]>
Commit | Line | Data |
---|---|---|
fc01f7e7 FB |
1 | /* |
2 | * QEMU System Emulator block driver | |
5fafdf24 | 3 | * |
fc01f7e7 | 4 | * Copyright (c) 2003 Fabrice Bellard |
c20555e1 | 5 | * Copyright (c) 2020 Virtuozzo International GmbH. |
5fafdf24 | 6 | * |
fc01f7e7 FB |
7 | * Permission is hereby granted, free of charge, to any person obtaining a copy |
8 | * of this software and associated documentation files (the "Software"), to deal | |
9 | * in the Software without restriction, including without limitation the rights | |
10 | * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell | |
11 | * copies of the Software, and to permit persons to whom the Software is | |
12 | * furnished to do so, subject to the following conditions: | |
13 | * | |
14 | * The above copyright notice and this permission notice shall be included in | |
15 | * all copies or substantial portions of the Software. | |
16 | * | |
17 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR | |
18 | * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, | |
19 | * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL | |
20 | * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER | |
21 | * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, | |
22 | * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN | |
23 | * THE SOFTWARE. | |
24 | */ | |
e688df6b | 25 | |
d38ea87a | 26 | #include "qemu/osdep.h" |
0ab8ed18 | 27 | #include "block/trace.h" |
737e150e PB |
28 | #include "block/block_int.h" |
29 | #include "block/blockjob.h" | |
0c9b70d5 | 30 | #include "block/fuse.h" |
cd7fca95 | 31 | #include "block/nbd.h" |
609f45ea | 32 | #include "block/qdict.h" |
d49b6836 | 33 | #include "qemu/error-report.h" |
5e5733e5 | 34 | #include "block/module_block.h" |
db725815 | 35 | #include "qemu/main-loop.h" |
1de7afc9 | 36 | #include "qemu/module.h" |
e688df6b | 37 | #include "qapi/error.h" |
452fcdbc | 38 | #include "qapi/qmp/qdict.h" |
7b1b5d19 | 39 | #include "qapi/qmp/qjson.h" |
e59a0cf1 | 40 | #include "qapi/qmp/qnull.h" |
fc81fa1e | 41 | #include "qapi/qmp/qstring.h" |
e1d74bc6 KW |
42 | #include "qapi/qobject-output-visitor.h" |
43 | #include "qapi/qapi-visit-block-core.h" | |
bfb197e0 | 44 | #include "sysemu/block-backend.h" |
1de7afc9 | 45 | #include "qemu/notify.h" |
922a01a0 | 46 | #include "qemu/option.h" |
10817bf0 | 47 | #include "qemu/coroutine.h" |
c13163fb | 48 | #include "block/qapi.h" |
1de7afc9 | 49 | #include "qemu/timer.h" |
f348b6d1 VB |
50 | #include "qemu/cutils.h" |
51 | #include "qemu/id.h" | |
21c2283e | 52 | #include "block/coroutines.h" |
fc01f7e7 | 53 | |
71e72a19 | 54 | #ifdef CONFIG_BSD |
7674e7bf | 55 | #include <sys/ioctl.h> |
72cf2d4f | 56 | #include <sys/queue.h> |
feccdcee | 57 | #if defined(HAVE_SYS_DISK_H) |
7674e7bf FB |
58 | #include <sys/disk.h> |
59 | #endif | |
c5e97233 | 60 | #endif |
7674e7bf | 61 | |
49dc768d AL |
62 | #ifdef _WIN32 |
63 | #include <windows.h> | |
64 | #endif | |
65 | ||
1c9805a3 SH |
66 | #define NOT_DONE 0x7fffffff /* used while emulated sync operation in progress */ |
67 | ||
dc364f4c BC |
68 | static QTAILQ_HEAD(, BlockDriverState) graph_bdrv_states = |
69 | QTAILQ_HEAD_INITIALIZER(graph_bdrv_states); | |
70 | ||
2c1d04e0 HR |
71 | static QTAILQ_HEAD(, BlockDriverState) all_bdrv_states = |
72 | QTAILQ_HEAD_INITIALIZER(all_bdrv_states); | |
73 | ||
8a22f02a SH |
74 | static QLIST_HEAD(, BlockDriver) bdrv_drivers = |
75 | QLIST_HEAD_INITIALIZER(bdrv_drivers); | |
ea2384d3 | 76 | |
5b363937 HR |
77 | static BlockDriverState *bdrv_open_inherit(const char *filename, |
78 | const char *reference, | |
79 | QDict *options, int flags, | |
80 | BlockDriverState *parent, | |
bd86fb99 | 81 | const BdrvChildClass *child_class, |
272c02ea | 82 | BdrvChildRole child_role, |
5b363937 | 83 | Error **errp); |
f3930ed0 | 84 | |
0978623e VSO |
85 | static void bdrv_replace_child_noperm(BdrvChild *child, |
86 | BlockDriverState *new_bs); | |
e9238278 VSO |
87 | static void bdrv_remove_file_or_backing_child(BlockDriverState *bs, |
88 | BdrvChild *child, | |
89 | Transaction *tran); | |
160333e1 VSO |
90 | static void bdrv_remove_filter_or_cow_child(BlockDriverState *bs, |
91 | Transaction *tran); | |
0978623e | 92 | |
72373e40 VSO |
93 | static int bdrv_reopen_prepare(BDRVReopenState *reopen_state, |
94 | BlockReopenQueue *queue, | |
ecd30d2d | 95 | Transaction *change_child_tran, Error **errp); |
53e96d1e VSO |
96 | static void bdrv_reopen_commit(BDRVReopenState *reopen_state); |
97 | static void bdrv_reopen_abort(BDRVReopenState *reopen_state); | |
98 | ||
eb852011 MA |
99 | /* If non-zero, use only whitelisted block drivers */ |
100 | static int use_bdrv_whitelist; | |
101 | ||
9e0b22f4 SH |
102 | #ifdef _WIN32 |
103 | static int is_windows_drive_prefix(const char *filename) | |
104 | { | |
105 | return (((filename[0] >= 'a' && filename[0] <= 'z') || | |
106 | (filename[0] >= 'A' && filename[0] <= 'Z')) && | |
107 | filename[1] == ':'); | |
108 | } | |
109 | ||
110 | int is_windows_drive(const char *filename) | |
111 | { | |
112 | if (is_windows_drive_prefix(filename) && | |
113 | filename[2] == '\0') | |
114 | return 1; | |
115 | if (strstart(filename, "\\\\.\\", NULL) || | |
116 | strstart(filename, "//./", NULL)) | |
117 | return 1; | |
118 | return 0; | |
119 | } | |
120 | #endif | |
121 | ||
339064d5 KW |
122 | size_t bdrv_opt_mem_align(BlockDriverState *bs) |
123 | { | |
124 | if (!bs || !bs->drv) { | |
459b4e66 | 125 | /* page size or 4k (hdd sector size) should be on the safe side */ |
038adc2f | 126 | return MAX(4096, qemu_real_host_page_size); |
339064d5 KW |
127 | } |
128 | ||
129 | return bs->bl.opt_mem_alignment; | |
130 | } | |
131 | ||
4196d2f0 DL |
132 | size_t bdrv_min_mem_align(BlockDriverState *bs) |
133 | { | |
134 | if (!bs || !bs->drv) { | |
459b4e66 | 135 | /* page size or 4k (hdd sector size) should be on the safe side */ |
038adc2f | 136 | return MAX(4096, qemu_real_host_page_size); |
4196d2f0 DL |
137 | } |
138 | ||
139 | return bs->bl.min_mem_alignment; | |
140 | } | |
141 | ||
9e0b22f4 | 142 | /* check if the path starts with "<protocol>:" */ |
5c98415b | 143 | int path_has_protocol(const char *path) |
9e0b22f4 | 144 | { |
947995c0 PB |
145 | const char *p; |
146 | ||
9e0b22f4 SH |
147 | #ifdef _WIN32 |
148 | if (is_windows_drive(path) || | |
149 | is_windows_drive_prefix(path)) { | |
150 | return 0; | |
151 | } | |
947995c0 PB |
152 | p = path + strcspn(path, ":/\\"); |
153 | #else | |
154 | p = path + strcspn(path, ":/"); | |
9e0b22f4 SH |
155 | #endif |
156 | ||
947995c0 | 157 | return *p == ':'; |
9e0b22f4 SH |
158 | } |
159 | ||
83f64091 | 160 | int path_is_absolute(const char *path) |
3b0d4f61 | 161 | { |
21664424 FB |
162 | #ifdef _WIN32 |
163 | /* specific case for names like: "\\.\d:" */ | |
f53f4da9 | 164 | if (is_windows_drive(path) || is_windows_drive_prefix(path)) { |
21664424 | 165 | return 1; |
f53f4da9 PB |
166 | } |
167 | return (*path == '/' || *path == '\\'); | |
3b9f94e1 | 168 | #else |
f53f4da9 | 169 | return (*path == '/'); |
3b9f94e1 | 170 | #endif |
3b0d4f61 FB |
171 | } |
172 | ||
009b03aa | 173 | /* if filename is absolute, just return its duplicate. Otherwise, build a |
83f64091 FB |
174 | path to it by considering it is relative to base_path. URL are |
175 | supported. */ | |
009b03aa | 176 | char *path_combine(const char *base_path, const char *filename) |
3b0d4f61 | 177 | { |
009b03aa | 178 | const char *protocol_stripped = NULL; |
83f64091 | 179 | const char *p, *p1; |
009b03aa | 180 | char *result; |
83f64091 FB |
181 | int len; |
182 | ||
83f64091 | 183 | if (path_is_absolute(filename)) { |
009b03aa HR |
184 | return g_strdup(filename); |
185 | } | |
0d54a6fe | 186 | |
009b03aa HR |
187 | if (path_has_protocol(base_path)) { |
188 | protocol_stripped = strchr(base_path, ':'); | |
189 | if (protocol_stripped) { | |
190 | protocol_stripped++; | |
0d54a6fe | 191 | } |
009b03aa HR |
192 | } |
193 | p = protocol_stripped ?: base_path; | |
0d54a6fe | 194 | |
009b03aa | 195 | p1 = strrchr(base_path, '/'); |
3b9f94e1 | 196 | #ifdef _WIN32 |
009b03aa HR |
197 | { |
198 | const char *p2; | |
199 | p2 = strrchr(base_path, '\\'); | |
200 | if (!p1 || p2 > p1) { | |
201 | p1 = p2; | |
3b9f94e1 | 202 | } |
009b03aa | 203 | } |
3b9f94e1 | 204 | #endif |
009b03aa HR |
205 | if (p1) { |
206 | p1++; | |
207 | } else { | |
208 | p1 = base_path; | |
209 | } | |
210 | if (p1 > p) { | |
211 | p = p1; | |
3b0d4f61 | 212 | } |
009b03aa HR |
213 | len = p - base_path; |
214 | ||
215 | result = g_malloc(len + strlen(filename) + 1); | |
216 | memcpy(result, base_path, len); | |
217 | strcpy(result + len, filename); | |
218 | ||
219 | return result; | |
220 | } | |
221 | ||
03c320d8 HR |
222 | /* |
223 | * Helper function for bdrv_parse_filename() implementations to remove optional | |
224 | * protocol prefixes (especially "file:") from a filename and for putting the | |
225 | * stripped filename into the options QDict if there is such a prefix. | |
226 | */ | |
227 | void bdrv_parse_filename_strip_prefix(const char *filename, const char *prefix, | |
228 | QDict *options) | |
229 | { | |
230 | if (strstart(filename, prefix, &filename)) { | |
231 | /* Stripping the explicit protocol prefix may result in a protocol | |
232 | * prefix being (wrongly) detected (if the filename contains a colon) */ | |
233 | if (path_has_protocol(filename)) { | |
18cf67c5 | 234 | GString *fat_filename; |
03c320d8 HR |
235 | |
236 | /* This means there is some colon before the first slash; therefore, | |
237 | * this cannot be an absolute path */ | |
238 | assert(!path_is_absolute(filename)); | |
239 | ||
240 | /* And we can thus fix the protocol detection issue by prefixing it | |
241 | * by "./" */ | |
18cf67c5 MA |
242 | fat_filename = g_string_new("./"); |
243 | g_string_append(fat_filename, filename); | |
03c320d8 | 244 | |
18cf67c5 | 245 | assert(!path_has_protocol(fat_filename->str)); |
03c320d8 | 246 | |
18cf67c5 MA |
247 | qdict_put(options, "filename", |
248 | qstring_from_gstring(fat_filename)); | |
03c320d8 HR |
249 | } else { |
250 | /* If no protocol prefix was detected, we can use the shortened | |
251 | * filename as-is */ | |
252 | qdict_put_str(options, "filename", filename); | |
253 | } | |
254 | } | |
255 | } | |
256 | ||
257 | ||
9c5e6594 KW |
258 | /* Returns whether the image file is opened as read-only. Note that this can |
259 | * return false and writing to the image file is still not possible because the | |
260 | * image is inactivated. */ | |
93ed524e JC |
261 | bool bdrv_is_read_only(BlockDriverState *bs) |
262 | { | |
975da073 | 263 | return !(bs->open_flags & BDRV_O_RDWR); |
93ed524e JC |
264 | } |
265 | ||
54a32bfe KW |
266 | int bdrv_can_set_read_only(BlockDriverState *bs, bool read_only, |
267 | bool ignore_allow_rdw, Error **errp) | |
fe5241bf | 268 | { |
e2b8247a JC |
269 | /* Do not set read_only if copy_on_read is enabled */ |
270 | if (bs->copy_on_read && read_only) { | |
271 | error_setg(errp, "Can't set node '%s' to r/o with copy-on-read enabled", | |
272 | bdrv_get_device_or_node_name(bs)); | |
273 | return -EINVAL; | |
274 | } | |
275 | ||
d6fcdf06 | 276 | /* Do not clear read_only if it is prohibited */ |
54a32bfe KW |
277 | if (!read_only && !(bs->open_flags & BDRV_O_ALLOW_RDWR) && |
278 | !ignore_allow_rdw) | |
279 | { | |
d6fcdf06 JC |
280 | error_setg(errp, "Node '%s' is read only", |
281 | bdrv_get_device_or_node_name(bs)); | |
282 | return -EPERM; | |
283 | } | |
284 | ||
45803a03 JC |
285 | return 0; |
286 | } | |
287 | ||
eaa2410f KW |
288 | /* |
289 | * Called by a driver that can only provide a read-only image. | |
290 | * | |
291 | * Returns 0 if the node is already read-only or it could switch the node to | |
292 | * read-only because BDRV_O_AUTO_RDONLY is set. | |
293 | * | |
294 | * Returns -EACCES if the node is read-write and BDRV_O_AUTO_RDONLY is not set | |
295 | * or bdrv_can_set_read_only() forbids making the node read-only. If @errmsg | |
296 | * is not NULL, it is used as the error message for the Error object. | |
297 | */ | |
298 | int bdrv_apply_auto_read_only(BlockDriverState *bs, const char *errmsg, | |
299 | Error **errp) | |
45803a03 JC |
300 | { |
301 | int ret = 0; | |
302 | ||
eaa2410f KW |
303 | if (!(bs->open_flags & BDRV_O_RDWR)) { |
304 | return 0; | |
305 | } | |
306 | if (!(bs->open_flags & BDRV_O_AUTO_RDONLY)) { | |
307 | goto fail; | |
45803a03 JC |
308 | } |
309 | ||
eaa2410f KW |
310 | ret = bdrv_can_set_read_only(bs, true, false, NULL); |
311 | if (ret < 0) { | |
312 | goto fail; | |
eeae6a59 KW |
313 | } |
314 | ||
eaa2410f KW |
315 | bs->open_flags &= ~BDRV_O_RDWR; |
316 | ||
e2b8247a | 317 | return 0; |
eaa2410f KW |
318 | |
319 | fail: | |
320 | error_setg(errp, "%s", errmsg ?: "Image is read-only"); | |
321 | return -EACCES; | |
fe5241bf JC |
322 | } |
323 | ||
645ae7d8 HR |
324 | /* |
325 | * If @backing is empty, this function returns NULL without setting | |
326 | * @errp. In all other cases, NULL will only be returned with @errp | |
327 | * set. | |
328 | * | |
329 | * Therefore, a return value of NULL without @errp set means that | |
330 | * there is no backing file; if @errp is set, there is one but its | |
331 | * absolute filename cannot be generated. | |
332 | */ | |
333 | char *bdrv_get_full_backing_filename_from_filename(const char *backed, | |
334 | const char *backing, | |
335 | Error **errp) | |
dc5a1371 | 336 | { |
645ae7d8 HR |
337 | if (backing[0] == '\0') { |
338 | return NULL; | |
339 | } else if (path_has_protocol(backing) || path_is_absolute(backing)) { | |
340 | return g_strdup(backing); | |
9f07429e HR |
341 | } else if (backed[0] == '\0' || strstart(backed, "json:", NULL)) { |
342 | error_setg(errp, "Cannot use relative backing file names for '%s'", | |
343 | backed); | |
645ae7d8 | 344 | return NULL; |
dc5a1371 | 345 | } else { |
645ae7d8 | 346 | return path_combine(backed, backing); |
dc5a1371 PB |
347 | } |
348 | } | |
349 | ||
9f4793d8 HR |
350 | /* |
351 | * If @filename is empty or NULL, this function returns NULL without | |
352 | * setting @errp. In all other cases, NULL will only be returned with | |
353 | * @errp set. | |
354 | */ | |
355 | static char *bdrv_make_absolute_filename(BlockDriverState *relative_to, | |
356 | const char *filename, Error **errp) | |
0a82855a | 357 | { |
8df68616 | 358 | char *dir, *full_name; |
9f4793d8 | 359 | |
8df68616 HR |
360 | if (!filename || filename[0] == '\0') { |
361 | return NULL; | |
362 | } else if (path_has_protocol(filename) || path_is_absolute(filename)) { | |
363 | return g_strdup(filename); | |
364 | } | |
9f07429e | 365 | |
8df68616 HR |
366 | dir = bdrv_dirname(relative_to, errp); |
367 | if (!dir) { | |
368 | return NULL; | |
369 | } | |
f30c66ba | 370 | |
8df68616 HR |
371 | full_name = g_strconcat(dir, filename, NULL); |
372 | g_free(dir); | |
373 | return full_name; | |
9f4793d8 HR |
374 | } |
375 | ||
376 | char *bdrv_get_full_backing_filename(BlockDriverState *bs, Error **errp) | |
377 | { | |
378 | return bdrv_make_absolute_filename(bs, bs->backing_file, errp); | |
0a82855a HR |
379 | } |
380 | ||
0eb7217e SH |
381 | void bdrv_register(BlockDriver *bdrv) |
382 | { | |
a15f08dc | 383 | assert(bdrv->format_name); |
8a22f02a | 384 | QLIST_INSERT_HEAD(&bdrv_drivers, bdrv, list); |
ea2384d3 | 385 | } |
b338082b | 386 | |
e4e9986b MA |
387 | BlockDriverState *bdrv_new(void) |
388 | { | |
389 | BlockDriverState *bs; | |
390 | int i; | |
391 | ||
5839e53b | 392 | bs = g_new0(BlockDriverState, 1); |
e4654d2d | 393 | QLIST_INIT(&bs->dirty_bitmaps); |
fbe40ff7 FZ |
394 | for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) { |
395 | QLIST_INIT(&bs->op_blockers[i]); | |
396 | } | |
3783fa3d | 397 | qemu_co_mutex_init(&bs->reqs_lock); |
2119882c | 398 | qemu_mutex_init(&bs->dirty_bitmap_mutex); |
9fcb0251 | 399 | bs->refcnt = 1; |
dcd04228 | 400 | bs->aio_context = qemu_get_aio_context(); |
d7d512f6 | 401 | |
3ff2f67a EY |
402 | qemu_co_queue_init(&bs->flush_queue); |
403 | ||
0f12264e KW |
404 | for (i = 0; i < bdrv_drain_all_count; i++) { |
405 | bdrv_drained_begin(bs); | |
406 | } | |
407 | ||
2c1d04e0 HR |
408 | QTAILQ_INSERT_TAIL(&all_bdrv_states, bs, bs_list); |
409 | ||
b338082b FB |
410 | return bs; |
411 | } | |
412 | ||
88d88798 | 413 | static BlockDriver *bdrv_do_find_format(const char *format_name) |
ea2384d3 FB |
414 | { |
415 | BlockDriver *drv1; | |
88d88798 | 416 | |
8a22f02a SH |
417 | QLIST_FOREACH(drv1, &bdrv_drivers, list) { |
418 | if (!strcmp(drv1->format_name, format_name)) { | |
ea2384d3 | 419 | return drv1; |
8a22f02a | 420 | } |
ea2384d3 | 421 | } |
88d88798 | 422 | |
ea2384d3 FB |
423 | return NULL; |
424 | } | |
425 | ||
88d88798 MM |
426 | BlockDriver *bdrv_find_format(const char *format_name) |
427 | { | |
428 | BlockDriver *drv1; | |
429 | int i; | |
430 | ||
431 | drv1 = bdrv_do_find_format(format_name); | |
432 | if (drv1) { | |
433 | return drv1; | |
434 | } | |
435 | ||
436 | /* The driver isn't registered, maybe we need to load a module */ | |
437 | for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); ++i) { | |
438 | if (!strcmp(block_driver_modules[i].format_name, format_name)) { | |
439 | block_module_load_one(block_driver_modules[i].library_name); | |
440 | break; | |
441 | } | |
442 | } | |
443 | ||
444 | return bdrv_do_find_format(format_name); | |
445 | } | |
446 | ||
9ac404c5 | 447 | static int bdrv_format_is_whitelisted(const char *format_name, bool read_only) |
eb852011 | 448 | { |
b64ec4e4 FZ |
449 | static const char *whitelist_rw[] = { |
450 | CONFIG_BDRV_RW_WHITELIST | |
859aef02 | 451 | NULL |
b64ec4e4 FZ |
452 | }; |
453 | static const char *whitelist_ro[] = { | |
454 | CONFIG_BDRV_RO_WHITELIST | |
859aef02 | 455 | NULL |
eb852011 MA |
456 | }; |
457 | const char **p; | |
458 | ||
b64ec4e4 | 459 | if (!whitelist_rw[0] && !whitelist_ro[0]) { |
eb852011 | 460 | return 1; /* no whitelist, anything goes */ |
b64ec4e4 | 461 | } |
eb852011 | 462 | |
b64ec4e4 | 463 | for (p = whitelist_rw; *p; p++) { |
9ac404c5 | 464 | if (!strcmp(format_name, *p)) { |
eb852011 MA |
465 | return 1; |
466 | } | |
467 | } | |
b64ec4e4 FZ |
468 | if (read_only) { |
469 | for (p = whitelist_ro; *p; p++) { | |
9ac404c5 | 470 | if (!strcmp(format_name, *p)) { |
b64ec4e4 FZ |
471 | return 1; |
472 | } | |
473 | } | |
474 | } | |
eb852011 MA |
475 | return 0; |
476 | } | |
477 | ||
9ac404c5 AS |
478 | int bdrv_is_whitelisted(BlockDriver *drv, bool read_only) |
479 | { | |
480 | return bdrv_format_is_whitelisted(drv->format_name, read_only); | |
481 | } | |
482 | ||
e6ff69bf DB |
483 | bool bdrv_uses_whitelist(void) |
484 | { | |
485 | return use_bdrv_whitelist; | |
486 | } | |
487 | ||
5b7e1542 ZYW |
488 | typedef struct CreateCo { |
489 | BlockDriver *drv; | |
490 | char *filename; | |
83d0521a | 491 | QemuOpts *opts; |
5b7e1542 | 492 | int ret; |
cc84d90f | 493 | Error *err; |
5b7e1542 ZYW |
494 | } CreateCo; |
495 | ||
496 | static void coroutine_fn bdrv_create_co_entry(void *opaque) | |
497 | { | |
cc84d90f HR |
498 | Error *local_err = NULL; |
499 | int ret; | |
500 | ||
5b7e1542 ZYW |
501 | CreateCo *cco = opaque; |
502 | assert(cco->drv); | |
503 | ||
b92902df ML |
504 | ret = cco->drv->bdrv_co_create_opts(cco->drv, |
505 | cco->filename, cco->opts, &local_err); | |
621ff94d | 506 | error_propagate(&cco->err, local_err); |
cc84d90f | 507 | cco->ret = ret; |
5b7e1542 ZYW |
508 | } |
509 | ||
0e7e1989 | 510 | int bdrv_create(BlockDriver *drv, const char* filename, |
83d0521a | 511 | QemuOpts *opts, Error **errp) |
ea2384d3 | 512 | { |
5b7e1542 ZYW |
513 | int ret; |
514 | ||
515 | Coroutine *co; | |
516 | CreateCo cco = { | |
517 | .drv = drv, | |
518 | .filename = g_strdup(filename), | |
83d0521a | 519 | .opts = opts, |
5b7e1542 | 520 | .ret = NOT_DONE, |
cc84d90f | 521 | .err = NULL, |
5b7e1542 ZYW |
522 | }; |
523 | ||
efc75e2a | 524 | if (!drv->bdrv_co_create_opts) { |
cc84d90f | 525 | error_setg(errp, "Driver '%s' does not support image creation", drv->format_name); |
80168bff LC |
526 | ret = -ENOTSUP; |
527 | goto out; | |
5b7e1542 ZYW |
528 | } |
529 | ||
530 | if (qemu_in_coroutine()) { | |
531 | /* Fast-path if already in coroutine context */ | |
532 | bdrv_create_co_entry(&cco); | |
533 | } else { | |
0b8b8753 PB |
534 | co = qemu_coroutine_create(bdrv_create_co_entry, &cco); |
535 | qemu_coroutine_enter(co); | |
5b7e1542 | 536 | while (cco.ret == NOT_DONE) { |
b47ec2c4 | 537 | aio_poll(qemu_get_aio_context(), true); |
5b7e1542 ZYW |
538 | } |
539 | } | |
540 | ||
541 | ret = cco.ret; | |
cc84d90f | 542 | if (ret < 0) { |
84d18f06 | 543 | if (cco.err) { |
cc84d90f HR |
544 | error_propagate(errp, cco.err); |
545 | } else { | |
546 | error_setg_errno(errp, -ret, "Could not create image"); | |
547 | } | |
548 | } | |
0e7e1989 | 549 | |
80168bff LC |
550 | out: |
551 | g_free(cco.filename); | |
5b7e1542 | 552 | return ret; |
ea2384d3 FB |
553 | } |
554 | ||
fd17146c HR |
555 | /** |
556 | * Helper function for bdrv_create_file_fallback(): Resize @blk to at | |
557 | * least the given @minimum_size. | |
558 | * | |
559 | * On success, return @blk's actual length. | |
560 | * Otherwise, return -errno. | |
561 | */ | |
562 | static int64_t create_file_fallback_truncate(BlockBackend *blk, | |
563 | int64_t minimum_size, Error **errp) | |
84a12e66 | 564 | { |
cc84d90f | 565 | Error *local_err = NULL; |
fd17146c | 566 | int64_t size; |
cc84d90f | 567 | int ret; |
84a12e66 | 568 | |
8c6242b6 KW |
569 | ret = blk_truncate(blk, minimum_size, false, PREALLOC_MODE_OFF, 0, |
570 | &local_err); | |
fd17146c HR |
571 | if (ret < 0 && ret != -ENOTSUP) { |
572 | error_propagate(errp, local_err); | |
573 | return ret; | |
574 | } | |
575 | ||
576 | size = blk_getlength(blk); | |
577 | if (size < 0) { | |
578 | error_free(local_err); | |
579 | error_setg_errno(errp, -size, | |
580 | "Failed to inquire the new image file's length"); | |
581 | return size; | |
582 | } | |
583 | ||
584 | if (size < minimum_size) { | |
585 | /* Need to grow the image, but we failed to do that */ | |
586 | error_propagate(errp, local_err); | |
587 | return -ENOTSUP; | |
588 | } | |
589 | ||
590 | error_free(local_err); | |
591 | local_err = NULL; | |
592 | ||
593 | return size; | |
594 | } | |
595 | ||
596 | /** | |
597 | * Helper function for bdrv_create_file_fallback(): Zero the first | |
598 | * sector to remove any potentially pre-existing image header. | |
599 | */ | |
600 | static int create_file_fallback_zero_first_sector(BlockBackend *blk, | |
601 | int64_t current_size, | |
602 | Error **errp) | |
603 | { | |
604 | int64_t bytes_to_clear; | |
605 | int ret; | |
606 | ||
607 | bytes_to_clear = MIN(current_size, BDRV_SECTOR_SIZE); | |
608 | if (bytes_to_clear) { | |
609 | ret = blk_pwrite_zeroes(blk, 0, bytes_to_clear, BDRV_REQ_MAY_UNMAP); | |
610 | if (ret < 0) { | |
611 | error_setg_errno(errp, -ret, | |
612 | "Failed to clear the new image's first sector"); | |
613 | return ret; | |
614 | } | |
615 | } | |
616 | ||
617 | return 0; | |
618 | } | |
619 | ||
5a5e7f8c ML |
620 | /** |
621 | * Simple implementation of bdrv_co_create_opts for protocol drivers | |
622 | * which only support creation via opening a file | |
623 | * (usually existing raw storage device) | |
624 | */ | |
625 | int coroutine_fn bdrv_co_create_opts_simple(BlockDriver *drv, | |
626 | const char *filename, | |
627 | QemuOpts *opts, | |
628 | Error **errp) | |
fd17146c HR |
629 | { |
630 | BlockBackend *blk; | |
eeea1faa | 631 | QDict *options; |
fd17146c HR |
632 | int64_t size = 0; |
633 | char *buf = NULL; | |
634 | PreallocMode prealloc; | |
635 | Error *local_err = NULL; | |
636 | int ret; | |
637 | ||
638 | size = qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0); | |
639 | buf = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC); | |
640 | prealloc = qapi_enum_parse(&PreallocMode_lookup, buf, | |
641 | PREALLOC_MODE_OFF, &local_err); | |
642 | g_free(buf); | |
643 | if (local_err) { | |
644 | error_propagate(errp, local_err); | |
645 | return -EINVAL; | |
646 | } | |
647 | ||
648 | if (prealloc != PREALLOC_MODE_OFF) { | |
649 | error_setg(errp, "Unsupported preallocation mode '%s'", | |
650 | PreallocMode_str(prealloc)); | |
651 | return -ENOTSUP; | |
652 | } | |
653 | ||
eeea1faa | 654 | options = qdict_new(); |
fd17146c HR |
655 | qdict_put_str(options, "driver", drv->format_name); |
656 | ||
657 | blk = blk_new_open(filename, NULL, options, | |
658 | BDRV_O_RDWR | BDRV_O_RESIZE, errp); | |
659 | if (!blk) { | |
660 | error_prepend(errp, "Protocol driver '%s' does not support image " | |
661 | "creation, and opening the image failed: ", | |
662 | drv->format_name); | |
663 | return -EINVAL; | |
664 | } | |
665 | ||
666 | size = create_file_fallback_truncate(blk, size, errp); | |
667 | if (size < 0) { | |
668 | ret = size; | |
669 | goto out; | |
670 | } | |
671 | ||
672 | ret = create_file_fallback_zero_first_sector(blk, size, errp); | |
673 | if (ret < 0) { | |
674 | goto out; | |
675 | } | |
676 | ||
677 | ret = 0; | |
678 | out: | |
679 | blk_unref(blk); | |
680 | return ret; | |
681 | } | |
682 | ||
683 | int bdrv_create_file(const char *filename, QemuOpts *opts, Error **errp) | |
684 | { | |
729222af | 685 | QemuOpts *protocol_opts; |
fd17146c | 686 | BlockDriver *drv; |
729222af SG |
687 | QDict *qdict; |
688 | int ret; | |
fd17146c | 689 | |
b65a5e12 | 690 | drv = bdrv_find_protocol(filename, true, errp); |
84a12e66 | 691 | if (drv == NULL) { |
16905d71 | 692 | return -ENOENT; |
84a12e66 CH |
693 | } |
694 | ||
729222af SG |
695 | if (!drv->create_opts) { |
696 | error_setg(errp, "Driver '%s' does not support image creation", | |
697 | drv->format_name); | |
698 | return -ENOTSUP; | |
699 | } | |
700 | ||
701 | /* | |
702 | * 'opts' contains a QemuOptsList with a combination of format and protocol | |
703 | * default values. | |
704 | * | |
705 | * The format properly removes its options, but the default values remain | |
706 | * in 'opts->list'. So if the protocol has options with the same name | |
707 | * (e.g. rbd has 'cluster_size' as qcow2), it will see the default values | |
708 | * of the format, since for overlapping options, the format wins. | |
709 | * | |
710 | * To avoid this issue, lets convert QemuOpts to QDict, in this way we take | |
711 | * only the set options, and then convert it back to QemuOpts, using the | |
712 | * create_opts of the protocol. So the new QemuOpts, will contain only the | |
713 | * protocol defaults. | |
714 | */ | |
715 | qdict = qemu_opts_to_qdict(opts, NULL); | |
716 | protocol_opts = qemu_opts_from_qdict(drv->create_opts, qdict, errp); | |
717 | if (protocol_opts == NULL) { | |
718 | ret = -EINVAL; | |
719 | goto out; | |
720 | } | |
721 | ||
722 | ret = bdrv_create(drv, filename, protocol_opts, errp); | |
723 | out: | |
724 | qemu_opts_del(protocol_opts); | |
725 | qobject_unref(qdict); | |
726 | return ret; | |
84a12e66 CH |
727 | } |
728 | ||
e1d7f8bb DHB |
729 | int coroutine_fn bdrv_co_delete_file(BlockDriverState *bs, Error **errp) |
730 | { | |
731 | Error *local_err = NULL; | |
732 | int ret; | |
733 | ||
734 | assert(bs != NULL); | |
735 | ||
736 | if (!bs->drv) { | |
737 | error_setg(errp, "Block node '%s' is not opened", bs->filename); | |
738 | return -ENOMEDIUM; | |
739 | } | |
740 | ||
741 | if (!bs->drv->bdrv_co_delete_file) { | |
742 | error_setg(errp, "Driver '%s' does not support image deletion", | |
743 | bs->drv->format_name); | |
744 | return -ENOTSUP; | |
745 | } | |
746 | ||
747 | ret = bs->drv->bdrv_co_delete_file(bs, &local_err); | |
748 | if (ret < 0) { | |
749 | error_propagate(errp, local_err); | |
750 | } | |
751 | ||
752 | return ret; | |
753 | } | |
754 | ||
a890f08e ML |
755 | void coroutine_fn bdrv_co_delete_file_noerr(BlockDriverState *bs) |
756 | { | |
757 | Error *local_err = NULL; | |
758 | int ret; | |
759 | ||
760 | if (!bs) { | |
761 | return; | |
762 | } | |
763 | ||
764 | ret = bdrv_co_delete_file(bs, &local_err); | |
765 | /* | |
766 | * ENOTSUP will happen if the block driver doesn't support | |
767 | * the 'bdrv_co_delete_file' interface. This is a predictable | |
768 | * scenario and shouldn't be reported back to the user. | |
769 | */ | |
770 | if (ret == -ENOTSUP) { | |
771 | error_free(local_err); | |
772 | } else if (ret < 0) { | |
773 | error_report_err(local_err); | |
774 | } | |
775 | } | |
776 | ||
892b7de8 ET |
777 | /** |
778 | * Try to get @bs's logical and physical block size. | |
779 | * On success, store them in @bsz struct and return 0. | |
780 | * On failure return -errno. | |
781 | * @bs must not be empty. | |
782 | */ | |
783 | int bdrv_probe_blocksizes(BlockDriverState *bs, BlockSizes *bsz) | |
784 | { | |
785 | BlockDriver *drv = bs->drv; | |
93393e69 | 786 | BlockDriverState *filtered = bdrv_filter_bs(bs); |
892b7de8 ET |
787 | |
788 | if (drv && drv->bdrv_probe_blocksizes) { | |
789 | return drv->bdrv_probe_blocksizes(bs, bsz); | |
93393e69 HR |
790 | } else if (filtered) { |
791 | return bdrv_probe_blocksizes(filtered, bsz); | |
892b7de8 ET |
792 | } |
793 | ||
794 | return -ENOTSUP; | |
795 | } | |
796 | ||
797 | /** | |
798 | * Try to get @bs's geometry (cyls, heads, sectors). | |
799 | * On success, store them in @geo struct and return 0. | |
800 | * On failure return -errno. | |
801 | * @bs must not be empty. | |
802 | */ | |
803 | int bdrv_probe_geometry(BlockDriverState *bs, HDGeometry *geo) | |
804 | { | |
805 | BlockDriver *drv = bs->drv; | |
93393e69 | 806 | BlockDriverState *filtered = bdrv_filter_bs(bs); |
892b7de8 ET |
807 | |
808 | if (drv && drv->bdrv_probe_geometry) { | |
809 | return drv->bdrv_probe_geometry(bs, geo); | |
93393e69 HR |
810 | } else if (filtered) { |
811 | return bdrv_probe_geometry(filtered, geo); | |
892b7de8 ET |
812 | } |
813 | ||
814 | return -ENOTSUP; | |
815 | } | |
816 | ||
eba25057 JM |
817 | /* |
818 | * Create a uniquely-named empty temporary file. | |
819 | * Return 0 upon success, otherwise a negative errno value. | |
820 | */ | |
821 | int get_tmp_filename(char *filename, int size) | |
d5249393 | 822 | { |
eba25057 | 823 | #ifdef _WIN32 |
3b9f94e1 | 824 | char temp_dir[MAX_PATH]; |
eba25057 JM |
825 | /* GetTempFileName requires that its output buffer (4th param) |
826 | have length MAX_PATH or greater. */ | |
827 | assert(size >= MAX_PATH); | |
828 | return (GetTempPath(MAX_PATH, temp_dir) | |
829 | && GetTempFileName(temp_dir, "qem", 0, filename) | |
830 | ? 0 : -GetLastError()); | |
d5249393 | 831 | #else |
67b915a5 | 832 | int fd; |
7ccfb2eb | 833 | const char *tmpdir; |
0badc1ee | 834 | tmpdir = getenv("TMPDIR"); |
69bef793 AS |
835 | if (!tmpdir) { |
836 | tmpdir = "/var/tmp"; | |
837 | } | |
eba25057 JM |
838 | if (snprintf(filename, size, "%s/vl.XXXXXX", tmpdir) >= size) { |
839 | return -EOVERFLOW; | |
840 | } | |
ea2384d3 | 841 | fd = mkstemp(filename); |
fe235a06 DH |
842 | if (fd < 0) { |
843 | return -errno; | |
844 | } | |
845 | if (close(fd) != 0) { | |
846 | unlink(filename); | |
eba25057 JM |
847 | return -errno; |
848 | } | |
849 | return 0; | |
d5249393 | 850 | #endif |
eba25057 | 851 | } |
fc01f7e7 | 852 | |
84a12e66 CH |
853 | /* |
854 | * Detect host devices. By convention, /dev/cdrom[N] is always | |
855 | * recognized as a host CDROM. | |
856 | */ | |
857 | static BlockDriver *find_hdev_driver(const char *filename) | |
858 | { | |
859 | int score_max = 0, score; | |
860 | BlockDriver *drv = NULL, *d; | |
861 | ||
862 | QLIST_FOREACH(d, &bdrv_drivers, list) { | |
863 | if (d->bdrv_probe_device) { | |
864 | score = d->bdrv_probe_device(filename); | |
865 | if (score > score_max) { | |
866 | score_max = score; | |
867 | drv = d; | |
868 | } | |
869 | } | |
870 | } | |
871 | ||
872 | return drv; | |
873 | } | |
874 | ||
88d88798 MM |
875 | static BlockDriver *bdrv_do_find_protocol(const char *protocol) |
876 | { | |
877 | BlockDriver *drv1; | |
878 | ||
879 | QLIST_FOREACH(drv1, &bdrv_drivers, list) { | |
880 | if (drv1->protocol_name && !strcmp(drv1->protocol_name, protocol)) { | |
881 | return drv1; | |
882 | } | |
883 | } | |
884 | ||
885 | return NULL; | |
886 | } | |
887 | ||
98289620 | 888 | BlockDriver *bdrv_find_protocol(const char *filename, |
b65a5e12 HR |
889 | bool allow_protocol_prefix, |
890 | Error **errp) | |
83f64091 FB |
891 | { |
892 | BlockDriver *drv1; | |
893 | char protocol[128]; | |
1cec71e3 | 894 | int len; |
83f64091 | 895 | const char *p; |
88d88798 | 896 | int i; |
19cb3738 | 897 | |
66f82cee KW |
898 | /* TODO Drivers without bdrv_file_open must be specified explicitly */ |
899 | ||
39508e7a CH |
900 | /* |
901 | * XXX(hch): we really should not let host device detection | |
902 | * override an explicit protocol specification, but moving this | |
903 | * later breaks access to device names with colons in them. | |
904 | * Thanks to the brain-dead persistent naming schemes on udev- | |
905 | * based Linux systems those actually are quite common. | |
906 | */ | |
907 | drv1 = find_hdev_driver(filename); | |
908 | if (drv1) { | |
909 | return drv1; | |
910 | } | |
911 | ||
98289620 | 912 | if (!path_has_protocol(filename) || !allow_protocol_prefix) { |
ef810437 | 913 | return &bdrv_file; |
84a12e66 | 914 | } |
98289620 | 915 | |
9e0b22f4 SH |
916 | p = strchr(filename, ':'); |
917 | assert(p != NULL); | |
1cec71e3 AL |
918 | len = p - filename; |
919 | if (len > sizeof(protocol) - 1) | |
920 | len = sizeof(protocol) - 1; | |
921 | memcpy(protocol, filename, len); | |
922 | protocol[len] = '\0'; | |
88d88798 MM |
923 | |
924 | drv1 = bdrv_do_find_protocol(protocol); | |
925 | if (drv1) { | |
926 | return drv1; | |
927 | } | |
928 | ||
929 | for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); ++i) { | |
930 | if (block_driver_modules[i].protocol_name && | |
931 | !strcmp(block_driver_modules[i].protocol_name, protocol)) { | |
932 | block_module_load_one(block_driver_modules[i].library_name); | |
933 | break; | |
8a22f02a | 934 | } |
83f64091 | 935 | } |
b65a5e12 | 936 | |
88d88798 MM |
937 | drv1 = bdrv_do_find_protocol(protocol); |
938 | if (!drv1) { | |
939 | error_setg(errp, "Unknown protocol '%s'", protocol); | |
940 | } | |
941 | return drv1; | |
83f64091 FB |
942 | } |
943 | ||
c6684249 MA |
944 | /* |
945 | * Guess image format by probing its contents. | |
946 | * This is not a good idea when your image is raw (CVE-2008-2004), but | |
947 | * we do it anyway for backward compatibility. | |
948 | * | |
949 | * @buf contains the image's first @buf_size bytes. | |
7cddd372 KW |
950 | * @buf_size is the buffer size in bytes (generally BLOCK_PROBE_BUF_SIZE, |
951 | * but can be smaller if the image file is smaller) | |
c6684249 MA |
952 | * @filename is its filename. |
953 | * | |
954 | * For all block drivers, call the bdrv_probe() method to get its | |
955 | * probing score. | |
956 | * Return the first block driver with the highest probing score. | |
957 | */ | |
38f3ef57 KW |
958 | BlockDriver *bdrv_probe_all(const uint8_t *buf, int buf_size, |
959 | const char *filename) | |
c6684249 MA |
960 | { |
961 | int score_max = 0, score; | |
962 | BlockDriver *drv = NULL, *d; | |
963 | ||
964 | QLIST_FOREACH(d, &bdrv_drivers, list) { | |
965 | if (d->bdrv_probe) { | |
966 | score = d->bdrv_probe(buf, buf_size, filename); | |
967 | if (score > score_max) { | |
968 | score_max = score; | |
969 | drv = d; | |
970 | } | |
971 | } | |
972 | } | |
973 | ||
974 | return drv; | |
975 | } | |
976 | ||
5696c6e3 | 977 | static int find_image_format(BlockBackend *file, const char *filename, |
34b5d2c6 | 978 | BlockDriver **pdrv, Error **errp) |
f3a5d3f8 | 979 | { |
c6684249 | 980 | BlockDriver *drv; |
7cddd372 | 981 | uint8_t buf[BLOCK_PROBE_BUF_SIZE]; |
f500a6d3 | 982 | int ret = 0; |
f8ea0b00 | 983 | |
08a00559 | 984 | /* Return the raw BlockDriver * to scsi-generic devices or empty drives */ |
5696c6e3 | 985 | if (blk_is_sg(file) || !blk_is_inserted(file) || blk_getlength(file) == 0) { |
ef810437 | 986 | *pdrv = &bdrv_raw; |
c98ac35d | 987 | return ret; |
1a396859 | 988 | } |
f8ea0b00 | 989 | |
5696c6e3 | 990 | ret = blk_pread(file, 0, buf, sizeof(buf)); |
83f64091 | 991 | if (ret < 0) { |
34b5d2c6 HR |
992 | error_setg_errno(errp, -ret, "Could not read image for determining its " |
993 | "format"); | |
c98ac35d SW |
994 | *pdrv = NULL; |
995 | return ret; | |
83f64091 FB |
996 | } |
997 | ||
c6684249 | 998 | drv = bdrv_probe_all(buf, ret, filename); |
c98ac35d | 999 | if (!drv) { |
34b5d2c6 HR |
1000 | error_setg(errp, "Could not determine image format: No compatible " |
1001 | "driver found"); | |
c98ac35d SW |
1002 | ret = -ENOENT; |
1003 | } | |
1004 | *pdrv = drv; | |
1005 | return ret; | |
ea2384d3 FB |
1006 | } |
1007 | ||
51762288 SH |
1008 | /** |
1009 | * Set the current 'total_sectors' value | |
65a9bb25 | 1010 | * Return 0 on success, -errno on error. |
51762288 | 1011 | */ |
3d9f2d2a | 1012 | int refresh_total_sectors(BlockDriverState *bs, int64_t hint) |
51762288 SH |
1013 | { |
1014 | BlockDriver *drv = bs->drv; | |
1015 | ||
d470ad42 HR |
1016 | if (!drv) { |
1017 | return -ENOMEDIUM; | |
1018 | } | |
1019 | ||
396759ad | 1020 | /* Do not attempt drv->bdrv_getlength() on scsi-generic devices */ |
b192af8a | 1021 | if (bdrv_is_sg(bs)) |
396759ad NB |
1022 | return 0; |
1023 | ||
51762288 SH |
1024 | /* query actual device if possible, otherwise just trust the hint */ |
