]>
Commit | Line | Data |
---|---|---|
1 | /* | |
2 | * QEMU System Emulator block driver | |
3 | * | |
4 | * Copyright (c) 2003 Fabrice Bellard | |
5 | * | |
6 | * Permission is hereby granted, free of charge, to any person obtaining a copy | |
7 | * of this software and associated documentation files (the "Software"), to deal | |
8 | * in the Software without restriction, including without limitation the rights | |
9 | * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell | |
10 | * copies of the Software, and to permit persons to whom the Software is | |
11 | * furnished to do so, subject to the following conditions: | |
12 | * | |
13 | * The above copyright notice and this permission notice shall be included in | |
14 | * all copies or substantial portions of the Software. | |
15 | * | |
16 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR | |
17 | * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, | |
18 | * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL | |
19 | * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER | |
20 | * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, | |
21 | * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN | |
22 | * THE SOFTWARE. | |
23 | */ | |
24 | #include "qemu/osdep.h" | |
25 | #include "block/trace.h" | |
26 | #include "block/block_int.h" | |
27 | #include "block/blockjob.h" | |
28 | #include "block/nbd.h" | |
29 | #include "qemu/error-report.h" | |
30 | #include "module_block.h" | |
31 | #include "qemu/module.h" | |
32 | #include "qapi/qmp/qerror.h" | |
33 | #include "qapi/qmp/qbool.h" | |
34 | #include "qapi/qmp/qjson.h" | |
35 | #include "sysemu/block-backend.h" | |
36 | #include "sysemu/sysemu.h" | |
37 | #include "qemu/notify.h" | |
38 | #include "qemu/coroutine.h" | |
39 | #include "block/qapi.h" | |
40 | #include "qmp-commands.h" | |
41 | #include "qemu/timer.h" | |
42 | #include "qapi-event.h" | |
43 | #include "qemu/cutils.h" | |
44 | #include "qemu/id.h" | |
45 | #include "qapi/util.h" | |
46 | ||
47 | #ifdef CONFIG_BSD | |
48 | #include <sys/ioctl.h> | |
49 | #include <sys/queue.h> | |
50 | #ifndef __DragonFly__ | |
51 | #include <sys/disk.h> | |
52 | #endif | |
53 | #endif | |
54 | ||
55 | #ifdef _WIN32 | |
56 | #include <windows.h> | |
57 | #endif | |
58 | ||
59 | #define NOT_DONE 0x7fffffff /* used while emulated sync operation in progress */ | |
60 | ||
61 | static QTAILQ_HEAD(, BlockDriverState) graph_bdrv_states = | |
62 | QTAILQ_HEAD_INITIALIZER(graph_bdrv_states); | |
63 | ||
64 | static QTAILQ_HEAD(, BlockDriverState) all_bdrv_states = | |
65 | QTAILQ_HEAD_INITIALIZER(all_bdrv_states); | |
66 | ||
67 | static QLIST_HEAD(, BlockDriver) bdrv_drivers = | |
68 | QLIST_HEAD_INITIALIZER(bdrv_drivers); | |
69 | ||
70 | static BlockDriverState *bdrv_open_inherit(const char *filename, | |
71 | const char *reference, | |
72 | QDict *options, int flags, | |
73 | BlockDriverState *parent, | |
74 | const BdrvChildRole *child_role, | |
75 | Error **errp); | |
76 | ||
77 | /* If non-zero, use only whitelisted block drivers */ | |
78 | static int use_bdrv_whitelist; | |
79 | ||
80 | #ifdef _WIN32 | |
81 | static int is_windows_drive_prefix(const char *filename) | |
82 | { | |
83 | return (((filename[0] >= 'a' && filename[0] <= 'z') || | |
84 | (filename[0] >= 'A' && filename[0] <= 'Z')) && | |
85 | filename[1] == ':'); | |
86 | } | |
87 | ||
88 | int is_windows_drive(const char *filename) | |
89 | { | |
90 | if (is_windows_drive_prefix(filename) && | |
91 | filename[2] == '\0') | |
92 | return 1; | |
93 | if (strstart(filename, "\\\\.\\", NULL) || | |
94 | strstart(filename, "//./", NULL)) | |
95 | return 1; | |
96 | return 0; | |
97 | } | |
98 | #endif | |
99 | ||
100 | size_t bdrv_opt_mem_align(BlockDriverState *bs) | |
101 | { | |
102 | if (!bs || !bs->drv) { | |
103 | /* page size or 4k (hdd sector size) should be on the safe side */ | |
104 | return MAX(4096, getpagesize()); | |
105 | } | |
106 | ||
107 | return bs->bl.opt_mem_alignment; | |
108 | } | |
109 | ||
110 | size_t bdrv_min_mem_align(BlockDriverState *bs) | |
111 | { | |
112 | if (!bs || !bs->drv) { | |
113 | /* page size or 4k (hdd sector size) should be on the safe side */ | |
114 | return MAX(4096, getpagesize()); | |
115 | } | |
116 | ||
117 | return bs->bl.min_mem_alignment; | |
118 | } | |
119 | ||
120 | /* check if the path starts with "<protocol>:" */ | |
121 | int path_has_protocol(const char *path) | |
122 | { | |
123 | const char *p; | |
124 | ||
125 | #ifdef _WIN32 | |
126 | if (is_windows_drive(path) || | |
127 | is_windows_drive_prefix(path)) { | |
128 | return 0; | |
129 | } | |
130 | p = path + strcspn(path, ":/\\"); | |
131 | #else | |
132 | p = path + strcspn(path, ":/"); | |
133 | #endif | |
134 | ||
135 | return *p == ':'; | |
136 | } | |
137 | ||
138 | int path_is_absolute(const char *path) | |
139 | { | |
140 | #ifdef _WIN32 | |
141 | /* specific case for names like: "\\.\d:" */ | |
142 | if (is_windows_drive(path) || is_windows_drive_prefix(path)) { | |
143 | return 1; | |
144 | } | |
145 | return (*path == '/' || *path == '\\'); | |
146 | #else | |
147 | return (*path == '/'); | |
148 | #endif | |
149 | } | |
150 | ||
151 | /* if filename is absolute, just copy it to dest. Otherwise, build a | |
152 | path to it by considering it is relative to base_path. URL are | |
153 | supported. */ | |
154 | void path_combine(char *dest, int dest_size, | |
155 | const char *base_path, | |
156 | const char *filename) | |
157 | { | |
158 | const char *p, *p1; | |
159 | int len; | |
160 | ||
161 | if (dest_size <= 0) | |
162 | return; | |
163 | if (path_is_absolute(filename)) { | |
164 | pstrcpy(dest, dest_size, filename); | |
165 | } else { | |
166 | p = strchr(base_path, ':'); | |
167 | if (p) | |
168 | p++; | |
169 | else | |
170 | p = base_path; | |
171 | p1 = strrchr(base_path, '/'); | |
172 | #ifdef _WIN32 | |
173 | { | |
174 | const char *p2; | |
175 | p2 = strrchr(base_path, '\\'); | |
176 | if (!p1 || p2 > p1) | |
177 | p1 = p2; | |
178 | } | |
179 | #endif | |
180 | if (p1) | |
181 | p1++; | |
182 | else | |
183 | p1 = base_path; | |
184 | if (p1 > p) | |
185 | p = p1; | |
186 | len = p - base_path; | |
187 | if (len > dest_size - 1) | |
188 | len = dest_size - 1; | |
189 | memcpy(dest, base_path, len); | |
190 | dest[len] = '\0'; | |
191 | pstrcat(dest, dest_size, filename); | |
192 | } | |
193 | } | |
194 | ||
195 | void bdrv_get_full_backing_filename_from_filename(const char *backed, | |
196 | const char *backing, | |
197 | char *dest, size_t sz, | |
198 | Error **errp) | |
199 | { | |
200 | if (backing[0] == '\0' || path_has_protocol(backing) || | |
201 | path_is_absolute(backing)) | |
202 | { | |
203 | pstrcpy(dest, sz, backing); | |
204 | } else if (backed[0] == '\0' || strstart(backed, "json:", NULL)) { | |
205 | error_setg(errp, "Cannot use relative backing file names for '%s'", | |
206 | backed); | |
207 | } else { | |
208 | path_combine(dest, sz, backed, backing); | |
209 | } | |
210 | } | |
211 | ||
212 | void bdrv_get_full_backing_filename(BlockDriverState *bs, char *dest, size_t sz, | |
213 | Error **errp) | |
214 | { | |
215 | char *backed = bs->exact_filename[0] ? bs->exact_filename : bs->filename; | |
216 | ||
217 | bdrv_get_full_backing_filename_from_filename(backed, bs->backing_file, | |
218 | dest, sz, errp); | |
219 | } | |
220 | ||
221 | void bdrv_register(BlockDriver *bdrv) | |
222 | { | |
223 | QLIST_INSERT_HEAD(&bdrv_drivers, bdrv, list); | |
224 | } | |
225 | ||
226 | BlockDriverState *bdrv_new(void) | |
227 | { | |
228 | BlockDriverState *bs; | |
229 | int i; | |
230 | ||
231 | bs = g_new0(BlockDriverState, 1); | |
232 | QLIST_INIT(&bs->dirty_bitmaps); | |
233 | for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) { | |
234 | QLIST_INIT(&bs->op_blockers[i]); | |
235 | } | |
236 | notifier_with_return_list_init(&bs->before_write_notifiers); | |
237 | bs->refcnt = 1; | |
238 | bs->aio_context = qemu_get_aio_context(); | |
239 | ||
240 | qemu_co_queue_init(&bs->flush_queue); | |
241 | ||
242 | QTAILQ_INSERT_TAIL(&all_bdrv_states, bs, bs_list); | |
243 | ||
244 | return bs; | |
245 | } | |
246 | ||
247 | static BlockDriver *bdrv_do_find_format(const char *format_name) | |
248 | { | |
249 | BlockDriver *drv1; | |
250 | ||
251 | QLIST_FOREACH(drv1, &bdrv_drivers, list) { | |
252 | if (!strcmp(drv1->format_name, format_name)) { | |
253 | return drv1; | |
254 | } | |
255 | } | |
256 | ||
257 | return NULL; | |
258 | } | |
259 | ||
260 | BlockDriver *bdrv_find_format(const char *format_name) | |
261 | { | |
262 | BlockDriver *drv1; | |
263 | int i; | |
264 | ||
265 | drv1 = bdrv_do_find_format(format_name); | |
266 | if (drv1) { | |
267 | return drv1; | |
268 | } | |
269 | ||
270 | /* The driver isn't registered, maybe we need to load a module */ | |
271 | for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); ++i) { | |
272 | if (!strcmp(block_driver_modules[i].format_name, format_name)) { | |
273 | block_module_load_one(block_driver_modules[i].library_name); | |
274 | break; | |
275 | } | |
276 | } | |
277 | ||
278 | return bdrv_do_find_format(format_name); | |
279 | } | |
280 | ||
281 | static int bdrv_is_whitelisted(BlockDriver *drv, bool read_only) | |
282 | { | |
283 | static const char *whitelist_rw[] = { | |
284 | CONFIG_BDRV_RW_WHITELIST | |
285 | }; | |
286 | static const char *whitelist_ro[] = { | |
287 | CONFIG_BDRV_RO_WHITELIST | |
288 | }; | |
289 | const char **p; | |
290 | ||
291 | if (!whitelist_rw[0] && !whitelist_ro[0]) { | |
292 | return 1; /* no whitelist, anything goes */ | |
293 | } | |
294 | ||
295 | for (p = whitelist_rw; *p; p++) { | |
296 | if (!strcmp(drv->format_name, *p)) { | |
297 | return 1; | |
298 | } | |
299 | } | |
300 | if (read_only) { | |
301 | for (p = whitelist_ro; *p; p++) { | |
302 | if (!strcmp(drv->format_name, *p)) { | |
303 | return 1; | |
304 | } | |
305 | } | |
306 | } | |
307 | return 0; | |
308 | } | |
309 | ||
310 | bool bdrv_uses_whitelist(void) | |
311 | { | |
312 | return use_bdrv_whitelist; | |
313 | } | |
314 | ||
315 | typedef struct CreateCo { | |
316 | BlockDriver *drv; | |
317 | char *filename; | |
318 | QemuOpts *opts; | |
319 | int ret; | |
320 | Error *err; | |
321 | } CreateCo; | |
322 | ||
323 | static void coroutine_fn bdrv_create_co_entry(void *opaque) | |
324 | { | |
325 | Error *local_err = NULL; | |
326 | int ret; | |
327 | ||
328 | CreateCo *cco = opaque; | |
329 | assert(cco->drv); | |
330 | ||
331 | ret = cco->drv->bdrv_create(cco->filename, cco->opts, &local_err); | |
332 | error_propagate(&cco->err, local_err); | |
333 | cco->ret = ret; | |
334 | } | |
335 | ||
336 | int bdrv_create(BlockDriver *drv, const char* filename, | |
337 | QemuOpts *opts, Error **errp) | |
338 | { | |
339 | int ret; | |
340 | ||
341 | Coroutine *co; | |
342 | CreateCo cco = { | |
343 | .drv = drv, | |
344 | .filename = g_strdup(filename), | |
345 | .opts = opts, | |
346 | .ret = NOT_DONE, | |
347 | .err = NULL, | |
348 | }; | |
349 | ||
350 | if (!drv->bdrv_create) { | |
351 | error_setg(errp, "Driver '%s' does not support image creation", drv->format_name); | |
352 | ret = -ENOTSUP; | |
353 | goto out; | |
354 | } | |
355 | ||
356 | if (qemu_in_coroutine()) { | |
357 | /* Fast-path if already in coroutine context */ | |
358 | bdrv_create_co_entry(&cco); | |
359 | } else { | |
360 | co = qemu_coroutine_create(bdrv_create_co_entry, &cco); | |
361 | qemu_coroutine_enter(co); | |
362 | while (cco.ret == NOT_DONE) { | |
363 | aio_poll(qemu_get_aio_context(), true); | |
364 | } | |
365 | } | |
366 | ||
367 | ret = cco.ret; | |
368 | if (ret < 0) { | |
369 | if (cco.err) { | |
370 | error_propagate(errp, cco.err); | |
371 | } else { | |
372 | error_setg_errno(errp, -ret, "Could not create image"); | |
373 | } | |
374 | } | |
375 | ||
376 | out: | |
377 | g_free(cco.filename); | |
378 | return ret; | |
379 | } | |
380 | ||
381 | int bdrv_create_file(const char *filename, QemuOpts *opts, Error **errp) | |
382 | { | |
383 | BlockDriver *drv; | |
384 | Error *local_err = NULL; | |
385 | int ret; | |
386 | ||
387 | drv = bdrv_find_protocol(filename, true, errp); | |
388 | if (drv == NULL) { | |
389 | return -ENOENT; | |
390 | } | |
391 | ||
392 | ret = bdrv_create(drv, filename, opts, &local_err); | |
393 | error_propagate(errp, local_err); | |
394 | return ret; | |
395 | } | |
396 | ||
397 | /** | |
398 | * Try to get @bs's logical and physical block size. | |
399 | * On success, store them in @bsz struct and return 0. | |
400 | * On failure return -errno. | |
401 | * @bs must not be empty. | |
402 | */ | |
403 | int bdrv_probe_blocksizes(BlockDriverState *bs, BlockSizes *bsz) | |
404 | { | |
405 | BlockDriver *drv = bs->drv; | |
406 | ||
407 | if (drv && drv->bdrv_probe_blocksizes) { | |
408 | return drv->bdrv_probe_blocksizes(bs, bsz); | |
409 | } | |
410 | ||
411 | return -ENOTSUP; | |
412 | } | |
413 | ||
414 | /** | |
415 | * Try to get @bs's geometry (cyls, heads, sectors). | |
416 | * On success, store them in @geo struct and return 0. | |
417 | * On failure return -errno. | |
418 | * @bs must not be empty. | |
419 | */ | |
420 | int bdrv_probe_geometry(BlockDriverState *bs, HDGeometry *geo) | |
421 | { | |
422 | BlockDriver *drv = bs->drv; | |
423 | ||
424 | if (drv && drv->bdrv_probe_geometry) { | |
425 | return drv->bdrv_probe_geometry(bs, geo); | |
426 | } | |
427 | ||
428 | return -ENOTSUP; | |
429 | } | |
430 | ||
431 | /* | |
432 | * Create a uniquely-named empty temporary file. | |
433 | * Return 0 upon success, otherwise a negative errno value. | |
434 | */ | |
435 | int get_tmp_filename(char *filename, int size) | |
436 | { | |
437 | #ifdef _WIN32 | |
438 | char temp_dir[MAX_PATH]; | |
439 | /* GetTempFileName requires that its output buffer (4th param) | |
440 | have length MAX_PATH or greater. */ | |
441 | assert(size >= MAX_PATH); | |
442 | return (GetTempPath(MAX_PATH, temp_dir) | |
443 | && GetTempFileName(temp_dir, "qem", 0, filename) | |
444 | ? 0 : -GetLastError()); | |
445 | #else | |
446 | int fd; | |
447 | const char *tmpdir; | |
448 | tmpdir = getenv("TMPDIR"); | |
449 | if (!tmpdir) { | |
450 | tmpdir = "/var/tmp"; | |
451 | } | |
452 | if (snprintf(filename, size, "%s/vl.XXXXXX", tmpdir) >= size) { | |
453 | return -EOVERFLOW; | |
454 | } | |
455 | fd = mkstemp(filename); | |
456 | if (fd < 0) { | |
457 | return -errno; | |
458 | } | |
459 | if (close(fd) != 0) { | |
460 | unlink(filename); | |
461 | return -errno; | |
462 | } | |
463 | return 0; | |
464 | #endif | |
465 | } | |
466 | ||
467 | /* | |
468 | * Detect host devices. By convention, /dev/cdrom[N] is always | |
469 | * recognized as a host CDROM. | |
470 | */ | |
471 | static BlockDriver *find_hdev_driver(const char *filename) | |
472 | { | |
473 | int score_max = 0, score; | |
474 | BlockDriver *drv = NULL, *d; | |
475 | ||
476 | QLIST_FOREACH(d, &bdrv_drivers, list) { | |
477 | if (d->bdrv_probe_device) { | |
478 | score = d->bdrv_probe_device(filename); | |
479 | if (score > score_max) { | |
480 | score_max = score; | |
481 | drv = d; | |
482 | } | |
483 | } | |
484 | } | |
485 | ||
486 | return drv; | |
487 | } | |
488 | ||
489 | static BlockDriver *bdrv_do_find_protocol(const char *protocol) | |
490 | { | |
491 | BlockDriver *drv1; | |
492 | ||
493 | QLIST_FOREACH(drv1, &bdrv_drivers, list) { | |
494 | if (drv1->protocol_name && !strcmp(drv1->protocol_name, protocol)) { | |
495 | return drv1; | |
496 | } | |
497 | } | |
498 | ||
499 | return NULL; | |
500 | } | |
501 | ||
502 | BlockDriver *bdrv_find_protocol(const char *filename, | |
503 | bool allow_protocol_prefix, | |
504 | Error **errp) | |
505 | { | |
506 | BlockDriver *drv1; | |
507 | char protocol[128]; | |
508 | int len; | |
509 | const char *p; | |
510 | int i; | |
511 | ||
512 | /* TODO Drivers without bdrv_file_open must be specified explicitly */ | |
513 | ||
514 | /* | |
515 | * XXX(hch): we really should not let host device detection | |
516 | * override an explicit protocol specification, but moving this | |
517 | * later breaks access to device names with colons in them. | |
518 | * Thanks to the brain-dead persistent naming schemes on udev- | |
519 | * based Linux systems those actually are quite common. | |
520 | */ | |
521 | drv1 = find_hdev_driver(filename); | |
522 | if (drv1) { | |
523 | return drv1; | |
524 | } | |
525 | ||
526 | if (!path_has_protocol(filename) || !allow_protocol_prefix) { | |
527 | return &bdrv_file; | |
528 | } | |
529 | ||
530 | p = strchr(filename, ':'); | |
531 | assert(p != NULL); | |
532 | len = p - filename; | |
533 | if (len > sizeof(protocol) - 1) | |
534 | len = sizeof(protocol) - 1; | |
535 | memcpy(protocol, filename, len); | |
536 | protocol[len] = '\0'; | |
537 | ||
538 | drv1 = bdrv_do_find_protocol(protocol); | |
539 | if (drv1) { | |
540 | return drv1; | |
541 | } | |
542 | ||
543 | for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); ++i) { | |
544 | if (block_driver_modules[i].protocol_name && | |
545 | !strcmp(block_driver_modules[i].protocol_name, protocol)) { | |
546 | block_module_load_one(block_driver_modules[i].library_name); | |
547 | break; | |
548 | } | |
549 | } | |
550 | ||
551 | drv1 = bdrv_do_find_protocol(protocol); | |
552 | if (!drv1) { | |
553 | error_setg(errp, "Unknown protocol '%s'", protocol); | |
554 | } | |
555 | return drv1; | |
556 | } | |
557 | ||
558 | /* | |
559 | * Guess image format by probing its contents. | |
560 | * This is not a good idea when your image is raw (CVE-2008-2004), but | |
561 | * we do it anyway for backward compatibility. | |
562 | * | |
563 | * @buf contains the image's first @buf_size bytes. | |
564 | * @buf_size is the buffer size in bytes (generally BLOCK_PROBE_BUF_SIZE, | |
565 | * but can be smaller if the image file is smaller) | |
566 | * @filename is its filename. | |
567 | * | |
568 | * For all block drivers, call the bdrv_probe() method to get its | |
569 | * probing score. | |
570 | * Return the first block driver with the highest probing score. | |
571 | */ | |
572 | BlockDriver *bdrv_probe_all(const uint8_t *buf, int buf_size, | |
573 | const char *filename) | |
574 | { | |
575 | int score_max = 0, score; | |
576 | BlockDriver *drv = NULL, *d; | |
577 | ||
578 | QLIST_FOREACH(d, &bdrv_drivers, list) { | |
579 | if (d->bdrv_probe) { | |
580 | score = d->bdrv_probe(buf, buf_size, filename); | |
581 | if (score > score_max) { | |
582 | score_max = score; | |
583 | drv = d; | |
584 | } | |
585 | } | |
586 | } | |
587 | ||
588 | return drv; | |
589 | } | |
590 | ||
591 | static int find_image_format(BlockBackend *file, const char *filename, | |
592 | BlockDriver **pdrv, Error **errp) | |
593 | { | |
594 | BlockDriver *drv; | |
595 | uint8_t buf[BLOCK_PROBE_BUF_SIZE]; | |
596 | int ret = 0; | |
597 | ||
598 | /* Return the raw BlockDriver * to scsi-generic devices or empty drives */ | |
599 | if (blk_is_sg(file) || !blk_is_inserted(file) || blk_getlength(file) == 0) { | |
600 | *pdrv = &bdrv_raw; | |
601 | return ret; | |
602 | } | |
603 | ||
604 | ret = blk_pread(file, 0, buf, sizeof(buf)); | |
605 | if (ret < 0) { | |
606 | error_setg_errno(errp, -ret, "Could not read image for determining its " | |
607 | "format"); | |
608 | *pdrv = NULL; | |
609 | return ret; | |
610 | } | |
611 | ||
612 | drv = bdrv_probe_all(buf, ret, filename); | |
613 | if (!drv) { | |
614 | error_setg(errp, "Could not determine image format: No compatible " | |
615 | "driver found"); | |
616 | ret = -ENOENT; | |
617 | } | |
618 | *pdrv = drv; | |
619 | return ret; | |
620 | } | |
621 | ||
622 | /** | |
