]>
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 | */ |
3d9f2d2a | 1038 | int refresh_total_sectors(BlockDriverState *bs, int64_t hint) |
51762288 SH |
1039 | { |
1040 | BlockDriver *drv = bs->drv; | |
967d7905 | 1041 | IO_CODE(); |
51762288 | 1042 | |
d470ad42 HR |
1043 | if (!drv) { |
1044 | return -ENOMEDIUM; | |
1045 | } | |
1046 | ||
396759ad | 1047 | /* Do not attempt drv->bdrv_getlength() on scsi-generic devices */ |
b192af8a | 1048 | if (bdrv_is_sg(bs)) |
396759ad NB |
1049 | return 0; |
1050 | ||
51762288 SH |
1051 | /* query actual device if possible, otherwise just trust the hint */ |
1052 | if (drv->bdrv_getlength) { | |
1053 | int64_t length = drv->bdrv_getlength(bs); | |
1054 | if (length < 0) { | |
1055 | return length; | |
1056 | } | |
7e382003 | 1057 | hint = DIV_ROUND_UP(length, BDRV_SECTOR_SIZE); |
51762288 SH |
1058 | } |
1059 | ||
1060 | bs->total_sectors = hint; | |
8b117001 VSO |
1061 | |
1062 | if (bs->total_sectors * BDRV_SECTOR_SIZE > BDRV_MAX_LENGTH) { | |
1063 | return -EFBIG; | |
1064 | } | |
1065 | ||
51762288 SH |
1066 | return 0; |
1067 | } | |
1068 | ||
cddff5ba KW |
1069 | /** |
1070 | * Combines a QDict of new block driver @options with any missing options taken | |
1071 | * from @old_options, so that leaving out an option defaults to its old value. | |
1072 | */ | |
1073 | static void bdrv_join_options(BlockDriverState *bs, QDict *options, | |
1074 | QDict *old_options) | |
1075 | { | |
da359909 | 1076 | GLOBAL_STATE_CODE(); |
cddff5ba KW |
1077 | if (bs->drv && bs->drv->bdrv_join_options) { |
1078 | bs->drv->bdrv_join_options(options, old_options); | |
1079 | } else { | |
1080 | qdict_join(options, old_options, false); | |
1081 | } | |
1082 | } | |
1083 | ||
543770bd AG |
1084 | static BlockdevDetectZeroesOptions bdrv_parse_detect_zeroes(QemuOpts *opts, |
1085 | int open_flags, | |
1086 | Error **errp) | |
1087 | { | |
1088 | Error *local_err = NULL; | |
1089 | char *value = qemu_opt_get_del(opts, "detect-zeroes"); | |
1090 | BlockdevDetectZeroesOptions detect_zeroes = | |
1091 | qapi_enum_parse(&BlockdevDetectZeroesOptions_lookup, value, | |
1092 | BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF, &local_err); | |
bdb73476 | 1093 | GLOBAL_STATE_CODE(); |
543770bd AG |
1094 | g_free(value); |
1095 | if (local_err) { | |
1096 | error_propagate(errp, local_err); | |
1097 | return detect_zeroes; | |
1098 | } | |
1099 | ||
1100 | if (detect_zeroes == BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP && | |
1101 | !(open_flags & BDRV_O_UNMAP)) | |
1102 | { | |
1103 | error_setg(errp, "setting detect-zeroes to unmap is not allowed " | |
1104 | "without setting discard operation to unmap"); | |
1105 | } | |
1106 | ||
1107 | return detect_zeroes; | |
1108 | } | |
1109 | ||
f80f2673 AM |
1110 | /** |
1111 | * Set open flags for aio engine | |
1112 | * | |
1113 | * Return 0 on success, -1 if the engine specified is invalid | |
1114 | */ | |
1115 | int bdrv_parse_aio(const char *mode, int *flags) | |
1116 | { | |
1117 | if (!strcmp(mode, "threads")) { | |
1118 | /* do nothing, default */ | |
1119 | } else if (!strcmp(mode, "native")) { | |
1120 | *flags |= BDRV_O_NATIVE_AIO; | |
1121 | #ifdef CONFIG_LINUX_IO_URING | |
1122 | } else if (!strcmp(mode, "io_uring")) { | |
1123 | *flags |= BDRV_O_IO_URING; | |
1124 | #endif | |
1125 | } else { | |
1126 | return -1; | |
1127 | } | |
1128 | ||
1129 | return 0; | |
1130 | } | |
1131 | ||
9e8f1835 PB |
1132 | /** |
1133 | * Set open flags for a given discard mode | |
1134 | * | |
1135 | * Return 0 on success, -1 if the discard mode was invalid. | |
1136 | */ | |
1137 | int bdrv_parse_discard_flags(const char *mode, int *flags) | |
1138 | { | |
1139 | *flags &= ~BDRV_O_UNMAP; | |
1140 | ||
1141 | if (!strcmp(mode, "off") || !strcmp(mode, "ignore")) { | |
1142 | /* do nothing */ | |
1143 | } else if (!strcmp(mode, "on") || !strcmp(mode, "unmap")) { | |
1144 | *flags |= BDRV_O_UNMAP; | |
1145 | } else { | |
1146 | return -1; | |
1147 | } | |
1148 | ||
1149 | return 0; | |
1150 | } | |
1151 | ||
c3993cdc SH |
1152 | /** |
1153 | * Set open flags for a given cache mode | |
1154 | * | |
1155 | * Return 0 on success, -1 if the cache mode was invalid. | |
1156 | */ | |
53e8ae01 | 1157 | int bdrv_parse_cache_mode(const char *mode, int *flags, bool *writethrough) |
c3993cdc SH |
1158 | { |
1159 | *flags &= ~BDRV_O_CACHE_MASK; | |
1160 | ||
1161 | if (!strcmp(mode, "off") || !strcmp(mode, "none")) { | |
53e8ae01 KW |
1162 | *writethrough = false; |
1163 | *flags |= BDRV_O_NOCACHE; | |
92196b2f | 1164 | } else if (!strcmp(mode, "directsync")) { |
53e8ae01 | 1165 | *writethrough = true; |
92196b2f | 1166 | *flags |= BDRV_O_NOCACHE; |
c3993cdc | 1167 | } else if (!strcmp(mode, "writeback")) { |
53e8ae01 | 1168 | *writethrough = false; |
c3993cdc | 1169 | } else if (!strcmp(mode, "unsafe")) { |
53e8ae01 | 1170 | *writethrough = false; |
c3993cdc SH |
1171 | *flags |= BDRV_O_NO_FLUSH; |
1172 | } else if (!strcmp(mode, "writethrough")) { | |
53e8ae01 | 1173 | *writethrough = true; |
c3993cdc SH |
1174 | } else { |
1175 | return -1; | |
1176 | } | |
1177 | ||
1178 | return 0; | |
1179 | } | |
1180 | ||
b5411555 KW |
1181 | static char *bdrv_child_get_parent_desc(BdrvChild *c) |
1182 | { | |
1183 | BlockDriverState *parent = c->opaque; | |
2c0a3acb | 1184 | return g_strdup_printf("node '%s'", bdrv_get_node_name(parent)); |
b5411555 KW |
1185 | } |
1186 | ||
20018e12 KW |
1187 | static void bdrv_child_cb_drained_begin(BdrvChild *child) |
1188 | { | |
1189 | BlockDriverState *bs = child->opaque; | |
a82a3bd1 | 1190 | bdrv_do_drained_begin_quiesce(bs, NULL); |
20018e12 KW |
1191 | } |
1192 | ||
89bd0305 KW |
1193 | static bool bdrv_child_cb_drained_poll(BdrvChild *child) |
1194 | { | |
1195 | BlockDriverState *bs = child->opaque; | |
299403ae | 1196 | return bdrv_drain_poll(bs, NULL, false); |
89bd0305 KW |
1197 | } |
1198 | ||
2f65df6e | 1199 | static void bdrv_child_cb_drained_end(BdrvChild *child) |
20018e12 KW |
1200 | { |
1201 | BlockDriverState *bs = child->opaque; | |
2f65df6e | 1202 | bdrv_drained_end(bs); |
20018e12 KW |
1203 | } |
1204 | ||
38701b6a KW |
1205 | static int bdrv_child_cb_inactivate(BdrvChild *child) |
1206 | { | |
1207 | BlockDriverState *bs = child->opaque; | |
bdb73476 | 1208 | GLOBAL_STATE_CODE(); |
38701b6a KW |
1209 | assert(bs->open_flags & BDRV_O_INACTIVE); |
1210 | return 0; | |
1211 | } | |
1212 | ||
27633e74 EGE |
1213 | static bool bdrv_child_cb_change_aio_ctx(BdrvChild *child, AioContext *ctx, |
1214 | GHashTable *visited, Transaction *tran, | |
1215 | Error **errp) | |
53a7d041 KW |
1216 | { |
1217 | BlockDriverState *bs = child->opaque; | |
27633e74 | 1218 | return bdrv_change_aio_context(bs, ctx, visited, tran, errp); |
53a7d041 KW |
1219 | } |
1220 | ||
b1e6fc08 | 1221 | /* |
73176bee KW |
1222 | * Returns the options and flags that a temporary snapshot should get, based on |
1223 | * the originally requested flags (the originally requested image will have | |
1224 | * flags like a backing file) | |
b1e6fc08 | 1225 | */ |
73176bee KW |
1226 | static void bdrv_temp_snapshot_options(int *child_flags, QDict *child_options, |
1227 | int parent_flags, QDict *parent_options) | |
b1e6fc08 | 1228 | { |
bdb73476 | 1229 | GLOBAL_STATE_CODE(); |
73176bee KW |
1230 | *child_flags = (parent_flags & ~BDRV_O_SNAPSHOT) | BDRV_O_TEMPORARY; |
1231 | ||
1232 | /* For temporary files, unconditional cache=unsafe is fine */ | |
73176bee KW |
1233 | qdict_set_default_str(child_options, BDRV_OPT_CACHE_DIRECT, "off"); |
1234 | qdict_set_default_str(child_options, BDRV_OPT_CACHE_NO_FLUSH, "on"); | |
41869044 | 1235 | |
3f48686f | 1236 | /* Copy the read-only and discard options from the parent */ |
f87a0e29 | 1237 | qdict_copy_default(child_options, parent_options, BDRV_OPT_READ_ONLY); |
3f48686f | 1238 | qdict_copy_default(child_options, parent_options, BDRV_OPT_DISCARD); |
f87a0e29 | 1239 | |
41869044 KW |
1240 | /* aio=native doesn't work for cache.direct=off, so disable it for the |
1241 | * temporary snapshot */ | |
1242 | *child_flags &= ~BDRV_O_NATIVE_AIO; | |
b1e6fc08 KW |
1243 | } |
1244 | ||
db95dbba KW |
1245 | static void bdrv_backing_attach(BdrvChild *c) |
1246 | { | |
1247 | BlockDriverState *parent = c->opaque; | |
1248 | BlockDriverState *backing_hd = c->bs; | |
1249 | ||
bdb73476 | 1250 | GLOBAL_STATE_CODE(); |
db95dbba KW |
1251 | assert(!parent->backing_blocker); |
1252 | error_setg(&parent->backing_blocker, | |
1253 | "node is used as backing hd of '%s'", | |
1254 | bdrv_get_device_or_node_name(parent)); | |
1255 | ||
f30c66ba HR |
1256 | bdrv_refresh_filename(backing_hd); |
1257 | ||
db95dbba | 1258 | parent->open_flags &= ~BDRV_O_NO_BACKING; |
db95dbba KW |
1259 | |
1260 | bdrv_op_block_all(backing_hd, parent->backing_blocker); | |
1261 | /* Otherwise we won't be able to commit or stream */ | |
1262 | bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_COMMIT_TARGET, | |
1263 | parent->backing_blocker); | |
1264 | bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_STREAM, | |
1265 | parent->backing_blocker); | |
1266 | /* | |
1267 | * We do backup in 3 ways: | |
1268 | * 1. drive backup | |
1269 | * The target bs is new opened, and the source is top BDS | |
1270 | * 2. blockdev backup | |
1271 | * Both the source and the target are top BDSes. | |
1272 | * 3. internal backup(used for block replication) | |
1273 | * Both the source and the target are backing file | |
1274 | * | |
1275 | * In case 1 and 2, neither the source nor the target is the backing file. | |
1276 | * In case 3, we will block the top BDS, so there is only one block job | |
1277 | * for the top BDS and its backing chain. | |
1278 | */ | |
1279 | bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_BACKUP_SOURCE, | |
1280 | parent->backing_blocker); | |
1281 | bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_BACKUP_TARGET, | |
1282 | parent->backing_blocker); | |
ca2f1234 | 1283 | } |
d736f119 | 1284 | |
db95dbba KW |
1285 | static void bdrv_backing_detach(BdrvChild *c) |
1286 | { | |
1287 | BlockDriverState *parent = c->opaque; | |
1288 | ||
bdb73476 | 1289 | GLOBAL_STATE_CODE(); |
db95dbba KW |
1290 | assert(parent->backing_blocker); |
1291 | bdrv_op_unblock_all(c->bs, parent->backing_blocker); | |
1292 | error_free(parent->backing_blocker); | |
1293 | parent->backing_blocker = NULL; | |
48e08288 | 1294 | } |
d736f119 | 1295 | |
6858eba0 KW |
1296 | static int bdrv_backing_update_filename(BdrvChild *c, BlockDriverState *base, |
1297 | const char *filename, Error **errp) | |
1298 | { | |
1299 | BlockDriverState *parent = c->opaque; | |
e94d3dba | 1300 | bool read_only = bdrv_is_read_only(parent); |
6858eba0 | 1301 | int ret; |
bdb73476 | 1302 | GLOBAL_STATE_CODE(); |
6858eba0 | 1303 | |
e94d3dba AG |
1304 | if (read_only) { |
1305 | ret = bdrv_reopen_set_read_only(parent, false, errp); | |
61f09cea KW |
1306 | if (ret < 0) { |
1307 | return ret; | |
1308 | } | |
1309 | } | |
1310 | ||
6858eba0 | 1311 | ret = bdrv_change_backing_file(parent, filename, |
e54ee1b3 EB |
1312 | base->drv ? base->drv->format_name : "", |
1313 | false); | |
6858eba0 | 1314 | if (ret < 0) { |
64730694 | 1315 | error_setg_errno(errp, -ret, "Could not update backing file link"); |
6858eba0 KW |
1316 | } |
1317 | ||
e94d3dba AG |
1318 | if (read_only) { |
1319 | bdrv_reopen_set_read_only(parent, true, NULL); | |
61f09cea KW |
1320 | } |
1321 | ||
6858eba0 KW |
1322 | return ret; |
1323 | } | |
1324 | ||
fae8bd39 HR |
1325 | /* |
1326 | * Returns the options and flags that a generic child of a BDS should | |
1327 | * get, based on the given options and flags for the parent BDS. | |
1328 | */ | |
00ff7ffd HR |
1329 | static void bdrv_inherited_options(BdrvChildRole role, bool parent_is_format, |
1330 | int *child_flags, QDict *child_options, | |
1331 | int parent_flags, QDict *parent_options) | |
fae8bd39 HR |
1332 | { |
1333 | int flags = parent_flags; | |
bdb73476 | 1334 | GLOBAL_STATE_CODE(); |
fae8bd39 HR |
1335 | |
1336 | /* | |
1337 | * First, decide whether to set, clear, or leave BDRV_O_PROTOCOL. | |
1338 | * Generally, the question to answer is: Should this child be | |
1339 | * format-probed by default? | |
1340 | */ | |
1341 | ||
1342 | /* | |
1343 | * Pure and non-filtered data children of non-format nodes should | |
1344 | * be probed by default (even when the node itself has BDRV_O_PROTOCOL | |
1345 | * set). This only affects a very limited set of drivers (namely | |
1346 | * quorum and blkverify when this comment was written). | |
1347 | * Force-clear BDRV_O_PROTOCOL then. | |
1348 | */ | |
1349 | if (!parent_is_format && | |
1350 | (role & BDRV_CHILD_DATA) && | |
1351 | !(role & (BDRV_CHILD_METADATA | BDRV_CHILD_FILTERED))) | |
1352 | { | |
1353 | flags &= ~BDRV_O_PROTOCOL; | |
1354 | } | |
1355 | ||
1356 | /* | |
1357 | * All children of format nodes (except for COW children) and all | |
1358 | * metadata children in general should never be format-probed. | |
1359 | * Force-set BDRV_O_PROTOCOL then. | |
1360 | */ | |
1361 | if ((parent_is_format && !(role & BDRV_CHILD_COW)) || | |
1362 | (role & BDRV_CHILD_METADATA)) | |
1363 | { | |
1364 | flags |= BDRV_O_PROTOCOL; | |
1365 | } | |
1366 | ||
1367 | /* | |
1368 | * If the cache mode isn't explicitly set, inherit direct and no-flush from | |
1369 | * the parent. | |
1370 | */ | |
1371 | qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_DIRECT); | |
1372 | qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_NO_FLUSH); | |
1373 | qdict_copy_default(child_options, parent_options, BDRV_OPT_FORCE_SHARE); | |
1374 | ||
1375 | if (role & BDRV_CHILD_COW) { | |
1376 | /* backing files are opened read-only by default */ | |
1377 | qdict_set_default_str(child_options, BDRV_OPT_READ_ONLY, "on"); | |
1378 | qdict_set_default_str(child_options, BDRV_OPT_AUTO_READ_ONLY, "off"); | |
1379 | } else { | |
1380 | /* Inherit the read-only option from the parent if it's not set */ | |
1381 | qdict_copy_default(child_options, parent_options, BDRV_OPT_READ_ONLY); | |
1382 | qdict_copy_default(child_options, parent_options, | |
1383 | BDRV_OPT_AUTO_READ_ONLY); | |
1384 | } | |
1385 | ||
1386 | /* | |
1387 | * bdrv_co_pdiscard() respects unmap policy for the parent, so we | |
1388 | * can default to enable it on lower layers regardless of the | |
1389 | * parent option. | |
1390 | */ | |
1391 | qdict_set_default_str(child_options, BDRV_OPT_DISCARD, "unmap"); | |
1392 | ||
1393 | /* Clear flags that only apply to the top layer */ | |
1394 | flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_COPY_ON_READ); | |
1395 | ||
1396 | if (role & BDRV_CHILD_METADATA) { | |
1397 | flags &= ~BDRV_O_NO_IO; | |
1398 | } | |
1399 | if (role & BDRV_CHILD_COW) { | |
1400 | flags &= ~BDRV_O_TEMPORARY; | |
1401 | } | |
1402 | ||
1403 | *child_flags = flags; | |
1404 | } | |
1405 | ||
303de47b | 1406 | static void GRAPH_WRLOCK bdrv_child_cb_attach(BdrvChild *child) |
ca2f1234 HR |
1407 | { |
1408 | BlockDriverState *bs = child->opaque; | |
1409 | ||
3f35f82e | 1410 | assert_bdrv_graph_writable(); |
a225369b | 1411 | QLIST_INSERT_HEAD(&bs->children, child, next); |
5bb04747 VSO |
1412 | if (bs->drv->is_filter || (child->role & BDRV_CHILD_FILTERED)) { |
1413 | /* | |
1414 | * Here we handle filters and block/raw-format.c when it behave like | |
1415 | * filter. They generally have a single PRIMARY child, which is also the | |
1416 | * FILTERED child, and that they may have multiple more children, which | |
1417 | * are neither PRIMARY nor FILTERED. And never we have a COW child here. | |
1418 | * So bs->file will be the PRIMARY child, unless the PRIMARY child goes | |
1419 | * into bs->backing on exceptional cases; and bs->backing will be | |
1420 | * nothing else. | |
1421 | */ | |
1422 | assert(!(child->role & BDRV_CHILD_COW)); | |
1423 | if (child->role & BDRV_CHILD_PRIMARY) { | |
1424 | assert(child->role & BDRV_CHILD_FILTERED); | |
1425 | assert(!bs->backing); | |
1426 | assert(!bs->file); | |
1427 | ||
1428 | if (bs->drv->filtered_child_is_backing) { | |
1429 | bs->backing = child; | |
1430 | } else { | |
1431 | bs->file = child; | |
1432 | } | |
1433 | } else { | |
1434 | assert(!(child->role & BDRV_CHILD_FILTERED)); | |
1435 | } | |
1436 | } else if (child->role & BDRV_CHILD_COW) { | |
1437 | assert(bs->drv->supports_backing); | |
1438 | assert(!(child->role & BDRV_CHILD_PRIMARY)); | |
1439 | assert(!bs->backing); | |
1440 | bs->backing = child; | |
ca2f1234 | 1441 | bdrv_backing_attach(child); |
5bb04747 VSO |
1442 | } else if (child->role & BDRV_CHILD_PRIMARY) { |
1443 | assert(!bs->file); | |
1444 | bs->file = child; | |
ca2f1234 | 1445 | } |
ca2f1234 HR |
1446 | } |
1447 | ||
303de47b | 1448 | static void GRAPH_WRLOCK bdrv_child_cb_detach(BdrvChild *child) |
48e08288 HR |
1449 | { |
1450 | BlockDriverState *bs = child->opaque; | |
1451 | ||
1452 | if (child->role & BDRV_CHILD_COW) { | |
1453 | bdrv_backing_detach(child); | |
1454 | } | |
1455 | ||
3f35f82e | 1456 | assert_bdrv_graph_writable(); |
a225369b | 1457 | QLIST_REMOVE(child, next); |
5bb04747 VSO |
1458 | if (child == bs->backing) { |
1459 | assert(child != bs->file); | |
1460 | bs->backing = NULL; | |
1461 | } else if (child == bs->file) { | |
1462 | bs->file = NULL; | |
1463 | } | |
48e08288 HR |
1464 | } |
1465 | ||
43483550 HR |
1466 | static int bdrv_child_cb_update_filename(BdrvChild *c, BlockDriverState *base, |
1467 | const char *filename, Error **errp) | |
1468 | { | |
1469 | if (c->role & BDRV_CHILD_COW) { | |
1470 | return bdrv_backing_update_filename(c, base, filename, errp); | |
1471 | } | |
1472 | return 0; | |
1473 | } | |
1474 | ||
fb62b588 | 1475 | AioContext *child_of_bds_get_parent_aio_context(BdrvChild *c) |
3ca1f322 VSO |
1476 | { |
1477 | BlockDriverState *bs = c->opaque; | |
384a48fb | 1478 | IO_CODE(); |
3ca1f322 VSO |
1479 | |
1480 | return bdrv_get_aio_context(bs); | |
1481 | } | |
1482 | ||
43483550 HR |
1483 | const BdrvChildClass child_of_bds = { |
1484 | .parent_is_bds = true, | |
1485 | .get_parent_desc = bdrv_child_get_parent_desc, | |
1486 | .inherit_options = bdrv_inherited_options, | |
1487 | .drained_begin = bdrv_child_cb_drained_begin, | |
1488 | .drained_poll = bdrv_child_cb_drained_poll, | |
1489 | .drained_end = bdrv_child_cb_drained_end, | |
1490 | .attach = bdrv_child_cb_attach, | |
1491 | .detach = bdrv_child_cb_detach, | |
1492 | .inactivate = bdrv_child_cb_inactivate, | |
27633e74 | 1493 | .change_aio_ctx = bdrv_child_cb_change_aio_ctx, |
43483550 | 1494 | .update_filename = bdrv_child_cb_update_filename, |
fb62b588 | 1495 | .get_parent_aio_context = child_of_bds_get_parent_aio_context, |
43483550 HR |
1496 | }; |
1497 | ||
3ca1f322 VSO |
1498 | AioContext *bdrv_child_get_parent_aio_context(BdrvChild *c) |
1499 | { | |
d5f8d79c | 1500 | IO_CODE(); |
3ca1f322 VSO |
1501 | return c->klass->get_parent_aio_context(c); |
1502 | } | |
1503 | ||
7b272452 KW |
1504 | static int bdrv_open_flags(BlockDriverState *bs, int flags) |
1505 | { | |
61de4c68 | 1506 | int open_flags = flags; |
bdb73476 | 1507 | GLOBAL_STATE_CODE(); |
7b272452 KW |
1508 | |
1509 | /* | |
1510 | * Clear flags that are internal to the block layer before opening the | |
1511 | * image. | |
1512 | */ | |
20cca275 | 1513 | open_flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_PROTOCOL); |
7b272452 | 1514 | |
7b272452 KW |
1515 | return open_flags; |
1516 | } | |
1517 | ||
91a097e7 KW |
1518 | static void update_flags_from_options(int *flags, QemuOpts *opts) |
1519 | { | |
bdb73476 EGE |
1520 | GLOBAL_STATE_CODE(); |
1521 | ||
2a3d4331 | 1522 | *flags &= ~(BDRV_O_CACHE_MASK | BDRV_O_RDWR | BDRV_O_AUTO_RDONLY); |
91a097e7 | 1523 | |
57f9db9a | 1524 | if (qemu_opt_get_bool_del(opts, BDRV_OPT_CACHE_NO_FLUSH, false)) { |
91a097e7 KW |
1525 | *flags |= BDRV_O_NO_FLUSH; |
1526 | } | |
1527 | ||
57f9db9a | 1528 | if (qemu_opt_get_bool_del(opts, BDRV_OPT_CACHE_DIRECT, false)) { |
91a097e7 KW |
1529 | *flags |= BDRV_O_NOCACHE; |
1530 | } | |
f87a0e29 | 1531 | |
57f9db9a | 1532 | if (!qemu_opt_get_bool_del(opts, BDRV_OPT_READ_ONLY, false)) { |
f87a0e29 AG |
1533 | *flags |= BDRV_O_RDWR; |
1534 | } | |
1535 | ||
e35bdc12 KW |
1536 | if (qemu_opt_get_bool_del(opts, BDRV_OPT_AUTO_READ_ONLY, false)) { |
1537 | *flags |= BDRV_O_AUTO_RDONLY; | |
1538 | } | |
91a097e7 KW |
1539 | } |
1540 | ||
1541 | static void update_options_from_flags(QDict *options, int flags) | |
1542 | { | |
bdb73476 | 1543 | GLOBAL_STATE_CODE(); |
91a097e7 | 1544 | if (!qdict_haskey(options, BDRV_OPT_CACHE_DIRECT)) { |
46f5ac20 | 1545 | qdict_put_bool(options, BDRV_OPT_CACHE_DIRECT, flags & BDRV_O_NOCACHE); |
91a097e7 KW |
1546 | } |
1547 | if (!qdict_haskey(options, BDRV_OPT_CACHE_NO_FLUSH)) { | |
46f5ac20 EB |
1548 | qdict_put_bool(options, BDRV_OPT_CACHE_NO_FLUSH, |
1549 | flags & BDRV_O_NO_FLUSH); | |
91a097e7 | 1550 | } |
f87a0e29 | 1551 | if (!qdict_haskey(options, BDRV_OPT_READ_ONLY)) { |
46f5ac20 | 1552 | qdict_put_bool(options, BDRV_OPT_READ_ONLY, !(flags & BDRV_O_RDWR)); |
f87a0e29 | 1553 | } |
e35bdc12 KW |
1554 | if (!qdict_haskey(options, BDRV_OPT_AUTO_READ_ONLY)) { |
1555 | qdict_put_bool(options, BDRV_OPT_AUTO_READ_ONLY, | |
1556 | flags & BDRV_O_AUTO_RDONLY); | |
1557 | } | |
91a097e7 KW |
1558 | } |
1559 | ||
636ea370 KW |
1560 | static void bdrv_assign_node_name(BlockDriverState *bs, |
1561 | const char *node_name, | |
1562 | Error **errp) | |
6913c0c2 | 1563 | { |
15489c76 | 1564 | char *gen_node_name = NULL; |
bdb73476 | 1565 | GLOBAL_STATE_CODE(); |
6913c0c2 | 1566 | |
15489c76 JC |
1567 | if (!node_name) { |
1568 | node_name = gen_node_name = id_generate(ID_BLOCK); | |
1569 | } else if (!id_wellformed(node_name)) { | |
1570 | /* | |
1571 | * Check for empty string or invalid characters, but not if it is | |
1572 | * generated (generated names use characters not available to the user) | |
1573 | */ | |
785ec4b1 | 1574 | error_setg(errp, "Invalid node-name: '%s'", node_name); |
636ea370 | 1575 | return; |
6913c0c2 BC |
1576 | } |
1577 | ||
0c5e94ee | 1578 | /* takes care of avoiding namespaces collisions */ |
7f06d47e | 1579 | if (blk_by_name(node_name)) { |
0c5e94ee BC |
1580 | error_setg(errp, "node-name=%s is conflicting with a device id", |
1581 | node_name); | |
15489c76 | 1582 | goto out; |
0c5e94ee BC |
1583 | } |
1584 | ||
6913c0c2 BC |
1585 | /* takes care of avoiding duplicates node names */ |
1586 | if (bdrv_find_node(node_name)) { | |
785ec4b1 | 1587 | error_setg(errp, "Duplicate nodes with node-name='%s'", node_name); |
15489c76 | 1588 | goto out; |
6913c0c2 BC |
1589 | } |
1590 | ||
824808dd KW |
1591 | /* Make sure that the node name isn't truncated */ |
1592 | if (strlen(node_name) >= sizeof(bs->node_name)) { | |
1593 | error_setg(errp, "Node name too long"); | |
1594 | goto out; | |
1595 | } | |
1596 | ||
6913c0c2 BC |
1597 | /* copy node name into the bs and insert it into the graph list */ |
1598 | pstrcpy(bs->node_name, sizeof(bs->node_name), node_name); | |
1599 | QTAILQ_INSERT_TAIL(&graph_bdrv_states, bs, node_list); | |
15489c76 JC |
1600 | out: |
1601 | g_free(gen_node_name); | |
6913c0c2 BC |
1602 | } |
1603 | ||
01a56501 KW |
1604 | static int bdrv_open_driver(BlockDriverState *bs, BlockDriver *drv, |
1605 | const char *node_name, QDict *options, | |
1606 | int open_flags, Error **errp) | |
1607 | { | |
1608 | Error *local_err = NULL; | |
0f12264e | 1609 | int i, ret; |
da359909 | 1610 | GLOBAL_STATE_CODE(); |
01a56501 KW |
1611 | |
1612 | bdrv_assign_node_name(bs, node_name, &local_err); | |
1613 | if (local_err) { | |
1614 | error_propagate(errp, local_err); | |
1615 | return -EINVAL; | |
1616 | } | |
1617 | ||
1618 | bs->drv = drv; | |
1619 | bs->opaque = g_malloc0(drv->instance_size); | |
1620 | ||
1621 | if (drv->bdrv_file_open) { | |
1622 | assert(!drv->bdrv_needs_filename || bs->filename[0]); | |
1623 | ret = drv->bdrv_file_open(bs, options, open_flags, &local_err); | |
680c7f96 | 1624 | } else if (drv->bdrv_open) { |
01a56501 | 1625 | ret = drv->bdrv_open(bs, options, open_flags, &local_err); |
680c7f96 KW |
1626 | } else { |
1627 | ret = 0; | |
01a56501 KW |
1628 | } |
1629 | ||
1630 | if (ret < 0) { | |
1631 | if (local_err) { | |
1632 | error_propagate(errp, local_err); | |
1633 | } else if (bs->filename[0]) { | |
1634 | error_setg_errno(errp, -ret, "Could not open '%s'", bs->filename); | |
1635 | } else { | |
1636 | error_setg_errno(errp, -ret, "Could not open image"); | |
1637 | } | |
180ca19a | 1638 | goto open_failed; |
01a56501 KW |
1639 | } |
1640 | ||
e8b65355 SH |
1641 | assert(!(bs->supported_read_flags & ~BDRV_REQ_MASK)); |
1642 | assert(!(bs->supported_write_flags & ~BDRV_REQ_MASK)); | |
1643 | ||
1644 | /* | |
1645 | * Always allow the BDRV_REQ_REGISTERED_BUF optimization hint. This saves | |
1646 | * drivers that pass read/write requests through to a child the trouble of | |
1647 | * declaring support explicitly. | |
1648 | * | |
1649 | * Drivers must not propagate this flag accidentally when they initiate I/O | |
1650 | * to a bounce buffer. That case should be rare though. | |
1651 | */ | |
1652 | bs->supported_read_flags |= BDRV_REQ_REGISTERED_BUF; | |
1653 | bs->supported_write_flags |= BDRV_REQ_REGISTERED_BUF; | |
1654 | ||
01a56501 KW |
1655 | ret = refresh_total_sectors(bs, bs->total_sectors); |
1656 | if (ret < 0) { | |
1657 | error_setg_errno(errp, -ret, "Could not refresh total sector count"); | |
180ca19a | 1658 | return ret; |
01a56501 KW |
1659 | } |
1660 | ||
1e4c797c | 1661 | bdrv_refresh_limits(bs, NULL, &local_err); |
01a56501 KW |
1662 | if (local_err) { |
1663 | error_propagate(errp, local_err); | |
180ca19a | 1664 | return -EINVAL; |
01a56501 KW |
1665 | } |
1666 | ||
1667 | assert(bdrv_opt_mem_align(bs) != 0); | |
1668 | assert(bdrv_min_mem_align(bs) != 0); | |
1669 | assert(is_power_of_2(bs->bl.request_alignment)); | |
1670 | ||
0f12264e | 1671 | for (i = 0; i < bs->quiesce_counter; i++) { |
5e8ac217 KW |
1672 | if (drv->bdrv_drain_begin) { |
1673 | drv->bdrv_drain_begin(bs); | |
0f12264e KW |
1674 | } |
1675 | } | |
1676 | ||
01a56501 | 1677 | return 0; |
180ca19a MP |
1678 | open_failed: |
1679 | bs->drv = NULL; | |
1680 | if (bs->file != NULL) { | |
1681 | bdrv_unref_child(bs, bs->file); | |
5bb04747 | 1682 | assert(!bs->file); |
180ca19a | 1683 | } |
01a56501 KW |
1684 | g_free(bs->opaque); |
1685 | bs->opaque = NULL; | |
01a56501 KW |
1686 | return ret; |
1687 | } | |
1688 | ||
621d1737 VSO |
1689 | /* |
1690 | * Create and open a block node. | |
1691 | * | |
1692 | * @options is a QDict of options to pass to the block drivers, or NULL for an | |
1693 | * empty set of options. The reference to the QDict belongs to the block layer | |
1694 | * after the call (even on failure), so if the caller intends to reuse the | |
1695 | * dictionary, it needs to use qobject_ref() before calling bdrv_open. | |
1696 | */ | |
1697 | BlockDriverState *bdrv_new_open_driver_opts(BlockDriver *drv, | |
1698 | const char *node_name, | |
1699 | QDict *options, int flags, | |
1700 | Error **errp) | |
680c7f96 KW |
1701 | { |
1702 | BlockDriverState *bs; | |
1703 | int ret; | |
1704 | ||
f791bf7f EGE |
1705 | GLOBAL_STATE_CODE(); |
1706 | ||
680c7f96 KW |
1707 | bs = bdrv_new(); |
1708 | bs->open_flags = flags; | |
621d1737 VSO |
1709 | bs->options = options ?: qdict_new(); |
1710 | bs->explicit_options = qdict_clone_shallow(bs->options); | |
680c7f96 KW |
1711 | bs->opaque = NULL; |
1712 | ||
1713 | update_options_from_flags(bs->options, flags); | |
1714 | ||
1715 | ret = bdrv_open_driver(bs, drv, node_name, bs->options, flags, errp); | |
1716 | if (ret < 0) { | |
cb3e7f08 | 1717 | qobject_unref(bs->explicit_options); |
180ca19a | 1718 | bs->explicit_options = NULL; |
cb3e7f08 | 1719 | qobject_unref(bs->options); |
180ca19a | 1720 | bs->options = NULL; |
680c7f96 KW |
1721 | bdrv_unref(bs); |
1722 | return NULL; | |
1723 | } | |
1724 | ||
1725 | return bs; | |
1726 | } | |
1727 | ||
621d1737 VSO |
1728 | /* Create and open a block node. */ |
1729 | BlockDriverState *bdrv_new_open_driver(BlockDriver *drv, const char *node_name, | |
1730 | int flags, Error **errp) | |
1731 | { | |
f791bf7f | 1732 | GLOBAL_STATE_CODE(); |
621d1737 VSO |
1733 | return bdrv_new_open_driver_opts(drv, node_name, NULL, flags, errp); |
1734 | } | |
1735 | ||
c5f3014b | 1736 | QemuOptsList bdrv_runtime_opts = { |
18edf289 KW |
1737 | .name = "bdrv_common", |
1738 | .head = QTAILQ_HEAD_INITIALIZER(bdrv_runtime_opts.head), | |
1739 | .desc = { | |
1740 | { | |
1741 | .name = "node-name", | |
1742 | .type = QEMU_OPT_STRING, | |
1743 | .help = "Node name of the block device node", | |
1744 | }, | |
62392ebb KW |
1745 | { |
1746 | .name = "driver", | |
1747 | .type = QEMU_OPT_STRING, | |
1748 | .help = "Block driver to use for the node", | |
1749 | }, | |
91a097e7 KW |
1750 | { |
1751 | .name = BDRV_OPT_CACHE_DIRECT, | |
1752 | .type = QEMU_OPT_BOOL, | |
1753 | .help = "Bypass software writeback cache on the host", | |
1754 | }, | |
1755 | { | |
1756 | .name = BDRV_OPT_CACHE_NO_FLUSH, | |
1757 | .type = QEMU_OPT_BOOL, | |
1758 | .help = "Ignore flush requests", | |
1759 | }, | |
f87a0e29 AG |
1760 | { |
1761 | .name = BDRV_OPT_READ_ONLY, | |
1762 | .type = QEMU_OPT_BOOL, | |
1763 | .help = "Node is opened in read-only mode", | |
1764 | }, | |
e35bdc12 KW |
1765 | { |
1766 | .name = BDRV_OPT_AUTO_READ_ONLY, | |
1767 | .type = QEMU_OPT_BOOL, | |
1768 | .help = "Node can become read-only if opening read-write fails", | |
1769 | }, | |
692e01a2 KW |
1770 | { |
1771 | .name = "detect-zeroes", | |
1772 | .type = QEMU_OPT_STRING, | |
1773 | .help = "try to optimize zero writes (off, on, unmap)", | |
1774 | }, | |
818584a4 | 1775 | { |
415bbca8 | 1776 | .name = BDRV_OPT_DISCARD, |
818584a4 KW |
1777 | .type = QEMU_OPT_STRING, |
1778 | .help = "discard operation (ignore/off, unmap/on)", | |
1779 | }, | |
5a9347c6 FZ |
1780 | { |
1781 | .name = BDRV_OPT_FORCE_SHARE, | |
1782 | .type = QEMU_OPT_BOOL, | |
1783 | .help = "always accept other writers (default: off)", | |
1784 | }, | |
18edf289 KW |
1785 | { /* end of list */ } |
1786 | }, | |
1787 | }; | |
1788 | ||
5a5e7f8c ML |
1789 | QemuOptsList bdrv_create_opts_simple = { |
1790 | .name = "simple-create-opts", | |
1791 | .head = QTAILQ_HEAD_INITIALIZER(bdrv_create_opts_simple.head), | |
fd17146c HR |
1792 | .desc = { |
1793 | { | |
1794 | .name = BLOCK_OPT_SIZE, | |
1795 | .type = QEMU_OPT_SIZE, | |
1796 | .help = "Virtual disk size" | |
1797 | }, | |
1798 | { | |
1799 | .name = BLOCK_OPT_PREALLOC, | |
1800 | .type = QEMU_OPT_STRING, | |
1801 | .help = "Preallocation mode (allowed values: off)" | |
1802 | }, | |
1803 | { /* end of list */ } | |
1804 | } | |
1805 | }; | |
1806 | ||
57915332 KW |
1807 | /* |
1808 | * Common part for opening disk images and files | |
b6ad491a KW |
1809 | * |
1810 | * Removes all processed options from *options. | |
57915332 | 1811 | */ |
5696c6e3 | 1812 | static int bdrv_open_common(BlockDriverState *bs, BlockBackend *file, |
82dc8b41 | 1813 | QDict *options, Error **errp) |
57915332 KW |
1814 | { |
1815 | int ret, open_flags; | |
035fccdf | 1816 | const char *filename; |
62392ebb | 1817 | const char *driver_name = NULL; |
6913c0c2 | 1818 | const char *node_name = NULL; |
818584a4 | 1819 | const char *discard; |
18edf289 | 1820 | QemuOpts *opts; |
62392ebb | 1821 | BlockDriver *drv; |
34b5d2c6 | 1822 | Error *local_err = NULL; |
307261b2 | 1823 | bool ro; |
57915332 | 1824 | |
6405875c | 1825 | assert(bs->file == NULL); |
707ff828 | 1826 | assert(options != NULL && bs->options != options); |
bdb73476 | 1827 | GLOBAL_STATE_CODE(); |
57915332 | 1828 | |
62392ebb | 1829 | opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort); |
af175e85 | 1830 | if (!qemu_opts_absorb_qdict(opts, options, errp)) { |
62392ebb KW |
1831 | ret = -EINVAL; |
1832 | goto fail_opts; | |
1833 | } | |
1834 | ||
9b7e8691 AG |
1835 | update_flags_from_options(&bs->open_flags, opts); |
1836 | ||
62392ebb KW |
1837 | driver_name = qemu_opt_get(opts, "driver"); |
1838 | drv = bdrv_find_format(driver_name); | |
1839 | assert(drv != NULL); | |
1840 | ||
5a9347c6 FZ |
1841 | bs->force_share = qemu_opt_get_bool(opts, BDRV_OPT_FORCE_SHARE, false); |
1842 | ||
1843 | if (bs->force_share && (bs->open_flags & BDRV_O_RDWR)) { | |
1844 | error_setg(errp, | |
1845 | BDRV_OPT_FORCE_SHARE | |
1846 | "=on can only be used with read-only images"); | |
1847 | ret = -EINVAL; | |
1848 | goto fail_opts; | |
1849 | } | |
1850 | ||
45673671 | 1851 | if (file != NULL) { |
f30c66ba | 1852 | bdrv_refresh_filename(blk_bs(file)); |
5696c6e3 | 1853 | filename = blk_bs(file)->filename; |
45673671 | 1854 | } else { |
129c7d1c MA |
1855 | /* |
1856 | * Caution: while qdict_get_try_str() is fine, getting | |
1857 | * non-string types would require more care. When @options | |
1858 | * come from -blockdev or blockdev_add, its members are typed | |
1859 | * according to the QAPI schema, but when they come from | |
1860 | * -drive, they're all QString. | |
1861 | */ | |
45673671 KW |
1862 | filename = qdict_get_try_str(options, "filename"); |
1863 | } | |
1864 | ||
4a008240 | 1865 | if (drv->bdrv_needs_filename && (!filename || !filename[0])) { |
765003db KW |
1866 | error_setg(errp, "The '%s' block driver requires a file name", |
1867 | drv->format_name); | |
18edf289 KW |
1868 | ret = -EINVAL; |
1869 | goto fail_opts; | |
6913c0c2 | 1870 | } |
6913c0c2 | 1871 | |
82dc8b41 KW |
1872 | trace_bdrv_open_common(bs, filename ?: "", bs->open_flags, |
1873 | drv->format_name); | |
62392ebb | 1874 | |
307261b2 VSO |
1875 | ro = bdrv_is_read_only(bs); |
1876 | ||
1877 | if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv, ro)) { | |
1878 | if (!ro && bdrv_is_whitelisted(drv, true)) { | |
8be25de6 KW |
1879 | ret = bdrv_apply_auto_read_only(bs, NULL, NULL); |
1880 | } else { | |
1881 | ret = -ENOTSUP; | |
1882 | } | |
1883 | if (ret < 0) { | |
1884 | error_setg(errp, | |
307261b2 | 1885 | !ro && bdrv_is_whitelisted(drv, true) |
8be25de6 KW |
1886 | ? "Driver '%s' can only be used for read-only devices" |
1887 | : "Driver '%s' is not whitelisted", | |
1888 | drv->format_name); | |
1889 | goto fail_opts; | |
1890 | } | |
b64ec4e4 | 1891 | } |
57915332 | 1892 | |
d3faa13e | 1893 | /* bdrv_new() and bdrv_close() make it so */ |
d73415a3 | 1894 | assert(qatomic_read(&bs->copy_on_read) == 0); |
d3faa13e | 1895 | |
82dc8b41 | 1896 | if (bs->open_flags & BDRV_O_COPY_ON_READ) { |
307261b2 | 1897 | if (!ro) { |
0ebd24e0 KW |
1898 | bdrv_enable_copy_on_read(bs); |
1899 | } else { | |
1900 | error_setg(errp, "Can't use copy-on-read on read-only device"); | |
18edf289 KW |
1901 | ret = -EINVAL; |
1902 | goto fail_opts; | |
0ebd24e0 | 1903 | } |
53fec9d3 SH |
1904 | } |
1905 | ||
415bbca8 | 1906 | discard = qemu_opt_get(opts, BDRV_OPT_DISCARD); |
818584a4 KW |
1907 | if (discard != NULL) { |
1908 | if (bdrv_parse_discard_flags(discard, &bs->open_flags) != 0) { | |
1909 | error_setg(errp, "Invalid discard option"); | |
1910 | ret = -EINVAL; | |
1911 | goto fail_opts; | |
1912 | } | |
1913 | } | |
1914 | ||
543770bd AG |
1915 | bs->detect_zeroes = |
1916 | bdrv_parse_detect_zeroes(opts, bs->open_flags, &local_err); | |
1917 | if (local_err) { | |
1918 | error_propagate(errp, local_err); | |
1919 | ret = -EINVAL; | |
1920 | goto fail_opts; | |
692e01a2 KW |
1921 | } |
1922 | ||
c2ad1b0c KW |
1923 | if (filename != NULL) { |
1924 | pstrcpy(bs->filename, sizeof(bs->filename), filename); | |
