]>
Commit | Line | Data |
---|---|---|
fc01f7e7 FB |
1 | /* |
2 | * QEMU System Emulator block driver | |
5fafdf24 | 3 | * |
fc01f7e7 | 4 | * Copyright (c) 2003 Fabrice Bellard |
5fafdf24 | 5 | * |
fc01f7e7 FB |
6 | * Permission is hereby granted, free of charge, to any person obtaining a copy |
7 | * of this software and associated documentation files (the "Software"), to deal | |
8 | * in the Software without restriction, including without limitation the rights | |
9 | * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell | |
10 | * copies of the Software, and to permit persons to whom the Software is | |
11 | * furnished to do so, subject to the following conditions: | |
12 | * | |
13 | * The above copyright notice and this permission notice shall be included in | |
14 | * all copies or substantial portions of the Software. | |
15 | * | |
16 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR | |
17 | * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, | |
18 | * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL | |
19 | * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER | |
20 | * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, | |
21 | * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN | |
22 | * THE SOFTWARE. | |
23 | */ | |
e688df6b | 24 | |
d38ea87a | 25 | #include "qemu/osdep.h" |
0ab8ed18 | 26 | #include "block/trace.h" |
737e150e PB |
27 | #include "block/block_int.h" |
28 | #include "block/blockjob.h" | |
cd7fca95 | 29 | #include "block/nbd.h" |
609f45ea | 30 | #include "block/qdict.h" |
d49b6836 | 31 | #include "qemu/error-report.h" |
88d88798 | 32 | #include "module_block.h" |
db725815 | 33 | #include "qemu/main-loop.h" |
1de7afc9 | 34 | #include "qemu/module.h" |
e688df6b | 35 | #include "qapi/error.h" |
452fcdbc | 36 | #include "qapi/qmp/qdict.h" |
7b1b5d19 | 37 | #include "qapi/qmp/qjson.h" |
e59a0cf1 | 38 | #include "qapi/qmp/qnull.h" |
fc81fa1e | 39 | #include "qapi/qmp/qstring.h" |
e1d74bc6 KW |
40 | #include "qapi/qobject-output-visitor.h" |
41 | #include "qapi/qapi-visit-block-core.h" | |
bfb197e0 | 42 | #include "sysemu/block-backend.h" |
9c17d615 | 43 | #include "sysemu/sysemu.h" |
1de7afc9 | 44 | #include "qemu/notify.h" |
922a01a0 | 45 | #include "qemu/option.h" |
10817bf0 | 46 | #include "qemu/coroutine.h" |
c13163fb | 47 | #include "block/qapi.h" |
1de7afc9 | 48 | #include "qemu/timer.h" |
f348b6d1 VB |
49 | #include "qemu/cutils.h" |
50 | #include "qemu/id.h" | |
fc01f7e7 | 51 | |
71e72a19 | 52 | #ifdef CONFIG_BSD |
7674e7bf | 53 | #include <sys/ioctl.h> |
72cf2d4f | 54 | #include <sys/queue.h> |
c5e97233 | 55 | #ifndef __DragonFly__ |
7674e7bf FB |
56 | #include <sys/disk.h> |
57 | #endif | |
c5e97233 | 58 | #endif |
7674e7bf | 59 | |
49dc768d AL |
60 | #ifdef _WIN32 |
61 | #include <windows.h> | |
62 | #endif | |
63 | ||
1c9805a3 SH |
64 | #define NOT_DONE 0x7fffffff /* used while emulated sync operation in progress */ |
65 | ||
dc364f4c BC |
66 | static QTAILQ_HEAD(, BlockDriverState) graph_bdrv_states = |
67 | QTAILQ_HEAD_INITIALIZER(graph_bdrv_states); | |
68 | ||
2c1d04e0 HR |
69 | static QTAILQ_HEAD(, BlockDriverState) all_bdrv_states = |
70 | QTAILQ_HEAD_INITIALIZER(all_bdrv_states); | |
71 | ||
8a22f02a SH |
72 | static QLIST_HEAD(, BlockDriver) bdrv_drivers = |
73 | QLIST_HEAD_INITIALIZER(bdrv_drivers); | |
ea2384d3 | 74 | |
5b363937 HR |
75 | static BlockDriverState *bdrv_open_inherit(const char *filename, |
76 | const char *reference, | |
77 | QDict *options, int flags, | |
78 | BlockDriverState *parent, | |
bd86fb99 | 79 | const BdrvChildClass *child_class, |
272c02ea | 80 | BdrvChildRole child_role, |
5b363937 | 81 | Error **errp); |
f3930ed0 | 82 | |
eb852011 MA |
83 | /* If non-zero, use only whitelisted block drivers */ |
84 | static int use_bdrv_whitelist; | |
85 | ||
9e0b22f4 SH |
86 | #ifdef _WIN32 |
87 | static int is_windows_drive_prefix(const char *filename) | |
88 | { | |
89 | return (((filename[0] >= 'a' && filename[0] <= 'z') || | |
90 | (filename[0] >= 'A' && filename[0] <= 'Z')) && | |
91 | filename[1] == ':'); | |
92 | } | |
93 | ||
94 | int is_windows_drive(const char *filename) | |
95 | { | |
96 | if (is_windows_drive_prefix(filename) && | |
97 | filename[2] == '\0') | |
98 | return 1; | |
99 | if (strstart(filename, "\\\\.\\", NULL) || | |
100 | strstart(filename, "//./", NULL)) | |
101 | return 1; | |
102 | return 0; | |
103 | } | |
104 | #endif | |
105 | ||
339064d5 KW |
106 | size_t bdrv_opt_mem_align(BlockDriverState *bs) |
107 | { | |
108 | if (!bs || !bs->drv) { | |
459b4e66 | 109 | /* page size or 4k (hdd sector size) should be on the safe side */ |
038adc2f | 110 | return MAX(4096, qemu_real_host_page_size); |
339064d5 KW |
111 | } |
112 | ||
113 | return bs->bl.opt_mem_alignment; | |
114 | } | |
115 | ||
4196d2f0 DL |
116 | size_t bdrv_min_mem_align(BlockDriverState *bs) |
117 | { | |
118 | if (!bs || !bs->drv) { | |
459b4e66 | 119 | /* page size or 4k (hdd sector size) should be on the safe side */ |
038adc2f | 120 | return MAX(4096, qemu_real_host_page_size); |
4196d2f0 DL |
121 | } |
122 | ||
123 | return bs->bl.min_mem_alignment; | |
124 | } | |
125 | ||
9e0b22f4 | 126 | /* check if the path starts with "<protocol>:" */ |
5c98415b | 127 | int path_has_protocol(const char *path) |
9e0b22f4 | 128 | { |
947995c0 PB |
129 | const char *p; |
130 | ||
9e0b22f4 SH |
131 | #ifdef _WIN32 |
132 | if (is_windows_drive(path) || | |
133 | is_windows_drive_prefix(path)) { | |
134 | return 0; | |
135 | } | |
947995c0 PB |
136 | p = path + strcspn(path, ":/\\"); |
137 | #else | |
138 | p = path + strcspn(path, ":/"); | |
9e0b22f4 SH |
139 | #endif |
140 | ||
947995c0 | 141 | return *p == ':'; |
9e0b22f4 SH |
142 | } |
143 | ||
83f64091 | 144 | int path_is_absolute(const char *path) |
3b0d4f61 | 145 | { |
21664424 FB |
146 | #ifdef _WIN32 |
147 | /* specific case for names like: "\\.\d:" */ | |
f53f4da9 | 148 | if (is_windows_drive(path) || is_windows_drive_prefix(path)) { |
21664424 | 149 | return 1; |
f53f4da9 PB |
150 | } |
151 | return (*path == '/' || *path == '\\'); | |
3b9f94e1 | 152 | #else |
f53f4da9 | 153 | return (*path == '/'); |
3b9f94e1 | 154 | #endif |
3b0d4f61 FB |
155 | } |
156 | ||
009b03aa | 157 | /* if filename is absolute, just return its duplicate. Otherwise, build a |
83f64091 FB |
158 | path to it by considering it is relative to base_path. URL are |
159 | supported. */ | |
009b03aa | 160 | char *path_combine(const char *base_path, const char *filename) |
3b0d4f61 | 161 | { |
009b03aa | 162 | const char *protocol_stripped = NULL; |
83f64091 | 163 | const char *p, *p1; |
009b03aa | 164 | char *result; |
83f64091 FB |
165 | int len; |
166 | ||
83f64091 | 167 | if (path_is_absolute(filename)) { |
009b03aa HR |
168 | return g_strdup(filename); |
169 | } | |
0d54a6fe | 170 | |
009b03aa HR |
171 | if (path_has_protocol(base_path)) { |
172 | protocol_stripped = strchr(base_path, ':'); | |
173 | if (protocol_stripped) { | |
174 | protocol_stripped++; | |
0d54a6fe | 175 | } |
009b03aa HR |
176 | } |
177 | p = protocol_stripped ?: base_path; | |
0d54a6fe | 178 | |
009b03aa | 179 | p1 = strrchr(base_path, '/'); |
3b9f94e1 | 180 | #ifdef _WIN32 |
009b03aa HR |
181 | { |
182 | const char *p2; | |
183 | p2 = strrchr(base_path, '\\'); | |
184 | if (!p1 || p2 > p1) { | |
185 | p1 = p2; | |
3b9f94e1 | 186 | } |
009b03aa | 187 | } |
3b9f94e1 | 188 | #endif |
009b03aa HR |
189 | if (p1) { |
190 | p1++; | |
191 | } else { | |
192 | p1 = base_path; | |
193 | } | |
194 | if (p1 > p) { | |
195 | p = p1; | |
3b0d4f61 | 196 | } |
009b03aa HR |
197 | len = p - base_path; |
198 | ||
199 | result = g_malloc(len + strlen(filename) + 1); | |
200 | memcpy(result, base_path, len); | |
201 | strcpy(result + len, filename); | |
202 | ||
203 | return result; | |
204 | } | |
205 | ||
03c320d8 HR |
206 | /* |
207 | * Helper function for bdrv_parse_filename() implementations to remove optional | |
208 | * protocol prefixes (especially "file:") from a filename and for putting the | |
209 | * stripped filename into the options QDict if there is such a prefix. | |
210 | */ | |
211 | void bdrv_parse_filename_strip_prefix(const char *filename, const char *prefix, | |
212 | QDict *options) | |
213 | { | |
214 | if (strstart(filename, prefix, &filename)) { | |
215 | /* Stripping the explicit protocol prefix may result in a protocol | |
216 | * prefix being (wrongly) detected (if the filename contains a colon) */ | |
217 | if (path_has_protocol(filename)) { | |
218 | QString *fat_filename; | |
219 | ||
220 | /* This means there is some colon before the first slash; therefore, | |
221 | * this cannot be an absolute path */ | |
222 | assert(!path_is_absolute(filename)); | |
223 | ||
224 | /* And we can thus fix the protocol detection issue by prefixing it | |
225 | * by "./" */ | |
226 | fat_filename = qstring_from_str("./"); | |
227 | qstring_append(fat_filename, filename); | |
228 | ||
229 | assert(!path_has_protocol(qstring_get_str(fat_filename))); | |
230 | ||
231 | qdict_put(options, "filename", fat_filename); | |
232 | } else { | |
233 | /* If no protocol prefix was detected, we can use the shortened | |
234 | * filename as-is */ | |
235 | qdict_put_str(options, "filename", filename); | |
236 | } | |
237 | } | |
238 | } | |
239 | ||
240 | ||
9c5e6594 KW |
241 | /* Returns whether the image file is opened as read-only. Note that this can |
242 | * return false and writing to the image file is still not possible because the | |
243 | * image is inactivated. */ | |
93ed524e JC |
244 | bool bdrv_is_read_only(BlockDriverState *bs) |
245 | { | |
246 | return bs->read_only; | |
247 | } | |
248 | ||
54a32bfe KW |
249 | int bdrv_can_set_read_only(BlockDriverState *bs, bool read_only, |
250 | bool ignore_allow_rdw, Error **errp) | |
fe5241bf | 251 | { |
e2b8247a JC |
252 | /* Do not set read_only if copy_on_read is enabled */ |
253 | if (bs->copy_on_read && read_only) { | |
254 | error_setg(errp, "Can't set node '%s' to r/o with copy-on-read enabled", | |
255 | bdrv_get_device_or_node_name(bs)); | |
256 | return -EINVAL; | |
257 | } | |
258 | ||
d6fcdf06 | 259 | /* Do not clear read_only if it is prohibited */ |
54a32bfe KW |
260 | if (!read_only && !(bs->open_flags & BDRV_O_ALLOW_RDWR) && |
261 | !ignore_allow_rdw) | |
262 | { | |
d6fcdf06 JC |
263 | error_setg(errp, "Node '%s' is read only", |
264 | bdrv_get_device_or_node_name(bs)); | |
265 | return -EPERM; | |
266 | } | |
267 | ||
45803a03 JC |
268 | return 0; |
269 | } | |
270 | ||
eaa2410f KW |
271 | /* |
272 | * Called by a driver that can only provide a read-only image. | |
273 | * | |
274 | * Returns 0 if the node is already read-only or it could switch the node to | |
275 | * read-only because BDRV_O_AUTO_RDONLY is set. | |
276 | * | |
277 | * Returns -EACCES if the node is read-write and BDRV_O_AUTO_RDONLY is not set | |
278 | * or bdrv_can_set_read_only() forbids making the node read-only. If @errmsg | |
279 | * is not NULL, it is used as the error message for the Error object. | |
280 | */ | |
281 | int bdrv_apply_auto_read_only(BlockDriverState *bs, const char *errmsg, | |
282 | Error **errp) | |
45803a03 JC |
283 | { |
284 | int ret = 0; | |
285 | ||
eaa2410f KW |
286 | if (!(bs->open_flags & BDRV_O_RDWR)) { |
287 | return 0; | |
288 | } | |
289 | if (!(bs->open_flags & BDRV_O_AUTO_RDONLY)) { | |
290 | goto fail; | |
45803a03 JC |
291 | } |
292 | ||
eaa2410f KW |
293 | ret = bdrv_can_set_read_only(bs, true, false, NULL); |
294 | if (ret < 0) { | |
295 | goto fail; | |
eeae6a59 KW |
296 | } |
297 | ||
eaa2410f KW |
298 | bs->read_only = true; |
299 | bs->open_flags &= ~BDRV_O_RDWR; | |
300 | ||
e2b8247a | 301 | return 0; |
eaa2410f KW |
302 | |
303 | fail: | |
304 | error_setg(errp, "%s", errmsg ?: "Image is read-only"); | |
305 | return -EACCES; | |
fe5241bf JC |
306 | } |
307 | ||
645ae7d8 HR |
308 | /* |
309 | * If @backing is empty, this function returns NULL without setting | |
310 | * @errp. In all other cases, NULL will only be returned with @errp | |
311 | * set. | |
312 | * | |
313 | * Therefore, a return value of NULL without @errp set means that | |
314 | * there is no backing file; if @errp is set, there is one but its | |
315 | * absolute filename cannot be generated. | |
316 | */ | |
317 | char *bdrv_get_full_backing_filename_from_filename(const char *backed, | |
318 | const char *backing, | |
319 | Error **errp) | |
dc5a1371 | 320 | { |
645ae7d8 HR |
321 | if (backing[0] == '\0') { |
322 | return NULL; | |
323 | } else if (path_has_protocol(backing) || path_is_absolute(backing)) { | |
324 | return g_strdup(backing); | |
9f07429e HR |
325 | } else if (backed[0] == '\0' || strstart(backed, "json:", NULL)) { |
326 | error_setg(errp, "Cannot use relative backing file names for '%s'", | |
327 | backed); | |
645ae7d8 | 328 | return NULL; |
dc5a1371 | 329 | } else { |
645ae7d8 | 330 | return path_combine(backed, backing); |
dc5a1371 PB |
331 | } |
332 | } | |
333 | ||
9f4793d8 HR |
334 | /* |
335 | * If @filename is empty or NULL, this function returns NULL without | |
336 | * setting @errp. In all other cases, NULL will only be returned with | |
337 | * @errp set. | |
338 | */ | |
339 | static char *bdrv_make_absolute_filename(BlockDriverState *relative_to, | |
340 | const char *filename, Error **errp) | |
0a82855a | 341 | { |
8df68616 | 342 | char *dir, *full_name; |
9f4793d8 | 343 | |
8df68616 HR |
344 | if (!filename || filename[0] == '\0') { |
345 | return NULL; | |
346 | } else if (path_has_protocol(filename) || path_is_absolute(filename)) { | |
347 | return g_strdup(filename); | |
348 | } | |
9f07429e | 349 | |
8df68616 HR |
350 | dir = bdrv_dirname(relative_to, errp); |
351 | if (!dir) { | |
352 | return NULL; | |
353 | } | |
f30c66ba | 354 | |
8df68616 HR |
355 | full_name = g_strconcat(dir, filename, NULL); |
356 | g_free(dir); | |
357 | return full_name; | |
9f4793d8 HR |
358 | } |
359 | ||
360 | char *bdrv_get_full_backing_filename(BlockDriverState *bs, Error **errp) | |
361 | { | |
362 | return bdrv_make_absolute_filename(bs, bs->backing_file, errp); | |
0a82855a HR |
363 | } |
364 | ||
0eb7217e SH |
365 | void bdrv_register(BlockDriver *bdrv) |
366 | { | |
a15f08dc | 367 | assert(bdrv->format_name); |
8a22f02a | 368 | QLIST_INSERT_HEAD(&bdrv_drivers, bdrv, list); |
ea2384d3 | 369 | } |
b338082b | 370 | |
e4e9986b MA |
371 | BlockDriverState *bdrv_new(void) |
372 | { | |
373 | BlockDriverState *bs; | |
374 | int i; | |
375 | ||
5839e53b | 376 | bs = g_new0(BlockDriverState, 1); |
e4654d2d | 377 | QLIST_INIT(&bs->dirty_bitmaps); |
fbe40ff7 FZ |
378 | for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) { |
379 | QLIST_INIT(&bs->op_blockers[i]); | |
380 | } | |
d616b224 | 381 | notifier_with_return_list_init(&bs->before_write_notifiers); |
3783fa3d | 382 | qemu_co_mutex_init(&bs->reqs_lock); |
2119882c | 383 | qemu_mutex_init(&bs->dirty_bitmap_mutex); |
9fcb0251 | 384 | bs->refcnt = 1; |
dcd04228 | 385 | bs->aio_context = qemu_get_aio_context(); |
d7d512f6 | 386 | |
3ff2f67a EY |
387 | qemu_co_queue_init(&bs->flush_queue); |
388 | ||
0f12264e KW |
389 | for (i = 0; i < bdrv_drain_all_count; i++) { |
390 | bdrv_drained_begin(bs); | |
391 | } | |
392 | ||
2c1d04e0 HR |
393 | QTAILQ_INSERT_TAIL(&all_bdrv_states, bs, bs_list); |
394 | ||
b338082b FB |
395 | return bs; |
396 | } | |
397 | ||
88d88798 | 398 | static BlockDriver *bdrv_do_find_format(const char *format_name) |
ea2384d3 FB |
399 | { |
400 | BlockDriver *drv1; | |
88d88798 | 401 | |
8a22f02a SH |
402 | QLIST_FOREACH(drv1, &bdrv_drivers, list) { |
403 | if (!strcmp(drv1->format_name, format_name)) { | |
ea2384d3 | 404 | return drv1; |
8a22f02a | 405 | } |
ea2384d3 | 406 | } |
88d88798 | 407 | |
ea2384d3 FB |
408 | return NULL; |
409 | } | |
410 | ||
88d88798 MM |
411 | BlockDriver *bdrv_find_format(const char *format_name) |
412 | { | |
413 | BlockDriver *drv1; | |
414 | int i; | |
415 | ||
416 | drv1 = bdrv_do_find_format(format_name); | |
417 | if (drv1) { | |
418 | return drv1; | |
419 | } | |
420 | ||
421 | /* The driver isn't registered, maybe we need to load a module */ | |
422 | for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); ++i) { | |
423 | if (!strcmp(block_driver_modules[i].format_name, format_name)) { | |
424 | block_module_load_one(block_driver_modules[i].library_name); | |
425 | break; | |
426 | } | |
427 | } | |
428 | ||
429 | return bdrv_do_find_format(format_name); | |
430 | } | |
431 | ||
9ac404c5 | 432 | static int bdrv_format_is_whitelisted(const char *format_name, bool read_only) |
eb852011 | 433 | { |
b64ec4e4 FZ |
434 | static const char *whitelist_rw[] = { |
435 | CONFIG_BDRV_RW_WHITELIST | |
436 | }; | |
437 | static const char *whitelist_ro[] = { | |
438 | CONFIG_BDRV_RO_WHITELIST | |
eb852011 MA |
439 | }; |
440 | const char **p; | |
441 | ||
b64ec4e4 | 442 | if (!whitelist_rw[0] && !whitelist_ro[0]) { |
eb852011 | 443 | return 1; /* no whitelist, anything goes */ |
b64ec4e4 | 444 | } |
eb852011 | 445 | |
b64ec4e4 | 446 | for (p = whitelist_rw; *p; p++) { |
9ac404c5 | 447 | if (!strcmp(format_name, *p)) { |
eb852011 MA |
448 | return 1; |
449 | } | |
450 | } | |
b64ec4e4 FZ |
451 | if (read_only) { |
452 | for (p = whitelist_ro; *p; p++) { | |
9ac404c5 | 453 | if (!strcmp(format_name, *p)) { |
b64ec4e4 FZ |
454 | return 1; |
455 | } | |
456 | } | |
457 | } | |
eb852011 MA |
458 | return 0; |
459 | } | |
460 | ||
9ac404c5 AS |
461 | int bdrv_is_whitelisted(BlockDriver *drv, bool read_only) |
462 | { | |
463 | return bdrv_format_is_whitelisted(drv->format_name, read_only); | |
464 | } | |
465 | ||
e6ff69bf DB |
466 | bool bdrv_uses_whitelist(void) |
467 | { | |
468 | return use_bdrv_whitelist; | |
469 | } | |
470 | ||
5b7e1542 ZYW |
471 | typedef struct CreateCo { |
472 | BlockDriver *drv; | |
473 | char *filename; | |
83d0521a | 474 | QemuOpts *opts; |
5b7e1542 | 475 | int ret; |
cc84d90f | 476 | Error *err; |
5b7e1542 ZYW |
477 | } CreateCo; |
478 | ||
479 | static void coroutine_fn bdrv_create_co_entry(void *opaque) | |
480 | { | |
cc84d90f HR |
481 | Error *local_err = NULL; |
482 | int ret; | |
483 | ||
5b7e1542 ZYW |
484 | CreateCo *cco = opaque; |
485 | assert(cco->drv); | |
486 | ||
b92902df ML |
487 | ret = cco->drv->bdrv_co_create_opts(cco->drv, |
488 | cco->filename, cco->opts, &local_err); | |
621ff94d | 489 | error_propagate(&cco->err, local_err); |
cc84d90f | 490 | cco->ret = ret; |
5b7e1542 ZYW |
491 | } |
492 | ||
0e7e1989 | 493 | int bdrv_create(BlockDriver *drv, const char* filename, |
83d0521a | 494 | QemuOpts *opts, Error **errp) |
ea2384d3 | 495 | { |
5b7e1542 ZYW |
496 | int ret; |
497 | ||
498 | Coroutine *co; | |
499 | CreateCo cco = { | |
500 | .drv = drv, | |
501 | .filename = g_strdup(filename), | |
83d0521a | 502 | .opts = opts, |
5b7e1542 | 503 | .ret = NOT_DONE, |
cc84d90f | 504 | .err = NULL, |
5b7e1542 ZYW |
505 | }; |
506 | ||
efc75e2a | 507 | if (!drv->bdrv_co_create_opts) { |
cc84d90f | 508 | error_setg(errp, "Driver '%s' does not support image creation", drv->format_name); |
80168bff LC |
509 | ret = -ENOTSUP; |
510 | goto out; | |
5b7e1542 ZYW |
511 | } |
512 | ||
513 | if (qemu_in_coroutine()) { | |
514 | /* Fast-path if already in coroutine context */ | |
515 | bdrv_create_co_entry(&cco); | |
516 | } else { | |
0b8b8753 PB |
517 | co = qemu_coroutine_create(bdrv_create_co_entry, &cco); |
518 | qemu_coroutine_enter(co); | |
5b7e1542 | 519 | while (cco.ret == NOT_DONE) { |
b47ec2c4 | 520 | aio_poll(qemu_get_aio_context(), true); |
5b7e1542 ZYW |
521 | } |
522 | } | |
523 | ||
524 | ret = cco.ret; | |
cc84d90f | 525 | if (ret < 0) { |
84d18f06 | 526 | if (cco.err) { |
cc84d90f HR |
527 | error_propagate(errp, cco.err); |
528 | } else { | |
529 | error_setg_errno(errp, -ret, "Could not create image"); | |
530 | } | |
531 | } | |
0e7e1989 | 532 | |
80168bff LC |
533 | out: |
534 | g_free(cco.filename); | |
5b7e1542 | 535 | return ret; |
ea2384d3 FB |
536 | } |
537 | ||
fd17146c HR |
538 | /** |
539 | * Helper function for bdrv_create_file_fallback(): Resize @blk to at | |
540 | * least the given @minimum_size. | |
541 | * | |
542 | * On success, return @blk's actual length. | |
543 | * Otherwise, return -errno. | |
544 | */ | |
545 | static int64_t create_file_fallback_truncate(BlockBackend *blk, | |
546 | int64_t minimum_size, Error **errp) | |
84a12e66 | 547 | { |
cc84d90f | 548 | Error *local_err = NULL; |
fd17146c | 549 | int64_t size; |
cc84d90f | 550 | int ret; |
84a12e66 | 551 | |
8c6242b6 KW |
552 | ret = blk_truncate(blk, minimum_size, false, PREALLOC_MODE_OFF, 0, |
553 | &local_err); | |
fd17146c HR |
554 | if (ret < 0 && ret != -ENOTSUP) { |
555 | error_propagate(errp, local_err); | |
556 | return ret; | |
557 | } | |
558 | ||
559 | size = blk_getlength(blk); | |
560 | if (size < 0) { | |
561 | error_free(local_err); | |
562 | error_setg_errno(errp, -size, | |
563 | "Failed to inquire the new image file's length"); | |
564 | return size; | |
565 | } | |
566 | ||
567 | if (size < minimum_size) { | |
568 | /* Need to grow the image, but we failed to do that */ | |
569 | error_propagate(errp, local_err); | |
570 | return -ENOTSUP; | |
571 | } | |
572 | ||
573 | error_free(local_err); | |
574 | local_err = NULL; | |
575 | ||
576 | return size; | |
577 | } | |
578 | ||
579 | /** | |
580 | * Helper function for bdrv_create_file_fallback(): Zero the first | |
581 | * sector to remove any potentially pre-existing image header. | |
582 | */ | |
583 | static int create_file_fallback_zero_first_sector(BlockBackend *blk, | |
584 | int64_t current_size, | |
585 | Error **errp) | |
586 | { | |
587 | int64_t bytes_to_clear; | |
588 | int ret; | |
589 | ||
590 | bytes_to_clear = MIN(current_size, BDRV_SECTOR_SIZE); | |
591 | if (bytes_to_clear) { | |
592 | ret = blk_pwrite_zeroes(blk, 0, bytes_to_clear, BDRV_REQ_MAY_UNMAP); | |
593 | if (ret < 0) { | |
594 | error_setg_errno(errp, -ret, | |
595 | "Failed to clear the new image's first sector"); | |
596 | return ret; | |
597 | } | |
598 | } | |
599 | ||
600 | return 0; | |
601 | } | |
602 | ||
5a5e7f8c ML |
603 | /** |
604 | * Simple implementation of bdrv_co_create_opts for protocol drivers | |
605 | * which only support creation via opening a file | |
606 | * (usually existing raw storage device) | |
607 | */ | |
608 | int coroutine_fn bdrv_co_create_opts_simple(BlockDriver *drv, | |
609 | const char *filename, | |
610 | QemuOpts *opts, | |
611 | Error **errp) | |
fd17146c HR |
612 | { |
613 | BlockBackend *blk; | |
eeea1faa | 614 | QDict *options; |
fd17146c HR |
615 | int64_t size = 0; |
616 | char *buf = NULL; | |
617 | PreallocMode prealloc; | |
618 | Error *local_err = NULL; | |
619 | int ret; | |
620 | ||
621 | size = qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0); | |
622 | buf = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC); | |
623 | prealloc = qapi_enum_parse(&PreallocMode_lookup, buf, | |
624 | PREALLOC_MODE_OFF, &local_err); | |
625 | g_free(buf); | |
626 | if (local_err) { | |
627 | error_propagate(errp, local_err); | |
628 | return -EINVAL; | |
629 | } | |
630 | ||
631 | if (prealloc != PREALLOC_MODE_OFF) { | |
632 | error_setg(errp, "Unsupported preallocation mode '%s'", | |
633 | PreallocMode_str(prealloc)); | |
634 | return -ENOTSUP; | |
635 | } | |
636 | ||
eeea1faa | 637 | options = qdict_new(); |
fd17146c HR |
638 | qdict_put_str(options, "driver", drv->format_name); |
639 | ||
640 | blk = blk_new_open(filename, NULL, options, | |
641 | BDRV_O_RDWR | BDRV_O_RESIZE, errp); | |
642 | if (!blk) { | |
643 | error_prepend(errp, "Protocol driver '%s' does not support image " | |
644 | "creation, and opening the image failed: ", | |
645 | drv->format_name); | |
646 | return -EINVAL; | |
647 | } | |
648 | ||
649 | size = create_file_fallback_truncate(blk, size, errp); | |
650 | if (size < 0) { | |
651 | ret = size; | |
652 | goto out; | |
653 | } | |
654 | ||
655 | ret = create_file_fallback_zero_first_sector(blk, size, errp); | |
656 | if (ret < 0) { | |
657 | goto out; | |
658 | } | |
659 | ||
660 | ret = 0; | |
661 | out: | |
662 | blk_unref(blk); | |
663 | return ret; | |
664 | } | |
665 | ||
666 | int bdrv_create_file(const char *filename, QemuOpts *opts, Error **errp) | |
667 | { | |
668 | BlockDriver *drv; | |
669 | ||
b65a5e12 | 670 | drv = bdrv_find_protocol(filename, true, errp); |
84a12e66 | 671 | if (drv == NULL) { |
16905d71 | 672 | return -ENOENT; |
84a12e66 CH |
673 | } |
674 | ||
5a5e7f8c | 675 | return bdrv_create(drv, filename, opts, errp); |
84a12e66 CH |
676 | } |
677 | ||
e1d7f8bb DHB |
678 | int coroutine_fn bdrv_co_delete_file(BlockDriverState *bs, Error **errp) |
679 | { | |
680 | Error *local_err = NULL; | |
681 | int ret; | |
682 | ||
683 | assert(bs != NULL); | |
684 | ||
685 | if (!bs->drv) { | |
686 | error_setg(errp, "Block node '%s' is not opened", bs->filename); | |
687 | return -ENOMEDIUM; | |
688 | } | |
689 | ||
690 | if (!bs->drv->bdrv_co_delete_file) { | |
691 | error_setg(errp, "Driver '%s' does not support image deletion", | |
692 | bs->drv->format_name); | |
693 | return -ENOTSUP; | |
694 | } | |
695 | ||
696 | ret = bs->drv->bdrv_co_delete_file(bs, &local_err); | |
697 | if (ret < 0) { | |
698 | error_propagate(errp, local_err); | |
699 | } | |
700 | ||
701 | return ret; | |
702 | } | |
703 | ||
892b7de8 ET |
704 | /** |
705 | * Try to get @bs's logical and physical block size. | |
706 | * On success, store them in @bsz struct and return 0. | |
707 | * On failure return -errno. | |
708 | * @bs must not be empty. | |
709 | */ | |
710 | int bdrv_probe_blocksizes(BlockDriverState *bs, BlockSizes *bsz) | |
711 | { | |
712 | BlockDriver *drv = bs->drv; | |
713 | ||
714 | if (drv && drv->bdrv_probe_blocksizes) { | |
715 | return drv->bdrv_probe_blocksizes(bs, bsz); | |
5a612c00 MP |
716 | } else if (drv && drv->is_filter && bs->file) { |
717 | return bdrv_probe_blocksizes(bs->file->bs, bsz); | |
892b7de8 ET |
718 | } |
719 | ||
720 | return -ENOTSUP; | |
721 | } | |
722 | ||
723 | /** | |
724 | * Try to get @bs's geometry (cyls, heads, sectors). | |
725 | * On success, store them in @geo struct and return 0. | |
726 | * On failure return -errno. | |
727 | * @bs must not be empty. | |
728 | */ | |
729 | int bdrv_probe_geometry(BlockDriverState *bs, HDGeometry *geo) | |
730 | { | |
731 | BlockDriver *drv = bs->drv; | |
732 | ||
733 | if (drv && drv->bdrv_probe_geometry) { | |
734 | return drv->bdrv_probe_geometry(bs, geo); | |
5a612c00 MP |
735 | } else if (drv && drv->is_filter && bs->file) { |
736 | return bdrv_probe_geometry(bs->file->bs, geo); | |
892b7de8 ET |
737 | } |
738 | ||
739 | return -ENOTSUP; | |
740 | } | |
741 | ||
eba25057 JM |
742 | /* |
743 | * Create a uniquely-named empty temporary file. | |
744 | * Return 0 upon success, otherwise a negative errno value. | |
745 | */ | |
746 | int get_tmp_filename(char *filename, int size) | |
d5249393 | 747 | { |
eba25057 | 748 | #ifdef _WIN32 |
3b9f94e1 | 749 | char temp_dir[MAX_PATH]; |
eba25057 JM |
750 | /* GetTempFileName requires that its output buffer (4th param) |
751 | have length MAX_PATH or greater. */ | |
752 | assert(size >= MAX_PATH); | |
753 | return (GetTempPath(MAX_PATH, temp_dir) | |
754 | && GetTempFileName(temp_dir, "qem", 0, filename) | |
755 | ? 0 : -GetLastError()); | |
d5249393 | 756 | #else |
67b915a5 | 757 | int fd; |
7ccfb2eb | 758 | const char *tmpdir; |
0badc1ee | 759 | tmpdir = getenv("TMPDIR"); |
69bef793 AS |
760 | if (!tmpdir) { |
761 | tmpdir = "/var/tmp"; | |
762 | } | |
eba25057 JM |
763 | if (snprintf(filename, size, "%s/vl.XXXXXX", tmpdir) >= size) { |
764 | return -EOVERFLOW; | |
765 | } | |
ea2384d3 | 766 | fd = mkstemp(filename); |
fe235a06 DH |
767 | if (fd < 0) { |
768 | return -errno; | |
769 | } | |
770 | if (close(fd) != 0) { | |
771 | unlink(filename); | |
eba25057 JM |
772 | return -errno; |
773 | } | |
774 | return 0; | |
d5249393 | 775 | #endif |
eba25057 | 776 | } |
fc01f7e7 | 777 | |
84a12e66 CH |
778 | /* |
779 | * Detect host devices. By convention, /dev/cdrom[N] is always | |
780 | * recognized as a host CDROM. | |
781 | */ | |
782 | static BlockDriver *find_hdev_driver(const char *filename) | |
783 | { | |
784 | int score_max = 0, score; | |
785 | BlockDriver *drv = NULL, *d; | |
786 | ||
787 | QLIST_FOREACH(d, &bdrv_drivers, list) { | |
788 | if (d->bdrv_probe_device) { | |
789 | score = d->bdrv_probe_device(filename); | |
790 | if (score > score_max) { | |
791 | score_max = score; | |
792 | drv = d; | |
793 | } | |
794 | } | |
795 | } | |
796 | ||
797 | return drv; | |
798 | } | |
799 | ||
88d88798 MM |
800 | static BlockDriver *bdrv_do_find_protocol(const char *protocol) |
801 | { | |
802 | BlockDriver *drv1; | |
803 | ||
804 | QLIST_FOREACH(drv1, &bdrv_drivers, list) { | |
805 | if (drv1->protocol_name && !strcmp(drv1->protocol_name, protocol)) { | |
806 | return drv1; | |
807 | } | |
808 | } | |
809 | ||
810 | return NULL; | |
811 | } | |
812 | ||
98289620 | 813 | BlockDriver *bdrv_find_protocol(const char *filename, |
b65a5e12 HR |
814 | bool allow_protocol_prefix, |
815 | Error **errp) | |
83f64091 FB |
816 | { |
817 | BlockDriver *drv1; | |
818 | char protocol[128]; | |
1cec71e3 | 819 | int len; |
83f64091 | 820 | const char *p; |
88d88798 | 821 | int i; |
19cb3738 | 822 | |
66f82cee KW |
823 | /* TODO Drivers without bdrv_file_open must be specified explicitly */ |
824 | ||
39508e7a CH |
825 | /* |
826 | * XXX(hch): we really should not let host device detection | |
827 | * override an explicit protocol specification, but moving this | |
828 | * later breaks access to device names with colons in them. | |
829 | * Thanks to the brain-dead persistent naming schemes on udev- | |
830 | * based Linux systems those actually are quite common. | |
831 | */ | |
832 | drv1 = find_hdev_driver(filename); | |
833 | if (drv1) { | |
834 | return drv1; | |
835 | } | |
836 | ||
98289620 | 837 | if (!path_has_protocol(filename) || !allow_protocol_prefix) { |
ef810437 | 838 | return &bdrv_file; |
84a12e66 | 839 | } |
98289620 | 840 | |
9e0b22f4 SH |
841 | p = strchr(filename, ':'); |
842 | assert(p != NULL); | |
1cec71e3 AL |
843 | len = p - filename; |
844 | if (len > sizeof(protocol) - 1) | |
845 | len = sizeof(protocol) - 1; | |
846 | memcpy(protocol, filename, len); | |
847 | protocol[len] = '\0'; | |
88d88798 MM |
848 | |
849 | drv1 = bdrv_do_find_protocol(protocol); | |
850 | if (drv1) { | |
851 | return drv1; | |
852 | } | |
853 | ||
854 | for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); ++i) { | |
855 | if (block_driver_modules[i].protocol_name && | |
856 | !strcmp(block_driver_modules[i].protocol_name, protocol)) { | |
857 | block_module_load_one(block_driver_modules[i].library_name); | |
858 | break; | |
8a22f02a | 859 | } |
83f64091 | 860 | } |
b65a5e12 | 861 | |
88d88798 MM |
862 | drv1 = bdrv_do_find_protocol(protocol); |
863 | if (!drv1) { | |
864 | error_setg(errp, "Unknown protocol '%s'", protocol); | |
865 | } | |
866 | return drv1; | |
83f64091 FB |
867 | } |
868 | ||
c6684249 MA |
869 | /* |
870 | * Guess image format by probing its contents. | |
871 | * This is not a good idea when your image is raw (CVE-2008-2004), but | |
872 | * we do it anyway for backward compatibility. | |
873 | * | |
874 | * @buf contains the image's first @buf_size bytes. | |
7cddd372 KW |
875 | * @buf_size is the buffer size in bytes (generally BLOCK_PROBE_BUF_SIZE, |
876 | * but can be smaller if the image file is smaller) | |
c6684249 MA |
877 | * @filename is its filename. |
878 | * | |
879 | * For all block drivers, call the bdrv_probe() method to get its | |
880 | * probing score. | |
881 | * Return the first block driver with the highest probing score. | |
882 | */ | |
38f3ef57 KW |
883 | BlockDriver *bdrv_probe_all(const uint8_t *buf, int buf_size, |
884 | const char *filename) | |
c6684249 MA |
885 | { |
886 | int score_max = 0, score; | |
887 | BlockDriver *drv = NULL, *d; | |
888 | ||
889 | QLIST_FOREACH(d, &bdrv_drivers, list) { | |
890 | if (d->bdrv_probe) { | |
891 | score = d->bdrv_probe(buf, buf_size, filename); | |
892 | if (score > score_max) { | |
893 | score_max = score; | |
894 | drv = d; | |
895 | } | |
896 | } | |
897 | } | |
898 | ||
899 | return drv; | |
900 | } | |
901 | ||
5696c6e3 | 902 | static int find_image_format(BlockBackend *file, const char *filename, |
34b5d2c6 | 903 | BlockDriver **pdrv, Error **errp) |
f3a5d3f8 | 904 | { |
c6684249 | 905 | BlockDriver *drv; |
7cddd372 | 906 | uint8_t buf[BLOCK_PROBE_BUF_SIZE]; |
f500a6d3 | 907 | int ret = 0; |
f8ea0b00 | 908 | |
08a00559 | 909 | /* Return the raw BlockDriver * to scsi-generic devices or empty drives */ |
5696c6e3 | 910 | if (blk_is_sg(file) || !blk_is_inserted(file) || blk_getlength(file) == 0) { |
ef810437 | 911 | *pdrv = &bdrv_raw; |
c98ac35d | 912 | return ret; |
1a396859 | 913 | } |
f8ea0b00 | 914 | |
5696c6e3 | 915 | ret = blk_pread(file, 0, buf, sizeof(buf)); |
83f64091 | 916 | if (ret < 0) { |
34b5d2c6 HR |
917 | error_setg_errno(errp, -ret, "Could not read image for determining its " |
918 | "format"); | |
c98ac35d SW |
919 | *pdrv = NULL; |
920 | return ret; | |
83f64091 FB |
921 | } |
922 | ||
c6684249 | 923 | drv = bdrv_probe_all(buf, ret, filename); |
c98ac35d | 924 | if (!drv) { |
34b5d2c6 HR |
925 | error_setg(errp, "Could not determine image format: No compatible " |
926 | "driver found"); | |
c98ac35d SW |
927 | ret = -ENOENT; |
928 | } | |
929 | *pdrv = drv; | |
930 | return ret; | |
ea2384d3 FB |
931 | } |
932 | ||
51762288 SH |
933 | /** |
934 | * Set the current 'total_sectors' value | |
65a9bb25 | 935 | * Return 0 on success, -errno on error. |
51762288 | 936 | */ |
3d9f2d2a | 937 | int refresh_total_sectors(BlockDriverState *bs, int64_t hint) |
51762288 SH |
938 | { |
939 | BlockDriver *drv = bs->drv; | |
940 | ||
d470ad42 HR |
941 | if (!drv) { |
942 | return -ENOMEDIUM; | |
943 | } | |
