]>
Commit | Line | Data |
---|---|---|
1 | /* | |
2 | * QEMU System Emulator block driver | |
3 | * | |
4 | * Copyright (c) 2003 Fabrice Bellard | |
5 | * Copyright (c) 2020 Virtuozzo International GmbH. | |
6 | * | |
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 | */ | |
25 | ||
26 | #include "qemu/osdep.h" | |
27 | #include "block/trace.h" | |
28 | #include "block/block_int.h" | |
29 | #include "block/blockjob.h" | |
30 | #include "block/dirty-bitmap.h" | |
31 | #include "block/fuse.h" | |
32 | #include "block/nbd.h" | |
33 | #include "block/qdict.h" | |
34 | #include "qemu/error-report.h" | |
35 | #include "block/module_block.h" | |
36 | #include "qemu/main-loop.h" | |
37 | #include "qemu/module.h" | |
38 | #include "qapi/error.h" | |
39 | #include "qapi/qmp/qdict.h" | |
40 | #include "qapi/qmp/qjson.h" | |
41 | #include "qapi/qmp/qnull.h" | |
42 | #include "qapi/qmp/qstring.h" | |
43 | #include "qapi/qobject-output-visitor.h" | |
44 | #include "qapi/qapi-visit-block-core.h" | |
45 | #include "sysemu/block-backend.h" | |
46 | #include "qemu/notify.h" | |
47 | #include "qemu/option.h" | |
48 | #include "qemu/coroutine.h" | |
49 | #include "block/qapi.h" | |
50 | #include "qemu/timer.h" | |
51 | #include "qemu/cutils.h" | |
52 | #include "qemu/id.h" | |
53 | #include "qemu/range.h" | |
54 | #include "qemu/rcu.h" | |
55 | #include "block/coroutines.h" | |
56 | ||
57 | #ifdef CONFIG_BSD | |
58 | #include <sys/ioctl.h> | |
59 | #include <sys/queue.h> | |
60 | #if defined(HAVE_SYS_DISK_H) | |
61 | #include <sys/disk.h> | |
62 | #endif | |
63 | #endif | |
64 | ||
65 | #ifdef _WIN32 | |
66 | #include <windows.h> | |
67 | #endif | |
68 | ||
69 | #define NOT_DONE 0x7fffffff /* used while emulated sync operation in progress */ | |
70 | ||
71 | /* Protected by BQL */ | |
72 | static QTAILQ_HEAD(, BlockDriverState) graph_bdrv_states = | |
73 | QTAILQ_HEAD_INITIALIZER(graph_bdrv_states); | |
74 | ||
75 | /* Protected by BQL */ | |
76 | static QTAILQ_HEAD(, BlockDriverState) all_bdrv_states = | |
77 | QTAILQ_HEAD_INITIALIZER(all_bdrv_states); | |
78 | ||
79 | /* Protected by BQL */ | |
80 | static QLIST_HEAD(, BlockDriver) bdrv_drivers = | |
81 | QLIST_HEAD_INITIALIZER(bdrv_drivers); | |
82 | ||
83 | static BlockDriverState *bdrv_open_inherit(const char *filename, | |
84 | const char *reference, | |
85 | QDict *options, int flags, | |
86 | BlockDriverState *parent, | |
87 | const BdrvChildClass *child_class, | |
88 | BdrvChildRole child_role, | |
89 | Error **errp); | |
90 | ||
91 | static bool bdrv_recurse_has_child(BlockDriverState *bs, | |
92 | BlockDriverState *child); | |
93 | ||
94 | static void GRAPH_WRLOCK | |
95 | bdrv_replace_child_noperm(BdrvChild *child, BlockDriverState *new_bs); | |
96 | ||
97 | static void GRAPH_WRLOCK | |
98 | bdrv_remove_child(BdrvChild *child, Transaction *tran); | |
99 | ||
100 | static int bdrv_reopen_prepare(BDRVReopenState *reopen_state, | |
101 | BlockReopenQueue *queue, | |
102 | Transaction *change_child_tran, Error **errp); | |
103 | static void bdrv_reopen_commit(BDRVReopenState *reopen_state); | |
104 | static void bdrv_reopen_abort(BDRVReopenState *reopen_state); | |
105 | ||
106 | static bool bdrv_backing_overridden(BlockDriverState *bs); | |
107 | ||
108 | static bool bdrv_change_aio_context(BlockDriverState *bs, AioContext *ctx, | |
109 | GHashTable *visited, Transaction *tran, | |
110 | Error **errp); | |
111 | ||
112 | /* If non-zero, use only whitelisted block drivers */ | |
113 | static int use_bdrv_whitelist; | |
114 | ||
115 | #ifdef _WIN32 | |
116 | static int is_windows_drive_prefix(const char *filename) | |
117 | { | |
118 | return (((filename[0] >= 'a' && filename[0] <= 'z') || | |
119 | (filename[0] >= 'A' && filename[0] <= 'Z')) && | |
120 | filename[1] == ':'); | |
121 | } | |
122 | ||
123 | int is_windows_drive(const char *filename) | |
124 | { | |
125 | if (is_windows_drive_prefix(filename) && | |
126 | filename[2] == '\0') | |
127 | return 1; | |
128 | if (strstart(filename, "\\\\.\\", NULL) || | |
129 | strstart(filename, "//./", NULL)) | |
130 | return 1; | |
131 | return 0; | |
132 | } | |
133 | #endif | |
134 | ||
135 | size_t bdrv_opt_mem_align(BlockDriverState *bs) | |
136 | { | |
137 | if (!bs || !bs->drv) { | |
138 | /* page size or 4k (hdd sector size) should be on the safe side */ | |
139 | return MAX(4096, qemu_real_host_page_size()); | |
140 | } | |
141 | IO_CODE(); | |
142 | ||
143 | return bs->bl.opt_mem_alignment; | |
144 | } | |
145 | ||
146 | size_t bdrv_min_mem_align(BlockDriverState *bs) | |
147 | { | |
148 | if (!bs || !bs->drv) { | |
149 | /* page size or 4k (hdd sector size) should be on the safe side */ | |
150 | return MAX(4096, qemu_real_host_page_size()); | |
151 | } | |
152 | IO_CODE(); | |
153 | ||
154 | return bs->bl.min_mem_alignment; | |
155 | } | |
156 | ||
157 | /* check if the path starts with "<protocol>:" */ | |
158 | int path_has_protocol(const char *path) | |
159 | { | |
160 | const char *p; | |
161 | ||
162 | #ifdef _WIN32 | |
163 | if (is_windows_drive(path) || | |
164 | is_windows_drive_prefix(path)) { | |
165 | return 0; | |
166 | } | |
167 | p = path + strcspn(path, ":/\\"); | |
168 | #else | |
169 | p = path + strcspn(path, ":/"); | |
170 | #endif | |
171 | ||
172 | return *p == ':'; | |
173 | } | |
174 | ||
175 | int path_is_absolute(const char *path) | |
176 | { | |
177 | #ifdef _WIN32 | |
178 | /* specific case for names like: "\\.\d:" */ | |
179 | if (is_windows_drive(path) || is_windows_drive_prefix(path)) { | |
180 | return 1; | |
181 | } | |
182 | return (*path == '/' || *path == '\\'); | |
183 | #else | |
184 | return (*path == '/'); | |
185 | #endif | |
186 | } | |
187 | ||
188 | /* if filename is absolute, just return its duplicate. Otherwise, build a | |
189 | path to it by considering it is relative to base_path. URL are | |
190 | supported. */ | |
191 | char *path_combine(const char *base_path, const char *filename) | |
192 | { | |
193 | const char *protocol_stripped = NULL; | |
194 | const char *p, *p1; | |
195 | char *result; | |
196 | int len; | |
197 | ||
198 | if (path_is_absolute(filename)) { | |
199 | return g_strdup(filename); | |
200 | } | |
201 | ||
202 | if (path_has_protocol(base_path)) { | |
203 | protocol_stripped = strchr(base_path, ':'); | |
204 | if (protocol_stripped) { | |
205 | protocol_stripped++; | |
206 | } | |
207 | } | |
208 | p = protocol_stripped ?: base_path; | |
209 | ||
210 | p1 = strrchr(base_path, '/'); | |
211 | #ifdef _WIN32 | |
212 | { | |
213 | const char *p2; | |
214 | p2 = strrchr(base_path, '\\'); | |
215 | if (!p1 || p2 > p1) { | |
216 | p1 = p2; | |
217 | } | |
218 | } | |
219 | #endif | |
220 | if (p1) { | |
221 | p1++; | |
222 | } else { | |
223 | p1 = base_path; | |
224 | } | |
225 | if (p1 > p) { | |
226 | p = p1; | |
227 | } | |
228 | len = p - base_path; | |
229 | ||
230 | result = g_malloc(len + strlen(filename) + 1); | |
231 | memcpy(result, base_path, len); | |
232 | strcpy(result + len, filename); | |
233 | ||
234 | return result; | |
235 | } | |
236 | ||
237 | /* | |
238 | * Helper function for bdrv_parse_filename() implementations to remove optional | |
239 | * protocol prefixes (especially "file:") from a filename and for putting the | |
240 | * stripped filename into the options QDict if there is such a prefix. | |
241 | */ | |
242 | void bdrv_parse_filename_strip_prefix(const char *filename, const char *prefix, | |
243 | QDict *options) | |
244 | { | |
245 | if (strstart(filename, prefix, &filename)) { | |
246 | /* Stripping the explicit protocol prefix may result in a protocol | |
247 | * prefix being (wrongly) detected (if the filename contains a colon) */ | |
248 | if (path_has_protocol(filename)) { | |
249 | GString *fat_filename; | |
250 | ||
251 | /* This means there is some colon before the first slash; therefore, | |
252 | * this cannot be an absolute path */ | |
253 | assert(!path_is_absolute(filename)); | |
254 | ||
255 | /* And we can thus fix the protocol detection issue by prefixing it | |
256 | * by "./" */ | |
257 | fat_filename = g_string_new("./"); | |
258 | g_string_append(fat_filename, filename); | |
259 | ||
260 | assert(!path_has_protocol(fat_filename->str)); | |
261 | ||
262 | qdict_put(options, "filename", | |
263 | qstring_from_gstring(fat_filename)); | |
264 | } else { | |
265 | /* If no protocol prefix was detected, we can use the shortened | |
266 | * filename as-is */ | |
267 | qdict_put_str(options, "filename", filename); | |
268 | } | |
269 | } | |
270 | } | |
271 | ||
272 | ||
273 | /* Returns whether the image file is opened as read-only. Note that this can | |
274 | * return false and writing to the image file is still not possible because the | |
275 | * image is inactivated. */ | |
276 | bool bdrv_is_read_only(BlockDriverState *bs) | |
277 | { | |
278 | IO_CODE(); | |
279 | return !(bs->open_flags & BDRV_O_RDWR); | |
280 | } | |
281 | ||
282 | static int GRAPH_RDLOCK | |
283 | bdrv_can_set_read_only(BlockDriverState *bs, bool read_only, | |
284 | bool ignore_allow_rdw, Error **errp) | |
285 | { | |
286 | IO_CODE(); | |
287 | ||
288 | /* Do not set read_only if copy_on_read is enabled */ | |
289 | if (bs->copy_on_read && read_only) { | |
290 | error_setg(errp, "Can't set node '%s' to r/o with copy-on-read enabled", | |
291 | bdrv_get_device_or_node_name(bs)); | |
292 | return -EINVAL; | |
293 | } | |
294 | ||
295 | /* Do not clear read_only if it is prohibited */ | |
296 | if (!read_only && !(bs->open_flags & BDRV_O_ALLOW_RDWR) && | |
297 | !ignore_allow_rdw) | |
298 | { | |
299 | error_setg(errp, "Node '%s' is read only", | |
300 | bdrv_get_device_or_node_name(bs)); | |
301 | return -EPERM; | |
302 | } | |
303 | ||
304 | return 0; | |
305 | } | |
306 | ||
307 | /* | |
308 | * Called by a driver that can only provide a read-only image. | |
309 | * | |
310 | * Returns 0 if the node is already read-only or it could switch the node to | |
311 | * read-only because BDRV_O_AUTO_RDONLY is set. | |
312 | * | |
313 | * Returns -EACCES if the node is read-write and BDRV_O_AUTO_RDONLY is not set | |
314 | * or bdrv_can_set_read_only() forbids making the node read-only. If @errmsg | |
315 | * is not NULL, it is used as the error message for the Error object. | |
316 | */ | |
317 | int bdrv_apply_auto_read_only(BlockDriverState *bs, const char *errmsg, | |
318 | Error **errp) | |
319 | { | |
320 | int ret = 0; | |
321 | IO_CODE(); | |
322 | ||
323 | if (!(bs->open_flags & BDRV_O_RDWR)) { | |
324 | return 0; | |
325 | } | |
326 | if (!(bs->open_flags & BDRV_O_AUTO_RDONLY)) { | |
327 | goto fail; | |
328 | } | |
329 | ||
330 | ret = bdrv_can_set_read_only(bs, true, false, NULL); | |
331 | if (ret < 0) { | |
332 | goto fail; | |
333 | } | |
334 | ||
335 | bs->open_flags &= ~BDRV_O_RDWR; | |
336 | ||
337 | return 0; | |
338 | ||
339 | fail: | |
340 | error_setg(errp, "%s", errmsg ?: "Image is read-only"); | |
341 | return -EACCES; | |
342 | } | |
343 | ||
344 | /* | |
345 | * If @backing is empty, this function returns NULL without setting | |
346 | * @errp. In all other cases, NULL will only be returned with @errp | |
347 | * set. | |
348 | * | |
349 | * Therefore, a return value of NULL without @errp set means that | |
350 | * there is no backing file; if @errp is set, there is one but its | |
351 | * absolute filename cannot be generated. | |
352 | */ | |
353 | char *bdrv_get_full_backing_filename_from_filename(const char *backed, | |
354 | const char *backing, | |
355 | Error **errp) | |
356 | { | |
357 | if (backing[0] == '\0') { | |
358 | return NULL; | |
359 | } else if (path_has_protocol(backing) || path_is_absolute(backing)) { | |
360 | return g_strdup(backing); | |
361 | } else if (backed[0] == '\0' || strstart(backed, "json:", NULL)) { | |
362 | error_setg(errp, "Cannot use relative backing file names for '%s'", | |
363 | backed); | |
364 | return NULL; | |
365 | } else { | |
366 | return path_combine(backed, backing); | |
367 | } | |
368 | } | |
369 | ||
370 | /* | |
371 | * If @filename is empty or NULL, this function returns NULL without | |
372 | * setting @errp. In all other cases, NULL will only be returned with | |
373 | * @errp set. | |
374 | */ | |
375 | static char * GRAPH_RDLOCK | |
376 | bdrv_make_absolute_filename(BlockDriverState *relative_to, | |
377 | const char *filename, Error **errp) | |
378 | { | |
379 | char *dir, *full_name; | |
380 | ||
381 | if (!filename || filename[0] == '\0') { | |
382 | return NULL; | |
383 | } else if (path_has_protocol(filename) || path_is_absolute(filename)) { | |
384 | return g_strdup(filename); | |
385 | } | |
386 | ||
387 | dir = bdrv_dirname(relative_to, errp); | |
388 | if (!dir) { | |
389 | return NULL; | |
390 | } | |
391 | ||
392 | full_name = g_strconcat(dir, filename, NULL); | |
393 | g_free(dir); | |
394 | return full_name; | |
395 | } | |
396 | ||
397 | char *bdrv_get_full_backing_filename(BlockDriverState *bs, Error **errp) | |
398 | { | |
399 | GLOBAL_STATE_CODE(); | |
400 | return bdrv_make_absolute_filename(bs, bs->backing_file, errp); | |
401 | } | |
402 | ||
403 | void bdrv_register(BlockDriver *bdrv) | |
404 | { | |
405 | assert(bdrv->format_name); | |
406 | GLOBAL_STATE_CODE(); | |
407 | QLIST_INSERT_HEAD(&bdrv_drivers, bdrv, list); | |
408 | } | |
409 | ||
410 | BlockDriverState *bdrv_new(void) | |
411 | { | |
412 | BlockDriverState *bs; | |
413 | int i; | |
414 | ||
415 | GLOBAL_STATE_CODE(); | |
416 | ||
417 | bs = g_new0(BlockDriverState, 1); | |
418 | QLIST_INIT(&bs->dirty_bitmaps); | |
419 | for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) { | |
420 | QLIST_INIT(&bs->op_blockers[i]); | |
421 | } | |
422 | qemu_mutex_init(&bs->reqs_lock); | |
423 | qemu_mutex_init(&bs->dirty_bitmap_mutex); | |
424 | bs->refcnt = 1; | |
425 | bs->aio_context = qemu_get_aio_context(); | |
426 | ||
427 | qemu_co_queue_init(&bs->flush_queue); | |
428 | ||
429 | qemu_co_mutex_init(&bs->bsc_modify_lock); | |
430 | bs->block_status_cache = g_new0(BdrvBlockStatusCache, 1); | |
431 | ||
432 | for (i = 0; i < bdrv_drain_all_count; i++) { | |
433 | bdrv_drained_begin(bs); | |
434 | } | |
435 | ||
436 | QTAILQ_INSERT_TAIL(&all_bdrv_states, bs, bs_list); | |
437 | ||
438 | return bs; | |
439 | } | |
440 | ||
441 | static BlockDriver *bdrv_do_find_format(const char *format_name) | |
442 | { | |
443 | BlockDriver *drv1; | |
444 | GLOBAL_STATE_CODE(); | |
445 | ||
446 | QLIST_FOREACH(drv1, &bdrv_drivers, list) { | |
447 | if (!strcmp(drv1->format_name, format_name)) { | |
448 | return drv1; | |
449 | } | |
450 | } | |
451 | ||
452 | return NULL; | |
453 | } | |
454 | ||
455 | BlockDriver *bdrv_find_format(const char *format_name) | |
456 | { | |
457 | BlockDriver *drv1; | |
458 | int i; | |
459 | ||
460 | GLOBAL_STATE_CODE(); | |
461 | ||
462 | drv1 = bdrv_do_find_format(format_name); | |
463 | if (drv1) { | |
464 | return drv1; | |
465 | } | |
466 | ||
467 | /* The driver isn't registered, maybe we need to load a module */ | |
468 | for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); ++i) { | |
469 | if (!strcmp(block_driver_modules[i].format_name, format_name)) { | |
470 | Error *local_err = NULL; | |
471 | int rv = block_module_load(block_driver_modules[i].library_name, | |
472 | &local_err); | |
473 | if (rv > 0) { | |
474 | return bdrv_do_find_format(format_name); | |
475 | } else if (rv < 0) { | |
476 | error_report_err(local_err); | |
477 | } | |
478 | break; | |
479 | } | |
480 | } | |
481 | return NULL; | |
482 | } | |
483 | ||
484 | static int bdrv_format_is_whitelisted(const char *format_name, bool read_only) | |
485 | { | |
486 | static const char *whitelist_rw[] = { | |
487 | CONFIG_BDRV_RW_WHITELIST | |
488 | NULL | |
489 | }; | |
490 | static const char *whitelist_ro[] = { | |
491 | CONFIG_BDRV_RO_WHITELIST | |
492 | NULL | |
493 | }; | |
494 | const char **p; | |
495 | ||
496 | if (!whitelist_rw[0] && !whitelist_ro[0]) { | |
497 | return 1; /* no whitelist, anything goes */ | |
498 | } | |
499 | ||
500 | for (p = whitelist_rw; *p; p++) { | |
501 | if (!strcmp(format_name, *p)) { | |
502 | return 1; | |
503 | } | |
504 | } | |
505 | if (read_only) { | |
506 | for (p = whitelist_ro; *p; p++) { | |
507 | if (!strcmp(format_name, *p)) { | |
508 | return 1; | |
509 | } | |
510 | } | |
511 | } | |
512 | return 0; | |
513 | } | |
514 | ||
515 | int bdrv_is_whitelisted(BlockDriver *drv, bool read_only) | |
516 | { | |
517 | GLOBAL_STATE_CODE(); | |
518 | return bdrv_format_is_whitelisted(drv->format_name, read_only); | |
519 | } | |
520 | ||
521 | bool bdrv_uses_whitelist(void) | |
522 | { | |
523 | return use_bdrv_whitelist; | |
524 | } | |
525 | ||
526 | typedef struct CreateCo { | |
527 | BlockDriver *drv; | |
528 | char *filename; | |
529 | QemuOpts *opts; | |
530 | int ret; | |
531 | Error *err; | |
532 | } CreateCo; | |
533 | ||
534 | int coroutine_fn bdrv_co_create(BlockDriver *drv, const char *filename, | |
535 | QemuOpts *opts, Error **errp) | |
536 | { | |
537 | int ret; | |
538 | GLOBAL_STATE_CODE(); | |
539 | ERRP_GUARD(); | |
540 | ||
541 | if (!drv->bdrv_co_create_opts) { | |
542 | error_setg(errp, "Driver '%s' does not support image creation", | |
543 | drv->format_name); | |
544 | return -ENOTSUP; | |
545 | } | |
546 | ||
547 | ret = drv->bdrv_co_create_opts(drv, filename, opts, errp); | |
548 | if (ret < 0 && !*errp) { | |
549 | error_setg_errno(errp, -ret, "Could not create image"); | |
550 | } | |
551 | ||
552 | return ret; | |
553 | } | |
554 | ||
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 coroutine_fn GRAPH_UNLOCKED | |
563 | create_file_fallback_truncate(BlockBackend *blk, int64_t minimum_size, | |
564 | Error **errp) | |
565 | { | |
566 | Error *local_err = NULL; | |
567 | int64_t size; | |
568 | int ret; | |
569 | ||
570 | GLOBAL_STATE_CODE(); | |
571 | ||
572 | ret = blk_co_truncate(blk, minimum_size, false, PREALLOC_MODE_OFF, 0, | |
573 | &local_err); | |
574 | if (ret < 0 && ret != -ENOTSUP) { | |
575 | error_propagate(errp, local_err); | |
576 | return ret; | |
577 | } | |
578 | ||
579 | size = blk_co_getlength(blk); | |
580 | if (size < 0) { | |
581 | error_free(local_err); | |
582 | error_setg_errno(errp, -size, | |
583 | "Failed to inquire the new image file's length"); | |
584 | return size; | |
585 | } | |
586 | ||
587 | if (size < minimum_size) { | |
588 | /* Need to grow the image, but we failed to do that */ | |
589 | error_propagate(errp, local_err); | |
590 | return -ENOTSUP; | |
591 | } | |
592 | ||
593 | error_free(local_err); | |
594 | local_err = NULL; | |
595 | ||
596 | return size; | |
597 | } | |
598 | ||
599 | /** | |
600 | * Helper function for bdrv_create_file_fallback(): Zero the first | |
601 | * sector to remove any potentially pre-existing image header. | |
602 | */ | |
603 | static int coroutine_fn | |
604 | create_file_fallback_zero_first_sector(BlockBackend *blk, | |
605 | int64_t current_size, | |
606 | Error **errp) | |
607 | { | |
608 | int64_t bytes_to_clear; | |
609 | int ret; | |
610 | ||
611 | GLOBAL_STATE_CODE(); | |
612 | ||
613 | bytes_to_clear = MIN(current_size, BDRV_SECTOR_SIZE); | |
614 | if (bytes_to_clear) { | |
615 | ret = blk_co_pwrite_zeroes(blk, 0, bytes_to_clear, BDRV_REQ_MAY_UNMAP); | |
616 | if (ret < 0) { | |
617 | error_setg_errno(errp, -ret, | |
618 | "Failed to clear the new image's first sector"); | |
619 | return ret; | |
620 | } | |
621 | } | |
622 | ||
623 | return 0; | |
624 | } | |
625 | ||
626 | /** | |
627 | * Simple implementation of bdrv_co_create_opts for protocol drivers | |
628 | * which only support creation via opening a file | |
629 | * (usually existing raw storage device) | |
630 | */ | |
631 | int coroutine_fn bdrv_co_create_opts_simple(BlockDriver *drv, | |
632 | const char *filename, | |
633 | QemuOpts *opts, | |
634 | Error **errp) | |
635 | { | |
636 | BlockBackend *blk; | |
637 | QDict *options; | |
638 | int64_t size = 0; | |
639 | char *buf = NULL; | |
640 | PreallocMode prealloc; | |
641 | Error *local_err = NULL; | |
642 | int ret; | |
643 | ||
644 | GLOBAL_STATE_CODE(); | |
645 | ||
646 | size = qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0); | |
647 | buf = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC); | |
648 | prealloc = qapi_enum_parse(&PreallocMode_lookup, buf, | |
649 | PREALLOC_MODE_OFF, &local_err); | |
650 | g_free(buf); | |
651 | if (local_err) { | |
652 | error_propagate(errp, local_err); | |
653 | return -EINVAL; | |
654 | } | |
655 | ||
656 | if (prealloc != PREALLOC_MODE_OFF) { | |
657 | error_setg(errp, "Unsupported preallocation mode '%s'", | |
658 | PreallocMode_str(prealloc)); | |
659 | return -ENOTSUP; | |
660 | } | |
661 | ||
662 | options = qdict_new(); | |
663 | qdict_put_str(options, "driver", drv->format_name); | |
664 | ||
665 | blk = blk_co_new_open(filename, NULL, options, | |
666 | BDRV_O_RDWR | BDRV_O_RESIZE, errp); | |
667 | if (!blk) { | |
668 | error_prepend(errp, "Protocol driver '%s' does not support creating " | |
669 | "new images, so an existing image must be selected as " | |
670 | "the target; however, opening the given target as an " | |
671 | "existing image failed: ", | |
672 | drv->format_name); | |
673 | return -EINVAL; | |
674 | } | |
675 | ||
676 | size = create_file_fallback_truncate(blk, size, errp); | |
677 | if (size < 0) { | |
678 | ret = size; | |
679 | goto out; | |
680 | } | |
681 | ||
682 | ret = create_file_fallback_zero_first_sector(blk, size, errp); | |
683 | if (ret < 0) { | |
684 | goto out; | |
685 | } | |
686 | ||
687 | ret = 0; | |
688 | out: | |
689 | blk_co_unref(blk); | |
690 | return ret; | |
691 | } | |
692 | ||
693 | int coroutine_fn bdrv_co_create_file(const char *filename, QemuOpts *opts, | |
694 | Error **errp) | |
695 | { | |
696 | QemuOpts *protocol_opts; | |
697 | BlockDriver *drv; | |
698 | QDict *qdict; | |
699 | int ret; | |
700 | ||
701 | GLOBAL_STATE_CODE(); | |
702 | ||
703 | drv = bdrv_find_protocol(filename, true, errp); | |
704 | if (drv == NULL) { | |
705 | return -ENOENT; | |
706 | } | |
707 | ||
708 | if (!drv->create_opts) { | |
709 | error_setg(errp, "Driver '%s' does not support image creation", | |
710 | drv->format_name); | |
711 | return -ENOTSUP; | |
712 | } | |
713 | ||
714 | /* | |
715 | * 'opts' contains a QemuOptsList with a combination of format and protocol | |
716 | * default values. | |
717 | * | |
718 | * The format properly removes its options, but the default values remain | |
719 | * in 'opts->list'. So if the protocol has options with the same name | |
720 | * (e.g. rbd has 'cluster_size' as qcow2), it will see the default values | |
721 | * of the format, since for overlapping options, the format wins. | |
722 | * | |
723 | * To avoid this issue, lets convert QemuOpts to QDict, in this way we take | |
724 | * only the set options, and then convert it back to QemuOpts, using the | |
725 | * create_opts of the protocol. So the new QemuOpts, will contain only the | |
726 | * protocol defaults. | |
727 | */ | |
728 | qdict = qemu_opts_to_qdict(opts, NULL); | |
729 | protocol_opts = qemu_opts_from_qdict(drv->create_opts, qdict, errp); | |
730 | if (protocol_opts == NULL) { | |
731 | ret = -EINVAL; | |
732 | goto out; | |
733 | } | |
734 | ||
735 | ret = bdrv_co_create(drv, filename, protocol_opts, errp); | |
736 | out: | |
737 | qemu_opts_del(protocol_opts); | |
738 | qobject_unref(qdict); | |
739 | return ret; | |
740 | } | |
741 | ||
742 | int coroutine_fn bdrv_co_delete_file(BlockDriverState *bs, Error **errp) | |
743 | { | |
744 | Error *local_err = NULL; | |
745 | int ret; | |
746 | ||
747 | IO_CODE(); | |
748 | assert(bs != NULL); | |
749 | assert_bdrv_graph_readable(); | |
750 | ||
751 | if (!bs->drv) { | |
752 | error_setg(errp, "Block node '%s' is not opened", bs->filename); | |
753 | return -ENOMEDIUM; | |
754 | } | |
755 | ||
756 | if (!bs->drv->bdrv_co_delete_file) { | |
757 | error_setg(errp, "Driver '%s' does not support image deletion", | |
758 | bs->drv->format_name); | |
759 | return -ENOTSUP; | |
760 | } | |
761 | ||
762 | ret = bs->drv->bdrv_co_delete_file(bs, &local_err); | |
763 | if (ret < 0) { | |
764 | error_propagate(errp, local_err); | |
765 | } | |
766 | ||
767 | return ret; | |
768 | } | |
769 | ||
770 | void coroutine_fn bdrv_co_delete_file_noerr(BlockDriverState *bs) | |
771 | { | |
772 | Error *local_err = NULL; | |
773 | int ret; | |
774 | IO_CODE(); | |
775 | ||
776 | if (!bs) { | |
777 | return; | |
778 | } | |
779 | ||
780 | ret = bdrv_co_delete_file(bs, &local_err); | |
781 | /* | |
782 | * ENOTSUP will happen if the block driver doesn't support | |
783 | * the 'bdrv_co_delete_file' interface. This is a predictable | |
784 | * scenario and shouldn't be reported back to the user. | |
785 | */ | |
786 | if (ret == -ENOTSUP) { | |
787 | error_free(local_err); | |
788 | } else if (ret < 0) { | |
789 | error_report_err(local_err); | |
790 | } | |
791 | } | |
792 | ||
793 | /** | |
794 | * Try to get @bs's logical and physical block size. | |
795 | * On success, store them in @bsz struct and return 0. | |
796 | * On failure return -errno. | |
797 | * @bs must not be empty. | |
798 | */ | |
799 | int bdrv_probe_blocksizes(BlockDriverState *bs, BlockSizes *bsz) | |
800 | { | |
801 | BlockDriver *drv = bs->drv; | |
802 | BlockDriverState *filtered = bdrv_filter_bs(bs); | |
803 | GLOBAL_STATE_CODE(); | |
804 | ||
805 | if (drv && drv->bdrv_probe_blocksizes) { | |
806 | return drv->bdrv_probe_blocksizes(bs, bsz); | |
807 | } else if (filtered) { | |
808 | return bdrv_probe_blocksizes(filtered, bsz); | |
809 | } | |
810 | ||
811 | return -ENOTSUP; | |
812 | } | |
813 | ||
814 | /** | |
815 | * Try to get @bs's geometry (cyls, heads, sectors). | |
816 | * On success, store them in @geo struct and return 0. | |
817 | * On failure return -errno. | |
818 | * @bs must not be empty. | |
819 | */ | |
820 | int bdrv_probe_geometry(BlockDriverState *bs, HDGeometry *geo) | |
821 | { | |
822 | BlockDriver *drv = bs->drv; | |
823 | BlockDriverState *filtered; | |
824 | ||
825 | GLOBAL_STATE_CODE(); | |
826 | GRAPH_RDLOCK_GUARD_MAINLOOP(); | |
827 | ||
828 | if (drv && drv->bdrv_probe_geometry) { | |
829 | return drv->bdrv_probe_geometry(bs, geo); | |
830 | } | |
831 | ||
832 | filtered = bdrv_filter_bs(bs); | |
833 | if (filtered) { | |
834 | return bdrv_probe_geometry(filtered, geo); | |
835 | } | |
836 | ||
837 | return -ENOTSUP; | |
838 | } | |
839 | ||
840 | /* | |
841 | * Create a uniquely-named empty temporary file. | |
842 | * Return the actual file name used upon success, otherwise NULL. | |
843 | * This string should be freed with g_free() when not needed any longer. | |
844 | * | |
845 | * Note: creating a temporary file for the caller to (re)open is | |
846 | * inherently racy. Use g_file_open_tmp() instead whenever practical. | |
847 | */ | |
848 | char *create_tmp_file(Error **errp) | |
849 | { | |
850 | int fd; | |
851 | const char *tmpdir; | |
852 | g_autofree char *filename = NULL; | |
853 | ||
854 | tmpdir = g_get_tmp_dir(); | |
855 | #ifndef _WIN32 | |
856 | /* | |
857 | * See commit 69bef79 ("block: use /var/tmp instead of /tmp for -snapshot") | |
858 | * | |
859 | * This function is used to create temporary disk images (like -snapshot), | |
860 | * so the files can become very large. /tmp is often a tmpfs where as | |
861 | * /var/tmp is usually on a disk, so more appropriate for disk images. | |
862 | */ | |
863 | if (!g_strcmp0(tmpdir, "/tmp")) { | |
864 | tmpdir = "/var/tmp"; | |
865 | } | |
866 | #endif | |
867 | ||
868 | filename = g_strdup_printf("%s/vl.XXXXXX", tmpdir); | |
869 | fd = g_mkstemp(filename); | |
870 | if (fd < 0) { | |
871 | error_setg_errno(errp, errno, "Could not open temporary file '%s'", | |
872 | filename); | |
873 | return NULL; | |
874 | } | |
875 | close(fd); | |
876 | ||
877 | return g_steal_pointer(&filename); | |
878 | } | |
879 | ||
880 | /* | |
881 | * Detect host devices. By convention, /dev/cdrom[N] is always | |
882 | * recognized as a host CDROM. | |
883 | */ | |
884 | static BlockDriver *find_hdev_driver(const char *filename) | |
885 | { | |
886 | int score_max = 0, score; | |
887 | BlockDriver *drv = NULL, *d; | |
888 | GLOBAL_STATE_CODE(); | |
889 | ||
890 | QLIST_FOREACH(d, &bdrv_drivers, list) { | |
891 | if (d->bdrv_probe_device) { | |
892 | score = d->bdrv_probe_device(filename); | |
893 | if (score > score_max) { | |
894 | score_max = score; | |
895 | drv = d; | |
896 | } | |
897 | } | |
898 | } | |
899 | ||
900 | return drv; | |
901 | } | |
902 | ||
903 | static BlockDriver *bdrv_do_find_protocol(const char *protocol) | |
904 | { | |
905 | BlockDriver *drv1; | |
906 | GLOBAL_STATE_CODE(); | |
907 | ||
908 | QLIST_FOREACH(drv1, &bdrv_drivers, list) { | |
909 | if (drv1->protocol_name && !strcmp(drv1->protocol_name, protocol)) { | |
910 | return drv1; | |
911 | } | |
912 | } | |
913 | ||
914 | return NULL; | |
915 | } | |
916 | ||
917 | BlockDriver *bdrv_find_protocol(const char *filename, | |
918 | bool allow_protocol_prefix, | |
919 | Error **errp) | |
920 | { | |
921 | BlockDriver *drv1; | |
922 | char protocol[128]; | |
923 | int len; | |
924 | const char *p; | |
925 | int i; | |
926 | ||
927 | GLOBAL_STATE_CODE(); | |
928 | /* TODO Drivers without bdrv_file_open must be specified explicitly */ | |
929 | ||
930 | /* | |
931 | * XXX(hch): we really should not let host device detection | |
932 | * override an explicit protocol specification, but moving this | |
933 | * later breaks access to device names with colons in them. | |
934 | * Thanks to the brain-dead persistent naming schemes on udev- | |
935 | * based Linux systems those actually are quite common. | |
936 | */ | |
937 | drv1 = find_hdev_driver(filename); | |
938 | if (drv1) { | |
939 | return drv1; | |
940 | } | |
941 | ||
942 | if (!path_has_protocol(filename) || !allow_protocol_prefix) { | |
943 | return &bdrv_file; | |
944 | } | |
945 | ||
946 | p = strchr(filename, ':'); | |
947 | assert(p != NULL); | |
948 | len = p - filename; | |
949 | if (len > sizeof(protocol) - 1) | |
950 | len = sizeof(protocol) - 1; | |
951 | memcpy(protocol, filename, len); | |
952 | protocol[len] = '\0'; | |
953 | ||
954 | drv1 = bdrv_do_find_protocol(protocol); | |
955 | if (drv1) { | |
956 | return drv1; | |
957 | } | |
958 | ||
959 | for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); ++i) { | |
960 | if (block_driver_modules[i].protocol_name && | |
961 | !strcmp(block_driver_modules[i].protocol_name, protocol)) { | |
962 | int rv = block_module_load(block_driver_modules[i].library_name, errp); | |
963 | if (rv > 0) { | |
964 | drv1 = bdrv_do_find_protocol(protocol); | |
965 | } else if (rv < 0) { | |
966 | return NULL; | |
967 | } | |
968 | break; | |
969 | } | |
970 | } | |
971 | ||
972 | if (!drv1) { | |
973 | error_setg(errp, "Unknown protocol '%s'", protocol); | |
974 | } | |
975 | return drv1; | |
976 | } | |
977 | ||
978 | /* | |
979 | * Guess image format by probing its contents. | |
980 | * This is not a good idea when your image is raw (CVE-2008-2004), but | |
981 | * we do it anyway for backward compatibility. | |
982 | * | |
983 | * @buf contains the image's first @buf_size bytes. | |
984 | * @buf_size is the buffer size in bytes (generally BLOCK_PROBE_BUF_SIZE, | |
985 | * but can be smaller if the image file is smaller) | |
986 | * @filename is its filename. | |
987 | * | |
988 | * For all block drivers, call the bdrv_probe() method to get its | |
989 | * probing score. | |
990 | * Return the first block driver with the highest probing score. | |
991 | */ | |
992 | BlockDriver *bdrv_probe_all(const uint8_t *buf, int buf_size, | |
993 | const char *filename) | |
994 | { | |
995 | int score_max = 0, score; | |
996 | BlockDriver *drv = NULL, *d; | |
997 | IO_CODE(); | |
998 | ||
999 | QLIST_FOREACH(d, &bdrv_drivers, list) { | |
1000 | if (d->bdrv_probe) { | |
1001 | score = d->bdrv_probe(buf, buf_size, filename); | |
1002 | if (score > score_max) { | |
1003 | score_max = score; | |
1004 | drv = d; | |
1005 | } | |
1006 | } | |
1007 | } | |
1008 | ||
1009 | return drv; | |
1010 | } | |
1011 | ||
1012 | static int find_image_format(BlockBackend *file, const char *filename, | |
1013 | BlockDriver **pdrv, Error **errp) | |
1014 | { | |
1015 | BlockDriver *drv; | |
1016 | uint8_t buf[BLOCK_PROBE_BUF_SIZE]; | |
1017 | int ret = 0; | |
1018 | ||
1019 | GLOBAL_STATE_CODE(); | |
1020 | ||
1021 | /* Return the raw BlockDriver * to scsi-generic devices or empty drives */ | |
1022 | if (blk_is_sg(file) || !blk_is_inserted(file) || blk_getlength(file) == 0) { | |
1023 | *pdrv = &bdrv_raw; | |
1024 | return ret; | |
1025 | } | |
1026 | ||
1027 | ret = blk_pread(file, 0, sizeof(buf), buf, 0); | |
1028 | if (ret < 0) { | |
1029 | error_setg_errno(errp, -ret, "Could not read image for determining its " | |
1030 | "format"); | |
1031 | *pdrv = NULL; | |
1032 | return ret; | |
1033 | } | |
1034 | ||
1035 | drv = bdrv_probe_all(buf, sizeof(buf), filename); | |
1036 | if (!drv) { | |
1037 | error_setg(errp, "Could not determine image format: No compatible " | |
1038 | "driver found"); | |
1039 | *pdrv = NULL; | |
1040 | return -ENOENT; | |
1041 | } | |
1042 | ||
1043 | *pdrv = drv; | |
1044 | return 0; | |
1045 | } | |
1046 | ||
1047 | /** | |
1048 | * Set the current 'total_sectors' value | |
1049 | * Return 0 on success, -errno on error. | |
1050 | */ | |
1051 | int coroutine_fn bdrv_co_refresh_total_sectors(BlockDriverState *bs, | |
1052 | int64_t hint) | |
1053 | { | |
1054 | BlockDriver *drv = bs->drv; | |
1055 | IO_CODE(); | |
1056 | assert_bdrv_graph_readable(); | |
1057 | ||
1058 | if (!drv) { | |
1059 | return -ENOMEDIUM; | |
1060 | } | |
1061 | ||
1062 | /* Do not attempt drv->bdrv_co_getlength() on scsi-generic devices */ | |
1063 | if (bdrv_is_sg(bs)) | |
1064 | return 0; | |
1065 | ||
1066 | /* query actual device if possible, otherwise just trust the hint */ | |
1067 | if (drv->bdrv_co_getlength) { | |
1068 | int64_t length = drv->bdrv_co_getlength(bs); | |
1069 | if (length < 0) { | |
1070 | return length; | |
1071 | } | |
1072 | hint = DIV_ROUND_UP(length, BDRV_SECTOR_SIZE); | |
1073 | } | |
1074 | ||
1075 | bs->total_sectors = hint; | |
1076 | ||
1077 | if (bs->total_sectors * BDRV_SECTOR_SIZE > BDRV_MAX_LENGTH) { | |
1078 | return -EFBIG; | |
1079 | } | |
1080 | ||
1081 | return 0; | |
1082 | } | |
1083 | ||
1084 | /** | |
1085 | * Combines a QDict of new block driver @options with any missing options taken | |
1086 | * from @old_options, so that leaving out an option defaults to its old value. | |
1087 | */ | |
1088 | static void bdrv_join_options(BlockDriverState *bs, QDict *options, | |
1089 | QDict *old_options) | |
1090 | { | |
1091 | GLOBAL_STATE_CODE(); | |
1092 | if (bs->drv && bs->drv->bdrv_join_options) { | |
1093 | bs->drv->bdrv_join_options(options, old_options); | |
1094 | } else { | |
1095 | qdict_join(options, old_options, false); | |
1096 | } | |
1097 | } | |
1098 | ||
1099 | static BlockdevDetectZeroesOptions bdrv_parse_detect_zeroes(QemuOpts *opts, | |
1100 | int open_flags, | |
1101 | Error **errp) | |
1102 | { | |
1103 | Error *local_err = NULL; | |
1104 | char *value = qemu_opt_get_del(opts, "detect-zeroes"); | |
1105 | BlockdevDetectZeroesOptions detect_zeroes = | |
1106 | qapi_enum_parse(&BlockdevDetectZeroesOptions_lookup, value, | |
1107 | BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF, &local_err); | |
1108 | GLOBAL_STATE_CODE(); | |
1109 | g_free(value); | |
1110 | if (local_err) { | |
1111 | error_propagate(errp, local_err); | |
1112 | return detect_zeroes; | |
1113 | } | |
1114 | ||
1115 | if (detect_zeroes == BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP && | |
1116 | !(open_flags & BDRV_O_UNMAP)) | |
1117 | { | |
1118 | error_setg(errp, "setting detect-zeroes to unmap is not allowed " | |
1119 | "without setting discard operation to unmap"); | |
1120 | } | |
1121 | ||
1122 | return detect_zeroes; | |
1123 | } | |
1124 | ||
1125 | /** | |
1126 | * Set open flags for aio engine | |
1127 | * | |
1128 | * Return 0 on success, -1 if the engine specified is invalid | |
1129 | */ | |
1130 | int bdrv_parse_aio(const char *mode, int *flags) | |
1131 | { | |
1132 | if (!strcmp(mode, "threads")) { | |
1133 | /* do nothing, default */ | |
1134 | } else if (!strcmp(mode, "native")) { | |
1135 | *flags |= BDRV_O_NATIVE_AIO; | |
1136 | #ifdef CONFIG_LINUX_IO_URING | |
1137 | } else if (!strcmp(mode, "io_uring")) { | |
1138 | *flags |= BDRV_O_IO_URING; | |
1139 | #endif | |
1140 | } else { | |
1141 | return -1; | |
1142 | } | |
1143 | ||
1144 | return 0; | |
1145 | } | |
1146 | ||
1147 | /** | |
1148 | * Set open flags for a given discard mode | |
1149 | * | |
1150 | * Return 0 on success, -1 if the discard mode was invalid. | |
1151 | */ | |
1152 | int bdrv_parse_discard_flags(const char *mode, int *flags) | |
1153 | { | |
1154 | *flags &= ~BDRV_O_UNMAP; | |
1155 | ||
1156 | if (!strcmp(mode, "off") || !strcmp(mode, "ignore")) { | |
1157 | /* do nothing */ | |
1158 | } else if (!strcmp(mode, "on") || !strcmp(mode, "unmap")) { | |
1159 | *flags |= BDRV_O_UNMAP; | |
1160 | } else { | |
1161 | return -1; | |
1162 | } | |
1163 | ||
1164 | return 0; | |
1165 | } | |
1166 | ||
1167 | /** | |
1168 | * Set open flags for a given cache mode | |
1169 | * | |
1170 | * Return 0 on success, -1 if the cache mode was invalid. | |
1171 | */ | |
1172 | int bdrv_parse_cache_mode(const char *mode, int *flags, bool *writethrough) | |
1173 | { | |
1174 | *flags &= ~BDRV_O_CACHE_MASK; | |
1175 | ||
1176 | if (!strcmp(mode, "off") || !strcmp(mode, "none")) { | |
1177 | *writethrough = false; | |
1178 | *flags |= BDRV_O_NOCACHE; | |
1179 | } else if (!strcmp(mode, "directsync")) { | |
1180 | *writethrough = true; | |
1181 | *flags |= BDRV_O_NOCACHE; | |
1182 | } else if (!strcmp(mode, "writeback")) { | |
1183 | *writethrough = false; | |
1184 | } else if (!strcmp(mode, "unsafe")) { | |
1185 | *writethrough = false; | |
1186 | *flags |= BDRV_O_NO_FLUSH; | |
1187 | } else if (!strcmp(mode, "writethrough")) { | |
1188 | *writethrough = true; | |
1189 | } else { | |
1190 | return -1; | |
1191 | } | |
1192 | ||
1193 | return 0; | |
1194 | } | |
1195 | ||
1196 | static char *bdrv_child_get_parent_desc(BdrvChild *c) | |
1197 | { | |
1198 | BlockDriverState *parent = c->opaque; | |
