]>
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 | */ | |
84569a7d PB |
558 | static int64_t coroutine_fn GRAPH_UNLOCKED |
559 | create_file_fallback_truncate(BlockBackend *blk, int64_t minimum_size, | |
560 | Error **errp) | |
84a12e66 | 561 | { |
cc84d90f | 562 | Error *local_err = NULL; |
fd17146c | 563 | int64_t size; |
cc84d90f | 564 | int ret; |
84a12e66 | 565 | |
bdb73476 EGE |
566 | GLOBAL_STATE_CODE(); |
567 | ||
84569a7d PB |
568 | ret = blk_co_truncate(blk, minimum_size, false, PREALLOC_MODE_OFF, 0, |
569 | &local_err); | |
fd17146c HR |
570 | if (ret < 0 && ret != -ENOTSUP) { |
571 | error_propagate(errp, local_err); | |
572 | return ret; | |
573 | } | |
574 | ||
84569a7d | 575 | size = blk_co_getlength(blk); |
fd17146c HR |
576 | if (size < 0) { |
577 | error_free(local_err); | |
578 | error_setg_errno(errp, -size, | |
579 | "Failed to inquire the new image file's length"); | |
580 | return size; | |
581 | } | |
582 | ||
583 | if (size < minimum_size) { | |
584 | /* Need to grow the image, but we failed to do that */ | |
585 | error_propagate(errp, local_err); | |
586 | return -ENOTSUP; | |
587 | } | |
588 | ||
589 | error_free(local_err); | |
590 | local_err = NULL; | |
591 | ||
592 | return size; | |
593 | } | |
594 | ||
595 | /** | |
596 | * Helper function for bdrv_create_file_fallback(): Zero the first | |
597 | * sector to remove any potentially pre-existing image header. | |
598 | */ | |
881a4c55 PB |
599 | static int coroutine_fn |
600 | create_file_fallback_zero_first_sector(BlockBackend *blk, | |
601 | int64_t current_size, | |
602 | Error **errp) | |
fd17146c HR |
603 | { |
604 | int64_t bytes_to_clear; | |
605 | int ret; | |
606 | ||
bdb73476 EGE |
607 | GLOBAL_STATE_CODE(); |
608 | ||
fd17146c HR |
609 | bytes_to_clear = MIN(current_size, BDRV_SECTOR_SIZE); |
610 | if (bytes_to_clear) { | |
ce47ff20 | 611 | ret = blk_co_pwrite_zeroes(blk, 0, bytes_to_clear, BDRV_REQ_MAY_UNMAP); |
fd17146c HR |
612 | if (ret < 0) { |
613 | error_setg_errno(errp, -ret, | |
614 | "Failed to clear the new image's first sector"); | |
615 | return ret; | |
616 | } | |
617 | } | |
618 | ||
619 | return 0; | |
620 | } | |
621 | ||
5a5e7f8c ML |
622 | /** |
623 | * Simple implementation of bdrv_co_create_opts for protocol drivers | |
624 | * which only support creation via opening a file | |
625 | * (usually existing raw storage device) | |
626 | */ | |
627 | int coroutine_fn bdrv_co_create_opts_simple(BlockDriver *drv, | |
628 | const char *filename, | |
629 | QemuOpts *opts, | |
630 | Error **errp) | |
fd17146c HR |
631 | { |
632 | BlockBackend *blk; | |
eeea1faa | 633 | QDict *options; |
fd17146c HR |
634 | int64_t size = 0; |
635 | char *buf = NULL; | |
636 | PreallocMode prealloc; | |
637 | Error *local_err = NULL; | |
638 | int ret; | |
639 | ||
b4ad82aa EGE |
640 | GLOBAL_STATE_CODE(); |
641 | ||
fd17146c HR |
642 | size = qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0); |
643 | buf = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC); | |
644 | prealloc = qapi_enum_parse(&PreallocMode_lookup, buf, | |
645 | PREALLOC_MODE_OFF, &local_err); | |
646 | g_free(buf); | |
647 | if (local_err) { | |
648 | error_propagate(errp, local_err); | |
649 | return -EINVAL; | |
650 | } | |
651 | ||
652 | if (prealloc != PREALLOC_MODE_OFF) { | |
653 | error_setg(errp, "Unsupported preallocation mode '%s'", | |
654 | PreallocMode_str(prealloc)); | |
655 | return -ENOTSUP; | |
656 | } | |
657 | ||
eeea1faa | 658 | options = qdict_new(); |
fd17146c HR |
659 | qdict_put_str(options, "driver", drv->format_name); |
660 | ||
be1a732c KW |
661 | blk = blk_co_new_open(filename, NULL, options, |
662 | BDRV_O_RDWR | BDRV_O_RESIZE, errp); | |
fd17146c HR |
663 | if (!blk) { |
664 | error_prepend(errp, "Protocol driver '%s' does not support image " | |
665 | "creation, and opening the image failed: ", | |
666 | drv->format_name); | |
667 | return -EINVAL; | |
668 | } | |
669 | ||
670 | size = create_file_fallback_truncate(blk, size, errp); | |
671 | if (size < 0) { | |
672 | ret = size; | |
673 | goto out; | |
674 | } | |
675 | ||
676 | ret = create_file_fallback_zero_first_sector(blk, size, errp); | |
677 | if (ret < 0) { | |
678 | goto out; | |
679 | } | |
680 | ||
681 | ret = 0; | |
682 | out: | |
b2ab5f54 | 683 | blk_co_unref(blk); |
fd17146c HR |
684 | return ret; |
685 | } | |
686 | ||
2475a0d0 EGE |
687 | int coroutine_fn bdrv_co_create_file(const char *filename, QemuOpts *opts, |
688 | Error **errp) | |
fd17146c | 689 | { |
729222af | 690 | QemuOpts *protocol_opts; |
fd17146c | 691 | BlockDriver *drv; |
729222af SG |
692 | QDict *qdict; |
693 | int ret; | |
fd17146c | 694 | |
f791bf7f EGE |
695 | GLOBAL_STATE_CODE(); |
696 | ||
b65a5e12 | 697 | drv = bdrv_find_protocol(filename, true, errp); |
84a12e66 | 698 | if (drv == NULL) { |
16905d71 | 699 | return -ENOENT; |
84a12e66 CH |
700 | } |
701 | ||
729222af SG |
702 | if (!drv->create_opts) { |
703 | error_setg(errp, "Driver '%s' does not support image creation", | |
704 | drv->format_name); | |
705 | return -ENOTSUP; | |
706 | } | |
707 | ||
708 | /* | |
709 | * 'opts' contains a QemuOptsList with a combination of format and protocol | |
710 | * default values. | |
711 | * | |
712 | * The format properly removes its options, but the default values remain | |
713 | * in 'opts->list'. So if the protocol has options with the same name | |
714 | * (e.g. rbd has 'cluster_size' as qcow2), it will see the default values | |
715 | * of the format, since for overlapping options, the format wins. | |
716 | * | |
717 | * To avoid this issue, lets convert QemuOpts to QDict, in this way we take | |
718 | * only the set options, and then convert it back to QemuOpts, using the | |
719 | * create_opts of the protocol. So the new QemuOpts, will contain only the | |
720 | * protocol defaults. | |
721 | */ | |
722 | qdict = qemu_opts_to_qdict(opts, NULL); | |
723 | protocol_opts = qemu_opts_from_qdict(drv->create_opts, qdict, errp); | |
724 | if (protocol_opts == NULL) { | |
725 | ret = -EINVAL; | |
726 | goto out; | |
727 | } | |
728 | ||
2475a0d0 | 729 | ret = bdrv_co_create(drv, filename, protocol_opts, errp); |
729222af SG |
730 | out: |
731 | qemu_opts_del(protocol_opts); | |
732 | qobject_unref(qdict); | |
733 | return ret; | |
84a12e66 CH |
734 | } |
735 | ||
e1d7f8bb DHB |
736 | int coroutine_fn bdrv_co_delete_file(BlockDriverState *bs, Error **errp) |
737 | { | |
738 | Error *local_err = NULL; | |
739 | int ret; | |
740 | ||
384a48fb | 741 | IO_CODE(); |
e1d7f8bb | 742 | assert(bs != NULL); |
48aef794 | 743 | assert_bdrv_graph_readable(); |
e1d7f8bb DHB |
744 | |
745 | if (!bs->drv) { | |
746 | error_setg(errp, "Block node '%s' is not opened", bs->filename); | |
747 | return -ENOMEDIUM; | |
748 | } | |
749 | ||
750 | if (!bs->drv->bdrv_co_delete_file) { | |
751 | error_setg(errp, "Driver '%s' does not support image deletion", | |
752 | bs->drv->format_name); | |
753 | return -ENOTSUP; | |
754 | } | |
755 | ||
756 | ret = bs->drv->bdrv_co_delete_file(bs, &local_err); | |
757 | if (ret < 0) { | |
758 | error_propagate(errp, local_err); | |
759 | } | |
760 | ||
761 | return ret; | |
762 | } | |
763 | ||
a890f08e ML |
764 | void coroutine_fn bdrv_co_delete_file_noerr(BlockDriverState *bs) |
765 | { | |
766 | Error *local_err = NULL; | |
767 | int ret; | |
384a48fb | 768 | IO_CODE(); |
a890f08e ML |
769 | |
770 | if (!bs) { | |
771 | return; | |
772 | } | |
773 | ||
774 | ret = bdrv_co_delete_file(bs, &local_err); | |
775 | /* | |
776 | * ENOTSUP will happen if the block driver doesn't support | |
777 | * the 'bdrv_co_delete_file' interface. This is a predictable | |
778 | * scenario and shouldn't be reported back to the user. | |
779 | */ | |
780 | if (ret == -ENOTSUP) { | |
781 | error_free(local_err); | |
782 | } else if (ret < 0) { | |
783 | error_report_err(local_err); | |
784 | } | |
785 | } | |
786 | ||
892b7de8 ET |
787 | /** |
788 | * Try to get @bs's logical and physical block size. | |
789 | * On success, store them in @bsz struct and return 0. | |
790 | * On failure return -errno. | |
791 | * @bs must not be empty. | |
792 | */ | |
793 | int bdrv_probe_blocksizes(BlockDriverState *bs, BlockSizes *bsz) | |
794 | { | |
795 | BlockDriver *drv = bs->drv; | |
93393e69 | 796 | BlockDriverState *filtered = bdrv_filter_bs(bs); |
f791bf7f | 797 | GLOBAL_STATE_CODE(); |
892b7de8 ET |
798 | |
799 | if (drv && drv->bdrv_probe_blocksizes) { | |
800 | return drv->bdrv_probe_blocksizes(bs, bsz); | |
93393e69 HR |
801 | } else if (filtered) { |
802 | return bdrv_probe_blocksizes(filtered, bsz); | |
892b7de8 ET |
803 | } |
804 | ||
805 | return -ENOTSUP; | |
806 | } | |
807 | ||
808 | /** | |
809 | * Try to get @bs's geometry (cyls, heads, sectors). | |
810 | * On success, store them in @geo struct and return 0. | |
811 | * On failure return -errno. | |
812 | * @bs must not be empty. | |
813 | */ | |
814 | int bdrv_probe_geometry(BlockDriverState *bs, HDGeometry *geo) | |
815 | { | |
816 | BlockDriver *drv = bs->drv; | |
93393e69 | 817 | BlockDriverState *filtered = bdrv_filter_bs(bs); |
f791bf7f | 818 | GLOBAL_STATE_CODE(); |
892b7de8 ET |
819 | |
820 | if (drv && drv->bdrv_probe_geometry) { | |
821 | return drv->bdrv_probe_geometry(bs, geo); | |
93393e69 HR |
822 | } else if (filtered) { |
823 | return bdrv_probe_geometry(filtered, geo); | |
892b7de8 ET |
824 | } |
825 | ||
826 | return -ENOTSUP; | |
827 | } | |
828 | ||
eba25057 JM |
829 | /* |
830 | * Create a uniquely-named empty temporary file. | |
69fbfff9 BM |
831 | * Return the actual file name used upon success, otherwise NULL. |
832 | * This string should be freed with g_free() when not needed any longer. | |
833 | * | |
834 | * Note: creating a temporary file for the caller to (re)open is | |
835 | * inherently racy. Use g_file_open_tmp() instead whenever practical. | |
eba25057 | 836 | */ |
69fbfff9 | 837 | char *create_tmp_file(Error **errp) |
d5249393 | 838 | { |
67b915a5 | 839 | int fd; |
7ccfb2eb | 840 | const char *tmpdir; |
69fbfff9 BM |
841 | g_autofree char *filename = NULL; |
842 | ||
843 | tmpdir = g_get_tmp_dir(); | |
844 | #ifndef _WIN32 | |
845 | /* | |
846 | * See commit 69bef79 ("block: use /var/tmp instead of /tmp for -snapshot") | |
847 | * | |
848 | * This function is used to create temporary disk images (like -snapshot), | |
849 | * so the files can become very large. /tmp is often a tmpfs where as | |
850 | * /var/tmp is usually on a disk, so more appropriate for disk images. | |
851 | */ | |
852 | if (!g_strcmp0(tmpdir, "/tmp")) { | |
69bef793 AS |
853 | tmpdir = "/var/tmp"; |
854 | } | |
69fbfff9 BM |
855 | #endif |
856 | ||
857 | filename = g_strdup_printf("%s/vl.XXXXXX", tmpdir); | |
858 | fd = g_mkstemp(filename); | |
fe235a06 | 859 | if (fd < 0) { |
69fbfff9 BM |
860 | error_setg_errno(errp, errno, "Could not open temporary file '%s'", |
861 | filename); | |
862 | return NULL; | |
eba25057 | 863 | } |
6b6471ee | 864 | close(fd); |
69fbfff9 BM |
865 | |
866 | return g_steal_pointer(&filename); | |
eba25057 | 867 | } |
fc01f7e7 | 868 | |
84a12e66 CH |
869 | /* |
870 | * Detect host devices. By convention, /dev/cdrom[N] is always | |
871 | * recognized as a host CDROM. | |
872 | */ | |
873 | static BlockDriver *find_hdev_driver(const char *filename) | |
874 | { | |
875 | int score_max = 0, score; | |
876 | BlockDriver *drv = NULL, *d; | |
bdb73476 | 877 | GLOBAL_STATE_CODE(); |
84a12e66 CH |
878 | |
879 | QLIST_FOREACH(d, &bdrv_drivers, list) { | |
880 | if (d->bdrv_probe_device) { | |
881 | score = d->bdrv_probe_device(filename); | |
882 | if (score > score_max) { | |
883 | score_max = score; | |
884 | drv = d; | |
885 | } | |
886 | } | |
887 | } | |
888 | ||
889 | return drv; | |
890 | } | |
891 | ||
88d88798 MM |
892 | static BlockDriver *bdrv_do_find_protocol(const char *protocol) |
893 | { | |
894 | BlockDriver *drv1; | |
bdb73476 | 895 | GLOBAL_STATE_CODE(); |
88d88798 MM |
896 | |
897 | QLIST_FOREACH(drv1, &bdrv_drivers, list) { | |
898 | if (drv1->protocol_name && !strcmp(drv1->protocol_name, protocol)) { | |
899 | return drv1; | |
900 | } | |
901 | } | |
902 | ||
903 | return NULL; | |
904 | } | |
905 | ||
98289620 | 906 | BlockDriver *bdrv_find_protocol(const char *filename, |
b65a5e12 HR |
907 | bool allow_protocol_prefix, |
908 | Error **errp) | |
83f64091 FB |
909 | { |
910 | BlockDriver *drv1; | |
911 | char protocol[128]; | |
1cec71e3 | 912 | int len; |
83f64091 | 913 | const char *p; |
88d88798 | 914 | int i; |
19cb3738 | 915 | |
f791bf7f | 916 | GLOBAL_STATE_CODE(); |
66f82cee KW |
917 | /* TODO Drivers without bdrv_file_open must be specified explicitly */ |
918 | ||
39508e7a CH |
919 | /* |
920 | * XXX(hch): we really should not let host device detection | |
921 | * override an explicit protocol specification, but moving this | |
922 | * later breaks access to device names with colons in them. | |
923 | * Thanks to the brain-dead persistent naming schemes on udev- | |
924 | * based Linux systems those actually are quite common. | |
925 | */ | |
926 | drv1 = find_hdev_driver(filename); | |
927 | if (drv1) { | |
928 | return drv1; | |
929 | } | |
930 | ||
98289620 | 931 | if (!path_has_protocol(filename) || !allow_protocol_prefix) { |
ef810437 | 932 | return &bdrv_file; |
84a12e66 | 933 | } |
98289620 | 934 | |
9e0b22f4 SH |
935 | p = strchr(filename, ':'); |
936 | assert(p != NULL); | |
1cec71e3 AL |
937 | len = p - filename; |
938 | if (len > sizeof(protocol) - 1) | |
939 | len = sizeof(protocol) - 1; | |
940 | memcpy(protocol, filename, len); | |
941 | protocol[len] = '\0'; | |
88d88798 MM |
942 | |
943 | drv1 = bdrv_do_find_protocol(protocol); | |
944 | if (drv1) { | |
945 | return drv1; | |
946 | } | |
947 | ||
948 | for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); ++i) { | |
949 | if (block_driver_modules[i].protocol_name && | |
950 | !strcmp(block_driver_modules[i].protocol_name, protocol)) { | |
c551fb0b CF |
951 | int rv = block_module_load(block_driver_modules[i].library_name, errp); |
952 | if (rv > 0) { | |
953 | drv1 = bdrv_do_find_protocol(protocol); | |
954 | } else if (rv < 0) { | |
955 | return NULL; | |
956 | } | |
88d88798 | 957 | break; |
8a22f02a | 958 | } |
83f64091 | 959 | } |
b65a5e12 | 960 | |
88d88798 MM |
961 | if (!drv1) { |
962 | error_setg(errp, "Unknown protocol '%s'", protocol); | |
963 | } | |
964 | return drv1; | |
83f64091 FB |
965 | } |
966 | ||
c6684249 MA |
967 | /* |
968 | * Guess image format by probing its contents. | |
969 | * This is not a good idea when your image is raw (CVE-2008-2004), but | |
970 | * we do it anyway for backward compatibility. | |
971 | * | |
972 | * @buf contains the image's first @buf_size bytes. | |
7cddd372 KW |
973 | * @buf_size is the buffer size in bytes (generally BLOCK_PROBE_BUF_SIZE, |
974 | * but can be smaller if the image file is smaller) | |
c6684249 MA |
975 | * @filename is its filename. |
976 | * | |
977 | * For all block drivers, call the bdrv_probe() method to get its | |
978 | * probing score. | |
979 | * Return the first block driver with the highest probing score. | |
980 | */ | |
38f3ef57 KW |
981 | BlockDriver *bdrv_probe_all(const uint8_t *buf, int buf_size, |
982 | const char *filename) | |
c6684249 MA |
983 | { |
984 | int score_max = 0, score; | |
985 | BlockDriver *drv = NULL, *d; | |
967d7905 | 986 | IO_CODE(); |
c6684249 MA |
987 | |
988 | QLIST_FOREACH(d, &bdrv_drivers, list) { | |
989 | if (d->bdrv_probe) { | |
990 | score = d->bdrv_probe(buf, buf_size, filename); | |
991 | if (score > score_max) { | |
992 | score_max = score; | |
993 | drv = d; | |
994 | } | |
995 | } | |
996 | } | |
997 | ||
998 | return drv; | |
999 | } | |
1000 | ||
5696c6e3 | 1001 | static int find_image_format(BlockBackend *file, const char *filename, |
34b5d2c6 | 1002 | BlockDriver **pdrv, Error **errp) |
f3a5d3f8 | 1003 | { |
c6684249 | 1004 | BlockDriver *drv; |
7cddd372 | 1005 | uint8_t buf[BLOCK_PROBE_BUF_SIZE]; |
f500a6d3 | 1006 | int ret = 0; |
f8ea0b00 | 1007 | |
bdb73476 EGE |
1008 | GLOBAL_STATE_CODE(); |
1009 | ||
08a00559 | 1010 | /* Return the raw BlockDriver * to scsi-generic devices or empty drives */ |
5696c6e3 | 1011 | if (blk_is_sg(file) || !blk_is_inserted(file) || blk_getlength(file) == 0) { |
ef810437 | 1012 | *pdrv = &bdrv_raw; |
c98ac35d | 1013 | return ret; |
1a396859 | 1014 | } |
f8ea0b00 | 1015 | |
a9262f55 | 1016 | ret = blk_pread(file, 0, sizeof(buf), buf, 0); |
83f64091 | 1017 | if (ret < 0) { |
34b5d2c6 HR |
1018 | error_setg_errno(errp, -ret, "Could not read image for determining its " |
1019 | "format"); | |
c98ac35d SW |
1020 | *pdrv = NULL; |
1021 | return ret; | |
83f64091 FB |
1022 | } |
1023 | ||
bf5b16fa | 1024 | drv = bdrv_probe_all(buf, sizeof(buf), filename); |
c98ac35d | 1025 | if (!drv) { |
34b5d2c6 HR |
1026 | error_setg(errp, "Could not determine image format: No compatible " |
1027 | "driver found"); | |
bf5b16fa AF |
1028 | *pdrv = NULL; |
1029 | return -ENOENT; | |
c98ac35d | 1030 | } |
bf5b16fa | 1031 | |
c98ac35d | 1032 | *pdrv = drv; |
bf5b16fa | 1033 | return 0; |
ea2384d3 FB |
1034 | } |
1035 | ||
51762288 SH |
1036 | /** |
1037 | * Set the current 'total_sectors' value | |
65a9bb25 | 1038 | * Return 0 on success, -errno on error. |
51762288 | 1039 | */ |
c86422c5 EGE |
1040 | int coroutine_fn bdrv_co_refresh_total_sectors(BlockDriverState *bs, |
1041 | int64_t hint) | |
51762288 SH |
1042 | { |
1043 | BlockDriver *drv = bs->drv; | |
967d7905 | 1044 | IO_CODE(); |
8ab8140a | 1045 | assert_bdrv_graph_readable(); |
51762288 | 1046 | |
d470ad42 HR |
1047 | if (!drv) { |
1048 | return -ENOMEDIUM; | |
1049 | } | |
1050 | ||
c86422c5 | 1051 | /* Do not attempt drv->bdrv_co_getlength() on scsi-generic devices */ |
b192af8a | 1052 | if (bdrv_is_sg(bs)) |
396759ad NB |
1053 | return 0; |
1054 | ||
51762288 | 1055 | /* query actual device if possible, otherwise just trust the hint */ |
c86422c5 EGE |
1056 | if (drv->bdrv_co_getlength) { |
1057 | int64_t length = drv->bdrv_co_getlength(bs); | |
51762288 SH |
1058 | if (length < 0) { |
1059 | return length; | |
1060 | } | |
7e382003 | 1061 | hint = DIV_ROUND_UP(length, BDRV_SECTOR_SIZE); |
51762288 SH |
1062 | } |
1063 | ||
1064 | bs->total_sectors = hint; | |
8b117001 VSO |
1065 | |
1066 | if (bs->total_sectors * BDRV_SECTOR_SIZE > BDRV_MAX_LENGTH) { | |
1067 | return -EFBIG; | |
1068 | } | |
1069 | ||
51762288 SH |
1070 | return 0; |
1071 | } | |
1072 | ||
cddff5ba KW |
1073 | /** |
1074 | * Combines a QDict of new block driver @options with any missing options taken | |
1075 | * from @old_options, so that leaving out an option defaults to its old value. | |
1076 | */ | |
1077 | static void bdrv_join_options(BlockDriverState *bs, QDict *options, | |
1078 | QDict *old_options) | |
1079 | { | |
da359909 | 1080 | GLOBAL_STATE_CODE(); |
cddff5ba KW |
1081 | if (bs->drv && bs->drv->bdrv_join_options) { |
1082 | bs->drv->bdrv_join_options(options, old_options); | |
1083 | } else { | |
1084 | qdict_join(options, old_options, false); | |
1085 | } | |
1086 | } | |
1087 | ||
543770bd AG |
1088 | static BlockdevDetectZeroesOptions bdrv_parse_detect_zeroes(QemuOpts *opts, |
1089 | int open_flags, | |
1090 | Error **errp) | |
1091 | { | |
1092 | Error *local_err = NULL; | |
1093 | char *value = qemu_opt_get_del(opts, "detect-zeroes"); | |
1094 | BlockdevDetectZeroesOptions detect_zeroes = | |
1095 | qapi_enum_parse(&BlockdevDetectZeroesOptions_lookup, value, | |
1096 | BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF, &local_err); | |
bdb73476 | 1097 | GLOBAL_STATE_CODE(); |
543770bd AG |
1098 | g_free(value); |
1099 | if (local_err) { | |
1100 | error_propagate(errp, local_err); | |
1101 | return detect_zeroes; | |
1102 | } | |
1103 | ||
1104 | if (detect_zeroes == BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP && | |
1105 | !(open_flags & BDRV_O_UNMAP)) | |
1106 | { | |
1107 | error_setg(errp, "setting detect-zeroes to unmap is not allowed " | |
1108 | "without setting discard operation to unmap"); | |
1109 | } | |
1110 | ||
1111 | return detect_zeroes; | |
1112 | } | |
1113 | ||
f80f2673 AM |
1114 | /** |
1115 | * Set open flags for aio engine | |
1116 | * | |
1117 | * Return 0 on success, -1 if the engine specified is invalid | |
1118 | */ | |
1119 | int bdrv_parse_aio(const char *mode, int *flags) | |
1120 | { | |
1121 | if (!strcmp(mode, "threads")) { | |
1122 | /* do nothing, default */ | |
1123 | } else if (!strcmp(mode, "native")) { | |
1124 | *flags |= BDRV_O_NATIVE_AIO; | |
1125 | #ifdef CONFIG_LINUX_IO_URING | |
1126 | } else if (!strcmp(mode, "io_uring")) { | |
1127 | *flags |= BDRV_O_IO_URING; | |
1128 | #endif | |
1129 | } else { | |
1130 | return -1; | |
1131 | } | |
1132 | ||
1133 | return 0; | |
1134 | } | |
1135 | ||
9e8f1835 PB |
1136 | /** |
1137 | * Set open flags for a given discard mode | |
1138 | * | |
1139 | * Return 0 on success, -1 if the discard mode was invalid. | |
1140 | */ | |
1141 | int bdrv_parse_discard_flags(const char *mode, int *flags) | |
1142 | { | |
1143 | *flags &= ~BDRV_O_UNMAP; | |
1144 | ||
1145 | if (!strcmp(mode, "off") || !strcmp(mode, "ignore")) { | |
1146 | /* do nothing */ | |
1147 | } else if (!strcmp(mode, "on") || !strcmp(mode, "unmap")) { | |
1148 | *flags |= BDRV_O_UNMAP; | |
1149 | } else { | |
1150 | return -1; | |
1151 | } | |
1152 | ||
1153 | return 0; | |
1154 | } | |
1155 | ||
c3993cdc SH |
1156 | /** |
1157 | * Set open flags for a given cache mode | |
1158 | * | |
1159 | * Return 0 on success, -1 if the cache mode was invalid. | |
1160 | */ | |
53e8ae01 | 1161 | int bdrv_parse_cache_mode(const char *mode, int *flags, bool *writethrough) |
c3993cdc SH |
1162 | { |
1163 | *flags &= ~BDRV_O_CACHE_MASK; | |
1164 | ||
1165 | if (!strcmp(mode, "off") || !strcmp(mode, "none")) { | |
53e8ae01 KW |
1166 | *writethrough = false; |
1167 | *flags |= BDRV_O_NOCACHE; | |
92196b2f | 1168 | } else if (!strcmp(mode, "directsync")) { |
53e8ae01 | 1169 | *writethrough = true; |
92196b2f | 1170 | *flags |= BDRV_O_NOCACHE; |
c3993cdc | 1171 | } else if (!strcmp(mode, "writeback")) { |
53e8ae01 | 1172 | *writethrough = false; |
c3993cdc | 1173 | } else if (!strcmp(mode, "unsafe")) { |
53e8ae01 | 1174 | *writethrough = false; |
c3993cdc SH |
1175 | *flags |= BDRV_O_NO_FLUSH; |
1176 | } else if (!strcmp(mode, "writethrough")) { | |
53e8ae01 | 1177 | *writethrough = true; |
c3993cdc SH |
1178 | } else { |
1179 | return -1; | |
1180 | } | |
1181 | ||
1182 | return 0; | |
1183 | } | |
1184 | ||
b5411555 KW |
1185 | static char *bdrv_child_get_parent_desc(BdrvChild *c) |
1186 | { | |
1187 | BlockDriverState *parent = c->opaque; | |
2c0a3acb | 1188 | return g_strdup_printf("node '%s'", bdrv_get_node_name(parent)); |
b5411555 KW |
1189 | } |
1190 | ||
20018e12 KW |
1191 | static void bdrv_child_cb_drained_begin(BdrvChild *child) |
1192 | { | |
1193 | BlockDriverState *bs = child->opaque; | |
a82a3bd1 | 1194 | bdrv_do_drained_begin_quiesce(bs, NULL); |
20018e12 KW |
1195 | } |
1196 | ||
89bd0305 KW |
1197 | static bool bdrv_child_cb_drained_poll(BdrvChild *child) |
1198 | { | |
1199 | BlockDriverState *bs = child->opaque; | |
299403ae | 1200 | return bdrv_drain_poll(bs, NULL, false); |
89bd0305 KW |
1201 | } |
1202 | ||
2f65df6e | 1203 | static void bdrv_child_cb_drained_end(BdrvChild *child) |
20018e12 KW |
1204 | { |
1205 | BlockDriverState *bs = child->opaque; | |
2f65df6e | 1206 | bdrv_drained_end(bs); |
20018e12 KW |
1207 | } |
1208 | ||
38701b6a KW |
1209 | static int bdrv_child_cb_inactivate(BdrvChild *child) |
1210 | { | |
1211 | BlockDriverState *bs = child->opaque; | |
bdb73476 | 1212 | GLOBAL_STATE_CODE(); |
38701b6a KW |
1213 | assert(bs->open_flags & BDRV_O_INACTIVE); |
1214 | return 0; | |
1215 | } | |
1216 | ||
27633e74 EGE |
1217 | static bool bdrv_child_cb_change_aio_ctx(BdrvChild *child, AioContext *ctx, |
1218 | GHashTable *visited, Transaction *tran, | |
1219 | Error **errp) | |
53a7d041 KW |
1220 | { |
1221 | BlockDriverState *bs = child->opaque; | |
27633e74 | 1222 | return bdrv_change_aio_context(bs, ctx, visited, tran, errp); |
53a7d041 KW |
1223 | } |
1224 | ||
b1e6fc08 | 1225 | /* |
73176bee KW |
1226 | * Returns the options and flags that a temporary snapshot should get, based on |
1227 | * the originally requested flags (the originally requested image will have | |
1228 | * flags like a backing file) | |
b1e6fc08 | 1229 | */ |
73176bee KW |
1230 | static void bdrv_temp_snapshot_options(int *child_flags, QDict *child_options, |
1231 | int parent_flags, QDict *parent_options) | |
b1e6fc08 | 1232 | { |
bdb73476 | 1233 | GLOBAL_STATE_CODE(); |
73176bee KW |
1234 | *child_flags = (parent_flags & ~BDRV_O_SNAPSHOT) | BDRV_O_TEMPORARY; |
1235 | ||
1236 | /* For temporary files, unconditional cache=unsafe is fine */ | |
73176bee KW |
1237 | qdict_set_default_str(child_options, BDRV_OPT_CACHE_DIRECT, "off"); |
1238 | qdict_set_default_str(child_options, BDRV_OPT_CACHE_NO_FLUSH, "on"); | |
41869044 | 1239 | |
3f48686f | 1240 | /* Copy the read-only and discard options from the parent */ |
f87a0e29 | 1241 | qdict_copy_default(child_options, parent_options, BDRV_OPT_READ_ONLY); |
3f48686f | 1242 | qdict_copy_default(child_options, parent_options, BDRV_OPT_DISCARD); |
f87a0e29 | 1243 | |
41869044 KW |
1244 | /* aio=native doesn't work for cache.direct=off, so disable it for the |
1245 | * temporary snapshot */ | |
1246 | *child_flags &= ~BDRV_O_NATIVE_AIO; | |
b1e6fc08 KW |
1247 | } |
1248 | ||
db95dbba KW |
1249 | static void bdrv_backing_attach(BdrvChild *c) |
1250 | { | |
1251 | BlockDriverState *parent = c->opaque; | |
1252 | BlockDriverState *backing_hd = c->bs; | |
1253 | ||
bdb73476 | 1254 | GLOBAL_STATE_CODE(); |
db95dbba KW |
1255 | assert(!parent->backing_blocker); |
1256 | error_setg(&parent->backing_blocker, | |
1257 | "node is used as backing hd of '%s'", | |
1258 | bdrv_get_device_or_node_name(parent)); | |
1259 | ||
f30c66ba HR |
1260 | bdrv_refresh_filename(backing_hd); |
1261 | ||
db95dbba | 1262 | parent->open_flags &= ~BDRV_O_NO_BACKING; |
db95dbba KW |
1263 | |
1264 | bdrv_op_block_all(backing_hd, parent->backing_blocker); | |
1265 | /* Otherwise we won't be able to commit or stream */ | |
1266 | bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_COMMIT_TARGET, | |
1267 | parent->backing_blocker); | |
1268 | bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_STREAM, | |
1269 | parent->backing_blocker); | |
1270 | /* | |
1271 | * We do backup in 3 ways: | |
1272 | * 1. drive backup | |
1273 | * The target bs is new opened, and the source is top BDS | |
1274 | * 2. blockdev backup | |
1275 | * Both the source and the target are top BDSes. | |
1276 | * 3. internal backup(used for block replication) | |
1277 | * Both the source and the target are backing file | |
1278 | * | |
1279 | * In case 1 and 2, neither the source nor the target is the backing file. | |
1280 | * In case 3, we will block the top BDS, so there is only one block job | |
1281 | * for the top BDS and its backing chain. | |
1282 | */ | |
1283 | bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_BACKUP_SOURCE, | |
1284 | parent->backing_blocker); | |
1285 | bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_BACKUP_TARGET, | |
1286 | parent->backing_blocker); | |
ca2f1234 | 1287 | } |
d736f119 | 1288 | |
db95dbba KW |
1289 | static void bdrv_backing_detach(BdrvChild *c) |
1290 | { | |
1291 | BlockDriverState *parent = c->opaque; | |
1292 | ||
bdb73476 | 1293 | GLOBAL_STATE_CODE(); |
db95dbba KW |
1294 | assert(parent->backing_blocker); |
1295 | bdrv_op_unblock_all(c->bs, parent->backing_blocker); | |
1296 | error_free(parent->backing_blocker); | |
1297 | parent->backing_blocker = NULL; | |
48e08288 | 1298 | } |
d736f119 | 1299 | |
6858eba0 KW |
1300 | static int bdrv_backing_update_filename(BdrvChild *c, BlockDriverState *base, |
1301 | const char *filename, Error **errp) | |
1302 | { | |
1303 | BlockDriverState *parent = c->opaque; | |
e94d3dba | 1304 | bool read_only = bdrv_is_read_only(parent); |
6858eba0 | 1305 | int ret; |
bdb73476 | 1306 | GLOBAL_STATE_CODE(); |
6858eba0 | 1307 | |
e94d3dba AG |
1308 | if (read_only) { |
1309 | ret = bdrv_reopen_set_read_only(parent, false, errp); | |
61f09cea KW |
1310 | if (ret < 0) { |
1311 | return ret; | |
1312 | } | |
1313 | } | |
1314 | ||
6858eba0 | 1315 | ret = bdrv_change_backing_file(parent, filename, |
e54ee1b3 EB |
1316 | base->drv ? base->drv->format_name : "", |
1317 | false); | |
6858eba0 | 1318 | if (ret < 0) { |
64730694 | 1319 | error_setg_errno(errp, -ret, "Could not update backing file link"); |
6858eba0 KW |
1320 | } |
1321 | ||
e94d3dba AG |
1322 | if (read_only) { |
1323 | bdrv_reopen_set_read_only(parent, true, NULL); | |
61f09cea KW |
1324 | } |
1325 | ||
6858eba0 KW |
1326 | return ret; |
1327 | } | |
1328 | ||
fae8bd39 HR |
1329 | /* |
1330 | * Returns the options and flags that a generic child of a BDS should | |
1331 | * get, based on the given options and flags for the parent BDS. | |
1332 | */ | |
00ff7ffd HR |
1333 | static void bdrv_inherited_options(BdrvChildRole role, bool parent_is_format, |
1334 | int *child_flags, QDict *child_options, | |
1335 | int parent_flags, QDict *parent_options) | |
fae8bd39 HR |
1336 | { |
1337 | int flags = parent_flags; | |
bdb73476 | 1338 | GLOBAL_STATE_CODE(); |
fae8bd39 HR |
1339 | |
1340 | /* | |
1341 | * First, decide whether to set, clear, or leave BDRV_O_PROTOCOL. | |
1342 | * Generally, the question to answer is: Should this child be | |
1343 | * format-probed by default? | |
1344 | */ | |
1345 | ||
1346 | /* | |
1347 | * Pure and non-filtered data children of non-format nodes should | |
1348 | * be probed by default (even when the node itself has BDRV_O_PROTOCOL | |
1349 | * set). This only affects a very limited set of drivers (namely | |
1350 | * quorum and blkverify when this comment was written). | |
1351 | * Force-clear BDRV_O_PROTOCOL then. | |
1352 | */ | |
1353 | if (!parent_is_format && | |
1354 | (role & BDRV_CHILD_DATA) && | |
1355 | !(role & (BDRV_CHILD_METADATA | BDRV_CHILD_FILTERED))) | |
1356 | { | |
1357 | flags &= ~BDRV_O_PROTOCOL; | |
1358 | } | |
1359 | ||
1360 | /* | |
1361 | * All children of format nodes (except for COW children) and all | |
1362 | * metadata children in general should never be format-probed. | |
1363 | * Force-set BDRV_O_PROTOCOL then. | |
1364 | */ | |
1365 | if ((parent_is_format && !(role & BDRV_CHILD_COW)) || | |
1366 | (role & BDRV_CHILD_METADATA)) | |
1367 | { | |
1368 | flags |= BDRV_O_PROTOCOL; | |
1369 | } | |
1370 | ||
1371 | /* | |
1372 | * If the cache mode isn't explicitly set, inherit direct and no-flush from | |
1373 | * the parent. | |
1374 | */ | |
1375 | qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_DIRECT); | |
1376 | qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_NO_FLUSH); | |
1377 | qdict_copy_default(child_options, parent_options, BDRV_OPT_FORCE_SHARE); | |
1378 | ||
1379 | if (role & BDRV_CHILD_COW) { | |
1380 | /* backing files are opened read-only by default */ | |
1381 | qdict_set_default_str(child_options, BDRV_OPT_READ_ONLY, "on"); | |
1382 | qdict_set_default_str(child_options, BDRV_OPT_AUTO_READ_ONLY, "off"); | |
1383 | } else { | |
1384 | /* Inherit the read-only option from the parent if it's not set */ | |
1385 | qdict_copy_default(child_options, parent_options, BDRV_OPT_READ_ONLY); | |
1386 | qdict_copy_default(child_options, parent_options, | |
1387 | BDRV_OPT_AUTO_READ_ONLY); | |
1388 | } | |
1389 | ||
1390 | /* | |
1391 | * bdrv_co_pdiscard() respects unmap policy for the parent, so we | |
1392 | * can default to enable it on lower layers regardless of the | |
1393 | * parent option. | |
1394 | */ | |
1395 | qdict_set_default_str(child_options, BDRV_OPT_DISCARD, "unmap"); | |
1396 | ||
1397 | /* Clear flags that only apply to the top layer */ | |
1398 | flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_COPY_ON_READ); | |
1399 | ||
1400 | if (role & BDRV_CHILD_METADATA) { | |
1401 | flags &= ~BDRV_O_NO_IO; | |
1402 | } | |
1403 | if (role & BDRV_CHILD_COW) { | |
1404 | flags &= ~BDRV_O_TEMPORARY; | |
1405 | } | |
1406 | ||
1407 | *child_flags = flags; | |
1408 | } | |
1409 | ||
303de47b | 1410 | static void GRAPH_WRLOCK bdrv_child_cb_attach(BdrvChild *child) |
ca2f1234 HR |
1411 | { |
1412 | BlockDriverState *bs = child->opaque; | |
1413 | ||
3f35f82e | 1414 | assert_bdrv_graph_writable(); |
a225369b | 1415 | QLIST_INSERT_HEAD(&bs->children, child, next); |
5bb04747 VSO |
1416 | if (bs->drv->is_filter || (child->role & BDRV_CHILD_FILTERED)) { |
1417 | /* | |
1418 | * Here we handle filters and block/raw-format.c when it behave like | |
1419 | * filter. They generally have a single PRIMARY child, which is also the | |
1420 | * FILTERED child, and that they may have multiple more children, which | |
1421 | * are neither PRIMARY nor FILTERED. And never we have a COW child here. | |
1422 | * So bs->file will be the PRIMARY child, unless the PRIMARY child goes | |
1423 | * into bs->backing on exceptional cases; and bs->backing will be | |
1424 | * nothing else. | |
1425 | */ | |
1426 | assert(!(child->role & BDRV_CHILD_COW)); | |
1427 | if (child->role & BDRV_CHILD_PRIMARY) { | |
1428 | assert(child->role & BDRV_CHILD_FILTERED); | |
1429 | assert(!bs->backing); | |
1430 | assert(!bs->file); | |
1431 | ||
1432 | if (bs->drv->filtered_child_is_backing) { | |
1433 | bs->backing = child; | |
1434 | } else { | |
1435 | bs->file = child; | |
1436 | } | |
1437 | } else { | |
1438 | assert(!(child->role & BDRV_CHILD_FILTERED)); | |
1439 | } | |
1440 | } else if (child->role & BDRV_CHILD_COW) { | |
1441 | assert(bs->drv->supports_backing); | |
1442 | assert(!(child->role & BDRV_CHILD_PRIMARY)); | |
1443 | assert(!bs->backing); | |
1444 | bs->backing = child; | |
ca2f1234 | 1445 | bdrv_backing_attach(child); |
5bb04747 VSO |
1446 | } else if (child->role & BDRV_CHILD_PRIMARY) { |
1447 | assert(!bs->file); | |
1448 | bs->file = child; | |
ca2f1234 | 1449 | } |
ca2f1234 HR |
1450 | } |
1451 | ||
303de47b | 1452 | static void GRAPH_WRLOCK bdrv_child_cb_detach(BdrvChild *child) |
48e08288 HR |
1453 | { |
1454 | BlockDriverState *bs = child->opaque; | |
1455 | ||
1456 | if (child->role & BDRV_CHILD_COW) { | |
1457 | bdrv_backing_detach(child); | |
1458 | } | |
1459 | ||
3f35f82e | 1460 | assert_bdrv_graph_writable(); |
a225369b | 1461 | QLIST_REMOVE(child, next); |
5bb04747 VSO |
1462 | if (child == bs->backing) { |
1463 | assert(child != bs->file); | |
1464 | bs->backing = NULL; | |
1465 | } else if (child == bs->file) { | |
1466 | bs->file = NULL; | |
1467 | } | |
48e08288 HR |
1468 | } |
1469 | ||
43483550 HR |
1470 | static int bdrv_child_cb_update_filename(BdrvChild *c, BlockDriverState *base, |
1471 | const char *filename, Error **errp) | |
1472 | { | |
1473 | if (c->role & BDRV_CHILD_COW) { | |
1474 | return bdrv_backing_update_filename(c, base, filename, errp); | |
1475 | } | |
1476 | return 0; | |
1477 | } | |
1478 | ||
fb62b588 | 1479 | AioContext *child_of_bds_get_parent_aio_context(BdrvChild *c) |
3ca1f322 VSO |
1480 | { |
1481 | BlockDriverState *bs = c->opaque; | |
384a48fb | 1482 | IO_CODE(); |
3ca1f322 VSO |
1483 | |
1484 | return bdrv_get_aio_context(bs); | |
1485 | } | |
1486 | ||
43483550 HR |
1487 | const BdrvChildClass child_of_bds = { |
1488 | .parent_is_bds = true, | |
1489 | .get_parent_desc = bdrv_child_get_parent_desc, | |
1490 | .inherit_options = bdrv_inherited_options, | |
1491 | .drained_begin = bdrv_child_cb_drained_begin, | |
1492 | .drained_poll = bdrv_child_cb_drained_poll, | |
1493 | .drained_end = bdrv_child_cb_drained_end, | |
1494 | .attach = bdrv_child_cb_attach, | |
1495 | .detach = bdrv_child_cb_detach, | |
1496 | .inactivate = bdrv_child_cb_inactivate, | |
27633e74 | 1497 | .change_aio_ctx = bdrv_child_cb_change_aio_ctx, |
43483550 | 1498 | .update_filename = bdrv_child_cb_update_filename, |
fb62b588 | 1499 | .get_parent_aio_context = child_of_bds_get_parent_aio_context, |
43483550 HR |
1500 | }; |
1501 | ||
3ca1f322 VSO |
1502 | AioContext *bdrv_child_get_parent_aio_context(BdrvChild *c) |
1503 | { | |
d5f8d79c | 1504 | IO_CODE(); |
3ca1f322 VSO |
1505 | return c->klass->get_parent_aio_context(c); |
1506 | } | |
1507 | ||
7b272452 KW |
1508 | static int bdrv_open_flags(BlockDriverState *bs, int flags) |
1509 | { | |
61de4c68 | 1510 | int open_flags = flags; |
bdb73476 | 1511 | GLOBAL_STATE_CODE(); |
7b272452 KW |
1512 | |
1513 | /* | |
1514 | * Clear flags that are internal to the block layer before opening the | |
1515 | * image. | |
1516 | */ | |
20cca275 | 1517 | open_flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_PROTOCOL); |
7b272452 | 1518 | |
7b272452 KW |
1519 | return open_flags; |
1520 | } | |
1521 | ||
91a097e7 KW |
1522 | static void update_flags_from_options(int *flags, QemuOpts *opts) |
1523 | { | |
bdb73476 EGE |
1524 | GLOBAL_STATE_CODE(); |
1525 | ||
2a3d4331 | 1526 | *flags &= ~(BDRV_O_CACHE_MASK | BDRV_O_RDWR | BDRV_O_AUTO_RDONLY); |
91a097e7 | 1527 | |
57f9db9a | 1528 | if (qemu_opt_get_bool_del(opts, BDRV_OPT_CACHE_NO_FLUSH, false)) { |
91a097e7 KW |
1529 | *flags |= BDRV_O_NO_FLUSH; |
1530 | } | |
1531 | ||
57f9db9a | 1532 | if (qemu_opt_get_bool_del(opts, BDRV_OPT_CACHE_DIRECT, false)) { |
91a097e7 KW |
1533 | *flags |= BDRV_O_NOCACHE; |
1534 | } | |
f87a0e29 | 1535 | |
57f9db9a | 1536 | if (!qemu_opt_get_bool_del(opts, BDRV_OPT_READ_ONLY, false)) { |
f87a0e29 AG |
1537 | *flags |= BDRV_O_RDWR; |
1538 | } | |
1539 | ||
e35bdc12 KW |
1540 | if (qemu_opt_get_bool_del(opts, BDRV_OPT_AUTO_READ_ONLY, false)) { |
1541 | *flags |= BDRV_O_AUTO_RDONLY; | |
1542 | } | |
91a097e7 KW |
1543 | } |
1544 | ||
1545 | static void update_options_from_flags(QDict *options, int flags) | |
1546 | { | |
bdb73476 | 1547 | GLOBAL_STATE_CODE(); |
91a097e7 | 1548 | if (!qdict_haskey(options, BDRV_OPT_CACHE_DIRECT)) { |
46f5ac20 | 1549 | qdict_put_bool(options, BDRV_OPT_CACHE_DIRECT, flags & BDRV_O_NOCACHE); |
91a097e7 KW |
1550 | } |
1551 | if (!qdict_haskey(options, BDRV_OPT_CACHE_NO_FLUSH)) { | |
46f5ac20 EB |
1552 | qdict_put_bool(options, BDRV_OPT_CACHE_NO_FLUSH, |
1553 | flags & BDRV_O_NO_FLUSH); | |
91a097e7 | 1554 | } |
f87a0e29 | 1555 | if (!qdict_haskey(options, BDRV_OPT_READ_ONLY)) { |
46f5ac20 | 1556 | qdict_put_bool(options, BDRV_OPT_READ_ONLY, !(flags & BDRV_O_RDWR)); |
f87a0e29 | 1557 | } |
e35bdc12 KW |
1558 | if (!qdict_haskey(options, BDRV_OPT_AUTO_READ_ONLY)) { |
1559 | qdict_put_bool(options, BDRV_OPT_AUTO_READ_ONLY, | |
1560 | flags & BDRV_O_AUTO_RDONLY); | |
1561 | } | |
91a097e7 KW |
1562 | } |
1563 | ||
636ea370 KW |
1564 | static void bdrv_assign_node_name(BlockDriverState *bs, |
1565 | const char *node_name, | |
1566 | Error **errp) | |
6913c0c2 | 1567 | { |
15489c76 | 1568 | char *gen_node_name = NULL; |