623 | * Set the current 'total_sectors' value | |
624 | * Return 0 on success, -errno on error. | |
625 | */ | |
626 | static int refresh_total_sectors(BlockDriverState *bs, int64_t hint) | |
627 | { | |
628 | BlockDriver *drv = bs->drv; | |
629 | ||
630 | /* Do not attempt drv->bdrv_getlength() on scsi-generic devices */ | |
631 | if (bdrv_is_sg(bs)) | |
632 | return 0; | |
633 | ||
634 | /* query actual device if possible, otherwise just trust the hint */ | |
635 | if (drv->bdrv_getlength) { | |
636 | int64_t length = drv->bdrv_getlength(bs); | |
637 | if (length < 0) { | |
638 | return length; | |
639 | } | |
640 | hint = DIV_ROUND_UP(length, BDRV_SECTOR_SIZE); | |
641 | } | |
642 | ||
643 | bs->total_sectors = hint; | |
644 | return 0; | |
645 | } | |
646 | ||
647 | /** | |
648 | * Combines a QDict of new block driver @options with any missing options taken | |
649 | * from @old_options, so that leaving out an option defaults to its old value. | |
650 | */ | |
651 | static void bdrv_join_options(BlockDriverState *bs, QDict *options, | |
652 | QDict *old_options) | |
653 | { | |
654 | if (bs->drv && bs->drv->bdrv_join_options) { | |
655 | bs->drv->bdrv_join_options(options, old_options); | |
656 | } else { | |
657 | qdict_join(options, old_options, false); | |
658 | } | |
659 | } | |
660 | ||
661 | /** | |
662 | * Set open flags for a given discard mode | |
663 | * | |
664 | * Return 0 on success, -1 if the discard mode was invalid. | |
665 | */ | |
666 | int bdrv_parse_discard_flags(const char *mode, int *flags) | |
667 | { | |
668 | *flags &= ~BDRV_O_UNMAP; | |
669 | ||
670 | if (!strcmp(mode, "off") || !strcmp(mode, "ignore")) { | |
671 | /* do nothing */ | |
672 | } else if (!strcmp(mode, "on") || !strcmp(mode, "unmap")) { | |
673 | *flags |= BDRV_O_UNMAP; | |
674 | } else { | |
675 | return -1; | |
676 | } | |
677 | ||
678 | return 0; | |
679 | } | |
680 | ||
681 | /** | |
682 | * Set open flags for a given cache mode | |
683 | * | |
684 | * Return 0 on success, -1 if the cache mode was invalid. | |
685 | */ | |
686 | int bdrv_parse_cache_mode(const char *mode, int *flags, bool *writethrough) | |
687 | { | |
688 | *flags &= ~BDRV_O_CACHE_MASK; | |
689 | ||
690 | if (!strcmp(mode, "off") || !strcmp(mode, "none")) { | |
691 | *writethrough = false; | |
692 | *flags |= BDRV_O_NOCACHE; | |
693 | } else if (!strcmp(mode, "directsync")) { | |
694 | *writethrough = true; | |
695 | *flags |= BDRV_O_NOCACHE; | |
696 | } else if (!strcmp(mode, "writeback")) { | |
697 | *writethrough = false; | |
698 | } else if (!strcmp(mode, "unsafe")) { | |
699 | *writethrough = false; | |
700 | *flags |= BDRV_O_NO_FLUSH; | |
701 | } else if (!strcmp(mode, "writethrough")) { | |
702 | *writethrough = true; | |
703 | } else { | |
704 | return -1; | |
705 | } | |
706 | ||
707 | return 0; | |
708 | } | |
709 | ||
710 | static void bdrv_child_cb_drained_begin(BdrvChild *child) | |
711 | { | |
712 | BlockDriverState *bs = child->opaque; | |
713 | bdrv_drained_begin(bs); | |
714 | } | |
715 | ||
716 | static void bdrv_child_cb_drained_end(BdrvChild *child) | |
717 | { | |
718 | BlockDriverState *bs = child->opaque; | |
719 | bdrv_drained_end(bs); | |
720 | } | |
721 | ||
722 | /* | |
723 | * Returns the options and flags that a temporary snapshot should get, based on | |
724 | * the originally requested flags (the originally requested image will have | |
725 | * flags like a backing file) | |
726 | */ | |
727 | static void bdrv_temp_snapshot_options(int *child_flags, QDict *child_options, | |
728 | int parent_flags, QDict *parent_options) | |
729 | { | |
730 | *child_flags = (parent_flags & ~BDRV_O_SNAPSHOT) | BDRV_O_TEMPORARY; | |
731 | ||
732 | /* For temporary files, unconditional cache=unsafe is fine */ | |
733 | qdict_set_default_str(child_options, BDRV_OPT_CACHE_DIRECT, "off"); | |
734 | qdict_set_default_str(child_options, BDRV_OPT_CACHE_NO_FLUSH, "on"); | |
735 | ||
736 | /* Copy the read-only option from the parent */ | |
737 | qdict_copy_default(child_options, parent_options, BDRV_OPT_READ_ONLY); | |
738 | ||
739 | /* aio=native doesn't work for cache.direct=off, so disable it for the | |
740 | * temporary snapshot */ | |
741 | *child_flags &= ~BDRV_O_NATIVE_AIO; | |
742 | } | |
743 | ||
744 | /* | |
745 | * Returns the options and flags that bs->file should get if a protocol driver | |
746 | * is expected, based on the given options and flags for the parent BDS | |
747 | */ | |
748 | static void bdrv_inherited_options(int *child_flags, QDict *child_options, | |
749 | int parent_flags, QDict *parent_options) | |
750 | { | |
751 | int flags = parent_flags; | |
752 | ||
753 | /* Enable protocol handling, disable format probing for bs->file */ | |
754 | flags |= BDRV_O_PROTOCOL; | |
755 | ||
756 | /* If the cache mode isn't explicitly set, inherit direct and no-flush from | |
757 | * the parent. */ | |
758 | qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_DIRECT); | |
759 | qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_NO_FLUSH); | |
760 | ||
761 | /* Inherit the read-only option from the parent if it's not set */ | |
762 | qdict_copy_default(child_options, parent_options, BDRV_OPT_READ_ONLY); | |
763 | ||
764 | /* Our block drivers take care to send flushes and respect unmap policy, | |
765 | * so we can default to enable both on lower layers regardless of the | |
766 | * corresponding parent options. */ | |
767 | qdict_set_default_str(child_options, BDRV_OPT_DISCARD, "unmap"); | |
768 | ||
769 | /* Clear flags that only apply to the top layer */ | |
770 | flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_COPY_ON_READ | | |
771 | BDRV_O_NO_IO); | |
772 | ||
773 | *child_flags = flags; | |
774 | } | |
775 | ||
776 | const BdrvChildRole child_file = { | |
777 | .inherit_options = bdrv_inherited_options, | |
778 | .drained_begin = bdrv_child_cb_drained_begin, | |
779 | .drained_end = bdrv_child_cb_drained_end, | |
780 | }; | |
781 | ||
782 | /* | |
783 | * Returns the options and flags that bs->file should get if the use of formats | |
784 | * (and not only protocols) is permitted for it, based on the given options and | |
785 | * flags for the parent BDS | |
786 | */ | |
787 | static void bdrv_inherited_fmt_options(int *child_flags, QDict *child_options, | |
788 | int parent_flags, QDict *parent_options) | |
789 | { | |
790 | child_file.inherit_options(child_flags, child_options, | |
791 | parent_flags, parent_options); | |
792 | ||
793 | *child_flags &= ~(BDRV_O_PROTOCOL | BDRV_O_NO_IO); | |
794 | } | |
795 | ||
796 | const BdrvChildRole child_format = { | |
797 | .inherit_options = bdrv_inherited_fmt_options, | |
798 | .drained_begin = bdrv_child_cb_drained_begin, | |
799 | .drained_end = bdrv_child_cb_drained_end, | |
800 | }; | |
801 | ||
802 | /* | |
803 | * Returns the options and flags that bs->backing should get, based on the | |
804 | * given options and flags for the parent BDS | |
805 | */ | |
806 | static void bdrv_backing_options(int *child_flags, QDict *child_options, | |
807 | int parent_flags, QDict *parent_options) | |
808 | { | |
809 | int flags = parent_flags; | |
810 | ||
811 | /* The cache mode is inherited unmodified for backing files; except WCE, | |
812 | * which is only applied on the top level (BlockBackend) */ | |
813 | qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_DIRECT); | |
814 | qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_NO_FLUSH); | |
815 | ||
816 | /* backing files always opened read-only */ | |
817 | qdict_set_default_str(child_options, BDRV_OPT_READ_ONLY, "on"); | |
818 | flags &= ~BDRV_O_COPY_ON_READ; | |
819 | ||
820 | /* snapshot=on is handled on the top layer */ | |
821 | flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_TEMPORARY); | |
822 | ||
823 | *child_flags = flags; | |
824 | } | |
825 | ||
826 | const BdrvChildRole child_backing = { | |
827 | .inherit_options = bdrv_backing_options, | |
828 | .drained_begin = bdrv_child_cb_drained_begin, | |
829 | .drained_end = bdrv_child_cb_drained_end, | |
830 | }; | |
831 | ||
832 | static int bdrv_open_flags(BlockDriverState *bs, int flags) | |
833 | { | |
834 | int open_flags = flags; | |
835 | ||
836 | /* | |
837 | * Clear flags that are internal to the block layer before opening the | |
838 | * image. | |
839 | */ | |
840 | open_flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_PROTOCOL); | |
841 | ||
842 | /* | |
843 | * Snapshots should be writable. | |
844 | */ | |
845 | if (flags & BDRV_O_TEMPORARY) { | |
846 | open_flags |= BDRV_O_RDWR; | |
847 | } | |
848 | ||
849 | return open_flags; | |
850 | } | |
851 | ||
852 | static void update_flags_from_options(int *flags, QemuOpts *opts) | |
853 | { | |
854 | *flags &= ~BDRV_O_CACHE_MASK; | |
855 | ||
856 | assert(qemu_opt_find(opts, BDRV_OPT_CACHE_NO_FLUSH)); | |
857 | if (qemu_opt_get_bool(opts, BDRV_OPT_CACHE_NO_FLUSH, false)) { | |
858 | *flags |= BDRV_O_NO_FLUSH; | |
859 | } | |
860 | ||
861 | assert(qemu_opt_find(opts, BDRV_OPT_CACHE_DIRECT)); | |
862 | if (qemu_opt_get_bool(opts, BDRV_OPT_CACHE_DIRECT, false)) { | |
863 | *flags |= BDRV_O_NOCACHE; | |
864 | } | |
865 | ||
866 | *flags &= ~BDRV_O_RDWR; | |
867 | ||
868 | assert(qemu_opt_find(opts, BDRV_OPT_READ_ONLY)); | |
869 | if (!qemu_opt_get_bool(opts, BDRV_OPT_READ_ONLY, false)) { | |
870 | *flags |= BDRV_O_RDWR; | |
871 | } | |
872 | ||
873 | } | |
874 | ||
875 | static void update_options_from_flags(QDict *options, int flags) | |
876 | { | |
877 | if (!qdict_haskey(options, BDRV_OPT_CACHE_DIRECT)) { | |
878 | qdict_put(options, BDRV_OPT_CACHE_DIRECT, | |
879 | qbool_from_bool(flags & BDRV_O_NOCACHE)); | |
880 | } | |
881 | if (!qdict_haskey(options, BDRV_OPT_CACHE_NO_FLUSH)) { | |
882 | qdict_put(options, BDRV_OPT_CACHE_NO_FLUSH, | |
883 | qbool_from_bool(flags & BDRV_O_NO_FLUSH)); | |
884 | } | |
885 | if (!qdict_haskey(options, BDRV_OPT_READ_ONLY)) { | |
886 | qdict_put(options, BDRV_OPT_READ_ONLY, | |
887 | qbool_from_bool(!(flags & BDRV_O_RDWR))); | |
888 | } | |
889 | } | |
890 | ||
891 | static void bdrv_assign_node_name(BlockDriverState *bs, | |
892 | const char *node_name, | |
893 | Error **errp) | |
894 | { | |
895 | char *gen_node_name = NULL; | |
896 | ||
897 | if (!node_name) { | |
898 | node_name = gen_node_name = id_generate(ID_BLOCK); | |
899 | } else if (!id_wellformed(node_name)) { | |
900 | /* | |
901 | * Check for empty string or invalid characters, but not if it is | |
902 | * generated (generated names use characters not available to the user) | |
903 | */ | |
904 | error_setg(errp, "Invalid node name"); | |
905 | return; | |
906 | } | |
907 | ||
908 | /* takes care of avoiding namespaces collisions */ | |
909 | if (blk_by_name(node_name)) { | |
910 | error_setg(errp, "node-name=%s is conflicting with a device id", | |
911 | node_name); | |
912 | goto out; | |
913 | } | |
914 | ||
915 | /* takes care of avoiding duplicates node names */ | |
916 | if (bdrv_find_node(node_name)) { | |
917 | error_setg(errp, "Duplicate node name"); | |
918 | goto out; | |
919 | } | |
920 | ||
921 | /* copy node name into the bs and insert it into the graph list */ | |
922 | pstrcpy(bs->node_name, sizeof(bs->node_name), node_name); | |
923 | QTAILQ_INSERT_TAIL(&graph_bdrv_states, bs, node_list); | |
924 | out: | |
925 | g_free(gen_node_name); | |
926 | } | |
927 | ||
928 | static int bdrv_open_driver(BlockDriverState *bs, BlockDriver *drv, | |
929 | const char *node_name, QDict *options, | |
930 | int open_flags, Error **errp) | |
931 | { | |
932 | Error *local_err = NULL; | |
933 | int ret; | |
934 | ||
935 | bdrv_assign_node_name(bs, node_name, &local_err); | |
936 | if (local_err) { | |
937 | error_propagate(errp, local_err); | |
938 | return -EINVAL; | |
939 | } | |
940 | ||
941 | bs->drv = drv; | |
942 | bs->read_only = !(bs->open_flags & BDRV_O_RDWR); | |
943 | bs->opaque = g_malloc0(drv->instance_size); | |
944 | ||
945 | if (drv->bdrv_file_open) { | |
946 | assert(!drv->bdrv_needs_filename || bs->filename[0]); | |
947 | ret = drv->bdrv_file_open(bs, options, open_flags, &local_err); | |
948 | } else if (drv->bdrv_open) { | |
949 | ret = drv->bdrv_open(bs, options, open_flags, &local_err); | |
950 | } else { | |
951 | ret = 0; | |
952 | } | |
953 | ||
954 | if (ret < 0) { | |
955 | if (local_err) { | |
956 | error_propagate(errp, local_err); | |
957 | } else if (bs->filename[0]) { | |
958 | error_setg_errno(errp, -ret, "Could not open '%s'", bs->filename); | |
959 | } else { | |
960 | error_setg_errno(errp, -ret, "Could not open image"); | |
961 | } | |
962 | goto free_and_fail; | |
963 | } | |
964 | ||
965 | ret = refresh_total_sectors(bs, bs->total_sectors); | |
966 | if (ret < 0) { | |
967 | error_setg_errno(errp, -ret, "Could not refresh total sector count"); | |
968 | goto free_and_fail; | |
969 | } | |
970 | ||
971 | bdrv_refresh_limits(bs, &local_err); | |
972 | if (local_err) { | |
973 | error_propagate(errp, local_err); | |
974 | ret = -EINVAL; | |
975 | goto free_and_fail; | |
976 | } | |
977 | ||
978 | assert(bdrv_opt_mem_align(bs) != 0); | |
979 | assert(bdrv_min_mem_align(bs) != 0); | |
980 | assert(is_power_of_2(bs->bl.request_alignment)); | |
981 | ||
982 | return 0; | |
983 | ||
984 | free_and_fail: | |
985 | /* FIXME Close bs first if already opened*/ | |
986 | g_free(bs->opaque); | |
987 | bs->opaque = NULL; | |
988 | bs->drv = NULL; | |
989 | return ret; | |
990 | } | |
991 | ||
992 | BlockDriverState *bdrv_new_open_driver(BlockDriver *drv, const char *node_name, | |
993 | int flags, Error **errp) | |
994 | { | |
995 | BlockDriverState *bs; | |
996 | int ret; | |
997 | ||
998 | bs = bdrv_new(); | |
999 | bs->open_flags = flags; | |
1000 | bs->explicit_options = qdict_new(); | |
1001 | bs->options = qdict_new(); | |
1002 | bs->opaque = NULL; | |
1003 | ||
1004 | update_options_from_flags(bs->options, flags); | |
1005 | ||
1006 | ret = bdrv_open_driver(bs, drv, node_name, bs->options, flags, errp); | |
1007 | if (ret < 0) { | |
1008 | QDECREF(bs->explicit_options); | |
1009 | QDECREF(bs->options); | |
1010 | bdrv_unref(bs); | |
1011 | return NULL; | |
1012 | } | |
1013 | ||
1014 | return bs; | |
1015 | } | |
1016 | ||
1017 | QemuOptsList bdrv_runtime_opts = { | |
1018 | .name = "bdrv_common", | |
1019 | .head = QTAILQ_HEAD_INITIALIZER(bdrv_runtime_opts.head), | |
1020 | .desc = { | |
1021 | { | |
1022 | .name = "node-name", | |
1023 | .type = QEMU_OPT_STRING, | |
1024 | .help = "Node name of the block device node", | |
1025 | }, | |
1026 | { | |
1027 | .name = "driver", | |
1028 | .type = QEMU_OPT_STRING, | |
1029 | .help = "Block driver to use for the node", | |
1030 | }, | |
1031 | { | |
1032 | .name = BDRV_OPT_CACHE_DIRECT, | |
1033 | .type = QEMU_OPT_BOOL, | |
1034 | .help = "Bypass software writeback cache on the host", | |
1035 | }, | |
1036 | { | |
1037 | .name = BDRV_OPT_CACHE_NO_FLUSH, | |
1038 | .type = QEMU_OPT_BOOL, | |
1039 | .help = "Ignore flush requests", | |
1040 | }, | |
1041 | { | |
1042 | .name = BDRV_OPT_READ_ONLY, | |
1043 | .type = QEMU_OPT_BOOL, | |
1044 | .help = "Node is opened in read-only mode", | |
1045 | }, | |
1046 | { | |
1047 | .name = "detect-zeroes", | |
1048 | .type = QEMU_OPT_STRING, | |
1049 | .help = "try to optimize zero writes (off, on, unmap)", | |
1050 | }, | |
1051 | { | |
1052 | .name = "discard", | |
1053 | .type = QEMU_OPT_STRING, | |
1054 | .help = "discard operation (ignore/off, unmap/on)", | |
1055 | }, | |
1056 | { /* end of list */ } | |
1057 | }, | |
1058 | }; | |
1059 | ||
1060 | /* | |
1061 | * Common part for opening disk images and files | |
1062 | * | |
1063 | * Removes all processed options from *options. | |
1064 | */ | |
1065 | static int bdrv_open_common(BlockDriverState *bs, BlockBackend *file, | |
1066 | QDict *options, Error **errp) | |
1067 | { | |
1068 | int ret, open_flags; | |
1069 | const char *filename; | |
1070 | const char *driver_name = NULL; | |
1071 | const char *node_name = NULL; | |
1072 | const char *discard; | |
1073 | const char *detect_zeroes; | |
1074 | QemuOpts *opts; | |
1075 | BlockDriver *drv; | |
1076 | Error *local_err = NULL; | |
1077 | ||
1078 | assert(bs->file == NULL); | |
1079 | assert(options != NULL && bs->options != options); | |
1080 | ||
1081 | opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort); | |
1082 | qemu_opts_absorb_qdict(opts, options, &local_err); | |
1083 | if (local_err) { | |
1084 | error_propagate(errp, local_err); | |
1085 | ret = -EINVAL; | |
1086 | goto fail_opts; | |
1087 | } | |
1088 | ||
1089 | update_flags_from_options(&bs->open_flags, opts); | |
1090 | ||
1091 | driver_name = qemu_opt_get(opts, "driver"); | |
1092 | drv = bdrv_find_format(driver_name); | |
1093 | assert(drv != NULL); | |
1094 | ||
1095 | if (file != NULL) { | |
1096 | filename = blk_bs(file)->filename; | |
1097 | } else { | |
1098 | filename = qdict_get_try_str(options, "filename"); | |
1099 | } | |
1100 | ||
1101 | if (drv->bdrv_needs_filename && !filename) { | |
1102 | error_setg(errp, "The '%s' block driver requires a file name", | |
1103 | drv->format_name); | |
1104 | ret = -EINVAL; | |
1105 | goto fail_opts; | |
1106 | } | |
1107 | ||
1108 | trace_bdrv_open_common(bs, filename ?: "", bs->open_flags, | |
1109 | drv->format_name); | |
1110 | ||
1111 | bs->read_only = !(bs->open_flags & BDRV_O_RDWR); | |
1112 | ||
1113 | if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv, bs->read_only)) { | |
1114 | error_setg(errp, | |
1115 | !bs->read_only && bdrv_is_whitelisted(drv, true) | |
1116 | ? "Driver '%s' can only be used for read-only devices" | |
1117 | : "Driver '%s' is not whitelisted", | |
1118 | drv->format_name); | |
1119 | ret = -ENOTSUP; | |
1120 | goto fail_opts; | |
1121 | } | |
1122 | ||
1123 | assert(bs->copy_on_read == 0); /* bdrv_new() and bdrv_close() make it so */ | |
1124 | if (bs->open_flags & BDRV_O_COPY_ON_READ) { | |
1125 | if (!bs->read_only) { | |
1126 | bdrv_enable_copy_on_read(bs); | |
1127 | } else { | |
1128 | error_setg(errp, "Can't use copy-on-read on read-only device"); | |
1129 | ret = -EINVAL; | |
1130 | goto fail_opts; | |
1131 | } | |
1132 | } | |
1133 | ||
1134 | discard = qemu_opt_get(opts, "discard"); | |
1135 | if (discard != NULL) { | |
1136 | if (bdrv_parse_discard_flags(discard, &bs->open_flags) != 0) { | |
1137 | error_setg(errp, "Invalid discard option"); | |
1138 | ret = -EINVAL; | |
1139 | goto fail_opts; | |
1140 | } | |
1141 | } | |
1142 | ||
1143 | detect_zeroes = qemu_opt_get(opts, "detect-zeroes"); | |
1144 | if (detect_zeroes) { | |
1145 | BlockdevDetectZeroesOptions value = | |
1146 | qapi_enum_parse(BlockdevDetectZeroesOptions_lookup, | |
1147 | detect_zeroes, | |
1148 | BLOCKDEV_DETECT_ZEROES_OPTIONS__MAX, | |
1149 | BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF, | |
1150 | &local_err); | |
1151 | if (local_err) { | |
1152 | error_propagate(errp, local_err); | |
1153 | ret = -EINVAL; | |
1154 | goto fail_opts; | |
1155 | } | |
1156 | ||
1157 | if (value == BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP && | |
1158 | !(bs->open_flags & BDRV_O_UNMAP)) | |
1159 | { | |
1160 | error_setg(errp, "setting detect-zeroes to unmap is not allowed " | |
1161 | "without setting discard operation to unmap"); | |
1162 | ret = -EINVAL; | |
1163 | goto fail_opts; | |
1164 | } | |
1165 | ||
1166 | bs->detect_zeroes = value; | |
1167 | } | |
1168 | ||
1169 | if (filename != NULL) { | |
1170 | pstrcpy(bs->filename, sizeof(bs->filename), filename); | |
1171 | } else { | |
1172 | bs->filename[0] = '\0'; | |
1173 | } | |
1174 | pstrcpy(bs->exact_filename, sizeof(bs->exact_filename), bs->filename); | |