1025 | if (drv->bdrv_getlength) { | |
1026 | int64_t length = drv->bdrv_getlength(bs); | |
1027 | if (length < 0) { | |
1028 | return length; | |
1029 | } | |
7e382003 | 1030 | hint = DIV_ROUND_UP(length, BDRV_SECTOR_SIZE); |
51762288 SH |
1031 | } |
1032 | ||
1033 | bs->total_sectors = hint; | |
8b117001 VSO |
1034 | |
1035 | if (bs->total_sectors * BDRV_SECTOR_SIZE > BDRV_MAX_LENGTH) { | |
1036 | return -EFBIG; | |
1037 | } | |
1038 | ||
51762288 SH |
1039 | return 0; |
1040 | } | |
1041 | ||
cddff5ba KW |
1042 | /** |
1043 | * Combines a QDict of new block driver @options with any missing options taken | |
1044 | * from @old_options, so that leaving out an option defaults to its old value. | |
1045 | */ | |
1046 | static void bdrv_join_options(BlockDriverState *bs, QDict *options, | |
1047 | QDict *old_options) | |
1048 | { | |
1049 | if (bs->drv && bs->drv->bdrv_join_options) { | |
1050 | bs->drv->bdrv_join_options(options, old_options); | |
1051 | } else { | |
1052 | qdict_join(options, old_options, false); | |
1053 | } | |
1054 | } | |
1055 | ||
543770bd AG |
1056 | static BlockdevDetectZeroesOptions bdrv_parse_detect_zeroes(QemuOpts *opts, |
1057 | int open_flags, | |
1058 | Error **errp) | |
1059 | { | |
1060 | Error *local_err = NULL; | |
1061 | char *value = qemu_opt_get_del(opts, "detect-zeroes"); | |
1062 | BlockdevDetectZeroesOptions detect_zeroes = | |
1063 | qapi_enum_parse(&BlockdevDetectZeroesOptions_lookup, value, | |
1064 | BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF, &local_err); | |
1065 | g_free(value); | |
1066 | if (local_err) { | |
1067 | error_propagate(errp, local_err); | |
1068 | return detect_zeroes; | |
1069 | } | |
1070 | ||
1071 | if (detect_zeroes == BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP && | |
1072 | !(open_flags & BDRV_O_UNMAP)) | |
1073 | { | |
1074 | error_setg(errp, "setting detect-zeroes to unmap is not allowed " | |
1075 | "without setting discard operation to unmap"); | |
1076 | } | |
1077 | ||
1078 | return detect_zeroes; | |
1079 | } | |
1080 | ||
f80f2673 AM |
1081 | /** |
1082 | * Set open flags for aio engine | |
1083 | * | |
1084 | * Return 0 on success, -1 if the engine specified is invalid | |
1085 | */ | |
1086 | int bdrv_parse_aio(const char *mode, int *flags) | |
1087 | { | |
1088 | if (!strcmp(mode, "threads")) { | |
1089 | /* do nothing, default */ | |
1090 | } else if (!strcmp(mode, "native")) { | |
1091 | *flags |= BDRV_O_NATIVE_AIO; | |
1092 | #ifdef CONFIG_LINUX_IO_URING | |
1093 | } else if (!strcmp(mode, "io_uring")) { | |
1094 | *flags |= BDRV_O_IO_URING; | |
1095 | #endif | |
1096 | } else { | |
1097 | return -1; | |
1098 | } | |
1099 | ||
1100 | return 0; | |
1101 | } | |
1102 | ||
9e8f1835 PB |
1103 | /** |
1104 | * Set open flags for a given discard mode | |
1105 | * | |
1106 | * Return 0 on success, -1 if the discard mode was invalid. | |
1107 | */ | |
1108 | int bdrv_parse_discard_flags(const char *mode, int *flags) | |
1109 | { | |
1110 | *flags &= ~BDRV_O_UNMAP; | |
1111 | ||
1112 | if (!strcmp(mode, "off") || !strcmp(mode, "ignore")) { | |
1113 | /* do nothing */ | |
1114 | } else if (!strcmp(mode, "on") || !strcmp(mode, "unmap")) { | |
1115 | *flags |= BDRV_O_UNMAP; | |
1116 | } else { | |
1117 | return -1; | |
1118 | } | |
1119 | ||
1120 | return 0; | |
1121 | } | |
1122 | ||
c3993cdc SH |
1123 | /** |
1124 | * Set open flags for a given cache mode | |
1125 | * | |
1126 | * Return 0 on success, -1 if the cache mode was invalid. | |
1127 | */ | |
53e8ae01 | 1128 | int bdrv_parse_cache_mode(const char *mode, int *flags, bool *writethrough) |
c3993cdc SH |
1129 | { |
1130 | *flags &= ~BDRV_O_CACHE_MASK; | |
1131 | ||
1132 | if (!strcmp(mode, "off") || !strcmp(mode, "none")) { | |
53e8ae01 KW |
1133 | *writethrough = false; |
1134 | *flags |= BDRV_O_NOCACHE; | |
92196b2f | 1135 | } else if (!strcmp(mode, "directsync")) { |
53e8ae01 | 1136 | *writethrough = true; |
92196b2f | 1137 | *flags |= BDRV_O_NOCACHE; |
c3993cdc | 1138 | } else if (!strcmp(mode, "writeback")) { |
53e8ae01 | 1139 | *writethrough = false; |
c3993cdc | 1140 | } else if (!strcmp(mode, "unsafe")) { |
53e8ae01 | 1141 | *writethrough = false; |
c3993cdc SH |
1142 | *flags |= BDRV_O_NO_FLUSH; |
1143 | } else if (!strcmp(mode, "writethrough")) { | |
53e8ae01 | 1144 | *writethrough = true; |
c3993cdc SH |
1145 | } else { |
1146 | return -1; | |
1147 | } | |
1148 | ||
1149 | return 0; | |
1150 | } | |
1151 | ||
b5411555 KW |
1152 | static char *bdrv_child_get_parent_desc(BdrvChild *c) |
1153 | { | |
1154 | BlockDriverState *parent = c->opaque; | |
2c0a3acb | 1155 | return g_strdup_printf("node '%s'", bdrv_get_node_name(parent)); |
b5411555 KW |
1156 | } |
1157 | ||
20018e12 KW |
1158 | static void bdrv_child_cb_drained_begin(BdrvChild *child) |
1159 | { | |
1160 | BlockDriverState *bs = child->opaque; | |
6cd5c9d7 | 1161 | bdrv_do_drained_begin_quiesce(bs, NULL, false); |
20018e12 KW |
1162 | } |
1163 | ||
89bd0305 KW |
1164 | static bool bdrv_child_cb_drained_poll(BdrvChild *child) |
1165 | { | |
1166 | BlockDriverState *bs = child->opaque; | |
6cd5c9d7 | 1167 | return bdrv_drain_poll(bs, false, NULL, false); |
89bd0305 KW |
1168 | } |
1169 | ||
e037c09c HR |
1170 | static void bdrv_child_cb_drained_end(BdrvChild *child, |
1171 | int *drained_end_counter) | |
20018e12 KW |
1172 | { |
1173 | BlockDriverState *bs = child->opaque; | |
e037c09c | 1174 | bdrv_drained_end_no_poll(bs, drained_end_counter); |
20018e12 KW |
1175 | } |
1176 | ||
38701b6a KW |
1177 | static int bdrv_child_cb_inactivate(BdrvChild *child) |
1178 | { | |
1179 | BlockDriverState *bs = child->opaque; | |
1180 | assert(bs->open_flags & BDRV_O_INACTIVE); | |
1181 | return 0; | |
1182 | } | |
1183 | ||
5d231849 KW |
1184 | static bool bdrv_child_cb_can_set_aio_ctx(BdrvChild *child, AioContext *ctx, |
1185 | GSList **ignore, Error **errp) | |
1186 | { | |
1187 | BlockDriverState *bs = child->opaque; | |
1188 | return bdrv_can_set_aio_context(bs, ctx, ignore, errp); | |
1189 | } | |
1190 | ||
53a7d041 KW |
1191 | static void bdrv_child_cb_set_aio_ctx(BdrvChild *child, AioContext *ctx, |
1192 | GSList **ignore) | |
1193 | { | |
1194 | BlockDriverState *bs = child->opaque; | |
1195 | return bdrv_set_aio_context_ignore(bs, ctx, ignore); | |
1196 | } | |
1197 | ||
b1e6fc08 | 1198 | /* |
73176bee KW |
1199 | * Returns the options and flags that a temporary snapshot should get, based on |
1200 | * the originally requested flags (the originally requested image will have | |
1201 | * flags like a backing file) | |
b1e6fc08 | 1202 | */ |
73176bee KW |
1203 | static void bdrv_temp_snapshot_options(int *child_flags, QDict *child_options, |
1204 | int parent_flags, QDict *parent_options) | |
b1e6fc08 | 1205 | { |
73176bee KW |
1206 | *child_flags = (parent_flags & ~BDRV_O_SNAPSHOT) | BDRV_O_TEMPORARY; |
1207 | ||
1208 | /* For temporary files, unconditional cache=unsafe is fine */ | |
73176bee KW |
1209 | qdict_set_default_str(child_options, BDRV_OPT_CACHE_DIRECT, "off"); |
1210 | qdict_set_default_str(child_options, BDRV_OPT_CACHE_NO_FLUSH, "on"); | |
41869044 | 1211 | |
3f48686f | 1212 | /* Copy the read-only and discard options from the parent */ |
f87a0e29 | 1213 | qdict_copy_default(child_options, parent_options, BDRV_OPT_READ_ONLY); |
3f48686f | 1214 | qdict_copy_default(child_options, parent_options, BDRV_OPT_DISCARD); |
f87a0e29 | 1215 | |
41869044 KW |
1216 | /* aio=native doesn't work for cache.direct=off, so disable it for the |
1217 | * temporary snapshot */ | |
1218 | *child_flags &= ~BDRV_O_NATIVE_AIO; | |
b1e6fc08 KW |
1219 | } |
1220 | ||
db95dbba KW |
1221 | static void bdrv_backing_attach(BdrvChild *c) |
1222 | { | |
1223 | BlockDriverState *parent = c->opaque; | |
1224 | BlockDriverState *backing_hd = c->bs; | |
1225 | ||
1226 | assert(!parent->backing_blocker); | |
1227 | error_setg(&parent->backing_blocker, | |
1228 | "node is used as backing hd of '%s'", | |
1229 | bdrv_get_device_or_node_name(parent)); | |
1230 | ||
f30c66ba HR |
1231 | bdrv_refresh_filename(backing_hd); |
1232 | ||
db95dbba | 1233 | parent->open_flags &= ~BDRV_O_NO_BACKING; |
db95dbba KW |
1234 | |
1235 | bdrv_op_block_all(backing_hd, parent->backing_blocker); | |
1236 | /* Otherwise we won't be able to commit or stream */ | |
1237 | bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_COMMIT_TARGET, | |
1238 | parent->backing_blocker); | |
1239 | bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_STREAM, | |
1240 | parent->backing_blocker); | |
1241 | /* | |
1242 | * We do backup in 3 ways: | |
1243 | * 1. drive backup | |
1244 | * The target bs is new opened, and the source is top BDS | |
1245 | * 2. blockdev backup | |
1246 | * Both the source and the target are top BDSes. | |
1247 | * 3. internal backup(used for block replication) | |
1248 | * Both the source and the target are backing file | |
1249 | * | |
1250 | * In case 1 and 2, neither the source nor the target is the backing file. | |
1251 | * In case 3, we will block the top BDS, so there is only one block job | |
1252 | * for the top BDS and its backing chain. | |
1253 | */ | |
1254 | bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_BACKUP_SOURCE, | |
1255 | parent->backing_blocker); | |
1256 | bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_BACKUP_TARGET, | |
1257 | parent->backing_blocker); | |
ca2f1234 | 1258 | } |
d736f119 | 1259 | |
db95dbba KW |
1260 | static void bdrv_backing_detach(BdrvChild *c) |
1261 | { | |
1262 | BlockDriverState *parent = c->opaque; | |
1263 | ||
1264 | assert(parent->backing_blocker); | |
1265 | bdrv_op_unblock_all(c->bs, parent->backing_blocker); | |
1266 | error_free(parent->backing_blocker); | |
1267 | parent->backing_blocker = NULL; | |
48e08288 | 1268 | } |
d736f119 | 1269 | |
6858eba0 KW |
1270 | static int bdrv_backing_update_filename(BdrvChild *c, BlockDriverState *base, |
1271 | const char *filename, Error **errp) | |
1272 | { | |
1273 | BlockDriverState *parent = c->opaque; | |
e94d3dba | 1274 | bool read_only = bdrv_is_read_only(parent); |
6858eba0 KW |
1275 | int ret; |
1276 | ||
e94d3dba AG |
1277 | if (read_only) { |
1278 | ret = bdrv_reopen_set_read_only(parent, false, errp); | |
61f09cea KW |
1279 | if (ret < 0) { |
1280 | return ret; | |
1281 | } | |
1282 | } | |
1283 | ||
6858eba0 | 1284 | ret = bdrv_change_backing_file(parent, filename, |
e54ee1b3 EB |
1285 | base->drv ? base->drv->format_name : "", |
1286 | false); | |
6858eba0 | 1287 | if (ret < 0) { |
64730694 | 1288 | error_setg_errno(errp, -ret, "Could not update backing file link"); |
6858eba0 KW |
1289 | } |
1290 | ||
e94d3dba AG |
1291 | if (read_only) { |
1292 | bdrv_reopen_set_read_only(parent, true, NULL); | |
61f09cea KW |
1293 | } |
1294 | ||
6858eba0 KW |
1295 | return ret; |
1296 | } | |
1297 | ||
fae8bd39 HR |
1298 | /* |
1299 | * Returns the options and flags that a generic child of a BDS should | |
1300 | * get, based on the given options and flags for the parent BDS. | |
1301 | */ | |
00ff7ffd HR |
1302 | static void bdrv_inherited_options(BdrvChildRole role, bool parent_is_format, |
1303 | int *child_flags, QDict *child_options, | |
1304 | int parent_flags, QDict *parent_options) | |
fae8bd39 HR |
1305 | { |
1306 | int flags = parent_flags; | |
1307 | ||
1308 | /* | |
1309 | * First, decide whether to set, clear, or leave BDRV_O_PROTOCOL. | |
1310 | * Generally, the question to answer is: Should this child be | |
1311 | * format-probed by default? | |
1312 | */ | |
1313 | ||
1314 | /* | |
1315 | * Pure and non-filtered data children of non-format nodes should | |
1316 | * be probed by default (even when the node itself has BDRV_O_PROTOCOL | |
1317 | * set). This only affects a very limited set of drivers (namely | |
1318 | * quorum and blkverify when this comment was written). | |
1319 | * Force-clear BDRV_O_PROTOCOL then. | |
1320 | */ | |
1321 | if (!parent_is_format && | |
1322 | (role & BDRV_CHILD_DATA) && | |
1323 | !(role & (BDRV_CHILD_METADATA | BDRV_CHILD_FILTERED))) | |
1324 | { | |
1325 | flags &= ~BDRV_O_PROTOCOL; | |
1326 | } | |
1327 | ||
1328 | /* | |
1329 | * All children of format nodes (except for COW children) and all | |
1330 | * metadata children in general should never be format-probed. | |
1331 | * Force-set BDRV_O_PROTOCOL then. | |
1332 | */ | |
1333 | if ((parent_is_format && !(role & BDRV_CHILD_COW)) || | |
1334 | (role & BDRV_CHILD_METADATA)) | |
1335 | { | |
1336 | flags |= BDRV_O_PROTOCOL; | |
1337 | } | |
1338 | ||
1339 | /* | |
1340 | * If the cache mode isn't explicitly set, inherit direct and no-flush from | |
1341 | * the parent. | |
1342 | */ | |
1343 | qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_DIRECT); | |
1344 | qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_NO_FLUSH); | |
1345 | qdict_copy_default(child_options, parent_options, BDRV_OPT_FORCE_SHARE); | |
1346 | ||
1347 | if (role & BDRV_CHILD_COW) { | |
1348 | /* backing files are opened read-only by default */ | |
1349 | qdict_set_default_str(child_options, BDRV_OPT_READ_ONLY, "on"); | |
1350 | qdict_set_default_str(child_options, BDRV_OPT_AUTO_READ_ONLY, "off"); | |
1351 | } else { | |
1352 | /* Inherit the read-only option from the parent if it's not set */ | |
1353 | qdict_copy_default(child_options, parent_options, BDRV_OPT_READ_ONLY); | |
1354 | qdict_copy_default(child_options, parent_options, | |
1355 | BDRV_OPT_AUTO_READ_ONLY); | |
1356 | } | |
1357 | ||
1358 | /* | |
1359 | * bdrv_co_pdiscard() respects unmap policy for the parent, so we | |
1360 | * can default to enable it on lower layers regardless of the | |
1361 | * parent option. | |
1362 | */ | |
1363 | qdict_set_default_str(child_options, BDRV_OPT_DISCARD, "unmap"); | |
1364 | ||
1365 | /* Clear flags that only apply to the top layer */ | |
1366 | flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_COPY_ON_READ); | |
1367 | ||
1368 | if (role & BDRV_CHILD_METADATA) { | |
1369 | flags &= ~BDRV_O_NO_IO; | |
1370 | } | |
1371 | if (role & BDRV_CHILD_COW) { | |
1372 | flags &= ~BDRV_O_TEMPORARY; | |
1373 | } | |
1374 | ||
1375 | *child_flags = flags; | |
1376 | } | |
1377 | ||
ca2f1234 HR |
1378 | static void bdrv_child_cb_attach(BdrvChild *child) |
1379 | { | |
1380 | BlockDriverState *bs = child->opaque; | |
1381 | ||
1382 | if (child->role & BDRV_CHILD_COW) { | |
1383 | bdrv_backing_attach(child); | |
1384 | } | |
1385 | ||
1386 | bdrv_apply_subtree_drain(child, bs); | |
1387 | } | |
1388 | ||
48e08288 HR |
1389 | static void bdrv_child_cb_detach(BdrvChild *child) |
1390 | { | |
1391 | BlockDriverState *bs = child->opaque; | |
1392 | ||
1393 | if (child->role & BDRV_CHILD_COW) { | |
1394 | bdrv_backing_detach(child); | |
1395 | } | |
1396 | ||
1397 | bdrv_unapply_subtree_drain(child, bs); | |
1398 | } | |
1399 | ||
43483550 HR |
1400 | static int bdrv_child_cb_update_filename(BdrvChild *c, BlockDriverState *base, |
1401 | const char *filename, Error **errp) | |
1402 | { | |
1403 | if (c->role & BDRV_CHILD_COW) { | |
1404 | return bdrv_backing_update_filename(c, base, filename, errp); | |
1405 | } | |
1406 | return 0; | |
1407 | } | |
1408 | ||
fb62b588 | 1409 | AioContext *child_of_bds_get_parent_aio_context(BdrvChild *c) |
3ca1f322 VSO |
1410 | { |
1411 | BlockDriverState *bs = c->opaque; | |
1412 | ||
1413 | return bdrv_get_aio_context(bs); | |
1414 | } | |
1415 | ||
43483550 HR |
1416 | const BdrvChildClass child_of_bds = { |
1417 | .parent_is_bds = true, | |
1418 | .get_parent_desc = bdrv_child_get_parent_desc, | |
1419 | .inherit_options = bdrv_inherited_options, | |
1420 | .drained_begin = bdrv_child_cb_drained_begin, | |
1421 | .drained_poll = bdrv_child_cb_drained_poll, | |
1422 | .drained_end = bdrv_child_cb_drained_end, | |
1423 | .attach = bdrv_child_cb_attach, | |
1424 | .detach = bdrv_child_cb_detach, | |
1425 | .inactivate = bdrv_child_cb_inactivate, | |
1426 | .can_set_aio_ctx = bdrv_child_cb_can_set_aio_ctx, | |
1427 | .set_aio_ctx = bdrv_child_cb_set_aio_ctx, | |
1428 | .update_filename = bdrv_child_cb_update_filename, | |
fb62b588 | 1429 | .get_parent_aio_context = child_of_bds_get_parent_aio_context, |
43483550 HR |
1430 | }; |
1431 | ||
3ca1f322 VSO |
1432 | AioContext *bdrv_child_get_parent_aio_context(BdrvChild *c) |
1433 | { | |
1434 | return c->klass->get_parent_aio_context(c); | |
1435 | } | |
1436 | ||
7b272452 KW |
1437 | static int bdrv_open_flags(BlockDriverState *bs, int flags) |
1438 | { | |
61de4c68 | 1439 | int open_flags = flags; |
7b272452 KW |
1440 | |
1441 | /* | |
1442 | * Clear flags that are internal to the block layer before opening the | |
1443 | * image. | |
1444 | */ | |
20cca275 | 1445 | open_flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_PROTOCOL); |
7b272452 | 1446 | |
7b272452 KW |
1447 | return open_flags; |
1448 | } | |
1449 | ||
91a097e7 KW |
1450 | static void update_flags_from_options(int *flags, QemuOpts *opts) |
1451 | { | |
2a3d4331 | 1452 | *flags &= ~(BDRV_O_CACHE_MASK | BDRV_O_RDWR | BDRV_O_AUTO_RDONLY); |
91a097e7 | 1453 | |
57f9db9a | 1454 | if (qemu_opt_get_bool_del(opts, BDRV_OPT_CACHE_NO_FLUSH, false)) { |
91a097e7 KW |
1455 | *flags |= BDRV_O_NO_FLUSH; |
1456 | } | |
1457 | ||
57f9db9a | 1458 | if (qemu_opt_get_bool_del(opts, BDRV_OPT_CACHE_DIRECT, false)) { |
91a097e7 KW |
1459 | *flags |= BDRV_O_NOCACHE; |
1460 | } | |
f87a0e29 | 1461 | |
57f9db9a | 1462 | if (!qemu_opt_get_bool_del(opts, BDRV_OPT_READ_ONLY, false)) { |
f87a0e29 AG |
1463 | *flags |= BDRV_O_RDWR; |
1464 | } | |
1465 | ||
e35bdc12 KW |
1466 | if (qemu_opt_get_bool_del(opts, BDRV_OPT_AUTO_READ_ONLY, false)) { |
1467 | *flags |= BDRV_O_AUTO_RDONLY; | |
1468 | } | |
91a097e7 KW |
1469 | } |
1470 | ||
1471 | static void update_options_from_flags(QDict *options, int flags) | |
1472 | { | |
91a097e7 | 1473 | if (!qdict_haskey(options, BDRV_OPT_CACHE_DIRECT)) { |
46f5ac20 | 1474 | qdict_put_bool(options, BDRV_OPT_CACHE_DIRECT, flags & BDRV_O_NOCACHE); |
91a097e7 KW |
1475 | } |
1476 | if (!qdict_haskey(options, BDRV_OPT_CACHE_NO_FLUSH)) { | |
46f5ac20 EB |
1477 | qdict_put_bool(options, BDRV_OPT_CACHE_NO_FLUSH, |
1478 | flags & BDRV_O_NO_FLUSH); | |
91a097e7 | 1479 | } |
f87a0e29 | 1480 | if (!qdict_haskey(options, BDRV_OPT_READ_ONLY)) { |
46f5ac20 | 1481 | qdict_put_bool(options, BDRV_OPT_READ_ONLY, !(flags & BDRV_O_RDWR)); |
f87a0e29 | 1482 | } |
e35bdc12 KW |
1483 | if (!qdict_haskey(options, BDRV_OPT_AUTO_READ_ONLY)) { |
1484 | qdict_put_bool(options, BDRV_OPT_AUTO_READ_ONLY, | |
1485 | flags & BDRV_O_AUTO_RDONLY); | |
1486 | } | |
91a097e7 KW |
1487 | } |
1488 | ||
636ea370 KW |
1489 | static void bdrv_assign_node_name(BlockDriverState *bs, |
1490 | const char *node_name, | |
1491 | Error **errp) | |
6913c0c2 | 1492 | { |
15489c76 | 1493 | char *gen_node_name = NULL; |
6913c0c2 | 1494 | |
15489c76 JC |
1495 | if (!node_name) { |
1496 | node_name = gen_node_name = id_generate(ID_BLOCK); | |
1497 | } else if (!id_wellformed(node_name)) { | |
1498 | /* | |
1499 | * Check for empty string or invalid characters, but not if it is | |
1500 | * generated (generated names use characters not available to the user) | |
1501 | */ | |
785ec4b1 | 1502 | error_setg(errp, "Invalid node-name: '%s'", node_name); |
636ea370 | 1503 | return; |
6913c0c2 BC |
1504 | } |
1505 | ||
0c5e94ee | 1506 | /* takes care of avoiding namespaces collisions */ |
7f06d47e | 1507 | if (blk_by_name(node_name)) { |
0c5e94ee BC |
1508 | error_setg(errp, "node-name=%s is conflicting with a device id", |
1509 | node_name); | |
15489c76 | 1510 | goto out; |
0c5e94ee BC |
1511 | } |
1512 | ||
6913c0c2 BC |
1513 | /* takes care of avoiding duplicates node names */ |
1514 | if (bdrv_find_node(node_name)) { | |
785ec4b1 | 1515 | error_setg(errp, "Duplicate nodes with node-name='%s'", node_name); |
15489c76 | 1516 | goto out; |
6913c0c2 BC |
1517 | } |
1518 | ||
824808dd KW |
1519 | /* Make sure that the node name isn't truncated */ |
1520 | if (strlen(node_name) >= sizeof(bs->node_name)) { | |
1521 | error_setg(errp, "Node name too long"); | |
1522 | goto out; | |
1523 | } | |
1524 | ||
6913c0c2 BC |
1525 | /* copy node name into the bs and insert it into the graph list */ |
1526 | pstrcpy(bs->node_name, sizeof(bs->node_name), node_name); | |
1527 | QTAILQ_INSERT_TAIL(&graph_bdrv_states, bs, node_list); | |
15489c76 JC |
1528 | out: |
1529 | g_free(gen_node_name); | |
6913c0c2 BC |
1530 | } |
1531 | ||
01a56501 KW |
1532 | static int bdrv_open_driver(BlockDriverState *bs, BlockDriver *drv, |
1533 | const char *node_name, QDict *options, | |
1534 | int open_flags, Error **errp) | |
1535 | { | |
1536 | Error *local_err = NULL; | |
0f12264e | 1537 | int i, ret; |
01a56501 KW |
1538 | |
1539 | bdrv_assign_node_name(bs, node_name, &local_err); | |
1540 | if (local_err) { | |
1541 | error_propagate(errp, local_err); | |
1542 | return -EINVAL; | |
1543 | } | |
1544 | ||
1545 | bs->drv = drv; | |
1546 | bs->opaque = g_malloc0(drv->instance_size); | |
1547 | ||
1548 | if (drv->bdrv_file_open) { | |
1549 | assert(!drv->bdrv_needs_filename || bs->filename[0]); | |
1550 | ret = drv->bdrv_file_open(bs, options, open_flags, &local_err); | |
680c7f96 | 1551 | } else if (drv->bdrv_open) { |
01a56501 | 1552 | ret = drv->bdrv_open(bs, options, open_flags, &local_err); |
680c7f96 KW |
1553 | } else { |
1554 | ret = 0; | |
01a56501 KW |
1555 | } |
1556 | ||
1557 | if (ret < 0) { | |
1558 | if (local_err) { | |
1559 | error_propagate(errp, local_err); | |
1560 | } else if (bs->filename[0]) { | |
1561 | error_setg_errno(errp, -ret, "Could not open '%s'", bs->filename); | |
1562 | } else { | |
1563 | error_setg_errno(errp, -ret, "Could not open image"); | |
1564 | } | |
180ca19a | 1565 | goto open_failed; |
01a56501 KW |
1566 | } |
1567 | ||
1568 | ret = refresh_total_sectors(bs, bs->total_sectors); | |
1569 | if (ret < 0) { | |
1570 | error_setg_errno(errp, -ret, "Could not refresh total sector count"); | |
180ca19a | 1571 | return ret; |
01a56501 KW |
1572 | } |
1573 | ||
1e4c797c | 1574 | bdrv_refresh_limits(bs, NULL, &local_err); |
01a56501 KW |
1575 | if (local_err) { |
1576 | error_propagate(errp, local_err); | |
180ca19a | 1577 | return -EINVAL; |
01a56501 KW |
1578 | } |
1579 | ||
1580 | assert(bdrv_opt_mem_align(bs) != 0); | |
1581 | assert(bdrv_min_mem_align(bs) != 0); | |
1582 | assert(is_power_of_2(bs->bl.request_alignment)); | |
1583 | ||
0f12264e KW |
1584 | for (i = 0; i < bs->quiesce_counter; i++) { |
1585 | if (drv->bdrv_co_drain_begin) { | |
1586 | drv->bdrv_co_drain_begin(bs); | |
1587 | } | |
1588 | } | |
1589 | ||
01a56501 | 1590 | return 0; |
180ca19a MP |
1591 | open_failed: |
1592 | bs->drv = NULL; | |
1593 | if (bs->file != NULL) { | |
1594 | bdrv_unref_child(bs, bs->file); | |
1595 | bs->file = NULL; | |
1596 | } | |
01a56501 KW |
1597 | g_free(bs->opaque); |
1598 | bs->opaque = NULL; | |
01a56501 KW |
1599 | return ret; |
1600 | } | |
1601 | ||
680c7f96 KW |
1602 | BlockDriverState *bdrv_new_open_driver(BlockDriver *drv, const char *node_name, |
1603 | int flags, Error **errp) | |
1604 | { | |
1605 | BlockDriverState *bs; | |
1606 | int ret; | |
1607 | ||
1608 | bs = bdrv_new(); | |
1609 | bs->open_flags = flags; | |
1610 | bs->explicit_options = qdict_new(); | |
1611 | bs->options = qdict_new(); | |
1612 | bs->opaque = NULL; | |
1613 | ||
1614 | update_options_from_flags(bs->options, flags); | |
1615 | ||
1616 | ret = bdrv_open_driver(bs, drv, node_name, bs->options, flags, errp); | |
1617 | if (ret < 0) { | |
cb3e7f08 | 1618 | qobject_unref(bs->explicit_options); |
180ca19a | 1619 | bs->explicit_options = NULL; |
cb3e7f08 | 1620 | qobject_unref(bs->options); |
180ca19a | 1621 | bs->options = NULL; |
680c7f96 KW |
1622 | bdrv_unref(bs); |
1623 | return NULL; | |
1624 | } | |
1625 | ||
1626 | return bs; | |
1627 | } | |
1628 | ||
c5f3014b | 1629 | QemuOptsList bdrv_runtime_opts = { |
18edf289 KW |
1630 | .name = "bdrv_common", |
1631 | .head = QTAILQ_HEAD_INITIALIZER(bdrv_runtime_opts.head), | |
1632 | .desc = { | |
1633 | { | |
1634 | .name = "node-name", | |
1635 | .type = QEMU_OPT_STRING, | |
1636 | .help = "Node name of the block device node", | |
1637 | }, | |
62392ebb KW |
1638 | { |
1639 | .name = "driver", | |
1640 | .type = QEMU_OPT_STRING, | |
1641 | .help = "Block driver to use for the node", | |
1642 | }, | |
91a097e7 KW |
1643 | { |
1644 | .name = BDRV_OPT_CACHE_DIRECT, | |
1645 | .type = QEMU_OPT_BOOL, | |
1646 | .help = "Bypass software writeback cache on the host", | |
1647 | }, | |
1648 | { | |
1649 | .name = BDRV_OPT_CACHE_NO_FLUSH, | |
1650 | .type = QEMU_OPT_BOOL, | |
1651 | .help = "Ignore flush requests", | |
1652 | }, | |
f87a0e29 AG |
1653 | { |
1654 | .name = BDRV_OPT_READ_ONLY, | |
1655 | .type = QEMU_OPT_BOOL, | |
1656 | .help = "Node is opened in read-only mode", | |
1657 | }, | |
e35bdc12 KW |
1658 | { |
1659 | .name = BDRV_OPT_AUTO_READ_ONLY, | |
1660 | .type = QEMU_OPT_BOOL, | |
1661 | .help = "Node can become read-only if opening read-write fails", | |
1662 | }, | |
692e01a2 KW |
1663 | { |
1664 | .name = "detect-zeroes", | |
1665 | .type = QEMU_OPT_STRING, | |
1666 | .help = "try to optimize zero writes (off, on, unmap)", | |
1667 | }, | |
818584a4 | 1668 | { |
415bbca8 | 1669 | .name = BDRV_OPT_DISCARD, |
818584a4 KW |
1670 | .type = QEMU_OPT_STRING, |
1671 | .help = "discard operation (ignore/off, unmap/on)", | |
1672 | }, | |
5a9347c6 FZ |
1673 | { |
1674 | .name = BDRV_OPT_FORCE_SHARE, | |
1675 | .type = QEMU_OPT_BOOL, | |
1676 | .help = "always accept other writers (default: off)", | |
1677 | }, | |
18edf289 KW |
1678 | { /* end of list */ } |
1679 | }, | |
1680 | }; | |
1681 | ||
5a5e7f8c ML |
1682 | QemuOptsList bdrv_create_opts_simple = { |
1683 | .name = "simple-create-opts", | |
1684 | .head = QTAILQ_HEAD_INITIALIZER(bdrv_create_opts_simple.head), | |
fd17146c HR |
1685 | .desc = { |
1686 | { | |
1687 | .name = BLOCK_OPT_SIZE, | |
1688 | .type = QEMU_OPT_SIZE, | |
1689 | .help = "Virtual disk size" | |
1690 | }, | |
1691 | { | |
1692 | .name = BLOCK_OPT_PREALLOC, | |
1693 | .type = QEMU_OPT_STRING, | |
1694 | .help = "Preallocation mode (allowed values: off)" | |
1695 | }, | |
1696 | { /* end of list */ } | |
1697 | } | |
1698 | }; | |
1699 | ||
57915332 KW |
1700 | /* |
1701 | * Common part for opening disk images and files | |
b6ad491a KW |
1702 | * |
1703 | * Removes all processed options from *options. | |
57915332 | 1704 | */ |
5696c6e3 | 1705 | static int bdrv_open_common(BlockDriverState *bs, BlockBackend *file, |
82dc8b41 | 1706 | QDict *options, Error **errp) |
57915332 KW |
1707 | { |
1708 | int ret, open_flags; | |
035fccdf | 1709 | const char *filename; |
62392ebb | 1710 | const char *driver_name = NULL; |
6913c0c2 | 1711 | const char *node_name = NULL; |
818584a4 | 1712 | const char *discard; |
18edf289 | 1713 | QemuOpts *opts; |
62392ebb | 1714 | BlockDriver *drv; |
34b5d2c6 | 1715 | Error *local_err = NULL; |
307261b2 | 1716 | bool ro; |
57915332 | 1717 | |
6405875c | 1718 | assert(bs->file == NULL); |
707ff828 | 1719 | assert(options != NULL && bs->options != options); |
57915332 | 1720 | |
62392ebb | 1721 | opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort); |
af175e85 | 1722 | if (!qemu_opts_absorb_qdict(opts, options, errp)) { |
62392ebb KW |
1723 | ret = -EINVAL; |
1724 | goto fail_opts; | |
1725 | } | |
1726 | ||
9b7e8691 AG |
1727 | update_flags_from_options(&bs->open_flags, opts); |
1728 | ||
62392ebb KW |
1729 | driver_name = qemu_opt_get(opts, "driver"); |
1730 | drv = bdrv_find_format(driver_name); | |
1731 | assert(drv != NULL); | |
1732 | ||
5a9347c6 FZ |
1733 | bs->force_share = qemu_opt_get_bool(opts, BDRV_OPT_FORCE_SHARE, false); |
1734 | ||
1735 | if (bs->force_share && (bs->open_flags & BDRV_O_RDWR)) { | |
1736 | error_setg(errp, | |
1737 | BDRV_OPT_FORCE_SHARE | |
1738 | "=on can only be used with read-only images"); | |
1739 | ret = -EINVAL; | |
1740 | goto fail_opts; | |
1741 | } | |
1742 | ||
45673671 | 1743 | if (file != NULL) { |
f30c66ba | 1744 | bdrv_refresh_filename(blk_bs(file)); |
5696c6e3 | 1745 | filename = blk_bs(file)->filename; |
45673671 | 1746 | } else { |
129c7d1c MA |
1747 | /* |
1748 | * Caution: while qdict_get_try_str() is fine, getting | |
1749 | * non-string types would require more care. When @options | |
1750 | * come from -blockdev or blockdev_add, its members are typed | |
1751 | * according to the QAPI schema, but when they come from | |
1752 | * -drive, they're all QString. | |
1753 | */ | |
45673671 KW |
1754 | filename = qdict_get_try_str(options, "filename"); |
1755 | } | |
1756 | ||
4a008240 | 1757 | if (drv->bdrv_needs_filename && (!filename || !filename[0])) { |
765003db KW |
1758 | error_setg(errp, "The '%s' block driver requires a file name", |
1759 | drv->format_name); | |
18edf289 KW |
1760 | ret = -EINVAL; |
1761 | goto fail_opts; | |
6913c0c2 | 1762 | } |
6913c0c2 | 1763 | |
82dc8b41 KW |
1764 | trace_bdrv_open_common(bs, filename ?: "", bs->open_flags, |
1765 | drv->format_name); | |
62392ebb | 1766 | |
307261b2 VSO |
1767 | ro = bdrv_is_read_only(bs); |
1768 | ||
1769 | if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv, ro)) { | |
1770 | if (!ro && bdrv_is_whitelisted(drv, true)) { | |
8be25de6 KW |
1771 | ret = bdrv_apply_auto_read_only(bs, NULL, NULL); |
1772 | } else { | |
1773 | ret = -ENOTSUP; | |
1774 | } | |
1775 | if (ret < 0) { | |
1776 | error_setg(errp, | |
307261b2 | 1777 | !ro && bdrv_is_whitelisted(drv, true) |
8be25de6 KW |
1778 | ? "Driver '%s' can only be used for read-only devices" |
1779 | : "Driver '%s' is not whitelisted", | |
1780 | drv->format_name); | |
1781 | goto fail_opts; | |
1782 | } | |
b64ec4e4 | 1783 | } |
57915332 | 1784 | |
d3faa13e | 1785 | /* bdrv_new() and bdrv_close() make it so */ |
d73415a3 | 1786 | assert(qatomic_read(&bs->copy_on_read) == 0); |
d3faa13e | 1787 | |
82dc8b41 | 1788 | if (bs->open_flags & BDRV_O_COPY_ON_READ) { |
307261b2 | 1789 | if (!ro) { |
0ebd24e0 KW |
1790 | bdrv_enable_copy_on_read(bs); |
1791 | } else { | |
1792 | error_setg(errp, "Can't use copy-on-read on read-only device"); | |
18edf289 KW |
1793 | ret = -EINVAL; |
1794 | goto fail_opts; | |
0ebd24e0 | 1795 | } |
53fec9d3 SH |
1796 | } |
1797 | ||
415bbca8 | 1798 | discard = qemu_opt_get(opts, BDRV_OPT_DISCARD); |
818584a4 KW |
1799 | if (discard != NULL) { |
1800 | if (bdrv_parse_discard_flags(discard, &bs->open_flags) != 0) { | |
1801 | error_setg(errp, "Invalid discard option"); | |
1802 | ret = -EINVAL; | |
1803 | goto fail_opts; | |
1804 | } | |
1805 | } | |
1806 | ||
543770bd AG |
1807 | bs->detect_zeroes = |
1808 | bdrv_parse_detect_zeroes(opts, bs->open_flags, &local_err); | |
1809 | if (local_err) { | |
1810 | error_propagate(errp, local_err); | |
1811 | ret = -EINVAL; | |
1812 | goto fail_opts; | |
692e01a2 KW |
1813 | } |
1814 | ||
c2ad1b0c KW |
1815 | if (filename != NULL) { |
1816 | pstrcpy(bs->filename, sizeof(bs->filename), filename); | |
1817 | } else { | |
1818 | bs->filename[0] = '\0'; | |
1819 | } | |
91af7014 | 1820 | pstrcpy(bs->exact_filename, sizeof(bs->exact_filename), bs->filename); |
57915332 | 1821 | |
66f82cee | 1822 | /* Open the image, either directly or using a protocol */ |
82dc8b41 | 1823 | open_flags = bdrv_open_flags(bs, bs->open_flags); |
01a56501 | 1824 | node_name = qemu_opt_get(opts, "node-name"); |
57915332 | 1825 | |
01a56501 KW |
1826 | assert(!drv->bdrv_file_open || file == NULL); |
1827 | ret = bdrv_open_driver(bs, drv, node_name, options, open_flags, errp); | |
51762288 | 1828 | if (ret < 0) { |
01a56501 | 1829 | goto fail_opts; |
3baca891 KW |
1830 | } |
1831 | ||
18edf289 | 1832 | qemu_opts_del(opts); |
57915332 KW |
1833 | return 0; |
1834 | ||
18edf289 KW |
1835 | fail_opts: |
1836 | qemu_opts_del(opts); | |
57915332 KW |
1837 | return ret; |
1838 | } | |
1839 | ||
5e5c4f63 KW |
1840 | static QDict *parse_json_filename(const char *filename, Error **errp) |
1841 | { | |
1842 | QObject *options_obj; | |
1843 | QDict *options; | |
1844 | int ret; | |
1845 | ||
1846 | ret = strstart(filename, "json:", &filename); | |
1847 | assert(ret); | |
1848 | ||
5577fff7 | 1849 | options_obj = qobject_from_json(filename, errp); |
5e5c4f63 | 1850 | if (!options_obj) { |
5577fff7 | 1851 | error_prepend(errp, "Could not parse the JSON options: "); |
5e5c4f63 KW |
1852 | return NULL; |
1853 | } | |
1854 | ||
7dc847eb | 1855 | options = qobject_to(QDict, options_obj); |
ca6b6e1e | 1856 | if (!options) { |
cb3e7f08 | 1857 | qobject_unref(options_obj); |
5e5c4f63 KW |
1858 | error_setg(errp, "Invalid JSON object given"); |
1859 | return NULL; | |
1860 | } | |
1861 | ||
5e5c4f63 KW |
1862 | qdict_flatten(options); |
1863 | ||
1864 | return options; | |
1865 | } | |
1866 | ||
de3b53f0 KW |
1867 | static void parse_json_protocol(QDict *options, const char **pfilename, |
1868 | Error **errp) | |
1869 | { | |
1870 | QDict *json_options; | |
1871 | Error *local_err = NULL; | |
1872 | ||
1873 | /* Parse json: pseudo-protocol */ | |
1874 | if (!*pfilename || !g_str_has_prefix(*pfilename, "json:")) { | |
1875 | return; | |
1876 | } | |
1877 | ||
1878 | json_options = parse_json_filename(*pfilename, &local_err); | |
1879 | if (local_err) { | |
1880 | error_propagate(errp, local_err); | |
1881 | return; | |
1882 | } | |
1883 | ||
1884 | /* Options given in the filename have lower priority than options | |
1885 | * specified directly */ | |
1886 | qdict_join(options, json_options, false); | |
cb3e7f08 | 1887 | qobject_unref(json_options); |
de3b53f0 KW |
1888 | *pfilename = NULL; |
1889 | } | |
1890 | ||
b6ce07aa | 1891 | /* |
f54120ff KW |
1892 | * Fills in default options for opening images and converts the legacy |
1893 | * filename/flags pair to option QDict entries. | |
53a29513 HR |
1894 | * The BDRV_O_PROTOCOL flag in *flags will be set or cleared accordingly if a |
1895 | * block driver has been specified explicitly. | |
b6ce07aa | 1896 | */ |
de3b53f0 | 1897 | static int bdrv_fill_options(QDict **options, const char *filename, |
053e1578 | 1898 | int *flags, Error **errp) |
ea2384d3 | 1899 | { |
c2ad1b0c | 1900 | const char *drvname; |
53a29513 | 1901 | bool protocol = *flags & BDRV_O_PROTOCOL; |
e3fa4bfa | 1902 | bool parse_filename = false; |
053e1578 | 1903 | BlockDriver *drv = NULL; |
34b5d2c6 | 1904 | Error *local_err = NULL; |
83f64091 | 1905 | |
129c7d1c MA |
1906 | /* |
1907 | * Caution: while qdict_get_try_str() is fine, getting non-string | |
1908 | * types would require more care. When @options come from | |
1909 | * -blockdev or blockdev_add, its members are typed according to | |
1910 | * the QAPI schema, but when they come from -drive, they're all | |
1911 | * QString. | |
1912 | */ | |
53a29513 | 1913 | drvname = qdict_get_try_str(*options, "driver"); |
053e1578 HR |
1914 | if (drvname) { |
1915 | drv = bdrv_find_format(drvname); | |
1916 | if (!drv) { | |
1917 | error_setg(errp, "Unknown driver '%s'", drvname); | |
1918 | return -ENOENT; | |
1919 | } | |
1920 | /* If the user has explicitly specified the driver, this choice should | |
1921 | * override the BDRV_O_PROTOCOL flag */ | |
1922 | protocol = drv->bdrv_file_open; | |
53a29513 HR |
1923 | } |
1924 | ||
1925 | if (protocol) { | |
1926 | *flags |= BDRV_O_PROTOCOL; | |
1927 | } else { | |
1928 | *flags &= ~BDRV_O_PROTOCOL; | |
1929 | } | |
1930 | ||
91a097e7 KW |
1931 | /* Translate cache options from flags into options */ |
1932 | update_options_from_flags(*options, *flags); | |
1933 | ||
035fccdf | 1934 | /* Fetch the file name from the options QDict if necessary */ |
17b005f1 | 1935 | if (protocol && filename) { |
f54120ff | 1936 | if (!qdict_haskey(*options, "filename")) { |
46f5ac20 | 1937 | qdict_put_str(*options, "filename", filename); |
f54120ff KW |
1938 | parse_filename = true; |
1939 | } else { | |
1940 | error_setg(errp, "Can't specify 'file' and 'filename' options at " | |
1941 | "the same time"); | |
1942 | return -EINVAL; | |
1943 | } | |
035fccdf KW |
1944 | } |
1945 | ||
c2ad1b0c | 1946 | /* Find the right block driver */ |
129c7d1c | 1947 | /* See cautionary note on accessing @options above */ |
f54120ff | 1948 | filename = qdict_get_try_str(*options, "filename"); |
f54120ff | 1949 | |
053e1578 HR |
1950 | if (!drvname && protocol) { |
1951 | if (filename) { | |
1952 | drv = bdrv_find_protocol(filename, parse_filename, errp); | |
17b005f1 | 1953 | if (!drv) { |
053e1578 | 1954 | return -EINVAL; |
17b005f1 | 1955 | } |
053e1578 HR |
1956 | |
1957 | drvname = drv->format_name; | |
46f5ac20 | 1958 | qdict_put_str(*options, "driver", drvname); |
053e1578 HR |
1959 | } else { |
1960 | error_setg(errp, "Must specify either driver or file"); | |
1961 | return -EINVAL; | |
98289620 | 1962 | } |
c2ad1b0c KW |
1963 | } |
1964 | ||
17b005f1 | 1965 | assert(drv || !protocol); |
c2ad1b0c | 1966 | |
f54120ff | 1967 | /* Driver-specific filename parsing */ |
17b005f1 | 1968 | if (drv && drv->bdrv_parse_filename && parse_filename) { |
5acd9d81 | 1969 | drv->bdrv_parse_filename(filename, *options, &local_err); |
84d18f06 | 1970 | if (local_err) { |
34b5d2c6 | 1971 | error_propagate(errp, local_err); |
f54120ff | 1972 | return -EINVAL; |
6963a30d | 1973 | } |
cd5d031e HR |
1974 | |
1975 | if (!drv->bdrv_needs_filename) { | |
1976 | qdict_del(*options, "filename"); | |
cd5d031e | 1977 | } |
6963a30d KW |
1978 | } |
1979 | ||
f54120ff KW |
1980 | return 0; |
1981 | } | |
1982 | ||
148eb13c KW |