bdb73476 | 1569 | GLOBAL_STATE_CODE(); |
6913c0c2 | 1570 | |
15489c76 JC |
1571 | if (!node_name) { |
1572 | node_name = gen_node_name = id_generate(ID_BLOCK); | |
1573 | } else if (!id_wellformed(node_name)) { | |
1574 | /* | |
1575 | * Check for empty string or invalid characters, but not if it is | |
1576 | * generated (generated names use characters not available to the user) | |
1577 | */ | |
785ec4b1 | 1578 | error_setg(errp, "Invalid node-name: '%s'", node_name); |
636ea370 | 1579 | return; |
6913c0c2 BC |
1580 | } |
1581 | ||
0c5e94ee | 1582 | /* takes care of avoiding namespaces collisions */ |
7f06d47e | 1583 | if (blk_by_name(node_name)) { |
0c5e94ee BC |
1584 | error_setg(errp, "node-name=%s is conflicting with a device id", |
1585 | node_name); | |
15489c76 | 1586 | goto out; |
0c5e94ee BC |
1587 | } |
1588 | ||
6913c0c2 BC |
1589 | /* takes care of avoiding duplicates node names */ |
1590 | if (bdrv_find_node(node_name)) { | |
785ec4b1 | 1591 | error_setg(errp, "Duplicate nodes with node-name='%s'", node_name); |
15489c76 | 1592 | goto out; |
6913c0c2 BC |
1593 | } |
1594 | ||
824808dd KW |
1595 | /* Make sure that the node name isn't truncated */ |
1596 | if (strlen(node_name) >= sizeof(bs->node_name)) { | |
1597 | error_setg(errp, "Node name too long"); | |
1598 | goto out; | |
1599 | } | |
1600 | ||
6913c0c2 BC |
1601 | /* copy node name into the bs and insert it into the graph list */ |
1602 | pstrcpy(bs->node_name, sizeof(bs->node_name), node_name); | |
1603 | QTAILQ_INSERT_TAIL(&graph_bdrv_states, bs, node_list); | |
15489c76 JC |
1604 | out: |
1605 | g_free(gen_node_name); | |
6913c0c2 BC |
1606 | } |
1607 | ||
c86422c5 EGE |
1608 | /* |
1609 | * The caller must always hold @bs AioContext lock, because this function calls | |
1610 | * bdrv_refresh_total_sectors() which polls when called from non-coroutine | |
1611 | * context. | |
1612 | */ | |
1a30b0f5 KW |
1613 | static int no_coroutine_fn GRAPH_UNLOCKED |
1614 | bdrv_open_driver(BlockDriverState *bs, BlockDriver *drv, const char *node_name, | |
1615 | QDict *options, int open_flags, Error **errp) | |
01a56501 | 1616 | { |
8dc8a60c | 1617 | AioContext *ctx; |
01a56501 | 1618 | Error *local_err = NULL; |
0f12264e | 1619 | int i, ret; |
da359909 | 1620 | GLOBAL_STATE_CODE(); |
01a56501 KW |
1621 | |
1622 | bdrv_assign_node_name(bs, node_name, &local_err); | |
1623 | if (local_err) { | |
1624 | error_propagate(errp, local_err); | |
1625 | return -EINVAL; | |
1626 | } | |
1627 | ||
1628 | bs->drv = drv; | |
1629 | bs->opaque = g_malloc0(drv->instance_size); | |
1630 | ||
1631 | if (drv->bdrv_file_open) { | |
1632 | assert(!drv->bdrv_needs_filename || bs->filename[0]); | |
1633 | ret = drv->bdrv_file_open(bs, options, open_flags, &local_err); | |
680c7f96 | 1634 | } else if (drv->bdrv_open) { |
01a56501 | 1635 | ret = drv->bdrv_open(bs, options, open_flags, &local_err); |
680c7f96 KW |
1636 | } else { |
1637 | ret = 0; | |
01a56501 KW |
1638 | } |
1639 | ||
1640 | if (ret < 0) { | |
1641 | if (local_err) { | |
1642 | error_propagate(errp, local_err); | |
1643 | } else if (bs->filename[0]) { | |
1644 | error_setg_errno(errp, -ret, "Could not open '%s'", bs->filename); | |
1645 | } else { | |
1646 | error_setg_errno(errp, -ret, "Could not open image"); | |
1647 | } | |
180ca19a | 1648 | goto open_failed; |
01a56501 KW |
1649 | } |
1650 | ||
e8b65355 SH |
1651 | assert(!(bs->supported_read_flags & ~BDRV_REQ_MASK)); |
1652 | assert(!(bs->supported_write_flags & ~BDRV_REQ_MASK)); | |
1653 | ||
1654 | /* | |
1655 | * Always allow the BDRV_REQ_REGISTERED_BUF optimization hint. This saves | |
1656 | * drivers that pass read/write requests through to a child the trouble of | |
1657 | * declaring support explicitly. | |
1658 | * | |
1659 | * Drivers must not propagate this flag accidentally when they initiate I/O | |
1660 | * to a bounce buffer. That case should be rare though. | |
1661 | */ | |
1662 | bs->supported_read_flags |= BDRV_REQ_REGISTERED_BUF; | |
1663 | bs->supported_write_flags |= BDRV_REQ_REGISTERED_BUF; | |
1664 | ||
8dc8a60c KW |
1665 | /* Get the context after .bdrv_open, it can change the context */ |
1666 | ctx = bdrv_get_aio_context(bs); | |
1667 | aio_context_acquire(ctx); | |
1668 | ||
c057960c | 1669 | ret = bdrv_refresh_total_sectors(bs, bs->total_sectors); |
01a56501 KW |
1670 | if (ret < 0) { |
1671 | error_setg_errno(errp, -ret, "Could not refresh total sector count"); | |
8dc8a60c | 1672 | aio_context_release(ctx); |
180ca19a | 1673 | return ret; |
01a56501 KW |
1674 | } |
1675 | ||
e19b157f | 1676 | bdrv_graph_rdlock_main_loop(); |
1e4c797c | 1677 | bdrv_refresh_limits(bs, NULL, &local_err); |
e19b157f | 1678 | bdrv_graph_rdunlock_main_loop(); |
8dc8a60c | 1679 | aio_context_release(ctx); |
e19b157f | 1680 | |
01a56501 KW |
1681 | if (local_err) { |
1682 | error_propagate(errp, local_err); | |
180ca19a | 1683 | return -EINVAL; |
01a56501 KW |
1684 | } |
1685 | ||
1686 | assert(bdrv_opt_mem_align(bs) != 0); | |
1687 | assert(bdrv_min_mem_align(bs) != 0); | |
1688 | assert(is_power_of_2(bs->bl.request_alignment)); | |
1689 | ||
0f12264e | 1690 | for (i = 0; i < bs->quiesce_counter; i++) { |
5e8ac217 KW |
1691 | if (drv->bdrv_drain_begin) { |
1692 | drv->bdrv_drain_begin(bs); | |
0f12264e KW |
1693 | } |
1694 | } | |
1695 | ||
01a56501 | 1696 | return 0; |
180ca19a MP |
1697 | open_failed: |
1698 | bs->drv = NULL; | |
1699 | if (bs->file != NULL) { | |
1700 | bdrv_unref_child(bs, bs->file); | |
5bb04747 | 1701 | assert(!bs->file); |
180ca19a | 1702 | } |
01a56501 KW |
1703 | g_free(bs->opaque); |
1704 | bs->opaque = NULL; | |
01a56501 KW |
1705 | return ret; |
1706 | } | |
1707 | ||
621d1737 VSO |
1708 | /* |
1709 | * Create and open a block node. | |
1710 | * | |
1711 | * @options is a QDict of options to pass to the block drivers, or NULL for an | |
1712 | * empty set of options. The reference to the QDict belongs to the block layer | |
1713 | * after the call (even on failure), so if the caller intends to reuse the | |
1714 | * dictionary, it needs to use qobject_ref() before calling bdrv_open. | |
1715 | */ | |
1716 | BlockDriverState *bdrv_new_open_driver_opts(BlockDriver *drv, | |
1717 | const char *node_name, | |
1718 | QDict *options, int flags, | |
1719 | Error **errp) | |
680c7f96 KW |
1720 | { |
1721 | BlockDriverState *bs; | |
1722 | int ret; | |
1723 | ||
f791bf7f EGE |
1724 | GLOBAL_STATE_CODE(); |
1725 | ||
680c7f96 KW |
1726 | bs = bdrv_new(); |
1727 | bs->open_flags = flags; | |
621d1737 VSO |
1728 | bs->options = options ?: qdict_new(); |
1729 | bs->explicit_options = qdict_clone_shallow(bs->options); | |
680c7f96 KW |
1730 | bs->opaque = NULL; |
1731 | ||
1732 | update_options_from_flags(bs->options, flags); | |
1733 | ||
1734 | ret = bdrv_open_driver(bs, drv, node_name, bs->options, flags, errp); | |
1735 | if (ret < 0) { | |
cb3e7f08 | 1736 | qobject_unref(bs->explicit_options); |
180ca19a | 1737 | bs->explicit_options = NULL; |
cb3e7f08 | 1738 | qobject_unref(bs->options); |
180ca19a | 1739 | bs->options = NULL; |
680c7f96 KW |
1740 | bdrv_unref(bs); |
1741 | return NULL; | |
1742 | } | |
1743 | ||
1744 | return bs; | |
1745 | } | |
1746 | ||
621d1737 VSO |
1747 | /* Create and open a block node. */ |
1748 | BlockDriverState *bdrv_new_open_driver(BlockDriver *drv, const char *node_name, | |
1749 | int flags, Error **errp) | |
1750 | { | |
f791bf7f | 1751 | GLOBAL_STATE_CODE(); |
621d1737 VSO |
1752 | return bdrv_new_open_driver_opts(drv, node_name, NULL, flags, errp); |
1753 | } | |
1754 | ||
c5f3014b | 1755 | QemuOptsList bdrv_runtime_opts = { |
18edf289 KW |
1756 | .name = "bdrv_common", |
1757 | .head = QTAILQ_HEAD_INITIALIZER(bdrv_runtime_opts.head), | |
1758 | .desc = { | |
1759 | { | |
1760 | .name = "node-name", | |
1761 | .type = QEMU_OPT_STRING, | |
1762 | .help = "Node name of the block device node", | |
1763 | }, | |
62392ebb KW |
1764 | { |
1765 | .name = "driver", | |
1766 | .type = QEMU_OPT_STRING, | |
1767 | .help = "Block driver to use for the node", | |
1768 | }, | |
91a097e7 KW |
1769 | { |
1770 | .name = BDRV_OPT_CACHE_DIRECT, | |
1771 | .type = QEMU_OPT_BOOL, | |
1772 | .help = "Bypass software writeback cache on the host", | |
1773 | }, | |
1774 | { | |
1775 | .name = BDRV_OPT_CACHE_NO_FLUSH, | |
1776 | .type = QEMU_OPT_BOOL, | |
1777 | .help = "Ignore flush requests", | |
1778 | }, | |
f87a0e29 AG |
1779 | { |
1780 | .name = BDRV_OPT_READ_ONLY, | |
1781 | .type = QEMU_OPT_BOOL, | |
1782 | .help = "Node is opened in read-only mode", | |
1783 | }, | |
e35bdc12 KW |
1784 | { |
1785 | .name = BDRV_OPT_AUTO_READ_ONLY, | |
1786 | .type = QEMU_OPT_BOOL, | |
1787 | .help = "Node can become read-only if opening read-write fails", | |
1788 | }, | |
692e01a2 KW |
1789 | { |
1790 | .name = "detect-zeroes", | |
1791 | .type = QEMU_OPT_STRING, | |
1792 | .help = "try to optimize zero writes (off, on, unmap)", | |
1793 | }, | |
818584a4 | 1794 | { |
415bbca8 | 1795 | .name = BDRV_OPT_DISCARD, |
818584a4 KW |
1796 | .type = QEMU_OPT_STRING, |
1797 | .help = "discard operation (ignore/off, unmap/on)", | |
1798 | }, | |
5a9347c6 FZ |
1799 | { |
1800 | .name = BDRV_OPT_FORCE_SHARE, | |
1801 | .type = QEMU_OPT_BOOL, | |
1802 | .help = "always accept other writers (default: off)", | |
1803 | }, | |
18edf289 KW |
1804 | { /* end of list */ } |
1805 | }, | |
1806 | }; | |
1807 | ||
5a5e7f8c ML |
1808 | QemuOptsList bdrv_create_opts_simple = { |
1809 | .name = "simple-create-opts", | |
1810 | .head = QTAILQ_HEAD_INITIALIZER(bdrv_create_opts_simple.head), | |
fd17146c HR |
1811 | .desc = { |
1812 | { | |
1813 | .name = BLOCK_OPT_SIZE, | |
1814 | .type = QEMU_OPT_SIZE, | |
1815 | .help = "Virtual disk size" | |
1816 | }, | |
1817 | { | |
1818 | .name = BLOCK_OPT_PREALLOC, | |
1819 | .type = QEMU_OPT_STRING, | |
1820 | .help = "Preallocation mode (allowed values: off)" | |
1821 | }, | |
1822 | { /* end of list */ } | |
1823 | } | |
1824 | }; | |
1825 | ||
57915332 KW |
1826 | /* |
1827 | * Common part for opening disk images and files | |
b6ad491a KW |
1828 | * |
1829 | * Removes all processed options from *options. | |
57915332 | 1830 | */ |
5696c6e3 | 1831 | static int bdrv_open_common(BlockDriverState *bs, BlockBackend *file, |
82dc8b41 | 1832 | QDict *options, Error **errp) |
57915332 KW |
1833 | { |
1834 | int ret, open_flags; | |
035fccdf | 1835 | const char *filename; |
62392ebb | 1836 | const char *driver_name = NULL; |
6913c0c2 | 1837 | const char *node_name = NULL; |
818584a4 | 1838 | const char *discard; |
18edf289 | 1839 | QemuOpts *opts; |
62392ebb | 1840 | BlockDriver *drv; |
34b5d2c6 | 1841 | Error *local_err = NULL; |
307261b2 | 1842 | bool ro; |
57915332 | 1843 | |
6405875c | 1844 | assert(bs->file == NULL); |
707ff828 | 1845 | assert(options != NULL && bs->options != options); |
bdb73476 | 1846 | GLOBAL_STATE_CODE(); |
57915332 | 1847 | |
62392ebb | 1848 | opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort); |
af175e85 | 1849 | if (!qemu_opts_absorb_qdict(opts, options, errp)) { |
62392ebb KW |
1850 | ret = -EINVAL; |
1851 | goto fail_opts; | |
1852 | } | |
1853 | ||
9b7e8691 AG |
1854 | update_flags_from_options(&bs->open_flags, opts); |
1855 | ||
62392ebb KW |
1856 | driver_name = qemu_opt_get(opts, "driver"); |
1857 | drv = bdrv_find_format(driver_name); | |
1858 | assert(drv != NULL); | |
1859 | ||
5a9347c6 FZ |
1860 | bs->force_share = qemu_opt_get_bool(opts, BDRV_OPT_FORCE_SHARE, false); |
1861 | ||
1862 | if (bs->force_share && (bs->open_flags & BDRV_O_RDWR)) { | |
1863 | error_setg(errp, | |
1864 | BDRV_OPT_FORCE_SHARE | |
1865 | "=on can only be used with read-only images"); | |
1866 | ret = -EINVAL; | |
1867 | goto fail_opts; | |
1868 | } | |
1869 | ||
45673671 | 1870 | if (file != NULL) { |
f30c66ba | 1871 | bdrv_refresh_filename(blk_bs(file)); |
5696c6e3 | 1872 | filename = blk_bs(file)->filename; |
45673671 | 1873 | } else { |
129c7d1c MA |
1874 | /* |
1875 | * Caution: while qdict_get_try_str() is fine, getting | |
1876 | * non-string types would require more care. When @options | |
1877 | * come from -blockdev or blockdev_add, its members are typed | |
1878 | * according to the QAPI schema, but when they come from | |
1879 | * -drive, they're all QString. | |
1880 | */ | |
45673671 KW |
1881 | filename = qdict_get_try_str(options, "filename"); |
1882 | } | |
1883 | ||
4a008240 | 1884 | if (drv->bdrv_needs_filename && (!filename || !filename[0])) { |
765003db KW |
1885 | error_setg(errp, "The '%s' block driver requires a file name", |
1886 | drv->format_name); | |
18edf289 KW |
1887 | ret = -EINVAL; |
1888 | goto fail_opts; | |
6913c0c2 | 1889 | } |
6913c0c2 | 1890 | |
82dc8b41 KW |
1891 | trace_bdrv_open_common(bs, filename ?: "", bs->open_flags, |
1892 | drv->format_name); | |
62392ebb | 1893 | |
307261b2 VSO |
1894 | ro = bdrv_is_read_only(bs); |
1895 | ||
1896 | if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv, ro)) { | |
1897 | if (!ro && bdrv_is_whitelisted(drv, true)) { | |
8be25de6 KW |
1898 | ret = bdrv_apply_auto_read_only(bs, NULL, NULL); |
1899 | } else { | |
1900 | ret = -ENOTSUP; | |
1901 | } | |
1902 | if (ret < 0) { | |
1903 | error_setg(errp, | |
307261b2 | 1904 | !ro && bdrv_is_whitelisted(drv, true) |
8be25de6 KW |
1905 | ? "Driver '%s' can only be used for read-only devices" |
1906 | : "Driver '%s' is not whitelisted", | |
1907 | drv->format_name); | |
1908 | goto fail_opts; | |
1909 | } | |
b64ec4e4 | 1910 | } |
57915332 | 1911 | |
d3faa13e | 1912 | /* bdrv_new() and bdrv_close() make it so */ |
d73415a3 | 1913 | assert(qatomic_read(&bs->copy_on_read) == 0); |
d3faa13e | 1914 | |
82dc8b41 | 1915 | if (bs->open_flags & BDRV_O_COPY_ON_READ) { |
307261b2 | 1916 | if (!ro) { |
0ebd24e0 KW |
1917 | bdrv_enable_copy_on_read(bs); |
1918 | } else { | |
1919 | error_setg(errp, "Can't use copy-on-read on read-only device"); | |
18edf289 KW |
1920 | ret = -EINVAL; |
1921 | goto fail_opts; | |
0ebd24e0 | 1922 | } |
53fec9d3 SH |
1923 | } |
1924 | ||
415bbca8 | 1925 | discard = qemu_opt_get(opts, BDRV_OPT_DISCARD); |
818584a4 KW |
1926 | if (discard != NULL) { |
1927 | if (bdrv_parse_discard_flags(discard, &bs->open_flags) != 0) { | |
1928 | error_setg(errp, "Invalid discard option"); | |
1929 | ret = -EINVAL; | |
1930 | goto fail_opts; | |
1931 | } | |
1932 | } | |
1933 | ||
543770bd AG |
1934 | bs->detect_zeroes = |
1935 | bdrv_parse_detect_zeroes(opts, bs->open_flags, &local_err); | |
1936 | if (local_err) { | |
1937 | error_propagate(errp, local_err); | |
1938 | ret = -EINVAL; | |
1939 | goto fail_opts; | |
692e01a2 KW |
1940 | } |
1941 | ||
c2ad1b0c KW |
1942 | if (filename != NULL) { |
1943 | pstrcpy(bs->filename, sizeof(bs->filename), filename); | |
1944 | } else { | |
1945 | bs->filename[0] = '\0'; | |
1946 | } | |
91af7014 | 1947 | pstrcpy(bs->exact_filename, sizeof(bs->exact_filename), bs->filename); |
57915332 | 1948 | |
66f82cee | 1949 | /* Open the image, either directly or using a protocol */ |
82dc8b41 | 1950 | open_flags = bdrv_open_flags(bs, bs->open_flags); |
01a56501 | 1951 | node_name = qemu_opt_get(opts, "node-name"); |
57915332 | 1952 | |
01a56501 KW |
1953 | assert(!drv->bdrv_file_open || file == NULL); |
1954 | ret = bdrv_open_driver(bs, drv, node_name, options, open_flags, errp); | |
51762288 | 1955 | if (ret < 0) { |
01a56501 | 1956 | goto fail_opts; |
3baca891 KW |
1957 | } |
1958 | ||
18edf289 | 1959 | qemu_opts_del(opts); |
57915332 KW |
1960 | return 0; |
1961 | ||
18edf289 KW |
1962 | fail_opts: |
1963 | qemu_opts_del(opts); | |
57915332 KW |
1964 | return ret; |
1965 | } | |
1966 | ||
5e5c4f63 KW |
1967 | static QDict *parse_json_filename(const char *filename, Error **errp) |
1968 | { | |
1969 | QObject *options_obj; | |
1970 | QDict *options; | |
1971 | int ret; | |
bdb73476 | 1972 | GLOBAL_STATE_CODE(); |
5e5c4f63 KW |
1973 | |
1974 | ret = strstart(filename, "json:", &filename); | |
1975 | assert(ret); | |
1976 | ||
5577fff7 | 1977 | options_obj = qobject_from_json(filename, errp); |
5e5c4f63 | 1978 | if (!options_obj) { |
5577fff7 | 1979 | error_prepend(errp, "Could not parse the JSON options: "); |
5e5c4f63 KW |
1980 | return NULL; |
1981 | } | |
1982 | ||
7dc847eb | 1983 | options = qobject_to(QDict, options_obj); |
ca6b6e1e | 1984 | if (!options) { |
cb3e7f08 | 1985 | qobject_unref(options_obj); |
5e5c4f63 KW |
1986 | error_setg(errp, "Invalid JSON object given"); |
1987 | return NULL; | |
1988 | } | |
1989 | ||
5e5c4f63 KW |
1990 | qdict_flatten(options); |
1991 | ||
1992 | return options; | |
1993 | } | |
1994 | ||
de3b53f0 KW |
1995 | static void parse_json_protocol(QDict *options, const char **pfilename, |
1996 | Error **errp) | |
1997 | { | |
1998 | QDict *json_options; | |
1999 | Error *local_err = NULL; | |
bdb73476 | 2000 | GLOBAL_STATE_CODE(); |
de3b53f0 KW |
2001 | |
2002 | /* Parse json: pseudo-protocol */ | |
2003 | if (!*pfilename || !g_str_has_prefix(*pfilename, "json:")) { | |
2004 | return; | |
2005 | } | |
2006 | ||
2007 | json_options = parse_json_filename(*pfilename, &local_err); | |
2008 | if (local_err) { | |
2009 | error_propagate(errp, local_err); | |
2010 | return; | |
2011 | } | |
2012 | ||
2013 | /* Options given in the filename have lower priority than options | |
2014 | * specified directly */ | |
2015 | qdict_join(options, json_options, false); | |
cb3e7f08 | 2016 | qobject_unref(json_options); |
de3b53f0 KW |
2017 | *pfilename = NULL; |
2018 | } | |
2019 | ||
b6ce07aa | 2020 | /* |
f54120ff KW |
2021 | * Fills in default options for opening images and converts the legacy |
2022 | * filename/flags pair to option QDict entries. | |
53a29513 HR |
2023 | * The BDRV_O_PROTOCOL flag in *flags will be set or cleared accordingly if a |
2024 | * block driver has been specified explicitly. | |
b6ce07aa | 2025 | */ |
de3b53f0 | 2026 | static int bdrv_fill_options(QDict **options, const char *filename, |
053e1578 | 2027 | int *flags, Error **errp) |
ea2384d3 | 2028 | { |
c2ad1b0c | 2029 | const char *drvname; |
53a29513 | 2030 | bool protocol = *flags & BDRV_O_PROTOCOL; |
e3fa4bfa | 2031 | bool parse_filename = false; |
053e1578 | 2032 | BlockDriver *drv = NULL; |
34b5d2c6 | 2033 | Error *local_err = NULL; |
83f64091 | 2034 | |
da359909 EGE |
2035 | GLOBAL_STATE_CODE(); |
2036 | ||
129c7d1c MA |
2037 | /* |
2038 | * Caution: while qdict_get_try_str() is fine, getting non-string | |
2039 | * types would require more care. When @options come from | |
2040 | * -blockdev or blockdev_add, its members are typed according to | |
2041 | * the QAPI schema, but when they come from -drive, they're all | |
2042 | * QString. | |
2043 | */ | |
53a29513 | 2044 | drvname = qdict_get_try_str(*options, "driver"); |
053e1578 HR |
2045 | if (drvname) { |
2046 | drv = bdrv_find_format(drvname); | |
2047 | if (!drv) { | |
2048 | error_setg(errp, "Unknown driver '%s'", drvname); | |
2049 | return -ENOENT; | |
2050 | } | |
2051 | /* If the user has explicitly specified the driver, this choice should | |
2052 | * override the BDRV_O_PROTOCOL flag */ | |
2053 | protocol = drv->bdrv_file_open; | |
53a29513 HR |
2054 | } |
2055 | ||
2056 | if (protocol) { | |
2057 | *flags |= BDRV_O_PROTOCOL; | |
2058 | } else { | |
2059 | *flags &= ~BDRV_O_PROTOCOL; | |
2060 | } | |
2061 | ||
91a097e7 KW |
2062 | /* Translate cache options from flags into options */ |
2063 | update_options_from_flags(*options, *flags); | |
2064 | ||
035fccdf | 2065 | /* Fetch the file name from the options QDict if necessary */ |
17b005f1 | 2066 | if (protocol && filename) { |
f54120ff | 2067 | if (!qdict_haskey(*options, "filename")) { |
46f5ac20 | 2068 | qdict_put_str(*options, "filename", filename); |
f54120ff KW |
2069 | parse_filename = true; |
2070 | } else { | |
2071 | error_setg(errp, "Can't specify 'file' and 'filename' options at " | |
2072 | "the same time"); | |
2073 | return -EINVAL; | |
2074 | } | |
035fccdf KW |
2075 | } |
2076 | ||
c2ad1b0c | 2077 | /* Find the right block driver */ |
129c7d1c | 2078 | /* See cautionary note on accessing @options above */ |
f54120ff | 2079 | filename = qdict_get_try_str(*options, "filename"); |
f54120ff | 2080 | |
053e1578 HR |
2081 | if (!drvname && protocol) { |
2082 | if (filename) { | |
2083 | drv = bdrv_find_protocol(filename, parse_filename, errp); | |
17b005f1 | 2084 | if (!drv) { |
053e1578 | 2085 | return -EINVAL; |
17b005f1 | 2086 | } |
053e1578 HR |
2087 | |
2088 | drvname = drv->format_name; | |
46f5ac20 | 2089 | qdict_put_str(*options, "driver", drvname); |
053e1578 HR |
2090 | } else { |
2091 | error_setg(errp, "Must specify either driver or file"); | |
2092 | return -EINVAL; | |
98289620 | 2093 | } |
c2ad1b0c KW |
2094 | } |
2095 | ||
17b005f1 | 2096 | assert(drv || !protocol); |
c2ad1b0c | 2097 | |
f54120ff | 2098 | /* Driver-specific filename parsing */ |
17b005f1 | 2099 | if (drv && drv->bdrv_parse_filename && parse_filename) { |
5acd9d81 | 2100 | drv->bdrv_parse_filename(filename, *options, &local_err); |
84d18f06 | 2101 | if (local_err) { |
34b5d2c6 | 2102 | error_propagate(errp, local_err); |
f54120ff | 2103 | return -EINVAL; |
6963a30d | 2104 | } |
cd5d031e HR |
2105 | |
2106 | if (!drv->bdrv_needs_filename) { | |
2107 | qdict_del(*options, "filename"); | |
cd5d031e | 2108 | } |
6963a30d KW |
2109 | } |
2110 | ||
f54120ff KW |
2111 | return 0; |
2112 | } | |
2113 | ||
148eb13c KW |
2114 | typedef struct BlockReopenQueueEntry { |
2115 | bool prepared; | |
69b736e7 | 2116 | bool perms_checked; |
148eb13c | 2117 | BDRVReopenState state; |
859443b0 | 2118 | QTAILQ_ENTRY(BlockReopenQueueEntry) entry; |
148eb13c KW |
2119 | } BlockReopenQueueEntry; |
2120 | ||
2121 | /* | |
2122 | * Return the flags that @bs will have after the reopens in @q have | |
2123 | * successfully completed. If @q is NULL (or @bs is not contained in @q), | |
2124 | * return the current flags. | |
2125 | */ | |
2126 | static int bdrv_reopen_get_flags(BlockReopenQueue *q, BlockDriverState *bs) | |
2127 | { | |
2128 | BlockReopenQueueEntry *entry; | |
2129 | ||
2130 | if (q != NULL) { | |
859443b0 | 2131 | QTAILQ_FOREACH(entry, q, entry) { |
148eb13c KW |
2132 | if (entry->state.bs == bs) { |
2133 | return entry->state.flags; | |
2134 | } | |
2135 | } | |
2136 | } | |
2137 | ||
2138 | return bs->open_flags; | |
2139 | } | |
2140 | ||
2141 | /* Returns whether the image file can be written to after the reopen queue @q | |
2142 | * has been successfully applied, or right now if @q is NULL. */ | |
cc022140 HR |
2143 | static bool bdrv_is_writable_after_reopen(BlockDriverState *bs, |
2144 | BlockReopenQueue *q) | |
148eb13c KW |
2145 | { |
2146 | int flags = bdrv_reopen_get_flags(q, bs); | |
2147 | ||
2148 | return (flags & (BDRV_O_RDWR | BDRV_O_INACTIVE)) == BDRV_O_RDWR; | |
2149 | } | |
2150 | ||
cc022140 HR |
2151 | /* |
2152 | * Return whether the BDS can be written to. This is not necessarily | |
2153 | * the same as !bdrv_is_read_only(bs), as inactivated images may not | |
2154 | * be written to but do not count as read-only images. | |
2155 | */ | |
2156 | bool bdrv_is_writable(BlockDriverState *bs) | |
2157 | { | |
384a48fb | 2158 | IO_CODE(); |
cc022140 HR |
2159 | return bdrv_is_writable_after_reopen(bs, NULL); |
2160 | } | |
2161 | ||
3bf416ba VSO |
2162 | static char *bdrv_child_user_desc(BdrvChild *c) |
2163 | { | |
f0c28327 | 2164 | GLOBAL_STATE_CODE(); |
da261b69 | 2165 | return c->klass->get_parent_desc(c); |
3bf416ba VSO |
2166 | } |
2167 | ||
30ebb9aa VSO |
2168 | /* |
2169 | * Check that @a allows everything that @b needs. @a and @b must reference same | |
2170 | * child node. | |
2171 | */ | |
3bf416ba VSO |
2172 | static bool bdrv_a_allow_b(BdrvChild *a, BdrvChild *b, Error **errp) |
2173 | { | |
30ebb9aa VSO |
2174 | const char *child_bs_name; |
2175 | g_autofree char *a_user = NULL; | |
2176 | g_autofree char *b_user = NULL; | |
2177 | g_autofree char *perms = NULL; | |
2178 | ||
2179 | assert(a->bs); | |
2180 | assert(a->bs == b->bs); | |
862fded9 | 2181 | GLOBAL_STATE_CODE(); |
3bf416ba VSO |
2182 | |
2183 | if ((b->perm & a->shared_perm) == b->perm) { | |
2184 | return true; | |
2185 | } | |
2186 | ||
30ebb9aa VSO |
2187 | child_bs_name = bdrv_get_node_name(b->bs); |
2188 | a_user = bdrv_child_user_desc(a); | |
2189 | b_user = bdrv_child_user_desc(b); | |
2190 | perms = bdrv_perm_names(b->perm & ~a->shared_perm); | |
2191 | ||
2192 | error_setg(errp, "Permission conflict on node '%s': permissions '%s' are " | |
2193 | "both required by %s (uses node '%s' as '%s' child) and " | |
2194 | "unshared by %s (uses node '%s' as '%s' child).", | |
2195 | child_bs_name, perms, | |
2196 | b_user, child_bs_name, b->name, | |
2197 | a_user, child_bs_name, a->name); | |
3bf416ba VSO |
2198 | |
2199 | return false; | |
2200 | } | |
2201 | ||
9397c14f | 2202 | static bool bdrv_parent_perms_conflict(BlockDriverState *bs, Error **errp) |
3bf416ba VSO |
2203 | { |
2204 | BdrvChild *a, *b; | |
862fded9 | 2205 | GLOBAL_STATE_CODE(); |
3bf416ba VSO |
2206 | |
2207 | /* | |
2208 | * During the loop we'll look at each pair twice. That's correct because | |
2209 | * bdrv_a_allow_b() is asymmetric and we should check each pair in both | |
2210 | * directions. | |
2211 | */ | |
2212 | QLIST_FOREACH(a, &bs->parents, next_parent) { | |
2213 | QLIST_FOREACH(b, &bs->parents, next_parent) { | |
9397c14f | 2214 | if (a == b) { |
3bf416ba VSO |
2215 | continue; |
2216 | } | |
2217 | ||
2218 | if (!bdrv_a_allow_b(a, b, errp)) { | |
2219 | return true; | |
2220 | } | |
2221 | } | |
2222 | } | |
2223 | ||
2224 | return false; | |
2225 | } | |
2226 | ||
ffd1a5a2 | 2227 | static void bdrv_child_perm(BlockDriverState *bs, BlockDriverState *child_bs, |
e5d8a406 HR |
2228 | BdrvChild *c, BdrvChildRole role, |
2229 | BlockReopenQueue *reopen_queue, | |
ffd1a5a2 FZ |
2230 | uint64_t parent_perm, uint64_t parent_shared, |
2231 | uint64_t *nperm, uint64_t *nshared) | |
2232 | { | |
0b3ca76e | 2233 | assert(bs->drv && bs->drv->bdrv_child_perm); |
da359909 | 2234 | GLOBAL_STATE_CODE(); |
e5d8a406 | 2235 | bs->drv->bdrv_child_perm(bs, c, role, reopen_queue, |
0b3ca76e AG |
2236 | parent_perm, parent_shared, |
2237 | nperm, nshared); | |
e0995dc3 | 2238 | /* TODO Take force_share from reopen_queue */ |
ffd1a5a2 FZ |
2239 | if (child_bs && child_bs->force_share) { |
2240 | *nshared = BLK_PERM_ALL; | |
2241 | } | |
2242 | } | |
2243 | ||
bd57f8f7 VSO |
2244 | /* |
2245 | * Adds the whole subtree of @bs (including @bs itself) to the @list (except for | |
2246 | * nodes that are already in the @list, of course) so that final list is | |
2247 | * topologically sorted. Return the result (GSList @list object is updated, so | |
2248 | * don't use old reference after function call). | |
2249 | * | |
2250 | * On function start @list must be already topologically sorted and for any node | |
2251 | * in the @list the whole subtree of the node must be in the @list as well. The | |
2252 | * simplest way to satisfy this criteria: use only result of | |
2253 | * bdrv_topological_dfs() or NULL as @list parameter. | |
2254 | */ | |
2255 | static GSList *bdrv_topological_dfs(GSList *list, GHashTable *found, | |
2256 | BlockDriverState *bs) | |
2257 | { | |
2258 | BdrvChild *child; | |
2259 | g_autoptr(GHashTable) local_found = NULL; | |
2260 | ||
bdb73476 EGE |
2261 | GLOBAL_STATE_CODE(); |
2262 | ||
bd57f8f7 VSO |
2263 | if (!found) { |
2264 | assert(!list); | |
2265 | found = local_found = g_hash_table_new(NULL, NULL); | |
2266 | } | |
2267 | ||
2268 | if (g_hash_table_contains(found, bs)) { | |
2269 | return list; | |
2270 | } | |
2271 | g_hash_table_add(found, bs); | |
2272 | ||
2273 | QLIST_FOREACH(child, &bs->children, next) { | |
2274 | list = bdrv_topological_dfs(list, found, child->bs); | |
2275 | } | |
2276 | ||
2277 | return g_slist_prepend(list, bs); | |
2278 | } | |
2279 | ||
ecb776bd VSO |
2280 | typedef struct BdrvChildSetPermState { |
2281 | BdrvChild *child; | |
2282 | uint64_t old_perm; | |
2283 | uint64_t old_shared_perm; | |
2284 | } BdrvChildSetPermState; | |
b0defa83 VSO |
2285 | |
2286 | static void bdrv_child_set_perm_abort(void *opaque) | |
2287 | { | |
ecb776bd VSO |
2288 | BdrvChildSetPermState *s = opaque; |
2289 | ||
862fded9 EGE |
2290 | GLOBAL_STATE_CODE(); |
2291 | ||
ecb776bd VSO |
2292 | s->child->perm = s->old_perm; |
2293 | s->child->shared_perm = s->old_shared_perm; | |
b0defa83 VSO |
2294 | } |
2295 | ||
2296 | static TransactionActionDrv bdrv_child_set_pem_drv = { | |
2297 | .abort = bdrv_child_set_perm_abort, | |
ecb776bd | 2298 | .clean = g_free, |
b0defa83 VSO |
2299 | }; |
2300 | ||
ecb776bd VSO |
2301 | static void bdrv_child_set_perm(BdrvChild *c, uint64_t perm, |
2302 | uint64_t shared, Transaction *tran) | |
b0defa83 | 2303 | { |
ecb776bd | 2304 | BdrvChildSetPermState *s = g_new(BdrvChildSetPermState, 1); |
862fded9 | 2305 | GLOBAL_STATE_CODE(); |
ecb776bd VSO |
2306 | |
2307 | *s = (BdrvChildSetPermState) { | |
2308 | .child = c, | |
2309 | .old_perm = c->perm, | |
2310 | .old_shared_perm = c->shared_perm, | |
2311 | }; | |
b0defa83 VSO |
2312 | |
2313 | c->perm = perm; | |
2314 | c->shared_perm = shared; | |
2315 | ||
ecb776bd | 2316 | tran_add(tran, &bdrv_child_set_pem_drv, s); |
b0defa83 VSO |
2317 | } |
2318 | ||
2513ef59 VSO |
2319 | static void bdrv_drv_set_perm_commit(void *opaque) |
2320 | { | |
2321 | BlockDriverState *bs = opaque; | |
2322 | uint64_t cumulative_perms, cumulative_shared_perms; | |
da359909 | 2323 | GLOBAL_STATE_CODE(); |
2513ef59 VSO |
2324 | |
2325 | if (bs->drv->bdrv_set_perm) { | |
2326 | bdrv_get_cumulative_perm(bs, &cumulative_perms, | |
2327 | &cumulative_shared_perms); | |
2328 | bs->drv->bdrv_set_perm(bs, cumulative_perms, cumulative_shared_perms); | |
2329 | } | |
2330 | } | |
2331 | ||
2332 | static void bdrv_drv_set_perm_abort(void *opaque) | |
2333 | { | |
2334 | BlockDriverState *bs = opaque; | |
da359909 | 2335 | GLOBAL_STATE_CODE(); |
2513ef59 VSO |
2336 | |
2337 | if (bs->drv->bdrv_abort_perm_update) { | |
2338 | bs->drv->bdrv_abort_perm_update(bs); | |
2339 | } | |
2340 | } | |
2341 | ||
2342 | TransactionActionDrv bdrv_drv_set_perm_drv = { | |
2343 | .abort = bdrv_drv_set_perm_abort, | |
2344 | .commit = bdrv_drv_set_perm_commit, | |
2345 | }; | |
2346 | ||
2347 | static int bdrv_drv_set_perm(BlockDriverState *bs, uint64_t perm, | |
2348 | uint64_t shared_perm, Transaction *tran, | |
2349 | Error **errp) | |
2350 | { | |
da359909 | 2351 | GLOBAL_STATE_CODE(); |
2513ef59 VSO |
2352 | if (!bs->drv) { |
2353 | return 0; | |
2354 | } | |
2355 | ||
2356 | if (bs->drv->bdrv_check_perm) { | |
2357 | int ret = bs->drv->bdrv_check_perm(bs, perm, shared_perm, errp); | |
2358 | if (ret < 0) { | |
2359 | return ret; | |
2360 | } | |
2361 | } | |
2362 | ||
2363 | if (tran) { | |
2364 | tran_add(tran, &bdrv_drv_set_perm_drv, bs); | |
2365 | } | |
2366 | ||
2367 | return 0; | |
2368 | } | |
2369 | ||
0978623e VSO |
2370 | typedef struct BdrvReplaceChildState { |
2371 | BdrvChild *child; | |
2372 | BlockDriverState *old_bs; | |
2373 | } BdrvReplaceChildState; | |
2374 | ||
2375 | static void bdrv_replace_child_commit(void *opaque) | |
2376 | { | |
2377 | BdrvReplaceChildState *s = opaque; | |
bdb73476 | 2378 | GLOBAL_STATE_CODE(); |
0978623e VSO |
2379 | |
2380 | bdrv_unref(s->old_bs); | |
2381 | } | |
2382 | ||
2383 | static void bdrv_replace_child_abort(void *opaque) | |
2384 | { | |
2385 | BdrvReplaceChildState *s = opaque; | |
2386 | BlockDriverState *new_bs = s->child->bs; | |
2387 | ||
bdb73476 | 2388 | GLOBAL_STATE_CODE(); |
0f0b1e29 | 2389 | /* old_bs reference is transparently moved from @s to @s->child */ |
23987471 KW |
2390 | if (!s->child->bs) { |
2391 | /* | |
2392 | * The parents were undrained when removing old_bs from the child. New | |
2393 | * requests can't have been made, though, because the child was empty. | |
2394 | * | |
2395 | * TODO Make bdrv_replace_child_noperm() transactionable to avoid | |
2396 | * undraining the parent in the first place. Once this is done, having | |
2397 | * new_bs drained when calling bdrv_replace_child_tran() is not a | |
2398 | * requirement any more. | |
2399 | */ | |
606ed756 | 2400 | bdrv_parent_drained_begin_single(s->child); |
23987471 KW |
2401 | assert(!bdrv_parent_drained_poll_single(s->child)); |
2402 | } | |
2403 | assert(s->child->quiesced_parent); | |
544acc7d | 2404 | bdrv_replace_child_noperm(s->child, s->old_bs); |
0978623e VSO |
2405 | bdrv_unref(new_bs); |
2406 | } | |
2407 | ||
2408 | static TransactionActionDrv bdrv_replace_child_drv = { | |
2409 | .commit = bdrv_replace_child_commit, | |
2410 | .abort = bdrv_replace_child_abort, | |
2411 | .clean = g_free, | |
2412 | }; | |
2413 | ||
2414 | /* | |
4bf021db | 2415 | * bdrv_replace_child_tran |
0978623e VSO |
2416 | * |
2417 | * Note: real unref of old_bs is done only on commit. | |
4bf021db | 2418 | * |
23987471 KW |
2419 | * Both @child->bs and @new_bs (if non-NULL) must be drained. @new_bs must be |
2420 | * kept drained until the transaction is completed. | |
2421 | * | |
4bf021db | 2422 | * The function doesn't update permissions, caller is responsible for this. |
0978623e | 2423 | */ |
0f0b1e29 | 2424 | static void bdrv_replace_child_tran(BdrvChild *child, BlockDriverState *new_bs, |
4eba825a | 2425 | Transaction *tran) |
0978623e VSO |
2426 | { |
2427 | BdrvReplaceChildState *s = g_new(BdrvReplaceChildState, 1); | |
23987471 KW |
2428 | |
2429 | assert(child->quiesced_parent); | |
2430 | assert(!new_bs || new_bs->quiesce_counter); | |
2431 | ||
0978623e | 2432 | *s = (BdrvReplaceChildState) { |
0f0b1e29 VSO |
2433 | .child = child, |
2434 | .old_bs = child->bs, | |
0978623e VSO |
2435 | }; |
2436 | tran_add(tran, &bdrv_replace_child_drv, s); | |
2437 | ||
2438 | if (new_bs) { | |
2439 | bdrv_ref(new_bs); | |
2440 | } | |
544acc7d | 2441 | bdrv_replace_child_noperm(child, new_bs); |
0f0b1e29 | 2442 | /* old_bs reference is transparently moved from @child to @s */ |
0978623e VSO |
2443 | } |
2444 | ||
33a610c3 | 2445 | /* |
c20555e1 VSO |
2446 | * Refresh permissions in @bs subtree. The function is intended to be called |
2447 | * after some graph modification that was done without permission update. | |
33a610c3 | 2448 | */ |
c20555e1 VSO |
2449 | static int bdrv_node_refresh_perm(BlockDriverState *bs, BlockReopenQueue *q, |
2450 | Transaction *tran, Error **errp) | |
33a610c3 KW |
2451 | { |
2452 | BlockDriver *drv = bs->drv; | |
2453 | BdrvChild *c; | |
2454 | int ret; | |
c20555e1 | 2455 | uint64_t cumulative_perms, cumulative_shared_perms; |
862fded9 | 2456 | GLOBAL_STATE_CODE(); |
c20555e1 VSO |
2457 | |
2458 | bdrv_get_cumulative_perm(bs, &cumulative_perms, &cumulative_shared_perms); | |
33a610c3 KW |
2459 | |
2460 | /* Write permissions never work with read-only images */ | |
2461 | if ((cumulative_perms & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)) && | |
cc022140 | 2462 | !bdrv_is_writable_after_reopen(bs, q)) |
33a610c3 | 2463 | { |
481e0eee HR |
2464 | if (!bdrv_is_writable_after_reopen(bs, NULL)) { |
2465 | error_setg(errp, "Block node is read-only"); | |
2466 | } else { | |
c20555e1 VSO |
2467 | error_setg(errp, "Read-only block node '%s' cannot support " |
2468 | "read-write users", bdrv_get_node_name(bs)); | |
481e0eee HR |
2469 | } |
2470 | ||
33a610c3 KW |
2471 | return -EPERM; |
2472 | } | |
2473 | ||
9c60a5d1 KW |
2474 | /* |
2475 | * Unaligned requests will automatically be aligned to bl.request_alignment | |
2476 | * and without RESIZE we can't extend requests to write to space beyond the | |
2477 | * end of the image, so it's required that the image size is aligned. | |
2478 | */ | |
2479 | if ((cumulative_perms & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)) && | |
2480 | !(cumulative_perms & BLK_PERM_RESIZE)) | |
2481 | { | |
2482 | if ((bs->total_sectors * BDRV_SECTOR_SIZE) % bs->bl.request_alignment) { | |
2483 | error_setg(errp, "Cannot get 'write' permission without 'resize': " | |
2484 | "Image size is not a multiple of request " | |
2485 | "alignment"); | |
2486 | return -EPERM; | |
2487 | } | |
2488 | } | |
2489 | ||
33a610c3 KW |
2490 | /* Check this node */ |
2491 | if (!drv) { | |
2492 | return 0; | |
2493 | } | |
2494 | ||
b1d2bbeb | 2495 | ret = bdrv_drv_set_perm(bs, cumulative_perms, cumulative_shared_perms, tran, |
2513ef59 VSO |
2496 | errp); |
2497 | if (ret < 0) { | |
2498 | return ret; | |
33a610c3 KW |
2499 | } |
2500 | ||
78e421c9 | 2501 | /* Drivers that never have children can omit .bdrv_child_perm() */ |
33a610c3 | 2502 | if (!drv->bdrv_child_perm) { |
78e421c9 | 2503 | assert(QLIST_EMPTY(&bs->children)); |
33a610c3 KW |
2504 | return 0; |
2505 | } | |
2506 | ||
2507 | /* Check all children */ | |
2508 | QLIST_FOREACH(c, &bs->children, next) { | |
2509 | uint64_t cur_perm, cur_shared; | |
9eab1544 | 2510 | |
e5d8a406 | 2511 | bdrv_child_perm(bs, c->bs, c, c->role, q, |
ffd1a5a2 FZ |
2512 | cumulative_perms, cumulative_shared_perms, |
2513 | &cur_perm, &cur_shared); | |
ecb776bd | 2514 | bdrv_child_set_perm(c, cur_perm, cur_shared, tran); |
bd57f8f7 VSO |
2515 | } |
2516 | ||
2517 | return 0; | |
2518 | } | |
3ef45e02 | 2519 | |
fb0ff4d1 VSO |
2520 | /* |
2521 | * @list is a product of bdrv_topological_dfs() (may be called several times) - | |
2522 | * a topologically sorted subgraph. | |
2523 | */ | |
2524 | static int bdrv_do_refresh_perms(GSList *list, BlockReopenQueue *q, | |
2525 | Transaction *tran, Error **errp) | |
bd57f8f7 VSO |
2526 | { |
2527 | int ret; | |
b1d2bbeb | 2528 | BlockDriverState *bs; |
862fded9 | 2529 | GLOBAL_STATE_CODE(); |
bd57f8f7 | 2530 | |
b1d2bbeb VSO |
2531 | for ( ; list; list = list->next) { |
2532 | bs = list->data; | |
2533 | ||
9397c14f | 2534 | if (bdrv_parent_perms_conflict(bs, errp)) { |
b1d2bbeb | 2535 | return -EINVAL; |
bd57f8f7 VSO |
2536 | } |
2537 | ||
c20555e1 | 2538 | ret = bdrv_node_refresh_perm(bs, q, tran, errp); |
33a610c3 KW |
2539 | if (ret < 0) { |
2540 | return ret; | |
2541 | } | |
2542 | } | |
2543 | ||
2544 | return 0; | |
2545 | } | |
2546 | ||
fb0ff4d1 VSO |
2547 | /* |
2548 | * @list is any list of nodes. List is completed by all subtrees and | |
2549 | * topologically sorted. It's not a problem if some node occurs in the @list | |
2550 | * several times. | |
2551 | */ | |
2552 | static int bdrv_list_refresh_perms(GSList *list, BlockReopenQueue *q, | |
2553 | Transaction *tran, Error **errp) | |
2554 | { | |
2555 | g_autoptr(GHashTable) found = g_hash_table_new(NULL, NULL); | |
2556 | g_autoptr(GSList) refresh_list = NULL; | |
2557 | ||
2558 | for ( ; list; list = list->next) { | |
2559 | refresh_list = bdrv_topological_dfs(refresh_list, found, list->data); | |
2560 | } | |
2561 | ||
2562 | return bdrv_do_refresh_perms(refresh_list, q, tran, errp); | |
2563 | } | |
2564 | ||
c7a0f2be KW |
