]>
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 | ||
54a32bfe KW |
280 | int bdrv_can_set_read_only(BlockDriverState *bs, bool read_only, |
281 | bool ignore_allow_rdw, Error **errp) | |
fe5241bf | 282 | { |
384a48fb EGE |
283 | IO_CODE(); |
284 | ||
e2b8247a JC |
285 | /* Do not set read_only if copy_on_read is enabled */ |
286 | if (bs->copy_on_read && read_only) { | |
287 | error_setg(errp, "Can't set node '%s' to r/o with copy-on-read enabled", | |
288 | bdrv_get_device_or_node_name(bs)); | |
289 | return -EINVAL; | |
290 | } | |
291 | ||
d6fcdf06 | 292 | /* Do not clear read_only if it is prohibited */ |
54a32bfe KW |
293 | if (!read_only && !(bs->open_flags & BDRV_O_ALLOW_RDWR) && |
294 | !ignore_allow_rdw) | |
295 | { | |
d6fcdf06 JC |
296 | error_setg(errp, "Node '%s' is read only", |
297 | bdrv_get_device_or_node_name(bs)); | |
298 | return -EPERM; | |
299 | } | |
300 | ||
45803a03 JC |
301 | return 0; |
302 | } | |
303 | ||
eaa2410f KW |
304 | /* |
305 | * Called by a driver that can only provide a read-only image. | |
306 | * | |
307 | * Returns 0 if the node is already read-only or it could switch the node to | |
308 | * read-only because BDRV_O_AUTO_RDONLY is set. | |
309 | * | |
310 | * Returns -EACCES if the node is read-write and BDRV_O_AUTO_RDONLY is not set | |
311 | * or bdrv_can_set_read_only() forbids making the node read-only. If @errmsg | |
312 | * is not NULL, it is used as the error message for the Error object. | |
313 | */ | |
314 | int bdrv_apply_auto_read_only(BlockDriverState *bs, const char *errmsg, | |
315 | Error **errp) | |
45803a03 JC |
316 | { |
317 | int ret = 0; | |
384a48fb | 318 | IO_CODE(); |
45803a03 | 319 | |
eaa2410f KW |
320 | if (!(bs->open_flags & BDRV_O_RDWR)) { |
321 | return 0; | |
322 | } | |
323 | if (!(bs->open_flags & BDRV_O_AUTO_RDONLY)) { | |
324 | goto fail; | |
45803a03 JC |
325 | } |
326 | ||
eaa2410f KW |
327 | ret = bdrv_can_set_read_only(bs, true, false, NULL); |
328 | if (ret < 0) { | |
329 | goto fail; | |
eeae6a59 KW |
330 | } |
331 | ||
eaa2410f KW |
332 | bs->open_flags &= ~BDRV_O_RDWR; |
333 | ||
e2b8247a | 334 | return 0; |
eaa2410f KW |
335 | |
336 | fail: | |
337 | error_setg(errp, "%s", errmsg ?: "Image is read-only"); | |
338 | return -EACCES; | |
fe5241bf JC |
339 | } |
340 | ||
645ae7d8 HR |
341 | /* |
342 | * If @backing is empty, this function returns NULL without setting | |
343 | * @errp. In all other cases, NULL will only be returned with @errp | |
344 | * set. | |
345 | * | |
346 | * Therefore, a return value of NULL without @errp set means that | |
347 | * there is no backing file; if @errp is set, there is one but its | |
348 | * absolute filename cannot be generated. | |
349 | */ | |
350 | char *bdrv_get_full_backing_filename_from_filename(const char *backed, | |
351 | const char *backing, | |
352 | Error **errp) | |
dc5a1371 | 353 | { |
645ae7d8 HR |
354 | if (backing[0] == '\0') { |
355 | return NULL; | |
356 | } else if (path_has_protocol(backing) || path_is_absolute(backing)) { | |
357 | return g_strdup(backing); | |
9f07429e HR |
358 | } else if (backed[0] == '\0' || strstart(backed, "json:", NULL)) { |
359 | error_setg(errp, "Cannot use relative backing file names for '%s'", | |
360 | backed); | |
645ae7d8 | 361 | return NULL; |
dc5a1371 | 362 | } else { |
645ae7d8 | 363 | return path_combine(backed, backing); |
dc5a1371 PB |
364 | } |
365 | } | |
366 | ||
9f4793d8 HR |
367 | /* |
368 | * If @filename is empty or NULL, this function returns NULL without | |
369 | * setting @errp. In all other cases, NULL will only be returned with | |
370 | * @errp set. | |
371 | */ | |
372 | static char *bdrv_make_absolute_filename(BlockDriverState *relative_to, | |
373 | const char *filename, Error **errp) | |
0a82855a | 374 | { |
8df68616 | 375 | char *dir, *full_name; |
9f4793d8 | 376 | |
8df68616 HR |
377 | if (!filename || filename[0] == '\0') { |
378 | return NULL; | |
379 | } else if (path_has_protocol(filename) || path_is_absolute(filename)) { | |
380 | return g_strdup(filename); | |
381 | } | |
9f07429e | 382 | |
8df68616 HR |
383 | dir = bdrv_dirname(relative_to, errp); |
384 | if (!dir) { | |
385 | return NULL; | |
386 | } | |
f30c66ba | 387 | |
8df68616 HR |
388 | full_name = g_strconcat(dir, filename, NULL); |
389 | g_free(dir); | |
390 | return full_name; | |
9f4793d8 HR |
391 | } |
392 | ||
393 | char *bdrv_get_full_backing_filename(BlockDriverState *bs, Error **errp) | |
394 | { | |
f791bf7f | 395 | GLOBAL_STATE_CODE(); |
9f4793d8 | 396 | return bdrv_make_absolute_filename(bs, bs->backing_file, errp); |
0a82855a HR |
397 | } |
398 | ||
0eb7217e SH |
399 | void bdrv_register(BlockDriver *bdrv) |
400 | { | |
a15f08dc | 401 | assert(bdrv->format_name); |
f791bf7f | 402 | GLOBAL_STATE_CODE(); |
8a22f02a | 403 | QLIST_INSERT_HEAD(&bdrv_drivers, bdrv, list); |
ea2384d3 | 404 | } |
b338082b | 405 | |
e4e9986b MA |
406 | BlockDriverState *bdrv_new(void) |
407 | { | |
408 | BlockDriverState *bs; | |
409 | int i; | |
410 | ||
f791bf7f EGE |
411 | GLOBAL_STATE_CODE(); |
412 | ||
5839e53b | 413 | bs = g_new0(BlockDriverState, 1); |
e4654d2d | 414 | QLIST_INIT(&bs->dirty_bitmaps); |
fbe40ff7 FZ |
415 | for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) { |
416 | QLIST_INIT(&bs->op_blockers[i]); | |
417 | } | |
3783fa3d | 418 | qemu_co_mutex_init(&bs->reqs_lock); |
2119882c | 419 | qemu_mutex_init(&bs->dirty_bitmap_mutex); |
9fcb0251 | 420 | bs->refcnt = 1; |
dcd04228 | 421 | bs->aio_context = qemu_get_aio_context(); |
d7d512f6 | 422 | |
3ff2f67a EY |
423 | qemu_co_queue_init(&bs->flush_queue); |
424 | ||
0bc329fb HR |
425 | qemu_co_mutex_init(&bs->bsc_modify_lock); |
426 | bs->block_status_cache = g_new0(BdrvBlockStatusCache, 1); | |
427 | ||
0f12264e KW |
428 | for (i = 0; i < bdrv_drain_all_count; i++) { |
429 | bdrv_drained_begin(bs); | |
430 | } | |
431 | ||
2c1d04e0 HR |
432 | QTAILQ_INSERT_TAIL(&all_bdrv_states, bs, bs_list); |
433 | ||
b338082b FB |
434 | return bs; |
435 | } | |
436 | ||
88d88798 | 437 | static BlockDriver *bdrv_do_find_format(const char *format_name) |
ea2384d3 FB |
438 | { |
439 | BlockDriver *drv1; | |
bdb73476 | 440 | GLOBAL_STATE_CODE(); |
88d88798 | 441 | |
8a22f02a SH |
442 | QLIST_FOREACH(drv1, &bdrv_drivers, list) { |
443 | if (!strcmp(drv1->format_name, format_name)) { | |
ea2384d3 | 444 | return drv1; |
8a22f02a | 445 | } |
ea2384d3 | 446 | } |
88d88798 | 447 | |
ea2384d3 FB |
448 | return NULL; |
449 | } | |
450 | ||
88d88798 MM |
451 | BlockDriver *bdrv_find_format(const char *format_name) |
452 | { | |
453 | BlockDriver *drv1; | |
454 | int i; | |
455 | ||
f791bf7f EGE |
456 | GLOBAL_STATE_CODE(); |
457 | ||
88d88798 MM |
458 | drv1 = bdrv_do_find_format(format_name); |
459 | if (drv1) { | |
460 | return drv1; | |
461 | } | |
462 | ||
463 | /* The driver isn't registered, maybe we need to load a module */ | |
464 | for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); ++i) { | |
465 | if (!strcmp(block_driver_modules[i].format_name, format_name)) { | |
c551fb0b CF |
466 | Error *local_err = NULL; |
467 | int rv = block_module_load(block_driver_modules[i].library_name, | |
468 | &local_err); | |
469 | if (rv > 0) { | |
470 | return bdrv_do_find_format(format_name); | |
471 | } else if (rv < 0) { | |
472 | error_report_err(local_err); | |
473 | } | |
88d88798 MM |
474 | break; |
475 | } | |
476 | } | |
c551fb0b | 477 | return NULL; |
88d88798 MM |
478 | } |
479 | ||
9ac404c5 | 480 | static int bdrv_format_is_whitelisted(const char *format_name, bool read_only) |
eb852011 | 481 | { |
b64ec4e4 FZ |
482 | static const char *whitelist_rw[] = { |
483 | CONFIG_BDRV_RW_WHITELIST | |
859aef02 | 484 | NULL |
b64ec4e4 FZ |
485 | }; |
486 | static const char *whitelist_ro[] = { | |
487 | CONFIG_BDRV_RO_WHITELIST | |
859aef02 | 488 | NULL |
eb852011 MA |
489 | }; |
490 | const char **p; | |
491 | ||
b64ec4e4 | 492 | if (!whitelist_rw[0] && !whitelist_ro[0]) { |
eb852011 | 493 | return 1; /* no whitelist, anything goes */ |
b64ec4e4 | 494 | } |
eb852011 | 495 | |
b64ec4e4 | 496 | for (p = whitelist_rw; *p; p++) { |
9ac404c5 | 497 | if (!strcmp(format_name, *p)) { |
eb852011 MA |
498 | return 1; |
499 | } | |
500 | } | |
b64ec4e4 FZ |
501 | if (read_only) { |
502 | for (p = whitelist_ro; *p; p++) { | |
9ac404c5 | 503 | if (!strcmp(format_name, *p)) { |
b64ec4e4 FZ |
504 | return 1; |
505 | } | |
506 | } | |
507 | } | |
eb852011 MA |
508 | return 0; |
509 | } | |
510 | ||
9ac404c5 AS |
511 | int bdrv_is_whitelisted(BlockDriver *drv, bool read_only) |
512 | { | |
f791bf7f | 513 | GLOBAL_STATE_CODE(); |
9ac404c5 AS |
514 | return bdrv_format_is_whitelisted(drv->format_name, read_only); |
515 | } | |
516 | ||
e6ff69bf DB |
517 | bool bdrv_uses_whitelist(void) |
518 | { | |
519 | return use_bdrv_whitelist; | |
520 | } | |
521 | ||
5b7e1542 ZYW |
522 | typedef struct CreateCo { |
523 | BlockDriver *drv; | |
524 | char *filename; | |
83d0521a | 525 | QemuOpts *opts; |
5b7e1542 | 526 | int ret; |
cc84d90f | 527 | Error *err; |
5b7e1542 ZYW |
528 | } CreateCo; |
529 | ||
741443eb EGE |
530 | int coroutine_fn bdrv_co_create(BlockDriver *drv, const char *filename, |
531 | QemuOpts *opts, Error **errp) | |
5b7e1542 | 532 | { |
cc84d90f | 533 | int ret; |
84bdf21f EGE |
534 | GLOBAL_STATE_CODE(); |
535 | ERRP_GUARD(); | |
536 | ||
537 | if (!drv->bdrv_co_create_opts) { | |
538 | error_setg(errp, "Driver '%s' does not support image creation", | |
539 | drv->format_name); | |
540 | return -ENOTSUP; | |
541 | } | |
542 | ||
543 | ret = drv->bdrv_co_create_opts(drv, filename, opts, errp); | |
544 | if (ret < 0 && !*errp) { | |
545 | error_setg_errno(errp, -ret, "Could not create image"); | |
546 | } | |
cc84d90f | 547 | |
84bdf21f EGE |
548 | return ret; |
549 | } | |
550 | ||
fd17146c HR |
551 | /** |
552 | * Helper function for bdrv_create_file_fallback(): Resize @blk to at | |
553 | * least the given @minimum_size. | |
554 | * | |
555 | * On success, return @blk's actual length. | |
556 | * Otherwise, return -errno. | |
557 | */ | |
558 | static int64_t create_file_fallback_truncate(BlockBackend *blk, | |
559 | int64_t minimum_size, Error **errp) | |
84a12e66 | 560 | { |
cc84d90f | 561 | Error *local_err = NULL; |
fd17146c | 562 | int64_t size; |
cc84d90f | 563 | int ret; |
84a12e66 | 564 | |
bdb73476 EGE |
565 | GLOBAL_STATE_CODE(); |
566 | ||
8c6242b6 KW |
567 | ret = blk_truncate(blk, minimum_size, false, PREALLOC_MODE_OFF, 0, |
568 | &local_err); | |
fd17146c HR |
569 | if (ret < 0 && ret != -ENOTSUP) { |
570 | error_propagate(errp, local_err); | |
571 | return ret; | |
572 | } | |
573 | ||
574 | size = blk_getlength(blk); | |
575 | if (size < 0) { | |
576 | error_free(local_err); | |
577 | error_setg_errno(errp, -size, | |
578 | "Failed to inquire the new image file's length"); | |
579 | return size; | |
580 | } | |
581 | ||
582 | if (size < minimum_size) { | |
583 | /* Need to grow the image, but we failed to do that */ | |
584 | error_propagate(errp, local_err); | |
585 | return -ENOTSUP; | |
586 | } | |
587 | ||
588 | error_free(local_err); | |
589 | local_err = NULL; | |
590 | ||
591 | return size; | |
592 | } | |
593 | ||
594 | /** | |
595 | * Helper function for bdrv_create_file_fallback(): Zero the first | |
596 | * sector to remove any potentially pre-existing image header. | |
597 | */ | |
881a4c55 PB |
598 | static int coroutine_fn |
599 | create_file_fallback_zero_first_sector(BlockBackend *blk, | |
600 | int64_t current_size, | |
601 | Error **errp) | |
fd17146c HR |
602 | { |
603 | int64_t bytes_to_clear; | |
604 | int ret; | |
605 | ||
bdb73476 EGE |
606 | GLOBAL_STATE_CODE(); |
607 | ||
fd17146c HR |
608 | bytes_to_clear = MIN(current_size, BDRV_SECTOR_SIZE); |
609 | if (bytes_to_clear) { | |
ce47ff20 | 610 | ret = blk_co_pwrite_zeroes(blk, 0, bytes_to_clear, BDRV_REQ_MAY_UNMAP); |
fd17146c HR |
611 | if (ret < 0) { |
612 | error_setg_errno(errp, -ret, | |
613 | "Failed to clear the new image's first sector"); | |
614 | return ret; | |
615 | } | |
616 | } | |
617 | ||
618 | return 0; | |
619 | } | |
620 | ||
5a5e7f8c ML |
621 | /** |
622 | * Simple implementation of bdrv_co_create_opts for protocol drivers | |
623 | * which only support creation via opening a file | |
624 | * (usually existing raw storage device) | |
625 | */ | |
626 | int coroutine_fn bdrv_co_create_opts_simple(BlockDriver *drv, | |
627 | const char *filename, | |
628 | QemuOpts *opts, | |
629 | Error **errp) | |
fd17146c HR |
630 | { |
631 | BlockBackend *blk; | |
eeea1faa | 632 | QDict *options; |
fd17146c HR |
633 | int64_t size = 0; |
634 | char *buf = NULL; | |
635 | PreallocMode prealloc; | |
636 | Error *local_err = NULL; | |
637 | int ret; | |
638 | ||
b4ad82aa EGE |
639 | GLOBAL_STATE_CODE(); |
640 | ||
fd17146c HR |
641 | size = qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0); |
642 | buf = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC); | |
643 | prealloc = qapi_enum_parse(&PreallocMode_lookup, buf, | |
644 | PREALLOC_MODE_OFF, &local_err); | |
645 | g_free(buf); | |
646 | if (local_err) { | |
647 | error_propagate(errp, local_err); | |
648 | return -EINVAL; | |
649 | } | |
650 | ||
651 | if (prealloc != PREALLOC_MODE_OFF) { | |
652 | error_setg(errp, "Unsupported preallocation mode '%s'", | |
653 | PreallocMode_str(prealloc)); | |
654 | return -ENOTSUP; | |
655 | } | |
656 | ||
eeea1faa | 657 | options = qdict_new(); |
fd17146c HR |
658 | qdict_put_str(options, "driver", drv->format_name); |
659 | ||
660 | blk = blk_new_open(filename, NULL, options, | |
661 | BDRV_O_RDWR | BDRV_O_RESIZE, errp); | |
662 | if (!blk) { | |
663 | error_prepend(errp, "Protocol driver '%s' does not support image " | |
664 | "creation, and opening the image failed: ", | |
665 | drv->format_name); | |
666 | return -EINVAL; | |
667 | } | |
668 | ||
669 | size = create_file_fallback_truncate(blk, size, errp); | |
670 | if (size < 0) { | |
671 | ret = size; | |
672 | goto out; | |
673 | } | |
674 | ||
675 | ret = create_file_fallback_zero_first_sector(blk, size, errp); | |
676 | if (ret < 0) { | |
677 | goto out; | |
678 | } | |
679 | ||
680 | ret = 0; | |
681 | out: | |
682 | blk_unref(blk); | |
683 | return ret; | |
684 | } | |
685 | ||
2475a0d0 EGE |
686 | int coroutine_fn bdrv_co_create_file(const char *filename, QemuOpts *opts, |
687 | Error **errp) | |
fd17146c | 688 | { |
729222af | 689 | QemuOpts *protocol_opts; |
fd17146c | 690 | BlockDriver *drv; |
729222af SG |
691 | QDict *qdict; |
692 | int ret; | |
fd17146c | 693 | |
f791bf7f EGE |
694 | GLOBAL_STATE_CODE(); |
695 | ||
b65a5e12 | 696 | drv = bdrv_find_protocol(filename, true, errp); |
84a12e66 | 697 | if (drv == NULL) { |
16905d71 | 698 | return -ENOENT; |
84a12e66 CH |
699 | } |
700 | ||
729222af SG |
701 | if (!drv->create_opts) { |
702 | error_setg(errp, "Driver '%s' does not support image creation", | |
703 | drv->format_name); | |
704 | return -ENOTSUP; | |
705 | } | |
706 | ||
707 | /* | |
708 | * 'opts' contains a QemuOptsList with a combination of format and protocol | |
709 | * default values. | |
710 | * | |
711 | * The format properly removes its options, but the default values remain | |
712 | * in 'opts->list'. So if the protocol has options with the same name | |
713 | * (e.g. rbd has 'cluster_size' as qcow2), it will see the default values | |
714 | * of the format, since for overlapping options, the format wins. | |
715 | * | |
716 | * To avoid this issue, lets convert QemuOpts to QDict, in this way we take | |
717 | * only the set options, and then convert it back to QemuOpts, using the | |
718 | * create_opts of the protocol. So the new QemuOpts, will contain only the | |
719 | * protocol defaults. | |
720 | */ | |
721 | qdict = qemu_opts_to_qdict(opts, NULL); | |
722 | protocol_opts = qemu_opts_from_qdict(drv->create_opts, qdict, errp); | |
723 | if (protocol_opts == NULL) { | |
724 | ret = -EINVAL; | |
725 | goto out; | |
726 | } | |
727 | ||
2475a0d0 | 728 | ret = bdrv_co_create(drv, filename, protocol_opts, errp); |
729222af SG |
729 | out: |
730 | qemu_opts_del(protocol_opts); | |
731 | qobject_unref(qdict); | |
732 | return ret; | |
84a12e66 CH |
733 | } |
734 | ||
e1d7f8bb DHB |
735 | int coroutine_fn bdrv_co_delete_file(BlockDriverState *bs, Error **errp) |
736 | { | |
737 | Error *local_err = NULL; | |
738 | int ret; | |
739 | ||
384a48fb | 740 | IO_CODE(); |
e1d7f8bb DHB |
741 | assert(bs != NULL); |
742 | ||
743 | if (!bs->drv) { | |
744 | error_setg(errp, "Block node '%s' is not opened", bs->filename); | |
745 | return -ENOMEDIUM; | |
746 | } | |
747 | ||
748 | if (!bs->drv->bdrv_co_delete_file) { | |
749 | error_setg(errp, "Driver '%s' does not support image deletion", | |
750 | bs->drv->format_name); | |
751 | return -ENOTSUP; | |
752 | } | |
753 | ||
754 | ret = bs->drv->bdrv_co_delete_file(bs, &local_err); | |
755 | if (ret < 0) { | |
756 | error_propagate(errp, local_err); | |
757 | } | |
758 | ||
759 | return ret; | |
760 | } | |
761 | ||
a890f08e ML |
762 | void coroutine_fn bdrv_co_delete_file_noerr(BlockDriverState *bs) |
763 | { | |
764 | Error *local_err = NULL; | |
765 | int ret; | |
384a48fb | 766 | IO_CODE(); |
a890f08e ML |
767 | |
768 | if (!bs) { | |
769 | return; | |
770 | } | |
771 | ||
772 | ret = bdrv_co_delete_file(bs, &local_err); | |
773 | /* | |
774 | * ENOTSUP will happen if the block driver doesn't support | |
775 | * the 'bdrv_co_delete_file' interface. This is a predictable | |
776 | * scenario and shouldn't be reported back to the user. | |
777 | */ | |
778 | if (ret == -ENOTSUP) { | |
779 | error_free(local_err); | |
780 | } else if (ret < 0) { | |
781 | error_report_err(local_err); | |
782 | } | |
783 | } | |
784 | ||
892b7de8 ET |
785 | /** |
786 | * Try to get @bs's logical and physical block size. | |
787 | * On success, store them in @bsz struct and return 0. | |
788 | * On failure return -errno. | |
789 | * @bs must not be empty. | |
790 | */ | |
791 | int bdrv_probe_blocksizes(BlockDriverState *bs, BlockSizes *bsz) | |
792 | { | |
793 | BlockDriver *drv = bs->drv; | |
93393e69 | 794 | BlockDriverState *filtered = bdrv_filter_bs(bs); |
f791bf7f | 795 | GLOBAL_STATE_CODE(); |
892b7de8 ET |
796 | |
797 | if (drv && drv->bdrv_probe_blocksizes) { | |
798 | return drv->bdrv_probe_blocksizes(bs, bsz); | |
93393e69 HR |
799 | } else if (filtered) { |
800 | return bdrv_probe_blocksizes(filtered, bsz); | |
892b7de8 ET |
801 | } |
802 | ||
803 | return -ENOTSUP; | |
804 | } | |
805 | ||
806 | /** | |
807 | * Try to get @bs's geometry (cyls, heads, sectors). | |
808 | * On success, store them in @geo struct and return 0. | |
809 | * On failure return -errno. | |
810 | * @bs must not be empty. | |
811 | */ | |
812 | int bdrv_probe_geometry(BlockDriverState *bs, HDGeometry *geo) | |
813 | { | |
814 | BlockDriver *drv = bs->drv; | |
93393e69 | 815 | BlockDriverState *filtered = bdrv_filter_bs(bs); |
f791bf7f | 816 | GLOBAL_STATE_CODE(); |
892b7de8 ET |
817 | |
818 | if (drv && drv->bdrv_probe_geometry) { | |
819 | return drv->bdrv_probe_geometry(bs, geo); | |
93393e69 HR |
820 | } else if (filtered) { |
821 | return bdrv_probe_geometry(filtered, geo); | |
892b7de8 ET |
822 | } |
823 | ||
824 | return -ENOTSUP; | |
825 | } | |
826 | ||
eba25057 JM |
827 | /* |
828 | * Create a uniquely-named empty temporary file. | |
69fbfff9 BM |
829 | * Return the actual file name used upon success, otherwise NULL. |
830 | * This string should be freed with g_free() when not needed any longer. | |
831 | * | |
832 | * Note: creating a temporary file for the caller to (re)open is | |
833 | * inherently racy. Use g_file_open_tmp() instead whenever practical. | |
eba25057 | 834 | */ |
69fbfff9 | 835 | char *create_tmp_file(Error **errp) |
d5249393 | 836 | { |
67b915a5 | 837 | int fd; |
7ccfb2eb | 838 | const char *tmpdir; |
69fbfff9 BM |
839 | g_autofree char *filename = NULL; |
840 | ||
841 | tmpdir = g_get_tmp_dir(); | |
842 | #ifndef _WIN32 | |
843 | /* | |
844 | * See commit 69bef79 ("block: use /var/tmp instead of /tmp for -snapshot") | |
845 | * | |
846 | * This function is used to create temporary disk images (like -snapshot), | |
847 | * so the files can become very large. /tmp is often a tmpfs where as | |
848 | * /var/tmp is usually on a disk, so more appropriate for disk images. | |
849 | */ | |
850 | if (!g_strcmp0(tmpdir, "/tmp")) { | |
69bef793 AS |
851 | tmpdir = "/var/tmp"; |
852 | } | |
69fbfff9 BM |
853 | #endif |
854 | ||
855 | filename = g_strdup_printf("%s/vl.XXXXXX", tmpdir); | |
856 | fd = g_mkstemp(filename); | |
fe235a06 | 857 | if (fd < 0) { |
69fbfff9 BM |
858 | error_setg_errno(errp, errno, "Could not open temporary file '%s'", |
859 | filename); | |
860 | return NULL; | |
eba25057 | 861 | } |
6b6471ee | 862 | close(fd); |
69fbfff9 BM |
863 | |
864 | return g_steal_pointer(&filename); | |
eba25057 | 865 | } |
fc01f7e7 | 866 | |
84a12e66 CH |
867 | /* |
868 | * Detect host devices. By convention, /dev/cdrom[N] is always | |
869 | * recognized as a host CDROM. | |
870 | */ | |
871 | static BlockDriver *find_hdev_driver(const char *filename) | |
872 | { | |
873 | int score_max = 0, score; | |
874 | BlockDriver *drv = NULL, *d; | |
bdb73476 | 875 | GLOBAL_STATE_CODE(); |
84a12e66 CH |
876 | |
877 | QLIST_FOREACH(d, &bdrv_drivers, list) { | |
878 | if (d->bdrv_probe_device) { | |
879 | score = d->bdrv_probe_device(filename); | |
880 | if (score > score_max) { | |
881 | score_max = score; | |
882 | drv = d; | |
883 | } | |
884 | } | |
885 | } | |
886 | ||
887 | return drv; | |
888 | } | |
889 | ||
88d88798 MM |
890 | static BlockDriver *bdrv_do_find_protocol(const char *protocol) |
891 | { | |
892 | BlockDriver *drv1; | |
bdb73476 | 893 | GLOBAL_STATE_CODE(); |
88d88798 MM |
894 | |
895 | QLIST_FOREACH(drv1, &bdrv_drivers, list) { | |
896 | if (drv1->protocol_name && !strcmp(drv1->protocol_name, protocol)) { | |
897 | return drv1; | |
898 | } | |
899 | } | |
900 | ||
901 | return NULL; | |
902 | } | |
903 | ||
98289620 | 904 | BlockDriver *bdrv_find_protocol(const char *filename, |
b65a5e12 HR |
905 | bool allow_protocol_prefix, |
906 | Error **errp) | |
83f64091 FB |
907 | { |
908 | BlockDriver *drv1; | |
909 | char protocol[128]; | |
1cec71e3 | 910 | int len; |
83f64091 | 911 | const char *p; |
88d88798 | 912 | int i; |
19cb3738 | 913 | |
f791bf7f | 914 | GLOBAL_STATE_CODE(); |
66f82cee KW |
915 | /* TODO Drivers without bdrv_file_open must be specified explicitly */ |
916 | ||
39508e7a CH |
917 | /* |
918 | * XXX(hch): we really should not let host device detection | |
919 | * override an explicit protocol specification, but moving this | |
920 | * later breaks access to device names with colons in them. | |
921 | * Thanks to the brain-dead persistent naming schemes on udev- | |
922 | * based Linux systems those actually are quite common. | |
923 | */ | |
924 | drv1 = find_hdev_driver(filename); | |
925 | if (drv1) { | |
926 | return drv1; | |
927 | } | |
928 | ||
98289620 | 929 | if (!path_has_protocol(filename) || !allow_protocol_prefix) { |
ef810437 | 930 | return &bdrv_file; |
84a12e66 | 931 | } |
98289620 | 932 | |
9e0b22f4 SH |
933 | p = strchr(filename, ':'); |
934 | assert(p != NULL); | |
1cec71e3 AL |
935 | len = p - filename; |
936 | if (len > sizeof(protocol) - 1) | |
937 | len = sizeof(protocol) - 1; | |
938 | memcpy(protocol, filename, len); | |
939 | protocol[len] = '\0'; | |
88d88798 MM |
940 | |
941 | drv1 = bdrv_do_find_protocol(protocol); | |
942 | if (drv1) { | |
943 | return drv1; | |
944 | } | |
945 | ||
946 | for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); ++i) { | |
947 | if (block_driver_modules[i].protocol_name && | |
948 | !strcmp(block_driver_modules[i].protocol_name, protocol)) { | |
c551fb0b CF |
949 | int rv = block_module_load(block_driver_modules[i].library_name, errp); |
950 | if (rv > 0) { | |
951 | drv1 = bdrv_do_find_protocol(protocol); | |
952 | } else if (rv < 0) { | |
953 | return NULL; | |
954 | } | |
88d88798 | 955 | break; |
8a22f02a | 956 | } |
83f64091 | 957 | } |
b65a5e12 | 958 | |
88d88798 MM |
959 | if (!drv1) { |
960 | error_setg(errp, "Unknown protocol '%s'", protocol); | |
961 | } | |
962 | return drv1; | |
83f64091 FB |
963 | } |
964 | ||
c6684249 MA |
965 | /* |
966 | * Guess image format by probing its contents. | |
967 | * This is not a good idea when your image is raw (CVE-2008-2004), but | |
968 | * we do it anyway for backward compatibility. | |
969 | * | |
970 | * @buf contains the image's first @buf_size bytes. | |
7cddd372 KW |
971 | * @buf_size is the buffer size in bytes (generally BLOCK_PROBE_BUF_SIZE, |
972 | * but can be smaller if the image file is smaller) | |
c6684249 MA |
973 | * @filename is its filename. |
974 | * | |
975 | * For all block drivers, call the bdrv_probe() method to get its | |
976 | * probing score. | |
977 | * Return the first block driver with the highest probing score. | |
978 | */ | |
38f3ef57 KW |
979 | BlockDriver *bdrv_probe_all(const uint8_t *buf, int buf_size, |
980 | const char *filename) | |
c6684249 MA |
981 | { |
982 | int score_max = 0, score; | |
983 | BlockDriver *drv = NULL, *d; | |
967d7905 | 984 | IO_CODE(); |
c6684249 MA |
985 | |
986 | QLIST_FOREACH(d, &bdrv_drivers, list) { | |
987 | if (d->bdrv_probe) { | |
988 | score = d->bdrv_probe(buf, buf_size, filename); | |
989 | if (score > score_max) { | |
990 | score_max = score; | |
991 | drv = d; | |
992 | } | |
993 | } | |
994 | } | |
995 | ||
996 | return drv; | |
997 | } | |
998 | ||
5696c6e3 | 999 | static int find_image_format(BlockBackend *file, const char *filename, |
34b5d2c6 | 1000 | BlockDriver **pdrv, Error **errp) |
f3a5d3f8 | 1001 | { |
c6684249 | 1002 | BlockDriver *drv; |
7cddd372 | 1003 | uint8_t buf[BLOCK_PROBE_BUF_SIZE]; |
f500a6d3 | 1004 | int ret = 0; |
f8ea0b00 | 1005 | |
bdb73476 EGE |
1006 | GLOBAL_STATE_CODE(); |
1007 | ||
08a00559 | 1008 | /* Return the raw BlockDriver * to scsi-generic devices or empty drives */ |
5696c6e3 | 1009 | if (blk_is_sg(file) || !blk_is_inserted(file) || blk_getlength(file) == 0) { |
ef810437 | 1010 | *pdrv = &bdrv_raw; |
c98ac35d | 1011 | return ret; |
1a396859 | 1012 | } |
f8ea0b00 | 1013 | |
a9262f55 | 1014 | ret = blk_pread(file, 0, sizeof(buf), buf, 0); |
83f64091 | 1015 | if (ret < 0) { |
34b5d2c6 HR |
1016 | error_setg_errno(errp, -ret, "Could not read image for determining its " |
1017 | "format"); | |
c98ac35d SW |
1018 | *pdrv = NULL; |
1019 | return ret; | |
83f64091 FB |
1020 | } |
1021 | ||
bf5b16fa | 1022 | drv = bdrv_probe_all(buf, sizeof(buf), filename); |
c98ac35d | 1023 | if (!drv) { |
34b5d2c6 HR |
1024 | error_setg(errp, "Could not determine image format: No compatible " |
1025 | "driver found"); | |
bf5b16fa AF |
1026 | *pdrv = NULL; |
1027 | return -ENOENT; | |
c98ac35d | 1028 | } |
bf5b16fa | 1029 | |
c98ac35d | 1030 | *pdrv = drv; |
bf5b16fa | 1031 | return 0; |
ea2384d3 FB |
1032 | } |
1033 | ||
51762288 SH |
1034 | /** |
1035 | * Set the current 'total_sectors' value | |
65a9bb25 | 1036 | * Return 0 on success, -errno on error. |
51762288 | 1037 | */ |
c86422c5 EGE |
1038 | int coroutine_fn bdrv_co_refresh_total_sectors(BlockDriverState *bs, |
1039 | int64_t hint) | |
51762288 SH |
1040 | { |
1041 | BlockDriver *drv = bs->drv; | |
967d7905 | 1042 | IO_CODE(); |
51762288 | 1043 | |
d470ad42 HR |
1044 | if (!drv) { |
1045 | return -ENOMEDIUM; | |
1046 | } | |
1047 | ||
c86422c5 | 1048 | /* Do not attempt drv->bdrv_co_getlength() on scsi-generic devices */ |
b192af8a | 1049 | if (bdrv_is_sg(bs)) |
396759ad NB |
1050 | return 0; |
1051 | ||
51762288 | 1052 | /* query actual device if possible, otherwise just trust the hint */ |
c86422c5 EGE |
1053 | if (drv->bdrv_co_getlength) { |
1054 | int64_t length = drv->bdrv_co_getlength(bs); | |
51762288 SH |
1055 | if (length < 0) { |
1056 | return length; | |
1057 | } | |
7e382003 | 1058 | hint = DIV_ROUND_UP(length, BDRV_SECTOR_SIZE); |
51762288 SH |
1059 | } |
1060 | ||
1061 | bs->total_sectors = hint; | |
8b117001 VSO |
1062 | |
1063 | if (bs->total_sectors * BDRV_SECTOR_SIZE > BDRV_MAX_LENGTH) { | |
1064 | return -EFBIG; | |
1065 | } | |
1066 | ||
51762288 SH |
1067 | return 0; |
1068 | } | |
1069 | ||
cddff5ba KW |
1070 | /** |
1071 | * Combines a QDict of new block driver @options with any missing options taken | |
1072 | * from @old_options, so that leaving out an option defaults to its old value. | |
1073 | */ | |
1074 | static void bdrv_join_options(BlockDriverState *bs, QDict *options, | |
1075 | QDict *old_options) | |
1076 | { | |
da359909 | 1077 | GLOBAL_STATE_CODE(); |
cddff5ba KW |
1078 | if (bs->drv && bs->drv->bdrv_join_options) { |
1079 | bs->drv->bdrv_join_options(options, old_options); | |
1080 | } else { | |
1081 | qdict_join(options, old_options, false); | |
1082 | } | |
1083 | } | |
1084 | ||
543770bd AG |
1085 | static BlockdevDetectZeroesOptions bdrv_parse_detect_zeroes(QemuOpts *opts, |
1086 | int open_flags, | |
1087 | Error **errp) | |
1088 | { | |
1089 | Error *local_err = NULL; | |
1090 | char *value = qemu_opt_get_del(opts, "detect-zeroes"); | |
1091 | BlockdevDetectZeroesOptions detect_zeroes = | |
1092 | qapi_enum_parse(&BlockdevDetectZeroesOptions_lookup, value, | |
1093 | BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF, &local_err); | |
bdb73476 | 1094 | GLOBAL_STATE_CODE(); |
543770bd AG |
1095 | g_free(value); |
1096 | if (local_err) { | |
1097 | error_propagate(errp, local_err); | |
1098 | return detect_zeroes; | |
1099 | } | |
1100 | ||
1101 | if (detect_zeroes == BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP && | |
1102 | !(open_flags & BDRV_O_UNMAP)) | |
1103 | { | |
1104 | error_setg(errp, "setting detect-zeroes to unmap is not allowed " | |
1105 | "without setting discard operation to unmap"); | |
1106 | } | |
1107 | ||
1108 | return detect_zeroes; | |
1109 | } | |
1110 | ||
f80f2673 AM |
1111 | /** |
1112 | * Set open flags for aio engine | |
1113 | * | |
1114 | * Return 0 on success, -1 if the engine specified is invalid | |
1115 | */ | |
1116 | int bdrv_parse_aio(const char *mode, int *flags) | |
1117 | { | |
1118 | if (!strcmp(mode, "threads")) { | |
1119 | /* do nothing, default */ | |
1120 | } else if (!strcmp(mode, "native")) { | |
1121 | *flags |= BDRV_O_NATIVE_AIO; | |
1122 | #ifdef CONFIG_LINUX_IO_URING | |
1123 | } else if (!strcmp(mode, "io_uring")) { | |
1124 | *flags |= BDRV_O_IO_URING; | |
1125 | #endif | |
1126 | } else { | |
1127 | return -1; | |
1128 | } | |
1129 | ||
1130 | return 0; | |
1131 | } | |
1132 | ||
9e8f1835 PB |
1133 | /** |
1134 | * Set open flags for a given discard mode | |
1135 | * | |
1136 | * Return 0 on success, -1 if the discard mode was invalid. | |
1137 | */ | |
1138 | int bdrv_parse_discard_flags(const char *mode, int *flags) | |
1139 | { | |
1140 | *flags &= ~BDRV_O_UNMAP; | |
1141 | ||
1142 | if (!strcmp(mode, "off") || !strcmp(mode, "ignore")) { | |
1143 | /* do nothing */ | |
1144 | } else if (!strcmp(mode, "on") || !strcmp(mode, "unmap")) { | |
1145 | *flags |= BDRV_O_UNMAP; | |
1146 | } else { | |
1147 | return -1; | |
1148 | } | |
1149 | ||
1150 | return 0; | |
1151 | } | |
1152 | ||
c3993cdc SH |
1153 | /** |
1154 | * Set open flags for a given cache mode | |
1155 | * | |
1156 | * Return 0 on success, -1 if the cache mode was invalid. | |
1157 | */ | |
53e8ae01 | 1158 | int bdrv_parse_cache_mode(const char *mode, int *flags, bool *writethrough) |
c3993cdc SH |
1159 | { |
1160 | *flags &= ~BDRV_O_CACHE_MASK; | |
1161 | ||
1162 | if (!strcmp(mode, "off") || !strcmp(mode, "none")) { | |
53e8ae01 KW |
1163 | *writethrough = false; |
1164 | *flags |= BDRV_O_NOCACHE; | |
92196b2f | 1165 | } else if (!strcmp(mode, "directsync")) { |
53e8ae01 | 1166 | *writethrough = true; |
92196b2f | 1167 | *flags |= BDRV_O_NOCACHE; |
c3993cdc | 1168 | } else if (!strcmp(mode, "writeback")) { |
53e8ae01 | 1169 | *writethrough = false; |
c3993cdc | 1170 | } else if (!strcmp(mode, "unsafe")) { |
53e8ae01 | 1171 | *writethrough = false; |
c3993cdc SH |
1172 | *flags |= BDRV_O_NO_FLUSH; |
1173 | } else if (!strcmp(mode, "writethrough")) { | |
53e8ae01 | 1174 | *writethrough = true; |
c3993cdc SH |
1175 | } else { |
1176 | return -1; | |
1177 | } | |
1178 | ||
1179 | return 0; | |
1180 | } | |
1181 | ||
b5411555 KW |
1182 | static char *bdrv_child_get_parent_desc(BdrvChild *c) |
1183 | { | |
1184 | BlockDriverState *parent = c->opaque; | |
2c0a3acb | 1185 | return g_strdup_printf("node '%s'", bdrv_get_node_name(parent)); |
b5411555 KW |
1186 | } |
1187 | ||
20018e12 KW |
1188 | static void bdrv_child_cb_drained_begin(BdrvChild *child) |
1189 | { | |
1190 | BlockDriverState *bs = child->opaque; | |
a82a3bd1 | 1191 | bdrv_do_drained_begin_quiesce(bs, NULL); |
20018e12 KW |
1192 | } |
1193 | ||
89bd0305 KW |
1194 | static bool bdrv_child_cb_drained_poll(BdrvChild *child) |
1195 | { | |
1196 | BlockDriverState *bs = child->opaque; | |
299403ae | 1197 | return bdrv_drain_poll(bs, NULL, false); |
89bd0305 KW |
1198 | } |
1199 | ||
2f65df6e | 1200 | static void bdrv_child_cb_drained_end(BdrvChild *child) |
20018e12 KW |
1201 | { |
1202 | BlockDriverState *bs = child->opaque; | |
2f65df6e | 1203 | bdrv_drained_end(bs); |
20018e12 KW |
1204 | } |
1205 | ||
38701b6a KW |
1206 | static int bdrv_child_cb_inactivate(BdrvChild *child) |
1207 | { | |
1208 | BlockDriverState *bs = child->opaque; | |
bdb73476 | 1209 | GLOBAL_STATE_CODE(); |
38701b6a KW |
1210 | assert(bs->open_flags & BDRV_O_INACTIVE); |
1211 | return 0; | |
1212 | } | |
1213 | ||
27633e74 EGE |
1214 | static bool bdrv_child_cb_change_aio_ctx(BdrvChild *child, AioContext *ctx, |
1215 | GHashTable *visited, Transaction *tran, | |
1216 | Error **errp) | |
53a7d041 KW |
1217 | { |
1218 | BlockDriverState *bs = child->opaque; | |
27633e74 | 1219 | return bdrv_change_aio_context(bs, ctx, visited, tran, errp); |
53a7d041 KW |
1220 | } |
1221 | ||
b1e6fc08 | 1222 | /* |
73176bee KW |
1223 | * Returns the options and flags that a temporary snapshot should get, based on |
1224 | * the originally requested flags (the originally requested image will have | |
1225 | * flags like a backing file) | |
b1e6fc08 | 1226 | */ |
73176bee KW |
1227 | static void bdrv_temp_snapshot_options(int *child_flags, QDict *child_options, |
1228 | int parent_flags, QDict *parent_options) | |
b1e6fc08 | 1229 | { |
bdb73476 | 1230 | GLOBAL_STATE_CODE(); |
73176bee KW |
1231 | *child_flags = (parent_flags & ~BDRV_O_SNAPSHOT) | BDRV_O_TEMPORARY; |
1232 | ||
1233 | /* For temporary files, unconditional cache=unsafe is fine */ | |
73176bee KW |
1234 | qdict_set_default_str(child_options, BDRV_OPT_CACHE_DIRECT, "off"); |
1235 | qdict_set_default_str(child_options, BDRV_OPT_CACHE_NO_FLUSH, "on"); | |
41869044 | 1236 | |
3f48686f | 1237 | /* Copy the read-only and discard options from the parent */ |
f87a0e29 | 1238 | qdict_copy_default(child_options, parent_options, BDRV_OPT_READ_ONLY); |
3f48686f | 1239 | qdict_copy_default(child_options, parent_options, BDRV_OPT_DISCARD); |
f87a0e29 | 1240 | |
41869044 KW |
1241 | /* aio=native doesn't work for cache.direct=off, so disable it for the |
1242 | * temporary snapshot */ | |
1243 | *child_flags &= ~BDRV_O_NATIVE_AIO; | |
b1e6fc08 KW |
1244 | } |
1245 | ||
db95dbba KW |
1246 | static void bdrv_backing_attach(BdrvChild *c) |
1247 | { | |
1248 | BlockDriverState *parent = c->opaque; | |
1249 | BlockDriverState *backing_hd = c->bs; | |
1250 | ||
bdb73476 | 1251 | GLOBAL_STATE_CODE(); |
db95dbba KW |
1252 | assert(!parent->backing_blocker); |
1253 | error_setg(&parent->backing_blocker, | |
1254 | "node is used as backing hd of '%s'", | |
1255 | bdrv_get_device_or_node_name(parent)); | |
1256 | ||
f30c66ba HR |
1257 | bdrv_refresh_filename(backing_hd); |
1258 | ||
db95dbba | 1259 | parent->open_flags &= ~BDRV_O_NO_BACKING; |
db95dbba KW |
1260 | |
1261 | bdrv_op_block_all(backing_hd, parent->backing_blocker); | |
1262 | /* Otherwise we won't be able to commit or stream */ | |
1263 | bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_COMMIT_TARGET, | |
1264 | parent->backing_blocker); | |
1265 | bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_STREAM, | |
1266 | parent->backing_blocker); | |
1267 | /* | |
1268 | * We do backup in 3 ways: | |
1269 | * 1. drive backup | |
1270 | * The target bs is new opened, and the source is top BDS | |
1271 | * 2. blockdev backup | |
1272 | * Both the source and the target are top BDSes. | |
1273 | * 3. internal backup(used for block replication) | |
1274 | * Both the source and the target are backing file | |
1275 | * | |
1276 | * In case 1 and 2, neither the source nor the target is the backing file. | |
1277 | * In case 3, we will block the top BDS, so there is only one block job | |
1278 | * for the top BDS and its backing chain. | |
1279 | */ | |
1280 | bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_BACKUP_SOURCE, | |
1281 | parent->backing_blocker); | |
1282 | bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_BACKUP_TARGET, | |
1283 | parent->backing_blocker); | |
ca2f1234 | 1284 | } |
d736f119 | 1285 | |
db95dbba KW |
1286 | static void bdrv_backing_detach(BdrvChild *c) |
1287 | { | |
1288 | BlockDriverState *parent = c->opaque; | |
1289 | ||
bdb73476 | 1290 | GLOBAL_STATE_CODE(); |
db95dbba KW |
1291 | assert(parent->backing_blocker); |
1292 | bdrv_op_unblock_all(c->bs, parent->backing_blocker); | |
1293 | error_free(parent->backing_blocker); | |
1294 | parent->backing_blocker = NULL; | |
48e08288 | 1295 | } |
d736f119 | 1296 | |
6858eba0 KW |
1297 | static int bdrv_backing_update_filename(BdrvChild *c, BlockDriverState *base, |
1298 | const char *filename, Error **errp) | |
1299 | { | |
1300 | BlockDriverState *parent = c->opaque; | |
e94d3dba | 1301 | bool read_only = bdrv_is_read_only(parent); |
6858eba0 | 1302 | int ret; |
bdb73476 | 1303 | GLOBAL_STATE_CODE(); |
6858eba0 | 1304 | |
e94d3dba AG |
1305 | if (read_only) { |
1306 | ret = bdrv_reopen_set_read_only(parent, false, errp); | |
61f09cea KW |
1307 | if (ret < 0) { |
1308 | return ret; | |
1309 | } | |
1310 | } | |
1311 | ||
6858eba0 | 1312 | ret = bdrv_change_backing_file(parent, filename, |
e54ee1b3 EB |
1313 | base->drv ? base->drv->format_name : "", |
1314 | false); | |
6858eba0 | 1315 | if (ret < 0) { |
64730694 | 1316 | error_setg_errno(errp, -ret, "Could not update backing file link"); |
6858eba0 KW |
1317 | } |
1318 | ||
e94d3dba AG |
1319 | if (read_only) { |
1320 | bdrv_reopen_set_read_only(parent, true, NULL); | |
61f09cea KW |
1321 | } |
1322 | ||
6858eba0 KW |
1323 | return ret; |
1324 | } | |
1325 | ||
fae8bd39 HR |
1326 | /* |
1327 | * Returns the options and flags that a generic child of a BDS should | |
1328 | * get, based on the given options and flags for the parent BDS. | |
1329 | */ | |
00ff7ffd HR |
1330 | static void bdrv_inherited_options(BdrvChildRole role, bool parent_is_format, |
1331 | int *child_flags, QDict *child_options, | |
1332 | int parent_flags, QDict *parent_options) | |
fae8bd39 HR |