1925 | } else { | |
1926 | bs->filename[0] = '\0'; | |
1927 | } | |
91af7014 | 1928 | pstrcpy(bs->exact_filename, sizeof(bs->exact_filename), bs->filename); |
57915332 | 1929 | |
66f82cee | 1930 | /* Open the image, either directly or using a protocol */ |
82dc8b41 | 1931 | open_flags = bdrv_open_flags(bs, bs->open_flags); |
01a56501 | 1932 | node_name = qemu_opt_get(opts, "node-name"); |
57915332 | 1933 | |
01a56501 KW |
1934 | assert(!drv->bdrv_file_open || file == NULL); |
1935 | ret = bdrv_open_driver(bs, drv, node_name, options, open_flags, errp); | |
51762288 | 1936 | if (ret < 0) { |
01a56501 | 1937 | goto fail_opts; |
3baca891 KW |
1938 | } |
1939 | ||
18edf289 | 1940 | qemu_opts_del(opts); |
57915332 KW |
1941 | return 0; |
1942 | ||
18edf289 KW |
1943 | fail_opts: |
1944 | qemu_opts_del(opts); | |
57915332 KW |
1945 | return ret; |
1946 | } | |
1947 | ||
5e5c4f63 KW |
1948 | static QDict *parse_json_filename(const char *filename, Error **errp) |
1949 | { | |
1950 | QObject *options_obj; | |
1951 | QDict *options; | |
1952 | int ret; | |
bdb73476 | 1953 | GLOBAL_STATE_CODE(); |
5e5c4f63 KW |
1954 | |
1955 | ret = strstart(filename, "json:", &filename); | |
1956 | assert(ret); | |
1957 | ||
5577fff7 | 1958 | options_obj = qobject_from_json(filename, errp); |
5e5c4f63 | 1959 | if (!options_obj) { |
5577fff7 | 1960 | error_prepend(errp, "Could not parse the JSON options: "); |
5e5c4f63 KW |
1961 | return NULL; |
1962 | } | |
1963 | ||
7dc847eb | 1964 | options = qobject_to(QDict, options_obj); |
ca6b6e1e | 1965 | if (!options) { |
cb3e7f08 | 1966 | qobject_unref(options_obj); |
5e5c4f63 KW |
1967 | error_setg(errp, "Invalid JSON object given"); |
1968 | return NULL; | |
1969 | } | |
1970 | ||
5e5c4f63 KW |
1971 | qdict_flatten(options); |
1972 | ||
1973 | return options; | |
1974 | } | |
1975 | ||
de3b53f0 KW |
1976 | static void parse_json_protocol(QDict *options, const char **pfilename, |
1977 | Error **errp) | |
1978 | { | |
1979 | QDict *json_options; | |
1980 | Error *local_err = NULL; | |
bdb73476 | 1981 | GLOBAL_STATE_CODE(); |
de3b53f0 KW |
1982 | |
1983 | /* Parse json: pseudo-protocol */ | |
1984 | if (!*pfilename || !g_str_has_prefix(*pfilename, "json:")) { | |
1985 | return; | |
1986 | } | |
1987 | ||
1988 | json_options = parse_json_filename(*pfilename, &local_err); | |
1989 | if (local_err) { | |
1990 | error_propagate(errp, local_err); | |
1991 | return; | |
1992 | } | |
1993 | ||
1994 | /* Options given in the filename have lower priority than options | |
1995 | * specified directly */ | |
1996 | qdict_join(options, json_options, false); | |
cb3e7f08 | 1997 | qobject_unref(json_options); |
de3b53f0 KW |
1998 | *pfilename = NULL; |
1999 | } | |
2000 | ||
b6ce07aa | 2001 | /* |
f54120ff KW |
2002 | * Fills in default options for opening images and converts the legacy |
2003 | * filename/flags pair to option QDict entries. | |
53a29513 HR |
2004 | * The BDRV_O_PROTOCOL flag in *flags will be set or cleared accordingly if a |
2005 | * block driver has been specified explicitly. | |
b6ce07aa | 2006 | */ |
de3b53f0 | 2007 | static int bdrv_fill_options(QDict **options, const char *filename, |
053e1578 | 2008 | int *flags, Error **errp) |
ea2384d3 | 2009 | { |
c2ad1b0c | 2010 | const char *drvname; |
53a29513 | 2011 | bool protocol = *flags & BDRV_O_PROTOCOL; |
e3fa4bfa | 2012 | bool parse_filename = false; |
053e1578 | 2013 | BlockDriver *drv = NULL; |
34b5d2c6 | 2014 | Error *local_err = NULL; |
83f64091 | 2015 | |
da359909 EGE |
2016 | GLOBAL_STATE_CODE(); |
2017 | ||
129c7d1c MA |
2018 | /* |
2019 | * Caution: while qdict_get_try_str() is fine, getting non-string | |
2020 | * types would require more care. When @options come from | |
2021 | * -blockdev or blockdev_add, its members are typed according to | |
2022 | * the QAPI schema, but when they come from -drive, they're all | |
2023 | * QString. | |
2024 | */ | |
53a29513 | 2025 | drvname = qdict_get_try_str(*options, "driver"); |
053e1578 HR |
2026 | if (drvname) { |
2027 | drv = bdrv_find_format(drvname); | |
2028 | if (!drv) { | |
2029 | error_setg(errp, "Unknown driver '%s'", drvname); | |
2030 | return -ENOENT; | |
2031 | } | |
2032 | /* If the user has explicitly specified the driver, this choice should | |
2033 | * override the BDRV_O_PROTOCOL flag */ | |
2034 | protocol = drv->bdrv_file_open; | |
53a29513 HR |
2035 | } |
2036 | ||
2037 | if (protocol) { | |
2038 | *flags |= BDRV_O_PROTOCOL; | |
2039 | } else { | |
2040 | *flags &= ~BDRV_O_PROTOCOL; | |
2041 | } | |
2042 | ||
91a097e7 KW |
2043 | /* Translate cache options from flags into options */ |
2044 | update_options_from_flags(*options, *flags); | |
2045 | ||
035fccdf | 2046 | /* Fetch the file name from the options QDict if necessary */ |
17b005f1 | 2047 | if (protocol && filename) { |
f54120ff | 2048 | if (!qdict_haskey(*options, "filename")) { |
46f5ac20 | 2049 | qdict_put_str(*options, "filename", filename); |
f54120ff KW |
2050 | parse_filename = true; |
2051 | } else { | |
2052 | error_setg(errp, "Can't specify 'file' and 'filename' options at " | |
2053 | "the same time"); | |
2054 | return -EINVAL; | |
2055 | } | |
035fccdf KW |
2056 | } |
2057 | ||
c2ad1b0c | 2058 | /* Find the right block driver */ |
129c7d1c | 2059 | /* See cautionary note on accessing @options above */ |
f54120ff | 2060 | filename = qdict_get_try_str(*options, "filename"); |
f54120ff | 2061 | |
053e1578 HR |
2062 | if (!drvname && protocol) { |
2063 | if (filename) { | |
2064 | drv = bdrv_find_protocol(filename, parse_filename, errp); | |
17b005f1 | 2065 | if (!drv) { |
053e1578 | 2066 | return -EINVAL; |
17b005f1 | 2067 | } |
053e1578 HR |
2068 | |
2069 | drvname = drv->format_name; | |
46f5ac20 | 2070 | qdict_put_str(*options, "driver", drvname); |
053e1578 HR |
2071 | } else { |
2072 | error_setg(errp, "Must specify either driver or file"); | |
2073 | return -EINVAL; | |
98289620 | 2074 | } |
c2ad1b0c KW |
2075 | } |
2076 | ||
17b005f1 | 2077 | assert(drv || !protocol); |
c2ad1b0c | 2078 | |
f54120ff | 2079 | /* Driver-specific filename parsing */ |
17b005f1 | 2080 | if (drv && drv->bdrv_parse_filename && parse_filename) { |
5acd9d81 | 2081 | drv->bdrv_parse_filename(filename, *options, &local_err); |
84d18f06 | 2082 | if (local_err) { |
34b5d2c6 | 2083 | error_propagate(errp, local_err); |
f54120ff | 2084 | return -EINVAL; |
6963a30d | 2085 | } |
cd5d031e HR |
2086 | |
2087 | if (!drv->bdrv_needs_filename) { | |
2088 | qdict_del(*options, "filename"); | |
cd5d031e | 2089 | } |
6963a30d KW |
2090 | } |
2091 | ||
f54120ff KW |
2092 | return 0; |
2093 | } | |
2094 | ||
148eb13c KW |
2095 | typedef struct BlockReopenQueueEntry { |
2096 | bool prepared; | |
69b736e7 | 2097 | bool perms_checked; |
148eb13c | 2098 | BDRVReopenState state; |
859443b0 | 2099 | QTAILQ_ENTRY(BlockReopenQueueEntry) entry; |
148eb13c KW |
2100 | } BlockReopenQueueEntry; |
2101 | ||
2102 | /* | |
2103 | * Return the flags that @bs will have after the reopens in @q have | |
2104 | * successfully completed. If @q is NULL (or @bs is not contained in @q), | |
2105 | * return the current flags. | |
2106 | */ | |
2107 | static int bdrv_reopen_get_flags(BlockReopenQueue *q, BlockDriverState *bs) | |
2108 | { | |
2109 | BlockReopenQueueEntry *entry; | |
2110 | ||
2111 | if (q != NULL) { | |
859443b0 | 2112 | QTAILQ_FOREACH(entry, q, entry) { |
148eb13c KW |
2113 | if (entry->state.bs == bs) { |
2114 | return entry->state.flags; | |
2115 | } | |
2116 | } | |
2117 | } | |
2118 | ||
2119 | return bs->open_flags; | |
2120 | } | |
2121 | ||
2122 | /* Returns whether the image file can be written to after the reopen queue @q | |
2123 | * has been successfully applied, or right now if @q is NULL. */ | |
cc022140 HR |
2124 | static bool bdrv_is_writable_after_reopen(BlockDriverState *bs, |
2125 | BlockReopenQueue *q) | |
148eb13c KW |
2126 | { |
2127 | int flags = bdrv_reopen_get_flags(q, bs); | |
2128 | ||
2129 | return (flags & (BDRV_O_RDWR | BDRV_O_INACTIVE)) == BDRV_O_RDWR; | |
2130 | } | |
2131 | ||
cc022140 HR |
2132 | /* |
2133 | * Return whether the BDS can be written to. This is not necessarily | |
2134 | * the same as !bdrv_is_read_only(bs), as inactivated images may not | |
2135 | * be written to but do not count as read-only images. | |
2136 | */ | |
2137 | bool bdrv_is_writable(BlockDriverState *bs) | |
2138 | { | |
384a48fb | 2139 | IO_CODE(); |
cc022140 HR |
2140 | return bdrv_is_writable_after_reopen(bs, NULL); |
2141 | } | |
2142 | ||
3bf416ba VSO |
2143 | static char *bdrv_child_user_desc(BdrvChild *c) |
2144 | { | |
f0c28327 | 2145 | GLOBAL_STATE_CODE(); |
da261b69 | 2146 | return c->klass->get_parent_desc(c); |
3bf416ba VSO |
2147 | } |
2148 | ||
30ebb9aa VSO |
2149 | /* |
2150 | * Check that @a allows everything that @b needs. @a and @b must reference same | |
2151 | * child node. | |
2152 | */ | |
3bf416ba VSO |
2153 | static bool bdrv_a_allow_b(BdrvChild *a, BdrvChild *b, Error **errp) |
2154 | { | |
30ebb9aa VSO |
2155 | const char *child_bs_name; |
2156 | g_autofree char *a_user = NULL; | |
2157 | g_autofree char *b_user = NULL; | |
2158 | g_autofree char *perms = NULL; | |
2159 | ||
2160 | assert(a->bs); | |
2161 | assert(a->bs == b->bs); | |
862fded9 | 2162 | GLOBAL_STATE_CODE(); |
3bf416ba VSO |
2163 | |
2164 | if ((b->perm & a->shared_perm) == b->perm) { | |
2165 | return true; | |
2166 | } | |
2167 | ||
30ebb9aa VSO |
2168 | child_bs_name = bdrv_get_node_name(b->bs); |
2169 | a_user = bdrv_child_user_desc(a); | |
2170 | b_user = bdrv_child_user_desc(b); | |
2171 | perms = bdrv_perm_names(b->perm & ~a->shared_perm); | |
2172 | ||
2173 | error_setg(errp, "Permission conflict on node '%s': permissions '%s' are " | |
2174 | "both required by %s (uses node '%s' as '%s' child) and " | |
2175 | "unshared by %s (uses node '%s' as '%s' child).", | |
2176 | child_bs_name, perms, | |
2177 | b_user, child_bs_name, b->name, | |
2178 | a_user, child_bs_name, a->name); | |
3bf416ba VSO |
2179 | |
2180 | return false; | |
2181 | } | |
2182 | ||
9397c14f | 2183 | static bool bdrv_parent_perms_conflict(BlockDriverState *bs, Error **errp) |
3bf416ba VSO |
2184 | { |
2185 | BdrvChild *a, *b; | |
862fded9 | 2186 | GLOBAL_STATE_CODE(); |
3bf416ba VSO |
2187 | |
2188 | /* | |
2189 | * During the loop we'll look at each pair twice. That's correct because | |
2190 | * bdrv_a_allow_b() is asymmetric and we should check each pair in both | |
2191 | * directions. | |
2192 | */ | |
2193 | QLIST_FOREACH(a, &bs->parents, next_parent) { | |
2194 | QLIST_FOREACH(b, &bs->parents, next_parent) { | |
9397c14f | 2195 | if (a == b) { |
3bf416ba VSO |
2196 | continue; |
2197 | } | |
2198 | ||
2199 | if (!bdrv_a_allow_b(a, b, errp)) { | |
2200 | return true; | |
2201 | } | |
2202 | } | |
2203 | } | |
2204 | ||
2205 | return false; | |
2206 | } | |
2207 | ||
ffd1a5a2 | 2208 | static void bdrv_child_perm(BlockDriverState *bs, BlockDriverState *child_bs, |
e5d8a406 HR |
2209 | BdrvChild *c, BdrvChildRole role, |
2210 | BlockReopenQueue *reopen_queue, | |
ffd1a5a2 FZ |
2211 | uint64_t parent_perm, uint64_t parent_shared, |
2212 | uint64_t *nperm, uint64_t *nshared) | |
2213 | { | |
0b3ca76e | 2214 | assert(bs->drv && bs->drv->bdrv_child_perm); |
da359909 | 2215 | GLOBAL_STATE_CODE(); |
e5d8a406 | 2216 | bs->drv->bdrv_child_perm(bs, c, role, reopen_queue, |
0b3ca76e AG |
2217 | parent_perm, parent_shared, |
2218 | nperm, nshared); | |
e0995dc3 | 2219 | /* TODO Take force_share from reopen_queue */ |
ffd1a5a2 FZ |
2220 | if (child_bs && child_bs->force_share) { |
2221 | *nshared = BLK_PERM_ALL; | |
2222 | } | |
2223 | } | |
2224 | ||
bd57f8f7 VSO |
2225 | /* |
2226 | * Adds the whole subtree of @bs (including @bs itself) to the @list (except for | |
2227 | * nodes that are already in the @list, of course) so that final list is | |
2228 | * topologically sorted. Return the result (GSList @list object is updated, so | |
2229 | * don't use old reference after function call). | |
2230 | * | |
2231 | * On function start @list must be already topologically sorted and for any node | |
2232 | * in the @list the whole subtree of the node must be in the @list as well. The | |
2233 | * simplest way to satisfy this criteria: use only result of | |
2234 | * bdrv_topological_dfs() or NULL as @list parameter. | |
2235 | */ | |
2236 | static GSList *bdrv_topological_dfs(GSList *list, GHashTable *found, | |
2237 | BlockDriverState *bs) | |
2238 | { | |
2239 | BdrvChild *child; | |
2240 | g_autoptr(GHashTable) local_found = NULL; | |
2241 | ||
bdb73476 EGE |
2242 | GLOBAL_STATE_CODE(); |
2243 | ||
bd57f8f7 VSO |
2244 | if (!found) { |
2245 | assert(!list); | |
2246 | found = local_found = g_hash_table_new(NULL, NULL); | |
2247 | } | |
2248 | ||
2249 | if (g_hash_table_contains(found, bs)) { | |
2250 | return list; | |
2251 | } | |
2252 | g_hash_table_add(found, bs); | |
2253 | ||
2254 | QLIST_FOREACH(child, &bs->children, next) { | |
2255 | list = bdrv_topological_dfs(list, found, child->bs); | |
2256 | } | |
2257 | ||
2258 | return g_slist_prepend(list, bs); | |
2259 | } | |
2260 | ||
ecb776bd VSO |
2261 | typedef struct BdrvChildSetPermState { |
2262 | BdrvChild *child; | |
2263 | uint64_t old_perm; | |
2264 | uint64_t old_shared_perm; | |
2265 | } BdrvChildSetPermState; | |
b0defa83 VSO |
2266 | |
2267 | static void bdrv_child_set_perm_abort(void *opaque) | |
2268 | { | |
ecb776bd VSO |
2269 | BdrvChildSetPermState *s = opaque; |
2270 | ||
862fded9 EGE |
2271 | GLOBAL_STATE_CODE(); |
2272 | ||
ecb776bd VSO |
2273 | s->child->perm = s->old_perm; |
2274 | s->child->shared_perm = s->old_shared_perm; | |
b0defa83 VSO |
2275 | } |
2276 | ||
2277 | static TransactionActionDrv bdrv_child_set_pem_drv = { | |
2278 | .abort = bdrv_child_set_perm_abort, | |
ecb776bd | 2279 | .clean = g_free, |
b0defa83 VSO |
2280 | }; |
2281 | ||
ecb776bd VSO |
2282 | static void bdrv_child_set_perm(BdrvChild *c, uint64_t perm, |
2283 | uint64_t shared, Transaction *tran) | |
b0defa83 | 2284 | { |
ecb776bd | 2285 | BdrvChildSetPermState *s = g_new(BdrvChildSetPermState, 1); |
862fded9 | 2286 | GLOBAL_STATE_CODE(); |
ecb776bd VSO |
2287 | |
2288 | *s = (BdrvChildSetPermState) { | |
2289 | .child = c, | |
2290 | .old_perm = c->perm, | |
2291 | .old_shared_perm = c->shared_perm, | |
2292 | }; | |
b0defa83 VSO |
2293 | |
2294 | c->perm = perm; | |
2295 | c->shared_perm = shared; | |
2296 | ||
ecb776bd | 2297 | tran_add(tran, &bdrv_child_set_pem_drv, s); |
b0defa83 VSO |
2298 | } |
2299 | ||
2513ef59 VSO |
2300 | static void bdrv_drv_set_perm_commit(void *opaque) |
2301 | { | |
2302 | BlockDriverState *bs = opaque; | |
2303 | uint64_t cumulative_perms, cumulative_shared_perms; | |
da359909 | 2304 | GLOBAL_STATE_CODE(); |
2513ef59 VSO |
2305 | |
2306 | if (bs->drv->bdrv_set_perm) { | |
2307 | bdrv_get_cumulative_perm(bs, &cumulative_perms, | |
2308 | &cumulative_shared_perms); | |
2309 | bs->drv->bdrv_set_perm(bs, cumulative_perms, cumulative_shared_perms); | |
2310 | } | |
2311 | } | |
2312 | ||
2313 | static void bdrv_drv_set_perm_abort(void *opaque) | |
2314 | { | |
2315 | BlockDriverState *bs = opaque; | |
da359909 | 2316 | GLOBAL_STATE_CODE(); |
2513ef59 VSO |
2317 | |
2318 | if (bs->drv->bdrv_abort_perm_update) { | |
2319 | bs->drv->bdrv_abort_perm_update(bs); | |
2320 | } | |
2321 | } | |
2322 | ||
2323 | TransactionActionDrv bdrv_drv_set_perm_drv = { | |
2324 | .abort = bdrv_drv_set_perm_abort, | |
2325 | .commit = bdrv_drv_set_perm_commit, | |
2326 | }; | |
2327 | ||
2328 | static int bdrv_drv_set_perm(BlockDriverState *bs, uint64_t perm, | |
2329 | uint64_t shared_perm, Transaction *tran, | |
2330 | Error **errp) | |
2331 | { | |
da359909 | 2332 | GLOBAL_STATE_CODE(); |
2513ef59 VSO |
2333 | if (!bs->drv) { |
2334 | return 0; | |
2335 | } | |
2336 | ||
2337 | if (bs->drv->bdrv_check_perm) { | |
2338 | int ret = bs->drv->bdrv_check_perm(bs, perm, shared_perm, errp); | |
2339 | if (ret < 0) { | |
2340 | return ret; | |
2341 | } | |
2342 | } | |
2343 | ||
2344 | if (tran) { | |
2345 | tran_add(tran, &bdrv_drv_set_perm_drv, bs); | |
2346 | } | |
2347 | ||
2348 | return 0; | |
2349 | } | |
2350 | ||
0978623e VSO |
2351 | typedef struct BdrvReplaceChildState { |
2352 | BdrvChild *child; | |
2353 | BlockDriverState *old_bs; | |
2354 | } BdrvReplaceChildState; | |
2355 | ||
2356 | static void bdrv_replace_child_commit(void *opaque) | |
2357 | { | |
2358 | BdrvReplaceChildState *s = opaque; | |
bdb73476 | 2359 | GLOBAL_STATE_CODE(); |
0978623e VSO |
2360 | |
2361 | bdrv_unref(s->old_bs); | |
2362 | } | |
2363 | ||
2364 | static void bdrv_replace_child_abort(void *opaque) | |
2365 | { | |
2366 | BdrvReplaceChildState *s = opaque; | |
2367 | BlockDriverState *new_bs = s->child->bs; | |
2368 | ||
bdb73476 | 2369 | GLOBAL_STATE_CODE(); |
0f0b1e29 | 2370 | /* old_bs reference is transparently moved from @s to @s->child */ |
23987471 KW |
2371 | if (!s->child->bs) { |
2372 | /* | |
2373 | * The parents were undrained when removing old_bs from the child. New | |
2374 | * requests can't have been made, though, because the child was empty. | |
2375 | * | |
2376 | * TODO Make bdrv_replace_child_noperm() transactionable to avoid | |
2377 | * undraining the parent in the first place. Once this is done, having | |
2378 | * new_bs drained when calling bdrv_replace_child_tran() is not a | |
2379 | * requirement any more. | |
2380 | */ | |
606ed756 | 2381 | bdrv_parent_drained_begin_single(s->child); |
23987471 KW |
2382 | assert(!bdrv_parent_drained_poll_single(s->child)); |
2383 | } | |
2384 | assert(s->child->quiesced_parent); | |
544acc7d | 2385 | bdrv_replace_child_noperm(s->child, s->old_bs); |
0978623e VSO |
2386 | bdrv_unref(new_bs); |
2387 | } | |
2388 | ||
2389 | static TransactionActionDrv bdrv_replace_child_drv = { | |
2390 | .commit = bdrv_replace_child_commit, | |
2391 | .abort = bdrv_replace_child_abort, | |
2392 | .clean = g_free, | |
2393 | }; | |
2394 | ||
2395 | /* | |
4bf021db | 2396 | * bdrv_replace_child_tran |
0978623e VSO |
2397 | * |
2398 | * Note: real unref of old_bs is done only on commit. | |
4bf021db | 2399 | * |
23987471 KW |
2400 | * Both @child->bs and @new_bs (if non-NULL) must be drained. @new_bs must be |
2401 | * kept drained until the transaction is completed. | |
2402 | * | |
4bf021db | 2403 | * The function doesn't update permissions, caller is responsible for this. |
0978623e | 2404 | */ |
0f0b1e29 | 2405 | static void bdrv_replace_child_tran(BdrvChild *child, BlockDriverState *new_bs, |
4eba825a | 2406 | Transaction *tran) |
0978623e VSO |
2407 | { |
2408 | BdrvReplaceChildState *s = g_new(BdrvReplaceChildState, 1); | |
23987471 KW |
2409 | |
2410 | assert(child->quiesced_parent); | |
2411 | assert(!new_bs || new_bs->quiesce_counter); | |
2412 | ||
0978623e | 2413 | *s = (BdrvReplaceChildState) { |
0f0b1e29 VSO |
2414 | .child = child, |
2415 | .old_bs = child->bs, | |
0978623e VSO |
2416 | }; |
2417 | tran_add(tran, &bdrv_replace_child_drv, s); | |
2418 | ||
2419 | if (new_bs) { | |
2420 | bdrv_ref(new_bs); | |
2421 | } | |
544acc7d | 2422 | bdrv_replace_child_noperm(child, new_bs); |
0f0b1e29 | 2423 | /* old_bs reference is transparently moved from @child to @s */ |
0978623e VSO |
2424 | } |
2425 | ||
33a610c3 | 2426 | /* |
c20555e1 VSO |
2427 | * Refresh permissions in @bs subtree. The function is intended to be called |
2428 | * after some graph modification that was done without permission update. | |
33a610c3 | 2429 | */ |
c20555e1 VSO |
2430 | static int bdrv_node_refresh_perm(BlockDriverState *bs, BlockReopenQueue *q, |
2431 | Transaction *tran, Error **errp) | |
33a610c3 KW |
2432 | { |
2433 | BlockDriver *drv = bs->drv; | |
2434 | BdrvChild *c; | |
2435 | int ret; | |
c20555e1 | 2436 | uint64_t cumulative_perms, cumulative_shared_perms; |
862fded9 | 2437 | GLOBAL_STATE_CODE(); |
c20555e1 VSO |
2438 | |
2439 | bdrv_get_cumulative_perm(bs, &cumulative_perms, &cumulative_shared_perms); | |
33a610c3 KW |
2440 | |
2441 | /* Write permissions never work with read-only images */ | |
2442 | if ((cumulative_perms & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)) && | |
cc022140 | 2443 | !bdrv_is_writable_after_reopen(bs, q)) |
33a610c3 | 2444 | { |
481e0eee HR |
2445 | if (!bdrv_is_writable_after_reopen(bs, NULL)) { |
2446 | error_setg(errp, "Block node is read-only"); | |
2447 | } else { | |
c20555e1 VSO |
2448 | error_setg(errp, "Read-only block node '%s' cannot support " |
2449 | "read-write users", bdrv_get_node_name(bs)); | |
481e0eee HR |
2450 | } |
2451 | ||
33a610c3 KW |
2452 | return -EPERM; |
2453 | } | |
2454 | ||
9c60a5d1 KW |
2455 | /* |
2456 | * Unaligned requests will automatically be aligned to bl.request_alignment | |
2457 | * and without RESIZE we can't extend requests to write to space beyond the | |
2458 | * end of the image, so it's required that the image size is aligned. | |
2459 | */ | |
2460 | if ((cumulative_perms & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)) && | |
2461 | !(cumulative_perms & BLK_PERM_RESIZE)) | |
2462 | { | |
2463 | if ((bs->total_sectors * BDRV_SECTOR_SIZE) % bs->bl.request_alignment) { | |
2464 | error_setg(errp, "Cannot get 'write' permission without 'resize': " | |
2465 | "Image size is not a multiple of request " | |
2466 | "alignment"); | |
2467 | return -EPERM; | |
2468 | } | |
2469 | } | |
2470 | ||
33a610c3 KW |
2471 | /* Check this node */ |
2472 | if (!drv) { | |
2473 | return 0; | |
2474 | } | |
2475 | ||
b1d2bbeb | 2476 | ret = bdrv_drv_set_perm(bs, cumulative_perms, cumulative_shared_perms, tran, |
2513ef59 VSO |
2477 | errp); |
2478 | if (ret < 0) { | |
2479 | return ret; | |
33a610c3 KW |
2480 | } |
2481 | ||
78e421c9 | 2482 | /* Drivers that never have children can omit .bdrv_child_perm() */ |
33a610c3 | 2483 | if (!drv->bdrv_child_perm) { |
78e421c9 | 2484 | assert(QLIST_EMPTY(&bs->children)); |
33a610c3 KW |
2485 | return 0; |
2486 | } | |
2487 | ||
2488 | /* Check all children */ | |
2489 | QLIST_FOREACH(c, &bs->children, next) { | |
2490 | uint64_t cur_perm, cur_shared; | |
9eab1544 | 2491 | |
e5d8a406 | 2492 | bdrv_child_perm(bs, c->bs, c, c->role, q, |
ffd1a5a2 FZ |
2493 | cumulative_perms, cumulative_shared_perms, |
2494 | &cur_perm, &cur_shared); | |
ecb776bd | 2495 | bdrv_child_set_perm(c, cur_perm, cur_shared, tran); |
bd57f8f7 VSO |
2496 | } |
2497 | ||
2498 | return 0; | |
2499 | } | |
3ef45e02 | 2500 | |
fb0ff4d1 VSO |
2501 | /* |
2502 | * @list is a product of bdrv_topological_dfs() (may be called several times) - | |
2503 | * a topologically sorted subgraph. | |
2504 | */ | |
2505 | static int bdrv_do_refresh_perms(GSList *list, BlockReopenQueue *q, | |
2506 | Transaction *tran, Error **errp) | |
bd57f8f7 VSO |
2507 | { |
2508 | int ret; | |
b1d2bbeb | 2509 | BlockDriverState *bs; |
862fded9 | 2510 | GLOBAL_STATE_CODE(); |
bd57f8f7 | 2511 | |
b1d2bbeb VSO |
2512 | for ( ; list; list = list->next) { |
2513 | bs = list->data; | |
2514 | ||
9397c14f | 2515 | if (bdrv_parent_perms_conflict(bs, errp)) { |
b1d2bbeb | 2516 | return -EINVAL; |
bd57f8f7 VSO |
2517 | } |
2518 | ||
c20555e1 | 2519 | ret = bdrv_node_refresh_perm(bs, q, tran, errp); |
33a610c3 KW |
2520 | if (ret < 0) { |
2521 | return ret; | |
2522 | } | |
2523 | } | |
2524 | ||
2525 | return 0; | |
2526 | } | |
2527 | ||
fb0ff4d1 VSO |
2528 | /* |
2529 | * @list is any list of nodes. List is completed by all subtrees and | |
2530 | * topologically sorted. It's not a problem if some node occurs in the @list | |
2531 | * several times. | |
2532 | */ | |
2533 | static int bdrv_list_refresh_perms(GSList *list, BlockReopenQueue *q, | |
2534 | Transaction *tran, Error **errp) | |
2535 | { | |
2536 | g_autoptr(GHashTable) found = g_hash_table_new(NULL, NULL); | |
2537 | g_autoptr(GSList) refresh_list = NULL; | |
2538 | ||
2539 | for ( ; list; list = list->next) { | |
2540 | refresh_list = bdrv_topological_dfs(refresh_list, found, list->data); | |
2541 | } | |
2542 | ||
2543 | return bdrv_do_refresh_perms(refresh_list, q, tran, errp); | |
2544 | } | |
2545 | ||
c7a0f2be KW |
2546 | void bdrv_get_cumulative_perm(BlockDriverState *bs, uint64_t *perm, |
2547 | uint64_t *shared_perm) | |
33a610c3 KW |
2548 | { |
2549 | BdrvChild *c; | |
2550 | uint64_t cumulative_perms = 0; | |
2551 | uint64_t cumulative_shared_perms = BLK_PERM_ALL; | |
2552 | ||
b4ad82aa EGE |
2553 | GLOBAL_STATE_CODE(); |
2554 | ||
33a610c3 KW |
2555 | QLIST_FOREACH(c, &bs->parents, next_parent) { |
2556 | cumulative_perms |= c->perm; | |
2557 | cumulative_shared_perms &= c->shared_perm; | |
2558 | } | |
2559 | ||
2560 | *perm = cumulative_perms; | |
2561 | *shared_perm = cumulative_shared_perms; | |
2562 | } | |
2563 | ||
5176196c | 2564 | char *bdrv_perm_names(uint64_t perm) |
d083319f KW |
2565 | { |
2566 | struct perm_name { | |
2567 | uint64_t perm; | |
2568 | const char *name; | |
2569 | } permissions[] = { | |
2570 | { BLK_PERM_CONSISTENT_READ, "consistent read" }, | |
2571 | { BLK_PERM_WRITE, "write" }, | |
2572 | { BLK_PERM_WRITE_UNCHANGED, "write unchanged" }, | |
2573 | { BLK_PERM_RESIZE, "resize" }, | |
d083319f KW |
2574 | { 0, NULL } |
2575 | }; | |
2576 | ||
e2a7423a | 2577 | GString *result = g_string_sized_new(30); |
d083319f KW |
2578 | struct perm_name *p; |
2579 | ||
2580 | for (p = permissions; p->name; p++) { | |
2581 | if (perm & p->perm) { | |
e2a7423a AG |
2582 | if (result->len > 0) { |
2583 | g_string_append(result, ", "); | |
2584 | } | |
2585 | g_string_append(result, p->name); | |
d083319f KW |
2586 | } |
2587 | } | |
2588 | ||
e2a7423a | 2589 | return g_string_free(result, FALSE); |
d083319f KW |
2590 | } |
2591 | ||
33a610c3 | 2592 | |
f1316edb VSO |
2593 | /* @tran is allowed to be NULL. In this case no rollback is possible */ |
2594 | static int bdrv_refresh_perms(BlockDriverState *bs, Transaction *tran, | |
2595 | Error **errp) | |
bb87e4d1 VSO |
2596 | { |
2597 | int ret; | |
f1316edb | 2598 | Transaction *local_tran = NULL; |
b1d2bbeb | 2599 | g_autoptr(GSList) list = bdrv_topological_dfs(NULL, NULL, bs); |
862fded9 | 2600 | GLOBAL_STATE_CODE(); |
bb87e4d1 | 2601 | |
f1316edb VSO |
2602 | if (!tran) { |
2603 | tran = local_tran = tran_new(); | |
2604 | } | |
2605 | ||
fb0ff4d1 | 2606 | ret = bdrv_do_refresh_perms(list, NULL, tran, errp); |
f1316edb VSO |
2607 | |
2608 | if (local_tran) { | |
2609 | tran_finalize(local_tran, ret); | |
2610 | } | |
bb87e4d1 | 2611 | |
b1d2bbeb | 2612 | return ret; |
bb87e4d1 VSO |
2613 | } |
2614 | ||
33a610c3 KW |
2615 | int bdrv_child_try_set_perm(BdrvChild *c, uint64_t perm, uint64_t shared, |
2616 | Error **errp) | |
2617 | { | |
1046779e | 2618 | Error *local_err = NULL; |
83928dc4 | 2619 | Transaction *tran = tran_new(); |
33a610c3 KW |
2620 | int ret; |
2621 | ||
b4ad82aa EGE |
2622 | GLOBAL_STATE_CODE(); |
2623 | ||
ecb776bd | 2624 | bdrv_child_set_perm(c, perm, shared, tran); |
83928dc4 | 2625 | |
f1316edb | 2626 | ret = bdrv_refresh_perms(c->bs, tran, &local_err); |
83928dc4 VSO |
2627 | |
2628 | tran_finalize(tran, ret); | |
2629 | ||
33a610c3 | 2630 | if (ret < 0) { |
071b474f VSO |
2631 | if ((perm & ~c->perm) || (c->shared_perm & ~shared)) { |
2632 | /* tighten permissions */ | |
1046779e HR |
2633 | error_propagate(errp, local_err); |
2634 | } else { | |
2635 | /* | |
2636 | * Our caller may intend to only loosen restrictions and | |
2637 | * does not expect this function to fail. Errors are not | |
2638 | * fatal in such a case, so we can just hide them from our | |
2639 | * caller. | |
2640 | */ | |
2641 | error_free(local_err); | |
2642 | ret = 0; | |
2643 | } | |
33a610c3 KW |
2644 | } |
2645 | ||
83928dc4 | 2646 | return ret; |
d5e6f437 KW |
2647 | } |
2648 | ||
c1087f12 HR |
2649 | int bdrv_child_refresh_perms(BlockDriverState *bs, BdrvChild *c, Error **errp) |
2650 | { | |
2651 | uint64_t parent_perms, parent_shared; | |
2652 | uint64_t perms, shared; | |
2653 | ||
b4ad82aa EGE |
2654 | GLOBAL_STATE_CODE(); |
2655 | ||
c1087f12 | 2656 | bdrv_get_cumulative_perm(bs, &parent_perms, &parent_shared); |
e5d8a406 | 2657 | bdrv_child_perm(bs, c->bs, c, c->role, NULL, |
bf8e925e | 2658 | parent_perms, parent_shared, &perms, &shared); |
c1087f12 HR |
2659 | |
2660 | return bdrv_child_try_set_perm(c, perms, shared, errp); | |
2661 | } | |
2662 | ||
87278af1 HR |
2663 | /* |
2664 | * Default implementation for .bdrv_child_perm() for block filters: | |
2665 | * Forward CONSISTENT_READ, WRITE, WRITE_UNCHANGED, and RESIZE to the | |
2666 | * filtered child. | |
2667 | */ | |
2668 | static void bdrv_filter_default_perms(BlockDriverState *bs, BdrvChild *c, | |
87278af1 HR |
2669 | BdrvChildRole role, |
2670 | BlockReopenQueue *reopen_queue, | |
2671 | uint64_t perm, uint64_t shared, | |
2672 | uint64_t *nperm, uint64_t *nshared) | |
6a1b9ee1 | 2673 | { |
862fded9 | 2674 | GLOBAL_STATE_CODE(); |
e444fa83 KW |
2675 | *nperm = perm & DEFAULT_PERM_PASSTHROUGH; |
2676 | *nshared = (shared & DEFAULT_PERM_PASSTHROUGH) | DEFAULT_PERM_UNCHANGED; | |
6a1b9ee1 KW |
2677 | } |
2678 | ||
70082db4 | 2679 | static void bdrv_default_perms_for_cow(BlockDriverState *bs, BdrvChild *c, |
70082db4 HR |
2680 | BdrvChildRole role, |
2681 | BlockReopenQueue *reopen_queue, | |
2682 | uint64_t perm, uint64_t shared, | |
2683 | uint64_t *nperm, uint64_t *nshared) | |
2684 | { | |
e5d8a406 | 2685 | assert(role & BDRV_CHILD_COW); |
862fded9 | 2686 | GLOBAL_STATE_CODE(); |
70082db4 HR |
2687 | |
2688 | /* | |
2689 | * We want consistent read from backing files if the parent needs it. | |
2690 | * No other operations are performed on backing files. | |
2691 | */ | |
2692 | perm &= BLK_PERM_CONSISTENT_READ; | |
2693 | ||
2694 | /* | |
2695 | * If the parent can deal with changing data, we're okay with a | |
2696 | * writable and resizable backing file. | |
2697 | * TODO Require !(perm & BLK_PERM_CONSISTENT_READ), too? | |
2698 | */ | |
2699 | if (shared & BLK_PERM_WRITE) { | |
2700 | shared = BLK_PERM_WRITE | BLK_PERM_RESIZE; | |
2701 | } else { | |
2702 | shared = 0; | |
2703 | } | |
2704 | ||
64631f36 | 2705 | shared |= BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE_UNCHANGED; |
70082db4 HR |
2706 | |
2707 | if (bs->open_flags & BDRV_O_INACTIVE) { | |
2708 | shared |= BLK_PERM_WRITE | BLK_PERM_RESIZE; | |
2709 | } | |
2710 | ||
2711 | *nperm = perm; | |
2712 | *nshared = shared; | |
2713 | } | |
2714 | ||
6f838a4b | 2715 | static void bdrv_default_perms_for_storage(BlockDriverState *bs, BdrvChild *c, |
6f838a4b HR |
2716 | BdrvChildRole role, |
2717 | BlockReopenQueue *reopen_queue, | |
2718 | uint64_t perm, uint64_t shared, | |
2719 | uint64_t *nperm, uint64_t *nshared) | |
2720 | { | |
2721 | int flags; | |
2722 | ||
862fded9 | 2723 | GLOBAL_STATE_CODE(); |
e5d8a406 | 2724 | assert(role & (BDRV_CHILD_METADATA | BDRV_CHILD_DATA)); |
6f838a4b HR |
2725 | |
2726 | flags = bdrv_reopen_get_flags(reopen_queue, bs); | |
2727 | ||
2728 | /* | |
2729 | * Apart from the modifications below, the same permissions are | |
2730 | * forwarded and left alone as for filters | |
2731 | */ | |
e5d8a406 | 2732 | bdrv_filter_default_perms(bs, c, role, reopen_queue, |
6f838a4b HR |
2733 | perm, shared, &perm, &shared); |
2734 | ||
f889054f HR |
2735 | if (role & BDRV_CHILD_METADATA) { |
2736 | /* Format drivers may touch metadata even if the guest doesn't write */ | |
2737 | if (bdrv_is_writable_after_reopen(bs, reopen_queue)) { | |
2738 | perm |= BLK_PERM_WRITE | BLK_PERM_RESIZE; | |
2739 | } | |
2740 | ||
2741 | /* | |
2742 | * bs->file always needs to be consistent because of the | |
2743 | * metadata. We can never allow other users to resize or write | |
2744 | * to it. | |
2745 | */ | |
2746 | if (!(flags & BDRV_O_NO_IO)) { | |
2747 | perm |= BLK_PERM_CONSISTENT_READ; | |
2748 | } | |
2749 | shared &= ~(BLK_PERM_WRITE | BLK_PERM_RESIZE); | |
6f838a4b HR |
2750 | } |
2751 | ||
f889054f HR |
2752 | if (role & BDRV_CHILD_DATA) { |
2753 | /* | |
2754 | * Technically, everything in this block is a subset of the | |
2755 | * BDRV_CHILD_METADATA path taken above, and so this could | |
2756 | * be an "else if" branch. However, that is not obvious, and | |
2757 | * this function is not performance critical, therefore we let | |
2758 | * this be an independent "if". | |
2759 | */ | |
2760 | ||
2761 | /* | |
2762 | * We cannot allow other users to resize the file because the | |
2763 | * format driver might have some assumptions about the size | |
2764 | * (e.g. because it is stored in metadata, or because the file | |
2765 | * is split into fixed-size data files). | |
2766 | */ | |
2767 | shared &= ~BLK_PERM_RESIZE; | |
2768 | ||
2769 | /* | |
2770 | * WRITE_UNCHANGED often cannot be performed as such on the | |
2771 | * data file. For example, the qcow2 driver may still need to | |
2772 | * write copied clusters on copy-on-read. | |
2773 | */ | |
2774 | if (perm & BLK_PERM_WRITE_UNCHANGED) { | |
2775 | perm |= BLK_PERM_WRITE; | |
2776 | } | |
2777 | ||
2778 | /* | |
2779 | * If the data file is written to, the format driver may | |
2780 | * expect to be able to resize it by writing beyond the EOF. | |
2781 | */ | |
2782 | if (perm & BLK_PERM_WRITE) { | |
2783 | perm |= BLK_PERM_RESIZE; | |
2784 | } | |
6f838a4b | 2785 | } |
6f838a4b HR |
2786 | |
2787 | if (bs->open_flags & BDRV_O_INACTIVE) { | |
2788 | shared |= BLK_PERM_WRITE | BLK_PERM_RESIZE; | |
2789 | } | |
2790 | ||
2791 | *nperm = perm; | |
2792 | *nshared = shared; | |
2793 | } | |
2794 | ||
2519f549 | 2795 | void bdrv_default_perms(BlockDriverState *bs, BdrvChild *c, |
e5d8a406 | 2796 | BdrvChildRole role, BlockReopenQueue *reopen_queue, |
2519f549 HR |
2797 | uint64_t perm, uint64_t shared, |
2798 | uint64_t *nperm, uint64_t *nshared) | |
2799 | { | |
b4ad82aa | 2800 | GLOBAL_STATE_CODE(); |
2519f549 HR |
2801 | if (role & BDRV_CHILD_FILTERED) { |
2802 | assert(!(role & (BDRV_CHILD_DATA | BDRV_CHILD_METADATA | | |
2803 | BDRV_CHILD_COW))); | |
e5d8a406 | 2804 | bdrv_filter_default_perms(bs, c, role, reopen_queue, |
2519f549 HR |
2805 | perm, shared, nperm, nshared); |
2806 | } else if (role & BDRV_CHILD_COW) { | |
2807 | assert(!(role & (BDRV_CHILD_DATA | BDRV_CHILD_METADATA))); | |
e5d8a406 | 2808 | bdrv_default_perms_for_cow(bs, c, role, reopen_queue, |
2519f549 HR |
2809 | perm, shared, nperm, nshared); |
2810 | } else if (role & (BDRV_CHILD_METADATA | BDRV_CHILD_DATA)) { | |
e5d8a406 | 2811 | bdrv_default_perms_for_storage(bs, c, role, reopen_queue, |
2519f549 HR |
2812 | perm, shared, nperm, nshared); |
2813 | } else { | |
2814 | g_assert_not_reached(); | |
2815 | } | |
2816 | } | |
2817 | ||
7b1d9c4d HR |
2818 | uint64_t bdrv_qapi_perm_to_blk_perm(BlockPermission qapi_perm) |
2819 | { | |
2820 | static const uint64_t permissions[] = { | |
2821 | [BLOCK_PERMISSION_CONSISTENT_READ] = BLK_PERM_CONSISTENT_READ, | |
2822 | [BLOCK_PERMISSION_WRITE] = BLK_PERM_WRITE, | |
2823 | [BLOCK_PERMISSION_WRITE_UNCHANGED] = BLK_PERM_WRITE_UNCHANGED, | |
2824 | [BLOCK_PERMISSION_RESIZE] = BLK_PERM_RESIZE, | |
7b1d9c4d HR |
2825 | }; |
2826 | ||
2827 | QEMU_BUILD_BUG_ON(ARRAY_SIZE(permissions) != BLOCK_PERMISSION__MAX); | |
2828 | QEMU_BUILD_BUG_ON(1UL << ARRAY_SIZE(permissions) != BLK_PERM_ALL + 1); | |
2829 | ||
2830 | assert(qapi_perm < BLOCK_PERMISSION__MAX); | |
2831 | ||
2832 | return permissions[qapi_perm]; | |
2833 | } | |
2834 | ||
23987471 KW |
2835 | /* |
2836 | * Replaces the node that a BdrvChild points to without updating permissions. | |
2837 | * | |
2838 | * If @new_bs is non-NULL, the parent of @child must already be drained through | |