944 | ||
396759ad | 945 | /* Do not attempt drv->bdrv_getlength() on scsi-generic devices */ |
b192af8a | 946 | if (bdrv_is_sg(bs)) |
396759ad NB |
947 | return 0; |
948 | ||
51762288 SH |
949 | /* query actual device if possible, otherwise just trust the hint */ |
950 | if (drv->bdrv_getlength) { | |
951 | int64_t length = drv->bdrv_getlength(bs); | |
952 | if (length < 0) { | |
953 | return length; | |
954 | } | |
7e382003 | 955 | hint = DIV_ROUND_UP(length, BDRV_SECTOR_SIZE); |
51762288 SH |
956 | } |
957 | ||
958 | bs->total_sectors = hint; | |
959 | return 0; | |
960 | } | |
961 | ||
cddff5ba KW |
962 | /** |
963 | * Combines a QDict of new block driver @options with any missing options taken | |
964 | * from @old_options, so that leaving out an option defaults to its old value. | |
965 | */ | |
966 | static void bdrv_join_options(BlockDriverState *bs, QDict *options, | |
967 | QDict *old_options) | |
968 | { | |
969 | if (bs->drv && bs->drv->bdrv_join_options) { | |
970 | bs->drv->bdrv_join_options(options, old_options); | |
971 | } else { | |
972 | qdict_join(options, old_options, false); | |
973 | } | |
974 | } | |
975 | ||
543770bd AG |
976 | static BlockdevDetectZeroesOptions bdrv_parse_detect_zeroes(QemuOpts *opts, |
977 | int open_flags, | |
978 | Error **errp) | |
979 | { | |
980 | Error *local_err = NULL; | |
981 | char *value = qemu_opt_get_del(opts, "detect-zeroes"); | |
982 | BlockdevDetectZeroesOptions detect_zeroes = | |
983 | qapi_enum_parse(&BlockdevDetectZeroesOptions_lookup, value, | |
984 | BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF, &local_err); | |
985 | g_free(value); | |
986 | if (local_err) { | |
987 | error_propagate(errp, local_err); | |
988 | return detect_zeroes; | |
989 | } | |
990 | ||
991 | if (detect_zeroes == BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP && | |
992 | !(open_flags & BDRV_O_UNMAP)) | |
993 | { | |
994 | error_setg(errp, "setting detect-zeroes to unmap is not allowed " | |
995 | "without setting discard operation to unmap"); | |
996 | } | |
997 | ||
998 | return detect_zeroes; | |
999 | } | |
1000 | ||
f80f2673 AM |
1001 | /** |
1002 | * Set open flags for aio engine | |
1003 | * | |
1004 | * Return 0 on success, -1 if the engine specified is invalid | |
1005 | */ | |
1006 | int bdrv_parse_aio(const char *mode, int *flags) | |
1007 | { | |
1008 | if (!strcmp(mode, "threads")) { | |
1009 | /* do nothing, default */ | |
1010 | } else if (!strcmp(mode, "native")) { | |
1011 | *flags |= BDRV_O_NATIVE_AIO; | |
1012 | #ifdef CONFIG_LINUX_IO_URING | |
1013 | } else if (!strcmp(mode, "io_uring")) { | |
1014 | *flags |= BDRV_O_IO_URING; | |
1015 | #endif | |
1016 | } else { | |
1017 | return -1; | |
1018 | } | |
1019 | ||
1020 | return 0; | |
1021 | } | |
1022 | ||
9e8f1835 PB |
1023 | /** |
1024 | * Set open flags for a given discard mode | |
1025 | * | |
1026 | * Return 0 on success, -1 if the discard mode was invalid. | |
1027 | */ | |
1028 | int bdrv_parse_discard_flags(const char *mode, int *flags) | |
1029 | { | |
1030 | *flags &= ~BDRV_O_UNMAP; | |
1031 | ||
1032 | if (!strcmp(mode, "off") || !strcmp(mode, "ignore")) { | |
1033 | /* do nothing */ | |
1034 | } else if (!strcmp(mode, "on") || !strcmp(mode, "unmap")) { | |
1035 | *flags |= BDRV_O_UNMAP; | |
1036 | } else { | |
1037 | return -1; | |
1038 | } | |
1039 | ||
1040 | return 0; | |
1041 | } | |
1042 | ||
c3993cdc SH |
1043 | /** |
1044 | * Set open flags for a given cache mode | |
1045 | * | |
1046 | * Return 0 on success, -1 if the cache mode was invalid. | |
1047 | */ | |
53e8ae01 | 1048 | int bdrv_parse_cache_mode(const char *mode, int *flags, bool *writethrough) |
c3993cdc SH |
1049 | { |
1050 | *flags &= ~BDRV_O_CACHE_MASK; | |
1051 | ||
1052 | if (!strcmp(mode, "off") || !strcmp(mode, "none")) { | |
53e8ae01 KW |
1053 | *writethrough = false; |
1054 | *flags |= BDRV_O_NOCACHE; | |
92196b2f | 1055 | } else if (!strcmp(mode, "directsync")) { |
53e8ae01 | 1056 | *writethrough = true; |
92196b2f | 1057 | *flags |= BDRV_O_NOCACHE; |
c3993cdc | 1058 | } else if (!strcmp(mode, "writeback")) { |
53e8ae01 | 1059 | *writethrough = false; |
c3993cdc | 1060 | } else if (!strcmp(mode, "unsafe")) { |
53e8ae01 | 1061 | *writethrough = false; |
c3993cdc SH |
1062 | *flags |= BDRV_O_NO_FLUSH; |
1063 | } else if (!strcmp(mode, "writethrough")) { | |
53e8ae01 | 1064 | *writethrough = true; |
c3993cdc SH |
1065 | } else { |
1066 | return -1; | |
1067 | } | |
1068 | ||
1069 | return 0; | |
1070 | } | |
1071 | ||
b5411555 KW |
1072 | static char *bdrv_child_get_parent_desc(BdrvChild *c) |
1073 | { | |
1074 | BlockDriverState *parent = c->opaque; | |
1075 | return g_strdup(bdrv_get_device_or_node_name(parent)); | |
1076 | } | |
1077 | ||
20018e12 KW |
1078 | static void bdrv_child_cb_drained_begin(BdrvChild *child) |
1079 | { | |
1080 | BlockDriverState *bs = child->opaque; | |
6cd5c9d7 | 1081 | bdrv_do_drained_begin_quiesce(bs, NULL, false); |
20018e12 KW |
1082 | } |
1083 | ||
89bd0305 KW |
1084 | static bool bdrv_child_cb_drained_poll(BdrvChild *child) |
1085 | { | |
1086 | BlockDriverState *bs = child->opaque; | |
6cd5c9d7 | 1087 | return bdrv_drain_poll(bs, false, NULL, false); |
89bd0305 KW |
1088 | } |
1089 | ||
e037c09c HR |
1090 | static void bdrv_child_cb_drained_end(BdrvChild *child, |
1091 | int *drained_end_counter) | |
20018e12 KW |
1092 | { |
1093 | BlockDriverState *bs = child->opaque; | |
e037c09c | 1094 | bdrv_drained_end_no_poll(bs, drained_end_counter); |
20018e12 KW |
1095 | } |
1096 | ||
38701b6a KW |
1097 | static int bdrv_child_cb_inactivate(BdrvChild *child) |
1098 | { | |
1099 | BlockDriverState *bs = child->opaque; | |
1100 | assert(bs->open_flags & BDRV_O_INACTIVE); | |
1101 | return 0; | |
1102 | } | |
1103 | ||
5d231849 KW |
1104 | static bool bdrv_child_cb_can_set_aio_ctx(BdrvChild *child, AioContext *ctx, |
1105 | GSList **ignore, Error **errp) | |
1106 | { | |
1107 | BlockDriverState *bs = child->opaque; | |
1108 | return bdrv_can_set_aio_context(bs, ctx, ignore, errp); | |
1109 | } | |
1110 | ||
53a7d041 KW |
1111 | static void bdrv_child_cb_set_aio_ctx(BdrvChild *child, AioContext *ctx, |
1112 | GSList **ignore) | |
1113 | { | |
1114 | BlockDriverState *bs = child->opaque; | |
1115 | return bdrv_set_aio_context_ignore(bs, ctx, ignore); | |
1116 | } | |
1117 | ||
b1e6fc08 | 1118 | /* |
73176bee KW |
1119 | * Returns the options and flags that a temporary snapshot should get, based on |
1120 | * the originally requested flags (the originally requested image will have | |
1121 | * flags like a backing file) | |
b1e6fc08 | 1122 | */ |
73176bee KW |
1123 | static void bdrv_temp_snapshot_options(int *child_flags, QDict *child_options, |
1124 | int parent_flags, QDict *parent_options) | |
b1e6fc08 | 1125 | { |
73176bee KW |
1126 | *child_flags = (parent_flags & ~BDRV_O_SNAPSHOT) | BDRV_O_TEMPORARY; |
1127 | ||
1128 | /* For temporary files, unconditional cache=unsafe is fine */ | |
73176bee KW |
1129 | qdict_set_default_str(child_options, BDRV_OPT_CACHE_DIRECT, "off"); |
1130 | qdict_set_default_str(child_options, BDRV_OPT_CACHE_NO_FLUSH, "on"); | |
41869044 | 1131 | |
3f48686f | 1132 | /* Copy the read-only and discard options from the parent */ |
f87a0e29 | 1133 | qdict_copy_default(child_options, parent_options, BDRV_OPT_READ_ONLY); |
3f48686f | 1134 | qdict_copy_default(child_options, parent_options, BDRV_OPT_DISCARD); |
f87a0e29 | 1135 | |
41869044 KW |
1136 | /* aio=native doesn't work for cache.direct=off, so disable it for the |
1137 | * temporary snapshot */ | |
1138 | *child_flags &= ~BDRV_O_NATIVE_AIO; | |
b1e6fc08 KW |
1139 | } |
1140 | ||
db95dbba KW |
1141 | static void bdrv_backing_attach(BdrvChild *c) |
1142 | { | |
1143 | BlockDriverState *parent = c->opaque; | |
1144 | BlockDriverState *backing_hd = c->bs; | |
1145 | ||
1146 | assert(!parent->backing_blocker); | |
1147 | error_setg(&parent->backing_blocker, | |
1148 | "node is used as backing hd of '%s'", | |
1149 | bdrv_get_device_or_node_name(parent)); | |
1150 | ||
f30c66ba HR |
1151 | bdrv_refresh_filename(backing_hd); |
1152 | ||
db95dbba KW |
1153 | parent->open_flags &= ~BDRV_O_NO_BACKING; |
1154 | pstrcpy(parent->backing_file, sizeof(parent->backing_file), | |
1155 | backing_hd->filename); | |
1156 | pstrcpy(parent->backing_format, sizeof(parent->backing_format), | |
1157 | backing_hd->drv ? backing_hd->drv->format_name : ""); | |
1158 | ||
1159 | bdrv_op_block_all(backing_hd, parent->backing_blocker); | |
1160 | /* Otherwise we won't be able to commit or stream */ | |
1161 | bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_COMMIT_TARGET, | |
1162 | parent->backing_blocker); | |
1163 | bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_STREAM, | |
1164 | parent->backing_blocker); | |
1165 | /* | |
1166 | * We do backup in 3 ways: | |
1167 | * 1. drive backup | |
1168 | * The target bs is new opened, and the source is top BDS | |
1169 | * 2. blockdev backup | |
1170 | * Both the source and the target are top BDSes. | |
1171 | * 3. internal backup(used for block replication) | |
1172 | * Both the source and the target are backing file | |
1173 | * | |
1174 | * In case 1 and 2, neither the source nor the target is the backing file. | |
1175 | * In case 3, we will block the top BDS, so there is only one block job | |
1176 | * for the top BDS and its backing chain. | |
1177 | */ | |
1178 | bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_BACKUP_SOURCE, | |
1179 | parent->backing_blocker); | |
1180 | bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_BACKUP_TARGET, | |
1181 | parent->backing_blocker); | |
ca2f1234 | 1182 | } |
d736f119 | 1183 | |
db95dbba KW |
1184 | static void bdrv_backing_detach(BdrvChild *c) |
1185 | { | |
1186 | BlockDriverState *parent = c->opaque; | |
1187 | ||
1188 | assert(parent->backing_blocker); | |
1189 | bdrv_op_unblock_all(c->bs, parent->backing_blocker); | |
1190 | error_free(parent->backing_blocker); | |
1191 | parent->backing_blocker = NULL; | |
48e08288 | 1192 | } |
d736f119 | 1193 | |
6858eba0 KW |
1194 | static int bdrv_backing_update_filename(BdrvChild *c, BlockDriverState *base, |
1195 | const char *filename, Error **errp) | |
1196 | { | |
1197 | BlockDriverState *parent = c->opaque; | |
e94d3dba | 1198 | bool read_only = bdrv_is_read_only(parent); |
6858eba0 KW |
1199 | int ret; |
1200 | ||
e94d3dba AG |
1201 | if (read_only) { |
1202 | ret = bdrv_reopen_set_read_only(parent, false, errp); | |
61f09cea KW |
1203 | if (ret < 0) { |
1204 | return ret; | |
1205 | } | |
1206 | } | |
1207 | ||
6858eba0 KW |
1208 | ret = bdrv_change_backing_file(parent, filename, |
1209 | base->drv ? base->drv->format_name : ""); | |
1210 | if (ret < 0) { | |
64730694 | 1211 | error_setg_errno(errp, -ret, "Could not update backing file link"); |
6858eba0 KW |
1212 | } |
1213 | ||
e94d3dba AG |
1214 | if (read_only) { |
1215 | bdrv_reopen_set_read_only(parent, true, NULL); | |
61f09cea KW |
1216 | } |
1217 | ||
6858eba0 KW |
1218 | return ret; |
1219 | } | |
1220 | ||
fae8bd39 HR |
1221 | /* |
1222 | * Returns the options and flags that a generic child of a BDS should | |
1223 | * get, based on the given options and flags for the parent BDS. | |
1224 | */ | |
00ff7ffd HR |
1225 | static void bdrv_inherited_options(BdrvChildRole role, bool parent_is_format, |
1226 | int *child_flags, QDict *child_options, | |
1227 | int parent_flags, QDict *parent_options) | |
fae8bd39 HR |
1228 | { |
1229 | int flags = parent_flags; | |
1230 | ||
1231 | /* | |
1232 | * First, decide whether to set, clear, or leave BDRV_O_PROTOCOL. | |
1233 | * Generally, the question to answer is: Should this child be | |
1234 | * format-probed by default? | |
1235 | */ | |
1236 | ||
1237 | /* | |
1238 | * Pure and non-filtered data children of non-format nodes should | |
1239 | * be probed by default (even when the node itself has BDRV_O_PROTOCOL | |
1240 | * set). This only affects a very limited set of drivers (namely | |
1241 | * quorum and blkverify when this comment was written). | |
1242 | * Force-clear BDRV_O_PROTOCOL then. | |
1243 | */ | |
1244 | if (!parent_is_format && | |
1245 | (role & BDRV_CHILD_DATA) && | |
1246 | !(role & (BDRV_CHILD_METADATA | BDRV_CHILD_FILTERED))) | |
1247 | { | |
1248 | flags &= ~BDRV_O_PROTOCOL; | |
1249 | } | |
1250 | ||
1251 | /* | |
1252 | * All children of format nodes (except for COW children) and all | |
1253 | * metadata children in general should never be format-probed. | |
1254 | * Force-set BDRV_O_PROTOCOL then. | |
1255 | */ | |
1256 | if ((parent_is_format && !(role & BDRV_CHILD_COW)) || | |
1257 | (role & BDRV_CHILD_METADATA)) | |
1258 | { | |
1259 | flags |= BDRV_O_PROTOCOL; | |
1260 | } | |
1261 | ||
1262 | /* | |
1263 | * If the cache mode isn't explicitly set, inherit direct and no-flush from | |
1264 | * the parent. | |
1265 | */ | |
1266 | qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_DIRECT); | |
1267 | qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_NO_FLUSH); | |
1268 | qdict_copy_default(child_options, parent_options, BDRV_OPT_FORCE_SHARE); | |
1269 | ||
1270 | if (role & BDRV_CHILD_COW) { | |
1271 | /* backing files are opened read-only by default */ | |
1272 | qdict_set_default_str(child_options, BDRV_OPT_READ_ONLY, "on"); | |
1273 | qdict_set_default_str(child_options, BDRV_OPT_AUTO_READ_ONLY, "off"); | |
1274 | } else { | |
1275 | /* Inherit the read-only option from the parent if it's not set */ | |
1276 | qdict_copy_default(child_options, parent_options, BDRV_OPT_READ_ONLY); | |
1277 | qdict_copy_default(child_options, parent_options, | |
1278 | BDRV_OPT_AUTO_READ_ONLY); | |
1279 | } | |
1280 | ||
1281 | /* | |
1282 | * bdrv_co_pdiscard() respects unmap policy for the parent, so we | |
1283 | * can default to enable it on lower layers regardless of the | |
1284 | * parent option. | |
1285 | */ | |
1286 | qdict_set_default_str(child_options, BDRV_OPT_DISCARD, "unmap"); | |
1287 | ||
1288 | /* Clear flags that only apply to the top layer */ | |
1289 | flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_COPY_ON_READ); | |
1290 | ||
1291 | if (role & BDRV_CHILD_METADATA) { | |
1292 | flags &= ~BDRV_O_NO_IO; | |
1293 | } | |
1294 | if (role & BDRV_CHILD_COW) { | |
1295 | flags &= ~BDRV_O_TEMPORARY; | |
1296 | } | |
1297 | ||
1298 | *child_flags = flags; | |
1299 | } | |
1300 | ||
ca2f1234 HR |
1301 | static void bdrv_child_cb_attach(BdrvChild *child) |
1302 | { | |
1303 | BlockDriverState *bs = child->opaque; | |
1304 | ||
1305 | if (child->role & BDRV_CHILD_COW) { | |
1306 | bdrv_backing_attach(child); | |
1307 | } | |
1308 | ||
1309 | bdrv_apply_subtree_drain(child, bs); | |
1310 | } | |
1311 | ||
48e08288 HR |
1312 | static void bdrv_child_cb_detach(BdrvChild *child) |
1313 | { | |
1314 | BlockDriverState *bs = child->opaque; | |
1315 | ||
1316 | if (child->role & BDRV_CHILD_COW) { | |
1317 | bdrv_backing_detach(child); | |
1318 | } | |
1319 | ||
1320 | bdrv_unapply_subtree_drain(child, bs); | |
1321 | } | |
1322 | ||
43483550 HR |
1323 | static int bdrv_child_cb_update_filename(BdrvChild *c, BlockDriverState *base, |
1324 | const char *filename, Error **errp) | |
1325 | { | |
1326 | if (c->role & BDRV_CHILD_COW) { | |
1327 | return bdrv_backing_update_filename(c, base, filename, errp); | |
1328 | } | |
1329 | return 0; | |
1330 | } | |
1331 | ||
1332 | const BdrvChildClass child_of_bds = { | |
1333 | .parent_is_bds = true, | |
1334 | .get_parent_desc = bdrv_child_get_parent_desc, | |
1335 | .inherit_options = bdrv_inherited_options, | |
1336 | .drained_begin = bdrv_child_cb_drained_begin, | |
1337 | .drained_poll = bdrv_child_cb_drained_poll, | |
1338 | .drained_end = bdrv_child_cb_drained_end, | |
1339 | .attach = bdrv_child_cb_attach, | |
1340 | .detach = bdrv_child_cb_detach, | |
1341 | .inactivate = bdrv_child_cb_inactivate, | |
1342 | .can_set_aio_ctx = bdrv_child_cb_can_set_aio_ctx, | |
1343 | .set_aio_ctx = bdrv_child_cb_set_aio_ctx, | |
1344 | .update_filename = bdrv_child_cb_update_filename, | |
1345 | }; | |
1346 | ||
7b272452 KW |
1347 | static int bdrv_open_flags(BlockDriverState *bs, int flags) |
1348 | { | |
61de4c68 | 1349 | int open_flags = flags; |
7b272452 KW |
1350 | |
1351 | /* | |
1352 | * Clear flags that are internal to the block layer before opening the | |
1353 | * image. | |
1354 | */ | |
20cca275 | 1355 | open_flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_PROTOCOL); |
7b272452 | 1356 | |
7b272452 KW |
1357 | return open_flags; |
1358 | } | |
1359 | ||
91a097e7 KW |
1360 | static void update_flags_from_options(int *flags, QemuOpts *opts) |
1361 | { | |
2a3d4331 | 1362 | *flags &= ~(BDRV_O_CACHE_MASK | BDRV_O_RDWR | BDRV_O_AUTO_RDONLY); |
91a097e7 | 1363 | |
57f9db9a | 1364 | if (qemu_opt_get_bool_del(opts, BDRV_OPT_CACHE_NO_FLUSH, false)) { |
91a097e7 KW |
1365 | *flags |= BDRV_O_NO_FLUSH; |
1366 | } | |
1367 | ||
57f9db9a | 1368 | if (qemu_opt_get_bool_del(opts, BDRV_OPT_CACHE_DIRECT, false)) { |
91a097e7 KW |
1369 | *flags |= BDRV_O_NOCACHE; |
1370 | } | |
f87a0e29 | 1371 | |
57f9db9a | 1372 | if (!qemu_opt_get_bool_del(opts, BDRV_OPT_READ_ONLY, false)) { |
f87a0e29 AG |
1373 | *flags |= BDRV_O_RDWR; |
1374 | } | |
1375 | ||
e35bdc12 KW |
1376 | if (qemu_opt_get_bool_del(opts, BDRV_OPT_AUTO_READ_ONLY, false)) { |
1377 | *flags |= BDRV_O_AUTO_RDONLY; | |
1378 | } | |
91a097e7 KW |
1379 | } |
1380 | ||
1381 | static void update_options_from_flags(QDict *options, int flags) | |
1382 | { | |
91a097e7 | 1383 | if (!qdict_haskey(options, BDRV_OPT_CACHE_DIRECT)) { |
46f5ac20 | 1384 | qdict_put_bool(options, BDRV_OPT_CACHE_DIRECT, flags & BDRV_O_NOCACHE); |
91a097e7 KW |
1385 | } |
1386 | if (!qdict_haskey(options, BDRV_OPT_CACHE_NO_FLUSH)) { | |
46f5ac20 EB |
1387 | qdict_put_bool(options, BDRV_OPT_CACHE_NO_FLUSH, |
1388 | flags & BDRV_O_NO_FLUSH); | |
91a097e7 | 1389 | } |
f87a0e29 | 1390 | if (!qdict_haskey(options, BDRV_OPT_READ_ONLY)) { |
46f5ac20 | 1391 | qdict_put_bool(options, BDRV_OPT_READ_ONLY, !(flags & BDRV_O_RDWR)); |
f87a0e29 | 1392 | } |
e35bdc12 KW |
1393 | if (!qdict_haskey(options, BDRV_OPT_AUTO_READ_ONLY)) { |
1394 | qdict_put_bool(options, BDRV_OPT_AUTO_READ_ONLY, | |
1395 | flags & BDRV_O_AUTO_RDONLY); | |
1396 | } | |
91a097e7 KW |
1397 | } |
1398 | ||
636ea370 KW |
1399 | static void bdrv_assign_node_name(BlockDriverState *bs, |
1400 | const char *node_name, | |
1401 | Error **errp) | |
6913c0c2 | 1402 | { |
15489c76 | 1403 | char *gen_node_name = NULL; |
6913c0c2 | 1404 | |
15489c76 JC |
1405 | if (!node_name) { |
1406 | node_name = gen_node_name = id_generate(ID_BLOCK); | |
1407 | } else if (!id_wellformed(node_name)) { | |
1408 | /* | |
1409 | * Check for empty string or invalid characters, but not if it is | |
1410 | * generated (generated names use characters not available to the user) | |
1411 | */ | |
9aebf3b8 | 1412 | error_setg(errp, "Invalid node name"); |
636ea370 | 1413 | return; |
6913c0c2 BC |
1414 | } |
1415 | ||
0c5e94ee | 1416 | /* takes care of avoiding namespaces collisions */ |
7f06d47e | 1417 | if (blk_by_name(node_name)) { |
0c5e94ee BC |
1418 | error_setg(errp, "node-name=%s is conflicting with a device id", |
1419 | node_name); | |
15489c76 | 1420 | goto out; |
0c5e94ee BC |
1421 | } |
1422 | ||
6913c0c2 BC |
1423 | /* takes care of avoiding duplicates node names */ |
1424 | if (bdrv_find_node(node_name)) { | |
1425 | error_setg(errp, "Duplicate node name"); | |
15489c76 | 1426 | goto out; |
6913c0c2 BC |
1427 | } |
1428 | ||
824808dd KW |
1429 | /* Make sure that the node name isn't truncated */ |
1430 | if (strlen(node_name) >= sizeof(bs->node_name)) { | |
1431 | error_setg(errp, "Node name too long"); | |
1432 | goto out; | |
1433 | } | |
1434 | ||
6913c0c2 BC |
1435 | /* copy node name into the bs and insert it into the graph list */ |
1436 | pstrcpy(bs->node_name, sizeof(bs->node_name), node_name); | |
1437 | QTAILQ_INSERT_TAIL(&graph_bdrv_states, bs, node_list); | |
15489c76 JC |
1438 | out: |
1439 | g_free(gen_node_name); | |
6913c0c2 BC |
1440 | } |
1441 | ||
01a56501 KW |
1442 | static int bdrv_open_driver(BlockDriverState *bs, BlockDriver *drv, |
1443 | const char *node_name, QDict *options, | |
1444 | int open_flags, Error **errp) | |
1445 | { | |
1446 | Error *local_err = NULL; | |
0f12264e | 1447 | int i, ret; |
01a56501 KW |
1448 | |
1449 | bdrv_assign_node_name(bs, node_name, &local_err); | |
1450 | if (local_err) { | |
1451 | error_propagate(errp, local_err); | |
1452 | return -EINVAL; | |
1453 | } | |
1454 | ||
1455 | bs->drv = drv; | |
680c7f96 | 1456 | bs->read_only = !(bs->open_flags & BDRV_O_RDWR); |
01a56501 KW |
1457 | bs->opaque = g_malloc0(drv->instance_size); |
1458 | ||
1459 | if (drv->bdrv_file_open) { | |
1460 | assert(!drv->bdrv_needs_filename || bs->filename[0]); | |
1461 | ret = drv->bdrv_file_open(bs, options, open_flags, &local_err); | |
680c7f96 | 1462 | } else if (drv->bdrv_open) { |
01a56501 | 1463 | ret = drv->bdrv_open(bs, options, open_flags, &local_err); |
680c7f96 KW |
1464 | } else { |
1465 | ret = 0; | |
01a56501 KW |
1466 | } |
1467 | ||
1468 | if (ret < 0) { | |
1469 | if (local_err) { | |
1470 | error_propagate(errp, local_err); | |
1471 | } else if (bs->filename[0]) { | |
1472 | error_setg_errno(errp, -ret, "Could not open '%s'", bs->filename); | |
1473 | } else { | |
1474 | error_setg_errno(errp, -ret, "Could not open image"); | |
1475 | } | |
180ca19a | 1476 | goto open_failed; |
01a56501 KW |
1477 | } |
1478 | ||
1479 | ret = refresh_total_sectors(bs, bs->total_sectors); | |
1480 | if (ret < 0) { | |
1481 | error_setg_errno(errp, -ret, "Could not refresh total sector count"); | |
180ca19a | 1482 | return ret; |
01a56501 KW |
1483 | } |
1484 | ||
1485 | bdrv_refresh_limits(bs, &local_err); | |
1486 | if (local_err) { | |
1487 | error_propagate(errp, local_err); | |
180ca19a | 1488 | return -EINVAL; |
01a56501 KW |
1489 | } |
1490 | ||
1491 | assert(bdrv_opt_mem_align(bs) != 0); | |
1492 | assert(bdrv_min_mem_align(bs) != 0); | |
1493 | assert(is_power_of_2(bs->bl.request_alignment)); | |
1494 | ||
0f12264e KW |
1495 | for (i = 0; i < bs->quiesce_counter; i++) { |
1496 | if (drv->bdrv_co_drain_begin) { | |
1497 | drv->bdrv_co_drain_begin(bs); | |
1498 | } | |
1499 | } | |
1500 | ||
01a56501 | 1501 | return 0; |
180ca19a MP |
1502 | open_failed: |
1503 | bs->drv = NULL; | |
1504 | if (bs->file != NULL) { | |
1505 | bdrv_unref_child(bs, bs->file); | |
1506 | bs->file = NULL; | |
1507 | } | |
01a56501 KW |
1508 | g_free(bs->opaque); |
1509 | bs->opaque = NULL; | |
01a56501 KW |
1510 | return ret; |
1511 | } | |
1512 | ||
680c7f96 KW |
1513 | BlockDriverState *bdrv_new_open_driver(BlockDriver *drv, const char *node_name, |
1514 | int flags, Error **errp) | |
1515 | { | |
1516 | BlockDriverState *bs; | |
1517 | int ret; | |
1518 | ||
1519 | bs = bdrv_new(); | |
1520 | bs->open_flags = flags; | |
1521 | bs->explicit_options = qdict_new(); | |
1522 | bs->options = qdict_new(); | |
1523 | bs->opaque = NULL; | |
1524 | ||
1525 | update_options_from_flags(bs->options, flags); | |
1526 | ||
1527 | ret = bdrv_open_driver(bs, drv, node_name, bs->options, flags, errp); | |
1528 | if (ret < 0) { | |
cb3e7f08 | 1529 | qobject_unref(bs->explicit_options); |
180ca19a | 1530 | bs->explicit_options = NULL; |
cb3e7f08 | 1531 | qobject_unref(bs->options); |
180ca19a | 1532 | bs->options = NULL; |
680c7f96 KW |
1533 | bdrv_unref(bs); |
1534 | return NULL; | |
1535 | } | |
1536 | ||
1537 | return bs; | |
1538 | } | |
1539 | ||
c5f3014b | 1540 | QemuOptsList bdrv_runtime_opts = { |
18edf289 KW |
1541 | .name = "bdrv_common", |
1542 | .head = QTAILQ_HEAD_INITIALIZER(bdrv_runtime_opts.head), | |
1543 | .desc = { | |
1544 | { | |
1545 | .name = "node-name", | |
1546 | .type = QEMU_OPT_STRING, | |
1547 | .help = "Node name of the block device node", | |
1548 | }, | |
62392ebb KW |
1549 | { |
1550 | .name = "driver", | |
1551 | .type = QEMU_OPT_STRING, | |
1552 | .help = "Block driver to use for the node", | |
1553 | }, | |
91a097e7 KW |
1554 | { |
1555 | .name = BDRV_OPT_CACHE_DIRECT, | |
1556 | .type = QEMU_OPT_BOOL, | |
1557 | .help = "Bypass software writeback cache on the host", | |
1558 | }, | |
1559 | { | |
1560 | .name = BDRV_OPT_CACHE_NO_FLUSH, | |
1561 | .type = QEMU_OPT_BOOL, | |
1562 | .help = "Ignore flush requests", | |
1563 | }, | |
f87a0e29 AG |
1564 | { |
1565 | .name = BDRV_OPT_READ_ONLY, | |
1566 | .type = QEMU_OPT_BOOL, | |
1567 | .help = "Node is opened in read-only mode", | |
1568 | }, | |
e35bdc12 KW |
1569 | { |
1570 | .name = BDRV_OPT_AUTO_READ_ONLY, | |
1571 | .type = QEMU_OPT_BOOL, | |
1572 | .help = "Node can become read-only if opening read-write fails", | |
1573 | }, | |
692e01a2 KW |
1574 | { |
1575 | .name = "detect-zeroes", | |
1576 | .type = QEMU_OPT_STRING, | |
1577 | .help = "try to optimize zero writes (off, on, unmap)", | |
1578 | }, | |
818584a4 | 1579 | { |
415bbca8 | 1580 | .name = BDRV_OPT_DISCARD, |
818584a4 KW |
1581 | .type = QEMU_OPT_STRING, |
1582 | .help = "discard operation (ignore/off, unmap/on)", | |
1583 | }, | |
5a9347c6 FZ |
1584 | { |
1585 | .name = BDRV_OPT_FORCE_SHARE, | |
1586 | .type = QEMU_OPT_BOOL, | |
1587 | .help = "always accept other writers (default: off)", | |
1588 | }, | |
18edf289 KW |
1589 | { /* end of list */ } |
1590 | }, | |
1591 | }; | |
1592 | ||
5a5e7f8c ML |
1593 | QemuOptsList bdrv_create_opts_simple = { |
1594 | .name = "simple-create-opts", | |
1595 | .head = QTAILQ_HEAD_INITIALIZER(bdrv_create_opts_simple.head), | |
fd17146c HR |
1596 | .desc = { |
1597 | { | |
1598 | .name = BLOCK_OPT_SIZE, | |
1599 | .type = QEMU_OPT_SIZE, | |
1600 | .help = "Virtual disk size" | |
1601 | }, | |
1602 | { | |
1603 | .name = BLOCK_OPT_PREALLOC, | |
1604 | .type = QEMU_OPT_STRING, | |
1605 | .help = "Preallocation mode (allowed values: off)" | |
1606 | }, | |
1607 | { /* end of list */ } | |
1608 | } | |
1609 | }; | |
1610 | ||
57915332 KW |
1611 | /* |
1612 | * Common part for opening disk images and files | |
b6ad491a KW |
1613 | * |
1614 | * Removes all processed options from *options. | |
57915332 | 1615 | */ |
5696c6e3 | 1616 | static int bdrv_open_common(BlockDriverState *bs, BlockBackend *file, |
82dc8b41 | 1617 | QDict *options, Error **errp) |
57915332 KW |
1618 | { |
1619 | int ret, open_flags; | |
035fccdf | 1620 | const char *filename; |
62392ebb | 1621 | const char *driver_name = NULL; |
6913c0c2 | 1622 | const char *node_name = NULL; |
818584a4 | 1623 | const char *discard; |
18edf289 | 1624 | QemuOpts *opts; |
62392ebb | 1625 | BlockDriver *drv; |
34b5d2c6 | 1626 | Error *local_err = NULL; |
57915332 | 1627 | |
6405875c | 1628 | assert(bs->file == NULL); |
707ff828 | 1629 | assert(options != NULL && bs->options != options); |
57915332 | 1630 | |
62392ebb | 1631 | opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort); |
af175e85 | 1632 | if (!qemu_opts_absorb_qdict(opts, options, errp)) { |
62392ebb KW |
1633 | ret = -EINVAL; |
1634 | goto fail_opts; | |
1635 | } | |
1636 | ||
9b7e8691 AG |
1637 | update_flags_from_options(&bs->open_flags, opts); |
1638 | ||
62392ebb KW |
1639 | driver_name = qemu_opt_get(opts, "driver"); |
1640 | drv = bdrv_find_format(driver_name); | |
1641 | assert(drv != NULL); | |
1642 | ||
5a9347c6 FZ |
1643 | bs->force_share = qemu_opt_get_bool(opts, BDRV_OPT_FORCE_SHARE, false); |
1644 | ||
1645 | if (bs->force_share && (bs->open_flags & BDRV_O_RDWR)) { | |
1646 | error_setg(errp, | |
1647 | BDRV_OPT_FORCE_SHARE | |
1648 | "=on can only be used with read-only images"); | |
1649 | ret = -EINVAL; | |
1650 | goto fail_opts; | |
1651 | } | |
1652 | ||
45673671 | 1653 | if (file != NULL) { |
f30c66ba | 1654 | bdrv_refresh_filename(blk_bs(file)); |
5696c6e3 | 1655 | filename = blk_bs(file)->filename; |
45673671 | 1656 | } else { |
129c7d1c MA |
1657 | /* |
1658 | * Caution: while qdict_get_try_str() is fine, getting | |
1659 | * non-string types would require more care. When @options | |
1660 | * come from -blockdev or blockdev_add, its members are typed | |
1661 | * according to the QAPI schema, but when they come from | |
1662 | * -drive, they're all QString. | |
1663 | */ | |
45673671 KW |
1664 | filename = qdict_get_try_str(options, "filename"); |
1665 | } | |
1666 | ||
4a008240 | 1667 | if (drv->bdrv_needs_filename && (!filename || !filename[0])) { |
765003db KW |
1668 | error_setg(errp, "The '%s' block driver requires a file name", |
1669 | drv->format_name); | |
18edf289 KW |
1670 | ret = -EINVAL; |
1671 | goto fail_opts; | |
6913c0c2 | 1672 | } |
6913c0c2 | 1673 | |
82dc8b41 KW |
1674 | trace_bdrv_open_common(bs, filename ?: "", bs->open_flags, |
1675 | drv->format_name); | |
62392ebb | 1676 | |
82dc8b41 | 1677 | bs->read_only = !(bs->open_flags & BDRV_O_RDWR); |
b64ec4e4 FZ |
1678 | |
1679 | if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv, bs->read_only)) { | |
8be25de6 KW |
1680 | if (!bs->read_only && bdrv_is_whitelisted(drv, true)) { |
1681 | ret = bdrv_apply_auto_read_only(bs, NULL, NULL); | |
1682 | } else { | |
1683 | ret = -ENOTSUP; | |
1684 | } | |
1685 | if (ret < 0) { | |
1686 | error_setg(errp, | |
1687 | !bs->read_only && bdrv_is_whitelisted(drv, true) | |
1688 | ? "Driver '%s' can only be used for read-only devices" | |
1689 | : "Driver '%s' is not whitelisted", | |
1690 | drv->format_name); | |
1691 | goto fail_opts; | |
1692 | } | |
b64ec4e4 | 1693 | } |
57915332 | 1694 | |
d3faa13e PB |
1695 | /* bdrv_new() and bdrv_close() make it so */ |
1696 | assert(atomic_read(&bs->copy_on_read) == 0); | |
1697 | ||
82dc8b41 | 1698 | if (bs->open_flags & BDRV_O_COPY_ON_READ) { |
0ebd24e0 KW |
1699 | if (!bs->read_only) { |
1700 | bdrv_enable_copy_on_read(bs); | |
1701 | } else { | |
1702 | error_setg(errp, "Can't use copy-on-read on read-only device"); | |
18edf289 KW |
1703 | ret = -EINVAL; |
1704 | goto fail_opts; | |
0ebd24e0 | 1705 | } |
53fec9d3 SH |
1706 | } |
1707 | ||
415bbca8 | 1708 | discard = qemu_opt_get(opts, BDRV_OPT_DISCARD); |
818584a4 KW |
1709 | if (discard != NULL) { |
1710 | if (bdrv_parse_discard_flags(discard, &bs->open_flags) != 0) { | |
1711 | error_setg(errp, "Invalid discard option"); | |
1712 | ret = -EINVAL; | |
1713 | goto fail_opts; | |
1714 | } | |
1715 | } | |
1716 | ||
543770bd AG |
1717 | bs->detect_zeroes = |
1718 | bdrv_parse_detect_zeroes(opts, bs->open_flags, &local_err); | |
1719 | if (local_err) { | |
1720 | error_propagate(errp, local_err); | |
1721 | ret = -EINVAL; | |
1722 | goto fail_opts; | |
692e01a2 KW |
1723 | } |
1724 | ||
c2ad1b0c KW |
1725 | if (filename != NULL) { |
1726 | pstrcpy(bs->filename, sizeof(bs->filename), filename); | |
1727 | } else { | |
1728 | bs->filename[0] = '\0'; | |
1729 | } | |
91af7014 | 1730 | pstrcpy(bs->exact_filename, sizeof(bs->exact_filename), bs->filename); |
57915332 | 1731 | |
66f82cee | 1732 | /* Open the image, either directly or using a protocol */ |
82dc8b41 | 1733 | open_flags = bdrv_open_flags(bs, bs->open_flags); |
01a56501 | 1734 | node_name = qemu_opt_get(opts, "node-name"); |
57915332 | 1735 | |
01a56501 KW |
1736 | assert(!drv->bdrv_file_open || file == NULL); |
1737 | ret = bdrv_open_driver(bs, drv, node_name, options, open_flags, errp); | |
51762288 | 1738 | if (ret < 0) { |
01a56501 | 1739 | goto fail_opts; |
3baca891 KW |
1740 | } |
1741 | ||
18edf289 | 1742 | qemu_opts_del(opts); |
57915332 KW |
1743 | return 0; |
1744 | ||
18edf289 KW |
1745 | fail_opts: |
1746 | qemu_opts_del(opts); | |
57915332 KW |
1747 | return ret; |
1748 | } | |
1749 | ||
5e5c4f63 KW |
1750 | static QDict *parse_json_filename(const char *filename, Error **errp) |
1751 | { | |
1752 | QObject *options_obj; | |
1753 | QDict *options; | |
1754 | int ret; | |
1755 | ||
1756 | ret = strstart(filename, "json:", &filename); | |
1757 | assert(ret); | |
1758 | ||
5577fff7 | 1759 | options_obj = qobject_from_json(filename, errp); |
5e5c4f63 | 1760 | if (!options_obj) { |
5577fff7 | 1761 | error_prepend(errp, "Could not parse the JSON options: "); |
5e5c4f63 KW |
1762 | return NULL; |
1763 | } | |
1764 | ||
7dc847eb | 1765 | options = qobject_to(QDict, options_obj); |
ca6b6e1e | 1766 | if (!options) { |
cb3e7f08 | 1767 | qobject_unref(options_obj); |
5e5c4f63 KW |
1768 | error_setg(errp, "Invalid JSON object given"); |
1769 | return NULL; | |
1770 | } | |
1771 | ||
5e5c4f63 KW |
1772 | qdict_flatten(options); |
1773 | ||
1774 | return options; | |
1775 | } | |
1776 | ||
de3b53f0 KW |
1777 | static void parse_json_protocol(QDict *options, const char **pfilename, |
1778 | Error **errp) | |
1779 | { | |
1780 | QDict *json_options; | |
1781 | Error *local_err = NULL; | |
1782 | ||
1783 | /* Parse json: pseudo-protocol */ | |
1784 | if (!*pfilename || !g_str_has_prefix(*pfilename, "json:")) { | |
1785 | return; | |
1786 | } | |
1787 | ||
1788 | json_options = parse_json_filename(*pfilename, &local_err); | |
1789 | if (local_err) { | |
1790 | error_propagate(errp, local_err); | |
1791 | return; | |
1792 | } | |
1793 | ||
1794 | /* Options given in the filename have lower priority than options | |
1795 | * specified directly */ | |
1796 | qdict_join(options, json_options, false); | |
cb3e7f08 | 1797 | qobject_unref(json_options); |
de3b53f0 KW |
1798 | *pfilename = NULL; |
1799 | } | |
1800 | ||
b6ce07aa | 1801 | /* |
f54120ff KW |
1802 | * Fills in default options for opening images and converts the legacy |
1803 | * filename/flags pair to option QDict entries. | |
53a29513 HR |
1804 | * The BDRV_O_PROTOCOL flag in *flags will be set or cleared accordingly if a |
1805 | * block driver has been specified explicitly. | |
b6ce07aa | 1806 | */ |
de3b53f0 | 1807 | static int bdrv_fill_options(QDict **options, const char *filename, |
053e1578 | 1808 | int *flags, Error **errp) |
ea2384d3 | 1809 | { |
c2ad1b0c | 1810 | const char *drvname; |
53a29513 | 1811 | bool protocol = *flags & BDRV_O_PROTOCOL; |