1333 | { |
1334 | int flags = parent_flags; | |
bdb73476 | 1335 | GLOBAL_STATE_CODE(); |
fae8bd39 HR |
1336 | |
1337 | /* | |
1338 | * First, decide whether to set, clear, or leave BDRV_O_PROTOCOL. | |
1339 | * Generally, the question to answer is: Should this child be | |
1340 | * format-probed by default? | |
1341 | */ | |
1342 | ||
1343 | /* | |
1344 | * Pure and non-filtered data children of non-format nodes should | |
1345 | * be probed by default (even when the node itself has BDRV_O_PROTOCOL | |
1346 | * set). This only affects a very limited set of drivers (namely | |
1347 | * quorum and blkverify when this comment was written). | |
1348 | * Force-clear BDRV_O_PROTOCOL then. | |
1349 | */ | |
1350 | if (!parent_is_format && | |
1351 | (role & BDRV_CHILD_DATA) && | |
1352 | !(role & (BDRV_CHILD_METADATA | BDRV_CHILD_FILTERED))) | |
1353 | { | |
1354 | flags &= ~BDRV_O_PROTOCOL; | |
1355 | } | |
1356 | ||
1357 | /* | |
1358 | * All children of format nodes (except for COW children) and all | |
1359 | * metadata children in general should never be format-probed. | |
1360 | * Force-set BDRV_O_PROTOCOL then. | |
1361 | */ | |
1362 | if ((parent_is_format && !(role & BDRV_CHILD_COW)) || | |
1363 | (role & BDRV_CHILD_METADATA)) | |
1364 | { | |
1365 | flags |= BDRV_O_PROTOCOL; | |
1366 | } | |
1367 | ||
1368 | /* | |
1369 | * If the cache mode isn't explicitly set, inherit direct and no-flush from | |
1370 | * the parent. | |
1371 | */ | |
1372 | qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_DIRECT); | |
1373 | qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_NO_FLUSH); | |
1374 | qdict_copy_default(child_options, parent_options, BDRV_OPT_FORCE_SHARE); | |
1375 | ||
1376 | if (role & BDRV_CHILD_COW) { | |
1377 | /* backing files are opened read-only by default */ | |
1378 | qdict_set_default_str(child_options, BDRV_OPT_READ_ONLY, "on"); | |
1379 | qdict_set_default_str(child_options, BDRV_OPT_AUTO_READ_ONLY, "off"); | |
1380 | } else { | |
1381 | /* Inherit the read-only option from the parent if it's not set */ | |
1382 | qdict_copy_default(child_options, parent_options, BDRV_OPT_READ_ONLY); | |
1383 | qdict_copy_default(child_options, parent_options, | |
1384 | BDRV_OPT_AUTO_READ_ONLY); | |
1385 | } | |
1386 | ||
1387 | /* | |
1388 | * bdrv_co_pdiscard() respects unmap policy for the parent, so we | |
1389 | * can default to enable it on lower layers regardless of the | |
1390 | * parent option. | |
1391 | */ | |
1392 | qdict_set_default_str(child_options, BDRV_OPT_DISCARD, "unmap"); | |
1393 | ||
1394 | /* Clear flags that only apply to the top layer */ | |
1395 | flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_COPY_ON_READ); | |
1396 | ||
1397 | if (role & BDRV_CHILD_METADATA) { | |
1398 | flags &= ~BDRV_O_NO_IO; | |
1399 | } | |
1400 | if (role & BDRV_CHILD_COW) { | |
1401 | flags &= ~BDRV_O_TEMPORARY; | |
1402 | } | |
1403 | ||
1404 | *child_flags = flags; | |
1405 | } | |
1406 | ||
303de47b | 1407 | static void GRAPH_WRLOCK bdrv_child_cb_attach(BdrvChild *child) |
ca2f1234 HR |
1408 | { |
1409 | BlockDriverState *bs = child->opaque; | |
1410 | ||
3f35f82e | 1411 | assert_bdrv_graph_writable(); |
a225369b | 1412 | QLIST_INSERT_HEAD(&bs->children, child, next); |
5bb04747 VSO |
1413 | if (bs->drv->is_filter || (child->role & BDRV_CHILD_FILTERED)) { |
1414 | /* | |
1415 | * Here we handle filters and block/raw-format.c when it behave like | |
1416 | * filter. They generally have a single PRIMARY child, which is also the | |
1417 | * FILTERED child, and that they may have multiple more children, which | |
1418 | * are neither PRIMARY nor FILTERED. And never we have a COW child here. | |
1419 | * So bs->file will be the PRIMARY child, unless the PRIMARY child goes | |
1420 | * into bs->backing on exceptional cases; and bs->backing will be | |
1421 | * nothing else. | |
1422 | */ | |
1423 | assert(!(child->role & BDRV_CHILD_COW)); | |
1424 | if (child->role & BDRV_CHILD_PRIMARY) { | |
1425 | assert(child->role & BDRV_CHILD_FILTERED); | |
1426 | assert(!bs->backing); | |
1427 | assert(!bs->file); | |
1428 | ||
1429 | if (bs->drv->filtered_child_is_backing) { | |
1430 | bs->backing = child; | |
1431 | } else { | |
1432 | bs->file = child; | |
1433 | } | |
1434 | } else { | |
1435 | assert(!(child->role & BDRV_CHILD_FILTERED)); | |
1436 | } | |
1437 | } else if (child->role & BDRV_CHILD_COW) { | |
1438 | assert(bs->drv->supports_backing); | |
1439 | assert(!(child->role & BDRV_CHILD_PRIMARY)); | |
1440 | assert(!bs->backing); | |
1441 | bs->backing = child; | |
ca2f1234 | 1442 | bdrv_backing_attach(child); |
5bb04747 VSO |
1443 | } else if (child->role & BDRV_CHILD_PRIMARY) { |
1444 | assert(!bs->file); | |
1445 | bs->file = child; | |
ca2f1234 | 1446 | } |
ca2f1234 HR |
1447 | } |
1448 | ||
303de47b | 1449 | static void GRAPH_WRLOCK bdrv_child_cb_detach(BdrvChild *child) |
48e08288 HR |
1450 | { |
1451 | BlockDriverState *bs = child->opaque; | |
1452 | ||
1453 | if (child->role & BDRV_CHILD_COW) { | |
1454 | bdrv_backing_detach(child); | |
1455 | } | |
1456 | ||
3f35f82e | 1457 | assert_bdrv_graph_writable(); |
a225369b | 1458 | QLIST_REMOVE(child, next); |
5bb04747 VSO |
1459 | if (child == bs->backing) { |
1460 | assert(child != bs->file); | |
1461 | bs->backing = NULL; | |
1462 | } else if (child == bs->file) { | |
1463 | bs->file = NULL; | |
1464 | } | |
48e08288 HR |
1465 | } |
1466 | ||
43483550 HR |
1467 | static int bdrv_child_cb_update_filename(BdrvChild *c, BlockDriverState *base, |
1468 | const char *filename, Error **errp) | |
1469 | { | |
1470 | if (c->role & BDRV_CHILD_COW) { | |
1471 | return bdrv_backing_update_filename(c, base, filename, errp); | |
1472 | } | |
1473 | return 0; | |
1474 | } | |
1475 | ||
fb62b588 | 1476 | AioContext *child_of_bds_get_parent_aio_context(BdrvChild *c) |
3ca1f322 VSO |
1477 | { |
1478 | BlockDriverState *bs = c->opaque; | |
384a48fb | 1479 | IO_CODE(); |
3ca1f322 VSO |
1480 | |
1481 | return bdrv_get_aio_context(bs); | |
1482 | } | |
1483 | ||
43483550 HR |
1484 | const BdrvChildClass child_of_bds = { |
1485 | .parent_is_bds = true, | |
1486 | .get_parent_desc = bdrv_child_get_parent_desc, | |
1487 | .inherit_options = bdrv_inherited_options, | |
1488 | .drained_begin = bdrv_child_cb_drained_begin, | |
1489 | .drained_poll = bdrv_child_cb_drained_poll, | |
1490 | .drained_end = bdrv_child_cb_drained_end, | |
1491 | .attach = bdrv_child_cb_attach, | |
1492 | .detach = bdrv_child_cb_detach, | |
1493 | .inactivate = bdrv_child_cb_inactivate, | |
27633e74 | 1494 | .change_aio_ctx = bdrv_child_cb_change_aio_ctx, |
43483550 | 1495 | .update_filename = bdrv_child_cb_update_filename, |
fb62b588 | 1496 | .get_parent_aio_context = child_of_bds_get_parent_aio_context, |
43483550 HR |
1497 | }; |
1498 | ||
3ca1f322 VSO |
1499 | AioContext *bdrv_child_get_parent_aio_context(BdrvChild *c) |
1500 | { | |
d5f8d79c | 1501 | IO_CODE(); |
3ca1f322 VSO |
1502 | return c->klass->get_parent_aio_context(c); |
1503 | } | |
1504 | ||
7b272452 KW |
1505 | static int bdrv_open_flags(BlockDriverState *bs, int flags) |
1506 | { | |
61de4c68 | 1507 | int open_flags = flags; |
bdb73476 | 1508 | GLOBAL_STATE_CODE(); |
7b272452 KW |
1509 | |
1510 | /* | |
1511 | * Clear flags that are internal to the block layer before opening the | |
1512 | * image. | |
1513 | */ | |
20cca275 | 1514 | open_flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_PROTOCOL); |
7b272452 | 1515 | |
7b272452 KW |
1516 | return open_flags; |
1517 | } | |
1518 | ||
91a097e7 KW |
1519 | static void update_flags_from_options(int *flags, QemuOpts *opts) |
1520 | { | |
bdb73476 EGE |
1521 | GLOBAL_STATE_CODE(); |
1522 | ||
2a3d4331 | 1523 | *flags &= ~(BDRV_O_CACHE_MASK | BDRV_O_RDWR | BDRV_O_AUTO_RDONLY); |
91a097e7 | 1524 | |
57f9db9a | 1525 | if (qemu_opt_get_bool_del(opts, BDRV_OPT_CACHE_NO_FLUSH, false)) { |
91a097e7 KW |
1526 | *flags |= BDRV_O_NO_FLUSH; |
1527 | } | |
1528 | ||
57f9db9a | 1529 | if (qemu_opt_get_bool_del(opts, BDRV_OPT_CACHE_DIRECT, false)) { |
91a097e7 KW |
1530 | *flags |= BDRV_O_NOCACHE; |
1531 | } | |
f87a0e29 | 1532 | |
57f9db9a | 1533 | if (!qemu_opt_get_bool_del(opts, BDRV_OPT_READ_ONLY, false)) { |
f87a0e29 AG |
1534 | *flags |= BDRV_O_RDWR; |
1535 | } | |
1536 | ||
e35bdc12 KW |
1537 | if (qemu_opt_get_bool_del(opts, BDRV_OPT_AUTO_READ_ONLY, false)) { |
1538 | *flags |= BDRV_O_AUTO_RDONLY; | |
1539 | } | |
91a097e7 KW |
1540 | } |
1541 | ||
1542 | static void update_options_from_flags(QDict *options, int flags) | |
1543 | { | |
bdb73476 | 1544 | GLOBAL_STATE_CODE(); |
91a097e7 | 1545 | if (!qdict_haskey(options, BDRV_OPT_CACHE_DIRECT)) { |
46f5ac20 | 1546 | qdict_put_bool(options, BDRV_OPT_CACHE_DIRECT, flags & BDRV_O_NOCACHE); |
91a097e7 KW |
1547 | } |
1548 | if (!qdict_haskey(options, BDRV_OPT_CACHE_NO_FLUSH)) { | |
46f5ac20 EB |
1549 | qdict_put_bool(options, BDRV_OPT_CACHE_NO_FLUSH, |
1550 | flags & BDRV_O_NO_FLUSH); | |
91a097e7 | 1551 | } |
f87a0e29 | 1552 | if (!qdict_haskey(options, BDRV_OPT_READ_ONLY)) { |
46f5ac20 | 1553 | qdict_put_bool(options, BDRV_OPT_READ_ONLY, !(flags & BDRV_O_RDWR)); |
f87a0e29 | 1554 | } |
e35bdc12 KW |
1555 | if (!qdict_haskey(options, BDRV_OPT_AUTO_READ_ONLY)) { |
1556 | qdict_put_bool(options, BDRV_OPT_AUTO_READ_ONLY, | |
1557 | flags & BDRV_O_AUTO_RDONLY); | |
1558 | } | |
91a097e7 KW |
1559 | } |
1560 | ||
636ea370 KW |
1561 | static void bdrv_assign_node_name(BlockDriverState *bs, |
1562 | const char *node_name, | |
1563 | Error **errp) | |
6913c0c2 | 1564 | { |
15489c76 | 1565 | char *gen_node_name = NULL; |
bdb73476 | 1566 | GLOBAL_STATE_CODE(); |
6913c0c2 | 1567 | |
15489c76 JC |
1568 | if (!node_name) { |
1569 | node_name = gen_node_name = id_generate(ID_BLOCK); | |
1570 | } else if (!id_wellformed(node_name)) { | |
1571 | /* | |
1572 | * Check for empty string or invalid characters, but not if it is | |
1573 | * generated (generated names use characters not available to the user) | |
1574 | */ | |
785ec4b1 | 1575 | error_setg(errp, "Invalid node-name: '%s'", node_name); |
636ea370 | 1576 | return; |
6913c0c2 BC |
1577 | } |
1578 | ||
0c5e94ee | 1579 | /* takes care of avoiding namespaces collisions */ |
7f06d47e | 1580 | if (blk_by_name(node_name)) { |
0c5e94ee BC |
1581 | error_setg(errp, "node-name=%s is conflicting with a device id", |
1582 | node_name); | |
15489c76 | 1583 | goto out; |
0c5e94ee BC |
1584 | } |
1585 | ||
6913c0c2 BC |
1586 | /* takes care of avoiding duplicates node names */ |
1587 | if (bdrv_find_node(node_name)) { | |
785ec4b1 | 1588 | error_setg(errp, "Duplicate nodes with node-name='%s'", node_name); |
15489c76 | 1589 | goto out; |
6913c0c2 BC |
1590 | } |
1591 | ||
824808dd KW |
1592 | /* Make sure that the node name isn't truncated */ |
1593 | if (strlen(node_name) >= sizeof(bs->node_name)) { | |
1594 | error_setg(errp, "Node name too long"); | |
1595 | goto out; | |
1596 | } | |
1597 | ||
6913c0c2 BC |
1598 | /* copy node name into the bs and insert it into the graph list */ |
1599 | pstrcpy(bs->node_name, sizeof(bs->node_name), node_name); | |
1600 | QTAILQ_INSERT_TAIL(&graph_bdrv_states, bs, node_list); | |
15489c76 JC |
1601 | out: |
1602 | g_free(gen_node_name); | |
6913c0c2 BC |
1603 | } |
1604 | ||
c86422c5 EGE |
1605 | /* |
1606 | * The caller must always hold @bs AioContext lock, because this function calls | |
1607 | * bdrv_refresh_total_sectors() which polls when called from non-coroutine | |
1608 | * context. | |
1609 | */ | |
01a56501 KW |
1610 | static int bdrv_open_driver(BlockDriverState *bs, BlockDriver *drv, |
1611 | const char *node_name, QDict *options, | |
1612 | int open_flags, Error **errp) | |
1613 | { | |
1614 | Error *local_err = NULL; | |
0f12264e | 1615 | int i, ret; |
da359909 | 1616 | GLOBAL_STATE_CODE(); |
01a56501 KW |
1617 | |
1618 | bdrv_assign_node_name(bs, node_name, &local_err); | |
1619 | if (local_err) { | |
1620 | error_propagate(errp, local_err); | |
1621 | return -EINVAL; | |
1622 | } | |
1623 | ||
1624 | bs->drv = drv; | |
1625 | bs->opaque = g_malloc0(drv->instance_size); | |
1626 | ||
1627 | if (drv->bdrv_file_open) { | |
1628 | assert(!drv->bdrv_needs_filename || bs->filename[0]); | |
1629 | ret = drv->bdrv_file_open(bs, options, open_flags, &local_err); | |
680c7f96 | 1630 | } else if (drv->bdrv_open) { |
01a56501 | 1631 | ret = drv->bdrv_open(bs, options, open_flags, &local_err); |
680c7f96 KW |
1632 | } else { |
1633 | ret = 0; | |
01a56501 KW |
1634 | } |
1635 | ||
1636 | if (ret < 0) { | |
1637 | if (local_err) { | |
1638 | error_propagate(errp, local_err); | |
1639 | } else if (bs->filename[0]) { | |
1640 | error_setg_errno(errp, -ret, "Could not open '%s'", bs->filename); | |
1641 | } else { | |
1642 | error_setg_errno(errp, -ret, "Could not open image"); | |
1643 | } | |
180ca19a | 1644 | goto open_failed; |
01a56501 KW |
1645 | } |
1646 | ||
e8b65355 SH |
1647 | assert(!(bs->supported_read_flags & ~BDRV_REQ_MASK)); |
1648 | assert(!(bs->supported_write_flags & ~BDRV_REQ_MASK)); | |
1649 | ||
1650 | /* | |
1651 | * Always allow the BDRV_REQ_REGISTERED_BUF optimization hint. This saves | |
1652 | * drivers that pass read/write requests through to a child the trouble of | |
1653 | * declaring support explicitly. | |
1654 | * | |
1655 | * Drivers must not propagate this flag accidentally when they initiate I/O | |
1656 | * to a bounce buffer. That case should be rare though. | |
1657 | */ | |
1658 | bs->supported_read_flags |= BDRV_REQ_REGISTERED_BUF; | |
1659 | bs->supported_write_flags |= BDRV_REQ_REGISTERED_BUF; | |
1660 | ||
c057960c | 1661 | ret = bdrv_refresh_total_sectors(bs, bs->total_sectors); |
01a56501 KW |
1662 | if (ret < 0) { |
1663 | error_setg_errno(errp, -ret, "Could not refresh total sector count"); | |
180ca19a | 1664 | return ret; |
01a56501 KW |
1665 | } |
1666 | ||
1e4c797c | 1667 | bdrv_refresh_limits(bs, NULL, &local_err); |
01a56501 KW |
1668 | if (local_err) { |
1669 | error_propagate(errp, local_err); | |
180ca19a | 1670 | return -EINVAL; |
01a56501 KW |
1671 | } |
1672 | ||
1673 | assert(bdrv_opt_mem_align(bs) != 0); | |
1674 | assert(bdrv_min_mem_align(bs) != 0); | |
1675 | assert(is_power_of_2(bs->bl.request_alignment)); | |
1676 | ||
0f12264e | 1677 | for (i = 0; i < bs->quiesce_counter; i++) { |
5e8ac217 KW |
1678 | if (drv->bdrv_drain_begin) { |
1679 | drv->bdrv_drain_begin(bs); | |
0f12264e KW |
1680 | } |
1681 | } | |
1682 | ||
01a56501 | 1683 | return 0; |
180ca19a MP |
1684 | open_failed: |
1685 | bs->drv = NULL; | |
1686 | if (bs->file != NULL) { | |
1687 | bdrv_unref_child(bs, bs->file); | |
5bb04747 | 1688 | assert(!bs->file); |
180ca19a | 1689 | } |
01a56501 KW |
1690 | g_free(bs->opaque); |
1691 | bs->opaque = NULL; | |
01a56501 KW |
1692 | return ret; |
1693 | } | |
1694 | ||
621d1737 VSO |
1695 | /* |
1696 | * Create and open a block node. | |
1697 | * | |
1698 | * @options is a QDict of options to pass to the block drivers, or NULL for an | |
1699 | * empty set of options. The reference to the QDict belongs to the block layer | |
1700 | * after the call (even on failure), so if the caller intends to reuse the | |
1701 | * dictionary, it needs to use qobject_ref() before calling bdrv_open. | |
1702 | */ | |
1703 | BlockDriverState *bdrv_new_open_driver_opts(BlockDriver *drv, | |
1704 | const char *node_name, | |
1705 | QDict *options, int flags, | |
1706 | Error **errp) | |
680c7f96 KW |
1707 | { |
1708 | BlockDriverState *bs; | |
1709 | int ret; | |
1710 | ||
f791bf7f EGE |
1711 | GLOBAL_STATE_CODE(); |
1712 | ||
680c7f96 KW |
1713 | bs = bdrv_new(); |
1714 | bs->open_flags = flags; | |
621d1737 VSO |
1715 | bs->options = options ?: qdict_new(); |
1716 | bs->explicit_options = qdict_clone_shallow(bs->options); | |
680c7f96 KW |
1717 | bs->opaque = NULL; |
1718 | ||
1719 | update_options_from_flags(bs->options, flags); | |
1720 | ||
1721 | ret = bdrv_open_driver(bs, drv, node_name, bs->options, flags, errp); | |
1722 | if (ret < 0) { | |
cb3e7f08 | 1723 | qobject_unref(bs->explicit_options); |
180ca19a | 1724 | bs->explicit_options = NULL; |
cb3e7f08 | 1725 | qobject_unref(bs->options); |
180ca19a | 1726 | bs->options = NULL; |
680c7f96 KW |
1727 | bdrv_unref(bs); |
1728 | return NULL; | |
1729 | } | |
1730 | ||
1731 | return bs; | |
1732 | } | |
1733 | ||
621d1737 VSO |
1734 | /* Create and open a block node. */ |
1735 | BlockDriverState *bdrv_new_open_driver(BlockDriver *drv, const char *node_name, | |
1736 | int flags, Error **errp) | |
1737 | { | |
f791bf7f | 1738 | GLOBAL_STATE_CODE(); |
621d1737 VSO |
1739 | return bdrv_new_open_driver_opts(drv, node_name, NULL, flags, errp); |
1740 | } | |
1741 | ||
c5f3014b | 1742 | QemuOptsList bdrv_runtime_opts = { |
18edf289 KW |
1743 | .name = "bdrv_common", |
1744 | .head = QTAILQ_HEAD_INITIALIZER(bdrv_runtime_opts.head), | |
1745 | .desc = { | |
1746 | { | |
1747 | .name = "node-name", | |
1748 | .type = QEMU_OPT_STRING, | |
1749 | .help = "Node name of the block device node", | |
1750 | }, | |
62392ebb KW |
1751 | { |
1752 | .name = "driver", | |
1753 | .type = QEMU_OPT_STRING, | |
1754 | .help = "Block driver to use for the node", | |
1755 | }, | |
91a097e7 KW |
1756 | { |
1757 | .name = BDRV_OPT_CACHE_DIRECT, | |
1758 | .type = QEMU_OPT_BOOL, | |
1759 | .help = "Bypass software writeback cache on the host", | |
1760 | }, | |
1761 | { | |
1762 | .name = BDRV_OPT_CACHE_NO_FLUSH, | |
1763 | .type = QEMU_OPT_BOOL, | |
1764 | .help = "Ignore flush requests", | |
1765 | }, | |
f87a0e29 AG |
1766 | { |
1767 | .name = BDRV_OPT_READ_ONLY, | |
1768 | .type = QEMU_OPT_BOOL, | |
1769 | .help = "Node is opened in read-only mode", | |
1770 | }, | |
e35bdc12 KW |
1771 | { |
1772 | .name = BDRV_OPT_AUTO_READ_ONLY, | |
1773 | .type = QEMU_OPT_BOOL, | |
1774 | .help = "Node can become read-only if opening read-write fails", | |
1775 | }, | |
692e01a2 KW |
1776 | { |
1777 | .name = "detect-zeroes", | |
1778 | .type = QEMU_OPT_STRING, | |
1779 | .help = "try to optimize zero writes (off, on, unmap)", | |
1780 | }, | |
818584a4 | 1781 | { |
415bbca8 | 1782 | .name = BDRV_OPT_DISCARD, |
818584a4 KW |
1783 | .type = QEMU_OPT_STRING, |
1784 | .help = "discard operation (ignore/off, unmap/on)", | |
1785 | }, | |
5a9347c6 FZ |
1786 | { |
1787 | .name = BDRV_OPT_FORCE_SHARE, | |
1788 | .type = QEMU_OPT_BOOL, | |
1789 | .help = "always accept other writers (default: off)", | |
1790 | }, | |
18edf289 KW |
1791 | { /* end of list */ } |
1792 | }, | |
1793 | }; | |
1794 | ||
5a5e7f8c ML |
1795 | QemuOptsList bdrv_create_opts_simple = { |
1796 | .name = "simple-create-opts", | |
1797 | .head = QTAILQ_HEAD_INITIALIZER(bdrv_create_opts_simple.head), | |
fd17146c HR |
1798 | .desc = { |
1799 | { | |
1800 | .name = BLOCK_OPT_SIZE, | |
1801 | .type = QEMU_OPT_SIZE, | |
1802 | .help = "Virtual disk size" | |
1803 | }, | |
1804 | { | |
1805 | .name = BLOCK_OPT_PREALLOC, | |
1806 | .type = QEMU_OPT_STRING, | |
1807 | .help = "Preallocation mode (allowed values: off)" | |
1808 | }, | |
1809 | { /* end of list */ } | |
1810 | } | |
1811 | }; | |
1812 | ||
57915332 KW |
1813 | /* |
1814 | * Common part for opening disk images and files | |
b6ad491a KW |
1815 | * |
1816 | * Removes all processed options from *options. | |
57915332 | 1817 | */ |
5696c6e3 | 1818 | static int bdrv_open_common(BlockDriverState *bs, BlockBackend *file, |
82dc8b41 | 1819 | QDict *options, Error **errp) |
57915332 KW |
1820 | { |
1821 | int ret, open_flags; | |
035fccdf | 1822 | const char *filename; |
62392ebb | 1823 | const char *driver_name = NULL; |
6913c0c2 | 1824 | const char *node_name = NULL; |
818584a4 | 1825 | const char *discard; |
18edf289 | 1826 | QemuOpts *opts; |
62392ebb | 1827 | BlockDriver *drv; |
34b5d2c6 | 1828 | Error *local_err = NULL; |
307261b2 | 1829 | bool ro; |
57915332 | 1830 | |
6405875c | 1831 | assert(bs->file == NULL); |
707ff828 | 1832 | assert(options != NULL && bs->options != options); |
bdb73476 | 1833 | GLOBAL_STATE_CODE(); |
57915332 | 1834 | |
62392ebb | 1835 | opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort); |
af175e85 | 1836 | if (!qemu_opts_absorb_qdict(opts, options, errp)) { |
62392ebb KW |
1837 | ret = -EINVAL; |
1838 | goto fail_opts; | |
1839 | } | |
1840 | ||
9b7e8691 AG |
1841 | update_flags_from_options(&bs->open_flags, opts); |
1842 | ||
62392ebb KW |
1843 | driver_name = qemu_opt_get(opts, "driver"); |
1844 | drv = bdrv_find_format(driver_name); | |
1845 | assert(drv != NULL); | |
1846 | ||
5a9347c6 FZ |
1847 | bs->force_share = qemu_opt_get_bool(opts, BDRV_OPT_FORCE_SHARE, false); |
1848 | ||
1849 | if (bs->force_share && (bs->open_flags & BDRV_O_RDWR)) { | |
1850 | error_setg(errp, | |
1851 | BDRV_OPT_FORCE_SHARE | |
1852 | "=on can only be used with read-only images"); | |
1853 | ret = -EINVAL; | |
1854 | goto fail_opts; | |
1855 | } | |
1856 | ||
45673671 | 1857 | if (file != NULL) { |
f30c66ba | 1858 | bdrv_refresh_filename(blk_bs(file)); |
5696c6e3 | 1859 | filename = blk_bs(file)->filename; |
45673671 | 1860 | } else { |
129c7d1c MA |
1861 | /* |
1862 | * Caution: while qdict_get_try_str() is fine, getting | |
1863 | * non-string types would require more care. When @options | |
1864 | * come from -blockdev or blockdev_add, its members are typed | |
1865 | * according to the QAPI schema, but when they come from | |
1866 | * -drive, they're all QString. | |
1867 | */ | |
45673671 KW |
1868 | filename = qdict_get_try_str(options, "filename"); |
1869 | } | |
1870 | ||
4a008240 | 1871 | if (drv->bdrv_needs_filename && (!filename || !filename[0])) { |
765003db KW |
1872 | error_setg(errp, "The '%s' block driver requires a file name", |
1873 | drv->format_name); | |
18edf289 KW |
1874 | ret = -EINVAL; |
1875 | goto fail_opts; | |
6913c0c2 | 1876 | } |
6913c0c2 | 1877 | |
82dc8b41 KW |
1878 | trace_bdrv_open_common(bs, filename ?: "", bs->open_flags, |
1879 | drv->format_name); | |
62392ebb | 1880 | |
307261b2 VSO |
1881 | ro = bdrv_is_read_only(bs); |
1882 | ||
1883 | if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv, ro)) { | |
1884 | if (!ro && bdrv_is_whitelisted(drv, true)) { | |
8be25de6 KW |
1885 | ret = bdrv_apply_auto_read_only(bs, NULL, NULL); |
1886 | } else { | |
1887 | ret = -ENOTSUP; | |
1888 | } | |
1889 | if (ret < 0) { | |
1890 | error_setg(errp, | |
307261b2 | 1891 | !ro && bdrv_is_whitelisted(drv, true) |
8be25de6 KW |
1892 | ? "Driver '%s' can only be used for read-only devices" |
1893 | : "Driver '%s' is not whitelisted", | |
1894 | drv->format_name); | |
1895 | goto fail_opts; | |
1896 | } | |
b64ec4e4 | 1897 | } |
57915332 | 1898 | |
d3faa13e | 1899 | /* bdrv_new() and bdrv_close() make it so */ |
d73415a3 | 1900 | assert(qatomic_read(&bs->copy_on_read) == 0); |
d3faa13e | 1901 | |
82dc8b41 | 1902 | if (bs->open_flags & BDRV_O_COPY_ON_READ) { |
307261b2 | 1903 | if (!ro) { |
0ebd24e0 KW |
1904 | bdrv_enable_copy_on_read(bs); |
1905 | } else { | |
1906 | error_setg(errp, "Can't use copy-on-read on read-only device"); | |
18edf289 KW |
1907 | ret = -EINVAL; |
1908 | goto fail_opts; | |
0ebd24e0 | 1909 | } |
53fec9d3 SH |
1910 | } |
1911 | ||
415bbca8 | 1912 | discard = qemu_opt_get(opts, BDRV_OPT_DISCARD); |
818584a4 KW |
1913 | if (discard != NULL) { |
1914 | if (bdrv_parse_discard_flags(discard, &bs->open_flags) != 0) { | |
1915 | error_setg(errp, "Invalid discard option"); | |
1916 | ret = -EINVAL; | |
1917 | goto fail_opts; | |
1918 | } | |
1919 | } | |
1920 | ||
543770bd AG |
1921 | bs->detect_zeroes = |
1922 | bdrv_parse_detect_zeroes(opts, bs->open_flags, &local_err); | |
1923 | if (local_err) { | |
1924 | error_propagate(errp, local_err); | |
1925 | ret = -EINVAL; | |
1926 | goto fail_opts; | |
692e01a2 KW |
1927 | } |
1928 | ||
c2ad1b0c KW |
1929 | if (filename != NULL) { |
1930 | pstrcpy(bs->filename, sizeof(bs->filename), filename); | |
1931 | } else { | |
1932 | bs->filename[0] = '\0'; | |
1933 | } | |
91af7014 | 1934 | pstrcpy(bs->exact_filename, sizeof(bs->exact_filename), bs->filename); |
57915332 | 1935 | |
66f82cee | 1936 | /* Open the image, either directly or using a protocol */ |
82dc8b41 | 1937 | open_flags = bdrv_open_flags(bs, bs->open_flags); |
01a56501 | 1938 | node_name = qemu_opt_get(opts, "node-name"); |
57915332 | 1939 | |
01a56501 KW |
1940 | assert(!drv->bdrv_file_open || file == NULL); |
1941 | ret = bdrv_open_driver(bs, drv, node_name, options, open_flags, errp); | |
51762288 | 1942 | if (ret < 0) { |
01a56501 | 1943 | goto fail_opts; |
3baca891 KW |
1944 | } |
1945 | ||
18edf289 | 1946 | qemu_opts_del(opts); |
57915332 KW |
1947 | return 0; |
1948 | ||
18edf289 KW |
1949 | fail_opts: |
1950 | qemu_opts_del(opts); | |
57915332 KW |
1951 | return ret; |
1952 | } | |
1953 | ||
5e5c4f63 KW |
1954 | static QDict *parse_json_filename(const char *filename, Error **errp) |
1955 | { | |
1956 | QObject *options_obj; | |
1957 | QDict *options; | |
1958 | int ret; | |
bdb73476 | 1959 | GLOBAL_STATE_CODE(); |
5e5c4f63 KW |
1960 | |
1961 | ret = strstart(filename, "json:", &filename); | |
1962 | assert(ret); | |
1963 | ||
5577fff7 | 1964 | options_obj = qobject_from_json(filename, errp); |
5e5c4f63 | 1965 | if (!options_obj) { |
5577fff7 | 1966 | error_prepend(errp, "Could not parse the JSON options: "); |
5e5c4f63 KW |
1967 | return NULL; |
1968 | } | |
1969 | ||
7dc847eb | 1970 | options = qobject_to(QDict, options_obj); |
ca6b6e1e | 1971 | if (!options) { |
cb3e7f08 | 1972 | qobject_unref(options_obj); |
5e5c4f63 KW |
1973 | error_setg(errp, "Invalid JSON object given"); |
1974 | return NULL; | |
1975 | } | |
1976 | ||
5e5c4f63 KW |
1977 | qdict_flatten(options); |
1978 | ||
1979 | return options; | |
1980 | } | |
1981 | ||
de3b53f0 KW |
1982 | static void parse_json_protocol(QDict *options, const char **pfilename, |
1983 | Error **errp) | |
1984 | { | |
1985 | QDict *json_options; | |
1986 | Error *local_err = NULL; | |
bdb73476 | 1987 | GLOBAL_STATE_CODE(); |
de3b53f0 KW |
1988 | |
1989 | /* Parse json: pseudo-protocol */ | |
1990 | if (!*pfilename || !g_str_has_prefix(*pfilename, "json:")) { | |
1991 | return; | |
1992 | } | |
1993 | ||
1994 | json_options = parse_json_filename(*pfilename, &local_err); | |
1995 | if (local_err) { | |
1996 | error_propagate(errp, local_err); | |
1997 | return; | |
1998 | } | |
1999 | ||
2000 | /* Options given in the filename have lower priority than options | |
2001 | * specified directly */ | |
2002 | qdict_join(options, json_options, false); | |
cb3e7f08 | 2003 | qobject_unref(json_options); |
de3b53f0 KW |
2004 | *pfilename = NULL; |
2005 | } | |
2006 | ||
b6ce07aa | 2007 | /* |
f54120ff KW |
2008 | * Fills in default options for opening images and converts the legacy |
2009 | * filename/flags pair to option QDict entries. | |
53a29513 HR |
2010 | * The BDRV_O_PROTOCOL flag in *flags will be set or cleared accordingly if a |
2011 | * block driver has been specified explicitly. | |
b6ce07aa | 2012 | */ |
de3b53f0 | 2013 | static int bdrv_fill_options(QDict **options, const char *filename, |
053e1578 | 2014 | int *flags, Error **errp) |
ea2384d3 | 2015 | { |
c2ad1b0c | 2016 | const char *drvname; |
53a29513 | 2017 | bool protocol = *flags & BDRV_O_PROTOCOL; |
e3fa4bfa | 2018 | bool parse_filename = false; |
053e1578 | 2019 | BlockDriver *drv = NULL; |
34b5d2c6 | 2020 | Error *local_err = NULL; |
83f64091 | 2021 | |
da359909 EGE |
2022 | GLOBAL_STATE_CODE(); |
2023 | ||
129c7d1c MA |
2024 | /* |
2025 | * Caution: while qdict_get_try_str() is fine, getting non-string | |
2026 | * types would require more care. When @options come from | |
2027 | * -blockdev or blockdev_add, its members are typed according to | |
2028 | * the QAPI schema, but when they come from -drive, they're all | |
2029 | * QString. | |
2030 | */ | |
53a29513 | 2031 | drvname = qdict_get_try_str(*options, "driver"); |
053e1578 HR |
2032 | if (drvname) { |
2033 | drv = bdrv_find_format(drvname); | |
2034 | if (!drv) { | |
2035 | error_setg(errp, "Unknown driver '%s'", drvname); | |
2036 | return -ENOENT; | |
2037 | } | |
2038 | /* If the user has explicitly specified the driver, this choice should | |
2039 | * override the BDRV_O_PROTOCOL flag */ | |
2040 | protocol = drv->bdrv_file_open; | |
53a29513 HR |
2041 | } |
2042 | ||
2043 | if (protocol) { | |
2044 | *flags |= BDRV_O_PROTOCOL; | |
2045 | } else { | |
2046 | *flags &= ~BDRV_O_PROTOCOL; | |
2047 | } | |
2048 | ||
91a097e7 KW |
2049 | /* Translate cache options from flags into options */ |
2050 | update_options_from_flags(*options, *flags); | |
2051 | ||
035fccdf | 2052 | /* Fetch the file name from the options QDict if necessary */ |
17b005f1 | 2053 | if (protocol && filename) { |
f54120ff | 2054 | if (!qdict_haskey(*options, "filename")) { |
46f5ac20 | 2055 | qdict_put_str(*options, "filename", filename); |
f54120ff KW |
2056 | parse_filename = true; |
2057 | } else { | |
2058 | error_setg(errp, "Can't specify 'file' and 'filename' options at " | |
2059 | "the same time"); | |
2060 | return -EINVAL; | |
2061 | } | |
035fccdf KW |
2062 | } |
2063 | ||
c2ad1b0c | 2064 | /* Find the right block driver */ |
129c7d1c | 2065 | /* See cautionary note on accessing @options above */ |
f54120ff | 2066 | filename = qdict_get_try_str(*options, "filename"); |
f54120ff | 2067 | |
053e1578 HR |
2068 | if (!drvname && protocol) { |
2069 | if (filename) { | |
2070 | drv = bdrv_find_protocol(filename, parse_filename, errp); | |
17b005f1 | 2071 | if (!drv) { |
053e1578 | 2072 | return -EINVAL; |
17b005f1 | 2073 | } |
053e1578 HR |
2074 | |
2075 | drvname = drv->format_name; | |
46f5ac20 | 2076 | qdict_put_str(*options, "driver", drvname); |
053e1578 HR |
2077 | } else { |
2078 | error_setg(errp, "Must specify either driver or file"); | |
2079 | return -EINVAL; | |
98289620 | 2080 | } |
c2ad1b0c KW |
2081 | } |
2082 | ||
17b005f1 | 2083 | assert(drv || !protocol); |
c2ad1b0c | 2084 | |
f54120ff | 2085 | /* Driver-specific filename parsing */ |
17b005f1 | 2086 | if (drv && drv->bdrv_parse_filename && parse_filename) { |
5acd9d81 | 2087 | drv->bdrv_parse_filename(filename, *options, &local_err); |
84d18f06 | 2088 | if (local_err) { |
34b5d2c6 | 2089 | error_propagate(errp, local_err); |
f54120ff | 2090 | return -EINVAL; |
6963a30d | 2091 | } |
cd5d031e HR |
2092 | |
2093 | if (!drv->bdrv_needs_filename) { | |
2094 | qdict_del(*options, "filename"); | |
cd5d031e | 2095 | } |
6963a30d KW |
2096 | } |
2097 | ||
f54120ff KW |
2098 | return 0; |
2099 | } | |
2100 | ||
148eb13c KW |
2101 | typedef struct BlockReopenQueueEntry { |
2102 | bool prepared; | |
69b736e7 | 2103 | bool perms_checked; |
148eb13c | 2104 | BDRVReopenState state; |
859443b0 | 2105 | QTAILQ_ENTRY(BlockReopenQueueEntry) entry; |
148eb13c KW |
2106 | } BlockReopenQueueEntry; |
2107 | ||
2108 | /* | |
2109 | * Return the flags that @bs will have after the reopens in @q have | |
2110 | * successfully completed. If @q is NULL (or @bs is not contained in @q), | |
2111 | * return the current flags. | |
2112 | */ | |
2113 | static int bdrv_reopen_get_flags(BlockReopenQueue *q, BlockDriverState *bs) | |
2114 | { | |
2115 | BlockReopenQueueEntry *entry; | |
2116 | ||
2117 | if (q != NULL) { | |
859443b0 | 2118 | QTAILQ_FOREACH(entry, q, entry) { |
148eb13c KW |
2119 | if (entry->state.bs == bs) { |
2120 | return entry->state.flags; | |
2121 | } | |
2122 | } | |
2123 | } | |
2124 | ||
2125 | return bs->open_flags; | |
2126 | } | |
2127 | ||
2128 | /* Returns whether the image file can be written to after the reopen queue @q | |
2129 | * has been successfully applied, or right now if @q is NULL. */ | |
cc022140 HR |
2130 | static bool bdrv_is_writable_after_reopen(BlockDriverState *bs, |
2131 | BlockReopenQueue *q) | |
148eb13c KW |
2132 | { |
2133 | int flags = bdrv_reopen_get_flags(q, bs); | |
2134 | ||
2135 | return (flags & (BDRV_O_RDWR | BDRV_O_INACTIVE)) == BDRV_O_RDWR; | |
2136 | } | |
2137 | ||
cc022140 HR |
2138 | /* |
2139 | * Return whether the BDS can be written to. This is not necessarily | |
2140 | * the same as !bdrv_is_read_only(bs), as inactivated images may not | |
2141 | * be written to but do not count as read-only images. | |
2142 | */ | |
2143 | bool bdrv_is_writable(BlockDriverState *bs) | |
2144 | { | |
384a48fb | 2145 | IO_CODE(); |
cc022140 HR |
2146 | return bdrv_is_writable_after_reopen(bs, NULL); |
2147 | } | |
2148 | ||
3bf416ba VSO |
2149 | static char *bdrv_child_user_desc(BdrvChild *c) |
2150 | { | |
f0c28327 | 2151 | GLOBAL_STATE_CODE(); |
da261b69 | 2152 | return c->klass->get_parent_desc(c); |
3bf416ba VSO |
2153 | } |
2154 | ||
30ebb9aa VSO |
2155 | /* |
2156 | * Check that @a allows everything that @b needs. @a and @b must reference same | |
2157 | * child node. | |
2158 | */ | |
3bf416ba VSO |
2159 | static bool bdrv_a_allow_b(BdrvChild *a, BdrvChild *b, Error **errp) |
2160 | { | |
30ebb9aa VSO |
2161 | const char *child_bs_name; |
2162 | g_autofree char *a_user = NULL; | |
2163 | g_autofree char *b_user = NULL; | |
2164 | g_autofree char *perms = NULL; | |
2165 | ||
2166 | assert(a->bs); | |
2167 | assert(a->bs == b->bs); | |
862fded9 | 2168 | GLOBAL_STATE_CODE(); |
3bf416ba VSO |
2169 | |
2170 | if ((b->perm & a->shared_perm) == b->perm) { | |
2171 | return true; | |
2172 | } | |
2173 | ||
30ebb9aa VSO |
2174 | child_bs_name = bdrv_get_node_name(b->bs); |
2175 | a_user = bdrv_child_user_desc(a); | |
2176 | b_user = bdrv_child_user_desc(b); | |
2177 | perms = bdrv_perm_names(b->perm & ~a->shared_perm); | |
2178 | ||
2179 | error_setg(errp, "Permission conflict on node '%s': permissions '%s' are " | |
2180 | "both required by %s (uses node '%s' as '%s' child) and " | |
2181 | "unshared by %s (uses node '%s' as '%s' child).", | |
2182 | child_bs_name, perms, | |
2183 | b_user, child_bs_name, b->name, | |
2184 | a_user, child_bs_name, a->name); | |
3bf416ba VSO |
2185 | |
2186 | return false; | |
2187 | } | |
2188 | ||
9397c14f | 2189 | static bool bdrv_parent_perms_conflict(BlockDriverState *bs, Error **errp) |
3bf416ba VSO |
2190 | { |
2191 | BdrvChild *a, *b; | |
862fded9 | 2192 | GLOBAL_STATE_CODE(); |
3bf416ba VSO |
2193 | |
2194 | /* | |
2195 | * During the loop we'll look at each pair twice. That's correct because | |
2196 | * bdrv_a_allow_b() is asymmetric and we should check each pair in both | |
2197 | * directions. | |
2198 | */ | |
2199 | QLIST_FOREACH(a, &bs->parents, next_parent) { | |
2200 | QLIST_FOREACH(b, &bs->parents, next_parent) { | |
9397c14f | 2201 | if (a == b) { |
3bf416ba VSO |
2202 | continue; |
2203 | } | |
2204 | ||
2205 | if (!bdrv_a_allow_b(a, b, errp)) { | |
2206 | return true; | |
2207 | } | |
2208 | } | |
2209 | } | |
2210 | ||
2211 | return false; | |
2212 | } | |
2213 | ||
ffd1a5a2 | 2214 | static void bdrv_child_perm(BlockDriverState *bs, BlockDriverState *child_bs, |
e5d8a406 HR |
2215 | BdrvChild *c, BdrvChildRole role, |
2216 | BlockReopenQueue *reopen_queue, | |
ffd1a5a2 FZ |
2217 | uint64_t parent_perm, uint64_t parent_shared, |
2218 | uint64_t *nperm, uint64_t *nshared) | |
2219 | { | |
0b3ca76e | 2220 | assert(bs->drv && bs->drv->bdrv_child_perm); |
da359909 | 2221 | GLOBAL_STATE_CODE(); |
e5d8a406 | 2222 | bs->drv->bdrv_child_perm(bs, c, role, reopen_queue, |
0b3ca76e AG |
2223 | parent_perm, parent_shared, |
2224 | nperm, nshared); | |
e0995dc3 | 2225 | /* TODO Take force_share from reopen_queue */ |
ffd1a5a2 FZ |
2226 | if (child_bs && child_bs->force_share) { |
2227 | *nshared = BLK_PERM_ALL; | |
2228 | } | |
2229 | } | |
2230 | ||
bd57f8f7 VSO |
2231 | /* |
2232 | * Adds the whole subtree of @bs (including @bs itself) to the @list (except for | |
2233 | * nodes that are already in the @list, of course) so that final list is | |
2234 | * topologically sorted. Return the result (GSList @list object is updated, so | |
2235 | * don't use old reference after function call). | |
2236 | * | |
2237 | * On function start @list must be already topologically sorted and for any node | |
2238 | * in the @list the whole subtree of the node must be in the @list as well. The | |
2239 | * simplest way to satisfy this criteria: use only result of | |
2240 | * bdrv_topological_dfs() or NULL as @list parameter. | |
2241 | */ | |
2242 | static GSList *bdrv_topological_dfs(GSList *list, GHashTable *found, | |
2243 | BlockDriverState *bs) | |
2244 | { | |
2245 | BdrvChild *child; | |
2246 | g_autoptr(GHashTable) local_found = NULL; | |
2247 | ||
bdb73476 EGE |
2248 | GLOBAL_STATE_CODE(); |
2249 | ||
bd57f8f7 VSO |
2250 | if (!found) { |
2251 | assert(!list); | |
2252 | found = local_found = g_hash_table_new(NULL, NULL); | |
2253 | } | |
2254 | ||
2255 | if (g_hash_table_contains(found, bs)) { | |
2256 | return list; | |
2257 | } | |
2258 | g_hash_table_add(found, bs); | |
2259 | ||
2260 | QLIST_FOREACH(child, &bs->children, next) { | |
2261 | list = bdrv_topological_dfs(list, found, child->bs); | |
2262 | } | |
2263 | ||
2264 | return g_slist_prepend(list, bs); | |
2265 | } | |
2266 | ||
ecb776bd VSO |
2267 | typedef struct BdrvChildSetPermState { |
2268 | BdrvChild *child; | |
2269 | uint64_t old_perm; | |
2270 | uint64_t old_shared_perm; | |
2271 | } BdrvChildSetPermState; | |
b0defa83 VSO |
2272 | |
2273 | static void bdrv_child_set_perm_abort(void *opaque) | |
2274 | { | |
ecb776bd VSO |
2275 | BdrvChildSetPermState *s = opaque; |
2276 | ||
862fded9 EGE |
2277 | GLOBAL_STATE_CODE(); |
2278 | ||
ecb776bd VSO |
2279 | s->child->perm = s->old_perm; |
2280 | s->child->shared_perm = s->old_shared_perm; | |
b0defa83 VSO |
2281 | } |
2282 | ||
2283 | static TransactionActionDrv bdrv_child_set_pem_drv = { | |
2284 | .abort = bdrv_child_set_perm_abort, | |
ecb776bd | 2285 | .clean = g_free, |
b0defa83 VSO |
2286 | }; |
2287 | ||
ecb776bd VSO |
2288 | static void bdrv_child_set_perm(BdrvChild *c, uint64_t perm, |
2289 | uint64_t shared, Transaction *tran) | |
b0defa83 | 2290 | { |
ecb776bd | 2291 | BdrvChildSetPermState *s = g_new(BdrvChildSetPermState, 1); |
862fded9 | 2292 | GLOBAL_STATE_CODE(); |
ecb776bd VSO |
2293 | |
2294 | *s = (BdrvChildSetPermState) { | |
2295 | .child = c, | |
2296 | .old_perm = c->perm, | |
2297 | .old_shared_perm = c->shared_perm, | |
2298 | }; | |
b0defa83 VSO |
2299 | |
2300 | c->perm = perm; | |
2301 | c->shared_perm = shared; | |
2302 | ||
ecb776bd | 2303 | tran_add(tran, &bdrv_child_set_pem_drv, s); |
b0defa83 VSO |
2304 | } |
2305 | ||
2513ef59 VSO |
2306 | static void bdrv_drv_set_perm_commit(void *opaque) |
2307 | { | |
2308 | BlockDriverState *bs = opaque; | |
2309 | uint64_t cumulative_perms, cumulative_shared_perms; | |
da359909 | 2310 | GLOBAL_STATE_CODE(); |
2513ef59 VSO |
2311 | |
2312 | if (bs->drv->bdrv_set_perm) { | |
2313 | bdrv_get_cumulative_perm(bs, &cumulative_perms, | |
2314 | &cumulative_shared_perms); | |
2315 | bs->drv->bdrv_set_perm(bs, cumulative_perms, cumulative_shared_perms); | |
2316 | } | |
2317 | } | |
2318 | ||
2319 | static void bdrv_drv_set_perm_abort(void *opaque) | |
2320 | { | |
2321 | BlockDriverState *bs = opaque; | |
