]> git.proxmox.com Git - mirror_qemu.git/blob - block.c
c043f5eadd209396167c6e629184bd14f21a7d8b
[mirror_qemu.git] / block.c
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
25 #include "qemu/osdep.h"
26 #include "block/trace.h"
27 #include "block/block_int.h"
28 #include "block/blockjob.h"
29 #include "block/nbd.h"
30 #include "block/qdict.h"
31 #include "qemu/error-report.h"
32 #include "module_block.h"
33 #include "qemu/module.h"
34 #include "qapi/error.h"
35 #include "qapi/qmp/qdict.h"
36 #include "qapi/qmp/qjson.h"
37 #include "qapi/qmp/qnull.h"
38 #include "qapi/qmp/qstring.h"
39 #include "qapi/qobject-output-visitor.h"
40 #include "qapi/qapi-visit-block-core.h"
41 #include "sysemu/block-backend.h"
42 #include "sysemu/sysemu.h"
43 #include "qemu/notify.h"
44 #include "qemu/option.h"
45 #include "qemu/coroutine.h"
46 #include "block/qapi.h"
47 #include "qemu/timer.h"
48 #include "qemu/cutils.h"
49 #include "qemu/id.h"
50
51 #ifdef CONFIG_BSD
52 #include <sys/ioctl.h>
53 #include <sys/queue.h>
54 #ifndef __DragonFly__
55 #include <sys/disk.h>
56 #endif
57 #endif
58
59 #ifdef _WIN32
60 #include <windows.h>
61 #endif
62
63 #define NOT_DONE 0x7fffffff /* used while emulated sync operation in progress */
64
65 static QTAILQ_HEAD(, BlockDriverState) graph_bdrv_states =
66 QTAILQ_HEAD_INITIALIZER(graph_bdrv_states);
67
68 static QTAILQ_HEAD(, BlockDriverState) all_bdrv_states =
69 QTAILQ_HEAD_INITIALIZER(all_bdrv_states);
70
71 static QLIST_HEAD(, BlockDriver) bdrv_drivers =
72 QLIST_HEAD_INITIALIZER(bdrv_drivers);
73
74 static BlockDriverState *bdrv_open_inherit(const char *filename,
75 const char *reference,
76 QDict *options, int flags,
77 BlockDriverState *parent,
78 const BdrvChildRole *child_role,
79 Error **errp);
80
81 /* If non-zero, use only whitelisted block drivers */
82 static int use_bdrv_whitelist;
83
84 #ifdef _WIN32
85 static int is_windows_drive_prefix(const char *filename)
86 {
87 return (((filename[0] >= 'a' && filename[0] <= 'z') ||
88 (filename[0] >= 'A' && filename[0] <= 'Z')) &&
89 filename[1] == ':');
90 }
91
92 int is_windows_drive(const char *filename)
93 {
94 if (is_windows_drive_prefix(filename) &&
95 filename[2] == '\0')
96 return 1;
97 if (strstart(filename, "\\\\.\\", NULL) ||
98 strstart(filename, "//./", NULL))
99 return 1;
100 return 0;
101 }
102 #endif
103
104 size_t bdrv_opt_mem_align(BlockDriverState *bs)
105 {
106 if (!bs || !bs->drv) {
107 /* page size or 4k (hdd sector size) should be on the safe side */
108 return MAX(4096, getpagesize());
109 }
110
111 return bs->bl.opt_mem_alignment;
112 }
113
114 size_t bdrv_min_mem_align(BlockDriverState *bs)
115 {
116 if (!bs || !bs->drv) {
117 /* page size or 4k (hdd sector size) should be on the safe side */
118 return MAX(4096, getpagesize());
119 }
120
121 return bs->bl.min_mem_alignment;
122 }
123
124 /* check if the path starts with "<protocol>:" */
125 int path_has_protocol(const char *path)
126 {
127 const char *p;
128
129 #ifdef _WIN32
130 if (is_windows_drive(path) ||
131 is_windows_drive_prefix(path)) {
132 return 0;
133 }
134 p = path + strcspn(path, ":/\\");
135 #else
136 p = path + strcspn(path, ":/");
137 #endif
138
139 return *p == ':';
140 }
141
142 int path_is_absolute(const char *path)
143 {
144 #ifdef _WIN32
145 /* specific case for names like: "\\.\d:" */
146 if (is_windows_drive(path) || is_windows_drive_prefix(path)) {
147 return 1;
148 }
149 return (*path == '/' || *path == '\\');
150 #else
151 return (*path == '/');
152 #endif
153 }
154
155 /* if filename is absolute, just return its duplicate. Otherwise, build a
156 path to it by considering it is relative to base_path. URL are
157 supported. */
158 char *path_combine(const char *base_path, const char *filename)
159 {
160 const char *protocol_stripped = NULL;
161 const char *p, *p1;
162 char *result;
163 int len;
164
165 if (path_is_absolute(filename)) {
166 return g_strdup(filename);
167 }
168
169 if (path_has_protocol(base_path)) {
170 protocol_stripped = strchr(base_path, ':');
171 if (protocol_stripped) {
172 protocol_stripped++;
173 }
174 }
175 p = protocol_stripped ?: base_path;
176
177 p1 = strrchr(base_path, '/');
178 #ifdef _WIN32
179 {
180 const char *p2;
181 p2 = strrchr(base_path, '\\');
182 if (!p1 || p2 > p1) {
183 p1 = p2;
184 }
185 }
186 #endif
187 if (p1) {
188 p1++;
189 } else {
190 p1 = base_path;
191 }
192 if (p1 > p) {
193 p = p1;
194 }
195 len = p - base_path;
196
197 result = g_malloc(len + strlen(filename) + 1);
198 memcpy(result, base_path, len);
199 strcpy(result + len, filename);
200
201 return result;
202 }
203
204 /*
205 * Helper function for bdrv_parse_filename() implementations to remove optional
206 * protocol prefixes (especially "file:") from a filename and for putting the
207 * stripped filename into the options QDict if there is such a prefix.
208 */
209 void bdrv_parse_filename_strip_prefix(const char *filename, const char *prefix,
210 QDict *options)
211 {
212 if (strstart(filename, prefix, &filename)) {
213 /* Stripping the explicit protocol prefix may result in a protocol
214 * prefix being (wrongly) detected (if the filename contains a colon) */
215 if (path_has_protocol(filename)) {
216 QString *fat_filename;
217
218 /* This means there is some colon before the first slash; therefore,
219 * this cannot be an absolute path */
220 assert(!path_is_absolute(filename));
221
222 /* And we can thus fix the protocol detection issue by prefixing it
223 * by "./" */
224 fat_filename = qstring_from_str("./");
225 qstring_append(fat_filename, filename);
226
227 assert(!path_has_protocol(qstring_get_str(fat_filename)));
228
229 qdict_put(options, "filename", fat_filename);
230 } else {
231 /* If no protocol prefix was detected, we can use the shortened
232 * filename as-is */
233 qdict_put_str(options, "filename", filename);
234 }
235 }
236 }
237
238
239 /* Returns whether the image file is opened as read-only. Note that this can
240 * return false and writing to the image file is still not possible because the
241 * image is inactivated. */
242 bool bdrv_is_read_only(BlockDriverState *bs)
243 {
244 return bs->read_only;
245 }
246
247 int bdrv_can_set_read_only(BlockDriverState *bs, bool read_only,
248 bool ignore_allow_rdw, Error **errp)
249 {
250 /* Do not set read_only if copy_on_read is enabled */
251 if (bs->copy_on_read && read_only) {
252 error_setg(errp, "Can't set node '%s' to r/o with copy-on-read enabled",
253 bdrv_get_device_or_node_name(bs));
254 return -EINVAL;
255 }
256
257 /* Do not clear read_only if it is prohibited */
258 if (!read_only && !(bs->open_flags & BDRV_O_ALLOW_RDWR) &&
259 !ignore_allow_rdw)
260 {
261 error_setg(errp, "Node '%s' is read only",
262 bdrv_get_device_or_node_name(bs));
263 return -EPERM;
264 }
265
266 return 0;
267 }
268
269 /*
270 * Called by a driver that can only provide a read-only image.
271 *
272 * Returns 0 if the node is already read-only or it could switch the node to
273 * read-only because BDRV_O_AUTO_RDONLY is set.
274 *
275 * Returns -EACCES if the node is read-write and BDRV_O_AUTO_RDONLY is not set
276 * or bdrv_can_set_read_only() forbids making the node read-only. If @errmsg
277 * is not NULL, it is used as the error message for the Error object.
278 */
279 int bdrv_apply_auto_read_only(BlockDriverState *bs, const char *errmsg,
280 Error **errp)
281 {
282 int ret = 0;
283
284 if (!(bs->open_flags & BDRV_O_RDWR)) {
285 return 0;
286 }
287 if (!(bs->open_flags & BDRV_O_AUTO_RDONLY)) {
288 goto fail;
289 }
290
291 ret = bdrv_can_set_read_only(bs, true, false, NULL);
292 if (ret < 0) {
293 goto fail;
294 }
295
296 bs->read_only = true;
297 bs->open_flags &= ~BDRV_O_RDWR;
298
299 return 0;
300
301 fail:
302 error_setg(errp, "%s", errmsg ?: "Image is read-only");
303 return -EACCES;
304 }
305
306 /*
307 * If @backing is empty, this function returns NULL without setting
308 * @errp. In all other cases, NULL will only be returned with @errp
309 * set.
310 *
311 * Therefore, a return value of NULL without @errp set means that
312 * there is no backing file; if @errp is set, there is one but its
313 * absolute filename cannot be generated.
314 */
315 char *bdrv_get_full_backing_filename_from_filename(const char *backed,
316 const char *backing,
317 Error **errp)
318 {
319 if (backing[0] == '\0') {
320 return NULL;
321 } else if (path_has_protocol(backing) || path_is_absolute(backing)) {
322 return g_strdup(backing);
323 } else if (backed[0] == '\0' || strstart(backed, "json:", NULL)) {
324 error_setg(errp, "Cannot use relative backing file names for '%s'",
325 backed);
326 return NULL;
327 } else {
328 return path_combine(backed, backing);
329 }
330 }
331
332 /*
333 * If @filename is empty or NULL, this function returns NULL without
334 * setting @errp. In all other cases, NULL will only be returned with
335 * @errp set.
336 */
337 static char *bdrv_make_absolute_filename(BlockDriverState *relative_to,
338 const char *filename, Error **errp)
339 {
340 char *dir, *full_name;
341
342 if (!filename || filename[0] == '\0') {
343 return NULL;
344 } else if (path_has_protocol(filename) || path_is_absolute(filename)) {
345 return g_strdup(filename);
346 }
347
348 dir = bdrv_dirname(relative_to, errp);
349 if (!dir) {
350 return NULL;
351 }
352
353 full_name = g_strconcat(dir, filename, NULL);
354 g_free(dir);
355 return full_name;
356 }
357
358 char *bdrv_get_full_backing_filename(BlockDriverState *bs, Error **errp)
359 {
360 return bdrv_make_absolute_filename(bs, bs->backing_file, errp);
361 }
362
363 void bdrv_register(BlockDriver *bdrv)
364 {
365 QLIST_INSERT_HEAD(&bdrv_drivers, bdrv, list);
366 }
367
368 BlockDriverState *bdrv_new(void)
369 {
370 BlockDriverState *bs;
371 int i;
372
373 bs = g_new0(BlockDriverState, 1);
374 QLIST_INIT(&bs->dirty_bitmaps);
375 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
376 QLIST_INIT(&bs->op_blockers[i]);
377 }
378 notifier_with_return_list_init(&bs->before_write_notifiers);
379 qemu_co_mutex_init(&bs->reqs_lock);
380 qemu_mutex_init(&bs->dirty_bitmap_mutex);
381 bs->refcnt = 1;
382 bs->aio_context = qemu_get_aio_context();
383
384 qemu_co_queue_init(&bs->flush_queue);
385
386 for (i = 0; i < bdrv_drain_all_count; i++) {
387 bdrv_drained_begin(bs);
388 }
389
390 QTAILQ_INSERT_TAIL(&all_bdrv_states, bs, bs_list);
391
392 return bs;
393 }
394
395 static BlockDriver *bdrv_do_find_format(const char *format_name)
396 {
397 BlockDriver *drv1;
398
399 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
400 if (!strcmp(drv1->format_name, format_name)) {
401 return drv1;
402 }
403 }
404
405 return NULL;
406 }
407
408 BlockDriver *bdrv_find_format(const char *format_name)
409 {
410 BlockDriver *drv1;
411 int i;
412
413 drv1 = bdrv_do_find_format(format_name);
414 if (drv1) {
415 return drv1;
416 }
417
418 /* The driver isn't registered, maybe we need to load a module */
419 for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); ++i) {
420 if (!strcmp(block_driver_modules[i].format_name, format_name)) {
421 block_module_load_one(block_driver_modules[i].library_name);
422 break;
423 }
424 }
425
426 return bdrv_do_find_format(format_name);
427 }
428
429 static int bdrv_format_is_whitelisted(const char *format_name, bool read_only)
430 {
431 static const char *whitelist_rw[] = {
432 CONFIG_BDRV_RW_WHITELIST
433 };
434 static const char *whitelist_ro[] = {
435 CONFIG_BDRV_RO_WHITELIST
436 };
437 const char **p;
438
439 if (!whitelist_rw[0] && !whitelist_ro[0]) {
440 return 1; /* no whitelist, anything goes */
441 }
442
443 for (p = whitelist_rw; *p; p++) {
444 if (!strcmp(format_name, *p)) {
445 return 1;
446 }
447 }
448 if (read_only) {
449 for (p = whitelist_ro; *p; p++) {
450 if (!strcmp(format_name, *p)) {
451 return 1;
452 }
453 }
454 }
455 return 0;
456 }
457
458 int bdrv_is_whitelisted(BlockDriver *drv, bool read_only)
459 {
460 return bdrv_format_is_whitelisted(drv->format_name, read_only);
461 }
462
463 bool bdrv_uses_whitelist(void)
464 {
465 return use_bdrv_whitelist;
466 }
467
468 typedef struct CreateCo {
469 BlockDriver *drv;
470 char *filename;
471 QemuOpts *opts;
472 int ret;
473 Error *err;
474 } CreateCo;
475
476 static void coroutine_fn bdrv_create_co_entry(void *opaque)
477 {
478 Error *local_err = NULL;
479 int ret;
480
481 CreateCo *cco = opaque;
482 assert(cco->drv);
483
484 ret = cco->drv->bdrv_co_create_opts(cco->filename, cco->opts, &local_err);
485 error_propagate(&cco->err, local_err);
486 cco->ret = ret;
487 }
488
489 int bdrv_create(BlockDriver *drv, const char* filename,
490 QemuOpts *opts, Error **errp)
491 {
492 int ret;
493
494 Coroutine *co;
495 CreateCo cco = {
496 .drv = drv,
497 .filename = g_strdup(filename),
498 .opts = opts,
499 .ret = NOT_DONE,
500 .err = NULL,
501 };
502
503 if (!drv->bdrv_co_create_opts) {
504 error_setg(errp, "Driver '%s' does not support image creation", drv->format_name);
505 ret = -ENOTSUP;
506 goto out;
507 }
508
509 if (qemu_in_coroutine()) {
510 /* Fast-path if already in coroutine context */
511 bdrv_create_co_entry(&cco);
512 } else {
513 co = qemu_coroutine_create(bdrv_create_co_entry, &cco);
514 qemu_coroutine_enter(co);
515 while (cco.ret == NOT_DONE) {
516 aio_poll(qemu_get_aio_context(), true);
517 }
518 }
519
520 ret = cco.ret;
521 if (ret < 0) {
522 if (cco.err) {
523 error_propagate(errp, cco.err);
524 } else {
525 error_setg_errno(errp, -ret, "Could not create image");
526 }
527 }
528
529 out:
530 g_free(cco.filename);
531 return ret;
532 }
533
534 int bdrv_create_file(const char *filename, QemuOpts *opts, Error **errp)
535 {
536 BlockDriver *drv;
537 Error *local_err = NULL;
538 int ret;
539
540 drv = bdrv_find_protocol(filename, true, errp);
541 if (drv == NULL) {
542 return -ENOENT;
543 }
544
545 ret = bdrv_create(drv, filename, opts, &local_err);
546 error_propagate(errp, local_err);
547 return ret;
548 }
549
550 /**
551 * Try to get @bs's logical and physical block size.
552 * On success, store them in @bsz struct and return 0.
553 * On failure return -errno.
554 * @bs must not be empty.
555 */
556 int bdrv_probe_blocksizes(BlockDriverState *bs, BlockSizes *bsz)
557 {
558 BlockDriver *drv = bs->drv;
559
560 if (drv && drv->bdrv_probe_blocksizes) {
561 return drv->bdrv_probe_blocksizes(bs, bsz);
562 } else if (drv && drv->is_filter && bs->file) {
563 return bdrv_probe_blocksizes(bs->file->bs, bsz);
564 }
565
566 return -ENOTSUP;
567 }
568
569 /**
570 * Try to get @bs's geometry (cyls, heads, sectors).
571 * On success, store them in @geo struct and return 0.
572 * On failure return -errno.
573 * @bs must not be empty.
574 */
575 int bdrv_probe_geometry(BlockDriverState *bs, HDGeometry *geo)
576 {
577 BlockDriver *drv = bs->drv;
578
579 if (drv && drv->bdrv_probe_geometry) {
580 return drv->bdrv_probe_geometry(bs, geo);
581 } else if (drv && drv->is_filter && bs->file) {
582 return bdrv_probe_geometry(bs->file->bs, geo);
583 }
584
585 return -ENOTSUP;
586 }
587
588 /*
589 * Create a uniquely-named empty temporary file.
590 * Return 0 upon success, otherwise a negative errno value.
591 */
592 int get_tmp_filename(char *filename, int size)
593 {
594 #ifdef _WIN32
595 char temp_dir[MAX_PATH];
596 /* GetTempFileName requires that its output buffer (4th param)
597 have length MAX_PATH or greater. */
598 assert(size >= MAX_PATH);
599 return (GetTempPath(MAX_PATH, temp_dir)
600 && GetTempFileName(temp_dir, "qem", 0, filename)
601 ? 0 : -GetLastError());
602 #else
603 int fd;
604 const char *tmpdir;
605 tmpdir = getenv("TMPDIR");
606 if (!tmpdir) {
607 tmpdir = "/var/tmp";
608 }
609 if (snprintf(filename, size, "%s/vl.XXXXXX", tmpdir) >= size) {
610 return -EOVERFLOW;
611 }
612 fd = mkstemp(filename);
613 if (fd < 0) {
614 return -errno;
615 }
616 if (close(fd) != 0) {
617 unlink(filename);
618 return -errno;
619 }
620 return 0;
621 #endif
622 }
623
624 /*
625 * Detect host devices. By convention, /dev/cdrom[N] is always
626 * recognized as a host CDROM.
627 */
628 static BlockDriver *find_hdev_driver(const char *filename)
629 {
630 int score_max = 0, score;
631 BlockDriver *drv = NULL, *d;
632
633 QLIST_FOREACH(d, &bdrv_drivers, list) {
634 if (d->bdrv_probe_device) {
635 score = d->bdrv_probe_device(filename);
636 if (score > score_max) {
637 score_max = score;
638 drv = d;
639 }
640 }
641 }
642
643 return drv;
644 }
645
646 static BlockDriver *bdrv_do_find_protocol(const char *protocol)
647 {
648 BlockDriver *drv1;
649
650 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
651 if (drv1->protocol_name && !strcmp(drv1->protocol_name, protocol)) {
652 return drv1;
653 }
654 }
655
656 return NULL;
657 }
658
659 BlockDriver *bdrv_find_protocol(const char *filename,
660 bool allow_protocol_prefix,
661 Error **errp)
662 {
663 BlockDriver *drv1;
664 char protocol[128];
665 int len;
666 const char *p;
667 int i;
668
669 /* TODO Drivers without bdrv_file_open must be specified explicitly */
670
671 /*
672 * XXX(hch): we really should not let host device detection
673 * override an explicit protocol specification, but moving this
674 * later breaks access to device names with colons in them.
675 * Thanks to the brain-dead persistent naming schemes on udev-
676 * based Linux systems those actually are quite common.
677 */
678 drv1 = find_hdev_driver(filename);
679 if (drv1) {
680 return drv1;
681 }
682
683 if (!path_has_protocol(filename) || !allow_protocol_prefix) {
684 return &bdrv_file;
685 }
686
687 p = strchr(filename, ':');
688 assert(p != NULL);
689 len = p - filename;
690 if (len > sizeof(protocol) - 1)
691 len = sizeof(protocol) - 1;
692 memcpy(protocol, filename, len);
693 protocol[len] = '\0';
694
695 drv1 = bdrv_do_find_protocol(protocol);
696 if (drv1) {
697 return drv1;
698 }
699
700 for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); ++i) {
701 if (block_driver_modules[i].protocol_name &&
702 !strcmp(block_driver_modules[i].protocol_name, protocol)) {
703 block_module_load_one(block_driver_modules[i].library_name);
704 break;
705 }
706 }
707
708 drv1 = bdrv_do_find_protocol(protocol);
709 if (!drv1) {
710 error_setg(errp, "Unknown protocol '%s'", protocol);
711 }
712 return drv1;
713 }
714
715 /*
716 * Guess image format by probing its contents.
717 * This is not a good idea when your image is raw (CVE-2008-2004), but
718 * we do it anyway for backward compatibility.
719 *
720 * @buf contains the image's first @buf_size bytes.
721 * @buf_size is the buffer size in bytes (generally BLOCK_PROBE_BUF_SIZE,
722 * but can be smaller if the image file is smaller)
723 * @filename is its filename.
724 *
725 * For all block drivers, call the bdrv_probe() method to get its
726 * probing score.
727 * Return the first block driver with the highest probing score.
728 */
729 BlockDriver *bdrv_probe_all(const uint8_t *buf, int buf_size,
730 const char *filename)
731 {
732 int score_max = 0, score;
733 BlockDriver *drv = NULL, *d;
734
735 QLIST_FOREACH(d, &bdrv_drivers, list) {
736 if (d->bdrv_probe) {
737 score = d->bdrv_probe(buf, buf_size, filename);
738 if (score > score_max) {
739 score_max = score;
740 drv = d;
741 }
742 }
743 }
744
745 return drv;
746 }
747
748 static int find_image_format(BlockBackend *file, const char *filename,
749 BlockDriver **pdrv, Error **errp)
750 {
751 BlockDriver *drv;
752 uint8_t buf[BLOCK_PROBE_BUF_SIZE];
753 int ret = 0;
754
755 /* Return the raw BlockDriver * to scsi-generic devices or empty drives */
756 if (blk_is_sg(file) || !blk_is_inserted(file) || blk_getlength(file) == 0) {
757 *pdrv = &bdrv_raw;
758 return ret;
759 }
760
761 ret = blk_pread(file, 0, buf, sizeof(buf));
762 if (ret < 0) {
763 error_setg_errno(errp, -ret, "Could not read image for determining its "
764 "format");
765 *pdrv = NULL;
766 return ret;
767 }
768
769 drv = bdrv_probe_all(buf, ret, filename);
770 if (!drv) {
771 error_setg(errp, "Could not determine image format: No compatible "
772 "driver found");
773 ret = -ENOENT;
774 }
775 *pdrv = drv;
776 return ret;
777 }
778
779 /**
780 * Set the current 'total_sectors' value
781 * Return 0 on success, -errno on error.
