachChildCommonState) { | |
5bb04747 | 3076 | .child = new_child, |
548a74c0 VSO |
3077 | .old_parent_ctx = parent_ctx, |
3078 | .old_child_ctx = child_ctx, | |
3079 | }; | |
3080 | tran_add(tran, &bdrv_attach_child_common_drv, s); | |
3081 | ||
5bb04747 | 3082 | return new_child; |
548a74c0 VSO |
3083 | } |
3084 | ||
f8d2ad78 | 3085 | /* |
7ec390d5 | 3086 | * Function doesn't update permissions, caller is responsible for this. |
f8d2ad78 | 3087 | */ |
5bb04747 VSO |
3088 | static BdrvChild *bdrv_attach_child_noperm(BlockDriverState *parent_bs, |
3089 | BlockDriverState *child_bs, | |
3090 | const char *child_name, | |
3091 | const BdrvChildClass *child_class, | |
3092 | BdrvChildRole child_role, | |
3093 | Transaction *tran, | |
3094 | Error **errp) | |
aa5a04c7 | 3095 | { |
aa5a04c7 VSO |
3096 | uint64_t perm, shared_perm; |
3097 | ||
3098 | assert(parent_bs->drv); | |
bdb73476 | 3099 | GLOBAL_STATE_CODE(); |
aa5a04c7 | 3100 | |
bfb8aa6d KW |
3101 | if (bdrv_recurse_has_child(child_bs, parent_bs)) { |
3102 | error_setg(errp, "Making '%s' a %s child of '%s' would create a cycle", | |
3103 | child_bs->node_name, child_name, parent_bs->node_name); | |
5bb04747 | 3104 | return NULL; |
bfb8aa6d KW |
3105 | } |
3106 | ||
aa5a04c7 VSO |
3107 | bdrv_get_cumulative_perm(parent_bs, &perm, &shared_perm); |
3108 | bdrv_child_perm(parent_bs, child_bs, NULL, child_role, NULL, | |
3109 | perm, shared_perm, &perm, &shared_perm); | |
3110 | ||
5bb04747 VSO |
3111 | return bdrv_attach_child_common(child_bs, child_name, child_class, |
3112 | child_role, perm, shared_perm, parent_bs, | |
3113 | tran, errp); | |
aa5a04c7 VSO |
3114 | } |
3115 | ||
548a74c0 VSO |
3116 | /* |
3117 | * This function steals the reference to child_bs from the caller. | |
3118 | * That reference is later dropped by bdrv_root_unref_child(). | |
3119 | * | |
3120 | * On failure NULL is returned, errp is set and the reference to | |
3121 | * child_bs is also dropped. | |
3122 | * | |
3123 | * The caller must hold the AioContext lock @child_bs, but not that of @ctx | |
3124 | * (unless @child_bs is already in @ctx). | |
3125 | */ | |
3126 | BdrvChild *bdrv_root_attach_child(BlockDriverState *child_bs, | |
3127 | const char *child_name, | |
3128 | const BdrvChildClass *child_class, | |
3129 | BdrvChildRole child_role, | |
3130 | uint64_t perm, uint64_t shared_perm, | |
3131 | void *opaque, Error **errp) | |
3132 | { | |
3133 | int ret; | |
5bb04747 | 3134 | BdrvChild *child; |
548a74c0 VSO |
3135 | Transaction *tran = tran_new(); |
3136 | ||
b4ad82aa EGE |
3137 | GLOBAL_STATE_CODE(); |
3138 | ||
5bb04747 | 3139 | child = bdrv_attach_child_common(child_bs, child_name, child_class, |
548a74c0 | 3140 | child_role, perm, shared_perm, opaque, |
5bb04747 VSO |
3141 | tran, errp); |
3142 | if (!child) { | |
3143 | ret = -EINVAL; | |
e878bb12 | 3144 | goto out; |
548a74c0 VSO |
3145 | } |
3146 | ||
f1316edb | 3147 | ret = bdrv_refresh_perms(child_bs, tran, errp); |
548a74c0 | 3148 | |
e878bb12 KW |
3149 | out: |
3150 | tran_finalize(tran, ret); | |
f8d2ad78 | 3151 | |
548a74c0 | 3152 | bdrv_unref(child_bs); |
5bb04747 VSO |
3153 | |
3154 | return ret < 0 ? NULL : child; | |
df581792 KW |
3155 | } |
3156 | ||
b441dc71 AG |
3157 | /* |
3158 | * This function transfers the reference to child_bs from the caller | |
3159 | * to parent_bs. That reference is later dropped by parent_bs on | |
3160 | * bdrv_close() or if someone calls bdrv_unref_child(). | |
3161 | * | |
3162 | * On failure NULL is returned, errp is set and the reference to | |
3163 | * child_bs is also dropped. | |
132ada80 KW |
3164 | * |
3165 | * If @parent_bs and @child_bs are in different AioContexts, the caller must | |
3166 | * hold the AioContext lock for @child_bs, but not for @parent_bs. | |
b441dc71 | 3167 | */ |
98292c61 WC |
3168 | BdrvChild *bdrv_attach_child(BlockDriverState *parent_bs, |
3169 | BlockDriverState *child_bs, | |
3170 | const char *child_name, | |
bd86fb99 | 3171 | const BdrvChildClass *child_class, |
258b7765 | 3172 | BdrvChildRole child_role, |
8b2ff529 | 3173 | Error **errp) |
f21d96d0 | 3174 | { |
aa5a04c7 | 3175 | int ret; |
5bb04747 | 3176 | BdrvChild *child; |
aa5a04c7 | 3177 | Transaction *tran = tran_new(); |
f68c598b | 3178 | |
f791bf7f EGE |
3179 | GLOBAL_STATE_CODE(); |
3180 | ||
5bb04747 VSO |
3181 | child = bdrv_attach_child_noperm(parent_bs, child_bs, child_name, |
3182 | child_class, child_role, tran, errp); | |
3183 | if (!child) { | |
3184 | ret = -EINVAL; | |
aa5a04c7 VSO |
3185 | goto out; |
3186 | } | |
d5e6f437 | 3187 | |
f1316edb | 3188 | ret = bdrv_refresh_perms(parent_bs, tran, errp); |
aa5a04c7 VSO |
3189 | if (ret < 0) { |
3190 | goto out; | |
d5e6f437 KW |
3191 | } |
3192 | ||
aa5a04c7 VSO |
3193 | out: |
3194 | tran_finalize(tran, ret); | |
3195 | ||
3196 | bdrv_unref(child_bs); | |
3197 | ||
5bb04747 | 3198 | return ret < 0 ? NULL : child; |
f21d96d0 KW |
3199 | } |
3200 | ||
7b99a266 | 3201 | /* Callers must ensure that child->frozen is false. */ |
f21d96d0 | 3202 | void bdrv_root_unref_child(BdrvChild *child) |
33a60407 | 3203 | { |
00eb93b5 | 3204 | BlockDriverState *child_bs = child->bs; |
779020cb | 3205 | |
f791bf7f | 3206 | GLOBAL_STATE_CODE(); |
00eb93b5 VSO |
3207 | bdrv_replace_child_noperm(child, NULL); |
3208 | bdrv_child_free(child); | |
3209 | ||
3210 | if (child_bs) { | |
3211 | /* | |
3212 | * Update permissions for old node. We're just taking a parent away, so | |
3213 | * we're loosening restrictions. Errors of permission update are not | |
3214 | * fatal in this case, ignore them. | |
3215 | */ | |
f1316edb | 3216 | bdrv_refresh_perms(child_bs, NULL, NULL); |
00eb93b5 VSO |
3217 | |
3218 | /* | |
3219 | * When the parent requiring a non-default AioContext is removed, the | |
3220 | * node moves back to the main AioContext | |
3221 | */ | |
3222 | bdrv_try_change_aio_context(child_bs, qemu_get_aio_context(), NULL, | |
3223 | NULL); | |
3224 | } | |
f791bf7f | 3225 | |
f21d96d0 KW |
3226 | bdrv_unref(child_bs); |
3227 | } | |
3228 | ||
332b3a17 VSO |
3229 | typedef struct BdrvSetInheritsFrom { |
3230 | BlockDriverState *bs; | |
3231 | BlockDriverState *old_inherits_from; | |
3232 | } BdrvSetInheritsFrom; | |
3233 | ||
3234 | static void bdrv_set_inherits_from_abort(void *opaque) | |
3235 | { | |
3236 | BdrvSetInheritsFrom *s = opaque; | |
3237 | ||
3238 | s->bs->inherits_from = s->old_inherits_from; | |
3239 | } | |
3240 | ||
3241 | static TransactionActionDrv bdrv_set_inherits_from_drv = { | |
3242 | .abort = bdrv_set_inherits_from_abort, | |
3243 | .clean = g_free, | |
3244 | }; | |
3245 | ||
3246 | /* @tran is allowed to be NULL. In this case no rollback is possible */ | |
3247 | static void bdrv_set_inherits_from(BlockDriverState *bs, | |
3248 | BlockDriverState *new_inherits_from, | |
3249 | Transaction *tran) | |
3250 | { | |
3251 | if (tran) { | |
3252 | BdrvSetInheritsFrom *s = g_new(BdrvSetInheritsFrom, 1); | |
3253 | ||
3254 | *s = (BdrvSetInheritsFrom) { | |
3255 | .bs = bs, | |
3256 | .old_inherits_from = bs->inherits_from, | |
3257 | }; | |
3258 | ||
3259 | tran_add(tran, &bdrv_set_inherits_from_drv, s); | |
3260 | } | |
3261 | ||
3262 | bs->inherits_from = new_inherits_from; | |
3263 | } | |
3264 | ||
3cf746b3 HR |
3265 | /** |
3266 | * Clear all inherits_from pointers from children and grandchildren of | |
3267 | * @root that point to @root, where necessary. | |
332b3a17 | 3268 | * @tran is allowed to be NULL. In this case no rollback is possible |
3cf746b3 | 3269 | */ |
332b3a17 VSO |
3270 | static void bdrv_unset_inherits_from(BlockDriverState *root, BdrvChild *child, |
3271 | Transaction *tran) | |
f21d96d0 | 3272 | { |
3cf746b3 | 3273 | BdrvChild *c; |
4e4bf5c4 | 3274 | |
3cf746b3 HR |
3275 | if (child->bs->inherits_from == root) { |
3276 | /* | |
3277 | * Remove inherits_from only when the last reference between root and | |
3278 | * child->bs goes away. | |
3279 | */ | |
3280 | QLIST_FOREACH(c, &root->children, next) { | |
4e4bf5c4 KW |
3281 | if (c != child && c->bs == child->bs) { |
3282 | break; | |
3283 | } | |
3284 | } | |
3285 | if (c == NULL) { | |
332b3a17 | 3286 | bdrv_set_inherits_from(child->bs, NULL, tran); |
4e4bf5c4 | 3287 | } |
33a60407 KW |
3288 | } |
3289 | ||
3cf746b3 | 3290 | QLIST_FOREACH(c, &child->bs->children, next) { |
332b3a17 | 3291 | bdrv_unset_inherits_from(root, c, tran); |
3cf746b3 HR |
3292 | } |
3293 | } | |
3294 | ||
7b99a266 | 3295 | /* Callers must ensure that child->frozen is false. */ |
3cf746b3 HR |
3296 | void bdrv_unref_child(BlockDriverState *parent, BdrvChild *child) |
3297 | { | |
f791bf7f | 3298 | GLOBAL_STATE_CODE(); |
3cf746b3 HR |
3299 | if (child == NULL) { |
3300 | return; | |
3301 | } | |
3302 | ||
332b3a17 | 3303 | bdrv_unset_inherits_from(parent, child, NULL); |
f21d96d0 | 3304 | bdrv_root_unref_child(child); |
33a60407 KW |
3305 | } |
3306 | ||
5c8cab48 KW |
3307 | |
3308 | static void bdrv_parent_cb_change_media(BlockDriverState *bs, bool load) | |
3309 | { | |
3310 | BdrvChild *c; | |
f0c28327 | 3311 | GLOBAL_STATE_CODE(); |
5c8cab48 | 3312 | QLIST_FOREACH(c, &bs->parents, next_parent) { |
bd86fb99 HR |
3313 | if (c->klass->change_media) { |
3314 | c->klass->change_media(c, load); | |
5c8cab48 KW |
3315 | } |
3316 | } | |
3317 | } | |
3318 | ||
0065c455 AG |
3319 | /* Return true if you can reach parent going through child->inherits_from |
3320 | * recursively. If parent or child are NULL, return false */ | |
3321 | static bool bdrv_inherits_from_recursive(BlockDriverState *child, | |
3322 | BlockDriverState *parent) | |
3323 | { | |
3324 | while (child && child != parent) { | |
3325 | child = child->inherits_from; | |
3326 | } | |
3327 | ||
3328 | return child != NULL; | |
3329 | } | |
3330 | ||
25191e5f HR |
3331 | /* |
3332 | * Return the BdrvChildRole for @bs's backing child. bs->backing is | |
3333 | * mostly used for COW backing children (role = COW), but also for | |
3334 | * filtered children (role = FILTERED | PRIMARY). | |
3335 | */ | |
3336 | static BdrvChildRole bdrv_backing_role(BlockDriverState *bs) | |
3337 | { | |
3338 | if (bs->drv && bs->drv->is_filter) { | |
3339 | return BDRV_CHILD_FILTERED | BDRV_CHILD_PRIMARY; | |
3340 | } else { | |
3341 | return BDRV_CHILD_COW; | |
3342 | } | |
3343 | } | |
3344 | ||
5db15a57 | 3345 | /* |
e9238278 VSO |
3346 | * Sets the bs->backing or bs->file link of a BDS. A new reference is created; |
3347 | * callers which don't need their own reference any more must call bdrv_unref(). | |
7ec390d5 VSO |
3348 | * |
3349 | * Function doesn't update permissions, caller is responsible for this. | |
5db15a57 | 3350 | */ |
e9238278 VSO |
3351 | static int bdrv_set_file_or_backing_noperm(BlockDriverState *parent_bs, |
3352 | BlockDriverState *child_bs, | |
3353 | bool is_backing, | |
3354 | Transaction *tran, Error **errp) | |
8d24cce1 | 3355 | { |
e9238278 VSO |
3356 | bool update_inherits_from = |
3357 | bdrv_inherits_from_recursive(child_bs, parent_bs); | |
3358 | BdrvChild *child = is_backing ? parent_bs->backing : parent_bs->file; | |
3359 | BdrvChildRole role; | |
0065c455 | 3360 | |
bdb73476 EGE |
3361 | GLOBAL_STATE_CODE(); |
3362 | ||
e9238278 VSO |
3363 | if (!parent_bs->drv) { |
3364 | /* | |
3365 | * Node without drv is an object without a class :/. TODO: finally fix | |
3366 | * qcow2 driver to never clear bs->drv and implement format corruption | |
3367 | * handling in other way. | |
3368 | */ | |
3369 | error_setg(errp, "Node corrupted"); | |
3370 | return -EINVAL; | |
3371 | } | |
3372 | ||
3373 | if (child && child->frozen) { | |
3374 | error_setg(errp, "Cannot change frozen '%s' link from '%s' to '%s'", | |
3375 | child->name, parent_bs->node_name, child->bs->node_name); | |
a1e708fc | 3376 | return -EPERM; |
2cad1ebe AG |
3377 | } |
3378 | ||
25f78d9e VSO |
3379 | if (is_backing && !parent_bs->drv->is_filter && |
3380 | !parent_bs->drv->supports_backing) | |
3381 | { | |
3382 | error_setg(errp, "Driver '%s' of node '%s' does not support backing " | |
3383 | "files", parent_bs->drv->format_name, parent_bs->node_name); | |
3384 | return -EINVAL; | |
3385 | } | |
3386 | ||
e9238278 VSO |
3387 | if (parent_bs->drv->is_filter) { |
3388 | role = BDRV_CHILD_FILTERED | BDRV_CHILD_PRIMARY; | |
3389 | } else if (is_backing) { | |
3390 | role = BDRV_CHILD_COW; | |
3391 | } else { | |
3392 | /* | |
3393 | * We only can use same role as it is in existing child. We don't have | |
3394 | * infrastructure to determine role of file child in generic way | |
3395 | */ | |
3396 | if (!child) { | |
3397 | error_setg(errp, "Cannot set file child to format node without " | |
3398 | "file child"); | |
3399 | return -EINVAL; | |
3400 | } | |
3401 | role = child->role; | |
826b6ca0 FZ |
3402 | } |
3403 | ||
e9238278 VSO |
3404 | if (child) { |
3405 | bdrv_unset_inherits_from(parent_bs, child, tran); | |
57f08941 | 3406 | bdrv_remove_child(child, tran); |
e9238278 VSO |
3407 | } |
3408 | ||
3409 | if (!child_bs) { | |
8d24cce1 FZ |
3410 | goto out; |
3411 | } | |
12fa4af6 | 3412 | |
5bb04747 VSO |
3413 | child = bdrv_attach_child_noperm(parent_bs, child_bs, |
3414 | is_backing ? "backing" : "file", | |
3415 | &child_of_bds, role, | |
3416 | tran, errp); | |
3417 | if (!child) { | |
3418 | return -EINVAL; | |
a1e708fc VSO |
3419 | } |
3420 | ||
160333e1 VSO |
3421 | |
3422 | /* | |
e9238278 | 3423 | * If inherits_from pointed recursively to bs then let's update it to |
160333e1 VSO |
3424 | * point directly to bs (else it will become NULL). |
3425 | */ | |
a1e708fc | 3426 | if (update_inherits_from) { |
e9238278 | 3427 | bdrv_set_inherits_from(child_bs, parent_bs, tran); |
0065c455 | 3428 | } |
826b6ca0 | 3429 | |
8d24cce1 | 3430 | out: |
e19b157f | 3431 | bdrv_graph_rdlock_main_loop(); |
e9238278 | 3432 | bdrv_refresh_limits(parent_bs, tran, NULL); |
e19b157f | 3433 | bdrv_graph_rdunlock_main_loop(); |
160333e1 VSO |
3434 | |
3435 | return 0; | |
3436 | } | |
3437 | ||
e9238278 VSO |
3438 | static int bdrv_set_backing_noperm(BlockDriverState *bs, |
3439 | BlockDriverState *backing_hd, | |
3440 | Transaction *tran, Error **errp) | |
3441 | { | |
bdb73476 | 3442 | GLOBAL_STATE_CODE(); |
e9238278 VSO |
3443 | return bdrv_set_file_or_backing_noperm(bs, backing_hd, true, tran, errp); |
3444 | } | |
3445 | ||
92140b9f KW |
3446 | int bdrv_set_backing_hd_drained(BlockDriverState *bs, |
3447 | BlockDriverState *backing_hd, | |
3448 | Error **errp) | |
160333e1 VSO |
3449 | { |
3450 | int ret; | |
3451 | Transaction *tran = tran_new(); | |
3452 | ||
f791bf7f | 3453 | GLOBAL_STATE_CODE(); |
92140b9f | 3454 | assert(bs->quiesce_counter > 0); |
c0829cb1 | 3455 | |
160333e1 VSO |
3456 | ret = bdrv_set_backing_noperm(bs, backing_hd, tran, errp); |
3457 | if (ret < 0) { | |
3458 | goto out; | |
3459 | } | |
3460 | ||
f1316edb | 3461 | ret = bdrv_refresh_perms(bs, tran, errp); |
160333e1 VSO |
3462 | out: |
3463 | tran_finalize(tran, ret); | |
92140b9f KW |
3464 | return ret; |
3465 | } | |
a1e708fc | 3466 | |
92140b9f KW |
3467 | int bdrv_set_backing_hd(BlockDriverState *bs, BlockDriverState *backing_hd, |
3468 | Error **errp) | |
3469 | { | |
3470 | int ret; | |
3471 | GLOBAL_STATE_CODE(); | |
3472 | ||
3473 | bdrv_drained_begin(bs); | |
3474 | ret = bdrv_set_backing_hd_drained(bs, backing_hd, errp); | |
c0829cb1 VSO |
3475 | bdrv_drained_end(bs); |
3476 | ||
a1e708fc | 3477 | return ret; |
8d24cce1 FZ |
3478 | } |
3479 | ||
31ca6d07 KW |
3480 | /* |
3481 | * Opens the backing file for a BlockDriverState if not yet open | |
3482 | * | |
d9b7b057 KW |
3483 | * bdref_key specifies the key for the image's BlockdevRef in the options QDict. |
3484 | * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict | |
3485 | * itself, all options starting with "${bdref_key}." are considered part of the | |
3486 | * BlockdevRef. | |
3487 | * | |
2626d27f KW |
3488 | * The caller must hold the main AioContext lock. |
3489 | * | |
d9b7b057 | 3490 | * TODO Can this be unified with bdrv_open_image()? |
31ca6d07 | 3491 | */ |
d9b7b057 KW |
3492 | int bdrv_open_backing_file(BlockDriverState *bs, QDict *parent_options, |
3493 | const char *bdref_key, Error **errp) | |
9156df12 | 3494 | { |
6b6833c1 | 3495 | char *backing_filename = NULL; |
d9b7b057 KW |
3496 | char *bdref_key_dot; |
3497 | const char *reference = NULL; | |
317fc44e | 3498 | int ret = 0; |
998c2019 | 3499 | bool implicit_backing = false; |
8d24cce1 | 3500 | BlockDriverState *backing_hd; |
d9b7b057 KW |
3501 | QDict *options; |
3502 | QDict *tmp_parent_options = NULL; | |
34b5d2c6 | 3503 | Error *local_err = NULL; |
9156df12 | 3504 | |
f791bf7f EGE |
3505 | GLOBAL_STATE_CODE(); |
3506 | ||
760e0063 | 3507 | if (bs->backing != NULL) { |
1ba4b6a5 | 3508 | goto free_exit; |
9156df12 PB |
3509 | } |
3510 | ||
31ca6d07 | 3511 | /* NULL means an empty set of options */ |
d9b7b057 KW |
3512 | if (parent_options == NULL) { |
3513 | tmp_parent_options = qdict_new(); | |
3514 | parent_options = tmp_parent_options; | |
31ca6d07 KW |
3515 | } |
3516 | ||
9156df12 | 3517 | bs->open_flags &= ~BDRV_O_NO_BACKING; |
d9b7b057 KW |
3518 | |
3519 | bdref_key_dot = g_strdup_printf("%s.", bdref_key); | |
3520 | qdict_extract_subqdict(parent_options, &options, bdref_key_dot); | |
3521 | g_free(bdref_key_dot); | |
3522 | ||
129c7d1c MA |
3523 | /* |
3524 | * Caution: while qdict_get_try_str() is fine, getting non-string | |
3525 | * types would require more care. When @parent_options come from | |
3526 | * -blockdev or blockdev_add, its members are typed according to | |
3527 | * the QAPI schema, but when they come from -drive, they're all | |
3528 | * QString. | |
3529 | */ | |
d9b7b057 KW |
3530 | reference = qdict_get_try_str(parent_options, bdref_key); |
3531 | if (reference || qdict_haskey(options, "file.filename")) { | |
6b6833c1 | 3532 | /* keep backing_filename NULL */ |
1cb6f506 | 3533 | } else if (bs->backing_file[0] == '\0' && qdict_size(options) == 0) { |
cb3e7f08 | 3534 | qobject_unref(options); |
1ba4b6a5 | 3535 | goto free_exit; |
dbecebdd | 3536 | } else { |
998c2019 HR |
3537 | if (qdict_size(options) == 0) { |
3538 | /* If the user specifies options that do not modify the | |
3539 | * backing file's behavior, we might still consider it the | |
3540 | * implicit backing file. But it's easier this way, and | |
3541 | * just specifying some of the backing BDS's options is | |
3542 | * only possible with -drive anyway (otherwise the QAPI | |
3543 | * schema forces the user to specify everything). */ | |
3544 | implicit_backing = !strcmp(bs->auto_backing_file, bs->backing_file); | |
3545 | } | |
3546 | ||
6b6833c1 | 3547 | backing_filename = bdrv_get_full_backing_filename(bs, &local_err); |
9f07429e HR |
3548 | if (local_err) { |
3549 | ret = -EINVAL; | |
3550 | error_propagate(errp, local_err); | |
cb3e7f08 | 3551 | qobject_unref(options); |
9f07429e HR |
3552 | goto free_exit; |
3553 | } | |
9156df12 PB |
3554 | } |
3555 | ||
8ee79e70 KW |
3556 | if (!bs->drv || !bs->drv->supports_backing) { |
3557 | ret = -EINVAL; | |
3558 | error_setg(errp, "Driver doesn't support backing files"); | |
cb3e7f08 | 3559 | qobject_unref(options); |
8ee79e70 KW |
3560 | goto free_exit; |
3561 | } | |
3562 | ||
6bff597b PK |
3563 | if (!reference && |
3564 | bs->backing_format[0] != '\0' && !qdict_haskey(options, "driver")) { | |
46f5ac20 | 3565 | qdict_put_str(options, "driver", bs->backing_format); |
9156df12 PB |
3566 | } |
3567 | ||
6b6833c1 | 3568 | backing_hd = bdrv_open_inherit(backing_filename, reference, options, 0, bs, |
25191e5f | 3569 | &child_of_bds, bdrv_backing_role(bs), errp); |
5b363937 | 3570 | if (!backing_hd) { |
9156df12 | 3571 | bs->open_flags |= BDRV_O_NO_BACKING; |
e43bfd9c | 3572 | error_prepend(errp, "Could not open backing file: "); |
5b363937 | 3573 | ret = -EINVAL; |
1ba4b6a5 | 3574 | goto free_exit; |
9156df12 | 3575 | } |
df581792 | 3576 | |
998c2019 HR |
3577 | if (implicit_backing) { |
3578 | bdrv_refresh_filename(backing_hd); | |
3579 | pstrcpy(bs->auto_backing_file, sizeof(bs->auto_backing_file), | |
3580 | backing_hd->filename); | |
3581 | } | |
3582 | ||
5db15a57 KW |
3583 | /* Hook up the backing file link; drop our reference, bs owns the |
3584 | * backing_hd reference now */ | |
dc9c10a1 | 3585 | ret = bdrv_set_backing_hd(bs, backing_hd, errp); |
5db15a57 | 3586 | bdrv_unref(backing_hd); |
dc9c10a1 | 3587 | if (ret < 0) { |
12fa4af6 KW |
3588 | goto free_exit; |
3589 | } | |
d80ac658 | 3590 | |
d9b7b057 KW |
3591 | qdict_del(parent_options, bdref_key); |
3592 | ||
1ba4b6a5 BC |
3593 | free_exit: |
3594 | g_free(backing_filename); | |
cb3e7f08 | 3595 | qobject_unref(tmp_parent_options); |
1ba4b6a5 | 3596 | return ret; |
9156df12 PB |
3597 | } |
3598 | ||
2d6b86af KW |
3599 | static BlockDriverState * |
3600 | bdrv_open_child_bs(const char *filename, QDict *options, const char *bdref_key, | |
bd86fb99 | 3601 | BlockDriverState *parent, const BdrvChildClass *child_class, |
272c02ea | 3602 | BdrvChildRole child_role, bool allow_none, Error **errp) |
da557aac | 3603 | { |
2d6b86af | 3604 | BlockDriverState *bs = NULL; |
da557aac | 3605 | QDict *image_options; |
da557aac HR |
3606 | char *bdref_key_dot; |
3607 | const char *reference; | |
3608 | ||
bd86fb99 | 3609 | assert(child_class != NULL); |
f67503e5 | 3610 | |
da557aac HR |
3611 | bdref_key_dot = g_strdup_printf("%s.", bdref_key); |
3612 | qdict_extract_subqdict(options, &image_options, bdref_key_dot); | |
3613 | g_free(bdref_key_dot); | |
3614 | ||
129c7d1c MA |
3615 | /* |
3616 | * Caution: while qdict_get_try_str() is fine, getting non-string | |
3617 | * types would require more care. When @options come from | |
3618 | * -blockdev or blockdev_add, its members are typed according to | |
3619 | * the QAPI schema, but when they come from -drive, they're all | |
3620 | * QString. | |
3621 | */ | |
da557aac HR |
3622 | reference = qdict_get_try_str(options, bdref_key); |
3623 | if (!filename && !reference && !qdict_size(image_options)) { | |
b4b059f6 | 3624 | if (!allow_none) { |
da557aac HR |
3625 | error_setg(errp, "A block device must be specified for \"%s\"", |
3626 | bdref_key); | |
da557aac | 3627 | } |
cb3e7f08 | 3628 | qobject_unref(image_options); |
da557aac HR |
3629 | goto done; |
3630 | } | |
3631 | ||
5b363937 | 3632 | bs = bdrv_open_inherit(filename, reference, image_options, 0, |
272c02ea | 3633 | parent, child_class, child_role, errp); |
5b363937 | 3634 | if (!bs) { |
df581792 KW |
3635 | goto done; |
3636 | } | |
3637 | ||
da557aac HR |
3638 | done: |
3639 | qdict_del(options, bdref_key); | |
2d6b86af KW |
3640 | return bs; |
3641 | } | |
3642 | ||
3643 | /* | |
3644 | * Opens a disk image whose options are given as BlockdevRef in another block | |
3645 | * device's options. | |
3646 | * | |
3647 | * If allow_none is true, no image will be opened if filename is false and no | |
3648 | * BlockdevRef is given. NULL will be returned, but errp remains unset. | |
3649 | * | |
3650 | * bdrev_key specifies the key for the image's BlockdevRef in the options QDict. | |
3651 | * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict | |
3652 | * itself, all options starting with "${bdref_key}." are considered part of the | |
3653 | * BlockdevRef. | |
3654 | * | |
3655 | * The BlockdevRef will be removed from the options QDict. | |
aa269ff8 KW |
3656 | * |
3657 | * @parent can move to a different AioContext in this function. Callers must | |
3658 | * make sure that their AioContext locking is still correct after this. | |
2d6b86af KW |
3659 | */ |
3660 | BdrvChild *bdrv_open_child(const char *filename, | |
3661 | QDict *options, const char *bdref_key, | |
3662 | BlockDriverState *parent, | |
bd86fb99 | 3663 | const BdrvChildClass *child_class, |
258b7765 | 3664 | BdrvChildRole child_role, |
2d6b86af KW |
3665 | bool allow_none, Error **errp) |
3666 | { | |
3667 | BlockDriverState *bs; | |
3668 | ||
f791bf7f EGE |
3669 | GLOBAL_STATE_CODE(); |
3670 | ||
bd86fb99 | 3671 | bs = bdrv_open_child_bs(filename, options, bdref_key, parent, child_class, |
272c02ea | 3672 | child_role, allow_none, errp); |
2d6b86af KW |
3673 | if (bs == NULL) { |
3674 | return NULL; | |
3675 | } | |
3676 | ||
258b7765 HR |
3677 | return bdrv_attach_child(parent, bs, bdref_key, child_class, child_role, |
3678 | errp); | |
b4b059f6 KW |
3679 | } |
3680 | ||
83930780 VSO |
3681 | /* |
3682 | * Wrapper on bdrv_open_child() for most popular case: open primary child of bs. | |
aa269ff8 KW |
3683 | * |
3684 | * @parent can move to a different AioContext in this function. Callers must | |
3685 | * make sure that their AioContext locking is still correct after this. | |
83930780 VSO |
3686 | */ |
3687 | int bdrv_open_file_child(const char *filename, | |
3688 | QDict *options, const char *bdref_key, | |
3689 | BlockDriverState *parent, Error **errp) | |
3690 | { | |
3691 | BdrvChildRole role; | |
3692 | ||
3693 | /* commit_top and mirror_top don't use this function */ | |
3694 | assert(!parent->drv->filtered_child_is_backing); | |
83930780 VSO |
3695 | role = parent->drv->is_filter ? |
3696 | (BDRV_CHILD_FILTERED | BDRV_CHILD_PRIMARY) : BDRV_CHILD_IMAGE; | |
3697 | ||
5bb04747 VSO |
3698 | if (!bdrv_open_child(filename, options, bdref_key, parent, |
3699 | &child_of_bds, role, false, errp)) | |
3700 | { | |
3701 | return -EINVAL; | |
3702 | } | |
83930780 | 3703 | |
5bb04747 | 3704 | return 0; |
83930780 VSO |
3705 | } |
3706 | ||
bd86fb99 HR |
3707 | /* |
3708 | * TODO Future callers may need to specify parent/child_class in order for | |
3709 | * option inheritance to work. Existing callers use it for the root node. | |
3710 | */ | |
e1d74bc6 KW |
3711 | BlockDriverState *bdrv_open_blockdev_ref(BlockdevRef *ref, Error **errp) |
3712 | { | |
3713 | BlockDriverState *bs = NULL; | |
e1d74bc6 KW |
3714 | QObject *obj = NULL; |
3715 | QDict *qdict = NULL; | |
3716 | const char *reference = NULL; | |
3717 | Visitor *v = NULL; | |
3718 | ||
f791bf7f EGE |