1175 | ||
1176 | /* Open the image, either directly or using a protocol */ | |
1177 | open_flags = bdrv_open_flags(bs, bs->open_flags); | |
1178 | node_name = qemu_opt_get(opts, "node-name"); | |
1179 | ||
1180 | assert(!drv->bdrv_file_open || file == NULL); | |
1181 | ret = bdrv_open_driver(bs, drv, node_name, options, open_flags, errp); | |
1182 | if (ret < 0) { | |
1183 | goto fail_opts; | |
1184 | } | |
1185 | ||
1186 | qemu_opts_del(opts); | |
1187 | return 0; | |
1188 | ||
1189 | fail_opts: | |
1190 | qemu_opts_del(opts); | |
1191 | return ret; | |
1192 | } | |
1193 | ||
1194 | static QDict *parse_json_filename(const char *filename, Error **errp) | |
1195 | { | |
1196 | QObject *options_obj; | |
1197 | QDict *options; | |
1198 | int ret; | |
1199 | ||
1200 | ret = strstart(filename, "json:", &filename); | |
1201 | assert(ret); | |
1202 | ||
1203 | options_obj = qobject_from_json(filename); | |
1204 | if (!options_obj) { | |
1205 | error_setg(errp, "Could not parse the JSON options"); | |
1206 | return NULL; | |
1207 | } | |
1208 | ||
1209 | options = qobject_to_qdict(options_obj); | |
1210 | if (!options) { | |
1211 | qobject_decref(options_obj); | |
1212 | error_setg(errp, "Invalid JSON object given"); | |
1213 | return NULL; | |
1214 | } | |
1215 | ||
1216 | qdict_flatten(options); | |
1217 | ||
1218 | return options; | |
1219 | } | |
1220 | ||
1221 | static void parse_json_protocol(QDict *options, const char **pfilename, | |
1222 | Error **errp) | |
1223 | { | |
1224 | QDict *json_options; | |
1225 | Error *local_err = NULL; | |
1226 | ||
1227 | /* Parse json: pseudo-protocol */ | |
1228 | if (!*pfilename || !g_str_has_prefix(*pfilename, "json:")) { | |
1229 | return; | |
1230 | } | |
1231 | ||
1232 | json_options = parse_json_filename(*pfilename, &local_err); | |
1233 | if (local_err) { | |
1234 | error_propagate(errp, local_err); | |
1235 | return; | |
1236 | } | |
1237 | ||
1238 | /* Options given in the filename have lower priority than options | |
1239 | * specified directly */ | |
1240 | qdict_join(options, json_options, false); | |
1241 | QDECREF(json_options); | |
1242 | *pfilename = NULL; | |
1243 | } | |
1244 | ||
1245 | /* | |
1246 | * Fills in default options for opening images and converts the legacy | |
1247 | * filename/flags pair to option QDict entries. | |
1248 | * The BDRV_O_PROTOCOL flag in *flags will be set or cleared accordingly if a | |
1249 | * block driver has been specified explicitly. | |
1250 | */ | |
1251 | static int bdrv_fill_options(QDict **options, const char *filename, | |
1252 | int *flags, Error **errp) | |
1253 | { | |
1254 | const char *drvname; | |
1255 | bool protocol = *flags & BDRV_O_PROTOCOL; | |
1256 | bool parse_filename = false; | |
1257 | BlockDriver *drv = NULL; | |
1258 | Error *local_err = NULL; | |
1259 | ||
1260 | drvname = qdict_get_try_str(*options, "driver"); | |
1261 | if (drvname) { | |
1262 | drv = bdrv_find_format(drvname); | |
1263 | if (!drv) { | |
1264 | error_setg(errp, "Unknown driver '%s'", drvname); | |
1265 | return -ENOENT; | |
1266 | } | |
1267 | /* If the user has explicitly specified the driver, this choice should | |
1268 | * override the BDRV_O_PROTOCOL flag */ | |
1269 | protocol = drv->bdrv_file_open; | |
1270 | } | |
1271 | ||
1272 | if (protocol) { | |
1273 | *flags |= BDRV_O_PROTOCOL; | |
1274 | } else { | |
1275 | *flags &= ~BDRV_O_PROTOCOL; | |
1276 | } | |
1277 | ||
1278 | /* Translate cache options from flags into options */ | |
1279 | update_options_from_flags(*options, *flags); | |
1280 | ||
1281 | /* Fetch the file name from the options QDict if necessary */ | |
1282 | if (protocol && filename) { | |
1283 | if (!qdict_haskey(*options, "filename")) { | |
1284 | qdict_put(*options, "filename", qstring_from_str(filename)); | |
1285 | parse_filename = true; | |
1286 | } else { | |
1287 | error_setg(errp, "Can't specify 'file' and 'filename' options at " | |
1288 | "the same time"); | |
1289 | return -EINVAL; | |
1290 | } | |
1291 | } | |
1292 | ||
1293 | /* Find the right block driver */ | |
1294 | filename = qdict_get_try_str(*options, "filename"); | |
1295 | ||
1296 | if (!drvname && protocol) { | |
1297 | if (filename) { | |
1298 | drv = bdrv_find_protocol(filename, parse_filename, errp); | |
1299 | if (!drv) { | |
1300 | return -EINVAL; | |
1301 | } | |
1302 | ||
1303 | drvname = drv->format_name; | |
1304 | qdict_put(*options, "driver", qstring_from_str(drvname)); | |
1305 | } else { | |
1306 | error_setg(errp, "Must specify either driver or file"); | |
1307 | return -EINVAL; | |
1308 | } | |
1309 | } | |
1310 | ||
1311 | assert(drv || !protocol); | |
1312 | ||
1313 | /* Driver-specific filename parsing */ | |
1314 | if (drv && drv->bdrv_parse_filename && parse_filename) { | |
1315 | drv->bdrv_parse_filename(filename, *options, &local_err); | |
1316 | if (local_err) { | |
1317 | error_propagate(errp, local_err); | |
1318 | return -EINVAL; | |
1319 | } | |
1320 | ||
1321 | if (!drv->bdrv_needs_filename) { | |
1322 | qdict_del(*options, "filename"); | |
1323 | } | |
1324 | } | |
1325 | ||
1326 | return 0; | |
1327 | } | |
1328 | ||
1329 | /* | |
1330 | * Check whether permissions on this node can be changed in a way that | |
1331 | * @cumulative_perms and @cumulative_shared_perms are the new cumulative | |
1332 | * permissions of all its parents. This involves checking whether all necessary | |
1333 | * permission changes to child nodes can be performed. | |
1334 | * | |
1335 | * A call to this function must always be followed by a call to bdrv_set_perm() | |
1336 | * or bdrv_abort_perm_update(). | |
1337 | */ | |
1338 | static int bdrv_check_perm(BlockDriverState *bs, uint64_t cumulative_perms, | |
1339 | uint64_t cumulative_shared_perms, Error **errp) | |
1340 | { | |
1341 | BlockDriver *drv = bs->drv; | |
1342 | BdrvChild *c; | |
1343 | int ret; | |
1344 | ||
1345 | /* Write permissions never work with read-only images */ | |
1346 | if ((cumulative_perms & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)) && | |
1347 | bdrv_is_read_only(bs)) | |
1348 | { | |
1349 | error_setg(errp, "Block node is read-only"); | |
1350 | return -EPERM; | |
1351 | } | |
1352 | ||
1353 | /* Check this node */ | |
1354 | if (!drv) { | |
1355 | return 0; | |
1356 | } | |
1357 | ||
1358 | if (drv->bdrv_check_perm) { | |
1359 | return drv->bdrv_check_perm(bs, cumulative_perms, | |
1360 | cumulative_shared_perms, errp); | |
1361 | } | |
1362 | ||
1363 | /* Drivers that never have children can omit .bdrv_child_perm() */ | |
1364 | if (!drv->bdrv_child_perm) { | |
1365 | assert(QLIST_EMPTY(&bs->children)); | |
1366 | return 0; | |
1367 | } | |
1368 | ||
1369 | /* Check all children */ | |
1370 | QLIST_FOREACH(c, &bs->children, next) { | |
1371 | uint64_t cur_perm, cur_shared; | |
1372 | drv->bdrv_child_perm(bs, c, c->role, | |
1373 | cumulative_perms, cumulative_shared_perms, | |
1374 | &cur_perm, &cur_shared); | |
1375 | ret = bdrv_child_check_perm(c, cur_perm, cur_shared, errp); | |
1376 | if (ret < 0) { | |
1377 | return ret; | |
1378 | } | |
1379 | } | |
1380 | ||
1381 | return 0; | |
1382 | } | |
1383 | ||
1384 | /* | |
1385 | * Notifies drivers that after a previous bdrv_check_perm() call, the | |
1386 | * permission update is not performed and any preparations made for it (e.g. | |
1387 | * taken file locks) need to be undone. | |
1388 | * | |
1389 | * This function recursively notifies all child nodes. | |
1390 | */ | |
1391 | static void bdrv_abort_perm_update(BlockDriverState *bs) | |
1392 | { | |
1393 | BlockDriver *drv = bs->drv; | |
1394 | BdrvChild *c; | |
1395 | ||
1396 | if (!drv) { | |
1397 | return; | |
1398 | } | |
1399 | ||
1400 | if (drv->bdrv_abort_perm_update) { | |
1401 | drv->bdrv_abort_perm_update(bs); | |
1402 | } | |
1403 | ||
1404 | QLIST_FOREACH(c, &bs->children, next) { | |
1405 | bdrv_child_abort_perm_update(c); | |
1406 | } | |
1407 | } | |
1408 | ||
1409 | static void bdrv_set_perm(BlockDriverState *bs, uint64_t cumulative_perms, | |
1410 | uint64_t cumulative_shared_perms) | |
1411 | { | |
1412 | BlockDriver *drv = bs->drv; | |
1413 | BdrvChild *c; | |
1414 | ||
1415 | if (!drv) { | |
1416 | return; | |
1417 | } | |
1418 | ||
1419 | /* Update this node */ | |
1420 | if (drv->bdrv_set_perm) { | |
1421 | drv->bdrv_set_perm(bs, cumulative_perms, cumulative_shared_perms); | |
1422 | } | |
1423 | ||
1424 | /* Drivers that never have children can omit .bdrv_child_perm() */ | |
1425 | if (!drv->bdrv_child_perm) { | |
1426 | assert(QLIST_EMPTY(&bs->children)); | |
1427 | return; | |
1428 | } | |
1429 | ||
1430 | /* Update all children */ | |
1431 | QLIST_FOREACH(c, &bs->children, next) { | |
1432 | uint64_t cur_perm, cur_shared; | |
1433 | drv->bdrv_child_perm(bs, c, c->role, | |
1434 | cumulative_perms, cumulative_shared_perms, | |
1435 | &cur_perm, &cur_shared); | |
1436 | bdrv_child_set_perm(c, cur_perm, cur_shared); | |
1437 | } | |
1438 | } | |
1439 | ||
1440 | static void bdrv_get_cumulative_perm(BlockDriverState *bs, uint64_t *perm, | |
1441 | uint64_t *shared_perm) | |
1442 | { | |
1443 | BdrvChild *c; | |
1444 | uint64_t cumulative_perms = 0; | |
1445 | uint64_t cumulative_shared_perms = BLK_PERM_ALL; | |
1446 | ||
1447 | QLIST_FOREACH(c, &bs->parents, next_parent) { | |
1448 | cumulative_perms |= c->perm; | |
1449 | cumulative_shared_perms &= c->shared_perm; | |
1450 | } | |
1451 | ||
1452 | *perm = cumulative_perms; | |
1453 | *shared_perm = cumulative_shared_perms; | |
1454 | } | |
1455 | ||
1456 | /* | |
1457 | * Checks whether a new reference to @bs can be added if the new user requires | |
1458 | * @new_used_perm/@new_shared_perm as its permissions. If @ignore_child is set, | |
1459 | * this old reference is ignored in the calculations; this allows checking | |
1460 | * permission updates for an existing reference. | |
1461 | * | |
1462 | * Needs to be followed by a call to either bdrv_set_perm() or | |
1463 | * bdrv_abort_perm_update(). */ | |
1464 | static int bdrv_check_update_perm(BlockDriverState *bs, uint64_t new_used_perm, | |
1465 | uint64_t new_shared_perm, | |
1466 | BdrvChild *ignore_child, Error **errp) | |
1467 | { | |
1468 | BdrvChild *c; | |
1469 | uint64_t cumulative_perms = new_used_perm; | |
1470 | uint64_t cumulative_shared_perms = new_shared_perm; | |
1471 | ||
1472 | /* There is no reason why anyone couldn't tolerate write_unchanged */ | |
1473 | assert(new_shared_perm & BLK_PERM_WRITE_UNCHANGED); | |
1474 | ||
1475 | QLIST_FOREACH(c, &bs->parents, next_parent) { | |
1476 | if (c == ignore_child) { | |
1477 | continue; | |
1478 | } | |
1479 | ||
1480 | if ((new_used_perm & c->shared_perm) != new_used_perm || | |
1481 | (c->perm & new_shared_perm) != c->perm) | |
1482 | { | |
1483 | const char *user = NULL; | |
1484 | if (c->role->get_name) { | |
1485 | user = c->role->get_name(c); | |
1486 | if (user && !*user) { | |
1487 | user = NULL; | |
1488 | } | |
1489 | } | |
1490 | error_setg(errp, "Conflicts with %s", user ?: "another operation"); | |
1491 | return -EPERM; | |
1492 | } | |
1493 | ||
1494 | cumulative_perms |= c->perm; | |
1495 | cumulative_shared_perms &= c->shared_perm; | |
1496 | } | |
1497 | ||
1498 | return bdrv_check_perm(bs, cumulative_perms, cumulative_shared_perms, errp); | |
1499 | } | |
1500 | ||
1501 | /* Needs to be followed by a call to either bdrv_child_set_perm() or | |
1502 | * bdrv_child_abort_perm_update(). */ | |
1503 | int bdrv_child_check_perm(BdrvChild *c, uint64_t perm, uint64_t shared, | |
1504 | Error **errp) | |
1505 | { | |
1506 | return bdrv_check_update_perm(c->bs, perm, shared, c, errp); | |
1507 | } | |
1508 | ||
1509 | void bdrv_child_set_perm(BdrvChild *c, uint64_t perm, uint64_t shared) | |
1510 | { | |
1511 | uint64_t cumulative_perms, cumulative_shared_perms; | |
1512 | ||
1513 | c->perm = perm; | |
1514 | c->shared_perm = shared; | |
1515 | ||
1516 | bdrv_get_cumulative_perm(c->bs, &cumulative_perms, | |
1517 | &cumulative_shared_perms); | |
1518 | bdrv_set_perm(c->bs, cumulative_perms, cumulative_shared_perms); | |
1519 | } | |
1520 | ||
1521 | void bdrv_child_abort_perm_update(BdrvChild *c) | |
1522 | { | |
1523 | bdrv_abort_perm_update(c->bs); | |
1524 | } | |
1525 | ||
1526 | int bdrv_child_try_set_perm(BdrvChild *c, uint64_t perm, uint64_t shared, | |
1527 | Error **errp) | |
1528 | { | |
1529 | int ret; | |
1530 | ||
1531 | ret = bdrv_child_check_perm(c, perm, shared, errp); | |
1532 | if (ret < 0) { | |
1533 | bdrv_child_abort_perm_update(c); | |
1534 | return ret; | |
1535 | } | |
1536 | ||
1537 | bdrv_child_set_perm(c, perm, shared); | |
1538 | ||
1539 | return 0; | |
1540 | } | |
1541 | ||
1542 | #define DEFAULT_PERM_PASSTHROUGH (BLK_PERM_CONSISTENT_READ \ | |
1543 | | BLK_PERM_WRITE \ | |
1544 | | BLK_PERM_WRITE_UNCHANGED \ | |
1545 | | BLK_PERM_RESIZE) | |
1546 | #define DEFAULT_PERM_UNCHANGED (BLK_PERM_ALL & ~DEFAULT_PERM_PASSTHROUGH) | |
1547 | ||
1548 | void bdrv_filter_default_perms(BlockDriverState *bs, BdrvChild *c, | |
1549 | const BdrvChildRole *role, | |
1550 | uint64_t perm, uint64_t shared, | |
1551 | uint64_t *nperm, uint64_t *nshared) | |
1552 | { | |
1553 | if (c == NULL) { | |
1554 | *nperm = perm & DEFAULT_PERM_PASSTHROUGH; | |
1555 | *nshared = (shared & DEFAULT_PERM_PASSTHROUGH) | DEFAULT_PERM_UNCHANGED; | |
1556 | return; | |
1557 | } | |
1558 | ||
1559 | *nperm = (perm & DEFAULT_PERM_PASSTHROUGH) | | |
1560 | (c->perm & DEFAULT_PERM_UNCHANGED); | |
1561 | *nshared = (shared & DEFAULT_PERM_PASSTHROUGH) | | |
1562 | (c->shared_perm & DEFAULT_PERM_UNCHANGED); | |
1563 | } | |
1564 | ||
1565 | void bdrv_format_default_perms(BlockDriverState *bs, BdrvChild *c, | |
1566 | const BdrvChildRole *role, | |
1567 | uint64_t perm, uint64_t shared, | |
1568 | uint64_t *nperm, uint64_t *nshared) | |
1569 | { | |
1570 | bool backing = (role == &child_backing); | |
1571 | assert(role == &child_backing || role == &child_file); | |
1572 | ||
1573 | if (!backing) { | |
1574 | /* Apart from the modifications below, the same permissions are | |
1575 | * forwarded and left alone as for filters */ | |
1576 | bdrv_filter_default_perms(bs, c, role, perm, shared, &perm, &shared); | |
1577 | ||
1578 | /* Format drivers may touch metadata even if the guest doesn't write */ | |
1579 | if (!bdrv_is_read_only(bs)) { | |
1580 | perm |= BLK_PERM_WRITE | BLK_PERM_RESIZE; | |
1581 | } | |
1582 | ||
1583 | /* bs->file always needs to be consistent because of the metadata. We | |
1584 | * can never allow other users to resize or write to it. */ | |
1585 | perm |= BLK_PERM_CONSISTENT_READ; | |
1586 | shared &= ~(BLK_PERM_WRITE | BLK_PERM_RESIZE); | |
1587 | } else { | |
1588 | /* We want consistent read from backing files if the parent needs it. | |
1589 | * No other operations are performed on backing files. */ | |
1590 | perm &= BLK_PERM_CONSISTENT_READ; | |
1591 | ||
1592 | /* If the parent can deal with changing data, we're okay with a | |
1593 | * writable and resizable backing file. */ | |
1594 | /* TODO Require !(perm & BLK_PERM_CONSISTENT_READ), too? */ | |
1595 | if (shared & BLK_PERM_WRITE) { | |
1596 | shared = BLK_PERM_WRITE | BLK_PERM_RESIZE; | |
1597 | } else { | |
1598 | shared = 0; | |
1599 | } | |
1600 | ||
1601 | shared |= BLK_PERM_CONSISTENT_READ | BLK_PERM_GRAPH_MOD | | |
1602 | BLK_PERM_WRITE_UNCHANGED; | |
1603 | } | |
1604 | ||
1605 | *nperm = perm; | |
1606 | *nshared = shared; | |
1607 | } | |
1608 | ||
1609 | static void bdrv_replace_child(BdrvChild *child, BlockDriverState *new_bs, | |
1610 | bool check_new_perm) | |
1611 | { | |
1612 | BlockDriverState *old_bs = child->bs; | |
1613 | uint64_t perm, shared_perm; | |
1614 | ||
1615 | if (old_bs) { | |
1616 | if (old_bs->quiesce_counter && child->role->drained_end) { | |
1617 | child->role->drained_end(child); | |
1618 | } | |
1619 | QLIST_REMOVE(child, next_parent); | |
1620 | ||
1621 | /* Update permissions for old node. This is guaranteed to succeed | |
1622 | * because we're just taking a parent away, so we're loosening | |
1623 | * restrictions. */ | |
1624 | bdrv_get_cumulative_perm(old_bs, &perm, &shared_perm); | |
1625 | bdrv_check_perm(old_bs, perm, shared_perm, &error_abort); | |
1626 | bdrv_set_perm(old_bs, perm, shared_perm); | |
1627 | } | |
1628 | ||
1629 | child->bs = new_bs; | |
1630 | ||
1631 | if (new_bs) { | |
1632 | QLIST_INSERT_HEAD(&new_bs->parents, child, next_parent); | |
1633 | if (new_bs->quiesce_counter && child->role->drained_begin) { | |
1634 | child->role->drained_begin(child); | |
1635 | } | |
1636 | ||
1637 | bdrv_get_cumulative_perm(new_bs, &perm, &shared_perm); | |
1638 | if (check_new_perm) { | |
1639 | bdrv_check_perm(new_bs, perm, shared_perm, &error_abort); | |
1640 | } | |
1641 | bdrv_set_perm(new_bs, perm, shared_perm); | |
1642 | } | |
1643 | } | |
1644 | ||
1645 | BdrvChild *bdrv_root_attach_child(BlockDriverState *child_bs, | |
1646 | const char *child_name, | |
1647 | const BdrvChildRole *child_role, | |
1648 | uint64_t perm, uint64_t shared_perm, | |
1649 | void *opaque, Error **errp) | |
1650 | { | |
1651 | BdrvChild *child; | |
1652 | int ret; | |
1653 | ||
1654 | ret = bdrv_check_update_perm(child_bs, perm, shared_perm, NULL, errp); | |
1655 | if (ret < 0) { | |
1656 | bdrv_abort_perm_update(child_bs); | |
1657 | return NULL; | |
1658 | } | |
1659 | ||
1660 | child = g_new(BdrvChild, 1); | |
1661 | *child = (BdrvChild) { | |
1662 | .bs = NULL, | |
1663 | .name = g_strdup(child_name), | |
1664 | .role = child_role, | |
1665 | .perm = perm, | |
1666 | .shared_perm = shared_perm, | |
1667 | .opaque = opaque, | |
1668 | }; | |
1669 | ||
1670 | /* This performs the matching bdrv_set_perm() for the above check. */ | |
1671 | bdrv_replace_child(child, child_bs, false); | |
1672 | ||
1673 | return child; | |
1674 | } | |
1675 | ||
1676 | BdrvChild *bdrv_attach_child(BlockDriverState *parent_bs, | |
1677 | BlockDriverState *child_bs, | |
1678 | const char *child_name, | |
1679 | const BdrvChildRole *child_role, | |
1680 | Error **errp) | |
1681 | { | |
1682 | BdrvChild *child; | |
1683 | uint64_t perm, shared_perm; | |
1684 | ||
1685 | bdrv_get_cumulative_perm(parent_bs, &perm, &shared_perm); | |
1686 | ||
1687 | assert(parent_bs->drv); | |
1688 | parent_bs->drv->bdrv_child_perm(parent_bs, NULL, child_role, | |
1689 | perm, shared_perm, &perm, &shared_perm); | |
1690 | ||
1691 | child = bdrv_root_attach_child(child_bs, child_name, child_role, | |
1692 | perm, shared_perm, parent_bs, errp); | |
1693 | if (child == NULL) { | |
1694 | return NULL; | |
1695 | } | |
1696 | ||
1697 | QLIST_INSERT_HEAD(&parent_bs->children, child, next); | |
1698 | return child; | |
1699 | } | |
1700 | ||
1701 | static void bdrv_detach_child(BdrvChild *child) | |
1702 | { | |
1703 | if (child->next.le_prev) { | |
1704 | QLIST_REMOVE(child, next); | |
1705 | child->next.le_prev = NULL; | |
1706 | } | |
1707 | ||
1708 | bdrv_replace_child(child, NULL, false); | |
1709 | ||
1710 | g_free(child->name); | |
1711 | g_free(child); | |
1712 | } | |
1713 | ||
1714 | void bdrv_root_unref_child(BdrvChild *child) | |
1715 | { | |
1716 | BlockDriverState *child_bs; | |
1717 | ||
1718 | child_bs = child->bs; | |
1719 | bdrv_detach_child(child); | |
1720 | bdrv_unref(child_bs); | |
1721 | } | |
1722 | ||
1723 | void bdrv_unref_child(BlockDriverState *parent, BdrvChild *child) | |
1724 | { | |
1725 | if (child == NULL) { | |
1726 | return; | |
1727 | } | |
1728 | ||
1729 | if (child->bs->inherits_from == parent) { | |
1730 | BdrvChild *c; | |
1731 | ||
1732 | /* Remove inherits_from only when the last reference between parent and | |