1983 | typedef struct BlockReopenQueueEntry { |
1984 | bool prepared; | |
69b736e7 | 1985 | bool perms_checked; |
148eb13c | 1986 | BDRVReopenState state; |
859443b0 | 1987 | QTAILQ_ENTRY(BlockReopenQueueEntry) entry; |
148eb13c KW |
1988 | } BlockReopenQueueEntry; |
1989 | ||
1990 | /* | |
1991 | * Return the flags that @bs will have after the reopens in @q have | |
1992 | * successfully completed. If @q is NULL (or @bs is not contained in @q), | |
1993 | * return the current flags. | |
1994 | */ | |
1995 | static int bdrv_reopen_get_flags(BlockReopenQueue *q, BlockDriverState *bs) | |
1996 | { | |
1997 | BlockReopenQueueEntry *entry; | |
1998 | ||
1999 | if (q != NULL) { | |
859443b0 | 2000 | QTAILQ_FOREACH(entry, q, entry) { |
148eb13c KW |
2001 | if (entry->state.bs == bs) { |
2002 | return entry->state.flags; | |
2003 | } | |
2004 | } | |
2005 | } | |
2006 | ||
2007 | return bs->open_flags; | |
2008 | } | |
2009 | ||
2010 | /* Returns whether the image file can be written to after the reopen queue @q | |
2011 | * has been successfully applied, or right now if @q is NULL. */ | |
cc022140 HR |
2012 | static bool bdrv_is_writable_after_reopen(BlockDriverState *bs, |
2013 | BlockReopenQueue *q) | |
148eb13c KW |
2014 | { |
2015 | int flags = bdrv_reopen_get_flags(q, bs); | |
2016 | ||
2017 | return (flags & (BDRV_O_RDWR | BDRV_O_INACTIVE)) == BDRV_O_RDWR; | |
2018 | } | |
2019 | ||
cc022140 HR |
2020 | /* |
2021 | * Return whether the BDS can be written to. This is not necessarily | |
2022 | * the same as !bdrv_is_read_only(bs), as inactivated images may not | |
2023 | * be written to but do not count as read-only images. | |
2024 | */ | |
2025 | bool bdrv_is_writable(BlockDriverState *bs) | |
2026 | { | |
2027 | return bdrv_is_writable_after_reopen(bs, NULL); | |
2028 | } | |
2029 | ||
3bf416ba VSO |
2030 | static char *bdrv_child_user_desc(BdrvChild *c) |
2031 | { | |
da261b69 | 2032 | return c->klass->get_parent_desc(c); |
3bf416ba VSO |
2033 | } |
2034 | ||
30ebb9aa VSO |
2035 | /* |
2036 | * Check that @a allows everything that @b needs. @a and @b must reference same | |
2037 | * child node. | |
2038 | */ | |
3bf416ba VSO |
2039 | static bool bdrv_a_allow_b(BdrvChild *a, BdrvChild *b, Error **errp) |
2040 | { | |
30ebb9aa VSO |
2041 | const char *child_bs_name; |
2042 | g_autofree char *a_user = NULL; | |
2043 | g_autofree char *b_user = NULL; | |
2044 | g_autofree char *perms = NULL; | |
2045 | ||
2046 | assert(a->bs); | |
2047 | assert(a->bs == b->bs); | |
3bf416ba VSO |
2048 | |
2049 | if ((b->perm & a->shared_perm) == b->perm) { | |
2050 | return true; | |
2051 | } | |
2052 | ||
30ebb9aa VSO |
2053 | child_bs_name = bdrv_get_node_name(b->bs); |
2054 | a_user = bdrv_child_user_desc(a); | |
2055 | b_user = bdrv_child_user_desc(b); | |
2056 | perms = bdrv_perm_names(b->perm & ~a->shared_perm); | |
2057 | ||
2058 | error_setg(errp, "Permission conflict on node '%s': permissions '%s' are " | |
2059 | "both required by %s (uses node '%s' as '%s' child) and " | |
2060 | "unshared by %s (uses node '%s' as '%s' child).", | |
2061 | child_bs_name, perms, | |
2062 | b_user, child_bs_name, b->name, | |
2063 | a_user, child_bs_name, a->name); | |
3bf416ba VSO |
2064 | |
2065 | return false; | |
2066 | } | |
2067 | ||
9397c14f | 2068 | static bool bdrv_parent_perms_conflict(BlockDriverState *bs, Error **errp) |
3bf416ba VSO |
2069 | { |
2070 | BdrvChild *a, *b; | |
2071 | ||
2072 | /* | |
2073 | * During the loop we'll look at each pair twice. That's correct because | |
2074 | * bdrv_a_allow_b() is asymmetric and we should check each pair in both | |
2075 | * directions. | |
2076 | */ | |
2077 | QLIST_FOREACH(a, &bs->parents, next_parent) { | |
2078 | QLIST_FOREACH(b, &bs->parents, next_parent) { | |
9397c14f | 2079 | if (a == b) { |
3bf416ba VSO |
2080 | continue; |
2081 | } | |
2082 | ||
2083 | if (!bdrv_a_allow_b(a, b, errp)) { | |
2084 | return true; | |
2085 | } | |
2086 | } | |
2087 | } | |
2088 | ||
2089 | return false; | |
2090 | } | |
2091 | ||
ffd1a5a2 | 2092 | static void bdrv_child_perm(BlockDriverState *bs, BlockDriverState *child_bs, |
e5d8a406 HR |
2093 | BdrvChild *c, BdrvChildRole role, |
2094 | BlockReopenQueue *reopen_queue, | |
ffd1a5a2 FZ |
2095 | uint64_t parent_perm, uint64_t parent_shared, |
2096 | uint64_t *nperm, uint64_t *nshared) | |
2097 | { | |
0b3ca76e | 2098 | assert(bs->drv && bs->drv->bdrv_child_perm); |
e5d8a406 | 2099 | bs->drv->bdrv_child_perm(bs, c, role, reopen_queue, |
0b3ca76e AG |
2100 | parent_perm, parent_shared, |
2101 | nperm, nshared); | |
e0995dc3 | 2102 | /* TODO Take force_share from reopen_queue */ |
ffd1a5a2 FZ |
2103 | if (child_bs && child_bs->force_share) { |
2104 | *nshared = BLK_PERM_ALL; | |
2105 | } | |
2106 | } | |
2107 | ||
bd57f8f7 VSO |
2108 | /* |
2109 | * Adds the whole subtree of @bs (including @bs itself) to the @list (except for | |
2110 | * nodes that are already in the @list, of course) so that final list is | |
2111 | * topologically sorted. Return the result (GSList @list object is updated, so | |
2112 | * don't use old reference after function call). | |
2113 | * | |
2114 | * On function start @list must be already topologically sorted and for any node | |
2115 | * in the @list the whole subtree of the node must be in the @list as well. The | |
2116 | * simplest way to satisfy this criteria: use only result of | |
2117 | * bdrv_topological_dfs() or NULL as @list parameter. | |
2118 | */ | |
2119 | static GSList *bdrv_topological_dfs(GSList *list, GHashTable *found, | |
2120 | BlockDriverState *bs) | |
2121 | { | |
2122 | BdrvChild *child; | |
2123 | g_autoptr(GHashTable) local_found = NULL; | |
2124 | ||
2125 | if (!found) { | |
2126 | assert(!list); | |
2127 | found = local_found = g_hash_table_new(NULL, NULL); | |
2128 | } | |
2129 | ||
2130 | if (g_hash_table_contains(found, bs)) { | |
2131 | return list; | |
2132 | } | |
2133 | g_hash_table_add(found, bs); | |
2134 | ||
2135 | QLIST_FOREACH(child, &bs->children, next) { | |
2136 | list = bdrv_topological_dfs(list, found, child->bs); | |
2137 | } | |
2138 | ||
2139 | return g_slist_prepend(list, bs); | |
2140 | } | |
2141 | ||
ecb776bd VSO |
2142 | typedef struct BdrvChildSetPermState { |
2143 | BdrvChild *child; | |
2144 | uint64_t old_perm; | |
2145 | uint64_t old_shared_perm; | |
2146 | } BdrvChildSetPermState; | |
b0defa83 VSO |
2147 | |
2148 | static void bdrv_child_set_perm_abort(void *opaque) | |
2149 | { | |
ecb776bd VSO |
2150 | BdrvChildSetPermState *s = opaque; |
2151 | ||
2152 | s->child->perm = s->old_perm; | |
2153 | s->child->shared_perm = s->old_shared_perm; | |
b0defa83 VSO |
2154 | } |
2155 | ||
2156 | static TransactionActionDrv bdrv_child_set_pem_drv = { | |
2157 | .abort = bdrv_child_set_perm_abort, | |
ecb776bd | 2158 | .clean = g_free, |
b0defa83 VSO |
2159 | }; |
2160 | ||
ecb776bd VSO |
2161 | static void bdrv_child_set_perm(BdrvChild *c, uint64_t perm, |
2162 | uint64_t shared, Transaction *tran) | |
b0defa83 | 2163 | { |
ecb776bd VSO |
2164 | BdrvChildSetPermState *s = g_new(BdrvChildSetPermState, 1); |
2165 | ||
2166 | *s = (BdrvChildSetPermState) { | |
2167 | .child = c, | |
2168 | .old_perm = c->perm, | |
2169 | .old_shared_perm = c->shared_perm, | |
2170 | }; | |
b0defa83 VSO |
2171 | |
2172 | c->perm = perm; | |
2173 | c->shared_perm = shared; | |
2174 | ||
ecb776bd | 2175 | tran_add(tran, &bdrv_child_set_pem_drv, s); |
b0defa83 VSO |
2176 | } |
2177 | ||
2513ef59 VSO |
2178 | static void bdrv_drv_set_perm_commit(void *opaque) |
2179 | { | |
2180 | BlockDriverState *bs = opaque; | |
2181 | uint64_t cumulative_perms, cumulative_shared_perms; | |
2182 | ||
2183 | if (bs->drv->bdrv_set_perm) { | |
2184 | bdrv_get_cumulative_perm(bs, &cumulative_perms, | |
2185 | &cumulative_shared_perms); | |
2186 | bs->drv->bdrv_set_perm(bs, cumulative_perms, cumulative_shared_perms); | |
2187 | } | |
2188 | } | |
2189 | ||
2190 | static void bdrv_drv_set_perm_abort(void *opaque) | |
2191 | { | |
2192 | BlockDriverState *bs = opaque; | |
2193 | ||
2194 | if (bs->drv->bdrv_abort_perm_update) { | |
2195 | bs->drv->bdrv_abort_perm_update(bs); | |
2196 | } | |
2197 | } | |
2198 | ||
2199 | TransactionActionDrv bdrv_drv_set_perm_drv = { | |
2200 | .abort = bdrv_drv_set_perm_abort, | |
2201 | .commit = bdrv_drv_set_perm_commit, | |
2202 | }; | |
2203 | ||
2204 | static int bdrv_drv_set_perm(BlockDriverState *bs, uint64_t perm, | |
2205 | uint64_t shared_perm, Transaction *tran, | |
2206 | Error **errp) | |
2207 | { | |
2208 | if (!bs->drv) { | |
2209 | return 0; | |
2210 | } | |
2211 | ||
2212 | if (bs->drv->bdrv_check_perm) { | |
2213 | int ret = bs->drv->bdrv_check_perm(bs, perm, shared_perm, errp); | |
2214 | if (ret < 0) { | |
2215 | return ret; | |
2216 | } | |
2217 | } | |
2218 | ||
2219 | if (tran) { | |
2220 | tran_add(tran, &bdrv_drv_set_perm_drv, bs); | |
2221 | } | |
2222 | ||
2223 | return 0; | |
2224 | } | |
2225 | ||
0978623e VSO |
2226 | typedef struct BdrvReplaceChildState { |
2227 | BdrvChild *child; | |
2228 | BlockDriverState *old_bs; | |
2229 | } BdrvReplaceChildState; | |
2230 | ||
2231 | static void bdrv_replace_child_commit(void *opaque) | |
2232 | { | |
2233 | BdrvReplaceChildState *s = opaque; | |
2234 | ||
2235 | bdrv_unref(s->old_bs); | |
2236 | } | |
2237 | ||
2238 | static void bdrv_replace_child_abort(void *opaque) | |
2239 | { | |
2240 | BdrvReplaceChildState *s = opaque; | |
2241 | BlockDriverState *new_bs = s->child->bs; | |
2242 | ||
2243 | /* old_bs reference is transparently moved from @s to @s->child */ | |
2244 | bdrv_replace_child_noperm(s->child, s->old_bs); | |
2245 | bdrv_unref(new_bs); | |
2246 | } | |
2247 | ||
2248 | static TransactionActionDrv bdrv_replace_child_drv = { | |
2249 | .commit = bdrv_replace_child_commit, | |
2250 | .abort = bdrv_replace_child_abort, | |
2251 | .clean = g_free, | |
2252 | }; | |
2253 | ||
2254 | /* | |
4bf021db | 2255 | * bdrv_replace_child_tran |
0978623e VSO |
2256 | * |
2257 | * Note: real unref of old_bs is done only on commit. | |
4bf021db VSO |
2258 | * |
2259 | * The function doesn't update permissions, caller is responsible for this. | |
0978623e | 2260 | */ |
4bf021db VSO |
2261 | static void bdrv_replace_child_tran(BdrvChild *child, BlockDriverState *new_bs, |
2262 | Transaction *tran) | |
0978623e VSO |
2263 | { |
2264 | BdrvReplaceChildState *s = g_new(BdrvReplaceChildState, 1); | |
2265 | *s = (BdrvReplaceChildState) { | |
2266 | .child = child, | |
2267 | .old_bs = child->bs, | |
2268 | }; | |
2269 | tran_add(tran, &bdrv_replace_child_drv, s); | |
2270 | ||
2271 | if (new_bs) { | |
2272 | bdrv_ref(new_bs); | |
2273 | } | |
2274 | bdrv_replace_child_noperm(child, new_bs); | |
2275 | /* old_bs reference is transparently moved from @child to @s */ | |
2276 | } | |
2277 | ||
33a610c3 | 2278 | /* |
c20555e1 VSO |
2279 | * Refresh permissions in @bs subtree. The function is intended to be called |
2280 | * after some graph modification that was done without permission update. | |
33a610c3 | 2281 | */ |
c20555e1 VSO |
2282 | static int bdrv_node_refresh_perm(BlockDriverState *bs, BlockReopenQueue *q, |
2283 | Transaction *tran, Error **errp) | |
33a610c3 KW |
2284 | { |
2285 | BlockDriver *drv = bs->drv; | |
2286 | BdrvChild *c; | |
2287 | int ret; | |
c20555e1 VSO |
2288 | uint64_t cumulative_perms, cumulative_shared_perms; |
2289 | ||
2290 | bdrv_get_cumulative_perm(bs, &cumulative_perms, &cumulative_shared_perms); | |
33a610c3 KW |
2291 | |
2292 | /* Write permissions never work with read-only images */ | |
2293 | if ((cumulative_perms & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)) && | |
cc022140 | 2294 | !bdrv_is_writable_after_reopen(bs, q)) |
33a610c3 | 2295 | { |
481e0eee HR |
2296 | if (!bdrv_is_writable_after_reopen(bs, NULL)) { |
2297 | error_setg(errp, "Block node is read-only"); | |
2298 | } else { | |
c20555e1 VSO |
2299 | error_setg(errp, "Read-only block node '%s' cannot support " |
2300 | "read-write users", bdrv_get_node_name(bs)); | |
481e0eee HR |
2301 | } |
2302 | ||
33a610c3 KW |
2303 | return -EPERM; |
2304 | } | |
2305 | ||
9c60a5d1 KW |
2306 | /* |
2307 | * Unaligned requests will automatically be aligned to bl.request_alignment | |
2308 | * and without RESIZE we can't extend requests to write to space beyond the | |
2309 | * end of the image, so it's required that the image size is aligned. | |
2310 | */ | |
2311 | if ((cumulative_perms & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)) && | |
2312 | !(cumulative_perms & BLK_PERM_RESIZE)) | |
2313 | { | |
2314 | if ((bs->total_sectors * BDRV_SECTOR_SIZE) % bs->bl.request_alignment) { | |
2315 | error_setg(errp, "Cannot get 'write' permission without 'resize': " | |
2316 | "Image size is not a multiple of request " | |
2317 | "alignment"); | |
2318 | return -EPERM; | |
2319 | } | |
2320 | } | |
2321 | ||
33a610c3 KW |
2322 | /* Check this node */ |
2323 | if (!drv) { | |
2324 | return 0; | |
2325 | } | |
2326 | ||
b1d2bbeb | 2327 | ret = bdrv_drv_set_perm(bs, cumulative_perms, cumulative_shared_perms, tran, |
2513ef59 VSO |
2328 | errp); |
2329 | if (ret < 0) { | |
2330 | return ret; | |
33a610c3 KW |
2331 | } |
2332 | ||
78e421c9 | 2333 | /* Drivers that never have children can omit .bdrv_child_perm() */ |
33a610c3 | 2334 | if (!drv->bdrv_child_perm) { |
78e421c9 | 2335 | assert(QLIST_EMPTY(&bs->children)); |
33a610c3 KW |
2336 | return 0; |
2337 | } | |
2338 | ||
2339 | /* Check all children */ | |
2340 | QLIST_FOREACH(c, &bs->children, next) { | |
2341 | uint64_t cur_perm, cur_shared; | |
9eab1544 | 2342 | |
e5d8a406 | 2343 | bdrv_child_perm(bs, c->bs, c, c->role, q, |
ffd1a5a2 FZ |
2344 | cumulative_perms, cumulative_shared_perms, |
2345 | &cur_perm, &cur_shared); | |
ecb776bd | 2346 | bdrv_child_set_perm(c, cur_perm, cur_shared, tran); |
bd57f8f7 VSO |
2347 | } |
2348 | ||
2349 | return 0; | |
2350 | } | |
3ef45e02 | 2351 | |
25409807 VSO |
2352 | static int bdrv_list_refresh_perms(GSList *list, BlockReopenQueue *q, |
2353 | Transaction *tran, Error **errp) | |
bd57f8f7 VSO |
2354 | { |
2355 | int ret; | |
b1d2bbeb | 2356 | BlockDriverState *bs; |
bd57f8f7 | 2357 | |
b1d2bbeb VSO |
2358 | for ( ; list; list = list->next) { |
2359 | bs = list->data; | |
2360 | ||
9397c14f | 2361 | if (bdrv_parent_perms_conflict(bs, errp)) { |
b1d2bbeb | 2362 | return -EINVAL; |
bd57f8f7 VSO |
2363 | } |
2364 | ||
c20555e1 | 2365 | ret = bdrv_node_refresh_perm(bs, q, tran, errp); |
33a610c3 KW |
2366 | if (ret < 0) { |
2367 | return ret; | |
2368 | } | |
2369 | } | |
2370 | ||
2371 | return 0; | |
2372 | } | |
2373 | ||
c7a0f2be KW |
2374 | void bdrv_get_cumulative_perm(BlockDriverState *bs, uint64_t *perm, |
2375 | uint64_t *shared_perm) | |
33a610c3 KW |
2376 | { |
2377 | BdrvChild *c; | |
2378 | uint64_t cumulative_perms = 0; | |
2379 | uint64_t cumulative_shared_perms = BLK_PERM_ALL; | |
2380 | ||
2381 | QLIST_FOREACH(c, &bs->parents, next_parent) { | |
2382 | cumulative_perms |= c->perm; | |
2383 | cumulative_shared_perms &= c->shared_perm; | |
2384 | } | |
2385 | ||
2386 | *perm = cumulative_perms; | |
2387 | *shared_perm = cumulative_shared_perms; | |
2388 | } | |
2389 | ||
5176196c | 2390 | char *bdrv_perm_names(uint64_t perm) |
d083319f KW |
2391 | { |
2392 | struct perm_name { | |
2393 | uint64_t perm; | |
2394 | const char *name; | |
2395 | } permissions[] = { | |
2396 | { BLK_PERM_CONSISTENT_READ, "consistent read" }, | |
2397 | { BLK_PERM_WRITE, "write" }, | |
2398 | { BLK_PERM_WRITE_UNCHANGED, "write unchanged" }, | |
2399 | { BLK_PERM_RESIZE, "resize" }, | |
2400 | { BLK_PERM_GRAPH_MOD, "change children" }, | |
2401 | { 0, NULL } | |
2402 | }; | |
2403 | ||
e2a7423a | 2404 | GString *result = g_string_sized_new(30); |
d083319f KW |
2405 | struct perm_name *p; |
2406 | ||
2407 | for (p = permissions; p->name; p++) { | |
2408 | if (perm & p->perm) { | |
e2a7423a AG |
2409 | if (result->len > 0) { |
2410 | g_string_append(result, ", "); | |
2411 | } | |
2412 | g_string_append(result, p->name); | |
d083319f KW |
2413 | } |
2414 | } | |
2415 | ||
e2a7423a | 2416 | return g_string_free(result, FALSE); |
d083319f KW |
2417 | } |
2418 | ||
33a610c3 | 2419 | |
071b474f | 2420 | static int bdrv_refresh_perms(BlockDriverState *bs, Error **errp) |
bb87e4d1 VSO |
2421 | { |
2422 | int ret; | |
b1d2bbeb VSO |
2423 | Transaction *tran = tran_new(); |
2424 | g_autoptr(GSList) list = bdrv_topological_dfs(NULL, NULL, bs); | |
bb87e4d1 | 2425 | |
b1d2bbeb VSO |
2426 | ret = bdrv_list_refresh_perms(list, NULL, tran, errp); |
2427 | tran_finalize(tran, ret); | |
bb87e4d1 | 2428 | |
b1d2bbeb | 2429 | return ret; |
bb87e4d1 VSO |
2430 | } |
2431 | ||
33a610c3 KW |
2432 | int bdrv_child_try_set_perm(BdrvChild *c, uint64_t perm, uint64_t shared, |
2433 | Error **errp) | |
2434 | { | |
1046779e | 2435 | Error *local_err = NULL; |
83928dc4 | 2436 | Transaction *tran = tran_new(); |
33a610c3 KW |
2437 | int ret; |
2438 | ||
ecb776bd | 2439 | bdrv_child_set_perm(c, perm, shared, tran); |
83928dc4 VSO |
2440 | |
2441 | ret = bdrv_refresh_perms(c->bs, &local_err); | |
2442 | ||
2443 | tran_finalize(tran, ret); | |
2444 | ||
33a610c3 | 2445 | if (ret < 0) { |
071b474f VSO |
2446 | if ((perm & ~c->perm) || (c->shared_perm & ~shared)) { |
2447 | /* tighten permissions */ | |
1046779e HR |
2448 | error_propagate(errp, local_err); |
2449 | } else { | |
2450 | /* | |
2451 | * Our caller may intend to only loosen restrictions and | |
2452 | * does not expect this function to fail. Errors are not | |
2453 | * fatal in such a case, so we can just hide them from our | |
2454 | * caller. | |
2455 | */ | |
2456 | error_free(local_err); | |
2457 | ret = 0; | |
2458 | } | |
33a610c3 KW |
2459 | } |
2460 | ||
83928dc4 | 2461 | return ret; |
d5e6f437 KW |
2462 | } |
2463 | ||
c1087f12 HR |
2464 | int bdrv_child_refresh_perms(BlockDriverState *bs, BdrvChild *c, Error **errp) |
2465 | { | |
2466 | uint64_t parent_perms, parent_shared; | |
2467 | uint64_t perms, shared; | |
2468 | ||
2469 | bdrv_get_cumulative_perm(bs, &parent_perms, &parent_shared); | |
e5d8a406 | 2470 | bdrv_child_perm(bs, c->bs, c, c->role, NULL, |
bf8e925e | 2471 | parent_perms, parent_shared, &perms, &shared); |
c1087f12 HR |
2472 | |
2473 | return bdrv_child_try_set_perm(c, perms, shared, errp); | |
2474 | } | |
2475 | ||
87278af1 HR |
2476 | /* |
2477 | * Default implementation for .bdrv_child_perm() for block filters: | |
2478 | * Forward CONSISTENT_READ, WRITE, WRITE_UNCHANGED, and RESIZE to the | |
2479 | * filtered child. | |
2480 | */ | |
2481 | static void bdrv_filter_default_perms(BlockDriverState *bs, BdrvChild *c, | |
87278af1 HR |
2482 | BdrvChildRole role, |
2483 | BlockReopenQueue *reopen_queue, | |
2484 | uint64_t perm, uint64_t shared, | |
2485 | uint64_t *nperm, uint64_t *nshared) | |
6a1b9ee1 | 2486 | { |
e444fa83 KW |
2487 | *nperm = perm & DEFAULT_PERM_PASSTHROUGH; |
2488 | *nshared = (shared & DEFAULT_PERM_PASSTHROUGH) | DEFAULT_PERM_UNCHANGED; | |
6a1b9ee1 KW |
2489 | } |
2490 | ||
70082db4 | 2491 | static void bdrv_default_perms_for_cow(BlockDriverState *bs, BdrvChild *c, |
70082db4 HR |
2492 | BdrvChildRole role, |
2493 | BlockReopenQueue *reopen_queue, | |
2494 | uint64_t perm, uint64_t shared, | |
2495 | uint64_t *nperm, uint64_t *nshared) | |
2496 | { | |
e5d8a406 | 2497 | assert(role & BDRV_CHILD_COW); |
70082db4 HR |
2498 | |
2499 | /* | |
2500 | * We want consistent read from backing files if the parent needs it. | |
2501 | * No other operations are performed on backing files. | |
2502 | */ | |
2503 | perm &= BLK_PERM_CONSISTENT_READ; | |
2504 | ||
2505 | /* | |
2506 | * If the parent can deal with changing data, we're okay with a | |
2507 | * writable and resizable backing file. | |
2508 | * TODO Require !(perm & BLK_PERM_CONSISTENT_READ), too? | |
2509 | */ | |
2510 | if (shared & BLK_PERM_WRITE) { | |
2511 | shared = BLK_PERM_WRITE | BLK_PERM_RESIZE; | |
2512 | } else { | |
2513 | shared = 0; | |
2514 | } | |
2515 | ||
2516 | shared |= BLK_PERM_CONSISTENT_READ | BLK_PERM_GRAPH_MOD | | |
2517 | BLK_PERM_WRITE_UNCHANGED; | |
2518 | ||
2519 | if (bs->open_flags & BDRV_O_INACTIVE) { | |
2520 | shared |= BLK_PERM_WRITE | BLK_PERM_RESIZE; | |
2521 | } | |
2522 | ||
2523 | *nperm = perm; | |
2524 | *nshared = shared; | |
2525 | } | |
2526 | ||
6f838a4b | 2527 | static void bdrv_default_perms_for_storage(BlockDriverState *bs, BdrvChild *c, |
6f838a4b HR |
2528 | BdrvChildRole role, |
2529 | BlockReopenQueue *reopen_queue, | |
2530 | uint64_t perm, uint64_t shared, | |
2531 | uint64_t *nperm, uint64_t *nshared) | |
2532 | { | |
2533 | int flags; | |
2534 | ||
e5d8a406 | 2535 | assert(role & (BDRV_CHILD_METADATA | BDRV_CHILD_DATA)); |
6f838a4b HR |
2536 | |
2537 | flags = bdrv_reopen_get_flags(reopen_queue, bs); | |
2538 | ||
2539 | /* | |
2540 | * Apart from the modifications below, the same permissions are | |
2541 | * forwarded and left alone as for filters | |
2542 | */ | |
e5d8a406 | 2543 | bdrv_filter_default_perms(bs, c, role, reopen_queue, |
6f838a4b HR |
2544 | perm, shared, &perm, &shared); |
2545 | ||
f889054f HR |
2546 | if (role & BDRV_CHILD_METADATA) { |
2547 | /* Format drivers may touch metadata even if the guest doesn't write */ | |
2548 | if (bdrv_is_writable_after_reopen(bs, reopen_queue)) { | |
2549 | perm |= BLK_PERM_WRITE | BLK_PERM_RESIZE; | |
2550 | } | |
2551 | ||
2552 | /* | |
2553 | * bs->file always needs to be consistent because of the | |
2554 | * metadata. We can never allow other users to resize or write | |
2555 | * to it. | |
2556 | */ | |
2557 | if (!(flags & BDRV_O_NO_IO)) { | |
2558 | perm |= BLK_PERM_CONSISTENT_READ; | |
2559 | } | |
2560 | shared &= ~(BLK_PERM_WRITE | BLK_PERM_RESIZE); | |
6f838a4b HR |
2561 | } |
2562 | ||
f889054f HR |
2563 | if (role & BDRV_CHILD_DATA) { |
2564 | /* | |
2565 | * Technically, everything in this block is a subset of the | |
2566 | * BDRV_CHILD_METADATA path taken above, and so this could | |
2567 | * be an "else if" branch. However, that is not obvious, and | |
2568 | * this function is not performance critical, therefore we let | |
2569 | * this be an independent "if". | |
2570 | */ | |
2571 | ||
2572 | /* | |
2573 | * We cannot allow other users to resize the file because the | |
2574 | * format driver might have some assumptions about the size | |
2575 | * (e.g. because it is stored in metadata, or because the file | |
2576 | * is split into fixed-size data files). | |
2577 | */ | |
2578 | shared &= ~BLK_PERM_RESIZE; | |
2579 | ||
2580 | /* | |
2581 | * WRITE_UNCHANGED often cannot be performed as such on the | |
2582 | * data file. For example, the qcow2 driver may still need to | |
2583 | * write copied clusters on copy-on-read. | |
2584 | */ | |
2585 | if (perm & BLK_PERM_WRITE_UNCHANGED) { | |
2586 | perm |= BLK_PERM_WRITE; | |
2587 | } | |
2588 | ||
2589 | /* | |
2590 | * If the data file is written to, the format driver may | |
2591 | * expect to be able to resize it by writing beyond the EOF. | |
2592 | */ | |
2593 | if (perm & BLK_PERM_WRITE) { | |
2594 | perm |= BLK_PERM_RESIZE; | |
2595 | } | |
6f838a4b | 2596 | } |
6f838a4b HR |
2597 | |
2598 | if (bs->open_flags & BDRV_O_INACTIVE) { | |
2599 | shared |= BLK_PERM_WRITE | BLK_PERM_RESIZE; | |
2600 | } | |
2601 | ||
2602 | *nperm = perm; | |
2603 | *nshared = shared; | |
2604 | } | |
2605 | ||
2519f549 | 2606 | void bdrv_default_perms(BlockDriverState *bs, BdrvChild *c, |
e5d8a406 | 2607 | BdrvChildRole role, BlockReopenQueue *reopen_queue, |
2519f549 HR |
2608 | uint64_t perm, uint64_t shared, |
2609 | uint64_t *nperm, uint64_t *nshared) | |
2610 | { | |
2519f549 HR |
2611 | if (role & BDRV_CHILD_FILTERED) { |
2612 | assert(!(role & (BDRV_CHILD_DATA | BDRV_CHILD_METADATA | | |
2613 | BDRV_CHILD_COW))); | |
e5d8a406 | 2614 | bdrv_filter_default_perms(bs, c, role, reopen_queue, |
2519f549 HR |
2615 | perm, shared, nperm, nshared); |
2616 | } else if (role & BDRV_CHILD_COW) { | |
2617 | assert(!(role & (BDRV_CHILD_DATA | BDRV_CHILD_METADATA))); | |
e5d8a406 | 2618 | bdrv_default_perms_for_cow(bs, c, role, reopen_queue, |
2519f549 HR |
2619 | perm, shared, nperm, nshared); |
2620 | } else if (role & (BDRV_CHILD_METADATA | BDRV_CHILD_DATA)) { | |
e5d8a406 | 2621 | bdrv_default_perms_for_storage(bs, c, role, reopen_queue, |
2519f549 HR |
2622 | perm, shared, nperm, nshared); |
2623 | } else { | |
2624 | g_assert_not_reached(); | |
2625 | } | |
2626 | } | |
2627 | ||
7b1d9c4d HR |
2628 | uint64_t bdrv_qapi_perm_to_blk_perm(BlockPermission qapi_perm) |
2629 | { | |
2630 | static const uint64_t permissions[] = { | |
2631 | [BLOCK_PERMISSION_CONSISTENT_READ] = BLK_PERM_CONSISTENT_READ, | |
2632 | [BLOCK_PERMISSION_WRITE] = BLK_PERM_WRITE, | |
2633 | [BLOCK_PERMISSION_WRITE_UNCHANGED] = BLK_PERM_WRITE_UNCHANGED, | |
2634 | [BLOCK_PERMISSION_RESIZE] = BLK_PERM_RESIZE, | |
2635 | [BLOCK_PERMISSION_GRAPH_MOD] = BLK_PERM_GRAPH_MOD, | |
2636 | }; | |
2637 | ||
2638 | QEMU_BUILD_BUG_ON(ARRAY_SIZE(permissions) != BLOCK_PERMISSION__MAX); | |
2639 | QEMU_BUILD_BUG_ON(1UL << ARRAY_SIZE(permissions) != BLK_PERM_ALL + 1); | |
2640 | ||
2641 | assert(qapi_perm < BLOCK_PERMISSION__MAX); | |
2642 | ||
2643 | return permissions[qapi_perm]; | |
2644 | } | |
2645 | ||
8ee03995 KW |
2646 | static void bdrv_replace_child_noperm(BdrvChild *child, |
2647 | BlockDriverState *new_bs) | |
e9740bc6 KW |
2648 | { |
2649 | BlockDriverState *old_bs = child->bs; | |
debc2927 HR |
2650 | int new_bs_quiesce_counter; |
2651 | int drain_saldo; | |
e9740bc6 | 2652 | |
2cad1ebe AG |
2653 | assert(!child->frozen); |
2654 | ||
bb2614e9 FZ |
2655 | if (old_bs && new_bs) { |
2656 | assert(bdrv_get_aio_context(old_bs) == bdrv_get_aio_context(new_bs)); | |
2657 | } | |
debc2927 HR |
2658 | |
2659 | new_bs_quiesce_counter = (new_bs ? new_bs->quiesce_counter : 0); | |
2660 | drain_saldo = new_bs_quiesce_counter - child->parent_quiesce_counter; | |
2661 | ||
2662 | /* | |
2663 | * If the new child node is drained but the old one was not, flush | |
2664 | * all outstanding requests to the old child node. | |
2665 | */ | |
bd86fb99 | 2666 | while (drain_saldo > 0 && child->klass->drained_begin) { |
debc2927 HR |
2667 | bdrv_parent_drained_begin_single(child, true); |
2668 | drain_saldo--; | |
2669 | } | |
2670 | ||
e9740bc6 | 2671 | if (old_bs) { |
d736f119 KW |
2672 | /* Detach first so that the recursive drain sections coming from @child |
2673 | * are already gone and we only end the drain sections that came from | |
2674 | * elsewhere. */ | |
bd86fb99 HR |
2675 | if (child->klass->detach) { |
2676 | child->klass->detach(child); | |
d736f119 | 2677 | } |
e9740bc6 KW |
2678 | QLIST_REMOVE(child, next_parent); |
2679 | } | |
36fe1331 KW |
2680 | |
2681 | child->bs = new_bs; | |
2682 | ||
e9740bc6 KW |
2683 | if (new_bs) { |
2684 | QLIST_INSERT_HEAD(&new_bs->parents, child, next_parent); | |
debc2927 HR |
2685 | |
2686 | /* | |
2687 | * Detaching the old node may have led to the new node's | |
2688 | * quiesce_counter having been decreased. Not a problem, we | |
2689 | * just need to recognize this here and then invoke | |
2690 | * drained_end appropriately more often. | |
2691 | */ | |
2692 | assert(new_bs->quiesce_counter <= new_bs_quiesce_counter); | |
2693 | drain_saldo += new_bs->quiesce_counter - new_bs_quiesce_counter; | |
33a610c3 | 2694 | |
d736f119 KW |
2695 | /* Attach only after starting new drained sections, so that recursive |
2696 | * drain sections coming from @child don't get an extra .drained_begin | |
2697 | * callback. */ | |
bd86fb99 HR |
2698 | if (child->klass->attach) { |
2699 | child->klass->attach(child); | |
8ee03995 KW |
2700 | } |
2701 | } | |
debc2927 HR |
2702 | |
2703 | /* | |
2704 | * If the old child node was drained but the new one is not, allow | |
2705 | * requests to come in only after the new node has been attached. | |
2706 | */ | |
bd86fb99 | 2707 | while (drain_saldo < 0 && child->klass->drained_end) { |
debc2927 HR |
2708 | bdrv_parent_drained_end_single(child); |
2709 | drain_saldo++; | |
2710 | } | |
8ee03995 | 2711 | } |
33a610c3 | 2712 | |
46541ee5 VSO |
2713 | static void bdrv_child_free(void *opaque) |
2714 | { | |
2715 | BdrvChild *c = opaque; | |
2716 | ||
2717 | g_free(c->name); | |
2718 | g_free(c); | |
2719 | } | |
2720 | ||
548a74c0 | 2721 | static void bdrv_remove_empty_child(BdrvChild *child) |
df581792 | 2722 | { |
548a74c0 VSO |
2723 | assert(!child->bs); |
2724 | QLIST_SAFE_REMOVE(child, next); | |
46541ee5 | 2725 | bdrv_child_free(child); |
548a74c0 | 2726 | } |
d5e6f437 | 2727 | |
548a74c0 VSO |
2728 | typedef struct BdrvAttachChildCommonState { |
2729 | BdrvChild **child; | |
2730 | AioContext *old_parent_ctx; | |
2731 | AioContext *old_child_ctx; | |
2732 | } BdrvAttachChildCommonState; | |
2733 | ||
2734 | static void bdrv_attach_child_common_abort(void *opaque) | |
2735 | { | |
2736 | BdrvAttachChildCommonState *s = opaque; | |
2737 | BdrvChild *child = *s->child; | |
2738 | BlockDriverState *bs = child->bs; | |
2739 | ||
2740 | bdrv_replace_child_noperm(child, NULL); | |
2741 | ||
2742 | if (bdrv_get_aio_context(bs) != s->old_child_ctx) { | |
2743 | bdrv_try_set_aio_context(bs, s->old_child_ctx, &error_abort); | |
2744 | } | |
2745 | ||
2746 | if (bdrv_child_get_parent_aio_context(child) != s->old_parent_ctx) { | |
2747 | GSList *ignore = g_slist_prepend(NULL, child); | |
2748 | ||
2749 | child->klass->can_set_aio_ctx(child, s->old_parent_ctx, &ignore, | |
2750 | &error_abort); | |
2751 | g_slist_free(ignore); | |
2752 | ignore = g_slist_prepend(NULL, child); | |
2753 | child->klass->set_aio_ctx(child, s->old_parent_ctx, &ignore); | |
2754 | ||
2755 | g_slist_free(ignore); | |
d5e6f437 KW |
2756 | } |
2757 | ||
548a74c0 VSO |
2758 | bdrv_unref(bs); |
2759 | bdrv_remove_empty_child(child); | |
2760 | *s->child = NULL; | |
2761 | } | |
2762 | ||
2763 | static TransactionActionDrv bdrv_attach_child_common_drv = { | |
2764 | .abort = bdrv_attach_child_common_abort, | |
2765 | .clean = g_free, | |
2766 | }; | |
2767 | ||
2768 | /* | |
2769 | * Common part of attaching bdrv child to bs or to blk or to job | |
f8d2ad78 VSO |
2770 | * |
2771 | * Resulting new child is returned through @child. | |
2772 | * At start *@child must be NULL. | |
2773 | * @child is saved to a new entry of @tran, so that *@child could be reverted to | |
2774 | * NULL on abort(). So referenced variable must live at least until transaction | |
2775 | * end. | |
7ec390d5 VSO |
2776 | * |
2777 | * Function doesn't update permissions, caller is responsible for this. | |
548a74c0 VSO |
2778 | */ |
2779 | static int bdrv_attach_child_common(BlockDriverState *child_bs, | |
2780 | const char *child_name, | |
2781 | const BdrvChildClass *child_class, | |
2782 | BdrvChildRole child_role, | |
2783 | uint64_t perm, uint64_t shared_perm, | |
2784 | void *opaque, BdrvChild **child, | |
2785 | Transaction *tran, Error **errp) | |
2786 | { | |
2787 | BdrvChild *new_child; | |
2788 | AioContext *parent_ctx; | |
2789 | AioContext *child_ctx = bdrv_get_aio_context(child_bs); | |
2790 | ||
2791 | assert(child); | |
2792 | assert(*child == NULL); | |
da261b69 | 2793 | assert(child_class->get_parent_desc); |
548a74c0 VSO |
2794 | |
2795 | new_child = g_new(BdrvChild, 1); | |
2796 | *new_child = (BdrvChild) { | |
d5e6f437 KW |
2797 | .bs = NULL, |
2798 | .name = g_strdup(child_name), | |
bd86fb99 | 2799 | .klass = child_class, |
258b7765 | 2800 | .role = child_role, |
d5e6f437 KW |
2801 | .perm = perm, |
2802 | .shared_perm = shared_perm, | |
2803 | .opaque = opaque, | |
df581792 KW |
2804 | }; |
2805 | ||
548a74c0 VSO |
2806 | /* |
2807 | * If the AioContexts don't match, first try to move the subtree of | |
132ada80 | 2808 | * child_bs into the AioContext of the new parent. If this doesn't work, |
548a74c0 VSO |
2809 | * try moving the parent into the AioContext of child_bs instead. |
2810 | */ | |
2811 | parent_ctx = bdrv_child_get_parent_aio_context(new_child); | |
2812 | if (child_ctx != parent_ctx) { | |
2813 | Error *local_err = NULL; | |
2814 | int ret = bdrv_try_set_aio_context(child_bs, parent_ctx, &local_err); | |
2815 | ||
bd86fb99 | 2816 | if (ret < 0 && child_class->can_set_aio_ctx) { |
548a74c0 VSO |
2817 | GSList *ignore = g_slist_prepend(NULL, new_child); |
2818 | if (child_class->can_set_aio_ctx(new_child, child_ctx, &ignore, | |
2819 | NULL)) | |
2820 | { | |
132ada80 KW |
2821 | error_free(local_err); |
2822 | ret = 0; | |
2823 | g_slist_free(ignore); | |
548a74c0 VSO |
2824 | ignore = g_slist_prepend(NULL, new_child); |
2825 | child_class->set_aio_ctx(new_child, child_ctx, &ignore); | |
132ada80 KW |
2826 | } |
2827 | g_slist_free(ignore); | |
2828 | } | |
548a74c0 | 2829 | |
132ada80 KW |
2830 | if (ret < 0) { |
2831 | error_propagate(errp, local_err); | |
548a74c0 VSO |
2832 | bdrv_remove_empty_child(new_child); |
2833 | return ret; | |
132ada80 KW |
2834 | } |
2835 | } | |
2836 | ||
548a74c0 VSO |
2837 | bdrv_ref(child_bs); |
2838 | bdrv_replace_child_noperm(new_child, child_bs); | |
2839 | ||
2840 | *child = new_child; | |
b4b059f6 | 2841 | |
548a74c0 VSO |
2842 | BdrvAttachChildCommonState *s = g_new(BdrvAttachChildCommonState, 1); |
2843 | *s = (BdrvAttachChildCommonState) { | |
2844 | .child = child, | |
2845 | .old_parent_ctx = parent_ctx, | |
2846 | .old_child_ctx = child_ctx, | |
2847 | }; | |
2848 | tran_add(tran, &bdrv_attach_child_common_drv, s); | |
2849 | ||
2850 | return 0; | |
2851 | } | |
2852 | ||
f8d2ad78 VSO |
2853 | /* |
2854 | * Variable referenced by @child must live at least until transaction end. | |
2855 | * (see bdrv_attach_child_common() doc for details) | |
7ec390d5 VSO |
2856 | * |
2857 | * Function doesn't update permissions, caller is responsible for this. | |
f8d2ad78 | 2858 | */ |
aa5a04c7 VSO |
2859 | static int bdrv_attach_child_noperm(BlockDriverState *parent_bs, |
2860 | BlockDriverState *child_bs, | |
2861 | const char *child_name, | |
2862 | const BdrvChildClass *child_class, | |
2863 | BdrvChildRole child_role, | |
2864 | BdrvChild **child, | |
2865 | Transaction *tran, | |
2866 | Error **errp) | |
2867 | { | |
2868 | int ret; | |
2869 | uint64_t perm, shared_perm; | |
2870 | ||
2871 | assert(parent_bs->drv); | |
2872 | ||
2873 | bdrv_get_cumulative_perm(parent_bs, &perm, &shared_perm); | |
2874 | bdrv_child_perm(parent_bs, child_bs, NULL, child_role, NULL, | |
2875 | perm, shared_perm, &perm, &shared_perm); | |
2876 | ||
2877 | ret = bdrv_attach_child_common(child_bs, child_name, child_class, | |
2878 | child_role, perm, shared_perm, parent_bs, | |
2879 | child, tran, errp); | |
2880 | if (ret < 0) { | |
2881 | return ret; | |
2882 | } | |
2883 | ||
2884 | QLIST_INSERT_HEAD(&parent_bs->children, *child, next); | |
2885 | /* | |
2886 | * child is removed in bdrv_attach_child_common_abort(), so don't care to | |
2887 | * abort this change separately. | |
2888 | */ | |
2889 | ||
2890 | return 0; | |
2891 | } | |
2892 | ||
548a74c0 VSO |
2893 | static void bdrv_detach_child(BdrvChild *child) |
2894 | { | |
4954aace VSO |
2895 | BlockDriverState *old_bs = child->bs; |
2896 | ||
2897 | bdrv_replace_child_noperm(child, NULL); | |
548a74c0 | 2898 | bdrv_remove_empty_child(child); |
4954aace VSO |
2899 | |
2900 | if (old_bs) { | |
2901 | /* | |
2902 | * Update permissions for old node. We're just taking a parent away, so | |
2903 | * we're loosening restrictions. Errors of permission update are not | |
2904 | * fatal in this case, ignore them. | |
2905 | */ | |
2906 | bdrv_refresh_perms(old_bs, NULL); | |
2907 | ||
2908 | /* | |
2909 | * When the parent requiring a non-default AioContext is removed, the | |
2910 | * node moves back to the main AioContext | |
2911 | */ | |
2912 | bdrv_try_set_aio_context(old_bs, qemu_get_aio_context(), NULL); | |
2913 | } | |
548a74c0 VSO |
2914 | } |
2915 | ||
2916 | /* | |
2917 | * This function steals the reference to child_bs from the caller. | |
2918 | * That reference is later dropped by bdrv_root_unref_child(). | |
2919 | * | |
2920 | * On failure NULL is returned, errp is set and the reference to | |
2921 | * child_bs is also dropped. | |
2922 | * | |
2923 | * The caller must hold the AioContext lock @child_bs, but not that of @ctx | |
2924 | * (unless @child_bs is already in @ctx). | |
2925 | */ | |
2926 | BdrvChild *bdrv_root_attach_child(BlockDriverState *child_bs, | |
2927 | const char *child_name, | |
2928 | const BdrvChildClass *child_class, | |
2929 | BdrvChildRole child_role, | |
2930 | uint64_t perm, uint64_t shared_perm, | |
2931 | void *opaque, Error **errp) | |
2932 | { | |
2933 | int ret; | |
2934 | BdrvChild *child = NULL; | |
2935 | Transaction *tran = tran_new(); | |
2936 | ||
2937 | ret = bdrv_attach_child_common(child_bs, child_name, child_class, | |
2938 | child_role, perm, shared_perm, opaque, | |
2939 | &child, tran, errp); | |
2940 | if (ret < 0) { | |
e878bb12 | 2941 | goto out; |
548a74c0 VSO |
2942 | } |
2943 | ||
2944 | ret = bdrv_refresh_perms(child_bs, errp); | |
548a74c0 | 2945 | |
e878bb12 KW |
2946 | out: |
2947 | tran_finalize(tran, ret); | |
f8d2ad78 VSO |
2948 | /* child is unset on failure by bdrv_attach_child_common_abort() */ |
2949 | assert((ret < 0) == !child); | |
2950 | ||
548a74c0 | 2951 | bdrv_unref(child_bs); |
b4b059f6 | 2952 | return child; |
df581792 KW |
2953 | } |
2954 | ||
b441dc71 AG |
2955 | /* |