2565 | void bdrv_get_cumulative_perm(BlockDriverState *bs, uint64_t *perm, |
2566 | uint64_t *shared_perm) | |
33a610c3 KW |
2567 | { |
2568 | BdrvChild *c; | |
2569 | uint64_t cumulative_perms = 0; | |
2570 | uint64_t cumulative_shared_perms = BLK_PERM_ALL; | |
2571 | ||
b4ad82aa EGE |
2572 | GLOBAL_STATE_CODE(); |
2573 | ||
33a610c3 KW |
2574 | QLIST_FOREACH(c, &bs->parents, next_parent) { |
2575 | cumulative_perms |= c->perm; | |
2576 | cumulative_shared_perms &= c->shared_perm; | |
2577 | } | |
2578 | ||
2579 | *perm = cumulative_perms; | |
2580 | *shared_perm = cumulative_shared_perms; | |
2581 | } | |
2582 | ||
5176196c | 2583 | char *bdrv_perm_names(uint64_t perm) |
d083319f KW |
2584 | { |
2585 | struct perm_name { | |
2586 | uint64_t perm; | |
2587 | const char *name; | |
2588 | } permissions[] = { | |
2589 | { BLK_PERM_CONSISTENT_READ, "consistent read" }, | |
2590 | { BLK_PERM_WRITE, "write" }, | |
2591 | { BLK_PERM_WRITE_UNCHANGED, "write unchanged" }, | |
2592 | { BLK_PERM_RESIZE, "resize" }, | |
d083319f KW |
2593 | { 0, NULL } |
2594 | }; | |
2595 | ||
e2a7423a | 2596 | GString *result = g_string_sized_new(30); |
d083319f KW |
2597 | struct perm_name *p; |
2598 | ||
2599 | for (p = permissions; p->name; p++) { | |
2600 | if (perm & p->perm) { | |
e2a7423a AG |
2601 | if (result->len > 0) { |
2602 | g_string_append(result, ", "); | |
2603 | } | |
2604 | g_string_append(result, p->name); | |
d083319f KW |
2605 | } |
2606 | } | |
2607 | ||
e2a7423a | 2608 | return g_string_free(result, FALSE); |
d083319f KW |
2609 | } |
2610 | ||
33a610c3 | 2611 | |
f1316edb VSO |
2612 | /* @tran is allowed to be NULL. In this case no rollback is possible */ |
2613 | static int bdrv_refresh_perms(BlockDriverState *bs, Transaction *tran, | |
2614 | Error **errp) | |
bb87e4d1 VSO |
2615 | { |
2616 | int ret; | |
f1316edb | 2617 | Transaction *local_tran = NULL; |
b1d2bbeb | 2618 | g_autoptr(GSList) list = bdrv_topological_dfs(NULL, NULL, bs); |
862fded9 | 2619 | GLOBAL_STATE_CODE(); |
bb87e4d1 | 2620 | |
f1316edb VSO |
2621 | if (!tran) { |
2622 | tran = local_tran = tran_new(); | |
2623 | } | |
2624 | ||
fb0ff4d1 | 2625 | ret = bdrv_do_refresh_perms(list, NULL, tran, errp); |
f1316edb VSO |
2626 | |
2627 | if (local_tran) { | |
2628 | tran_finalize(local_tran, ret); | |
2629 | } | |
bb87e4d1 | 2630 | |
b1d2bbeb | 2631 | return ret; |
bb87e4d1 VSO |
2632 | } |
2633 | ||
33a610c3 KW |
2634 | int bdrv_child_try_set_perm(BdrvChild *c, uint64_t perm, uint64_t shared, |
2635 | Error **errp) | |
2636 | { | |
1046779e | 2637 | Error *local_err = NULL; |
83928dc4 | 2638 | Transaction *tran = tran_new(); |
33a610c3 KW |
2639 | int ret; |
2640 | ||
b4ad82aa EGE |
2641 | GLOBAL_STATE_CODE(); |
2642 | ||
ecb776bd | 2643 | bdrv_child_set_perm(c, perm, shared, tran); |
83928dc4 | 2644 | |
f1316edb | 2645 | ret = bdrv_refresh_perms(c->bs, tran, &local_err); |
83928dc4 VSO |
2646 | |
2647 | tran_finalize(tran, ret); | |
2648 | ||
33a610c3 | 2649 | if (ret < 0) { |
071b474f VSO |
2650 | if ((perm & ~c->perm) || (c->shared_perm & ~shared)) { |
2651 | /* tighten permissions */ | |
1046779e HR |
2652 | error_propagate(errp, local_err); |
2653 | } else { | |
2654 | /* | |
2655 | * Our caller may intend to only loosen restrictions and | |
2656 | * does not expect this function to fail. Errors are not | |
2657 | * fatal in such a case, so we can just hide them from our | |
2658 | * caller. | |
2659 | */ | |
2660 | error_free(local_err); | |
2661 | ret = 0; | |
2662 | } | |
33a610c3 KW |
2663 | } |
2664 | ||
83928dc4 | 2665 | return ret; |
d5e6f437 KW |
2666 | } |
2667 | ||
c1087f12 HR |
2668 | int bdrv_child_refresh_perms(BlockDriverState *bs, BdrvChild *c, Error **errp) |
2669 | { | |
2670 | uint64_t parent_perms, parent_shared; | |
2671 | uint64_t perms, shared; | |
2672 | ||
b4ad82aa EGE |
2673 | GLOBAL_STATE_CODE(); |
2674 | ||
c1087f12 | 2675 | bdrv_get_cumulative_perm(bs, &parent_perms, &parent_shared); |
e5d8a406 | 2676 | bdrv_child_perm(bs, c->bs, c, c->role, NULL, |
bf8e925e | 2677 | parent_perms, parent_shared, &perms, &shared); |
c1087f12 HR |
2678 | |
2679 | return bdrv_child_try_set_perm(c, perms, shared, errp); | |
2680 | } | |
2681 | ||
87278af1 HR |
2682 | /* |
2683 | * Default implementation for .bdrv_child_perm() for block filters: | |
2684 | * Forward CONSISTENT_READ, WRITE, WRITE_UNCHANGED, and RESIZE to the | |
2685 | * filtered child. | |
2686 | */ | |
2687 | static void bdrv_filter_default_perms(BlockDriverState *bs, BdrvChild *c, | |
87278af1 HR |
2688 | BdrvChildRole role, |
2689 | BlockReopenQueue *reopen_queue, | |
2690 | uint64_t perm, uint64_t shared, | |
2691 | uint64_t *nperm, uint64_t *nshared) | |
6a1b9ee1 | 2692 | { |
862fded9 | 2693 | GLOBAL_STATE_CODE(); |
e444fa83 KW |
2694 | *nperm = perm & DEFAULT_PERM_PASSTHROUGH; |
2695 | *nshared = (shared & DEFAULT_PERM_PASSTHROUGH) | DEFAULT_PERM_UNCHANGED; | |
6a1b9ee1 KW |
2696 | } |
2697 | ||
70082db4 | 2698 | static void bdrv_default_perms_for_cow(BlockDriverState *bs, BdrvChild *c, |
70082db4 HR |
2699 | BdrvChildRole role, |
2700 | BlockReopenQueue *reopen_queue, | |
2701 | uint64_t perm, uint64_t shared, | |
2702 | uint64_t *nperm, uint64_t *nshared) | |
2703 | { | |
e5d8a406 | 2704 | assert(role & BDRV_CHILD_COW); |
862fded9 | 2705 | GLOBAL_STATE_CODE(); |
70082db4 HR |
2706 | |
2707 | /* | |
2708 | * We want consistent read from backing files if the parent needs it. | |
2709 | * No other operations are performed on backing files. | |
2710 | */ | |
2711 | perm &= BLK_PERM_CONSISTENT_READ; | |
2712 | ||
2713 | /* | |
2714 | * If the parent can deal with changing data, we're okay with a | |
2715 | * writable and resizable backing file. | |
2716 | * TODO Require !(perm & BLK_PERM_CONSISTENT_READ), too? | |
2717 | */ | |
2718 | if (shared & BLK_PERM_WRITE) { | |
2719 | shared = BLK_PERM_WRITE | BLK_PERM_RESIZE; | |
2720 | } else { | |
2721 | shared = 0; | |
2722 | } | |
2723 | ||
64631f36 | 2724 | shared |= BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE_UNCHANGED; |
70082db4 HR |
2725 | |
2726 | if (bs->open_flags & BDRV_O_INACTIVE) { | |
2727 | shared |= BLK_PERM_WRITE | BLK_PERM_RESIZE; | |
2728 | } | |
2729 | ||
2730 | *nperm = perm; | |
2731 | *nshared = shared; | |
2732 | } | |
2733 | ||
6f838a4b | 2734 | static void bdrv_default_perms_for_storage(BlockDriverState *bs, BdrvChild *c, |
6f838a4b HR |
2735 | BdrvChildRole role, |
2736 | BlockReopenQueue *reopen_queue, | |
2737 | uint64_t perm, uint64_t shared, | |
2738 | uint64_t *nperm, uint64_t *nshared) | |
2739 | { | |
2740 | int flags; | |
2741 | ||
862fded9 | 2742 | GLOBAL_STATE_CODE(); |
e5d8a406 | 2743 | assert(role & (BDRV_CHILD_METADATA | BDRV_CHILD_DATA)); |
6f838a4b HR |
2744 | |
2745 | flags = bdrv_reopen_get_flags(reopen_queue, bs); | |
2746 | ||
2747 | /* | |
2748 | * Apart from the modifications below, the same permissions are | |
2749 | * forwarded and left alone as for filters | |
2750 | */ | |
e5d8a406 | 2751 | bdrv_filter_default_perms(bs, c, role, reopen_queue, |
6f838a4b HR |
2752 | perm, shared, &perm, &shared); |
2753 | ||
f889054f HR |
2754 | if (role & BDRV_CHILD_METADATA) { |
2755 | /* Format drivers may touch metadata even if the guest doesn't write */ | |
2756 | if (bdrv_is_writable_after_reopen(bs, reopen_queue)) { | |
2757 | perm |= BLK_PERM_WRITE | BLK_PERM_RESIZE; | |
2758 | } | |
2759 | ||
2760 | /* | |
2761 | * bs->file always needs to be consistent because of the | |
2762 | * metadata. We can never allow other users to resize or write | |
2763 | * to it. | |
2764 | */ | |
2765 | if (!(flags & BDRV_O_NO_IO)) { | |
2766 | perm |= BLK_PERM_CONSISTENT_READ; | |
2767 | } | |
2768 | shared &= ~(BLK_PERM_WRITE | BLK_PERM_RESIZE); | |
6f838a4b HR |
2769 | } |
2770 | ||
f889054f HR |
2771 | if (role & BDRV_CHILD_DATA) { |
2772 | /* | |
2773 | * Technically, everything in this block is a subset of the | |
2774 | * BDRV_CHILD_METADATA path taken above, and so this could | |
2775 | * be an "else if" branch. However, that is not obvious, and | |
2776 | * this function is not performance critical, therefore we let | |
2777 | * this be an independent "if". | |
2778 | */ | |
2779 | ||
2780 | /* | |
2781 | * We cannot allow other users to resize the file because the | |
2782 | * format driver might have some assumptions about the size | |
2783 | * (e.g. because it is stored in metadata, or because the file | |
2784 | * is split into fixed-size data files). | |
2785 | */ | |
2786 | shared &= ~BLK_PERM_RESIZE; | |
2787 | ||
2788 | /* | |
2789 | * WRITE_UNCHANGED often cannot be performed as such on the | |
2790 | * data file. For example, the qcow2 driver may still need to | |
2791 | * write copied clusters on copy-on-read. | |
2792 | */ | |
2793 | if (perm & BLK_PERM_WRITE_UNCHANGED) { | |
2794 | perm |= BLK_PERM_WRITE; | |
2795 | } | |
2796 | ||
2797 | /* | |
2798 | * If the data file is written to, the format driver may | |
2799 | * expect to be able to resize it by writing beyond the EOF. | |
2800 | */ | |
2801 | if (perm & BLK_PERM_WRITE) { | |
2802 | perm |= BLK_PERM_RESIZE; | |
2803 | } | |
6f838a4b | 2804 | } |
6f838a4b HR |
2805 | |
2806 | if (bs->open_flags & BDRV_O_INACTIVE) { | |
2807 | shared |= BLK_PERM_WRITE | BLK_PERM_RESIZE; | |
2808 | } | |
2809 | ||
2810 | *nperm = perm; | |
2811 | *nshared = shared; | |
2812 | } | |
2813 | ||
2519f549 | 2814 | void bdrv_default_perms(BlockDriverState *bs, BdrvChild *c, |
e5d8a406 | 2815 | BdrvChildRole role, BlockReopenQueue *reopen_queue, |
2519f549 HR |
2816 | uint64_t perm, uint64_t shared, |
2817 | uint64_t *nperm, uint64_t *nshared) | |
2818 | { | |
b4ad82aa | 2819 | GLOBAL_STATE_CODE(); |
2519f549 HR |
2820 | if (role & BDRV_CHILD_FILTERED) { |
2821 | assert(!(role & (BDRV_CHILD_DATA | BDRV_CHILD_METADATA | | |
2822 | BDRV_CHILD_COW))); | |
e5d8a406 | 2823 | bdrv_filter_default_perms(bs, c, role, reopen_queue, |
2519f549 HR |
2824 | perm, shared, nperm, nshared); |
2825 | } else if (role & BDRV_CHILD_COW) { | |
2826 | assert(!(role & (BDRV_CHILD_DATA | BDRV_CHILD_METADATA))); | |
e5d8a406 | 2827 | bdrv_default_perms_for_cow(bs, c, role, reopen_queue, |
2519f549 HR |
2828 | perm, shared, nperm, nshared); |
2829 | } else if (role & (BDRV_CHILD_METADATA | BDRV_CHILD_DATA)) { | |
e5d8a406 | 2830 | bdrv_default_perms_for_storage(bs, c, role, reopen_queue, |
2519f549 HR |
2831 | perm, shared, nperm, nshared); |
2832 | } else { | |
2833 | g_assert_not_reached(); | |
2834 | } | |
2835 | } | |
2836 | ||
7b1d9c4d HR |
2837 | uint64_t bdrv_qapi_perm_to_blk_perm(BlockPermission qapi_perm) |
2838 | { | |
2839 | static const uint64_t permissions[] = { | |
2840 | [BLOCK_PERMISSION_CONSISTENT_READ] = BLK_PERM_CONSISTENT_READ, | |
2841 | [BLOCK_PERMISSION_WRITE] = BLK_PERM_WRITE, | |
2842 | [BLOCK_PERMISSION_WRITE_UNCHANGED] = BLK_PERM_WRITE_UNCHANGED, | |
2843 | [BLOCK_PERMISSION_RESIZE] = BLK_PERM_RESIZE, | |
7b1d9c4d HR |
2844 | }; |
2845 | ||
2846 | QEMU_BUILD_BUG_ON(ARRAY_SIZE(permissions) != BLOCK_PERMISSION__MAX); | |
2847 | QEMU_BUILD_BUG_ON(1UL << ARRAY_SIZE(permissions) != BLK_PERM_ALL + 1); | |
2848 | ||
2849 | assert(qapi_perm < BLOCK_PERMISSION__MAX); | |
2850 | ||
2851 | return permissions[qapi_perm]; | |
2852 | } | |
2853 | ||
23987471 KW |
2854 | /* |
2855 | * Replaces the node that a BdrvChild points to without updating permissions. | |
2856 | * | |
2857 | * If @new_bs is non-NULL, the parent of @child must already be drained through | |
31b2ddfe | 2858 | * @child and the caller must hold the AioContext lock for @new_bs. |
23987471 | 2859 | */ |
544acc7d | 2860 | static void bdrv_replace_child_noperm(BdrvChild *child, |
4eba825a | 2861 | BlockDriverState *new_bs) |
e9740bc6 KW |
2862 | { |
2863 | BlockDriverState *old_bs = child->bs; | |
debc2927 | 2864 | int new_bs_quiesce_counter; |
e9740bc6 | 2865 | |
2cad1ebe | 2866 | assert(!child->frozen); |
23987471 KW |
2867 | |
2868 | /* | |
2869 | * If we want to change the BdrvChild to point to a drained node as its new | |
2870 | * child->bs, we need to make sure that its new parent is drained, too. In | |
2871 | * other words, either child->quiesce_parent must already be true or we must | |
2872 | * be able to set it and keep the parent's quiesce_counter consistent with | |
2873 | * that, but without polling or starting new requests (this function | |
2874 | * guarantees that it doesn't poll, and starting new requests would be | |
2875 | * against the invariants of drain sections). | |
2876 | * | |
2877 | * To keep things simple, we pick the first option (child->quiesce_parent | |
2878 | * must already be true). We also generalise the rule a bit to make it | |
2879 | * easier to verify in callers and more likely to be covered in test cases: | |
2880 | * The parent must be quiesced through this child even if new_bs isn't | |
2881 | * currently drained. | |
2882 | * | |
2883 | * The only exception is for callers that always pass new_bs == NULL. In | |
2884 | * this case, we obviously never need to consider the case of a drained | |
2885 | * new_bs, so we can keep the callers simpler by allowing them not to drain | |
2886 | * the parent. | |
2887 | */ | |
2888 | assert(!new_bs || child->quiesced_parent); | |
bfb8aa6d | 2889 | assert(old_bs != new_bs); |
f0c28327 | 2890 | GLOBAL_STATE_CODE(); |
2cad1ebe | 2891 | |
bb2614e9 FZ |
2892 | if (old_bs && new_bs) { |
2893 | assert(bdrv_get_aio_context(old_bs) == bdrv_get_aio_context(new_bs)); | |
2894 | } | |
debc2927 | 2895 | |
29312507 | 2896 | /* TODO Pull this up into the callers to avoid polling here */ |
31b2ddfe | 2897 | bdrv_graph_wrlock(new_bs); |
e9740bc6 | 2898 | if (old_bs) { |
bd86fb99 HR |
2899 | if (child->klass->detach) { |
2900 | child->klass->detach(child); | |
d736f119 | 2901 | } |
e9740bc6 KW |
2902 | QLIST_REMOVE(child, next_parent); |
2903 | } | |
36fe1331 KW |
2904 | |
2905 | child->bs = new_bs; | |
2906 | ||
e9740bc6 KW |
2907 | if (new_bs) { |
2908 | QLIST_INSERT_HEAD(&new_bs->parents, child, next_parent); | |
bd86fb99 HR |
2909 | if (child->klass->attach) { |
2910 | child->klass->attach(child); | |
8ee03995 KW |
2911 | } |
2912 | } | |
29312507 | 2913 | bdrv_graph_wrunlock(); |
debc2927 HR |
2914 | |
2915 | /* | |
23987471 KW |
2916 | * If the parent was drained through this BdrvChild previously, but new_bs |
2917 | * is not drained, allow requests to come in only after the new node has | |
2918 | * been attached. | |
debc2927 | 2919 | */ |
57e05be3 KW |
2920 | new_bs_quiesce_counter = (new_bs ? new_bs->quiesce_counter : 0); |
2921 | if (!new_bs_quiesce_counter && child->quiesced_parent) { | |
debc2927 | 2922 | bdrv_parent_drained_end_single(child); |
debc2927 | 2923 | } |
8ee03995 | 2924 | } |
33a610c3 | 2925 | |
04c9c3a5 HR |
2926 | /** |
2927 | * Free the given @child. | |
2928 | * | |
2929 | * The child must be empty (i.e. `child->bs == NULL`) and it must be | |
2930 | * unused (i.e. not in a children list). | |
2931 | */ | |
2932 | static void bdrv_child_free(BdrvChild *child) | |
df581792 | 2933 | { |
548a74c0 | 2934 | assert(!child->bs); |
bdb73476 | 2935 | GLOBAL_STATE_CODE(); |
a225369b | 2936 | assert(!child->next.le_prev); /* not in children list */ |
04c9c3a5 HR |
2937 | |
2938 | g_free(child->name); | |
2939 | g_free(child); | |
548a74c0 | 2940 | } |
d5e6f437 | 2941 | |
548a74c0 | 2942 | typedef struct BdrvAttachChildCommonState { |
5bb04747 | 2943 | BdrvChild *child; |
548a74c0 VSO |
2944 | AioContext *old_parent_ctx; |
2945 | AioContext *old_child_ctx; | |
2946 | } BdrvAttachChildCommonState; | |
2947 | ||
2948 | static void bdrv_attach_child_common_abort(void *opaque) | |
2949 | { | |
2950 | BdrvAttachChildCommonState *s = opaque; | |
5bb04747 | 2951 | BlockDriverState *bs = s->child->bs; |
548a74c0 | 2952 | |
f0c28327 | 2953 | GLOBAL_STATE_CODE(); |
5bb04747 | 2954 | bdrv_replace_child_noperm(s->child, NULL); |
548a74c0 VSO |
2955 | |
2956 | if (bdrv_get_aio_context(bs) != s->old_child_ctx) { | |
142e6907 | 2957 | bdrv_try_change_aio_context(bs, s->old_child_ctx, NULL, &error_abort); |
548a74c0 VSO |
2958 | } |
2959 | ||
5bb04747 | 2960 | if (bdrv_child_get_parent_aio_context(s->child) != s->old_parent_ctx) { |
f8be48ad EGE |
2961 | Transaction *tran; |
2962 | GHashTable *visited; | |
2963 | bool ret; | |
548a74c0 | 2964 | |
f8be48ad | 2965 | tran = tran_new(); |
548a74c0 | 2966 | |
f8be48ad EGE |
2967 | /* No need to visit `child`, because it has been detached already */ |
2968 | visited = g_hash_table_new(NULL, NULL); | |
2969 | ret = s->child->klass->change_aio_ctx(s->child, s->old_parent_ctx, | |
2970 | visited, tran, &error_abort); | |
2971 | g_hash_table_destroy(visited); | |
2972 | ||
2973 | /* transaction is supposed to always succeed */ | |
2974 | assert(ret == true); | |
2975 | tran_commit(tran); | |
d5e6f437 KW |
2976 | } |
2977 | ||
548a74c0 | 2978 | bdrv_unref(bs); |
5bb04747 | 2979 | bdrv_child_free(s->child); |
548a74c0 VSO |
2980 | } |
2981 | ||
2982 | static TransactionActionDrv bdrv_attach_child_common_drv = { | |
2983 | .abort = bdrv_attach_child_common_abort, | |
2984 | .clean = g_free, | |
2985 | }; | |
2986 | ||
2987 | /* | |
2988 | * Common part of attaching bdrv child to bs or to blk or to job | |
f8d2ad78 | 2989 | * |
7ec390d5 | 2990 | * Function doesn't update permissions, caller is responsible for this. |
5bb04747 VSO |
2991 | * |
2992 | * Returns new created child. | |
c066e808 KW |
2993 | * |
2994 | * The caller must hold the AioContext lock for @child_bs. Both @parent_bs and | |
2995 | * @child_bs can move to a different AioContext in this function. Callers must | |
2996 | * make sure that their AioContext locking is still correct after this. | |
548a74c0 | 2997 | */ |
5bb04747 VSO |
2998 | static BdrvChild *bdrv_attach_child_common(BlockDriverState *child_bs, |
2999 | const char *child_name, | |
3000 | const BdrvChildClass *child_class, | |
3001 | BdrvChildRole child_role, | |
3002 | uint64_t perm, uint64_t shared_perm, | |
3003 | void *opaque, | |
3004 | Transaction *tran, Error **errp) | |
548a74c0 VSO |
3005 | { |
3006 | BdrvChild *new_child; | |
c066e808 | 3007 | AioContext *parent_ctx, *new_child_ctx; |
548a74c0 VSO |
3008 | AioContext *child_ctx = bdrv_get_aio_context(child_bs); |
3009 | ||
da261b69 | 3010 | assert(child_class->get_parent_desc); |
bdb73476 | 3011 | GLOBAL_STATE_CODE(); |
548a74c0 VSO |
3012 | |
3013 | new_child = g_new(BdrvChild, 1); | |
3014 | *new_child = (BdrvChild) { | |
d5e6f437 KW |
3015 | .bs = NULL, |
3016 | .name = g_strdup(child_name), | |
bd86fb99 | 3017 | .klass = child_class, |
258b7765 | 3018 | .role = child_role, |
d5e6f437 KW |
3019 | .perm = perm, |
3020 | .shared_perm = shared_perm, | |
3021 | .opaque = opaque, | |
df581792 KW |
3022 | }; |
3023 | ||
548a74c0 VSO |
3024 | /* |
3025 | * If the AioContexts don't match, first try to move the subtree of | |
132ada80 | 3026 | * child_bs into the AioContext of the new parent. If this doesn't work, |
548a74c0 VSO |
3027 | * try moving the parent into the AioContext of child_bs instead. |
3028 | */ | |
3029 | parent_ctx = bdrv_child_get_parent_aio_context(new_child); | |
3030 | if (child_ctx != parent_ctx) { | |
3031 | Error *local_err = NULL; | |
142e6907 EGE |
3032 | int ret = bdrv_try_change_aio_context(child_bs, parent_ctx, NULL, |
3033 | &local_err); | |
548a74c0 | 3034 | |
f8be48ad EGE |
3035 | if (ret < 0 && child_class->change_aio_ctx) { |
3036 | Transaction *tran = tran_new(); | |
3037 | GHashTable *visited = g_hash_table_new(NULL, NULL); | |
3038 | bool ret_child; | |
3039 | ||
3040 | g_hash_table_add(visited, new_child); | |
3041 | ret_child = child_class->change_aio_ctx(new_child, child_ctx, | |
3042 | visited, tran, NULL); | |
3043 | if (ret_child == true) { | |
132ada80 KW |
3044 | error_free(local_err); |
3045 | ret = 0; | |
132ada80 | 3046 | } |
f8be48ad EGE |
3047 | tran_finalize(tran, ret_child == true ? 0 : -1); |
3048 | g_hash_table_destroy(visited); | |
132ada80 | 3049 | } |
548a74c0 | 3050 | |
132ada80 KW |
3051 | if (ret < 0) { |
3052 | error_propagate(errp, local_err); | |
04c9c3a5 | 3053 | bdrv_child_free(new_child); |
5bb04747 | 3054 | return NULL; |
132ada80 KW |
3055 | } |
3056 | } | |
3057 | ||
c066e808 KW |
3058 | new_child_ctx = bdrv_get_aio_context(child_bs); |
3059 | if (new_child_ctx != child_ctx) { | |
3060 | aio_context_release(child_ctx); | |
3061 | aio_context_acquire(new_child_ctx); | |
3062 | } | |
3063 | ||
548a74c0 | 3064 | bdrv_ref(child_bs); |
23987471 KW |
3065 | /* |
3066 | * Let every new BdrvChild start with a drained parent. Inserting the child | |
3067 | * in the graph with bdrv_replace_child_noperm() will undrain it if | |
3068 | * @child_bs is not drained. | |
3069 | * | |
3070 | * The child was only just created and is not yet visible in global state | |
3071 | * until bdrv_replace_child_noperm() inserts it into the graph, so nobody | |
3072 | * could have sent requests and polling is not necessary. | |
3073 | * | |
3074 | * Note that this means that the parent isn't fully drained yet, we only | |
3075 | * stop new requests from coming in. This is fine, we don't care about the | |
3076 | * old requests here, they are not for this child. If another place enters a | |
3077 | * drain section for the same parent, but wants it to be fully quiesced, it | |
3078 | * will not run most of the the code in .drained_begin() again (which is not | |
3079 | * a problem, we already did this), but it will still poll until the parent | |
3080 | * is fully quiesced, so it will not be negatively affected either. | |
3081 | */ | |
606ed756 | 3082 | bdrv_parent_drained_begin_single(new_child); |
544acc7d | 3083 | bdrv_replace_child_noperm(new_child, child_bs); |
b4b059f6 | 3084 | |
548a74c0 VSO |
3085 | BdrvAttachChildCommonState *s = g_new(BdrvAttachChildCommonState, 1); |
3086 | *s = (BdrvAttachChildCommonState) { | |
5bb04747 | 3087 | .child = new_child, |
548a74c0 VSO |
3088 | .old_parent_ctx = parent_ctx, |
3089 | .old_child_ctx = child_ctx, | |
3090 | }; | |
3091 | tran_add(tran, &bdrv_attach_child_common_drv, s); | |
3092 | ||
c066e808 KW |
3093 | if (new_child_ctx != child_ctx) { |
3094 | aio_context_release(new_child_ctx); | |
3095 | aio_context_acquire(child_ctx); | |
3096 | } | |
3097 | ||
5bb04747 | 3098 | return new_child; |
548a74c0 VSO |
3099 | } |
3100 | ||
f8d2ad78 | 3101 | /* |
7ec390d5 | 3102 | * Function doesn't update permissions, caller is responsible for this. |
c066e808 KW |
3103 | * |
3104 | * The caller must hold the AioContext lock for @child_bs. Both @parent_bs and | |
3105 | * @child_bs can move to a different AioContext in this function. Callers must | |
3106 | * make sure that their AioContext locking is still correct after this. | |
f8d2ad78 | 3107 | */ |
5bb04747 VSO |
3108 | static BdrvChild *bdrv_attach_child_noperm(BlockDriverState *parent_bs, |
3109 | BlockDriverState *child_bs, | |
3110 | const char *child_name, | |
3111 | const BdrvChildClass *child_class, | |
3112 | BdrvChildRole child_role, | |
3113 | Transaction *tran, | |
3114 | Error **errp) | |
aa5a04c7 | 3115 | { |
aa5a04c7 VSO |
3116 | uint64_t perm, shared_perm; |
3117 | ||
3118 | assert(parent_bs->drv); | |
bdb73476 | 3119 | GLOBAL_STATE_CODE(); |
aa5a04c7 | 3120 | |
bfb8aa6d KW |
3121 | if (bdrv_recurse_has_child(child_bs, parent_bs)) { |
3122 | error_setg(errp, "Making '%s' a %s child of '%s' would create a cycle", | |
3123 | child_bs->node_name, child_name, parent_bs->node_name); | |
5bb04747 | 3124 | return NULL; |
bfb8aa6d KW |
3125 | } |
3126 | ||
aa5a04c7 VSO |
3127 | bdrv_get_cumulative_perm(parent_bs, &perm, &shared_perm); |
3128 | bdrv_child_perm(parent_bs, child_bs, NULL, child_role, NULL, | |
3129 | perm, shared_perm, &perm, &shared_perm); | |
3130 | ||
5bb04747 VSO |
3131 | return bdrv_attach_child_common(child_bs, child_name, child_class, |
3132 | child_role, perm, shared_perm, parent_bs, | |
3133 | tran, errp); | |
aa5a04c7 VSO |
3134 | } |
3135 | ||
548a74c0 VSO |
3136 | /* |
3137 | * This function steals the reference to child_bs from the caller. | |
3138 | * That reference is later dropped by bdrv_root_unref_child(). | |
3139 | * | |
3140 | * On failure NULL is returned, errp is set and the reference to | |
3141 | * child_bs is also dropped. | |
3142 | * | |
3143 | * The caller must hold the AioContext lock @child_bs, but not that of @ctx | |
3144 | * (unless @child_bs is already in @ctx). | |
3145 | */ | |
3146 | BdrvChild *bdrv_root_attach_child(BlockDriverState *child_bs, | |
3147 | const char *child_name, | |
3148 | const BdrvChildClass *child_class, | |
3149 | BdrvChildRole child_role, | |
3150 | uint64_t perm, uint64_t shared_perm, | |
3151 | void *opaque, Error **errp) | |
3152 | { | |
3153 | int ret; | |
5bb04747 | 3154 | BdrvChild *child; |
548a74c0 VSO |
3155 | Transaction *tran = tran_new(); |
3156 | ||
b4ad82aa EGE |
3157 | GLOBAL_STATE_CODE(); |
3158 | ||
5bb04747 | 3159 | child = bdrv_attach_child_common(child_bs, child_name, child_class, |
548a74c0 | 3160 | child_role, perm, shared_perm, opaque, |
5bb04747 VSO |
3161 | tran, errp); |
3162 | if (!child) { | |
3163 | ret = -EINVAL; | |
e878bb12 | 3164 | goto out; |
548a74c0 VSO |
3165 | } |
3166 | ||
f1316edb | 3167 | ret = bdrv_refresh_perms(child_bs, tran, errp); |
548a74c0 | 3168 | |
e878bb12 KW |
3169 | out: |
3170 | tran_finalize(tran, ret); | |
f8d2ad78 | 3171 | |
548a74c0 | 3172 | bdrv_unref(child_bs); |
5bb04747 VSO |
3173 | |
3174 | return ret < 0 ? NULL : child; | |
df581792 KW |
3175 | } |
3176 | ||
b441dc71 AG |
3177 | /* |
3178 | * This function transfers the reference to child_bs from the caller | |
3179 | * to parent_bs. That reference is later dropped by parent_bs on | |
3180 | * bdrv_close() or if someone calls bdrv_unref_child(). | |
3181 | * | |
3182 | * On failure NULL is returned, errp is set and the reference to | |
3183 | * child_bs is also dropped. | |
132ada80 KW |
3184 | * |
3185 | * If @parent_bs and @child_bs are in different AioContexts, the caller must | |
3186 | * hold the AioContext lock for @child_bs, but not for @parent_bs. | |
b441dc71 | 3187 | */ |
98292c61 WC |
3188 | BdrvChild *bdrv_attach_child(BlockDriverState *parent_bs, |
3189 | BlockDriverState *child_bs, | |
3190 | const char *child_name, | |
bd86fb99 | 3191 | const BdrvChildClass *child_class, |
258b7765 | 3192 | BdrvChildRole child_role, |
8b2ff529 | 3193 | Error **errp) |
f21d96d0 | 3194 | { |
aa5a04c7 | 3195 | int ret; |
5bb04747 | 3196 | BdrvChild *child; |
aa5a04c7 | 3197 | Transaction *tran = tran_new(); |
f68c598b | 3198 | |
f791bf7f EGE |
3199 | GLOBAL_STATE_CODE(); |
3200 | ||
5bb04747 VSO |
3201 | child = bdrv_attach_child_noperm(parent_bs, child_bs, child_name, |
3202 | child_class, child_role, tran, errp); | |
3203 | if (!child) { | |
3204 | ret = -EINVAL; | |
aa5a04c7 VSO |
3205 | goto out; |
3206 | } | |
d5e6f437 | 3207 | |
f1316edb | 3208 | ret = bdrv_refresh_perms(parent_bs, tran, errp); |
aa5a04c7 VSO |
3209 | if (ret < 0) { |
3210 | goto out; | |
d5e6f437 KW |
3211 | } |
3212 | ||
aa5a04c7 VSO |
3213 | out: |
3214 | tran_finalize(tran, ret); | |
3215 | ||
3216 | bdrv_unref(child_bs); | |
3217 | ||
5bb04747 | 3218 | return ret < 0 ? NULL : child; |
f21d96d0 KW |
3219 | } |
3220 | ||
7b99a266 | 3221 | /* Callers must ensure that child->frozen is false. */ |
f21d96d0 | 3222 | void bdrv_root_unref_child(BdrvChild *child) |
33a60407 | 3223 | { |
00eb93b5 | 3224 | BlockDriverState *child_bs = child->bs; |
779020cb | 3225 | |
f791bf7f | 3226 | GLOBAL_STATE_CODE(); |
00eb93b5 VSO |
3227 | bdrv_replace_child_noperm(child, NULL); |
3228 | bdrv_child_free(child); | |
3229 | ||
3230 | if (child_bs) { | |
3231 | /* | |
3232 | * Update permissions for old node. We're just taking a parent away, so | |
3233 | * we're loosening restrictions. Errors of permission update are not | |
3234 | * fatal in this case, ignore them. | |
3235 | */ | |
f1316edb | 3236 | bdrv_refresh_perms(child_bs, NULL, NULL); |
00eb93b5 VSO |
3237 | |
3238 | /* | |
3239 | * When the parent requiring a non-default AioContext is removed, the | |
3240 | * node moves back to the main AioContext | |
3241 | */ | |
3242 | bdrv_try_change_aio_context(child_bs, qemu_get_aio_context(), NULL, | |
3243 | NULL); | |
3244 | } | |
f791bf7f | 3245 | |
f21d96d0 KW |
3246 | bdrv_unref(child_bs); |
3247 | } | |
3248 | ||
332b3a17 VSO |
3249 | typedef struct BdrvSetInheritsFrom { |
3250 | BlockDriverState *bs; | |
3251 | BlockDriverState *old_inherits_from; | |
3252 | } BdrvSetInheritsFrom; | |
3253 | ||
3254 | static void bdrv_set_inherits_from_abort(void *opaque) | |
3255 | { | |
3256 | BdrvSetInheritsFrom *s = opaque; | |
3257 | ||
3258 | s->bs->inherits_from = s->old_inherits_from; | |
3259 | } | |
3260 | ||
3261 | static TransactionActionDrv bdrv_set_inherits_from_drv = { | |
3262 | .abort = bdrv_set_inherits_from_abort, | |
3263 | .clean = g_free, | |
3264 | }; | |
3265 | ||
3266 | /* @tran is allowed to be NULL. In this case no rollback is possible */ | |
3267 | static void bdrv_set_inherits_from(BlockDriverState *bs, | |
3268 | BlockDriverState *new_inherits_from, | |
3269 | Transaction *tran) | |
3270 | { | |
3271 | if (tran) { | |
3272 | BdrvSetInheritsFrom *s = g_new(BdrvSetInheritsFrom, 1); | |
3273 | ||
3274 | *s = (BdrvSetInheritsFrom) { | |
3275 | .bs = bs, | |
3276 | .old_inherits_from = bs->inherits_from, | |
3277 | }; | |
3278 | ||
3279 | tran_add(tran, &bdrv_set_inherits_from_drv, s); | |
3280 | } | |
3281 | ||
3282 | bs->inherits_from = new_inherits_from; | |
3283 | } | |
3284 | ||
3cf746b3 HR |
3285 | /** |
3286 | * Clear all inherits_from pointers from children and grandchildren of | |
3287 | * @root that point to @root, where necessary. | |
332b3a17 | 3288 | * @tran is allowed to be NULL. In this case no rollback is possible |
3cf746b3 | 3289 | */ |
332b3a17 VSO |
3290 | static void bdrv_unset_inherits_from(BlockDriverState *root, BdrvChild *child, |
3291 | Transaction *tran) | |
f21d96d0 | 3292 | { |
3cf746b3 | 3293 | BdrvChild *c; |
4e4bf5c4 | 3294 | |
3cf746b3 HR |
3295 | if (child->bs->inherits_from == root) { |
3296 | /* | |
3297 | * Remove inherits_from only when the last reference between root and | |
3298 | * child->bs goes away. | |
3299 | */ | |
3300 | QLIST_FOREACH(c, &root->children, next) { | |
4e4bf5c4 KW |
3301 | if (c != child && c->bs == child->bs) { |
3302 | break; | |
3303 | } | |
3304 | } | |
3305 | if (c == NULL) { | |
332b3a17 | 3306 | bdrv_set_inherits_from(child->bs, NULL, tran); |
4e4bf5c4 | 3307 | } |
33a60407 KW |
3308 | } |
3309 | ||
3cf746b3 | 3310 | QLIST_FOREACH(c, &child->bs->children, next) { |
332b3a17 | 3311 | bdrv_unset_inherits_from(root, c, tran); |
3cf746b3 HR |
3312 | } |
3313 | } | |
3314 | ||
7b99a266 | 3315 | /* Callers must ensure that child->frozen is false. */ |
3cf746b3 HR |
3316 | void bdrv_unref_child(BlockDriverState *parent, BdrvChild *child) |
3317 | { | |
f791bf7f | 3318 | GLOBAL_STATE_CODE(); |
3cf746b3 HR |
3319 | if (child == NULL) { |
3320 | return; | |
3321 | } | |
3322 | ||
332b3a17 | 3323 | bdrv_unset_inherits_from(parent, child, NULL); |
f21d96d0 | 3324 | bdrv_root_unref_child(child); |
33a60407 KW |
3325 | } |
3326 | ||
5c8cab48 KW |
3327 | |
3328 | static void bdrv_parent_cb_change_media(BlockDriverState *bs, bool load) | |
3329 | { | |
3330 | BdrvChild *c; | |
f0c28327 | 3331 | GLOBAL_STATE_CODE(); |
5c8cab48 | 3332 | QLIST_FOREACH(c, &bs->parents, next_parent) { |
bd86fb99 HR |
3333 | if (c->klass->change_media) { |
3334 | c->klass->change_media(c, load); | |
5c8cab48 KW |
3335 | } |
3336 | } | |
3337 | } | |
3338 | ||
0065c455 AG |
3339 | /* Return true if you can reach parent going through child->inherits_from |
3340 | * recursively. If parent or child are NULL, return false */ | |
3341 | static bool bdrv_inherits_from_recursive(BlockDriverState *child, | |
3342 | BlockDriverState *parent) | |
3343 | { | |
3344 | while (child && child != parent) { | |
3345 | child = child->inherits_from; | |
3346 | } | |
3347 | ||
3348 | return child != NULL; | |
3349 | } | |
3350 | ||
25191e5f HR |
3351 | /* |
3352 | * Return the BdrvChildRole for @bs's backing child. bs->backing is | |
3353 | * mostly used for COW backing children (role = COW), but also for | |
3354 | * filtered children (role = FILTERED | PRIMARY). | |
3355 | */ | |
3356 | static BdrvChildRole bdrv_backing_role(BlockDriverState *bs) | |
3357 | { | |
3358 | if (bs->drv && bs->drv->is_filter) { | |
3359 | return BDRV_CHILD_FILTERED | BDRV_CHILD_PRIMARY; | |
3360 | } else { | |
3361 | return BDRV_CHILD_COW; | |
3362 | } | |
3363 | } | |
3364 | ||
5db15a57 | 3365 | /* |
e9238278 VSO |
3366 | * Sets the bs->backing or bs->file link of a BDS. A new reference is created; |
3367 | * callers which don't need their own reference any more must call bdrv_unref(). | |
7ec390d5 VSO |
3368 | * |
3369 | * Function doesn't update permissions, caller is responsible for this. | |
4b408668 KW |
3370 | * |
3371 | * The caller must hold the AioContext lock for @child_bs. Both @parent_bs and | |
3372 | * @child_bs can move to a different AioContext in this function. Callers must | |
3373 | * make sure that their AioContext locking is still correct after this. | |
5db15a57 | 3374 | */ |
e9238278 VSO |
3375 | static int bdrv_set_file_or_backing_noperm(BlockDriverState *parent_bs, |
3376 | BlockDriverState *child_bs, | |
3377 | bool is_backing, | |
3378 | Transaction *tran, Error **errp) | |
8d24cce1 | 3379 | { |
e9238278 VSO |
3380 | bool update_inherits_from = |
3381 | bdrv_inherits_from_recursive(child_bs, parent_bs); | |
3382 | BdrvChild *child = is_backing ? parent_bs->backing : parent_bs->file; | |
3383 | BdrvChildRole role; | |
0065c455 | 3384 | |
bdb73476 EGE |
3385 | GLOBAL_STATE_CODE(); |
3386 | ||
e9238278 VSO |
3387 | if (!parent_bs->drv) { |
3388 | /* | |
3389 | * Node without drv is an object without a class :/. TODO: finally fix | |
3390 | * qcow2 driver to never clear bs->drv and implement format corruption | |
3391 | * handling in other way. | |
3392 | */ | |
3393 | error_setg(errp, "Node corrupted"); | |
3394 | return -EINVAL; | |
3395 | } | |
3396 | ||
3397 | if (child && child->frozen) { | |
3398 | error_setg(errp, "Cannot change frozen '%s' link from '%s' to '%s'", | |
3399 | child->name, parent_bs->node_name, child->bs->node_name); | |
a1e708fc | 3400 | return -EPERM; |
2cad1ebe AG |
3401 | } |
3402 | ||
25f78d9e VSO |
3403 | if (is_backing && !parent_bs->drv->is_filter && |
3404 | !parent_bs->drv->supports_backing) | |
3405 | { | |
3406 | error_setg(errp, "Driver '%s' of node '%s' does not support backing " | |
3407 | "files", parent_bs->drv->format_name, parent_bs->node_name); | |
3408 | return -EINVAL; | |
3409 | } | |
3410 | ||
e9238278 VSO |
3411 | if (parent_bs->drv->is_filter) { |
3412 | role = BDRV_CHILD_FILTERED | BDRV_CHILD_PRIMARY; | |
3413 | } else if (is_backing) { | |
3414 | role = BDRV_CHILD_COW; | |
3415 | } else { | |
3416 | /* | |
3417 | * We only can use same role as it is in existing child. We don't have | |
3418 | * infrastructure to determine role of file child in generic way | |
3419 | */ | |
3420 | if (!child) { | |
3421 | error_setg(errp, "Cannot set file child to format node without " | |
3422 | "file child"); | |
3423 | return -EINVAL; | |
3424 | } | |
3425 | role = child->role; | |
826b6ca0 FZ |
3426 | } |
3427 | ||
e9238278 VSO |
3428 | if (child) { |
3429 | bdrv_unset_inherits_from(parent_bs, child, tran); | |
57f08941 | 3430 | bdrv_remove_child(child, tran); |
e9238278 VSO |
3431 | } |
3432 | ||
3433 | if (!child_bs) { | |
8d24cce1 FZ |
3434 | goto out; |
3435 | } | |
12fa4af6 | 3436 | |
5bb04747 VSO |
3437 | child = bdrv_attach_child_noperm(parent_bs, child_bs, |
3438 | is_backing ? "backing" : "file", | |
3439 | &child_of_bds, role, | |
3440 | tran, errp); | |
3441 | if (!child) { | |
3442 | return -EINVAL; | |
a1e708fc VSO |
3443 | } |
3444 | ||
160333e1 VSO |
3445 | |
3446 | /* | |
e9238278 | 3447 | * If inherits_from pointed recursively to bs then let's update it to |
160333e1 VSO |
3448 | * point directly to bs (else it will become NULL). |
3449 | */ | |
a1e708fc | 3450 | if (update_inherits_from) { |
e9238278 | 3451 | bdrv_set_inherits_from(child_bs, parent_bs, tran); |
0065c455 | 3452 | } |
826b6ca0 | 3453 | |
8d24cce1 | 3454 | out: |
e19b157f | 3455 | bdrv_graph_rdlock_main_loop(); |
e9238278 | 3456 | bdrv_refresh_limits(parent_bs, tran, NULL); |
e19b157f | 3457 | bdrv_graph_rdunlock_main_loop(); |
160333e1 VSO |
3458 | |
3459 | return 0; | |
3460 | } | |
3461 | ||
4b408668 KW |
3462 | /* |
3463 | * The caller must hold the AioContext lock for @backing_hd. Both @bs and | |
3464 | * @backing_hd can move to a different AioContext in this function. Callers must | |
3465 | * make sure that their AioContext locking is still correct after this. | |
3466 | */ | |
e9238278 VSO |
3467 | static int bdrv_set_backing_noperm(BlockDriverState *bs, |
3468 | BlockDriverState *backing_hd, | |
3469 | Transaction *tran, Error **errp) | |
3470 | { | |
bdb73476 | 3471 | GLOBAL_STATE_CODE(); |
e9238278 VSO |
3472 | return bdrv_set_file_or_backing_noperm(bs, backing_hd, true, tran, errp); |
3473 | } | |
3474 | ||
92140b9f KW |
3475 | int bdrv_set_backing_hd_drained(BlockDriverState *bs, |
3476 | BlockDriverState *backing_hd, | |
3477 | Error **errp) | |
160333e1 VSO |
3478 | { |
3479 | int ret; | |
3480 | Transaction *tran = tran_new(); | |
3481 | ||
f791bf7f | 3482 | GLOBAL_STATE_CODE(); |
92140b9f | 3483 | assert(bs->quiesce_counter > 0); |
c0829cb1 | 3484 | |
160333e1 VSO |
3485 | ret = bdrv_set_backing_noperm(bs, backing_hd, tran, errp); |
3486 | if (ret < 0) { | |
3487 | goto out; | |
3488 | } | |
3489 | ||
f1316edb | 3490 | ret = bdrv_refresh_perms(bs, tran, errp); |
160333e1 VSO |
3491 | out: |
3492 | tran_finalize(tran, ret); | |
92140b9f KW |
3493 | return ret; |
3494 | } | |
a1e708fc | 3495 | |
92140b9f KW |
3496 | int bdrv_set_backing_hd(BlockDriverState *bs, BlockDriverState *backing_hd, |
3497 | Error **errp) | |
3498 | { | |
3499 | int ret; | |
3500 | GLOBAL_STATE_CODE(); | |
3501 | ||
3502 | bdrv_drained_begin(bs); | |
3503 | ret = bdrv_set_backing_hd_drained(bs, backing_hd, errp); | |
c0829cb1 VSO |
3504 | bdrv_drained_end(bs); |
3505 | ||
a1e708fc | 3506 | return ret; |
8d24cce1 FZ |
3507 | } |
3508 | ||
31ca6d07 KW |
3509 | /* |
3510 | * Opens the backing file for a BlockDriverState if not yet open | |
3511 | * | |
d9b7b057 KW |
3512 | * bdref_key specifies the key for the image's BlockdevRef in the options QDict. |
3513 | * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict | |
3514 | * itself, all options starting with "${bdref_key}." are considered part of the | |
3515 | * BlockdevRef. | |
3516 | * | |
2626d27f KW |
3517 | * The caller must hold the main AioContext lock. |
3518 | * | |
d9b7b057 | 3519 | * TODO Can this be unified with bdrv_open_image()? |
31ca6d07 | 3520 | */ |