782 */
783 int refresh_total_sectors(BlockDriverState *bs, int64_t hint)
784 {
785 BlockDriver *drv = bs->drv;
786
787 if (!drv) {
788 return -ENOMEDIUM;
789 }
790
791 /* Do not attempt drv->bdrv_getlength() on scsi-generic devices */
792 if (bdrv_is_sg(bs))
793 return 0;
794
795 /* query actual device if possible, otherwise just trust the hint */
796 if (drv->bdrv_getlength) {
797 int64_t length = drv->bdrv_getlength(bs);
798 if (length < 0) {
799 return length;
800 }
801 hint = DIV_ROUND_UP(length, BDRV_SECTOR_SIZE);
802 }
803
804 bs->total_sectors = hint;
805 return 0;
806 }
807
808 /**
809 * Combines a QDict of new block driver @options with any missing options taken
810 * from @old_options, so that leaving out an option defaults to its old value.
811 */
812 static void bdrv_join_options(BlockDriverState *bs, QDict *options,
813 QDict *old_options)
814 {
815 if (bs->drv && bs->drv->bdrv_join_options) {
816 bs->drv->bdrv_join_options(options, old_options);
817 } else {
818 qdict_join(options, old_options, false);
819 }
820 }
821
822 static BlockdevDetectZeroesOptions bdrv_parse_detect_zeroes(QemuOpts *opts,
823 int open_flags,
824 Error **errp)
825 {
826 Error *local_err = NULL;
827 char *value = qemu_opt_get_del(opts, "detect-zeroes");
828 BlockdevDetectZeroesOptions detect_zeroes =
829 qapi_enum_parse(&BlockdevDetectZeroesOptions_lookup, value,
830 BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF, &local_err);
831 g_free(value);
832 if (local_err) {
833 error_propagate(errp, local_err);
834 return detect_zeroes;
835 }
836
837 if (detect_zeroes == BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP &&
838 !(open_flags & BDRV_O_UNMAP))
839 {
840 error_setg(errp, "setting detect-zeroes to unmap is not allowed "
841 "without setting discard operation to unmap");
842 }
843
844 return detect_zeroes;
845 }
846
847 /**
848 * Set open flags for a given discard mode
849 *
850 * Return 0 on success, -1 if the discard mode was invalid.
851 */
852 int bdrv_parse_discard_flags(const char *mode, int *flags)
853 {
854 *flags &= ~BDRV_O_UNMAP;
855
856 if (!strcmp(mode, "off") || !strcmp(mode, "ignore")) {
857 /* do nothing */
858 } else if (!strcmp(mode, "on") || !strcmp(mode, "unmap")) {
859 *flags |= BDRV_O_UNMAP;
860 } else {
861 return -1;
862 }
863
864 return 0;
865 }
866
867 /**
868 * Set open flags for a given cache mode
869 *
870 * Return 0 on success, -1 if the cache mode was invalid.
871 */
872 int bdrv_parse_cache_mode(const char *mode, int *flags, bool *writethrough)
873 {
874 *flags &= ~BDRV_O_CACHE_MASK;
875
876 if (!strcmp(mode, "off") || !strcmp(mode, "none")) {
877 *writethrough = false;
878 *flags |= BDRV_O_NOCACHE;
879 } else if (!strcmp(mode, "directsync")) {
880 *writethrough = true;
881 *flags |= BDRV_O_NOCACHE;
882 } else if (!strcmp(mode, "writeback")) {
883 *writethrough = false;
884 } else if (!strcmp(mode, "unsafe")) {
885 *writethrough = false;
886 *flags |= BDRV_O_NO_FLUSH;
887 } else if (!strcmp(mode, "writethrough")) {
888 *writethrough = true;
889 } else {
890 return -1;
891 }
892
893 return 0;
894 }
895
896 static char *bdrv_child_get_parent_desc(BdrvChild *c)
897 {
898 BlockDriverState *parent = c->opaque;
899 return g_strdup(bdrv_get_device_or_node_name(parent));
900 }
901
902 static void bdrv_child_cb_drained_begin(BdrvChild *child)
903 {
904 BlockDriverState *bs = child->opaque;
905 bdrv_do_drained_begin_quiesce(bs, NULL, false);
906 }
907
908 static bool bdrv_child_cb_drained_poll(BdrvChild *child)
909 {
910 BlockDriverState *bs = child->opaque;
911 return bdrv_drain_poll(bs, false, NULL, false);
912 }
913
914 static void bdrv_child_cb_drained_end(BdrvChild *child)
915 {
916 BlockDriverState *bs = child->opaque;
917 bdrv_drained_end(bs);
918 }
919
920 static void bdrv_child_cb_attach(BdrvChild *child)
921 {
922 BlockDriverState *bs = child->opaque;
923 bdrv_apply_subtree_drain(child, bs);
924 }
925
926 static void bdrv_child_cb_detach(BdrvChild *child)
927 {
928 BlockDriverState *bs = child->opaque;
929 bdrv_unapply_subtree_drain(child, bs);
930 }
931
932 static int bdrv_child_cb_inactivate(BdrvChild *child)
933 {
934 BlockDriverState *bs = child->opaque;
935 assert(bs->open_flags & BDRV_O_INACTIVE);
936 return 0;
937 }
938
939 /*
940 * Returns the options and flags that a temporary snapshot should get, based on
941 * the originally requested flags (the originally requested image will have
942 * flags like a backing file)
943 */
944 static void bdrv_temp_snapshot_options(int *child_flags, QDict *child_options,
945 int parent_flags, QDict *parent_options)
946 {
947 *child_flags = (parent_flags & ~BDRV_O_SNAPSHOT) | BDRV_O_TEMPORARY;
948
949 /* For temporary files, unconditional cache=unsafe is fine */
950 qdict_set_default_str(child_options, BDRV_OPT_CACHE_DIRECT, "off");
951 qdict_set_default_str(child_options, BDRV_OPT_CACHE_NO_FLUSH, "on");
952
953 /* Copy the read-only option from the parent */
954 qdict_copy_default(child_options, parent_options, BDRV_OPT_READ_ONLY);
955
956 /* aio=native doesn't work for cache.direct=off, so disable it for the
957 * temporary snapshot */
958 *child_flags &= ~BDRV_O_NATIVE_AIO;
959 }
960
961 /*
962 * Returns the options and flags that bs->file should get if a protocol driver
963 * is expected, based on the given options and flags for the parent BDS
964 */
965 static void bdrv_inherited_options(int *child_flags, QDict *child_options,
966 int parent_flags, QDict *parent_options)
967 {
968 int flags = parent_flags;
969
970 /* Enable protocol handling, disable format probing for bs->file */
971 flags |= BDRV_O_PROTOCOL;
972
973 /* If the cache mode isn't explicitly set, inherit direct and no-flush from
974 * the parent. */
975 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_DIRECT);
976 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_NO_FLUSH);
977 qdict_copy_default(child_options, parent_options, BDRV_OPT_FORCE_SHARE);
978
979 /* Inherit the read-only option from the parent if it's not set */
980 qdict_copy_default(child_options, parent_options, BDRV_OPT_READ_ONLY);
981 qdict_copy_default(child_options, parent_options, BDRV_OPT_AUTO_READ_ONLY);
982
983 /* Our block drivers take care to send flushes and respect unmap policy,
984 * so we can default to enable both on lower layers regardless of the
985 * corresponding parent options. */
986 qdict_set_default_str(child_options, BDRV_OPT_DISCARD, "unmap");
987
988 /* Clear flags that only apply to the top layer */
989 flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_COPY_ON_READ |
990 BDRV_O_NO_IO);
991
992 *child_flags = flags;
993 }
994
995 const BdrvChildRole child_file = {
996 .parent_is_bds = true,
997 .get_parent_desc = bdrv_child_get_parent_desc,
998 .inherit_options = bdrv_inherited_options,
999 .drained_begin = bdrv_child_cb_drained_begin,
1000 .drained_poll = bdrv_child_cb_drained_poll,
1001 .drained_end = bdrv_child_cb_drained_end,
1002 .attach = bdrv_child_cb_attach,
1003 .detach = bdrv_child_cb_detach,
1004 .inactivate = bdrv_child_cb_inactivate,
1005 };
1006
1007 /*
1008 * Returns the options and flags that bs->file should get if the use of formats
1009 * (and not only protocols) is permitted for it, based on the given options and
1010 * flags for the parent BDS
1011 */
1012 static void bdrv_inherited_fmt_options(int *child_flags, QDict *child_options,
1013 int parent_flags, QDict *parent_options)
1014 {
1015 child_file.inherit_options(child_flags, child_options,
1016 parent_flags, parent_options);
1017
1018 *child_flags &= ~(BDRV_O_PROTOCOL | BDRV_O_NO_IO);
1019 }
1020
1021 const BdrvChildRole child_format = {
1022 .parent_is_bds = true,
1023 .get_parent_desc = bdrv_child_get_parent_desc,
1024 .inherit_options = bdrv_inherited_fmt_options,
1025 .drained_begin = bdrv_child_cb_drained_begin,
1026 .drained_poll = bdrv_child_cb_drained_poll,
1027 .drained_end = bdrv_child_cb_drained_end,
1028 .attach = bdrv_child_cb_attach,
1029 .detach = bdrv_child_cb_detach,
1030 .inactivate = bdrv_child_cb_inactivate,
1031 };
1032
1033 static void bdrv_backing_attach(BdrvChild *c)
1034 {
1035 BlockDriverState *parent = c->opaque;
1036 BlockDriverState *backing_hd = c->bs;
1037
1038 assert(!parent->backing_blocker);
1039 error_setg(&parent->backing_blocker,
1040 "node is used as backing hd of '%s'",
1041 bdrv_get_device_or_node_name(parent));
1042
1043 bdrv_refresh_filename(backing_hd);
1044
1045 parent->open_flags &= ~BDRV_O_NO_BACKING;
1046 pstrcpy(parent->backing_file, sizeof(parent->backing_file),
1047 backing_hd->filename);
1048 pstrcpy(parent->backing_format, sizeof(parent->backing_format),
1049 backing_hd->drv ? backing_hd->drv->format_name : "");
1050
1051 bdrv_op_block_all(backing_hd, parent->backing_blocker);
1052 /* Otherwise we won't be able to commit or stream */
1053 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_COMMIT_TARGET,
1054 parent->backing_blocker);
1055 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_STREAM,
1056 parent->backing_blocker);
1057 /*
1058 * We do backup in 3 ways:
1059 * 1. drive backup
1060 * The target bs is new opened, and the source is top BDS
1061 * 2. blockdev backup
1062 * Both the source and the target are top BDSes.
1063 * 3. internal backup(used for block replication)
1064 * Both the source and the target are backing file
1065 *
1066 * In case 1 and 2, neither the source nor the target is the backing file.
1067 * In case 3, we will block the top BDS, so there is only one block job
1068 * for the top BDS and its backing chain.
1069 */
1070 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_BACKUP_SOURCE,
1071 parent->backing_blocker);
1072 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_BACKUP_TARGET,
1073 parent->backing_blocker);
1074
1075 bdrv_child_cb_attach(c);
1076 }
1077
1078 static void bdrv_backing_detach(BdrvChild *c)
1079 {
1080 BlockDriverState *parent = c->opaque;
1081
1082 assert(parent->backing_blocker);
1083 bdrv_op_unblock_all(c->bs, parent->backing_blocker);
1084 error_free(parent->backing_blocker);
1085 parent->backing_blocker = NULL;
1086
1087 bdrv_child_cb_detach(c);
1088 }
1089
1090 /*
1091 * Returns the options and flags that bs->backing should get, based on the
1092 * given options and flags for the parent BDS
1093 */
1094 static void bdrv_backing_options(int *child_flags, QDict *child_options,
1095 int parent_flags, QDict *parent_options)
1096 {
1097 int flags = parent_flags;
1098
1099 /* The cache mode is inherited unmodified for backing files; except WCE,
1100 * which is only applied on the top level (BlockBackend) */
1101 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_DIRECT);
1102 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_NO_FLUSH);
1103 qdict_copy_default(child_options, parent_options, BDRV_OPT_FORCE_SHARE);
1104
1105 /* backing files always opened read-only */
1106 qdict_set_default_str(child_options, BDRV_OPT_READ_ONLY, "on");
1107 qdict_set_default_str(child_options, BDRV_OPT_AUTO_READ_ONLY, "off");
1108 flags &= ~BDRV_O_COPY_ON_READ;
1109
1110 /* snapshot=on is handled on the top layer */
1111 flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_TEMPORARY);
1112
1113 *child_flags = flags;
1114 }
1115
1116 static int bdrv_backing_update_filename(BdrvChild *c, BlockDriverState *base,
1117 const char *filename, Error **errp)
1118 {
1119 BlockDriverState *parent = c->opaque;
1120 bool read_only = bdrv_is_read_only(parent);
1121 int ret;
1122
1123 if (read_only) {
1124 ret = bdrv_reopen_set_read_only(parent, false, errp);
1125 if (ret < 0) {
1126 return ret;
1127 }
1128 }
1129
1130 ret = bdrv_change_backing_file(parent, filename,
1131 base->drv ? base->drv->format_name : "");
1132 if (ret < 0) {
1133 error_setg_errno(errp, -ret, "Could not update backing file link");
1134 }
1135
1136 if (read_only) {
1137 bdrv_reopen_set_read_only(parent, true, NULL);
1138 }
1139
1140 return ret;
1141 }
1142
1143 const BdrvChildRole child_backing = {
1144 .parent_is_bds = true,
1145 .get_parent_desc = bdrv_child_get_parent_desc,
1146 .attach = bdrv_backing_attach,
1147 .detach = bdrv_backing_detach,
1148 .inherit_options = bdrv_backing_options,
1149 .drained_begin = bdrv_child_cb_drained_begin,
1150 .drained_poll = bdrv_child_cb_drained_poll,
1151 .drained_end = bdrv_child_cb_drained_end,
1152 .inactivate = bdrv_child_cb_inactivate,
1153 .update_filename = bdrv_backing_update_filename,
1154 };
1155
1156 static int bdrv_open_flags(BlockDriverState *bs, int flags)
1157 {
1158 int open_flags = flags;
1159
1160 /*
1161 * Clear flags that are internal to the block layer before opening the
1162 * image.
1163 */
1164 open_flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_PROTOCOL);
1165
1166 return open_flags;
1167 }
1168
1169 static void update_flags_from_options(int *flags, QemuOpts *opts)
1170 {
1171 *flags &= ~(BDRV_O_CACHE_MASK | BDRV_O_RDWR | BDRV_O_AUTO_RDONLY);
1172
1173 if (qemu_opt_get_bool_del(opts, BDRV_OPT_CACHE_NO_FLUSH, false)) {
1174 *flags |= BDRV_O_NO_FLUSH;
1175 }
1176
1177 if (qemu_opt_get_bool_del(opts, BDRV_OPT_CACHE_DIRECT, false)) {
1178 *flags |= BDRV_O_NOCACHE;
1179 }
1180
1181 if (!qemu_opt_get_bool_del(opts, BDRV_OPT_READ_ONLY, false)) {
1182 *flags |= BDRV_O_RDWR;
1183 }
1184
1185 if (qemu_opt_get_bool_del(opts, BDRV_OPT_AUTO_READ_ONLY, false)) {
1186 *flags |= BDRV_O_AUTO_RDONLY;
1187 }
1188 }
1189
1190 static void update_options_from_flags(QDict *options, int flags)
1191 {
1192 if (!qdict_haskey(options, BDRV_OPT_CACHE_DIRECT)) {
1193 qdict_put_bool(options, BDRV_OPT_CACHE_DIRECT, flags & BDRV_O_NOCACHE);
1194 }
1195 if (!qdict_haskey(options, BDRV_OPT_CACHE_NO_FLUSH)) {
1196 qdict_put_bool(options, BDRV_OPT_CACHE_NO_FLUSH,
1197 flags & BDRV_O_NO_FLUSH);
1198 }
1199 if (!qdict_haskey(options, BDRV_OPT_READ_ONLY)) {
1200 qdict_put_bool(options, BDRV_OPT_READ_ONLY, !(flags & BDRV_O_RDWR));
1201 }
1202 if (!qdict_haskey(options, BDRV_OPT_AUTO_READ_ONLY)) {
1203 qdict_put_bool(options, BDRV_OPT_AUTO_READ_ONLY,
1204 flags & BDRV_O_AUTO_RDONLY);
1205 }
1206 }
1207
1208 static void bdrv_assign_node_name(BlockDriverState *bs,
1209 const char *node_name,
1210 Error **errp)
1211 {
1212 char *gen_node_name = NULL;
1213
1214 if (!node_name) {
1215 node_name = gen_node_name = id_generate(ID_BLOCK);
1216 } else if (!id_wellformed(node_name)) {
1217 /*
1218 * Check for empty string or invalid characters, but not if it is
1219 * generated (generated names use characters not available to the user)
1220 */
1221 error_setg(errp, "Invalid node name");
1222 return;
1223 }
1224
1225 /* takes care of avoiding namespaces collisions */
1226 if (blk_by_name(node_name)) {
1227 error_setg(errp, "node-name=%s is conflicting with a device id",
1228 node_name);
1229 goto out;
1230 }
1231
1232 /* takes care of avoiding duplicates node names */
1233 if (bdrv_find_node(node_name)) {
1234 error_setg(errp, "Duplicate node name");
1235 goto out;
1236 }
1237
1238 /* Make sure that the node name isn't truncated */
1239 if (strlen(node_name) >= sizeof(bs->node_name)) {
1240 error_setg(errp, "Node name too long");
1241 goto out;
1242 }
1243
1244 /* copy node name into the bs and insert it into the graph list */
1245 pstrcpy(bs->node_name, sizeof(bs->node_name), node_name);
1246 QTAILQ_INSERT_TAIL(&graph_bdrv_states, bs, node_list);
1247 out:
1248 g_free(gen_node_name);
1249 }
1250
1251 static int bdrv_open_driver(BlockDriverState *bs, BlockDriver *drv,
1252 const char *node_name, QDict *options,
1253 int open_flags, Error **errp)
1254 {
1255 Error *local_err = NULL;
1256 int i, ret;
1257
1258 bdrv_assign_node_name(bs, node_name, &local_err);
1259 if (local_err) {
1260 error_propagate(errp, local_err);
1261 return -EINVAL;
1262 }
1263
1264 bs->drv = drv;
1265 bs->read_only = !(bs->open_flags & BDRV_O_RDWR);
1266 bs->opaque = g_malloc0(drv->instance_size);
1267
1268 if (drv->bdrv_file_open) {
1269 assert(!drv->bdrv_needs_filename || bs->filename[0]);
1270 ret = drv->bdrv_file_open(bs, options, open_flags, &local_err);
1271 } else if (drv->bdrv_open) {
1272 ret = drv->bdrv_open(bs, options, open_flags, &local_err);
1273 } else {
1274 ret = 0;
1275 }
1276
1277 if (ret < 0) {
1278 if (local_err) {
1279 error_propagate(errp, local_err);
1280 } else if (bs->filename[0]) {
1281 error_setg_errno(errp, -ret, "Could not open '%s'", bs->filename);
1282 } else {
1283 error_setg_errno(errp, -ret, "Could not open image");
1284 }
1285 goto open_failed;
1286 }
1287
1288 ret = refresh_total_sectors(bs, bs->total_sectors);
1289 if (ret < 0) {
1290 error_setg_errno(errp, -ret, "Could not refresh total sector count");
1291 return ret;
1292 }
1293
1294 bdrv_refresh_limits(bs, &local_err);
1295 if (local_err) {
1296 error_propagate(errp, local_err);
1297 return -EINVAL;
1298 }
1299
1300 assert(bdrv_opt_mem_align(bs) != 0);
1301 assert(bdrv_min_mem_align(bs) != 0);
1302 assert(is_power_of_2(bs->bl.request_alignment));
1303
1304 for (i = 0; i < bs->quiesce_counter; i++) {
1305 if (drv->bdrv_co_drain_begin) {
1306 drv->bdrv_co_drain_begin(bs);
1307 }
1308 }
1309
1310 return 0;
1311 open_failed:
1312 bs->drv = NULL;
1313 if (bs->file != NULL) {
1314 bdrv_unref_child(bs, bs->file);
1315 bs->file = NULL;
1316 }
1317 g_free(bs->opaque);
1318 bs->opaque = NULL;
1319 return ret;
1320 }
1321
1322 BlockDriverState *bdrv_new_open_driver(BlockDriver *drv, const char *node_name,
1323 int flags, Error **errp)
1324 {
1325 BlockDriverState *bs;
1326 int ret;
1327
1328 bs = bdrv_new();
1329 bs->open_flags = flags;
1330 bs->explicit_options = qdict_new();
1331 bs->options = qdict_new();
1332 bs->opaque = NULL;
1333
1334 update_options_from_flags(bs->options, flags);
1335
1336 ret = bdrv_open_driver(bs, drv, node_name, bs->options, flags, errp);
1337 if (ret < 0) {
1338 qobject_unref(bs->explicit_options);
1339 bs->explicit_options = NULL;
1340 qobject_unref(bs->options);
1341 bs->options = NULL;
1342 bdrv_unref(bs);
1343 return NULL;
1344 }
1345
1346 return bs;
1347 }
1348
1349 QemuOptsList bdrv_runtime_opts = {
1350 .name = "bdrv_common",
1351 .head = QTAILQ_HEAD_INITIALIZER(bdrv_runtime_opts.head),
1352 .desc = {
1353 {
1354 .name = "node-name",
1355 .type = QEMU_OPT_STRING,
1356 .help = "Node name of the block device node",
1357 },
1358 {
1359 .name = "driver",
1360 .type = QEMU_OPT_STRING,
1361 .help = "Block driver to use for the node",
1362 },
1363 {
1364 .name = BDRV_OPT_CACHE_DIRECT,
1365 .type = QEMU_OPT_BOOL,
1366 .help = "Bypass software writeback cache on the host",
1367 },
1368 {
1369 .name = BDRV_OPT_CACHE_NO_FLUSH,
1370 .type = QEMU_OPT_BOOL,
1371 .help = "Ignore flush requests",
1372 },
1373 {
1374 .name = BDRV_OPT_READ_ONLY,
1375 .type = QEMU_OPT_BOOL,
1376 .help = "Node is opened in read-only mode",
1377 },
1378 {
1379 .name = BDRV_OPT_AUTO_READ_ONLY,
1380 .type = QEMU_OPT_BOOL,
1381 .help = "Node can become read-only if opening read-write fails",
1382 },
1383 {
1384 .name = "detect-zeroes",
1385 .type = QEMU_OPT_STRING,
1386 .help = "try to optimize zero writes (off, on, unmap)",
1387 },
1388 {
1389 .name = BDRV_OPT_DISCARD,
1390 .type = QEMU_OPT_STRING,
1391 .help = "discard operation (ignore/off, unmap/on)",
1392 },
1393 {
1394 .name = BDRV_OPT_FORCE_SHARE,
1395 .type = QEMU_OPT_BOOL,
1396 .help = "always accept other writers (default: off)",
1397 },
1398 { /* end of list */ }
1399 },
1400 };
1401
1402 /*
1403 * Common part for opening disk images and files
1404 *
1405 * Removes all processed options from *options.