e3fa4bfa | 1812 | bool parse_filename = false; |
053e1578 | 1813 | BlockDriver *drv = NULL; |
34b5d2c6 | 1814 | Error *local_err = NULL; |
83f64091 | 1815 | |
129c7d1c MA |
1816 | /* |
1817 | * Caution: while qdict_get_try_str() is fine, getting non-string | |
1818 | * types would require more care. When @options come from | |
1819 | * -blockdev or blockdev_add, its members are typed according to | |
1820 | * the QAPI schema, but when they come from -drive, they're all | |
1821 | * QString. | |
1822 | */ | |
53a29513 | 1823 | drvname = qdict_get_try_str(*options, "driver"); |
053e1578 HR |
1824 | if (drvname) { |
1825 | drv = bdrv_find_format(drvname); | |
1826 | if (!drv) { | |
1827 | error_setg(errp, "Unknown driver '%s'", drvname); | |
1828 | return -ENOENT; | |
1829 | } | |
1830 | /* If the user has explicitly specified the driver, this choice should | |
1831 | * override the BDRV_O_PROTOCOL flag */ | |
1832 | protocol = drv->bdrv_file_open; | |
53a29513 HR |
1833 | } |
1834 | ||
1835 | if (protocol) { | |
1836 | *flags |= BDRV_O_PROTOCOL; | |
1837 | } else { | |
1838 | *flags &= ~BDRV_O_PROTOCOL; | |
1839 | } | |
1840 | ||
91a097e7 KW |
1841 | /* Translate cache options from flags into options */ |
1842 | update_options_from_flags(*options, *flags); | |
1843 | ||
035fccdf | 1844 | /* Fetch the file name from the options QDict if necessary */ |
17b005f1 | 1845 | if (protocol && filename) { |
f54120ff | 1846 | if (!qdict_haskey(*options, "filename")) { |
46f5ac20 | 1847 | qdict_put_str(*options, "filename", filename); |
f54120ff KW |
1848 | parse_filename = true; |
1849 | } else { | |
1850 | error_setg(errp, "Can't specify 'file' and 'filename' options at " | |
1851 | "the same time"); | |
1852 | return -EINVAL; | |
1853 | } | |
035fccdf KW |
1854 | } |
1855 | ||
c2ad1b0c | 1856 | /* Find the right block driver */ |
129c7d1c | 1857 | /* See cautionary note on accessing @options above */ |
f54120ff | 1858 | filename = qdict_get_try_str(*options, "filename"); |
f54120ff | 1859 | |
053e1578 HR |
1860 | if (!drvname && protocol) { |
1861 | if (filename) { | |
1862 | drv = bdrv_find_protocol(filename, parse_filename, errp); | |
17b005f1 | 1863 | if (!drv) { |
053e1578 | 1864 | return -EINVAL; |
17b005f1 | 1865 | } |
053e1578 HR |
1866 | |
1867 | drvname = drv->format_name; | |
46f5ac20 | 1868 | qdict_put_str(*options, "driver", drvname); |
053e1578 HR |
1869 | } else { |
1870 | error_setg(errp, "Must specify either driver or file"); | |
1871 | return -EINVAL; | |
98289620 | 1872 | } |
c2ad1b0c KW |
1873 | } |
1874 | ||
17b005f1 | 1875 | assert(drv || !protocol); |
c2ad1b0c | 1876 | |
f54120ff | 1877 | /* Driver-specific filename parsing */ |
17b005f1 | 1878 | if (drv && drv->bdrv_parse_filename && parse_filename) { |
5acd9d81 | 1879 | drv->bdrv_parse_filename(filename, *options, &local_err); |
84d18f06 | 1880 | if (local_err) { |
34b5d2c6 | 1881 | error_propagate(errp, local_err); |
f54120ff | 1882 | return -EINVAL; |
6963a30d | 1883 | } |
cd5d031e HR |
1884 | |
1885 | if (!drv->bdrv_needs_filename) { | |
1886 | qdict_del(*options, "filename"); | |
cd5d031e | 1887 | } |
6963a30d KW |
1888 | } |
1889 | ||
f54120ff KW |
1890 | return 0; |
1891 | } | |
1892 | ||
3121fb45 KW |
1893 | static int bdrv_child_check_perm(BdrvChild *c, BlockReopenQueue *q, |
1894 | uint64_t perm, uint64_t shared, | |
9eab1544 HR |
1895 | GSList *ignore_children, |
1896 | bool *tighten_restrictions, Error **errp); | |
c1cef672 FZ |
1897 | static void bdrv_child_abort_perm_update(BdrvChild *c); |
1898 | static void bdrv_child_set_perm(BdrvChild *c, uint64_t perm, uint64_t shared); | |
1899 | ||
148eb13c KW |
1900 | typedef struct BlockReopenQueueEntry { |
1901 | bool prepared; | |
69b736e7 | 1902 | bool perms_checked; |
148eb13c | 1903 | BDRVReopenState state; |
859443b0 | 1904 | QTAILQ_ENTRY(BlockReopenQueueEntry) entry; |
148eb13c KW |
1905 | } BlockReopenQueueEntry; |
1906 | ||
1907 | /* | |
1908 | * Return the flags that @bs will have after the reopens in @q have | |
1909 | * successfully completed. If @q is NULL (or @bs is not contained in @q), | |
1910 | * return the current flags. | |
1911 | */ | |
1912 | static int bdrv_reopen_get_flags(BlockReopenQueue *q, BlockDriverState *bs) | |
1913 | { | |
1914 | BlockReopenQueueEntry *entry; | |
1915 | ||
1916 | if (q != NULL) { | |
859443b0 | 1917 | QTAILQ_FOREACH(entry, q, entry) { |
148eb13c KW |
1918 | if (entry->state.bs == bs) { |
1919 | return entry->state.flags; | |
1920 | } | |
1921 | } | |
1922 | } | |
1923 | ||
1924 | return bs->open_flags; | |
1925 | } | |
1926 | ||
1927 | /* Returns whether the image file can be written to after the reopen queue @q | |
1928 | * has been successfully applied, or right now if @q is NULL. */ | |
cc022140 HR |
1929 | static bool bdrv_is_writable_after_reopen(BlockDriverState *bs, |
1930 | BlockReopenQueue *q) | |
148eb13c KW |
1931 | { |
1932 | int flags = bdrv_reopen_get_flags(q, bs); | |
1933 | ||
1934 | return (flags & (BDRV_O_RDWR | BDRV_O_INACTIVE)) == BDRV_O_RDWR; | |
1935 | } | |
1936 | ||
cc022140 HR |
1937 | /* |
1938 | * Return whether the BDS can be written to. This is not necessarily | |
1939 | * the same as !bdrv_is_read_only(bs), as inactivated images may not | |
1940 | * be written to but do not count as read-only images. | |
1941 | */ | |
1942 | bool bdrv_is_writable(BlockDriverState *bs) | |
1943 | { | |
1944 | return bdrv_is_writable_after_reopen(bs, NULL); | |
1945 | } | |
1946 | ||
ffd1a5a2 | 1947 | static void bdrv_child_perm(BlockDriverState *bs, BlockDriverState *child_bs, |
e5d8a406 HR |
1948 | BdrvChild *c, BdrvChildRole role, |
1949 | BlockReopenQueue *reopen_queue, | |
ffd1a5a2 FZ |
1950 | uint64_t parent_perm, uint64_t parent_shared, |
1951 | uint64_t *nperm, uint64_t *nshared) | |
1952 | { | |
0b3ca76e | 1953 | assert(bs->drv && bs->drv->bdrv_child_perm); |
e5d8a406 | 1954 | bs->drv->bdrv_child_perm(bs, c, role, reopen_queue, |
0b3ca76e AG |
1955 | parent_perm, parent_shared, |
1956 | nperm, nshared); | |
e0995dc3 | 1957 | /* TODO Take force_share from reopen_queue */ |
ffd1a5a2 FZ |
1958 | if (child_bs && child_bs->force_share) { |
1959 | *nshared = BLK_PERM_ALL; | |
1960 | } | |
1961 | } | |
1962 | ||
33a610c3 KW |
1963 | /* |
1964 | * Check whether permissions on this node can be changed in a way that | |
1965 | * @cumulative_perms and @cumulative_shared_perms are the new cumulative | |
1966 | * permissions of all its parents. This involves checking whether all necessary | |
1967 | * permission changes to child nodes can be performed. | |
1968 | * | |
9eab1544 HR |
1969 | * Will set *tighten_restrictions to true if and only if new permissions have to |
1970 | * be taken or currently shared permissions are to be unshared. Otherwise, | |
1971 | * errors are not fatal as long as the caller accepts that the restrictions | |
1972 | * remain tighter than they need to be. The caller still has to abort the | |
1973 | * transaction. | |
1974 | * @tighten_restrictions cannot be used together with @q: When reopening, we may | |
1975 | * encounter fatal errors even though no restrictions are to be tightened. For | |
1976 | * example, changing a node from RW to RO will fail if the WRITE permission is | |
1977 | * to be kept. | |
1978 | * | |
33a610c3 KW |
1979 | * A call to this function must always be followed by a call to bdrv_set_perm() |
1980 | * or bdrv_abort_perm_update(). | |
1981 | */ | |
3121fb45 KW |
1982 | static int bdrv_check_perm(BlockDriverState *bs, BlockReopenQueue *q, |
1983 | uint64_t cumulative_perms, | |
46181129 | 1984 | uint64_t cumulative_shared_perms, |
9eab1544 HR |
1985 | GSList *ignore_children, |
1986 | bool *tighten_restrictions, Error **errp) | |
33a610c3 KW |
1987 | { |
1988 | BlockDriver *drv = bs->drv; | |
1989 | BdrvChild *c; | |
1990 | int ret; | |
1991 | ||
9eab1544 HR |
1992 | assert(!q || !tighten_restrictions); |
1993 | ||
1994 | if (tighten_restrictions) { | |
1995 | uint64_t current_perms, current_shared; | |
1996 | uint64_t added_perms, removed_shared_perms; | |
1997 | ||
1998 | bdrv_get_cumulative_perm(bs, ¤t_perms, ¤t_shared); | |
1999 | ||
2000 | added_perms = cumulative_perms & ~current_perms; | |
2001 | removed_shared_perms = current_shared & ~cumulative_shared_perms; | |
2002 | ||
2003 | *tighten_restrictions = added_perms || removed_shared_perms; | |
2004 | } | |
2005 | ||
33a610c3 KW |
2006 | /* Write permissions never work with read-only images */ |
2007 | if ((cumulative_perms & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)) && | |
cc022140 | 2008 | !bdrv_is_writable_after_reopen(bs, q)) |
33a610c3 | 2009 | { |
481e0eee HR |
2010 | if (!bdrv_is_writable_after_reopen(bs, NULL)) { |
2011 | error_setg(errp, "Block node is read-only"); | |
2012 | } else { | |
2013 | uint64_t current_perms, current_shared; | |
2014 | bdrv_get_cumulative_perm(bs, ¤t_perms, ¤t_shared); | |
2015 | if (current_perms & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)) { | |
2016 | error_setg(errp, "Cannot make block node read-only, there is " | |
2017 | "a writer on it"); | |
2018 | } else { | |
2019 | error_setg(errp, "Cannot make block node read-only and create " | |
2020 | "a writer on it"); | |
2021 | } | |
2022 | } | |
2023 | ||
33a610c3 KW |
2024 | return -EPERM; |
2025 | } | |
2026 | ||
2027 | /* Check this node */ | |
2028 | if (!drv) { | |
2029 | return 0; | |
2030 | } | |
2031 | ||
2032 | if (drv->bdrv_check_perm) { | |
2033 | return drv->bdrv_check_perm(bs, cumulative_perms, | |
2034 | cumulative_shared_perms, errp); | |
2035 | } | |
2036 | ||
78e421c9 | 2037 | /* Drivers that never have children can omit .bdrv_child_perm() */ |
33a610c3 | 2038 | if (!drv->bdrv_child_perm) { |
78e421c9 | 2039 | assert(QLIST_EMPTY(&bs->children)); |
33a610c3 KW |
2040 | return 0; |
2041 | } | |
2042 | ||
2043 | /* Check all children */ | |
2044 | QLIST_FOREACH(c, &bs->children, next) { | |
2045 | uint64_t cur_perm, cur_shared; | |
9eab1544 HR |
2046 | bool child_tighten_restr; |
2047 | ||
e5d8a406 | 2048 | bdrv_child_perm(bs, c->bs, c, c->role, q, |
ffd1a5a2 FZ |
2049 | cumulative_perms, cumulative_shared_perms, |
2050 | &cur_perm, &cur_shared); | |
9eab1544 HR |
2051 | ret = bdrv_child_check_perm(c, q, cur_perm, cur_shared, ignore_children, |
2052 | tighten_restrictions ? &child_tighten_restr | |
2053 | : NULL, | |
2054 | errp); | |
2055 | if (tighten_restrictions) { | |
2056 | *tighten_restrictions |= child_tighten_restr; | |
2057 | } | |
33a610c3 KW |
2058 | if (ret < 0) { |
2059 | return ret; | |
2060 | } | |
2061 | } | |
2062 | ||
2063 | return 0; | |
2064 | } | |
2065 | ||
2066 | /* | |
2067 | * Notifies drivers that after a previous bdrv_check_perm() call, the | |
2068 | * permission update is not performed and any preparations made for it (e.g. | |
2069 | * taken file locks) need to be undone. | |
2070 | * | |
2071 | * This function recursively notifies all child nodes. | |
2072 | */ | |
2073 | static void bdrv_abort_perm_update(BlockDriverState *bs) | |
2074 | { | |
2075 | BlockDriver *drv = bs->drv; | |
2076 | BdrvChild *c; | |
2077 | ||
2078 | if (!drv) { | |
2079 | return; | |
2080 | } | |
2081 | ||
2082 | if (drv->bdrv_abort_perm_update) { | |
2083 | drv->bdrv_abort_perm_update(bs); | |
2084 | } | |
2085 | ||
2086 | QLIST_FOREACH(c, &bs->children, next) { | |
2087 | bdrv_child_abort_perm_update(c); | |
2088 | } | |
2089 | } | |
2090 | ||
2091 | static void bdrv_set_perm(BlockDriverState *bs, uint64_t cumulative_perms, | |
2092 | uint64_t cumulative_shared_perms) | |
2093 | { | |
2094 | BlockDriver *drv = bs->drv; | |
2095 | BdrvChild *c; | |
2096 | ||
2097 | if (!drv) { | |
2098 | return; | |
2099 | } | |
2100 | ||
2101 | /* Update this node */ | |
2102 | if (drv->bdrv_set_perm) { | |
2103 | drv->bdrv_set_perm(bs, cumulative_perms, cumulative_shared_perms); | |
2104 | } | |
2105 | ||
78e421c9 | 2106 | /* Drivers that never have children can omit .bdrv_child_perm() */ |
33a610c3 | 2107 | if (!drv->bdrv_child_perm) { |
78e421c9 | 2108 | assert(QLIST_EMPTY(&bs->children)); |
33a610c3 KW |
2109 | return; |
2110 | } | |
2111 | ||
2112 | /* Update all children */ | |
2113 | QLIST_FOREACH(c, &bs->children, next) { | |
2114 | uint64_t cur_perm, cur_shared; | |
e5d8a406 | 2115 | bdrv_child_perm(bs, c->bs, c, c->role, NULL, |
ffd1a5a2 FZ |
2116 | cumulative_perms, cumulative_shared_perms, |
2117 | &cur_perm, &cur_shared); | |
33a610c3 KW |
2118 | bdrv_child_set_perm(c, cur_perm, cur_shared); |
2119 | } | |
2120 | } | |
2121 | ||
c7a0f2be KW |
2122 | void bdrv_get_cumulative_perm(BlockDriverState *bs, uint64_t *perm, |
2123 | uint64_t *shared_perm) | |
33a610c3 KW |
2124 | { |
2125 | BdrvChild *c; | |
2126 | uint64_t cumulative_perms = 0; | |
2127 | uint64_t cumulative_shared_perms = BLK_PERM_ALL; | |
2128 | ||
2129 | QLIST_FOREACH(c, &bs->parents, next_parent) { | |
2130 | cumulative_perms |= c->perm; | |
2131 | cumulative_shared_perms &= c->shared_perm; | |
2132 | } | |
2133 | ||
2134 | *perm = cumulative_perms; | |
2135 | *shared_perm = cumulative_shared_perms; | |
2136 | } | |
2137 | ||
d083319f KW |
2138 | static char *bdrv_child_user_desc(BdrvChild *c) |
2139 | { | |
bd86fb99 HR |
2140 | if (c->klass->get_parent_desc) { |
2141 | return c->klass->get_parent_desc(c); | |
d083319f KW |
2142 | } |
2143 | ||
2144 | return g_strdup("another user"); | |
2145 | } | |
2146 | ||
5176196c | 2147 | char *bdrv_perm_names(uint64_t perm) |
d083319f KW |
2148 | { |
2149 | struct perm_name { | |
2150 | uint64_t perm; | |
2151 | const char *name; | |
2152 | } permissions[] = { | |
2153 | { BLK_PERM_CONSISTENT_READ, "consistent read" }, | |
2154 | { BLK_PERM_WRITE, "write" }, | |
2155 | { BLK_PERM_WRITE_UNCHANGED, "write unchanged" }, | |
2156 | { BLK_PERM_RESIZE, "resize" }, | |
2157 | { BLK_PERM_GRAPH_MOD, "change children" }, | |
2158 | { 0, NULL } | |
2159 | }; | |
2160 | ||
e2a7423a | 2161 | GString *result = g_string_sized_new(30); |
d083319f KW |
2162 | struct perm_name *p; |
2163 | ||
2164 | for (p = permissions; p->name; p++) { | |
2165 | if (perm & p->perm) { | |
e2a7423a AG |
2166 | if (result->len > 0) { |
2167 | g_string_append(result, ", "); | |
2168 | } | |
2169 | g_string_append(result, p->name); | |
d083319f KW |
2170 | } |
2171 | } | |
2172 | ||
e2a7423a | 2173 | return g_string_free(result, FALSE); |
d083319f KW |
2174 | } |
2175 | ||
33a610c3 KW |
2176 | /* |
2177 | * Checks whether a new reference to @bs can be added if the new user requires | |
46181129 KW |
2178 | * @new_used_perm/@new_shared_perm as its permissions. If @ignore_children is |
2179 | * set, the BdrvChild objects in this list are ignored in the calculations; | |
2180 | * this allows checking permission updates for an existing reference. | |
33a610c3 | 2181 | * |
9eab1544 HR |
2182 | * See bdrv_check_perm() for the semantics of @tighten_restrictions. |
2183 | * | |
33a610c3 KW |
2184 | * Needs to be followed by a call to either bdrv_set_perm() or |
2185 | * bdrv_abort_perm_update(). */ | |
3121fb45 KW |
2186 | static int bdrv_check_update_perm(BlockDriverState *bs, BlockReopenQueue *q, |
2187 | uint64_t new_used_perm, | |
d5e6f437 | 2188 | uint64_t new_shared_perm, |
9eab1544 HR |
2189 | GSList *ignore_children, |
2190 | bool *tighten_restrictions, | |
2191 | Error **errp) | |
d5e6f437 KW |
2192 | { |
2193 | BdrvChild *c; | |
33a610c3 KW |
2194 | uint64_t cumulative_perms = new_used_perm; |
2195 | uint64_t cumulative_shared_perms = new_shared_perm; | |
d5e6f437 | 2196 | |
9eab1544 HR |
2197 | assert(!q || !tighten_restrictions); |
2198 | ||
d5e6f437 KW |
2199 | /* There is no reason why anyone couldn't tolerate write_unchanged */ |
2200 | assert(new_shared_perm & BLK_PERM_WRITE_UNCHANGED); | |
2201 | ||
2202 | QLIST_FOREACH(c, &bs->parents, next_parent) { | |
46181129 | 2203 | if (g_slist_find(ignore_children, c)) { |
d5e6f437 KW |
2204 | continue; |
2205 | } | |
2206 | ||
d083319f KW |
2207 | if ((new_used_perm & c->shared_perm) != new_used_perm) { |
2208 | char *user = bdrv_child_user_desc(c); | |
2209 | char *perm_names = bdrv_perm_names(new_used_perm & ~c->shared_perm); | |
9eab1544 HR |
2210 | |
2211 | if (tighten_restrictions) { | |
2212 | *tighten_restrictions = true; | |
2213 | } | |
2214 | ||
d083319f KW |
2215 | error_setg(errp, "Conflicts with use by %s as '%s', which does not " |
2216 | "allow '%s' on %s", | |
2217 | user, c->name, perm_names, bdrv_get_node_name(c->bs)); | |
2218 | g_free(user); | |
2219 | g_free(perm_names); | |
2220 | return -EPERM; | |
2221 | } | |
2222 | ||
2223 | if ((c->perm & new_shared_perm) != c->perm) { | |
2224 | char *user = bdrv_child_user_desc(c); | |
2225 | char *perm_names = bdrv_perm_names(c->perm & ~new_shared_perm); | |
9eab1544 HR |
2226 | |
2227 | if (tighten_restrictions) { | |
2228 | *tighten_restrictions = true; | |
2229 | } | |
2230 | ||
d083319f KW |
2231 | error_setg(errp, "Conflicts with use by %s as '%s', which uses " |
2232 | "'%s' on %s", | |
2233 | user, c->name, perm_names, bdrv_get_node_name(c->bs)); | |
2234 | g_free(user); | |
2235 | g_free(perm_names); | |
d5e6f437 KW |
2236 | return -EPERM; |
2237 | } | |
33a610c3 KW |
2238 | |
2239 | cumulative_perms |= c->perm; | |
2240 | cumulative_shared_perms &= c->shared_perm; | |
d5e6f437 KW |
2241 | } |
2242 | ||
3121fb45 | 2243 | return bdrv_check_perm(bs, q, cumulative_perms, cumulative_shared_perms, |
9eab1544 | 2244 | ignore_children, tighten_restrictions, errp); |
33a610c3 KW |
2245 | } |
2246 | ||
2247 | /* Needs to be followed by a call to either bdrv_child_set_perm() or | |
2248 | * bdrv_child_abort_perm_update(). */ | |
3121fb45 KW |
2249 | static int bdrv_child_check_perm(BdrvChild *c, BlockReopenQueue *q, |
2250 | uint64_t perm, uint64_t shared, | |
9eab1544 HR |
2251 | GSList *ignore_children, |
2252 | bool *tighten_restrictions, Error **errp) | |
33a610c3 | 2253 | { |
46181129 KW |
2254 | int ret; |
2255 | ||
2256 | ignore_children = g_slist_prepend(g_slist_copy(ignore_children), c); | |
9eab1544 HR |
2257 | ret = bdrv_check_update_perm(c->bs, q, perm, shared, ignore_children, |
2258 | tighten_restrictions, errp); | |
46181129 KW |
2259 | g_slist_free(ignore_children); |
2260 | ||
f962e961 VSO |
2261 | if (ret < 0) { |
2262 | return ret; | |
2263 | } | |
2264 | ||
2265 | if (!c->has_backup_perm) { | |
2266 | c->has_backup_perm = true; | |
2267 | c->backup_perm = c->perm; | |
2268 | c->backup_shared_perm = c->shared_perm; | |
2269 | } | |
2270 | /* | |
2271 | * Note: it's OK if c->has_backup_perm was already set, as we can find the | |
2272 | * same child twice during check_perm procedure | |
2273 | */ | |
2274 | ||
2275 | c->perm = perm; | |
2276 | c->shared_perm = shared; | |
2277 | ||
2278 | return 0; | |
33a610c3 KW |
2279 | } |
2280 | ||
c1cef672 | 2281 | static void bdrv_child_set_perm(BdrvChild *c, uint64_t perm, uint64_t shared) |
33a610c3 KW |
2282 | { |
2283 | uint64_t cumulative_perms, cumulative_shared_perms; | |
2284 | ||
f962e961 VSO |
2285 | c->has_backup_perm = false; |
2286 | ||
33a610c3 KW |
2287 | c->perm = perm; |
2288 | c->shared_perm = shared; | |
2289 | ||
2290 | bdrv_get_cumulative_perm(c->bs, &cumulative_perms, | |
2291 | &cumulative_shared_perms); | |
2292 | bdrv_set_perm(c->bs, cumulative_perms, cumulative_shared_perms); | |
2293 | } | |
2294 | ||
c1cef672 | 2295 | static void bdrv_child_abort_perm_update(BdrvChild *c) |
33a610c3 | 2296 | { |
f962e961 VSO |
2297 | if (c->has_backup_perm) { |
2298 | c->perm = c->backup_perm; | |
2299 | c->shared_perm = c->backup_shared_perm; | |
2300 | c->has_backup_perm = false; | |
2301 | } | |
2302 | ||
33a610c3 KW |
2303 | bdrv_abort_perm_update(c->bs); |
2304 | } | |
2305 | ||
2306 | int bdrv_child_try_set_perm(BdrvChild *c, uint64_t perm, uint64_t shared, | |
2307 | Error **errp) | |
2308 | { | |
1046779e | 2309 | Error *local_err = NULL; |
33a610c3 | 2310 | int ret; |
1046779e | 2311 | bool tighten_restrictions; |
33a610c3 | 2312 | |
1046779e HR |
2313 | ret = bdrv_child_check_perm(c, NULL, perm, shared, NULL, |
2314 | &tighten_restrictions, &local_err); | |
33a610c3 KW |
2315 | if (ret < 0) { |
2316 | bdrv_child_abort_perm_update(c); | |
1046779e HR |
2317 | if (tighten_restrictions) { |
2318 | error_propagate(errp, local_err); | |
2319 | } else { | |
2320 | /* | |
2321 | * Our caller may intend to only loosen restrictions and | |
2322 | * does not expect this function to fail. Errors are not | |
2323 | * fatal in such a case, so we can just hide them from our | |
2324 | * caller. | |
2325 | */ | |
2326 | error_free(local_err); | |
2327 | ret = 0; | |
2328 | } | |
33a610c3 KW |
2329 | return ret; |
2330 | } | |
2331 | ||
2332 | bdrv_child_set_perm(c, perm, shared); | |
2333 | ||
d5e6f437 KW |
2334 | return 0; |
2335 | } | |
2336 | ||
c1087f12 HR |
2337 | int bdrv_child_refresh_perms(BlockDriverState *bs, BdrvChild *c, Error **errp) |
2338 | { | |
2339 | uint64_t parent_perms, parent_shared; | |
2340 | uint64_t perms, shared; | |
2341 | ||
2342 | bdrv_get_cumulative_perm(bs, &parent_perms, &parent_shared); | |
e5d8a406 | 2343 | bdrv_child_perm(bs, c->bs, c, c->role, NULL, |
bf8e925e | 2344 | parent_perms, parent_shared, &perms, &shared); |
c1087f12 HR |
2345 | |
2346 | return bdrv_child_try_set_perm(c, perms, shared, errp); | |
2347 | } | |
2348 | ||
87278af1 HR |
2349 | /* |
2350 | * Default implementation for .bdrv_child_perm() for block filters: | |
2351 | * Forward CONSISTENT_READ, WRITE, WRITE_UNCHANGED, and RESIZE to the | |
2352 | * filtered child. | |
2353 | */ | |
2354 | static void bdrv_filter_default_perms(BlockDriverState *bs, BdrvChild *c, | |
87278af1 HR |
2355 | BdrvChildRole role, |
2356 | BlockReopenQueue *reopen_queue, | |
2357 | uint64_t perm, uint64_t shared, | |
2358 | uint64_t *nperm, uint64_t *nshared) | |
6a1b9ee1 | 2359 | { |
e444fa83 KW |
2360 | *nperm = perm & DEFAULT_PERM_PASSTHROUGH; |
2361 | *nshared = (shared & DEFAULT_PERM_PASSTHROUGH) | DEFAULT_PERM_UNCHANGED; | |
6a1b9ee1 KW |
2362 | } |
2363 | ||
70082db4 | 2364 | static void bdrv_default_perms_for_cow(BlockDriverState *bs, BdrvChild *c, |
70082db4 HR |
2365 | BdrvChildRole role, |
2366 | BlockReopenQueue *reopen_queue, | |
2367 | uint64_t perm, uint64_t shared, | |
2368 | uint64_t *nperm, uint64_t *nshared) | |
2369 | { | |
e5d8a406 | 2370 | assert(role & BDRV_CHILD_COW); |
70082db4 HR |
2371 | |
2372 | /* | |
2373 | * We want consistent read from backing files if the parent needs it. | |
2374 | * No other operations are performed on backing files. | |
2375 | */ | |
2376 | perm &= BLK_PERM_CONSISTENT_READ; | |
2377 | ||
2378 | /* | |
2379 | * If the parent can deal with changing data, we're okay with a | |
2380 | * writable and resizable backing file. | |
2381 | * TODO Require !(perm & BLK_PERM_CONSISTENT_READ), too? | |
2382 | */ | |
2383 | if (shared & BLK_PERM_WRITE) { | |
2384 | shared = BLK_PERM_WRITE | BLK_PERM_RESIZE; | |
2385 | } else { | |
2386 | shared = 0; | |
2387 | } | |
2388 | ||
2389 | shared |= BLK_PERM_CONSISTENT_READ | BLK_PERM_GRAPH_MOD | | |
2390 | BLK_PERM_WRITE_UNCHANGED; | |
2391 | ||
2392 | if (bs->open_flags & BDRV_O_INACTIVE) { | |
2393 | shared |= BLK_PERM_WRITE | BLK_PERM_RESIZE; | |
2394 | } | |
2395 | ||
2396 | *nperm = perm; | |
2397 | *nshared = shared; | |
2398 | } | |
2399 | ||
6f838a4b | 2400 | static void bdrv_default_perms_for_storage(BlockDriverState *bs, BdrvChild *c, |
6f838a4b HR |
2401 | BdrvChildRole role, |
2402 | BlockReopenQueue *reopen_queue, | |
2403 | uint64_t perm, uint64_t shared, | |
2404 | uint64_t *nperm, uint64_t *nshared) | |
2405 | { | |
2406 | int flags; | |
2407 | ||
e5d8a406 | 2408 | assert(role & (BDRV_CHILD_METADATA | BDRV_CHILD_DATA)); |
6f838a4b HR |
2409 | |
2410 | flags = bdrv_reopen_get_flags(reopen_queue, bs); | |
2411 | ||
2412 | /* | |
2413 | * Apart from the modifications below, the same permissions are | |
2414 | * forwarded and left alone as for filters | |
2415 | */ | |
e5d8a406 | 2416 | bdrv_filter_default_perms(bs, c, role, reopen_queue, |
6f838a4b HR |
2417 | perm, shared, &perm, &shared); |
2418 | ||
f889054f HR |
2419 | if (role & BDRV_CHILD_METADATA) { |
2420 | /* Format drivers may touch metadata even if the guest doesn't write */ | |
2421 | if (bdrv_is_writable_after_reopen(bs, reopen_queue)) { | |
2422 | perm |= BLK_PERM_WRITE | BLK_PERM_RESIZE; | |
2423 | } | |
2424 | ||
2425 | /* | |
2426 | * bs->file always needs to be consistent because of the | |
2427 | * metadata. We can never allow other users to resize or write | |
2428 | * to it. | |
2429 | */ | |
2430 | if (!(flags & BDRV_O_NO_IO)) { | |
2431 | perm |= BLK_PERM_CONSISTENT_READ; | |
2432 | } | |
2433 | shared &= ~(BLK_PERM_WRITE | BLK_PERM_RESIZE); | |
6f838a4b HR |
2434 | } |
2435 | ||
f889054f HR |
2436 | if (role & BDRV_CHILD_DATA) { |
2437 | /* | |
2438 | * Technically, everything in this block is a subset of the | |
2439 | * BDRV_CHILD_METADATA path taken above, and so this could | |
2440 | * be an "else if" branch. However, that is not obvious, and | |
2441 | * this function is not performance critical, therefore we let | |
2442 | * this be an independent "if". | |
2443 | */ | |
2444 | ||
2445 | /* | |
2446 | * We cannot allow other users to resize the file because the | |
2447 | * format driver might have some assumptions about the size | |
2448 | * (e.g. because it is stored in metadata, or because the file | |
2449 | * is split into fixed-size data files). | |
2450 | */ | |
2451 | shared &= ~BLK_PERM_RESIZE; | |
2452 | ||
2453 | /* | |
2454 | * WRITE_UNCHANGED often cannot be performed as such on the | |
2455 | * data file. For example, the qcow2 driver may still need to | |
2456 | * write copied clusters on copy-on-read. | |
2457 | */ | |
2458 | if (perm & BLK_PERM_WRITE_UNCHANGED) { | |
2459 | perm |= BLK_PERM_WRITE; | |
2460 | } | |
2461 | ||
2462 | /* | |
2463 | * If the data file is written to, the format driver may | |
2464 | * expect to be able to resize it by writing beyond the EOF. | |
2465 | */ | |
2466 | if (perm & BLK_PERM_WRITE) { | |
2467 | perm |= BLK_PERM_RESIZE; | |
2468 | } | |
6f838a4b | 2469 | } |
6f838a4b HR |
2470 | |
2471 | if (bs->open_flags & BDRV_O_INACTIVE) { | |
2472 | shared |= BLK_PERM_WRITE | BLK_PERM_RESIZE; | |
2473 | } | |
2474 | ||
2475 | *nperm = perm; | |
2476 | *nshared = shared; | |
2477 | } | |
2478 | ||
2519f549 | 2479 | void bdrv_default_perms(BlockDriverState *bs, BdrvChild *c, |
e5d8a406 | 2480 | BdrvChildRole role, BlockReopenQueue *reopen_queue, |
2519f549 HR |
2481 | uint64_t perm, uint64_t shared, |
2482 | uint64_t *nperm, uint64_t *nshared) | |
2483 | { | |
2519f549 HR |
2484 | if (role & BDRV_CHILD_FILTERED) { |
2485 | assert(!(role & (BDRV_CHILD_DATA | BDRV_CHILD_METADATA | | |
2486 | BDRV_CHILD_COW))); | |
e5d8a406 | 2487 | bdrv_filter_default_perms(bs, c, role, reopen_queue, |
2519f549 HR |
2488 | perm, shared, nperm, nshared); |
2489 | } else if (role & BDRV_CHILD_COW) { | |
2490 | assert(!(role & (BDRV_CHILD_DATA | BDRV_CHILD_METADATA))); | |
e5d8a406 | 2491 | bdrv_default_perms_for_cow(bs, c, role, reopen_queue, |
2519f549 HR |
2492 | perm, shared, nperm, nshared); |
2493 | } else if (role & (BDRV_CHILD_METADATA | BDRV_CHILD_DATA)) { | |
e5d8a406 | 2494 | bdrv_default_perms_for_storage(bs, c, role, reopen_queue, |
2519f549 HR |
2495 | perm, shared, nperm, nshared); |
2496 | } else { | |
2497 | g_assert_not_reached(); | |
2498 | } | |
2499 | } | |
2500 | ||
7b1d9c4d HR |
2501 | uint64_t bdrv_qapi_perm_to_blk_perm(BlockPermission qapi_perm) |
2502 | { | |
2503 | static const uint64_t permissions[] = { | |
2504 | [BLOCK_PERMISSION_CONSISTENT_READ] = BLK_PERM_CONSISTENT_READ, | |
2505 | [BLOCK_PERMISSION_WRITE] = BLK_PERM_WRITE, | |
2506 | [BLOCK_PERMISSION_WRITE_UNCHANGED] = BLK_PERM_WRITE_UNCHANGED, | |
2507 | [BLOCK_PERMISSION_RESIZE] = BLK_PERM_RESIZE, | |
2508 | [BLOCK_PERMISSION_GRAPH_MOD] = BLK_PERM_GRAPH_MOD, | |
2509 | }; | |
2510 | ||
2511 | QEMU_BUILD_BUG_ON(ARRAY_SIZE(permissions) != BLOCK_PERMISSION__MAX); | |
2512 | QEMU_BUILD_BUG_ON(1UL << ARRAY_SIZE(permissions) != BLK_PERM_ALL + 1); | |
2513 | ||
2514 | assert(qapi_perm < BLOCK_PERMISSION__MAX); | |
2515 | ||
2516 | return permissions[qapi_perm]; | |
2517 | } | |
2518 | ||
8ee03995 KW |
2519 | static void bdrv_replace_child_noperm(BdrvChild *child, |
2520 | BlockDriverState *new_bs) | |
e9740bc6 KW |
2521 | { |
2522 | BlockDriverState *old_bs = child->bs; | |
debc2927 HR |
2523 | int new_bs_quiesce_counter; |
2524 | int drain_saldo; | |
e9740bc6 | 2525 | |
2cad1ebe AG |
2526 | assert(!child->frozen); |
2527 | ||
bb2614e9 FZ |
2528 | if (old_bs && new_bs) { |
2529 | assert(bdrv_get_aio_context(old_bs) == bdrv_get_aio_context(new_bs)); | |
2530 | } | |
debc2927 HR |
2531 | |
2532 | new_bs_quiesce_counter = (new_bs ? new_bs->quiesce_counter : 0); | |
2533 | drain_saldo = new_bs_quiesce_counter - child->parent_quiesce_counter; | |
2534 | ||
2535 | /* | |
2536 | * If the new child node is drained but the old one was not, flush | |
2537 | * all outstanding requests to the old child node. | |
2538 | */ | |
bd86fb99 | 2539 | while (drain_saldo > 0 && child->klass->drained_begin) { |
debc2927 HR |
2540 | bdrv_parent_drained_begin_single(child, true); |
2541 | drain_saldo--; | |
2542 | } | |
2543 | ||
e9740bc6 | 2544 | if (old_bs) { |
d736f119 KW |
2545 | /* Detach first so that the recursive drain sections coming from @child |
2546 | * are already gone and we only end the drain sections that came from | |
2547 | * elsewhere. */ | |
bd86fb99 HR |
2548 | if (child->klass->detach) { |
2549 | child->klass->detach(child); | |
d736f119 | 2550 | } |
e9740bc6 KW |
2551 | QLIST_REMOVE(child, next_parent); |
2552 | } | |
36fe1331 KW |
2553 | |
2554 | child->bs = new_bs; | |
2555 | ||
e9740bc6 KW |
2556 | if (new_bs) { |
2557 | QLIST_INSERT_HEAD(&new_bs->parents, child, next_parent); | |
debc2927 HR |
2558 | |
2559 | /* | |
2560 | * Detaching the old node may have led to the new node's | |
2561 | * quiesce_counter having been decreased. Not a problem, we | |
2562 | * just need to recognize this here and then invoke | |
2563 | * drained_end appropriately more often. | |
2564 | */ | |
2565 | assert(new_bs->quiesce_counter <= new_bs_quiesce_counter); | |
2566 | drain_saldo += new_bs->quiesce_counter - new_bs_quiesce_counter; | |
33a610c3 | 2567 | |
d736f119 KW |
2568 | /* Attach only after starting new drained sections, so that recursive |
2569 | * drain sections coming from @child don't get an extra .drained_begin | |
2570 | * callback. */ | |
bd86fb99 HR |
2571 | if (child->klass->attach) { |
2572 | child->klass->attach(child); | |
8ee03995 KW |
2573 | } |
2574 | } | |
debc2927 HR |
2575 | |
2576 | /* | |
2577 | * If the old child node was drained but the new one is not, allow | |
2578 | * requests to come in only after the new node has been attached. | |
2579 | */ | |
bd86fb99 | 2580 | while (drain_saldo < 0 && child->klass->drained_end) { |
debc2927 HR |
2581 | bdrv_parent_drained_end_single(child); |
2582 | drain_saldo++; | |
2583 | } | |
8ee03995 | 2584 | } |
33a610c3 | 2585 | |
466787fb KW |
2586 | /* |
2587 | * Updates @child to change its reference to point to @new_bs, including | |
2588 | * checking and applying the necessary permisson updates both to the old node | |
2589 | * and to @new_bs. | |
2590 | * | |
2591 | * NULL is passed as @new_bs for removing the reference before freeing @child. | |
2592 | * | |
2593 | * If @new_bs is not NULL, bdrv_check_perm() must be called beforehand, as this | |
2594 | * function uses bdrv_set_perm() to update the permissions according to the new | |
2595 | * reference that @new_bs gets. | |
2596 | */ | |
2597 | static void bdrv_replace_child(BdrvChild *child, BlockDriverState *new_bs) | |
8ee03995 KW |
2598 | { |
2599 | BlockDriverState *old_bs = child->bs; | |
2600 | uint64_t perm, shared_perm; | |
2601 | ||
8aecf1d1 KW |
2602 | bdrv_replace_child_noperm(child, new_bs); |
2603 | ||
87ace5f8 HR |
2604 | /* |
2605 | * Start with the new node's permissions. If @new_bs is a (direct | |
2606 | * or indirect) child of @old_bs, we must complete the permission | |
2607 | * update on @new_bs before we loosen the restrictions on @old_bs. | |
2608 | * Otherwise, bdrv_check_perm() on @old_bs would re-initiate | |
2609 | * updating the permissions of @new_bs, and thus not purely loosen | |
2610 | * restrictions. | |
2611 | */ | |
2612 | if (new_bs) { | |
2613 | bdrv_get_cumulative_perm(new_bs, &perm, &shared_perm); | |
2614 | bdrv_set_perm(new_bs, perm, shared_perm); | |
2615 | } | |
2616 | ||
8ee03995 | 2617 | if (old_bs) { |
33a610c3 KW |
2618 | /* Update permissions for old node. This is guaranteed to succeed |
2619 | * because we're just taking a parent away, so we're loosening | |
2620 | * restrictions. */ | |
1046779e HR |
2621 | bool tighten_restrictions; |
2622 | int ret; | |
2623 | ||
33a610c3 | 2624 | bdrv_get_cumulative_perm(old_bs, &perm, &shared_perm); |
1046779e HR |
2625 | ret = bdrv_check_perm(old_bs, NULL, perm, shared_perm, NULL, |
2626 | &tighten_restrictions, NULL); | |
2627 | assert(tighten_restrictions == false); | |
2628 | if (ret < 0) { | |
2629 | /* We only tried to loosen restrictions, so errors are not fatal */ | |
2630 | bdrv_abort_perm_update(old_bs); | |
2631 | } else { | |
2632 | bdrv_set_perm(old_bs, perm, shared_perm); | |
2633 | } | |
ad943dcb KW |
2634 | |
2635 | /* When the parent requiring a non-default AioContext is removed, the | |
2636 | * node moves back to the main AioContext */ | |
2637 | bdrv_try_set_aio_context(old_bs, qemu_get_aio_context(), NULL); | |
e9740bc6 | 2638 | } |
e9740bc6 KW |
2639 | } |
2640 | ||
b441dc71 AG |
2641 | /* |
2642 | * This function steals the reference to child_bs from the caller. | |
2643 | * That reference is later dropped by bdrv_root_unref_child(). | |
2644 | * | |
2645 | * On failure NULL is returned, errp is set and the reference to | |
2646 | * child_bs is also dropped. | |
132ada80 KW |