1733 | * child->bs goes away. */ | |
1734 | QLIST_FOREACH(c, &parent->children, next) { | |
1735 | if (c != child && c->bs == child->bs) { | |
1736 | break; | |
1737 | } | |
1738 | } | |
1739 | if (c == NULL) { | |
1740 | child->bs->inherits_from = NULL; | |
1741 | } | |
1742 | } | |
1743 | ||
1744 | bdrv_root_unref_child(child); | |
1745 | } | |
1746 | ||
1747 | ||
1748 | static void bdrv_parent_cb_change_media(BlockDriverState *bs, bool load) | |
1749 | { | |
1750 | BdrvChild *c; | |
1751 | QLIST_FOREACH(c, &bs->parents, next_parent) { | |
1752 | if (c->role->change_media) { | |
1753 | c->role->change_media(c, load); | |
1754 | } | |
1755 | } | |
1756 | } | |
1757 | ||
1758 | static void bdrv_parent_cb_resize(BlockDriverState *bs) | |
1759 | { | |
1760 | BdrvChild *c; | |
1761 | QLIST_FOREACH(c, &bs->parents, next_parent) { | |
1762 | if (c->role->resize) { | |
1763 | c->role->resize(c); | |
1764 | } | |
1765 | } | |
1766 | } | |
1767 | ||
1768 | /* | |
1769 | * Sets the backing file link of a BDS. A new reference is created; callers | |
1770 | * which don't need their own reference any more must call bdrv_unref(). | |
1771 | */ | |
1772 | void bdrv_set_backing_hd(BlockDriverState *bs, BlockDriverState *backing_hd) | |
1773 | { | |
1774 | if (backing_hd) { | |
1775 | bdrv_ref(backing_hd); | |
1776 | } | |
1777 | ||
1778 | if (bs->backing) { | |
1779 | assert(bs->backing_blocker); | |
1780 | bdrv_op_unblock_all(bs->backing->bs, bs->backing_blocker); | |
1781 | bdrv_unref_child(bs, bs->backing); | |
1782 | } else if (backing_hd) { | |
1783 | error_setg(&bs->backing_blocker, | |
1784 | "node is used as backing hd of '%s'", | |
1785 | bdrv_get_device_or_node_name(bs)); | |
1786 | } | |
1787 | ||
1788 | if (!backing_hd) { | |
1789 | error_free(bs->backing_blocker); | |
1790 | bs->backing_blocker = NULL; | |
1791 | bs->backing = NULL; | |
1792 | goto out; | |
1793 | } | |
1794 | /* FIXME Error handling */ | |
1795 | bs->backing = bdrv_attach_child(bs, backing_hd, "backing", &child_backing, | |
1796 | &error_abort); | |
1797 | bs->open_flags &= ~BDRV_O_NO_BACKING; | |
1798 | pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_hd->filename); | |
1799 | pstrcpy(bs->backing_format, sizeof(bs->backing_format), | |
1800 | backing_hd->drv ? backing_hd->drv->format_name : ""); | |
1801 | ||
1802 | bdrv_op_block_all(backing_hd, bs->backing_blocker); | |
1803 | /* Otherwise we won't be able to commit or stream */ | |
1804 | bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_COMMIT_TARGET, | |
1805 | bs->backing_blocker); | |
1806 | bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_STREAM, | |
1807 | bs->backing_blocker); | |
1808 | /* | |
1809 | * We do backup in 3 ways: | |
1810 | * 1. drive backup | |
1811 | * The target bs is new opened, and the source is top BDS | |
1812 | * 2. blockdev backup | |
1813 | * Both the source and the target are top BDSes. | |
1814 | * 3. internal backup(used for block replication) | |
1815 | * Both the source and the target are backing file | |
1816 | * | |
1817 | * In case 1 and 2, neither the source nor the target is the backing file. | |
1818 | * In case 3, we will block the top BDS, so there is only one block job | |
1819 | * for the top BDS and its backing chain. | |
1820 | */ | |
1821 | bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_BACKUP_SOURCE, | |
1822 | bs->backing_blocker); | |
1823 | bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_BACKUP_TARGET, | |
1824 | bs->backing_blocker); | |
1825 | out: | |
1826 | bdrv_refresh_limits(bs, NULL); | |
1827 | } | |
1828 | ||
1829 | /* | |
1830 | * Opens the backing file for a BlockDriverState if not yet open | |
1831 | * | |
1832 | * bdref_key specifies the key for the image's BlockdevRef in the options QDict. | |
1833 | * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict | |
1834 | * itself, all options starting with "${bdref_key}." are considered part of the | |
1835 | * BlockdevRef. | |
1836 | * | |
1837 | * TODO Can this be unified with bdrv_open_image()? | |
1838 | */ | |
1839 | int bdrv_open_backing_file(BlockDriverState *bs, QDict *parent_options, | |
1840 | const char *bdref_key, Error **errp) | |
1841 | { | |
1842 | char *backing_filename = g_malloc0(PATH_MAX); | |
1843 | char *bdref_key_dot; | |
1844 | const char *reference = NULL; | |
1845 | int ret = 0; | |
1846 | BlockDriverState *backing_hd; | |
1847 | QDict *options; | |
1848 | QDict *tmp_parent_options = NULL; | |
1849 | Error *local_err = NULL; | |
1850 | ||
1851 | if (bs->backing != NULL) { | |
1852 | goto free_exit; | |
1853 | } | |
1854 | ||
1855 | /* NULL means an empty set of options */ | |
1856 | if (parent_options == NULL) { | |
1857 | tmp_parent_options = qdict_new(); | |
1858 | parent_options = tmp_parent_options; | |
1859 | } | |
1860 | ||
1861 | bs->open_flags &= ~BDRV_O_NO_BACKING; | |
1862 | ||
1863 | bdref_key_dot = g_strdup_printf("%s.", bdref_key); | |
1864 | qdict_extract_subqdict(parent_options, &options, bdref_key_dot); | |
1865 | g_free(bdref_key_dot); | |
1866 | ||
1867 | reference = qdict_get_try_str(parent_options, bdref_key); | |
1868 | if (reference || qdict_haskey(options, "file.filename")) { | |
1869 | backing_filename[0] = '\0'; | |
1870 | } else if (bs->backing_file[0] == '\0' && qdict_size(options) == 0) { | |
1871 | QDECREF(options); | |
1872 | goto free_exit; | |
1873 | } else { | |
1874 | bdrv_get_full_backing_filename(bs, backing_filename, PATH_MAX, | |
1875 | &local_err); | |
1876 | if (local_err) { | |
1877 | ret = -EINVAL; | |
1878 | error_propagate(errp, local_err); | |
1879 | QDECREF(options); | |
1880 | goto free_exit; | |
1881 | } | |
1882 | } | |
1883 | ||
1884 | if (!bs->drv || !bs->drv->supports_backing) { | |
1885 | ret = -EINVAL; | |
1886 | error_setg(errp, "Driver doesn't support backing files"); | |
1887 | QDECREF(options); | |
1888 | goto free_exit; | |
1889 | } | |
1890 | ||
1891 | if (bs->backing_format[0] != '\0' && !qdict_haskey(options, "driver")) { | |
1892 | qdict_put(options, "driver", qstring_from_str(bs->backing_format)); | |
1893 | } | |
1894 | ||
1895 | backing_hd = bdrv_open_inherit(*backing_filename ? backing_filename : NULL, | |
1896 | reference, options, 0, bs, &child_backing, | |
1897 | errp); | |
1898 | if (!backing_hd) { | |
1899 | bs->open_flags |= BDRV_O_NO_BACKING; | |
1900 | error_prepend(errp, "Could not open backing file: "); | |
1901 | ret = -EINVAL; | |
1902 | goto free_exit; | |
1903 | } | |
1904 | ||
1905 | /* Hook up the backing file link; drop our reference, bs owns the | |
1906 | * backing_hd reference now */ | |
1907 | bdrv_set_backing_hd(bs, backing_hd); | |
1908 | bdrv_unref(backing_hd); | |
1909 | ||
1910 | qdict_del(parent_options, bdref_key); | |
1911 | ||
1912 | free_exit: | |
1913 | g_free(backing_filename); | |
1914 | QDECREF(tmp_parent_options); | |
1915 | return ret; | |
1916 | } | |
1917 | ||
1918 | static BlockDriverState * | |
1919 | bdrv_open_child_bs(const char *filename, QDict *options, const char *bdref_key, | |
1920 | BlockDriverState *parent, const BdrvChildRole *child_role, | |
1921 | bool allow_none, Error **errp) | |
1922 | { | |
1923 | BlockDriverState *bs = NULL; | |
1924 | QDict *image_options; | |
1925 | char *bdref_key_dot; | |
1926 | const char *reference; | |
1927 | ||
1928 | assert(child_role != NULL); | |
1929 | ||
1930 | bdref_key_dot = g_strdup_printf("%s.", bdref_key); | |
1931 | qdict_extract_subqdict(options, &image_options, bdref_key_dot); | |
1932 | g_free(bdref_key_dot); | |
1933 | ||
1934 | reference = qdict_get_try_str(options, bdref_key); | |
1935 | if (!filename && !reference && !qdict_size(image_options)) { | |
1936 | if (!allow_none) { | |
1937 | error_setg(errp, "A block device must be specified for \"%s\"", | |
1938 | bdref_key); | |
1939 | } | |
1940 | QDECREF(image_options); | |
1941 | goto done; | |
1942 | } | |
1943 | ||
1944 | bs = bdrv_open_inherit(filename, reference, image_options, 0, | |
1945 | parent, child_role, errp); | |
1946 | if (!bs) { | |
1947 | goto done; | |
1948 | } | |
1949 | ||
1950 | done: | |
1951 | qdict_del(options, bdref_key); | |
1952 | return bs; | |
1953 | } | |
1954 | ||
1955 | /* | |
1956 | * Opens a disk image whose options are given as BlockdevRef in another block | |
1957 | * device's options. | |
1958 | * | |
1959 | * If allow_none is true, no image will be opened if filename is false and no | |
1960 | * BlockdevRef is given. NULL will be returned, but errp remains unset. | |
1961 | * | |
1962 | * bdrev_key specifies the key for the image's BlockdevRef in the options QDict. | |
1963 | * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict | |
1964 | * itself, all options starting with "${bdref_key}." are considered part of the | |
1965 | * BlockdevRef. | |
1966 | * | |
1967 | * The BlockdevRef will be removed from the options QDict. | |
1968 | */ | |
1969 | BdrvChild *bdrv_open_child(const char *filename, | |
1970 | QDict *options, const char *bdref_key, | |
1971 | BlockDriverState *parent, | |
1972 | const BdrvChildRole *child_role, | |
1973 | bool allow_none, Error **errp) | |
1974 | { | |
1975 | BdrvChild *c; | |
1976 | BlockDriverState *bs; | |
1977 | ||
1978 | bs = bdrv_open_child_bs(filename, options, bdref_key, parent, child_role, | |
1979 | allow_none, errp); | |
1980 | if (bs == NULL) { | |
1981 | return NULL; | |
1982 | } | |
1983 | ||
1984 | c = bdrv_attach_child(parent, bs, bdref_key, child_role, errp); | |
1985 | if (!c) { | |
1986 | bdrv_unref(bs); | |
1987 | return NULL; | |
1988 | } | |
1989 | ||
1990 | return c; | |
1991 | } | |
1992 | ||
1993 | static BlockDriverState *bdrv_append_temp_snapshot(BlockDriverState *bs, | |
1994 | int flags, | |
1995 | QDict *snapshot_options, | |
1996 | Error **errp) | |
1997 | { | |
1998 | /* TODO: extra byte is a hack to ensure MAX_PATH space on Windows. */ | |
1999 | char *tmp_filename = g_malloc0(PATH_MAX + 1); | |
2000 | int64_t total_size; | |
2001 | QemuOpts *opts = NULL; | |
2002 | BlockDriverState *bs_snapshot; | |
2003 | int ret; | |
2004 | ||
2005 | /* if snapshot, we create a temporary backing file and open it | |
2006 | instead of opening 'filename' directly */ | |
2007 | ||
2008 | /* Get the required size from the image */ | |
2009 | total_size = bdrv_getlength(bs); | |
2010 | if (total_size < 0) { | |
2011 | error_setg_errno(errp, -total_size, "Could not get image size"); | |
2012 | goto out; | |
2013 | } | |
2014 | ||
2015 | /* Create the temporary image */ | |
2016 | ret = get_tmp_filename(tmp_filename, PATH_MAX + 1); | |
2017 | if (ret < 0) { | |
2018 | error_setg_errno(errp, -ret, "Could not get temporary filename"); | |
2019 | goto out; | |
2020 | } | |
2021 | ||
2022 | opts = qemu_opts_create(bdrv_qcow2.create_opts, NULL, 0, | |
2023 | &error_abort); | |
2024 | qemu_opt_set_number(opts, BLOCK_OPT_SIZE, total_size, &error_abort); | |
2025 | ret = bdrv_create(&bdrv_qcow2, tmp_filename, opts, errp); | |
2026 | qemu_opts_del(opts); | |
2027 | if (ret < 0) { | |
2028 | error_prepend(errp, "Could not create temporary overlay '%s': ", | |
2029 | tmp_filename); | |
2030 | goto out; | |
2031 | } | |
2032 | ||
2033 | /* Prepare options QDict for the temporary file */ | |
2034 | qdict_put(snapshot_options, "file.driver", | |
2035 | qstring_from_str("file")); | |
2036 | qdict_put(snapshot_options, "file.filename", | |
2037 | qstring_from_str(tmp_filename)); | |
2038 | qdict_put(snapshot_options, "driver", | |
2039 | qstring_from_str("qcow2")); | |
2040 | ||
2041 | bs_snapshot = bdrv_open(NULL, NULL, snapshot_options, flags, errp); | |
2042 | snapshot_options = NULL; | |
2043 | if (!bs_snapshot) { | |
2044 | ret = -EINVAL; | |
2045 | goto out; | |
2046 | } | |
2047 | ||
2048 | /* bdrv_append() consumes a strong reference to bs_snapshot (i.e. it will | |
2049 | * call bdrv_unref() on it), so in order to be able to return one, we have | |
2050 | * to increase bs_snapshot's refcount here */ | |
2051 | bdrv_ref(bs_snapshot); | |
2052 | bdrv_append(bs_snapshot, bs); | |
2053 | ||
2054 | g_free(tmp_filename); | |
2055 | return bs_snapshot; | |
2056 | ||
2057 | out: | |
2058 | QDECREF(snapshot_options); | |
2059 | g_free(tmp_filename); | |
2060 | return NULL; | |
2061 | } | |
2062 | ||
2063 | /* | |
2064 | * Opens a disk image (raw, qcow2, vmdk, ...) | |
2065 | * | |
2066 | * options is a QDict of options to pass to the block drivers, or NULL for an | |
2067 | * empty set of options. The reference to the QDict belongs to the block layer | |
2068 | * after the call (even on failure), so if the caller intends to reuse the | |
2069 | * dictionary, it needs to use QINCREF() before calling bdrv_open. | |
2070 | * | |
2071 | * If *pbs is NULL, a new BDS will be created with a pointer to it stored there. | |
2072 | * If it is not NULL, the referenced BDS will be reused. | |
2073 | * | |
2074 | * The reference parameter may be used to specify an existing block device which | |
2075 | * should be opened. If specified, neither options nor a filename may be given, | |
2076 | * nor can an existing BDS be reused (that is, *pbs has to be NULL). | |
2077 | */ | |
2078 | static BlockDriverState *bdrv_open_inherit(const char *filename, | |
2079 | const char *reference, | |
2080 | QDict *options, int flags, | |
2081 | BlockDriverState *parent, | |
2082 | const BdrvChildRole *child_role, | |
2083 | Error **errp) | |
2084 | { | |
2085 | int ret; | |
2086 | BlockBackend *file = NULL; | |
2087 | BlockDriverState *bs; | |
2088 | BlockDriver *drv = NULL; | |
2089 | const char *drvname; | |
2090 | const char *backing; | |
2091 | Error *local_err = NULL; | |
2092 | QDict *snapshot_options = NULL; | |
2093 | int snapshot_flags = 0; | |
2094 | ||
2095 | assert(!child_role || !flags); | |
2096 | assert(!child_role == !parent); | |
2097 | ||
2098 | if (reference) { | |
2099 | bool options_non_empty = options ? qdict_size(options) : false; | |
2100 | QDECREF(options); | |
2101 | ||
2102 | if (filename || options_non_empty) { | |
2103 | error_setg(errp, "Cannot reference an existing block device with " | |
2104 | "additional options or a new filename"); | |
2105 | return NULL; | |
2106 | } | |
2107 | ||
2108 | bs = bdrv_lookup_bs(reference, reference, errp); | |
2109 | if (!bs) { | |
2110 | return NULL; | |
2111 | } | |
2112 | ||
2113 | bdrv_ref(bs); | |
2114 | return bs; | |
2115 | } | |
2116 | ||
2117 | bs = bdrv_new(); | |
2118 | ||
2119 | /* NULL means an empty set of options */ | |
2120 | if (options == NULL) { | |
2121 | options = qdict_new(); | |
2122 | } | |
2123 | ||
2124 | /* json: syntax counts as explicit options, as if in the QDict */ | |
2125 | parse_json_protocol(options, &filename, &local_err); | |
2126 | if (local_err) { | |
2127 | goto fail; | |
2128 | } | |
2129 | ||
2130 | bs->explicit_options = qdict_clone_shallow(options); | |
2131 | ||
2132 | if (child_role) { | |
2133 | bs->inherits_from = parent; | |
2134 | child_role->inherit_options(&flags, options, | |
2135 | parent->open_flags, parent->options); | |
2136 | } | |
2137 | ||
2138 | ret = bdrv_fill_options(&options, filename, &flags, &local_err); | |
2139 | if (local_err) { | |
2140 | goto fail; | |
2141 | } | |
2142 | ||
2143 | /* Set the BDRV_O_RDWR and BDRV_O_ALLOW_RDWR flags. | |
2144 | * FIXME: we're parsing the QDict to avoid having to create a | |
2145 | * QemuOpts just for this, but neither option is optimal. */ | |
2146 | if (g_strcmp0(qdict_get_try_str(options, BDRV_OPT_READ_ONLY), "on") && | |
2147 | !qdict_get_try_bool(options, BDRV_OPT_READ_ONLY, false)) { | |
2148 | flags |= (BDRV_O_RDWR | BDRV_O_ALLOW_RDWR); | |
2149 | } else { | |
2150 | flags &= ~BDRV_O_RDWR; | |
2151 | } | |
2152 | ||
2153 | if (flags & BDRV_O_SNAPSHOT) { | |
2154 | snapshot_options = qdict_new(); | |
2155 | bdrv_temp_snapshot_options(&snapshot_flags, snapshot_options, | |
2156 | flags, options); | |
2157 | /* Let bdrv_backing_options() override "read-only" */ | |
2158 | qdict_del(options, BDRV_OPT_READ_ONLY); | |
2159 | bdrv_backing_options(&flags, options, flags, options); | |
2160 | } | |
2161 | ||
2162 | bs->open_flags = flags; | |
2163 | bs->options = options; | |
2164 | options = qdict_clone_shallow(options); | |
2165 | ||
2166 | /* Find the right image format driver */ | |
2167 | drvname = qdict_get_try_str(options, "driver"); | |
2168 | if (drvname) { | |
2169 | drv = bdrv_find_format(drvname); | |
2170 | if (!drv) { | |
2171 | error_setg(errp, "Unknown driver: '%s'", drvname); | |
2172 | goto fail; | |
2173 | } | |
2174 | } | |
2175 | ||
2176 | assert(drvname || !(flags & BDRV_O_PROTOCOL)); | |
2177 | ||
2178 | backing = qdict_get_try_str(options, "backing"); | |
2179 | if (backing && *backing == '\0') { | |
2180 | flags |= BDRV_O_NO_BACKING; | |
2181 | qdict_del(options, "backing"); | |
2182 | } | |
2183 | ||
2184 | /* Open image file without format layer. This BlockBackend is only used for | |
2185 | * probing, the block drivers will do their own bdrv_open_child() for the | |
2186 | * same BDS, which is why we put the node name back into options. */ | |
2187 | if ((flags & BDRV_O_PROTOCOL) == 0) { | |
2188 | BlockDriverState *file_bs; | |
2189 | ||
2190 | file_bs = bdrv_open_child_bs(filename, options, "file", bs, | |
2191 | &child_file, true, &local_err); | |
2192 | if (local_err) { | |
2193 | goto fail; | |
2194 | } | |
2195 | if (file_bs != NULL) { | |
2196 | file = blk_new(BLK_PERM_CONSISTENT_READ, BLK_PERM_ALL); | |
2197 | blk_insert_bs(file, file_bs, &local_err); | |
2198 | bdrv_unref(file_bs); | |
2199 | if (local_err) { | |
2200 | goto fail; | |
2201 | } | |
2202 | ||
2203 | qdict_put(options, "file", | |
2204 | qstring_from_str(bdrv_get_node_name(file_bs))); | |
2205 | } | |
2206 | } | |
2207 | ||
2208 | /* Image format probing */ | |
2209 | bs->probed = !drv; | |
2210 | if (!drv && file) { | |
2211 | ret = find_image_format(file, filename, &drv, &local_err); | |
2212 | if (ret < 0) { | |
2213 | goto fail; | |
2214 | } | |
2215 | /* | |
2216 | * This option update would logically belong in bdrv_fill_options(), | |
2217 | * but we first need to open bs->file for the probing to work, while | |
2218 | * opening bs->file already requires the (mostly) final set of options | |
2219 | * so that cache mode etc. can be inherited. | |
2220 | * | |
2221 | * Adding the driver later is somewhat ugly, but it's not an option | |
2222 | * that would ever be inherited, so it's correct. We just need to make | |
2223 | * sure to update both bs->options (which has the full effective | |
2224 | * options for bs) and options (which has file.* already removed). | |
2225 | */ | |
2226 | qdict_put(bs->options, "driver", qstring_from_str(drv->format_name)); | |
2227 | qdict_put(options, "driver", qstring_from_str(drv->format_name)); | |
2228 | } else if (!drv) { | |
2229 | error_setg(errp, "Must specify either driver or file"); | |
2230 | goto fail; | |
2231 | } | |
2232 | ||
2233 | /* BDRV_O_PROTOCOL must be set iff a protocol BDS is about to be created */ | |
2234 | assert(!!(flags & BDRV_O_PROTOCOL) == !!drv->bdrv_file_open); | |
2235 | /* file must be NULL if a protocol BDS is about to be created | |
2236 | * (the inverse results in an error message from bdrv_open_common()) */ | |
2237 | assert(!(flags & BDRV_O_PROTOCOL) || !file); | |
2238 | ||
2239 | /* Open the image */ | |
2240 | ret = bdrv_open_common(bs, file, options, &local_err); | |
2241 | if (ret < 0) { | |
2242 | goto fail; | |
2243 | } | |
2244 | ||
2245 | if (file) { | |
2246 | blk_unref(file); | |
2247 | file = NULL; | |
2248 | } | |
2249 | ||
2250 | /* If there is a backing file, use it */ | |
2251 | if ((flags & BDRV_O_NO_BACKING) == 0) { | |