1406 */
1407 static int bdrv_open_common(BlockDriverState *bs, BlockBackend *file,
1408 QDict *options, Error **errp)
1409 {
1410 int ret, open_flags;
1411 const char *filename;
1412 const char *driver_name = NULL;
1413 const char *node_name = NULL;
1414 const char *discard;
1415 QemuOpts *opts;
1416 BlockDriver *drv;
1417 Error *local_err = NULL;
1418
1419 assert(bs->file == NULL);
1420 assert(options != NULL && bs->options != options);
1421
1422 opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
1423 qemu_opts_absorb_qdict(opts, options, &local_err);
1424 if (local_err) {
1425 error_propagate(errp, local_err);
1426 ret = -EINVAL;
1427 goto fail_opts;
1428 }
1429
1430 update_flags_from_options(&bs->open_flags, opts);
1431
1432 driver_name = qemu_opt_get(opts, "driver");
1433 drv = bdrv_find_format(driver_name);
1434 assert(drv != NULL);
1435
1436 bs->force_share = qemu_opt_get_bool(opts, BDRV_OPT_FORCE_SHARE, false);
1437
1438 if (bs->force_share && (bs->open_flags & BDRV_O_RDWR)) {
1439 error_setg(errp,
1440 BDRV_OPT_FORCE_SHARE
1441 "=on can only be used with read-only images");
1442 ret = -EINVAL;
1443 goto fail_opts;
1444 }
1445
1446 if (file != NULL) {
1447 bdrv_refresh_filename(blk_bs(file));
1448 filename = blk_bs(file)->filename;
1449 } else {
1450 /*
1451 * Caution: while qdict_get_try_str() is fine, getting
1452 * non-string types would require more care. When @options
1453 * come from -blockdev or blockdev_add, its members are typed
1454 * according to the QAPI schema, but when they come from
1455 * -drive, they're all QString.
1456 */
1457 filename = qdict_get_try_str(options, "filename");
1458 }
1459
1460 if (drv->bdrv_needs_filename && (!filename || !filename[0])) {
1461 error_setg(errp, "The '%s' block driver requires a file name",
1462 drv->format_name);
1463 ret = -EINVAL;
1464 goto fail_opts;
1465 }
1466
1467 trace_bdrv_open_common(bs, filename ?: "", bs->open_flags,
1468 drv->format_name);
1469
1470 bs->read_only = !(bs->open_flags & BDRV_O_RDWR);
1471
1472 if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv, bs->read_only)) {
1473 if (!bs->read_only && bdrv_is_whitelisted(drv, true)) {
1474 ret = bdrv_apply_auto_read_only(bs, NULL, NULL);
1475 } else {
1476 ret = -ENOTSUP;
1477 }
1478 if (ret < 0) {
1479 error_setg(errp,
1480 !bs->read_only && bdrv_is_whitelisted(drv, true)
1481 ? "Driver '%s' can only be used for read-only devices"
1482 : "Driver '%s' is not whitelisted",
1483 drv->format_name);
1484 goto fail_opts;
1485 }
1486 }
1487
1488 /* bdrv_new() and bdrv_close() make it so */
1489 assert(atomic_read(&bs->copy_on_read) == 0);
1490
1491 if (bs->open_flags & BDRV_O_COPY_ON_READ) {
1492 if (!bs->read_only) {
1493 bdrv_enable_copy_on_read(bs);
1494 } else {
1495 error_setg(errp, "Can't use copy-on-read on read-only device");
1496 ret = -EINVAL;
1497 goto fail_opts;
1498 }
1499 }
1500
1501 discard = qemu_opt_get(opts, BDRV_OPT_DISCARD);
1502 if (discard != NULL) {
1503 if (bdrv_parse_discard_flags(discard, &bs->open_flags) != 0) {
1504 error_setg(errp, "Invalid discard option");
1505 ret = -EINVAL;
1506 goto fail_opts;
1507 }
1508 }
1509
1510 bs->detect_zeroes =
1511 bdrv_parse_detect_zeroes(opts, bs->open_flags, &local_err);
1512 if (local_err) {
1513 error_propagate(errp, local_err);
1514 ret = -EINVAL;
1515 goto fail_opts;
1516 }
1517
1518 if (filename != NULL) {
1519 pstrcpy(bs->filename, sizeof(bs->filename), filename);
1520 } else {
1521 bs->filename[0] = '\0';
1522 }
1523 pstrcpy(bs->exact_filename, sizeof(bs->exact_filename), bs->filename);
1524
1525 /* Open the image, either directly or using a protocol */
1526 open_flags = bdrv_open_flags(bs, bs->open_flags);
1527 node_name = qemu_opt_get(opts, "node-name");
1528
1529 assert(!drv->bdrv_file_open || file == NULL);
1530 ret = bdrv_open_driver(bs, drv, node_name, options, open_flags, errp);
1531 if (ret < 0) {
1532 goto fail_opts;
1533 }
1534
1535 qemu_opts_del(opts);
1536 return 0;
1537
1538 fail_opts:
1539 qemu_opts_del(opts);
1540 return ret;
1541 }
1542
1543 static QDict *parse_json_filename(const char *filename, Error **errp)
1544 {
1545 QObject *options_obj;
1546 QDict *options;
1547 int ret;
1548
1549 ret = strstart(filename, "json:", &filename);
1550 assert(ret);
1551
1552 options_obj = qobject_from_json(filename, errp);
1553 if (!options_obj) {
1554 error_prepend(errp, "Could not parse the JSON options: ");
1555 return NULL;
1556 }
1557
1558 options = qobject_to(QDict, options_obj);
1559 if (!options) {
1560 qobject_unref(options_obj);
1561 error_setg(errp, "Invalid JSON object given");
1562 return NULL;
1563 }
1564
1565 qdict_flatten(options);
1566
1567 return options;
1568 }
1569
1570 static void parse_json_protocol(QDict *options, const char **pfilename,
1571 Error **errp)
1572 {
1573 QDict *json_options;
1574 Error *local_err = NULL;
1575
1576 /* Parse json: pseudo-protocol */
1577 if (!*pfilename || !g_str_has_prefix(*pfilename, "json:")) {
1578 return;
1579 }
1580
1581 json_options = parse_json_filename(*pfilename, &local_err);
1582 if (local_err) {
1583 error_propagate(errp, local_err);
1584 return;
1585 }
1586
1587 /* Options given in the filename have lower priority than options
1588 * specified directly */
1589 qdict_join(options, json_options, false);
1590 qobject_unref(json_options);
1591 *pfilename = NULL;
1592 }
1593
1594 /*
1595 * Fills in default options for opening images and converts the legacy
1596 * filename/flags pair to option QDict entries.
1597 * The BDRV_O_PROTOCOL flag in *flags will be set or cleared accordingly if a
1598 * block driver has been specified explicitly.
1599 */
1600 static int bdrv_fill_options(QDict **options, const char *filename,
1601 int *flags, Error **errp)
1602 {
1603 const char *drvname;
1604 bool protocol = *flags & BDRV_O_PROTOCOL;
1605 bool parse_filename = false;
1606 BlockDriver *drv = NULL;
1607 Error *local_err = NULL;
1608
1609 /*
1610 * Caution: while qdict_get_try_str() is fine, getting non-string
1611 * types would require more care. When @options come from
1612 * -blockdev or blockdev_add, its members are typed according to
1613 * the QAPI schema, but when they come from -drive, they're all
1614 * QString.
1615 */
1616 drvname = qdict_get_try_str(*options, "driver");
1617 if (drvname) {
1618 drv = bdrv_find_format(drvname);
1619 if (!drv) {
1620 error_setg(errp, "Unknown driver '%s'", drvname);
1621 return -ENOENT;
1622 }
1623 /* If the user has explicitly specified the driver, this choice should
1624 * override the BDRV_O_PROTOCOL flag */
1625 protocol = drv->bdrv_file_open;
1626 }
1627
1628 if (protocol) {
1629 *flags |= BDRV_O_PROTOCOL;
1630 } else {
1631 *flags &= ~BDRV_O_PROTOCOL;
1632 }
1633
1634 /* Translate cache options from flags into options */
1635 update_options_from_flags(*options, *flags);
1636
1637 /* Fetch the file name from the options QDict if necessary */
1638 if (protocol && filename) {
1639 if (!qdict_haskey(*options, "filename")) {
1640 qdict_put_str(*options, "filename", filename);
1641 parse_filename = true;
1642 } else {
1643 error_setg(errp, "Can't specify 'file' and 'filename' options at "
1644 "the same time");
1645 return -EINVAL;
1646 }
1647 }
1648
1649 /* Find the right block driver */
1650 /* See cautionary note on accessing @options above */
1651 filename = qdict_get_try_str(*options, "filename");
1652
1653 if (!drvname && protocol) {
1654 if (filename) {
1655 drv = bdrv_find_protocol(filename, parse_filename, errp);
1656 if (!drv) {
1657 return -EINVAL;
1658 }
1659
1660 drvname = drv->format_name;
1661 qdict_put_str(*options, "driver", drvname);
1662 } else {
1663 error_setg(errp, "Must specify either driver or file");
1664 return -EINVAL;
1665 }
1666 }
1667
1668 assert(drv || !protocol);
1669
1670 /* Driver-specific filename parsing */
1671 if (drv && drv->bdrv_parse_filename && parse_filename) {
1672 drv->bdrv_parse_filename(filename, *options, &local_err);
1673 if (local_err) {
1674 error_propagate(errp, local_err);
1675 return -EINVAL;
1676 }
1677
1678 if (!drv->bdrv_needs_filename) {
1679 qdict_del(*options, "filename");
1680 }
1681 }
1682
1683 return 0;
1684 }
1685
1686 static int bdrv_child_check_perm(BdrvChild *c, BlockReopenQueue *q,
1687 uint64_t perm, uint64_t shared,
1688 GSList *ignore_children, Error **errp);
1689 static void bdrv_child_abort_perm_update(BdrvChild *c);
1690 static void bdrv_child_set_perm(BdrvChild *c, uint64_t perm, uint64_t shared);
1691
1692 typedef struct BlockReopenQueueEntry {
1693 bool prepared;
1694 bool perms_checked;
1695 BDRVReopenState state;
1696 QSIMPLEQ_ENTRY(BlockReopenQueueEntry) entry;
1697 } BlockReopenQueueEntry;
1698
1699 /*
1700 * Return the flags that @bs will have after the reopens in @q have
1701 * successfully completed. If @q is NULL (or @bs is not contained in @q),
1702 * return the current flags.
1703 */
1704 static int bdrv_reopen_get_flags(BlockReopenQueue *q, BlockDriverState *bs)
1705 {
1706 BlockReopenQueueEntry *entry;
1707
1708 if (q != NULL) {
1709 QSIMPLEQ_FOREACH(entry, q, entry) {
1710 if (entry->state.bs == bs) {
1711 return entry->state.flags;
1712 }
1713 }
1714 }
1715
1716 return bs->open_flags;
1717 }
1718
1719 /* Returns whether the image file can be written to after the reopen queue @q
1720 * has been successfully applied, or right now if @q is NULL. */
1721 static bool bdrv_is_writable_after_reopen(BlockDriverState *bs,
1722 BlockReopenQueue *q)
1723 {
1724 int flags = bdrv_reopen_get_flags(q, bs);
1725
1726 return (flags & (BDRV_O_RDWR | BDRV_O_INACTIVE)) == BDRV_O_RDWR;
1727 }
1728
1729 /*
1730 * Return whether the BDS can be written to. This is not necessarily
1731 * the same as !bdrv_is_read_only(bs), as inactivated images may not
1732 * be written to but do not count as read-only images.
1733 */
1734 bool bdrv_is_writable(BlockDriverState *bs)
1735 {
1736 return bdrv_is_writable_after_reopen(bs, NULL);
1737 }
1738
1739 static void bdrv_child_perm(BlockDriverState *bs, BlockDriverState *child_bs,
1740 BdrvChild *c, const BdrvChildRole *role,
1741 BlockReopenQueue *reopen_queue,
1742 uint64_t parent_perm, uint64_t parent_shared,
1743 uint64_t *nperm, uint64_t *nshared)
1744 {
1745 if (bs->drv && bs->drv->bdrv_child_perm) {
1746 bs->drv->bdrv_child_perm(bs, c, role, reopen_queue,
1747 parent_perm, parent_shared,
1748 nperm, nshared);
1749 }
1750 /* TODO Take force_share from reopen_queue */
1751 if (child_bs && child_bs->force_share) {
1752 *nshared = BLK_PERM_ALL;
1753 }
1754 }
1755
1756 /*
1757 * Check whether permissions on this node can be changed in a way that
1758 * @cumulative_perms and @cumulative_shared_perms are the new cumulative
1759 * permissions of all its parents. This involves checking whether all necessary
1760 * permission changes to child nodes can be performed.
1761 *
1762 * A call to this function must always be followed by a call to bdrv_set_perm()
1763 * or bdrv_abort_perm_update().
1764 */
1765 static int bdrv_check_perm(BlockDriverState *bs, BlockReopenQueue *q,
1766 uint64_t cumulative_perms,
1767 uint64_t cumulative_shared_perms,
1768 GSList *ignore_children, Error **errp)
1769 {
1770 BlockDriver *drv = bs->drv;
1771 BdrvChild *c;
1772 int ret;
1773
1774 /* Write permissions never work with read-only images */
1775 if ((cumulative_perms & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)) &&
1776 !bdrv_is_writable_after_reopen(bs, q))
1777 {
1778 error_setg(errp, "Block node is read-only");
1779 return -EPERM;
1780 }
1781
1782 /* Check this node */
1783 if (!drv) {
1784 return 0;
1785 }
1786
1787 if (drv->bdrv_check_perm) {
1788 return drv->bdrv_check_perm(bs, cumulative_perms,
1789 cumulative_shared_perms, errp);
1790 }
1791
1792 /* Drivers that never have children can omit .bdrv_child_perm() */
1793 if (!drv->bdrv_child_perm) {
1794 assert(QLIST_EMPTY(&bs->children));
1795 return 0;
1796 }
1797
1798 /* Check all children */
1799 QLIST_FOREACH(c, &bs->children, next) {
1800 uint64_t cur_perm, cur_shared;
1801 bdrv_child_perm(bs, c->bs, c, c->role, q,
1802 cumulative_perms, cumulative_shared_perms,
1803 &cur_perm, &cur_shared);
1804 ret = bdrv_child_check_perm(c, q, cur_perm, cur_shared,
1805 ignore_children, errp);
1806 if (ret < 0) {
1807 return ret;
1808 }
1809 }
1810
1811 return 0;
1812 }
1813
1814 /*
1815 * Notifies drivers that after a previous bdrv_check_perm() call, the
1816 * permission update is not performed and any preparations made for it (e.g.
1817 * taken file locks) need to be undone.
1818 *
1819 * This function recursively notifies all child nodes.
1820 */
1821 static void bdrv_abort_perm_update(BlockDriverState *bs)
1822 {
1823 BlockDriver *drv = bs->drv;
1824 BdrvChild *c;
1825
1826 if (!drv) {
1827 return;
1828 }
1829
1830 if (drv->bdrv_abort_perm_update) {
1831 drv->bdrv_abort_perm_update(bs);
1832 }
1833
1834 QLIST_FOREACH(c, &bs->children, next) {
1835 bdrv_child_abort_perm_update(c);
1836 }
1837 }
1838
1839 static void bdrv_set_perm(BlockDriverState *bs, uint64_t cumulative_perms,
1840 uint64_t cumulative_shared_perms)
1841 {
1842 BlockDriver *drv = bs->drv;
1843 BdrvChild *c;
1844
1845 if (!drv) {
1846 return;
1847 }
1848
1849 /* Update this node */
1850 if (drv->bdrv_set_perm) {
1851 drv->bdrv_set_perm(bs, cumulative_perms, cumulative_shared_perms);
1852 }
1853
1854 /* Drivers that never have children can omit .bdrv_child_perm() */
1855 if (!drv->bdrv_child_perm) {
1856 assert(QLIST_EMPTY(&bs->children));
1857 return;
1858 }
1859
1860 /* Update all children */
1861 QLIST_FOREACH(c, &bs->children, next) {
1862 uint64_t cur_perm, cur_shared;
1863 bdrv_child_perm(bs, c->bs, c, c->role, NULL,
1864 cumulative_perms, cumulative_shared_perms,
1865 &cur_perm, &cur_shared);
1866 bdrv_child_set_perm(c, cur_perm, cur_shared);
1867 }
1868 }
1869
1870 static void bdrv_get_cumulative_perm(BlockDriverState *bs, uint64_t *perm,
1871 uint64_t *shared_perm)
1872 {
1873 BdrvChild *c;
1874 uint64_t cumulative_perms = 0;
1875 uint64_t cumulative_shared_perms = BLK_PERM_ALL;
1876
1877 QLIST_FOREACH(c, &bs->parents, next_parent) {
1878 cumulative_perms |= c->perm;
1879 cumulative_shared_perms &= c->shared_perm;
1880 }
1881
1882 *perm = cumulative_perms;
1883 *shared_perm = cumulative_shared_perms;
1884 }
1885
1886 static char *bdrv_child_user_desc(BdrvChild *c)
1887 {
1888 if (c->role->get_parent_desc) {
1889 return c->role->get_parent_desc(c);
1890 }
1891
1892 return g_strdup("another user");
1893 }
1894
1895 char *bdrv_perm_names(uint64_t perm)
1896 {
1897 struct perm_name {
1898 uint64_t perm;
1899 const char *name;
1900 } permissions[] = {
1901 { BLK_PERM_CONSISTENT_READ, "consistent read" },
1902 { BLK_PERM_WRITE, "write" },
1903 { BLK_PERM_WRITE_UNCHANGED, "write unchanged" },
1904 { BLK_PERM_RESIZE, "resize" },
1905 { BLK_PERM_GRAPH_MOD, "change children" },
1906 { 0, NULL }
1907 };
1908
1909 char *result = g_strdup("");
1910 struct perm_name *p;
1911
1912 for (p = permissions; p->name; p++) {
1913 if (perm & p->perm) {
1914 char *old = result;
1915 result = g_strdup_printf("%s%s%s", old, *old ? ", " : "", p->name);
1916 g_free(old);
1917 }
1918 }
1919
1920 return result;
1921 }
1922
1923 /*
1924 * Checks whether a new reference to @bs can be added if the new user requires
1925 * @new_used_perm/@new_shared_perm as its permissions. If @ignore_children is
1926 * set, the BdrvChild objects in this list are ignored in the calculations;
1927 * this allows checking permission updates for an existing reference.
1928 *
1929 * Needs to be followed by a call to either bdrv_set_perm() or
1930 * bdrv_abort_perm_update(). */
1931 static int bdrv_check_update_perm(BlockDriverState *bs, BlockReopenQueue *q,
1932 uint64_t new_used_perm,
1933 uint64_t new_shared_perm,
1934 GSList *ignore_children, Error **errp)
1935 {
1936 BdrvChild *c;
1937 uint64_t cumulative_perms = new_used_perm;
1938 uint64_t cumulative_shared_perms = new_shared_perm;
1939
1940 /* There is no reason why anyone couldn't tolerate write_unchanged */
1941 assert(new_shared_perm & BLK_PERM_WRITE_UNCHANGED);
1942
1943 QLIST_FOREACH(c, &bs->parents, next_parent) {
1944 if (g_slist_find(ignore_children, c)) {
1945 continue;
1946 }
1947
1948 if ((new_used_perm & c->shared_perm) != new_used_perm) {
1949 char *user = bdrv_child_user_desc(c);
1950 char *perm_names = bdrv_perm_names(new_used_perm & ~c->shared_perm);
1951 error_setg(errp, "Conflicts with use by %s as '%s', which does not "
1952 "allow '%s' on %s",
1953 user, c->name, perm_names, bdrv_get_node_name(c->bs));
1954 g_free(user);
1955 g_free(perm_names);
1956 return -EPERM;
1957 }
1958
1959 if ((c->perm & new_shared_perm) != c->perm) {
1960 char *user = bdrv_child_user_desc(c);
1961 char *perm_names = bdrv_perm_names(c->perm & ~new_shared_perm);
1962 error_setg(errp, "Conflicts with use by %s as '%s', which uses "
1963 "'%s' on %s",
1964 user, c->name, perm_names, bdrv_get_node_name(c->bs));
1965 g_free(user);
1966 g_free(perm_names);
1967 return -EPERM;
1968 }
1969
1970 cumulative_perms |= c->perm;
1971 cumulative_shared_perms &= c->shared_perm;
1972 }
1973
1974 return bdrv_check_perm(bs, q, cumulative_perms, cumulative_shared_perms,
1975 ignore_children, errp);
1976 }
1977
1978 /* Needs to be followed by a call to either bdrv_child_set_perm() or
1979 * bdrv_child_abort_perm_update(). */
1980 static int bdrv_child_check_perm(BdrvChild *c, BlockReopenQueue *q,
1981 uint64_t perm, uint64_t shared,
1982 GSList *ignore_children, Error **errp)
1983 {
1984 int ret;
1985
1986 ignore_children = g_slist_prepend(g_slist_copy(ignore_children), c);
1987 ret = bdrv_check_update_perm(c->bs, q, perm, shared, ignore_children, errp);
1988 g_slist_free(ignore_children);
1989
1990 if (ret < 0) {
1991 return ret;
1992 }
1993
1994 if (!c->has_backup_perm) {
1995 c->has_backup_perm = true;
1996 c->backup_perm = c->perm;
1997 c->backup_shared_perm = c->shared_perm;
1998 }
1999 /*
2000 * Note: it's OK if c->has_backup_perm was already set, as we can find the
2001 * same child twice during check_perm procedure
2002 */
2003
2004 c->perm = perm;
2005 c->shared_perm = shared;
2006
2007 return 0;
2008 }
2009
2010 static void bdrv_child_set_perm(BdrvChild *c, uint64_t perm, uint64_t shared)
2011 {
2012 uint64_t cumulative_perms, cumulative_shared_perms;
2013
2014 c->has_backup_perm = false;
2015
2016 c->perm = perm;
2017 c->shared_perm = shared;
2018
2019 bdrv_get_cumulative_perm(c->bs, &cumulative_perms,
2020 &cumulative_shared_perms);
2021 bdrv_set_perm(c->bs, cumulative_perms, cumulative_shared_perms);
2022 }
2023
2024 static void bdrv_child_abort_perm_update(BdrvChild *c)
2025 {
2026 if (c->has_backup_perm) {
2027 c->perm = c->backup_perm;
2028 c->shared_perm = c->backup_shared_perm;
2029 c->has_backup_perm = false;
2030 }
2031
2032 bdrv_abort_perm_update(c->bs);
2033 }
2034
2035 int bdrv_child_try_set_perm(BdrvChild *c, uint64_t perm, uint64_t shared,
2036 Error **errp)
2037 {
2038 int ret;
2039
2040 ret = bdrv_child_check_perm(c, NULL, perm, shared, NULL, errp);
2041 if (ret < 0) {
2042 bdrv_child_abort_perm_update(c);
2043 return ret;
2044 }
2045
2046 bdrv_child_set_perm(c, perm, shared);
2047
2048 return 0;
2049 }
2050
2051 void bdrv_filter_default_perms(BlockDriverState *bs, BdrvChild *c,
2052 const BdrvChildRole *role,
2053 BlockReopenQueue *reopen_queue,
2054 uint64_t perm, uint64_t shared,
2055 uint64_t *nperm, uint64_t *nshared)
2056 {
2057 if (c == NULL) {
2058 *nperm = perm & DEFAULT_PERM_PASSTHROUGH;
2059 *nshared = (shared & DEFAULT_PERM_PASSTHROUGH) | DEFAULT_PERM_UNCHANGED;
2060 return;
2061 }
2062
2063 *nperm = (perm & DEFAULT_PERM_PASSTHROUGH) |
2064 (c->perm & DEFAULT_PERM_UNCHANGED);
2065 *nshared = (shared & DEFAULT_PERM_PASSTHROUGH) |
2066 (c->shared_perm & DEFAULT_PERM_UNCHANGED);
2067 }
2068
2069 void bdrv_format_default_perms(BlockDriverState *bs, BdrvChild *c,
2070 const BdrvChildRole *role,
2071 BlockReopenQueue *reopen_queue,
2072 uint64_t perm, uint64_t shared,
2073 uint64_t *nperm, uint64_t *nshared)
2074 {
2075 bool backing = (role == &child_backing);
2076 assert(role == &child_backing || role == &child_file);
2077
2078 if (!backing) {
2079 int flags = bdrv_reopen_get_flags(reopen_queue, bs);
2080
2081 /* Apart from the modifications below, the same permissions are
2082 * forwarded and left alone as for filters */
2083 bdrv_filter_default_perms(bs, c, role, reopen_queue, perm, shared,
2084 &perm, &shared);
2085
2086 /* Format drivers may touch metadata even if the guest doesn't write */
2087 if (bdrv_is_writable_after_reopen(bs, reopen_queue)) {
2088 perm |= BLK_PERM_WRITE | BLK_PERM_RESIZE;
2089 }
2090
2091 /* bs->file always needs to be consistent because of the metadata. We
2092 * can never allow other users to resize or write to it. */
2093 if (!(flags & BDRV_O_NO_IO)) {
2094 perm |= BLK_PERM_CONSISTENT_READ;
2095 }
2096 shared &= ~(BLK_PERM_WRITE | BLK_PERM_RESIZE);
2097 } else {
2098 /* We want consistent read from backing files if the parent needs it.