2647 | * |
2648 | * The caller must hold the AioContext lock @child_bs, but not that of @ctx | |
2649 | * (unless @child_bs is already in @ctx). | |
b441dc71 | 2650 | */ |
f21d96d0 KW |
2651 | BdrvChild *bdrv_root_attach_child(BlockDriverState *child_bs, |
2652 | const char *child_name, | |
bd86fb99 | 2653 | const BdrvChildClass *child_class, |
258b7765 | 2654 | BdrvChildRole child_role, |
132ada80 | 2655 | AioContext *ctx, |
d5e6f437 KW |
2656 | uint64_t perm, uint64_t shared_perm, |
2657 | void *opaque, Error **errp) | |
df581792 | 2658 | { |
d5e6f437 | 2659 | BdrvChild *child; |
132ada80 | 2660 | Error *local_err = NULL; |
d5e6f437 KW |
2661 | int ret; |
2662 | ||
9eab1544 HR |
2663 | ret = bdrv_check_update_perm(child_bs, NULL, perm, shared_perm, NULL, NULL, |
2664 | errp); | |
d5e6f437 | 2665 | if (ret < 0) { |
33a610c3 | 2666 | bdrv_abort_perm_update(child_bs); |
b441dc71 | 2667 | bdrv_unref(child_bs); |
d5e6f437 KW |
2668 | return NULL; |
2669 | } | |
2670 | ||
2671 | child = g_new(BdrvChild, 1); | |
df581792 | 2672 | *child = (BdrvChild) { |
d5e6f437 KW |
2673 | .bs = NULL, |
2674 | .name = g_strdup(child_name), | |
bd86fb99 | 2675 | .klass = child_class, |
258b7765 | 2676 | .role = child_role, |
d5e6f437 KW |
2677 | .perm = perm, |
2678 | .shared_perm = shared_perm, | |
2679 | .opaque = opaque, | |
df581792 KW |
2680 | }; |
2681 | ||
132ada80 KW |
2682 | /* If the AioContexts don't match, first try to move the subtree of |
2683 | * child_bs into the AioContext of the new parent. If this doesn't work, | |
2684 | * try moving the parent into the AioContext of child_bs instead. */ | |
2685 | if (bdrv_get_aio_context(child_bs) != ctx) { | |
2686 | ret = bdrv_try_set_aio_context(child_bs, ctx, &local_err); | |
bd86fb99 | 2687 | if (ret < 0 && child_class->can_set_aio_ctx) { |
0beab811 | 2688 | GSList *ignore = g_slist_prepend(NULL, child); |
132ada80 | 2689 | ctx = bdrv_get_aio_context(child_bs); |
bd86fb99 | 2690 | if (child_class->can_set_aio_ctx(child, ctx, &ignore, NULL)) { |
132ada80 KW |
2691 | error_free(local_err); |
2692 | ret = 0; | |
2693 | g_slist_free(ignore); | |
0beab811 | 2694 | ignore = g_slist_prepend(NULL, child); |
bd86fb99 | 2695 | child_class->set_aio_ctx(child, ctx, &ignore); |
132ada80 KW |
2696 | } |
2697 | g_slist_free(ignore); | |
2698 | } | |
2699 | if (ret < 0) { | |
2700 | error_propagate(errp, local_err); | |
2701 | g_free(child); | |
2702 | bdrv_abort_perm_update(child_bs); | |
7a26df20 | 2703 | bdrv_unref(child_bs); |
132ada80 KW |
2704 | return NULL; |
2705 | } | |
2706 | } | |
2707 | ||
33a610c3 | 2708 | /* This performs the matching bdrv_set_perm() for the above check. */ |
466787fb | 2709 | bdrv_replace_child(child, child_bs); |
b4b059f6 KW |
2710 | |
2711 | return child; | |
df581792 KW |
2712 | } |
2713 | ||
b441dc71 AG |
2714 | /* |
2715 | * This function transfers the reference to child_bs from the caller | |
2716 | * to parent_bs. That reference is later dropped by parent_bs on | |
2717 | * bdrv_close() or if someone calls bdrv_unref_child(). | |
2718 | * | |
2719 | * On failure NULL is returned, errp is set and the reference to | |
2720 | * child_bs is also dropped. | |
132ada80 KW |
2721 | * |
2722 | * If @parent_bs and @child_bs are in different AioContexts, the caller must | |
2723 | * hold the AioContext lock for @child_bs, but not for @parent_bs. | |
b441dc71 | 2724 | */ |
98292c61 WC |
2725 | BdrvChild *bdrv_attach_child(BlockDriverState *parent_bs, |
2726 | BlockDriverState *child_bs, | |
2727 | const char *child_name, | |
bd86fb99 | 2728 | const BdrvChildClass *child_class, |
258b7765 | 2729 | BdrvChildRole child_role, |
8b2ff529 | 2730 | Error **errp) |
f21d96d0 | 2731 | { |
d5e6f437 | 2732 | BdrvChild *child; |
f68c598b KW |
2733 | uint64_t perm, shared_perm; |
2734 | ||
2735 | bdrv_get_cumulative_perm(parent_bs, &perm, &shared_perm); | |
2736 | ||
2737 | assert(parent_bs->drv); | |
e5d8a406 | 2738 | bdrv_child_perm(parent_bs, child_bs, NULL, child_role, NULL, |
ffd1a5a2 | 2739 | perm, shared_perm, &perm, &shared_perm); |
d5e6f437 | 2740 | |
bd86fb99 | 2741 | child = bdrv_root_attach_child(child_bs, child_name, child_class, |
258b7765 | 2742 | child_role, bdrv_get_aio_context(parent_bs), |
f68c598b | 2743 | perm, shared_perm, parent_bs, errp); |
d5e6f437 KW |
2744 | if (child == NULL) { |
2745 | return NULL; | |
2746 | } | |
2747 | ||
f21d96d0 KW |
2748 | QLIST_INSERT_HEAD(&parent_bs->children, child, next); |
2749 | return child; | |
2750 | } | |
2751 | ||
3f09bfbc | 2752 | static void bdrv_detach_child(BdrvChild *child) |
33a60407 | 2753 | { |
195ed8cb | 2754 | QLIST_SAFE_REMOVE(child, next); |
e9740bc6 | 2755 | |
466787fb | 2756 | bdrv_replace_child(child, NULL); |
e9740bc6 | 2757 | |
260fecf1 | 2758 | g_free(child->name); |
33a60407 KW |
2759 | g_free(child); |
2760 | } | |
2761 | ||
f21d96d0 | 2762 | void bdrv_root_unref_child(BdrvChild *child) |
33a60407 | 2763 | { |
779020cb KW |
2764 | BlockDriverState *child_bs; |
2765 | ||
f21d96d0 KW |
2766 | child_bs = child->bs; |
2767 | bdrv_detach_child(child); | |
2768 | bdrv_unref(child_bs); | |
2769 | } | |
2770 | ||
3cf746b3 HR |
2771 | /** |
2772 | * Clear all inherits_from pointers from children and grandchildren of | |
2773 | * @root that point to @root, where necessary. | |
2774 | */ | |
2775 | static void bdrv_unset_inherits_from(BlockDriverState *root, BdrvChild *child) | |
f21d96d0 | 2776 | { |
3cf746b3 | 2777 | BdrvChild *c; |
4e4bf5c4 | 2778 | |
3cf746b3 HR |
2779 | if (child->bs->inherits_from == root) { |
2780 | /* | |
2781 | * Remove inherits_from only when the last reference between root and | |
2782 | * child->bs goes away. | |
2783 | */ | |
2784 | QLIST_FOREACH(c, &root->children, next) { | |
4e4bf5c4 KW |
2785 | if (c != child && c->bs == child->bs) { |
2786 | break; | |
2787 | } | |
2788 | } | |
2789 | if (c == NULL) { | |
2790 | child->bs->inherits_from = NULL; | |
2791 | } | |
33a60407 KW |
2792 | } |
2793 | ||
3cf746b3 HR |
2794 | QLIST_FOREACH(c, &child->bs->children, next) { |
2795 | bdrv_unset_inherits_from(root, c); | |
2796 | } | |
2797 | } | |
2798 | ||
2799 | void bdrv_unref_child(BlockDriverState *parent, BdrvChild *child) | |
2800 | { | |
2801 | if (child == NULL) { | |
2802 | return; | |
2803 | } | |
2804 | ||
2805 | bdrv_unset_inherits_from(parent, child); | |
f21d96d0 | 2806 | bdrv_root_unref_child(child); |
33a60407 KW |
2807 | } |
2808 | ||
5c8cab48 KW |
2809 | |
2810 | static void bdrv_parent_cb_change_media(BlockDriverState *bs, bool load) | |
2811 | { | |
2812 | BdrvChild *c; | |
2813 | QLIST_FOREACH(c, &bs->parents, next_parent) { | |
bd86fb99 HR |
2814 | if (c->klass->change_media) { |
2815 | c->klass->change_media(c, load); | |
5c8cab48 KW |
2816 | } |
2817 | } | |
2818 | } | |
2819 | ||
0065c455 AG |
2820 | /* Return true if you can reach parent going through child->inherits_from |
2821 | * recursively. If parent or child are NULL, return false */ | |
2822 | static bool bdrv_inherits_from_recursive(BlockDriverState *child, | |
2823 | BlockDriverState *parent) | |
2824 | { | |
2825 | while (child && child != parent) { | |
2826 | child = child->inherits_from; | |
2827 | } | |
2828 | ||
2829 | return child != NULL; | |
2830 | } | |
2831 | ||
25191e5f HR |
2832 | /* |
2833 | * Return the BdrvChildRole for @bs's backing child. bs->backing is | |
2834 | * mostly used for COW backing children (role = COW), but also for | |
2835 | * filtered children (role = FILTERED | PRIMARY). | |
2836 | */ | |
2837 | static BdrvChildRole bdrv_backing_role(BlockDriverState *bs) | |
2838 | { | |
2839 | if (bs->drv && bs->drv->is_filter) { | |
2840 | return BDRV_CHILD_FILTERED | BDRV_CHILD_PRIMARY; | |
2841 | } else { | |
2842 | return BDRV_CHILD_COW; | |
2843 | } | |
2844 | } | |
2845 | ||
5db15a57 KW |
2846 | /* |
2847 | * Sets the backing file link of a BDS. A new reference is created; callers | |
2848 | * which don't need their own reference any more must call bdrv_unref(). | |
2849 | */ | |
12fa4af6 KW |
2850 | void bdrv_set_backing_hd(BlockDriverState *bs, BlockDriverState *backing_hd, |
2851 | Error **errp) | |
8d24cce1 | 2852 | { |
0065c455 AG |
2853 | bool update_inherits_from = bdrv_chain_contains(bs, backing_hd) && |
2854 | bdrv_inherits_from_recursive(backing_hd, bs); | |
2855 | ||
2cad1ebe AG |
2856 | if (bdrv_is_backing_chain_frozen(bs, backing_bs(bs), errp)) { |
2857 | return; | |
2858 | } | |
2859 | ||
5db15a57 KW |
2860 | if (backing_hd) { |
2861 | bdrv_ref(backing_hd); | |
2862 | } | |
8d24cce1 | 2863 | |
760e0063 | 2864 | if (bs->backing) { |
5db15a57 | 2865 | bdrv_unref_child(bs, bs->backing); |
6e57963a | 2866 | bs->backing = NULL; |
826b6ca0 FZ |
2867 | } |
2868 | ||
8d24cce1 FZ |
2869 | if (!backing_hd) { |
2870 | goto out; | |
2871 | } | |
12fa4af6 | 2872 | |
25191e5f HR |
2873 | bs->backing = bdrv_attach_child(bs, backing_hd, "backing", &child_of_bds, |
2874 | bdrv_backing_role(bs), errp); | |
0065c455 AG |
2875 | /* If backing_hd was already part of bs's backing chain, and |
2876 | * inherits_from pointed recursively to bs then let's update it to | |
2877 | * point directly to bs (else it will become NULL). */ | |
b441dc71 | 2878 | if (bs->backing && update_inherits_from) { |
0065c455 AG |
2879 | backing_hd->inherits_from = bs; |
2880 | } | |
826b6ca0 | 2881 | |
8d24cce1 | 2882 | out: |
3baca891 | 2883 | bdrv_refresh_limits(bs, NULL); |
8d24cce1 FZ |
2884 | } |
2885 | ||
31ca6d07 KW |
2886 | /* |
2887 | * Opens the backing file for a BlockDriverState if not yet open | |
2888 | * | |
d9b7b057 KW |
2889 | * bdref_key specifies the key for the image's BlockdevRef in the options QDict. |
2890 | * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict | |
2891 | * itself, all options starting with "${bdref_key}." are considered part of the | |
2892 | * BlockdevRef. | |
2893 | * | |
2894 | * TODO Can this be unified with bdrv_open_image()? | |
31ca6d07 | 2895 | */ |
d9b7b057 KW |
2896 | int bdrv_open_backing_file(BlockDriverState *bs, QDict *parent_options, |
2897 | const char *bdref_key, Error **errp) | |
9156df12 | 2898 | { |
6b6833c1 | 2899 | char *backing_filename = NULL; |
d9b7b057 KW |
2900 | char *bdref_key_dot; |
2901 | const char *reference = NULL; | |
317fc44e | 2902 | int ret = 0; |
998c2019 | 2903 | bool implicit_backing = false; |
8d24cce1 | 2904 | BlockDriverState *backing_hd; |
d9b7b057 KW |
2905 | QDict *options; |
2906 | QDict *tmp_parent_options = NULL; | |
34b5d2c6 | 2907 | Error *local_err = NULL; |
9156df12 | 2908 | |
760e0063 | 2909 | if (bs->backing != NULL) { |
1ba4b6a5 | 2910 | goto free_exit; |
9156df12 PB |
2911 | } |
2912 | ||
31ca6d07 | 2913 | /* NULL means an empty set of options */ |
d9b7b057 KW |
2914 | if (parent_options == NULL) { |
2915 | tmp_parent_options = qdict_new(); | |
2916 | parent_options = tmp_parent_options; | |
31ca6d07 KW |
2917 | } |
2918 | ||
9156df12 | 2919 | bs->open_flags &= ~BDRV_O_NO_BACKING; |
d9b7b057 KW |
2920 | |
2921 | bdref_key_dot = g_strdup_printf("%s.", bdref_key); | |
2922 | qdict_extract_subqdict(parent_options, &options, bdref_key_dot); | |
2923 | g_free(bdref_key_dot); | |
2924 | ||
129c7d1c MA |
2925 | /* |
2926 | * Caution: while qdict_get_try_str() is fine, getting non-string | |
2927 | * types would require more care. When @parent_options come from | |
2928 | * -blockdev or blockdev_add, its members are typed according to | |
2929 | * the QAPI schema, but when they come from -drive, they're all | |
2930 | * QString. | |
2931 | */ | |
d9b7b057 KW |
2932 | reference = qdict_get_try_str(parent_options, bdref_key); |
2933 | if (reference || qdict_haskey(options, "file.filename")) { | |
6b6833c1 | 2934 | /* keep backing_filename NULL */ |
1cb6f506 | 2935 | } else if (bs->backing_file[0] == '\0' && qdict_size(options) == 0) { |
cb3e7f08 | 2936 | qobject_unref(options); |
1ba4b6a5 | 2937 | goto free_exit; |
dbecebdd | 2938 | } else { |
998c2019 HR |
2939 | if (qdict_size(options) == 0) { |
2940 | /* If the user specifies options that do not modify the | |
2941 | * backing file's behavior, we might still consider it the | |
2942 | * implicit backing file. But it's easier this way, and | |
2943 | * just specifying some of the backing BDS's options is | |
2944 | * only possible with -drive anyway (otherwise the QAPI | |
2945 | * schema forces the user to specify everything). */ | |
2946 | implicit_backing = !strcmp(bs->auto_backing_file, bs->backing_file); | |
2947 | } | |
2948 | ||
6b6833c1 | 2949 | backing_filename = bdrv_get_full_backing_filename(bs, &local_err); |
9f07429e HR |
2950 | if (local_err) { |
2951 | ret = -EINVAL; | |
2952 | error_propagate(errp, local_err); | |
cb3e7f08 | 2953 | qobject_unref(options); |
9f07429e HR |
2954 | goto free_exit; |
2955 | } | |
9156df12 PB |
2956 | } |
2957 | ||
8ee79e70 KW |
2958 | if (!bs->drv || !bs->drv->supports_backing) { |
2959 | ret = -EINVAL; | |
2960 | error_setg(errp, "Driver doesn't support backing files"); | |
cb3e7f08 | 2961 | qobject_unref(options); |
8ee79e70 KW |
2962 | goto free_exit; |
2963 | } | |
2964 | ||
6bff597b PK |
2965 | if (!reference && |
2966 | bs->backing_format[0] != '\0' && !qdict_haskey(options, "driver")) { | |
46f5ac20 | 2967 | qdict_put_str(options, "driver", bs->backing_format); |
9156df12 PB |
2968 | } |
2969 | ||
6b6833c1 | 2970 | backing_hd = bdrv_open_inherit(backing_filename, reference, options, 0, bs, |
25191e5f | 2971 | &child_of_bds, bdrv_backing_role(bs), errp); |
5b363937 | 2972 | if (!backing_hd) { |
9156df12 | 2973 | bs->open_flags |= BDRV_O_NO_BACKING; |
e43bfd9c | 2974 | error_prepend(errp, "Could not open backing file: "); |
5b363937 | 2975 | ret = -EINVAL; |
1ba4b6a5 | 2976 | goto free_exit; |
9156df12 | 2977 | } |
df581792 | 2978 | |
998c2019 HR |
2979 | if (implicit_backing) { |
2980 | bdrv_refresh_filename(backing_hd); | |
2981 | pstrcpy(bs->auto_backing_file, sizeof(bs->auto_backing_file), | |
2982 | backing_hd->filename); | |
2983 | } | |
2984 | ||
5db15a57 KW |
2985 | /* Hook up the backing file link; drop our reference, bs owns the |
2986 | * backing_hd reference now */ | |
12fa4af6 | 2987 | bdrv_set_backing_hd(bs, backing_hd, &local_err); |
5db15a57 | 2988 | bdrv_unref(backing_hd); |
12fa4af6 | 2989 | if (local_err) { |
8cd1a3e4 | 2990 | error_propagate(errp, local_err); |
12fa4af6 KW |
2991 | ret = -EINVAL; |
2992 | goto free_exit; | |
2993 | } | |
d80ac658 | 2994 | |
d9b7b057 KW |
2995 | qdict_del(parent_options, bdref_key); |
2996 | ||
1ba4b6a5 BC |
2997 | free_exit: |
2998 | g_free(backing_filename); | |
cb3e7f08 | 2999 | qobject_unref(tmp_parent_options); |
1ba4b6a5 | 3000 | return ret; |
9156df12 PB |
3001 | } |
3002 | ||
2d6b86af KW |
3003 | static BlockDriverState * |
3004 | bdrv_open_child_bs(const char *filename, QDict *options, const char *bdref_key, | |
bd86fb99 | 3005 | BlockDriverState *parent, const BdrvChildClass *child_class, |
272c02ea | 3006 | BdrvChildRole child_role, bool allow_none, Error **errp) |
da557aac | 3007 | { |
2d6b86af | 3008 | BlockDriverState *bs = NULL; |
da557aac | 3009 | QDict *image_options; |
da557aac HR |
3010 | char *bdref_key_dot; |
3011 | const char *reference; | |
3012 | ||
bd86fb99 | 3013 | assert(child_class != NULL); |
f67503e5 | 3014 | |
da557aac HR |
3015 | bdref_key_dot = g_strdup_printf("%s.", bdref_key); |
3016 | qdict_extract_subqdict(options, &image_options, bdref_key_dot); | |
3017 | g_free(bdref_key_dot); | |
3018 | ||
129c7d1c MA |
3019 | /* |
3020 | * Caution: while qdict_get_try_str() is fine, getting non-string | |
3021 | * types would require more care. When @options come from | |
3022 | * -blockdev or blockdev_add, its members are typed according to | |
3023 | * the QAPI schema, but when they come from -drive, they're all | |
3024 | * QString. | |
3025 | */ | |
da557aac HR |
3026 | reference = qdict_get_try_str(options, bdref_key); |
3027 | if (!filename && !reference && !qdict_size(image_options)) { | |
b4b059f6 | 3028 | if (!allow_none) { |
da557aac HR |
3029 | error_setg(errp, "A block device must be specified for \"%s\"", |
3030 | bdref_key); | |
da557aac | 3031 | } |
cb3e7f08 | 3032 | qobject_unref(image_options); |
da557aac HR |
3033 | goto done; |
3034 | } | |
3035 | ||
5b363937 | 3036 | bs = bdrv_open_inherit(filename, reference, image_options, 0, |
272c02ea | 3037 | parent, child_class, child_role, errp); |
5b363937 | 3038 | if (!bs) { |
df581792 KW |
3039 | goto done; |
3040 | } | |
3041 | ||
da557aac HR |
3042 | done: |
3043 | qdict_del(options, bdref_key); | |
2d6b86af KW |
3044 | return bs; |
3045 | } | |
3046 | ||
3047 | /* | |
3048 | * Opens a disk image whose options are given as BlockdevRef in another block | |
3049 | * device's options. | |
3050 | * | |
3051 | * If allow_none is true, no image will be opened if filename is false and no | |
3052 | * BlockdevRef is given. NULL will be returned, but errp remains unset. | |
3053 | * | |
3054 | * bdrev_key specifies the key for the image's BlockdevRef in the options QDict. | |
3055 | * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict | |
3056 | * itself, all options starting with "${bdref_key}." are considered part of the | |
3057 | * BlockdevRef. | |
3058 | * | |
3059 | * The BlockdevRef will be removed from the options QDict. | |
3060 | */ | |
3061 | BdrvChild *bdrv_open_child(const char *filename, | |
3062 | QDict *options, const char *bdref_key, | |
3063 | BlockDriverState *parent, | |
bd86fb99 | 3064 | const BdrvChildClass *child_class, |
258b7765 | 3065 | BdrvChildRole child_role, |
2d6b86af KW |
3066 | bool allow_none, Error **errp) |
3067 | { | |
3068 | BlockDriverState *bs; | |
3069 | ||
bd86fb99 | 3070 | bs = bdrv_open_child_bs(filename, options, bdref_key, parent, child_class, |
272c02ea | 3071 | child_role, allow_none, errp); |
2d6b86af KW |
3072 | if (bs == NULL) { |
3073 | return NULL; | |
3074 | } | |
3075 | ||
258b7765 HR |
3076 | return bdrv_attach_child(parent, bs, bdref_key, child_class, child_role, |
3077 | errp); | |
b4b059f6 KW |
3078 | } |
3079 | ||
bd86fb99 HR |
3080 | /* |
3081 | * TODO Future callers may need to specify parent/child_class in order for | |
3082 | * option inheritance to work. Existing callers use it for the root node. | |
3083 | */ | |
e1d74bc6 KW |
3084 | BlockDriverState *bdrv_open_blockdev_ref(BlockdevRef *ref, Error **errp) |
3085 | { | |
3086 | BlockDriverState *bs = NULL; | |
e1d74bc6 KW |
3087 | QObject *obj = NULL; |
3088 | QDict *qdict = NULL; | |
3089 | const char *reference = NULL; | |
3090 | Visitor *v = NULL; | |
3091 | ||
3092 | if (ref->type == QTYPE_QSTRING) { | |
3093 | reference = ref->u.reference; | |
3094 | } else { | |
3095 | BlockdevOptions *options = &ref->u.definition; | |
3096 | assert(ref->type == QTYPE_QDICT); | |
3097 | ||
3098 | v = qobject_output_visitor_new(&obj); | |
1f584248 | 3099 | visit_type_BlockdevOptions(v, NULL, &options, &error_abort); |
e1d74bc6 KW |
3100 | visit_complete(v, &obj); |
3101 | ||
7dc847eb | 3102 | qdict = qobject_to(QDict, obj); |
e1d74bc6 KW |
3103 | qdict_flatten(qdict); |
3104 | ||
3105 | /* bdrv_open_inherit() defaults to the values in bdrv_flags (for | |
3106 | * compatibility with other callers) rather than what we want as the | |
3107 | * real defaults. Apply the defaults here instead. */ | |
3108 | qdict_set_default_str(qdict, BDRV_OPT_CACHE_DIRECT, "off"); | |
3109 | qdict_set_default_str(qdict, BDRV_OPT_CACHE_NO_FLUSH, "off"); | |
3110 | qdict_set_default_str(qdict, BDRV_OPT_READ_ONLY, "off"); | |
e35bdc12 KW |
3111 | qdict_set_default_str(qdict, BDRV_OPT_AUTO_READ_ONLY, "off"); |
3112 | ||
e1d74bc6 KW |
3113 | } |
3114 | ||
272c02ea | 3115 | bs = bdrv_open_inherit(NULL, reference, qdict, 0, NULL, NULL, 0, errp); |
e1d74bc6 | 3116 | obj = NULL; |
cb3e7f08 | 3117 | qobject_unref(obj); |
e1d74bc6 KW |
3118 | visit_free(v); |
3119 | return bs; | |
3120 | } | |
3121 | ||
66836189 HR |
3122 | static BlockDriverState *bdrv_append_temp_snapshot(BlockDriverState *bs, |
3123 | int flags, | |
3124 | QDict *snapshot_options, | |
3125 | Error **errp) | |
b998875d KW |
3126 | { |
3127 | /* TODO: extra byte is a hack to ensure MAX_PATH space on Windows. */ | |
1ba4b6a5 | 3128 | char *tmp_filename = g_malloc0(PATH_MAX + 1); |
b998875d | 3129 | int64_t total_size; |
83d0521a | 3130 | QemuOpts *opts = NULL; |
ff6ed714 | 3131 | BlockDriverState *bs_snapshot = NULL; |
b2c2832c | 3132 | Error *local_err = NULL; |
b998875d KW |
3133 | int ret; |
3134 | ||
3135 | /* if snapshot, we create a temporary backing file and open it | |
3136 | instead of opening 'filename' directly */ | |
3137 | ||
3138 | /* Get the required size from the image */ | |
f187743a KW |
3139 | total_size = bdrv_getlength(bs); |
3140 | if (total_size < 0) { | |
3141 | error_setg_errno(errp, -total_size, "Could not get image size"); | |
1ba4b6a5 | 3142 | goto out; |
f187743a | 3143 | } |
b998875d KW |
3144 | |
3145 | /* Create the temporary image */ | |
1ba4b6a5 | 3146 | ret = get_tmp_filename(tmp_filename, PATH_MAX + 1); |
b998875d KW |
3147 | if (ret < 0) { |
3148 | error_setg_errno(errp, -ret, "Could not get temporary filename"); | |
1ba4b6a5 | 3149 | goto out; |
b998875d KW |
3150 | } |
3151 | ||
ef810437 | 3152 | opts = qemu_opts_create(bdrv_qcow2.create_opts, NULL, 0, |
c282e1fd | 3153 | &error_abort); |
39101f25 | 3154 | qemu_opt_set_number(opts, BLOCK_OPT_SIZE, total_size, &error_abort); |
e43bfd9c | 3155 | ret = bdrv_create(&bdrv_qcow2, tmp_filename, opts, errp); |
83d0521a | 3156 | qemu_opts_del(opts); |
b998875d | 3157 | if (ret < 0) { |
e43bfd9c MA |
3158 | error_prepend(errp, "Could not create temporary overlay '%s': ", |
3159 | tmp_filename); | |
1ba4b6a5 | 3160 | goto out; |
b998875d KW |
3161 | } |
3162 | ||
73176bee | 3163 | /* Prepare options QDict for the temporary file */ |
46f5ac20 EB |
3164 | qdict_put_str(snapshot_options, "file.driver", "file"); |
3165 | qdict_put_str(snapshot_options, "file.filename", tmp_filename); | |
3166 | qdict_put_str(snapshot_options, "driver", "qcow2"); | |
b998875d | 3167 | |
5b363937 | 3168 | bs_snapshot = bdrv_open(NULL, NULL, snapshot_options, flags, errp); |
73176bee | 3169 | snapshot_options = NULL; |
5b363937 | 3170 | if (!bs_snapshot) { |
1ba4b6a5 | 3171 | goto out; |
b998875d KW |
3172 | } |
3173 | ||
ff6ed714 EB |
3174 | /* bdrv_append() consumes a strong reference to bs_snapshot |
3175 | * (i.e. it will call bdrv_unref() on it) even on error, so in | |
3176 | * order to be able to return one, we have to increase | |
3177 | * bs_snapshot's refcount here */ | |
66836189 | 3178 | bdrv_ref(bs_snapshot); |
b2c2832c KW |
3179 | bdrv_append(bs_snapshot, bs, &local_err); |
3180 | if (local_err) { | |
3181 | error_propagate(errp, local_err); | |
ff6ed714 | 3182 | bs_snapshot = NULL; |
b2c2832c KW |
3183 | goto out; |
3184 | } | |
1ba4b6a5 BC |
3185 | |
3186 | out: | |
cb3e7f08 | 3187 | qobject_unref(snapshot_options); |
1ba4b6a5 | 3188 | g_free(tmp_filename); |
ff6ed714 | 3189 | return bs_snapshot; |
b998875d KW |
3190 | } |
3191 | ||
b6ce07aa KW |
3192 | /* |
3193 | * Opens a disk image (raw, qcow2, vmdk, ...) | |
de9c0cec KW |
3194 | * |
3195 | * options is a QDict of options to pass to the block drivers, or NULL for an | |
3196 | * empty set of options. The reference to the QDict belongs to the block layer | |
3197 | * after the call (even on failure), so if the caller intends to reuse the | |
cb3e7f08 | 3198 | * dictionary, it needs to use qobject_ref() before calling bdrv_open. |
f67503e5 HR |
3199 | * |
3200 | * If *pbs is NULL, a new BDS will be created with a pointer to it stored there. | |
3201 | * If it is not NULL, the referenced BDS will be reused. | |
ddf5636d HR |
3202 | * |
3203 | * The reference parameter may be used to specify an existing block device which | |
3204 | * should be opened. If specified, neither options nor a filename may be given, | |
3205 | * nor can an existing BDS be reused (that is, *pbs has to be NULL). | |
b6ce07aa | 3206 | */ |
5b363937 HR |
3207 | static BlockDriverState *bdrv_open_inherit(const char *filename, |
3208 | const char *reference, | |
3209 | QDict *options, int flags, | |
3210 | BlockDriverState *parent, | |
bd86fb99 | 3211 | const BdrvChildClass *child_class, |
272c02ea | 3212 | BdrvChildRole child_role, |
5b363937 | 3213 | Error **errp) |
ea2384d3 | 3214 | { |
b6ce07aa | 3215 | int ret; |
5696c6e3 | 3216 | BlockBackend *file = NULL; |
9a4f4c31 | 3217 | BlockDriverState *bs; |
ce343771 | 3218 | BlockDriver *drv = NULL; |
2f624b80 | 3219 | BdrvChild *child; |
74fe54f2 | 3220 | const char *drvname; |
3e8c2e57 | 3221 | const char *backing; |
34b5d2c6 | 3222 | Error *local_err = NULL; |
73176bee | 3223 | QDict *snapshot_options = NULL; |
b1e6fc08 | 3224 | int snapshot_flags = 0; |
712e7874 | 3225 | |
bd86fb99 HR |
3226 | assert(!child_class || !flags); |
3227 | assert(!child_class == !parent); | |
f67503e5 | 3228 | |
ddf5636d HR |
3229 | if (reference) { |
3230 | bool options_non_empty = options ? qdict_size(options) : false; | |
cb3e7f08 | 3231 | qobject_unref(options); |
ddf5636d | 3232 | |
ddf5636d HR |
3233 | if (filename || options_non_empty) { |
3234 | error_setg(errp, "Cannot reference an existing block device with " | |
3235 | "additional options or a new filename"); | |
5b363937 | 3236 | return NULL; |
ddf5636d HR |
3237 | } |
3238 | ||
3239 | bs = bdrv_lookup_bs(reference, reference, errp); | |
3240 | if (!bs) { | |
5b363937 | 3241 | return NULL; |
ddf5636d | 3242 | } |
76b22320 | 3243 | |
ddf5636d | 3244 | bdrv_ref(bs); |
5b363937 | 3245 | return bs; |
ddf5636d HR |
3246 | } |
3247 | ||
5b363937 | 3248 | bs = bdrv_new(); |
f67503e5 | 3249 | |
de9c0cec KW |
3250 | /* NULL means an empty set of options */ |
3251 | if (options == NULL) { | |
3252 | options = qdict_new(); | |
3253 | } | |
3254 | ||
145f598e | 3255 | /* json: syntax counts as explicit options, as if in the QDict */ |
de3b53f0 KW |
3256 | parse_json_protocol(options, &filename, &local_err); |
3257 | if (local_err) { | |
de3b53f0 KW |
3258 | goto fail; |
3259 | } | |
3260 | ||
145f598e KW |
3261 | bs->explicit_options = qdict_clone_shallow(options); |
3262 | ||
bd86fb99 | 3263 | if (child_class) { |
3cdc69d3 HR |
3264 | bool parent_is_format; |
3265 | ||
3266 | if (parent->drv) { | |
3267 | parent_is_format = parent->drv->is_format; | |
3268 | } else { | |
3269 | /* | |
3270 | * parent->drv is not set yet because this node is opened for | |
3271 | * (potential) format probing. That means that @parent is going | |
3272 | * to be a format node. | |
3273 | */ | |
3274 | parent_is_format = true; | |
3275 | } | |
3276 | ||
bddcec37 | 3277 | bs->inherits_from = parent; |
3cdc69d3 HR |
3278 | child_class->inherit_options(child_role, parent_is_format, |
3279 | &flags, options, | |
bd86fb99 | 3280 | parent->open_flags, parent->options); |
f3930ed0 KW |
3281 | } |
3282 | ||
de3b53f0 | 3283 | ret = bdrv_fill_options(&options, filename, &flags, &local_err); |
dfde483e | 3284 | if (ret < 0) { |
462f5bcf KW |
3285 | goto fail; |
3286 | } | |
3287 | ||
129c7d1c MA |
3288 | /* |
3289 | * Set the BDRV_O_RDWR and BDRV_O_ALLOW_RDWR flags. | |
3290 | * Caution: getting a boolean member of @options requires care. | |
3291 | * When @options come from -blockdev or blockdev_add, members are | |
3292 | * typed according to the QAPI schema, but when they come from | |
3293 | * -drive, they're all QString. | |
3294 | */ | |
f87a0e29 AG |
3295 | if (g_strcmp0(qdict_get_try_str(options, BDRV_OPT_READ_ONLY), "on") && |
3296 | !qdict_get_try_bool(options, BDRV_OPT_READ_ONLY, false)) { | |
3297 | flags |= (BDRV_O_RDWR | BDRV_O_ALLOW_RDWR); | |
3298 | } else { | |
3299 | flags &= ~BDRV_O_RDWR; | |
14499ea5 AG |
3300 | } |
3301 | ||
3302 | if (flags & BDRV_O_SNAPSHOT) { | |
3303 | snapshot_options = qdict_new(); | |
3304 | bdrv_temp_snapshot_options(&snapshot_flags, snapshot_options, | |
3305 | flags, options); | |
f87a0e29 AG |
3306 | /* Let bdrv_backing_options() override "read-only" */ |
3307 | qdict_del(options, BDRV_OPT_READ_ONLY); | |
00ff7ffd HR |
3308 | bdrv_inherited_options(BDRV_CHILD_COW, true, |
3309 | &flags, options, flags, options); | |
14499ea5 AG |
3310 | } |
3311 | ||
62392ebb KW |
3312 | bs->open_flags = flags; |
3313 | bs->options = options; | |
3314 | options = qdict_clone_shallow(options); | |
3315 | ||
76c591b0 | 3316 | /* Find the right image format driver */ |
129c7d1c | 3317 | /* See cautionary note on accessing @options above */ |
76c591b0 KW |
3318 | drvname = qdict_get_try_str(options, "driver"); |
3319 | if (drvname) { | |
3320 | drv = bdrv_find_format(drvname); | |
76c591b0 KW |
3321 | if (!drv) { |
3322 | error_setg(errp, "Unknown driver: '%s'", drvname); | |
76c591b0 KW |
3323 | goto fail; |
3324 | } | |
3325 | } | |
3326 | ||
3327 | assert(drvname || !(flags & BDRV_O_PROTOCOL)); | |
76c591b0 | 3328 | |
129c7d1c | 3329 | /* See cautionary note on accessing @options above */ |
3e8c2e57 | 3330 | backing = qdict_get_try_str(options, "backing"); |
e59a0cf1 HR |
3331 | if (qobject_to(QNull, qdict_get(options, "backing")) != NULL || |
3332 | (backing && *backing == '\0')) | |
3333 | { | |
4f7be280 HR |
3334 | if (backing) { |
3335 | warn_report("Use of \"backing\": \"\" is deprecated; " | |
3336 | "use \"backing\": null instead"); | |
3337 | } | |
3e8c2e57 | 3338 | flags |= BDRV_O_NO_BACKING; |
ae0f57f0 KW |
3339 | qdict_del(bs->explicit_options, "backing"); |
3340 | qdict_del(bs->options, "backing"); | |
3e8c2e57 AG |
3341 | qdict_del(options, "backing"); |
3342 | } | |
3343 | ||
5696c6e3 | 3344 | /* Open image file without format layer. This BlockBackend is only used for |
4e4bf5c4 KW |
3345 | * probing, the block drivers will do their own bdrv_open_child() for the |
3346 | * same BDS, which is why we put the node name back into options. */ | |
f4788adc | 3347 | if ((flags & BDRV_O_PROTOCOL) == 0) { |
5696c6e3 KW |
3348 | BlockDriverState *file_bs; |
3349 | ||
3350 | file_bs = bdrv_open_child_bs(filename, options, "file", bs, | |
58944401 HR |
3351 | &child_of_bds, BDRV_CHILD_IMAGE, |
3352 | true, &local_err); | |
1fdd6933 | 3353 | if (local_err) { |
f4788adc KW |
3354 | goto fail; |
3355 | } | |
5696c6e3 | 3356 | if (file_bs != NULL) { |
dacaa162 KW |
3357 | /* Not requesting BLK_PERM_CONSISTENT_READ because we're only |
3358 | * looking at the header to guess the image format. This works even | |
3359 | * in cases where a guest would not see a consistent state. */ | |
d861ab3a | 3360 | file = blk_new(bdrv_get_aio_context(file_bs), 0, BLK_PERM_ALL); |
d7086422 | 3361 | blk_insert_bs(file, file_bs, &local_err); |
5696c6e3 | 3362 | bdrv_unref(file_bs); |
d7086422 KW |
3363 | if (local_err) { |
3364 | goto fail; | |
3365 | } | |
5696c6e3 | 3366 | |
46f5ac20 | 3367 | qdict_put_str(options, "file", bdrv_get_node_name(file_bs)); |
4e4bf5c4 | 3368 | } |
f500a6d3 KW |
3369 | } |
3370 | ||
76c591b0 | 3371 | /* Image format probing */ |
38f3ef57 | 3372 | bs->probed = !drv; |
76c591b0 | 3373 | if (!drv && file) { |
cf2ab8fc | 3374 | ret = find_image_format(file, filename, &drv, &local_err); |
17b005f1 | 3375 | if (ret < 0) { |
8bfea15d | 3376 | goto fail; |
2a05cbe4 | 3377 | } |
62392ebb KW |
3378 | /* |
3379 | * This option update would logically belong in bdrv_fill_options(), | |
3380 | * but we first need to open bs->file for the probing to work, while | |
3381 | * opening bs->file already requires the (mostly) final set of options | |
3382 | * so that cache mode etc. can be inherited. | |
3383 | * | |
3384 | * Adding the driver later is somewhat ugly, but it's not an option | |
3385 | * that would ever be inherited, so it's correct. We just need to make | |
3386 | * sure to update both bs->options (which has the full effective | |
3387 | * options for bs) and options (which has file.* already removed). | |
3388 | */ | |
46f5ac20 EB |
3389 | qdict_put_str(bs->options, "driver", drv->format_name); |
3390 | qdict_put_str(options, "driver", drv->format_name); | |
76c591b0 | 3391 | } else if (!drv) { |
17b005f1 | 3392 | error_setg(errp, "Must specify either driver or file"); |
8bfea15d | 3393 | goto fail; |
ea2384d3 | 3394 | } |
b6ce07aa | 3395 | |
53a29513 HR |
3396 | /* BDRV_O_PROTOCOL must be set iff a protocol BDS is about to be created */ |
3397 | assert(!!(flags & BDRV_O_PROTOCOL) == !!drv->bdrv_file_open); | |
3398 | /* file must be NULL if a protocol BDS is about to be created | |
3399 | * (the inverse results in an error message from bdrv_open_common()) */ | |
3400 | assert(!(flags & BDRV_O_PROTOCOL) || !file); | |
3401 | ||
b6ce07aa | 3402 | /* Open the image */ |
82dc8b41 | 3403 | ret = bdrv_open_common(bs, file, options, &local_err); |
b6ce07aa | 3404 | if (ret < 0) { |
8bfea15d | 3405 | goto fail; |
6987307c CH |
3406 | } |
3407 | ||
4e4bf5c4 | 3408 | if (file) { |
5696c6e3 | 3409 | blk_unref(file); |
f500a6d3 KW |
3410 | file = NULL; |
3411 | } | |
3412 | ||
b6ce07aa | 3413 | /* If there is a backing file, use it */ |
9156df12 | 3414 | if ((flags & BDRV_O_NO_BACKING) == 0) { |
d9b7b057 | 3415 | ret = bdrv_open_backing_file(bs, options, "backing", &local_err); |
b6ce07aa | 3416 | if (ret < 0) { |
b6ad491a | 3417 | goto close_and_fail; |
b6ce07aa | 3418 | } |
b6ce07aa KW |
3419 | } |
3420 | ||
50196d7a AG |
3421 | /* Remove all children options and references |
3422 | * from bs->options and bs->explicit_options */ | |
2f624b80 AG |
3423 | QLIST_FOREACH(child, &bs->children, next) { |
3424 | char *child_key_dot; | |
3425 | child_key_dot = g_strdup_printf("%s.", child->name); | |
3426 | qdict_extract_subqdict(bs->explicit_options, NULL, child_key_dot); | |
3427 | qdict_extract_subqdict(bs->options, NULL, child_key_dot); | |
50196d7a AG |
3428 | qdict_del(bs->explicit_options, child->name); |
3429 | qdict_del(bs->options, child->name); | |
2f624b80 AG |
3430 | g_free(child_key_dot); |
3431 | } | |
3432 | ||
b6ad491a | 3433 | /* Check if any unknown options were used */ |
7ad2757f | 3434 | if (qdict_size(options) != 0) { |