3719 | GLOBAL_STATE_CODE(); |
3720 | ||
e1d74bc6 KW |
3721 | if (ref->type == QTYPE_QSTRING) { |
3722 | reference = ref->u.reference; | |
3723 | } else { | |
3724 | BlockdevOptions *options = &ref->u.definition; | |
3725 | assert(ref->type == QTYPE_QDICT); | |
3726 | ||
3727 | v = qobject_output_visitor_new(&obj); | |
1f584248 | 3728 | visit_type_BlockdevOptions(v, NULL, &options, &error_abort); |
e1d74bc6 KW |
3729 | visit_complete(v, &obj); |
3730 | ||
7dc847eb | 3731 | qdict = qobject_to(QDict, obj); |
e1d74bc6 KW |
3732 | qdict_flatten(qdict); |
3733 | ||
3734 | /* bdrv_open_inherit() defaults to the values in bdrv_flags (for | |
3735 | * compatibility with other callers) rather than what we want as the | |
3736 | * real defaults. Apply the defaults here instead. */ | |
3737 | qdict_set_default_str(qdict, BDRV_OPT_CACHE_DIRECT, "off"); | |
3738 | qdict_set_default_str(qdict, BDRV_OPT_CACHE_NO_FLUSH, "off"); | |
3739 | qdict_set_default_str(qdict, BDRV_OPT_READ_ONLY, "off"); | |
e35bdc12 KW |
3740 | qdict_set_default_str(qdict, BDRV_OPT_AUTO_READ_ONLY, "off"); |
3741 | ||
e1d74bc6 KW |
3742 | } |
3743 | ||
272c02ea | 3744 | bs = bdrv_open_inherit(NULL, reference, qdict, 0, NULL, NULL, 0, errp); |
e1d74bc6 | 3745 | obj = NULL; |
cb3e7f08 | 3746 | qobject_unref(obj); |
e1d74bc6 KW |
3747 | visit_free(v); |
3748 | return bs; | |
3749 | } | |
3750 | ||
66836189 HR |
3751 | static BlockDriverState *bdrv_append_temp_snapshot(BlockDriverState *bs, |
3752 | int flags, | |
3753 | QDict *snapshot_options, | |
3754 | Error **errp) | |
b998875d | 3755 | { |
69fbfff9 | 3756 | g_autofree char *tmp_filename = NULL; |
b998875d | 3757 | int64_t total_size; |
83d0521a | 3758 | QemuOpts *opts = NULL; |
ff6ed714 | 3759 | BlockDriverState *bs_snapshot = NULL; |
b998875d KW |
3760 | int ret; |
3761 | ||
bdb73476 EGE |
3762 | GLOBAL_STATE_CODE(); |
3763 | ||
b998875d KW |
3764 | /* if snapshot, we create a temporary backing file and open it |
3765 | instead of opening 'filename' directly */ | |
3766 | ||
3767 | /* Get the required size from the image */ | |
f187743a KW |
3768 | total_size = bdrv_getlength(bs); |
3769 | if (total_size < 0) { | |
3770 | error_setg_errno(errp, -total_size, "Could not get image size"); | |
1ba4b6a5 | 3771 | goto out; |
f187743a | 3772 | } |
b998875d KW |
3773 | |
3774 | /* Create the temporary image */ | |
69fbfff9 BM |
3775 | tmp_filename = create_tmp_file(errp); |
3776 | if (!tmp_filename) { | |
1ba4b6a5 | 3777 | goto out; |
b998875d KW |
3778 | } |
3779 | ||
ef810437 | 3780 | opts = qemu_opts_create(bdrv_qcow2.create_opts, NULL, 0, |
c282e1fd | 3781 | &error_abort); |
39101f25 | 3782 | qemu_opt_set_number(opts, BLOCK_OPT_SIZE, total_size, &error_abort); |
e43bfd9c | 3783 | ret = bdrv_create(&bdrv_qcow2, tmp_filename, opts, errp); |
83d0521a | 3784 | qemu_opts_del(opts); |
b998875d | 3785 | if (ret < 0) { |
e43bfd9c MA |
3786 | error_prepend(errp, "Could not create temporary overlay '%s': ", |
3787 | tmp_filename); | |
1ba4b6a5 | 3788 | goto out; |
b998875d KW |
3789 | } |
3790 | ||
73176bee | 3791 | /* Prepare options QDict for the temporary file */ |
46f5ac20 EB |
3792 | qdict_put_str(snapshot_options, "file.driver", "file"); |
3793 | qdict_put_str(snapshot_options, "file.filename", tmp_filename); | |
3794 | qdict_put_str(snapshot_options, "driver", "qcow2"); | |
b998875d | 3795 | |
5b363937 | 3796 | bs_snapshot = bdrv_open(NULL, NULL, snapshot_options, flags, errp); |
73176bee | 3797 | snapshot_options = NULL; |
5b363937 | 3798 | if (!bs_snapshot) { |
1ba4b6a5 | 3799 | goto out; |
b998875d KW |
3800 | } |
3801 | ||
934aee14 VSO |
3802 | ret = bdrv_append(bs_snapshot, bs, errp); |
3803 | if (ret < 0) { | |
ff6ed714 | 3804 | bs_snapshot = NULL; |
b2c2832c KW |
3805 | goto out; |
3806 | } | |
1ba4b6a5 BC |
3807 | |
3808 | out: | |
cb3e7f08 | 3809 | qobject_unref(snapshot_options); |
ff6ed714 | 3810 | return bs_snapshot; |
b998875d KW |
3811 | } |
3812 | ||
b6ce07aa KW |
3813 | /* |
3814 | * Opens a disk image (raw, qcow2, vmdk, ...) | |
de9c0cec KW |
3815 | * |
3816 | * options is a QDict of options to pass to the block drivers, or NULL for an | |
3817 | * empty set of options. The reference to the QDict belongs to the block layer | |
3818 | * after the call (even on failure), so if the caller intends to reuse the | |
cb3e7f08 | 3819 | * dictionary, it needs to use qobject_ref() before calling bdrv_open. |
f67503e5 HR |
3820 | * |
3821 | * If *pbs is NULL, a new BDS will be created with a pointer to it stored there. | |
3822 | * If it is not NULL, the referenced BDS will be reused. | |
ddf5636d HR |
3823 | * |
3824 | * The reference parameter may be used to specify an existing block device which | |
3825 | * should be opened. If specified, neither options nor a filename may be given, | |
3826 | * nor can an existing BDS be reused (that is, *pbs has to be NULL). | |
c86422c5 | 3827 | * |
ae400dbb | 3828 | * The caller must always hold the main AioContext lock. |
b6ce07aa | 3829 | */ |
32192301 KW |
3830 | static BlockDriverState * no_coroutine_fn |
3831 | bdrv_open_inherit(const char *filename, const char *reference, QDict *options, | |
3832 | int flags, BlockDriverState *parent, | |
3833 | const BdrvChildClass *child_class, BdrvChildRole child_role, | |
3834 | Error **errp) | |
ea2384d3 | 3835 | { |
b6ce07aa | 3836 | int ret; |
5696c6e3 | 3837 | BlockBackend *file = NULL; |
9a4f4c31 | 3838 | BlockDriverState *bs; |
ce343771 | 3839 | BlockDriver *drv = NULL; |
2f624b80 | 3840 | BdrvChild *child; |
74fe54f2 | 3841 | const char *drvname; |
3e8c2e57 | 3842 | const char *backing; |
34b5d2c6 | 3843 | Error *local_err = NULL; |
73176bee | 3844 | QDict *snapshot_options = NULL; |
b1e6fc08 | 3845 | int snapshot_flags = 0; |
712e7874 | 3846 | |
bd86fb99 HR |
3847 | assert(!child_class || !flags); |
3848 | assert(!child_class == !parent); | |
f0c28327 | 3849 | GLOBAL_STATE_CODE(); |
32192301 | 3850 | assert(!qemu_in_coroutine()); |
f67503e5 | 3851 | |
ddf5636d HR |
3852 | if (reference) { |
3853 | bool options_non_empty = options ? qdict_size(options) : false; | |
cb3e7f08 | 3854 | qobject_unref(options); |
ddf5636d | 3855 | |
ddf5636d HR |
3856 | if (filename || options_non_empty) { |
3857 | error_setg(errp, "Cannot reference an existing block device with " | |
3858 | "additional options or a new filename"); | |
5b363937 | 3859 | return NULL; |
ddf5636d HR |
3860 | } |
3861 | ||
3862 | bs = bdrv_lookup_bs(reference, reference, errp); | |
3863 | if (!bs) { | |
5b363937 | 3864 | return NULL; |
ddf5636d | 3865 | } |
76b22320 | 3866 | |
ddf5636d | 3867 | bdrv_ref(bs); |
5b363937 | 3868 | return bs; |
ddf5636d HR |
3869 | } |
3870 | ||
5b363937 | 3871 | bs = bdrv_new(); |
f67503e5 | 3872 | |
de9c0cec KW |
3873 | /* NULL means an empty set of options */ |
3874 | if (options == NULL) { | |
3875 | options = qdict_new(); | |
3876 | } | |
3877 | ||
145f598e | 3878 | /* json: syntax counts as explicit options, as if in the QDict */ |
de3b53f0 KW |
3879 | parse_json_protocol(options, &filename, &local_err); |
3880 | if (local_err) { | |
de3b53f0 KW |
3881 | goto fail; |
3882 | } | |
3883 | ||
145f598e KW |
3884 | bs->explicit_options = qdict_clone_shallow(options); |
3885 | ||
bd86fb99 | 3886 | if (child_class) { |
3cdc69d3 HR |
3887 | bool parent_is_format; |
3888 | ||
3889 | if (parent->drv) { | |
3890 | parent_is_format = parent->drv->is_format; | |
3891 | } else { | |
3892 | /* | |
3893 | * parent->drv is not set yet because this node is opened for | |
3894 | * (potential) format probing. That means that @parent is going | |
3895 | * to be a format node. | |
3896 | */ | |
3897 | parent_is_format = true; | |
3898 | } | |
3899 | ||
bddcec37 | 3900 | bs->inherits_from = parent; |
3cdc69d3 HR |
3901 | child_class->inherit_options(child_role, parent_is_format, |
3902 | &flags, options, | |
bd86fb99 | 3903 | parent->open_flags, parent->options); |
f3930ed0 KW |
3904 | } |
3905 | ||
de3b53f0 | 3906 | ret = bdrv_fill_options(&options, filename, &flags, &local_err); |
dfde483e | 3907 | if (ret < 0) { |
462f5bcf KW |
3908 | goto fail; |
3909 | } | |
3910 | ||
129c7d1c MA |
3911 | /* |
3912 | * Set the BDRV_O_RDWR and BDRV_O_ALLOW_RDWR flags. | |
3913 | * Caution: getting a boolean member of @options requires care. | |
3914 | * When @options come from -blockdev or blockdev_add, members are | |
3915 | * typed according to the QAPI schema, but when they come from | |
3916 | * -drive, they're all QString. | |
3917 | */ | |
f87a0e29 AG |
3918 | if (g_strcmp0(qdict_get_try_str(options, BDRV_OPT_READ_ONLY), "on") && |
3919 | !qdict_get_try_bool(options, BDRV_OPT_READ_ONLY, false)) { | |
3920 | flags |= (BDRV_O_RDWR | BDRV_O_ALLOW_RDWR); | |
3921 | } else { | |
3922 | flags &= ~BDRV_O_RDWR; | |
14499ea5 AG |
3923 | } |
3924 | ||
3925 | if (flags & BDRV_O_SNAPSHOT) { | |
3926 | snapshot_options = qdict_new(); | |
3927 | bdrv_temp_snapshot_options(&snapshot_flags, snapshot_options, | |
3928 | flags, options); | |
f87a0e29 AG |
3929 | /* Let bdrv_backing_options() override "read-only" */ |
3930 | qdict_del(options, BDRV_OPT_READ_ONLY); | |
00ff7ffd HR |
3931 | bdrv_inherited_options(BDRV_CHILD_COW, true, |
3932 | &flags, options, flags, options); | |
14499ea5 AG |
3933 | } |
3934 | ||
62392ebb KW |
3935 | bs->open_flags = flags; |
3936 | bs->options = options; | |
3937 | options = qdict_clone_shallow(options); | |
3938 | ||
76c591b0 | 3939 | /* Find the right image format driver */ |
129c7d1c | 3940 | /* See cautionary note on accessing @options above */ |
76c591b0 KW |
3941 | drvname = qdict_get_try_str(options, "driver"); |
3942 | if (drvname) { | |
3943 | drv = bdrv_find_format(drvname); | |
76c591b0 KW |
3944 | if (!drv) { |
3945 | error_setg(errp, "Unknown driver: '%s'", drvname); | |
76c591b0 KW |
3946 | goto fail; |
3947 | } | |
3948 | } | |
3949 | ||
3950 | assert(drvname || !(flags & BDRV_O_PROTOCOL)); | |
76c591b0 | 3951 | |
129c7d1c | 3952 | /* See cautionary note on accessing @options above */ |
3e8c2e57 | 3953 | backing = qdict_get_try_str(options, "backing"); |
e59a0cf1 HR |
3954 | if (qobject_to(QNull, qdict_get(options, "backing")) != NULL || |
3955 | (backing && *backing == '\0')) | |
3956 | { | |
4f7be280 HR |
3957 | if (backing) { |
3958 | warn_report("Use of \"backing\": \"\" is deprecated; " | |
3959 | "use \"backing\": null instead"); | |
3960 | } | |
3e8c2e57 | 3961 | flags |= BDRV_O_NO_BACKING; |
ae0f57f0 KW |
3962 | qdict_del(bs->explicit_options, "backing"); |
3963 | qdict_del(bs->options, "backing"); | |
3e8c2e57 AG |
3964 | qdict_del(options, "backing"); |
3965 | } | |
3966 | ||
5696c6e3 | 3967 | /* Open image file without format layer. This BlockBackend is only used for |
4e4bf5c4 KW |
3968 | * probing, the block drivers will do their own bdrv_open_child() for the |
3969 | * same BDS, which is why we put the node name back into options. */ | |
f4788adc | 3970 | if ((flags & BDRV_O_PROTOCOL) == 0) { |
5696c6e3 KW |
3971 | BlockDriverState *file_bs; |
3972 | ||
3973 | file_bs = bdrv_open_child_bs(filename, options, "file", bs, | |
58944401 HR |
3974 | &child_of_bds, BDRV_CHILD_IMAGE, |
3975 | true, &local_err); | |
1fdd6933 | 3976 | if (local_err) { |
f4788adc KW |
3977 | goto fail; |
3978 | } | |
5696c6e3 | 3979 | if (file_bs != NULL) { |
dacaa162 KW |
3980 | /* Not requesting BLK_PERM_CONSISTENT_READ because we're only |
3981 | * looking at the header to guess the image format. This works even | |
3982 | * in cases where a guest would not see a consistent state. */ | |
d861ab3a | 3983 | file = blk_new(bdrv_get_aio_context(file_bs), 0, BLK_PERM_ALL); |
d7086422 | 3984 | blk_insert_bs(file, file_bs, &local_err); |
5696c6e3 | 3985 | bdrv_unref(file_bs); |
d7086422 KW |
3986 | if (local_err) { |
3987 | goto fail; | |
3988 | } | |
5696c6e3 | 3989 | |
46f5ac20 | 3990 | qdict_put_str(options, "file", bdrv_get_node_name(file_bs)); |
4e4bf5c4 | 3991 | } |
f500a6d3 KW |
3992 | } |
3993 | ||
76c591b0 | 3994 | /* Image format probing */ |
38f3ef57 | 3995 | bs->probed = !drv; |
76c591b0 | 3996 | if (!drv && file) { |
cf2ab8fc | 3997 | ret = find_image_format(file, filename, &drv, &local_err); |
17b005f1 | 3998 | if (ret < 0) { |
8bfea15d | 3999 | goto fail; |
2a05cbe4 | 4000 | } |
62392ebb KW |
4001 | /* |
4002 | * This option update would logically belong in bdrv_fill_options(), | |
4003 | * but we first need to open bs->file for the probing to work, while | |
4004 | * opening bs->file already requires the (mostly) final set of options | |
4005 | * so that cache mode etc. can be inherited. | |
4006 | * | |
4007 | * Adding the driver later is somewhat ugly, but it's not an option | |
4008 | * that would ever be inherited, so it's correct. We just need to make | |
4009 | * sure to update both bs->options (which has the full effective | |
4010 | * options for bs) and options (which has file.* already removed). | |
4011 | */ | |
46f5ac20 EB |
4012 | qdict_put_str(bs->options, "driver", drv->format_name); |
4013 | qdict_put_str(options, "driver", drv->format_name); | |
76c591b0 | 4014 | } else if (!drv) { |
17b005f1 | 4015 | error_setg(errp, "Must specify either driver or file"); |
8bfea15d | 4016 | goto fail; |
ea2384d3 | 4017 | } |
b6ce07aa | 4018 | |
53a29513 HR |
4019 | /* BDRV_O_PROTOCOL must be set iff a protocol BDS is about to be created */ |
4020 | assert(!!(flags & BDRV_O_PROTOCOL) == !!drv->bdrv_file_open); | |
4021 | /* file must be NULL if a protocol BDS is about to be created | |
4022 | * (the inverse results in an error message from bdrv_open_common()) */ | |
4023 | assert(!(flags & BDRV_O_PROTOCOL) || !file); | |
4024 | ||
b6ce07aa | 4025 | /* Open the image */ |
82dc8b41 | 4026 | ret = bdrv_open_common(bs, file, options, &local_err); |
b6ce07aa | 4027 | if (ret < 0) { |
8bfea15d | 4028 | goto fail; |
6987307c CH |
4029 | } |
4030 | ||
4e4bf5c4 | 4031 | if (file) { |
5696c6e3 | 4032 | blk_unref(file); |
f500a6d3 KW |
4033 | file = NULL; |
4034 | } | |
4035 | ||
b6ce07aa | 4036 | /* If there is a backing file, use it */ |
9156df12 | 4037 | if ((flags & BDRV_O_NO_BACKING) == 0) { |
d9b7b057 | 4038 | ret = bdrv_open_backing_file(bs, options, "backing", &local_err); |
b6ce07aa | 4039 | if (ret < 0) { |
b6ad491a | 4040 | goto close_and_fail; |
b6ce07aa | 4041 | } |
b6ce07aa KW |
4042 | } |
4043 | ||
50196d7a AG |
4044 | /* Remove all children options and references |
4045 | * from bs->options and bs->explicit_options */ | |
2f624b80 AG |
4046 | QLIST_FOREACH(child, &bs->children, next) { |
4047 | char *child_key_dot; | |
4048 | child_key_dot = g_strdup_printf("%s.", child->name); | |
4049 | qdict_extract_subqdict(bs->explicit_options, NULL, child_key_dot); | |
4050 | qdict_extract_subqdict(bs->options, NULL, child_key_dot); | |
50196d7a AG |
4051 | qdict_del(bs->explicit_options, child->name); |
4052 | qdict_del(bs->options, child->name); | |
2f624b80 AG |
4053 | g_free(child_key_dot); |
4054 | } | |
4055 | ||
b6ad491a | 4056 | /* Check if any unknown options were used */ |
7ad2757f | 4057 | if (qdict_size(options) != 0) { |
b6ad491a | 4058 | const QDictEntry *entry = qdict_first(options); |
5acd9d81 HR |
4059 | if (flags & BDRV_O_PROTOCOL) { |
4060 | error_setg(errp, "Block protocol '%s' doesn't support the option " | |
4061 | "'%s'", drv->format_name, entry->key); | |
4062 | } else { | |
d0e46a55 HR |
4063 | error_setg(errp, |
4064 | "Block format '%s' does not support the option '%s'", | |
4065 | drv->format_name, entry->key); | |
5acd9d81 | 4066 | } |
b6ad491a | 4067 | |
b6ad491a KW |
4068 | goto close_and_fail; |
4069 | } | |
b6ad491a | 4070 | |
c01c214b | 4071 | bdrv_parent_cb_change_media(bs, true); |
b6ce07aa | 4072 | |
cb3e7f08 | 4073 | qobject_unref(options); |
8961be33 | 4074 | options = NULL; |
dd62f1ca KW |
4075 | |
4076 | /* For snapshot=on, create a temporary qcow2 overlay. bs points to the | |
4077 | * temporary snapshot afterwards. */ | |
4078 | if (snapshot_flags) { | |
66836189 HR |
4079 | BlockDriverState *snapshot_bs; |
4080 | snapshot_bs = bdrv_append_temp_snapshot(bs, snapshot_flags, | |
4081 | snapshot_options, &local_err); | |
73176bee | 4082 | snapshot_options = NULL; |
dd62f1ca KW |
4083 | if (local_err) { |
4084 | goto close_and_fail; | |
4085 | } | |
5b363937 HR |
4086 | /* We are not going to return bs but the overlay on top of it |
4087 | * (snapshot_bs); thus, we have to drop the strong reference to bs | |
4088 | * (which we obtained by calling bdrv_new()). bs will not be deleted, | |
4089 | * though, because the overlay still has a reference to it. */ | |
4090 | bdrv_unref(bs); | |
4091 | bs = snapshot_bs; | |
dd62f1ca KW |
4092 | } |
4093 | ||
5b363937 | 4094 | return bs; |
b6ce07aa | 4095 | |
8bfea15d | 4096 | fail: |
5696c6e3 | 4097 | blk_unref(file); |
cb3e7f08 MAL |
4098 | qobject_unref(snapshot_options); |
4099 | qobject_unref(bs->explicit_options); | |
4100 | qobject_unref(bs->options); | |
4101 | qobject_unref(options); | |
de9c0cec | 4102 | bs->options = NULL; |
998cbd6a | 4103 | bs->explicit_options = NULL; |
5b363937 | 4104 | bdrv_unref(bs); |
621ff94d | 4105 | error_propagate(errp, local_err); |
5b363937 | 4106 | return NULL; |
de9c0cec | 4107 | |
b6ad491a | 4108 | close_and_fail: |
5b363937 | 4109 | bdrv_unref(bs); |
cb3e7f08 MAL |
4110 | qobject_unref(snapshot_options); |
4111 | qobject_unref(options); | |
621ff94d | 4112 | error_propagate(errp, local_err); |
5b363937 | 4113 | return NULL; |
b6ce07aa KW |
4114 | } |
4115 | ||
ae400dbb | 4116 | /* The caller must always hold the main AioContext lock. */ |
5b363937 HR |
4117 | BlockDriverState *bdrv_open(const char *filename, const char *reference, |
4118 | QDict *options, int flags, Error **errp) | |
f3930ed0 | 4119 | { |
f791bf7f EGE |
4120 | GLOBAL_STATE_CODE(); |
4121 | ||
5b363937 | 4122 | return bdrv_open_inherit(filename, reference, options, flags, NULL, |
272c02ea | 4123 | NULL, 0, errp); |
f3930ed0 KW |
4124 | } |
4125 | ||
faf116b4 AG |
4126 | /* Return true if the NULL-terminated @list contains @str */ |
4127 | static bool is_str_in_list(const char *str, const char *const *list) | |
4128 | { | |
4129 | if (str && list) { | |
4130 | int i; | |
4131 | for (i = 0; list[i] != NULL; i++) { | |
4132 | if (!strcmp(str, list[i])) { | |
4133 | return true; | |
4134 | } | |
4135 | } | |
4136 | } | |
4137 | return false; | |
4138 | } | |
4139 | ||
4140 | /* | |
4141 | * Check that every option set in @bs->options is also set in | |
4142 | * @new_opts. | |
4143 | * | |
4144 | * Options listed in the common_options list and in | |
4145 | * @bs->drv->mutable_opts are skipped. | |
4146 | * | |
4147 | * Return 0 on success, otherwise return -EINVAL and set @errp. | |
4148 | */ | |
4149 | static int bdrv_reset_options_allowed(BlockDriverState *bs, | |
4150 | const QDict *new_opts, Error **errp) | |
4151 | { | |
4152 | const QDictEntry *e; | |
4153 | /* These options are common to all block drivers and are handled | |
4154 | * in bdrv_reopen_prepare() so they can be left out of @new_opts */ | |
4155 | const char *const common_options[] = { | |
4156 | "node-name", "discard", "cache.direct", "cache.no-flush", | |
4157 | "read-only", "auto-read-only", "detect-zeroes", NULL | |
4158 | }; | |
4159 | ||
4160 | for (e = qdict_first(bs->options); e; e = qdict_next(bs->options, e)) { | |
4161 | if (!qdict_haskey(new_opts, e->key) && | |
4162 | !is_str_in_list(e->key, common_options) && | |
4163 | !is_str_in_list(e->key, bs->drv->mutable_opts)) { | |
4164 | error_setg(errp, "Option '%s' cannot be reset " | |
4165 | "to its default value", e->key); | |
4166 | return -EINVAL; | |
4167 | } | |
4168 | } | |
4169 | ||
4170 | return 0; | |
4171 | } | |
4172 | ||
cb828c31 AG |
4173 | /* |
4174 | * Returns true if @child can be reached recursively from @bs | |
4175 | */ | |
4176 | static bool bdrv_recurse_has_child(BlockDriverState *bs, | |
4177 | BlockDriverState *child) | |
4178 | { | |
4179 | BdrvChild *c; | |
4180 | ||
4181 | if (bs == child) { | |
4182 | return true; | |
4183 | } | |
4184 | ||
4185 | QLIST_FOREACH(c, &bs->children, next) { | |
4186 | if (bdrv_recurse_has_child(c->bs, child)) { | |
4187 | return true; | |
4188 | } | |
4189 | } | |
4190 | ||
4191 | return false; | |
4192 | } | |
4193 | ||
e971aa12 JC |
4194 | /* |
4195 | * Adds a BlockDriverState to a simple queue for an atomic, transactional | |
4196 | * reopen of multiple devices. | |
4197 | * | |
859443b0 | 4198 | * bs_queue can either be an existing BlockReopenQueue that has had QTAILQ_INIT |
e971aa12 JC |
4199 | * already performed, or alternatively may be NULL a new BlockReopenQueue will |
4200 | * be created and initialized. This newly created BlockReopenQueue should be | |
4201 | * passed back in for subsequent calls that are intended to be of the same | |
4202 | * atomic 'set'. | |
4203 | * | |
4204 | * bs is the BlockDriverState to add to the reopen queue. | |
4205 | * | |
4d2cb092 KW |
4206 | * options contains the changed options for the associated bs |
4207 | * (the BlockReopenQueue takes ownership) | |
4208 | * | |
e971aa12 JC |
4209 | * flags contains the open flags for the associated bs |
4210 | * | |
4211 | * returns a pointer to bs_queue, which is either the newly allocated | |
4212 | * bs_queue, or the existing bs_queue being used. | |
4213 | * | |
d22933ac | 4214 | * bs is drained here and undrained by bdrv_reopen_queue_free(). |
2e117866 KW |
4215 | * |
4216 | * To be called with bs->aio_context locked. | |
e971aa12 | 4217 | */ |
28518102 KW |
4218 | static BlockReopenQueue *bdrv_reopen_queue_child(BlockReopenQueue *bs_queue, |
4219 | BlockDriverState *bs, | |
4220 | QDict *options, | |
bd86fb99 | 4221 | const BdrvChildClass *klass, |
272c02ea | 4222 | BdrvChildRole role, |
3cdc69d3 | 4223 | bool parent_is_format, |
28518102 | 4224 | QDict *parent_options, |
077e8e20 AG |
4225 | int parent_flags, |
4226 | bool keep_old_opts) | |
e971aa12 JC |
4227 | { |
4228 | assert(bs != NULL); | |
4229 | ||
4230 | BlockReopenQueueEntry *bs_entry; | |
67251a31 | 4231 | BdrvChild *child; |
9aa09ddd AG |
4232 | QDict *old_options, *explicit_options, *options_copy; |
4233 | int flags; | |
4234 | QemuOpts *opts; | |
67251a31 | 4235 | |
f0c28327 | 4236 | GLOBAL_STATE_CODE(); |
1a63a907 | 4237 | |
d22933ac KW |
4238 | bdrv_drained_begin(bs); |
4239 | ||
e971aa12 JC |
4240 | if (bs_queue == NULL) { |
4241 | bs_queue = g_new0(BlockReopenQueue, 1); | |
859443b0 | 4242 | QTAILQ_INIT(bs_queue); |
e971aa12 JC |
4243 | } |
4244 | ||
4d2cb092 KW |
4245 | if (!options) { |
4246 | options = qdict_new(); | |
4247 | } | |
4248 | ||
5b7ba05f | 4249 | /* Check if this BlockDriverState is already in the queue */ |
859443b0 | 4250 | QTAILQ_FOREACH(bs_entry, bs_queue, entry) { |
5b7ba05f AG |
4251 | if (bs == bs_entry->state.bs) { |
4252 | break; | |
4253 | } | |
4254 | } | |
4255 | ||
28518102 KW |
4256 | /* |
4257 | * Precedence of options: | |
4258 | * 1. Explicitly passed in options (highest) | |
9aa09ddd AG |
4259 | * 2. Retained from explicitly set options of bs |
4260 | * 3. Inherited from parent node | |
4261 | * 4. Retained from effective options of bs | |
28518102 KW |
4262 | */ |
4263 | ||
145f598e | 4264 | /* Old explicitly set values (don't overwrite by inherited value) */ |
077e8e20 AG |
4265 | if (bs_entry || keep_old_opts) { |
4266 | old_options = qdict_clone_shallow(bs_entry ? | |
4267 | bs_entry->state.explicit_options : | |
4268 | bs->explicit_options); | |
4269 | bdrv_join_options(bs, options, old_options); | |
4270 | qobject_unref(old_options); | |
5b7ba05f | 4271 | } |
145f598e KW |
4272 | |
4273 | explicit_options = qdict_clone_shallow(options); | |
4274 | ||
28518102 KW |
4275 | /* Inherit from parent node */ |
4276 | if (parent_options) { | |
9aa09ddd | 4277 | flags = 0; |
3cdc69d3 | 4278 | klass->inherit_options(role, parent_is_format, &flags, options, |
272c02ea | 4279 | parent_flags, parent_options); |
9aa09ddd AG |
4280 | } else { |
4281 | flags = bdrv_get_flags(bs); | |
28518102 KW |
4282 | } |
4283 | ||
077e8e20 AG |
4284 | if (keep_old_opts) { |
4285 | /* Old values are used for options that aren't set yet */ | |
4286 | old_options = qdict_clone_shallow(bs->options); | |
4287 | bdrv_join_options(bs, options, old_options); | |
4288 | qobject_unref(old_options); | |
4289 | } | |
4d2cb092 | 4290 | |
9aa09ddd AG |
4291 | /* We have the final set of options so let's update the flags */ |
4292 | options_copy = qdict_clone_shallow(options); | |
4293 | opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort); | |
4294 | qemu_opts_absorb_qdict(opts, options_copy, NULL); | |
4295 | update_flags_from_options(&flags, opts); | |
4296 | qemu_opts_del(opts); | |
4297 | qobject_unref(options_copy); | |
4298 | ||
fd452021 | 4299 | /* bdrv_open_inherit() sets and clears some additional flags internally */ |
f1f25a2e | 4300 | flags &= ~BDRV_O_PROTOCOL; |
fd452021 KW |
4301 | if (flags & BDRV_O_RDWR) { |
4302 | flags |= BDRV_O_ALLOW_RDWR; | |
4303 | } | |
f1f25a2e | 4304 | |
1857c97b KW |
4305 | if (!bs_entry) { |
4306 | bs_entry = g_new0(BlockReopenQueueEntry, 1); | |
859443b0 | 4307 | QTAILQ_INSERT_TAIL(bs_queue, bs_entry, entry); |
1857c97b | 4308 | } else { |
cb3e7f08 MAL |
4309 | qobject_unref(bs_entry->state.options); |
4310 | qobject_unref(bs_entry->state.explicit_options); | |
1857c97b KW |
4311 | } |
4312 | ||
4313 | bs_entry->state.bs = bs; | |
4314 | bs_entry->state.options = options; | |
4315 | bs_entry->state.explicit_options = explicit_options; | |
4316 | bs_entry->state.flags = flags; | |
4317 | ||
8546632e AG |
4318 | /* |
4319 | * If keep_old_opts is false then it means that unspecified | |
4320 | * options must be reset to their original value. We don't allow | |
4321 | * resetting 'backing' but we need to know if the option is | |
4322 | * missing in order to decide if we have to return an error. | |
4323 | */ | |
4324 | if (!keep_old_opts) { | |
4325 | bs_entry->state.backing_missing = | |
4326 | !qdict_haskey(options, "backing") && | |