1199 | return g_strdup_printf("node '%s'", bdrv_get_node_name(parent)); | |
1200 | } | |
1201 | ||
1202 | static void GRAPH_RDLOCK bdrv_child_cb_drained_begin(BdrvChild *child) | |
1203 | { | |
1204 | BlockDriverState *bs = child->opaque; | |
1205 | bdrv_do_drained_begin_quiesce(bs, NULL); | |
1206 | } | |
1207 | ||
1208 | static bool GRAPH_RDLOCK bdrv_child_cb_drained_poll(BdrvChild *child) | |
1209 | { | |
1210 | BlockDriverState *bs = child->opaque; | |
1211 | return bdrv_drain_poll(bs, NULL, false); | |
1212 | } | |
1213 | ||
1214 | static void GRAPH_RDLOCK bdrv_child_cb_drained_end(BdrvChild *child) | |
1215 | { | |
1216 | BlockDriverState *bs = child->opaque; | |
1217 | bdrv_drained_end(bs); | |
1218 | } | |
1219 | ||
1220 | static int bdrv_child_cb_inactivate(BdrvChild *child) | |
1221 | { | |
1222 | BlockDriverState *bs = child->opaque; | |
1223 | GLOBAL_STATE_CODE(); | |
1224 | assert(bs->open_flags & BDRV_O_INACTIVE); | |
1225 | return 0; | |
1226 | } | |
1227 | ||
1228 | static bool bdrv_child_cb_change_aio_ctx(BdrvChild *child, AioContext *ctx, | |
1229 | GHashTable *visited, Transaction *tran, | |
1230 | Error **errp) | |
1231 | { | |
1232 | BlockDriverState *bs = child->opaque; | |
1233 | return bdrv_change_aio_context(bs, ctx, visited, tran, errp); | |
1234 | } | |
1235 | ||
1236 | /* | |
1237 | * Returns the options and flags that a temporary snapshot should get, based on | |
1238 | * the originally requested flags (the originally requested image will have | |
1239 | * flags like a backing file) | |
1240 | */ | |
1241 | static void bdrv_temp_snapshot_options(int *child_flags, QDict *child_options, | |
1242 | int parent_flags, QDict *parent_options) | |
1243 | { | |
1244 | GLOBAL_STATE_CODE(); | |
1245 | *child_flags = (parent_flags & ~BDRV_O_SNAPSHOT) | BDRV_O_TEMPORARY; | |
1246 | ||
1247 | /* For temporary files, unconditional cache=unsafe is fine */ | |
1248 | qdict_set_default_str(child_options, BDRV_OPT_CACHE_DIRECT, "off"); | |
1249 | qdict_set_default_str(child_options, BDRV_OPT_CACHE_NO_FLUSH, "on"); | |
1250 | ||
1251 | /* Copy the read-only and discard options from the parent */ | |
1252 | qdict_copy_default(child_options, parent_options, BDRV_OPT_READ_ONLY); | |
1253 | qdict_copy_default(child_options, parent_options, BDRV_OPT_DISCARD); | |
1254 | ||
1255 | /* aio=native doesn't work for cache.direct=off, so disable it for the | |
1256 | * temporary snapshot */ | |
1257 | *child_flags &= ~BDRV_O_NATIVE_AIO; | |
1258 | } | |
1259 | ||
1260 | static void GRAPH_WRLOCK bdrv_backing_attach(BdrvChild *c) | |
1261 | { | |
1262 | BlockDriverState *parent = c->opaque; | |
1263 | BlockDriverState *backing_hd = c->bs; | |
1264 | ||
1265 | GLOBAL_STATE_CODE(); | |
1266 | assert(!parent->backing_blocker); | |
1267 | error_setg(&parent->backing_blocker, | |
1268 | "node is used as backing hd of '%s'", | |
1269 | bdrv_get_device_or_node_name(parent)); | |
1270 | ||
1271 | bdrv_refresh_filename(backing_hd); | |
1272 | ||
1273 | parent->open_flags &= ~BDRV_O_NO_BACKING; | |
1274 | ||
1275 | bdrv_op_block_all(backing_hd, parent->backing_blocker); | |
1276 | /* Otherwise we won't be able to commit or stream */ | |
1277 | bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_COMMIT_TARGET, | |
1278 | parent->backing_blocker); | |
1279 | bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_STREAM, | |
1280 | parent->backing_blocker); | |
1281 | /* | |
1282 | * We do backup in 3 ways: | |
1283 | * 1. drive backup | |
1284 | * The target bs is new opened, and the source is top BDS | |
1285 | * 2. blockdev backup | |
1286 | * Both the source and the target are top BDSes. | |
1287 | * 3. internal backup(used for block replication) | |
1288 | * Both the source and the target are backing file | |
1289 | * | |
1290 | * In case 1 and 2, neither the source nor the target is the backing file. | |
1291 | * In case 3, we will block the top BDS, so there is only one block job | |
1292 | * for the top BDS and its backing chain. | |
1293 | */ | |
1294 | bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_BACKUP_SOURCE, | |
1295 | parent->backing_blocker); | |
1296 | bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_BACKUP_TARGET, | |
1297 | parent->backing_blocker); | |
1298 | } | |
1299 | ||
1300 | static void bdrv_backing_detach(BdrvChild *c) | |
1301 | { | |
1302 | BlockDriverState *parent = c->opaque; | |
1303 | ||
1304 | GLOBAL_STATE_CODE(); | |
1305 | assert(parent->backing_blocker); | |
1306 | bdrv_op_unblock_all(c->bs, parent->backing_blocker); | |
1307 | error_free(parent->backing_blocker); | |
1308 | parent->backing_blocker = NULL; | |
1309 | } | |
1310 | ||
1311 | static int bdrv_backing_update_filename(BdrvChild *c, BlockDriverState *base, | |
1312 | const char *filename, Error **errp) | |
1313 | { | |
1314 | BlockDriverState *parent = c->opaque; | |
1315 | bool read_only = bdrv_is_read_only(parent); | |
1316 | int ret; | |
1317 | GLOBAL_STATE_CODE(); | |
1318 | ||
1319 | if (read_only) { | |
1320 | ret = bdrv_reopen_set_read_only(parent, false, errp); | |
1321 | if (ret < 0) { | |
1322 | return ret; | |
1323 | } | |
1324 | } | |
1325 | ||
1326 | ret = bdrv_change_backing_file(parent, filename, | |
1327 | base->drv ? base->drv->format_name : "", | |
1328 | false); | |
1329 | if (ret < 0) { | |
1330 | error_setg_errno(errp, -ret, "Could not update backing file link"); | |
1331 | } | |
1332 | ||
1333 | if (read_only) { | |
1334 | bdrv_reopen_set_read_only(parent, true, NULL); | |
1335 | } | |
1336 | ||
1337 | return ret; | |
1338 | } | |
1339 | ||
1340 | /* | |
1341 | * Returns the options and flags that a generic child of a BDS should | |
1342 | * get, based on the given options and flags for the parent BDS. | |
1343 | */ | |
1344 | static void bdrv_inherited_options(BdrvChildRole role, bool parent_is_format, | |
1345 | int *child_flags, QDict *child_options, | |
1346 | int parent_flags, QDict *parent_options) | |
1347 | { | |
1348 | int flags = parent_flags; | |
1349 | GLOBAL_STATE_CODE(); | |
1350 | ||
1351 | /* | |
1352 | * First, decide whether to set, clear, or leave BDRV_O_PROTOCOL. | |
1353 | * Generally, the question to answer is: Should this child be | |
1354 | * format-probed by default? | |
1355 | */ | |
1356 | ||
1357 | /* | |
1358 | * Pure and non-filtered data children of non-format nodes should | |
1359 | * be probed by default (even when the node itself has BDRV_O_PROTOCOL | |
1360 | * set). This only affects a very limited set of drivers (namely | |
1361 | * quorum and blkverify when this comment was written). | |
1362 | * Force-clear BDRV_O_PROTOCOL then. | |
1363 | */ | |
1364 | if (!parent_is_format && | |
1365 | (role & BDRV_CHILD_DATA) && | |
1366 | !(role & (BDRV_CHILD_METADATA | BDRV_CHILD_FILTERED))) | |
1367 | { | |
1368 | flags &= ~BDRV_O_PROTOCOL; | |
1369 | } | |
1370 | ||
1371 | /* | |
1372 | * All children of format nodes (except for COW children) and all | |
1373 | * metadata children in general should never be format-probed. | |
1374 | * Force-set BDRV_O_PROTOCOL then. | |
1375 | */ | |
1376 | if ((parent_is_format && !(role & BDRV_CHILD_COW)) || | |
1377 | (role & BDRV_CHILD_METADATA)) | |
1378 | { | |
1379 | flags |= BDRV_O_PROTOCOL; | |
1380 | } | |
1381 | ||
1382 | /* | |
1383 | * If the cache mode isn't explicitly set, inherit direct and no-flush from | |
1384 | * the parent. | |
1385 | */ | |
1386 | qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_DIRECT); | |
1387 | qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_NO_FLUSH); | |
1388 | qdict_copy_default(child_options, parent_options, BDRV_OPT_FORCE_SHARE); | |
1389 | ||
1390 | if (role & BDRV_CHILD_COW) { | |
1391 | /* backing files are opened read-only by default */ | |
1392 | qdict_set_default_str(child_options, BDRV_OPT_READ_ONLY, "on"); | |
1393 | qdict_set_default_str(child_options, BDRV_OPT_AUTO_READ_ONLY, "off"); | |
1394 | } else { | |
1395 | /* Inherit the read-only option from the parent if it's not set */ | |
1396 | qdict_copy_default(child_options, parent_options, BDRV_OPT_READ_ONLY); | |
1397 | qdict_copy_default(child_options, parent_options, | |
1398 | BDRV_OPT_AUTO_READ_ONLY); | |
1399 | } | |
1400 | ||
1401 | /* | |
1402 | * bdrv_co_pdiscard() respects unmap policy for the parent, so we | |
1403 | * can default to enable it on lower layers regardless of the | |
1404 | * parent option. | |
1405 | */ | |
1406 | qdict_set_default_str(child_options, BDRV_OPT_DISCARD, "unmap"); | |
1407 | ||
1408 | /* Clear flags that only apply to the top layer */ | |
1409 | flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_COPY_ON_READ); | |
1410 | ||
1411 | if (role & BDRV_CHILD_METADATA) { | |
1412 | flags &= ~BDRV_O_NO_IO; | |
1413 | } | |
1414 | if (role & BDRV_CHILD_COW) { | |
1415 | flags &= ~BDRV_O_TEMPORARY; | |
1416 | } | |
1417 | ||
1418 | *child_flags = flags; | |
1419 | } | |
1420 | ||
1421 | static void GRAPH_WRLOCK bdrv_child_cb_attach(BdrvChild *child) | |
1422 | { | |
1423 | BlockDriverState *bs = child->opaque; | |
1424 | ||
1425 | assert_bdrv_graph_writable(); | |
1426 | QLIST_INSERT_HEAD(&bs->children, child, next); | |
1427 | if (bs->drv->is_filter || (child->role & BDRV_CHILD_FILTERED)) { | |
1428 | /* | |
1429 | * Here we handle filters and block/raw-format.c when it behave like | |
1430 | * filter. They generally have a single PRIMARY child, which is also the | |
1431 | * FILTERED child, and that they may have multiple more children, which | |
1432 | * are neither PRIMARY nor FILTERED. And never we have a COW child here. | |
1433 | * So bs->file will be the PRIMARY child, unless the PRIMARY child goes | |
1434 | * into bs->backing on exceptional cases; and bs->backing will be | |
1435 | * nothing else. | |
1436 | */ | |
1437 | assert(!(child->role & BDRV_CHILD_COW)); | |
1438 | if (child->role & BDRV_CHILD_PRIMARY) { | |
1439 | assert(child->role & BDRV_CHILD_FILTERED); | |
1440 | assert(!bs->backing); | |
1441 | assert(!bs->file); | |
1442 | ||
1443 | if (bs->drv->filtered_child_is_backing) { | |
1444 | bs->backing = child; | |
1445 | } else { | |
1446 | bs->file = child; | |
1447 | } | |
1448 | } else { | |
1449 | assert(!(child->role & BDRV_CHILD_FILTERED)); | |
1450 | } | |
1451 | } else if (child->role & BDRV_CHILD_COW) { | |
1452 | assert(bs->drv->supports_backing); | |
1453 | assert(!(child->role & BDRV_CHILD_PRIMARY)); | |
1454 | assert(!bs->backing); | |
1455 | bs->backing = child; | |
1456 | bdrv_backing_attach(child); | |
1457 | } else if (child->role & BDRV_CHILD_PRIMARY) { | |
1458 | assert(!bs->file); | |
1459 | bs->file = child; | |
1460 | } | |
1461 | } | |
1462 | ||
1463 | static void GRAPH_WRLOCK bdrv_child_cb_detach(BdrvChild *child) | |
1464 | { | |
1465 | BlockDriverState *bs = child->opaque; | |
1466 | ||
1467 | if (child->role & BDRV_CHILD_COW) { | |
1468 | bdrv_backing_detach(child); | |
1469 | } | |
1470 | ||
1471 | assert_bdrv_graph_writable(); | |
1472 | QLIST_REMOVE(child, next); | |
1473 | if (child == bs->backing) { | |
1474 | assert(child != bs->file); | |
1475 | bs->backing = NULL; | |
1476 | } else if (child == bs->file) { | |
1477 | bs->file = NULL; | |
1478 | } | |
1479 | } | |
1480 | ||
1481 | static int bdrv_child_cb_update_filename(BdrvChild *c, BlockDriverState *base, | |
1482 | const char *filename, Error **errp) | |
1483 | { | |
1484 | if (c->role & BDRV_CHILD_COW) { | |
1485 | return bdrv_backing_update_filename(c, base, filename, errp); | |
1486 | } | |
1487 | return 0; | |
1488 | } | |
1489 | ||
1490 | AioContext *child_of_bds_get_parent_aio_context(BdrvChild *c) | |
1491 | { | |
1492 | BlockDriverState *bs = c->opaque; | |
1493 | IO_CODE(); | |
1494 | ||
1495 | return bdrv_get_aio_context(bs); | |
1496 | } | |
1497 | ||
1498 | const BdrvChildClass child_of_bds = { | |
1499 | .parent_is_bds = true, | |
1500 | .get_parent_desc = bdrv_child_get_parent_desc, | |
1501 | .inherit_options = bdrv_inherited_options, | |
1502 | .drained_begin = bdrv_child_cb_drained_begin, | |
1503 | .drained_poll = bdrv_child_cb_drained_poll, | |
1504 | .drained_end = bdrv_child_cb_drained_end, | |
1505 | .attach = bdrv_child_cb_attach, | |
1506 | .detach = bdrv_child_cb_detach, | |
1507 | .inactivate = bdrv_child_cb_inactivate, | |
1508 | .change_aio_ctx = bdrv_child_cb_change_aio_ctx, | |
1509 | .update_filename = bdrv_child_cb_update_filename, | |
1510 | .get_parent_aio_context = child_of_bds_get_parent_aio_context, | |
1511 | }; | |
1512 | ||
1513 | AioContext *bdrv_child_get_parent_aio_context(BdrvChild *c) | |
1514 | { | |
1515 | IO_CODE(); | |
1516 | return c->klass->get_parent_aio_context(c); | |
1517 | } | |
1518 | ||
1519 | static int bdrv_open_flags(BlockDriverState *bs, int flags) | |
1520 | { | |
1521 | int open_flags = flags; | |
1522 | GLOBAL_STATE_CODE(); | |
1523 | ||
1524 | /* | |
1525 | * Clear flags that are internal to the block layer before opening the | |
1526 | * image. | |
1527 | */ | |
1528 | open_flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_PROTOCOL); | |
1529 | ||
1530 | return open_flags; | |
1531 | } | |
1532 | ||
1533 | static void update_flags_from_options(int *flags, QemuOpts *opts) | |
1534 | { | |
1535 | GLOBAL_STATE_CODE(); | |
1536 | ||
1537 | *flags &= ~(BDRV_O_CACHE_MASK | BDRV_O_RDWR | BDRV_O_AUTO_RDONLY); | |
1538 | ||
1539 | if (qemu_opt_get_bool_del(opts, BDRV_OPT_CACHE_NO_FLUSH, false)) { | |
1540 | *flags |= BDRV_O_NO_FLUSH; | |
1541 | } | |
1542 | ||
1543 | if (qemu_opt_get_bool_del(opts, BDRV_OPT_CACHE_DIRECT, false)) { | |
1544 | *flags |= BDRV_O_NOCACHE; | |
1545 | } | |
1546 | ||
1547 | if (!qemu_opt_get_bool_del(opts, BDRV_OPT_READ_ONLY, false)) { | |
1548 | *flags |= BDRV_O_RDWR; | |
1549 | } | |
1550 | ||
1551 | if (qemu_opt_get_bool_del(opts, BDRV_OPT_AUTO_READ_ONLY, false)) { | |
1552 | *flags |= BDRV_O_AUTO_RDONLY; | |
1553 | } | |
1554 | } | |
1555 | ||
1556 | static void update_options_from_flags(QDict *options, int flags) | |
1557 | { | |
1558 | GLOBAL_STATE_CODE(); | |
1559 | if (!qdict_haskey(options, BDRV_OPT_CACHE_DIRECT)) { | |
1560 | qdict_put_bool(options, BDRV_OPT_CACHE_DIRECT, flags & BDRV_O_NOCACHE); | |
1561 | } | |
1562 | if (!qdict_haskey(options, BDRV_OPT_CACHE_NO_FLUSH)) { | |
1563 | qdict_put_bool(options, BDRV_OPT_CACHE_NO_FLUSH, | |
1564 | flags & BDRV_O_NO_FLUSH); | |
1565 | } | |
1566 | if (!qdict_haskey(options, BDRV_OPT_READ_ONLY)) { | |
1567 | qdict_put_bool(options, BDRV_OPT_READ_ONLY, !(flags & BDRV_O_RDWR)); | |
1568 | } | |
1569 | if (!qdict_haskey(options, BDRV_OPT_AUTO_READ_ONLY)) { | |
1570 | qdict_put_bool(options, BDRV_OPT_AUTO_READ_ONLY, | |
1571 | flags & BDRV_O_AUTO_RDONLY); | |
1572 | } | |
1573 | } | |
1574 | ||
1575 | static void bdrv_assign_node_name(BlockDriverState *bs, | |
1576 | const char *node_name, | |
1577 | Error **errp) | |
1578 | { | |
1579 | char *gen_node_name = NULL; | |
1580 | GLOBAL_STATE_CODE(); | |
1581 | ||
1582 | if (!node_name) { | |
1583 | node_name = gen_node_name = id_generate(ID_BLOCK); | |
1584 | } else if (!id_wellformed(node_name)) { | |
1585 | /* | |
1586 | * Check for empty string or invalid characters, but not if it is | |
1587 | * generated (generated names use characters not available to the user) | |
1588 | */ | |
1589 | error_setg(errp, "Invalid node-name: '%s'", node_name); | |
1590 | return; | |
1591 | } | |
1592 | ||
1593 | /* takes care of avoiding namespaces collisions */ | |
1594 | if (blk_by_name(node_name)) { | |
1595 | error_setg(errp, "node-name=%s is conflicting with a device id", | |
1596 | node_name); | |
1597 | goto out; | |
1598 | } | |
1599 | ||
1600 | /* takes care of avoiding duplicates node names */ | |
1601 | if (bdrv_find_node(node_name)) { | |
1602 | error_setg(errp, "Duplicate nodes with node-name='%s'", node_name); | |
1603 | goto out; | |
1604 | } | |
1605 | ||
1606 | /* Make sure that the node name isn't truncated */ | |
1607 | if (strlen(node_name) >= sizeof(bs->node_name)) { | |
1608 | error_setg(errp, "Node name too long"); | |
1609 | goto out; | |
1610 | } | |
1611 | ||
1612 | /* copy node name into the bs and insert it into the graph list */ | |
1613 | pstrcpy(bs->node_name, sizeof(bs->node_name), node_name); | |
1614 | QTAILQ_INSERT_TAIL(&graph_bdrv_states, bs, node_list); | |
1615 | out: | |
1616 | g_free(gen_node_name); | |
1617 | } | |
1618 | ||
1619 | /* | |
1620 | * The caller must always hold @bs AioContext lock, because this function calls | |
1621 | * bdrv_refresh_total_sectors() which polls when called from non-coroutine | |
1622 | * context. | |
1623 | */ | |
1624 | static int no_coroutine_fn GRAPH_UNLOCKED | |
1625 | bdrv_open_driver(BlockDriverState *bs, BlockDriver *drv, const char *node_name, | |
1626 | QDict *options, int open_flags, Error **errp) | |
1627 | { | |
1628 | AioContext *ctx; | |
1629 | Error *local_err = NULL; | |
1630 | int i, ret; | |
1631 | GLOBAL_STATE_CODE(); | |
1632 | ||
1633 | bdrv_assign_node_name(bs, node_name, &local_err); | |
1634 | if (local_err) { | |
1635 | error_propagate(errp, local_err); | |
1636 | return -EINVAL; | |
1637 | } | |
1638 | ||
1639 | bs->drv = drv; | |
1640 | bs->opaque = g_malloc0(drv->instance_size); | |
1641 | ||
1642 | if (drv->bdrv_file_open) { | |
1643 | assert(!drv->bdrv_needs_filename || bs->filename[0]); | |
1644 | ret = drv->bdrv_file_open(bs, options, open_flags, &local_err); | |
1645 | } else if (drv->bdrv_open) { | |
1646 | ret = drv->bdrv_open(bs, options, open_flags, &local_err); | |
1647 | } else { | |
1648 | ret = 0; | |
1649 | } | |
1650 | ||
1651 | if (ret < 0) { | |
1652 | if (local_err) { | |
1653 | error_propagate(errp, local_err); | |
1654 | } else if (bs->filename[0]) { | |
1655 | error_setg_errno(errp, -ret, "Could not open '%s'", bs->filename); | |
1656 | } else { | |
1657 | error_setg_errno(errp, -ret, "Could not open image"); | |
1658 | } | |
1659 | goto open_failed; | |
1660 | } | |
1661 | ||
1662 | assert(!(bs->supported_read_flags & ~BDRV_REQ_MASK)); | |
1663 | assert(!(bs->supported_write_flags & ~BDRV_REQ_MASK)); | |
1664 | ||
1665 | /* | |
1666 | * Always allow the BDRV_REQ_REGISTERED_BUF optimization hint. This saves | |
1667 | * drivers that pass read/write requests through to a child the trouble of | |
1668 | * declaring support explicitly. | |
1669 | * | |
1670 | * Drivers must not propagate this flag accidentally when they initiate I/O | |
1671 | * to a bounce buffer. That case should be rare though. | |
1672 | */ | |
1673 | bs->supported_read_flags |= BDRV_REQ_REGISTERED_BUF; | |
1674 | bs->supported_write_flags |= BDRV_REQ_REGISTERED_BUF; | |
1675 | ||
1676 | /* Get the context after .bdrv_open, it can change the context */ | |
1677 | ctx = bdrv_get_aio_context(bs); | |
1678 | aio_context_acquire(ctx); | |
1679 | ||
1680 | ret = bdrv_refresh_total_sectors(bs, bs->total_sectors); | |
1681 | if (ret < 0) { | |
1682 | error_setg_errno(errp, -ret, "Could not refresh total sector count"); | |
1683 | aio_context_release(ctx); | |
1684 | return ret; | |
1685 | } | |
1686 | ||
1687 | bdrv_graph_rdlock_main_loop(); | |
1688 | bdrv_refresh_limits(bs, NULL, &local_err); | |
1689 | bdrv_graph_rdunlock_main_loop(); | |
1690 | aio_context_release(ctx); | |
1691 | ||
1692 | if (local_err) { | |
1693 | error_propagate(errp, local_err); | |
1694 | return -EINVAL; | |
1695 | } | |
1696 | ||
1697 | assert(bdrv_opt_mem_align(bs) != 0); | |
1698 | assert(bdrv_min_mem_align(bs) != 0); | |
1699 | assert(is_power_of_2(bs->bl.request_alignment)); | |
1700 | ||
1701 | for (i = 0; i < bs->quiesce_counter; i++) { | |
1702 | if (drv->bdrv_drain_begin) { | |
1703 | drv->bdrv_drain_begin(bs); | |
1704 | } | |
1705 | } | |
1706 | ||
1707 | return 0; | |
1708 | open_failed: | |
1709 | bs->drv = NULL; | |
1710 | ||
1711 | bdrv_graph_wrlock(NULL); | |
1712 | if (bs->file != NULL) { | |
1713 | bdrv_unref_child(bs, bs->file); | |
1714 | assert(!bs->file); | |
1715 | } | |
1716 | bdrv_graph_wrunlock(); | |
1717 | ||
1718 | g_free(bs->opaque); | |
1719 | bs->opaque = NULL; | |
1720 | return ret; | |
1721 | } | |
1722 | ||
1723 | /* | |
1724 | * Create and open a block node. | |
1725 | * | |
1726 | * @options is a QDict of options to pass to the block drivers, or NULL for an | |
1727 | * empty set of options. The reference to the QDict belongs to the block layer | |
1728 | * after the call (even on failure), so if the caller intends to reuse the | |
1729 | * dictionary, it needs to use qobject_ref() before calling bdrv_open. | |
1730 | */ | |
1731 | BlockDriverState *bdrv_new_open_driver_opts(BlockDriver *drv, | |
1732 | const char *node_name, | |
1733 | QDict *options, int flags, | |
1734 | Error **errp) | |
1735 | { | |
1736 | BlockDriverState *bs; | |
1737 | int ret; | |
1738 | ||
1739 | GLOBAL_STATE_CODE(); | |
1740 | ||
1741 | bs = bdrv_new(); | |
1742 | bs->open_flags = flags; | |
1743 | bs->options = options ?: qdict_new(); | |
1744 | bs->explicit_options = qdict_clone_shallow(bs->options); | |
1745 | bs->opaque = NULL; | |
1746 | ||
1747 | update_options_from_flags(bs->options, flags); | |
1748 | ||
1749 | ret = bdrv_open_driver(bs, drv, node_name, bs->options, flags, errp); | |
1750 | if (ret < 0) { | |
1751 | qobject_unref(bs->explicit_options); | |
1752 | bs->explicit_options = NULL; | |
1753 | qobject_unref(bs->options); | |
1754 | bs->options = NULL; | |
1755 | bdrv_unref(bs); | |
1756 | return NULL; | |
1757 | } | |
1758 | ||
1759 | return bs; | |
1760 | } | |
1761 | ||
1762 | /* Create and open a block node. */ | |
1763 | BlockDriverState *bdrv_new_open_driver(BlockDriver *drv, const char *node_name, | |
1764 | int flags, Error **errp) | |
1765 | { | |
1766 | GLOBAL_STATE_CODE(); | |
1767 | return bdrv_new_open_driver_opts(drv, node_name, NULL, flags, errp); | |
1768 | } | |
1769 | ||
1770 | QemuOptsList bdrv_runtime_opts = { | |
1771 | .name = "bdrv_common", | |
1772 | .head = QTAILQ_HEAD_INITIALIZER(bdrv_runtime_opts.head), | |
1773 | .desc = { | |
1774 | { | |
1775 | .name = "node-name", | |
1776 | .type = QEMU_OPT_STRING, | |
1777 | .help = "Node name of the block device node", | |
1778 | }, | |
1779 | { | |
1780 | .name = "driver", | |
1781 | .type = QEMU_OPT_STRING, | |
1782 | .help = "Block driver to use for the node", | |
1783 | }, | |
1784 | { | |
1785 | .name = BDRV_OPT_CACHE_DIRECT, | |
1786 | .type = QEMU_OPT_BOOL, | |
1787 | .help = "Bypass software writeback cache on the host", | |
1788 | }, | |
1789 | { | |
1790 | .name = BDRV_OPT_CACHE_NO_FLUSH, | |
1791 | .type = QEMU_OPT_BOOL, | |
1792 | .help = "Ignore flush requests", | |
1793 | }, | |
1794 | { | |
1795 | .name = BDRV_OPT_READ_ONLY, | |
1796 | .type = QEMU_OPT_BOOL, | |
1797 | .help = "Node is opened in read-only mode", | |
1798 | }, | |
1799 | { | |
1800 | .name = BDRV_OPT_AUTO_READ_ONLY, | |
1801 | .type = QEMU_OPT_BOOL, | |
1802 | .help = "Node can become read-only if opening read-write fails", | |
1803 | }, | |
1804 | { | |
1805 | .name = "detect-zeroes", | |
1806 | .type = QEMU_OPT_STRING, | |
1807 | .help = "try to optimize zero writes (off, on, unmap)", | |
1808 | }, | |
1809 | { | |
1810 | .name = BDRV_OPT_DISCARD, | |
1811 | .type = QEMU_OPT_STRING, | |
1812 | .help = "discard operation (ignore/off, unmap/on)", | |
1813 | }, | |
1814 | { | |
1815 | .name = BDRV_OPT_FORCE_SHARE, | |
1816 | .type = QEMU_OPT_BOOL, | |
1817 | .help = "always accept other writers (default: off)", | |
1818 | }, | |
1819 | { /* end of list */ } | |
1820 | }, | |
1821 | }; | |
1822 | ||
1823 | QemuOptsList bdrv_create_opts_simple = { | |
1824 | .name = "simple-create-opts", | |
1825 | .head = QTAILQ_HEAD_INITIALIZER(bdrv_create_opts_simple.head), | |
1826 | .desc = { | |
1827 | { | |
1828 | .name = BLOCK_OPT_SIZE, | |
1829 | .type = QEMU_OPT_SIZE, | |
1830 | .help = "Virtual disk size" | |
1831 | }, | |
1832 | { | |
1833 | .name = BLOCK_OPT_PREALLOC, | |
1834 | .type = QEMU_OPT_STRING, | |
1835 | .help = "Preallocation mode (allowed values: off)" | |
1836 | }, | |
1837 | { /* end of list */ } | |
1838 | } | |
1839 | }; | |
1840 | ||
1841 | /* | |
1842 | * Common part for opening disk images and files | |
1843 | * | |
1844 | * Removes all processed options from *options. | |
1845 | */ | |
1846 | static int bdrv_open_common(BlockDriverState *bs, BlockBackend *file, | |
1847 | QDict *options, Error **errp) | |
1848 | { | |
1849 | int ret, open_flags; | |
1850 | const char *filename; | |
1851 | const char *driver_name = NULL; | |
1852 | const char *node_name = NULL; | |
1853 | const char *discard; | |
1854 | QemuOpts *opts; | |
1855 | BlockDriver *drv; | |
1856 | Error *local_err = NULL; | |
1857 | bool ro; | |
1858 | ||
1859 | GLOBAL_STATE_CODE(); | |
1860 | ||
1861 | bdrv_graph_rdlock_main_loop(); | |
1862 | assert(bs->file == NULL); | |
1863 | assert(options != NULL && bs->options != options); | |
1864 | bdrv_graph_rdunlock_main_loop(); | |
1865 | ||
1866 | opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort); | |
1867 | if (!qemu_opts_absorb_qdict(opts, options, errp)) { | |
1868 | ret = -EINVAL; | |
1869 | goto fail_opts; | |
1870 | } | |
1871 | ||
1872 | update_flags_from_options(&bs->open_flags, opts); | |
1873 | ||
1874 | driver_name = qemu_opt_get(opts, "driver"); | |
1875 | drv = bdrv_find_format(driver_name); | |
1876 | assert(drv != NULL); | |
1877 | ||
1878 | bs->force_share = qemu_opt_get_bool(opts, BDRV_OPT_FORCE_SHARE, false); | |
1879 | ||
1880 | if (bs->force_share && (bs->open_flags & BDRV_O_RDWR)) { | |
1881 | error_setg(errp, | |
1882 | BDRV_OPT_FORCE_SHARE | |
1883 | "=on can only be used with read-only images"); | |
1884 | ret = -EINVAL; | |
1885 | goto fail_opts; | |
1886 | } | |
1887 | ||
1888 | if (file != NULL) { | |
1889 | bdrv_graph_rdlock_main_loop(); | |
1890 | bdrv_refresh_filename(blk_bs(file)); | |
1891 | bdrv_graph_rdunlock_main_loop(); | |
1892 | ||
1893 | filename = blk_bs(file)->filename; | |
1894 | } else { | |
1895 | /* | |
1896 | * Caution: while qdict_get_try_str() is fine, getting | |
1897 | * non-string types would require more care. When @options | |
1898 | * come from -blockdev or blockdev_add, its members are typed | |
1899 | * according to the QAPI schema, but when they come from | |
1900 | * -drive, they're all QString. | |
1901 | */ | |
1902 | filename = qdict_get_try_str(options, "filename"); | |
1903 | } | |
1904 | ||
1905 | if (drv->bdrv_needs_filename && (!filename || !filename[0])) { | |
1906 | error_setg(errp, "The '%s' block driver requires a file name", | |
1907 | drv->format_name); | |
1908 | ret = -EINVAL; | |
1909 | goto fail_opts; | |
1910 | } | |
1911 | ||
1912 | trace_bdrv_open_common(bs, filename ?: "", bs->open_flags, | |
1913 | drv->format_name); | |
1914 | ||
1915 | ro = bdrv_is_read_only(bs); | |
1916 | ||
1917 | if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv, ro)) { | |
1918 | if (!ro && bdrv_is_whitelisted(drv, true)) { | |
1919 | bdrv_graph_rdlock_main_loop(); | |
1920 | ret = bdrv_apply_auto_read_only(bs, NULL, NULL); | |
1921 | bdrv_graph_rdunlock_main_loop(); | |
1922 | } else { | |
1923 | ret = -ENOTSUP; | |
1924 | } | |
1925 | if (ret < 0) { | |
1926 | error_setg(errp, | |
1927 | !ro && bdrv_is_whitelisted(drv, true) | |
1928 | ? "Driver '%s' can only be used for read-only devices" | |
1929 | : "Driver '%s' is not whitelisted", | |
1930 | drv->format_name); | |
1931 | goto fail_opts; | |
1932 | } | |
1933 | } | |
1934 | ||
1935 | /* bdrv_new() and bdrv_close() make it so */ | |
1936 | assert(qatomic_read(&bs->copy_on_read) == 0); | |
1937 | ||
1938 | if (bs->open_flags & BDRV_O_COPY_ON_READ) { | |
1939 | if (!ro) { | |
1940 | bdrv_enable_copy_on_read(bs); | |
1941 | } else { | |
1942 | error_setg(errp, "Can't use copy-on-read on read-only device"); | |
1943 | ret = -EINVAL; | |
1944 | goto fail_opts; | |
1945 | } | |
1946 | } | |
1947 | ||
1948 | discard = qemu_opt_get(opts, BDRV_OPT_DISCARD); | |
1949 | if (discard != NULL) { | |
1950 | if (bdrv_parse_discard_flags(discard, &bs->open_flags) != 0) { | |
1951 | error_setg(errp, "Invalid discard option"); | |
1952 | ret = -EINVAL; | |
1953 | goto fail_opts; | |
1954 | } | |
1955 | } | |
1956 | ||
1957 | bs->detect_zeroes = | |
1958 | bdrv_parse_detect_zeroes(opts, bs->open_flags, &local_err); | |
1959 | if (local_err) { | |
1960 | error_propagate(errp, local_err); | |
1961 | ret = -EINVAL; | |
1962 | goto fail_opts; | |
1963 | } | |
1964 | ||
1965 | if (filename != NULL) { | |
1966 | pstrcpy(bs->filename, sizeof(bs->filename), filename); | |
1967 | } else { | |
1968 | bs->filename[0] = '\0'; | |
1969 | } | |
1970 | pstrcpy(bs->exact_filename, sizeof(bs->exact_filename), bs->filename); | |
1971 | ||
1972 | /* Open the image, either directly or using a protocol */ | |
1973 | open_flags = bdrv_open_flags(bs, bs->open_flags); | |
1974 | node_name = qemu_opt_get(opts, "node-name"); | |
1975 | ||
1976 | assert(!drv->bdrv_file_open || file == NULL); | |
1977 | ret = bdrv_open_driver(bs, drv, node_name, options, open_flags, errp); | |
1978 | if (ret < 0) { | |
1979 | goto fail_opts; | |
1980 | } | |
1981 | ||
1982 | qemu_opts_del(opts); | |
1983 | return 0; | |
1984 | ||
1985 | fail_opts: | |
1986 | qemu_opts_del(opts); | |
1987 | return ret; | |
1988 | } | |
1989 | ||
1990 | static QDict *parse_json_filename(const char *filename, Error **errp) | |
1991 | { | |
1992 | QObject *options_obj; | |
1993 | QDict *options; | |
1994 | int ret; | |
1995 | GLOBAL_STATE_CODE(); | |
1996 | ||
1997 | ret = strstart(filename, "json:", &filename); | |
1998 | assert(ret); | |
1999 | ||
2000 | options_obj = qobject_from_json(filename, errp); | |
2001 | if (!options_obj) { | |
2002 | error_prepend(errp, "Could not parse the JSON options: "); | |
2003 | return NULL; | |
2004 | } | |
2005 | ||
2006 | options = qobject_to(QDict, options_obj); | |
2007 | if (!options) { | |
2008 | qobject_unref(options_obj); | |
2009 | error_setg(errp, "Invalid JSON object given"); | |
2010 | return NULL; | |
2011 | } | |
2012 | ||
2013 | qdict_flatten(options); | |
2014 | ||
2015 | return options; | |
2016 | } | |
2017 | ||
2018 | static void parse_json_protocol(QDict *options, const char **pfilename, | |
2019 | Error **errp) | |
2020 | { | |
2021 | QDict *json_options; | |
2022 | Error *local_err = NULL; | |
2023 | GLOBAL_STATE_CODE(); | |
2024 | ||
2025 | /* Parse json: pseudo-protocol */ | |
2026 | if (!*pfilename || !g_str_has_prefix(*pfilename, "json:")) { | |
2027 | return; | |
2028 | } | |
2029 | ||
2030 | json_options = parse_json_filename(*pfilename, &local_err); | |
2031 | if (local_err) { | |
2032 | error_propagate(errp, local_err); | |
2033 | return; | |
2034 | } | |
2035 | ||
2036 | /* Options given in the filename have lower priority than options | |
2037 | * specified directly */ | |
2038 | qdict_join(options, json_options, false); | |
2039 | qobject_unref(json_options); | |
2040 | *pfilename = NULL; | |
2041 | } | |
2042 | ||
2043 | /* | |
2044 | * Fills in default options for opening images and converts the legacy | |
2045 | * filename/flags pair to option QDict entries. | |
2046 | * The BDRV_O_PROTOCOL flag in *flags will be set or cleared accordingly if a | |
2047 | * block driver has been specified explicitly. | |
2048 | */ | |
2049 | static int bdrv_fill_options(QDict **options, const char *filename, | |
2050 | int *flags, Error **errp) | |
2051 | { | |
2052 | const char *drvname; | |
2053 | bool protocol = *flags & BDRV_O_PROTOCOL; | |
2054 | bool parse_filename = false; | |
2055 | BlockDriver *drv = NULL; | |
2056 | Error *local_err = NULL; | |
2057 | ||
2058 | GLOBAL_STATE_CODE(); | |
2059 | ||
2060 | /* | |
2061 | * Caution: while qdict_get_try_str() is fine, getting non-string | |
2062 | * types would require more care. When @options come from | |
2063 | * -blockdev or blockdev_add, its members are typed according to | |
2064 | * the QAPI schema, but when they come from -drive, they're all | |
2065 | * QString. | |
2066 | */ | |
2067 | drvname = qdict_get_try_str(*options, "driver"); | |
2068 | if (drvname) { | |
2069 | drv = bdrv_find_format(drvname); | |
2070 | if (!drv) { | |
2071 | error_setg(errp, "Unknown driver '%s'", drvname); | |
2072 | return -ENOENT; | |
2073 | } | |
2074 | /* If the user has explicitly specified the driver, this choice should | |
2075 | * override the BDRV_O_PROTOCOL flag */ | |
2076 | protocol = drv->bdrv_file_open; | |
2077 | } | |
2078 | ||
2079 | if (protocol) { | |
2080 | *flags |= BDRV_O_PROTOCOL; | |
2081 | } else { | |
2082 | *flags &= ~BDRV_O_PROTOCOL; | |
2083 | } | |
2084 | ||
2085 | /* Translate cache options from flags into options */ | |
2086 | update_options_from_flags(*options, *flags); | |
2087 | ||
2088 | /* Fetch the file name from the options QDict if necessary */ | |
2089 | if (protocol && filename) { | |
2090 | if (!qdict_haskey(*options, "filename")) { | |
2091 | qdict_put_str(*options, "filename", filename); | |
2092 | parse_filename = true; | |
2093 | } else { | |
2094 | error_setg(errp, "Can't specify 'file' and 'filename' options at " | |
2095 | "the same time"); | |
2096 | return -EINVAL; | |
2097 | } | |
2098 | } | |
2099 | ||
2100 | /* Find the right block driver */ | |
2101 | /* See cautionary note on accessing @options above */ | |
2102 | filename = qdict_get_try_str(*options, "filename"); | |
2103 | ||
2104 | if (!drvname && protocol) { | |
2105 | if (filename) { | |
2106 | drv = bdrv_find_protocol(filename, parse_filename, errp); | |
2107 | if (!drv) { | |
2108 | return -EINVAL; | |
2109 | } | |
2110 | ||
2111 | drvname = drv->format_name; | |
2112 | qdict_put_str(*options, "driver", drvname); | |
2113 | } else { | |
2114 | error_setg(errp, "Must specify either driver or file"); | |
2115 | return -EINVAL; | |
2116 | } | |
2117 | } | |
2118 | ||
2119 | assert(drv || !protocol); | |
2120 | ||
2121 | /* Driver-specific filename parsing */ | |
2122 | if (drv && drv->bdrv_parse_filename && parse_filename) { | |
2123 | drv->bdrv_parse_filename(filename, *options, &local_err); | |
2124 | if (local_err) { | |
2125 | error_propagate(errp, local_err); | |
2126 | return -EINVAL; | |
2127 | } | |
2128 | ||
2129 | if (!drv->bdrv_needs_filename) { | |
2130 | qdict_del(*options, "filename"); | |
2131 | } | |
2132 | } | |
2133 | ||
2134 | return 0; | |
2135 | } | |
2136 | ||
2137 | typedef struct BlockReopenQueueEntry { | |
2138 | bool prepared; | |
2139 | BDRVReopenState state; | |
2140 | QTAILQ_ENTRY(BlockReopenQueueEntry) entry; | |
2141 | } BlockReopenQueueEntry; | |
2142 | ||
2143 | /* | |
2144 | * Return the flags that @bs will have after the reopens in @q have | |
2145 | * successfully completed. If @q is NULL (or @bs is not contained in @q), | |
2146 | * return the current flags. | |
2147 | */ | |
2148 | static int bdrv_reopen_get_flags(BlockReopenQueue *q, BlockDriverState *bs) | |
2149 | { | |
2150 | BlockReopenQueueEntry *entry; | |
2151 | ||
2152 | if (q != NULL) { | |
2153 | QTAILQ_FOREACH(entry, q, entry) { | |
2154 | if (entry->state.bs == bs) { | |
2155 | return entry->state.flags; | |
2156 | } | |
2157 | } | |
2158 | } | |
2159 | ||
2160 | return bs->open_flags; | |
2161 | } | |
2162 | ||
2163 | /* Returns whether the image file can be written to after the reopen queue @q | |
2164 | * has been successfully applied, or right now if @q is NULL. */ | |
2165 | static bool bdrv_is_writable_after_reopen(BlockDriverState *bs, | |
2166 | BlockReopenQueue *q) | |
2167 | { | |
2168 | int flags = bdrv_reopen_get_flags(q, bs); | |
2169 | ||
2170 | return (flags & (BDRV_O_RDWR | BDRV_O_INACTIVE)) == BDRV_O_RDWR; | |
2171 | } | |
2172 | ||
2173 | /* | |
2174 | * Return whether the BDS can be written to. This is not necessarily | |
2175 | * the same as !bdrv_is_read_only(bs), as inactivated images may not | |
2176 | * be written to but do not count as read-only images. | |
2177 | */ | |
2178 | bool bdrv_is_writable(BlockDriverState *bs) | |
2179 | { | |
2180 | IO_CODE(); | |
2181 | return bdrv_is_writable_after_reopen(bs, NULL); | |
2182 | } | |
2183 | ||
2184 | static char *bdrv_child_user_desc(BdrvChild *c) | |
2185 | { | |
2186 | GLOBAL_STATE_CODE(); | |
2187 | return c->klass->get_parent_desc(c); | |
2188 | } | |
2189 | ||
2190 | /* | |
2191 | * Check that @a allows everything that @b needs. @a and @b must reference same | |
2192 | * child node. | |
2193 | */ | |
2194 | static bool bdrv_a_allow_b(BdrvChild *a, BdrvChild *b, Error **errp) | |
2195 | { | |
2196 | const char *child_bs_name; | |
2197 | g_autofree char *a_user = NULL; | |
2198 | g_autofree char *b_user = NULL; | |
2199 | g_autofree char *perms = NULL; | |
2200 | ||
2201 | assert(a->bs); | |
2202 | assert(a->bs == b->bs); | |
2203 | GLOBAL_STATE_CODE(); | |
2204 | ||
2205 | if ((b->perm & a->shared_perm) == b->perm) { | |
2206 | return true; | |
2207 | } | |
2208 | ||
2209 | child_bs_name = bdrv_get_node_name(b->bs); | |
2210 | a_user = bdrv_child_user_desc(a); | |
2211 | b_user = bdrv_child_user_desc(b); | |
2212 | perms = bdrv_perm_names(b->perm & ~a->shared_perm); | |
2213 | ||
2214 | error_setg(errp, "Permission conflict on node '%s': permissions '%s' are " | |
2215 | "both required by %s (uses node '%s' as '%s' child) and " | |
2216 | "unshared by %s (uses node '%s' as '%s' child).", | |
2217 | child_bs_name, perms, | |
2218 | b_user, child_bs_name, b->name, | |
2219 | a_user, child_bs_name, a->name); | |
2220 | ||
2221 | return false; | |
2222 | } | |
2223 | ||
2224 | static bool GRAPH_RDLOCK | |
2225 | bdrv_parent_perms_conflict(BlockDriverState *bs, Error **errp) | |
2226 | { | |
2227 | BdrvChild *a, *b; | |
2228 | GLOBAL_STATE_CODE(); | |
2229 | ||
2230 | /* | |
2231 | * During the loop we'll look at each pair twice. That's correct because | |
2232 | * bdrv_a_allow_b() is asymmetric and we should check each pair in both | |
2233 | * directions. | |
2234 | */ | |
2235 | QLIST_FOREACH(a, &bs->parents, next_parent) { | |
2236 | QLIST_FOREACH(b, &bs->parents, next_parent) { | |
2237 | if (a == b) { | |
2238 | continue; | |
2239 | } | |
2240 | ||
2241 | if (!bdrv_a_allow_b(a, b, errp)) { | |
2242 | return true; | |
2243 | } | |
2244 | } | |
2245 | } | |
2246 | ||
2247 | return false; | |
2248 | } | |
2249 | ||
2250 | static void GRAPH_RDLOCK | |
2251 | bdrv_child_perm(BlockDriverState *bs, BlockDriverState *child_bs, | |
2252 | BdrvChild *c, BdrvChildRole role, | |
2253 | BlockReopenQueue *reopen_queue, | |