2252 | ret = bdrv_open_backing_file(bs, options, "backing", &local_err); | |
2253 | if (ret < 0) { | |
2254 | goto close_and_fail; | |
2255 | } | |
2256 | } | |
2257 | ||
2258 | bdrv_refresh_filename(bs); | |
2259 | ||
2260 | /* Check if any unknown options were used */ | |
2261 | if (qdict_size(options) != 0) { | |
2262 | const QDictEntry *entry = qdict_first(options); | |
2263 | if (flags & BDRV_O_PROTOCOL) { | |
2264 | error_setg(errp, "Block protocol '%s' doesn't support the option " | |
2265 | "'%s'", drv->format_name, entry->key); | |
2266 | } else { | |
2267 | error_setg(errp, | |
2268 | "Block format '%s' does not support the option '%s'", | |
2269 | drv->format_name, entry->key); | |
2270 | } | |
2271 | ||
2272 | goto close_and_fail; | |
2273 | } | |
2274 | ||
2275 | if (!bdrv_key_required(bs)) { | |
2276 | bdrv_parent_cb_change_media(bs, true); | |
2277 | } else if (!runstate_check(RUN_STATE_PRELAUNCH) | |
2278 | && !runstate_check(RUN_STATE_INMIGRATE) | |
2279 | && !runstate_check(RUN_STATE_PAUSED)) { /* HACK */ | |
2280 | error_setg(errp, | |
2281 | "Guest must be stopped for opening of encrypted image"); | |
2282 | goto close_and_fail; | |
2283 | } | |
2284 | ||
2285 | QDECREF(options); | |
2286 | ||
2287 | /* For snapshot=on, create a temporary qcow2 overlay. bs points to the | |
2288 | * temporary snapshot afterwards. */ | |
2289 | if (snapshot_flags) { | |
2290 | BlockDriverState *snapshot_bs; | |
2291 | snapshot_bs = bdrv_append_temp_snapshot(bs, snapshot_flags, | |
2292 | snapshot_options, &local_err); | |
2293 | snapshot_options = NULL; | |
2294 | if (local_err) { | |
2295 | goto close_and_fail; | |
2296 | } | |
2297 | /* We are not going to return bs but the overlay on top of it | |
2298 | * (snapshot_bs); thus, we have to drop the strong reference to bs | |
2299 | * (which we obtained by calling bdrv_new()). bs will not be deleted, | |
2300 | * though, because the overlay still has a reference to it. */ | |
2301 | bdrv_unref(bs); | |
2302 | bs = snapshot_bs; | |
2303 | } | |
2304 | ||
2305 | return bs; | |
2306 | ||
2307 | fail: | |
2308 | blk_unref(file); | |
2309 | if (bs->file != NULL) { | |
2310 | bdrv_unref_child(bs, bs->file); | |
2311 | } | |
2312 | QDECREF(snapshot_options); | |
2313 | QDECREF(bs->explicit_options); | |
2314 | QDECREF(bs->options); | |
2315 | QDECREF(options); | |
2316 | bs->options = NULL; | |
2317 | bdrv_unref(bs); | |
2318 | error_propagate(errp, local_err); | |
2319 | return NULL; | |
2320 | ||
2321 | close_and_fail: | |
2322 | bdrv_unref(bs); | |
2323 | QDECREF(snapshot_options); | |
2324 | QDECREF(options); | |
2325 | error_propagate(errp, local_err); | |
2326 | return NULL; | |
2327 | } | |
2328 | ||
2329 | BlockDriverState *bdrv_open(const char *filename, const char *reference, | |
2330 | QDict *options, int flags, Error **errp) | |
2331 | { | |
2332 | return bdrv_open_inherit(filename, reference, options, flags, NULL, | |
2333 | NULL, errp); | |
2334 | } | |
2335 | ||
2336 | typedef struct BlockReopenQueueEntry { | |
2337 | bool prepared; | |
2338 | BDRVReopenState state; | |
2339 | QSIMPLEQ_ENTRY(BlockReopenQueueEntry) entry; | |
2340 | } BlockReopenQueueEntry; | |
2341 | ||
2342 | /* | |
2343 | * Adds a BlockDriverState to a simple queue for an atomic, transactional | |
2344 | * reopen of multiple devices. | |
2345 | * | |
2346 | * bs_queue can either be an existing BlockReopenQueue that has had QSIMPLE_INIT | |
2347 | * already performed, or alternatively may be NULL a new BlockReopenQueue will | |
2348 | * be created and initialized. This newly created BlockReopenQueue should be | |
2349 | * passed back in for subsequent calls that are intended to be of the same | |
2350 | * atomic 'set'. | |
2351 | * | |
2352 | * bs is the BlockDriverState to add to the reopen queue. | |
2353 | * | |
2354 | * options contains the changed options for the associated bs | |
2355 | * (the BlockReopenQueue takes ownership) | |
2356 | * | |
2357 | * flags contains the open flags for the associated bs | |
2358 | * | |
2359 | * returns a pointer to bs_queue, which is either the newly allocated | |
2360 | * bs_queue, or the existing bs_queue being used. | |
2361 | * | |
2362 | */ | |
2363 | static BlockReopenQueue *bdrv_reopen_queue_child(BlockReopenQueue *bs_queue, | |
2364 | BlockDriverState *bs, | |
2365 | QDict *options, | |
2366 | int flags, | |
2367 | const BdrvChildRole *role, | |
2368 | QDict *parent_options, | |
2369 | int parent_flags) | |
2370 | { | |
2371 | assert(bs != NULL); | |
2372 | ||
2373 | BlockReopenQueueEntry *bs_entry; | |
2374 | BdrvChild *child; | |
2375 | QDict *old_options, *explicit_options; | |
2376 | ||
2377 | if (bs_queue == NULL) { | |
2378 | bs_queue = g_new0(BlockReopenQueue, 1); | |
2379 | QSIMPLEQ_INIT(bs_queue); | |
2380 | } | |
2381 | ||
2382 | if (!options) { | |
2383 | options = qdict_new(); | |
2384 | } | |
2385 | ||
2386 | /* Check if this BlockDriverState is already in the queue */ | |
2387 | QSIMPLEQ_FOREACH(bs_entry, bs_queue, entry) { | |
2388 | if (bs == bs_entry->state.bs) { | |
2389 | break; | |
2390 | } | |
2391 | } | |
2392 | ||
2393 | /* | |
2394 | * Precedence of options: | |
2395 | * 1. Explicitly passed in options (highest) | |
2396 | * 2. Set in flags (only for top level) | |
2397 | * 3. Retained from explicitly set options of bs | |
2398 | * 4. Inherited from parent node | |
2399 | * 5. Retained from effective options of bs | |
2400 | */ | |
2401 | ||
2402 | if (!parent_options) { | |
2403 | /* | |
2404 | * Any setting represented by flags is always updated. If the | |
2405 | * corresponding QDict option is set, it takes precedence. Otherwise | |
2406 | * the flag is translated into a QDict option. The old setting of bs is | |
2407 | * not considered. | |
2408 | */ | |
2409 | update_options_from_flags(options, flags); | |
2410 | } | |
2411 | ||
2412 | /* Old explicitly set values (don't overwrite by inherited value) */ | |
2413 | if (bs_entry) { | |
2414 | old_options = qdict_clone_shallow(bs_entry->state.explicit_options); | |
2415 | } else { | |
2416 | old_options = qdict_clone_shallow(bs->explicit_options); | |
2417 | } | |
2418 | bdrv_join_options(bs, options, old_options); | |
2419 | QDECREF(old_options); | |
2420 | ||
2421 | explicit_options = qdict_clone_shallow(options); | |
2422 | ||
2423 | /* Inherit from parent node */ | |
2424 | if (parent_options) { | |
2425 | assert(!flags); | |
2426 | role->inherit_options(&flags, options, parent_flags, parent_options); | |
2427 | } | |
2428 | ||
2429 | /* Old values are used for options that aren't set yet */ | |
2430 | old_options = qdict_clone_shallow(bs->options); | |
2431 | bdrv_join_options(bs, options, old_options); | |
2432 | QDECREF(old_options); | |
2433 | ||
2434 | /* bdrv_open() masks this flag out */ | |
2435 | flags &= ~BDRV_O_PROTOCOL; | |
2436 | ||
2437 | QLIST_FOREACH(child, &bs->children, next) { | |
2438 | QDict *new_child_options; | |
2439 | char *child_key_dot; | |
2440 | ||
2441 | /* reopen can only change the options of block devices that were | |
2442 | * implicitly created and inherited options. For other (referenced) | |
2443 | * block devices, a syntax like "backing.foo" results in an error. */ | |
2444 | if (child->bs->inherits_from != bs) { | |
2445 | continue; | |
2446 | } | |
2447 | ||
2448 | child_key_dot = g_strdup_printf("%s.", child->name); | |
2449 | qdict_extract_subqdict(options, &new_child_options, child_key_dot); | |
2450 | g_free(child_key_dot); | |
2451 | ||
2452 | bdrv_reopen_queue_child(bs_queue, child->bs, new_child_options, 0, | |
2453 | child->role, options, flags); | |
2454 | } | |
2455 | ||
2456 | if (!bs_entry) { | |
2457 | bs_entry = g_new0(BlockReopenQueueEntry, 1); | |
2458 | QSIMPLEQ_INSERT_TAIL(bs_queue, bs_entry, entry); | |
2459 | } else { | |
2460 | QDECREF(bs_entry->state.options); | |
2461 | QDECREF(bs_entry->state.explicit_options); | |
2462 | } | |
2463 | ||
2464 | bs_entry->state.bs = bs; | |
2465 | bs_entry->state.options = options; | |
2466 | bs_entry->state.explicit_options = explicit_options; | |
2467 | bs_entry->state.flags = flags; | |
2468 | ||
2469 | return bs_queue; | |
2470 | } | |
2471 | ||
2472 | BlockReopenQueue *bdrv_reopen_queue(BlockReopenQueue *bs_queue, | |
2473 | BlockDriverState *bs, | |
2474 | QDict *options, int flags) | |
2475 | { | |
2476 | return bdrv_reopen_queue_child(bs_queue, bs, options, flags, | |
2477 | NULL, NULL, 0); | |
2478 | } | |
2479 | ||
2480 | /* | |
2481 | * Reopen multiple BlockDriverStates atomically & transactionally. | |
2482 | * | |
2483 | * The queue passed in (bs_queue) must have been built up previous | |
2484 | * via bdrv_reopen_queue(). | |
2485 | * | |
2486 | * Reopens all BDS specified in the queue, with the appropriate | |
2487 | * flags. All devices are prepared for reopen, and failure of any | |
2488 | * device will cause all device changes to be abandonded, and intermediate | |
2489 | * data cleaned up. | |
2490 | * | |
2491 | * If all devices prepare successfully, then the changes are committed | |
2492 | * to all devices. | |
2493 | * | |
2494 | */ | |
2495 | int bdrv_reopen_multiple(AioContext *ctx, BlockReopenQueue *bs_queue, Error **errp) | |
2496 | { | |
2497 | int ret = -1; | |
2498 | BlockReopenQueueEntry *bs_entry, *next; | |
2499 | Error *local_err = NULL; | |
2500 | ||
2501 | assert(bs_queue != NULL); | |
2502 | ||
2503 | aio_context_release(ctx); | |
2504 | bdrv_drain_all_begin(); | |
2505 | aio_context_acquire(ctx); | |
2506 | ||
2507 | QSIMPLEQ_FOREACH(bs_entry, bs_queue, entry) { | |
2508 | if (bdrv_reopen_prepare(&bs_entry->state, bs_queue, &local_err)) { | |
2509 | error_propagate(errp, local_err); | |
2510 | goto cleanup; | |
2511 | } | |
2512 | bs_entry->prepared = true; | |
2513 | } | |
2514 | ||
2515 | /* If we reach this point, we have success and just need to apply the | |
2516 | * changes | |
2517 | */ | |
2518 | QSIMPLEQ_FOREACH(bs_entry, bs_queue, entry) { | |
2519 | bdrv_reopen_commit(&bs_entry->state); | |
2520 | } | |
2521 | ||
2522 | ret = 0; | |
2523 | ||
2524 | cleanup: | |
2525 | QSIMPLEQ_FOREACH_SAFE(bs_entry, bs_queue, entry, next) { | |
2526 | if (ret && bs_entry->prepared) { | |
2527 | bdrv_reopen_abort(&bs_entry->state); | |
2528 | } else if (ret) { | |
2529 | QDECREF(bs_entry->state.explicit_options); | |
2530 | } | |
2531 | QDECREF(bs_entry->state.options); | |
2532 | g_free(bs_entry); | |
2533 | } | |
2534 | g_free(bs_queue); | |
2535 | ||
2536 | bdrv_drain_all_end(); | |
2537 | ||
2538 | return ret; | |
2539 | } | |
2540 | ||
2541 | ||
2542 | /* Reopen a single BlockDriverState with the specified flags. */ | |
2543 | int bdrv_reopen(BlockDriverState *bs, int bdrv_flags, Error **errp) | |
2544 | { | |
2545 | int ret = -1; | |
2546 | Error *local_err = NULL; | |
2547 | BlockReopenQueue *queue = bdrv_reopen_queue(NULL, bs, NULL, bdrv_flags); | |
2548 | ||
2549 | ret = bdrv_reopen_multiple(bdrv_get_aio_context(bs), queue, &local_err); | |
2550 | if (local_err != NULL) { | |
2551 | error_propagate(errp, local_err); | |
2552 | } | |
2553 | return ret; | |
2554 | } | |
2555 | ||
2556 | ||
2557 | /* | |
2558 | * Prepares a BlockDriverState for reopen. All changes are staged in the | |
2559 | * 'opaque' field of the BDRVReopenState, which is used and allocated by | |
2560 | * the block driver layer .bdrv_reopen_prepare() | |
2561 | * | |
2562 | * bs is the BlockDriverState to reopen | |
2563 | * flags are the new open flags | |
2564 | * queue is the reopen queue | |
2565 | * | |
2566 | * Returns 0 on success, non-zero on error. On error errp will be set | |
2567 | * as well. | |
2568 | * | |
2569 | * On failure, bdrv_reopen_abort() will be called to clean up any data. | |
2570 | * It is the responsibility of the caller to then call the abort() or | |
2571 | * commit() for any other BDS that have been left in a prepare() state | |
2572 | * | |
2573 | */ | |
2574 | int bdrv_reopen_prepare(BDRVReopenState *reopen_state, BlockReopenQueue *queue, | |
2575 | Error **errp) | |
2576 | { | |
2577 | int ret = -1; | |
2578 | Error *local_err = NULL; | |
2579 | BlockDriver *drv; | |
2580 | QemuOpts *opts; | |
2581 | const char *value; | |
2582 | ||
2583 | assert(reopen_state != NULL); | |
2584 | assert(reopen_state->bs->drv != NULL); | |
2585 | drv = reopen_state->bs->drv; | |
2586 | ||
2587 | /* Process generic block layer options */ | |
2588 | opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort); | |
2589 | qemu_opts_absorb_qdict(opts, reopen_state->options, &local_err); | |
2590 | if (local_err) { | |
2591 | error_propagate(errp, local_err); | |
2592 | ret = -EINVAL; | |
2593 | goto error; | |
2594 | } | |
2595 | ||
2596 | update_flags_from_options(&reopen_state->flags, opts); | |
2597 | ||
2598 | /* node-name and driver must be unchanged. Put them back into the QDict, so | |
2599 | * that they are checked at the end of this function. */ | |
2600 | value = qemu_opt_get(opts, "node-name"); | |
2601 | if (value) { | |
2602 | qdict_put(reopen_state->options, "node-name", qstring_from_str(value)); | |
2603 | } | |
2604 | ||
2605 | value = qemu_opt_get(opts, "driver"); | |
2606 | if (value) { | |
2607 | qdict_put(reopen_state->options, "driver", qstring_from_str(value)); | |
2608 | } | |
2609 | ||
2610 | /* if we are to stay read-only, do not allow permission change | |
2611 | * to r/w */ | |
2612 | if (!(reopen_state->bs->open_flags & BDRV_O_ALLOW_RDWR) && | |
2613 | reopen_state->flags & BDRV_O_RDWR) { | |
2614 | error_setg(errp, "Node '%s' is read only", | |
2615 | bdrv_get_device_or_node_name(reopen_state->bs)); | |
2616 | goto error; | |
2617 | } | |
2618 | ||
2619 | ||
2620 | ret = bdrv_flush(reopen_state->bs); | |
2621 | if (ret) { | |
2622 | error_setg_errno(errp, -ret, "Error flushing drive"); | |
2623 | goto error; | |
2624 | } | |
2625 | ||
2626 | if (drv->bdrv_reopen_prepare) { | |
2627 | ret = drv->bdrv_reopen_prepare(reopen_state, queue, &local_err); | |
2628 | if (ret) { | |
2629 | if (local_err != NULL) { | |
2630 | error_propagate(errp, local_err); | |
2631 | } else { | |
2632 | error_setg(errp, "failed while preparing to reopen image '%s'", | |
2633 | reopen_state->bs->filename); | |
2634 | } | |
2635 | goto error; | |
2636 | } | |
2637 | } else { | |
2638 | /* It is currently mandatory to have a bdrv_reopen_prepare() | |
2639 | * handler for each supported drv. */ | |
2640 | error_setg(errp, "Block format '%s' used by node '%s' " | |
2641 | "does not support reopening files", drv->format_name, | |
2642 | bdrv_get_device_or_node_name(reopen_state->bs)); | |
2643 | ret = -1; | |
2644 | goto error; | |
2645 | } | |
2646 | ||
2647 | /* Options that are not handled are only okay if they are unchanged | |
2648 | * compared to the old state. It is expected that some options are only | |
2649 | * used for the initial open, but not reopen (e.g. filename) */ | |
2650 | if (qdict_size(reopen_state->options)) { | |
2651 | const QDictEntry *entry = qdict_first(reopen_state->options); | |
2652 | ||
2653 | do { | |
2654 | QString *new_obj = qobject_to_qstring(entry->value); | |
2655 | const char *new = qstring_get_str(new_obj); | |
2656 | const char *old = qdict_get_try_str(reopen_state->bs->options, | |
2657 | entry->key); | |
2658 | ||
2659 | if (!old || strcmp(new, old)) { | |
2660 | error_setg(errp, "Cannot change the option '%s'", entry->key); | |
2661 | ret = -EINVAL; | |
2662 | goto error; | |
2663 | } | |
2664 | } while ((entry = qdict_next(reopen_state->options, entry))); | |
2665 | } | |
2666 | ||
2667 | ret = 0; | |
2668 | ||
2669 | error: | |
2670 | qemu_opts_del(opts); | |
2671 | return ret; | |
2672 | } | |
2673 | ||
2674 | /* | |
2675 | * Takes the staged changes for the reopen from bdrv_reopen_prepare(), and | |
2676 | * makes them final by swapping the staging BlockDriverState contents into | |
2677 | * the active BlockDriverState contents. | |
2678 | */ | |
2679 | void bdrv_reopen_commit(BDRVReopenState *reopen_state) | |
2680 | { | |
2681 | BlockDriver *drv; | |
2682 | ||
2683 | assert(reopen_state != NULL); | |
2684 | drv = reopen_state->bs->drv; | |
2685 | assert(drv != NULL); | |
2686 | ||
2687 | /* If there are any driver level actions to take */ | |
2688 | if (drv->bdrv_reopen_commit) { | |
2689 | drv->bdrv_reopen_commit(reopen_state); | |
2690 | } | |
2691 | ||
2692 | /* set BDS specific flags now */ | |
2693 | QDECREF(reopen_state->bs->explicit_options); | |
2694 | ||
2695 | reopen_state->bs->explicit_options = reopen_state->explicit_options; | |
2696 | reopen_state->bs->open_flags = reopen_state->flags; | |
2697 | reopen_state->bs->read_only = !(reopen_state->flags & BDRV_O_RDWR); | |
2698 | ||
2699 | bdrv_refresh_limits(reopen_state->bs, NULL); | |
2700 | } | |
2701 | ||
2702 | /* | |
2703 | * Abort the reopen, and delete and free the staged changes in | |
2704 | * reopen_state | |
2705 | */ | |
2706 | void bdrv_reopen_abort(BDRVReopenState *reopen_state) | |
2707 | { | |
2708 | BlockDriver *drv; | |
2709 | ||
2710 | assert(reopen_state != NULL); | |
2711 | drv = reopen_state->bs->drv; | |
2712 | assert(drv != NULL); | |
2713 | ||
2714 | if (drv->bdrv_reopen_abort) { | |
2715 | drv->bdrv_reopen_abort(reopen_state); | |
2716 | } | |
2717 | ||
2718 | QDECREF(reopen_state->explicit_options); | |
2719 | } | |
2720 | ||
2721 | ||
2722 | static void bdrv_close(BlockDriverState *bs) | |
2723 | { | |
2724 | BdrvAioNotifier *ban, *ban_next; | |
2725 | ||
2726 | assert(!bs->job); | |
2727 | assert(!bs->refcnt); | |
2728 | ||
2729 | bdrv_drained_begin(bs); /* complete I/O */ | |
2730 | bdrv_flush(bs); | |
2731 | bdrv_drain(bs); /* in case flush left pending I/O */ | |
2732 | ||
2733 | bdrv_release_named_dirty_bitmaps(bs); | |
2734 | assert(QLIST_EMPTY(&bs->dirty_bitmaps)); | |
2735 | ||
2736 | if (bs->drv) { | |
2737 | BdrvChild *child, *next; | |
2738 | ||
2739 | bs->drv->bdrv_close(bs); | |
2740 | bs->drv = NULL; | |
2741 | ||
2742 | bdrv_set_backing_hd(bs, NULL); | |
2743 | ||
2744 | if (bs->file != NULL) { | |
2745 | bdrv_unref_child(bs, bs->file); | |
2746 | bs->file = NULL; | |
2747 | } | |
2748 | ||
2749 | QLIST_FOREACH_SAFE(child, &bs->children, next, next) { | |
2750 | /* TODO Remove bdrv_unref() from drivers' close function and use | |
2751 | * bdrv_unref_child() here */ | |
2752 | if (child->bs->inherits_from == bs) { | |
2753 | child->bs->inherits_from = NULL; | |
2754 | } | |
2755 | bdrv_detach_child(child); | |
2756 | } | |
2757 | ||
2758 | g_free(bs->opaque); | |
2759 | bs->opaque = NULL; | |
2760 | bs->copy_on_read = 0; | |
2761 | bs->backing_file[0] = '\0'; | |
2762 | bs->backing_format[0] = '\0'; | |
2763 | bs->total_sectors = 0; | |
2764 | bs->encrypted = false; | |
2765 | bs->valid_key = false; | |
2766 | bs->sg = false; | |
2767 | QDECREF(bs->options); | |
2768 | QDECREF(bs->explicit_options); | |
2769 | bs->options = NULL; | |
2770 | QDECREF(bs->full_open_options); | |
2771 | bs->full_open_options = NULL; | |
2772 | } | |
2773 | ||
2774 | QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_next) { | |
2775 | g_free(ban); | |
2776 | } | |
2777 | QLIST_INIT(&bs->aio_notifiers); | |
2778 | bdrv_drained_end(bs); | |
2779 | } | |
2780 | ||
2781 | void bdrv_close_all(void) | |
2782 | { | |
2783 | block_job_cancel_sync_all(); | |
2784 | nbd_export_close_all(); | |
2785 | ||
2786 | /* Drop references from requests still in flight, such as canceled block | |
2787 | * jobs whose AIO context has not been polled yet */ | |
2788 | bdrv_drain_all(); | |
2789 | ||
2790 | blk_remove_all_bs(); | |
2791 | blockdev_close_all_bdrv_states(); | |
2792 | ||