2839 | * @child. | |
23987471 | 2840 | */ |
544acc7d | 2841 | static void bdrv_replace_child_noperm(BdrvChild *child, |
4eba825a | 2842 | BlockDriverState *new_bs) |
e9740bc6 KW |
2843 | { |
2844 | BlockDriverState *old_bs = child->bs; | |
debc2927 | 2845 | int new_bs_quiesce_counter; |
e9740bc6 | 2846 | |
2cad1ebe | 2847 | assert(!child->frozen); |
23987471 KW |
2848 | |
2849 | /* | |
2850 | * If we want to change the BdrvChild to point to a drained node as its new | |
2851 | * child->bs, we need to make sure that its new parent is drained, too. In | |
2852 | * other words, either child->quiesce_parent must already be true or we must | |
2853 | * be able to set it and keep the parent's quiesce_counter consistent with | |
2854 | * that, but without polling or starting new requests (this function | |
2855 | * guarantees that it doesn't poll, and starting new requests would be | |
2856 | * against the invariants of drain sections). | |
2857 | * | |
2858 | * To keep things simple, we pick the first option (child->quiesce_parent | |
2859 | * must already be true). We also generalise the rule a bit to make it | |
2860 | * easier to verify in callers and more likely to be covered in test cases: | |
2861 | * The parent must be quiesced through this child even if new_bs isn't | |
2862 | * currently drained. | |
2863 | * | |
2864 | * The only exception is for callers that always pass new_bs == NULL. In | |
2865 | * this case, we obviously never need to consider the case of a drained | |
2866 | * new_bs, so we can keep the callers simpler by allowing them not to drain | |
2867 | * the parent. | |
2868 | */ | |
2869 | assert(!new_bs || child->quiesced_parent); | |
bfb8aa6d | 2870 | assert(old_bs != new_bs); |
f0c28327 | 2871 | GLOBAL_STATE_CODE(); |
2cad1ebe | 2872 | |
bb2614e9 FZ |
2873 | if (old_bs && new_bs) { |
2874 | assert(bdrv_get_aio_context(old_bs) == bdrv_get_aio_context(new_bs)); | |
2875 | } | |
debc2927 | 2876 | |
29312507 EGE |
2877 | /* TODO Pull this up into the callers to avoid polling here */ |
2878 | bdrv_graph_wrlock(); | |
e9740bc6 | 2879 | if (old_bs) { |
bd86fb99 HR |
2880 | if (child->klass->detach) { |
2881 | child->klass->detach(child); | |
d736f119 | 2882 | } |
e9740bc6 KW |
2883 | QLIST_REMOVE(child, next_parent); |
2884 | } | |
36fe1331 KW |
2885 | |
2886 | child->bs = new_bs; | |
2887 | ||
e9740bc6 KW |
2888 | if (new_bs) { |
2889 | QLIST_INSERT_HEAD(&new_bs->parents, child, next_parent); | |
bd86fb99 HR |
2890 | if (child->klass->attach) { |
2891 | child->klass->attach(child); | |
8ee03995 KW |
2892 | } |
2893 | } | |
29312507 | 2894 | bdrv_graph_wrunlock(); |
debc2927 HR |
2895 | |
2896 | /* | |
23987471 KW |
2897 | * If the parent was drained through this BdrvChild previously, but new_bs |
2898 | * is not drained, allow requests to come in only after the new node has | |
2899 | * been attached. | |
debc2927 | 2900 | */ |
57e05be3 KW |
2901 | new_bs_quiesce_counter = (new_bs ? new_bs->quiesce_counter : 0); |
2902 | if (!new_bs_quiesce_counter && child->quiesced_parent) { | |
debc2927 | 2903 | bdrv_parent_drained_end_single(child); |
debc2927 | 2904 | } |
8ee03995 | 2905 | } |
33a610c3 | 2906 | |
04c9c3a5 HR |
2907 | /** |
2908 | * Free the given @child. | |
2909 | * | |
2910 | * The child must be empty (i.e. `child->bs == NULL`) and it must be | |
2911 | * unused (i.e. not in a children list). | |
2912 | */ | |
2913 | static void bdrv_child_free(BdrvChild *child) | |
df581792 | 2914 | { |
548a74c0 | 2915 | assert(!child->bs); |
bdb73476 | 2916 | GLOBAL_STATE_CODE(); |
a225369b | 2917 | assert(!child->next.le_prev); /* not in children list */ |
04c9c3a5 HR |
2918 | |
2919 | g_free(child->name); | |
2920 | g_free(child); | |
548a74c0 | 2921 | } |
d5e6f437 | 2922 | |
548a74c0 | 2923 | typedef struct BdrvAttachChildCommonState { |
5bb04747 | 2924 | BdrvChild *child; |
548a74c0 VSO |
2925 | AioContext *old_parent_ctx; |
2926 | AioContext *old_child_ctx; | |
2927 | } BdrvAttachChildCommonState; | |
2928 | ||
2929 | static void bdrv_attach_child_common_abort(void *opaque) | |
2930 | { | |
2931 | BdrvAttachChildCommonState *s = opaque; | |
5bb04747 | 2932 | BlockDriverState *bs = s->child->bs; |
548a74c0 | 2933 | |
f0c28327 | 2934 | GLOBAL_STATE_CODE(); |
5bb04747 | 2935 | bdrv_replace_child_noperm(s->child, NULL); |
548a74c0 VSO |
2936 | |
2937 | if (bdrv_get_aio_context(bs) != s->old_child_ctx) { | |
142e6907 | 2938 | bdrv_try_change_aio_context(bs, s->old_child_ctx, NULL, &error_abort); |
548a74c0 VSO |
2939 | } |
2940 | ||
5bb04747 | 2941 | if (bdrv_child_get_parent_aio_context(s->child) != s->old_parent_ctx) { |
f8be48ad EGE |
2942 | Transaction *tran; |
2943 | GHashTable *visited; | |
2944 | bool ret; | |
548a74c0 | 2945 | |
f8be48ad | 2946 | tran = tran_new(); |
548a74c0 | 2947 | |
f8be48ad EGE |
2948 | /* No need to visit `child`, because it has been detached already */ |
2949 | visited = g_hash_table_new(NULL, NULL); | |
2950 | ret = s->child->klass->change_aio_ctx(s->child, s->old_parent_ctx, | |
2951 | visited, tran, &error_abort); | |
2952 | g_hash_table_destroy(visited); | |
2953 | ||
2954 | /* transaction is supposed to always succeed */ | |
2955 | assert(ret == true); | |
2956 | tran_commit(tran); | |
d5e6f437 KW |
2957 | } |
2958 | ||
548a74c0 | 2959 | bdrv_unref(bs); |
5bb04747 | 2960 | bdrv_child_free(s->child); |
548a74c0 VSO |
2961 | } |
2962 | ||
2963 | static TransactionActionDrv bdrv_attach_child_common_drv = { | |
2964 | .abort = bdrv_attach_child_common_abort, | |
2965 | .clean = g_free, | |
2966 | }; | |
2967 | ||
2968 | /* | |
2969 | * Common part of attaching bdrv child to bs or to blk or to job | |
f8d2ad78 | 2970 | * |
7ec390d5 | 2971 | * Function doesn't update permissions, caller is responsible for this. |
5bb04747 VSO |
2972 | * |
2973 | * Returns new created child. | |
548a74c0 | 2974 | */ |
5bb04747 VSO |
2975 | static BdrvChild *bdrv_attach_child_common(BlockDriverState *child_bs, |
2976 | const char *child_name, | |
2977 | const BdrvChildClass *child_class, | |
2978 | BdrvChildRole child_role, | |
2979 | uint64_t perm, uint64_t shared_perm, | |
2980 | void *opaque, | |
2981 | Transaction *tran, Error **errp) | |
548a74c0 VSO |
2982 | { |
2983 | BdrvChild *new_child; | |
2984 | AioContext *parent_ctx; | |
2985 | AioContext *child_ctx = bdrv_get_aio_context(child_bs); | |
2986 | ||
da261b69 | 2987 | assert(child_class->get_parent_desc); |
bdb73476 | 2988 | GLOBAL_STATE_CODE(); |
548a74c0 VSO |
2989 | |
2990 | new_child = g_new(BdrvChild, 1); | |
2991 | *new_child = (BdrvChild) { | |
d5e6f437 KW |
2992 | .bs = NULL, |
2993 | .name = g_strdup(child_name), | |
bd86fb99 | 2994 | .klass = child_class, |
258b7765 | 2995 | .role = child_role, |
d5e6f437 KW |
2996 | .perm = perm, |
2997 | .shared_perm = shared_perm, | |
2998 | .opaque = opaque, | |
df581792 KW |
2999 | }; |
3000 | ||
548a74c0 VSO |
3001 | /* |
3002 | * If the AioContexts don't match, first try to move the subtree of | |
132ada80 | 3003 | * child_bs into the AioContext of the new parent. If this doesn't work, |
548a74c0 VSO |
3004 | * try moving the parent into the AioContext of child_bs instead. |
3005 | */ | |
3006 | parent_ctx = bdrv_child_get_parent_aio_context(new_child); | |
3007 | if (child_ctx != parent_ctx) { | |
3008 | Error *local_err = NULL; | |
142e6907 EGE |
3009 | int ret = bdrv_try_change_aio_context(child_bs, parent_ctx, NULL, |
3010 | &local_err); | |
548a74c0 | 3011 | |
f8be48ad EGE |
3012 | if (ret < 0 && child_class->change_aio_ctx) { |
3013 | Transaction *tran = tran_new(); | |
3014 | GHashTable *visited = g_hash_table_new(NULL, NULL); | |
3015 | bool ret_child; | |
3016 | ||
3017 | g_hash_table_add(visited, new_child); | |
3018 | ret_child = child_class->change_aio_ctx(new_child, child_ctx, | |
3019 | visited, tran, NULL); | |
3020 | if (ret_child == true) { | |
132ada80 KW |
3021 | error_free(local_err); |
3022 | ret = 0; | |
132ada80 | 3023 | } |
f8be48ad EGE |
3024 | tran_finalize(tran, ret_child == true ? 0 : -1); |
3025 | g_hash_table_destroy(visited); | |
132ada80 | 3026 | } |
548a74c0 | 3027 | |
132ada80 KW |
3028 | if (ret < 0) { |
3029 | error_propagate(errp, local_err); | |
04c9c3a5 | 3030 | bdrv_child_free(new_child); |
5bb04747 | 3031 | return NULL; |
132ada80 KW |
3032 | } |
3033 | } | |
3034 | ||
548a74c0 | 3035 | bdrv_ref(child_bs); |
23987471 KW |
3036 | /* |
3037 | * Let every new BdrvChild start with a drained parent. Inserting the child | |
3038 | * in the graph with bdrv_replace_child_noperm() will undrain it if | |
3039 | * @child_bs is not drained. | |
3040 | * | |
3041 | * The child was only just created and is not yet visible in global state | |
3042 | * until bdrv_replace_child_noperm() inserts it into the graph, so nobody | |
3043 | * could have sent requests and polling is not necessary. | |
3044 | * | |
3045 | * Note that this means that the parent isn't fully drained yet, we only | |
3046 | * stop new requests from coming in. This is fine, we don't care about the | |
3047 | * old requests here, they are not for this child. If another place enters a | |
3048 | * drain section for the same parent, but wants it to be fully quiesced, it | |
3049 | * will not run most of the the code in .drained_begin() again (which is not | |
3050 | * a problem, we already did this), but it will still poll until the parent | |
3051 | * is fully quiesced, so it will not be negatively affected either. | |
3052 | */ | |
606ed756 | 3053 | bdrv_parent_drained_begin_single(new_child); |
544acc7d | 3054 | bdrv_replace_child_noperm(new_child, child_bs); |
b4b059f6 | 3055 | |
548a74c0 VSO |
3056 | BdrvAttachChildCommonState *s = g_new(BdrvAttachChildCommonState, 1); |
3057 | *s = (BdrvAttachChildCommonState) { | |
5bb04747 | 3058 | .child = new_child, |
548a74c0 VSO |
3059 | .old_parent_ctx = parent_ctx, |
3060 | .old_child_ctx = child_ctx, | |
3061 | }; | |
3062 | tran_add(tran, &bdrv_attach_child_common_drv, s); | |
3063 | ||
5bb04747 | 3064 | return new_child; |
548a74c0 VSO |
3065 | } |
3066 | ||
f8d2ad78 | 3067 | /* |
7ec390d5 | 3068 | * Function doesn't update permissions, caller is responsible for this. |
f8d2ad78 | 3069 | */ |
5bb04747 VSO |
3070 | static BdrvChild *bdrv_attach_child_noperm(BlockDriverState *parent_bs, |
3071 | BlockDriverState *child_bs, | |
3072 | const char *child_name, | |
3073 | const BdrvChildClass *child_class, | |
3074 | BdrvChildRole child_role, | |
3075 | Transaction *tran, | |
3076 | Error **errp) | |
aa5a04c7 | 3077 | { |
aa5a04c7 VSO |
3078 | uint64_t perm, shared_perm; |
3079 | ||
3080 | assert(parent_bs->drv); | |
bdb73476 | 3081 | GLOBAL_STATE_CODE(); |
aa5a04c7 | 3082 | |
bfb8aa6d KW |
3083 | if (bdrv_recurse_has_child(child_bs, parent_bs)) { |
3084 | error_setg(errp, "Making '%s' a %s child of '%s' would create a cycle", | |
3085 | child_bs->node_name, child_name, parent_bs->node_name); | |
5bb04747 | 3086 | return NULL; |
bfb8aa6d KW |
3087 | } |
3088 | ||
aa5a04c7 VSO |
3089 | bdrv_get_cumulative_perm(parent_bs, &perm, &shared_perm); |
3090 | bdrv_child_perm(parent_bs, child_bs, NULL, child_role, NULL, | |
3091 | perm, shared_perm, &perm, &shared_perm); | |
3092 | ||
5bb04747 VSO |
3093 | return bdrv_attach_child_common(child_bs, child_name, child_class, |
3094 | child_role, perm, shared_perm, parent_bs, | |
3095 | tran, errp); | |
aa5a04c7 VSO |
3096 | } |
3097 | ||
548a74c0 VSO |
3098 | /* |
3099 | * This function steals the reference to child_bs from the caller. | |
3100 | * That reference is later dropped by bdrv_root_unref_child(). | |
3101 | * | |
3102 | * On failure NULL is returned, errp is set and the reference to | |
3103 | * child_bs is also dropped. | |
3104 | * | |
3105 | * The caller must hold the AioContext lock @child_bs, but not that of @ctx | |
3106 | * (unless @child_bs is already in @ctx). | |
3107 | */ | |
3108 | BdrvChild *bdrv_root_attach_child(BlockDriverState *child_bs, | |
3109 | const char *child_name, | |
3110 | const BdrvChildClass *child_class, | |
3111 | BdrvChildRole child_role, | |
3112 | uint64_t perm, uint64_t shared_perm, | |
3113 | void *opaque, Error **errp) | |
3114 | { | |
3115 | int ret; | |
5bb04747 | 3116 | BdrvChild *child; |
548a74c0 VSO |
3117 | Transaction *tran = tran_new(); |
3118 | ||
b4ad82aa EGE |
3119 | GLOBAL_STATE_CODE(); |
3120 | ||
5bb04747 | 3121 | child = bdrv_attach_child_common(child_bs, child_name, child_class, |
548a74c0 | 3122 | child_role, perm, shared_perm, opaque, |
5bb04747 VSO |
3123 | tran, errp); |
3124 | if (!child) { | |
3125 | ret = -EINVAL; | |
e878bb12 | 3126 | goto out; |
548a74c0 VSO |
3127 | } |
3128 | ||
f1316edb | 3129 | ret = bdrv_refresh_perms(child_bs, tran, errp); |
548a74c0 | 3130 | |
e878bb12 KW |
3131 | out: |
3132 | tran_finalize(tran, ret); | |
f8d2ad78 | 3133 | |
548a74c0 | 3134 | bdrv_unref(child_bs); |
5bb04747 VSO |
3135 | |
3136 | return ret < 0 ? NULL : child; | |
df581792 KW |
3137 | } |
3138 | ||
b441dc71 AG |
3139 | /* |
3140 | * This function transfers the reference to child_bs from the caller | |
3141 | * to parent_bs. That reference is later dropped by parent_bs on | |
3142 | * bdrv_close() or if someone calls bdrv_unref_child(). | |
3143 | * | |
3144 | * On failure NULL is returned, errp is set and the reference to | |
3145 | * child_bs is also dropped. | |
132ada80 KW |
3146 | * |
3147 | * If @parent_bs and @child_bs are in different AioContexts, the caller must | |
3148 | * hold the AioContext lock for @child_bs, but not for @parent_bs. | |
b441dc71 | 3149 | */ |
98292c61 WC |
3150 | BdrvChild *bdrv_attach_child(BlockDriverState *parent_bs, |
3151 | BlockDriverState *child_bs, | |
3152 | const char *child_name, | |
bd86fb99 | 3153 | const BdrvChildClass *child_class, |
258b7765 | 3154 | BdrvChildRole child_role, |
8b2ff529 | 3155 | Error **errp) |
f21d96d0 | 3156 | { |
aa5a04c7 | 3157 | int ret; |
5bb04747 | 3158 | BdrvChild *child; |
aa5a04c7 | 3159 | Transaction *tran = tran_new(); |
f68c598b | 3160 | |
f791bf7f EGE |
3161 | GLOBAL_STATE_CODE(); |
3162 | ||
5bb04747 VSO |
3163 | child = bdrv_attach_child_noperm(parent_bs, child_bs, child_name, |
3164 | child_class, child_role, tran, errp); | |
3165 | if (!child) { | |
3166 | ret = -EINVAL; | |
aa5a04c7 VSO |
3167 | goto out; |
3168 | } | |
d5e6f437 | 3169 | |
f1316edb | 3170 | ret = bdrv_refresh_perms(parent_bs, tran, errp); |
aa5a04c7 VSO |
3171 | if (ret < 0) { |
3172 | goto out; | |
d5e6f437 KW |
3173 | } |
3174 | ||
aa5a04c7 VSO |
3175 | out: |
3176 | tran_finalize(tran, ret); | |
3177 | ||
3178 | bdrv_unref(child_bs); | |
3179 | ||
5bb04747 | 3180 | return ret < 0 ? NULL : child; |
f21d96d0 KW |
3181 | } |
3182 | ||
7b99a266 | 3183 | /* Callers must ensure that child->frozen is false. */ |
f21d96d0 | 3184 | void bdrv_root_unref_child(BdrvChild *child) |
33a60407 | 3185 | { |
00eb93b5 | 3186 | BlockDriverState *child_bs = child->bs; |
779020cb | 3187 | |
f791bf7f | 3188 | GLOBAL_STATE_CODE(); |
00eb93b5 VSO |
3189 | bdrv_replace_child_noperm(child, NULL); |
3190 | bdrv_child_free(child); | |
3191 | ||
3192 | if (child_bs) { | |
3193 | /* | |
3194 | * Update permissions for old node. We're just taking a parent away, so | |
3195 | * we're loosening restrictions. Errors of permission update are not | |
3196 | * fatal in this case, ignore them. | |
3197 | */ | |
f1316edb | 3198 | bdrv_refresh_perms(child_bs, NULL, NULL); |
00eb93b5 VSO |
3199 | |
3200 | /* | |
3201 | * When the parent requiring a non-default AioContext is removed, the | |
3202 | * node moves back to the main AioContext | |
3203 | */ | |
3204 | bdrv_try_change_aio_context(child_bs, qemu_get_aio_context(), NULL, | |
3205 | NULL); | |
3206 | } | |
f791bf7f | 3207 | |
f21d96d0 KW |
3208 | bdrv_unref(child_bs); |
3209 | } | |
3210 | ||
332b3a17 VSO |
3211 | typedef struct BdrvSetInheritsFrom { |
3212 | BlockDriverState *bs; | |
3213 | BlockDriverState *old_inherits_from; | |
3214 | } BdrvSetInheritsFrom; | |
3215 | ||
3216 | static void bdrv_set_inherits_from_abort(void *opaque) | |
3217 | { | |
3218 | BdrvSetInheritsFrom *s = opaque; | |
3219 | ||
3220 | s->bs->inherits_from = s->old_inherits_from; | |
3221 | } | |
3222 | ||
3223 | static TransactionActionDrv bdrv_set_inherits_from_drv = { | |
3224 | .abort = bdrv_set_inherits_from_abort, | |
3225 | .clean = g_free, | |
3226 | }; | |
3227 | ||
3228 | /* @tran is allowed to be NULL. In this case no rollback is possible */ | |
3229 | static void bdrv_set_inherits_from(BlockDriverState *bs, | |
3230 | BlockDriverState *new_inherits_from, | |
3231 | Transaction *tran) | |
3232 | { | |
3233 | if (tran) { | |
3234 | BdrvSetInheritsFrom *s = g_new(BdrvSetInheritsFrom, 1); | |
3235 | ||
3236 | *s = (BdrvSetInheritsFrom) { | |
3237 | .bs = bs, | |
3238 | .old_inherits_from = bs->inherits_from, | |
3239 | }; | |
3240 | ||
3241 | tran_add(tran, &bdrv_set_inherits_from_drv, s); | |
3242 | } | |
3243 | ||
3244 | bs->inherits_from = new_inherits_from; | |
3245 | } | |
3246 | ||
3cf746b3 HR |
3247 | /** |
3248 | * Clear all inherits_from pointers from children and grandchildren of | |
3249 | * @root that point to @root, where necessary. | |
332b3a17 | 3250 | * @tran is allowed to be NULL. In this case no rollback is possible |
3cf746b3 | 3251 | */ |
332b3a17 VSO |
3252 | static void bdrv_unset_inherits_from(BlockDriverState *root, BdrvChild *child, |
3253 | Transaction *tran) | |
f21d96d0 | 3254 | { |
3cf746b3 | 3255 | BdrvChild *c; |
4e4bf5c4 | 3256 | |
3cf746b3 HR |
3257 | if (child->bs->inherits_from == root) { |
3258 | /* | |
3259 | * Remove inherits_from only when the last reference between root and | |
3260 | * child->bs goes away. | |
3261 | */ | |
3262 | QLIST_FOREACH(c, &root->children, next) { | |
4e4bf5c4 KW |
3263 | if (c != child && c->bs == child->bs) { |
3264 | break; | |
3265 | } | |
3266 | } | |
3267 | if (c == NULL) { | |
332b3a17 | 3268 | bdrv_set_inherits_from(child->bs, NULL, tran); |
4e4bf5c4 | 3269 | } |
33a60407 KW |
3270 | } |
3271 | ||
3cf746b3 | 3272 | QLIST_FOREACH(c, &child->bs->children, next) { |
332b3a17 | 3273 | bdrv_unset_inherits_from(root, c, tran); |
3cf746b3 HR |
3274 | } |
3275 | } | |
3276 | ||
7b99a266 | 3277 | /* Callers must ensure that child->frozen is false. */ |
3cf746b3 HR |
3278 | void bdrv_unref_child(BlockDriverState *parent, BdrvChild *child) |
3279 | { | |
f791bf7f | 3280 | GLOBAL_STATE_CODE(); |
3cf746b3 HR |
3281 | if (child == NULL) { |
3282 | return; | |
3283 | } | |
3284 | ||
332b3a17 | 3285 | bdrv_unset_inherits_from(parent, child, NULL); |
f21d96d0 | 3286 | bdrv_root_unref_child(child); |
33a60407 KW |
3287 | } |
3288 | ||
5c8cab48 KW |
3289 | |
3290 | static void bdrv_parent_cb_change_media(BlockDriverState *bs, bool load) | |
3291 | { | |
3292 | BdrvChild *c; | |
f0c28327 | 3293 | GLOBAL_STATE_CODE(); |
5c8cab48 | 3294 | QLIST_FOREACH(c, &bs->parents, next_parent) { |
bd86fb99 HR |
3295 | if (c->klass->change_media) { |
3296 | c->klass->change_media(c, load); | |
5c8cab48 KW |
3297 | } |
3298 | } | |
3299 | } | |
3300 | ||
0065c455 AG |
3301 | /* Return true if you can reach parent going through child->inherits_from |
3302 | * recursively. If parent or child are NULL, return false */ | |
3303 | static bool bdrv_inherits_from_recursive(BlockDriverState *child, | |
3304 | BlockDriverState *parent) | |
3305 | { | |
3306 | while (child && child != parent) { | |
3307 | child = child->inherits_from; | |
3308 | } | |
3309 | ||
3310 | return child != NULL; | |
3311 | } | |
3312 | ||
25191e5f HR |
3313 | /* |
3314 | * Return the BdrvChildRole for @bs's backing child. bs->backing is | |
3315 | * mostly used for COW backing children (role = COW), but also for | |
3316 | * filtered children (role = FILTERED | PRIMARY). | |
3317 | */ | |
3318 | static BdrvChildRole bdrv_backing_role(BlockDriverState *bs) | |
3319 | { | |
3320 | if (bs->drv && bs->drv->is_filter) { | |
3321 | return BDRV_CHILD_FILTERED | BDRV_CHILD_PRIMARY; | |
3322 | } else { | |
3323 | return BDRV_CHILD_COW; | |
3324 | } | |
3325 | } | |
3326 | ||
5db15a57 | 3327 | /* |
e9238278 VSO |
3328 | * Sets the bs->backing or bs->file link of a BDS. A new reference is created; |
3329 | * callers which don't need their own reference any more must call bdrv_unref(). | |
7ec390d5 VSO |
3330 | * |
3331 | * Function doesn't update permissions, caller is responsible for this. | |
5db15a57 | 3332 | */ |
e9238278 VSO |
3333 | static int bdrv_set_file_or_backing_noperm(BlockDriverState *parent_bs, |
3334 | BlockDriverState *child_bs, | |
3335 | bool is_backing, | |
3336 | Transaction *tran, Error **errp) | |
8d24cce1 | 3337 | { |
e9238278 VSO |
3338 | bool update_inherits_from = |
3339 | bdrv_inherits_from_recursive(child_bs, parent_bs); | |
3340 | BdrvChild *child = is_backing ? parent_bs->backing : parent_bs->file; | |
3341 | BdrvChildRole role; | |
0065c455 | 3342 | |
bdb73476 EGE |
3343 | GLOBAL_STATE_CODE(); |
3344 | ||
e9238278 VSO |
3345 | if (!parent_bs->drv) { |
3346 | /* | |
3347 | * Node without drv is an object without a class :/. TODO: finally fix | |
3348 | * qcow2 driver to never clear bs->drv and implement format corruption | |
3349 | * handling in other way. | |
3350 | */ | |
3351 | error_setg(errp, "Node corrupted"); | |
3352 | return -EINVAL; | |
3353 | } | |
3354 | ||
3355 | if (child && child->frozen) { | |
3356 | error_setg(errp, "Cannot change frozen '%s' link from '%s' to '%s'", | |
3357 | child->name, parent_bs->node_name, child->bs->node_name); | |
a1e708fc | 3358 | return -EPERM; |
2cad1ebe AG |
3359 | } |
3360 | ||
25f78d9e VSO |
3361 | if (is_backing && !parent_bs->drv->is_filter && |
3362 | !parent_bs->drv->supports_backing) | |
3363 | { | |
3364 | error_setg(errp, "Driver '%s' of node '%s' does not support backing " | |
3365 | "files", parent_bs->drv->format_name, parent_bs->node_name); | |
3366 | return -EINVAL; | |
3367 | } | |
3368 | ||
e9238278 VSO |
3369 | if (parent_bs->drv->is_filter) { |
3370 | role = BDRV_CHILD_FILTERED | BDRV_CHILD_PRIMARY; | |
3371 | } else if (is_backing) { | |
3372 | role = BDRV_CHILD_COW; | |
3373 | } else { | |
3374 | /* | |
3375 | * We only can use same role as it is in existing child. We don't have | |
3376 | * infrastructure to determine role of file child in generic way | |
3377 | */ | |
3378 | if (!child) { | |
3379 | error_setg(errp, "Cannot set file child to format node without " | |
3380 | "file child"); | |
3381 | return -EINVAL; | |
3382 | } | |
3383 | role = child->role; | |
826b6ca0 FZ |
3384 | } |
3385 | ||
e9238278 VSO |
3386 | if (child) { |
3387 | bdrv_unset_inherits_from(parent_bs, child, tran); | |
57f08941 | 3388 | bdrv_remove_child(child, tran); |
e9238278 VSO |
3389 | } |
3390 | ||
3391 | if (!child_bs) { | |
8d24cce1 FZ |
3392 | goto out; |
3393 | } | |
12fa4af6 | 3394 | |
5bb04747 VSO |
3395 | child = bdrv_attach_child_noperm(parent_bs, child_bs, |
3396 | is_backing ? "backing" : "file", | |
3397 | &child_of_bds, role, | |
3398 | tran, errp); | |
3399 | if (!child) { | |
3400 | return -EINVAL; | |
a1e708fc VSO |
3401 | } |
3402 | ||
160333e1 VSO |
3403 | |
3404 | /* | |
e9238278 | 3405 | * If inherits_from pointed recursively to bs then let's update it to |
160333e1 VSO |
3406 | * point directly to bs (else it will become NULL). |
3407 | */ | |
a1e708fc | 3408 | if (update_inherits_from) { |
e9238278 | 3409 | bdrv_set_inherits_from(child_bs, parent_bs, tran); |
0065c455 | 3410 | } |
826b6ca0 | 3411 | |
8d24cce1 | 3412 | out: |
e9238278 | 3413 | bdrv_refresh_limits(parent_bs, tran, NULL); |
160333e1 VSO |
3414 | |
3415 | return 0; | |
3416 | } | |
3417 | ||
e9238278 VSO |
3418 | static int bdrv_set_backing_noperm(BlockDriverState *bs, |
3419 | BlockDriverState *backing_hd, | |
3420 | Transaction *tran, Error **errp) | |
3421 | { | |
bdb73476 | 3422 | GLOBAL_STATE_CODE(); |
e9238278 VSO |
3423 | return bdrv_set_file_or_backing_noperm(bs, backing_hd, true, tran, errp); |
3424 | } | |
3425 | ||
92140b9f KW |
3426 | int bdrv_set_backing_hd_drained(BlockDriverState *bs, |
3427 | BlockDriverState *backing_hd, | |
3428 | Error **errp) | |
160333e1 VSO |
3429 | { |
3430 | int ret; | |
3431 | Transaction *tran = tran_new(); | |
3432 | ||
f791bf7f | 3433 | GLOBAL_STATE_CODE(); |
92140b9f | 3434 | assert(bs->quiesce_counter > 0); |
c0829cb1 | 3435 | |
160333e1 VSO |
3436 | ret = bdrv_set_backing_noperm(bs, backing_hd, tran, errp); |
3437 | if (ret < 0) { | |
3438 | goto out; | |
3439 | } | |
3440 | ||
f1316edb | 3441 | ret = bdrv_refresh_perms(bs, tran, errp); |
160333e1 VSO |
3442 | out: |
3443 | tran_finalize(tran, ret); | |
92140b9f KW |
3444 | return ret; |
3445 | } | |
a1e708fc | 3446 | |
92140b9f KW |
3447 | int bdrv_set_backing_hd(BlockDriverState *bs, BlockDriverState *backing_hd, |
3448 | Error **errp) | |
3449 | { | |
3450 | int ret; | |
3451 | GLOBAL_STATE_CODE(); | |
3452 | ||
3453 | bdrv_drained_begin(bs); | |
3454 | ret = bdrv_set_backing_hd_drained(bs, backing_hd, errp); | |
c0829cb1 VSO |
3455 | bdrv_drained_end(bs); |
3456 | ||
a1e708fc | 3457 | return ret; |
8d24cce1 FZ |
3458 | } |
3459 | ||
31ca6d07 KW |
3460 | /* |
3461 | * Opens the backing file for a BlockDriverState if not yet open | |
3462 | * | |
d9b7b057 KW |
3463 | * bdref_key specifies the key for the image's BlockdevRef in the options QDict. |
3464 | * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict | |
3465 | * itself, all options starting with "${bdref_key}." are considered part of the | |
3466 | * BlockdevRef. | |
3467 | * | |
3468 | * TODO Can this be unified with bdrv_open_image()? | |
31ca6d07 | 3469 | */ |
d9b7b057 KW |
3470 | int bdrv_open_backing_file(BlockDriverState *bs, QDict *parent_options, |
3471 | const char *bdref_key, Error **errp) | |
9156df12 | 3472 | { |
6b6833c1 | 3473 | char *backing_filename = NULL; |
d9b7b057 KW |
3474 | char *bdref_key_dot; |
3475 | const char *reference = NULL; | |
317fc44e | 3476 | int ret = 0; |
998c2019 | 3477 | bool implicit_backing = false; |
8d24cce1 | 3478 | BlockDriverState *backing_hd; |
d9b7b057 KW |
3479 | QDict *options; |
3480 | QDict *tmp_parent_options = NULL; | |
34b5d2c6 | 3481 | Error *local_err = NULL; |
9156df12 | 3482 | |
f791bf7f EGE |
3483 | GLOBAL_STATE_CODE(); |
3484 | ||
760e0063 | 3485 | if (bs->backing != NULL) { |
1ba4b6a5 | 3486 | goto free_exit; |
9156df12 PB |
3487 | } |
3488 | ||
31ca6d07 | 3489 | /* NULL means an empty set of options */ |
d9b7b057 KW |
3490 | if (parent_options == NULL) { |
3491 | tmp_parent_options = qdict_new(); | |
3492 | parent_options = tmp_parent_options; | |
31ca6d07 KW |
3493 | } |
3494 | ||
9156df12 | 3495 | bs->open_flags &= ~BDRV_O_NO_BACKING; |
d9b7b057 KW |
3496 | |
3497 | bdref_key_dot = g_strdup_printf("%s.", bdref_key); | |
3498 | qdict_extract_subqdict(parent_options, &options, bdref_key_dot); | |
3499 | g_free(bdref_key_dot); | |
3500 | ||
129c7d1c MA |
3501 | /* |
3502 | * Caution: while qdict_get_try_str() is fine, getting non-string | |
3503 | * types would require more care. When @parent_options come from | |
3504 | * -blockdev or blockdev_add, its members are typed according to | |
3505 | * the QAPI schema, but when they come from -drive, they're all | |
3506 | * QString. | |
3507 | */ | |
d9b7b057 KW |
3508 | reference = qdict_get_try_str(parent_options, bdref_key); |
3509 | if (reference || qdict_haskey(options, "file.filename")) { | |
6b6833c1 | 3510 | /* keep backing_filename NULL */ |
1cb6f506 | 3511 | } else if (bs->backing_file[0] == '\0' && qdict_size(options) == 0) { |
cb3e7f08 | 3512 | qobject_unref(options); |
1ba4b6a5 | 3513 | goto free_exit; |
dbecebdd | 3514 | } else { |
998c2019 HR |
3515 | if (qdict_size(options) == 0) { |
3516 | /* If the user specifies options that do not modify the | |
3517 | * backing file's behavior, we might still consider it the | |
3518 | * implicit backing file. But it's easier this way, and | |
3519 | * just specifying some of the backing BDS's options is | |
3520 | * only possible with -drive anyway (otherwise the QAPI | |
3521 | * schema forces the user to specify everything). */ | |
3522 | implicit_backing = !strcmp(bs->auto_backing_file, bs->backing_file); | |
3523 | } | |
3524 | ||
6b6833c1 | 3525 | backing_filename = bdrv_get_full_backing_filename(bs, &local_err); |
9f07429e HR |
3526 | if (local_err) { |
3527 | ret = -EINVAL; | |
3528 | error_propagate(errp, local_err); | |
cb3e7f08 | 3529 | qobject_unref(options); |
9f07429e HR |
3530 | goto free_exit; |
3531 | } | |
9156df12 PB |
3532 | } |
3533 | ||
8ee79e70 KW |
3534 | if (!bs->drv || !bs->drv->supports_backing) { |
3535 | ret = -EINVAL; | |
3536 | error_setg(errp, "Driver doesn't support backing files"); | |
cb3e7f08 | 3537 | qobject_unref(options); |
8ee79e70 KW |
3538 | goto free_exit; |
3539 | } | |
3540 | ||
6bff597b PK |
3541 | if (!reference && |
3542 | bs->backing_format[0] != '\0' && !qdict_haskey(options, "driver")) { | |
46f5ac20 | 3543 | qdict_put_str(options, "driver", bs->backing_format); |
9156df12 PB |
3544 | } |
3545 | ||
6b6833c1 | 3546 | backing_hd = bdrv_open_inherit(backing_filename, reference, options, 0, bs, |
25191e5f | 3547 | &child_of_bds, bdrv_backing_role(bs), errp); |
5b363937 | 3548 | if (!backing_hd) { |
9156df12 | 3549 | bs->open_flags |= BDRV_O_NO_BACKING; |
e43bfd9c | 3550 | error_prepend(errp, "Could not open backing file: "); |
5b363937 | 3551 | ret = -EINVAL; |
1ba4b6a5 | 3552 | goto free_exit; |
9156df12 | 3553 | } |
df581792 | 3554 | |
998c2019 HR |
3555 | if (implicit_backing) { |
3556 | bdrv_refresh_filename(backing_hd); | |
3557 | pstrcpy(bs->auto_backing_file, sizeof(bs->auto_backing_file), | |
3558 | backing_hd->filename); | |
3559 | } | |
3560 | ||
5db15a57 KW |
3561 | /* Hook up the backing file link; drop our reference, bs owns the |
3562 | * backing_hd reference now */ | |
dc9c10a1 | 3563 | ret = bdrv_set_backing_hd(bs, backing_hd, errp); |
5db15a57 | 3564 | bdrv_unref(backing_hd); |
dc9c10a1 | 3565 | if (ret < 0) { |
12fa4af6 KW |
3566 | goto free_exit; |
3567 | } | |
d80ac658 | 3568 | |
d9b7b057 KW |
3569 | qdict_del(parent_options, bdref_key); |
3570 | ||
1ba4b6a5 BC |
3571 | free_exit: |
3572 | g_free(backing_filename); | |
cb3e7f08 | 3573 | qobject_unref(tmp_parent_options); |
1ba4b6a5 | 3574 | return ret; |
9156df12 PB |
3575 | } |
3576 | ||
2d6b86af KW |
3577 | static BlockDriverState * |
3578 | bdrv_open_child_bs(const char *filename, QDict *options, const char *bdref_key, | |
bd86fb99 | 3579 | BlockDriverState *parent, const BdrvChildClass *child_class, |
272c02ea | 3580 | BdrvChildRole child_role, bool allow_none, Error **errp) |
da557aac | 3581 | { |
2d6b86af | 3582 | BlockDriverState *bs = NULL; |
da557aac | 3583 | QDict *image_options; |
da557aac HR |
3584 | char *bdref_key_dot; |
3585 | const char *reference; | |
3586 | ||
bd86fb99 | 3587 | assert(child_class != NULL); |
f67503e5 | 3588 | |
da557aac HR |
3589 | bdref_key_dot = g_strdup_printf("%s.", bdref_key); |
3590 | qdict_extract_subqdict(options, &image_options, bdref_key_dot); | |
3591 | g_free(bdref_key_dot); | |
3592 | ||
129c7d1c MA |
3593 | /* |
3594 | * Caution: while qdict_get_try_str() is fine, getting non-string | |
3595 | * types would require more care. When @options come from | |
3596 | * -blockdev or blockdev_add, its members are typed according to | |
3597 | * the QAPI schema, but when they come from -drive, they're all | |
3598 | * QString. | |
3599 | */ | |
da557aac HR |
3600 | reference = qdict_get_try_str(options, bdref_key); |
3601 | if (!filename && !reference && !qdict_size(image_options)) { | |
b4b059f6 | 3602 | if (!allow_none) { |
da557aac HR |
3603 | error_setg(errp, "A block device must be specified for \"%s\"", |
3604 | bdref_key); | |
da557aac | 3605 | } |
cb3e7f08 | 3606 | qobject_unref(image_options); |
da557aac HR |
3607 | goto done; |
3608 | } | |
3609 | ||
5b363937 | 3610 | bs = bdrv_open_inherit(filename, reference, image_options, 0, |
272c02ea | 3611 | parent, child_class, child_role, errp); |
5b363937 | 3612 | if (!bs) { |
df581792 KW |
3613 | goto done; |
3614 | } | |
3615 | ||
da557aac HR |
3616 | done: |
3617 | qdict_del(options, bdref_key); | |
2d6b86af KW |
3618 | return bs; |
3619 | } | |
3620 | ||
3621 | /* | |
3622 | * Opens a disk image whose options are given as BlockdevRef in another block | |
3623 | * device's options. | |
3624 | * | |
3625 | * If allow_none is true, no image will be opened if filename is false and no | |
3626 | * BlockdevRef is given. NULL will be returned, but errp remains unset. | |
3627 | * | |
3628 | * bdrev_key specifies the key for the image's BlockdevRef in the options QDict. | |
3629 | * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict | |
3630 | * itself, all options starting with "${bdref_key}." are considered part of the | |
3631 | * BlockdevRef. | |
3632 | * | |
3633 | * The BlockdevRef will be removed from the options QDict. | |
3634 | */ | |
3635 | BdrvChild *bdrv_open_child(const char *filename, | |
3636 | QDict *options, const char *bdref_key, | |
3637 | BlockDriverState *parent, | |
bd86fb99 | 3638 | const BdrvChildClass *child_class, |
258b7765 | 3639 | BdrvChildRole child_role, |
2d6b86af KW |
3640 | bool allow_none, Error **errp) |
3641 | { | |
3642 | BlockDriverState *bs; | |
3643 | ||
f791bf7f EGE |
3644 | GLOBAL_STATE_CODE(); |
3645 | ||
bd86fb99 | 3646 | bs = bdrv_open_child_bs(filename, options, bdref_key, parent, child_class, |
272c02ea | 3647 | child_role, allow_none, errp); |
2d6b86af KW |
3648 | if (bs == NULL) { |
3649 | return NULL; | |
3650 | } | |
3651 | ||
258b7765 HR |
3652 | return bdrv_attach_child(parent, bs, bdref_key, child_class, child_role, |
3653 | errp); | |
b4b059f6 KW |
3654 | } |
3655 | ||
83930780 VSO |
3656 | /* |
3657 | * Wrapper on bdrv_open_child() for most popular case: open primary child of bs. | |
3658 | */ | |
3659 | int bdrv_open_file_child(const char *filename, | |
3660 | QDict *options, const char *bdref_key, | |
3661 | BlockDriverState *parent, Error **errp) | |
3662 | { | |
3663 | BdrvChildRole role; | |
3664 | ||
3665 | /* commit_top and mirror_top don't use this function */ | |
3666 | assert(!parent->drv->filtered_child_is_backing); | |
83930780 VSO |
3667 | role = parent->drv->is_filter ? |
3668 | (BDRV_CHILD_FILTERED | BDRV_CHILD_PRIMARY) : BDRV_CHILD_IMAGE; | |
3669 | ||
5bb04747 VSO |
3670 | if (!bdrv_open_child(filename, options, bdref_key, parent, |
3671 | &child_of_bds, role, false, errp)) | |
3672 | { | |
3673 | return -EINVAL; | |
3674 | } | |
83930780 | 3675 | |
5bb04747 | 3676 | return 0; |
83930780 VSO |
3677 | } |
3678 | ||
bd86fb99 HR |
3679 | /* |
3680 | * TODO Future callers may need to specify parent/child_class in order for | |
3681 | * option inheritance to work. Existing callers use it for the root node. | |
3682 | */ | |
e1d74bc6 KW |
3683 | BlockDriverState *bdrv_open_blockdev_ref(BlockdevRef *ref, Error **errp) |
3684 | { | |
3685 | BlockDriverState *bs = NULL; | |
e1d74bc6 KW |
3686 | QObject *obj = NULL; |
3687 | QDict *qdict = NULL; | |
3688 | const char *reference = NULL; | |
3689 | Visitor *v = NULL; | |
3690 | ||
f791bf7f EGE |
3691 | GLOBAL_STATE_CODE(); |
3692 | ||
e1d74bc6 KW |
3693 | if (ref->type == QTYPE_QSTRING) { |
3694 | reference = ref->u.reference; | |
3695 | } else { | |
3696 | BlockdevOptions *options = &ref->u.definition; | |
3697 | assert(ref->type == QTYPE_QDICT); | |
3698 | ||
3699 | v = qobject_output_visitor_new(&obj); | |
1f584248 | 3700 | visit_type_BlockdevOptions(v, NULL, &options, &error_abort); |