2099 * No other operations are performed on backing files. */
2100 perm &= BLK_PERM_CONSISTENT_READ;
2101
2102 /* If the parent can deal with changing data, we're okay with a
2103 * writable and resizable backing file. */
2104 /* TODO Require !(perm & BLK_PERM_CONSISTENT_READ), too? */
2105 if (shared & BLK_PERM_WRITE) {
2106 shared = BLK_PERM_WRITE | BLK_PERM_RESIZE;
2107 } else {
2108 shared = 0;
2109 }
2110
2111 shared |= BLK_PERM_CONSISTENT_READ | BLK_PERM_GRAPH_MOD |
2112 BLK_PERM_WRITE_UNCHANGED;
2113 }
2114
2115 if (bs->open_flags & BDRV_O_INACTIVE) {
2116 shared |= BLK_PERM_WRITE | BLK_PERM_RESIZE;
2117 }
2118
2119 *nperm = perm;
2120 *nshared = shared;
2121 }
2122
2123 static void bdrv_replace_child_noperm(BdrvChild *child,
2124 BlockDriverState *new_bs)
2125 {
2126 BlockDriverState *old_bs = child->bs;
2127 int i;
2128
2129 assert(!child->frozen);
2130
2131 if (old_bs && new_bs) {
2132 assert(bdrv_get_aio_context(old_bs) == bdrv_get_aio_context(new_bs));
2133 }
2134 if (old_bs) {
2135 /* Detach first so that the recursive drain sections coming from @child
2136 * are already gone and we only end the drain sections that came from
2137 * elsewhere. */
2138 if (child->role->detach) {
2139 child->role->detach(child);
2140 }
2141 if (old_bs->quiesce_counter && child->role->drained_end) {
2142 int num = old_bs->quiesce_counter;
2143 if (child->role->parent_is_bds) {
2144 num -= bdrv_drain_all_count;
2145 }
2146 assert(num >= 0);
2147 for (i = 0; i < num; i++) {
2148 child->role->drained_end(child);
2149 }
2150 }
2151 QLIST_REMOVE(child, next_parent);
2152 }
2153
2154 child->bs = new_bs;
2155
2156 if (new_bs) {
2157 QLIST_INSERT_HEAD(&new_bs->parents, child, next_parent);
2158 if (new_bs->quiesce_counter && child->role->drained_begin) {
2159 int num = new_bs->quiesce_counter;
2160 if (child->role->parent_is_bds) {
2161 num -= bdrv_drain_all_count;
2162 }
2163 assert(num >= 0);
2164 for (i = 0; i < num; i++) {
2165 bdrv_parent_drained_begin_single(child, true);
2166 }
2167 }
2168
2169 /* Attach only after starting new drained sections, so that recursive
2170 * drain sections coming from @child don't get an extra .drained_begin
2171 * callback. */
2172 if (child->role->attach) {
2173 child->role->attach(child);
2174 }
2175 }
2176 }
2177
2178 /*
2179 * Updates @child to change its reference to point to @new_bs, including
2180 * checking and applying the necessary permisson updates both to the old node
2181 * and to @new_bs.
2182 *
2183 * NULL is passed as @new_bs for removing the reference before freeing @child.
2184 *
2185 * If @new_bs is not NULL, bdrv_check_perm() must be called beforehand, as this
2186 * function uses bdrv_set_perm() to update the permissions according to the new
2187 * reference that @new_bs gets.
2188 */
2189 static void bdrv_replace_child(BdrvChild *child, BlockDriverState *new_bs)
2190 {
2191 BlockDriverState *old_bs = child->bs;
2192 uint64_t perm, shared_perm;
2193
2194 bdrv_replace_child_noperm(child, new_bs);
2195
2196 if (old_bs) {
2197 /* Update permissions for old node. This is guaranteed to succeed
2198 * because we're just taking a parent away, so we're loosening
2199 * restrictions. */
2200 bdrv_get_cumulative_perm(old_bs, &perm, &shared_perm);
2201 bdrv_check_perm(old_bs, NULL, perm, shared_perm, NULL, &error_abort);
2202 bdrv_set_perm(old_bs, perm, shared_perm);
2203 }
2204
2205 if (new_bs) {
2206 bdrv_get_cumulative_perm(new_bs, &perm, &shared_perm);
2207 bdrv_set_perm(new_bs, perm, shared_perm);
2208 }
2209 }
2210
2211 BdrvChild *bdrv_root_attach_child(BlockDriverState *child_bs,
2212 const char *child_name,
2213 const BdrvChildRole *child_role,
2214 uint64_t perm, uint64_t shared_perm,
2215 void *opaque, Error **errp)
2216 {
2217 BdrvChild *child;
2218 int ret;
2219
2220 ret = bdrv_check_update_perm(child_bs, NULL, perm, shared_perm, NULL, errp);
2221 if (ret < 0) {
2222 bdrv_abort_perm_update(child_bs);
2223 return NULL;
2224 }
2225
2226 child = g_new(BdrvChild, 1);
2227 *child = (BdrvChild) {
2228 .bs = NULL,
2229 .name = g_strdup(child_name),
2230 .role = child_role,
2231 .perm = perm,
2232 .shared_perm = shared_perm,
2233 .opaque = opaque,
2234 };
2235
2236 /* This performs the matching bdrv_set_perm() for the above check. */
2237 bdrv_replace_child(child, child_bs);
2238
2239 return child;
2240 }
2241
2242 BdrvChild *bdrv_attach_child(BlockDriverState *parent_bs,
2243 BlockDriverState *child_bs,
2244 const char *child_name,
2245 const BdrvChildRole *child_role,
2246 Error **errp)
2247 {
2248 BdrvChild *child;
2249 uint64_t perm, shared_perm;
2250
2251 bdrv_get_cumulative_perm(parent_bs, &perm, &shared_perm);
2252
2253 assert(parent_bs->drv);
2254 assert(bdrv_get_aio_context(parent_bs) == bdrv_get_aio_context(child_bs));
2255 bdrv_child_perm(parent_bs, child_bs, NULL, child_role, NULL,
2256 perm, shared_perm, &perm, &shared_perm);
2257
2258 child = bdrv_root_attach_child(child_bs, child_name, child_role,
2259 perm, shared_perm, parent_bs, errp);
2260 if (child == NULL) {
2261 return NULL;
2262 }
2263
2264 QLIST_INSERT_HEAD(&parent_bs->children, child, next);
2265 return child;
2266 }
2267
2268 static void bdrv_detach_child(BdrvChild *child)
2269 {
2270 if (child->next.le_prev) {
2271 QLIST_REMOVE(child, next);
2272 child->next.le_prev = NULL;
2273 }
2274
2275 bdrv_replace_child(child, NULL);
2276
2277 g_free(child->name);
2278 g_free(child);
2279 }
2280
2281 void bdrv_root_unref_child(BdrvChild *child)
2282 {
2283 BlockDriverState *child_bs;
2284
2285 child_bs = child->bs;
2286 bdrv_detach_child(child);
2287 bdrv_unref(child_bs);
2288 }
2289
2290 void bdrv_unref_child(BlockDriverState *parent, BdrvChild *child)
2291 {
2292 if (child == NULL) {
2293 return;
2294 }
2295
2296 if (child->bs->inherits_from == parent) {
2297 BdrvChild *c;
2298
2299 /* Remove inherits_from only when the last reference between parent and
2300 * child->bs goes away. */
2301 QLIST_FOREACH(c, &parent->children, next) {
2302 if (c != child && c->bs == child->bs) {
2303 break;
2304 }
2305 }
2306 if (c == NULL) {
2307 child->bs->inherits_from = NULL;
2308 }
2309 }
2310
2311 bdrv_root_unref_child(child);
2312 }
2313
2314
2315 static void bdrv_parent_cb_change_media(BlockDriverState *bs, bool load)
2316 {
2317 BdrvChild *c;
2318 QLIST_FOREACH(c, &bs->parents, next_parent) {
2319 if (c->role->change_media) {
2320 c->role->change_media(c, load);
2321 }
2322 }
2323 }
2324
2325 /* Return true if you can reach parent going through child->inherits_from
2326 * recursively. If parent or child are NULL, return false */
2327 static bool bdrv_inherits_from_recursive(BlockDriverState *child,
2328 BlockDriverState *parent)
2329 {
2330 while (child && child != parent) {
2331 child = child->inherits_from;
2332 }
2333
2334 return child != NULL;
2335 }
2336
2337 /*
2338 * Sets the backing file link of a BDS. A new reference is created; callers
2339 * which don't need their own reference any more must call bdrv_unref().
2340 */
2341 void bdrv_set_backing_hd(BlockDriverState *bs, BlockDriverState *backing_hd,
2342 Error **errp)
2343 {
2344 bool update_inherits_from = bdrv_chain_contains(bs, backing_hd) &&
2345 bdrv_inherits_from_recursive(backing_hd, bs);
2346
2347 if (bdrv_is_backing_chain_frozen(bs, backing_bs(bs), errp)) {
2348 return;
2349 }
2350
2351 if (backing_hd) {
2352 bdrv_ref(backing_hd);
2353 }
2354
2355 if (bs->backing) {
2356 bdrv_unref_child(bs, bs->backing);
2357 }
2358
2359 if (!backing_hd) {
2360 bs->backing = NULL;
2361 goto out;
2362 }
2363
2364 bs->backing = bdrv_attach_child(bs, backing_hd, "backing", &child_backing,
2365 errp);
2366 /* If backing_hd was already part of bs's backing chain, and
2367 * inherits_from pointed recursively to bs then let's update it to
2368 * point directly to bs (else it will become NULL). */
2369 if (update_inherits_from) {
2370 backing_hd->inherits_from = bs;
2371 }
2372 if (!bs->backing) {
2373 bdrv_unref(backing_hd);
2374 }
2375
2376 out:
2377 bdrv_refresh_limits(bs, NULL);
2378 }
2379
2380 /*
2381 * Opens the backing file for a BlockDriverState if not yet open
2382 *
2383 * bdref_key specifies the key for the image's BlockdevRef in the options QDict.
2384 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
2385 * itself, all options starting with "${bdref_key}." are considered part of the
2386 * BlockdevRef.
2387 *
2388 * TODO Can this be unified with bdrv_open_image()?
2389 */
2390 int bdrv_open_backing_file(BlockDriverState *bs, QDict *parent_options,
2391 const char *bdref_key, Error **errp)
2392 {
2393 char *backing_filename = NULL;
2394 char *bdref_key_dot;
2395 const char *reference = NULL;
2396 int ret = 0;
2397 bool implicit_backing = false;
2398 BlockDriverState *backing_hd;
2399 QDict *options;
2400 QDict *tmp_parent_options = NULL;
2401 Error *local_err = NULL;
2402
2403 if (bs->backing != NULL) {
2404 goto free_exit;
2405 }
2406
2407 /* NULL means an empty set of options */
2408 if (parent_options == NULL) {
2409 tmp_parent_options = qdict_new();
2410 parent_options = tmp_parent_options;
2411 }
2412
2413 bs->open_flags &= ~BDRV_O_NO_BACKING;
2414
2415 bdref_key_dot = g_strdup_printf("%s.", bdref_key);
2416 qdict_extract_subqdict(parent_options, &options, bdref_key_dot);
2417 g_free(bdref_key_dot);
2418
2419 /*
2420 * Caution: while qdict_get_try_str() is fine, getting non-string
2421 * types would require more care. When @parent_options come from
2422 * -blockdev or blockdev_add, its members are typed according to
2423 * the QAPI schema, but when they come from -drive, they're all
2424 * QString.
2425 */
2426 reference = qdict_get_try_str(parent_options, bdref_key);
2427 if (reference || qdict_haskey(options, "file.filename")) {
2428 /* keep backing_filename NULL */
2429 } else if (bs->backing_file[0] == '\0' && qdict_size(options) == 0) {
2430 qobject_unref(options);
2431 goto free_exit;
2432 } else {
2433 if (qdict_size(options) == 0) {
2434 /* If the user specifies options that do not modify the
2435 * backing file's behavior, we might still consider it the
2436 * implicit backing file. But it's easier this way, and
2437 * just specifying some of the backing BDS's options is
2438 * only possible with -drive anyway (otherwise the QAPI
2439 * schema forces the user to specify everything). */
2440 implicit_backing = !strcmp(bs->auto_backing_file, bs->backing_file);
2441 }
2442
2443 backing_filename = bdrv_get_full_backing_filename(bs, &local_err);
2444 if (local_err) {
2445 ret = -EINVAL;
2446 error_propagate(errp, local_err);
2447 qobject_unref(options);
2448 goto free_exit;
2449 }
2450 }
2451
2452 if (!bs->drv || !bs->drv->supports_backing) {
2453 ret = -EINVAL;
2454 error_setg(errp, "Driver doesn't support backing files");
2455 qobject_unref(options);
2456 goto free_exit;
2457 }
2458
2459 if (!reference &&
2460 bs->backing_format[0] != '\0' && !qdict_haskey(options, "driver")) {
2461 qdict_put_str(options, "driver", bs->backing_format);
2462 }
2463
2464 backing_hd = bdrv_open_inherit(backing_filename, reference, options, 0, bs,
2465 &child_backing, errp);
2466 if (!backing_hd) {
2467 bs->open_flags |= BDRV_O_NO_BACKING;
2468 error_prepend(errp, "Could not open backing file: ");
2469 ret = -EINVAL;
2470 goto free_exit;
2471 }
2472 bdrv_set_aio_context(backing_hd, bdrv_get_aio_context(bs));
2473
2474 if (implicit_backing) {
2475 bdrv_refresh_filename(backing_hd);
2476 pstrcpy(bs->auto_backing_file, sizeof(bs->auto_backing_file),
2477 backing_hd->filename);
2478 }
2479
2480 /* Hook up the backing file link; drop our reference, bs owns the
2481 * backing_hd reference now */
2482 bdrv_set_backing_hd(bs, backing_hd, &local_err);
2483 bdrv_unref(backing_hd);
2484 if (local_err) {
2485 error_propagate(errp, local_err);
2486 ret = -EINVAL;
2487 goto free_exit;
2488 }
2489
2490 qdict_del(parent_options, bdref_key);
2491
2492 free_exit:
2493 g_free(backing_filename);
2494 qobject_unref(tmp_parent_options);
2495 return ret;
2496 }
2497
2498 static BlockDriverState *
2499 bdrv_open_child_bs(const char *filename, QDict *options, const char *bdref_key,
2500 BlockDriverState *parent, const BdrvChildRole *child_role,
2501 bool allow_none, Error **errp)
2502 {
2503 BlockDriverState *bs = NULL;
2504 QDict *image_options;
2505 char *bdref_key_dot;
2506 const char *reference;
2507
2508 assert(child_role != NULL);
2509
2510 bdref_key_dot = g_strdup_printf("%s.", bdref_key);
2511 qdict_extract_subqdict(options, &image_options, bdref_key_dot);
2512 g_free(bdref_key_dot);
2513
2514 /*
2515 * Caution: while qdict_get_try_str() is fine, getting non-string
2516 * types would require more care. When @options come from
2517 * -blockdev or blockdev_add, its members are typed according to
2518 * the QAPI schema, but when they come from -drive, they're all
2519 * QString.
2520 */
2521 reference = qdict_get_try_str(options, bdref_key);
2522 if (!filename && !reference && !qdict_size(image_options)) {
2523 if (!allow_none) {
2524 error_setg(errp, "A block device must be specified for \"%s\"",
2525 bdref_key);
2526 }
2527 qobject_unref(image_options);
2528 goto done;
2529 }
2530
2531 bs = bdrv_open_inherit(filename, reference, image_options, 0,
2532 parent, child_role, errp);
2533 if (!bs) {
2534 goto done;
2535 }
2536
2537 done:
2538 qdict_del(options, bdref_key);
2539 return bs;
2540 }
2541
2542 /*
2543 * Opens a disk image whose options are given as BlockdevRef in another block
2544 * device's options.
2545 *
2546 * If allow_none is true, no image will be opened if filename is false and no
2547 * BlockdevRef is given. NULL will be returned, but errp remains unset.
2548 *
2549 * bdrev_key specifies the key for the image's BlockdevRef in the options QDict.
2550 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
2551 * itself, all options starting with "${bdref_key}." are considered part of the
2552 * BlockdevRef.
2553 *
2554 * The BlockdevRef will be removed from the options QDict.
2555 */
2556 BdrvChild *bdrv_open_child(const char *filename,
2557 QDict *options, const char *bdref_key,
2558 BlockDriverState *parent,
2559 const BdrvChildRole *child_role,
2560 bool allow_none, Error **errp)
2561 {
2562 BdrvChild *c;
2563 BlockDriverState *bs;
2564
2565 bs = bdrv_open_child_bs(filename, options, bdref_key, parent, child_role,
2566 allow_none, errp);
2567 if (bs == NULL) {
2568 return NULL;
2569 }
2570
2571 c = bdrv_attach_child(parent, bs, bdref_key, child_role, errp);
2572 if (!c) {
2573 bdrv_unref(bs);
2574 return NULL;
2575 }
2576
2577 return c;
2578 }
2579
2580 /* TODO Future callers may need to specify parent/child_role in order for
2581 * option inheritance to work. Existing callers use it for the root node. */
2582 BlockDriverState *bdrv_open_blockdev_ref(BlockdevRef *ref, Error **errp)
2583 {
2584 BlockDriverState *bs = NULL;
2585 Error *local_err = NULL;
2586 QObject *obj = NULL;
2587 QDict *qdict = NULL;
2588 const char *reference = NULL;
2589 Visitor *v = NULL;
2590
2591 if (ref->type == QTYPE_QSTRING) {
2592 reference = ref->u.reference;
2593 } else {
2594 BlockdevOptions *options = &ref->u.definition;
2595 assert(ref->type == QTYPE_QDICT);
2596
2597 v = qobject_output_visitor_new(&obj);
2598 visit_type_BlockdevOptions(v, NULL, &options, &local_err);
2599 if (local_err) {
2600 error_propagate(errp, local_err);
2601 goto fail;
2602 }
2603 visit_complete(v, &obj);
2604
2605 qdict = qobject_to(QDict, obj);
2606 qdict_flatten(qdict);
2607
2608 /* bdrv_open_inherit() defaults to the values in bdrv_flags (for
2609 * compatibility with other callers) rather than what we want as the
2610 * real defaults. Apply the defaults here instead. */
2611 qdict_set_default_str(qdict, BDRV_OPT_CACHE_DIRECT, "off");
2612 qdict_set_default_str(qdict, BDRV_OPT_CACHE_NO_FLUSH, "off");
2613 qdict_set_default_str(qdict, BDRV_OPT_READ_ONLY, "off");
2614 qdict_set_default_str(qdict, BDRV_OPT_AUTO_READ_ONLY, "off");
2615
2616 }
2617
2618 bs = bdrv_open_inherit(NULL, reference, qdict, 0, NULL, NULL, errp);
2619 obj = NULL;
2620
2621 fail:
2622 qobject_unref(obj);
2623 visit_free(v);
2624 return bs;
2625 }
2626
2627 static BlockDriverState *bdrv_append_temp_snapshot(BlockDriverState *bs,
2628 int flags,
2629 QDict *snapshot_options,
2630 Error **errp)
2631 {
2632 /* TODO: extra byte is a hack to ensure MAX_PATH space on Windows. */
2633 char *tmp_filename = g_malloc0(PATH_MAX + 1);
2634 int64_t total_size;
2635 QemuOpts *opts = NULL;
2636 BlockDriverState *bs_snapshot = NULL;
2637 Error *local_err = NULL;
2638 int ret;
2639
2640 /* if snapshot, we create a temporary backing file and open it
2641 instead of opening 'filename' directly */
2642
2643 /* Get the required size from the image */
2644 total_size = bdrv_getlength(bs);
2645 if (total_size < 0) {
2646 error_setg_errno(errp, -total_size, "Could not get image size");
2647 goto out;
2648 }
2649
2650 /* Create the temporary image */
2651 ret = get_tmp_filename(tmp_filename, PATH_MAX + 1);
2652 if (ret < 0) {
2653 error_setg_errno(errp, -ret, "Could not get temporary filename");
2654 goto out;
2655 }
2656
2657 opts = qemu_opts_create(bdrv_qcow2.create_opts, NULL, 0,
2658 &error_abort);
2659 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, total_size, &error_abort);
2660 ret = bdrv_create(&bdrv_qcow2, tmp_filename, opts, errp);
2661 qemu_opts_del(opts);
2662 if (ret < 0) {
2663 error_prepend(errp, "Could not create temporary overlay '%s': ",
2664 tmp_filename);
2665 goto out;
2666 }
2667
2668 /* Prepare options QDict for the temporary file */
2669 qdict_put_str(snapshot_options, "file.driver", "file");
2670 qdict_put_str(snapshot_options, "file.filename", tmp_filename);
2671 qdict_put_str(snapshot_options, "driver", "qcow2");
2672
2673 bs_snapshot = bdrv_open(NULL, NULL, snapshot_options, flags, errp);
2674 snapshot_options = NULL;
2675 if (!bs_snapshot) {
2676 goto out;
2677 }
2678
2679 /* bdrv_append() consumes a strong reference to bs_snapshot
2680 * (i.e. it will call bdrv_unref() on it) even on error, so in
2681 * order to be able to return one, we have to increase
2682 * bs_snapshot's refcount here */
2683 bdrv_ref(bs_snapshot);
2684 bdrv_append(bs_snapshot, bs, &local_err);
2685 if (local_err) {
2686 error_propagate(errp, local_err);
2687 bs_snapshot = NULL;
2688 goto out;
2689 }
2690
2691 out:
2692 qobject_unref(snapshot_options);
2693 g_free(tmp_filename);
2694 return bs_snapshot;
2695 }
2696
2697 /*
2698 * Opens a disk image (raw, qcow2, vmdk, ...)