2254 | uint64_t parent_perm, uint64_t parent_shared, | |
2255 | uint64_t *nperm, uint64_t *nshared) | |
2256 | { | |
2257 | assert(bs->drv && bs->drv->bdrv_child_perm); | |
2258 | GLOBAL_STATE_CODE(); | |
2259 | bs->drv->bdrv_child_perm(bs, c, role, reopen_queue, | |
2260 | parent_perm, parent_shared, | |
2261 | nperm, nshared); | |
2262 | /* TODO Take force_share from reopen_queue */ | |
2263 | if (child_bs && child_bs->force_share) { | |
2264 | *nshared = BLK_PERM_ALL; | |
2265 | } | |
2266 | } | |
2267 | ||
2268 | /* | |
2269 | * Adds the whole subtree of @bs (including @bs itself) to the @list (except for | |
2270 | * nodes that are already in the @list, of course) so that final list is | |
2271 | * topologically sorted. Return the result (GSList @list object is updated, so | |
2272 | * don't use old reference after function call). | |
2273 | * | |
2274 | * On function start @list must be already topologically sorted and for any node | |
2275 | * in the @list the whole subtree of the node must be in the @list as well. The | |
2276 | * simplest way to satisfy this criteria: use only result of | |
2277 | * bdrv_topological_dfs() or NULL as @list parameter. | |
2278 | */ | |
2279 | static GSList * GRAPH_RDLOCK | |
2280 | bdrv_topological_dfs(GSList *list, GHashTable *found, BlockDriverState *bs) | |
2281 | { | |
2282 | BdrvChild *child; | |
2283 | g_autoptr(GHashTable) local_found = NULL; | |
2284 | ||
2285 | GLOBAL_STATE_CODE(); | |
2286 | ||
2287 | if (!found) { | |
2288 | assert(!list); | |
2289 | found = local_found = g_hash_table_new(NULL, NULL); | |
2290 | } | |
2291 | ||
2292 | if (g_hash_table_contains(found, bs)) { | |
2293 | return list; | |
2294 | } | |
2295 | g_hash_table_add(found, bs); | |
2296 | ||
2297 | QLIST_FOREACH(child, &bs->children, next) { | |
2298 | list = bdrv_topological_dfs(list, found, child->bs); | |
2299 | } | |
2300 | ||
2301 | return g_slist_prepend(list, bs); | |
2302 | } | |
2303 | ||
2304 | typedef struct BdrvChildSetPermState { | |
2305 | BdrvChild *child; | |
2306 | uint64_t old_perm; | |
2307 | uint64_t old_shared_perm; | |
2308 | } BdrvChildSetPermState; | |
2309 | ||
2310 | static void bdrv_child_set_perm_abort(void *opaque) | |
2311 | { | |
2312 | BdrvChildSetPermState *s = opaque; | |
2313 | ||
2314 | GLOBAL_STATE_CODE(); | |
2315 | ||
2316 | s->child->perm = s->old_perm; | |
2317 | s->child->shared_perm = s->old_shared_perm; | |
2318 | } | |
2319 | ||
2320 | static TransactionActionDrv bdrv_child_set_pem_drv = { | |
2321 | .abort = bdrv_child_set_perm_abort, | |
2322 | .clean = g_free, | |
2323 | }; | |
2324 | ||
2325 | static void bdrv_child_set_perm(BdrvChild *c, uint64_t perm, | |
2326 | uint64_t shared, Transaction *tran) | |
2327 | { | |
2328 | BdrvChildSetPermState *s = g_new(BdrvChildSetPermState, 1); | |
2329 | GLOBAL_STATE_CODE(); | |
2330 | ||
2331 | *s = (BdrvChildSetPermState) { | |
2332 | .child = c, | |
2333 | .old_perm = c->perm, | |
2334 | .old_shared_perm = c->shared_perm, | |
2335 | }; | |
2336 | ||
2337 | c->perm = perm; | |
2338 | c->shared_perm = shared; | |
2339 | ||
2340 | tran_add(tran, &bdrv_child_set_pem_drv, s); | |
2341 | } | |
2342 | ||
2343 | static void GRAPH_RDLOCK bdrv_drv_set_perm_commit(void *opaque) | |
2344 | { | |
2345 | BlockDriverState *bs = opaque; | |
2346 | uint64_t cumulative_perms, cumulative_shared_perms; | |
2347 | GLOBAL_STATE_CODE(); | |
2348 | ||
2349 | if (bs->drv->bdrv_set_perm) { | |
2350 | bdrv_get_cumulative_perm(bs, &cumulative_perms, | |
2351 | &cumulative_shared_perms); | |
2352 | bs->drv->bdrv_set_perm(bs, cumulative_perms, cumulative_shared_perms); | |
2353 | } | |
2354 | } | |
2355 | ||
2356 | static void GRAPH_RDLOCK bdrv_drv_set_perm_abort(void *opaque) | |
2357 | { | |
2358 | BlockDriverState *bs = opaque; | |
2359 | GLOBAL_STATE_CODE(); | |
2360 | ||
2361 | if (bs->drv->bdrv_abort_perm_update) { | |
2362 | bs->drv->bdrv_abort_perm_update(bs); | |
2363 | } | |
2364 | } | |
2365 | ||
2366 | TransactionActionDrv bdrv_drv_set_perm_drv = { | |
2367 | .abort = bdrv_drv_set_perm_abort, | |
2368 | .commit = bdrv_drv_set_perm_commit, | |
2369 | }; | |
2370 | ||
2371 | /* | |
2372 | * After calling this function, the transaction @tran may only be completed | |
2373 | * while holding a reader lock for the graph. | |
2374 | */ | |
2375 | static int GRAPH_RDLOCK | |
2376 | bdrv_drv_set_perm(BlockDriverState *bs, uint64_t perm, uint64_t shared_perm, | |
2377 | Transaction *tran, Error **errp) | |
2378 | { | |
2379 | GLOBAL_STATE_CODE(); | |
2380 | if (!bs->drv) { | |
2381 | return 0; | |
2382 | } | |
2383 | ||
2384 | if (bs->drv->bdrv_check_perm) { | |
2385 | int ret = bs->drv->bdrv_check_perm(bs, perm, shared_perm, errp); | |
2386 | if (ret < 0) { | |
2387 | return ret; | |
2388 | } | |
2389 | } | |
2390 | ||
2391 | if (tran) { | |
2392 | tran_add(tran, &bdrv_drv_set_perm_drv, bs); | |
2393 | } | |
2394 | ||
2395 | return 0; | |
2396 | } | |
2397 | ||
2398 | typedef struct BdrvReplaceChildState { | |
2399 | BdrvChild *child; | |
2400 | BlockDriverState *old_bs; | |
2401 | } BdrvReplaceChildState; | |
2402 | ||
2403 | static void GRAPH_WRLOCK bdrv_replace_child_commit(void *opaque) | |
2404 | { | |
2405 | BdrvReplaceChildState *s = opaque; | |
2406 | GLOBAL_STATE_CODE(); | |
2407 | ||
2408 | bdrv_schedule_unref(s->old_bs); | |
2409 | } | |
2410 | ||
2411 | static void GRAPH_WRLOCK bdrv_replace_child_abort(void *opaque) | |
2412 | { | |
2413 | BdrvReplaceChildState *s = opaque; | |
2414 | BlockDriverState *new_bs = s->child->bs; | |
2415 | ||
2416 | GLOBAL_STATE_CODE(); | |
2417 | assert_bdrv_graph_writable(); | |
2418 | ||
2419 | /* old_bs reference is transparently moved from @s to @s->child */ | |
2420 | if (!s->child->bs) { | |
2421 | /* | |
2422 | * The parents were undrained when removing old_bs from the child. New | |
2423 | * requests can't have been made, though, because the child was empty. | |
2424 | * | |
2425 | * TODO Make bdrv_replace_child_noperm() transactionable to avoid | |
2426 | * undraining the parent in the first place. Once this is done, having | |
2427 | * new_bs drained when calling bdrv_replace_child_tran() is not a | |
2428 | * requirement any more. | |
2429 | */ | |
2430 | bdrv_parent_drained_begin_single(s->child); | |
2431 | assert(!bdrv_parent_drained_poll_single(s->child)); | |
2432 | } | |
2433 | assert(s->child->quiesced_parent); | |
2434 | bdrv_replace_child_noperm(s->child, s->old_bs); | |
2435 | ||
2436 | bdrv_unref(new_bs); | |
2437 | } | |
2438 | ||
2439 | static TransactionActionDrv bdrv_replace_child_drv = { | |
2440 | .commit = bdrv_replace_child_commit, | |
2441 | .abort = bdrv_replace_child_abort, | |
2442 | .clean = g_free, | |
2443 | }; | |
2444 | ||
2445 | /* | |
2446 | * bdrv_replace_child_tran | |
2447 | * | |
2448 | * Note: real unref of old_bs is done only on commit. | |
2449 | * | |
2450 | * Both @child->bs and @new_bs (if non-NULL) must be drained. @new_bs must be | |
2451 | * kept drained until the transaction is completed. | |
2452 | * | |
2453 | * After calling this function, the transaction @tran may only be completed | |
2454 | * while holding a writer lock for the graph. | |
2455 | * | |
2456 | * The function doesn't update permissions, caller is responsible for this. | |
2457 | */ | |
2458 | static void GRAPH_WRLOCK | |
2459 | bdrv_replace_child_tran(BdrvChild *child, BlockDriverState *new_bs, | |
2460 | Transaction *tran) | |
2461 | { | |
2462 | BdrvReplaceChildState *s = g_new(BdrvReplaceChildState, 1); | |
2463 | ||
2464 | assert(child->quiesced_parent); | |
2465 | assert(!new_bs || new_bs->quiesce_counter); | |
2466 | ||
2467 | *s = (BdrvReplaceChildState) { | |
2468 | .child = child, | |
2469 | .old_bs = child->bs, | |
2470 | }; | |
2471 | tran_add(tran, &bdrv_replace_child_drv, s); | |
2472 | ||
2473 | if (new_bs) { | |
2474 | bdrv_ref(new_bs); | |
2475 | } | |
2476 | ||
2477 | bdrv_replace_child_noperm(child, new_bs); | |
2478 | /* old_bs reference is transparently moved from @child to @s */ | |
2479 | } | |
2480 | ||
2481 | /* | |
2482 | * Refresh permissions in @bs subtree. The function is intended to be called | |
2483 | * after some graph modification that was done without permission update. | |
2484 | * | |
2485 | * After calling this function, the transaction @tran may only be completed | |
2486 | * while holding a reader lock for the graph. | |
2487 | */ | |
2488 | static int GRAPH_RDLOCK | |
2489 | bdrv_node_refresh_perm(BlockDriverState *bs, BlockReopenQueue *q, | |
2490 | Transaction *tran, Error **errp) | |
2491 | { | |
2492 | BlockDriver *drv = bs->drv; | |
2493 | BdrvChild *c; | |
2494 | int ret; | |
2495 | uint64_t cumulative_perms, cumulative_shared_perms; | |
2496 | GLOBAL_STATE_CODE(); | |
2497 | ||
2498 | bdrv_get_cumulative_perm(bs, &cumulative_perms, &cumulative_shared_perms); | |
2499 | ||
2500 | /* Write permissions never work with read-only images */ | |
2501 | if ((cumulative_perms & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)) && | |
2502 | !bdrv_is_writable_after_reopen(bs, q)) | |
2503 | { | |
2504 | if (!bdrv_is_writable_after_reopen(bs, NULL)) { | |
2505 | error_setg(errp, "Block node is read-only"); | |
2506 | } else { | |
2507 | error_setg(errp, "Read-only block node '%s' cannot support " | |
2508 | "read-write users", bdrv_get_node_name(bs)); | |
2509 | } | |
2510 | ||
2511 | return -EPERM; | |
2512 | } | |
2513 | ||
2514 | /* | |
2515 | * Unaligned requests will automatically be aligned to bl.request_alignment | |
2516 | * and without RESIZE we can't extend requests to write to space beyond the | |
2517 | * end of the image, so it's required that the image size is aligned. | |
2518 | */ | |
2519 | if ((cumulative_perms & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)) && | |
2520 | !(cumulative_perms & BLK_PERM_RESIZE)) | |
2521 | { | |
2522 | if ((bs->total_sectors * BDRV_SECTOR_SIZE) % bs->bl.request_alignment) { | |
2523 | error_setg(errp, "Cannot get 'write' permission without 'resize': " | |
2524 | "Image size is not a multiple of request " | |
2525 | "alignment"); | |
2526 | return -EPERM; | |
2527 | } | |
2528 | } | |
2529 | ||
2530 | /* Check this node */ | |
2531 | if (!drv) { | |
2532 | return 0; | |
2533 | } | |
2534 | ||
2535 | ret = bdrv_drv_set_perm(bs, cumulative_perms, cumulative_shared_perms, tran, | |
2536 | errp); | |
2537 | if (ret < 0) { | |
2538 | return ret; | |
2539 | } | |
2540 | ||
2541 | /* Drivers that never have children can omit .bdrv_child_perm() */ | |
2542 | if (!drv->bdrv_child_perm) { | |
2543 | assert(QLIST_EMPTY(&bs->children)); | |
2544 | return 0; | |
2545 | } | |
2546 | ||
2547 | /* Check all children */ | |
2548 | QLIST_FOREACH(c, &bs->children, next) { | |
2549 | uint64_t cur_perm, cur_shared; | |
2550 | ||
2551 | bdrv_child_perm(bs, c->bs, c, c->role, q, | |
2552 | cumulative_perms, cumulative_shared_perms, | |
2553 | &cur_perm, &cur_shared); | |
2554 | bdrv_child_set_perm(c, cur_perm, cur_shared, tran); | |
2555 | } | |
2556 | ||
2557 | return 0; | |
2558 | } | |
2559 | ||
2560 | /* | |
2561 | * @list is a product of bdrv_topological_dfs() (may be called several times) - | |
2562 | * a topologically sorted subgraph. | |
2563 | * | |
2564 | * After calling this function, the transaction @tran may only be completed | |
2565 | * while holding a reader lock for the graph. | |
2566 | */ | |
2567 | static int GRAPH_RDLOCK | |
2568 | bdrv_do_refresh_perms(GSList *list, BlockReopenQueue *q, Transaction *tran, | |
2569 | Error **errp) | |
2570 | { | |
2571 | int ret; | |
2572 | BlockDriverState *bs; | |
2573 | GLOBAL_STATE_CODE(); | |
2574 | ||
2575 | for ( ; list; list = list->next) { | |
2576 | bs = list->data; | |
2577 | ||
2578 | if (bdrv_parent_perms_conflict(bs, errp)) { | |
2579 | return -EINVAL; | |
2580 | } | |
2581 | ||
2582 | ret = bdrv_node_refresh_perm(bs, q, tran, errp); | |
2583 | if (ret < 0) { | |
2584 | return ret; | |
2585 | } | |
2586 | } | |
2587 | ||
2588 | return 0; | |
2589 | } | |
2590 | ||
2591 | /* | |
2592 | * @list is any list of nodes. List is completed by all subtrees and | |
2593 | * topologically sorted. It's not a problem if some node occurs in the @list | |
2594 | * several times. | |
2595 | * | |
2596 | * After calling this function, the transaction @tran may only be completed | |
2597 | * while holding a reader lock for the graph. | |
2598 | */ | |
2599 | static int GRAPH_RDLOCK | |
2600 | bdrv_list_refresh_perms(GSList *list, BlockReopenQueue *q, Transaction *tran, | |
2601 | Error **errp) | |
2602 | { | |
2603 | g_autoptr(GHashTable) found = g_hash_table_new(NULL, NULL); | |
2604 | g_autoptr(GSList) refresh_list = NULL; | |
2605 | ||
2606 | for ( ; list; list = list->next) { | |
2607 | refresh_list = bdrv_topological_dfs(refresh_list, found, list->data); | |
2608 | } | |
2609 | ||
2610 | return bdrv_do_refresh_perms(refresh_list, q, tran, errp); | |
2611 | } | |
2612 | ||
2613 | void bdrv_get_cumulative_perm(BlockDriverState *bs, uint64_t *perm, | |
2614 | uint64_t *shared_perm) | |
2615 | { | |
2616 | BdrvChild *c; | |
2617 | uint64_t cumulative_perms = 0; | |
2618 | uint64_t cumulative_shared_perms = BLK_PERM_ALL; | |
2619 | ||
2620 | GLOBAL_STATE_CODE(); | |
2621 | ||
2622 | QLIST_FOREACH(c, &bs->parents, next_parent) { | |
2623 | cumulative_perms |= c->perm; | |
2624 | cumulative_shared_perms &= c->shared_perm; | |
2625 | } | |
2626 | ||
2627 | *perm = cumulative_perms; | |
2628 | *shared_perm = cumulative_shared_perms; | |
2629 | } | |
2630 | ||
2631 | char *bdrv_perm_names(uint64_t perm) | |
2632 | { | |
2633 | struct perm_name { | |
2634 | uint64_t perm; | |
2635 | const char *name; | |
2636 | } permissions[] = { | |
2637 | { BLK_PERM_CONSISTENT_READ, "consistent read" }, | |
2638 | { BLK_PERM_WRITE, "write" }, | |
2639 | { BLK_PERM_WRITE_UNCHANGED, "write unchanged" }, | |
2640 | { BLK_PERM_RESIZE, "resize" }, | |
2641 | { 0, NULL } | |
2642 | }; | |
2643 | ||
2644 | GString *result = g_string_sized_new(30); | |
2645 | struct perm_name *p; | |
2646 | ||
2647 | for (p = permissions; p->name; p++) { | |
2648 | if (perm & p->perm) { | |
2649 | if (result->len > 0) { | |
2650 | g_string_append(result, ", "); | |
2651 | } | |
2652 | g_string_append(result, p->name); | |
2653 | } | |
2654 | } | |
2655 | ||
2656 | return g_string_free(result, FALSE); | |
2657 | } | |
2658 | ||
2659 | ||
2660 | /* | |
2661 | * @tran is allowed to be NULL. In this case no rollback is possible. | |
2662 | * | |
2663 | * After calling this function, the transaction @tran may only be completed | |
2664 | * while holding a reader lock for the graph. | |
2665 | */ | |
2666 | static int GRAPH_RDLOCK | |
2667 | bdrv_refresh_perms(BlockDriverState *bs, Transaction *tran, Error **errp) | |
2668 | { | |
2669 | int ret; | |
2670 | Transaction *local_tran = NULL; | |
2671 | g_autoptr(GSList) list = bdrv_topological_dfs(NULL, NULL, bs); | |
2672 | GLOBAL_STATE_CODE(); | |
2673 | ||
2674 | if (!tran) { | |
2675 | tran = local_tran = tran_new(); | |
2676 | } | |
2677 | ||
2678 | ret = bdrv_do_refresh_perms(list, NULL, tran, errp); | |
2679 | ||
2680 | if (local_tran) { | |
2681 | tran_finalize(local_tran, ret); | |
2682 | } | |
2683 | ||
2684 | return ret; | |
2685 | } | |
2686 | ||
2687 | int bdrv_child_try_set_perm(BdrvChild *c, uint64_t perm, uint64_t shared, | |
2688 | Error **errp) | |
2689 | { | |
2690 | Error *local_err = NULL; | |
2691 | Transaction *tran = tran_new(); | |
2692 | int ret; | |
2693 | ||
2694 | GLOBAL_STATE_CODE(); | |
2695 | ||
2696 | bdrv_child_set_perm(c, perm, shared, tran); | |
2697 | ||
2698 | ret = bdrv_refresh_perms(c->bs, tran, &local_err); | |
2699 | ||
2700 | tran_finalize(tran, ret); | |
2701 | ||
2702 | if (ret < 0) { | |
2703 | if ((perm & ~c->perm) || (c->shared_perm & ~shared)) { | |
2704 | /* tighten permissions */ | |
2705 | error_propagate(errp, local_err); | |
2706 | } else { | |
2707 | /* | |
2708 | * Our caller may intend to only loosen restrictions and | |
2709 | * does not expect this function to fail. Errors are not | |
2710 | * fatal in such a case, so we can just hide them from our | |
2711 | * caller. | |
2712 | */ | |
2713 | error_free(local_err); | |
2714 | ret = 0; | |
2715 | } | |
2716 | } | |
2717 | ||
2718 | return ret; | |
2719 | } | |
2720 | ||
2721 | int bdrv_child_refresh_perms(BlockDriverState *bs, BdrvChild *c, Error **errp) | |
2722 | { | |
2723 | uint64_t parent_perms, parent_shared; | |
2724 | uint64_t perms, shared; | |
2725 | ||
2726 | GLOBAL_STATE_CODE(); | |
2727 | ||
2728 | bdrv_get_cumulative_perm(bs, &parent_perms, &parent_shared); | |
2729 | bdrv_child_perm(bs, c->bs, c, c->role, NULL, | |
2730 | parent_perms, parent_shared, &perms, &shared); | |
2731 | ||
2732 | return bdrv_child_try_set_perm(c, perms, shared, errp); | |
2733 | } | |
2734 | ||
2735 | /* | |
2736 | * Default implementation for .bdrv_child_perm() for block filters: | |
2737 | * Forward CONSISTENT_READ, WRITE, WRITE_UNCHANGED, and RESIZE to the | |
2738 | * filtered child. | |
2739 | */ | |
2740 | static void bdrv_filter_default_perms(BlockDriverState *bs, BdrvChild *c, | |
2741 | BdrvChildRole role, | |
2742 | BlockReopenQueue *reopen_queue, | |
2743 | uint64_t perm, uint64_t shared, | |
2744 | uint64_t *nperm, uint64_t *nshared) | |
2745 | { | |
2746 | GLOBAL_STATE_CODE(); | |
2747 | *nperm = perm & DEFAULT_PERM_PASSTHROUGH; | |
2748 | *nshared = (shared & DEFAULT_PERM_PASSTHROUGH) | DEFAULT_PERM_UNCHANGED; | |
2749 | } | |
2750 | ||
2751 | static void bdrv_default_perms_for_cow(BlockDriverState *bs, BdrvChild *c, | |
2752 | BdrvChildRole role, | |
2753 | BlockReopenQueue *reopen_queue, | |
2754 | uint64_t perm, uint64_t shared, | |
2755 | uint64_t *nperm, uint64_t *nshared) | |
2756 | { | |
2757 | assert(role & BDRV_CHILD_COW); | |
2758 | GLOBAL_STATE_CODE(); | |
2759 | ||
2760 | /* | |
2761 | * We want consistent read from backing files if the parent needs it. | |
2762 | * No other operations are performed on backing files. | |
2763 | */ | |
2764 | perm &= BLK_PERM_CONSISTENT_READ; | |
2765 | ||
2766 | /* | |
2767 | * If the parent can deal with changing data, we're okay with a | |
2768 | * writable and resizable backing file. | |
2769 | * TODO Require !(perm & BLK_PERM_CONSISTENT_READ), too? | |
2770 | */ | |
2771 | if (shared & BLK_PERM_WRITE) { | |
2772 | shared = BLK_PERM_WRITE | BLK_PERM_RESIZE; | |
2773 | } else { | |
2774 | shared = 0; | |
2775 | } | |
2776 | ||
2777 | shared |= BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE_UNCHANGED; | |
2778 | ||
2779 | if (bs->open_flags & BDRV_O_INACTIVE) { | |
2780 | shared |= BLK_PERM_WRITE | BLK_PERM_RESIZE; | |
2781 | } | |
2782 | ||
2783 | *nperm = perm; | |
2784 | *nshared = shared; | |
2785 | } | |
2786 | ||
2787 | static void bdrv_default_perms_for_storage(BlockDriverState *bs, BdrvChild *c, | |
2788 | BdrvChildRole role, | |
2789 | BlockReopenQueue *reopen_queue, | |
2790 | uint64_t perm, uint64_t shared, | |
2791 | uint64_t *nperm, uint64_t *nshared) | |
2792 | { | |
2793 | int flags; | |
2794 | ||
2795 | GLOBAL_STATE_CODE(); | |
2796 | assert(role & (BDRV_CHILD_METADATA | BDRV_CHILD_DATA)); | |
2797 | ||
2798 | flags = bdrv_reopen_get_flags(reopen_queue, bs); | |
2799 | ||
2800 | /* | |
2801 | * Apart from the modifications below, the same permissions are | |
2802 | * forwarded and left alone as for filters | |
2803 | */ | |
2804 | bdrv_filter_default_perms(bs, c, role, reopen_queue, | |
2805 | perm, shared, &perm, &shared); | |
2806 | ||
2807 | if (role & BDRV_CHILD_METADATA) { | |
2808 | /* Format drivers may touch metadata even if the guest doesn't write */ | |
2809 | if (bdrv_is_writable_after_reopen(bs, reopen_queue)) { | |
2810 | perm |= BLK_PERM_WRITE | BLK_PERM_RESIZE; | |
2811 | } | |
2812 | ||
2813 | /* | |
2814 | * bs->file always needs to be consistent because of the | |
2815 | * metadata. We can never allow other users to resize or write | |
2816 | * to it. | |
2817 | */ | |
2818 | if (!(flags & BDRV_O_NO_IO)) { | |
2819 | perm |= BLK_PERM_CONSISTENT_READ; | |
2820 | } | |
2821 | shared &= ~(BLK_PERM_WRITE | BLK_PERM_RESIZE); | |
2822 | } | |
2823 | ||
2824 | if (role & BDRV_CHILD_DATA) { | |
2825 | /* | |
2826 | * Technically, everything in this block is a subset of the | |
2827 | * BDRV_CHILD_METADATA path taken above, and so this could | |
2828 | * be an "else if" branch. However, that is not obvious, and | |
2829 | * this function is not performance critical, therefore we let | |
2830 | * this be an independent "if". | |
2831 | */ | |
2832 | ||
2833 | /* | |
2834 | * We cannot allow other users to resize the file because the | |
2835 | * format driver might have some assumptions about the size | |
2836 | * (e.g. because it is stored in metadata, or because the file | |
2837 | * is split into fixed-size data files). | |
2838 | */ | |
2839 | shared &= ~BLK_PERM_RESIZE; | |
2840 | ||
2841 | /* | |
2842 | * WRITE_UNCHANGED often cannot be performed as such on the | |
2843 | * data file. For example, the qcow2 driver may still need to | |
2844 | * write copied clusters on copy-on-read. | |
2845 | */ | |
2846 | if (perm & BLK_PERM_WRITE_UNCHANGED) { | |
2847 | perm |= BLK_PERM_WRITE; | |
2848 | } | |
2849 | ||
2850 | /* | |
2851 | * If the data file is written to, the format driver may | |
2852 | * expect to be able to resize it by writing beyond the EOF. | |
2853 | */ | |
2854 | if (perm & BLK_PERM_WRITE) { | |
2855 | perm |= BLK_PERM_RESIZE; | |
2856 | } | |
2857 | } | |
2858 | ||
2859 | if (bs->open_flags & BDRV_O_INACTIVE) { | |
2860 | shared |= BLK_PERM_WRITE | BLK_PERM_RESIZE; | |
2861 | } | |
2862 | ||
2863 | *nperm = perm; | |
2864 | *nshared = shared; | |
2865 | } | |
2866 | ||
2867 | void bdrv_default_perms(BlockDriverState *bs, BdrvChild *c, | |
2868 | BdrvChildRole role, BlockReopenQueue *reopen_queue, | |
2869 | uint64_t perm, uint64_t shared, | |
2870 | uint64_t *nperm, uint64_t *nshared) | |
2871 | { | |
2872 | GLOBAL_STATE_CODE(); | |
2873 | if (role & BDRV_CHILD_FILTERED) { | |
2874 | assert(!(role & (BDRV_CHILD_DATA | BDRV_CHILD_METADATA | | |
2875 | BDRV_CHILD_COW))); | |
2876 | bdrv_filter_default_perms(bs, c, role, reopen_queue, | |
2877 | perm, shared, nperm, nshared); | |
2878 | } else if (role & BDRV_CHILD_COW) { | |
2879 | assert(!(role & (BDRV_CHILD_DATA | BDRV_CHILD_METADATA))); | |
2880 | bdrv_default_perms_for_cow(bs, c, role, reopen_queue, | |
2881 | perm, shared, nperm, nshared); | |
2882 | } else if (role & (BDRV_CHILD_METADATA | BDRV_CHILD_DATA)) { | |
2883 | bdrv_default_perms_for_storage(bs, c, role, reopen_queue, | |
2884 | perm, shared, nperm, nshared); | |
2885 | } else { | |
2886 | g_assert_not_reached(); | |
2887 | } | |
2888 | } | |
2889 | ||
2890 | uint64_t bdrv_qapi_perm_to_blk_perm(BlockPermission qapi_perm) | |
2891 | { | |
2892 | static const uint64_t permissions[] = { | |
2893 | [BLOCK_PERMISSION_CONSISTENT_READ] = BLK_PERM_CONSISTENT_READ, | |
2894 | [BLOCK_PERMISSION_WRITE] = BLK_PERM_WRITE, | |
2895 | [BLOCK_PERMISSION_WRITE_UNCHANGED] = BLK_PERM_WRITE_UNCHANGED, | |
2896 | [BLOCK_PERMISSION_RESIZE] = BLK_PERM_RESIZE, | |
2897 | }; | |
2898 | ||
2899 | QEMU_BUILD_BUG_ON(ARRAY_SIZE(permissions) != BLOCK_PERMISSION__MAX); | |
2900 | QEMU_BUILD_BUG_ON(1UL << ARRAY_SIZE(permissions) != BLK_PERM_ALL + 1); | |
2901 | ||
2902 | assert(qapi_perm < BLOCK_PERMISSION__MAX); | |
2903 | ||
2904 | return permissions[qapi_perm]; | |
2905 | } | |
2906 | ||
2907 | /* | |
2908 | * Replaces the node that a BdrvChild points to without updating permissions. | |
2909 | * | |
2910 | * If @new_bs is non-NULL, the parent of @child must already be drained through | |
2911 | * @child and the caller must hold the AioContext lock for @new_bs. | |
2912 | */ | |
2913 | static void GRAPH_WRLOCK | |
2914 | bdrv_replace_child_noperm(BdrvChild *child, BlockDriverState *new_bs) | |
2915 | { | |
2916 | BlockDriverState *old_bs = child->bs; | |
2917 | int new_bs_quiesce_counter; | |
2918 | ||
2919 | assert(!child->frozen); | |
2920 | ||
2921 | /* | |
2922 | * If we want to change the BdrvChild to point to a drained node as its new | |
2923 | * child->bs, we need to make sure that its new parent is drained, too. In | |
2924 | * other words, either child->quiesce_parent must already be true or we must | |
2925 | * be able to set it and keep the parent's quiesce_counter consistent with | |
2926 | * that, but without polling or starting new requests (this function | |
2927 | * guarantees that it doesn't poll, and starting new requests would be | |
2928 | * against the invariants of drain sections). | |
2929 | * | |
2930 | * To keep things simple, we pick the first option (child->quiesce_parent | |
2931 | * must already be true). We also generalise the rule a bit to make it | |
2932 | * easier to verify in callers and more likely to be covered in test cases: | |
2933 | * The parent must be quiesced through this child even if new_bs isn't | |
2934 | * currently drained. | |
2935 | * | |
2936 | * The only exception is for callers that always pass new_bs == NULL. In | |
2937 | * this case, we obviously never need to consider the case of a drained | |
2938 | * new_bs, so we can keep the callers simpler by allowing them not to drain | |
2939 | * the parent. | |
2940 | */ | |
2941 | assert(!new_bs || child->quiesced_parent); | |
2942 | assert(old_bs != new_bs); | |
2943 | GLOBAL_STATE_CODE(); | |
2944 | ||
2945 | if (old_bs && new_bs) { | |
2946 | assert(bdrv_get_aio_context(old_bs) == bdrv_get_aio_context(new_bs)); | |
2947 | } | |
2948 | ||
2949 | if (old_bs) { | |
2950 | if (child->klass->detach) { | |
2951 | child->klass->detach(child); | |
2952 | } | |
2953 | QLIST_REMOVE(child, next_parent); | |
2954 | } | |
2955 | ||
2956 | child->bs = new_bs; | |
2957 | ||
2958 | if (new_bs) { | |
2959 | QLIST_INSERT_HEAD(&new_bs->parents, child, next_parent); | |
2960 | if (child->klass->attach) { | |
2961 | child->klass->attach(child); | |
2962 | } | |
2963 | } | |
2964 | ||
2965 | /* | |
2966 | * If the parent was drained through this BdrvChild previously, but new_bs | |
2967 | * is not drained, allow requests to come in only after the new node has | |
2968 | * been attached. | |
2969 | */ | |
2970 | new_bs_quiesce_counter = (new_bs ? new_bs->quiesce_counter : 0); | |
2971 | if (!new_bs_quiesce_counter && child->quiesced_parent) { | |
2972 | bdrv_parent_drained_end_single(child); | |
2973 | } | |
2974 | } | |
2975 | ||
2976 | /** | |
2977 | * Free the given @child. | |
2978 | * | |
2979 | * The child must be empty (i.e. `child->bs == NULL`) and it must be | |
2980 | * unused (i.e. not in a children list). | |
2981 | */ | |
2982 | static void bdrv_child_free(BdrvChild *child) | |
2983 | { | |
2984 | assert(!child->bs); | |
2985 | GLOBAL_STATE_CODE(); | |
2986 | GRAPH_RDLOCK_GUARD_MAINLOOP(); | |
2987 | ||
2988 | assert(!child->next.le_prev); /* not in children list */ | |
2989 | ||
2990 | g_free(child->name); | |
2991 | g_free(child); | |
2992 | } | |
2993 | ||
2994 | typedef struct BdrvAttachChildCommonState { | |
2995 | BdrvChild *child; | |
2996 | AioContext *old_parent_ctx; | |
2997 | AioContext *old_child_ctx; | |
2998 | } BdrvAttachChildCommonState; | |
2999 | ||
3000 | static void GRAPH_WRLOCK bdrv_attach_child_common_abort(void *opaque) | |
3001 | { | |
3002 | BdrvAttachChildCommonState *s = opaque; | |
3003 | BlockDriverState *bs = s->child->bs; | |
3004 | ||
3005 | GLOBAL_STATE_CODE(); | |
3006 | assert_bdrv_graph_writable(); | |
3007 | ||
3008 | bdrv_replace_child_noperm(s->child, NULL); | |
3009 | ||
3010 | if (bdrv_get_aio_context(bs) != s->old_child_ctx) { | |
3011 | bdrv_try_change_aio_context(bs, s->old_child_ctx, NULL, &error_abort); | |
3012 | } | |
3013 | ||
3014 | if (bdrv_child_get_parent_aio_context(s->child) != s->old_parent_ctx) { | |
3015 | Transaction *tran; | |
3016 | GHashTable *visited; | |
3017 | bool ret; | |
3018 | ||
3019 | tran = tran_new(); | |
3020 | ||
3021 | /* No need to visit `child`, because it has been detached already */ | |
3022 | visited = g_hash_table_new(NULL, NULL); | |
3023 | ret = s->child->klass->change_aio_ctx(s->child, s->old_parent_ctx, | |
3024 | visited, tran, &error_abort); | |
3025 | g_hash_table_destroy(visited); | |
3026 | ||
3027 | /* transaction is supposed to always succeed */ | |
3028 | assert(ret == true); | |
3029 | tran_commit(tran); | |
3030 | } | |
3031 | ||
3032 | bdrv_schedule_unref(bs); | |
3033 | bdrv_child_free(s->child); | |
3034 | } | |
3035 | ||
3036 | static TransactionActionDrv bdrv_attach_child_common_drv = { | |
3037 | .abort = bdrv_attach_child_common_abort, | |
3038 | .clean = g_free, | |
3039 | }; | |
3040 | ||
3041 | /* | |
3042 | * Common part of attaching bdrv child to bs or to blk or to job | |
3043 | * | |
3044 | * Function doesn't update permissions, caller is responsible for this. | |
3045 | * | |
3046 | * After calling this function, the transaction @tran may only be completed | |
3047 | * while holding a writer lock for the graph. | |
3048 | * | |
3049 | * Returns new created child. | |
3050 | * | |
3051 | * The caller must hold the AioContext lock for @child_bs. Both @parent_bs and | |
3052 | * @child_bs can move to a different AioContext in this function. Callers must | |
3053 | * make sure that their AioContext locking is still correct after this. | |
3054 | */ | |
3055 | static BdrvChild * GRAPH_WRLOCK | |
3056 | bdrv_attach_child_common(BlockDriverState *child_bs, | |
3057 | const char *child_name, | |
3058 | const BdrvChildClass *child_class, | |
3059 | BdrvChildRole child_role, | |
3060 | uint64_t perm, uint64_t shared_perm, | |
3061 | void *opaque, | |
3062 | Transaction *tran, Error **errp) | |
3063 | { | |
3064 | BdrvChild *new_child; | |
3065 | AioContext *parent_ctx, *new_child_ctx; | |
3066 | AioContext *child_ctx = bdrv_get_aio_context(child_bs); | |
3067 | ||
3068 | assert(child_class->get_parent_desc); | |
3069 | GLOBAL_STATE_CODE(); | |
3070 | ||
3071 | new_child = g_new(BdrvChild, 1); | |
3072 | *new_child = (BdrvChild) { | |
3073 | .bs = NULL, | |
3074 | .name = g_strdup(child_name), | |
3075 | .klass = child_class, | |
3076 | .role = child_role, | |
3077 | .perm = perm, | |
3078 | .shared_perm = shared_perm, | |
3079 | .opaque = opaque, | |
3080 | }; | |
3081 | ||
3082 | /* | |
3083 | * If the AioContexts don't match, first try to move the subtree of | |
3084 | * child_bs into the AioContext of the new parent. If this doesn't work, | |
3085 | * try moving the parent into the AioContext of child_bs instead. | |
3086 | */ | |
3087 | parent_ctx = bdrv_child_get_parent_aio_context(new_child); | |
3088 | if (child_ctx != parent_ctx) { | |
3089 | Error *local_err = NULL; | |
3090 | int ret = bdrv_try_change_aio_context(child_bs, parent_ctx, NULL, | |
3091 | &local_err); | |
3092 | ||
3093 | if (ret < 0 && child_class->change_aio_ctx) { | |
3094 | Transaction *aio_ctx_tran = tran_new(); | |
3095 | GHashTable *visited = g_hash_table_new(NULL, NULL); | |
3096 | bool ret_child; | |
3097 | ||
3098 | g_hash_table_add(visited, new_child); | |
3099 | ret_child = child_class->change_aio_ctx(new_child, child_ctx, | |
3100 | visited, aio_ctx_tran, | |
3101 | NULL); | |
3102 | if (ret_child == true) { | |
3103 | error_free(local_err); | |
3104 | ret = 0; | |
3105 | } | |
3106 | tran_finalize(aio_ctx_tran, ret_child == true ? 0 : -1); | |
3107 | g_hash_table_destroy(visited); | |
3108 | } | |
3109 | ||
3110 | if (ret < 0) { | |
3111 | error_propagate(errp, local_err); | |
3112 | bdrv_child_free(new_child); | |
3113 | return NULL; | |
3114 | } | |
3115 | } | |
3116 | ||
3117 | new_child_ctx = bdrv_get_aio_context(child_bs); | |
3118 | if (new_child_ctx != child_ctx) { | |
3119 | aio_context_release(child_ctx); | |
3120 | aio_context_acquire(new_child_ctx); | |
3121 | } | |
3122 | ||
3123 | bdrv_ref(child_bs); | |
3124 | /* | |
3125 | * Let every new BdrvChild start with a drained parent. Inserting the child | |
3126 | * in the graph with bdrv_replace_child_noperm() will undrain it if | |
3127 | * @child_bs is not drained. | |
3128 | * | |
3129 | * The child was only just created and is not yet visible in global state | |
3130 | * until bdrv_replace_child_noperm() inserts it into the graph, so nobody | |
3131 | * could have sent requests and polling is not necessary. | |
3132 | * | |
3133 | * Note that this means that the parent isn't fully drained yet, we only | |
3134 | * stop new requests from coming in. This is fine, we don't care about the | |
3135 | * old requests here, they are not for this child. If another place enters a | |
3136 | * drain section for the same parent, but wants it to be fully quiesced, it | |
3137 | * will not run most of the the code in .drained_begin() again (which is not | |
3138 | * a problem, we already did this), but it will still poll until the parent | |
3139 | * is fully quiesced, so it will not be negatively affected either. | |
3140 | */ | |
3141 | bdrv_parent_drained_begin_single(new_child); | |
3142 | bdrv_replace_child_noperm(new_child, child_bs); | |
3143 | ||
3144 | BdrvAttachChildCommonState *s = g_new(BdrvAttachChildCommonState, 1); | |
3145 | *s = (BdrvAttachChildCommonState) { | |
3146 | .child = new_child, | |
3147 | .old_parent_ctx = parent_ctx, | |
3148 | .old_child_ctx = child_ctx, | |
3149 | }; | |
3150 | tran_add(tran, &bdrv_attach_child_common_drv, s); | |
3151 | ||
3152 | if (new_child_ctx != child_ctx) { | |
3153 | aio_context_release(new_child_ctx); | |
3154 | aio_context_acquire(child_ctx); | |
3155 | } | |
3156 | ||
3157 | return new_child; | |
3158 | } | |
3159 | ||
3160 | /* | |
3161 | * Function doesn't update permissions, caller is responsible for this. | |
3162 | * | |
3163 | * The caller must hold the AioContext lock for @child_bs. Both @parent_bs and | |
3164 | * @child_bs can move to a different AioContext in this function. Callers must | |
3165 | * make sure that their AioContext locking is still correct after this. | |
3166 | * | |
3167 | * After calling this function, the transaction @tran may only be completed | |
3168 | * while holding a writer lock for the graph. | |
3169 | */ | |
3170 | static BdrvChild * GRAPH_WRLOCK | |
3171 | bdrv_attach_child_noperm(BlockDriverState *parent_bs, | |
3172 | BlockDriverState *child_bs, | |
3173 | const char *child_name, | |
3174 | const BdrvChildClass *child_class, | |
3175 | BdrvChildRole child_role, | |
3176 | Transaction *tran, | |
3177 | Error **errp) | |
3178 | { | |
3179 | uint64_t perm, shared_perm; | |
3180 | ||
3181 | assert(parent_bs->drv); | |
3182 | GLOBAL_STATE_CODE(); | |
3183 | ||
3184 | if (bdrv_recurse_has_child(child_bs, parent_bs)) { | |
3185 | error_setg(errp, "Making '%s' a %s child of '%s' would create a cycle", | |
3186 | child_bs->node_name, child_name, parent_bs->node_name); | |
3187 | return NULL; | |
3188 | } | |
3189 | ||
3190 | bdrv_get_cumulative_perm(parent_bs, &perm, &shared_perm); | |
3191 | bdrv_child_perm(parent_bs, child_bs, NULL, child_role, NULL, | |
3192 | perm, shared_perm, &perm, &shared_perm); | |
3193 | ||
3194 | return bdrv_attach_child_common(child_bs, child_name, child_class, | |
3195 | child_role, perm, shared_perm, parent_bs, | |
3196 | tran, errp); | |
3197 | } | |
3198 | ||
3199 | /* | |
3200 | * This function steals the reference to child_bs from the caller. | |
3201 | * That reference is later dropped by bdrv_root_unref_child(). | |
3202 | * | |
3203 | * On failure NULL is returned, errp is set and the reference to | |
3204 | * child_bs is also dropped. | |
3205 | * | |
3206 | * The caller must hold the AioContext lock @child_bs, but not that of @ctx | |
3207 | * (unless @child_bs is already in @ctx). | |
3208 | */ | |
3209 | BdrvChild *bdrv_root_attach_child(BlockDriverState *child_bs, | |
3210 | const char *child_name, | |
3211 | const BdrvChildClass *child_class, | |
3212 | BdrvChildRole child_role, | |
3213 | uint64_t perm, uint64_t shared_perm, | |
3214 | void *opaque, Error **errp) | |
3215 | { | |
3216 | int ret; | |
3217 | BdrvChild *child; | |
3218 | Transaction *tran = tran_new(); | |
3219 | ||
3220 | GLOBAL_STATE_CODE(); | |
3221 | ||
3222 | child = bdrv_attach_child_common(child_bs, child_name, child_class, | |
3223 | child_role, perm, shared_perm, opaque, | |
3224 | tran, errp); | |
3225 | if (!child) { | |
3226 | ret = -EINVAL; | |
3227 | goto out; | |
3228 | } | |
3229 | ||
3230 | ret = bdrv_refresh_perms(child_bs, tran, errp); | |
3231 | ||
3232 | out: | |
3233 | tran_finalize(tran, ret); | |
3234 | ||
3235 | bdrv_schedule_unref(child_bs); | |
3236 | ||
3237 | return ret < 0 ? NULL : child; | |
3238 | } | |
3239 | ||
3240 | /* | |
3241 | * This function transfers the reference to child_bs from the caller | |
3242 | * to parent_bs. That reference is later dropped by parent_bs on | |
3243 | * bdrv_close() or if someone calls bdrv_unref_child(). | |
3244 | * | |
3245 | * On failure NULL is returned, errp is set and the reference to | |
3246 | * child_bs is also dropped. | |
3247 | * | |
3248 | * If @parent_bs and @child_bs are in different AioContexts, the caller must | |
3249 | * hold the AioContext lock for @child_bs, but not for @parent_bs. | |
3250 | */ | |
3251 | BdrvChild *bdrv_attach_child(BlockDriverState *parent_bs, | |
3252 | BlockDriverState *child_bs, | |
3253 | const char *child_name, | |
3254 | const BdrvChildClass *child_class, | |
3255 | BdrvChildRole child_role, | |
3256 | Error **errp) | |
3257 | { | |
3258 | int ret; | |
3259 | BdrvChild *child; | |
3260 | Transaction *tran = tran_new(); | |
3261 | ||
3262 | GLOBAL_STATE_CODE(); | |
3263 | ||
3264 | child = bdrv_attach_child_noperm(parent_bs, child_bs, child_name, | |
3265 | child_class, child_role, tran, errp); | |
3266 | if (!child) { | |
3267 | ret = -EINVAL; | |
3268 | goto out; | |
3269 | } | |
3270 | ||
3271 | ret = bdrv_refresh_perms(parent_bs, tran, errp); | |
3272 | if (ret < 0) { | |
3273 | goto out; | |
3274 | } | |
3275 | ||
3276 | out: | |
3277 | tran_finalize(tran, ret); | |
3278 | ||
3279 | bdrv_schedule_unref(child_bs); | |
3280 | ||
3281 | return ret < 0 ? NULL : child; | |
3282 | } | |
3283 | ||
3284 | /* Callers must ensure that child->frozen is false. */ | |
3285 | void bdrv_root_unref_child(BdrvChild *child) | |
3286 | { | |
3287 | BlockDriverState *child_bs = child->bs; | |
3288 | ||
3289 | GLOBAL_STATE_CODE(); | |