da359909 | 2322 | GLOBAL_STATE_CODE(); |
2513ef59 VSO |
2323 | |
2324 | if (bs->drv->bdrv_abort_perm_update) { | |
2325 | bs->drv->bdrv_abort_perm_update(bs); | |
2326 | } | |
2327 | } | |
2328 | ||
2329 | TransactionActionDrv bdrv_drv_set_perm_drv = { | |
2330 | .abort = bdrv_drv_set_perm_abort, | |
2331 | .commit = bdrv_drv_set_perm_commit, | |
2332 | }; | |
2333 | ||
2334 | static int bdrv_drv_set_perm(BlockDriverState *bs, uint64_t perm, | |
2335 | uint64_t shared_perm, Transaction *tran, | |
2336 | Error **errp) | |
2337 | { | |
da359909 | 2338 | GLOBAL_STATE_CODE(); |
2513ef59 VSO |
2339 | if (!bs->drv) { |
2340 | return 0; | |
2341 | } | |
2342 | ||
2343 | if (bs->drv->bdrv_check_perm) { | |
2344 | int ret = bs->drv->bdrv_check_perm(bs, perm, shared_perm, errp); | |
2345 | if (ret < 0) { | |
2346 | return ret; | |
2347 | } | |
2348 | } | |
2349 | ||
2350 | if (tran) { | |
2351 | tran_add(tran, &bdrv_drv_set_perm_drv, bs); | |
2352 | } | |
2353 | ||
2354 | return 0; | |
2355 | } | |
2356 | ||
0978623e VSO |
2357 | typedef struct BdrvReplaceChildState { |
2358 | BdrvChild *child; | |
2359 | BlockDriverState *old_bs; | |
2360 | } BdrvReplaceChildState; | |
2361 | ||
2362 | static void bdrv_replace_child_commit(void *opaque) | |
2363 | { | |
2364 | BdrvReplaceChildState *s = opaque; | |
bdb73476 | 2365 | GLOBAL_STATE_CODE(); |
0978623e VSO |
2366 | |
2367 | bdrv_unref(s->old_bs); | |
2368 | } | |
2369 | ||
2370 | static void bdrv_replace_child_abort(void *opaque) | |
2371 | { | |
2372 | BdrvReplaceChildState *s = opaque; | |
2373 | BlockDriverState *new_bs = s->child->bs; | |
2374 | ||
bdb73476 | 2375 | GLOBAL_STATE_CODE(); |
0f0b1e29 | 2376 | /* old_bs reference is transparently moved from @s to @s->child */ |
23987471 KW |
2377 | if (!s->child->bs) { |
2378 | /* | |
2379 | * The parents were undrained when removing old_bs from the child. New | |
2380 | * requests can't have been made, though, because the child was empty. | |
2381 | * | |
2382 | * TODO Make bdrv_replace_child_noperm() transactionable to avoid | |
2383 | * undraining the parent in the first place. Once this is done, having | |
2384 | * new_bs drained when calling bdrv_replace_child_tran() is not a | |
2385 | * requirement any more. | |
2386 | */ | |
606ed756 | 2387 | bdrv_parent_drained_begin_single(s->child); |
23987471 KW |
2388 | assert(!bdrv_parent_drained_poll_single(s->child)); |
2389 | } | |
2390 | assert(s->child->quiesced_parent); | |
544acc7d | 2391 | bdrv_replace_child_noperm(s->child, s->old_bs); |
0978623e VSO |
2392 | bdrv_unref(new_bs); |
2393 | } | |
2394 | ||
2395 | static TransactionActionDrv bdrv_replace_child_drv = { | |
2396 | .commit = bdrv_replace_child_commit, | |
2397 | .abort = bdrv_replace_child_abort, | |
2398 | .clean = g_free, | |
2399 | }; | |
2400 | ||
2401 | /* | |
4bf021db | 2402 | * bdrv_replace_child_tran |
0978623e VSO |
2403 | * |
2404 | * Note: real unref of old_bs is done only on commit. | |
4bf021db | 2405 | * |
23987471 KW |
2406 | * Both @child->bs and @new_bs (if non-NULL) must be drained. @new_bs must be |
2407 | * kept drained until the transaction is completed. | |
2408 | * | |
4bf021db | 2409 | * The function doesn't update permissions, caller is responsible for this. |
0978623e | 2410 | */ |
0f0b1e29 | 2411 | static void bdrv_replace_child_tran(BdrvChild *child, BlockDriverState *new_bs, |
4eba825a | 2412 | Transaction *tran) |
0978623e VSO |
2413 | { |
2414 | BdrvReplaceChildState *s = g_new(BdrvReplaceChildState, 1); | |
23987471 KW |
2415 | |
2416 | assert(child->quiesced_parent); | |
2417 | assert(!new_bs || new_bs->quiesce_counter); | |
2418 | ||
0978623e | 2419 | *s = (BdrvReplaceChildState) { |
0f0b1e29 VSO |
2420 | .child = child, |
2421 | .old_bs = child->bs, | |
0978623e VSO |
2422 | }; |
2423 | tran_add(tran, &bdrv_replace_child_drv, s); | |
2424 | ||
2425 | if (new_bs) { | |
2426 | bdrv_ref(new_bs); | |
2427 | } | |
544acc7d | 2428 | bdrv_replace_child_noperm(child, new_bs); |
0f0b1e29 | 2429 | /* old_bs reference is transparently moved from @child to @s */ |
0978623e VSO |
2430 | } |
2431 | ||
33a610c3 | 2432 | /* |
c20555e1 VSO |
2433 | * Refresh permissions in @bs subtree. The function is intended to be called |
2434 | * after some graph modification that was done without permission update. | |
33a610c3 | 2435 | */ |
c20555e1 VSO |
2436 | static int bdrv_node_refresh_perm(BlockDriverState *bs, BlockReopenQueue *q, |
2437 | Transaction *tran, Error **errp) | |
33a610c3 KW |
2438 | { |
2439 | BlockDriver *drv = bs->drv; | |
2440 | BdrvChild *c; | |
2441 | int ret; | |
c20555e1 | 2442 | uint64_t cumulative_perms, cumulative_shared_perms; |
862fded9 | 2443 | GLOBAL_STATE_CODE(); |
c20555e1 VSO |
2444 | |
2445 | bdrv_get_cumulative_perm(bs, &cumulative_perms, &cumulative_shared_perms); | |
33a610c3 KW |
2446 | |
2447 | /* Write permissions never work with read-only images */ | |
2448 | if ((cumulative_perms & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)) && | |
cc022140 | 2449 | !bdrv_is_writable_after_reopen(bs, q)) |
33a610c3 | 2450 | { |
481e0eee HR |
2451 | if (!bdrv_is_writable_after_reopen(bs, NULL)) { |
2452 | error_setg(errp, "Block node is read-only"); | |
2453 | } else { | |
c20555e1 VSO |
2454 | error_setg(errp, "Read-only block node '%s' cannot support " |
2455 | "read-write users", bdrv_get_node_name(bs)); | |
481e0eee HR |
2456 | } |
2457 | ||
33a610c3 KW |
2458 | return -EPERM; |
2459 | } | |
2460 | ||
9c60a5d1 KW |
2461 | /* |
2462 | * Unaligned requests will automatically be aligned to bl.request_alignment | |
2463 | * and without RESIZE we can't extend requests to write to space beyond the | |
2464 | * end of the image, so it's required that the image size is aligned. | |
2465 | */ | |
2466 | if ((cumulative_perms & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)) && | |
2467 | !(cumulative_perms & BLK_PERM_RESIZE)) | |
2468 | { | |
2469 | if ((bs->total_sectors * BDRV_SECTOR_SIZE) % bs->bl.request_alignment) { | |
2470 | error_setg(errp, "Cannot get 'write' permission without 'resize': " | |
2471 | "Image size is not a multiple of request " | |
2472 | "alignment"); | |
2473 | return -EPERM; | |
2474 | } | |
2475 | } | |
2476 | ||
33a610c3 KW |
2477 | /* Check this node */ |
2478 | if (!drv) { | |
2479 | return 0; | |
2480 | } | |
2481 | ||
b1d2bbeb | 2482 | ret = bdrv_drv_set_perm(bs, cumulative_perms, cumulative_shared_perms, tran, |
2513ef59 VSO |
2483 | errp); |
2484 | if (ret < 0) { | |
2485 | return ret; | |
33a610c3 KW |
2486 | } |
2487 | ||
78e421c9 | 2488 | /* Drivers that never have children can omit .bdrv_child_perm() */ |
33a610c3 | 2489 | if (!drv->bdrv_child_perm) { |
78e421c9 | 2490 | assert(QLIST_EMPTY(&bs->children)); |
33a610c3 KW |
2491 | return 0; |
2492 | } | |
2493 | ||
2494 | /* Check all children */ | |
2495 | QLIST_FOREACH(c, &bs->children, next) { | |
2496 | uint64_t cur_perm, cur_shared; | |
9eab1544 | 2497 | |
e5d8a406 | 2498 | bdrv_child_perm(bs, c->bs, c, c->role, q, |
ffd1a5a2 FZ |
2499 | cumulative_perms, cumulative_shared_perms, |
2500 | &cur_perm, &cur_shared); | |
ecb776bd | 2501 | bdrv_child_set_perm(c, cur_perm, cur_shared, tran); |
bd57f8f7 VSO |
2502 | } |
2503 | ||
2504 | return 0; | |
2505 | } | |
3ef45e02 | 2506 | |
fb0ff4d1 VSO |
2507 | /* |
2508 | * @list is a product of bdrv_topological_dfs() (may be called several times) - | |
2509 | * a topologically sorted subgraph. | |
2510 | */ | |
2511 | static int bdrv_do_refresh_perms(GSList *list, BlockReopenQueue *q, | |
2512 | Transaction *tran, Error **errp) | |
bd57f8f7 VSO |
2513 | { |
2514 | int ret; | |
b1d2bbeb | 2515 | BlockDriverState *bs; |
862fded9 | 2516 | GLOBAL_STATE_CODE(); |
bd57f8f7 | 2517 | |
b1d2bbeb VSO |
2518 | for ( ; list; list = list->next) { |
2519 | bs = list->data; | |
2520 | ||
9397c14f | 2521 | if (bdrv_parent_perms_conflict(bs, errp)) { |
b1d2bbeb | 2522 | return -EINVAL; |
bd57f8f7 VSO |
2523 | } |
2524 | ||
c20555e1 | 2525 | ret = bdrv_node_refresh_perm(bs, q, tran, errp); |
33a610c3 KW |
2526 | if (ret < 0) { |
2527 | return ret; | |
2528 | } | |
2529 | } | |
2530 | ||
2531 | return 0; | |
2532 | } | |
2533 | ||
fb0ff4d1 VSO |
2534 | /* |
2535 | * @list is any list of nodes. List is completed by all subtrees and | |
2536 | * topologically sorted. It's not a problem if some node occurs in the @list | |
2537 | * several times. | |
2538 | */ | |
2539 | static int bdrv_list_refresh_perms(GSList *list, BlockReopenQueue *q, | |
2540 | Transaction *tran, Error **errp) | |
2541 | { | |
2542 | g_autoptr(GHashTable) found = g_hash_table_new(NULL, NULL); | |
2543 | g_autoptr(GSList) refresh_list = NULL; | |
2544 | ||
2545 | for ( ; list; list = list->next) { | |
2546 | refresh_list = bdrv_topological_dfs(refresh_list, found, list->data); | |
2547 | } | |
2548 | ||
2549 | return bdrv_do_refresh_perms(refresh_list, q, tran, errp); | |
2550 | } | |
2551 | ||
c7a0f2be KW |
2552 | void bdrv_get_cumulative_perm(BlockDriverState *bs, uint64_t *perm, |
2553 | uint64_t *shared_perm) | |
33a610c3 KW |
2554 | { |
2555 | BdrvChild *c; | |
2556 | uint64_t cumulative_perms = 0; | |
2557 | uint64_t cumulative_shared_perms = BLK_PERM_ALL; | |
2558 | ||
b4ad82aa EGE |
2559 | GLOBAL_STATE_CODE(); |
2560 | ||
33a610c3 KW |
2561 | QLIST_FOREACH(c, &bs->parents, next_parent) { |
2562 | cumulative_perms |= c->perm; | |
2563 | cumulative_shared_perms &= c->shared_perm; | |
2564 | } | |
2565 | ||
2566 | *perm = cumulative_perms; | |
2567 | *shared_perm = cumulative_shared_perms; | |
2568 | } | |
2569 | ||
5176196c | 2570 | char *bdrv_perm_names(uint64_t perm) |
d083319f KW |
2571 | { |
2572 | struct perm_name { | |
2573 | uint64_t perm; | |
2574 | const char *name; | |
2575 | } permissions[] = { | |
2576 | { BLK_PERM_CONSISTENT_READ, "consistent read" }, | |
2577 | { BLK_PERM_WRITE, "write" }, | |
2578 | { BLK_PERM_WRITE_UNCHANGED, "write unchanged" }, | |
2579 | { BLK_PERM_RESIZE, "resize" }, | |
d083319f KW |
2580 | { 0, NULL } |
2581 | }; | |
2582 | ||
e2a7423a | 2583 | GString *result = g_string_sized_new(30); |
d083319f KW |
2584 | struct perm_name *p; |
2585 | ||
2586 | for (p = permissions; p->name; p++) { | |
2587 | if (perm & p->perm) { | |
e2a7423a AG |
2588 | if (result->len > 0) { |
2589 | g_string_append(result, ", "); | |
2590 | } | |
2591 | g_string_append(result, p->name); | |
d083319f KW |
2592 | } |
2593 | } | |
2594 | ||
e2a7423a | 2595 | return g_string_free(result, FALSE); |
d083319f KW |
2596 | } |
2597 | ||
33a610c3 | 2598 | |
f1316edb VSO |
2599 | /* @tran is allowed to be NULL. In this case no rollback is possible */ |
2600 | static int bdrv_refresh_perms(BlockDriverState *bs, Transaction *tran, | |
2601 | Error **errp) | |
bb87e4d1 VSO |
2602 | { |
2603 | int ret; | |
f1316edb | 2604 | Transaction *local_tran = NULL; |
b1d2bbeb | 2605 | g_autoptr(GSList) list = bdrv_topological_dfs(NULL, NULL, bs); |
862fded9 | 2606 | GLOBAL_STATE_CODE(); |
bb87e4d1 | 2607 | |
f1316edb VSO |
2608 | if (!tran) { |
2609 | tran = local_tran = tran_new(); | |
2610 | } | |
2611 | ||
fb0ff4d1 | 2612 | ret = bdrv_do_refresh_perms(list, NULL, tran, errp); |
f1316edb VSO |
2613 | |
2614 | if (local_tran) { | |
2615 | tran_finalize(local_tran, ret); | |
2616 | } | |
bb87e4d1 | 2617 | |
b1d2bbeb | 2618 | return ret; |
bb87e4d1 VSO |
2619 | } |
2620 | ||
33a610c3 KW |
2621 | int bdrv_child_try_set_perm(BdrvChild *c, uint64_t perm, uint64_t shared, |
2622 | Error **errp) | |
2623 | { | |
1046779e | 2624 | Error *local_err = NULL; |
83928dc4 | 2625 | Transaction *tran = tran_new(); |
33a610c3 KW |
2626 | int ret; |
2627 | ||
b4ad82aa EGE |
2628 | GLOBAL_STATE_CODE(); |
2629 | ||
ecb776bd | 2630 | bdrv_child_set_perm(c, perm, shared, tran); |
83928dc4 | 2631 | |
f1316edb | 2632 | ret = bdrv_refresh_perms(c->bs, tran, &local_err); |
83928dc4 VSO |
2633 | |
2634 | tran_finalize(tran, ret); | |
2635 | ||
33a610c3 | 2636 | if (ret < 0) { |
071b474f VSO |
2637 | if ((perm & ~c->perm) || (c->shared_perm & ~shared)) { |
2638 | /* tighten permissions */ | |
1046779e HR |
2639 | error_propagate(errp, local_err); |
2640 | } else { | |
2641 | /* | |
2642 | * Our caller may intend to only loosen restrictions and | |
2643 | * does not expect this function to fail. Errors are not | |
2644 | * fatal in such a case, so we can just hide them from our | |
2645 | * caller. | |
2646 | */ | |
2647 | error_free(local_err); | |
2648 | ret = 0; | |
2649 | } | |
33a610c3 KW |
2650 | } |
2651 | ||
83928dc4 | 2652 | return ret; |
d5e6f437 KW |
2653 | } |
2654 | ||
c1087f12 HR |
2655 | int bdrv_child_refresh_perms(BlockDriverState *bs, BdrvChild *c, Error **errp) |
2656 | { | |
2657 | uint64_t parent_perms, parent_shared; | |
2658 | uint64_t perms, shared; | |
2659 | ||
b4ad82aa EGE |
2660 | GLOBAL_STATE_CODE(); |
2661 | ||
c1087f12 | 2662 | bdrv_get_cumulative_perm(bs, &parent_perms, &parent_shared); |
e5d8a406 | 2663 | bdrv_child_perm(bs, c->bs, c, c->role, NULL, |
bf8e925e | 2664 | parent_perms, parent_shared, &perms, &shared); |
c1087f12 HR |
2665 | |
2666 | return bdrv_child_try_set_perm(c, perms, shared, errp); | |
2667 | } | |
2668 | ||
87278af1 HR |
2669 | /* |
2670 | * Default implementation for .bdrv_child_perm() for block filters: | |
2671 | * Forward CONSISTENT_READ, WRITE, WRITE_UNCHANGED, and RESIZE to the | |
2672 | * filtered child. | |
2673 | */ | |
2674 | static void bdrv_filter_default_perms(BlockDriverState *bs, BdrvChild *c, | |
87278af1 HR |
2675 | BdrvChildRole role, |
2676 | BlockReopenQueue *reopen_queue, | |
2677 | uint64_t perm, uint64_t shared, | |
2678 | uint64_t *nperm, uint64_t *nshared) | |
6a1b9ee1 | 2679 | { |
862fded9 | 2680 | GLOBAL_STATE_CODE(); |
e444fa83 KW |
2681 | *nperm = perm & DEFAULT_PERM_PASSTHROUGH; |
2682 | *nshared = (shared & DEFAULT_PERM_PASSTHROUGH) | DEFAULT_PERM_UNCHANGED; | |
6a1b9ee1 KW |
2683 | } |
2684 | ||
70082db4 | 2685 | static void bdrv_default_perms_for_cow(BlockDriverState *bs, BdrvChild *c, |
70082db4 HR |
2686 | BdrvChildRole role, |
2687 | BlockReopenQueue *reopen_queue, | |
2688 | uint64_t perm, uint64_t shared, | |
2689 | uint64_t *nperm, uint64_t *nshared) | |
2690 | { | |
e5d8a406 | 2691 | assert(role & BDRV_CHILD_COW); |
862fded9 | 2692 | GLOBAL_STATE_CODE(); |
70082db4 HR |
2693 | |
2694 | /* | |
2695 | * We want consistent read from backing files if the parent needs it. | |
2696 | * No other operations are performed on backing files. | |
2697 | */ | |
2698 | perm &= BLK_PERM_CONSISTENT_READ; | |
2699 | ||
2700 | /* | |
2701 | * If the parent can deal with changing data, we're okay with a | |
2702 | * writable and resizable backing file. | |
2703 | * TODO Require !(perm & BLK_PERM_CONSISTENT_READ), too? | |
2704 | */ | |
2705 | if (shared & BLK_PERM_WRITE) { | |
2706 | shared = BLK_PERM_WRITE | BLK_PERM_RESIZE; | |
2707 | } else { | |
2708 | shared = 0; | |
2709 | } | |
2710 | ||
64631f36 | 2711 | shared |= BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE_UNCHANGED; |
70082db4 HR |
2712 | |
2713 | if (bs->open_flags & BDRV_O_INACTIVE) { | |
2714 | shared |= BLK_PERM_WRITE | BLK_PERM_RESIZE; | |
2715 | } | |
2716 | ||
2717 | *nperm = perm; | |
2718 | *nshared = shared; | |
2719 | } | |
2720 | ||
6f838a4b | 2721 | static void bdrv_default_perms_for_storage(BlockDriverState *bs, BdrvChild *c, |
6f838a4b HR |
2722 | BdrvChildRole role, |
2723 | BlockReopenQueue *reopen_queue, | |
2724 | uint64_t perm, uint64_t shared, | |
2725 | uint64_t *nperm, uint64_t *nshared) | |
2726 | { | |
2727 | int flags; | |
2728 | ||
862fded9 | 2729 | GLOBAL_STATE_CODE(); |
e5d8a406 | 2730 | assert(role & (BDRV_CHILD_METADATA | BDRV_CHILD_DATA)); |
6f838a4b HR |
2731 | |
2732 | flags = bdrv_reopen_get_flags(reopen_queue, bs); | |
2733 | ||
2734 | /* | |
2735 | * Apart from the modifications below, the same permissions are | |
2736 | * forwarded and left alone as for filters | |
2737 | */ | |
e5d8a406 | 2738 | bdrv_filter_default_perms(bs, c, role, reopen_queue, |
6f838a4b HR |
2739 | perm, shared, &perm, &shared); |
2740 | ||
f889054f HR |
2741 | if (role & BDRV_CHILD_METADATA) { |
2742 | /* Format drivers may touch metadata even if the guest doesn't write */ | |
2743 | if (bdrv_is_writable_after_reopen(bs, reopen_queue)) { | |
2744 | perm |= BLK_PERM_WRITE | BLK_PERM_RESIZE; | |
2745 | } | |
2746 | ||
2747 | /* | |
2748 | * bs->file always needs to be consistent because of the | |
2749 | * metadata. We can never allow other users to resize or write | |
2750 | * to it. | |
2751 | */ | |
2752 | if (!(flags & BDRV_O_NO_IO)) { | |
2753 | perm |= BLK_PERM_CONSISTENT_READ; | |
2754 | } | |
2755 | shared &= ~(BLK_PERM_WRITE | BLK_PERM_RESIZE); | |
6f838a4b HR |
2756 | } |
2757 | ||
f889054f HR |
2758 | if (role & BDRV_CHILD_DATA) { |
2759 | /* | |
2760 | * Technically, everything in this block is a subset of the | |
2761 | * BDRV_CHILD_METADATA path taken above, and so this could | |
2762 | * be an "else if" branch. However, that is not obvious, and | |
2763 | * this function is not performance critical, therefore we let | |
2764 | * this be an independent "if". | |
2765 | */ | |
2766 | ||
2767 | /* | |
2768 | * We cannot allow other users to resize the file because the | |
2769 | * format driver might have some assumptions about the size | |
2770 | * (e.g. because it is stored in metadata, or because the file | |
2771 | * is split into fixed-size data files). | |
2772 | */ | |
2773 | shared &= ~BLK_PERM_RESIZE; | |
2774 | ||
2775 | /* | |
2776 | * WRITE_UNCHANGED often cannot be performed as such on the | |
2777 | * data file. For example, the qcow2 driver may still need to | |
2778 | * write copied clusters on copy-on-read. | |
2779 | */ | |
2780 | if (perm & BLK_PERM_WRITE_UNCHANGED) { | |
2781 | perm |= BLK_PERM_WRITE; | |
2782 | } | |
2783 | ||
2784 | /* | |
2785 | * If the data file is written to, the format driver may | |
2786 | * expect to be able to resize it by writing beyond the EOF. | |
2787 | */ | |
2788 | if (perm & BLK_PERM_WRITE) { | |
2789 | perm |= BLK_PERM_RESIZE; | |
2790 | } | |
6f838a4b | 2791 | } |
6f838a4b HR |
2792 | |
2793 | if (bs->open_flags & BDRV_O_INACTIVE) { | |
2794 | shared |= BLK_PERM_WRITE | BLK_PERM_RESIZE; | |
2795 | } | |
2796 | ||
2797 | *nperm = perm; | |
2798 | *nshared = shared; | |
2799 | } | |
2800 | ||
2519f549 | 2801 | void bdrv_default_perms(BlockDriverState *bs, BdrvChild *c, |
e5d8a406 | 2802 | BdrvChildRole role, BlockReopenQueue *reopen_queue, |
2519f549 HR |
2803 | uint64_t perm, uint64_t shared, |
2804 | uint64_t *nperm, uint64_t *nshared) | |
2805 | { | |
b4ad82aa | 2806 | GLOBAL_STATE_CODE(); |
2519f549 HR |
2807 | if (role & BDRV_CHILD_FILTERED) { |
2808 | assert(!(role & (BDRV_CHILD_DATA | BDRV_CHILD_METADATA | | |
2809 | BDRV_CHILD_COW))); | |
e5d8a406 | 2810 | bdrv_filter_default_perms(bs, c, role, reopen_queue, |
2519f549 HR |
2811 | perm, shared, nperm, nshared); |
2812 | } else if (role & BDRV_CHILD_COW) { | |
2813 | assert(!(role & (BDRV_CHILD_DATA | BDRV_CHILD_METADATA))); | |
e5d8a406 | 2814 | bdrv_default_perms_for_cow(bs, c, role, reopen_queue, |
2519f549 HR |
2815 | perm, shared, nperm, nshared); |
2816 | } else if (role & (BDRV_CHILD_METADATA | BDRV_CHILD_DATA)) { | |
e5d8a406 | 2817 | bdrv_default_perms_for_storage(bs, c, role, reopen_queue, |
2519f549 HR |
2818 | perm, shared, nperm, nshared); |
2819 | } else { | |
2820 | g_assert_not_reached(); | |
2821 | } | |
2822 | } | |
2823 | ||
7b1d9c4d HR |
2824 | uint64_t bdrv_qapi_perm_to_blk_perm(BlockPermission qapi_perm) |
2825 | { | |
2826 | static const uint64_t permissions[] = { | |
2827 | [BLOCK_PERMISSION_CONSISTENT_READ] = BLK_PERM_CONSISTENT_READ, | |
2828 | [BLOCK_PERMISSION_WRITE] = BLK_PERM_WRITE, | |
2829 | [BLOCK_PERMISSION_WRITE_UNCHANGED] = BLK_PERM_WRITE_UNCHANGED, | |
2830 | [BLOCK_PERMISSION_RESIZE] = BLK_PERM_RESIZE, | |
7b1d9c4d HR |
2831 | }; |
2832 | ||
2833 | QEMU_BUILD_BUG_ON(ARRAY_SIZE(permissions) != BLOCK_PERMISSION__MAX); | |
2834 | QEMU_BUILD_BUG_ON(1UL << ARRAY_SIZE(permissions) != BLK_PERM_ALL + 1); | |
2835 | ||
2836 | assert(qapi_perm < BLOCK_PERMISSION__MAX); | |
2837 | ||
2838 | return permissions[qapi_perm]; | |
2839 | } | |
2840 | ||
23987471 KW |
2841 | /* |
2842 | * Replaces the node that a BdrvChild points to without updating permissions. | |
2843 | * | |
2844 | * If @new_bs is non-NULL, the parent of @child must already be drained through | |
2845 | * @child. | |
23987471 | 2846 | */ |
544acc7d | 2847 | static void bdrv_replace_child_noperm(BdrvChild *child, |
4eba825a | 2848 | BlockDriverState *new_bs) |
e9740bc6 KW |
2849 | { |
2850 | BlockDriverState *old_bs = child->bs; | |
debc2927 | 2851 | int new_bs_quiesce_counter; |
e9740bc6 | 2852 | |
2cad1ebe | 2853 | assert(!child->frozen); |
23987471 KW |
2854 | |
2855 | /* | |
2856 | * If we want to change the BdrvChild to point to a drained node as its new | |
2857 | * child->bs, we need to make sure that its new parent is drained, too. In | |
2858 | * other words, either child->quiesce_parent must already be true or we must | |
2859 | * be able to set it and keep the parent's quiesce_counter consistent with | |
2860 | * that, but without polling or starting new requests (this function | |
2861 | * guarantees that it doesn't poll, and starting new requests would be | |
2862 | * against the invariants of drain sections). | |
2863 | * | |
2864 | * To keep things simple, we pick the first option (child->quiesce_parent | |
2865 | * must already be true). We also generalise the rule a bit to make it | |
2866 | * easier to verify in callers and more likely to be covered in test cases: | |
2867 | * The parent must be quiesced through this child even if new_bs isn't | |
2868 | * currently drained. | |
2869 | * | |
2870 | * The only exception is for callers that always pass new_bs == NULL. In | |
2871 | * this case, we obviously never need to consider the case of a drained | |
2872 | * new_bs, so we can keep the callers simpler by allowing them not to drain | |
2873 | * the parent. | |
2874 | */ | |
2875 | assert(!new_bs || child->quiesced_parent); | |
bfb8aa6d | 2876 | assert(old_bs != new_bs); |
f0c28327 | 2877 | GLOBAL_STATE_CODE(); |
2cad1ebe | 2878 | |
bb2614e9 FZ |
2879 | if (old_bs && new_bs) { |
2880 | assert(bdrv_get_aio_context(old_bs) == bdrv_get_aio_context(new_bs)); | |
2881 | } | |
debc2927 | 2882 | |
29312507 EGE |
2883 | /* TODO Pull this up into the callers to avoid polling here */ |
2884 | bdrv_graph_wrlock(); | |
e9740bc6 | 2885 | if (old_bs) { |
bd86fb99 HR |
2886 | if (child->klass->detach) { |
2887 | child->klass->detach(child); | |
d736f119 | 2888 | } |
e9740bc6 KW |
2889 | QLIST_REMOVE(child, next_parent); |
2890 | } | |
36fe1331 KW |
2891 | |
2892 | child->bs = new_bs; | |
2893 | ||
e9740bc6 KW |
2894 | if (new_bs) { |
2895 | QLIST_INSERT_HEAD(&new_bs->parents, child, next_parent); | |
bd86fb99 HR |
2896 | if (child->klass->attach) { |
2897 | child->klass->attach(child); | |
8ee03995 KW |
2898 | } |
2899 | } | |
29312507 | 2900 | bdrv_graph_wrunlock(); |
debc2927 HR |
2901 | |
2902 | /* | |
23987471 KW |
2903 | * If the parent was drained through this BdrvChild previously, but new_bs |
2904 | * is not drained, allow requests to come in only after the new node has | |
2905 | * been attached. | |
debc2927 | 2906 | */ |
57e05be3 KW |
2907 | new_bs_quiesce_counter = (new_bs ? new_bs->quiesce_counter : 0); |
2908 | if (!new_bs_quiesce_counter && child->quiesced_parent) { | |
debc2927 | 2909 | bdrv_parent_drained_end_single(child); |
debc2927 | 2910 | } |
8ee03995 | 2911 | } |
33a610c3 | 2912 | |
04c9c3a5 HR |
2913 | /** |
2914 | * Free the given @child. | |
2915 | * | |
2916 | * The child must be empty (i.e. `child->bs == NULL`) and it must be | |
2917 | * unused (i.e. not in a children list). | |
2918 | */ | |
2919 | static void bdrv_child_free(BdrvChild *child) | |
df581792 | 2920 | { |
548a74c0 | 2921 | assert(!child->bs); |
bdb73476 | 2922 | GLOBAL_STATE_CODE(); |
a225369b | 2923 | assert(!child->next.le_prev); /* not in children list */ |
04c9c3a5 HR |
2924 | |
2925 | g_free(child->name); | |
2926 | g_free(child); | |
548a74c0 | 2927 | } |
d5e6f437 | 2928 | |
548a74c0 | 2929 | typedef struct BdrvAttachChildCommonState { |
5bb04747 | 2930 | BdrvChild *child; |
548a74c0 VSO |
2931 | AioContext *old_parent_ctx; |
2932 | AioContext *old_child_ctx; | |
2933 | } BdrvAttachChildCommonState; | |
2934 | ||
2935 | static void bdrv_attach_child_common_abort(void *opaque) | |
2936 | { | |
2937 | BdrvAttachChildCommonState *s = opaque; | |
5bb04747 | 2938 | BlockDriverState *bs = s->child->bs; |
548a74c0 | 2939 | |
f0c28327 | 2940 | GLOBAL_STATE_CODE(); |
5bb04747 | 2941 | bdrv_replace_child_noperm(s->child, NULL); |
548a74c0 VSO |
2942 | |
2943 | if (bdrv_get_aio_context(bs) != s->old_child_ctx) { | |
142e6907 | 2944 | bdrv_try_change_aio_context(bs, s->old_child_ctx, NULL, &error_abort); |
548a74c0 VSO |
2945 | } |
2946 | ||
5bb04747 | 2947 | if (bdrv_child_get_parent_aio_context(s->child) != s->old_parent_ctx) { |
f8be48ad EGE |
2948 | Transaction *tran; |
2949 | GHashTable *visited; | |
2950 | bool ret; | |
548a74c0 | 2951 | |
f8be48ad | 2952 | tran = tran_new(); |
548a74c0 | 2953 | |
f8be48ad EGE |
2954 | /* No need to visit `child`, because it has been detached already */ |
2955 | visited = g_hash_table_new(NULL, NULL); | |
2956 | ret = s->child->klass->change_aio_ctx(s->child, s->old_parent_ctx, | |
2957 | visited, tran, &error_abort); | |
2958 | g_hash_table_destroy(visited); | |
2959 | ||
2960 | /* transaction is supposed to always succeed */ | |
2961 | assert(ret == true); | |
2962 | tran_commit(tran); | |
d5e6f437 KW |
2963 | } |
2964 | ||
548a74c0 | 2965 | bdrv_unref(bs); |
5bb04747 | 2966 | bdrv_child_free(s->child); |
548a74c0 VSO |
2967 | } |
2968 | ||
2969 | static TransactionActionDrv bdrv_attach_child_common_drv = { | |
2970 | .abort = bdrv_attach_child_common_abort, | |
2971 | .clean = g_free, | |
2972 | }; | |
2973 | ||
2974 | /* | |
2975 | * Common part of attaching bdrv child to bs or to blk or to job | |
f8d2ad78 | 2976 | * |
7ec390d5 | 2977 | * Function doesn't update permissions, caller is responsible for this. |
5bb04747 VSO |
2978 | * |
2979 | * Returns new created child. | |
548a74c0 | 2980 | */ |
5bb04747 VSO |
2981 | static BdrvChild *bdrv_attach_child_common(BlockDriverState *child_bs, |
2982 | const char *child_name, | |
2983 | const BdrvChildClass *child_class, | |
2984 | BdrvChildRole child_role, | |
2985 | uint64_t perm, uint64_t shared_perm, | |
2986 | void *opaque, | |
2987 | Transaction *tran, Error **errp) | |
548a74c0 VSO |
2988 | { |
2989 | BdrvChild *new_child; | |
2990 | AioContext *parent_ctx; | |
2991 | AioContext *child_ctx = bdrv_get_aio_context(child_bs); | |
2992 | ||
da261b69 | 2993 | assert(child_class->get_parent_desc); |
bdb73476 | 2994 | GLOBAL_STATE_CODE(); |
548a74c0 VSO |
2995 | |
2996 | new_child = g_new(BdrvChild, 1); | |
2997 | *new_child = (BdrvChild) { | |
d5e6f437 KW |
2998 | .bs = NULL, |
2999 | .name = g_strdup(child_name), | |
bd86fb99 | 3000 | .klass = child_class, |
258b7765 | 3001 | .role = child_role, |
d5e6f437 KW |
3002 | .perm = perm, |
3003 | .shared_perm = shared_perm, | |
3004 | .opaque = opaque, | |
df581792 KW |
3005 | }; |
3006 | ||
548a74c0 VSO |
3007 | /* |
3008 | * If the AioContexts don't match, first try to move the subtree of | |
132ada80 | 3009 | * child_bs into the AioContext of the new parent. If this doesn't work, |
548a74c0 VSO |
3010 | * try moving the parent into the AioContext of child_bs instead. |
3011 | */ | |
3012 | parent_ctx = bdrv_child_get_parent_aio_context(new_child); | |
3013 | if (child_ctx != parent_ctx) { | |
3014 | Error *local_err = NULL; | |
142e6907 EGE |
3015 | int ret = bdrv_try_change_aio_context(child_bs, parent_ctx, NULL, |
3016 | &local_err); | |
548a74c0 | 3017 | |
f8be48ad EGE |
3018 | if (ret < 0 && child_class->change_aio_ctx) { |
3019 | Transaction *tran = tran_new(); | |
3020 | GHashTable *visited = g_hash_table_new(NULL, NULL); | |
3021 | bool ret_child; | |
3022 | ||
3023 | g_hash_table_add(visited, new_child); | |
3024 | ret_child = child_class->change_aio_ctx(new_child, child_ctx, | |
3025 | visited, tran, NULL); | |
3026 | if (ret_child == true) { | |
132ada80 KW |
3027 | error_free(local_err); |
3028 | ret = 0; | |
132ada80 | 3029 | } |
f8be48ad EGE |
3030 | tran_finalize(tran, ret_child == true ? 0 : -1); |
3031 | g_hash_table_destroy(visited); | |
132ada80 | 3032 | } |
548a74c0 | 3033 | |
132ada80 KW |
3034 | if (ret < 0) { |
3035 | error_propagate(errp, local_err); | |
04c9c3a5 | 3036 | bdrv_child_free(new_child); |
5bb04747 | 3037 | return NULL; |
132ada80 KW |
3038 | } |
3039 | } | |
3040 | ||
548a74c0 | 3041 | bdrv_ref(child_bs); |
23987471 KW |
3042 | /* |
3043 | * Let every new BdrvChild start with a drained parent. Inserting the child | |
3044 | * in the graph with bdrv_replace_child_noperm() will undrain it if | |
3045 | * @child_bs is not drained. | |
3046 | * | |
3047 | * The child was only just created and is not yet visible in global state | |
3048 | * until bdrv_replace_child_noperm() inserts it into the graph, so nobody | |
3049 | * could have sent requests and polling is not necessary. | |
3050 | * | |
3051 | * Note that this means that the parent isn't fully drained yet, we only | |
3052 | * stop new requests from coming in. This is fine, we don't care about the | |
3053 | * old requests here, they are not for this child. If another place enters a | |
3054 | * drain section for the same parent, but wants it to be fully quiesced, it | |
3055 | * will not run most of the the code in .drained_begin() again (which is not | |
3056 | * a problem, we already did this), but it will still poll until the parent | |
3057 | * is fully quiesced, so it will not be negatively affected either. | |
3058 | */ | |
606ed756 | 3059 | bdrv_parent_drained_begin_single(new_child); |
544acc7d | 3060 | bdrv_replace_child_noperm(new_child, child_bs); |
b4b059f6 | 3061 | |
548a74c0 VSO |
3062 | BdrvAttachChildCommonState *s = g_new(BdrvAttachChildCommonState, 1); |
3063 | *s = (BdrvAttachChildCommonState) { | |
5bb04747 | 3064 | .child = new_child, |
548a74c0 VSO |
3065 | .old_parent_ctx = parent_ctx, |
3066 | .old_child_ctx = child_ctx, | |
3067 | }; | |
3068 | tran_add(tran, &bdrv_attach_child_common_drv, s); | |
3069 | ||
5bb04747 | 3070 | return new_child; |
548a74c0 VSO |
3071 | } |
3072 | ||
f8d2ad78 | 3073 | /* |
7ec390d5 | 3074 | * Function doesn't update permissions, caller is responsible for this. |
f8d2ad78 | 3075 | */ |
5bb04747 VSO |
3076 | static BdrvChild *bdrv_attach_child_noperm(BlockDriverState *parent_bs, |
3077 | BlockDriverState *child_bs, | |
3078 | const char *child_name, | |
3079 | const BdrvChildClass *child_class, | |
3080 | BdrvChildRole child_role, | |
3081 | Transaction *tran, | |
3082 | Error **errp) | |
aa5a04c7 | 3083 | { |
aa5a04c7 VSO |
3084 | uint64_t perm, shared_perm; |
3085 | ||
3086 | assert(parent_bs->drv); | |
bdb73476 | 3087 | GLOBAL_STATE_CODE(); |
aa5a04c7 | 3088 | |
bfb8aa6d KW |
3089 | if (bdrv_recurse_has_child(child_bs, parent_bs)) { |
3090 | error_setg(errp, "Making '%s' a %s child of '%s' would create a cycle", | |
3091 | child_bs->node_name, child_name, parent_bs->node_name); | |
5bb04747 | 3092 | return NULL; |
bfb8aa6d KW |
3093 | } |
3094 | ||
aa5a04c7 VSO |
3095 | bdrv_get_cumulative_perm(parent_bs, &perm, &shared_perm); |
3096 | bdrv_child_perm(parent_bs, child_bs, NULL, child_role, NULL, | |
3097 | perm, shared_perm, &perm, &shared_perm); | |
3098 | ||
5bb04747 VSO |
3099 | return bdrv_attach_child_common(child_bs, child_name, child_class, |
3100 | child_role, perm, shared_perm, parent_bs, | |
3101 | tran, errp); | |
aa5a04c7 VSO |
3102 | } |
3103 | ||
548a74c0 VSO |
3104 | /* |
3105 | * This function steals the reference to child_bs from the caller. | |
3106 | * That reference is later dropped by bdrv_root_unref_child(). | |
3107 | * | |
3108 | * On failure NULL is returned, errp is set and the reference to | |
3109 | * child_bs is also dropped. | |
3110 | * | |
3111 | * The caller must hold the AioContext lock @child_bs, but not that of @ctx | |
3112 | * (unless @child_bs is already in @ctx). | |
3113 | */ | |
3114 | BdrvChild *bdrv_root_attach_child(BlockDriverState *child_bs, | |
3115 | const char *child_name, | |
3116 | const BdrvChildClass *child_class, | |
3117 | BdrvChildRole child_role, | |
3118 | uint64_t perm, uint64_t shared_perm, | |
3119 | void *opaque, Error **errp) | |
3120 | { | |
3121 | int ret; | |
5bb04747 | 3122 | BdrvChild *child; |
548a74c0 VSO |
3123 | Transaction *tran = tran_new(); |
3124 | ||
b4ad82aa EGE |
3125 | GLOBAL_STATE_CODE(); |
3126 | ||
5bb04747 | 3127 | child = bdrv_attach_child_common(child_bs, child_name, child_class, |
548a74c0 | 3128 | child_role, perm, shared_perm, opaque, |
5bb04747 VSO |
3129 | tran, errp); |
3130 | if (!child) { | |
3131 | ret = -EINVAL; | |
e878bb12 | 3132 | goto out; |
548a74c0 VSO |
3133 | } |
3134 | ||
f1316edb | 3135 | ret = bdrv_refresh_perms(child_bs, tran, errp); |
548a74c0 | 3136 | |
e878bb12 KW |
3137 | out: |
3138 | tran_finalize(tran, ret); | |
f8d2ad78 | 3139 | |
548a74c0 | 3140 | bdrv_unref(child_bs); |
5bb04747 VSO |
3141 | |
3142 | return ret < 0 ? NULL : child; | |
df581792 KW |
3143 | } |
3144 | ||
b441dc71 AG |
3145 | /* |
3146 | * This function transfers the reference to child_bs from the caller | |
3147 | * to parent_bs. That reference is later dropped by parent_bs on | |
3148 | * bdrv_close() or if someone calls bdrv_unref_child(). | |
3149 | * | |
3150 | * On failure NULL is returned, errp is set and the reference to | |
3151 | * child_bs is also dropped. | |
132ada80 KW |
3152 | * |
3153 | * If @parent_bs and @child_bs are in different AioContexts, the caller must | |
3154 | * hold the AioContext lock for @child_bs, but not for @parent_bs. | |
b441dc71 | 3155 | */ |
98292c61 WC |
3156 | BdrvChild *bdrv_attach_child(BlockDriverState *parent_bs, |
3157 | BlockDriverState *child_bs, | |
3158 | const char *child_name, | |
bd86fb99 | 3159 | const BdrvChildClass *child_class, |
258b7765 | 3160 | BdrvChildRole child_role, |
8b2ff529 | 3161 | Error **errp) |
f21d96d0 | 3162 | { |
aa5a04c7 | 3163 | int ret; |
5bb04747 | 3164 | BdrvChild *child; |
aa5a04c7 | 3165 | Transaction *tran = tran_new(); |
f68c598b | 3166 | |
f791bf7f EGE |
3167 | GLOBAL_STATE_CODE(); |
3168 | ||
5bb04747 VSO |
3169 | child = bdrv_attach_child_noperm(parent_bs, child_bs, child_name, |
3170 | child_class, child_role, tran, errp); | |
3171 | if (!child) { | |
3172 | ret = -EINVAL; | |
aa5a04c7 VSO |
3173 | goto out; |
3174 | } | |
d5e6f437 | 3175 | |
f1316edb | 3176 | ret = bdrv_refresh_perms(parent_bs, tran, errp); |
aa5a04c7 VSO |
3177 | if (ret < 0) { |
3178 | goto out; | |
d5e6f437 KW |
3179 | } |
3180 | ||
aa5a04c7 VSO |
3181 | out: |
3182 | tran_finalize(tran, ret); | |
3183 | ||
3184 | bdrv_unref(child_bs); | |
3185 | ||
5bb04747 | 3186 | return ret < 0 ? NULL : child; |
f21d96d0 KW |
3187 | } |
3188 | ||
7b99a266 | 3189 | /* Callers must ensure that child->frozen is false. */ |
f21d96d0 | 3190 | void bdrv_root_unref_child(BdrvChild *child) |
33a60407 | 3191 | { |
00eb93b5 | 3192 | BlockDriverState *child_bs = child->bs; |
779020cb | 3193 | |
f791bf7f | 3194 | GLOBAL_STATE_CODE(); |
00eb93b5 VSO |
3195 | bdrv_replace_child_noperm(child, NULL); |
3196 | bdrv_child_free(child); | |
3197 | ||
3198 | if (child_bs) { | |
3199 | /* | |
3200 | * Update permissions for old node. We're just taking a parent away, so | |
3201 | * we're loosening restrictions. Errors of permission update are not | |
3202 | * fatal in this case, ignore them. | |
3203 | */ | |
f1316edb | 3204 | bdrv_refresh_perms(child_bs, NULL, NULL); |
00eb93b5 VSO |
3205 | |
3206 | /* | |
3207 | * When the parent requiring a non-default AioContext is removed, the | |
3208 | * node moves back to the main AioContext | |
3209 | */ | |
3210 | bdrv_try_change_aio_context(child_bs, qemu_get_aio_context(), NULL, | |
3211 | NULL); | |
3212 | } | |
f791bf7f | 3213 | |
f21d96d0 KW |
3214 | bdrv_unref(child_bs); |
3215 | } | |
3216 | ||
332b3a17 VSO |
3217 | typedef struct BdrvSetInheritsFrom { |
3218 | BlockDriverState *bs; | |
3219 | BlockDriverState *old_inherits_from; | |
3220 | } BdrvSetInheritsFrom; | |
3221 | ||
3222 | static void bdrv_set_inherits_from_abort(void *opaque) | |
3223 | { | |
3224 | BdrvSetInheritsFrom *s = opaque; | |
3225 | ||
3226 | s->bs->inherits_from = s->old_inherits_from; | |
3227 | } | |
3228 | ||
3229 | static TransactionActionDrv bdrv_set_inherits_from_drv = { | |
3230 | .abort = bdrv_set_inherits_from_abort, | |
3231 | .clean = g_free, | |
3232 | }; | |
3233 | ||
3234 | /* @tran is allowed to be NULL. In this case no rollback is possible */ | |
3235 | static void bdrv_set_inherits_from(BlockDriverState *bs, | |
3236 | BlockDriverState *new_inherits_from, | |
3237 | Transaction *tran) | |
3238 | { | |
3239 | if (tran) { | |
3240 | BdrvSetInheritsFrom *s = g_new(BdrvSetInheritsFrom, 1); | |
3241 | ||
3242 | *s = (BdrvSetInheritsFrom) { | |
3243 | .bs = bs, | |
3244 | .old_inherits_from = bs->inherits_from, | |
3245 | }; | |
3246 | ||
3247 | tran_add(tran, &bdrv_set_inherits_from_drv, s); | |
3248 | } | |
3249 | ||
3250 | bs->inherits_from = new_inherits_from; | |
3251 | } | |
3252 | ||
3cf746b3 HR |
3253 | /** |
3254 | * Clear all inherits_from pointers from children and grandchildren of | |
3255 | * @root that point to @root, where necessary. | |
332b3a17 | 3256 | * @tran is allowed to be NULL. In this case no rollback is possible |
3cf746b3 | 3257 | */ |
332b3a17 VSO |
3258 | static void bdrv_unset_inherits_from(BlockDriverState *root, BdrvChild *child, |
3259 | Transaction *tran) | |
f21d96d0 | 3260 | { |
3cf746b3 | 3261 | BdrvChild *c; |
4e4bf5c4 | 3262 | |
3cf746b3 HR |
3263 | if (child->bs->inherits_from == root) { |
3264 | /* | |
3265 | * Remove inherits_from only when the last reference between root and | |
3266 | * child->bs goes away. | |
3267 | */ | |
3268 | QLIST_FOREACH(c, &root->children, next) { | |
4e4bf5c4 KW |
3269 | if (c != child && c->bs == child->bs) { |
3270 | break; | |
3271 | } | |
3272 | } | |
3273 | if (c == NULL) { | |
332b3a17 | 3274 | bdrv_set_inherits_from(child->bs, NULL, tran); |
4e4bf5c4 | 3275 | } |
33a60407 KW |
3276 | } |
3277 | ||
3cf746b3 | 3278 | QLIST_FOREACH(c, &child->bs->children, next) { |
332b3a17 | 3279 | bdrv_unset_inherits_from(root, c, tran); |
3cf746b3 HR |
3280 | } |
3281 | } | |
3282 | ||
7b99a266 | 3283 | /* Callers must ensure that child->frozen is false. */ |
3cf746b3 HR |
3284 | void bdrv_unref_child(BlockDriverState *parent, BdrvChild *child) |
3285 | { | |
f791bf7f | 3286 | GLOBAL_STATE_CODE(); |
3cf746b3 HR |
3287 | if (child == NULL) { |
3288 | return; | |
3289 | } | |
3290 | ||
332b3a17 | 3291 | bdrv_unset_inherits_from(parent, child, NULL); |
f21d96d0 | 3292 | bdrv_root_unref_child(child); |
33a60407 KW |
3293 | } |
3294 | ||
5c8cab48 KW |
3295 | |