4327 | !qdict_haskey(options, "backing.driver"); | |
4328 | } | |
4329 | ||
67251a31 | 4330 | QLIST_FOREACH(child, &bs->children, next) { |
8546632e AG |
4331 | QDict *new_child_options = NULL; |
4332 | bool child_keep_old = keep_old_opts; | |
67251a31 | 4333 | |
4c9dfe5d KW |
4334 | /* reopen can only change the options of block devices that were |
4335 | * implicitly created and inherited options. For other (referenced) | |
4336 | * block devices, a syntax like "backing.foo" results in an error. */ | |
67251a31 KW |
4337 | if (child->bs->inherits_from != bs) { |
4338 | continue; | |
4339 | } | |
4340 | ||
8546632e AG |
4341 | /* Check if the options contain a child reference */ |
4342 | if (qdict_haskey(options, child->name)) { | |
4343 | const char *childref = qdict_get_try_str(options, child->name); | |
4344 | /* | |
4345 | * The current child must not be reopened if the child | |
4346 | * reference is null or points to a different node. | |
4347 | */ | |
4348 | if (g_strcmp0(childref, child->bs->node_name)) { | |
4349 | continue; | |
4350 | } | |
4351 | /* | |
4352 | * If the child reference points to the current child then | |
4353 | * reopen it with its existing set of options (note that | |
4354 | * it can still inherit new options from the parent). | |
4355 | */ | |
4356 | child_keep_old = true; | |
4357 | } else { | |
4358 | /* Extract child options ("child-name.*") */ | |
4359 | char *child_key_dot = g_strdup_printf("%s.", child->name); | |
4360 | qdict_extract_subqdict(explicit_options, NULL, child_key_dot); | |
4361 | qdict_extract_subqdict(options, &new_child_options, child_key_dot); | |
4362 | g_free(child_key_dot); | |
4363 | } | |
4c9dfe5d | 4364 | |
9aa09ddd | 4365 | bdrv_reopen_queue_child(bs_queue, child->bs, new_child_options, |
3cdc69d3 HR |
4366 | child->klass, child->role, bs->drv->is_format, |
4367 | options, flags, child_keep_old); | |
e971aa12 JC |
4368 | } |
4369 | ||
e971aa12 JC |
4370 | return bs_queue; |
4371 | } | |
4372 | ||
2e117866 | 4373 | /* To be called with bs->aio_context locked */ |
28518102 KW |
4374 | BlockReopenQueue *bdrv_reopen_queue(BlockReopenQueue *bs_queue, |
4375 | BlockDriverState *bs, | |
077e8e20 | 4376 | QDict *options, bool keep_old_opts) |
28518102 | 4377 | { |
f791bf7f EGE |
4378 | GLOBAL_STATE_CODE(); |
4379 | ||
3cdc69d3 HR |
4380 | return bdrv_reopen_queue_child(bs_queue, bs, options, NULL, 0, false, |
4381 | NULL, 0, keep_old_opts); | |
28518102 KW |
4382 | } |
4383 | ||
ab5b5228 AG |
4384 | void bdrv_reopen_queue_free(BlockReopenQueue *bs_queue) |
4385 | { | |
f791bf7f | 4386 | GLOBAL_STATE_CODE(); |
ab5b5228 AG |
4387 | if (bs_queue) { |
4388 | BlockReopenQueueEntry *bs_entry, *next; | |
4389 | QTAILQ_FOREACH_SAFE(bs_entry, bs_queue, entry, next) { | |
d22933ac KW |
4390 | AioContext *ctx = bdrv_get_aio_context(bs_entry->state.bs); |
4391 | ||
4392 | aio_context_acquire(ctx); | |
4393 | bdrv_drained_end(bs_entry->state.bs); | |
4394 | aio_context_release(ctx); | |
4395 | ||
ab5b5228 AG |
4396 | qobject_unref(bs_entry->state.explicit_options); |
4397 | qobject_unref(bs_entry->state.options); | |
4398 | g_free(bs_entry); | |
4399 | } | |
4400 | g_free(bs_queue); | |
4401 | } | |
4402 | } | |
4403 | ||
e971aa12 JC |
4404 | /* |
4405 | * Reopen multiple BlockDriverStates atomically & transactionally. | |
4406 | * | |
4407 | * The queue passed in (bs_queue) must have been built up previous | |
4408 | * via bdrv_reopen_queue(). | |
4409 | * | |
4410 | * Reopens all BDS specified in the queue, with the appropriate | |
4411 | * flags. All devices are prepared for reopen, and failure of any | |
50d6a8a3 | 4412 | * device will cause all device changes to be abandoned, and intermediate |
e971aa12 JC |
4413 | * data cleaned up. |
4414 | * | |
4415 | * If all devices prepare successfully, then the changes are committed | |
4416 | * to all devices. | |
4417 | * | |
1a63a907 KW |
4418 | * All affected nodes must be drained between bdrv_reopen_queue() and |
4419 | * bdrv_reopen_multiple(). | |
6cf42ca2 KW |
4420 | * |
4421 | * To be called from the main thread, with all other AioContexts unlocked. | |
e971aa12 | 4422 | */ |
5019aece | 4423 | int bdrv_reopen_multiple(BlockReopenQueue *bs_queue, Error **errp) |
e971aa12 JC |
4424 | { |
4425 | int ret = -1; | |
4426 | BlockReopenQueueEntry *bs_entry, *next; | |
6cf42ca2 | 4427 | AioContext *ctx; |
72373e40 | 4428 | Transaction *tran = tran_new(); |
72373e40 | 4429 | g_autoptr(GSList) refresh_list = NULL; |
e971aa12 | 4430 | |
6cf42ca2 | 4431 | assert(qemu_get_current_aio_context() == qemu_get_aio_context()); |
e971aa12 | 4432 | assert(bs_queue != NULL); |
da359909 | 4433 | GLOBAL_STATE_CODE(); |
e971aa12 | 4434 | |
a2aabf88 | 4435 | QTAILQ_FOREACH(bs_entry, bs_queue, entry) { |
6cf42ca2 KW |
4436 | ctx = bdrv_get_aio_context(bs_entry->state.bs); |
4437 | aio_context_acquire(ctx); | |
a2aabf88 | 4438 | ret = bdrv_flush(bs_entry->state.bs); |
6cf42ca2 | 4439 | aio_context_release(ctx); |
a2aabf88 VSO |
4440 | if (ret < 0) { |
4441 | error_setg_errno(errp, -ret, "Error flushing drive"); | |
e3fc91aa | 4442 | goto abort; |
a2aabf88 VSO |
4443 | } |
4444 | } | |
4445 | ||
859443b0 | 4446 | QTAILQ_FOREACH(bs_entry, bs_queue, entry) { |
1a63a907 | 4447 | assert(bs_entry->state.bs->quiesce_counter > 0); |
6cf42ca2 KW |
4448 | ctx = bdrv_get_aio_context(bs_entry->state.bs); |
4449 | aio_context_acquire(ctx); | |
72373e40 | 4450 | ret = bdrv_reopen_prepare(&bs_entry->state, bs_queue, tran, errp); |
6cf42ca2 | 4451 | aio_context_release(ctx); |
72373e40 VSO |
4452 | if (ret < 0) { |
4453 | goto abort; | |
e971aa12 JC |
4454 | } |
4455 | bs_entry->prepared = true; | |
4456 | } | |
4457 | ||
859443b0 | 4458 | QTAILQ_FOREACH(bs_entry, bs_queue, entry) { |
69b736e7 | 4459 | BDRVReopenState *state = &bs_entry->state; |
72373e40 | 4460 | |
fb0ff4d1 | 4461 | refresh_list = g_slist_prepend(refresh_list, state->bs); |
72373e40 | 4462 | if (state->old_backing_bs) { |
fb0ff4d1 | 4463 | refresh_list = g_slist_prepend(refresh_list, state->old_backing_bs); |
cb828c31 | 4464 | } |
ecd30d2d | 4465 | if (state->old_file_bs) { |
fb0ff4d1 | 4466 | refresh_list = g_slist_prepend(refresh_list, state->old_file_bs); |
ecd30d2d | 4467 | } |
72373e40 VSO |
4468 | } |
4469 | ||
4470 | /* | |
4471 | * Note that file-posix driver rely on permission update done during reopen | |
4472 | * (even if no permission changed), because it wants "new" permissions for | |
4473 | * reconfiguring the fd and that's why it does it in raw_check_perm(), not | |
4474 | * in raw_reopen_prepare() which is called with "old" permissions. | |
4475 | */ | |
4476 | ret = bdrv_list_refresh_perms(refresh_list, bs_queue, tran, errp); | |
4477 | if (ret < 0) { | |
4478 | goto abort; | |
69b736e7 KW |
4479 | } |
4480 | ||
fcd6a4f4 VSO |
4481 | /* |
4482 | * If we reach this point, we have success and just need to apply the | |
4483 | * changes. | |
4484 | * | |
4485 | * Reverse order is used to comfort qcow2 driver: on commit it need to write | |
4486 | * IN_USE flag to the image, to mark bitmaps in the image as invalid. But | |
4487 | * children are usually goes after parents in reopen-queue, so go from last | |
4488 | * to first element. | |
e971aa12 | 4489 | */ |
fcd6a4f4 | 4490 | QTAILQ_FOREACH_REVERSE(bs_entry, bs_queue, entry) { |
6cf42ca2 KW |
4491 | ctx = bdrv_get_aio_context(bs_entry->state.bs); |
4492 | aio_context_acquire(ctx); | |
e971aa12 | 4493 | bdrv_reopen_commit(&bs_entry->state); |
6cf42ca2 | 4494 | aio_context_release(ctx); |
e971aa12 JC |
4495 | } |
4496 | ||
72373e40 | 4497 | tran_commit(tran); |
69b736e7 | 4498 | |
72373e40 VSO |
4499 | QTAILQ_FOREACH_REVERSE(bs_entry, bs_queue, entry) { |
4500 | BlockDriverState *bs = bs_entry->state.bs; | |
74ad9a3b | 4501 | |
72373e40 | 4502 | if (bs->drv->bdrv_reopen_commit_post) { |
6cf42ca2 KW |
4503 | ctx = bdrv_get_aio_context(bs); |
4504 | aio_context_acquire(ctx); | |
72373e40 | 4505 | bs->drv->bdrv_reopen_commit_post(&bs_entry->state); |
6cf42ca2 | 4506 | aio_context_release(ctx); |
69b736e7 KW |
4507 | } |
4508 | } | |
17e1e2be | 4509 | |
72373e40 VSO |
4510 | ret = 0; |
4511 | goto cleanup; | |
17e1e2be | 4512 | |
72373e40 VSO |
4513 | abort: |
4514 | tran_abort(tran); | |
4515 | QTAILQ_FOREACH_SAFE(bs_entry, bs_queue, entry, next) { | |
4516 | if (bs_entry->prepared) { | |
6cf42ca2 KW |
4517 | ctx = bdrv_get_aio_context(bs_entry->state.bs); |
4518 | aio_context_acquire(ctx); | |
72373e40 | 4519 | bdrv_reopen_abort(&bs_entry->state); |
6cf42ca2 | 4520 | aio_context_release(ctx); |
17e1e2be PK |
4521 | } |
4522 | } | |
72373e40 | 4523 | |
e971aa12 | 4524 | cleanup: |
ab5b5228 | 4525 | bdrv_reopen_queue_free(bs_queue); |
40840e41 | 4526 | |
e971aa12 JC |
4527 | return ret; |
4528 | } | |
4529 | ||
6cf42ca2 KW |
4530 | int bdrv_reopen(BlockDriverState *bs, QDict *opts, bool keep_old_opts, |
4531 | Error **errp) | |
6e1000a8 | 4532 | { |
6cf42ca2 | 4533 | AioContext *ctx = bdrv_get_aio_context(bs); |
6e1000a8 | 4534 | BlockReopenQueue *queue; |
6cf42ca2 | 4535 | int ret; |
6e1000a8 | 4536 | |
f791bf7f EGE |
4537 | GLOBAL_STATE_CODE(); |
4538 | ||
2e117866 KW |
4539 | queue = bdrv_reopen_queue(NULL, bs, opts, keep_old_opts); |
4540 | ||
6cf42ca2 KW |
4541 | if (ctx != qemu_get_aio_context()) { |
4542 | aio_context_release(ctx); | |
4543 | } | |
5019aece | 4544 | ret = bdrv_reopen_multiple(queue, errp); |
6cf42ca2 KW |
4545 | |
4546 | if (ctx != qemu_get_aio_context()) { | |
4547 | aio_context_acquire(ctx); | |
4548 | } | |
6e1000a8 AG |
4549 | |
4550 | return ret; | |
4551 | } | |
4552 | ||
6cf42ca2 KW |
4553 | int bdrv_reopen_set_read_only(BlockDriverState *bs, bool read_only, |
4554 | Error **errp) | |
4555 | { | |
4556 | QDict *opts = qdict_new(); | |
4557 | ||
f791bf7f EGE |
4558 | GLOBAL_STATE_CODE(); |
4559 | ||
6cf42ca2 KW |
4560 | qdict_put_bool(opts, BDRV_OPT_READ_ONLY, read_only); |
4561 | ||
4562 | return bdrv_reopen(bs, opts, true, errp); | |
4563 | } | |
4564 | ||
cb828c31 AG |
4565 | /* |
4566 | * Take a BDRVReopenState and check if the value of 'backing' in the | |
4567 | * reopen_state->options QDict is valid or not. | |
4568 | * | |
4569 | * If 'backing' is missing from the QDict then return 0. | |
4570 | * | |
4571 | * If 'backing' contains the node name of the backing file of | |
4572 | * reopen_state->bs then return 0. | |
4573 | * | |
4574 | * If 'backing' contains a different node name (or is null) then check | |
4575 | * whether the current backing file can be replaced with the new one. | |
4576 | * If that's the case then reopen_state->replace_backing_bs is set to | |
4577 | * true and reopen_state->new_backing_bs contains a pointer to the new | |
4578 | * backing BlockDriverState (or NULL). | |
4579 | * | |
4580 | * Return 0 on success, otherwise return < 0 and set @errp. | |
4581 | */ | |
ecd30d2d AG |
4582 | static int bdrv_reopen_parse_file_or_backing(BDRVReopenState *reopen_state, |
4583 | bool is_backing, Transaction *tran, | |
4584 | Error **errp) | |
cb828c31 AG |
4585 | { |
4586 | BlockDriverState *bs = reopen_state->bs; | |
ecd30d2d AG |
4587 | BlockDriverState *new_child_bs; |
4588 | BlockDriverState *old_child_bs = is_backing ? child_bs(bs->backing) : | |
4589 | child_bs(bs->file); | |
4590 | const char *child_name = is_backing ? "backing" : "file"; | |
cb828c31 AG |
4591 | QObject *value; |
4592 | const char *str; | |
4593 | ||
bdb73476 EGE |
4594 | GLOBAL_STATE_CODE(); |
4595 | ||
ecd30d2d | 4596 | value = qdict_get(reopen_state->options, child_name); |
cb828c31 AG |
4597 | if (value == NULL) { |
4598 | return 0; | |
4599 | } | |
4600 | ||
4601 | switch (qobject_type(value)) { | |
4602 | case QTYPE_QNULL: | |
ecd30d2d AG |
4603 | assert(is_backing); /* The 'file' option does not allow a null value */ |
4604 | new_child_bs = NULL; | |
cb828c31 AG |
4605 | break; |
4606 | case QTYPE_QSTRING: | |
410f44f5 | 4607 | str = qstring_get_str(qobject_to(QString, value)); |
ecd30d2d AG |
4608 | new_child_bs = bdrv_lookup_bs(NULL, str, errp); |
4609 | if (new_child_bs == NULL) { | |
cb828c31 | 4610 | return -EINVAL; |
ecd30d2d AG |
4611 | } else if (bdrv_recurse_has_child(new_child_bs, bs)) { |
4612 | error_setg(errp, "Making '%s' a %s child of '%s' would create a " | |
4613 | "cycle", str, child_name, bs->node_name); | |
cb828c31 AG |
4614 | return -EINVAL; |
4615 | } | |
4616 | break; | |
4617 | default: | |
ecd30d2d AG |
4618 | /* |
4619 | * The options QDict has been flattened, so 'backing' and 'file' | |
4620 | * do not allow any other data type here. | |
4621 | */ | |
cb828c31 AG |
4622 | g_assert_not_reached(); |
4623 | } | |
4624 | ||
ecd30d2d AG |
4625 | if (old_child_bs == new_child_bs) { |
4626 | return 0; | |
4627 | } | |
4628 | ||
4629 | if (old_child_bs) { | |
4630 | if (bdrv_skip_implicit_filters(old_child_bs) == new_child_bs) { | |
cbfdb98c VSO |
4631 | return 0; |
4632 | } | |
4633 | ||
ecd30d2d AG |
4634 | if (old_child_bs->implicit) { |
4635 | error_setg(errp, "Cannot replace implicit %s child of %s", | |
4636 | child_name, bs->node_name); | |
cbfdb98c VSO |
4637 | return -EPERM; |
4638 | } | |
4639 | } | |
4640 | ||
ecd30d2d | 4641 | if (bs->drv->is_filter && !old_child_bs) { |
25f78d9e VSO |
4642 | /* |
4643 | * Filters always have a file or a backing child, so we are trying to | |
4644 | * change wrong child | |
4645 | */ | |
4646 | error_setg(errp, "'%s' is a %s filter node that does not support a " | |
ecd30d2d | 4647 | "%s child", bs->node_name, bs->drv->format_name, child_name); |
1d42f48c HR |
4648 | return -EINVAL; |
4649 | } | |
4650 | ||
ecd30d2d AG |
4651 | if (is_backing) { |
4652 | reopen_state->old_backing_bs = old_child_bs; | |
4653 | } else { | |
4654 | reopen_state->old_file_bs = old_child_bs; | |
4655 | } | |
4656 | ||
4657 | return bdrv_set_file_or_backing_noperm(bs, new_child_bs, is_backing, | |
4658 | tran, errp); | |
cb828c31 AG |
4659 | } |
4660 | ||
e971aa12 JC |
4661 | /* |
4662 | * Prepares a BlockDriverState for reopen. All changes are staged in the | |
4663 | * 'opaque' field of the BDRVReopenState, which is used and allocated by | |
4664 | * the block driver layer .bdrv_reopen_prepare() | |
4665 | * | |
4666 | * bs is the BlockDriverState to reopen | |
4667 | * flags are the new open flags | |
4668 | * queue is the reopen queue | |
4669 | * | |
4670 | * Returns 0 on success, non-zero on error. On error errp will be set | |
4671 | * as well. | |
4672 | * | |
4673 | * On failure, bdrv_reopen_abort() will be called to clean up any data. | |
4674 | * It is the responsibility of the caller to then call the abort() or | |
4675 | * commit() for any other BDS that have been left in a prepare() state | |
4676 | * | |
4677 | */ | |
53e96d1e | 4678 | static int bdrv_reopen_prepare(BDRVReopenState *reopen_state, |
72373e40 | 4679 | BlockReopenQueue *queue, |
ecd30d2d | 4680 | Transaction *change_child_tran, Error **errp) |
e971aa12 JC |
4681 | { |
4682 | int ret = -1; | |
e6d79c41 | 4683 | int old_flags; |
e971aa12 JC |
4684 | Error *local_err = NULL; |
4685 | BlockDriver *drv; | |
ccf9dc07 | 4686 | QemuOpts *opts; |
4c8350fe | 4687 | QDict *orig_reopen_opts; |
593b3071 | 4688 | char *discard = NULL; |
3d8ce171 | 4689 | bool read_only; |
9ad08c44 | 4690 | bool drv_prepared = false; |
e971aa12 JC |
4691 | |
4692 | assert(reopen_state != NULL); | |
4693 | assert(reopen_state->bs->drv != NULL); | |
da359909 | 4694 | GLOBAL_STATE_CODE(); |
e971aa12 JC |
4695 | drv = reopen_state->bs->drv; |
4696 | ||
4c8350fe AG |
4697 | /* This function and each driver's bdrv_reopen_prepare() remove |
4698 | * entries from reopen_state->options as they are processed, so | |
4699 | * we need to make a copy of the original QDict. */ | |
4700 | orig_reopen_opts = qdict_clone_shallow(reopen_state->options); | |
4701 | ||
ccf9dc07 KW |
4702 | /* Process generic block layer options */ |
4703 | opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort); | |
af175e85 | 4704 | if (!qemu_opts_absorb_qdict(opts, reopen_state->options, errp)) { |
ccf9dc07 KW |
4705 | ret = -EINVAL; |
4706 | goto error; | |
4707 | } | |
4708 | ||
e6d79c41 AG |
4709 | /* This was already called in bdrv_reopen_queue_child() so the flags |
4710 | * are up-to-date. This time we simply want to remove the options from | |
4711 | * QemuOpts in order to indicate that they have been processed. */ | |
4712 | old_flags = reopen_state->flags; | |
91a097e7 | 4713 | update_flags_from_options(&reopen_state->flags, opts); |
e6d79c41 | 4714 | assert(old_flags == reopen_state->flags); |
91a097e7 | 4715 | |
415bbca8 | 4716 | discard = qemu_opt_get_del(opts, BDRV_OPT_DISCARD); |
593b3071 AG |
4717 | if (discard != NULL) { |
4718 | if (bdrv_parse_discard_flags(discard, &reopen_state->flags) != 0) { | |
4719 | error_setg(errp, "Invalid discard option"); | |
4720 | ret = -EINVAL; | |
4721 | goto error; | |
4722 | } | |
4723 | } | |
4724 | ||
543770bd AG |
4725 | reopen_state->detect_zeroes = |
4726 | bdrv_parse_detect_zeroes(opts, reopen_state->flags, &local_err); | |
4727 | if (local_err) { | |
4728 | error_propagate(errp, local_err); | |
4729 | ret = -EINVAL; | |
4730 | goto error; | |
4731 | } | |
4732 | ||
57f9db9a AG |
4733 | /* All other options (including node-name and driver) must be unchanged. |
4734 | * Put them back into the QDict, so that they are checked at the end | |
4735 | * of this function. */ | |
4736 | qemu_opts_to_qdict(opts, reopen_state->options); | |
ccf9dc07 | 4737 | |
3d8ce171 JC |
4738 | /* If we are to stay read-only, do not allow permission change |
4739 | * to r/w. Attempting to set to r/w may fail if either BDRV_O_ALLOW_RDWR is | |
4740 | * not set, or if the BDS still has copy_on_read enabled */ | |
4741 | read_only = !(reopen_state->flags & BDRV_O_RDWR); | |
54a32bfe | 4742 | ret = bdrv_can_set_read_only(reopen_state->bs, read_only, true, &local_err); |
3d8ce171 JC |
4743 | if (local_err) { |
4744 | error_propagate(errp, local_err); | |
e971aa12 JC |
4745 | goto error; |
4746 | } | |
4747 | ||
e971aa12 | 4748 | if (drv->bdrv_reopen_prepare) { |
faf116b4 AG |
4749 | /* |
4750 | * If a driver-specific option is missing, it means that we | |
4751 | * should reset it to its default value. | |
4752 | * But not all options allow that, so we need to check it first. | |
4753 | */ | |
4754 | ret = bdrv_reset_options_allowed(reopen_state->bs, | |
4755 | reopen_state->options, errp); | |
4756 | if (ret) { | |
4757 | goto error; | |
4758 | } | |
4759 | ||
e971aa12 JC |
4760 | ret = drv->bdrv_reopen_prepare(reopen_state, queue, &local_err); |
4761 | if (ret) { | |
4762 | if (local_err != NULL) { | |
4763 | error_propagate(errp, local_err); | |
4764 | } else { | |
f30c66ba | 4765 | bdrv_refresh_filename(reopen_state->bs); |
d8b6895f LC |
4766 | error_setg(errp, "failed while preparing to reopen image '%s'", |
4767 | reopen_state->bs->filename); | |
e971aa12 JC |
4768 | } |
4769 | goto error; | |
4770 | } | |
4771 | } else { | |
4772 | /* It is currently mandatory to have a bdrv_reopen_prepare() | |
4773 | * handler for each supported drv. */ | |
81e5f78a AG |
4774 | error_setg(errp, "Block format '%s' used by node '%s' " |
4775 | "does not support reopening files", drv->format_name, | |
4776 | bdrv_get_device_or_node_name(reopen_state->bs)); | |
e971aa12 JC |
4777 | ret = -1; |
4778 | goto error; | |
4779 | } | |
4780 | ||
9ad08c44 HR |
4781 | drv_prepared = true; |
4782 | ||
bacd9b87 AG |
4783 | /* |
4784 | * We must provide the 'backing' option if the BDS has a backing | |
4785 | * file or if the image file has a backing file name as part of | |
4786 | * its metadata. Otherwise the 'backing' option can be omitted. | |
4787 | */ | |
4788 | if (drv->supports_backing && reopen_state->backing_missing && | |
1d42f48c | 4789 | (reopen_state->bs->backing || reopen_state->bs->backing_file[0])) { |
8546632e AG |
4790 | error_setg(errp, "backing is missing for '%s'", |
4791 | reopen_state->bs->node_name); | |
4792 | ret = -EINVAL; | |
4793 | goto error; | |
4794 | } | |
4795 | ||
cb828c31 AG |
4796 | /* |
4797 | * Allow changing the 'backing' option. The new value can be | |
4798 | * either a reference to an existing node (using its node name) | |
4799 | * or NULL to simply detach the current backing file. | |
4800 | */ | |
ecd30d2d AG |
4801 | ret = bdrv_reopen_parse_file_or_backing(reopen_state, true, |
4802 | change_child_tran, errp); | |
cb828c31 AG |
4803 | if (ret < 0) { |
4804 | goto error; | |
4805 | } | |
4806 | qdict_del(reopen_state->options, "backing"); | |
4807 | ||
ecd30d2d AG |
4808 | /* Allow changing the 'file' option. In this case NULL is not allowed */ |
4809 | ret = bdrv_reopen_parse_file_or_backing(reopen_state, false, | |
4810 | change_child_tran, errp); | |
4811 | if (ret < 0) { | |
4812 | goto error; | |
4813 | } | |
4814 | qdict_del(reopen_state->options, "file"); | |
4815 | ||
4d2cb092 KW |
4816 | /* Options that are not handled are only okay if they are unchanged |
4817 | * compared to the old state. It is expected that some options are only | |
4818 | * used for the initial open, but not reopen (e.g. filename) */ | |
4819 | if (qdict_size(reopen_state->options)) { | |
4820 | const QDictEntry *entry = qdict_first(reopen_state->options); | |
4821 | ||
4822 | do { | |
54fd1b0d HR |
4823 | QObject *new = entry->value; |
4824 | QObject *old = qdict_get(reopen_state->bs->options, entry->key); | |
4825 | ||
db905283 AG |
4826 | /* Allow child references (child_name=node_name) as long as they |
4827 | * point to the current child (i.e. everything stays the same). */ | |
4828 | if (qobject_type(new) == QTYPE_QSTRING) { | |
4829 | BdrvChild *child; | |
4830 | QLIST_FOREACH(child, &reopen_state->bs->children, next) { | |
4831 | if (!strcmp(child->name, entry->key)) { | |
4832 | break; | |
4833 | } | |
4834 | } | |
4835 | ||
4836 | if (child) { | |
410f44f5 MA |
4837 | if (!strcmp(child->bs->node_name, |
4838 | qstring_get_str(qobject_to(QString, new)))) { | |
db905283 AG |
4839 | continue; /* Found child with this name, skip option */ |
4840 | } | |
4841 | } | |
4842 | } | |
4843 | ||
129c7d1c | 4844 | /* |
54fd1b0d HR |
4845 | * TODO: When using -drive to specify blockdev options, all values |
4846 | * will be strings; however, when using -blockdev, blockdev-add or | |
4847 | * filenames using the json:{} pseudo-protocol, they will be | |
4848 | * correctly typed. | |
4849 | * In contrast, reopening options are (currently) always strings | |
4850 | * (because you can only specify them through qemu-io; all other | |
4851 | * callers do not specify any options). | |
4852 | * Therefore, when using anything other than -drive to create a BDS, | |
4853 | * this cannot detect non-string options as unchanged, because | |
4854 | * qobject_is_equal() always returns false for objects of different | |
4855 | * type. In the future, this should be remedied by correctly typing | |
4856 | * all options. For now, this is not too big of an issue because | |
4857 | * the user can simply omit options which cannot be changed anyway, | |
4858 | * so they will stay unchanged. | |
129c7d1c | 4859 | */ |
54fd1b0d | 4860 | if (!qobject_is_equal(new, old)) { |
4d2cb092 KW |
4861 | error_setg(errp, "Cannot change the option '%s'", entry->key); |
4862 | ret = -EINVAL; | |
4863 | goto error; | |
4864 | } | |
4865 | } while ((entry = qdict_next(reopen_state->options, entry))); | |
4866 | } | |
4867 | ||
e971aa12 JC |
4868 | ret = 0; |
4869 | ||
4c8350fe AG |
4870 | /* Restore the original reopen_state->options QDict */ |
4871 | qobject_unref(reopen_state->options); | |
4872 | reopen_state->options = qobject_ref(orig_reopen_opts); | |
4873 | ||
e971aa12 | 4874 | error: |
9ad08c44 HR |
4875 | if (ret < 0 && drv_prepared) { |
4876 | /* drv->bdrv_reopen_prepare() has succeeded, so we need to | |
4877 | * call drv->bdrv_reopen_abort() before signaling an error | |
4878 | * (bdrv_reopen_multiple() will not call bdrv_reopen_abort() | |
4879 | * when the respective bdrv_reopen_prepare() has failed) */ | |
4880 | if (drv->bdrv_reopen_abort) { | |
4881 | drv->bdrv_reopen_abort(reopen_state); | |
4882 | } | |
4883 | } | |
ccf9dc07 | 4884 | qemu_opts_del(opts); |
4c8350fe | 4885 | qobject_unref(orig_reopen_opts); |
593b3071 | 4886 | g_free(discard); |
e971aa12 JC |
4887 | return ret; |
4888 | } | |
4889 | ||
4890 | /* | |
4891 | * Takes the staged changes for the reopen from bdrv_reopen_prepare(), and | |
4892 | * makes them final by swapping the staging BlockDriverState contents into | |
4893 | * the active BlockDriverState contents. | |
4894 | */ | |
53e96d1e | 4895 | static void bdrv_reopen_commit(BDRVReopenState *reopen_state) |
e971aa12 JC |
4896 | { |
4897 | BlockDriver *drv; | |
50bf65ba | 4898 | BlockDriverState *bs; |
50196d7a | 4899 | BdrvChild *child; |
e971aa12 JC |
4900 | |
4901 | assert(reopen_state != NULL); | |
50bf65ba VSO |
4902 | bs = reopen_state->bs; |
4903 | drv = bs->drv; | |
e971aa12 | 4904 | assert(drv != NULL); |
da359909 | 4905 | GLOBAL_STATE_CODE(); |
e971aa12 JC |
4906 | |
4907 | /* If there are any driver level actions to take */ | |
4908 | if (drv->bdrv_reopen_commit) { | |
4909 | drv->bdrv_reopen_commit(reopen_state); | |
4910 | } | |
4911 | ||
4912 | /* set BDS specific flags now */ | |
cb3e7f08 | 4913 | qobject_unref(bs->explicit_options); |
4c8350fe | 4914 | qobject_unref(bs->options); |
ab5b5228 AG |
4915 | qobject_ref(reopen_state->explicit_options); |
4916 | qobject_ref(reopen_state->options); | |
145f598e | 4917 | |
50bf65ba | 4918 | bs->explicit_options = reopen_state->explicit_options; |
4c8350fe | 4919 | bs->options = reopen_state->options; |
50bf65ba | 4920 | bs->open_flags = reopen_state->flags; |
543770bd | 4921 | bs->detect_zeroes = reopen_state->detect_zeroes; |
355ef4ac | 4922 | |
50196d7a AG |
4923 | /* Remove child references from bs->options and bs->explicit_options. |
4924 | * Child options were already removed in bdrv_reopen_queue_child() */ | |
4925 | QLIST_FOREACH(child, &bs->children, next) { | |