e1d74bc6 KW |
3701 | visit_complete(v, &obj); |
3702 | ||
7dc847eb | 3703 | qdict = qobject_to(QDict, obj); |
e1d74bc6 KW |
3704 | qdict_flatten(qdict); |
3705 | ||
3706 | /* bdrv_open_inherit() defaults to the values in bdrv_flags (for | |
3707 | * compatibility with other callers) rather than what we want as the | |
3708 | * real defaults. Apply the defaults here instead. */ | |
3709 | qdict_set_default_str(qdict, BDRV_OPT_CACHE_DIRECT, "off"); | |
3710 | qdict_set_default_str(qdict, BDRV_OPT_CACHE_NO_FLUSH, "off"); | |
3711 | qdict_set_default_str(qdict, BDRV_OPT_READ_ONLY, "off"); | |
e35bdc12 KW |
3712 | qdict_set_default_str(qdict, BDRV_OPT_AUTO_READ_ONLY, "off"); |
3713 | ||
e1d74bc6 KW |
3714 | } |
3715 | ||
272c02ea | 3716 | bs = bdrv_open_inherit(NULL, reference, qdict, 0, NULL, NULL, 0, errp); |
e1d74bc6 | 3717 | obj = NULL; |
cb3e7f08 | 3718 | qobject_unref(obj); |
e1d74bc6 KW |
3719 | visit_free(v); |
3720 | return bs; | |
3721 | } | |
3722 | ||
66836189 HR |
3723 | static BlockDriverState *bdrv_append_temp_snapshot(BlockDriverState *bs, |
3724 | int flags, | |
3725 | QDict *snapshot_options, | |
3726 | Error **errp) | |
b998875d | 3727 | { |
69fbfff9 | 3728 | g_autofree char *tmp_filename = NULL; |
b998875d | 3729 | int64_t total_size; |
83d0521a | 3730 | QemuOpts *opts = NULL; |
ff6ed714 | 3731 | BlockDriverState *bs_snapshot = NULL; |
b998875d KW |
3732 | int ret; |
3733 | ||
bdb73476 EGE |
3734 | GLOBAL_STATE_CODE(); |
3735 | ||
b998875d KW |
3736 | /* if snapshot, we create a temporary backing file and open it |
3737 | instead of opening 'filename' directly */ | |
3738 | ||
3739 | /* Get the required size from the image */ | |
f187743a KW |
3740 | total_size = bdrv_getlength(bs); |
3741 | if (total_size < 0) { | |
3742 | error_setg_errno(errp, -total_size, "Could not get image size"); | |
1ba4b6a5 | 3743 | goto out; |
f187743a | 3744 | } |
b998875d KW |
3745 | |
3746 | /* Create the temporary image */ | |
69fbfff9 BM |
3747 | tmp_filename = create_tmp_file(errp); |
3748 | if (!tmp_filename) { | |
1ba4b6a5 | 3749 | goto out; |
b998875d KW |
3750 | } |
3751 | ||
ef810437 | 3752 | opts = qemu_opts_create(bdrv_qcow2.create_opts, NULL, 0, |
c282e1fd | 3753 | &error_abort); |
39101f25 | 3754 | qemu_opt_set_number(opts, BLOCK_OPT_SIZE, total_size, &error_abort); |
e43bfd9c | 3755 | ret = bdrv_create(&bdrv_qcow2, tmp_filename, opts, errp); |
83d0521a | 3756 | qemu_opts_del(opts); |
b998875d | 3757 | if (ret < 0) { |
e43bfd9c MA |
3758 | error_prepend(errp, "Could not create temporary overlay '%s': ", |
3759 | tmp_filename); | |
1ba4b6a5 | 3760 | goto out; |
b998875d KW |
3761 | } |
3762 | ||
73176bee | 3763 | /* Prepare options QDict for the temporary file */ |
46f5ac20 EB |
3764 | qdict_put_str(snapshot_options, "file.driver", "file"); |
3765 | qdict_put_str(snapshot_options, "file.filename", tmp_filename); | |
3766 | qdict_put_str(snapshot_options, "driver", "qcow2"); | |
b998875d | 3767 | |
5b363937 | 3768 | bs_snapshot = bdrv_open(NULL, NULL, snapshot_options, flags, errp); |
73176bee | 3769 | snapshot_options = NULL; |
5b363937 | 3770 | if (!bs_snapshot) { |
1ba4b6a5 | 3771 | goto out; |
b998875d KW |
3772 | } |
3773 | ||
934aee14 VSO |
3774 | ret = bdrv_append(bs_snapshot, bs, errp); |
3775 | if (ret < 0) { | |
ff6ed714 | 3776 | bs_snapshot = NULL; |
b2c2832c KW |
3777 | goto out; |
3778 | } | |
1ba4b6a5 BC |
3779 | |
3780 | out: | |
cb3e7f08 | 3781 | qobject_unref(snapshot_options); |
ff6ed714 | 3782 | return bs_snapshot; |
b998875d KW |
3783 | } |
3784 | ||
b6ce07aa KW |
3785 | /* |
3786 | * Opens a disk image (raw, qcow2, vmdk, ...) | |
de9c0cec KW |
3787 | * |
3788 | * options is a QDict of options to pass to the block drivers, or NULL for an | |
3789 | * empty set of options. The reference to the QDict belongs to the block layer | |
3790 | * after the call (even on failure), so if the caller intends to reuse the | |
cb3e7f08 | 3791 | * dictionary, it needs to use qobject_ref() before calling bdrv_open. |
f67503e5 HR |
3792 | * |
3793 | * If *pbs is NULL, a new BDS will be created with a pointer to it stored there. | |
3794 | * If it is not NULL, the referenced BDS will be reused. | |
ddf5636d HR |
3795 | * |
3796 | * The reference parameter may be used to specify an existing block device which | |
3797 | * should be opened. If specified, neither options nor a filename may be given, | |
3798 | * nor can an existing BDS be reused (that is, *pbs has to be NULL). | |
b6ce07aa | 3799 | */ |
5b363937 HR |
3800 | static BlockDriverState *bdrv_open_inherit(const char *filename, |
3801 | const char *reference, | |
3802 | QDict *options, int flags, | |
3803 | BlockDriverState *parent, | |
bd86fb99 | 3804 | const BdrvChildClass *child_class, |
272c02ea | 3805 | BdrvChildRole child_role, |
5b363937 | 3806 | Error **errp) |
ea2384d3 | 3807 | { |
b6ce07aa | 3808 | int ret; |
5696c6e3 | 3809 | BlockBackend *file = NULL; |
9a4f4c31 | 3810 | BlockDriverState *bs; |
ce343771 | 3811 | BlockDriver *drv = NULL; |
2f624b80 | 3812 | BdrvChild *child; |
74fe54f2 | 3813 | const char *drvname; |
3e8c2e57 | 3814 | const char *backing; |
34b5d2c6 | 3815 | Error *local_err = NULL; |
73176bee | 3816 | QDict *snapshot_options = NULL; |
b1e6fc08 | 3817 | int snapshot_flags = 0; |
712e7874 | 3818 | |
bd86fb99 HR |
3819 | assert(!child_class || !flags); |
3820 | assert(!child_class == !parent); | |
f0c28327 | 3821 | GLOBAL_STATE_CODE(); |
f67503e5 | 3822 | |
ddf5636d HR |
3823 | if (reference) { |
3824 | bool options_non_empty = options ? qdict_size(options) : false; | |
cb3e7f08 | 3825 | qobject_unref(options); |
ddf5636d | 3826 | |
ddf5636d HR |
3827 | if (filename || options_non_empty) { |
3828 | error_setg(errp, "Cannot reference an existing block device with " | |
3829 | "additional options or a new filename"); | |
5b363937 | 3830 | return NULL; |
ddf5636d HR |
3831 | } |
3832 | ||
3833 | bs = bdrv_lookup_bs(reference, reference, errp); | |
3834 | if (!bs) { | |
5b363937 | 3835 | return NULL; |
ddf5636d | 3836 | } |
76b22320 | 3837 | |
ddf5636d | 3838 | bdrv_ref(bs); |
5b363937 | 3839 | return bs; |
ddf5636d HR |
3840 | } |
3841 | ||
5b363937 | 3842 | bs = bdrv_new(); |
f67503e5 | 3843 | |
de9c0cec KW |
3844 | /* NULL means an empty set of options */ |
3845 | if (options == NULL) { | |
3846 | options = qdict_new(); | |
3847 | } | |
3848 | ||
145f598e | 3849 | /* json: syntax counts as explicit options, as if in the QDict */ |
de3b53f0 KW |
3850 | parse_json_protocol(options, &filename, &local_err); |
3851 | if (local_err) { | |
de3b53f0 KW |
3852 | goto fail; |
3853 | } | |
3854 | ||
145f598e KW |
3855 | bs->explicit_options = qdict_clone_shallow(options); |
3856 | ||
bd86fb99 | 3857 | if (child_class) { |
3cdc69d3 HR |
3858 | bool parent_is_format; |
3859 | ||
3860 | if (parent->drv) { | |
3861 | parent_is_format = parent->drv->is_format; | |
3862 | } else { | |
3863 | /* | |
3864 | * parent->drv is not set yet because this node is opened for | |
3865 | * (potential) format probing. That means that @parent is going | |
3866 | * to be a format node. | |
3867 | */ | |
3868 | parent_is_format = true; | |
3869 | } | |
3870 | ||
bddcec37 | 3871 | bs->inherits_from = parent; |
3cdc69d3 HR |
3872 | child_class->inherit_options(child_role, parent_is_format, |
3873 | &flags, options, | |
bd86fb99 | 3874 | parent->open_flags, parent->options); |
f3930ed0 KW |
3875 | } |
3876 | ||
de3b53f0 | 3877 | ret = bdrv_fill_options(&options, filename, &flags, &local_err); |
dfde483e | 3878 | if (ret < 0) { |
462f5bcf KW |
3879 | goto fail; |
3880 | } | |
3881 | ||
129c7d1c MA |
3882 | /* |
3883 | * Set the BDRV_O_RDWR and BDRV_O_ALLOW_RDWR flags. | |
3884 | * Caution: getting a boolean member of @options requires care. | |
3885 | * When @options come from -blockdev or blockdev_add, members are | |
3886 | * typed according to the QAPI schema, but when they come from | |
3887 | * -drive, they're all QString. | |
3888 | */ | |
f87a0e29 AG |
3889 | if (g_strcmp0(qdict_get_try_str(options, BDRV_OPT_READ_ONLY), "on") && |
3890 | !qdict_get_try_bool(options, BDRV_OPT_READ_ONLY, false)) { | |
3891 | flags |= (BDRV_O_RDWR | BDRV_O_ALLOW_RDWR); | |
3892 | } else { | |
3893 | flags &= ~BDRV_O_RDWR; | |
14499ea5 AG |
3894 | } |
3895 | ||
3896 | if (flags & BDRV_O_SNAPSHOT) { | |
3897 | snapshot_options = qdict_new(); | |
3898 | bdrv_temp_snapshot_options(&snapshot_flags, snapshot_options, | |
3899 | flags, options); | |
f87a0e29 AG |
3900 | /* Let bdrv_backing_options() override "read-only" */ |
3901 | qdict_del(options, BDRV_OPT_READ_ONLY); | |
00ff7ffd HR |
3902 | bdrv_inherited_options(BDRV_CHILD_COW, true, |
3903 | &flags, options, flags, options); | |
14499ea5 AG |
3904 | } |
3905 | ||
62392ebb KW |
3906 | bs->open_flags = flags; |
3907 | bs->options = options; | |
3908 | options = qdict_clone_shallow(options); | |
3909 | ||
76c591b0 | 3910 | /* Find the right image format driver */ |
129c7d1c | 3911 | /* See cautionary note on accessing @options above */ |
76c591b0 KW |
3912 | drvname = qdict_get_try_str(options, "driver"); |
3913 | if (drvname) { | |
3914 | drv = bdrv_find_format(drvname); | |
76c591b0 KW |
3915 | if (!drv) { |
3916 | error_setg(errp, "Unknown driver: '%s'", drvname); | |
76c591b0 KW |
3917 | goto fail; |
3918 | } | |
3919 | } | |
3920 | ||
3921 | assert(drvname || !(flags & BDRV_O_PROTOCOL)); | |
76c591b0 | 3922 | |
129c7d1c | 3923 | /* See cautionary note on accessing @options above */ |
3e8c2e57 | 3924 | backing = qdict_get_try_str(options, "backing"); |
e59a0cf1 HR |
3925 | if (qobject_to(QNull, qdict_get(options, "backing")) != NULL || |
3926 | (backing && *backing == '\0')) | |
3927 | { | |
4f7be280 HR |
3928 | if (backing) { |
3929 | warn_report("Use of \"backing\": \"\" is deprecated; " | |
3930 | "use \"backing\": null instead"); | |
3931 | } | |
3e8c2e57 | 3932 | flags |= BDRV_O_NO_BACKING; |
ae0f57f0 KW |
3933 | qdict_del(bs->explicit_options, "backing"); |
3934 | qdict_del(bs->options, "backing"); | |
3e8c2e57 AG |
3935 | qdict_del(options, "backing"); |
3936 | } | |
3937 | ||
5696c6e3 | 3938 | /* Open image file without format layer. This BlockBackend is only used for |
4e4bf5c4 KW |
3939 | * probing, the block drivers will do their own bdrv_open_child() for the |
3940 | * same BDS, which is why we put the node name back into options. */ | |
f4788adc | 3941 | if ((flags & BDRV_O_PROTOCOL) == 0) { |
5696c6e3 KW |
3942 | BlockDriverState *file_bs; |
3943 | ||
3944 | file_bs = bdrv_open_child_bs(filename, options, "file", bs, | |
58944401 HR |
3945 | &child_of_bds, BDRV_CHILD_IMAGE, |
3946 | true, &local_err); | |
1fdd6933 | 3947 | if (local_err) { |
f4788adc KW |
3948 | goto fail; |
3949 | } | |
5696c6e3 | 3950 | if (file_bs != NULL) { |
dacaa162 KW |
3951 | /* Not requesting BLK_PERM_CONSISTENT_READ because we're only |
3952 | * looking at the header to guess the image format. This works even | |
3953 | * in cases where a guest would not see a consistent state. */ | |
d861ab3a | 3954 | file = blk_new(bdrv_get_aio_context(file_bs), 0, BLK_PERM_ALL); |
d7086422 | 3955 | blk_insert_bs(file, file_bs, &local_err); |
5696c6e3 | 3956 | bdrv_unref(file_bs); |
d7086422 KW |
3957 | if (local_err) { |
3958 | goto fail; | |
3959 | } | |
5696c6e3 | 3960 | |
46f5ac20 | 3961 | qdict_put_str(options, "file", bdrv_get_node_name(file_bs)); |
4e4bf5c4 | 3962 | } |
f500a6d3 KW |
3963 | } |
3964 | ||
76c591b0 | 3965 | /* Image format probing */ |
38f3ef57 | 3966 | bs->probed = !drv; |
76c591b0 | 3967 | if (!drv && file) { |
cf2ab8fc | 3968 | ret = find_image_format(file, filename, &drv, &local_err); |
17b005f1 | 3969 | if (ret < 0) { |
8bfea15d | 3970 | goto fail; |
2a05cbe4 | 3971 | } |
62392ebb KW |
3972 | /* |
3973 | * This option update would logically belong in bdrv_fill_options(), | |
3974 | * but we first need to open bs->file for the probing to work, while | |
3975 | * opening bs->file already requires the (mostly) final set of options | |
3976 | * so that cache mode etc. can be inherited. | |
3977 | * | |
3978 | * Adding the driver later is somewhat ugly, but it's not an option | |
3979 | * that would ever be inherited, so it's correct. We just need to make | |
3980 | * sure to update both bs->options (which has the full effective | |
3981 | * options for bs) and options (which has file.* already removed). | |
3982 | */ | |
46f5ac20 EB |
3983 | qdict_put_str(bs->options, "driver", drv->format_name); |
3984 | qdict_put_str(options, "driver", drv->format_name); | |
76c591b0 | 3985 | } else if (!drv) { |
17b005f1 | 3986 | error_setg(errp, "Must specify either driver or file"); |
8bfea15d | 3987 | goto fail; |
ea2384d3 | 3988 | } |
b6ce07aa | 3989 | |
53a29513 HR |
3990 | /* BDRV_O_PROTOCOL must be set iff a protocol BDS is about to be created */ |
3991 | assert(!!(flags & BDRV_O_PROTOCOL) == !!drv->bdrv_file_open); | |
3992 | /* file must be NULL if a protocol BDS is about to be created | |
3993 | * (the inverse results in an error message from bdrv_open_common()) */ | |
3994 | assert(!(flags & BDRV_O_PROTOCOL) || !file); | |
3995 | ||
b6ce07aa | 3996 | /* Open the image */ |
82dc8b41 | 3997 | ret = bdrv_open_common(bs, file, options, &local_err); |
b6ce07aa | 3998 | if (ret < 0) { |
8bfea15d | 3999 | goto fail; |
6987307c CH |
4000 | } |
4001 | ||
4e4bf5c4 | 4002 | if (file) { |
5696c6e3 | 4003 | blk_unref(file); |
f500a6d3 KW |
4004 | file = NULL; |
4005 | } | |
4006 | ||
b6ce07aa | 4007 | /* If there is a backing file, use it */ |
9156df12 | 4008 | if ((flags & BDRV_O_NO_BACKING) == 0) { |
d9b7b057 | 4009 | ret = bdrv_open_backing_file(bs, options, "backing", &local_err); |
b6ce07aa | 4010 | if (ret < 0) { |
b6ad491a | 4011 | goto close_and_fail; |
b6ce07aa | 4012 | } |
b6ce07aa KW |
4013 | } |
4014 | ||
50196d7a AG |
4015 | /* Remove all children options and references |
4016 | * from bs->options and bs->explicit_options */ | |
2f624b80 AG |
4017 | QLIST_FOREACH(child, &bs->children, next) { |
4018 | char *child_key_dot; | |
4019 | child_key_dot = g_strdup_printf("%s.", child->name); | |
4020 | qdict_extract_subqdict(bs->explicit_options, NULL, child_key_dot); | |
4021 | qdict_extract_subqdict(bs->options, NULL, child_key_dot); | |
50196d7a AG |
4022 | qdict_del(bs->explicit_options, child->name); |
4023 | qdict_del(bs->options, child->name); | |
2f624b80 AG |
4024 | g_free(child_key_dot); |
4025 | } | |
4026 | ||
b6ad491a | 4027 | /* Check if any unknown options were used */ |
7ad2757f | 4028 | if (qdict_size(options) != 0) { |
b6ad491a | 4029 | const QDictEntry *entry = qdict_first(options); |
5acd9d81 HR |
4030 | if (flags & BDRV_O_PROTOCOL) { |
4031 | error_setg(errp, "Block protocol '%s' doesn't support the option " | |
4032 | "'%s'", drv->format_name, entry->key); | |
4033 | } else { | |
d0e46a55 HR |
4034 | error_setg(errp, |
4035 | "Block format '%s' does not support the option '%s'", | |
4036 | drv->format_name, entry->key); | |
5acd9d81 | 4037 | } |
b6ad491a | 4038 | |
b6ad491a KW |
4039 | goto close_and_fail; |
4040 | } | |
b6ad491a | 4041 | |
c01c214b | 4042 | bdrv_parent_cb_change_media(bs, true); |
b6ce07aa | 4043 | |
cb3e7f08 | 4044 | qobject_unref(options); |
8961be33 | 4045 | options = NULL; |
dd62f1ca KW |
4046 | |
4047 | /* For snapshot=on, create a temporary qcow2 overlay. bs points to the | |
4048 | * temporary snapshot afterwards. */ | |
4049 | if (snapshot_flags) { | |
66836189 HR |
4050 | BlockDriverState *snapshot_bs; |
4051 | snapshot_bs = bdrv_append_temp_snapshot(bs, snapshot_flags, | |
4052 | snapshot_options, &local_err); | |
73176bee | 4053 | snapshot_options = NULL; |
dd62f1ca KW |
4054 | if (local_err) { |
4055 | goto close_and_fail; | |
4056 | } | |
5b363937 HR |
4057 | /* We are not going to return bs but the overlay on top of it |
4058 | * (snapshot_bs); thus, we have to drop the strong reference to bs | |
4059 | * (which we obtained by calling bdrv_new()). bs will not be deleted, | |
4060 | * though, because the overlay still has a reference to it. */ | |
4061 | bdrv_unref(bs); | |
4062 | bs = snapshot_bs; | |
dd62f1ca KW |
4063 | } |
4064 | ||
5b363937 | 4065 | return bs; |
b6ce07aa | 4066 | |
8bfea15d | 4067 | fail: |
5696c6e3 | 4068 | blk_unref(file); |
cb3e7f08 MAL |
4069 | qobject_unref(snapshot_options); |
4070 | qobject_unref(bs->explicit_options); | |
4071 | qobject_unref(bs->options); | |
4072 | qobject_unref(options); | |
de9c0cec | 4073 | bs->options = NULL; |
998cbd6a | 4074 | bs->explicit_options = NULL; |
5b363937 | 4075 | bdrv_unref(bs); |
621ff94d | 4076 | error_propagate(errp, local_err); |
5b363937 | 4077 | return NULL; |
de9c0cec | 4078 | |
b6ad491a | 4079 | close_and_fail: |
5b363937 | 4080 | bdrv_unref(bs); |
cb3e7f08 MAL |
4081 | qobject_unref(snapshot_options); |
4082 | qobject_unref(options); | |
621ff94d | 4083 | error_propagate(errp, local_err); |
5b363937 | 4084 | return NULL; |
b6ce07aa KW |
4085 | } |
4086 | ||
5b363937 HR |
4087 | BlockDriverState *bdrv_open(const char *filename, const char *reference, |
4088 | QDict *options, int flags, Error **errp) | |
f3930ed0 | 4089 | { |
f791bf7f EGE |
4090 | GLOBAL_STATE_CODE(); |
4091 | ||
5b363937 | 4092 | return bdrv_open_inherit(filename, reference, options, flags, NULL, |
272c02ea | 4093 | NULL, 0, errp); |
f3930ed0 KW |
4094 | } |
4095 | ||
faf116b4 AG |
4096 | /* Return true if the NULL-terminated @list contains @str */ |
4097 | static bool is_str_in_list(const char *str, const char *const *list) | |
4098 | { | |
4099 | if (str && list) { | |
4100 | int i; | |
4101 | for (i = 0; list[i] != NULL; i++) { | |
4102 | if (!strcmp(str, list[i])) { | |
4103 | return true; | |
4104 | } | |
4105 | } | |
4106 | } | |
4107 | return false; | |
4108 | } | |
4109 | ||
4110 | /* | |
4111 | * Check that every option set in @bs->options is also set in | |
4112 | * @new_opts. | |
4113 | * | |
4114 | * Options listed in the common_options list and in | |
4115 | * @bs->drv->mutable_opts are skipped. | |
4116 | * | |
4117 | * Return 0 on success, otherwise return -EINVAL and set @errp. | |
4118 | */ | |
4119 | static int bdrv_reset_options_allowed(BlockDriverState *bs, | |
4120 | const QDict *new_opts, Error **errp) | |
4121 | { | |
4122 | const QDictEntry *e; | |
4123 | /* These options are common to all block drivers and are handled | |
4124 | * in bdrv_reopen_prepare() so they can be left out of @new_opts */ | |
4125 | const char *const common_options[] = { | |
4126 | "node-name", "discard", "cache.direct", "cache.no-flush", | |
4127 | "read-only", "auto-read-only", "detect-zeroes", NULL | |
4128 | }; | |
4129 | ||
4130 | for (e = qdict_first(bs->options); e; e = qdict_next(bs->options, e)) { | |
4131 | if (!qdict_haskey(new_opts, e->key) && | |
4132 | !is_str_in_list(e->key, common_options) && | |
4133 | !is_str_in_list(e->key, bs->drv->mutable_opts)) { | |
4134 | error_setg(errp, "Option '%s' cannot be reset " | |
4135 | "to its default value", e->key); | |
4136 | return -EINVAL; | |
4137 | } | |
4138 | } | |
4139 | ||
4140 | return 0; | |
4141 | } | |
4142 | ||
cb828c31 AG |
4143 | /* |
4144 | * Returns true if @child can be reached recursively from @bs | |
4145 | */ | |
4146 | static bool bdrv_recurse_has_child(BlockDriverState *bs, | |
4147 | BlockDriverState *child) | |
4148 | { | |
4149 | BdrvChild *c; | |
4150 | ||
4151 | if (bs == child) { | |
4152 | return true; | |
4153 | } | |
4154 | ||
4155 | QLIST_FOREACH(c, &bs->children, next) { | |
4156 | if (bdrv_recurse_has_child(c->bs, child)) { | |
4157 | return true; | |
4158 | } | |
4159 | } | |
4160 | ||
4161 | return false; | |
4162 | } | |
4163 | ||
e971aa12 JC |
4164 | /* |
4165 | * Adds a BlockDriverState to a simple queue for an atomic, transactional | |
4166 | * reopen of multiple devices. | |
4167 | * | |
859443b0 | 4168 | * bs_queue can either be an existing BlockReopenQueue that has had QTAILQ_INIT |
e971aa12 JC |
4169 | * already performed, or alternatively may be NULL a new BlockReopenQueue will |
4170 | * be created and initialized. This newly created BlockReopenQueue should be | |
4171 | * passed back in for subsequent calls that are intended to be of the same | |
4172 | * atomic 'set'. | |
4173 | * | |
4174 | * bs is the BlockDriverState to add to the reopen queue. | |
4175 | * | |
4d2cb092 KW |
4176 | * options contains the changed options for the associated bs |
4177 | * (the BlockReopenQueue takes ownership) | |
4178 | * | |
e971aa12 JC |
4179 | * flags contains the open flags for the associated bs |
4180 | * | |
4181 | * returns a pointer to bs_queue, which is either the newly allocated | |
4182 | * bs_queue, or the existing bs_queue being used. | |
4183 | * | |
d22933ac | 4184 | * bs is drained here and undrained by bdrv_reopen_queue_free(). |
2e117866 KW |
4185 | * |
4186 | * To be called with bs->aio_context locked. | |
e971aa12 | 4187 | */ |
28518102 KW |
4188 | static BlockReopenQueue *bdrv_reopen_queue_child(BlockReopenQueue *bs_queue, |
4189 | BlockDriverState *bs, | |
4190 | QDict *options, | |
bd86fb99 | 4191 | const BdrvChildClass *klass, |
272c02ea | 4192 | BdrvChildRole role, |
3cdc69d3 | 4193 | bool parent_is_format, |
28518102 | 4194 | QDict *parent_options, |
077e8e20 AG |
4195 | int parent_flags, |
4196 | bool keep_old_opts) | |
e971aa12 JC |
4197 | { |
4198 | assert(bs != NULL); | |
4199 | ||
4200 | BlockReopenQueueEntry *bs_entry; | |
67251a31 | 4201 | BdrvChild *child; |
9aa09ddd AG |
4202 | QDict *old_options, *explicit_options, *options_copy; |
4203 | int flags; | |
4204 | QemuOpts *opts; | |
67251a31 | 4205 | |
f0c28327 | 4206 | GLOBAL_STATE_CODE(); |
1a63a907 | 4207 | |
d22933ac KW |
4208 | bdrv_drained_begin(bs); |
4209 | ||
e971aa12 JC |
4210 | if (bs_queue == NULL) { |
4211 | bs_queue = g_new0(BlockReopenQueue, 1); | |
859443b0 | 4212 | QTAILQ_INIT(bs_queue); |
e971aa12 JC |
4213 | } |
4214 | ||
4d2cb092 KW |
4215 | if (!options) { |
4216 | options = qdict_new(); | |
4217 | } | |
4218 | ||
5b7ba05f | 4219 | /* Check if this BlockDriverState is already in the queue */ |
859443b0 | 4220 | QTAILQ_FOREACH(bs_entry, bs_queue, entry) { |
5b7ba05f AG |
4221 | if (bs == bs_entry->state.bs) { |
4222 | break; | |
4223 | } | |
4224 | } | |
4225 | ||
28518102 KW |
4226 | /* |
4227 | * Precedence of options: | |
4228 | * 1. Explicitly passed in options (highest) | |
9aa09ddd AG |
4229 | * 2. Retained from explicitly set options of bs |
4230 | * 3. Inherited from parent node | |
4231 | * 4. Retained from effective options of bs | |
28518102 KW |
4232 | */ |
4233 | ||
145f598e | 4234 | /* Old explicitly set values (don't overwrite by inherited value) */ |
077e8e20 AG |
4235 | if (bs_entry || keep_old_opts) { |
4236 | old_options = qdict_clone_shallow(bs_entry ? | |
4237 | bs_entry->state.explicit_options : | |
4238 | bs->explicit_options); | |
4239 | bdrv_join_options(bs, options, old_options); | |
4240 | qobject_unref(old_options); | |
5b7ba05f | 4241 | } |
145f598e KW |
4242 | |
4243 | explicit_options = qdict_clone_shallow(options); | |
4244 | ||
28518102 KW |
4245 | /* Inherit from parent node */ |
4246 | if (parent_options) { | |
9aa09ddd | 4247 | flags = 0; |
3cdc69d3 | 4248 | klass->inherit_options(role, parent_is_format, &flags, options, |
272c02ea | 4249 | parent_flags, parent_options); |
9aa09ddd AG |
4250 | } else { |
4251 | flags = bdrv_get_flags(bs); | |
28518102 KW |
4252 | } |
4253 | ||
077e8e20 AG |
4254 | if (keep_old_opts) { |
4255 | /* Old values are used for options that aren't set yet */ | |
4256 | old_options = qdict_clone_shallow(bs->options); | |
4257 | bdrv_join_options(bs, options, old_options); | |
4258 | qobject_unref(old_options); | |
4259 | } | |
4d2cb092 | 4260 | |
9aa09ddd AG |
4261 | /* We have the final set of options so let's update the flags */ |
4262 | options_copy = qdict_clone_shallow(options); | |
4263 | opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort); | |
4264 | qemu_opts_absorb_qdict(opts, options_copy, NULL); | |
4265 | update_flags_from_options(&flags, opts); | |
4266 | qemu_opts_del(opts); | |
4267 | qobject_unref(options_copy); | |
4268 | ||
fd452021 | 4269 | /* bdrv_open_inherit() sets and clears some additional flags internally */ |
f1f25a2e | 4270 | flags &= ~BDRV_O_PROTOCOL; |
fd452021 KW |
4271 | if (flags & BDRV_O_RDWR) { |
4272 | flags |= BDRV_O_ALLOW_RDWR; | |
4273 | } | |
f1f25a2e | 4274 | |
1857c97b KW |
4275 | if (!bs_entry) { |
4276 | bs_entry = g_new0(BlockReopenQueueEntry, 1); | |
859443b0 | 4277 | QTAILQ_INSERT_TAIL(bs_queue, bs_entry, entry); |
1857c97b | 4278 | } else { |
cb3e7f08 MAL |
4279 | qobject_unref(bs_entry->state.options); |
4280 | qobject_unref(bs_entry->state.explicit_options); | |
1857c97b KW |
4281 | } |
4282 | ||
4283 | bs_entry->state.bs = bs; | |
4284 | bs_entry->state.options = options; | |
4285 | bs_entry->state.explicit_options = explicit_options; | |
4286 | bs_entry->state.flags = flags; | |
4287 | ||
8546632e AG |
4288 | /* |
4289 | * If keep_old_opts is false then it means that unspecified | |
4290 | * options must be reset to their original value. We don't allow | |
4291 | * resetting 'backing' but we need to know if the option is | |
4292 | * missing in order to decide if we have to return an error. | |
4293 | */ | |
4294 | if (!keep_old_opts) { | |
4295 | bs_entry->state.backing_missing = | |
4296 | !qdict_haskey(options, "backing") && | |
4297 | !qdict_haskey(options, "backing.driver"); | |
4298 | } | |
4299 | ||
67251a31 | 4300 | QLIST_FOREACH(child, &bs->children, next) { |
8546632e AG |
4301 | QDict *new_child_options = NULL; |
4302 | bool child_keep_old = keep_old_opts; | |
67251a31 | 4303 | |
4c9dfe5d KW |
4304 | /* reopen can only change the options of block devices that were |
4305 | * implicitly created and inherited options. For other (referenced) | |
4306 | * block devices, a syntax like "backing.foo" results in an error. */ | |
67251a31 KW |
4307 | if (child->bs->inherits_from != bs) { |
4308 | continue; | |
4309 | } | |
4310 | ||
8546632e AG |
4311 | /* Check if the options contain a child reference */ |
4312 | if (qdict_haskey(options, child->name)) { | |
4313 | const char *childref = qdict_get_try_str(options, child->name); | |
4314 | /* | |
4315 | * The current child must not be reopened if the child | |
4316 | * reference is null or points to a different node. | |
4317 | */ | |
4318 | if (g_strcmp0(childref, child->bs->node_name)) { | |
4319 | continue; | |
4320 | } | |
4321 | /* | |
4322 | * If the child reference points to the current child then | |
4323 | * reopen it with its existing set of options (note that | |
4324 | * it can still inherit new options from the parent). | |
4325 | */ | |
4326 | child_keep_old = true; | |
4327 | } else { | |
4328 | /* Extract child options ("child-name.*") */ | |
4329 | char *child_key_dot = g_strdup_printf("%s.", child->name); | |
4330 | qdict_extract_subqdict(explicit_options, NULL, child_key_dot); | |
4331 | qdict_extract_subqdict(options, &new_child_options, child_key_dot); | |
4332 | g_free(child_key_dot); | |
4333 | } | |
4c9dfe5d | 4334 | |
9aa09ddd | 4335 | bdrv_reopen_queue_child(bs_queue, child->bs, new_child_options, |
3cdc69d3 HR |
4336 | child->klass, child->role, bs->drv->is_format, |
4337 | options, flags, child_keep_old); | |
e971aa12 JC |
4338 | } |
4339 | ||
e971aa12 JC |
4340 | return bs_queue; |
4341 | } | |
4342 | ||
2e117866 | 4343 | /* To be called with bs->aio_context locked */ |
28518102 KW |
4344 | BlockReopenQueue *bdrv_reopen_queue(BlockReopenQueue *bs_queue, |
4345 | BlockDriverState *bs, | |
077e8e20 | 4346 | QDict *options, bool keep_old_opts) |
28518102 | 4347 | { |
f791bf7f EGE |
4348 | GLOBAL_STATE_CODE(); |
4349 | ||
3cdc69d3 HR |
4350 | return bdrv_reopen_queue_child(bs_queue, bs, options, NULL, 0, false, |
4351 | NULL, 0, keep_old_opts); | |
28518102 KW |
4352 | } |
4353 | ||
ab5b5228 AG |
4354 | void bdrv_reopen_queue_free(BlockReopenQueue *bs_queue) |
4355 | { | |
f791bf7f | 4356 | GLOBAL_STATE_CODE(); |
ab5b5228 AG |
4357 | if (bs_queue) { |
4358 | BlockReopenQueueEntry *bs_entry, *next; | |
4359 | QTAILQ_FOREACH_SAFE(bs_entry, bs_queue, entry, next) { | |
d22933ac KW |
4360 | AioContext *ctx = bdrv_get_aio_context(bs_entry->state.bs); |
4361 | ||
4362 | aio_context_acquire(ctx); | |
4363 | bdrv_drained_end(bs_entry->state.bs); | |
4364 | aio_context_release(ctx); | |
4365 | ||
ab5b5228 AG |
4366 | qobject_unref(bs_entry->state.explicit_options); |
4367 | qobject_unref(bs_entry->state.options); | |
4368 | g_free(bs_entry); | |
4369 | } | |
4370 | g_free(bs_queue); | |
4371 | } | |
4372 | } | |
4373 | ||
e971aa12 JC |
4374 | /* |
4375 | * Reopen multiple BlockDriverStates atomically & transactionally. | |
4376 | * | |
4377 | * The queue passed in (bs_queue) must have been built up previous | |
4378 | * via bdrv_reopen_queue(). | |
4379 | * | |
4380 | * Reopens all BDS specified in the queue, with the appropriate | |
4381 | * flags. All devices are prepared for reopen, and failure of any | |
50d6a8a3 | 4382 | * device will cause all device changes to be abandoned, and intermediate |
e971aa12 JC |
4383 | * data cleaned up. |
4384 | * | |
4385 | * If all devices prepare successfully, then the changes are committed | |
4386 | * to all devices. | |
4387 | * | |
1a63a907 KW |
4388 | * All affected nodes must be drained between bdrv_reopen_queue() and |
4389 | * bdrv_reopen_multiple(). | |
6cf42ca2 KW |
4390 | * |
4391 | * To be called from the main thread, with all other AioContexts unlocked. | |
e971aa12 | 4392 | */ |
5019aece | 4393 | int bdrv_reopen_multiple(BlockReopenQueue *bs_queue, Error **errp) |
e971aa12 JC |
4394 | { |
4395 | int ret = -1; | |
4396 | BlockReopenQueueEntry *bs_entry, *next; | |
6cf42ca2 | 4397 | AioContext *ctx; |
72373e40 | 4398 | Transaction *tran = tran_new(); |
72373e40 | 4399 | g_autoptr(GSList) refresh_list = NULL; |
e971aa12 | 4400 | |
6cf42ca2 | 4401 | assert(qemu_get_current_aio_context() == qemu_get_aio_context()); |
e971aa12 | 4402 | assert(bs_queue != NULL); |
da359909 | 4403 | GLOBAL_STATE_CODE(); |
e971aa12 | 4404 | |
a2aabf88 | 4405 | QTAILQ_FOREACH(bs_entry, bs_queue, entry) { |
6cf42ca2 KW |
4406 | ctx = bdrv_get_aio_context(bs_entry->state.bs); |
4407 | aio_context_acquire(ctx); | |
a2aabf88 | 4408 | ret = bdrv_flush(bs_entry->state.bs); |
6cf42ca2 | 4409 | aio_context_release(ctx); |
a2aabf88 VSO |
4410 | if (ret < 0) { |
4411 | error_setg_errno(errp, -ret, "Error flushing drive"); | |
e3fc91aa | 4412 | goto abort; |
a2aabf88 VSO |
4413 | } |
4414 | } | |
4415 | ||
859443b0 | 4416 | QTAILQ_FOREACH(bs_entry, bs_queue, entry) { |
1a63a907 | 4417 | assert(bs_entry->state.bs->quiesce_counter > 0); |
6cf42ca2 KW |
4418 | ctx = bdrv_get_aio_context(bs_entry->state.bs); |
4419 | aio_context_acquire(ctx); | |
72373e40 | 4420 | ret = bdrv_reopen_prepare(&bs_entry->state, bs_queue, tran, errp); |
6cf42ca2 | 4421 | aio_context_release(ctx); |
72373e40 VSO |
4422 | if (ret < 0) { |
4423 | goto abort; | |
e971aa12 JC |
4424 | } |
4425 | bs_entry->prepared = true; | |
4426 | } | |
4427 | ||
859443b0 | 4428 | QTAILQ_FOREACH(bs_entry, bs_queue, entry) { |
69b736e7 | 4429 | BDRVReopenState *state = &bs_entry->state; |
72373e40 | 4430 | |
fb0ff4d1 | 4431 | refresh_list = g_slist_prepend(refresh_list, state->bs); |
72373e40 | 4432 | if (state->old_backing_bs) { |
fb0ff4d1 | 4433 | refresh_list = g_slist_prepend(refresh_list, state->old_backing_bs); |
cb828c31 | 4434 | } |
ecd30d2d | 4435 | if (state->old_file_bs) { |
fb0ff4d1 | 4436 | refresh_list = g_slist_prepend(refresh_list, state->old_file_bs); |
ecd30d2d | 4437 | } |
72373e40 VSO |
4438 | } |
4439 | ||
4440 | /* | |
4441 | * Note that file-posix driver rely on permission update done during reopen | |
4442 | * (even if no permission changed), because it wants "new" permissions for | |
4443 | * reconfiguring the fd and that's why it does it in raw_check_perm(), not | |
4444 | * in raw_reopen_prepare() which is called with "old" permissions. | |
4445 | */ | |
4446 | ret = bdrv_list_refresh_perms(refresh_list, bs_queue, tran, errp); | |
4447 | if (ret < 0) { | |
4448 | goto abort; | |
69b736e7 KW |
4449 | } |
4450 | ||
fcd6a4f4 VSO |
4451 | /* |
4452 | * If we reach this point, we have success and just need to apply the | |
4453 | * changes. | |
4454 | * | |
4455 | * Reverse order is used to comfort qcow2 driver: on commit it need to write | |
4456 | * IN_USE flag to the image, to mark bitmaps in the image as invalid. But | |
4457 | * children are usually goes after parents in reopen-queue, so go from last | |
4458 | * to first element. | |
e971aa12 | 4459 | */ |
fcd6a4f4 | 4460 | QTAILQ_FOREACH_REVERSE(bs_entry, bs_queue, entry) { |
6cf42ca2 KW |
4461 | ctx = bdrv_get_aio_context(bs_entry->state.bs); |
4462 | aio_context_acquire(ctx); | |
e971aa12 | 4463 | bdrv_reopen_commit(&bs_entry->state); |
6cf42ca2 | 4464 | aio_context_release(ctx); |
e971aa12 JC |
4465 | } |
4466 | ||
72373e40 | 4467 | tran_commit(tran); |
69b736e7 | 4468 | |
72373e40 VSO |
4469 | QTAILQ_FOREACH_REVERSE(bs_entry, bs_queue, entry) { |
4470 | BlockDriverState *bs = bs_entry->state.bs; | |
74ad9a3b | 4471 | |
72373e40 | 4472 | if (bs->drv->bdrv_reopen_commit_post) { |
6cf42ca2 KW |
4473 | ctx = bdrv_get_aio_context(bs); |
4474 | aio_context_acquire(ctx); | |
72373e40 | 4475 | bs->drv->bdrv_reopen_commit_post(&bs_entry->state); |
6cf42ca2 | 4476 | aio_context_release(ctx); |
69b736e7 KW |
4477 | } |
4478 | } | |
17e1e2be | 4479 | |
72373e40 VSO |
4480 | ret = 0; |
4481 | goto cleanup; | |
17e1e2be | 4482 | |
72373e40 VSO |
4483 | abort: |
4484 | tran_abort(tran); | |
4485 | QTAILQ_FOREACH_SAFE(bs_entry, bs_queue, entry, next) { | |
4486 | if (bs_entry->prepared) { | |
6cf42ca2 KW |
4487 | ctx = bdrv_get_aio_context(bs_entry->state.bs); |
4488 | aio_context_acquire(ctx); | |
72373e40 | 4489 | bdrv_reopen_abort(&bs_entry->state); |
6cf42ca2 | 4490 | aio_context_release(ctx); |
17e1e2be PK |
4491 | } |
4492 | } | |
72373e40 | 4493 | |
e971aa12 | 4494 | cleanup: |
ab5b5228 | 4495 | bdrv_reopen_queue_free(bs_queue); |
40840e41 | 4496 | |
e971aa12 JC |
4497 | return ret; |
4498 | } | |
4499 | ||
6cf42ca2 KW |
4500 | int bdrv_reopen(BlockDriverState *bs, QDict *opts, bool keep_old_opts, |
4501 | Error **errp) | |
6e1000a8 | 4502 | { |
6cf42ca2 | 4503 | AioContext *ctx = bdrv_get_aio_context(bs); |
6e1000a8 | 4504 | BlockReopenQueue *queue; |
6cf42ca2 | 4505 | int ret; |
6e1000a8 | 4506 | |
f791bf7f EGE |
4507 | GLOBAL_STATE_CODE(); |
4508 | ||
2e117866 KW |
4509 | queue = bdrv_reopen_queue(NULL, bs, opts, keep_old_opts); |
4510 | ||
6cf42ca2 KW |
4511 | if (ctx != qemu_get_aio_context()) { |
4512 | aio_context_release(ctx); | |
4513 | } | |
5019aece | 4514 | ret = bdrv_reopen_multiple(queue, errp); |
6cf42ca2 KW |
4515 | |
4516 | if (ctx != qemu_get_aio_context()) { | |
4517 | aio_context_acquire(ctx); | |
4518 | } | |
6e1000a8 AG |
4519 | |
4520 | return ret; | |
4521 | } | |
4522 | ||
6cf42ca2 KW |
4523 | int bdrv_reopen_set_read_only(BlockDriverState *bs, bool read_only, |
4524 | Error **errp) | |
4525 | { | |
4526 | QDict *opts = qdict_new(); | |
4527 | ||
f791bf7f EGE |
4528 | GLOBAL_STATE_CODE(); |
4529 | ||
6cf42ca2 KW |
4530 | qdict_put_bool(opts, BDRV_OPT_READ_ONLY, read_only); |
4531 | ||
4532 | return bdrv_reopen(bs, opts, true, errp); | |
4533 | } | |
4534 | ||
cb828c31 AG |
4535 | /* |
4536 | * Take a BDRVReopenState and check if the value of 'backing' in the | |
4537 | * reopen_state->options QDict is valid or not. | |
4538 | * | |