2956 | * This function transfers the reference to child_bs from the caller | |
2957 | * to parent_bs. That reference is later dropped by parent_bs on | |
2958 | * bdrv_close() or if someone calls bdrv_unref_child(). | |
2959 | * | |
2960 | * On failure NULL is returned, errp is set and the reference to | |
2961 | * child_bs is also dropped. | |
132ada80 KW |
2962 | * |
2963 | * If @parent_bs and @child_bs are in different AioContexts, the caller must | |
2964 | * hold the AioContext lock for @child_bs, but not for @parent_bs. | |
b441dc71 | 2965 | */ |
98292c61 WC |
2966 | BdrvChild *bdrv_attach_child(BlockDriverState *parent_bs, |
2967 | BlockDriverState *child_bs, | |
2968 | const char *child_name, | |
bd86fb99 | 2969 | const BdrvChildClass *child_class, |
258b7765 | 2970 | BdrvChildRole child_role, |
8b2ff529 | 2971 | Error **errp) |
f21d96d0 | 2972 | { |
aa5a04c7 VSO |
2973 | int ret; |
2974 | BdrvChild *child = NULL; | |
2975 | Transaction *tran = tran_new(); | |
f68c598b | 2976 | |
aa5a04c7 VSO |
2977 | ret = bdrv_attach_child_noperm(parent_bs, child_bs, child_name, child_class, |
2978 | child_role, &child, tran, errp); | |
2979 | if (ret < 0) { | |
2980 | goto out; | |
2981 | } | |
d5e6f437 | 2982 | |
aa5a04c7 VSO |
2983 | ret = bdrv_refresh_perms(parent_bs, errp); |
2984 | if (ret < 0) { | |
2985 | goto out; | |
d5e6f437 KW |
2986 | } |
2987 | ||
aa5a04c7 VSO |
2988 | out: |
2989 | tran_finalize(tran, ret); | |
f8d2ad78 VSO |
2990 | /* child is unset on failure by bdrv_attach_child_common_abort() */ |
2991 | assert((ret < 0) == !child); | |
aa5a04c7 VSO |
2992 | |
2993 | bdrv_unref(child_bs); | |
2994 | ||
f21d96d0 KW |
2995 | return child; |
2996 | } | |
2997 | ||
7b99a266 | 2998 | /* Callers must ensure that child->frozen is false. */ |
f21d96d0 | 2999 | void bdrv_root_unref_child(BdrvChild *child) |
33a60407 | 3000 | { |
779020cb KW |
3001 | BlockDriverState *child_bs; |
3002 | ||
f21d96d0 KW |
3003 | child_bs = child->bs; |
3004 | bdrv_detach_child(child); | |
3005 | bdrv_unref(child_bs); | |
3006 | } | |
3007 | ||
332b3a17 VSO |
3008 | typedef struct BdrvSetInheritsFrom { |
3009 | BlockDriverState *bs; | |
3010 | BlockDriverState *old_inherits_from; | |
3011 | } BdrvSetInheritsFrom; | |
3012 | ||
3013 | static void bdrv_set_inherits_from_abort(void *opaque) | |
3014 | { | |
3015 | BdrvSetInheritsFrom *s = opaque; | |
3016 | ||
3017 | s->bs->inherits_from = s->old_inherits_from; | |
3018 | } | |
3019 | ||
3020 | static TransactionActionDrv bdrv_set_inherits_from_drv = { | |
3021 | .abort = bdrv_set_inherits_from_abort, | |
3022 | .clean = g_free, | |
3023 | }; | |
3024 | ||
3025 | /* @tran is allowed to be NULL. In this case no rollback is possible */ | |
3026 | static void bdrv_set_inherits_from(BlockDriverState *bs, | |
3027 | BlockDriverState *new_inherits_from, | |
3028 | Transaction *tran) | |
3029 | { | |
3030 | if (tran) { | |
3031 | BdrvSetInheritsFrom *s = g_new(BdrvSetInheritsFrom, 1); | |
3032 | ||
3033 | *s = (BdrvSetInheritsFrom) { | |
3034 | .bs = bs, | |
3035 | .old_inherits_from = bs->inherits_from, | |
3036 | }; | |
3037 | ||
3038 | tran_add(tran, &bdrv_set_inherits_from_drv, s); | |
3039 | } | |
3040 | ||
3041 | bs->inherits_from = new_inherits_from; | |
3042 | } | |
3043 | ||
3cf746b3 HR |
3044 | /** |
3045 | * Clear all inherits_from pointers from children and grandchildren of | |
3046 | * @root that point to @root, where necessary. | |
332b3a17 | 3047 | * @tran is allowed to be NULL. In this case no rollback is possible |
3cf746b3 | 3048 | */ |
332b3a17 VSO |
3049 | static void bdrv_unset_inherits_from(BlockDriverState *root, BdrvChild *child, |
3050 | Transaction *tran) | |
f21d96d0 | 3051 | { |
3cf746b3 | 3052 | BdrvChild *c; |
4e4bf5c4 | 3053 | |
3cf746b3 HR |
3054 | if (child->bs->inherits_from == root) { |
3055 | /* | |
3056 | * Remove inherits_from only when the last reference between root and | |
3057 | * child->bs goes away. | |
3058 | */ | |
3059 | QLIST_FOREACH(c, &root->children, next) { | |
4e4bf5c4 KW |
3060 | if (c != child && c->bs == child->bs) { |
3061 | break; | |
3062 | } | |
3063 | } | |
3064 | if (c == NULL) { | |
332b3a17 | 3065 | bdrv_set_inherits_from(child->bs, NULL, tran); |
4e4bf5c4 | 3066 | } |
33a60407 KW |
3067 | } |
3068 | ||
3cf746b3 | 3069 | QLIST_FOREACH(c, &child->bs->children, next) { |
332b3a17 | 3070 | bdrv_unset_inherits_from(root, c, tran); |
3cf746b3 HR |
3071 | } |
3072 | } | |
3073 | ||
7b99a266 | 3074 | /* Callers must ensure that child->frozen is false. */ |
3cf746b3 HR |
3075 | void bdrv_unref_child(BlockDriverState *parent, BdrvChild *child) |
3076 | { | |
3077 | if (child == NULL) { | |
3078 | return; | |
3079 | } | |
3080 | ||
332b3a17 | 3081 | bdrv_unset_inherits_from(parent, child, NULL); |
f21d96d0 | 3082 | bdrv_root_unref_child(child); |
33a60407 KW |
3083 | } |
3084 | ||
5c8cab48 KW |
3085 | |
3086 | static void bdrv_parent_cb_change_media(BlockDriverState *bs, bool load) | |
3087 | { | |
3088 | BdrvChild *c; | |
3089 | QLIST_FOREACH(c, &bs->parents, next_parent) { | |
bd86fb99 HR |
3090 | if (c->klass->change_media) { |
3091 | c->klass->change_media(c, load); | |
5c8cab48 KW |
3092 | } |
3093 | } | |
3094 | } | |
3095 | ||
0065c455 AG |
3096 | /* Return true if you can reach parent going through child->inherits_from |
3097 | * recursively. If parent or child are NULL, return false */ | |
3098 | static bool bdrv_inherits_from_recursive(BlockDriverState *child, | |
3099 | BlockDriverState *parent) | |
3100 | { | |
3101 | while (child && child != parent) { | |
3102 | child = child->inherits_from; | |
3103 | } | |
3104 | ||
3105 | return child != NULL; | |
3106 | } | |
3107 | ||
25191e5f HR |
3108 | /* |
3109 | * Return the BdrvChildRole for @bs's backing child. bs->backing is | |
3110 | * mostly used for COW backing children (role = COW), but also for | |
3111 | * filtered children (role = FILTERED | PRIMARY). | |
3112 | */ | |
3113 | static BdrvChildRole bdrv_backing_role(BlockDriverState *bs) | |
3114 | { | |
3115 | if (bs->drv && bs->drv->is_filter) { | |
3116 | return BDRV_CHILD_FILTERED | BDRV_CHILD_PRIMARY; | |
3117 | } else { | |
3118 | return BDRV_CHILD_COW; | |
3119 | } | |
3120 | } | |
3121 | ||
5db15a57 | 3122 | /* |
e9238278 VSO |
3123 | * Sets the bs->backing or bs->file link of a BDS. A new reference is created; |
3124 | * callers which don't need their own reference any more must call bdrv_unref(). | |
7ec390d5 VSO |
3125 | * |
3126 | * Function doesn't update permissions, caller is responsible for this. | |
5db15a57 | 3127 | */ |
e9238278 VSO |
3128 | static int bdrv_set_file_or_backing_noperm(BlockDriverState *parent_bs, |
3129 | BlockDriverState *child_bs, | |
3130 | bool is_backing, | |
3131 | Transaction *tran, Error **errp) | |
8d24cce1 | 3132 | { |
a1e708fc | 3133 | int ret = 0; |
e9238278 VSO |
3134 | bool update_inherits_from = |
3135 | bdrv_inherits_from_recursive(child_bs, parent_bs); | |
3136 | BdrvChild *child = is_backing ? parent_bs->backing : parent_bs->file; | |
3137 | BdrvChildRole role; | |
0065c455 | 3138 | |
e9238278 VSO |
3139 | if (!parent_bs->drv) { |
3140 | /* | |
3141 | * Node without drv is an object without a class :/. TODO: finally fix | |
3142 | * qcow2 driver to never clear bs->drv and implement format corruption | |
3143 | * handling in other way. | |
3144 | */ | |
3145 | error_setg(errp, "Node corrupted"); | |
3146 | return -EINVAL; | |
3147 | } | |
3148 | ||
3149 | if (child && child->frozen) { | |
3150 | error_setg(errp, "Cannot change frozen '%s' link from '%s' to '%s'", | |
3151 | child->name, parent_bs->node_name, child->bs->node_name); | |
a1e708fc | 3152 | return -EPERM; |
2cad1ebe AG |
3153 | } |
3154 | ||
25f78d9e VSO |
3155 | if (is_backing && !parent_bs->drv->is_filter && |
3156 | !parent_bs->drv->supports_backing) | |
3157 | { | |
3158 | error_setg(errp, "Driver '%s' of node '%s' does not support backing " | |
3159 | "files", parent_bs->drv->format_name, parent_bs->node_name); | |
3160 | return -EINVAL; | |
3161 | } | |
3162 | ||
e9238278 VSO |
3163 | if (parent_bs->drv->is_filter) { |
3164 | role = BDRV_CHILD_FILTERED | BDRV_CHILD_PRIMARY; | |
3165 | } else if (is_backing) { | |
3166 | role = BDRV_CHILD_COW; | |
3167 | } else { | |
3168 | /* | |
3169 | * We only can use same role as it is in existing child. We don't have | |
3170 | * infrastructure to determine role of file child in generic way | |
3171 | */ | |
3172 | if (!child) { | |
3173 | error_setg(errp, "Cannot set file child to format node without " | |
3174 | "file child"); | |
3175 | return -EINVAL; | |
3176 | } | |
3177 | role = child->role; | |
826b6ca0 FZ |
3178 | } |
3179 | ||
e9238278 VSO |
3180 | if (child) { |
3181 | bdrv_unset_inherits_from(parent_bs, child, tran); | |
3182 | bdrv_remove_file_or_backing_child(parent_bs, child, tran); | |
3183 | } | |
3184 | ||
3185 | if (!child_bs) { | |
8d24cce1 FZ |
3186 | goto out; |
3187 | } | |
12fa4af6 | 3188 | |
e9238278 VSO |
3189 | ret = bdrv_attach_child_noperm(parent_bs, child_bs, |
3190 | is_backing ? "backing" : "file", | |
3191 | &child_of_bds, role, | |
3192 | is_backing ? &parent_bs->backing : | |
3193 | &parent_bs->file, | |
3194 | tran, errp); | |
160333e1 VSO |
3195 | if (ret < 0) { |
3196 | return ret; | |
a1e708fc VSO |
3197 | } |
3198 | ||
160333e1 VSO |
3199 | |
3200 | /* | |
e9238278 | 3201 | * If inherits_from pointed recursively to bs then let's update it to |
160333e1 VSO |
3202 | * point directly to bs (else it will become NULL). |
3203 | */ | |
a1e708fc | 3204 | if (update_inherits_from) { |
e9238278 | 3205 | bdrv_set_inherits_from(child_bs, parent_bs, tran); |
0065c455 | 3206 | } |
826b6ca0 | 3207 | |
8d24cce1 | 3208 | out: |
e9238278 | 3209 | bdrv_refresh_limits(parent_bs, tran, NULL); |
160333e1 VSO |
3210 | |
3211 | return 0; | |
3212 | } | |
3213 | ||
e9238278 VSO |
3214 | static int bdrv_set_backing_noperm(BlockDriverState *bs, |
3215 | BlockDriverState *backing_hd, | |
3216 | Transaction *tran, Error **errp) | |
3217 | { | |
3218 | return bdrv_set_file_or_backing_noperm(bs, backing_hd, true, tran, errp); | |
3219 | } | |
3220 | ||
160333e1 VSO |
3221 | int bdrv_set_backing_hd(BlockDriverState *bs, BlockDriverState *backing_hd, |
3222 | Error **errp) | |
3223 | { | |
3224 | int ret; | |
3225 | Transaction *tran = tran_new(); | |
3226 | ||
3227 | ret = bdrv_set_backing_noperm(bs, backing_hd, tran, errp); | |
3228 | if (ret < 0) { | |
3229 | goto out; | |
3230 | } | |
3231 | ||
3232 | ret = bdrv_refresh_perms(bs, errp); | |
3233 | out: | |
3234 | tran_finalize(tran, ret); | |
a1e708fc VSO |
3235 | |
3236 | return ret; | |
8d24cce1 FZ |
3237 | } |
3238 | ||
31ca6d07 KW |
3239 | /* |
3240 | * Opens the backing file for a BlockDriverState if not yet open | |
3241 | * | |
d9b7b057 KW |
3242 | * bdref_key specifies the key for the image's BlockdevRef in the options QDict. |
3243 | * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict | |
3244 | * itself, all options starting with "${bdref_key}." are considered part of the | |
3245 | * BlockdevRef. | |
3246 | * | |
3247 | * TODO Can this be unified with bdrv_open_image()? | |
31ca6d07 | 3248 | */ |
d9b7b057 KW |
3249 | int bdrv_open_backing_file(BlockDriverState *bs, QDict *parent_options, |
3250 | const char *bdref_key, Error **errp) | |
9156df12 | 3251 | { |
6b6833c1 | 3252 | char *backing_filename = NULL; |
d9b7b057 KW |
3253 | char *bdref_key_dot; |
3254 | const char *reference = NULL; | |
317fc44e | 3255 | int ret = 0; |
998c2019 | 3256 | bool implicit_backing = false; |
8d24cce1 | 3257 | BlockDriverState *backing_hd; |
d9b7b057 KW |
3258 | QDict *options; |
3259 | QDict *tmp_parent_options = NULL; | |
34b5d2c6 | 3260 | Error *local_err = NULL; |
9156df12 | 3261 | |
760e0063 | 3262 | if (bs->backing != NULL) { |
1ba4b6a5 | 3263 | goto free_exit; |
9156df12 PB |
3264 | } |
3265 | ||
31ca6d07 | 3266 | /* NULL means an empty set of options */ |
d9b7b057 KW |
3267 | if (parent_options == NULL) { |
3268 | tmp_parent_options = qdict_new(); | |
3269 | parent_options = tmp_parent_options; | |
31ca6d07 KW |
3270 | } |
3271 | ||
9156df12 | 3272 | bs->open_flags &= ~BDRV_O_NO_BACKING; |
d9b7b057 KW |
3273 | |
3274 | bdref_key_dot = g_strdup_printf("%s.", bdref_key); | |
3275 | qdict_extract_subqdict(parent_options, &options, bdref_key_dot); | |
3276 | g_free(bdref_key_dot); | |
3277 | ||
129c7d1c MA |
3278 | /* |
3279 | * Caution: while qdict_get_try_str() is fine, getting non-string | |
3280 | * types would require more care. When @parent_options come from | |
3281 | * -blockdev or blockdev_add, its members are typed according to | |
3282 | * the QAPI schema, but when they come from -drive, they're all | |
3283 | * QString. | |
3284 | */ | |
d9b7b057 KW |
3285 | reference = qdict_get_try_str(parent_options, bdref_key); |
3286 | if (reference || qdict_haskey(options, "file.filename")) { | |
6b6833c1 | 3287 | /* keep backing_filename NULL */ |
1cb6f506 | 3288 | } else if (bs->backing_file[0] == '\0' && qdict_size(options) == 0) { |
cb3e7f08 | 3289 | qobject_unref(options); |
1ba4b6a5 | 3290 | goto free_exit; |
dbecebdd | 3291 | } else { |
998c2019 HR |
3292 | if (qdict_size(options) == 0) { |
3293 | /* If the user specifies options that do not modify the | |
3294 | * backing file's behavior, we might still consider it the | |
3295 | * implicit backing file. But it's easier this way, and | |
3296 | * just specifying some of the backing BDS's options is | |
3297 | * only possible with -drive anyway (otherwise the QAPI | |
3298 | * schema forces the user to specify everything). */ | |
3299 | implicit_backing = !strcmp(bs->auto_backing_file, bs->backing_file); | |
3300 | } | |
3301 | ||
6b6833c1 | 3302 | backing_filename = bdrv_get_full_backing_filename(bs, &local_err); |
9f07429e HR |
3303 | if (local_err) { |
3304 | ret = -EINVAL; | |
3305 | error_propagate(errp, local_err); | |
cb3e7f08 | 3306 | qobject_unref(options); |
9f07429e HR |
3307 | goto free_exit; |
3308 | } | |
9156df12 PB |
3309 | } |
3310 | ||
8ee79e70 KW |
3311 | if (!bs->drv || !bs->drv->supports_backing) { |
3312 | ret = -EINVAL; | |
3313 | error_setg(errp, "Driver doesn't support backing files"); | |
cb3e7f08 | 3314 | qobject_unref(options); |
8ee79e70 KW |
3315 | goto free_exit; |
3316 | } | |
3317 | ||
6bff597b PK |
3318 | if (!reference && |
3319 | bs->backing_format[0] != '\0' && !qdict_haskey(options, "driver")) { | |
46f5ac20 | 3320 | qdict_put_str(options, "driver", bs->backing_format); |
9156df12 PB |
3321 | } |
3322 | ||
6b6833c1 | 3323 | backing_hd = bdrv_open_inherit(backing_filename, reference, options, 0, bs, |
25191e5f | 3324 | &child_of_bds, bdrv_backing_role(bs), errp); |
5b363937 | 3325 | if (!backing_hd) { |
9156df12 | 3326 | bs->open_flags |= BDRV_O_NO_BACKING; |
e43bfd9c | 3327 | error_prepend(errp, "Could not open backing file: "); |
5b363937 | 3328 | ret = -EINVAL; |
1ba4b6a5 | 3329 | goto free_exit; |
9156df12 | 3330 | } |
df581792 | 3331 | |
998c2019 HR |
3332 | if (implicit_backing) { |
3333 | bdrv_refresh_filename(backing_hd); | |
3334 | pstrcpy(bs->auto_backing_file, sizeof(bs->auto_backing_file), | |
3335 | backing_hd->filename); | |
3336 | } | |
3337 | ||
5db15a57 KW |
3338 | /* Hook up the backing file link; drop our reference, bs owns the |
3339 | * backing_hd reference now */ | |
dc9c10a1 | 3340 | ret = bdrv_set_backing_hd(bs, backing_hd, errp); |
5db15a57 | 3341 | bdrv_unref(backing_hd); |
dc9c10a1 | 3342 | if (ret < 0) { |
12fa4af6 KW |
3343 | goto free_exit; |
3344 | } | |
d80ac658 | 3345 | |
d9b7b057 KW |
3346 | qdict_del(parent_options, bdref_key); |
3347 | ||
1ba4b6a5 BC |
3348 | free_exit: |
3349 | g_free(backing_filename); | |
cb3e7f08 | 3350 | qobject_unref(tmp_parent_options); |
1ba4b6a5 | 3351 | return ret; |
9156df12 PB |
3352 | } |
3353 | ||
2d6b86af KW |
3354 | static BlockDriverState * |
3355 | bdrv_open_child_bs(const char *filename, QDict *options, const char *bdref_key, | |
bd86fb99 | 3356 | BlockDriverState *parent, const BdrvChildClass *child_class, |
272c02ea | 3357 | BdrvChildRole child_role, bool allow_none, Error **errp) |
da557aac | 3358 | { |
2d6b86af | 3359 | BlockDriverState *bs = NULL; |
da557aac | 3360 | QDict *image_options; |
da557aac HR |
3361 | char *bdref_key_dot; |
3362 | const char *reference; | |
3363 | ||
bd86fb99 | 3364 | assert(child_class != NULL); |
f67503e5 | 3365 | |
da557aac HR |
3366 | bdref_key_dot = g_strdup_printf("%s.", bdref_key); |
3367 | qdict_extract_subqdict(options, &image_options, bdref_key_dot); | |
3368 | g_free(bdref_key_dot); | |
3369 | ||
129c7d1c MA |
3370 | /* |
3371 | * Caution: while qdict_get_try_str() is fine, getting non-string | |
3372 | * types would require more care. When @options come from | |
3373 | * -blockdev or blockdev_add, its members are typed according to | |
3374 | * the QAPI schema, but when they come from -drive, they're all | |
3375 | * QString. | |
3376 | */ | |
da557aac HR |
3377 | reference = qdict_get_try_str(options, bdref_key); |
3378 | if (!filename && !reference && !qdict_size(image_options)) { | |
b4b059f6 | 3379 | if (!allow_none) { |
da557aac HR |
3380 | error_setg(errp, "A block device must be specified for \"%s\"", |
3381 | bdref_key); | |
da557aac | 3382 | } |
cb3e7f08 | 3383 | qobject_unref(image_options); |
da557aac HR |
3384 | goto done; |
3385 | } | |
3386 | ||
5b363937 | 3387 | bs = bdrv_open_inherit(filename, reference, image_options, 0, |
272c02ea | 3388 | parent, child_class, child_role, errp); |
5b363937 | 3389 | if (!bs) { |
df581792 KW |
3390 | goto done; |
3391 | } | |
3392 | ||
da557aac HR |
3393 | done: |
3394 | qdict_del(options, bdref_key); | |
2d6b86af KW |
3395 | return bs; |
3396 | } | |
3397 | ||
3398 | /* | |
3399 | * Opens a disk image whose options are given as BlockdevRef in another block | |
3400 | * device's options. | |
3401 | * | |
3402 | * If allow_none is true, no image will be opened if filename is false and no | |
3403 | * BlockdevRef is given. NULL will be returned, but errp remains unset. | |
3404 | * | |
3405 | * bdrev_key specifies the key for the image's BlockdevRef in the options QDict. | |
3406 | * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict | |
3407 | * itself, all options starting with "${bdref_key}." are considered part of the | |
3408 | * BlockdevRef. | |
3409 | * | |
3410 | * The BlockdevRef will be removed from the options QDict. | |
3411 | */ | |
3412 | BdrvChild *bdrv_open_child(const char *filename, | |
3413 | QDict *options, const char *bdref_key, | |
3414 | BlockDriverState *parent, | |
bd86fb99 | 3415 | const BdrvChildClass *child_class, |
258b7765 | 3416 | BdrvChildRole child_role, |
2d6b86af KW |
3417 | bool allow_none, Error **errp) |
3418 | { | |
3419 | BlockDriverState *bs; | |
3420 | ||
bd86fb99 | 3421 | bs = bdrv_open_child_bs(filename, options, bdref_key, parent, child_class, |
272c02ea | 3422 | child_role, allow_none, errp); |
2d6b86af KW |
3423 | if (bs == NULL) { |
3424 | return NULL; | |
3425 | } | |
3426 | ||
258b7765 HR |
3427 | return bdrv_attach_child(parent, bs, bdref_key, child_class, child_role, |
3428 | errp); | |
b4b059f6 KW |
3429 | } |
3430 | ||
bd86fb99 HR |
3431 | /* |
3432 | * TODO Future callers may need to specify parent/child_class in order for | |
3433 | * option inheritance to work. Existing callers use it for the root node. | |
3434 | */ | |
e1d74bc6 KW |
3435 | BlockDriverState *bdrv_open_blockdev_ref(BlockdevRef *ref, Error **errp) |
3436 | { | |
3437 | BlockDriverState *bs = NULL; | |
e1d74bc6 KW |
3438 | QObject *obj = NULL; |
3439 | QDict *qdict = NULL; | |
3440 | const char *reference = NULL; | |
3441 | Visitor *v = NULL; | |
3442 | ||
3443 | if (ref->type == QTYPE_QSTRING) { | |
3444 | reference = ref->u.reference; | |
3445 | } else { | |
3446 | BlockdevOptions *options = &ref->u.definition; | |
3447 | assert(ref->type == QTYPE_QDICT); | |
3448 | ||
3449 | v = qobject_output_visitor_new(&obj); | |
1f584248 | 3450 | visit_type_BlockdevOptions(v, NULL, &options, &error_abort); |
e1d74bc6 KW |
3451 | visit_complete(v, &obj); |
3452 | ||
7dc847eb | 3453 | qdict = qobject_to(QDict, obj); |
e1d74bc6 KW |
3454 | qdict_flatten(qdict); |
3455 | ||
3456 | /* bdrv_open_inherit() defaults to the values in bdrv_flags (for | |
3457 | * compatibility with other callers) rather than what we want as the | |
3458 | * real defaults. Apply the defaults here instead. */ | |
3459 | qdict_set_default_str(qdict, BDRV_OPT_CACHE_DIRECT, "off"); | |
3460 | qdict_set_default_str(qdict, BDRV_OPT_CACHE_NO_FLUSH, "off"); | |
3461 | qdict_set_default_str(qdict, BDRV_OPT_READ_ONLY, "off"); | |
e35bdc12 KW |
3462 | qdict_set_default_str(qdict, BDRV_OPT_AUTO_READ_ONLY, "off"); |
3463 | ||
e1d74bc6 KW |
3464 | } |
3465 | ||
272c02ea | 3466 | bs = bdrv_open_inherit(NULL, reference, qdict, 0, NULL, NULL, 0, errp); |
e1d74bc6 | 3467 | obj = NULL; |
cb3e7f08 | 3468 | qobject_unref(obj); |
e1d74bc6 KW |
3469 | visit_free(v); |
3470 | return bs; | |
3471 | } | |
3472 | ||
66836189 HR |
3473 | static BlockDriverState *bdrv_append_temp_snapshot(BlockDriverState *bs, |
3474 | int flags, | |
3475 | QDict *snapshot_options, | |
3476 | Error **errp) | |
b998875d KW |
3477 | { |
3478 | /* TODO: extra byte is a hack to ensure MAX_PATH space on Windows. */ | |
1ba4b6a5 | 3479 | char *tmp_filename = g_malloc0(PATH_MAX + 1); |
b998875d | 3480 | int64_t total_size; |
83d0521a | 3481 | QemuOpts *opts = NULL; |
ff6ed714 | 3482 | BlockDriverState *bs_snapshot = NULL; |
b998875d KW |
3483 | int ret; |
3484 | ||
3485 | /* if snapshot, we create a temporary backing file and open it | |
3486 | instead of opening 'filename' directly */ | |
3487 | ||
3488 | /* Get the required size from the image */ | |
f187743a KW |
3489 | total_size = bdrv_getlength(bs); |
3490 | if (total_size < 0) { | |
3491 | error_setg_errno(errp, -total_size, "Could not get image size"); | |
1ba4b6a5 | 3492 | goto out; |
f187743a | 3493 | } |
b998875d KW |
3494 | |
3495 | /* Create the temporary image */ | |
1ba4b6a5 | 3496 | ret = get_tmp_filename(tmp_filename, PATH_MAX + 1); |
b998875d KW |
3497 | if (ret < 0) { |
3498 | error_setg_errno(errp, -ret, "Could not get temporary filename"); | |
1ba4b6a5 | 3499 | goto out; |
b998875d KW |
3500 | } |
3501 | ||
ef810437 | 3502 | opts = qemu_opts_create(bdrv_qcow2.create_opts, NULL, 0, |
c282e1fd | 3503 | &error_abort); |
39101f25 | 3504 | qemu_opt_set_number(opts, BLOCK_OPT_SIZE, total_size, &error_abort); |
e43bfd9c | 3505 | ret = bdrv_create(&bdrv_qcow2, tmp_filename, opts, errp); |
83d0521a | 3506 | qemu_opts_del(opts); |
b998875d | 3507 | if (ret < 0) { |
e43bfd9c MA |
3508 | error_prepend(errp, "Could not create temporary overlay '%s': ", |
3509 | tmp_filename); | |
1ba4b6a5 | 3510 | goto out; |
b998875d KW |
3511 | } |
3512 | ||
73176bee | 3513 | /* Prepare options QDict for the temporary file */ |
46f5ac20 EB |
3514 | qdict_put_str(snapshot_options, "file.driver", "file"); |
3515 | qdict_put_str(snapshot_options, "file.filename", tmp_filename); | |
3516 | qdict_put_str(snapshot_options, "driver", "qcow2"); | |
b998875d | 3517 | |
5b363937 | 3518 | bs_snapshot = bdrv_open(NULL, NULL, snapshot_options, flags, errp); |
73176bee | 3519 | snapshot_options = NULL; |
5b363937 | 3520 | if (!bs_snapshot) { |
1ba4b6a5 | 3521 | goto out; |
b998875d KW |
3522 | } |
3523 | ||
934aee14 VSO |
3524 | ret = bdrv_append(bs_snapshot, bs, errp); |
3525 | if (ret < 0) { | |
ff6ed714 | 3526 | bs_snapshot = NULL; |
b2c2832c KW |
3527 | goto out; |
3528 | } | |
1ba4b6a5 BC |
3529 | |
3530 | out: | |
cb3e7f08 | 3531 | qobject_unref(snapshot_options); |
1ba4b6a5 | 3532 | g_free(tmp_filename); |
ff6ed714 | 3533 | return bs_snapshot; |
b998875d KW |
3534 | } |
3535 | ||
b6ce07aa KW |
3536 | /* |
3537 | * Opens a disk image (raw, qcow2, vmdk, ...) | |
de9c0cec KW |
3538 | * |
3539 | * options is a QDict of options to pass to the block drivers, or NULL for an | |
3540 | * empty set of options. The reference to the QDict belongs to the block layer | |
3541 | * after the call (even on failure), so if the caller intends to reuse the | |
cb3e7f08 | 3542 | * dictionary, it needs to use qobject_ref() before calling bdrv_open. |
f67503e5 HR |
3543 | * |
3544 | * If *pbs is NULL, a new BDS will be created with a pointer to it stored there. | |
3545 | * If it is not NULL, the referenced BDS will be reused. | |
ddf5636d HR |
3546 | * |
3547 | * The reference parameter may be used to specify an existing block device which | |
3548 | * should be opened. If specified, neither options nor a filename may be given, | |
3549 | * nor can an existing BDS be reused (that is, *pbs has to be NULL). | |
b6ce07aa | 3550 | */ |
5b363937 HR |
3551 | static BlockDriverState *bdrv_open_inherit(const char *filename, |
3552 | const char *reference, | |
3553 | QDict *options, int flags, | |
3554 | BlockDriverState *parent, | |
bd86fb99 | 3555 | const BdrvChildClass *child_class, |
272c02ea | 3556 | BdrvChildRole child_role, |
5b363937 | 3557 | Error **errp) |
ea2384d3 | 3558 | { |
b6ce07aa | 3559 | int ret; |
5696c6e3 | 3560 | BlockBackend *file = NULL; |
9a4f4c31 | 3561 | BlockDriverState *bs; |
ce343771 | 3562 | BlockDriver *drv = NULL; |
2f624b80 | 3563 | BdrvChild *child; |
74fe54f2 | 3564 | const char *drvname; |
3e8c2e57 | 3565 | const char *backing; |
34b5d2c6 | 3566 | Error *local_err = NULL; |
73176bee | 3567 | QDict *snapshot_options = NULL; |
b1e6fc08 | 3568 | int snapshot_flags = 0; |
712e7874 | 3569 | |
bd86fb99 HR |
3570 | assert(!child_class || !flags); |
3571 | assert(!child_class == !parent); | |
f67503e5 | 3572 | |
ddf5636d HR |
3573 | if (reference) { |
3574 | bool options_non_empty = options ? qdict_size(options) : false; | |
cb3e7f08 | 3575 | qobject_unref(options); |
ddf5636d | 3576 | |
ddf5636d HR |
3577 | if (filename || options_non_empty) { |
3578 | error_setg(errp, "Cannot reference an existing block device with " | |
3579 | "additional options or a new filename"); | |
5b363937 | 3580 | return NULL; |
ddf5636d HR |
3581 | } |
3582 | ||
3583 | bs = bdrv_lookup_bs(reference, reference, errp); | |
3584 | if (!bs) { | |
5b363937 | 3585 | return NULL; |
ddf5636d | 3586 | } |
76b22320 | 3587 | |
ddf5636d | 3588 | bdrv_ref(bs); |
5b363937 | 3589 | return bs; |
ddf5636d HR |
3590 | } |
3591 | ||
5b363937 | 3592 | bs = bdrv_new(); |
f67503e5 | 3593 | |
de9c0cec KW |
3594 | /* NULL means an empty set of options */ |
3595 | if (options == NULL) { | |
3596 | options = qdict_new(); | |
3597 | } | |
3598 | ||
145f598e | 3599 | /* json: syntax counts as explicit options, as if in the QDict */ |
de3b53f0 KW |
3600 | parse_json_protocol(options, &filename, &local_err); |
3601 | if (local_err) { | |
de3b53f0 KW |
3602 | goto fail; |
3603 | } | |
3604 | ||
145f598e KW |
3605 | bs->explicit_options = qdict_clone_shallow(options); |
3606 | ||
bd86fb99 | 3607 | if (child_class) { |
3cdc69d3 HR |
3608 | bool parent_is_format; |
3609 | ||
3610 | if (parent->drv) { | |
3611 | parent_is_format = parent->drv->is_format; | |
3612 | } else { | |
3613 | /* | |
3614 | * parent->drv is not set yet because this node is opened for | |
3615 | * (potential) format probing. That means that @parent is going | |
3616 | * to be a format node. | |
3617 | */ | |
3618 | parent_is_format = true; | |
3619 | } | |
3620 | ||
bddcec37 | 3621 | bs->inherits_from = parent; |
3cdc69d3 HR |
3622 | child_class->inherit_options(child_role, parent_is_format, |
3623 | &flags, options, | |
bd86fb99 | 3624 | parent->open_flags, parent->options); |
f3930ed0 KW |
3625 | } |
3626 | ||
de3b53f0 | 3627 | ret = bdrv_fill_options(&options, filename, &flags, &local_err); |
dfde483e | 3628 | if (ret < 0) { |
462f5bcf KW |
3629 | goto fail; |
3630 | } | |
3631 | ||
129c7d1c MA |
3632 | /* |
3633 | * Set the BDRV_O_RDWR and BDRV_O_ALLOW_RDWR flags. | |
3634 | * Caution: getting a boolean member of @options requires care. | |
3635 | * When @options come from -blockdev or blockdev_add, members are | |
3636 | * typed according to the QAPI schema, but when they come from | |
3637 | * -drive, they're all QString. | |
3638 | */ | |
f87a0e29 AG |
3639 | if (g_strcmp0(qdict_get_try_str(options, BDRV_OPT_READ_ONLY), "on") && |
3640 | !qdict_get_try_bool(options, BDRV_OPT_READ_ONLY, false)) { | |
3641 | flags |= (BDRV_O_RDWR | BDRV_O_ALLOW_RDWR); | |
3642 | } else { | |
3643 | flags &= ~BDRV_O_RDWR; | |
14499ea5 AG |
3644 | } |
3645 | ||
3646 | if (flags & BDRV_O_SNAPSHOT) { | |
3647 | snapshot_options = qdict_new(); | |
3648 | bdrv_temp_snapshot_options(&snapshot_flags, snapshot_options, | |
3649 | flags, options); | |
f87a0e29 AG |
3650 | /* Let bdrv_backing_options() override "read-only" */ |
3651 | qdict_del(options, BDRV_OPT_READ_ONLY); | |
00ff7ffd HR |
3652 | bdrv_inherited_options(BDRV_CHILD_COW, true, |
3653 | &flags, options, flags, options); | |
14499ea5 AG |
3654 | } |
3655 | ||
62392ebb KW |
3656 | bs->open_flags = flags; |
3657 | bs->options = options; | |
3658 | options = qdict_clone_shallow(options); | |
3659 | ||
76c591b0 | 3660 | /* Find the right image format driver */ |
129c7d1c | 3661 | /* See cautionary note on accessing @options above */ |
76c591b0 KW |
3662 | drvname = qdict_get_try_str(options, "driver"); |
3663 | if (drvname) { | |
3664 | drv = bdrv_find_format(drvname); | |
76c591b0 KW |
3665 | if (!drv) { |
3666 | error_setg(errp, "Unknown driver: '%s'", drvname); | |
76c591b0 KW |
3667 | goto fail; |
3668 | } | |
3669 | } | |
3670 | ||
3671 | assert(drvname || !(flags & BDRV_O_PROTOCOL)); | |
76c591b0 | 3672 | |
129c7d1c | 3673 | /* See cautionary note on accessing @options above */ |
3e8c2e57 | 3674 | backing = qdict_get_try_str(options, "backing"); |
e59a0cf1 HR |
3675 | if (qobject_to(QNull, qdict_get(options, "backing")) != NULL || |
3676 | (backing && *backing == '\0')) | |
3677 | { | |
4f7be280 HR |
3678 | if (backing) { |
3679 | warn_report("Use of \"backing\": \"\" is deprecated; " | |
3680 | "use \"backing\": null instead"); | |
3681 | } | |
3e8c2e57 | 3682 | flags |= BDRV_O_NO_BACKING; |
ae0f57f0 KW |
3683 | qdict_del(bs->explicit_options, "backing"); |
3684 | qdict_del(bs->options, "backing"); | |
3e8c2e57 AG |
3685 | qdict_del(options, "backing"); |
3686 | } | |
3687 | ||
5696c6e3 | 3688 | /* Open image file without format layer. This BlockBackend is only used for |
4e4bf5c4 KW |
3689 | * probing, the block drivers will do their own bdrv_open_child() for the |
3690 | * same BDS, which is why we put the node name back into options. */ | |
f4788adc | 3691 | if ((flags & BDRV_O_PROTOCOL) == 0) { |
5696c6e3 KW |
3692 | BlockDriverState *file_bs; |
3693 | ||
3694 | file_bs = bdrv_open_child_bs(filename, options, "file", bs, | |
58944401 HR |
3695 | &child_of_bds, BDRV_CHILD_IMAGE, |
3696 | true, &local_err); | |
1fdd6933 | 3697 | if (local_err) { |
f4788adc KW |
3698 | goto fail; |
3699 | } | |
5696c6e3 | 3700 | if (file_bs != NULL) { |
dacaa162 KW |
3701 | /* Not requesting BLK_PERM_CONSISTENT_READ because we're only |
3702 | * looking at the header to guess the image format. This works even | |
3703 | * in cases where a guest would not see a consistent state. */ | |
d861ab3a | 3704 | file = blk_new(bdrv_get_aio_context(file_bs), 0, BLK_PERM_ALL); |
d7086422 | 3705 | blk_insert_bs(file, file_bs, &local_err); |
5696c6e3 | 3706 | bdrv_unref(file_bs); |
d7086422 KW |
3707 | if (local_err) { |
3708 | goto fail; | |
3709 | } | |
5696c6e3 | 3710 | |
46f5ac20 | 3711 | qdict_put_str(options, "file", bdrv_get_node_name(file_bs)); |
4e4bf5c4 | 3712 | } |
f500a6d3 KW |
3713 | } |
3714 | ||
76c591b0 | 3715 | /* Image format probing */ |
38f3ef57 | 3716 | bs->probed = !drv; |
76c591b0 | 3717 | if (!drv && file) { |
cf2ab8fc | 3718 | ret = find_image_format(file, filename, &drv, &local_err); |
17b005f1 | 3719 | if (ret < 0) { |
8bfea15d | 3720 | goto fail; |
2a05cbe4 | 3721 | } |
62392ebb KW |
3722 | /* |
3723 | * This option update would logically belong in bdrv_fill_options(), | |
3724 | * but we first need to open bs->file for the probing to work, while | |
3725 | * opening bs->file already requires the (mostly) final set of options | |
3726 | * so that cache mode etc. can be inherited. | |
3727 | * | |
3728 | * Adding the driver later is somewhat ugly, but it's not an option | |
3729 | * that would ever be inherited, so it's correct. We just need to make | |
3730 | * sure to update both bs->options (which has the full effective | |
3731 | * options for bs) and options (which has file.* already removed). | |
3732 | */ | |
46f5ac20 EB |
3733 | qdict_put_str(bs->options, "driver", drv->format_name); |
3734 | qdict_put_str(options, "driver", drv->format_name); | |
76c591b0 | 3735 | } else if (!drv) { |
17b005f1 | 3736 | error_setg(errp, "Must specify either driver or file"); |
8bfea15d | 3737 | goto fail; |
ea2384d3 | 3738 | } |
b6ce07aa | 3739 | |
53a29513 HR |
3740 | /* BDRV_O_PROTOCOL must be set iff a protocol BDS is about to be created */ |
3741 | assert(!!(flags & BDRV_O_PROTOCOL) == !!drv->bdrv_file_open); | |
3742 | /* file must be NULL if a protocol BDS is about to be created | |
3743 | * (the inverse results in an error message from bdrv_open_common()) */ | |
3744 | assert(!(flags & BDRV_O_PROTOCOL) || !file); | |
3745 | ||
b6ce07aa | 3746 | /* Open the image */ |
82dc8b41 | 3747 | ret = bdrv_open_common(bs, file, options, &local_err); |
b6ce07aa | 3748 | if (ret < 0) { |
8bfea15d | 3749 | goto fail; |
6987307c CH |
3750 | } |
3751 | ||
4e4bf5c4 | 3752 | if (file) { |
5696c6e3 | 3753 | blk_unref(file); |
f500a6d3 KW |
3754 | file = NULL; |
3755 | } | |
3756 | ||
b6ce07aa | 3757 | /* If there is a backing file, use it */ |
9156df12 | 3758 | if ((flags & BDRV_O_NO_BACKING) == 0) { |
d9b7b057 | 3759 | ret = bdrv_open_backing_file(bs, options, "backing", &local_err); |
b6ce07aa | 3760 | if (ret < 0) { |
b6ad491a | 3761 | goto close_and_fail; |
b6ce07aa | 3762 | } |
b6ce07aa KW |
3763 | } |
3764 | ||
50196d7a AG |
3765 | /* Remove all children options and references |
3766 | * from bs->options and bs->explicit_options */ | |
2f624b80 AG |
3767 | QLIST_FOREACH(child, &bs->children, next) { |
3768 | char *child_key_dot; | |
3769 | child_key_dot = g_strdup_printf("%s.", child->name); | |
3770 | qdict_extract_subqdict(bs->explicit_options, NULL, child_key_dot); | |
3771 | qdict_extract_subqdict(bs->options, NULL, child_key_dot); | |
50196d7a AG |
3772 | qdict_del(bs->explicit_options, child->name); |
3773 | qdict_del(bs->options, child->name); | |
2f624b80 AG |
3774 | g_free(child_key_dot); |
3775 | } | |
3776 | ||
b6ad491a | 3777 | /* Check if any unknown options were used */ |
7ad2757f | 3778 | if (qdict_size(options) != 0) { |
b6ad491a | 3779 | const QDictEntry *entry = qdict_first(options); |
5acd9d81 HR |
3780 | if (flags & BDRV_O_PROTOCOL) { |
3781 | error_setg(errp, "Block protocol '%s' doesn't support the option " | |
3782 | "'%s'", drv->format_name, entry->key); | |
3783 | } else { | |
d0e46a55 HR |
3784 | error_setg(errp, |
3785 | "Block format '%s' does not support the option '%s'", | |
3786 | drv->format_name, entry->key); | |
5acd9d81 | 3787 | } |
b6ad491a | 3788 | |
b6ad491a KW |
3789 | goto close_and_fail; |
3790 | } | |
b6ad491a | 3791 | |
c01c214b | 3792 | bdrv_parent_cb_change_media(bs, true); |
b6ce07aa | 3793 | |
cb3e7f08 | 3794 | qobject_unref(options); |
8961be33 | 3795 | options = NULL; |
dd62f1ca KW |
3796 | |
3797 | /* For snapshot=on, create a temporary qcow2 overlay. bs points to the | |
3798 | * temporary snapshot afterwards. */ | |
3799 | if (snapshot_flags) { | |
66836189 HR |
3800 | BlockDriverState *snapshot_bs; |
3801 | snapshot_bs = bdrv_append_temp_snapshot(bs, snapshot_flags, | |
3802 | snapshot_options, &local_err); | |
73176bee | 3803 | snapshot_options = NULL; |
dd62f1ca KW |
3804 | if (local_err) { |
3805 | goto close_and_fail; | |
3806 | } | |
5b363937 HR |
3807 | /* We are not going to return bs but the overlay on top of it |
3808 | * (snapshot_bs); thus, we have to drop the strong reference to bs | |
3809 | * (which we obtained by calling bdrv_new()). bs will not be deleted, | |
3810 | * though, because the overlay still has a reference to it. */ | |
3811 | bdrv_unref(bs); | |
3812 | bs = snapshot_bs; | |
dd62f1ca KW |
3813 | } |
3814 | ||
5b363937 | 3815 | return bs; |
b6ce07aa | 3816 | |
8bfea15d | 3817 | fail: |
5696c6e3 | 3818 | blk_unref(file); |
cb3e7f08 MAL |
3819 | qobject_unref(snapshot_options); |
3820 | qobject_unref(bs->explicit_options); | |
3821 | qobject_unref(bs->options); | |
3822 | qobject_unref(options); | |
de9c0cec | 3823 | bs->options = NULL; |
998cbd6a | 3824 | bs->explicit_options = NULL; |
5b363937 | 3825 | bdrv_unref(bs); |
621ff94d | 3826 | error_propagate(errp, local_err); |
5b363937 | 3827 | return NULL; |
de9c0cec | 3828 | |
b6ad491a | 3829 | close_and_fail: |
5b363937 | 3830 | bdrv_unref(bs); |
cb3e7f08 MAL |
3831 | qobject_unref(snapshot_options); |
3832 | qobject_unref(options); | |
621ff94d | 3833 | error_propagate(errp, local_err); |
5b363937 | 3834 | return NULL; |
b6ce07aa KW |
3835 | } |
3836 | ||
5b363937 HR |
3837 | BlockDriverState *bdrv_open(const char *filename, const char *reference, |
3838 | QDict *options, int flags, Error **errp) | |
f3930ed0 | 3839 | { |
5b363937 | 3840 | return bdrv_open_inherit(filename, reference, options, flags, NULL, |
272c02ea | 3841 | NULL, 0, errp); |
f3930ed0 KW |
3842 | } |
3843 | ||
faf116b4 AG |