3290 | bdrv_replace_child_noperm(child, NULL); | |
3291 | bdrv_child_free(child); | |
3292 | ||
3293 | if (child_bs) { | |
3294 | /* | |
3295 | * Update permissions for old node. We're just taking a parent away, so | |
3296 | * we're loosening restrictions. Errors of permission update are not | |
3297 | * fatal in this case, ignore them. | |
3298 | */ | |
3299 | bdrv_refresh_perms(child_bs, NULL, NULL); | |
3300 | ||
3301 | /* | |
3302 | * When the parent requiring a non-default AioContext is removed, the | |
3303 | * node moves back to the main AioContext | |
3304 | */ | |
3305 | bdrv_try_change_aio_context(child_bs, qemu_get_aio_context(), NULL, | |
3306 | NULL); | |
3307 | } | |
3308 | ||
3309 | bdrv_schedule_unref(child_bs); | |
3310 | } | |
3311 | ||
3312 | typedef struct BdrvSetInheritsFrom { | |
3313 | BlockDriverState *bs; | |
3314 | BlockDriverState *old_inherits_from; | |
3315 | } BdrvSetInheritsFrom; | |
3316 | ||
3317 | static void bdrv_set_inherits_from_abort(void *opaque) | |
3318 | { | |
3319 | BdrvSetInheritsFrom *s = opaque; | |
3320 | ||
3321 | s->bs->inherits_from = s->old_inherits_from; | |
3322 | } | |
3323 | ||
3324 | static TransactionActionDrv bdrv_set_inherits_from_drv = { | |
3325 | .abort = bdrv_set_inherits_from_abort, | |
3326 | .clean = g_free, | |
3327 | }; | |
3328 | ||
3329 | /* @tran is allowed to be NULL. In this case no rollback is possible */ | |
3330 | static void bdrv_set_inherits_from(BlockDriverState *bs, | |
3331 | BlockDriverState *new_inherits_from, | |
3332 | Transaction *tran) | |
3333 | { | |
3334 | if (tran) { | |
3335 | BdrvSetInheritsFrom *s = g_new(BdrvSetInheritsFrom, 1); | |
3336 | ||
3337 | *s = (BdrvSetInheritsFrom) { | |
3338 | .bs = bs, | |
3339 | .old_inherits_from = bs->inherits_from, | |
3340 | }; | |
3341 | ||
3342 | tran_add(tran, &bdrv_set_inherits_from_drv, s); | |
3343 | } | |
3344 | ||
3345 | bs->inherits_from = new_inherits_from; | |
3346 | } | |
3347 | ||
3348 | /** | |
3349 | * Clear all inherits_from pointers from children and grandchildren of | |
3350 | * @root that point to @root, where necessary. | |
3351 | * @tran is allowed to be NULL. In this case no rollback is possible | |
3352 | */ | |
3353 | static void GRAPH_WRLOCK | |
3354 | bdrv_unset_inherits_from(BlockDriverState *root, BdrvChild *child, | |
3355 | Transaction *tran) | |
3356 | { | |
3357 | BdrvChild *c; | |
3358 | ||
3359 | if (child->bs->inherits_from == root) { | |
3360 | /* | |
3361 | * Remove inherits_from only when the last reference between root and | |
3362 | * child->bs goes away. | |
3363 | */ | |
3364 | QLIST_FOREACH(c, &root->children, next) { | |
3365 | if (c != child && c->bs == child->bs) { | |
3366 | break; | |
3367 | } | |
3368 | } | |
3369 | if (c == NULL) { | |
3370 | bdrv_set_inherits_from(child->bs, NULL, tran); | |
3371 | } | |
3372 | } | |
3373 | ||
3374 | QLIST_FOREACH(c, &child->bs->children, next) { | |
3375 | bdrv_unset_inherits_from(root, c, tran); | |
3376 | } | |
3377 | } | |
3378 | ||
3379 | /* Callers must ensure that child->frozen is false. */ | |
3380 | void bdrv_unref_child(BlockDriverState *parent, BdrvChild *child) | |
3381 | { | |
3382 | GLOBAL_STATE_CODE(); | |
3383 | if (child == NULL) { | |
3384 | return; | |
3385 | } | |
3386 | ||
3387 | bdrv_unset_inherits_from(parent, child, NULL); | |
3388 | bdrv_root_unref_child(child); | |
3389 | } | |
3390 | ||
3391 | ||
3392 | static void GRAPH_RDLOCK | |
3393 | bdrv_parent_cb_change_media(BlockDriverState *bs, bool load) | |
3394 | { | |
3395 | BdrvChild *c; | |
3396 | GLOBAL_STATE_CODE(); | |
3397 | QLIST_FOREACH(c, &bs->parents, next_parent) { | |
3398 | if (c->klass->change_media) { | |
3399 | c->klass->change_media(c, load); | |
3400 | } | |
3401 | } | |
3402 | } | |
3403 | ||
3404 | /* Return true if you can reach parent going through child->inherits_from | |
3405 | * recursively. If parent or child are NULL, return false */ | |
3406 | static bool bdrv_inherits_from_recursive(BlockDriverState *child, | |
3407 | BlockDriverState *parent) | |
3408 | { | |
3409 | while (child && child != parent) { | |
3410 | child = child->inherits_from; | |
3411 | } | |
3412 | ||
3413 | return child != NULL; | |
3414 | } | |
3415 | ||
3416 | /* | |
3417 | * Return the BdrvChildRole for @bs's backing child. bs->backing is | |
3418 | * mostly used for COW backing children (role = COW), but also for | |
3419 | * filtered children (role = FILTERED | PRIMARY). | |
3420 | */ | |
3421 | static BdrvChildRole bdrv_backing_role(BlockDriverState *bs) | |
3422 | { | |
3423 | if (bs->drv && bs->drv->is_filter) { | |
3424 | return BDRV_CHILD_FILTERED | BDRV_CHILD_PRIMARY; | |
3425 | } else { | |
3426 | return BDRV_CHILD_COW; | |
3427 | } | |
3428 | } | |
3429 | ||
3430 | /* | |
3431 | * Sets the bs->backing or bs->file link of a BDS. A new reference is created; | |
3432 | * callers which don't need their own reference any more must call bdrv_unref(). | |
3433 | * | |
3434 | * If the respective child is already present (i.e. we're detaching a node), | |
3435 | * that child node must be drained. | |
3436 | * | |
3437 | * Function doesn't update permissions, caller is responsible for this. | |
3438 | * | |
3439 | * The caller must hold the AioContext lock for @child_bs. Both @parent_bs and | |
3440 | * @child_bs can move to a different AioContext in this function. Callers must | |
3441 | * make sure that their AioContext locking is still correct after this. | |
3442 | * | |
3443 | * After calling this function, the transaction @tran may only be completed | |
3444 | * while holding a writer lock for the graph. | |
3445 | */ | |
3446 | static int GRAPH_WRLOCK | |
3447 | bdrv_set_file_or_backing_noperm(BlockDriverState *parent_bs, | |
3448 | BlockDriverState *child_bs, | |
3449 | bool is_backing, | |
3450 | Transaction *tran, Error **errp) | |
3451 | { | |
3452 | bool update_inherits_from = | |
3453 | bdrv_inherits_from_recursive(child_bs, parent_bs); | |
3454 | BdrvChild *child = is_backing ? parent_bs->backing : parent_bs->file; | |
3455 | BdrvChildRole role; | |
3456 | ||
3457 | GLOBAL_STATE_CODE(); | |
3458 | ||
3459 | if (!parent_bs->drv) { | |
3460 | /* | |
3461 | * Node without drv is an object without a class :/. TODO: finally fix | |
3462 | * qcow2 driver to never clear bs->drv and implement format corruption | |
3463 | * handling in other way. | |
3464 | */ | |
3465 | error_setg(errp, "Node corrupted"); | |
3466 | return -EINVAL; | |
3467 | } | |
3468 | ||
3469 | if (child && child->frozen) { | |
3470 | error_setg(errp, "Cannot change frozen '%s' link from '%s' to '%s'", | |
3471 | child->name, parent_bs->node_name, child->bs->node_name); | |
3472 | return -EPERM; | |
3473 | } | |
3474 | ||
3475 | if (is_backing && !parent_bs->drv->is_filter && | |
3476 | !parent_bs->drv->supports_backing) | |
3477 | { | |
3478 | error_setg(errp, "Driver '%s' of node '%s' does not support backing " | |
3479 | "files", parent_bs->drv->format_name, parent_bs->node_name); | |
3480 | return -EINVAL; | |
3481 | } | |
3482 | ||
3483 | if (parent_bs->drv->is_filter) { | |
3484 | role = BDRV_CHILD_FILTERED | BDRV_CHILD_PRIMARY; | |
3485 | } else if (is_backing) { | |
3486 | role = BDRV_CHILD_COW; | |
3487 | } else { | |
3488 | /* | |
3489 | * We only can use same role as it is in existing child. We don't have | |
3490 | * infrastructure to determine role of file child in generic way | |
3491 | */ | |
3492 | if (!child) { | |
3493 | error_setg(errp, "Cannot set file child to format node without " | |
3494 | "file child"); | |
3495 | return -EINVAL; | |
3496 | } | |
3497 | role = child->role; | |
3498 | } | |
3499 | ||
3500 | if (child) { | |
3501 | assert(child->bs->quiesce_counter); | |
3502 | bdrv_unset_inherits_from(parent_bs, child, tran); | |
3503 | bdrv_remove_child(child, tran); | |
3504 | } | |
3505 | ||
3506 | if (!child_bs) { | |
3507 | goto out; | |
3508 | } | |
3509 | ||
3510 | child = bdrv_attach_child_noperm(parent_bs, child_bs, | |
3511 | is_backing ? "backing" : "file", | |
3512 | &child_of_bds, role, | |
3513 | tran, errp); | |
3514 | if (!child) { | |
3515 | return -EINVAL; | |
3516 | } | |
3517 | ||
3518 | ||
3519 | /* | |
3520 | * If inherits_from pointed recursively to bs then let's update it to | |
3521 | * point directly to bs (else it will become NULL). | |
3522 | */ | |
3523 | if (update_inherits_from) { | |
3524 | bdrv_set_inherits_from(child_bs, parent_bs, tran); | |
3525 | } | |
3526 | ||
3527 | out: | |
3528 | bdrv_refresh_limits(parent_bs, tran, NULL); | |
3529 | ||
3530 | return 0; | |
3531 | } | |
3532 | ||
3533 | /* | |
3534 | * The caller must hold the AioContext lock for @backing_hd. Both @bs and | |
3535 | * @backing_hd can move to a different AioContext in this function. Callers must | |
3536 | * make sure that their AioContext locking is still correct after this. | |
3537 | * | |
3538 | * If a backing child is already present (i.e. we're detaching a node), that | |
3539 | * child node must be drained. | |
3540 | */ | |
3541 | int bdrv_set_backing_hd_drained(BlockDriverState *bs, | |
3542 | BlockDriverState *backing_hd, | |
3543 | Error **errp) | |
3544 | { | |
3545 | int ret; | |
3546 | Transaction *tran = tran_new(); | |
3547 | ||
3548 | GLOBAL_STATE_CODE(); | |
3549 | assert(bs->quiesce_counter > 0); | |
3550 | if (bs->backing) { | |
3551 | assert(bs->backing->bs->quiesce_counter > 0); | |
3552 | } | |
3553 | ||
3554 | ret = bdrv_set_file_or_backing_noperm(bs, backing_hd, true, tran, errp); | |
3555 | if (ret < 0) { | |
3556 | goto out; | |
3557 | } | |
3558 | ||
3559 | ret = bdrv_refresh_perms(bs, tran, errp); | |
3560 | out: | |
3561 | tran_finalize(tran, ret); | |
3562 | return ret; | |
3563 | } | |
3564 | ||
3565 | int bdrv_set_backing_hd(BlockDriverState *bs, BlockDriverState *backing_hd, | |
3566 | Error **errp) | |
3567 | { | |
3568 | BlockDriverState *drain_bs; | |
3569 | int ret; | |
3570 | GLOBAL_STATE_CODE(); | |
3571 | ||
3572 | bdrv_graph_rdlock_main_loop(); | |
3573 | drain_bs = bs->backing ? bs->backing->bs : bs; | |
3574 | bdrv_graph_rdunlock_main_loop(); | |
3575 | ||
3576 | bdrv_ref(drain_bs); | |
3577 | bdrv_drained_begin(drain_bs); | |
3578 | bdrv_graph_wrlock(backing_hd); | |
3579 | ret = bdrv_set_backing_hd_drained(bs, backing_hd, errp); | |
3580 | bdrv_graph_wrunlock(); | |
3581 | bdrv_drained_end(drain_bs); | |
3582 | bdrv_unref(drain_bs); | |
3583 | ||
3584 | return ret; | |
3585 | } | |
3586 | ||
3587 | /* | |
3588 | * Opens the backing file for a BlockDriverState if not yet open | |
3589 | * | |
3590 | * bdref_key specifies the key for the image's BlockdevRef in the options QDict. | |
3591 | * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict | |
3592 | * itself, all options starting with "${bdref_key}." are considered part of the | |
3593 | * BlockdevRef. | |
3594 | * | |
3595 | * The caller must hold the main AioContext lock. | |
3596 | * | |
3597 | * TODO Can this be unified with bdrv_open_image()? | |
3598 | */ | |
3599 | int bdrv_open_backing_file(BlockDriverState *bs, QDict *parent_options, | |
3600 | const char *bdref_key, Error **errp) | |
3601 | { | |
3602 | char *backing_filename = NULL; | |
3603 | char *bdref_key_dot; | |
3604 | const char *reference = NULL; | |
3605 | int ret = 0; | |
3606 | bool implicit_backing = false; | |
3607 | BlockDriverState *backing_hd; | |
3608 | AioContext *backing_hd_ctx; | |
3609 | QDict *options; | |
3610 | QDict *tmp_parent_options = NULL; | |
3611 | Error *local_err = NULL; | |
3612 | ||
3613 | GLOBAL_STATE_CODE(); | |
3614 | GRAPH_RDLOCK_GUARD_MAINLOOP(); | |
3615 | ||
3616 | if (bs->backing != NULL) { | |
3617 | goto free_exit; | |
3618 | } | |
3619 | ||
3620 | /* NULL means an empty set of options */ | |
3621 | if (parent_options == NULL) { | |
3622 | tmp_parent_options = qdict_new(); | |
3623 | parent_options = tmp_parent_options; | |
3624 | } | |
3625 | ||
3626 | bs->open_flags &= ~BDRV_O_NO_BACKING; | |
3627 | ||
3628 | bdref_key_dot = g_strdup_printf("%s.", bdref_key); | |
3629 | qdict_extract_subqdict(parent_options, &options, bdref_key_dot); | |
3630 | g_free(bdref_key_dot); | |
3631 | ||
3632 | /* | |
3633 | * Caution: while qdict_get_try_str() is fine, getting non-string | |
3634 | * types would require more care. When @parent_options come from | |
3635 | * -blockdev or blockdev_add, its members are typed according to | |
3636 | * the QAPI schema, but when they come from -drive, they're all | |
3637 | * QString. | |
3638 | */ | |
3639 | reference = qdict_get_try_str(parent_options, bdref_key); | |
3640 | if (reference || qdict_haskey(options, "file.filename")) { | |
3641 | /* keep backing_filename NULL */ | |
3642 | } else if (bs->backing_file[0] == '\0' && qdict_size(options) == 0) { | |
3643 | qobject_unref(options); | |
3644 | goto free_exit; | |
3645 | } else { | |
3646 | if (qdict_size(options) == 0) { | |
3647 | /* If the user specifies options that do not modify the | |
3648 | * backing file's behavior, we might still consider it the | |
3649 | * implicit backing file. But it's easier this way, and | |
3650 | * just specifying some of the backing BDS's options is | |
3651 | * only possible with -drive anyway (otherwise the QAPI | |
3652 | * schema forces the user to specify everything). */ | |
3653 | implicit_backing = !strcmp(bs->auto_backing_file, bs->backing_file); | |
3654 | } | |
3655 | ||
3656 | backing_filename = bdrv_get_full_backing_filename(bs, &local_err); | |
3657 | if (local_err) { | |
3658 | ret = -EINVAL; | |
3659 | error_propagate(errp, local_err); | |
3660 | qobject_unref(options); | |
3661 | goto free_exit; | |
3662 | } | |
3663 | } | |
3664 | ||
3665 | if (!bs->drv || !bs->drv->supports_backing) { | |
3666 | ret = -EINVAL; | |
3667 | error_setg(errp, "Driver doesn't support backing files"); | |
3668 | qobject_unref(options); | |
3669 | goto free_exit; | |
3670 | } | |
3671 | ||
3672 | if (!reference && | |
3673 | bs->backing_format[0] != '\0' && !qdict_haskey(options, "driver")) { | |
3674 | qdict_put_str(options, "driver", bs->backing_format); | |
3675 | } | |
3676 | ||
3677 | backing_hd = bdrv_open_inherit(backing_filename, reference, options, 0, bs, | |
3678 | &child_of_bds, bdrv_backing_role(bs), errp); | |
3679 | if (!backing_hd) { | |
3680 | bs->open_flags |= BDRV_O_NO_BACKING; | |
3681 | error_prepend(errp, "Could not open backing file: "); | |
3682 | ret = -EINVAL; | |
3683 | goto free_exit; | |
3684 | } | |
3685 | ||
3686 | if (implicit_backing) { | |
3687 | bdrv_refresh_filename(backing_hd); | |
3688 | pstrcpy(bs->auto_backing_file, sizeof(bs->auto_backing_file), | |
3689 | backing_hd->filename); | |
3690 | } | |
3691 | ||
3692 | /* Hook up the backing file link; drop our reference, bs owns the | |
3693 | * backing_hd reference now */ | |
3694 | backing_hd_ctx = bdrv_get_aio_context(backing_hd); | |
3695 | aio_context_acquire(backing_hd_ctx); | |
3696 | ret = bdrv_set_backing_hd(bs, backing_hd, errp); | |
3697 | bdrv_unref(backing_hd); | |
3698 | aio_context_release(backing_hd_ctx); | |
3699 | ||
3700 | if (ret < 0) { | |
3701 | goto free_exit; | |
3702 | } | |
3703 | ||
3704 | qdict_del(parent_options, bdref_key); | |
3705 | ||
3706 | free_exit: | |
3707 | g_free(backing_filename); | |
3708 | qobject_unref(tmp_parent_options); | |
3709 | return ret; | |
3710 | } | |
3711 | ||
3712 | static BlockDriverState * | |
3713 | bdrv_open_child_bs(const char *filename, QDict *options, const char *bdref_key, | |
3714 | BlockDriverState *parent, const BdrvChildClass *child_class, | |
3715 | BdrvChildRole child_role, bool allow_none, Error **errp) | |
3716 | { | |
3717 | BlockDriverState *bs = NULL; | |
3718 | QDict *image_options; | |
3719 | char *bdref_key_dot; | |
3720 | const char *reference; | |
3721 | ||
3722 | assert(child_class != NULL); | |
3723 | ||
3724 | bdref_key_dot = g_strdup_printf("%s.", bdref_key); | |
3725 | qdict_extract_subqdict(options, &image_options, bdref_key_dot); | |
3726 | g_free(bdref_key_dot); | |
3727 | ||
3728 | /* | |
3729 | * Caution: while qdict_get_try_str() is fine, getting non-string | |
3730 | * types would require more care. When @options come from | |
3731 | * -blockdev or blockdev_add, its members are typed according to | |
3732 | * the QAPI schema, but when they come from -drive, they're all | |
3733 | * QString. | |
3734 | */ | |
3735 | reference = qdict_get_try_str(options, bdref_key); | |
3736 | if (!filename && !reference && !qdict_size(image_options)) { | |
3737 | if (!allow_none) { | |
3738 | error_setg(errp, "A block device must be specified for \"%s\"", | |
3739 | bdref_key); | |
3740 | } | |
3741 | qobject_unref(image_options); | |
3742 | goto done; | |
3743 | } | |
3744 | ||
3745 | bs = bdrv_open_inherit(filename, reference, image_options, 0, | |
3746 | parent, child_class, child_role, errp); | |
3747 | if (!bs) { | |
3748 | goto done; | |
3749 | } | |
3750 | ||
3751 | done: | |
3752 | qdict_del(options, bdref_key); | |
3753 | return bs; | |
3754 | } | |
3755 | ||
3756 | /* | |
3757 | * Opens a disk image whose options are given as BlockdevRef in another block | |
3758 | * device's options. | |
3759 | * | |
3760 | * If allow_none is true, no image will be opened if filename is false and no | |
3761 | * BlockdevRef is given. NULL will be returned, but errp remains unset. | |
3762 | * | |
3763 | * bdrev_key specifies the key for the image's BlockdevRef in the options QDict. | |
3764 | * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict | |
3765 | * itself, all options starting with "${bdref_key}." are considered part of the | |
3766 | * BlockdevRef. | |
3767 | * | |
3768 | * The BlockdevRef will be removed from the options QDict. | |
3769 | * | |
3770 | * The caller must hold the lock of the main AioContext and no other AioContext. | |
3771 | * @parent can move to a different AioContext in this function. Callers must | |
3772 | * make sure that their AioContext locking is still correct after this. | |
3773 | */ | |
3774 | BdrvChild *bdrv_open_child(const char *filename, | |
3775 | QDict *options, const char *bdref_key, | |
3776 | BlockDriverState *parent, | |
3777 | const BdrvChildClass *child_class, | |
3778 | BdrvChildRole child_role, | |
3779 | bool allow_none, Error **errp) | |
3780 | { | |
3781 | BlockDriverState *bs; | |
3782 | BdrvChild *child; | |
3783 | AioContext *ctx; | |
3784 | ||
3785 | GLOBAL_STATE_CODE(); | |
3786 | ||
3787 | bs = bdrv_open_child_bs(filename, options, bdref_key, parent, child_class, | |
3788 | child_role, allow_none, errp); | |
3789 | if (bs == NULL) { | |
3790 | return NULL; | |
3791 | } | |
3792 | ||
3793 | bdrv_graph_wrlock(NULL); | |
3794 | ctx = bdrv_get_aio_context(bs); | |
3795 | aio_context_acquire(ctx); | |
3796 | child = bdrv_attach_child(parent, bs, bdref_key, child_class, child_role, | |
3797 | errp); | |
3798 | aio_context_release(ctx); | |
3799 | bdrv_graph_wrunlock(); | |
3800 | ||
3801 | return child; | |
3802 | } | |
3803 | ||
3804 | /* | |
3805 | * Wrapper on bdrv_open_child() for most popular case: open primary child of bs. | |
3806 | * | |
3807 | * The caller must hold the lock of the main AioContext and no other AioContext. | |
3808 | * @parent can move to a different AioContext in this function. Callers must | |
3809 | * make sure that their AioContext locking is still correct after this. | |
3810 | */ | |
3811 | int bdrv_open_file_child(const char *filename, | |
3812 | QDict *options, const char *bdref_key, | |
3813 | BlockDriverState *parent, Error **errp) | |
3814 | { | |
3815 | BdrvChildRole role; | |
3816 | ||
3817 | /* commit_top and mirror_top don't use this function */ | |
3818 | assert(!parent->drv->filtered_child_is_backing); | |
3819 | role = parent->drv->is_filter ? | |
3820 | (BDRV_CHILD_FILTERED | BDRV_CHILD_PRIMARY) : BDRV_CHILD_IMAGE; | |
3821 | ||
3822 | if (!bdrv_open_child(filename, options, bdref_key, parent, | |
3823 | &child_of_bds, role, false, errp)) | |
3824 | { | |
3825 | return -EINVAL; | |
3826 | } | |
3827 | ||
3828 | return 0; | |
3829 | } | |
3830 | ||
3831 | /* | |
3832 | * TODO Future callers may need to specify parent/child_class in order for | |
3833 | * option inheritance to work. Existing callers use it for the root node. | |
3834 | */ | |
3835 | BlockDriverState *bdrv_open_blockdev_ref(BlockdevRef *ref, Error **errp) | |
3836 | { | |
3837 | BlockDriverState *bs = NULL; | |
3838 | QObject *obj = NULL; | |
3839 | QDict *qdict = NULL; | |
3840 | const char *reference = NULL; | |
3841 | Visitor *v = NULL; | |
3842 | ||
3843 | GLOBAL_STATE_CODE(); | |
3844 | ||
3845 | if (ref->type == QTYPE_QSTRING) { | |
3846 | reference = ref->u.reference; | |
3847 | } else { | |
3848 | BlockdevOptions *options = &ref->u.definition; | |
3849 | assert(ref->type == QTYPE_QDICT); | |
3850 | ||
3851 | v = qobject_output_visitor_new(&obj); | |
3852 | visit_type_BlockdevOptions(v, NULL, &options, &error_abort); | |
3853 | visit_complete(v, &obj); | |
3854 | ||
3855 | qdict = qobject_to(QDict, obj); | |
3856 | qdict_flatten(qdict); | |
3857 | ||
3858 | /* bdrv_open_inherit() defaults to the values in bdrv_flags (for | |
3859 | * compatibility with other callers) rather than what we want as the | |
3860 | * real defaults. Apply the defaults here instead. */ | |
3861 | qdict_set_default_str(qdict, BDRV_OPT_CACHE_DIRECT, "off"); | |
3862 | qdict_set_default_str(qdict, BDRV_OPT_CACHE_NO_FLUSH, "off"); | |
3863 | qdict_set_default_str(qdict, BDRV_OPT_READ_ONLY, "off"); | |
3864 | qdict_set_default_str(qdict, BDRV_OPT_AUTO_READ_ONLY, "off"); | |
3865 | ||
3866 | } | |
3867 | ||
3868 | bs = bdrv_open_inherit(NULL, reference, qdict, 0, NULL, NULL, 0, errp); | |
3869 | obj = NULL; | |
3870 | qobject_unref(obj); | |
3871 | visit_free(v); | |
3872 | return bs; | |
3873 | } | |
3874 | ||
3875 | static BlockDriverState *bdrv_append_temp_snapshot(BlockDriverState *bs, | |
3876 | int flags, | |
3877 | QDict *snapshot_options, | |
3878 | Error **errp) | |
3879 | { | |
3880 | g_autofree char *tmp_filename = NULL; | |
3881 | int64_t total_size; | |
3882 | QemuOpts *opts = NULL; | |
3883 | BlockDriverState *bs_snapshot = NULL; | |
3884 | AioContext *ctx = bdrv_get_aio_context(bs); | |
3885 | int ret; | |
3886 | ||
3887 | GLOBAL_STATE_CODE(); | |
3888 | ||
3889 | /* if snapshot, we create a temporary backing file and open it | |
3890 | instead of opening 'filename' directly */ | |
3891 | ||
3892 | /* Get the required size from the image */ | |
3893 | aio_context_acquire(ctx); | |
3894 | total_size = bdrv_getlength(bs); | |
3895 | aio_context_release(ctx); | |
3896 | ||
3897 | if (total_size < 0) { | |
3898 | error_setg_errno(errp, -total_size, "Could not get image size"); | |
3899 | goto out; | |
3900 | } | |
3901 | ||
3902 | /* Create the temporary image */ | |
3903 | tmp_filename = create_tmp_file(errp); | |
3904 | if (!tmp_filename) { | |
3905 | goto out; | |
3906 | } | |
3907 | ||
3908 | opts = qemu_opts_create(bdrv_qcow2.create_opts, NULL, 0, | |
3909 | &error_abort); | |
3910 | qemu_opt_set_number(opts, BLOCK_OPT_SIZE, total_size, &error_abort); | |
3911 | ret = bdrv_create(&bdrv_qcow2, tmp_filename, opts, errp); | |
3912 | qemu_opts_del(opts); | |
3913 | if (ret < 0) { | |
3914 | error_prepend(errp, "Could not create temporary overlay '%s': ", | |
3915 | tmp_filename); | |
3916 | goto out; | |
3917 | } | |
3918 | ||
3919 | /* Prepare options QDict for the temporary file */ | |
3920 | qdict_put_str(snapshot_options, "file.driver", "file"); | |
3921 | qdict_put_str(snapshot_options, "file.filename", tmp_filename); | |
3922 | qdict_put_str(snapshot_options, "driver", "qcow2"); | |
3923 | ||
3924 | bs_snapshot = bdrv_open(NULL, NULL, snapshot_options, flags, errp); | |
3925 | snapshot_options = NULL; | |
3926 | if (!bs_snapshot) { | |
3927 | goto out; | |
3928 | } | |
3929 | ||
3930 | aio_context_acquire(ctx); | |
3931 | ret = bdrv_append(bs_snapshot, bs, errp); | |
3932 | aio_context_release(ctx); | |
3933 | ||
3934 | if (ret < 0) { | |
3935 | bs_snapshot = NULL; | |
3936 | goto out; | |
3937 | } | |
3938 | ||
3939 | out: | |
3940 | qobject_unref(snapshot_options); | |
3941 | return bs_snapshot; | |
3942 | } | |
3943 | ||
3944 | /* | |
3945 | * Opens a disk image (raw, qcow2, vmdk, ...) | |
3946 | * | |
3947 | * options is a QDict of options to pass to the block drivers, or NULL for an | |
3948 | * empty set of options. The reference to the QDict belongs to the block layer | |
3949 | * after the call (even on failure), so if the caller intends to reuse the | |
3950 | * dictionary, it needs to use qobject_ref() before calling bdrv_open. | |
3951 | * | |
3952 | * If *pbs is NULL, a new BDS will be created with a pointer to it stored there. | |
3953 | * If it is not NULL, the referenced BDS will be reused. | |
3954 | * | |
3955 | * The reference parameter may be used to specify an existing block device which | |
3956 | * should be opened. If specified, neither options nor a filename may be given, | |
3957 | * nor can an existing BDS be reused (that is, *pbs has to be NULL). | |
3958 | * | |
3959 | * The caller must always hold the main AioContext lock. | |
3960 | */ | |
3961 | static BlockDriverState * no_coroutine_fn | |
3962 | bdrv_open_inherit(const char *filename, const char *reference, QDict *options, | |
3963 | int flags, BlockDriverState *parent, | |
3964 | const BdrvChildClass *child_class, BdrvChildRole child_role, | |
3965 | Error **errp) | |
3966 | { | |
3967 | int ret; | |
3968 | BlockBackend *file = NULL; | |
3969 | BlockDriverState *bs; | |
3970 | BlockDriver *drv = NULL; | |
3971 | BdrvChild *child; | |
3972 | const char *drvname; | |
3973 | const char *backing; | |
3974 | Error *local_err = NULL; | |
3975 | QDict *snapshot_options = NULL; | |
3976 | int snapshot_flags = 0; | |
3977 | AioContext *ctx = qemu_get_aio_context(); | |
3978 | ||
3979 | assert(!child_class || !flags); | |
3980 | assert(!child_class == !parent); | |
3981 | GLOBAL_STATE_CODE(); | |
3982 | assert(!qemu_in_coroutine()); | |
3983 | ||
3984 | /* TODO We'll eventually have to take a writer lock in this function */ | |
3985 | GRAPH_RDLOCK_GUARD_MAINLOOP(); | |
3986 | ||
3987 | if (reference) { | |
3988 | bool options_non_empty = options ? qdict_size(options) : false; | |
3989 | qobject_unref(options); | |
3990 | ||
3991 | if (filename || options_non_empty) { | |
3992 | error_setg(errp, "Cannot reference an existing block device with " | |
3993 | "additional options or a new filename"); | |
3994 | return NULL; | |
3995 | } | |
3996 | ||
3997 | bs = bdrv_lookup_bs(reference, reference, errp); | |
3998 | if (!bs) { | |
3999 | return NULL; | |
4000 | } | |
4001 | ||
4002 | bdrv_ref(bs); | |
4003 | return bs; | |
4004 | } | |
4005 | ||
4006 | bs = bdrv_new(); | |
4007 | ||
4008 | /* NULL means an empty set of options */ | |
4009 | if (options == NULL) { | |
4010 | options = qdict_new(); | |
4011 | } | |
4012 | ||
4013 | /* json: syntax counts as explicit options, as if in the QDict */ | |
4014 | parse_json_protocol(options, &filename, &local_err); | |
4015 | if (local_err) { | |
4016 | goto fail; | |
4017 | } | |
4018 | ||
4019 | bs->explicit_options = qdict_clone_shallow(options); | |
4020 | ||
4021 | if (child_class) { | |
4022 | bool parent_is_format; | |
4023 | ||
4024 | if (parent->drv) { | |
4025 | parent_is_format = parent->drv->is_format; | |
4026 | } else { | |
4027 | /* | |
4028 | * parent->drv is not set yet because this node is opened for | |
4029 | * (potential) format probing. That means that @parent is going | |
4030 | * to be a format node. | |
4031 | */ | |
4032 | parent_is_format = true; | |
4033 | } | |
4034 | ||
4035 | bs->inherits_from = parent; | |
4036 | child_class->inherit_options(child_role, parent_is_format, | |
4037 | &flags, options, | |
4038 | parent->open_flags, parent->options); | |
4039 | } | |
4040 | ||
4041 | ret = bdrv_fill_options(&options, filename, &flags, &local_err); | |
4042 | if (ret < 0) { | |
4043 | goto fail; | |
4044 | } | |
4045 | ||
4046 | /* | |
4047 | * Set the BDRV_O_RDWR and BDRV_O_ALLOW_RDWR flags. | |
4048 | * Caution: getting a boolean member of @options requires care. | |
4049 | * When @options come from -blockdev or blockdev_add, members are | |
4050 | * typed according to the QAPI schema, but when they come from | |
4051 | * -drive, they're all QString. | |
4052 | */ | |
4053 | if (g_strcmp0(qdict_get_try_str(options, BDRV_OPT_READ_ONLY), "on") && | |
4054 | !qdict_get_try_bool(options, BDRV_OPT_READ_ONLY, false)) { | |
4055 | flags |= (BDRV_O_RDWR | BDRV_O_ALLOW_RDWR); | |
4056 | } else { | |
4057 | flags &= ~BDRV_O_RDWR; | |
4058 | } | |
4059 | ||
4060 | if (flags & BDRV_O_SNAPSHOT) { | |
4061 | snapshot_options = qdict_new(); | |
4062 | bdrv_temp_snapshot_options(&snapshot_flags, snapshot_options, | |
4063 | flags, options); | |
4064 | /* Let bdrv_backing_options() override "read-only" */ | |
4065 | qdict_del(options, BDRV_OPT_READ_ONLY); | |
4066 | bdrv_inherited_options(BDRV_CHILD_COW, true, | |
4067 | &flags, options, flags, options); | |
4068 | } | |
4069 | ||
4070 | bs->open_flags = flags; | |
4071 | bs->options = options; | |
4072 | options = qdict_clone_shallow(options); | |
4073 | ||
4074 | /* Find the right image format driver */ | |
4075 | /* See cautionary note on accessing @options above */ | |
4076 | drvname = qdict_get_try_str(options, "driver"); | |
4077 | if (drvname) { | |
4078 | drv = bdrv_find_format(drvname); | |
4079 | if (!drv) { | |
4080 | error_setg(errp, "Unknown driver: '%s'", drvname); | |
4081 | goto fail; | |
4082 | } | |
4083 | } | |
4084 | ||
4085 | assert(drvname || !(flags & BDRV_O_PROTOCOL)); | |
4086 | ||
4087 | /* See cautionary note on accessing @options above */ | |
4088 | backing = qdict_get_try_str(options, "backing"); | |
4089 | if (qobject_to(QNull, qdict_get(options, "backing")) != NULL || | |
4090 | (backing && *backing == '\0')) | |
4091 | { | |
4092 | if (backing) { | |
4093 | warn_report("Use of \"backing\": \"\" is deprecated; " | |
4094 | "use \"backing\": null instead"); | |
4095 | } | |
4096 | flags |= BDRV_O_NO_BACKING; | |
4097 | qdict_del(bs->explicit_options, "backing"); | |
4098 | qdict_del(bs->options, "backing"); | |
4099 | qdict_del(options, "backing"); | |
4100 | } | |
4101 | ||
4102 | /* Open image file without format layer. This BlockBackend is only used for | |
4103 | * probing, the block drivers will do their own bdrv_open_child() for the | |
4104 | * same BDS, which is why we put the node name back into options. */ | |
4105 | if ((flags & BDRV_O_PROTOCOL) == 0) { | |
4106 | BlockDriverState *file_bs; | |
4107 | ||
4108 | file_bs = bdrv_open_child_bs(filename, options, "file", bs, | |
4109 | &child_of_bds, BDRV_CHILD_IMAGE, | |
4110 | true, &local_err); | |
4111 | if (local_err) { | |
4112 | goto fail; | |
4113 | } | |
4114 | if (file_bs != NULL) { | |
4115 | /* Not requesting BLK_PERM_CONSISTENT_READ because we're only | |
4116 | * looking at the header to guess the image format. This works even | |
4117 | * in cases where a guest would not see a consistent state. */ | |
4118 | ctx = bdrv_get_aio_context(file_bs); | |
4119 | aio_context_acquire(ctx); | |
4120 | file = blk_new(ctx, 0, BLK_PERM_ALL); | |
4121 | blk_insert_bs(file, file_bs, &local_err); | |
4122 | bdrv_unref(file_bs); | |
4123 | aio_context_release(ctx); | |
4124 | ||
4125 | if (local_err) { | |
4126 | goto fail; | |
4127 | } | |
4128 | ||
4129 | qdict_put_str(options, "file", bdrv_get_node_name(file_bs)); | |
4130 | } | |
4131 | } | |
4132 | ||
4133 | /* Image format probing */ | |
4134 | bs->probed = !drv; | |
4135 | if (!drv && file) { | |
4136 | ret = find_image_format(file, filename, &drv, &local_err); | |
4137 | if (ret < 0) { | |
4138 | goto fail; | |
4139 | } | |
4140 | /* | |
4141 | * This option update would logically belong in bdrv_fill_options(), | |
4142 | * but we first need to open bs->file for the probing to work, while | |
4143 | * opening bs->file already requires the (mostly) final set of options | |
4144 | * so that cache mode etc. can be inherited. | |
4145 | * | |
4146 | * Adding the driver later is somewhat ugly, but it's not an option | |
4147 | * that would ever be inherited, so it's correct. We just need to make | |
4148 | * sure to update both bs->options (which has the full effective | |
4149 | * options for bs) and options (which has file.* already removed). | |
4150 | */ | |
4151 | qdict_put_str(bs->options, "driver", drv->format_name); | |
4152 | qdict_put_str(options, "driver", drv->format_name); | |
4153 | } else if (!drv) { | |
4154 | error_setg(errp, "Must specify either driver or file"); | |
4155 | goto fail; | |
4156 | } | |
4157 | ||
4158 | /* BDRV_O_PROTOCOL must be set iff a protocol BDS is about to be created */ | |
4159 | assert(!!(flags & BDRV_O_PROTOCOL) == !!drv->bdrv_file_open); | |
4160 | /* file must be NULL if a protocol BDS is about to be created | |
4161 | * (the inverse results in an error message from bdrv_open_common()) */ | |
4162 | assert(!(flags & BDRV_O_PROTOCOL) || !file); | |
4163 | ||
4164 | /* Open the image */ | |
4165 | ret = bdrv_open_common(bs, file, options, &local_err); | |
4166 | if (ret < 0) { | |
4167 | goto fail; | |
4168 | } | |
4169 | ||
4170 | /* The AioContext could have changed during bdrv_open_common() */ | |
4171 | ctx = bdrv_get_aio_context(bs); | |
4172 | ||
4173 | if (file) { | |
4174 | aio_context_acquire(ctx); | |
4175 | blk_unref(file); | |
4176 | aio_context_release(ctx); | |
4177 | file = NULL; | |
4178 | } | |
4179 | ||
4180 | /* If there is a backing file, use it */ | |
4181 | if ((flags & BDRV_O_NO_BACKING) == 0) { | |
4182 | ret = bdrv_open_backing_file(bs, options, "backing", &local_err); | |
4183 | if (ret < 0) { | |
4184 | goto close_and_fail; | |
4185 | } | |
4186 | } | |
4187 | ||
4188 | /* Remove all children options and references | |
4189 | * from bs->options and bs->explicit_options */ | |
4190 | QLIST_FOREACH(child, &bs->children, next) { | |
4191 | char *child_key_dot; | |
4192 | child_key_dot = g_strdup_printf("%s.", child->name); | |
4193 | qdict_extract_subqdict(bs->explicit_options, NULL, child_key_dot); | |
4194 | qdict_extract_subqdict(bs->options, NULL, child_key_dot); | |
4195 | qdict_del(bs->explicit_options, child->name); | |
4196 | qdict_del(bs->options, child->name); | |
4197 | g_free(child_key_dot); | |
4198 | } | |
4199 | ||
4200 | /* Check if any unknown options were used */ | |
4201 | if (qdict_size(options) != 0) { | |
4202 | const QDictEntry *entry = qdict_first(options); | |
4203 | if (flags & BDRV_O_PROTOCOL) { | |
4204 | error_setg(errp, "Block protocol '%s' doesn't support the option " | |
4205 | "'%s'", drv->format_name, entry->key); | |
4206 | } else { | |
4207 | error_setg(errp, | |
4208 | "Block format '%s' does not support the option '%s'", | |
4209 | drv->format_name, entry->key); | |
4210 | } | |
4211 | ||
4212 | goto close_and_fail; | |
4213 | } | |
4214 | ||
4215 | bdrv_parent_cb_change_media(bs, true); | |
4216 | ||
4217 | qobject_unref(options); | |
4218 | options = NULL; | |
4219 | ||
4220 | /* For snapshot=on, create a temporary qcow2 overlay. bs points to the | |
4221 | * temporary snapshot afterwards. */ | |
4222 | if (snapshot_flags) { | |
4223 | BlockDriverState *snapshot_bs; | |
4224 | snapshot_bs = bdrv_append_temp_snapshot(bs, snapshot_flags, | |
4225 | snapshot_options, &local_err); | |
4226 | snapshot_options = NULL; | |
4227 | if (local_err) { | |
4228 | goto close_and_fail; | |
4229 | } | |
4230 | /* We are not going to return bs but the overlay on top of it | |
4231 | * (snapshot_bs); thus, we have to drop the strong reference to bs | |
4232 | * (which we obtained by calling bdrv_new()). bs will not be deleted, | |
4233 | * though, because the overlay still has a reference to it. */ | |
4234 | aio_context_acquire(ctx); | |
4235 | bdrv_unref(bs); | |
4236 | aio_context_release(ctx); | |
4237 | bs = snapshot_bs; | |
4238 | } | |
4239 | ||
4240 | return bs; | |
4241 | ||
4242 | fail: | |
4243 | aio_context_acquire(ctx); | |
4244 | blk_unref(file); | |
4245 | qobject_unref(snapshot_options); | |
4246 | qobject_unref(bs->explicit_options); | |
4247 | qobject_unref(bs->options); | |
4248 | qobject_unref(options); | |
4249 | bs->options = NULL; | |
4250 | bs->explicit_options = NULL; | |
4251 | bdrv_unref(bs); | |
4252 | aio_context_release(ctx); | |
4253 | error_propagate(errp, local_err); | |
4254 | return NULL; | |
4255 | ||
4256 | close_and_fail: | |
4257 | aio_context_acquire(ctx); | |
4258 | bdrv_unref(bs); | |
4259 | aio_context_release(ctx); | |
4260 | qobject_unref(snapshot_options); | |
4261 | qobject_unref(options); | |
4262 | error_propagate(errp, local_err); | |
4263 | return NULL; | |
4264 | } | |
4265 | ||
4266 | /* The caller must always hold the main AioContext lock. */ | |
4267 | BlockDriverState *bdrv_open(const char *filename, const char *reference, | |
4268 | QDict *options, int flags, Error **errp) | |
4269 | { | |
4270 | GLOBAL_STATE_CODE(); | |
4271 | ||
4272 | return bdrv_open_inherit(filename, reference, options, flags, NULL, | |
4273 | NULL, 0, errp); | |
4274 | } | |
4275 | ||
4276 | /* Return true if the NULL-terminated @list contains @str */ | |
4277 | static bool is_str_in_list(const char *str, const char *const *list) | |
4278 | { | |
4279 | if (str && list) { | |
4280 | int i; | |
4281 | for (i = 0; list[i] != NULL; i++) { | |
4282 | if (!strcmp(str, list[i])) { | |
4283 | return true; | |
4284 | } | |
4285 | } | |
4286 | } | |
4287 | return false; | |
4288 | } | |
4289 | ||
4290 | /* | |
4291 | * Check that every option set in @bs->options is also set in | |
4292 | * @new_opts. | |
4293 | * | |
4294 | * Options listed in the common_options list and in | |
4295 | * @bs->drv->mutable_opts are skipped. | |
4296 | * | |
4297 | * Return 0 on success, otherwise return -EINVAL and set @errp. | |
4298 | */ | |
4299 | static int bdrv_reset_options_allowed(BlockDriverState *bs, | |
4300 | const QDict *new_opts, Error **errp) | |
4301 | { | |
4302 | const QDictEntry *e; | |
4303 | /* These options are common to all block drivers and are handled | |
4304 | * in bdrv_reopen_prepare() so they can be left out of @new_opts */ | |
4305 | const char *const common_options[] = { | |
4306 | "node-name", "discard", "cache.direct", "cache.no-flush", | |
4307 | "read-only", "auto-read-only", "detect-zeroes", NULL | |