4539 | * If 'backing' is missing from the QDict then return 0. | |
4540 | * | |
4541 | * If 'backing' contains the node name of the backing file of | |
4542 | * reopen_state->bs then return 0. | |
4543 | * | |
4544 | * If 'backing' contains a different node name (or is null) then check | |
4545 | * whether the current backing file can be replaced with the new one. | |
4546 | * If that's the case then reopen_state->replace_backing_bs is set to | |
4547 | * true and reopen_state->new_backing_bs contains a pointer to the new | |
4548 | * backing BlockDriverState (or NULL). | |
4549 | * | |
4550 | * Return 0 on success, otherwise return < 0 and set @errp. | |
4551 | */ | |
ecd30d2d AG |
4552 | static int bdrv_reopen_parse_file_or_backing(BDRVReopenState *reopen_state, |
4553 | bool is_backing, Transaction *tran, | |
4554 | Error **errp) | |
cb828c31 AG |
4555 | { |
4556 | BlockDriverState *bs = reopen_state->bs; | |
ecd30d2d AG |
4557 | BlockDriverState *new_child_bs; |
4558 | BlockDriverState *old_child_bs = is_backing ? child_bs(bs->backing) : | |
4559 | child_bs(bs->file); | |
4560 | const char *child_name = is_backing ? "backing" : "file"; | |
cb828c31 AG |
4561 | QObject *value; |
4562 | const char *str; | |
4563 | ||
bdb73476 EGE |
4564 | GLOBAL_STATE_CODE(); |
4565 | ||
ecd30d2d | 4566 | value = qdict_get(reopen_state->options, child_name); |
cb828c31 AG |
4567 | if (value == NULL) { |
4568 | return 0; | |
4569 | } | |
4570 | ||
4571 | switch (qobject_type(value)) { | |
4572 | case QTYPE_QNULL: | |
ecd30d2d AG |
4573 | assert(is_backing); /* The 'file' option does not allow a null value */ |
4574 | new_child_bs = NULL; | |
cb828c31 AG |
4575 | break; |
4576 | case QTYPE_QSTRING: | |
410f44f5 | 4577 | str = qstring_get_str(qobject_to(QString, value)); |
ecd30d2d AG |
4578 | new_child_bs = bdrv_lookup_bs(NULL, str, errp); |
4579 | if (new_child_bs == NULL) { | |
cb828c31 | 4580 | return -EINVAL; |
ecd30d2d AG |
4581 | } else if (bdrv_recurse_has_child(new_child_bs, bs)) { |
4582 | error_setg(errp, "Making '%s' a %s child of '%s' would create a " | |
4583 | "cycle", str, child_name, bs->node_name); | |
cb828c31 AG |
4584 | return -EINVAL; |
4585 | } | |
4586 | break; | |
4587 | default: | |
ecd30d2d AG |
4588 | /* |
4589 | * The options QDict has been flattened, so 'backing' and 'file' | |
4590 | * do not allow any other data type here. | |
4591 | */ | |
cb828c31 AG |
4592 | g_assert_not_reached(); |
4593 | } | |
4594 | ||
ecd30d2d AG |
4595 | if (old_child_bs == new_child_bs) { |
4596 | return 0; | |
4597 | } | |
4598 | ||
4599 | if (old_child_bs) { | |
4600 | if (bdrv_skip_implicit_filters(old_child_bs) == new_child_bs) { | |
cbfdb98c VSO |
4601 | return 0; |
4602 | } | |
4603 | ||
ecd30d2d AG |
4604 | if (old_child_bs->implicit) { |
4605 | error_setg(errp, "Cannot replace implicit %s child of %s", | |
4606 | child_name, bs->node_name); | |
cbfdb98c VSO |
4607 | return -EPERM; |
4608 | } | |
4609 | } | |
4610 | ||
ecd30d2d | 4611 | if (bs->drv->is_filter && !old_child_bs) { |
25f78d9e VSO |
4612 | /* |
4613 | * Filters always have a file or a backing child, so we are trying to | |
4614 | * change wrong child | |
4615 | */ | |
4616 | error_setg(errp, "'%s' is a %s filter node that does not support a " | |
ecd30d2d | 4617 | "%s child", bs->node_name, bs->drv->format_name, child_name); |
1d42f48c HR |
4618 | return -EINVAL; |
4619 | } | |
4620 | ||
ecd30d2d AG |
4621 | if (is_backing) { |
4622 | reopen_state->old_backing_bs = old_child_bs; | |
4623 | } else { | |
4624 | reopen_state->old_file_bs = old_child_bs; | |
4625 | } | |
4626 | ||
4627 | return bdrv_set_file_or_backing_noperm(bs, new_child_bs, is_backing, | |
4628 | tran, errp); | |
cb828c31 AG |
4629 | } |
4630 | ||
e971aa12 JC |
4631 | /* |
4632 | * Prepares a BlockDriverState for reopen. All changes are staged in the | |
4633 | * 'opaque' field of the BDRVReopenState, which is used and allocated by | |
4634 | * the block driver layer .bdrv_reopen_prepare() | |
4635 | * | |
4636 | * bs is the BlockDriverState to reopen | |
4637 | * flags are the new open flags | |
4638 | * queue is the reopen queue | |
4639 | * | |
4640 | * Returns 0 on success, non-zero on error. On error errp will be set | |
4641 | * as well. | |
4642 | * | |
4643 | * On failure, bdrv_reopen_abort() will be called to clean up any data. | |
4644 | * It is the responsibility of the caller to then call the abort() or | |
4645 | * commit() for any other BDS that have been left in a prepare() state | |
4646 | * | |
4647 | */ | |
53e96d1e | 4648 | static int bdrv_reopen_prepare(BDRVReopenState *reopen_state, |
72373e40 | 4649 | BlockReopenQueue *queue, |
ecd30d2d | 4650 | Transaction *change_child_tran, Error **errp) |
e971aa12 JC |
4651 | { |
4652 | int ret = -1; | |
e6d79c41 | 4653 | int old_flags; |
e971aa12 JC |
4654 | Error *local_err = NULL; |
4655 | BlockDriver *drv; | |
ccf9dc07 | 4656 | QemuOpts *opts; |
4c8350fe | 4657 | QDict *orig_reopen_opts; |
593b3071 | 4658 | char *discard = NULL; |
3d8ce171 | 4659 | bool read_only; |
9ad08c44 | 4660 | bool drv_prepared = false; |
e971aa12 JC |
4661 | |
4662 | assert(reopen_state != NULL); | |
4663 | assert(reopen_state->bs->drv != NULL); | |
da359909 | 4664 | GLOBAL_STATE_CODE(); |
e971aa12 JC |
4665 | drv = reopen_state->bs->drv; |
4666 | ||
4c8350fe AG |
4667 | /* This function and each driver's bdrv_reopen_prepare() remove |
4668 | * entries from reopen_state->options as they are processed, so | |
4669 | * we need to make a copy of the original QDict. */ | |
4670 | orig_reopen_opts = qdict_clone_shallow(reopen_state->options); | |
4671 | ||
ccf9dc07 KW |
4672 | /* Process generic block layer options */ |
4673 | opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort); | |
af175e85 | 4674 | if (!qemu_opts_absorb_qdict(opts, reopen_state->options, errp)) { |
ccf9dc07 KW |
4675 | ret = -EINVAL; |
4676 | goto error; | |
4677 | } | |
4678 | ||
e6d79c41 AG |
4679 | /* This was already called in bdrv_reopen_queue_child() so the flags |
4680 | * are up-to-date. This time we simply want to remove the options from | |
4681 | * QemuOpts in order to indicate that they have been processed. */ | |
4682 | old_flags = reopen_state->flags; | |
91a097e7 | 4683 | update_flags_from_options(&reopen_state->flags, opts); |
e6d79c41 | 4684 | assert(old_flags == reopen_state->flags); |
91a097e7 | 4685 | |
415bbca8 | 4686 | discard = qemu_opt_get_del(opts, BDRV_OPT_DISCARD); |
593b3071 AG |
4687 | if (discard != NULL) { |
4688 | if (bdrv_parse_discard_flags(discard, &reopen_state->flags) != 0) { | |
4689 | error_setg(errp, "Invalid discard option"); | |
4690 | ret = -EINVAL; | |
4691 | goto error; | |
4692 | } | |
4693 | } | |
4694 | ||
543770bd AG |
4695 | reopen_state->detect_zeroes = |
4696 | bdrv_parse_detect_zeroes(opts, reopen_state->flags, &local_err); | |
4697 | if (local_err) { | |
4698 | error_propagate(errp, local_err); | |
4699 | ret = -EINVAL; | |
4700 | goto error; | |
4701 | } | |
4702 | ||
57f9db9a AG |
4703 | /* All other options (including node-name and driver) must be unchanged. |
4704 | * Put them back into the QDict, so that they are checked at the end | |
4705 | * of this function. */ | |
4706 | qemu_opts_to_qdict(opts, reopen_state->options); | |
ccf9dc07 | 4707 | |
3d8ce171 JC |
4708 | /* If we are to stay read-only, do not allow permission change |
4709 | * to r/w. Attempting to set to r/w may fail if either BDRV_O_ALLOW_RDWR is | |
4710 | * not set, or if the BDS still has copy_on_read enabled */ | |
4711 | read_only = !(reopen_state->flags & BDRV_O_RDWR); | |
54a32bfe | 4712 | ret = bdrv_can_set_read_only(reopen_state->bs, read_only, true, &local_err); |
3d8ce171 JC |
4713 | if (local_err) { |
4714 | error_propagate(errp, local_err); | |
e971aa12 JC |
4715 | goto error; |
4716 | } | |
4717 | ||
e971aa12 | 4718 | if (drv->bdrv_reopen_prepare) { |
faf116b4 AG |
4719 | /* |
4720 | * If a driver-specific option is missing, it means that we | |
4721 | * should reset it to its default value. | |
4722 | * But not all options allow that, so we need to check it first. | |
4723 | */ | |
4724 | ret = bdrv_reset_options_allowed(reopen_state->bs, | |
4725 | reopen_state->options, errp); | |
4726 | if (ret) { | |
4727 | goto error; | |
4728 | } | |
4729 | ||
e971aa12 JC |
4730 | ret = drv->bdrv_reopen_prepare(reopen_state, queue, &local_err); |
4731 | if (ret) { | |
4732 | if (local_err != NULL) { | |
4733 | error_propagate(errp, local_err); | |
4734 | } else { | |
f30c66ba | 4735 | bdrv_refresh_filename(reopen_state->bs); |
d8b6895f LC |
4736 | error_setg(errp, "failed while preparing to reopen image '%s'", |
4737 | reopen_state->bs->filename); | |
e971aa12 JC |
4738 | } |
4739 | goto error; | |
4740 | } | |
4741 | } else { | |
4742 | /* It is currently mandatory to have a bdrv_reopen_prepare() | |
4743 | * handler for each supported drv. */ | |
81e5f78a AG |
4744 | error_setg(errp, "Block format '%s' used by node '%s' " |
4745 | "does not support reopening files", drv->format_name, | |
4746 | bdrv_get_device_or_node_name(reopen_state->bs)); | |
e971aa12 JC |
4747 | ret = -1; |
4748 | goto error; | |
4749 | } | |
4750 | ||
9ad08c44 HR |
4751 | drv_prepared = true; |
4752 | ||
bacd9b87 AG |
4753 | /* |
4754 | * We must provide the 'backing' option if the BDS has a backing | |
4755 | * file or if the image file has a backing file name as part of | |
4756 | * its metadata. Otherwise the 'backing' option can be omitted. | |
4757 | */ | |
4758 | if (drv->supports_backing && reopen_state->backing_missing && | |
1d42f48c | 4759 | (reopen_state->bs->backing || reopen_state->bs->backing_file[0])) { |
8546632e AG |
4760 | error_setg(errp, "backing is missing for '%s'", |
4761 | reopen_state->bs->node_name); | |
4762 | ret = -EINVAL; | |
4763 | goto error; | |
4764 | } | |
4765 | ||
cb828c31 AG |
4766 | /* |
4767 | * Allow changing the 'backing' option. The new value can be | |
4768 | * either a reference to an existing node (using its node name) | |
4769 | * or NULL to simply detach the current backing file. | |
4770 | */ | |
ecd30d2d AG |
4771 | ret = bdrv_reopen_parse_file_or_backing(reopen_state, true, |
4772 | change_child_tran, errp); | |
cb828c31 AG |
4773 | if (ret < 0) { |
4774 | goto error; | |
4775 | } | |
4776 | qdict_del(reopen_state->options, "backing"); | |
4777 | ||
ecd30d2d AG |
4778 | /* Allow changing the 'file' option. In this case NULL is not allowed */ |
4779 | ret = bdrv_reopen_parse_file_or_backing(reopen_state, false, | |
4780 | change_child_tran, errp); | |
4781 | if (ret < 0) { | |
4782 | goto error; | |
4783 | } | |
4784 | qdict_del(reopen_state->options, "file"); | |
4785 | ||
4d2cb092 KW |
4786 | /* Options that are not handled are only okay if they are unchanged |
4787 | * compared to the old state. It is expected that some options are only | |
4788 | * used for the initial open, but not reopen (e.g. filename) */ | |
4789 | if (qdict_size(reopen_state->options)) { | |
4790 | const QDictEntry *entry = qdict_first(reopen_state->options); | |
4791 | ||
4792 | do { | |
54fd1b0d HR |
4793 | QObject *new = entry->value; |
4794 | QObject *old = qdict_get(reopen_state->bs->options, entry->key); | |
4795 | ||
db905283 AG |
4796 | /* Allow child references (child_name=node_name) as long as they |
4797 | * point to the current child (i.e. everything stays the same). */ | |
4798 | if (qobject_type(new) == QTYPE_QSTRING) { | |
4799 | BdrvChild *child; | |
4800 | QLIST_FOREACH(child, &reopen_state->bs->children, next) { | |
4801 | if (!strcmp(child->name, entry->key)) { | |
4802 | break; | |
4803 | } | |
4804 | } | |
4805 | ||
4806 | if (child) { | |
410f44f5 MA |
4807 | if (!strcmp(child->bs->node_name, |
4808 | qstring_get_str(qobject_to(QString, new)))) { | |
db905283 AG |
4809 | continue; /* Found child with this name, skip option */ |
4810 | } | |
4811 | } | |
4812 | } | |
4813 | ||
129c7d1c | 4814 | /* |
54fd1b0d HR |
4815 | * TODO: When using -drive to specify blockdev options, all values |
4816 | * will be strings; however, when using -blockdev, blockdev-add or | |
4817 | * filenames using the json:{} pseudo-protocol, they will be | |
4818 | * correctly typed. | |
4819 | * In contrast, reopening options are (currently) always strings | |
4820 | * (because you can only specify them through qemu-io; all other | |
4821 | * callers do not specify any options). | |
4822 | * Therefore, when using anything other than -drive to create a BDS, | |
4823 | * this cannot detect non-string options as unchanged, because | |
4824 | * qobject_is_equal() always returns false for objects of different | |
4825 | * type. In the future, this should be remedied by correctly typing | |
4826 | * all options. For now, this is not too big of an issue because | |
4827 | * the user can simply omit options which cannot be changed anyway, | |
4828 | * so they will stay unchanged. | |
129c7d1c | 4829 | */ |
54fd1b0d | 4830 | if (!qobject_is_equal(new, old)) { |
4d2cb092 KW |
4831 | error_setg(errp, "Cannot change the option '%s'", entry->key); |
4832 | ret = -EINVAL; | |
4833 | goto error; | |
4834 | } | |
4835 | } while ((entry = qdict_next(reopen_state->options, entry))); | |
4836 | } | |
4837 | ||
e971aa12 JC |
4838 | ret = 0; |
4839 | ||
4c8350fe AG |
4840 | /* Restore the original reopen_state->options QDict */ |
4841 | qobject_unref(reopen_state->options); | |
4842 | reopen_state->options = qobject_ref(orig_reopen_opts); | |
4843 | ||
e971aa12 | 4844 | error: |
9ad08c44 HR |
4845 | if (ret < 0 && drv_prepared) { |
4846 | /* drv->bdrv_reopen_prepare() has succeeded, so we need to | |
4847 | * call drv->bdrv_reopen_abort() before signaling an error | |
4848 | * (bdrv_reopen_multiple() will not call bdrv_reopen_abort() | |
4849 | * when the respective bdrv_reopen_prepare() has failed) */ | |
4850 | if (drv->bdrv_reopen_abort) { | |
4851 | drv->bdrv_reopen_abort(reopen_state); | |
4852 | } | |
4853 | } | |
ccf9dc07 | 4854 | qemu_opts_del(opts); |
4c8350fe | 4855 | qobject_unref(orig_reopen_opts); |
593b3071 | 4856 | g_free(discard); |
e971aa12 JC |
4857 | return ret; |
4858 | } | |
4859 | ||
4860 | /* | |
4861 | * Takes the staged changes for the reopen from bdrv_reopen_prepare(), and | |
4862 | * makes them final by swapping the staging BlockDriverState contents into | |
4863 | * the active BlockDriverState contents. | |
4864 | */ | |
53e96d1e | 4865 | static void bdrv_reopen_commit(BDRVReopenState *reopen_state) |
e971aa12 JC |
4866 | { |
4867 | BlockDriver *drv; | |
50bf65ba | 4868 | BlockDriverState *bs; |
50196d7a | 4869 | BdrvChild *child; |
e971aa12 JC |
4870 | |
4871 | assert(reopen_state != NULL); | |
50bf65ba VSO |
4872 | bs = reopen_state->bs; |
4873 | drv = bs->drv; | |
e971aa12 | 4874 | assert(drv != NULL); |
da359909 | 4875 | GLOBAL_STATE_CODE(); |
e971aa12 JC |
4876 | |
4877 | /* If there are any driver level actions to take */ | |
4878 | if (drv->bdrv_reopen_commit) { | |
4879 | drv->bdrv_reopen_commit(reopen_state); | |
4880 | } | |
4881 | ||
4882 | /* set BDS specific flags now */ | |
cb3e7f08 | 4883 | qobject_unref(bs->explicit_options); |
4c8350fe | 4884 | qobject_unref(bs->options); |
ab5b5228 AG |
4885 | qobject_ref(reopen_state->explicit_options); |
4886 | qobject_ref(reopen_state->options); | |
145f598e | 4887 | |
50bf65ba | 4888 | bs->explicit_options = reopen_state->explicit_options; |
4c8350fe | 4889 | bs->options = reopen_state->options; |
50bf65ba | 4890 | bs->open_flags = reopen_state->flags; |
543770bd | 4891 | bs->detect_zeroes = reopen_state->detect_zeroes; |
355ef4ac | 4892 | |
50196d7a AG |
4893 | /* Remove child references from bs->options and bs->explicit_options. |
4894 | * Child options were already removed in bdrv_reopen_queue_child() */ | |
4895 | QLIST_FOREACH(child, &bs->children, next) { | |
4896 | qdict_del(bs->explicit_options, child->name); | |
4897 | qdict_del(bs->options, child->name); | |
4898 | } | |
3d0e8743 VSO |
4899 | /* backing is probably removed, so it's not handled by previous loop */ |
4900 | qdict_del(bs->explicit_options, "backing"); | |
4901 | qdict_del(bs->options, "backing"); | |
4902 | ||
1e4c797c | 4903 | bdrv_refresh_limits(bs, NULL, NULL); |
e971aa12 JC |
4904 | } |
4905 | ||
4906 | /* | |
4907 | * Abort the reopen, and delete and free the staged changes in | |
4908 | * reopen_state | |
4909 | */ | |
53e96d1e | 4910 | static void bdrv_reopen_abort(BDRVReopenState *reopen_state) |
e971aa12 JC |
4911 | { |
4912 | BlockDriver *drv; | |
4913 | ||
4914 | assert(reopen_state != NULL); | |
4915 | drv = reopen_state->bs->drv; | |
4916 | assert(drv != NULL); | |
da359909 | 4917 | GLOBAL_STATE_CODE(); |
e971aa12 JC |
4918 | |
4919 | if (drv->bdrv_reopen_abort) { | |
4920 | drv->bdrv_reopen_abort(reopen_state); | |
4921 | } | |
4922 | } | |
4923 | ||
4924 | ||
64dff520 | 4925 | static void bdrv_close(BlockDriverState *bs) |
fc01f7e7 | 4926 | { |
33384421 | 4927 | BdrvAioNotifier *ban, *ban_next; |
50a3efb0 | 4928 | BdrvChild *child, *next; |
33384421 | 4929 | |
f791bf7f | 4930 | GLOBAL_STATE_CODE(); |
30f55fb8 | 4931 | assert(!bs->refcnt); |
99b7e775 | 4932 | |
fc27291d | 4933 | bdrv_drained_begin(bs); /* complete I/O */ |
58fda173 | 4934 | bdrv_flush(bs); |
53ec73e2 | 4935 | bdrv_drain(bs); /* in case flush left pending I/O */ |
fc27291d | 4936 | |
3cbc002c | 4937 | if (bs->drv) { |
3c005293 | 4938 | if (bs->drv->bdrv_close) { |
7b99a266 | 4939 | /* Must unfreeze all children, so bdrv_unref_child() works */ |
3c005293 VSO |
4940 | bs->drv->bdrv_close(bs); |
4941 | } | |
9a4f4c31 | 4942 | bs->drv = NULL; |
50a3efb0 | 4943 | } |
9a7dedbc | 4944 | |
50a3efb0 | 4945 | QLIST_FOREACH_SAFE(child, &bs->children, next, next) { |
dd4118c7 | 4946 | bdrv_unref_child(bs, child); |
b338082b | 4947 | } |
98f90dba | 4948 | |
5bb04747 VSO |
4949 | assert(!bs->backing); |
4950 | assert(!bs->file); | |
50a3efb0 AG |
4951 | g_free(bs->opaque); |
4952 | bs->opaque = NULL; | |
d73415a3 | 4953 | qatomic_set(&bs->copy_on_read, 0); |
50a3efb0 AG |
4954 | bs->backing_file[0] = '\0'; |
4955 | bs->backing_format[0] = '\0'; | |
4956 | bs->total_sectors = 0; | |
4957 | bs->encrypted = false; | |
4958 | bs->sg = false; | |
cb3e7f08 MAL |
4959 | qobject_unref(bs->options); |
4960 | qobject_unref(bs->explicit_options); | |
50a3efb0 AG |
4961 | bs->options = NULL; |
4962 | bs->explicit_options = NULL; | |
cb3e7f08 | 4963 | qobject_unref(bs->full_open_options); |
50a3efb0 | 4964 | bs->full_open_options = NULL; |
0bc329fb HR |
4965 | g_free(bs->block_status_cache); |
4966 | bs->block_status_cache = NULL; | |
50a3efb0 | 4967 | |
cca43ae1 VSO |
4968 | bdrv_release_named_dirty_bitmaps(bs); |
4969 | assert(QLIST_EMPTY(&bs->dirty_bitmaps)); | |
4970 | ||
33384421 HR |
4971 | QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_next) { |
4972 | g_free(ban); | |
4973 | } | |
4974 | QLIST_INIT(&bs->aio_notifiers); | |
fc27291d | 4975 | bdrv_drained_end(bs); |
1a6d3bd2 GK |
4976 | |
4977 | /* | |
4978 | * If we're still inside some bdrv_drain_all_begin()/end() sections, end | |
4979 | * them now since this BDS won't exist anymore when bdrv_drain_all_end() | |
4980 | * gets called. | |
4981 | */ | |
4982 | if (bs->quiesce_counter) { | |
4983 | bdrv_drain_all_end_quiesce(bs); | |
4984 | } | |
b338082b FB |
4985 | } |
4986 | ||
2bc93fed MK |
4987 | void bdrv_close_all(void) |
4988 | { | |
f791bf7f | 4989 | GLOBAL_STATE_CODE(); |
880eeec6 | 4990 | assert(job_next(NULL) == NULL); |
ca9bd24c HR |
4991 | |
4992 | /* Drop references from requests still in flight, such as canceled block | |
4993 | * jobs whose AIO context has not been polled yet */ | |
4994 | bdrv_drain_all(); | |
2bc93fed | 4995 | |
ca9bd24c HR |
4996 | blk_remove_all_bs(); |
4997 | blockdev_close_all_bdrv_states(); | |
ed78cda3 | 4998 | |
a1a2af07 | 4999 | assert(QTAILQ_EMPTY(&all_bdrv_states)); |
2bc93fed MK |
5000 | } |
5001 | ||
d0ac0380 KW |
5002 | static bool should_update_child(BdrvChild *c, BlockDriverState *to) |
5003 | { | |
2f30b7c3 VSO |
5004 | GQueue *queue; |
5005 | GHashTable *found; | |
5006 | bool ret; | |
d0ac0380 | 5007 | |
bd86fb99 | 5008 | if (c->klass->stay_at_node) { |
d0ac0380 KW |
5009 | return false; |
5010 | } | |
5011 | ||
ec9f10fe HR |
5012 | /* If the child @c belongs to the BDS @to, replacing the current |
5013 | * c->bs by @to would mean to create a loop. | |
5014 | * | |
5015 | * Such a case occurs when appending a BDS to a backing chain. | |
5016 | * For instance, imagine the following chain: | |
5017 | * | |
5018 | * guest device -> node A -> further backing chain... | |
5019 | * | |
5020 | * Now we create a new BDS B which we want to put on top of this | |
5021 | * chain, so we first attach A as its backing node: | |
5022 | * | |
5023 | * node B | |
5024 | * | | |
5025 | * v | |
5026 | * guest device -> node A -> further backing chain... | |
5027 | * | |
5028 | * Finally we want to replace A by B. When doing that, we want to | |
5029 | * replace all pointers to A by pointers to B -- except for the | |
5030 | * pointer from B because (1) that would create a loop, and (2) | |
5031 | * that pointer should simply stay intact: | |
5032 | * | |
5033 | * guest device -> node B | |
5034 | * | | |
5035 | * v | |
5036 | * node A -> further backing chain... | |
5037 | * | |
5038 | * In general, when replacing a node A (c->bs) by a node B (@to), | |
5039 | * if A is a child of B, that means we cannot replace A by B there | |
5040 | * because that would create a loop. Silently detaching A from B | |
5041 | * is also not really an option. So overall just leaving A in | |
2f30b7c3 VSO |
5042 | * place there is the most sensible choice. |
5043 | * | |
5044 | * We would also create a loop in any cases where @c is only | |
5045 | * indirectly referenced by @to. Prevent this by returning false | |
5046 | * if @c is found (by breadth-first search) anywhere in the whole | |
5047 | * subtree of @to. | |
5048 | */ | |
5049 | ||
5050 | ret = true; | |
5051 | found = g_hash_table_new(NULL, NULL); | |
5052 | g_hash_table_add(found, to); | |
5053 | queue = g_queue_new(); | |
5054 | g_queue_push_tail(queue, to); | |
5055 | ||
5056 | while (!g_queue_is_empty(queue)) { | |
5057 | BlockDriverState *v = g_queue_pop_head(queue); | |
5058 | BdrvChild *c2; | |
5059 | ||
5060 | QLIST_FOREACH(c2, &v->children, next) { | |
5061 | if (c2 == c) { | |
5062 | ret = false; | |
5063 | break; | |
5064 | } | |
5065 | ||
5066 | if (g_hash_table_contains(found, c2->bs)) { | |
5067 | continue; | |
5068 | } | |
5069 | ||
5070 | g_queue_push_tail(queue, c2->bs); | |
5071 | g_hash_table_add(found, c2->bs); | |
d0ac0380 KW |
5072 | } |
5073 | } | |
5074 | ||
2f30b7c3 VSO |
5075 | g_queue_free(queue); |
5076 | g_hash_table_destroy(found); | |
5077 | ||
5078 | return ret; | |
d0ac0380 KW |
5079 | } |
5080 | ||
57f08941 | 5081 | static void bdrv_remove_child_commit(void *opaque) |
46541ee5 | 5082 | { |
bdb73476 | 5083 | GLOBAL_STATE_CODE(); |
5bb04747 | 5084 | bdrv_child_free(opaque); |
82b54cf5 HR |
5085 | } |
5086 | ||
57f08941 VSO |
5087 | static TransactionActionDrv bdrv_remove_child_drv = { |
5088 | .commit = bdrv_remove_child_commit, | |
46541ee5 VSO |
5089 | }; |
5090 | ||
57f08941 VSO |
5091 | /* Function doesn't update permissions, caller is responsible for this. */ |
5092 | static void bdrv_remove_child(BdrvChild *child, Transaction *tran) | |
46541ee5 | 5093 | { |
46541ee5 VSO |
5094 | if (!child) { |
5095 | return; | |
5096 | } | |
5097 | ||
5098 | if (child->bs) { | |
23987471 KW |
5099 | BlockDriverState *bs = child->bs; |
5100 | bdrv_drained_begin(bs); | |
a2c37a30 | 5101 | bdrv_replace_child_tran(child, NULL, tran); |
23987471 | 5102 | bdrv_drained_end(bs); |
46541ee5 VSO |
5103 | } |
5104 | ||
57f08941 | 5105 | tran_add(tran, &bdrv_remove_child_drv, child); |
46541ee5 VSO |
5106 | } |
5107 | ||
23987471 KW |
5108 | static void undrain_on_clean_cb(void *opaque) |
5109 | { | |
5110 | bdrv_drained_end(opaque); | |
5111 | } | |
5112 | ||
5113 | static TransactionActionDrv undrain_on_clean = { | |
5114 | .clean = undrain_on_clean_cb, | |
5115 | }; | |
5116 | ||
117caba9 VSO |
5117 | static int bdrv_replace_node_noperm(BlockDriverState *from, |
5118 | BlockDriverState *to, | |
5119 | bool auto_skip, Transaction *tran, | |
5120 | Error **errp) | |
5121 | { | |
5122 | BdrvChild *c, *next; | |
5123 | ||
bdb73476 | 5124 | GLOBAL_STATE_CODE(); |
82b54cf5 | 5125 | |
23987471 KW |
5126 | bdrv_drained_begin(from); |
5127 | bdrv_drained_begin(to); | |
5128 | tran_add(tran, &undrain_on_clean, from); | |
5129 | tran_add(tran, &undrain_on_clean, to); | |
5130 | ||
117caba9 VSO |
5131 | QLIST_FOREACH_SAFE(c, &from->parents, next_parent, next) { |
5132 | assert(c->bs == from); | |
5133 | if (!should_update_child(c, to)) { | |
5134 | if (auto_skip) { | |
5135 | continue; | |
5136 | } | |
5137 | error_setg(errp, "Should not change '%s' link to '%s'", | |
5138 | c->name, from->node_name); | |
5139 | return -EINVAL; | |
5140 | } | |
5141 | if (c->frozen) { | |
5142 | error_setg(errp, "Cannot change '%s' link to '%s'", | |
5143 | c->name, from->node_name); | |
5144 | return -EPERM; | |
5145 | } | |
0f0b1e29 | 5146 | bdrv_replace_child_tran(c, to, tran); |
117caba9 VSO |
5147 | } |
5148 | ||
5149 | return 0; | |
5150 | } | |
5151 | ||
313274bb VSO |
5152 | /* |
5153 | * With auto_skip=true bdrv_replace_node_common skips updating from parents | |
5154 | * if it creates a parent-child relation loop or if parent is block-job. | |
5155 | * | |
5156 | * With auto_skip=false the error is returned if from has a parent which should | |
5157 | * not be updated. | |
3108a15c VSO |
5158 | * |
5159 | * With @detach_subchain=true @to must be in a backing chain of @from. In this | |
5160 | * case backing link of the cow-parent of @to is removed. | |
313274bb | 5161 | */ |
a1e708fc VSO |
5162 | static int bdrv_replace_node_common(BlockDriverState *from, |
5163 | BlockDriverState *to, | |
3108a15c VSO |
5164 | bool auto_skip, bool detach_subchain, |
5165 | Error **errp) | |
dd62f1ca | 5166 | { |
3bb0e298 | 5167 | Transaction *tran = tran_new(); |
3bb0e298 | 5168 | g_autoptr(GSList) refresh_list = NULL; |
2d369d6e | 5169 | BlockDriverState *to_cow_parent = NULL; |
234ac1a9 KW |
5170 | int ret; |
5171 | ||
bdb73476 | 5172 | GLOBAL_STATE_CODE(); |
82b54cf5 | 5173 | |
3108a15c VSO |
5174 | if (detach_subchain) { |
5175 | assert(bdrv_chain_contains(from, to)); | |
5176 | assert(from != to); | |
5177 | for (to_cow_parent = from; | |
5178 | bdrv_filter_or_cow_bs(to_cow_parent) != to; | |
5179 | to_cow_parent = bdrv_filter_or_cow_bs(to_cow_parent)) | |
5180 | { | |
5181 | ; | |
5182 | } | |
5183 | } | |
5184 | ||
234ac1a9 KW |
5185 | /* Make sure that @from doesn't go away until we have successfully attached |
5186 | * all of its parents to @to. */ | |
5187 | bdrv_ref(from); | |
dd62f1ca | 5188 | |
f871abd6 | 5189 | assert(qemu_get_current_aio_context() == qemu_get_aio_context()); |
30dd65f3 | 5190 | assert(bdrv_get_aio_context(from) == bdrv_get_aio_context(to)); |
f871abd6 KW |
5191 | bdrv_drained_begin(from); |
5192 | ||
3bb0e298 VSO |
5193 | /* |
5194 | * Do the replacement without permission update. | |
5195 | * Replacement may influence the permissions, we should calculate new | |
5196 | * permissions based on new graph. If we fail, we'll roll-back the | |
5197 | * replacement. | |
5198 | */ | |
117caba9 VSO |
5199 | ret = bdrv_replace_node_noperm(from, to, auto_skip, tran, errp); |
5200 | if (ret < 0) { | |
5201 | goto out; | |
234ac1a9 KW |
5202 | } |
5203 | ||
3108a15c | 5204 | if (detach_subchain) { |
f38eaec4 | 5205 | bdrv_remove_child(bdrv_filter_or_cow_child(to_cow_parent), tran); |
3108a15c VSO |
5206 | } |
5207 | ||
fb0ff4d1 VSO |
5208 | refresh_list = g_slist_prepend(refresh_list, to); |
5209 | refresh_list = g_slist_prepend(refresh_list, from); | |
9bd910e2 | 5210 | |
3bb0e298 VSO |
5211 | ret = bdrv_list_refresh_perms(refresh_list, NULL, tran, errp); |
5212 | if (ret < 0) { | |
5213 | goto out; | |
dd62f1ca | 5214 | } |
234ac1a9 | 5215 | |
a1e708fc VSO |
5216 | ret = 0; |
5217 | ||
234ac1a9 | 5218 | out: |
3bb0e298 VSO |
5219 | tran_finalize(tran, ret); |
5220 | ||
f871abd6 | 5221 | bdrv_drained_end(from); |
234ac1a9 | 5222 | bdrv_unref(from); |
a1e708fc VSO |
5223 | |
5224 | return ret; | |
dd62f1ca KW |
5225 | } |
5226 | ||
a1e708fc VSO |
5227 | int bdrv_replace_node(BlockDriverState *from, BlockDriverState *to, |
5228 | Error **errp) | |
313274bb | 5229 | { |
f791bf7f EGE |
5230 | GLOBAL_STATE_CODE(); |
5231 | ||
3108a15c VSO |
5232 | return bdrv_replace_node_common(from, to, true, false, errp); |
5233 | } | |
5234 | ||
5235 | int bdrv_drop_filter(BlockDriverState *bs, Error **errp) | |
5236 | { | |
f791bf7f EGE |
5237 | GLOBAL_STATE_CODE(); |
5238 | ||
3108a15c VSO |
5239 | return bdrv_replace_node_common(bs, bdrv_filter_or_cow_bs(bs), true, true, |
5240 | errp); | |
313274bb VSO |
5241 | } |
5242 | ||
4ddc07ca PB |
5243 | /* |
5244 | * Add new bs contents at the top of an image chain while the chain is | |
5245 | * live, while keeping required fields on the top layer. | |
5246 | * | |
5247 | * This will modify the BlockDriverState fields, and swap contents | |
5248 | * between bs_new and bs_top. Both bs_new and bs_top are modified. | |
5249 | * | |
2272edcf VSO |
5250 | * bs_new must not be attached to a BlockBackend and must not have backing |
5251 | * child. | |
4ddc07ca PB |
5252 | * |
5253 | * This function does not create any image files. | |
5254 | */ | |
a1e708fc VSO |
5255 | int bdrv_append(BlockDriverState *bs_new, BlockDriverState *bs_top, |
5256 | Error **errp) | |
4ddc07ca | 5257 | { |
2272edcf | 5258 | int ret; |
5bb04747 | 5259 | BdrvChild *child; |
2272edcf VSO |
5260 | Transaction *tran = tran_new(); |
5261 | ||
f791bf7f EGE |
5262 | GLOBAL_STATE_CODE(); |
5263 | ||
2272edcf VSO |
5264 | assert(!bs_new->backing); |
5265 | ||
5bb04747 VSO |
5266 | child = bdrv_attach_child_noperm(bs_new, bs_top, "backing", |
5267 | &child_of_bds, bdrv_backing_role(bs_new), | |
5268 | tran, errp); | |
5269 | if (!child) { | |
5270 | ret = -EINVAL; | |
2272edcf | 5271 | goto out; |
b2c2832c | 5272 | } |
dd62f1ca | 5273 | |
2272edcf | 5274 | ret = bdrv_replace_node_noperm(bs_top, bs_new, true, tran, errp); |
a1e708fc | 5275 | if (ret < 0) { |
2272edcf | 5276 | goto out; |
234ac1a9 | 5277 | } |
4ddc07ca | 5278 | |
f1316edb | 5279 | ret = bdrv_refresh_perms(bs_new, tran, errp); |
2272edcf VSO |
5280 | out: |
5281 | tran_finalize(tran, ret); | |
5282 | ||
1e4c797c | 5283 | bdrv_refresh_limits(bs_top, NULL, NULL); |
2272edcf VSO |
5284 | |
5285 | return ret; | |
8802d1fd JC |
5286 | } |
5287 | ||
bd8f4c42 VSO |
5288 | /* Not for empty child */ |
5289 | int bdrv_replace_child_bs(BdrvChild *child, BlockDriverState *new_bs, | |
5290 | Error **errp) | |
5291 | { | |
5292 | int ret; | |
5293 | Transaction *tran = tran_new(); | |
bd8f4c42 VSO |
5294 | g_autoptr(GSList) refresh_list = NULL; |
5295 | BlockDriverState *old_bs = child->bs; | |
5296 | ||
f791bf7f EGE |
5297 | GLOBAL_STATE_CODE(); |
5298 | ||
bd8f4c42 VSO |
5299 | bdrv_ref(old_bs); |
5300 | bdrv_drained_begin(old_bs); | |
5301 | bdrv_drained_begin(new_bs); | |
5302 | ||
0f0b1e29 | 5303 | bdrv_replace_child_tran(child, new_bs, tran); |
bd8f4c42 | 5304 | |
fb0ff4d1 VSO |
5305 | refresh_list = g_slist_prepend(refresh_list, old_bs); |
5306 | refresh_list = g_slist_prepend(refresh_list, new_bs); | |
bd8f4c42 VSO |
5307 | |
5308 | ret = bdrv_list_refresh_perms(refresh_list, NULL, tran, errp); | |
5309 | ||
5310 | tran_finalize(tran, ret); | |
5311 | ||
5312 | bdrv_drained_end(old_bs); | |
5313 | bdrv_drained_end(new_bs); | |
5314 | bdrv_unref(old_bs); | |
5315 | ||
5316 | return ret; | |
5317 | } | |
5318 | ||
4f6fd349 | 5319 | static void bdrv_delete(BlockDriverState *bs) |
b338082b | 5320 | { |
3718d8ab | 5321 | assert(bdrv_op_blocker_is_empty(bs)); |
4f6fd349 | 5322 | assert(!bs->refcnt); |
f791bf7f | 5323 | GLOBAL_STATE_CODE(); |
18846dee | 5324 | |
1b7bdbc1 | 5325 | /* remove from list, if necessary */ |
63eaaae0 KW |
5326 | if (bs->node_name[0] != '\0') { |
5327 | QTAILQ_REMOVE(&graph_bdrv_states, bs, node_list); | |
5328 | } | |
2c1d04e0 HR |
5329 | QTAILQ_REMOVE(&all_bdrv_states, bs, bs_list); |
5330 | ||
30c321f9 AK |
5331 | bdrv_close(bs); |
5332 | ||
7267c094 | 5333 | g_free(bs); |
fc01f7e7 FB |
5334 | } |
5335 | ||
96796fae VSO |
5336 | |
5337 | /* | |
5338 | * Replace @bs by newly created block node. | |
5339 | * | |
5340 | * @options is a QDict of options to pass to the block drivers, or NULL for an | |
5341 | * empty set of options. The reference to the QDict belongs to the block layer | |
5342 | * after the call (even on failure), so if the caller intends to reuse the | |
5343 | * dictionary, it needs to use qobject_ref() before calling bdrv_open. | |
5344 | */ | |
5345 | BlockDriverState *bdrv_insert_node(BlockDriverState *bs, QDict *options, | |
8872ef78 AS |
5346 | int flags, Error **errp) |
5347 | { | |
f053b7e8 VSO |
5348 | ERRP_GUARD(); |
5349 | int ret; | |
b11c8739 VSO |
5350 | BlockDriverState *new_node_bs = NULL; |
5351 | const char *drvname, *node_name; | |
5352 | BlockDriver *drv; | |
5353 | ||
5354 | drvname = qdict_get_try_str(options, "driver"); | |
5355 | if (!drvname) { | |
5356 | error_setg(errp, "driver is not specified"); | |
5357 | goto fail; | |
5358 | } | |
5359 | ||
5360 | drv = bdrv_find_format(drvname); | |
5361 | if (!drv) { | |
5362 | error_setg(errp, "Unknown driver: '%s'", drvname); | |
5363 | goto fail; | |
5364 | } | |
8872ef78 | 5365 | |
b11c8739 VSO |
5366 | node_name = qdict_get_try_str(options, "node-name"); |
5367 | ||
f791bf7f EGE |
5368 | GLOBAL_STATE_CODE(); |
5369 | ||
b11c8739 VSO |
5370 | new_node_bs = bdrv_new_open_driver_opts(drv, node_name, options, flags, |
5371 | errp); | |
5372 | options = NULL; /* bdrv_new_open_driver() eats options */ | |
5373 | if (!new_node_bs) { | |
8872ef78 | 5374 | error_prepend(errp, "Could not create node: "); |
b11c8739 | 5375 | goto fail; |
8872ef78 AS |
5376 | } |
5377 | ||
5378 | bdrv_drained_begin(bs); | |
f053b7e8 | 5379 | ret = bdrv_replace_node(bs, new_node_bs, errp); |
8872ef78 AS |
5380 | bdrv_drained_end(bs); |
5381 | ||
f053b7e8 VSO |
5382 | if (ret < 0) { |
5383 | error_prepend(errp, "Could not replace node: "); | |
b11c8739 | 5384 | goto fail; |
8872ef78 AS |
5385 | } |
5386 | ||
5387 | return new_node_bs; | |
b11c8739 VSO |
5388 | |
5389 | fail: | |
5390 | qobject_unref(options); | |
5391 | bdrv_unref(new_node_bs); | |
5392 | return NULL; | |
8872ef78 AS |
5393 | } |
5394 | ||
e97fc193 AL |
5395 | /* |
5396 | * Run consistency checks on an image | |
5397 | * | |
e076f338 | 5398 | * Returns 0 if the check could be completed (it doesn't mean that the image is |
a1c7273b | 5399 | * free of errors) or -errno when an internal error occurred. The results of the |
e076f338 | 5400 | * check are stored in res. |
e97fc193 | 5401 | */ |
21c2283e VSO |
5402 | int coroutine_fn bdrv_co_check(BlockDriverState *bs, |
5403 | BdrvCheckResult *res, BdrvCheckMode fix) | |
e97fc193 | 5404 | { |
1581a70d | 5405 | IO_CODE(); |
1b3ff9fe | 5406 | assert_bdrv_graph_readable(); |
908bcd54 HR |
5407 | if (bs->drv == NULL) { |
5408 | return -ENOMEDIUM; | |
5409 | } | |
2fd61638 | 5410 | if (bs->drv->bdrv_co_check == NULL) { |
e97fc193 AL |
5411 | return -ENOTSUP; |
5412 | } | |
5413 | ||
e076f338 | 5414 | memset(res, 0, sizeof(*res)); |
2fd61638 PB |
5415 | return bs->drv->bdrv_co_check(bs, res, fix); |
5416 | } | |
5417 | ||