3844 | /* Return true if the NULL-terminated @list contains @str */ |
3845 | static bool is_str_in_list(const char *str, const char *const *list) | |
3846 | { | |
3847 | if (str && list) { | |
3848 | int i; | |
3849 | for (i = 0; list[i] != NULL; i++) { | |
3850 | if (!strcmp(str, list[i])) { | |
3851 | return true; | |
3852 | } | |
3853 | } | |
3854 | } | |
3855 | return false; | |
3856 | } | |
3857 | ||
3858 | /* | |
3859 | * Check that every option set in @bs->options is also set in | |
3860 | * @new_opts. | |
3861 | * | |
3862 | * Options listed in the common_options list and in | |
3863 | * @bs->drv->mutable_opts are skipped. | |
3864 | * | |
3865 | * Return 0 on success, otherwise return -EINVAL and set @errp. | |
3866 | */ | |
3867 | static int bdrv_reset_options_allowed(BlockDriverState *bs, | |
3868 | const QDict *new_opts, Error **errp) | |
3869 | { | |
3870 | const QDictEntry *e; | |
3871 | /* These options are common to all block drivers and are handled | |
3872 | * in bdrv_reopen_prepare() so they can be left out of @new_opts */ | |
3873 | const char *const common_options[] = { | |
3874 | "node-name", "discard", "cache.direct", "cache.no-flush", | |
3875 | "read-only", "auto-read-only", "detect-zeroes", NULL | |
3876 | }; | |
3877 | ||
3878 | for (e = qdict_first(bs->options); e; e = qdict_next(bs->options, e)) { | |
3879 | if (!qdict_haskey(new_opts, e->key) && | |
3880 | !is_str_in_list(e->key, common_options) && | |
3881 | !is_str_in_list(e->key, bs->drv->mutable_opts)) { | |
3882 | error_setg(errp, "Option '%s' cannot be reset " | |
3883 | "to its default value", e->key); | |
3884 | return -EINVAL; | |
3885 | } | |
3886 | } | |
3887 | ||
3888 | return 0; | |
3889 | } | |
3890 | ||
cb828c31 AG |
3891 | /* |
3892 | * Returns true if @child can be reached recursively from @bs | |
3893 | */ | |
3894 | static bool bdrv_recurse_has_child(BlockDriverState *bs, | |
3895 | BlockDriverState *child) | |
3896 | { | |
3897 | BdrvChild *c; | |
3898 | ||
3899 | if (bs == child) { | |
3900 | return true; | |
3901 | } | |
3902 | ||
3903 | QLIST_FOREACH(c, &bs->children, next) { | |
3904 | if (bdrv_recurse_has_child(c->bs, child)) { | |
3905 | return true; | |
3906 | } | |
3907 | } | |
3908 | ||
3909 | return false; | |
3910 | } | |
3911 | ||
e971aa12 JC |
3912 | /* |
3913 | * Adds a BlockDriverState to a simple queue for an atomic, transactional | |
3914 | * reopen of multiple devices. | |
3915 | * | |
859443b0 | 3916 | * bs_queue can either be an existing BlockReopenQueue that has had QTAILQ_INIT |
e971aa12 JC |
3917 | * already performed, or alternatively may be NULL a new BlockReopenQueue will |
3918 | * be created and initialized. This newly created BlockReopenQueue should be | |
3919 | * passed back in for subsequent calls that are intended to be of the same | |
3920 | * atomic 'set'. | |
3921 | * | |
3922 | * bs is the BlockDriverState to add to the reopen queue. | |
3923 | * | |
4d2cb092 KW |
3924 | * options contains the changed options for the associated bs |
3925 | * (the BlockReopenQueue takes ownership) | |
3926 | * | |
e971aa12 JC |
3927 | * flags contains the open flags for the associated bs |
3928 | * | |
3929 | * returns a pointer to bs_queue, which is either the newly allocated | |
3930 | * bs_queue, or the existing bs_queue being used. | |
3931 | * | |
1a63a907 | 3932 | * bs must be drained between bdrv_reopen_queue() and bdrv_reopen_multiple(). |
e971aa12 | 3933 | */ |
28518102 KW |
3934 | static BlockReopenQueue *bdrv_reopen_queue_child(BlockReopenQueue *bs_queue, |
3935 | BlockDriverState *bs, | |
3936 | QDict *options, | |
bd86fb99 | 3937 | const BdrvChildClass *klass, |
272c02ea | 3938 | BdrvChildRole role, |
3cdc69d3 | 3939 | bool parent_is_format, |
28518102 | 3940 | QDict *parent_options, |
077e8e20 AG |
3941 | int parent_flags, |
3942 | bool keep_old_opts) | |
e971aa12 JC |
3943 | { |
3944 | assert(bs != NULL); | |
3945 | ||
3946 | BlockReopenQueueEntry *bs_entry; | |
67251a31 | 3947 | BdrvChild *child; |
9aa09ddd AG |
3948 | QDict *old_options, *explicit_options, *options_copy; |
3949 | int flags; | |
3950 | QemuOpts *opts; | |
67251a31 | 3951 | |
1a63a907 KW |
3952 | /* Make sure that the caller remembered to use a drained section. This is |
3953 | * important to avoid graph changes between the recursive queuing here and | |
3954 | * bdrv_reopen_multiple(). */ | |
3955 | assert(bs->quiesce_counter > 0); | |
3956 | ||
e971aa12 JC |
3957 | if (bs_queue == NULL) { |
3958 | bs_queue = g_new0(BlockReopenQueue, 1); | |
859443b0 | 3959 | QTAILQ_INIT(bs_queue); |
e971aa12 JC |
3960 | } |
3961 | ||
4d2cb092 KW |
3962 | if (!options) { |
3963 | options = qdict_new(); | |
3964 | } | |
3965 | ||
5b7ba05f | 3966 | /* Check if this BlockDriverState is already in the queue */ |
859443b0 | 3967 | QTAILQ_FOREACH(bs_entry, bs_queue, entry) { |
5b7ba05f AG |
3968 | if (bs == bs_entry->state.bs) { |
3969 | break; | |
3970 | } | |
3971 | } | |
3972 | ||
28518102 KW |
3973 | /* |
3974 | * Precedence of options: | |
3975 | * 1. Explicitly passed in options (highest) | |
9aa09ddd AG |
3976 | * 2. Retained from explicitly set options of bs |
3977 | * 3. Inherited from parent node | |
3978 | * 4. Retained from effective options of bs | |
28518102 KW |
3979 | */ |
3980 | ||
145f598e | 3981 | /* Old explicitly set values (don't overwrite by inherited value) */ |
077e8e20 AG |
3982 | if (bs_entry || keep_old_opts) { |
3983 | old_options = qdict_clone_shallow(bs_entry ? | |
3984 | bs_entry->state.explicit_options : | |
3985 | bs->explicit_options); | |
3986 | bdrv_join_options(bs, options, old_options); | |
3987 | qobject_unref(old_options); | |
5b7ba05f | 3988 | } |
145f598e KW |
3989 | |
3990 | explicit_options = qdict_clone_shallow(options); | |
3991 | ||
28518102 KW |
3992 | /* Inherit from parent node */ |
3993 | if (parent_options) { | |
9aa09ddd | 3994 | flags = 0; |
3cdc69d3 | 3995 | klass->inherit_options(role, parent_is_format, &flags, options, |
272c02ea | 3996 | parent_flags, parent_options); |
9aa09ddd AG |
3997 | } else { |
3998 | flags = bdrv_get_flags(bs); | |
28518102 KW |
3999 | } |
4000 | ||
077e8e20 AG |
4001 | if (keep_old_opts) { |
4002 | /* Old values are used for options that aren't set yet */ | |
4003 | old_options = qdict_clone_shallow(bs->options); | |
4004 | bdrv_join_options(bs, options, old_options); | |
4005 | qobject_unref(old_options); | |
4006 | } | |
4d2cb092 | 4007 | |
9aa09ddd AG |
4008 | /* We have the final set of options so let's update the flags */ |
4009 | options_copy = qdict_clone_shallow(options); | |
4010 | opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort); | |
4011 | qemu_opts_absorb_qdict(opts, options_copy, NULL); | |
4012 | update_flags_from_options(&flags, opts); | |
4013 | qemu_opts_del(opts); | |
4014 | qobject_unref(options_copy); | |
4015 | ||
fd452021 | 4016 | /* bdrv_open_inherit() sets and clears some additional flags internally */ |
f1f25a2e | 4017 | flags &= ~BDRV_O_PROTOCOL; |
fd452021 KW |
4018 | if (flags & BDRV_O_RDWR) { |
4019 | flags |= BDRV_O_ALLOW_RDWR; | |
4020 | } | |
f1f25a2e | 4021 | |
1857c97b KW |
4022 | if (!bs_entry) { |
4023 | bs_entry = g_new0(BlockReopenQueueEntry, 1); | |
859443b0 | 4024 | QTAILQ_INSERT_TAIL(bs_queue, bs_entry, entry); |
1857c97b | 4025 | } else { |
cb3e7f08 MAL |
4026 | qobject_unref(bs_entry->state.options); |
4027 | qobject_unref(bs_entry->state.explicit_options); | |
1857c97b KW |
4028 | } |
4029 | ||
4030 | bs_entry->state.bs = bs; | |
4031 | bs_entry->state.options = options; | |
4032 | bs_entry->state.explicit_options = explicit_options; | |
4033 | bs_entry->state.flags = flags; | |
4034 | ||
8546632e AG |
4035 | /* |
4036 | * If keep_old_opts is false then it means that unspecified | |
4037 | * options must be reset to their original value. We don't allow | |
4038 | * resetting 'backing' but we need to know if the option is | |
4039 | * missing in order to decide if we have to return an error. | |
4040 | */ | |
4041 | if (!keep_old_opts) { | |
4042 | bs_entry->state.backing_missing = | |
4043 | !qdict_haskey(options, "backing") && | |
4044 | !qdict_haskey(options, "backing.driver"); | |
4045 | } | |
4046 | ||
67251a31 | 4047 | QLIST_FOREACH(child, &bs->children, next) { |
8546632e AG |
4048 | QDict *new_child_options = NULL; |
4049 | bool child_keep_old = keep_old_opts; | |
67251a31 | 4050 | |
4c9dfe5d KW |
4051 | /* reopen can only change the options of block devices that were |
4052 | * implicitly created and inherited options. For other (referenced) | |
4053 | * block devices, a syntax like "backing.foo" results in an error. */ | |
67251a31 KW |
4054 | if (child->bs->inherits_from != bs) { |
4055 | continue; | |
4056 | } | |
4057 | ||
8546632e AG |
4058 | /* Check if the options contain a child reference */ |
4059 | if (qdict_haskey(options, child->name)) { | |
4060 | const char *childref = qdict_get_try_str(options, child->name); | |
4061 | /* | |
4062 | * The current child must not be reopened if the child | |
4063 | * reference is null or points to a different node. | |
4064 | */ | |
4065 | if (g_strcmp0(childref, child->bs->node_name)) { | |
4066 | continue; | |
4067 | } | |
4068 | /* | |
4069 | * If the child reference points to the current child then | |
4070 | * reopen it with its existing set of options (note that | |
4071 | * it can still inherit new options from the parent). | |
4072 | */ | |
4073 | child_keep_old = true; | |
4074 | } else { | |
4075 | /* Extract child options ("child-name.*") */ | |
4076 | char *child_key_dot = g_strdup_printf("%s.", child->name); | |
4077 | qdict_extract_subqdict(explicit_options, NULL, child_key_dot); | |
4078 | qdict_extract_subqdict(options, &new_child_options, child_key_dot); | |
4079 | g_free(child_key_dot); | |
4080 | } | |
4c9dfe5d | 4081 | |
9aa09ddd | 4082 | bdrv_reopen_queue_child(bs_queue, child->bs, new_child_options, |
3cdc69d3 HR |
4083 | child->klass, child->role, bs->drv->is_format, |
4084 | options, flags, child_keep_old); | |
e971aa12 JC |
4085 | } |
4086 | ||
e971aa12 JC |
4087 | return bs_queue; |
4088 | } | |
4089 | ||
28518102 KW |
4090 | BlockReopenQueue *bdrv_reopen_queue(BlockReopenQueue *bs_queue, |
4091 | BlockDriverState *bs, | |
077e8e20 | 4092 | QDict *options, bool keep_old_opts) |
28518102 | 4093 | { |
3cdc69d3 HR |
4094 | return bdrv_reopen_queue_child(bs_queue, bs, options, NULL, 0, false, |
4095 | NULL, 0, keep_old_opts); | |
28518102 KW |
4096 | } |
4097 | ||
ab5b5228 AG |
4098 | void bdrv_reopen_queue_free(BlockReopenQueue *bs_queue) |
4099 | { | |
4100 | if (bs_queue) { | |
4101 | BlockReopenQueueEntry *bs_entry, *next; | |
4102 | QTAILQ_FOREACH_SAFE(bs_entry, bs_queue, entry, next) { | |
4103 | qobject_unref(bs_entry->state.explicit_options); | |
4104 | qobject_unref(bs_entry->state.options); | |
4105 | g_free(bs_entry); | |
4106 | } | |
4107 | g_free(bs_queue); | |
4108 | } | |
4109 | } | |
4110 | ||
e971aa12 JC |
4111 | /* |
4112 | * Reopen multiple BlockDriverStates atomically & transactionally. | |
4113 | * | |
4114 | * The queue passed in (bs_queue) must have been built up previous | |
4115 | * via bdrv_reopen_queue(). | |
4116 | * | |
4117 | * Reopens all BDS specified in the queue, with the appropriate | |
4118 | * flags. All devices are prepared for reopen, and failure of any | |
50d6a8a3 | 4119 | * device will cause all device changes to be abandoned, and intermediate |
e971aa12 JC |
4120 | * data cleaned up. |
4121 | * | |
4122 | * If all devices prepare successfully, then the changes are committed | |
4123 | * to all devices. | |
4124 | * | |
1a63a907 KW |
4125 | * All affected nodes must be drained between bdrv_reopen_queue() and |
4126 | * bdrv_reopen_multiple(). | |
6cf42ca2 KW |
4127 | * |
4128 | * To be called from the main thread, with all other AioContexts unlocked. | |
e971aa12 | 4129 | */ |
5019aece | 4130 | int bdrv_reopen_multiple(BlockReopenQueue *bs_queue, Error **errp) |
e971aa12 JC |
4131 | { |
4132 | int ret = -1; | |
4133 | BlockReopenQueueEntry *bs_entry, *next; | |
6cf42ca2 | 4134 | AioContext *ctx; |
72373e40 VSO |
4135 | Transaction *tran = tran_new(); |
4136 | g_autoptr(GHashTable) found = NULL; | |
4137 | g_autoptr(GSList) refresh_list = NULL; | |
e971aa12 | 4138 | |
6cf42ca2 | 4139 | assert(qemu_get_current_aio_context() == qemu_get_aio_context()); |
e971aa12 JC |
4140 | assert(bs_queue != NULL); |
4141 | ||
a2aabf88 | 4142 | QTAILQ_FOREACH(bs_entry, bs_queue, entry) { |
6cf42ca2 KW |
4143 | ctx = bdrv_get_aio_context(bs_entry->state.bs); |
4144 | aio_context_acquire(ctx); | |
a2aabf88 | 4145 | ret = bdrv_flush(bs_entry->state.bs); |
6cf42ca2 | 4146 | aio_context_release(ctx); |
a2aabf88 VSO |
4147 | if (ret < 0) { |
4148 | error_setg_errno(errp, -ret, "Error flushing drive"); | |
e3fc91aa | 4149 | goto abort; |
a2aabf88 VSO |
4150 | } |
4151 | } | |
4152 | ||
859443b0 | 4153 | QTAILQ_FOREACH(bs_entry, bs_queue, entry) { |
1a63a907 | 4154 | assert(bs_entry->state.bs->quiesce_counter > 0); |
6cf42ca2 KW |
4155 | ctx = bdrv_get_aio_context(bs_entry->state.bs); |
4156 | aio_context_acquire(ctx); | |
72373e40 | 4157 | ret = bdrv_reopen_prepare(&bs_entry->state, bs_queue, tran, errp); |
6cf42ca2 | 4158 | aio_context_release(ctx); |
72373e40 VSO |
4159 | if (ret < 0) { |
4160 | goto abort; | |
e971aa12 JC |
4161 | } |
4162 | bs_entry->prepared = true; | |
4163 | } | |
4164 | ||
72373e40 | 4165 | found = g_hash_table_new(NULL, NULL); |
859443b0 | 4166 | QTAILQ_FOREACH(bs_entry, bs_queue, entry) { |
69b736e7 | 4167 | BDRVReopenState *state = &bs_entry->state; |
72373e40 VSO |
4168 | |
4169 | refresh_list = bdrv_topological_dfs(refresh_list, found, state->bs); | |
4170 | if (state->old_backing_bs) { | |
4171 | refresh_list = bdrv_topological_dfs(refresh_list, found, | |
4172 | state->old_backing_bs); | |
cb828c31 | 4173 | } |
ecd30d2d AG |
4174 | if (state->old_file_bs) { |
4175 | refresh_list = bdrv_topological_dfs(refresh_list, found, | |
4176 | state->old_file_bs); | |
4177 | } | |
72373e40 VSO |
4178 | } |
4179 | ||
4180 | /* | |
4181 | * Note that file-posix driver rely on permission update done during reopen | |
4182 | * (even if no permission changed), because it wants "new" permissions for | |
4183 | * reconfiguring the fd and that's why it does it in raw_check_perm(), not | |
4184 | * in raw_reopen_prepare() which is called with "old" permissions. | |
4185 | */ | |
4186 | ret = bdrv_list_refresh_perms(refresh_list, bs_queue, tran, errp); | |
4187 | if (ret < 0) { | |
4188 | goto abort; | |
69b736e7 KW |
4189 | } |
4190 | ||
fcd6a4f4 VSO |
4191 | /* |
4192 | * If we reach this point, we have success and just need to apply the | |
4193 | * changes. | |
4194 | * | |
4195 | * Reverse order is used to comfort qcow2 driver: on commit it need to write | |
4196 | * IN_USE flag to the image, to mark bitmaps in the image as invalid. But | |
4197 | * children are usually goes after parents in reopen-queue, so go from last | |
4198 | * to first element. | |
e971aa12 | 4199 | */ |
fcd6a4f4 | 4200 | QTAILQ_FOREACH_REVERSE(bs_entry, bs_queue, entry) { |
6cf42ca2 KW |
4201 | ctx = bdrv_get_aio_context(bs_entry->state.bs); |
4202 | aio_context_acquire(ctx); | |
e971aa12 | 4203 | bdrv_reopen_commit(&bs_entry->state); |
6cf42ca2 | 4204 | aio_context_release(ctx); |
e971aa12 JC |
4205 | } |
4206 | ||
72373e40 | 4207 | tran_commit(tran); |
69b736e7 | 4208 | |
72373e40 VSO |
4209 | QTAILQ_FOREACH_REVERSE(bs_entry, bs_queue, entry) { |
4210 | BlockDriverState *bs = bs_entry->state.bs; | |
74ad9a3b | 4211 | |
72373e40 | 4212 | if (bs->drv->bdrv_reopen_commit_post) { |
6cf42ca2 KW |
4213 | ctx = bdrv_get_aio_context(bs); |
4214 | aio_context_acquire(ctx); | |
72373e40 | 4215 | bs->drv->bdrv_reopen_commit_post(&bs_entry->state); |
6cf42ca2 | 4216 | aio_context_release(ctx); |
69b736e7 KW |
4217 | } |
4218 | } | |
17e1e2be | 4219 | |
72373e40 VSO |
4220 | ret = 0; |
4221 | goto cleanup; | |
17e1e2be | 4222 | |
72373e40 VSO |
4223 | abort: |
4224 | tran_abort(tran); | |
4225 | QTAILQ_FOREACH_SAFE(bs_entry, bs_queue, entry, next) { | |
4226 | if (bs_entry->prepared) { | |
6cf42ca2 KW |
4227 | ctx = bdrv_get_aio_context(bs_entry->state.bs); |
4228 | aio_context_acquire(ctx); | |
72373e40 | 4229 | bdrv_reopen_abort(&bs_entry->state); |
6cf42ca2 | 4230 | aio_context_release(ctx); |
17e1e2be PK |
4231 | } |
4232 | } | |
72373e40 | 4233 | |
e971aa12 | 4234 | cleanup: |
ab5b5228 | 4235 | bdrv_reopen_queue_free(bs_queue); |
40840e41 | 4236 | |
e971aa12 JC |
4237 | return ret; |
4238 | } | |
4239 | ||
6cf42ca2 KW |
4240 | int bdrv_reopen(BlockDriverState *bs, QDict *opts, bool keep_old_opts, |
4241 | Error **errp) | |
6e1000a8 | 4242 | { |
6cf42ca2 | 4243 | AioContext *ctx = bdrv_get_aio_context(bs); |
6e1000a8 | 4244 | BlockReopenQueue *queue; |
6cf42ca2 | 4245 | int ret; |
6e1000a8 AG |
4246 | |
4247 | bdrv_subtree_drained_begin(bs); | |
6cf42ca2 KW |
4248 | if (ctx != qemu_get_aio_context()) { |
4249 | aio_context_release(ctx); | |
4250 | } | |
4251 | ||
4252 | queue = bdrv_reopen_queue(NULL, bs, opts, keep_old_opts); | |
5019aece | 4253 | ret = bdrv_reopen_multiple(queue, errp); |
6cf42ca2 KW |
4254 | |
4255 | if (ctx != qemu_get_aio_context()) { | |
4256 | aio_context_acquire(ctx); | |
4257 | } | |
6e1000a8 AG |
4258 | bdrv_subtree_drained_end(bs); |
4259 | ||
4260 | return ret; | |
4261 | } | |
4262 | ||
6cf42ca2 KW |
4263 | int bdrv_reopen_set_read_only(BlockDriverState *bs, bool read_only, |
4264 | Error **errp) | |
4265 | { | |
4266 | QDict *opts = qdict_new(); | |
4267 | ||
4268 | qdict_put_bool(opts, BDRV_OPT_READ_ONLY, read_only); | |
4269 | ||
4270 | return bdrv_reopen(bs, opts, true, errp); | |
4271 | } | |
4272 | ||
cb828c31 AG |
4273 | /* |
4274 | * Take a BDRVReopenState and check if the value of 'backing' in the | |
4275 | * reopen_state->options QDict is valid or not. | |
4276 | * | |
4277 | * If 'backing' is missing from the QDict then return 0. | |
4278 | * | |
4279 | * If 'backing' contains the node name of the backing file of | |
4280 | * reopen_state->bs then return 0. | |
4281 | * | |
4282 | * If 'backing' contains a different node name (or is null) then check | |
4283 | * whether the current backing file can be replaced with the new one. | |
4284 | * If that's the case then reopen_state->replace_backing_bs is set to | |
4285 | * true and reopen_state->new_backing_bs contains a pointer to the new | |
4286 | * backing BlockDriverState (or NULL). | |
4287 | * | |
4288 | * Return 0 on success, otherwise return < 0 and set @errp. | |
4289 | */ | |
ecd30d2d AG |
4290 | static int bdrv_reopen_parse_file_or_backing(BDRVReopenState *reopen_state, |
4291 | bool is_backing, Transaction *tran, | |
4292 | Error **errp) | |
cb828c31 AG |
4293 | { |
4294 | BlockDriverState *bs = reopen_state->bs; | |
ecd30d2d AG |
4295 | BlockDriverState *new_child_bs; |
4296 | BlockDriverState *old_child_bs = is_backing ? child_bs(bs->backing) : | |
4297 | child_bs(bs->file); | |
4298 | const char *child_name = is_backing ? "backing" : "file"; | |
cb828c31 AG |
4299 | QObject *value; |
4300 | const char *str; | |
4301 | ||
ecd30d2d | 4302 | value = qdict_get(reopen_state->options, child_name); |
cb828c31 AG |
4303 | if (value == NULL) { |
4304 | return 0; | |
4305 | } | |
4306 | ||
4307 | switch (qobject_type(value)) { | |
4308 | case QTYPE_QNULL: | |
ecd30d2d AG |
4309 | assert(is_backing); /* The 'file' option does not allow a null value */ |
4310 | new_child_bs = NULL; | |
cb828c31 AG |
4311 | break; |
4312 | case QTYPE_QSTRING: | |
410f44f5 | 4313 | str = qstring_get_str(qobject_to(QString, value)); |
ecd30d2d AG |
4314 | new_child_bs = bdrv_lookup_bs(NULL, str, errp); |
4315 | if (new_child_bs == NULL) { | |
cb828c31 | 4316 | return -EINVAL; |
ecd30d2d AG |
4317 | } else if (bdrv_recurse_has_child(new_child_bs, bs)) { |
4318 | error_setg(errp, "Making '%s' a %s child of '%s' would create a " | |
4319 | "cycle", str, child_name, bs->node_name); | |
cb828c31 AG |
4320 | return -EINVAL; |
4321 | } | |
4322 | break; | |
4323 | default: | |
ecd30d2d AG |
4324 | /* |
4325 | * The options QDict has been flattened, so 'backing' and 'file' | |
4326 | * do not allow any other data type here. | |
4327 | */ | |
cb828c31 AG |
4328 | g_assert_not_reached(); |
4329 | } | |
4330 | ||
ecd30d2d AG |
4331 | if (old_child_bs == new_child_bs) { |
4332 | return 0; | |
4333 | } | |
4334 | ||
4335 | if (old_child_bs) { | |
4336 | if (bdrv_skip_implicit_filters(old_child_bs) == new_child_bs) { | |
cbfdb98c VSO |
4337 | return 0; |
4338 | } | |
4339 | ||
ecd30d2d AG |
4340 | if (old_child_bs->implicit) { |
4341 | error_setg(errp, "Cannot replace implicit %s child of %s", | |
4342 | child_name, bs->node_name); | |
cbfdb98c VSO |
4343 | return -EPERM; |
4344 | } | |
4345 | } | |
4346 | ||
ecd30d2d | 4347 | if (bs->drv->is_filter && !old_child_bs) { |
25f78d9e VSO |
4348 | /* |
4349 | * Filters always have a file or a backing child, so we are trying to | |
4350 | * change wrong child | |
4351 | */ | |
4352 | error_setg(errp, "'%s' is a %s filter node that does not support a " | |
ecd30d2d | 4353 | "%s child", bs->node_name, bs->drv->format_name, child_name); |
1d42f48c HR |
4354 | return -EINVAL; |
4355 | } | |
4356 | ||
ecd30d2d AG |
4357 | if (is_backing) { |
4358 | reopen_state->old_backing_bs = old_child_bs; | |
4359 | } else { | |
4360 | reopen_state->old_file_bs = old_child_bs; | |
4361 | } | |
4362 | ||
4363 | return bdrv_set_file_or_backing_noperm(bs, new_child_bs, is_backing, | |
4364 | tran, errp); | |
cb828c31 AG |
4365 | } |
4366 | ||
e971aa12 JC |
4367 | /* |
4368 | * Prepares a BlockDriverState for reopen. All changes are staged in the | |
4369 | * 'opaque' field of the BDRVReopenState, which is used and allocated by | |
4370 | * the block driver layer .bdrv_reopen_prepare() | |
4371 | * | |
4372 | * bs is the BlockDriverState to reopen | |
4373 | * flags are the new open flags | |
4374 | * queue is the reopen queue | |
4375 | * | |
4376 | * Returns 0 on success, non-zero on error. On error errp will be set | |
4377 | * as well. | |
4378 | * | |
4379 | * On failure, bdrv_reopen_abort() will be called to clean up any data. | |
4380 | * It is the responsibility of the caller to then call the abort() or | |
4381 | * commit() for any other BDS that have been left in a prepare() state | |
4382 | * | |
4383 | */ | |
53e96d1e | 4384 | static int bdrv_reopen_prepare(BDRVReopenState *reopen_state, |
72373e40 | 4385 | BlockReopenQueue *queue, |
ecd30d2d | 4386 | Transaction *change_child_tran, Error **errp) |
e971aa12 JC |
4387 | { |
4388 | int ret = -1; | |
e6d79c41 | 4389 | int old_flags; |
e971aa12 JC |
4390 | Error *local_err = NULL; |
4391 | BlockDriver *drv; | |
ccf9dc07 | 4392 | QemuOpts *opts; |
4c8350fe | 4393 | QDict *orig_reopen_opts; |
593b3071 | 4394 | char *discard = NULL; |
3d8ce171 | 4395 | bool read_only; |
9ad08c44 | 4396 | bool drv_prepared = false; |
e971aa12 JC |
4397 | |
4398 | assert(reopen_state != NULL); | |
4399 | assert(reopen_state->bs->drv != NULL); | |
4400 | drv = reopen_state->bs->drv; | |
4401 | ||
4c8350fe AG |
4402 | /* This function and each driver's bdrv_reopen_prepare() remove |
4403 | * entries from reopen_state->options as they are processed, so | |
4404 | * we need to make a copy of the original QDict. */ | |
4405 | orig_reopen_opts = qdict_clone_shallow(reopen_state->options); | |
4406 | ||
ccf9dc07 KW |
4407 | /* Process generic block layer options */ |
4408 | opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort); | |
af175e85 | 4409 | if (!qemu_opts_absorb_qdict(opts, reopen_state->options, errp)) { |
ccf9dc07 KW |
4410 | ret = -EINVAL; |
4411 | goto error; | |
4412 | } | |
4413 | ||
e6d79c41 AG |
4414 | /* This was already called in bdrv_reopen_queue_child() so the flags |
4415 | * are up-to-date. This time we simply want to remove the options from | |
4416 | * QemuOpts in order to indicate that they have been processed. */ | |
4417 | old_flags = reopen_state->flags; | |
91a097e7 | 4418 | update_flags_from_options(&reopen_state->flags, opts); |
e6d79c41 | 4419 | assert(old_flags == reopen_state->flags); |
91a097e7 | 4420 | |
415bbca8 | 4421 | discard = qemu_opt_get_del(opts, BDRV_OPT_DISCARD); |
593b3071 AG |
4422 | if (discard != NULL) { |
4423 | if (bdrv_parse_discard_flags(discard, &reopen_state->flags) != 0) { | |
4424 | error_setg(errp, "Invalid discard option"); | |
4425 | ret = -EINVAL; | |
4426 | goto error; | |
4427 | } | |
4428 | } | |
4429 | ||
543770bd AG |
4430 | reopen_state->detect_zeroes = |
4431 | bdrv_parse_detect_zeroes(opts, reopen_state->flags, &local_err); | |
4432 | if (local_err) { | |
4433 | error_propagate(errp, local_err); | |
4434 | ret = -EINVAL; | |
4435 | goto error; | |
4436 | } | |
4437 | ||
57f9db9a AG |
4438 | /* All other options (including node-name and driver) must be unchanged. |
4439 | * Put them back into the QDict, so that they are checked at the end | |
4440 | * of this function. */ | |
4441 | qemu_opts_to_qdict(opts, reopen_state->options); | |
ccf9dc07 | 4442 | |
3d8ce171 JC |
4443 | /* If we are to stay read-only, do not allow permission change |
4444 | * to r/w. Attempting to set to r/w may fail if either BDRV_O_ALLOW_RDWR is | |
4445 | * not set, or if the BDS still has copy_on_read enabled */ | |
4446 | read_only = !(reopen_state->flags & BDRV_O_RDWR); | |
54a32bfe | 4447 | ret = bdrv_can_set_read_only(reopen_state->bs, read_only, true, &local_err); |
3d8ce171 JC |
4448 | if (local_err) { |
4449 | error_propagate(errp, local_err); | |
e971aa12 JC |
4450 | goto error; |
4451 | } | |
4452 | ||
e971aa12 | 4453 | if (drv->bdrv_reopen_prepare) { |
faf116b4 AG |
4454 | /* |
4455 | * If a driver-specific option is missing, it means that we | |
4456 | * should reset it to its default value. | |
4457 | * But not all options allow that, so we need to check it first. | |
4458 | */ | |
4459 | ret = bdrv_reset_options_allowed(reopen_state->bs, | |
4460 | reopen_state->options, errp); | |
4461 | if (ret) { | |
4462 | goto error; | |
4463 | } | |
4464 | ||
e971aa12 JC |
4465 | ret = drv->bdrv_reopen_prepare(reopen_state, queue, &local_err); |
4466 | if (ret) { | |
4467 | if (local_err != NULL) { | |
4468 | error_propagate(errp, local_err); | |
4469 | } else { | |
f30c66ba | 4470 | bdrv_refresh_filename(reopen_state->bs); |
d8b6895f LC |
4471 | error_setg(errp, "failed while preparing to reopen image '%s'", |
4472 | reopen_state->bs->filename); | |
e971aa12 JC |
4473 | } |
4474 | goto error; | |
4475 | } | |
4476 | } else { | |
4477 | /* It is currently mandatory to have a bdrv_reopen_prepare() | |
4478 | * handler for each supported drv. */ | |
81e5f78a AG |
4479 | error_setg(errp, "Block format '%s' used by node '%s' " |
4480 | "does not support reopening files", drv->format_name, | |
4481 | bdrv_get_device_or_node_name(reopen_state->bs)); | |
e971aa12 JC |
4482 | ret = -1; |
4483 | goto error; | |
4484 | } | |
4485 | ||
9ad08c44 HR |
4486 | drv_prepared = true; |
4487 | ||
bacd9b87 AG |
4488 | /* |
4489 | * We must provide the 'backing' option if the BDS has a backing | |
4490 | * file or if the image file has a backing file name as part of | |
4491 | * its metadata. Otherwise the 'backing' option can be omitted. | |
4492 | */ | |
4493 | if (drv->supports_backing && reopen_state->backing_missing && | |
1d42f48c | 4494 | (reopen_state->bs->backing || reopen_state->bs->backing_file[0])) { |
8546632e AG |
4495 | error_setg(errp, "backing is missing for '%s'", |
4496 | reopen_state->bs->node_name); | |
4497 | ret = -EINVAL; | |
4498 | goto error; | |
4499 | } | |
4500 | ||
cb828c31 AG |
4501 | /* |
4502 | * Allow changing the 'backing' option. The new value can be | |
4503 | * either a reference to an existing node (using its node name) | |
4504 | * or NULL to simply detach the current backing file. | |
4505 | */ | |
ecd30d2d AG |
4506 | ret = bdrv_reopen_parse_file_or_backing(reopen_state, true, |
4507 | change_child_tran, errp); | |
cb828c31 AG |
4508 | if (ret < 0) { |
4509 | goto error; | |
4510 | } | |
4511 | qdict_del(reopen_state->options, "backing"); | |
4512 | ||
ecd30d2d AG |
4513 | /* Allow changing the 'file' option. In this case NULL is not allowed */ |
4514 | ret = bdrv_reopen_parse_file_or_backing(reopen_state, false, | |
4515 | change_child_tran, errp); | |
4516 | if (ret < 0) { | |
4517 | goto error; | |
4518 | } | |
4519 | qdict_del(reopen_state->options, "file"); | |
4520 | ||
4d2cb092 KW |
4521 | /* Options that are not handled are only okay if they are unchanged |
4522 | * compared to the old state. It is expected that some options are only | |
4523 | * used for the initial open, but not reopen (e.g. filename) */ | |
4524 | if (qdict_size(reopen_state->options)) { | |
4525 | const QDictEntry *entry = qdict_first(reopen_state->options); | |
4526 | ||
4527 | do { | |
54fd1b0d HR |
4528 | QObject *new = entry->value; |
4529 | QObject *old = qdict_get(reopen_state->bs->options, entry->key); | |
4530 | ||
db905283 AG |
4531 | /* Allow child references (child_name=node_name) as long as they |
4532 | * point to the current child (i.e. everything stays the same). */ | |
4533 | if (qobject_type(new) == QTYPE_QSTRING) { | |
4534 | BdrvChild *child; | |
4535 | QLIST_FOREACH(child, &reopen_state->bs->children, next) { | |
4536 | if (!strcmp(child->name, entry->key)) { | |
4537 | break; | |
4538 | } | |
4539 | } | |
4540 | ||
4541 | if (child) { | |
410f44f5 MA |
4542 | if (!strcmp(child->bs->node_name, |
4543 | qstring_get_str(qobject_to(QString, new)))) { | |
db905283 AG |
4544 | continue; /* Found child with this name, skip option */ |
4545 | } | |
4546 | } | |
4547 | } | |
4548 | ||
129c7d1c | 4549 | /* |
54fd1b0d HR |
4550 | * TODO: When using -drive to specify blockdev options, all values |
4551 | * will be strings; however, when using -blockdev, blockdev-add or | |
4552 | * filenames using the json:{} pseudo-protocol, they will be | |
4553 | * correctly typed. | |
4554 | * In contrast, reopening options are (currently) always strings | |
4555 | * (because you can only specify them through qemu-io; all other | |
4556 | * callers do not specify any options). | |
4557 | * Therefore, when using anything other than -drive to create a BDS, | |
4558 | * this cannot detect non-string options as unchanged, because | |
4559 | * qobject_is_equal() always returns false for objects of different | |
4560 | * type. In the future, this should be remedied by correctly typing | |
4561 | * all options. For now, this is not too big of an issue because | |
4562 | * the user can simply omit options which cannot be changed anyway, | |
4563 | * so they will stay unchanged. | |
129c7d1c | 4564 | */ |
54fd1b0d | 4565 | if (!qobject_is_equal(new, old)) { |
4d2cb092 KW |
4566 | error_setg(errp, "Cannot change the option '%s'", entry->key); |
4567 | ret = -EINVAL; | |
4568 | goto error; | |
4569 | } | |
4570 | } while ((entry = qdict_next(reopen_state->options, entry))); | |
4571 | } | |
4572 | ||
e971aa12 JC |
4573 | ret = 0; |
4574 | ||
4c8350fe AG |
4575 | /* Restore the original reopen_state->options QDict */ |
4576 | qobject_unref(reopen_state->options); | |
4577 | reopen_state->options = qobject_ref(orig_reopen_opts); | |
4578 | ||
e971aa12 | 4579 | error: |
9ad08c44 HR |
4580 | if (ret < 0 && drv_prepared) { |
4581 | /* drv->bdrv_reopen_prepare() has succeeded, so we need to | |
4582 | * call drv->bdrv_reopen_abort() before signaling an error | |
4583 | * (bdrv_reopen_multiple() will not call bdrv_reopen_abort() | |
4584 | * when the respective bdrv_reopen_prepare() has failed) */ | |
4585 | if (drv->bdrv_reopen_abort) { | |
4586 | drv->bdrv_reopen_abort(reopen_state); | |
4587 | } | |
4588 | } | |
ccf9dc07 | 4589 | qemu_opts_del(opts); |
4c8350fe | 4590 | qobject_unref(orig_reopen_opts); |
593b3071 | 4591 | g_free(discard); |
e971aa12 JC |
4592 | return ret; |
4593 | } | |
4594 | ||
4595 | /* | |
4596 | * Takes the staged changes for the reopen from bdrv_reopen_prepare(), and | |
4597 | * makes them final by swapping the staging BlockDriverState contents into | |
4598 | * the active BlockDriverState contents. | |
4599 | */ | |
53e96d1e | 4600 | static void bdrv_reopen_commit(BDRVReopenState *reopen_state) |
e971aa12 JC |
4601 | { |
4602 | BlockDriver *drv; | |
50bf65ba | 4603 | BlockDriverState *bs; |
50196d7a | 4604 | BdrvChild *child; |
e971aa12 JC |
4605 | |
4606 | assert(reopen_state != NULL); | |
50bf65ba VSO |
4607 | bs = reopen_state->bs; |
4608 | drv = bs->drv; | |
e971aa12 JC |
4609 | assert(drv != NULL); |
4610 | ||
4611 | /* If there are any driver level actions to take */ | |
4612 | if (drv->bdrv_reopen_commit) { | |
4613 | drv->bdrv_reopen_commit(reopen_state); | |
4614 | } | |
4615 | ||
4616 | /* set BDS specific flags now */ | |
cb3e7f08 | 4617 | qobject_unref(bs->explicit_options); |
4c8350fe | 4618 | qobject_unref(bs->options); |
ab5b5228 AG |
4619 | qobject_ref(reopen_state->explicit_options); |
4620 | qobject_ref(reopen_state->options); | |
145f598e | 4621 | |
50bf65ba | 4622 | bs->explicit_options = reopen_state->explicit_options; |
4c8350fe | 4623 | bs->options = reopen_state->options; |
50bf65ba | 4624 | bs->open_flags = reopen_state->flags; |
543770bd | 4625 | bs->detect_zeroes = reopen_state->detect_zeroes; |
355ef4ac | 4626 | |
50196d7a AG |
4627 | /* Remove child references from bs->options and bs->explicit_options. |
4628 | * Child options were already removed in bdrv_reopen_queue_child() */ | |
4629 | QLIST_FOREACH(child, &bs->children, next) { | |
4630 | qdict_del(bs->explicit_options, child->name); | |
4631 | qdict_del(bs->options, child->name); | |
4632 | } | |
3d0e8743 VSO |
4633 | /* backing is probably removed, so it's not handled by previous loop */ |
4634 | qdict_del(bs->explicit_options, "backing"); | |
4635 | qdict_del(bs->options, "backing"); | |
4636 | ||
1e4c797c | 4637 | bdrv_refresh_limits(bs, NULL, NULL); |
e971aa12 JC |
4638 | } |
4639 | ||
4640 | /* | |
4641 | * Abort the reopen, and delete and free the staged changes in | |
4642 | * reopen_state | |
4643 | */ | |
53e96d1e | 4644 | static void bdrv_reopen_abort(BDRVReopenState *reopen_state) |
e971aa12 JC |
4645 | { |
4646 | BlockDriver *drv; | |
4647 | ||
4648 | assert(reopen_state != NULL); | |
4649 | drv = reopen_state->bs->drv; | |
4650 | assert(drv != NULL); | |
4651 | ||
4652 | if (drv->bdrv_reopen_abort) { | |
4653 | drv->bdrv_reopen_abort(reopen_state); | |
4654 | } | |
4655 | } | |
4656 | ||
4657 | ||
64dff520 | 4658 | static void bdrv_close(BlockDriverState *bs) |
fc01f7e7 | 4659 | { |
33384421 | 4660 | BdrvAioNotifier *ban, *ban_next; |
50a3efb0 | 4661 | BdrvChild *child, *next; |
33384421 | 4662 | |
30f55fb8 | 4663 | assert(!bs->refcnt); |
99b7e775 | 4664 | |
fc27291d | 4665 | bdrv_drained_begin(bs); /* complete I/O */ |
58fda173 | 4666 | bdrv_flush(bs); |
53ec73e2 | 4667 | bdrv_drain(bs); /* in case flush left pending I/O */ |
fc27291d | 4668 | |
3cbc002c | 4669 | if (bs->drv) { |
3c005293 | 4670 | if (bs->drv->bdrv_close) { |
7b99a266 | 4671 | /* Must unfreeze all children, so bdrv_unref_child() works */ |
3c005293 VSO |
4672 | bs->drv->bdrv_close(bs); |
4673 | } | |
9a4f4c31 | 4674 | bs->drv = NULL; |
50a3efb0 | 4675 | } |
9a7dedbc | 4676 | |
50a3efb0 | 4677 | QLIST_FOREACH_SAFE(child, &bs->children, next, next) { |
dd4118c7 | 4678 | bdrv_unref_child(bs, child); |
b338082b | 4679 | } |
98f90dba | 4680 | |
dd4118c7 AG |
4681 | bs->backing = NULL; |
4682 | bs->file = NULL; | |
50a3efb0 AG |
4683 | g_free(bs->opaque); |
4684 | bs->opaque = NULL; | |
d73415a3 | 4685 | qatomic_set(&bs->copy_on_read, 0); |
50a3efb0 AG |
4686 | bs->backing_file[0] = '\0'; |
4687 | bs->backing_format[0] = '\0'; | |
4688 | bs->total_sectors = 0; | |
4689 | bs->encrypted = false; | |
4690 | bs->sg = false; | |
cb3e7f08 MAL |
4691 | qobject_unref(bs->options); |
4692 | qobject_unref(bs->explicit_options); | |
50a3efb0 AG |
4693 | bs->options = NULL; |
4694 | bs->explicit_options = NULL; | |
cb3e7f08 | 4695 | qobject_unref(bs->full_open_options); |
50a3efb0 AG |
4696 | bs->full_open_options = NULL; |
4697 | ||
cca43ae1 VSO |
4698 | bdrv_release_named_dirty_bitmaps(bs); |
4699 | assert(QLIST_EMPTY(&bs->dirty_bitmaps)); | |
4700 | ||
33384421 HR |
4701 | QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_next) { |
4702 | g_free(ban); | |
4703 | } | |
4704 | QLIST_INIT(&bs->aio_notifiers); | |
fc27291d | 4705 | bdrv_drained_end(bs); |
1a6d3bd2 GK |
4706 | |
4707 | /* | |
4708 | * If we're still inside some bdrv_drain_all_begin()/end() sections, end | |
4709 | * them now since this BDS won't exist anymore when bdrv_drain_all_end() | |
4710 | * gets called. | |
4711 | */ | |
4712 | if (bs->quiesce_counter) { | |
4713 | bdrv_drain_all_end_quiesce(bs); | |
4714 | } | |
b338082b FB |
4715 | } |
4716 | ||
2bc93fed MK |
4717 | void bdrv_close_all(void) |
4718 | { | |
b3b5299d | 4719 | assert(job_next(NULL) == NULL); |
ca9bd24c HR |
4720 | |
4721 | /* Drop references from requests still in flight, such as canceled block | |
4722 | * jobs whose AIO context has not been polled yet */ | |
4723 | bdrv_drain_all(); | |
2bc93fed | 4724 | |
ca9bd24c HR |
4725 | blk_remove_all_bs(); |
4726 | blockdev_close_all_bdrv_states(); | |
ed78cda3 | 4727 | |
a1a2af07 | 4728 | assert(QTAILQ_EMPTY(&all_bdrv_states)); |
2bc93fed MK |
4729 | } |
4730 | ||
d0ac0380 KW |
4731 | static bool should_update_child(BdrvChild *c, BlockDriverState *to) |