d9b7b057 KW |
3521 | int bdrv_open_backing_file(BlockDriverState *bs, QDict *parent_options, |
3522 | const char *bdref_key, Error **errp) | |
9156df12 | 3523 | { |
6b6833c1 | 3524 | char *backing_filename = NULL; |
d9b7b057 KW |
3525 | char *bdref_key_dot; |
3526 | const char *reference = NULL; | |
317fc44e | 3527 | int ret = 0; |
998c2019 | 3528 | bool implicit_backing = false; |
8d24cce1 | 3529 | BlockDriverState *backing_hd; |
8aa04542 | 3530 | AioContext *backing_hd_ctx; |
d9b7b057 KW |
3531 | QDict *options; |
3532 | QDict *tmp_parent_options = NULL; | |
34b5d2c6 | 3533 | Error *local_err = NULL; |
9156df12 | 3534 | |
f791bf7f EGE |
3535 | GLOBAL_STATE_CODE(); |
3536 | ||
760e0063 | 3537 | if (bs->backing != NULL) { |
1ba4b6a5 | 3538 | goto free_exit; |
9156df12 PB |
3539 | } |
3540 | ||
31ca6d07 | 3541 | /* NULL means an empty set of options */ |
d9b7b057 KW |
3542 | if (parent_options == NULL) { |
3543 | tmp_parent_options = qdict_new(); | |
3544 | parent_options = tmp_parent_options; | |
31ca6d07 KW |
3545 | } |
3546 | ||
9156df12 | 3547 | bs->open_flags &= ~BDRV_O_NO_BACKING; |
d9b7b057 KW |
3548 | |
3549 | bdref_key_dot = g_strdup_printf("%s.", bdref_key); | |
3550 | qdict_extract_subqdict(parent_options, &options, bdref_key_dot); | |
3551 | g_free(bdref_key_dot); | |
3552 | ||
129c7d1c MA |
3553 | /* |
3554 | * Caution: while qdict_get_try_str() is fine, getting non-string | |
3555 | * types would require more care. When @parent_options come from | |
3556 | * -blockdev or blockdev_add, its members are typed according to | |
3557 | * the QAPI schema, but when they come from -drive, they're all | |
3558 | * QString. | |
3559 | */ | |
d9b7b057 KW |
3560 | reference = qdict_get_try_str(parent_options, bdref_key); |
3561 | if (reference || qdict_haskey(options, "file.filename")) { | |
6b6833c1 | 3562 | /* keep backing_filename NULL */ |
1cb6f506 | 3563 | } else if (bs->backing_file[0] == '\0' && qdict_size(options) == 0) { |
cb3e7f08 | 3564 | qobject_unref(options); |
1ba4b6a5 | 3565 | goto free_exit; |
dbecebdd | 3566 | } else { |
998c2019 HR |
3567 | if (qdict_size(options) == 0) { |
3568 | /* If the user specifies options that do not modify the | |
3569 | * backing file's behavior, we might still consider it the | |
3570 | * implicit backing file. But it's easier this way, and | |
3571 | * just specifying some of the backing BDS's options is | |
3572 | * only possible with -drive anyway (otherwise the QAPI | |
3573 | * schema forces the user to specify everything). */ | |
3574 | implicit_backing = !strcmp(bs->auto_backing_file, bs->backing_file); | |
3575 | } | |
3576 | ||
6b6833c1 | 3577 | backing_filename = bdrv_get_full_backing_filename(bs, &local_err); |
9f07429e HR |
3578 | if (local_err) { |
3579 | ret = -EINVAL; | |
3580 | error_propagate(errp, local_err); | |
cb3e7f08 | 3581 | qobject_unref(options); |
9f07429e HR |
3582 | goto free_exit; |
3583 | } | |
9156df12 PB |
3584 | } |
3585 | ||
8ee79e70 KW |
3586 | if (!bs->drv || !bs->drv->supports_backing) { |
3587 | ret = -EINVAL; | |
3588 | error_setg(errp, "Driver doesn't support backing files"); | |
cb3e7f08 | 3589 | qobject_unref(options); |
8ee79e70 KW |
3590 | goto free_exit; |
3591 | } | |
3592 | ||
6bff597b PK |
3593 | if (!reference && |
3594 | bs->backing_format[0] != '\0' && !qdict_haskey(options, "driver")) { | |
46f5ac20 | 3595 | qdict_put_str(options, "driver", bs->backing_format); |
9156df12 PB |
3596 | } |
3597 | ||
6b6833c1 | 3598 | backing_hd = bdrv_open_inherit(backing_filename, reference, options, 0, bs, |
25191e5f | 3599 | &child_of_bds, bdrv_backing_role(bs), errp); |
5b363937 | 3600 | if (!backing_hd) { |
9156df12 | 3601 | bs->open_flags |= BDRV_O_NO_BACKING; |
e43bfd9c | 3602 | error_prepend(errp, "Could not open backing file: "); |
5b363937 | 3603 | ret = -EINVAL; |
1ba4b6a5 | 3604 | goto free_exit; |
9156df12 | 3605 | } |
df581792 | 3606 | |
998c2019 HR |
3607 | if (implicit_backing) { |
3608 | bdrv_refresh_filename(backing_hd); | |
3609 | pstrcpy(bs->auto_backing_file, sizeof(bs->auto_backing_file), | |
3610 | backing_hd->filename); | |
3611 | } | |
3612 | ||
5db15a57 KW |
3613 | /* Hook up the backing file link; drop our reference, bs owns the |
3614 | * backing_hd reference now */ | |
8aa04542 KW |
3615 | backing_hd_ctx = bdrv_get_aio_context(backing_hd); |
3616 | aio_context_acquire(backing_hd_ctx); | |
dc9c10a1 | 3617 | ret = bdrv_set_backing_hd(bs, backing_hd, errp); |
5db15a57 | 3618 | bdrv_unref(backing_hd); |
8aa04542 KW |
3619 | aio_context_release(backing_hd_ctx); |
3620 | ||
dc9c10a1 | 3621 | if (ret < 0) { |
12fa4af6 KW |
3622 | goto free_exit; |
3623 | } | |
d80ac658 | 3624 | |
d9b7b057 KW |
3625 | qdict_del(parent_options, bdref_key); |
3626 | ||
1ba4b6a5 BC |
3627 | free_exit: |
3628 | g_free(backing_filename); | |
cb3e7f08 | 3629 | qobject_unref(tmp_parent_options); |
1ba4b6a5 | 3630 | return ret; |
9156df12 PB |
3631 | } |
3632 | ||
2d6b86af KW |
3633 | static BlockDriverState * |
3634 | bdrv_open_child_bs(const char *filename, QDict *options, const char *bdref_key, | |
bd86fb99 | 3635 | BlockDriverState *parent, const BdrvChildClass *child_class, |
272c02ea | 3636 | BdrvChildRole child_role, bool allow_none, Error **errp) |
da557aac | 3637 | { |
2d6b86af | 3638 | BlockDriverState *bs = NULL; |
da557aac | 3639 | QDict *image_options; |
da557aac HR |
3640 | char *bdref_key_dot; |
3641 | const char *reference; | |
3642 | ||
bd86fb99 | 3643 | assert(child_class != NULL); |
f67503e5 | 3644 | |
da557aac HR |
3645 | bdref_key_dot = g_strdup_printf("%s.", bdref_key); |
3646 | qdict_extract_subqdict(options, &image_options, bdref_key_dot); | |
3647 | g_free(bdref_key_dot); | |
3648 | ||
129c7d1c MA |
3649 | /* |
3650 | * Caution: while qdict_get_try_str() is fine, getting non-string | |
3651 | * types would require more care. When @options come from | |
3652 | * -blockdev or blockdev_add, its members are typed according to | |
3653 | * the QAPI schema, but when they come from -drive, they're all | |
3654 | * QString. | |
3655 | */ | |
da557aac HR |
3656 | reference = qdict_get_try_str(options, bdref_key); |
3657 | if (!filename && !reference && !qdict_size(image_options)) { | |
b4b059f6 | 3658 | if (!allow_none) { |
da557aac HR |
3659 | error_setg(errp, "A block device must be specified for \"%s\"", |
3660 | bdref_key); | |
da557aac | 3661 | } |
cb3e7f08 | 3662 | qobject_unref(image_options); |
da557aac HR |
3663 | goto done; |
3664 | } | |
3665 | ||
5b363937 | 3666 | bs = bdrv_open_inherit(filename, reference, image_options, 0, |
272c02ea | 3667 | parent, child_class, child_role, errp); |
5b363937 | 3668 | if (!bs) { |
df581792 KW |
3669 | goto done; |
3670 | } | |
3671 | ||
da557aac HR |
3672 | done: |
3673 | qdict_del(options, bdref_key); | |
2d6b86af KW |
3674 | return bs; |
3675 | } | |
3676 | ||
3677 | /* | |
3678 | * Opens a disk image whose options are given as BlockdevRef in another block | |
3679 | * device's options. | |
3680 | * | |
3681 | * If allow_none is true, no image will be opened if filename is false and no | |
3682 | * BlockdevRef is given. NULL will be returned, but errp remains unset. | |
3683 | * | |
3684 | * bdrev_key specifies the key for the image's BlockdevRef in the options QDict. | |
3685 | * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict | |
3686 | * itself, all options starting with "${bdref_key}." are considered part of the | |
3687 | * BlockdevRef. | |
3688 | * | |
3689 | * The BlockdevRef will be removed from the options QDict. | |
aa269ff8 | 3690 | * |
8394c35e | 3691 | * The caller must hold the lock of the main AioContext and no other AioContext. |
aa269ff8 KW |
3692 | * @parent can move to a different AioContext in this function. Callers must |
3693 | * make sure that their AioContext locking is still correct after this. | |
2d6b86af KW |
3694 | */ |
3695 | BdrvChild *bdrv_open_child(const char *filename, | |
3696 | QDict *options, const char *bdref_key, | |
3697 | BlockDriverState *parent, | |
bd86fb99 | 3698 | const BdrvChildClass *child_class, |
258b7765 | 3699 | BdrvChildRole child_role, |
2d6b86af KW |
3700 | bool allow_none, Error **errp) |
3701 | { | |
3702 | BlockDriverState *bs; | |
8394c35e KW |
3703 | BdrvChild *child; |
3704 | AioContext *ctx; | |
2d6b86af | 3705 | |
f791bf7f EGE |
3706 | GLOBAL_STATE_CODE(); |
3707 | ||
bd86fb99 | 3708 | bs = bdrv_open_child_bs(filename, options, bdref_key, parent, child_class, |
272c02ea | 3709 | child_role, allow_none, errp); |
2d6b86af KW |
3710 | if (bs == NULL) { |
3711 | return NULL; | |
3712 | } | |
3713 | ||
8394c35e KW |
3714 | ctx = bdrv_get_aio_context(bs); |
3715 | aio_context_acquire(ctx); | |
3716 | child = bdrv_attach_child(parent, bs, bdref_key, child_class, child_role, | |
3717 | errp); | |
3718 | aio_context_release(ctx); | |
3719 | ||
3720 | return child; | |
b4b059f6 KW |
3721 | } |
3722 | ||
83930780 VSO |
3723 | /* |
3724 | * Wrapper on bdrv_open_child() for most popular case: open primary child of bs. | |
aa269ff8 | 3725 | * |
8394c35e | 3726 | * The caller must hold the lock of the main AioContext and no other AioContext. |
aa269ff8 KW |
3727 | * @parent can move to a different AioContext in this function. Callers must |
3728 | * make sure that their AioContext locking is still correct after this. | |
83930780 VSO |
3729 | */ |
3730 | int bdrv_open_file_child(const char *filename, | |
3731 | QDict *options, const char *bdref_key, | |
3732 | BlockDriverState *parent, Error **errp) | |
3733 | { | |
3734 | BdrvChildRole role; | |
3735 | ||
3736 | /* commit_top and mirror_top don't use this function */ | |
3737 | assert(!parent->drv->filtered_child_is_backing); | |
83930780 VSO |
3738 | role = parent->drv->is_filter ? |
3739 | (BDRV_CHILD_FILTERED | BDRV_CHILD_PRIMARY) : BDRV_CHILD_IMAGE; | |
3740 | ||
5bb04747 VSO |
3741 | if (!bdrv_open_child(filename, options, bdref_key, parent, |
3742 | &child_of_bds, role, false, errp)) | |
3743 | { | |
3744 | return -EINVAL; | |
3745 | } | |
83930780 | 3746 | |
5bb04747 | 3747 | return 0; |
83930780 VSO |
3748 | } |
3749 | ||
bd86fb99 HR |
3750 | /* |
3751 | * TODO Future callers may need to specify parent/child_class in order for | |
3752 | * option inheritance to work. Existing callers use it for the root node. | |
3753 | */ | |
e1d74bc6 KW |
3754 | BlockDriverState *bdrv_open_blockdev_ref(BlockdevRef *ref, Error **errp) |
3755 | { | |
3756 | BlockDriverState *bs = NULL; | |
e1d74bc6 KW |
3757 | QObject *obj = NULL; |
3758 | QDict *qdict = NULL; | |
3759 | const char *reference = NULL; | |
3760 | Visitor *v = NULL; | |
3761 | ||
f791bf7f EGE |
3762 | GLOBAL_STATE_CODE(); |
3763 | ||
e1d74bc6 KW |
3764 | if (ref->type == QTYPE_QSTRING) { |
3765 | reference = ref->u.reference; | |
3766 | } else { | |
3767 | BlockdevOptions *options = &ref->u.definition; | |
3768 | assert(ref->type == QTYPE_QDICT); | |
3769 | ||
3770 | v = qobject_output_visitor_new(&obj); | |
1f584248 | 3771 | visit_type_BlockdevOptions(v, NULL, &options, &error_abort); |
e1d74bc6 KW |
3772 | visit_complete(v, &obj); |
3773 | ||
7dc847eb | 3774 | qdict = qobject_to(QDict, obj); |
e1d74bc6 KW |
3775 | qdict_flatten(qdict); |
3776 | ||
3777 | /* bdrv_open_inherit() defaults to the values in bdrv_flags (for | |
3778 | * compatibility with other callers) rather than what we want as the | |
3779 | * real defaults. Apply the defaults here instead. */ | |
3780 | qdict_set_default_str(qdict, BDRV_OPT_CACHE_DIRECT, "off"); | |
3781 | qdict_set_default_str(qdict, BDRV_OPT_CACHE_NO_FLUSH, "off"); | |
3782 | qdict_set_default_str(qdict, BDRV_OPT_READ_ONLY, "off"); | |
e35bdc12 KW |
3783 | qdict_set_default_str(qdict, BDRV_OPT_AUTO_READ_ONLY, "off"); |
3784 | ||
e1d74bc6 KW |
3785 | } |
3786 | ||
272c02ea | 3787 | bs = bdrv_open_inherit(NULL, reference, qdict, 0, NULL, NULL, 0, errp); |
e1d74bc6 | 3788 | obj = NULL; |
cb3e7f08 | 3789 | qobject_unref(obj); |
e1d74bc6 KW |
3790 | visit_free(v); |
3791 | return bs; | |
3792 | } | |
3793 | ||
66836189 HR |
3794 | static BlockDriverState *bdrv_append_temp_snapshot(BlockDriverState *bs, |
3795 | int flags, | |
3796 | QDict *snapshot_options, | |
3797 | Error **errp) | |
b998875d | 3798 | { |
69fbfff9 | 3799 | g_autofree char *tmp_filename = NULL; |
b998875d | 3800 | int64_t total_size; |
83d0521a | 3801 | QemuOpts *opts = NULL; |
ff6ed714 | 3802 | BlockDriverState *bs_snapshot = NULL; |
f665f01f | 3803 | AioContext *ctx = bdrv_get_aio_context(bs); |
b998875d KW |
3804 | int ret; |
3805 | ||
bdb73476 EGE |
3806 | GLOBAL_STATE_CODE(); |
3807 | ||
b998875d KW |
3808 | /* if snapshot, we create a temporary backing file and open it |
3809 | instead of opening 'filename' directly */ | |
3810 | ||
3811 | /* Get the required size from the image */ | |
f665f01f | 3812 | aio_context_acquire(ctx); |
f187743a | 3813 | total_size = bdrv_getlength(bs); |
f665f01f KW |
3814 | aio_context_release(ctx); |
3815 | ||
f187743a KW |
3816 | if (total_size < 0) { |
3817 | error_setg_errno(errp, -total_size, "Could not get image size"); | |
1ba4b6a5 | 3818 | goto out; |
f187743a | 3819 | } |
b998875d KW |
3820 | |
3821 | /* Create the temporary image */ | |
69fbfff9 BM |
3822 | tmp_filename = create_tmp_file(errp); |
3823 | if (!tmp_filename) { | |
1ba4b6a5 | 3824 | goto out; |
b998875d KW |
3825 | } |
3826 | ||
ef810437 | 3827 | opts = qemu_opts_create(bdrv_qcow2.create_opts, NULL, 0, |
c282e1fd | 3828 | &error_abort); |
39101f25 | 3829 | qemu_opt_set_number(opts, BLOCK_OPT_SIZE, total_size, &error_abort); |
e43bfd9c | 3830 | ret = bdrv_create(&bdrv_qcow2, tmp_filename, opts, errp); |
83d0521a | 3831 | qemu_opts_del(opts); |
b998875d | 3832 | if (ret < 0) { |
e43bfd9c MA |
3833 | error_prepend(errp, "Could not create temporary overlay '%s': ", |
3834 | tmp_filename); | |
1ba4b6a5 | 3835 | goto out; |
b998875d KW |
3836 | } |
3837 | ||
73176bee | 3838 | /* Prepare options QDict for the temporary file */ |
46f5ac20 EB |
3839 | qdict_put_str(snapshot_options, "file.driver", "file"); |
3840 | qdict_put_str(snapshot_options, "file.filename", tmp_filename); | |
3841 | qdict_put_str(snapshot_options, "driver", "qcow2"); | |
b998875d | 3842 | |
5b363937 | 3843 | bs_snapshot = bdrv_open(NULL, NULL, snapshot_options, flags, errp); |
73176bee | 3844 | snapshot_options = NULL; |
5b363937 | 3845 | if (!bs_snapshot) { |
1ba4b6a5 | 3846 | goto out; |
b998875d KW |
3847 | } |
3848 | ||
f665f01f | 3849 | aio_context_acquire(ctx); |
934aee14 | 3850 | ret = bdrv_append(bs_snapshot, bs, errp); |
f665f01f KW |
3851 | aio_context_release(ctx); |
3852 | ||
934aee14 | 3853 | if (ret < 0) { |
ff6ed714 | 3854 | bs_snapshot = NULL; |
b2c2832c KW |
3855 | goto out; |
3856 | } | |
1ba4b6a5 BC |
3857 | |
3858 | out: | |
cb3e7f08 | 3859 | qobject_unref(snapshot_options); |
ff6ed714 | 3860 | return bs_snapshot; |
b998875d KW |
3861 | } |
3862 | ||
b6ce07aa KW |
3863 | /* |
3864 | * Opens a disk image (raw, qcow2, vmdk, ...) | |
de9c0cec KW |
3865 | * |
3866 | * options is a QDict of options to pass to the block drivers, or NULL for an | |
3867 | * empty set of options. The reference to the QDict belongs to the block layer | |
3868 | * after the call (even on failure), so if the caller intends to reuse the | |
cb3e7f08 | 3869 | * dictionary, it needs to use qobject_ref() before calling bdrv_open. |
f67503e5 HR |
3870 | * |
3871 | * If *pbs is NULL, a new BDS will be created with a pointer to it stored there. | |
3872 | * If it is not NULL, the referenced BDS will be reused. | |
ddf5636d HR |
3873 | * |
3874 | * The reference parameter may be used to specify an existing block device which | |
3875 | * should be opened. If specified, neither options nor a filename may be given, | |
3876 | * nor can an existing BDS be reused (that is, *pbs has to be NULL). | |
c86422c5 | 3877 | * |
ae400dbb | 3878 | * The caller must always hold the main AioContext lock. |
b6ce07aa | 3879 | */ |
32192301 KW |
3880 | static BlockDriverState * no_coroutine_fn |
3881 | bdrv_open_inherit(const char *filename, const char *reference, QDict *options, | |
3882 | int flags, BlockDriverState *parent, | |
3883 | const BdrvChildClass *child_class, BdrvChildRole child_role, | |
3884 | Error **errp) | |
ea2384d3 | 3885 | { |
b6ce07aa | 3886 | int ret; |
5696c6e3 | 3887 | BlockBackend *file = NULL; |
9a4f4c31 | 3888 | BlockDriverState *bs; |
ce343771 | 3889 | BlockDriver *drv = NULL; |
2f624b80 | 3890 | BdrvChild *child; |
74fe54f2 | 3891 | const char *drvname; |
3e8c2e57 | 3892 | const char *backing; |
34b5d2c6 | 3893 | Error *local_err = NULL; |
73176bee | 3894 | QDict *snapshot_options = NULL; |
b1e6fc08 | 3895 | int snapshot_flags = 0; |
f665f01f | 3896 | AioContext *ctx = qemu_get_aio_context(); |
712e7874 | 3897 | |
bd86fb99 HR |
3898 | assert(!child_class || !flags); |
3899 | assert(!child_class == !parent); | |
f0c28327 | 3900 | GLOBAL_STATE_CODE(); |
32192301 | 3901 | assert(!qemu_in_coroutine()); |
f67503e5 | 3902 | |
ddf5636d HR |
3903 | if (reference) { |
3904 | bool options_non_empty = options ? qdict_size(options) : false; | |
cb3e7f08 | 3905 | qobject_unref(options); |
ddf5636d | 3906 | |
ddf5636d HR |
3907 | if (filename || options_non_empty) { |
3908 | error_setg(errp, "Cannot reference an existing block device with " | |
3909 | "additional options or a new filename"); | |
5b363937 | 3910 | return NULL; |
ddf5636d HR |
3911 | } |
3912 | ||
3913 | bs = bdrv_lookup_bs(reference, reference, errp); | |
3914 | if (!bs) { | |
5b363937 | 3915 | return NULL; |
ddf5636d | 3916 | } |
76b22320 | 3917 | |
ddf5636d | 3918 | bdrv_ref(bs); |
5b363937 | 3919 | return bs; |
ddf5636d HR |
3920 | } |
3921 | ||
5b363937 | 3922 | bs = bdrv_new(); |
f67503e5 | 3923 | |
de9c0cec KW |
3924 | /* NULL means an empty set of options */ |
3925 | if (options == NULL) { | |
3926 | options = qdict_new(); | |
3927 | } | |
3928 | ||
145f598e | 3929 | /* json: syntax counts as explicit options, as if in the QDict */ |
de3b53f0 KW |
3930 | parse_json_protocol(options, &filename, &local_err); |
3931 | if (local_err) { | |
de3b53f0 KW |
3932 | goto fail; |
3933 | } | |
3934 | ||
145f598e KW |
3935 | bs->explicit_options = qdict_clone_shallow(options); |
3936 | ||
bd86fb99 | 3937 | if (child_class) { |
3cdc69d3 HR |
3938 | bool parent_is_format; |
3939 | ||
3940 | if (parent->drv) { | |
3941 | parent_is_format = parent->drv->is_format; | |
3942 | } else { | |
3943 | /* | |
3944 | * parent->drv is not set yet because this node is opened for | |
3945 | * (potential) format probing. That means that @parent is going | |
3946 | * to be a format node. | |
3947 | */ | |
3948 | parent_is_format = true; | |
3949 | } | |
3950 | ||
bddcec37 | 3951 | bs->inherits_from = parent; |
3cdc69d3 HR |
3952 | child_class->inherit_options(child_role, parent_is_format, |
3953 | &flags, options, | |
bd86fb99 | 3954 | parent->open_flags, parent->options); |
f3930ed0 KW |
3955 | } |
3956 | ||
de3b53f0 | 3957 | ret = bdrv_fill_options(&options, filename, &flags, &local_err); |
dfde483e | 3958 | if (ret < 0) { |
462f5bcf KW |
3959 | goto fail; |
3960 | } | |
3961 | ||
129c7d1c MA |
3962 | /* |
3963 | * Set the BDRV_O_RDWR and BDRV_O_ALLOW_RDWR flags. | |
3964 | * Caution: getting a boolean member of @options requires care. | |
3965 | * When @options come from -blockdev or blockdev_add, members are | |
3966 | * typed according to the QAPI schema, but when they come from | |
3967 | * -drive, they're all QString. | |
3968 | */ | |
f87a0e29 AG |
3969 | if (g_strcmp0(qdict_get_try_str(options, BDRV_OPT_READ_ONLY), "on") && |
3970 | !qdict_get_try_bool(options, BDRV_OPT_READ_ONLY, false)) { | |
3971 | flags |= (BDRV_O_RDWR | BDRV_O_ALLOW_RDWR); | |
3972 | } else { | |
3973 | flags &= ~BDRV_O_RDWR; | |
14499ea5 AG |
3974 | } |
3975 | ||
3976 | if (flags & BDRV_O_SNAPSHOT) { | |
3977 | snapshot_options = qdict_new(); | |
3978 | bdrv_temp_snapshot_options(&snapshot_flags, snapshot_options, | |
3979 | flags, options); | |
f87a0e29 AG |
3980 | /* Let bdrv_backing_options() override "read-only" */ |
3981 | qdict_del(options, BDRV_OPT_READ_ONLY); | |
00ff7ffd HR |
3982 | bdrv_inherited_options(BDRV_CHILD_COW, true, |
3983 | &flags, options, flags, options); | |
14499ea5 AG |
3984 | } |
3985 | ||
62392ebb KW |
3986 | bs->open_flags = flags; |
3987 | bs->options = options; | |
3988 | options = qdict_clone_shallow(options); | |
3989 | ||
76c591b0 | 3990 | /* Find the right image format driver */ |
129c7d1c | 3991 | /* See cautionary note on accessing @options above */ |
76c591b0 KW |
3992 | drvname = qdict_get_try_str(options, "driver"); |
3993 | if (drvname) { | |
3994 | drv = bdrv_find_format(drvname); | |
76c591b0 KW |
3995 | if (!drv) { |
3996 | error_setg(errp, "Unknown driver: '%s'", drvname); | |
76c591b0 KW |
3997 | goto fail; |
3998 | } | |
3999 | } | |
4000 | ||
4001 | assert(drvname || !(flags & BDRV_O_PROTOCOL)); | |
76c591b0 | 4002 | |
129c7d1c | 4003 | /* See cautionary note on accessing @options above */ |
3e8c2e57 | 4004 | backing = qdict_get_try_str(options, "backing"); |
e59a0cf1 HR |
4005 | if (qobject_to(QNull, qdict_get(options, "backing")) != NULL || |
4006 | (backing && *backing == '\0')) | |
4007 | { | |
4f7be280 HR |
4008 | if (backing) { |
4009 | warn_report("Use of \"backing\": \"\" is deprecated; " | |
4010 | "use \"backing\": null instead"); | |
4011 | } | |
3e8c2e57 | 4012 | flags |= BDRV_O_NO_BACKING; |
ae0f57f0 KW |
4013 | qdict_del(bs->explicit_options, "backing"); |
4014 | qdict_del(bs->options, "backing"); | |
3e8c2e57 AG |
4015 | qdict_del(options, "backing"); |
4016 | } | |
4017 | ||
5696c6e3 | 4018 | /* Open image file without format layer. This BlockBackend is only used for |
4e4bf5c4 KW |
4019 | * probing, the block drivers will do their own bdrv_open_child() for the |
4020 | * same BDS, which is why we put the node name back into options. */ | |
f4788adc | 4021 | if ((flags & BDRV_O_PROTOCOL) == 0) { |
5696c6e3 KW |
4022 | BlockDriverState *file_bs; |
4023 | ||
4024 | file_bs = bdrv_open_child_bs(filename, options, "file", bs, | |
58944401 HR |
4025 | &child_of_bds, BDRV_CHILD_IMAGE, |
4026 | true, &local_err); | |
1fdd6933 | 4027 | if (local_err) { |
f4788adc KW |
4028 | goto fail; |
4029 | } | |
5696c6e3 | 4030 | if (file_bs != NULL) { |
dacaa162 KW |
4031 | /* Not requesting BLK_PERM_CONSISTENT_READ because we're only |
4032 | * looking at the header to guess the image format. This works even | |
4033 | * in cases where a guest would not see a consistent state. */ | |
f665f01f KW |
4034 | ctx = bdrv_get_aio_context(file_bs); |
4035 | aio_context_acquire(ctx); | |
4036 | file = blk_new(ctx, 0, BLK_PERM_ALL); | |
d7086422 | 4037 | blk_insert_bs(file, file_bs, &local_err); |
5696c6e3 | 4038 | bdrv_unref(file_bs); |
f665f01f KW |
4039 | aio_context_release(ctx); |
4040 | ||
d7086422 KW |
4041 | if (local_err) { |
4042 | goto fail; | |
4043 | } | |
5696c6e3 | 4044 | |
46f5ac20 | 4045 | qdict_put_str(options, "file", bdrv_get_node_name(file_bs)); |
4e4bf5c4 | 4046 | } |
f500a6d3 KW |
4047 | } |
4048 | ||
76c591b0 | 4049 | /* Image format probing */ |
38f3ef57 | 4050 | bs->probed = !drv; |
76c591b0 | 4051 | if (!drv && file) { |
cf2ab8fc | 4052 | ret = find_image_format(file, filename, &drv, &local_err); |
17b005f1 | 4053 | if (ret < 0) { |
8bfea15d | 4054 | goto fail; |
2a05cbe4 | 4055 | } |
62392ebb KW |
4056 | /* |
4057 | * This option update would logically belong in bdrv_fill_options(), | |
4058 | * but we first need to open bs->file for the probing to work, while | |
4059 | * opening bs->file already requires the (mostly) final set of options | |
4060 | * so that cache mode etc. can be inherited. | |
4061 | * | |
4062 | * Adding the driver later is somewhat ugly, but it's not an option | |
4063 | * that would ever be inherited, so it's correct. We just need to make | |
4064 | * sure to update both bs->options (which has the full effective | |
4065 | * options for bs) and options (which has file.* already removed). | |
4066 | */ | |
46f5ac20 EB |
4067 | qdict_put_str(bs->options, "driver", drv->format_name); |
4068 | qdict_put_str(options, "driver", drv->format_name); | |
76c591b0 | 4069 | } else if (!drv) { |
17b005f1 | 4070 | error_setg(errp, "Must specify either driver or file"); |
8bfea15d | 4071 | goto fail; |
ea2384d3 | 4072 | } |
b6ce07aa | 4073 | |
53a29513 HR |
4074 | /* BDRV_O_PROTOCOL must be set iff a protocol BDS is about to be created */ |
4075 | assert(!!(flags & BDRV_O_PROTOCOL) == !!drv->bdrv_file_open); | |
4076 | /* file must be NULL if a protocol BDS is about to be created | |
4077 | * (the inverse results in an error message from bdrv_open_common()) */ | |
4078 | assert(!(flags & BDRV_O_PROTOCOL) || !file); | |
4079 | ||
b6ce07aa | 4080 | /* Open the image */ |
82dc8b41 | 4081 | ret = bdrv_open_common(bs, file, options, &local_err); |
b6ce07aa | 4082 | if (ret < 0) { |
8bfea15d | 4083 | goto fail; |
6987307c CH |
4084 | } |
4085 | ||
f665f01f KW |
4086 | /* The AioContext could have changed during bdrv_open_common() */ |
4087 | ctx = bdrv_get_aio_context(bs); | |
4088 | ||
4e4bf5c4 | 4089 | if (file) { |
f665f01f | 4090 | aio_context_acquire(ctx); |
5696c6e3 | 4091 | blk_unref(file); |
f665f01f | 4092 | aio_context_release(ctx); |
f500a6d3 KW |
4093 | file = NULL; |
4094 | } | |
4095 | ||
b6ce07aa | 4096 | /* If there is a backing file, use it */ |
9156df12 | 4097 | if ((flags & BDRV_O_NO_BACKING) == 0) { |
d9b7b057 | 4098 | ret = bdrv_open_backing_file(bs, options, "backing", &local_err); |
b6ce07aa | 4099 | if (ret < 0) { |
b6ad491a | 4100 | goto close_and_fail; |
b6ce07aa | 4101 | } |
b6ce07aa KW |
4102 | } |
4103 | ||
50196d7a AG |
4104 | /* Remove all children options and references |
4105 | * from bs->options and bs->explicit_options */ | |
2f624b80 AG |
4106 | QLIST_FOREACH(child, &bs->children, next) { |
4107 | char *child_key_dot; | |
4108 | child_key_dot = g_strdup_printf("%s.", child->name); | |
4109 | qdict_extract_subqdict(bs->explicit_options, NULL, child_key_dot); | |
4110 | qdict_extract_subqdict(bs->options, NULL, child_key_dot); | |
50196d7a AG |
4111 | qdict_del(bs->explicit_options, child->name); |
4112 | qdict_del(bs->options, child->name); | |
2f624b80 AG |
4113 | g_free(child_key_dot); |
4114 | } | |
4115 | ||
b6ad491a | 4116 | /* Check if any unknown options were used */ |
7ad2757f | 4117 | if (qdict_size(options) != 0) { |
b6ad491a | 4118 | const QDictEntry *entry = qdict_first(options); |
5acd9d81 HR |
4119 | if (flags & BDRV_O_PROTOCOL) { |
4120 | error_setg(errp, "Block protocol '%s' doesn't support the option " | |
4121 | "'%s'", drv->format_name, entry->key); | |
4122 | } else { | |
d0e46a55 HR |
4123 | error_setg(errp, |
4124 | "Block format '%s' does not support the option '%s'", | |
4125 | drv->format_name, entry->key); | |
5acd9d81 | 4126 | } |
b6ad491a | 4127 | |
b6ad491a KW |
4128 | goto close_and_fail; |
4129 | } | |
b6ad491a | 4130 | |
c01c214b | 4131 | bdrv_parent_cb_change_media(bs, true); |
b6ce07aa | 4132 | |
cb3e7f08 | 4133 | qobject_unref(options); |
8961be33 | 4134 | options = NULL; |
dd62f1ca KW |
4135 | |
4136 | /* For snapshot=on, create a temporary qcow2 overlay. bs points to the | |
4137 | * temporary snapshot afterwards. */ | |
4138 | if (snapshot_flags) { | |
66836189 HR |
4139 | BlockDriverState *snapshot_bs; |
4140 | snapshot_bs = bdrv_append_temp_snapshot(bs, snapshot_flags, | |
4141 | snapshot_options, &local_err); | |
73176bee | 4142 | snapshot_options = NULL; |
dd62f1ca KW |
4143 | if (local_err) { |
4144 | goto close_and_fail; | |
4145 | } | |
5b363937 HR |
4146 | /* We are not going to return bs but the overlay on top of it |
4147 | * (snapshot_bs); thus, we have to drop the strong reference to bs | |
4148 | * (which we obtained by calling bdrv_new()). bs will not be deleted, | |
4149 | * though, because the overlay still has a reference to it. */ | |
f665f01f | 4150 | aio_context_acquire(ctx); |
5b363937 | 4151 | bdrv_unref(bs); |
f665f01f | 4152 | aio_context_release(ctx); |
5b363937 | 4153 | bs = snapshot_bs; |
dd62f1ca KW |
4154 | } |
4155 | ||
5b363937 | 4156 | return bs; |
b6ce07aa | 4157 | |
8bfea15d | 4158 | fail: |
f665f01f | 4159 | aio_context_acquire(ctx); |
5696c6e3 | 4160 | blk_unref(file); |
cb3e7f08 MAL |
4161 | qobject_unref(snapshot_options); |
4162 | qobject_unref(bs->explicit_options); | |
4163 | qobject_unref(bs->options); | |
4164 | qobject_unref(options); | |
de9c0cec | 4165 | bs->options = NULL; |
998cbd6a | 4166 | bs->explicit_options = NULL; |
5b363937 | 4167 | bdrv_unref(bs); |
f665f01f | 4168 | aio_context_release(ctx); |
621ff94d | 4169 | error_propagate(errp, local_err); |
5b363937 | 4170 | return NULL; |
de9c0cec | 4171 | |
b6ad491a | 4172 | close_and_fail: |
f665f01f | 4173 | aio_context_acquire(ctx); |
5b363937 | 4174 | bdrv_unref(bs); |
f665f01f | 4175 | aio_context_release(ctx); |
cb3e7f08 MAL |
4176 | qobject_unref(snapshot_options); |
4177 | qobject_unref(options); | |
621ff94d | 4178 | error_propagate(errp, local_err); |
5b363937 | 4179 | return NULL; |
b6ce07aa KW |
4180 | } |
4181 | ||
ae400dbb | 4182 | /* The caller must always hold the main AioContext lock. */ |
5b363937 HR |
4183 | BlockDriverState *bdrv_open(const char *filename, const char *reference, |
4184 | QDict *options, int flags, Error **errp) | |
f3930ed0 | 4185 | { |
f791bf7f EGE |
4186 | GLOBAL_STATE_CODE(); |
4187 | ||
5b363937 | 4188 | return bdrv_open_inherit(filename, reference, options, flags, NULL, |
272c02ea | 4189 | NULL, 0, errp); |
f3930ed0 KW |
4190 | } |
4191 | ||
faf116b4 AG |
4192 | /* Return true if the NULL-terminated @list contains @str */ |
4193 | static bool is_str_in_list(const char *str, const char *const *list) | |
4194 | { | |
4195 | if (str && list) { | |
4196 | int i; | |
4197 | for (i = 0; list[i] != NULL; i++) { | |
4198 | if (!strcmp(str, list[i])) { | |
4199 | return true; | |
4200 | } | |
4201 | } | |
4202 | } | |
4203 | return false; | |
4204 | } | |
4205 | ||
4206 | /* | |
4207 | * Check that every option set in @bs->options is also set in | |
4208 | * @new_opts. | |
4209 | * | |
4210 | * Options listed in the common_options list and in | |
4211 | * @bs->drv->mutable_opts are skipped. | |
4212 | * | |
4213 | * Return 0 on success, otherwise return -EINVAL and set @errp. | |
4214 | */ | |
4215 | static int bdrv_reset_options_allowed(BlockDriverState *bs, | |
4216 | const QDict *new_opts, Error **errp) | |
4217 | { | |
4218 | const QDictEntry *e; | |
4219 | /* These options are common to all block drivers and are handled | |
4220 | * in bdrv_reopen_prepare() so they can be left out of @new_opts */ | |
4221 | const char *const common_options[] = { | |
4222 | "node-name", "discard", "cache.direct", "cache.no-flush", | |
4223 | "read-only", "auto-read-only", "detect-zeroes", NULL | |
4224 | }; | |
4225 | ||
4226 | for (e = qdict_first(bs->options); e; e = qdict_next(bs->options, e)) { | |
4227 | if (!qdict_haskey(new_opts, e->key) && | |
4228 | !is_str_in_list(e->key, common_options) && | |
4229 | !is_str_in_list(e->key, bs->drv->mutable_opts)) { | |
4230 | error_setg(errp, "Option '%s' cannot be reset " | |
4231 | "to its default value", e->key); | |
4232 | return -EINVAL; | |
4233 | } | |
4234 | } | |
4235 | ||
4236 | return 0; | |
4237 | } | |
4238 | ||
cb828c31 AG |
4239 | /* |
4240 | * Returns true if @child can be reached recursively from @bs | |
4241 | */ | |
4242 | static bool bdrv_recurse_has_child(BlockDriverState *bs, | |
4243 | BlockDriverState *child) | |
4244 | { | |
4245 | BdrvChild *c; | |
4246 | ||
4247 | if (bs == child) { | |
4248 | return true; | |
4249 | } | |
4250 | ||
4251 | QLIST_FOREACH(c, &bs->children, next) { | |
4252 | if (bdrv_recurse_has_child(c->bs, child)) { | |
4253 | return true; | |
4254 | } | |
4255 | } | |
4256 | ||
4257 | return false; | |
4258 | } | |
4259 | ||
e971aa12 JC |
4260 | /* |
4261 | * Adds a BlockDriverState to a simple queue for an atomic, transactional | |
4262 | * reopen of multiple devices. | |
4263 | * | |
859443b0 | 4264 | * bs_queue can either be an existing BlockReopenQueue that has had QTAILQ_INIT |
e971aa12 JC |
4265 | * already performed, or alternatively may be NULL a new BlockReopenQueue will |
4266 | * be created and initialized. This newly created BlockReopenQueue should be | |
4267 | * passed back in for subsequent calls that are intended to be of the same | |
4268 | * atomic 'set'. | |
4269 | * | |
4270 | * bs is the BlockDriverState to add to the reopen queue. | |
4271 | * | |
4d2cb092 KW |
4272 | * options contains the changed options for the associated bs |
4273 | * (the BlockReopenQueue takes ownership) | |
4274 | * | |
e971aa12 JC |
4275 | * flags contains the open flags for the associated bs |
4276 | * | |
4277 | * returns a pointer to bs_queue, which is either the newly allocated | |
4278 | * bs_queue, or the existing bs_queue being used. | |
4279 | * | |
d22933ac | 4280 | * bs is drained here and undrained by bdrv_reopen_queue_free(). |
2e117866 KW |
4281 | * |
4282 | * To be called with bs->aio_context locked. | |
e971aa12 | 4283 | */ |
28518102 KW |
4284 | static BlockReopenQueue *bdrv_reopen_queue_child(BlockReopenQueue *bs_queue, |
4285 | BlockDriverState *bs, | |
4286 | QDict *options, | |
bd86fb99 | 4287 | const BdrvChildClass *klass, |
272c02ea | 4288 | BdrvChildRole role, |
3cdc69d3 | 4289 | bool parent_is_format, |
28518102 | 4290 | QDict *parent_options, |
077e8e20 AG |
4291 | int parent_flags, |
4292 | bool keep_old_opts) | |
e971aa12 JC |
4293 | { |
4294 | assert(bs != NULL); | |
4295 | ||
4296 | BlockReopenQueueEntry *bs_entry; | |
67251a31 | 4297 | BdrvChild *child; |
9aa09ddd AG |
4298 | QDict *old_options, *explicit_options, *options_copy; |
4299 | int flags; | |
4300 | QemuOpts *opts; | |
67251a31 | 4301 | |
f0c28327 | 4302 | GLOBAL_STATE_CODE(); |
1a63a907 | 4303 | |
d22933ac KW |
4304 | bdrv_drained_begin(bs); |
4305 | ||
e971aa12 JC |
4306 | if (bs_queue == NULL) { |
4307 | bs_queue = g_new0(BlockReopenQueue, 1); | |
859443b0 | 4308 | QTAILQ_INIT(bs_queue); |
e971aa12 JC |
4309 | } |
4310 | ||
4d2cb092 KW |
4311 | if (!options) { |
4312 | options = qdict_new(); | |
4313 | } | |
4314 | ||
5b7ba05f | 4315 | /* Check if this BlockDriverState is already in the queue */ |
859443b0 | 4316 | QTAILQ_FOREACH(bs_entry, bs_queue, entry) { |
5b7ba05f AG |
4317 | if (bs == bs_entry->state.bs) { |
4318 | break; | |
4319 | } | |
4320 | } | |
4321 | ||
28518102 KW |
4322 | /* |
4323 | * Precedence of options: | |
4324 | * 1. Explicitly passed in options (highest) | |
9aa09ddd AG |
4325 | * 2. Retained from explicitly set options of bs |
4326 | * 3. Inherited from parent node | |
4327 | * 4. Retained from effective options of bs | |
28518102 KW |
4328 | */ |
4329 | ||
145f598e | 4330 | /* Old explicitly set values (don't overwrite by inherited value) */ |
077e8e20 AG |
4331 | if (bs_entry || keep_old_opts) { |
4332 | old_options = qdict_clone_shallow(bs_entry ? | |
4333 | bs_entry->state.explicit_options : | |
4334 | bs->explicit_options); | |
4335 | bdrv_join_options(bs, options, old_options); | |
4336 | qobject_unref(old_options); | |
5b7ba05f | 4337 | } |
145f598e KW |
4338 | |
4339 | explicit_options = qdict_clone_shallow(options); | |
4340 | ||
28518102 KW |
4341 | /* Inherit from parent node */ |
4342 | if (parent_options) { | |
9aa09ddd | 4343 | flags = 0; |
3cdc69d3 | 4344 | klass->inherit_options(role, parent_is_format, &flags, options, |
272c02ea | 4345 | parent_flags, parent_options); |
9aa09ddd AG |
4346 | } else { |
4347 | flags = bdrv_get_flags(bs); | |
28518102 KW |
4348 | } |
4349 | ||
077e8e20 AG |
4350 | if (keep_old_opts) { |
4351 | /* Old values are used for options that aren't set yet */ | |
4352 | old_options = qdict_clone_shallow(bs->options); | |
4353 | bdrv_join_options(bs, options, old_options); | |
4354 | qobject_unref(old_options); | |
4355 | } | |
4d2cb092 | 4356 | |
9aa09ddd AG |
4357 | /* We have the final set of options so let's update the flags */ |
4358 | options_copy = qdict_clone_shallow(options); | |
4359 | opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort); | |
4360 | qemu_opts_absorb_qdict(opts, options_copy, NULL); | |
4361 | update_flags_from_options(&flags, opts); | |
4362 | qemu_opts_del(opts); | |
4363 | qobject_unref(options_copy); | |
4364 | ||
fd452021 | 4365 | /* bdrv_open_inherit() sets and clears some additional flags internally */ |
f1f25a2e | 4366 | flags &= ~BDRV_O_PROTOCOL; |
fd452021 KW |
4367 | if (flags & BDRV_O_RDWR) { |
4368 | flags |= BDRV_O_ALLOW_RDWR; | |
4369 | } | |
f1f25a2e | 4370 | |
1857c97b KW |
4371 | if (!bs_entry) { |
4372 | bs_entry = g_new0(BlockReopenQueueEntry, 1); | |
859443b0 | 4373 | QTAILQ_INSERT_TAIL(bs_queue, bs_entry, entry); |
1857c97b | 4374 | } else { |
cb3e7f08 MAL |
4375 | qobject_unref(bs_entry->state.options); |
4376 | qobject_unref(bs_entry->state.explicit_options); | |
1857c97b KW |
4377 | } |
4378 | ||
4379 | bs_entry->state.bs = bs; | |
4380 | bs_entry->state.options = options; | |
4381 | bs_entry->state.explicit_options = explicit_options; | |
4382 | bs_entry->state.flags = flags; | |
4383 | ||
8546632e AG |
4384 | /* |
4385 | * If keep_old_opts is false then it means that unspecified | |
4386 | * options must be reset to their original value. We don't allow | |
4387 | * resetting 'backing' but we need to know if the option is | |
4388 | * missing in order to decide if we have to return an error. | |
4389 | */ | |
4390 | if (!keep_old_opts) { | |
4391 | bs_entry->state.backing_missing = | |
4392 | !qdict_haskey(options, "backing") && | |
4393 | !qdict_haskey(options, "backing.driver"); | |
4394 | } | |
4395 | ||
67251a31 | 4396 | QLIST_FOREACH(child, &bs->children, next) { |
8546632e AG |
4397 | QDict *new_child_options = NULL; |
4398 | bool child_keep_old = keep_old_opts; | |
67251a31 | 4399 | |
4c9dfe5d KW |
4400 | /* reopen can only change the options of block devices that were |
4401 | * implicitly created and inherited options. For other (referenced) | |
4402 | * block devices, a syntax like "backing.foo" results in an error. */ | |
67251a31 KW |
4403 | if (child->bs->inherits_from != bs) { |
4404 | continue; | |
4405 | } | |
4406 | ||
8546632e AG |
4407 | /* Check if the options contain a child reference */ |
4408 | if (qdict_haskey(options, child->name)) { | |
4409 | const char *childref = qdict_get_try_str(options, child->name); | |
4410 | /* | |
4411 | * The current child must not be reopened if the child | |
4412 | * reference is null or points to a different node. | |
4413 | */ | |
4414 | if (g_strcmp0(childref, child->bs->node_name)) { | |
4415 | continue; | |
4416 | } | |
4417 | /* | |
4418 | * If the child reference points to the current child then | |
4419 | * reopen it with its existing set of options (note that | |
4420 | * it can still inherit new options from the parent). | |