4308 | }; | |
4309 | ||
4310 | for (e = qdict_first(bs->options); e; e = qdict_next(bs->options, e)) { | |
4311 | if (!qdict_haskey(new_opts, e->key) && | |
4312 | !is_str_in_list(e->key, common_options) && | |
4313 | !is_str_in_list(e->key, bs->drv->mutable_opts)) { | |
4314 | error_setg(errp, "Option '%s' cannot be reset " | |
4315 | "to its default value", e->key); | |
4316 | return -EINVAL; | |
4317 | } | |
4318 | } | |
4319 | ||
4320 | return 0; | |
4321 | } | |
4322 | ||
4323 | /* | |
4324 | * Returns true if @child can be reached recursively from @bs | |
4325 | */ | |
4326 | static bool GRAPH_RDLOCK | |
4327 | bdrv_recurse_has_child(BlockDriverState *bs, BlockDriverState *child) | |
4328 | { | |
4329 | BdrvChild *c; | |
4330 | ||
4331 | if (bs == child) { | |
4332 | return true; | |
4333 | } | |
4334 | ||
4335 | QLIST_FOREACH(c, &bs->children, next) { | |
4336 | if (bdrv_recurse_has_child(c->bs, child)) { | |
4337 | return true; | |
4338 | } | |
4339 | } | |
4340 | ||
4341 | return false; | |
4342 | } | |
4343 | ||
4344 | /* | |
4345 | * Adds a BlockDriverState to a simple queue for an atomic, transactional | |
4346 | * reopen of multiple devices. | |
4347 | * | |
4348 | * bs_queue can either be an existing BlockReopenQueue that has had QTAILQ_INIT | |
4349 | * already performed, or alternatively may be NULL a new BlockReopenQueue will | |
4350 | * be created and initialized. This newly created BlockReopenQueue should be | |
4351 | * passed back in for subsequent calls that are intended to be of the same | |
4352 | * atomic 'set'. | |
4353 | * | |
4354 | * bs is the BlockDriverState to add to the reopen queue. | |
4355 | * | |
4356 | * options contains the changed options for the associated bs | |
4357 | * (the BlockReopenQueue takes ownership) | |
4358 | * | |
4359 | * flags contains the open flags for the associated bs | |
4360 | * | |
4361 | * returns a pointer to bs_queue, which is either the newly allocated | |
4362 | * bs_queue, or the existing bs_queue being used. | |
4363 | * | |
4364 | * bs is drained here and undrained by bdrv_reopen_queue_free(). | |
4365 | * | |
4366 | * To be called with bs->aio_context locked. | |
4367 | */ | |
4368 | static BlockReopenQueue * GRAPH_RDLOCK | |
4369 | bdrv_reopen_queue_child(BlockReopenQueue *bs_queue, BlockDriverState *bs, | |
4370 | QDict *options, const BdrvChildClass *klass, | |
4371 | BdrvChildRole role, bool parent_is_format, | |
4372 | QDict *parent_options, int parent_flags, | |
4373 | bool keep_old_opts) | |
4374 | { | |
4375 | assert(bs != NULL); | |
4376 | ||
4377 | BlockReopenQueueEntry *bs_entry; | |
4378 | BdrvChild *child; | |
4379 | QDict *old_options, *explicit_options, *options_copy; | |
4380 | int flags; | |
4381 | QemuOpts *opts; | |
4382 | ||
4383 | GLOBAL_STATE_CODE(); | |
4384 | ||
4385 | /* | |
4386 | * Strictly speaking, draining is illegal under GRAPH_RDLOCK. We know that | |
4387 | * we've been called with bdrv_graph_rdlock_main_loop(), though, so it's ok | |
4388 | * in practice. | |
4389 | */ | |
4390 | bdrv_drained_begin(bs); | |
4391 | ||
4392 | if (bs_queue == NULL) { | |
4393 | bs_queue = g_new0(BlockReopenQueue, 1); | |
4394 | QTAILQ_INIT(bs_queue); | |
4395 | } | |
4396 | ||
4397 | if (!options) { | |
4398 | options = qdict_new(); | |
4399 | } | |
4400 | ||
4401 | /* Check if this BlockDriverState is already in the queue */ | |
4402 | QTAILQ_FOREACH(bs_entry, bs_queue, entry) { | |
4403 | if (bs == bs_entry->state.bs) { | |
4404 | break; | |
4405 | } | |
4406 | } | |
4407 | ||
4408 | /* | |
4409 | * Precedence of options: | |
4410 | * 1. Explicitly passed in options (highest) | |
4411 | * 2. Retained from explicitly set options of bs | |
4412 | * 3. Inherited from parent node | |
4413 | * 4. Retained from effective options of bs | |
4414 | */ | |
4415 | ||
4416 | /* Old explicitly set values (don't overwrite by inherited value) */ | |
4417 | if (bs_entry || keep_old_opts) { | |
4418 | old_options = qdict_clone_shallow(bs_entry ? | |
4419 | bs_entry->state.explicit_options : | |
4420 | bs->explicit_options); | |
4421 | bdrv_join_options(bs, options, old_options); | |
4422 | qobject_unref(old_options); | |
4423 | } | |
4424 | ||
4425 | explicit_options = qdict_clone_shallow(options); | |
4426 | ||
4427 | /* Inherit from parent node */ | |
4428 | if (parent_options) { | |
4429 | flags = 0; | |
4430 | klass->inherit_options(role, parent_is_format, &flags, options, | |
4431 | parent_flags, parent_options); | |
4432 | } else { | |
4433 | flags = bdrv_get_flags(bs); | |
4434 | } | |
4435 | ||
4436 | if (keep_old_opts) { | |
4437 | /* Old values are used for options that aren't set yet */ | |
4438 | old_options = qdict_clone_shallow(bs->options); | |
4439 | bdrv_join_options(bs, options, old_options); | |
4440 | qobject_unref(old_options); | |
4441 | } | |
4442 | ||
4443 | /* We have the final set of options so let's update the flags */ | |
4444 | options_copy = qdict_clone_shallow(options); | |
4445 | opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort); | |
4446 | qemu_opts_absorb_qdict(opts, options_copy, NULL); | |
4447 | update_flags_from_options(&flags, opts); | |
4448 | qemu_opts_del(opts); | |
4449 | qobject_unref(options_copy); | |
4450 | ||
4451 | /* bdrv_open_inherit() sets and clears some additional flags internally */ | |
4452 | flags &= ~BDRV_O_PROTOCOL; | |
4453 | if (flags & BDRV_O_RDWR) { | |
4454 | flags |= BDRV_O_ALLOW_RDWR; | |
4455 | } | |
4456 | ||
4457 | if (!bs_entry) { | |
4458 | bs_entry = g_new0(BlockReopenQueueEntry, 1); | |
4459 | QTAILQ_INSERT_TAIL(bs_queue, bs_entry, entry); | |
4460 | } else { | |
4461 | qobject_unref(bs_entry->state.options); | |
4462 | qobject_unref(bs_entry->state.explicit_options); | |
4463 | } | |
4464 | ||
4465 | bs_entry->state.bs = bs; | |
4466 | bs_entry->state.options = options; | |
4467 | bs_entry->state.explicit_options = explicit_options; | |
4468 | bs_entry->state.flags = flags; | |
4469 | ||
4470 | /* | |
4471 | * If keep_old_opts is false then it means that unspecified | |
4472 | * options must be reset to their original value. We don't allow | |
4473 | * resetting 'backing' but we need to know if the option is | |
4474 | * missing in order to decide if we have to return an error. | |
4475 | */ | |
4476 | if (!keep_old_opts) { | |
4477 | bs_entry->state.backing_missing = | |
4478 | !qdict_haskey(options, "backing") && | |
4479 | !qdict_haskey(options, "backing.driver"); | |
4480 | } | |
4481 | ||
4482 | QLIST_FOREACH(child, &bs->children, next) { | |
4483 | QDict *new_child_options = NULL; | |
4484 | bool child_keep_old = keep_old_opts; | |
4485 | ||
4486 | /* reopen can only change the options of block devices that were | |
4487 | * implicitly created and inherited options. For other (referenced) | |
4488 | * block devices, a syntax like "backing.foo" results in an error. */ | |
4489 | if (child->bs->inherits_from != bs) { | |
4490 | continue; | |
4491 | } | |
4492 | ||
4493 | /* Check if the options contain a child reference */ | |
4494 | if (qdict_haskey(options, child->name)) { | |
4495 | const char *childref = qdict_get_try_str(options, child->name); | |
4496 | /* | |
4497 | * The current child must not be reopened if the child | |
4498 | * reference is null or points to a different node. | |
4499 | */ | |
4500 | if (g_strcmp0(childref, child->bs->node_name)) { | |
4501 | continue; | |
4502 | } | |
4503 | /* | |
4504 | * If the child reference points to the current child then | |
4505 | * reopen it with its existing set of options (note that | |
4506 | * it can still inherit new options from the parent). | |
4507 | */ | |
4508 | child_keep_old = true; | |
4509 | } else { | |
4510 | /* Extract child options ("child-name.*") */ | |
4511 | char *child_key_dot = g_strdup_printf("%s.", child->name); | |
4512 | qdict_extract_subqdict(explicit_options, NULL, child_key_dot); | |
4513 | qdict_extract_subqdict(options, &new_child_options, child_key_dot); | |
4514 | g_free(child_key_dot); | |
4515 | } | |
4516 | ||
4517 | bdrv_reopen_queue_child(bs_queue, child->bs, new_child_options, | |
4518 | child->klass, child->role, bs->drv->is_format, | |
4519 | options, flags, child_keep_old); | |
4520 | } | |
4521 | ||
4522 | return bs_queue; | |
4523 | } | |
4524 | ||
4525 | /* To be called with bs->aio_context locked */ | |
4526 | BlockReopenQueue *bdrv_reopen_queue(BlockReopenQueue *bs_queue, | |
4527 | BlockDriverState *bs, | |
4528 | QDict *options, bool keep_old_opts) | |
4529 | { | |
4530 | GLOBAL_STATE_CODE(); | |
4531 | GRAPH_RDLOCK_GUARD_MAINLOOP(); | |
4532 | ||
4533 | return bdrv_reopen_queue_child(bs_queue, bs, options, NULL, 0, false, | |
4534 | NULL, 0, keep_old_opts); | |
4535 | } | |
4536 | ||
4537 | void bdrv_reopen_queue_free(BlockReopenQueue *bs_queue) | |
4538 | { | |
4539 | GLOBAL_STATE_CODE(); | |
4540 | if (bs_queue) { | |
4541 | BlockReopenQueueEntry *bs_entry, *next; | |
4542 | QTAILQ_FOREACH_SAFE(bs_entry, bs_queue, entry, next) { | |
4543 | AioContext *ctx = bdrv_get_aio_context(bs_entry->state.bs); | |
4544 | ||
4545 | aio_context_acquire(ctx); | |
4546 | bdrv_drained_end(bs_entry->state.bs); | |
4547 | aio_context_release(ctx); | |
4548 | ||
4549 | qobject_unref(bs_entry->state.explicit_options); | |
4550 | qobject_unref(bs_entry->state.options); | |
4551 | g_free(bs_entry); | |
4552 | } | |
4553 | g_free(bs_queue); | |
4554 | } | |
4555 | } | |
4556 | ||
4557 | /* | |
4558 | * Reopen multiple BlockDriverStates atomically & transactionally. | |
4559 | * | |
4560 | * The queue passed in (bs_queue) must have been built up previous | |
4561 | * via bdrv_reopen_queue(). | |
4562 | * | |
4563 | * Reopens all BDS specified in the queue, with the appropriate | |
4564 | * flags. All devices are prepared for reopen, and failure of any | |
4565 | * device will cause all device changes to be abandoned, and intermediate | |
4566 | * data cleaned up. | |
4567 | * | |
4568 | * If all devices prepare successfully, then the changes are committed | |
4569 | * to all devices. | |
4570 | * | |
4571 | * All affected nodes must be drained between bdrv_reopen_queue() and | |
4572 | * bdrv_reopen_multiple(). | |
4573 | * | |
4574 | * To be called from the main thread, with all other AioContexts unlocked. | |
4575 | */ | |
4576 | int bdrv_reopen_multiple(BlockReopenQueue *bs_queue, Error **errp) | |
4577 | { | |
4578 | int ret = -1; | |
4579 | BlockReopenQueueEntry *bs_entry, *next; | |
4580 | AioContext *ctx; | |
4581 | Transaction *tran = tran_new(); | |
4582 | g_autoptr(GSList) refresh_list = NULL; | |
4583 | ||
4584 | assert(qemu_get_current_aio_context() == qemu_get_aio_context()); | |
4585 | assert(bs_queue != NULL); | |
4586 | GLOBAL_STATE_CODE(); | |
4587 | ||
4588 | QTAILQ_FOREACH(bs_entry, bs_queue, entry) { | |
4589 | ctx = bdrv_get_aio_context(bs_entry->state.bs); | |
4590 | aio_context_acquire(ctx); | |
4591 | ret = bdrv_flush(bs_entry->state.bs); | |
4592 | aio_context_release(ctx); | |
4593 | if (ret < 0) { | |
4594 | error_setg_errno(errp, -ret, "Error flushing drive"); | |
4595 | goto abort; | |
4596 | } | |
4597 | } | |
4598 | ||
4599 | QTAILQ_FOREACH(bs_entry, bs_queue, entry) { | |
4600 | assert(bs_entry->state.bs->quiesce_counter > 0); | |
4601 | ctx = bdrv_get_aio_context(bs_entry->state.bs); | |
4602 | aio_context_acquire(ctx); | |
4603 | ret = bdrv_reopen_prepare(&bs_entry->state, bs_queue, tran, errp); | |
4604 | aio_context_release(ctx); | |
4605 | if (ret < 0) { | |
4606 | goto abort; | |
4607 | } | |
4608 | bs_entry->prepared = true; | |
4609 | } | |
4610 | ||
4611 | QTAILQ_FOREACH(bs_entry, bs_queue, entry) { | |
4612 | BDRVReopenState *state = &bs_entry->state; | |
4613 | ||
4614 | refresh_list = g_slist_prepend(refresh_list, state->bs); | |
4615 | if (state->old_backing_bs) { | |
4616 | refresh_list = g_slist_prepend(refresh_list, state->old_backing_bs); | |
4617 | } | |
4618 | if (state->old_file_bs) { | |
4619 | refresh_list = g_slist_prepend(refresh_list, state->old_file_bs); | |
4620 | } | |
4621 | } | |
4622 | ||
4623 | /* | |
4624 | * Note that file-posix driver rely on permission update done during reopen | |
4625 | * (even if no permission changed), because it wants "new" permissions for | |
4626 | * reconfiguring the fd and that's why it does it in raw_check_perm(), not | |
4627 | * in raw_reopen_prepare() which is called with "old" permissions. | |
4628 | */ | |
4629 | bdrv_graph_rdlock_main_loop(); | |
4630 | ret = bdrv_list_refresh_perms(refresh_list, bs_queue, tran, errp); | |
4631 | bdrv_graph_rdunlock_main_loop(); | |
4632 | ||
4633 | if (ret < 0) { | |
4634 | goto abort; | |
4635 | } | |
4636 | ||
4637 | /* | |
4638 | * If we reach this point, we have success and just need to apply the | |
4639 | * changes. | |
4640 | * | |
4641 | * Reverse order is used to comfort qcow2 driver: on commit it need to write | |
4642 | * IN_USE flag to the image, to mark bitmaps in the image as invalid. But | |
4643 | * children are usually goes after parents in reopen-queue, so go from last | |
4644 | * to first element. | |
4645 | */ | |
4646 | QTAILQ_FOREACH_REVERSE(bs_entry, bs_queue, entry) { | |
4647 | ctx = bdrv_get_aio_context(bs_entry->state.bs); | |
4648 | aio_context_acquire(ctx); | |
4649 | bdrv_reopen_commit(&bs_entry->state); | |
4650 | aio_context_release(ctx); | |
4651 | } | |
4652 | ||
4653 | bdrv_graph_wrlock(NULL); | |
4654 | tran_commit(tran); | |
4655 | bdrv_graph_wrunlock(); | |
4656 | ||
4657 | QTAILQ_FOREACH_REVERSE(bs_entry, bs_queue, entry) { | |
4658 | BlockDriverState *bs = bs_entry->state.bs; | |
4659 | ||
4660 | if (bs->drv->bdrv_reopen_commit_post) { | |
4661 | ctx = bdrv_get_aio_context(bs); | |
4662 | aio_context_acquire(ctx); | |
4663 | bs->drv->bdrv_reopen_commit_post(&bs_entry->state); | |
4664 | aio_context_release(ctx); | |
4665 | } | |
4666 | } | |
4667 | ||
4668 | ret = 0; | |
4669 | goto cleanup; | |
4670 | ||
4671 | abort: | |
4672 | bdrv_graph_wrlock(NULL); | |
4673 | tran_abort(tran); | |
4674 | bdrv_graph_wrunlock(); | |
4675 | ||
4676 | QTAILQ_FOREACH_SAFE(bs_entry, bs_queue, entry, next) { | |
4677 | if (bs_entry->prepared) { | |
4678 | ctx = bdrv_get_aio_context(bs_entry->state.bs); | |
4679 | aio_context_acquire(ctx); | |
4680 | bdrv_reopen_abort(&bs_entry->state); | |
4681 | aio_context_release(ctx); | |
4682 | } | |
4683 | } | |
4684 | ||
4685 | cleanup: | |
4686 | bdrv_reopen_queue_free(bs_queue); | |
4687 | ||
4688 | return ret; | |
4689 | } | |
4690 | ||
4691 | int bdrv_reopen(BlockDriverState *bs, QDict *opts, bool keep_old_opts, | |
4692 | Error **errp) | |
4693 | { | |
4694 | AioContext *ctx = bdrv_get_aio_context(bs); | |
4695 | BlockReopenQueue *queue; | |
4696 | int ret; | |
4697 | ||
4698 | GLOBAL_STATE_CODE(); | |
4699 | ||
4700 | queue = bdrv_reopen_queue(NULL, bs, opts, keep_old_opts); | |
4701 | ||
4702 | if (ctx != qemu_get_aio_context()) { | |
4703 | aio_context_release(ctx); | |
4704 | } | |
4705 | ret = bdrv_reopen_multiple(queue, errp); | |
4706 | ||
4707 | if (ctx != qemu_get_aio_context()) { | |
4708 | aio_context_acquire(ctx); | |
4709 | } | |
4710 | ||
4711 | return ret; | |
4712 | } | |
4713 | ||
4714 | int bdrv_reopen_set_read_only(BlockDriverState *bs, bool read_only, | |
4715 | Error **errp) | |
4716 | { | |
4717 | QDict *opts = qdict_new(); | |
4718 | ||
4719 | GLOBAL_STATE_CODE(); | |
4720 | ||
4721 | qdict_put_bool(opts, BDRV_OPT_READ_ONLY, read_only); | |
4722 | ||
4723 | return bdrv_reopen(bs, opts, true, errp); | |
4724 | } | |
4725 | ||
4726 | /* | |
4727 | * Take a BDRVReopenState and check if the value of 'backing' in the | |
4728 | * reopen_state->options QDict is valid or not. | |
4729 | * | |
4730 | * If 'backing' is missing from the QDict then return 0. | |
4731 | * | |
4732 | * If 'backing' contains the node name of the backing file of | |
4733 | * reopen_state->bs then return 0. | |
4734 | * | |
4735 | * If 'backing' contains a different node name (or is null) then check | |
4736 | * whether the current backing file can be replaced with the new one. | |
4737 | * If that's the case then reopen_state->replace_backing_bs is set to | |
4738 | * true and reopen_state->new_backing_bs contains a pointer to the new | |
4739 | * backing BlockDriverState (or NULL). | |
4740 | * | |
4741 | * After calling this function, the transaction @tran may only be completed | |
4742 | * while holding a writer lock for the graph. | |
4743 | * | |
4744 | * Return 0 on success, otherwise return < 0 and set @errp. | |
4745 | * | |
4746 | * The caller must hold the AioContext lock of @reopen_state->bs. | |
4747 | * @reopen_state->bs can move to a different AioContext in this function. | |
4748 | * Callers must make sure that their AioContext locking is still correct after | |
4749 | * this. | |
4750 | */ | |
4751 | static int GRAPH_UNLOCKED | |
4752 | bdrv_reopen_parse_file_or_backing(BDRVReopenState *reopen_state, | |
4753 | bool is_backing, Transaction *tran, | |
4754 | Error **errp) | |
4755 | { | |
4756 | BlockDriverState *bs = reopen_state->bs; | |
4757 | BlockDriverState *new_child_bs; | |
4758 | BlockDriverState *old_child_bs; | |
4759 | ||
4760 | const char *child_name = is_backing ? "backing" : "file"; | |
4761 | QObject *value; | |
4762 | const char *str; | |
4763 | AioContext *ctx, *old_ctx; | |
4764 | bool has_child; | |
4765 | int ret; | |
4766 | ||
4767 | GLOBAL_STATE_CODE(); | |
4768 | ||
4769 | value = qdict_get(reopen_state->options, child_name); | |
4770 | if (value == NULL) { | |
4771 | return 0; | |
4772 | } | |
4773 | ||
4774 | bdrv_graph_rdlock_main_loop(); | |
4775 | ||
4776 | switch (qobject_type(value)) { | |
4777 | case QTYPE_QNULL: | |
4778 | assert(is_backing); /* The 'file' option does not allow a null value */ | |
4779 | new_child_bs = NULL; | |
4780 | break; | |
4781 | case QTYPE_QSTRING: | |
4782 | str = qstring_get_str(qobject_to(QString, value)); | |
4783 | new_child_bs = bdrv_lookup_bs(NULL, str, errp); | |
4784 | if (new_child_bs == NULL) { | |
4785 | ret = -EINVAL; | |
4786 | goto out_rdlock; | |
4787 | } | |
4788 | ||
4789 | has_child = bdrv_recurse_has_child(new_child_bs, bs); | |
4790 | if (has_child) { | |
4791 | error_setg(errp, "Making '%s' a %s child of '%s' would create a " | |
4792 | "cycle", str, child_name, bs->node_name); | |
4793 | ret = -EINVAL; | |
4794 | goto out_rdlock; | |
4795 | } | |
4796 | break; | |
4797 | default: | |
4798 | /* | |
4799 | * The options QDict has been flattened, so 'backing' and 'file' | |
4800 | * do not allow any other data type here. | |
4801 | */ | |
4802 | g_assert_not_reached(); | |
4803 | } | |
4804 | ||
4805 | old_child_bs = is_backing ? child_bs(bs->backing) : child_bs(bs->file); | |
4806 | if (old_child_bs == new_child_bs) { | |
4807 | ret = 0; | |
4808 | goto out_rdlock; | |
4809 | } | |
4810 | ||
4811 | if (old_child_bs) { | |
4812 | if (bdrv_skip_implicit_filters(old_child_bs) == new_child_bs) { | |
4813 | ret = 0; | |
4814 | goto out_rdlock; | |
4815 | } | |
4816 | ||
4817 | if (old_child_bs->implicit) { | |
4818 | error_setg(errp, "Cannot replace implicit %s child of %s", | |
4819 | child_name, bs->node_name); | |
4820 | ret = -EPERM; | |
4821 | goto out_rdlock; | |
4822 | } | |
4823 | } | |
4824 | ||
4825 | if (bs->drv->is_filter && !old_child_bs) { | |
4826 | /* | |
4827 | * Filters always have a file or a backing child, so we are trying to | |
4828 | * change wrong child | |
4829 | */ | |
4830 | error_setg(errp, "'%s' is a %s filter node that does not support a " | |
4831 | "%s child", bs->node_name, bs->drv->format_name, child_name); | |
4832 | ret = -EINVAL; | |
4833 | goto out_rdlock; | |
4834 | } | |
4835 | ||
4836 | if (is_backing) { | |
4837 | reopen_state->old_backing_bs = old_child_bs; | |
4838 | } else { | |
4839 | reopen_state->old_file_bs = old_child_bs; | |
4840 | } | |
4841 | ||
4842 | if (old_child_bs) { | |
4843 | bdrv_ref(old_child_bs); | |
4844 | bdrv_drained_begin(old_child_bs); | |
4845 | } | |
4846 | ||
4847 | old_ctx = bdrv_get_aio_context(bs); | |
4848 | ctx = bdrv_get_aio_context(new_child_bs); | |
4849 | if (old_ctx != ctx) { | |
4850 | aio_context_release(old_ctx); | |
4851 | aio_context_acquire(ctx); | |
4852 | } | |
4853 | ||
4854 | bdrv_graph_rdunlock_main_loop(); | |
4855 | bdrv_graph_wrlock(new_child_bs); | |
4856 | ||
4857 | ret = bdrv_set_file_or_backing_noperm(bs, new_child_bs, is_backing, | |
4858 | tran, errp); | |
4859 | ||
4860 | bdrv_graph_wrunlock(); | |
4861 | ||
4862 | if (old_ctx != ctx) { | |
4863 | aio_context_release(ctx); | |
4864 | aio_context_acquire(old_ctx); | |
4865 | } | |
4866 | ||
4867 | if (old_child_bs) { | |
4868 | bdrv_drained_end(old_child_bs); | |
4869 | bdrv_unref(old_child_bs); | |
4870 | } | |
4871 | ||
4872 | return ret; | |
4873 | ||
4874 | out_rdlock: | |
4875 | bdrv_graph_rdunlock_main_loop(); | |
4876 | return ret; | |
4877 | } | |
4878 | ||
4879 | /* | |
4880 | * Prepares a BlockDriverState for reopen. All changes are staged in the | |
4881 | * 'opaque' field of the BDRVReopenState, which is used and allocated by | |
4882 | * the block driver layer .bdrv_reopen_prepare() | |
4883 | * | |
4884 | * bs is the BlockDriverState to reopen | |
4885 | * flags are the new open flags | |
4886 | * queue is the reopen queue | |
4887 | * | |
4888 | * Returns 0 on success, non-zero on error. On error errp will be set | |
4889 | * as well. | |
4890 | * | |
4891 | * On failure, bdrv_reopen_abort() will be called to clean up any data. | |
4892 | * It is the responsibility of the caller to then call the abort() or | |
4893 | * commit() for any other BDS that have been left in a prepare() state | |
4894 | * | |
4895 | * The caller must hold the AioContext lock of @reopen_state->bs. | |
4896 | * | |
4897 | * After calling this function, the transaction @change_child_tran may only be | |
4898 | * completed while holding a writer lock for the graph. | |
4899 | */ | |
4900 | static int GRAPH_UNLOCKED | |
4901 | bdrv_reopen_prepare(BDRVReopenState *reopen_state, BlockReopenQueue *queue, | |
4902 | Transaction *change_child_tran, Error **errp) | |
4903 | { | |
4904 | int ret = -1; | |
4905 | int old_flags; | |
4906 | Error *local_err = NULL; | |
4907 | BlockDriver *drv; | |
4908 | QemuOpts *opts; | |
4909 | QDict *orig_reopen_opts; | |
4910 | char *discard = NULL; | |
4911 | bool read_only; | |
4912 | bool drv_prepared = false; | |
4913 | ||
4914 | assert(reopen_state != NULL); | |
4915 | assert(reopen_state->bs->drv != NULL); | |
4916 | GLOBAL_STATE_CODE(); | |
4917 | drv = reopen_state->bs->drv; | |
4918 | ||
4919 | /* This function and each driver's bdrv_reopen_prepare() remove | |
4920 | * entries from reopen_state->options as they are processed, so | |
4921 | * we need to make a copy of the original QDict. */ | |
4922 | orig_reopen_opts = qdict_clone_shallow(reopen_state->options); | |
4923 | ||
4924 | /* Process generic block layer options */ | |
4925 | opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort); | |
4926 | if (!qemu_opts_absorb_qdict(opts, reopen_state->options, errp)) { | |
4927 | ret = -EINVAL; | |
4928 | goto error; | |
4929 | } | |
4930 | ||
4931 | /* This was already called in bdrv_reopen_queue_child() so the flags | |
4932 | * are up-to-date. This time we simply want to remove the options from | |
4933 | * QemuOpts in order to indicate that they have been processed. */ | |
4934 | old_flags = reopen_state->flags; | |
4935 | update_flags_from_options(&reopen_state->flags, opts); | |
4936 | assert(old_flags == reopen_state->flags); | |
4937 | ||
4938 | discard = qemu_opt_get_del(opts, BDRV_OPT_DISCARD); | |
4939 | if (discard != NULL) { | |
4940 | if (bdrv_parse_discard_flags(discard, &reopen_state->flags) != 0) { | |
4941 | error_setg(errp, "Invalid discard option"); | |
4942 | ret = -EINVAL; | |
4943 | goto error; | |
4944 | } | |
4945 | } | |
4946 | ||
4947 | reopen_state->detect_zeroes = | |
4948 | bdrv_parse_detect_zeroes(opts, reopen_state->flags, &local_err); | |
4949 | if (local_err) { | |
4950 | error_propagate(errp, local_err); | |
4951 | ret = -EINVAL; | |
4952 | goto error; | |
4953 | } | |
4954 | ||
4955 | /* All other options (including node-name and driver) must be unchanged. | |
4956 | * Put them back into the QDict, so that they are checked at the end | |
4957 | * of this function. */ | |
4958 | qemu_opts_to_qdict(opts, reopen_state->options); | |
4959 | ||
4960 | /* If we are to stay read-only, do not allow permission change | |
4961 | * to r/w. Attempting to set to r/w may fail if either BDRV_O_ALLOW_RDWR is | |
4962 | * not set, or if the BDS still has copy_on_read enabled */ | |
4963 | read_only = !(reopen_state->flags & BDRV_O_RDWR); | |
4964 | ||
4965 | bdrv_graph_rdlock_main_loop(); | |
4966 | ret = bdrv_can_set_read_only(reopen_state->bs, read_only, true, &local_err); | |
4967 | bdrv_graph_rdunlock_main_loop(); | |
4968 | if (local_err) { | |
4969 | error_propagate(errp, local_err); | |
4970 | goto error; | |
4971 | } | |
4972 | ||
4973 | if (drv->bdrv_reopen_prepare) { | |
4974 | /* | |
4975 | * If a driver-specific option is missing, it means that we | |
4976 | * should reset it to its default value. | |
4977 | * But not all options allow that, so we need to check it first. | |
4978 | */ | |
4979 | ret = bdrv_reset_options_allowed(reopen_state->bs, | |
4980 | reopen_state->options, errp); | |
4981 | if (ret) { | |
4982 | goto error; | |
4983 | } | |
4984 | ||
4985 | ret = drv->bdrv_reopen_prepare(reopen_state, queue, &local_err); | |
4986 | if (ret) { | |
4987 | if (local_err != NULL) { | |
4988 | error_propagate(errp, local_err); | |
4989 | } else { | |
4990 | bdrv_graph_rdlock_main_loop(); | |
4991 | bdrv_refresh_filename(reopen_state->bs); | |
4992 | bdrv_graph_rdunlock_main_loop(); | |
4993 | error_setg(errp, "failed while preparing to reopen image '%s'", | |
4994 | reopen_state->bs->filename); | |
4995 | } | |
4996 | goto error; | |
4997 | } | |
4998 | } else { | |
4999 | /* It is currently mandatory to have a bdrv_reopen_prepare() | |
5000 | * handler for each supported drv. */ | |
5001 | bdrv_graph_rdlock_main_loop(); | |
5002 | error_setg(errp, "Block format '%s' used by node '%s' " | |
5003 | "does not support reopening files", drv->format_name, | |
5004 | bdrv_get_device_or_node_name(reopen_state->bs)); | |
5005 | bdrv_graph_rdunlock_main_loop(); | |
5006 | ret = -1; | |
5007 | goto error; | |
5008 | } | |
5009 | ||
5010 | drv_prepared = true; | |
5011 | ||
5012 | /* | |
5013 | * We must provide the 'backing' option if the BDS has a backing | |
5014 | * file or if the image file has a backing file name as part of | |
5015 | * its metadata. Otherwise the 'backing' option can be omitted. | |
5016 | */ | |
5017 | bdrv_graph_rdlock_main_loop(); | |
5018 | if (drv->supports_backing && reopen_state->backing_missing && | |
5019 | (reopen_state->bs->backing || reopen_state->bs->backing_file[0])) { | |
5020 | error_setg(errp, "backing is missing for '%s'", | |
5021 | reopen_state->bs->node_name); | |
5022 | bdrv_graph_rdunlock_main_loop(); | |
5023 | ret = -EINVAL; | |
5024 | goto error; | |
5025 | } | |
5026 | bdrv_graph_rdunlock_main_loop(); | |
5027 | ||
5028 | /* | |
5029 | * Allow changing the 'backing' option. The new value can be | |
5030 | * either a reference to an existing node (using its node name) | |
5031 | * or NULL to simply detach the current backing file. | |
5032 | */ | |
5033 | ret = bdrv_reopen_parse_file_or_backing(reopen_state, true, | |
5034 | change_child_tran, errp); | |
5035 | if (ret < 0) { | |
5036 | goto error; | |
5037 | } | |
5038 | qdict_del(reopen_state->options, "backing"); | |
5039 | ||
5040 | /* Allow changing the 'file' option. In this case NULL is not allowed */ | |
5041 | ret = bdrv_reopen_parse_file_or_backing(reopen_state, false, | |
5042 | change_child_tran, errp); | |
5043 | if (ret < 0) { | |
5044 | goto error; | |
5045 | } | |
5046 | qdict_del(reopen_state->options, "file"); | |
5047 | ||
5048 | /* Options that are not handled are only okay if they are unchanged | |
5049 | * compared to the old state. It is expected that some options are only | |
5050 | * used for the initial open, but not reopen (e.g. filename) */ | |
5051 | if (qdict_size(reopen_state->options)) { | |
5052 | const QDictEntry *entry = qdict_first(reopen_state->options); | |
5053 | ||
5054 | GRAPH_RDLOCK_GUARD_MAINLOOP(); | |
5055 | ||
5056 | do { | |
5057 | QObject *new = entry->value; | |
5058 | QObject *old = qdict_get(reopen_state->bs->options, entry->key); | |
5059 | ||
5060 | /* Allow child references (child_name=node_name) as long as they | |
5061 | * point to the current child (i.e. everything stays the same). */ | |
5062 | if (qobject_type(new) == QTYPE_QSTRING) { | |
5063 | BdrvChild *child; | |
5064 | QLIST_FOREACH(child, &reopen_state->bs->children, next) { | |
5065 | if (!strcmp(child->name, entry->key)) { | |
5066 | break; | |
5067 | } | |
5068 | } | |
5069 | ||
5070 | if (child) { | |
5071 | if (!strcmp(child->bs->node_name, | |
5072 | qstring_get_str(qobject_to(QString, new)))) { | |
5073 | continue; /* Found child with this name, skip option */ | |
5074 | } | |
5075 | } | |
5076 | } | |
5077 | ||
5078 | /* | |
5079 | * TODO: When using -drive to specify blockdev options, all values | |
5080 | * will be strings; however, when using -blockdev, blockdev-add or | |
5081 | * filenames using the json:{} pseudo-protocol, they will be | |
5082 | * correctly typed. | |
5083 | * In contrast, reopening options are (currently) always strings | |
5084 | * (because you can only specify them through qemu-io; all other | |
5085 | * callers do not specify any options). | |
5086 | * Therefore, when using anything other than -drive to create a BDS, | |
5087 | * this cannot detect non-string options as unchanged, because | |
5088 | * qobject_is_equal() always returns false for objects of different | |
5089 | * type. In the future, this should be remedied by correctly typing | |
5090 | * all options. For now, this is not too big of an issue because | |
5091 | * the user can simply omit options which cannot be changed anyway, | |
5092 | * so they will stay unchanged. | |
5093 | */ | |
5094 | if (!qobject_is_equal(new, old)) { | |
5095 | error_setg(errp, "Cannot change the option '%s'", entry->key); | |
5096 | ret = -EINVAL; | |
5097 | goto error; | |
5098 | } | |
5099 | } while ((entry = qdict_next(reopen_state->options, entry))); | |
5100 | } | |
5101 | ||
5102 | ret = 0; | |
5103 | ||
5104 | /* Restore the original reopen_state->options QDict */ | |
5105 | qobject_unref(reopen_state->options); | |
5106 | reopen_state->options = qobject_ref(orig_reopen_opts); | |
5107 | ||
5108 | error: | |
5109 | if (ret < 0 && drv_prepared) { | |
5110 | /* drv->bdrv_reopen_prepare() has succeeded, so we need to | |
5111 | * call drv->bdrv_reopen_abort() before signaling an error | |
5112 | * (bdrv_reopen_multiple() will not call bdrv_reopen_abort() | |
5113 | * when the respective bdrv_reopen_prepare() has failed) */ | |
5114 | if (drv->bdrv_reopen_abort) { | |
5115 | drv->bdrv_reopen_abort(reopen_state); | |
5116 | } | |
5117 | } | |
5118 | qemu_opts_del(opts); | |
5119 | qobject_unref(orig_reopen_opts); | |
5120 | g_free(discard); | |
5121 | return ret; | |
5122 | } | |
5123 | ||
5124 | /* | |
5125 | * Takes the staged changes for the reopen from bdrv_reopen_prepare(), and | |
5126 | * makes them final by swapping the staging BlockDriverState contents into | |
5127 | * the active BlockDriverState contents. | |
5128 | */ | |
5129 | static void GRAPH_UNLOCKED bdrv_reopen_commit(BDRVReopenState *reopen_state) | |
5130 | { | |
5131 | BlockDriver *drv; | |
5132 | BlockDriverState *bs; | |
5133 | BdrvChild *child; | |
5134 | ||
5135 | assert(reopen_state != NULL); | |
5136 | bs = reopen_state->bs; | |
5137 | drv = bs->drv; | |
5138 | assert(drv != NULL); | |
5139 | GLOBAL_STATE_CODE(); | |
5140 | ||
5141 | /* If there are any driver level actions to take */ | |
5142 | if (drv->bdrv_reopen_commit) { | |
5143 | drv->bdrv_reopen_commit(reopen_state); | |
5144 | } | |
5145 | ||
5146 | GRAPH_RDLOCK_GUARD_MAINLOOP(); | |
5147 | ||
5148 | /* set BDS specific flags now */ | |
5149 | qobject_unref(bs->explicit_options); | |
5150 | qobject_unref(bs->options); | |
5151 | qobject_ref(reopen_state->explicit_options); | |
5152 | qobject_ref(reopen_state->options); | |
5153 | ||
5154 | bs->explicit_options = reopen_state->explicit_options; | |
5155 | bs->options = reopen_state->options; | |
5156 | bs->open_flags = reopen_state->flags; | |
5157 | bs->detect_zeroes = reopen_state->detect_zeroes; | |
5158 | ||
5159 | /* Remove child references from bs->options and bs->explicit_options. | |
5160 | * Child options were already removed in bdrv_reopen_queue_child() */ | |
5161 | QLIST_FOREACH(child, &bs->children, next) { | |
5162 | qdict_del(bs->explicit_options, child->name); | |
5163 | qdict_del(bs->options, child->name); | |
5164 | } | |
5165 | /* backing is probably removed, so it's not handled by previous loop */ | |
5166 | qdict_del(bs->explicit_options, "backing"); | |
5167 | qdict_del(bs->options, "backing"); | |
5168 | ||
5169 | bdrv_refresh_limits(bs, NULL, NULL); | |
5170 | bdrv_refresh_total_sectors(bs, bs->total_sectors); | |
5171 | } | |
5172 | ||
5173 | /* | |
5174 | * Abort the reopen, and delete and free the staged changes in | |
5175 | * reopen_state | |
5176 | */ | |
5177 | static void GRAPH_UNLOCKED bdrv_reopen_abort(BDRVReopenState *reopen_state) | |
5178 | { | |
5179 | BlockDriver *drv; | |
5180 | ||
5181 | assert(reopen_state != NULL); | |
5182 | drv = reopen_state->bs->drv; | |
5183 | assert(drv != NULL); | |
5184 | GLOBAL_STATE_CODE(); | |
5185 | ||
5186 | if (drv->bdrv_reopen_abort) { | |
5187 | drv->bdrv_reopen_abort(reopen_state); | |
5188 | } | |
5189 | } | |
5190 | ||
5191 | ||
5192 | static void bdrv_close(BlockDriverState *bs) | |
5193 | { | |
5194 | BdrvAioNotifier *ban, *ban_next; | |
5195 | BdrvChild *child, *next; | |
5196 | ||
5197 | GLOBAL_STATE_CODE(); | |
5198 | assert(!bs->refcnt); | |
5199 | ||
5200 | bdrv_drained_begin(bs); /* complete I/O */ | |
5201 | bdrv_flush(bs); | |
5202 | bdrv_drain(bs); /* in case flush left pending I/O */ | |
5203 | ||
5204 | if (bs->drv) { | |
5205 | if (bs->drv->bdrv_close) { | |
5206 | /* Must unfreeze all children, so bdrv_unref_child() works */ | |
5207 | bs->drv->bdrv_close(bs); | |
5208 | } | |
5209 | bs->drv = NULL; | |
5210 | } | |
5211 | ||
5212 | bdrv_graph_wrlock(bs); | |
5213 | QLIST_FOREACH_SAFE(child, &bs->children, next, next) { | |
5214 | bdrv_unref_child(bs, child); | |
5215 | } | |
5216 | ||
5217 | assert(!bs->backing); | |
5218 | assert(!bs->file); | |
5219 | bdrv_graph_wrunlock(); | |
5220 | ||
5221 | g_free(bs->opaque); | |
5222 | bs->opaque = NULL; | |
5223 | qatomic_set(&bs->copy_on_read, 0); | |
5224 | bs->backing_file[0] = '\0'; | |
5225 | bs->backing_format[0] = '\0'; | |
5226 | bs->total_sectors = 0; | |
5227 | bs->encrypted = false; | |
5228 | bs->sg = false; | |
5229 | qobject_unref(bs->options); | |
5230 | qobject_unref(bs->explicit_options); | |
5231 | bs->options = NULL; | |
5232 | bs->explicit_options = NULL; | |
5233 | qobject_unref(bs->full_open_options); | |
5234 | bs->full_open_options = NULL; | |
5235 | g_free(bs->block_status_cache); | |
5236 | bs->block_status_cache = NULL; | |
5237 | ||
5238 | bdrv_release_named_dirty_bitmaps(bs); | |
5239 | assert(QLIST_EMPTY(&bs->dirty_bitmaps)); | |
5240 | ||
5241 | QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_next) { | |
5242 | g_free(ban); | |
5243 | } | |
5244 | QLIST_INIT(&bs->aio_notifiers); | |
5245 | bdrv_drained_end(bs); | |
5246 | ||
5247 | /* | |
5248 | * If we're still inside some bdrv_drain_all_begin()/end() sections, end | |
5249 | * them now since this BDS won't exist anymore when bdrv_drain_all_end() | |
5250 | * gets called. | |
5251 | */ | |
5252 | if (bs->quiesce_counter) { | |
5253 | bdrv_drain_all_end_quiesce(bs); | |
5254 | } | |
5255 | } | |
5256 | ||
5257 | void bdrv_close_all(void) | |
5258 | { | |
5259 | GLOBAL_STATE_CODE(); | |
5260 | assert(job_next(NULL) == NULL); | |
5261 | ||
5262 | /* Drop references from requests still in flight, such as canceled block | |
5263 | * jobs whose AIO context has not been polled yet */ | |
5264 | bdrv_drain_all(); | |
5265 | ||
5266 | blk_remove_all_bs(); | |
5267 | blockdev_close_all_bdrv_states(); | |
5268 | ||
5269 | assert(QTAILQ_EMPTY(&all_bdrv_states)); | |
5270 | } | |
5271 | ||
5272 | static bool GRAPH_RDLOCK should_update_child(BdrvChild *c, BlockDriverState *to) | |
5273 | { | |
5274 | GQueue *queue; | |
5275 | GHashTable *found; | |
5276 | bool ret; | |
5277 | ||
5278 | if (c->klass->stay_at_node) { | |
5279 | return false; | |
5280 | } | |
5281 | ||
5282 | /* If the child @c belongs to the BDS @to, replacing the current | |
5283 | * c->bs by @to would mean to create a loop. | |
5284 | * | |
5285 | * Such a case occurs when appending a BDS to a backing chain. | |
5286 | * For instance, imagine the following chain: | |
5287 | * | |
5288 | * guest device -> node A -> further backing chain... | |
5289 | * | |
5290 | * Now we create a new BDS B which we want to put on top of this | |
5291 | * chain, so we first attach A as its backing node: | |
5292 | * | |
5293 | * node B | |
5294 | * | | |
5295 | * v | |
5296 | * guest device -> node A -> further backing chain... | |
5297 | * | |
5298 | * Finally we want to replace A by B. When doing that, we want to | |
5299 | * replace all pointers to A by pointers to B -- except for the | |
5300 | * pointer from B because (1) that would create a loop, and (2) | |
5301 | * that pointer should simply stay intact: | |
5302 | * | |
5303 | * guest device -> node B | |
5304 | * | | |
5305 | * v | |
5306 | * node A -> further backing chain... | |
5307 | * | |
5308 | * In general, when replacing a node A (c->bs) by a node B (@to), | |
5309 | * if A is a child of B, that means we cannot replace A by B there | |
5310 | * because that would create a loop. Silently detaching A from B | |
5311 | * is also not really an option. So overall just leaving A in | |
5312 | * place there is the most sensible choice. | |
5313 | * | |
5314 | * We would also create a loop in any cases where @c is only | |
5315 | * indirectly referenced by @to. Prevent this by returning false | |