4732 | { | |
2f30b7c3 VSO |
4733 | GQueue *queue; |
4734 | GHashTable *found; | |
4735 | bool ret; | |
d0ac0380 | 4736 | |
bd86fb99 | 4737 | if (c->klass->stay_at_node) { |
d0ac0380 KW |
4738 | return false; |
4739 | } | |
4740 | ||
ec9f10fe HR |
4741 | /* If the child @c belongs to the BDS @to, replacing the current |
4742 | * c->bs by @to would mean to create a loop. | |
4743 | * | |
4744 | * Such a case occurs when appending a BDS to a backing chain. | |
4745 | * For instance, imagine the following chain: | |
4746 | * | |
4747 | * guest device -> node A -> further backing chain... | |
4748 | * | |
4749 | * Now we create a new BDS B which we want to put on top of this | |
4750 | * chain, so we first attach A as its backing node: | |
4751 | * | |
4752 | * node B | |
4753 | * | | |
4754 | * v | |
4755 | * guest device -> node A -> further backing chain... | |
4756 | * | |
4757 | * Finally we want to replace A by B. When doing that, we want to | |
4758 | * replace all pointers to A by pointers to B -- except for the | |
4759 | * pointer from B because (1) that would create a loop, and (2) | |
4760 | * that pointer should simply stay intact: | |
4761 | * | |
4762 | * guest device -> node B | |
4763 | * | | |
4764 | * v | |
4765 | * node A -> further backing chain... | |
4766 | * | |
4767 | * In general, when replacing a node A (c->bs) by a node B (@to), | |
4768 | * if A is a child of B, that means we cannot replace A by B there | |
4769 | * because that would create a loop. Silently detaching A from B | |
4770 | * is also not really an option. So overall just leaving A in | |
2f30b7c3 VSO |
4771 | * place there is the most sensible choice. |
4772 | * | |
4773 | * We would also create a loop in any cases where @c is only | |
4774 | * indirectly referenced by @to. Prevent this by returning false | |
4775 | * if @c is found (by breadth-first search) anywhere in the whole | |
4776 | * subtree of @to. | |
4777 | */ | |
4778 | ||
4779 | ret = true; | |
4780 | found = g_hash_table_new(NULL, NULL); | |
4781 | g_hash_table_add(found, to); | |
4782 | queue = g_queue_new(); | |
4783 | g_queue_push_tail(queue, to); | |
4784 | ||
4785 | while (!g_queue_is_empty(queue)) { | |
4786 | BlockDriverState *v = g_queue_pop_head(queue); | |
4787 | BdrvChild *c2; | |
4788 | ||
4789 | QLIST_FOREACH(c2, &v->children, next) { | |
4790 | if (c2 == c) { | |
4791 | ret = false; | |
4792 | break; | |
4793 | } | |
4794 | ||
4795 | if (g_hash_table_contains(found, c2->bs)) { | |
4796 | continue; | |
4797 | } | |
4798 | ||
4799 | g_queue_push_tail(queue, c2->bs); | |
4800 | g_hash_table_add(found, c2->bs); | |
d0ac0380 KW |
4801 | } |
4802 | } | |
4803 | ||
2f30b7c3 VSO |
4804 | g_queue_free(queue); |
4805 | g_hash_table_destroy(found); | |
4806 | ||
4807 | return ret; | |
d0ac0380 KW |
4808 | } |
4809 | ||
46541ee5 VSO |
4810 | typedef struct BdrvRemoveFilterOrCowChild { |
4811 | BdrvChild *child; | |
4812 | bool is_backing; | |
4813 | } BdrvRemoveFilterOrCowChild; | |
4814 | ||
4815 | static void bdrv_remove_filter_or_cow_child_abort(void *opaque) | |
4816 | { | |
4817 | BdrvRemoveFilterOrCowChild *s = opaque; | |
4818 | BlockDriverState *parent_bs = s->child->opaque; | |
4819 | ||
4820 | QLIST_INSERT_HEAD(&parent_bs->children, s->child, next); | |
4821 | if (s->is_backing) { | |
4822 | parent_bs->backing = s->child; | |
4823 | } else { | |
4824 | parent_bs->file = s->child; | |
4825 | } | |
4826 | ||
4827 | /* | |
4bf021db | 4828 | * We don't have to restore child->bs here to undo bdrv_replace_child_tran() |
46541ee5 VSO |
4829 | * because that function is transactionable and it registered own completion |
4830 | * entries in @tran, so .abort() for bdrv_replace_child_safe() will be | |
4831 | * called automatically. | |
4832 | */ | |
4833 | } | |
4834 | ||
4835 | static void bdrv_remove_filter_or_cow_child_commit(void *opaque) | |
4836 | { | |
4837 | BdrvRemoveFilterOrCowChild *s = opaque; | |
4838 | ||
4839 | bdrv_child_free(s->child); | |
4840 | } | |
4841 | ||
4842 | static TransactionActionDrv bdrv_remove_filter_or_cow_child_drv = { | |
4843 | .abort = bdrv_remove_filter_or_cow_child_abort, | |
4844 | .commit = bdrv_remove_filter_or_cow_child_commit, | |
4845 | .clean = g_free, | |
4846 | }; | |
4847 | ||
4848 | /* | |
5b995019 | 4849 | * A function to remove backing or file child of @bs. |
7ec390d5 | 4850 | * Function doesn't update permissions, caller is responsible for this. |
46541ee5 | 4851 | */ |
5b995019 VSO |
4852 | static void bdrv_remove_file_or_backing_child(BlockDriverState *bs, |
4853 | BdrvChild *child, | |
4854 | Transaction *tran) | |
46541ee5 VSO |
4855 | { |
4856 | BdrvRemoveFilterOrCowChild *s; | |
5b995019 VSO |
4857 | |
4858 | assert(child == bs->backing || child == bs->file); | |
46541ee5 VSO |
4859 | |
4860 | if (!child) { | |
4861 | return; | |
4862 | } | |
4863 | ||
4864 | if (child->bs) { | |
4bf021db | 4865 | bdrv_replace_child_tran(child, NULL, tran); |
46541ee5 VSO |
4866 | } |
4867 | ||
4868 | s = g_new(BdrvRemoveFilterOrCowChild, 1); | |
4869 | *s = (BdrvRemoveFilterOrCowChild) { | |
4870 | .child = child, | |
4871 | .is_backing = (child == bs->backing), | |
4872 | }; | |
4873 | tran_add(tran, &bdrv_remove_filter_or_cow_child_drv, s); | |
4874 | ||
4875 | QLIST_SAFE_REMOVE(child, next); | |
4876 | if (s->is_backing) { | |
4877 | bs->backing = NULL; | |
4878 | } else { | |
4879 | bs->file = NULL; | |
4880 | } | |
4881 | } | |
4882 | ||
5b995019 VSO |
4883 | /* |
4884 | * A function to remove backing-chain child of @bs if exists: cow child for | |
4885 | * format nodes (always .backing) and filter child for filters (may be .file or | |
4886 | * .backing) | |
4887 | */ | |
4888 | static void bdrv_remove_filter_or_cow_child(BlockDriverState *bs, | |
4889 | Transaction *tran) | |
4890 | { | |
4891 | bdrv_remove_file_or_backing_child(bs, bdrv_filter_or_cow_child(bs), tran); | |
4892 | } | |
4893 | ||
117caba9 VSO |
4894 | static int bdrv_replace_node_noperm(BlockDriverState *from, |
4895 | BlockDriverState *to, | |
4896 | bool auto_skip, Transaction *tran, | |
4897 | Error **errp) | |
4898 | { | |
4899 | BdrvChild *c, *next; | |
4900 | ||
4901 | QLIST_FOREACH_SAFE(c, &from->parents, next_parent, next) { | |
4902 | assert(c->bs == from); | |
4903 | if (!should_update_child(c, to)) { | |
4904 | if (auto_skip) { | |
4905 | continue; | |
4906 | } | |
4907 | error_setg(errp, "Should not change '%s' link to '%s'", | |
4908 | c->name, from->node_name); | |
4909 | return -EINVAL; | |
4910 | } | |
4911 | if (c->frozen) { | |
4912 | error_setg(errp, "Cannot change '%s' link to '%s'", | |
4913 | c->name, from->node_name); | |
4914 | return -EPERM; | |
4915 | } | |
4bf021db | 4916 | bdrv_replace_child_tran(c, to, tran); |
117caba9 VSO |
4917 | } |
4918 | ||
4919 | return 0; | |
4920 | } | |
4921 | ||
313274bb VSO |
4922 | /* |
4923 | * With auto_skip=true bdrv_replace_node_common skips updating from parents | |
4924 | * if it creates a parent-child relation loop or if parent is block-job. | |
4925 | * | |
4926 | * With auto_skip=false the error is returned if from has a parent which should | |
4927 | * not be updated. | |
3108a15c VSO |
4928 | * |
4929 | * With @detach_subchain=true @to must be in a backing chain of @from. In this | |
4930 | * case backing link of the cow-parent of @to is removed. | |
313274bb | 4931 | */ |
a1e708fc VSO |
4932 | static int bdrv_replace_node_common(BlockDriverState *from, |
4933 | BlockDriverState *to, | |
3108a15c VSO |
4934 | bool auto_skip, bool detach_subchain, |
4935 | Error **errp) | |
dd62f1ca | 4936 | { |
3bb0e298 VSO |
4937 | Transaction *tran = tran_new(); |
4938 | g_autoptr(GHashTable) found = NULL; | |
4939 | g_autoptr(GSList) refresh_list = NULL; | |
2d369d6e | 4940 | BlockDriverState *to_cow_parent = NULL; |
234ac1a9 KW |
4941 | int ret; |
4942 | ||
3108a15c VSO |
4943 | if (detach_subchain) { |
4944 | assert(bdrv_chain_contains(from, to)); | |
4945 | assert(from != to); | |
4946 | for (to_cow_parent = from; | |
4947 | bdrv_filter_or_cow_bs(to_cow_parent) != to; | |
4948 | to_cow_parent = bdrv_filter_or_cow_bs(to_cow_parent)) | |
4949 | { | |
4950 | ; | |
4951 | } | |
4952 | } | |
4953 | ||
234ac1a9 KW |
4954 | /* Make sure that @from doesn't go away until we have successfully attached |
4955 | * all of its parents to @to. */ | |
4956 | bdrv_ref(from); | |
dd62f1ca | 4957 | |
f871abd6 | 4958 | assert(qemu_get_current_aio_context() == qemu_get_aio_context()); |
30dd65f3 | 4959 | assert(bdrv_get_aio_context(from) == bdrv_get_aio_context(to)); |
f871abd6 KW |
4960 | bdrv_drained_begin(from); |
4961 | ||
3bb0e298 VSO |
4962 | /* |
4963 | * Do the replacement without permission update. | |
4964 | * Replacement may influence the permissions, we should calculate new | |
4965 | * permissions based on new graph. If we fail, we'll roll-back the | |
4966 | * replacement. | |
4967 | */ | |
117caba9 VSO |
4968 | ret = bdrv_replace_node_noperm(from, to, auto_skip, tran, errp); |
4969 | if (ret < 0) { | |
4970 | goto out; | |
234ac1a9 KW |
4971 | } |
4972 | ||
3108a15c VSO |
4973 | if (detach_subchain) { |
4974 | bdrv_remove_filter_or_cow_child(to_cow_parent, tran); | |
4975 | } | |
4976 | ||
3bb0e298 | 4977 | found = g_hash_table_new(NULL, NULL); |
234ac1a9 | 4978 | |
3bb0e298 VSO |
4979 | refresh_list = bdrv_topological_dfs(refresh_list, found, to); |
4980 | refresh_list = bdrv_topological_dfs(refresh_list, found, from); | |
9bd910e2 | 4981 | |
3bb0e298 VSO |
4982 | ret = bdrv_list_refresh_perms(refresh_list, NULL, tran, errp); |
4983 | if (ret < 0) { | |
4984 | goto out; | |
dd62f1ca | 4985 | } |
234ac1a9 | 4986 | |
a1e708fc VSO |
4987 | ret = 0; |
4988 | ||
234ac1a9 | 4989 | out: |
3bb0e298 VSO |
4990 | tran_finalize(tran, ret); |
4991 | ||
f871abd6 | 4992 | bdrv_drained_end(from); |
234ac1a9 | 4993 | bdrv_unref(from); |
a1e708fc VSO |
4994 | |
4995 | return ret; | |
dd62f1ca KW |
4996 | } |
4997 | ||
a1e708fc VSO |
4998 | int bdrv_replace_node(BlockDriverState *from, BlockDriverState *to, |
4999 | Error **errp) | |
313274bb | 5000 | { |
3108a15c VSO |
5001 | return bdrv_replace_node_common(from, to, true, false, errp); |
5002 | } | |
5003 | ||
5004 | int bdrv_drop_filter(BlockDriverState *bs, Error **errp) | |
5005 | { | |
5006 | return bdrv_replace_node_common(bs, bdrv_filter_or_cow_bs(bs), true, true, | |
5007 | errp); | |
313274bb VSO |
5008 | } |
5009 | ||
4ddc07ca PB |
5010 | /* |
5011 | * Add new bs contents at the top of an image chain while the chain is | |
5012 | * live, while keeping required fields on the top layer. | |
5013 | * | |
5014 | * This will modify the BlockDriverState fields, and swap contents | |
5015 | * between bs_new and bs_top. Both bs_new and bs_top are modified. | |
5016 | * | |
2272edcf VSO |
5017 | * bs_new must not be attached to a BlockBackend and must not have backing |
5018 | * child. | |
4ddc07ca PB |
5019 | * |
5020 | * This function does not create any image files. | |
5021 | */ | |
a1e708fc VSO |
5022 | int bdrv_append(BlockDriverState *bs_new, BlockDriverState *bs_top, |
5023 | Error **errp) | |
4ddc07ca | 5024 | { |
2272edcf VSO |
5025 | int ret; |
5026 | Transaction *tran = tran_new(); | |
5027 | ||
5028 | assert(!bs_new->backing); | |
5029 | ||
5030 | ret = bdrv_attach_child_noperm(bs_new, bs_top, "backing", | |
5031 | &child_of_bds, bdrv_backing_role(bs_new), | |
5032 | &bs_new->backing, tran, errp); | |
a1e708fc | 5033 | if (ret < 0) { |
2272edcf | 5034 | goto out; |
b2c2832c | 5035 | } |
dd62f1ca | 5036 | |
2272edcf | 5037 | ret = bdrv_replace_node_noperm(bs_top, bs_new, true, tran, errp); |
a1e708fc | 5038 | if (ret < 0) { |
2272edcf | 5039 | goto out; |
234ac1a9 | 5040 | } |
4ddc07ca | 5041 | |
2272edcf VSO |
5042 | ret = bdrv_refresh_perms(bs_new, errp); |
5043 | out: | |
5044 | tran_finalize(tran, ret); | |
5045 | ||
1e4c797c | 5046 | bdrv_refresh_limits(bs_top, NULL, NULL); |
2272edcf VSO |
5047 | |
5048 | return ret; | |
8802d1fd JC |
5049 | } |
5050 | ||
bd8f4c42 VSO |
5051 | /* Not for empty child */ |
5052 | int bdrv_replace_child_bs(BdrvChild *child, BlockDriverState *new_bs, | |
5053 | Error **errp) | |
5054 | { | |
5055 | int ret; | |
5056 | Transaction *tran = tran_new(); | |
5057 | g_autoptr(GHashTable) found = NULL; | |
5058 | g_autoptr(GSList) refresh_list = NULL; | |
5059 | BlockDriverState *old_bs = child->bs; | |
5060 | ||
5061 | bdrv_ref(old_bs); | |
5062 | bdrv_drained_begin(old_bs); | |
5063 | bdrv_drained_begin(new_bs); | |
5064 | ||
5065 | bdrv_replace_child_tran(child, new_bs, tran); | |
5066 | ||
5067 | found = g_hash_table_new(NULL, NULL); | |
5068 | refresh_list = bdrv_topological_dfs(refresh_list, found, old_bs); | |
5069 | refresh_list = bdrv_topological_dfs(refresh_list, found, new_bs); | |
5070 | ||
5071 | ret = bdrv_list_refresh_perms(refresh_list, NULL, tran, errp); | |
5072 | ||
5073 | tran_finalize(tran, ret); | |
5074 | ||
5075 | bdrv_drained_end(old_bs); | |
5076 | bdrv_drained_end(new_bs); | |
5077 | bdrv_unref(old_bs); | |
5078 | ||
5079 | return ret; | |
5080 | } | |
5081 | ||
4f6fd349 | 5082 | static void bdrv_delete(BlockDriverState *bs) |
b338082b | 5083 | { |
3718d8ab | 5084 | assert(bdrv_op_blocker_is_empty(bs)); |
4f6fd349 | 5085 | assert(!bs->refcnt); |
18846dee | 5086 | |
1b7bdbc1 | 5087 | /* remove from list, if necessary */ |
63eaaae0 KW |
5088 | if (bs->node_name[0] != '\0') { |
5089 | QTAILQ_REMOVE(&graph_bdrv_states, bs, node_list); | |
5090 | } | |
2c1d04e0 HR |
5091 | QTAILQ_REMOVE(&all_bdrv_states, bs, bs_list); |
5092 | ||
30c321f9 AK |
5093 | bdrv_close(bs); |
5094 | ||
7267c094 | 5095 | g_free(bs); |
fc01f7e7 FB |
5096 | } |
5097 | ||
8872ef78 AS |
5098 | BlockDriverState *bdrv_insert_node(BlockDriverState *bs, QDict *node_options, |
5099 | int flags, Error **errp) | |
5100 | { | |
5101 | BlockDriverState *new_node_bs; | |
5102 | Error *local_err = NULL; | |
5103 | ||
5104 | new_node_bs = bdrv_open(NULL, NULL, node_options, flags, errp); | |
5105 | if (new_node_bs == NULL) { | |
5106 | error_prepend(errp, "Could not create node: "); | |
5107 | return NULL; | |
5108 | } | |
5109 | ||
5110 | bdrv_drained_begin(bs); | |
5111 | bdrv_replace_node(bs, new_node_bs, &local_err); | |
5112 | bdrv_drained_end(bs); | |
5113 | ||
5114 | if (local_err) { | |
5115 | bdrv_unref(new_node_bs); | |
5116 | error_propagate(errp, local_err); | |
5117 | return NULL; | |
5118 | } | |
5119 | ||
5120 | return new_node_bs; | |
5121 | } | |
5122 | ||
e97fc193 AL |
5123 | /* |
5124 | * Run consistency checks on an image | |
5125 | * | |
e076f338 | 5126 | * Returns 0 if the check could be completed (it doesn't mean that the image is |
a1c7273b | 5127 | * free of errors) or -errno when an internal error occurred. The results of the |
e076f338 | 5128 | * check are stored in res. |
e97fc193 | 5129 | */ |
21c2283e VSO |
5130 | int coroutine_fn bdrv_co_check(BlockDriverState *bs, |
5131 | BdrvCheckResult *res, BdrvCheckMode fix) | |
e97fc193 | 5132 | { |
908bcd54 HR |
5133 | if (bs->drv == NULL) { |
5134 | return -ENOMEDIUM; | |
5135 | } | |
2fd61638 | 5136 | if (bs->drv->bdrv_co_check == NULL) { |
e97fc193 AL |
5137 | return -ENOTSUP; |
5138 | } | |
5139 | ||
e076f338 | 5140 | memset(res, 0, sizeof(*res)); |
2fd61638 PB |
5141 | return bs->drv->bdrv_co_check(bs, res, fix); |
5142 | } | |
5143 | ||
756e6736 KW |
5144 | /* |
5145 | * Return values: | |
5146 | * 0 - success | |
5147 | * -EINVAL - backing format specified, but no file | |
5148 | * -ENOSPC - can't update the backing file because no space is left in the | |
5149 | * image file header | |
5150 | * -ENOTSUP - format driver doesn't support changing the backing file | |
5151 | */ | |
e54ee1b3 | 5152 | int bdrv_change_backing_file(BlockDriverState *bs, const char *backing_file, |
497a30db | 5153 | const char *backing_fmt, bool require) |
756e6736 KW |
5154 | { |
5155 | BlockDriver *drv = bs->drv; | |
469ef350 | 5156 | int ret; |
756e6736 | 5157 | |
d470ad42 HR |
5158 | if (!drv) { |
5159 | return -ENOMEDIUM; | |
5160 | } | |
5161 | ||
5f377794 PB |
5162 | /* Backing file format doesn't make sense without a backing file */ |
5163 | if (backing_fmt && !backing_file) { | |
5164 | return -EINVAL; | |
5165 | } | |
5166 | ||
497a30db EB |
5167 | if (require && backing_file && !backing_fmt) { |
5168 | return -EINVAL; | |
e54ee1b3 EB |
5169 | } |
5170 | ||
756e6736 | 5171 | if (drv->bdrv_change_backing_file != NULL) { |
469ef350 | 5172 | ret = drv->bdrv_change_backing_file(bs, backing_file, backing_fmt); |
756e6736 | 5173 | } else { |
469ef350 | 5174 | ret = -ENOTSUP; |
756e6736 | 5175 | } |
469ef350 PB |
5176 | |
5177 | if (ret == 0) { | |
5178 | pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_file ?: ""); | |
5179 | pstrcpy(bs->backing_format, sizeof(bs->backing_format), backing_fmt ?: ""); | |
998c2019 HR |
5180 | pstrcpy(bs->auto_backing_file, sizeof(bs->auto_backing_file), |
5181 | backing_file ?: ""); | |
469ef350 PB |
5182 | } |
5183 | return ret; | |
756e6736 KW |
5184 | } |
5185 | ||
6ebdcee2 | 5186 | /* |
dcf3f9b2 HR |
5187 | * Finds the first non-filter node above bs in the chain between |
5188 | * active and bs. The returned node is either an immediate parent of | |
5189 | * bs, or there are only filter nodes between the two. | |
6ebdcee2 JC |
5190 | * |
5191 | * Returns NULL if bs is not found in active's image chain, | |
5192 | * or if active == bs. | |
4caf0fcd JC |
5193 | * |
5194 | * Returns the bottommost base image if bs == NULL. | |
6ebdcee2 JC |
5195 | */ |
5196 | BlockDriverState *bdrv_find_overlay(BlockDriverState *active, | |
5197 | BlockDriverState *bs) | |
5198 | { | |
dcf3f9b2 HR |
5199 | bs = bdrv_skip_filters(bs); |
5200 | active = bdrv_skip_filters(active); | |
5201 | ||
5202 | while (active) { | |
5203 | BlockDriverState *next = bdrv_backing_chain_next(active); | |
5204 | if (bs == next) { | |
5205 | return active; | |
5206 | } | |
5207 | active = next; | |
6ebdcee2 JC |
5208 | } |
5209 | ||
dcf3f9b2 | 5210 | return NULL; |
4caf0fcd | 5211 | } |
6ebdcee2 | 5212 | |
4caf0fcd JC |
5213 | /* Given a BDS, searches for the base layer. */ |
5214 | BlockDriverState *bdrv_find_base(BlockDriverState *bs) | |
5215 | { | |
5216 | return bdrv_find_overlay(bs, NULL); | |
6ebdcee2 JC |
5217 | } |
5218 | ||
2cad1ebe | 5219 | /* |
7b99a266 HR |
5220 | * Return true if at least one of the COW (backing) and filter links |
5221 | * between @bs and @base is frozen. @errp is set if that's the case. | |
0f0998f6 | 5222 | * @base must be reachable from @bs, or NULL. |
2cad1ebe AG |
5223 | */ |
5224 | bool bdrv_is_backing_chain_frozen(BlockDriverState *bs, BlockDriverState *base, | |
5225 | Error **errp) | |
5226 | { | |
5227 | BlockDriverState *i; | |
7b99a266 | 5228 | BdrvChild *child; |
2cad1ebe | 5229 | |
7b99a266 HR |
5230 | for (i = bs; i != base; i = child_bs(child)) { |
5231 | child = bdrv_filter_or_cow_child(i); | |
5232 | ||
5233 | if (child && child->frozen) { | |
2cad1ebe | 5234 | error_setg(errp, "Cannot change '%s' link from '%s' to '%s'", |
7b99a266 | 5235 | child->name, i->node_name, child->bs->node_name); |
2cad1ebe AG |
5236 | return true; |
5237 | } | |
5238 | } | |
5239 | ||
5240 | return false; | |
5241 | } | |
5242 | ||
5243 | /* | |
7b99a266 | 5244 | * Freeze all COW (backing) and filter links between @bs and @base. |
2cad1ebe AG |
5245 | * If any of the links is already frozen the operation is aborted and |
5246 | * none of the links are modified. | |
0f0998f6 | 5247 | * @base must be reachable from @bs, or NULL. |
2cad1ebe AG |
5248 | * Returns 0 on success. On failure returns < 0 and sets @errp. |
5249 | */ | |
5250 | int bdrv_freeze_backing_chain(BlockDriverState *bs, BlockDriverState *base, | |
5251 | Error **errp) | |
5252 | { | |
5253 | BlockDriverState *i; | |
7b99a266 | 5254 | BdrvChild *child; |
2cad1ebe AG |
5255 | |
5256 | if (bdrv_is_backing_chain_frozen(bs, base, errp)) { | |
5257 | return -EPERM; | |
5258 | } | |
5259 | ||
7b99a266 HR |
5260 | for (i = bs; i != base; i = child_bs(child)) { |
5261 | child = bdrv_filter_or_cow_child(i); | |
5262 | if (child && child->bs->never_freeze) { | |
e5182c1c | 5263 | error_setg(errp, "Cannot freeze '%s' link to '%s'", |
7b99a266 | 5264 | child->name, child->bs->node_name); |
e5182c1c HR |
5265 | return -EPERM; |
5266 | } | |
5267 | } | |
5268 | ||
7b99a266 HR |
5269 | for (i = bs; i != base; i = child_bs(child)) { |
5270 | child = bdrv_filter_or_cow_child(i); | |
5271 | if (child) { | |
5272 | child->frozen = true; | |
0f0998f6 | 5273 | } |
2cad1ebe AG |
5274 | } |
5275 | ||
5276 | return 0; | |
5277 | } | |
5278 | ||
5279 | /* | |
7b99a266 HR |
5280 | * Unfreeze all COW (backing) and filter links between @bs and @base. |
5281 | * The caller must ensure that all links are frozen before using this | |
5282 | * function. | |
0f0998f6 | 5283 | * @base must be reachable from @bs, or NULL. |
2cad1ebe AG |
5284 | */ |
5285 | void bdrv_unfreeze_backing_chain(BlockDriverState *bs, BlockDriverState *base) | |
5286 | { | |
5287 | BlockDriverState *i; | |
7b99a266 | 5288 | BdrvChild *child; |
2cad1ebe | 5289 | |
7b99a266 HR |
5290 | for (i = bs; i != base; i = child_bs(child)) { |
5291 | child = bdrv_filter_or_cow_child(i); | |
5292 | if (child) { | |
5293 | assert(child->frozen); | |
5294 | child->frozen = false; | |
0f0998f6 | 5295 | } |
2cad1ebe AG |
5296 | } |
5297 | } | |
5298 | ||
6ebdcee2 JC |
5299 | /* |
5300 | * Drops images above 'base' up to and including 'top', and sets the image | |
5301 | * above 'top' to have base as its backing file. | |
5302 | * | |
5303 | * Requires that the overlay to 'top' is opened r/w, so that the backing file | |
5304 | * information in 'bs' can be properly updated. | |
5305 | * | |
5306 | * E.g., this will convert the following chain: | |
5307 | * bottom <- base <- intermediate <- top <- active | |
5308 | * | |
5309 | * to | |
5310 | * | |
5311 | * bottom <- base <- active | |
5312 | * | |
5313 | * It is allowed for bottom==base, in which case it converts: | |
5314 | * | |
5315 | * base <- intermediate <- top <- active | |
5316 | * | |
5317 | * to | |
5318 | * | |
5319 | * base <- active | |
5320 | * | |
54e26900 JC |
5321 | * If backing_file_str is non-NULL, it will be used when modifying top's |
5322 | * overlay image metadata. | |
5323 | * | |
6ebdcee2 JC |
5324 | * Error conditions: |
5325 | * if active == top, that is considered an error | |
5326 | * | |
5327 | */ | |
bde70715 KW |
5328 | int bdrv_drop_intermediate(BlockDriverState *top, BlockDriverState *base, |
5329 | const char *backing_file_str) | |
6ebdcee2 | 5330 | { |
6bd858b3 AG |
5331 | BlockDriverState *explicit_top = top; |
5332 | bool update_inherits_from; | |
d669ed6a | 5333 | BdrvChild *c; |
12fa4af6 | 5334 | Error *local_err = NULL; |
6ebdcee2 | 5335 | int ret = -EIO; |
d669ed6a VSO |
5336 | g_autoptr(GSList) updated_children = NULL; |
5337 | GSList *p; | |
6ebdcee2 | 5338 | |
6858eba0 | 5339 | bdrv_ref(top); |
637d54a5 | 5340 | bdrv_subtree_drained_begin(top); |
6858eba0 | 5341 | |
6ebdcee2 JC |
5342 | if (!top->drv || !base->drv) { |
5343 | goto exit; | |
5344 | } | |
5345 | ||
5db15a57 KW |
5346 | /* Make sure that base is in the backing chain of top */ |
5347 | if (!bdrv_chain_contains(top, base)) { | |
6ebdcee2 JC |
5348 | goto exit; |
5349 | } | |
5350 | ||
6bd858b3 AG |
5351 | /* If 'base' recursively inherits from 'top' then we should set |
5352 | * base->inherits_from to top->inherits_from after 'top' and all | |
5353 | * other intermediate nodes have been dropped. | |
5354 | * If 'top' is an implicit node (e.g. "commit_top") we should skip | |
5355 | * it because no one inherits from it. We use explicit_top for that. */ | |
dcf3f9b2 | 5356 | explicit_top = bdrv_skip_implicit_filters(explicit_top); |
6bd858b3 AG |
5357 | update_inherits_from = bdrv_inherits_from_recursive(base, explicit_top); |
5358 | ||
6ebdcee2 | 5359 | /* success - we can delete the intermediate states, and link top->base */ |
bde70715 KW |
5360 | /* TODO Check graph modification op blockers (BLK_PERM_GRAPH_MOD) once |
5361 | * we've figured out how they should work. */ | |
f30c66ba HR |
5362 | if (!backing_file_str) { |
5363 | bdrv_refresh_filename(base); | |
5364 | backing_file_str = base->filename; | |
5365 | } | |
61f09cea | 5366 | |
d669ed6a VSO |
5367 | QLIST_FOREACH(c, &top->parents, next_parent) { |
5368 | updated_children = g_slist_prepend(updated_children, c); | |
5369 | } | |
5370 | ||
3108a15c VSO |
5371 | /* |
5372 | * It seems correct to pass detach_subchain=true here, but it triggers | |
5373 | * one more yet not fixed bug, when due to nested aio_poll loop we switch to | |
5374 | * another drained section, which modify the graph (for example, removing | |
5375 | * the child, which we keep in updated_children list). So, it's a TODO. | |
5376 | * | |
5377 | * Note, bug triggered if pass detach_subchain=true here and run | |
5378 | * test-bdrv-drain. test_drop_intermediate_poll() test-case will crash. | |
5379 | * That's a FIXME. | |
5380 | */ | |
5381 | bdrv_replace_node_common(top, base, false, false, &local_err); | |
d669ed6a VSO |
5382 | if (local_err) { |
5383 | error_report_err(local_err); | |
5384 | goto exit; | |
5385 | } | |
5386 | ||
5387 | for (p = updated_children; p; p = p->next) { | |
5388 | c = p->data; | |
12fa4af6 | 5389 | |
bd86fb99 HR |
5390 | if (c->klass->update_filename) { |
5391 | ret = c->klass->update_filename(c, base, backing_file_str, | |
5392 | &local_err); | |
61f09cea | 5393 | if (ret < 0) { |
d669ed6a VSO |
5394 | /* |
5395 | * TODO: Actually, we want to rollback all previous iterations | |
5396 | * of this loop, and (which is almost impossible) previous | |
5397 | * bdrv_replace_node()... | |
5398 | * | |
5399 | * Note, that c->klass->update_filename may lead to permission | |
5400 | * update, so it's a bad idea to call it inside permission | |
5401 | * update transaction of bdrv_replace_node. | |
5402 | */ | |
61f09cea KW |
5403 | error_report_err(local_err); |
5404 | goto exit; | |
5405 | } | |
5406 | } | |
12fa4af6 | 5407 | } |
6ebdcee2 | 5408 | |
6bd858b3 AG |
5409 | if (update_inherits_from) { |
5410 | base->inherits_from = explicit_top->inherits_from; | |
5411 | } | |
5412 | ||
6ebdcee2 | 5413 | ret = 0; |
6ebdcee2 | 5414 | exit: |
637d54a5 | 5415 | bdrv_subtree_drained_end(top); |
6858eba0 | 5416 | bdrv_unref(top); |
6ebdcee2 JC |
5417 | return ret; |
5418 | } | |
5419 | ||
081e4650 HR |
5420 | /** |
5421 | * Implementation of BlockDriver.bdrv_get_allocated_file_size() that | |
5422 | * sums the size of all data-bearing children. (This excludes backing | |
5423 | * children.) | |
5424 | */ | |
5425 | static int64_t bdrv_sum_allocated_file_size(BlockDriverState *bs) | |
5426 | { | |
5427 | BdrvChild *child; | |
5428 | int64_t child_size, sum = 0; | |
5429 | ||
5430 | QLIST_FOREACH(child, &bs->children, next) { | |
5431 | if (child->role & (BDRV_CHILD_DATA | BDRV_CHILD_METADATA | | |
5432 | BDRV_CHILD_FILTERED)) | |
5433 | { | |
5434 | child_size = bdrv_get_allocated_file_size(child->bs); | |
5435 | if (child_size < 0) { | |
5436 | return child_size; | |
5437 | } | |
5438 | sum += child_size; | |
5439 | } | |
5440 | } | |
5441 | ||
5442 | return sum; | |
5443 | } | |
5444 | ||
61007b31 SH |
5445 | /** |
5446 | * Length of a allocated file in bytes. Sparse files are counted by actual | |
5447 | * allocated space. Return < 0 if error or unknown. | |
5448 | */ | |
5449 | int64_t bdrv_get_allocated_file_size(BlockDriverState *bs) | |
71d0770c | 5450 | { |
61007b31 SH |
5451 | BlockDriver *drv = bs->drv; |
5452 | if (!drv) { | |
5453 | return -ENOMEDIUM; | |
8f4754ed | 5454 | } |
61007b31 SH |
5455 | if (drv->bdrv_get_allocated_file_size) { |
5456 | return drv->bdrv_get_allocated_file_size(bs); | |
5457 | } | |
081e4650 HR |
5458 | |
5459 | if (drv->bdrv_file_open) { | |
5460 | /* | |
5461 | * Protocol drivers default to -ENOTSUP (most of their data is | |
5462 | * not stored in any of their children (if they even have any), | |
5463 | * so there is no generic way to figure it out). | |
5464 | */ | |
5465 | return -ENOTSUP; | |
5466 | } else if (drv->is_filter) { | |
5467 | /* Filter drivers default to the size of their filtered child */ | |
5468 | return bdrv_get_allocated_file_size(bdrv_filter_bs(bs)); | |
5469 | } else { | |
5470 | /* Other drivers default to summing their children's sizes */ | |
5471 | return bdrv_sum_allocated_file_size(bs); | |
1c9805a3 SH |
5472 | } |
5473 | } | |
e7a8a783 | 5474 | |
90880ff1 SH |
5475 | /* |
5476 | * bdrv_measure: | |
5477 | * @drv: Format driver | |
5478 | * @opts: Creation options for new image | |
5479 | * @in_bs: Existing image containing data for new image (may be NULL) | |
5480 | * @errp: Error object | |
5481 | * Returns: A #BlockMeasureInfo (free using qapi_free_BlockMeasureInfo()) | |
5482 | * or NULL on error | |
5483 | * | |
5484 | * Calculate file size required to create a new image. | |
5485 | * | |
5486 | * If @in_bs is given then space for allocated clusters and zero clusters | |
5487 | * from that image are included in the calculation. If @opts contains a | |
5488 | * backing file that is shared by @in_bs then backing clusters may be omitted | |
5489 | * from the calculation. | |
5490 | * | |
5491 | * If @in_bs is NULL then the calculation includes no allocated clusters | |
5492 | * unless a preallocation option is given in @opts. | |
5493 | * | |
5494 | * Note that @in_bs may use a different BlockDriver from @drv. | |
5495 | * | |
5496 | * If an error occurs the @errp pointer is set. | |
5497 | */ | |
5498 | BlockMeasureInfo *bdrv_measure(BlockDriver *drv, QemuOpts *opts, | |
5499 | BlockDriverState *in_bs, Error **errp) | |
5500 | { | |
5501 | if (!drv->bdrv_measure) { | |
5502 | error_setg(errp, "Block driver '%s' does not support size measurement", | |
5503 | drv->format_name); | |
5504 | return NULL; | |
5505 | } | |
5506 | ||
5507 | return drv->bdrv_measure(opts, in_bs, errp); | |
5508 | } | |
5509 | ||
61007b31 SH |
5510 | /** |
5511 | * Return number of sectors on success, -errno on error. | |
1c9805a3 | 5512 | */ |
61007b31 | 5513 | int64_t bdrv_nb_sectors(BlockDriverState *bs) |
1c9805a3 | 5514 | { |
61007b31 | 5515 | BlockDriver *drv = bs->drv; |
498e386c | 5516 | |
61007b31 SH |
5517 | if (!drv) |
5518 | return -ENOMEDIUM; | |
2572b37a | 5519 | |
61007b31 SH |
5520 | if (drv->has_variable_length) { |
5521 | int ret = refresh_total_sectors(bs, bs->total_sectors); | |
5522 | if (ret < 0) { | |
5523 | return ret; | |
1c9805a3 SH |
5524 | } |
5525 | } | |
61007b31 | 5526 | return bs->total_sectors; |
1c9805a3 | 5527 | } |
b338082b | 5528 | |
61007b31 SH |
5529 | /** |
5530 | * Return length in bytes on success, -errno on error. | |
5531 | * The length is always a multiple of BDRV_SECTOR_SIZE. | |
8d3b1a2d | 5532 | */ |
61007b31 | 5533 | int64_t bdrv_getlength(BlockDriverState *bs) |
8d3b1a2d | 5534 | { |
61007b31 | 5535 | int64_t ret = bdrv_nb_sectors(bs); |
8d3b1a2d | 5536 | |
122860ba EB |
5537 | if (ret < 0) { |
5538 | return ret; | |
5539 | } | |
5540 | if (ret > INT64_MAX / BDRV_SECTOR_SIZE) { | |
5541 | return -EFBIG; | |
5542 | } | |
5543 | return ret * BDRV_SECTOR_SIZE; | |
fc01f7e7 FB |
5544 | } |
5545 | ||
61007b31 SH |
5546 | /* return 0 as number of sectors if no device present or error */ |
5547 | void bdrv_get_geometry(BlockDriverState *bs, uint64_t *nb_sectors_ptr) | |
07d27a44 | 5548 | { |
61007b31 | 5549 | int64_t nb_sectors = bdrv_nb_sectors(bs); |
07d27a44 | 5550 | |
61007b31 | 5551 | *nb_sectors_ptr = nb_sectors < 0 ? 0 : nb_sectors; |
07d27a44 MA |
5552 | } |
5553 | ||
54115412 | 5554 | bool bdrv_is_sg(BlockDriverState *bs) |
f08145fe | 5555 | { |
61007b31 | 5556 | return bs->sg; |
f08145fe KW |
5557 | } |
5558 | ||
ae23f786 HR |
5559 | /** |
5560 | * Return whether the given node supports compressed writes. | |
5561 | */ | |
5562 | bool bdrv_supports_compressed_writes(BlockDriverState *bs) | |
5563 | { | |
5564 | BlockDriverState *filtered; | |
5565 | ||
5566 | if (!bs->drv || !block_driver_can_compress(bs->drv)) { | |
5567 | return false; | |
5568 | } | |
5569 | ||
5570 | filtered = bdrv_filter_bs(bs); | |
5571 | if (filtered) { | |
5572 | /* | |
5573 | * Filters can only forward compressed writes, so we have to | |
5574 | * check the child. | |
5575 | */ | |
5576 | return bdrv_supports_compressed_writes(filtered); | |
5577 | } | |
5578 | ||
5579 | return true; | |
5580 | } | |
5581 | ||
61007b31 | 5582 | const char *bdrv_get_format_name(BlockDriverState *bs) |
40b4f539 | 5583 | { |
61007b31 | 5584 | return bs->drv ? bs->drv->format_name : NULL; |
40b4f539 KW |
5585 | } |
5586 | ||
61007b31 | 5587 | static int qsort_strcmp(const void *a, const void *b) |
40b4f539 | 5588 | { |
ceff5bd7 | 5589 | return strcmp(*(char *const *)a, *(char *const *)b); |
40b4f539 KW |
5590 | } |
5591 | ||
61007b31 | 5592 | void bdrv_iterate_format(void (*it)(void *opaque, const char *name), |
9ac404c5 | 5593 | void *opaque, bool read_only) |
40b4f539 | 5594 | { |
61007b31 SH |
5595 | BlockDriver *drv; |
5596 | int count = 0; | |
5597 | int i; | |
5598 | const char **formats = NULL; | |
40b4f539 | 5599 | |
61007b31 SH |
5600 | QLIST_FOREACH(drv, &bdrv_drivers, list) { |
5601 | if (drv->format_name) { | |
5602 | bool found = false; | |
5603 | int i = count; | |
9ac404c5 AS |
5604 | |
5605 | if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv, read_only)) { | |
5606 | continue; | |
5607 | } | |
5608 | ||
61007b31 SH |
5609 | while (formats && i && !found) { |
5610 | found = !strcmp(formats[--i], drv->format_name); | |
5611 | } | |
e2a305fb | 5612 | |
61007b31 SH |
5613 | if (!found) { |
5614 | formats = g_renew(const char *, formats, count + 1); | |
5615 | formats[count++] = drv->format_name; | |
5616 | } | |
6c5a42ac | 5617 | } |
61007b31 | 5618 | } |
6c5a42ac | 5619 | |
eb0df69f HR |
5620 | for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); i++) { |
5621 | const char *format_name = block_driver_modules[i].format_name; | |
5622 | ||
5623 | if (format_name) { | |
5624 | bool found = false; | |
5625 | int j = count; | |
5626 | ||
9ac404c5 AS |
5627 | if (use_bdrv_whitelist && |
5628 | !bdrv_format_is_whitelisted(format_name, read_only)) { | |
5629 | continue; | |
5630 | } | |
5631 | ||
eb0df69f HR |
5632 | while (formats && j && !found) { |
5633 | found = !strcmp(formats[--j], format_name); | |
5634 | } | |
5635 | ||
5636 | if (!found) { | |
5637 | formats = g_renew(const char *, formats, count + 1); | |
5638 | formats[count++] = format_name; | |
5639 | } | |
5640 | } | |
5641 | } | |
5642 | ||
61007b31 | 5643 | qsort(formats, count, sizeof(formats[0]), qsort_strcmp); |
40b4f539 | 5644 | |
61007b31 SH |
5645 | for (i = 0; i < count; i++) { |
5646 | it(opaque, formats[i]); | |
5647 | } | |
40b4f539 | 5648 | |
61007b31 SH |
5649 | g_free(formats); |
5650 | } | |
40b4f539 | 5651 | |
61007b31 SH |
5652 | /* This function is to find a node in the bs graph */ |
5653 | BlockDriverState *bdrv_find_node(const char *node_name) | |
5654 | { | |
5655 | BlockDriverState *bs; | |
391827eb | 5656 | |
61007b31 | 5657 | assert(node_name); |
40b4f539 | 5658 | |
61007b31 SH |
5659 | QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) { |
5660 | if (!strcmp(node_name, bs->node_name)) { | |
5661 | return bs; | |
40b4f539 KW |
5662 | } |
5663 | } | |
61007b31 | 5664 | return NULL; |
40b4f539 KW |
5665 | } |
5666 | ||
61007b31 | 5667 | /* Put this QMP function here so it can access the static graph_bdrv_states. */ |
facda544 PK |
5668 | BlockDeviceInfoList *bdrv_named_nodes_list(bool flat, |
5669 | Error **errp) | |
40b4f539 | 5670 | { |
9812e712 | 5671 | BlockDeviceInfoList *list; |
61007b31 | 5672 | BlockDriverState *bs; |
40b4f539 | 5673 | |
61007b31 SH |
5674 | list = NULL; |
5675 | QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) { | |
facda544 | 5676 | BlockDeviceInfo *info = bdrv_block_device_info(NULL, bs, flat, errp); |
61007b31 SH |
5677 | if (!info) { |
5678 | qapi_free_BlockDeviceInfoList(list); | |
5679 | return NULL; | |
301db7c2 | 5680 | } |
9812e712 | 5681 | QAPI_LIST_PREPEND(list, info); |
301db7c2 RH |
5682 | } |
5683 | ||
61007b31 SH |
5684 | return list; |
5685 | } | |
40b4f539 | 5686 | |
5d3b4e99 VSO |
5687 | typedef struct XDbgBlockGraphConstructor { |
5688 | XDbgBlockGraph *graph; | |
5689 | GHashTable *graph_nodes; | |
5690 | } XDbgBlockGraphConstructor; | |
5691 | ||
5692 | static XDbgBlockGraphConstructor *xdbg_graph_new(void) | |
5693 | { | |
5694 | XDbgBlockGraphConstructor *gr = g_new(XDbgBlockGraphConstructor, 1); | |
5695 | ||
5696 | gr->graph = g_new0(XDbgBlockGraph, 1); | |
5697 | gr->graph_nodes = g_hash_table_new(NULL, NULL); | |
5698 | ||
5699 | return gr; | |
5700 | } | |
5701 | ||
5702 | static XDbgBlockGraph *xdbg_graph_finalize(XDbgBlockGraphConstructor *gr) | |
5703 | { | |
5704 | XDbgBlockGraph *graph = gr->graph; | |
5705 | ||
5706 | g_hash_table_destroy(gr->graph_nodes); | |
5707 | g_free(gr); | |
5708 | ||
5709 | return graph; | |
5710 | } | |
5711 | ||
5712 | static uintptr_t xdbg_graph_node_num(XDbgBlockGraphConstructor *gr, void *node) | |
5713 | { | |
5714 | uintptr_t ret = (uintptr_t)g_hash_table_lookup(gr->graph_nodes, node); | |
5715 | ||
5716 | if (ret != 0) { | |
5717 | return ret; | |
5718 | } | |
5719 | ||
5720 | /* | |
5721 | * Start counting from 1, not 0, because 0 interferes with not-found (NULL) | |