3296 | static void bdrv_parent_cb_change_media(BlockDriverState *bs, bool load) | |
3297 | { | |
3298 | BdrvChild *c; | |
f0c28327 | 3299 | GLOBAL_STATE_CODE(); |
5c8cab48 | 3300 | QLIST_FOREACH(c, &bs->parents, next_parent) { |
bd86fb99 HR |
3301 | if (c->klass->change_media) { |
3302 | c->klass->change_media(c, load); | |
5c8cab48 KW |
3303 | } |
3304 | } | |
3305 | } | |
3306 | ||
0065c455 AG |
3307 | /* Return true if you can reach parent going through child->inherits_from |
3308 | * recursively. If parent or child are NULL, return false */ | |
3309 | static bool bdrv_inherits_from_recursive(BlockDriverState *child, | |
3310 | BlockDriverState *parent) | |
3311 | { | |
3312 | while (child && child != parent) { | |
3313 | child = child->inherits_from; | |
3314 | } | |
3315 | ||
3316 | return child != NULL; | |
3317 | } | |
3318 | ||
25191e5f HR |
3319 | /* |
3320 | * Return the BdrvChildRole for @bs's backing child. bs->backing is | |
3321 | * mostly used for COW backing children (role = COW), but also for | |
3322 | * filtered children (role = FILTERED | PRIMARY). | |
3323 | */ | |
3324 | static BdrvChildRole bdrv_backing_role(BlockDriverState *bs) | |
3325 | { | |
3326 | if (bs->drv && bs->drv->is_filter) { | |
3327 | return BDRV_CHILD_FILTERED | BDRV_CHILD_PRIMARY; | |
3328 | } else { | |
3329 | return BDRV_CHILD_COW; | |
3330 | } | |
3331 | } | |
3332 | ||
5db15a57 | 3333 | /* |
e9238278 VSO |
3334 | * Sets the bs->backing or bs->file link of a BDS. A new reference is created; |
3335 | * callers which don't need their own reference any more must call bdrv_unref(). | |
7ec390d5 VSO |
3336 | * |
3337 | * Function doesn't update permissions, caller is responsible for this. | |
5db15a57 | 3338 | */ |
e9238278 VSO |
3339 | static int bdrv_set_file_or_backing_noperm(BlockDriverState *parent_bs, |
3340 | BlockDriverState *child_bs, | |
3341 | bool is_backing, | |
3342 | Transaction *tran, Error **errp) | |
8d24cce1 | 3343 | { |
e9238278 VSO |
3344 | bool update_inherits_from = |
3345 | bdrv_inherits_from_recursive(child_bs, parent_bs); | |
3346 | BdrvChild *child = is_backing ? parent_bs->backing : parent_bs->file; | |
3347 | BdrvChildRole role; | |
0065c455 | 3348 | |
bdb73476 EGE |
3349 | GLOBAL_STATE_CODE(); |
3350 | ||
e9238278 VSO |
3351 | if (!parent_bs->drv) { |
3352 | /* | |
3353 | * Node without drv is an object without a class :/. TODO: finally fix | |
3354 | * qcow2 driver to never clear bs->drv and implement format corruption | |
3355 | * handling in other way. | |
3356 | */ | |
3357 | error_setg(errp, "Node corrupted"); | |
3358 | return -EINVAL; | |
3359 | } | |
3360 | ||
3361 | if (child && child->frozen) { | |
3362 | error_setg(errp, "Cannot change frozen '%s' link from '%s' to '%s'", | |
3363 | child->name, parent_bs->node_name, child->bs->node_name); | |
a1e708fc | 3364 | return -EPERM; |
2cad1ebe AG |
3365 | } |
3366 | ||
25f78d9e VSO |
3367 | if (is_backing && !parent_bs->drv->is_filter && |
3368 | !parent_bs->drv->supports_backing) | |
3369 | { | |
3370 | error_setg(errp, "Driver '%s' of node '%s' does not support backing " | |
3371 | "files", parent_bs->drv->format_name, parent_bs->node_name); | |
3372 | return -EINVAL; | |
3373 | } | |
3374 | ||
e9238278 VSO |
3375 | if (parent_bs->drv->is_filter) { |
3376 | role = BDRV_CHILD_FILTERED | BDRV_CHILD_PRIMARY; | |
3377 | } else if (is_backing) { | |
3378 | role = BDRV_CHILD_COW; | |
3379 | } else { | |
3380 | /* | |
3381 | * We only can use same role as it is in existing child. We don't have | |
3382 | * infrastructure to determine role of file child in generic way | |
3383 | */ | |
3384 | if (!child) { | |
3385 | error_setg(errp, "Cannot set file child to format node without " | |
3386 | "file child"); | |
3387 | return -EINVAL; | |
3388 | } | |
3389 | role = child->role; | |
826b6ca0 FZ |
3390 | } |
3391 | ||
e9238278 VSO |
3392 | if (child) { |
3393 | bdrv_unset_inherits_from(parent_bs, child, tran); | |
57f08941 | 3394 | bdrv_remove_child(child, tran); |
e9238278 VSO |
3395 | } |
3396 | ||
3397 | if (!child_bs) { | |
8d24cce1 FZ |
3398 | goto out; |
3399 | } | |
12fa4af6 | 3400 | |
5bb04747 VSO |
3401 | child = bdrv_attach_child_noperm(parent_bs, child_bs, |
3402 | is_backing ? "backing" : "file", | |
3403 | &child_of_bds, role, | |
3404 | tran, errp); | |
3405 | if (!child) { | |
3406 | return -EINVAL; | |
a1e708fc VSO |
3407 | } |
3408 | ||
160333e1 VSO |
3409 | |
3410 | /* | |
e9238278 | 3411 | * If inherits_from pointed recursively to bs then let's update it to |
160333e1 VSO |
3412 | * point directly to bs (else it will become NULL). |
3413 | */ | |
a1e708fc | 3414 | if (update_inherits_from) { |
e9238278 | 3415 | bdrv_set_inherits_from(child_bs, parent_bs, tran); |
0065c455 | 3416 | } |
826b6ca0 | 3417 | |
8d24cce1 | 3418 | out: |
e9238278 | 3419 | bdrv_refresh_limits(parent_bs, tran, NULL); |
160333e1 VSO |
3420 | |
3421 | return 0; | |
3422 | } | |
3423 | ||
e9238278 VSO |
3424 | static int bdrv_set_backing_noperm(BlockDriverState *bs, |
3425 | BlockDriverState *backing_hd, | |
3426 | Transaction *tran, Error **errp) | |
3427 | { | |
bdb73476 | 3428 | GLOBAL_STATE_CODE(); |
e9238278 VSO |
3429 | return bdrv_set_file_or_backing_noperm(bs, backing_hd, true, tran, errp); |
3430 | } | |
3431 | ||
92140b9f KW |
3432 | int bdrv_set_backing_hd_drained(BlockDriverState *bs, |
3433 | BlockDriverState *backing_hd, | |
3434 | Error **errp) | |
160333e1 VSO |
3435 | { |
3436 | int ret; | |
3437 | Transaction *tran = tran_new(); | |
3438 | ||
f791bf7f | 3439 | GLOBAL_STATE_CODE(); |
92140b9f | 3440 | assert(bs->quiesce_counter > 0); |
c0829cb1 | 3441 | |
160333e1 VSO |
3442 | ret = bdrv_set_backing_noperm(bs, backing_hd, tran, errp); |
3443 | if (ret < 0) { | |
3444 | goto out; | |
3445 | } | |
3446 | ||
f1316edb | 3447 | ret = bdrv_refresh_perms(bs, tran, errp); |
160333e1 VSO |
3448 | out: |
3449 | tran_finalize(tran, ret); | |
92140b9f KW |
3450 | return ret; |
3451 | } | |
a1e708fc | 3452 | |
92140b9f KW |
3453 | int bdrv_set_backing_hd(BlockDriverState *bs, BlockDriverState *backing_hd, |
3454 | Error **errp) | |
3455 | { | |
3456 | int ret; | |
3457 | GLOBAL_STATE_CODE(); | |
3458 | ||
3459 | bdrv_drained_begin(bs); | |
3460 | ret = bdrv_set_backing_hd_drained(bs, backing_hd, errp); | |
c0829cb1 VSO |
3461 | bdrv_drained_end(bs); |
3462 | ||
a1e708fc | 3463 | return ret; |
8d24cce1 FZ |
3464 | } |
3465 | ||
31ca6d07 KW |
3466 | /* |
3467 | * Opens the backing file for a BlockDriverState if not yet open | |
3468 | * | |
d9b7b057 KW |
3469 | * bdref_key specifies the key for the image's BlockdevRef in the options QDict. |
3470 | * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict | |
3471 | * itself, all options starting with "${bdref_key}." are considered part of the | |
3472 | * BlockdevRef. | |
3473 | * | |
3474 | * TODO Can this be unified with bdrv_open_image()? | |
31ca6d07 | 3475 | */ |
d9b7b057 KW |
3476 | int bdrv_open_backing_file(BlockDriverState *bs, QDict *parent_options, |
3477 | const char *bdref_key, Error **errp) | |
9156df12 | 3478 | { |
6b6833c1 | 3479 | char *backing_filename = NULL; |
d9b7b057 KW |
3480 | char *bdref_key_dot; |
3481 | const char *reference = NULL; | |
317fc44e | 3482 | int ret = 0; |
998c2019 | 3483 | bool implicit_backing = false; |
8d24cce1 | 3484 | BlockDriverState *backing_hd; |
d9b7b057 KW |
3485 | QDict *options; |
3486 | QDict *tmp_parent_options = NULL; | |
34b5d2c6 | 3487 | Error *local_err = NULL; |
9156df12 | 3488 | |
f791bf7f EGE |
3489 | GLOBAL_STATE_CODE(); |
3490 | ||
760e0063 | 3491 | if (bs->backing != NULL) { |
1ba4b6a5 | 3492 | goto free_exit; |
9156df12 PB |
3493 | } |
3494 | ||
31ca6d07 | 3495 | /* NULL means an empty set of options */ |
d9b7b057 KW |
3496 | if (parent_options == NULL) { |
3497 | tmp_parent_options = qdict_new(); | |
3498 | parent_options = tmp_parent_options; | |
31ca6d07 KW |
3499 | } |
3500 | ||
9156df12 | 3501 | bs->open_flags &= ~BDRV_O_NO_BACKING; |
d9b7b057 KW |
3502 | |
3503 | bdref_key_dot = g_strdup_printf("%s.", bdref_key); | |
3504 | qdict_extract_subqdict(parent_options, &options, bdref_key_dot); | |
3505 | g_free(bdref_key_dot); | |
3506 | ||
129c7d1c MA |
3507 | /* |
3508 | * Caution: while qdict_get_try_str() is fine, getting non-string | |
3509 | * types would require more care. When @parent_options come from | |
3510 | * -blockdev or blockdev_add, its members are typed according to | |
3511 | * the QAPI schema, but when they come from -drive, they're all | |
3512 | * QString. | |
3513 | */ | |
d9b7b057 KW |
3514 | reference = qdict_get_try_str(parent_options, bdref_key); |
3515 | if (reference || qdict_haskey(options, "file.filename")) { | |
6b6833c1 | 3516 | /* keep backing_filename NULL */ |
1cb6f506 | 3517 | } else if (bs->backing_file[0] == '\0' && qdict_size(options) == 0) { |
cb3e7f08 | 3518 | qobject_unref(options); |
1ba4b6a5 | 3519 | goto free_exit; |
dbecebdd | 3520 | } else { |
998c2019 HR |
3521 | if (qdict_size(options) == 0) { |
3522 | /* If the user specifies options that do not modify the | |
3523 | * backing file's behavior, we might still consider it the | |
3524 | * implicit backing file. But it's easier this way, and | |
3525 | * just specifying some of the backing BDS's options is | |
3526 | * only possible with -drive anyway (otherwise the QAPI | |
3527 | * schema forces the user to specify everything). */ | |
3528 | implicit_backing = !strcmp(bs->auto_backing_file, bs->backing_file); | |
3529 | } | |
3530 | ||
6b6833c1 | 3531 | backing_filename = bdrv_get_full_backing_filename(bs, &local_err); |
9f07429e HR |
3532 | if (local_err) { |
3533 | ret = -EINVAL; | |
3534 | error_propagate(errp, local_err); | |
cb3e7f08 | 3535 | qobject_unref(options); |
9f07429e HR |
3536 | goto free_exit; |
3537 | } | |
9156df12 PB |
3538 | } |
3539 | ||
8ee79e70 KW |
3540 | if (!bs->drv || !bs->drv->supports_backing) { |
3541 | ret = -EINVAL; | |
3542 | error_setg(errp, "Driver doesn't support backing files"); | |
cb3e7f08 | 3543 | qobject_unref(options); |
8ee79e70 KW |
3544 | goto free_exit; |
3545 | } | |
3546 | ||
6bff597b PK |
3547 | if (!reference && |
3548 | bs->backing_format[0] != '\0' && !qdict_haskey(options, "driver")) { | |
46f5ac20 | 3549 | qdict_put_str(options, "driver", bs->backing_format); |
9156df12 PB |
3550 | } |
3551 | ||
6b6833c1 | 3552 | backing_hd = bdrv_open_inherit(backing_filename, reference, options, 0, bs, |
25191e5f | 3553 | &child_of_bds, bdrv_backing_role(bs), errp); |
5b363937 | 3554 | if (!backing_hd) { |
9156df12 | 3555 | bs->open_flags |= BDRV_O_NO_BACKING; |
e43bfd9c | 3556 | error_prepend(errp, "Could not open backing file: "); |
5b363937 | 3557 | ret = -EINVAL; |
1ba4b6a5 | 3558 | goto free_exit; |
9156df12 | 3559 | } |
df581792 | 3560 | |
998c2019 HR |
3561 | if (implicit_backing) { |
3562 | bdrv_refresh_filename(backing_hd); | |
3563 | pstrcpy(bs->auto_backing_file, sizeof(bs->auto_backing_file), | |
3564 | backing_hd->filename); | |
3565 | } | |
3566 | ||
5db15a57 KW |
3567 | /* Hook up the backing file link; drop our reference, bs owns the |
3568 | * backing_hd reference now */ | |
dc9c10a1 | 3569 | ret = bdrv_set_backing_hd(bs, backing_hd, errp); |
5db15a57 | 3570 | bdrv_unref(backing_hd); |
dc9c10a1 | 3571 | if (ret < 0) { |
12fa4af6 KW |
3572 | goto free_exit; |
3573 | } | |
d80ac658 | 3574 | |
d9b7b057 KW |
3575 | qdict_del(parent_options, bdref_key); |
3576 | ||
1ba4b6a5 BC |
3577 | free_exit: |
3578 | g_free(backing_filename); | |
cb3e7f08 | 3579 | qobject_unref(tmp_parent_options); |
1ba4b6a5 | 3580 | return ret; |
9156df12 PB |
3581 | } |
3582 | ||
2d6b86af KW |
3583 | static BlockDriverState * |
3584 | bdrv_open_child_bs(const char *filename, QDict *options, const char *bdref_key, | |
bd86fb99 | 3585 | BlockDriverState *parent, const BdrvChildClass *child_class, |
272c02ea | 3586 | BdrvChildRole child_role, bool allow_none, Error **errp) |
da557aac | 3587 | { |
2d6b86af | 3588 | BlockDriverState *bs = NULL; |
da557aac | 3589 | QDict *image_options; |
da557aac HR |
3590 | char *bdref_key_dot; |
3591 | const char *reference; | |
3592 | ||
bd86fb99 | 3593 | assert(child_class != NULL); |
f67503e5 | 3594 | |
da557aac HR |
3595 | bdref_key_dot = g_strdup_printf("%s.", bdref_key); |
3596 | qdict_extract_subqdict(options, &image_options, bdref_key_dot); | |
3597 | g_free(bdref_key_dot); | |
3598 | ||
129c7d1c MA |
3599 | /* |
3600 | * Caution: while qdict_get_try_str() is fine, getting non-string | |
3601 | * types would require more care. When @options come from | |
3602 | * -blockdev or blockdev_add, its members are typed according to | |
3603 | * the QAPI schema, but when they come from -drive, they're all | |
3604 | * QString. | |
3605 | */ | |
da557aac HR |
3606 | reference = qdict_get_try_str(options, bdref_key); |
3607 | if (!filename && !reference && !qdict_size(image_options)) { | |
b4b059f6 | 3608 | if (!allow_none) { |
da557aac HR |
3609 | error_setg(errp, "A block device must be specified for \"%s\"", |
3610 | bdref_key); | |
da557aac | 3611 | } |
cb3e7f08 | 3612 | qobject_unref(image_options); |
da557aac HR |
3613 | goto done; |
3614 | } | |
3615 | ||
5b363937 | 3616 | bs = bdrv_open_inherit(filename, reference, image_options, 0, |
272c02ea | 3617 | parent, child_class, child_role, errp); |
5b363937 | 3618 | if (!bs) { |
df581792 KW |
3619 | goto done; |
3620 | } | |
3621 | ||
da557aac HR |
3622 | done: |
3623 | qdict_del(options, bdref_key); | |
2d6b86af KW |
3624 | return bs; |
3625 | } | |
3626 | ||
3627 | /* | |
3628 | * Opens a disk image whose options are given as BlockdevRef in another block | |
3629 | * device's options. | |
3630 | * | |
3631 | * If allow_none is true, no image will be opened if filename is false and no | |
3632 | * BlockdevRef is given. NULL will be returned, but errp remains unset. | |
3633 | * | |
3634 | * bdrev_key specifies the key for the image's BlockdevRef in the options QDict. | |
3635 | * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict | |
3636 | * itself, all options starting with "${bdref_key}." are considered part of the | |
3637 | * BlockdevRef. | |
3638 | * | |
3639 | * The BlockdevRef will be removed from the options QDict. | |
3640 | */ | |
3641 | BdrvChild *bdrv_open_child(const char *filename, | |
3642 | QDict *options, const char *bdref_key, | |
3643 | BlockDriverState *parent, | |
bd86fb99 | 3644 | const BdrvChildClass *child_class, |
258b7765 | 3645 | BdrvChildRole child_role, |
2d6b86af KW |
3646 | bool allow_none, Error **errp) |
3647 | { | |
3648 | BlockDriverState *bs; | |
3649 | ||
f791bf7f EGE |
3650 | GLOBAL_STATE_CODE(); |
3651 | ||
bd86fb99 | 3652 | bs = bdrv_open_child_bs(filename, options, bdref_key, parent, child_class, |
272c02ea | 3653 | child_role, allow_none, errp); |
2d6b86af KW |
3654 | if (bs == NULL) { |
3655 | return NULL; | |
3656 | } | |
3657 | ||
258b7765 HR |
3658 | return bdrv_attach_child(parent, bs, bdref_key, child_class, child_role, |
3659 | errp); | |
b4b059f6 KW |
3660 | } |
3661 | ||
83930780 VSO |
3662 | /* |
3663 | * Wrapper on bdrv_open_child() for most popular case: open primary child of bs. | |
3664 | */ | |
3665 | int bdrv_open_file_child(const char *filename, | |
3666 | QDict *options, const char *bdref_key, | |
3667 | BlockDriverState *parent, Error **errp) | |
3668 | { | |
3669 | BdrvChildRole role; | |
3670 | ||
3671 | /* commit_top and mirror_top don't use this function */ | |
3672 | assert(!parent->drv->filtered_child_is_backing); | |
83930780 VSO |
3673 | role = parent->drv->is_filter ? |
3674 | (BDRV_CHILD_FILTERED | BDRV_CHILD_PRIMARY) : BDRV_CHILD_IMAGE; | |
3675 | ||
5bb04747 VSO |
3676 | if (!bdrv_open_child(filename, options, bdref_key, parent, |
3677 | &child_of_bds, role, false, errp)) | |
3678 | { | |
3679 | return -EINVAL; | |
3680 | } | |
83930780 | 3681 | |
5bb04747 | 3682 | return 0; |
83930780 VSO |
3683 | } |
3684 | ||
bd86fb99 HR |
3685 | /* |
3686 | * TODO Future callers may need to specify parent/child_class in order for | |
3687 | * option inheritance to work. Existing callers use it for the root node. | |
3688 | */ | |
e1d74bc6 KW |
3689 | BlockDriverState *bdrv_open_blockdev_ref(BlockdevRef *ref, Error **errp) |
3690 | { | |
3691 | BlockDriverState *bs = NULL; | |
e1d74bc6 KW |
3692 | QObject *obj = NULL; |
3693 | QDict *qdict = NULL; | |
3694 | const char *reference = NULL; | |
3695 | Visitor *v = NULL; | |
3696 | ||
f791bf7f EGE |
3697 | GLOBAL_STATE_CODE(); |
3698 | ||
e1d74bc6 KW |
3699 | if (ref->type == QTYPE_QSTRING) { |
3700 | reference = ref->u.reference; | |
3701 | } else { | |
3702 | BlockdevOptions *options = &ref->u.definition; | |
3703 | assert(ref->type == QTYPE_QDICT); | |
3704 | ||
3705 | v = qobject_output_visitor_new(&obj); | |
1f584248 | 3706 | visit_type_BlockdevOptions(v, NULL, &options, &error_abort); |
e1d74bc6 KW |
3707 | visit_complete(v, &obj); |
3708 | ||
7dc847eb | 3709 | qdict = qobject_to(QDict, obj); |
e1d74bc6 KW |
3710 | qdict_flatten(qdict); |
3711 | ||
3712 | /* bdrv_open_inherit() defaults to the values in bdrv_flags (for | |
3713 | * compatibility with other callers) rather than what we want as the | |
3714 | * real defaults. Apply the defaults here instead. */ | |
3715 | qdict_set_default_str(qdict, BDRV_OPT_CACHE_DIRECT, "off"); | |
3716 | qdict_set_default_str(qdict, BDRV_OPT_CACHE_NO_FLUSH, "off"); | |
3717 | qdict_set_default_str(qdict, BDRV_OPT_READ_ONLY, "off"); | |
e35bdc12 KW |
3718 | qdict_set_default_str(qdict, BDRV_OPT_AUTO_READ_ONLY, "off"); |
3719 | ||
e1d74bc6 KW |
3720 | } |
3721 | ||
272c02ea | 3722 | bs = bdrv_open_inherit(NULL, reference, qdict, 0, NULL, NULL, 0, errp); |
e1d74bc6 | 3723 | obj = NULL; |
cb3e7f08 | 3724 | qobject_unref(obj); |
e1d74bc6 KW |
3725 | visit_free(v); |
3726 | return bs; | |
3727 | } | |
3728 | ||
66836189 HR |
3729 | static BlockDriverState *bdrv_append_temp_snapshot(BlockDriverState *bs, |
3730 | int flags, | |
3731 | QDict *snapshot_options, | |
3732 | Error **errp) | |
b998875d | 3733 | { |
69fbfff9 | 3734 | g_autofree char *tmp_filename = NULL; |
b998875d | 3735 | int64_t total_size; |
83d0521a | 3736 | QemuOpts *opts = NULL; |
ff6ed714 | 3737 | BlockDriverState *bs_snapshot = NULL; |
b998875d KW |
3738 | int ret; |
3739 | ||
bdb73476 EGE |
3740 | GLOBAL_STATE_CODE(); |
3741 | ||
b998875d KW |
3742 | /* if snapshot, we create a temporary backing file and open it |
3743 | instead of opening 'filename' directly */ | |
3744 | ||
3745 | /* Get the required size from the image */ | |
f187743a KW |
3746 | total_size = bdrv_getlength(bs); |
3747 | if (total_size < 0) { | |
3748 | error_setg_errno(errp, -total_size, "Could not get image size"); | |
1ba4b6a5 | 3749 | goto out; |
f187743a | 3750 | } |
b998875d KW |
3751 | |
3752 | /* Create the temporary image */ | |
69fbfff9 BM |
3753 | tmp_filename = create_tmp_file(errp); |
3754 | if (!tmp_filename) { | |
1ba4b6a5 | 3755 | goto out; |
b998875d KW |
3756 | } |
3757 | ||
ef810437 | 3758 | opts = qemu_opts_create(bdrv_qcow2.create_opts, NULL, 0, |
c282e1fd | 3759 | &error_abort); |
39101f25 | 3760 | qemu_opt_set_number(opts, BLOCK_OPT_SIZE, total_size, &error_abort); |
e43bfd9c | 3761 | ret = bdrv_create(&bdrv_qcow2, tmp_filename, opts, errp); |
83d0521a | 3762 | qemu_opts_del(opts); |
b998875d | 3763 | if (ret < 0) { |
e43bfd9c MA |
3764 | error_prepend(errp, "Could not create temporary overlay '%s': ", |
3765 | tmp_filename); | |
1ba4b6a5 | 3766 | goto out; |
b998875d KW |
3767 | } |
3768 | ||
73176bee | 3769 | /* Prepare options QDict for the temporary file */ |
46f5ac20 EB |
3770 | qdict_put_str(snapshot_options, "file.driver", "file"); |
3771 | qdict_put_str(snapshot_options, "file.filename", tmp_filename); | |
3772 | qdict_put_str(snapshot_options, "driver", "qcow2"); | |
b998875d | 3773 | |
5b363937 | 3774 | bs_snapshot = bdrv_open(NULL, NULL, snapshot_options, flags, errp); |
73176bee | 3775 | snapshot_options = NULL; |
5b363937 | 3776 | if (!bs_snapshot) { |
1ba4b6a5 | 3777 | goto out; |
b998875d KW |
3778 | } |
3779 | ||
934aee14 VSO |
3780 | ret = bdrv_append(bs_snapshot, bs, errp); |
3781 | if (ret < 0) { | |
ff6ed714 | 3782 | bs_snapshot = NULL; |
b2c2832c KW |
3783 | goto out; |
3784 | } | |
1ba4b6a5 BC |
3785 | |
3786 | out: | |
cb3e7f08 | 3787 | qobject_unref(snapshot_options); |
ff6ed714 | 3788 | return bs_snapshot; |
b998875d KW |
3789 | } |
3790 | ||
b6ce07aa KW |
3791 | /* |
3792 | * Opens a disk image (raw, qcow2, vmdk, ...) | |
de9c0cec KW |
3793 | * |
3794 | * options is a QDict of options to pass to the block drivers, or NULL for an | |
3795 | * empty set of options. The reference to the QDict belongs to the block layer | |
3796 | * after the call (even on failure), so if the caller intends to reuse the | |
cb3e7f08 | 3797 | * dictionary, it needs to use qobject_ref() before calling bdrv_open. |
f67503e5 HR |
3798 | * |
3799 | * If *pbs is NULL, a new BDS will be created with a pointer to it stored there. | |
3800 | * If it is not NULL, the referenced BDS will be reused. | |
ddf5636d HR |
3801 | * |
3802 | * The reference parameter may be used to specify an existing block device which | |
3803 | * should be opened. If specified, neither options nor a filename may be given, | |
3804 | * nor can an existing BDS be reused (that is, *pbs has to be NULL). | |
c86422c5 EGE |
3805 | * |
3806 | * The caller must always hold @filename AioContext lock, because this | |
3807 | * function eventually calls bdrv_refresh_total_sectors() which polls | |
3808 | * when called from non-coroutine context. | |
b6ce07aa | 3809 | */ |
5b363937 HR |
3810 | static BlockDriverState *bdrv_open_inherit(const char *filename, |
3811 | const char *reference, | |
3812 | QDict *options, int flags, | |
3813 | BlockDriverState *parent, | |
bd86fb99 | 3814 | const BdrvChildClass *child_class, |
272c02ea | 3815 | BdrvChildRole child_role, |
5b363937 | 3816 | Error **errp) |
ea2384d3 | 3817 | { |
b6ce07aa | 3818 | int ret; |
5696c6e3 | 3819 | BlockBackend *file = NULL; |
9a4f4c31 | 3820 | BlockDriverState *bs; |
ce343771 | 3821 | BlockDriver *drv = NULL; |
2f624b80 | 3822 | BdrvChild *child; |
74fe54f2 | 3823 | const char *drvname; |
3e8c2e57 | 3824 | const char *backing; |
34b5d2c6 | 3825 | Error *local_err = NULL; |
73176bee | 3826 | QDict *snapshot_options = NULL; |
b1e6fc08 | 3827 | int snapshot_flags = 0; |
712e7874 | 3828 | |
bd86fb99 HR |
3829 | assert(!child_class || !flags); |
3830 | assert(!child_class == !parent); | |
f0c28327 | 3831 | GLOBAL_STATE_CODE(); |
f67503e5 | 3832 | |
ddf5636d HR |
3833 | if (reference) { |
3834 | bool options_non_empty = options ? qdict_size(options) : false; | |
cb3e7f08 | 3835 | qobject_unref(options); |
ddf5636d | 3836 | |
ddf5636d HR |
3837 | if (filename || options_non_empty) { |
3838 | error_setg(errp, "Cannot reference an existing block device with " | |
3839 | "additional options or a new filename"); | |
5b363937 | 3840 | return NULL; |
ddf5636d HR |
3841 | } |
3842 | ||
3843 | bs = bdrv_lookup_bs(reference, reference, errp); | |
3844 | if (!bs) { | |
5b363937 | 3845 | return NULL; |
ddf5636d | 3846 | } |
76b22320 | 3847 | |
ddf5636d | 3848 | bdrv_ref(bs); |
5b363937 | 3849 | return bs; |
ddf5636d HR |
3850 | } |
3851 | ||
5b363937 | 3852 | bs = bdrv_new(); |
f67503e5 | 3853 | |
de9c0cec KW |
3854 | /* NULL means an empty set of options */ |
3855 | if (options == NULL) { | |
3856 | options = qdict_new(); | |
3857 | } | |
3858 | ||
145f598e | 3859 | /* json: syntax counts as explicit options, as if in the QDict */ |
de3b53f0 KW |
3860 | parse_json_protocol(options, &filename, &local_err); |
3861 | if (local_err) { | |
de3b53f0 KW |
3862 | goto fail; |
3863 | } | |
3864 | ||
145f598e KW |
3865 | bs->explicit_options = qdict_clone_shallow(options); |
3866 | ||
bd86fb99 | 3867 | if (child_class) { |
3cdc69d3 HR |
3868 | bool parent_is_format; |
3869 | ||
3870 | if (parent->drv) { | |
3871 | parent_is_format = parent->drv->is_format; | |
3872 | } else { | |
3873 | /* | |
3874 | * parent->drv is not set yet because this node is opened for | |
3875 | * (potential) format probing. That means that @parent is going | |
3876 | * to be a format node. | |
3877 | */ | |
3878 | parent_is_format = true; | |
3879 | } | |
3880 | ||
bddcec37 | 3881 | bs->inherits_from = parent; |
3cdc69d3 HR |
3882 | child_class->inherit_options(child_role, parent_is_format, |
3883 | &flags, options, | |
bd86fb99 | 3884 | parent->open_flags, parent->options); |
f3930ed0 KW |
3885 | } |
3886 | ||
de3b53f0 | 3887 | ret = bdrv_fill_options(&options, filename, &flags, &local_err); |
dfde483e | 3888 | if (ret < 0) { |
462f5bcf KW |
3889 | goto fail; |
3890 | } | |
3891 | ||
129c7d1c MA |
3892 | /* |
3893 | * Set the BDRV_O_RDWR and BDRV_O_ALLOW_RDWR flags. | |
3894 | * Caution: getting a boolean member of @options requires care. | |
3895 | * When @options come from -blockdev or blockdev_add, members are | |
3896 | * typed according to the QAPI schema, but when they come from | |
3897 | * -drive, they're all QString. | |
3898 | */ | |
f87a0e29 AG |
3899 | if (g_strcmp0(qdict_get_try_str(options, BDRV_OPT_READ_ONLY), "on") && |
3900 | !qdict_get_try_bool(options, BDRV_OPT_READ_ONLY, false)) { | |
3901 | flags |= (BDRV_O_RDWR | BDRV_O_ALLOW_RDWR); | |
3902 | } else { | |
3903 | flags &= ~BDRV_O_RDWR; | |
14499ea5 AG |
3904 | } |
3905 | ||
3906 | if (flags & BDRV_O_SNAPSHOT) { | |
3907 | snapshot_options = qdict_new(); | |
3908 | bdrv_temp_snapshot_options(&snapshot_flags, snapshot_options, | |
3909 | flags, options); | |
f87a0e29 AG |
3910 | /* Let bdrv_backing_options() override "read-only" */ |
3911 | qdict_del(options, BDRV_OPT_READ_ONLY); | |
00ff7ffd HR |
3912 | bdrv_inherited_options(BDRV_CHILD_COW, true, |
3913 | &flags, options, flags, options); | |
14499ea5 AG |
3914 | } |
3915 | ||
62392ebb KW |
3916 | bs->open_flags = flags; |
3917 | bs->options = options; | |
3918 | options = qdict_clone_shallow(options); | |
3919 | ||
76c591b0 | 3920 | /* Find the right image format driver */ |
129c7d1c | 3921 | /* See cautionary note on accessing @options above */ |
76c591b0 KW |
3922 | drvname = qdict_get_try_str(options, "driver"); |
3923 | if (drvname) { | |
3924 | drv = bdrv_find_format(drvname); | |
76c591b0 KW |
3925 | if (!drv) { |
3926 | error_setg(errp, "Unknown driver: '%s'", drvname); | |
76c591b0 KW |
3927 | goto fail; |
3928 | } | |
3929 | } | |
3930 | ||
3931 | assert(drvname || !(flags & BDRV_O_PROTOCOL)); | |
76c591b0 | 3932 | |
129c7d1c | 3933 | /* See cautionary note on accessing @options above */ |
3e8c2e57 | 3934 | backing = qdict_get_try_str(options, "backing"); |
e59a0cf1 HR |
3935 | if (qobject_to(QNull, qdict_get(options, "backing")) != NULL || |
3936 | (backing && *backing == '\0')) | |
3937 | { | |
4f7be280 HR |
3938 | if (backing) { |
3939 | warn_report("Use of \"backing\": \"\" is deprecated; " | |
3940 | "use \"backing\": null instead"); | |
3941 | } | |
3e8c2e57 | 3942 | flags |= BDRV_O_NO_BACKING; |
ae0f57f0 KW |
3943 | qdict_del(bs->explicit_options, "backing"); |
3944 | qdict_del(bs->options, "backing"); | |
3e8c2e57 AG |
3945 | qdict_del(options, "backing"); |
3946 | } | |
3947 | ||
5696c6e3 | 3948 | /* Open image file without format layer. This BlockBackend is only used for |
4e4bf5c4 KW |
3949 | * probing, the block drivers will do their own bdrv_open_child() for the |
3950 | * same BDS, which is why we put the node name back into options. */ | |
f4788adc | 3951 | if ((flags & BDRV_O_PROTOCOL) == 0) { |
5696c6e3 KW |
3952 | BlockDriverState *file_bs; |
3953 | ||
3954 | file_bs = bdrv_open_child_bs(filename, options, "file", bs, | |
58944401 HR |
3955 | &child_of_bds, BDRV_CHILD_IMAGE, |
3956 | true, &local_err); | |
1fdd6933 | 3957 | if (local_err) { |
f4788adc KW |
3958 | goto fail; |
3959 | } | |
5696c6e3 | 3960 | if (file_bs != NULL) { |
dacaa162 KW |
3961 | /* Not requesting BLK_PERM_CONSISTENT_READ because we're only |
3962 | * looking at the header to guess the image format. This works even | |
3963 | * in cases where a guest would not see a consistent state. */ | |
d861ab3a | 3964 | file = blk_new(bdrv_get_aio_context(file_bs), 0, BLK_PERM_ALL); |
d7086422 | 3965 | blk_insert_bs(file, file_bs, &local_err); |
5696c6e3 | 3966 | bdrv_unref(file_bs); |
d7086422 KW |
3967 | if (local_err) { |
3968 | goto fail; | |
3969 | } | |
5696c6e3 | 3970 | |
46f5ac20 | 3971 | qdict_put_str(options, "file", bdrv_get_node_name(file_bs)); |
4e4bf5c4 | 3972 | } |
f500a6d3 KW |
3973 | } |
3974 | ||
76c591b0 | 3975 | /* Image format probing */ |
38f3ef57 | 3976 | bs->probed = !drv; |
76c591b0 | 3977 | if (!drv && file) { |
cf2ab8fc | 3978 | ret = find_image_format(file, filename, &drv, &local_err); |
17b005f1 | 3979 | if (ret < 0) { |
8bfea15d | 3980 | goto fail; |
2a05cbe4 | 3981 | } |
62392ebb KW |
3982 | /* |
3983 | * This option update would logically belong in bdrv_fill_options(), | |
3984 | * but we first need to open bs->file for the probing to work, while | |
3985 | * opening bs->file already requires the (mostly) final set of options | |
3986 | * so that cache mode etc. can be inherited. | |
3987 | * | |
3988 | * Adding the driver later is somewhat ugly, but it's not an option | |
3989 | * that would ever be inherited, so it's correct. We just need to make | |
3990 | * sure to update both bs->options (which has the full effective | |
3991 | * options for bs) and options (which has file.* already removed). | |
3992 | */ | |
46f5ac20 EB |
3993 | qdict_put_str(bs->options, "driver", drv->format_name); |
3994 | qdict_put_str(options, "driver", drv->format_name); | |
76c591b0 | 3995 | } else if (!drv) { |
17b005f1 | 3996 | error_setg(errp, "Must specify either driver or file"); |
8bfea15d | 3997 | goto fail; |
ea2384d3 | 3998 | } |
b6ce07aa | 3999 | |
53a29513 HR |
4000 | /* BDRV_O_PROTOCOL must be set iff a protocol BDS is about to be created */ |
4001 | assert(!!(flags & BDRV_O_PROTOCOL) == !!drv->bdrv_file_open); | |
4002 | /* file must be NULL if a protocol BDS is about to be created | |
4003 | * (the inverse results in an error message from bdrv_open_common()) */ | |
4004 | assert(!(flags & BDRV_O_PROTOCOL) || !file); | |
4005 | ||
b6ce07aa | 4006 | /* Open the image */ |
82dc8b41 | 4007 | ret = bdrv_open_common(bs, file, options, &local_err); |
b6ce07aa | 4008 | if (ret < 0) { |
8bfea15d | 4009 | goto fail; |
6987307c CH |
4010 | } |
4011 | ||
4e4bf5c4 | 4012 | if (file) { |
5696c6e3 | 4013 | blk_unref(file); |
f500a6d3 KW |
4014 | file = NULL; |
4015 | } | |
4016 | ||
b6ce07aa | 4017 | /* If there is a backing file, use it */ |
9156df12 | 4018 | if ((flags & BDRV_O_NO_BACKING) == 0) { |
d9b7b057 | 4019 | ret = bdrv_open_backing_file(bs, options, "backing", &local_err); |
b6ce07aa | 4020 | if (ret < 0) { |
b6ad491a | 4021 | goto close_and_fail; |
b6ce07aa | 4022 | } |
b6ce07aa KW |
4023 | } |
4024 | ||
50196d7a AG |
4025 | /* Remove all children options and references |
4026 | * from bs->options and bs->explicit_options */ | |
2f624b80 AG |
4027 | QLIST_FOREACH(child, &bs->children, next) { |
4028 | char *child_key_dot; | |
4029 | child_key_dot = g_strdup_printf("%s.", child->name); | |
4030 | qdict_extract_subqdict(bs->explicit_options, NULL, child_key_dot); | |
4031 | qdict_extract_subqdict(bs->options, NULL, child_key_dot); | |
50196d7a AG |
4032 | qdict_del(bs->explicit_options, child->name); |
4033 | qdict_del(bs->options, child->name); | |
2f624b80 AG |
4034 | g_free(child_key_dot); |
4035 | } | |
4036 | ||
b6ad491a | 4037 | /* Check if any unknown options were used */ |
7ad2757f | 4038 | if (qdict_size(options) != 0) { |
b6ad491a | 4039 | const QDictEntry *entry = qdict_first(options); |
5acd9d81 HR |
4040 | if (flags & BDRV_O_PROTOCOL) { |
4041 | error_setg(errp, "Block protocol '%s' doesn't support the option " | |
4042 | "'%s'", drv->format_name, entry->key); | |
4043 | } else { | |
d0e46a55 HR |
4044 | error_setg(errp, |
4045 | "Block format '%s' does not support the option '%s'", | |
4046 | drv->format_name, entry->key); | |
5acd9d81 | 4047 | } |
b6ad491a | 4048 | |
b6ad491a KW |
4049 | goto close_and_fail; |
4050 | } | |
b6ad491a | 4051 | |
c01c214b | 4052 | bdrv_parent_cb_change_media(bs, true); |
b6ce07aa | 4053 | |
cb3e7f08 | 4054 | qobject_unref(options); |
8961be33 | 4055 | options = NULL; |
dd62f1ca KW |
4056 | |
4057 | /* For snapshot=on, create a temporary qcow2 overlay. bs points to the | |
4058 | * temporary snapshot afterwards. */ | |
4059 | if (snapshot_flags) { | |
66836189 HR |
4060 | BlockDriverState *snapshot_bs; |
4061 | snapshot_bs = bdrv_append_temp_snapshot(bs, snapshot_flags, | |
4062 | snapshot_options, &local_err); | |
73176bee | 4063 | snapshot_options = NULL; |
dd62f1ca KW |
4064 | if (local_err) { |
4065 | goto close_and_fail; | |
4066 | } | |
5b363937 HR |
4067 | /* We are not going to return bs but the overlay on top of it |
4068 | * (snapshot_bs); thus, we have to drop the strong reference to bs | |
4069 | * (which we obtained by calling bdrv_new()). bs will not be deleted, | |
4070 | * though, because the overlay still has a reference to it. */ | |
4071 | bdrv_unref(bs); | |
4072 | bs = snapshot_bs; | |
dd62f1ca KW |
4073 | } |
4074 | ||
5b363937 | 4075 | return bs; |
b6ce07aa | 4076 | |
8bfea15d | 4077 | fail: |
5696c6e3 | 4078 | blk_unref(file); |
cb3e7f08 MAL |
4079 | qobject_unref(snapshot_options); |
4080 | qobject_unref(bs->explicit_options); | |
4081 | qobject_unref(bs->options); | |
4082 | qobject_unref(options); | |
de9c0cec | 4083 | bs->options = NULL; |
998cbd6a | 4084 | bs->explicit_options = NULL; |
5b363937 | 4085 | bdrv_unref(bs); |
621ff94d | 4086 | error_propagate(errp, local_err); |
5b363937 | 4087 | return NULL; |
de9c0cec | 4088 | |
b6ad491a | 4089 | close_and_fail: |
5b363937 | 4090 | bdrv_unref(bs); |
cb3e7f08 MAL |
4091 | qobject_unref(snapshot_options); |
4092 | qobject_unref(options); | |
621ff94d | 4093 | error_propagate(errp, local_err); |
5b363937 | 4094 | return NULL; |
b6ce07aa KW |
4095 | } |
4096 | ||
c86422c5 EGE |
4097 | /* |
4098 | * The caller must always hold @filename AioContext lock, because this | |
4099 | * function eventually calls bdrv_refresh_total_sectors() which polls | |
4100 | * when called from non-coroutine context. | |
4101 | */ | |
5b363937 HR |
4102 | BlockDriverState *bdrv_open(const char *filename, const char *reference, |
4103 | QDict *options, int flags, Error **errp) | |
f3930ed0 | 4104 | { |
f791bf7f EGE |
4105 | GLOBAL_STATE_CODE(); |
4106 | ||
5b363937 | 4107 | return bdrv_open_inherit(filename, reference, options, flags, NULL, |
272c02ea | 4108 | NULL, 0, errp); |
f3930ed0 KW |
4109 | } |
4110 | ||
faf116b4 AG |
4111 | /* Return true if the NULL-terminated @list contains @str */ |
4112 | static bool is_str_in_list(const char *str, const char *const *list) | |
4113 | { | |
4114 | if (str && list) { | |
4115 | int i; | |
4116 | for (i = 0; list[i] != NULL; i++) { | |
4117 | if (!strcmp(str, list[i])) { | |
4118 | return true; | |
4119 | } | |
4120 | } | |
4121 | } | |
4122 | return false; | |
4123 | } | |
4124 | ||
4125 | /* | |
4126 | * Check that every option set in @bs->options is also set in | |
4127 | * @new_opts. | |
4128 | * | |
4129 | * Options listed in the common_options list and in | |
4130 | * @bs->drv->mutable_opts are skipped. | |
4131 | * | |
4132 | * Return 0 on success, otherwise return -EINVAL and set @errp. | |
4133 | */ | |
4134 | static int bdrv_reset_options_allowed(BlockDriverState *bs, | |
4135 | const QDict *new_opts, Error **errp) | |
4136 | { | |
4137 | const QDictEntry *e; | |
4138 | /* These options are common to all block drivers and are handled | |
4139 | * in bdrv_reopen_prepare() so they can be left out of @new_opts */ | |
4140 | const char *const common_options[] = { | |
4141 | "node-name", "discard", "cache.direct", "cache.no-flush", | |
4142 | "read-only", "auto-read-only", "detect-zeroes", NULL | |
4143 | }; | |
4144 | ||
4145 | for (e = qdict_first(bs->options); e; e = qdict_next(bs->options, e)) { | |
4146 | if (!qdict_haskey(new_opts, e->key) && | |
4147 | !is_str_in_list(e->key, common_options) && | |
4148 | !is_str_in_list(e->key, bs->drv->mutable_opts)) { | |
4149 | error_setg(errp, "Option '%s' cannot be reset " | |
4150 | "to its default value", e->key); | |
4151 | return -EINVAL; | |
4152 | } | |
4153 | } | |
4154 | ||
4155 | return 0; | |
4156 | } | |
4157 | ||
cb828c31 AG |
4158 | /* |
4159 | * Returns true if @child can be reached recursively from @bs | |
4160 | */ | |
4161 | static bool bdrv_recurse_has_child(BlockDriverState *bs, | |
4162 | BlockDriverState *child) | |
4163 | { | |
4164 | BdrvChild *c; | |
4165 | ||
4166 | if (bs == child) { | |
4167 | return true; | |
4168 | } | |
4169 | ||
4170 | QLIST_FOREACH(c, &bs->children, next) { | |
4171 | if (bdrv_recurse_has_child(c->bs, child)) { | |
4172 | return true; | |
4173 | } | |
4174 | } | |
4175 | ||
4176 | return false; | |
4177 | } | |
4178 | ||
e971aa12 JC |
4179 | /* |
4180 | * Adds a BlockDriverState to a simple queue for an atomic, transactional | |
4181 | * reopen of multiple devices. | |
4182 | * | |
859443b0 | 4183 | * bs_queue can either be an existing BlockReopenQueue that has had QTAILQ_INIT |
e971aa12 JC |
4184 | * already performed, or alternatively may be NULL a new BlockReopenQueue will |
4185 | * be created and initialized. This newly created BlockReopenQueue should be | |
4186 | * passed back in for subsequent calls that are intended to be of the same | |
4187 | * atomic 'set'. | |
4188 | * | |
4189 | * bs is the BlockDriverState to add to the reopen queue. | |
4190 | * | |
4d2cb092 KW |
4191 | * options contains the changed options for the associated bs |
4192 | * (the BlockReopenQueue takes ownership) | |
4193 | * | |
e971aa12 JC |
4194 | * flags contains the open flags for the associated bs |
4195 | * | |
4196 | * returns a pointer to bs_queue, which is either the newly allocated | |
4197 | * bs_queue, or the existing bs_queue being used. | |
4198 | * | |
d22933ac | 4199 | * bs is drained here and undrained by bdrv_reopen_queue_free(). |
2e117866 KW |
4200 | * |
4201 | * To be called with bs->aio_context locked. | |
e971aa12 | 4202 | */ |
28518102 KW |
4203 | static BlockReopenQueue *bdrv_reopen_queue_child(BlockReopenQueue *bs_queue, |
4204 | BlockDriverState *bs, | |
4205 | QDict *options, | |
bd86fb99 | 4206 | const BdrvChildClass *klass, |
272c02ea | 4207 | BdrvChildRole role, |
3cdc69d3 | 4208 | bool parent_is_format, |
28518102 | 4209 | QDict *parent_options, |
077e8e20 AG |
4210 | int parent_flags, |
4211 | bool keep_old_opts) | |
e971aa12 JC |
4212 | { |
4213 | assert(bs != NULL); | |
4214 | ||
4215 | BlockReopenQueueEntry *bs_entry; | |
67251a31 | 4216 | BdrvChild *child; |
9aa09ddd AG |
4217 | QDict *old_options, *explicit_options, *options_copy; |
4218 | int flags; | |
4219 | QemuOpts *opts; | |
67251a31 | 4220 | |
f0c28327 | 4221 | GLOBAL_STATE_CODE(); |
1a63a907 | 4222 | |
d22933ac KW |
4223 | bdrv_drained_begin(bs); |
4224 | ||