5316 | * if @c is found (by breadth-first search) anywhere in the whole | |
5317 | * subtree of @to. | |
5318 | */ | |
5319 | ||
5320 | ret = true; | |
5321 | found = g_hash_table_new(NULL, NULL); | |
5322 | g_hash_table_add(found, to); | |
5323 | queue = g_queue_new(); | |
5324 | g_queue_push_tail(queue, to); | |
5325 | ||
5326 | while (!g_queue_is_empty(queue)) { | |
5327 | BlockDriverState *v = g_queue_pop_head(queue); | |
5328 | BdrvChild *c2; | |
5329 | ||
5330 | QLIST_FOREACH(c2, &v->children, next) { | |
5331 | if (c2 == c) { | |
5332 | ret = false; | |
5333 | break; | |
5334 | } | |
5335 | ||
5336 | if (g_hash_table_contains(found, c2->bs)) { | |
5337 | continue; | |
5338 | } | |
5339 | ||
5340 | g_queue_push_tail(queue, c2->bs); | |
5341 | g_hash_table_add(found, c2->bs); | |
5342 | } | |
5343 | } | |
5344 | ||
5345 | g_queue_free(queue); | |
5346 | g_hash_table_destroy(found); | |
5347 | ||
5348 | return ret; | |
5349 | } | |
5350 | ||
5351 | static void bdrv_remove_child_commit(void *opaque) | |
5352 | { | |
5353 | GLOBAL_STATE_CODE(); | |
5354 | bdrv_child_free(opaque); | |
5355 | } | |
5356 | ||
5357 | static TransactionActionDrv bdrv_remove_child_drv = { | |
5358 | .commit = bdrv_remove_child_commit, | |
5359 | }; | |
5360 | ||
5361 | /* | |
5362 | * Function doesn't update permissions, caller is responsible for this. | |
5363 | * | |
5364 | * @child->bs (if non-NULL) must be drained. | |
5365 | * | |
5366 | * After calling this function, the transaction @tran may only be completed | |
5367 | * while holding a writer lock for the graph. | |
5368 | */ | |
5369 | static void GRAPH_WRLOCK bdrv_remove_child(BdrvChild *child, Transaction *tran) | |
5370 | { | |
5371 | if (!child) { | |
5372 | return; | |
5373 | } | |
5374 | ||
5375 | if (child->bs) { | |
5376 | assert(child->quiesced_parent); | |
5377 | bdrv_replace_child_tran(child, NULL, tran); | |
5378 | } | |
5379 | ||
5380 | tran_add(tran, &bdrv_remove_child_drv, child); | |
5381 | } | |
5382 | ||
5383 | /* | |
5384 | * Both @from and @to (if non-NULL) must be drained. @to must be kept drained | |
5385 | * until the transaction is completed. | |
5386 | * | |
5387 | * After calling this function, the transaction @tran may only be completed | |
5388 | * while holding a writer lock for the graph. | |
5389 | */ | |
5390 | static int GRAPH_WRLOCK | |
5391 | bdrv_replace_node_noperm(BlockDriverState *from, | |
5392 | BlockDriverState *to, | |
5393 | bool auto_skip, Transaction *tran, | |
5394 | Error **errp) | |
5395 | { | |
5396 | BdrvChild *c, *next; | |
5397 | ||
5398 | GLOBAL_STATE_CODE(); | |
5399 | ||
5400 | assert(from->quiesce_counter); | |
5401 | assert(to->quiesce_counter); | |
5402 | ||
5403 | QLIST_FOREACH_SAFE(c, &from->parents, next_parent, next) { | |
5404 | assert(c->bs == from); | |
5405 | if (!should_update_child(c, to)) { | |
5406 | if (auto_skip) { | |
5407 | continue; | |
5408 | } | |
5409 | error_setg(errp, "Should not change '%s' link to '%s'", | |
5410 | c->name, from->node_name); | |
5411 | return -EINVAL; | |
5412 | } | |
5413 | if (c->frozen) { | |
5414 | error_setg(errp, "Cannot change '%s' link to '%s'", | |
5415 | c->name, from->node_name); | |
5416 | return -EPERM; | |
5417 | } | |
5418 | bdrv_replace_child_tran(c, to, tran); | |
5419 | } | |
5420 | ||
5421 | return 0; | |
5422 | } | |
5423 | ||
5424 | /* | |
5425 | * Switch all parents of @from to point to @to instead. @from and @to must be in | |
5426 | * the same AioContext and both must be drained. | |
5427 | * | |
5428 | * With auto_skip=true bdrv_replace_node_common skips updating from parents | |
5429 | * if it creates a parent-child relation loop or if parent is block-job. | |
5430 | * | |
5431 | * With auto_skip=false the error is returned if from has a parent which should | |
5432 | * not be updated. | |
5433 | * | |
5434 | * With @detach_subchain=true @to must be in a backing chain of @from. In this | |
5435 | * case backing link of the cow-parent of @to is removed. | |
5436 | */ | |
5437 | static int GRAPH_WRLOCK | |
5438 | bdrv_replace_node_common(BlockDriverState *from, BlockDriverState *to, | |
5439 | bool auto_skip, bool detach_subchain, Error **errp) | |
5440 | { | |
5441 | Transaction *tran = tran_new(); | |
5442 | g_autoptr(GSList) refresh_list = NULL; | |
5443 | BlockDriverState *to_cow_parent = NULL; | |
5444 | int ret; | |
5445 | ||
5446 | GLOBAL_STATE_CODE(); | |
5447 | ||
5448 | assert(from->quiesce_counter); | |
5449 | assert(to->quiesce_counter); | |
5450 | assert(bdrv_get_aio_context(from) == bdrv_get_aio_context(to)); | |
5451 | ||
5452 | if (detach_subchain) { | |
5453 | assert(bdrv_chain_contains(from, to)); | |
5454 | assert(from != to); | |
5455 | for (to_cow_parent = from; | |
5456 | bdrv_filter_or_cow_bs(to_cow_parent) != to; | |
5457 | to_cow_parent = bdrv_filter_or_cow_bs(to_cow_parent)) | |
5458 | { | |
5459 | ; | |
5460 | } | |
5461 | } | |
5462 | ||
5463 | /* | |
5464 | * Do the replacement without permission update. | |
5465 | * Replacement may influence the permissions, we should calculate new | |
5466 | * permissions based on new graph. If we fail, we'll roll-back the | |
5467 | * replacement. | |
5468 | */ | |
5469 | ret = bdrv_replace_node_noperm(from, to, auto_skip, tran, errp); | |
5470 | if (ret < 0) { | |
5471 | goto out; | |
5472 | } | |
5473 | ||
5474 | if (detach_subchain) { | |
5475 | /* to_cow_parent is already drained because from is drained */ | |
5476 | bdrv_remove_child(bdrv_filter_or_cow_child(to_cow_parent), tran); | |
5477 | } | |
5478 | ||
5479 | refresh_list = g_slist_prepend(refresh_list, to); | |
5480 | refresh_list = g_slist_prepend(refresh_list, from); | |
5481 | ||
5482 | ret = bdrv_list_refresh_perms(refresh_list, NULL, tran, errp); | |
5483 | if (ret < 0) { | |
5484 | goto out; | |
5485 | } | |
5486 | ||
5487 | ret = 0; | |
5488 | ||
5489 | out: | |
5490 | tran_finalize(tran, ret); | |
5491 | return ret; | |
5492 | } | |
5493 | ||
5494 | int bdrv_replace_node(BlockDriverState *from, BlockDriverState *to, | |
5495 | Error **errp) | |
5496 | { | |
5497 | return bdrv_replace_node_common(from, to, true, false, errp); | |
5498 | } | |
5499 | ||
5500 | int bdrv_drop_filter(BlockDriverState *bs, Error **errp) | |
5501 | { | |
5502 | BlockDriverState *child_bs; | |
5503 | int ret; | |
5504 | ||
5505 | GLOBAL_STATE_CODE(); | |
5506 | ||
5507 | bdrv_graph_rdlock_main_loop(); | |
5508 | child_bs = bdrv_filter_or_cow_bs(bs); | |
5509 | bdrv_graph_rdunlock_main_loop(); | |
5510 | ||
5511 | bdrv_drained_begin(child_bs); | |
5512 | bdrv_graph_wrlock(bs); | |
5513 | ret = bdrv_replace_node_common(bs, child_bs, true, true, errp); | |
5514 | bdrv_graph_wrunlock(); | |
5515 | bdrv_drained_end(child_bs); | |
5516 | ||
5517 | return ret; | |
5518 | } | |
5519 | ||
5520 | /* | |
5521 | * Add new bs contents at the top of an image chain while the chain is | |
5522 | * live, while keeping required fields on the top layer. | |
5523 | * | |
5524 | * This will modify the BlockDriverState fields, and swap contents | |
5525 | * between bs_new and bs_top. Both bs_new and bs_top are modified. | |
5526 | * | |
5527 | * bs_new must not be attached to a BlockBackend and must not have backing | |
5528 | * child. | |
5529 | * | |
5530 | * This function does not create any image files. | |
5531 | * | |
5532 | * The caller must hold the AioContext lock for @bs_top. | |
5533 | */ | |
5534 | int bdrv_append(BlockDriverState *bs_new, BlockDriverState *bs_top, | |
5535 | Error **errp) | |
5536 | { | |
5537 | int ret; | |
5538 | BdrvChild *child; | |
5539 | Transaction *tran = tran_new(); | |
5540 | AioContext *old_context, *new_context = NULL; | |
5541 | ||
5542 | GLOBAL_STATE_CODE(); | |
5543 | ||
5544 | bdrv_graph_rdlock_main_loop(); | |
5545 | assert(!bs_new->backing); | |
5546 | bdrv_graph_rdunlock_main_loop(); | |
5547 | ||
5548 | old_context = bdrv_get_aio_context(bs_top); | |
5549 | bdrv_drained_begin(bs_top); | |
5550 | ||
5551 | /* | |
5552 | * bdrv_drained_begin() requires that only the AioContext of the drained | |
5553 | * node is locked, and at this point it can still differ from the AioContext | |
5554 | * of bs_top. | |
5555 | */ | |
5556 | new_context = bdrv_get_aio_context(bs_new); | |
5557 | aio_context_release(old_context); | |
5558 | aio_context_acquire(new_context); | |
5559 | bdrv_drained_begin(bs_new); | |
5560 | aio_context_release(new_context); | |
5561 | aio_context_acquire(old_context); | |
5562 | new_context = NULL; | |
5563 | ||
5564 | bdrv_graph_wrlock(bs_top); | |
5565 | ||
5566 | child = bdrv_attach_child_noperm(bs_new, bs_top, "backing", | |
5567 | &child_of_bds, bdrv_backing_role(bs_new), | |
5568 | tran, errp); | |
5569 | if (!child) { | |
5570 | ret = -EINVAL; | |
5571 | goto out; | |
5572 | } | |
5573 | ||
5574 | /* | |
5575 | * bdrv_attach_child_noperm could change the AioContext of bs_top and | |
5576 | * bs_new, but at least they are in the same AioContext now. This is the | |
5577 | * AioContext that we need to lock for the rest of the function. | |
5578 | */ | |
5579 | new_context = bdrv_get_aio_context(bs_top); | |
5580 | ||
5581 | if (old_context != new_context) { | |
5582 | aio_context_release(old_context); | |
5583 | aio_context_acquire(new_context); | |
5584 | } | |
5585 | ||
5586 | ret = bdrv_replace_node_noperm(bs_top, bs_new, true, tran, errp); | |
5587 | if (ret < 0) { | |
5588 | goto out; | |
5589 | } | |
5590 | ||
5591 | ret = bdrv_refresh_perms(bs_new, tran, errp); | |
5592 | out: | |
5593 | tran_finalize(tran, ret); | |
5594 | ||
5595 | bdrv_refresh_limits(bs_top, NULL, NULL); | |
5596 | bdrv_graph_wrunlock(); | |
5597 | ||
5598 | bdrv_drained_end(bs_top); | |
5599 | bdrv_drained_end(bs_new); | |
5600 | ||
5601 | if (new_context && old_context != new_context) { | |
5602 | aio_context_release(new_context); | |
5603 | aio_context_acquire(old_context); | |
5604 | } | |
5605 | ||
5606 | return ret; | |
5607 | } | |
5608 | ||
5609 | /* Not for empty child */ | |
5610 | int bdrv_replace_child_bs(BdrvChild *child, BlockDriverState *new_bs, | |
5611 | Error **errp) | |
5612 | { | |
5613 | int ret; | |
5614 | Transaction *tran = tran_new(); | |
5615 | g_autoptr(GSList) refresh_list = NULL; | |
5616 | BlockDriverState *old_bs = child->bs; | |
5617 | ||
5618 | GLOBAL_STATE_CODE(); | |
5619 | ||
5620 | bdrv_ref(old_bs); | |
5621 | bdrv_drained_begin(old_bs); | |
5622 | bdrv_drained_begin(new_bs); | |
5623 | bdrv_graph_wrlock(new_bs); | |
5624 | ||
5625 | bdrv_replace_child_tran(child, new_bs, tran); | |
5626 | ||
5627 | refresh_list = g_slist_prepend(refresh_list, old_bs); | |
5628 | refresh_list = g_slist_prepend(refresh_list, new_bs); | |
5629 | ||
5630 | ret = bdrv_list_refresh_perms(refresh_list, NULL, tran, errp); | |
5631 | ||
5632 | tran_finalize(tran, ret); | |
5633 | ||
5634 | bdrv_graph_wrunlock(); | |
5635 | bdrv_drained_end(old_bs); | |
5636 | bdrv_drained_end(new_bs); | |
5637 | bdrv_unref(old_bs); | |
5638 | ||
5639 | return ret; | |
5640 | } | |
5641 | ||
5642 | static void bdrv_delete(BlockDriverState *bs) | |
5643 | { | |
5644 | assert(bdrv_op_blocker_is_empty(bs)); | |
5645 | assert(!bs->refcnt); | |
5646 | GLOBAL_STATE_CODE(); | |
5647 | ||
5648 | /* remove from list, if necessary */ | |
5649 | if (bs->node_name[0] != '\0') { | |
5650 | QTAILQ_REMOVE(&graph_bdrv_states, bs, node_list); | |
5651 | } | |
5652 | QTAILQ_REMOVE(&all_bdrv_states, bs, bs_list); | |
5653 | ||
5654 | bdrv_close(bs); | |
5655 | ||
5656 | qemu_mutex_destroy(&bs->reqs_lock); | |
5657 | ||
5658 | g_free(bs); | |
5659 | } | |
5660 | ||
5661 | ||
5662 | /* | |
5663 | * Replace @bs by newly created block node. | |
5664 | * | |
5665 | * @options is a QDict of options to pass to the block drivers, or NULL for an | |
5666 | * empty set of options. The reference to the QDict belongs to the block layer | |
5667 | * after the call (even on failure), so if the caller intends to reuse the | |
5668 | * dictionary, it needs to use qobject_ref() before calling bdrv_open. | |
5669 | * | |
5670 | * The caller holds the AioContext lock for @bs. It must make sure that @bs | |
5671 | * stays in the same AioContext, i.e. @options must not refer to nodes in a | |
5672 | * different AioContext. | |
5673 | */ | |
5674 | BlockDriverState *bdrv_insert_node(BlockDriverState *bs, QDict *options, | |
5675 | int flags, Error **errp) | |
5676 | { | |
5677 | ERRP_GUARD(); | |
5678 | int ret; | |
5679 | AioContext *ctx = bdrv_get_aio_context(bs); | |
5680 | BlockDriverState *new_node_bs = NULL; | |
5681 | const char *drvname, *node_name; | |
5682 | BlockDriver *drv; | |
5683 | ||
5684 | drvname = qdict_get_try_str(options, "driver"); | |
5685 | if (!drvname) { | |
5686 | error_setg(errp, "driver is not specified"); | |
5687 | goto fail; | |
5688 | } | |
5689 | ||
5690 | drv = bdrv_find_format(drvname); | |
5691 | if (!drv) { | |
5692 | error_setg(errp, "Unknown driver: '%s'", drvname); | |
5693 | goto fail; | |
5694 | } | |
5695 | ||
5696 | node_name = qdict_get_try_str(options, "node-name"); | |
5697 | ||
5698 | GLOBAL_STATE_CODE(); | |
5699 | ||
5700 | aio_context_release(ctx); | |
5701 | aio_context_acquire(qemu_get_aio_context()); | |
5702 | new_node_bs = bdrv_new_open_driver_opts(drv, node_name, options, flags, | |
5703 | errp); | |
5704 | aio_context_release(qemu_get_aio_context()); | |
5705 | aio_context_acquire(ctx); | |
5706 | assert(bdrv_get_aio_context(bs) == ctx); | |
5707 | ||
5708 | options = NULL; /* bdrv_new_open_driver() eats options */ | |
5709 | if (!new_node_bs) { | |
5710 | error_prepend(errp, "Could not create node: "); | |
5711 | goto fail; | |
5712 | } | |
5713 | ||
5714 | /* | |
5715 | * Make sure that @bs doesn't go away until we have successfully attached | |
5716 | * all of its parents to @new_node_bs and undrained it again. | |
5717 | */ | |
5718 | bdrv_ref(bs); | |
5719 | bdrv_drained_begin(bs); | |
5720 | bdrv_drained_begin(new_node_bs); | |
5721 | bdrv_graph_wrlock(new_node_bs); | |
5722 | ret = bdrv_replace_node(bs, new_node_bs, errp); | |
5723 | bdrv_graph_wrunlock(); | |
5724 | bdrv_drained_end(new_node_bs); | |
5725 | bdrv_drained_end(bs); | |
5726 | bdrv_unref(bs); | |
5727 | ||
5728 | if (ret < 0) { | |
5729 | error_prepend(errp, "Could not replace node: "); | |
5730 | goto fail; | |
5731 | } | |
5732 | ||
5733 | return new_node_bs; | |
5734 | ||
5735 | fail: | |
5736 | qobject_unref(options); | |
5737 | bdrv_unref(new_node_bs); | |
5738 | return NULL; | |
5739 | } | |
5740 | ||
5741 | /* | |
5742 | * Run consistency checks on an image | |
5743 | * | |
5744 | * Returns 0 if the check could be completed (it doesn't mean that the image is | |
5745 | * free of errors) or -errno when an internal error occurred. The results of the | |
5746 | * check are stored in res. | |
5747 | */ | |
5748 | int coroutine_fn bdrv_co_check(BlockDriverState *bs, | |
5749 | BdrvCheckResult *res, BdrvCheckMode fix) | |
5750 | { | |
5751 | IO_CODE(); | |
5752 | assert_bdrv_graph_readable(); | |
5753 | if (bs->drv == NULL) { | |
5754 | return -ENOMEDIUM; | |
5755 | } | |
5756 | if (bs->drv->bdrv_co_check == NULL) { | |
5757 | return -ENOTSUP; | |
5758 | } | |
5759 | ||
5760 | memset(res, 0, sizeof(*res)); | |
5761 | return bs->drv->bdrv_co_check(bs, res, fix); | |
5762 | } | |
5763 | ||
5764 | /* | |
5765 | * Return values: | |
5766 | * 0 - success | |
5767 | * -EINVAL - backing format specified, but no file | |
5768 | * -ENOSPC - can't update the backing file because no space is left in the | |
5769 | * image file header | |
5770 | * -ENOTSUP - format driver doesn't support changing the backing file | |
5771 | */ | |
5772 | int coroutine_fn | |
5773 | bdrv_co_change_backing_file(BlockDriverState *bs, const char *backing_file, | |
5774 | const char *backing_fmt, bool require) | |
5775 | { | |
5776 | BlockDriver *drv = bs->drv; | |
5777 | int ret; | |
5778 | ||
5779 | IO_CODE(); | |
5780 | ||
5781 | if (!drv) { | |
5782 | return -ENOMEDIUM; | |
5783 | } | |
5784 | ||
5785 | /* Backing file format doesn't make sense without a backing file */ | |
5786 | if (backing_fmt && !backing_file) { | |
5787 | return -EINVAL; | |
5788 | } | |
5789 | ||
5790 | if (require && backing_file && !backing_fmt) { | |
5791 | return -EINVAL; | |
5792 | } | |
5793 | ||
5794 | if (drv->bdrv_co_change_backing_file != NULL) { | |
5795 | ret = drv->bdrv_co_change_backing_file(bs, backing_file, backing_fmt); | |
5796 | } else { | |
5797 | ret = -ENOTSUP; | |
5798 | } | |
5799 | ||
5800 | if (ret == 0) { | |
5801 | pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_file ?: ""); | |
5802 | pstrcpy(bs->backing_format, sizeof(bs->backing_format), backing_fmt ?: ""); | |
5803 | pstrcpy(bs->auto_backing_file, sizeof(bs->auto_backing_file), | |
5804 | backing_file ?: ""); | |
5805 | } | |
5806 | return ret; | |
5807 | } | |
5808 | ||
5809 | /* | |
5810 | * Finds the first non-filter node above bs in the chain between | |
5811 | * active and bs. The returned node is either an immediate parent of | |
5812 | * bs, or there are only filter nodes between the two. | |
5813 | * | |
5814 | * Returns NULL if bs is not found in active's image chain, | |
5815 | * or if active == bs. | |
5816 | * | |
5817 | * Returns the bottommost base image if bs == NULL. | |
5818 | */ | |
5819 | BlockDriverState *bdrv_find_overlay(BlockDriverState *active, | |
5820 | BlockDriverState *bs) | |
5821 | { | |
5822 | ||
5823 | GLOBAL_STATE_CODE(); | |
5824 | ||
5825 | bs = bdrv_skip_filters(bs); | |
5826 | active = bdrv_skip_filters(active); | |
5827 | ||
5828 | while (active) { | |
5829 | BlockDriverState *next = bdrv_backing_chain_next(active); | |
5830 | if (bs == next) { | |
5831 | return active; | |
5832 | } | |
5833 | active = next; | |
5834 | } | |
5835 | ||
5836 | return NULL; | |
5837 | } | |
5838 | ||
5839 | /* Given a BDS, searches for the base layer. */ | |
5840 | BlockDriverState *bdrv_find_base(BlockDriverState *bs) | |
5841 | { | |
5842 | GLOBAL_STATE_CODE(); | |
5843 | ||
5844 | return bdrv_find_overlay(bs, NULL); | |
5845 | } | |
5846 | ||
5847 | /* | |
5848 | * Return true if at least one of the COW (backing) and filter links | |
5849 | * between @bs and @base is frozen. @errp is set if that's the case. | |
5850 | * @base must be reachable from @bs, or NULL. | |
5851 | */ | |
5852 | static bool GRAPH_RDLOCK | |
5853 | bdrv_is_backing_chain_frozen(BlockDriverState *bs, BlockDriverState *base, | |
5854 | Error **errp) | |
5855 | { | |
5856 | BlockDriverState *i; | |
5857 | BdrvChild *child; | |
5858 | ||
5859 | GLOBAL_STATE_CODE(); | |
5860 | ||
5861 | for (i = bs; i != base; i = child_bs(child)) { | |
5862 | child = bdrv_filter_or_cow_child(i); | |
5863 | ||
5864 | if (child && child->frozen) { | |
5865 | error_setg(errp, "Cannot change '%s' link from '%s' to '%s'", | |
5866 | child->name, i->node_name, child->bs->node_name); | |
5867 | return true; | |
5868 | } | |
5869 | } | |
5870 | ||
5871 | return false; | |
5872 | } | |
5873 | ||
5874 | /* | |
5875 | * Freeze all COW (backing) and filter links between @bs and @base. | |
5876 | * If any of the links is already frozen the operation is aborted and | |
5877 | * none of the links are modified. | |
5878 | * @base must be reachable from @bs, or NULL. | |
5879 | * Returns 0 on success. On failure returns < 0 and sets @errp. | |
5880 | */ | |
5881 | int bdrv_freeze_backing_chain(BlockDriverState *bs, BlockDriverState *base, | |
5882 | Error **errp) | |
5883 | { | |
5884 | BlockDriverState *i; | |
5885 | BdrvChild *child; | |
5886 | ||
5887 | GLOBAL_STATE_CODE(); | |
5888 | ||
5889 | if (bdrv_is_backing_chain_frozen(bs, base, errp)) { | |
5890 | return -EPERM; | |
5891 | } | |
5892 | ||
5893 | for (i = bs; i != base; i = child_bs(child)) { | |
5894 | child = bdrv_filter_or_cow_child(i); | |
5895 | if (child && child->bs->never_freeze) { | |
5896 | error_setg(errp, "Cannot freeze '%s' link to '%s'", | |
5897 | child->name, child->bs->node_name); | |
5898 | return -EPERM; | |
5899 | } | |
5900 | } | |
5901 | ||
5902 | for (i = bs; i != base; i = child_bs(child)) { | |
5903 | child = bdrv_filter_or_cow_child(i); | |
5904 | if (child) { | |
5905 | child->frozen = true; | |
5906 | } | |
5907 | } | |
5908 | ||
5909 | return 0; | |
5910 | } | |
5911 | ||
5912 | /* | |
5913 | * Unfreeze all COW (backing) and filter links between @bs and @base. | |
5914 | * The caller must ensure that all links are frozen before using this | |
5915 | * function. | |
5916 | * @base must be reachable from @bs, or NULL. | |
5917 | */ | |
5918 | void bdrv_unfreeze_backing_chain(BlockDriverState *bs, BlockDriverState *base) | |
5919 | { | |
5920 | BlockDriverState *i; | |
5921 | BdrvChild *child; | |
5922 | ||
5923 | GLOBAL_STATE_CODE(); | |
5924 | ||
5925 | for (i = bs; i != base; i = child_bs(child)) { | |
5926 | child = bdrv_filter_or_cow_child(i); | |
5927 | if (child) { | |
5928 | assert(child->frozen); | |
5929 | child->frozen = false; | |
5930 | } | |
5931 | } | |
5932 | } | |
5933 | ||
5934 | /* | |
5935 | * Drops images above 'base' up to and including 'top', and sets the image | |
5936 | * above 'top' to have base as its backing file. | |
5937 | * | |
5938 | * Requires that the overlay to 'top' is opened r/w, so that the backing file | |
5939 | * information in 'bs' can be properly updated. | |
5940 | * | |
5941 | * E.g., this will convert the following chain: | |
5942 | * bottom <- base <- intermediate <- top <- active | |
5943 | * | |
5944 | * to | |
5945 | * | |
5946 | * bottom <- base <- active | |
5947 | * | |
5948 | * It is allowed for bottom==base, in which case it converts: | |
5949 | * | |
5950 | * base <- intermediate <- top <- active | |
5951 | * | |
5952 | * to | |
5953 | * | |
5954 | * base <- active | |
5955 | * | |
5956 | * If backing_file_str is non-NULL, it will be used when modifying top's | |
5957 | * overlay image metadata. | |
5958 | * | |
5959 | * Error conditions: | |
5960 | * if active == top, that is considered an error | |
5961 | * | |
5962 | */ | |
5963 | int bdrv_drop_intermediate(BlockDriverState *top, BlockDriverState *base, | |
5964 | const char *backing_file_str) | |
5965 | { | |
5966 | BlockDriverState *explicit_top = top; | |
5967 | bool update_inherits_from; | |
5968 | BdrvChild *c; | |
5969 | Error *local_err = NULL; | |
5970 | int ret = -EIO; | |
5971 | g_autoptr(GSList) updated_children = NULL; | |
5972 | GSList *p; | |
5973 | ||
5974 | GLOBAL_STATE_CODE(); | |
5975 | ||
5976 | bdrv_ref(top); | |
5977 | bdrv_drained_begin(base); | |
5978 | bdrv_graph_wrlock(base); | |
5979 | ||
5980 | if (!top->drv || !base->drv) { | |
5981 | goto exit_wrlock; | |
5982 | } | |
5983 | ||
5984 | /* Make sure that base is in the backing chain of top */ | |
5985 | if (!bdrv_chain_contains(top, base)) { | |
5986 | goto exit_wrlock; | |
5987 | } | |
5988 | ||
5989 | /* If 'base' recursively inherits from 'top' then we should set | |
5990 | * base->inherits_from to top->inherits_from after 'top' and all | |
5991 | * other intermediate nodes have been dropped. | |
5992 | * If 'top' is an implicit node (e.g. "commit_top") we should skip | |
5993 | * it because no one inherits from it. We use explicit_top for that. */ | |
5994 | explicit_top = bdrv_skip_implicit_filters(explicit_top); | |
5995 | update_inherits_from = bdrv_inherits_from_recursive(base, explicit_top); | |
5996 | ||
5997 | /* success - we can delete the intermediate states, and link top->base */ | |
5998 | if (!backing_file_str) { | |
5999 | bdrv_refresh_filename(base); | |
6000 | backing_file_str = base->filename; | |
6001 | } | |
6002 | ||
6003 | QLIST_FOREACH(c, &top->parents, next_parent) { | |
6004 | updated_children = g_slist_prepend(updated_children, c); | |
6005 | } | |
6006 | ||
6007 | /* | |
6008 | * It seems correct to pass detach_subchain=true here, but it triggers | |
6009 | * one more yet not fixed bug, when due to nested aio_poll loop we switch to | |
6010 | * another drained section, which modify the graph (for example, removing | |
6011 | * the child, which we keep in updated_children list). So, it's a TODO. | |
6012 | * | |
6013 | * Note, bug triggered if pass detach_subchain=true here and run | |
6014 | * test-bdrv-drain. test_drop_intermediate_poll() test-case will crash. | |
6015 | * That's a FIXME. | |
6016 | */ | |
6017 | bdrv_replace_node_common(top, base, false, false, &local_err); | |
6018 | bdrv_graph_wrunlock(); | |
6019 | ||
6020 | if (local_err) { | |
6021 | error_report_err(local_err); | |
6022 | goto exit; | |
6023 | } | |
6024 | ||
6025 | for (p = updated_children; p; p = p->next) { | |
6026 | c = p->data; | |
6027 | ||
6028 | if (c->klass->update_filename) { | |
6029 | ret = c->klass->update_filename(c, base, backing_file_str, | |
6030 | &local_err); | |
6031 | if (ret < 0) { | |
6032 | /* | |
6033 | * TODO: Actually, we want to rollback all previous iterations | |
6034 | * of this loop, and (which is almost impossible) previous | |
6035 | * bdrv_replace_node()... | |
6036 | * | |
6037 | * Note, that c->klass->update_filename may lead to permission | |
6038 | * update, so it's a bad idea to call it inside permission | |
6039 | * update transaction of bdrv_replace_node. | |
6040 | */ | |
6041 | error_report_err(local_err); | |
6042 | goto exit; | |
6043 | } | |
6044 | } | |
6045 | } | |
6046 | ||
6047 | if (update_inherits_from) { | |
6048 | base->inherits_from = explicit_top->inherits_from; | |
6049 | } | |
6050 | ||
6051 | ret = 0; | |
6052 | goto exit; | |
6053 | ||
6054 | exit_wrlock: | |
6055 | bdrv_graph_wrunlock(); | |
6056 | exit: | |
6057 | bdrv_drained_end(base); | |
6058 | bdrv_unref(top); | |
6059 | return ret; | |
6060 | } | |
6061 | ||
6062 | /** | |
6063 | * Implementation of BlockDriver.bdrv_co_get_allocated_file_size() that | |
6064 | * sums the size of all data-bearing children. (This excludes backing | |
6065 | * children.) | |
6066 | */ | |
6067 | static int64_t coroutine_fn GRAPH_RDLOCK | |
6068 | bdrv_sum_allocated_file_size(BlockDriverState *bs) | |
6069 | { | |
6070 | BdrvChild *child; | |
6071 | int64_t child_size, sum = 0; | |
6072 | ||
6073 | QLIST_FOREACH(child, &bs->children, next) { | |
6074 | if (child->role & (BDRV_CHILD_DATA | BDRV_CHILD_METADATA | | |
6075 | BDRV_CHILD_FILTERED)) | |
6076 | { | |
6077 | child_size = bdrv_co_get_allocated_file_size(child->bs); | |
6078 | if (child_size < 0) { | |
6079 | return child_size; | |
6080 | } | |
6081 | sum += child_size; | |
6082 | } | |
6083 | } | |
6084 | ||
6085 | return sum; | |
6086 | } | |
6087 | ||
6088 | /** | |
6089 | * Length of a allocated file in bytes. Sparse files are counted by actual | |
6090 | * allocated space. Return < 0 if error or unknown. | |
6091 | */ | |
6092 | int64_t coroutine_fn bdrv_co_get_allocated_file_size(BlockDriverState *bs) | |
6093 | { | |
6094 | BlockDriver *drv = bs->drv; | |
6095 | IO_CODE(); | |
6096 | assert_bdrv_graph_readable(); | |
6097 | ||
6098 | if (!drv) { | |
6099 | return -ENOMEDIUM; | |
6100 | } | |
6101 | if (drv->bdrv_co_get_allocated_file_size) { | |
6102 | return drv->bdrv_co_get_allocated_file_size(bs); | |
6103 | } | |
6104 | ||
6105 | if (drv->bdrv_file_open) { | |
6106 | /* | |
6107 | * Protocol drivers default to -ENOTSUP (most of their data is | |
6108 | * not stored in any of their children (if they even have any), | |
6109 | * so there is no generic way to figure it out). | |
6110 | */ | |
6111 | return -ENOTSUP; | |
6112 | } else if (drv->is_filter) { | |
6113 | /* Filter drivers default to the size of their filtered child */ | |
6114 | return bdrv_co_get_allocated_file_size(bdrv_filter_bs(bs)); | |
6115 | } else { | |
6116 | /* Other drivers default to summing their children's sizes */ | |
6117 | return bdrv_sum_allocated_file_size(bs); | |
6118 | } | |
6119 | } | |
6120 | ||
6121 | /* | |
6122 | * bdrv_measure: | |
6123 | * @drv: Format driver | |
6124 | * @opts: Creation options for new image | |
6125 | * @in_bs: Existing image containing data for new image (may be NULL) | |
6126 | * @errp: Error object | |
6127 | * Returns: A #BlockMeasureInfo (free using qapi_free_BlockMeasureInfo()) | |
6128 | * or NULL on error | |
6129 | * | |
6130 | * Calculate file size required to create a new image. | |
6131 | * | |
6132 | * If @in_bs is given then space for allocated clusters and zero clusters | |
6133 | * from that image are included in the calculation. If @opts contains a | |
6134 | * backing file that is shared by @in_bs then backing clusters may be omitted | |
6135 | * from the calculation. | |
6136 | * | |
6137 | * If @in_bs is NULL then the calculation includes no allocated clusters | |
6138 | * unless a preallocation option is given in @opts. | |
6139 | * | |
6140 | * Note that @in_bs may use a different BlockDriver from @drv. | |
6141 | * | |
6142 | * If an error occurs the @errp pointer is set. | |
6143 | */ | |
6144 | BlockMeasureInfo *bdrv_measure(BlockDriver *drv, QemuOpts *opts, | |
6145 | BlockDriverState *in_bs, Error **errp) | |
6146 | { | |
6147 | IO_CODE(); | |
6148 | if (!drv->bdrv_measure) { | |
6149 | error_setg(errp, "Block driver '%s' does not support size measurement", | |
6150 | drv->format_name); | |
6151 | return NULL; | |
6152 | } | |
6153 | ||
6154 | return drv->bdrv_measure(opts, in_bs, errp); | |
6155 | } | |
6156 | ||
6157 | /** | |
6158 | * Return number of sectors on success, -errno on error. | |
6159 | */ | |
6160 | int64_t coroutine_fn bdrv_co_nb_sectors(BlockDriverState *bs) | |
6161 | { | |
6162 | BlockDriver *drv = bs->drv; | |
6163 | IO_CODE(); | |
6164 | assert_bdrv_graph_readable(); | |
6165 | ||
6166 | if (!drv) | |
6167 | return -ENOMEDIUM; | |
6168 | ||
6169 | if (bs->bl.has_variable_length) { | |
6170 | int ret = bdrv_co_refresh_total_sectors(bs, bs->total_sectors); | |
6171 | if (ret < 0) { | |
6172 | return ret; | |
6173 | } | |
6174 | } | |
6175 | return bs->total_sectors; | |
6176 | } | |
6177 | ||
6178 | /* | |
6179 | * This wrapper is written by hand because this function is in the hot I/O path, | |
6180 | * via blk_get_geometry. | |
6181 | */ | |
6182 | int64_t coroutine_mixed_fn bdrv_nb_sectors(BlockDriverState *bs) | |
6183 | { | |
6184 | BlockDriver *drv = bs->drv; | |
6185 | IO_CODE(); | |
6186 | ||
6187 | if (!drv) | |
6188 | return -ENOMEDIUM; | |
6189 | ||
6190 | if (bs->bl.has_variable_length) { | |
6191 | int ret = bdrv_refresh_total_sectors(bs, bs->total_sectors); | |
6192 | if (ret < 0) { | |
6193 | return ret; | |
6194 | } | |
6195 | } | |
6196 | ||
6197 | return bs->total_sectors; | |
6198 | } | |
6199 | ||
6200 | /** | |
6201 | * Return length in bytes on success, -errno on error. | |
6202 | * The length is always a multiple of BDRV_SECTOR_SIZE. | |
6203 | */ | |
6204 | int64_t coroutine_fn bdrv_co_getlength(BlockDriverState *bs) | |
6205 | { | |
6206 | int64_t ret; | |
6207 | IO_CODE(); | |
6208 | assert_bdrv_graph_readable(); | |
6209 | ||
6210 | ret = bdrv_co_nb_sectors(bs); | |
6211 | if (ret < 0) { | |
6212 | return ret; | |
6213 | } | |
6214 | if (ret > INT64_MAX / BDRV_SECTOR_SIZE) { | |
6215 | return -EFBIG; | |
6216 | } | |
6217 | return ret * BDRV_SECTOR_SIZE; | |
6218 | } | |
6219 | ||
6220 | bool bdrv_is_sg(BlockDriverState *bs) | |
6221 | { | |
6222 | IO_CODE(); | |
6223 | return bs->sg; | |
6224 | } | |
6225 | ||
6226 | /** | |
6227 | * Return whether the given node supports compressed writes. | |
6228 | */ | |
6229 | bool bdrv_supports_compressed_writes(BlockDriverState *bs) | |
6230 | { | |
6231 | BlockDriverState *filtered; | |
6232 | IO_CODE(); | |
6233 | ||
6234 | if (!bs->drv || !block_driver_can_compress(bs->drv)) { | |
6235 | return false; | |
6236 | } | |
6237 | ||
6238 | filtered = bdrv_filter_bs(bs); | |
6239 | if (filtered) { | |
6240 | /* | |
6241 | * Filters can only forward compressed writes, so we have to | |
6242 | * check the child. | |
6243 | */ | |
6244 | return bdrv_supports_compressed_writes(filtered); | |
6245 | } | |
6246 | ||
6247 | return true; | |
6248 | } | |
6249 | ||
6250 | const char *bdrv_get_format_name(BlockDriverState *bs) | |
6251 | { | |
6252 | IO_CODE(); | |
6253 | return bs->drv ? bs->drv->format_name : NULL; | |
6254 | } | |
6255 | ||
6256 | static int qsort_strcmp(const void *a, const void *b) | |
6257 | { | |
6258 | return strcmp(*(char *const *)a, *(char *const *)b); | |
6259 | } | |
6260 | ||
6261 | void bdrv_iterate_format(void (*it)(void *opaque, const char *name), | |
6262 | void *opaque, bool read_only) | |
6263 | { | |
6264 | BlockDriver *drv; | |
6265 | int count = 0; | |
6266 | int i; | |
6267 | const char **formats = NULL; | |
6268 | ||
6269 | GLOBAL_STATE_CODE(); | |
6270 | ||
6271 | QLIST_FOREACH(drv, &bdrv_drivers, list) { | |
6272 | if (drv->format_name) { | |
6273 | bool found = false; | |
6274 | ||
6275 | if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv, read_only)) { | |
6276 | continue; | |
6277 | } | |
6278 | ||
6279 | i = count; | |
6280 | while (formats && i && !found) { | |
6281 | found = !strcmp(formats[--i], drv->format_name); | |
6282 | } | |
6283 | ||
6284 | if (!found) { | |
6285 | formats = g_renew(const char *, formats, count + 1); | |
6286 | formats[count++] = drv->format_name; | |
6287 | } | |
6288 | } | |
6289 | } | |
6290 | ||
6291 | for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); i++) { | |
6292 | const char *format_name = block_driver_modules[i].format_name; | |
6293 | ||
6294 | if (format_name) { | |
6295 | bool found = false; | |
6296 | int j = count; | |
6297 | ||
6298 | if (use_bdrv_whitelist && | |
6299 | !bdrv_format_is_whitelisted(format_name, read_only)) { | |
6300 | continue; | |
6301 | } | |
6302 | ||
6303 | while (formats && j && !found) { | |
6304 | found = !strcmp(formats[--j], format_name); | |
6305 | } | |
6306 | ||
6307 | if (!found) { | |
6308 | formats = g_renew(const char *, formats, count + 1); | |
6309 | formats[count++] = format_name; | |
6310 | } | |
6311 | } | |
6312 | } | |
6313 | ||
6314 | qsort(formats, count, sizeof(formats[0]), qsort_strcmp); | |
6315 | ||
6316 | for (i = 0; i < count; i++) { | |
6317 | it(opaque, formats[i]); | |
6318 | } | |
6319 | ||
6320 | g_free(formats); | |
6321 | } | |
6322 | ||
6323 | /* This function is to find a node in the bs graph */ | |
6324 | BlockDriverState *bdrv_find_node(const char *node_name) | |
6325 | { | |
6326 | BlockDriverState *bs; | |
6327 | ||
6328 | assert(node_name); | |
6329 | GLOBAL_STATE_CODE(); | |
6330 | ||
6331 | QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) { | |
6332 | if (!strcmp(node_name, bs->node_name)) { | |
6333 | return bs; | |
6334 | } | |
6335 | } | |
6336 | return NULL; | |
6337 | } | |
6338 | ||
6339 | /* Put this QMP function here so it can access the static graph_bdrv_states. */ | |
6340 | BlockDeviceInfoList *bdrv_named_nodes_list(bool flat, | |
6341 | Error **errp) | |
6342 | { | |
6343 | BlockDeviceInfoList *list; | |
6344 | BlockDriverState *bs; | |
6345 | ||
6346 | GLOBAL_STATE_CODE(); | |
6347 | GRAPH_RDLOCK_GUARD_MAINLOOP(); | |
6348 | ||
6349 | list = NULL; | |
6350 | QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) { | |
6351 | BlockDeviceInfo *info = bdrv_block_device_info(NULL, bs, flat, errp); | |
6352 | if (!info) { | |
6353 | qapi_free_BlockDeviceInfoList(list); | |
6354 | return NULL; | |
6355 | } | |
6356 | QAPI_LIST_PREPEND(list, info); | |
6357 | } | |
6358 | ||
6359 | return list; | |
6360 | } | |
6361 | ||
6362 | typedef struct XDbgBlockGraphConstructor { | |
6363 | XDbgBlockGraph *graph; | |
6364 | GHashTable *graph_nodes; | |
6365 | } XDbgBlockGraphConstructor; | |
6366 | ||
6367 | static XDbgBlockGraphConstructor *xdbg_graph_new(void) | |
6368 | { | |
6369 | XDbgBlockGraphConstructor *gr = g_new(XDbgBlockGraphConstructor, 1); | |
6370 | ||
6371 | gr->graph = g_new0(XDbgBlockGraph, 1); | |
6372 | gr->graph_nodes = g_hash_table_new(NULL, NULL); | |
6373 | ||
6374 | return gr; | |
6375 | } | |
6376 | ||
6377 | static XDbgBlockGraph *xdbg_graph_finalize(XDbgBlockGraphConstructor *gr) | |
6378 | { | |
6379 | XDbgBlockGraph *graph = gr->graph; | |
6380 | ||
6381 | g_hash_table_destroy(gr->graph_nodes); | |
6382 | g_free(gr); | |
6383 | ||
6384 | return graph; | |
6385 | } | |
6386 | ||
6387 | static uintptr_t xdbg_graph_node_num(XDbgBlockGraphConstructor *gr, void *node) | |
6388 | { | |
6389 | uintptr_t ret = (uintptr_t)g_hash_table_lookup(gr->graph_nodes, node); | |
6390 | ||
6391 | if (ret != 0) { | |
6392 | return ret; | |
6393 | } | |
6394 | ||
6395 | /* | |
6396 | * Start counting from 1, not 0, because 0 interferes with not-found (NULL) | |
6397 | * answer of g_hash_table_lookup. | |
6398 | */ | |
6399 | ret = g_hash_table_size(gr->graph_nodes) + 1; | |
6400 | g_hash_table_insert(gr->graph_nodes, node, (void *)ret); | |
6401 | ||
6402 | return ret; | |
6403 | } | |
6404 | ||
6405 | static void xdbg_graph_add_node(XDbgBlockGraphConstructor *gr, void *node, | |
6406 | XDbgBlockGraphNodeType type, const char *name) | |
6407 | { | |
6408 | XDbgBlockGraphNode *n; | |
6409 | ||
6410 | n = g_new0(XDbgBlockGraphNode, 1); | |
6411 | ||
6412 | n->id = xdbg_graph_node_num(gr, node); | |
6413 | n->type = type; | |
6414 | n->name = g_strdup(name); | |
6415 | ||
6416 | QAPI_LIST_PREPEND(gr->graph->nodes, n); | |
6417 | } | |
6418 | ||
6419 | static void xdbg_graph_add_edge(XDbgBlockGraphConstructor *gr, void *parent, | |
6420 | const BdrvChild *child) | |
6421 | { | |
6422 | BlockPermission qapi_perm; | |
6423 | XDbgBlockGraphEdge *edge; | |
6424 | GLOBAL_STATE_CODE(); | |
6425 | ||
6426 | edge = g_new0(XDbgBlockGraphEdge, 1); | |
6427 | ||
6428 | edge->parent = xdbg_graph_node_num(gr, parent); | |
6429 | edge->child = xdbg_graph_node_num(gr, child->bs); | |
6430 | edge->name = g_strdup(child->name); | |
6431 | ||
6432 | for (qapi_perm = 0; qapi_perm < BLOCK_PERMISSION__MAX; qapi_perm++) { | |
6433 | uint64_t flag = bdrv_qapi_perm_to_blk_perm(qapi_perm); | |
6434 | ||
6435 | if (flag & child->perm) { | |
6436 | QAPI_LIST_PREPEND(edge->perm, qapi_perm); | |
6437 | } | |
6438 | if (flag & child->shared_perm) { | |
6439 | QAPI_LIST_PREPEND(edge->shared_perm, qapi_perm); | |
6440 | } | |
6441 | } | |
6442 | ||
6443 | QAPI_LIST_PREPEND(gr->graph->edges, edge); | |
6444 | } | |
6445 | ||
6446 | ||
6447 | XDbgBlockGraph *bdrv_get_xdbg_block_graph(Error **errp) | |
6448 | { | |
6449 | BlockBackend *blk; | |