5722 | * answer of g_hash_table_lookup. | |
5723 | */ | |
5724 | ret = g_hash_table_size(gr->graph_nodes) + 1; | |
5725 | g_hash_table_insert(gr->graph_nodes, node, (void *)ret); | |
5726 | ||
5727 | return ret; | |
5728 | } | |
5729 | ||
5730 | static void xdbg_graph_add_node(XDbgBlockGraphConstructor *gr, void *node, | |
5731 | XDbgBlockGraphNodeType type, const char *name) | |
5732 | { | |
5733 | XDbgBlockGraphNode *n; | |
5734 | ||
5735 | n = g_new0(XDbgBlockGraphNode, 1); | |
5736 | ||
5737 | n->id = xdbg_graph_node_num(gr, node); | |
5738 | n->type = type; | |
5739 | n->name = g_strdup(name); | |
5740 | ||
9812e712 | 5741 | QAPI_LIST_PREPEND(gr->graph->nodes, n); |
5d3b4e99 VSO |
5742 | } |
5743 | ||
5744 | static void xdbg_graph_add_edge(XDbgBlockGraphConstructor *gr, void *parent, | |
5745 | const BdrvChild *child) | |
5746 | { | |
cdb1cec8 | 5747 | BlockPermission qapi_perm; |
5d3b4e99 VSO |
5748 | XDbgBlockGraphEdge *edge; |
5749 | ||
5d3b4e99 VSO |
5750 | edge = g_new0(XDbgBlockGraphEdge, 1); |
5751 | ||
5752 | edge->parent = xdbg_graph_node_num(gr, parent); | |
5753 | edge->child = xdbg_graph_node_num(gr, child->bs); | |
5754 | edge->name = g_strdup(child->name); | |
5755 | ||
cdb1cec8 HR |
5756 | for (qapi_perm = 0; qapi_perm < BLOCK_PERMISSION__MAX; qapi_perm++) { |
5757 | uint64_t flag = bdrv_qapi_perm_to_blk_perm(qapi_perm); | |
5758 | ||
5759 | if (flag & child->perm) { | |
9812e712 | 5760 | QAPI_LIST_PREPEND(edge->perm, qapi_perm); |
5d3b4e99 | 5761 | } |
cdb1cec8 | 5762 | if (flag & child->shared_perm) { |
9812e712 | 5763 | QAPI_LIST_PREPEND(edge->shared_perm, qapi_perm); |
5d3b4e99 VSO |
5764 | } |
5765 | } | |
5766 | ||
9812e712 | 5767 | QAPI_LIST_PREPEND(gr->graph->edges, edge); |
5d3b4e99 VSO |
5768 | } |
5769 | ||
5770 | ||
5771 | XDbgBlockGraph *bdrv_get_xdbg_block_graph(Error **errp) | |
5772 | { | |
5773 | BlockBackend *blk; | |
5774 | BlockJob *job; | |
5775 | BlockDriverState *bs; | |
5776 | BdrvChild *child; | |
5777 | XDbgBlockGraphConstructor *gr = xdbg_graph_new(); | |
5778 | ||
5779 | for (blk = blk_all_next(NULL); blk; blk = blk_all_next(blk)) { | |
5780 | char *allocated_name = NULL; | |
5781 | const char *name = blk_name(blk); | |
5782 | ||
5783 | if (!*name) { | |
5784 | name = allocated_name = blk_get_attached_dev_id(blk); | |
5785 | } | |
5786 | xdbg_graph_add_node(gr, blk, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_BACKEND, | |
5787 | name); | |
5788 | g_free(allocated_name); | |
5789 | if (blk_root(blk)) { | |
5790 | xdbg_graph_add_edge(gr, blk, blk_root(blk)); | |
5791 | } | |
5792 | } | |
5793 | ||
5794 | for (job = block_job_next(NULL); job; job = block_job_next(job)) { | |
5795 | GSList *el; | |
5796 | ||
5797 | xdbg_graph_add_node(gr, job, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_JOB, | |
5798 | job->job.id); | |
5799 | for (el = job->nodes; el; el = el->next) { | |
5800 | xdbg_graph_add_edge(gr, job, (BdrvChild *)el->data); | |
5801 | } | |
5802 | } | |
5803 | ||
5804 | QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) { | |
5805 | xdbg_graph_add_node(gr, bs, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_DRIVER, | |
5806 | bs->node_name); | |
5807 | QLIST_FOREACH(child, &bs->children, next) { | |
5808 | xdbg_graph_add_edge(gr, bs, child); | |
5809 | } | |
5810 | } | |
5811 | ||
5812 | return xdbg_graph_finalize(gr); | |
5813 | } | |
5814 | ||
61007b31 SH |
5815 | BlockDriverState *bdrv_lookup_bs(const char *device, |
5816 | const char *node_name, | |
5817 | Error **errp) | |
5818 | { | |
5819 | BlockBackend *blk; | |
5820 | BlockDriverState *bs; | |
40b4f539 | 5821 | |
61007b31 SH |
5822 | if (device) { |
5823 | blk = blk_by_name(device); | |
40b4f539 | 5824 | |
61007b31 | 5825 | if (blk) { |
9f4ed6fb AG |
5826 | bs = blk_bs(blk); |
5827 | if (!bs) { | |
5433c24f | 5828 | error_setg(errp, "Device '%s' has no medium", device); |
5433c24f HR |
5829 | } |
5830 | ||
9f4ed6fb | 5831 | return bs; |
61007b31 SH |
5832 | } |
5833 | } | |
40b4f539 | 5834 | |
61007b31 SH |
5835 | if (node_name) { |
5836 | bs = bdrv_find_node(node_name); | |
6d519a5f | 5837 | |
61007b31 SH |
5838 | if (bs) { |
5839 | return bs; | |
5840 | } | |
40b4f539 KW |
5841 | } |
5842 | ||
785ec4b1 | 5843 | error_setg(errp, "Cannot find device=\'%s\' nor node-name=\'%s\'", |
61007b31 SH |
5844 | device ? device : "", |
5845 | node_name ? node_name : ""); | |
5846 | return NULL; | |
40b4f539 KW |
5847 | } |
5848 | ||
61007b31 SH |
5849 | /* If 'base' is in the same chain as 'top', return true. Otherwise, |
5850 | * return false. If either argument is NULL, return false. */ | |
5851 | bool bdrv_chain_contains(BlockDriverState *top, BlockDriverState *base) | |
83f64091 | 5852 | { |
61007b31 | 5853 | while (top && top != base) { |
dcf3f9b2 | 5854 | top = bdrv_filter_or_cow_bs(top); |
02c50efe | 5855 | } |
61007b31 SH |
5856 | |
5857 | return top != NULL; | |
02c50efe FZ |
5858 | } |
5859 | ||
61007b31 | 5860 | BlockDriverState *bdrv_next_node(BlockDriverState *bs) |
02c50efe | 5861 | { |
61007b31 SH |
5862 | if (!bs) { |
5863 | return QTAILQ_FIRST(&graph_bdrv_states); | |
02c50efe | 5864 | } |
61007b31 | 5865 | return QTAILQ_NEXT(bs, node_list); |
83f64091 FB |
5866 | } |
5867 | ||
0f12264e KW |
5868 | BlockDriverState *bdrv_next_all_states(BlockDriverState *bs) |
5869 | { | |
5870 | if (!bs) { | |
5871 | return QTAILQ_FIRST(&all_bdrv_states); | |
5872 | } | |
5873 | return QTAILQ_NEXT(bs, bs_list); | |
5874 | } | |
5875 | ||
61007b31 | 5876 | const char *bdrv_get_node_name(const BlockDriverState *bs) |
83f64091 | 5877 | { |
61007b31 | 5878 | return bs->node_name; |
beac80cd FB |
5879 | } |
5880 | ||
1f0c461b | 5881 | const char *bdrv_get_parent_name(const BlockDriverState *bs) |
4c265bf9 KW |
5882 | { |
5883 | BdrvChild *c; | |
5884 | const char *name; | |
5885 | ||
5886 | /* If multiple parents have a name, just pick the first one. */ | |
5887 | QLIST_FOREACH(c, &bs->parents, next_parent) { | |
bd86fb99 HR |
5888 | if (c->klass->get_name) { |
5889 | name = c->klass->get_name(c); | |
4c265bf9 KW |
5890 | if (name && *name) { |
5891 | return name; | |
5892 | } | |
5893 | } | |
5894 | } | |
5895 | ||
5896 | return NULL; | |
5897 | } | |
5898 | ||
61007b31 SH |
5899 | /* TODO check what callers really want: bs->node_name or blk_name() */ |
5900 | const char *bdrv_get_device_name(const BlockDriverState *bs) | |
beac80cd | 5901 | { |
4c265bf9 | 5902 | return bdrv_get_parent_name(bs) ?: ""; |
f141eafe | 5903 | } |
83f64091 | 5904 | |
61007b31 SH |
5905 | /* This can be used to identify nodes that might not have a device |
5906 | * name associated. Since node and device names live in the same | |
5907 | * namespace, the result is unambiguous. The exception is if both are | |
5908 | * absent, then this returns an empty (non-null) string. */ | |
5909 | const char *bdrv_get_device_or_node_name(const BlockDriverState *bs) | |
f141eafe | 5910 | { |
4c265bf9 | 5911 | return bdrv_get_parent_name(bs) ?: bs->node_name; |
beac80cd | 5912 | } |
beac80cd | 5913 | |
61007b31 | 5914 | int bdrv_get_flags(BlockDriverState *bs) |
0b5a2445 | 5915 | { |
61007b31 | 5916 | return bs->open_flags; |
0b5a2445 PB |
5917 | } |
5918 | ||
61007b31 | 5919 | int bdrv_has_zero_init_1(BlockDriverState *bs) |
68485420 | 5920 | { |
61007b31 | 5921 | return 1; |
0b5a2445 PB |
5922 | } |
5923 | ||
61007b31 | 5924 | int bdrv_has_zero_init(BlockDriverState *bs) |
0b5a2445 | 5925 | { |
93393e69 HR |
5926 | BlockDriverState *filtered; |
5927 | ||
d470ad42 HR |
5928 | if (!bs->drv) { |
5929 | return 0; | |
5930 | } | |
0b5a2445 | 5931 | |
61007b31 SH |
5932 | /* If BS is a copy on write image, it is initialized to |
5933 | the contents of the base image, which may not be zeroes. */ | |
34778172 | 5934 | if (bdrv_cow_child(bs)) { |
61007b31 SH |
5935 | return 0; |
5936 | } | |
5937 | if (bs->drv->bdrv_has_zero_init) { | |
5938 | return bs->drv->bdrv_has_zero_init(bs); | |
0b5a2445 | 5939 | } |
93393e69 HR |
5940 | |
5941 | filtered = bdrv_filter_bs(bs); | |
5942 | if (filtered) { | |
5943 | return bdrv_has_zero_init(filtered); | |
5a612c00 | 5944 | } |
61007b31 SH |
5945 | |
5946 | /* safe default */ | |
5947 | return 0; | |
68485420 KW |
5948 | } |
5949 | ||
61007b31 | 5950 | bool bdrv_can_write_zeroes_with_unmap(BlockDriverState *bs) |
68485420 | 5951 | { |
2f0342ef | 5952 | if (!(bs->open_flags & BDRV_O_UNMAP)) { |
61007b31 SH |
5953 | return false; |
5954 | } | |
68485420 | 5955 | |
e24d813b | 5956 | return bs->supported_zero_flags & BDRV_REQ_MAY_UNMAP; |
68485420 KW |
5957 | } |
5958 | ||
61007b31 SH |
5959 | void bdrv_get_backing_filename(BlockDriverState *bs, |
5960 | char *filename, int filename_size) | |
016f5cf6 | 5961 | { |
61007b31 SH |
5962 | pstrcpy(filename, filename_size, bs->backing_file); |
5963 | } | |
d318aea9 | 5964 | |
61007b31 SH |
5965 | int bdrv_get_info(BlockDriverState *bs, BlockDriverInfo *bdi) |
5966 | { | |
8b117001 | 5967 | int ret; |
61007b31 | 5968 | BlockDriver *drv = bs->drv; |
5a612c00 MP |
5969 | /* if bs->drv == NULL, bs is closed, so there's nothing to do here */ |
5970 | if (!drv) { | |
61007b31 | 5971 | return -ENOMEDIUM; |
5a612c00 MP |
5972 | } |
5973 | if (!drv->bdrv_get_info) { | |
93393e69 HR |
5974 | BlockDriverState *filtered = bdrv_filter_bs(bs); |
5975 | if (filtered) { | |
5976 | return bdrv_get_info(filtered, bdi); | |
5a612c00 | 5977 | } |
61007b31 | 5978 | return -ENOTSUP; |
5a612c00 | 5979 | } |
61007b31 | 5980 | memset(bdi, 0, sizeof(*bdi)); |
8b117001 VSO |
5981 | ret = drv->bdrv_get_info(bs, bdi); |
5982 | if (ret < 0) { | |
5983 | return ret; | |
5984 | } | |
5985 | ||
5986 | if (bdi->cluster_size > BDRV_MAX_ALIGNMENT) { | |
5987 | return -EINVAL; | |
5988 | } | |
5989 | ||
5990 | return 0; | |
61007b31 | 5991 | } |
016f5cf6 | 5992 | |
1bf6e9ca AS |
5993 | ImageInfoSpecific *bdrv_get_specific_info(BlockDriverState *bs, |
5994 | Error **errp) | |
61007b31 SH |
5995 | { |
5996 | BlockDriver *drv = bs->drv; | |
5997 | if (drv && drv->bdrv_get_specific_info) { | |
1bf6e9ca | 5998 | return drv->bdrv_get_specific_info(bs, errp); |
61007b31 SH |
5999 | } |
6000 | return NULL; | |
016f5cf6 AG |
6001 | } |
6002 | ||
d9245599 AN |
6003 | BlockStatsSpecific *bdrv_get_specific_stats(BlockDriverState *bs) |
6004 | { | |
6005 | BlockDriver *drv = bs->drv; | |
6006 | if (!drv || !drv->bdrv_get_specific_stats) { | |
6007 | return NULL; | |
6008 | } | |
6009 | return drv->bdrv_get_specific_stats(bs); | |
6010 | } | |
6011 | ||
a31939e6 | 6012 | void bdrv_debug_event(BlockDriverState *bs, BlkdebugEvent event) |
4265d620 | 6013 | { |
61007b31 SH |
6014 | if (!bs || !bs->drv || !bs->drv->bdrv_debug_event) { |
6015 | return; | |
6016 | } | |
4265d620 | 6017 | |
61007b31 | 6018 | bs->drv->bdrv_debug_event(bs, event); |
4265d620 PB |
6019 | } |
6020 | ||
d10529a2 | 6021 | static BlockDriverState *bdrv_find_debug_node(BlockDriverState *bs) |
4265d620 | 6022 | { |
61007b31 | 6023 | while (bs && bs->drv && !bs->drv->bdrv_debug_breakpoint) { |
f706a92f | 6024 | bs = bdrv_primary_bs(bs); |
61007b31 | 6025 | } |
4265d620 | 6026 | |
61007b31 | 6027 | if (bs && bs->drv && bs->drv->bdrv_debug_breakpoint) { |
d10529a2 VSO |
6028 | assert(bs->drv->bdrv_debug_remove_breakpoint); |
6029 | return bs; | |
6030 | } | |
6031 | ||
6032 | return NULL; | |
6033 | } | |
6034 | ||
6035 | int bdrv_debug_breakpoint(BlockDriverState *bs, const char *event, | |
6036 | const char *tag) | |
6037 | { | |
6038 | bs = bdrv_find_debug_node(bs); | |
6039 | if (bs) { | |
61007b31 SH |
6040 | return bs->drv->bdrv_debug_breakpoint(bs, event, tag); |
6041 | } | |
4265d620 | 6042 | |
61007b31 | 6043 | return -ENOTSUP; |
4265d620 PB |
6044 | } |
6045 | ||
61007b31 | 6046 | int bdrv_debug_remove_breakpoint(BlockDriverState *bs, const char *tag) |
ea2384d3 | 6047 | { |
d10529a2 VSO |
6048 | bs = bdrv_find_debug_node(bs); |
6049 | if (bs) { | |
61007b31 SH |
6050 | return bs->drv->bdrv_debug_remove_breakpoint(bs, tag); |
6051 | } | |
6052 | ||
6053 | return -ENOTSUP; | |
eb852011 MA |
6054 | } |
6055 | ||
61007b31 | 6056 | int bdrv_debug_resume(BlockDriverState *bs, const char *tag) |
ce1a14dc | 6057 | { |
61007b31 | 6058 | while (bs && (!bs->drv || !bs->drv->bdrv_debug_resume)) { |
f706a92f | 6059 | bs = bdrv_primary_bs(bs); |
61007b31 | 6060 | } |
ce1a14dc | 6061 | |
61007b31 SH |
6062 | if (bs && bs->drv && bs->drv->bdrv_debug_resume) { |
6063 | return bs->drv->bdrv_debug_resume(bs, tag); | |
6064 | } | |
ce1a14dc | 6065 | |
61007b31 | 6066 | return -ENOTSUP; |
f197fe2b FZ |
6067 | } |
6068 | ||
61007b31 | 6069 | bool bdrv_debug_is_suspended(BlockDriverState *bs, const char *tag) |
ce1a14dc | 6070 | { |
61007b31 | 6071 | while (bs && bs->drv && !bs->drv->bdrv_debug_is_suspended) { |
f706a92f | 6072 | bs = bdrv_primary_bs(bs); |
f197fe2b | 6073 | } |
19cb3738 | 6074 | |
61007b31 SH |
6075 | if (bs && bs->drv && bs->drv->bdrv_debug_is_suspended) { |
6076 | return bs->drv->bdrv_debug_is_suspended(bs, tag); | |
6077 | } | |
f9f05dc5 | 6078 | |
61007b31 SH |
6079 | return false; |
6080 | } | |
f9f05dc5 | 6081 | |
61007b31 SH |
6082 | /* backing_file can either be relative, or absolute, or a protocol. If it is |
6083 | * relative, it must be relative to the chain. So, passing in bs->filename | |
6084 | * from a BDS as backing_file should not be done, as that may be relative to | |
6085 | * the CWD rather than the chain. */ | |
6086 | BlockDriverState *bdrv_find_backing_image(BlockDriverState *bs, | |
6087 | const char *backing_file) | |
f9f05dc5 | 6088 | { |
61007b31 SH |
6089 | char *filename_full = NULL; |
6090 | char *backing_file_full = NULL; | |
6091 | char *filename_tmp = NULL; | |
6092 | int is_protocol = 0; | |
0b877d09 | 6093 | bool filenames_refreshed = false; |
61007b31 SH |
6094 | BlockDriverState *curr_bs = NULL; |
6095 | BlockDriverState *retval = NULL; | |
dcf3f9b2 | 6096 | BlockDriverState *bs_below; |
f9f05dc5 | 6097 | |
61007b31 SH |
6098 | if (!bs || !bs->drv || !backing_file) { |
6099 | return NULL; | |
f9f05dc5 KW |
6100 | } |
6101 | ||
61007b31 SH |
6102 | filename_full = g_malloc(PATH_MAX); |
6103 | backing_file_full = g_malloc(PATH_MAX); | |
f9f05dc5 | 6104 | |
61007b31 | 6105 | is_protocol = path_has_protocol(backing_file); |
f9f05dc5 | 6106 | |
dcf3f9b2 HR |
6107 | /* |
6108 | * Being largely a legacy function, skip any filters here | |
6109 | * (because filters do not have normal filenames, so they cannot | |
6110 | * match anyway; and allowing json:{} filenames is a bit out of | |
6111 | * scope). | |
6112 | */ | |
6113 | for (curr_bs = bdrv_skip_filters(bs); | |
6114 | bdrv_cow_child(curr_bs) != NULL; | |
6115 | curr_bs = bs_below) | |
6116 | { | |
6117 | bs_below = bdrv_backing_chain_next(curr_bs); | |
f9f05dc5 | 6118 | |
0b877d09 HR |
6119 | if (bdrv_backing_overridden(curr_bs)) { |
6120 | /* | |
6121 | * If the backing file was overridden, we can only compare | |
6122 | * directly against the backing node's filename. | |
6123 | */ | |
6124 | ||
6125 | if (!filenames_refreshed) { | |
6126 | /* | |
6127 | * This will automatically refresh all of the | |
6128 | * filenames in the rest of the backing chain, so we | |
6129 | * only need to do this once. | |
6130 | */ | |
6131 | bdrv_refresh_filename(bs_below); | |
6132 | filenames_refreshed = true; | |
6133 | } | |
6134 | ||
6135 | if (strcmp(backing_file, bs_below->filename) == 0) { | |
6136 | retval = bs_below; | |
6137 | break; | |
6138 | } | |
6139 | } else if (is_protocol || path_has_protocol(curr_bs->backing_file)) { | |
6140 | /* | |
6141 | * If either of the filename paths is actually a protocol, then | |
6142 | * compare unmodified paths; otherwise make paths relative. | |
6143 | */ | |
6b6833c1 HR |
6144 | char *backing_file_full_ret; |
6145 | ||
61007b31 | 6146 | if (strcmp(backing_file, curr_bs->backing_file) == 0) { |
dcf3f9b2 | 6147 | retval = bs_below; |
61007b31 SH |
6148 | break; |
6149 | } | |
418661e0 | 6150 | /* Also check against the full backing filename for the image */ |
6b6833c1 HR |
6151 | backing_file_full_ret = bdrv_get_full_backing_filename(curr_bs, |
6152 | NULL); | |
6153 | if (backing_file_full_ret) { | |
6154 | bool equal = strcmp(backing_file, backing_file_full_ret) == 0; | |
6155 | g_free(backing_file_full_ret); | |
6156 | if (equal) { | |
dcf3f9b2 | 6157 | retval = bs_below; |
418661e0 JC |
6158 | break; |
6159 | } | |
418661e0 | 6160 | } |
61007b31 SH |
6161 | } else { |
6162 | /* If not an absolute filename path, make it relative to the current | |
6163 | * image's filename path */ | |
2d9158ce HR |
6164 | filename_tmp = bdrv_make_absolute_filename(curr_bs, backing_file, |
6165 | NULL); | |
6166 | /* We are going to compare canonicalized absolute pathnames */ | |
6167 | if (!filename_tmp || !realpath(filename_tmp, filename_full)) { | |
6168 | g_free(filename_tmp); | |
61007b31 SH |
6169 | continue; |
6170 | } | |
2d9158ce | 6171 | g_free(filename_tmp); |
07f07615 | 6172 | |
61007b31 SH |
6173 | /* We need to make sure the backing filename we are comparing against |
6174 | * is relative to the current image filename (or absolute) */ | |
2d9158ce HR |
6175 | filename_tmp = bdrv_get_full_backing_filename(curr_bs, NULL); |
6176 | if (!filename_tmp || !realpath(filename_tmp, backing_file_full)) { | |
6177 | g_free(filename_tmp); | |
61007b31 SH |
6178 | continue; |
6179 | } | |
2d9158ce | 6180 | g_free(filename_tmp); |
eb489bb1 | 6181 | |
61007b31 | 6182 | if (strcmp(backing_file_full, filename_full) == 0) { |
dcf3f9b2 | 6183 | retval = bs_below; |
61007b31 SH |
6184 | break; |
6185 | } | |
6186 | } | |
eb489bb1 KW |
6187 | } |
6188 | ||
61007b31 SH |
6189 | g_free(filename_full); |
6190 | g_free(backing_file_full); | |
61007b31 SH |
6191 | return retval; |
6192 | } | |
6193 | ||
61007b31 SH |
6194 | void bdrv_init(void) |
6195 | { | |
e5f05f8c KW |
6196 | #ifdef CONFIG_BDRV_WHITELIST_TOOLS |
6197 | use_bdrv_whitelist = 1; | |
6198 | #endif | |
61007b31 SH |
6199 | module_call_init(MODULE_INIT_BLOCK); |
6200 | } | |
29cdb251 | 6201 | |
61007b31 SH |
6202 | void bdrv_init_with_whitelist(void) |
6203 | { | |
6204 | use_bdrv_whitelist = 1; | |
6205 | bdrv_init(); | |
07f07615 PB |
6206 | } |
6207 | ||
21c2283e | 6208 | int coroutine_fn bdrv_co_invalidate_cache(BlockDriverState *bs, Error **errp) |
0f15423c | 6209 | { |
4417ab7a | 6210 | BdrvChild *child, *parent; |
5a8a30db KW |
6211 | Error *local_err = NULL; |
6212 | int ret; | |
9c98f145 | 6213 | BdrvDirtyBitmap *bm; |
5a8a30db | 6214 | |
3456a8d1 | 6215 | if (!bs->drv) { |
5416645f | 6216 | return -ENOMEDIUM; |
3456a8d1 KW |
6217 | } |
6218 | ||
16e977d5 | 6219 | QLIST_FOREACH(child, &bs->children, next) { |
2b148f39 | 6220 | bdrv_co_invalidate_cache(child->bs, &local_err); |
0d1c5c91 | 6221 | if (local_err) { |
0d1c5c91 | 6222 | error_propagate(errp, local_err); |
5416645f | 6223 | return -EINVAL; |
0d1c5c91 | 6224 | } |
5a8a30db | 6225 | } |
0d1c5c91 | 6226 | |
dafe0960 KW |
6227 | /* |
6228 | * Update permissions, they may differ for inactive nodes. | |
6229 | * | |
6230 | * Note that the required permissions of inactive images are always a | |
6231 | * subset of the permissions required after activating the image. This | |
6232 | * allows us to just get the permissions upfront without restricting | |
6233 | * drv->bdrv_invalidate_cache(). | |
6234 | * | |
6235 | * It also means that in error cases, we don't have to try and revert to | |
6236 | * the old permissions (which is an operation that could fail, too). We can | |
6237 | * just keep the extended permissions for the next time that an activation | |
6238 | * of the image is tried. | |
6239 | */ | |
7bb4941a KW |
6240 | if (bs->open_flags & BDRV_O_INACTIVE) { |
6241 | bs->open_flags &= ~BDRV_O_INACTIVE; | |
071b474f | 6242 | ret = bdrv_refresh_perms(bs, errp); |
7bb4941a | 6243 | if (ret < 0) { |
0d1c5c91 | 6244 | bs->open_flags |= BDRV_O_INACTIVE; |
5416645f | 6245 | return ret; |
0d1c5c91 | 6246 | } |
3456a8d1 | 6247 | |
7bb4941a KW |
6248 | if (bs->drv->bdrv_co_invalidate_cache) { |
6249 | bs->drv->bdrv_co_invalidate_cache(bs, &local_err); | |
6250 | if (local_err) { | |
6251 | bs->open_flags |= BDRV_O_INACTIVE; | |
6252 | error_propagate(errp, local_err); | |
5416645f | 6253 | return -EINVAL; |
7bb4941a KW |
6254 | } |
6255 | } | |
9c98f145 | 6256 | |
7bb4941a KW |
6257 | FOR_EACH_DIRTY_BITMAP(bs, bm) { |
6258 | bdrv_dirty_bitmap_skip_store(bm, false); | |
6259 | } | |
6260 | ||
6261 | ret = refresh_total_sectors(bs, bs->total_sectors); | |
6262 | if (ret < 0) { | |
6263 | bs->open_flags |= BDRV_O_INACTIVE; | |
6264 | error_setg_errno(errp, -ret, "Could not refresh total sector count"); | |
5416645f | 6265 | return ret; |
7bb4941a | 6266 | } |
5a8a30db | 6267 | } |
4417ab7a KW |
6268 | |
6269 | QLIST_FOREACH(parent, &bs->parents, next_parent) { | |
bd86fb99 HR |
6270 | if (parent->klass->activate) { |
6271 | parent->klass->activate(parent, &local_err); | |
4417ab7a | 6272 | if (local_err) { |
78fc3b3a | 6273 | bs->open_flags |= BDRV_O_INACTIVE; |
4417ab7a | 6274 | error_propagate(errp, local_err); |
5416645f | 6275 | return -EINVAL; |
4417ab7a KW |
6276 | } |
6277 | } | |
6278 | } | |
5416645f VSO |
6279 | |
6280 | return 0; | |
0f15423c AL |
6281 | } |
6282 | ||
5a8a30db | 6283 | void bdrv_invalidate_cache_all(Error **errp) |
0f15423c | 6284 | { |
7c8eece4 | 6285 | BlockDriverState *bs; |
88be7b4b | 6286 | BdrvNextIterator it; |
0f15423c | 6287 | |
88be7b4b | 6288 | for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) { |
ed78cda3 | 6289 | AioContext *aio_context = bdrv_get_aio_context(bs); |
5416645f | 6290 | int ret; |
ed78cda3 SH |
6291 | |
6292 | aio_context_acquire(aio_context); | |
5416645f | 6293 | ret = bdrv_invalidate_cache(bs, errp); |
ed78cda3 | 6294 | aio_context_release(aio_context); |
5416645f | 6295 | if (ret < 0) { |
5e003f17 | 6296 | bdrv_next_cleanup(&it); |
5a8a30db KW |
6297 | return; |
6298 | } | |
0f15423c AL |
6299 | } |
6300 | } | |
6301 | ||
9e37271f KW |
6302 | static bool bdrv_has_bds_parent(BlockDriverState *bs, bool only_active) |
6303 | { | |
6304 | BdrvChild *parent; | |
6305 | ||
6306 | QLIST_FOREACH(parent, &bs->parents, next_parent) { | |
bd86fb99 | 6307 | if (parent->klass->parent_is_bds) { |
9e37271f KW |
6308 | BlockDriverState *parent_bs = parent->opaque; |
6309 | if (!only_active || !(parent_bs->open_flags & BDRV_O_INACTIVE)) { | |
6310 | return true; | |
6311 | } | |
6312 | } | |
6313 | } | |
6314 | ||
6315 | return false; | |
6316 | } | |
6317 | ||
6318 | static int bdrv_inactivate_recurse(BlockDriverState *bs) | |
76b1c7fe | 6319 | { |
cfa1a572 | 6320 | BdrvChild *child, *parent; |
76b1c7fe KW |
6321 | int ret; |
6322 | ||
d470ad42 HR |
6323 | if (!bs->drv) { |
6324 | return -ENOMEDIUM; | |
6325 | } | |
6326 | ||
9e37271f KW |
6327 | /* Make sure that we don't inactivate a child before its parent. |
6328 | * It will be covered by recursion from the yet active parent. */ | |
6329 | if (bdrv_has_bds_parent(bs, true)) { | |
6330 | return 0; | |
6331 | } | |
6332 | ||
6333 | assert(!(bs->open_flags & BDRV_O_INACTIVE)); | |
6334 | ||
6335 | /* Inactivate this node */ | |
6336 | if (bs->drv->bdrv_inactivate) { | |
76b1c7fe KW |
6337 | ret = bs->drv->bdrv_inactivate(bs); |
6338 | if (ret < 0) { | |
6339 | return ret; | |
6340 | } | |
6341 | } | |
6342 | ||
9e37271f | 6343 | QLIST_FOREACH(parent, &bs->parents, next_parent) { |
bd86fb99 HR |
6344 | if (parent->klass->inactivate) { |
6345 | ret = parent->klass->inactivate(parent); | |
9e37271f KW |
6346 | if (ret < 0) { |
6347 | return ret; | |
cfa1a572 KW |
6348 | } |
6349 | } | |
9e37271f | 6350 | } |
9c5e6594 | 6351 | |
9e37271f | 6352 | bs->open_flags |= BDRV_O_INACTIVE; |
7d5b5261 | 6353 | |
bb87e4d1 VSO |
6354 | /* |
6355 | * Update permissions, they may differ for inactive nodes. | |
6356 | * We only tried to loosen restrictions, so errors are not fatal, ignore | |
6357 | * them. | |
6358 | */ | |
071b474f | 6359 | bdrv_refresh_perms(bs, NULL); |
9e37271f KW |
6360 | |
6361 | /* Recursively inactivate children */ | |
38701b6a | 6362 | QLIST_FOREACH(child, &bs->children, next) { |
9e37271f | 6363 | ret = bdrv_inactivate_recurse(child->bs); |
38701b6a KW |
6364 | if (ret < 0) { |
6365 | return ret; | |
6366 | } | |
6367 | } | |
6368 | ||
76b1c7fe KW |
6369 | return 0; |
6370 | } | |
6371 | ||
6372 | int bdrv_inactivate_all(void) | |
6373 | { | |
79720af6 | 6374 | BlockDriverState *bs = NULL; |
88be7b4b | 6375 | BdrvNextIterator it; |
aad0b7a0 | 6376 | int ret = 0; |
bd6458e4 | 6377 | GSList *aio_ctxs = NULL, *ctx; |
76b1c7fe | 6378 | |
88be7b4b | 6379 | for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) { |
bd6458e4 PB |
6380 | AioContext *aio_context = bdrv_get_aio_context(bs); |
6381 | ||
6382 | if (!g_slist_find(aio_ctxs, aio_context)) { | |
6383 | aio_ctxs = g_slist_prepend(aio_ctxs, aio_context); | |
6384 | aio_context_acquire(aio_context); | |
6385 | } | |
aad0b7a0 | 6386 | } |
76b1c7fe | 6387 | |
9e37271f KW |
6388 | for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) { |
6389 | /* Nodes with BDS parents are covered by recursion from the last | |
6390 | * parent that gets inactivated. Don't inactivate them a second | |
6391 | * time if that has already happened. */ | |
6392 | if (bdrv_has_bds_parent(bs, false)) { | |
6393 | continue; | |
6394 | } | |
6395 | ret = bdrv_inactivate_recurse(bs); | |
6396 | if (ret < 0) { | |
6397 | bdrv_next_cleanup(&it); | |
6398 | goto out; | |
76b1c7fe KW |
6399 | } |
6400 | } | |
6401 | ||
aad0b7a0 | 6402 | out: |
bd6458e4 PB |
6403 | for (ctx = aio_ctxs; ctx != NULL; ctx = ctx->next) { |
6404 | AioContext *aio_context = ctx->data; | |
6405 | aio_context_release(aio_context); | |
aad0b7a0 | 6406 | } |
bd6458e4 | 6407 | g_slist_free(aio_ctxs); |
aad0b7a0 FZ |
6408 | |
6409 | return ret; | |
76b1c7fe KW |
6410 | } |
6411 | ||
19cb3738 FB |
6412 | /**************************************************************/ |
6413 | /* removable device support */ | |
6414 | ||
6415 | /** | |
6416 | * Return TRUE if the media is present | |
6417 | */ | |
e031f750 | 6418 | bool bdrv_is_inserted(BlockDriverState *bs) |
19cb3738 FB |
6419 | { |
6420 | BlockDriver *drv = bs->drv; | |
28d7a789 | 6421 | BdrvChild *child; |
a1aff5bf | 6422 | |
e031f750 HR |
6423 | if (!drv) { |
6424 | return false; | |
6425 | } | |
28d7a789 HR |
6426 | if (drv->bdrv_is_inserted) { |
6427 | return drv->bdrv_is_inserted(bs); | |
6428 | } | |
6429 | QLIST_FOREACH(child, &bs->children, next) { | |
6430 | if (!bdrv_is_inserted(child->bs)) { | |
6431 | return false; | |
6432 | } | |
e031f750 | 6433 | } |
28d7a789 | 6434 | return true; |
19cb3738 FB |
6435 | } |
6436 | ||
19cb3738 FB |
6437 | /** |
6438 | * If eject_flag is TRUE, eject the media. Otherwise, close the tray | |
6439 | */ | |
f36f3949 | 6440 | void bdrv_eject(BlockDriverState *bs, bool eject_flag) |
19cb3738 FB |
6441 | { |
6442 | BlockDriver *drv = bs->drv; | |
19cb3738 | 6443 | |
822e1cd1 MA |
6444 | if (drv && drv->bdrv_eject) { |
6445 | drv->bdrv_eject(bs, eject_flag); | |
19cb3738 FB |
6446 | } |
6447 | } | |
6448 | ||
19cb3738 FB |
6449 | /** |
6450 | * Lock or unlock the media (if it is locked, the user won't be able | |
6451 | * to eject it manually). | |
6452 | */ | |
025e849a | 6453 | void bdrv_lock_medium(BlockDriverState *bs, bool locked) |
19cb3738 FB |
6454 | { |
6455 | BlockDriver *drv = bs->drv; | |
6456 | ||
025e849a | 6457 | trace_bdrv_lock_medium(bs, locked); |
b8c6d095 | 6458 | |
025e849a MA |
6459 | if (drv && drv->bdrv_lock_medium) { |
6460 | drv->bdrv_lock_medium(bs, locked); | |
19cb3738 FB |
6461 | } |
6462 | } | |
985a03b0 | 6463 | |
9fcb0251 FZ |
6464 | /* Get a reference to bs */ |
6465 | void bdrv_ref(BlockDriverState *bs) | |
6466 | { | |
6467 | bs->refcnt++; | |
6468 | } | |
6469 | ||
6470 | /* Release a previously grabbed reference to bs. | |
6471 | * If after releasing, reference count is zero, the BlockDriverState is | |
6472 | * deleted. */ | |
6473 | void bdrv_unref(BlockDriverState *bs) | |
6474 | { | |
9a4d5ca6 JC |
6475 | if (!bs) { |
6476 | return; | |
6477 | } | |
9fcb0251 FZ |
6478 | assert(bs->refcnt > 0); |
6479 | if (--bs->refcnt == 0) { | |
6480 | bdrv_delete(bs); | |
6481 | } | |
6482 | } | |
6483 | ||
fbe40ff7 FZ |
6484 | struct BdrvOpBlocker { |
6485 | Error *reason; | |
6486 | QLIST_ENTRY(BdrvOpBlocker) list; | |
6487 | }; | |
6488 | ||
6489 | bool bdrv_op_is_blocked(BlockDriverState *bs, BlockOpType op, Error **errp) | |
6490 | { | |
6491 | BdrvOpBlocker *blocker; | |
6492 | assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX); | |
6493 | if (!QLIST_EMPTY(&bs->op_blockers[op])) { | |
6494 | blocker = QLIST_FIRST(&bs->op_blockers[op]); | |
4b576648 MA |
6495 | error_propagate_prepend(errp, error_copy(blocker->reason), |
6496 | "Node '%s' is busy: ", | |
6497 | bdrv_get_device_or_node_name(bs)); | |
fbe40ff7 FZ |
6498 | return true; |
6499 | } | |
6500 | return false; | |
6501 | } | |
6502 | ||
6503 | void bdrv_op_block(BlockDriverState *bs, BlockOpType op, Error *reason) | |
6504 | { | |
6505 | BdrvOpBlocker *blocker; | |
6506 | assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX); | |
6507 | ||
5839e53b | 6508 | blocker = g_new0(BdrvOpBlocker, 1); |
fbe40ff7 FZ |
6509 | blocker->reason = reason; |
6510 | QLIST_INSERT_HEAD(&bs->op_blockers[op], blocker, list); | |
6511 | } | |
6512 | ||
6513 | void bdrv_op_unblock(BlockDriverState *bs, BlockOpType op, Error *reason) | |
6514 | { | |
6515 | BdrvOpBlocker *blocker, *next; | |
6516 | assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX); | |
6517 | QLIST_FOREACH_SAFE(blocker, &bs->op_blockers[op], list, next) { | |
6518 | if (blocker->reason == reason) { | |
6519 | QLIST_REMOVE(blocker, list); | |
6520 | g_free(blocker); | |
6521 | } | |
6522 | } | |
6523 | } | |
6524 | ||
6525 | void bdrv_op_block_all(BlockDriverState *bs, Error *reason) | |
6526 | { | |
6527 | int i; | |
6528 | for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) { | |
6529 | bdrv_op_block(bs, i, reason); | |
6530 | } | |
6531 | } | |
6532 | ||
6533 | void bdrv_op_unblock_all(BlockDriverState *bs, Error *reason) | |
6534 | { | |
6535 | int i; | |
6536 | for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) { | |
6537 | bdrv_op_unblock(bs, i, reason); | |
6538 | } | |
6539 | } | |
6540 | ||
6541 | bool bdrv_op_blocker_is_empty(BlockDriverState *bs) | |
6542 | { | |
6543 | int i; | |
6544 | ||
6545 | for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) { | |
6546 | if (!QLIST_EMPTY(&bs->op_blockers[i])) { | |
6547 | return false; | |
6548 | } | |
6549 | } | |
6550 | return true; | |
6551 | } | |
6552 | ||
d92ada22 LC |
6553 | void bdrv_img_create(const char *filename, const char *fmt, |
6554 | const char *base_filename, const char *base_fmt, | |
9217283d FZ |
6555 | char *options, uint64_t img_size, int flags, bool quiet, |
6556 | Error **errp) | |
f88e1a42 | 6557 | { |
83d0521a CL |
6558 | QemuOptsList *create_opts = NULL; |
6559 | QemuOpts *opts = NULL; | |
6560 | const char *backing_fmt, *backing_file; | |
6561 | int64_t size; | |
f88e1a42 | 6562 | BlockDriver *drv, *proto_drv; |
cc84d90f | 6563 | Error *local_err = NULL; |
f88e1a42 JS |
6564 | int ret = 0; |
6565 | ||
6566 | /* Find driver and parse its options */ | |
6567 | drv = bdrv_find_format(fmt); | |
6568 | if (!drv) { | |
71c79813 | 6569 | error_setg(errp, "Unknown file format '%s'", fmt); |
d92ada22 | 6570 | return; |
f88e1a42 JS |
6571 | } |
6572 | ||
b65a5e12 | 6573 | proto_drv = bdrv_find_protocol(filename, true, errp); |
f88e1a42 | 6574 | if (!proto_drv) { |
d92ada22 | 6575 | return; |
f88e1a42 JS |
6576 | } |
6577 | ||
c6149724 HR |
6578 | if (!drv->create_opts) { |
6579 | error_setg(errp, "Format driver '%s' does not support image creation", | |
6580 | drv->format_name); | |
6581 | return; | |
6582 | } | |
6583 | ||
5a5e7f8c ML |
6584 | if (!proto_drv->create_opts) { |
6585 | error_setg(errp, "Protocol driver '%s' does not support image creation", | |
6586 | proto_drv->format_name); | |
6587 | return; | |
6588 | } | |
6589 | ||
f6dc1c31 | 6590 | /* Create parameter list */ |
c282e1fd | 6591 | create_opts = qemu_opts_append(create_opts, drv->create_opts); |
5a5e7f8c | 6592 | create_opts = qemu_opts_append(create_opts, proto_drv->create_opts); |
f88e1a42 | 6593 | |
83d0521a | 6594 | opts = qemu_opts_create(create_opts, NULL, 0, &error_abort); |
f88e1a42 JS |
6595 | |
6596 | /* Parse -o options */ | |
6597 | if (options) { | |
a5f9b9df | 6598 | if (!qemu_opts_do_parse(opts, options, NULL, errp)) { |
f88e1a42 JS |
6599 | goto out; |
6600 | } | |
6601 | } | |
6602 | ||
f6dc1c31 KW |
6603 | if (!qemu_opt_get(opts, BLOCK_OPT_SIZE)) { |
6604 | qemu_opt_set_number(opts, BLOCK_OPT_SIZE, img_size, &error_abort); | |
6605 | } else if (img_size != UINT64_C(-1)) { | |
6606 | error_setg(errp, "The image size must be specified only once"); | |
6607 | goto out; | |
6608 | } | |
6609 | ||
f88e1a42 | 6610 | if (base_filename) { |
235e59cf | 6611 | if (!qemu_opt_set(opts, BLOCK_OPT_BACKING_FILE, base_filename, |
3882578b | 6612 | NULL)) { |
71c79813 LC |
6613 | error_setg(errp, "Backing file not supported for file format '%s'", |
6614 | fmt); | |
f88e1a42 JS |
6615 | goto out; |
6616 | } | |
6617 | } | |
6618 | ||
6619 | if (base_fmt) { | |
3882578b | 6620 | if (!qemu_opt_set(opts, BLOCK_OPT_BACKING_FMT, base_fmt, NULL)) { |
71c79813 LC |
6621 | error_setg(errp, "Backing file format not supported for file " |
6622 | "format '%s'", fmt); | |
f88e1a42 JS |
6623 | goto out; |
6624 | } | |
6625 | } | |
6626 | ||
83d0521a CL |
6627 | backing_file = qemu_opt_get(opts, BLOCK_OPT_BACKING_FILE); |
6628 | if (backing_file) { | |
6629 | if (!strcmp(filename, backing_file)) { | |
71c79813 LC |
6630 | error_setg(errp, "Error: Trying to create an image with the " |
6631 | "same filename as the backing file"); | |
792da93a JS |
6632 | goto out; |
6633 | } | |
975a7bd2 CK |
6634 | if (backing_file[0] == '\0') { |
6635 | error_setg(errp, "Expected backing file name, got empty string"); | |
6636 | goto out; | |
6637 | } | |
792da93a JS |
6638 | } |
6639 | ||
83d0521a | 6640 | backing_fmt = qemu_opt_get(opts, BLOCK_OPT_BACKING_FMT); |
f88e1a42 | 6641 | |
6e6e55f5 JS |
6642 | /* The size for the image must always be specified, unless we have a backing |
6643 | * file and we have not been forbidden from opening it. */ | |
a8b42a1c | 6644 | size = qemu_opt_get_size(opts, BLOCK_OPT_SIZE, img_size); |
6e6e55f5 JS |
6645 | if (backing_file && !(flags & BDRV_O_NO_BACKING)) { |
6646 | BlockDriverState *bs; | |
645ae7d8 | 6647 | char *full_backing; |
6e6e55f5 JS |
6648 | int back_flags; |
6649 | QDict *backing_options = NULL; | |
6650 | ||
645ae7d8 HR |
6651 | full_backing = |
6652 | bdrv_get_full_backing_filename_from_filename(filename, backing_file, | |
6653 | &local_err); | |
6e6e55f5 | 6654 | if (local_err) { |
6e6e55f5 JS |
6655 | goto out; |
6656 | } | |
645ae7d8 | 6657 | assert(full_backing); |
29168018 | 6658 | |
d5b23994 HR |
6659 | /* |
6660 | * No need to do I/O here, which allows us to open encrypted | |
6661 | * backing images without needing the secret | |
6662 | */ | |
6e6e55f5 JS |
6663 | back_flags = flags; |
6664 | back_flags &= ~(BDRV_O_RDWR | BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING); | |
d5b23994 | 6665 | back_flags |= BDRV_O_NO_IO; |
f88e1a42 | 6666 | |
cc954f01 | 6667 | backing_options = qdict_new(); |
6e6e55f5 | 6668 | if (backing_fmt) { |
6e6e55f5 JS |
6669 | qdict_put_str(backing_options, "driver", backing_fmt); |
6670 | } | |
cc954f01 | 6671 | qdict_put_bool(backing_options, BDRV_OPT_FORCE_SHARE, true); |
e6641719 | 6672 | |
6e6e55f5 JS |
6673 | bs = bdrv_open(full_backing, NULL, backing_options, back_flags, |
6674 | &local_err); | |
6675 | g_free(full_backing); | |
add8200d EB |
6676 | if (!bs) { |
6677 | error_append_hint(&local_err, "Could not open backing image.\n"); | |
6e6e55f5 JS |
6678 | goto out; |
6679 | } else { | |
d9f059aa | 6680 | if (!backing_fmt) { |
497a30db EB |
6681 | error_setg(&local_err, |
6682 | "Backing file specified without backing format"); | |
6683 | error_append_hint(&local_err, "Detected format of %s.", | |
6684 | bs->drv->format_name); | |
6685 | goto out; | |
d9f059aa | 6686 | } |
6e6e55f5 JS |
6687 | if (size == -1) { |
6688 | /* Opened BS, have no size */ | |
6689 | size = bdrv_getlength(bs); | |
6690 | if (size < 0) { | |
6691 | error_setg_errno(errp, -size, "Could not get size of '%s'", | |
6692 | backing_file); | |
6693 | bdrv_unref(bs); | |
6694 | goto out; | |
6695 | } | |
6696 | qemu_opt_set_number(opts, BLOCK_OPT_SIZE, size, &error_abort); | |
52bf1e72 | 6697 | } |
66f6b814 | 6698 | bdrv_unref(bs); |
f88e1a42 | 6699 | } |
d9f059aa EB |
6700 | /* (backing_file && !(flags & BDRV_O_NO_BACKING)) */ |
6701 | } else if (backing_file && !backing_fmt) { | |
497a30db EB |
6702 | error_setg(&local_err, |
6703 | "Backing file specified without backing format"); | |
6704 | goto out; | |
d9f059aa | 6705 | } |
6e6e55f5 JS |
6706 | |
6707 | if (size == -1) { | |
6708 | error_setg(errp, "Image creation needs a size parameter"); | |
6709 | goto out; | |
f88e1a42 JS |
6710 | } |
6711 | ||
f382d43a | 6712 | if (!quiet) { |
fe646693 | 6713 | printf("Formatting '%s', fmt=%s ", filename, fmt); |
43c5d8f8 | 6714 | qemu_opts_print(opts, " "); |
f382d43a | 6715 | puts(""); |
4e2f4418 | 6716 | fflush(stdout); |
f382d43a | 6717 | } |
83d0521a | 6718 | |