b6ad491a | 3435 | const QDictEntry *entry = qdict_first(options); |
5acd9d81 HR |
3436 | if (flags & BDRV_O_PROTOCOL) { |
3437 | error_setg(errp, "Block protocol '%s' doesn't support the option " | |
3438 | "'%s'", drv->format_name, entry->key); | |
3439 | } else { | |
d0e46a55 HR |
3440 | error_setg(errp, |
3441 | "Block format '%s' does not support the option '%s'", | |
3442 | drv->format_name, entry->key); | |
5acd9d81 | 3443 | } |
b6ad491a | 3444 | |
b6ad491a KW |
3445 | goto close_and_fail; |
3446 | } | |
b6ad491a | 3447 | |
c01c214b | 3448 | bdrv_parent_cb_change_media(bs, true); |
b6ce07aa | 3449 | |
cb3e7f08 | 3450 | qobject_unref(options); |
8961be33 | 3451 | options = NULL; |
dd62f1ca KW |
3452 | |
3453 | /* For snapshot=on, create a temporary qcow2 overlay. bs points to the | |
3454 | * temporary snapshot afterwards. */ | |
3455 | if (snapshot_flags) { | |
66836189 HR |
3456 | BlockDriverState *snapshot_bs; |
3457 | snapshot_bs = bdrv_append_temp_snapshot(bs, snapshot_flags, | |
3458 | snapshot_options, &local_err); | |
73176bee | 3459 | snapshot_options = NULL; |
dd62f1ca KW |
3460 | if (local_err) { |
3461 | goto close_and_fail; | |
3462 | } | |
5b363937 HR |
3463 | /* We are not going to return bs but the overlay on top of it |
3464 | * (snapshot_bs); thus, we have to drop the strong reference to bs | |
3465 | * (which we obtained by calling bdrv_new()). bs will not be deleted, | |
3466 | * though, because the overlay still has a reference to it. */ | |
3467 | bdrv_unref(bs); | |
3468 | bs = snapshot_bs; | |
dd62f1ca KW |
3469 | } |
3470 | ||
5b363937 | 3471 | return bs; |
b6ce07aa | 3472 | |
8bfea15d | 3473 | fail: |
5696c6e3 | 3474 | blk_unref(file); |
cb3e7f08 MAL |
3475 | qobject_unref(snapshot_options); |
3476 | qobject_unref(bs->explicit_options); | |
3477 | qobject_unref(bs->options); | |
3478 | qobject_unref(options); | |
de9c0cec | 3479 | bs->options = NULL; |
998cbd6a | 3480 | bs->explicit_options = NULL; |
5b363937 | 3481 | bdrv_unref(bs); |
621ff94d | 3482 | error_propagate(errp, local_err); |
5b363937 | 3483 | return NULL; |
de9c0cec | 3484 | |
b6ad491a | 3485 | close_and_fail: |
5b363937 | 3486 | bdrv_unref(bs); |
cb3e7f08 MAL |
3487 | qobject_unref(snapshot_options); |
3488 | qobject_unref(options); | |
621ff94d | 3489 | error_propagate(errp, local_err); |
5b363937 | 3490 | return NULL; |
b6ce07aa KW |
3491 | } |
3492 | ||
5b363937 HR |
3493 | BlockDriverState *bdrv_open(const char *filename, const char *reference, |
3494 | QDict *options, int flags, Error **errp) | |
f3930ed0 | 3495 | { |
5b363937 | 3496 | return bdrv_open_inherit(filename, reference, options, flags, NULL, |
272c02ea | 3497 | NULL, 0, errp); |
f3930ed0 KW |
3498 | } |
3499 | ||
faf116b4 AG |
3500 | /* Return true if the NULL-terminated @list contains @str */ |
3501 | static bool is_str_in_list(const char *str, const char *const *list) | |
3502 | { | |
3503 | if (str && list) { | |
3504 | int i; | |
3505 | for (i = 0; list[i] != NULL; i++) { | |
3506 | if (!strcmp(str, list[i])) { | |
3507 | return true; | |
3508 | } | |
3509 | } | |
3510 | } | |
3511 | return false; | |
3512 | } | |
3513 | ||
3514 | /* | |
3515 | * Check that every option set in @bs->options is also set in | |
3516 | * @new_opts. | |
3517 | * | |
3518 | * Options listed in the common_options list and in | |
3519 | * @bs->drv->mutable_opts are skipped. | |
3520 | * | |
3521 | * Return 0 on success, otherwise return -EINVAL and set @errp. | |
3522 | */ | |
3523 | static int bdrv_reset_options_allowed(BlockDriverState *bs, | |
3524 | const QDict *new_opts, Error **errp) | |
3525 | { | |
3526 | const QDictEntry *e; | |
3527 | /* These options are common to all block drivers and are handled | |
3528 | * in bdrv_reopen_prepare() so they can be left out of @new_opts */ | |
3529 | const char *const common_options[] = { | |
3530 | "node-name", "discard", "cache.direct", "cache.no-flush", | |
3531 | "read-only", "auto-read-only", "detect-zeroes", NULL | |
3532 | }; | |
3533 | ||
3534 | for (e = qdict_first(bs->options); e; e = qdict_next(bs->options, e)) { | |
3535 | if (!qdict_haskey(new_opts, e->key) && | |
3536 | !is_str_in_list(e->key, common_options) && | |
3537 | !is_str_in_list(e->key, bs->drv->mutable_opts)) { | |
3538 | error_setg(errp, "Option '%s' cannot be reset " | |
3539 | "to its default value", e->key); | |
3540 | return -EINVAL; | |
3541 | } | |
3542 | } | |
3543 | ||
3544 | return 0; | |
3545 | } | |
3546 | ||
cb828c31 AG |
3547 | /* |
3548 | * Returns true if @child can be reached recursively from @bs | |
3549 | */ | |
3550 | static bool bdrv_recurse_has_child(BlockDriverState *bs, | |
3551 | BlockDriverState *child) | |
3552 | { | |
3553 | BdrvChild *c; | |
3554 | ||
3555 | if (bs == child) { | |
3556 | return true; | |
3557 | } | |
3558 | ||
3559 | QLIST_FOREACH(c, &bs->children, next) { | |
3560 | if (bdrv_recurse_has_child(c->bs, child)) { | |
3561 | return true; | |
3562 | } | |
3563 | } | |
3564 | ||
3565 | return false; | |
3566 | } | |
3567 | ||
e971aa12 JC |
3568 | /* |
3569 | * Adds a BlockDriverState to a simple queue for an atomic, transactional | |
3570 | * reopen of multiple devices. | |
3571 | * | |
859443b0 | 3572 | * bs_queue can either be an existing BlockReopenQueue that has had QTAILQ_INIT |
e971aa12 JC |
3573 | * already performed, or alternatively may be NULL a new BlockReopenQueue will |
3574 | * be created and initialized. This newly created BlockReopenQueue should be | |
3575 | * passed back in for subsequent calls that are intended to be of the same | |
3576 | * atomic 'set'. | |
3577 | * | |
3578 | * bs is the BlockDriverState to add to the reopen queue. | |
3579 | * | |
4d2cb092 KW |
3580 | * options contains the changed options for the associated bs |
3581 | * (the BlockReopenQueue takes ownership) | |
3582 | * | |
e971aa12 JC |
3583 | * flags contains the open flags for the associated bs |
3584 | * | |
3585 | * returns a pointer to bs_queue, which is either the newly allocated | |
3586 | * bs_queue, or the existing bs_queue being used. | |
3587 | * | |
1a63a907 | 3588 | * bs must be drained between bdrv_reopen_queue() and bdrv_reopen_multiple(). |
e971aa12 | 3589 | */ |
28518102 KW |
3590 | static BlockReopenQueue *bdrv_reopen_queue_child(BlockReopenQueue *bs_queue, |
3591 | BlockDriverState *bs, | |
3592 | QDict *options, | |
bd86fb99 | 3593 | const BdrvChildClass *klass, |
272c02ea | 3594 | BdrvChildRole role, |
3cdc69d3 | 3595 | bool parent_is_format, |
28518102 | 3596 | QDict *parent_options, |
077e8e20 AG |
3597 | int parent_flags, |
3598 | bool keep_old_opts) | |
e971aa12 JC |
3599 | { |
3600 | assert(bs != NULL); | |
3601 | ||
3602 | BlockReopenQueueEntry *bs_entry; | |
67251a31 | 3603 | BdrvChild *child; |
9aa09ddd AG |
3604 | QDict *old_options, *explicit_options, *options_copy; |
3605 | int flags; | |
3606 | QemuOpts *opts; | |
67251a31 | 3607 | |
1a63a907 KW |
3608 | /* Make sure that the caller remembered to use a drained section. This is |
3609 | * important to avoid graph changes between the recursive queuing here and | |
3610 | * bdrv_reopen_multiple(). */ | |
3611 | assert(bs->quiesce_counter > 0); | |
3612 | ||
e971aa12 JC |
3613 | if (bs_queue == NULL) { |
3614 | bs_queue = g_new0(BlockReopenQueue, 1); | |
859443b0 | 3615 | QTAILQ_INIT(bs_queue); |
e971aa12 JC |
3616 | } |
3617 | ||
4d2cb092 KW |
3618 | if (!options) { |
3619 | options = qdict_new(); | |
3620 | } | |
3621 | ||
5b7ba05f | 3622 | /* Check if this BlockDriverState is already in the queue */ |
859443b0 | 3623 | QTAILQ_FOREACH(bs_entry, bs_queue, entry) { |
5b7ba05f AG |
3624 | if (bs == bs_entry->state.bs) { |
3625 | break; | |
3626 | } | |
3627 | } | |
3628 | ||
28518102 KW |
3629 | /* |
3630 | * Precedence of options: | |
3631 | * 1. Explicitly passed in options (highest) | |
9aa09ddd AG |
3632 | * 2. Retained from explicitly set options of bs |
3633 | * 3. Inherited from parent node | |
3634 | * 4. Retained from effective options of bs | |
28518102 KW |
3635 | */ |
3636 | ||
145f598e | 3637 | /* Old explicitly set values (don't overwrite by inherited value) */ |
077e8e20 AG |
3638 | if (bs_entry || keep_old_opts) { |
3639 | old_options = qdict_clone_shallow(bs_entry ? | |
3640 | bs_entry->state.explicit_options : | |
3641 | bs->explicit_options); | |
3642 | bdrv_join_options(bs, options, old_options); | |
3643 | qobject_unref(old_options); | |
5b7ba05f | 3644 | } |
145f598e KW |
3645 | |
3646 | explicit_options = qdict_clone_shallow(options); | |
3647 | ||
28518102 KW |
3648 | /* Inherit from parent node */ |
3649 | if (parent_options) { | |
9aa09ddd | 3650 | flags = 0; |
3cdc69d3 | 3651 | klass->inherit_options(role, parent_is_format, &flags, options, |
272c02ea | 3652 | parent_flags, parent_options); |
9aa09ddd AG |
3653 | } else { |
3654 | flags = bdrv_get_flags(bs); | |
28518102 KW |
3655 | } |
3656 | ||
077e8e20 AG |
3657 | if (keep_old_opts) { |
3658 | /* Old values are used for options that aren't set yet */ | |
3659 | old_options = qdict_clone_shallow(bs->options); | |
3660 | bdrv_join_options(bs, options, old_options); | |
3661 | qobject_unref(old_options); | |
3662 | } | |
4d2cb092 | 3663 | |
9aa09ddd AG |
3664 | /* We have the final set of options so let's update the flags */ |
3665 | options_copy = qdict_clone_shallow(options); | |
3666 | opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort); | |
3667 | qemu_opts_absorb_qdict(opts, options_copy, NULL); | |
3668 | update_flags_from_options(&flags, opts); | |
3669 | qemu_opts_del(opts); | |
3670 | qobject_unref(options_copy); | |
3671 | ||
fd452021 | 3672 | /* bdrv_open_inherit() sets and clears some additional flags internally */ |
f1f25a2e | 3673 | flags &= ~BDRV_O_PROTOCOL; |
fd452021 KW |
3674 | if (flags & BDRV_O_RDWR) { |
3675 | flags |= BDRV_O_ALLOW_RDWR; | |
3676 | } | |
f1f25a2e | 3677 | |
1857c97b KW |
3678 | if (!bs_entry) { |
3679 | bs_entry = g_new0(BlockReopenQueueEntry, 1); | |
859443b0 | 3680 | QTAILQ_INSERT_TAIL(bs_queue, bs_entry, entry); |
1857c97b | 3681 | } else { |
cb3e7f08 MAL |
3682 | qobject_unref(bs_entry->state.options); |
3683 | qobject_unref(bs_entry->state.explicit_options); | |
1857c97b KW |
3684 | } |
3685 | ||
3686 | bs_entry->state.bs = bs; | |
3687 | bs_entry->state.options = options; | |
3688 | bs_entry->state.explicit_options = explicit_options; | |
3689 | bs_entry->state.flags = flags; | |
3690 | ||
30450259 KW |
3691 | /* This needs to be overwritten in bdrv_reopen_prepare() */ |
3692 | bs_entry->state.perm = UINT64_MAX; | |
3693 | bs_entry->state.shared_perm = 0; | |
3694 | ||
8546632e AG |
3695 | /* |
3696 | * If keep_old_opts is false then it means that unspecified | |
3697 | * options must be reset to their original value. We don't allow | |
3698 | * resetting 'backing' but we need to know if the option is | |
3699 | * missing in order to decide if we have to return an error. | |
3700 | */ | |
3701 | if (!keep_old_opts) { | |
3702 | bs_entry->state.backing_missing = | |
3703 | !qdict_haskey(options, "backing") && | |
3704 | !qdict_haskey(options, "backing.driver"); | |
3705 | } | |
3706 | ||
67251a31 | 3707 | QLIST_FOREACH(child, &bs->children, next) { |
8546632e AG |
3708 | QDict *new_child_options = NULL; |
3709 | bool child_keep_old = keep_old_opts; | |
67251a31 | 3710 | |
4c9dfe5d KW |
3711 | /* reopen can only change the options of block devices that were |
3712 | * implicitly created and inherited options. For other (referenced) | |
3713 | * block devices, a syntax like "backing.foo" results in an error. */ | |
67251a31 KW |
3714 | if (child->bs->inherits_from != bs) { |
3715 | continue; | |
3716 | } | |
3717 | ||
8546632e AG |
3718 | /* Check if the options contain a child reference */ |
3719 | if (qdict_haskey(options, child->name)) { | |
3720 | const char *childref = qdict_get_try_str(options, child->name); | |
3721 | /* | |
3722 | * The current child must not be reopened if the child | |
3723 | * reference is null or points to a different node. | |
3724 | */ | |
3725 | if (g_strcmp0(childref, child->bs->node_name)) { | |
3726 | continue; | |
3727 | } | |
3728 | /* | |
3729 | * If the child reference points to the current child then | |
3730 | * reopen it with its existing set of options (note that | |
3731 | * it can still inherit new options from the parent). | |
3732 | */ | |
3733 | child_keep_old = true; | |
3734 | } else { | |
3735 | /* Extract child options ("child-name.*") */ | |
3736 | char *child_key_dot = g_strdup_printf("%s.", child->name); | |
3737 | qdict_extract_subqdict(explicit_options, NULL, child_key_dot); | |
3738 | qdict_extract_subqdict(options, &new_child_options, child_key_dot); | |
3739 | g_free(child_key_dot); | |
3740 | } | |
4c9dfe5d | 3741 | |
9aa09ddd | 3742 | bdrv_reopen_queue_child(bs_queue, child->bs, new_child_options, |
3cdc69d3 HR |
3743 | child->klass, child->role, bs->drv->is_format, |
3744 | options, flags, child_keep_old); | |
e971aa12 JC |
3745 | } |
3746 | ||
e971aa12 JC |
3747 | return bs_queue; |
3748 | } | |
3749 | ||
28518102 KW |
3750 | BlockReopenQueue *bdrv_reopen_queue(BlockReopenQueue *bs_queue, |
3751 | BlockDriverState *bs, | |
077e8e20 | 3752 | QDict *options, bool keep_old_opts) |
28518102 | 3753 | { |
3cdc69d3 HR |
3754 | return bdrv_reopen_queue_child(bs_queue, bs, options, NULL, 0, false, |
3755 | NULL, 0, keep_old_opts); | |
28518102 KW |
3756 | } |
3757 | ||
e971aa12 JC |
3758 | /* |
3759 | * Reopen multiple BlockDriverStates atomically & transactionally. | |
3760 | * | |
3761 | * The queue passed in (bs_queue) must have been built up previous | |
3762 | * via bdrv_reopen_queue(). | |
3763 | * | |
3764 | * Reopens all BDS specified in the queue, with the appropriate | |
3765 | * flags. All devices are prepared for reopen, and failure of any | |
50d6a8a3 | 3766 | * device will cause all device changes to be abandoned, and intermediate |
e971aa12 JC |
3767 | * data cleaned up. |
3768 | * | |
3769 | * If all devices prepare successfully, then the changes are committed | |
3770 | * to all devices. | |
3771 | * | |
1a63a907 KW |
3772 | * All affected nodes must be drained between bdrv_reopen_queue() and |
3773 | * bdrv_reopen_multiple(). | |
e971aa12 | 3774 | */ |
5019aece | 3775 | int bdrv_reopen_multiple(BlockReopenQueue *bs_queue, Error **errp) |
e971aa12 JC |
3776 | { |
3777 | int ret = -1; | |
3778 | BlockReopenQueueEntry *bs_entry, *next; | |
e971aa12 JC |
3779 | |
3780 | assert(bs_queue != NULL); | |
3781 | ||
859443b0 | 3782 | QTAILQ_FOREACH(bs_entry, bs_queue, entry) { |
1a63a907 | 3783 | assert(bs_entry->state.bs->quiesce_counter > 0); |
a4615ab3 | 3784 | if (bdrv_reopen_prepare(&bs_entry->state, bs_queue, errp)) { |
e971aa12 JC |
3785 | goto cleanup; |
3786 | } | |
3787 | bs_entry->prepared = true; | |
3788 | } | |
3789 | ||
859443b0 | 3790 | QTAILQ_FOREACH(bs_entry, bs_queue, entry) { |
69b736e7 KW |
3791 | BDRVReopenState *state = &bs_entry->state; |
3792 | ret = bdrv_check_perm(state->bs, bs_queue, state->perm, | |
9eab1544 | 3793 | state->shared_perm, NULL, NULL, errp); |
69b736e7 KW |
3794 | if (ret < 0) { |
3795 | goto cleanup_perm; | |
3796 | } | |
cb828c31 AG |
3797 | /* Check if new_backing_bs would accept the new permissions */ |
3798 | if (state->replace_backing_bs && state->new_backing_bs) { | |
3799 | uint64_t nperm, nshared; | |
3800 | bdrv_child_perm(state->bs, state->new_backing_bs, | |
e5d8a406 | 3801 | NULL, bdrv_backing_role(state->bs), |
25191e5f | 3802 | bs_queue, state->perm, state->shared_perm, |
cb828c31 AG |
3803 | &nperm, &nshared); |
3804 | ret = bdrv_check_update_perm(state->new_backing_bs, NULL, | |
9eab1544 | 3805 | nperm, nshared, NULL, NULL, errp); |
cb828c31 AG |
3806 | if (ret < 0) { |
3807 | goto cleanup_perm; | |
3808 | } | |
3809 | } | |
69b736e7 KW |
3810 | bs_entry->perms_checked = true; |
3811 | } | |
3812 | ||
fcd6a4f4 VSO |
3813 | /* |
3814 | * If we reach this point, we have success and just need to apply the | |
3815 | * changes. | |
3816 | * | |
3817 | * Reverse order is used to comfort qcow2 driver: on commit it need to write | |
3818 | * IN_USE flag to the image, to mark bitmaps in the image as invalid. But | |
3819 | * children are usually goes after parents in reopen-queue, so go from last | |
3820 | * to first element. | |
e971aa12 | 3821 | */ |
fcd6a4f4 | 3822 | QTAILQ_FOREACH_REVERSE(bs_entry, bs_queue, entry) { |
e971aa12 JC |
3823 | bdrv_reopen_commit(&bs_entry->state); |
3824 | } | |
3825 | ||
3826 | ret = 0; | |
69b736e7 | 3827 | cleanup_perm: |
859443b0 | 3828 | QTAILQ_FOREACH_SAFE(bs_entry, bs_queue, entry, next) { |
69b736e7 KW |
3829 | BDRVReopenState *state = &bs_entry->state; |
3830 | ||
3831 | if (!bs_entry->perms_checked) { | |
3832 | continue; | |
3833 | } | |
e971aa12 | 3834 | |
69b736e7 KW |
3835 | if (ret == 0) { |
3836 | bdrv_set_perm(state->bs, state->perm, state->shared_perm); | |
3837 | } else { | |
3838 | bdrv_abort_perm_update(state->bs); | |
cb828c31 AG |
3839 | if (state->replace_backing_bs && state->new_backing_bs) { |
3840 | bdrv_abort_perm_update(state->new_backing_bs); | |
3841 | } | |
69b736e7 KW |
3842 | } |
3843 | } | |
17e1e2be PK |
3844 | |
3845 | if (ret == 0) { | |
3846 | QTAILQ_FOREACH_REVERSE(bs_entry, bs_queue, entry) { | |
3847 | BlockDriverState *bs = bs_entry->state.bs; | |
3848 | ||
3849 | if (bs->drv->bdrv_reopen_commit_post) | |
3850 | bs->drv->bdrv_reopen_commit_post(&bs_entry->state); | |
3851 | } | |
3852 | } | |
e971aa12 | 3853 | cleanup: |
859443b0 | 3854 | QTAILQ_FOREACH_SAFE(bs_entry, bs_queue, entry, next) { |
1bab38e7 AG |
3855 | if (ret) { |
3856 | if (bs_entry->prepared) { | |
3857 | bdrv_reopen_abort(&bs_entry->state); | |
3858 | } | |
cb3e7f08 | 3859 | qobject_unref(bs_entry->state.explicit_options); |
4c8350fe | 3860 | qobject_unref(bs_entry->state.options); |
e971aa12 | 3861 | } |
cb828c31 AG |
3862 | if (bs_entry->state.new_backing_bs) { |
3863 | bdrv_unref(bs_entry->state.new_backing_bs); | |
3864 | } | |
e971aa12 JC |
3865 | g_free(bs_entry); |
3866 | } | |
3867 | g_free(bs_queue); | |
40840e41 | 3868 | |
e971aa12 JC |
3869 | return ret; |
3870 | } | |
3871 | ||
6e1000a8 AG |
3872 | int bdrv_reopen_set_read_only(BlockDriverState *bs, bool read_only, |
3873 | Error **errp) | |
3874 | { | |
3875 | int ret; | |
3876 | BlockReopenQueue *queue; | |
3877 | QDict *opts = qdict_new(); | |
3878 | ||
3879 | qdict_put_bool(opts, BDRV_OPT_READ_ONLY, read_only); | |
3880 | ||
3881 | bdrv_subtree_drained_begin(bs); | |
077e8e20 | 3882 | queue = bdrv_reopen_queue(NULL, bs, opts, true); |
5019aece | 3883 | ret = bdrv_reopen_multiple(queue, errp); |
6e1000a8 AG |
3884 | bdrv_subtree_drained_end(bs); |
3885 | ||
3886 | return ret; | |
3887 | } | |
3888 | ||
30450259 KW |
3889 | static BlockReopenQueueEntry *find_parent_in_reopen_queue(BlockReopenQueue *q, |
3890 | BdrvChild *c) | |
3891 | { | |
3892 | BlockReopenQueueEntry *entry; | |
3893 | ||
859443b0 | 3894 | QTAILQ_FOREACH(entry, q, entry) { |
30450259 KW |
3895 | BlockDriverState *bs = entry->state.bs; |
3896 | BdrvChild *child; | |
3897 | ||
3898 | QLIST_FOREACH(child, &bs->children, next) { | |
3899 | if (child == c) { | |
3900 | return entry; | |
3901 | } | |
3902 | } | |
3903 | } | |
3904 | ||
3905 | return NULL; | |
3906 | } | |
3907 | ||
3908 | static void bdrv_reopen_perm(BlockReopenQueue *q, BlockDriverState *bs, | |
3909 | uint64_t *perm, uint64_t *shared) | |
3910 | { | |
3911 | BdrvChild *c; | |
3912 | BlockReopenQueueEntry *parent; | |
3913 | uint64_t cumulative_perms = 0; | |
3914 | uint64_t cumulative_shared_perms = BLK_PERM_ALL; | |
3915 | ||
3916 | QLIST_FOREACH(c, &bs->parents, next_parent) { | |
3917 | parent = find_parent_in_reopen_queue(q, c); | |
3918 | if (!parent) { | |
3919 | cumulative_perms |= c->perm; | |
3920 | cumulative_shared_perms &= c->shared_perm; | |
3921 | } else { | |
3922 | uint64_t nperm, nshared; | |
3923 | ||
e5d8a406 | 3924 | bdrv_child_perm(parent->state.bs, bs, c, c->role, q, |
30450259 KW |
3925 | parent->state.perm, parent->state.shared_perm, |
3926 | &nperm, &nshared); | |
3927 | ||
3928 | cumulative_perms |= nperm; | |
3929 | cumulative_shared_perms &= nshared; | |
3930 | } | |
3931 | } | |
3932 | *perm = cumulative_perms; | |
3933 | *shared = cumulative_shared_perms; | |
3934 | } | |
e971aa12 | 3935 | |
1de6b45f KW |
3936 | static bool bdrv_reopen_can_attach(BlockDriverState *parent, |
3937 | BdrvChild *child, | |
3938 | BlockDriverState *new_child, | |
3939 | Error **errp) | |
3940 | { | |
3941 | AioContext *parent_ctx = bdrv_get_aio_context(parent); | |
3942 | AioContext *child_ctx = bdrv_get_aio_context(new_child); | |
3943 | GSList *ignore; | |
3944 | bool ret; | |
3945 | ||
3946 | ignore = g_slist_prepend(NULL, child); | |
3947 | ret = bdrv_can_set_aio_context(new_child, parent_ctx, &ignore, NULL); | |
3948 | g_slist_free(ignore); | |
3949 | if (ret) { | |
3950 | return ret; | |
3951 | } | |
3952 | ||
3953 | ignore = g_slist_prepend(NULL, child); | |
3954 | ret = bdrv_can_set_aio_context(parent, child_ctx, &ignore, errp); | |
3955 | g_slist_free(ignore); | |
3956 | return ret; | |
3957 | } | |
3958 | ||
cb828c31 AG |
3959 | /* |
3960 | * Take a BDRVReopenState and check if the value of 'backing' in the | |
3961 | * reopen_state->options QDict is valid or not. | |
3962 | * | |
3963 | * If 'backing' is missing from the QDict then return 0. | |
3964 | * | |
3965 | * If 'backing' contains the node name of the backing file of | |
3966 | * reopen_state->bs then return 0. | |
3967 | * | |
3968 | * If 'backing' contains a different node name (or is null) then check | |
3969 | * whether the current backing file can be replaced with the new one. | |
3970 | * If that's the case then reopen_state->replace_backing_bs is set to | |
3971 | * true and reopen_state->new_backing_bs contains a pointer to the new | |
3972 | * backing BlockDriverState (or NULL). | |
3973 | * | |
3974 | * Return 0 on success, otherwise return < 0 and set @errp. | |
3975 | */ | |
3976 | static int bdrv_reopen_parse_backing(BDRVReopenState *reopen_state, | |
3977 | Error **errp) | |
3978 | { | |
3979 | BlockDriverState *bs = reopen_state->bs; | |
3980 | BlockDriverState *overlay_bs, *new_backing_bs; | |
3981 | QObject *value; | |
3982 | const char *str; | |
3983 | ||
3984 | value = qdict_get(reopen_state->options, "backing"); | |
3985 | if (value == NULL) { | |
3986 | return 0; | |
3987 | } | |
3988 | ||
3989 | switch (qobject_type(value)) { | |
3990 | case QTYPE_QNULL: | |
3991 | new_backing_bs = NULL; | |
3992 | break; | |
3993 | case QTYPE_QSTRING: | |
3994 | str = qobject_get_try_str(value); | |
3995 | new_backing_bs = bdrv_lookup_bs(NULL, str, errp); | |
3996 | if (new_backing_bs == NULL) { | |
3997 | return -EINVAL; | |
3998 | } else if (bdrv_recurse_has_child(new_backing_bs, bs)) { | |
3999 | error_setg(errp, "Making '%s' a backing file of '%s' " | |
4000 | "would create a cycle", str, bs->node_name); | |
4001 | return -EINVAL; | |
4002 | } | |
4003 | break; | |
4004 | default: | |
4005 | /* 'backing' does not allow any other data type */ | |
4006 | g_assert_not_reached(); | |
4007 | } | |
4008 | ||
4009 | /* | |
1de6b45f KW |
4010 | * Check AioContext compatibility so that the bdrv_set_backing_hd() call in |
4011 | * bdrv_reopen_commit() won't fail. | |
cb828c31 AG |
4012 | */ |
4013 | if (new_backing_bs) { | |
1de6b45f | 4014 | if (!bdrv_reopen_can_attach(bs, bs->backing, new_backing_bs, errp)) { |
cb828c31 AG |
4015 | return -EINVAL; |
4016 | } | |
4017 | } | |
4018 | ||
4019 | /* | |
4020 | * Find the "actual" backing file by skipping all links that point | |
4021 | * to an implicit node, if any (e.g. a commit filter node). | |
4022 | */ | |
4023 | overlay_bs = bs; | |
4024 | while (backing_bs(overlay_bs) && backing_bs(overlay_bs)->implicit) { | |
4025 | overlay_bs = backing_bs(overlay_bs); | |
4026 | } | |
4027 | ||
4028 | /* If we want to replace the backing file we need some extra checks */ | |
4029 | if (new_backing_bs != backing_bs(overlay_bs)) { | |
4030 | /* Check for implicit nodes between bs and its backing file */ | |
4031 | if (bs != overlay_bs) { | |
4032 | error_setg(errp, "Cannot change backing link if '%s' has " | |
4033 | "an implicit backing file", bs->node_name); | |
4034 | return -EPERM; | |
4035 | } | |
4036 | /* Check if the backing link that we want to replace is frozen */ | |
4037 | if (bdrv_is_backing_chain_frozen(overlay_bs, backing_bs(overlay_bs), | |
4038 | errp)) { | |
4039 | return -EPERM; | |
4040 | } | |
4041 | reopen_state->replace_backing_bs = true; | |
4042 | if (new_backing_bs) { | |
4043 | bdrv_ref(new_backing_bs); | |
4044 | reopen_state->new_backing_bs = new_backing_bs; | |
4045 | } | |
4046 | } | |
4047 | ||
4048 | return 0; | |
4049 | } | |
4050 | ||
e971aa12 JC |
4051 | /* |
4052 | * Prepares a BlockDriverState for reopen. All changes are staged in the | |
4053 | * 'opaque' field of the BDRVReopenState, which is used and allocated by | |
4054 | * the block driver layer .bdrv_reopen_prepare() | |
4055 | * | |
4056 | * bs is the BlockDriverState to reopen | |
4057 | * flags are the new open flags | |
4058 | * queue is the reopen queue | |
4059 | * | |
4060 | * Returns 0 on success, non-zero on error. On error errp will be set | |
4061 | * as well. | |
4062 | * | |
4063 | * On failure, bdrv_reopen_abort() will be called to clean up any data. | |
4064 | * It is the responsibility of the caller to then call the abort() or | |
4065 | * commit() for any other BDS that have been left in a prepare() state | |
4066 | * | |
4067 | */ | |
4068 | int bdrv_reopen_prepare(BDRVReopenState *reopen_state, BlockReopenQueue *queue, | |
4069 | Error **errp) | |
4070 | { | |
4071 | int ret = -1; | |
e6d79c41 | 4072 | int old_flags; |
e971aa12 JC |
4073 | Error *local_err = NULL; |
4074 | BlockDriver *drv; | |
ccf9dc07 | 4075 | QemuOpts *opts; |
4c8350fe | 4076 | QDict *orig_reopen_opts; |
593b3071 | 4077 | char *discard = NULL; |
3d8ce171 | 4078 | bool read_only; |
9ad08c44 | 4079 | bool drv_prepared = false; |
e971aa12 JC |
4080 | |
4081 | assert(reopen_state != NULL); | |
4082 | assert(reopen_state->bs->drv != NULL); | |
4083 | drv = reopen_state->bs->drv; | |
4084 | ||
4c8350fe AG |
4085 | /* This function and each driver's bdrv_reopen_prepare() remove |
4086 | * entries from reopen_state->options as they are processed, so | |
4087 | * we need to make a copy of the original QDict. */ | |
4088 | orig_reopen_opts = qdict_clone_shallow(reopen_state->options); | |
4089 | ||
ccf9dc07 KW |
4090 | /* Process generic block layer options */ |
4091 | opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort); | |
af175e85 | 4092 | if (!qemu_opts_absorb_qdict(opts, reopen_state->options, errp)) { |
ccf9dc07 KW |
4093 | ret = -EINVAL; |
4094 | goto error; | |
4095 | } | |
4096 | ||
e6d79c41 AG |
4097 | /* This was already called in bdrv_reopen_queue_child() so the flags |
4098 | * are up-to-date. This time we simply want to remove the options from | |
4099 | * QemuOpts in order to indicate that they have been processed. */ | |
4100 | old_flags = reopen_state->flags; | |
91a097e7 | 4101 | update_flags_from_options(&reopen_state->flags, opts); |
e6d79c41 | 4102 | assert(old_flags == reopen_state->flags); |
91a097e7 | 4103 | |
415bbca8 | 4104 | discard = qemu_opt_get_del(opts, BDRV_OPT_DISCARD); |
593b3071 AG |
4105 | if (discard != NULL) { |
4106 | if (bdrv_parse_discard_flags(discard, &reopen_state->flags) != 0) { | |
4107 | error_setg(errp, "Invalid discard option"); | |
4108 | ret = -EINVAL; | |
4109 | goto error; | |
4110 | } | |
4111 | } | |
4112 | ||
543770bd AG |
4113 | reopen_state->detect_zeroes = |
4114 | bdrv_parse_detect_zeroes(opts, reopen_state->flags, &local_err); | |
4115 | if (local_err) { | |
4116 | error_propagate(errp, local_err); | |
4117 | ret = -EINVAL; | |
4118 | goto error; | |
4119 | } | |
4120 | ||
57f9db9a AG |
4121 | /* All other options (including node-name and driver) must be unchanged. |
4122 | * Put them back into the QDict, so that they are checked at the end | |
4123 | * of this function. */ | |
4124 | qemu_opts_to_qdict(opts, reopen_state->options); | |
ccf9dc07 | 4125 | |
3d8ce171 JC |
4126 | /* If we are to stay read-only, do not allow permission change |
4127 | * to r/w. Attempting to set to r/w may fail if either BDRV_O_ALLOW_RDWR is | |
4128 | * not set, or if the BDS still has copy_on_read enabled */ | |
4129 | read_only = !(reopen_state->flags & BDRV_O_RDWR); | |
54a32bfe | 4130 | ret = bdrv_can_set_read_only(reopen_state->bs, read_only, true, &local_err); |
3d8ce171 JC |
4131 | if (local_err) { |
4132 | error_propagate(errp, local_err); | |
e971aa12 JC |
4133 | goto error; |
4134 | } | |
4135 | ||
30450259 KW |
4136 | /* Calculate required permissions after reopening */ |
4137 | bdrv_reopen_perm(queue, reopen_state->bs, | |
4138 | &reopen_state->perm, &reopen_state->shared_perm); | |
e971aa12 JC |
4139 | |
4140 | ret = bdrv_flush(reopen_state->bs); | |
4141 | if (ret) { | |
455b0fde | 4142 | error_setg_errno(errp, -ret, "Error flushing drive"); |
e971aa12 JC |
4143 | goto error; |
4144 | } | |
4145 | ||
4146 | if (drv->bdrv_reopen_prepare) { | |
faf116b4 AG |
4147 | /* |
4148 | * If a driver-specific option is missing, it means that we | |
4149 | * should reset it to its default value. | |
4150 | * But not all options allow that, so we need to check it first. | |
4151 | */ | |
4152 | ret = bdrv_reset_options_allowed(reopen_state->bs, | |
4153 | reopen_state->options, errp); | |
4154 | if (ret) { | |
4155 | goto error; | |
4156 | } | |
4157 | ||
e971aa12 JC |
4158 | ret = drv->bdrv_reopen_prepare(reopen_state, queue, &local_err); |
4159 | if (ret) { | |
4160 | if (local_err != NULL) { | |
4161 | error_propagate(errp, local_err); | |
4162 | } else { | |
f30c66ba | 4163 | bdrv_refresh_filename(reopen_state->bs); |
d8b6895f LC |
4164 | error_setg(errp, "failed while preparing to reopen image '%s'", |
4165 | reopen_state->bs->filename); | |
e971aa12 JC |
4166 | } |
4167 | goto error; | |
4168 | } | |
4169 | } else { | |
4170 | /* It is currently mandatory to have a bdrv_reopen_prepare() | |
4171 | * handler for each supported drv. */ | |
81e5f78a AG |
4172 | error_setg(errp, "Block format '%s' used by node '%s' " |
4173 | "does not support reopening files", drv->format_name, | |
4174 | bdrv_get_device_or_node_name(reopen_state->bs)); | |
e971aa12 JC |
4175 | ret = -1; |
4176 | goto error; | |
4177 | } | |
4178 | ||
9ad08c44 HR |
4179 | drv_prepared = true; |
4180 | ||
bacd9b87 AG |
4181 | /* |
4182 | * We must provide the 'backing' option if the BDS has a backing | |
4183 | * file or if the image file has a backing file name as part of | |
4184 | * its metadata. Otherwise the 'backing' option can be omitted. | |
4185 | */ | |
4186 | if (drv->supports_backing && reopen_state->backing_missing && | |
4187 | (backing_bs(reopen_state->bs) || reopen_state->bs->backing_file[0])) { | |
8546632e AG |
4188 | error_setg(errp, "backing is missing for '%s'", |
4189 | reopen_state->bs->node_name); | |
4190 | ret = -EINVAL; | |
4191 | goto error; | |
4192 | } | |
4193 | ||
cb828c31 AG |
4194 | /* |
4195 | * Allow changing the 'backing' option. The new value can be | |
4196 | * either a reference to an existing node (using its node name) | |
4197 | * or NULL to simply detach the current backing file. | |
4198 | */ | |
4199 | ret = bdrv_reopen_parse_backing(reopen_state, errp); | |
4200 | if (ret < 0) { | |
4201 | goto error; | |
4202 | } | |
4203 | qdict_del(reopen_state->options, "backing"); | |
4204 | ||
4d2cb092 KW |
4205 | /* Options that are not handled are only okay if they are unchanged |
4206 | * compared to the old state. It is expected that some options are only | |
4207 | * used for the initial open, but not reopen (e.g. filename) */ | |
4208 | if (qdict_size(reopen_state->options)) { | |
4209 | const QDictEntry *entry = qdict_first(reopen_state->options); | |
4210 | ||
4211 | do { | |
54fd1b0d HR |
4212 | QObject *new = entry->value; |
4213 | QObject *old = qdict_get(reopen_state->bs->options, entry->key); | |
4214 | ||
db905283 AG |
4215 | /* Allow child references (child_name=node_name) as long as they |
4216 | * point to the current child (i.e. everything stays the same). */ | |
4217 | if (qobject_type(new) == QTYPE_QSTRING) { | |
4218 | BdrvChild *child; | |
4219 | QLIST_FOREACH(child, &reopen_state->bs->children, next) { | |
4220 | if (!strcmp(child->name, entry->key)) { | |
4221 | break; | |
4222 | } | |
4223 | } | |
4224 | ||
4225 | if (child) { | |
4226 | const char *str = qobject_get_try_str(new); | |
4227 | if (!strcmp(child->bs->node_name, str)) { | |
4228 | continue; /* Found child with this name, skip option */ | |
4229 | } | |
4230 | } | |
4231 | } | |
4232 | ||
129c7d1c | 4233 | /* |
54fd1b0d HR |
4234 | * TODO: When using -drive to specify blockdev options, all values |
4235 | * will be strings; however, when using -blockdev, blockdev-add or | |
4236 | * filenames using the json:{} pseudo-protocol, they will be | |
4237 | * correctly typed. | |
4238 | * In contrast, reopening options are (currently) always strings | |
4239 | * (because you can only specify them through qemu-io; all other | |
4240 | * callers do not specify any options). | |
4241 | * Therefore, when using anything other than -drive to create a BDS, | |
4242 | * this cannot detect non-string options as unchanged, because | |
4243 | * qobject_is_equal() always returns false for objects of different | |
4244 | * type. In the future, this should be remedied by correctly typing | |
4245 | * all options. For now, this is not too big of an issue because | |
4246 | * the user can simply omit options which cannot be changed anyway, | |
4247 | * so they will stay unchanged. | |
129c7d1c | 4248 | */ |
54fd1b0d | 4249 | if (!qobject_is_equal(new, old)) { |
4d2cb092 KW |
4250 | error_setg(errp, "Cannot change the option '%s'", entry->key); |