756e6736 KW |
5418 | /* |
5419 | * Return values: | |
5420 | * 0 - success | |
5421 | * -EINVAL - backing format specified, but no file | |
5422 | * -ENOSPC - can't update the backing file because no space is left in the | |
5423 | * image file header | |
5424 | * -ENOTSUP - format driver doesn't support changing the backing file | |
5425 | */ | |
e54ee1b3 | 5426 | int bdrv_change_backing_file(BlockDriverState *bs, const char *backing_file, |
497a30db | 5427 | const char *backing_fmt, bool require) |
756e6736 KW |
5428 | { |
5429 | BlockDriver *drv = bs->drv; | |
469ef350 | 5430 | int ret; |
756e6736 | 5431 | |
f791bf7f EGE |
5432 | GLOBAL_STATE_CODE(); |
5433 | ||
d470ad42 HR |
5434 | if (!drv) { |
5435 | return -ENOMEDIUM; | |
5436 | } | |
5437 | ||
5f377794 PB |
5438 | /* Backing file format doesn't make sense without a backing file */ |
5439 | if (backing_fmt && !backing_file) { | |
5440 | return -EINVAL; | |
5441 | } | |
5442 | ||
497a30db EB |
5443 | if (require && backing_file && !backing_fmt) { |
5444 | return -EINVAL; | |
e54ee1b3 EB |
5445 | } |
5446 | ||
756e6736 | 5447 | if (drv->bdrv_change_backing_file != NULL) { |
469ef350 | 5448 | ret = drv->bdrv_change_backing_file(bs, backing_file, backing_fmt); |
756e6736 | 5449 | } else { |
469ef350 | 5450 | ret = -ENOTSUP; |
756e6736 | 5451 | } |
469ef350 PB |
5452 | |
5453 | if (ret == 0) { | |
5454 | pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_file ?: ""); | |
5455 | pstrcpy(bs->backing_format, sizeof(bs->backing_format), backing_fmt ?: ""); | |
998c2019 HR |
5456 | pstrcpy(bs->auto_backing_file, sizeof(bs->auto_backing_file), |
5457 | backing_file ?: ""); | |
469ef350 PB |
5458 | } |
5459 | return ret; | |
756e6736 KW |
5460 | } |
5461 | ||
6ebdcee2 | 5462 | /* |
dcf3f9b2 HR |
5463 | * Finds the first non-filter node above bs in the chain between |
5464 | * active and bs. The returned node is either an immediate parent of | |
5465 | * bs, or there are only filter nodes between the two. | |
6ebdcee2 JC |
5466 | * |
5467 | * Returns NULL if bs is not found in active's image chain, | |
5468 | * or if active == bs. | |
4caf0fcd JC |
5469 | * |
5470 | * Returns the bottommost base image if bs == NULL. | |
6ebdcee2 JC |
5471 | */ |
5472 | BlockDriverState *bdrv_find_overlay(BlockDriverState *active, | |
5473 | BlockDriverState *bs) | |
5474 | { | |
f791bf7f EGE |
5475 | |
5476 | GLOBAL_STATE_CODE(); | |
5477 | ||
dcf3f9b2 HR |
5478 | bs = bdrv_skip_filters(bs); |
5479 | active = bdrv_skip_filters(active); | |
5480 | ||
5481 | while (active) { | |
5482 | BlockDriverState *next = bdrv_backing_chain_next(active); | |
5483 | if (bs == next) { | |
5484 | return active; | |
5485 | } | |
5486 | active = next; | |
6ebdcee2 JC |
5487 | } |
5488 | ||
dcf3f9b2 | 5489 | return NULL; |
4caf0fcd | 5490 | } |
6ebdcee2 | 5491 | |
4caf0fcd JC |
5492 | /* Given a BDS, searches for the base layer. */ |
5493 | BlockDriverState *bdrv_find_base(BlockDriverState *bs) | |
5494 | { | |
f791bf7f EGE |
5495 | GLOBAL_STATE_CODE(); |
5496 | ||
4caf0fcd | 5497 | return bdrv_find_overlay(bs, NULL); |
6ebdcee2 JC |
5498 | } |
5499 | ||
2cad1ebe | 5500 | /* |
7b99a266 HR |
5501 | * Return true if at least one of the COW (backing) and filter links |
5502 | * between @bs and @base is frozen. @errp is set if that's the case. | |
0f0998f6 | 5503 | * @base must be reachable from @bs, or NULL. |
2cad1ebe AG |
5504 | */ |
5505 | bool bdrv_is_backing_chain_frozen(BlockDriverState *bs, BlockDriverState *base, | |
5506 | Error **errp) | |
5507 | { | |
5508 | BlockDriverState *i; | |
7b99a266 | 5509 | BdrvChild *child; |
2cad1ebe | 5510 | |
f791bf7f EGE |
5511 | GLOBAL_STATE_CODE(); |
5512 | ||
7b99a266 HR |
5513 | for (i = bs; i != base; i = child_bs(child)) { |
5514 | child = bdrv_filter_or_cow_child(i); | |
5515 | ||
5516 | if (child && child->frozen) { | |
2cad1ebe | 5517 | error_setg(errp, "Cannot change '%s' link from '%s' to '%s'", |
7b99a266 | 5518 | child->name, i->node_name, child->bs->node_name); |
2cad1ebe AG |
5519 | return true; |
5520 | } | |
5521 | } | |
5522 | ||
5523 | return false; | |
5524 | } | |
5525 | ||
5526 | /* | |
7b99a266 | 5527 | * Freeze all COW (backing) and filter links between @bs and @base. |
2cad1ebe AG |
5528 | * If any of the links is already frozen the operation is aborted and |
5529 | * none of the links are modified. | |
0f0998f6 | 5530 | * @base must be reachable from @bs, or NULL. |
2cad1ebe AG |
5531 | * Returns 0 on success. On failure returns < 0 and sets @errp. |
5532 | */ | |
5533 | int bdrv_freeze_backing_chain(BlockDriverState *bs, BlockDriverState *base, | |
5534 | Error **errp) | |
5535 | { | |
5536 | BlockDriverState *i; | |
7b99a266 | 5537 | BdrvChild *child; |
2cad1ebe | 5538 | |
f791bf7f EGE |
5539 | GLOBAL_STATE_CODE(); |
5540 | ||
2cad1ebe AG |
5541 | if (bdrv_is_backing_chain_frozen(bs, base, errp)) { |
5542 | return -EPERM; | |
5543 | } | |
5544 | ||
7b99a266 HR |
5545 | for (i = bs; i != base; i = child_bs(child)) { |
5546 | child = bdrv_filter_or_cow_child(i); | |
5547 | if (child && child->bs->never_freeze) { | |
e5182c1c | 5548 | error_setg(errp, "Cannot freeze '%s' link to '%s'", |
7b99a266 | 5549 | child->name, child->bs->node_name); |
e5182c1c HR |
5550 | return -EPERM; |
5551 | } | |
5552 | } | |
5553 | ||
7b99a266 HR |
5554 | for (i = bs; i != base; i = child_bs(child)) { |
5555 | child = bdrv_filter_or_cow_child(i); | |
5556 | if (child) { | |
5557 | child->frozen = true; | |
0f0998f6 | 5558 | } |
2cad1ebe AG |
5559 | } |
5560 | ||
5561 | return 0; | |
5562 | } | |
5563 | ||
5564 | /* | |
7b99a266 HR |
5565 | * Unfreeze all COW (backing) and filter links between @bs and @base. |
5566 | * The caller must ensure that all links are frozen before using this | |
5567 | * function. | |
0f0998f6 | 5568 | * @base must be reachable from @bs, or NULL. |
2cad1ebe AG |
5569 | */ |
5570 | void bdrv_unfreeze_backing_chain(BlockDriverState *bs, BlockDriverState *base) | |
5571 | { | |
5572 | BlockDriverState *i; | |
7b99a266 | 5573 | BdrvChild *child; |
2cad1ebe | 5574 | |
f791bf7f EGE |
5575 | GLOBAL_STATE_CODE(); |
5576 | ||
7b99a266 HR |
5577 | for (i = bs; i != base; i = child_bs(child)) { |
5578 | child = bdrv_filter_or_cow_child(i); | |
5579 | if (child) { | |
5580 | assert(child->frozen); | |
5581 | child->frozen = false; | |
0f0998f6 | 5582 | } |
2cad1ebe AG |
5583 | } |
5584 | } | |
5585 | ||
6ebdcee2 JC |
5586 | /* |
5587 | * Drops images above 'base' up to and including 'top', and sets the image | |
5588 | * above 'top' to have base as its backing file. | |
5589 | * | |
5590 | * Requires that the overlay to 'top' is opened r/w, so that the backing file | |
5591 | * information in 'bs' can be properly updated. | |
5592 | * | |
5593 | * E.g., this will convert the following chain: | |
5594 | * bottom <- base <- intermediate <- top <- active | |
5595 | * | |
5596 | * to | |
5597 | * | |
5598 | * bottom <- base <- active | |
5599 | * | |
5600 | * It is allowed for bottom==base, in which case it converts: | |
5601 | * | |
5602 | * base <- intermediate <- top <- active | |
5603 | * | |
5604 | * to | |
5605 | * | |
5606 | * base <- active | |
5607 | * | |
54e26900 JC |
5608 | * If backing_file_str is non-NULL, it will be used when modifying top's |
5609 | * overlay image metadata. | |
5610 | * | |
6ebdcee2 JC |
5611 | * Error conditions: |
5612 | * if active == top, that is considered an error | |
5613 | * | |
5614 | */ | |
bde70715 KW |
5615 | int bdrv_drop_intermediate(BlockDriverState *top, BlockDriverState *base, |
5616 | const char *backing_file_str) | |
6ebdcee2 | 5617 | { |
6bd858b3 AG |
5618 | BlockDriverState *explicit_top = top; |
5619 | bool update_inherits_from; | |
d669ed6a | 5620 | BdrvChild *c; |
12fa4af6 | 5621 | Error *local_err = NULL; |
6ebdcee2 | 5622 | int ret = -EIO; |
d669ed6a VSO |
5623 | g_autoptr(GSList) updated_children = NULL; |
5624 | GSList *p; | |
6ebdcee2 | 5625 | |
f791bf7f EGE |
5626 | GLOBAL_STATE_CODE(); |
5627 | ||
6858eba0 | 5628 | bdrv_ref(top); |
631086de | 5629 | bdrv_drained_begin(base); |
6858eba0 | 5630 | |
6ebdcee2 JC |
5631 | if (!top->drv || !base->drv) { |
5632 | goto exit; | |
5633 | } | |
5634 | ||
5db15a57 KW |
5635 | /* Make sure that base is in the backing chain of top */ |
5636 | if (!bdrv_chain_contains(top, base)) { | |
6ebdcee2 JC |
5637 | goto exit; |
5638 | } | |
5639 | ||
6bd858b3 AG |
5640 | /* If 'base' recursively inherits from 'top' then we should set |
5641 | * base->inherits_from to top->inherits_from after 'top' and all | |
5642 | * other intermediate nodes have been dropped. | |
5643 | * If 'top' is an implicit node (e.g. "commit_top") we should skip | |
5644 | * it because no one inherits from it. We use explicit_top for that. */ | |
dcf3f9b2 | 5645 | explicit_top = bdrv_skip_implicit_filters(explicit_top); |
6bd858b3 AG |
5646 | update_inherits_from = bdrv_inherits_from_recursive(base, explicit_top); |
5647 | ||
6ebdcee2 | 5648 | /* success - we can delete the intermediate states, and link top->base */ |
f30c66ba HR |
5649 | if (!backing_file_str) { |
5650 | bdrv_refresh_filename(base); | |
5651 | backing_file_str = base->filename; | |
5652 | } | |
61f09cea | 5653 | |
d669ed6a VSO |
5654 | QLIST_FOREACH(c, &top->parents, next_parent) { |
5655 | updated_children = g_slist_prepend(updated_children, c); | |
5656 | } | |
5657 | ||
3108a15c VSO |
5658 | /* |
5659 | * It seems correct to pass detach_subchain=true here, but it triggers | |
5660 | * one more yet not fixed bug, when due to nested aio_poll loop we switch to | |
5661 | * another drained section, which modify the graph (for example, removing | |
5662 | * the child, which we keep in updated_children list). So, it's a TODO. | |
5663 | * | |
5664 | * Note, bug triggered if pass detach_subchain=true here and run | |
5665 | * test-bdrv-drain. test_drop_intermediate_poll() test-case will crash. | |
5666 | * That's a FIXME. | |
5667 | */ | |
5668 | bdrv_replace_node_common(top, base, false, false, &local_err); | |
d669ed6a VSO |
5669 | if (local_err) { |
5670 | error_report_err(local_err); | |
5671 | goto exit; | |
5672 | } | |
5673 | ||
5674 | for (p = updated_children; p; p = p->next) { | |
5675 | c = p->data; | |
12fa4af6 | 5676 | |
bd86fb99 HR |
5677 | if (c->klass->update_filename) { |
5678 | ret = c->klass->update_filename(c, base, backing_file_str, | |
5679 | &local_err); | |
61f09cea | 5680 | if (ret < 0) { |
d669ed6a VSO |
5681 | /* |
5682 | * TODO: Actually, we want to rollback all previous iterations | |
5683 | * of this loop, and (which is almost impossible) previous | |
5684 | * bdrv_replace_node()... | |
5685 | * | |
5686 | * Note, that c->klass->update_filename may lead to permission | |
5687 | * update, so it's a bad idea to call it inside permission | |
5688 | * update transaction of bdrv_replace_node. | |
5689 | */ | |
61f09cea KW |
5690 | error_report_err(local_err); |
5691 | goto exit; | |
5692 | } | |
5693 | } | |
12fa4af6 | 5694 | } |
6ebdcee2 | 5695 | |
6bd858b3 AG |
5696 | if (update_inherits_from) { |
5697 | base->inherits_from = explicit_top->inherits_from; | |
5698 | } | |
5699 | ||
6ebdcee2 | 5700 | ret = 0; |
6ebdcee2 | 5701 | exit: |
631086de | 5702 | bdrv_drained_end(base); |
6858eba0 | 5703 | bdrv_unref(top); |
6ebdcee2 JC |
5704 | return ret; |
5705 | } | |
5706 | ||
081e4650 HR |
5707 | /** |
5708 | * Implementation of BlockDriver.bdrv_get_allocated_file_size() that | |
5709 | * sums the size of all data-bearing children. (This excludes backing | |
5710 | * children.) | |
5711 | */ | |
5712 | static int64_t bdrv_sum_allocated_file_size(BlockDriverState *bs) | |
5713 | { | |
5714 | BdrvChild *child; | |
5715 | int64_t child_size, sum = 0; | |
5716 | ||
5717 | QLIST_FOREACH(child, &bs->children, next) { | |
5718 | if (child->role & (BDRV_CHILD_DATA | BDRV_CHILD_METADATA | | |
5719 | BDRV_CHILD_FILTERED)) | |
5720 | { | |
5721 | child_size = bdrv_get_allocated_file_size(child->bs); | |
5722 | if (child_size < 0) { | |
5723 | return child_size; | |
5724 | } | |
5725 | sum += child_size; | |
5726 | } | |
5727 | } | |
5728 | ||
5729 | return sum; | |
5730 | } | |
5731 | ||
61007b31 SH |
5732 | /** |
5733 | * Length of a allocated file in bytes. Sparse files are counted by actual | |
5734 | * allocated space. Return < 0 if error or unknown. | |
5735 | */ | |
5736 | int64_t bdrv_get_allocated_file_size(BlockDriverState *bs) | |
71d0770c | 5737 | { |
61007b31 | 5738 | BlockDriver *drv = bs->drv; |
384a48fb EGE |
5739 | IO_CODE(); |
5740 | ||
61007b31 SH |
5741 | if (!drv) { |
5742 | return -ENOMEDIUM; | |
8f4754ed | 5743 | } |
61007b31 SH |
5744 | if (drv->bdrv_get_allocated_file_size) { |
5745 | return drv->bdrv_get_allocated_file_size(bs); | |
5746 | } | |
081e4650 HR |
5747 | |
5748 | if (drv->bdrv_file_open) { | |
5749 | /* | |
5750 | * Protocol drivers default to -ENOTSUP (most of their data is | |
5751 | * not stored in any of their children (if they even have any), | |
5752 | * so there is no generic way to figure it out). | |
5753 | */ | |
5754 | return -ENOTSUP; | |
5755 | } else if (drv->is_filter) { | |
5756 | /* Filter drivers default to the size of their filtered child */ | |
5757 | return bdrv_get_allocated_file_size(bdrv_filter_bs(bs)); | |
5758 | } else { | |
5759 | /* Other drivers default to summing their children's sizes */ | |
5760 | return bdrv_sum_allocated_file_size(bs); | |
1c9805a3 SH |
5761 | } |
5762 | } | |
e7a8a783 | 5763 | |
90880ff1 SH |
5764 | /* |
5765 | * bdrv_measure: | |
5766 | * @drv: Format driver | |
5767 | * @opts: Creation options for new image | |
5768 | * @in_bs: Existing image containing data for new image (may be NULL) | |
5769 | * @errp: Error object | |
5770 | * Returns: A #BlockMeasureInfo (free using qapi_free_BlockMeasureInfo()) | |
5771 | * or NULL on error | |
5772 | * | |
5773 | * Calculate file size required to create a new image. | |
5774 | * | |
5775 | * If @in_bs is given then space for allocated clusters and zero clusters | |
5776 | * from that image are included in the calculation. If @opts contains a | |
5777 | * backing file that is shared by @in_bs then backing clusters may be omitted | |
5778 | * from the calculation. | |
5779 | * | |
5780 | * If @in_bs is NULL then the calculation includes no allocated clusters | |
5781 | * unless a preallocation option is given in @opts. | |
5782 | * | |
5783 | * Note that @in_bs may use a different BlockDriver from @drv. | |
5784 | * | |
5785 | * If an error occurs the @errp pointer is set. | |
5786 | */ | |
5787 | BlockMeasureInfo *bdrv_measure(BlockDriver *drv, QemuOpts *opts, | |
5788 | BlockDriverState *in_bs, Error **errp) | |
5789 | { | |
384a48fb | 5790 | IO_CODE(); |
90880ff1 SH |
5791 | if (!drv->bdrv_measure) { |
5792 | error_setg(errp, "Block driver '%s' does not support size measurement", | |
5793 | drv->format_name); | |
5794 | return NULL; | |
5795 | } | |
5796 | ||
5797 | return drv->bdrv_measure(opts, in_bs, errp); | |
5798 | } | |
5799 | ||
61007b31 SH |
5800 | /** |
5801 | * Return number of sectors on success, -errno on error. | |
1c9805a3 | 5802 | */ |
61007b31 | 5803 | int64_t bdrv_nb_sectors(BlockDriverState *bs) |
1c9805a3 | 5804 | { |
61007b31 | 5805 | BlockDriver *drv = bs->drv; |
384a48fb | 5806 | IO_CODE(); |
498e386c | 5807 | |
61007b31 SH |
5808 | if (!drv) |
5809 | return -ENOMEDIUM; | |
2572b37a | 5810 | |
61007b31 SH |
5811 | if (drv->has_variable_length) { |
5812 | int ret = refresh_total_sectors(bs, bs->total_sectors); | |
5813 | if (ret < 0) { | |
5814 | return ret; | |
1c9805a3 SH |
5815 | } |
5816 | } | |
61007b31 | 5817 | return bs->total_sectors; |
1c9805a3 | 5818 | } |
b338082b | 5819 | |
61007b31 SH |
5820 | /** |
5821 | * Return length in bytes on success, -errno on error. | |
5822 | * The length is always a multiple of BDRV_SECTOR_SIZE. | |
8d3b1a2d | 5823 | */ |
61007b31 | 5824 | int64_t bdrv_getlength(BlockDriverState *bs) |
8d3b1a2d | 5825 | { |
61007b31 | 5826 | int64_t ret = bdrv_nb_sectors(bs); |
384a48fb | 5827 | IO_CODE(); |
8d3b1a2d | 5828 | |
122860ba EB |
5829 | if (ret < 0) { |
5830 | return ret; | |
5831 | } | |
5832 | if (ret > INT64_MAX / BDRV_SECTOR_SIZE) { | |
5833 | return -EFBIG; | |
5834 | } | |
5835 | return ret * BDRV_SECTOR_SIZE; | |
fc01f7e7 FB |
5836 | } |
5837 | ||
61007b31 SH |
5838 | /* return 0 as number of sectors if no device present or error */ |
5839 | void bdrv_get_geometry(BlockDriverState *bs, uint64_t *nb_sectors_ptr) | |
07d27a44 | 5840 | { |
61007b31 | 5841 | int64_t nb_sectors = bdrv_nb_sectors(bs); |
384a48fb | 5842 | IO_CODE(); |
07d27a44 | 5843 | |
61007b31 | 5844 | *nb_sectors_ptr = nb_sectors < 0 ? 0 : nb_sectors; |
07d27a44 MA |
5845 | } |
5846 | ||
54115412 | 5847 | bool bdrv_is_sg(BlockDriverState *bs) |
f08145fe | 5848 | { |
384a48fb | 5849 | IO_CODE(); |
61007b31 | 5850 | return bs->sg; |
f08145fe KW |
5851 | } |
5852 | ||
ae23f786 HR |
5853 | /** |
5854 | * Return whether the given node supports compressed writes. | |
5855 | */ | |
5856 | bool bdrv_supports_compressed_writes(BlockDriverState *bs) | |
5857 | { | |
5858 | BlockDriverState *filtered; | |
384a48fb | 5859 | IO_CODE(); |
ae23f786 HR |
5860 | |
5861 | if (!bs->drv || !block_driver_can_compress(bs->drv)) { | |
5862 | return false; | |
5863 | } | |
5864 | ||
5865 | filtered = bdrv_filter_bs(bs); | |
5866 | if (filtered) { | |
5867 | /* | |
5868 | * Filters can only forward compressed writes, so we have to | |
5869 | * check the child. | |
5870 | */ | |
5871 | return bdrv_supports_compressed_writes(filtered); | |
5872 | } | |
5873 | ||
5874 | return true; | |
5875 | } | |
5876 | ||
61007b31 | 5877 | const char *bdrv_get_format_name(BlockDriverState *bs) |
40b4f539 | 5878 | { |
384a48fb | 5879 | IO_CODE(); |
61007b31 | 5880 | return bs->drv ? bs->drv->format_name : NULL; |
40b4f539 KW |
5881 | } |
5882 | ||
61007b31 | 5883 | static int qsort_strcmp(const void *a, const void *b) |
40b4f539 | 5884 | { |
ceff5bd7 | 5885 | return strcmp(*(char *const *)a, *(char *const *)b); |
40b4f539 KW |
5886 | } |
5887 | ||
61007b31 | 5888 | void bdrv_iterate_format(void (*it)(void *opaque, const char *name), |
9ac404c5 | 5889 | void *opaque, bool read_only) |
40b4f539 | 5890 | { |
61007b31 SH |
5891 | BlockDriver *drv; |
5892 | int count = 0; | |
5893 | int i; | |
5894 | const char **formats = NULL; | |
40b4f539 | 5895 | |
f791bf7f EGE |
5896 | GLOBAL_STATE_CODE(); |
5897 | ||
61007b31 SH |
5898 | QLIST_FOREACH(drv, &bdrv_drivers, list) { |
5899 | if (drv->format_name) { | |
5900 | bool found = false; | |
5901 | int i = count; | |
9ac404c5 AS |
5902 | |
5903 | if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv, read_only)) { | |
5904 | continue; | |
5905 | } | |
5906 | ||
61007b31 SH |
5907 | while (formats && i && !found) { |
5908 | found = !strcmp(formats[--i], drv->format_name); | |
5909 | } | |
e2a305fb | 5910 | |
61007b31 SH |
5911 | if (!found) { |
5912 | formats = g_renew(const char *, formats, count + 1); | |
5913 | formats[count++] = drv->format_name; | |
5914 | } | |
6c5a42ac | 5915 | } |
61007b31 | 5916 | } |
6c5a42ac | 5917 | |
eb0df69f HR |
5918 | for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); i++) { |
5919 | const char *format_name = block_driver_modules[i].format_name; | |
5920 | ||
5921 | if (format_name) { | |
5922 | bool found = false; | |
5923 | int j = count; | |
5924 | ||
9ac404c5 AS |
5925 | if (use_bdrv_whitelist && |
5926 | !bdrv_format_is_whitelisted(format_name, read_only)) { | |
5927 | continue; | |
5928 | } | |
5929 | ||
eb0df69f HR |
5930 | while (formats && j && !found) { |
5931 | found = !strcmp(formats[--j], format_name); | |
5932 | } | |
5933 | ||
5934 | if (!found) { | |
5935 | formats = g_renew(const char *, formats, count + 1); | |
5936 | formats[count++] = format_name; | |
5937 | } | |
5938 | } | |
5939 | } | |
5940 | ||
61007b31 | 5941 | qsort(formats, count, sizeof(formats[0]), qsort_strcmp); |
40b4f539 | 5942 | |
61007b31 SH |
5943 | for (i = 0; i < count; i++) { |
5944 | it(opaque, formats[i]); | |
5945 | } | |
40b4f539 | 5946 | |
61007b31 SH |
5947 | g_free(formats); |
5948 | } | |
40b4f539 | 5949 | |
61007b31 SH |
5950 | /* This function is to find a node in the bs graph */ |
5951 | BlockDriverState *bdrv_find_node(const char *node_name) | |
5952 | { | |
5953 | BlockDriverState *bs; | |
391827eb | 5954 | |
61007b31 | 5955 | assert(node_name); |
f791bf7f | 5956 | GLOBAL_STATE_CODE(); |
40b4f539 | 5957 | |
61007b31 SH |
5958 | QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) { |
5959 | if (!strcmp(node_name, bs->node_name)) { | |
5960 | return bs; | |
40b4f539 KW |
5961 | } |
5962 | } | |
61007b31 | 5963 | return NULL; |
40b4f539 KW |
5964 | } |
5965 | ||
61007b31 | 5966 | /* Put this QMP function here so it can access the static graph_bdrv_states. */ |
facda544 PK |
5967 | BlockDeviceInfoList *bdrv_named_nodes_list(bool flat, |
5968 | Error **errp) | |
40b4f539 | 5969 | { |
9812e712 | 5970 | BlockDeviceInfoList *list; |
61007b31 | 5971 | BlockDriverState *bs; |
40b4f539 | 5972 | |
f791bf7f EGE |
5973 | GLOBAL_STATE_CODE(); |
5974 | ||
61007b31 SH |
5975 | list = NULL; |
5976 | QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) { | |
facda544 | 5977 | BlockDeviceInfo *info = bdrv_block_device_info(NULL, bs, flat, errp); |
61007b31 SH |
5978 | if (!info) { |
5979 | qapi_free_BlockDeviceInfoList(list); | |
5980 | return NULL; | |
301db7c2 | 5981 | } |
9812e712 | 5982 | QAPI_LIST_PREPEND(list, info); |
301db7c2 RH |
5983 | } |
5984 | ||
61007b31 SH |
5985 | return list; |
5986 | } | |
40b4f539 | 5987 | |
5d3b4e99 VSO |
5988 | typedef struct XDbgBlockGraphConstructor { |
5989 | XDbgBlockGraph *graph; | |
5990 | GHashTable *graph_nodes; | |
5991 | } XDbgBlockGraphConstructor; | |
5992 | ||
5993 | static XDbgBlockGraphConstructor *xdbg_graph_new(void) | |
5994 | { | |
5995 | XDbgBlockGraphConstructor *gr = g_new(XDbgBlockGraphConstructor, 1); | |
5996 | ||
5997 | gr->graph = g_new0(XDbgBlockGraph, 1); | |
5998 | gr->graph_nodes = g_hash_table_new(NULL, NULL); | |
5999 | ||
6000 | return gr; | |
6001 | } | |
6002 | ||
6003 | static XDbgBlockGraph *xdbg_graph_finalize(XDbgBlockGraphConstructor *gr) | |
6004 | { | |
6005 | XDbgBlockGraph *graph = gr->graph; | |
6006 | ||
6007 | g_hash_table_destroy(gr->graph_nodes); | |
6008 | g_free(gr); | |
6009 | ||
6010 | return graph; | |
6011 | } | |
6012 | ||
6013 | static uintptr_t xdbg_graph_node_num(XDbgBlockGraphConstructor *gr, void *node) | |
6014 | { | |
6015 | uintptr_t ret = (uintptr_t)g_hash_table_lookup(gr->graph_nodes, node); | |
6016 | ||
6017 | if (ret != 0) { | |
6018 | return ret; | |
6019 | } | |
6020 | ||
6021 | /* | |
6022 | * Start counting from 1, not 0, because 0 interferes with not-found (NULL) | |
6023 | * answer of g_hash_table_lookup. | |
6024 | */ | |
6025 | ret = g_hash_table_size(gr->graph_nodes) + 1; | |
6026 | g_hash_table_insert(gr->graph_nodes, node, (void *)ret); | |
6027 | ||
6028 | return ret; | |
6029 | } | |
6030 | ||
6031 | static void xdbg_graph_add_node(XDbgBlockGraphConstructor *gr, void *node, | |
6032 | XDbgBlockGraphNodeType type, const char *name) | |
6033 | { | |
6034 | XDbgBlockGraphNode *n; | |
6035 | ||
6036 | n = g_new0(XDbgBlockGraphNode, 1); | |
6037 | ||
6038 | n->id = xdbg_graph_node_num(gr, node); | |
6039 | n->type = type; | |
6040 | n->name = g_strdup(name); | |
6041 | ||
9812e712 | 6042 | QAPI_LIST_PREPEND(gr->graph->nodes, n); |
5d3b4e99 VSO |
6043 | } |
6044 | ||
6045 | static void xdbg_graph_add_edge(XDbgBlockGraphConstructor *gr, void *parent, | |
6046 | const BdrvChild *child) | |
6047 | { | |
cdb1cec8 | 6048 | BlockPermission qapi_perm; |
5d3b4e99 | 6049 | XDbgBlockGraphEdge *edge; |
862fded9 | 6050 | GLOBAL_STATE_CODE(); |
5d3b4e99 | 6051 | |
5d3b4e99 VSO |
6052 | edge = g_new0(XDbgBlockGraphEdge, 1); |
6053 | ||
6054 | edge->parent = xdbg_graph_node_num(gr, parent); | |
6055 | edge->child = xdbg_graph_node_num(gr, child->bs); | |
6056 | edge->name = g_strdup(child->name); | |
6057 | ||
cdb1cec8 HR |
6058 | for (qapi_perm = 0; qapi_perm < BLOCK_PERMISSION__MAX; qapi_perm++) { |
6059 | uint64_t flag = bdrv_qapi_perm_to_blk_perm(qapi_perm); | |
6060 | ||
6061 | if (flag & child->perm) { | |
9812e712 | 6062 | QAPI_LIST_PREPEND(edge->perm, qapi_perm); |
5d3b4e99 | 6063 | } |
cdb1cec8 | 6064 | if (flag & child->shared_perm) { |
9812e712 | 6065 | QAPI_LIST_PREPEND(edge->shared_perm, qapi_perm); |
5d3b4e99 VSO |
6066 | } |
6067 | } | |
6068 | ||
9812e712 | 6069 | QAPI_LIST_PREPEND(gr->graph->edges, edge); |
5d3b4e99 VSO |
6070 | } |
6071 | ||
6072 | ||
6073 | XDbgBlockGraph *bdrv_get_xdbg_block_graph(Error **errp) | |
6074 | { | |
6075 | BlockBackend *blk; | |
6076 | BlockJob *job; | |
6077 | BlockDriverState *bs; | |
6078 | BdrvChild *child; | |
6079 | XDbgBlockGraphConstructor *gr = xdbg_graph_new(); | |
6080 | ||
f791bf7f EGE |
6081 | GLOBAL_STATE_CODE(); |
6082 | ||
5d3b4e99 VSO |
6083 | for (blk = blk_all_next(NULL); blk; blk = blk_all_next(blk)) { |
6084 | char *allocated_name = NULL; | |
6085 | const char *name = blk_name(blk); | |
6086 | ||
6087 | if (!*name) { | |
6088 | name = allocated_name = blk_get_attached_dev_id(blk); | |
6089 | } | |
6090 | xdbg_graph_add_node(gr, blk, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_BACKEND, | |
6091 | name); | |
6092 | g_free(allocated_name); | |
6093 | if (blk_root(blk)) { | |
6094 | xdbg_graph_add_edge(gr, blk, blk_root(blk)); | |
6095 | } | |
6096 | } | |
6097 | ||
880eeec6 EGE |
6098 | WITH_JOB_LOCK_GUARD() { |
6099 | for (job = block_job_next_locked(NULL); job; | |
6100 | job = block_job_next_locked(job)) { | |
6101 | GSList *el; | |
5d3b4e99 | 6102 | |
880eeec6 EGE |
6103 | xdbg_graph_add_node(gr, job, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_JOB, |
6104 | job->job.id); | |
6105 | for (el = job->nodes; el; el = el->next) { | |
6106 | xdbg_graph_add_edge(gr, job, (BdrvChild *)el->data); | |
6107 | } | |
5d3b4e99 VSO |
6108 | } |
6109 | } | |
6110 | ||
6111 | QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) { | |
6112 | xdbg_graph_add_node(gr, bs, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_DRIVER, | |
6113 | bs->node_name); | |
6114 | QLIST_FOREACH(child, &bs->children, next) { | |
6115 | xdbg_graph_add_edge(gr, bs, child); | |
6116 | } | |
6117 | } | |
6118 | ||
6119 | return xdbg_graph_finalize(gr); | |
6120 | } | |
6121 | ||
61007b31 SH |
6122 | BlockDriverState *bdrv_lookup_bs(const char *device, |
6123 | const char *node_name, | |
6124 | Error **errp) | |
6125 | { | |
6126 | BlockBackend *blk; | |
6127 | BlockDriverState *bs; | |
40b4f539 | 6128 | |
f791bf7f EGE |
6129 | GLOBAL_STATE_CODE(); |
6130 | ||
61007b31 SH |
6131 | if (device) { |
6132 | blk = blk_by_name(device); | |
40b4f539 | 6133 | |
61007b31 | 6134 | if (blk) { |
9f4ed6fb AG |
6135 | bs = blk_bs(blk); |
6136 | if (!bs) { | |
5433c24f | 6137 | error_setg(errp, "Device '%s' has no medium", device); |
5433c24f HR |
6138 | } |
6139 | ||
9f4ed6fb | 6140 | return bs; |
61007b31 SH |
6141 | } |
6142 | } | |
40b4f539 | 6143 | |
61007b31 SH |
6144 | if (node_name) { |
6145 | bs = bdrv_find_node(node_name); | |
6d519a5f | 6146 | |
61007b31 SH |
6147 | if (bs) { |
6148 | return bs; | |
6149 | } | |
40b4f539 KW |
6150 | } |
6151 | ||
785ec4b1 | 6152 | error_setg(errp, "Cannot find device=\'%s\' nor node-name=\'%s\'", |
61007b31 SH |
6153 | device ? device : "", |
6154 | node_name ? node_name : ""); | |
6155 | return NULL; | |
40b4f539 KW |
6156 | } |
6157 | ||
61007b31 SH |
6158 | /* If 'base' is in the same chain as 'top', return true. Otherwise, |
6159 | * return false. If either argument is NULL, return false. */ | |
6160 | bool bdrv_chain_contains(BlockDriverState *top, BlockDriverState *base) | |
83f64091 | 6161 | { |
f791bf7f EGE |
6162 | |
6163 | GLOBAL_STATE_CODE(); | |
6164 | ||
61007b31 | 6165 | while (top && top != base) { |
dcf3f9b2 | 6166 | top = bdrv_filter_or_cow_bs(top); |
02c50efe | 6167 | } |
61007b31 SH |
6168 | |
6169 | return top != NULL; | |
02c50efe FZ |
6170 | } |
6171 | ||
61007b31 | 6172 | BlockDriverState *bdrv_next_node(BlockDriverState *bs) |
02c50efe | 6173 | { |
f791bf7f | 6174 | GLOBAL_STATE_CODE(); |
61007b31 SH |
6175 | if (!bs) { |
6176 | return QTAILQ_FIRST(&graph_bdrv_states); | |
02c50efe | 6177 | } |
61007b31 | 6178 | return QTAILQ_NEXT(bs, node_list); |
83f64091 FB |
6179 | } |
6180 | ||
0f12264e KW |
6181 | BlockDriverState *bdrv_next_all_states(BlockDriverState *bs) |
6182 | { | |
f791bf7f | 6183 | GLOBAL_STATE_CODE(); |
0f12264e KW |
6184 | if (!bs) { |
6185 | return QTAILQ_FIRST(&all_bdrv_states); | |
6186 | } | |
6187 | return QTAILQ_NEXT(bs, bs_list); | |
6188 | } | |
6189 | ||
61007b31 | 6190 | const char *bdrv_get_node_name(const BlockDriverState *bs) |
83f64091 | 6191 | { |
384a48fb | 6192 | IO_CODE(); |
61007b31 | 6193 | return bs->node_name; |
beac80cd FB |
6194 | } |
6195 | ||
1f0c461b | 6196 | const char *bdrv_get_parent_name(const BlockDriverState *bs) |
4c265bf9 KW |
6197 | { |
6198 | BdrvChild *c; | |
6199 | const char *name; | |
967d7905 | 6200 | IO_CODE(); |
4c265bf9 KW |
6201 | |
6202 | /* If multiple parents have a name, just pick the first one. */ | |
6203 | QLIST_FOREACH(c, &bs->parents, next_parent) { | |
bd86fb99 HR |
6204 | if (c->klass->get_name) { |
6205 | name = c->klass->get_name(c); | |
4c265bf9 KW |
6206 | if (name && *name) { |
6207 | return name; | |
6208 | } | |
6209 | } | |
6210 | } | |
6211 | ||
6212 | return NULL; | |
6213 | } | |
6214 | ||
61007b31 SH |
6215 | /* TODO check what callers really want: bs->node_name or blk_name() */ |
6216 | const char *bdrv_get_device_name(const BlockDriverState *bs) | |
beac80cd | 6217 | { |
384a48fb | 6218 | IO_CODE(); |
4c265bf9 | 6219 | return bdrv_get_parent_name(bs) ?: ""; |
f141eafe | 6220 | } |
83f64091 | 6221 | |
61007b31 SH |
6222 | /* This can be used to identify nodes that might not have a device |
6223 | * name associated. Since node and device names live in the same | |
6224 | * namespace, the result is unambiguous. The exception is if both are | |
6225 | * absent, then this returns an empty (non-null) string. */ | |
6226 | const char *bdrv_get_device_or_node_name(const BlockDriverState *bs) | |
f141eafe | 6227 | { |
384a48fb | 6228 | IO_CODE(); |
4c265bf9 | 6229 | return bdrv_get_parent_name(bs) ?: bs->node_name; |
beac80cd | 6230 | } |
beac80cd | 6231 | |
61007b31 | 6232 | int bdrv_get_flags(BlockDriverState *bs) |
0b5a2445 | 6233 | { |
15aee7ac | 6234 | IO_CODE(); |
61007b31 | 6235 | return bs->open_flags; |
0b5a2445 PB |
6236 | } |
6237 | ||
61007b31 | 6238 | int bdrv_has_zero_init_1(BlockDriverState *bs) |
68485420 | 6239 | { |
f791bf7f | 6240 | GLOBAL_STATE_CODE(); |
61007b31 | 6241 | return 1; |
0b5a2445 PB |
6242 | } |
6243 | ||
61007b31 | 6244 | int bdrv_has_zero_init(BlockDriverState *bs) |
0b5a2445 | 6245 | { |
93393e69 | 6246 | BlockDriverState *filtered; |
f791bf7f | 6247 | GLOBAL_STATE_CODE(); |
93393e69 | 6248 | |
d470ad42 HR |
6249 | if (!bs->drv) { |
6250 | return 0; | |
6251 | } | |
0b5a2445 | 6252 | |
61007b31 SH |
6253 | /* If BS is a copy on write image, it is initialized to |
6254 | the contents of the base image, which may not be zeroes. */ | |
34778172 | 6255 | if (bdrv_cow_child(bs)) { |
61007b31 SH |
6256 | return 0; |
6257 | } | |
6258 | if (bs->drv->bdrv_has_zero_init) { | |
6259 | return bs->drv->bdrv_has_zero_init(bs); | |
0b5a2445 | 6260 | } |
93393e69 HR |
6261 | |
6262 | filtered = bdrv_filter_bs(bs); | |
6263 | if (filtered) { | |
6264 | return bdrv_has_zero_init(filtered); | |
5a612c00 | 6265 | } |
61007b31 SH |
6266 | |
6267 | /* safe default */ | |
6268 | return 0; | |
68485420 KW |
6269 | } |
6270 | ||
61007b31 | 6271 | bool bdrv_can_write_zeroes_with_unmap(BlockDriverState *bs) |
68485420 | 6272 | { |
384a48fb | 6273 | IO_CODE(); |
2f0342ef | 6274 | if (!(bs->open_flags & BDRV_O_UNMAP)) { |
61007b31 SH |
6275 | return false; |
6276 | } | |
68485420 | 6277 | |
e24d813b | 6278 | return bs->supported_zero_flags & BDRV_REQ_MAY_UNMAP; |
68485420 KW |
6279 | } |
6280 | ||
61007b31 SH |
6281 | void bdrv_get_backing_filename(BlockDriverState *bs, |
6282 | char *filename, int filename_size) | |
016f5cf6 | 6283 | { |
384a48fb | 6284 | IO_CODE(); |
61007b31 SH |
6285 | pstrcpy(filename, filename_size, bs->backing_file); |
6286 | } | |
d318aea9 | 6287 | |
61007b31 SH |
6288 | int bdrv_get_info(BlockDriverState *bs, BlockDriverInfo *bdi) |
6289 | { | |
8b117001 | 6290 | int ret; |
61007b31 | 6291 | BlockDriver *drv = bs->drv; |
384a48fb | 6292 | IO_CODE(); |
5a612c00 MP |
6293 | /* if bs->drv == NULL, bs is closed, so there's nothing to do here */ |
6294 | if (!drv) { | |
61007b31 | 6295 | return -ENOMEDIUM; |
5a612c00 MP |
6296 | } |
6297 | if (!drv->bdrv_get_info) { | |
93393e69 HR |
6298 | BlockDriverState *filtered = bdrv_filter_bs(bs); |
6299 | if (filtered) { | |
6300 | return bdrv_get_info(filtered, bdi); | |
5a612c00 | 6301 | } |
61007b31 | 6302 | return -ENOTSUP; |
5a612c00 | 6303 | } |
61007b31 | 6304 | memset(bdi, 0, sizeof(*bdi)); |
8b117001 VSO |
6305 | ret = drv->bdrv_get_info(bs, bdi); |
6306 | if (ret < 0) { | |
6307 | return ret; | |
6308 | } | |
6309 | ||
6310 | if (bdi->cluster_size > BDRV_MAX_ALIGNMENT) { | |
6311 | return -EINVAL; | |
6312 | } | |
6313 | ||
6314 | return 0; | |
61007b31 | 6315 | } |
016f5cf6 | 6316 | |
1bf6e9ca AS |
6317 | ImageInfoSpecific *bdrv_get_specific_info(BlockDriverState *bs, |
6318 | Error **errp) | |
61007b31 SH |
6319 | { |
6320 | BlockDriver *drv = bs->drv; | |
384a48fb | 6321 | IO_CODE(); |
61007b31 | 6322 | if (drv && drv->bdrv_get_specific_info) { |
1bf6e9ca | 6323 | return drv->bdrv_get_specific_info(bs, errp); |
61007b31 SH |
6324 | } |
6325 | return NULL; | |
016f5cf6 AG |
6326 | } |
6327 | ||
d9245599 AN |
6328 | BlockStatsSpecific *bdrv_get_specific_stats(BlockDriverState *bs) |
6329 | { | |
6330 | BlockDriver *drv = bs->drv; | |
384a48fb | 6331 | IO_CODE(); |
d9245599 AN |
6332 | if (!drv || !drv->bdrv_get_specific_stats) { |
6333 | return NULL; | |
6334 | } | |
6335 | return drv->bdrv_get_specific_stats(bs); | |
6336 | } | |
6337 | ||
a31939e6 | 6338 | void bdrv_debug_event(BlockDriverState *bs, BlkdebugEvent event) |
4265d620 | 6339 | { |
384a48fb | 6340 | IO_CODE(); |
61007b31 SH |
6341 | if (!bs || !bs->drv || !bs->drv->bdrv_debug_event) { |
6342 | return; | |
6343 | } | |
4265d620 | 6344 | |
61007b31 | 6345 | bs->drv->bdrv_debug_event(bs, event); |
4265d620 PB |
6346 | } |
6347 | ||
d10529a2 | 6348 | static BlockDriverState *bdrv_find_debug_node(BlockDriverState *bs) |
4265d620 | 6349 | { |
bdb73476 | 6350 | GLOBAL_STATE_CODE(); |
61007b31 | 6351 | while (bs && bs->drv && !bs->drv->bdrv_debug_breakpoint) { |
f706a92f | 6352 | bs = bdrv_primary_bs(bs); |
61007b31 | 6353 | } |
4265d620 | 6354 | |
61007b31 | 6355 | if (bs && bs->drv && bs->drv->bdrv_debug_breakpoint) { |
d10529a2 VSO |
6356 | assert(bs->drv->bdrv_debug_remove_breakpoint); |
6357 | return bs; | |
6358 | } | |
6359 | ||
6360 | return NULL; | |
6361 | } | |
6362 | ||
6363 | int bdrv_debug_breakpoint(BlockDriverState *bs, const char *event, | |
6364 | const char *tag) | |
6365 | { | |
f791bf7f | 6366 | GLOBAL_STATE_CODE(); |
d10529a2 VSO |
6367 | bs = bdrv_find_debug_node(bs); |
6368 | if (bs) { | |
61007b31 SH |
6369 | return bs->drv->bdrv_debug_breakpoint(bs, event, tag); |
6370 | } | |
4265d620 | 6371 | |
61007b31 | 6372 | return -ENOTSUP; |
4265d620 PB |
6373 | } |
6374 | ||