e971aa12 JC |
4225 | if (bs_queue == NULL) { |
4226 | bs_queue = g_new0(BlockReopenQueue, 1); | |
859443b0 | 4227 | QTAILQ_INIT(bs_queue); |
e971aa12 JC |
4228 | } |
4229 | ||
4d2cb092 KW |
4230 | if (!options) { |
4231 | options = qdict_new(); | |
4232 | } | |
4233 | ||
5b7ba05f | 4234 | /* Check if this BlockDriverState is already in the queue */ |
859443b0 | 4235 | QTAILQ_FOREACH(bs_entry, bs_queue, entry) { |
5b7ba05f AG |
4236 | if (bs == bs_entry->state.bs) { |
4237 | break; | |
4238 | } | |
4239 | } | |
4240 | ||
28518102 KW |
4241 | /* |
4242 | * Precedence of options: | |
4243 | * 1. Explicitly passed in options (highest) | |
9aa09ddd AG |
4244 | * 2. Retained from explicitly set options of bs |
4245 | * 3. Inherited from parent node | |
4246 | * 4. Retained from effective options of bs | |
28518102 KW |
4247 | */ |
4248 | ||
145f598e | 4249 | /* Old explicitly set values (don't overwrite by inherited value) */ |
077e8e20 AG |
4250 | if (bs_entry || keep_old_opts) { |
4251 | old_options = qdict_clone_shallow(bs_entry ? | |
4252 | bs_entry->state.explicit_options : | |
4253 | bs->explicit_options); | |
4254 | bdrv_join_options(bs, options, old_options); | |
4255 | qobject_unref(old_options); | |
5b7ba05f | 4256 | } |
145f598e KW |
4257 | |
4258 | explicit_options = qdict_clone_shallow(options); | |
4259 | ||
28518102 KW |
4260 | /* Inherit from parent node */ |
4261 | if (parent_options) { | |
9aa09ddd | 4262 | flags = 0; |
3cdc69d3 | 4263 | klass->inherit_options(role, parent_is_format, &flags, options, |
272c02ea | 4264 | parent_flags, parent_options); |
9aa09ddd AG |
4265 | } else { |
4266 | flags = bdrv_get_flags(bs); | |
28518102 KW |
4267 | } |
4268 | ||
077e8e20 AG |
4269 | if (keep_old_opts) { |
4270 | /* Old values are used for options that aren't set yet */ | |
4271 | old_options = qdict_clone_shallow(bs->options); | |
4272 | bdrv_join_options(bs, options, old_options); | |
4273 | qobject_unref(old_options); | |
4274 | } | |
4d2cb092 | 4275 | |
9aa09ddd AG |
4276 | /* We have the final set of options so let's update the flags */ |
4277 | options_copy = qdict_clone_shallow(options); | |
4278 | opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort); | |
4279 | qemu_opts_absorb_qdict(opts, options_copy, NULL); | |
4280 | update_flags_from_options(&flags, opts); | |
4281 | qemu_opts_del(opts); | |
4282 | qobject_unref(options_copy); | |
4283 | ||
fd452021 | 4284 | /* bdrv_open_inherit() sets and clears some additional flags internally */ |
f1f25a2e | 4285 | flags &= ~BDRV_O_PROTOCOL; |
fd452021 KW |
4286 | if (flags & BDRV_O_RDWR) { |
4287 | flags |= BDRV_O_ALLOW_RDWR; | |
4288 | } | |
f1f25a2e | 4289 | |
1857c97b KW |
4290 | if (!bs_entry) { |
4291 | bs_entry = g_new0(BlockReopenQueueEntry, 1); | |
859443b0 | 4292 | QTAILQ_INSERT_TAIL(bs_queue, bs_entry, entry); |
1857c97b | 4293 | } else { |
cb3e7f08 MAL |
4294 | qobject_unref(bs_entry->state.options); |
4295 | qobject_unref(bs_entry->state.explicit_options); | |
1857c97b KW |
4296 | } |
4297 | ||
4298 | bs_entry->state.bs = bs; | |
4299 | bs_entry->state.options = options; | |
4300 | bs_entry->state.explicit_options = explicit_options; | |
4301 | bs_entry->state.flags = flags; | |
4302 | ||
8546632e AG |
4303 | /* |
4304 | * If keep_old_opts is false then it means that unspecified | |
4305 | * options must be reset to their original value. We don't allow | |
4306 | * resetting 'backing' but we need to know if the option is | |
4307 | * missing in order to decide if we have to return an error. | |
4308 | */ | |
4309 | if (!keep_old_opts) { | |
4310 | bs_entry->state.backing_missing = | |
4311 | !qdict_haskey(options, "backing") && | |
4312 | !qdict_haskey(options, "backing.driver"); | |
4313 | } | |
4314 | ||
67251a31 | 4315 | QLIST_FOREACH(child, &bs->children, next) { |
8546632e AG |
4316 | QDict *new_child_options = NULL; |
4317 | bool child_keep_old = keep_old_opts; | |
67251a31 | 4318 | |
4c9dfe5d KW |
4319 | /* reopen can only change the options of block devices that were |
4320 | * implicitly created and inherited options. For other (referenced) | |
4321 | * block devices, a syntax like "backing.foo" results in an error. */ | |
67251a31 KW |
4322 | if (child->bs->inherits_from != bs) { |
4323 | continue; | |
4324 | } | |
4325 | ||
8546632e AG |
4326 | /* Check if the options contain a child reference */ |
4327 | if (qdict_haskey(options, child->name)) { | |
4328 | const char *childref = qdict_get_try_str(options, child->name); | |
4329 | /* | |
4330 | * The current child must not be reopened if the child | |
4331 | * reference is null or points to a different node. | |
4332 | */ | |
4333 | if (g_strcmp0(childref, child->bs->node_name)) { | |
4334 | continue; | |
4335 | } | |
4336 | /* | |
4337 | * If the child reference points to the current child then | |
4338 | * reopen it with its existing set of options (note that | |
4339 | * it can still inherit new options from the parent). | |
4340 | */ | |
4341 | child_keep_old = true; | |
4342 | } else { | |
4343 | /* Extract child options ("child-name.*") */ | |
4344 | char *child_key_dot = g_strdup_printf("%s.", child->name); | |
4345 | qdict_extract_subqdict(explicit_options, NULL, child_key_dot); | |
4346 | qdict_extract_subqdict(options, &new_child_options, child_key_dot); | |
4347 | g_free(child_key_dot); | |
4348 | } | |
4c9dfe5d | 4349 | |
9aa09ddd | 4350 | bdrv_reopen_queue_child(bs_queue, child->bs, new_child_options, |
3cdc69d3 HR |
4351 | child->klass, child->role, bs->drv->is_format, |
4352 | options, flags, child_keep_old); | |
e971aa12 JC |
4353 | } |
4354 | ||
e971aa12 JC |
4355 | return bs_queue; |
4356 | } | |
4357 | ||
2e117866 | 4358 | /* To be called with bs->aio_context locked */ |
28518102 KW |
4359 | BlockReopenQueue *bdrv_reopen_queue(BlockReopenQueue *bs_queue, |
4360 | BlockDriverState *bs, | |
077e8e20 | 4361 | QDict *options, bool keep_old_opts) |
28518102 | 4362 | { |
f791bf7f EGE |
4363 | GLOBAL_STATE_CODE(); |
4364 | ||
3cdc69d3 HR |
4365 | return bdrv_reopen_queue_child(bs_queue, bs, options, NULL, 0, false, |
4366 | NULL, 0, keep_old_opts); | |
28518102 KW |
4367 | } |
4368 | ||
ab5b5228 AG |
4369 | void bdrv_reopen_queue_free(BlockReopenQueue *bs_queue) |
4370 | { | |
f791bf7f | 4371 | GLOBAL_STATE_CODE(); |
ab5b5228 AG |
4372 | if (bs_queue) { |
4373 | BlockReopenQueueEntry *bs_entry, *next; | |
4374 | QTAILQ_FOREACH_SAFE(bs_entry, bs_queue, entry, next) { | |
d22933ac KW |
4375 | AioContext *ctx = bdrv_get_aio_context(bs_entry->state.bs); |
4376 | ||
4377 | aio_context_acquire(ctx); | |
4378 | bdrv_drained_end(bs_entry->state.bs); | |
4379 | aio_context_release(ctx); | |
4380 | ||
ab5b5228 AG |
4381 | qobject_unref(bs_entry->state.explicit_options); |
4382 | qobject_unref(bs_entry->state.options); | |
4383 | g_free(bs_entry); | |
4384 | } | |
4385 | g_free(bs_queue); | |
4386 | } | |
4387 | } | |
4388 | ||
e971aa12 JC |
4389 | /* |
4390 | * Reopen multiple BlockDriverStates atomically & transactionally. | |
4391 | * | |
4392 | * The queue passed in (bs_queue) must have been built up previous | |
4393 | * via bdrv_reopen_queue(). | |
4394 | * | |
4395 | * Reopens all BDS specified in the queue, with the appropriate | |
4396 | * flags. All devices are prepared for reopen, and failure of any | |
50d6a8a3 | 4397 | * device will cause all device changes to be abandoned, and intermediate |
e971aa12 JC |
4398 | * data cleaned up. |
4399 | * | |
4400 | * If all devices prepare successfully, then the changes are committed | |
4401 | * to all devices. | |
4402 | * | |
1a63a907 KW |
4403 | * All affected nodes must be drained between bdrv_reopen_queue() and |
4404 | * bdrv_reopen_multiple(). | |
6cf42ca2 KW |
4405 | * |
4406 | * To be called from the main thread, with all other AioContexts unlocked. | |
e971aa12 | 4407 | */ |
5019aece | 4408 | int bdrv_reopen_multiple(BlockReopenQueue *bs_queue, Error **errp) |
e971aa12 JC |
4409 | { |
4410 | int ret = -1; | |
4411 | BlockReopenQueueEntry *bs_entry, *next; | |
6cf42ca2 | 4412 | AioContext *ctx; |
72373e40 | 4413 | Transaction *tran = tran_new(); |
72373e40 | 4414 | g_autoptr(GSList) refresh_list = NULL; |
e971aa12 | 4415 | |
6cf42ca2 | 4416 | assert(qemu_get_current_aio_context() == qemu_get_aio_context()); |
e971aa12 | 4417 | assert(bs_queue != NULL); |
da359909 | 4418 | GLOBAL_STATE_CODE(); |
e971aa12 | 4419 | |
a2aabf88 | 4420 | QTAILQ_FOREACH(bs_entry, bs_queue, entry) { |
6cf42ca2 KW |
4421 | ctx = bdrv_get_aio_context(bs_entry->state.bs); |
4422 | aio_context_acquire(ctx); | |
a2aabf88 | 4423 | ret = bdrv_flush(bs_entry->state.bs); |
6cf42ca2 | 4424 | aio_context_release(ctx); |
a2aabf88 VSO |
4425 | if (ret < 0) { |
4426 | error_setg_errno(errp, -ret, "Error flushing drive"); | |
e3fc91aa | 4427 | goto abort; |
a2aabf88 VSO |
4428 | } |
4429 | } | |
4430 | ||
859443b0 | 4431 | QTAILQ_FOREACH(bs_entry, bs_queue, entry) { |
1a63a907 | 4432 | assert(bs_entry->state.bs->quiesce_counter > 0); |
6cf42ca2 KW |
4433 | ctx = bdrv_get_aio_context(bs_entry->state.bs); |
4434 | aio_context_acquire(ctx); | |
72373e40 | 4435 | ret = bdrv_reopen_prepare(&bs_entry->state, bs_queue, tran, errp); |
6cf42ca2 | 4436 | aio_context_release(ctx); |
72373e40 VSO |
4437 | if (ret < 0) { |
4438 | goto abort; | |
e971aa12 JC |
4439 | } |
4440 | bs_entry->prepared = true; | |
4441 | } | |
4442 | ||
859443b0 | 4443 | QTAILQ_FOREACH(bs_entry, bs_queue, entry) { |
69b736e7 | 4444 | BDRVReopenState *state = &bs_entry->state; |
72373e40 | 4445 | |
fb0ff4d1 | 4446 | refresh_list = g_slist_prepend(refresh_list, state->bs); |
72373e40 | 4447 | if (state->old_backing_bs) { |
fb0ff4d1 | 4448 | refresh_list = g_slist_prepend(refresh_list, state->old_backing_bs); |
cb828c31 | 4449 | } |
ecd30d2d | 4450 | if (state->old_file_bs) { |
fb0ff4d1 | 4451 | refresh_list = g_slist_prepend(refresh_list, state->old_file_bs); |
ecd30d2d | 4452 | } |
72373e40 VSO |
4453 | } |
4454 | ||
4455 | /* | |
4456 | * Note that file-posix driver rely on permission update done during reopen | |
4457 | * (even if no permission changed), because it wants "new" permissions for | |
4458 | * reconfiguring the fd and that's why it does it in raw_check_perm(), not | |
4459 | * in raw_reopen_prepare() which is called with "old" permissions. | |
4460 | */ | |
4461 | ret = bdrv_list_refresh_perms(refresh_list, bs_queue, tran, errp); | |
4462 | if (ret < 0) { | |
4463 | goto abort; | |
69b736e7 KW |
4464 | } |
4465 | ||
fcd6a4f4 VSO |
4466 | /* |
4467 | * If we reach this point, we have success and just need to apply the | |
4468 | * changes. | |
4469 | * | |
4470 | * Reverse order is used to comfort qcow2 driver: on commit it need to write | |
4471 | * IN_USE flag to the image, to mark bitmaps in the image as invalid. But | |
4472 | * children are usually goes after parents in reopen-queue, so go from last | |
4473 | * to first element. | |
e971aa12 | 4474 | */ |
fcd6a4f4 | 4475 | QTAILQ_FOREACH_REVERSE(bs_entry, bs_queue, entry) { |
6cf42ca2 KW |
4476 | ctx = bdrv_get_aio_context(bs_entry->state.bs); |
4477 | aio_context_acquire(ctx); | |
e971aa12 | 4478 | bdrv_reopen_commit(&bs_entry->state); |
6cf42ca2 | 4479 | aio_context_release(ctx); |
e971aa12 JC |
4480 | } |
4481 | ||
72373e40 | 4482 | tran_commit(tran); |
69b736e7 | 4483 | |
72373e40 VSO |
4484 | QTAILQ_FOREACH_REVERSE(bs_entry, bs_queue, entry) { |
4485 | BlockDriverState *bs = bs_entry->state.bs; | |
74ad9a3b | 4486 | |
72373e40 | 4487 | if (bs->drv->bdrv_reopen_commit_post) { |
6cf42ca2 KW |
4488 | ctx = bdrv_get_aio_context(bs); |
4489 | aio_context_acquire(ctx); | |
72373e40 | 4490 | bs->drv->bdrv_reopen_commit_post(&bs_entry->state); |
6cf42ca2 | 4491 | aio_context_release(ctx); |
69b736e7 KW |
4492 | } |
4493 | } | |
17e1e2be | 4494 | |
72373e40 VSO |
4495 | ret = 0; |
4496 | goto cleanup; | |
17e1e2be | 4497 | |
72373e40 VSO |
4498 | abort: |
4499 | tran_abort(tran); | |
4500 | QTAILQ_FOREACH_SAFE(bs_entry, bs_queue, entry, next) { | |
4501 | if (bs_entry->prepared) { | |
6cf42ca2 KW |
4502 | ctx = bdrv_get_aio_context(bs_entry->state.bs); |
4503 | aio_context_acquire(ctx); | |
72373e40 | 4504 | bdrv_reopen_abort(&bs_entry->state); |
6cf42ca2 | 4505 | aio_context_release(ctx); |
17e1e2be PK |
4506 | } |
4507 | } | |
72373e40 | 4508 | |
e971aa12 | 4509 | cleanup: |
ab5b5228 | 4510 | bdrv_reopen_queue_free(bs_queue); |
40840e41 | 4511 | |
e971aa12 JC |
4512 | return ret; |
4513 | } | |
4514 | ||
6cf42ca2 KW |
4515 | int bdrv_reopen(BlockDriverState *bs, QDict *opts, bool keep_old_opts, |
4516 | Error **errp) | |
6e1000a8 | 4517 | { |
6cf42ca2 | 4518 | AioContext *ctx = bdrv_get_aio_context(bs); |
6e1000a8 | 4519 | BlockReopenQueue *queue; |
6cf42ca2 | 4520 | int ret; |
6e1000a8 | 4521 | |
f791bf7f EGE |
4522 | GLOBAL_STATE_CODE(); |
4523 | ||
2e117866 KW |
4524 | queue = bdrv_reopen_queue(NULL, bs, opts, keep_old_opts); |
4525 | ||
6cf42ca2 KW |
4526 | if (ctx != qemu_get_aio_context()) { |
4527 | aio_context_release(ctx); | |
4528 | } | |
5019aece | 4529 | ret = bdrv_reopen_multiple(queue, errp); |
6cf42ca2 KW |
4530 | |
4531 | if (ctx != qemu_get_aio_context()) { | |
4532 | aio_context_acquire(ctx); | |
4533 | } | |
6e1000a8 AG |
4534 | |
4535 | return ret; | |
4536 | } | |
4537 | ||
6cf42ca2 KW |
4538 | int bdrv_reopen_set_read_only(BlockDriverState *bs, bool read_only, |
4539 | Error **errp) | |
4540 | { | |
4541 | QDict *opts = qdict_new(); | |
4542 | ||
f791bf7f EGE |
4543 | GLOBAL_STATE_CODE(); |
4544 | ||
6cf42ca2 KW |
4545 | qdict_put_bool(opts, BDRV_OPT_READ_ONLY, read_only); |
4546 | ||
4547 | return bdrv_reopen(bs, opts, true, errp); | |
4548 | } | |
4549 | ||
cb828c31 AG |
4550 | /* |
4551 | * Take a BDRVReopenState and check if the value of 'backing' in the | |
4552 | * reopen_state->options QDict is valid or not. | |
4553 | * | |
4554 | * If 'backing' is missing from the QDict then return 0. | |
4555 | * | |
4556 | * If 'backing' contains the node name of the backing file of | |
4557 | * reopen_state->bs then return 0. | |
4558 | * | |
4559 | * If 'backing' contains a different node name (or is null) then check | |
4560 | * whether the current backing file can be replaced with the new one. | |
4561 | * If that's the case then reopen_state->replace_backing_bs is set to | |
4562 | * true and reopen_state->new_backing_bs contains a pointer to the new | |
4563 | * backing BlockDriverState (or NULL). | |
4564 | * | |
4565 | * Return 0 on success, otherwise return < 0 and set @errp. | |
4566 | */ | |
ecd30d2d AG |
4567 | static int bdrv_reopen_parse_file_or_backing(BDRVReopenState *reopen_state, |
4568 | bool is_backing, Transaction *tran, | |
4569 | Error **errp) | |
cb828c31 AG |
4570 | { |
4571 | BlockDriverState *bs = reopen_state->bs; | |
ecd30d2d AG |
4572 | BlockDriverState *new_child_bs; |
4573 | BlockDriverState *old_child_bs = is_backing ? child_bs(bs->backing) : | |
4574 | child_bs(bs->file); | |
4575 | const char *child_name = is_backing ? "backing" : "file"; | |
cb828c31 AG |
4576 | QObject *value; |
4577 | const char *str; | |
4578 | ||
bdb73476 EGE |
4579 | GLOBAL_STATE_CODE(); |
4580 | ||
ecd30d2d | 4581 | value = qdict_get(reopen_state->options, child_name); |
cb828c31 AG |
4582 | if (value == NULL) { |
4583 | return 0; | |
4584 | } | |
4585 | ||
4586 | switch (qobject_type(value)) { | |
4587 | case QTYPE_QNULL: | |
ecd30d2d AG |
4588 | assert(is_backing); /* The 'file' option does not allow a null value */ |
4589 | new_child_bs = NULL; | |
cb828c31 AG |
4590 | break; |
4591 | case QTYPE_QSTRING: | |
410f44f5 | 4592 | str = qstring_get_str(qobject_to(QString, value)); |
ecd30d2d AG |
4593 | new_child_bs = bdrv_lookup_bs(NULL, str, errp); |
4594 | if (new_child_bs == NULL) { | |
cb828c31 | 4595 | return -EINVAL; |
ecd30d2d AG |
4596 | } else if (bdrv_recurse_has_child(new_child_bs, bs)) { |
4597 | error_setg(errp, "Making '%s' a %s child of '%s' would create a " | |
4598 | "cycle", str, child_name, bs->node_name); | |
cb828c31 AG |
4599 | return -EINVAL; |
4600 | } | |
4601 | break; | |
4602 | default: | |
ecd30d2d AG |
4603 | /* |
4604 | * The options QDict has been flattened, so 'backing' and 'file' | |
4605 | * do not allow any other data type here. | |
4606 | */ | |
cb828c31 AG |
4607 | g_assert_not_reached(); |
4608 | } | |
4609 | ||
ecd30d2d AG |
4610 | if (old_child_bs == new_child_bs) { |
4611 | return 0; | |
4612 | } | |
4613 | ||
4614 | if (old_child_bs) { | |
4615 | if (bdrv_skip_implicit_filters(old_child_bs) == new_child_bs) { | |
cbfdb98c VSO |
4616 | return 0; |
4617 | } | |
4618 | ||
ecd30d2d AG |
4619 | if (old_child_bs->implicit) { |
4620 | error_setg(errp, "Cannot replace implicit %s child of %s", | |
4621 | child_name, bs->node_name); | |
cbfdb98c VSO |
4622 | return -EPERM; |
4623 | } | |
4624 | } | |
4625 | ||
ecd30d2d | 4626 | if (bs->drv->is_filter && !old_child_bs) { |
25f78d9e VSO |
4627 | /* |
4628 | * Filters always have a file or a backing child, so we are trying to | |
4629 | * change wrong child | |
4630 | */ | |
4631 | error_setg(errp, "'%s' is a %s filter node that does not support a " | |
ecd30d2d | 4632 | "%s child", bs->node_name, bs->drv->format_name, child_name); |
1d42f48c HR |
4633 | return -EINVAL; |
4634 | } | |
4635 | ||
ecd30d2d AG |
4636 | if (is_backing) { |
4637 | reopen_state->old_backing_bs = old_child_bs; | |
4638 | } else { | |
4639 | reopen_state->old_file_bs = old_child_bs; | |
4640 | } | |
4641 | ||
4642 | return bdrv_set_file_or_backing_noperm(bs, new_child_bs, is_backing, | |
4643 | tran, errp); | |
cb828c31 AG |
4644 | } |
4645 | ||
e971aa12 JC |
4646 | /* |
4647 | * Prepares a BlockDriverState for reopen. All changes are staged in the | |
4648 | * 'opaque' field of the BDRVReopenState, which is used and allocated by | |
4649 | * the block driver layer .bdrv_reopen_prepare() | |
4650 | * | |
4651 | * bs is the BlockDriverState to reopen | |
4652 | * flags are the new open flags | |
4653 | * queue is the reopen queue | |
4654 | * | |
4655 | * Returns 0 on success, non-zero on error. On error errp will be set | |
4656 | * as well. | |
4657 | * | |
4658 | * On failure, bdrv_reopen_abort() will be called to clean up any data. | |
4659 | * It is the responsibility of the caller to then call the abort() or | |
4660 | * commit() for any other BDS that have been left in a prepare() state | |
4661 | * | |
4662 | */ | |
53e96d1e | 4663 | static int bdrv_reopen_prepare(BDRVReopenState *reopen_state, |
72373e40 | 4664 | BlockReopenQueue *queue, |
ecd30d2d | 4665 | Transaction *change_child_tran, Error **errp) |
e971aa12 JC |
4666 | { |
4667 | int ret = -1; | |
e6d79c41 | 4668 | int old_flags; |
e971aa12 JC |
4669 | Error *local_err = NULL; |
4670 | BlockDriver *drv; | |
ccf9dc07 | 4671 | QemuOpts *opts; |
4c8350fe | 4672 | QDict *orig_reopen_opts; |
593b3071 | 4673 | char *discard = NULL; |
3d8ce171 | 4674 | bool read_only; |
9ad08c44 | 4675 | bool drv_prepared = false; |
e971aa12 JC |
4676 | |
4677 | assert(reopen_state != NULL); | |
4678 | assert(reopen_state->bs->drv != NULL); | |
da359909 | 4679 | GLOBAL_STATE_CODE(); |
e971aa12 JC |
4680 | drv = reopen_state->bs->drv; |
4681 | ||
4c8350fe AG |
4682 | /* This function and each driver's bdrv_reopen_prepare() remove |
4683 | * entries from reopen_state->options as they are processed, so | |
4684 | * we need to make a copy of the original QDict. */ | |
4685 | orig_reopen_opts = qdict_clone_shallow(reopen_state->options); | |
4686 | ||
ccf9dc07 KW |
4687 | /* Process generic block layer options */ |
4688 | opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort); | |
af175e85 | 4689 | if (!qemu_opts_absorb_qdict(opts, reopen_state->options, errp)) { |
ccf9dc07 KW |
4690 | ret = -EINVAL; |
4691 | goto error; | |
4692 | } | |
4693 | ||
e6d79c41 AG |
4694 | /* This was already called in bdrv_reopen_queue_child() so the flags |
4695 | * are up-to-date. This time we simply want to remove the options from | |
4696 | * QemuOpts in order to indicate that they have been processed. */ | |
4697 | old_flags = reopen_state->flags; | |
91a097e7 | 4698 | update_flags_from_options(&reopen_state->flags, opts); |
e6d79c41 | 4699 | assert(old_flags == reopen_state->flags); |
91a097e7 | 4700 | |
415bbca8 | 4701 | discard = qemu_opt_get_del(opts, BDRV_OPT_DISCARD); |
593b3071 AG |
4702 | if (discard != NULL) { |
4703 | if (bdrv_parse_discard_flags(discard, &reopen_state->flags) != 0) { | |
4704 | error_setg(errp, "Invalid discard option"); | |
4705 | ret = -EINVAL; | |
4706 | goto error; | |
4707 | } | |
4708 | } | |
4709 | ||
543770bd AG |
4710 | reopen_state->detect_zeroes = |
4711 | bdrv_parse_detect_zeroes(opts, reopen_state->flags, &local_err); | |
4712 | if (local_err) { | |
4713 | error_propagate(errp, local_err); | |
4714 | ret = -EINVAL; | |
4715 | goto error; | |
4716 | } | |
4717 | ||
57f9db9a AG |
4718 | /* All other options (including node-name and driver) must be unchanged. |
4719 | * Put them back into the QDict, so that they are checked at the end | |
4720 | * of this function. */ | |
4721 | qemu_opts_to_qdict(opts, reopen_state->options); | |
ccf9dc07 | 4722 | |
3d8ce171 JC |
4723 | /* If we are to stay read-only, do not allow permission change |
4724 | * to r/w. Attempting to set to r/w may fail if either BDRV_O_ALLOW_RDWR is | |
4725 | * not set, or if the BDS still has copy_on_read enabled */ | |
4726 | read_only = !(reopen_state->flags & BDRV_O_RDWR); | |
54a32bfe | 4727 | ret = bdrv_can_set_read_only(reopen_state->bs, read_only, true, &local_err); |
3d8ce171 JC |
4728 | if (local_err) { |
4729 | error_propagate(errp, local_err); | |
e971aa12 JC |
4730 | goto error; |
4731 | } | |
4732 | ||
e971aa12 | 4733 | if (drv->bdrv_reopen_prepare) { |
faf116b4 AG |
4734 | /* |
4735 | * If a driver-specific option is missing, it means that we | |
4736 | * should reset it to its default value. | |
4737 | * But not all options allow that, so we need to check it first. | |
4738 | */ | |
4739 | ret = bdrv_reset_options_allowed(reopen_state->bs, | |
4740 | reopen_state->options, errp); | |
4741 | if (ret) { | |
4742 | goto error; | |
4743 | } | |
4744 | ||
e971aa12 JC |
4745 | ret = drv->bdrv_reopen_prepare(reopen_state, queue, &local_err); |
4746 | if (ret) { | |
4747 | if (local_err != NULL) { | |
4748 | error_propagate(errp, local_err); | |
4749 | } else { | |
f30c66ba | 4750 | bdrv_refresh_filename(reopen_state->bs); |
d8b6895f LC |
4751 | error_setg(errp, "failed while preparing to reopen image '%s'", |
4752 | reopen_state->bs->filename); | |
e971aa12 JC |
4753 | } |
4754 | goto error; | |
4755 | } | |
4756 | } else { | |
4757 | /* It is currently mandatory to have a bdrv_reopen_prepare() | |
4758 | * handler for each supported drv. */ | |
81e5f78a AG |
4759 | error_setg(errp, "Block format '%s' used by node '%s' " |
4760 | "does not support reopening files", drv->format_name, | |
4761 | bdrv_get_device_or_node_name(reopen_state->bs)); | |
e971aa12 JC |
4762 | ret = -1; |
4763 | goto error; | |
4764 | } | |
4765 | ||
9ad08c44 HR |
4766 | drv_prepared = true; |
4767 | ||
bacd9b87 AG |
4768 | /* |
4769 | * We must provide the 'backing' option if the BDS has a backing | |
4770 | * file or if the image file has a backing file name as part of | |
4771 | * its metadata. Otherwise the 'backing' option can be omitted. | |
4772 | */ | |
4773 | if (drv->supports_backing && reopen_state->backing_missing && | |
1d42f48c | 4774 | (reopen_state->bs->backing || reopen_state->bs->backing_file[0])) { |
8546632e AG |
4775 | error_setg(errp, "backing is missing for '%s'", |
4776 | reopen_state->bs->node_name); | |
4777 | ret = -EINVAL; | |
4778 | goto error; | |
4779 | } | |
4780 | ||
cb828c31 AG |
4781 | /* |
4782 | * Allow changing the 'backing' option. The new value can be | |
4783 | * either a reference to an existing node (using its node name) | |
4784 | * or NULL to simply detach the current backing file. | |
4785 | */ | |
ecd30d2d AG |
4786 | ret = bdrv_reopen_parse_file_or_backing(reopen_state, true, |
4787 | change_child_tran, errp); | |
cb828c31 AG |
4788 | if (ret < 0) { |
4789 | goto error; | |
4790 | } | |
4791 | qdict_del(reopen_state->options, "backing"); | |
4792 | ||
ecd30d2d AG |
4793 | /* Allow changing the 'file' option. In this case NULL is not allowed */ |
4794 | ret = bdrv_reopen_parse_file_or_backing(reopen_state, false, | |
4795 | change_child_tran, errp); | |
4796 | if (ret < 0) { | |
4797 | goto error; | |
4798 | } | |
4799 | qdict_del(reopen_state->options, "file"); | |
4800 | ||
4d2cb092 KW |
4801 | /* Options that are not handled are only okay if they are unchanged |
4802 | * compared to the old state. It is expected that some options are only | |
4803 | * used for the initial open, but not reopen (e.g. filename) */ | |
4804 | if (qdict_size(reopen_state->options)) { | |
4805 | const QDictEntry *entry = qdict_first(reopen_state->options); | |
4806 | ||
4807 | do { | |
54fd1b0d HR |
4808 | QObject *new = entry->value; |
4809 | QObject *old = qdict_get(reopen_state->bs->options, entry->key); | |
4810 | ||
db905283 AG |
4811 | /* Allow child references (child_name=node_name) as long as they |
4812 | * point to the current child (i.e. everything stays the same). */ | |
4813 | if (qobject_type(new) == QTYPE_QSTRING) { | |
4814 | BdrvChild *child; | |
4815 | QLIST_FOREACH(child, &reopen_state->bs->children, next) { | |
4816 | if (!strcmp(child->name, entry->key)) { | |
4817 | break; | |
4818 | } | |
4819 | } | |
4820 | ||
4821 | if (child) { | |
410f44f5 MA |
4822 | if (!strcmp(child->bs->node_name, |
4823 | qstring_get_str(qobject_to(QString, new)))) { | |
db905283 AG |
4824 | continue; /* Found child with this name, skip option */ |
4825 | } | |
4826 | } | |
4827 | } | |
4828 | ||
129c7d1c | 4829 | /* |
54fd1b0d HR |
4830 | * TODO: When using -drive to specify blockdev options, all values |
4831 | * will be strings; however, when using -blockdev, blockdev-add or | |
4832 | * filenames using the json:{} pseudo-protocol, they will be | |
4833 | * correctly typed. | |
4834 | * In contrast, reopening options are (currently) always strings | |
4835 | * (because you can only specify them through qemu-io; all other | |
4836 | * callers do not specify any options). | |
4837 | * Therefore, when using anything other than -drive to create a BDS, | |
4838 | * this cannot detect non-string options as unchanged, because | |
4839 | * qobject_is_equal() always returns false for objects of different | |
4840 | * type. In the future, this should be remedied by correctly typing | |
4841 | * all options. For now, this is not too big of an issue because | |
4842 | * the user can simply omit options which cannot be changed anyway, | |
4843 | * so they will stay unchanged. | |
129c7d1c | 4844 | */ |
54fd1b0d | 4845 | if (!qobject_is_equal(new, old)) { |
4d2cb092 KW |
4846 | error_setg(errp, "Cannot change the option '%s'", entry->key); |
4847 | ret = -EINVAL; | |
4848 | goto error; | |
4849 | } | |
4850 | } while ((entry = qdict_next(reopen_state->options, entry))); | |
4851 | } | |
4852 | ||
e971aa12 JC |
4853 | ret = 0; |
4854 | ||
4c8350fe AG |
4855 | /* Restore the original reopen_state->options QDict */ |
4856 | qobject_unref(reopen_state->options); | |
4857 | reopen_state->options = qobject_ref(orig_reopen_opts); | |
4858 | ||
e971aa12 | 4859 | error: |
9ad08c44 HR |
4860 | if (ret < 0 && drv_prepared) { |
4861 | /* drv->bdrv_reopen_prepare() has succeeded, so we need to | |
4862 | * call drv->bdrv_reopen_abort() before signaling an error | |
4863 | * (bdrv_reopen_multiple() will not call bdrv_reopen_abort() | |
4864 | * when the respective bdrv_reopen_prepare() has failed) */ | |
4865 | if (drv->bdrv_reopen_abort) { | |
4866 | drv->bdrv_reopen_abort(reopen_state); | |
4867 | } | |
4868 | } | |
ccf9dc07 | 4869 | qemu_opts_del(opts); |
4c8350fe | 4870 | qobject_unref(orig_reopen_opts); |
593b3071 | 4871 | g_free(discard); |
e971aa12 JC |
4872 | return ret; |
4873 | } | |
4874 | ||
4875 | /* | |
4876 | * Takes the staged changes for the reopen from bdrv_reopen_prepare(), and | |
4877 | * makes them final by swapping the staging BlockDriverState contents into | |
4878 | * the active BlockDriverState contents. | |
4879 | */ | |
53e96d1e | 4880 | static void bdrv_reopen_commit(BDRVReopenState *reopen_state) |
e971aa12 JC |
4881 | { |
4882 | BlockDriver *drv; | |
50bf65ba | 4883 | BlockDriverState *bs; |
50196d7a | 4884 | BdrvChild *child; |
e971aa12 JC |
4885 | |
4886 | assert(reopen_state != NULL); | |
50bf65ba VSO |
4887 | bs = reopen_state->bs; |
4888 | drv = bs->drv; | |
e971aa12 | 4889 | assert(drv != NULL); |
da359909 | 4890 | GLOBAL_STATE_CODE(); |
e971aa12 JC |
4891 | |
4892 | /* If there are any driver level actions to take */ | |
4893 | if (drv->bdrv_reopen_commit) { | |
4894 | drv->bdrv_reopen_commit(reopen_state); | |
4895 | } | |
4896 | ||
4897 | /* set BDS specific flags now */ | |
cb3e7f08 | 4898 | qobject_unref(bs->explicit_options); |
4c8350fe | 4899 | qobject_unref(bs->options); |
ab5b5228 AG |
4900 | qobject_ref(reopen_state->explicit_options); |
4901 | qobject_ref(reopen_state->options); | |
145f598e | 4902 | |
50bf65ba | 4903 | bs->explicit_options = reopen_state->explicit_options; |
4c8350fe | 4904 | bs->options = reopen_state->options; |
50bf65ba | 4905 | bs->open_flags = reopen_state->flags; |
543770bd | 4906 | bs->detect_zeroes = reopen_state->detect_zeroes; |
355ef4ac | 4907 | |
50196d7a AG |
4908 | /* Remove child references from bs->options and bs->explicit_options. |
4909 | * Child options were already removed in bdrv_reopen_queue_child() */ | |
4910 | QLIST_FOREACH(child, &bs->children, next) { | |
4911 | qdict_del(bs->explicit_options, child->name); | |
4912 | qdict_del(bs->options, child->name); | |
4913 | } | |
3d0e8743 VSO |
4914 | /* backing is probably removed, so it's not handled by previous loop */ |
4915 | qdict_del(bs->explicit_options, "backing"); | |
4916 | qdict_del(bs->options, "backing"); | |
4917 | ||
1e4c797c | 4918 | bdrv_refresh_limits(bs, NULL, NULL); |
e971aa12 JC |
4919 | } |
4920 | ||
4921 | /* | |
4922 | * Abort the reopen, and delete and free the staged changes in | |
4923 | * reopen_state | |
4924 | */ | |
53e96d1e | 4925 | static void bdrv_reopen_abort(BDRVReopenState *reopen_state) |
e971aa12 JC |
4926 | { |
4927 | BlockDriver *drv; | |
4928 | ||
4929 | assert(reopen_state != NULL); | |
4930 | drv = reopen_state->bs->drv; | |
4931 | assert(drv != NULL); | |
da359909 | 4932 | GLOBAL_STATE_CODE(); |
e971aa12 JC |
4933 | |
4934 | if (drv->bdrv_reopen_abort) { | |
4935 | drv->bdrv_reopen_abort(reopen_state); | |
4936 | } | |
4937 | } | |
4938 | ||
4939 | ||
64dff520 | 4940 | static void bdrv_close(BlockDriverState *bs) |
fc01f7e7 | 4941 | { |
33384421 | 4942 | BdrvAioNotifier *ban, *ban_next; |
50a3efb0 | 4943 | BdrvChild *child, *next; |
33384421 | 4944 | |
f791bf7f | 4945 | GLOBAL_STATE_CODE(); |
30f55fb8 | 4946 | assert(!bs->refcnt); |
99b7e775 | 4947 | |
fc27291d | 4948 | bdrv_drained_begin(bs); /* complete I/O */ |
58fda173 | 4949 | bdrv_flush(bs); |
53ec73e2 | 4950 | bdrv_drain(bs); /* in case flush left pending I/O */ |
fc27291d | 4951 | |
3cbc002c | 4952 | if (bs->drv) { |
3c005293 | 4953 | if (bs->drv->bdrv_close) { |
7b99a266 | 4954 | /* Must unfreeze all children, so bdrv_unref_child() works */ |
3c005293 VSO |
4955 | bs->drv->bdrv_close(bs); |
4956 | } | |
9a4f4c31 | 4957 | bs->drv = NULL; |
50a3efb0 | 4958 | } |
9a7dedbc | 4959 | |
50a3efb0 | 4960 | QLIST_FOREACH_SAFE(child, &bs->children, next, next) { |
dd4118c7 | 4961 | bdrv_unref_child(bs, child); |
b338082b | 4962 | } |
98f90dba | 4963 | |
5bb04747 VSO |
4964 | assert(!bs->backing); |
4965 | assert(!bs->file); | |
50a3efb0 AG |
4966 | g_free(bs->opaque); |
4967 | bs->opaque = NULL; | |
d73415a3 | 4968 | qatomic_set(&bs->copy_on_read, 0); |
50a3efb0 AG |
4969 | bs->backing_file[0] = '\0'; |
4970 | bs->backing_format[0] = '\0'; | |
4971 | bs->total_sectors = 0; | |
4972 | bs->encrypted = false; | |
4973 | bs->sg = false; | |
cb3e7f08 MAL |
4974 | qobject_unref(bs->options); |
4975 | qobject_unref(bs->explicit_options); | |
50a3efb0 AG |
4976 | bs->options = NULL; |
4977 | bs->explicit_options = NULL; | |
cb3e7f08 | 4978 | qobject_unref(bs->full_open_options); |
50a3efb0 | 4979 | bs->full_open_options = NULL; |
0bc329fb HR |
4980 | g_free(bs->block_status_cache); |
4981 | bs->block_status_cache = NULL; | |
50a3efb0 | 4982 | |
cca43ae1 VSO |
4983 | bdrv_release_named_dirty_bitmaps(bs); |
4984 | assert(QLIST_EMPTY(&bs->dirty_bitmaps)); | |
4985 | ||
33384421 HR |
4986 | QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_next) { |
4987 | g_free(ban); | |
4988 | } | |
4989 | QLIST_INIT(&bs->aio_notifiers); | |
fc27291d | 4990 | bdrv_drained_end(bs); |
1a6d3bd2 GK |
4991 | |
4992 | /* | |
4993 | * If we're still inside some bdrv_drain_all_begin()/end() sections, end | |
4994 | * them now since this BDS won't exist anymore when bdrv_drain_all_end() | |
4995 | * gets called. | |
4996 | */ | |
4997 | if (bs->quiesce_counter) { | |
4998 | bdrv_drain_all_end_quiesce(bs); | |
4999 | } | |
b338082b FB |
5000 | } |
5001 | ||
2bc93fed MK |
5002 | void bdrv_close_all(void) |
5003 | { | |
f791bf7f | 5004 | GLOBAL_STATE_CODE(); |
880eeec6 | 5005 | assert(job_next(NULL) == NULL); |
ca9bd24c HR |
5006 | |
5007 | /* Drop references from requests still in flight, such as canceled block | |
5008 | * jobs whose AIO context has not been polled yet */ | |
5009 | bdrv_drain_all(); | |
2bc93fed | 5010 | |
ca9bd24c HR |
5011 | blk_remove_all_bs(); |
5012 | blockdev_close_all_bdrv_states(); | |
ed78cda3 | 5013 | |
a1a2af07 | 5014 | assert(QTAILQ_EMPTY(&all_bdrv_states)); |
2bc93fed MK |
5015 | } |
5016 | ||
d0ac0380 KW |
5017 | static bool should_update_child(BdrvChild *c, BlockDriverState *to) |
5018 | { | |
2f30b7c3 VSO |
5019 | GQueue *queue; |
5020 | GHashTable *found; | |
5021 | bool ret; | |
d0ac0380 | 5022 | |
bd86fb99 | 5023 | if (c->klass->stay_at_node) { |
d0ac0380 KW |
5024 | return false; |
5025 | } | |
5026 | ||
ec9f10fe HR |
5027 | /* If the child @c belongs to the BDS @to, replacing the current |
5028 | * c->bs by @to would mean to create a loop. | |
5029 | * | |
5030 | * Such a case occurs when appending a BDS to a backing chain. | |
5031 | * For instance, imagine the following chain: | |
5032 | * | |
5033 | * guest device -> node A -> further backing chain... | |
5034 | * | |
5035 | * Now we create a new BDS B which we want to put on top of this | |
5036 | * chain, so we first attach A as its backing node: | |
5037 | * | |
5038 | * node B | |
5039 | * | | |
5040 | * v | |
5041 | * guest device -> node A -> further backing chain... | |
5042 | * | |
5043 | * Finally we want to replace A by B. When doing that, we want to | |
5044 | * replace all pointers to A by pointers to B -- except for the | |
5045 | * pointer from B because (1) that would create a loop, and (2) | |
5046 | * that pointer should simply stay intact: | |
5047 | * | |
5048 | * guest device -> node B | |
5049 | * | | |
5050 | * v | |
5051 | * node A -> further backing chain... | |
5052 | * | |
5053 | * In general, when replacing a node A (c->bs) by a node B (@to), | |
5054 | * if A is a child of B, that means we cannot replace A by B there | |
5055 | * because that would create a loop. Silently detaching A from B | |
5056 | * is also not really an option. So overall just leaving A in | |
2f30b7c3 VSO |
5057 | * place there is the most sensible choice. |
5058 | * | |
5059 | * We would also create a loop in any cases where @c is only | |
5060 | * indirectly referenced by @to. Prevent this by returning false | |
5061 | * if @c is found (by breadth-first search) anywhere in the whole | |
5062 | * subtree of @to. | |
5063 | */ | |
5064 | ||
5065 | ret = true; | |
5066 | found = g_hash_table_new(NULL, NULL); | |
5067 | g_hash_table_add(found, to); | |
5068 | queue = g_queue_new(); | |
5069 | g_queue_push_tail(queue, to); | |
5070 | ||
5071 | while (!g_queue_is_empty(queue)) { | |
5072 | BlockDriverState *v = g_queue_pop_head(queue); | |
5073 | BdrvChild *c2; | |
5074 | ||
5075 | QLIST_FOREACH(c2, &v->children, next) { | |
5076 | if (c2 == c) { | |
5077 | ret = false; | |
5078 | break; | |
5079 | } | |
5080 | ||
5081 | if (g_hash_table_contains(found, c2->bs)) { | |
5082 | continue; | |
5083 | } | |
5084 | ||
5085 | g_queue_push_tail(queue, c2->bs); | |
5086 | g_hash_table_add(found, c2->bs); | |
d0ac0380 KW |
5087 | } |
5088 | } | |
5089 | ||
2f30b7c3 VSO |
5090 | g_queue_free(queue); |
5091 | g_hash_table_destroy(found); | |
5092 | ||
5093 | return ret; | |
d0ac0380 KW |
5094 | } |
5095 | ||
57f08941 | 5096 | static void bdrv_remove_child_commit(void *opaque) |
46541ee5 | 5097 | { |
bdb73476 | 5098 | GLOBAL_STATE_CODE(); |
5bb04747 | 5099 | bdrv_child_free(opaque); |
82b54cf5 HR |
5100 | } |
5101 | ||
57f08941 VSO |
5102 | static TransactionActionDrv bdrv_remove_child_drv = { |
5103 | .commit = bdrv_remove_child_commit, | |
46541ee5 VSO |
5104 | }; |
5105 | ||
57f08941 VSO |
5106 | /* Function doesn't update permissions, caller is responsible for this. */ |
5107 | static void bdrv_remove_child(BdrvChild *child, Transaction *tran) | |
46541ee5 | 5108 | { |
46541ee5 VSO |
5109 | if (!child) { |
5110 | return; | |
5111 | } | |
5112 | ||
5113 | if (child->bs) { | |
23987471 KW |
5114 | BlockDriverState *bs = child->bs; |
5115 | bdrv_drained_begin(bs); | |
a2c37a30 | 5116 | bdrv_replace_child_tran(child, NULL, tran); |
23987471 | 5117 | bdrv_drained_end(bs); |
46541ee5 VSO |
5118 | } |
5119 | ||
57f08941 | 5120 | tran_add(tran, &bdrv_remove_child_drv, child); |
46541ee5 VSO |
5121 | } |
5122 | ||
23987471 KW |
5123 | static void undrain_on_clean_cb(void *opaque) |
5124 | { | |
5125 | bdrv_drained_end(opaque); | |
5126 | } | |
5127 | ||
5128 | static TransactionActionDrv undrain_on_clean = { | |
5129 | .clean = undrain_on_clean_cb, | |
5130 | }; | |
5131 | ||
117caba9 VSO |
5132 | static int bdrv_replace_node_noperm(BlockDriverState *from, |
5133 | BlockDriverState *to, | |
5134 | bool auto_skip, Transaction *tran, | |
5135 | Error **errp) | |
5136 | { | |
5137 | BdrvChild *c, *next; | |
5138 | ||
bdb73476 | 5139 | GLOBAL_STATE_CODE(); |
82b54cf5 | 5140 | |
23987471 KW |
5141 | bdrv_drained_begin(from); |
5142 | bdrv_drained_begin(to); | |
5143 | tran_add(tran, &undrain_on_clean, from); | |
5144 | tran_add(tran, &undrain_on_clean, to); | |
5145 | ||
117caba9 VSO |
5146 | QLIST_FOREACH_SAFE(c, &from->parents, next_parent, next) { |
5147 | assert(c->bs == from); | |
5148 | if (!should_update_child(c, to)) { | |
5149 | if (auto_skip) { | |
5150 | continue; | |
5151 | } | |
5152 | error_setg(errp, "Should not change '%s' link to '%s'", | |
5153 | c->name, from->node_name); | |
5154 | return -EINVAL; | |
5155 | } | |
5156 | if (c->frozen) { | |
5157 | error_setg(errp, "Cannot change '%s' link to '%s'", | |
5158 | c->name, from->node_name); | |
5159 | return -EPERM; | |
5160 | } | |
0f0b1e29 | 5161 | bdrv_replace_child_tran(c, to, tran); |