2793 | assert(QTAILQ_EMPTY(&all_bdrv_states)); | |
2794 | } | |
2795 | ||
2796 | static void change_parent_backing_link(BlockDriverState *from, | |
2797 | BlockDriverState *to) | |
2798 | { | |
2799 | BdrvChild *c, *next, *to_c; | |
2800 | ||
2801 | QLIST_FOREACH_SAFE(c, &from->parents, next_parent, next) { | |
2802 | if (c->role == &child_backing) { | |
2803 | /* @from is generally not allowed to be a backing file, except for | |
2804 | * when @to is the overlay. In that case, @from may not be replaced | |
2805 | * by @to as @to's backing node. */ | |
2806 | QLIST_FOREACH(to_c, &to->children, next) { | |
2807 | if (to_c == c) { | |
2808 | break; | |
2809 | } | |
2810 | } | |
2811 | if (to_c) { | |
2812 | continue; | |
2813 | } | |
2814 | } | |
2815 | ||
2816 | assert(c->role != &child_backing); | |
2817 | bdrv_ref(to); | |
2818 | /* FIXME Are we sure that bdrv_replace_child() can't run into | |
2819 | * &error_abort because of permissions? */ | |
2820 | bdrv_replace_child(c, to, true); | |
2821 | bdrv_unref(from); | |
2822 | } | |
2823 | } | |
2824 | ||
2825 | /* | |
2826 | * Add new bs contents at the top of an image chain while the chain is | |
2827 | * live, while keeping required fields on the top layer. | |
2828 | * | |
2829 | * This will modify the BlockDriverState fields, and swap contents | |
2830 | * between bs_new and bs_top. Both bs_new and bs_top are modified. | |
2831 | * | |
2832 | * bs_new must not be attached to a BlockBackend. | |
2833 | * | |
2834 | * This function does not create any image files. | |
2835 | * | |
2836 | * bdrv_append() takes ownership of a bs_new reference and unrefs it because | |
2837 | * that's what the callers commonly need. bs_new will be referenced by the old | |
2838 | * parents of bs_top after bdrv_append() returns. If the caller needs to keep a | |
2839 | * reference of its own, it must call bdrv_ref(). | |
2840 | */ | |
2841 | void bdrv_append(BlockDriverState *bs_new, BlockDriverState *bs_top) | |
2842 | { | |
2843 | assert(!bdrv_requests_pending(bs_top)); | |
2844 | assert(!bdrv_requests_pending(bs_new)); | |
2845 | ||
2846 | bdrv_ref(bs_top); | |
2847 | ||
2848 | change_parent_backing_link(bs_top, bs_new); | |
2849 | bdrv_set_backing_hd(bs_new, bs_top); | |
2850 | bdrv_unref(bs_top); | |
2851 | ||
2852 | /* bs_new is now referenced by its new parents, we don't need the | |
2853 | * additional reference any more. */ | |
2854 | bdrv_unref(bs_new); | |
2855 | } | |
2856 | ||
2857 | void bdrv_replace_in_backing_chain(BlockDriverState *old, BlockDriverState *new) | |
2858 | { | |
2859 | assert(!bdrv_requests_pending(old)); | |
2860 | assert(!bdrv_requests_pending(new)); | |
2861 | ||
2862 | bdrv_ref(old); | |
2863 | ||
2864 | change_parent_backing_link(old, new); | |
2865 | ||
2866 | bdrv_unref(old); | |
2867 | } | |
2868 | ||
2869 | static void bdrv_delete(BlockDriverState *bs) | |
2870 | { | |
2871 | assert(!bs->job); | |
2872 | assert(bdrv_op_blocker_is_empty(bs)); | |
2873 | assert(!bs->refcnt); | |
2874 | ||
2875 | bdrv_close(bs); | |
2876 | ||
2877 | /* remove from list, if necessary */ | |
2878 | if (bs->node_name[0] != '\0') { | |
2879 | QTAILQ_REMOVE(&graph_bdrv_states, bs, node_list); | |
2880 | } | |
2881 | QTAILQ_REMOVE(&all_bdrv_states, bs, bs_list); | |
2882 | ||
2883 | g_free(bs); | |
2884 | } | |
2885 | ||
2886 | /* | |
2887 | * Run consistency checks on an image | |
2888 | * | |
2889 | * Returns 0 if the check could be completed (it doesn't mean that the image is | |
2890 | * free of errors) or -errno when an internal error occurred. The results of the | |
2891 | * check are stored in res. | |
2892 | */ | |
2893 | int bdrv_check(BlockDriverState *bs, BdrvCheckResult *res, BdrvCheckMode fix) | |
2894 | { | |
2895 | if (bs->drv == NULL) { | |
2896 | return -ENOMEDIUM; | |
2897 | } | |
2898 | if (bs->drv->bdrv_check == NULL) { | |
2899 | return -ENOTSUP; | |
2900 | } | |
2901 | ||
2902 | memset(res, 0, sizeof(*res)); | |
2903 | return bs->drv->bdrv_check(bs, res, fix); | |
2904 | } | |
2905 | ||
2906 | /* | |
2907 | * Return values: | |
2908 | * 0 - success | |
2909 | * -EINVAL - backing format specified, but no file | |
2910 | * -ENOSPC - can't update the backing file because no space is left in the | |
2911 | * image file header | |
2912 | * -ENOTSUP - format driver doesn't support changing the backing file | |
2913 | */ | |
2914 | int bdrv_change_backing_file(BlockDriverState *bs, | |
2915 | const char *backing_file, const char *backing_fmt) | |
2916 | { | |
2917 | BlockDriver *drv = bs->drv; | |
2918 | int ret; | |
2919 | ||
2920 | /* Backing file format doesn't make sense without a backing file */ | |
2921 | if (backing_fmt && !backing_file) { | |
2922 | return -EINVAL; | |
2923 | } | |
2924 | ||
2925 | if (drv->bdrv_change_backing_file != NULL) { | |
2926 | ret = drv->bdrv_change_backing_file(bs, backing_file, backing_fmt); | |
2927 | } else { | |
2928 | ret = -ENOTSUP; | |
2929 | } | |
2930 | ||
2931 | if (ret == 0) { | |
2932 | pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_file ?: ""); | |
2933 | pstrcpy(bs->backing_format, sizeof(bs->backing_format), backing_fmt ?: ""); | |
2934 | } | |
2935 | return ret; | |
2936 | } | |
2937 | ||
2938 | /* | |
2939 | * Finds the image layer in the chain that has 'bs' as its backing file. | |
2940 | * | |
2941 | * active is the current topmost image. | |
2942 | * | |
2943 | * Returns NULL if bs is not found in active's image chain, | |
2944 | * or if active == bs. | |
2945 | * | |
2946 | * Returns the bottommost base image if bs == NULL. | |
2947 | */ | |
2948 | BlockDriverState *bdrv_find_overlay(BlockDriverState *active, | |
2949 | BlockDriverState *bs) | |
2950 | { | |
2951 | while (active && bs != backing_bs(active)) { | |
2952 | active = backing_bs(active); | |
2953 | } | |
2954 | ||
2955 | return active; | |
2956 | } | |
2957 | ||
2958 | /* Given a BDS, searches for the base layer. */ | |
2959 | BlockDriverState *bdrv_find_base(BlockDriverState *bs) | |
2960 | { | |
2961 | return bdrv_find_overlay(bs, NULL); | |
2962 | } | |
2963 | ||
2964 | /* | |
2965 | * Drops images above 'base' up to and including 'top', and sets the image | |
2966 | * above 'top' to have base as its backing file. | |
2967 | * | |
2968 | * Requires that the overlay to 'top' is opened r/w, so that the backing file | |
2969 | * information in 'bs' can be properly updated. | |
2970 | * | |
2971 | * E.g., this will convert the following chain: | |
2972 | * bottom <- base <- intermediate <- top <- active | |
2973 | * | |
2974 | * to | |
2975 | * | |
2976 | * bottom <- base <- active | |
2977 | * | |
2978 | * It is allowed for bottom==base, in which case it converts: | |
2979 | * | |
2980 | * base <- intermediate <- top <- active | |
2981 | * | |
2982 | * to | |
2983 | * | |
2984 | * base <- active | |
2985 | * | |
2986 | * If backing_file_str is non-NULL, it will be used when modifying top's | |
2987 | * overlay image metadata. | |
2988 | * | |
2989 | * Error conditions: | |
2990 | * if active == top, that is considered an error | |
2991 | * | |
2992 | */ | |
2993 | int bdrv_drop_intermediate(BlockDriverState *active, BlockDriverState *top, | |
2994 | BlockDriverState *base, const char *backing_file_str) | |
2995 | { | |
2996 | BlockDriverState *new_top_bs = NULL; | |
2997 | int ret = -EIO; | |
2998 | ||
2999 | if (!top->drv || !base->drv) { | |
3000 | goto exit; | |
3001 | } | |
3002 | ||
3003 | new_top_bs = bdrv_find_overlay(active, top); | |
3004 | ||
3005 | if (new_top_bs == NULL) { | |
3006 | /* we could not find the image above 'top', this is an error */ | |
3007 | goto exit; | |
3008 | } | |
3009 | ||
3010 | /* special case of new_top_bs->backing->bs already pointing to base - nothing | |
3011 | * to do, no intermediate images */ | |
3012 | if (backing_bs(new_top_bs) == base) { | |
3013 | ret = 0; | |
3014 | goto exit; | |
3015 | } | |
3016 | ||
3017 | /* Make sure that base is in the backing chain of top */ | |
3018 | if (!bdrv_chain_contains(top, base)) { | |
3019 | goto exit; | |
3020 | } | |
3021 | ||
3022 | /* success - we can delete the intermediate states, and link top->base */ | |
3023 | backing_file_str = backing_file_str ? backing_file_str : base->filename; | |
3024 | ret = bdrv_change_backing_file(new_top_bs, backing_file_str, | |
3025 | base->drv ? base->drv->format_name : ""); | |
3026 | if (ret) { | |
3027 | goto exit; | |
3028 | } | |
3029 | bdrv_set_backing_hd(new_top_bs, base); | |
3030 | ||
3031 | ret = 0; | |
3032 | exit: | |
3033 | return ret; | |
3034 | } | |
3035 | ||
3036 | /** | |
3037 | * Truncate file to 'offset' bytes (needed only for file protocols) | |
3038 | */ | |
3039 | int bdrv_truncate(BdrvChild *child, int64_t offset) | |
3040 | { | |
3041 | BlockDriverState *bs = child->bs; | |
3042 | BlockDriver *drv = bs->drv; | |
3043 | int ret; | |
3044 | if (!drv) | |
3045 | return -ENOMEDIUM; | |
3046 | if (!drv->bdrv_truncate) | |
3047 | return -ENOTSUP; | |
3048 | if (bs->read_only) | |
3049 | return -EACCES; | |
3050 | ||
3051 | ret = drv->bdrv_truncate(bs, offset); | |
3052 | if (ret == 0) { | |
3053 | ret = refresh_total_sectors(bs, offset >> BDRV_SECTOR_BITS); | |
3054 | bdrv_dirty_bitmap_truncate(bs); | |
3055 | bdrv_parent_cb_resize(bs); | |
3056 | ++bs->write_gen; | |
3057 | } | |
3058 | return ret; | |
3059 | } | |
3060 | ||
3061 | /** | |
3062 | * Length of a allocated file in bytes. Sparse files are counted by actual | |
3063 | * allocated space. Return < 0 if error or unknown. | |
3064 | */ | |
3065 | int64_t bdrv_get_allocated_file_size(BlockDriverState *bs) | |
3066 | { | |
3067 | BlockDriver *drv = bs->drv; | |
3068 | if (!drv) { | |
3069 | return -ENOMEDIUM; | |
3070 | } | |
3071 | if (drv->bdrv_get_allocated_file_size) { | |
3072 | return drv->bdrv_get_allocated_file_size(bs); | |
3073 | } | |
3074 | if (bs->file) { | |
3075 | return bdrv_get_allocated_file_size(bs->file->bs); | |
3076 | } | |
3077 | return -ENOTSUP; | |
3078 | } | |
3079 | ||
3080 | /** | |
3081 | * Return number of sectors on success, -errno on error. | |
3082 | */ | |
3083 | int64_t bdrv_nb_sectors(BlockDriverState *bs) | |
3084 | { | |
3085 | BlockDriver *drv = bs->drv; | |
3086 | ||
3087 | if (!drv) | |
3088 | return -ENOMEDIUM; | |
3089 | ||
3090 | if (drv->has_variable_length) { | |
3091 | int ret = refresh_total_sectors(bs, bs->total_sectors); | |
3092 | if (ret < 0) { | |
3093 | return ret; | |
3094 | } | |
3095 | } | |
3096 | return bs->total_sectors; | |
3097 | } | |
3098 | ||
3099 | /** | |
3100 | * Return length in bytes on success, -errno on error. | |
3101 | * The length is always a multiple of BDRV_SECTOR_SIZE. | |
3102 | */ | |
3103 | int64_t bdrv_getlength(BlockDriverState *bs) | |
3104 | { | |
3105 | int64_t ret = bdrv_nb_sectors(bs); | |
3106 | ||
3107 | ret = ret > INT64_MAX / BDRV_SECTOR_SIZE ? -EFBIG : ret; | |
3108 | return ret < 0 ? ret : ret * BDRV_SECTOR_SIZE; | |
3109 | } | |
3110 | ||
3111 | /* return 0 as number of sectors if no device present or error */ | |
3112 | void bdrv_get_geometry(BlockDriverState *bs, uint64_t *nb_sectors_ptr) | |
3113 | { | |
3114 | int64_t nb_sectors = bdrv_nb_sectors(bs); | |
3115 | ||
3116 | *nb_sectors_ptr = nb_sectors < 0 ? 0 : nb_sectors; | |
3117 | } | |
3118 | ||
3119 | bool bdrv_is_read_only(BlockDriverState *bs) | |
3120 | { | |
3121 | return bs->read_only; | |
3122 | } | |
3123 | ||
3124 | bool bdrv_is_sg(BlockDriverState *bs) | |
3125 | { | |
3126 | return bs->sg; | |
3127 | } | |
3128 | ||
3129 | bool bdrv_is_encrypted(BlockDriverState *bs) | |
3130 | { | |
3131 | if (bs->backing && bs->backing->bs->encrypted) { | |
3132 | return true; | |
3133 | } | |
3134 | return bs->encrypted; | |
3135 | } | |
3136 | ||
3137 | bool bdrv_key_required(BlockDriverState *bs) | |
3138 | { | |
3139 | BdrvChild *backing = bs->backing; | |
3140 | ||
3141 | if (backing && backing->bs->encrypted && !backing->bs->valid_key) { | |
3142 | return true; | |
3143 | } | |
3144 | return (bs->encrypted && !bs->valid_key); | |
3145 | } | |
3146 | ||
3147 | int bdrv_set_key(BlockDriverState *bs, const char *key) | |
3148 | { | |
3149 | int ret; | |
3150 | if (bs->backing && bs->backing->bs->encrypted) { | |
3151 | ret = bdrv_set_key(bs->backing->bs, key); | |
3152 | if (ret < 0) | |
3153 | return ret; | |
3154 | if (!bs->encrypted) | |
3155 | return 0; | |
3156 | } | |
3157 | if (!bs->encrypted) { | |
3158 | return -EINVAL; | |
3159 | } else if (!bs->drv || !bs->drv->bdrv_set_key) { | |
3160 | return -ENOMEDIUM; | |
3161 | } | |
3162 | ret = bs->drv->bdrv_set_key(bs, key); | |
3163 | if (ret < 0) { | |
3164 | bs->valid_key = false; | |
3165 | } else if (!bs->valid_key) { | |
3166 | /* call the change callback now, we skipped it on open */ | |
3167 | bs->valid_key = true; | |
3168 | bdrv_parent_cb_change_media(bs, true); | |
3169 | } | |
3170 | return ret; | |
3171 | } | |
3172 | ||
3173 | /* | |
3174 | * Provide an encryption key for @bs. | |
3175 | * If @key is non-null: | |
3176 | * If @bs is not encrypted, fail. | |
3177 | * Else if the key is invalid, fail. | |
3178 | * Else set @bs's key to @key, replacing the existing key, if any. | |
3179 | * If @key is null: | |
3180 | * If @bs is encrypted and still lacks a key, fail. | |
3181 | * Else do nothing. | |
3182 | * On failure, store an error object through @errp if non-null. | |
3183 | */ | |
3184 | void bdrv_add_key(BlockDriverState *bs, const char *key, Error **errp) | |
3185 | { | |
3186 | if (key) { | |
3187 | if (!bdrv_is_encrypted(bs)) { | |
3188 | error_setg(errp, "Node '%s' is not encrypted", | |
3189 | bdrv_get_device_or_node_name(bs)); | |
3190 | } else if (bdrv_set_key(bs, key) < 0) { | |
3191 | error_setg(errp, QERR_INVALID_PASSWORD); | |
3192 | } | |
3193 | } else { | |
3194 | if (bdrv_key_required(bs)) { | |
3195 | error_set(errp, ERROR_CLASS_DEVICE_ENCRYPTED, | |
3196 | "'%s' (%s) is encrypted", | |
3197 | bdrv_get_device_or_node_name(bs), | |
3198 | bdrv_get_encrypted_filename(bs)); | |
3199 | } | |
3200 | } | |
3201 | } | |
3202 | ||
3203 | const char *bdrv_get_format_name(BlockDriverState *bs) | |
3204 | { | |
3205 | return bs->drv ? bs->drv->format_name : NULL; | |
3206 | } | |
3207 | ||
3208 | static int qsort_strcmp(const void *a, const void *b) | |
3209 | { | |
3210 | return strcmp(*(char *const *)a, *(char *const *)b); | |
3211 | } | |
3212 | ||
3213 | void bdrv_iterate_format(void (*it)(void *opaque, const char *name), | |
3214 | void *opaque) | |
3215 | { | |
3216 | BlockDriver *drv; | |
3217 | int count = 0; | |
3218 | int i; | |
3219 | const char **formats = NULL; | |
3220 | ||
3221 | QLIST_FOREACH(drv, &bdrv_drivers, list) { | |
3222 | if (drv->format_name) { | |
3223 | bool found = false; | |
3224 | int i = count; | |
3225 | while (formats && i && !found) { | |
3226 | found = !strcmp(formats[--i], drv->format_name); | |
3227 | } | |
3228 | ||
3229 | if (!found) { | |
3230 | formats = g_renew(const char *, formats, count + 1); | |
3231 | formats[count++] = drv->format_name; | |
3232 | } | |
3233 | } | |
3234 | } | |
3235 | ||
3236 | for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); i++) { | |
3237 | const char *format_name = block_driver_modules[i].format_name; | |
3238 | ||
3239 | if (format_name) { | |
3240 | bool found = false; | |
3241 | int j = count; | |
3242 | ||
3243 | while (formats && j && !found) { | |
3244 | found = !strcmp(formats[--j], format_name); | |
3245 | } | |
3246 | ||
3247 | if (!found) { | |
3248 | formats = g_renew(const char *, formats, count + 1); | |
3249 | formats[count++] = format_name; | |
3250 | } | |
3251 | } | |
3252 | } | |
3253 | ||
3254 | qsort(formats, count, sizeof(formats[0]), qsort_strcmp); | |
3255 | ||
3256 | for (i = 0; i < count; i++) { | |
3257 | it(opaque, formats[i]); | |
3258 | } | |
3259 | ||
3260 | g_free(formats); | |
3261 | } | |
3262 | ||
3263 | /* This function is to find a node in the bs graph */ | |
3264 | BlockDriverState *bdrv_find_node(const char *node_name) | |
3265 | { | |
3266 | BlockDriverState *bs; | |
3267 | ||
3268 | assert(node_name); | |
3269 | ||
3270 | QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) { | |
3271 | if (!strcmp(node_name, bs->node_name)) { | |
3272 | return bs; | |
3273 | } | |
3274 | } | |
3275 | return NULL; | |
3276 | } | |
3277 | ||
3278 | /* Put this QMP function here so it can access the static graph_bdrv_states. */ | |
3279 | BlockDeviceInfoList *bdrv_named_nodes_list(Error **errp) | |
3280 | { | |
3281 | BlockDeviceInfoList *list, *entry; | |
3282 | BlockDriverState *bs; | |
3283 | ||
3284 | list = NULL; | |
3285 | QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) { | |
3286 | BlockDeviceInfo *info = bdrv_block_device_info(NULL, bs, errp); | |
3287 | if (!info) { | |
3288 | qapi_free_BlockDeviceInfoList(list); | |
3289 | return NULL; | |
3290 | } | |
3291 | entry = g_malloc0(sizeof(*entry)); | |
3292 | entry->value = info; | |
3293 | entry->next = list; | |
3294 | list = entry; | |
3295 | } | |
3296 | ||
3297 | return list; | |
3298 | } | |
3299 | ||
3300 | BlockDriverState *bdrv_lookup_bs(const char *device, | |
3301 | const char *node_name, | |
3302 | Error **errp) | |
3303 | { | |
3304 | BlockBackend *blk; | |
3305 | BlockDriverState *bs; | |
3306 | ||
3307 | if (device) { | |
3308 | blk = blk_by_name(device); | |
3309 | ||
3310 | if (blk) { | |
3311 | bs = blk_bs(blk); | |
3312 | if (!bs) { | |
3313 | error_setg(errp, "Device '%s' has no medium", device); | |
3314 | } | |
3315 | ||
3316 | return bs; | |
3317 | } | |
3318 | } | |
3319 | ||
3320 | if (node_name) { | |
3321 | bs = bdrv_find_node(node_name); | |
3322 | ||
3323 | if (bs) { | |
3324 | return bs; | |
3325 | } | |
3326 | } | |
3327 | ||
3328 | error_setg(errp, "Cannot find device=%s nor node_name=%s", | |
3329 | device ? device : "", | |
3330 | node_name ? node_name : ""); | |
3331 | return NULL; | |
3332 | } | |
3333 | ||
3334 | /* If 'base' is in the same chain as 'top', return true. Otherwise, | |
3335 | * return false. If either argument is NULL, return false. */ | |
3336 | bool bdrv_chain_contains(BlockDriverState *top, BlockDriverState *base) | |
3337 | { | |
3338 | while (top && top != base) { | |
3339 | top = backing_bs(top); | |
3340 | } | |
3341 | ||
3342 | return top != NULL; | |
3343 | } | |
3344 | ||
3345 | BlockDriverState *bdrv_next_node(BlockDriverState *bs) | |
3346 | { | |
3347 | if (!bs) { | |
3348 | return QTAILQ_FIRST(&graph_bdrv_states); | |
3349 | } | |
3350 | return QTAILQ_NEXT(bs, node_list); | |
3351 | } | |
3352 | ||
3353 | const char *bdrv_get_node_name(const BlockDriverState *bs) | |
3354 | { | |
3355 | return bs->node_name; | |
3356 | } | |
3357 | ||
3358 | const char *bdrv_get_parent_name(const BlockDriverState *bs) | |
3359 | { | |
3360 | BdrvChild *c; | |
3361 | const char *name; | |
3362 | ||
3363 | /* If multiple parents have a name, just pick the first one. */ | |
3364 | QLIST_FOREACH(c, &bs->parents, next_parent) { | |
3365 | if (c->role->get_name) { | |
3366 | name = c->role->get_name(c); | |
3367 | if (name && *name) { | |
3368 | return name; | |
3369 | } | |
3370 | } | |
3371 | } | |
3372 | ||
3373 | return NULL; | |
3374 | } | |
3375 | ||
3376 | /* TODO check what callers really want: bs->node_name or blk_name() */ | |
3377 | const char *bdrv_get_device_name(const BlockDriverState *bs) | |
3378 | { | |
3379 | return bdrv_get_parent_name(bs) ?: ""; | |
3380 | } | |
3381 | ||
3382 | /* This can be used to identify nodes that might not have a device | |
3383 | * name associated. Since node and device names live in the same | |