c282e1fd | 6719 | ret = bdrv_create(drv, filename, opts, &local_err); |
83d0521a | 6720 | |
cc84d90f HR |
6721 | if (ret == -EFBIG) { |
6722 | /* This is generally a better message than whatever the driver would | |
6723 | * deliver (especially because of the cluster_size_hint), since that | |
6724 | * is most probably not much different from "image too large". */ | |
6725 | const char *cluster_size_hint = ""; | |
83d0521a | 6726 | if (qemu_opt_get_size(opts, BLOCK_OPT_CLUSTER_SIZE, 0)) { |
cc84d90f | 6727 | cluster_size_hint = " (try using a larger cluster size)"; |
f88e1a42 | 6728 | } |
cc84d90f HR |
6729 | error_setg(errp, "The image size is too large for file format '%s'" |
6730 | "%s", fmt, cluster_size_hint); | |
6731 | error_free(local_err); | |
6732 | local_err = NULL; | |
f88e1a42 JS |
6733 | } |
6734 | ||
6735 | out: | |
83d0521a CL |
6736 | qemu_opts_del(opts); |
6737 | qemu_opts_free(create_opts); | |
621ff94d | 6738 | error_propagate(errp, local_err); |
f88e1a42 | 6739 | } |
85d126f3 SH |
6740 | |
6741 | AioContext *bdrv_get_aio_context(BlockDriverState *bs) | |
6742 | { | |
33f2a757 | 6743 | return bs ? bs->aio_context : qemu_get_aio_context(); |
dcd04228 SH |
6744 | } |
6745 | ||
e336fd4c KW |
6746 | AioContext *coroutine_fn bdrv_co_enter(BlockDriverState *bs) |
6747 | { | |
6748 | Coroutine *self = qemu_coroutine_self(); | |
6749 | AioContext *old_ctx = qemu_coroutine_get_aio_context(self); | |
6750 | AioContext *new_ctx; | |
6751 | ||
6752 | /* | |
6753 | * Increase bs->in_flight to ensure that this operation is completed before | |
6754 | * moving the node to a different AioContext. Read new_ctx only afterwards. | |
6755 | */ | |
6756 | bdrv_inc_in_flight(bs); | |
6757 | ||
6758 | new_ctx = bdrv_get_aio_context(bs); | |
6759 | aio_co_reschedule_self(new_ctx); | |
6760 | return old_ctx; | |
6761 | } | |
6762 | ||
6763 | void coroutine_fn bdrv_co_leave(BlockDriverState *bs, AioContext *old_ctx) | |
6764 | { | |
6765 | aio_co_reschedule_self(old_ctx); | |
6766 | bdrv_dec_in_flight(bs); | |
6767 | } | |
6768 | ||
18c6ac1c KW |
6769 | void coroutine_fn bdrv_co_lock(BlockDriverState *bs) |
6770 | { | |
6771 | AioContext *ctx = bdrv_get_aio_context(bs); | |
6772 | ||
6773 | /* In the main thread, bs->aio_context won't change concurrently */ | |
6774 | assert(qemu_get_current_aio_context() == qemu_get_aio_context()); | |
6775 | ||
6776 | /* | |
6777 | * We're in coroutine context, so we already hold the lock of the main | |
6778 | * loop AioContext. Don't lock it twice to avoid deadlocks. | |
6779 | */ | |
6780 | assert(qemu_in_coroutine()); | |
6781 | if (ctx != qemu_get_aio_context()) { | |
6782 | aio_context_acquire(ctx); | |
6783 | } | |
6784 | } | |
6785 | ||
6786 | void coroutine_fn bdrv_co_unlock(BlockDriverState *bs) | |
6787 | { | |
6788 | AioContext *ctx = bdrv_get_aio_context(bs); | |
6789 | ||
6790 | assert(qemu_in_coroutine()); | |
6791 | if (ctx != qemu_get_aio_context()) { | |
6792 | aio_context_release(ctx); | |
6793 | } | |
6794 | } | |
6795 | ||
052a7572 FZ |
6796 | void bdrv_coroutine_enter(BlockDriverState *bs, Coroutine *co) |
6797 | { | |
6798 | aio_co_enter(bdrv_get_aio_context(bs), co); | |
6799 | } | |
6800 | ||
e8a095da SH |
6801 | static void bdrv_do_remove_aio_context_notifier(BdrvAioNotifier *ban) |
6802 | { | |
6803 | QLIST_REMOVE(ban, list); | |
6804 | g_free(ban); | |
6805 | } | |
6806 | ||
a3a683c3 | 6807 | static void bdrv_detach_aio_context(BlockDriverState *bs) |
dcd04228 | 6808 | { |
e8a095da | 6809 | BdrvAioNotifier *baf, *baf_tmp; |
33384421 | 6810 | |
e8a095da SH |
6811 | assert(!bs->walking_aio_notifiers); |
6812 | bs->walking_aio_notifiers = true; | |
6813 | QLIST_FOREACH_SAFE(baf, &bs->aio_notifiers, list, baf_tmp) { | |
6814 | if (baf->deleted) { | |
6815 | bdrv_do_remove_aio_context_notifier(baf); | |
6816 | } else { | |
6817 | baf->detach_aio_context(baf->opaque); | |
6818 | } | |
33384421 | 6819 | } |
e8a095da SH |
6820 | /* Never mind iterating again to check for ->deleted. bdrv_close() will |
6821 | * remove remaining aio notifiers if we aren't called again. | |
6822 | */ | |
6823 | bs->walking_aio_notifiers = false; | |
33384421 | 6824 | |
1bffe1ae | 6825 | if (bs->drv && bs->drv->bdrv_detach_aio_context) { |
dcd04228 SH |
6826 | bs->drv->bdrv_detach_aio_context(bs); |
6827 | } | |
dcd04228 | 6828 | |
e64f25f3 KW |
6829 | if (bs->quiesce_counter) { |
6830 | aio_enable_external(bs->aio_context); | |
6831 | } | |
dcd04228 SH |
6832 | bs->aio_context = NULL; |
6833 | } | |
6834 | ||
a3a683c3 KW |
6835 | static void bdrv_attach_aio_context(BlockDriverState *bs, |
6836 | AioContext *new_context) | |
dcd04228 | 6837 | { |
e8a095da | 6838 | BdrvAioNotifier *ban, *ban_tmp; |
33384421 | 6839 | |
e64f25f3 KW |
6840 | if (bs->quiesce_counter) { |
6841 | aio_disable_external(new_context); | |
6842 | } | |
6843 | ||
dcd04228 SH |
6844 | bs->aio_context = new_context; |
6845 | ||
1bffe1ae | 6846 | if (bs->drv && bs->drv->bdrv_attach_aio_context) { |
dcd04228 SH |
6847 | bs->drv->bdrv_attach_aio_context(bs, new_context); |
6848 | } | |
33384421 | 6849 | |
e8a095da SH |
6850 | assert(!bs->walking_aio_notifiers); |
6851 | bs->walking_aio_notifiers = true; | |
6852 | QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_tmp) { | |
6853 | if (ban->deleted) { | |
6854 | bdrv_do_remove_aio_context_notifier(ban); | |
6855 | } else { | |
6856 | ban->attached_aio_context(new_context, ban->opaque); | |
6857 | } | |
33384421 | 6858 | } |
e8a095da | 6859 | bs->walking_aio_notifiers = false; |
dcd04228 SH |
6860 | } |
6861 | ||
42a65f02 KW |
6862 | /* |
6863 | * Changes the AioContext used for fd handlers, timers, and BHs by this | |
6864 | * BlockDriverState and all its children and parents. | |
6865 | * | |
43eaaaef HR |
6866 | * Must be called from the main AioContext. |
6867 | * | |
42a65f02 KW |
6868 | * The caller must own the AioContext lock for the old AioContext of bs, but it |
6869 | * must not own the AioContext lock for new_context (unless new_context is the | |
6870 | * same as the current context of bs). | |
6871 | * | |
6872 | * @ignore will accumulate all visited BdrvChild object. The caller is | |
6873 | * responsible for freeing the list afterwards. | |
6874 | */ | |
53a7d041 KW |
6875 | void bdrv_set_aio_context_ignore(BlockDriverState *bs, |
6876 | AioContext *new_context, GSList **ignore) | |
dcd04228 | 6877 | { |
e037c09c | 6878 | AioContext *old_context = bdrv_get_aio_context(bs); |
722d8e73 SL |
6879 | GSList *children_to_process = NULL; |
6880 | GSList *parents_to_process = NULL; | |
6881 | GSList *entry; | |
6882 | BdrvChild *child, *parent; | |
0d83708a | 6883 | |
43eaaaef HR |
6884 | g_assert(qemu_get_current_aio_context() == qemu_get_aio_context()); |
6885 | ||
e037c09c | 6886 | if (old_context == new_context) { |
57830a49 DP |
6887 | return; |
6888 | } | |
6889 | ||
d70d5954 | 6890 | bdrv_drained_begin(bs); |
0d83708a KW |
6891 | |
6892 | QLIST_FOREACH(child, &bs->children, next) { | |
53a7d041 KW |
6893 | if (g_slist_find(*ignore, child)) { |
6894 | continue; | |
6895 | } | |
6896 | *ignore = g_slist_prepend(*ignore, child); | |
722d8e73 | 6897 | children_to_process = g_slist_prepend(children_to_process, child); |
53a7d041 | 6898 | } |
722d8e73 SL |
6899 | |
6900 | QLIST_FOREACH(parent, &bs->parents, next_parent) { | |
6901 | if (g_slist_find(*ignore, parent)) { | |
53a7d041 KW |
6902 | continue; |
6903 | } | |
722d8e73 SL |
6904 | *ignore = g_slist_prepend(*ignore, parent); |
6905 | parents_to_process = g_slist_prepend(parents_to_process, parent); | |
6906 | } | |
6907 | ||
6908 | for (entry = children_to_process; | |
6909 | entry != NULL; | |
6910 | entry = g_slist_next(entry)) { | |
6911 | child = entry->data; | |
6912 | bdrv_set_aio_context_ignore(child->bs, new_context, ignore); | |
6913 | } | |
6914 | g_slist_free(children_to_process); | |
6915 | ||
6916 | for (entry = parents_to_process; | |
6917 | entry != NULL; | |
6918 | entry = g_slist_next(entry)) { | |
6919 | parent = entry->data; | |
6920 | assert(parent->klass->set_aio_ctx); | |
6921 | parent->klass->set_aio_ctx(parent, new_context, ignore); | |
0d83708a | 6922 | } |
722d8e73 | 6923 | g_slist_free(parents_to_process); |
0d83708a | 6924 | |
dcd04228 SH |
6925 | bdrv_detach_aio_context(bs); |
6926 | ||
e037c09c | 6927 | /* Acquire the new context, if necessary */ |
43eaaaef | 6928 | if (qemu_get_aio_context() != new_context) { |
e037c09c HR |
6929 | aio_context_acquire(new_context); |
6930 | } | |
6931 | ||
dcd04228 | 6932 | bdrv_attach_aio_context(bs, new_context); |
e037c09c HR |
6933 | |
6934 | /* | |
6935 | * If this function was recursively called from | |
6936 | * bdrv_set_aio_context_ignore(), there may be nodes in the | |
6937 | * subtree that have not yet been moved to the new AioContext. | |
6938 | * Release the old one so bdrv_drained_end() can poll them. | |
6939 | */ | |
43eaaaef | 6940 | if (qemu_get_aio_context() != old_context) { |
e037c09c HR |
6941 | aio_context_release(old_context); |
6942 | } | |
6943 | ||
d70d5954 | 6944 | bdrv_drained_end(bs); |
e037c09c | 6945 | |
43eaaaef | 6946 | if (qemu_get_aio_context() != old_context) { |
e037c09c HR |
6947 | aio_context_acquire(old_context); |
6948 | } | |
43eaaaef | 6949 | if (qemu_get_aio_context() != new_context) { |
e037c09c HR |
6950 | aio_context_release(new_context); |
6951 | } | |
85d126f3 | 6952 | } |
d616b224 | 6953 | |
5d231849 KW |
6954 | static bool bdrv_parent_can_set_aio_context(BdrvChild *c, AioContext *ctx, |
6955 | GSList **ignore, Error **errp) | |
6956 | { | |
6957 | if (g_slist_find(*ignore, c)) { | |
6958 | return true; | |
6959 | } | |
6960 | *ignore = g_slist_prepend(*ignore, c); | |
6961 | ||
bd86fb99 HR |
6962 | /* |
6963 | * A BdrvChildClass that doesn't handle AioContext changes cannot | |
6964 | * tolerate any AioContext changes | |
6965 | */ | |
6966 | if (!c->klass->can_set_aio_ctx) { | |
5d231849 KW |
6967 | char *user = bdrv_child_user_desc(c); |
6968 | error_setg(errp, "Changing iothreads is not supported by %s", user); | |
6969 | g_free(user); | |
6970 | return false; | |
6971 | } | |
bd86fb99 | 6972 | if (!c->klass->can_set_aio_ctx(c, ctx, ignore, errp)) { |
5d231849 KW |
6973 | assert(!errp || *errp); |
6974 | return false; | |
6975 | } | |
6976 | return true; | |
6977 | } | |
6978 | ||
6979 | bool bdrv_child_can_set_aio_context(BdrvChild *c, AioContext *ctx, | |
6980 | GSList **ignore, Error **errp) | |
6981 | { | |
6982 | if (g_slist_find(*ignore, c)) { | |
6983 | return true; | |
6984 | } | |
6985 | *ignore = g_slist_prepend(*ignore, c); | |
6986 | return bdrv_can_set_aio_context(c->bs, ctx, ignore, errp); | |
6987 | } | |
6988 | ||
6989 | /* @ignore will accumulate all visited BdrvChild object. The caller is | |
6990 | * responsible for freeing the list afterwards. */ | |
6991 | bool bdrv_can_set_aio_context(BlockDriverState *bs, AioContext *ctx, | |
6992 | GSList **ignore, Error **errp) | |
6993 | { | |
6994 | BdrvChild *c; | |
6995 | ||
6996 | if (bdrv_get_aio_context(bs) == ctx) { | |
6997 | return true; | |
6998 | } | |
6999 | ||
7000 | QLIST_FOREACH(c, &bs->parents, next_parent) { | |
7001 | if (!bdrv_parent_can_set_aio_context(c, ctx, ignore, errp)) { | |
7002 | return false; | |
7003 | } | |
7004 | } | |
7005 | QLIST_FOREACH(c, &bs->children, next) { | |
7006 | if (!bdrv_child_can_set_aio_context(c, ctx, ignore, errp)) { | |
7007 | return false; | |
7008 | } | |
7009 | } | |
7010 | ||
7011 | return true; | |
7012 | } | |
7013 | ||
7014 | int bdrv_child_try_set_aio_context(BlockDriverState *bs, AioContext *ctx, | |
7015 | BdrvChild *ignore_child, Error **errp) | |
7016 | { | |
7017 | GSList *ignore; | |
7018 | bool ret; | |
7019 | ||
7020 | ignore = ignore_child ? g_slist_prepend(NULL, ignore_child) : NULL; | |
7021 | ret = bdrv_can_set_aio_context(bs, ctx, &ignore, errp); | |
7022 | g_slist_free(ignore); | |
7023 | ||
7024 | if (!ret) { | |
7025 | return -EPERM; | |
7026 | } | |
7027 | ||
53a7d041 KW |
7028 | ignore = ignore_child ? g_slist_prepend(NULL, ignore_child) : NULL; |
7029 | bdrv_set_aio_context_ignore(bs, ctx, &ignore); | |
7030 | g_slist_free(ignore); | |
7031 | ||
5d231849 KW |
7032 | return 0; |
7033 | } | |
7034 | ||
7035 | int bdrv_try_set_aio_context(BlockDriverState *bs, AioContext *ctx, | |
7036 | Error **errp) | |
7037 | { | |
7038 | return bdrv_child_try_set_aio_context(bs, ctx, NULL, errp); | |
7039 | } | |
7040 | ||
33384421 HR |
7041 | void bdrv_add_aio_context_notifier(BlockDriverState *bs, |
7042 | void (*attached_aio_context)(AioContext *new_context, void *opaque), | |
7043 | void (*detach_aio_context)(void *opaque), void *opaque) | |
7044 | { | |
7045 | BdrvAioNotifier *ban = g_new(BdrvAioNotifier, 1); | |
7046 | *ban = (BdrvAioNotifier){ | |
7047 | .attached_aio_context = attached_aio_context, | |
7048 | .detach_aio_context = detach_aio_context, | |
7049 | .opaque = opaque | |
7050 | }; | |
7051 | ||
7052 | QLIST_INSERT_HEAD(&bs->aio_notifiers, ban, list); | |
7053 | } | |
7054 | ||
7055 | void bdrv_remove_aio_context_notifier(BlockDriverState *bs, | |
7056 | void (*attached_aio_context)(AioContext *, | |
7057 | void *), | |
7058 | void (*detach_aio_context)(void *), | |
7059 | void *opaque) | |
7060 | { | |
7061 | BdrvAioNotifier *ban, *ban_next; | |
7062 | ||
7063 | QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_next) { | |
7064 | if (ban->attached_aio_context == attached_aio_context && | |
7065 | ban->detach_aio_context == detach_aio_context && | |
e8a095da SH |
7066 | ban->opaque == opaque && |
7067 | ban->deleted == false) | |
33384421 | 7068 | { |
e8a095da SH |
7069 | if (bs->walking_aio_notifiers) { |
7070 | ban->deleted = true; | |
7071 | } else { | |
7072 | bdrv_do_remove_aio_context_notifier(ban); | |
7073 | } | |
33384421 HR |
7074 | return; |
7075 | } | |
7076 | } | |
7077 | ||
7078 | abort(); | |
7079 | } | |
7080 | ||
77485434 | 7081 | int bdrv_amend_options(BlockDriverState *bs, QemuOpts *opts, |
d1402b50 | 7082 | BlockDriverAmendStatusCB *status_cb, void *cb_opaque, |
a3579bfa | 7083 | bool force, |
d1402b50 | 7084 | Error **errp) |
6f176b48 | 7085 | { |
d470ad42 | 7086 | if (!bs->drv) { |
d1402b50 | 7087 | error_setg(errp, "Node is ejected"); |
d470ad42 HR |
7088 | return -ENOMEDIUM; |
7089 | } | |
c282e1fd | 7090 | if (!bs->drv->bdrv_amend_options) { |
d1402b50 HR |
7091 | error_setg(errp, "Block driver '%s' does not support option amendment", |
7092 | bs->drv->format_name); | |
6f176b48 HR |
7093 | return -ENOTSUP; |
7094 | } | |
a3579bfa ML |
7095 | return bs->drv->bdrv_amend_options(bs, opts, status_cb, |
7096 | cb_opaque, force, errp); | |
6f176b48 | 7097 | } |
f6186f49 | 7098 | |
5d69b5ab HR |
7099 | /* |
7100 | * This function checks whether the given @to_replace is allowed to be | |
7101 | * replaced by a node that always shows the same data as @bs. This is | |
7102 | * used for example to verify whether the mirror job can replace | |
7103 | * @to_replace by the target mirrored from @bs. | |
7104 | * To be replaceable, @bs and @to_replace may either be guaranteed to | |
7105 | * always show the same data (because they are only connected through | |
7106 | * filters), or some driver may allow replacing one of its children | |
7107 | * because it can guarantee that this child's data is not visible at | |
7108 | * all (for example, for dissenting quorum children that have no other | |
7109 | * parents). | |
7110 | */ | |
7111 | bool bdrv_recurse_can_replace(BlockDriverState *bs, | |
7112 | BlockDriverState *to_replace) | |
7113 | { | |
93393e69 HR |
7114 | BlockDriverState *filtered; |
7115 | ||
5d69b5ab HR |
7116 | if (!bs || !bs->drv) { |
7117 | return false; | |
7118 | } | |
7119 | ||
7120 | if (bs == to_replace) { | |
7121 | return true; | |
7122 | } | |
7123 | ||
7124 | /* See what the driver can do */ | |
7125 | if (bs->drv->bdrv_recurse_can_replace) { | |
7126 | return bs->drv->bdrv_recurse_can_replace(bs, to_replace); | |
7127 | } | |
7128 | ||
7129 | /* For filters without an own implementation, we can recurse on our own */ | |
93393e69 HR |
7130 | filtered = bdrv_filter_bs(bs); |
7131 | if (filtered) { | |
7132 | return bdrv_recurse_can_replace(filtered, to_replace); | |
5d69b5ab HR |
7133 | } |
7134 | ||
7135 | /* Safe default */ | |
7136 | return false; | |
7137 | } | |
7138 | ||
810803a8 HR |
7139 | /* |
7140 | * Check whether the given @node_name can be replaced by a node that | |
7141 | * has the same data as @parent_bs. If so, return @node_name's BDS; | |
7142 | * NULL otherwise. | |
7143 | * | |
7144 | * @node_name must be a (recursive) *child of @parent_bs (or this | |
7145 | * function will return NULL). | |
7146 | * | |
7147 | * The result (whether the node can be replaced or not) is only valid | |
7148 | * for as long as no graph or permission changes occur. | |
7149 | */ | |
e12f3784 WC |
7150 | BlockDriverState *check_to_replace_node(BlockDriverState *parent_bs, |
7151 | const char *node_name, Error **errp) | |
09158f00 BC |
7152 | { |
7153 | BlockDriverState *to_replace_bs = bdrv_find_node(node_name); | |
5a7e7a0b SH |
7154 | AioContext *aio_context; |
7155 | ||
09158f00 | 7156 | if (!to_replace_bs) { |
785ec4b1 | 7157 | error_setg(errp, "Failed to find node with node-name='%s'", node_name); |
09158f00 BC |
7158 | return NULL; |
7159 | } | |
7160 | ||
5a7e7a0b SH |
7161 | aio_context = bdrv_get_aio_context(to_replace_bs); |
7162 | aio_context_acquire(aio_context); | |
7163 | ||
09158f00 | 7164 | if (bdrv_op_is_blocked(to_replace_bs, BLOCK_OP_TYPE_REPLACE, errp)) { |
5a7e7a0b SH |
7165 | to_replace_bs = NULL; |
7166 | goto out; | |
09158f00 BC |
7167 | } |
7168 | ||
7169 | /* We don't want arbitrary node of the BDS chain to be replaced only the top | |
7170 | * most non filter in order to prevent data corruption. | |
7171 | * Another benefit is that this tests exclude backing files which are | |
7172 | * blocked by the backing blockers. | |
7173 | */ | |
810803a8 HR |
7174 | if (!bdrv_recurse_can_replace(parent_bs, to_replace_bs)) { |
7175 | error_setg(errp, "Cannot replace '%s' by a node mirrored from '%s', " | |
7176 | "because it cannot be guaranteed that doing so would not " | |
7177 | "lead to an abrupt change of visible data", | |
7178 | node_name, parent_bs->node_name); | |
5a7e7a0b SH |
7179 | to_replace_bs = NULL; |
7180 | goto out; | |
09158f00 BC |
7181 | } |
7182 | ||
5a7e7a0b SH |
7183 | out: |
7184 | aio_context_release(aio_context); | |
09158f00 BC |
7185 | return to_replace_bs; |
7186 | } | |
448ad91d | 7187 | |
97e2f021 HR |
7188 | /** |
7189 | * Iterates through the list of runtime option keys that are said to | |
7190 | * be "strong" for a BDS. An option is called "strong" if it changes | |
7191 | * a BDS's data. For example, the null block driver's "size" and | |
7192 | * "read-zeroes" options are strong, but its "latency-ns" option is | |
7193 | * not. | |
7194 | * | |
7195 | * If a key returned by this function ends with a dot, all options | |
7196 | * starting with that prefix are strong. | |
7197 | */ | |
7198 | static const char *const *strong_options(BlockDriverState *bs, | |
7199 | const char *const *curopt) | |
7200 | { | |
7201 | static const char *const global_options[] = { | |
7202 | "driver", "filename", NULL | |
7203 | }; | |
7204 | ||
7205 | if (!curopt) { | |
7206 | return &global_options[0]; | |
7207 | } | |
7208 | ||
7209 | curopt++; | |
7210 | if (curopt == &global_options[ARRAY_SIZE(global_options) - 1] && bs->drv) { | |
7211 | curopt = bs->drv->strong_runtime_opts; | |
7212 | } | |
7213 | ||
7214 | return (curopt && *curopt) ? curopt : NULL; | |
7215 | } | |
7216 | ||
7217 | /** | |
7218 | * Copies all strong runtime options from bs->options to the given | |
7219 | * QDict. The set of strong option keys is determined by invoking | |
7220 | * strong_options(). | |
7221 | * | |
7222 | * Returns true iff any strong option was present in bs->options (and | |
7223 | * thus copied to the target QDict) with the exception of "filename" | |
7224 | * and "driver". The caller is expected to use this value to decide | |
7225 | * whether the existence of strong options prevents the generation of | |
7226 | * a plain filename. | |
7227 | */ | |
7228 | static bool append_strong_runtime_options(QDict *d, BlockDriverState *bs) | |
7229 | { | |
7230 | bool found_any = false; | |
7231 | const char *const *option_name = NULL; | |
7232 | ||
7233 | if (!bs->drv) { | |
7234 | return false; | |
7235 | } | |
7236 | ||
7237 | while ((option_name = strong_options(bs, option_name))) { | |
7238 | bool option_given = false; | |
7239 | ||
7240 | assert(strlen(*option_name) > 0); | |
7241 | if ((*option_name)[strlen(*option_name) - 1] != '.') { | |
7242 | QObject *entry = qdict_get(bs->options, *option_name); | |
7243 | if (!entry) { | |
7244 | continue; | |
7245 | } | |
7246 | ||
7247 | qdict_put_obj(d, *option_name, qobject_ref(entry)); | |
7248 | option_given = true; | |
7249 | } else { | |
7250 | const QDictEntry *entry; | |
7251 | for (entry = qdict_first(bs->options); entry; | |
7252 | entry = qdict_next(bs->options, entry)) | |
7253 | { | |
7254 | if (strstart(qdict_entry_key(entry), *option_name, NULL)) { | |
7255 | qdict_put_obj(d, qdict_entry_key(entry), | |
7256 | qobject_ref(qdict_entry_value(entry))); | |
7257 | option_given = true; | |
7258 | } | |
7259 | } | |
7260 | } | |
7261 | ||
7262 | /* While "driver" and "filename" need to be included in a JSON filename, | |
7263 | * their existence does not prohibit generation of a plain filename. */ | |
7264 | if (!found_any && option_given && | |
7265 | strcmp(*option_name, "driver") && strcmp(*option_name, "filename")) | |
7266 | { | |
7267 | found_any = true; | |
7268 | } | |
7269 | } | |
7270 | ||
62a01a27 HR |
7271 | if (!qdict_haskey(d, "driver")) { |
7272 | /* Drivers created with bdrv_new_open_driver() may not have a | |
7273 | * @driver option. Add it here. */ | |
7274 | qdict_put_str(d, "driver", bs->drv->format_name); | |
7275 | } | |
7276 | ||
97e2f021 HR |
7277 | return found_any; |
7278 | } | |
7279 | ||
90993623 HR |
7280 | /* Note: This function may return false positives; it may return true |
7281 | * even if opening the backing file specified by bs's image header | |
7282 | * would result in exactly bs->backing. */ | |
0b877d09 | 7283 | bool bdrv_backing_overridden(BlockDriverState *bs) |
90993623 HR |
7284 | { |
7285 | if (bs->backing) { | |
7286 | return strcmp(bs->auto_backing_file, | |
7287 | bs->backing->bs->filename); | |
7288 | } else { | |
7289 | /* No backing BDS, so if the image header reports any backing | |
7290 | * file, it must have been suppressed */ | |
7291 | return bs->auto_backing_file[0] != '\0'; | |
7292 | } | |
7293 | } | |
7294 | ||
91af7014 HR |
7295 | /* Updates the following BDS fields: |
7296 | * - exact_filename: A filename which may be used for opening a block device | |
7297 | * which (mostly) equals the given BDS (even without any | |
7298 | * other options; so reading and writing must return the same | |
7299 | * results, but caching etc. may be different) | |
7300 | * - full_open_options: Options which, when given when opening a block device | |
7301 | * (without a filename), result in a BDS (mostly) | |
7302 | * equalling the given one | |
7303 | * - filename: If exact_filename is set, it is copied here. Otherwise, | |
7304 | * full_open_options is converted to a JSON object, prefixed with | |
7305 | * "json:" (for use through the JSON pseudo protocol) and put here. | |
7306 | */ | |
7307 | void bdrv_refresh_filename(BlockDriverState *bs) | |
7308 | { | |
7309 | BlockDriver *drv = bs->drv; | |
e24518e3 | 7310 | BdrvChild *child; |
52f72d6f | 7311 | BlockDriverState *primary_child_bs; |
91af7014 | 7312 | QDict *opts; |
90993623 | 7313 | bool backing_overridden; |
998b3a1e HR |
7314 | bool generate_json_filename; /* Whether our default implementation should |
7315 | fill exact_filename (false) or not (true) */ | |
91af7014 HR |
7316 | |
7317 | if (!drv) { | |
7318 | return; | |
7319 | } | |
7320 | ||
e24518e3 HR |
7321 | /* This BDS's file name may depend on any of its children's file names, so |
7322 | * refresh those first */ | |
7323 | QLIST_FOREACH(child, &bs->children, next) { | |
7324 | bdrv_refresh_filename(child->bs); | |
91af7014 HR |
7325 | } |
7326 | ||
bb808d5f HR |
7327 | if (bs->implicit) { |
7328 | /* For implicit nodes, just copy everything from the single child */ | |
7329 | child = QLIST_FIRST(&bs->children); | |
7330 | assert(QLIST_NEXT(child, next) == NULL); | |
7331 | ||
7332 | pstrcpy(bs->exact_filename, sizeof(bs->exact_filename), | |
7333 | child->bs->exact_filename); | |
7334 | pstrcpy(bs->filename, sizeof(bs->filename), child->bs->filename); | |
7335 | ||
cb895614 | 7336 | qobject_unref(bs->full_open_options); |
bb808d5f HR |
7337 | bs->full_open_options = qobject_ref(child->bs->full_open_options); |
7338 | ||
7339 | return; | |
7340 | } | |
7341 | ||
90993623 HR |
7342 | backing_overridden = bdrv_backing_overridden(bs); |
7343 | ||
7344 | if (bs->open_flags & BDRV_O_NO_IO) { | |
7345 | /* Without I/O, the backing file does not change anything. | |
7346 | * Therefore, in such a case (primarily qemu-img), we can | |
7347 | * pretend the backing file has not been overridden even if | |
7348 | * it technically has been. */ | |
7349 | backing_overridden = false; | |
7350 | } | |
7351 | ||
97e2f021 HR |
7352 | /* Gather the options QDict */ |
7353 | opts = qdict_new(); | |
998b3a1e HR |
7354 | generate_json_filename = append_strong_runtime_options(opts, bs); |
7355 | generate_json_filename |= backing_overridden; | |
97e2f021 HR |
7356 | |
7357 | if (drv->bdrv_gather_child_options) { | |
7358 | /* Some block drivers may not want to present all of their children's | |
7359 | * options, or name them differently from BdrvChild.name */ | |
7360 | drv->bdrv_gather_child_options(bs, opts, backing_overridden); | |
7361 | } else { | |
7362 | QLIST_FOREACH(child, &bs->children, next) { | |
25191e5f | 7363 | if (child == bs->backing && !backing_overridden) { |
97e2f021 HR |
7364 | /* We can skip the backing BDS if it has not been overridden */ |
7365 | continue; | |
7366 | } | |
7367 | ||
7368 | qdict_put(opts, child->name, | |
7369 | qobject_ref(child->bs->full_open_options)); | |
7370 | } | |
7371 | ||
7372 | if (backing_overridden && !bs->backing) { | |
7373 | /* Force no backing file */ | |
7374 | qdict_put_null(opts, "backing"); | |
7375 | } | |
7376 | } | |
7377 | ||
7378 | qobject_unref(bs->full_open_options); | |
7379 | bs->full_open_options = opts; | |
7380 | ||
52f72d6f HR |
7381 | primary_child_bs = bdrv_primary_bs(bs); |
7382 | ||
998b3a1e HR |
7383 | if (drv->bdrv_refresh_filename) { |
7384 | /* Obsolete information is of no use here, so drop the old file name | |
7385 | * information before refreshing it */ | |
7386 | bs->exact_filename[0] = '\0'; | |
7387 | ||
7388 | drv->bdrv_refresh_filename(bs); | |
52f72d6f HR |
7389 | } else if (primary_child_bs) { |
7390 | /* | |
7391 | * Try to reconstruct valid information from the underlying | |
7392 | * file -- this only works for format nodes (filter nodes | |
7393 | * cannot be probed and as such must be selected by the user | |
7394 | * either through an options dict, or through a special | |
7395 | * filename which the filter driver must construct in its | |
7396 | * .bdrv_refresh_filename() implementation). | |
7397 | */ | |
998b3a1e HR |
7398 | |
7399 | bs->exact_filename[0] = '\0'; | |
7400 | ||
fb695c74 HR |
7401 | /* |
7402 | * We can use the underlying file's filename if: | |
7403 | * - it has a filename, | |
52f72d6f | 7404 | * - the current BDS is not a filter, |
fb695c74 HR |
7405 | * - the file is a protocol BDS, and |
7406 | * - opening that file (as this BDS's format) will automatically create | |
7407 | * the BDS tree we have right now, that is: | |
7408 | * - the user did not significantly change this BDS's behavior with | |
7409 | * some explicit (strong) options | |
7410 | * - no non-file child of this BDS has been overridden by the user | |
7411 | * Both of these conditions are represented by generate_json_filename. | |
7412 | */ | |
52f72d6f HR |
7413 | if (primary_child_bs->exact_filename[0] && |
7414 | primary_child_bs->drv->bdrv_file_open && | |
7415 | !drv->is_filter && !generate_json_filename) | |
fb695c74 | 7416 | { |
52f72d6f | 7417 | strcpy(bs->exact_filename, primary_child_bs->exact_filename); |
998b3a1e HR |
7418 | } |
7419 | } | |
7420 | ||
91af7014 HR |
7421 | if (bs->exact_filename[0]) { |
7422 | pstrcpy(bs->filename, sizeof(bs->filename), bs->exact_filename); | |
97e2f021 | 7423 | } else { |
eab3a467 | 7424 | GString *json = qobject_to_json(QOBJECT(bs->full_open_options)); |
5c86bdf1 | 7425 | if (snprintf(bs->filename, sizeof(bs->filename), "json:%s", |
eab3a467 | 7426 | json->str) >= sizeof(bs->filename)) { |
5c86bdf1 EB |
7427 | /* Give user a hint if we truncated things. */ |
7428 | strcpy(bs->filename + sizeof(bs->filename) - 4, "..."); | |
7429 | } | |
eab3a467 | 7430 | g_string_free(json, true); |
91af7014 HR |
7431 | } |
7432 | } | |
e06018ad | 7433 | |
1e89d0f9 HR |
7434 | char *bdrv_dirname(BlockDriverState *bs, Error **errp) |
7435 | { | |
7436 | BlockDriver *drv = bs->drv; | |
52f72d6f | 7437 | BlockDriverState *child_bs; |
1e89d0f9 HR |
7438 | |
7439 | if (!drv) { | |
7440 | error_setg(errp, "Node '%s' is ejected", bs->node_name); | |
7441 | return NULL; | |
7442 | } | |
7443 | ||
7444 | if (drv->bdrv_dirname) { | |
7445 | return drv->bdrv_dirname(bs, errp); | |
7446 | } | |
7447 | ||
52f72d6f HR |
7448 | child_bs = bdrv_primary_bs(bs); |
7449 | if (child_bs) { | |
7450 | return bdrv_dirname(child_bs, errp); | |
1e89d0f9 HR |
7451 | } |
7452 | ||
7453 | bdrv_refresh_filename(bs); | |
7454 | if (bs->exact_filename[0] != '\0') { | |
7455 | return path_combine(bs->exact_filename, ""); | |
7456 | } | |
7457 | ||
7458 | error_setg(errp, "Cannot generate a base directory for %s nodes", | |
7459 | drv->format_name); | |
7460 | return NULL; | |
7461 | } | |
7462 | ||
e06018ad WC |
7463 | /* |
7464 | * Hot add/remove a BDS's child. So the user can take a child offline when | |
7465 | * it is broken and take a new child online | |
7466 | */ | |
7467 | void bdrv_add_child(BlockDriverState *parent_bs, BlockDriverState *child_bs, | |
7468 | Error **errp) | |
7469 | { | |
7470 | ||
7471 | if (!parent_bs->drv || !parent_bs->drv->bdrv_add_child) { | |
7472 | error_setg(errp, "The node %s does not support adding a child", | |
7473 | bdrv_get_device_or_node_name(parent_bs)); | |
7474 | return; | |
7475 | } | |
7476 | ||
7477 | if (!QLIST_EMPTY(&child_bs->parents)) { | |
7478 | error_setg(errp, "The node %s already has a parent", | |
7479 | child_bs->node_name); | |
7480 | return; | |
7481 | } | |
7482 | ||
7483 | parent_bs->drv->bdrv_add_child(parent_bs, child_bs, errp); | |
7484 | } | |
7485 | ||
7486 | void bdrv_del_child(BlockDriverState *parent_bs, BdrvChild *child, Error **errp) | |
7487 | { | |
7488 | BdrvChild *tmp; | |
7489 | ||
7490 | if (!parent_bs->drv || !parent_bs->drv->bdrv_del_child) { | |
7491 | error_setg(errp, "The node %s does not support removing a child", | |
7492 | bdrv_get_device_or_node_name(parent_bs)); | |
7493 | return; | |
7494 | } | |
7495 | ||
7496 | QLIST_FOREACH(tmp, &parent_bs->children, next) { | |
7497 | if (tmp == child) { | |
7498 | break; | |
7499 | } | |
7500 | } | |
7501 | ||
7502 | if (!tmp) { | |
7503 | error_setg(errp, "The node %s does not have a child named %s", | |
7504 | bdrv_get_device_or_node_name(parent_bs), | |
7505 | bdrv_get_device_or_node_name(child->bs)); | |
7506 | return; | |
7507 | } | |
7508 | ||
7509 | parent_bs->drv->bdrv_del_child(parent_bs, child, errp); | |
7510 | } | |
6f7a3b53 HR |
7511 | |
7512 | int bdrv_make_empty(BdrvChild *c, Error **errp) | |
7513 | { | |
7514 | BlockDriver *drv = c->bs->drv; | |
7515 | int ret; | |
7516 | ||
7517 | assert(c->perm & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)); | |
7518 | ||
7519 | if (!drv->bdrv_make_empty) { | |
7520 | error_setg(errp, "%s does not support emptying nodes", | |
7521 | drv->format_name); | |
7522 | return -ENOTSUP; | |
7523 | } | |
7524 | ||
7525 | ret = drv->bdrv_make_empty(c->bs); | |
7526 | if (ret < 0) { | |
7527 | error_setg_errno(errp, -ret, "Failed to empty %s", | |
7528 | c->bs->filename); | |
7529 | return ret; | |
7530 | } | |
7531 | ||
7532 | return 0; | |
7533 | } | |
9a6fc887 HR |
7534 | |
7535 | /* | |
7536 | * Return the child that @bs acts as an overlay for, and from which data may be | |
7537 | * copied in COW or COR operations. Usually this is the backing file. | |
7538 | */ | |
7539 | BdrvChild *bdrv_cow_child(BlockDriverState *bs) | |
7540 | { | |
7541 | if (!bs || !bs->drv) { | |
7542 | return NULL; | |
7543 | } | |
7544 | ||
7545 | if (bs->drv->is_filter) { | |
7546 | return NULL; | |
7547 | } | |
7548 | ||
7549 | if (!bs->backing) { | |
7550 | return NULL; | |
7551 | } | |
7552 | ||
7553 | assert(bs->backing->role & BDRV_CHILD_COW); | |
7554 | return bs->backing; | |
7555 | } | |
7556 | ||
7557 | /* | |
7558 | * If @bs acts as a filter for exactly one of its children, return | |
7559 | * that child. | |
7560 | */ | |
7561 | BdrvChild *bdrv_filter_child(BlockDriverState *bs) | |
7562 | { | |
7563 | BdrvChild *c; | |
7564 | ||
7565 | if (!bs || !bs->drv) { | |
7566 | return NULL; | |
7567 | } | |
7568 | ||
7569 | if (!bs->drv->is_filter) { | |
7570 | return NULL; | |
7571 | } | |
7572 | ||
7573 | /* Only one of @backing or @file may be used */ | |
7574 | assert(!(bs->backing && bs->file)); | |
7575 | ||
7576 | c = bs->backing ?: bs->file; | |
7577 | if (!c) { | |
7578 | return NULL; | |
7579 | } | |
7580 | ||
7581 | assert(c->role & BDRV_CHILD_FILTERED); | |
7582 | return c; | |
7583 | } | |
7584 | ||
7585 | /* | |
7586 | * Return either the result of bdrv_cow_child() or bdrv_filter_child(), | |
7587 | * whichever is non-NULL. | |
7588 | * | |
7589 | * Return NULL if both are NULL. | |
7590 | */ | |
7591 | BdrvChild *bdrv_filter_or_cow_child(BlockDriverState *bs) | |
7592 | { | |
7593 | BdrvChild *cow_child = bdrv_cow_child(bs); | |
7594 | BdrvChild *filter_child = bdrv_filter_child(bs); | |
7595 | ||
7596 | /* Filter nodes cannot have COW backing files */ | |
7597 | assert(!(cow_child && filter_child)); | |
7598 | ||
7599 | return cow_child ?: filter_child; | |
7600 | } | |
7601 | ||
7602 | /* | |
7603 | * Return the primary child of this node: For filters, that is the | |
7604 | * filtered child. For other nodes, that is usually the child storing | |
7605 | * metadata. | |
7606 | * (A generally more helpful description is that this is (usually) the | |
7607 | * child that has the same filename as @bs.) | |
7608 | * | |
7609 | * Drivers do not necessarily have a primary child; for example quorum | |
7610 | * does not. | |
7611 | */ | |
7612 | BdrvChild *bdrv_primary_child(BlockDriverState *bs) | |
7613 | { | |
7614 | BdrvChild *c, *found = NULL; | |
7615 | ||
7616 | QLIST_FOREACH(c, &bs->children, next) { | |
7617 | if (c->role & BDRV_CHILD_PRIMARY) { | |
7618 | assert(!found); | |
7619 | found = c; | |
7620 | } | |
7621 | } | |
7622 | ||
7623 | return found; | |
7624 | } | |
d38d7eb8 HR |
7625 | |
7626 | static BlockDriverState *bdrv_do_skip_filters(BlockDriverState *bs, | |
7627 | bool stop_on_explicit_filter) | |
7628 | { | |
7629 | BdrvChild *c; | |
7630 | ||
7631 | if (!bs) { | |
7632 | return NULL; | |
7633 | } | |
7634 | ||
7635 | while (!(stop_on_explicit_filter && !bs->implicit)) { | |
7636 | c = bdrv_filter_child(bs); | |
7637 | if (!c) { | |
7638 | /* | |
7639 | * A filter that is embedded in a working block graph must | |
7640 | * have a child. Assert this here so this function does | |
7641 | * not return a filter node that is not expected by the | |
7642 | * caller. | |
7643 | */ | |
7644 | assert(!bs->drv || !bs->drv->is_filter); | |
7645 | break; | |
7646 | } | |
7647 | bs = c->bs; | |
7648 | } | |
7649 | /* | |
7650 | * Note that this treats nodes with bs->drv == NULL as not being | |
7651 | * filters (bs->drv == NULL should be replaced by something else | |
7652 | * anyway). | |
7653 | * The advantage of this behavior is that this function will thus | |
7654 | * always return a non-NULL value (given a non-NULL @bs). | |
7655 | */ | |
7656 | ||
7657 | return bs; | |
7658 | } | |
7659 | ||
7660 | /* | |
7661 | * Return the first BDS that has not been added implicitly or that | |
7662 | * does not have a filtered child down the chain starting from @bs | |
7663 | * (including @bs itself). | |
7664 | */ | |
7665 | BlockDriverState *bdrv_skip_implicit_filters(BlockDriverState *bs) | |
7666 | { | |
7667 | return bdrv_do_skip_filters(bs, true); | |
7668 | } | |
7669 | ||
7670 | /* | |
7671 | * Return the first BDS that does not have a filtered child down the | |
7672 | * chain starting from @bs (including @bs itself). | |
7673 | */ | |
7674 | BlockDriverState *bdrv_skip_filters(BlockDriverState *bs) | |
7675 | { | |
7676 | return bdrv_do_skip_filters(bs, false); | |
7677 | } | |
7678 | ||
7679 | /* | |
7680 | * For a backing chain, return the first non-filter backing image of | |
7681 | * the first non-filter image. | |
7682 | */ | |
7683 | BlockDriverState *bdrv_backing_chain_next(BlockDriverState *bs) | |
7684 | { | |
7685 | return bdrv_skip_filters(bdrv_cow_bs(bdrv_skip_filters(bs))); | |
7686 | } |