117caba9 VSO |
5162 | } |
5163 | ||
5164 | return 0; | |
5165 | } | |
5166 | ||
313274bb VSO |
5167 | /* |
5168 | * With auto_skip=true bdrv_replace_node_common skips updating from parents | |
5169 | * if it creates a parent-child relation loop or if parent is block-job. | |
5170 | * | |
5171 | * With auto_skip=false the error is returned if from has a parent which should | |
5172 | * not be updated. | |
3108a15c VSO |
5173 | * |
5174 | * With @detach_subchain=true @to must be in a backing chain of @from. In this | |
5175 | * case backing link of the cow-parent of @to is removed. | |
313274bb | 5176 | */ |
a1e708fc VSO |
5177 | static int bdrv_replace_node_common(BlockDriverState *from, |
5178 | BlockDriverState *to, | |
3108a15c VSO |
5179 | bool auto_skip, bool detach_subchain, |
5180 | Error **errp) | |
dd62f1ca | 5181 | { |
3bb0e298 | 5182 | Transaction *tran = tran_new(); |
3bb0e298 | 5183 | g_autoptr(GSList) refresh_list = NULL; |
2d369d6e | 5184 | BlockDriverState *to_cow_parent = NULL; |
234ac1a9 KW |
5185 | int ret; |
5186 | ||
bdb73476 | 5187 | GLOBAL_STATE_CODE(); |
82b54cf5 | 5188 | |
3108a15c VSO |
5189 | if (detach_subchain) { |
5190 | assert(bdrv_chain_contains(from, to)); | |
5191 | assert(from != to); | |
5192 | for (to_cow_parent = from; | |
5193 | bdrv_filter_or_cow_bs(to_cow_parent) != to; | |
5194 | to_cow_parent = bdrv_filter_or_cow_bs(to_cow_parent)) | |
5195 | { | |
5196 | ; | |
5197 | } | |
5198 | } | |
5199 | ||
234ac1a9 KW |
5200 | /* Make sure that @from doesn't go away until we have successfully attached |
5201 | * all of its parents to @to. */ | |
5202 | bdrv_ref(from); | |
dd62f1ca | 5203 | |
f871abd6 | 5204 | assert(qemu_get_current_aio_context() == qemu_get_aio_context()); |
30dd65f3 | 5205 | assert(bdrv_get_aio_context(from) == bdrv_get_aio_context(to)); |
f871abd6 KW |
5206 | bdrv_drained_begin(from); |
5207 | ||
3bb0e298 VSO |
5208 | /* |
5209 | * Do the replacement without permission update. | |
5210 | * Replacement may influence the permissions, we should calculate new | |
5211 | * permissions based on new graph. If we fail, we'll roll-back the | |
5212 | * replacement. | |
5213 | */ | |
117caba9 VSO |
5214 | ret = bdrv_replace_node_noperm(from, to, auto_skip, tran, errp); |
5215 | if (ret < 0) { | |
5216 | goto out; | |
234ac1a9 KW |
5217 | } |
5218 | ||
3108a15c | 5219 | if (detach_subchain) { |
f38eaec4 | 5220 | bdrv_remove_child(bdrv_filter_or_cow_child(to_cow_parent), tran); |
3108a15c VSO |
5221 | } |
5222 | ||
fb0ff4d1 VSO |
5223 | refresh_list = g_slist_prepend(refresh_list, to); |
5224 | refresh_list = g_slist_prepend(refresh_list, from); | |
9bd910e2 | 5225 | |
3bb0e298 VSO |
5226 | ret = bdrv_list_refresh_perms(refresh_list, NULL, tran, errp); |
5227 | if (ret < 0) { | |
5228 | goto out; | |
dd62f1ca | 5229 | } |
234ac1a9 | 5230 | |
a1e708fc VSO |
5231 | ret = 0; |
5232 | ||
234ac1a9 | 5233 | out: |
3bb0e298 VSO |
5234 | tran_finalize(tran, ret); |
5235 | ||
f871abd6 | 5236 | bdrv_drained_end(from); |
234ac1a9 | 5237 | bdrv_unref(from); |
a1e708fc VSO |
5238 | |
5239 | return ret; | |
dd62f1ca KW |
5240 | } |
5241 | ||
a1e708fc VSO |
5242 | int bdrv_replace_node(BlockDriverState *from, BlockDriverState *to, |
5243 | Error **errp) | |
313274bb | 5244 | { |
f791bf7f EGE |
5245 | GLOBAL_STATE_CODE(); |
5246 | ||
3108a15c VSO |
5247 | return bdrv_replace_node_common(from, to, true, false, errp); |
5248 | } | |
5249 | ||
5250 | int bdrv_drop_filter(BlockDriverState *bs, Error **errp) | |
5251 | { | |
f791bf7f EGE |
5252 | GLOBAL_STATE_CODE(); |
5253 | ||
3108a15c VSO |
5254 | return bdrv_replace_node_common(bs, bdrv_filter_or_cow_bs(bs), true, true, |
5255 | errp); | |
313274bb VSO |
5256 | } |
5257 | ||
4ddc07ca PB |
5258 | /* |
5259 | * Add new bs contents at the top of an image chain while the chain is | |
5260 | * live, while keeping required fields on the top layer. | |
5261 | * | |
5262 | * This will modify the BlockDriverState fields, and swap contents | |
5263 | * between bs_new and bs_top. Both bs_new and bs_top are modified. | |
5264 | * | |
2272edcf VSO |
5265 | * bs_new must not be attached to a BlockBackend and must not have backing |
5266 | * child. | |
4ddc07ca PB |
5267 | * |
5268 | * This function does not create any image files. | |
5269 | */ | |
a1e708fc VSO |
5270 | int bdrv_append(BlockDriverState *bs_new, BlockDriverState *bs_top, |
5271 | Error **errp) | |
4ddc07ca | 5272 | { |
2272edcf | 5273 | int ret; |
5bb04747 | 5274 | BdrvChild *child; |
2272edcf VSO |
5275 | Transaction *tran = tran_new(); |
5276 | ||
f791bf7f EGE |
5277 | GLOBAL_STATE_CODE(); |
5278 | ||
2272edcf VSO |
5279 | assert(!bs_new->backing); |
5280 | ||
5bb04747 VSO |
5281 | child = bdrv_attach_child_noperm(bs_new, bs_top, "backing", |
5282 | &child_of_bds, bdrv_backing_role(bs_new), | |
5283 | tran, errp); | |
5284 | if (!child) { | |
5285 | ret = -EINVAL; | |
2272edcf | 5286 | goto out; |
b2c2832c | 5287 | } |
dd62f1ca | 5288 | |
2272edcf | 5289 | ret = bdrv_replace_node_noperm(bs_top, bs_new, true, tran, errp); |
a1e708fc | 5290 | if (ret < 0) { |
2272edcf | 5291 | goto out; |
234ac1a9 | 5292 | } |
4ddc07ca | 5293 | |
f1316edb | 5294 | ret = bdrv_refresh_perms(bs_new, tran, errp); |
2272edcf VSO |
5295 | out: |
5296 | tran_finalize(tran, ret); | |
5297 | ||
1e4c797c | 5298 | bdrv_refresh_limits(bs_top, NULL, NULL); |
2272edcf VSO |
5299 | |
5300 | return ret; | |
8802d1fd JC |
5301 | } |
5302 | ||
bd8f4c42 VSO |
5303 | /* Not for empty child */ |
5304 | int bdrv_replace_child_bs(BdrvChild *child, BlockDriverState *new_bs, | |
5305 | Error **errp) | |
5306 | { | |
5307 | int ret; | |
5308 | Transaction *tran = tran_new(); | |
bd8f4c42 VSO |
5309 | g_autoptr(GSList) refresh_list = NULL; |
5310 | BlockDriverState *old_bs = child->bs; | |
5311 | ||
f791bf7f EGE |
5312 | GLOBAL_STATE_CODE(); |
5313 | ||
bd8f4c42 VSO |
5314 | bdrv_ref(old_bs); |
5315 | bdrv_drained_begin(old_bs); | |
5316 | bdrv_drained_begin(new_bs); | |
5317 | ||
0f0b1e29 | 5318 | bdrv_replace_child_tran(child, new_bs, tran); |
bd8f4c42 | 5319 | |
fb0ff4d1 VSO |
5320 | refresh_list = g_slist_prepend(refresh_list, old_bs); |
5321 | refresh_list = g_slist_prepend(refresh_list, new_bs); | |
bd8f4c42 VSO |
5322 | |
5323 | ret = bdrv_list_refresh_perms(refresh_list, NULL, tran, errp); | |
5324 | ||
5325 | tran_finalize(tran, ret); | |
5326 | ||
5327 | bdrv_drained_end(old_bs); | |
5328 | bdrv_drained_end(new_bs); | |
5329 | bdrv_unref(old_bs); | |
5330 | ||
5331 | return ret; | |
5332 | } | |
5333 | ||
4f6fd349 | 5334 | static void bdrv_delete(BlockDriverState *bs) |
b338082b | 5335 | { |
3718d8ab | 5336 | assert(bdrv_op_blocker_is_empty(bs)); |
4f6fd349 | 5337 | assert(!bs->refcnt); |
f791bf7f | 5338 | GLOBAL_STATE_CODE(); |
18846dee | 5339 | |
1b7bdbc1 | 5340 | /* remove from list, if necessary */ |
63eaaae0 KW |
5341 | if (bs->node_name[0] != '\0') { |
5342 | QTAILQ_REMOVE(&graph_bdrv_states, bs, node_list); | |
5343 | } | |
2c1d04e0 HR |
5344 | QTAILQ_REMOVE(&all_bdrv_states, bs, bs_list); |
5345 | ||
30c321f9 AK |
5346 | bdrv_close(bs); |
5347 | ||
7267c094 | 5348 | g_free(bs); |
fc01f7e7 FB |
5349 | } |
5350 | ||
96796fae VSO |
5351 | |
5352 | /* | |
5353 | * Replace @bs by newly created block node. | |
5354 | * | |
5355 | * @options is a QDict of options to pass to the block drivers, or NULL for an | |
5356 | * empty set of options. The reference to the QDict belongs to the block layer | |
5357 | * after the call (even on failure), so if the caller intends to reuse the | |
5358 | * dictionary, it needs to use qobject_ref() before calling bdrv_open. | |
5359 | */ | |
5360 | BlockDriverState *bdrv_insert_node(BlockDriverState *bs, QDict *options, | |
8872ef78 AS |
5361 | int flags, Error **errp) |
5362 | { | |
f053b7e8 VSO |
5363 | ERRP_GUARD(); |
5364 | int ret; | |
b11c8739 VSO |
5365 | BlockDriverState *new_node_bs = NULL; |
5366 | const char *drvname, *node_name; | |
5367 | BlockDriver *drv; | |
5368 | ||
5369 | drvname = qdict_get_try_str(options, "driver"); | |
5370 | if (!drvname) { | |
5371 | error_setg(errp, "driver is not specified"); | |
5372 | goto fail; | |
5373 | } | |
5374 | ||
5375 | drv = bdrv_find_format(drvname); | |
5376 | if (!drv) { | |
5377 | error_setg(errp, "Unknown driver: '%s'", drvname); | |
5378 | goto fail; | |
5379 | } | |
8872ef78 | 5380 | |
b11c8739 VSO |
5381 | node_name = qdict_get_try_str(options, "node-name"); |
5382 | ||
f791bf7f EGE |
5383 | GLOBAL_STATE_CODE(); |
5384 | ||
b11c8739 VSO |
5385 | new_node_bs = bdrv_new_open_driver_opts(drv, node_name, options, flags, |
5386 | errp); | |
5387 | options = NULL; /* bdrv_new_open_driver() eats options */ | |
5388 | if (!new_node_bs) { | |
8872ef78 | 5389 | error_prepend(errp, "Could not create node: "); |
b11c8739 | 5390 | goto fail; |
8872ef78 AS |
5391 | } |
5392 | ||
5393 | bdrv_drained_begin(bs); | |
f053b7e8 | 5394 | ret = bdrv_replace_node(bs, new_node_bs, errp); |
8872ef78 AS |
5395 | bdrv_drained_end(bs); |
5396 | ||
f053b7e8 VSO |
5397 | if (ret < 0) { |
5398 | error_prepend(errp, "Could not replace node: "); | |
b11c8739 | 5399 | goto fail; |
8872ef78 AS |
5400 | } |
5401 | ||
5402 | return new_node_bs; | |
b11c8739 VSO |
5403 | |
5404 | fail: | |
5405 | qobject_unref(options); | |
5406 | bdrv_unref(new_node_bs); | |
5407 | return NULL; | |
8872ef78 AS |
5408 | } |
5409 | ||
e97fc193 AL |
5410 | /* |
5411 | * Run consistency checks on an image | |
5412 | * | |
e076f338 | 5413 | * Returns 0 if the check could be completed (it doesn't mean that the image is |
a1c7273b | 5414 | * free of errors) or -errno when an internal error occurred. The results of the |
e076f338 | 5415 | * check are stored in res. |
e97fc193 | 5416 | */ |
21c2283e VSO |
5417 | int coroutine_fn bdrv_co_check(BlockDriverState *bs, |
5418 | BdrvCheckResult *res, BdrvCheckMode fix) | |
e97fc193 | 5419 | { |
1581a70d | 5420 | IO_CODE(); |
1b3ff9fe | 5421 | assert_bdrv_graph_readable(); |
908bcd54 HR |
5422 | if (bs->drv == NULL) { |
5423 | return -ENOMEDIUM; | |
5424 | } | |
2fd61638 | 5425 | if (bs->drv->bdrv_co_check == NULL) { |
e97fc193 AL |
5426 | return -ENOTSUP; |
5427 | } | |
5428 | ||
e076f338 | 5429 | memset(res, 0, sizeof(*res)); |
2fd61638 PB |
5430 | return bs->drv->bdrv_co_check(bs, res, fix); |
5431 | } | |
5432 | ||
756e6736 KW |
5433 | /* |
5434 | * Return values: | |
5435 | * 0 - success | |
5436 | * -EINVAL - backing format specified, but no file | |
5437 | * -ENOSPC - can't update the backing file because no space is left in the | |
5438 | * image file header | |
5439 | * -ENOTSUP - format driver doesn't support changing the backing file | |
5440 | */ | |
e54ee1b3 | 5441 | int bdrv_change_backing_file(BlockDriverState *bs, const char *backing_file, |
497a30db | 5442 | const char *backing_fmt, bool require) |
756e6736 KW |
5443 | { |
5444 | BlockDriver *drv = bs->drv; | |
469ef350 | 5445 | int ret; |
756e6736 | 5446 | |
f791bf7f EGE |
5447 | GLOBAL_STATE_CODE(); |
5448 | ||
d470ad42 HR |
5449 | if (!drv) { |
5450 | return -ENOMEDIUM; | |
5451 | } | |
5452 | ||
5f377794 PB |
5453 | /* Backing file format doesn't make sense without a backing file */ |
5454 | if (backing_fmt && !backing_file) { | |
5455 | return -EINVAL; | |
5456 | } | |
5457 | ||
497a30db EB |
5458 | if (require && backing_file && !backing_fmt) { |
5459 | return -EINVAL; | |
e54ee1b3 EB |
5460 | } |
5461 | ||
756e6736 | 5462 | if (drv->bdrv_change_backing_file != NULL) { |
469ef350 | 5463 | ret = drv->bdrv_change_backing_file(bs, backing_file, backing_fmt); |
756e6736 | 5464 | } else { |
469ef350 | 5465 | ret = -ENOTSUP; |
756e6736 | 5466 | } |
469ef350 PB |
5467 | |
5468 | if (ret == 0) { | |
5469 | pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_file ?: ""); | |
5470 | pstrcpy(bs->backing_format, sizeof(bs->backing_format), backing_fmt ?: ""); | |
998c2019 HR |
5471 | pstrcpy(bs->auto_backing_file, sizeof(bs->auto_backing_file), |
5472 | backing_file ?: ""); | |
469ef350 PB |
5473 | } |
5474 | return ret; | |
756e6736 KW |
5475 | } |
5476 | ||
6ebdcee2 | 5477 | /* |
dcf3f9b2 HR |
5478 | * Finds the first non-filter node above bs in the chain between |
5479 | * active and bs. The returned node is either an immediate parent of | |
5480 | * bs, or there are only filter nodes between the two. | |
6ebdcee2 JC |
5481 | * |
5482 | * Returns NULL if bs is not found in active's image chain, | |
5483 | * or if active == bs. | |
4caf0fcd JC |
5484 | * |
5485 | * Returns the bottommost base image if bs == NULL. | |
6ebdcee2 JC |
5486 | */ |
5487 | BlockDriverState *bdrv_find_overlay(BlockDriverState *active, | |
5488 | BlockDriverState *bs) | |
5489 | { | |
f791bf7f EGE |
5490 | |
5491 | GLOBAL_STATE_CODE(); | |
5492 | ||
dcf3f9b2 HR |
5493 | bs = bdrv_skip_filters(bs); |
5494 | active = bdrv_skip_filters(active); | |
5495 | ||
5496 | while (active) { | |
5497 | BlockDriverState *next = bdrv_backing_chain_next(active); | |
5498 | if (bs == next) { | |
5499 | return active; | |
5500 | } | |
5501 | active = next; | |
6ebdcee2 JC |
5502 | } |
5503 | ||
dcf3f9b2 | 5504 | return NULL; |
4caf0fcd | 5505 | } |
6ebdcee2 | 5506 | |
4caf0fcd JC |
5507 | /* Given a BDS, searches for the base layer. */ |
5508 | BlockDriverState *bdrv_find_base(BlockDriverState *bs) | |
5509 | { | |
f791bf7f EGE |
5510 | GLOBAL_STATE_CODE(); |
5511 | ||
4caf0fcd | 5512 | return bdrv_find_overlay(bs, NULL); |
6ebdcee2 JC |
5513 | } |
5514 | ||
2cad1ebe | 5515 | /* |
7b99a266 HR |
5516 | * Return true if at least one of the COW (backing) and filter links |
5517 | * between @bs and @base is frozen. @errp is set if that's the case. | |
0f0998f6 | 5518 | * @base must be reachable from @bs, or NULL. |
2cad1ebe AG |
5519 | */ |
5520 | bool bdrv_is_backing_chain_frozen(BlockDriverState *bs, BlockDriverState *base, | |
5521 | Error **errp) | |
5522 | { | |
5523 | BlockDriverState *i; | |
7b99a266 | 5524 | BdrvChild *child; |
2cad1ebe | 5525 | |
f791bf7f EGE |
5526 | GLOBAL_STATE_CODE(); |
5527 | ||
7b99a266 HR |
5528 | for (i = bs; i != base; i = child_bs(child)) { |
5529 | child = bdrv_filter_or_cow_child(i); | |
5530 | ||
5531 | if (child && child->frozen) { | |
2cad1ebe | 5532 | error_setg(errp, "Cannot change '%s' link from '%s' to '%s'", |
7b99a266 | 5533 | child->name, i->node_name, child->bs->node_name); |
2cad1ebe AG |
5534 | return true; |
5535 | } | |
5536 | } | |
5537 | ||
5538 | return false; | |
5539 | } | |
5540 | ||
5541 | /* | |
7b99a266 | 5542 | * Freeze all COW (backing) and filter links between @bs and @base. |
2cad1ebe AG |
5543 | * If any of the links is already frozen the operation is aborted and |
5544 | * none of the links are modified. | |
0f0998f6 | 5545 | * @base must be reachable from @bs, or NULL. |
2cad1ebe AG |
5546 | * Returns 0 on success. On failure returns < 0 and sets @errp. |
5547 | */ | |
5548 | int bdrv_freeze_backing_chain(BlockDriverState *bs, BlockDriverState *base, | |
5549 | Error **errp) | |
5550 | { | |
5551 | BlockDriverState *i; | |
7b99a266 | 5552 | BdrvChild *child; |
2cad1ebe | 5553 | |
f791bf7f EGE |
5554 | GLOBAL_STATE_CODE(); |
5555 | ||
2cad1ebe AG |
5556 | if (bdrv_is_backing_chain_frozen(bs, base, errp)) { |
5557 | return -EPERM; | |
5558 | } | |
5559 | ||
7b99a266 HR |
5560 | for (i = bs; i != base; i = child_bs(child)) { |
5561 | child = bdrv_filter_or_cow_child(i); | |
5562 | if (child && child->bs->never_freeze) { | |
e5182c1c | 5563 | error_setg(errp, "Cannot freeze '%s' link to '%s'", |
7b99a266 | 5564 | child->name, child->bs->node_name); |
e5182c1c HR |
5565 | return -EPERM; |
5566 | } | |
5567 | } | |
5568 | ||
7b99a266 HR |
5569 | for (i = bs; i != base; i = child_bs(child)) { |
5570 | child = bdrv_filter_or_cow_child(i); | |
5571 | if (child) { | |
5572 | child->frozen = true; | |
0f0998f6 | 5573 | } |
2cad1ebe AG |
5574 | } |
5575 | ||
5576 | return 0; | |
5577 | } | |
5578 | ||
5579 | /* | |
7b99a266 HR |
5580 | * Unfreeze all COW (backing) and filter links between @bs and @base. |
5581 | * The caller must ensure that all links are frozen before using this | |
5582 | * function. | |
0f0998f6 | 5583 | * @base must be reachable from @bs, or NULL. |
2cad1ebe AG |
5584 | */ |
5585 | void bdrv_unfreeze_backing_chain(BlockDriverState *bs, BlockDriverState *base) | |
5586 | { | |
5587 | BlockDriverState *i; | |
7b99a266 | 5588 | BdrvChild *child; |
2cad1ebe | 5589 | |
f791bf7f EGE |
5590 | GLOBAL_STATE_CODE(); |
5591 | ||
7b99a266 HR |
5592 | for (i = bs; i != base; i = child_bs(child)) { |
5593 | child = bdrv_filter_or_cow_child(i); | |
5594 | if (child) { | |
5595 | assert(child->frozen); | |
5596 | child->frozen = false; | |
0f0998f6 | 5597 | } |
2cad1ebe AG |
5598 | } |
5599 | } | |
5600 | ||
6ebdcee2 JC |
5601 | /* |
5602 | * Drops images above 'base' up to and including 'top', and sets the image | |
5603 | * above 'top' to have base as its backing file. | |
5604 | * | |
5605 | * Requires that the overlay to 'top' is opened r/w, so that the backing file | |
5606 | * information in 'bs' can be properly updated. | |
5607 | * | |
5608 | * E.g., this will convert the following chain: | |
5609 | * bottom <- base <- intermediate <- top <- active | |
5610 | * | |
5611 | * to | |
5612 | * | |
5613 | * bottom <- base <- active | |
5614 | * | |
5615 | * It is allowed for bottom==base, in which case it converts: | |
5616 | * | |
5617 | * base <- intermediate <- top <- active | |
5618 | * | |
5619 | * to | |
5620 | * | |
5621 | * base <- active | |
5622 | * | |
54e26900 JC |
5623 | * If backing_file_str is non-NULL, it will be used when modifying top's |
5624 | * overlay image metadata. | |
5625 | * | |
6ebdcee2 JC |
5626 | * Error conditions: |
5627 | * if active == top, that is considered an error | |
5628 | * | |
5629 | */ | |
bde70715 KW |
5630 | int bdrv_drop_intermediate(BlockDriverState *top, BlockDriverState *base, |
5631 | const char *backing_file_str) | |
6ebdcee2 | 5632 | { |
6bd858b3 AG |
5633 | BlockDriverState *explicit_top = top; |
5634 | bool update_inherits_from; | |
d669ed6a | 5635 | BdrvChild *c; |
12fa4af6 | 5636 | Error *local_err = NULL; |
6ebdcee2 | 5637 | int ret = -EIO; |
d669ed6a VSO |
5638 | g_autoptr(GSList) updated_children = NULL; |
5639 | GSList *p; | |
6ebdcee2 | 5640 | |
f791bf7f EGE |
5641 | GLOBAL_STATE_CODE(); |
5642 | ||
6858eba0 | 5643 | bdrv_ref(top); |
631086de | 5644 | bdrv_drained_begin(base); |
6858eba0 | 5645 | |
6ebdcee2 JC |
5646 | if (!top->drv || !base->drv) { |
5647 | goto exit; | |
5648 | } | |
5649 | ||
5db15a57 KW |
5650 | /* Make sure that base is in the backing chain of top */ |
5651 | if (!bdrv_chain_contains(top, base)) { | |
6ebdcee2 JC |
5652 | goto exit; |
5653 | } | |
5654 | ||
6bd858b3 AG |
5655 | /* If 'base' recursively inherits from 'top' then we should set |
5656 | * base->inherits_from to top->inherits_from after 'top' and all | |
5657 | * other intermediate nodes have been dropped. | |
5658 | * If 'top' is an implicit node (e.g. "commit_top") we should skip | |
5659 | * it because no one inherits from it. We use explicit_top for that. */ | |
dcf3f9b2 | 5660 | explicit_top = bdrv_skip_implicit_filters(explicit_top); |
6bd858b3 AG |
5661 | update_inherits_from = bdrv_inherits_from_recursive(base, explicit_top); |
5662 | ||
6ebdcee2 | 5663 | /* success - we can delete the intermediate states, and link top->base */ |
f30c66ba HR |
5664 | if (!backing_file_str) { |
5665 | bdrv_refresh_filename(base); | |
5666 | backing_file_str = base->filename; | |
5667 | } | |
61f09cea | 5668 | |
d669ed6a VSO |
5669 | QLIST_FOREACH(c, &top->parents, next_parent) { |
5670 | updated_children = g_slist_prepend(updated_children, c); | |
5671 | } | |
5672 | ||
3108a15c VSO |
5673 | /* |
5674 | * It seems correct to pass detach_subchain=true here, but it triggers | |
5675 | * one more yet not fixed bug, when due to nested aio_poll loop we switch to | |
5676 | * another drained section, which modify the graph (for example, removing | |
5677 | * the child, which we keep in updated_children list). So, it's a TODO. | |
5678 | * | |
5679 | * Note, bug triggered if pass detach_subchain=true here and run | |
5680 | * test-bdrv-drain. test_drop_intermediate_poll() test-case will crash. | |
5681 | * That's a FIXME. | |
5682 | */ | |
5683 | bdrv_replace_node_common(top, base, false, false, &local_err); | |
d669ed6a VSO |
5684 | if (local_err) { |
5685 | error_report_err(local_err); | |
5686 | goto exit; | |
5687 | } | |
5688 | ||
5689 | for (p = updated_children; p; p = p->next) { | |
5690 | c = p->data; | |
12fa4af6 | 5691 | |
bd86fb99 HR |
5692 | if (c->klass->update_filename) { |
5693 | ret = c->klass->update_filename(c, base, backing_file_str, | |
5694 | &local_err); | |
61f09cea | 5695 | if (ret < 0) { |
d669ed6a VSO |
5696 | /* |
5697 | * TODO: Actually, we want to rollback all previous iterations | |
5698 | * of this loop, and (which is almost impossible) previous | |
5699 | * bdrv_replace_node()... | |
5700 | * | |
5701 | * Note, that c->klass->update_filename may lead to permission | |
5702 | * update, so it's a bad idea to call it inside permission | |
5703 | * update transaction of bdrv_replace_node. | |
5704 | */ | |
61f09cea KW |
5705 | error_report_err(local_err); |
5706 | goto exit; | |
5707 | } | |
5708 | } | |
12fa4af6 | 5709 | } |
6ebdcee2 | 5710 | |
6bd858b3 AG |
5711 | if (update_inherits_from) { |
5712 | base->inherits_from = explicit_top->inherits_from; | |
5713 | } | |
5714 | ||
6ebdcee2 | 5715 | ret = 0; |
6ebdcee2 | 5716 | exit: |
631086de | 5717 | bdrv_drained_end(base); |
6858eba0 | 5718 | bdrv_unref(top); |
6ebdcee2 JC |
5719 | return ret; |
5720 | } | |
5721 | ||
081e4650 | 5722 | /** |
82618d7b | 5723 | * Implementation of BlockDriver.bdrv_co_get_allocated_file_size() that |
081e4650 HR |
5724 | * sums the size of all data-bearing children. (This excludes backing |
5725 | * children.) | |
5726 | */ | |
5727 | static int64_t bdrv_sum_allocated_file_size(BlockDriverState *bs) | |
5728 | { | |
5729 | BdrvChild *child; | |
5730 | int64_t child_size, sum = 0; | |
5731 | ||
5732 | QLIST_FOREACH(child, &bs->children, next) { | |
5733 | if (child->role & (BDRV_CHILD_DATA | BDRV_CHILD_METADATA | | |
5734 | BDRV_CHILD_FILTERED)) | |
5735 | { | |
82618d7b | 5736 | child_size = bdrv_co_get_allocated_file_size(child->bs); |
081e4650 HR |
5737 | if (child_size < 0) { |
5738 | return child_size; | |
5739 | } | |
5740 | sum += child_size; | |
5741 | } | |
5742 | } | |
5743 | ||
5744 | return sum; | |
5745 | } | |
5746 | ||
61007b31 SH |
5747 | /** |
5748 | * Length of a allocated file in bytes. Sparse files are counted by actual | |
5749 | * allocated space. Return < 0 if error or unknown. | |
5750 | */ | |
82618d7b | 5751 | int64_t coroutine_fn bdrv_co_get_allocated_file_size(BlockDriverState *bs) |
71d0770c | 5752 | { |
61007b31 | 5753 | BlockDriver *drv = bs->drv; |
384a48fb EGE |
5754 | IO_CODE(); |
5755 | ||
61007b31 SH |
5756 | if (!drv) { |
5757 | return -ENOMEDIUM; | |
8f4754ed | 5758 | } |
82618d7b EGE |
5759 | if (drv->bdrv_co_get_allocated_file_size) { |
5760 | return drv->bdrv_co_get_allocated_file_size(bs); | |
61007b31 | 5761 | } |
081e4650 HR |
5762 | |
5763 | if (drv->bdrv_file_open) { | |
5764 | /* | |
5765 | * Protocol drivers default to -ENOTSUP (most of their data is | |
5766 | * not stored in any of their children (if they even have any), | |
5767 | * so there is no generic way to figure it out). | |
5768 | */ | |
5769 | return -ENOTSUP; | |
5770 | } else if (drv->is_filter) { | |
5771 | /* Filter drivers default to the size of their filtered child */ | |
82618d7b | 5772 | return bdrv_co_get_allocated_file_size(bdrv_filter_bs(bs)); |
081e4650 HR |
5773 | } else { |
5774 | /* Other drivers default to summing their children's sizes */ | |
5775 | return bdrv_sum_allocated_file_size(bs); | |
1c9805a3 SH |
5776 | } |
5777 | } | |
e7a8a783 | 5778 | |
90880ff1 SH |
5779 | /* |
5780 | * bdrv_measure: | |
5781 | * @drv: Format driver | |
5782 | * @opts: Creation options for new image | |
5783 | * @in_bs: Existing image containing data for new image (may be NULL) | |
5784 | * @errp: Error object | |
5785 | * Returns: A #BlockMeasureInfo (free using qapi_free_BlockMeasureInfo()) | |
5786 | * or NULL on error | |
5787 | * | |
5788 | * Calculate file size required to create a new image. | |
5789 | * | |
5790 | * If @in_bs is given then space for allocated clusters and zero clusters | |
5791 | * from that image are included in the calculation. If @opts contains a | |
5792 | * backing file that is shared by @in_bs then backing clusters may be omitted | |
5793 | * from the calculation. | |
5794 | * | |
5795 | * If @in_bs is NULL then the calculation includes no allocated clusters | |
5796 | * unless a preallocation option is given in @opts. | |
5797 | * | |
5798 | * Note that @in_bs may use a different BlockDriver from @drv. | |
5799 | * | |
5800 | * If an error occurs the @errp pointer is set. | |
5801 | */ | |
5802 | BlockMeasureInfo *bdrv_measure(BlockDriver *drv, QemuOpts *opts, | |
5803 | BlockDriverState *in_bs, Error **errp) | |
5804 | { | |
384a48fb | 5805 | IO_CODE(); |
90880ff1 SH |
5806 | if (!drv->bdrv_measure) { |
5807 | error_setg(errp, "Block driver '%s' does not support size measurement", | |
5808 | drv->format_name); | |
5809 | return NULL; | |
5810 | } | |
5811 | ||
5812 | return drv->bdrv_measure(opts, in_bs, errp); | |
5813 | } | |
5814 | ||
61007b31 SH |
5815 | /** |
5816 | * Return number of sectors on success, -errno on error. | |
1c9805a3 | 5817 | */ |
c86422c5 | 5818 | int64_t coroutine_fn bdrv_co_nb_sectors(BlockDriverState *bs) |
1c9805a3 | 5819 | { |
61007b31 | 5820 | BlockDriver *drv = bs->drv; |
384a48fb | 5821 | IO_CODE(); |
498e386c | 5822 | |
61007b31 SH |
5823 | if (!drv) |
5824 | return -ENOMEDIUM; | |
2572b37a | 5825 | |
61007b31 | 5826 | if (drv->has_variable_length) { |
c86422c5 | 5827 | int ret = bdrv_co_refresh_total_sectors(bs, bs->total_sectors); |
61007b31 SH |
5828 | if (ret < 0) { |
5829 | return ret; | |
1c9805a3 SH |
5830 | } |
5831 | } | |
61007b31 | 5832 | return bs->total_sectors; |
1c9805a3 | 5833 | } |
b338082b | 5834 | |
61007b31 SH |
5835 | /** |
5836 | * Return length in bytes on success, -errno on error. | |
5837 | * The length is always a multiple of BDRV_SECTOR_SIZE. | |
8d3b1a2d | 5838 | */ |
c86422c5 | 5839 | int64_t coroutine_fn bdrv_co_getlength(BlockDriverState *bs) |
8d3b1a2d | 5840 | { |
c86422c5 | 5841 | int64_t ret; |
384a48fb | 5842 | IO_CODE(); |
8d3b1a2d | 5843 | |
c86422c5 | 5844 | ret = bdrv_co_nb_sectors(bs); |
122860ba EB |
5845 | if (ret < 0) { |
5846 | return ret; | |
5847 | } | |
5848 | if (ret > INT64_MAX / BDRV_SECTOR_SIZE) { | |
5849 | return -EFBIG; | |
5850 | } | |
5851 | return ret * BDRV_SECTOR_SIZE; | |
fc01f7e7 FB |
5852 | } |
5853 | ||
61007b31 SH |
5854 | /* return 0 as number of sectors if no device present or error */ |
5855 | void bdrv_get_geometry(BlockDriverState *bs, uint64_t *nb_sectors_ptr) | |
07d27a44 | 5856 | { |
61007b31 | 5857 | int64_t nb_sectors = bdrv_nb_sectors(bs); |
384a48fb | 5858 | IO_CODE(); |
07d27a44 | 5859 | |
61007b31 | 5860 | *nb_sectors_ptr = nb_sectors < 0 ? 0 : nb_sectors; |
07d27a44 MA |
5861 | } |
5862 | ||
54115412 | 5863 | bool bdrv_is_sg(BlockDriverState *bs) |
f08145fe | 5864 | { |
384a48fb | 5865 | IO_CODE(); |
61007b31 | 5866 | return bs->sg; |
f08145fe KW |
5867 | } |
5868 | ||
ae23f786 HR |
5869 | /** |
5870 | * Return whether the given node supports compressed writes. | |
5871 | */ | |
5872 | bool bdrv_supports_compressed_writes(BlockDriverState *bs) | |
5873 | { | |
5874 | BlockDriverState *filtered; | |
384a48fb | 5875 | IO_CODE(); |
ae23f786 HR |
5876 | |
5877 | if (!bs->drv || !block_driver_can_compress(bs->drv)) { | |
5878 | return false; | |
5879 | } | |
5880 | ||
5881 | filtered = bdrv_filter_bs(bs); | |
5882 | if (filtered) { | |
5883 | /* | |
5884 | * Filters can only forward compressed writes, so we have to | |
5885 | * check the child. | |
5886 | */ | |
5887 | return bdrv_supports_compressed_writes(filtered); | |
5888 | } | |
5889 | ||
5890 | return true; | |
5891 | } | |
5892 | ||
61007b31 | 5893 | const char *bdrv_get_format_name(BlockDriverState *bs) |
40b4f539 | 5894 | { |
384a48fb | 5895 | IO_CODE(); |
61007b31 | 5896 | return bs->drv ? bs->drv->format_name : NULL; |
40b4f539 KW |
5897 | } |
5898 | ||
61007b31 | 5899 | static int qsort_strcmp(const void *a, const void *b) |
40b4f539 | 5900 | { |
ceff5bd7 | 5901 | return strcmp(*(char *const *)a, *(char *const *)b); |
40b4f539 KW |
5902 | } |
5903 | ||
61007b31 | 5904 | void bdrv_iterate_format(void (*it)(void *opaque, const char *name), |
9ac404c5 | 5905 | void *opaque, bool read_only) |
40b4f539 | 5906 | { |
61007b31 SH |
5907 | BlockDriver *drv; |
5908 | int count = 0; | |
5909 | int i; | |
5910 | const char **formats = NULL; | |
40b4f539 | 5911 | |
f791bf7f EGE |
5912 | GLOBAL_STATE_CODE(); |
5913 | ||
61007b31 SH |
5914 | QLIST_FOREACH(drv, &bdrv_drivers, list) { |
5915 | if (drv->format_name) { | |
5916 | bool found = false; | |
5917 | int i = count; | |
9ac404c5 AS |
5918 | |
5919 | if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv, read_only)) { | |
5920 | continue; | |
5921 | } | |
5922 | ||
61007b31 SH |
5923 | while (formats && i && !found) { |
5924 | found = !strcmp(formats[--i], drv->format_name); | |
5925 | } | |
e2a305fb | 5926 | |
61007b31 SH |
5927 | if (!found) { |
5928 | formats = g_renew(const char *, formats, count + 1); | |
5929 | formats[count++] = drv->format_name; | |
5930 | } | |
6c5a42ac | 5931 | } |
61007b31 | 5932 | } |
6c5a42ac | 5933 | |
eb0df69f HR |
5934 | for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); i++) { |
5935 | const char *format_name = block_driver_modules[i].format_name; | |
5936 | ||
5937 | if (format_name) { | |
5938 | bool found = false; | |
5939 | int j = count; | |
5940 | ||
9ac404c5 AS |
5941 | if (use_bdrv_whitelist && |
5942 | !bdrv_format_is_whitelisted(format_name, read_only)) { | |
5943 | continue; | |
5944 | } | |
5945 | ||
eb0df69f HR |
5946 | while (formats && j && !found) { |
5947 | found = !strcmp(formats[--j], format_name); | |
5948 | } | |
5949 | ||
5950 | if (!found) { | |
5951 | formats = g_renew(const char *, formats, count + 1); | |
5952 | formats[count++] = format_name; | |
5953 | } | |
5954 | } | |
5955 | } | |
5956 | ||
61007b31 | 5957 | qsort(formats, count, sizeof(formats[0]), qsort_strcmp); |
40b4f539 | 5958 | |
61007b31 SH |
5959 | for (i = 0; i < count; i++) { |
5960 | it(opaque, formats[i]); | |
5961 | } | |
40b4f539 | 5962 | |
61007b31 SH |
5963 | g_free(formats); |
5964 | } | |
40b4f539 | 5965 | |
61007b31 SH |
5966 | /* This function is to find a node in the bs graph */ |
5967 | BlockDriverState *bdrv_find_node(const char *node_name) | |
5968 | { | |
5969 | BlockDriverState *bs; | |
391827eb | 5970 | |
61007b31 | 5971 | assert(node_name); |
f791bf7f | 5972 | GLOBAL_STATE_CODE(); |
40b4f539 | 5973 | |
61007b31 SH |
5974 | QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) { |
5975 | if (!strcmp(node_name, bs->node_name)) { | |
5976 | return bs; | |
40b4f539 KW |
5977 | } |
5978 | } | |
61007b31 | 5979 | return NULL; |
40b4f539 KW |
5980 | } |
5981 | ||
61007b31 | 5982 | /* Put this QMP function here so it can access the static graph_bdrv_states. */ |
facda544 PK |
5983 | BlockDeviceInfoList *bdrv_named_nodes_list(bool flat, |
5984 | Error **errp) | |
40b4f539 | 5985 | { |
9812e712 | 5986 | BlockDeviceInfoList *list; |
61007b31 | 5987 | BlockDriverState *bs; |
40b4f539 | 5988 | |
f791bf7f EGE |
5989 | GLOBAL_STATE_CODE(); |
5990 | ||
61007b31 SH |
5991 | list = NULL; |
5992 | QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) { | |
facda544 | 5993 | BlockDeviceInfo *info = bdrv_block_device_info(NULL, bs, flat, errp); |
61007b31 SH |
5994 | if (!info) { |
5995 | qapi_free_BlockDeviceInfoList(list); | |
5996 | return NULL; | |
301db7c2 | 5997 | } |
9812e712 | 5998 | QAPI_LIST_PREPEND(list, info); |
301db7c2 RH |
5999 | } |
6000 | ||
61007b31 SH |
6001 | return list; |
6002 | } | |
40b4f539 | 6003 | |
5d3b4e99 VSO |
6004 | typedef struct XDbgBlockGraphConstructor { |
6005 | XDbgBlockGraph *graph; | |
6006 | GHashTable *graph_nodes; | |
6007 | } XDbgBlockGraphConstructor; | |
6008 | ||
6009 | static XDbgBlockGraphConstructor *xdbg_graph_new(void) | |
6010 | { | |
6011 | XDbgBlockGraphConstructor *gr = g_new(XDbgBlockGraphConstructor, 1); | |
6012 | ||
6013 | gr->graph = g_new0(XDbgBlockGraph, 1); | |
6014 | gr->graph_nodes = g_hash_table_new(NULL, NULL); | |
6015 | ||
6016 | return gr; | |
6017 | } | |
6018 | ||
6019 | static XDbgBlockGraph *xdbg_graph_finalize(XDbgBlockGraphConstructor *gr) | |
6020 | { | |
6021 | XDbgBlockGraph *graph = gr->graph; | |
6022 | ||
6023 | g_hash_table_destroy(gr->graph_nodes); | |
6024 | g_free(gr); | |
6025 | ||
6026 | return graph; | |
6027 | } | |
6028 | ||
6029 | static uintptr_t xdbg_graph_node_num(XDbgBlockGraphConstructor *gr, void *node) | |
6030 | { | |
6031 | uintptr_t ret = (uintptr_t)g_hash_table_lookup(gr->graph_nodes, node); | |
6032 | ||
6033 | if (ret != 0) { | |
6034 | return ret; | |
6035 | } | |
6036 | ||
6037 | /* | |
6038 | * Start counting from 1, not 0, because 0 interferes with not-found (NULL) | |
6039 | * answer of g_hash_table_lookup. | |
6040 | */ | |
6041 | ret = g_hash_table_size(gr->graph_nodes) + 1; | |
6042 | g_hash_table_insert(gr->graph_nodes, node, (void *)ret); | |
6043 | ||
6044 | return ret; | |
6045 | } | |
6046 | ||
6047 | static void xdbg_graph_add_node(XDbgBlockGraphConstructor *gr, void *node, | |
6048 | XDbgBlockGraphNodeType type, const char *name) | |
6049 | { | |
6050 | XDbgBlockGraphNode *n; | |
6051 | ||
6052 | n = g_new0(XDbgBlockGraphNode, 1); | |
6053 | ||
6054 | n->id = xdbg_graph_node_num(gr, node); | |
6055 | n->type = type; | |
6056 | n->name = g_strdup(name); | |
6057 | ||
9812e712 | 6058 | QAPI_LIST_PREPEND(gr->graph->nodes, n); |
5d3b4e99 VSO |
6059 | } |
6060 | ||
6061 | static void xdbg_graph_add_edge(XDbgBlockGraphConstructor *gr, void *parent, | |
6062 | const BdrvChild *child) | |
6063 | { | |
cdb1cec8 | 6064 | BlockPermission qapi_perm; |
5d3b4e99 | 6065 | XDbgBlockGraphEdge *edge; |
862fded9 | 6066 | GLOBAL_STATE_CODE(); |
5d3b4e99 | 6067 | |
5d3b4e99 VSO |
6068 | edge = g_new0(XDbgBlockGraphEdge, 1); |
6069 | ||
6070 | edge->parent = xdbg_graph_node_num(gr, parent); | |
6071 | edge->child = xdbg_graph_node_num(gr, child->bs); | |
6072 | edge->name = g_strdup(child->name); | |
6073 | ||
cdb1cec8 HR |
6074 | for (qapi_perm = 0; qapi_perm < BLOCK_PERMISSION__MAX; qapi_perm++) { |
6075 | uint64_t flag = bdrv_qapi_perm_to_blk_perm(qapi_perm); | |
6076 | ||
6077 | if (flag & child->perm) { | |
9812e712 | 6078 | QAPI_LIST_PREPEND(edge->perm, qapi_perm); |
5d3b4e99 | 6079 | } |
cdb1cec8 | 6080 | if (flag & child->shared_perm) { |
9812e712 | 6081 | QAPI_LIST_PREPEND(edge->shared_perm, qapi_perm); |
5d3b4e99 VSO |
6082 | } |
6083 | } | |
6084 | ||
9812e712 | 6085 | QAPI_LIST_PREPEND(gr->graph->edges, edge); |
5d3b4e99 VSO |
6086 | } |
6087 | ||
6088 | ||
6089 | XDbgBlockGraph *bdrv_get_xdbg_block_graph(Error **errp) | |
6090 | { | |
6091 | BlockBackend *blk; | |
6092 | BlockJob *job; | |
6093 | BlockDriverState *bs; | |
6094 | BdrvChild *child; | |
6095 | XDbgBlockGraphConstructor *gr = xdbg_graph_new(); | |
6096 | ||
f791bf7f EGE |
6097 | GLOBAL_STATE_CODE(); |
6098 | ||
5d3b4e99 VSO |
6099 | for (blk = blk_all_next(NULL); blk; blk = blk_all_next(blk)) { |
6100 | char *allocated_name = NULL; | |
6101 | const char *name = blk_name(blk); | |
6102 | ||
6103 | if (!*name) { | |
6104 | name = allocated_name = blk_get_attached_dev_id(blk); | |
6105 | } | |
6106 | xdbg_graph_add_node(gr, blk, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_BACKEND, | |
6107 | name); | |
6108 | g_free(allocated_name); | |
6109 | if (blk_root(blk)) { | |
6110 | xdbg_graph_add_edge(gr, blk, blk_root(blk)); | |
6111 | } | |
6112 | } | |
6113 | ||
880eeec6 EGE |
6114 | WITH_JOB_LOCK_GUARD() { |
6115 | for (job = block_job_next_locked(NULL); job; | |
6116 | job = block_job_next_locked(job)) { | |
6117 | GSList *el; | |
5d3b4e99 | 6118 | |
880eeec6 EGE |
6119 | xdbg_graph_add_node(gr, job, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_JOB, |
6120 | job->job.id); | |
6121 | for (el = job->nodes; el; el = el->next) { | |
6122 | xdbg_graph_add_edge(gr, job, (BdrvChild *)el->data); | |
6123 | } | |
5d3b4e99 VSO |
6124 | } |
6125 | } | |
6126 | ||
6127 | QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) { | |
6128 | xdbg_graph_add_node(gr, bs, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_DRIVER, | |
6129 | bs->node_name); | |
6130 | QLIST_FOREACH(child, &bs->children, next) { | |
6131 | xdbg_graph_add_edge(gr, bs, child); | |
6132 | } | |
6133 | } | |
6134 | ||
6135 | return xdbg_graph_finalize(gr); | |
6136 | } | |
6137 | ||
61007b31 SH |
6138 | BlockDriverState *bdrv_lookup_bs(const char *device, |
6139 | const char *node_name, | |
6140 | Error **errp) | |
6141 | { | |
6142 | BlockBackend *blk; | |
6143 | BlockDriverState *bs; | |
40b4f539 | 6144 | |
f791bf7f EGE |
6145 | GLOBAL_STATE_CODE(); |
6146 | ||
61007b31 SH |
6147 | if (device) { |
6148 | blk = blk_by_name(device); | |
40b4f539 | 6149 | |
61007b31 | 6150 | if (blk) { |
9f4ed6fb AG |
6151 | bs = blk_bs(blk); |
6152 | if (!bs) { | |
5433c24f | 6153 | error_setg(errp, "Device '%s' has no medium", device); |
5433c24f HR |
6154 | } |
6155 | ||
9f4ed6fb | 6156 | return bs; |
61007b31 SH |
6157 | } |
6158 | } | |
40b4f539 | 6159 | |
61007b31 SH |
6160 | if (node_name) { |
6161 | bs = bdrv_find_node(node_name); | |
6d519a5f | 6162 | |
61007b31 SH |
6163 | if (bs) { |
6164 | return bs; | |
6165 | } | |
40b4f539 KW |
6166 | } |
6167 | ||
785ec4b1 | 6168 | error_setg(errp, "Cannot find device=\'%s\' nor node-name=\'%s\'", |
61007b31 SH |
6169 | device ? device : "", |
6170 | node_name ? node_name : ""); | |
6171 | return NULL; | |
40b4f539 KW |
6172 | } |
6173 | ||
61007b31 SH |
6174 | /* If 'base' is in the same chain as 'top', return true. Otherwise, |
6175 | * return false. If either argument is NULL, return false. */ | |
6176 | bool bdrv_chain_contains(BlockDriverState *top, BlockDriverState *base) | |
83f64091 | 6177 | { |
f791bf7f EGE |
6178 | |
6179 | GLOBAL_STATE_CODE(); | |
6180 | ||
61007b31 | 6181 | while (top && top != base) { |
dcf3f9b2 | 6182 | top = bdrv_filter_or_cow_bs(top); |
02c50efe | 6183 | } |
61007b31 SH |
6184 | |
6185 | return top != NULL; | |
02c50efe FZ |
6186 | } |
6187 | ||
61007b31 | 6188 | BlockDriverState *bdrv_next_node(BlockDriverState *bs) |
02c50efe | 6189 | { |
f791bf7f | 6190 | GLOBAL_STATE_CODE(); |
61007b31 SH |
6191 | if (!bs) { |
6192 | return QTAILQ_FIRST(&graph_bdrv_states); | |
02c50efe | 6193 | } |
61007b31 | 6194 | return QTAILQ_NEXT(bs, node_list); |
83f64091 FB |
6195 | } |
6196 | ||
0f12264e KW |
6197 | BlockDriverState *bdrv_next_all_states(BlockDriverState *bs) |
6198 | { | |