4926 | qdict_del(bs->explicit_options, child->name); | |
4927 | qdict_del(bs->options, child->name); | |
4928 | } | |
3d0e8743 VSO |
4929 | /* backing is probably removed, so it's not handled by previous loop */ |
4930 | qdict_del(bs->explicit_options, "backing"); | |
4931 | qdict_del(bs->options, "backing"); | |
4932 | ||
e19b157f | 4933 | bdrv_graph_rdlock_main_loop(); |
1e4c797c | 4934 | bdrv_refresh_limits(bs, NULL, NULL); |
e19b157f | 4935 | bdrv_graph_rdunlock_main_loop(); |
439cc330 | 4936 | bdrv_refresh_total_sectors(bs, bs->total_sectors); |
e971aa12 JC |
4937 | } |
4938 | ||
4939 | /* | |
4940 | * Abort the reopen, and delete and free the staged changes in | |
4941 | * reopen_state | |
4942 | */ | |
53e96d1e | 4943 | static void bdrv_reopen_abort(BDRVReopenState *reopen_state) |
e971aa12 JC |
4944 | { |
4945 | BlockDriver *drv; | |
4946 | ||
4947 | assert(reopen_state != NULL); | |
4948 | drv = reopen_state->bs->drv; | |
4949 | assert(drv != NULL); | |
da359909 | 4950 | GLOBAL_STATE_CODE(); |
e971aa12 JC |
4951 | |
4952 | if (drv->bdrv_reopen_abort) { | |
4953 | drv->bdrv_reopen_abort(reopen_state); | |
4954 | } | |
4955 | } | |
4956 | ||
4957 | ||
64dff520 | 4958 | static void bdrv_close(BlockDriverState *bs) |
fc01f7e7 | 4959 | { |
33384421 | 4960 | BdrvAioNotifier *ban, *ban_next; |
50a3efb0 | 4961 | BdrvChild *child, *next; |
33384421 | 4962 | |
f791bf7f | 4963 | GLOBAL_STATE_CODE(); |
30f55fb8 | 4964 | assert(!bs->refcnt); |
99b7e775 | 4965 | |
fc27291d | 4966 | bdrv_drained_begin(bs); /* complete I/O */ |
58fda173 | 4967 | bdrv_flush(bs); |
53ec73e2 | 4968 | bdrv_drain(bs); /* in case flush left pending I/O */ |
fc27291d | 4969 | |
3cbc002c | 4970 | if (bs->drv) { |
3c005293 | 4971 | if (bs->drv->bdrv_close) { |
7b99a266 | 4972 | /* Must unfreeze all children, so bdrv_unref_child() works */ |
3c005293 VSO |
4973 | bs->drv->bdrv_close(bs); |
4974 | } | |
9a4f4c31 | 4975 | bs->drv = NULL; |
50a3efb0 | 4976 | } |
9a7dedbc | 4977 | |
50a3efb0 | 4978 | QLIST_FOREACH_SAFE(child, &bs->children, next, next) { |
dd4118c7 | 4979 | bdrv_unref_child(bs, child); |
b338082b | 4980 | } |
98f90dba | 4981 | |
5bb04747 VSO |
4982 | assert(!bs->backing); |
4983 | assert(!bs->file); | |
50a3efb0 AG |
4984 | g_free(bs->opaque); |
4985 | bs->opaque = NULL; | |
d73415a3 | 4986 | qatomic_set(&bs->copy_on_read, 0); |
50a3efb0 AG |
4987 | bs->backing_file[0] = '\0'; |
4988 | bs->backing_format[0] = '\0'; | |
4989 | bs->total_sectors = 0; | |
4990 | bs->encrypted = false; | |
4991 | bs->sg = false; | |
cb3e7f08 MAL |
4992 | qobject_unref(bs->options); |
4993 | qobject_unref(bs->explicit_options); | |
50a3efb0 AG |
4994 | bs->options = NULL; |
4995 | bs->explicit_options = NULL; | |
cb3e7f08 | 4996 | qobject_unref(bs->full_open_options); |
50a3efb0 | 4997 | bs->full_open_options = NULL; |
0bc329fb HR |
4998 | g_free(bs->block_status_cache); |
4999 | bs->block_status_cache = NULL; | |
50a3efb0 | 5000 | |
cca43ae1 VSO |
5001 | bdrv_release_named_dirty_bitmaps(bs); |
5002 | assert(QLIST_EMPTY(&bs->dirty_bitmaps)); | |
5003 | ||
33384421 HR |
5004 | QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_next) { |
5005 | g_free(ban); | |
5006 | } | |
5007 | QLIST_INIT(&bs->aio_notifiers); | |
fc27291d | 5008 | bdrv_drained_end(bs); |
1a6d3bd2 GK |
5009 | |
5010 | /* | |
5011 | * If we're still inside some bdrv_drain_all_begin()/end() sections, end | |
5012 | * them now since this BDS won't exist anymore when bdrv_drain_all_end() | |
5013 | * gets called. | |
5014 | */ | |
5015 | if (bs->quiesce_counter) { | |
5016 | bdrv_drain_all_end_quiesce(bs); | |
5017 | } | |
b338082b FB |
5018 | } |
5019 | ||
2bc93fed MK |
5020 | void bdrv_close_all(void) |
5021 | { | |
f791bf7f | 5022 | GLOBAL_STATE_CODE(); |
880eeec6 | 5023 | assert(job_next(NULL) == NULL); |
ca9bd24c HR |
5024 | |
5025 | /* Drop references from requests still in flight, such as canceled block | |
5026 | * jobs whose AIO context has not been polled yet */ | |
5027 | bdrv_drain_all(); | |
2bc93fed | 5028 | |
ca9bd24c HR |
5029 | blk_remove_all_bs(); |
5030 | blockdev_close_all_bdrv_states(); | |
ed78cda3 | 5031 | |
a1a2af07 | 5032 | assert(QTAILQ_EMPTY(&all_bdrv_states)); |
2bc93fed MK |
5033 | } |
5034 | ||
d0ac0380 KW |
5035 | static bool should_update_child(BdrvChild *c, BlockDriverState *to) |
5036 | { | |
2f30b7c3 VSO |
5037 | GQueue *queue; |
5038 | GHashTable *found; | |
5039 | bool ret; | |
d0ac0380 | 5040 | |
bd86fb99 | 5041 | if (c->klass->stay_at_node) { |
d0ac0380 KW |
5042 | return false; |
5043 | } | |
5044 | ||
ec9f10fe HR |
5045 | /* If the child @c belongs to the BDS @to, replacing the current |
5046 | * c->bs by @to would mean to create a loop. | |
5047 | * | |
5048 | * Such a case occurs when appending a BDS to a backing chain. | |
5049 | * For instance, imagine the following chain: | |
5050 | * | |
5051 | * guest device -> node A -> further backing chain... | |
5052 | * | |
5053 | * Now we create a new BDS B which we want to put on top of this | |
5054 | * chain, so we first attach A as its backing node: | |
5055 | * | |
5056 | * node B | |
5057 | * | | |
5058 | * v | |
5059 | * guest device -> node A -> further backing chain... | |
5060 | * | |
5061 | * Finally we want to replace A by B. When doing that, we want to | |
5062 | * replace all pointers to A by pointers to B -- except for the | |
5063 | * pointer from B because (1) that would create a loop, and (2) | |
5064 | * that pointer should simply stay intact: | |
5065 | * | |
5066 | * guest device -> node B | |
5067 | * | | |
5068 | * v | |
5069 | * node A -> further backing chain... | |
5070 | * | |
5071 | * In general, when replacing a node A (c->bs) by a node B (@to), | |
5072 | * if A is a child of B, that means we cannot replace A by B there | |
5073 | * because that would create a loop. Silently detaching A from B | |
5074 | * is also not really an option. So overall just leaving A in | |
2f30b7c3 VSO |
5075 | * place there is the most sensible choice. |
5076 | * | |
5077 | * We would also create a loop in any cases where @c is only | |
5078 | * indirectly referenced by @to. Prevent this by returning false | |
5079 | * if @c is found (by breadth-first search) anywhere in the whole | |
5080 | * subtree of @to. | |
5081 | */ | |
5082 | ||
5083 | ret = true; | |
5084 | found = g_hash_table_new(NULL, NULL); | |
5085 | g_hash_table_add(found, to); | |
5086 | queue = g_queue_new(); | |
5087 | g_queue_push_tail(queue, to); | |
5088 | ||
5089 | while (!g_queue_is_empty(queue)) { | |
5090 | BlockDriverState *v = g_queue_pop_head(queue); | |
5091 | BdrvChild *c2; | |
5092 | ||
5093 | QLIST_FOREACH(c2, &v->children, next) { | |
5094 | if (c2 == c) { | |
5095 | ret = false; | |
5096 | break; | |
5097 | } | |
5098 | ||
5099 | if (g_hash_table_contains(found, c2->bs)) { | |
5100 | continue; | |
5101 | } | |
5102 | ||
5103 | g_queue_push_tail(queue, c2->bs); | |
5104 | g_hash_table_add(found, c2->bs); | |
d0ac0380 KW |
5105 | } |
5106 | } | |
5107 | ||
2f30b7c3 VSO |
5108 | g_queue_free(queue); |
5109 | g_hash_table_destroy(found); | |
5110 | ||
5111 | return ret; | |
d0ac0380 KW |
5112 | } |
5113 | ||
57f08941 | 5114 | static void bdrv_remove_child_commit(void *opaque) |
46541ee5 | 5115 | { |
bdb73476 | 5116 | GLOBAL_STATE_CODE(); |
5bb04747 | 5117 | bdrv_child_free(opaque); |
82b54cf5 HR |
5118 | } |
5119 | ||
57f08941 VSO |
5120 | static TransactionActionDrv bdrv_remove_child_drv = { |
5121 | .commit = bdrv_remove_child_commit, | |
46541ee5 VSO |
5122 | }; |
5123 | ||
57f08941 VSO |
5124 | /* Function doesn't update permissions, caller is responsible for this. */ |
5125 | static void bdrv_remove_child(BdrvChild *child, Transaction *tran) | |
46541ee5 | 5126 | { |
46541ee5 VSO |
5127 | if (!child) { |
5128 | return; | |
5129 | } | |
5130 | ||
5131 | if (child->bs) { | |
23987471 KW |
5132 | BlockDriverState *bs = child->bs; |
5133 | bdrv_drained_begin(bs); | |
a2c37a30 | 5134 | bdrv_replace_child_tran(child, NULL, tran); |
23987471 | 5135 | bdrv_drained_end(bs); |
46541ee5 VSO |
5136 | } |
5137 | ||
57f08941 | 5138 | tran_add(tran, &bdrv_remove_child_drv, child); |
46541ee5 VSO |
5139 | } |
5140 | ||
23987471 KW |
5141 | static void undrain_on_clean_cb(void *opaque) |
5142 | { | |
5143 | bdrv_drained_end(opaque); | |
5144 | } | |
5145 | ||
5146 | static TransactionActionDrv undrain_on_clean = { | |
5147 | .clean = undrain_on_clean_cb, | |
5148 | }; | |
5149 | ||
117caba9 VSO |
5150 | static int bdrv_replace_node_noperm(BlockDriverState *from, |
5151 | BlockDriverState *to, | |
5152 | bool auto_skip, Transaction *tran, | |
5153 | Error **errp) | |
5154 | { | |
5155 | BdrvChild *c, *next; | |
5156 | ||
bdb73476 | 5157 | GLOBAL_STATE_CODE(); |
82b54cf5 | 5158 | |
23987471 KW |
5159 | bdrv_drained_begin(from); |
5160 | bdrv_drained_begin(to); | |
5161 | tran_add(tran, &undrain_on_clean, from); | |
5162 | tran_add(tran, &undrain_on_clean, to); | |
5163 | ||
117caba9 VSO |
5164 | QLIST_FOREACH_SAFE(c, &from->parents, next_parent, next) { |
5165 | assert(c->bs == from); | |
5166 | if (!should_update_child(c, to)) { | |
5167 | if (auto_skip) { | |
5168 | continue; | |
5169 | } | |
5170 | error_setg(errp, "Should not change '%s' link to '%s'", | |
5171 | c->name, from->node_name); | |
5172 | return -EINVAL; | |
5173 | } | |
5174 | if (c->frozen) { | |
5175 | error_setg(errp, "Cannot change '%s' link to '%s'", | |
5176 | c->name, from->node_name); | |
5177 | return -EPERM; | |
5178 | } | |
0f0b1e29 | 5179 | bdrv_replace_child_tran(c, to, tran); |
117caba9 VSO |
5180 | } |
5181 | ||
5182 | return 0; | |
5183 | } | |
5184 | ||
313274bb VSO |
5185 | /* |
5186 | * With auto_skip=true bdrv_replace_node_common skips updating from parents | |
5187 | * if it creates a parent-child relation loop or if parent is block-job. | |
5188 | * | |
5189 | * With auto_skip=false the error is returned if from has a parent which should | |
5190 | * not be updated. | |
3108a15c VSO |
5191 | * |
5192 | * With @detach_subchain=true @to must be in a backing chain of @from. In this | |
5193 | * case backing link of the cow-parent of @to is removed. | |
313274bb | 5194 | */ |
a1e708fc VSO |
5195 | static int bdrv_replace_node_common(BlockDriverState *from, |
5196 | BlockDriverState *to, | |
3108a15c VSO |
5197 | bool auto_skip, bool detach_subchain, |
5198 | Error **errp) | |
dd62f1ca | 5199 | { |
3bb0e298 | 5200 | Transaction *tran = tran_new(); |
3bb0e298 | 5201 | g_autoptr(GSList) refresh_list = NULL; |
2d369d6e | 5202 | BlockDriverState *to_cow_parent = NULL; |
234ac1a9 KW |
5203 | int ret; |
5204 | ||
bdb73476 | 5205 | GLOBAL_STATE_CODE(); |
82b54cf5 | 5206 | |
3108a15c VSO |
5207 | if (detach_subchain) { |
5208 | assert(bdrv_chain_contains(from, to)); | |
5209 | assert(from != to); | |
5210 | for (to_cow_parent = from; | |
5211 | bdrv_filter_or_cow_bs(to_cow_parent) != to; | |
5212 | to_cow_parent = bdrv_filter_or_cow_bs(to_cow_parent)) | |
5213 | { | |
5214 | ; | |
5215 | } | |
5216 | } | |
5217 | ||
234ac1a9 KW |
5218 | /* Make sure that @from doesn't go away until we have successfully attached |
5219 | * all of its parents to @to. */ | |
5220 | bdrv_ref(from); | |
dd62f1ca | 5221 | |
f871abd6 | 5222 | assert(qemu_get_current_aio_context() == qemu_get_aio_context()); |
30dd65f3 | 5223 | assert(bdrv_get_aio_context(from) == bdrv_get_aio_context(to)); |
f871abd6 KW |
5224 | bdrv_drained_begin(from); |
5225 | ||
3bb0e298 VSO |
5226 | /* |
5227 | * Do the replacement without permission update. | |
5228 | * Replacement may influence the permissions, we should calculate new | |
5229 | * permissions based on new graph. If we fail, we'll roll-back the | |
5230 | * replacement. | |
5231 | */ | |
117caba9 VSO |
5232 | ret = bdrv_replace_node_noperm(from, to, auto_skip, tran, errp); |
5233 | if (ret < 0) { | |
5234 | goto out; | |
234ac1a9 KW |
5235 | } |
5236 | ||
3108a15c | 5237 | if (detach_subchain) { |
f38eaec4 | 5238 | bdrv_remove_child(bdrv_filter_or_cow_child(to_cow_parent), tran); |
3108a15c VSO |
5239 | } |
5240 | ||
fb0ff4d1 VSO |
5241 | refresh_list = g_slist_prepend(refresh_list, to); |
5242 | refresh_list = g_slist_prepend(refresh_list, from); | |
9bd910e2 | 5243 | |
3bb0e298 VSO |
5244 | ret = bdrv_list_refresh_perms(refresh_list, NULL, tran, errp); |
5245 | if (ret < 0) { | |
5246 | goto out; | |
dd62f1ca | 5247 | } |
234ac1a9 | 5248 | |
a1e708fc VSO |
5249 | ret = 0; |
5250 | ||
234ac1a9 | 5251 | out: |
3bb0e298 VSO |
5252 | tran_finalize(tran, ret); |
5253 | ||
f871abd6 | 5254 | bdrv_drained_end(from); |
234ac1a9 | 5255 | bdrv_unref(from); |
a1e708fc VSO |
5256 | |
5257 | return ret; | |
dd62f1ca KW |
5258 | } |
5259 | ||
a1e708fc VSO |
5260 | int bdrv_replace_node(BlockDriverState *from, BlockDriverState *to, |
5261 | Error **errp) | |
313274bb | 5262 | { |
f791bf7f EGE |
5263 | GLOBAL_STATE_CODE(); |
5264 | ||
3108a15c VSO |
5265 | return bdrv_replace_node_common(from, to, true, false, errp); |
5266 | } | |
5267 | ||
5268 | int bdrv_drop_filter(BlockDriverState *bs, Error **errp) | |
5269 | { | |
f791bf7f EGE |
5270 | GLOBAL_STATE_CODE(); |
5271 | ||
3108a15c VSO |
5272 | return bdrv_replace_node_common(bs, bdrv_filter_or_cow_bs(bs), true, true, |
5273 | errp); | |
313274bb VSO |
5274 | } |
5275 | ||
4ddc07ca PB |
5276 | /* |
5277 | * Add new bs contents at the top of an image chain while the chain is | |
5278 | * live, while keeping required fields on the top layer. | |
5279 | * | |
5280 | * This will modify the BlockDriverState fields, and swap contents | |
5281 | * between bs_new and bs_top. Both bs_new and bs_top are modified. | |
5282 | * | |
2272edcf VSO |
5283 | * bs_new must not be attached to a BlockBackend and must not have backing |
5284 | * child. | |
4ddc07ca PB |
5285 | * |
5286 | * This function does not create any image files. | |
60d90bf4 SG |
5287 | * |
5288 | * The caller must hold the AioContext lock for @bs_top. | |
4ddc07ca | 5289 | */ |
a1e708fc VSO |
5290 | int bdrv_append(BlockDriverState *bs_new, BlockDriverState *bs_top, |
5291 | Error **errp) | |
4ddc07ca | 5292 | { |
2272edcf | 5293 | int ret; |
5bb04747 | 5294 | BdrvChild *child; |
2272edcf | 5295 | Transaction *tran = tran_new(); |
60d90bf4 | 5296 | AioContext *old_context, *new_context = NULL; |
2272edcf | 5297 | |
f791bf7f EGE |
5298 | GLOBAL_STATE_CODE(); |
5299 | ||
2272edcf VSO |
5300 | assert(!bs_new->backing); |
5301 | ||
60d90bf4 SG |
5302 | old_context = bdrv_get_aio_context(bs_top); |
5303 | ||
5bb04747 VSO |
5304 | child = bdrv_attach_child_noperm(bs_new, bs_top, "backing", |
5305 | &child_of_bds, bdrv_backing_role(bs_new), | |
5306 | tran, errp); | |
5307 | if (!child) { | |
5308 | ret = -EINVAL; | |
2272edcf | 5309 | goto out; |
b2c2832c | 5310 | } |
dd62f1ca | 5311 | |
60d90bf4 SG |
5312 | /* |
5313 | * bdrv_attach_child_noperm could change the AioContext of bs_top. | |
5314 | * bdrv_replace_node_noperm calls bdrv_drained_begin, so let's temporarily | |
5315 | * hold the new AioContext, since bdrv_drained_begin calls BDRV_POLL_WHILE | |
5316 | * that assumes the new lock is taken. | |
5317 | */ | |
5318 | new_context = bdrv_get_aio_context(bs_top); | |
5319 | ||
5320 | if (old_context != new_context) { | |
5321 | aio_context_release(old_context); | |
5322 | aio_context_acquire(new_context); | |
5323 | } | |
5324 | ||
2272edcf | 5325 | ret = bdrv_replace_node_noperm(bs_top, bs_new, true, tran, errp); |
a1e708fc | 5326 | if (ret < 0) { |
2272edcf | 5327 | goto out; |
234ac1a9 | 5328 | } |
4ddc07ca | 5329 | |
f1316edb | 5330 | ret = bdrv_refresh_perms(bs_new, tran, errp); |
2272edcf VSO |
5331 | out: |
5332 | tran_finalize(tran, ret); | |
5333 | ||
e19b157f | 5334 | bdrv_graph_rdlock_main_loop(); |
1e4c797c | 5335 | bdrv_refresh_limits(bs_top, NULL, NULL); |
e19b157f | 5336 | bdrv_graph_rdunlock_main_loop(); |
2272edcf | 5337 | |
60d90bf4 SG |
5338 | if (new_context && old_context != new_context) { |
5339 | aio_context_release(new_context); | |
5340 | aio_context_acquire(old_context); | |
5341 | } | |
5342 | ||
2272edcf | 5343 | return ret; |
8802d1fd JC |
5344 | } |
5345 | ||
bd8f4c42 VSO |
5346 | /* Not for empty child */ |
5347 | int bdrv_replace_child_bs(BdrvChild *child, BlockDriverState *new_bs, | |
5348 | Error **errp) | |
5349 | { | |
5350 | int ret; | |
5351 | Transaction *tran = tran_new(); | |
bd8f4c42 VSO |
5352 | g_autoptr(GSList) refresh_list = NULL; |
5353 | BlockDriverState *old_bs = child->bs; | |
5354 | ||
f791bf7f EGE |
5355 | GLOBAL_STATE_CODE(); |
5356 | ||
bd8f4c42 VSO |
5357 | bdrv_ref(old_bs); |
5358 | bdrv_drained_begin(old_bs); | |
5359 | bdrv_drained_begin(new_bs); | |
5360 | ||
0f0b1e29 | 5361 | bdrv_replace_child_tran(child, new_bs, tran); |
bd8f4c42 | 5362 | |
fb0ff4d1 VSO |
5363 | refresh_list = g_slist_prepend(refresh_list, old_bs); |
5364 | refresh_list = g_slist_prepend(refresh_list, new_bs); | |
bd8f4c42 VSO |
5365 | |
5366 | ret = bdrv_list_refresh_perms(refresh_list, NULL, tran, errp); | |
5367 | ||
5368 | tran_finalize(tran, ret); | |
5369 | ||
5370 | bdrv_drained_end(old_bs); | |
5371 | bdrv_drained_end(new_bs); | |
5372 | bdrv_unref(old_bs); | |
5373 | ||
5374 | return ret; | |
5375 | } | |
5376 | ||
4f6fd349 | 5377 | static void bdrv_delete(BlockDriverState *bs) |
b338082b | 5378 | { |
3718d8ab | 5379 | assert(bdrv_op_blocker_is_empty(bs)); |
4f6fd349 | 5380 | assert(!bs->refcnt); |
f791bf7f | 5381 | GLOBAL_STATE_CODE(); |
18846dee | 5382 | |
1b7bdbc1 | 5383 | /* remove from list, if necessary */ |
63eaaae0 KW |
5384 | if (bs->node_name[0] != '\0') { |
5385 | QTAILQ_REMOVE(&graph_bdrv_states, bs, node_list); | |
5386 | } | |
2c1d04e0 HR |
5387 | QTAILQ_REMOVE(&all_bdrv_states, bs, bs_list); |
5388 | ||
30c321f9 AK |
5389 | bdrv_close(bs); |
5390 | ||
7267c094 | 5391 | g_free(bs); |
fc01f7e7 FB |
5392 | } |
5393 | ||
96796fae VSO |
5394 | |
5395 | /* | |
5396 | * Replace @bs by newly created block node. | |
5397 | * | |
5398 | * @options is a QDict of options to pass to the block drivers, or NULL for an | |
5399 | * empty set of options. The reference to the QDict belongs to the block layer | |
5400 | * after the call (even on failure), so if the caller intends to reuse the | |
5401 | * dictionary, it needs to use qobject_ref() before calling bdrv_open. | |
8823407c KW |
5402 | * |
5403 | * The caller holds the AioContext lock for @bs. It must make sure that @bs | |
5404 | * stays in the same AioContext, i.e. @options must not refer to nodes in a | |
5405 | * different AioContext. | |
96796fae VSO |
5406 | */ |
5407 | BlockDriverState *bdrv_insert_node(BlockDriverState *bs, QDict *options, | |
8872ef78 AS |
5408 | int flags, Error **errp) |
5409 | { | |
f053b7e8 VSO |
5410 | ERRP_GUARD(); |
5411 | int ret; | |
8823407c | 5412 | AioContext *ctx = bdrv_get_aio_context(bs); |
b11c8739 VSO |
5413 | BlockDriverState *new_node_bs = NULL; |
5414 | const char *drvname, *node_name; | |
5415 | BlockDriver *drv; | |
5416 | ||
5417 | drvname = qdict_get_try_str(options, "driver"); | |
5418 | if (!drvname) { | |
5419 | error_setg(errp, "driver is not specified"); | |
5420 | goto fail; | |
5421 | } | |
5422 | ||
5423 | drv = bdrv_find_format(drvname); | |
5424 | if (!drv) { | |
5425 | error_setg(errp, "Unknown driver: '%s'", drvname); | |
5426 | goto fail; | |
5427 | } | |
8872ef78 | 5428 | |
b11c8739 VSO |
5429 | node_name = qdict_get_try_str(options, "node-name"); |
5430 | ||
f791bf7f EGE |
5431 | GLOBAL_STATE_CODE(); |
5432 | ||
8823407c KW |
5433 | aio_context_release(ctx); |
5434 | aio_context_acquire(qemu_get_aio_context()); | |
b11c8739 VSO |
5435 | new_node_bs = bdrv_new_open_driver_opts(drv, node_name, options, flags, |
5436 | errp); | |
8823407c KW |
5437 | aio_context_release(qemu_get_aio_context()); |
5438 | aio_context_acquire(ctx); | |
5439 | assert(bdrv_get_aio_context(bs) == ctx); | |
5440 | ||
b11c8739 VSO |
5441 | options = NULL; /* bdrv_new_open_driver() eats options */ |
5442 | if (!new_node_bs) { | |
8872ef78 | 5443 | error_prepend(errp, "Could not create node: "); |
b11c8739 | 5444 | goto fail; |
8872ef78 AS |
5445 | } |
5446 | ||
5447 | bdrv_drained_begin(bs); | |
f053b7e8 | 5448 | ret = bdrv_replace_node(bs, new_node_bs, errp); |
8872ef78 AS |
5449 | bdrv_drained_end(bs); |
5450 | ||
f053b7e8 VSO |
5451 | if (ret < 0) { |
5452 | error_prepend(errp, "Could not replace node: "); | |
b11c8739 | 5453 | goto fail; |
8872ef78 AS |
5454 | } |
5455 | ||
5456 | return new_node_bs; | |
b11c8739 VSO |
5457 | |
5458 | fail: | |
5459 | qobject_unref(options); | |
5460 | bdrv_unref(new_node_bs); | |
5461 | return NULL; | |
8872ef78 AS |
5462 | } |
5463 | ||
e97fc193 AL |
5464 | /* |
5465 | * Run consistency checks on an image | |
5466 | * | |
e076f338 | 5467 | * Returns 0 if the check could be completed (it doesn't mean that the image is |
a1c7273b | 5468 | * free of errors) or -errno when an internal error occurred. The results of the |
e076f338 | 5469 | * check are stored in res. |
e97fc193 | 5470 | */ |
21c2283e VSO |
5471 | int coroutine_fn bdrv_co_check(BlockDriverState *bs, |
5472 | BdrvCheckResult *res, BdrvCheckMode fix) | |
e97fc193 | 5473 | { |
1581a70d | 5474 | IO_CODE(); |
1b3ff9fe | 5475 | assert_bdrv_graph_readable(); |
908bcd54 HR |
5476 | if (bs->drv == NULL) { |
5477 | return -ENOMEDIUM; | |
5478 | } | |
2fd61638 | 5479 | if (bs->drv->bdrv_co_check == NULL) { |
e97fc193 AL |
5480 | return -ENOTSUP; |
5481 | } | |
5482 | ||
e076f338 | 5483 | memset(res, 0, sizeof(*res)); |
2fd61638 PB |
5484 | return bs->drv->bdrv_co_check(bs, res, fix); |
5485 | } | |
5486 | ||
756e6736 KW |
5487 | /* |
5488 | * Return values: | |
5489 | * 0 - success | |
5490 | * -EINVAL - backing format specified, but no file | |
5491 | * -ENOSPC - can't update the backing file because no space is left in the | |
5492 | * image file header | |
5493 | * -ENOTSUP - format driver doesn't support changing the backing file | |
5494 | */ | |
e54ee1b3 | 5495 | int bdrv_change_backing_file(BlockDriverState *bs, const char *backing_file, |
497a30db | 5496 | const char *backing_fmt, bool require) |
756e6736 KW |
5497 | { |
5498 | BlockDriver *drv = bs->drv; | |
469ef350 | 5499 | int ret; |
756e6736 | 5500 | |
f791bf7f EGE |
5501 | GLOBAL_STATE_CODE(); |
5502 | ||
d470ad42 HR |
5503 | if (!drv) { |
5504 | return -ENOMEDIUM; | |
5505 | } | |
5506 | ||
5f377794 PB |
5507 | /* Backing file format doesn't make sense without a backing file */ |
5508 | if (backing_fmt && !backing_file) { | |
5509 | return -EINVAL; | |
5510 | } | |
5511 | ||
497a30db EB |
5512 | if (require && backing_file && !backing_fmt) { |
5513 | return -EINVAL; | |
e54ee1b3 EB |
5514 | } |
5515 | ||
756e6736 | 5516 | if (drv->bdrv_change_backing_file != NULL) { |
469ef350 | 5517 | ret = drv->bdrv_change_backing_file(bs, backing_file, backing_fmt); |
756e6736 | 5518 | } else { |
469ef350 | 5519 | ret = -ENOTSUP; |
756e6736 | 5520 | } |
469ef350 PB |
5521 | |
5522 | if (ret == 0) { | |
5523 | pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_file ?: ""); | |
5524 | pstrcpy(bs->backing_format, sizeof(bs->backing_format), backing_fmt ?: ""); | |
998c2019 HR |
5525 | pstrcpy(bs->auto_backing_file, sizeof(bs->auto_backing_file), |
5526 | backing_file ?: ""); | |
469ef350 PB |
5527 | } |
5528 | return ret; | |
756e6736 KW |
5529 | } |
5530 | ||
6ebdcee2 | 5531 | /* |
dcf3f9b2 HR |
5532 | * Finds the first non-filter node above bs in the chain between |
5533 | * active and bs. The returned node is either an immediate parent of | |
5534 | * bs, or there are only filter nodes between the two. | |
6ebdcee2 JC |
5535 | * |
5536 | * Returns NULL if bs is not found in active's image chain, | |
5537 | * or if active == bs. | |
4caf0fcd JC |
5538 | * |
5539 | * Returns the bottommost base image if bs == NULL. | |
6ebdcee2 JC |
5540 | */ |
5541 | BlockDriverState *bdrv_find_overlay(BlockDriverState *active, | |
5542 | BlockDriverState *bs) | |
5543 | { | |
f791bf7f EGE |
5544 | |
5545 | GLOBAL_STATE_CODE(); | |
5546 | ||
dcf3f9b2 HR |
5547 | bs = bdrv_skip_filters(bs); |
5548 | active = bdrv_skip_filters(active); | |
5549 | ||
5550 | while (active) { | |
5551 | BlockDriverState *next = bdrv_backing_chain_next(active); | |
5552 | if (bs == next) { | |
5553 | return active; | |
5554 | } | |
5555 | active = next; | |
6ebdcee2 JC |
5556 | } |
5557 | ||
dcf3f9b2 | 5558 | return NULL; |
4caf0fcd | 5559 | } |
6ebdcee2 | 5560 | |
4caf0fcd JC |
5561 | /* Given a BDS, searches for the base layer. */ |
5562 | BlockDriverState *bdrv_find_base(BlockDriverState *bs) | |
5563 | { | |
f791bf7f EGE |
5564 | GLOBAL_STATE_CODE(); |
5565 | ||
4caf0fcd | 5566 | return bdrv_find_overlay(bs, NULL); |
6ebdcee2 JC |
5567 | } |
5568 | ||
2cad1ebe | 5569 | /* |
7b99a266 HR |
5570 | * Return true if at least one of the COW (backing) and filter links |
5571 | * between @bs and @base is frozen. @errp is set if that's the case. | |
0f0998f6 | 5572 | * @base must be reachable from @bs, or NULL. |
2cad1ebe AG |
5573 | */ |
5574 | bool bdrv_is_backing_chain_frozen(BlockDriverState *bs, BlockDriverState *base, | |
5575 | Error **errp) | |
5576 | { | |
5577 | BlockDriverState *i; | |
7b99a266 | 5578 | BdrvChild *child; |
2cad1ebe | 5579 | |
f791bf7f EGE |
5580 | GLOBAL_STATE_CODE(); |
5581 | ||
7b99a266 HR |
5582 | for (i = bs; i != base; i = child_bs(child)) { |
5583 | child = bdrv_filter_or_cow_child(i); | |
5584 | ||
5585 | if (child && child->frozen) { | |
2cad1ebe | 5586 | error_setg(errp, "Cannot change '%s' link from '%s' to '%s'", |
7b99a266 | 5587 | child->name, i->node_name, child->bs->node_name); |
2cad1ebe AG |
5588 | return true; |
5589 | } | |
5590 | } | |
5591 | ||
5592 | return false; | |
5593 | } | |
5594 | ||
5595 | /* | |
7b99a266 | 5596 | * Freeze all COW (backing) and filter links between @bs and @base. |
2cad1ebe AG |
5597 | * If any of the links is already frozen the operation is aborted and |
5598 | * none of the links are modified. | |