2699 *
2700 * options is a QDict of options to pass to the block drivers, or NULL for an
2701 * empty set of options. The reference to the QDict belongs to the block layer
2702 * after the call (even on failure), so if the caller intends to reuse the
2703 * dictionary, it needs to use qobject_ref() before calling bdrv_open.
2704 *
2705 * If *pbs is NULL, a new BDS will be created with a pointer to it stored there.
2706 * If it is not NULL, the referenced BDS will be reused.
2707 *
2708 * The reference parameter may be used to specify an existing block device which
2709 * should be opened. If specified, neither options nor a filename may be given,
2710 * nor can an existing BDS be reused (that is, *pbs has to be NULL).
2711 */
2712 static BlockDriverState *bdrv_open_inherit(const char *filename,
2713 const char *reference,
2714 QDict *options, int flags,
2715 BlockDriverState *parent,
2716 const BdrvChildRole *child_role,
2717 Error **errp)
2718 {
2719 int ret;
2720 BlockBackend *file = NULL;
2721 BlockDriverState *bs;
2722 BlockDriver *drv = NULL;
2723 BdrvChild *child;
2724 const char *drvname;
2725 const char *backing;
2726 Error *local_err = NULL;
2727 QDict *snapshot_options = NULL;
2728 int snapshot_flags = 0;
2729
2730 assert(!child_role || !flags);
2731 assert(!child_role == !parent);
2732
2733 if (reference) {
2734 bool options_non_empty = options ? qdict_size(options) : false;
2735 qobject_unref(options);
2736
2737 if (filename || options_non_empty) {
2738 error_setg(errp, "Cannot reference an existing block device with "
2739 "additional options or a new filename");
2740 return NULL;
2741 }
2742
2743 bs = bdrv_lookup_bs(reference, reference, errp);
2744 if (!bs) {
2745 return NULL;
2746 }
2747
2748 bdrv_ref(bs);
2749 return bs;
2750 }
2751
2752 bs = bdrv_new();
2753
2754 /* NULL means an empty set of options */
2755 if (options == NULL) {
2756 options = qdict_new();
2757 }
2758
2759 /* json: syntax counts as explicit options, as if in the QDict */
2760 parse_json_protocol(options, &filename, &local_err);
2761 if (local_err) {
2762 goto fail;
2763 }
2764
2765 bs->explicit_options = qdict_clone_shallow(options);
2766
2767 if (child_role) {
2768 bs->inherits_from = parent;
2769 child_role->inherit_options(&flags, options,
2770 parent->open_flags, parent->options);
2771 }
2772
2773 ret = bdrv_fill_options(&options, filename, &flags, &local_err);
2774 if (local_err) {
2775 goto fail;
2776 }
2777
2778 /*
2779 * Set the BDRV_O_RDWR and BDRV_O_ALLOW_RDWR flags.
2780 * Caution: getting a boolean member of @options requires care.
2781 * When @options come from -blockdev or blockdev_add, members are
2782 * typed according to the QAPI schema, but when they come from
2783 * -drive, they're all QString.
2784 */
2785 if (g_strcmp0(qdict_get_try_str(options, BDRV_OPT_READ_ONLY), "on") &&
2786 !qdict_get_try_bool(options, BDRV_OPT_READ_ONLY, false)) {
2787 flags |= (BDRV_O_RDWR | BDRV_O_ALLOW_RDWR);
2788 } else {
2789 flags &= ~BDRV_O_RDWR;
2790 }
2791
2792 if (flags & BDRV_O_SNAPSHOT) {
2793 snapshot_options = qdict_new();
2794 bdrv_temp_snapshot_options(&snapshot_flags, snapshot_options,
2795 flags, options);
2796 /* Let bdrv_backing_options() override "read-only" */
2797 qdict_del(options, BDRV_OPT_READ_ONLY);
2798 bdrv_backing_options(&flags, options, flags, options);
2799 }
2800
2801 bs->open_flags = flags;
2802 bs->options = options;
2803 options = qdict_clone_shallow(options);
2804
2805 /* Find the right image format driver */
2806 /* See cautionary note on accessing @options above */
2807 drvname = qdict_get_try_str(options, "driver");
2808 if (drvname) {
2809 drv = bdrv_find_format(drvname);
2810 if (!drv) {
2811 error_setg(errp, "Unknown driver: '%s'", drvname);
2812 goto fail;
2813 }
2814 }
2815
2816 assert(drvname || !(flags & BDRV_O_PROTOCOL));
2817
2818 /* See cautionary note on accessing @options above */
2819 backing = qdict_get_try_str(options, "backing");
2820 if (qobject_to(QNull, qdict_get(options, "backing")) != NULL ||
2821 (backing && *backing == '\0'))
2822 {
2823 if (backing) {
2824 warn_report("Use of \"backing\": \"\" is deprecated; "
2825 "use \"backing\": null instead");
2826 }
2827 flags |= BDRV_O_NO_BACKING;
2828 qdict_del(options, "backing");
2829 }
2830
2831 /* Open image file without format layer. This BlockBackend is only used for
2832 * probing, the block drivers will do their own bdrv_open_child() for the
2833 * same BDS, which is why we put the node name back into options. */
2834 if ((flags & BDRV_O_PROTOCOL) == 0) {
2835 BlockDriverState *file_bs;
2836
2837 file_bs = bdrv_open_child_bs(filename, options, "file", bs,
2838 &child_file, true, &local_err);
2839 if (local_err) {
2840 goto fail;
2841 }
2842 if (file_bs != NULL) {
2843 /* Not requesting BLK_PERM_CONSISTENT_READ because we're only
2844 * looking at the header to guess the image format. This works even
2845 * in cases where a guest would not see a consistent state. */
2846 file = blk_new(0, BLK_PERM_ALL);
2847 blk_insert_bs(file, file_bs, &local_err);
2848 bdrv_unref(file_bs);
2849 if (local_err) {
2850 goto fail;
2851 }
2852
2853 qdict_put_str(options, "file", bdrv_get_node_name(file_bs));
2854 }
2855 }
2856
2857 /* Image format probing */
2858 bs->probed = !drv;
2859 if (!drv && file) {
2860 ret = find_image_format(file, filename, &drv, &local_err);
2861 if (ret < 0) {
2862 goto fail;
2863 }
2864 /*
2865 * This option update would logically belong in bdrv_fill_options(),
2866 * but we first need to open bs->file for the probing to work, while
2867 * opening bs->file already requires the (mostly) final set of options
2868 * so that cache mode etc. can be inherited.
2869 *
2870 * Adding the driver later is somewhat ugly, but it's not an option
2871 * that would ever be inherited, so it's correct. We just need to make
2872 * sure to update both bs->options (which has the full effective
2873 * options for bs) and options (which has file.* already removed).
2874 */
2875 qdict_put_str(bs->options, "driver", drv->format_name);
2876 qdict_put_str(options, "driver", drv->format_name);
2877 } else if (!drv) {
2878 error_setg(errp, "Must specify either driver or file");
2879 goto fail;
2880 }
2881
2882 /* BDRV_O_PROTOCOL must be set iff a protocol BDS is about to be created */
2883 assert(!!(flags & BDRV_O_PROTOCOL) == !!drv->bdrv_file_open);
2884 /* file must be NULL if a protocol BDS is about to be created
2885 * (the inverse results in an error message from bdrv_open_common()) */
2886 assert(!(flags & BDRV_O_PROTOCOL) || !file);
2887
2888 /* Open the image */
2889 ret = bdrv_open_common(bs, file, options, &local_err);
2890 if (ret < 0) {
2891 goto fail;
2892 }
2893
2894 if (file) {
2895 blk_unref(file);
2896 file = NULL;
2897 }
2898
2899 /* If there is a backing file, use it */
2900 if ((flags & BDRV_O_NO_BACKING) == 0) {
2901 ret = bdrv_open_backing_file(bs, options, "backing", &local_err);
2902 if (ret < 0) {
2903 goto close_and_fail;
2904 }
2905 }
2906
2907 /* Remove all children options and references
2908 * from bs->options and bs->explicit_options */
2909 QLIST_FOREACH(child, &bs->children, next) {
2910 char *child_key_dot;
2911 child_key_dot = g_strdup_printf("%s.", child->name);
2912 qdict_extract_subqdict(bs->explicit_options, NULL, child_key_dot);
2913 qdict_extract_subqdict(bs->options, NULL, child_key_dot);
2914 qdict_del(bs->explicit_options, child->name);
2915 qdict_del(bs->options, child->name);
2916 g_free(child_key_dot);
2917 }
2918
2919 /* Check if any unknown options were used */
2920 if (qdict_size(options) != 0) {
2921 const QDictEntry *entry = qdict_first(options);
2922 if (flags & BDRV_O_PROTOCOL) {
2923 error_setg(errp, "Block protocol '%s' doesn't support the option "
2924 "'%s'", drv->format_name, entry->key);
2925 } else {
2926 error_setg(errp,
2927 "Block format '%s' does not support the option '%s'",
2928 drv->format_name, entry->key);
2929 }
2930
2931 goto close_and_fail;
2932 }
2933
2934 bdrv_parent_cb_change_media(bs, true);
2935
2936 qobject_unref(options);
2937 options = NULL;
2938
2939 /* For snapshot=on, create a temporary qcow2 overlay. bs points to the
2940 * temporary snapshot afterwards. */
2941 if (snapshot_flags) {
2942 BlockDriverState *snapshot_bs;
2943 snapshot_bs = bdrv_append_temp_snapshot(bs, snapshot_flags,
2944 snapshot_options, &local_err);
2945 snapshot_options = NULL;
2946 if (local_err) {
2947 goto close_and_fail;
2948 }
2949 /* We are not going to return bs but the overlay on top of it
2950 * (snapshot_bs); thus, we have to drop the strong reference to bs
2951 * (which we obtained by calling bdrv_new()). bs will not be deleted,
2952 * though, because the overlay still has a reference to it. */
2953 bdrv_unref(bs);
2954 bs = snapshot_bs;
2955 }
2956
2957 return bs;
2958
2959 fail:
2960 blk_unref(file);
2961 qobject_unref(snapshot_options);
2962 qobject_unref(bs->explicit_options);
2963 qobject_unref(bs->options);
2964 qobject_unref(options);
2965 bs->options = NULL;
2966 bs->explicit_options = NULL;
2967 bdrv_unref(bs);
2968 error_propagate(errp, local_err);
2969 return NULL;
2970
2971 close_and_fail:
2972 bdrv_unref(bs);
2973 qobject_unref(snapshot_options);
2974 qobject_unref(options);
2975 error_propagate(errp, local_err);
2976 return NULL;
2977 }
2978
2979 BlockDriverState *bdrv_open(const char *filename, const char *reference,
2980 QDict *options, int flags, Error **errp)
2981 {
2982 return bdrv_open_inherit(filename, reference, options, flags, NULL,
2983 NULL, errp);
2984 }
2985
2986 /*
2987 * Adds a BlockDriverState to a simple queue for an atomic, transactional
2988 * reopen of multiple devices.
2989 *
2990 * bs_queue can either be an existing BlockReopenQueue that has had QSIMPLE_INIT
2991 * already performed, or alternatively may be NULL a new BlockReopenQueue will
2992 * be created and initialized. This newly created BlockReopenQueue should be
2993 * passed back in for subsequent calls that are intended to be of the same
2994 * atomic 'set'.
2995 *
2996 * bs is the BlockDriverState to add to the reopen queue.
2997 *
2998 * options contains the changed options for the associated bs
2999 * (the BlockReopenQueue takes ownership)
3000 *
3001 * flags contains the open flags for the associated bs
3002 *
3003 * returns a pointer to bs_queue, which is either the newly allocated
3004 * bs_queue, or the existing bs_queue being used.
3005 *
3006 * bs must be drained between bdrv_reopen_queue() and bdrv_reopen_multiple().
3007 */
3008 static BlockReopenQueue *bdrv_reopen_queue_child(BlockReopenQueue *bs_queue,
3009 BlockDriverState *bs,
3010 QDict *options,
3011 const BdrvChildRole *role,
3012 QDict *parent_options,
3013 int parent_flags,
3014 bool keep_old_opts)
3015 {
3016 assert(bs != NULL);
3017
3018 BlockReopenQueueEntry *bs_entry;
3019 BdrvChild *child;
3020 QDict *old_options, *explicit_options, *options_copy;
3021 int flags;
3022 QemuOpts *opts;
3023
3024 /* Make sure that the caller remembered to use a drained section. This is
3025 * important to avoid graph changes between the recursive queuing here and
3026 * bdrv_reopen_multiple(). */
3027 assert(bs->quiesce_counter > 0);
3028
3029 if (bs_queue == NULL) {
3030 bs_queue = g_new0(BlockReopenQueue, 1);
3031 QSIMPLEQ_INIT(bs_queue);
3032 }
3033
3034 if (!options) {
3035 options = qdict_new();
3036 }
3037
3038 /* Check if this BlockDriverState is already in the queue */
3039 QSIMPLEQ_FOREACH(bs_entry, bs_queue, entry) {
3040 if (bs == bs_entry->state.bs) {
3041 break;
3042 }
3043 }
3044
3045 /*
3046 * Precedence of options:
3047 * 1. Explicitly passed in options (highest)
3048 * 2. Retained from explicitly set options of bs
3049 * 3. Inherited from parent node
3050 * 4. Retained from effective options of bs
3051 */
3052
3053 /* Old explicitly set values (don't overwrite by inherited value) */
3054 if (bs_entry || keep_old_opts) {
3055 old_options = qdict_clone_shallow(bs_entry ?
3056 bs_entry->state.explicit_options :
3057 bs->explicit_options);
3058 bdrv_join_options(bs, options, old_options);
3059 qobject_unref(old_options);
3060 }
3061
3062 explicit_options = qdict_clone_shallow(options);
3063
3064 /* Inherit from parent node */
3065 if (parent_options) {
3066 flags = 0;
3067 role->inherit_options(&flags, options, parent_flags, parent_options);
3068 } else {
3069 flags = bdrv_get_flags(bs);
3070 }
3071
3072 if (keep_old_opts) {
3073 /* Old values are used for options that aren't set yet */
3074 old_options = qdict_clone_shallow(bs->options);
3075 bdrv_join_options(bs, options, old_options);
3076 qobject_unref(old_options);
3077 }
3078
3079 /* We have the final set of options so let's update the flags */
3080 options_copy = qdict_clone_shallow(options);
3081 opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
3082 qemu_opts_absorb_qdict(opts, options_copy, NULL);
3083 update_flags_from_options(&flags, opts);
3084 qemu_opts_del(opts);
3085 qobject_unref(options_copy);
3086
3087 /* bdrv_open_inherit() sets and clears some additional flags internally */
3088 flags &= ~BDRV_O_PROTOCOL;
3089 if (flags & BDRV_O_RDWR) {
3090 flags |= BDRV_O_ALLOW_RDWR;
3091 }
3092
3093 if (!bs_entry) {
3094 bs_entry = g_new0(BlockReopenQueueEntry, 1);
3095 QSIMPLEQ_INSERT_TAIL(bs_queue, bs_entry, entry);
3096 } else {
3097 qobject_unref(bs_entry->state.options);
3098 qobject_unref(bs_entry->state.explicit_options);
3099 }
3100
3101 bs_entry->state.bs = bs;
3102 bs_entry->state.options = options;
3103 bs_entry->state.explicit_options = explicit_options;
3104 bs_entry->state.flags = flags;
3105
3106 /* This needs to be overwritten in bdrv_reopen_prepare() */
3107 bs_entry->state.perm = UINT64_MAX;
3108 bs_entry->state.shared_perm = 0;
3109
3110 /*
3111 * If keep_old_opts is false then it means that unspecified
3112 * options must be reset to their original value. We don't allow
3113 * resetting 'backing' but we need to know if the option is
3114 * missing in order to decide if we have to return an error.
3115 */
3116 if (!keep_old_opts) {
3117 bs_entry->state.backing_missing =
3118 !qdict_haskey(options, "backing") &&
3119 !qdict_haskey(options, "backing.driver");
3120 }
3121
3122 QLIST_FOREACH(child, &bs->children, next) {
3123 QDict *new_child_options = NULL;
3124 bool child_keep_old = keep_old_opts;
3125
3126 /* reopen can only change the options of block devices that were
3127 * implicitly created and inherited options. For other (referenced)
3128 * block devices, a syntax like "backing.foo" results in an error. */
3129 if (child->bs->inherits_from != bs) {
3130 continue;
3131 }
3132
3133 /* Check if the options contain a child reference */
3134 if (qdict_haskey(options, child->name)) {
3135 const char *childref = qdict_get_try_str(options, child->name);
3136 /*
3137 * The current child must not be reopened if the child
3138 * reference is null or points to a different node.
3139 */
3140 if (g_strcmp0(childref, child->bs->node_name)) {
3141 continue;
3142 }
3143 /*
3144 * If the child reference points to the current child then
3145 * reopen it with its existing set of options (note that
3146 * it can still inherit new options from the parent).
3147 */
3148 child_keep_old = true;
3149 } else {
3150 /* Extract child options ("child-name.*") */
3151 char *child_key_dot = g_strdup_printf("%s.", child->name);
3152 qdict_extract_subqdict(explicit_options, NULL, child_key_dot);
3153 qdict_extract_subqdict(options, &new_child_options, child_key_dot);
3154 g_free(child_key_dot);
3155 }
3156
3157 bdrv_reopen_queue_child(bs_queue, child->bs, new_child_options,
3158 child->role, options, flags, child_keep_old);
3159 }
3160
3161 return bs_queue;
3162 }
3163
3164 BlockReopenQueue *bdrv_reopen_queue(BlockReopenQueue *bs_queue,
3165 BlockDriverState *bs,
3166 QDict *options, bool keep_old_opts)
3167 {
3168 return bdrv_reopen_queue_child(bs_queue, bs, options, NULL, NULL, 0,
3169 keep_old_opts);
3170 }
3171
3172 /*
3173 * Reopen multiple BlockDriverStates atomically & transactionally.
3174 *
3175 * The queue passed in (bs_queue) must have been built up previous
3176 * via bdrv_reopen_queue().
3177 *
3178 * Reopens all BDS specified in the queue, with the appropriate
3179 * flags. All devices are prepared for reopen, and failure of any
3180 * device will cause all device changes to be abandoned, and intermediate
3181 * data cleaned up.
3182 *
3183 * If all devices prepare successfully, then the changes are committed
3184 * to all devices.
3185 *
3186 * All affected nodes must be drained between bdrv_reopen_queue() and
3187 * bdrv_reopen_multiple().
3188 */
3189 int bdrv_reopen_multiple(AioContext *ctx, BlockReopenQueue *bs_queue, Error **errp)
3190 {
3191 int ret = -1;
3192 BlockReopenQueueEntry *bs_entry, *next;
3193
3194 assert(bs_queue != NULL);
3195
3196 QSIMPLEQ_FOREACH(bs_entry, bs_queue, entry) {
3197 assert(bs_entry->state.bs->quiesce_counter > 0);
3198 if (bdrv_reopen_prepare(&bs_entry->state, bs_queue, errp)) {
3199 goto cleanup;
3200 }
3201 bs_entry->prepared = true;
3202 }
3203
3204 QSIMPLEQ_FOREACH(bs_entry, bs_queue, entry) {
3205 BDRVReopenState *state = &bs_entry->state;
3206 ret = bdrv_check_perm(state->bs, bs_queue, state->perm,
3207 state->shared_perm, NULL, errp);
3208 if (ret < 0) {
3209 goto cleanup_perm;
3210 }
3211 bs_entry->perms_checked = true;
3212 }
3213
3214 /* If we reach this point, we have success and just need to apply the
3215 * changes
3216 */
3217 QSIMPLEQ_FOREACH(bs_entry, bs_queue, entry) {
3218 bdrv_reopen_commit(&bs_entry->state);
3219 }
3220
3221 ret = 0;
3222 cleanup_perm:
3223 QSIMPLEQ_FOREACH_SAFE(bs_entry, bs_queue, entry, next) {
3224 BDRVReopenState *state = &bs_entry->state;
3225
3226 if (!bs_entry->perms_checked) {
3227 continue;
3228 }
3229
3230 if (ret == 0) {
3231 bdrv_set_perm(state->bs, state->perm, state->shared_perm);
3232 } else {
3233 bdrv_abort_perm_update(state->bs);
3234 }
3235 }
3236 cleanup:
3237 QSIMPLEQ_FOREACH_SAFE(bs_entry, bs_queue, entry, next) {
3238 if (ret) {
3239 if (bs_entry->prepared) {
3240 bdrv_reopen_abort(&bs_entry->state);
3241 }
3242 qobject_unref(bs_entry->state.explicit_options);
3243 qobject_unref(bs_entry->state.options);
3244 }
3245 g_free(bs_entry);
3246 }
3247 g_free(bs_queue);
3248
3249 return ret;
3250 }
3251
3252 int bdrv_reopen_set_read_only(BlockDriverState *bs, bool read_only,
3253 Error **errp)
3254 {
3255 int ret;
3256 BlockReopenQueue *queue;
3257 QDict *opts = qdict_new();
3258
3259 qdict_put_bool(opts, BDRV_OPT_READ_ONLY, read_only);
3260
3261 bdrv_subtree_drained_begin(bs);
3262 queue = bdrv_reopen_queue(NULL, bs, opts, true);
3263 ret = bdrv_reopen_multiple(bdrv_get_aio_context(bs), queue, errp);
3264 bdrv_subtree_drained_end(bs);
3265
3266 return ret;
3267 }
3268
3269 static BlockReopenQueueEntry *find_parent_in_reopen_queue(BlockReopenQueue *q,
3270 BdrvChild *c)
3271 {
3272 BlockReopenQueueEntry *entry;
3273
3274 QSIMPLEQ_FOREACH(entry, q, entry) {
3275 BlockDriverState *bs = entry->state.bs;
3276 BdrvChild *child;
3277
3278 QLIST_FOREACH(child, &bs->children, next) {
3279 if (child == c) {
3280 return entry;
3281 }
3282 }
3283 }
3284
3285 return NULL;
3286 }
3287
3288 static void bdrv_reopen_perm(BlockReopenQueue *q, BlockDriverState *bs,
3289 uint64_t *perm, uint64_t *shared)
3290 {
3291 BdrvChild *c;
3292 BlockReopenQueueEntry *parent;
3293 uint64_t cumulative_perms = 0;
3294 uint64_t cumulative_shared_perms = BLK_PERM_ALL;
3295
3296 QLIST_FOREACH(c, &bs->parents, next_parent) {
3297 parent = find_parent_in_reopen_queue(q, c);
3298 if (!parent) {
3299 cumulative_perms |= c->perm;
3300 cumulative_shared_perms &= c->shared_perm;
3301 } else {
3302 uint64_t nperm, nshared;
3303
3304 bdrv_child_perm(parent->state.bs, bs, c, c->role, q,
3305 parent->state.perm, parent->state.shared_perm,
3306 &nperm, &nshared);
3307
3308 cumulative_perms |= nperm;
3309 cumulative_shared_perms &= nshared;
3310 }
3311 }
3312 *perm = cumulative_perms;
3313 *shared = cumulative_shared_perms;
3314 }
3315
3316 /*
3317 * Prepares a BlockDriverState for reopen. All changes are staged in the
3318 * 'opaque' field of the BDRVReopenState, which is used and allocated by
3319 * the block driver layer .bdrv_reopen_prepare()
3320 *
3321 * bs is the BlockDriverState to reopen
3322 * flags are the new open flags
3323 * queue is the reopen queue
3324 *
3325 * Returns 0 on success, non-zero on error. On error errp will be set
3326 * as well.