61007b31 | 6375 | int bdrv_debug_remove_breakpoint(BlockDriverState *bs, const char *tag) |
ea2384d3 | 6376 | { |
f791bf7f | 6377 | GLOBAL_STATE_CODE(); |
d10529a2 VSO |
6378 | bs = bdrv_find_debug_node(bs); |
6379 | if (bs) { | |
61007b31 SH |
6380 | return bs->drv->bdrv_debug_remove_breakpoint(bs, tag); |
6381 | } | |
6382 | ||
6383 | return -ENOTSUP; | |
eb852011 MA |
6384 | } |
6385 | ||
61007b31 | 6386 | int bdrv_debug_resume(BlockDriverState *bs, const char *tag) |
ce1a14dc | 6387 | { |
f791bf7f | 6388 | GLOBAL_STATE_CODE(); |
61007b31 | 6389 | while (bs && (!bs->drv || !bs->drv->bdrv_debug_resume)) { |
f706a92f | 6390 | bs = bdrv_primary_bs(bs); |
61007b31 | 6391 | } |
ce1a14dc | 6392 | |
61007b31 SH |
6393 | if (bs && bs->drv && bs->drv->bdrv_debug_resume) { |
6394 | return bs->drv->bdrv_debug_resume(bs, tag); | |
6395 | } | |
ce1a14dc | 6396 | |
61007b31 | 6397 | return -ENOTSUP; |
f197fe2b FZ |
6398 | } |
6399 | ||
61007b31 | 6400 | bool bdrv_debug_is_suspended(BlockDriverState *bs, const char *tag) |
ce1a14dc | 6401 | { |
f791bf7f | 6402 | GLOBAL_STATE_CODE(); |
61007b31 | 6403 | while (bs && bs->drv && !bs->drv->bdrv_debug_is_suspended) { |
f706a92f | 6404 | bs = bdrv_primary_bs(bs); |
f197fe2b | 6405 | } |
19cb3738 | 6406 | |
61007b31 SH |
6407 | if (bs && bs->drv && bs->drv->bdrv_debug_is_suspended) { |
6408 | return bs->drv->bdrv_debug_is_suspended(bs, tag); | |
6409 | } | |
f9f05dc5 | 6410 | |
61007b31 SH |
6411 | return false; |
6412 | } | |
f9f05dc5 | 6413 | |
61007b31 SH |
6414 | /* backing_file can either be relative, or absolute, or a protocol. If it is |
6415 | * relative, it must be relative to the chain. So, passing in bs->filename | |
6416 | * from a BDS as backing_file should not be done, as that may be relative to | |
6417 | * the CWD rather than the chain. */ | |
6418 | BlockDriverState *bdrv_find_backing_image(BlockDriverState *bs, | |
6419 | const char *backing_file) | |
f9f05dc5 | 6420 | { |
61007b31 SH |
6421 | char *filename_full = NULL; |
6422 | char *backing_file_full = NULL; | |
6423 | char *filename_tmp = NULL; | |
6424 | int is_protocol = 0; | |
0b877d09 | 6425 | bool filenames_refreshed = false; |
61007b31 SH |
6426 | BlockDriverState *curr_bs = NULL; |
6427 | BlockDriverState *retval = NULL; | |
dcf3f9b2 | 6428 | BlockDriverState *bs_below; |
f9f05dc5 | 6429 | |
f791bf7f EGE |
6430 | GLOBAL_STATE_CODE(); |
6431 | ||
61007b31 SH |
6432 | if (!bs || !bs->drv || !backing_file) { |
6433 | return NULL; | |
f9f05dc5 KW |
6434 | } |
6435 | ||
61007b31 SH |
6436 | filename_full = g_malloc(PATH_MAX); |
6437 | backing_file_full = g_malloc(PATH_MAX); | |
f9f05dc5 | 6438 | |
61007b31 | 6439 | is_protocol = path_has_protocol(backing_file); |
f9f05dc5 | 6440 | |
dcf3f9b2 HR |
6441 | /* |
6442 | * Being largely a legacy function, skip any filters here | |
6443 | * (because filters do not have normal filenames, so they cannot | |
6444 | * match anyway; and allowing json:{} filenames is a bit out of | |
6445 | * scope). | |
6446 | */ | |
6447 | for (curr_bs = bdrv_skip_filters(bs); | |
6448 | bdrv_cow_child(curr_bs) != NULL; | |
6449 | curr_bs = bs_below) | |
6450 | { | |
6451 | bs_below = bdrv_backing_chain_next(curr_bs); | |
f9f05dc5 | 6452 | |
0b877d09 HR |
6453 | if (bdrv_backing_overridden(curr_bs)) { |
6454 | /* | |
6455 | * If the backing file was overridden, we can only compare | |
6456 | * directly against the backing node's filename. | |
6457 | */ | |
6458 | ||
6459 | if (!filenames_refreshed) { | |
6460 | /* | |
6461 | * This will automatically refresh all of the | |
6462 | * filenames in the rest of the backing chain, so we | |
6463 | * only need to do this once. | |
6464 | */ | |
6465 | bdrv_refresh_filename(bs_below); | |
6466 | filenames_refreshed = true; | |
6467 | } | |
6468 | ||
6469 | if (strcmp(backing_file, bs_below->filename) == 0) { | |
6470 | retval = bs_below; | |
6471 | break; | |
6472 | } | |
6473 | } else if (is_protocol || path_has_protocol(curr_bs->backing_file)) { | |
6474 | /* | |
6475 | * If either of the filename paths is actually a protocol, then | |
6476 | * compare unmodified paths; otherwise make paths relative. | |
6477 | */ | |
6b6833c1 HR |
6478 | char *backing_file_full_ret; |
6479 | ||
61007b31 | 6480 | if (strcmp(backing_file, curr_bs->backing_file) == 0) { |
dcf3f9b2 | 6481 | retval = bs_below; |
61007b31 SH |
6482 | break; |
6483 | } | |
418661e0 | 6484 | /* Also check against the full backing filename for the image */ |
6b6833c1 HR |
6485 | backing_file_full_ret = bdrv_get_full_backing_filename(curr_bs, |
6486 | NULL); | |
6487 | if (backing_file_full_ret) { | |
6488 | bool equal = strcmp(backing_file, backing_file_full_ret) == 0; | |
6489 | g_free(backing_file_full_ret); | |
6490 | if (equal) { | |
dcf3f9b2 | 6491 | retval = bs_below; |
418661e0 JC |
6492 | break; |
6493 | } | |
418661e0 | 6494 | } |
61007b31 SH |
6495 | } else { |
6496 | /* If not an absolute filename path, make it relative to the current | |
6497 | * image's filename path */ | |
2d9158ce HR |
6498 | filename_tmp = bdrv_make_absolute_filename(curr_bs, backing_file, |
6499 | NULL); | |
6500 | /* We are going to compare canonicalized absolute pathnames */ | |
6501 | if (!filename_tmp || !realpath(filename_tmp, filename_full)) { | |
6502 | g_free(filename_tmp); | |
61007b31 SH |
6503 | continue; |
6504 | } | |
2d9158ce | 6505 | g_free(filename_tmp); |
07f07615 | 6506 | |
61007b31 SH |
6507 | /* We need to make sure the backing filename we are comparing against |
6508 | * is relative to the current image filename (or absolute) */ | |
2d9158ce HR |
6509 | filename_tmp = bdrv_get_full_backing_filename(curr_bs, NULL); |
6510 | if (!filename_tmp || !realpath(filename_tmp, backing_file_full)) { | |
6511 | g_free(filename_tmp); | |
61007b31 SH |
6512 | continue; |
6513 | } | |
2d9158ce | 6514 | g_free(filename_tmp); |
eb489bb1 | 6515 | |
61007b31 | 6516 | if (strcmp(backing_file_full, filename_full) == 0) { |
dcf3f9b2 | 6517 | retval = bs_below; |
61007b31 SH |
6518 | break; |
6519 | } | |
6520 | } | |
eb489bb1 KW |
6521 | } |
6522 | ||
61007b31 SH |
6523 | g_free(filename_full); |
6524 | g_free(backing_file_full); | |
61007b31 SH |
6525 | return retval; |
6526 | } | |
6527 | ||
61007b31 SH |
6528 | void bdrv_init(void) |
6529 | { | |
e5f05f8c KW |
6530 | #ifdef CONFIG_BDRV_WHITELIST_TOOLS |
6531 | use_bdrv_whitelist = 1; | |
6532 | #endif | |
61007b31 SH |
6533 | module_call_init(MODULE_INIT_BLOCK); |
6534 | } | |
29cdb251 | 6535 | |
61007b31 SH |
6536 | void bdrv_init_with_whitelist(void) |
6537 | { | |
6538 | use_bdrv_whitelist = 1; | |
6539 | bdrv_init(); | |
07f07615 PB |
6540 | } |
6541 | ||
a94750d9 | 6542 | int bdrv_activate(BlockDriverState *bs, Error **errp) |
0f15423c | 6543 | { |
4417ab7a | 6544 | BdrvChild *child, *parent; |
5a8a30db KW |
6545 | Error *local_err = NULL; |
6546 | int ret; | |
9c98f145 | 6547 | BdrvDirtyBitmap *bm; |
5a8a30db | 6548 | |
f791bf7f EGE |
6549 | GLOBAL_STATE_CODE(); |
6550 | ||
3456a8d1 | 6551 | if (!bs->drv) { |
5416645f | 6552 | return -ENOMEDIUM; |
3456a8d1 KW |
6553 | } |
6554 | ||
16e977d5 | 6555 | QLIST_FOREACH(child, &bs->children, next) { |
11d0c9b3 | 6556 | bdrv_activate(child->bs, &local_err); |
0d1c5c91 | 6557 | if (local_err) { |
0d1c5c91 | 6558 | error_propagate(errp, local_err); |
5416645f | 6559 | return -EINVAL; |
0d1c5c91 | 6560 | } |
5a8a30db | 6561 | } |
0d1c5c91 | 6562 | |
dafe0960 KW |
6563 | /* |
6564 | * Update permissions, they may differ for inactive nodes. | |
6565 | * | |
6566 | * Note that the required permissions of inactive images are always a | |
6567 | * subset of the permissions required after activating the image. This | |
6568 | * allows us to just get the permissions upfront without restricting | |
11d0c9b3 | 6569 | * bdrv_co_invalidate_cache(). |
dafe0960 KW |
6570 | * |
6571 | * It also means that in error cases, we don't have to try and revert to | |
6572 | * the old permissions (which is an operation that could fail, too). We can | |
6573 | * just keep the extended permissions for the next time that an activation | |
6574 | * of the image is tried. | |
6575 | */ | |
7bb4941a KW |
6576 | if (bs->open_flags & BDRV_O_INACTIVE) { |
6577 | bs->open_flags &= ~BDRV_O_INACTIVE; | |
f1316edb | 6578 | ret = bdrv_refresh_perms(bs, NULL, errp); |
7bb4941a | 6579 | if (ret < 0) { |
0d1c5c91 | 6580 | bs->open_flags |= BDRV_O_INACTIVE; |
5416645f | 6581 | return ret; |
0d1c5c91 | 6582 | } |
3456a8d1 | 6583 | |
11d0c9b3 EGE |
6584 | ret = bdrv_invalidate_cache(bs, errp); |
6585 | if (ret < 0) { | |
6586 | bs->open_flags |= BDRV_O_INACTIVE; | |
6587 | return ret; | |
7bb4941a | 6588 | } |
9c98f145 | 6589 | |
7bb4941a KW |
6590 | FOR_EACH_DIRTY_BITMAP(bs, bm) { |
6591 | bdrv_dirty_bitmap_skip_store(bm, false); | |
6592 | } | |
6593 | ||
6594 | ret = refresh_total_sectors(bs, bs->total_sectors); | |
6595 | if (ret < 0) { | |
6596 | bs->open_flags |= BDRV_O_INACTIVE; | |
6597 | error_setg_errno(errp, -ret, "Could not refresh total sector count"); | |
5416645f | 6598 | return ret; |
7bb4941a | 6599 | } |
5a8a30db | 6600 | } |
4417ab7a KW |
6601 | |
6602 | QLIST_FOREACH(parent, &bs->parents, next_parent) { | |
bd86fb99 HR |
6603 | if (parent->klass->activate) { |
6604 | parent->klass->activate(parent, &local_err); | |
4417ab7a | 6605 | if (local_err) { |
78fc3b3a | 6606 | bs->open_flags |= BDRV_O_INACTIVE; |
4417ab7a | 6607 | error_propagate(errp, local_err); |
5416645f | 6608 | return -EINVAL; |
4417ab7a KW |
6609 | } |
6610 | } | |
6611 | } | |
5416645f VSO |
6612 | |
6613 | return 0; | |
0f15423c AL |
6614 | } |
6615 | ||
11d0c9b3 EGE |
6616 | int coroutine_fn bdrv_co_invalidate_cache(BlockDriverState *bs, Error **errp) |
6617 | { | |
6618 | Error *local_err = NULL; | |
1581a70d | 6619 | IO_CODE(); |
11d0c9b3 EGE |
6620 | |
6621 | assert(!(bs->open_flags & BDRV_O_INACTIVE)); | |
1b3ff9fe | 6622 | assert_bdrv_graph_readable(); |
11d0c9b3 EGE |
6623 | |
6624 | if (bs->drv->bdrv_co_invalidate_cache) { | |
6625 | bs->drv->bdrv_co_invalidate_cache(bs, &local_err); | |
6626 | if (local_err) { | |
6627 | error_propagate(errp, local_err); | |
6628 | return -EINVAL; | |
6629 | } | |
6630 | } | |
6631 | ||
6632 | return 0; | |
6633 | } | |
6634 | ||
3b717194 | 6635 | void bdrv_activate_all(Error **errp) |
0f15423c | 6636 | { |
7c8eece4 | 6637 | BlockDriverState *bs; |
88be7b4b | 6638 | BdrvNextIterator it; |
0f15423c | 6639 | |
f791bf7f EGE |
6640 | GLOBAL_STATE_CODE(); |
6641 | ||
88be7b4b | 6642 | for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) { |
ed78cda3 | 6643 | AioContext *aio_context = bdrv_get_aio_context(bs); |
5416645f | 6644 | int ret; |
ed78cda3 SH |
6645 | |
6646 | aio_context_acquire(aio_context); | |
a94750d9 | 6647 | ret = bdrv_activate(bs, errp); |
ed78cda3 | 6648 | aio_context_release(aio_context); |
5416645f | 6649 | if (ret < 0) { |
5e003f17 | 6650 | bdrv_next_cleanup(&it); |
5a8a30db KW |
6651 | return; |
6652 | } | |
0f15423c AL |
6653 | } |
6654 | } | |
6655 | ||
9e37271f KW |
6656 | static bool bdrv_has_bds_parent(BlockDriverState *bs, bool only_active) |
6657 | { | |
6658 | BdrvChild *parent; | |
bdb73476 | 6659 | GLOBAL_STATE_CODE(); |
9e37271f KW |
6660 | |
6661 | QLIST_FOREACH(parent, &bs->parents, next_parent) { | |
bd86fb99 | 6662 | if (parent->klass->parent_is_bds) { |
9e37271f KW |
6663 | BlockDriverState *parent_bs = parent->opaque; |
6664 | if (!only_active || !(parent_bs->open_flags & BDRV_O_INACTIVE)) { | |
6665 | return true; | |
6666 | } | |
6667 | } | |
6668 | } | |
6669 | ||
6670 | return false; | |
6671 | } | |
6672 | ||
6673 | static int bdrv_inactivate_recurse(BlockDriverState *bs) | |
76b1c7fe | 6674 | { |
cfa1a572 | 6675 | BdrvChild *child, *parent; |
76b1c7fe | 6676 | int ret; |
a13de40a | 6677 | uint64_t cumulative_perms, cumulative_shared_perms; |
76b1c7fe | 6678 | |
da359909 EGE |
6679 | GLOBAL_STATE_CODE(); |
6680 | ||
d470ad42 HR |
6681 | if (!bs->drv) { |
6682 | return -ENOMEDIUM; | |
6683 | } | |
6684 | ||
9e37271f KW |
6685 | /* Make sure that we don't inactivate a child before its parent. |
6686 | * It will be covered by recursion from the yet active parent. */ | |
6687 | if (bdrv_has_bds_parent(bs, true)) { | |
6688 | return 0; | |
6689 | } | |
6690 | ||
6691 | assert(!(bs->open_flags & BDRV_O_INACTIVE)); | |
6692 | ||
6693 | /* Inactivate this node */ | |
6694 | if (bs->drv->bdrv_inactivate) { | |
76b1c7fe KW |
6695 | ret = bs->drv->bdrv_inactivate(bs); |
6696 | if (ret < 0) { | |
6697 | return ret; | |
6698 | } | |
6699 | } | |
6700 | ||
9e37271f | 6701 | QLIST_FOREACH(parent, &bs->parents, next_parent) { |
bd86fb99 HR |
6702 | if (parent->klass->inactivate) { |
6703 | ret = parent->klass->inactivate(parent); | |
9e37271f KW |
6704 | if (ret < 0) { |
6705 | return ret; | |
cfa1a572 KW |
6706 | } |
6707 | } | |
9e37271f | 6708 | } |
9c5e6594 | 6709 | |
a13de40a VSO |
6710 | bdrv_get_cumulative_perm(bs, &cumulative_perms, |
6711 | &cumulative_shared_perms); | |
6712 | if (cumulative_perms & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)) { | |
6713 | /* Our inactive parents still need write access. Inactivation failed. */ | |
6714 | return -EPERM; | |
6715 | } | |
6716 | ||
9e37271f | 6717 | bs->open_flags |= BDRV_O_INACTIVE; |
7d5b5261 | 6718 | |
bb87e4d1 VSO |
6719 | /* |
6720 | * Update permissions, they may differ for inactive nodes. | |
6721 | * We only tried to loosen restrictions, so errors are not fatal, ignore | |
6722 | * them. | |
6723 | */ | |
f1316edb | 6724 | bdrv_refresh_perms(bs, NULL, NULL); |
9e37271f KW |
6725 | |
6726 | /* Recursively inactivate children */ | |
38701b6a | 6727 | QLIST_FOREACH(child, &bs->children, next) { |
9e37271f | 6728 | ret = bdrv_inactivate_recurse(child->bs); |
38701b6a KW |
6729 | if (ret < 0) { |
6730 | return ret; | |
6731 | } | |
6732 | } | |
6733 | ||
76b1c7fe KW |
6734 | return 0; |
6735 | } | |
6736 | ||
6737 | int bdrv_inactivate_all(void) | |
6738 | { | |
79720af6 | 6739 | BlockDriverState *bs = NULL; |
88be7b4b | 6740 | BdrvNextIterator it; |
aad0b7a0 | 6741 | int ret = 0; |
bd6458e4 | 6742 | GSList *aio_ctxs = NULL, *ctx; |
76b1c7fe | 6743 | |
f791bf7f EGE |
6744 | GLOBAL_STATE_CODE(); |
6745 | ||
88be7b4b | 6746 | for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) { |
bd6458e4 PB |
6747 | AioContext *aio_context = bdrv_get_aio_context(bs); |
6748 | ||
6749 | if (!g_slist_find(aio_ctxs, aio_context)) { | |
6750 | aio_ctxs = g_slist_prepend(aio_ctxs, aio_context); | |
6751 | aio_context_acquire(aio_context); | |
6752 | } | |
aad0b7a0 | 6753 | } |
76b1c7fe | 6754 | |
9e37271f KW |
6755 | for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) { |
6756 | /* Nodes with BDS parents are covered by recursion from the last | |
6757 | * parent that gets inactivated. Don't inactivate them a second | |
6758 | * time if that has already happened. */ | |
6759 | if (bdrv_has_bds_parent(bs, false)) { | |
6760 | continue; | |
6761 | } | |
6762 | ret = bdrv_inactivate_recurse(bs); | |
6763 | if (ret < 0) { | |
6764 | bdrv_next_cleanup(&it); | |
6765 | goto out; | |
76b1c7fe KW |
6766 | } |
6767 | } | |
6768 | ||
aad0b7a0 | 6769 | out: |
bd6458e4 PB |
6770 | for (ctx = aio_ctxs; ctx != NULL; ctx = ctx->next) { |
6771 | AioContext *aio_context = ctx->data; | |
6772 | aio_context_release(aio_context); | |
aad0b7a0 | 6773 | } |
bd6458e4 | 6774 | g_slist_free(aio_ctxs); |
aad0b7a0 FZ |
6775 | |
6776 | return ret; | |
76b1c7fe KW |
6777 | } |
6778 | ||
19cb3738 FB |
6779 | /**************************************************************/ |
6780 | /* removable device support */ | |
6781 | ||
6782 | /** | |
6783 | * Return TRUE if the media is present | |
6784 | */ | |
e031f750 | 6785 | bool bdrv_is_inserted(BlockDriverState *bs) |
19cb3738 FB |
6786 | { |
6787 | BlockDriver *drv = bs->drv; | |
28d7a789 | 6788 | BdrvChild *child; |
384a48fb | 6789 | IO_CODE(); |
a1aff5bf | 6790 | |
e031f750 HR |
6791 | if (!drv) { |
6792 | return false; | |
6793 | } | |
28d7a789 HR |
6794 | if (drv->bdrv_is_inserted) { |
6795 | return drv->bdrv_is_inserted(bs); | |
6796 | } | |
6797 | QLIST_FOREACH(child, &bs->children, next) { | |
6798 | if (!bdrv_is_inserted(child->bs)) { | |
6799 | return false; | |
6800 | } | |
e031f750 | 6801 | } |
28d7a789 | 6802 | return true; |
19cb3738 FB |
6803 | } |
6804 | ||
19cb3738 FB |
6805 | /** |
6806 | * If eject_flag is TRUE, eject the media. Otherwise, close the tray | |
6807 | */ | |
f36f3949 | 6808 | void bdrv_eject(BlockDriverState *bs, bool eject_flag) |
19cb3738 FB |
6809 | { |
6810 | BlockDriver *drv = bs->drv; | |
384a48fb | 6811 | IO_CODE(); |
19cb3738 | 6812 | |
822e1cd1 MA |
6813 | if (drv && drv->bdrv_eject) { |
6814 | drv->bdrv_eject(bs, eject_flag); | |
19cb3738 FB |
6815 | } |
6816 | } | |
6817 | ||
19cb3738 FB |
6818 | /** |
6819 | * Lock or unlock the media (if it is locked, the user won't be able | |
6820 | * to eject it manually). | |
6821 | */ | |
025e849a | 6822 | void bdrv_lock_medium(BlockDriverState *bs, bool locked) |
19cb3738 FB |
6823 | { |
6824 | BlockDriver *drv = bs->drv; | |
384a48fb | 6825 | IO_CODE(); |
025e849a | 6826 | trace_bdrv_lock_medium(bs, locked); |
b8c6d095 | 6827 | |
025e849a MA |
6828 | if (drv && drv->bdrv_lock_medium) { |
6829 | drv->bdrv_lock_medium(bs, locked); | |
19cb3738 FB |
6830 | } |
6831 | } | |
985a03b0 | 6832 | |
9fcb0251 FZ |
6833 | /* Get a reference to bs */ |
6834 | void bdrv_ref(BlockDriverState *bs) | |
6835 | { | |
f791bf7f | 6836 | GLOBAL_STATE_CODE(); |
9fcb0251 FZ |
6837 | bs->refcnt++; |
6838 | } | |
6839 | ||
6840 | /* Release a previously grabbed reference to bs. | |
6841 | * If after releasing, reference count is zero, the BlockDriverState is | |
6842 | * deleted. */ | |
6843 | void bdrv_unref(BlockDriverState *bs) | |
6844 | { | |
f791bf7f | 6845 | GLOBAL_STATE_CODE(); |
9a4d5ca6 JC |
6846 | if (!bs) { |
6847 | return; | |
6848 | } | |
9fcb0251 FZ |
6849 | assert(bs->refcnt > 0); |
6850 | if (--bs->refcnt == 0) { | |
6851 | bdrv_delete(bs); | |
6852 | } | |
6853 | } | |
6854 | ||
fbe40ff7 FZ |
6855 | struct BdrvOpBlocker { |
6856 | Error *reason; | |
6857 | QLIST_ENTRY(BdrvOpBlocker) list; | |
6858 | }; | |
6859 | ||
6860 | bool bdrv_op_is_blocked(BlockDriverState *bs, BlockOpType op, Error **errp) | |
6861 | { | |
6862 | BdrvOpBlocker *blocker; | |
f791bf7f | 6863 | GLOBAL_STATE_CODE(); |
fbe40ff7 FZ |
6864 | assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX); |
6865 | if (!QLIST_EMPTY(&bs->op_blockers[op])) { | |
6866 | blocker = QLIST_FIRST(&bs->op_blockers[op]); | |
4b576648 MA |
6867 | error_propagate_prepend(errp, error_copy(blocker->reason), |
6868 | "Node '%s' is busy: ", | |
6869 | bdrv_get_device_or_node_name(bs)); | |
fbe40ff7 FZ |
6870 | return true; |
6871 | } | |
6872 | return false; | |
6873 | } | |
6874 | ||
6875 | void bdrv_op_block(BlockDriverState *bs, BlockOpType op, Error *reason) | |
6876 | { | |
6877 | BdrvOpBlocker *blocker; | |
f791bf7f | 6878 | GLOBAL_STATE_CODE(); |
fbe40ff7 FZ |
6879 | assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX); |
6880 | ||
5839e53b | 6881 | blocker = g_new0(BdrvOpBlocker, 1); |
fbe40ff7 FZ |
6882 | blocker->reason = reason; |
6883 | QLIST_INSERT_HEAD(&bs->op_blockers[op], blocker, list); | |
6884 | } | |
6885 | ||
6886 | void bdrv_op_unblock(BlockDriverState *bs, BlockOpType op, Error *reason) | |
6887 | { | |
6888 | BdrvOpBlocker *blocker, *next; | |
f791bf7f | 6889 | GLOBAL_STATE_CODE(); |
fbe40ff7 FZ |
6890 | assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX); |
6891 | QLIST_FOREACH_SAFE(blocker, &bs->op_blockers[op], list, next) { | |
6892 | if (blocker->reason == reason) { | |
6893 | QLIST_REMOVE(blocker, list); | |
6894 | g_free(blocker); | |
6895 | } | |
6896 | } | |
6897 | } | |
6898 | ||
6899 | void bdrv_op_block_all(BlockDriverState *bs, Error *reason) | |
6900 | { | |
6901 | int i; | |
f791bf7f | 6902 | GLOBAL_STATE_CODE(); |
fbe40ff7 FZ |
6903 | for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) { |
6904 | bdrv_op_block(bs, i, reason); | |
6905 | } | |
6906 | } | |
6907 | ||
6908 | void bdrv_op_unblock_all(BlockDriverState *bs, Error *reason) | |
6909 | { | |
6910 | int i; | |
f791bf7f | 6911 | GLOBAL_STATE_CODE(); |
fbe40ff7 FZ |
6912 | for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) { |
6913 | bdrv_op_unblock(bs, i, reason); | |
6914 | } | |
6915 | } | |
6916 | ||
6917 | bool bdrv_op_blocker_is_empty(BlockDriverState *bs) | |
6918 | { | |
6919 | int i; | |
f791bf7f | 6920 | GLOBAL_STATE_CODE(); |
fbe40ff7 FZ |
6921 | for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) { |
6922 | if (!QLIST_EMPTY(&bs->op_blockers[i])) { | |
6923 | return false; | |
6924 | } | |
6925 | } | |
6926 | return true; | |
6927 | } | |
6928 | ||
e1355055 KW |
6929 | /* |
6930 | * Must not be called while holding the lock of an AioContext other than the | |
6931 | * current one. | |
6932 | */ | |
d92ada22 LC |
6933 | void bdrv_img_create(const char *filename, const char *fmt, |
6934 | const char *base_filename, const char *base_fmt, | |
9217283d FZ |
6935 | char *options, uint64_t img_size, int flags, bool quiet, |
6936 | Error **errp) | |
f88e1a42 | 6937 | { |
83d0521a CL |
6938 | QemuOptsList *create_opts = NULL; |
6939 | QemuOpts *opts = NULL; | |
6940 | const char *backing_fmt, *backing_file; | |
6941 | int64_t size; | |
f88e1a42 | 6942 | BlockDriver *drv, *proto_drv; |
cc84d90f | 6943 | Error *local_err = NULL; |
f88e1a42 JS |
6944 | int ret = 0; |
6945 | ||
f791bf7f EGE |
6946 | GLOBAL_STATE_CODE(); |
6947 | ||
f88e1a42 JS |
6948 | /* Find driver and parse its options */ |
6949 | drv = bdrv_find_format(fmt); | |
6950 | if (!drv) { | |
71c79813 | 6951 | error_setg(errp, "Unknown file format '%s'", fmt); |
d92ada22 | 6952 | return; |
f88e1a42 JS |
6953 | } |
6954 | ||
b65a5e12 | 6955 | proto_drv = bdrv_find_protocol(filename, true, errp); |
f88e1a42 | 6956 | if (!proto_drv) { |
d92ada22 | 6957 | return; |
f88e1a42 JS |
6958 | } |
6959 | ||
c6149724 HR |
6960 | if (!drv->create_opts) { |
6961 | error_setg(errp, "Format driver '%s' does not support image creation", | |
6962 | drv->format_name); | |
6963 | return; | |
6964 | } | |
6965 | ||
5a5e7f8c ML |
6966 | if (!proto_drv->create_opts) { |
6967 | error_setg(errp, "Protocol driver '%s' does not support image creation", | |
6968 | proto_drv->format_name); | |
6969 | return; | |
6970 | } | |
6971 | ||
f6dc1c31 | 6972 | /* Create parameter list */ |
c282e1fd | 6973 | create_opts = qemu_opts_append(create_opts, drv->create_opts); |
5a5e7f8c | 6974 | create_opts = qemu_opts_append(create_opts, proto_drv->create_opts); |
f88e1a42 | 6975 | |
83d0521a | 6976 | opts = qemu_opts_create(create_opts, NULL, 0, &error_abort); |
f88e1a42 JS |
6977 | |
6978 | /* Parse -o options */ | |
6979 | if (options) { | |
a5f9b9df | 6980 | if (!qemu_opts_do_parse(opts, options, NULL, errp)) { |
f88e1a42 JS |
6981 | goto out; |
6982 | } | |
6983 | } | |
6984 | ||
f6dc1c31 KW |
6985 | if (!qemu_opt_get(opts, BLOCK_OPT_SIZE)) { |
6986 | qemu_opt_set_number(opts, BLOCK_OPT_SIZE, img_size, &error_abort); | |
6987 | } else if (img_size != UINT64_C(-1)) { | |
6988 | error_setg(errp, "The image size must be specified only once"); | |
6989 | goto out; | |
6990 | } | |
6991 | ||
f88e1a42 | 6992 | if (base_filename) { |
235e59cf | 6993 | if (!qemu_opt_set(opts, BLOCK_OPT_BACKING_FILE, base_filename, |
3882578b | 6994 | NULL)) { |
71c79813 LC |
6995 | error_setg(errp, "Backing file not supported for file format '%s'", |
6996 | fmt); | |
f88e1a42 JS |
6997 | goto out; |
6998 | } | |
6999 | } | |
7000 | ||
7001 | if (base_fmt) { | |
3882578b | 7002 | if (!qemu_opt_set(opts, BLOCK_OPT_BACKING_FMT, base_fmt, NULL)) { |
71c79813 LC |
7003 | error_setg(errp, "Backing file format not supported for file " |
7004 | "format '%s'", fmt); | |
f88e1a42 JS |
7005 | goto out; |
7006 | } | |
7007 | } | |
7008 | ||
83d0521a CL |
7009 | backing_file = qemu_opt_get(opts, BLOCK_OPT_BACKING_FILE); |
7010 | if (backing_file) { | |
7011 | if (!strcmp(filename, backing_file)) { | |
71c79813 LC |
7012 | error_setg(errp, "Error: Trying to create an image with the " |
7013 | "same filename as the backing file"); | |
792da93a JS |
7014 | goto out; |
7015 | } | |
975a7bd2 CK |
7016 | if (backing_file[0] == '\0') { |
7017 | error_setg(errp, "Expected backing file name, got empty string"); | |
7018 | goto out; | |
7019 | } | |
792da93a JS |
7020 | } |
7021 | ||
83d0521a | 7022 | backing_fmt = qemu_opt_get(opts, BLOCK_OPT_BACKING_FMT); |
f88e1a42 | 7023 | |
6e6e55f5 JS |
7024 | /* The size for the image must always be specified, unless we have a backing |
7025 | * file and we have not been forbidden from opening it. */ | |
a8b42a1c | 7026 | size = qemu_opt_get_size(opts, BLOCK_OPT_SIZE, img_size); |
6e6e55f5 JS |
7027 | if (backing_file && !(flags & BDRV_O_NO_BACKING)) { |
7028 | BlockDriverState *bs; | |
645ae7d8 | 7029 | char *full_backing; |
6e6e55f5 JS |
7030 | int back_flags; |
7031 | QDict *backing_options = NULL; | |
7032 | ||
645ae7d8 HR |
7033 | full_backing = |
7034 | bdrv_get_full_backing_filename_from_filename(filename, backing_file, | |
7035 | &local_err); | |
6e6e55f5 | 7036 | if (local_err) { |
6e6e55f5 JS |
7037 | goto out; |
7038 | } | |
645ae7d8 | 7039 | assert(full_backing); |
29168018 | 7040 | |
d5b23994 HR |
7041 | /* |
7042 | * No need to do I/O here, which allows us to open encrypted | |
7043 | * backing images without needing the secret | |
7044 | */ | |
6e6e55f5 JS |
7045 | back_flags = flags; |
7046 | back_flags &= ~(BDRV_O_RDWR | BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING); | |
d5b23994 | 7047 | back_flags |= BDRV_O_NO_IO; |
f88e1a42 | 7048 | |
cc954f01 | 7049 | backing_options = qdict_new(); |
6e6e55f5 | 7050 | if (backing_fmt) { |
6e6e55f5 JS |
7051 | qdict_put_str(backing_options, "driver", backing_fmt); |
7052 | } | |
cc954f01 | 7053 | qdict_put_bool(backing_options, BDRV_OPT_FORCE_SHARE, true); |
e6641719 | 7054 | |
6e6e55f5 JS |
7055 | bs = bdrv_open(full_backing, NULL, backing_options, back_flags, |
7056 | &local_err); | |
7057 | g_free(full_backing); | |
add8200d EB |
7058 | if (!bs) { |
7059 | error_append_hint(&local_err, "Could not open backing image.\n"); | |
6e6e55f5 JS |
7060 | goto out; |
7061 | } else { | |
d9f059aa | 7062 | if (!backing_fmt) { |
497a30db EB |
7063 | error_setg(&local_err, |
7064 | "Backing file specified without backing format"); | |
7065 | error_append_hint(&local_err, "Detected format of %s.", | |
7066 | bs->drv->format_name); | |
7067 | goto out; | |
d9f059aa | 7068 | } |
6e6e55f5 JS |
7069 | if (size == -1) { |
7070 | /* Opened BS, have no size */ | |
7071 | size = bdrv_getlength(bs); | |
7072 | if (size < 0) { | |
7073 | error_setg_errno(errp, -size, "Could not get size of '%s'", | |
7074 | backing_file); | |
7075 | bdrv_unref(bs); | |
7076 | goto out; | |
7077 | } | |
7078 | qemu_opt_set_number(opts, BLOCK_OPT_SIZE, size, &error_abort); | |
52bf1e72 | 7079 | } |
66f6b814 | 7080 | bdrv_unref(bs); |
f88e1a42 | 7081 | } |
d9f059aa EB |
7082 | /* (backing_file && !(flags & BDRV_O_NO_BACKING)) */ |
7083 | } else if (backing_file && !backing_fmt) { | |
497a30db EB |
7084 | error_setg(&local_err, |
7085 | "Backing file specified without backing format"); | |
7086 | goto out; | |
d9f059aa | 7087 | } |
6e6e55f5 JS |
7088 | |
7089 | if (size == -1) { | |
7090 | error_setg(errp, "Image creation needs a size parameter"); | |
7091 | goto out; | |
f88e1a42 JS |
7092 | } |
7093 | ||
f382d43a | 7094 | if (!quiet) { |
fe646693 | 7095 | printf("Formatting '%s', fmt=%s ", filename, fmt); |
43c5d8f8 | 7096 | qemu_opts_print(opts, " "); |
f382d43a | 7097 | puts(""); |
4e2f4418 | 7098 | fflush(stdout); |
f382d43a | 7099 | } |
83d0521a | 7100 | |
c282e1fd | 7101 | ret = bdrv_create(drv, filename, opts, &local_err); |
83d0521a | 7102 | |
cc84d90f HR |
7103 | if (ret == -EFBIG) { |
7104 | /* This is generally a better message than whatever the driver would | |
7105 | * deliver (especially because of the cluster_size_hint), since that | |
7106 | * is most probably not much different from "image too large". */ | |
7107 | const char *cluster_size_hint = ""; | |
83d0521a | 7108 | if (qemu_opt_get_size(opts, BLOCK_OPT_CLUSTER_SIZE, 0)) { |
cc84d90f | 7109 | cluster_size_hint = " (try using a larger cluster size)"; |
f88e1a42 | 7110 | } |
cc84d90f HR |
7111 | error_setg(errp, "The image size is too large for file format '%s'" |
7112 | "%s", fmt, cluster_size_hint); | |
7113 | error_free(local_err); | |
7114 | local_err = NULL; | |
f88e1a42 JS |
7115 | } |
7116 | ||
7117 | out: | |
83d0521a CL |
7118 | qemu_opts_del(opts); |
7119 | qemu_opts_free(create_opts); | |
621ff94d | 7120 | error_propagate(errp, local_err); |
f88e1a42 | 7121 | } |
85d126f3 SH |
7122 | |
7123 | AioContext *bdrv_get_aio_context(BlockDriverState *bs) | |
7124 | { | |
384a48fb | 7125 | IO_CODE(); |
33f2a757 | 7126 | return bs ? bs->aio_context : qemu_get_aio_context(); |
dcd04228 SH |
7127 | } |
7128 | ||
e336fd4c KW |
7129 | AioContext *coroutine_fn bdrv_co_enter(BlockDriverState *bs) |
7130 | { | |
7131 | Coroutine *self = qemu_coroutine_self(); | |
7132 | AioContext *old_ctx = qemu_coroutine_get_aio_context(self); | |
7133 | AioContext *new_ctx; | |
384a48fb | 7134 | IO_CODE(); |
e336fd4c KW |
7135 | |
7136 | /* | |
7137 | * Increase bs->in_flight to ensure that this operation is completed before | |
7138 | * moving the node to a different AioContext. Read new_ctx only afterwards. | |
7139 | */ | |
7140 | bdrv_inc_in_flight(bs); | |
7141 | ||
7142 | new_ctx = bdrv_get_aio_context(bs); | |
7143 | aio_co_reschedule_self(new_ctx); | |
7144 | return old_ctx; | |
7145 | } | |
7146 | ||
7147 | void coroutine_fn bdrv_co_leave(BlockDriverState *bs, AioContext *old_ctx) | |
7148 | { | |
384a48fb | 7149 | IO_CODE(); |
e336fd4c KW |
7150 | aio_co_reschedule_self(old_ctx); |
7151 | bdrv_dec_in_flight(bs); | |
7152 | } | |
7153 | ||
18c6ac1c KW |
7154 | void coroutine_fn bdrv_co_lock(BlockDriverState *bs) |
7155 | { | |
7156 | AioContext *ctx = bdrv_get_aio_context(bs); | |
7157 | ||
7158 | /* In the main thread, bs->aio_context won't change concurrently */ | |
7159 | assert(qemu_get_current_aio_context() == qemu_get_aio_context()); | |
7160 | ||
7161 | /* | |
7162 | * We're in coroutine context, so we already hold the lock of the main | |
7163 | * loop AioContext. Don't lock it twice to avoid deadlocks. | |
7164 | */ | |
7165 | assert(qemu_in_coroutine()); | |
7166 | if (ctx != qemu_get_aio_context()) { | |
7167 | aio_context_acquire(ctx); | |
7168 | } | |
7169 | } | |
7170 | ||
7171 | void coroutine_fn bdrv_co_unlock(BlockDriverState *bs) | |
7172 | { | |
7173 | AioContext *ctx = bdrv_get_aio_context(bs); | |
7174 | ||
7175 | assert(qemu_in_coroutine()); | |
7176 | if (ctx != qemu_get_aio_context()) { | |
7177 | aio_context_release(ctx); | |
7178 | } | |
7179 | } | |
7180 | ||
052a7572 FZ |
7181 | void bdrv_coroutine_enter(BlockDriverState *bs, Coroutine *co) |
7182 | { | |
384a48fb | 7183 | IO_CODE(); |
052a7572 FZ |
7184 | aio_co_enter(bdrv_get_aio_context(bs), co); |
7185 | } | |
7186 | ||
e8a095da SH |
7187 | static void bdrv_do_remove_aio_context_notifier(BdrvAioNotifier *ban) |
7188 | { | |
bdb73476 | 7189 | GLOBAL_STATE_CODE(); |
e8a095da SH |
7190 | QLIST_REMOVE(ban, list); |
7191 | g_free(ban); | |
7192 | } | |
7193 | ||
a3a683c3 | 7194 | static void bdrv_detach_aio_context(BlockDriverState *bs) |
dcd04228 | 7195 | { |
e8a095da | 7196 | BdrvAioNotifier *baf, *baf_tmp; |
33384421 | 7197 | |
e8a095da | 7198 | assert(!bs->walking_aio_notifiers); |
da359909 | 7199 | GLOBAL_STATE_CODE(); |
e8a095da SH |
7200 | bs->walking_aio_notifiers = true; |
7201 | QLIST_FOREACH_SAFE(baf, &bs->aio_notifiers, list, baf_tmp) { | |
7202 | if (baf->deleted) { | |
7203 | bdrv_do_remove_aio_context_notifier(baf); | |
7204 | } else { | |
7205 | baf->detach_aio_context(baf->opaque); | |
7206 | } | |
33384421 | 7207 | } |
e8a095da SH |
7208 | /* Never mind iterating again to check for ->deleted. bdrv_close() will |
7209 | * remove remaining aio notifiers if we aren't called again. | |
7210 | */ | |
7211 | bs->walking_aio_notifiers = false; | |
33384421 | 7212 | |
1bffe1ae | 7213 | if (bs->drv && bs->drv->bdrv_detach_aio_context) { |
dcd04228 SH |
7214 | bs->drv->bdrv_detach_aio_context(bs); |
7215 | } | |
dcd04228 | 7216 | |
e64f25f3 KW |
7217 | if (bs->quiesce_counter) { |
7218 | aio_enable_external(bs->aio_context); | |
7219 | } | |
dcd04228 SH |
7220 | bs->aio_context = NULL; |
7221 | } | |
7222 | ||
a3a683c3 KW |
7223 | static void bdrv_attach_aio_context(BlockDriverState *bs, |
7224 | AioContext *new_context) | |
dcd04228 | 7225 | { |
e8a095da | 7226 | BdrvAioNotifier *ban, *ban_tmp; |
da359909 | 7227 | GLOBAL_STATE_CODE(); |
33384421 | 7228 | |
e64f25f3 KW |
7229 | if (bs->quiesce_counter) { |
7230 | aio_disable_external(new_context); | |
7231 | } | |
7232 | ||
dcd04228 SH |
7233 | bs->aio_context = new_context; |
7234 | ||
1bffe1ae | 7235 | if (bs->drv && bs->drv->bdrv_attach_aio_context) { |
dcd04228 SH |
7236 | bs->drv->bdrv_attach_aio_context(bs, new_context); |
7237 | } | |
33384421 | 7238 | |
e8a095da SH |
7239 | assert(!bs->walking_aio_notifiers); |
7240 | bs->walking_aio_notifiers = true; | |
7241 | QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_tmp) { | |
7242 | if (ban->deleted) { | |
7243 | bdrv_do_remove_aio_context_notifier(ban); | |
7244 | } else { | |
7245 | ban->attached_aio_context(new_context, ban->opaque); | |
7246 | } | |
33384421 | 7247 | } |
e8a095da | 7248 | bs->walking_aio_notifiers = false; |
dcd04228 SH |
7249 | } |
7250 | ||
7e8c182f EGE |
7251 | typedef struct BdrvStateSetAioContext { |
7252 | AioContext *new_ctx; | |
7253 | BlockDriverState *bs; | |
7254 | } BdrvStateSetAioContext; | |
d616b224 | 7255 | |
7e8c182f | 7256 | static bool bdrv_parent_change_aio_context(BdrvChild *c, AioContext *ctx, |
e08cc001 EGE |
7257 | GHashTable *visited, |
7258 | Transaction *tran, | |
7e8c182f | 7259 | Error **errp) |
5d231849 | 7260 | { |
f0c28327 | 7261 | GLOBAL_STATE_CODE(); |
e08cc001 | 7262 | if (g_hash_table_contains(visited, c)) { |
5d231849 KW |
7263 | return true; |
7264 | } | |
e08cc001 | 7265 | g_hash_table_add(visited, c); |
5d231849 | 7266 | |
bd86fb99 HR |
7267 | /* |
7268 | * A BdrvChildClass that doesn't handle AioContext changes cannot | |
7269 | * tolerate any AioContext changes | |
7270 | */ | |
7e8c182f | 7271 | if (!c->klass->change_aio_ctx) { |
5d231849 KW |
7272 | char *user = bdrv_child_user_desc(c); |
7273 | error_setg(errp, "Changing iothreads is not supported by %s", user); | |
7274 | g_free(user); | |
7275 | return false; | |
7276 | } | |
7e8c182f | 7277 | if (!c->klass->change_aio_ctx(c, ctx, visited, tran, errp)) { |
5d231849 KW |
7278 | assert(!errp || *errp); |
7279 | return false; | |
7280 | } | |
7281 | return true; | |
7282 | } | |
7283 | ||
7e8c182f | 7284 | bool bdrv_child_change_aio_context(BdrvChild *c, AioContext *ctx, |
e08cc001 | 7285 | GHashTable *visited, Transaction *tran, |
7e8c182f | 7286 | Error **errp) |
5d231849 | 7287 | { |
f791bf7f | 7288 | GLOBAL_STATE_CODE(); |
e08cc001 | 7289 | if (g_hash_table_contains(visited, c)) { |
5d231849 KW |
7290 | return true; |
7291 | } | |
e08cc001 | 7292 | g_hash_table_add(visited, c); |
7e8c182f | 7293 | return bdrv_change_aio_context(c->bs, ctx, visited, tran, errp); |
5d231849 KW |
7294 | } |
7295 | ||
7e8c182f EGE |
7296 | static void bdrv_set_aio_context_clean(void *opaque) |
7297 | { | |