6450 | BlockJob *job; | |
6451 | BlockDriverState *bs; | |
6452 | BdrvChild *child; | |
6453 | XDbgBlockGraphConstructor *gr = xdbg_graph_new(); | |
6454 | ||
6455 | GLOBAL_STATE_CODE(); | |
6456 | ||
6457 | for (blk = blk_all_next(NULL); blk; blk = blk_all_next(blk)) { | |
6458 | char *allocated_name = NULL; | |
6459 | const char *name = blk_name(blk); | |
6460 | ||
6461 | if (!*name) { | |
6462 | name = allocated_name = blk_get_attached_dev_id(blk); | |
6463 | } | |
6464 | xdbg_graph_add_node(gr, blk, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_BACKEND, | |
6465 | name); | |
6466 | g_free(allocated_name); | |
6467 | if (blk_root(blk)) { | |
6468 | xdbg_graph_add_edge(gr, blk, blk_root(blk)); | |
6469 | } | |
6470 | } | |
6471 | ||
6472 | WITH_JOB_LOCK_GUARD() { | |
6473 | for (job = block_job_next_locked(NULL); job; | |
6474 | job = block_job_next_locked(job)) { | |
6475 | GSList *el; | |
6476 | ||
6477 | xdbg_graph_add_node(gr, job, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_JOB, | |
6478 | job->job.id); | |
6479 | for (el = job->nodes; el; el = el->next) { | |
6480 | xdbg_graph_add_edge(gr, job, (BdrvChild *)el->data); | |
6481 | } | |
6482 | } | |
6483 | } | |
6484 | ||
6485 | QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) { | |
6486 | xdbg_graph_add_node(gr, bs, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_DRIVER, | |
6487 | bs->node_name); | |
6488 | QLIST_FOREACH(child, &bs->children, next) { | |
6489 | xdbg_graph_add_edge(gr, bs, child); | |
6490 | } | |
6491 | } | |
6492 | ||
6493 | return xdbg_graph_finalize(gr); | |
6494 | } | |
6495 | ||
6496 | BlockDriverState *bdrv_lookup_bs(const char *device, | |
6497 | const char *node_name, | |
6498 | Error **errp) | |
6499 | { | |
6500 | BlockBackend *blk; | |
6501 | BlockDriverState *bs; | |
6502 | ||
6503 | GLOBAL_STATE_CODE(); | |
6504 | ||
6505 | if (device) { | |
6506 | blk = blk_by_name(device); | |
6507 | ||
6508 | if (blk) { | |
6509 | bs = blk_bs(blk); | |
6510 | if (!bs) { | |
6511 | error_setg(errp, "Device '%s' has no medium", device); | |
6512 | } | |
6513 | ||
6514 | return bs; | |
6515 | } | |
6516 | } | |
6517 | ||
6518 | if (node_name) { | |
6519 | bs = bdrv_find_node(node_name); | |
6520 | ||
6521 | if (bs) { | |
6522 | return bs; | |
6523 | } | |
6524 | } | |
6525 | ||
6526 | error_setg(errp, "Cannot find device=\'%s\' nor node-name=\'%s\'", | |
6527 | device ? device : "", | |
6528 | node_name ? node_name : ""); | |
6529 | return NULL; | |
6530 | } | |
6531 | ||
6532 | /* If 'base' is in the same chain as 'top', return true. Otherwise, | |
6533 | * return false. If either argument is NULL, return false. */ | |
6534 | bool bdrv_chain_contains(BlockDriverState *top, BlockDriverState *base) | |
6535 | { | |
6536 | ||
6537 | GLOBAL_STATE_CODE(); | |
6538 | ||
6539 | while (top && top != base) { | |
6540 | top = bdrv_filter_or_cow_bs(top); | |
6541 | } | |
6542 | ||
6543 | return top != NULL; | |
6544 | } | |
6545 | ||
6546 | BlockDriverState *bdrv_next_node(BlockDriverState *bs) | |
6547 | { | |
6548 | GLOBAL_STATE_CODE(); | |
6549 | if (!bs) { | |
6550 | return QTAILQ_FIRST(&graph_bdrv_states); | |
6551 | } | |
6552 | return QTAILQ_NEXT(bs, node_list); | |
6553 | } | |
6554 | ||
6555 | BlockDriverState *bdrv_next_all_states(BlockDriverState *bs) | |
6556 | { | |
6557 | GLOBAL_STATE_CODE(); | |
6558 | if (!bs) { | |
6559 | return QTAILQ_FIRST(&all_bdrv_states); | |
6560 | } | |
6561 | return QTAILQ_NEXT(bs, bs_list); | |
6562 | } | |
6563 | ||
6564 | const char *bdrv_get_node_name(const BlockDriverState *bs) | |
6565 | { | |
6566 | IO_CODE(); | |
6567 | return bs->node_name; | |
6568 | } | |
6569 | ||
6570 | const char *bdrv_get_parent_name(const BlockDriverState *bs) | |
6571 | { | |
6572 | BdrvChild *c; | |
6573 | const char *name; | |
6574 | IO_CODE(); | |
6575 | ||
6576 | /* If multiple parents have a name, just pick the first one. */ | |
6577 | QLIST_FOREACH(c, &bs->parents, next_parent) { | |
6578 | if (c->klass->get_name) { | |
6579 | name = c->klass->get_name(c); | |
6580 | if (name && *name) { | |
6581 | return name; | |
6582 | } | |
6583 | } | |
6584 | } | |
6585 | ||
6586 | return NULL; | |
6587 | } | |
6588 | ||
6589 | /* TODO check what callers really want: bs->node_name or blk_name() */ | |
6590 | const char *bdrv_get_device_name(const BlockDriverState *bs) | |
6591 | { | |
6592 | IO_CODE(); | |
6593 | return bdrv_get_parent_name(bs) ?: ""; | |
6594 | } | |
6595 | ||
6596 | /* This can be used to identify nodes that might not have a device | |
6597 | * name associated. Since node and device names live in the same | |
6598 | * namespace, the result is unambiguous. The exception is if both are | |
6599 | * absent, then this returns an empty (non-null) string. */ | |
6600 | const char *bdrv_get_device_or_node_name(const BlockDriverState *bs) | |
6601 | { | |
6602 | IO_CODE(); | |
6603 | return bdrv_get_parent_name(bs) ?: bs->node_name; | |
6604 | } | |
6605 | ||
6606 | int bdrv_get_flags(BlockDriverState *bs) | |
6607 | { | |
6608 | IO_CODE(); | |
6609 | return bs->open_flags; | |
6610 | } | |
6611 | ||
6612 | int bdrv_has_zero_init_1(BlockDriverState *bs) | |
6613 | { | |
6614 | GLOBAL_STATE_CODE(); | |
6615 | return 1; | |
6616 | } | |
6617 | ||
6618 | int coroutine_mixed_fn bdrv_has_zero_init(BlockDriverState *bs) | |
6619 | { | |
6620 | BlockDriverState *filtered; | |
6621 | GLOBAL_STATE_CODE(); | |
6622 | ||
6623 | if (!bs->drv) { | |
6624 | return 0; | |
6625 | } | |
6626 | ||
6627 | /* If BS is a copy on write image, it is initialized to | |
6628 | the contents of the base image, which may not be zeroes. */ | |
6629 | if (bdrv_cow_child(bs)) { | |
6630 | return 0; | |
6631 | } | |
6632 | if (bs->drv->bdrv_has_zero_init) { | |
6633 | return bs->drv->bdrv_has_zero_init(bs); | |
6634 | } | |
6635 | ||
6636 | filtered = bdrv_filter_bs(bs); | |
6637 | if (filtered) { | |
6638 | return bdrv_has_zero_init(filtered); | |
6639 | } | |
6640 | ||
6641 | /* safe default */ | |
6642 | return 0; | |
6643 | } | |
6644 | ||
6645 | bool bdrv_can_write_zeroes_with_unmap(BlockDriverState *bs) | |
6646 | { | |
6647 | IO_CODE(); | |
6648 | if (!(bs->open_flags & BDRV_O_UNMAP)) { | |
6649 | return false; | |
6650 | } | |
6651 | ||
6652 | return bs->supported_zero_flags & BDRV_REQ_MAY_UNMAP; | |
6653 | } | |
6654 | ||
6655 | void bdrv_get_backing_filename(BlockDriverState *bs, | |
6656 | char *filename, int filename_size) | |
6657 | { | |
6658 | IO_CODE(); | |
6659 | pstrcpy(filename, filename_size, bs->backing_file); | |
6660 | } | |
6661 | ||
6662 | int coroutine_fn bdrv_co_get_info(BlockDriverState *bs, BlockDriverInfo *bdi) | |
6663 | { | |
6664 | int ret; | |
6665 | BlockDriver *drv = bs->drv; | |
6666 | IO_CODE(); | |
6667 | assert_bdrv_graph_readable(); | |
6668 | ||
6669 | /* if bs->drv == NULL, bs is closed, so there's nothing to do here */ | |
6670 | if (!drv) { | |
6671 | return -ENOMEDIUM; | |
6672 | } | |
6673 | if (!drv->bdrv_co_get_info) { | |
6674 | BlockDriverState *filtered = bdrv_filter_bs(bs); | |
6675 | if (filtered) { | |
6676 | return bdrv_co_get_info(filtered, bdi); | |
6677 | } | |
6678 | return -ENOTSUP; | |
6679 | } | |
6680 | memset(bdi, 0, sizeof(*bdi)); | |
6681 | ret = drv->bdrv_co_get_info(bs, bdi); | |
6682 | if (bdi->subcluster_size == 0) { | |
6683 | /* | |
6684 | * If the driver left this unset, subclusters are not supported. | |
6685 | * Then it is safe to treat each cluster as having only one subcluster. | |
6686 | */ | |
6687 | bdi->subcluster_size = bdi->cluster_size; | |
6688 | } | |
6689 | if (ret < 0) { | |
6690 | return ret; | |
6691 | } | |
6692 | ||
6693 | if (bdi->cluster_size > BDRV_MAX_ALIGNMENT) { | |
6694 | return -EINVAL; | |
6695 | } | |
6696 | ||
6697 | return 0; | |
6698 | } | |
6699 | ||
6700 | ImageInfoSpecific *bdrv_get_specific_info(BlockDriverState *bs, | |
6701 | Error **errp) | |
6702 | { | |
6703 | BlockDriver *drv = bs->drv; | |
6704 | IO_CODE(); | |
6705 | if (drv && drv->bdrv_get_specific_info) { | |
6706 | return drv->bdrv_get_specific_info(bs, errp); | |
6707 | } | |
6708 | return NULL; | |
6709 | } | |
6710 | ||
6711 | BlockStatsSpecific *bdrv_get_specific_stats(BlockDriverState *bs) | |
6712 | { | |
6713 | BlockDriver *drv = bs->drv; | |
6714 | IO_CODE(); | |
6715 | if (!drv || !drv->bdrv_get_specific_stats) { | |
6716 | return NULL; | |
6717 | } | |
6718 | return drv->bdrv_get_specific_stats(bs); | |
6719 | } | |
6720 | ||
6721 | void coroutine_fn bdrv_co_debug_event(BlockDriverState *bs, BlkdebugEvent event) | |
6722 | { | |
6723 | IO_CODE(); | |
6724 | assert_bdrv_graph_readable(); | |
6725 | ||
6726 | if (!bs || !bs->drv || !bs->drv->bdrv_co_debug_event) { | |
6727 | return; | |
6728 | } | |
6729 | ||
6730 | bs->drv->bdrv_co_debug_event(bs, event); | |
6731 | } | |
6732 | ||
6733 | static BlockDriverState * GRAPH_RDLOCK | |
6734 | bdrv_find_debug_node(BlockDriverState *bs) | |
6735 | { | |
6736 | GLOBAL_STATE_CODE(); | |
6737 | while (bs && bs->drv && !bs->drv->bdrv_debug_breakpoint) { | |
6738 | bs = bdrv_primary_bs(bs); | |
6739 | } | |
6740 | ||
6741 | if (bs && bs->drv && bs->drv->bdrv_debug_breakpoint) { | |
6742 | assert(bs->drv->bdrv_debug_remove_breakpoint); | |
6743 | return bs; | |
6744 | } | |
6745 | ||
6746 | return NULL; | |
6747 | } | |
6748 | ||
6749 | int bdrv_debug_breakpoint(BlockDriverState *bs, const char *event, | |
6750 | const char *tag) | |
6751 | { | |
6752 | GLOBAL_STATE_CODE(); | |
6753 | GRAPH_RDLOCK_GUARD_MAINLOOP(); | |
6754 | ||
6755 | bs = bdrv_find_debug_node(bs); | |
6756 | if (bs) { | |
6757 | return bs->drv->bdrv_debug_breakpoint(bs, event, tag); | |
6758 | } | |
6759 | ||
6760 | return -ENOTSUP; | |
6761 | } | |
6762 | ||
6763 | int bdrv_debug_remove_breakpoint(BlockDriverState *bs, const char *tag) | |
6764 | { | |
6765 | GLOBAL_STATE_CODE(); | |
6766 | GRAPH_RDLOCK_GUARD_MAINLOOP(); | |
6767 | ||
6768 | bs = bdrv_find_debug_node(bs); | |
6769 | if (bs) { | |
6770 | return bs->drv->bdrv_debug_remove_breakpoint(bs, tag); | |
6771 | } | |
6772 | ||
6773 | return -ENOTSUP; | |
6774 | } | |
6775 | ||
6776 | int bdrv_debug_resume(BlockDriverState *bs, const char *tag) | |
6777 | { | |
6778 | GLOBAL_STATE_CODE(); | |
6779 | GRAPH_RDLOCK_GUARD_MAINLOOP(); | |
6780 | ||
6781 | while (bs && (!bs->drv || !bs->drv->bdrv_debug_resume)) { | |
6782 | bs = bdrv_primary_bs(bs); | |
6783 | } | |
6784 | ||
6785 | if (bs && bs->drv && bs->drv->bdrv_debug_resume) { | |
6786 | return bs->drv->bdrv_debug_resume(bs, tag); | |
6787 | } | |
6788 | ||
6789 | return -ENOTSUP; | |
6790 | } | |
6791 | ||
6792 | bool bdrv_debug_is_suspended(BlockDriverState *bs, const char *tag) | |
6793 | { | |
6794 | GLOBAL_STATE_CODE(); | |
6795 | GRAPH_RDLOCK_GUARD_MAINLOOP(); | |
6796 | ||
6797 | while (bs && bs->drv && !bs->drv->bdrv_debug_is_suspended) { | |
6798 | bs = bdrv_primary_bs(bs); | |
6799 | } | |
6800 | ||
6801 | if (bs && bs->drv && bs->drv->bdrv_debug_is_suspended) { | |
6802 | return bs->drv->bdrv_debug_is_suspended(bs, tag); | |
6803 | } | |
6804 | ||
6805 | return false; | |
6806 | } | |
6807 | ||
6808 | /* backing_file can either be relative, or absolute, or a protocol. If it is | |
6809 | * relative, it must be relative to the chain. So, passing in bs->filename | |
6810 | * from a BDS as backing_file should not be done, as that may be relative to | |
6811 | * the CWD rather than the chain. */ | |
6812 | BlockDriverState *bdrv_find_backing_image(BlockDriverState *bs, | |
6813 | const char *backing_file) | |
6814 | { | |
6815 | char *filename_full = NULL; | |
6816 | char *backing_file_full = NULL; | |
6817 | char *filename_tmp = NULL; | |
6818 | int is_protocol = 0; | |
6819 | bool filenames_refreshed = false; | |
6820 | BlockDriverState *curr_bs = NULL; | |
6821 | BlockDriverState *retval = NULL; | |
6822 | BlockDriverState *bs_below; | |
6823 | ||
6824 | GLOBAL_STATE_CODE(); | |
6825 | GRAPH_RDLOCK_GUARD_MAINLOOP(); | |
6826 | ||
6827 | if (!bs || !bs->drv || !backing_file) { | |
6828 | return NULL; | |
6829 | } | |
6830 | ||
6831 | filename_full = g_malloc(PATH_MAX); | |
6832 | backing_file_full = g_malloc(PATH_MAX); | |
6833 | ||
6834 | is_protocol = path_has_protocol(backing_file); | |
6835 | ||
6836 | /* | |
6837 | * Being largely a legacy function, skip any filters here | |
6838 | * (because filters do not have normal filenames, so they cannot | |
6839 | * match anyway; and allowing json:{} filenames is a bit out of | |
6840 | * scope). | |
6841 | */ | |
6842 | for (curr_bs = bdrv_skip_filters(bs); | |
6843 | bdrv_cow_child(curr_bs) != NULL; | |
6844 | curr_bs = bs_below) | |
6845 | { | |
6846 | bs_below = bdrv_backing_chain_next(curr_bs); | |
6847 | ||
6848 | if (bdrv_backing_overridden(curr_bs)) { | |
6849 | /* | |
6850 | * If the backing file was overridden, we can only compare | |
6851 | * directly against the backing node's filename. | |
6852 | */ | |
6853 | ||
6854 | if (!filenames_refreshed) { | |
6855 | /* | |
6856 | * This will automatically refresh all of the | |
6857 | * filenames in the rest of the backing chain, so we | |
6858 | * only need to do this once. | |
6859 | */ | |
6860 | bdrv_refresh_filename(bs_below); | |
6861 | filenames_refreshed = true; | |
6862 | } | |
6863 | ||
6864 | if (strcmp(backing_file, bs_below->filename) == 0) { | |
6865 | retval = bs_below; | |
6866 | break; | |
6867 | } | |
6868 | } else if (is_protocol || path_has_protocol(curr_bs->backing_file)) { | |
6869 | /* | |
6870 | * If either of the filename paths is actually a protocol, then | |
6871 | * compare unmodified paths; otherwise make paths relative. | |
6872 | */ | |
6873 | char *backing_file_full_ret; | |
6874 | ||
6875 | if (strcmp(backing_file, curr_bs->backing_file) == 0) { | |
6876 | retval = bs_below; | |
6877 | break; | |
6878 | } | |
6879 | /* Also check against the full backing filename for the image */ | |
6880 | backing_file_full_ret = bdrv_get_full_backing_filename(curr_bs, | |
6881 | NULL); | |
6882 | if (backing_file_full_ret) { | |
6883 | bool equal = strcmp(backing_file, backing_file_full_ret) == 0; | |
6884 | g_free(backing_file_full_ret); | |
6885 | if (equal) { | |
6886 | retval = bs_below; | |
6887 | break; | |
6888 | } | |
6889 | } | |
6890 | } else { | |
6891 | /* If not an absolute filename path, make it relative to the current | |
6892 | * image's filename path */ | |
6893 | filename_tmp = bdrv_make_absolute_filename(curr_bs, backing_file, | |
6894 | NULL); | |
6895 | /* We are going to compare canonicalized absolute pathnames */ | |
6896 | if (!filename_tmp || !realpath(filename_tmp, filename_full)) { | |
6897 | g_free(filename_tmp); | |
6898 | continue; | |
6899 | } | |
6900 | g_free(filename_tmp); | |
6901 | ||
6902 | /* We need to make sure the backing filename we are comparing against | |
6903 | * is relative to the current image filename (or absolute) */ | |
6904 | filename_tmp = bdrv_get_full_backing_filename(curr_bs, NULL); | |
6905 | if (!filename_tmp || !realpath(filename_tmp, backing_file_full)) { | |
6906 | g_free(filename_tmp); | |
6907 | continue; | |
6908 | } | |
6909 | g_free(filename_tmp); | |
6910 | ||
6911 | if (strcmp(backing_file_full, filename_full) == 0) { | |
6912 | retval = bs_below; | |
6913 | break; | |
6914 | } | |
6915 | } | |
6916 | } | |
6917 | ||
6918 | g_free(filename_full); | |
6919 | g_free(backing_file_full); | |
6920 | return retval; | |
6921 | } | |
6922 | ||
6923 | void bdrv_init(void) | |
6924 | { | |
6925 | #ifdef CONFIG_BDRV_WHITELIST_TOOLS | |
6926 | use_bdrv_whitelist = 1; | |
6927 | #endif | |
6928 | module_call_init(MODULE_INIT_BLOCK); | |
6929 | } | |
6930 | ||
6931 | void bdrv_init_with_whitelist(void) | |
6932 | { | |
6933 | use_bdrv_whitelist = 1; | |
6934 | bdrv_init(); | |
6935 | } | |
6936 | ||
6937 | int bdrv_activate(BlockDriverState *bs, Error **errp) | |
6938 | { | |
6939 | BdrvChild *child, *parent; | |
6940 | Error *local_err = NULL; | |
6941 | int ret; | |
6942 | BdrvDirtyBitmap *bm; | |
6943 | ||
6944 | GLOBAL_STATE_CODE(); | |
6945 | GRAPH_RDLOCK_GUARD_MAINLOOP(); | |
6946 | ||
6947 | if (!bs->drv) { | |
6948 | return -ENOMEDIUM; | |
6949 | } | |
6950 | ||
6951 | QLIST_FOREACH(child, &bs->children, next) { | |
6952 | bdrv_activate(child->bs, &local_err); | |
6953 | if (local_err) { | |
6954 | error_propagate(errp, local_err); | |
6955 | return -EINVAL; | |
6956 | } | |
6957 | } | |
6958 | ||
6959 | /* | |
6960 | * Update permissions, they may differ for inactive nodes. | |
6961 | * | |
6962 | * Note that the required permissions of inactive images are always a | |
6963 | * subset of the permissions required after activating the image. This | |
6964 | * allows us to just get the permissions upfront without restricting | |
6965 | * bdrv_co_invalidate_cache(). | |
6966 | * | |
6967 | * It also means that in error cases, we don't have to try and revert to | |
6968 | * the old permissions (which is an operation that could fail, too). We can | |
6969 | * just keep the extended permissions for the next time that an activation | |
6970 | * of the image is tried. | |
6971 | */ | |
6972 | if (bs->open_flags & BDRV_O_INACTIVE) { | |
6973 | bs->open_flags &= ~BDRV_O_INACTIVE; | |
6974 | ret = bdrv_refresh_perms(bs, NULL, errp); | |
6975 | if (ret < 0) { | |
6976 | bs->open_flags |= BDRV_O_INACTIVE; | |
6977 | return ret; | |
6978 | } | |
6979 | ||
6980 | ret = bdrv_invalidate_cache(bs, errp); | |
6981 | if (ret < 0) { | |
6982 | bs->open_flags |= BDRV_O_INACTIVE; | |
6983 | return ret; | |
6984 | } | |
6985 | ||
6986 | FOR_EACH_DIRTY_BITMAP(bs, bm) { | |
6987 | bdrv_dirty_bitmap_skip_store(bm, false); | |
6988 | } | |
6989 | ||
6990 | ret = bdrv_refresh_total_sectors(bs, bs->total_sectors); | |
6991 | if (ret < 0) { | |
6992 | bs->open_flags |= BDRV_O_INACTIVE; | |
6993 | error_setg_errno(errp, -ret, "Could not refresh total sector count"); | |
6994 | return ret; | |
6995 | } | |
6996 | } | |
6997 | ||
6998 | QLIST_FOREACH(parent, &bs->parents, next_parent) { | |
6999 | if (parent->klass->activate) { | |
7000 | parent->klass->activate(parent, &local_err); | |
7001 | if (local_err) { | |
7002 | bs->open_flags |= BDRV_O_INACTIVE; | |
7003 | error_propagate(errp, local_err); | |
7004 | return -EINVAL; | |
7005 | } | |
7006 | } | |
7007 | } | |
7008 | ||
7009 | return 0; | |
7010 | } | |
7011 | ||
7012 | int coroutine_fn bdrv_co_invalidate_cache(BlockDriverState *bs, Error **errp) | |
7013 | { | |
7014 | Error *local_err = NULL; | |
7015 | IO_CODE(); | |
7016 | ||
7017 | assert(!(bs->open_flags & BDRV_O_INACTIVE)); | |
7018 | assert_bdrv_graph_readable(); | |
7019 | ||
7020 | if (bs->drv->bdrv_co_invalidate_cache) { | |
7021 | bs->drv->bdrv_co_invalidate_cache(bs, &local_err); | |
7022 | if (local_err) { | |
7023 | error_propagate(errp, local_err); | |
7024 | return -EINVAL; | |
7025 | } | |
7026 | } | |
7027 | ||
7028 | return 0; | |
7029 | } | |
7030 | ||
7031 | void bdrv_activate_all(Error **errp) | |
7032 | { | |
7033 | BlockDriverState *bs; | |
7034 | BdrvNextIterator it; | |
7035 | ||
7036 | GLOBAL_STATE_CODE(); | |
7037 | GRAPH_RDLOCK_GUARD_MAINLOOP(); | |
7038 | ||
7039 | for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) { | |
7040 | AioContext *aio_context = bdrv_get_aio_context(bs); | |
7041 | int ret; | |
7042 | ||
7043 | aio_context_acquire(aio_context); | |
7044 | ret = bdrv_activate(bs, errp); | |
7045 | aio_context_release(aio_context); | |
7046 | if (ret < 0) { | |
7047 | bdrv_next_cleanup(&it); | |
7048 | return; | |
7049 | } | |
7050 | } | |
7051 | } | |
7052 | ||
7053 | static bool GRAPH_RDLOCK | |
7054 | bdrv_has_bds_parent(BlockDriverState *bs, bool only_active) | |
7055 | { | |
7056 | BdrvChild *parent; | |
7057 | GLOBAL_STATE_CODE(); | |
7058 | ||
7059 | QLIST_FOREACH(parent, &bs->parents, next_parent) { | |
7060 | if (parent->klass->parent_is_bds) { | |
7061 | BlockDriverState *parent_bs = parent->opaque; | |
7062 | if (!only_active || !(parent_bs->open_flags & BDRV_O_INACTIVE)) { | |
7063 | return true; | |
7064 | } | |
7065 | } | |
7066 | } | |
7067 | ||
7068 | return false; | |
7069 | } | |
7070 | ||
7071 | static int GRAPH_RDLOCK bdrv_inactivate_recurse(BlockDriverState *bs) | |
7072 | { | |
7073 | BdrvChild *child, *parent; | |
7074 | int ret; | |
7075 | uint64_t cumulative_perms, cumulative_shared_perms; | |
7076 | ||
7077 | GLOBAL_STATE_CODE(); | |
7078 | ||
7079 | if (!bs->drv) { | |
7080 | return -ENOMEDIUM; | |
7081 | } | |
7082 | ||
7083 | /* Make sure that we don't inactivate a child before its parent. | |
7084 | * It will be covered by recursion from the yet active parent. */ | |
7085 | if (bdrv_has_bds_parent(bs, true)) { | |
7086 | return 0; | |
7087 | } | |
7088 | ||
7089 | assert(!(bs->open_flags & BDRV_O_INACTIVE)); | |
7090 | ||
7091 | /* Inactivate this node */ | |
7092 | if (bs->drv->bdrv_inactivate) { | |
7093 | ret = bs->drv->bdrv_inactivate(bs); | |
7094 | if (ret < 0) { | |
7095 | return ret; | |
7096 | } | |
7097 | } | |
7098 | ||
7099 | QLIST_FOREACH(parent, &bs->parents, next_parent) { | |
7100 | if (parent->klass->inactivate) { | |
7101 | ret = parent->klass->inactivate(parent); | |
7102 | if (ret < 0) { | |
7103 | return ret; | |
7104 | } | |
7105 | } | |
7106 | } | |
7107 | ||
7108 | bdrv_get_cumulative_perm(bs, &cumulative_perms, | |
7109 | &cumulative_shared_perms); | |
7110 | if (cumulative_perms & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)) { | |
7111 | /* Our inactive parents still need write access. Inactivation failed. */ | |
7112 | return -EPERM; | |
7113 | } | |
7114 | ||
7115 | bs->open_flags |= BDRV_O_INACTIVE; | |
7116 | ||
7117 | /* | |
7118 | * Update permissions, they may differ for inactive nodes. | |
7119 | * We only tried to loosen restrictions, so errors are not fatal, ignore | |
7120 | * them. | |
7121 | */ | |
7122 | bdrv_refresh_perms(bs, NULL, NULL); | |
7123 | ||
7124 | /* Recursively inactivate children */ | |
7125 | QLIST_FOREACH(child, &bs->children, next) { | |
7126 | ret = bdrv_inactivate_recurse(child->bs); | |
7127 | if (ret < 0) { | |
7128 | return ret; | |
7129 | } | |
7130 | } | |
7131 | ||
7132 | return 0; | |
7133 | } | |
7134 | ||
7135 | int bdrv_inactivate_all(void) | |
7136 | { | |
7137 | BlockDriverState *bs = NULL; | |
7138 | BdrvNextIterator it; | |
7139 | int ret = 0; | |
7140 | GSList *aio_ctxs = NULL, *ctx; | |
7141 | ||
7142 | GLOBAL_STATE_CODE(); | |
7143 | GRAPH_RDLOCK_GUARD_MAINLOOP(); | |
7144 | ||
7145 | for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) { | |
7146 | AioContext *aio_context = bdrv_get_aio_context(bs); | |
7147 | ||
7148 | if (!g_slist_find(aio_ctxs, aio_context)) { | |
7149 | aio_ctxs = g_slist_prepend(aio_ctxs, aio_context); | |
7150 | aio_context_acquire(aio_context); | |
7151 | } | |
7152 | } | |
7153 | ||
7154 | for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) { | |
7155 | /* Nodes with BDS parents are covered by recursion from the last | |
7156 | * parent that gets inactivated. Don't inactivate them a second | |
7157 | * time if that has already happened. */ | |
7158 | if (bdrv_has_bds_parent(bs, false)) { | |
7159 | continue; | |
7160 | } | |
7161 | ret = bdrv_inactivate_recurse(bs); | |
7162 | if (ret < 0) { | |
7163 | bdrv_next_cleanup(&it); | |
7164 | goto out; | |
7165 | } | |
7166 | } | |
7167 | ||
7168 | out: | |
7169 | for (ctx = aio_ctxs; ctx != NULL; ctx = ctx->next) { | |
7170 | AioContext *aio_context = ctx->data; | |
7171 | aio_context_release(aio_context); | |
7172 | } | |
7173 | g_slist_free(aio_ctxs); | |
7174 | ||
7175 | return ret; | |
7176 | } | |
7177 | ||
7178 | /**************************************************************/ | |
7179 | /* removable device support */ | |
7180 | ||
7181 | /** | |
7182 | * Return TRUE if the media is present | |
7183 | */ | |
7184 | bool coroutine_fn bdrv_co_is_inserted(BlockDriverState *bs) | |
7185 | { | |
7186 | BlockDriver *drv = bs->drv; | |
7187 | BdrvChild *child; | |
7188 | IO_CODE(); | |
7189 | assert_bdrv_graph_readable(); | |
7190 | ||
7191 | if (!drv) { | |
7192 | return false; | |
7193 | } | |
7194 | if (drv->bdrv_co_is_inserted) { | |
7195 | return drv->bdrv_co_is_inserted(bs); | |
7196 | } | |
7197 | QLIST_FOREACH(child, &bs->children, next) { | |
7198 | if (!bdrv_co_is_inserted(child->bs)) { | |
7199 | return false; | |
7200 | } | |
7201 | } | |
7202 | return true; | |
7203 | } | |
7204 | ||
7205 | /** | |
7206 | * If eject_flag is TRUE, eject the media. Otherwise, close the tray | |
7207 | */ | |
7208 | void coroutine_fn bdrv_co_eject(BlockDriverState *bs, bool eject_flag) | |
7209 | { | |
7210 | BlockDriver *drv = bs->drv; | |
7211 | IO_CODE(); | |
7212 | assert_bdrv_graph_readable(); | |
7213 | ||
7214 | if (drv && drv->bdrv_co_eject) { | |
7215 | drv->bdrv_co_eject(bs, eject_flag); | |
7216 | } | |
7217 | } | |
7218 | ||
7219 | /** | |
7220 | * Lock or unlock the media (if it is locked, the user won't be able | |
7221 | * to eject it manually). | |
7222 | */ | |
7223 | void coroutine_fn bdrv_co_lock_medium(BlockDriverState *bs, bool locked) | |
7224 | { | |
7225 | BlockDriver *drv = bs->drv; | |
7226 | IO_CODE(); | |
7227 | assert_bdrv_graph_readable(); | |
7228 | trace_bdrv_lock_medium(bs, locked); | |
7229 | ||
7230 | if (drv && drv->bdrv_co_lock_medium) { | |
7231 | drv->bdrv_co_lock_medium(bs, locked); | |
7232 | } | |
7233 | } | |
7234 | ||
7235 | /* Get a reference to bs */ | |
7236 | void bdrv_ref(BlockDriverState *bs) | |
7237 | { | |
7238 | GLOBAL_STATE_CODE(); | |
7239 | bs->refcnt++; | |
7240 | } | |
7241 | ||
7242 | /* Release a previously grabbed reference to bs. | |
7243 | * If after releasing, reference count is zero, the BlockDriverState is | |
7244 | * deleted. */ | |
7245 | void bdrv_unref(BlockDriverState *bs) | |
7246 | { | |
7247 | GLOBAL_STATE_CODE(); | |
7248 | if (!bs) { | |
7249 | return; | |
7250 | } | |
7251 | assert(bs->refcnt > 0); | |
7252 | if (--bs->refcnt == 0) { | |
7253 | bdrv_delete(bs); | |
7254 | } | |
7255 | } | |
7256 | ||
7257 | /* | |
7258 | * Release a BlockDriverState reference while holding the graph write lock. | |
7259 | * | |
7260 | * Calling bdrv_unref() directly is forbidden while holding the graph lock | |
7261 | * because bdrv_close() both involves polling and taking the graph lock | |
7262 | * internally. bdrv_schedule_unref() instead delays decreasing the refcount and | |
7263 | * possibly closing @bs until the graph lock is released. | |
7264 | */ | |
7265 | void bdrv_schedule_unref(BlockDriverState *bs) | |
7266 | { | |
7267 | if (!bs) { | |
7268 | return; | |
7269 | } | |
7270 | aio_bh_schedule_oneshot(qemu_get_aio_context(), | |
7271 | (QEMUBHFunc *) bdrv_unref, bs); | |
7272 | } | |
7273 | ||
7274 | struct BdrvOpBlocker { | |
7275 | Error *reason; | |
7276 | QLIST_ENTRY(BdrvOpBlocker) list; | |
7277 | }; | |
7278 | ||
7279 | bool bdrv_op_is_blocked(BlockDriverState *bs, BlockOpType op, Error **errp) | |
7280 | { | |
7281 | BdrvOpBlocker *blocker; | |
7282 | GLOBAL_STATE_CODE(); | |
7283 | ||
7284 | assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX); | |
7285 | if (!QLIST_EMPTY(&bs->op_blockers[op])) { | |
7286 | blocker = QLIST_FIRST(&bs->op_blockers[op]); | |
7287 | error_propagate_prepend(errp, error_copy(blocker->reason), | |
7288 | "Node '%s' is busy: ", | |
7289 | bdrv_get_device_or_node_name(bs)); | |
7290 | return true; | |
7291 | } | |
7292 | return false; | |
7293 | } | |
7294 | ||
7295 | void bdrv_op_block(BlockDriverState *bs, BlockOpType op, Error *reason) | |
7296 | { | |
7297 | BdrvOpBlocker *blocker; | |
7298 | GLOBAL_STATE_CODE(); | |
7299 | assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX); | |
7300 | ||
7301 | blocker = g_new0(BdrvOpBlocker, 1); | |
7302 | blocker->reason = reason; | |
7303 | QLIST_INSERT_HEAD(&bs->op_blockers[op], blocker, list); | |
7304 | } | |
7305 | ||
7306 | void bdrv_op_unblock(BlockDriverState *bs, BlockOpType op, Error *reason) | |
7307 | { | |
7308 | BdrvOpBlocker *blocker, *next; | |
7309 | GLOBAL_STATE_CODE(); | |
7310 | assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX); | |
7311 | QLIST_FOREACH_SAFE(blocker, &bs->op_blockers[op], list, next) { | |
7312 | if (blocker->reason == reason) { | |
7313 | QLIST_REMOVE(blocker, list); | |
7314 | g_free(blocker); | |
7315 | } | |
7316 | } | |
7317 | } | |
7318 | ||
7319 | void bdrv_op_block_all(BlockDriverState *bs, Error *reason) | |
7320 | { | |
7321 | int i; | |
7322 | GLOBAL_STATE_CODE(); | |
7323 | for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) { | |
7324 | bdrv_op_block(bs, i, reason); | |
7325 | } | |
7326 | } | |
7327 | ||
7328 | void bdrv_op_unblock_all(BlockDriverState *bs, Error *reason) | |
7329 | { | |
7330 | int i; | |
7331 | GLOBAL_STATE_CODE(); | |
7332 | for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) { | |
7333 | bdrv_op_unblock(bs, i, reason); | |
7334 | } | |
7335 | } | |
7336 | ||
7337 | bool bdrv_op_blocker_is_empty(BlockDriverState *bs) | |
7338 | { | |
7339 | int i; | |
7340 | GLOBAL_STATE_CODE(); | |
7341 | for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) { | |
7342 | if (!QLIST_EMPTY(&bs->op_blockers[i])) { | |
7343 | return false; | |
7344 | } | |
7345 | } | |
7346 | return true; | |
7347 | } | |
7348 | ||
7349 | /* | |
7350 | * Must not be called while holding the lock of an AioContext other than the | |
7351 | * current one. | |
7352 | */ | |
7353 | void bdrv_img_create(const char *filename, const char *fmt, | |
7354 | const char *base_filename, const char *base_fmt, | |
7355 | char *options, uint64_t img_size, int flags, bool quiet, | |
7356 | Error **errp) | |
7357 | { | |
7358 | QemuOptsList *create_opts = NULL; | |
7359 | QemuOpts *opts = NULL; | |
7360 | const char *backing_fmt, *backing_file; | |
7361 | int64_t size; | |
7362 | BlockDriver *drv, *proto_drv; | |
7363 | Error *local_err = NULL; | |
7364 | int ret = 0; | |
7365 | ||
7366 | GLOBAL_STATE_CODE(); | |
7367 | ||
7368 | /* Find driver and parse its options */ | |
7369 | drv = bdrv_find_format(fmt); | |
7370 | if (!drv) { | |
7371 | error_setg(errp, "Unknown file format '%s'", fmt); | |
7372 | return; | |
7373 | } | |
7374 | ||
7375 | proto_drv = bdrv_find_protocol(filename, true, errp); | |
7376 | if (!proto_drv) { | |
7377 | return; | |
7378 | } | |
7379 | ||
7380 | if (!drv->create_opts) { | |
7381 | error_setg(errp, "Format driver '%s' does not support image creation", | |
7382 | drv->format_name); | |
7383 | return; | |
7384 | } | |
7385 | ||
7386 | if (!proto_drv->create_opts) { | |
7387 | error_setg(errp, "Protocol driver '%s' does not support image creation", | |
7388 | proto_drv->format_name); | |
7389 | return; | |
7390 | } | |
7391 | ||
7392 | aio_context_acquire(qemu_get_aio_context()); | |
7393 | ||
7394 | /* Create parameter list */ | |
7395 | create_opts = qemu_opts_append(create_opts, drv->create_opts); | |
7396 | create_opts = qemu_opts_append(create_opts, proto_drv->create_opts); | |
7397 | ||
7398 | opts = qemu_opts_create(create_opts, NULL, 0, &error_abort); | |
7399 | ||
7400 | /* Parse -o options */ | |
7401 | if (options) { | |
7402 | if (!qemu_opts_do_parse(opts, options, NULL, errp)) { | |
7403 | goto out; | |
7404 | } | |
7405 | } | |
7406 | ||
7407 | if (!qemu_opt_get(opts, BLOCK_OPT_SIZE)) { | |
7408 | qemu_opt_set_number(opts, BLOCK_OPT_SIZE, img_size, &error_abort); | |
7409 | } else if (img_size != UINT64_C(-1)) { | |
7410 | error_setg(errp, "The image size must be specified only once"); | |
7411 | goto out; | |
7412 | } | |
7413 | ||
7414 | if (base_filename) { | |
7415 | if (!qemu_opt_set(opts, BLOCK_OPT_BACKING_FILE, base_filename, | |
7416 | NULL)) { | |
7417 | error_setg(errp, "Backing file not supported for file format '%s'", | |
7418 | fmt); | |
7419 | goto out; | |
7420 | } | |
7421 | } | |
7422 | ||
7423 | if (base_fmt) { | |
7424 | if (!qemu_opt_set(opts, BLOCK_OPT_BACKING_FMT, base_fmt, NULL)) { | |
7425 | error_setg(errp, "Backing file format not supported for file " | |
7426 | "format '%s'", fmt); | |
7427 | goto out; | |
7428 | } | |
7429 | } | |
7430 | ||
7431 | backing_file = qemu_opt_get(opts, BLOCK_OPT_BACKING_FILE); | |
7432 | if (backing_file) { | |
7433 | if (!strcmp(filename, backing_file)) { | |
7434 | error_setg(errp, "Error: Trying to create an image with the " | |
7435 | "same filename as the backing file"); | |
7436 | goto out; | |
7437 | } | |
7438 | if (backing_file[0] == '\0') { | |
7439 | error_setg(errp, "Expected backing file name, got empty string"); | |
7440 | goto out; | |
7441 | } | |
7442 | } | |
7443 | ||
7444 | backing_fmt = qemu_opt_get(opts, BLOCK_OPT_BACKING_FMT); | |
7445 | ||
7446 | /* The size for the image must always be specified, unless we have a backing | |
7447 | * file and we have not been forbidden from opening it. */ | |
7448 | size = qemu_opt_get_size(opts, BLOCK_OPT_SIZE, img_size); | |
7449 | if (backing_file && !(flags & BDRV_O_NO_BACKING)) { | |
7450 | BlockDriverState *bs; | |
7451 | char *full_backing; | |
7452 | int back_flags; | |
7453 | QDict *backing_options = NULL; | |
7454 | ||
7455 | full_backing = | |
7456 | bdrv_get_full_backing_filename_from_filename(filename, backing_file, | |
7457 | &local_err); | |
7458 | if (local_err) { | |
7459 | goto out; | |
7460 | } | |
7461 | assert(full_backing); | |
7462 | ||
7463 | /* | |
7464 | * No need to do I/O here, which allows us to open encrypted | |
7465 | * backing images without needing the secret | |
7466 | */ | |
7467 | back_flags = flags; | |
7468 | back_flags &= ~(BDRV_O_RDWR | BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING); | |
7469 | back_flags |= BDRV_O_NO_IO; | |
7470 | ||
7471 | backing_options = qdict_new(); | |
7472 | if (backing_fmt) { | |
7473 | qdict_put_str(backing_options, "driver", backing_fmt); | |
7474 | } | |
7475 | qdict_put_bool(backing_options, BDRV_OPT_FORCE_SHARE, true); | |
7476 | ||
7477 | bs = bdrv_open(full_backing, NULL, backing_options, back_flags, | |
7478 | &local_err); | |
7479 | g_free(full_backing); | |
7480 | if (!bs) { | |
7481 | error_append_hint(&local_err, "Could not open backing image.\n"); | |
7482 | goto out; | |
7483 | } else { | |
7484 | if (!backing_fmt) { | |
7485 | error_setg(&local_err, | |
7486 | "Backing file specified without backing format"); | |
7487 | error_append_hint(&local_err, "Detected format of %s.\n", | |
7488 | bs->drv->format_name); | |
7489 | goto out; | |
7490 | } | |
7491 | if (size == -1) { | |
7492 | /* Opened BS, have no size */ | |
7493 | size = bdrv_getlength(bs); | |
7494 | if (size < 0) { | |
7495 | error_setg_errno(errp, -size, "Could not get size of '%s'", | |
7496 | backing_file); | |
7497 | bdrv_unref(bs); | |
7498 | goto out; | |
7499 | } | |
7500 | qemu_opt_set_number(opts, BLOCK_OPT_SIZE, size, &error_abort); | |
7501 | } | |
7502 | bdrv_unref(bs); | |
7503 | } | |
7504 | /* (backing_file && !(flags & BDRV_O_NO_BACKING)) */ | |
7505 | } else if (backing_file && !backing_fmt) { | |
7506 | error_setg(&local_err, | |
7507 | "Backing file specified without backing format"); | |
7508 | goto out; | |
7509 | } | |
7510 | ||
7511 | if (size == -1) { | |
7512 | error_setg(errp, "Image creation needs a size parameter"); | |
7513 | goto out; | |
7514 | } | |
7515 | ||
7516 | if (!quiet) { | |
7517 | printf("Formatting '%s', fmt=%s ", filename, fmt); | |
7518 | qemu_opts_print(opts, " "); | |
7519 | puts(""); | |
7520 | fflush(stdout); | |
7521 | } | |
7522 | ||
7523 | ret = bdrv_create(drv, filename, opts, &local_err); | |
7524 | ||
7525 | if (ret == -EFBIG) { | |
7526 | /* This is generally a better message than whatever the driver would | |
7527 | * deliver (especially because of the cluster_size_hint), since that | |
7528 | * is most probably not much different from "image too large". */ | |
7529 | const char *cluster_size_hint = ""; | |
7530 | if (qemu_opt_get_size(opts, BLOCK_OPT_CLUSTER_SIZE, 0)) { | |
7531 | cluster_size_hint = " (try using a larger cluster size)"; | |
7532 | } | |
7533 | error_setg(errp, "The image size is too large for file format '%s'" | |
7534 | "%s", fmt, cluster_size_hint); | |
7535 | error_free(local_err); | |
7536 | local_err = NULL; | |
7537 | } | |
7538 | ||
7539 | out: | |
7540 | qemu_opts_del(opts); | |
7541 | qemu_opts_free(create_opts); | |
7542 | error_propagate(errp, local_err); | |
7543 | aio_context_release(qemu_get_aio_context()); | |
7544 | } | |
7545 | ||
7546 | AioContext *bdrv_get_aio_context(BlockDriverState *bs) | |
7547 | { | |
7548 | IO_CODE(); | |
7549 | return bs ? bs->aio_context : qemu_get_aio_context(); | |
7550 | } | |
7551 | ||
7552 | AioContext *coroutine_fn bdrv_co_enter(BlockDriverState *bs) | |
7553 | { | |
7554 | Coroutine *self = qemu_coroutine_self(); | |
7555 | AioContext *old_ctx = qemu_coroutine_get_aio_context(self); | |
7556 | AioContext *new_ctx; | |
7557 | IO_CODE(); | |
7558 | ||
7559 | /* | |
7560 | * Increase bs->in_flight to ensure that this operation is completed before | |
7561 | * moving the node to a different AioContext. Read new_ctx only afterwards. | |
7562 | */ | |
7563 | bdrv_inc_in_flight(bs); | |
7564 | ||
7565 | new_ctx = bdrv_get_aio_context(bs); | |
7566 | aio_co_reschedule_self(new_ctx); | |
7567 | return old_ctx; | |
7568 | } | |
7569 | ||
7570 | void coroutine_fn bdrv_co_leave(BlockDriverState *bs, AioContext *old_ctx) | |
7571 | { | |
7572 | IO_CODE(); | |
7573 | aio_co_reschedule_self(old_ctx); | |
7574 | bdrv_dec_in_flight(bs); | |
7575 | } | |
7576 | ||
7577 | void coroutine_fn bdrv_co_lock(BlockDriverState *bs) | |
7578 | { | |
7579 | AioContext *ctx = bdrv_get_aio_context(bs); | |
7580 | ||
7581 | /* In the main thread, bs->aio_context won't change concurrently */ | |
7582 | assert(qemu_get_current_aio_context() == qemu_get_aio_context()); | |
7583 | ||
7584 | /* | |
7585 | * We're in coroutine context, so we already hold the lock of the main | |
7586 | * loop AioContext. Don't lock it twice to avoid deadlocks. | |
7587 | */ | |
7588 | assert(qemu_in_coroutine()); | |
7589 | if (ctx != qemu_get_aio_context()) { | |
7590 | aio_context_acquire(ctx); | |
7591 | } | |
7592 | } | |
7593 | ||
7594 | void coroutine_fn bdrv_co_unlock(BlockDriverState *bs) | |
7595 | { | |
7596 | AioContext *ctx = bdrv_get_aio_context(bs); | |
7597 | ||
7598 | assert(qemu_in_coroutine()); | |
7599 | if (ctx != qemu_get_aio_context()) { | |