3384 | * namespace, the result is unambiguous. The exception is if both are | |
3385 | * absent, then this returns an empty (non-null) string. */ | |
3386 | const char *bdrv_get_device_or_node_name(const BlockDriverState *bs) | |
3387 | { | |
3388 | return bdrv_get_parent_name(bs) ?: bs->node_name; | |
3389 | } | |
3390 | ||
3391 | int bdrv_get_flags(BlockDriverState *bs) | |
3392 | { | |
3393 | return bs->open_flags; | |
3394 | } | |
3395 | ||
3396 | int bdrv_has_zero_init_1(BlockDriverState *bs) | |
3397 | { | |
3398 | return 1; | |
3399 | } | |
3400 | ||
3401 | int bdrv_has_zero_init(BlockDriverState *bs) | |
3402 | { | |
3403 | assert(bs->drv); | |
3404 | ||
3405 | /* If BS is a copy on write image, it is initialized to | |
3406 | the contents of the base image, which may not be zeroes. */ | |
3407 | if (bs->backing) { | |
3408 | return 0; | |
3409 | } | |
3410 | if (bs->drv->bdrv_has_zero_init) { | |
3411 | return bs->drv->bdrv_has_zero_init(bs); | |
3412 | } | |
3413 | ||
3414 | /* safe default */ | |
3415 | return 0; | |
3416 | } | |
3417 | ||
3418 | bool bdrv_unallocated_blocks_are_zero(BlockDriverState *bs) | |
3419 | { | |
3420 | BlockDriverInfo bdi; | |
3421 | ||
3422 | if (bs->backing) { | |
3423 | return false; | |
3424 | } | |
3425 | ||
3426 | if (bdrv_get_info(bs, &bdi) == 0) { | |
3427 | return bdi.unallocated_blocks_are_zero; | |
3428 | } | |
3429 | ||
3430 | return false; | |
3431 | } | |
3432 | ||
3433 | bool bdrv_can_write_zeroes_with_unmap(BlockDriverState *bs) | |
3434 | { | |
3435 | BlockDriverInfo bdi; | |
3436 | ||
3437 | if (!(bs->open_flags & BDRV_O_UNMAP)) { | |
3438 | return false; | |
3439 | } | |
3440 | ||
3441 | if (bdrv_get_info(bs, &bdi) == 0) { | |
3442 | return bdi.can_write_zeroes_with_unmap; | |
3443 | } | |
3444 | ||
3445 | return false; | |
3446 | } | |
3447 | ||
3448 | const char *bdrv_get_encrypted_filename(BlockDriverState *bs) | |
3449 | { | |
3450 | if (bs->backing && bs->backing->bs->encrypted) | |
3451 | return bs->backing_file; | |
3452 | else if (bs->encrypted) | |
3453 | return bs->filename; | |
3454 | else | |
3455 | return NULL; | |
3456 | } | |
3457 | ||
3458 | void bdrv_get_backing_filename(BlockDriverState *bs, | |
3459 | char *filename, int filename_size) | |
3460 | { | |
3461 | pstrcpy(filename, filename_size, bs->backing_file); | |
3462 | } | |
3463 | ||
3464 | int bdrv_get_info(BlockDriverState *bs, BlockDriverInfo *bdi) | |
3465 | { | |
3466 | BlockDriver *drv = bs->drv; | |
3467 | if (!drv) | |
3468 | return -ENOMEDIUM; | |
3469 | if (!drv->bdrv_get_info) | |
3470 | return -ENOTSUP; | |
3471 | memset(bdi, 0, sizeof(*bdi)); | |
3472 | return drv->bdrv_get_info(bs, bdi); | |
3473 | } | |
3474 | ||
3475 | ImageInfoSpecific *bdrv_get_specific_info(BlockDriverState *bs) | |
3476 | { | |
3477 | BlockDriver *drv = bs->drv; | |
3478 | if (drv && drv->bdrv_get_specific_info) { | |
3479 | return drv->bdrv_get_specific_info(bs); | |
3480 | } | |
3481 | return NULL; | |
3482 | } | |
3483 | ||
3484 | void bdrv_debug_event(BlockDriverState *bs, BlkdebugEvent event) | |
3485 | { | |
3486 | if (!bs || !bs->drv || !bs->drv->bdrv_debug_event) { | |
3487 | return; | |
3488 | } | |
3489 | ||
3490 | bs->drv->bdrv_debug_event(bs, event); | |
3491 | } | |
3492 | ||
3493 | int bdrv_debug_breakpoint(BlockDriverState *bs, const char *event, | |
3494 | const char *tag) | |
3495 | { | |
3496 | while (bs && bs->drv && !bs->drv->bdrv_debug_breakpoint) { | |
3497 | bs = bs->file ? bs->file->bs : NULL; | |
3498 | } | |
3499 | ||
3500 | if (bs && bs->drv && bs->drv->bdrv_debug_breakpoint) { | |
3501 | return bs->drv->bdrv_debug_breakpoint(bs, event, tag); | |
3502 | } | |
3503 | ||
3504 | return -ENOTSUP; | |
3505 | } | |
3506 | ||
3507 | int bdrv_debug_remove_breakpoint(BlockDriverState *bs, const char *tag) | |
3508 | { | |
3509 | while (bs && bs->drv && !bs->drv->bdrv_debug_remove_breakpoint) { | |
3510 | bs = bs->file ? bs->file->bs : NULL; | |
3511 | } | |
3512 | ||
3513 | if (bs && bs->drv && bs->drv->bdrv_debug_remove_breakpoint) { | |
3514 | return bs->drv->bdrv_debug_remove_breakpoint(bs, tag); | |
3515 | } | |
3516 | ||
3517 | return -ENOTSUP; | |
3518 | } | |
3519 | ||
3520 | int bdrv_debug_resume(BlockDriverState *bs, const char *tag) | |
3521 | { | |
3522 | while (bs && (!bs->drv || !bs->drv->bdrv_debug_resume)) { | |
3523 | bs = bs->file ? bs->file->bs : NULL; | |
3524 | } | |
3525 | ||
3526 | if (bs && bs->drv && bs->drv->bdrv_debug_resume) { | |
3527 | return bs->drv->bdrv_debug_resume(bs, tag); | |
3528 | } | |
3529 | ||
3530 | return -ENOTSUP; | |
3531 | } | |
3532 | ||
3533 | bool bdrv_debug_is_suspended(BlockDriverState *bs, const char *tag) | |
3534 | { | |
3535 | while (bs && bs->drv && !bs->drv->bdrv_debug_is_suspended) { | |
3536 | bs = bs->file ? bs->file->bs : NULL; | |
3537 | } | |
3538 | ||
3539 | if (bs && bs->drv && bs->drv->bdrv_debug_is_suspended) { | |
3540 | return bs->drv->bdrv_debug_is_suspended(bs, tag); | |
3541 | } | |
3542 | ||
3543 | return false; | |
3544 | } | |
3545 | ||
3546 | /* backing_file can either be relative, or absolute, or a protocol. If it is | |
3547 | * relative, it must be relative to the chain. So, passing in bs->filename | |
3548 | * from a BDS as backing_file should not be done, as that may be relative to | |
3549 | * the CWD rather than the chain. */ | |
3550 | BlockDriverState *bdrv_find_backing_image(BlockDriverState *bs, | |
3551 | const char *backing_file) | |
3552 | { | |
3553 | char *filename_full = NULL; | |
3554 | char *backing_file_full = NULL; | |
3555 | char *filename_tmp = NULL; | |
3556 | int is_protocol = 0; | |
3557 | BlockDriverState *curr_bs = NULL; | |
3558 | BlockDriverState *retval = NULL; | |
3559 | Error *local_error = NULL; | |
3560 | ||
3561 | if (!bs || !bs->drv || !backing_file) { | |
3562 | return NULL; | |
3563 | } | |
3564 | ||
3565 | filename_full = g_malloc(PATH_MAX); | |
3566 | backing_file_full = g_malloc(PATH_MAX); | |
3567 | filename_tmp = g_malloc(PATH_MAX); | |
3568 | ||
3569 | is_protocol = path_has_protocol(backing_file); | |
3570 | ||
3571 | for (curr_bs = bs; curr_bs->backing; curr_bs = curr_bs->backing->bs) { | |
3572 | ||
3573 | /* If either of the filename paths is actually a protocol, then | |
3574 | * compare unmodified paths; otherwise make paths relative */ | |
3575 | if (is_protocol || path_has_protocol(curr_bs->backing_file)) { | |
3576 | if (strcmp(backing_file, curr_bs->backing_file) == 0) { | |
3577 | retval = curr_bs->backing->bs; | |
3578 | break; | |
3579 | } | |
3580 | /* Also check against the full backing filename for the image */ | |
3581 | bdrv_get_full_backing_filename(curr_bs, backing_file_full, PATH_MAX, | |
3582 | &local_error); | |
3583 | if (local_error == NULL) { | |
3584 | if (strcmp(backing_file, backing_file_full) == 0) { | |
3585 | retval = curr_bs->backing->bs; | |
3586 | break; | |
3587 | } | |
3588 | } else { | |
3589 | error_free(local_error); | |
3590 | local_error = NULL; | |
3591 | } | |
3592 | } else { | |
3593 | /* If not an absolute filename path, make it relative to the current | |
3594 | * image's filename path */ | |
3595 | path_combine(filename_tmp, PATH_MAX, curr_bs->filename, | |
3596 | backing_file); | |
3597 | ||
3598 | /* We are going to compare absolute pathnames */ | |
3599 | if (!realpath(filename_tmp, filename_full)) { | |
3600 | continue; | |
3601 | } | |
3602 | ||
3603 | /* We need to make sure the backing filename we are comparing against | |
3604 | * is relative to the current image filename (or absolute) */ | |
3605 | path_combine(filename_tmp, PATH_MAX, curr_bs->filename, | |
3606 | curr_bs->backing_file); | |
3607 | ||
3608 | if (!realpath(filename_tmp, backing_file_full)) { | |
3609 | continue; | |
3610 | } | |
3611 | ||
3612 | if (strcmp(backing_file_full, filename_full) == 0) { | |
3613 | retval = curr_bs->backing->bs; | |
3614 | break; | |
3615 | } | |
3616 | } | |
3617 | } | |
3618 | ||
3619 | g_free(filename_full); | |
3620 | g_free(backing_file_full); | |
3621 | g_free(filename_tmp); | |
3622 | return retval; | |
3623 | } | |
3624 | ||
3625 | int bdrv_get_backing_file_depth(BlockDriverState *bs) | |
3626 | { | |
3627 | if (!bs->drv) { | |
3628 | return 0; | |
3629 | } | |
3630 | ||
3631 | if (!bs->backing) { | |
3632 | return 0; | |
3633 | } | |
3634 | ||
3635 | return 1 + bdrv_get_backing_file_depth(bs->backing->bs); | |
3636 | } | |
3637 | ||
3638 | void bdrv_init(void) | |
3639 | { | |
3640 | module_call_init(MODULE_INIT_BLOCK); | |
3641 | } | |
3642 | ||
3643 | void bdrv_init_with_whitelist(void) | |
3644 | { | |
3645 | use_bdrv_whitelist = 1; | |
3646 | bdrv_init(); | |
3647 | } | |
3648 | ||
3649 | void bdrv_invalidate_cache(BlockDriverState *bs, Error **errp) | |
3650 | { | |
3651 | BdrvChild *child; | |
3652 | Error *local_err = NULL; | |
3653 | int ret; | |
3654 | ||
3655 | if (!bs->drv) { | |
3656 | return; | |
3657 | } | |
3658 | ||
3659 | if (!(bs->open_flags & BDRV_O_INACTIVE)) { | |
3660 | return; | |
3661 | } | |
3662 | ||
3663 | QLIST_FOREACH(child, &bs->children, next) { | |
3664 | bdrv_invalidate_cache(child->bs, &local_err); | |
3665 | if (local_err) { | |
3666 | error_propagate(errp, local_err); | |
3667 | return; | |
3668 | } | |
3669 | } | |
3670 | ||
3671 | bs->open_flags &= ~BDRV_O_INACTIVE; | |
3672 | if (bs->drv->bdrv_invalidate_cache) { | |
3673 | bs->drv->bdrv_invalidate_cache(bs, &local_err); | |
3674 | if (local_err) { | |
3675 | bs->open_flags |= BDRV_O_INACTIVE; | |
3676 | error_propagate(errp, local_err); | |
3677 | return; | |
3678 | } | |
3679 | } | |
3680 | ||
3681 | ret = refresh_total_sectors(bs, bs->total_sectors); | |
3682 | if (ret < 0) { | |
3683 | bs->open_flags |= BDRV_O_INACTIVE; | |
3684 | error_setg_errno(errp, -ret, "Could not refresh total sector count"); | |
3685 | return; | |
3686 | } | |
3687 | } | |
3688 | ||
3689 | void bdrv_invalidate_cache_all(Error **errp) | |
3690 | { | |
3691 | BlockDriverState *bs; | |
3692 | Error *local_err = NULL; | |
3693 | BdrvNextIterator it; | |
3694 | ||
3695 | for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) { | |
3696 | AioContext *aio_context = bdrv_get_aio_context(bs); | |
3697 | ||
3698 | aio_context_acquire(aio_context); | |
3699 | bdrv_invalidate_cache(bs, &local_err); | |
3700 | aio_context_release(aio_context); | |
3701 | if (local_err) { | |
3702 | error_propagate(errp, local_err); | |
3703 | return; | |
3704 | } | |
3705 | } | |
3706 | } | |
3707 | ||
3708 | static int bdrv_inactivate_recurse(BlockDriverState *bs, | |
3709 | bool setting_flag) | |
3710 | { | |
3711 | BdrvChild *child; | |
3712 | int ret; | |
3713 | ||
3714 | if (!setting_flag && bs->drv->bdrv_inactivate) { | |
3715 | ret = bs->drv->bdrv_inactivate(bs); | |
3716 | if (ret < 0) { | |
3717 | return ret; | |
3718 | } | |
3719 | } | |
3720 | ||
3721 | QLIST_FOREACH(child, &bs->children, next) { | |
3722 | ret = bdrv_inactivate_recurse(child->bs, setting_flag); | |
3723 | if (ret < 0) { | |
3724 | return ret; | |
3725 | } | |
3726 | } | |
3727 | ||
3728 | if (setting_flag) { | |
3729 | bs->open_flags |= BDRV_O_INACTIVE; | |
3730 | } | |
3731 | return 0; | |
3732 | } | |
3733 | ||
3734 | int bdrv_inactivate_all(void) | |
3735 | { | |
3736 | BlockDriverState *bs = NULL; | |
3737 | BdrvNextIterator it; | |
3738 | int ret = 0; | |
3739 | int pass; | |
3740 | ||
3741 | for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) { | |
3742 | aio_context_acquire(bdrv_get_aio_context(bs)); | |
3743 | } | |
3744 | ||
3745 | /* We do two passes of inactivation. The first pass calls to drivers' | |
3746 | * .bdrv_inactivate callbacks recursively so all cache is flushed to disk; | |
3747 | * the second pass sets the BDRV_O_INACTIVE flag so that no further write | |
3748 | * is allowed. */ | |
3749 | for (pass = 0; pass < 2; pass++) { | |
3750 | for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) { | |
3751 | ret = bdrv_inactivate_recurse(bs, pass); | |
3752 | if (ret < 0) { | |
3753 | goto out; | |
3754 | } | |
3755 | } | |
3756 | } | |
3757 | ||
3758 | out: | |
3759 | for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) { | |
3760 | aio_context_release(bdrv_get_aio_context(bs)); | |
3761 | } | |
3762 | ||
3763 | return ret; | |
3764 | } | |
3765 | ||
3766 | /**************************************************************/ | |
3767 | /* removable device support */ | |
3768 | ||
3769 | /** | |
3770 | * Return TRUE if the media is present | |
3771 | */ | |
3772 | bool bdrv_is_inserted(BlockDriverState *bs) | |
3773 | { | |
3774 | BlockDriver *drv = bs->drv; | |
3775 | BdrvChild *child; | |
3776 | ||
3777 | if (!drv) { | |
3778 | return false; | |
3779 | } | |
3780 | if (drv->bdrv_is_inserted) { | |
3781 | return drv->bdrv_is_inserted(bs); | |
3782 | } | |
3783 | QLIST_FOREACH(child, &bs->children, next) { | |
3784 | if (!bdrv_is_inserted(child->bs)) { | |
3785 | return false; | |
3786 | } | |
3787 | } | |
3788 | return true; | |
3789 | } | |
3790 | ||
3791 | /** | |
3792 | * Return whether the media changed since the last call to this | |
3793 | * function, or -ENOTSUP if we don't know. Most drivers don't know. | |
3794 | */ | |
3795 | int bdrv_media_changed(BlockDriverState *bs) | |
3796 | { | |
3797 | BlockDriver *drv = bs->drv; | |
3798 | ||
3799 | if (drv && drv->bdrv_media_changed) { | |
3800 | return drv->bdrv_media_changed(bs); | |
3801 | } | |
3802 | return -ENOTSUP; | |
3803 | } | |
3804 | ||
3805 | /** | |
3806 | * If eject_flag is TRUE, eject the media. Otherwise, close the tray | |
3807 | */ | |
3808 | void bdrv_eject(BlockDriverState *bs, bool eject_flag) | |
3809 | { | |
3810 | BlockDriver *drv = bs->drv; | |
3811 | ||
3812 | if (drv && drv->bdrv_eject) { | |
3813 | drv->bdrv_eject(bs, eject_flag); | |
3814 | } | |
3815 | } | |
3816 | ||
3817 | /** | |
3818 | * Lock or unlock the media (if it is locked, the user won't be able | |
3819 | * to eject it manually). | |
3820 | */ | |
3821 | void bdrv_lock_medium(BlockDriverState *bs, bool locked) | |
3822 | { | |
3823 | BlockDriver *drv = bs->drv; | |
3824 | ||
3825 | trace_bdrv_lock_medium(bs, locked); | |
3826 | ||
3827 | if (drv && drv->bdrv_lock_medium) { | |
3828 | drv->bdrv_lock_medium(bs, locked); | |
3829 | } | |
3830 | } | |
3831 | ||
3832 | /* Get a reference to bs */ | |
3833 | void bdrv_ref(BlockDriverState *bs) | |
3834 | { | |
3835 | bs->refcnt++; | |
3836 | } | |
3837 | ||
3838 | /* Release a previously grabbed reference to bs. | |
3839 | * If after releasing, reference count is zero, the BlockDriverState is | |
3840 | * deleted. */ | |
3841 | void bdrv_unref(BlockDriverState *bs) | |
3842 | { | |
3843 | if (!bs) { | |
3844 | return; | |
3845 | } | |
3846 | assert(bs->refcnt > 0); | |
3847 | if (--bs->refcnt == 0) { | |
3848 | bdrv_delete(bs); | |
3849 | } | |
3850 | } | |
3851 | ||
3852 | struct BdrvOpBlocker { | |
3853 | Error *reason; | |
3854 | QLIST_ENTRY(BdrvOpBlocker) list; | |
3855 | }; | |
3856 | ||
3857 | bool bdrv_op_is_blocked(BlockDriverState *bs, BlockOpType op, Error **errp) | |
3858 | { | |
3859 | BdrvOpBlocker *blocker; | |
3860 | assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX); | |
3861 | if (!QLIST_EMPTY(&bs->op_blockers[op])) { | |
3862 | blocker = QLIST_FIRST(&bs->op_blockers[op]); | |
3863 | if (errp) { | |
3864 | *errp = error_copy(blocker->reason); | |
3865 | error_prepend(errp, "Node '%s' is busy: ", | |
3866 | bdrv_get_device_or_node_name(bs)); | |
3867 | } | |
3868 | return true; | |
3869 | } | |
3870 | return false; | |
3871 | } | |
3872 | ||
3873 | void bdrv_op_block(BlockDriverState *bs, BlockOpType op, Error *reason) | |
3874 | { | |
3875 | BdrvOpBlocker *blocker; | |
3876 | assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX); | |
3877 | ||
3878 | blocker = g_new0(BdrvOpBlocker, 1); | |
3879 | blocker->reason = reason; | |
3880 | QLIST_INSERT_HEAD(&bs->op_blockers[op], blocker, list); | |
3881 | } | |
3882 | ||
3883 | void bdrv_op_unblock(BlockDriverState *bs, BlockOpType op, Error *reason) | |
3884 | { | |
3885 | BdrvOpBlocker *blocker, *next; | |
3886 | assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX); | |
3887 | QLIST_FOREACH_SAFE(blocker, &bs->op_blockers[op], list, next) { | |
3888 | if (blocker->reason == reason) { | |
3889 | QLIST_REMOVE(blocker, list); | |
3890 | g_free(blocker); | |
3891 | } | |
3892 | } | |
3893 | } | |
3894 | ||
3895 | void bdrv_op_block_all(BlockDriverState *bs, Error *reason) | |
3896 | { | |
3897 | int i; | |
3898 | for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) { | |
3899 | bdrv_op_block(bs, i, reason); | |
3900 | } | |
3901 | } | |
3902 | ||
3903 | void bdrv_op_unblock_all(BlockDriverState *bs, Error *reason) | |
3904 | { | |
3905 | int i; | |
3906 | for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) { | |
3907 | bdrv_op_unblock(bs, i, reason); | |
3908 | } | |
3909 | } | |
3910 | ||
3911 | bool bdrv_op_blocker_is_empty(BlockDriverState *bs) | |
3912 | { | |
3913 | int i; | |
3914 | ||
3915 | for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) { | |
3916 | if (!QLIST_EMPTY(&bs->op_blockers[i])) { | |
3917 | return false; | |
3918 | } | |
3919 | } | |
3920 | return true; | |
3921 | } | |
3922 | ||
3923 | void bdrv_img_create(const char *filename, const char *fmt, | |
3924 | const char *base_filename, const char *base_fmt, | |
3925 | char *options, uint64_t img_size, int flags, | |
3926 | Error **errp, bool quiet) | |
3927 | { | |
3928 | QemuOptsList *create_opts = NULL; | |
3929 | QemuOpts *opts = NULL; | |
3930 | const char *backing_fmt, *backing_file; | |
3931 | int64_t size; | |
3932 | BlockDriver *drv, *proto_drv; | |
3933 | Error *local_err = NULL; | |
3934 | int ret = 0; | |
3935 | ||
3936 | /* Find driver and parse its options */ | |
3937 | drv = bdrv_find_format(fmt); | |
3938 | if (!drv) { | |
3939 | error_setg(errp, "Unknown file format '%s'", fmt); | |
3940 | return; | |
3941 | } | |
3942 | ||
3943 | proto_drv = bdrv_find_protocol(filename, true, errp); | |
3944 | if (!proto_drv) { | |
3945 | return; | |
3946 | } | |
3947 | ||
3948 | if (!drv->create_opts) { | |
3949 | error_setg(errp, "Format driver '%s' does not support image creation", | |
3950 | drv->format_name); | |
3951 | return; | |
3952 | } | |
3953 | ||
3954 | if (!proto_drv->create_opts) { | |
3955 | error_setg(errp, "Protocol driver '%s' does not support image creation", | |
3956 | proto_drv->format_name); | |
3957 | return; | |
3958 | } | |
3959 | ||
3960 | create_opts = qemu_opts_append(create_opts, drv->create_opts); | |
3961 | create_opts = qemu_opts_append(create_opts, proto_drv->create_opts); | |
3962 | ||
3963 | /* Create parameter list with default values */ | |
3964 | opts = qemu_opts_create(create_opts, NULL, 0, &error_abort); | |
3965 | qemu_opt_set_number(opts, BLOCK_OPT_SIZE, img_size, &error_abort); | |
3966 | ||
3967 | /* Parse -o options */ | |
3968 | if (options) { | |
3969 | qemu_opts_do_parse(opts, options, NULL, &local_err); | |
3970 | if (local_err) { | |
3971 | error_report_err(local_err); | |
3972 | local_err = NULL; | |
3973 | error_setg(errp, "Invalid options for file format '%s'", fmt); | |
3974 | goto out; | |
3975 | } | |
3976 | } | |
3977 | ||
3978 | if (base_filename) { | |
3979 | qemu_opt_set(opts, BLOCK_OPT_BACKING_FILE, base_filename, &local_err); | |
3980 | if (local_err) { | |
3981 | error_setg(errp, "Backing file not supported for file format '%s'", | |
3982 | fmt); | |
3983 | goto out; | |
3984 | } | |
3985 | } | |
3986 | ||
3987 | if (base_fmt) { | |
3988 | qemu_opt_set(opts, BLOCK_OPT_BACKING_FMT, base_fmt, &local_err); | |
3989 | if (local_err) { | |