0f0998f6 | 5599 | * @base must be reachable from @bs, or NULL. |
2cad1ebe AG |
5600 | * Returns 0 on success. On failure returns < 0 and sets @errp. |
5601 | */ | |
5602 | int bdrv_freeze_backing_chain(BlockDriverState *bs, BlockDriverState *base, | |
5603 | Error **errp) | |
5604 | { | |
5605 | BlockDriverState *i; | |
7b99a266 | 5606 | BdrvChild *child; |
2cad1ebe | 5607 | |
f791bf7f EGE |
5608 | GLOBAL_STATE_CODE(); |
5609 | ||
2cad1ebe AG |
5610 | if (bdrv_is_backing_chain_frozen(bs, base, errp)) { |
5611 | return -EPERM; | |
5612 | } | |
5613 | ||
7b99a266 HR |
5614 | for (i = bs; i != base; i = child_bs(child)) { |
5615 | child = bdrv_filter_or_cow_child(i); | |
5616 | if (child && child->bs->never_freeze) { | |
e5182c1c | 5617 | error_setg(errp, "Cannot freeze '%s' link to '%s'", |
7b99a266 | 5618 | child->name, child->bs->node_name); |
e5182c1c HR |
5619 | return -EPERM; |
5620 | } | |
5621 | } | |
5622 | ||
7b99a266 HR |
5623 | for (i = bs; i != base; i = child_bs(child)) { |
5624 | child = bdrv_filter_or_cow_child(i); | |
5625 | if (child) { | |
5626 | child->frozen = true; | |
0f0998f6 | 5627 | } |
2cad1ebe AG |
5628 | } |
5629 | ||
5630 | return 0; | |
5631 | } | |
5632 | ||
5633 | /* | |
7b99a266 HR |
5634 | * Unfreeze all COW (backing) and filter links between @bs and @base. |
5635 | * The caller must ensure that all links are frozen before using this | |
5636 | * function. | |
0f0998f6 | 5637 | * @base must be reachable from @bs, or NULL. |
2cad1ebe AG |
5638 | */ |
5639 | void bdrv_unfreeze_backing_chain(BlockDriverState *bs, BlockDriverState *base) | |
5640 | { | |
5641 | BlockDriverState *i; | |
7b99a266 | 5642 | BdrvChild *child; |
2cad1ebe | 5643 | |
f791bf7f EGE |
5644 | GLOBAL_STATE_CODE(); |
5645 | ||
7b99a266 HR |
5646 | for (i = bs; i != base; i = child_bs(child)) { |
5647 | child = bdrv_filter_or_cow_child(i); | |
5648 | if (child) { | |
5649 | assert(child->frozen); | |
5650 | child->frozen = false; | |
0f0998f6 | 5651 | } |
2cad1ebe AG |
5652 | } |
5653 | } | |
5654 | ||
6ebdcee2 JC |
5655 | /* |
5656 | * Drops images above 'base' up to and including 'top', and sets the image | |
5657 | * above 'top' to have base as its backing file. | |
5658 | * | |
5659 | * Requires that the overlay to 'top' is opened r/w, so that the backing file | |
5660 | * information in 'bs' can be properly updated. | |
5661 | * | |
5662 | * E.g., this will convert the following chain: | |
5663 | * bottom <- base <- intermediate <- top <- active | |
5664 | * | |
5665 | * to | |
5666 | * | |
5667 | * bottom <- base <- active | |
5668 | * | |
5669 | * It is allowed for bottom==base, in which case it converts: | |
5670 | * | |
5671 | * base <- intermediate <- top <- active | |
5672 | * | |
5673 | * to | |
5674 | * | |
5675 | * base <- active | |
5676 | * | |
54e26900 JC |
5677 | * If backing_file_str is non-NULL, it will be used when modifying top's |
5678 | * overlay image metadata. | |
5679 | * | |
6ebdcee2 JC |
5680 | * Error conditions: |
5681 | * if active == top, that is considered an error | |
5682 | * | |
5683 | */ | |
bde70715 KW |
5684 | int bdrv_drop_intermediate(BlockDriverState *top, BlockDriverState *base, |
5685 | const char *backing_file_str) | |
6ebdcee2 | 5686 | { |
6bd858b3 AG |
5687 | BlockDriverState *explicit_top = top; |
5688 | bool update_inherits_from; | |
d669ed6a | 5689 | BdrvChild *c; |
12fa4af6 | 5690 | Error *local_err = NULL; |
6ebdcee2 | 5691 | int ret = -EIO; |
d669ed6a VSO |
5692 | g_autoptr(GSList) updated_children = NULL; |
5693 | GSList *p; | |
6ebdcee2 | 5694 | |
f791bf7f EGE |
5695 | GLOBAL_STATE_CODE(); |
5696 | ||
6858eba0 | 5697 | bdrv_ref(top); |
631086de | 5698 | bdrv_drained_begin(base); |
6858eba0 | 5699 | |
6ebdcee2 JC |
5700 | if (!top->drv || !base->drv) { |
5701 | goto exit; | |
5702 | } | |
5703 | ||
5db15a57 KW |
5704 | /* Make sure that base is in the backing chain of top */ |
5705 | if (!bdrv_chain_contains(top, base)) { | |
6ebdcee2 JC |
5706 | goto exit; |
5707 | } | |
5708 | ||
6bd858b3 AG |
5709 | /* If 'base' recursively inherits from 'top' then we should set |
5710 | * base->inherits_from to top->inherits_from after 'top' and all | |
5711 | * other intermediate nodes have been dropped. | |
5712 | * If 'top' is an implicit node (e.g. "commit_top") we should skip | |
5713 | * it because no one inherits from it. We use explicit_top for that. */ | |
dcf3f9b2 | 5714 | explicit_top = bdrv_skip_implicit_filters(explicit_top); |
6bd858b3 AG |
5715 | update_inherits_from = bdrv_inherits_from_recursive(base, explicit_top); |
5716 | ||
6ebdcee2 | 5717 | /* success - we can delete the intermediate states, and link top->base */ |
f30c66ba HR |
5718 | if (!backing_file_str) { |
5719 | bdrv_refresh_filename(base); | |
5720 | backing_file_str = base->filename; | |
5721 | } | |
61f09cea | 5722 | |
d669ed6a VSO |
5723 | QLIST_FOREACH(c, &top->parents, next_parent) { |
5724 | updated_children = g_slist_prepend(updated_children, c); | |
5725 | } | |
5726 | ||
3108a15c VSO |
5727 | /* |
5728 | * It seems correct to pass detach_subchain=true here, but it triggers | |
5729 | * one more yet not fixed bug, when due to nested aio_poll loop we switch to | |
5730 | * another drained section, which modify the graph (for example, removing | |
5731 | * the child, which we keep in updated_children list). So, it's a TODO. | |
5732 | * | |
5733 | * Note, bug triggered if pass detach_subchain=true here and run | |
5734 | * test-bdrv-drain. test_drop_intermediate_poll() test-case will crash. | |
5735 | * That's a FIXME. | |
5736 | */ | |
5737 | bdrv_replace_node_common(top, base, false, false, &local_err); | |
d669ed6a VSO |
5738 | if (local_err) { |
5739 | error_report_err(local_err); | |
5740 | goto exit; | |
5741 | } | |
5742 | ||
5743 | for (p = updated_children; p; p = p->next) { | |
5744 | c = p->data; | |
12fa4af6 | 5745 | |
bd86fb99 HR |
5746 | if (c->klass->update_filename) { |
5747 | ret = c->klass->update_filename(c, base, backing_file_str, | |
5748 | &local_err); | |
61f09cea | 5749 | if (ret < 0) { |
d669ed6a VSO |
5750 | /* |
5751 | * TODO: Actually, we want to rollback all previous iterations | |
5752 | * of this loop, and (which is almost impossible) previous | |
5753 | * bdrv_replace_node()... | |
5754 | * | |
5755 | * Note, that c->klass->update_filename may lead to permission | |
5756 | * update, so it's a bad idea to call it inside permission | |
5757 | * update transaction of bdrv_replace_node. | |
5758 | */ | |
61f09cea KW |
5759 | error_report_err(local_err); |
5760 | goto exit; | |
5761 | } | |
5762 | } | |
12fa4af6 | 5763 | } |
6ebdcee2 | 5764 | |
6bd858b3 AG |
5765 | if (update_inherits_from) { |
5766 | base->inherits_from = explicit_top->inherits_from; | |
5767 | } | |
5768 | ||
6ebdcee2 | 5769 | ret = 0; |
6ebdcee2 | 5770 | exit: |
631086de | 5771 | bdrv_drained_end(base); |
6858eba0 | 5772 | bdrv_unref(top); |
6ebdcee2 JC |
5773 | return ret; |
5774 | } | |
5775 | ||
081e4650 | 5776 | /** |
82618d7b | 5777 | * Implementation of BlockDriver.bdrv_co_get_allocated_file_size() that |
081e4650 HR |
5778 | * sums the size of all data-bearing children. (This excludes backing |
5779 | * children.) | |
5780 | */ | |
de335638 EGE |
5781 | static int64_t coroutine_fn GRAPH_RDLOCK |
5782 | bdrv_sum_allocated_file_size(BlockDriverState *bs) | |
081e4650 HR |
5783 | { |
5784 | BdrvChild *child; | |
5785 | int64_t child_size, sum = 0; | |
5786 | ||
5787 | QLIST_FOREACH(child, &bs->children, next) { | |
5788 | if (child->role & (BDRV_CHILD_DATA | BDRV_CHILD_METADATA | | |
5789 | BDRV_CHILD_FILTERED)) | |
5790 | { | |
82618d7b | 5791 | child_size = bdrv_co_get_allocated_file_size(child->bs); |
081e4650 HR |
5792 | if (child_size < 0) { |
5793 | return child_size; | |
5794 | } | |
5795 | sum += child_size; | |
5796 | } | |
5797 | } | |
5798 | ||
5799 | return sum; | |
5800 | } | |
5801 | ||
61007b31 SH |
5802 | /** |
5803 | * Length of a allocated file in bytes. Sparse files are counted by actual | |
5804 | * allocated space. Return < 0 if error or unknown. | |
5805 | */ | |
82618d7b | 5806 | int64_t coroutine_fn bdrv_co_get_allocated_file_size(BlockDriverState *bs) |
71d0770c | 5807 | { |
61007b31 | 5808 | BlockDriver *drv = bs->drv; |
384a48fb | 5809 | IO_CODE(); |
de335638 | 5810 | assert_bdrv_graph_readable(); |
384a48fb | 5811 | |
61007b31 SH |
5812 | if (!drv) { |
5813 | return -ENOMEDIUM; | |
8f4754ed | 5814 | } |
82618d7b EGE |
5815 | if (drv->bdrv_co_get_allocated_file_size) { |
5816 | return drv->bdrv_co_get_allocated_file_size(bs); | |
61007b31 | 5817 | } |
081e4650 HR |
5818 | |
5819 | if (drv->bdrv_file_open) { | |
5820 | /* | |
5821 | * Protocol drivers default to -ENOTSUP (most of their data is | |
5822 | * not stored in any of their children (if they even have any), | |
5823 | * so there is no generic way to figure it out). | |
5824 | */ | |
5825 | return -ENOTSUP; | |
5826 | } else if (drv->is_filter) { | |
5827 | /* Filter drivers default to the size of their filtered child */ | |
82618d7b | 5828 | return bdrv_co_get_allocated_file_size(bdrv_filter_bs(bs)); |
081e4650 HR |
5829 | } else { |
5830 | /* Other drivers default to summing their children's sizes */ | |
5831 | return bdrv_sum_allocated_file_size(bs); | |
1c9805a3 SH |
5832 | } |
5833 | } | |
e7a8a783 | 5834 | |
90880ff1 SH |
5835 | /* |
5836 | * bdrv_measure: | |
5837 | * @drv: Format driver | |
5838 | * @opts: Creation options for new image | |
5839 | * @in_bs: Existing image containing data for new image (may be NULL) | |
5840 | * @errp: Error object | |
5841 | * Returns: A #BlockMeasureInfo (free using qapi_free_BlockMeasureInfo()) | |
5842 | * or NULL on error | |
5843 | * | |
5844 | * Calculate file size required to create a new image. | |
5845 | * | |
5846 | * If @in_bs is given then space for allocated clusters and zero clusters | |
5847 | * from that image are included in the calculation. If @opts contains a | |
5848 | * backing file that is shared by @in_bs then backing clusters may be omitted | |
5849 | * from the calculation. | |
5850 | * | |
5851 | * If @in_bs is NULL then the calculation includes no allocated clusters | |
5852 | * unless a preallocation option is given in @opts. | |
5853 | * | |
5854 | * Note that @in_bs may use a different BlockDriver from @drv. | |
5855 | * | |
5856 | * If an error occurs the @errp pointer is set. | |
5857 | */ | |
5858 | BlockMeasureInfo *bdrv_measure(BlockDriver *drv, QemuOpts *opts, | |
5859 | BlockDriverState *in_bs, Error **errp) | |
5860 | { | |
384a48fb | 5861 | IO_CODE(); |
90880ff1 SH |
5862 | if (!drv->bdrv_measure) { |
5863 | error_setg(errp, "Block driver '%s' does not support size measurement", | |
5864 | drv->format_name); | |
5865 | return NULL; | |
5866 | } | |
5867 | ||
5868 | return drv->bdrv_measure(opts, in_bs, errp); | |
5869 | } | |
5870 | ||
61007b31 SH |
5871 | /** |
5872 | * Return number of sectors on success, -errno on error. | |
1c9805a3 | 5873 | */ |
c86422c5 | 5874 | int64_t coroutine_fn bdrv_co_nb_sectors(BlockDriverState *bs) |
1c9805a3 | 5875 | { |
61007b31 | 5876 | BlockDriver *drv = bs->drv; |
384a48fb | 5877 | IO_CODE(); |
8ab8140a | 5878 | assert_bdrv_graph_readable(); |
498e386c | 5879 | |
61007b31 SH |
5880 | if (!drv) |
5881 | return -ENOMEDIUM; | |
2572b37a | 5882 | |
160a29e2 | 5883 | if (bs->bl.has_variable_length) { |
c86422c5 | 5884 | int ret = bdrv_co_refresh_total_sectors(bs, bs->total_sectors); |
61007b31 SH |
5885 | if (ret < 0) { |
5886 | return ret; | |
1c9805a3 SH |
5887 | } |
5888 | } | |
61007b31 | 5889 | return bs->total_sectors; |
1c9805a3 | 5890 | } |
b338082b | 5891 | |
81f730d4 PB |
5892 | /* |
5893 | * This wrapper is written by hand because this function is in the hot I/O path, | |
5894 | * via blk_get_geometry. | |
5895 | */ | |
5896 | int64_t coroutine_mixed_fn bdrv_nb_sectors(BlockDriverState *bs) | |
5897 | { | |
5898 | BlockDriver *drv = bs->drv; | |
5899 | IO_CODE(); | |
5900 | ||
5901 | if (!drv) | |
5902 | return -ENOMEDIUM; | |
5903 | ||
5904 | if (bs->bl.has_variable_length) { | |
5905 | int ret = bdrv_refresh_total_sectors(bs, bs->total_sectors); | |
5906 | if (ret < 0) { | |
5907 | return ret; | |
5908 | } | |
5909 | } | |
5910 | ||
5911 | return bs->total_sectors; | |
5912 | } | |
5913 | ||
61007b31 SH |
5914 | /** |
5915 | * Return length in bytes on success, -errno on error. | |
5916 | * The length is always a multiple of BDRV_SECTOR_SIZE. | |
8d3b1a2d | 5917 | */ |
c86422c5 | 5918 | int64_t coroutine_fn bdrv_co_getlength(BlockDriverState *bs) |
8d3b1a2d | 5919 | { |
c86422c5 | 5920 | int64_t ret; |
384a48fb | 5921 | IO_CODE(); |
8ab8140a | 5922 | assert_bdrv_graph_readable(); |
8d3b1a2d | 5923 | |
c86422c5 | 5924 | ret = bdrv_co_nb_sectors(bs); |
122860ba EB |
5925 | if (ret < 0) { |
5926 | return ret; | |
5927 | } | |
5928 | if (ret > INT64_MAX / BDRV_SECTOR_SIZE) { | |
5929 | return -EFBIG; | |
5930 | } | |
5931 | return ret * BDRV_SECTOR_SIZE; | |
fc01f7e7 FB |
5932 | } |
5933 | ||
54115412 | 5934 | bool bdrv_is_sg(BlockDriverState *bs) |
f08145fe | 5935 | { |
384a48fb | 5936 | IO_CODE(); |
61007b31 | 5937 | return bs->sg; |
f08145fe KW |
5938 | } |
5939 | ||
ae23f786 HR |
5940 | /** |
5941 | * Return whether the given node supports compressed writes. | |
5942 | */ | |
5943 | bool bdrv_supports_compressed_writes(BlockDriverState *bs) | |
5944 | { | |
5945 | BlockDriverState *filtered; | |
384a48fb | 5946 | IO_CODE(); |
ae23f786 HR |
5947 | |
5948 | if (!bs->drv || !block_driver_can_compress(bs->drv)) { | |
5949 | return false; | |
5950 | } | |
5951 | ||
5952 | filtered = bdrv_filter_bs(bs); | |
5953 | if (filtered) { | |
5954 | /* | |
5955 | * Filters can only forward compressed writes, so we have to | |
5956 | * check the child. | |
5957 | */ | |
5958 | return bdrv_supports_compressed_writes(filtered); | |
5959 | } | |
5960 | ||
5961 | return true; | |
5962 | } | |
5963 | ||
61007b31 | 5964 | const char *bdrv_get_format_name(BlockDriverState *bs) |
40b4f539 | 5965 | { |
384a48fb | 5966 | IO_CODE(); |
61007b31 | 5967 | return bs->drv ? bs->drv->format_name : NULL; |
40b4f539 KW |
5968 | } |
5969 | ||
61007b31 | 5970 | static int qsort_strcmp(const void *a, const void *b) |
40b4f539 | 5971 | { |
ceff5bd7 | 5972 | return strcmp(*(char *const *)a, *(char *const *)b); |
40b4f539 KW |
5973 | } |
5974 | ||
61007b31 | 5975 | void bdrv_iterate_format(void (*it)(void *opaque, const char *name), |
9ac404c5 | 5976 | void *opaque, bool read_only) |
40b4f539 | 5977 | { |
61007b31 SH |
5978 | BlockDriver *drv; |
5979 | int count = 0; | |
5980 | int i; | |
5981 | const char **formats = NULL; | |
40b4f539 | 5982 | |
f791bf7f EGE |
5983 | GLOBAL_STATE_CODE(); |
5984 | ||
61007b31 SH |
5985 | QLIST_FOREACH(drv, &bdrv_drivers, list) { |
5986 | if (drv->format_name) { | |
5987 | bool found = false; | |
5988 | int i = count; | |
9ac404c5 AS |
5989 | |
5990 | if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv, read_only)) { | |
5991 | continue; | |
5992 | } | |
5993 | ||
61007b31 SH |
5994 | while (formats && i && !found) { |
5995 | found = !strcmp(formats[--i], drv->format_name); | |
5996 | } | |
e2a305fb | 5997 | |
61007b31 SH |
5998 | if (!found) { |
5999 | formats = g_renew(const char *, formats, count + 1); | |
6000 | formats[count++] = drv->format_name; | |
6001 | } | |
6c5a42ac | 6002 | } |
61007b31 | 6003 | } |
6c5a42ac | 6004 | |
eb0df69f HR |
6005 | for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); i++) { |
6006 | const char *format_name = block_driver_modules[i].format_name; | |
6007 | ||
6008 | if (format_name) { | |
6009 | bool found = false; | |
6010 | int j = count; | |
6011 | ||
9ac404c5 AS |
6012 | if (use_bdrv_whitelist && |
6013 | !bdrv_format_is_whitelisted(format_name, read_only)) { | |
6014 | continue; | |
6015 | } | |
6016 | ||
eb0df69f HR |
6017 | while (formats && j && !found) { |
6018 | found = !strcmp(formats[--j], format_name); | |
6019 | } | |
6020 | ||
6021 | if (!found) { | |
6022 | formats = g_renew(const char *, formats, count + 1); | |
6023 | formats[count++] = format_name; | |
6024 | } | |
6025 | } | |
6026 | } | |
6027 | ||
61007b31 | 6028 | qsort(formats, count, sizeof(formats[0]), qsort_strcmp); |
40b4f539 | 6029 | |
61007b31 SH |
6030 | for (i = 0; i < count; i++) { |
6031 | it(opaque, formats[i]); | |
6032 | } | |
40b4f539 | 6033 | |
61007b31 SH |
6034 | g_free(formats); |
6035 | } | |
40b4f539 | 6036 | |
61007b31 SH |
6037 | /* This function is to find a node in the bs graph */ |
6038 | BlockDriverState *bdrv_find_node(const char *node_name) | |
6039 | { | |
6040 | BlockDriverState *bs; | |
391827eb | 6041 | |
61007b31 | 6042 | assert(node_name); |
f791bf7f | 6043 | GLOBAL_STATE_CODE(); |
40b4f539 | 6044 | |
61007b31 SH |
6045 | QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) { |
6046 | if (!strcmp(node_name, bs->node_name)) { | |
6047 | return bs; | |
40b4f539 KW |
6048 | } |
6049 | } | |
61007b31 | 6050 | return NULL; |
40b4f539 KW |
6051 | } |
6052 | ||
61007b31 | 6053 | /* Put this QMP function here so it can access the static graph_bdrv_states. */ |
facda544 PK |
6054 | BlockDeviceInfoList *bdrv_named_nodes_list(bool flat, |
6055 | Error **errp) | |
40b4f539 | 6056 | { |
9812e712 | 6057 | BlockDeviceInfoList *list; |
61007b31 | 6058 | BlockDriverState *bs; |
40b4f539 | 6059 | |
f791bf7f EGE |
6060 | GLOBAL_STATE_CODE(); |
6061 | ||
61007b31 SH |
6062 | list = NULL; |
6063 | QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) { | |
facda544 | 6064 | BlockDeviceInfo *info = bdrv_block_device_info(NULL, bs, flat, errp); |
61007b31 SH |
6065 | if (!info) { |
6066 | qapi_free_BlockDeviceInfoList(list); | |
6067 | return NULL; | |
301db7c2 | 6068 | } |
9812e712 | 6069 | QAPI_LIST_PREPEND(list, info); |
301db7c2 RH |
6070 | } |
6071 | ||
61007b31 SH |
6072 | return list; |
6073 | } | |
40b4f539 | 6074 | |
5d3b4e99 VSO |
6075 | typedef struct XDbgBlockGraphConstructor { |
6076 | XDbgBlockGraph *graph; | |
6077 | GHashTable *graph_nodes; | |
6078 | } XDbgBlockGraphConstructor; | |
6079 | ||
6080 | static XDbgBlockGraphConstructor *xdbg_graph_new(void) | |
6081 | { | |
6082 | XDbgBlockGraphConstructor *gr = g_new(XDbgBlockGraphConstructor, 1); | |
6083 | ||
6084 | gr->graph = g_new0(XDbgBlockGraph, 1); | |
6085 | gr->graph_nodes = g_hash_table_new(NULL, NULL); | |
6086 | ||
6087 | return gr; | |
6088 | } | |
6089 | ||
6090 | static XDbgBlockGraph *xdbg_graph_finalize(XDbgBlockGraphConstructor *gr) | |
6091 | { | |
6092 | XDbgBlockGraph *graph = gr->graph; | |
6093 | ||
6094 | g_hash_table_destroy(gr->graph_nodes); | |
6095 | g_free(gr); | |
6096 | ||
6097 | return graph; | |
6098 | } | |
6099 | ||
6100 | static uintptr_t xdbg_graph_node_num(XDbgBlockGraphConstructor *gr, void *node) | |
6101 | { | |
6102 | uintptr_t ret = (uintptr_t)g_hash_table_lookup(gr->graph_nodes, node); | |
6103 | ||
6104 | if (ret != 0) { | |
6105 | return ret; | |
6106 | } | |
6107 | ||
6108 | /* | |
6109 | * Start counting from 1, not 0, because 0 interferes with not-found (NULL) | |
6110 | * answer of g_hash_table_lookup. | |
6111 | */ | |
6112 | ret = g_hash_table_size(gr->graph_nodes) + 1; | |
6113 | g_hash_table_insert(gr->graph_nodes, node, (void *)ret); | |
6114 | ||
6115 | return ret; | |
6116 | } | |
6117 | ||
6118 | static void xdbg_graph_add_node(XDbgBlockGraphConstructor *gr, void *node, | |
6119 | XDbgBlockGraphNodeType type, const char *name) | |
6120 | { | |
6121 | XDbgBlockGraphNode *n; | |
6122 | ||
6123 | n = g_new0(XDbgBlockGraphNode, 1); | |
6124 | ||
6125 | n->id = xdbg_graph_node_num(gr, node); | |
6126 | n->type = type; | |
6127 | n->name = g_strdup(name); | |
6128 | ||
9812e712 | 6129 | QAPI_LIST_PREPEND(gr->graph->nodes, n); |
5d3b4e99 VSO |
6130 | } |
6131 | ||
6132 | static void xdbg_graph_add_edge(XDbgBlockGraphConstructor *gr, void *parent, | |
6133 | const BdrvChild *child) | |
6134 | { | |
cdb1cec8 | 6135 | BlockPermission qapi_perm; |
5d3b4e99 | 6136 | XDbgBlockGraphEdge *edge; |
862fded9 | 6137 | GLOBAL_STATE_CODE(); |
5d3b4e99 | 6138 | |
5d3b4e99 VSO |
6139 | edge = g_new0(XDbgBlockGraphEdge, 1); |
6140 | ||
6141 | edge->parent = xdbg_graph_node_num(gr, parent); | |
6142 | edge->child = xdbg_graph_node_num(gr, child->bs); | |
6143 | edge->name = g_strdup(child->name); | |
6144 | ||
cdb1cec8 HR |
6145 | for (qapi_perm = 0; qapi_perm < BLOCK_PERMISSION__MAX; qapi_perm++) { |
6146 | uint64_t flag = bdrv_qapi_perm_to_blk_perm(qapi_perm); | |
6147 | ||
6148 | if (flag & child->perm) { | |
9812e712 | 6149 | QAPI_LIST_PREPEND(edge->perm, qapi_perm); |
5d3b4e99 | 6150 | } |
cdb1cec8 | 6151 | if (flag & child->shared_perm) { |
9812e712 | 6152 | QAPI_LIST_PREPEND(edge->shared_perm, qapi_perm); |
5d3b4e99 VSO |
6153 | } |
6154 | } | |
6155 | ||
9812e712 | 6156 | QAPI_LIST_PREPEND(gr->graph->edges, edge); |
5d3b4e99 VSO |
6157 | } |
6158 | ||
6159 | ||
6160 | XDbgBlockGraph *bdrv_get_xdbg_block_graph(Error **errp) | |
6161 | { | |
6162 | BlockBackend *blk; | |
6163 | BlockJob *job; | |
6164 | BlockDriverState *bs; | |
6165 | BdrvChild *child; | |
6166 | XDbgBlockGraphConstructor *gr = xdbg_graph_new(); | |
6167 | ||
f791bf7f EGE |
6168 | GLOBAL_STATE_CODE(); |
6169 | ||
5d3b4e99 VSO |
6170 | for (blk = blk_all_next(NULL); blk; blk = blk_all_next(blk)) { |
6171 | char *allocated_name = NULL; | |
6172 | const char *name = blk_name(blk); | |
6173 | ||
6174 | if (!*name) { | |
6175 | name = allocated_name = blk_get_attached_dev_id(blk); | |
6176 | } | |
6177 | xdbg_graph_add_node(gr, blk, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_BACKEND, | |
6178 | name); | |
6179 | g_free(allocated_name); | |
6180 | if (blk_root(blk)) { | |
6181 | xdbg_graph_add_edge(gr, blk, blk_root(blk)); | |
6182 | } | |
6183 | } | |
6184 | ||
880eeec6 EGE |
6185 | WITH_JOB_LOCK_GUARD() { |
6186 | for (job = block_job_next_locked(NULL); job; | |
6187 | job = block_job_next_locked(job)) { | |
6188 | GSList *el; | |
5d3b4e99 | 6189 | |
880eeec6 EGE |
6190 | xdbg_graph_add_node(gr, job, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_JOB, |
6191 | job->job.id); | |
6192 | for (el = job->nodes; el; el = el->next) { | |
6193 | xdbg_graph_add_edge(gr, job, (BdrvChild *)el->data); | |
6194 | } | |
5d3b4e99 VSO |
6195 | } |
6196 | } | |
6197 | ||
6198 | QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) { | |
6199 | xdbg_graph_add_node(gr, bs, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_DRIVER, | |
6200 | bs->node_name); | |
6201 | QLIST_FOREACH(child, &bs->children, next) { | |
6202 | xdbg_graph_add_edge(gr, bs, child); | |
6203 | } | |
6204 | } | |
6205 | ||
6206 | return xdbg_graph_finalize(gr); | |
6207 | } | |
6208 | ||
61007b31 SH |
6209 | BlockDriverState *bdrv_lookup_bs(const char *device, |
6210 | const char *node_name, | |
6211 | Error **errp) | |
6212 | { | |
6213 | BlockBackend *blk; | |
6214 | BlockDriverState *bs; | |
40b4f539 | 6215 | |
f791bf7f EGE |
6216 | GLOBAL_STATE_CODE(); |
6217 | ||
61007b31 SH |
6218 | if (device) { |
6219 | blk = blk_by_name(device); | |
40b4f539 | 6220 | |
61007b31 | 6221 | if (blk) { |
9f4ed6fb AG |
6222 | bs = blk_bs(blk); |
6223 | if (!bs) { | |
5433c24f | 6224 | error_setg(errp, "Device '%s' has no medium", device); |
5433c24f HR |
6225 | } |
6226 | ||
9f4ed6fb | 6227 | return bs; |
61007b31 SH |
6228 | } |
6229 | } | |
40b4f539 | 6230 | |
61007b31 SH |
6231 | if (node_name) { |
6232 | bs = bdrv_find_node(node_name); | |
6d519a5f | 6233 | |
61007b31 SH |
6234 | if (bs) { |
6235 | return bs; | |
6236 | } | |
40b4f539 KW |
6237 | } |
6238 | ||
785ec4b1 | 6239 | error_setg(errp, "Cannot find device=\'%s\' nor node-name=\'%s\'", |
61007b31 SH |
6240 | device ? device : "", |
6241 | node_name ? node_name : ""); | |
6242 | return NULL; | |
40b4f539 KW |
6243 | } |
6244 | ||
61007b31 SH |
6245 | /* If 'base' is in the same chain as 'top', return true. Otherwise, |
6246 | * return false. If either argument is NULL, return false. */ | |
6247 | bool bdrv_chain_contains(BlockDriverState *top, BlockDriverState *base) | |
83f64091 | 6248 | { |
f791bf7f EGE |
6249 | |
6250 | GLOBAL_STATE_CODE(); | |
6251 | ||
61007b31 | 6252 | while (top && top != base) { |
dcf3f9b2 | 6253 | top = bdrv_filter_or_cow_bs(top); |
02c50efe | 6254 | } |
61007b31 SH |
6255 | |
6256 | return top != NULL; | |
02c50efe FZ |
6257 | } |
6258 | ||
61007b31 | 6259 | BlockDriverState *bdrv_next_node(BlockDriverState *bs) |
02c50efe | 6260 | { |
f791bf7f | 6261 | GLOBAL_STATE_CODE(); |
61007b31 SH |
6262 | if (!bs) { |
6263 | return QTAILQ_FIRST(&graph_bdrv_states); | |
02c50efe | 6264 | } |
61007b31 | 6265 | return QTAILQ_NEXT(bs, node_list); |
83f64091 FB |
6266 | } |
6267 | ||
0f12264e KW |
6268 | BlockDriverState *bdrv_next_all_states(BlockDriverState *bs) |
6269 | { | |
f791bf7f | 6270 | GLOBAL_STATE_CODE(); |
0f12264e KW |
6271 | if (!bs) { |
6272 | return QTAILQ_FIRST(&all_bdrv_states); | |
6273 | } | |
6274 | return QTAILQ_NEXT(bs, bs_list); | |
6275 | } | |
6276 | ||
61007b31 | 6277 | const char *bdrv_get_node_name(const BlockDriverState *bs) |
83f64091 | 6278 | { |
384a48fb | 6279 | IO_CODE(); |
61007b31 | 6280 | return bs->node_name; |
beac80cd FB |
6281 | } |
6282 | ||
1f0c461b | 6283 | const char *bdrv_get_parent_name(const BlockDriverState *bs) |
4c265bf9 KW |
6284 | { |
6285 | BdrvChild *c; | |
6286 | const char *name; | |
967d7905 | 6287 | IO_CODE(); |
4c265bf9 KW |
6288 | |
6289 | /* If multiple parents have a name, just pick the first one. */ | |
6290 | QLIST_FOREACH(c, &bs->parents, next_parent) { | |
bd86fb99 HR |
6291 | if (c->klass->get_name) { |
6292 | name = c->klass->get_name(c); | |
4c265bf9 KW |
6293 | if (name && *name) { |
6294 | return name; | |
6295 | } | |
6296 | } | |
6297 | } | |
6298 | ||
6299 | return NULL; | |
6300 | } | |
6301 | ||
61007b31 SH |
6302 | /* TODO check what callers really want: bs->node_name or blk_name() */ |
6303 | const char *bdrv_get_device_name(const BlockDriverState *bs) | |
beac80cd | 6304 | { |