3327 *
3328 * On failure, bdrv_reopen_abort() will be called to clean up any data.
3329 * It is the responsibility of the caller to then call the abort() or
3330 * commit() for any other BDS that have been left in a prepare() state
3331 *
3332 */
3333 int bdrv_reopen_prepare(BDRVReopenState *reopen_state, BlockReopenQueue *queue,
3334 Error **errp)
3335 {
3336 int ret = -1;
3337 int old_flags;
3338 Error *local_err = NULL;
3339 BlockDriver *drv;
3340 QemuOpts *opts;
3341 QDict *orig_reopen_opts;
3342 char *discard = NULL;
3343 bool read_only;
3344 bool drv_prepared = false;
3345
3346 assert(reopen_state != NULL);
3347 assert(reopen_state->bs->drv != NULL);
3348 drv = reopen_state->bs->drv;
3349
3350 /* This function and each driver's bdrv_reopen_prepare() remove
3351 * entries from reopen_state->options as they are processed, so
3352 * we need to make a copy of the original QDict. */
3353 orig_reopen_opts = qdict_clone_shallow(reopen_state->options);
3354
3355 /* Process generic block layer options */
3356 opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
3357 qemu_opts_absorb_qdict(opts, reopen_state->options, &local_err);
3358 if (local_err) {
3359 error_propagate(errp, local_err);
3360 ret = -EINVAL;
3361 goto error;
3362 }
3363
3364 /* This was already called in bdrv_reopen_queue_child() so the flags
3365 * are up-to-date. This time we simply want to remove the options from
3366 * QemuOpts in order to indicate that they have been processed. */
3367 old_flags = reopen_state->flags;
3368 update_flags_from_options(&reopen_state->flags, opts);
3369 assert(old_flags == reopen_state->flags);
3370
3371 discard = qemu_opt_get_del(opts, BDRV_OPT_DISCARD);
3372 if (discard != NULL) {
3373 if (bdrv_parse_discard_flags(discard, &reopen_state->flags) != 0) {
3374 error_setg(errp, "Invalid discard option");
3375 ret = -EINVAL;
3376 goto error;
3377 }
3378 }
3379
3380 reopen_state->detect_zeroes =
3381 bdrv_parse_detect_zeroes(opts, reopen_state->flags, &local_err);
3382 if (local_err) {
3383 error_propagate(errp, local_err);
3384 ret = -EINVAL;
3385 goto error;
3386 }
3387
3388 /* All other options (including node-name and driver) must be unchanged.
3389 * Put them back into the QDict, so that they are checked at the end
3390 * of this function. */
3391 qemu_opts_to_qdict(opts, reopen_state->options);
3392
3393 /* If we are to stay read-only, do not allow permission change
3394 * to r/w. Attempting to set to r/w may fail if either BDRV_O_ALLOW_RDWR is
3395 * not set, or if the BDS still has copy_on_read enabled */
3396 read_only = !(reopen_state->flags & BDRV_O_RDWR);
3397 ret = bdrv_can_set_read_only(reopen_state->bs, read_only, true, &local_err);
3398 if (local_err) {
3399 error_propagate(errp, local_err);
3400 goto error;
3401 }
3402
3403 /* Calculate required permissions after reopening */
3404 bdrv_reopen_perm(queue, reopen_state->bs,
3405 &reopen_state->perm, &reopen_state->shared_perm);
3406
3407 ret = bdrv_flush(reopen_state->bs);
3408 if (ret) {
3409 error_setg_errno(errp, -ret, "Error flushing drive");
3410 goto error;
3411 }
3412
3413 if (drv->bdrv_reopen_prepare) {
3414 ret = drv->bdrv_reopen_prepare(reopen_state, queue, &local_err);
3415 if (ret) {
3416 if (local_err != NULL) {
3417 error_propagate(errp, local_err);
3418 } else {
3419 bdrv_refresh_filename(reopen_state->bs);
3420 error_setg(errp, "failed while preparing to reopen image '%s'",
3421 reopen_state->bs->filename);
3422 }
3423 goto error;
3424 }
3425 } else {
3426 /* It is currently mandatory to have a bdrv_reopen_prepare()
3427 * handler for each supported drv. */
3428 error_setg(errp, "Block format '%s' used by node '%s' "
3429 "does not support reopening files", drv->format_name,
3430 bdrv_get_device_or_node_name(reopen_state->bs));
3431 ret = -1;
3432 goto error;
3433 }
3434
3435 drv_prepared = true;
3436
3437 /*
3438 * We must provide the 'backing' option if the BDS has a backing
3439 * file or if the image file has a backing file name as part of
3440 * its metadata. Otherwise the 'backing' option can be omitted.
3441 */
3442 if (drv->supports_backing && reopen_state->backing_missing &&
3443 (backing_bs(reopen_state->bs) || reopen_state->bs->backing_file[0])) {
3444 error_setg(errp, "backing is missing for '%s'",
3445 reopen_state->bs->node_name);
3446 ret = -EINVAL;
3447 goto error;
3448 }
3449
3450 /* Options that are not handled are only okay if they are unchanged
3451 * compared to the old state. It is expected that some options are only
3452 * used for the initial open, but not reopen (e.g. filename) */
3453 if (qdict_size(reopen_state->options)) {
3454 const QDictEntry *entry = qdict_first(reopen_state->options);
3455
3456 do {
3457 QObject *new = entry->value;
3458 QObject *old = qdict_get(reopen_state->bs->options, entry->key);
3459
3460 /* Allow child references (child_name=node_name) as long as they
3461 * point to the current child (i.e. everything stays the same). */
3462 if (qobject_type(new) == QTYPE_QSTRING) {
3463 BdrvChild *child;
3464 QLIST_FOREACH(child, &reopen_state->bs->children, next) {
3465 if (!strcmp(child->name, entry->key)) {
3466 break;
3467 }
3468 }
3469
3470 if (child) {
3471 const char *str = qobject_get_try_str(new);
3472 if (!strcmp(child->bs->node_name, str)) {
3473 continue; /* Found child with this name, skip option */
3474 }
3475 }
3476 }
3477
3478 /*
3479 * TODO: When using -drive to specify blockdev options, all values
3480 * will be strings; however, when using -blockdev, blockdev-add or
3481 * filenames using the json:{} pseudo-protocol, they will be
3482 * correctly typed.
3483 * In contrast, reopening options are (currently) always strings
3484 * (because you can only specify them through qemu-io; all other
3485 * callers do not specify any options).
3486 * Therefore, when using anything other than -drive to create a BDS,
3487 * this cannot detect non-string options as unchanged, because
3488 * qobject_is_equal() always returns false for objects of different
3489 * type. In the future, this should be remedied by correctly typing
3490 * all options. For now, this is not too big of an issue because
3491 * the user can simply omit options which cannot be changed anyway,
3492 * so they will stay unchanged.
3493 */
3494 if (!qobject_is_equal(new, old)) {
3495 error_setg(errp, "Cannot change the option '%s'", entry->key);
3496 ret = -EINVAL;
3497 goto error;
3498 }
3499 } while ((entry = qdict_next(reopen_state->options, entry)));
3500 }
3501
3502 ret = 0;
3503
3504 /* Restore the original reopen_state->options QDict */
3505 qobject_unref(reopen_state->options);
3506 reopen_state->options = qobject_ref(orig_reopen_opts);
3507
3508 error:
3509 if (ret < 0 && drv_prepared) {
3510 /* drv->bdrv_reopen_prepare() has succeeded, so we need to
3511 * call drv->bdrv_reopen_abort() before signaling an error
3512 * (bdrv_reopen_multiple() will not call bdrv_reopen_abort()
3513 * when the respective bdrv_reopen_prepare() has failed) */
3514 if (drv->bdrv_reopen_abort) {
3515 drv->bdrv_reopen_abort(reopen_state);
3516 }
3517 }
3518 qemu_opts_del(opts);
3519 qobject_unref(orig_reopen_opts);
3520 g_free(discard);
3521 return ret;
3522 }
3523
3524 /*
3525 * Takes the staged changes for the reopen from bdrv_reopen_prepare(), and
3526 * makes them final by swapping the staging BlockDriverState contents into
3527 * the active BlockDriverState contents.
3528 */
3529 void bdrv_reopen_commit(BDRVReopenState *reopen_state)
3530 {
3531 BlockDriver *drv;
3532 BlockDriverState *bs;
3533 BdrvChild *child;
3534 bool old_can_write, new_can_write;
3535
3536 assert(reopen_state != NULL);
3537 bs = reopen_state->bs;
3538 drv = bs->drv;
3539 assert(drv != NULL);
3540
3541 old_can_write =
3542 !bdrv_is_read_only(bs) && !(bdrv_get_flags(bs) & BDRV_O_INACTIVE);
3543
3544 /* If there are any driver level actions to take */
3545 if (drv->bdrv_reopen_commit) {
3546 drv->bdrv_reopen_commit(reopen_state);
3547 }
3548
3549 /* set BDS specific flags now */
3550 qobject_unref(bs->explicit_options);
3551 qobject_unref(bs->options);
3552
3553 bs->explicit_options = reopen_state->explicit_options;
3554 bs->options = reopen_state->options;
3555 bs->open_flags = reopen_state->flags;
3556 bs->read_only = !(reopen_state->flags & BDRV_O_RDWR);
3557 bs->detect_zeroes = reopen_state->detect_zeroes;
3558
3559 /* Remove child references from bs->options and bs->explicit_options.
3560 * Child options were already removed in bdrv_reopen_queue_child() */
3561 QLIST_FOREACH(child, &bs->children, next) {
3562 qdict_del(bs->explicit_options, child->name);
3563 qdict_del(bs->options, child->name);
3564 }
3565
3566 bdrv_refresh_limits(bs, NULL);
3567
3568 new_can_write =
3569 !bdrv_is_read_only(bs) && !(bdrv_get_flags(bs) & BDRV_O_INACTIVE);
3570 if (!old_can_write && new_can_write && drv->bdrv_reopen_bitmaps_rw) {
3571 Error *local_err = NULL;
3572 if (drv->bdrv_reopen_bitmaps_rw(bs, &local_err) < 0) {
3573 /* This is not fatal, bitmaps just left read-only, so all following
3574 * writes will fail. User can remove read-only bitmaps to unblock
3575 * writes.
3576 */
3577 error_reportf_err(local_err,
3578 "%s: Failed to make dirty bitmaps writable: ",
3579 bdrv_get_node_name(bs));
3580 }
3581 }
3582 }
3583
3584 /*
3585 * Abort the reopen, and delete and free the staged changes in
3586 * reopen_state
3587 */
3588 void bdrv_reopen_abort(BDRVReopenState *reopen_state)
3589 {
3590 BlockDriver *drv;
3591
3592 assert(reopen_state != NULL);
3593 drv = reopen_state->bs->drv;
3594 assert(drv != NULL);
3595
3596 if (drv->bdrv_reopen_abort) {
3597 drv->bdrv_reopen_abort(reopen_state);
3598 }
3599 }
3600
3601
3602 static void bdrv_close(BlockDriverState *bs)
3603 {
3604 BdrvAioNotifier *ban, *ban_next;
3605 BdrvChild *child, *next;
3606
3607 assert(!bs->job);
3608 assert(!bs->refcnt);
3609
3610 bdrv_drained_begin(bs); /* complete I/O */
3611 bdrv_flush(bs);
3612 bdrv_drain(bs); /* in case flush left pending I/O */
3613
3614 if (bs->drv) {
3615 if (bs->drv->bdrv_close) {
3616 bs->drv->bdrv_close(bs);
3617 }
3618 bs->drv = NULL;
3619 }
3620
3621 bdrv_set_backing_hd(bs, NULL, &error_abort);
3622
3623 if (bs->file != NULL) {
3624 bdrv_unref_child(bs, bs->file);
3625 bs->file = NULL;
3626 }
3627
3628 QLIST_FOREACH_SAFE(child, &bs->children, next, next) {
3629 /* TODO Remove bdrv_unref() from drivers' close function and use
3630 * bdrv_unref_child() here */
3631 if (child->bs->inherits_from == bs) {
3632 child->bs->inherits_from = NULL;
3633 }
3634 bdrv_detach_child(child);
3635 }
3636
3637 g_free(bs->opaque);
3638 bs->opaque = NULL;
3639 atomic_set(&bs->copy_on_read, 0);
3640 bs->backing_file[0] = '\0';
3641 bs->backing_format[0] = '\0';
3642 bs->total_sectors = 0;
3643 bs->encrypted = false;
3644 bs->sg = false;
3645 qobject_unref(bs->options);
3646 qobject_unref(bs->explicit_options);
3647 bs->options = NULL;
3648 bs->explicit_options = NULL;
3649 qobject_unref(bs->full_open_options);
3650 bs->full_open_options = NULL;
3651
3652 bdrv_release_named_dirty_bitmaps(bs);
3653 assert(QLIST_EMPTY(&bs->dirty_bitmaps));
3654
3655 QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_next) {
3656 g_free(ban);
3657 }
3658 QLIST_INIT(&bs->aio_notifiers);
3659 bdrv_drained_end(bs);
3660 }
3661
3662 void bdrv_close_all(void)
3663 {
3664 assert(job_next(NULL) == NULL);
3665 nbd_export_close_all();
3666
3667 /* Drop references from requests still in flight, such as canceled block
3668 * jobs whose AIO context has not been polled yet */
3669 bdrv_drain_all();
3670
3671 blk_remove_all_bs();
3672 blockdev_close_all_bdrv_states();
3673
3674 assert(QTAILQ_EMPTY(&all_bdrv_states));
3675 }
3676
3677 static bool should_update_child(BdrvChild *c, BlockDriverState *to)
3678 {
3679 GQueue *queue;
3680 GHashTable *found;
3681 bool ret;
3682
3683 if (c->role->stay_at_node) {
3684 return false;
3685 }
3686
3687 /* If the child @c belongs to the BDS @to, replacing the current
3688 * c->bs by @to would mean to create a loop.
3689 *
3690 * Such a case occurs when appending a BDS to a backing chain.
3691 * For instance, imagine the following chain:
3692 *
3693 * guest device -> node A -> further backing chain...
3694 *
3695 * Now we create a new BDS B which we want to put on top of this
3696 * chain, so we first attach A as its backing node:
3697 *
3698 * node B
3699 * |
3700 * v
3701 * guest device -> node A -> further backing chain...
3702 *
3703 * Finally we want to replace A by B. When doing that, we want to
3704 * replace all pointers to A by pointers to B -- except for the
3705 * pointer from B because (1) that would create a loop, and (2)
3706 * that pointer should simply stay intact:
3707 *
3708 * guest device -> node B
3709 * |
3710 * v
3711 * node A -> further backing chain...
3712 *
3713 * In general, when replacing a node A (c->bs) by a node B (@to),
3714 * if A is a child of B, that means we cannot replace A by B there
3715 * because that would create a loop. Silently detaching A from B
3716 * is also not really an option. So overall just leaving A in
3717 * place there is the most sensible choice.
3718 *
3719 * We would also create a loop in any cases where @c is only
3720 * indirectly referenced by @to. Prevent this by returning false
3721 * if @c is found (by breadth-first search) anywhere in the whole
3722 * subtree of @to.
3723 */
3724
3725 ret = true;
3726 found = g_hash_table_new(NULL, NULL);
3727 g_hash_table_add(found, to);
3728 queue = g_queue_new();
3729 g_queue_push_tail(queue, to);
3730
3731 while (!g_queue_is_empty(queue)) {
3732 BlockDriverState *v = g_queue_pop_head(queue);
3733 BdrvChild *c2;
3734
3735 QLIST_FOREACH(c2, &v->children, next) {
3736 if (c2 == c) {
3737 ret = false;
3738 break;
3739 }
3740
3741 if (g_hash_table_contains(found, c2->bs)) {
3742 continue;
3743 }
3744
3745 g_queue_push_tail(queue, c2->bs);
3746 g_hash_table_add(found, c2->bs);
3747 }
3748 }
3749
3750 g_queue_free(queue);
3751 g_hash_table_destroy(found);
3752
3753 return ret;
3754 }
3755
3756 void bdrv_replace_node(BlockDriverState *from, BlockDriverState *to,
3757 Error **errp)
3758 {
3759 BdrvChild *c, *next;
3760 GSList *list = NULL, *p;
3761 uint64_t old_perm, old_shared;
3762 uint64_t perm = 0, shared = BLK_PERM_ALL;
3763 int ret;
3764
3765 assert(!atomic_read(&from->in_flight));
3766 assert(!atomic_read(&to->in_flight));
3767
3768 /* Make sure that @from doesn't go away until we have successfully attached
3769 * all of its parents to @to. */
3770 bdrv_ref(from);
3771
3772 /* Put all parents into @list and calculate their cumulative permissions */
3773 QLIST_FOREACH_SAFE(c, &from->parents, next_parent, next) {
3774 assert(c->bs == from);
3775 if (!should_update_child(c, to)) {
3776 continue;
3777 }
3778 if (c->frozen) {
3779 error_setg(errp, "Cannot change '%s' link to '%s'",
3780 c->name, from->node_name);
3781 goto out;
3782 }
3783 list = g_slist_prepend(list, c);
3784 perm |= c->perm;
3785 shared &= c->shared_perm;
3786 }
3787
3788 /* Check whether the required permissions can be granted on @to, ignoring
3789 * all BdrvChild in @list so that they can't block themselves. */
3790 ret = bdrv_check_update_perm(to, NULL, perm, shared, list, errp);
3791 if (ret < 0) {
3792 bdrv_abort_perm_update(to);
3793 goto out;
3794 }
3795
3796 /* Now actually perform the change. We performed the permission check for
3797 * all elements of @list at once, so set the permissions all at once at the
3798 * very end. */
3799 for (p = list; p != NULL; p = p->next) {
3800 c = p->data;
3801
3802 bdrv_ref(to);
3803 bdrv_replace_child_noperm(c, to);
3804 bdrv_unref(from);
3805 }
3806
3807 bdrv_get_cumulative_perm(to, &old_perm, &old_shared);
3808 bdrv_set_perm(to, old_perm | perm, old_shared | shared);
3809
3810 out:
3811 g_slist_free(list);
3812 bdrv_unref(from);
3813 }
3814
3815 /*
3816 * Add new bs contents at the top of an image chain while the chain is
3817 * live, while keeping required fields on the top layer.
3818 *
3819 * This will modify the BlockDriverState fields, and swap contents
3820 * between bs_new and bs_top. Both bs_new and bs_top are modified.
3821 *
3822 * bs_new must not be attached to a BlockBackend.
3823 *
3824 * This function does not create any image files.
3825 *
3826 * bdrv_append() takes ownership of a bs_new reference and unrefs it because
3827 * that's what the callers commonly need. bs_new will be referenced by the old
3828 * parents of bs_top after bdrv_append() returns. If the caller needs to keep a
3829 * reference of its own, it must call bdrv_ref().
3830 */
3831 void bdrv_append(BlockDriverState *bs_new, BlockDriverState *bs_top,
3832 Error **errp)
3833 {
3834 Error *local_err = NULL;
3835
3836 bdrv_set_backing_hd(bs_new, bs_top, &local_err);
3837 if (local_err) {
3838 error_propagate(errp, local_err);
3839 goto out;
3840 }
3841
3842 bdrv_replace_node(bs_top, bs_new, &local_err);
3843 if (local_err) {
3844 error_propagate(errp, local_err);
3845 bdrv_set_backing_hd(bs_new, NULL, &error_abort);
3846 goto out;
3847 }
3848
3849 /* bs_new is now referenced by its new parents, we don't need the
3850 * additional reference any more. */
3851 out:
3852 bdrv_unref(bs_new);
3853 }
3854
3855 static void bdrv_delete(BlockDriverState *bs)
3856 {
3857 assert(!bs->job);
3858 assert(bdrv_op_blocker_is_empty(bs));
3859 assert(!bs->refcnt);
3860
3861 bdrv_close(bs);
3862
3863 /* remove from list, if necessary */
3864 if (bs->node_name[0] != '\0') {
3865 QTAILQ_REMOVE(&graph_bdrv_states, bs, node_list);
3866 }
3867 QTAILQ_REMOVE(&all_bdrv_states, bs, bs_list);
3868
3869 g_free(bs);
3870 }
3871
3872 /*
3873 * Run consistency checks on an image
3874 *
3875 * Returns 0 if the check could be completed (it doesn't mean that the image is
3876 * free of errors) or -errno when an internal error occurred. The results of the
3877 * check are stored in res.
3878 */
3879 static int coroutine_fn bdrv_co_check(BlockDriverState *bs,
3880 BdrvCheckResult *res, BdrvCheckMode fix)
3881 {
3882 if (bs->drv == NULL) {
3883 return -ENOMEDIUM;
3884 }
3885 if (bs->drv->bdrv_co_check == NULL) {
3886 return -ENOTSUP;
3887 }
3888
3889 memset(res, 0, sizeof(*res));
3890 return bs->drv->bdrv_co_check(bs, res, fix);
3891 }
3892
3893 typedef struct CheckCo {
3894 BlockDriverState *bs;
3895 BdrvCheckResult *res;
3896 BdrvCheckMode fix;
3897 int ret;
3898 } CheckCo;
3899
3900 static void bdrv_check_co_entry(void *opaque)
3901 {
3902 CheckCo *cco = opaque;
3903 cco->ret = bdrv_co_check(cco->bs, cco->res, cco->fix);
3904 aio_wait_kick();
3905 }
3906
3907 int bdrv_check(BlockDriverState *bs,
3908 BdrvCheckResult *res, BdrvCheckMode fix)
3909 {
3910 Coroutine *co;
3911 CheckCo cco = {
3912 .bs = bs,
3913 .res = res,
3914 .ret = -EINPROGRESS,
3915 .fix = fix,
3916 };
3917
3918 if (qemu_in_coroutine()) {
3919 /* Fast-path if already in coroutine context */
3920 bdrv_check_co_entry(&cco);
3921 } else {
3922 co = qemu_coroutine_create(bdrv_check_co_entry, &cco);
3923 bdrv_coroutine_enter(bs, co);
3924 BDRV_POLL_WHILE(bs, cco.ret == -EINPROGRESS);
3925 }
3926
3927 return cco.ret;
3928 }
3929
3930 /*
3931 * Return values:
3932 * 0 - success
3933 * -EINVAL - backing format specified, but no file
3934 * -ENOSPC - can't update the backing file because no space is left in the
3935 * image file header
3936 * -ENOTSUP - format driver doesn't support changing the backing file
3937 */
3938 int bdrv_change_backing_file(BlockDriverState *bs,
3939 const char *backing_file, const char *backing_fmt)
3940 {
3941 BlockDriver *drv = bs->drv;
3942 int ret;
3943
3944 if (!drv) {
3945 return -ENOMEDIUM;
3946 }
3947
3948 /* Backing file format doesn't make sense without a backing file */
3949 if (backing_fmt && !backing_file) {
3950 return -EINVAL;
3951 }
3952
3953 if (drv->bdrv_change_backing_file != NULL) {
3954 ret = drv->bdrv_change_backing_file(bs, backing_file, backing_fmt);
3955 } else {
3956 ret = -ENOTSUP;
3957 }
3958
3959 if (ret == 0) {
3960 pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_file ?: "");
3961 pstrcpy(bs->backing_format, sizeof(bs->backing_format), backing_fmt ?: "");
3962 pstrcpy(bs->auto_backing_file, sizeof(bs->auto_backing_file),
3963 backing_file ?: "");
3964 }
3965 return ret;
3966 }
3967
3968 /*
3969 * Finds the image layer in the chain that has 'bs' as its backing file.