7298 | BdrvStateSetAioContext *state = (BdrvStateSetAioContext *) opaque; | |
7299 | BlockDriverState *bs = (BlockDriverState *) state->bs; | |
7300 | ||
7301 | /* Paired with bdrv_drained_begin in bdrv_change_aio_context() */ | |
7302 | bdrv_drained_end(bs); | |
7303 | ||
7304 | g_free(state); | |
7305 | } | |
7306 | ||
7307 | static void bdrv_set_aio_context_commit(void *opaque) | |
7308 | { | |
7309 | BdrvStateSetAioContext *state = (BdrvStateSetAioContext *) opaque; | |
7310 | BlockDriverState *bs = (BlockDriverState *) state->bs; | |
7311 | AioContext *new_context = state->new_ctx; | |
7312 | AioContext *old_context = bdrv_get_aio_context(bs); | |
7e8c182f EGE |
7313 | |
7314 | /* | |
7315 | * Take the old AioContex when detaching it from bs. | |
7316 | * At this point, new_context lock is already acquired, and we are now | |
7317 | * also taking old_context. This is safe as long as bdrv_detach_aio_context | |
7318 | * does not call AIO_POLL_WHILE(). | |
7319 | */ | |
7320 | if (old_context != qemu_get_aio_context()) { | |
7321 | aio_context_acquire(old_context); | |
7322 | } | |
7323 | bdrv_detach_aio_context(bs); | |
7324 | if (old_context != qemu_get_aio_context()) { | |
7325 | aio_context_release(old_context); | |
7326 | } | |
7327 | bdrv_attach_aio_context(bs, new_context); | |
7328 | } | |
7329 | ||
7330 | static TransactionActionDrv set_aio_context = { | |
7331 | .commit = bdrv_set_aio_context_commit, | |
7332 | .clean = bdrv_set_aio_context_clean, | |
7333 | }; | |
7334 | ||
7335 | /* | |
7336 | * Changes the AioContext used for fd handlers, timers, and BHs by this | |
7337 | * BlockDriverState and all its children and parents. | |
7338 | * | |
7339 | * Must be called from the main AioContext. | |
7340 | * | |
7341 | * The caller must own the AioContext lock for the old AioContext of bs, but it | |
7342 | * must not own the AioContext lock for new_context (unless new_context is the | |
7343 | * same as the current context of bs). | |
7344 | * | |
7345 | * @visited will accumulate all visited BdrvChild objects. The caller is | |
7346 | * responsible for freeing the list afterwards. | |
7347 | */ | |
7348 | static bool bdrv_change_aio_context(BlockDriverState *bs, AioContext *ctx, | |
e08cc001 | 7349 | GHashTable *visited, Transaction *tran, |
7e8c182f | 7350 | Error **errp) |
5d231849 KW |
7351 | { |
7352 | BdrvChild *c; | |
7e8c182f EGE |
7353 | BdrvStateSetAioContext *state; |
7354 | ||
7355 | GLOBAL_STATE_CODE(); | |
5d231849 KW |
7356 | |
7357 | if (bdrv_get_aio_context(bs) == ctx) { | |
7358 | return true; | |
7359 | } | |
7360 | ||
7361 | QLIST_FOREACH(c, &bs->parents, next_parent) { | |
7e8c182f | 7362 | if (!bdrv_parent_change_aio_context(c, ctx, visited, tran, errp)) { |
5d231849 KW |
7363 | return false; |
7364 | } | |
7365 | } | |
7e8c182f | 7366 | |
5d231849 | 7367 | QLIST_FOREACH(c, &bs->children, next) { |
7e8c182f | 7368 | if (!bdrv_child_change_aio_context(c, ctx, visited, tran, errp)) { |
5d231849 KW |
7369 | return false; |
7370 | } | |
7371 | } | |
7372 | ||
7e8c182f EGE |
7373 | state = g_new(BdrvStateSetAioContext, 1); |
7374 | *state = (BdrvStateSetAioContext) { | |
7375 | .new_ctx = ctx, | |
7376 | .bs = bs, | |
7377 | }; | |
7378 | ||
7379 | /* Paired with bdrv_drained_end in bdrv_set_aio_context_clean() */ | |
7380 | bdrv_drained_begin(bs); | |
7381 | ||
7382 | tran_add(tran, &set_aio_context, state); | |
7383 | ||
5d231849 KW |
7384 | return true; |
7385 | } | |
7386 | ||
7e8c182f EGE |
7387 | /* |
7388 | * Change bs's and recursively all of its parents' and children's AioContext | |
7389 | * to the given new context, returning an error if that isn't possible. | |
7390 | * | |
7391 | * If ignore_child is not NULL, that child (and its subgraph) will not | |
7392 | * be touched. | |
7393 | * | |
7394 | * This function still requires the caller to take the bs current | |
7395 | * AioContext lock, otherwise draining will fail since AIO_WAIT_WHILE | |
7396 | * assumes the lock is always held if bs is in another AioContext. | |
7397 | * For the same reason, it temporarily also holds the new AioContext, since | |
7398 | * bdrv_drained_end calls BDRV_POLL_WHILE that assumes the lock is taken too. | |
7399 | * Therefore the new AioContext lock must not be taken by the caller. | |
7400 | */ | |
a41cfda1 EGE |
7401 | int bdrv_try_change_aio_context(BlockDriverState *bs, AioContext *ctx, |
7402 | BdrvChild *ignore_child, Error **errp) | |
5d231849 | 7403 | { |
7e8c182f | 7404 | Transaction *tran; |
e08cc001 | 7405 | GHashTable *visited; |
7e8c182f EGE |
7406 | int ret; |
7407 | AioContext *old_context = bdrv_get_aio_context(bs); | |
f791bf7f EGE |
7408 | GLOBAL_STATE_CODE(); |
7409 | ||
7e8c182f EGE |
7410 | /* |
7411 | * Recursion phase: go through all nodes of the graph. | |
7412 | * Take care of checking that all nodes support changing AioContext | |
7413 | * and drain them, builing a linear list of callbacks to run if everything | |
7414 | * is successful (the transaction itself). | |
7415 | */ | |
7416 | tran = tran_new(); | |
e08cc001 EGE |
7417 | visited = g_hash_table_new(NULL, NULL); |
7418 | if (ignore_child) { | |
7419 | g_hash_table_add(visited, ignore_child); | |
7420 | } | |
7421 | ret = bdrv_change_aio_context(bs, ctx, visited, tran, errp); | |
7422 | g_hash_table_destroy(visited); | |
7e8c182f EGE |
7423 | |
7424 | /* | |
7425 | * Linear phase: go through all callbacks collected in the transaction. | |
7426 | * Run all callbacks collected in the recursion to switch all nodes | |
7427 | * AioContext lock (transaction commit), or undo all changes done in the | |
7428 | * recursion (transaction abort). | |
7429 | */ | |
5d231849 KW |
7430 | |
7431 | if (!ret) { | |
7e8c182f EGE |
7432 | /* Just run clean() callbacks. No AioContext changed. */ |
7433 | tran_abort(tran); | |
5d231849 KW |
7434 | return -EPERM; |
7435 | } | |
7436 | ||
7e8c182f EGE |
7437 | /* |
7438 | * Release old AioContext, it won't be needed anymore, as all | |
7439 | * bdrv_drained_begin() have been called already. | |
7440 | */ | |
7441 | if (qemu_get_aio_context() != old_context) { | |
7442 | aio_context_release(old_context); | |
7443 | } | |
7444 | ||
7445 | /* | |
7446 | * Acquire new AioContext since bdrv_drained_end() is going to be called | |
7447 | * after we switched all nodes in the new AioContext, and the function | |
7448 | * assumes that the lock of the bs is always taken. | |
7449 | */ | |
7450 | if (qemu_get_aio_context() != ctx) { | |
7451 | aio_context_acquire(ctx); | |
7452 | } | |
53a7d041 | 7453 | |
7e8c182f | 7454 | tran_commit(tran); |
5d231849 | 7455 | |
7e8c182f EGE |
7456 | if (qemu_get_aio_context() != ctx) { |
7457 | aio_context_release(ctx); | |
7458 | } | |
7459 | ||
7460 | /* Re-acquire the old AioContext, since the caller takes and releases it. */ | |
7461 | if (qemu_get_aio_context() != old_context) { | |
7462 | aio_context_acquire(old_context); | |
7463 | } | |
7464 | ||
7465 | return 0; | |
5d231849 KW |
7466 | } |
7467 | ||
33384421 HR |
7468 | void bdrv_add_aio_context_notifier(BlockDriverState *bs, |
7469 | void (*attached_aio_context)(AioContext *new_context, void *opaque), | |
7470 | void (*detach_aio_context)(void *opaque), void *opaque) | |
7471 | { | |
7472 | BdrvAioNotifier *ban = g_new(BdrvAioNotifier, 1); | |
7473 | *ban = (BdrvAioNotifier){ | |
7474 | .attached_aio_context = attached_aio_context, | |
7475 | .detach_aio_context = detach_aio_context, | |
7476 | .opaque = opaque | |
7477 | }; | |
f791bf7f | 7478 | GLOBAL_STATE_CODE(); |
33384421 HR |
7479 | |
7480 | QLIST_INSERT_HEAD(&bs->aio_notifiers, ban, list); | |
7481 | } | |
7482 | ||
7483 | void bdrv_remove_aio_context_notifier(BlockDriverState *bs, | |
7484 | void (*attached_aio_context)(AioContext *, | |
7485 | void *), | |
7486 | void (*detach_aio_context)(void *), | |
7487 | void *opaque) | |
7488 | { | |
7489 | BdrvAioNotifier *ban, *ban_next; | |
f791bf7f | 7490 | GLOBAL_STATE_CODE(); |
33384421 HR |
7491 | |
7492 | QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_next) { | |
7493 | if (ban->attached_aio_context == attached_aio_context && | |
7494 | ban->detach_aio_context == detach_aio_context && | |
e8a095da SH |
7495 | ban->opaque == opaque && |
7496 | ban->deleted == false) | |
33384421 | 7497 | { |
e8a095da SH |
7498 | if (bs->walking_aio_notifiers) { |
7499 | ban->deleted = true; | |
7500 | } else { | |
7501 | bdrv_do_remove_aio_context_notifier(ban); | |
7502 | } | |
33384421 HR |
7503 | return; |
7504 | } | |
7505 | } | |
7506 | ||
7507 | abort(); | |
7508 | } | |
7509 | ||
77485434 | 7510 | int bdrv_amend_options(BlockDriverState *bs, QemuOpts *opts, |
d1402b50 | 7511 | BlockDriverAmendStatusCB *status_cb, void *cb_opaque, |
a3579bfa | 7512 | bool force, |
d1402b50 | 7513 | Error **errp) |
6f176b48 | 7514 | { |
f791bf7f | 7515 | GLOBAL_STATE_CODE(); |
d470ad42 | 7516 | if (!bs->drv) { |
d1402b50 | 7517 | error_setg(errp, "Node is ejected"); |
d470ad42 HR |
7518 | return -ENOMEDIUM; |
7519 | } | |
c282e1fd | 7520 | if (!bs->drv->bdrv_amend_options) { |
d1402b50 HR |
7521 | error_setg(errp, "Block driver '%s' does not support option amendment", |
7522 | bs->drv->format_name); | |
6f176b48 HR |
7523 | return -ENOTSUP; |
7524 | } | |
a3579bfa ML |
7525 | return bs->drv->bdrv_amend_options(bs, opts, status_cb, |
7526 | cb_opaque, force, errp); | |
6f176b48 | 7527 | } |
f6186f49 | 7528 | |
5d69b5ab HR |
7529 | /* |
7530 | * This function checks whether the given @to_replace is allowed to be | |
7531 | * replaced by a node that always shows the same data as @bs. This is | |
7532 | * used for example to verify whether the mirror job can replace | |
7533 | * @to_replace by the target mirrored from @bs. | |
7534 | * To be replaceable, @bs and @to_replace may either be guaranteed to | |
7535 | * always show the same data (because they are only connected through | |
7536 | * filters), or some driver may allow replacing one of its children | |
7537 | * because it can guarantee that this child's data is not visible at | |
7538 | * all (for example, for dissenting quorum children that have no other | |
7539 | * parents). | |
7540 | */ | |
7541 | bool bdrv_recurse_can_replace(BlockDriverState *bs, | |
7542 | BlockDriverState *to_replace) | |
7543 | { | |
93393e69 HR |
7544 | BlockDriverState *filtered; |
7545 | ||
b4ad82aa EGE |
7546 | GLOBAL_STATE_CODE(); |
7547 | ||
5d69b5ab HR |
7548 | if (!bs || !bs->drv) { |
7549 | return false; | |
7550 | } | |
7551 | ||
7552 | if (bs == to_replace) { | |
7553 | return true; | |
7554 | } | |
7555 | ||
7556 | /* See what the driver can do */ | |
7557 | if (bs->drv->bdrv_recurse_can_replace) { | |
7558 | return bs->drv->bdrv_recurse_can_replace(bs, to_replace); | |
7559 | } | |
7560 | ||
7561 | /* For filters without an own implementation, we can recurse on our own */ | |
93393e69 HR |
7562 | filtered = bdrv_filter_bs(bs); |
7563 | if (filtered) { | |
7564 | return bdrv_recurse_can_replace(filtered, to_replace); | |
5d69b5ab HR |
7565 | } |
7566 | ||
7567 | /* Safe default */ | |
7568 | return false; | |
7569 | } | |
7570 | ||
810803a8 HR |
7571 | /* |
7572 | * Check whether the given @node_name can be replaced by a node that | |
7573 | * has the same data as @parent_bs. If so, return @node_name's BDS; | |
7574 | * NULL otherwise. | |
7575 | * | |
7576 | * @node_name must be a (recursive) *child of @parent_bs (or this | |
7577 | * function will return NULL). | |
7578 | * | |
7579 | * The result (whether the node can be replaced or not) is only valid | |
7580 | * for as long as no graph or permission changes occur. | |
7581 | */ | |
e12f3784 WC |
7582 | BlockDriverState *check_to_replace_node(BlockDriverState *parent_bs, |
7583 | const char *node_name, Error **errp) | |
09158f00 BC |
7584 | { |
7585 | BlockDriverState *to_replace_bs = bdrv_find_node(node_name); | |
5a7e7a0b SH |
7586 | AioContext *aio_context; |
7587 | ||
f791bf7f EGE |
7588 | GLOBAL_STATE_CODE(); |
7589 | ||
09158f00 | 7590 | if (!to_replace_bs) { |
785ec4b1 | 7591 | error_setg(errp, "Failed to find node with node-name='%s'", node_name); |
09158f00 BC |
7592 | return NULL; |
7593 | } | |
7594 | ||
5a7e7a0b SH |
7595 | aio_context = bdrv_get_aio_context(to_replace_bs); |
7596 | aio_context_acquire(aio_context); | |
7597 | ||
09158f00 | 7598 | if (bdrv_op_is_blocked(to_replace_bs, BLOCK_OP_TYPE_REPLACE, errp)) { |
5a7e7a0b SH |
7599 | to_replace_bs = NULL; |
7600 | goto out; | |
09158f00 BC |
7601 | } |
7602 | ||
7603 | /* We don't want arbitrary node of the BDS chain to be replaced only the top | |
7604 | * most non filter in order to prevent data corruption. | |
7605 | * Another benefit is that this tests exclude backing files which are | |
7606 | * blocked by the backing blockers. | |
7607 | */ | |
810803a8 HR |
7608 | if (!bdrv_recurse_can_replace(parent_bs, to_replace_bs)) { |
7609 | error_setg(errp, "Cannot replace '%s' by a node mirrored from '%s', " | |
7610 | "because it cannot be guaranteed that doing so would not " | |
7611 | "lead to an abrupt change of visible data", | |
7612 | node_name, parent_bs->node_name); | |
5a7e7a0b SH |
7613 | to_replace_bs = NULL; |
7614 | goto out; | |
09158f00 BC |
7615 | } |
7616 | ||
5a7e7a0b SH |
7617 | out: |
7618 | aio_context_release(aio_context); | |
09158f00 BC |
7619 | return to_replace_bs; |
7620 | } | |
448ad91d | 7621 | |
97e2f021 HR |
7622 | /** |
7623 | * Iterates through the list of runtime option keys that are said to | |
7624 | * be "strong" for a BDS. An option is called "strong" if it changes | |
7625 | * a BDS's data. For example, the null block driver's "size" and | |
7626 | * "read-zeroes" options are strong, but its "latency-ns" option is | |
7627 | * not. | |
7628 | * | |
7629 | * If a key returned by this function ends with a dot, all options | |
7630 | * starting with that prefix are strong. | |
7631 | */ | |
7632 | static const char *const *strong_options(BlockDriverState *bs, | |
7633 | const char *const *curopt) | |
7634 | { | |
7635 | static const char *const global_options[] = { | |
7636 | "driver", "filename", NULL | |
7637 | }; | |
7638 | ||
7639 | if (!curopt) { | |
7640 | return &global_options[0]; | |
7641 | } | |
7642 | ||
7643 | curopt++; | |
7644 | if (curopt == &global_options[ARRAY_SIZE(global_options) - 1] && bs->drv) { | |
7645 | curopt = bs->drv->strong_runtime_opts; | |
7646 | } | |
7647 | ||
7648 | return (curopt && *curopt) ? curopt : NULL; | |
7649 | } | |
7650 | ||
7651 | /** | |
7652 | * Copies all strong runtime options from bs->options to the given | |
7653 | * QDict. The set of strong option keys is determined by invoking | |
7654 | * strong_options(). | |
7655 | * | |
7656 | * Returns true iff any strong option was present in bs->options (and | |
7657 | * thus copied to the target QDict) with the exception of "filename" | |
7658 | * and "driver". The caller is expected to use this value to decide | |
7659 | * whether the existence of strong options prevents the generation of | |
7660 | * a plain filename. | |
7661 | */ | |
7662 | static bool append_strong_runtime_options(QDict *d, BlockDriverState *bs) | |
7663 | { | |
7664 | bool found_any = false; | |
7665 | const char *const *option_name = NULL; | |
7666 | ||
7667 | if (!bs->drv) { | |
7668 | return false; | |
7669 | } | |
7670 | ||
7671 | while ((option_name = strong_options(bs, option_name))) { | |
7672 | bool option_given = false; | |
7673 | ||
7674 | assert(strlen(*option_name) > 0); | |
7675 | if ((*option_name)[strlen(*option_name) - 1] != '.') { | |
7676 | QObject *entry = qdict_get(bs->options, *option_name); | |
7677 | if (!entry) { | |
7678 | continue; | |
7679 | } | |
7680 | ||
7681 | qdict_put_obj(d, *option_name, qobject_ref(entry)); | |
7682 | option_given = true; | |
7683 | } else { | |
7684 | const QDictEntry *entry; | |
7685 | for (entry = qdict_first(bs->options); entry; | |
7686 | entry = qdict_next(bs->options, entry)) | |
7687 | { | |
7688 | if (strstart(qdict_entry_key(entry), *option_name, NULL)) { | |
7689 | qdict_put_obj(d, qdict_entry_key(entry), | |
7690 | qobject_ref(qdict_entry_value(entry))); | |
7691 | option_given = true; | |
7692 | } | |
7693 | } | |
7694 | } | |
7695 | ||
7696 | /* While "driver" and "filename" need to be included in a JSON filename, | |
7697 | * their existence does not prohibit generation of a plain filename. */ | |
7698 | if (!found_any && option_given && | |
7699 | strcmp(*option_name, "driver") && strcmp(*option_name, "filename")) | |
7700 | { | |
7701 | found_any = true; | |
7702 | } | |
7703 | } | |
7704 | ||
62a01a27 HR |
7705 | if (!qdict_haskey(d, "driver")) { |
7706 | /* Drivers created with bdrv_new_open_driver() may not have a | |
7707 | * @driver option. Add it here. */ | |
7708 | qdict_put_str(d, "driver", bs->drv->format_name); | |
7709 | } | |
7710 | ||
97e2f021 HR |
7711 | return found_any; |
7712 | } | |
7713 | ||
90993623 HR |
7714 | /* Note: This function may return false positives; it may return true |
7715 | * even if opening the backing file specified by bs's image header | |
7716 | * would result in exactly bs->backing. */ | |
fa8fc1d0 | 7717 | static bool bdrv_backing_overridden(BlockDriverState *bs) |
90993623 | 7718 | { |
b4ad82aa | 7719 | GLOBAL_STATE_CODE(); |
90993623 HR |
7720 | if (bs->backing) { |
7721 | return strcmp(bs->auto_backing_file, | |
7722 | bs->backing->bs->filename); | |
7723 | } else { | |
7724 | /* No backing BDS, so if the image header reports any backing | |
7725 | * file, it must have been suppressed */ | |
7726 | return bs->auto_backing_file[0] != '\0'; | |
7727 | } | |
7728 | } | |
7729 | ||
91af7014 HR |
7730 | /* Updates the following BDS fields: |
7731 | * - exact_filename: A filename which may be used for opening a block device | |
7732 | * which (mostly) equals the given BDS (even without any | |
7733 | * other options; so reading and writing must return the same | |
7734 | * results, but caching etc. may be different) | |
7735 | * - full_open_options: Options which, when given when opening a block device | |
7736 | * (without a filename), result in a BDS (mostly) | |
7737 | * equalling the given one | |
7738 | * - filename: If exact_filename is set, it is copied here. Otherwise, | |
7739 | * full_open_options is converted to a JSON object, prefixed with | |
7740 | * "json:" (for use through the JSON pseudo protocol) and put here. | |
7741 | */ | |
7742 | void bdrv_refresh_filename(BlockDriverState *bs) | |
7743 | { | |
7744 | BlockDriver *drv = bs->drv; | |
e24518e3 | 7745 | BdrvChild *child; |
52f72d6f | 7746 | BlockDriverState *primary_child_bs; |
91af7014 | 7747 | QDict *opts; |
90993623 | 7748 | bool backing_overridden; |
998b3a1e HR |
7749 | bool generate_json_filename; /* Whether our default implementation should |
7750 | fill exact_filename (false) or not (true) */ | |
91af7014 | 7751 | |
f791bf7f EGE |
7752 | GLOBAL_STATE_CODE(); |
7753 | ||
91af7014 HR |
7754 | if (!drv) { |
7755 | return; | |
7756 | } | |
7757 | ||
e24518e3 HR |
7758 | /* This BDS's file name may depend on any of its children's file names, so |
7759 | * refresh those first */ | |
7760 | QLIST_FOREACH(child, &bs->children, next) { | |
7761 | bdrv_refresh_filename(child->bs); | |
91af7014 HR |
7762 | } |
7763 | ||
bb808d5f HR |
7764 | if (bs->implicit) { |
7765 | /* For implicit nodes, just copy everything from the single child */ | |
7766 | child = QLIST_FIRST(&bs->children); | |
7767 | assert(QLIST_NEXT(child, next) == NULL); | |
7768 | ||
7769 | pstrcpy(bs->exact_filename, sizeof(bs->exact_filename), | |
7770 | child->bs->exact_filename); | |
7771 | pstrcpy(bs->filename, sizeof(bs->filename), child->bs->filename); | |
7772 | ||
cb895614 | 7773 | qobject_unref(bs->full_open_options); |
bb808d5f HR |
7774 | bs->full_open_options = qobject_ref(child->bs->full_open_options); |
7775 | ||
7776 | return; | |
7777 | } | |
7778 | ||
90993623 HR |
7779 | backing_overridden = bdrv_backing_overridden(bs); |
7780 | ||
7781 | if (bs->open_flags & BDRV_O_NO_IO) { | |
7782 | /* Without I/O, the backing file does not change anything. | |
7783 | * Therefore, in such a case (primarily qemu-img), we can | |
7784 | * pretend the backing file has not been overridden even if | |
7785 | * it technically has been. */ | |
7786 | backing_overridden = false; | |
7787 | } | |
7788 | ||
97e2f021 HR |
7789 | /* Gather the options QDict */ |
7790 | opts = qdict_new(); | |
998b3a1e HR |
7791 | generate_json_filename = append_strong_runtime_options(opts, bs); |
7792 | generate_json_filename |= backing_overridden; | |
97e2f021 HR |
7793 | |
7794 | if (drv->bdrv_gather_child_options) { | |
7795 | /* Some block drivers may not want to present all of their children's | |
7796 | * options, or name them differently from BdrvChild.name */ | |
7797 | drv->bdrv_gather_child_options(bs, opts, backing_overridden); | |
7798 | } else { | |
7799 | QLIST_FOREACH(child, &bs->children, next) { | |
25191e5f | 7800 | if (child == bs->backing && !backing_overridden) { |
97e2f021 HR |
7801 | /* We can skip the backing BDS if it has not been overridden */ |
7802 | continue; | |
7803 | } | |
7804 | ||
7805 | qdict_put(opts, child->name, | |
7806 | qobject_ref(child->bs->full_open_options)); | |
7807 | } | |
7808 | ||
7809 | if (backing_overridden && !bs->backing) { | |
7810 | /* Force no backing file */ | |
7811 | qdict_put_null(opts, "backing"); | |
7812 | } | |
7813 | } | |
7814 | ||
7815 | qobject_unref(bs->full_open_options); | |
7816 | bs->full_open_options = opts; | |
7817 | ||
52f72d6f HR |
7818 | primary_child_bs = bdrv_primary_bs(bs); |
7819 | ||
998b3a1e HR |
7820 | if (drv->bdrv_refresh_filename) { |
7821 | /* Obsolete information is of no use here, so drop the old file name | |
7822 | * information before refreshing it */ | |
7823 | bs->exact_filename[0] = '\0'; | |
7824 | ||
7825 | drv->bdrv_refresh_filename(bs); | |
52f72d6f HR |
7826 | } else if (primary_child_bs) { |
7827 | /* | |
7828 | * Try to reconstruct valid information from the underlying | |
7829 | * file -- this only works for format nodes (filter nodes | |
7830 | * cannot be probed and as such must be selected by the user | |
7831 | * either through an options dict, or through a special | |
7832 | * filename which the filter driver must construct in its | |
7833 | * .bdrv_refresh_filename() implementation). | |
7834 | */ | |
998b3a1e HR |
7835 | |
7836 | bs->exact_filename[0] = '\0'; | |
7837 | ||
fb695c74 HR |
7838 | /* |
7839 | * We can use the underlying file's filename if: | |
7840 | * - it has a filename, | |
52f72d6f | 7841 | * - the current BDS is not a filter, |
fb695c74 HR |
7842 | * - the file is a protocol BDS, and |
7843 | * - opening that file (as this BDS's format) will automatically create | |
7844 | * the BDS tree we have right now, that is: | |
7845 | * - the user did not significantly change this BDS's behavior with | |
7846 | * some explicit (strong) options | |
7847 | * - no non-file child of this BDS has been overridden by the user | |
7848 | * Both of these conditions are represented by generate_json_filename. | |
7849 | */ | |
52f72d6f HR |
7850 | if (primary_child_bs->exact_filename[0] && |
7851 | primary_child_bs->drv->bdrv_file_open && | |
7852 | !drv->is_filter && !generate_json_filename) | |
fb695c74 | 7853 | { |
52f72d6f | 7854 | strcpy(bs->exact_filename, primary_child_bs->exact_filename); |
998b3a1e HR |
7855 | } |
7856 | } | |
7857 | ||
91af7014 HR |
7858 | if (bs->exact_filename[0]) { |
7859 | pstrcpy(bs->filename, sizeof(bs->filename), bs->exact_filename); | |
97e2f021 | 7860 | } else { |
eab3a467 | 7861 | GString *json = qobject_to_json(QOBJECT(bs->full_open_options)); |
5c86bdf1 | 7862 | if (snprintf(bs->filename, sizeof(bs->filename), "json:%s", |
eab3a467 | 7863 | json->str) >= sizeof(bs->filename)) { |
5c86bdf1 EB |
7864 | /* Give user a hint if we truncated things. */ |
7865 | strcpy(bs->filename + sizeof(bs->filename) - 4, "..."); | |
7866 | } | |
eab3a467 | 7867 | g_string_free(json, true); |
91af7014 HR |
7868 | } |
7869 | } | |
e06018ad | 7870 | |
1e89d0f9 HR |
7871 | char *bdrv_dirname(BlockDriverState *bs, Error **errp) |
7872 | { | |
7873 | BlockDriver *drv = bs->drv; | |
52f72d6f | 7874 | BlockDriverState *child_bs; |
1e89d0f9 | 7875 | |
f791bf7f EGE |
7876 | GLOBAL_STATE_CODE(); |
7877 | ||
1e89d0f9 HR |
7878 | if (!drv) { |
7879 | error_setg(errp, "Node '%s' is ejected", bs->node_name); | |
7880 | return NULL; | |
7881 | } | |
7882 | ||
7883 | if (drv->bdrv_dirname) { | |
7884 | return drv->bdrv_dirname(bs, errp); | |
7885 | } | |
7886 | ||
52f72d6f HR |
7887 | child_bs = bdrv_primary_bs(bs); |
7888 | if (child_bs) { | |
7889 | return bdrv_dirname(child_bs, errp); | |
1e89d0f9 HR |
7890 | } |
7891 | ||
7892 | bdrv_refresh_filename(bs); | |
7893 | if (bs->exact_filename[0] != '\0') { | |
7894 | return path_combine(bs->exact_filename, ""); | |
7895 | } | |
7896 | ||
7897 | error_setg(errp, "Cannot generate a base directory for %s nodes", | |
7898 | drv->format_name); | |
7899 | return NULL; | |
7900 | } | |
7901 | ||
e06018ad WC |
7902 | /* |
7903 | * Hot add/remove a BDS's child. So the user can take a child offline when | |
7904 | * it is broken and take a new child online | |
7905 | */ | |
7906 | void bdrv_add_child(BlockDriverState *parent_bs, BlockDriverState *child_bs, | |
7907 | Error **errp) | |
7908 | { | |
f791bf7f | 7909 | GLOBAL_STATE_CODE(); |
e06018ad WC |
7910 | if (!parent_bs->drv || !parent_bs->drv->bdrv_add_child) { |
7911 | error_setg(errp, "The node %s does not support adding a child", | |
7912 | bdrv_get_device_or_node_name(parent_bs)); | |
7913 | return; | |
7914 | } | |
7915 | ||
7916 | if (!QLIST_EMPTY(&child_bs->parents)) { | |
7917 | error_setg(errp, "The node %s already has a parent", | |
7918 | child_bs->node_name); | |
7919 | return; | |
7920 | } | |
7921 | ||
7922 | parent_bs->drv->bdrv_add_child(parent_bs, child_bs, errp); | |
7923 | } | |
7924 | ||
7925 | void bdrv_del_child(BlockDriverState *parent_bs, BdrvChild *child, Error **errp) | |
7926 | { | |
7927 | BdrvChild *tmp; | |
7928 | ||
f791bf7f | 7929 | GLOBAL_STATE_CODE(); |
e06018ad WC |
7930 | if (!parent_bs->drv || !parent_bs->drv->bdrv_del_child) { |
7931 | error_setg(errp, "The node %s does not support removing a child", | |
7932 | bdrv_get_device_or_node_name(parent_bs)); | |
7933 | return; | |
7934 | } | |
7935 | ||
7936 | QLIST_FOREACH(tmp, &parent_bs->children, next) { | |
7937 | if (tmp == child) { | |
7938 | break; | |
7939 | } | |
7940 | } | |
7941 | ||
7942 | if (!tmp) { | |
7943 | error_setg(errp, "The node %s does not have a child named %s", | |
7944 | bdrv_get_device_or_node_name(parent_bs), | |
7945 | bdrv_get_device_or_node_name(child->bs)); | |
7946 | return; | |
7947 | } | |
7948 | ||
7949 | parent_bs->drv->bdrv_del_child(parent_bs, child, errp); | |
7950 | } | |
6f7a3b53 HR |
7951 | |
7952 | int bdrv_make_empty(BdrvChild *c, Error **errp) | |
7953 | { | |
7954 | BlockDriver *drv = c->bs->drv; | |
7955 | int ret; | |
7956 | ||
f791bf7f | 7957 | GLOBAL_STATE_CODE(); |
6f7a3b53 HR |
7958 | assert(c->perm & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)); |
7959 | ||
7960 | if (!drv->bdrv_make_empty) { | |
7961 | error_setg(errp, "%s does not support emptying nodes", | |
7962 | drv->format_name); | |
7963 | return -ENOTSUP; | |
7964 | } | |
7965 | ||
7966 | ret = drv->bdrv_make_empty(c->bs); | |
7967 | if (ret < 0) { | |
7968 | error_setg_errno(errp, -ret, "Failed to empty %s", | |
7969 | c->bs->filename); | |
7970 | return ret; | |
7971 | } | |
7972 | ||
7973 | return 0; | |
7974 | } | |
9a6fc887 HR |
7975 | |
7976 | /* | |
7977 | * Return the child that @bs acts as an overlay for, and from which data may be | |
7978 | * copied in COW or COR operations. Usually this is the backing file. | |
7979 | */ | |
7980 | BdrvChild *bdrv_cow_child(BlockDriverState *bs) | |
7981 | { | |
967d7905 EGE |
7982 | IO_CODE(); |
7983 | ||
9a6fc887 HR |
7984 | if (!bs || !bs->drv) { |
7985 | return NULL; | |
7986 | } | |
7987 | ||
7988 | if (bs->drv->is_filter) { | |
7989 | return NULL; | |
7990 | } | |
7991 | ||
7992 | if (!bs->backing) { | |
7993 | return NULL; | |
7994 | } | |
7995 | ||
7996 | assert(bs->backing->role & BDRV_CHILD_COW); | |
7997 | return bs->backing; | |
7998 | } | |
7999 | ||
8000 | /* | |
8001 | * If @bs acts as a filter for exactly one of its children, return | |
8002 | * that child. | |
8003 | */ | |
8004 | BdrvChild *bdrv_filter_child(BlockDriverState *bs) | |
8005 | { | |
8006 | BdrvChild *c; | |
967d7905 | 8007 | IO_CODE(); |
9a6fc887 HR |
8008 | |
8009 | if (!bs || !bs->drv) { | |
8010 | return NULL; | |
8011 | } | |
8012 | ||
8013 | if (!bs->drv->is_filter) { | |
8014 | return NULL; | |
8015 | } | |
8016 | ||
8017 | /* Only one of @backing or @file may be used */ | |
8018 | assert(!(bs->backing && bs->file)); | |
8019 | ||
8020 | c = bs->backing ?: bs->file; | |
8021 | if (!c) { | |
8022 | return NULL; | |
8023 | } | |
8024 | ||
8025 | assert(c->role & BDRV_CHILD_FILTERED); | |
8026 | return c; | |
8027 | } | |
8028 | ||
8029 | /* | |
8030 | * Return either the result of bdrv_cow_child() or bdrv_filter_child(), | |
8031 | * whichever is non-NULL. | |
8032 | * | |
8033 | * Return NULL if both are NULL. | |
8034 | */ | |
8035 | BdrvChild *bdrv_filter_or_cow_child(BlockDriverState *bs) | |
8036 | { | |
8037 | BdrvChild *cow_child = bdrv_cow_child(bs); | |
8038 | BdrvChild *filter_child = bdrv_filter_child(bs); | |
967d7905 | 8039 | IO_CODE(); |
9a6fc887 HR |
8040 | |
8041 | /* Filter nodes cannot have COW backing files */ | |
8042 | assert(!(cow_child && filter_child)); | |
8043 | ||
8044 | return cow_child ?: filter_child; | |
8045 | } | |
8046 | ||
8047 | /* | |
8048 | * Return the primary child of this node: For filters, that is the | |
8049 | * filtered child. For other nodes, that is usually the child storing | |
8050 | * metadata. | |
8051 | * (A generally more helpful description is that this is (usually) the | |
8052 | * child that has the same filename as @bs.) | |
8053 | * | |
8054 | * Drivers do not necessarily have a primary child; for example quorum | |
8055 | * does not. | |
8056 | */ | |
8057 | BdrvChild *bdrv_primary_child(BlockDriverState *bs) | |
8058 | { | |
8059 | BdrvChild *c, *found = NULL; | |
967d7905 | 8060 | IO_CODE(); |
9a6fc887 HR |
8061 | |
8062 | QLIST_FOREACH(c, &bs->children, next) { | |
8063 | if (c->role & BDRV_CHILD_PRIMARY) { | |
8064 | assert(!found); | |
8065 | found = c; | |
8066 | } | |
8067 | } | |
8068 | ||
8069 | return found; | |
8070 | } | |
d38d7eb8 HR |
8071 | |
8072 | static BlockDriverState *bdrv_do_skip_filters(BlockDriverState *bs, | |
8073 | bool stop_on_explicit_filter) | |
8074 | { | |
8075 | BdrvChild *c; | |
8076 | ||
8077 | if (!bs) { | |
8078 | return NULL; | |
8079 | } | |
8080 | ||
8081 | while (!(stop_on_explicit_filter && !bs->implicit)) { | |
8082 | c = bdrv_filter_child(bs); | |
8083 | if (!c) { | |
8084 | /* | |
8085 | * A filter that is embedded in a working block graph must | |
8086 | * have a child. Assert this here so this function does | |
8087 | * not return a filter node that is not expected by the | |
8088 | * caller. | |
8089 | */ | |
8090 | assert(!bs->drv || !bs->drv->is_filter); | |
8091 | break; | |
8092 | } | |
8093 | bs = c->bs; | |
8094 | } | |
8095 | /* | |
8096 | * Note that this treats nodes with bs->drv == NULL as not being | |
8097 | * filters (bs->drv == NULL should be replaced by something else | |
8098 | * anyway). | |
8099 | * The advantage of this behavior is that this function will thus | |
8100 | * always return a non-NULL value (given a non-NULL @bs). | |
8101 | */ | |
8102 | ||
8103 | return bs; | |
8104 | } | |
8105 | ||
8106 | /* | |
8107 | * Return the first BDS that has not been added implicitly or that | |
8108 | * does not have a filtered child down the chain starting from @bs | |
8109 | * (including @bs itself). | |
8110 | */ | |
8111 | BlockDriverState *bdrv_skip_implicit_filters(BlockDriverState *bs) | |
8112 | { | |
b4ad82aa | 8113 | GLOBAL_STATE_CODE(); |
d38d7eb8 HR |
8114 | return bdrv_do_skip_filters(bs, true); |
8115 | } | |
8116 | ||
8117 | /* | |
8118 | * Return the first BDS that does not have a filtered child down the | |
8119 | * chain starting from @bs (including @bs itself). | |
8120 | */ | |
8121 | BlockDriverState *bdrv_skip_filters(BlockDriverState *bs) | |
8122 | { | |
967d7905 | 8123 | IO_CODE(); |
d38d7eb8 HR |
8124 | return bdrv_do_skip_filters(bs, false); |
8125 | } | |
8126 | ||
8127 | /* | |
8128 | * For a backing chain, return the first non-filter backing image of | |
8129 | * the first non-filter image. | |
8130 | */ | |
8131 | BlockDriverState *bdrv_backing_chain_next(BlockDriverState *bs) | |
8132 | { | |
967d7905 | 8133 | IO_CODE(); |
d38d7eb8 HR |
8134 | return bdrv_skip_filters(bdrv_cow_bs(bdrv_skip_filters(bs))); |
8135 | } | |
0bc329fb HR |
8136 | |
8137 | /** | |
8138 | * Check whether [offset, offset + bytes) overlaps with the cached | |
8139 | * block-status data region. | |
8140 | * | |
8141 | * If so, and @pnum is not NULL, set *pnum to `bsc.data_end - offset`, | |
8142 | * which is what bdrv_bsc_is_data()'s interface needs. | |
8143 | * Otherwise, *pnum is not touched. | |
8144 | */ | |
8145 | static bool bdrv_bsc_range_overlaps_locked(BlockDriverState *bs, | |
8146 | int64_t offset, int64_t bytes, | |
8147 | int64_t *pnum) | |
8148 | { | |
8149 | BdrvBlockStatusCache *bsc = qatomic_rcu_read(&bs->block_status_cache); | |
8150 | bool overlaps; | |
8151 | ||
8152 | overlaps = | |
8153 | qatomic_read(&bsc->valid) && | |
8154 | ranges_overlap(offset, bytes, bsc->data_start, | |
8155 | bsc->data_end - bsc->data_start); | |
8156 | ||
8157 | if (overlaps && pnum) { | |
8158 | *pnum = bsc->data_end - offset; | |
8159 | } | |
8160 | ||
8161 | return overlaps; | |
8162 | } | |
8163 | ||
8164 | /** | |
8165 | * See block_int.h for this function's documentation. | |
8166 | */ | |
8167 | bool bdrv_bsc_is_data(BlockDriverState *bs, int64_t offset, int64_t *pnum) | |
8168 | { | |
967d7905 | 8169 | IO_CODE(); |
0bc329fb | 8170 | RCU_READ_LOCK_GUARD(); |
0bc329fb HR |
8171 | return bdrv_bsc_range_overlaps_locked(bs, offset, 1, pnum); |
8172 | } | |
8173 | ||
8174 | /** | |
8175 | * See block_int.h for this function's documentation. | |
8176 | */ | |
8177 | void bdrv_bsc_invalidate_range(BlockDriverState *bs, | |
8178 | int64_t offset, int64_t bytes) | |
8179 | { | |
967d7905 | 8180 | IO_CODE(); |
0bc329fb HR |
8181 | RCU_READ_LOCK_GUARD(); |
8182 | ||
8183 | if (bdrv_bsc_range_overlaps_locked(bs, offset, bytes, NULL)) { | |
8184 | qatomic_set(&bs->block_status_cache->valid, false); | |
8185 | } | |
8186 | } | |
8187 | ||
8188 | /** | |
8189 | * See block_int.h for this function's documentation. | |
8190 | */ | |
8191 | void bdrv_bsc_fill(BlockDriverState *bs, int64_t offset, int64_t bytes) | |
8192 | { | |
8193 | BdrvBlockStatusCache *new_bsc = g_new(BdrvBlockStatusCache, 1); | |
8194 | BdrvBlockStatusCache *old_bsc; | |
967d7905 | 8195 | IO_CODE(); |
0bc329fb HR |
8196 | |
8197 | *new_bsc = (BdrvBlockStatusCache) { | |
8198 | .valid = true, | |
8199 | .data_start = offset, | |
8200 | .data_end = offset + bytes, | |
8201 | }; | |
8202 | ||
8203 | QEMU_LOCK_GUARD(&bs->bsc_modify_lock); | |
8204 | ||
8205 | old_bsc = qatomic_rcu_read(&bs->block_status_cache); | |
8206 | qatomic_rcu_set(&bs->block_status_cache, new_bsc); | |
8207 | if (old_bsc) { | |
8208 | g_free_rcu(old_bsc, rcu); | |
8209 | } | |
8210 | } |