3990 | error_setg(errp, "Backing file format not supported for file " | |
3991 | "format '%s'", fmt); | |
3992 | goto out; | |
3993 | } | |
3994 | } | |
3995 | ||
3996 | backing_file = qemu_opt_get(opts, BLOCK_OPT_BACKING_FILE); | |
3997 | if (backing_file) { | |
3998 | if (!strcmp(filename, backing_file)) { | |
3999 | error_setg(errp, "Error: Trying to create an image with the " | |
4000 | "same filename as the backing file"); | |
4001 | goto out; | |
4002 | } | |
4003 | } | |
4004 | ||
4005 | backing_fmt = qemu_opt_get(opts, BLOCK_OPT_BACKING_FMT); | |
4006 | ||
4007 | // The size for the image must always be specified, with one exception: | |
4008 | // If we are using a backing file, we can obtain the size from there | |
4009 | size = qemu_opt_get_size(opts, BLOCK_OPT_SIZE, 0); | |
4010 | if (size == -1) { | |
4011 | if (backing_file) { | |
4012 | BlockDriverState *bs; | |
4013 | char *full_backing = g_new0(char, PATH_MAX); | |
4014 | int64_t size; | |
4015 | int back_flags; | |
4016 | QDict *backing_options = NULL; | |
4017 | ||
4018 | bdrv_get_full_backing_filename_from_filename(filename, backing_file, | |
4019 | full_backing, PATH_MAX, | |
4020 | &local_err); | |
4021 | if (local_err) { | |
4022 | g_free(full_backing); | |
4023 | goto out; | |
4024 | } | |
4025 | ||
4026 | /* backing files always opened read-only */ | |
4027 | back_flags = flags; | |
4028 | back_flags &= ~(BDRV_O_RDWR | BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING); | |
4029 | ||
4030 | if (backing_fmt) { | |
4031 | backing_options = qdict_new(); | |
4032 | qdict_put(backing_options, "driver", | |
4033 | qstring_from_str(backing_fmt)); | |
4034 | } | |
4035 | ||
4036 | bs = bdrv_open(full_backing, NULL, backing_options, back_flags, | |
4037 | &local_err); | |
4038 | g_free(full_backing); | |
4039 | if (!bs) { | |
4040 | goto out; | |
4041 | } | |
4042 | size = bdrv_getlength(bs); | |
4043 | if (size < 0) { | |
4044 | error_setg_errno(errp, -size, "Could not get size of '%s'", | |
4045 | backing_file); | |
4046 | bdrv_unref(bs); | |
4047 | goto out; | |
4048 | } | |
4049 | ||
4050 | qemu_opt_set_number(opts, BLOCK_OPT_SIZE, size, &error_abort); | |
4051 | ||
4052 | bdrv_unref(bs); | |
4053 | } else { | |
4054 | error_setg(errp, "Image creation needs a size parameter"); | |
4055 | goto out; | |
4056 | } | |
4057 | } | |
4058 | ||
4059 | if (!quiet) { | |
4060 | printf("Formatting '%s', fmt=%s ", filename, fmt); | |
4061 | qemu_opts_print(opts, " "); | |
4062 | puts(""); | |
4063 | } | |
4064 | ||
4065 | ret = bdrv_create(drv, filename, opts, &local_err); | |
4066 | ||
4067 | if (ret == -EFBIG) { | |
4068 | /* This is generally a better message than whatever the driver would | |
4069 | * deliver (especially because of the cluster_size_hint), since that | |
4070 | * is most probably not much different from "image too large". */ | |
4071 | const char *cluster_size_hint = ""; | |
4072 | if (qemu_opt_get_size(opts, BLOCK_OPT_CLUSTER_SIZE, 0)) { | |
4073 | cluster_size_hint = " (try using a larger cluster size)"; | |
4074 | } | |
4075 | error_setg(errp, "The image size is too large for file format '%s'" | |
4076 | "%s", fmt, cluster_size_hint); | |
4077 | error_free(local_err); | |
4078 | local_err = NULL; | |
4079 | } | |
4080 | ||
4081 | out: | |
4082 | qemu_opts_del(opts); | |
4083 | qemu_opts_free(create_opts); | |
4084 | error_propagate(errp, local_err); | |
4085 | } | |
4086 | ||
4087 | AioContext *bdrv_get_aio_context(BlockDriverState *bs) | |
4088 | { | |
4089 | return bs->aio_context; | |
4090 | } | |
4091 | ||
4092 | static void bdrv_do_remove_aio_context_notifier(BdrvAioNotifier *ban) | |
4093 | { | |
4094 | QLIST_REMOVE(ban, list); | |
4095 | g_free(ban); | |
4096 | } | |
4097 | ||
4098 | void bdrv_detach_aio_context(BlockDriverState *bs) | |
4099 | { | |
4100 | BdrvAioNotifier *baf, *baf_tmp; | |
4101 | BdrvChild *child; | |
4102 | ||
4103 | if (!bs->drv) { | |
4104 | return; | |
4105 | } | |
4106 | ||
4107 | assert(!bs->walking_aio_notifiers); | |
4108 | bs->walking_aio_notifiers = true; | |
4109 | QLIST_FOREACH_SAFE(baf, &bs->aio_notifiers, list, baf_tmp) { | |
4110 | if (baf->deleted) { | |
4111 | bdrv_do_remove_aio_context_notifier(baf); | |
4112 | } else { | |
4113 | baf->detach_aio_context(baf->opaque); | |
4114 | } | |
4115 | } | |
4116 | /* Never mind iterating again to check for ->deleted. bdrv_close() will | |
4117 | * remove remaining aio notifiers if we aren't called again. | |
4118 | */ | |
4119 | bs->walking_aio_notifiers = false; | |
4120 | ||
4121 | if (bs->drv->bdrv_detach_aio_context) { | |
4122 | bs->drv->bdrv_detach_aio_context(bs); | |
4123 | } | |
4124 | QLIST_FOREACH(child, &bs->children, next) { | |
4125 | bdrv_detach_aio_context(child->bs); | |
4126 | } | |
4127 | ||
4128 | bs->aio_context = NULL; | |
4129 | } | |
4130 | ||
4131 | void bdrv_attach_aio_context(BlockDriverState *bs, | |
4132 | AioContext *new_context) | |
4133 | { | |
4134 | BdrvAioNotifier *ban, *ban_tmp; | |
4135 | BdrvChild *child; | |
4136 | ||
4137 | if (!bs->drv) { | |
4138 | return; | |
4139 | } | |
4140 | ||
4141 | bs->aio_context = new_context; | |
4142 | ||
4143 | QLIST_FOREACH(child, &bs->children, next) { | |
4144 | bdrv_attach_aio_context(child->bs, new_context); | |
4145 | } | |
4146 | if (bs->drv->bdrv_attach_aio_context) { | |
4147 | bs->drv->bdrv_attach_aio_context(bs, new_context); | |
4148 | } | |
4149 | ||
4150 | assert(!bs->walking_aio_notifiers); | |
4151 | bs->walking_aio_notifiers = true; | |
4152 | QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_tmp) { | |
4153 | if (ban->deleted) { | |
4154 | bdrv_do_remove_aio_context_notifier(ban); | |
4155 | } else { | |
4156 | ban->attached_aio_context(new_context, ban->opaque); | |
4157 | } | |
4158 | } | |
4159 | bs->walking_aio_notifiers = false; | |
4160 | } | |
4161 | ||
4162 | void bdrv_set_aio_context(BlockDriverState *bs, AioContext *new_context) | |
4163 | { | |
4164 | bdrv_drain(bs); /* ensure there are no in-flight requests */ | |
4165 | ||
4166 | bdrv_detach_aio_context(bs); | |
4167 | ||
4168 | /* This function executes in the old AioContext so acquire the new one in | |
4169 | * case it runs in a different thread. | |
4170 | */ | |
4171 | aio_context_acquire(new_context); | |
4172 | bdrv_attach_aio_context(bs, new_context); | |
4173 | aio_context_release(new_context); | |
4174 | } | |
4175 | ||
4176 | void bdrv_add_aio_context_notifier(BlockDriverState *bs, | |
4177 | void (*attached_aio_context)(AioContext *new_context, void *opaque), | |
4178 | void (*detach_aio_context)(void *opaque), void *opaque) | |
4179 | { | |
4180 | BdrvAioNotifier *ban = g_new(BdrvAioNotifier, 1); | |
4181 | *ban = (BdrvAioNotifier){ | |
4182 | .attached_aio_context = attached_aio_context, | |
4183 | .detach_aio_context = detach_aio_context, | |
4184 | .opaque = opaque | |
4185 | }; | |
4186 | ||
4187 | QLIST_INSERT_HEAD(&bs->aio_notifiers, ban, list); | |
4188 | } | |
4189 | ||
4190 | void bdrv_remove_aio_context_notifier(BlockDriverState *bs, | |
4191 | void (*attached_aio_context)(AioContext *, | |
4192 | void *), | |
4193 | void (*detach_aio_context)(void *), | |
4194 | void *opaque) | |
4195 | { | |
4196 | BdrvAioNotifier *ban, *ban_next; | |
4197 | ||
4198 | QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_next) { | |
4199 | if (ban->attached_aio_context == attached_aio_context && | |
4200 | ban->detach_aio_context == detach_aio_context && | |
4201 | ban->opaque == opaque && | |
4202 | ban->deleted == false) | |
4203 | { | |
4204 | if (bs->walking_aio_notifiers) { | |
4205 | ban->deleted = true; | |
4206 | } else { | |
4207 | bdrv_do_remove_aio_context_notifier(ban); | |
4208 | } | |
4209 | return; | |
4210 | } | |
4211 | } | |
4212 | ||
4213 | abort(); | |
4214 | } | |
4215 | ||
4216 | int bdrv_amend_options(BlockDriverState *bs, QemuOpts *opts, | |
4217 | BlockDriverAmendStatusCB *status_cb, void *cb_opaque) | |
4218 | { | |
4219 | if (!bs->drv->bdrv_amend_options) { | |
4220 | return -ENOTSUP; | |
4221 | } | |
4222 | return bs->drv->bdrv_amend_options(bs, opts, status_cb, cb_opaque); | |
4223 | } | |
4224 | ||
4225 | /* This function will be called by the bdrv_recurse_is_first_non_filter method | |
4226 | * of block filter and by bdrv_is_first_non_filter. | |
4227 | * It is used to test if the given bs is the candidate or recurse more in the | |
4228 | * node graph. | |
4229 | */ | |
4230 | bool bdrv_recurse_is_first_non_filter(BlockDriverState *bs, | |
4231 | BlockDriverState *candidate) | |
4232 | { | |
4233 | /* return false if basic checks fails */ | |
4234 | if (!bs || !bs->drv) { | |
4235 | return false; | |
4236 | } | |
4237 | ||
4238 | /* the code reached a non block filter driver -> check if the bs is | |
4239 | * the same as the candidate. It's the recursion termination condition. | |
4240 | */ | |
4241 | if (!bs->drv->is_filter) { | |
4242 | return bs == candidate; | |
4243 | } | |
4244 | /* Down this path the driver is a block filter driver */ | |
4245 | ||
4246 | /* If the block filter recursion method is defined use it to recurse down | |
4247 | * the node graph. | |
4248 | */ | |
4249 | if (bs->drv->bdrv_recurse_is_first_non_filter) { | |
4250 | return bs->drv->bdrv_recurse_is_first_non_filter(bs, candidate); | |
4251 | } | |
4252 | ||
4253 | /* the driver is a block filter but don't allow to recurse -> return false | |
4254 | */ | |
4255 | return false; | |
4256 | } | |
4257 | ||
4258 | /* This function checks if the candidate is the first non filter bs down it's | |
4259 | * bs chain. Since we don't have pointers to parents it explore all bs chains | |
4260 | * from the top. Some filters can choose not to pass down the recursion. | |
4261 | */ | |
4262 | bool bdrv_is_first_non_filter(BlockDriverState *candidate) | |
4263 | { | |
4264 | BlockDriverState *bs; | |
4265 | BdrvNextIterator it; | |
4266 | ||
4267 | /* walk down the bs forest recursively */ | |
4268 | for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) { | |
4269 | bool perm; | |
4270 | ||
4271 | /* try to recurse in this top level bs */ | |
4272 | perm = bdrv_recurse_is_first_non_filter(bs, candidate); | |
4273 | ||
4274 | /* candidate is the first non filter */ | |
4275 | if (perm) { | |
4276 | return true; | |
4277 | } | |
4278 | } | |
4279 | ||
4280 | return false; | |
4281 | } | |
4282 | ||
4283 | BlockDriverState *check_to_replace_node(BlockDriverState *parent_bs, | |
4284 | const char *node_name, Error **errp) | |
4285 | { | |
4286 | BlockDriverState *to_replace_bs = bdrv_find_node(node_name); | |
4287 | AioContext *aio_context; | |
4288 | ||
4289 | if (!to_replace_bs) { | |
4290 | error_setg(errp, "Node name '%s' not found", node_name); | |
4291 | return NULL; | |
4292 | } | |
4293 | ||
4294 | aio_context = bdrv_get_aio_context(to_replace_bs); | |
4295 | aio_context_acquire(aio_context); | |
4296 | ||
4297 | if (bdrv_op_is_blocked(to_replace_bs, BLOCK_OP_TYPE_REPLACE, errp)) { | |
4298 | to_replace_bs = NULL; | |
4299 | goto out; | |
4300 | } | |
4301 | ||
4302 | /* We don't want arbitrary node of the BDS chain to be replaced only the top | |
4303 | * most non filter in order to prevent data corruption. | |
4304 | * Another benefit is that this tests exclude backing files which are | |
4305 | * blocked by the backing blockers. | |
4306 | */ | |
4307 | if (!bdrv_recurse_is_first_non_filter(parent_bs, to_replace_bs)) { | |
4308 | error_setg(errp, "Only top most non filter can be replaced"); | |
4309 | to_replace_bs = NULL; | |
4310 | goto out; | |
4311 | } | |
4312 | ||
4313 | out: | |
4314 | aio_context_release(aio_context); | |
4315 | return to_replace_bs; | |
4316 | } | |
4317 | ||
4318 | static bool append_open_options(QDict *d, BlockDriverState *bs) | |
4319 | { | |
4320 | const QDictEntry *entry; | |
4321 | QemuOptDesc *desc; | |
4322 | BdrvChild *child; | |
4323 | bool found_any = false; | |
4324 | const char *p; | |
4325 | ||
4326 | for (entry = qdict_first(bs->options); entry; | |
4327 | entry = qdict_next(bs->options, entry)) | |
4328 | { | |
4329 | /* Exclude options for children */ | |
4330 | QLIST_FOREACH(child, &bs->children, next) { | |
4331 | if (strstart(qdict_entry_key(entry), child->name, &p) | |
4332 | && (!*p || *p == '.')) | |
4333 | { | |
4334 | break; | |
4335 | } | |
4336 | } | |
4337 | if (child) { | |
4338 | continue; | |
4339 | } | |
4340 | ||
4341 | /* And exclude all non-driver-specific options */ | |
4342 | for (desc = bdrv_runtime_opts.desc; desc->name; desc++) { | |
4343 | if (!strcmp(qdict_entry_key(entry), desc->name)) { | |
4344 | break; | |
4345 | } | |
4346 | } | |
4347 | if (desc->name) { | |
4348 | continue; | |
4349 | } | |
4350 | ||
4351 | qobject_incref(qdict_entry_value(entry)); | |
4352 | qdict_put_obj(d, qdict_entry_key(entry), qdict_entry_value(entry)); | |
4353 | found_any = true; | |
4354 | } | |
4355 | ||
4356 | return found_any; | |
4357 | } | |
4358 | ||
4359 | /* Updates the following BDS fields: | |
4360 | * - exact_filename: A filename which may be used for opening a block device | |
4361 | * which (mostly) equals the given BDS (even without any | |
4362 | * other options; so reading and writing must return the same | |
4363 | * results, but caching etc. may be different) | |
4364 | * - full_open_options: Options which, when given when opening a block device | |
4365 | * (without a filename), result in a BDS (mostly) | |
4366 | * equalling the given one | |
4367 | * - filename: If exact_filename is set, it is copied here. Otherwise, | |
4368 | * full_open_options is converted to a JSON object, prefixed with | |
4369 | * "json:" (for use through the JSON pseudo protocol) and put here. | |
4370 | */ | |
4371 | void bdrv_refresh_filename(BlockDriverState *bs) | |
4372 | { | |
4373 | BlockDriver *drv = bs->drv; | |
4374 | QDict *opts; | |
4375 | ||
4376 | if (!drv) { | |
4377 | return; | |
4378 | } | |
4379 | ||
4380 | /* This BDS's file name will most probably depend on its file's name, so | |
4381 | * refresh that first */ | |
4382 | if (bs->file) { | |
4383 | bdrv_refresh_filename(bs->file->bs); | |
4384 | } | |
4385 | ||
4386 | if (drv->bdrv_refresh_filename) { | |
4387 | /* Obsolete information is of no use here, so drop the old file name | |
4388 | * information before refreshing it */ | |
4389 | bs->exact_filename[0] = '\0'; | |
4390 | if (bs->full_open_options) { | |
4391 | QDECREF(bs->full_open_options); | |
4392 | bs->full_open_options = NULL; | |
4393 | } | |
4394 | ||
4395 | opts = qdict_new(); | |
4396 | append_open_options(opts, bs); | |
4397 | drv->bdrv_refresh_filename(bs, opts); | |
4398 | QDECREF(opts); | |
4399 | } else if (bs->file) { | |
4400 | /* Try to reconstruct valid information from the underlying file */ | |
4401 | bool has_open_options; | |
4402 | ||
4403 | bs->exact_filename[0] = '\0'; | |
4404 | if (bs->full_open_options) { | |
4405 | QDECREF(bs->full_open_options); | |
4406 | bs->full_open_options = NULL; | |
4407 | } | |
4408 | ||
4409 | opts = qdict_new(); | |
4410 | has_open_options = append_open_options(opts, bs); | |
4411 | ||
4412 | /* If no specific options have been given for this BDS, the filename of | |
4413 | * the underlying file should suffice for this one as well */ | |
4414 | if (bs->file->bs->exact_filename[0] && !has_open_options) { | |
4415 | strcpy(bs->exact_filename, bs->file->bs->exact_filename); | |
4416 | } | |
4417 | /* Reconstructing the full options QDict is simple for most format block | |
4418 | * drivers, as long as the full options are known for the underlying | |
4419 | * file BDS. The full options QDict of that file BDS should somehow | |
4420 | * contain a representation of the filename, therefore the following | |
4421 | * suffices without querying the (exact_)filename of this BDS. */ | |
4422 | if (bs->file->bs->full_open_options) { | |
4423 | qdict_put_obj(opts, "driver", | |
4424 | QOBJECT(qstring_from_str(drv->format_name))); | |
4425 | QINCREF(bs->file->bs->full_open_options); | |
4426 | qdict_put_obj(opts, "file", | |
4427 | QOBJECT(bs->file->bs->full_open_options)); | |
4428 | ||
4429 | bs->full_open_options = opts; | |
4430 | } else { | |
4431 | QDECREF(opts); | |
4432 | } | |
4433 | } else if (!bs->full_open_options && qdict_size(bs->options)) { | |
4434 | /* There is no underlying file BDS (at least referenced by BDS.file), | |
4435 | * so the full options QDict should be equal to the options given | |
4436 | * specifically for this block device when it was opened (plus the | |
4437 | * driver specification). | |
4438 | * Because those options don't change, there is no need to update | |
4439 | * full_open_options when it's already set. */ | |
4440 | ||
4441 | opts = qdict_new(); | |
4442 | append_open_options(opts, bs); | |
4443 | qdict_put_obj(opts, "driver", | |
4444 | QOBJECT(qstring_from_str(drv->format_name))); | |
4445 | ||
4446 | if (bs->exact_filename[0]) { | |
4447 | /* This may not work for all block protocol drivers (some may | |
4448 | * require this filename to be parsed), but we have to find some | |
4449 | * default solution here, so just include it. If some block driver | |
4450 | * does not support pure options without any filename at all or | |
4451 | * needs some special format of the options QDict, it needs to | |
4452 | * implement the driver-specific bdrv_refresh_filename() function. | |
4453 | */ | |
4454 | qdict_put_obj(opts, "filename", | |
4455 | QOBJECT(qstring_from_str(bs->exact_filename))); | |
4456 | } | |
4457 | ||
4458 | bs->full_open_options = opts; | |
4459 | } | |
4460 | ||
4461 | if (bs->exact_filename[0]) { | |
4462 | pstrcpy(bs->filename, sizeof(bs->filename), bs->exact_filename); | |
4463 | } else if (bs->full_open_options) { | |
4464 | QString *json = qobject_to_json(QOBJECT(bs->full_open_options)); | |
4465 | snprintf(bs->filename, sizeof(bs->filename), "json:%s", | |
4466 | qstring_get_str(json)); | |
4467 | QDECREF(json); | |
4468 | } | |
4469 | } | |
4470 | ||
4471 | /* | |
4472 | * Hot add/remove a BDS's child. So the user can take a child offline when | |
4473 | * it is broken and take a new child online | |
4474 | */ | |
4475 | void bdrv_add_child(BlockDriverState *parent_bs, BlockDriverState *child_bs, | |
4476 | Error **errp) | |
4477 | { | |
4478 | ||
4479 | if (!parent_bs->drv || !parent_bs->drv->bdrv_add_child) { | |
4480 | error_setg(errp, "The node %s does not support adding a child", | |
4481 | bdrv_get_device_or_node_name(parent_bs)); | |
4482 | return; | |
4483 | } | |
4484 | ||
4485 | if (!QLIST_EMPTY(&child_bs->parents)) { | |
4486 | error_setg(errp, "The node %s already has a parent", | |
4487 | child_bs->node_name); | |
4488 | return; | |
4489 | } | |
4490 | ||
4491 | parent_bs->drv->bdrv_add_child(parent_bs, child_bs, errp); | |
4492 | } | |
4493 | ||
4494 | void bdrv_del_child(BlockDriverState *parent_bs, BdrvChild *child, Error **errp) | |
4495 | { | |
4496 | BdrvChild *tmp; | |
4497 | ||
4498 | if (!parent_bs->drv || !parent_bs->drv->bdrv_del_child) { | |
4499 | error_setg(errp, "The node %s does not support removing a child", | |
4500 | bdrv_get_device_or_node_name(parent_bs)); | |
4501 | return; | |
4502 | } | |
4503 | ||
4504 | QLIST_FOREACH(tmp, &parent_bs->children, next) { | |
4505 | if (tmp == child) { | |
4506 | break; | |
4507 | } | |
4508 | } | |
4509 | ||
4510 | if (!tmp) { | |
4511 | error_setg(errp, "The node %s does not have a child named %s", | |
4512 | bdrv_get_device_or_node_name(parent_bs), | |
4513 | bdrv_get_device_or_node_name(child->bs)); | |
4514 | return; | |
4515 | } | |
4516 | ||
4517 | parent_bs->drv->bdrv_del_child(parent_bs, child, errp); | |
4518 | } |