f791bf7f | 6199 | GLOBAL_STATE_CODE(); |
0f12264e KW |
6200 | if (!bs) { |
6201 | return QTAILQ_FIRST(&all_bdrv_states); | |
6202 | } | |
6203 | return QTAILQ_NEXT(bs, bs_list); | |
6204 | } | |
6205 | ||
61007b31 | 6206 | const char *bdrv_get_node_name(const BlockDriverState *bs) |
83f64091 | 6207 | { |
384a48fb | 6208 | IO_CODE(); |
61007b31 | 6209 | return bs->node_name; |
beac80cd FB |
6210 | } |
6211 | ||
1f0c461b | 6212 | const char *bdrv_get_parent_name(const BlockDriverState *bs) |
4c265bf9 KW |
6213 | { |
6214 | BdrvChild *c; | |
6215 | const char *name; | |
967d7905 | 6216 | IO_CODE(); |
4c265bf9 KW |
6217 | |
6218 | /* If multiple parents have a name, just pick the first one. */ | |
6219 | QLIST_FOREACH(c, &bs->parents, next_parent) { | |
bd86fb99 HR |
6220 | if (c->klass->get_name) { |
6221 | name = c->klass->get_name(c); | |
4c265bf9 KW |
6222 | if (name && *name) { |
6223 | return name; | |
6224 | } | |
6225 | } | |
6226 | } | |
6227 | ||
6228 | return NULL; | |
6229 | } | |
6230 | ||
61007b31 SH |
6231 | /* TODO check what callers really want: bs->node_name or blk_name() */ |
6232 | const char *bdrv_get_device_name(const BlockDriverState *bs) | |
beac80cd | 6233 | { |
384a48fb | 6234 | IO_CODE(); |
4c265bf9 | 6235 | return bdrv_get_parent_name(bs) ?: ""; |
f141eafe | 6236 | } |
83f64091 | 6237 | |
61007b31 SH |
6238 | /* This can be used to identify nodes that might not have a device |
6239 | * name associated. Since node and device names live in the same | |
6240 | * namespace, the result is unambiguous. The exception is if both are | |
6241 | * absent, then this returns an empty (non-null) string. */ | |
6242 | const char *bdrv_get_device_or_node_name(const BlockDriverState *bs) | |
f141eafe | 6243 | { |
384a48fb | 6244 | IO_CODE(); |
4c265bf9 | 6245 | return bdrv_get_parent_name(bs) ?: bs->node_name; |
beac80cd | 6246 | } |
beac80cd | 6247 | |
61007b31 | 6248 | int bdrv_get_flags(BlockDriverState *bs) |
0b5a2445 | 6249 | { |
15aee7ac | 6250 | IO_CODE(); |
61007b31 | 6251 | return bs->open_flags; |
0b5a2445 PB |
6252 | } |
6253 | ||
61007b31 | 6254 | int bdrv_has_zero_init_1(BlockDriverState *bs) |
68485420 | 6255 | { |
f791bf7f | 6256 | GLOBAL_STATE_CODE(); |
61007b31 | 6257 | return 1; |
0b5a2445 PB |
6258 | } |
6259 | ||
61007b31 | 6260 | int bdrv_has_zero_init(BlockDriverState *bs) |
0b5a2445 | 6261 | { |
93393e69 | 6262 | BlockDriverState *filtered; |
f791bf7f | 6263 | GLOBAL_STATE_CODE(); |
93393e69 | 6264 | |
d470ad42 HR |
6265 | if (!bs->drv) { |
6266 | return 0; | |
6267 | } | |
0b5a2445 | 6268 | |
61007b31 SH |
6269 | /* If BS is a copy on write image, it is initialized to |
6270 | the contents of the base image, which may not be zeroes. */ | |
34778172 | 6271 | if (bdrv_cow_child(bs)) { |
61007b31 SH |
6272 | return 0; |
6273 | } | |
6274 | if (bs->drv->bdrv_has_zero_init) { | |
6275 | return bs->drv->bdrv_has_zero_init(bs); | |
0b5a2445 | 6276 | } |
93393e69 HR |
6277 | |
6278 | filtered = bdrv_filter_bs(bs); | |
6279 | if (filtered) { | |
6280 | return bdrv_has_zero_init(filtered); | |
5a612c00 | 6281 | } |
61007b31 SH |
6282 | |
6283 | /* safe default */ | |
6284 | return 0; | |
68485420 KW |
6285 | } |
6286 | ||
61007b31 | 6287 | bool bdrv_can_write_zeroes_with_unmap(BlockDriverState *bs) |
68485420 | 6288 | { |
384a48fb | 6289 | IO_CODE(); |
2f0342ef | 6290 | if (!(bs->open_flags & BDRV_O_UNMAP)) { |
61007b31 SH |
6291 | return false; |
6292 | } | |
68485420 | 6293 | |
e24d813b | 6294 | return bs->supported_zero_flags & BDRV_REQ_MAY_UNMAP; |
68485420 KW |
6295 | } |
6296 | ||
61007b31 SH |
6297 | void bdrv_get_backing_filename(BlockDriverState *bs, |
6298 | char *filename, int filename_size) | |
016f5cf6 | 6299 | { |
384a48fb | 6300 | IO_CODE(); |
61007b31 SH |
6301 | pstrcpy(filename, filename_size, bs->backing_file); |
6302 | } | |
d318aea9 | 6303 | |
3d47eb0a | 6304 | int coroutine_fn bdrv_co_get_info(BlockDriverState *bs, BlockDriverInfo *bdi) |
61007b31 | 6305 | { |
8b117001 | 6306 | int ret; |
61007b31 | 6307 | BlockDriver *drv = bs->drv; |
384a48fb | 6308 | IO_CODE(); |
5a612c00 MP |
6309 | /* if bs->drv == NULL, bs is closed, so there's nothing to do here */ |
6310 | if (!drv) { | |
61007b31 | 6311 | return -ENOMEDIUM; |
5a612c00 | 6312 | } |
3d47eb0a | 6313 | if (!drv->bdrv_co_get_info) { |
93393e69 HR |
6314 | BlockDriverState *filtered = bdrv_filter_bs(bs); |
6315 | if (filtered) { | |
3d47eb0a | 6316 | return bdrv_co_get_info(filtered, bdi); |
5a612c00 | 6317 | } |
61007b31 | 6318 | return -ENOTSUP; |
5a612c00 | 6319 | } |
61007b31 | 6320 | memset(bdi, 0, sizeof(*bdi)); |
3d47eb0a | 6321 | ret = drv->bdrv_co_get_info(bs, bdi); |
8b117001 VSO |
6322 | if (ret < 0) { |
6323 | return ret; | |
6324 | } | |
6325 | ||
6326 | if (bdi->cluster_size > BDRV_MAX_ALIGNMENT) { | |
6327 | return -EINVAL; | |
6328 | } | |
6329 | ||
6330 | return 0; | |
61007b31 | 6331 | } |
016f5cf6 | 6332 | |
1bf6e9ca AS |
6333 | ImageInfoSpecific *bdrv_get_specific_info(BlockDriverState *bs, |
6334 | Error **errp) | |
61007b31 SH |
6335 | { |
6336 | BlockDriver *drv = bs->drv; | |
384a48fb | 6337 | IO_CODE(); |
61007b31 | 6338 | if (drv && drv->bdrv_get_specific_info) { |
1bf6e9ca | 6339 | return drv->bdrv_get_specific_info(bs, errp); |
61007b31 SH |
6340 | } |
6341 | return NULL; | |
016f5cf6 AG |
6342 | } |
6343 | ||
d9245599 AN |
6344 | BlockStatsSpecific *bdrv_get_specific_stats(BlockDriverState *bs) |
6345 | { | |
6346 | BlockDriver *drv = bs->drv; | |
384a48fb | 6347 | IO_CODE(); |
d9245599 AN |
6348 | if (!drv || !drv->bdrv_get_specific_stats) { |
6349 | return NULL; | |
6350 | } | |
6351 | return drv->bdrv_get_specific_stats(bs); | |
6352 | } | |
6353 | ||
c834dc05 | 6354 | void coroutine_fn bdrv_co_debug_event(BlockDriverState *bs, BlkdebugEvent event) |
4265d620 | 6355 | { |
384a48fb | 6356 | IO_CODE(); |
c834dc05 | 6357 | if (!bs || !bs->drv || !bs->drv->bdrv_co_debug_event) { |
61007b31 SH |
6358 | return; |
6359 | } | |
4265d620 | 6360 | |
c834dc05 | 6361 | bs->drv->bdrv_co_debug_event(bs, event); |
4265d620 PB |
6362 | } |
6363 | ||
d10529a2 | 6364 | static BlockDriverState *bdrv_find_debug_node(BlockDriverState *bs) |
4265d620 | 6365 | { |
bdb73476 | 6366 | GLOBAL_STATE_CODE(); |
61007b31 | 6367 | while (bs && bs->drv && !bs->drv->bdrv_debug_breakpoint) { |
f706a92f | 6368 | bs = bdrv_primary_bs(bs); |
61007b31 | 6369 | } |
4265d620 | 6370 | |
61007b31 | 6371 | if (bs && bs->drv && bs->drv->bdrv_debug_breakpoint) { |
d10529a2 VSO |
6372 | assert(bs->drv->bdrv_debug_remove_breakpoint); |
6373 | return bs; | |
6374 | } | |
6375 | ||
6376 | return NULL; | |
6377 | } | |
6378 | ||
6379 | int bdrv_debug_breakpoint(BlockDriverState *bs, const char *event, | |
6380 | const char *tag) | |
6381 | { | |
f791bf7f | 6382 | GLOBAL_STATE_CODE(); |
d10529a2 VSO |
6383 | bs = bdrv_find_debug_node(bs); |
6384 | if (bs) { | |
61007b31 SH |
6385 | return bs->drv->bdrv_debug_breakpoint(bs, event, tag); |
6386 | } | |
4265d620 | 6387 | |
61007b31 | 6388 | return -ENOTSUP; |
4265d620 PB |
6389 | } |
6390 | ||
61007b31 | 6391 | int bdrv_debug_remove_breakpoint(BlockDriverState *bs, const char *tag) |
ea2384d3 | 6392 | { |
f791bf7f | 6393 | GLOBAL_STATE_CODE(); |
d10529a2 VSO |
6394 | bs = bdrv_find_debug_node(bs); |
6395 | if (bs) { | |
61007b31 SH |
6396 | return bs->drv->bdrv_debug_remove_breakpoint(bs, tag); |
6397 | } | |
6398 | ||
6399 | return -ENOTSUP; | |
eb852011 MA |
6400 | } |
6401 | ||
61007b31 | 6402 | int bdrv_debug_resume(BlockDriverState *bs, const char *tag) |
ce1a14dc | 6403 | { |
f791bf7f | 6404 | GLOBAL_STATE_CODE(); |
61007b31 | 6405 | while (bs && (!bs->drv || !bs->drv->bdrv_debug_resume)) { |
f706a92f | 6406 | bs = bdrv_primary_bs(bs); |
61007b31 | 6407 | } |
ce1a14dc | 6408 | |
61007b31 SH |
6409 | if (bs && bs->drv && bs->drv->bdrv_debug_resume) { |
6410 | return bs->drv->bdrv_debug_resume(bs, tag); | |
6411 | } | |
ce1a14dc | 6412 | |
61007b31 | 6413 | return -ENOTSUP; |
f197fe2b FZ |
6414 | } |
6415 | ||
61007b31 | 6416 | bool bdrv_debug_is_suspended(BlockDriverState *bs, const char *tag) |
ce1a14dc | 6417 | { |
f791bf7f | 6418 | GLOBAL_STATE_CODE(); |
61007b31 | 6419 | while (bs && bs->drv && !bs->drv->bdrv_debug_is_suspended) { |
f706a92f | 6420 | bs = bdrv_primary_bs(bs); |
f197fe2b | 6421 | } |
19cb3738 | 6422 | |
61007b31 SH |
6423 | if (bs && bs->drv && bs->drv->bdrv_debug_is_suspended) { |
6424 | return bs->drv->bdrv_debug_is_suspended(bs, tag); | |
6425 | } | |
f9f05dc5 | 6426 | |
61007b31 SH |
6427 | return false; |
6428 | } | |
f9f05dc5 | 6429 | |
61007b31 SH |
6430 | /* backing_file can either be relative, or absolute, or a protocol. If it is |
6431 | * relative, it must be relative to the chain. So, passing in bs->filename | |
6432 | * from a BDS as backing_file should not be done, as that may be relative to | |
6433 | * the CWD rather than the chain. */ | |
6434 | BlockDriverState *bdrv_find_backing_image(BlockDriverState *bs, | |
6435 | const char *backing_file) | |
f9f05dc5 | 6436 | { |
61007b31 SH |
6437 | char *filename_full = NULL; |
6438 | char *backing_file_full = NULL; | |
6439 | char *filename_tmp = NULL; | |
6440 | int is_protocol = 0; | |
0b877d09 | 6441 | bool filenames_refreshed = false; |
61007b31 SH |
6442 | BlockDriverState *curr_bs = NULL; |
6443 | BlockDriverState *retval = NULL; | |
dcf3f9b2 | 6444 | BlockDriverState *bs_below; |
f9f05dc5 | 6445 | |
f791bf7f EGE |
6446 | GLOBAL_STATE_CODE(); |
6447 | ||
61007b31 SH |
6448 | if (!bs || !bs->drv || !backing_file) { |
6449 | return NULL; | |
f9f05dc5 KW |
6450 | } |
6451 | ||
61007b31 SH |
6452 | filename_full = g_malloc(PATH_MAX); |
6453 | backing_file_full = g_malloc(PATH_MAX); | |
f9f05dc5 | 6454 | |
61007b31 | 6455 | is_protocol = path_has_protocol(backing_file); |
f9f05dc5 | 6456 | |
dcf3f9b2 HR |
6457 | /* |
6458 | * Being largely a legacy function, skip any filters here | |
6459 | * (because filters do not have normal filenames, so they cannot | |
6460 | * match anyway; and allowing json:{} filenames is a bit out of | |
6461 | * scope). | |
6462 | */ | |
6463 | for (curr_bs = bdrv_skip_filters(bs); | |
6464 | bdrv_cow_child(curr_bs) != NULL; | |
6465 | curr_bs = bs_below) | |
6466 | { | |
6467 | bs_below = bdrv_backing_chain_next(curr_bs); | |
f9f05dc5 | 6468 | |
0b877d09 HR |
6469 | if (bdrv_backing_overridden(curr_bs)) { |
6470 | /* | |
6471 | * If the backing file was overridden, we can only compare | |
6472 | * directly against the backing node's filename. | |
6473 | */ | |
6474 | ||
6475 | if (!filenames_refreshed) { | |
6476 | /* | |
6477 | * This will automatically refresh all of the | |
6478 | * filenames in the rest of the backing chain, so we | |
6479 | * only need to do this once. | |
6480 | */ | |
6481 | bdrv_refresh_filename(bs_below); | |
6482 | filenames_refreshed = true; | |
6483 | } | |
6484 | ||
6485 | if (strcmp(backing_file, bs_below->filename) == 0) { | |
6486 | retval = bs_below; | |
6487 | break; | |
6488 | } | |
6489 | } else if (is_protocol || path_has_protocol(curr_bs->backing_file)) { | |
6490 | /* | |
6491 | * If either of the filename paths is actually a protocol, then | |
6492 | * compare unmodified paths; otherwise make paths relative. | |
6493 | */ | |
6b6833c1 HR |
6494 | char *backing_file_full_ret; |
6495 | ||
61007b31 | 6496 | if (strcmp(backing_file, curr_bs->backing_file) == 0) { |
dcf3f9b2 | 6497 | retval = bs_below; |
61007b31 SH |
6498 | break; |
6499 | } | |
418661e0 | 6500 | /* Also check against the full backing filename for the image */ |
6b6833c1 HR |
6501 | backing_file_full_ret = bdrv_get_full_backing_filename(curr_bs, |
6502 | NULL); | |
6503 | if (backing_file_full_ret) { | |
6504 | bool equal = strcmp(backing_file, backing_file_full_ret) == 0; | |
6505 | g_free(backing_file_full_ret); | |
6506 | if (equal) { | |
dcf3f9b2 | 6507 | retval = bs_below; |
418661e0 JC |
6508 | break; |
6509 | } | |
418661e0 | 6510 | } |
61007b31 SH |
6511 | } else { |
6512 | /* If not an absolute filename path, make it relative to the current | |
6513 | * image's filename path */ | |
2d9158ce HR |
6514 | filename_tmp = bdrv_make_absolute_filename(curr_bs, backing_file, |
6515 | NULL); | |
6516 | /* We are going to compare canonicalized absolute pathnames */ | |
6517 | if (!filename_tmp || !realpath(filename_tmp, filename_full)) { | |
6518 | g_free(filename_tmp); | |
61007b31 SH |
6519 | continue; |
6520 | } | |
2d9158ce | 6521 | g_free(filename_tmp); |
07f07615 | 6522 | |
61007b31 SH |
6523 | /* We need to make sure the backing filename we are comparing against |
6524 | * is relative to the current image filename (or absolute) */ | |
2d9158ce HR |
6525 | filename_tmp = bdrv_get_full_backing_filename(curr_bs, NULL); |
6526 | if (!filename_tmp || !realpath(filename_tmp, backing_file_full)) { | |
6527 | g_free(filename_tmp); | |
61007b31 SH |
6528 | continue; |
6529 | } | |
2d9158ce | 6530 | g_free(filename_tmp); |
eb489bb1 | 6531 | |
61007b31 | 6532 | if (strcmp(backing_file_full, filename_full) == 0) { |
dcf3f9b2 | 6533 | retval = bs_below; |
61007b31 SH |
6534 | break; |
6535 | } | |
6536 | } | |
eb489bb1 KW |
6537 | } |
6538 | ||
61007b31 SH |
6539 | g_free(filename_full); |
6540 | g_free(backing_file_full); | |
61007b31 SH |
6541 | return retval; |
6542 | } | |
6543 | ||
61007b31 SH |
6544 | void bdrv_init(void) |
6545 | { | |
e5f05f8c KW |
6546 | #ifdef CONFIG_BDRV_WHITELIST_TOOLS |
6547 | use_bdrv_whitelist = 1; | |
6548 | #endif | |
61007b31 SH |
6549 | module_call_init(MODULE_INIT_BLOCK); |
6550 | } | |
29cdb251 | 6551 | |
61007b31 SH |
6552 | void bdrv_init_with_whitelist(void) |
6553 | { | |
6554 | use_bdrv_whitelist = 1; | |
6555 | bdrv_init(); | |
07f07615 PB |
6556 | } |
6557 | ||
a94750d9 | 6558 | int bdrv_activate(BlockDriverState *bs, Error **errp) |
0f15423c | 6559 | { |
4417ab7a | 6560 | BdrvChild *child, *parent; |
5a8a30db KW |
6561 | Error *local_err = NULL; |
6562 | int ret; | |
9c98f145 | 6563 | BdrvDirtyBitmap *bm; |
5a8a30db | 6564 | |
f791bf7f EGE |
6565 | GLOBAL_STATE_CODE(); |
6566 | ||
3456a8d1 | 6567 | if (!bs->drv) { |
5416645f | 6568 | return -ENOMEDIUM; |
3456a8d1 KW |
6569 | } |
6570 | ||
16e977d5 | 6571 | QLIST_FOREACH(child, &bs->children, next) { |
11d0c9b3 | 6572 | bdrv_activate(child->bs, &local_err); |
0d1c5c91 | 6573 | if (local_err) { |
0d1c5c91 | 6574 | error_propagate(errp, local_err); |
5416645f | 6575 | return -EINVAL; |
0d1c5c91 | 6576 | } |
5a8a30db | 6577 | } |
0d1c5c91 | 6578 | |
dafe0960 KW |
6579 | /* |
6580 | * Update permissions, they may differ for inactive nodes. | |
6581 | * | |
6582 | * Note that the required permissions of inactive images are always a | |
6583 | * subset of the permissions required after activating the image. This | |
6584 | * allows us to just get the permissions upfront without restricting | |
11d0c9b3 | 6585 | * bdrv_co_invalidate_cache(). |
dafe0960 KW |
6586 | * |
6587 | * It also means that in error cases, we don't have to try and revert to | |
6588 | * the old permissions (which is an operation that could fail, too). We can | |
6589 | * just keep the extended permissions for the next time that an activation | |
6590 | * of the image is tried. | |
6591 | */ | |
7bb4941a KW |
6592 | if (bs->open_flags & BDRV_O_INACTIVE) { |
6593 | bs->open_flags &= ~BDRV_O_INACTIVE; | |
f1316edb | 6594 | ret = bdrv_refresh_perms(bs, NULL, errp); |
7bb4941a | 6595 | if (ret < 0) { |
0d1c5c91 | 6596 | bs->open_flags |= BDRV_O_INACTIVE; |
5416645f | 6597 | return ret; |
0d1c5c91 | 6598 | } |
3456a8d1 | 6599 | |
11d0c9b3 EGE |
6600 | ret = bdrv_invalidate_cache(bs, errp); |
6601 | if (ret < 0) { | |
6602 | bs->open_flags |= BDRV_O_INACTIVE; | |
6603 | return ret; | |
7bb4941a | 6604 | } |
9c98f145 | 6605 | |
7bb4941a KW |
6606 | FOR_EACH_DIRTY_BITMAP(bs, bm) { |
6607 | bdrv_dirty_bitmap_skip_store(bm, false); | |
6608 | } | |
6609 | ||
c057960c | 6610 | ret = bdrv_refresh_total_sectors(bs, bs->total_sectors); |
7bb4941a KW |
6611 | if (ret < 0) { |
6612 | bs->open_flags |= BDRV_O_INACTIVE; | |
6613 | error_setg_errno(errp, -ret, "Could not refresh total sector count"); | |
5416645f | 6614 | return ret; |
7bb4941a | 6615 | } |
5a8a30db | 6616 | } |
4417ab7a KW |
6617 | |
6618 | QLIST_FOREACH(parent, &bs->parents, next_parent) { | |
bd86fb99 HR |
6619 | if (parent->klass->activate) { |
6620 | parent->klass->activate(parent, &local_err); | |
4417ab7a | 6621 | if (local_err) { |
78fc3b3a | 6622 | bs->open_flags |= BDRV_O_INACTIVE; |
4417ab7a | 6623 | error_propagate(errp, local_err); |
5416645f | 6624 | return -EINVAL; |
4417ab7a KW |
6625 | } |
6626 | } | |
6627 | } | |
5416645f VSO |
6628 | |
6629 | return 0; | |
0f15423c AL |
6630 | } |
6631 | ||
11d0c9b3 EGE |
6632 | int coroutine_fn bdrv_co_invalidate_cache(BlockDriverState *bs, Error **errp) |
6633 | { | |
6634 | Error *local_err = NULL; | |
1581a70d | 6635 | IO_CODE(); |
11d0c9b3 EGE |
6636 | |
6637 | assert(!(bs->open_flags & BDRV_O_INACTIVE)); | |
1b3ff9fe | 6638 | assert_bdrv_graph_readable(); |
11d0c9b3 EGE |
6639 | |
6640 | if (bs->drv->bdrv_co_invalidate_cache) { | |
6641 | bs->drv->bdrv_co_invalidate_cache(bs, &local_err); | |
6642 | if (local_err) { | |
6643 | error_propagate(errp, local_err); | |
6644 | return -EINVAL; | |
6645 | } | |
6646 | } | |
6647 | ||
6648 | return 0; | |
6649 | } | |
6650 | ||
3b717194 | 6651 | void bdrv_activate_all(Error **errp) |
0f15423c | 6652 | { |
7c8eece4 | 6653 | BlockDriverState *bs; |
88be7b4b | 6654 | BdrvNextIterator it; |
0f15423c | 6655 | |
f791bf7f EGE |
6656 | GLOBAL_STATE_CODE(); |
6657 | ||
88be7b4b | 6658 | for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) { |
ed78cda3 | 6659 | AioContext *aio_context = bdrv_get_aio_context(bs); |
5416645f | 6660 | int ret; |
ed78cda3 SH |
6661 | |
6662 | aio_context_acquire(aio_context); | |
a94750d9 | 6663 | ret = bdrv_activate(bs, errp); |
ed78cda3 | 6664 | aio_context_release(aio_context); |
5416645f | 6665 | if (ret < 0) { |
5e003f17 | 6666 | bdrv_next_cleanup(&it); |
5a8a30db KW |
6667 | return; |
6668 | } | |
0f15423c AL |
6669 | } |
6670 | } | |
6671 | ||
9e37271f KW |
6672 | static bool bdrv_has_bds_parent(BlockDriverState *bs, bool only_active) |
6673 | { | |
6674 | BdrvChild *parent; | |
bdb73476 | 6675 | GLOBAL_STATE_CODE(); |
9e37271f KW |
6676 | |
6677 | QLIST_FOREACH(parent, &bs->parents, next_parent) { | |
bd86fb99 | 6678 | if (parent->klass->parent_is_bds) { |
9e37271f KW |
6679 | BlockDriverState *parent_bs = parent->opaque; |
6680 | if (!only_active || !(parent_bs->open_flags & BDRV_O_INACTIVE)) { | |
6681 | return true; | |
6682 | } | |
6683 | } | |
6684 | } | |
6685 | ||
6686 | return false; | |
6687 | } | |
6688 | ||
6689 | static int bdrv_inactivate_recurse(BlockDriverState *bs) | |
76b1c7fe | 6690 | { |
cfa1a572 | 6691 | BdrvChild *child, *parent; |
76b1c7fe | 6692 | int ret; |
a13de40a | 6693 | uint64_t cumulative_perms, cumulative_shared_perms; |
76b1c7fe | 6694 | |
da359909 EGE |
6695 | GLOBAL_STATE_CODE(); |
6696 | ||
d470ad42 HR |
6697 | if (!bs->drv) { |
6698 | return -ENOMEDIUM; | |
6699 | } | |
6700 | ||
9e37271f KW |
6701 | /* Make sure that we don't inactivate a child before its parent. |
6702 | * It will be covered by recursion from the yet active parent. */ | |
6703 | if (bdrv_has_bds_parent(bs, true)) { | |
6704 | return 0; | |
6705 | } | |
6706 | ||
6707 | assert(!(bs->open_flags & BDRV_O_INACTIVE)); | |
6708 | ||
6709 | /* Inactivate this node */ | |
6710 | if (bs->drv->bdrv_inactivate) { | |
76b1c7fe KW |
6711 | ret = bs->drv->bdrv_inactivate(bs); |
6712 | if (ret < 0) { | |
6713 | return ret; | |
6714 | } | |
6715 | } | |
6716 | ||
9e37271f | 6717 | QLIST_FOREACH(parent, &bs->parents, next_parent) { |
bd86fb99 HR |
6718 | if (parent->klass->inactivate) { |
6719 | ret = parent->klass->inactivate(parent); | |
9e37271f KW |
6720 | if (ret < 0) { |
6721 | return ret; | |
cfa1a572 KW |
6722 | } |
6723 | } | |
9e37271f | 6724 | } |
9c5e6594 | 6725 | |
a13de40a VSO |
6726 | bdrv_get_cumulative_perm(bs, &cumulative_perms, |
6727 | &cumulative_shared_perms); | |
6728 | if (cumulative_perms & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)) { | |
6729 | /* Our inactive parents still need write access. Inactivation failed. */ | |
6730 | return -EPERM; | |
6731 | } | |
6732 | ||
9e37271f | 6733 | bs->open_flags |= BDRV_O_INACTIVE; |
7d5b5261 | 6734 | |
bb87e4d1 VSO |
6735 | /* |
6736 | * Update permissions, they may differ for inactive nodes. | |
6737 | * We only tried to loosen restrictions, so errors are not fatal, ignore | |
6738 | * them. | |
6739 | */ | |
f1316edb | 6740 | bdrv_refresh_perms(bs, NULL, NULL); |
9e37271f KW |
6741 | |
6742 | /* Recursively inactivate children */ | |
38701b6a | 6743 | QLIST_FOREACH(child, &bs->children, next) { |
9e37271f | 6744 | ret = bdrv_inactivate_recurse(child->bs); |
38701b6a KW |
6745 | if (ret < 0) { |
6746 | return ret; | |
6747 | } | |
6748 | } | |
6749 | ||
76b1c7fe KW |
6750 | return 0; |
6751 | } | |
6752 | ||
6753 | int bdrv_inactivate_all(void) | |
6754 | { | |
79720af6 | 6755 | BlockDriverState *bs = NULL; |
88be7b4b | 6756 | BdrvNextIterator it; |
aad0b7a0 | 6757 | int ret = 0; |
bd6458e4 | 6758 | GSList *aio_ctxs = NULL, *ctx; |
76b1c7fe | 6759 | |
f791bf7f EGE |
6760 | GLOBAL_STATE_CODE(); |
6761 | ||
88be7b4b | 6762 | for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) { |
bd6458e4 PB |
6763 | AioContext *aio_context = bdrv_get_aio_context(bs); |
6764 | ||
6765 | if (!g_slist_find(aio_ctxs, aio_context)) { | |
6766 | aio_ctxs = g_slist_prepend(aio_ctxs, aio_context); | |
6767 | aio_context_acquire(aio_context); | |
6768 | } | |
aad0b7a0 | 6769 | } |
76b1c7fe | 6770 | |
9e37271f KW |
6771 | for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) { |
6772 | /* Nodes with BDS parents are covered by recursion from the last | |
6773 | * parent that gets inactivated. Don't inactivate them a second | |
6774 | * time if that has already happened. */ | |
6775 | if (bdrv_has_bds_parent(bs, false)) { | |
6776 | continue; | |
6777 | } | |
6778 | ret = bdrv_inactivate_recurse(bs); | |
6779 | if (ret < 0) { | |
6780 | bdrv_next_cleanup(&it); | |
6781 | goto out; | |
76b1c7fe KW |
6782 | } |
6783 | } | |
6784 | ||
aad0b7a0 | 6785 | out: |
bd6458e4 PB |
6786 | for (ctx = aio_ctxs; ctx != NULL; ctx = ctx->next) { |
6787 | AioContext *aio_context = ctx->data; | |
6788 | aio_context_release(aio_context); | |
aad0b7a0 | 6789 | } |
bd6458e4 | 6790 | g_slist_free(aio_ctxs); |
aad0b7a0 FZ |
6791 | |
6792 | return ret; | |
76b1c7fe KW |
6793 | } |
6794 | ||
19cb3738 FB |
6795 | /**************************************************************/ |
6796 | /* removable device support */ | |
6797 | ||
6798 | /** | |
6799 | * Return TRUE if the media is present | |
6800 | */ | |
1e97be91 | 6801 | bool coroutine_fn bdrv_co_is_inserted(BlockDriverState *bs) |
19cb3738 FB |
6802 | { |
6803 | BlockDriver *drv = bs->drv; | |
28d7a789 | 6804 | BdrvChild *child; |
384a48fb | 6805 | IO_CODE(); |
a1aff5bf | 6806 | |
e031f750 HR |
6807 | if (!drv) { |
6808 | return false; | |
6809 | } | |
1e97be91 EGE |
6810 | if (drv->bdrv_co_is_inserted) { |
6811 | return drv->bdrv_co_is_inserted(bs); | |
28d7a789 HR |
6812 | } |
6813 | QLIST_FOREACH(child, &bs->children, next) { | |
1e97be91 | 6814 | if (!bdrv_co_is_inserted(child->bs)) { |
28d7a789 HR |
6815 | return false; |
6816 | } | |
e031f750 | 6817 | } |
28d7a789 | 6818 | return true; |
19cb3738 FB |
6819 | } |
6820 | ||
19cb3738 FB |
6821 | /** |
6822 | * If eject_flag is TRUE, eject the media. Otherwise, close the tray | |
6823 | */ | |
2531b390 | 6824 | void coroutine_fn bdrv_co_eject(BlockDriverState *bs, bool eject_flag) |
19cb3738 FB |
6825 | { |
6826 | BlockDriver *drv = bs->drv; | |
384a48fb | 6827 | IO_CODE(); |
19cb3738 | 6828 | |
2531b390 EGE |
6829 | if (drv && drv->bdrv_co_eject) { |
6830 | drv->bdrv_co_eject(bs, eject_flag); | |
19cb3738 FB |
6831 | } |
6832 | } | |
6833 | ||
19cb3738 FB |
6834 | /** |
6835 | * Lock or unlock the media (if it is locked, the user won't be able | |
6836 | * to eject it manually). | |
6837 | */ | |
2c75261c | 6838 | void coroutine_fn bdrv_co_lock_medium(BlockDriverState *bs, bool locked) |
19cb3738 FB |
6839 | { |
6840 | BlockDriver *drv = bs->drv; | |
384a48fb | 6841 | IO_CODE(); |
025e849a | 6842 | trace_bdrv_lock_medium(bs, locked); |
b8c6d095 | 6843 | |
2c75261c EGE |
6844 | if (drv && drv->bdrv_co_lock_medium) { |
6845 | drv->bdrv_co_lock_medium(bs, locked); | |
19cb3738 FB |
6846 | } |
6847 | } | |
985a03b0 | 6848 | |
9fcb0251 FZ |
6849 | /* Get a reference to bs */ |
6850 | void bdrv_ref(BlockDriverState *bs) | |
6851 | { | |
f791bf7f | 6852 | GLOBAL_STATE_CODE(); |
9fcb0251 FZ |
6853 | bs->refcnt++; |
6854 | } | |
6855 | ||
6856 | /* Release a previously grabbed reference to bs. | |
6857 | * If after releasing, reference count is zero, the BlockDriverState is | |
6858 | * deleted. */ | |
6859 | void bdrv_unref(BlockDriverState *bs) | |
6860 | { | |
f791bf7f | 6861 | GLOBAL_STATE_CODE(); |
9a4d5ca6 JC |
6862 | if (!bs) { |
6863 | return; | |
6864 | } | |
9fcb0251 FZ |
6865 | assert(bs->refcnt > 0); |
6866 | if (--bs->refcnt == 0) { | |
6867 | bdrv_delete(bs); | |
6868 | } | |
6869 | } | |
6870 | ||
fbe40ff7 FZ |
6871 | struct BdrvOpBlocker { |
6872 | Error *reason; | |
6873 | QLIST_ENTRY(BdrvOpBlocker) list; | |
6874 | }; | |
6875 | ||
6876 | bool bdrv_op_is_blocked(BlockDriverState *bs, BlockOpType op, Error **errp) | |
6877 | { | |
6878 | BdrvOpBlocker *blocker; | |
f791bf7f | 6879 | GLOBAL_STATE_CODE(); |
fbe40ff7 FZ |
6880 | assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX); |
6881 | if (!QLIST_EMPTY(&bs->op_blockers[op])) { | |
6882 | blocker = QLIST_FIRST(&bs->op_blockers[op]); | |
4b576648 MA |
6883 | error_propagate_prepend(errp, error_copy(blocker->reason), |
6884 | "Node '%s' is busy: ", | |
6885 | bdrv_get_device_or_node_name(bs)); | |
fbe40ff7 FZ |
6886 | return true; |
6887 | } | |
6888 | return false; | |
6889 | } | |
6890 | ||
6891 | void bdrv_op_block(BlockDriverState *bs, BlockOpType op, Error *reason) | |
6892 | { | |
6893 | BdrvOpBlocker *blocker; | |
f791bf7f | 6894 | GLOBAL_STATE_CODE(); |
fbe40ff7 FZ |
6895 | assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX); |
6896 | ||
5839e53b | 6897 | blocker = g_new0(BdrvOpBlocker, 1); |
fbe40ff7 FZ |
6898 | blocker->reason = reason; |
6899 | QLIST_INSERT_HEAD(&bs->op_blockers[op], blocker, list); | |
6900 | } | |
6901 | ||
6902 | void bdrv_op_unblock(BlockDriverState *bs, BlockOpType op, Error *reason) | |
6903 | { | |
6904 | BdrvOpBlocker *blocker, *next; | |
f791bf7f | 6905 | GLOBAL_STATE_CODE(); |
fbe40ff7 FZ |
6906 | assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX); |
6907 | QLIST_FOREACH_SAFE(blocker, &bs->op_blockers[op], list, next) { | |
6908 | if (blocker->reason == reason) { | |
6909 | QLIST_REMOVE(blocker, list); | |
6910 | g_free(blocker); | |
6911 | } | |
6912 | } | |
6913 | } | |
6914 | ||
6915 | void bdrv_op_block_all(BlockDriverState *bs, Error *reason) | |
6916 | { | |
6917 | int i; | |
f791bf7f | 6918 | GLOBAL_STATE_CODE(); |
fbe40ff7 FZ |
6919 | for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) { |
6920 | bdrv_op_block(bs, i, reason); | |
6921 | } | |
6922 | } | |
6923 | ||
6924 | void bdrv_op_unblock_all(BlockDriverState *bs, Error *reason) | |
6925 | { | |
6926 | int i; | |
f791bf7f | 6927 | GLOBAL_STATE_CODE(); |
fbe40ff7 FZ |
6928 | for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) { |
6929 | bdrv_op_unblock(bs, i, reason); | |
6930 | } | |
6931 | } | |
6932 | ||
6933 | bool bdrv_op_blocker_is_empty(BlockDriverState *bs) | |
6934 | { | |
6935 | int i; | |
f791bf7f | 6936 | GLOBAL_STATE_CODE(); |
fbe40ff7 FZ |
6937 | for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) { |
6938 | if (!QLIST_EMPTY(&bs->op_blockers[i])) { | |
6939 | return false; | |
6940 | } | |
6941 | } | |
6942 | return true; | |
6943 | } | |
6944 | ||
e1355055 KW |
6945 | /* |
6946 | * Must not be called while holding the lock of an AioContext other than the | |
6947 | * current one. | |
6948 | */ | |
d92ada22 LC |
6949 | void bdrv_img_create(const char *filename, const char *fmt, |
6950 | const char *base_filename, const char *base_fmt, | |
9217283d FZ |
6951 | char *options, uint64_t img_size, int flags, bool quiet, |
6952 | Error **errp) | |
f88e1a42 | 6953 | { |
83d0521a CL |
6954 | QemuOptsList *create_opts = NULL; |
6955 | QemuOpts *opts = NULL; | |
6956 | const char *backing_fmt, *backing_file; | |
6957 | int64_t size; | |
f88e1a42 | 6958 | BlockDriver *drv, *proto_drv; |
cc84d90f | 6959 | Error *local_err = NULL; |
f88e1a42 JS |
6960 | int ret = 0; |
6961 | ||
f791bf7f EGE |
6962 | GLOBAL_STATE_CODE(); |
6963 | ||
f88e1a42 JS |
6964 | /* Find driver and parse its options */ |
6965 | drv = bdrv_find_format(fmt); | |
6966 | if (!drv) { | |
71c79813 | 6967 | error_setg(errp, "Unknown file format '%s'", fmt); |
d92ada22 | 6968 | return; |
f88e1a42 JS |
6969 | } |
6970 | ||
b65a5e12 | 6971 | proto_drv = bdrv_find_protocol(filename, true, errp); |
f88e1a42 | 6972 | if (!proto_drv) { |
d92ada22 | 6973 | return; |
f88e1a42 JS |
6974 | } |
6975 | ||
c6149724 HR |
6976 | if (!drv->create_opts) { |
6977 | error_setg(errp, "Format driver '%s' does not support image creation", | |
6978 | drv->format_name); | |
6979 | return; | |
6980 | } | |
6981 | ||
5a5e7f8c ML |
6982 | if (!proto_drv->create_opts) { |
6983 | error_setg(errp, "Protocol driver '%s' does not support image creation", | |
6984 | proto_drv->format_name); | |
6985 | return; | |
6986 | } | |
6987 | ||
f6dc1c31 | 6988 | /* Create parameter list */ |
c282e1fd | 6989 | create_opts = qemu_opts_append(create_opts, drv->create_opts); |
5a5e7f8c | 6990 | create_opts = qemu_opts_append(create_opts, proto_drv->create_opts); |
f88e1a42 | 6991 | |
83d0521a | 6992 | opts = qemu_opts_create(create_opts, NULL, 0, &error_abort); |
f88e1a42 JS |
6993 | |
6994 | /* Parse -o options */ | |
6995 | if (options) { | |
a5f9b9df | 6996 | if (!qemu_opts_do_parse(opts, options, NULL, errp)) { |
f88e1a42 JS |
6997 | goto out; |
6998 | } | |
6999 | } | |
7000 | ||
f6dc1c31 KW |
7001 | if (!qemu_opt_get(opts, BLOCK_OPT_SIZE)) { |
7002 | qemu_opt_set_number(opts, BLOCK_OPT_SIZE, img_size, &error_abort); | |
7003 | } else if (img_size != UINT64_C(-1)) { | |
7004 | error_setg(errp, "The image size must be specified only once"); | |
7005 | goto out; | |
7006 | } | |
7007 | ||
f88e1a42 | 7008 | if (base_filename) { |
235e59cf | 7009 | if (!qemu_opt_set(opts, BLOCK_OPT_BACKING_FILE, base_filename, |
3882578b | 7010 | NULL)) { |
71c79813 LC |
7011 | error_setg(errp, "Backing file not supported for file format '%s'", |
7012 | fmt); | |
f88e1a42 JS |
7013 | goto out; |
7014 | } | |
7015 | } | |
7016 | ||
7017 | if (base_fmt) { | |
3882578b | 7018 | if (!qemu_opt_set(opts, BLOCK_OPT_BACKING_FMT, base_fmt, NULL)) { |
71c79813 LC |
7019 | error_setg(errp, "Backing file format not supported for file " |
7020 | "format '%s'", fmt); | |
f88e1a42 JS |
7021 | goto out; |
7022 | } | |
7023 | } | |
7024 | ||
83d0521a CL |
7025 | backing_file = qemu_opt_get(opts, BLOCK_OPT_BACKING_FILE); |
7026 | if (backing_file) { | |
7027 | if (!strcmp(filename, backing_file)) { | |
71c79813 LC |
7028 | error_setg(errp, "Error: Trying to create an image with the " |
7029 | "same filename as the backing file"); | |
792da93a JS |
7030 | goto out; |
7031 | } | |
975a7bd2 CK |
7032 | if (backing_file[0] == '\0') { |
7033 | error_setg(errp, "Expected backing file name, got empty string"); | |
7034 | goto out; | |
7035 | } | |
792da93a JS |
7036 | } |
7037 | ||
83d0521a | 7038 | backing_fmt = qemu_opt_get(opts, BLOCK_OPT_BACKING_FMT); |
f88e1a42 | 7039 | |
6e6e55f5 JS |
7040 | /* The size for the image must always be specified, unless we have a backing |
7041 | * file and we have not been forbidden from opening it. */ | |
a8b42a1c | 7042 | size = qemu_opt_get_size(opts, BLOCK_OPT_SIZE, img_size); |
6e6e55f5 JS |
7043 | if (backing_file && !(flags & BDRV_O_NO_BACKING)) { |
7044 | BlockDriverState *bs; | |
645ae7d8 | 7045 | char *full_backing; |
6e6e55f5 JS |
7046 | int back_flags; |
7047 | QDict *backing_options = NULL; | |
7048 | ||
645ae7d8 HR |
7049 | full_backing = |
7050 | bdrv_get_full_backing_filename_from_filename(filename, backing_file, | |
7051 | &local_err); | |
6e6e55f5 | 7052 | if (local_err) { |
6e6e55f5 JS |
7053 | goto out; |
7054 | } | |
645ae7d8 | 7055 | assert(full_backing); |
29168018 | 7056 | |
d5b23994 HR |
7057 | /* |
7058 | * No need to do I/O here, which allows us to open encrypted | |
7059 | * backing images without needing the secret | |
7060 | */ | |
6e6e55f5 JS |
7061 | back_flags = flags; |
7062 | back_flags &= ~(BDRV_O_RDWR | BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING); | |
d5b23994 | 7063 | back_flags |= BDRV_O_NO_IO; |
f88e1a42 | 7064 | |
cc954f01 | 7065 | backing_options = qdict_new(); |
6e6e55f5 | 7066 | if (backing_fmt) { |
6e6e55f5 JS |
7067 | qdict_put_str(backing_options, "driver", backing_fmt); |
7068 | } | |
cc954f01 | 7069 | qdict_put_bool(backing_options, BDRV_OPT_FORCE_SHARE, true); |
e6641719 | 7070 | |
6e6e55f5 JS |
7071 | bs = bdrv_open(full_backing, NULL, backing_options, back_flags, |
7072 | &local_err); | |
7073 | g_free(full_backing); | |
add8200d EB |
7074 | if (!bs) { |
7075 | error_append_hint(&local_err, "Could not open backing image.\n"); | |
6e6e55f5 JS |
7076 | goto out; |
7077 | } else { | |
d9f059aa | 7078 | if (!backing_fmt) { |
497a30db EB |
7079 | error_setg(&local_err, |
7080 | "Backing file specified without backing format"); | |
7081 | error_append_hint(&local_err, "Detected format of %s.", | |
7082 | bs->drv->format_name); | |
7083 | goto out; | |
d9f059aa | 7084 | } |
6e6e55f5 JS |
7085 | if (size == -1) { |
7086 | /* Opened BS, have no size */ | |
7087 | size = bdrv_getlength(bs); | |
7088 | if (size < 0) { | |
7089 | error_setg_errno(errp, -size, "Could not get size of '%s'", | |
7090 | backing_file); | |
7091 | bdrv_unref(bs); | |
7092 | goto out; | |
7093 | } | |
7094 | qemu_opt_set_number(opts, BLOCK_OPT_SIZE, size, &error_abort); | |
52bf1e72 | 7095 | } |
66f6b814 | 7096 | bdrv_unref(bs); |
f88e1a42 | 7097 | } |
d9f059aa EB |
7098 | /* (backing_file && !(flags & BDRV_O_NO_BACKING)) */ |
7099 | } else if (backing_file && !backing_fmt) { | |
497a30db EB |
7100 | error_setg(&local_err, |
7101 | "Backing file specified without backing format"); | |
7102 | goto out; | |
d9f059aa | 7103 | } |
6e6e55f5 JS |
7104 | |
7105 | if (size == -1) { | |
7106 | error_setg(errp, "Image creation needs a size parameter"); | |
7107 | goto out; | |
f88e1a42 JS |
7108 | } |
7109 | ||
f382d43a | 7110 | if (!quiet) { |
fe646693 | 7111 | printf("Formatting '%s', fmt=%s ", filename, fmt); |
43c5d8f8 | 7112 | qemu_opts_print(opts, " "); |
f382d43a | 7113 | puts(""); |
4e2f4418 | 7114 | fflush(stdout); |
f382d43a | 7115 | } |
83d0521a | 7116 | |
c282e1fd | 7117 | ret = bdrv_create(drv, filename, opts, &local_err); |
83d0521a | 7118 | |
cc84d90f HR |
7119 | if (ret == -EFBIG) { |
7120 | /* This is generally a better message than whatever the driver would | |
7121 | * deliver (especially because of the cluster_size_hint), since that | |
7122 | * is most probably not much different from "image too large". */ | |
7123 | const char *cluster_size_hint = ""; | |
83d0521a | 7124 | if (qemu_opt_get_size(opts, BLOCK_OPT_CLUSTER_SIZE, 0)) { |
cc84d90f | 7125 | cluster_size_hint = " (try using a larger cluster size)"; |
f88e1a42 | 7126 | } |
cc84d90f HR |
7127 | error_setg(errp, "The image size is too large for file format '%s'" |
7128 | "%s", fmt, cluster_size_hint); | |
7129 | error_free(local_err); | |
7130 | local_err = NULL; | |
f88e1a42 JS |
7131 | } |
7132 | ||
7133 | out: | |
83d0521a CL |
7134 | qemu_opts_del(opts); |
7135 | qemu_opts_free(create_opts); | |
621ff94d | 7136 | error_propagate(errp, local_err); |
f88e1a42 | 7137 | } |
85d126f3 SH |
7138 | |
7139 | AioContext *bdrv_get_aio_context(BlockDriverState *bs) | |
7140 | { | |
384a48fb | 7141 | IO_CODE(); |
33f2a757 | 7142 | return bs ? bs->aio_context : qemu_get_aio_context(); |
dcd04228 SH |
7143 | } |
7144 | ||
e336fd4c KW |
7145 | AioContext *coroutine_fn bdrv_co_enter(BlockDriverState *bs) |
7146 | { | |
7147 | Coroutine *self = qemu_coroutine_self(); | |
7148 | AioContext *old_ctx = qemu_coroutine_get_aio_context(self); | |
7149 | AioContext *new_ctx; | |
384a48fb | 7150 | IO_CODE(); |
e336fd4c KW |
7151 | |
7152 | /* | |
7153 | * Increase bs->in_flight to ensure that this operation is completed before | |
7154 | * moving the node to a different AioContext. Read new_ctx only afterwards. | |
7155 | */ | |
7156 | bdrv_inc_in_flight(bs); | |
7157 | ||
7158 | new_ctx = bdrv_get_aio_context(bs); | |
7159 | aio_co_reschedule_self(new_ctx); | |
7160 | return old_ctx; | |
7161 | } | |
7162 | ||
7163 | void coroutine_fn bdrv_co_leave(BlockDriverState *bs, AioContext *old_ctx) | |
7164 | { | |
384a48fb | 7165 | IO_CODE(); |
e336fd4c KW |
7166 | aio_co_reschedule_self(old_ctx); |
7167 | bdrv_dec_in_flight(bs); | |
7168 | } | |
7169 | ||
18c6ac1c KW |
7170 | void coroutine_fn bdrv_co_lock(BlockDriverState *bs) |
7171 | { | |
7172 | AioContext *ctx = bdrv_get_aio_context(bs); | |
7173 | ||
7174 | /* In the main thread, bs->aio_context won't change concurrently */ | |
7175 | assert(qemu_get_current_aio_context() == qemu_get_aio_context()); | |
7176 | ||
7177 | /* | |
7178 | * We're in coroutine context, so we already hold the lock of the main | |
7179 | * loop AioContext. Don't lock it twice to avoid deadlocks. | |
7180 | */ | |
7181 | assert(qemu_in_coroutine()); | |
7182 | if (ctx != qemu_get_aio_context()) { | |
7183 | aio_context_acquire(ctx); | |
7184 | } | |
7185 | } | |
7186 | ||
7187 | void coroutine_fn bdrv_co_unlock(BlockDriverState *bs) | |
7188 | { | |
7189 | AioContext *ctx = bdrv_get_aio_context(bs); | |
7190 | ||
7191 | assert(qemu_in_coroutine()); | |
7192 | if (ctx != qemu_get_aio_context()) { | |
7193 | aio_context_release(ctx); | |
7194 | } | |
7195 | } | |
7196 | ||
e8a095da SH |
7197 | static void bdrv_do_remove_aio_context_notifier(BdrvAioNotifier *ban) |
7198 | { | |
bdb73476 | 7199 | GLOBAL_STATE_CODE(); |
e8a095da SH |
7200 | QLIST_REMOVE(ban, list); |
7201 | g_free(ban); | |
7202 | } | |
7203 | ||
a3a683c3 | 7204 | static void bdrv_detach_aio_context(BlockDriverState *bs) |
dcd04228 | 7205 | { |
e8a095da | 7206 | BdrvAioNotifier *baf, *baf_tmp; |
33384421 | 7207 | |
e8a095da | 7208 | assert(!bs->walking_aio_notifiers); |
da359909 | 7209 | GLOBAL_STATE_CODE(); |
e8a095da SH |
7210 | bs->walking_aio_notifiers = true; |
7211 | QLIST_FOREACH_SAFE(baf, &bs->aio_notifiers, list, baf_tmp) { | |
7212 | if (baf->deleted) { | |
7213 | bdrv_do_remove_aio_context_notifier(baf); | |
7214 | } else { | |
7215 | baf->detach_aio_context(baf->opaque); | |
7216 | } | |
33384421 | 7217 | } |
e8a095da SH |
7218 | /* Never mind iterating again to check for ->deleted. bdrv_close() will |