4251 | ret = -EINVAL; | |
4252 | goto error; | |
4253 | } | |
4254 | } while ((entry = qdict_next(reopen_state->options, entry))); | |
4255 | } | |
4256 | ||
e971aa12 JC |
4257 | ret = 0; |
4258 | ||
4c8350fe AG |
4259 | /* Restore the original reopen_state->options QDict */ |
4260 | qobject_unref(reopen_state->options); | |
4261 | reopen_state->options = qobject_ref(orig_reopen_opts); | |
4262 | ||
e971aa12 | 4263 | error: |
9ad08c44 HR |
4264 | if (ret < 0 && drv_prepared) { |
4265 | /* drv->bdrv_reopen_prepare() has succeeded, so we need to | |
4266 | * call drv->bdrv_reopen_abort() before signaling an error | |
4267 | * (bdrv_reopen_multiple() will not call bdrv_reopen_abort() | |
4268 | * when the respective bdrv_reopen_prepare() has failed) */ | |
4269 | if (drv->bdrv_reopen_abort) { | |
4270 | drv->bdrv_reopen_abort(reopen_state); | |
4271 | } | |
4272 | } | |
ccf9dc07 | 4273 | qemu_opts_del(opts); |
4c8350fe | 4274 | qobject_unref(orig_reopen_opts); |
593b3071 | 4275 | g_free(discard); |
e971aa12 JC |
4276 | return ret; |
4277 | } | |
4278 | ||
4279 | /* | |
4280 | * Takes the staged changes for the reopen from bdrv_reopen_prepare(), and | |
4281 | * makes them final by swapping the staging BlockDriverState contents into | |
4282 | * the active BlockDriverState contents. | |
4283 | */ | |
4284 | void bdrv_reopen_commit(BDRVReopenState *reopen_state) | |
4285 | { | |
4286 | BlockDriver *drv; | |
50bf65ba | 4287 | BlockDriverState *bs; |
50196d7a | 4288 | BdrvChild *child; |
e971aa12 JC |
4289 | |
4290 | assert(reopen_state != NULL); | |
50bf65ba VSO |
4291 | bs = reopen_state->bs; |
4292 | drv = bs->drv; | |
e971aa12 JC |
4293 | assert(drv != NULL); |
4294 | ||
4295 | /* If there are any driver level actions to take */ | |
4296 | if (drv->bdrv_reopen_commit) { | |
4297 | drv->bdrv_reopen_commit(reopen_state); | |
4298 | } | |
4299 | ||
4300 | /* set BDS specific flags now */ | |
cb3e7f08 | 4301 | qobject_unref(bs->explicit_options); |
4c8350fe | 4302 | qobject_unref(bs->options); |
145f598e | 4303 | |
50bf65ba | 4304 | bs->explicit_options = reopen_state->explicit_options; |
4c8350fe | 4305 | bs->options = reopen_state->options; |
50bf65ba VSO |
4306 | bs->open_flags = reopen_state->flags; |
4307 | bs->read_only = !(reopen_state->flags & BDRV_O_RDWR); | |
543770bd | 4308 | bs->detect_zeroes = reopen_state->detect_zeroes; |
355ef4ac | 4309 | |
cb828c31 AG |
4310 | if (reopen_state->replace_backing_bs) { |
4311 | qdict_del(bs->explicit_options, "backing"); | |
4312 | qdict_del(bs->options, "backing"); | |
4313 | } | |
4314 | ||
50196d7a AG |
4315 | /* Remove child references from bs->options and bs->explicit_options. |
4316 | * Child options were already removed in bdrv_reopen_queue_child() */ | |
4317 | QLIST_FOREACH(child, &bs->children, next) { | |
4318 | qdict_del(bs->explicit_options, child->name); | |
4319 | qdict_del(bs->options, child->name); | |
4320 | } | |
4321 | ||
cb828c31 AG |
4322 | /* |
4323 | * Change the backing file if a new one was specified. We do this | |
4324 | * after updating bs->options, so bdrv_refresh_filename() (called | |
4325 | * from bdrv_set_backing_hd()) has the new values. | |
4326 | */ | |
4327 | if (reopen_state->replace_backing_bs) { | |
4328 | BlockDriverState *old_backing_bs = backing_bs(bs); | |
4329 | assert(!old_backing_bs || !old_backing_bs->implicit); | |
4330 | /* Abort the permission update on the backing bs we're detaching */ | |
4331 | if (old_backing_bs) { | |
4332 | bdrv_abort_perm_update(old_backing_bs); | |
4333 | } | |
4334 | bdrv_set_backing_hd(bs, reopen_state->new_backing_bs, &error_abort); | |
4335 | } | |
4336 | ||
50bf65ba | 4337 | bdrv_refresh_limits(bs, NULL); |
e971aa12 JC |
4338 | } |
4339 | ||
4340 | /* | |
4341 | * Abort the reopen, and delete and free the staged changes in | |
4342 | * reopen_state | |
4343 | */ | |
4344 | void bdrv_reopen_abort(BDRVReopenState *reopen_state) | |
4345 | { | |
4346 | BlockDriver *drv; | |
4347 | ||
4348 | assert(reopen_state != NULL); | |
4349 | drv = reopen_state->bs->drv; | |
4350 | assert(drv != NULL); | |
4351 | ||
4352 | if (drv->bdrv_reopen_abort) { | |
4353 | drv->bdrv_reopen_abort(reopen_state); | |
4354 | } | |
4355 | } | |
4356 | ||
4357 | ||
64dff520 | 4358 | static void bdrv_close(BlockDriverState *bs) |
fc01f7e7 | 4359 | { |
33384421 | 4360 | BdrvAioNotifier *ban, *ban_next; |
50a3efb0 | 4361 | BdrvChild *child, *next; |
33384421 | 4362 | |
30f55fb8 | 4363 | assert(!bs->refcnt); |
99b7e775 | 4364 | |
fc27291d | 4365 | bdrv_drained_begin(bs); /* complete I/O */ |
58fda173 | 4366 | bdrv_flush(bs); |
53ec73e2 | 4367 | bdrv_drain(bs); /* in case flush left pending I/O */ |
fc27291d | 4368 | |
3cbc002c | 4369 | if (bs->drv) { |
3c005293 VSO |
4370 | if (bs->drv->bdrv_close) { |
4371 | bs->drv->bdrv_close(bs); | |
4372 | } | |
9a4f4c31 | 4373 | bs->drv = NULL; |
50a3efb0 | 4374 | } |
9a7dedbc | 4375 | |
50a3efb0 | 4376 | QLIST_FOREACH_SAFE(child, &bs->children, next, next) { |
dd4118c7 | 4377 | bdrv_unref_child(bs, child); |
b338082b | 4378 | } |
98f90dba | 4379 | |
dd4118c7 AG |
4380 | bs->backing = NULL; |
4381 | bs->file = NULL; | |
50a3efb0 AG |
4382 | g_free(bs->opaque); |
4383 | bs->opaque = NULL; | |
4384 | atomic_set(&bs->copy_on_read, 0); | |
4385 | bs->backing_file[0] = '\0'; | |
4386 | bs->backing_format[0] = '\0'; | |
4387 | bs->total_sectors = 0; | |
4388 | bs->encrypted = false; | |
4389 | bs->sg = false; | |
cb3e7f08 MAL |
4390 | qobject_unref(bs->options); |
4391 | qobject_unref(bs->explicit_options); | |
50a3efb0 AG |
4392 | bs->options = NULL; |
4393 | bs->explicit_options = NULL; | |
cb3e7f08 | 4394 | qobject_unref(bs->full_open_options); |
50a3efb0 AG |
4395 | bs->full_open_options = NULL; |
4396 | ||
cca43ae1 VSO |
4397 | bdrv_release_named_dirty_bitmaps(bs); |
4398 | assert(QLIST_EMPTY(&bs->dirty_bitmaps)); | |
4399 | ||
33384421 HR |
4400 | QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_next) { |
4401 | g_free(ban); | |
4402 | } | |
4403 | QLIST_INIT(&bs->aio_notifiers); | |
fc27291d | 4404 | bdrv_drained_end(bs); |
b338082b FB |
4405 | } |
4406 | ||
2bc93fed MK |
4407 | void bdrv_close_all(void) |
4408 | { | |
b3b5299d | 4409 | assert(job_next(NULL) == NULL); |
cd7fca95 | 4410 | nbd_export_close_all(); |
ca9bd24c HR |
4411 | |
4412 | /* Drop references from requests still in flight, such as canceled block | |
4413 | * jobs whose AIO context has not been polled yet */ | |
4414 | bdrv_drain_all(); | |
2bc93fed | 4415 | |
ca9bd24c HR |
4416 | blk_remove_all_bs(); |
4417 | blockdev_close_all_bdrv_states(); | |
ed78cda3 | 4418 | |
a1a2af07 | 4419 | assert(QTAILQ_EMPTY(&all_bdrv_states)); |
2bc93fed MK |
4420 | } |
4421 | ||
d0ac0380 KW |
4422 | static bool should_update_child(BdrvChild *c, BlockDriverState *to) |
4423 | { | |
2f30b7c3 VSO |
4424 | GQueue *queue; |
4425 | GHashTable *found; | |
4426 | bool ret; | |
d0ac0380 | 4427 | |
bd86fb99 | 4428 | if (c->klass->stay_at_node) { |
d0ac0380 KW |
4429 | return false; |
4430 | } | |
4431 | ||
ec9f10fe HR |
4432 | /* If the child @c belongs to the BDS @to, replacing the current |
4433 | * c->bs by @to would mean to create a loop. | |
4434 | * | |
4435 | * Such a case occurs when appending a BDS to a backing chain. | |
4436 | * For instance, imagine the following chain: | |
4437 | * | |
4438 | * guest device -> node A -> further backing chain... | |
4439 | * | |
4440 | * Now we create a new BDS B which we want to put on top of this | |
4441 | * chain, so we first attach A as its backing node: | |
4442 | * | |
4443 | * node B | |
4444 | * | | |
4445 | * v | |
4446 | * guest device -> node A -> further backing chain... | |
4447 | * | |
4448 | * Finally we want to replace A by B. When doing that, we want to | |
4449 | * replace all pointers to A by pointers to B -- except for the | |
4450 | * pointer from B because (1) that would create a loop, and (2) | |
4451 | * that pointer should simply stay intact: | |
4452 | * | |
4453 | * guest device -> node B | |
4454 | * | | |
4455 | * v | |
4456 | * node A -> further backing chain... | |
4457 | * | |
4458 | * In general, when replacing a node A (c->bs) by a node B (@to), | |
4459 | * if A is a child of B, that means we cannot replace A by B there | |
4460 | * because that would create a loop. Silently detaching A from B | |
4461 | * is also not really an option. So overall just leaving A in | |
2f30b7c3 VSO |
4462 | * place there is the most sensible choice. |
4463 | * | |
4464 | * We would also create a loop in any cases where @c is only | |
4465 | * indirectly referenced by @to. Prevent this by returning false | |
4466 | * if @c is found (by breadth-first search) anywhere in the whole | |
4467 | * subtree of @to. | |
4468 | */ | |
4469 | ||
4470 | ret = true; | |
4471 | found = g_hash_table_new(NULL, NULL); | |
4472 | g_hash_table_add(found, to); | |
4473 | queue = g_queue_new(); | |
4474 | g_queue_push_tail(queue, to); | |
4475 | ||
4476 | while (!g_queue_is_empty(queue)) { | |
4477 | BlockDriverState *v = g_queue_pop_head(queue); | |
4478 | BdrvChild *c2; | |
4479 | ||
4480 | QLIST_FOREACH(c2, &v->children, next) { | |
4481 | if (c2 == c) { | |
4482 | ret = false; | |
4483 | break; | |
4484 | } | |
4485 | ||
4486 | if (g_hash_table_contains(found, c2->bs)) { | |
4487 | continue; | |
4488 | } | |
4489 | ||
4490 | g_queue_push_tail(queue, c2->bs); | |
4491 | g_hash_table_add(found, c2->bs); | |
d0ac0380 KW |
4492 | } |
4493 | } | |
4494 | ||
2f30b7c3 VSO |
4495 | g_queue_free(queue); |
4496 | g_hash_table_destroy(found); | |
4497 | ||
4498 | return ret; | |
d0ac0380 KW |
4499 | } |
4500 | ||
5fe31c25 KW |
4501 | void bdrv_replace_node(BlockDriverState *from, BlockDriverState *to, |
4502 | Error **errp) | |
dd62f1ca | 4503 | { |
d0ac0380 | 4504 | BdrvChild *c, *next; |
234ac1a9 | 4505 | GSList *list = NULL, *p; |
234ac1a9 KW |
4506 | uint64_t perm = 0, shared = BLK_PERM_ALL; |
4507 | int ret; | |
4508 | ||
4509 | /* Make sure that @from doesn't go away until we have successfully attached | |
4510 | * all of its parents to @to. */ | |
4511 | bdrv_ref(from); | |
dd62f1ca | 4512 | |
f871abd6 | 4513 | assert(qemu_get_current_aio_context() == qemu_get_aio_context()); |
30dd65f3 | 4514 | assert(bdrv_get_aio_context(from) == bdrv_get_aio_context(to)); |
f871abd6 KW |
4515 | bdrv_drained_begin(from); |
4516 | ||
234ac1a9 | 4517 | /* Put all parents into @list and calculate their cumulative permissions */ |
dd62f1ca | 4518 | QLIST_FOREACH_SAFE(c, &from->parents, next_parent, next) { |
ec9f10fe | 4519 | assert(c->bs == from); |
d0ac0380 | 4520 | if (!should_update_child(c, to)) { |
26de9438 KW |
4521 | continue; |
4522 | } | |
2cad1ebe AG |
4523 | if (c->frozen) { |
4524 | error_setg(errp, "Cannot change '%s' link to '%s'", | |
4525 | c->name, from->node_name); | |
4526 | goto out; | |
4527 | } | |
234ac1a9 KW |
4528 | list = g_slist_prepend(list, c); |
4529 | perm |= c->perm; | |
4530 | shared &= c->shared_perm; | |
4531 | } | |
4532 | ||
4533 | /* Check whether the required permissions can be granted on @to, ignoring | |
4534 | * all BdrvChild in @list so that they can't block themselves. */ | |
9eab1544 | 4535 | ret = bdrv_check_update_perm(to, NULL, perm, shared, list, NULL, errp); |
234ac1a9 KW |
4536 | if (ret < 0) { |
4537 | bdrv_abort_perm_update(to); | |
4538 | goto out; | |
4539 | } | |
4540 | ||
4541 | /* Now actually perform the change. We performed the permission check for | |
4542 | * all elements of @list at once, so set the permissions all at once at the | |
4543 | * very end. */ | |
4544 | for (p = list; p != NULL; p = p->next) { | |
4545 | c = p->data; | |
9bd910e2 | 4546 | |
dd62f1ca | 4547 | bdrv_ref(to); |
234ac1a9 | 4548 | bdrv_replace_child_noperm(c, to); |
dd62f1ca KW |
4549 | bdrv_unref(from); |
4550 | } | |
234ac1a9 | 4551 | |
b503de61 VSO |
4552 | bdrv_get_cumulative_perm(to, &perm, &shared); |
4553 | bdrv_set_perm(to, perm, shared); | |
234ac1a9 KW |
4554 | |
4555 | out: | |
4556 | g_slist_free(list); | |
f871abd6 | 4557 | bdrv_drained_end(from); |
234ac1a9 | 4558 | bdrv_unref(from); |
dd62f1ca KW |
4559 | } |
4560 | ||
4ddc07ca PB |
4561 | /* |
4562 | * Add new bs contents at the top of an image chain while the chain is | |
4563 | * live, while keeping required fields on the top layer. | |
4564 | * | |
4565 | * This will modify the BlockDriverState fields, and swap contents | |
4566 | * between bs_new and bs_top. Both bs_new and bs_top are modified. | |
4567 | * | |
bfb197e0 | 4568 | * bs_new must not be attached to a BlockBackend. |
4ddc07ca PB |
4569 | * |
4570 | * This function does not create any image files. | |
dd62f1ca KW |
4571 | * |
4572 | * bdrv_append() takes ownership of a bs_new reference and unrefs it because | |
4573 | * that's what the callers commonly need. bs_new will be referenced by the old | |
4574 | * parents of bs_top after bdrv_append() returns. If the caller needs to keep a | |
4575 | * reference of its own, it must call bdrv_ref(). | |
4ddc07ca | 4576 | */ |
b2c2832c KW |
4577 | void bdrv_append(BlockDriverState *bs_new, BlockDriverState *bs_top, |
4578 | Error **errp) | |
4ddc07ca | 4579 | { |
b2c2832c KW |
4580 | Error *local_err = NULL; |
4581 | ||
b2c2832c KW |
4582 | bdrv_set_backing_hd(bs_new, bs_top, &local_err); |
4583 | if (local_err) { | |
4584 | error_propagate(errp, local_err); | |
4585 | goto out; | |
4586 | } | |
dd62f1ca | 4587 | |
5fe31c25 | 4588 | bdrv_replace_node(bs_top, bs_new, &local_err); |
234ac1a9 KW |
4589 | if (local_err) { |
4590 | error_propagate(errp, local_err); | |
4591 | bdrv_set_backing_hd(bs_new, NULL, &error_abort); | |
4592 | goto out; | |
4593 | } | |
4ddc07ca | 4594 | |
dd62f1ca KW |
4595 | /* bs_new is now referenced by its new parents, we don't need the |
4596 | * additional reference any more. */ | |
b2c2832c | 4597 | out: |
dd62f1ca | 4598 | bdrv_unref(bs_new); |
8802d1fd JC |
4599 | } |
4600 | ||
4f6fd349 | 4601 | static void bdrv_delete(BlockDriverState *bs) |
b338082b | 4602 | { |
3718d8ab | 4603 | assert(bdrv_op_blocker_is_empty(bs)); |
4f6fd349 | 4604 | assert(!bs->refcnt); |
18846dee | 4605 | |
1b7bdbc1 | 4606 | /* remove from list, if necessary */ |
63eaaae0 KW |
4607 | if (bs->node_name[0] != '\0') { |
4608 | QTAILQ_REMOVE(&graph_bdrv_states, bs, node_list); | |
4609 | } | |
2c1d04e0 HR |
4610 | QTAILQ_REMOVE(&all_bdrv_states, bs, bs_list); |
4611 | ||
30c321f9 AK |
4612 | bdrv_close(bs); |
4613 | ||
7267c094 | 4614 | g_free(bs); |
fc01f7e7 FB |
4615 | } |
4616 | ||
e97fc193 AL |
4617 | /* |
4618 | * Run consistency checks on an image | |
4619 | * | |
e076f338 | 4620 | * Returns 0 if the check could be completed (it doesn't mean that the image is |
a1c7273b | 4621 | * free of errors) or -errno when an internal error occurred. The results of the |
e076f338 | 4622 | * check are stored in res. |
e97fc193 | 4623 | */ |
2fd61638 PB |
4624 | static int coroutine_fn bdrv_co_check(BlockDriverState *bs, |
4625 | BdrvCheckResult *res, BdrvCheckMode fix) | |
e97fc193 | 4626 | { |
908bcd54 HR |
4627 | if (bs->drv == NULL) { |
4628 | return -ENOMEDIUM; | |
4629 | } | |
2fd61638 | 4630 | if (bs->drv->bdrv_co_check == NULL) { |
e97fc193 AL |
4631 | return -ENOTSUP; |
4632 | } | |
4633 | ||
e076f338 | 4634 | memset(res, 0, sizeof(*res)); |
2fd61638 PB |
4635 | return bs->drv->bdrv_co_check(bs, res, fix); |
4636 | } | |
4637 | ||
4638 | typedef struct CheckCo { | |
4639 | BlockDriverState *bs; | |
4640 | BdrvCheckResult *res; | |
4641 | BdrvCheckMode fix; | |
4642 | int ret; | |
4643 | } CheckCo; | |
4644 | ||
66a5bdf3 | 4645 | static void coroutine_fn bdrv_check_co_entry(void *opaque) |
2fd61638 PB |
4646 | { |
4647 | CheckCo *cco = opaque; | |
4648 | cco->ret = bdrv_co_check(cco->bs, cco->res, cco->fix); | |
4720cbee | 4649 | aio_wait_kick(); |
2fd61638 PB |
4650 | } |
4651 | ||
4652 | int bdrv_check(BlockDriverState *bs, | |
4653 | BdrvCheckResult *res, BdrvCheckMode fix) | |
4654 | { | |
4655 | Coroutine *co; | |
4656 | CheckCo cco = { | |
4657 | .bs = bs, | |
4658 | .res = res, | |
4659 | .ret = -EINPROGRESS, | |
4660 | .fix = fix, | |
4661 | }; | |
4662 | ||
4663 | if (qemu_in_coroutine()) { | |
4664 | /* Fast-path if already in coroutine context */ | |
4665 | bdrv_check_co_entry(&cco); | |
4666 | } else { | |
4667 | co = qemu_coroutine_create(bdrv_check_co_entry, &cco); | |
4720cbee | 4668 | bdrv_coroutine_enter(bs, co); |
2fd61638 PB |
4669 | BDRV_POLL_WHILE(bs, cco.ret == -EINPROGRESS); |
4670 | } | |
4671 | ||
4672 | return cco.ret; | |
e97fc193 AL |
4673 | } |
4674 | ||
756e6736 KW |
4675 | /* |
4676 | * Return values: | |
4677 | * 0 - success | |
4678 | * -EINVAL - backing format specified, but no file | |
4679 | * -ENOSPC - can't update the backing file because no space is left in the | |
4680 | * image file header | |
4681 | * -ENOTSUP - format driver doesn't support changing the backing file | |
4682 | */ | |
4683 | int bdrv_change_backing_file(BlockDriverState *bs, | |
4684 | const char *backing_file, const char *backing_fmt) | |
4685 | { | |
4686 | BlockDriver *drv = bs->drv; | |
469ef350 | 4687 | int ret; |
756e6736 | 4688 | |
d470ad42 HR |
4689 | if (!drv) { |
4690 | return -ENOMEDIUM; | |
4691 | } | |
4692 | ||
5f377794 PB |
4693 | /* Backing file format doesn't make sense without a backing file */ |
4694 | if (backing_fmt && !backing_file) { | |
4695 | return -EINVAL; | |
4696 | } | |
4697 | ||
756e6736 | 4698 | if (drv->bdrv_change_backing_file != NULL) { |
469ef350 | 4699 | ret = drv->bdrv_change_backing_file(bs, backing_file, backing_fmt); |
756e6736 | 4700 | } else { |
469ef350 | 4701 | ret = -ENOTSUP; |
756e6736 | 4702 | } |
469ef350 PB |
4703 | |
4704 | if (ret == 0) { | |
4705 | pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_file ?: ""); | |
4706 | pstrcpy(bs->backing_format, sizeof(bs->backing_format), backing_fmt ?: ""); | |
998c2019 HR |
4707 | pstrcpy(bs->auto_backing_file, sizeof(bs->auto_backing_file), |
4708 | backing_file ?: ""); | |
469ef350 PB |
4709 | } |
4710 | return ret; | |
756e6736 KW |
4711 | } |
4712 | ||
6ebdcee2 JC |
4713 | /* |
4714 | * Finds the image layer in the chain that has 'bs' as its backing file. | |
4715 | * | |
4716 | * active is the current topmost image. | |
4717 | * | |
4718 | * Returns NULL if bs is not found in active's image chain, | |
4719 | * or if active == bs. | |
4caf0fcd JC |
4720 | * |
4721 | * Returns the bottommost base image if bs == NULL. | |
6ebdcee2 JC |
4722 | */ |
4723 | BlockDriverState *bdrv_find_overlay(BlockDriverState *active, | |
4724 | BlockDriverState *bs) | |
4725 | { | |
760e0063 KW |
4726 | while (active && bs != backing_bs(active)) { |
4727 | active = backing_bs(active); | |
6ebdcee2 JC |
4728 | } |
4729 | ||
4caf0fcd JC |
4730 | return active; |
4731 | } | |
6ebdcee2 | 4732 | |
4caf0fcd JC |
4733 | /* Given a BDS, searches for the base layer. */ |
4734 | BlockDriverState *bdrv_find_base(BlockDriverState *bs) | |
4735 | { | |
4736 | return bdrv_find_overlay(bs, NULL); | |
6ebdcee2 JC |
4737 | } |
4738 | ||
2cad1ebe AG |
4739 | /* |
4740 | * Return true if at least one of the backing links between @bs and | |
4741 | * @base is frozen. @errp is set if that's the case. | |
0f0998f6 | 4742 | * @base must be reachable from @bs, or NULL. |
2cad1ebe AG |
4743 | */ |
4744 | bool bdrv_is_backing_chain_frozen(BlockDriverState *bs, BlockDriverState *base, | |
4745 | Error **errp) | |
4746 | { | |
4747 | BlockDriverState *i; | |
4748 | ||
0f0998f6 AG |
4749 | for (i = bs; i != base; i = backing_bs(i)) { |
4750 | if (i->backing && i->backing->frozen) { | |
2cad1ebe AG |
4751 | error_setg(errp, "Cannot change '%s' link from '%s' to '%s'", |
4752 | i->backing->name, i->node_name, | |
4753 | backing_bs(i)->node_name); | |
4754 | return true; | |
4755 | } | |
4756 | } | |
4757 | ||
4758 | return false; | |
4759 | } | |
4760 | ||
4761 | /* | |
4762 | * Freeze all backing links between @bs and @base. | |
4763 | * If any of the links is already frozen the operation is aborted and | |
4764 | * none of the links are modified. | |
0f0998f6 | 4765 | * @base must be reachable from @bs, or NULL. |
2cad1ebe AG |
4766 | * Returns 0 on success. On failure returns < 0 and sets @errp. |
4767 | */ | |
4768 | int bdrv_freeze_backing_chain(BlockDriverState *bs, BlockDriverState *base, | |
4769 | Error **errp) | |
4770 | { | |
4771 | BlockDriverState *i; | |
4772 | ||
4773 | if (bdrv_is_backing_chain_frozen(bs, base, errp)) { | |
4774 | return -EPERM; | |
4775 | } | |
4776 | ||
e5182c1c HR |
4777 | for (i = bs; i != base; i = backing_bs(i)) { |
4778 | if (i->backing && backing_bs(i)->never_freeze) { | |
4779 | error_setg(errp, "Cannot freeze '%s' link to '%s'", | |
4780 | i->backing->name, backing_bs(i)->node_name); | |
4781 | return -EPERM; | |
4782 | } | |
4783 | } | |
4784 | ||
0f0998f6 AG |
4785 | for (i = bs; i != base; i = backing_bs(i)) { |
4786 | if (i->backing) { | |
4787 | i->backing->frozen = true; | |
4788 | } | |
2cad1ebe AG |
4789 | } |
4790 | ||
4791 | return 0; | |
4792 | } | |
4793 | ||
4794 | /* | |
4795 | * Unfreeze all backing links between @bs and @base. The caller must | |
4796 | * ensure that all links are frozen before using this function. | |
0f0998f6 | 4797 | * @base must be reachable from @bs, or NULL. |
2cad1ebe AG |
4798 | */ |
4799 | void bdrv_unfreeze_backing_chain(BlockDriverState *bs, BlockDriverState *base) | |
4800 | { | |
4801 | BlockDriverState *i; | |
4802 | ||
0f0998f6 AG |
4803 | for (i = bs; i != base; i = backing_bs(i)) { |
4804 | if (i->backing) { | |
4805 | assert(i->backing->frozen); | |
4806 | i->backing->frozen = false; | |
4807 | } | |
2cad1ebe AG |
4808 | } |
4809 | } | |
4810 | ||
6ebdcee2 JC |
4811 | /* |
4812 | * Drops images above 'base' up to and including 'top', and sets the image | |
4813 | * above 'top' to have base as its backing file. | |
4814 | * | |
4815 | * Requires that the overlay to 'top' is opened r/w, so that the backing file | |
4816 | * information in 'bs' can be properly updated. | |
4817 | * | |
4818 | * E.g., this will convert the following chain: | |
4819 | * bottom <- base <- intermediate <- top <- active | |
4820 | * | |
4821 | * to | |
4822 | * | |
4823 | * bottom <- base <- active | |
4824 | * | |
4825 | * It is allowed for bottom==base, in which case it converts: | |
4826 | * | |
4827 | * base <- intermediate <- top <- active | |
4828 | * | |
4829 | * to | |
4830 | * | |
4831 | * base <- active | |
4832 | * | |
54e26900 JC |
4833 | * If backing_file_str is non-NULL, it will be used when modifying top's |
4834 | * overlay image metadata. | |
4835 | * | |
6ebdcee2 JC |
4836 | * Error conditions: |
4837 | * if active == top, that is considered an error | |
4838 | * | |
4839 | */ | |
bde70715 KW |
4840 | int bdrv_drop_intermediate(BlockDriverState *top, BlockDriverState *base, |
4841 | const char *backing_file_str) | |
6ebdcee2 | 4842 | { |
6bd858b3 AG |
4843 | BlockDriverState *explicit_top = top; |
4844 | bool update_inherits_from; | |
61f09cea | 4845 | BdrvChild *c, *next; |
12fa4af6 | 4846 | Error *local_err = NULL; |
6ebdcee2 JC |
4847 | int ret = -EIO; |
4848 | ||
6858eba0 | 4849 | bdrv_ref(top); |
637d54a5 | 4850 | bdrv_subtree_drained_begin(top); |
6858eba0 | 4851 | |
6ebdcee2 JC |
4852 | if (!top->drv || !base->drv) { |
4853 | goto exit; | |
4854 | } | |
4855 | ||
5db15a57 KW |
4856 | /* Make sure that base is in the backing chain of top */ |
4857 | if (!bdrv_chain_contains(top, base)) { | |
6ebdcee2 JC |
4858 | goto exit; |
4859 | } | |
4860 | ||
2cad1ebe AG |
4861 | /* This function changes all links that point to top and makes |
4862 | * them point to base. Check that none of them is frozen. */ | |
4863 | QLIST_FOREACH(c, &top->parents, next_parent) { | |
4864 | if (c->frozen) { | |
4865 | goto exit; | |
4866 | } | |
4867 | } | |
4868 | ||
6bd858b3 AG |
4869 | /* If 'base' recursively inherits from 'top' then we should set |
4870 | * base->inherits_from to top->inherits_from after 'top' and all | |
4871 | * other intermediate nodes have been dropped. | |
4872 | * If 'top' is an implicit node (e.g. "commit_top") we should skip | |
4873 | * it because no one inherits from it. We use explicit_top for that. */ | |
4874 | while (explicit_top && explicit_top->implicit) { | |
4875 | explicit_top = backing_bs(explicit_top); | |
4876 | } | |
4877 | update_inherits_from = bdrv_inherits_from_recursive(base, explicit_top); | |
4878 | ||
6ebdcee2 | 4879 | /* success - we can delete the intermediate states, and link top->base */ |
bde70715 KW |
4880 | /* TODO Check graph modification op blockers (BLK_PERM_GRAPH_MOD) once |
4881 | * we've figured out how they should work. */ | |
f30c66ba HR |
4882 | if (!backing_file_str) { |
4883 | bdrv_refresh_filename(base); | |
4884 | backing_file_str = base->filename; | |
4885 | } | |
61f09cea KW |
4886 | |
4887 | QLIST_FOREACH_SAFE(c, &top->parents, next_parent, next) { | |
4888 | /* Check whether we are allowed to switch c from top to base */ | |
4889 | GSList *ignore_children = g_slist_prepend(NULL, c); | |
2345bde6 | 4890 | ret = bdrv_check_update_perm(base, NULL, c->perm, c->shared_perm, |
9eab1544 | 4891 | ignore_children, NULL, &local_err); |
2c860e79 | 4892 | g_slist_free(ignore_children); |
2345bde6 | 4893 | if (ret < 0) { |
61f09cea | 4894 | error_report_err(local_err); |
6858eba0 KW |
4895 | goto exit; |
4896 | } | |
12fa4af6 | 4897 | |
61f09cea | 4898 | /* If so, update the backing file path in the image file */ |
bd86fb99 HR |
4899 | if (c->klass->update_filename) { |
4900 | ret = c->klass->update_filename(c, base, backing_file_str, | |
4901 | &local_err); | |
61f09cea KW |
4902 | if (ret < 0) { |
4903 | bdrv_abort_perm_update(base); | |
4904 | error_report_err(local_err); | |
4905 | goto exit; | |
4906 | } | |
4907 | } | |
4908 | ||
4909 | /* Do the actual switch in the in-memory graph. | |
4910 | * Completes bdrv_check_update_perm() transaction internally. */ | |
4911 | bdrv_ref(base); | |
4912 | bdrv_replace_child(c, base); | |
4913 | bdrv_unref(top); | |
12fa4af6 | 4914 | } |
6ebdcee2 | 4915 | |
6bd858b3 AG |
4916 | if (update_inherits_from) { |
4917 | base->inherits_from = explicit_top->inherits_from; | |
4918 | } | |
4919 | ||
6ebdcee2 | 4920 | ret = 0; |
6ebdcee2 | 4921 | exit: |
637d54a5 | 4922 | bdrv_subtree_drained_end(top); |
6858eba0 | 4923 | bdrv_unref(top); |
6ebdcee2 JC |
4924 | return ret; |
4925 | } | |
4926 | ||
61007b31 SH |
4927 | /** |
4928 | * Length of a allocated file in bytes. Sparse files are counted by actual | |
4929 | * allocated space. Return < 0 if error or unknown. | |
4930 | */ | |
4931 | int64_t bdrv_get_allocated_file_size(BlockDriverState *bs) | |
71d0770c | 4932 | { |
61007b31 SH |
4933 | BlockDriver *drv = bs->drv; |
4934 | if (!drv) { | |
4935 | return -ENOMEDIUM; | |
8f4754ed | 4936 | } |
61007b31 SH |
4937 | if (drv->bdrv_get_allocated_file_size) { |
4938 | return drv->bdrv_get_allocated_file_size(bs); | |
4939 | } | |
4940 | if (bs->file) { | |
9a4f4c31 | 4941 | return bdrv_get_allocated_file_size(bs->file->bs); |
1c9805a3 | 4942 | } |
61007b31 | 4943 | return -ENOTSUP; |
1c9805a3 | 4944 | } |
e7a8a783 | 4945 | |
90880ff1 SH |
4946 | /* |
4947 | * bdrv_measure: | |
4948 | * @drv: Format driver | |
4949 | * @opts: Creation options for new image | |
4950 | * @in_bs: Existing image containing data for new image (may be NULL) | |
4951 | * @errp: Error object | |
4952 | * Returns: A #BlockMeasureInfo (free using qapi_free_BlockMeasureInfo()) | |
4953 | * or NULL on error | |
4954 | * | |
4955 | * Calculate file size required to create a new image. | |
4956 | * | |
4957 | * If @in_bs is given then space for allocated clusters and zero clusters | |
4958 | * from that image are included in the calculation. If @opts contains a | |
4959 | * backing file that is shared by @in_bs then backing clusters may be omitted | |
4960 | * from the calculation. | |
4961 | * | |
4962 | * If @in_bs is NULL then the calculation includes no allocated clusters | |
4963 | * unless a preallocation option is given in @opts. | |
4964 | * | |
4965 | * Note that @in_bs may use a different BlockDriver from @drv. | |
4966 | * | |
4967 | * If an error occurs the @errp pointer is set. | |
4968 | */ | |
4969 | BlockMeasureInfo *bdrv_measure(BlockDriver *drv, QemuOpts *opts, | |
4970 | BlockDriverState *in_bs, Error **errp) | |
4971 | { | |
4972 | if (!drv->bdrv_measure) { | |
4973 | error_setg(errp, "Block driver '%s' does not support size measurement", | |
4974 | drv->format_name); | |
4975 | return NULL; | |
4976 | } | |
4977 | ||
4978 | return drv->bdrv_measure(opts, in_bs, errp); | |
4979 | } | |
4980 | ||
61007b31 SH |
4981 | /** |
4982 | * Return number of sectors on success, -errno on error. | |
1c9805a3 | 4983 | */ |
61007b31 | 4984 | int64_t bdrv_nb_sectors(BlockDriverState *bs) |
1c9805a3 | 4985 | { |
61007b31 | 4986 | BlockDriver *drv = bs->drv; |
498e386c | 4987 | |
61007b31 SH |
4988 | if (!drv) |
4989 | return -ENOMEDIUM; | |
2572b37a | 4990 | |
61007b31 SH |
4991 | if (drv->has_variable_length) { |
4992 | int ret = refresh_total_sectors(bs, bs->total_sectors); | |
4993 | if (ret < 0) { | |
4994 | return ret; | |
1c9805a3 SH |
4995 | } |
4996 | } | |
61007b31 | 4997 | return bs->total_sectors; |
1c9805a3 | 4998 | } |
b338082b | 4999 | |
61007b31 SH |
5000 | /** |
5001 | * Return length in bytes on success, -errno on error. | |
5002 | * The length is always a multiple of BDRV_SECTOR_SIZE. | |
8d3b1a2d | 5003 | */ |
61007b31 | 5004 | int64_t bdrv_getlength(BlockDriverState *bs) |
8d3b1a2d | 5005 | { |
61007b31 | 5006 | int64_t ret = bdrv_nb_sectors(bs); |
8d3b1a2d | 5007 | |
4a9c9ea0 | 5008 | ret = ret > INT64_MAX / BDRV_SECTOR_SIZE ? -EFBIG : ret; |
61007b31 | 5009 | return ret < 0 ? ret : ret * BDRV_SECTOR_SIZE; |
fc01f7e7 FB |
5010 | } |
5011 | ||
61007b31 SH |
5012 | /* return 0 as number of sectors if no device present or error */ |
5013 | void bdrv_get_geometry(BlockDriverState *bs, uint64_t *nb_sectors_ptr) | |
07d27a44 | 5014 | { |
61007b31 | 5015 | int64_t nb_sectors = bdrv_nb_sectors(bs); |
07d27a44 | 5016 | |
61007b31 | 5017 | *nb_sectors_ptr = nb_sectors < 0 ? 0 : nb_sectors; |
07d27a44 MA |
5018 | } |
5019 | ||
54115412 | 5020 | bool bdrv_is_sg(BlockDriverState *bs) |
f08145fe | 5021 | { |
61007b31 | 5022 | return bs->sg; |
f08145fe KW |
5023 | } |
5024 | ||
54115412 | 5025 | bool bdrv_is_encrypted(BlockDriverState *bs) |
fc3959e4 | 5026 | { |
760e0063 | 5027 | if (bs->backing && bs->backing->bs->encrypted) { |
54115412 | 5028 | return true; |
760e0063 | 5029 | } |
61007b31 | 5030 | return bs->encrypted; |
fc3959e4 FZ |
5031 | } |
5032 | ||
61007b31 | 5033 | const char *bdrv_get_format_name(BlockDriverState *bs) |
40b4f539 | 5034 | { |
61007b31 | 5035 | return bs->drv ? bs->drv->format_name : NULL; |
40b4f539 KW |
5036 | } |
5037 | ||
61007b31 | 5038 | static int qsort_strcmp(const void *a, const void *b) |
40b4f539 | 5039 | { |
ceff5bd7 | 5040 | return strcmp(*(char *const *)a, *(char *const *)b); |
40b4f539 KW |
5041 | } |
5042 | ||
61007b31 | 5043 | void bdrv_iterate_format(void (*it)(void *opaque, const char *name), |
9ac404c5 | 5044 | void *opaque, bool read_only) |
40b4f539 | 5045 | { |
61007b31 SH |
5046 | BlockDriver *drv; |
5047 | int count = 0; | |
5048 | int i; | |
5049 | const char **formats = NULL; | |
40b4f539 | 5050 | |
61007b31 SH |
5051 | QLIST_FOREACH(drv, &bdrv_drivers, list) { |
5052 | if (drv->format_name) { | |
5053 | bool found = false; | |
5054 | int i = count; | |
9ac404c5 AS |
5055 | |
5056 | if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv, read_only)) { | |
5057 | continue; | |
5058 | } | |
5059 | ||
61007b31 SH |
5060 | while (formats && i && !found) { |
5061 | found = !strcmp(formats[--i], drv->format_name); | |
5062 | } | |
e2a305fb | 5063 | |
61007b31 SH |
5064 | if (!found) { |
5065 | formats = g_renew(const char *, formats, count + 1); | |
5066 | formats[count++] = drv->format_name; | |
5067 | } | |
6c5a42ac | 5068 | } |
61007b31 | 5069 | } |
6c5a42ac | 5070 | |
eb0df69f HR |
5071 | for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); i++) { |
5072 | const char *format_name = block_driver_modules[i].format_name; | |
5073 | ||
5074 | if (format_name) { | |
5075 | bool found = false; | |
5076 | int j = count; | |
5077 | ||
9ac404c5 AS |
5078 | if (use_bdrv_whitelist && |
5079 | !bdrv_format_is_whitelisted(format_name, read_only)) { | |
5080 | continue; | |
5081 | } | |
5082 | ||
eb0df69f HR |
5083 | while (formats && j && !found) { |
5084 | found = !strcmp(formats[--j], format_name); | |
5085 | } | |
5086 | ||
5087 | if (!found) { | |
5088 | formats = g_renew(const char *, formats, count + 1); | |
5089 | formats[count++] = format_name; | |
5090 | } | |
5091 | } | |
5092 | } | |
5093 | ||
61007b31 | 5094 | qsort(formats, count, sizeof(formats[0]), qsort_strcmp); |
40b4f539 | 5095 | |
61007b31 SH |
5096 | for (i = 0; i < count; i++) { |
5097 | it(opaque, formats[i]); | |
5098 | } | |
40b4f539 | 5099 | |
61007b31 SH |
5100 | g_free(formats); |
5101 | } | |
40b4f539 | 5102 | |
61007b31 SH |
5103 | /* This function is to find a node in the bs graph */ |
5104 | BlockDriverState *bdrv_find_node(const char *node_name) | |
5105 | { | |
5106 | BlockDriverState *bs; | |
391827eb | 5107 | |
61007b31 | 5108 | assert(node_name); |
40b4f539 | 5109 | |
61007b31 SH |
5110 | QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) { |
5111 | if (!strcmp(node_name, bs->node_name)) { | |
5112 | return bs; | |
40b4f539 KW |
5113 | } |
5114 | } | |
61007b31 | 5115 | return NULL; |
40b4f539 KW |
5116 | } |
5117 | ||
61007b31 | 5118 | /* Put this QMP function here so it can access the static graph_bdrv_states. */ |
facda544 PK |
5119 | BlockDeviceInfoList *bdrv_named_nodes_list(bool flat, |
5120 | Error **errp) | |
40b4f539 | 5121 | { |
61007b31 SH |
5122 | BlockDeviceInfoList *list, *entry; |
5123 | BlockDriverState *bs; | |