384a48fb | 6305 | IO_CODE(); |
4c265bf9 | 6306 | return bdrv_get_parent_name(bs) ?: ""; |
f141eafe | 6307 | } |
83f64091 | 6308 | |
61007b31 SH |
6309 | /* This can be used to identify nodes that might not have a device |
6310 | * name associated. Since node and device names live in the same | |
6311 | * namespace, the result is unambiguous. The exception is if both are | |
6312 | * absent, then this returns an empty (non-null) string. */ | |
6313 | const char *bdrv_get_device_or_node_name(const BlockDriverState *bs) | |
f141eafe | 6314 | { |
384a48fb | 6315 | IO_CODE(); |
4c265bf9 | 6316 | return bdrv_get_parent_name(bs) ?: bs->node_name; |
beac80cd | 6317 | } |
beac80cd | 6318 | |
61007b31 | 6319 | int bdrv_get_flags(BlockDriverState *bs) |
0b5a2445 | 6320 | { |
15aee7ac | 6321 | IO_CODE(); |
61007b31 | 6322 | return bs->open_flags; |
0b5a2445 PB |
6323 | } |
6324 | ||
61007b31 | 6325 | int bdrv_has_zero_init_1(BlockDriverState *bs) |
68485420 | 6326 | { |
f791bf7f | 6327 | GLOBAL_STATE_CODE(); |
61007b31 | 6328 | return 1; |
0b5a2445 PB |
6329 | } |
6330 | ||
61007b31 | 6331 | int bdrv_has_zero_init(BlockDriverState *bs) |
0b5a2445 | 6332 | { |
93393e69 | 6333 | BlockDriverState *filtered; |
f791bf7f | 6334 | GLOBAL_STATE_CODE(); |
93393e69 | 6335 | |
d470ad42 HR |
6336 | if (!bs->drv) { |
6337 | return 0; | |
6338 | } | |
0b5a2445 | 6339 | |
61007b31 SH |
6340 | /* If BS is a copy on write image, it is initialized to |
6341 | the contents of the base image, which may not be zeroes. */ | |
34778172 | 6342 | if (bdrv_cow_child(bs)) { |
61007b31 SH |
6343 | return 0; |
6344 | } | |
6345 | if (bs->drv->bdrv_has_zero_init) { | |
6346 | return bs->drv->bdrv_has_zero_init(bs); | |
0b5a2445 | 6347 | } |
93393e69 HR |
6348 | |
6349 | filtered = bdrv_filter_bs(bs); | |
6350 | if (filtered) { | |
6351 | return bdrv_has_zero_init(filtered); | |
5a612c00 | 6352 | } |
61007b31 SH |
6353 | |
6354 | /* safe default */ | |
6355 | return 0; | |
68485420 KW |
6356 | } |
6357 | ||
61007b31 | 6358 | bool bdrv_can_write_zeroes_with_unmap(BlockDriverState *bs) |
68485420 | 6359 | { |
384a48fb | 6360 | IO_CODE(); |
2f0342ef | 6361 | if (!(bs->open_flags & BDRV_O_UNMAP)) { |
61007b31 SH |
6362 | return false; |
6363 | } | |
68485420 | 6364 | |
e24d813b | 6365 | return bs->supported_zero_flags & BDRV_REQ_MAY_UNMAP; |
68485420 KW |
6366 | } |
6367 | ||
61007b31 SH |
6368 | void bdrv_get_backing_filename(BlockDriverState *bs, |
6369 | char *filename, int filename_size) | |
016f5cf6 | 6370 | { |
384a48fb | 6371 | IO_CODE(); |
61007b31 SH |
6372 | pstrcpy(filename, filename_size, bs->backing_file); |
6373 | } | |
d318aea9 | 6374 | |
3d47eb0a | 6375 | int coroutine_fn bdrv_co_get_info(BlockDriverState *bs, BlockDriverInfo *bdi) |
61007b31 | 6376 | { |
8b117001 | 6377 | int ret; |
61007b31 | 6378 | BlockDriver *drv = bs->drv; |
384a48fb | 6379 | IO_CODE(); |
a00e70c0 EGE |
6380 | assert_bdrv_graph_readable(); |
6381 | ||
5a612c00 MP |
6382 | /* if bs->drv == NULL, bs is closed, so there's nothing to do here */ |
6383 | if (!drv) { | |
61007b31 | 6384 | return -ENOMEDIUM; |
5a612c00 | 6385 | } |
3d47eb0a | 6386 | if (!drv->bdrv_co_get_info) { |
93393e69 HR |
6387 | BlockDriverState *filtered = bdrv_filter_bs(bs); |
6388 | if (filtered) { | |
3d47eb0a | 6389 | return bdrv_co_get_info(filtered, bdi); |
5a612c00 | 6390 | } |
61007b31 | 6391 | return -ENOTSUP; |
5a612c00 | 6392 | } |
61007b31 | 6393 | memset(bdi, 0, sizeof(*bdi)); |
3d47eb0a | 6394 | ret = drv->bdrv_co_get_info(bs, bdi); |
8b117001 VSO |
6395 | if (ret < 0) { |
6396 | return ret; | |
6397 | } | |
6398 | ||
6399 | if (bdi->cluster_size > BDRV_MAX_ALIGNMENT) { | |
6400 | return -EINVAL; | |
6401 | } | |
6402 | ||
6403 | return 0; | |
61007b31 | 6404 | } |
016f5cf6 | 6405 | |
1bf6e9ca AS |
6406 | ImageInfoSpecific *bdrv_get_specific_info(BlockDriverState *bs, |
6407 | Error **errp) | |
61007b31 SH |
6408 | { |
6409 | BlockDriver *drv = bs->drv; | |
384a48fb | 6410 | IO_CODE(); |
61007b31 | 6411 | if (drv && drv->bdrv_get_specific_info) { |
1bf6e9ca | 6412 | return drv->bdrv_get_specific_info(bs, errp); |
61007b31 SH |
6413 | } |
6414 | return NULL; | |
016f5cf6 AG |
6415 | } |
6416 | ||
d9245599 AN |
6417 | BlockStatsSpecific *bdrv_get_specific_stats(BlockDriverState *bs) |
6418 | { | |
6419 | BlockDriver *drv = bs->drv; | |
384a48fb | 6420 | IO_CODE(); |
d9245599 AN |
6421 | if (!drv || !drv->bdrv_get_specific_stats) { |
6422 | return NULL; | |
6423 | } | |
6424 | return drv->bdrv_get_specific_stats(bs); | |
6425 | } | |
6426 | ||
c834dc05 | 6427 | void coroutine_fn bdrv_co_debug_event(BlockDriverState *bs, BlkdebugEvent event) |
4265d620 | 6428 | { |
384a48fb | 6429 | IO_CODE(); |
cb2bfaa4 EGE |
6430 | assert_bdrv_graph_readable(); |
6431 | ||
c834dc05 | 6432 | if (!bs || !bs->drv || !bs->drv->bdrv_co_debug_event) { |
61007b31 SH |
6433 | return; |
6434 | } | |
4265d620 | 6435 | |
c834dc05 | 6436 | bs->drv->bdrv_co_debug_event(bs, event); |
4265d620 PB |
6437 | } |
6438 | ||
d10529a2 | 6439 | static BlockDriverState *bdrv_find_debug_node(BlockDriverState *bs) |
4265d620 | 6440 | { |
bdb73476 | 6441 | GLOBAL_STATE_CODE(); |
61007b31 | 6442 | while (bs && bs->drv && !bs->drv->bdrv_debug_breakpoint) { |
f706a92f | 6443 | bs = bdrv_primary_bs(bs); |
61007b31 | 6444 | } |
4265d620 | 6445 | |
61007b31 | 6446 | if (bs && bs->drv && bs->drv->bdrv_debug_breakpoint) { |
d10529a2 VSO |
6447 | assert(bs->drv->bdrv_debug_remove_breakpoint); |
6448 | return bs; | |
6449 | } | |
6450 | ||
6451 | return NULL; | |
6452 | } | |
6453 | ||
6454 | int bdrv_debug_breakpoint(BlockDriverState *bs, const char *event, | |
6455 | const char *tag) | |
6456 | { | |
f791bf7f | 6457 | GLOBAL_STATE_CODE(); |
d10529a2 VSO |
6458 | bs = bdrv_find_debug_node(bs); |
6459 | if (bs) { | |
61007b31 SH |
6460 | return bs->drv->bdrv_debug_breakpoint(bs, event, tag); |
6461 | } | |
4265d620 | 6462 | |
61007b31 | 6463 | return -ENOTSUP; |
4265d620 PB |
6464 | } |
6465 | ||
61007b31 | 6466 | int bdrv_debug_remove_breakpoint(BlockDriverState *bs, const char *tag) |
ea2384d3 | 6467 | { |
f791bf7f | 6468 | GLOBAL_STATE_CODE(); |
d10529a2 VSO |
6469 | bs = bdrv_find_debug_node(bs); |
6470 | if (bs) { | |
61007b31 SH |
6471 | return bs->drv->bdrv_debug_remove_breakpoint(bs, tag); |
6472 | } | |
6473 | ||
6474 | return -ENOTSUP; | |
eb852011 MA |
6475 | } |
6476 | ||
61007b31 | 6477 | int bdrv_debug_resume(BlockDriverState *bs, const char *tag) |
ce1a14dc | 6478 | { |
f791bf7f | 6479 | GLOBAL_STATE_CODE(); |
61007b31 | 6480 | while (bs && (!bs->drv || !bs->drv->bdrv_debug_resume)) { |
f706a92f | 6481 | bs = bdrv_primary_bs(bs); |
61007b31 | 6482 | } |
ce1a14dc | 6483 | |
61007b31 SH |
6484 | if (bs && bs->drv && bs->drv->bdrv_debug_resume) { |
6485 | return bs->drv->bdrv_debug_resume(bs, tag); | |
6486 | } | |
ce1a14dc | 6487 | |
61007b31 | 6488 | return -ENOTSUP; |
f197fe2b FZ |
6489 | } |
6490 | ||
61007b31 | 6491 | bool bdrv_debug_is_suspended(BlockDriverState *bs, const char *tag) |
ce1a14dc | 6492 | { |
f791bf7f | 6493 | GLOBAL_STATE_CODE(); |
61007b31 | 6494 | while (bs && bs->drv && !bs->drv->bdrv_debug_is_suspended) { |
f706a92f | 6495 | bs = bdrv_primary_bs(bs); |
f197fe2b | 6496 | } |
19cb3738 | 6497 | |
61007b31 SH |
6498 | if (bs && bs->drv && bs->drv->bdrv_debug_is_suspended) { |
6499 | return bs->drv->bdrv_debug_is_suspended(bs, tag); | |
6500 | } | |
f9f05dc5 | 6501 | |
61007b31 SH |
6502 | return false; |
6503 | } | |
f9f05dc5 | 6504 | |
61007b31 SH |
6505 | /* backing_file can either be relative, or absolute, or a protocol. If it is |
6506 | * relative, it must be relative to the chain. So, passing in bs->filename | |
6507 | * from a BDS as backing_file should not be done, as that may be relative to | |
6508 | * the CWD rather than the chain. */ | |
6509 | BlockDriverState *bdrv_find_backing_image(BlockDriverState *bs, | |
6510 | const char *backing_file) | |
f9f05dc5 | 6511 | { |
61007b31 SH |
6512 | char *filename_full = NULL; |
6513 | char *backing_file_full = NULL; | |
6514 | char *filename_tmp = NULL; | |
6515 | int is_protocol = 0; | |
0b877d09 | 6516 | bool filenames_refreshed = false; |
61007b31 SH |
6517 | BlockDriverState *curr_bs = NULL; |
6518 | BlockDriverState *retval = NULL; | |
dcf3f9b2 | 6519 | BlockDriverState *bs_below; |
f9f05dc5 | 6520 | |
f791bf7f EGE |
6521 | GLOBAL_STATE_CODE(); |
6522 | ||
61007b31 SH |
6523 | if (!bs || !bs->drv || !backing_file) { |
6524 | return NULL; | |
f9f05dc5 KW |
6525 | } |
6526 | ||
61007b31 SH |
6527 | filename_full = g_malloc(PATH_MAX); |
6528 | backing_file_full = g_malloc(PATH_MAX); | |
f9f05dc5 | 6529 | |
61007b31 | 6530 | is_protocol = path_has_protocol(backing_file); |
f9f05dc5 | 6531 | |
dcf3f9b2 HR |
6532 | /* |
6533 | * Being largely a legacy function, skip any filters here | |
6534 | * (because filters do not have normal filenames, so they cannot | |
6535 | * match anyway; and allowing json:{} filenames is a bit out of | |
6536 | * scope). | |
6537 | */ | |
6538 | for (curr_bs = bdrv_skip_filters(bs); | |
6539 | bdrv_cow_child(curr_bs) != NULL; | |
6540 | curr_bs = bs_below) | |
6541 | { | |
6542 | bs_below = bdrv_backing_chain_next(curr_bs); | |
f9f05dc5 | 6543 | |
0b877d09 HR |
6544 | if (bdrv_backing_overridden(curr_bs)) { |
6545 | /* | |
6546 | * If the backing file was overridden, we can only compare | |
6547 | * directly against the backing node's filename. | |
6548 | */ | |
6549 | ||
6550 | if (!filenames_refreshed) { | |
6551 | /* | |
6552 | * This will automatically refresh all of the | |
6553 | * filenames in the rest of the backing chain, so we | |
6554 | * only need to do this once. | |
6555 | */ | |
6556 | bdrv_refresh_filename(bs_below); | |
6557 | filenames_refreshed = true; | |
6558 | } | |
6559 | ||
6560 | if (strcmp(backing_file, bs_below->filename) == 0) { | |
6561 | retval = bs_below; | |
6562 | break; | |
6563 | } | |
6564 | } else if (is_protocol || path_has_protocol(curr_bs->backing_file)) { | |
6565 | /* | |
6566 | * If either of the filename paths is actually a protocol, then | |
6567 | * compare unmodified paths; otherwise make paths relative. | |
6568 | */ | |
6b6833c1 HR |
6569 | char *backing_file_full_ret; |
6570 | ||
61007b31 | 6571 | if (strcmp(backing_file, curr_bs->backing_file) == 0) { |
dcf3f9b2 | 6572 | retval = bs_below; |
61007b31 SH |
6573 | break; |
6574 | } | |
418661e0 | 6575 | /* Also check against the full backing filename for the image */ |
6b6833c1 HR |
6576 | backing_file_full_ret = bdrv_get_full_backing_filename(curr_bs, |
6577 | NULL); | |
6578 | if (backing_file_full_ret) { | |
6579 | bool equal = strcmp(backing_file, backing_file_full_ret) == 0; | |
6580 | g_free(backing_file_full_ret); | |
6581 | if (equal) { | |
dcf3f9b2 | 6582 | retval = bs_below; |
418661e0 JC |
6583 | break; |
6584 | } | |
418661e0 | 6585 | } |
61007b31 SH |
6586 | } else { |
6587 | /* If not an absolute filename path, make it relative to the current | |
6588 | * image's filename path */ | |
2d9158ce HR |
6589 | filename_tmp = bdrv_make_absolute_filename(curr_bs, backing_file, |
6590 | NULL); | |
6591 | /* We are going to compare canonicalized absolute pathnames */ | |
6592 | if (!filename_tmp || !realpath(filename_tmp, filename_full)) { | |
6593 | g_free(filename_tmp); | |
61007b31 SH |
6594 | continue; |
6595 | } | |
2d9158ce | 6596 | g_free(filename_tmp); |
07f07615 | 6597 | |
61007b31 SH |
6598 | /* We need to make sure the backing filename we are comparing against |
6599 | * is relative to the current image filename (or absolute) */ | |
2d9158ce HR |
6600 | filename_tmp = bdrv_get_full_backing_filename(curr_bs, NULL); |
6601 | if (!filename_tmp || !realpath(filename_tmp, backing_file_full)) { | |
6602 | g_free(filename_tmp); | |
61007b31 SH |
6603 | continue; |
6604 | } | |
2d9158ce | 6605 | g_free(filename_tmp); |
eb489bb1 | 6606 | |
61007b31 | 6607 | if (strcmp(backing_file_full, filename_full) == 0) { |
dcf3f9b2 | 6608 | retval = bs_below; |
61007b31 SH |
6609 | break; |
6610 | } | |
6611 | } | |
eb489bb1 KW |
6612 | } |
6613 | ||
61007b31 SH |
6614 | g_free(filename_full); |
6615 | g_free(backing_file_full); | |
61007b31 SH |
6616 | return retval; |
6617 | } | |
6618 | ||
61007b31 SH |
6619 | void bdrv_init(void) |
6620 | { | |
e5f05f8c KW |
6621 | #ifdef CONFIG_BDRV_WHITELIST_TOOLS |
6622 | use_bdrv_whitelist = 1; | |
6623 | #endif | |
61007b31 SH |
6624 | module_call_init(MODULE_INIT_BLOCK); |
6625 | } | |
29cdb251 | 6626 | |
61007b31 SH |
6627 | void bdrv_init_with_whitelist(void) |
6628 | { | |
6629 | use_bdrv_whitelist = 1; | |
6630 | bdrv_init(); | |
07f07615 PB |
6631 | } |
6632 | ||
a94750d9 | 6633 | int bdrv_activate(BlockDriverState *bs, Error **errp) |
0f15423c | 6634 | { |
4417ab7a | 6635 | BdrvChild *child, *parent; |
5a8a30db KW |
6636 | Error *local_err = NULL; |
6637 | int ret; | |
9c98f145 | 6638 | BdrvDirtyBitmap *bm; |
5a8a30db | 6639 | |
f791bf7f EGE |
6640 | GLOBAL_STATE_CODE(); |
6641 | ||
3456a8d1 | 6642 | if (!bs->drv) { |
5416645f | 6643 | return -ENOMEDIUM; |
3456a8d1 KW |
6644 | } |
6645 | ||
16e977d5 | 6646 | QLIST_FOREACH(child, &bs->children, next) { |
11d0c9b3 | 6647 | bdrv_activate(child->bs, &local_err); |
0d1c5c91 | 6648 | if (local_err) { |
0d1c5c91 | 6649 | error_propagate(errp, local_err); |
5416645f | 6650 | return -EINVAL; |
0d1c5c91 | 6651 | } |
5a8a30db | 6652 | } |
0d1c5c91 | 6653 | |
dafe0960 KW |
6654 | /* |
6655 | * Update permissions, they may differ for inactive nodes. | |
6656 | * | |
6657 | * Note that the required permissions of inactive images are always a | |
6658 | * subset of the permissions required after activating the image. This | |
6659 | * allows us to just get the permissions upfront without restricting | |
11d0c9b3 | 6660 | * bdrv_co_invalidate_cache(). |
dafe0960 KW |
6661 | * |
6662 | * It also means that in error cases, we don't have to try and revert to | |
6663 | * the old permissions (which is an operation that could fail, too). We can | |
6664 | * just keep the extended permissions for the next time that an activation | |
6665 | * of the image is tried. | |
6666 | */ | |
7bb4941a KW |
6667 | if (bs->open_flags & BDRV_O_INACTIVE) { |
6668 | bs->open_flags &= ~BDRV_O_INACTIVE; | |
f1316edb | 6669 | ret = bdrv_refresh_perms(bs, NULL, errp); |
7bb4941a | 6670 | if (ret < 0) { |
0d1c5c91 | 6671 | bs->open_flags |= BDRV_O_INACTIVE; |
5416645f | 6672 | return ret; |
0d1c5c91 | 6673 | } |
3456a8d1 | 6674 | |
11d0c9b3 EGE |
6675 | ret = bdrv_invalidate_cache(bs, errp); |
6676 | if (ret < 0) { | |
6677 | bs->open_flags |= BDRV_O_INACTIVE; | |
6678 | return ret; | |
7bb4941a | 6679 | } |
9c98f145 | 6680 | |
7bb4941a KW |
6681 | FOR_EACH_DIRTY_BITMAP(bs, bm) { |
6682 | bdrv_dirty_bitmap_skip_store(bm, false); | |
6683 | } | |
6684 | ||
c057960c | 6685 | ret = bdrv_refresh_total_sectors(bs, bs->total_sectors); |
7bb4941a KW |
6686 | if (ret < 0) { |
6687 | bs->open_flags |= BDRV_O_INACTIVE; | |
6688 | error_setg_errno(errp, -ret, "Could not refresh total sector count"); | |
5416645f | 6689 | return ret; |
7bb4941a | 6690 | } |
5a8a30db | 6691 | } |
4417ab7a KW |
6692 | |
6693 | QLIST_FOREACH(parent, &bs->parents, next_parent) { | |
bd86fb99 HR |
6694 | if (parent->klass->activate) { |
6695 | parent->klass->activate(parent, &local_err); | |
4417ab7a | 6696 | if (local_err) { |
78fc3b3a | 6697 | bs->open_flags |= BDRV_O_INACTIVE; |
4417ab7a | 6698 | error_propagate(errp, local_err); |
5416645f | 6699 | return -EINVAL; |
4417ab7a KW |
6700 | } |
6701 | } | |
6702 | } | |
5416645f VSO |
6703 | |
6704 | return 0; | |
0f15423c AL |
6705 | } |
6706 | ||
11d0c9b3 EGE |
6707 | int coroutine_fn bdrv_co_invalidate_cache(BlockDriverState *bs, Error **errp) |
6708 | { | |
6709 | Error *local_err = NULL; | |
1581a70d | 6710 | IO_CODE(); |
11d0c9b3 EGE |
6711 | |
6712 | assert(!(bs->open_flags & BDRV_O_INACTIVE)); | |
1b3ff9fe | 6713 | assert_bdrv_graph_readable(); |
11d0c9b3 EGE |
6714 | |
6715 | if (bs->drv->bdrv_co_invalidate_cache) { | |
6716 | bs->drv->bdrv_co_invalidate_cache(bs, &local_err); | |
6717 | if (local_err) { | |
6718 | error_propagate(errp, local_err); | |
6719 | return -EINVAL; | |
6720 | } | |
6721 | } | |
6722 | ||
6723 | return 0; | |
6724 | } | |
6725 | ||
3b717194 | 6726 | void bdrv_activate_all(Error **errp) |
0f15423c | 6727 | { |
7c8eece4 | 6728 | BlockDriverState *bs; |
88be7b4b | 6729 | BdrvNextIterator it; |
0f15423c | 6730 | |
f791bf7f EGE |
6731 | GLOBAL_STATE_CODE(); |
6732 | ||
88be7b4b | 6733 | for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) { |
ed78cda3 | 6734 | AioContext *aio_context = bdrv_get_aio_context(bs); |
5416645f | 6735 | int ret; |
ed78cda3 SH |
6736 | |
6737 | aio_context_acquire(aio_context); | |
a94750d9 | 6738 | ret = bdrv_activate(bs, errp); |
ed78cda3 | 6739 | aio_context_release(aio_context); |
5416645f | 6740 | if (ret < 0) { |
5e003f17 | 6741 | bdrv_next_cleanup(&it); |
5a8a30db KW |
6742 | return; |
6743 | } | |
0f15423c AL |
6744 | } |
6745 | } | |
6746 | ||
9e37271f KW |
6747 | static bool bdrv_has_bds_parent(BlockDriverState *bs, bool only_active) |
6748 | { | |
6749 | BdrvChild *parent; | |
bdb73476 | 6750 | GLOBAL_STATE_CODE(); |
9e37271f KW |
6751 | |
6752 | QLIST_FOREACH(parent, &bs->parents, next_parent) { | |
bd86fb99 | 6753 | if (parent->klass->parent_is_bds) { |
9e37271f KW |
6754 | BlockDriverState *parent_bs = parent->opaque; |
6755 | if (!only_active || !(parent_bs->open_flags & BDRV_O_INACTIVE)) { | |
6756 | return true; | |
6757 | } | |
6758 | } | |
6759 | } | |
6760 | ||
6761 | return false; | |
6762 | } | |
6763 | ||
6764 | static int bdrv_inactivate_recurse(BlockDriverState *bs) | |
76b1c7fe | 6765 | { |
cfa1a572 | 6766 | BdrvChild *child, *parent; |
76b1c7fe | 6767 | int ret; |
a13de40a | 6768 | uint64_t cumulative_perms, cumulative_shared_perms; |
76b1c7fe | 6769 | |
da359909 EGE |
6770 | GLOBAL_STATE_CODE(); |
6771 | ||
d470ad42 HR |
6772 | if (!bs->drv) { |
6773 | return -ENOMEDIUM; | |
6774 | } | |
6775 | ||
9e37271f KW |
6776 | /* Make sure that we don't inactivate a child before its parent. |
6777 | * It will be covered by recursion from the yet active parent. */ | |
6778 | if (bdrv_has_bds_parent(bs, true)) { | |
6779 | return 0; | |
6780 | } | |
6781 | ||
6782 | assert(!(bs->open_flags & BDRV_O_INACTIVE)); | |
6783 | ||
6784 | /* Inactivate this node */ | |
6785 | if (bs->drv->bdrv_inactivate) { | |
76b1c7fe KW |
6786 | ret = bs->drv->bdrv_inactivate(bs); |
6787 | if (ret < 0) { | |
6788 | return ret; | |
6789 | } | |
6790 | } | |
6791 | ||
9e37271f | 6792 | QLIST_FOREACH(parent, &bs->parents, next_parent) { |
bd86fb99 HR |
6793 | if (parent->klass->inactivate) { |
6794 | ret = parent->klass->inactivate(parent); | |
9e37271f KW |
6795 | if (ret < 0) { |
6796 | return ret; | |
cfa1a572 KW |
6797 | } |
6798 | } | |
9e37271f | 6799 | } |
9c5e6594 | 6800 | |
a13de40a VSO |
6801 | bdrv_get_cumulative_perm(bs, &cumulative_perms, |
6802 | &cumulative_shared_perms); | |
6803 | if (cumulative_perms & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)) { | |
6804 | /* Our inactive parents still need write access. Inactivation failed. */ | |
6805 | return -EPERM; | |
6806 | } | |
6807 | ||
9e37271f | 6808 | bs->open_flags |= BDRV_O_INACTIVE; |
7d5b5261 | 6809 | |
bb87e4d1 VSO |
6810 | /* |
6811 | * Update permissions, they may differ for inactive nodes. | |
6812 | * We only tried to loosen restrictions, so errors are not fatal, ignore | |
6813 | * them. | |
6814 | */ | |
f1316edb | 6815 | bdrv_refresh_perms(bs, NULL, NULL); |
9e37271f KW |
6816 | |
6817 | /* Recursively inactivate children */ | |
38701b6a | 6818 | QLIST_FOREACH(child, &bs->children, next) { |
9e37271f | 6819 | ret = bdrv_inactivate_recurse(child->bs); |
38701b6a KW |
6820 | if (ret < 0) { |
6821 | return ret; | |
6822 | } | |
6823 | } | |
6824 | ||
76b1c7fe KW |
6825 | return 0; |
6826 | } | |
6827 | ||
6828 | int bdrv_inactivate_all(void) | |
6829 | { | |
79720af6 | 6830 | BlockDriverState *bs = NULL; |
88be7b4b | 6831 | BdrvNextIterator it; |
aad0b7a0 | 6832 | int ret = 0; |
bd6458e4 | 6833 | GSList *aio_ctxs = NULL, *ctx; |
76b1c7fe | 6834 | |
f791bf7f EGE |
6835 | GLOBAL_STATE_CODE(); |
6836 | ||
88be7b4b | 6837 | for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) { |
bd6458e4 PB |
6838 | AioContext *aio_context = bdrv_get_aio_context(bs); |
6839 | ||
6840 | if (!g_slist_find(aio_ctxs, aio_context)) { | |
6841 | aio_ctxs = g_slist_prepend(aio_ctxs, aio_context); | |
6842 | aio_context_acquire(aio_context); | |
6843 | } | |
aad0b7a0 | 6844 | } |
76b1c7fe | 6845 | |
9e37271f KW |
6846 | for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) { |
6847 | /* Nodes with BDS parents are covered by recursion from the last | |
6848 | * parent that gets inactivated. Don't inactivate them a second | |
6849 | * time if that has already happened. */ | |
6850 | if (bdrv_has_bds_parent(bs, false)) { | |
6851 | continue; | |
6852 | } | |
6853 | ret = bdrv_inactivate_recurse(bs); | |
6854 | if (ret < 0) { | |
6855 | bdrv_next_cleanup(&it); | |
6856 | goto out; | |
76b1c7fe KW |
6857 | } |
6858 | } | |
6859 | ||
aad0b7a0 | 6860 | out: |
bd6458e4 PB |
6861 | for (ctx = aio_ctxs; ctx != NULL; ctx = ctx->next) { |
6862 | AioContext *aio_context = ctx->data; | |
6863 | aio_context_release(aio_context); | |
aad0b7a0 | 6864 | } |
bd6458e4 | 6865 | g_slist_free(aio_ctxs); |
aad0b7a0 FZ |
6866 | |
6867 | return ret; | |
76b1c7fe KW |
6868 | } |
6869 | ||
19cb3738 FB |
6870 | /**************************************************************/ |
6871 | /* removable device support */ | |
6872 | ||
6873 | /** | |
6874 | * Return TRUE if the media is present | |
6875 | */ | |
1e97be91 | 6876 | bool coroutine_fn bdrv_co_is_inserted(BlockDriverState *bs) |
19cb3738 FB |
6877 | { |
6878 | BlockDriver *drv = bs->drv; | |
28d7a789 | 6879 | BdrvChild *child; |
384a48fb | 6880 | IO_CODE(); |
c73ff92c | 6881 | assert_bdrv_graph_readable(); |
a1aff5bf | 6882 | |
e031f750 HR |
6883 | if (!drv) { |
6884 | return false; | |
6885 | } | |
1e97be91 EGE |
6886 | if (drv->bdrv_co_is_inserted) { |
6887 | return drv->bdrv_co_is_inserted(bs); | |
28d7a789 HR |
6888 | } |
6889 | QLIST_FOREACH(child, &bs->children, next) { | |
1e97be91 | 6890 | if (!bdrv_co_is_inserted(child->bs)) { |
28d7a789 HR |
6891 | return false; |
6892 | } | |
e031f750 | 6893 | } |
28d7a789 | 6894 | return true; |
19cb3738 FB |
6895 | } |
6896 | ||
19cb3738 FB |
6897 | /** |
6898 | * If eject_flag is TRUE, eject the media. Otherwise, close the tray | |
6899 | */ | |
2531b390 | 6900 | void coroutine_fn bdrv_co_eject(BlockDriverState *bs, bool eject_flag) |
19cb3738 FB |
6901 | { |
6902 | BlockDriver *drv = bs->drv; | |
384a48fb | 6903 | IO_CODE(); |
79a292e5 | 6904 | assert_bdrv_graph_readable(); |
19cb3738 | 6905 | |
2531b390 EGE |
6906 | if (drv && drv->bdrv_co_eject) { |
6907 | drv->bdrv_co_eject(bs, eject_flag); | |
19cb3738 FB |
6908 | } |
6909 | } | |
6910 | ||
19cb3738 FB |
6911 | /** |
6912 | * Lock or unlock the media (if it is locked, the user won't be able | |
6913 | * to eject it manually). | |
6914 | */ | |
2c75261c | 6915 | void coroutine_fn bdrv_co_lock_medium(BlockDriverState *bs, bool locked) |
19cb3738 FB |
6916 | { |
6917 | BlockDriver *drv = bs->drv; | |
384a48fb | 6918 | IO_CODE(); |
79a292e5 | 6919 | assert_bdrv_graph_readable(); |
025e849a | 6920 | trace_bdrv_lock_medium(bs, locked); |
b8c6d095 | 6921 | |
2c75261c EGE |
6922 | if (drv && drv->bdrv_co_lock_medium) { |
6923 | drv->bdrv_co_lock_medium(bs, locked); | |
19cb3738 FB |
6924 | } |
6925 | } | |
985a03b0 | 6926 | |
9fcb0251 FZ |
6927 | /* Get a reference to bs */ |
6928 | void bdrv_ref(BlockDriverState *bs) | |
6929 | { | |
f791bf7f | 6930 | GLOBAL_STATE_CODE(); |
9fcb0251 FZ |
6931 | bs->refcnt++; |
6932 | } | |
6933 | ||
6934 | /* Release a previously grabbed reference to bs. | |
6935 | * If after releasing, reference count is zero, the BlockDriverState is | |
6936 | * deleted. */ | |
6937 | void bdrv_unref(BlockDriverState *bs) | |
6938 | { | |
f791bf7f | 6939 | GLOBAL_STATE_CODE(); |
9a4d5ca6 JC |
6940 | if (!bs) { |
6941 | return; | |
6942 | } | |
9fcb0251 FZ |
6943 | assert(bs->refcnt > 0); |
6944 | if (--bs->refcnt == 0) { | |
6945 | bdrv_delete(bs); | |
6946 | } | |
6947 | } | |
6948 | ||
fbe40ff7 FZ |
6949 | struct BdrvOpBlocker { |
6950 | Error *reason; | |
6951 | QLIST_ENTRY(BdrvOpBlocker) list; | |
6952 | }; | |
6953 | ||
6954 | bool bdrv_op_is_blocked(BlockDriverState *bs, BlockOpType op, Error **errp) | |
6955 | { | |
6956 | BdrvOpBlocker *blocker; | |
f791bf7f | 6957 | GLOBAL_STATE_CODE(); |
fbe40ff7 FZ |
6958 | assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX); |
6959 | if (!QLIST_EMPTY(&bs->op_blockers[op])) { | |
6960 | blocker = QLIST_FIRST(&bs->op_blockers[op]); | |
4b576648 MA |
6961 | error_propagate_prepend(errp, error_copy(blocker->reason), |
6962 | "Node '%s' is busy: ", | |
6963 | bdrv_get_device_or_node_name(bs)); | |
fbe40ff7 FZ |
6964 | return true; |
6965 | } | |
6966 | return false; | |
6967 | } | |
6968 | ||
6969 | void bdrv_op_block(BlockDriverState *bs, BlockOpType op, Error *reason) | |
6970 | { | |
6971 | BdrvOpBlocker *blocker; | |
f791bf7f | 6972 | GLOBAL_STATE_CODE(); |
fbe40ff7 FZ |
6973 | assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX); |
6974 | ||
5839e53b | 6975 | blocker = g_new0(BdrvOpBlocker, 1); |
fbe40ff7 FZ |
6976 | blocker->reason = reason; |
6977 | QLIST_INSERT_HEAD(&bs->op_blockers[op], blocker, list); | |
6978 | } | |
6979 | ||
6980 | void bdrv_op_unblock(BlockDriverState *bs, BlockOpType op, Error *reason) | |
6981 | { | |
6982 | BdrvOpBlocker *blocker, *next; | |
f791bf7f | 6983 | GLOBAL_STATE_CODE(); |
fbe40ff7 FZ |
6984 | assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX); |
6985 | QLIST_FOREACH_SAFE(blocker, &bs->op_blockers[op], list, next) { | |
6986 | if (blocker->reason == reason) { | |
6987 | QLIST_REMOVE(blocker, list); | |