4421 | */ | |
4422 | child_keep_old = true; | |
4423 | } else { | |
4424 | /* Extract child options ("child-name.*") */ | |
4425 | char *child_key_dot = g_strdup_printf("%s.", child->name); | |
4426 | qdict_extract_subqdict(explicit_options, NULL, child_key_dot); | |
4427 | qdict_extract_subqdict(options, &new_child_options, child_key_dot); | |
4428 | g_free(child_key_dot); | |
4429 | } | |
4c9dfe5d | 4430 | |
9aa09ddd | 4431 | bdrv_reopen_queue_child(bs_queue, child->bs, new_child_options, |
3cdc69d3 HR |
4432 | child->klass, child->role, bs->drv->is_format, |
4433 | options, flags, child_keep_old); | |
e971aa12 JC |
4434 | } |
4435 | ||
e971aa12 JC |
4436 | return bs_queue; |
4437 | } | |
4438 | ||
2e117866 | 4439 | /* To be called with bs->aio_context locked */ |
28518102 KW |
4440 | BlockReopenQueue *bdrv_reopen_queue(BlockReopenQueue *bs_queue, |
4441 | BlockDriverState *bs, | |
077e8e20 | 4442 | QDict *options, bool keep_old_opts) |
28518102 | 4443 | { |
f791bf7f EGE |
4444 | GLOBAL_STATE_CODE(); |
4445 | ||
3cdc69d3 HR |
4446 | return bdrv_reopen_queue_child(bs_queue, bs, options, NULL, 0, false, |
4447 | NULL, 0, keep_old_opts); | |
28518102 KW |
4448 | } |
4449 | ||
ab5b5228 AG |
4450 | void bdrv_reopen_queue_free(BlockReopenQueue *bs_queue) |
4451 | { | |
f791bf7f | 4452 | GLOBAL_STATE_CODE(); |
ab5b5228 AG |
4453 | if (bs_queue) { |
4454 | BlockReopenQueueEntry *bs_entry, *next; | |
4455 | QTAILQ_FOREACH_SAFE(bs_entry, bs_queue, entry, next) { | |
d22933ac KW |
4456 | AioContext *ctx = bdrv_get_aio_context(bs_entry->state.bs); |
4457 | ||
4458 | aio_context_acquire(ctx); | |
4459 | bdrv_drained_end(bs_entry->state.bs); | |
4460 | aio_context_release(ctx); | |
4461 | ||
ab5b5228 AG |
4462 | qobject_unref(bs_entry->state.explicit_options); |
4463 | qobject_unref(bs_entry->state.options); | |
4464 | g_free(bs_entry); | |
4465 | } | |
4466 | g_free(bs_queue); | |
4467 | } | |
4468 | } | |
4469 | ||
e971aa12 JC |
4470 | /* |
4471 | * Reopen multiple BlockDriverStates atomically & transactionally. | |
4472 | * | |
4473 | * The queue passed in (bs_queue) must have been built up previous | |
4474 | * via bdrv_reopen_queue(). | |
4475 | * | |
4476 | * Reopens all BDS specified in the queue, with the appropriate | |
4477 | * flags. All devices are prepared for reopen, and failure of any | |
50d6a8a3 | 4478 | * device will cause all device changes to be abandoned, and intermediate |
e971aa12 JC |
4479 | * data cleaned up. |
4480 | * | |
4481 | * If all devices prepare successfully, then the changes are committed | |
4482 | * to all devices. | |
4483 | * | |
1a63a907 KW |
4484 | * All affected nodes must be drained between bdrv_reopen_queue() and |
4485 | * bdrv_reopen_multiple(). | |
6cf42ca2 KW |
4486 | * |
4487 | * To be called from the main thread, with all other AioContexts unlocked. | |
e971aa12 | 4488 | */ |
5019aece | 4489 | int bdrv_reopen_multiple(BlockReopenQueue *bs_queue, Error **errp) |
e971aa12 JC |
4490 | { |
4491 | int ret = -1; | |
4492 | BlockReopenQueueEntry *bs_entry, *next; | |
6cf42ca2 | 4493 | AioContext *ctx; |
72373e40 | 4494 | Transaction *tran = tran_new(); |
72373e40 | 4495 | g_autoptr(GSList) refresh_list = NULL; |
e971aa12 | 4496 | |
6cf42ca2 | 4497 | assert(qemu_get_current_aio_context() == qemu_get_aio_context()); |
e971aa12 | 4498 | assert(bs_queue != NULL); |
da359909 | 4499 | GLOBAL_STATE_CODE(); |
e971aa12 | 4500 | |
a2aabf88 | 4501 | QTAILQ_FOREACH(bs_entry, bs_queue, entry) { |
6cf42ca2 KW |
4502 | ctx = bdrv_get_aio_context(bs_entry->state.bs); |
4503 | aio_context_acquire(ctx); | |
a2aabf88 | 4504 | ret = bdrv_flush(bs_entry->state.bs); |
6cf42ca2 | 4505 | aio_context_release(ctx); |
a2aabf88 VSO |
4506 | if (ret < 0) { |
4507 | error_setg_errno(errp, -ret, "Error flushing drive"); | |
e3fc91aa | 4508 | goto abort; |
a2aabf88 VSO |
4509 | } |
4510 | } | |
4511 | ||
859443b0 | 4512 | QTAILQ_FOREACH(bs_entry, bs_queue, entry) { |
1a63a907 | 4513 | assert(bs_entry->state.bs->quiesce_counter > 0); |
6cf42ca2 KW |
4514 | ctx = bdrv_get_aio_context(bs_entry->state.bs); |
4515 | aio_context_acquire(ctx); | |
72373e40 | 4516 | ret = bdrv_reopen_prepare(&bs_entry->state, bs_queue, tran, errp); |
6cf42ca2 | 4517 | aio_context_release(ctx); |
72373e40 VSO |
4518 | if (ret < 0) { |
4519 | goto abort; | |
e971aa12 JC |
4520 | } |
4521 | bs_entry->prepared = true; | |
4522 | } | |
4523 | ||
859443b0 | 4524 | QTAILQ_FOREACH(bs_entry, bs_queue, entry) { |
69b736e7 | 4525 | BDRVReopenState *state = &bs_entry->state; |
72373e40 | 4526 | |
fb0ff4d1 | 4527 | refresh_list = g_slist_prepend(refresh_list, state->bs); |
72373e40 | 4528 | if (state->old_backing_bs) { |
fb0ff4d1 | 4529 | refresh_list = g_slist_prepend(refresh_list, state->old_backing_bs); |
cb828c31 | 4530 | } |
ecd30d2d | 4531 | if (state->old_file_bs) { |
fb0ff4d1 | 4532 | refresh_list = g_slist_prepend(refresh_list, state->old_file_bs); |
ecd30d2d | 4533 | } |
72373e40 VSO |
4534 | } |
4535 | ||
4536 | /* | |
4537 | * Note that file-posix driver rely on permission update done during reopen | |
4538 | * (even if no permission changed), because it wants "new" permissions for | |
4539 | * reconfiguring the fd and that's why it does it in raw_check_perm(), not | |
4540 | * in raw_reopen_prepare() which is called with "old" permissions. | |
4541 | */ | |
4542 | ret = bdrv_list_refresh_perms(refresh_list, bs_queue, tran, errp); | |
4543 | if (ret < 0) { | |
4544 | goto abort; | |
69b736e7 KW |
4545 | } |
4546 | ||
fcd6a4f4 VSO |
4547 | /* |
4548 | * If we reach this point, we have success and just need to apply the | |
4549 | * changes. | |
4550 | * | |
4551 | * Reverse order is used to comfort qcow2 driver: on commit it need to write | |
4552 | * IN_USE flag to the image, to mark bitmaps in the image as invalid. But | |
4553 | * children are usually goes after parents in reopen-queue, so go from last | |
4554 | * to first element. | |
e971aa12 | 4555 | */ |
fcd6a4f4 | 4556 | QTAILQ_FOREACH_REVERSE(bs_entry, bs_queue, entry) { |
6cf42ca2 KW |
4557 | ctx = bdrv_get_aio_context(bs_entry->state.bs); |
4558 | aio_context_acquire(ctx); | |
e971aa12 | 4559 | bdrv_reopen_commit(&bs_entry->state); |
6cf42ca2 | 4560 | aio_context_release(ctx); |
e971aa12 JC |
4561 | } |
4562 | ||
72373e40 | 4563 | tran_commit(tran); |
69b736e7 | 4564 | |
72373e40 VSO |
4565 | QTAILQ_FOREACH_REVERSE(bs_entry, bs_queue, entry) { |
4566 | BlockDriverState *bs = bs_entry->state.bs; | |
74ad9a3b | 4567 | |
72373e40 | 4568 | if (bs->drv->bdrv_reopen_commit_post) { |
6cf42ca2 KW |
4569 | ctx = bdrv_get_aio_context(bs); |
4570 | aio_context_acquire(ctx); | |
72373e40 | 4571 | bs->drv->bdrv_reopen_commit_post(&bs_entry->state); |
6cf42ca2 | 4572 | aio_context_release(ctx); |
69b736e7 KW |
4573 | } |
4574 | } | |
17e1e2be | 4575 | |
72373e40 VSO |
4576 | ret = 0; |
4577 | goto cleanup; | |
17e1e2be | 4578 | |
72373e40 VSO |
4579 | abort: |
4580 | tran_abort(tran); | |
4581 | QTAILQ_FOREACH_SAFE(bs_entry, bs_queue, entry, next) { | |
4582 | if (bs_entry->prepared) { | |
6cf42ca2 KW |
4583 | ctx = bdrv_get_aio_context(bs_entry->state.bs); |
4584 | aio_context_acquire(ctx); | |
72373e40 | 4585 | bdrv_reopen_abort(&bs_entry->state); |
6cf42ca2 | 4586 | aio_context_release(ctx); |
17e1e2be PK |
4587 | } |
4588 | } | |
72373e40 | 4589 | |
e971aa12 | 4590 | cleanup: |
ab5b5228 | 4591 | bdrv_reopen_queue_free(bs_queue); |
40840e41 | 4592 | |
e971aa12 JC |
4593 | return ret; |
4594 | } | |
4595 | ||
6cf42ca2 KW |
4596 | int bdrv_reopen(BlockDriverState *bs, QDict *opts, bool keep_old_opts, |
4597 | Error **errp) | |
6e1000a8 | 4598 | { |
6cf42ca2 | 4599 | AioContext *ctx = bdrv_get_aio_context(bs); |
6e1000a8 | 4600 | BlockReopenQueue *queue; |
6cf42ca2 | 4601 | int ret; |
6e1000a8 | 4602 | |
f791bf7f EGE |
4603 | GLOBAL_STATE_CODE(); |
4604 | ||
2e117866 KW |
4605 | queue = bdrv_reopen_queue(NULL, bs, opts, keep_old_opts); |
4606 | ||
6cf42ca2 KW |
4607 | if (ctx != qemu_get_aio_context()) { |
4608 | aio_context_release(ctx); | |
4609 | } | |
5019aece | 4610 | ret = bdrv_reopen_multiple(queue, errp); |
6cf42ca2 KW |
4611 | |
4612 | if (ctx != qemu_get_aio_context()) { | |
4613 | aio_context_acquire(ctx); | |
4614 | } | |
6e1000a8 AG |
4615 | |
4616 | return ret; | |
4617 | } | |
4618 | ||
6cf42ca2 KW |
4619 | int bdrv_reopen_set_read_only(BlockDriverState *bs, bool read_only, |
4620 | Error **errp) | |
4621 | { | |
4622 | QDict *opts = qdict_new(); | |
4623 | ||
f791bf7f EGE |
4624 | GLOBAL_STATE_CODE(); |
4625 | ||
6cf42ca2 KW |
4626 | qdict_put_bool(opts, BDRV_OPT_READ_ONLY, read_only); |
4627 | ||
4628 | return bdrv_reopen(bs, opts, true, errp); | |
4629 | } | |
4630 | ||
cb828c31 AG |
4631 | /* |
4632 | * Take a BDRVReopenState and check if the value of 'backing' in the | |
4633 | * reopen_state->options QDict is valid or not. | |
4634 | * | |
4635 | * If 'backing' is missing from the QDict then return 0. | |
4636 | * | |
4637 | * If 'backing' contains the node name of the backing file of | |
4638 | * reopen_state->bs then return 0. | |
4639 | * | |
4640 | * If 'backing' contains a different node name (or is null) then check | |
4641 | * whether the current backing file can be replaced with the new one. | |
4642 | * If that's the case then reopen_state->replace_backing_bs is set to | |
4643 | * true and reopen_state->new_backing_bs contains a pointer to the new | |
4644 | * backing BlockDriverState (or NULL). | |
4645 | * | |
4646 | * Return 0 on success, otherwise return < 0 and set @errp. | |
4b408668 KW |
4647 | * |
4648 | * The caller must hold the AioContext lock of @reopen_state->bs. | |
4649 | * @reopen_state->bs can move to a different AioContext in this function. | |
4650 | * Callers must make sure that their AioContext locking is still correct after | |
4651 | * this. | |
cb828c31 | 4652 | */ |
ecd30d2d AG |
4653 | static int bdrv_reopen_parse_file_or_backing(BDRVReopenState *reopen_state, |
4654 | bool is_backing, Transaction *tran, | |
4655 | Error **errp) | |
cb828c31 AG |
4656 | { |
4657 | BlockDriverState *bs = reopen_state->bs; | |
ecd30d2d AG |
4658 | BlockDriverState *new_child_bs; |
4659 | BlockDriverState *old_child_bs = is_backing ? child_bs(bs->backing) : | |
4660 | child_bs(bs->file); | |
4661 | const char *child_name = is_backing ? "backing" : "file"; | |
cb828c31 AG |
4662 | QObject *value; |
4663 | const char *str; | |
4b408668 KW |
4664 | AioContext *ctx, *old_ctx; |
4665 | int ret; | |
cb828c31 | 4666 | |
bdb73476 EGE |
4667 | GLOBAL_STATE_CODE(); |
4668 | ||
ecd30d2d | 4669 | value = qdict_get(reopen_state->options, child_name); |
cb828c31 AG |
4670 | if (value == NULL) { |
4671 | return 0; | |
4672 | } | |
4673 | ||
4674 | switch (qobject_type(value)) { | |
4675 | case QTYPE_QNULL: | |
ecd30d2d AG |
4676 | assert(is_backing); /* The 'file' option does not allow a null value */ |
4677 | new_child_bs = NULL; | |
cb828c31 AG |
4678 | break; |
4679 | case QTYPE_QSTRING: | |
410f44f5 | 4680 | str = qstring_get_str(qobject_to(QString, value)); |
ecd30d2d AG |
4681 | new_child_bs = bdrv_lookup_bs(NULL, str, errp); |
4682 | if (new_child_bs == NULL) { | |
cb828c31 | 4683 | return -EINVAL; |
ecd30d2d AG |
4684 | } else if (bdrv_recurse_has_child(new_child_bs, bs)) { |
4685 | error_setg(errp, "Making '%s' a %s child of '%s' would create a " | |
4686 | "cycle", str, child_name, bs->node_name); | |
cb828c31 AG |
4687 | return -EINVAL; |
4688 | } | |
4689 | break; | |
4690 | default: | |
ecd30d2d AG |
4691 | /* |
4692 | * The options QDict has been flattened, so 'backing' and 'file' | |
4693 | * do not allow any other data type here. | |
4694 | */ | |
cb828c31 AG |
4695 | g_assert_not_reached(); |
4696 | } | |
4697 | ||
ecd30d2d AG |
4698 | if (old_child_bs == new_child_bs) { |
4699 | return 0; | |
4700 | } | |
4701 | ||
4702 | if (old_child_bs) { | |
4703 | if (bdrv_skip_implicit_filters(old_child_bs) == new_child_bs) { | |
cbfdb98c VSO |
4704 | return 0; |
4705 | } | |
4706 | ||
ecd30d2d AG |
4707 | if (old_child_bs->implicit) { |
4708 | error_setg(errp, "Cannot replace implicit %s child of %s", | |
4709 | child_name, bs->node_name); | |
cbfdb98c VSO |
4710 | return -EPERM; |
4711 | } | |
4712 | } | |
4713 | ||
ecd30d2d | 4714 | if (bs->drv->is_filter && !old_child_bs) { |
25f78d9e VSO |
4715 | /* |
4716 | * Filters always have a file or a backing child, so we are trying to | |
4717 | * change wrong child | |
4718 | */ | |
4719 | error_setg(errp, "'%s' is a %s filter node that does not support a " | |
ecd30d2d | 4720 | "%s child", bs->node_name, bs->drv->format_name, child_name); |
1d42f48c HR |
4721 | return -EINVAL; |
4722 | } | |
4723 | ||
ecd30d2d AG |
4724 | if (is_backing) { |
4725 | reopen_state->old_backing_bs = old_child_bs; | |
4726 | } else { | |
4727 | reopen_state->old_file_bs = old_child_bs; | |
4728 | } | |
4729 | ||
4b408668 KW |
4730 | old_ctx = bdrv_get_aio_context(bs); |
4731 | ctx = bdrv_get_aio_context(new_child_bs); | |
4732 | if (old_ctx != ctx) { | |
4733 | aio_context_release(old_ctx); | |
4734 | aio_context_acquire(ctx); | |
4735 | } | |
4736 | ||
4737 | ret = bdrv_set_file_or_backing_noperm(bs, new_child_bs, is_backing, | |
4738 | tran, errp); | |
4739 | ||
4740 | if (old_ctx != ctx) { | |
4741 | aio_context_release(ctx); | |
4742 | aio_context_acquire(old_ctx); | |
4743 | } | |
4744 | ||
4745 | return ret; | |
cb828c31 AG |
4746 | } |
4747 | ||
e971aa12 JC |
4748 | /* |
4749 | * Prepares a BlockDriverState for reopen. All changes are staged in the | |
4750 | * 'opaque' field of the BDRVReopenState, which is used and allocated by | |
4751 | * the block driver layer .bdrv_reopen_prepare() | |
4752 | * | |
4753 | * bs is the BlockDriverState to reopen | |
4754 | * flags are the new open flags | |
4755 | * queue is the reopen queue | |
4756 | * | |
4757 | * Returns 0 on success, non-zero on error. On error errp will be set | |
4758 | * as well. | |
4759 | * | |
4760 | * On failure, bdrv_reopen_abort() will be called to clean up any data. | |
4761 | * It is the responsibility of the caller to then call the abort() or | |
4762 | * commit() for any other BDS that have been left in a prepare() state | |
4763 | * | |
4b408668 | 4764 | * The caller must hold the AioContext lock of @reopen_state->bs. |
e971aa12 | 4765 | */ |
53e96d1e | 4766 | static int bdrv_reopen_prepare(BDRVReopenState *reopen_state, |
72373e40 | 4767 | BlockReopenQueue *queue, |
ecd30d2d | 4768 | Transaction *change_child_tran, Error **errp) |
e971aa12 JC |
4769 | { |
4770 | int ret = -1; | |
e6d79c41 | 4771 | int old_flags; |
e971aa12 JC |
4772 | Error *local_err = NULL; |
4773 | BlockDriver *drv; | |
ccf9dc07 | 4774 | QemuOpts *opts; |
4c8350fe | 4775 | QDict *orig_reopen_opts; |
593b3071 | 4776 | char *discard = NULL; |
3d8ce171 | 4777 | bool read_only; |
9ad08c44 | 4778 | bool drv_prepared = false; |
e971aa12 JC |
4779 | |
4780 | assert(reopen_state != NULL); | |
4781 | assert(reopen_state->bs->drv != NULL); | |
da359909 | 4782 | GLOBAL_STATE_CODE(); |
e971aa12 JC |
4783 | drv = reopen_state->bs->drv; |
4784 | ||
4c8350fe AG |
4785 | /* This function and each driver's bdrv_reopen_prepare() remove |
4786 | * entries from reopen_state->options as they are processed, so | |
4787 | * we need to make a copy of the original QDict. */ | |
4788 | orig_reopen_opts = qdict_clone_shallow(reopen_state->options); | |
4789 | ||
ccf9dc07 KW |
4790 | /* Process generic block layer options */ |
4791 | opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort); | |
af175e85 | 4792 | if (!qemu_opts_absorb_qdict(opts, reopen_state->options, errp)) { |
ccf9dc07 KW |
4793 | ret = -EINVAL; |
4794 | goto error; | |
4795 | } | |
4796 | ||
e6d79c41 AG |
4797 | /* This was already called in bdrv_reopen_queue_child() so the flags |
4798 | * are up-to-date. This time we simply want to remove the options from | |
4799 | * QemuOpts in order to indicate that they have been processed. */ | |
4800 | old_flags = reopen_state->flags; | |
91a097e7 | 4801 | update_flags_from_options(&reopen_state->flags, opts); |
e6d79c41 | 4802 | assert(old_flags == reopen_state->flags); |
91a097e7 | 4803 | |
415bbca8 | 4804 | discard = qemu_opt_get_del(opts, BDRV_OPT_DISCARD); |
593b3071 AG |
4805 | if (discard != NULL) { |
4806 | if (bdrv_parse_discard_flags(discard, &reopen_state->flags) != 0) { | |
4807 | error_setg(errp, "Invalid discard option"); | |
4808 | ret = -EINVAL; | |
4809 | goto error; | |
4810 | } | |
4811 | } | |
4812 | ||
543770bd AG |
4813 | reopen_state->detect_zeroes = |
4814 | bdrv_parse_detect_zeroes(opts, reopen_state->flags, &local_err); | |
4815 | if (local_err) { | |
4816 | error_propagate(errp, local_err); | |
4817 | ret = -EINVAL; | |
4818 | goto error; | |
4819 | } | |
4820 | ||
57f9db9a AG |
4821 | /* All other options (including node-name and driver) must be unchanged. |
4822 | * Put them back into the QDict, so that they are checked at the end | |
4823 | * of this function. */ | |
4824 | qemu_opts_to_qdict(opts, reopen_state->options); | |
ccf9dc07 | 4825 | |
3d8ce171 JC |
4826 | /* If we are to stay read-only, do not allow permission change |
4827 | * to r/w. Attempting to set to r/w may fail if either BDRV_O_ALLOW_RDWR is | |
4828 | * not set, or if the BDS still has copy_on_read enabled */ | |
4829 | read_only = !(reopen_state->flags & BDRV_O_RDWR); | |
54a32bfe | 4830 | ret = bdrv_can_set_read_only(reopen_state->bs, read_only, true, &local_err); |
3d8ce171 JC |
4831 | if (local_err) { |
4832 | error_propagate(errp, local_err); | |
e971aa12 JC |
4833 | goto error; |
4834 | } | |
4835 | ||
e971aa12 | 4836 | if (drv->bdrv_reopen_prepare) { |
faf116b4 AG |
4837 | /* |
4838 | * If a driver-specific option is missing, it means that we | |
4839 | * should reset it to its default value. | |
4840 | * But not all options allow that, so we need to check it first. | |
4841 | */ | |
4842 | ret = bdrv_reset_options_allowed(reopen_state->bs, | |
4843 | reopen_state->options, errp); | |
4844 | if (ret) { | |
4845 | goto error; | |
4846 | } | |
4847 | ||
e971aa12 JC |
4848 | ret = drv->bdrv_reopen_prepare(reopen_state, queue, &local_err); |
4849 | if (ret) { | |
4850 | if (local_err != NULL) { | |
4851 | error_propagate(errp, local_err); | |
4852 | } else { | |
f30c66ba | 4853 | bdrv_refresh_filename(reopen_state->bs); |
d8b6895f LC |
4854 | error_setg(errp, "failed while preparing to reopen image '%s'", |
4855 | reopen_state->bs->filename); | |
e971aa12 JC |
4856 | } |
4857 | goto error; | |
4858 | } | |
4859 | } else { | |
4860 | /* It is currently mandatory to have a bdrv_reopen_prepare() | |
4861 | * handler for each supported drv. */ | |
81e5f78a AG |
4862 | error_setg(errp, "Block format '%s' used by node '%s' " |
4863 | "does not support reopening files", drv->format_name, | |
4864 | bdrv_get_device_or_node_name(reopen_state->bs)); | |
e971aa12 JC |
4865 | ret = -1; |
4866 | goto error; | |
4867 | } | |
4868 | ||
9ad08c44 HR |
4869 | drv_prepared = true; |
4870 | ||
bacd9b87 AG |
4871 | /* |
4872 | * We must provide the 'backing' option if the BDS has a backing | |
4873 | * file or if the image file has a backing file name as part of | |
4874 | * its metadata. Otherwise the 'backing' option can be omitted. | |
4875 | */ | |
4876 | if (drv->supports_backing && reopen_state->backing_missing && | |
1d42f48c | 4877 | (reopen_state->bs->backing || reopen_state->bs->backing_file[0])) { |
8546632e AG |
4878 | error_setg(errp, "backing is missing for '%s'", |
4879 | reopen_state->bs->node_name); | |
4880 | ret = -EINVAL; | |
4881 | goto error; | |
4882 | } | |
4883 | ||
cb828c31 AG |
4884 | /* |
4885 | * Allow changing the 'backing' option. The new value can be | |
4886 | * either a reference to an existing node (using its node name) | |
4887 | * or NULL to simply detach the current backing file. | |
4888 | */ | |
ecd30d2d AG |
4889 | ret = bdrv_reopen_parse_file_or_backing(reopen_state, true, |
4890 | change_child_tran, errp); | |
cb828c31 AG |
4891 | if (ret < 0) { |
4892 | goto error; | |
4893 | } | |
4894 | qdict_del(reopen_state->options, "backing"); | |
4895 | ||
ecd30d2d AG |
4896 | /* Allow changing the 'file' option. In this case NULL is not allowed */ |
4897 | ret = bdrv_reopen_parse_file_or_backing(reopen_state, false, | |
4898 | change_child_tran, errp); | |
4899 | if (ret < 0) { | |
4900 | goto error; | |
4901 | } | |
4902 | qdict_del(reopen_state->options, "file"); | |
4903 | ||
4d2cb092 KW |
4904 | /* Options that are not handled are only okay if they are unchanged |
4905 | * compared to the old state. It is expected that some options are only | |
4906 | * used for the initial open, but not reopen (e.g. filename) */ | |
4907 | if (qdict_size(reopen_state->options)) { | |
4908 | const QDictEntry *entry = qdict_first(reopen_state->options); | |
4909 | ||
4910 | do { | |
54fd1b0d HR |
4911 | QObject *new = entry->value; |
4912 | QObject *old = qdict_get(reopen_state->bs->options, entry->key); | |
4913 | ||
db905283 AG |
4914 | /* Allow child references (child_name=node_name) as long as they |
4915 | * point to the current child (i.e. everything stays the same). */ | |
4916 | if (qobject_type(new) == QTYPE_QSTRING) { | |
4917 | BdrvChild *child; | |
4918 | QLIST_FOREACH(child, &reopen_state->bs->children, next) { | |
4919 | if (!strcmp(child->name, entry->key)) { | |
4920 | break; | |
4921 | } | |
4922 | } | |
4923 | ||
4924 | if (child) { | |
410f44f5 MA |
4925 | if (!strcmp(child->bs->node_name, |
4926 | qstring_get_str(qobject_to(QString, new)))) { | |
db905283 AG |
4927 | continue; /* Found child with this name, skip option */ |
4928 | } | |
4929 | } | |
4930 | } | |
4931 | ||
129c7d1c | 4932 | /* |
54fd1b0d HR |
4933 | * TODO: When using -drive to specify blockdev options, all values |
4934 | * will be strings; however, when using -blockdev, blockdev-add or | |
4935 | * filenames using the json:{} pseudo-protocol, they will be | |
4936 | * correctly typed. | |
4937 | * In contrast, reopening options are (currently) always strings | |
4938 | * (because you can only specify them through qemu-io; all other | |
4939 | * callers do not specify any options). | |
4940 | * Therefore, when using anything other than -drive to create a BDS, | |
4941 | * this cannot detect non-string options as unchanged, because | |
4942 | * qobject_is_equal() always returns false for objects of different | |
4943 | * type. In the future, this should be remedied by correctly typing | |
4944 | * all options. For now, this is not too big of an issue because | |
4945 | * the user can simply omit options which cannot be changed anyway, | |
4946 | * so they will stay unchanged. | |
129c7d1c | 4947 | */ |
54fd1b0d | 4948 | if (!qobject_is_equal(new, old)) { |
4d2cb092 KW |
4949 | error_setg(errp, "Cannot change the option '%s'", entry->key); |
4950 | ret = -EINVAL; | |
4951 | goto error; | |
4952 | } | |
4953 | } while ((entry = qdict_next(reopen_state->options, entry))); | |
4954 | } | |
4955 | ||
e971aa12 JC |
4956 | ret = 0; |
4957 | ||
4c8350fe AG |
4958 | /* Restore the original reopen_state->options QDict */ |
4959 | qobject_unref(reopen_state->options); | |
4960 | reopen_state->options = qobject_ref(orig_reopen_opts); | |
4961 | ||
e971aa12 | 4962 | error: |
9ad08c44 HR |
4963 | if (ret < 0 && drv_prepared) { |
4964 | /* drv->bdrv_reopen_prepare() has succeeded, so we need to | |
4965 | * call drv->bdrv_reopen_abort() before signaling an error | |
4966 | * (bdrv_reopen_multiple() will not call bdrv_reopen_abort() | |
4967 | * when the respective bdrv_reopen_prepare() has failed) */ | |
4968 | if (drv->bdrv_reopen_abort) { | |
4969 | drv->bdrv_reopen_abort(reopen_state); | |
4970 | } | |
4971 | } | |
ccf9dc07 | 4972 | qemu_opts_del(opts); |
4c8350fe | 4973 | qobject_unref(orig_reopen_opts); |
593b3071 | 4974 | g_free(discard); |
e971aa12 JC |
4975 | return ret; |
4976 | } | |
4977 | ||
4978 | /* | |
4979 | * Takes the staged changes for the reopen from bdrv_reopen_prepare(), and | |
4980 | * makes them final by swapping the staging BlockDriverState contents into | |
4981 | * the active BlockDriverState contents. | |
4982 | */ | |
53e96d1e | 4983 | static void bdrv_reopen_commit(BDRVReopenState *reopen_state) |
e971aa12 JC |
4984 | { |
4985 | BlockDriver *drv; | |
50bf65ba | 4986 | BlockDriverState *bs; |
50196d7a | 4987 | BdrvChild *child; |
e971aa12 JC |
4988 | |
4989 | assert(reopen_state != NULL); | |
50bf65ba VSO |
4990 | bs = reopen_state->bs; |
4991 | drv = bs->drv; | |
e971aa12 | 4992 | assert(drv != NULL); |
da359909 | 4993 | GLOBAL_STATE_CODE(); |
e971aa12 JC |
4994 | |
4995 | /* If there are any driver level actions to take */ | |
4996 | if (drv->bdrv_reopen_commit) { | |
4997 | drv->bdrv_reopen_commit(reopen_state); | |
4998 | } | |
4999 | ||
5000 | /* set BDS specific flags now */ | |
cb3e7f08 | 5001 | qobject_unref(bs->explicit_options); |
4c8350fe | 5002 | qobject_unref(bs->options); |
ab5b5228 AG |
5003 | qobject_ref(reopen_state->explicit_options); |
5004 | qobject_ref(reopen_state->options); | |
145f598e | 5005 | |
50bf65ba | 5006 | bs->explicit_options = reopen_state->explicit_options; |
4c8350fe | 5007 | bs->options = reopen_state->options; |
50bf65ba | 5008 | bs->open_flags = reopen_state->flags; |
543770bd | 5009 | bs->detect_zeroes = reopen_state->detect_zeroes; |
355ef4ac | 5010 | |
50196d7a AG |
5011 | /* Remove child references from bs->options and bs->explicit_options. |
5012 | * Child options were already removed in bdrv_reopen_queue_child() */ | |
5013 | QLIST_FOREACH(child, &bs->children, next) { | |
5014 | qdict_del(bs->explicit_options, child->name); | |
5015 | qdict_del(bs->options, child->name); | |
5016 | } | |
3d0e8743 VSO |
5017 | /* backing is probably removed, so it's not handled by previous loop */ |
5018 | qdict_del(bs->explicit_options, "backing"); | |
5019 | qdict_del(bs->options, "backing"); | |
5020 | ||
e19b157f | 5021 | bdrv_graph_rdlock_main_loop(); |
1e4c797c | 5022 | bdrv_refresh_limits(bs, NULL, NULL); |
e19b157f | 5023 | bdrv_graph_rdunlock_main_loop(); |
439cc330 | 5024 | bdrv_refresh_total_sectors(bs, bs->total_sectors); |
e971aa12 JC |
5025 | } |
5026 | ||
5027 | /* | |
5028 | * Abort the reopen, and delete and free the staged changes in | |
5029 | * reopen_state | |
5030 | */ | |
53e96d1e | 5031 | static void bdrv_reopen_abort(BDRVReopenState *reopen_state) |
e971aa12 JC |
5032 | { |
5033 | BlockDriver *drv; | |
5034 | ||
5035 | assert(reopen_state != NULL); | |
5036 | drv = reopen_state->bs->drv; | |
5037 | assert(drv != NULL); | |
da359909 | 5038 | GLOBAL_STATE_CODE(); |
e971aa12 JC |
5039 | |
5040 | if (drv->bdrv_reopen_abort) { | |
5041 | drv->bdrv_reopen_abort(reopen_state); | |
5042 | } | |
5043 | } | |
5044 | ||
5045 | ||
64dff520 | 5046 | static void bdrv_close(BlockDriverState *bs) |
fc01f7e7 | 5047 | { |
33384421 | 5048 | BdrvAioNotifier *ban, *ban_next; |
50a3efb0 | 5049 | BdrvChild *child, *next; |
33384421 | 5050 | |
f791bf7f | 5051 | GLOBAL_STATE_CODE(); |
30f55fb8 | 5052 | assert(!bs->refcnt); |
99b7e775 | 5053 | |
fc27291d | 5054 | bdrv_drained_begin(bs); /* complete I/O */ |
58fda173 | 5055 | bdrv_flush(bs); |
53ec73e2 | 5056 | bdrv_drain(bs); /* in case flush left pending I/O */ |
fc27291d | 5057 | |
3cbc002c | 5058 | if (bs->drv) { |
3c005293 | 5059 | if (bs->drv->bdrv_close) { |
7b99a266 | 5060 | /* Must unfreeze all children, so bdrv_unref_child() works */ |
3c005293 VSO |
5061 | bs->drv->bdrv_close(bs); |
5062 | } | |
9a4f4c31 | 5063 | bs->drv = NULL; |
50a3efb0 | 5064 | } |
9a7dedbc | 5065 | |
50a3efb0 | 5066 | QLIST_FOREACH_SAFE(child, &bs->children, next, next) { |
dd4118c7 | 5067 | bdrv_unref_child(bs, child); |
b338082b | 5068 | } |
98f90dba | 5069 | |
5bb04747 VSO |
5070 | assert(!bs->backing); |
5071 | assert(!bs->file); | |
50a3efb0 AG |
5072 | g_free(bs->opaque); |
5073 | bs->opaque = NULL; | |
d73415a3 | 5074 | qatomic_set(&bs->copy_on_read, 0); |
50a3efb0 AG |
5075 | bs->backing_file[0] = '\0'; |
5076 | bs->backing_format[0] = '\0'; | |
5077 | bs->total_sectors = 0; | |
5078 | bs->encrypted = false; | |
5079 | bs->sg = false; | |
cb3e7f08 MAL |
5080 | qobject_unref(bs->options); |
5081 | qobject_unref(bs->explicit_options); | |
50a3efb0 AG |
5082 | bs->options = NULL; |
5083 | bs->explicit_options = NULL; | |
cb3e7f08 | 5084 | qobject_unref(bs->full_open_options); |
50a3efb0 | 5085 | bs->full_open_options = NULL; |
0bc329fb HR |
5086 | g_free(bs->block_status_cache); |
5087 | bs->block_status_cache = NULL; | |
50a3efb0 | 5088 | |
cca43ae1 VSO |
5089 | bdrv_release_named_dirty_bitmaps(bs); |
5090 | assert(QLIST_EMPTY(&bs->dirty_bitmaps)); | |
5091 | ||
33384421 HR |
5092 | QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_next) { |
5093 | g_free(ban); | |
5094 | } | |
5095 | QLIST_INIT(&bs->aio_notifiers); | |
fc27291d | 5096 | bdrv_drained_end(bs); |
1a6d3bd2 GK |
5097 | |
5098 | /* | |
5099 | * If we're still inside some bdrv_drain_all_begin()/end() sections, end | |
5100 | * them now since this BDS won't exist anymore when bdrv_drain_all_end() | |
5101 | * gets called. | |
5102 | */ | |
5103 | if (bs->quiesce_counter) { | |
5104 | bdrv_drain_all_end_quiesce(bs); | |
5105 | } | |
b338082b FB |
5106 | } |
5107 | ||
2bc93fed MK |
5108 | void bdrv_close_all(void) |
5109 | { | |
f791bf7f | 5110 | GLOBAL_STATE_CODE(); |
880eeec6 | 5111 | assert(job_next(NULL) == NULL); |
ca9bd24c HR |
5112 | |
5113 | /* Drop references from requests still in flight, such as canceled block | |
5114 | * jobs whose AIO context has not been polled yet */ | |
5115 | bdrv_drain_all(); | |
2bc93fed | 5116 | |
ca9bd24c HR |
5117 | blk_remove_all_bs(); |
5118 | blockdev_close_all_bdrv_states(); | |
ed78cda3 | 5119 | |
a1a2af07 | 5120 | assert(QTAILQ_EMPTY(&all_bdrv_states)); |
2bc93fed MK |
5121 | } |
5122 | ||
d0ac0380 KW |
5123 | static bool should_update_child(BdrvChild *c, BlockDriverState *to) |
5124 | { | |
2f30b7c3 VSO |
5125 | GQueue *queue; |
5126 | GHashTable *found; | |
5127 | bool ret; | |
d0ac0380 | 5128 | |
bd86fb99 | 5129 | if (c->klass->stay_at_node) { |
d0ac0380 KW |
5130 | return false; |
5131 | } | |
5132 | ||
ec9f10fe HR |
5133 | /* If the child @c belongs to the BDS @to, replacing the current |
5134 | * c->bs by @to would mean to create a loop. | |
5135 | * | |
5136 | * Such a case occurs when appending a BDS to a backing chain. | |
5137 | * For instance, imagine the following chain: | |
5138 | * | |
5139 | * guest device -> node A -> further backing chain... | |
5140 | * | |
5141 | * Now we create a new BDS B which we want to put on top of this | |
5142 | * chain, so we first attach A as its backing node: | |
5143 | * | |
5144 | * node B | |
5145 | * | | |
5146 | * v | |
5147 | * guest device -> node A -> further backing chain... | |
5148 | * | |
5149 | * Finally we want to replace A by B. When doing that, we want to | |
5150 | * replace all pointers to A by pointers to B -- except for the | |
5151 | * pointer from B because (1) that would create a loop, and (2) | |
5152 | * that pointer should simply stay intact: | |
5153 | * | |
5154 | * guest device -> node B | |
5155 | * | | |
5156 | * v | |
5157 | * node A -> further backing chain... | |
5158 | * | |
5159 | * In general, when replacing a node A (c->bs) by a node B (@to), | |
5160 | * if A is a child of B, that means we cannot replace A by B there | |
5161 | * because that would create a loop. Silently detaching A from B | |
5162 | * is also not really an option. So overall just leaving A in | |
2f30b7c3 VSO |
5163 | * place there is the most sensible choice. |
5164 | * | |
5165 | * We would also create a loop in any cases where @c is only | |
5166 | * indirectly referenced by @to. Prevent this by returning false | |
5167 | * if @c is found (by breadth-first search) anywhere in the whole | |
5168 | * subtree of @to. | |
5169 | */ | |
5170 | ||
5171 | ret = true; | |
5172 | found = g_hash_table_new(NULL, NULL); | |
5173 | g_hash_table_add(found, to); | |
5174 | queue = g_queue_new(); | |
5175 | g_queue_push_tail(queue, to); | |
5176 | ||
5177 | while (!g_queue_is_empty(queue)) { | |
5178 | BlockDriverState *v = g_queue_pop_head(queue); | |
5179 | BdrvChild *c2; | |
5180 | ||
5181 | QLIST_FOREACH(c2, &v->children, next) { | |
5182 | if (c2 == c) { | |
5183 | ret = false; | |
5184 | break; | |
5185 | } | |
5186 | ||
5187 | if (g_hash_table_contains(found, c2->bs)) { | |
5188 | continue; | |
5189 | } | |
5190 | ||
5191 | g_queue_push_tail(queue, c2->bs); | |
5192 | g_hash_table_add(found, c2->bs); | |
d0ac0380 KW |
5193 | } |
5194 | } | |
5195 | ||
2f30b7c3 VSO |
5196 | g_queue_free(queue); |
5197 | g_hash_table_destroy(found); | |
5198 | ||
5199 | return ret; | |
d0ac0380 KW |
5200 | } |
5201 | ||
57f08941 | 5202 | static void bdrv_remove_child_commit(void *opaque) |
46541ee5 | 5203 | { |
bdb73476 | 5204 | GLOBAL_STATE_CODE(); |
5bb04747 | 5205 | bdrv_child_free(opaque); |
82b54cf5 HR |
5206 | } |
5207 | ||
57f08941 VSO |
5208 | static TransactionActionDrv bdrv_remove_child_drv = { |
5209 | .commit = bdrv_remove_child_commit, | |
46541ee5 VSO |
5210 | }; |
5211 | ||
57f08941 VSO |
5212 | /* Function doesn't update permissions, caller is responsible for this. */ |
5213 | static void bdrv_remove_child(BdrvChild *child, Transaction *tran) | |
46541ee5 | 5214 | { |
46541ee5 VSO |
5215 | if (!child) { |
5216 | return; | |
5217 | } | |
5218 | ||
5219 | if (child->bs) { | |
23987471 KW |
5220 | BlockDriverState *bs = child->bs; |
5221 | bdrv_drained_begin(bs); | |
a2c37a30 | 5222 | bdrv_replace_child_tran(child, NULL, tran); |
23987471 | 5223 | bdrv_drained_end(bs); |
46541ee5 VSO |
5224 | } |
5225 | ||
57f08941 | 5226 | tran_add(tran, &bdrv_remove_child_drv, child); |
46541ee5 VSO |
5227 | } |
5228 | ||
23987471 KW |
5229 | static void undrain_on_clean_cb(void *opaque) |
5230 | { | |
5231 | bdrv_drained_end(opaque); | |
5232 | } | |
5233 | ||
5234 | static TransactionActionDrv undrain_on_clean = { | |
5235 | .clean = undrain_on_clean_cb, | |
5236 | }; | |
5237 | ||
117caba9 VSO |
5238 | static int bdrv_replace_node_noperm(BlockDriverState *from, |
5239 | BlockDriverState *to, | |
5240 | bool auto_skip, Transaction *tran, | |
5241 | Error **errp) | |
5242 | { | |
5243 | BdrvChild *c, *next; | |
5244 | ||
bdb73476 | 5245 | GLOBAL_STATE_CODE(); |
82b54cf5 | 5246 | |
23987471 KW |
5247 | bdrv_drained_begin(from); |
5248 | bdrv_drained_begin(to); | |
5249 | tran_add(tran, &undrain_on_clean, from); | |
5250 | tran_add(tran, &undrain_on_clean, to); | |
5251 | ||
117caba9 VSO |
5252 | QLIST_FOREACH_SAFE(c, &from->parents, next_parent, next) { |
5253 | assert(c->bs == from); | |
5254 | if (!should_update_child(c, to)) { | |
5255 | if (auto_skip) { | |
5256 | continue; | |
5257 | } | |
5258 | error_setg(errp, "Should not change '%s' link to '%s'", | |
5259 | c->name, from->node_name); | |
5260 | return -EINVAL; | |
5261 | } | |
5262 | if (c->frozen) { | |
5263 | error_setg(errp, "Cannot change '%s' link to '%s'", | |
5264 | c->name, from->node_name); | |
5265 | return -EPERM; | |
5266 | } | |
0f0b1e29 | 5267 | bdrv_replace_child_tran(c, to, tran); |
117caba9 VSO |
5268 | } |
5269 | ||
5270 | return 0; | |
5271 | } | |
5272 | ||
313274bb VSO |
5273 | /* |
5274 | * With auto_skip=true bdrv_replace_node_common skips updating from parents | |
5275 | * if it creates a parent-child relation loop or if parent is block-job. | |
5276 | * | |
5277 | * With auto_skip=false the error is returned if from has a parent which should | |
5278 | * not be updated. | |
3108a15c VSO |
5279 | * |
5280 | * With @detach_subchain=true @to must be in a backing chain of @from. In this | |
5281 | * case backing link of the cow-parent of @to is removed. | |
313274bb | 5282 | */ |
a1e708fc VSO |
5283 | static int bdrv_replace_node_common(BlockDriverState *from, |
5284 | BlockDriverState *to, | |
3108a15c VSO |
5285 | bool auto_skip, bool detach_subchain, |
5286 | Error **errp) | |
dd62f1ca | 5287 | { |
3bb0e298 | 5288 | Transaction *tran = tran_new(); |
3bb0e298 | 5289 | g_autoptr(GSList) refresh_list = NULL; |
2d369d6e | 5290 | BlockDriverState *to_cow_parent = NULL; |
234ac1a9 KW |
5291 | int ret; |
5292 | ||
bdb73476 | 5293 | GLOBAL_STATE_CODE(); |
82b54cf5 | 5294 | |
3108a15c VSO |
5295 | if (detach_subchain) { |
5296 | assert(bdrv_chain_contains(from, to)); | |
5297 | assert(from != to); | |
5298 | for (to_cow_parent = from; | |
5299 | bdrv_filter_or_cow_bs(to_cow_parent) != to; | |
5300 | to_cow_parent = bdrv_filter_or_cow_bs(to_cow_parent)) | |
5301 | { | |
5302 | ; | |
5303 | } | |
5304 | } | |
5305 | ||
234ac1a9 KW |
5306 | /* Make sure that @from doesn't go away until we have successfully attached |
5307 | * all of its parents to @to. */ | |
5308 | bdrv_ref(from); | |
dd62f1ca | 5309 | |
f871abd6 | 5310 | assert(qemu_get_current_aio_context() == qemu_get_aio_context()); |
30dd65f3 | 5311 | assert(bdrv_get_aio_context(from) == bdrv_get_aio_context(to)); |
f871abd6 KW |
5312 | bdrv_drained_begin(from); |
5313 | ||
3bb0e298 VSO |
5314 | /* |
5315 | * Do the replacement without permission update. | |
5316 | * Replacement may influence the permissions, we should calculate new | |
5317 | * permissions based on new graph. If we fail, we'll roll-back the | |
5318 | * replacement. | |
5319 | */ | |
117caba9 VSO |
5320 | ret = bdrv_replace_node_noperm(from, to, auto_skip, tran, errp); |
5321 | if (ret < 0) { | |
5322 | goto out; | |
234ac1a9 KW |
5323 | } |
5324 | ||
3108a15c | 5325 | if (detach_subchain) { |
f38eaec4 | 5326 | bdrv_remove_child(bdrv_filter_or_cow_child(to_cow_parent), tran); |
3108a15c VSO |
5327 | } |
5328 | ||
fb0ff4d1 VSO |
5329 | refresh_list = g_slist_prepend(refresh_list, to); |
5330 | refresh_list = g_slist_prepend(refresh_list, from); | |
9bd910e2 | 5331 | |
3bb0e298 VSO |
5332 | ret = bdrv_list_refresh_perms(refresh_list, NULL, tran, errp); |
5333 | if (ret < 0) { | |
5334 | goto out; | |
dd62f1ca | 5335 | } |
234ac1a9 | 5336 | |
a1e708fc VSO |
5337 | ret = 0; |
5338 | ||
234ac1a9 | 5339 | out: |
3bb0e298 VSO |
5340 | tran_finalize(tran, ret); |
5341 | ||
f871abd6 | 5342 | bdrv_drained_end(from); |
234ac1a9 | 5343 | bdrv_unref(from); |
a1e708fc VSO |
5344 | |
5345 | return ret; | |
dd62f1ca KW |
5346 | } |
5347 | ||
a1e708fc VSO |
5348 | int bdrv_replace_node(BlockDriverState *from, BlockDriverState *to, |
5349 | Error **errp) | |
313274bb | 5350 | { |
f791bf7f EGE |
5351 | GLOBAL_STATE_CODE(); |
5352 | ||
3108a15c VSO |
5353 | return bdrv_replace_node_common(from, to, true, false, errp); |
5354 | } | |
5355 | ||
5356 | int bdrv_drop_filter(BlockDriverState *bs, Error **errp) | |
5357 | { | |
f791bf7f EGE |
5358 | GLOBAL_STATE_CODE(); |
5359 | ||
3108a15c VSO |
5360 | return bdrv_replace_node_common(bs, bdrv_filter_or_cow_bs(bs), true, true, |
5361 | errp); | |
313274bb VSO |