40b4f539 | 5124 | |
61007b31 SH |
5125 | list = NULL; |
5126 | QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) { | |
facda544 | 5127 | BlockDeviceInfo *info = bdrv_block_device_info(NULL, bs, flat, errp); |
61007b31 SH |
5128 | if (!info) { |
5129 | qapi_free_BlockDeviceInfoList(list); | |
5130 | return NULL; | |
301db7c2 | 5131 | } |
61007b31 SH |
5132 | entry = g_malloc0(sizeof(*entry)); |
5133 | entry->value = info; | |
5134 | entry->next = list; | |
5135 | list = entry; | |
301db7c2 RH |
5136 | } |
5137 | ||
61007b31 SH |
5138 | return list; |
5139 | } | |
40b4f539 | 5140 | |
5d3b4e99 VSO |
5141 | #define QAPI_LIST_ADD(list, element) do { \ |
5142 | typeof(list) _tmp = g_new(typeof(*(list)), 1); \ | |
5143 | _tmp->value = (element); \ | |
5144 | _tmp->next = (list); \ | |
5145 | (list) = _tmp; \ | |
5146 | } while (0) | |
5147 | ||
5148 | typedef struct XDbgBlockGraphConstructor { | |
5149 | XDbgBlockGraph *graph; | |
5150 | GHashTable *graph_nodes; | |
5151 | } XDbgBlockGraphConstructor; | |
5152 | ||
5153 | static XDbgBlockGraphConstructor *xdbg_graph_new(void) | |
5154 | { | |
5155 | XDbgBlockGraphConstructor *gr = g_new(XDbgBlockGraphConstructor, 1); | |
5156 | ||
5157 | gr->graph = g_new0(XDbgBlockGraph, 1); | |
5158 | gr->graph_nodes = g_hash_table_new(NULL, NULL); | |
5159 | ||
5160 | return gr; | |
5161 | } | |
5162 | ||
5163 | static XDbgBlockGraph *xdbg_graph_finalize(XDbgBlockGraphConstructor *gr) | |
5164 | { | |
5165 | XDbgBlockGraph *graph = gr->graph; | |
5166 | ||
5167 | g_hash_table_destroy(gr->graph_nodes); | |
5168 | g_free(gr); | |
5169 | ||
5170 | return graph; | |
5171 | } | |
5172 | ||
5173 | static uintptr_t xdbg_graph_node_num(XDbgBlockGraphConstructor *gr, void *node) | |
5174 | { | |
5175 | uintptr_t ret = (uintptr_t)g_hash_table_lookup(gr->graph_nodes, node); | |
5176 | ||
5177 | if (ret != 0) { | |
5178 | return ret; | |
5179 | } | |
5180 | ||
5181 | /* | |
5182 | * Start counting from 1, not 0, because 0 interferes with not-found (NULL) | |
5183 | * answer of g_hash_table_lookup. | |
5184 | */ | |
5185 | ret = g_hash_table_size(gr->graph_nodes) + 1; | |
5186 | g_hash_table_insert(gr->graph_nodes, node, (void *)ret); | |
5187 | ||
5188 | return ret; | |
5189 | } | |
5190 | ||
5191 | static void xdbg_graph_add_node(XDbgBlockGraphConstructor *gr, void *node, | |
5192 | XDbgBlockGraphNodeType type, const char *name) | |
5193 | { | |
5194 | XDbgBlockGraphNode *n; | |
5195 | ||
5196 | n = g_new0(XDbgBlockGraphNode, 1); | |
5197 | ||
5198 | n->id = xdbg_graph_node_num(gr, node); | |
5199 | n->type = type; | |
5200 | n->name = g_strdup(name); | |
5201 | ||
5202 | QAPI_LIST_ADD(gr->graph->nodes, n); | |
5203 | } | |
5204 | ||
5205 | static void xdbg_graph_add_edge(XDbgBlockGraphConstructor *gr, void *parent, | |
5206 | const BdrvChild *child) | |
5207 | { | |
cdb1cec8 | 5208 | BlockPermission qapi_perm; |
5d3b4e99 VSO |
5209 | XDbgBlockGraphEdge *edge; |
5210 | ||
5d3b4e99 VSO |
5211 | edge = g_new0(XDbgBlockGraphEdge, 1); |
5212 | ||
5213 | edge->parent = xdbg_graph_node_num(gr, parent); | |
5214 | edge->child = xdbg_graph_node_num(gr, child->bs); | |
5215 | edge->name = g_strdup(child->name); | |
5216 | ||
cdb1cec8 HR |
5217 | for (qapi_perm = 0; qapi_perm < BLOCK_PERMISSION__MAX; qapi_perm++) { |
5218 | uint64_t flag = bdrv_qapi_perm_to_blk_perm(qapi_perm); | |
5219 | ||
5220 | if (flag & child->perm) { | |
5221 | QAPI_LIST_ADD(edge->perm, qapi_perm); | |
5d3b4e99 | 5222 | } |
cdb1cec8 HR |
5223 | if (flag & child->shared_perm) { |
5224 | QAPI_LIST_ADD(edge->shared_perm, qapi_perm); | |
5d3b4e99 VSO |
5225 | } |
5226 | } | |
5227 | ||
5228 | QAPI_LIST_ADD(gr->graph->edges, edge); | |
5229 | } | |
5230 | ||
5231 | ||
5232 | XDbgBlockGraph *bdrv_get_xdbg_block_graph(Error **errp) | |
5233 | { | |
5234 | BlockBackend *blk; | |
5235 | BlockJob *job; | |
5236 | BlockDriverState *bs; | |
5237 | BdrvChild *child; | |
5238 | XDbgBlockGraphConstructor *gr = xdbg_graph_new(); | |
5239 | ||
5240 | for (blk = blk_all_next(NULL); blk; blk = blk_all_next(blk)) { | |
5241 | char *allocated_name = NULL; | |
5242 | const char *name = blk_name(blk); | |
5243 | ||
5244 | if (!*name) { | |
5245 | name = allocated_name = blk_get_attached_dev_id(blk); | |
5246 | } | |
5247 | xdbg_graph_add_node(gr, blk, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_BACKEND, | |
5248 | name); | |
5249 | g_free(allocated_name); | |
5250 | if (blk_root(blk)) { | |
5251 | xdbg_graph_add_edge(gr, blk, blk_root(blk)); | |
5252 | } | |
5253 | } | |
5254 | ||
5255 | for (job = block_job_next(NULL); job; job = block_job_next(job)) { | |
5256 | GSList *el; | |
5257 | ||
5258 | xdbg_graph_add_node(gr, job, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_JOB, | |
5259 | job->job.id); | |
5260 | for (el = job->nodes; el; el = el->next) { | |
5261 | xdbg_graph_add_edge(gr, job, (BdrvChild *)el->data); | |
5262 | } | |
5263 | } | |
5264 | ||
5265 | QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) { | |
5266 | xdbg_graph_add_node(gr, bs, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_DRIVER, | |
5267 | bs->node_name); | |
5268 | QLIST_FOREACH(child, &bs->children, next) { | |
5269 | xdbg_graph_add_edge(gr, bs, child); | |
5270 | } | |
5271 | } | |
5272 | ||
5273 | return xdbg_graph_finalize(gr); | |
5274 | } | |
5275 | ||
61007b31 SH |
5276 | BlockDriverState *bdrv_lookup_bs(const char *device, |
5277 | const char *node_name, | |
5278 | Error **errp) | |
5279 | { | |
5280 | BlockBackend *blk; | |
5281 | BlockDriverState *bs; | |
40b4f539 | 5282 | |
61007b31 SH |
5283 | if (device) { |
5284 | blk = blk_by_name(device); | |
40b4f539 | 5285 | |
61007b31 | 5286 | if (blk) { |
9f4ed6fb AG |
5287 | bs = blk_bs(blk); |
5288 | if (!bs) { | |
5433c24f | 5289 | error_setg(errp, "Device '%s' has no medium", device); |
5433c24f HR |
5290 | } |
5291 | ||
9f4ed6fb | 5292 | return bs; |
61007b31 SH |
5293 | } |
5294 | } | |
40b4f539 | 5295 | |
61007b31 SH |
5296 | if (node_name) { |
5297 | bs = bdrv_find_node(node_name); | |
6d519a5f | 5298 | |
61007b31 SH |
5299 | if (bs) { |
5300 | return bs; | |
5301 | } | |
40b4f539 KW |
5302 | } |
5303 | ||
61007b31 SH |
5304 | error_setg(errp, "Cannot find device=%s nor node_name=%s", |
5305 | device ? device : "", | |
5306 | node_name ? node_name : ""); | |
5307 | return NULL; | |
40b4f539 KW |
5308 | } |
5309 | ||
61007b31 SH |
5310 | /* If 'base' is in the same chain as 'top', return true. Otherwise, |
5311 | * return false. If either argument is NULL, return false. */ | |
5312 | bool bdrv_chain_contains(BlockDriverState *top, BlockDriverState *base) | |
83f64091 | 5313 | { |
61007b31 | 5314 | while (top && top != base) { |
760e0063 | 5315 | top = backing_bs(top); |
02c50efe | 5316 | } |
61007b31 SH |
5317 | |
5318 | return top != NULL; | |
02c50efe FZ |
5319 | } |
5320 | ||
61007b31 | 5321 | BlockDriverState *bdrv_next_node(BlockDriverState *bs) |
02c50efe | 5322 | { |
61007b31 SH |
5323 | if (!bs) { |
5324 | return QTAILQ_FIRST(&graph_bdrv_states); | |
02c50efe | 5325 | } |
61007b31 | 5326 | return QTAILQ_NEXT(bs, node_list); |
83f64091 FB |
5327 | } |
5328 | ||
0f12264e KW |
5329 | BlockDriverState *bdrv_next_all_states(BlockDriverState *bs) |
5330 | { | |
5331 | if (!bs) { | |
5332 | return QTAILQ_FIRST(&all_bdrv_states); | |
5333 | } | |
5334 | return QTAILQ_NEXT(bs, bs_list); | |
5335 | } | |
5336 | ||
61007b31 | 5337 | const char *bdrv_get_node_name(const BlockDriverState *bs) |
83f64091 | 5338 | { |
61007b31 | 5339 | return bs->node_name; |
beac80cd FB |
5340 | } |
5341 | ||
1f0c461b | 5342 | const char *bdrv_get_parent_name(const BlockDriverState *bs) |
4c265bf9 KW |
5343 | { |
5344 | BdrvChild *c; | |
5345 | const char *name; | |
5346 | ||
5347 | /* If multiple parents have a name, just pick the first one. */ | |
5348 | QLIST_FOREACH(c, &bs->parents, next_parent) { | |
bd86fb99 HR |
5349 | if (c->klass->get_name) { |
5350 | name = c->klass->get_name(c); | |
4c265bf9 KW |
5351 | if (name && *name) { |
5352 | return name; | |
5353 | } | |
5354 | } | |
5355 | } | |
5356 | ||
5357 | return NULL; | |
5358 | } | |
5359 | ||
61007b31 SH |
5360 | /* TODO check what callers really want: bs->node_name or blk_name() */ |
5361 | const char *bdrv_get_device_name(const BlockDriverState *bs) | |
beac80cd | 5362 | { |
4c265bf9 | 5363 | return bdrv_get_parent_name(bs) ?: ""; |
f141eafe | 5364 | } |
83f64091 | 5365 | |
61007b31 SH |
5366 | /* This can be used to identify nodes that might not have a device |
5367 | * name associated. Since node and device names live in the same | |
5368 | * namespace, the result is unambiguous. The exception is if both are | |
5369 | * absent, then this returns an empty (non-null) string. */ | |
5370 | const char *bdrv_get_device_or_node_name(const BlockDriverState *bs) | |
f141eafe | 5371 | { |
4c265bf9 | 5372 | return bdrv_get_parent_name(bs) ?: bs->node_name; |
beac80cd | 5373 | } |
beac80cd | 5374 | |
61007b31 | 5375 | int bdrv_get_flags(BlockDriverState *bs) |
0b5a2445 | 5376 | { |
61007b31 | 5377 | return bs->open_flags; |
0b5a2445 PB |
5378 | } |
5379 | ||
61007b31 | 5380 | int bdrv_has_zero_init_1(BlockDriverState *bs) |
68485420 | 5381 | { |
61007b31 | 5382 | return 1; |
0b5a2445 PB |
5383 | } |
5384 | ||
61007b31 | 5385 | int bdrv_has_zero_init(BlockDriverState *bs) |
0b5a2445 | 5386 | { |
d470ad42 HR |
5387 | if (!bs->drv) { |
5388 | return 0; | |
5389 | } | |
0b5a2445 | 5390 | |
61007b31 SH |
5391 | /* If BS is a copy on write image, it is initialized to |
5392 | the contents of the base image, which may not be zeroes. */ | |
760e0063 | 5393 | if (bs->backing) { |
61007b31 SH |
5394 | return 0; |
5395 | } | |
5396 | if (bs->drv->bdrv_has_zero_init) { | |
5397 | return bs->drv->bdrv_has_zero_init(bs); | |
0b5a2445 | 5398 | } |
5a612c00 MP |
5399 | if (bs->file && bs->drv->is_filter) { |
5400 | return bdrv_has_zero_init(bs->file->bs); | |
5401 | } | |
61007b31 SH |
5402 | |
5403 | /* safe default */ | |
5404 | return 0; | |
68485420 KW |
5405 | } |
5406 | ||
61007b31 | 5407 | bool bdrv_can_write_zeroes_with_unmap(BlockDriverState *bs) |
68485420 | 5408 | { |
2f0342ef | 5409 | if (!(bs->open_flags & BDRV_O_UNMAP)) { |
61007b31 SH |
5410 | return false; |
5411 | } | |
68485420 | 5412 | |
e24d813b | 5413 | return bs->supported_zero_flags & BDRV_REQ_MAY_UNMAP; |
68485420 KW |
5414 | } |
5415 | ||
61007b31 SH |
5416 | void bdrv_get_backing_filename(BlockDriverState *bs, |
5417 | char *filename, int filename_size) | |
016f5cf6 | 5418 | { |
61007b31 SH |
5419 | pstrcpy(filename, filename_size, bs->backing_file); |
5420 | } | |
d318aea9 | 5421 | |
61007b31 SH |
5422 | int bdrv_get_info(BlockDriverState *bs, BlockDriverInfo *bdi) |
5423 | { | |
5424 | BlockDriver *drv = bs->drv; | |
5a612c00 MP |
5425 | /* if bs->drv == NULL, bs is closed, so there's nothing to do here */ |
5426 | if (!drv) { | |
61007b31 | 5427 | return -ENOMEDIUM; |
5a612c00 MP |
5428 | } |
5429 | if (!drv->bdrv_get_info) { | |
5430 | if (bs->file && drv->is_filter) { | |
5431 | return bdrv_get_info(bs->file->bs, bdi); | |
5432 | } | |
61007b31 | 5433 | return -ENOTSUP; |
5a612c00 | 5434 | } |
61007b31 SH |
5435 | memset(bdi, 0, sizeof(*bdi)); |
5436 | return drv->bdrv_get_info(bs, bdi); | |
5437 | } | |
016f5cf6 | 5438 | |
1bf6e9ca AS |
5439 | ImageInfoSpecific *bdrv_get_specific_info(BlockDriverState *bs, |
5440 | Error **errp) | |
61007b31 SH |
5441 | { |
5442 | BlockDriver *drv = bs->drv; | |
5443 | if (drv && drv->bdrv_get_specific_info) { | |
1bf6e9ca | 5444 | return drv->bdrv_get_specific_info(bs, errp); |
61007b31 SH |
5445 | } |
5446 | return NULL; | |
016f5cf6 AG |
5447 | } |
5448 | ||
d9245599 AN |
5449 | BlockStatsSpecific *bdrv_get_specific_stats(BlockDriverState *bs) |
5450 | { | |
5451 | BlockDriver *drv = bs->drv; | |
5452 | if (!drv || !drv->bdrv_get_specific_stats) { | |
5453 | return NULL; | |
5454 | } | |
5455 | return drv->bdrv_get_specific_stats(bs); | |
5456 | } | |
5457 | ||
a31939e6 | 5458 | void bdrv_debug_event(BlockDriverState *bs, BlkdebugEvent event) |
4265d620 | 5459 | { |
61007b31 SH |
5460 | if (!bs || !bs->drv || !bs->drv->bdrv_debug_event) { |
5461 | return; | |
5462 | } | |
4265d620 | 5463 | |
61007b31 | 5464 | bs->drv->bdrv_debug_event(bs, event); |
4265d620 PB |
5465 | } |
5466 | ||
d10529a2 | 5467 | static BlockDriverState *bdrv_find_debug_node(BlockDriverState *bs) |
4265d620 | 5468 | { |
61007b31 | 5469 | while (bs && bs->drv && !bs->drv->bdrv_debug_breakpoint) { |
d10529a2 VSO |
5470 | if (bs->file) { |
5471 | bs = bs->file->bs; | |
5472 | continue; | |
5473 | } | |
5474 | ||
5475 | if (bs->drv->is_filter && bs->backing) { | |
5476 | bs = bs->backing->bs; | |
5477 | continue; | |
5478 | } | |
5479 | ||
5480 | break; | |
61007b31 | 5481 | } |
4265d620 | 5482 | |
61007b31 | 5483 | if (bs && bs->drv && bs->drv->bdrv_debug_breakpoint) { |
d10529a2 VSO |
5484 | assert(bs->drv->bdrv_debug_remove_breakpoint); |
5485 | return bs; | |
5486 | } | |
5487 | ||
5488 | return NULL; | |
5489 | } | |
5490 | ||
5491 | int bdrv_debug_breakpoint(BlockDriverState *bs, const char *event, | |
5492 | const char *tag) | |
5493 | { | |
5494 | bs = bdrv_find_debug_node(bs); | |
5495 | if (bs) { | |
61007b31 SH |
5496 | return bs->drv->bdrv_debug_breakpoint(bs, event, tag); |
5497 | } | |
4265d620 | 5498 | |
61007b31 | 5499 | return -ENOTSUP; |
4265d620 PB |
5500 | } |
5501 | ||
61007b31 | 5502 | int bdrv_debug_remove_breakpoint(BlockDriverState *bs, const char *tag) |
ea2384d3 | 5503 | { |
d10529a2 VSO |
5504 | bs = bdrv_find_debug_node(bs); |
5505 | if (bs) { | |
61007b31 SH |
5506 | return bs->drv->bdrv_debug_remove_breakpoint(bs, tag); |
5507 | } | |
5508 | ||
5509 | return -ENOTSUP; | |
eb852011 MA |
5510 | } |
5511 | ||
61007b31 | 5512 | int bdrv_debug_resume(BlockDriverState *bs, const char *tag) |
ce1a14dc | 5513 | { |
61007b31 | 5514 | while (bs && (!bs->drv || !bs->drv->bdrv_debug_resume)) { |
9a4f4c31 | 5515 | bs = bs->file ? bs->file->bs : NULL; |
61007b31 | 5516 | } |
ce1a14dc | 5517 | |
61007b31 SH |
5518 | if (bs && bs->drv && bs->drv->bdrv_debug_resume) { |
5519 | return bs->drv->bdrv_debug_resume(bs, tag); | |
5520 | } | |
ce1a14dc | 5521 | |
61007b31 | 5522 | return -ENOTSUP; |
f197fe2b FZ |
5523 | } |
5524 | ||
61007b31 | 5525 | bool bdrv_debug_is_suspended(BlockDriverState *bs, const char *tag) |
ce1a14dc | 5526 | { |
61007b31 | 5527 | while (bs && bs->drv && !bs->drv->bdrv_debug_is_suspended) { |
9a4f4c31 | 5528 | bs = bs->file ? bs->file->bs : NULL; |
f197fe2b | 5529 | } |
19cb3738 | 5530 | |
61007b31 SH |
5531 | if (bs && bs->drv && bs->drv->bdrv_debug_is_suspended) { |
5532 | return bs->drv->bdrv_debug_is_suspended(bs, tag); | |
5533 | } | |
f9f05dc5 | 5534 | |
61007b31 SH |
5535 | return false; |
5536 | } | |
f9f05dc5 | 5537 | |
61007b31 SH |
5538 | /* backing_file can either be relative, or absolute, or a protocol. If it is |
5539 | * relative, it must be relative to the chain. So, passing in bs->filename | |
5540 | * from a BDS as backing_file should not be done, as that may be relative to | |
5541 | * the CWD rather than the chain. */ | |
5542 | BlockDriverState *bdrv_find_backing_image(BlockDriverState *bs, | |
5543 | const char *backing_file) | |
f9f05dc5 | 5544 | { |
61007b31 SH |
5545 | char *filename_full = NULL; |
5546 | char *backing_file_full = NULL; | |
5547 | char *filename_tmp = NULL; | |
5548 | int is_protocol = 0; | |
5549 | BlockDriverState *curr_bs = NULL; | |
5550 | BlockDriverState *retval = NULL; | |
f9f05dc5 | 5551 | |
61007b31 SH |
5552 | if (!bs || !bs->drv || !backing_file) { |
5553 | return NULL; | |
f9f05dc5 KW |
5554 | } |
5555 | ||
61007b31 SH |
5556 | filename_full = g_malloc(PATH_MAX); |
5557 | backing_file_full = g_malloc(PATH_MAX); | |
f9f05dc5 | 5558 | |
61007b31 | 5559 | is_protocol = path_has_protocol(backing_file); |
f9f05dc5 | 5560 | |
760e0063 | 5561 | for (curr_bs = bs; curr_bs->backing; curr_bs = curr_bs->backing->bs) { |
f9f05dc5 | 5562 | |
61007b31 SH |
5563 | /* If either of the filename paths is actually a protocol, then |
5564 | * compare unmodified paths; otherwise make paths relative */ | |
5565 | if (is_protocol || path_has_protocol(curr_bs->backing_file)) { | |
6b6833c1 HR |
5566 | char *backing_file_full_ret; |
5567 | ||
61007b31 | 5568 | if (strcmp(backing_file, curr_bs->backing_file) == 0) { |
760e0063 | 5569 | retval = curr_bs->backing->bs; |
61007b31 SH |
5570 | break; |
5571 | } | |
418661e0 | 5572 | /* Also check against the full backing filename for the image */ |
6b6833c1 HR |
5573 | backing_file_full_ret = bdrv_get_full_backing_filename(curr_bs, |
5574 | NULL); | |
5575 | if (backing_file_full_ret) { | |
5576 | bool equal = strcmp(backing_file, backing_file_full_ret) == 0; | |
5577 | g_free(backing_file_full_ret); | |
5578 | if (equal) { | |
418661e0 JC |
5579 | retval = curr_bs->backing->bs; |
5580 | break; | |
5581 | } | |
418661e0 | 5582 | } |
61007b31 SH |
5583 | } else { |
5584 | /* If not an absolute filename path, make it relative to the current | |
5585 | * image's filename path */ | |
2d9158ce HR |
5586 | filename_tmp = bdrv_make_absolute_filename(curr_bs, backing_file, |
5587 | NULL); | |
5588 | /* We are going to compare canonicalized absolute pathnames */ | |
5589 | if (!filename_tmp || !realpath(filename_tmp, filename_full)) { | |
5590 | g_free(filename_tmp); | |
61007b31 SH |
5591 | continue; |
5592 | } | |
2d9158ce | 5593 | g_free(filename_tmp); |
07f07615 | 5594 | |
61007b31 SH |
5595 | /* We need to make sure the backing filename we are comparing against |
5596 | * is relative to the current image filename (or absolute) */ | |
2d9158ce HR |
5597 | filename_tmp = bdrv_get_full_backing_filename(curr_bs, NULL); |
5598 | if (!filename_tmp || !realpath(filename_tmp, backing_file_full)) { | |
5599 | g_free(filename_tmp); | |
61007b31 SH |
5600 | continue; |
5601 | } | |
2d9158ce | 5602 | g_free(filename_tmp); |
eb489bb1 | 5603 | |
61007b31 | 5604 | if (strcmp(backing_file_full, filename_full) == 0) { |
760e0063 | 5605 | retval = curr_bs->backing->bs; |
61007b31 SH |
5606 | break; |
5607 | } | |
5608 | } | |
eb489bb1 KW |
5609 | } |
5610 | ||
61007b31 SH |
5611 | g_free(filename_full); |
5612 | g_free(backing_file_full); | |
61007b31 SH |
5613 | return retval; |
5614 | } | |
5615 | ||
61007b31 SH |
5616 | void bdrv_init(void) |
5617 | { | |
5618 | module_call_init(MODULE_INIT_BLOCK); | |
5619 | } | |
29cdb251 | 5620 | |
61007b31 SH |
5621 | void bdrv_init_with_whitelist(void) |
5622 | { | |
5623 | use_bdrv_whitelist = 1; | |
5624 | bdrv_init(); | |
07f07615 PB |
5625 | } |
5626 | ||
2b148f39 PB |
5627 | static void coroutine_fn bdrv_co_invalidate_cache(BlockDriverState *bs, |
5628 | Error **errp) | |
0f15423c | 5629 | { |
4417ab7a | 5630 | BdrvChild *child, *parent; |
9c5e6594 | 5631 | uint64_t perm, shared_perm; |
5a8a30db KW |
5632 | Error *local_err = NULL; |
5633 | int ret; | |
9c98f145 | 5634 | BdrvDirtyBitmap *bm; |
5a8a30db | 5635 | |
3456a8d1 KW |
5636 | if (!bs->drv) { |
5637 | return; | |
5638 | } | |
5639 | ||
16e977d5 | 5640 | QLIST_FOREACH(child, &bs->children, next) { |
2b148f39 | 5641 | bdrv_co_invalidate_cache(child->bs, &local_err); |
0d1c5c91 | 5642 | if (local_err) { |
0d1c5c91 FZ |
5643 | error_propagate(errp, local_err); |
5644 | return; | |
5645 | } | |
5a8a30db | 5646 | } |
0d1c5c91 | 5647 | |
dafe0960 KW |
5648 | /* |
5649 | * Update permissions, they may differ for inactive nodes. | |
5650 | * | |
5651 | * Note that the required permissions of inactive images are always a | |
5652 | * subset of the permissions required after activating the image. This | |
5653 | * allows us to just get the permissions upfront without restricting | |
5654 | * drv->bdrv_invalidate_cache(). | |
5655 | * | |
5656 | * It also means that in error cases, we don't have to try and revert to | |
5657 | * the old permissions (which is an operation that could fail, too). We can | |
5658 | * just keep the extended permissions for the next time that an activation | |
5659 | * of the image is tried. | |
5660 | */ | |
7bb4941a KW |
5661 | if (bs->open_flags & BDRV_O_INACTIVE) { |
5662 | bs->open_flags &= ~BDRV_O_INACTIVE; | |
5663 | bdrv_get_cumulative_perm(bs, &perm, &shared_perm); | |
668f62ec | 5664 | ret = bdrv_check_perm(bs, NULL, perm, shared_perm, NULL, NULL, errp); |
7bb4941a | 5665 | if (ret < 0) { |
0d1c5c91 | 5666 | bs->open_flags |= BDRV_O_INACTIVE; |
0d1c5c91 FZ |
5667 | return; |
5668 | } | |
7bb4941a | 5669 | bdrv_set_perm(bs, perm, shared_perm); |
3456a8d1 | 5670 | |
7bb4941a KW |
5671 | if (bs->drv->bdrv_co_invalidate_cache) { |
5672 | bs->drv->bdrv_co_invalidate_cache(bs, &local_err); | |
5673 | if (local_err) { | |
5674 | bs->open_flags |= BDRV_O_INACTIVE; | |
5675 | error_propagate(errp, local_err); | |
5676 | return; | |
5677 | } | |
5678 | } | |
9c98f145 | 5679 | |
7bb4941a KW |
5680 | FOR_EACH_DIRTY_BITMAP(bs, bm) { |
5681 | bdrv_dirty_bitmap_skip_store(bm, false); | |
5682 | } | |
5683 | ||
5684 | ret = refresh_total_sectors(bs, bs->total_sectors); | |
5685 | if (ret < 0) { | |
5686 | bs->open_flags |= BDRV_O_INACTIVE; | |
5687 | error_setg_errno(errp, -ret, "Could not refresh total sector count"); | |
5688 | return; | |
5689 | } | |
5a8a30db | 5690 | } |
4417ab7a KW |
5691 | |
5692 | QLIST_FOREACH(parent, &bs->parents, next_parent) { | |
bd86fb99 HR |
5693 | if (parent->klass->activate) { |
5694 | parent->klass->activate(parent, &local_err); | |
4417ab7a | 5695 | if (local_err) { |
78fc3b3a | 5696 | bs->open_flags |= BDRV_O_INACTIVE; |
4417ab7a KW |
5697 | error_propagate(errp, local_err); |
5698 | return; | |
5699 | } | |
5700 | } | |
5701 | } | |
0f15423c AL |
5702 | } |
5703 | ||
2b148f39 PB |
5704 | typedef struct InvalidateCacheCo { |
5705 | BlockDriverState *bs; | |
5706 | Error **errp; | |
5707 | bool done; | |
5708 | } InvalidateCacheCo; | |
5709 | ||
5710 | static void coroutine_fn bdrv_invalidate_cache_co_entry(void *opaque) | |
5711 | { | |
5712 | InvalidateCacheCo *ico = opaque; | |
5713 | bdrv_co_invalidate_cache(ico->bs, ico->errp); | |
5714 | ico->done = true; | |
4720cbee | 5715 | aio_wait_kick(); |
2b148f39 PB |
5716 | } |
5717 | ||
5718 | void bdrv_invalidate_cache(BlockDriverState *bs, Error **errp) | |
5719 | { | |
5720 | Coroutine *co; | |
5721 | InvalidateCacheCo ico = { | |
5722 | .bs = bs, | |
5723 | .done = false, | |
5724 | .errp = errp | |
5725 | }; | |
5726 | ||
5727 | if (qemu_in_coroutine()) { | |
5728 | /* Fast-path if already in coroutine context */ | |
5729 | bdrv_invalidate_cache_co_entry(&ico); | |
5730 | } else { | |
5731 | co = qemu_coroutine_create(bdrv_invalidate_cache_co_entry, &ico); | |
4720cbee | 5732 | bdrv_coroutine_enter(bs, co); |
2b148f39 PB |
5733 | BDRV_POLL_WHILE(bs, !ico.done); |
5734 | } | |
5735 | } | |
5736 | ||
5a8a30db | 5737 | void bdrv_invalidate_cache_all(Error **errp) |
0f15423c | 5738 | { |
7c8eece4 | 5739 | BlockDriverState *bs; |
5a8a30db | 5740 | Error *local_err = NULL; |
88be7b4b | 5741 | BdrvNextIterator it; |
0f15423c | 5742 | |
88be7b4b | 5743 | for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) { |
ed78cda3 SH |
5744 | AioContext *aio_context = bdrv_get_aio_context(bs); |
5745 | ||
5746 | aio_context_acquire(aio_context); | |
5a8a30db | 5747 | bdrv_invalidate_cache(bs, &local_err); |
ed78cda3 | 5748 | aio_context_release(aio_context); |
5a8a30db KW |
5749 | if (local_err) { |
5750 | error_propagate(errp, local_err); | |
5e003f17 | 5751 | bdrv_next_cleanup(&it); |
5a8a30db KW |
5752 | return; |
5753 | } | |
0f15423c AL |
5754 | } |
5755 | } | |
5756 | ||
9e37271f KW |
5757 | static bool bdrv_has_bds_parent(BlockDriverState *bs, bool only_active) |
5758 | { | |
5759 | BdrvChild *parent; | |
5760 | ||
5761 | QLIST_FOREACH(parent, &bs->parents, next_parent) { | |
bd86fb99 | 5762 | if (parent->klass->parent_is_bds) { |
9e37271f KW |
5763 | BlockDriverState *parent_bs = parent->opaque; |
5764 | if (!only_active || !(parent_bs->open_flags & BDRV_O_INACTIVE)) { | |
5765 | return true; | |
5766 | } | |
5767 | } | |
5768 | } | |
5769 | ||
5770 | return false; | |
5771 | } | |
5772 | ||
5773 | static int bdrv_inactivate_recurse(BlockDriverState *bs) | |
76b1c7fe | 5774 | { |
cfa1a572 | 5775 | BdrvChild *child, *parent; |
1046779e | 5776 | bool tighten_restrictions; |
9e37271f | 5777 | uint64_t perm, shared_perm; |
76b1c7fe KW |
5778 | int ret; |
5779 | ||
d470ad42 HR |
5780 | if (!bs->drv) { |
5781 | return -ENOMEDIUM; | |
5782 | } | |
5783 | ||
9e37271f KW |
5784 | /* Make sure that we don't inactivate a child before its parent. |
5785 | * It will be covered by recursion from the yet active parent. */ | |
5786 | if (bdrv_has_bds_parent(bs, true)) { | |
5787 | return 0; | |
5788 | } | |
5789 | ||
5790 | assert(!(bs->open_flags & BDRV_O_INACTIVE)); | |
5791 | ||
5792 | /* Inactivate this node */ | |
5793 | if (bs->drv->bdrv_inactivate) { | |
76b1c7fe KW |
5794 | ret = bs->drv->bdrv_inactivate(bs); |
5795 | if (ret < 0) { | |
5796 | return ret; | |
5797 | } | |
5798 | } | |
5799 | ||
9e37271f | 5800 | QLIST_FOREACH(parent, &bs->parents, next_parent) { |
bd86fb99 HR |
5801 | if (parent->klass->inactivate) { |
5802 | ret = parent->klass->inactivate(parent); | |
9e37271f KW |
5803 | if (ret < 0) { |
5804 | return ret; | |
cfa1a572 KW |
5805 | } |
5806 | } | |
9e37271f | 5807 | } |
9c5e6594 | 5808 | |
9e37271f | 5809 | bs->open_flags |= BDRV_O_INACTIVE; |
7d5b5261 | 5810 | |
9e37271f KW |
5811 | /* Update permissions, they may differ for inactive nodes */ |
5812 | bdrv_get_cumulative_perm(bs, &perm, &shared_perm); | |
1046779e HR |
5813 | ret = bdrv_check_perm(bs, NULL, perm, shared_perm, NULL, |
5814 | &tighten_restrictions, NULL); | |
5815 | assert(tighten_restrictions == false); | |
5816 | if (ret < 0) { | |
5817 | /* We only tried to loosen restrictions, so errors are not fatal */ | |
5818 | bdrv_abort_perm_update(bs); | |
5819 | } else { | |
5820 | bdrv_set_perm(bs, perm, shared_perm); | |
5821 | } | |
38701b6a | 5822 | |
9e37271f KW |
5823 | |
5824 | /* Recursively inactivate children */ | |
38701b6a | 5825 | QLIST_FOREACH(child, &bs->children, next) { |
9e37271f | 5826 | ret = bdrv_inactivate_recurse(child->bs); |
38701b6a KW |
5827 | if (ret < 0) { |
5828 | return ret; | |
5829 | } | |
5830 | } | |
5831 | ||
76b1c7fe KW |
5832 | return 0; |
5833 | } | |
5834 | ||
5835 | int bdrv_inactivate_all(void) | |
5836 | { | |
79720af6 | 5837 | BlockDriverState *bs = NULL; |
88be7b4b | 5838 | BdrvNextIterator it; |
aad0b7a0 | 5839 | int ret = 0; |
bd6458e4 | 5840 | GSList *aio_ctxs = NULL, *ctx; |
76b1c7fe | 5841 | |
88be7b4b | 5842 | for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) { |
bd6458e4 PB |
5843 | AioContext *aio_context = bdrv_get_aio_context(bs); |
5844 | ||
5845 | if (!g_slist_find(aio_ctxs, aio_context)) { | |
5846 | aio_ctxs = g_slist_prepend(aio_ctxs, aio_context); | |
5847 | aio_context_acquire(aio_context); | |
5848 | } | |
aad0b7a0 | 5849 | } |
76b1c7fe | 5850 | |
9e37271f KW |
5851 | for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) { |
5852 | /* Nodes with BDS parents are covered by recursion from the last | |
5853 | * parent that gets inactivated. Don't inactivate them a second | |
5854 | * time if that has already happened. */ | |
5855 | if (bdrv_has_bds_parent(bs, false)) { | |
5856 | continue; | |
5857 | } | |
5858 | ret = bdrv_inactivate_recurse(bs); | |
5859 | if (ret < 0) { | |
5860 | bdrv_next_cleanup(&it); | |
5861 | goto out; | |
76b1c7fe KW |
5862 | } |
5863 | } | |
5864 | ||
aad0b7a0 | 5865 | out: |
bd6458e4 PB |
5866 | for (ctx = aio_ctxs; ctx != NULL; ctx = ctx->next) { |
5867 | AioContext *aio_context = ctx->data; | |
5868 | aio_context_release(aio_context); | |
aad0b7a0 | 5869 | } |
bd6458e4 | 5870 | g_slist_free(aio_ctxs); |
aad0b7a0 FZ |
5871 | |
5872 | return ret; | |
76b1c7fe KW |
5873 | } |
5874 | ||
19cb3738 FB |
5875 | /**************************************************************/ |
5876 | /* removable device support */ | |
5877 | ||
5878 | /** | |
5879 | * Return TRUE if the media is present | |
5880 | */ | |
e031f750 | 5881 | bool bdrv_is_inserted(BlockDriverState *bs) |
19cb3738 FB |
5882 | { |
5883 | BlockDriver *drv = bs->drv; | |
28d7a789 | 5884 | BdrvChild *child; |
a1aff5bf | 5885 | |
e031f750 HR |
5886 | if (!drv) { |
5887 | return false; | |
5888 | } | |
28d7a789 HR |
5889 | if (drv->bdrv_is_inserted) { |
5890 | return drv->bdrv_is_inserted(bs); | |
5891 | } | |
5892 | QLIST_FOREACH(child, &bs->children, next) { | |
5893 | if (!bdrv_is_inserted(child->bs)) { | |
5894 | return false; | |
5895 | } | |
e031f750 | 5896 | } |
28d7a789 | 5897 | return true; |
19cb3738 FB |
5898 | } |
5899 | ||
19cb3738 FB |
5900 | /** |
5901 | * If eject_flag is TRUE, eject the media. Otherwise, close the tray | |
5902 | */ | |
f36f3949 | 5903 | void bdrv_eject(BlockDriverState *bs, bool eject_flag) |
19cb3738 FB |
5904 | { |
5905 | BlockDriver *drv = bs->drv; | |
19cb3738 | 5906 | |
822e1cd1 MA |
5907 | if (drv && drv->bdrv_eject) { |
5908 | drv->bdrv_eject(bs, eject_flag); | |
19cb3738 FB |
5909 | } |
5910 | } | |
5911 | ||
19cb3738 FB |
5912 | /** |
5913 | * Lock or unlock the media (if it is locked, the user won't be able | |
5914 | * to eject it manually). | |
5915 | */ | |
025e849a | 5916 | void bdrv_lock_medium(BlockDriverState *bs, bool locked) |
19cb3738 FB |
5917 | { |
5918 | BlockDriver *drv = bs->drv; | |
5919 | ||
025e849a | 5920 | trace_bdrv_lock_medium(bs, locked); |
b8c6d095 | 5921 | |
025e849a MA |
5922 | if (drv && drv->bdrv_lock_medium) { |
5923 | drv->bdrv_lock_medium(bs, locked); | |
19cb3738 FB |
5924 | } |
5925 | } | |
985a03b0 | 5926 | |
9fcb0251 FZ |
5927 | /* Get a reference to bs */ |
5928 | void bdrv_ref(BlockDriverState *bs) | |
5929 | { | |
5930 | bs->refcnt++; | |
5931 | } | |
5932 | ||
5933 | /* Release a previously grabbed reference to bs. | |
5934 | * If after releasing, reference count is zero, the BlockDriverState is | |
5935 | * deleted. */ | |
5936 | void bdrv_unref(BlockDriverState *bs) | |
5937 | { | |
9a4d5ca6 JC |
5938 | if (!bs) { |
5939 | return; | |
5940 | } | |
9fcb0251 FZ |
5941 | assert(bs->refcnt > 0); |
5942 | if (--bs->refcnt == 0) { | |
5943 | bdrv_delete(bs); | |
5944 | } | |
5945 | } | |
5946 | ||
fbe40ff7 FZ |
5947 | struct BdrvOpBlocker { |
5948 | Error *reason; | |
5949 | QLIST_ENTRY(BdrvOpBlocker) list; | |
5950 | }; | |
5951 | ||
5952 | bool bdrv_op_is_blocked(BlockDriverState *bs, BlockOpType op, Error **errp) | |
5953 | { | |
5954 | BdrvOpBlocker *blocker; | |
5955 | assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX); | |
5956 | if (!QLIST_EMPTY(&bs->op_blockers[op])) { | |
5957 | blocker = QLIST_FIRST(&bs->op_blockers[op]); | |
4b576648 MA |
5958 | error_propagate_prepend(errp, error_copy(blocker->reason), |
5959 | "Node '%s' is busy: ", | |
5960 | bdrv_get_device_or_node_name(bs)); | |
fbe40ff7 FZ |
5961 | return true; |
5962 | } | |
5963 | return false; | |
5964 | } | |
5965 | ||
5966 | void bdrv_op_block(BlockDriverState *bs, BlockOpType op, Error *reason) | |
5967 | { | |
5968 | BdrvOpBlocker *blocker; | |
5969 | assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX); | |
5970 | ||
5839e53b | 5971 | blocker = g_new0(BdrvOpBlocker, 1); |
fbe40ff7 FZ |
5972 | blocker->reason = reason; |
5973 | QLIST_INSERT_HEAD(&bs->op_blockers[op], blocker, list); | |
5974 | } | |
5975 | ||
5976 | void bdrv_op_unblock(BlockDriverState *bs, BlockOpType op, Error *reason) | |
5977 | { | |
5978 | BdrvOpBlocker *blocker, *next; | |
5979 | assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX); | |
5980 | QLIST_FOREACH_SAFE(blocker, &bs->op_blockers[op], list, next) { | |
5981 | if (blocker->reason == reason) { | |
5982 | QLIST_REMOVE(blocker, list); | |
5983 | g_free(blocker); | |
5984 | } | |
5985 | } | |
5986 | } | |
5987 | ||
5988 | void bdrv_op_block_all(BlockDriverState *bs, Error *reason) | |
5989 | { | |
5990 | int i; | |
5991 | for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) { | |
5992 | bdrv_op_block(bs, i, reason); | |
5993 | } | |
5994 | } | |
5995 | ||
5996 | void bdrv_op_unblock_all(BlockDriverState *bs, Error *reason) | |
5997 | { | |
5998 | int i; | |
5999 | for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) { | |
6000 | bdrv_op_unblock(bs, i, reason); | |
6001 | } | |
6002 | } | |
6003 | ||
6004 | bool bdrv_op_blocker_is_empty(BlockDriverState *bs) | |
6005 | { | |
6006 | int i; | |
6007 | ||
6008 | for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) { | |
6009 | if (!QLIST_EMPTY(&bs->op_blockers[i])) { | |
6010 | return false; | |
6011 | } | |
6012 | } | |
6013 | return true; | |
6014 | } | |
6015 | ||
d92ada22 LC |
6016 | void bdrv_img_create(const char *filename, const char *fmt, |
6017 | const char *base_filename, const char *base_fmt, | |
9217283d FZ |
6018 | char *options, uint64_t img_size, int flags, bool quiet, |
6019 | Error **errp) | |
f88e1a42 | 6020 | { |
83d0521a CL |
6021 | QemuOptsList *create_opts = NULL; |
6022 | QemuOpts *opts = NULL; | |
6023 | const char *backing_fmt, *backing_file; | |
6024 | int64_t size; | |
f88e1a42 | 6025 | BlockDriver *drv, *proto_drv; |
cc84d90f | 6026 | Error *local_err = NULL; |
f88e1a42 JS |
6027 | int ret = 0; |
6028 | ||
6029 | /* Find driver and parse its options */ | |
6030 | drv = bdrv_find_format(fmt); | |
6031 | if (!drv) { | |
71c79813 | 6032 | error_setg(errp, "Unknown file format '%s'", fmt); |
d92ada22 | 6033 | return; |
f88e1a42 JS |
6034 | } |
6035 | ||
b65a5e12 | 6036 | proto_drv = bdrv_find_protocol(filename, true, errp); |
f88e1a42 | 6037 | if (!proto_drv) { |
d92ada22 | 6038 | return; |
f88e1a42 JS |
6039 | } |
6040 | ||
c6149724 HR |
6041 | if (!drv->create_opts) { |
6042 | error_setg(errp, "Format driver '%s' does not support image creation", | |
6043 | drv->format_name); | |
6044 | return; | |
6045 | } | |
6046 | ||
5a5e7f8c ML |
6047 | if (!proto_drv->create_opts) { |