3970 *
3971 * active is the current topmost image.
3972 *
3973 * Returns NULL if bs is not found in active's image chain,
3974 * or if active == bs.
3975 *
3976 * Returns the bottommost base image if bs == NULL.
3977 */
3978 BlockDriverState *bdrv_find_overlay(BlockDriverState *active,
3979 BlockDriverState *bs)
3980 {
3981 while (active && bs != backing_bs(active)) {
3982 active = backing_bs(active);
3983 }
3984
3985 return active;
3986 }
3987
3988 /* Given a BDS, searches for the base layer. */
3989 BlockDriverState *bdrv_find_base(BlockDriverState *bs)
3990 {
3991 return bdrv_find_overlay(bs, NULL);
3992 }
3993
3994 /*
3995 * Return true if at least one of the backing links between @bs and
3996 * @base is frozen. @errp is set if that's the case.
3997 */
3998 bool bdrv_is_backing_chain_frozen(BlockDriverState *bs, BlockDriverState *base,
3999 Error **errp)
4000 {
4001 BlockDriverState *i;
4002
4003 for (i = bs; i != base && i->backing; i = backing_bs(i)) {
4004 if (i->backing->frozen) {
4005 error_setg(errp, "Cannot change '%s' link from '%s' to '%s'",
4006 i->backing->name, i->node_name,
4007 backing_bs(i)->node_name);
4008 return true;
4009 }
4010 }
4011
4012 return false;
4013 }
4014
4015 /*
4016 * Freeze all backing links between @bs and @base.
4017 * If any of the links is already frozen the operation is aborted and
4018 * none of the links are modified.
4019 * Returns 0 on success. On failure returns < 0 and sets @errp.
4020 */
4021 int bdrv_freeze_backing_chain(BlockDriverState *bs, BlockDriverState *base,
4022 Error **errp)
4023 {
4024 BlockDriverState *i;
4025
4026 if (bdrv_is_backing_chain_frozen(bs, base, errp)) {
4027 return -EPERM;
4028 }
4029
4030 for (i = bs; i != base && i->backing; i = backing_bs(i)) {
4031 i->backing->frozen = true;
4032 }
4033
4034 return 0;
4035 }
4036
4037 /*
4038 * Unfreeze all backing links between @bs and @base. The caller must
4039 * ensure that all links are frozen before using this function.
4040 */
4041 void bdrv_unfreeze_backing_chain(BlockDriverState *bs, BlockDriverState *base)
4042 {
4043 BlockDriverState *i;
4044
4045 for (i = bs; i != base && i->backing; i = backing_bs(i)) {
4046 assert(i->backing->frozen);
4047 i->backing->frozen = false;
4048 }
4049 }
4050
4051 /*
4052 * Drops images above 'base' up to and including 'top', and sets the image
4053 * above 'top' to have base as its backing file.
4054 *
4055 * Requires that the overlay to 'top' is opened r/w, so that the backing file
4056 * information in 'bs' can be properly updated.
4057 *
4058 * E.g., this will convert the following chain:
4059 * bottom <- base <- intermediate <- top <- active
4060 *
4061 * to
4062 *
4063 * bottom <- base <- active
4064 *
4065 * It is allowed for bottom==base, in which case it converts:
4066 *
4067 * base <- intermediate <- top <- active
4068 *
4069 * to
4070 *
4071 * base <- active
4072 *
4073 * If backing_file_str is non-NULL, it will be used when modifying top's
4074 * overlay image metadata.
4075 *
4076 * Error conditions:
4077 * if active == top, that is considered an error
4078 *
4079 */
4080 int bdrv_drop_intermediate(BlockDriverState *top, BlockDriverState *base,
4081 const char *backing_file_str)
4082 {
4083 BlockDriverState *explicit_top = top;
4084 bool update_inherits_from;
4085 BdrvChild *c, *next;
4086 Error *local_err = NULL;
4087 int ret = -EIO;
4088
4089 bdrv_ref(top);
4090
4091 if (!top->drv || !base->drv) {
4092 goto exit;
4093 }
4094
4095 /* Make sure that base is in the backing chain of top */
4096 if (!bdrv_chain_contains(top, base)) {
4097 goto exit;
4098 }
4099
4100 /* This function changes all links that point to top and makes
4101 * them point to base. Check that none of them is frozen. */
4102 QLIST_FOREACH(c, &top->parents, next_parent) {
4103 if (c->frozen) {
4104 goto exit;
4105 }
4106 }
4107
4108 /* If 'base' recursively inherits from 'top' then we should set
4109 * base->inherits_from to top->inherits_from after 'top' and all
4110 * other intermediate nodes have been dropped.
4111 * If 'top' is an implicit node (e.g. "commit_top") we should skip
4112 * it because no one inherits from it. We use explicit_top for that. */
4113 while (explicit_top && explicit_top->implicit) {
4114 explicit_top = backing_bs(explicit_top);
4115 }
4116 update_inherits_from = bdrv_inherits_from_recursive(base, explicit_top);
4117
4118 /* success - we can delete the intermediate states, and link top->base */
4119 /* TODO Check graph modification op blockers (BLK_PERM_GRAPH_MOD) once
4120 * we've figured out how they should work. */
4121 if (!backing_file_str) {
4122 bdrv_refresh_filename(base);
4123 backing_file_str = base->filename;
4124 }
4125
4126 QLIST_FOREACH_SAFE(c, &top->parents, next_parent, next) {
4127 /* Check whether we are allowed to switch c from top to base */
4128 GSList *ignore_children = g_slist_prepend(NULL, c);
4129 bdrv_check_update_perm(base, NULL, c->perm, c->shared_perm,
4130 ignore_children, &local_err);
4131 g_slist_free(ignore_children);
4132 if (local_err) {
4133 ret = -EPERM;
4134 error_report_err(local_err);
4135 goto exit;
4136 }
4137
4138 /* If so, update the backing file path in the image file */
4139 if (c->role->update_filename) {
4140 ret = c->role->update_filename(c, base, backing_file_str,
4141 &local_err);
4142 if (ret < 0) {
4143 bdrv_abort_perm_update(base);
4144 error_report_err(local_err);
4145 goto exit;
4146 }
4147 }
4148
4149 /* Do the actual switch in the in-memory graph.
4150 * Completes bdrv_check_update_perm() transaction internally. */
4151 bdrv_ref(base);
4152 bdrv_replace_child(c, base);
4153 bdrv_unref(top);
4154 }
4155
4156 if (update_inherits_from) {
4157 base->inherits_from = explicit_top->inherits_from;
4158 }
4159
4160 ret = 0;
4161 exit:
4162 bdrv_unref(top);
4163 return ret;
4164 }
4165
4166 /**
4167 * Length of a allocated file in bytes. Sparse files are counted by actual
4168 * allocated space. Return < 0 if error or unknown.
4169 */
4170 int64_t bdrv_get_allocated_file_size(BlockDriverState *bs)
4171 {
4172 BlockDriver *drv = bs->drv;
4173 if (!drv) {
4174 return -ENOMEDIUM;
4175 }
4176 if (drv->bdrv_get_allocated_file_size) {
4177 return drv->bdrv_get_allocated_file_size(bs);
4178 }
4179 if (bs->file) {
4180 return bdrv_get_allocated_file_size(bs->file->bs);
4181 }
4182 return -ENOTSUP;
4183 }
4184
4185 /*
4186 * bdrv_measure:
4187 * @drv: Format driver
4188 * @opts: Creation options for new image
4189 * @in_bs: Existing image containing data for new image (may be NULL)
4190 * @errp: Error object
4191 * Returns: A #BlockMeasureInfo (free using qapi_free_BlockMeasureInfo())
4192 * or NULL on error
4193 *
4194 * Calculate file size required to create a new image.
4195 *
4196 * If @in_bs is given then space for allocated clusters and zero clusters
4197 * from that image are included in the calculation. If @opts contains a
4198 * backing file that is shared by @in_bs then backing clusters may be omitted
4199 * from the calculation.
4200 *
4201 * If @in_bs is NULL then the calculation includes no allocated clusters
4202 * unless a preallocation option is given in @opts.
4203 *
4204 * Note that @in_bs may use a different BlockDriver from @drv.
4205 *
4206 * If an error occurs the @errp pointer is set.
4207 */
4208 BlockMeasureInfo *bdrv_measure(BlockDriver *drv, QemuOpts *opts,
4209 BlockDriverState *in_bs, Error **errp)
4210 {
4211 if (!drv->bdrv_measure) {
4212 error_setg(errp, "Block driver '%s' does not support size measurement",
4213 drv->format_name);
4214 return NULL;
4215 }
4216
4217 return drv->bdrv_measure(opts, in_bs, errp);
4218 }
4219
4220 /**
4221 * Return number of sectors on success, -errno on error.
4222 */
4223 int64_t bdrv_nb_sectors(BlockDriverState *bs)
4224 {
4225 BlockDriver *drv = bs->drv;
4226
4227 if (!drv)
4228 return -ENOMEDIUM;
4229
4230 if (drv->has_variable_length) {
4231 int ret = refresh_total_sectors(bs, bs->total_sectors);
4232 if (ret < 0) {
4233 return ret;
4234 }
4235 }
4236 return bs->total_sectors;
4237 }
4238
4239 /**
4240 * Return length in bytes on success, -errno on error.
4241 * The length is always a multiple of BDRV_SECTOR_SIZE.
4242 */
4243 int64_t bdrv_getlength(BlockDriverState *bs)
4244 {
4245 int64_t ret = bdrv_nb_sectors(bs);
4246
4247 ret = ret > INT64_MAX / BDRV_SECTOR_SIZE ? -EFBIG : ret;
4248 return ret < 0 ? ret : ret * BDRV_SECTOR_SIZE;
4249 }
4250
4251 /* return 0 as number of sectors if no device present or error */
4252 void bdrv_get_geometry(BlockDriverState *bs, uint64_t *nb_sectors_ptr)
4253 {
4254 int64_t nb_sectors = bdrv_nb_sectors(bs);
4255
4256 *nb_sectors_ptr = nb_sectors < 0 ? 0 : nb_sectors;
4257 }
4258
4259 bool bdrv_is_sg(BlockDriverState *bs)
4260 {
4261 return bs->sg;
4262 }
4263
4264 bool bdrv_is_encrypted(BlockDriverState *bs)
4265 {
4266 if (bs->backing && bs->backing->bs->encrypted) {
4267 return true;
4268 }
4269 return bs->encrypted;
4270 }
4271
4272 const char *bdrv_get_format_name(BlockDriverState *bs)
4273 {
4274 return bs->drv ? bs->drv->format_name : NULL;
4275 }
4276
4277 static int qsort_strcmp(const void *a, const void *b)
4278 {
4279 return strcmp(*(char *const *)a, *(char *const *)b);
4280 }
4281
4282 void bdrv_iterate_format(void (*it)(void *opaque, const char *name),
4283 void *opaque, bool read_only)
4284 {
4285 BlockDriver *drv;
4286 int count = 0;
4287 int i;
4288 const char **formats = NULL;
4289
4290 QLIST_FOREACH(drv, &bdrv_drivers, list) {
4291 if (drv->format_name) {
4292 bool found = false;
4293 int i = count;
4294
4295 if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv, read_only)) {
4296 continue;
4297 }
4298
4299 while (formats && i && !found) {
4300 found = !strcmp(formats[--i], drv->format_name);
4301 }
4302
4303 if (!found) {
4304 formats = g_renew(const char *, formats, count + 1);
4305 formats[count++] = drv->format_name;
4306 }
4307 }
4308 }
4309
4310 for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); i++) {
4311 const char *format_name = block_driver_modules[i].format_name;
4312
4313 if (format_name) {
4314 bool found = false;
4315 int j = count;
4316
4317 if (use_bdrv_whitelist &&
4318 !bdrv_format_is_whitelisted(format_name, read_only)) {
4319 continue;
4320 }
4321
4322 while (formats && j && !found) {
4323 found = !strcmp(formats[--j], format_name);
4324 }
4325
4326 if (!found) {
4327 formats = g_renew(const char *, formats, count + 1);
4328 formats[count++] = format_name;
4329 }
4330 }
4331 }
4332
4333 qsort(formats, count, sizeof(formats[0]), qsort_strcmp);
4334
4335 for (i = 0; i < count; i++) {
4336 it(opaque, formats[i]);
4337 }
4338
4339 g_free(formats);
4340 }
4341
4342 /* This function is to find a node in the bs graph */
4343 BlockDriverState *bdrv_find_node(const char *node_name)
4344 {
4345 BlockDriverState *bs;
4346
4347 assert(node_name);
4348
4349 QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
4350 if (!strcmp(node_name, bs->node_name)) {
4351 return bs;
4352 }
4353 }
4354 return NULL;
4355 }
4356
4357 /* Put this QMP function here so it can access the static graph_bdrv_states. */
4358 BlockDeviceInfoList *bdrv_named_nodes_list(Error **errp)
4359 {
4360 BlockDeviceInfoList *list, *entry;
4361 BlockDriverState *bs;
4362
4363 list = NULL;
4364 QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
4365 BlockDeviceInfo *info = bdrv_block_device_info(NULL, bs, errp);
4366 if (!info) {
4367 qapi_free_BlockDeviceInfoList(list);
4368 return NULL;
4369 }
4370 entry = g_malloc0(sizeof(*entry));
4371 entry->value = info;
4372 entry->next = list;
4373 list = entry;
4374 }
4375
4376 return list;
4377 }
4378
4379 #define QAPI_LIST_ADD(list, element) do { \
4380 typeof(list) _tmp = g_new(typeof(*(list)), 1); \
4381 _tmp->value = (element); \
4382 _tmp->next = (list); \
4383 (list) = _tmp; \
4384 } while (0)
4385
4386 typedef struct XDbgBlockGraphConstructor {
4387 XDbgBlockGraph *graph;
4388 GHashTable *graph_nodes;
4389 } XDbgBlockGraphConstructor;
4390
4391 static XDbgBlockGraphConstructor *xdbg_graph_new(void)
4392 {
4393 XDbgBlockGraphConstructor *gr = g_new(XDbgBlockGraphConstructor, 1);
4394
4395 gr->graph = g_new0(XDbgBlockGraph, 1);
4396 gr->graph_nodes = g_hash_table_new(NULL, NULL);
4397
4398 return gr;
4399 }
4400
4401 static XDbgBlockGraph *xdbg_graph_finalize(XDbgBlockGraphConstructor *gr)
4402 {
4403 XDbgBlockGraph *graph = gr->graph;
4404
4405 g_hash_table_destroy(gr->graph_nodes);
4406 g_free(gr);
4407
4408 return graph;
4409 }
4410
4411 static uintptr_t xdbg_graph_node_num(XDbgBlockGraphConstructor *gr, void *node)
4412 {
4413 uintptr_t ret = (uintptr_t)g_hash_table_lookup(gr->graph_nodes, node);
4414
4415 if (ret != 0) {
4416 return ret;
4417 }
4418
4419 /*
4420 * Start counting from 1, not 0, because 0 interferes with not-found (NULL)
4421 * answer of g_hash_table_lookup.
4422 */
4423 ret = g_hash_table_size(gr->graph_nodes) + 1;
4424 g_hash_table_insert(gr->graph_nodes, node, (void *)ret);
4425
4426 return ret;
4427 }
4428
4429 static void xdbg_graph_add_node(XDbgBlockGraphConstructor *gr, void *node,
4430 XDbgBlockGraphNodeType type, const char *name)
4431 {
4432 XDbgBlockGraphNode *n;
4433
4434 n = g_new0(XDbgBlockGraphNode, 1);
4435
4436 n->id = xdbg_graph_node_num(gr, node);
4437 n->type = type;
4438 n->name = g_strdup(name);
4439
4440 QAPI_LIST_ADD(gr->graph->nodes, n);
4441 }
4442
4443 static void xdbg_graph_add_edge(XDbgBlockGraphConstructor *gr, void *parent,
4444 const BdrvChild *child)
4445 {
4446 typedef struct {
4447 unsigned int flag;
4448 BlockPermission num;
4449 } PermissionMap;
4450
4451 static const PermissionMap permissions[] = {
4452 { BLK_PERM_CONSISTENT_READ, BLOCK_PERMISSION_CONSISTENT_READ },
4453 { BLK_PERM_WRITE, BLOCK_PERMISSION_WRITE },
4454 { BLK_PERM_WRITE_UNCHANGED, BLOCK_PERMISSION_WRITE_UNCHANGED },
4455 { BLK_PERM_RESIZE, BLOCK_PERMISSION_RESIZE },
4456 { BLK_PERM_GRAPH_MOD, BLOCK_PERMISSION_GRAPH_MOD },
4457 { 0, 0 }
4458 };
4459 const PermissionMap *p;
4460 XDbgBlockGraphEdge *edge;
4461
4462 QEMU_BUILD_BUG_ON(1UL << (ARRAY_SIZE(permissions) - 1) != BLK_PERM_ALL + 1);
4463
4464 edge = g_new0(XDbgBlockGraphEdge, 1);
4465
4466 edge->parent = xdbg_graph_node_num(gr, parent);
4467 edge->child = xdbg_graph_node_num(gr, child->bs);
4468 edge->name = g_strdup(child->name);
4469
4470 for (p = permissions; p->flag; p++) {
4471 if (p->flag & child->perm) {
4472 QAPI_LIST_ADD(edge->perm, p->num);
4473 }
4474 if (p->flag & child->shared_perm) {
4475 QAPI_LIST_ADD(edge->shared_perm, p->num);
4476 }
4477 }
4478
4479 QAPI_LIST_ADD(gr->graph->edges, edge);
4480 }
4481
4482
4483 XDbgBlockGraph *bdrv_get_xdbg_block_graph(Error **errp)
4484 {
4485 BlockBackend *blk;
4486 BlockJob *job;
4487 BlockDriverState *bs;
4488 BdrvChild *child;
4489 XDbgBlockGraphConstructor *gr = xdbg_graph_new();
4490
4491 for (blk = blk_all_next(NULL); blk; blk = blk_all_next(blk)) {
4492 char *allocated_name = NULL;
4493 const char *name = blk_name(blk);
4494
4495 if (!*name) {
4496 name = allocated_name = blk_get_attached_dev_id(blk);
4497 }
4498 xdbg_graph_add_node(gr, blk, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_BACKEND,
4499 name);
4500 g_free(allocated_name);
4501 if (blk_root(blk)) {
4502 xdbg_graph_add_edge(gr, blk, blk_root(blk));
4503 }
4504 }
4505
4506 for (job = block_job_next(NULL); job; job = block_job_next(job)) {
4507 GSList *el;
4508
4509 xdbg_graph_add_node(gr, job, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_JOB,
4510 job->job.id);
4511 for (el = job->nodes; el; el = el->next) {
4512 xdbg_graph_add_edge(gr, job, (BdrvChild *)el->data);
4513 }
4514 }
4515
4516 QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
4517 xdbg_graph_add_node(gr, bs, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_DRIVER,
4518 bs->node_name);
4519 QLIST_FOREACH(child, &bs->children, next) {
4520 xdbg_graph_add_edge(gr, bs, child);
4521 }
4522 }
4523
4524 return xdbg_graph_finalize(gr);
4525 }
4526
4527 BlockDriverState *bdrv_lookup_bs(const char *device,
4528 const char *node_name,
4529 Error **errp)
4530 {
4531 BlockBackend *blk;
4532 BlockDriverState *bs;
4533
4534 if (device) {
4535 blk = blk_by_name(device);
4536
4537 if (blk) {
4538 bs = blk_bs(blk);
4539 if (!bs) {
4540 error_setg(errp, "Device '%s' has no medium", device);
4541 }
4542
4543 return bs;
4544 }
4545 }
4546
4547 if (node_name) {
4548 bs = bdrv_find_node(node_name);
4549
4550 if (bs) {
4551 return bs;
4552 }
4553 }
4554
4555 error_setg(errp, "Cannot find device=%s nor node_name=%s",
4556 device ? device : "",
4557 node_name ? node_name : "");
4558 return NULL;
4559 }
4560
4561 /* If 'base' is in the same chain as 'top', return true. Otherwise,
4562 * return false. If either argument is NULL, return false. */
4563 bool bdrv_chain_contains(BlockDriverState *top, BlockDriverState *base)
4564 {
4565 while (top && top != base) {
4566 top = backing_bs(top);
4567 }
4568
4569 return top != NULL;
4570 }
4571
4572 BlockDriverState *bdrv_next_node(BlockDriverState *bs)
4573 {
4574 if (!bs) {
4575 return QTAILQ_FIRST(&graph_bdrv_states);
4576 }
4577 return QTAILQ_NEXT(bs, node_list);
4578 }
4579
4580 BlockDriverState *bdrv_next_all_states(BlockDriverState *bs)
4581 {
4582 if (!bs) {
4583 return QTAILQ_FIRST(&all_bdrv_states);
4584 }
4585 return QTAILQ_NEXT(bs, bs_list);
4586 }
4587
4588 const char *bdrv_get_node_name(const BlockDriverState *bs)
4589 {
4590 return bs->node_name;
4591 }
4592
4593 const char *bdrv_get_parent_name(const BlockDriverState *bs)
4594 {
4595 BdrvChild *c;
4596 const char *name;
4597
4598 /* If multiple parents have a name, just pick the first one. */
4599 QLIST_FOREACH(c, &bs->parents, next_parent) {
4600 if (c->role->get_name) {
4601 name = c->role->get_name(c);
4602 if (name && *name) {
4603 return name;
4604 }
4605 }
4606 }
4607
4608 return NULL;
4609 }
4610
4611 /* TODO check what callers really want: bs->node_name or blk_name() */
4612 const char *bdrv_get_device_name(const BlockDriverState *bs)
4613 {
4614 return bdrv_get_parent_name(bs) ?: "";
4615 }
4616
4617 /* This can be used to identify nodes that might not have a device
4618 * name associated. Since node and device names live in the same
4619 * namespace, the result is unambiguous. The exception is if both are
4620 * absent, then this returns an empty (non-null) string. */
4621 const char *bdrv_get_device_or_node_name(const BlockDriverState *bs)
4622 {
4623 return bdrv_get_parent_name(bs) ?: bs->node_name;
4624 }
4625
4626 int bdrv_get_flags(BlockDriverState *bs)
4627 {
4628 return bs->open_flags;
4629 }
4630
4631 int bdrv_has_zero_init_1(BlockDriverState *bs)
4632 {
4633 return 1;
4634 }
4635
4636 int bdrv_has_zero_init(BlockDriverState *bs)
4637 {
4638 if (!bs->drv) {
4639 return 0;
4640 }
4641
4642 /* If BS is a copy on write image, it is initialized to
4643 the contents of the base image, which may not be zeroes. */
4644 if (bs->backing) {
4645 return 0;
4646 }
4647 if (bs->drv->bdrv_has_zero_init) {