5362 | } |
5363 | ||
4ddc07ca PB |
5364 | /* |
5365 | * Add new bs contents at the top of an image chain while the chain is | |
5366 | * live, while keeping required fields on the top layer. | |
5367 | * | |
5368 | * This will modify the BlockDriverState fields, and swap contents | |
5369 | * between bs_new and bs_top. Both bs_new and bs_top are modified. | |
5370 | * | |
2272edcf VSO |
5371 | * bs_new must not be attached to a BlockBackend and must not have backing |
5372 | * child. | |
4ddc07ca PB |
5373 | * |
5374 | * This function does not create any image files. | |
60d90bf4 SG |
5375 | * |
5376 | * The caller must hold the AioContext lock for @bs_top. | |
4ddc07ca | 5377 | */ |
a1e708fc VSO |
5378 | int bdrv_append(BlockDriverState *bs_new, BlockDriverState *bs_top, |
5379 | Error **errp) | |
4ddc07ca | 5380 | { |
2272edcf | 5381 | int ret; |
5bb04747 | 5382 | BdrvChild *child; |
2272edcf | 5383 | Transaction *tran = tran_new(); |
60d90bf4 | 5384 | AioContext *old_context, *new_context = NULL; |
2272edcf | 5385 | |
f791bf7f EGE |
5386 | GLOBAL_STATE_CODE(); |
5387 | ||
2272edcf VSO |
5388 | assert(!bs_new->backing); |
5389 | ||
60d90bf4 SG |
5390 | old_context = bdrv_get_aio_context(bs_top); |
5391 | ||
5bb04747 VSO |
5392 | child = bdrv_attach_child_noperm(bs_new, bs_top, "backing", |
5393 | &child_of_bds, bdrv_backing_role(bs_new), | |
5394 | tran, errp); | |
5395 | if (!child) { | |
5396 | ret = -EINVAL; | |
2272edcf | 5397 | goto out; |
b2c2832c | 5398 | } |
dd62f1ca | 5399 | |
60d90bf4 SG |
5400 | /* |
5401 | * bdrv_attach_child_noperm could change the AioContext of bs_top. | |
5402 | * bdrv_replace_node_noperm calls bdrv_drained_begin, so let's temporarily | |
5403 | * hold the new AioContext, since bdrv_drained_begin calls BDRV_POLL_WHILE | |
5404 | * that assumes the new lock is taken. | |
5405 | */ | |
5406 | new_context = bdrv_get_aio_context(bs_top); | |
5407 | ||
5408 | if (old_context != new_context) { | |
5409 | aio_context_release(old_context); | |
5410 | aio_context_acquire(new_context); | |
5411 | } | |
5412 | ||
2272edcf | 5413 | ret = bdrv_replace_node_noperm(bs_top, bs_new, true, tran, errp); |
a1e708fc | 5414 | if (ret < 0) { |
2272edcf | 5415 | goto out; |
234ac1a9 | 5416 | } |
4ddc07ca | 5417 | |
f1316edb | 5418 | ret = bdrv_refresh_perms(bs_new, tran, errp); |
2272edcf VSO |
5419 | out: |
5420 | tran_finalize(tran, ret); | |
5421 | ||
e19b157f | 5422 | bdrv_graph_rdlock_main_loop(); |
1e4c797c | 5423 | bdrv_refresh_limits(bs_top, NULL, NULL); |
e19b157f | 5424 | bdrv_graph_rdunlock_main_loop(); |
2272edcf | 5425 | |
60d90bf4 SG |
5426 | if (new_context && old_context != new_context) { |
5427 | aio_context_release(new_context); | |
5428 | aio_context_acquire(old_context); | |
5429 | } | |
5430 | ||
2272edcf | 5431 | return ret; |
8802d1fd JC |
5432 | } |
5433 | ||
bd8f4c42 VSO |
5434 | /* Not for empty child */ |
5435 | int bdrv_replace_child_bs(BdrvChild *child, BlockDriverState *new_bs, | |
5436 | Error **errp) | |
5437 | { | |
5438 | int ret; | |
5439 | Transaction *tran = tran_new(); | |
bd8f4c42 VSO |
5440 | g_autoptr(GSList) refresh_list = NULL; |
5441 | BlockDriverState *old_bs = child->bs; | |
5442 | ||
f791bf7f EGE |
5443 | GLOBAL_STATE_CODE(); |
5444 | ||
bd8f4c42 VSO |
5445 | bdrv_ref(old_bs); |
5446 | bdrv_drained_begin(old_bs); | |
5447 | bdrv_drained_begin(new_bs); | |
5448 | ||
0f0b1e29 | 5449 | bdrv_replace_child_tran(child, new_bs, tran); |
bd8f4c42 | 5450 | |
fb0ff4d1 VSO |
5451 | refresh_list = g_slist_prepend(refresh_list, old_bs); |
5452 | refresh_list = g_slist_prepend(refresh_list, new_bs); | |
bd8f4c42 VSO |
5453 | |
5454 | ret = bdrv_list_refresh_perms(refresh_list, NULL, tran, errp); | |
5455 | ||
5456 | tran_finalize(tran, ret); | |
5457 | ||
5458 | bdrv_drained_end(old_bs); | |
5459 | bdrv_drained_end(new_bs); | |
5460 | bdrv_unref(old_bs); | |
5461 | ||
5462 | return ret; | |
5463 | } | |
5464 | ||
4f6fd349 | 5465 | static void bdrv_delete(BlockDriverState *bs) |
b338082b | 5466 | { |
3718d8ab | 5467 | assert(bdrv_op_blocker_is_empty(bs)); |
4f6fd349 | 5468 | assert(!bs->refcnt); |
f791bf7f | 5469 | GLOBAL_STATE_CODE(); |
18846dee | 5470 | |
1b7bdbc1 | 5471 | /* remove from list, if necessary */ |
63eaaae0 KW |
5472 | if (bs->node_name[0] != '\0') { |
5473 | QTAILQ_REMOVE(&graph_bdrv_states, bs, node_list); | |
5474 | } | |
2c1d04e0 HR |
5475 | QTAILQ_REMOVE(&all_bdrv_states, bs, bs_list); |
5476 | ||
30c321f9 AK |
5477 | bdrv_close(bs); |
5478 | ||
7267c094 | 5479 | g_free(bs); |
fc01f7e7 FB |
5480 | } |
5481 | ||
96796fae VSO |
5482 | |
5483 | /* | |
5484 | * Replace @bs by newly created block node. | |
5485 | * | |
5486 | * @options is a QDict of options to pass to the block drivers, or NULL for an | |
5487 | * empty set of options. The reference to the QDict belongs to the block layer | |
5488 | * after the call (even on failure), so if the caller intends to reuse the | |
5489 | * dictionary, it needs to use qobject_ref() before calling bdrv_open. | |
8823407c KW |
5490 | * |
5491 | * The caller holds the AioContext lock for @bs. It must make sure that @bs | |
5492 | * stays in the same AioContext, i.e. @options must not refer to nodes in a | |
5493 | * different AioContext. | |
96796fae VSO |
5494 | */ |
5495 | BlockDriverState *bdrv_insert_node(BlockDriverState *bs, QDict *options, | |
8872ef78 AS |
5496 | int flags, Error **errp) |
5497 | { | |
f053b7e8 VSO |
5498 | ERRP_GUARD(); |
5499 | int ret; | |
8823407c | 5500 | AioContext *ctx = bdrv_get_aio_context(bs); |
b11c8739 VSO |
5501 | BlockDriverState *new_node_bs = NULL; |
5502 | const char *drvname, *node_name; | |
5503 | BlockDriver *drv; | |
5504 | ||
5505 | drvname = qdict_get_try_str(options, "driver"); | |
5506 | if (!drvname) { | |
5507 | error_setg(errp, "driver is not specified"); | |
5508 | goto fail; | |
5509 | } | |
5510 | ||
5511 | drv = bdrv_find_format(drvname); | |
5512 | if (!drv) { | |
5513 | error_setg(errp, "Unknown driver: '%s'", drvname); | |
5514 | goto fail; | |
5515 | } | |
8872ef78 | 5516 | |
b11c8739 VSO |
5517 | node_name = qdict_get_try_str(options, "node-name"); |
5518 | ||
f791bf7f EGE |
5519 | GLOBAL_STATE_CODE(); |
5520 | ||
8823407c KW |
5521 | aio_context_release(ctx); |
5522 | aio_context_acquire(qemu_get_aio_context()); | |
b11c8739 VSO |
5523 | new_node_bs = bdrv_new_open_driver_opts(drv, node_name, options, flags, |
5524 | errp); | |
8823407c KW |
5525 | aio_context_release(qemu_get_aio_context()); |
5526 | aio_context_acquire(ctx); | |
5527 | assert(bdrv_get_aio_context(bs) == ctx); | |
5528 | ||
b11c8739 VSO |
5529 | options = NULL; /* bdrv_new_open_driver() eats options */ |
5530 | if (!new_node_bs) { | |
8872ef78 | 5531 | error_prepend(errp, "Could not create node: "); |
b11c8739 | 5532 | goto fail; |
8872ef78 AS |
5533 | } |
5534 | ||
5535 | bdrv_drained_begin(bs); | |
f053b7e8 | 5536 | ret = bdrv_replace_node(bs, new_node_bs, errp); |
8872ef78 AS |
5537 | bdrv_drained_end(bs); |
5538 | ||
f053b7e8 VSO |
5539 | if (ret < 0) { |
5540 | error_prepend(errp, "Could not replace node: "); | |
b11c8739 | 5541 | goto fail; |
8872ef78 AS |
5542 | } |
5543 | ||
5544 | return new_node_bs; | |
b11c8739 VSO |
5545 | |
5546 | fail: | |
5547 | qobject_unref(options); | |
5548 | bdrv_unref(new_node_bs); | |
5549 | return NULL; | |
8872ef78 AS |
5550 | } |
5551 | ||
e97fc193 AL |
5552 | /* |
5553 | * Run consistency checks on an image | |
5554 | * | |
e076f338 | 5555 | * Returns 0 if the check could be completed (it doesn't mean that the image is |
a1c7273b | 5556 | * free of errors) or -errno when an internal error occurred. The results of the |
e076f338 | 5557 | * check are stored in res. |
e97fc193 | 5558 | */ |
21c2283e VSO |
5559 | int coroutine_fn bdrv_co_check(BlockDriverState *bs, |
5560 | BdrvCheckResult *res, BdrvCheckMode fix) | |
e97fc193 | 5561 | { |
1581a70d | 5562 | IO_CODE(); |
1b3ff9fe | 5563 | assert_bdrv_graph_readable(); |
908bcd54 HR |
5564 | if (bs->drv == NULL) { |
5565 | return -ENOMEDIUM; | |
5566 | } | |
2fd61638 | 5567 | if (bs->drv->bdrv_co_check == NULL) { |
e97fc193 AL |
5568 | return -ENOTSUP; |
5569 | } | |
5570 | ||
e076f338 | 5571 | memset(res, 0, sizeof(*res)); |
2fd61638 PB |
5572 | return bs->drv->bdrv_co_check(bs, res, fix); |
5573 | } | |
5574 | ||
756e6736 KW |
5575 | /* |
5576 | * Return values: | |
5577 | * 0 - success | |
5578 | * -EINVAL - backing format specified, but no file | |
5579 | * -ENOSPC - can't update the backing file because no space is left in the | |
5580 | * image file header | |
5581 | * -ENOTSUP - format driver doesn't support changing the backing file | |
5582 | */ | |
e54ee1b3 | 5583 | int bdrv_change_backing_file(BlockDriverState *bs, const char *backing_file, |
497a30db | 5584 | const char *backing_fmt, bool require) |
756e6736 KW |
5585 | { |
5586 | BlockDriver *drv = bs->drv; | |
469ef350 | 5587 | int ret; |
756e6736 | 5588 | |
f791bf7f EGE |
5589 | GLOBAL_STATE_CODE(); |
5590 | ||
d470ad42 HR |
5591 | if (!drv) { |
5592 | return -ENOMEDIUM; | |
5593 | } | |
5594 | ||
5f377794 PB |
5595 | /* Backing file format doesn't make sense without a backing file */ |
5596 | if (backing_fmt && !backing_file) { | |
5597 | return -EINVAL; | |
5598 | } | |
5599 | ||
497a30db EB |
5600 | if (require && backing_file && !backing_fmt) { |
5601 | return -EINVAL; | |
e54ee1b3 EB |
5602 | } |
5603 | ||
756e6736 | 5604 | if (drv->bdrv_change_backing_file != NULL) { |
469ef350 | 5605 | ret = drv->bdrv_change_backing_file(bs, backing_file, backing_fmt); |
756e6736 | 5606 | } else { |
469ef350 | 5607 | ret = -ENOTSUP; |
756e6736 | 5608 | } |
469ef350 PB |
5609 | |
5610 | if (ret == 0) { | |
5611 | pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_file ?: ""); | |
5612 | pstrcpy(bs->backing_format, sizeof(bs->backing_format), backing_fmt ?: ""); | |
998c2019 HR |
5613 | pstrcpy(bs->auto_backing_file, sizeof(bs->auto_backing_file), |
5614 | backing_file ?: ""); | |
469ef350 PB |
5615 | } |
5616 | return ret; | |
756e6736 KW |
5617 | } |
5618 | ||
6ebdcee2 | 5619 | /* |
dcf3f9b2 HR |
5620 | * Finds the first non-filter node above bs in the chain between |
5621 | * active and bs. The returned node is either an immediate parent of | |
5622 | * bs, or there are only filter nodes between the two. | |
6ebdcee2 JC |
5623 | * |
5624 | * Returns NULL if bs is not found in active's image chain, | |
5625 | * or if active == bs. | |
4caf0fcd JC |
5626 | * |
5627 | * Returns the bottommost base image if bs == NULL. | |
6ebdcee2 JC |
5628 | */ |
5629 | BlockDriverState *bdrv_find_overlay(BlockDriverState *active, | |
5630 | BlockDriverState *bs) | |
5631 | { | |
f791bf7f EGE |
5632 | |
5633 | GLOBAL_STATE_CODE(); | |
5634 | ||
dcf3f9b2 HR |
5635 | bs = bdrv_skip_filters(bs); |
5636 | active = bdrv_skip_filters(active); | |
5637 | ||
5638 | while (active) { | |
5639 | BlockDriverState *next = bdrv_backing_chain_next(active); | |
5640 | if (bs == next) { | |
5641 | return active; | |
5642 | } | |
5643 | active = next; | |
6ebdcee2 JC |
5644 | } |
5645 | ||
dcf3f9b2 | 5646 | return NULL; |
4caf0fcd | 5647 | } |
6ebdcee2 | 5648 | |
4caf0fcd JC |
5649 | /* Given a BDS, searches for the base layer. */ |
5650 | BlockDriverState *bdrv_find_base(BlockDriverState *bs) | |
5651 | { | |
f791bf7f EGE |
5652 | GLOBAL_STATE_CODE(); |
5653 | ||
4caf0fcd | 5654 | return bdrv_find_overlay(bs, NULL); |
6ebdcee2 JC |
5655 | } |
5656 | ||
2cad1ebe | 5657 | /* |
7b99a266 HR |
5658 | * Return true if at least one of the COW (backing) and filter links |
5659 | * between @bs and @base is frozen. @errp is set if that's the case. | |
0f0998f6 | 5660 | * @base must be reachable from @bs, or NULL. |
2cad1ebe AG |
5661 | */ |
5662 | bool bdrv_is_backing_chain_frozen(BlockDriverState *bs, BlockDriverState *base, | |
5663 | Error **errp) | |
5664 | { | |
5665 | BlockDriverState *i; | |
7b99a266 | 5666 | BdrvChild *child; |
2cad1ebe | 5667 | |
f791bf7f EGE |
5668 | GLOBAL_STATE_CODE(); |
5669 | ||
7b99a266 HR |
5670 | for (i = bs; i != base; i = child_bs(child)) { |
5671 | child = bdrv_filter_or_cow_child(i); | |
5672 | ||
5673 | if (child && child->frozen) { | |
2cad1ebe | 5674 | error_setg(errp, "Cannot change '%s' link from '%s' to '%s'", |
7b99a266 | 5675 | child->name, i->node_name, child->bs->node_name); |
2cad1ebe AG |
5676 | return true; |
5677 | } | |
5678 | } | |
5679 | ||
5680 | return false; | |
5681 | } | |
5682 | ||
5683 | /* | |
7b99a266 | 5684 | * Freeze all COW (backing) and filter links between @bs and @base. |
2cad1ebe AG |
5685 | * If any of the links is already frozen the operation is aborted and |
5686 | * none of the links are modified. | |
0f0998f6 | 5687 | * @base must be reachable from @bs, or NULL. |
2cad1ebe AG |
5688 | * Returns 0 on success. On failure returns < 0 and sets @errp. |
5689 | */ | |
5690 | int bdrv_freeze_backing_chain(BlockDriverState *bs, BlockDriverState *base, | |
5691 | Error **errp) | |
5692 | { | |
5693 | BlockDriverState *i; | |
7b99a266 | 5694 | BdrvChild *child; |
2cad1ebe | 5695 | |
f791bf7f EGE |
5696 | GLOBAL_STATE_CODE(); |
5697 | ||
2cad1ebe AG |
5698 | if (bdrv_is_backing_chain_frozen(bs, base, errp)) { |
5699 | return -EPERM; | |
5700 | } | |
5701 | ||
7b99a266 HR |
5702 | for (i = bs; i != base; i = child_bs(child)) { |
5703 | child = bdrv_filter_or_cow_child(i); | |
5704 | if (child && child->bs->never_freeze) { | |
e5182c1c | 5705 | error_setg(errp, "Cannot freeze '%s' link to '%s'", |
7b99a266 | 5706 | child->name, child->bs->node_name); |
e5182c1c HR |
5707 | return -EPERM; |
5708 | } | |
5709 | } | |
5710 | ||
7b99a266 HR |
5711 | for (i = bs; i != base; i = child_bs(child)) { |
5712 | child = bdrv_filter_or_cow_child(i); | |
5713 | if (child) { | |
5714 | child->frozen = true; | |
0f0998f6 | 5715 | } |
2cad1ebe AG |
5716 | } |
5717 | ||
5718 | return 0; | |
5719 | } | |
5720 | ||
5721 | /* | |
7b99a266 HR |
5722 | * Unfreeze all COW (backing) and filter links between @bs and @base. |
5723 | * The caller must ensure that all links are frozen before using this | |
5724 | * function. | |
0f0998f6 | 5725 | * @base must be reachable from @bs, or NULL. |
2cad1ebe AG |
5726 | */ |
5727 | void bdrv_unfreeze_backing_chain(BlockDriverState *bs, BlockDriverState *base) | |
5728 | { | |
5729 | BlockDriverState *i; | |
7b99a266 | 5730 | BdrvChild *child; |
2cad1ebe | 5731 | |
f791bf7f EGE |
5732 | GLOBAL_STATE_CODE(); |
5733 | ||
7b99a266 HR |
5734 | for (i = bs; i != base; i = child_bs(child)) { |
5735 | child = bdrv_filter_or_cow_child(i); | |
5736 | if (child) { | |
5737 | assert(child->frozen); | |
5738 | child->frozen = false; | |
0f0998f6 | 5739 | } |
2cad1ebe AG |
5740 | } |
5741 | } | |
5742 | ||
6ebdcee2 JC |
5743 | /* |
5744 | * Drops images above 'base' up to and including 'top', and sets the image | |
5745 | * above 'top' to have base as its backing file. | |
5746 | * | |
5747 | * Requires that the overlay to 'top' is opened r/w, so that the backing file | |
5748 | * information in 'bs' can be properly updated. | |
5749 | * | |
5750 | * E.g., this will convert the following chain: | |
5751 | * bottom <- base <- intermediate <- top <- active | |
5752 | * | |
5753 | * to | |
5754 | * | |
5755 | * bottom <- base <- active | |
5756 | * | |
5757 | * It is allowed for bottom==base, in which case it converts: | |
5758 | * | |
5759 | * base <- intermediate <- top <- active | |
5760 | * | |
5761 | * to | |
5762 | * | |
5763 | * base <- active | |
5764 | * | |
54e26900 JC |
5765 | * If backing_file_str is non-NULL, it will be used when modifying top's |
5766 | * overlay image metadata. | |
5767 | * | |
6ebdcee2 JC |
5768 | * Error conditions: |
5769 | * if active == top, that is considered an error | |
5770 | * | |
5771 | */ | |
bde70715 KW |
5772 | int bdrv_drop_intermediate(BlockDriverState *top, BlockDriverState *base, |
5773 | const char *backing_file_str) | |
6ebdcee2 | 5774 | { |
6bd858b3 AG |
5775 | BlockDriverState *explicit_top = top; |
5776 | bool update_inherits_from; | |
d669ed6a | 5777 | BdrvChild *c; |
12fa4af6 | 5778 | Error *local_err = NULL; |
6ebdcee2 | 5779 | int ret = -EIO; |
d669ed6a VSO |
5780 | g_autoptr(GSList) updated_children = NULL; |
5781 | GSList *p; | |
6ebdcee2 | 5782 | |
f791bf7f EGE |
5783 | GLOBAL_STATE_CODE(); |
5784 | ||
6858eba0 | 5785 | bdrv_ref(top); |
631086de | 5786 | bdrv_drained_begin(base); |
6858eba0 | 5787 | |
6ebdcee2 JC |
5788 | if (!top->drv || !base->drv) { |
5789 | goto exit; | |
5790 | } | |
5791 | ||
5db15a57 KW |
5792 | /* Make sure that base is in the backing chain of top */ |
5793 | if (!bdrv_chain_contains(top, base)) { | |
6ebdcee2 JC |
5794 | goto exit; |
5795 | } | |
5796 | ||
6bd858b3 AG |
5797 | /* If 'base' recursively inherits from 'top' then we should set |
5798 | * base->inherits_from to top->inherits_from after 'top' and all | |
5799 | * other intermediate nodes have been dropped. | |
5800 | * If 'top' is an implicit node (e.g. "commit_top") we should skip | |
5801 | * it because no one inherits from it. We use explicit_top for that. */ | |
dcf3f9b2 | 5802 | explicit_top = bdrv_skip_implicit_filters(explicit_top); |
6bd858b3 AG |
5803 | update_inherits_from = bdrv_inherits_from_recursive(base, explicit_top); |
5804 | ||
6ebdcee2 | 5805 | /* success - we can delete the intermediate states, and link top->base */ |
f30c66ba HR |
5806 | if (!backing_file_str) { |
5807 | bdrv_refresh_filename(base); | |
5808 | backing_file_str = base->filename; | |
5809 | } | |
61f09cea | 5810 | |
d669ed6a VSO |
5811 | QLIST_FOREACH(c, &top->parents, next_parent) { |
5812 | updated_children = g_slist_prepend(updated_children, c); | |
5813 | } | |
5814 | ||
3108a15c VSO |
5815 | /* |
5816 | * It seems correct to pass detach_subchain=true here, but it triggers | |
5817 | * one more yet not fixed bug, when due to nested aio_poll loop we switch to | |
5818 | * another drained section, which modify the graph (for example, removing | |
5819 | * the child, which we keep in updated_children list). So, it's a TODO. | |
5820 | * | |
5821 | * Note, bug triggered if pass detach_subchain=true here and run | |
5822 | * test-bdrv-drain. test_drop_intermediate_poll() test-case will crash. | |
5823 | * That's a FIXME. | |
5824 | */ | |
5825 | bdrv_replace_node_common(top, base, false, false, &local_err); | |
d669ed6a VSO |
5826 | if (local_err) { |
5827 | error_report_err(local_err); | |
5828 | goto exit; | |
5829 | } | |
5830 | ||
5831 | for (p = updated_children; p; p = p->next) { | |
5832 | c = p->data; | |
12fa4af6 | 5833 | |
bd86fb99 HR |
5834 | if (c->klass->update_filename) { |
5835 | ret = c->klass->update_filename(c, base, backing_file_str, | |
5836 | &local_err); | |
61f09cea | 5837 | if (ret < 0) { |
d669ed6a VSO |
5838 | /* |
5839 | * TODO: Actually, we want to rollback all previous iterations | |
5840 | * of this loop, and (which is almost impossible) previous | |
5841 | * bdrv_replace_node()... | |
5842 | * | |
5843 | * Note, that c->klass->update_filename may lead to permission | |
5844 | * update, so it's a bad idea to call it inside permission | |
5845 | * update transaction of bdrv_replace_node. | |
5846 | */ | |
61f09cea KW |
5847 | error_report_err(local_err); |
5848 | goto exit; | |
5849 | } | |
5850 | } | |
12fa4af6 | 5851 | } |
6ebdcee2 | 5852 | |
6bd858b3 AG |
5853 | if (update_inherits_from) { |
5854 | base->inherits_from = explicit_top->inherits_from; | |
5855 | } | |
5856 | ||
6ebdcee2 | 5857 | ret = 0; |
6ebdcee2 | 5858 | exit: |
631086de | 5859 | bdrv_drained_end(base); |
6858eba0 | 5860 | bdrv_unref(top); |
6ebdcee2 JC |
5861 | return ret; |
5862 | } | |
5863 | ||
081e4650 | 5864 | /** |
82618d7b | 5865 | * Implementation of BlockDriver.bdrv_co_get_allocated_file_size() that |
081e4650 HR |
5866 | * sums the size of all data-bearing children. (This excludes backing |
5867 | * children.) | |
5868 | */ | |
de335638 EGE |
5869 | static int64_t coroutine_fn GRAPH_RDLOCK |
5870 | bdrv_sum_allocated_file_size(BlockDriverState *bs) | |
081e4650 HR |
5871 | { |
5872 | BdrvChild *child; | |
5873 | int64_t child_size, sum = 0; | |
5874 | ||
5875 | QLIST_FOREACH(child, &bs->children, next) { | |
5876 | if (child->role & (BDRV_CHILD_DATA | BDRV_CHILD_METADATA | | |
5877 | BDRV_CHILD_FILTERED)) | |
5878 | { | |
82618d7b | 5879 | child_size = bdrv_co_get_allocated_file_size(child->bs); |
081e4650 HR |
5880 | if (child_size < 0) { |
5881 | return child_size; | |
5882 | } | |
5883 | sum += child_size; | |
5884 | } | |
5885 | } | |
5886 | ||
5887 | return sum; | |
5888 | } | |
5889 | ||
61007b31 SH |
5890 | /** |
5891 | * Length of a allocated file in bytes. Sparse files are counted by actual | |
5892 | * allocated space. Return < 0 if error or unknown. | |
5893 | */ | |
82618d7b | 5894 | int64_t coroutine_fn bdrv_co_get_allocated_file_size(BlockDriverState *bs) |
71d0770c | 5895 | { |
61007b31 | 5896 | BlockDriver *drv = bs->drv; |
384a48fb | 5897 | IO_CODE(); |
de335638 | 5898 | assert_bdrv_graph_readable(); |
384a48fb | 5899 | |
61007b31 SH |
5900 | if (!drv) { |
5901 | return -ENOMEDIUM; | |
8f4754ed | 5902 | } |
82618d7b EGE |
5903 | if (drv->bdrv_co_get_allocated_file_size) { |
5904 | return drv->bdrv_co_get_allocated_file_size(bs); | |
61007b31 | 5905 | } |
081e4650 HR |
5906 | |
5907 | if (drv->bdrv_file_open) { | |
5908 | /* | |
5909 | * Protocol drivers default to -ENOTSUP (most of their data is | |
5910 | * not stored in any of their children (if they even have any), | |
5911 | * so there is no generic way to figure it out). | |
5912 | */ | |
5913 | return -ENOTSUP; | |
5914 | } else if (drv->is_filter) { | |
5915 | /* Filter drivers default to the size of their filtered child */ | |
82618d7b | 5916 | return bdrv_co_get_allocated_file_size(bdrv_filter_bs(bs)); |
081e4650 HR |
5917 | } else { |
5918 | /* Other drivers default to summing their children's sizes */ | |
5919 | return bdrv_sum_allocated_file_size(bs); | |
1c9805a3 SH |
5920 | } |
5921 | } | |
e7a8a783 | 5922 | |
90880ff1 SH |
5923 | /* |
5924 | * bdrv_measure: | |
5925 | * @drv: Format driver | |
5926 | * @opts: Creation options for new image | |
5927 | * @in_bs: Existing image containing data for new image (may be NULL) | |
5928 | * @errp: Error object | |
5929 | * Returns: A #BlockMeasureInfo (free using qapi_free_BlockMeasureInfo()) | |
5930 | * or NULL on error | |
5931 | * | |
5932 | * Calculate file size required to create a new image. | |
5933 | * | |
5934 | * If @in_bs is given then space for allocated clusters and zero clusters | |
5935 | * from that image are included in the calculation. If @opts contains a | |
5936 | * backing file that is shared by @in_bs then backing clusters may be omitted | |
5937 | * from the calculation. | |
5938 | * | |
5939 | * If @in_bs is NULL then the calculation includes no allocated clusters | |
5940 | * unless a preallocation option is given in @opts. | |
5941 | * | |
5942 | * Note that @in_bs may use a different BlockDriver from @drv. | |
5943 | * | |
5944 | * If an error occurs the @errp pointer is set. | |
5945 | */ | |
5946 | BlockMeasureInfo *bdrv_measure(BlockDriver *drv, QemuOpts *opts, | |
5947 | BlockDriverState *in_bs, Error **errp) | |
5948 | { | |
384a48fb | 5949 | IO_CODE(); |
90880ff1 SH |
5950 | if (!drv->bdrv_measure) { |
5951 | error_setg(errp, "Block driver '%s' does not support size measurement", | |
5952 | drv->format_name); | |
5953 | return NULL; | |
5954 | } | |
5955 | ||
5956 | return drv->bdrv_measure(opts, in_bs, errp); | |
5957 | } | |
5958 | ||
61007b31 SH |
5959 | /** |
5960 | * Return number of sectors on success, -errno on error. | |
1c9805a3 | 5961 | */ |
c86422c5 | 5962 | int64_t coroutine_fn bdrv_co_nb_sectors(BlockDriverState *bs) |
1c9805a3 | 5963 | { |
61007b31 | 5964 | BlockDriver *drv = bs->drv; |
384a48fb | 5965 | IO_CODE(); |
8ab8140a | 5966 | assert_bdrv_graph_readable(); |
498e386c | 5967 | |
61007b31 SH |
5968 | if (!drv) |
5969 | return -ENOMEDIUM; | |
2572b37a | 5970 | |
160a29e2 | 5971 | if (bs->bl.has_variable_length) { |
c86422c5 | 5972 | int ret = bdrv_co_refresh_total_sectors(bs, bs->total_sectors); |
61007b31 SH |
5973 | if (ret < 0) { |
5974 | return ret; | |
1c9805a3 SH |
5975 | } |
5976 | } | |
61007b31 | 5977 | return bs->total_sectors; |
1c9805a3 | 5978 | } |
b338082b | 5979 | |
81f730d4 PB |
5980 | /* |
5981 | * This wrapper is written by hand because this function is in the hot I/O path, | |
5982 | * via blk_get_geometry. | |
5983 | */ | |
5984 | int64_t coroutine_mixed_fn bdrv_nb_sectors(BlockDriverState *bs) | |
5985 | { | |
5986 | BlockDriver *drv = bs->drv; | |
5987 | IO_CODE(); | |
5988 | ||
5989 | if (!drv) | |
5990 | return -ENOMEDIUM; | |
5991 | ||
5992 | if (bs->bl.has_variable_length) { | |
5993 | int ret = bdrv_refresh_total_sectors(bs, bs->total_sectors); | |
5994 | if (ret < 0) { | |
5995 | return ret; | |
5996 | } | |
5997 | } | |
5998 | ||
5999 | return bs->total_sectors; | |
6000 | } | |
6001 | ||
61007b31 SH |
6002 | /** |
6003 | * Return length in bytes on success, -errno on error. | |
6004 | * The length is always a multiple of BDRV_SECTOR_SIZE. | |
8d3b1a2d | 6005 | */ |
c86422c5 | 6006 | int64_t coroutine_fn bdrv_co_getlength(BlockDriverState *bs) |
8d3b1a2d | 6007 | { |
c86422c5 | 6008 | int64_t ret; |
384a48fb | 6009 | IO_CODE(); |
8ab8140a | 6010 | assert_bdrv_graph_readable(); |
8d3b1a2d | 6011 | |
c86422c5 | 6012 | ret = bdrv_co_nb_sectors(bs); |
122860ba EB |
6013 | if (ret < 0) { |
6014 | return ret; | |
6015 | } | |
6016 | if (ret > INT64_MAX / BDRV_SECTOR_SIZE) { | |
6017 | return -EFBIG; | |
6018 | } | |
6019 | return ret * BDRV_SECTOR_SIZE; | |
fc01f7e7 FB |
6020 | } |
6021 | ||
54115412 | 6022 | bool bdrv_is_sg(BlockDriverState *bs) |
f08145fe | 6023 | { |
384a48fb | 6024 | IO_CODE(); |
61007b31 | 6025 | return bs->sg; |
f08145fe KW |
6026 | } |
6027 | ||
ae23f786 HR |
6028 | /** |
6029 | * Return whether the given node supports compressed writes. | |
6030 | */ | |
6031 | bool bdrv_supports_compressed_writes(BlockDriverState *bs) | |
6032 | { | |
6033 | BlockDriverState *filtered; | |
384a48fb | 6034 | IO_CODE(); |
ae23f786 HR |
6035 | |
6036 | if (!bs->drv || !block_driver_can_compress(bs->drv)) { | |
6037 | return false; | |
6038 | } | |
6039 | ||
6040 | filtered = bdrv_filter_bs(bs); | |
6041 | if (filtered) { | |
6042 | /* | |
6043 | * Filters can only forward compressed writes, so we have to | |
6044 | * check the child. | |
6045 | */ | |
6046 | return bdrv_supports_compressed_writes(filtered); | |
6047 | } | |
6048 | ||
6049 | return true; | |
6050 | } | |
6051 | ||
61007b31 | 6052 | const char *bdrv_get_format_name(BlockDriverState *bs) |
40b4f539 | 6053 | { |
384a48fb | 6054 | IO_CODE(); |
61007b31 | 6055 | return bs->drv ? bs->drv->format_name : NULL; |
40b4f539 KW |
6056 | } |
6057 | ||
61007b31 | 6058 | static int qsort_strcmp(const void *a, const void *b) |
40b4f539 | 6059 | { |
ceff5bd7 | 6060 | return strcmp(*(char *const *)a, *(char *const *)b); |
40b4f539 KW |
6061 | } |
6062 | ||
61007b31 | 6063 | void bdrv_iterate_format(void (*it)(void *opaque, const char *name), |
9ac404c5 | 6064 | void *opaque, bool read_only) |
40b4f539 | 6065 | { |
61007b31 SH |
6066 | BlockDriver *drv; |
6067 | int count = 0; | |
6068 | int i; | |
6069 | const char **formats = NULL; | |
40b4f539 | 6070 | |
f791bf7f EGE |
6071 | GLOBAL_STATE_CODE(); |
6072 | ||
61007b31 SH |
6073 | QLIST_FOREACH(drv, &bdrv_drivers, list) { |
6074 | if (drv->format_name) { | |
6075 | bool found = false; | |
6076 | int i = count; | |
9ac404c5 AS |
6077 | |
6078 | if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv, read_only)) { | |
6079 | continue; | |
6080 | } | |
6081 | ||
61007b31 SH |
6082 | while (formats && i && !found) { |
6083 | found = !strcmp(formats[--i], drv->format_name); | |
6084 | } | |
e2a305fb | 6085 | |
61007b31 SH |
6086 | if (!found) { |
6087 | formats = g_renew(const char *, formats, count + 1); | |
6088 | formats[count++] = drv->format_name; | |
6089 | } | |
6c5a42ac | 6090 | } |
61007b31 | 6091 | } |
6c5a42ac | 6092 | |
eb0df69f HR |
6093 | for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); i++) { |
6094 | const char *format_name = block_driver_modules[i].format_name; | |
6095 | ||
6096 | if (format_name) { | |
6097 | bool found = false; | |
6098 | int j = count; | |
6099 | ||
9ac404c5 AS |
6100 | if (use_bdrv_whitelist && |
6101 | !bdrv_format_is_whitelisted(format_name, read_only)) { | |
6102 | continue; | |
6103 | } | |
6104 | ||
eb0df69f HR |
6105 | while (formats && j && !found) { |
6106 | found = !strcmp(formats[--j], format_name); | |
6107 | } | |
6108 | ||
6109 | if (!found) { | |
6110 | formats = g_renew(const char *, formats, count + 1); | |
6111 | formats[count++] = format_name; | |
6112 | } | |
6113 | } | |
6114 | } | |
6115 | ||
61007b31 | 6116 | qsort(formats, count, sizeof(formats[0]), qsort_strcmp); |
40b4f539 | 6117 | |
61007b31 SH |
6118 | for (i = 0; i < count; i++) { |
6119 | it(opaque, formats[i]); | |
6120 | } | |
40b4f539 | 6121 | |
61007b31 SH |
6122 | g_free(formats); |
6123 | } | |
40b4f539 | 6124 | |
61007b31 SH |
6125 | /* This function is to find a node in the bs graph */ |
6126 | BlockDriverState *bdrv_find_node(const char *node_name) | |
6127 | { | |
6128 | BlockDriverState *bs; | |
391827eb | 6129 | |
61007b31 | 6130 | assert(node_name); |
f791bf7f | 6131 | GLOBAL_STATE_CODE(); |
40b4f539 | 6132 | |
61007b31 SH |
6133 | QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) { |
6134 | if (!strcmp(node_name, bs->node_name)) { | |
6135 | return bs; | |
40b4f539 KW |
6136 | } |
6137 | } | |
61007b31 | 6138 | return NULL; |
40b4f539 KW |
6139 | } |
6140 | ||
61007b31 | 6141 | /* Put this QMP function here so it can access the static graph_bdrv_states. */ |
facda544 PK |
6142 | BlockDeviceInfoList *bdrv_named_nodes_list(bool flat, |
6143 | Error **errp) | |
40b4f539 | 6144 | { |
9812e712 | 6145 | BlockDeviceInfoList *list; |
61007b31 | 6146 | BlockDriverState *bs; |
40b4f539 | 6147 | |
f791bf7f EGE |
6148 | GLOBAL_STATE_CODE(); |
6149 | ||
61007b31 SH |
6150 | list = NULL; |
6151 | QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) { | |
facda544 | 6152 | BlockDeviceInfo *info = bdrv_block_device_info(NULL, bs, flat, errp); |
61007b31 SH |
6153 | if (!info) { |
6154 | qapi_free_BlockDeviceInfoList(list); | |
6155 | return NULL; | |
301db7c2 | 6156 | } |
9812e712 | 6157 | QAPI_LIST_PREPEND(list, info); |
301db7c2 RH |
6158 | } |
6159 | ||
61007b31 SH |
6160 | return list; |
6161 | } | |
40b4f539 | 6162 | |
5d3b4e99 VSO |
6163 | typedef struct XDbgBlockGraphConstructor { |
6164 | XDbgBlockGraph *graph; | |
6165 | GHashTable *graph_nodes; | |
6166 | } XDbgBlockGraphConstructor; | |
6167 | ||
6168 | static XDbgBlockGraphConstructor *xdbg_graph_new(void) | |
6169 | { | |
6170 | XDbgBlockGraphConstructor *gr = g_new(XDbgBlockGraphConstructor, 1); | |
6171 | ||
6172 | gr->graph = g_new0(XDbgBlockGraph, 1); | |
6173 | gr->graph_nodes = g_hash_table_new(NULL, NULL); | |
6174 | ||
6175 | return gr; | |
6176 | } | |
6177 | ||
6178 | static XDbgBlockGraph *xdbg_graph_finalize(XDbgBlockGraphConstructor *gr) | |
6179 | { | |
6180 | XDbgBlockGraph *graph = gr->graph; | |
6181 | ||
6182 | g_hash_table_destroy(gr->graph_nodes); | |
6183 | g_free(gr); | |
6184 | ||
6185 | return graph; | |
6186 | } | |
6187 | ||
6188 | static uintptr_t xdbg_graph_node_num(XDbgBlockGraphConstructor *gr, void *node) | |
6189 | { | |
6190 | uintptr_t ret = (uintptr_t)g_hash_table_lookup(gr->graph_nodes, node); | |
6191 | ||
6192 | if (ret != 0) { | |
6193 | return ret; | |
6194 | } | |
6195 | ||
6196 | /* | |
6197 | * Start counting from 1, not 0, because 0 interferes with not-found (NULL) | |
6198 | * answer of g_hash_table_lookup. | |
6199 | */ | |
6200 | ret = g_hash_table_size(gr->graph_nodes) + 1; | |
6201 | g_hash_table_insert(gr->graph_nodes, node, (void *)ret); | |
6202 | ||
6203 | return ret; | |
6204 | } | |
6205 | ||
6206 | static void xdbg_graph_add_node(XDbgBlockGraphConstructor *gr, void *node, | |
6207 | XDbgBlockGraphNodeType type, const char *name) | |
6208 | { | |
6209 | XDbgBlockGraphNode *n; | |
6210 | ||
6211 | n = g_new0(XDbgBlockGraphNode, 1); | |
6212 | ||
6213 | n->id = xdbg_graph_node_num(gr, node); | |
6214 | n->type = type; | |
6215 | n->name = g_strdup(name); | |
6216 | ||
9812e712 | 6217 | QAPI_LIST_PREPEND(gr->graph->nodes, n); |
5d3b4e99 VSO |
6218 | } |
6219 | ||
6220 | static void xdbg_graph_add_edge(XDbgBlockGraphConstructor *gr, void *parent, | |
6221 | const BdrvChild *child) | |
6222 | { | |
cdb1cec8 | 6223 | BlockPermission qapi_perm; |
5d3b4e99 | 6224 | XDbgBlockGraphEdge *edge; |
862fded9 | 6225 | GLOBAL_STATE_CODE(); |
5d3b4e99 | 6226 | |
5d3b4e99 VSO |
6227 | edge = g_new0(XDbgBlockGraphEdge, 1); |
6228 | ||
6229 | edge->parent = xdbg_graph_node_num(gr, parent); | |
6230 | edge->child = xdbg_graph_node_num(gr, child->bs); | |
6231 | edge->name = g_strdup(child->name); | |
6232 | ||
cdb1cec8 HR |
6233 | for (qapi_perm = 0; qapi_perm < BLOCK_PERMISSION__MAX; qapi_perm++) { |
6234 | uint64_t flag = bdrv_qapi_perm_to_blk_perm(qapi_perm); | |
6235 | ||
6236 | if (flag & child->perm) { | |
9812e712 | 6237 | QAPI_LIST_PREPEND(edge->perm, qapi_perm); |
5d3b4e99 | 6238 | } |
cdb1cec8 | 6239 | if (flag & child->shared_perm) { |
9812e712 | 6240 | QAPI_LIST_PREPEND(edge->shared_perm, qapi_perm); |
5d3b4e99 VSO |
6241 | } |
6242 | } | |
6243 | ||
9812e712 | 6244 | QAPI_LIST_PREPEND(gr->graph->edges, edge); |
5d3b4e99 VSO |
6245 | } |
6246 | ||
6247 | ||
6248 | XDbgBlockGraph *bdrv_get_xdbg_block_graph(Error **errp) | |
6249 | { | |
6250 | BlockBackend *blk; | |
6251 | BlockJob *job; | |
6252 | BlockDriverState *bs; | |
6253 | BdrvChild *child; | |
6254 | XDbgBlockGraphConstructor *gr = xdbg_graph_new(); | |
6255 | ||
f791bf7f EGE |
6256 | GLOBAL_STATE_CODE(); |
6257 | ||
5d3b4e99 VSO |
6258 | for (blk = blk_all_next(NULL); blk; blk = blk_all_next(blk)) { |
6259 | char *allocated_name = NULL; | |
6260 | const char *name = blk_name(blk); | |
6261 | ||
6262 | if (!*name) { | |
6263 | name = allocated_name = blk_get_attached_dev_id(blk); | |
6264 | } | |
6265 | xdbg_graph_add_node(gr, blk, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_BACKEND, | |
6266 | name); | |
6267 | g_free(allocated_name); | |
6268 | if (blk_root(blk)) { | |
6269 | xdbg_graph_add_edge(gr, blk, blk_root(blk)); | |
6270 | } | |
6271 | } | |
6272 | ||
880eeec6 EGE |
6273 | WITH_JOB_LOCK_GUARD() { |
6274 | for (job = block_job_next_locked(NULL); job; | |
6275 | job = block_job_next_locked(job)) { | |
6276 | GSList *el; | |
5d3b4e99 | 6277 | |
880eeec6 EGE |
6278 | xdbg_graph_add_node(gr, job, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_JOB, |
6279 | job->job.id); | |
6280 | for (el = job->nodes; el; el = el->next) { | |
6281 | xdbg_graph_add_edge(gr, job, (BdrvChild *)el->data); | |
6282 | } | |
5d3b4e99 VSO |
6283 | } |
6284 | } | |
6285 | ||
6286 | QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) { | |
6287 | xdbg_graph_add_node(gr, bs, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_DRIVER, | |
6288 | bs->node_name); | |
6289 | QLIST_FOREACH(child, &bs->children, next) { | |
6290 | xdbg_graph_add_edge(gr, bs, child); | |
6291 | } | |
6292 | } | |
6293 | ||
6294 | return xdbg_graph_finalize(gr); | |
6295 | } | |
6296 | ||
61007b31 SH |
6297 | BlockDriverState *bdrv_lookup_bs(const char *device, |
6298 | const char *node_name, | |
6299 | Error **errp) | |
6300 | { | |
6301 | BlockBackend *blk; | |
6302 | BlockDriverState *bs; | |
40b4f539 | 6303 | |
f791bf7f EGE |
6304 | GLOBAL_STATE_CODE(); |
6305 | ||
61007b31 SH |
6306 | if (device) { |
6307 | blk = blk_by_name(device); | |
40b4f539 | 6308 | |
61007b31 | 6309 | if (blk) { |
9f4ed6fb AG |
6310 | bs = blk_bs(blk); |
6311 | if (!bs) { | |
5433c24f | 6312 | error_setg(errp, "Device '%s' has no medium", device); |
5433c24f HR |
6313 | } |
6314 | ||
9f4ed6fb | 6315 | return bs; |
61007b31 SH |
6316 | } |
6317 | } | |
40b4f539 | 6318 | |
61007b31 SH |
6319 | if (node_name) { |
6320 | bs = bdrv_find_node(node_name); | |
6d519a5f | 6321 | |
61007b31 SH |
6322 | if (bs) { |
6323 | return bs; | |
6324 | } | |
40b4f539 KW |
6325 | } |
6326 | ||
785ec4b1 | 6327 | error_setg(errp, "Cannot find device=\'%s\' nor node-name=\'%s\'", |
61007b31 SH |
6328 | device ? device : "", |
6329 | node_name ? node_name : ""); | |
6330 | return NULL; | |
40b4f539 KW |
6331 | } |
6332 | ||
61007b31 SH |
6333 | /* If 'base' is in the same chain as 'top', return true. Otherwise, |
6334 | * return false. If either argument is NULL, return false. */ | |
6335 | bool bdrv_chain_contains(BlockDriverState *top, BlockDriverState *base) | |
83f64091 | 6336 | { |
f791bf7f EGE |
6337 | |
6338 | GLOBAL_STATE_CODE(); | |
6339 | ||
61007b31 | 6340 | while (top && top != base) { |
dcf3f9b2 | 6341 | top = bdrv_filter_or_cow_bs(top); |
02c50efe | 6342 | } |
61007b31 SH |
6343 | |
6344 | return top != NULL; | |
02c50efe FZ |
6345 | } |
6346 | ||
61007b31 | 6347 | BlockDriverState *bdrv_next_node(BlockDriverState *bs) |
02c50efe | 6348 | { |
f791bf7f | 6349 | GLOBAL_STATE_CODE(); |
61007b31 SH |
6350 | if (!bs) { |
6351 | return QTAILQ_FIRST(&graph_bdrv_states); | |
02c50efe | 6352 | } |
61007b31 | 6353 | return QTAILQ_NEXT(bs, node_list); |
83f64091 FB |
6354 | } |
6355 | ||
0f12264e KW |
6356 | BlockDriverState *bdrv_next_all_states(BlockDriverState *bs) |
6357 | { | |
f791bf7f | 6358 | GLOBAL_STATE_CODE(); |
0f12264e KW |
6359 | if (!bs) { |
6360 | return QTAILQ_FIRST(&all_bdrv_states); | |
6361 | } | |
6362 | return QTAILQ_NEXT(bs, bs_list); | |
6363 | } | |
6364 | ||
61007b31 | 6365 | const char *bdrv_get_node_name(const BlockDriverState *bs) |
83f64091 | 6366 | { |
384a48fb | 6367 | IO_CODE(); |
61007b31 | 6368 | return bs->node_name; |
beac80cd FB |
6369 | } |
6370 | ||
1f0c461b | 6371 | const char *bdrv_get_parent_name(const BlockDriverState *bs) |
4c265bf9 KW |
6372 | { |
6373 | BdrvChild *c; | |
6374 | const char *name; | |
967d7905 | 6375 | IO_CODE(); |
4c265bf9 KW |
6376 | |
6377 | /* If multiple parents have a name, just pick the first one. */ | |
6378 | QLIST_FOREACH(c, &bs->parents, next_parent) { | |
bd86fb99 HR |
6379 | if (c->klass->get_name) { |
6380 | name = c->klass->get_name(c); | |
4c265bf9 KW |
6381 | if (name && *name) { |
6382 | return name; | |
6383 | } | |
6384 | } | |
6385 | } | |
6386 | ||
6387 | return NULL; | |
6388 | } | |
6389 | ||
61007b31 SH |
6390 | /* TODO check what callers really want: bs->node_name or blk_name() */ |
6391 | const char *bdrv_get_device_name(const BlockDriverState *bs) | |
beac80cd | 6392 | { |
384a48fb | 6393 | IO_CODE(); |
4c265bf9 | 6394 | return bdrv_get_parent_name(bs) ?: ""; |
f141eafe | 6395 | } |
83f64091 | 6396 | |
61007b31 SH |
6397 | /* This can be used to identify nodes that might not have a device |