6048 | error_setg(errp, "Protocol driver '%s' does not support image creation", | |
6049 | proto_drv->format_name); | |
6050 | return; | |
6051 | } | |
6052 | ||
f6dc1c31 | 6053 | /* Create parameter list */ |
c282e1fd | 6054 | create_opts = qemu_opts_append(create_opts, drv->create_opts); |
5a5e7f8c | 6055 | create_opts = qemu_opts_append(create_opts, proto_drv->create_opts); |
f88e1a42 | 6056 | |
83d0521a | 6057 | opts = qemu_opts_create(create_opts, NULL, 0, &error_abort); |
f88e1a42 JS |
6058 | |
6059 | /* Parse -o options */ | |
6060 | if (options) { | |
a5f9b9df | 6061 | if (!qemu_opts_do_parse(opts, options, NULL, errp)) { |
f88e1a42 JS |
6062 | goto out; |
6063 | } | |
6064 | } | |
6065 | ||
f6dc1c31 KW |
6066 | if (!qemu_opt_get(opts, BLOCK_OPT_SIZE)) { |
6067 | qemu_opt_set_number(opts, BLOCK_OPT_SIZE, img_size, &error_abort); | |
6068 | } else if (img_size != UINT64_C(-1)) { | |
6069 | error_setg(errp, "The image size must be specified only once"); | |
6070 | goto out; | |
6071 | } | |
6072 | ||
f88e1a42 | 6073 | if (base_filename) { |
235e59cf | 6074 | if (!qemu_opt_set(opts, BLOCK_OPT_BACKING_FILE, base_filename, |
3882578b | 6075 | NULL)) { |
71c79813 LC |
6076 | error_setg(errp, "Backing file not supported for file format '%s'", |
6077 | fmt); | |
f88e1a42 JS |
6078 | goto out; |
6079 | } | |
6080 | } | |
6081 | ||
6082 | if (base_fmt) { | |
3882578b | 6083 | if (!qemu_opt_set(opts, BLOCK_OPT_BACKING_FMT, base_fmt, NULL)) { |
71c79813 LC |
6084 | error_setg(errp, "Backing file format not supported for file " |
6085 | "format '%s'", fmt); | |
f88e1a42 JS |
6086 | goto out; |
6087 | } | |
6088 | } | |
6089 | ||
83d0521a CL |
6090 | backing_file = qemu_opt_get(opts, BLOCK_OPT_BACKING_FILE); |
6091 | if (backing_file) { | |
6092 | if (!strcmp(filename, backing_file)) { | |
71c79813 LC |
6093 | error_setg(errp, "Error: Trying to create an image with the " |
6094 | "same filename as the backing file"); | |
792da93a JS |
6095 | goto out; |
6096 | } | |
6097 | } | |
6098 | ||
83d0521a | 6099 | backing_fmt = qemu_opt_get(opts, BLOCK_OPT_BACKING_FMT); |
f88e1a42 | 6100 | |
6e6e55f5 JS |
6101 | /* The size for the image must always be specified, unless we have a backing |
6102 | * file and we have not been forbidden from opening it. */ | |
a8b42a1c | 6103 | size = qemu_opt_get_size(opts, BLOCK_OPT_SIZE, img_size); |
6e6e55f5 JS |
6104 | if (backing_file && !(flags & BDRV_O_NO_BACKING)) { |
6105 | BlockDriverState *bs; | |
645ae7d8 | 6106 | char *full_backing; |
6e6e55f5 JS |
6107 | int back_flags; |
6108 | QDict *backing_options = NULL; | |
6109 | ||
645ae7d8 HR |
6110 | full_backing = |
6111 | bdrv_get_full_backing_filename_from_filename(filename, backing_file, | |
6112 | &local_err); | |
6e6e55f5 | 6113 | if (local_err) { |
6e6e55f5 JS |
6114 | goto out; |
6115 | } | |
645ae7d8 | 6116 | assert(full_backing); |
29168018 | 6117 | |
6e6e55f5 JS |
6118 | /* backing files always opened read-only */ |
6119 | back_flags = flags; | |
6120 | back_flags &= ~(BDRV_O_RDWR | BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING); | |
f88e1a42 | 6121 | |
cc954f01 | 6122 | backing_options = qdict_new(); |
6e6e55f5 | 6123 | if (backing_fmt) { |
6e6e55f5 JS |
6124 | qdict_put_str(backing_options, "driver", backing_fmt); |
6125 | } | |
cc954f01 | 6126 | qdict_put_bool(backing_options, BDRV_OPT_FORCE_SHARE, true); |
e6641719 | 6127 | |
6e6e55f5 JS |
6128 | bs = bdrv_open(full_backing, NULL, backing_options, back_flags, |
6129 | &local_err); | |
6130 | g_free(full_backing); | |
6131 | if (!bs && size != -1) { | |
6132 | /* Couldn't open BS, but we have a size, so it's nonfatal */ | |
6133 | warn_reportf_err(local_err, | |
6134 | "Could not verify backing image. " | |
6135 | "This may become an error in future versions.\n"); | |
6136 | local_err = NULL; | |
6137 | } else if (!bs) { | |
6138 | /* Couldn't open bs, do not have size */ | |
6139 | error_append_hint(&local_err, | |
6140 | "Could not open backing image to determine size.\n"); | |
6141 | goto out; | |
6142 | } else { | |
6143 | if (size == -1) { | |
6144 | /* Opened BS, have no size */ | |
6145 | size = bdrv_getlength(bs); | |
6146 | if (size < 0) { | |
6147 | error_setg_errno(errp, -size, "Could not get size of '%s'", | |
6148 | backing_file); | |
6149 | bdrv_unref(bs); | |
6150 | goto out; | |
6151 | } | |
6152 | qemu_opt_set_number(opts, BLOCK_OPT_SIZE, size, &error_abort); | |
52bf1e72 | 6153 | } |
66f6b814 | 6154 | bdrv_unref(bs); |
f88e1a42 | 6155 | } |
6e6e55f5 JS |
6156 | } /* (backing_file && !(flags & BDRV_O_NO_BACKING)) */ |
6157 | ||
6158 | if (size == -1) { | |
6159 | error_setg(errp, "Image creation needs a size parameter"); | |
6160 | goto out; | |
f88e1a42 JS |
6161 | } |
6162 | ||
f382d43a | 6163 | if (!quiet) { |
fe646693 | 6164 | printf("Formatting '%s', fmt=%s ", filename, fmt); |
43c5d8f8 | 6165 | qemu_opts_print(opts, " "); |
f382d43a MR |
6166 | puts(""); |
6167 | } | |
83d0521a | 6168 | |
c282e1fd | 6169 | ret = bdrv_create(drv, filename, opts, &local_err); |
83d0521a | 6170 | |
cc84d90f HR |
6171 | if (ret == -EFBIG) { |
6172 | /* This is generally a better message than whatever the driver would | |
6173 | * deliver (especially because of the cluster_size_hint), since that | |
6174 | * is most probably not much different from "image too large". */ | |
6175 | const char *cluster_size_hint = ""; | |
83d0521a | 6176 | if (qemu_opt_get_size(opts, BLOCK_OPT_CLUSTER_SIZE, 0)) { |
cc84d90f | 6177 | cluster_size_hint = " (try using a larger cluster size)"; |
f88e1a42 | 6178 | } |
cc84d90f HR |
6179 | error_setg(errp, "The image size is too large for file format '%s'" |
6180 | "%s", fmt, cluster_size_hint); | |
6181 | error_free(local_err); | |
6182 | local_err = NULL; | |
f88e1a42 JS |
6183 | } |
6184 | ||
6185 | out: | |
83d0521a CL |
6186 | qemu_opts_del(opts); |
6187 | qemu_opts_free(create_opts); | |
621ff94d | 6188 | error_propagate(errp, local_err); |
f88e1a42 | 6189 | } |
85d126f3 SH |
6190 | |
6191 | AioContext *bdrv_get_aio_context(BlockDriverState *bs) | |
6192 | { | |
33f2a757 | 6193 | return bs ? bs->aio_context : qemu_get_aio_context(); |
dcd04228 SH |
6194 | } |
6195 | ||
052a7572 FZ |
6196 | void bdrv_coroutine_enter(BlockDriverState *bs, Coroutine *co) |
6197 | { | |
6198 | aio_co_enter(bdrv_get_aio_context(bs), co); | |
6199 | } | |
6200 | ||
e8a095da SH |
6201 | static void bdrv_do_remove_aio_context_notifier(BdrvAioNotifier *ban) |
6202 | { | |
6203 | QLIST_REMOVE(ban, list); | |
6204 | g_free(ban); | |
6205 | } | |
6206 | ||
a3a683c3 | 6207 | static void bdrv_detach_aio_context(BlockDriverState *bs) |
dcd04228 | 6208 | { |
e8a095da | 6209 | BdrvAioNotifier *baf, *baf_tmp; |
33384421 | 6210 | |
e8a095da SH |
6211 | assert(!bs->walking_aio_notifiers); |
6212 | bs->walking_aio_notifiers = true; | |
6213 | QLIST_FOREACH_SAFE(baf, &bs->aio_notifiers, list, baf_tmp) { | |
6214 | if (baf->deleted) { | |
6215 | bdrv_do_remove_aio_context_notifier(baf); | |
6216 | } else { | |
6217 | baf->detach_aio_context(baf->opaque); | |
6218 | } | |
33384421 | 6219 | } |
e8a095da SH |
6220 | /* Never mind iterating again to check for ->deleted. bdrv_close() will |
6221 | * remove remaining aio notifiers if we aren't called again. | |
6222 | */ | |
6223 | bs->walking_aio_notifiers = false; | |
33384421 | 6224 | |
1bffe1ae | 6225 | if (bs->drv && bs->drv->bdrv_detach_aio_context) { |
dcd04228 SH |
6226 | bs->drv->bdrv_detach_aio_context(bs); |
6227 | } | |
dcd04228 | 6228 | |
e64f25f3 KW |
6229 | if (bs->quiesce_counter) { |
6230 | aio_enable_external(bs->aio_context); | |
6231 | } | |
dcd04228 SH |
6232 | bs->aio_context = NULL; |
6233 | } | |
6234 | ||
a3a683c3 KW |
6235 | static void bdrv_attach_aio_context(BlockDriverState *bs, |
6236 | AioContext *new_context) | |
dcd04228 | 6237 | { |
e8a095da | 6238 | BdrvAioNotifier *ban, *ban_tmp; |
33384421 | 6239 | |
e64f25f3 KW |
6240 | if (bs->quiesce_counter) { |
6241 | aio_disable_external(new_context); | |
6242 | } | |
6243 | ||
dcd04228 SH |
6244 | bs->aio_context = new_context; |
6245 | ||
1bffe1ae | 6246 | if (bs->drv && bs->drv->bdrv_attach_aio_context) { |
dcd04228 SH |
6247 | bs->drv->bdrv_attach_aio_context(bs, new_context); |
6248 | } | |
33384421 | 6249 | |
e8a095da SH |
6250 | assert(!bs->walking_aio_notifiers); |
6251 | bs->walking_aio_notifiers = true; | |
6252 | QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_tmp) { | |
6253 | if (ban->deleted) { | |
6254 | bdrv_do_remove_aio_context_notifier(ban); | |
6255 | } else { | |
6256 | ban->attached_aio_context(new_context, ban->opaque); | |
6257 | } | |
33384421 | 6258 | } |
e8a095da | 6259 | bs->walking_aio_notifiers = false; |
dcd04228 SH |
6260 | } |
6261 | ||
42a65f02 KW |
6262 | /* |
6263 | * Changes the AioContext used for fd handlers, timers, and BHs by this | |
6264 | * BlockDriverState and all its children and parents. | |
6265 | * | |
43eaaaef HR |
6266 | * Must be called from the main AioContext. |
6267 | * | |
42a65f02 KW |
6268 | * The caller must own the AioContext lock for the old AioContext of bs, but it |
6269 | * must not own the AioContext lock for new_context (unless new_context is the | |
6270 | * same as the current context of bs). | |
6271 | * | |
6272 | * @ignore will accumulate all visited BdrvChild object. The caller is | |
6273 | * responsible for freeing the list afterwards. | |
6274 | */ | |
53a7d041 KW |
6275 | void bdrv_set_aio_context_ignore(BlockDriverState *bs, |
6276 | AioContext *new_context, GSList **ignore) | |
dcd04228 | 6277 | { |
e037c09c | 6278 | AioContext *old_context = bdrv_get_aio_context(bs); |
0d83708a KW |
6279 | BdrvChild *child; |
6280 | ||
43eaaaef HR |
6281 | g_assert(qemu_get_current_aio_context() == qemu_get_aio_context()); |
6282 | ||
e037c09c | 6283 | if (old_context == new_context) { |
57830a49 DP |
6284 | return; |
6285 | } | |
6286 | ||
d70d5954 | 6287 | bdrv_drained_begin(bs); |
0d83708a KW |
6288 | |
6289 | QLIST_FOREACH(child, &bs->children, next) { | |
53a7d041 KW |
6290 | if (g_slist_find(*ignore, child)) { |
6291 | continue; | |
6292 | } | |
6293 | *ignore = g_slist_prepend(*ignore, child); | |
6294 | bdrv_set_aio_context_ignore(child->bs, new_context, ignore); | |
6295 | } | |
6296 | QLIST_FOREACH(child, &bs->parents, next_parent) { | |
6297 | if (g_slist_find(*ignore, child)) { | |
6298 | continue; | |
6299 | } | |
bd86fb99 | 6300 | assert(child->klass->set_aio_ctx); |
42a65f02 | 6301 | *ignore = g_slist_prepend(*ignore, child); |
bd86fb99 | 6302 | child->klass->set_aio_ctx(child, new_context, ignore); |
0d83708a KW |
6303 | } |
6304 | ||
dcd04228 SH |
6305 | bdrv_detach_aio_context(bs); |
6306 | ||
e037c09c | 6307 | /* Acquire the new context, if necessary */ |
43eaaaef | 6308 | if (qemu_get_aio_context() != new_context) { |
e037c09c HR |
6309 | aio_context_acquire(new_context); |
6310 | } | |
6311 | ||
dcd04228 | 6312 | bdrv_attach_aio_context(bs, new_context); |
e037c09c HR |
6313 | |
6314 | /* | |
6315 | * If this function was recursively called from | |
6316 | * bdrv_set_aio_context_ignore(), there may be nodes in the | |
6317 | * subtree that have not yet been moved to the new AioContext. | |
6318 | * Release the old one so bdrv_drained_end() can poll them. | |
6319 | */ | |
43eaaaef | 6320 | if (qemu_get_aio_context() != old_context) { |
e037c09c HR |
6321 | aio_context_release(old_context); |
6322 | } | |
6323 | ||
d70d5954 | 6324 | bdrv_drained_end(bs); |
e037c09c | 6325 | |
43eaaaef | 6326 | if (qemu_get_aio_context() != old_context) { |
e037c09c HR |
6327 | aio_context_acquire(old_context); |
6328 | } | |
43eaaaef | 6329 | if (qemu_get_aio_context() != new_context) { |
e037c09c HR |
6330 | aio_context_release(new_context); |
6331 | } | |
85d126f3 | 6332 | } |
d616b224 | 6333 | |
5d231849 KW |
6334 | static bool bdrv_parent_can_set_aio_context(BdrvChild *c, AioContext *ctx, |
6335 | GSList **ignore, Error **errp) | |
6336 | { | |
6337 | if (g_slist_find(*ignore, c)) { | |
6338 | return true; | |
6339 | } | |
6340 | *ignore = g_slist_prepend(*ignore, c); | |
6341 | ||
bd86fb99 HR |
6342 | /* |
6343 | * A BdrvChildClass that doesn't handle AioContext changes cannot | |
6344 | * tolerate any AioContext changes | |
6345 | */ | |
6346 | if (!c->klass->can_set_aio_ctx) { | |
5d231849 KW |
6347 | char *user = bdrv_child_user_desc(c); |
6348 | error_setg(errp, "Changing iothreads is not supported by %s", user); | |
6349 | g_free(user); | |
6350 | return false; | |
6351 | } | |
bd86fb99 | 6352 | if (!c->klass->can_set_aio_ctx(c, ctx, ignore, errp)) { |
5d231849 KW |
6353 | assert(!errp || *errp); |
6354 | return false; | |
6355 | } | |
6356 | return true; | |
6357 | } | |
6358 | ||
6359 | bool bdrv_child_can_set_aio_context(BdrvChild *c, AioContext *ctx, | |
6360 | GSList **ignore, Error **errp) | |
6361 | { | |
6362 | if (g_slist_find(*ignore, c)) { | |
6363 | return true; | |
6364 | } | |
6365 | *ignore = g_slist_prepend(*ignore, c); | |
6366 | return bdrv_can_set_aio_context(c->bs, ctx, ignore, errp); | |
6367 | } | |
6368 | ||
6369 | /* @ignore will accumulate all visited BdrvChild object. The caller is | |
6370 | * responsible for freeing the list afterwards. */ | |
6371 | bool bdrv_can_set_aio_context(BlockDriverState *bs, AioContext *ctx, | |
6372 | GSList **ignore, Error **errp) | |
6373 | { | |
6374 | BdrvChild *c; | |
6375 | ||
6376 | if (bdrv_get_aio_context(bs) == ctx) { | |
6377 | return true; | |
6378 | } | |
6379 | ||
6380 | QLIST_FOREACH(c, &bs->parents, next_parent) { | |
6381 | if (!bdrv_parent_can_set_aio_context(c, ctx, ignore, errp)) { | |
6382 | return false; | |
6383 | } | |
6384 | } | |
6385 | QLIST_FOREACH(c, &bs->children, next) { | |
6386 | if (!bdrv_child_can_set_aio_context(c, ctx, ignore, errp)) { | |
6387 | return false; | |
6388 | } | |
6389 | } | |
6390 | ||
6391 | return true; | |
6392 | } | |
6393 | ||
6394 | int bdrv_child_try_set_aio_context(BlockDriverState *bs, AioContext *ctx, | |
6395 | BdrvChild *ignore_child, Error **errp) | |
6396 | { | |
6397 | GSList *ignore; | |
6398 | bool ret; | |
6399 | ||
6400 | ignore = ignore_child ? g_slist_prepend(NULL, ignore_child) : NULL; | |
6401 | ret = bdrv_can_set_aio_context(bs, ctx, &ignore, errp); | |
6402 | g_slist_free(ignore); | |
6403 | ||
6404 | if (!ret) { | |
6405 | return -EPERM; | |
6406 | } | |
6407 | ||
53a7d041 KW |
6408 | ignore = ignore_child ? g_slist_prepend(NULL, ignore_child) : NULL; |
6409 | bdrv_set_aio_context_ignore(bs, ctx, &ignore); | |
6410 | g_slist_free(ignore); | |
6411 | ||
5d231849 KW |
6412 | return 0; |
6413 | } | |
6414 | ||
6415 | int bdrv_try_set_aio_context(BlockDriverState *bs, AioContext *ctx, | |
6416 | Error **errp) | |
6417 | { | |
6418 | return bdrv_child_try_set_aio_context(bs, ctx, NULL, errp); | |
6419 | } | |
6420 | ||
33384421 HR |
6421 | void bdrv_add_aio_context_notifier(BlockDriverState *bs, |
6422 | void (*attached_aio_context)(AioContext *new_context, void *opaque), | |
6423 | void (*detach_aio_context)(void *opaque), void *opaque) | |
6424 | { | |
6425 | BdrvAioNotifier *ban = g_new(BdrvAioNotifier, 1); | |
6426 | *ban = (BdrvAioNotifier){ | |
6427 | .attached_aio_context = attached_aio_context, | |
6428 | .detach_aio_context = detach_aio_context, | |
6429 | .opaque = opaque | |
6430 | }; | |
6431 | ||
6432 | QLIST_INSERT_HEAD(&bs->aio_notifiers, ban, list); | |
6433 | } | |
6434 | ||
6435 | void bdrv_remove_aio_context_notifier(BlockDriverState *bs, | |
6436 | void (*attached_aio_context)(AioContext *, | |
6437 | void *), | |
6438 | void (*detach_aio_context)(void *), | |
6439 | void *opaque) | |
6440 | { | |
6441 | BdrvAioNotifier *ban, *ban_next; | |
6442 | ||
6443 | QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_next) { | |
6444 | if (ban->attached_aio_context == attached_aio_context && | |
6445 | ban->detach_aio_context == detach_aio_context && | |
e8a095da SH |
6446 | ban->opaque == opaque && |
6447 | ban->deleted == false) | |
33384421 | 6448 | { |
e8a095da SH |
6449 | if (bs->walking_aio_notifiers) { |
6450 | ban->deleted = true; | |
6451 | } else { | |
6452 | bdrv_do_remove_aio_context_notifier(ban); | |
6453 | } | |
33384421 HR |
6454 | return; |
6455 | } | |
6456 | } | |
6457 | ||
6458 | abort(); | |
6459 | } | |
6460 | ||
77485434 | 6461 | int bdrv_amend_options(BlockDriverState *bs, QemuOpts *opts, |
d1402b50 | 6462 | BlockDriverAmendStatusCB *status_cb, void *cb_opaque, |
a3579bfa | 6463 | bool force, |
d1402b50 | 6464 | Error **errp) |
6f176b48 | 6465 | { |
d470ad42 | 6466 | if (!bs->drv) { |
d1402b50 | 6467 | error_setg(errp, "Node is ejected"); |
d470ad42 HR |
6468 | return -ENOMEDIUM; |
6469 | } | |
c282e1fd | 6470 | if (!bs->drv->bdrv_amend_options) { |
d1402b50 HR |
6471 | error_setg(errp, "Block driver '%s' does not support option amendment", |
6472 | bs->drv->format_name); | |
6f176b48 HR |
6473 | return -ENOTSUP; |
6474 | } | |
a3579bfa ML |
6475 | return bs->drv->bdrv_amend_options(bs, opts, status_cb, |
6476 | cb_opaque, force, errp); | |
6f176b48 | 6477 | } |
f6186f49 | 6478 | |
5d69b5ab HR |
6479 | /* |
6480 | * This function checks whether the given @to_replace is allowed to be | |
6481 | * replaced by a node that always shows the same data as @bs. This is | |
6482 | * used for example to verify whether the mirror job can replace | |
6483 | * @to_replace by the target mirrored from @bs. | |
6484 | * To be replaceable, @bs and @to_replace may either be guaranteed to | |
6485 | * always show the same data (because they are only connected through | |
6486 | * filters), or some driver may allow replacing one of its children | |
6487 | * because it can guarantee that this child's data is not visible at | |
6488 | * all (for example, for dissenting quorum children that have no other | |
6489 | * parents). | |
6490 | */ | |
6491 | bool bdrv_recurse_can_replace(BlockDriverState *bs, | |
6492 | BlockDriverState *to_replace) | |
6493 | { | |
6494 | if (!bs || !bs->drv) { | |
6495 | return false; | |
6496 | } | |
6497 | ||
6498 | if (bs == to_replace) { | |
6499 | return true; | |
6500 | } | |
6501 | ||
6502 | /* See what the driver can do */ | |
6503 | if (bs->drv->bdrv_recurse_can_replace) { | |
6504 | return bs->drv->bdrv_recurse_can_replace(bs, to_replace); | |
6505 | } | |
6506 | ||
6507 | /* For filters without an own implementation, we can recurse on our own */ | |
6508 | if (bs->drv->is_filter) { | |
6509 | BdrvChild *child = bs->file ?: bs->backing; | |
6510 | return bdrv_recurse_can_replace(child->bs, to_replace); | |
6511 | } | |
6512 | ||
6513 | /* Safe default */ | |
6514 | return false; | |
6515 | } | |
6516 | ||
810803a8 HR |
6517 | /* |
6518 | * Check whether the given @node_name can be replaced by a node that | |
6519 | * has the same data as @parent_bs. If so, return @node_name's BDS; | |
6520 | * NULL otherwise. | |
6521 | * | |
6522 | * @node_name must be a (recursive) *child of @parent_bs (or this | |
6523 | * function will return NULL). | |
6524 | * | |
6525 | * The result (whether the node can be replaced or not) is only valid | |
6526 | * for as long as no graph or permission changes occur. | |
6527 | */ | |
e12f3784 WC |
6528 | BlockDriverState *check_to_replace_node(BlockDriverState *parent_bs, |
6529 | const char *node_name, Error **errp) | |
09158f00 BC |
6530 | { |
6531 | BlockDriverState *to_replace_bs = bdrv_find_node(node_name); | |
5a7e7a0b SH |
6532 | AioContext *aio_context; |
6533 | ||
09158f00 BC |
6534 | if (!to_replace_bs) { |
6535 | error_setg(errp, "Node name '%s' not found", node_name); | |
6536 | return NULL; | |
6537 | } | |
6538 | ||
5a7e7a0b SH |
6539 | aio_context = bdrv_get_aio_context(to_replace_bs); |
6540 | aio_context_acquire(aio_context); | |
6541 | ||
09158f00 | 6542 | if (bdrv_op_is_blocked(to_replace_bs, BLOCK_OP_TYPE_REPLACE, errp)) { |
5a7e7a0b SH |
6543 | to_replace_bs = NULL; |
6544 | goto out; | |
09158f00 BC |
6545 | } |
6546 | ||
6547 | /* We don't want arbitrary node of the BDS chain to be replaced only the top | |
6548 | * most non filter in order to prevent data corruption. | |
6549 | * Another benefit is that this tests exclude backing files which are | |
6550 | * blocked by the backing blockers. | |
6551 | */ | |
810803a8 HR |
6552 | if (!bdrv_recurse_can_replace(parent_bs, to_replace_bs)) { |
6553 | error_setg(errp, "Cannot replace '%s' by a node mirrored from '%s', " | |
6554 | "because it cannot be guaranteed that doing so would not " | |
6555 | "lead to an abrupt change of visible data", | |
6556 | node_name, parent_bs->node_name); | |
5a7e7a0b SH |
6557 | to_replace_bs = NULL; |
6558 | goto out; | |
09158f00 BC |
6559 | } |
6560 | ||
5a7e7a0b SH |
6561 | out: |
6562 | aio_context_release(aio_context); | |
09158f00 BC |
6563 | return to_replace_bs; |
6564 | } | |
448ad91d | 6565 | |
97e2f021 HR |
6566 | /** |
6567 | * Iterates through the list of runtime option keys that are said to | |
6568 | * be "strong" for a BDS. An option is called "strong" if it changes | |
6569 | * a BDS's data. For example, the null block driver's "size" and | |
6570 | * "read-zeroes" options are strong, but its "latency-ns" option is | |
6571 | * not. | |
6572 | * | |
6573 | * If a key returned by this function ends with a dot, all options | |
6574 | * starting with that prefix are strong. | |
6575 | */ | |
6576 | static const char *const *strong_options(BlockDriverState *bs, | |
6577 | const char *const *curopt) | |
6578 | { | |
6579 | static const char *const global_options[] = { | |
6580 | "driver", "filename", NULL | |
6581 | }; | |
6582 | ||
6583 | if (!curopt) { | |
6584 | return &global_options[0]; | |
6585 | } | |
6586 | ||
6587 | curopt++; | |
6588 | if (curopt == &global_options[ARRAY_SIZE(global_options) - 1] && bs->drv) { | |
6589 | curopt = bs->drv->strong_runtime_opts; | |
6590 | } | |
6591 | ||
6592 | return (curopt && *curopt) ? curopt : NULL; | |
6593 | } | |
6594 | ||
6595 | /** | |
6596 | * Copies all strong runtime options from bs->options to the given | |
6597 | * QDict. The set of strong option keys is determined by invoking | |
6598 | * strong_options(). | |
6599 | * | |
6600 | * Returns true iff any strong option was present in bs->options (and | |
6601 | * thus copied to the target QDict) with the exception of "filename" | |
6602 | * and "driver". The caller is expected to use this value to decide | |
6603 | * whether the existence of strong options prevents the generation of | |
6604 | * a plain filename. | |
6605 | */ | |
6606 | static bool append_strong_runtime_options(QDict *d, BlockDriverState *bs) | |
6607 | { | |
6608 | bool found_any = false; | |
6609 | const char *const *option_name = NULL; | |
6610 | ||
6611 | if (!bs->drv) { | |
6612 | return false; | |
6613 | } | |
6614 | ||
6615 | while ((option_name = strong_options(bs, option_name))) { | |
6616 | bool option_given = false; | |
6617 | ||
6618 | assert(strlen(*option_name) > 0); | |
6619 | if ((*option_name)[strlen(*option_name) - 1] != '.') { | |
6620 | QObject *entry = qdict_get(bs->options, *option_name); | |
6621 | if (!entry) { | |
6622 | continue; | |
6623 | } | |
6624 | ||
6625 | qdict_put_obj(d, *option_name, qobject_ref(entry)); | |
6626 | option_given = true; | |
6627 | } else { | |
6628 | const QDictEntry *entry; | |
6629 | for (entry = qdict_first(bs->options); entry; | |
6630 | entry = qdict_next(bs->options, entry)) | |
6631 | { | |
6632 | if (strstart(qdict_entry_key(entry), *option_name, NULL)) { | |
6633 | qdict_put_obj(d, qdict_entry_key(entry), | |
6634 | qobject_ref(qdict_entry_value(entry))); | |
6635 | option_given = true; | |
6636 | } | |
6637 | } | |
6638 | } | |
6639 | ||
6640 | /* While "driver" and "filename" need to be included in a JSON filename, | |
6641 | * their existence does not prohibit generation of a plain filename. */ | |
6642 | if (!found_any && option_given && | |
6643 | strcmp(*option_name, "driver") && strcmp(*option_name, "filename")) | |
6644 | { | |
6645 | found_any = true; | |
6646 | } | |
6647 | } | |
6648 | ||
62a01a27 HR |
6649 | if (!qdict_haskey(d, "driver")) { |
6650 | /* Drivers created with bdrv_new_open_driver() may not have a | |
6651 | * @driver option. Add it here. */ | |
6652 | qdict_put_str(d, "driver", bs->drv->format_name); | |
6653 | } | |
6654 | ||
97e2f021 HR |
6655 | return found_any; |
6656 | } | |
6657 | ||
90993623 HR |
6658 | /* Note: This function may return false positives; it may return true |
6659 | * even if opening the backing file specified by bs's image header | |
6660 | * would result in exactly bs->backing. */ | |
6661 | static bool bdrv_backing_overridden(BlockDriverState *bs) | |
6662 | { | |
6663 | if (bs->backing) { | |
6664 | return strcmp(bs->auto_backing_file, | |
6665 | bs->backing->bs->filename); | |
6666 | } else { | |
6667 | /* No backing BDS, so if the image header reports any backing | |
6668 | * file, it must have been suppressed */ | |
6669 | return bs->auto_backing_file[0] != '\0'; | |
6670 | } | |
6671 | } | |
6672 | ||
91af7014 HR |
6673 | /* Updates the following BDS fields: |
6674 | * - exact_filename: A filename which may be used for opening a block device | |
6675 | * which (mostly) equals the given BDS (even without any | |
6676 | * other options; so reading and writing must return the same | |
6677 | * results, but caching etc. may be different) | |
6678 | * - full_open_options: Options which, when given when opening a block device | |
6679 | * (without a filename), result in a BDS (mostly) | |
6680 | * equalling the given one | |
6681 | * - filename: If exact_filename is set, it is copied here. Otherwise, | |
6682 | * full_open_options is converted to a JSON object, prefixed with | |
6683 | * "json:" (for use through the JSON pseudo protocol) and put here. | |
6684 | */ | |
6685 | void bdrv_refresh_filename(BlockDriverState *bs) | |
6686 | { | |
6687 | BlockDriver *drv = bs->drv; | |
e24518e3 | 6688 | BdrvChild *child; |
91af7014 | 6689 | QDict *opts; |
90993623 | 6690 | bool backing_overridden; |
998b3a1e HR |
6691 | bool generate_json_filename; /* Whether our default implementation should |
6692 | fill exact_filename (false) or not (true) */ | |
91af7014 HR |
6693 | |
6694 | if (!drv) { | |
6695 | return; | |
6696 | } | |
6697 | ||
e24518e3 HR |
6698 | /* This BDS's file name may depend on any of its children's file names, so |
6699 | * refresh those first */ | |
6700 | QLIST_FOREACH(child, &bs->children, next) { | |
6701 | bdrv_refresh_filename(child->bs); | |
91af7014 HR |
6702 | } |
6703 | ||
bb808d5f HR |
6704 | if (bs->implicit) { |
6705 | /* For implicit nodes, just copy everything from the single child */ | |
6706 | child = QLIST_FIRST(&bs->children); | |
6707 | assert(QLIST_NEXT(child, next) == NULL); | |
6708 | ||
6709 | pstrcpy(bs->exact_filename, sizeof(bs->exact_filename), | |
6710 | child->bs->exact_filename); | |
6711 | pstrcpy(bs->filename, sizeof(bs->filename), child->bs->filename); | |
6712 | ||
cb895614 | 6713 | qobject_unref(bs->full_open_options); |
bb808d5f HR |
6714 | bs->full_open_options = qobject_ref(child->bs->full_open_options); |
6715 | ||
6716 | return; | |
6717 | } | |
6718 | ||
90993623 HR |
6719 | backing_overridden = bdrv_backing_overridden(bs); |
6720 | ||
6721 | if (bs->open_flags & BDRV_O_NO_IO) { | |
6722 | /* Without I/O, the backing file does not change anything. | |
6723 | * Therefore, in such a case (primarily qemu-img), we can | |
6724 | * pretend the backing file has not been overridden even if | |
6725 | * it technically has been. */ | |
6726 | backing_overridden = false; | |
6727 | } | |
6728 | ||
97e2f021 HR |
6729 | /* Gather the options QDict */ |
6730 | opts = qdict_new(); | |
998b3a1e HR |
6731 | generate_json_filename = append_strong_runtime_options(opts, bs); |
6732 | generate_json_filename |= backing_overridden; | |
97e2f021 HR |
6733 | |
6734 | if (drv->bdrv_gather_child_options) { | |
6735 | /* Some block drivers may not want to present all of their children's | |
6736 | * options, or name them differently from BdrvChild.name */ | |
6737 | drv->bdrv_gather_child_options(bs, opts, backing_overridden); | |
6738 | } else { | |
6739 | QLIST_FOREACH(child, &bs->children, next) { | |
25191e5f | 6740 | if (child == bs->backing && !backing_overridden) { |
97e2f021 HR |
6741 | /* We can skip the backing BDS if it has not been overridden */ |
6742 | continue; | |
6743 | } | |
6744 | ||
6745 | qdict_put(opts, child->name, | |
6746 | qobject_ref(child->bs->full_open_options)); | |
6747 | } | |
6748 | ||
6749 | if (backing_overridden && !bs->backing) { | |
6750 | /* Force no backing file */ | |
6751 | qdict_put_null(opts, "backing"); | |
6752 | } | |
6753 | } | |
6754 | ||
6755 | qobject_unref(bs->full_open_options); | |
6756 | bs->full_open_options = opts; | |
6757 | ||
998b3a1e HR |
6758 | if (drv->bdrv_refresh_filename) { |
6759 | /* Obsolete information is of no use here, so drop the old file name | |
6760 | * information before refreshing it */ | |
6761 | bs->exact_filename[0] = '\0'; | |
6762 | ||
6763 | drv->bdrv_refresh_filename(bs); | |
6764 | } else if (bs->file) { | |
6765 | /* Try to reconstruct valid information from the underlying file */ | |
6766 | ||
6767 | bs->exact_filename[0] = '\0'; | |
6768 | ||
fb695c74 HR |
6769 | /* |
6770 | * We can use the underlying file's filename if: | |
6771 | * - it has a filename, | |
6772 | * - the file is a protocol BDS, and | |
6773 | * - opening that file (as this BDS's format) will automatically create | |
6774 | * the BDS tree we have right now, that is: | |
6775 | * - the user did not significantly change this BDS's behavior with | |
6776 | * some explicit (strong) options | |
6777 | * - no non-file child of this BDS has been overridden by the user | |
6778 | * Both of these conditions are represented by generate_json_filename. | |
6779 | */ | |
6780 | if (bs->file->bs->exact_filename[0] && | |
6781 | bs->file->bs->drv->bdrv_file_open && | |
6782 | !generate_json_filename) | |
6783 | { | |
998b3a1e HR |
6784 | strcpy(bs->exact_filename, bs->file->bs->exact_filename); |
6785 | } | |
6786 | } | |
6787 | ||
91af7014 HR |
6788 | if (bs->exact_filename[0]) { |
6789 | pstrcpy(bs->filename, sizeof(bs->filename), bs->exact_filename); | |
97e2f021 | 6790 | } else { |
91af7014 | 6791 | QString *json = qobject_to_json(QOBJECT(bs->full_open_options)); |
5c86bdf1 EB |
6792 | if (snprintf(bs->filename, sizeof(bs->filename), "json:%s", |
6793 | qstring_get_str(json)) >= sizeof(bs->filename)) { | |
6794 | /* Give user a hint if we truncated things. */ | |
6795 | strcpy(bs->filename + sizeof(bs->filename) - 4, "..."); | |
6796 | } | |
cb3e7f08 | 6797 | qobject_unref(json); |
91af7014 HR |
6798 | } |
6799 | } | |
e06018ad | 6800 | |
1e89d0f9 HR |
6801 | char *bdrv_dirname(BlockDriverState *bs, Error **errp) |
6802 | { | |
6803 | BlockDriver *drv = bs->drv; | |
6804 | ||
6805 | if (!drv) { | |
6806 | error_setg(errp, "Node '%s' is ejected", bs->node_name); | |
6807 | return NULL; | |
6808 | } | |
6809 | ||
6810 | if (drv->bdrv_dirname) { | |
6811 | return drv->bdrv_dirname(bs, errp); | |
6812 | } | |
6813 | ||
6814 | if (bs->file) { | |
6815 | return bdrv_dirname(bs->file->bs, errp); | |
6816 | } | |
6817 | ||
6818 | bdrv_refresh_filename(bs); | |
6819 | if (bs->exact_filename[0] != '\0') { | |
6820 | return path_combine(bs->exact_filename, ""); | |
6821 | } | |
6822 | ||
6823 | error_setg(errp, "Cannot generate a base directory for %s nodes", | |
6824 | drv->format_name); | |
6825 | return NULL; | |
6826 | } | |
6827 | ||
e06018ad WC |
6828 | /* |
6829 | * Hot add/remove a BDS's child. So the user can take a child offline when | |
6830 | * it is broken and take a new child online | |
6831 | */ | |
6832 | void bdrv_add_child(BlockDriverState *parent_bs, BlockDriverState *child_bs, | |
6833 | Error **errp) | |
6834 | { | |
6835 | ||
6836 | if (!parent_bs->drv || !parent_bs->drv->bdrv_add_child) { | |
6837 | error_setg(errp, "The node %s does not support adding a child", | |
6838 | bdrv_get_device_or_node_name(parent_bs)); | |
6839 | return; | |
6840 | } | |
6841 | ||
6842 | if (!QLIST_EMPTY(&child_bs->parents)) { | |
6843 | error_setg(errp, "The node %s already has a parent", | |
6844 | child_bs->node_name); | |
6845 | return; | |
6846 | } | |
6847 | ||
6848 | parent_bs->drv->bdrv_add_child(parent_bs, child_bs, errp); | |
6849 | } | |
6850 | ||
6851 | void bdrv_del_child(BlockDriverState *parent_bs, BdrvChild *child, Error **errp) | |
6852 | { | |
6853 | BdrvChild *tmp; | |
6854 | ||
6855 | if (!parent_bs->drv || !parent_bs->drv->bdrv_del_child) { | |
6856 | error_setg(errp, "The node %s does not support removing a child", | |
6857 | bdrv_get_device_or_node_name(parent_bs)); | |
6858 | return; | |
6859 | } | |
6860 | ||
6861 | QLIST_FOREACH(tmp, &parent_bs->children, next) { | |
6862 | if (tmp == child) { | |
6863 | break; | |
6864 | } | |
6865 | } | |
6866 | ||
6867 | if (!tmp) { | |
6868 | error_setg(errp, "The node %s does not have a child named %s", | |
6869 | bdrv_get_device_or_node_name(parent_bs), | |
6870 | bdrv_get_device_or_node_name(child->bs)); | |
6871 | return; | |
6872 | } | |
6873 | ||
6874 | parent_bs->drv->bdrv_del_child(parent_bs, child, errp); | |
6875 | } | |
6f7a3b53 HR |
6876 | |
6877 | int bdrv_make_empty(BdrvChild *c, Error **errp) | |
6878 | { | |
6879 | BlockDriver *drv = c->bs->drv; | |
6880 | int ret; | |
6881 | ||
6882 | assert(c->perm & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)); | |
6883 | ||
6884 | if (!drv->bdrv_make_empty) { | |
6885 | error_setg(errp, "%s does not support emptying nodes", | |
6886 | drv->format_name); | |
6887 | return -ENOTSUP; | |
6888 | } | |
6889 | ||
6890 | ret = drv->bdrv_make_empty(c->bs); | |
6891 | if (ret < 0) { | |
6892 | error_setg_errno(errp, -ret, "Failed to empty %s", | |
6893 | c->bs->filename); | |
6894 | return ret; | |
6895 | } | |
6896 | ||
6897 | return 0; | |
6898 | } |