7600 | aio_context_release(ctx); | |
7601 | } | |
7602 | } | |
7603 | ||
7604 | static void bdrv_do_remove_aio_context_notifier(BdrvAioNotifier *ban) | |
7605 | { | |
7606 | GLOBAL_STATE_CODE(); | |
7607 | QLIST_REMOVE(ban, list); | |
7608 | g_free(ban); | |
7609 | } | |
7610 | ||
7611 | static void bdrv_detach_aio_context(BlockDriverState *bs) | |
7612 | { | |
7613 | BdrvAioNotifier *baf, *baf_tmp; | |
7614 | ||
7615 | assert(!bs->walking_aio_notifiers); | |
7616 | GLOBAL_STATE_CODE(); | |
7617 | bs->walking_aio_notifiers = true; | |
7618 | QLIST_FOREACH_SAFE(baf, &bs->aio_notifiers, list, baf_tmp) { | |
7619 | if (baf->deleted) { | |
7620 | bdrv_do_remove_aio_context_notifier(baf); | |
7621 | } else { | |
7622 | baf->detach_aio_context(baf->opaque); | |
7623 | } | |
7624 | } | |
7625 | /* Never mind iterating again to check for ->deleted. bdrv_close() will | |
7626 | * remove remaining aio notifiers if we aren't called again. | |
7627 | */ | |
7628 | bs->walking_aio_notifiers = false; | |
7629 | ||
7630 | if (bs->drv && bs->drv->bdrv_detach_aio_context) { | |
7631 | bs->drv->bdrv_detach_aio_context(bs); | |
7632 | } | |
7633 | ||
7634 | bs->aio_context = NULL; | |
7635 | } | |
7636 | ||
7637 | static void bdrv_attach_aio_context(BlockDriverState *bs, | |
7638 | AioContext *new_context) | |
7639 | { | |
7640 | BdrvAioNotifier *ban, *ban_tmp; | |
7641 | GLOBAL_STATE_CODE(); | |
7642 | ||
7643 | bs->aio_context = new_context; | |
7644 | ||
7645 | if (bs->drv && bs->drv->bdrv_attach_aio_context) { | |
7646 | bs->drv->bdrv_attach_aio_context(bs, new_context); | |
7647 | } | |
7648 | ||
7649 | assert(!bs->walking_aio_notifiers); | |
7650 | bs->walking_aio_notifiers = true; | |
7651 | QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_tmp) { | |
7652 | if (ban->deleted) { | |
7653 | bdrv_do_remove_aio_context_notifier(ban); | |
7654 | } else { | |
7655 | ban->attached_aio_context(new_context, ban->opaque); | |
7656 | } | |
7657 | } | |
7658 | bs->walking_aio_notifiers = false; | |
7659 | } | |
7660 | ||
7661 | typedef struct BdrvStateSetAioContext { | |
7662 | AioContext *new_ctx; | |
7663 | BlockDriverState *bs; | |
7664 | } BdrvStateSetAioContext; | |
7665 | ||
7666 | static bool bdrv_parent_change_aio_context(BdrvChild *c, AioContext *ctx, | |
7667 | GHashTable *visited, | |
7668 | Transaction *tran, | |
7669 | Error **errp) | |
7670 | { | |
7671 | GLOBAL_STATE_CODE(); | |
7672 | if (g_hash_table_contains(visited, c)) { | |
7673 | return true; | |
7674 | } | |
7675 | g_hash_table_add(visited, c); | |
7676 | ||
7677 | /* | |
7678 | * A BdrvChildClass that doesn't handle AioContext changes cannot | |
7679 | * tolerate any AioContext changes | |
7680 | */ | |
7681 | if (!c->klass->change_aio_ctx) { | |
7682 | char *user = bdrv_child_user_desc(c); | |
7683 | error_setg(errp, "Changing iothreads is not supported by %s", user); | |
7684 | g_free(user); | |
7685 | return false; | |
7686 | } | |
7687 | if (!c->klass->change_aio_ctx(c, ctx, visited, tran, errp)) { | |
7688 | assert(!errp || *errp); | |
7689 | return false; | |
7690 | } | |
7691 | return true; | |
7692 | } | |
7693 | ||
7694 | bool bdrv_child_change_aio_context(BdrvChild *c, AioContext *ctx, | |
7695 | GHashTable *visited, Transaction *tran, | |
7696 | Error **errp) | |
7697 | { | |
7698 | GLOBAL_STATE_CODE(); | |
7699 | if (g_hash_table_contains(visited, c)) { | |
7700 | return true; | |
7701 | } | |
7702 | g_hash_table_add(visited, c); | |
7703 | return bdrv_change_aio_context(c->bs, ctx, visited, tran, errp); | |
7704 | } | |
7705 | ||
7706 | static void bdrv_set_aio_context_clean(void *opaque) | |
7707 | { | |
7708 | BdrvStateSetAioContext *state = (BdrvStateSetAioContext *) opaque; | |
7709 | BlockDriverState *bs = (BlockDriverState *) state->bs; | |
7710 | ||
7711 | /* Paired with bdrv_drained_begin in bdrv_change_aio_context() */ | |
7712 | bdrv_drained_end(bs); | |
7713 | ||
7714 | g_free(state); | |
7715 | } | |
7716 | ||
7717 | static void bdrv_set_aio_context_commit(void *opaque) | |
7718 | { | |
7719 | BdrvStateSetAioContext *state = (BdrvStateSetAioContext *) opaque; | |
7720 | BlockDriverState *bs = (BlockDriverState *) state->bs; | |
7721 | AioContext *new_context = state->new_ctx; | |
7722 | AioContext *old_context = bdrv_get_aio_context(bs); | |
7723 | ||
7724 | /* | |
7725 | * Take the old AioContex when detaching it from bs. | |
7726 | * At this point, new_context lock is already acquired, and we are now | |
7727 | * also taking old_context. This is safe as long as bdrv_detach_aio_context | |
7728 | * does not call AIO_POLL_WHILE(). | |
7729 | */ | |
7730 | if (old_context != qemu_get_aio_context()) { | |
7731 | aio_context_acquire(old_context); | |
7732 | } | |
7733 | bdrv_detach_aio_context(bs); | |
7734 | if (old_context != qemu_get_aio_context()) { | |
7735 | aio_context_release(old_context); | |
7736 | } | |
7737 | bdrv_attach_aio_context(bs, new_context); | |
7738 | } | |
7739 | ||
7740 | static TransactionActionDrv set_aio_context = { | |
7741 | .commit = bdrv_set_aio_context_commit, | |
7742 | .clean = bdrv_set_aio_context_clean, | |
7743 | }; | |
7744 | ||
7745 | /* | |
7746 | * Changes the AioContext used for fd handlers, timers, and BHs by this | |
7747 | * BlockDriverState and all its children and parents. | |
7748 | * | |
7749 | * Must be called from the main AioContext. | |
7750 | * | |
7751 | * The caller must own the AioContext lock for the old AioContext of bs, but it | |
7752 | * must not own the AioContext lock for new_context (unless new_context is the | |
7753 | * same as the current context of bs). | |
7754 | * | |
7755 | * @visited will accumulate all visited BdrvChild objects. The caller is | |
7756 | * responsible for freeing the list afterwards. | |
7757 | */ | |
7758 | static bool bdrv_change_aio_context(BlockDriverState *bs, AioContext *ctx, | |
7759 | GHashTable *visited, Transaction *tran, | |
7760 | Error **errp) | |
7761 | { | |
7762 | BdrvChild *c; | |
7763 | BdrvStateSetAioContext *state; | |
7764 | ||
7765 | GLOBAL_STATE_CODE(); | |
7766 | ||
7767 | if (bdrv_get_aio_context(bs) == ctx) { | |
7768 | return true; | |
7769 | } | |
7770 | ||
7771 | bdrv_graph_rdlock_main_loop(); | |
7772 | QLIST_FOREACH(c, &bs->parents, next_parent) { | |
7773 | if (!bdrv_parent_change_aio_context(c, ctx, visited, tran, errp)) { | |
7774 | bdrv_graph_rdunlock_main_loop(); | |
7775 | return false; | |
7776 | } | |
7777 | } | |
7778 | ||
7779 | QLIST_FOREACH(c, &bs->children, next) { | |
7780 | if (!bdrv_child_change_aio_context(c, ctx, visited, tran, errp)) { | |
7781 | bdrv_graph_rdunlock_main_loop(); | |
7782 | return false; | |
7783 | } | |
7784 | } | |
7785 | bdrv_graph_rdunlock_main_loop(); | |
7786 | ||
7787 | state = g_new(BdrvStateSetAioContext, 1); | |
7788 | *state = (BdrvStateSetAioContext) { | |
7789 | .new_ctx = ctx, | |
7790 | .bs = bs, | |
7791 | }; | |
7792 | ||
7793 | /* Paired with bdrv_drained_end in bdrv_set_aio_context_clean() */ | |
7794 | bdrv_drained_begin(bs); | |
7795 | ||
7796 | tran_add(tran, &set_aio_context, state); | |
7797 | ||
7798 | return true; | |
7799 | } | |
7800 | ||
7801 | /* | |
7802 | * Change bs's and recursively all of its parents' and children's AioContext | |
7803 | * to the given new context, returning an error if that isn't possible. | |
7804 | * | |
7805 | * If ignore_child is not NULL, that child (and its subgraph) will not | |
7806 | * be touched. | |
7807 | * | |
7808 | * This function still requires the caller to take the bs current | |
7809 | * AioContext lock, otherwise draining will fail since AIO_WAIT_WHILE | |
7810 | * assumes the lock is always held if bs is in another AioContext. | |
7811 | * For the same reason, it temporarily also holds the new AioContext, since | |
7812 | * bdrv_drained_end calls BDRV_POLL_WHILE that assumes the lock is taken too. | |
7813 | * Therefore the new AioContext lock must not be taken by the caller. | |
7814 | */ | |
7815 | int bdrv_try_change_aio_context(BlockDriverState *bs, AioContext *ctx, | |
7816 | BdrvChild *ignore_child, Error **errp) | |
7817 | { | |
7818 | Transaction *tran; | |
7819 | GHashTable *visited; | |
7820 | int ret; | |
7821 | AioContext *old_context = bdrv_get_aio_context(bs); | |
7822 | GLOBAL_STATE_CODE(); | |
7823 | ||
7824 | /* | |
7825 | * Recursion phase: go through all nodes of the graph. | |
7826 | * Take care of checking that all nodes support changing AioContext | |
7827 | * and drain them, building a linear list of callbacks to run if everything | |
7828 | * is successful (the transaction itself). | |
7829 | */ | |
7830 | tran = tran_new(); | |
7831 | visited = g_hash_table_new(NULL, NULL); | |
7832 | if (ignore_child) { | |
7833 | g_hash_table_add(visited, ignore_child); | |
7834 | } | |
7835 | ret = bdrv_change_aio_context(bs, ctx, visited, tran, errp); | |
7836 | g_hash_table_destroy(visited); | |
7837 | ||
7838 | /* | |
7839 | * Linear phase: go through all callbacks collected in the transaction. | |
7840 | * Run all callbacks collected in the recursion to switch all nodes | |
7841 | * AioContext lock (transaction commit), or undo all changes done in the | |
7842 | * recursion (transaction abort). | |
7843 | */ | |
7844 | ||
7845 | if (!ret) { | |
7846 | /* Just run clean() callbacks. No AioContext changed. */ | |
7847 | tran_abort(tran); | |
7848 | return -EPERM; | |
7849 | } | |
7850 | ||
7851 | /* | |
7852 | * Release old AioContext, it won't be needed anymore, as all | |
7853 | * bdrv_drained_begin() have been called already. | |
7854 | */ | |
7855 | if (qemu_get_aio_context() != old_context) { | |
7856 | aio_context_release(old_context); | |
7857 | } | |
7858 | ||
7859 | /* | |
7860 | * Acquire new AioContext since bdrv_drained_end() is going to be called | |
7861 | * after we switched all nodes in the new AioContext, and the function | |
7862 | * assumes that the lock of the bs is always taken. | |
7863 | */ | |
7864 | if (qemu_get_aio_context() != ctx) { | |
7865 | aio_context_acquire(ctx); | |
7866 | } | |
7867 | ||
7868 | tran_commit(tran); | |
7869 | ||
7870 | if (qemu_get_aio_context() != ctx) { | |
7871 | aio_context_release(ctx); | |
7872 | } | |
7873 | ||
7874 | /* Re-acquire the old AioContext, since the caller takes and releases it. */ | |
7875 | if (qemu_get_aio_context() != old_context) { | |
7876 | aio_context_acquire(old_context); | |
7877 | } | |
7878 | ||
7879 | return 0; | |
7880 | } | |
7881 | ||
7882 | void bdrv_add_aio_context_notifier(BlockDriverState *bs, | |
7883 | void (*attached_aio_context)(AioContext *new_context, void *opaque), | |
7884 | void (*detach_aio_context)(void *opaque), void *opaque) | |
7885 | { | |
7886 | BdrvAioNotifier *ban = g_new(BdrvAioNotifier, 1); | |
7887 | *ban = (BdrvAioNotifier){ | |
7888 | .attached_aio_context = attached_aio_context, | |
7889 | .detach_aio_context = detach_aio_context, | |
7890 | .opaque = opaque | |
7891 | }; | |
7892 | GLOBAL_STATE_CODE(); | |
7893 | ||
7894 | QLIST_INSERT_HEAD(&bs->aio_notifiers, ban, list); | |
7895 | } | |
7896 | ||
7897 | void bdrv_remove_aio_context_notifier(BlockDriverState *bs, | |
7898 | void (*attached_aio_context)(AioContext *, | |
7899 | void *), | |
7900 | void (*detach_aio_context)(void *), | |
7901 | void *opaque) | |
7902 | { | |
7903 | BdrvAioNotifier *ban, *ban_next; | |
7904 | GLOBAL_STATE_CODE(); | |
7905 | ||
7906 | QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_next) { | |
7907 | if (ban->attached_aio_context == attached_aio_context && | |
7908 | ban->detach_aio_context == detach_aio_context && | |
7909 | ban->opaque == opaque && | |
7910 | ban->deleted == false) | |
7911 | { | |
7912 | if (bs->walking_aio_notifiers) { | |
7913 | ban->deleted = true; | |
7914 | } else { | |
7915 | bdrv_do_remove_aio_context_notifier(ban); | |
7916 | } | |
7917 | return; | |
7918 | } | |
7919 | } | |
7920 | ||
7921 | abort(); | |
7922 | } | |
7923 | ||
7924 | int bdrv_amend_options(BlockDriverState *bs, QemuOpts *opts, | |
7925 | BlockDriverAmendStatusCB *status_cb, void *cb_opaque, | |
7926 | bool force, | |
7927 | Error **errp) | |
7928 | { | |
7929 | GLOBAL_STATE_CODE(); | |
7930 | if (!bs->drv) { | |
7931 | error_setg(errp, "Node is ejected"); | |
7932 | return -ENOMEDIUM; | |
7933 | } | |
7934 | if (!bs->drv->bdrv_amend_options) { | |
7935 | error_setg(errp, "Block driver '%s' does not support option amendment", | |
7936 | bs->drv->format_name); | |
7937 | return -ENOTSUP; | |
7938 | } | |
7939 | return bs->drv->bdrv_amend_options(bs, opts, status_cb, | |
7940 | cb_opaque, force, errp); | |
7941 | } | |
7942 | ||
7943 | /* | |
7944 | * This function checks whether the given @to_replace is allowed to be | |
7945 | * replaced by a node that always shows the same data as @bs. This is | |
7946 | * used for example to verify whether the mirror job can replace | |
7947 | * @to_replace by the target mirrored from @bs. | |
7948 | * To be replaceable, @bs and @to_replace may either be guaranteed to | |
7949 | * always show the same data (because they are only connected through | |
7950 | * filters), or some driver may allow replacing one of its children | |
7951 | * because it can guarantee that this child's data is not visible at | |
7952 | * all (for example, for dissenting quorum children that have no other | |
7953 | * parents). | |
7954 | */ | |
7955 | bool bdrv_recurse_can_replace(BlockDriverState *bs, | |
7956 | BlockDriverState *to_replace) | |
7957 | { | |
7958 | BlockDriverState *filtered; | |
7959 | ||
7960 | GLOBAL_STATE_CODE(); | |
7961 | ||
7962 | if (!bs || !bs->drv) { | |
7963 | return false; | |
7964 | } | |
7965 | ||
7966 | if (bs == to_replace) { | |
7967 | return true; | |
7968 | } | |
7969 | ||
7970 | /* See what the driver can do */ | |
7971 | if (bs->drv->bdrv_recurse_can_replace) { | |
7972 | return bs->drv->bdrv_recurse_can_replace(bs, to_replace); | |
7973 | } | |
7974 | ||
7975 | /* For filters without an own implementation, we can recurse on our own */ | |
7976 | filtered = bdrv_filter_bs(bs); | |
7977 | if (filtered) { | |
7978 | return bdrv_recurse_can_replace(filtered, to_replace); | |
7979 | } | |
7980 | ||
7981 | /* Safe default */ | |
7982 | return false; | |
7983 | } | |
7984 | ||
7985 | /* | |
7986 | * Check whether the given @node_name can be replaced by a node that | |
7987 | * has the same data as @parent_bs. If so, return @node_name's BDS; | |
7988 | * NULL otherwise. | |
7989 | * | |
7990 | * @node_name must be a (recursive) *child of @parent_bs (or this | |
7991 | * function will return NULL). | |
7992 | * | |
7993 | * The result (whether the node can be replaced or not) is only valid | |
7994 | * for as long as no graph or permission changes occur. | |
7995 | */ | |
7996 | BlockDriverState *check_to_replace_node(BlockDriverState *parent_bs, | |
7997 | const char *node_name, Error **errp) | |
7998 | { | |
7999 | BlockDriverState *to_replace_bs = bdrv_find_node(node_name); | |
8000 | AioContext *aio_context; | |
8001 | ||
8002 | GLOBAL_STATE_CODE(); | |
8003 | ||
8004 | if (!to_replace_bs) { | |
8005 | error_setg(errp, "Failed to find node with node-name='%s'", node_name); | |
8006 | return NULL; | |
8007 | } | |
8008 | ||
8009 | aio_context = bdrv_get_aio_context(to_replace_bs); | |
8010 | aio_context_acquire(aio_context); | |
8011 | ||
8012 | if (bdrv_op_is_blocked(to_replace_bs, BLOCK_OP_TYPE_REPLACE, errp)) { | |
8013 | to_replace_bs = NULL; | |
8014 | goto out; | |
8015 | } | |
8016 | ||
8017 | /* We don't want arbitrary node of the BDS chain to be replaced only the top | |
8018 | * most non filter in order to prevent data corruption. | |
8019 | * Another benefit is that this tests exclude backing files which are | |
8020 | * blocked by the backing blockers. | |
8021 | */ | |
8022 | if (!bdrv_recurse_can_replace(parent_bs, to_replace_bs)) { | |
8023 | error_setg(errp, "Cannot replace '%s' by a node mirrored from '%s', " | |
8024 | "because it cannot be guaranteed that doing so would not " | |
8025 | "lead to an abrupt change of visible data", | |
8026 | node_name, parent_bs->node_name); | |
8027 | to_replace_bs = NULL; | |
8028 | goto out; | |
8029 | } | |
8030 | ||
8031 | out: | |
8032 | aio_context_release(aio_context); | |
8033 | return to_replace_bs; | |
8034 | } | |
8035 | ||
8036 | /** | |
8037 | * Iterates through the list of runtime option keys that are said to | |
8038 | * be "strong" for a BDS. An option is called "strong" if it changes | |
8039 | * a BDS's data. For example, the null block driver's "size" and | |
8040 | * "read-zeroes" options are strong, but its "latency-ns" option is | |
8041 | * not. | |
8042 | * | |
8043 | * If a key returned by this function ends with a dot, all options | |
8044 | * starting with that prefix are strong. | |
8045 | */ | |
8046 | static const char *const *strong_options(BlockDriverState *bs, | |
8047 | const char *const *curopt) | |
8048 | { | |
8049 | static const char *const global_options[] = { | |
8050 | "driver", "filename", NULL | |
8051 | }; | |
8052 | ||
8053 | if (!curopt) { | |
8054 | return &global_options[0]; | |
8055 | } | |
8056 | ||
8057 | curopt++; | |
8058 | if (curopt == &global_options[ARRAY_SIZE(global_options) - 1] && bs->drv) { | |
8059 | curopt = bs->drv->strong_runtime_opts; | |
8060 | } | |
8061 | ||
8062 | return (curopt && *curopt) ? curopt : NULL; | |
8063 | } | |
8064 | ||
8065 | /** | |
8066 | * Copies all strong runtime options from bs->options to the given | |
8067 | * QDict. The set of strong option keys is determined by invoking | |
8068 | * strong_options(). | |
8069 | * | |
8070 | * Returns true iff any strong option was present in bs->options (and | |
8071 | * thus copied to the target QDict) with the exception of "filename" | |
8072 | * and "driver". The caller is expected to use this value to decide | |
8073 | * whether the existence of strong options prevents the generation of | |
8074 | * a plain filename. | |
8075 | */ | |
8076 | static bool append_strong_runtime_options(QDict *d, BlockDriverState *bs) | |
8077 | { | |
8078 | bool found_any = false; | |
8079 | const char *const *option_name = NULL; | |
8080 | ||
8081 | if (!bs->drv) { | |
8082 | return false; | |
8083 | } | |
8084 | ||
8085 | while ((option_name = strong_options(bs, option_name))) { | |
8086 | bool option_given = false; | |
8087 | ||
8088 | assert(strlen(*option_name) > 0); | |
8089 | if ((*option_name)[strlen(*option_name) - 1] != '.') { | |
8090 | QObject *entry = qdict_get(bs->options, *option_name); | |
8091 | if (!entry) { | |
8092 | continue; | |
8093 | } | |
8094 | ||
8095 | qdict_put_obj(d, *option_name, qobject_ref(entry)); | |
8096 | option_given = true; | |
8097 | } else { | |
8098 | const QDictEntry *entry; | |
8099 | for (entry = qdict_first(bs->options); entry; | |
8100 | entry = qdict_next(bs->options, entry)) | |
8101 | { | |
8102 | if (strstart(qdict_entry_key(entry), *option_name, NULL)) { | |
8103 | qdict_put_obj(d, qdict_entry_key(entry), | |
8104 | qobject_ref(qdict_entry_value(entry))); | |
8105 | option_given = true; | |
8106 | } | |
8107 | } | |
8108 | } | |
8109 | ||
8110 | /* While "driver" and "filename" need to be included in a JSON filename, | |
8111 | * their existence does not prohibit generation of a plain filename. */ | |
8112 | if (!found_any && option_given && | |
8113 | strcmp(*option_name, "driver") && strcmp(*option_name, "filename")) | |
8114 | { | |
8115 | found_any = true; | |
8116 | } | |
8117 | } | |
8118 | ||
8119 | if (!qdict_haskey(d, "driver")) { | |
8120 | /* Drivers created with bdrv_new_open_driver() may not have a | |
8121 | * @driver option. Add it here. */ | |
8122 | qdict_put_str(d, "driver", bs->drv->format_name); | |
8123 | } | |
8124 | ||
8125 | return found_any; | |
8126 | } | |
8127 | ||
8128 | /* Note: This function may return false positives; it may return true | |
8129 | * even if opening the backing file specified by bs's image header | |
8130 | * would result in exactly bs->backing. */ | |
8131 | static bool GRAPH_RDLOCK bdrv_backing_overridden(BlockDriverState *bs) | |
8132 | { | |
8133 | GLOBAL_STATE_CODE(); | |
8134 | if (bs->backing) { | |
8135 | return strcmp(bs->auto_backing_file, | |
8136 | bs->backing->bs->filename); | |
8137 | } else { | |
8138 | /* No backing BDS, so if the image header reports any backing | |
8139 | * file, it must have been suppressed */ | |
8140 | return bs->auto_backing_file[0] != '\0'; | |
8141 | } | |
8142 | } | |
8143 | ||
8144 | /* Updates the following BDS fields: | |
8145 | * - exact_filename: A filename which may be used for opening a block device | |
8146 | * which (mostly) equals the given BDS (even without any | |
8147 | * other options; so reading and writing must return the same | |
8148 | * results, but caching etc. may be different) | |
8149 | * - full_open_options: Options which, when given when opening a block device | |
8150 | * (without a filename), result in a BDS (mostly) | |
8151 | * equalling the given one | |
8152 | * - filename: If exact_filename is set, it is copied here. Otherwise, | |
8153 | * full_open_options is converted to a JSON object, prefixed with | |
8154 | * "json:" (for use through the JSON pseudo protocol) and put here. | |
8155 | */ | |
8156 | void bdrv_refresh_filename(BlockDriverState *bs) | |
8157 | { | |
8158 | BlockDriver *drv = bs->drv; | |
8159 | BdrvChild *child; | |
8160 | BlockDriverState *primary_child_bs; | |
8161 | QDict *opts; | |
8162 | bool backing_overridden; | |
8163 | bool generate_json_filename; /* Whether our default implementation should | |
8164 | fill exact_filename (false) or not (true) */ | |
8165 | ||
8166 | GLOBAL_STATE_CODE(); | |
8167 | ||
8168 | if (!drv) { | |
8169 | return; | |
8170 | } | |
8171 | ||
8172 | /* This BDS's file name may depend on any of its children's file names, so | |
8173 | * refresh those first */ | |
8174 | QLIST_FOREACH(child, &bs->children, next) { | |
8175 | bdrv_refresh_filename(child->bs); | |
8176 | } | |
8177 | ||
8178 | if (bs->implicit) { | |
8179 | /* For implicit nodes, just copy everything from the single child */ | |
8180 | child = QLIST_FIRST(&bs->children); | |
8181 | assert(QLIST_NEXT(child, next) == NULL); | |
8182 | ||
8183 | pstrcpy(bs->exact_filename, sizeof(bs->exact_filename), | |
8184 | child->bs->exact_filename); | |
8185 | pstrcpy(bs->filename, sizeof(bs->filename), child->bs->filename); | |
8186 | ||
8187 | qobject_unref(bs->full_open_options); | |
8188 | bs->full_open_options = qobject_ref(child->bs->full_open_options); | |
8189 | ||
8190 | return; | |
8191 | } | |
8192 | ||
8193 | backing_overridden = bdrv_backing_overridden(bs); | |
8194 | ||
8195 | if (bs->open_flags & BDRV_O_NO_IO) { | |
8196 | /* Without I/O, the backing file does not change anything. | |
8197 | * Therefore, in such a case (primarily qemu-img), we can | |
8198 | * pretend the backing file has not been overridden even if | |
8199 | * it technically has been. */ | |
8200 | backing_overridden = false; | |
8201 | } | |
8202 | ||
8203 | /* Gather the options QDict */ | |
8204 | opts = qdict_new(); | |
8205 | generate_json_filename = append_strong_runtime_options(opts, bs); | |
8206 | generate_json_filename |= backing_overridden; | |
8207 | ||
8208 | if (drv->bdrv_gather_child_options) { | |
8209 | /* Some block drivers may not want to present all of their children's | |
8210 | * options, or name them differently from BdrvChild.name */ | |
8211 | drv->bdrv_gather_child_options(bs, opts, backing_overridden); | |
8212 | } else { | |
8213 | QLIST_FOREACH(child, &bs->children, next) { | |
8214 | if (child == bs->backing && !backing_overridden) { | |
8215 | /* We can skip the backing BDS if it has not been overridden */ | |
8216 | continue; | |
8217 | } | |
8218 | ||
8219 | qdict_put(opts, child->name, | |
8220 | qobject_ref(child->bs->full_open_options)); | |
8221 | } | |
8222 | ||
8223 | if (backing_overridden && !bs->backing) { | |
8224 | /* Force no backing file */ | |
8225 | qdict_put_null(opts, "backing"); | |
8226 | } | |
8227 | } | |
8228 | ||
8229 | qobject_unref(bs->full_open_options); | |
8230 | bs->full_open_options = opts; | |
8231 | ||
8232 | primary_child_bs = bdrv_primary_bs(bs); | |
8233 | ||
8234 | if (drv->bdrv_refresh_filename) { | |
8235 | /* Obsolete information is of no use here, so drop the old file name | |
8236 | * information before refreshing it */ | |
8237 | bs->exact_filename[0] = '\0'; | |
8238 | ||
8239 | drv->bdrv_refresh_filename(bs); | |
8240 | } else if (primary_child_bs) { | |
8241 | /* | |
8242 | * Try to reconstruct valid information from the underlying | |
8243 | * file -- this only works for format nodes (filter nodes | |
8244 | * cannot be probed and as such must be selected by the user | |
8245 | * either through an options dict, or through a special | |
8246 | * filename which the filter driver must construct in its | |
8247 | * .bdrv_refresh_filename() implementation). | |
8248 | */ | |
8249 | ||
8250 | bs->exact_filename[0] = '\0'; | |
8251 | ||
8252 | /* | |
8253 | * We can use the underlying file's filename if: | |
8254 | * - it has a filename, | |
8255 | * - the current BDS is not a filter, | |
8256 | * - the file is a protocol BDS, and | |
8257 | * - opening that file (as this BDS's format) will automatically create | |
8258 | * the BDS tree we have right now, that is: | |
8259 | * - the user did not significantly change this BDS's behavior with | |
8260 | * some explicit (strong) options | |
8261 | * - no non-file child of this BDS has been overridden by the user | |
8262 | * Both of these conditions are represented by generate_json_filename. | |
8263 | */ | |
8264 | if (primary_child_bs->exact_filename[0] && | |
8265 | primary_child_bs->drv->bdrv_file_open && | |
8266 | !drv->is_filter && !generate_json_filename) | |
8267 | { | |
8268 | strcpy(bs->exact_filename, primary_child_bs->exact_filename); | |
8269 | } | |
8270 | } | |
8271 | ||
8272 | if (bs->exact_filename[0]) { | |
8273 | pstrcpy(bs->filename, sizeof(bs->filename), bs->exact_filename); | |
8274 | } else { | |
8275 | GString *json = qobject_to_json(QOBJECT(bs->full_open_options)); | |
8276 | if (snprintf(bs->filename, sizeof(bs->filename), "json:%s", | |
8277 | json->str) >= sizeof(bs->filename)) { | |
8278 | /* Give user a hint if we truncated things. */ | |
8279 | strcpy(bs->filename + sizeof(bs->filename) - 4, "..."); | |
8280 | } | |
8281 | g_string_free(json, true); | |
8282 | } | |
8283 | } | |
8284 | ||
8285 | char *bdrv_dirname(BlockDriverState *bs, Error **errp) | |
8286 | { | |
8287 | BlockDriver *drv = bs->drv; | |
8288 | BlockDriverState *child_bs; | |
8289 | ||
8290 | GLOBAL_STATE_CODE(); | |
8291 | ||
8292 | if (!drv) { | |
8293 | error_setg(errp, "Node '%s' is ejected", bs->node_name); | |
8294 | return NULL; | |
8295 | } | |
8296 | ||
8297 | if (drv->bdrv_dirname) { | |
8298 | return drv->bdrv_dirname(bs, errp); | |
8299 | } | |
8300 | ||
8301 | child_bs = bdrv_primary_bs(bs); | |
8302 | if (child_bs) { | |
8303 | return bdrv_dirname(child_bs, errp); | |
8304 | } | |
8305 | ||
8306 | bdrv_refresh_filename(bs); | |
8307 | if (bs->exact_filename[0] != '\0') { | |
8308 | return path_combine(bs->exact_filename, ""); | |
8309 | } | |
8310 | ||
8311 | error_setg(errp, "Cannot generate a base directory for %s nodes", | |
8312 | drv->format_name); | |
8313 | return NULL; | |
8314 | } | |
8315 | ||
8316 | /* | |
8317 | * Hot add/remove a BDS's child. So the user can take a child offline when | |
8318 | * it is broken and take a new child online | |
8319 | */ | |
8320 | void bdrv_add_child(BlockDriverState *parent_bs, BlockDriverState *child_bs, | |
8321 | Error **errp) | |
8322 | { | |
8323 | GLOBAL_STATE_CODE(); | |
8324 | if (!parent_bs->drv || !parent_bs->drv->bdrv_add_child) { | |
8325 | error_setg(errp, "The node %s does not support adding a child", | |
8326 | bdrv_get_device_or_node_name(parent_bs)); | |
8327 | return; | |
8328 | } | |
8329 | ||
8330 | /* | |
8331 | * Non-zoned block drivers do not follow zoned storage constraints | |
8332 | * (i.e. sequential writes to zones). Refuse mixing zoned and non-zoned | |
8333 | * drivers in a graph. | |
8334 | */ | |
8335 | if (!parent_bs->drv->supports_zoned_children && | |
8336 | child_bs->bl.zoned == BLK_Z_HM) { | |
8337 | /* | |
8338 | * The host-aware model allows zoned storage constraints and random | |
8339 | * write. Allow mixing host-aware and non-zoned drivers. Using | |
8340 | * host-aware device as a regular device. | |
8341 | */ | |
8342 | error_setg(errp, "Cannot add a %s child to a %s parent", | |
8343 | child_bs->bl.zoned == BLK_Z_HM ? "zoned" : "non-zoned", | |
8344 | parent_bs->drv->supports_zoned_children ? | |
8345 | "support zoned children" : "not support zoned children"); | |
8346 | return; | |
8347 | } | |
8348 | ||
8349 | if (!QLIST_EMPTY(&child_bs->parents)) { | |
8350 | error_setg(errp, "The node %s already has a parent", | |
8351 | child_bs->node_name); | |
8352 | return; | |
8353 | } | |
8354 | ||
8355 | parent_bs->drv->bdrv_add_child(parent_bs, child_bs, errp); | |
8356 | } | |
8357 | ||
8358 | void bdrv_del_child(BlockDriverState *parent_bs, BdrvChild *child, Error **errp) | |
8359 | { | |
8360 | BdrvChild *tmp; | |
8361 | ||
8362 | GLOBAL_STATE_CODE(); | |
8363 | if (!parent_bs->drv || !parent_bs->drv->bdrv_del_child) { | |
8364 | error_setg(errp, "The node %s does not support removing a child", | |
8365 | bdrv_get_device_or_node_name(parent_bs)); | |
8366 | return; | |
8367 | } | |
8368 | ||
8369 | QLIST_FOREACH(tmp, &parent_bs->children, next) { | |
8370 | if (tmp == child) { | |
8371 | break; | |
8372 | } | |
8373 | } | |
8374 | ||
8375 | if (!tmp) { | |
8376 | error_setg(errp, "The node %s does not have a child named %s", | |
8377 | bdrv_get_device_or_node_name(parent_bs), | |
8378 | bdrv_get_device_or_node_name(child->bs)); | |
8379 | return; | |
8380 | } | |
8381 | ||
8382 | parent_bs->drv->bdrv_del_child(parent_bs, child, errp); | |
8383 | } | |
8384 | ||
8385 | int bdrv_make_empty(BdrvChild *c, Error **errp) | |
8386 | { | |
8387 | BlockDriver *drv = c->bs->drv; | |
8388 | int ret; | |
8389 | ||
8390 | GLOBAL_STATE_CODE(); | |
8391 | assert(c->perm & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)); | |
8392 | ||
8393 | if (!drv->bdrv_make_empty) { | |
8394 | error_setg(errp, "%s does not support emptying nodes", | |
8395 | drv->format_name); | |
8396 | return -ENOTSUP; | |
8397 | } | |
8398 | ||
8399 | ret = drv->bdrv_make_empty(c->bs); | |
8400 | if (ret < 0) { | |
8401 | error_setg_errno(errp, -ret, "Failed to empty %s", | |
8402 | c->bs->filename); | |
8403 | return ret; | |
8404 | } | |
8405 | ||
8406 | return 0; | |
8407 | } | |
8408 | ||
8409 | /* | |
8410 | * Return the child that @bs acts as an overlay for, and from which data may be | |
8411 | * copied in COW or COR operations. Usually this is the backing file. | |
8412 | */ | |
8413 | BdrvChild *bdrv_cow_child(BlockDriverState *bs) | |
8414 | { | |
8415 | IO_CODE(); | |
8416 | ||
8417 | if (!bs || !bs->drv) { | |
8418 | return NULL; | |
8419 | } | |
8420 | ||
8421 | if (bs->drv->is_filter) { | |
8422 | return NULL; | |
8423 | } | |
8424 | ||
8425 | if (!bs->backing) { | |
8426 | return NULL; | |
8427 | } | |
8428 | ||
8429 | assert(bs->backing->role & BDRV_CHILD_COW); | |
8430 | return bs->backing; | |
8431 | } | |
8432 | ||
8433 | /* | |
8434 | * If @bs acts as a filter for exactly one of its children, return | |
8435 | * that child. | |
8436 | */ | |
8437 | BdrvChild *bdrv_filter_child(BlockDriverState *bs) | |
8438 | { | |
8439 | BdrvChild *c; | |
8440 | IO_CODE(); | |
8441 | ||
8442 | if (!bs || !bs->drv) { | |
8443 | return NULL; | |
8444 | } | |
8445 | ||
8446 | if (!bs->drv->is_filter) { | |
8447 | return NULL; | |
8448 | } | |
8449 | ||
8450 | /* Only one of @backing or @file may be used */ | |
8451 | assert(!(bs->backing && bs->file)); | |
8452 | ||
8453 | c = bs->backing ?: bs->file; | |
8454 | if (!c) { | |
8455 | return NULL; | |
8456 | } | |
8457 | ||
8458 | assert(c->role & BDRV_CHILD_FILTERED); | |
8459 | return c; | |
8460 | } | |
8461 | ||
8462 | /* | |
8463 | * Return either the result of bdrv_cow_child() or bdrv_filter_child(), | |
8464 | * whichever is non-NULL. | |
8465 | * | |
8466 | * Return NULL if both are NULL. | |
8467 | */ | |
8468 | BdrvChild *bdrv_filter_or_cow_child(BlockDriverState *bs) | |
8469 | { | |
8470 | BdrvChild *cow_child = bdrv_cow_child(bs); | |
8471 | BdrvChild *filter_child = bdrv_filter_child(bs); | |
8472 | IO_CODE(); | |
8473 | ||
8474 | /* Filter nodes cannot have COW backing files */ | |
8475 | assert(!(cow_child && filter_child)); | |
8476 | ||
8477 | return cow_child ?: filter_child; | |
8478 | } | |
8479 | ||
8480 | /* | |
8481 | * Return the primary child of this node: For filters, that is the | |
8482 | * filtered child. For other nodes, that is usually the child storing | |
8483 | * metadata. | |
8484 | * (A generally more helpful description is that this is (usually) the | |
8485 | * child that has the same filename as @bs.) | |
8486 | * | |
8487 | * Drivers do not necessarily have a primary child; for example quorum | |
8488 | * does not. | |
8489 | */ | |
8490 | BdrvChild *bdrv_primary_child(BlockDriverState *bs) | |
8491 | { | |
8492 | BdrvChild *c, *found = NULL; | |
8493 | IO_CODE(); | |
8494 | ||
8495 | QLIST_FOREACH(c, &bs->children, next) { | |
8496 | if (c->role & BDRV_CHILD_PRIMARY) { | |
8497 | assert(!found); | |
8498 | found = c; | |
8499 | } | |
8500 | } | |
8501 | ||
8502 | return found; | |
8503 | } | |
8504 | ||
8505 | static BlockDriverState * GRAPH_RDLOCK | |
8506 | bdrv_do_skip_filters(BlockDriverState *bs, bool stop_on_explicit_filter) | |
8507 | { | |
8508 | BdrvChild *c; | |
8509 | ||
8510 | if (!bs) { | |
8511 | return NULL; | |
8512 | } | |
8513 | ||
8514 | while (!(stop_on_explicit_filter && !bs->implicit)) { | |
8515 | c = bdrv_filter_child(bs); | |
8516 | if (!c) { | |
8517 | /* | |
8518 | * A filter that is embedded in a working block graph must | |
8519 | * have a child. Assert this here so this function does | |
8520 | * not return a filter node that is not expected by the | |
8521 | * caller. | |
8522 | */ | |
8523 | assert(!bs->drv || !bs->drv->is_filter); | |
8524 | break; | |
8525 | } | |
8526 | bs = c->bs; | |
8527 | } | |
8528 | /* | |
8529 | * Note that this treats nodes with bs->drv == NULL as not being | |
8530 | * filters (bs->drv == NULL should be replaced by something else | |
8531 | * anyway). | |
8532 | * The advantage of this behavior is that this function will thus | |
8533 | * always return a non-NULL value (given a non-NULL @bs). | |
8534 | */ | |
8535 | ||
8536 | return bs; | |
8537 | } | |
8538 | ||
8539 | /* | |
8540 | * Return the first BDS that has not been added implicitly or that | |
8541 | * does not have a filtered child down the chain starting from @bs | |
8542 | * (including @bs itself). | |
8543 | */ | |
8544 | BlockDriverState *bdrv_skip_implicit_filters(BlockDriverState *bs) | |
8545 | { | |
8546 | GLOBAL_STATE_CODE(); | |
8547 | return bdrv_do_skip_filters(bs, true); | |
8548 | } | |
8549 | ||
8550 | /* | |
8551 | * Return the first BDS that does not have a filtered child down the | |
8552 | * chain starting from @bs (including @bs itself). | |
8553 | */ | |
8554 | BlockDriverState *bdrv_skip_filters(BlockDriverState *bs) | |
8555 | { | |
8556 | IO_CODE(); | |
8557 | return bdrv_do_skip_filters(bs, false); | |
8558 | } | |
8559 | ||
8560 | /* | |
8561 | * For a backing chain, return the first non-filter backing image of | |
8562 | * the first non-filter image. | |
8563 | */ | |
8564 | BlockDriverState *bdrv_backing_chain_next(BlockDriverState *bs) | |
8565 | { | |
8566 | IO_CODE(); | |
8567 | return bdrv_skip_filters(bdrv_cow_bs(bdrv_skip_filters(bs))); | |
8568 | } | |
8569 | ||
8570 | /** | |
8571 | * Check whether [offset, offset + bytes) overlaps with the cached | |
8572 | * block-status data region. | |
8573 | * | |
8574 | * If so, and @pnum is not NULL, set *pnum to `bsc.data_end - offset`, | |
8575 | * which is what bdrv_bsc_is_data()'s interface needs. | |
8576 | * Otherwise, *pnum is not touched. | |
8577 | */ | |
8578 | static bool bdrv_bsc_range_overlaps_locked(BlockDriverState *bs, | |
8579 | int64_t offset, int64_t bytes, | |
8580 | int64_t *pnum) | |
8581 | { | |
8582 | BdrvBlockStatusCache *bsc = qatomic_rcu_read(&bs->block_status_cache); | |
8583 | bool overlaps; | |
8584 | ||
8585 | overlaps = | |
8586 | qatomic_read(&bsc->valid) && | |
8587 | ranges_overlap(offset, bytes, bsc->data_start, | |
8588 | bsc->data_end - bsc->data_start); | |
8589 | ||
8590 | if (overlaps && pnum) { | |
8591 | *pnum = bsc->data_end - offset; | |
8592 | } | |
8593 | ||
8594 | return overlaps; | |
8595 | } | |
8596 | ||
8597 | /** | |
8598 | * See block_int.h for this function's documentation. | |
8599 | */ | |
8600 | bool bdrv_bsc_is_data(BlockDriverState *bs, int64_t offset, int64_t *pnum) | |
8601 | { | |
8602 | IO_CODE(); | |
8603 | RCU_READ_LOCK_GUARD(); | |
8604 | return bdrv_bsc_range_overlaps_locked(bs, offset, 1, pnum); | |
8605 | } | |
8606 | ||
8607 | /** | |
8608 | * See block_int.h for this function's documentation. | |
8609 | */ | |
8610 | void bdrv_bsc_invalidate_range(BlockDriverState *bs, | |
8611 | int64_t offset, int64_t bytes) | |
8612 | { | |
8613 | IO_CODE(); | |
8614 | RCU_READ_LOCK_GUARD(); | |
8615 | ||
8616 | if (bdrv_bsc_range_overlaps_locked(bs, offset, bytes, NULL)) { | |
8617 | qatomic_set(&bs->block_status_cache->valid, false); | |
8618 | } | |
8619 | } | |
8620 | ||
8621 | /** | |
8622 | * See block_int.h for this function's documentation. | |
8623 | */ | |
8624 | void bdrv_bsc_fill(BlockDriverState *bs, int64_t offset, int64_t bytes) | |
8625 | { | |
8626 | BdrvBlockStatusCache *new_bsc = g_new(BdrvBlockStatusCache, 1); | |
8627 | BdrvBlockStatusCache *old_bsc; | |
8628 | IO_CODE(); | |
8629 | ||
8630 | *new_bsc = (BdrvBlockStatusCache) { | |
8631 | .valid = true, | |
8632 | .data_start = offset, | |
8633 | .data_end = offset + bytes, | |
8634 | }; | |
8635 | ||
8636 | QEMU_LOCK_GUARD(&bs->bsc_modify_lock); | |
8637 | ||
8638 | old_bsc = qatomic_rcu_read(&bs->block_status_cache); | |
8639 | qatomic_rcu_set(&bs->block_status_cache, new_bsc); | |
8640 | if (old_bsc) { | |
8641 | g_free_rcu(old_bsc, rcu); | |
8642 | } | |
8643 | } |