6988 | g_free(blocker); | |
6989 | } | |
6990 | } | |
6991 | } | |
6992 | ||
6993 | void bdrv_op_block_all(BlockDriverState *bs, Error *reason) | |
6994 | { | |
6995 | int i; | |
f791bf7f | 6996 | GLOBAL_STATE_CODE(); |
fbe40ff7 FZ |
6997 | for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) { |
6998 | bdrv_op_block(bs, i, reason); | |
6999 | } | |
7000 | } | |
7001 | ||
7002 | void bdrv_op_unblock_all(BlockDriverState *bs, Error *reason) | |
7003 | { | |
7004 | int i; | |
f791bf7f | 7005 | GLOBAL_STATE_CODE(); |
fbe40ff7 FZ |
7006 | for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) { |
7007 | bdrv_op_unblock(bs, i, reason); | |
7008 | } | |
7009 | } | |
7010 | ||
7011 | bool bdrv_op_blocker_is_empty(BlockDriverState *bs) | |
7012 | { | |
7013 | int i; | |
f791bf7f | 7014 | GLOBAL_STATE_CODE(); |
fbe40ff7 FZ |
7015 | for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) { |
7016 | if (!QLIST_EMPTY(&bs->op_blockers[i])) { | |
7017 | return false; | |
7018 | } | |
7019 | } | |
7020 | return true; | |
7021 | } | |
7022 | ||
e1355055 KW |
7023 | /* |
7024 | * Must not be called while holding the lock of an AioContext other than the | |
7025 | * current one. | |
7026 | */ | |
d92ada22 LC |
7027 | void bdrv_img_create(const char *filename, const char *fmt, |
7028 | const char *base_filename, const char *base_fmt, | |
9217283d FZ |
7029 | char *options, uint64_t img_size, int flags, bool quiet, |
7030 | Error **errp) | |
f88e1a42 | 7031 | { |
83d0521a CL |
7032 | QemuOptsList *create_opts = NULL; |
7033 | QemuOpts *opts = NULL; | |
7034 | const char *backing_fmt, *backing_file; | |
7035 | int64_t size; | |
f88e1a42 | 7036 | BlockDriver *drv, *proto_drv; |
cc84d90f | 7037 | Error *local_err = NULL; |
f88e1a42 JS |
7038 | int ret = 0; |
7039 | ||
f791bf7f EGE |
7040 | GLOBAL_STATE_CODE(); |
7041 | ||
f88e1a42 JS |
7042 | /* Find driver and parse its options */ |
7043 | drv = bdrv_find_format(fmt); | |
7044 | if (!drv) { | |
71c79813 | 7045 | error_setg(errp, "Unknown file format '%s'", fmt); |
d92ada22 | 7046 | return; |
f88e1a42 JS |
7047 | } |
7048 | ||
b65a5e12 | 7049 | proto_drv = bdrv_find_protocol(filename, true, errp); |
f88e1a42 | 7050 | if (!proto_drv) { |
d92ada22 | 7051 | return; |
f88e1a42 JS |
7052 | } |
7053 | ||
c6149724 HR |
7054 | if (!drv->create_opts) { |
7055 | error_setg(errp, "Format driver '%s' does not support image creation", | |
7056 | drv->format_name); | |
7057 | return; | |
7058 | } | |
7059 | ||
5a5e7f8c ML |
7060 | if (!proto_drv->create_opts) { |
7061 | error_setg(errp, "Protocol driver '%s' does not support image creation", | |
7062 | proto_drv->format_name); | |
7063 | return; | |
7064 | } | |
7065 | ||
c6e0a6de KW |
7066 | aio_context_acquire(qemu_get_aio_context()); |
7067 | ||
f6dc1c31 | 7068 | /* Create parameter list */ |
c282e1fd | 7069 | create_opts = qemu_opts_append(create_opts, drv->create_opts); |
5a5e7f8c | 7070 | create_opts = qemu_opts_append(create_opts, proto_drv->create_opts); |
f88e1a42 | 7071 | |
83d0521a | 7072 | opts = qemu_opts_create(create_opts, NULL, 0, &error_abort); |
f88e1a42 JS |
7073 | |
7074 | /* Parse -o options */ | |
7075 | if (options) { | |
a5f9b9df | 7076 | if (!qemu_opts_do_parse(opts, options, NULL, errp)) { |
f88e1a42 JS |
7077 | goto out; |
7078 | } | |
7079 | } | |
7080 | ||
f6dc1c31 KW |
7081 | if (!qemu_opt_get(opts, BLOCK_OPT_SIZE)) { |
7082 | qemu_opt_set_number(opts, BLOCK_OPT_SIZE, img_size, &error_abort); | |
7083 | } else if (img_size != UINT64_C(-1)) { | |
7084 | error_setg(errp, "The image size must be specified only once"); | |
7085 | goto out; | |
7086 | } | |
7087 | ||
f88e1a42 | 7088 | if (base_filename) { |
235e59cf | 7089 | if (!qemu_opt_set(opts, BLOCK_OPT_BACKING_FILE, base_filename, |
3882578b | 7090 | NULL)) { |
71c79813 LC |
7091 | error_setg(errp, "Backing file not supported for file format '%s'", |
7092 | fmt); | |
f88e1a42 JS |
7093 | goto out; |
7094 | } | |
7095 | } | |
7096 | ||
7097 | if (base_fmt) { | |
3882578b | 7098 | if (!qemu_opt_set(opts, BLOCK_OPT_BACKING_FMT, base_fmt, NULL)) { |
71c79813 LC |
7099 | error_setg(errp, "Backing file format not supported for file " |
7100 | "format '%s'", fmt); | |
f88e1a42 JS |
7101 | goto out; |
7102 | } | |
7103 | } | |
7104 | ||
83d0521a CL |
7105 | backing_file = qemu_opt_get(opts, BLOCK_OPT_BACKING_FILE); |
7106 | if (backing_file) { | |
7107 | if (!strcmp(filename, backing_file)) { | |
71c79813 LC |
7108 | error_setg(errp, "Error: Trying to create an image with the " |
7109 | "same filename as the backing file"); | |
792da93a JS |
7110 | goto out; |
7111 | } | |
975a7bd2 CK |
7112 | if (backing_file[0] == '\0') { |
7113 | error_setg(errp, "Expected backing file name, got empty string"); | |
7114 | goto out; | |
7115 | } | |
792da93a JS |
7116 | } |
7117 | ||
83d0521a | 7118 | backing_fmt = qemu_opt_get(opts, BLOCK_OPT_BACKING_FMT); |
f88e1a42 | 7119 | |
6e6e55f5 JS |
7120 | /* The size for the image must always be specified, unless we have a backing |
7121 | * file and we have not been forbidden from opening it. */ | |
a8b42a1c | 7122 | size = qemu_opt_get_size(opts, BLOCK_OPT_SIZE, img_size); |
6e6e55f5 JS |
7123 | if (backing_file && !(flags & BDRV_O_NO_BACKING)) { |
7124 | BlockDriverState *bs; | |
645ae7d8 | 7125 | char *full_backing; |
6e6e55f5 JS |
7126 | int back_flags; |
7127 | QDict *backing_options = NULL; | |
7128 | ||
645ae7d8 HR |
7129 | full_backing = |
7130 | bdrv_get_full_backing_filename_from_filename(filename, backing_file, | |
7131 | &local_err); | |
6e6e55f5 | 7132 | if (local_err) { |
6e6e55f5 JS |
7133 | goto out; |
7134 | } | |
645ae7d8 | 7135 | assert(full_backing); |
29168018 | 7136 | |
d5b23994 HR |
7137 | /* |
7138 | * No need to do I/O here, which allows us to open encrypted | |
7139 | * backing images without needing the secret | |
7140 | */ | |
6e6e55f5 JS |
7141 | back_flags = flags; |
7142 | back_flags &= ~(BDRV_O_RDWR | BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING); | |
d5b23994 | 7143 | back_flags |= BDRV_O_NO_IO; |
f88e1a42 | 7144 | |
cc954f01 | 7145 | backing_options = qdict_new(); |
6e6e55f5 | 7146 | if (backing_fmt) { |
6e6e55f5 JS |
7147 | qdict_put_str(backing_options, "driver", backing_fmt); |
7148 | } | |
cc954f01 | 7149 | qdict_put_bool(backing_options, BDRV_OPT_FORCE_SHARE, true); |
e6641719 | 7150 | |
6e6e55f5 JS |
7151 | bs = bdrv_open(full_backing, NULL, backing_options, back_flags, |
7152 | &local_err); | |
7153 | g_free(full_backing); | |
add8200d EB |
7154 | if (!bs) { |
7155 | error_append_hint(&local_err, "Could not open backing image.\n"); | |
6e6e55f5 JS |
7156 | goto out; |
7157 | } else { | |
d9f059aa | 7158 | if (!backing_fmt) { |
497a30db EB |
7159 | error_setg(&local_err, |
7160 | "Backing file specified without backing format"); | |
7161 | error_append_hint(&local_err, "Detected format of %s.", | |
7162 | bs->drv->format_name); | |
7163 | goto out; | |
d9f059aa | 7164 | } |
6e6e55f5 JS |
7165 | if (size == -1) { |
7166 | /* Opened BS, have no size */ | |
7167 | size = bdrv_getlength(bs); | |
7168 | if (size < 0) { | |
7169 | error_setg_errno(errp, -size, "Could not get size of '%s'", | |
7170 | backing_file); | |
7171 | bdrv_unref(bs); | |
7172 | goto out; | |
7173 | } | |
7174 | qemu_opt_set_number(opts, BLOCK_OPT_SIZE, size, &error_abort); | |
52bf1e72 | 7175 | } |
66f6b814 | 7176 | bdrv_unref(bs); |
f88e1a42 | 7177 | } |
d9f059aa EB |
7178 | /* (backing_file && !(flags & BDRV_O_NO_BACKING)) */ |
7179 | } else if (backing_file && !backing_fmt) { | |
497a30db EB |
7180 | error_setg(&local_err, |
7181 | "Backing file specified without backing format"); | |
7182 | goto out; | |
d9f059aa | 7183 | } |
6e6e55f5 JS |
7184 | |
7185 | if (size == -1) { | |
7186 | error_setg(errp, "Image creation needs a size parameter"); | |
7187 | goto out; | |
f88e1a42 JS |
7188 | } |
7189 | ||
f382d43a | 7190 | if (!quiet) { |
fe646693 | 7191 | printf("Formatting '%s', fmt=%s ", filename, fmt); |
43c5d8f8 | 7192 | qemu_opts_print(opts, " "); |
f382d43a | 7193 | puts(""); |
4e2f4418 | 7194 | fflush(stdout); |
f382d43a | 7195 | } |
83d0521a | 7196 | |
c282e1fd | 7197 | ret = bdrv_create(drv, filename, opts, &local_err); |
83d0521a | 7198 | |
cc84d90f HR |
7199 | if (ret == -EFBIG) { |
7200 | /* This is generally a better message than whatever the driver would | |
7201 | * deliver (especially because of the cluster_size_hint), since that | |
7202 | * is most probably not much different from "image too large". */ | |
7203 | const char *cluster_size_hint = ""; | |
83d0521a | 7204 | if (qemu_opt_get_size(opts, BLOCK_OPT_CLUSTER_SIZE, 0)) { |
cc84d90f | 7205 | cluster_size_hint = " (try using a larger cluster size)"; |
f88e1a42 | 7206 | } |
cc84d90f HR |
7207 | error_setg(errp, "The image size is too large for file format '%s'" |
7208 | "%s", fmt, cluster_size_hint); | |
7209 | error_free(local_err); | |
7210 | local_err = NULL; | |
f88e1a42 JS |
7211 | } |
7212 | ||
7213 | out: | |
83d0521a CL |
7214 | qemu_opts_del(opts); |
7215 | qemu_opts_free(create_opts); | |
621ff94d | 7216 | error_propagate(errp, local_err); |
c6e0a6de | 7217 | aio_context_release(qemu_get_aio_context()); |
f88e1a42 | 7218 | } |
85d126f3 SH |
7219 | |
7220 | AioContext *bdrv_get_aio_context(BlockDriverState *bs) | |
7221 | { | |
384a48fb | 7222 | IO_CODE(); |
33f2a757 | 7223 | return bs ? bs->aio_context : qemu_get_aio_context(); |
dcd04228 SH |
7224 | } |
7225 | ||
e336fd4c KW |
7226 | AioContext *coroutine_fn bdrv_co_enter(BlockDriverState *bs) |
7227 | { | |
7228 | Coroutine *self = qemu_coroutine_self(); | |
7229 | AioContext *old_ctx = qemu_coroutine_get_aio_context(self); | |
7230 | AioContext *new_ctx; | |
384a48fb | 7231 | IO_CODE(); |
e336fd4c KW |
7232 | |
7233 | /* | |
7234 | * Increase bs->in_flight to ensure that this operation is completed before | |
7235 | * moving the node to a different AioContext. Read new_ctx only afterwards. | |
7236 | */ | |
7237 | bdrv_inc_in_flight(bs); | |
7238 | ||
7239 | new_ctx = bdrv_get_aio_context(bs); | |
7240 | aio_co_reschedule_self(new_ctx); | |
7241 | return old_ctx; | |
7242 | } | |
7243 | ||
7244 | void coroutine_fn bdrv_co_leave(BlockDriverState *bs, AioContext *old_ctx) | |
7245 | { | |
384a48fb | 7246 | IO_CODE(); |
e336fd4c KW |
7247 | aio_co_reschedule_self(old_ctx); |
7248 | bdrv_dec_in_flight(bs); | |
7249 | } | |
7250 | ||
18c6ac1c KW |
7251 | void coroutine_fn bdrv_co_lock(BlockDriverState *bs) |
7252 | { | |
7253 | AioContext *ctx = bdrv_get_aio_context(bs); | |
7254 | ||
7255 | /* In the main thread, bs->aio_context won't change concurrently */ | |
7256 | assert(qemu_get_current_aio_context() == qemu_get_aio_context()); | |
7257 | ||
7258 | /* | |
7259 | * We're in coroutine context, so we already hold the lock of the main | |
7260 | * loop AioContext. Don't lock it twice to avoid deadlocks. | |
7261 | */ | |
7262 | assert(qemu_in_coroutine()); | |
7263 | if (ctx != qemu_get_aio_context()) { | |
7264 | aio_context_acquire(ctx); | |
7265 | } | |
7266 | } | |
7267 | ||
7268 | void coroutine_fn bdrv_co_unlock(BlockDriverState *bs) | |
7269 | { | |
7270 | AioContext *ctx = bdrv_get_aio_context(bs); | |
7271 | ||
7272 | assert(qemu_in_coroutine()); | |
7273 | if (ctx != qemu_get_aio_context()) { | |
7274 | aio_context_release(ctx); | |
7275 | } | |
7276 | } | |
7277 | ||
e8a095da SH |
7278 | static void bdrv_do_remove_aio_context_notifier(BdrvAioNotifier *ban) |
7279 | { | |
bdb73476 | 7280 | GLOBAL_STATE_CODE(); |
e8a095da SH |
7281 | QLIST_REMOVE(ban, list); |
7282 | g_free(ban); | |
7283 | } | |
7284 | ||
a3a683c3 | 7285 | static void bdrv_detach_aio_context(BlockDriverState *bs) |
dcd04228 | 7286 | { |
e8a095da | 7287 | BdrvAioNotifier *baf, *baf_tmp; |
33384421 | 7288 | |
e8a095da | 7289 | assert(!bs->walking_aio_notifiers); |
da359909 | 7290 | GLOBAL_STATE_CODE(); |
e8a095da SH |
7291 | bs->walking_aio_notifiers = true; |
7292 | QLIST_FOREACH_SAFE(baf, &bs->aio_notifiers, list, baf_tmp) { | |
7293 | if (baf->deleted) { | |
7294 | bdrv_do_remove_aio_context_notifier(baf); | |
7295 | } else { | |
7296 | baf->detach_aio_context(baf->opaque); | |
7297 | } | |
33384421 | 7298 | } |
e8a095da SH |
7299 | /* Never mind iterating again to check for ->deleted. bdrv_close() will |
7300 | * remove remaining aio notifiers if we aren't called again. | |
7301 | */ | |
7302 | bs->walking_aio_notifiers = false; | |
33384421 | 7303 | |
1bffe1ae | 7304 | if (bs->drv && bs->drv->bdrv_detach_aio_context) { |
dcd04228 SH |
7305 | bs->drv->bdrv_detach_aio_context(bs); |
7306 | } | |
dcd04228 SH |
7307 | |
7308 | bs->aio_context = NULL; | |
7309 | } | |
7310 | ||
a3a683c3 KW |
7311 | static void bdrv_attach_aio_context(BlockDriverState *bs, |
7312 | AioContext *new_context) | |
dcd04228 | 7313 | { |
e8a095da | 7314 | BdrvAioNotifier *ban, *ban_tmp; |
da359909 | 7315 | GLOBAL_STATE_CODE(); |
33384421 | 7316 | |
dcd04228 SH |
7317 | bs->aio_context = new_context; |
7318 | ||
1bffe1ae | 7319 | if (bs->drv && bs->drv->bdrv_attach_aio_context) { |
dcd04228 SH |
7320 | bs->drv->bdrv_attach_aio_context(bs, new_context); |
7321 | } | |
33384421 | 7322 | |
e8a095da SH |
7323 | assert(!bs->walking_aio_notifiers); |
7324 | bs->walking_aio_notifiers = true; | |
7325 | QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_tmp) { | |
7326 | if (ban->deleted) { | |
7327 | bdrv_do_remove_aio_context_notifier(ban); | |
7328 | } else { | |
7329 | ban->attached_aio_context(new_context, ban->opaque); | |
7330 | } | |
33384421 | 7331 | } |
e8a095da | 7332 | bs->walking_aio_notifiers = false; |
dcd04228 SH |
7333 | } |
7334 | ||
7e8c182f EGE |
7335 | typedef struct BdrvStateSetAioContext { |
7336 | AioContext *new_ctx; | |
7337 | BlockDriverState *bs; | |
7338 | } BdrvStateSetAioContext; | |
d616b224 | 7339 | |
7e8c182f | 7340 | static bool bdrv_parent_change_aio_context(BdrvChild *c, AioContext *ctx, |
e08cc001 EGE |
7341 | GHashTable *visited, |
7342 | Transaction *tran, | |
7e8c182f | 7343 | Error **errp) |
5d231849 | 7344 | { |
f0c28327 | 7345 | GLOBAL_STATE_CODE(); |
e08cc001 | 7346 | if (g_hash_table_contains(visited, c)) { |
5d231849 KW |
7347 | return true; |
7348 | } | |
e08cc001 | 7349 | g_hash_table_add(visited, c); |
5d231849 | 7350 | |
bd86fb99 HR |
7351 | /* |
7352 | * A BdrvChildClass that doesn't handle AioContext changes cannot | |
7353 | * tolerate any AioContext changes | |
7354 | */ | |
7e8c182f | 7355 | if (!c->klass->change_aio_ctx) { |
5d231849 KW |
7356 | char *user = bdrv_child_user_desc(c); |
7357 | error_setg(errp, "Changing iothreads is not supported by %s", user); | |
7358 | g_free(user); | |
7359 | return false; | |
7360 | } | |
7e8c182f | 7361 | if (!c->klass->change_aio_ctx(c, ctx, visited, tran, errp)) { |
5d231849 KW |
7362 | assert(!errp || *errp); |
7363 | return false; | |
7364 | } | |
7365 | return true; | |
7366 | } | |
7367 | ||
7e8c182f | 7368 | bool bdrv_child_change_aio_context(BdrvChild *c, AioContext *ctx, |
e08cc001 | 7369 | GHashTable *visited, Transaction *tran, |
7e8c182f | 7370 | Error **errp) |
5d231849 | 7371 | { |
f791bf7f | 7372 | GLOBAL_STATE_CODE(); |
e08cc001 | 7373 | if (g_hash_table_contains(visited, c)) { |
5d231849 KW |
7374 | return true; |
7375 | } | |
e08cc001 | 7376 | g_hash_table_add(visited, c); |
7e8c182f | 7377 | return bdrv_change_aio_context(c->bs, ctx, visited, tran, errp); |
5d231849 KW |
7378 | } |
7379 | ||
7e8c182f EGE |
7380 | static void bdrv_set_aio_context_clean(void *opaque) |
7381 | { | |
7382 | BdrvStateSetAioContext *state = (BdrvStateSetAioContext *) opaque; | |
7383 | BlockDriverState *bs = (BlockDriverState *) state->bs; | |
7384 | ||
7385 | /* Paired with bdrv_drained_begin in bdrv_change_aio_context() */ | |
7386 | bdrv_drained_end(bs); | |
7387 | ||
7388 | g_free(state); | |
7389 | } | |
7390 | ||
7391 | static void bdrv_set_aio_context_commit(void *opaque) | |
7392 | { | |
7393 | BdrvStateSetAioContext *state = (BdrvStateSetAioContext *) opaque; | |
7394 | BlockDriverState *bs = (BlockDriverState *) state->bs; | |
7395 | AioContext *new_context = state->new_ctx; | |
7396 | AioContext *old_context = bdrv_get_aio_context(bs); | |
7e8c182f EGE |
7397 | |
7398 | /* | |
7399 | * Take the old AioContex when detaching it from bs. | |
7400 | * At this point, new_context lock is already acquired, and we are now | |
7401 | * also taking old_context. This is safe as long as bdrv_detach_aio_context | |
7402 | * does not call AIO_POLL_WHILE(). | |
7403 | */ | |
7404 | if (old_context != qemu_get_aio_context()) { | |
7405 | aio_context_acquire(old_context); | |
7406 | } | |
7407 | bdrv_detach_aio_context(bs); | |
7408 | if (old_context != qemu_get_aio_context()) { | |
7409 | aio_context_release(old_context); | |
7410 | } | |
7411 | bdrv_attach_aio_context(bs, new_context); | |
7412 | } | |
7413 | ||
7414 | static TransactionActionDrv set_aio_context = { | |
7415 | .commit = bdrv_set_aio_context_commit, | |
7416 | .clean = bdrv_set_aio_context_clean, | |
7417 | }; | |
7418 | ||
7419 | /* | |
7420 | * Changes the AioContext used for fd handlers, timers, and BHs by this | |
7421 | * BlockDriverState and all its children and parents. | |
7422 | * | |
7423 | * Must be called from the main AioContext. | |
7424 | * | |
7425 | * The caller must own the AioContext lock for the old AioContext of bs, but it | |
7426 | * must not own the AioContext lock for new_context (unless new_context is the | |
7427 | * same as the current context of bs). | |
7428 | * | |
7429 | * @visited will accumulate all visited BdrvChild objects. The caller is | |
7430 | * responsible for freeing the list afterwards. | |
7431 | */ | |
7432 | static bool bdrv_change_aio_context(BlockDriverState *bs, AioContext *ctx, | |
e08cc001 | 7433 | GHashTable *visited, Transaction *tran, |
7e8c182f | 7434 | Error **errp) |
5d231849 KW |
7435 | { |
7436 | BdrvChild *c; | |
7e8c182f EGE |
7437 | BdrvStateSetAioContext *state; |
7438 | ||
7439 | GLOBAL_STATE_CODE(); | |
5d231849 KW |
7440 | |
7441 | if (bdrv_get_aio_context(bs) == ctx) { | |
7442 | return true; | |
7443 | } | |
7444 | ||
7445 | QLIST_FOREACH(c, &bs->parents, next_parent) { | |
7e8c182f | 7446 | if (!bdrv_parent_change_aio_context(c, ctx, visited, tran, errp)) { |
5d231849 KW |
7447 | return false; |
7448 | } | |
7449 | } | |
7e8c182f | 7450 | |
5d231849 | 7451 | QLIST_FOREACH(c, &bs->children, next) { |
7e8c182f | 7452 | if (!bdrv_child_change_aio_context(c, ctx, visited, tran, errp)) { |
5d231849 KW |
7453 | return false; |
7454 | } | |
7455 | } | |
7456 | ||
7e8c182f EGE |
7457 | state = g_new(BdrvStateSetAioContext, 1); |
7458 | *state = (BdrvStateSetAioContext) { | |
7459 | .new_ctx = ctx, | |
7460 | .bs = bs, | |
7461 | }; | |
7462 | ||
7463 | /* Paired with bdrv_drained_end in bdrv_set_aio_context_clean() */ | |
7464 | bdrv_drained_begin(bs); | |
7465 | ||
7466 | tran_add(tran, &set_aio_context, state); | |
7467 | ||
5d231849 KW |
7468 | return true; |
7469 | } | |
7470 | ||
7e8c182f EGE |
7471 | /* |
7472 | * Change bs's and recursively all of its parents' and children's AioContext | |
7473 | * to the given new context, returning an error if that isn't possible. | |
7474 | * | |
7475 | * If ignore_child is not NULL, that child (and its subgraph) will not | |
7476 | * be touched. | |
7477 | * | |
7478 | * This function still requires the caller to take the bs current | |
7479 | * AioContext lock, otherwise draining will fail since AIO_WAIT_WHILE | |
7480 | * assumes the lock is always held if bs is in another AioContext. | |
7481 | * For the same reason, it temporarily also holds the new AioContext, since | |
7482 | * bdrv_drained_end calls BDRV_POLL_WHILE that assumes the lock is taken too. | |
7483 | * Therefore the new AioContext lock must not be taken by the caller. | |
7484 | */ | |
a41cfda1 EGE |
7485 | int bdrv_try_change_aio_context(BlockDriverState *bs, AioContext *ctx, |
7486 | BdrvChild *ignore_child, Error **errp) | |
5d231849 | 7487 | { |
7e8c182f | 7488 | Transaction *tran; |
e08cc001 | 7489 | GHashTable *visited; |
7e8c182f EGE |
7490 | int ret; |
7491 | AioContext *old_context = bdrv_get_aio_context(bs); | |
f791bf7f EGE |
7492 | GLOBAL_STATE_CODE(); |
7493 | ||
7e8c182f EGE |
7494 | /* |
7495 | * Recursion phase: go through all nodes of the graph. | |
7496 | * Take care of checking that all nodes support changing AioContext | |
7497 | * and drain them, builing a linear list of callbacks to run if everything | |
7498 | * is successful (the transaction itself). | |
7499 | */ | |
7500 | tran = tran_new(); | |
e08cc001 EGE |
7501 | visited = g_hash_table_new(NULL, NULL); |
7502 | if (ignore_child) { | |
7503 | g_hash_table_add(visited, ignore_child); | |
7504 | } | |
7505 | ret = bdrv_change_aio_context(bs, ctx, visited, tran, errp); | |
7506 | g_hash_table_destroy(visited); | |
7e8c182f EGE |
7507 | |
7508 | /* | |
7509 | * Linear phase: go through all callbacks collected in the transaction. | |
7510 | * Run all callbacks collected in the recursion to switch all nodes | |
7511 | * AioContext lock (transaction commit), or undo all changes done in the | |
7512 | * recursion (transaction abort). | |
7513 | */ | |
5d231849 KW |
7514 | |
7515 | if (!ret) { | |
7e8c182f EGE |
7516 | /* Just run clean() callbacks. No AioContext changed. */ |
7517 | tran_abort(tran); | |
5d231849 KW |
7518 | return -EPERM; |
7519 | } | |
7520 | ||
7e8c182f EGE |
7521 | /* |
7522 | * Release old AioContext, it won't be needed anymore, as all | |
7523 | * bdrv_drained_begin() have been called already. | |
7524 | */ | |
7525 | if (qemu_get_aio_context() != old_context) { | |
7526 | aio_context_release(old_context); | |
7527 | } | |
7528 | ||
7529 | /* | |
7530 | * Acquire new AioContext since bdrv_drained_end() is going to be called | |
7531 | * after we switched all nodes in the new AioContext, and the function | |
7532 | * assumes that the lock of the bs is always taken. | |
7533 | */ | |
7534 | if (qemu_get_aio_context() != ctx) { | |
7535 | aio_context_acquire(ctx); | |
7536 | } | |
53a7d041 | 7537 | |
7e8c182f | 7538 | tran_commit(tran); |
5d231849 | 7539 | |
7e8c182f EGE |
7540 | if (qemu_get_aio_context() != ctx) { |
7541 | aio_context_release(ctx); | |
7542 | } | |
7543 | ||
7544 | /* Re-acquire the old AioContext, since the caller takes and releases it. */ | |
7545 | if (qemu_get_aio_context() != old_context) { | |
7546 | aio_context_acquire(old_context); | |
7547 | } | |
7548 | ||
7549 | return 0; | |
5d231849 KW |
7550 | } |
7551 | ||
33384421 HR |
7552 | void bdrv_add_aio_context_notifier(BlockDriverState *bs, |
7553 | void (*attached_aio_context)(AioContext *new_context, void *opaque), | |
7554 | void (*detach_aio_context)(void *opaque), void *opaque) | |
7555 | { | |
7556 | BdrvAioNotifier *ban = g_new(BdrvAioNotifier, 1); | |
7557 | *ban = (BdrvAioNotifier){ | |
7558 | .attached_aio_context = attached_aio_context, | |
7559 | .detach_aio_context = detach_aio_context, | |
7560 | .opaque = opaque | |
7561 | }; | |
f791bf7f | 7562 | GLOBAL_STATE_CODE(); |
33384421 HR |
7563 | |
7564 | QLIST_INSERT_HEAD(&bs->aio_notifiers, ban, list); | |
7565 | } | |
7566 | ||
7567 | void bdrv_remove_aio_context_notifier(BlockDriverState *bs, | |
7568 | void (*attached_aio_context)(AioContext *, | |
7569 | void *), | |
7570 | void (*detach_aio_context)(void *), | |
7571 | void *opaque) | |
7572 | { | |
7573 | BdrvAioNotifier *ban, *ban_next; | |
f791bf7f | 7574 | GLOBAL_STATE_CODE(); |
33384421 HR |
7575 | |
7576 | QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_next) { | |
7577 | if (ban->attached_aio_context == attached_aio_context && | |
7578 | ban->detach_aio_context == detach_aio_context && | |
e8a095da SH |
7579 | ban->opaque == opaque && |
7580 | ban->deleted == false) | |
33384421 | 7581 | { |
e8a095da SH |
7582 | if (bs->walking_aio_notifiers) { |
7583 | ban->deleted = true; | |
7584 | } else { | |
7585 | bdrv_do_remove_aio_context_notifier(ban); | |
7586 | } | |
33384421 HR |
7587 | return; |
7588 | } | |
7589 | } | |
7590 | ||
7591 | abort(); | |
7592 | } | |
7593 | ||
77485434 | 7594 | int bdrv_amend_options(BlockDriverState *bs, QemuOpts *opts, |
d1402b50 | 7595 | BlockDriverAmendStatusCB *status_cb, void *cb_opaque, |
a3579bfa | 7596 | bool force, |
d1402b50 | 7597 | Error **errp) |
6f176b48 | 7598 | { |
f791bf7f | 7599 | GLOBAL_STATE_CODE(); |
d470ad42 | 7600 | if (!bs->drv) { |
d1402b50 | 7601 | error_setg(errp, "Node is ejected"); |
d470ad42 HR |
7602 | return -ENOMEDIUM; |
7603 | } | |
c282e1fd | 7604 | if (!bs->drv->bdrv_amend_options) { |
d1402b50 HR |
7605 | error_setg(errp, "Block driver '%s' does not support option amendment", |
7606 | bs->drv->format_name); | |
6f176b48 HR |
7607 | return -ENOTSUP; |
7608 | } | |
a3579bfa ML |
7609 | return bs->drv->bdrv_amend_options(bs, opts, status_cb, |
7610 | cb_opaque, force, errp); | |
6f176b48 | 7611 | } |
f6186f49 | 7612 | |
5d69b5ab HR |
7613 | /* |
7614 | * This function checks whether the given @to_replace is allowed to be | |
7615 | * replaced by a node that always shows the same data as @bs. This is | |
7616 | * used for example to verify whether the mirror job can replace | |
7617 | * @to_replace by the target mirrored from @bs. | |
7618 | * To be replaceable, @bs and @to_replace may either be guaranteed to | |
7619 | * always show the same data (because they are only connected through | |
7620 | * filters), or some driver may allow replacing one of its children | |
7621 | * because it can guarantee that this child's data is not visible at | |
7622 | * all (for example, for dissenting quorum children that have no other | |
7623 | * parents). | |
7624 | */ | |
7625 | bool bdrv_recurse_can_replace(BlockDriverState *bs, | |
7626 | BlockDriverState *to_replace) | |
7627 | { | |
93393e69 HR |
7628 | BlockDriverState *filtered; |
7629 | ||
b4ad82aa EGE |
7630 | GLOBAL_STATE_CODE(); |
7631 | ||
5d69b5ab HR |
7632 | if (!bs || !bs->drv) { |