6398 | * name associated. Since node and device names live in the same | |
6399 | * namespace, the result is unambiguous. The exception is if both are | |
6400 | * absent, then this returns an empty (non-null) string. */ | |
6401 | const char *bdrv_get_device_or_node_name(const BlockDriverState *bs) | |
f141eafe | 6402 | { |
384a48fb | 6403 | IO_CODE(); |
4c265bf9 | 6404 | return bdrv_get_parent_name(bs) ?: bs->node_name; |
beac80cd | 6405 | } |
beac80cd | 6406 | |
61007b31 | 6407 | int bdrv_get_flags(BlockDriverState *bs) |
0b5a2445 | 6408 | { |
15aee7ac | 6409 | IO_CODE(); |
61007b31 | 6410 | return bs->open_flags; |
0b5a2445 PB |
6411 | } |
6412 | ||
61007b31 | 6413 | int bdrv_has_zero_init_1(BlockDriverState *bs) |
68485420 | 6414 | { |
f791bf7f | 6415 | GLOBAL_STATE_CODE(); |
61007b31 | 6416 | return 1; |
0b5a2445 PB |
6417 | } |
6418 | ||
61007b31 | 6419 | int bdrv_has_zero_init(BlockDriverState *bs) |
0b5a2445 | 6420 | { |
93393e69 | 6421 | BlockDriverState *filtered; |
f791bf7f | 6422 | GLOBAL_STATE_CODE(); |
93393e69 | 6423 | |
d470ad42 HR |
6424 | if (!bs->drv) { |
6425 | return 0; | |
6426 | } | |
0b5a2445 | 6427 | |
61007b31 SH |
6428 | /* If BS is a copy on write image, it is initialized to |
6429 | the contents of the base image, which may not be zeroes. */ | |
34778172 | 6430 | if (bdrv_cow_child(bs)) { |
61007b31 SH |
6431 | return 0; |
6432 | } | |
6433 | if (bs->drv->bdrv_has_zero_init) { | |
6434 | return bs->drv->bdrv_has_zero_init(bs); | |
0b5a2445 | 6435 | } |
93393e69 HR |
6436 | |
6437 | filtered = bdrv_filter_bs(bs); | |
6438 | if (filtered) { | |
6439 | return bdrv_has_zero_init(filtered); | |
5a612c00 | 6440 | } |
61007b31 SH |
6441 | |
6442 | /* safe default */ | |
6443 | return 0; | |
68485420 KW |
6444 | } |
6445 | ||
61007b31 | 6446 | bool bdrv_can_write_zeroes_with_unmap(BlockDriverState *bs) |
68485420 | 6447 | { |
384a48fb | 6448 | IO_CODE(); |
2f0342ef | 6449 | if (!(bs->open_flags & BDRV_O_UNMAP)) { |
61007b31 SH |
6450 | return false; |
6451 | } | |
68485420 | 6452 | |
e24d813b | 6453 | return bs->supported_zero_flags & BDRV_REQ_MAY_UNMAP; |
68485420 KW |
6454 | } |
6455 | ||
61007b31 SH |
6456 | void bdrv_get_backing_filename(BlockDriverState *bs, |
6457 | char *filename, int filename_size) | |
016f5cf6 | 6458 | { |
384a48fb | 6459 | IO_CODE(); |
61007b31 SH |
6460 | pstrcpy(filename, filename_size, bs->backing_file); |
6461 | } | |
d318aea9 | 6462 | |
3d47eb0a | 6463 | int coroutine_fn bdrv_co_get_info(BlockDriverState *bs, BlockDriverInfo *bdi) |
61007b31 | 6464 | { |
8b117001 | 6465 | int ret; |
61007b31 | 6466 | BlockDriver *drv = bs->drv; |
384a48fb | 6467 | IO_CODE(); |
a00e70c0 EGE |
6468 | assert_bdrv_graph_readable(); |
6469 | ||
5a612c00 MP |
6470 | /* if bs->drv == NULL, bs is closed, so there's nothing to do here */ |
6471 | if (!drv) { | |
61007b31 | 6472 | return -ENOMEDIUM; |
5a612c00 | 6473 | } |
3d47eb0a | 6474 | if (!drv->bdrv_co_get_info) { |
93393e69 HR |
6475 | BlockDriverState *filtered = bdrv_filter_bs(bs); |
6476 | if (filtered) { | |
3d47eb0a | 6477 | return bdrv_co_get_info(filtered, bdi); |
5a612c00 | 6478 | } |
61007b31 | 6479 | return -ENOTSUP; |
5a612c00 | 6480 | } |
61007b31 | 6481 | memset(bdi, 0, sizeof(*bdi)); |
3d47eb0a | 6482 | ret = drv->bdrv_co_get_info(bs, bdi); |
8b117001 VSO |
6483 | if (ret < 0) { |
6484 | return ret; | |
6485 | } | |
6486 | ||
6487 | if (bdi->cluster_size > BDRV_MAX_ALIGNMENT) { | |
6488 | return -EINVAL; | |
6489 | } | |
6490 | ||
6491 | return 0; | |
61007b31 | 6492 | } |
016f5cf6 | 6493 | |
1bf6e9ca AS |
6494 | ImageInfoSpecific *bdrv_get_specific_info(BlockDriverState *bs, |
6495 | Error **errp) | |
61007b31 SH |
6496 | { |
6497 | BlockDriver *drv = bs->drv; | |
384a48fb | 6498 | IO_CODE(); |
61007b31 | 6499 | if (drv && drv->bdrv_get_specific_info) { |
1bf6e9ca | 6500 | return drv->bdrv_get_specific_info(bs, errp); |
61007b31 SH |
6501 | } |
6502 | return NULL; | |
016f5cf6 AG |
6503 | } |
6504 | ||
d9245599 AN |
6505 | BlockStatsSpecific *bdrv_get_specific_stats(BlockDriverState *bs) |
6506 | { | |
6507 | BlockDriver *drv = bs->drv; | |
384a48fb | 6508 | IO_CODE(); |
d9245599 AN |
6509 | if (!drv || !drv->bdrv_get_specific_stats) { |
6510 | return NULL; | |
6511 | } | |
6512 | return drv->bdrv_get_specific_stats(bs); | |
6513 | } | |
6514 | ||
c834dc05 | 6515 | void coroutine_fn bdrv_co_debug_event(BlockDriverState *bs, BlkdebugEvent event) |
4265d620 | 6516 | { |
384a48fb | 6517 | IO_CODE(); |
cb2bfaa4 EGE |
6518 | assert_bdrv_graph_readable(); |
6519 | ||
c834dc05 | 6520 | if (!bs || !bs->drv || !bs->drv->bdrv_co_debug_event) { |
61007b31 SH |
6521 | return; |
6522 | } | |
4265d620 | 6523 | |
c834dc05 | 6524 | bs->drv->bdrv_co_debug_event(bs, event); |
4265d620 PB |
6525 | } |
6526 | ||
d10529a2 | 6527 | static BlockDriverState *bdrv_find_debug_node(BlockDriverState *bs) |
4265d620 | 6528 | { |
bdb73476 | 6529 | GLOBAL_STATE_CODE(); |
61007b31 | 6530 | while (bs && bs->drv && !bs->drv->bdrv_debug_breakpoint) { |
f706a92f | 6531 | bs = bdrv_primary_bs(bs); |
61007b31 | 6532 | } |
4265d620 | 6533 | |
61007b31 | 6534 | if (bs && bs->drv && bs->drv->bdrv_debug_breakpoint) { |
d10529a2 VSO |
6535 | assert(bs->drv->bdrv_debug_remove_breakpoint); |
6536 | return bs; | |
6537 | } | |
6538 | ||
6539 | return NULL; | |
6540 | } | |
6541 | ||
6542 | int bdrv_debug_breakpoint(BlockDriverState *bs, const char *event, | |
6543 | const char *tag) | |
6544 | { | |
f791bf7f | 6545 | GLOBAL_STATE_CODE(); |
d10529a2 VSO |
6546 | bs = bdrv_find_debug_node(bs); |
6547 | if (bs) { | |
61007b31 SH |
6548 | return bs->drv->bdrv_debug_breakpoint(bs, event, tag); |
6549 | } | |
4265d620 | 6550 | |
61007b31 | 6551 | return -ENOTSUP; |
4265d620 PB |
6552 | } |
6553 | ||
61007b31 | 6554 | int bdrv_debug_remove_breakpoint(BlockDriverState *bs, const char *tag) |
ea2384d3 | 6555 | { |
f791bf7f | 6556 | GLOBAL_STATE_CODE(); |
d10529a2 VSO |
6557 | bs = bdrv_find_debug_node(bs); |
6558 | if (bs) { | |
61007b31 SH |
6559 | return bs->drv->bdrv_debug_remove_breakpoint(bs, tag); |
6560 | } | |
6561 | ||
6562 | return -ENOTSUP; | |
eb852011 MA |
6563 | } |
6564 | ||
61007b31 | 6565 | int bdrv_debug_resume(BlockDriverState *bs, const char *tag) |
ce1a14dc | 6566 | { |
f791bf7f | 6567 | GLOBAL_STATE_CODE(); |
61007b31 | 6568 | while (bs && (!bs->drv || !bs->drv->bdrv_debug_resume)) { |
f706a92f | 6569 | bs = bdrv_primary_bs(bs); |
61007b31 | 6570 | } |
ce1a14dc | 6571 | |
61007b31 SH |
6572 | if (bs && bs->drv && bs->drv->bdrv_debug_resume) { |
6573 | return bs->drv->bdrv_debug_resume(bs, tag); | |
6574 | } | |
ce1a14dc | 6575 | |
61007b31 | 6576 | return -ENOTSUP; |
f197fe2b FZ |
6577 | } |
6578 | ||
61007b31 | 6579 | bool bdrv_debug_is_suspended(BlockDriverState *bs, const char *tag) |
ce1a14dc | 6580 | { |
f791bf7f | 6581 | GLOBAL_STATE_CODE(); |
61007b31 | 6582 | while (bs && bs->drv && !bs->drv->bdrv_debug_is_suspended) { |
f706a92f | 6583 | bs = bdrv_primary_bs(bs); |
f197fe2b | 6584 | } |
19cb3738 | 6585 | |
61007b31 SH |
6586 | if (bs && bs->drv && bs->drv->bdrv_debug_is_suspended) { |
6587 | return bs->drv->bdrv_debug_is_suspended(bs, tag); | |
6588 | } | |
f9f05dc5 | 6589 | |
61007b31 SH |
6590 | return false; |
6591 | } | |
f9f05dc5 | 6592 | |
61007b31 SH |
6593 | /* backing_file can either be relative, or absolute, or a protocol. If it is |
6594 | * relative, it must be relative to the chain. So, passing in bs->filename | |
6595 | * from a BDS as backing_file should not be done, as that may be relative to | |
6596 | * the CWD rather than the chain. */ | |
6597 | BlockDriverState *bdrv_find_backing_image(BlockDriverState *bs, | |
6598 | const char *backing_file) | |
f9f05dc5 | 6599 | { |
61007b31 SH |
6600 | char *filename_full = NULL; |
6601 | char *backing_file_full = NULL; | |
6602 | char *filename_tmp = NULL; | |
6603 | int is_protocol = 0; | |
0b877d09 | 6604 | bool filenames_refreshed = false; |
61007b31 SH |
6605 | BlockDriverState *curr_bs = NULL; |
6606 | BlockDriverState *retval = NULL; | |
dcf3f9b2 | 6607 | BlockDriverState *bs_below; |
f9f05dc5 | 6608 | |
f791bf7f EGE |
6609 | GLOBAL_STATE_CODE(); |
6610 | ||
61007b31 SH |
6611 | if (!bs || !bs->drv || !backing_file) { |
6612 | return NULL; | |
f9f05dc5 KW |
6613 | } |
6614 | ||
61007b31 SH |
6615 | filename_full = g_malloc(PATH_MAX); |
6616 | backing_file_full = g_malloc(PATH_MAX); | |
f9f05dc5 | 6617 | |
61007b31 | 6618 | is_protocol = path_has_protocol(backing_file); |
f9f05dc5 | 6619 | |
dcf3f9b2 HR |
6620 | /* |
6621 | * Being largely a legacy function, skip any filters here | |
6622 | * (because filters do not have normal filenames, so they cannot | |
6623 | * match anyway; and allowing json:{} filenames is a bit out of | |
6624 | * scope). | |
6625 | */ | |
6626 | for (curr_bs = bdrv_skip_filters(bs); | |
6627 | bdrv_cow_child(curr_bs) != NULL; | |
6628 | curr_bs = bs_below) | |
6629 | { | |
6630 | bs_below = bdrv_backing_chain_next(curr_bs); | |
f9f05dc5 | 6631 | |
0b877d09 HR |
6632 | if (bdrv_backing_overridden(curr_bs)) { |
6633 | /* | |
6634 | * If the backing file was overridden, we can only compare | |
6635 | * directly against the backing node's filename. | |
6636 | */ | |
6637 | ||
6638 | if (!filenames_refreshed) { | |
6639 | /* | |
6640 | * This will automatically refresh all of the | |
6641 | * filenames in the rest of the backing chain, so we | |
6642 | * only need to do this once. | |
6643 | */ | |
6644 | bdrv_refresh_filename(bs_below); | |
6645 | filenames_refreshed = true; | |
6646 | } | |
6647 | ||
6648 | if (strcmp(backing_file, bs_below->filename) == 0) { | |
6649 | retval = bs_below; | |
6650 | break; | |
6651 | } | |
6652 | } else if (is_protocol || path_has_protocol(curr_bs->backing_file)) { | |
6653 | /* | |
6654 | * If either of the filename paths is actually a protocol, then | |
6655 | * compare unmodified paths; otherwise make paths relative. | |
6656 | */ | |
6b6833c1 HR |
6657 | char *backing_file_full_ret; |
6658 | ||
61007b31 | 6659 | if (strcmp(backing_file, curr_bs->backing_file) == 0) { |
dcf3f9b2 | 6660 | retval = bs_below; |
61007b31 SH |
6661 | break; |
6662 | } | |
418661e0 | 6663 | /* Also check against the full backing filename for the image */ |
6b6833c1 HR |
6664 | backing_file_full_ret = bdrv_get_full_backing_filename(curr_bs, |
6665 | NULL); | |
6666 | if (backing_file_full_ret) { | |
6667 | bool equal = strcmp(backing_file, backing_file_full_ret) == 0; | |
6668 | g_free(backing_file_full_ret); | |
6669 | if (equal) { | |
dcf3f9b2 | 6670 | retval = bs_below; |
418661e0 JC |
6671 | break; |
6672 | } | |
418661e0 | 6673 | } |
61007b31 SH |
6674 | } else { |
6675 | /* If not an absolute filename path, make it relative to the current | |
6676 | * image's filename path */ | |
2d9158ce HR |
6677 | filename_tmp = bdrv_make_absolute_filename(curr_bs, backing_file, |
6678 | NULL); | |
6679 | /* We are going to compare canonicalized absolute pathnames */ | |
6680 | if (!filename_tmp || !realpath(filename_tmp, filename_full)) { | |
6681 | g_free(filename_tmp); | |
61007b31 SH |
6682 | continue; |
6683 | } | |
2d9158ce | 6684 | g_free(filename_tmp); |
07f07615 | 6685 | |
61007b31 SH |
6686 | /* We need to make sure the backing filename we are comparing against |
6687 | * is relative to the current image filename (or absolute) */ | |
2d9158ce HR |
6688 | filename_tmp = bdrv_get_full_backing_filename(curr_bs, NULL); |
6689 | if (!filename_tmp || !realpath(filename_tmp, backing_file_full)) { | |
6690 | g_free(filename_tmp); | |
61007b31 SH |
6691 | continue; |
6692 | } | |
2d9158ce | 6693 | g_free(filename_tmp); |
eb489bb1 | 6694 | |
61007b31 | 6695 | if (strcmp(backing_file_full, filename_full) == 0) { |
dcf3f9b2 | 6696 | retval = bs_below; |
61007b31 SH |
6697 | break; |
6698 | } | |
6699 | } | |
eb489bb1 KW |
6700 | } |
6701 | ||
61007b31 SH |
6702 | g_free(filename_full); |
6703 | g_free(backing_file_full); | |
61007b31 SH |
6704 | return retval; |
6705 | } | |
6706 | ||
61007b31 SH |
6707 | void bdrv_init(void) |
6708 | { | |
e5f05f8c KW |
6709 | #ifdef CONFIG_BDRV_WHITELIST_TOOLS |
6710 | use_bdrv_whitelist = 1; | |
6711 | #endif | |
61007b31 SH |
6712 | module_call_init(MODULE_INIT_BLOCK); |
6713 | } | |
29cdb251 | 6714 | |
61007b31 SH |
6715 | void bdrv_init_with_whitelist(void) |
6716 | { | |
6717 | use_bdrv_whitelist = 1; | |
6718 | bdrv_init(); | |
07f07615 PB |
6719 | } |
6720 | ||
a94750d9 | 6721 | int bdrv_activate(BlockDriverState *bs, Error **errp) |
0f15423c | 6722 | { |
4417ab7a | 6723 | BdrvChild *child, *parent; |
5a8a30db KW |
6724 | Error *local_err = NULL; |
6725 | int ret; | |
9c98f145 | 6726 | BdrvDirtyBitmap *bm; |
5a8a30db | 6727 | |
f791bf7f EGE |
6728 | GLOBAL_STATE_CODE(); |
6729 | ||
3456a8d1 | 6730 | if (!bs->drv) { |
5416645f | 6731 | return -ENOMEDIUM; |
3456a8d1 KW |
6732 | } |
6733 | ||
16e977d5 | 6734 | QLIST_FOREACH(child, &bs->children, next) { |
11d0c9b3 | 6735 | bdrv_activate(child->bs, &local_err); |
0d1c5c91 | 6736 | if (local_err) { |
0d1c5c91 | 6737 | error_propagate(errp, local_err); |
5416645f | 6738 | return -EINVAL; |
0d1c5c91 | 6739 | } |
5a8a30db | 6740 | } |
0d1c5c91 | 6741 | |
dafe0960 KW |
6742 | /* |
6743 | * Update permissions, they may differ for inactive nodes. | |
6744 | * | |
6745 | * Note that the required permissions of inactive images are always a | |
6746 | * subset of the permissions required after activating the image. This | |
6747 | * allows us to just get the permissions upfront without restricting | |
11d0c9b3 | 6748 | * bdrv_co_invalidate_cache(). |
dafe0960 KW |
6749 | * |
6750 | * It also means that in error cases, we don't have to try and revert to | |
6751 | * the old permissions (which is an operation that could fail, too). We can | |
6752 | * just keep the extended permissions for the next time that an activation | |
6753 | * of the image is tried. | |
6754 | */ | |
7bb4941a KW |
6755 | if (bs->open_flags & BDRV_O_INACTIVE) { |
6756 | bs->open_flags &= ~BDRV_O_INACTIVE; | |
f1316edb | 6757 | ret = bdrv_refresh_perms(bs, NULL, errp); |
7bb4941a | 6758 | if (ret < 0) { |
0d1c5c91 | 6759 | bs->open_flags |= BDRV_O_INACTIVE; |
5416645f | 6760 | return ret; |
0d1c5c91 | 6761 | } |
3456a8d1 | 6762 | |
11d0c9b3 EGE |
6763 | ret = bdrv_invalidate_cache(bs, errp); |
6764 | if (ret < 0) { | |
6765 | bs->open_flags |= BDRV_O_INACTIVE; | |
6766 | return ret; | |
7bb4941a | 6767 | } |
9c98f145 | 6768 | |
7bb4941a KW |
6769 | FOR_EACH_DIRTY_BITMAP(bs, bm) { |
6770 | bdrv_dirty_bitmap_skip_store(bm, false); | |
6771 | } | |
6772 | ||
c057960c | 6773 | ret = bdrv_refresh_total_sectors(bs, bs->total_sectors); |
7bb4941a KW |
6774 | if (ret < 0) { |
6775 | bs->open_flags |= BDRV_O_INACTIVE; | |
6776 | error_setg_errno(errp, -ret, "Could not refresh total sector count"); | |
5416645f | 6777 | return ret; |
7bb4941a | 6778 | } |
5a8a30db | 6779 | } |
4417ab7a KW |
6780 | |
6781 | QLIST_FOREACH(parent, &bs->parents, next_parent) { | |
bd86fb99 HR |
6782 | if (parent->klass->activate) { |
6783 | parent->klass->activate(parent, &local_err); | |
4417ab7a | 6784 | if (local_err) { |
78fc3b3a | 6785 | bs->open_flags |= BDRV_O_INACTIVE; |
4417ab7a | 6786 | error_propagate(errp, local_err); |
5416645f | 6787 | return -EINVAL; |
4417ab7a KW |
6788 | } |
6789 | } | |
6790 | } | |
5416645f VSO |
6791 | |
6792 | return 0; | |
0f15423c AL |
6793 | } |
6794 | ||
11d0c9b3 EGE |
6795 | int coroutine_fn bdrv_co_invalidate_cache(BlockDriverState *bs, Error **errp) |
6796 | { | |
6797 | Error *local_err = NULL; | |
1581a70d | 6798 | IO_CODE(); |
11d0c9b3 EGE |
6799 | |
6800 | assert(!(bs->open_flags & BDRV_O_INACTIVE)); | |
1b3ff9fe | 6801 | assert_bdrv_graph_readable(); |
11d0c9b3 EGE |
6802 | |
6803 | if (bs->drv->bdrv_co_invalidate_cache) { | |
6804 | bs->drv->bdrv_co_invalidate_cache(bs, &local_err); | |
6805 | if (local_err) { | |
6806 | error_propagate(errp, local_err); | |
6807 | return -EINVAL; | |
6808 | } | |
6809 | } | |
6810 | ||
6811 | return 0; | |
6812 | } | |
6813 | ||
3b717194 | 6814 | void bdrv_activate_all(Error **errp) |
0f15423c | 6815 | { |
7c8eece4 | 6816 | BlockDriverState *bs; |
88be7b4b | 6817 | BdrvNextIterator it; |
0f15423c | 6818 | |
f791bf7f EGE |
6819 | GLOBAL_STATE_CODE(); |
6820 | ||
88be7b4b | 6821 | for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) { |
ed78cda3 | 6822 | AioContext *aio_context = bdrv_get_aio_context(bs); |
5416645f | 6823 | int ret; |
ed78cda3 SH |
6824 | |
6825 | aio_context_acquire(aio_context); | |
a94750d9 | 6826 | ret = bdrv_activate(bs, errp); |
ed78cda3 | 6827 | aio_context_release(aio_context); |
5416645f | 6828 | if (ret < 0) { |
5e003f17 | 6829 | bdrv_next_cleanup(&it); |
5a8a30db KW |
6830 | return; |
6831 | } | |
0f15423c AL |
6832 | } |
6833 | } | |
6834 | ||
9e37271f KW |
6835 | static bool bdrv_has_bds_parent(BlockDriverState *bs, bool only_active) |
6836 | { | |
6837 | BdrvChild *parent; | |
bdb73476 | 6838 | GLOBAL_STATE_CODE(); |
9e37271f KW |
6839 | |
6840 | QLIST_FOREACH(parent, &bs->parents, next_parent) { | |
bd86fb99 | 6841 | if (parent->klass->parent_is_bds) { |
9e37271f KW |
6842 | BlockDriverState *parent_bs = parent->opaque; |
6843 | if (!only_active || !(parent_bs->open_flags & BDRV_O_INACTIVE)) { | |
6844 | return true; | |
6845 | } | |
6846 | } | |
6847 | } | |
6848 | ||
6849 | return false; | |
6850 | } | |
6851 | ||
6852 | static int bdrv_inactivate_recurse(BlockDriverState *bs) | |
76b1c7fe | 6853 | { |
cfa1a572 | 6854 | BdrvChild *child, *parent; |
76b1c7fe | 6855 | int ret; |
a13de40a | 6856 | uint64_t cumulative_perms, cumulative_shared_perms; |
76b1c7fe | 6857 | |
da359909 EGE |
6858 | GLOBAL_STATE_CODE(); |
6859 | ||
d470ad42 HR |
6860 | if (!bs->drv) { |
6861 | return -ENOMEDIUM; | |
6862 | } | |
6863 | ||
9e37271f KW |
6864 | /* Make sure that we don't inactivate a child before its parent. |
6865 | * It will be covered by recursion from the yet active parent. */ | |
6866 | if (bdrv_has_bds_parent(bs, true)) { | |
6867 | return 0; | |
6868 | } | |
6869 | ||
6870 | assert(!(bs->open_flags & BDRV_O_INACTIVE)); | |
6871 | ||
6872 | /* Inactivate this node */ | |
6873 | if (bs->drv->bdrv_inactivate) { | |
76b1c7fe KW |
6874 | ret = bs->drv->bdrv_inactivate(bs); |
6875 | if (ret < 0) { | |
6876 | return ret; | |
6877 | } | |
6878 | } | |
6879 | ||
9e37271f | 6880 | QLIST_FOREACH(parent, &bs->parents, next_parent) { |
bd86fb99 HR |
6881 | if (parent->klass->inactivate) { |
6882 | ret = parent->klass->inactivate(parent); | |
9e37271f KW |
6883 | if (ret < 0) { |
6884 | return ret; | |
cfa1a572 KW |
6885 | } |
6886 | } | |
9e37271f | 6887 | } |
9c5e6594 | 6888 | |
a13de40a VSO |
6889 | bdrv_get_cumulative_perm(bs, &cumulative_perms, |
6890 | &cumulative_shared_perms); | |
6891 | if (cumulative_perms & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)) { | |
6892 | /* Our inactive parents still need write access. Inactivation failed. */ | |
6893 | return -EPERM; | |
6894 | } | |
6895 | ||
9e37271f | 6896 | bs->open_flags |= BDRV_O_INACTIVE; |
7d5b5261 | 6897 | |
bb87e4d1 VSO |
6898 | /* |
6899 | * Update permissions, they may differ for inactive nodes. | |
6900 | * We only tried to loosen restrictions, so errors are not fatal, ignore | |
6901 | * them. | |
6902 | */ | |
f1316edb | 6903 | bdrv_refresh_perms(bs, NULL, NULL); |
9e37271f KW |
6904 | |
6905 | /* Recursively inactivate children */ | |
38701b6a | 6906 | QLIST_FOREACH(child, &bs->children, next) { |
9e37271f | 6907 | ret = bdrv_inactivate_recurse(child->bs); |
38701b6a KW |
6908 | if (ret < 0) { |
6909 | return ret; | |
6910 | } | |
6911 | } | |
6912 | ||
76b1c7fe KW |
6913 | return 0; |
6914 | } | |
6915 | ||
6916 | int bdrv_inactivate_all(void) | |
6917 | { | |
79720af6 | 6918 | BlockDriverState *bs = NULL; |
88be7b4b | 6919 | BdrvNextIterator it; |
aad0b7a0 | 6920 | int ret = 0; |
bd6458e4 | 6921 | GSList *aio_ctxs = NULL, *ctx; |
76b1c7fe | 6922 | |
f791bf7f EGE |
6923 | GLOBAL_STATE_CODE(); |
6924 | ||
88be7b4b | 6925 | for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) { |
bd6458e4 PB |
6926 | AioContext *aio_context = bdrv_get_aio_context(bs); |
6927 | ||
6928 | if (!g_slist_find(aio_ctxs, aio_context)) { | |
6929 | aio_ctxs = g_slist_prepend(aio_ctxs, aio_context); | |
6930 | aio_context_acquire(aio_context); | |
6931 | } | |
aad0b7a0 | 6932 | } |
76b1c7fe | 6933 | |
9e37271f KW |
6934 | for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) { |
6935 | /* Nodes with BDS parents are covered by recursion from the last | |
6936 | * parent that gets inactivated. Don't inactivate them a second | |
6937 | * time if that has already happened. */ | |
6938 | if (bdrv_has_bds_parent(bs, false)) { | |
6939 | continue; | |
6940 | } | |
6941 | ret = bdrv_inactivate_recurse(bs); | |
6942 | if (ret < 0) { | |
6943 | bdrv_next_cleanup(&it); | |
6944 | goto out; | |
76b1c7fe KW |
6945 | } |
6946 | } | |
6947 | ||
aad0b7a0 | 6948 | out: |
bd6458e4 PB |
6949 | for (ctx = aio_ctxs; ctx != NULL; ctx = ctx->next) { |
6950 | AioContext *aio_context = ctx->data; | |
6951 | aio_context_release(aio_context); | |
aad0b7a0 | 6952 | } |
bd6458e4 | 6953 | g_slist_free(aio_ctxs); |
aad0b7a0 FZ |
6954 | |
6955 | return ret; | |
76b1c7fe KW |
6956 | } |
6957 | ||
19cb3738 FB |
6958 | /**************************************************************/ |
6959 | /* removable device support */ | |
6960 | ||
6961 | /** | |
6962 | * Return TRUE if the media is present | |
6963 | */ | |
1e97be91 | 6964 | bool coroutine_fn bdrv_co_is_inserted(BlockDriverState *bs) |
19cb3738 FB |
6965 | { |
6966 | BlockDriver *drv = bs->drv; | |
28d7a789 | 6967 | BdrvChild *child; |
384a48fb | 6968 | IO_CODE(); |
c73ff92c | 6969 | assert_bdrv_graph_readable(); |
a1aff5bf | 6970 | |
e031f750 HR |
6971 | if (!drv) { |
6972 | return false; | |
6973 | } | |
1e97be91 EGE |
6974 | if (drv->bdrv_co_is_inserted) { |
6975 | return drv->bdrv_co_is_inserted(bs); | |
28d7a789 HR |
6976 | } |
6977 | QLIST_FOREACH(child, &bs->children, next) { | |
1e97be91 | 6978 | if (!bdrv_co_is_inserted(child->bs)) { |
28d7a789 HR |
6979 | return false; |
6980 | } | |
e031f750 | 6981 | } |
28d7a789 | 6982 | return true; |
19cb3738 FB |
6983 | } |
6984 | ||
19cb3738 FB |
6985 | /** |
6986 | * If eject_flag is TRUE, eject the media. Otherwise, close the tray | |
6987 | */ | |
2531b390 | 6988 | void coroutine_fn bdrv_co_eject(BlockDriverState *bs, bool eject_flag) |
19cb3738 FB |
6989 | { |
6990 | BlockDriver *drv = bs->drv; | |
384a48fb | 6991 | IO_CODE(); |
79a292e5 | 6992 | assert_bdrv_graph_readable(); |
19cb3738 | 6993 | |
2531b390 EGE |
6994 | if (drv && drv->bdrv_co_eject) { |
6995 | drv->bdrv_co_eject(bs, eject_flag); | |
19cb3738 FB |
6996 | } |
6997 | } | |
6998 | ||
19cb3738 FB |
6999 | /** |
7000 | * Lock or unlock the media (if it is locked, the user won't be able | |
7001 | * to eject it manually). | |
7002 | */ | |
2c75261c | 7003 | void coroutine_fn bdrv_co_lock_medium(BlockDriverState *bs, bool locked) |
19cb3738 FB |
7004 | { |
7005 | BlockDriver *drv = bs->drv; | |
384a48fb | 7006 | IO_CODE(); |
79a292e5 | 7007 | assert_bdrv_graph_readable(); |
025e849a | 7008 | trace_bdrv_lock_medium(bs, locked); |
b8c6d095 | 7009 | |
2c75261c EGE |
7010 | if (drv && drv->bdrv_co_lock_medium) { |
7011 | drv->bdrv_co_lock_medium(bs, locked); | |
19cb3738 FB |
7012 | } |
7013 | } | |
985a03b0 | 7014 | |
9fcb0251 FZ |
7015 | /* Get a reference to bs */ |
7016 | void bdrv_ref(BlockDriverState *bs) | |
7017 | { | |
f791bf7f | 7018 | GLOBAL_STATE_CODE(); |
9fcb0251 FZ |
7019 | bs->refcnt++; |
7020 | } | |
7021 | ||
7022 | /* Release a previously grabbed reference to bs. | |
7023 | * If after releasing, reference count is zero, the BlockDriverState is | |
7024 | * deleted. */ | |
7025 | void bdrv_unref(BlockDriverState *bs) | |
7026 | { | |
f791bf7f | 7027 | GLOBAL_STATE_CODE(); |
9a4d5ca6 JC |
7028 | if (!bs) { |
7029 | return; | |
7030 | } | |
9fcb0251 FZ |
7031 | assert(bs->refcnt > 0); |
7032 | if (--bs->refcnt == 0) { | |
7033 | bdrv_delete(bs); | |
7034 | } | |
7035 | } | |
7036 | ||
fbe40ff7 FZ |
7037 | struct BdrvOpBlocker { |
7038 | Error *reason; | |
7039 | QLIST_ENTRY(BdrvOpBlocker) list; | |
7040 | }; | |
7041 | ||
7042 | bool bdrv_op_is_blocked(BlockDriverState *bs, BlockOpType op, Error **errp) | |
7043 | { | |
7044 | BdrvOpBlocker *blocker; | |
f791bf7f | 7045 | GLOBAL_STATE_CODE(); |
fbe40ff7 FZ |
7046 | assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX); |
7047 | if (!QLIST_EMPTY(&bs->op_blockers[op])) { | |
7048 | blocker = QLIST_FIRST(&bs->op_blockers[op]); | |
4b576648 MA |
7049 | error_propagate_prepend(errp, error_copy(blocker->reason), |
7050 | "Node '%s' is busy: ", | |
7051 | bdrv_get_device_or_node_name(bs)); | |
fbe40ff7 FZ |
7052 | return true; |
7053 | } | |
7054 | return false; | |
7055 | } | |
7056 | ||
7057 | void bdrv_op_block(BlockDriverState *bs, BlockOpType op, Error *reason) | |
7058 | { | |
7059 | BdrvOpBlocker *blocker; | |
f791bf7f | 7060 | GLOBAL_STATE_CODE(); |
fbe40ff7 FZ |
7061 | assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX); |
7062 | ||
5839e53b | 7063 | blocker = g_new0(BdrvOpBlocker, 1); |
fbe40ff7 FZ |
7064 | blocker->reason = reason; |
7065 | QLIST_INSERT_HEAD(&bs->op_blockers[op], blocker, list); | |
7066 | } | |
7067 | ||
7068 | void bdrv_op_unblock(BlockDriverState *bs, BlockOpType op, Error *reason) | |
7069 | { | |
7070 | BdrvOpBlocker *blocker, *next; | |
f791bf7f | 7071 | GLOBAL_STATE_CODE(); |
fbe40ff7 FZ |
7072 | assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX); |
7073 | QLIST_FOREACH_SAFE(blocker, &bs->op_blockers[op], list, next) { | |
7074 | if (blocker->reason == reason) { | |
7075 | QLIST_REMOVE(blocker, list); | |
7076 | g_free(blocker); | |
7077 | } | |
7078 | } | |
7079 | } | |
7080 | ||
7081 | void bdrv_op_block_all(BlockDriverState *bs, Error *reason) | |
7082 | { | |
7083 | int i; | |
f791bf7f | 7084 | GLOBAL_STATE_CODE(); |
fbe40ff7 FZ |
7085 | for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) { |
7086 | bdrv_op_block(bs, i, reason); | |
7087 | } | |
7088 | } | |
7089 | ||
7090 | void bdrv_op_unblock_all(BlockDriverState *bs, Error *reason) | |
7091 | { | |
7092 | int i; | |
f791bf7f | 7093 | GLOBAL_STATE_CODE(); |
fbe40ff7 FZ |
7094 | for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) { |
7095 | bdrv_op_unblock(bs, i, reason); | |
7096 | } | |
7097 | } | |
7098 | ||
7099 | bool bdrv_op_blocker_is_empty(BlockDriverState *bs) | |
7100 | { | |
7101 | int i; | |
f791bf7f | 7102 | GLOBAL_STATE_CODE(); |
fbe40ff7 FZ |
7103 | for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) { |
7104 | if (!QLIST_EMPTY(&bs->op_blockers[i])) { | |
7105 | return false; | |
7106 | } | |
7107 | } | |
7108 | return true; | |
7109 | } | |
7110 | ||
e1355055 KW |
7111 | /* |
7112 | * Must not be called while holding the lock of an AioContext other than the | |
7113 | * current one. | |
7114 | */ | |
d92ada22 LC |
7115 | void bdrv_img_create(const char *filename, const char *fmt, |
7116 | const char *base_filename, const char *base_fmt, | |
9217283d FZ |
7117 | char *options, uint64_t img_size, int flags, bool quiet, |
7118 | Error **errp) | |
f88e1a42 | 7119 | { |
83d0521a CL |
7120 | QemuOptsList *create_opts = NULL; |
7121 | QemuOpts *opts = NULL; | |
7122 | const char *backing_fmt, *backing_file; | |
7123 | int64_t size; | |
f88e1a42 | 7124 | BlockDriver *drv, *proto_drv; |
cc84d90f | 7125 | Error *local_err = NULL; |
f88e1a42 JS |
7126 | int ret = 0; |
7127 | ||
f791bf7f EGE |
7128 | GLOBAL_STATE_CODE(); |
7129 | ||
f88e1a42 JS |
7130 | /* Find driver and parse its options */ |
7131 | drv = bdrv_find_format(fmt); | |
7132 | if (!drv) { | |
71c79813 | 7133 | error_setg(errp, "Unknown file format '%s'", fmt); |
d92ada22 | 7134 | return; |
f88e1a42 JS |
7135 | } |
7136 | ||
b65a5e12 | 7137 | proto_drv = bdrv_find_protocol(filename, true, errp); |
f88e1a42 | 7138 | if (!proto_drv) { |
d92ada22 | 7139 | return; |
f88e1a42 JS |
7140 | } |
7141 | ||
c6149724 HR |
7142 | if (!drv->create_opts) { |
7143 | error_setg(errp, "Format driver '%s' does not support image creation", | |
7144 | drv->format_name); | |
7145 | return; | |
7146 | } | |
7147 | ||
5a5e7f8c ML |
7148 | if (!proto_drv->create_opts) { |
7149 | error_setg(errp, "Protocol driver '%s' does not support image creation", | |
7150 | proto_drv->format_name); | |
7151 | return; | |
7152 | } | |
7153 | ||
c6e0a6de KW |
7154 | aio_context_acquire(qemu_get_aio_context()); |
7155 | ||
f6dc1c31 | 7156 | /* Create parameter list */ |
c282e1fd | 7157 | create_opts = qemu_opts_append(create_opts, drv->create_opts); |
5a5e7f8c | 7158 | create_opts = qemu_opts_append(create_opts, proto_drv->create_opts); |
f88e1a42 | 7159 | |
83d0521a | 7160 | opts = qemu_opts_create(create_opts, NULL, 0, &error_abort); |
f88e1a42 JS |
7161 | |
7162 | /* Parse -o options */ | |
7163 | if (options) { | |
a5f9b9df | 7164 | if (!qemu_opts_do_parse(opts, options, NULL, errp)) { |
f88e1a42 JS |
7165 | goto out; |
7166 | } | |
7167 | } | |
7168 | ||
f6dc1c31 KW |
7169 | if (!qemu_opt_get(opts, BLOCK_OPT_SIZE)) { |
7170 | qemu_opt_set_number(opts, BLOCK_OPT_SIZE, img_size, &error_abort); | |
7171 | } else if (img_size != UINT64_C(-1)) { | |
7172 | error_setg(errp, "The image size must be specified only once"); | |
7173 | goto out; | |
7174 | } | |
7175 | ||
f88e1a42 | 7176 | if (base_filename) { |
235e59cf | 7177 | if (!qemu_opt_set(opts, BLOCK_OPT_BACKING_FILE, base_filename, |
3882578b | 7178 | NULL)) { |
71c79813 LC |
7179 | error_setg(errp, "Backing file not supported for file format '%s'", |
7180 | fmt); | |
f88e1a42 JS |
7181 | goto out; |
7182 | } | |
7183 | } | |
7184 | ||
7185 | if (base_fmt) { | |
3882578b | 7186 | if (!qemu_opt_set(opts, BLOCK_OPT_BACKING_FMT, base_fmt, NULL)) { |
71c79813 LC |
7187 | error_setg(errp, "Backing file format not supported for file " |
7188 | "format '%s'", fmt); | |
f88e1a42 JS |
7189 | goto out; |
7190 | } | |
7191 | } | |
7192 | ||
83d0521a CL |
7193 | backing_file = qemu_opt_get(opts, BLOCK_OPT_BACKING_FILE); |
7194 | if (backing_file) { | |
7195 | if (!strcmp(filename, backing_file)) { | |
71c79813 LC |
7196 | error_setg(errp, "Error: Trying to create an image with the " |
7197 | "same filename as the backing file"); | |
792da93a JS |
7198 | goto out; |
7199 | } | |
975a7bd2 CK |
7200 | if (backing_file[0] == '\0') { |
7201 | error_setg(errp, "Expected backing file name, got empty string"); | |
7202 | goto out; | |
7203 | } | |
792da93a JS |
7204 | } |
7205 | ||
83d0521a | 7206 | backing_fmt = qemu_opt_get(opts, BLOCK_OPT_BACKING_FMT); |
f88e1a42 | 7207 | |
6e6e55f5 JS |
7208 | /* The size for the image must always be specified, unless we have a backing |
7209 | * file and we have not been forbidden from opening it. */ | |
a8b42a1c | 7210 | size = qemu_opt_get_size(opts, BLOCK_OPT_SIZE, img_size); |
6e6e55f5 JS |
7211 | if (backing_file && !(flags & BDRV_O_NO_BACKING)) { |
7212 | BlockDriverState *bs; | |
645ae7d8 | 7213 | char *full_backing; |
6e6e55f5 JS |
7214 | int back_flags; |
7215 | QDict *backing_options = NULL; | |
7216 | ||
645ae7d8 HR |
7217 | full_backing = |
7218 | bdrv_get_full_backing_filename_from_filename(filename, backing_file, | |
7219 | &local_err); | |
6e6e55f5 | 7220 | if (local_err) { |
6e6e55f5 JS |
7221 | goto out; |
7222 | } | |
645ae7d8 | 7223 | assert(full_backing); |
29168018 | 7224 | |
d5b23994 HR |
7225 | /* |
7226 | * No need to do I/O here, which allows us to open encrypted | |
7227 | * backing images without needing the secret | |
7228 | */ | |
6e6e55f5 JS |
7229 | back_flags = flags; |
7230 | back_flags &= ~(BDRV_O_RDWR | BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING); | |
d5b23994 | 7231 | back_flags |= BDRV_O_NO_IO; |
f88e1a42 | 7232 | |
cc954f01 | 7233 | backing_options = qdict_new(); |
6e6e55f5 | 7234 | if (backing_fmt) { |
6e6e55f5 JS |
7235 | qdict_put_str(backing_options, "driver", backing_fmt); |
7236 | } | |
cc954f01 | 7237 | qdict_put_bool(backing_options, BDRV_OPT_FORCE_SHARE, true); |
e6641719 | 7238 | |
6e6e55f5 JS |
7239 | bs = bdrv_open(full_backing, NULL, backing_options, back_flags, |
7240 | &local_err); | |
7241 | g_free(full_backing); | |
add8200d EB |
7242 | if (!bs) { |
7243 | error_append_hint(&local_err, "Could not open backing image.\n"); | |
6e6e55f5 JS |
7244 | goto out; |
7245 | } else { | |
d9f059aa | 7246 | if (!backing_fmt) { |
497a30db EB |
7247 | error_setg(&local_err, |
7248 | "Backing file specified without backing format"); | |
fbdffb08 | 7249 | error_append_hint(&local_err, "Detected format of %s.\n", |
497a30db EB |
7250 | bs->drv->format_name); |
7251 | goto out; | |
d9f059aa | 7252 | } |
6e6e55f5 JS |
7253 | if (size == -1) { |
7254 | /* Opened BS, have no size */ | |
7255 | size = bdrv_getlength(bs); | |
7256 | if (size < 0) { | |
7257 | error_setg_errno(errp, -size, "Could not get size of '%s'", | |
7258 | backing_file); | |
7259 | bdrv_unref(bs); | |
7260 | goto out; | |
7261 | } | |
7262 | qemu_opt_set_number(opts, BLOCK_OPT_SIZE, size, &error_abort); | |
52bf1e72 | 7263 | } |
66f6b814 | 7264 | bdrv_unref(bs); |
f88e1a42 | 7265 | } |
d9f059aa EB |
7266 | /* (backing_file && !(flags & BDRV_O_NO_BACKING)) */ |
7267 | } else if (backing_file && !backing_fmt) { | |
497a30db EB |
7268 | error_setg(&local_err, |
7269 | "Backing file specified without backing format"); | |
7270 | goto out; | |
d9f059aa | 7271 | } |
6e6e55f5 JS |
7272 | |
7273 | if (size == -1) { | |
7274 | error_setg(errp, "Image creation needs a size parameter"); | |
7275 | goto out; | |
f88e1a42 JS |
7276 | } |
7277 | ||
f382d43a | 7278 | if (!quiet) { |
fe646693 | 7279 | printf("Formatting '%s', fmt=%s ", filename, fmt); |
43c5d8f8 | 7280 | qemu_opts_print(opts, " "); |
f382d43a | 7281 | puts(""); |
4e2f4418 | 7282 | fflush(stdout); |
f382d43a | 7283 | } |
83d0521a | 7284 | |
c282e1fd | 7285 | ret = bdrv_create(drv, filename, opts, &local_err); |
83d0521a | 7286 | |
cc84d90f HR |
7287 | if (ret == -EFBIG) { |
7288 | /* This is generally a better message than whatever the driver would | |
7289 | * deliver (especially because of the cluster_size_hint), since that | |
7290 | * is most probably not much different from "image too large". */ | |
7291 | const char *cluster_size_hint = ""; | |
83d0521a | 7292 | if (qemu_opt_get_size(opts, BLOCK_OPT_CLUSTER_SIZE, 0)) { |
cc84d90f | 7293 | cluster_size_hint = " (try using a larger cluster size)"; |
f88e1a42 | 7294 | } |
cc84d90f HR |
7295 | error_setg(errp, "The image size is too large for file format '%s'" |
7296 | "%s", fmt, cluster_size_hint); | |
7297 | error_free(local_err); | |
7298 | local_err = NULL; | |
f88e1a42 JS |
7299 | } |
7300 | ||
7301 | out: | |
83d0521a CL |
7302 | qemu_opts_del(opts); |
7303 | qemu_opts_free(create_opts); | |
621ff94d | 7304 | error_propagate(errp, local_err); |
c6e0a6de | 7305 | aio_context_release(qemu_get_aio_context()); |
f88e1a42 | 7306 | } |
85d126f3 SH |
7307 | |
7308 | AioContext *bdrv_get_aio_context(BlockDriverState *bs) | |
7309 | { | |
384a48fb | 7310 | IO_CODE(); |
33f2a757 | 7311 | return bs ? bs->aio_context : qemu_get_aio_context(); |
dcd04228 SH |
7312 | } |
7313 | ||
e336fd4c KW |
7314 | AioContext *coroutine_fn bdrv_co_enter(BlockDriverState *bs) |
7315 | { | |
7316 | Coroutine *self = qemu_coroutine_self(); | |
7317 | AioContext *old_ctx = qemu_coroutine_get_aio_context(self); | |
7318 | AioContext *new_ctx; | |
384a48fb | 7319 | IO_CODE(); |
e336fd4c KW |
7320 | |
7321 | /* | |
7322 | * Increase bs->in_flight to ensure that this operation is completed before | |
7323 | * moving the node to a different AioContext. Read new_ctx only afterwards. | |
7324 | */ | |
7325 | bdrv_inc_in_flight(bs); | |
7326 | ||
7327 | new_ctx = bdrv_get_aio_context(bs); | |
7328 | aio_co_reschedule_self(new_ctx); | |
7329 | return old_ctx; | |
7330 | } | |
7331 | ||
7332 | void coroutine_fn bdrv_co_leave(BlockDriverState *bs, AioContext *old_ctx) | |
7333 | { | |
384a48fb | 7334 | IO_CODE(); |
e336fd4c KW |
7335 | aio_co_reschedule_self(old_ctx); |
7336 | bdrv_dec_in_flight(bs); | |
7337 | } | |
7338 | ||
18c6ac1c KW |
7339 | void coroutine_fn bdrv_co_lock(BlockDriverState *bs) |
7340 | { | |
7341 | AioContext *ctx = bdrv_get_aio_context(bs); | |
7342 | ||
7343 | /* In the main thread, bs->aio_context won't change concurrently */ | |
7344 | assert(qemu_get_current_aio_context() == qemu_get_aio_context()); | |
7345 | ||
7346 | /* | |
7347 | * We're in coroutine context, so we already hold the lock of the main | |
7348 | * loop AioContext. Don't lock it twice to avoid deadlocks. | |
7349 | */ | |
7350 | assert(qemu_in_coroutine()); | |
7351 | if (ctx != qemu_get_aio_context()) { | |
7352 | aio_context_acquire(ctx); | |
7353 | } | |
7354 | } | |
7355 | ||
7356 | void coroutine_fn bdrv_co_unlock(BlockDriverState *bs) | |
7357 | { | |
7358 | AioContext *ctx = bdrv_get_aio_context(bs); | |
7359 | ||
7360 | assert(qemu_in_coroutine()); | |
7361 | if (ctx != qemu_get_aio_context()) { | |
7362 | aio_context_release(ctx); | |
7363 | } | |
7364 | } | |
7365 | ||
e8a095da SH |
7366 | static void bdrv_do_remove_aio_context_notifier(BdrvAioNotifier *ban) |
7367 | { | |
bdb73476 | 7368 | GLOBAL_STATE_CODE(); |
e8a095da SH |
7369 | QLIST_REMOVE(ban, list); |
7370 | g_free(ban); | |
7371 | } | |
7372 | ||
a3a683c3 | 7373 | static void bdrv_detach_aio_context(BlockDriverState *bs) |
dcd04228 | 7374 | { |
e8a095da | 7375 | BdrvAioNotifier *baf, *baf_tmp; |
33384421 | 7376 | |
e8a095da | 7377 | assert(!bs->walking_aio_notifiers); |
da359909 | 7378 | GLOBAL_STATE_CODE(); |
e8a095da SH |
7379 | bs->walking_aio_notifiers = true; |
7380 | QLIST_FOREACH_SAFE(baf, &bs->aio_notifiers, list, baf_tmp) { | |
7381 | if (baf->deleted) { | |
7382 | bdrv_do_remove_aio_context_notifier(baf); | |
7383 | } else { | |
7384 | baf->detach_aio_context(baf->opaque); | |
7385 | } | |
33384421 | 7386 | } |
e8a095da SH |
7387 | /* Never mind iterating again to check for ->deleted. bdrv_close() will |
7388 | * remove remaining aio notifiers if we aren't called again. | |
7389 | */ | |
7390 | bs->walking_aio_notifiers = false; | |
33384421 | 7391 | |
1bffe1ae | 7392 | if (bs->drv && bs->drv->bdrv_detach_aio_context) { |
dcd04228 SH |
7393 | bs->drv->bdrv_detach_aio_context(bs); |
7394 | } | |
dcd04228 SH |
7395 | |
7396 | bs->aio_context = NULL; | |
7397 | } | |
7398 | ||
a3a683c3 KW |
7399 | static void bdrv_attach_aio_context(BlockDriverState *bs, |
7400 | AioContext *new_context) | |
dcd04228 | 7401 | { |
e8a095da | 7402 | BdrvAioNotifier *ban, *ban_tmp; |
da359909 | 7403 | GLOBAL_STATE_CODE(); |
33384421 | 7404 | |
dcd04228 SH |
7405 | bs->aio_context = new_context; |
7406 | ||
1bffe1ae | 7407 | if (bs->drv && bs->drv->bdrv_attach_aio_context) { |
dcd04228 SH |