7219 | * remove remaining aio notifiers if we aren't called again. | |
7220 | */ | |
7221 | bs->walking_aio_notifiers = false; | |
33384421 | 7222 | |
1bffe1ae | 7223 | if (bs->drv && bs->drv->bdrv_detach_aio_context) { |
dcd04228 SH |
7224 | bs->drv->bdrv_detach_aio_context(bs); |
7225 | } | |
dcd04228 | 7226 | |
e64f25f3 KW |
7227 | if (bs->quiesce_counter) { |
7228 | aio_enable_external(bs->aio_context); | |
7229 | } | |
dcd04228 SH |
7230 | bs->aio_context = NULL; |
7231 | } | |
7232 | ||
a3a683c3 KW |
7233 | static void bdrv_attach_aio_context(BlockDriverState *bs, |
7234 | AioContext *new_context) | |
dcd04228 | 7235 | { |
e8a095da | 7236 | BdrvAioNotifier *ban, *ban_tmp; |
da359909 | 7237 | GLOBAL_STATE_CODE(); |
33384421 | 7238 | |
e64f25f3 KW |
7239 | if (bs->quiesce_counter) { |
7240 | aio_disable_external(new_context); | |
7241 | } | |
7242 | ||
dcd04228 SH |
7243 | bs->aio_context = new_context; |
7244 | ||
1bffe1ae | 7245 | if (bs->drv && bs->drv->bdrv_attach_aio_context) { |
dcd04228 SH |
7246 | bs->drv->bdrv_attach_aio_context(bs, new_context); |
7247 | } | |
33384421 | 7248 | |
e8a095da SH |
7249 | assert(!bs->walking_aio_notifiers); |
7250 | bs->walking_aio_notifiers = true; | |
7251 | QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_tmp) { | |
7252 | if (ban->deleted) { | |
7253 | bdrv_do_remove_aio_context_notifier(ban); | |
7254 | } else { | |
7255 | ban->attached_aio_context(new_context, ban->opaque); | |
7256 | } | |
33384421 | 7257 | } |
e8a095da | 7258 | bs->walking_aio_notifiers = false; |
dcd04228 SH |
7259 | } |
7260 | ||
7e8c182f EGE |
7261 | typedef struct BdrvStateSetAioContext { |
7262 | AioContext *new_ctx; | |
7263 | BlockDriverState *bs; | |
7264 | } BdrvStateSetAioContext; | |
d616b224 | 7265 | |
7e8c182f | 7266 | static bool bdrv_parent_change_aio_context(BdrvChild *c, AioContext *ctx, |
e08cc001 EGE |
7267 | GHashTable *visited, |
7268 | Transaction *tran, | |
7e8c182f | 7269 | Error **errp) |
5d231849 | 7270 | { |
f0c28327 | 7271 | GLOBAL_STATE_CODE(); |
e08cc001 | 7272 | if (g_hash_table_contains(visited, c)) { |
5d231849 KW |
7273 | return true; |
7274 | } | |
e08cc001 | 7275 | g_hash_table_add(visited, c); |
5d231849 | 7276 | |
bd86fb99 HR |
7277 | /* |
7278 | * A BdrvChildClass that doesn't handle AioContext changes cannot | |
7279 | * tolerate any AioContext changes | |
7280 | */ | |
7e8c182f | 7281 | if (!c->klass->change_aio_ctx) { |
5d231849 KW |
7282 | char *user = bdrv_child_user_desc(c); |
7283 | error_setg(errp, "Changing iothreads is not supported by %s", user); | |
7284 | g_free(user); | |
7285 | return false; | |
7286 | } | |
7e8c182f | 7287 | if (!c->klass->change_aio_ctx(c, ctx, visited, tran, errp)) { |
5d231849 KW |
7288 | assert(!errp || *errp); |
7289 | return false; | |
7290 | } | |
7291 | return true; | |
7292 | } | |
7293 | ||
7e8c182f | 7294 | bool bdrv_child_change_aio_context(BdrvChild *c, AioContext *ctx, |
e08cc001 | 7295 | GHashTable *visited, Transaction *tran, |
7e8c182f | 7296 | Error **errp) |
5d231849 | 7297 | { |
f791bf7f | 7298 | GLOBAL_STATE_CODE(); |
e08cc001 | 7299 | if (g_hash_table_contains(visited, c)) { |
5d231849 KW |
7300 | return true; |
7301 | } | |
e08cc001 | 7302 | g_hash_table_add(visited, c); |
7e8c182f | 7303 | return bdrv_change_aio_context(c->bs, ctx, visited, tran, errp); |
5d231849 KW |
7304 | } |
7305 | ||
7e8c182f EGE |
7306 | static void bdrv_set_aio_context_clean(void *opaque) |
7307 | { | |
7308 | BdrvStateSetAioContext *state = (BdrvStateSetAioContext *) opaque; | |
7309 | BlockDriverState *bs = (BlockDriverState *) state->bs; | |
7310 | ||
7311 | /* Paired with bdrv_drained_begin in bdrv_change_aio_context() */ | |
7312 | bdrv_drained_end(bs); | |
7313 | ||
7314 | g_free(state); | |
7315 | } | |
7316 | ||
7317 | static void bdrv_set_aio_context_commit(void *opaque) | |
7318 | { | |
7319 | BdrvStateSetAioContext *state = (BdrvStateSetAioContext *) opaque; | |
7320 | BlockDriverState *bs = (BlockDriverState *) state->bs; | |
7321 | AioContext *new_context = state->new_ctx; | |
7322 | AioContext *old_context = bdrv_get_aio_context(bs); | |
7e8c182f EGE |
7323 | |
7324 | /* | |
7325 | * Take the old AioContex when detaching it from bs. | |
7326 | * At this point, new_context lock is already acquired, and we are now | |
7327 | * also taking old_context. This is safe as long as bdrv_detach_aio_context | |
7328 | * does not call AIO_POLL_WHILE(). | |
7329 | */ | |
7330 | if (old_context != qemu_get_aio_context()) { | |
7331 | aio_context_acquire(old_context); | |
7332 | } | |
7333 | bdrv_detach_aio_context(bs); | |
7334 | if (old_context != qemu_get_aio_context()) { | |
7335 | aio_context_release(old_context); | |
7336 | } | |
7337 | bdrv_attach_aio_context(bs, new_context); | |
7338 | } | |
7339 | ||
7340 | static TransactionActionDrv set_aio_context = { | |
7341 | .commit = bdrv_set_aio_context_commit, | |
7342 | .clean = bdrv_set_aio_context_clean, | |
7343 | }; | |
7344 | ||
7345 | /* | |
7346 | * Changes the AioContext used for fd handlers, timers, and BHs by this | |
7347 | * BlockDriverState and all its children and parents. | |
7348 | * | |
7349 | * Must be called from the main AioContext. | |
7350 | * | |
7351 | * The caller must own the AioContext lock for the old AioContext of bs, but it | |
7352 | * must not own the AioContext lock for new_context (unless new_context is the | |
7353 | * same as the current context of bs). | |
7354 | * | |
7355 | * @visited will accumulate all visited BdrvChild objects. The caller is | |
7356 | * responsible for freeing the list afterwards. | |
7357 | */ | |
7358 | static bool bdrv_change_aio_context(BlockDriverState *bs, AioContext *ctx, | |
e08cc001 | 7359 | GHashTable *visited, Transaction *tran, |
7e8c182f | 7360 | Error **errp) |
5d231849 KW |
7361 | { |
7362 | BdrvChild *c; | |
7e8c182f EGE |
7363 | BdrvStateSetAioContext *state; |
7364 | ||
7365 | GLOBAL_STATE_CODE(); | |
5d231849 KW |
7366 | |
7367 | if (bdrv_get_aio_context(bs) == ctx) { | |
7368 | return true; | |
7369 | } | |
7370 | ||
7371 | QLIST_FOREACH(c, &bs->parents, next_parent) { | |
7e8c182f | 7372 | if (!bdrv_parent_change_aio_context(c, ctx, visited, tran, errp)) { |
5d231849 KW |
7373 | return false; |
7374 | } | |
7375 | } | |
7e8c182f | 7376 | |
5d231849 | 7377 | QLIST_FOREACH(c, &bs->children, next) { |
7e8c182f | 7378 | if (!bdrv_child_change_aio_context(c, ctx, visited, tran, errp)) { |
5d231849 KW |
7379 | return false; |
7380 | } | |
7381 | } | |
7382 | ||
7e8c182f EGE |
7383 | state = g_new(BdrvStateSetAioContext, 1); |
7384 | *state = (BdrvStateSetAioContext) { | |
7385 | .new_ctx = ctx, | |
7386 | .bs = bs, | |
7387 | }; | |
7388 | ||
7389 | /* Paired with bdrv_drained_end in bdrv_set_aio_context_clean() */ | |
7390 | bdrv_drained_begin(bs); | |
7391 | ||
7392 | tran_add(tran, &set_aio_context, state); | |
7393 | ||
5d231849 KW |
7394 | return true; |
7395 | } | |
7396 | ||
7e8c182f EGE |
7397 | /* |
7398 | * Change bs's and recursively all of its parents' and children's AioContext | |
7399 | * to the given new context, returning an error if that isn't possible. | |
7400 | * | |
7401 | * If ignore_child is not NULL, that child (and its subgraph) will not | |
7402 | * be touched. | |
7403 | * | |
7404 | * This function still requires the caller to take the bs current | |
7405 | * AioContext lock, otherwise draining will fail since AIO_WAIT_WHILE | |
7406 | * assumes the lock is always held if bs is in another AioContext. | |
7407 | * For the same reason, it temporarily also holds the new AioContext, since | |
7408 | * bdrv_drained_end calls BDRV_POLL_WHILE that assumes the lock is taken too. | |
7409 | * Therefore the new AioContext lock must not be taken by the caller. | |
7410 | */ | |
a41cfda1 EGE |
7411 | int bdrv_try_change_aio_context(BlockDriverState *bs, AioContext *ctx, |
7412 | BdrvChild *ignore_child, Error **errp) | |
5d231849 | 7413 | { |
7e8c182f | 7414 | Transaction *tran; |
e08cc001 | 7415 | GHashTable *visited; |
7e8c182f EGE |
7416 | int ret; |
7417 | AioContext *old_context = bdrv_get_aio_context(bs); | |
f791bf7f EGE |
7418 | GLOBAL_STATE_CODE(); |
7419 | ||
7e8c182f EGE |
7420 | /* |
7421 | * Recursion phase: go through all nodes of the graph. | |
7422 | * Take care of checking that all nodes support changing AioContext | |
7423 | * and drain them, builing a linear list of callbacks to run if everything | |
7424 | * is successful (the transaction itself). | |
7425 | */ | |
7426 | tran = tran_new(); | |
e08cc001 EGE |
7427 | visited = g_hash_table_new(NULL, NULL); |
7428 | if (ignore_child) { | |
7429 | g_hash_table_add(visited, ignore_child); | |
7430 | } | |
7431 | ret = bdrv_change_aio_context(bs, ctx, visited, tran, errp); | |
7432 | g_hash_table_destroy(visited); | |
7e8c182f EGE |
7433 | |
7434 | /* | |
7435 | * Linear phase: go through all callbacks collected in the transaction. | |
7436 | * Run all callbacks collected in the recursion to switch all nodes | |
7437 | * AioContext lock (transaction commit), or undo all changes done in the | |
7438 | * recursion (transaction abort). | |
7439 | */ | |
5d231849 KW |
7440 | |
7441 | if (!ret) { | |
7e8c182f EGE |
7442 | /* Just run clean() callbacks. No AioContext changed. */ |
7443 | tran_abort(tran); | |
5d231849 KW |
7444 | return -EPERM; |
7445 | } | |
7446 | ||
7e8c182f EGE |
7447 | /* |
7448 | * Release old AioContext, it won't be needed anymore, as all | |
7449 | * bdrv_drained_begin() have been called already. | |
7450 | */ | |
7451 | if (qemu_get_aio_context() != old_context) { | |
7452 | aio_context_release(old_context); | |
7453 | } | |
7454 | ||
7455 | /* | |
7456 | * Acquire new AioContext since bdrv_drained_end() is going to be called | |
7457 | * after we switched all nodes in the new AioContext, and the function | |
7458 | * assumes that the lock of the bs is always taken. | |
7459 | */ | |
7460 | if (qemu_get_aio_context() != ctx) { | |
7461 | aio_context_acquire(ctx); | |
7462 | } | |
53a7d041 | 7463 | |
7e8c182f | 7464 | tran_commit(tran); |
5d231849 | 7465 | |
7e8c182f EGE |
7466 | if (qemu_get_aio_context() != ctx) { |
7467 | aio_context_release(ctx); | |
7468 | } | |
7469 | ||
7470 | /* Re-acquire the old AioContext, since the caller takes and releases it. */ | |
7471 | if (qemu_get_aio_context() != old_context) { | |
7472 | aio_context_acquire(old_context); | |
7473 | } | |
7474 | ||
7475 | return 0; | |
5d231849 KW |
7476 | } |
7477 | ||
33384421 HR |
7478 | void bdrv_add_aio_context_notifier(BlockDriverState *bs, |
7479 | void (*attached_aio_context)(AioContext *new_context, void *opaque), | |
7480 | void (*detach_aio_context)(void *opaque), void *opaque) | |
7481 | { | |
7482 | BdrvAioNotifier *ban = g_new(BdrvAioNotifier, 1); | |
7483 | *ban = (BdrvAioNotifier){ | |
7484 | .attached_aio_context = attached_aio_context, | |
7485 | .detach_aio_context = detach_aio_context, | |
7486 | .opaque = opaque | |
7487 | }; | |
f791bf7f | 7488 | GLOBAL_STATE_CODE(); |
33384421 HR |
7489 | |
7490 | QLIST_INSERT_HEAD(&bs->aio_notifiers, ban, list); | |
7491 | } | |
7492 | ||
7493 | void bdrv_remove_aio_context_notifier(BlockDriverState *bs, | |
7494 | void (*attached_aio_context)(AioContext *, | |
7495 | void *), | |
7496 | void (*detach_aio_context)(void *), | |
7497 | void *opaque) | |
7498 | { | |
7499 | BdrvAioNotifier *ban, *ban_next; | |
f791bf7f | 7500 | GLOBAL_STATE_CODE(); |
33384421 HR |
7501 | |
7502 | QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_next) { | |
7503 | if (ban->attached_aio_context == attached_aio_context && | |
7504 | ban->detach_aio_context == detach_aio_context && | |
e8a095da SH |
7505 | ban->opaque == opaque && |
7506 | ban->deleted == false) | |
33384421 | 7507 | { |
e8a095da SH |
7508 | if (bs->walking_aio_notifiers) { |
7509 | ban->deleted = true; | |
7510 | } else { | |
7511 | bdrv_do_remove_aio_context_notifier(ban); | |
7512 | } | |
33384421 HR |
7513 | return; |
7514 | } | |
7515 | } | |
7516 | ||
7517 | abort(); | |
7518 | } | |
7519 | ||
77485434 | 7520 | int bdrv_amend_options(BlockDriverState *bs, QemuOpts *opts, |
d1402b50 | 7521 | BlockDriverAmendStatusCB *status_cb, void *cb_opaque, |
a3579bfa | 7522 | bool force, |
d1402b50 | 7523 | Error **errp) |
6f176b48 | 7524 | { |
f791bf7f | 7525 | GLOBAL_STATE_CODE(); |
d470ad42 | 7526 | if (!bs->drv) { |
d1402b50 | 7527 | error_setg(errp, "Node is ejected"); |
d470ad42 HR |
7528 | return -ENOMEDIUM; |
7529 | } | |
c282e1fd | 7530 | if (!bs->drv->bdrv_amend_options) { |
d1402b50 HR |
7531 | error_setg(errp, "Block driver '%s' does not support option amendment", |
7532 | bs->drv->format_name); | |
6f176b48 HR |
7533 | return -ENOTSUP; |
7534 | } | |
a3579bfa ML |
7535 | return bs->drv->bdrv_amend_options(bs, opts, status_cb, |
7536 | cb_opaque, force, errp); | |
6f176b48 | 7537 | } |
f6186f49 | 7538 | |
5d69b5ab HR |
7539 | /* |
7540 | * This function checks whether the given @to_replace is allowed to be | |
7541 | * replaced by a node that always shows the same data as @bs. This is | |
7542 | * used for example to verify whether the mirror job can replace | |
7543 | * @to_replace by the target mirrored from @bs. | |
7544 | * To be replaceable, @bs and @to_replace may either be guaranteed to | |
7545 | * always show the same data (because they are only connected through | |
7546 | * filters), or some driver may allow replacing one of its children | |
7547 | * because it can guarantee that this child's data is not visible at | |
7548 | * all (for example, for dissenting quorum children that have no other | |
7549 | * parents). | |
7550 | */ | |
7551 | bool bdrv_recurse_can_replace(BlockDriverState *bs, | |
7552 | BlockDriverState *to_replace) | |
7553 | { | |
93393e69 HR |
7554 | BlockDriverState *filtered; |
7555 | ||
b4ad82aa EGE |
7556 | GLOBAL_STATE_CODE(); |
7557 | ||
5d69b5ab HR |
7558 | if (!bs || !bs->drv) { |
7559 | return false; | |
7560 | } | |
7561 | ||
7562 | if (bs == to_replace) { | |
7563 | return true; | |
7564 | } | |
7565 | ||
7566 | /* See what the driver can do */ | |
7567 | if (bs->drv->bdrv_recurse_can_replace) { | |
7568 | return bs->drv->bdrv_recurse_can_replace(bs, to_replace); | |
7569 | } | |
7570 | ||
7571 | /* For filters without an own implementation, we can recurse on our own */ | |
93393e69 HR |
7572 | filtered = bdrv_filter_bs(bs); |
7573 | if (filtered) { | |
7574 | return bdrv_recurse_can_replace(filtered, to_replace); | |
5d69b5ab HR |
7575 | } |
7576 | ||
7577 | /* Safe default */ | |
7578 | return false; | |
7579 | } | |
7580 | ||
810803a8 HR |
7581 | /* |
7582 | * Check whether the given @node_name can be replaced by a node that | |
7583 | * has the same data as @parent_bs. If so, return @node_name's BDS; | |
7584 | * NULL otherwise. | |
7585 | * | |
7586 | * @node_name must be a (recursive) *child of @parent_bs (or this | |
7587 | * function will return NULL). | |
7588 | * | |
7589 | * The result (whether the node can be replaced or not) is only valid | |
7590 | * for as long as no graph or permission changes occur. | |
7591 | */ | |
e12f3784 WC |
7592 | BlockDriverState *check_to_replace_node(BlockDriverState *parent_bs, |
7593 | const char *node_name, Error **errp) | |
09158f00 BC |
7594 | { |
7595 | BlockDriverState *to_replace_bs = bdrv_find_node(node_name); | |
5a7e7a0b SH |
7596 | AioContext *aio_context; |
7597 | ||
f791bf7f EGE |
7598 | GLOBAL_STATE_CODE(); |
7599 | ||
09158f00 | 7600 | if (!to_replace_bs) { |
785ec4b1 | 7601 | error_setg(errp, "Failed to find node with node-name='%s'", node_name); |
09158f00 BC |
7602 | return NULL; |
7603 | } | |
7604 | ||
5a7e7a0b SH |
7605 | aio_context = bdrv_get_aio_context(to_replace_bs); |
7606 | aio_context_acquire(aio_context); | |
7607 | ||
09158f00 | 7608 | if (bdrv_op_is_blocked(to_replace_bs, BLOCK_OP_TYPE_REPLACE, errp)) { |
5a7e7a0b SH |
7609 | to_replace_bs = NULL; |
7610 | goto out; | |
09158f00 BC |
7611 | } |
7612 | ||
7613 | /* We don't want arbitrary node of the BDS chain to be replaced only the top | |
7614 | * most non filter in order to prevent data corruption. | |
7615 | * Another benefit is that this tests exclude backing files which are | |
7616 | * blocked by the backing blockers. | |
7617 | */ | |
810803a8 HR |
7618 | if (!bdrv_recurse_can_replace(parent_bs, to_replace_bs)) { |
7619 | error_setg(errp, "Cannot replace '%s' by a node mirrored from '%s', " | |
7620 | "because it cannot be guaranteed that doing so would not " | |
7621 | "lead to an abrupt change of visible data", | |
7622 | node_name, parent_bs->node_name); | |
5a7e7a0b SH |
7623 | to_replace_bs = NULL; |
7624 | goto out; | |
09158f00 BC |
7625 | } |
7626 | ||
5a7e7a0b SH |
7627 | out: |
7628 | aio_context_release(aio_context); | |
09158f00 BC |
7629 | return to_replace_bs; |
7630 | } | |
448ad91d | 7631 | |
97e2f021 HR |
7632 | /** |
7633 | * Iterates through the list of runtime option keys that are said to | |
7634 | * be "strong" for a BDS. An option is called "strong" if it changes | |
7635 | * a BDS's data. For example, the null block driver's "size" and | |
7636 | * "read-zeroes" options are strong, but its "latency-ns" option is | |
7637 | * not. | |
7638 | * | |
7639 | * If a key returned by this function ends with a dot, all options | |
7640 | * starting with that prefix are strong. | |
7641 | */ | |
7642 | static const char *const *strong_options(BlockDriverState *bs, | |
7643 | const char *const *curopt) | |
7644 | { | |
7645 | static const char *const global_options[] = { | |
7646 | "driver", "filename", NULL | |
7647 | }; | |
7648 | ||
7649 | if (!curopt) { | |
7650 | return &global_options[0]; | |
7651 | } | |
7652 | ||
7653 | curopt++; | |
7654 | if (curopt == &global_options[ARRAY_SIZE(global_options) - 1] && bs->drv) { | |
7655 | curopt = bs->drv->strong_runtime_opts; | |
7656 | } | |
7657 | ||
7658 | return (curopt && *curopt) ? curopt : NULL; | |
7659 | } | |
7660 | ||
7661 | /** | |
7662 | * Copies all strong runtime options from bs->options to the given | |
7663 | * QDict. The set of strong option keys is determined by invoking | |
7664 | * strong_options(). | |
7665 | * | |
7666 | * Returns true iff any strong option was present in bs->options (and | |
7667 | * thus copied to the target QDict) with the exception of "filename" | |
7668 | * and "driver". The caller is expected to use this value to decide | |
7669 | * whether the existence of strong options prevents the generation of | |
7670 | * a plain filename. | |
7671 | */ | |
7672 | static bool append_strong_runtime_options(QDict *d, BlockDriverState *bs) | |
7673 | { | |
7674 | bool found_any = false; | |
7675 | const char *const *option_name = NULL; | |
7676 | ||
7677 | if (!bs->drv) { | |
7678 | return false; | |
7679 | } | |
7680 | ||
7681 | while ((option_name = strong_options(bs, option_name))) { | |
7682 | bool option_given = false; | |
7683 | ||
7684 | assert(strlen(*option_name) > 0); | |
7685 | if ((*option_name)[strlen(*option_name) - 1] != '.') { | |
7686 | QObject *entry = qdict_get(bs->options, *option_name); | |
7687 | if (!entry) { | |
7688 | continue; | |
7689 | } | |
7690 | ||
7691 | qdict_put_obj(d, *option_name, qobject_ref(entry)); | |
7692 | option_given = true; | |
7693 | } else { | |
7694 | const QDictEntry *entry; | |
7695 | for (entry = qdict_first(bs->options); entry; | |
7696 | entry = qdict_next(bs->options, entry)) | |
7697 | { | |
7698 | if (strstart(qdict_entry_key(entry), *option_name, NULL)) { | |
7699 | qdict_put_obj(d, qdict_entry_key(entry), | |
7700 | qobject_ref(qdict_entry_value(entry))); | |
7701 | option_given = true; | |
7702 | } | |
7703 | } | |
7704 | } | |
7705 | ||
7706 | /* While "driver" and "filename" need to be included in a JSON filename, | |
7707 | * their existence does not prohibit generation of a plain filename. */ | |
7708 | if (!found_any && option_given && | |
7709 | strcmp(*option_name, "driver") && strcmp(*option_name, "filename")) | |
7710 | { | |
7711 | found_any = true; | |
7712 | } | |
7713 | } | |
7714 | ||
62a01a27 HR |
7715 | if (!qdict_haskey(d, "driver")) { |
7716 | /* Drivers created with bdrv_new_open_driver() may not have a | |
7717 | * @driver option. Add it here. */ | |
7718 | qdict_put_str(d, "driver", bs->drv->format_name); | |
7719 | } | |
7720 | ||
97e2f021 HR |
7721 | return found_any; |
7722 | } | |
7723 | ||
90993623 HR |
7724 | /* Note: This function may return false positives; it may return true |
7725 | * even if opening the backing file specified by bs's image header | |
7726 | * would result in exactly bs->backing. */ | |
fa8fc1d0 | 7727 | static bool bdrv_backing_overridden(BlockDriverState *bs) |
90993623 | 7728 | { |
b4ad82aa | 7729 | GLOBAL_STATE_CODE(); |
90993623 HR |
7730 | if (bs->backing) { |
7731 | return strcmp(bs->auto_backing_file, | |
7732 | bs->backing->bs->filename); | |
7733 | } else { | |
7734 | /* No backing BDS, so if the image header reports any backing | |
7735 | * file, it must have been suppressed */ | |
7736 | return bs->auto_backing_file[0] != '\0'; | |
7737 | } | |
7738 | } | |
7739 | ||
91af7014 HR |
7740 | /* Updates the following BDS fields: |
7741 | * - exact_filename: A filename which may be used for opening a block device | |
7742 | * which (mostly) equals the given BDS (even without any | |
7743 | * other options; so reading and writing must return the same | |
7744 | * results, but caching etc. may be different) | |
7745 | * - full_open_options: Options which, when given when opening a block device | |
7746 | * (without a filename), result in a BDS (mostly) | |
7747 | * equalling the given one | |
7748 | * - filename: If exact_filename is set, it is copied here. Otherwise, | |
7749 | * full_open_options is converted to a JSON object, prefixed with | |
7750 | * "json:" (for use through the JSON pseudo protocol) and put here. | |
7751 | */ | |
7752 | void bdrv_refresh_filename(BlockDriverState *bs) | |
7753 | { | |
7754 | BlockDriver *drv = bs->drv; | |
e24518e3 | 7755 | BdrvChild *child; |
52f72d6f | 7756 | BlockDriverState *primary_child_bs; |
91af7014 | 7757 | QDict *opts; |
90993623 | 7758 | bool backing_overridden; |
998b3a1e HR |
7759 | bool generate_json_filename; /* Whether our default implementation should |
7760 | fill exact_filename (false) or not (true) */ | |
91af7014 | 7761 | |
f791bf7f EGE |
7762 | GLOBAL_STATE_CODE(); |
7763 | ||
91af7014 HR |
7764 | if (!drv) { |
7765 | return; | |
7766 | } | |
7767 | ||
e24518e3 HR |
7768 | /* This BDS's file name may depend on any of its children's file names, so |
7769 | * refresh those first */ | |
7770 | QLIST_FOREACH(child, &bs->children, next) { | |
7771 | bdrv_refresh_filename(child->bs); | |
91af7014 HR |
7772 | } |
7773 | ||
bb808d5f HR |
7774 | if (bs->implicit) { |
7775 | /* For implicit nodes, just copy everything from the single child */ | |
7776 | child = QLIST_FIRST(&bs->children); | |
7777 | assert(QLIST_NEXT(child, next) == NULL); | |
7778 | ||
7779 | pstrcpy(bs->exact_filename, sizeof(bs->exact_filename), | |
7780 | child->bs->exact_filename); | |
7781 | pstrcpy(bs->filename, sizeof(bs->filename), child->bs->filename); | |
7782 | ||
cb895614 | 7783 | qobject_unref(bs->full_open_options); |
bb808d5f HR |
7784 | bs->full_open_options = qobject_ref(child->bs->full_open_options); |
7785 | ||
7786 | return; | |
7787 | } | |
7788 | ||
90993623 HR |
7789 | backing_overridden = bdrv_backing_overridden(bs); |
7790 | ||
7791 | if (bs->open_flags & BDRV_O_NO_IO) { | |
7792 | /* Without I/O, the backing file does not change anything. | |
7793 | * Therefore, in such a case (primarily qemu-img), we can | |
7794 | * pretend the backing file has not been overridden even if | |
7795 | * it technically has been. */ | |
7796 | backing_overridden = false; | |
7797 | } | |
7798 | ||
97e2f021 HR |
7799 | /* Gather the options QDict */ |
7800 | opts = qdict_new(); | |
998b3a1e HR |
7801 | generate_json_filename = append_strong_runtime_options(opts, bs); |
7802 | generate_json_filename |= backing_overridden; | |
97e2f021 HR |
7803 | |
7804 | if (drv->bdrv_gather_child_options) { | |
7805 | /* Some block drivers may not want to present all of their children's | |
7806 | * options, or name them differently from BdrvChild.name */ | |
7807 | drv->bdrv_gather_child_options(bs, opts, backing_overridden); | |
7808 | } else { | |
7809 | QLIST_FOREACH(child, &bs->children, next) { | |
25191e5f | 7810 | if (child == bs->backing && !backing_overridden) { |
97e2f021 HR |
7811 | /* We can skip the backing BDS if it has not been overridden */ |
7812 | continue; | |
7813 | } | |
7814 | ||
7815 | qdict_put(opts, child->name, | |
7816 | qobject_ref(child->bs->full_open_options)); | |
7817 | } | |
7818 | ||
7819 | if (backing_overridden && !bs->backing) { | |
7820 | /* Force no backing file */ | |
7821 | qdict_put_null(opts, "backing"); | |
7822 | } | |
7823 | } | |
7824 | ||
7825 | qobject_unref(bs->full_open_options); | |
7826 | bs->full_open_options = opts; | |
7827 | ||
52f72d6f HR |
7828 | primary_child_bs = bdrv_primary_bs(bs); |
7829 | ||
998b3a1e HR |
7830 | if (drv->bdrv_refresh_filename) { |
7831 | /* Obsolete information is of no use here, so drop the old file name | |
7832 | * information before refreshing it */ | |
7833 | bs->exact_filename[0] = '\0'; | |
7834 | ||
7835 | drv->bdrv_refresh_filename(bs); | |
52f72d6f HR |
7836 | } else if (primary_child_bs) { |
7837 | /* | |
7838 | * Try to reconstruct valid information from the underlying | |
7839 | * file -- this only works for format nodes (filter nodes | |
7840 | * cannot be probed and as such must be selected by the user | |
7841 | * either through an options dict, or through a special | |
7842 | * filename which the filter driver must construct in its | |
7843 | * .bdrv_refresh_filename() implementation). | |
7844 | */ | |
998b3a1e HR |
7845 | |
7846 | bs->exact_filename[0] = '\0'; | |
7847 | ||
fb695c74 HR |
7848 | /* |
7849 | * We can use the underlying file's filename if: | |
7850 | * - it has a filename, | |
52f72d6f | 7851 | * - the current BDS is not a filter, |
fb695c74 HR |
7852 | * - the file is a protocol BDS, and |
7853 | * - opening that file (as this BDS's format) will automatically create | |
7854 | * the BDS tree we have right now, that is: | |
7855 | * - the user did not significantly change this BDS's behavior with | |
7856 | * some explicit (strong) options | |
7857 | * - no non-file child of this BDS has been overridden by the user | |
7858 | * Both of these conditions are represented by generate_json_filename. | |
7859 | */ | |
52f72d6f HR |
7860 | if (primary_child_bs->exact_filename[0] && |
7861 | primary_child_bs->drv->bdrv_file_open && | |
7862 | !drv->is_filter && !generate_json_filename) | |
fb695c74 | 7863 | { |
52f72d6f | 7864 | strcpy(bs->exact_filename, primary_child_bs->exact_filename); |
998b3a1e HR |
7865 | } |
7866 | } | |
7867 | ||
91af7014 HR |
7868 | if (bs->exact_filename[0]) { |
7869 | pstrcpy(bs->filename, sizeof(bs->filename), bs->exact_filename); | |
97e2f021 | 7870 | } else { |
eab3a467 | 7871 | GString *json = qobject_to_json(QOBJECT(bs->full_open_options)); |
5c86bdf1 | 7872 | if (snprintf(bs->filename, sizeof(bs->filename), "json:%s", |
eab3a467 | 7873 | json->str) >= sizeof(bs->filename)) { |
5c86bdf1 EB |
7874 | /* Give user a hint if we truncated things. */ |
7875 | strcpy(bs->filename + sizeof(bs->filename) - 4, "..."); | |
7876 | } | |
eab3a467 | 7877 | g_string_free(json, true); |
91af7014 HR |
7878 | } |
7879 | } | |
e06018ad | 7880 | |
1e89d0f9 HR |
7881 | char *bdrv_dirname(BlockDriverState *bs, Error **errp) |
7882 | { | |
7883 | BlockDriver *drv = bs->drv; | |
52f72d6f | 7884 | BlockDriverState *child_bs; |
1e89d0f9 | 7885 | |
f791bf7f EGE |
7886 | GLOBAL_STATE_CODE(); |
7887 | ||
1e89d0f9 HR |
7888 | if (!drv) { |
7889 | error_setg(errp, "Node '%s' is ejected", bs->node_name); | |
7890 | return NULL; | |
7891 | } | |
7892 | ||
7893 | if (drv->bdrv_dirname) { | |
7894 | return drv->bdrv_dirname(bs, errp); | |
7895 | } | |
7896 | ||
52f72d6f HR |
7897 | child_bs = bdrv_primary_bs(bs); |
7898 | if (child_bs) { | |
7899 | return bdrv_dirname(child_bs, errp); | |
1e89d0f9 HR |
7900 | } |
7901 | ||
7902 | bdrv_refresh_filename(bs); | |
7903 | if (bs->exact_filename[0] != '\0') { | |
7904 | return path_combine(bs->exact_filename, ""); | |
7905 | } | |
7906 | ||
7907 | error_setg(errp, "Cannot generate a base directory for %s nodes", | |
7908 | drv->format_name); | |
7909 | return NULL; | |
7910 | } | |
7911 | ||
e06018ad WC |
7912 | /* |
7913 | * Hot add/remove a BDS's child. So the user can take a child offline when | |
7914 | * it is broken and take a new child online | |
7915 | */ | |
7916 | void bdrv_add_child(BlockDriverState *parent_bs, BlockDriverState *child_bs, | |
7917 | Error **errp) | |
7918 | { | |
f791bf7f | 7919 | GLOBAL_STATE_CODE(); |
e06018ad WC |
7920 | if (!parent_bs->drv || !parent_bs->drv->bdrv_add_child) { |
7921 | error_setg(errp, "The node %s does not support adding a child", | |
7922 | bdrv_get_device_or_node_name(parent_bs)); | |
7923 | return; | |
7924 | } | |
7925 | ||
7926 | if (!QLIST_EMPTY(&child_bs->parents)) { | |
7927 | error_setg(errp, "The node %s already has a parent", | |
7928 | child_bs->node_name); | |
7929 | return; | |
7930 | } | |
7931 | ||
7932 | parent_bs->drv->bdrv_add_child(parent_bs, child_bs, errp); | |
7933 | } | |
7934 | ||
7935 | void bdrv_del_child(BlockDriverState *parent_bs, BdrvChild *child, Error **errp) | |
7936 | { | |
7937 | BdrvChild *tmp; | |
7938 | ||
f791bf7f | 7939 | GLOBAL_STATE_CODE(); |
e06018ad WC |
7940 | if (!parent_bs->drv || !parent_bs->drv->bdrv_del_child) { |
7941 | error_setg(errp, "The node %s does not support removing a child", | |
7942 | bdrv_get_device_or_node_name(parent_bs)); | |
7943 | return; | |
7944 | } | |
7945 | ||
7946 | QLIST_FOREACH(tmp, &parent_bs->children, next) { | |
7947 | if (tmp == child) { | |
7948 | break; | |
7949 | } | |
7950 | } | |
7951 | ||
7952 | if (!tmp) { | |
7953 | error_setg(errp, "The node %s does not have a child named %s", | |
7954 | bdrv_get_device_or_node_name(parent_bs), | |
7955 | bdrv_get_device_or_node_name(child->bs)); | |
7956 | return; | |
7957 | } | |
7958 | ||
7959 | parent_bs->drv->bdrv_del_child(parent_bs, child, errp); | |
7960 | } | |
6f7a3b53 HR |
7961 | |
7962 | int bdrv_make_empty(BdrvChild *c, Error **errp) | |
7963 | { | |
7964 | BlockDriver *drv = c->bs->drv; | |
7965 | int ret; | |
7966 | ||
f791bf7f | 7967 | GLOBAL_STATE_CODE(); |
6f7a3b53 HR |
7968 | assert(c->perm & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)); |
7969 | ||
7970 | if (!drv->bdrv_make_empty) { | |
7971 | error_setg(errp, "%s does not support emptying nodes", | |
7972 | drv->format_name); | |
7973 | return -ENOTSUP; | |
7974 | } | |
7975 | ||
7976 | ret = drv->bdrv_make_empty(c->bs); | |
7977 | if (ret < 0) { | |
7978 | error_setg_errno(errp, -ret, "Failed to empty %s", | |
7979 | c->bs->filename); | |
7980 | return ret; | |
7981 | } | |
7982 | ||
7983 | return 0; | |
7984 | } | |
9a6fc887 HR |
7985 | |
7986 | /* | |
7987 | * Return the child that @bs acts as an overlay for, and from which data may be | |
7988 | * copied in COW or COR operations. Usually this is the backing file. | |
7989 | */ | |
7990 | BdrvChild *bdrv_cow_child(BlockDriverState *bs) | |
7991 | { | |
967d7905 EGE |
7992 | IO_CODE(); |
7993 | ||
9a6fc887 HR |
7994 | if (!bs || !bs->drv) { |
7995 | return NULL; | |
7996 | } | |
7997 | ||
7998 | if (bs->drv->is_filter) { | |
7999 | return NULL; | |
8000 | } | |
8001 | ||
8002 | if (!bs->backing) { | |
8003 | return NULL; | |
8004 | } | |
8005 | ||
8006 | assert(bs->backing->role & BDRV_CHILD_COW); | |
8007 | return bs->backing; | |
8008 | } | |
8009 | ||
8010 | /* | |
8011 | * If @bs acts as a filter for exactly one of its children, return | |
8012 | * that child. | |
8013 | */ | |
8014 | BdrvChild *bdrv_filter_child(BlockDriverState *bs) | |
8015 | { | |
8016 | BdrvChild *c; | |
967d7905 | 8017 | IO_CODE(); |
9a6fc887 HR |
8018 | |
8019 | if (!bs || !bs->drv) { | |
8020 | return NULL; | |
8021 | } | |
8022 | ||
8023 | if (!bs->drv->is_filter) { | |
8024 | return NULL; | |
8025 | } | |
8026 | ||
8027 | /* Only one of @backing or @file may be used */ | |
8028 | assert(!(bs->backing && bs->file)); | |
8029 | ||
8030 | c = bs->backing ?: bs->file; | |
8031 | if (!c) { | |
8032 | return NULL; | |
8033 | } | |
8034 | ||
8035 | assert(c->role & BDRV_CHILD_FILTERED); | |
8036 | return c; | |
8037 | } | |
8038 | ||
8039 | /* | |
8040 | * Return either the result of bdrv_cow_child() or bdrv_filter_child(), | |
8041 | * whichever is non-NULL. | |
8042 | * | |
8043 | * Return NULL if both are NULL. | |
8044 | */ | |
8045 | BdrvChild *bdrv_filter_or_cow_child(BlockDriverState *bs) | |
8046 | { | |
8047 | BdrvChild *cow_child = bdrv_cow_child(bs); | |
8048 | BdrvChild *filter_child = bdrv_filter_child(bs); | |
967d7905 | 8049 | IO_CODE(); |
9a6fc887 HR |
8050 | |
8051 | /* Filter nodes cannot have COW backing files */ | |
8052 | assert(!(cow_child && filter_child)); | |
8053 | ||
8054 | return cow_child ?: filter_child; | |
8055 | } | |
8056 | ||
8057 | /* | |
8058 | * Return the primary child of this node: For filters, that is the | |
8059 | * filtered child. For other nodes, that is usually the child storing | |
8060 | * metadata. | |
8061 | * (A generally more helpful description is that this is (usually) the | |
8062 | * child that has the same filename as @bs.) | |
8063 | * | |
8064 | * Drivers do not necessarily have a primary child; for example quorum | |
8065 | * does not. | |
8066 | */ | |
8067 | BdrvChild *bdrv_primary_child(BlockDriverState *bs) | |
8068 | { | |
8069 | BdrvChild *c, *found = NULL; | |
967d7905 | 8070 | IO_CODE(); |
9a6fc887 HR |
8071 | |
8072 | QLIST_FOREACH(c, &bs->children, next) { | |
8073 | if (c->role & BDRV_CHILD_PRIMARY) { | |
8074 | assert(!found); | |
8075 | found = c; | |
8076 | } | |
8077 | } | |
8078 | ||
8079 | return found; | |
8080 | } | |
d38d7eb8 HR |
8081 | |
8082 | static BlockDriverState *bdrv_do_skip_filters(BlockDriverState *bs, | |
8083 | bool stop_on_explicit_filter) | |
8084 | { | |
8085 | BdrvChild *c; | |
8086 | ||
8087 | if (!bs) { | |
8088 | return NULL; | |
8089 | } | |
8090 | ||
8091 | while (!(stop_on_explicit_filter && !bs->implicit)) { | |
8092 | c = bdrv_filter_child(bs); | |
8093 | if (!c) { | |
8094 | /* | |
8095 | * A filter that is embedded in a working block graph must | |
8096 | * have a child. Assert this here so this function does | |
8097 | * not return a filter node that is not expected by the | |
8098 | * caller. | |
8099 | */ | |
8100 | assert(!bs->drv || !bs->drv->is_filter); | |
8101 | break; | |
8102 | } | |
8103 | bs = c->bs; | |
8104 | } | |
8105 | /* | |
8106 | * Note that this treats nodes with bs->drv == NULL as not being | |
8107 | * filters (bs->drv == NULL should be replaced by something else | |
8108 | * anyway). | |
8109 | * The advantage of this behavior is that this function will thus | |
8110 | * always return a non-NULL value (given a non-NULL @bs). | |
8111 | */ | |
8112 | ||
8113 | return bs; | |
8114 | } | |
8115 | ||
8116 | /* | |
8117 | * Return the first BDS that has not been added implicitly or that | |
8118 | * does not have a filtered child down the chain starting from @bs | |
8119 | * (including @bs itself). | |
8120 | */ | |
8121 | BlockDriverState *bdrv_skip_implicit_filters(BlockDriverState *bs) | |
8122 | { | |
b4ad82aa | 8123 | GLOBAL_STATE_CODE(); |
d38d7eb8 HR |
8124 | return bdrv_do_skip_filters(bs, true); |
8125 | } | |
8126 | ||
8127 | /* | |
8128 | * Return the first BDS that does not have a filtered child down the | |
8129 | * chain starting from @bs (including @bs itself). | |
8130 | */ | |
8131 | BlockDriverState *bdrv_skip_filters(BlockDriverState *bs) | |
8132 | { | |
967d7905 | 8133 | IO_CODE(); |
d38d7eb8 HR |
8134 | return bdrv_do_skip_filters(bs, false); |
8135 | } | |
8136 | ||
8137 | /* | |
8138 | * For a backing chain, return the first non-filter backing image of | |
8139 | * the first non-filter image. | |
8140 | */ | |
8141 | BlockDriverState *bdrv_backing_chain_next(BlockDriverState *bs) | |
8142 | { | |
967d7905 | 8143 | IO_CODE(); |
d38d7eb8 HR |
8144 | return bdrv_skip_filters(bdrv_cow_bs(bdrv_skip_filters(bs))); |
8145 | } | |
0bc329fb HR |
8146 | |
8147 | /** | |
8148 | * Check whether [offset, offset + bytes) overlaps with the cached | |
8149 | * block-status data region. | |
8150 | * | |
8151 | * If so, and @pnum is not NULL, set *pnum to `bsc.data_end - offset`, | |
8152 | * which is what bdrv_bsc_is_data()'s interface needs. | |
8153 | * Otherwise, *pnum is not touched. | |
8154 | */ | |
8155 | static bool bdrv_bsc_range_overlaps_locked(BlockDriverState *bs, | |
8156 | int64_t offset, int64_t bytes, | |
8157 | int64_t *pnum) | |
8158 | { | |
8159 | BdrvBlockStatusCache *bsc = qatomic_rcu_read(&bs->block_status_cache); | |
8160 | bool overlaps; | |
8161 | ||
8162 | overlaps = | |
8163 | qatomic_read(&bsc->valid) && | |
8164 | ranges_overlap(offset, bytes, bsc->data_start, | |
8165 | bsc->data_end - bsc->data_start); | |
8166 | ||
8167 | if (overlaps && pnum) { | |
8168 | *pnum = bsc->data_end - offset; | |
8169 | } | |
8170 | ||
8171 | return overlaps; | |
8172 | } | |
8173 | ||
8174 | /** | |
8175 | * See block_int.h for this function's documentation. | |
8176 | */ | |
8177 | bool bdrv_bsc_is_data(BlockDriverState *bs, int64_t offset, int64_t *pnum) | |
8178 | { | |
967d7905 | 8179 | IO_CODE(); |
0bc329fb | 8180 | RCU_READ_LOCK_GUARD(); |
0bc329fb HR |
8181 | return bdrv_bsc_range_overlaps_locked(bs, offset, 1, pnum); |
8182 | } | |
8183 | ||
8184 | /** | |
8185 | * See block_int.h for this function's documentation. | |
8186 | */ | |
8187 | void bdrv_bsc_invalidate_range(BlockDriverState *bs, | |
8188 | int64_t offset, int64_t bytes) | |
8189 | { | |
967d7905 | 8190 | IO_CODE(); |
0bc329fb HR |
8191 | RCU_READ_LOCK_GUARD(); |
8192 | ||
8193 | if (bdrv_bsc_range_overlaps_locked(bs, offset, bytes, NULL)) { | |
8194 | qatomic_set(&bs->block_status_cache->valid, false); | |
8195 | } | |
8196 | } | |
8197 | ||
8198 | /** | |
8199 | * See block_int.h for this function's documentation. | |
8200 | */ | |
8201 | void bdrv_bsc_fill(BlockDriverState *bs, int64_t offset, int64_t bytes) | |
8202 | { | |
8203 | BdrvBlockStatusCache *new_bsc = g_new(BdrvBlockStatusCache, 1); | |
8204 | BdrvBlockStatusCache *old_bsc; | |
967d7905 | 8205 | IO_CODE(); |
0bc329fb HR |
8206 | |
8207 | *new_bsc = (BdrvBlockStatusCache) { | |
8208 | .valid = true, | |
8209 | .data_start = offset, | |
8210 | .data_end = offset + bytes, | |
8211 | }; | |
8212 | ||
8213 | QEMU_LOCK_GUARD(&bs->bsc_modify_lock); | |
8214 | ||
8215 | old_bsc = qatomic_rcu_read(&bs->block_status_cache); | |
8216 | qatomic_rcu_set(&bs->block_status_cache, new_bsc); | |
8217 | if (old_bsc) { | |
8218 | g_free_rcu(old_bsc, rcu); | |
8219 | } | |
8220 | } |