7633 | return false; | |
7634 | } | |
7635 | ||
7636 | if (bs == to_replace) { | |
7637 | return true; | |
7638 | } | |
7639 | ||
7640 | /* See what the driver can do */ | |
7641 | if (bs->drv->bdrv_recurse_can_replace) { | |
7642 | return bs->drv->bdrv_recurse_can_replace(bs, to_replace); | |
7643 | } | |
7644 | ||
7645 | /* For filters without an own implementation, we can recurse on our own */ | |
93393e69 HR |
7646 | filtered = bdrv_filter_bs(bs); |
7647 | if (filtered) { | |
7648 | return bdrv_recurse_can_replace(filtered, to_replace); | |
5d69b5ab HR |
7649 | } |
7650 | ||
7651 | /* Safe default */ | |
7652 | return false; | |
7653 | } | |
7654 | ||
810803a8 HR |
7655 | /* |
7656 | * Check whether the given @node_name can be replaced by a node that | |
7657 | * has the same data as @parent_bs. If so, return @node_name's BDS; | |
7658 | * NULL otherwise. | |
7659 | * | |
7660 | * @node_name must be a (recursive) *child of @parent_bs (or this | |
7661 | * function will return NULL). | |
7662 | * | |
7663 | * The result (whether the node can be replaced or not) is only valid | |
7664 | * for as long as no graph or permission changes occur. | |
7665 | */ | |
e12f3784 WC |
7666 | BlockDriverState *check_to_replace_node(BlockDriverState *parent_bs, |
7667 | const char *node_name, Error **errp) | |
09158f00 BC |
7668 | { |
7669 | BlockDriverState *to_replace_bs = bdrv_find_node(node_name); | |
5a7e7a0b SH |
7670 | AioContext *aio_context; |
7671 | ||
f791bf7f EGE |
7672 | GLOBAL_STATE_CODE(); |
7673 | ||
09158f00 | 7674 | if (!to_replace_bs) { |
785ec4b1 | 7675 | error_setg(errp, "Failed to find node with node-name='%s'", node_name); |
09158f00 BC |
7676 | return NULL; |
7677 | } | |
7678 | ||
5a7e7a0b SH |
7679 | aio_context = bdrv_get_aio_context(to_replace_bs); |
7680 | aio_context_acquire(aio_context); | |
7681 | ||
09158f00 | 7682 | if (bdrv_op_is_blocked(to_replace_bs, BLOCK_OP_TYPE_REPLACE, errp)) { |
5a7e7a0b SH |
7683 | to_replace_bs = NULL; |
7684 | goto out; | |
09158f00 BC |
7685 | } |
7686 | ||
7687 | /* We don't want arbitrary node of the BDS chain to be replaced only the top | |
7688 | * most non filter in order to prevent data corruption. | |
7689 | * Another benefit is that this tests exclude backing files which are | |
7690 | * blocked by the backing blockers. | |
7691 | */ | |
810803a8 HR |
7692 | if (!bdrv_recurse_can_replace(parent_bs, to_replace_bs)) { |
7693 | error_setg(errp, "Cannot replace '%s' by a node mirrored from '%s', " | |
7694 | "because it cannot be guaranteed that doing so would not " | |
7695 | "lead to an abrupt change of visible data", | |
7696 | node_name, parent_bs->node_name); | |
5a7e7a0b SH |
7697 | to_replace_bs = NULL; |
7698 | goto out; | |
09158f00 BC |
7699 | } |
7700 | ||
5a7e7a0b SH |
7701 | out: |
7702 | aio_context_release(aio_context); | |
09158f00 BC |
7703 | return to_replace_bs; |
7704 | } | |
448ad91d | 7705 | |
97e2f021 HR |
7706 | /** |
7707 | * Iterates through the list of runtime option keys that are said to | |
7708 | * be "strong" for a BDS. An option is called "strong" if it changes | |
7709 | * a BDS's data. For example, the null block driver's "size" and | |
7710 | * "read-zeroes" options are strong, but its "latency-ns" option is | |
7711 | * not. | |
7712 | * | |
7713 | * If a key returned by this function ends with a dot, all options | |
7714 | * starting with that prefix are strong. | |
7715 | */ | |
7716 | static const char *const *strong_options(BlockDriverState *bs, | |
7717 | const char *const *curopt) | |
7718 | { | |
7719 | static const char *const global_options[] = { | |
7720 | "driver", "filename", NULL | |
7721 | }; | |
7722 | ||
7723 | if (!curopt) { | |
7724 | return &global_options[0]; | |
7725 | } | |
7726 | ||
7727 | curopt++; | |
7728 | if (curopt == &global_options[ARRAY_SIZE(global_options) - 1] && bs->drv) { | |
7729 | curopt = bs->drv->strong_runtime_opts; | |
7730 | } | |
7731 | ||
7732 | return (curopt && *curopt) ? curopt : NULL; | |
7733 | } | |
7734 | ||
7735 | /** | |
7736 | * Copies all strong runtime options from bs->options to the given | |
7737 | * QDict. The set of strong option keys is determined by invoking | |
7738 | * strong_options(). | |
7739 | * | |
7740 | * Returns true iff any strong option was present in bs->options (and | |
7741 | * thus copied to the target QDict) with the exception of "filename" | |
7742 | * and "driver". The caller is expected to use this value to decide | |
7743 | * whether the existence of strong options prevents the generation of | |
7744 | * a plain filename. | |
7745 | */ | |
7746 | static bool append_strong_runtime_options(QDict *d, BlockDriverState *bs) | |
7747 | { | |
7748 | bool found_any = false; | |
7749 | const char *const *option_name = NULL; | |
7750 | ||
7751 | if (!bs->drv) { | |
7752 | return false; | |
7753 | } | |
7754 | ||
7755 | while ((option_name = strong_options(bs, option_name))) { | |
7756 | bool option_given = false; | |
7757 | ||
7758 | assert(strlen(*option_name) > 0); | |
7759 | if ((*option_name)[strlen(*option_name) - 1] != '.') { | |
7760 | QObject *entry = qdict_get(bs->options, *option_name); | |
7761 | if (!entry) { | |
7762 | continue; | |
7763 | } | |
7764 | ||
7765 | qdict_put_obj(d, *option_name, qobject_ref(entry)); | |
7766 | option_given = true; | |
7767 | } else { | |
7768 | const QDictEntry *entry; | |
7769 | for (entry = qdict_first(bs->options); entry; | |
7770 | entry = qdict_next(bs->options, entry)) | |
7771 | { | |
7772 | if (strstart(qdict_entry_key(entry), *option_name, NULL)) { | |
7773 | qdict_put_obj(d, qdict_entry_key(entry), | |
7774 | qobject_ref(qdict_entry_value(entry))); | |
7775 | option_given = true; | |
7776 | } | |
7777 | } | |
7778 | } | |
7779 | ||
7780 | /* While "driver" and "filename" need to be included in a JSON filename, | |
7781 | * their existence does not prohibit generation of a plain filename. */ | |
7782 | if (!found_any && option_given && | |
7783 | strcmp(*option_name, "driver") && strcmp(*option_name, "filename")) | |
7784 | { | |
7785 | found_any = true; | |
7786 | } | |
7787 | } | |
7788 | ||
62a01a27 HR |
7789 | if (!qdict_haskey(d, "driver")) { |
7790 | /* Drivers created with bdrv_new_open_driver() may not have a | |
7791 | * @driver option. Add it here. */ | |
7792 | qdict_put_str(d, "driver", bs->drv->format_name); | |
7793 | } | |
7794 | ||
97e2f021 HR |
7795 | return found_any; |
7796 | } | |
7797 | ||
90993623 HR |
7798 | /* Note: This function may return false positives; it may return true |
7799 | * even if opening the backing file specified by bs's image header | |
7800 | * would result in exactly bs->backing. */ | |
fa8fc1d0 | 7801 | static bool bdrv_backing_overridden(BlockDriverState *bs) |
90993623 | 7802 | { |
b4ad82aa | 7803 | GLOBAL_STATE_CODE(); |
90993623 HR |
7804 | if (bs->backing) { |
7805 | return strcmp(bs->auto_backing_file, | |
7806 | bs->backing->bs->filename); | |
7807 | } else { | |
7808 | /* No backing BDS, so if the image header reports any backing | |
7809 | * file, it must have been suppressed */ | |
7810 | return bs->auto_backing_file[0] != '\0'; | |
7811 | } | |
7812 | } | |
7813 | ||
91af7014 HR |
7814 | /* Updates the following BDS fields: |
7815 | * - exact_filename: A filename which may be used for opening a block device | |
7816 | * which (mostly) equals the given BDS (even without any | |
7817 | * other options; so reading and writing must return the same | |
7818 | * results, but caching etc. may be different) | |
7819 | * - full_open_options: Options which, when given when opening a block device | |
7820 | * (without a filename), result in a BDS (mostly) | |
7821 | * equalling the given one | |
7822 | * - filename: If exact_filename is set, it is copied here. Otherwise, | |
7823 | * full_open_options is converted to a JSON object, prefixed with | |
7824 | * "json:" (for use through the JSON pseudo protocol) and put here. | |
7825 | */ | |
7826 | void bdrv_refresh_filename(BlockDriverState *bs) | |
7827 | { | |
7828 | BlockDriver *drv = bs->drv; | |
e24518e3 | 7829 | BdrvChild *child; |
52f72d6f | 7830 | BlockDriverState *primary_child_bs; |
91af7014 | 7831 | QDict *opts; |
90993623 | 7832 | bool backing_overridden; |
998b3a1e HR |
7833 | bool generate_json_filename; /* Whether our default implementation should |
7834 | fill exact_filename (false) or not (true) */ | |
91af7014 | 7835 | |
f791bf7f EGE |
7836 | GLOBAL_STATE_CODE(); |
7837 | ||
91af7014 HR |
7838 | if (!drv) { |
7839 | return; | |
7840 | } | |
7841 | ||
e24518e3 HR |
7842 | /* This BDS's file name may depend on any of its children's file names, so |
7843 | * refresh those first */ | |
7844 | QLIST_FOREACH(child, &bs->children, next) { | |
7845 | bdrv_refresh_filename(child->bs); | |
91af7014 HR |
7846 | } |
7847 | ||
bb808d5f HR |
7848 | if (bs->implicit) { |
7849 | /* For implicit nodes, just copy everything from the single child */ | |
7850 | child = QLIST_FIRST(&bs->children); | |
7851 | assert(QLIST_NEXT(child, next) == NULL); | |
7852 | ||
7853 | pstrcpy(bs->exact_filename, sizeof(bs->exact_filename), | |
7854 | child->bs->exact_filename); | |
7855 | pstrcpy(bs->filename, sizeof(bs->filename), child->bs->filename); | |
7856 | ||
cb895614 | 7857 | qobject_unref(bs->full_open_options); |
bb808d5f HR |
7858 | bs->full_open_options = qobject_ref(child->bs->full_open_options); |
7859 | ||
7860 | return; | |
7861 | } | |
7862 | ||
90993623 HR |
7863 | backing_overridden = bdrv_backing_overridden(bs); |
7864 | ||
7865 | if (bs->open_flags & BDRV_O_NO_IO) { | |
7866 | /* Without I/O, the backing file does not change anything. | |
7867 | * Therefore, in such a case (primarily qemu-img), we can | |
7868 | * pretend the backing file has not been overridden even if | |
7869 | * it technically has been. */ | |
7870 | backing_overridden = false; | |
7871 | } | |
7872 | ||
97e2f021 HR |
7873 | /* Gather the options QDict */ |
7874 | opts = qdict_new(); | |
998b3a1e HR |
7875 | generate_json_filename = append_strong_runtime_options(opts, bs); |
7876 | generate_json_filename |= backing_overridden; | |
97e2f021 HR |
7877 | |
7878 | if (drv->bdrv_gather_child_options) { | |
7879 | /* Some block drivers may not want to present all of their children's | |
7880 | * options, or name them differently from BdrvChild.name */ | |
7881 | drv->bdrv_gather_child_options(bs, opts, backing_overridden); | |
7882 | } else { | |
7883 | QLIST_FOREACH(child, &bs->children, next) { | |
25191e5f | 7884 | if (child == bs->backing && !backing_overridden) { |
97e2f021 HR |
7885 | /* We can skip the backing BDS if it has not been overridden */ |
7886 | continue; | |
7887 | } | |
7888 | ||
7889 | qdict_put(opts, child->name, | |
7890 | qobject_ref(child->bs->full_open_options)); | |
7891 | } | |
7892 | ||
7893 | if (backing_overridden && !bs->backing) { | |
7894 | /* Force no backing file */ | |
7895 | qdict_put_null(opts, "backing"); | |
7896 | } | |
7897 | } | |
7898 | ||
7899 | qobject_unref(bs->full_open_options); | |
7900 | bs->full_open_options = opts; | |
7901 | ||
52f72d6f HR |
7902 | primary_child_bs = bdrv_primary_bs(bs); |
7903 | ||
998b3a1e HR |
7904 | if (drv->bdrv_refresh_filename) { |
7905 | /* Obsolete information is of no use here, so drop the old file name | |
7906 | * information before refreshing it */ | |
7907 | bs->exact_filename[0] = '\0'; | |
7908 | ||
7909 | drv->bdrv_refresh_filename(bs); | |
52f72d6f HR |
7910 | } else if (primary_child_bs) { |
7911 | /* | |
7912 | * Try to reconstruct valid information from the underlying | |
7913 | * file -- this only works for format nodes (filter nodes | |
7914 | * cannot be probed and as such must be selected by the user | |
7915 | * either through an options dict, or through a special | |
7916 | * filename which the filter driver must construct in its | |
7917 | * .bdrv_refresh_filename() implementation). | |
7918 | */ | |
998b3a1e HR |
7919 | |
7920 | bs->exact_filename[0] = '\0'; | |
7921 | ||
fb695c74 HR |
7922 | /* |
7923 | * We can use the underlying file's filename if: | |
7924 | * - it has a filename, | |
52f72d6f | 7925 | * - the current BDS is not a filter, |
fb695c74 HR |
7926 | * - the file is a protocol BDS, and |
7927 | * - opening that file (as this BDS's format) will automatically create | |
7928 | * the BDS tree we have right now, that is: | |
7929 | * - the user did not significantly change this BDS's behavior with | |
7930 | * some explicit (strong) options | |
7931 | * - no non-file child of this BDS has been overridden by the user | |
7932 | * Both of these conditions are represented by generate_json_filename. | |
7933 | */ | |
52f72d6f HR |
7934 | if (primary_child_bs->exact_filename[0] && |
7935 | primary_child_bs->drv->bdrv_file_open && | |
7936 | !drv->is_filter && !generate_json_filename) | |
fb695c74 | 7937 | { |
52f72d6f | 7938 | strcpy(bs->exact_filename, primary_child_bs->exact_filename); |
998b3a1e HR |
7939 | } |
7940 | } | |
7941 | ||
91af7014 HR |
7942 | if (bs->exact_filename[0]) { |
7943 | pstrcpy(bs->filename, sizeof(bs->filename), bs->exact_filename); | |
97e2f021 | 7944 | } else { |
eab3a467 | 7945 | GString *json = qobject_to_json(QOBJECT(bs->full_open_options)); |
5c86bdf1 | 7946 | if (snprintf(bs->filename, sizeof(bs->filename), "json:%s", |
eab3a467 | 7947 | json->str) >= sizeof(bs->filename)) { |
5c86bdf1 EB |
7948 | /* Give user a hint if we truncated things. */ |
7949 | strcpy(bs->filename + sizeof(bs->filename) - 4, "..."); | |
7950 | } | |
eab3a467 | 7951 | g_string_free(json, true); |
91af7014 HR |
7952 | } |
7953 | } | |
e06018ad | 7954 | |
1e89d0f9 HR |
7955 | char *bdrv_dirname(BlockDriverState *bs, Error **errp) |
7956 | { | |
7957 | BlockDriver *drv = bs->drv; | |
52f72d6f | 7958 | BlockDriverState *child_bs; |
1e89d0f9 | 7959 | |
f791bf7f EGE |
7960 | GLOBAL_STATE_CODE(); |
7961 | ||
1e89d0f9 HR |
7962 | if (!drv) { |
7963 | error_setg(errp, "Node '%s' is ejected", bs->node_name); | |
7964 | return NULL; | |
7965 | } | |
7966 | ||
7967 | if (drv->bdrv_dirname) { | |
7968 | return drv->bdrv_dirname(bs, errp); | |
7969 | } | |
7970 | ||
52f72d6f HR |
7971 | child_bs = bdrv_primary_bs(bs); |
7972 | if (child_bs) { | |
7973 | return bdrv_dirname(child_bs, errp); | |
1e89d0f9 HR |
7974 | } |
7975 | ||
7976 | bdrv_refresh_filename(bs); | |
7977 | if (bs->exact_filename[0] != '\0') { | |
7978 | return path_combine(bs->exact_filename, ""); | |
7979 | } | |
7980 | ||
7981 | error_setg(errp, "Cannot generate a base directory for %s nodes", | |
7982 | drv->format_name); | |
7983 | return NULL; | |
7984 | } | |
7985 | ||
e06018ad WC |
7986 | /* |
7987 | * Hot add/remove a BDS's child. So the user can take a child offline when | |
7988 | * it is broken and take a new child online | |
7989 | */ | |
7990 | void bdrv_add_child(BlockDriverState *parent_bs, BlockDriverState *child_bs, | |
7991 | Error **errp) | |
7992 | { | |
f791bf7f | 7993 | GLOBAL_STATE_CODE(); |
e06018ad WC |
7994 | if (!parent_bs->drv || !parent_bs->drv->bdrv_add_child) { |
7995 | error_setg(errp, "The node %s does not support adding a child", | |
7996 | bdrv_get_device_or_node_name(parent_bs)); | |
7997 | return; | |
7998 | } | |
7999 | ||
774c726c SL |
8000 | /* |
8001 | * Non-zoned block drivers do not follow zoned storage constraints | |
8002 | * (i.e. sequential writes to zones). Refuse mixing zoned and non-zoned | |
8003 | * drivers in a graph. | |
8004 | */ | |
8005 | if (!parent_bs->drv->supports_zoned_children && | |
8006 | child_bs->bl.zoned == BLK_Z_HM) { | |
8007 | /* | |
8008 | * The host-aware model allows zoned storage constraints and random | |
8009 | * write. Allow mixing host-aware and non-zoned drivers. Using | |
8010 | * host-aware device as a regular device. | |
8011 | */ | |
8012 | error_setg(errp, "Cannot add a %s child to a %s parent", | |
8013 | child_bs->bl.zoned == BLK_Z_HM ? "zoned" : "non-zoned", | |
8014 | parent_bs->drv->supports_zoned_children ? | |
8015 | "support zoned children" : "not support zoned children"); | |
8016 | return; | |
8017 | } | |
8018 | ||
e06018ad WC |
8019 | if (!QLIST_EMPTY(&child_bs->parents)) { |
8020 | error_setg(errp, "The node %s already has a parent", | |
8021 | child_bs->node_name); | |
8022 | return; | |
8023 | } | |
8024 | ||
8025 | parent_bs->drv->bdrv_add_child(parent_bs, child_bs, errp); | |
8026 | } | |
8027 | ||
8028 | void bdrv_del_child(BlockDriverState *parent_bs, BdrvChild *child, Error **errp) | |
8029 | { | |
8030 | BdrvChild *tmp; | |
8031 | ||
f791bf7f | 8032 | GLOBAL_STATE_CODE(); |
e06018ad WC |
8033 | if (!parent_bs->drv || !parent_bs->drv->bdrv_del_child) { |
8034 | error_setg(errp, "The node %s does not support removing a child", | |
8035 | bdrv_get_device_or_node_name(parent_bs)); | |
8036 | return; | |
8037 | } | |
8038 | ||
8039 | QLIST_FOREACH(tmp, &parent_bs->children, next) { | |
8040 | if (tmp == child) { | |
8041 | break; | |
8042 | } | |
8043 | } | |
8044 | ||
8045 | if (!tmp) { | |
8046 | error_setg(errp, "The node %s does not have a child named %s", | |
8047 | bdrv_get_device_or_node_name(parent_bs), | |
8048 | bdrv_get_device_or_node_name(child->bs)); | |
8049 | return; | |
8050 | } | |
8051 | ||
8052 | parent_bs->drv->bdrv_del_child(parent_bs, child, errp); | |
8053 | } | |
6f7a3b53 HR |
8054 | |
8055 | int bdrv_make_empty(BdrvChild *c, Error **errp) | |
8056 | { | |
8057 | BlockDriver *drv = c->bs->drv; | |
8058 | int ret; | |
8059 | ||
f791bf7f | 8060 | GLOBAL_STATE_CODE(); |
6f7a3b53 HR |
8061 | assert(c->perm & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)); |
8062 | ||
8063 | if (!drv->bdrv_make_empty) { | |
8064 | error_setg(errp, "%s does not support emptying nodes", | |
8065 | drv->format_name); | |
8066 | return -ENOTSUP; | |
8067 | } | |
8068 | ||
8069 | ret = drv->bdrv_make_empty(c->bs); | |
8070 | if (ret < 0) { | |
8071 | error_setg_errno(errp, -ret, "Failed to empty %s", | |
8072 | c->bs->filename); | |
8073 | return ret; | |
8074 | } | |
8075 | ||
8076 | return 0; | |
8077 | } | |
9a6fc887 HR |
8078 | |
8079 | /* | |
8080 | * Return the child that @bs acts as an overlay for, and from which data may be | |
8081 | * copied in COW or COR operations. Usually this is the backing file. | |
8082 | */ | |
8083 | BdrvChild *bdrv_cow_child(BlockDriverState *bs) | |
8084 | { | |
967d7905 EGE |
8085 | IO_CODE(); |
8086 | ||
9a6fc887 HR |
8087 | if (!bs || !bs->drv) { |
8088 | return NULL; | |
8089 | } | |
8090 | ||
8091 | if (bs->drv->is_filter) { | |
8092 | return NULL; | |
8093 | } | |
8094 | ||
8095 | if (!bs->backing) { | |
8096 | return NULL; | |
8097 | } | |
8098 | ||
8099 | assert(bs->backing->role & BDRV_CHILD_COW); | |
8100 | return bs->backing; | |
8101 | } | |
8102 | ||
8103 | /* | |
8104 | * If @bs acts as a filter for exactly one of its children, return | |
8105 | * that child. | |
8106 | */ | |
8107 | BdrvChild *bdrv_filter_child(BlockDriverState *bs) | |
8108 | { | |
8109 | BdrvChild *c; | |
967d7905 | 8110 | IO_CODE(); |
9a6fc887 HR |
8111 | |
8112 | if (!bs || !bs->drv) { | |
8113 | return NULL; | |
8114 | } | |
8115 | ||
8116 | if (!bs->drv->is_filter) { | |
8117 | return NULL; | |
8118 | } | |
8119 | ||
8120 | /* Only one of @backing or @file may be used */ | |
8121 | assert(!(bs->backing && bs->file)); | |
8122 | ||
8123 | c = bs->backing ?: bs->file; | |
8124 | if (!c) { | |
8125 | return NULL; | |
8126 | } | |
8127 | ||
8128 | assert(c->role & BDRV_CHILD_FILTERED); | |
8129 | return c; | |
8130 | } | |
8131 | ||
8132 | /* | |
8133 | * Return either the result of bdrv_cow_child() or bdrv_filter_child(), | |
8134 | * whichever is non-NULL. | |
8135 | * | |
8136 | * Return NULL if both are NULL. | |
8137 | */ | |
8138 | BdrvChild *bdrv_filter_or_cow_child(BlockDriverState *bs) | |
8139 | { | |
8140 | BdrvChild *cow_child = bdrv_cow_child(bs); | |
8141 | BdrvChild *filter_child = bdrv_filter_child(bs); | |
967d7905 | 8142 | IO_CODE(); |
9a6fc887 HR |
8143 | |
8144 | /* Filter nodes cannot have COW backing files */ | |
8145 | assert(!(cow_child && filter_child)); | |
8146 | ||
8147 | return cow_child ?: filter_child; | |
8148 | } | |
8149 | ||
8150 | /* | |
8151 | * Return the primary child of this node: For filters, that is the | |
8152 | * filtered child. For other nodes, that is usually the child storing | |
8153 | * metadata. | |
8154 | * (A generally more helpful description is that this is (usually) the | |
8155 | * child that has the same filename as @bs.) | |
8156 | * | |
8157 | * Drivers do not necessarily have a primary child; for example quorum | |
8158 | * does not. | |
8159 | */ | |
8160 | BdrvChild *bdrv_primary_child(BlockDriverState *bs) | |
8161 | { | |
8162 | BdrvChild *c, *found = NULL; | |
967d7905 | 8163 | IO_CODE(); |
9a6fc887 HR |
8164 | |
8165 | QLIST_FOREACH(c, &bs->children, next) { | |
8166 | if (c->role & BDRV_CHILD_PRIMARY) { | |
8167 | assert(!found); | |
8168 | found = c; | |
8169 | } | |
8170 | } | |
8171 | ||
8172 | return found; | |
8173 | } | |
d38d7eb8 HR |
8174 | |
8175 | static BlockDriverState *bdrv_do_skip_filters(BlockDriverState *bs, | |
8176 | bool stop_on_explicit_filter) | |
8177 | { | |
8178 | BdrvChild *c; | |
8179 | ||
8180 | if (!bs) { | |
8181 | return NULL; | |
8182 | } | |
8183 | ||
8184 | while (!(stop_on_explicit_filter && !bs->implicit)) { | |
8185 | c = bdrv_filter_child(bs); | |
8186 | if (!c) { | |
8187 | /* | |
8188 | * A filter that is embedded in a working block graph must | |
8189 | * have a child. Assert this here so this function does | |
8190 | * not return a filter node that is not expected by the | |
8191 | * caller. | |
8192 | */ | |
8193 | assert(!bs->drv || !bs->drv->is_filter); | |
8194 | break; | |
8195 | } | |
8196 | bs = c->bs; | |
8197 | } | |
8198 | /* | |
8199 | * Note that this treats nodes with bs->drv == NULL as not being | |
8200 | * filters (bs->drv == NULL should be replaced by something else | |
8201 | * anyway). | |
8202 | * The advantage of this behavior is that this function will thus | |
8203 | * always return a non-NULL value (given a non-NULL @bs). | |
8204 | */ | |
8205 | ||
8206 | return bs; | |
8207 | } | |
8208 | ||
8209 | /* | |
8210 | * Return the first BDS that has not been added implicitly or that | |
8211 | * does not have a filtered child down the chain starting from @bs | |
8212 | * (including @bs itself). | |
8213 | */ | |
8214 | BlockDriverState *bdrv_skip_implicit_filters(BlockDriverState *bs) | |
8215 | { | |
b4ad82aa | 8216 | GLOBAL_STATE_CODE(); |
d38d7eb8 HR |
8217 | return bdrv_do_skip_filters(bs, true); |
8218 | } | |
8219 | ||
8220 | /* | |
8221 | * Return the first BDS that does not have a filtered child down the | |
8222 | * chain starting from @bs (including @bs itself). | |
8223 | */ | |
8224 | BlockDriverState *bdrv_skip_filters(BlockDriverState *bs) | |
8225 | { | |
967d7905 | 8226 | IO_CODE(); |
d38d7eb8 HR |
8227 | return bdrv_do_skip_filters(bs, false); |
8228 | } | |
8229 | ||
8230 | /* | |
8231 | * For a backing chain, return the first non-filter backing image of | |
8232 | * the first non-filter image. | |
8233 | */ | |
8234 | BlockDriverState *bdrv_backing_chain_next(BlockDriverState *bs) | |
8235 | { | |
967d7905 | 8236 | IO_CODE(); |
d38d7eb8 HR |
8237 | return bdrv_skip_filters(bdrv_cow_bs(bdrv_skip_filters(bs))); |
8238 | } | |
0bc329fb HR |
8239 | |
8240 | /** | |
8241 | * Check whether [offset, offset + bytes) overlaps with the cached | |
8242 | * block-status data region. | |
8243 | * | |
8244 | * If so, and @pnum is not NULL, set *pnum to `bsc.data_end - offset`, | |
8245 | * which is what bdrv_bsc_is_data()'s interface needs. | |
8246 | * Otherwise, *pnum is not touched. | |
8247 | */ | |
8248 | static bool bdrv_bsc_range_overlaps_locked(BlockDriverState *bs, | |
8249 | int64_t offset, int64_t bytes, | |
8250 | int64_t *pnum) | |
8251 | { | |
8252 | BdrvBlockStatusCache *bsc = qatomic_rcu_read(&bs->block_status_cache); | |
8253 | bool overlaps; | |
8254 | ||
8255 | overlaps = | |
8256 | qatomic_read(&bsc->valid) && | |
8257 | ranges_overlap(offset, bytes, bsc->data_start, | |
8258 | bsc->data_end - bsc->data_start); | |
8259 | ||
8260 | if (overlaps && pnum) { | |
8261 | *pnum = bsc->data_end - offset; | |
8262 | } | |
8263 | ||
8264 | return overlaps; | |
8265 | } | |
8266 | ||
8267 | /** | |
8268 | * See block_int.h for this function's documentation. | |
8269 | */ | |
8270 | bool bdrv_bsc_is_data(BlockDriverState *bs, int64_t offset, int64_t *pnum) | |
8271 | { | |
967d7905 | 8272 | IO_CODE(); |
0bc329fb | 8273 | RCU_READ_LOCK_GUARD(); |
0bc329fb HR |
8274 | return bdrv_bsc_range_overlaps_locked(bs, offset, 1, pnum); |
8275 | } | |
8276 | ||
8277 | /** | |
8278 | * See block_int.h for this function's documentation. | |
8279 | */ | |
8280 | void bdrv_bsc_invalidate_range(BlockDriverState *bs, | |
8281 | int64_t offset, int64_t bytes) | |
8282 | { | |
967d7905 | 8283 | IO_CODE(); |
0bc329fb HR |
8284 | RCU_READ_LOCK_GUARD(); |
8285 | ||
8286 | if (bdrv_bsc_range_overlaps_locked(bs, offset, bytes, NULL)) { | |
8287 | qatomic_set(&bs->block_status_cache->valid, false); | |
8288 | } | |
8289 | } | |
8290 | ||
8291 | /** | |
8292 | * See block_int.h for this function's documentation. | |
8293 | */ | |
8294 | void bdrv_bsc_fill(BlockDriverState *bs, int64_t offset, int64_t bytes) | |
8295 | { | |
8296 | BdrvBlockStatusCache *new_bsc = g_new(BdrvBlockStatusCache, 1); | |
8297 | BdrvBlockStatusCache *old_bsc; | |
967d7905 | 8298 | IO_CODE(); |
0bc329fb HR |
8299 | |
8300 | *new_bsc = (BdrvBlockStatusCache) { | |
8301 | .valid = true, | |
8302 | .data_start = offset, | |
8303 | .data_end = offset + bytes, | |
8304 | }; | |
8305 | ||
8306 | QEMU_LOCK_GUARD(&bs->bsc_modify_lock); | |
8307 | ||
8308 | old_bsc = qatomic_rcu_read(&bs->block_status_cache); | |
8309 | qatomic_rcu_set(&bs->block_status_cache, new_bsc); | |
8310 | if (old_bsc) { | |
8311 | g_free_rcu(old_bsc, rcu); | |
8312 | } | |
8313 | } |