7408 | bs->drv->bdrv_attach_aio_context(bs, new_context); |
7409 | } | |
33384421 | 7410 | |
e8a095da SH |
7411 | assert(!bs->walking_aio_notifiers); |
7412 | bs->walking_aio_notifiers = true; | |
7413 | QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_tmp) { | |
7414 | if (ban->deleted) { | |
7415 | bdrv_do_remove_aio_context_notifier(ban); | |
7416 | } else { | |
7417 | ban->attached_aio_context(new_context, ban->opaque); | |
7418 | } | |
33384421 | 7419 | } |
e8a095da | 7420 | bs->walking_aio_notifiers = false; |
dcd04228 SH |
7421 | } |
7422 | ||
7e8c182f EGE |
7423 | typedef struct BdrvStateSetAioContext { |
7424 | AioContext *new_ctx; | |
7425 | BlockDriverState *bs; | |
7426 | } BdrvStateSetAioContext; | |
d616b224 | 7427 | |
7e8c182f | 7428 | static bool bdrv_parent_change_aio_context(BdrvChild *c, AioContext *ctx, |
e08cc001 EGE |
7429 | GHashTable *visited, |
7430 | Transaction *tran, | |
7e8c182f | 7431 | Error **errp) |
5d231849 | 7432 | { |
f0c28327 | 7433 | GLOBAL_STATE_CODE(); |
e08cc001 | 7434 | if (g_hash_table_contains(visited, c)) { |
5d231849 KW |
7435 | return true; |
7436 | } | |
e08cc001 | 7437 | g_hash_table_add(visited, c); |
5d231849 | 7438 | |
bd86fb99 HR |
7439 | /* |
7440 | * A BdrvChildClass that doesn't handle AioContext changes cannot | |
7441 | * tolerate any AioContext changes | |
7442 | */ | |
7e8c182f | 7443 | if (!c->klass->change_aio_ctx) { |
5d231849 KW |
7444 | char *user = bdrv_child_user_desc(c); |
7445 | error_setg(errp, "Changing iothreads is not supported by %s", user); | |
7446 | g_free(user); | |
7447 | return false; | |
7448 | } | |
7e8c182f | 7449 | if (!c->klass->change_aio_ctx(c, ctx, visited, tran, errp)) { |
5d231849 KW |
7450 | assert(!errp || *errp); |
7451 | return false; | |
7452 | } | |
7453 | return true; | |
7454 | } | |
7455 | ||
7e8c182f | 7456 | bool bdrv_child_change_aio_context(BdrvChild *c, AioContext *ctx, |
e08cc001 | 7457 | GHashTable *visited, Transaction *tran, |
7e8c182f | 7458 | Error **errp) |
5d231849 | 7459 | { |
f791bf7f | 7460 | GLOBAL_STATE_CODE(); |
e08cc001 | 7461 | if (g_hash_table_contains(visited, c)) { |
5d231849 KW |
7462 | return true; |
7463 | } | |
e08cc001 | 7464 | g_hash_table_add(visited, c); |
7e8c182f | 7465 | return bdrv_change_aio_context(c->bs, ctx, visited, tran, errp); |
5d231849 KW |
7466 | } |
7467 | ||
7e8c182f EGE |
7468 | static void bdrv_set_aio_context_clean(void *opaque) |
7469 | { | |
7470 | BdrvStateSetAioContext *state = (BdrvStateSetAioContext *) opaque; | |
7471 | BlockDriverState *bs = (BlockDriverState *) state->bs; | |
7472 | ||
7473 | /* Paired with bdrv_drained_begin in bdrv_change_aio_context() */ | |
7474 | bdrv_drained_end(bs); | |
7475 | ||
7476 | g_free(state); | |
7477 | } | |
7478 | ||
7479 | static void bdrv_set_aio_context_commit(void *opaque) | |
7480 | { | |
7481 | BdrvStateSetAioContext *state = (BdrvStateSetAioContext *) opaque; | |
7482 | BlockDriverState *bs = (BlockDriverState *) state->bs; | |
7483 | AioContext *new_context = state->new_ctx; | |
7484 | AioContext *old_context = bdrv_get_aio_context(bs); | |
7e8c182f EGE |
7485 | |
7486 | /* | |
7487 | * Take the old AioContex when detaching it from bs. | |
7488 | * At this point, new_context lock is already acquired, and we are now | |
7489 | * also taking old_context. This is safe as long as bdrv_detach_aio_context | |
7490 | * does not call AIO_POLL_WHILE(). | |
7491 | */ | |
7492 | if (old_context != qemu_get_aio_context()) { | |
7493 | aio_context_acquire(old_context); | |
7494 | } | |
7495 | bdrv_detach_aio_context(bs); | |
7496 | if (old_context != qemu_get_aio_context()) { | |
7497 | aio_context_release(old_context); | |
7498 | } | |
7499 | bdrv_attach_aio_context(bs, new_context); | |
7500 | } | |
7501 | ||
7502 | static TransactionActionDrv set_aio_context = { | |
7503 | .commit = bdrv_set_aio_context_commit, | |
7504 | .clean = bdrv_set_aio_context_clean, | |
7505 | }; | |
7506 | ||
7507 | /* | |
7508 | * Changes the AioContext used for fd handlers, timers, and BHs by this | |
7509 | * BlockDriverState and all its children and parents. | |
7510 | * | |
7511 | * Must be called from the main AioContext. | |
7512 | * | |
7513 | * The caller must own the AioContext lock for the old AioContext of bs, but it | |
7514 | * must not own the AioContext lock for new_context (unless new_context is the | |
7515 | * same as the current context of bs). | |
7516 | * | |
7517 | * @visited will accumulate all visited BdrvChild objects. The caller is | |
7518 | * responsible for freeing the list afterwards. | |
7519 | */ | |
7520 | static bool bdrv_change_aio_context(BlockDriverState *bs, AioContext *ctx, | |
e08cc001 | 7521 | GHashTable *visited, Transaction *tran, |
7e8c182f | 7522 | Error **errp) |
5d231849 KW |
7523 | { |
7524 | BdrvChild *c; | |
7e8c182f EGE |
7525 | BdrvStateSetAioContext *state; |
7526 | ||
7527 | GLOBAL_STATE_CODE(); | |
5d231849 KW |
7528 | |
7529 | if (bdrv_get_aio_context(bs) == ctx) { | |
7530 | return true; | |
7531 | } | |
7532 | ||
7533 | QLIST_FOREACH(c, &bs->parents, next_parent) { | |
7e8c182f | 7534 | if (!bdrv_parent_change_aio_context(c, ctx, visited, tran, errp)) { |
5d231849 KW |
7535 | return false; |
7536 | } | |
7537 | } | |
7e8c182f | 7538 | |
5d231849 | 7539 | QLIST_FOREACH(c, &bs->children, next) { |
7e8c182f | 7540 | if (!bdrv_child_change_aio_context(c, ctx, visited, tran, errp)) { |
5d231849 KW |
7541 | return false; |
7542 | } | |
7543 | } | |
7544 | ||
7e8c182f EGE |
7545 | state = g_new(BdrvStateSetAioContext, 1); |
7546 | *state = (BdrvStateSetAioContext) { | |
7547 | .new_ctx = ctx, | |
7548 | .bs = bs, | |
7549 | }; | |
7550 | ||
7551 | /* Paired with bdrv_drained_end in bdrv_set_aio_context_clean() */ | |
7552 | bdrv_drained_begin(bs); | |
7553 | ||
7554 | tran_add(tran, &set_aio_context, state); | |
7555 | ||
5d231849 KW |
7556 | return true; |
7557 | } | |
7558 | ||
7e8c182f EGE |
7559 | /* |
7560 | * Change bs's and recursively all of its parents' and children's AioContext | |
7561 | * to the given new context, returning an error if that isn't possible. | |
7562 | * | |
7563 | * If ignore_child is not NULL, that child (and its subgraph) will not | |
7564 | * be touched. | |
7565 | * | |
7566 | * This function still requires the caller to take the bs current | |
7567 | * AioContext lock, otherwise draining will fail since AIO_WAIT_WHILE | |
7568 | * assumes the lock is always held if bs is in another AioContext. | |
7569 | * For the same reason, it temporarily also holds the new AioContext, since | |
7570 | * bdrv_drained_end calls BDRV_POLL_WHILE that assumes the lock is taken too. | |
7571 | * Therefore the new AioContext lock must not be taken by the caller. | |
7572 | */ | |
a41cfda1 EGE |
7573 | int bdrv_try_change_aio_context(BlockDriverState *bs, AioContext *ctx, |
7574 | BdrvChild *ignore_child, Error **errp) | |
5d231849 | 7575 | { |
7e8c182f | 7576 | Transaction *tran; |
e08cc001 | 7577 | GHashTable *visited; |
7e8c182f EGE |
7578 | int ret; |
7579 | AioContext *old_context = bdrv_get_aio_context(bs); | |
f791bf7f EGE |
7580 | GLOBAL_STATE_CODE(); |
7581 | ||
7e8c182f EGE |
7582 | /* |
7583 | * Recursion phase: go through all nodes of the graph. | |
7584 | * Take care of checking that all nodes support changing AioContext | |
7585 | * and drain them, builing a linear list of callbacks to run if everything | |
7586 | * is successful (the transaction itself). | |
7587 | */ | |
7588 | tran = tran_new(); | |
e08cc001 EGE |
7589 | visited = g_hash_table_new(NULL, NULL); |
7590 | if (ignore_child) { | |
7591 | g_hash_table_add(visited, ignore_child); | |
7592 | } | |
7593 | ret = bdrv_change_aio_context(bs, ctx, visited, tran, errp); | |
7594 | g_hash_table_destroy(visited); | |
7e8c182f EGE |
7595 | |
7596 | /* | |
7597 | * Linear phase: go through all callbacks collected in the transaction. | |
7598 | * Run all callbacks collected in the recursion to switch all nodes | |
7599 | * AioContext lock (transaction commit), or undo all changes done in the | |
7600 | * recursion (transaction abort). | |
7601 | */ | |
5d231849 KW |
7602 | |
7603 | if (!ret) { | |
7e8c182f EGE |
7604 | /* Just run clean() callbacks. No AioContext changed. */ |
7605 | tran_abort(tran); | |
5d231849 KW |
7606 | return -EPERM; |
7607 | } | |
7608 | ||
7e8c182f EGE |
7609 | /* |
7610 | * Release old AioContext, it won't be needed anymore, as all | |
7611 | * bdrv_drained_begin() have been called already. | |
7612 | */ | |
7613 | if (qemu_get_aio_context() != old_context) { | |
7614 | aio_context_release(old_context); | |
7615 | } | |
7616 | ||
7617 | /* | |
7618 | * Acquire new AioContext since bdrv_drained_end() is going to be called | |
7619 | * after we switched all nodes in the new AioContext, and the function | |
7620 | * assumes that the lock of the bs is always taken. | |
7621 | */ | |
7622 | if (qemu_get_aio_context() != ctx) { | |
7623 | aio_context_acquire(ctx); | |
7624 | } | |
53a7d041 | 7625 | |
7e8c182f | 7626 | tran_commit(tran); |
5d231849 | 7627 | |
7e8c182f EGE |
7628 | if (qemu_get_aio_context() != ctx) { |
7629 | aio_context_release(ctx); | |
7630 | } | |
7631 | ||
7632 | /* Re-acquire the old AioContext, since the caller takes and releases it. */ | |
7633 | if (qemu_get_aio_context() != old_context) { | |
7634 | aio_context_acquire(old_context); | |
7635 | } | |
7636 | ||
7637 | return 0; | |
5d231849 KW |
7638 | } |
7639 | ||
33384421 HR |
7640 | void bdrv_add_aio_context_notifier(BlockDriverState *bs, |
7641 | void (*attached_aio_context)(AioContext *new_context, void *opaque), | |
7642 | void (*detach_aio_context)(void *opaque), void *opaque) | |
7643 | { | |
7644 | BdrvAioNotifier *ban = g_new(BdrvAioNotifier, 1); | |
7645 | *ban = (BdrvAioNotifier){ | |
7646 | .attached_aio_context = attached_aio_context, | |
7647 | .detach_aio_context = detach_aio_context, | |
7648 | .opaque = opaque | |
7649 | }; | |
f791bf7f | 7650 | GLOBAL_STATE_CODE(); |
33384421 HR |
7651 | |
7652 | QLIST_INSERT_HEAD(&bs->aio_notifiers, ban, list); | |
7653 | } | |
7654 | ||
7655 | void bdrv_remove_aio_context_notifier(BlockDriverState *bs, | |
7656 | void (*attached_aio_context)(AioContext *, | |
7657 | void *), | |
7658 | void (*detach_aio_context)(void *), | |
7659 | void *opaque) | |
7660 | { | |
7661 | BdrvAioNotifier *ban, *ban_next; | |
f791bf7f | 7662 | GLOBAL_STATE_CODE(); |
33384421 HR |
7663 | |
7664 | QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_next) { | |
7665 | if (ban->attached_aio_context == attached_aio_context && | |
7666 | ban->detach_aio_context == detach_aio_context && | |
e8a095da SH |
7667 | ban->opaque == opaque && |
7668 | ban->deleted == false) | |
33384421 | 7669 | { |
e8a095da SH |
7670 | if (bs->walking_aio_notifiers) { |
7671 | ban->deleted = true; | |
7672 | } else { | |
7673 | bdrv_do_remove_aio_context_notifier(ban); | |
7674 | } | |
33384421 HR |
7675 | return; |
7676 | } | |
7677 | } | |
7678 | ||
7679 | abort(); | |
7680 | } | |
7681 | ||
77485434 | 7682 | int bdrv_amend_options(BlockDriverState *bs, QemuOpts *opts, |
d1402b50 | 7683 | BlockDriverAmendStatusCB *status_cb, void *cb_opaque, |
a3579bfa | 7684 | bool force, |
d1402b50 | 7685 | Error **errp) |
6f176b48 | 7686 | { |
f791bf7f | 7687 | GLOBAL_STATE_CODE(); |
d470ad42 | 7688 | if (!bs->drv) { |
d1402b50 | 7689 | error_setg(errp, "Node is ejected"); |
d470ad42 HR |
7690 | return -ENOMEDIUM; |
7691 | } | |
c282e1fd | 7692 | if (!bs->drv->bdrv_amend_options) { |
d1402b50 HR |
7693 | error_setg(errp, "Block driver '%s' does not support option amendment", |
7694 | bs->drv->format_name); | |
6f176b48 HR |
7695 | return -ENOTSUP; |
7696 | } | |
a3579bfa ML |
7697 | return bs->drv->bdrv_amend_options(bs, opts, status_cb, |
7698 | cb_opaque, force, errp); | |
6f176b48 | 7699 | } |
f6186f49 | 7700 | |
5d69b5ab HR |
7701 | /* |
7702 | * This function checks whether the given @to_replace is allowed to be | |
7703 | * replaced by a node that always shows the same data as @bs. This is | |
7704 | * used for example to verify whether the mirror job can replace | |
7705 | * @to_replace by the target mirrored from @bs. | |
7706 | * To be replaceable, @bs and @to_replace may either be guaranteed to | |
7707 | * always show the same data (because they are only connected through | |
7708 | * filters), or some driver may allow replacing one of its children | |
7709 | * because it can guarantee that this child's data is not visible at | |
7710 | * all (for example, for dissenting quorum children that have no other | |
7711 | * parents). | |
7712 | */ | |
7713 | bool bdrv_recurse_can_replace(BlockDriverState *bs, | |
7714 | BlockDriverState *to_replace) | |
7715 | { | |
93393e69 HR |
7716 | BlockDriverState *filtered; |
7717 | ||
b4ad82aa EGE |
7718 | GLOBAL_STATE_CODE(); |
7719 | ||
5d69b5ab HR |
7720 | if (!bs || !bs->drv) { |
7721 | return false; | |
7722 | } | |
7723 | ||
7724 | if (bs == to_replace) { | |
7725 | return true; | |
7726 | } | |
7727 | ||
7728 | /* See what the driver can do */ | |
7729 | if (bs->drv->bdrv_recurse_can_replace) { | |
7730 | return bs->drv->bdrv_recurse_can_replace(bs, to_replace); | |
7731 | } | |
7732 | ||
7733 | /* For filters without an own implementation, we can recurse on our own */ | |
93393e69 HR |
7734 | filtered = bdrv_filter_bs(bs); |
7735 | if (filtered) { | |
7736 | return bdrv_recurse_can_replace(filtered, to_replace); | |
5d69b5ab HR |
7737 | } |
7738 | ||
7739 | /* Safe default */ | |
7740 | return false; | |
7741 | } | |
7742 | ||
810803a8 HR |
7743 | /* |
7744 | * Check whether the given @node_name can be replaced by a node that | |
7745 | * has the same data as @parent_bs. If so, return @node_name's BDS; | |
7746 | * NULL otherwise. | |
7747 | * | |
7748 | * @node_name must be a (recursive) *child of @parent_bs (or this | |
7749 | * function will return NULL). | |
7750 | * | |
7751 | * The result (whether the node can be replaced or not) is only valid | |
7752 | * for as long as no graph or permission changes occur. | |
7753 | */ | |
e12f3784 WC |
7754 | BlockDriverState *check_to_replace_node(BlockDriverState *parent_bs, |
7755 | const char *node_name, Error **errp) | |
09158f00 BC |
7756 | { |
7757 | BlockDriverState *to_replace_bs = bdrv_find_node(node_name); | |
5a7e7a0b SH |
7758 | AioContext *aio_context; |
7759 | ||
f791bf7f EGE |
7760 | GLOBAL_STATE_CODE(); |
7761 | ||
09158f00 | 7762 | if (!to_replace_bs) { |
785ec4b1 | 7763 | error_setg(errp, "Failed to find node with node-name='%s'", node_name); |
09158f00 BC |
7764 | return NULL; |
7765 | } | |
7766 | ||
5a7e7a0b SH |
7767 | aio_context = bdrv_get_aio_context(to_replace_bs); |
7768 | aio_context_acquire(aio_context); | |
7769 | ||
09158f00 | 7770 | if (bdrv_op_is_blocked(to_replace_bs, BLOCK_OP_TYPE_REPLACE, errp)) { |
5a7e7a0b SH |
7771 | to_replace_bs = NULL; |
7772 | goto out; | |
09158f00 BC |
7773 | } |
7774 | ||
7775 | /* We don't want arbitrary node of the BDS chain to be replaced only the top | |
7776 | * most non filter in order to prevent data corruption. | |
7777 | * Another benefit is that this tests exclude backing files which are | |
7778 | * blocked by the backing blockers. | |
7779 | */ | |
810803a8 HR |
7780 | if (!bdrv_recurse_can_replace(parent_bs, to_replace_bs)) { |
7781 | error_setg(errp, "Cannot replace '%s' by a node mirrored from '%s', " | |
7782 | "because it cannot be guaranteed that doing so would not " | |
7783 | "lead to an abrupt change of visible data", | |
7784 | node_name, parent_bs->node_name); | |
5a7e7a0b SH |
7785 | to_replace_bs = NULL; |
7786 | goto out; | |
09158f00 BC |
7787 | } |
7788 | ||
5a7e7a0b SH |
7789 | out: |
7790 | aio_context_release(aio_context); | |
09158f00 BC |
7791 | return to_replace_bs; |
7792 | } | |
448ad91d | 7793 | |
97e2f021 HR |
7794 | /** |
7795 | * Iterates through the list of runtime option keys that are said to | |
7796 | * be "strong" for a BDS. An option is called "strong" if it changes | |
7797 | * a BDS's data. For example, the null block driver's "size" and | |
7798 | * "read-zeroes" options are strong, but its "latency-ns" option is | |
7799 | * not. | |
7800 | * | |
7801 | * If a key returned by this function ends with a dot, all options | |
7802 | * starting with that prefix are strong. | |
7803 | */ | |
7804 | static const char *const *strong_options(BlockDriverState *bs, | |
7805 | const char *const *curopt) | |
7806 | { | |
7807 | static const char *const global_options[] = { | |
7808 | "driver", "filename", NULL | |
7809 | }; | |
7810 | ||
7811 | if (!curopt) { | |
7812 | return &global_options[0]; | |
7813 | } | |
7814 | ||
7815 | curopt++; | |
7816 | if (curopt == &global_options[ARRAY_SIZE(global_options) - 1] && bs->drv) { | |
7817 | curopt = bs->drv->strong_runtime_opts; | |
7818 | } | |
7819 | ||
7820 | return (curopt && *curopt) ? curopt : NULL; | |
7821 | } | |
7822 | ||
7823 | /** | |
7824 | * Copies all strong runtime options from bs->options to the given | |
7825 | * QDict. The set of strong option keys is determined by invoking | |
7826 | * strong_options(). | |
7827 | * | |
7828 | * Returns true iff any strong option was present in bs->options (and | |
7829 | * thus copied to the target QDict) with the exception of "filename" | |
7830 | * and "driver". The caller is expected to use this value to decide | |
7831 | * whether the existence of strong options prevents the generation of | |
7832 | * a plain filename. | |
7833 | */ | |
7834 | static bool append_strong_runtime_options(QDict *d, BlockDriverState *bs) | |
7835 | { | |
7836 | bool found_any = false; | |
7837 | const char *const *option_name = NULL; | |
7838 | ||
7839 | if (!bs->drv) { | |
7840 | return false; | |
7841 | } | |
7842 | ||
7843 | while ((option_name = strong_options(bs, option_name))) { | |
7844 | bool option_given = false; | |
7845 | ||
7846 | assert(strlen(*option_name) > 0); | |
7847 | if ((*option_name)[strlen(*option_name) - 1] != '.') { | |
7848 | QObject *entry = qdict_get(bs->options, *option_name); | |
7849 | if (!entry) { | |
7850 | continue; | |
7851 | } | |
7852 | ||
7853 | qdict_put_obj(d, *option_name, qobject_ref(entry)); | |
7854 | option_given = true; | |
7855 | } else { | |
7856 | const QDictEntry *entry; | |
7857 | for (entry = qdict_first(bs->options); entry; | |
7858 | entry = qdict_next(bs->options, entry)) | |
7859 | { | |
7860 | if (strstart(qdict_entry_key(entry), *option_name, NULL)) { | |
7861 | qdict_put_obj(d, qdict_entry_key(entry), | |
7862 | qobject_ref(qdict_entry_value(entry))); | |
7863 | option_given = true; | |
7864 | } | |
7865 | } | |
7866 | } | |
7867 | ||
7868 | /* While "driver" and "filename" need to be included in a JSON filename, | |
7869 | * their existence does not prohibit generation of a plain filename. */ | |
7870 | if (!found_any && option_given && | |
7871 | strcmp(*option_name, "driver") && strcmp(*option_name, "filename")) | |
7872 | { | |
7873 | found_any = true; | |
7874 | } | |
7875 | } | |
7876 | ||
62a01a27 HR |
7877 | if (!qdict_haskey(d, "driver")) { |
7878 | /* Drivers created with bdrv_new_open_driver() may not have a | |
7879 | * @driver option. Add it here. */ | |
7880 | qdict_put_str(d, "driver", bs->drv->format_name); | |
7881 | } | |
7882 | ||
97e2f021 HR |
7883 | return found_any; |
7884 | } | |
7885 | ||
90993623 HR |
7886 | /* Note: This function may return false positives; it may return true |
7887 | * even if opening the backing file specified by bs's image header | |
7888 | * would result in exactly bs->backing. */ | |
fa8fc1d0 | 7889 | static bool bdrv_backing_overridden(BlockDriverState *bs) |
90993623 | 7890 | { |
b4ad82aa | 7891 | GLOBAL_STATE_CODE(); |
90993623 HR |
7892 | if (bs->backing) { |
7893 | return strcmp(bs->auto_backing_file, | |
7894 | bs->backing->bs->filename); | |
7895 | } else { | |
7896 | /* No backing BDS, so if the image header reports any backing | |
7897 | * file, it must have been suppressed */ | |
7898 | return bs->auto_backing_file[0] != '\0'; | |
7899 | } | |
7900 | } | |
7901 | ||
91af7014 HR |
7902 | /* Updates the following BDS fields: |
7903 | * - exact_filename: A filename which may be used for opening a block device | |
7904 | * which (mostly) equals the given BDS (even without any | |
7905 | * other options; so reading and writing must return the same | |
7906 | * results, but caching etc. may be different) | |
7907 | * - full_open_options: Options which, when given when opening a block device | |
7908 | * (without a filename), result in a BDS (mostly) | |
7909 | * equalling the given one | |
7910 | * - filename: If exact_filename is set, it is copied here. Otherwise, | |
7911 | * full_open_options is converted to a JSON object, prefixed with | |
7912 | * "json:" (for use through the JSON pseudo protocol) and put here. | |
7913 | */ | |
7914 | void bdrv_refresh_filename(BlockDriverState *bs) | |
7915 | { | |
7916 | BlockDriver *drv = bs->drv; | |
e24518e3 | 7917 | BdrvChild *child; |
52f72d6f | 7918 | BlockDriverState *primary_child_bs; |
91af7014 | 7919 | QDict *opts; |
90993623 | 7920 | bool backing_overridden; |
998b3a1e HR |
7921 | bool generate_json_filename; /* Whether our default implementation should |
7922 | fill exact_filename (false) or not (true) */ | |
91af7014 | 7923 | |
f791bf7f EGE |
7924 | GLOBAL_STATE_CODE(); |
7925 | ||
91af7014 HR |
7926 | if (!drv) { |
7927 | return; | |
7928 | } | |
7929 | ||
e24518e3 HR |
7930 | /* This BDS's file name may depend on any of its children's file names, so |
7931 | * refresh those first */ | |
7932 | QLIST_FOREACH(child, &bs->children, next) { | |
7933 | bdrv_refresh_filename(child->bs); | |
91af7014 HR |
7934 | } |
7935 | ||
bb808d5f HR |
7936 | if (bs->implicit) { |
7937 | /* For implicit nodes, just copy everything from the single child */ | |
7938 | child = QLIST_FIRST(&bs->children); | |
7939 | assert(QLIST_NEXT(child, next) == NULL); | |
7940 | ||
7941 | pstrcpy(bs->exact_filename, sizeof(bs->exact_filename), | |
7942 | child->bs->exact_filename); | |
7943 | pstrcpy(bs->filename, sizeof(bs->filename), child->bs->filename); | |
7944 | ||
cb895614 | 7945 | qobject_unref(bs->full_open_options); |
bb808d5f HR |
7946 | bs->full_open_options = qobject_ref(child->bs->full_open_options); |
7947 | ||
7948 | return; | |
7949 | } | |
7950 | ||
90993623 HR |
7951 | backing_overridden = bdrv_backing_overridden(bs); |
7952 | ||
7953 | if (bs->open_flags & BDRV_O_NO_IO) { | |
7954 | /* Without I/O, the backing file does not change anything. | |
7955 | * Therefore, in such a case (primarily qemu-img), we can | |
7956 | * pretend the backing file has not been overridden even if | |
7957 | * it technically has been. */ | |
7958 | backing_overridden = false; | |
7959 | } | |
7960 | ||
97e2f021 HR |
7961 | /* Gather the options QDict */ |
7962 | opts = qdict_new(); | |
998b3a1e HR |
7963 | generate_json_filename = append_strong_runtime_options(opts, bs); |
7964 | generate_json_filename |= backing_overridden; | |
97e2f021 HR |
7965 | |
7966 | if (drv->bdrv_gather_child_options) { | |
7967 | /* Some block drivers may not want to present all of their children's | |
7968 | * options, or name them differently from BdrvChild.name */ | |
7969 | drv->bdrv_gather_child_options(bs, opts, backing_overridden); | |
7970 | } else { | |
7971 | QLIST_FOREACH(child, &bs->children, next) { | |
25191e5f | 7972 | if (child == bs->backing && !backing_overridden) { |
97e2f021 HR |
7973 | /* We can skip the backing BDS if it has not been overridden */ |
7974 | continue; | |
7975 | } | |
7976 | ||
7977 | qdict_put(opts, child->name, | |
7978 | qobject_ref(child->bs->full_open_options)); | |
7979 | } | |
7980 | ||
7981 | if (backing_overridden && !bs->backing) { | |
7982 | /* Force no backing file */ | |
7983 | qdict_put_null(opts, "backing"); | |
7984 | } | |
7985 | } | |
7986 | ||
7987 | qobject_unref(bs->full_open_options); | |
7988 | bs->full_open_options = opts; | |
7989 | ||
52f72d6f HR |
7990 | primary_child_bs = bdrv_primary_bs(bs); |
7991 | ||
998b3a1e HR |
7992 | if (drv->bdrv_refresh_filename) { |
7993 | /* Obsolete information is of no use here, so drop the old file name | |
7994 | * information before refreshing it */ | |
7995 | bs->exact_filename[0] = '\0'; | |
7996 | ||
7997 | drv->bdrv_refresh_filename(bs); | |
52f72d6f HR |
7998 | } else if (primary_child_bs) { |
7999 | /* | |
8000 | * Try to reconstruct valid information from the underlying | |
8001 | * file -- this only works for format nodes (filter nodes | |
8002 | * cannot be probed and as such must be selected by the user | |
8003 | * either through an options dict, or through a special | |
8004 | * filename which the filter driver must construct in its | |
8005 | * .bdrv_refresh_filename() implementation). | |
8006 | */ | |
998b3a1e HR |
8007 | |
8008 | bs->exact_filename[0] = '\0'; | |
8009 | ||
fb695c74 HR |
8010 | /* |
8011 | * We can use the underlying file's filename if: | |
8012 | * - it has a filename, | |
52f72d6f | 8013 | * - the current BDS is not a filter, |
fb695c74 HR |
8014 | * - the file is a protocol BDS, and |
8015 | * - opening that file (as this BDS's format) will automatically create | |
8016 | * the BDS tree we have right now, that is: | |
8017 | * - the user did not significantly change this BDS's behavior with | |
8018 | * some explicit (strong) options | |
8019 | * - no non-file child of this BDS has been overridden by the user | |
8020 | * Both of these conditions are represented by generate_json_filename. | |
8021 | */ | |
52f72d6f HR |
8022 | if (primary_child_bs->exact_filename[0] && |
8023 | primary_child_bs->drv->bdrv_file_open && | |
8024 | !drv->is_filter && !generate_json_filename) | |
fb695c74 | 8025 | { |
52f72d6f | 8026 | strcpy(bs->exact_filename, primary_child_bs->exact_filename); |
998b3a1e HR |
8027 | } |
8028 | } | |
8029 | ||
91af7014 HR |
8030 | if (bs->exact_filename[0]) { |
8031 | pstrcpy(bs->filename, sizeof(bs->filename), bs->exact_filename); | |
97e2f021 | 8032 | } else { |
eab3a467 | 8033 | GString *json = qobject_to_json(QOBJECT(bs->full_open_options)); |
5c86bdf1 | 8034 | if (snprintf(bs->filename, sizeof(bs->filename), "json:%s", |
eab3a467 | 8035 | json->str) >= sizeof(bs->filename)) { |
5c86bdf1 EB |
8036 | /* Give user a hint if we truncated things. */ |
8037 | strcpy(bs->filename + sizeof(bs->filename) - 4, "..."); | |
8038 | } | |
eab3a467 | 8039 | g_string_free(json, true); |
91af7014 HR |
8040 | } |
8041 | } | |
e06018ad | 8042 | |
1e89d0f9 HR |
8043 | char *bdrv_dirname(BlockDriverState *bs, Error **errp) |
8044 | { | |
8045 | BlockDriver *drv = bs->drv; | |
52f72d6f | 8046 | BlockDriverState *child_bs; |
1e89d0f9 | 8047 | |
f791bf7f EGE |
8048 | GLOBAL_STATE_CODE(); |
8049 | ||
1e89d0f9 HR |
8050 | if (!drv) { |
8051 | error_setg(errp, "Node '%s' is ejected", bs->node_name); | |
8052 | return NULL; | |
8053 | } | |
8054 | ||
8055 | if (drv->bdrv_dirname) { | |
8056 | return drv->bdrv_dirname(bs, errp); | |
8057 | } | |
8058 | ||
52f72d6f HR |
8059 | child_bs = bdrv_primary_bs(bs); |
8060 | if (child_bs) { | |
8061 | return bdrv_dirname(child_bs, errp); | |
1e89d0f9 HR |
8062 | } |
8063 | ||
8064 | bdrv_refresh_filename(bs); | |
8065 | if (bs->exact_filename[0] != '\0') { | |
8066 | return path_combine(bs->exact_filename, ""); | |
8067 | } | |
8068 | ||
8069 | error_setg(errp, "Cannot generate a base directory for %s nodes", | |
8070 | drv->format_name); | |
8071 | return NULL; | |
8072 | } | |
8073 | ||
e06018ad WC |
8074 | /* |
8075 | * Hot add/remove a BDS's child. So the user can take a child offline when | |
8076 | * it is broken and take a new child online | |
8077 | */ | |
8078 | void bdrv_add_child(BlockDriverState *parent_bs, BlockDriverState *child_bs, | |
8079 | Error **errp) | |
8080 | { | |
f791bf7f | 8081 | GLOBAL_STATE_CODE(); |
e06018ad WC |
8082 | if (!parent_bs->drv || !parent_bs->drv->bdrv_add_child) { |
8083 | error_setg(errp, "The node %s does not support adding a child", | |
8084 | bdrv_get_device_or_node_name(parent_bs)); | |
8085 | return; | |
8086 | } | |
8087 | ||
774c726c SL |
8088 | /* |
8089 | * Non-zoned block drivers do not follow zoned storage constraints | |
8090 | * (i.e. sequential writes to zones). Refuse mixing zoned and non-zoned | |
8091 | * drivers in a graph. | |
8092 | */ | |
8093 | if (!parent_bs->drv->supports_zoned_children && | |
8094 | child_bs->bl.zoned == BLK_Z_HM) { | |
8095 | /* | |
8096 | * The host-aware model allows zoned storage constraints and random | |
8097 | * write. Allow mixing host-aware and non-zoned drivers. Using | |
8098 | * host-aware device as a regular device. | |
8099 | */ | |
8100 | error_setg(errp, "Cannot add a %s child to a %s parent", | |
8101 | child_bs->bl.zoned == BLK_Z_HM ? "zoned" : "non-zoned", | |
8102 | parent_bs->drv->supports_zoned_children ? | |
8103 | "support zoned children" : "not support zoned children"); | |
8104 | return; | |
8105 | } | |
8106 | ||
e06018ad WC |
8107 | if (!QLIST_EMPTY(&child_bs->parents)) { |
8108 | error_setg(errp, "The node %s already has a parent", | |
8109 | child_bs->node_name); | |
8110 | return; | |
8111 | } | |
8112 | ||
8113 | parent_bs->drv->bdrv_add_child(parent_bs, child_bs, errp); | |
8114 | } | |
8115 | ||
8116 | void bdrv_del_child(BlockDriverState *parent_bs, BdrvChild *child, Error **errp) | |
8117 | { | |
8118 | BdrvChild *tmp; | |
8119 | ||
f791bf7f | 8120 | GLOBAL_STATE_CODE(); |
e06018ad WC |
8121 | if (!parent_bs->drv || !parent_bs->drv->bdrv_del_child) { |
8122 | error_setg(errp, "The node %s does not support removing a child", | |
8123 | bdrv_get_device_or_node_name(parent_bs)); | |
8124 | return; | |
8125 | } | |
8126 | ||
8127 | QLIST_FOREACH(tmp, &parent_bs->children, next) { | |
8128 | if (tmp == child) { | |
8129 | break; | |
8130 | } | |
8131 | } | |
8132 | ||
8133 | if (!tmp) { | |
8134 | error_setg(errp, "The node %s does not have a child named %s", | |
8135 | bdrv_get_device_or_node_name(parent_bs), | |
8136 | bdrv_get_device_or_node_name(child->bs)); | |
8137 | return; | |
8138 | } | |
8139 | ||
8140 | parent_bs->drv->bdrv_del_child(parent_bs, child, errp); | |
8141 | } | |
6f7a3b53 HR |
8142 | |
8143 | int bdrv_make_empty(BdrvChild *c, Error **errp) | |
8144 | { | |
8145 | BlockDriver *drv = c->bs->drv; | |
8146 | int ret; | |
8147 | ||
f791bf7f | 8148 | GLOBAL_STATE_CODE(); |
6f7a3b53 HR |
8149 | assert(c->perm & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)); |
8150 | ||
8151 | if (!drv->bdrv_make_empty) { | |
8152 | error_setg(errp, "%s does not support emptying nodes", | |
8153 | drv->format_name); | |
8154 | return -ENOTSUP; | |
8155 | } | |
8156 | ||
8157 | ret = drv->bdrv_make_empty(c->bs); | |
8158 | if (ret < 0) { | |
8159 | error_setg_errno(errp, -ret, "Failed to empty %s", | |
8160 | c->bs->filename); | |
8161 | return ret; | |
8162 | } | |
8163 | ||
8164 | return 0; | |
8165 | } | |
9a6fc887 HR |
8166 | |
8167 | /* | |
8168 | * Return the child that @bs acts as an overlay for, and from which data may be | |
8169 | * copied in COW or COR operations. Usually this is the backing file. | |
8170 | */ | |
8171 | BdrvChild *bdrv_cow_child(BlockDriverState *bs) | |
8172 | { | |
967d7905 EGE |
8173 | IO_CODE(); |
8174 | ||
9a6fc887 HR |
8175 | if (!bs || !bs->drv) { |
8176 | return NULL; | |
8177 | } | |
8178 | ||
8179 | if (bs->drv->is_filter) { | |
8180 | return NULL; | |
8181 | } | |
8182 | ||
8183 | if (!bs->backing) { | |
8184 | return NULL; | |
8185 | } | |
8186 | ||
8187 | assert(bs->backing->role & BDRV_CHILD_COW); | |
8188 | return bs->backing; | |
8189 | } | |
8190 | ||
8191 | /* | |
8192 | * If @bs acts as a filter for exactly one of its children, return | |
8193 | * that child. | |
8194 | */ | |
8195 | BdrvChild *bdrv_filter_child(BlockDriverState *bs) | |
8196 | { | |
8197 | BdrvChild *c; | |
967d7905 | 8198 | IO_CODE(); |
9a6fc887 HR |
8199 | |
8200 | if (!bs || !bs->drv) { | |
8201 | return NULL; | |
8202 | } | |
8203 | ||
8204 | if (!bs->drv->is_filter) { | |
8205 | return NULL; | |
8206 | } | |
8207 | ||
8208 | /* Only one of @backing or @file may be used */ | |
8209 | assert(!(bs->backing && bs->file)); | |
8210 | ||
8211 | c = bs->backing ?: bs->file; | |
8212 | if (!c) { | |
8213 | return NULL; | |
8214 | } | |
8215 | ||
8216 | assert(c->role & BDRV_CHILD_FILTERED); | |
8217 | return c; | |
8218 | } | |
8219 | ||
8220 | /* | |
8221 | * Return either the result of bdrv_cow_child() or bdrv_filter_child(), | |
8222 | * whichever is non-NULL. | |
8223 | * | |
8224 | * Return NULL if both are NULL. | |
8225 | */ | |
8226 | BdrvChild *bdrv_filter_or_cow_child(BlockDriverState *bs) | |
8227 | { | |
8228 | BdrvChild *cow_child = bdrv_cow_child(bs); | |
8229 | BdrvChild *filter_child = bdrv_filter_child(bs); | |
967d7905 | 8230 | IO_CODE(); |
9a6fc887 HR |
8231 | |
8232 | /* Filter nodes cannot have COW backing files */ | |
8233 | assert(!(cow_child && filter_child)); | |
8234 | ||
8235 | return cow_child ?: filter_child; | |
8236 | } | |
8237 | ||
8238 | /* | |
8239 | * Return the primary child of this node: For filters, that is the | |
8240 | * filtered child. For other nodes, that is usually the child storing | |
8241 | * metadata. | |
8242 | * (A generally more helpful description is that this is (usually) the | |
8243 | * child that has the same filename as @bs.) | |
8244 | * | |
8245 | * Drivers do not necessarily have a primary child; for example quorum | |
8246 | * does not. | |
8247 | */ | |
8248 | BdrvChild *bdrv_primary_child(BlockDriverState *bs) | |
8249 | { | |
8250 | BdrvChild *c, *found = NULL; | |
967d7905 | 8251 | IO_CODE(); |
9a6fc887 HR |
8252 | |
8253 | QLIST_FOREACH(c, &bs->children, next) { | |
8254 | if (c->role & BDRV_CHILD_PRIMARY) { | |
8255 | assert(!found); | |
8256 | found = c; | |
8257 | } | |
8258 | } | |
8259 | ||
8260 | return found; | |
8261 | } | |
d38d7eb8 HR |
8262 | |
8263 | static BlockDriverState *bdrv_do_skip_filters(BlockDriverState *bs, | |
8264 | bool stop_on_explicit_filter) | |
8265 | { | |
8266 | BdrvChild *c; | |
8267 | ||
8268 | if (!bs) { | |
8269 | return NULL; | |
8270 | } | |
8271 | ||
8272 | while (!(stop_on_explicit_filter && !bs->implicit)) { | |
8273 | c = bdrv_filter_child(bs); | |
8274 | if (!c) { | |
8275 | /* | |
8276 | * A filter that is embedded in a working block graph must | |
8277 | * have a child. Assert this here so this function does | |
8278 | * not return a filter node that is not expected by the | |
8279 | * caller. | |
8280 | */ | |
8281 | assert(!bs->drv || !bs->drv->is_filter); | |
8282 | break; | |
8283 | } | |
8284 | bs = c->bs; | |
8285 | } | |
8286 | /* | |
8287 | * Note that this treats nodes with bs->drv == NULL as not being | |
8288 | * filters (bs->drv == NULL should be replaced by something else | |
8289 | * anyway). | |
8290 | * The advantage of this behavior is that this function will thus | |
8291 | * always return a non-NULL value (given a non-NULL @bs). | |
8292 | */ | |
8293 | ||
8294 | return bs; | |
8295 | } | |
8296 | ||
8297 | /* | |
8298 | * Return the first BDS that has not been added implicitly or that | |
8299 | * does not have a filtered child down the chain starting from @bs | |
8300 | * (including @bs itself). | |
8301 | */ | |
8302 | BlockDriverState *bdrv_skip_implicit_filters(BlockDriverState *bs) | |
8303 | { | |
b4ad82aa | 8304 | GLOBAL_STATE_CODE(); |
d38d7eb8 HR |
8305 | return bdrv_do_skip_filters(bs, true); |
8306 | } | |
8307 | ||
8308 | /* | |
8309 | * Return the first BDS that does not have a filtered child down the | |
8310 | * chain starting from @bs (including @bs itself). | |
8311 | */ | |
8312 | BlockDriverState *bdrv_skip_filters(BlockDriverState *bs) | |
8313 | { | |
967d7905 | 8314 | IO_CODE(); |
d38d7eb8 HR |
8315 | return bdrv_do_skip_filters(bs, false); |
8316 | } | |
8317 | ||
8318 | /* | |
8319 | * For a backing chain, return the first non-filter backing image of | |
8320 | * the first non-filter image. | |
8321 | */ | |
8322 | BlockDriverState *bdrv_backing_chain_next(BlockDriverState *bs) | |
8323 | { | |
967d7905 | 8324 | IO_CODE(); |
d38d7eb8 HR |
8325 | return bdrv_skip_filters(bdrv_cow_bs(bdrv_skip_filters(bs))); |
8326 | } | |
0bc329fb HR |
8327 | |
8328 | /** | |
8329 | * Check whether [offset, offset + bytes) overlaps with the cached | |
8330 | * block-status data region. | |
8331 | * | |
8332 | * If so, and @pnum is not NULL, set *pnum to `bsc.data_end - offset`, | |
8333 | * which is what bdrv_bsc_is_data()'s interface needs. | |
8334 | * Otherwise, *pnum is not touched. | |
8335 | */ | |
8336 | static bool bdrv_bsc_range_overlaps_locked(BlockDriverState *bs, | |
8337 | int64_t offset, int64_t bytes, | |
8338 | int64_t *pnum) | |
8339 | { | |
8340 | BdrvBlockStatusCache *bsc = qatomic_rcu_read(&bs->block_status_cache); | |
8341 | bool overlaps; | |
8342 | ||
8343 | overlaps = | |
8344 | qatomic_read(&bsc->valid) && | |
8345 | ranges_overlap(offset, bytes, bsc->data_start, | |
8346 | bsc->data_end - bsc->data_start); | |
8347 | ||
8348 | if (overlaps && pnum) { | |
8349 | *pnum = bsc->data_end - offset; | |
8350 | } | |
8351 | ||
8352 | return overlaps; | |
8353 | } | |
8354 | ||
8355 | /** | |
8356 | * See block_int.h for this function's documentation. | |
8357 | */ | |
8358 | bool bdrv_bsc_is_data(BlockDriverState *bs, int64_t offset, int64_t *pnum) | |
8359 | { | |
967d7905 | 8360 | IO_CODE(); |
0bc329fb | 8361 | RCU_READ_LOCK_GUARD(); |
0bc329fb HR |
8362 | return bdrv_bsc_range_overlaps_locked(bs, offset, 1, pnum); |
8363 | } | |
8364 | ||
8365 | /** | |
8366 | * See block_int.h for this function's documentation. | |
8367 | */ | |
8368 | void bdrv_bsc_invalidate_range(BlockDriverState *bs, | |
8369 | int64_t offset, int64_t bytes) | |
8370 | { | |
967d7905 | 8371 | IO_CODE(); |
0bc329fb HR |
8372 | RCU_READ_LOCK_GUARD(); |
8373 | ||
8374 | if (bdrv_bsc_range_overlaps_locked(bs, offset, bytes, NULL)) { | |
8375 | qatomic_set(&bs->block_status_cache->valid, false); | |
8376 | } | |
8377 | } | |
8378 | ||
8379 | /** | |
8380 | * See block_int.h for this function's documentation. | |
8381 | */ | |
8382 | void bdrv_bsc_fill(BlockDriverState *bs, int64_t offset, int64_t bytes) | |
8383 | { | |
8384 | BdrvBlockStatusCache *new_bsc = g_new(BdrvBlockStatusCache, 1); | |
8385 | BdrvBlockStatusCache *old_bsc; | |
967d7905 | 8386 | IO_CODE(); |
0bc329fb HR |
8387 | |
8388 | *new_bsc = (BdrvBlockStatusCache) { | |
8389 | .valid = true, | |
8390 | .data_start = offset, | |
8391 | .data_end = offset + bytes, | |
8392 | }; | |
8393 | ||
8394 | QEMU_LOCK_GUARD(&bs->bsc_modify_lock); | |
8395 | ||
8396 | old_bsc = qatomic_rcu_read(&bs->block_status_cache); | |
8397 | qatomic_rcu_set(&bs->block_status_cache, new_bsc); | |
8398 | if (old_bsc) { | |
8399 | g_free_rcu(old_bsc, rcu); | |
8400 | } | |
8401 | } |