4648 return bs->drv->bdrv_has_zero_init(bs);
4649 }
4650 if (bs->file && bs->drv->is_filter) {
4651 return bdrv_has_zero_init(bs->file->bs);
4652 }
4653
4654 /* safe default */
4655 return 0;
4656 }
4657
4658 bool bdrv_unallocated_blocks_are_zero(BlockDriverState *bs)
4659 {
4660 BlockDriverInfo bdi;
4661
4662 if (bs->backing) {
4663 return false;
4664 }
4665
4666 if (bdrv_get_info(bs, &bdi) == 0) {
4667 return bdi.unallocated_blocks_are_zero;
4668 }
4669
4670 return false;
4671 }
4672
4673 bool bdrv_can_write_zeroes_with_unmap(BlockDriverState *bs)
4674 {
4675 if (!(bs->open_flags & BDRV_O_UNMAP)) {
4676 return false;
4677 }
4678
4679 return bs->supported_zero_flags & BDRV_REQ_MAY_UNMAP;
4680 }
4681
4682 void bdrv_get_backing_filename(BlockDriverState *bs,
4683 char *filename, int filename_size)
4684 {
4685 pstrcpy(filename, filename_size, bs->backing_file);
4686 }
4687
4688 int bdrv_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
4689 {
4690 BlockDriver *drv = bs->drv;
4691 /* if bs->drv == NULL, bs is closed, so there's nothing to do here */
4692 if (!drv) {
4693 return -ENOMEDIUM;
4694 }
4695 if (!drv->bdrv_get_info) {
4696 if (bs->file && drv->is_filter) {
4697 return bdrv_get_info(bs->file->bs, bdi);
4698 }
4699 return -ENOTSUP;
4700 }
4701 memset(bdi, 0, sizeof(*bdi));
4702 return drv->bdrv_get_info(bs, bdi);
4703 }
4704
4705 ImageInfoSpecific *bdrv_get_specific_info(BlockDriverState *bs,
4706 Error **errp)
4707 {
4708 BlockDriver *drv = bs->drv;
4709 if (drv && drv->bdrv_get_specific_info) {
4710 return drv->bdrv_get_specific_info(bs, errp);
4711 }
4712 return NULL;
4713 }
4714
4715 void bdrv_debug_event(BlockDriverState *bs, BlkdebugEvent event)
4716 {
4717 if (!bs || !bs->drv || !bs->drv->bdrv_debug_event) {
4718 return;
4719 }
4720
4721 bs->drv->bdrv_debug_event(bs, event);
4722 }
4723
4724 int bdrv_debug_breakpoint(BlockDriverState *bs, const char *event,
4725 const char *tag)
4726 {
4727 while (bs && bs->drv && !bs->drv->bdrv_debug_breakpoint) {
4728 bs = bs->file ? bs->file->bs : NULL;
4729 }
4730
4731 if (bs && bs->drv && bs->drv->bdrv_debug_breakpoint) {
4732 return bs->drv->bdrv_debug_breakpoint(bs, event, tag);
4733 }
4734
4735 return -ENOTSUP;
4736 }
4737
4738 int bdrv_debug_remove_breakpoint(BlockDriverState *bs, const char *tag)
4739 {
4740 while (bs && bs->drv && !bs->drv->bdrv_debug_remove_breakpoint) {
4741 bs = bs->file ? bs->file->bs : NULL;
4742 }
4743
4744 if (bs && bs->drv && bs->drv->bdrv_debug_remove_breakpoint) {
4745 return bs->drv->bdrv_debug_remove_breakpoint(bs, tag);
4746 }
4747
4748 return -ENOTSUP;
4749 }
4750
4751 int bdrv_debug_resume(BlockDriverState *bs, const char *tag)
4752 {
4753 while (bs && (!bs->drv || !bs->drv->bdrv_debug_resume)) {
4754 bs = bs->file ? bs->file->bs : NULL;
4755 }
4756
4757 if (bs && bs->drv && bs->drv->bdrv_debug_resume) {
4758 return bs->drv->bdrv_debug_resume(bs, tag);
4759 }
4760
4761 return -ENOTSUP;
4762 }
4763
4764 bool bdrv_debug_is_suspended(BlockDriverState *bs, const char *tag)
4765 {
4766 while (bs && bs->drv && !bs->drv->bdrv_debug_is_suspended) {
4767 bs = bs->file ? bs->file->bs : NULL;
4768 }
4769
4770 if (bs && bs->drv && bs->drv->bdrv_debug_is_suspended) {
4771 return bs->drv->bdrv_debug_is_suspended(bs, tag);
4772 }
4773
4774 return false;
4775 }
4776
4777 /* backing_file can either be relative, or absolute, or a protocol. If it is
4778 * relative, it must be relative to the chain. So, passing in bs->filename
4779 * from a BDS as backing_file should not be done, as that may be relative to
4780 * the CWD rather than the chain. */
4781 BlockDriverState *bdrv_find_backing_image(BlockDriverState *bs,
4782 const char *backing_file)
4783 {
4784 char *filename_full = NULL;
4785 char *backing_file_full = NULL;
4786 char *filename_tmp = NULL;
4787 int is_protocol = 0;
4788 BlockDriverState *curr_bs = NULL;
4789 BlockDriverState *retval = NULL;
4790
4791 if (!bs || !bs->drv || !backing_file) {
4792 return NULL;
4793 }
4794
4795 filename_full = g_malloc(PATH_MAX);
4796 backing_file_full = g_malloc(PATH_MAX);
4797
4798 is_protocol = path_has_protocol(backing_file);
4799
4800 for (curr_bs = bs; curr_bs->backing; curr_bs = curr_bs->backing->bs) {
4801
4802 /* If either of the filename paths is actually a protocol, then
4803 * compare unmodified paths; otherwise make paths relative */
4804 if (is_protocol || path_has_protocol(curr_bs->backing_file)) {
4805 char *backing_file_full_ret;
4806
4807 if (strcmp(backing_file, curr_bs->backing_file) == 0) {
4808 retval = curr_bs->backing->bs;
4809 break;
4810 }
4811 /* Also check against the full backing filename for the image */
4812 backing_file_full_ret = bdrv_get_full_backing_filename(curr_bs,
4813 NULL);
4814 if (backing_file_full_ret) {
4815 bool equal = strcmp(backing_file, backing_file_full_ret) == 0;
4816 g_free(backing_file_full_ret);
4817 if (equal) {
4818 retval = curr_bs->backing->bs;
4819 break;
4820 }
4821 }
4822 } else {
4823 /* If not an absolute filename path, make it relative to the current
4824 * image's filename path */
4825 filename_tmp = bdrv_make_absolute_filename(curr_bs, backing_file,
4826 NULL);
4827 /* We are going to compare canonicalized absolute pathnames */
4828 if (!filename_tmp || !realpath(filename_tmp, filename_full)) {
4829 g_free(filename_tmp);
4830 continue;
4831 }
4832 g_free(filename_tmp);
4833
4834 /* We need to make sure the backing filename we are comparing against
4835 * is relative to the current image filename (or absolute) */
4836 filename_tmp = bdrv_get_full_backing_filename(curr_bs, NULL);
4837 if (!filename_tmp || !realpath(filename_tmp, backing_file_full)) {
4838 g_free(filename_tmp);
4839 continue;
4840 }
4841 g_free(filename_tmp);
4842
4843 if (strcmp(backing_file_full, filename_full) == 0) {
4844 retval = curr_bs->backing->bs;
4845 break;
4846 }
4847 }
4848 }
4849
4850 g_free(filename_full);
4851 g_free(backing_file_full);
4852 return retval;
4853 }
4854
4855 void bdrv_init(void)
4856 {
4857 module_call_init(MODULE_INIT_BLOCK);
4858 }
4859
4860 void bdrv_init_with_whitelist(void)
4861 {
4862 use_bdrv_whitelist = 1;
4863 bdrv_init();
4864 }
4865
4866 static void coroutine_fn bdrv_co_invalidate_cache(BlockDriverState *bs,
4867 Error **errp)
4868 {
4869 BdrvChild *child, *parent;
4870 uint64_t perm, shared_perm;
4871 Error *local_err = NULL;
4872 int ret;
4873 BdrvDirtyBitmap *bm;
4874
4875 if (!bs->drv) {
4876 return;
4877 }
4878
4879 if (!(bs->open_flags & BDRV_O_INACTIVE)) {
4880 return;
4881 }
4882
4883 QLIST_FOREACH(child, &bs->children, next) {
4884 bdrv_co_invalidate_cache(child->bs, &local_err);
4885 if (local_err) {
4886 error_propagate(errp, local_err);
4887 return;
4888 }
4889 }
4890
4891 /*
4892 * Update permissions, they may differ for inactive nodes.
4893 *
4894 * Note that the required permissions of inactive images are always a
4895 * subset of the permissions required after activating the image. This
4896 * allows us to just get the permissions upfront without restricting
4897 * drv->bdrv_invalidate_cache().
4898 *
4899 * It also means that in error cases, we don't have to try and revert to
4900 * the old permissions (which is an operation that could fail, too). We can
4901 * just keep the extended permissions for the next time that an activation
4902 * of the image is tried.
4903 */
4904 bs->open_flags &= ~BDRV_O_INACTIVE;
4905 bdrv_get_cumulative_perm(bs, &perm, &shared_perm);
4906 ret = bdrv_check_perm(bs, NULL, perm, shared_perm, NULL, &local_err);
4907 if (ret < 0) {
4908 bs->open_flags |= BDRV_O_INACTIVE;
4909 error_propagate(errp, local_err);
4910 return;
4911 }
4912 bdrv_set_perm(bs, perm, shared_perm);
4913
4914 if (bs->drv->bdrv_co_invalidate_cache) {
4915 bs->drv->bdrv_co_invalidate_cache(bs, &local_err);
4916 if (local_err) {
4917 bs->open_flags |= BDRV_O_INACTIVE;
4918 error_propagate(errp, local_err);
4919 return;
4920 }
4921 }
4922
4923 for (bm = bdrv_dirty_bitmap_next(bs, NULL); bm;
4924 bm = bdrv_dirty_bitmap_next(bs, bm))
4925 {
4926 bdrv_dirty_bitmap_set_migration(bm, false);
4927 }
4928
4929 ret = refresh_total_sectors(bs, bs->total_sectors);
4930 if (ret < 0) {
4931 bs->open_flags |= BDRV_O_INACTIVE;
4932 error_setg_errno(errp, -ret, "Could not refresh total sector count");
4933 return;
4934 }
4935
4936 QLIST_FOREACH(parent, &bs->parents, next_parent) {
4937 if (parent->role->activate) {
4938 parent->role->activate(parent, &local_err);
4939 if (local_err) {
4940 bs->open_flags |= BDRV_O_INACTIVE;
4941 error_propagate(errp, local_err);
4942 return;
4943 }
4944 }
4945 }
4946 }
4947
4948 typedef struct InvalidateCacheCo {
4949 BlockDriverState *bs;
4950 Error **errp;
4951 bool done;
4952 } InvalidateCacheCo;
4953
4954 static void coroutine_fn bdrv_invalidate_cache_co_entry(void *opaque)
4955 {
4956 InvalidateCacheCo *ico = opaque;
4957 bdrv_co_invalidate_cache(ico->bs, ico->errp);
4958 ico->done = true;
4959 aio_wait_kick();
4960 }
4961
4962 void bdrv_invalidate_cache(BlockDriverState *bs, Error **errp)
4963 {
4964 Coroutine *co;
4965 InvalidateCacheCo ico = {
4966 .bs = bs,
4967 .done = false,
4968 .errp = errp
4969 };
4970
4971 if (qemu_in_coroutine()) {
4972 /* Fast-path if already in coroutine context */
4973 bdrv_invalidate_cache_co_entry(&ico);
4974 } else {
4975 co = qemu_coroutine_create(bdrv_invalidate_cache_co_entry, &ico);
4976 bdrv_coroutine_enter(bs, co);
4977 BDRV_POLL_WHILE(bs, !ico.done);
4978 }
4979 }
4980
4981 void bdrv_invalidate_cache_all(Error **errp)
4982 {
4983 BlockDriverState *bs;
4984 Error *local_err = NULL;
4985 BdrvNextIterator it;
4986
4987 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
4988 AioContext *aio_context = bdrv_get_aio_context(bs);
4989
4990 aio_context_acquire(aio_context);
4991 bdrv_invalidate_cache(bs, &local_err);
4992 aio_context_release(aio_context);
4993 if (local_err) {
4994 error_propagate(errp, local_err);
4995 bdrv_next_cleanup(&it);
4996 return;
4997 }
4998 }
4999 }
5000
5001 static bool bdrv_has_bds_parent(BlockDriverState *bs, bool only_active)
5002 {
5003 BdrvChild *parent;
5004
5005 QLIST_FOREACH(parent, &bs->parents, next_parent) {
5006 if (parent->role->parent_is_bds) {
5007 BlockDriverState *parent_bs = parent->opaque;
5008 if (!only_active || !(parent_bs->open_flags & BDRV_O_INACTIVE)) {
5009 return true;
5010 }
5011 }
5012 }
5013
5014 return false;
5015 }
5016
5017 static int bdrv_inactivate_recurse(BlockDriverState *bs)
5018 {
5019 BdrvChild *child, *parent;
5020 uint64_t perm, shared_perm;
5021 int ret;
5022
5023 if (!bs->drv) {
5024 return -ENOMEDIUM;
5025 }
5026
5027 /* Make sure that we don't inactivate a child before its parent.
5028 * It will be covered by recursion from the yet active parent. */
5029 if (bdrv_has_bds_parent(bs, true)) {
5030 return 0;
5031 }
5032
5033 assert(!(bs->open_flags & BDRV_O_INACTIVE));
5034
5035 /* Inactivate this node */
5036 if (bs->drv->bdrv_inactivate) {
5037 ret = bs->drv->bdrv_inactivate(bs);
5038 if (ret < 0) {
5039 return ret;
5040 }
5041 }
5042
5043 QLIST_FOREACH(parent, &bs->parents, next_parent) {
5044 if (parent->role->inactivate) {
5045 ret = parent->role->inactivate(parent);
5046 if (ret < 0) {
5047 return ret;
5048 }
5049 }
5050 }
5051
5052 bs->open_flags |= BDRV_O_INACTIVE;
5053
5054 /* Update permissions, they may differ for inactive nodes */
5055 bdrv_get_cumulative_perm(bs, &perm, &shared_perm);
5056 bdrv_check_perm(bs, NULL, perm, shared_perm, NULL, &error_abort);
5057 bdrv_set_perm(bs, perm, shared_perm);
5058
5059
5060 /* Recursively inactivate children */
5061 QLIST_FOREACH(child, &bs->children, next) {
5062 ret = bdrv_inactivate_recurse(child->bs);
5063 if (ret < 0) {
5064 return ret;
5065 }
5066 }
5067
5068 return 0;
5069 }
5070
5071 int bdrv_inactivate_all(void)
5072 {
5073 BlockDriverState *bs = NULL;
5074 BdrvNextIterator it;
5075 int ret = 0;
5076 GSList *aio_ctxs = NULL, *ctx;
5077
5078 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
5079 AioContext *aio_context = bdrv_get_aio_context(bs);
5080
5081 if (!g_slist_find(aio_ctxs, aio_context)) {
5082 aio_ctxs = g_slist_prepend(aio_ctxs, aio_context);
5083 aio_context_acquire(aio_context);
5084 }
5085 }
5086
5087 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
5088 /* Nodes with BDS parents are covered by recursion from the last
5089 * parent that gets inactivated. Don't inactivate them a second
5090 * time if that has already happened. */
5091 if (bdrv_has_bds_parent(bs, false)) {
5092 continue;
5093 }
5094 ret = bdrv_inactivate_recurse(bs);
5095 if (ret < 0) {
5096 bdrv_next_cleanup(&it);
5097 goto out;
5098 }
5099 }
5100
5101 out:
5102 for (ctx = aio_ctxs; ctx != NULL; ctx = ctx->next) {
5103 AioContext *aio_context = ctx->data;
5104 aio_context_release(aio_context);
5105 }
5106 g_slist_free(aio_ctxs);
5107
5108 return ret;
5109 }
5110
5111 /**************************************************************/
5112 /* removable device support */
5113
5114 /**
5115 * Return TRUE if the media is present
5116 */
5117 bool bdrv_is_inserted(BlockDriverState *bs)
5118 {
5119 BlockDriver *drv = bs->drv;
5120 BdrvChild *child;
5121
5122 if (!drv) {
5123 return false;
5124 }
5125 if (drv->bdrv_is_inserted) {
5126 return drv->bdrv_is_inserted(bs);
5127 }
5128 QLIST_FOREACH(child, &bs->children, next) {
5129 if (!bdrv_is_inserted(child->bs)) {
5130 return false;
5131 }
5132 }
5133 return true;
5134 }
5135
5136 /**
5137 * If eject_flag is TRUE, eject the media. Otherwise, close the tray
5138 */
5139 void bdrv_eject(BlockDriverState *bs, bool eject_flag)
5140 {
5141 BlockDriver *drv = bs->drv;
5142
5143 if (drv && drv->bdrv_eject) {
5144 drv->bdrv_eject(bs, eject_flag);
5145 }
5146 }
5147
5148 /**
5149 * Lock or unlock the media (if it is locked, the user won't be able
5150 * to eject it manually).
5151 */
5152 void bdrv_lock_medium(BlockDriverState *bs, bool locked)
5153 {
5154 BlockDriver *drv = bs->drv;
5155
5156 trace_bdrv_lock_medium(bs, locked);
5157
5158 if (drv && drv->bdrv_lock_medium) {
5159 drv->bdrv_lock_medium(bs, locked);
5160 }
5161 }
5162
5163 /* Get a reference to bs */
5164 void bdrv_ref(BlockDriverState *bs)
5165 {
5166 bs->refcnt++;
5167 }
5168
5169 /* Release a previously grabbed reference to bs.
5170 * If after releasing, reference count is zero, the BlockDriverState is
5171 * deleted. */
5172 void bdrv_unref(BlockDriverState *bs)
5173 {
5174 if (!bs) {
5175 return;
5176 }
5177 assert(bs->refcnt > 0);
5178 if (--bs->refcnt == 0) {
5179 bdrv_delete(bs);
5180 }
5181 }
5182
5183 struct BdrvOpBlocker {
5184 Error *reason;
5185 QLIST_ENTRY(BdrvOpBlocker) list;
5186 };
5187
5188 bool bdrv_op_is_blocked(BlockDriverState *bs, BlockOpType op, Error **errp)
5189 {
5190 BdrvOpBlocker *blocker;
5191 assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
5192 if (!QLIST_EMPTY(&bs->op_blockers[op])) {
5193 blocker = QLIST_FIRST(&bs->op_blockers[op]);
5194 error_propagate_prepend(errp, error_copy(blocker->reason),
5195 "Node '%s' is busy: ",
5196 bdrv_get_device_or_node_name(bs));
5197 return true;
5198 }
5199 return false;
5200 }
5201
5202 void bdrv_op_block(BlockDriverState *bs, BlockOpType op, Error *reason)
5203 {
5204 BdrvOpBlocker *blocker;
5205 assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
5206
5207 blocker = g_new0(BdrvOpBlocker, 1);
5208 blocker->reason = reason;
5209 QLIST_INSERT_HEAD(&bs->op_blockers[op], blocker, list);
5210 }
5211
5212 void bdrv_op_unblock(BlockDriverState *bs, BlockOpType op, Error *reason)
5213 {
5214 BdrvOpBlocker *blocker, *next;
5215 assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
5216 QLIST_FOREACH_SAFE(blocker, &bs->op_blockers[op], list, next) {
5217 if (blocker->reason == reason) {
5218 QLIST_REMOVE(blocker, list);
5219 g_free(blocker);
5220 }
5221 }
5222 }
5223
5224 void bdrv_op_block_all(BlockDriverState *bs, Error *reason)
5225 {
5226 int i;
5227 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
5228 bdrv_op_block(bs, i, reason);
5229 }
5230 }
5231
5232 void bdrv_op_unblock_all(BlockDriverState *bs, Error *reason)
5233 {
5234 int i;
5235 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
5236 bdrv_op_unblock(bs, i, reason);
5237 }
5238 }
5239
5240 bool bdrv_op_blocker_is_empty(BlockDriverState *bs)
5241 {
5242 int i;
5243
5244 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
5245 if (!QLIST_EMPTY(&bs->op_blockers[i])) {
5246 return false;
5247 }
5248 }
5249 return true;
5250 }
5251
5252 void bdrv_img_create(const char *filename, const char *fmt,
5253 const char *base_filename, const char *base_fmt,
5254 char *options, uint64_t img_size, int flags, bool quiet,
5255 Error **errp)
5256 {
5257 QemuOptsList *create_opts = NULL;
5258 QemuOpts *opts = NULL;
5259 const char *backing_fmt, *backing_file;
5260 int64_t size;
5261 BlockDriver *drv, *proto_drv;
5262 Error *local_err = NULL;
5263 int ret = 0;
5264
5265 /* Find driver and parse its options */
5266 drv = bdrv_find_format(fmt);
5267 if (!drv) {
5268 error_setg(errp, "Unknown file format '%s'", fmt);
5269 return;
5270 }
5271
5272 proto_drv = bdrv_find_protocol(filename, true, errp);
5273 if (!proto_drv) {
5274 return;
5275 }
5276
5277 if (!drv->create_opts) {
5278 error_setg(errp, "Format driver '%s' does not support image creation",
5279 drv->format_name);
5280 return;
5281 }
5282
5283 if (!proto_drv->create_opts) {
5284 error_setg(errp, "Protocol driver '%s' does not support image creation",
5285 proto_drv->format_name);
5286 return;
5287 }
5288
5289 create_opts = qemu_opts_append(create_opts, drv->create_opts);
5290 create_opts = qemu_opts_append(create_opts, proto_drv->create_opts);
5291
5292 /* Create parameter list with default values */
5293 opts = qemu_opts_create(create_opts, NULL, 0, &error_abort);
5294 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, img_size, &error_abort);
5295
5296 /* Parse -o options */
5297 if (options) {
5298 qemu_opts_do_parse(opts, options, NULL, &local_err);
5299 if (local_err) {
5300 goto out;
5301 }
5302 }
5303
5304 if (base_filename) {
5305 qemu_opt_set(opts, BLOCK_OPT_BACKING_FILE, base_filename, &local_err);
5306 if (local_err) {
5307 error_setg(errp, "Backing file not supported for file format '%s'",
5308 fmt);
5309 goto out;
5310 }
5311 }
5312
5313 if (base_fmt) {
5314 qemu_opt_set(opts, BLOCK_OPT_BACKING_FMT, base_fmt, &local_err);
5315 if (local_err) {
5316 error_setg(errp, "Backing file format not supported for file "
5317 "format '%s'", fmt);
5318 goto out;
5319 }
5320 }
5321
5322 backing_file = qemu_opt_get(opts, BLOCK_OPT_BACKING_FILE);
5323 if (backing_file) {
5324 if (!strcmp(filename, backing_file)) {
5325 error_setg(errp, "Error: Trying to create an image with the "
5326 "same filename as the backing file");
5327 goto out;
5328 }
5329 }
5330
5331 backing_fmt = qemu_opt_get(opts, BLOCK_OPT_BACKING_FMT);
5332
5333 /* The size for the image must always be specified, unless we have a backing
5334 * file and we have not been forbidden from opening it. */
5335 size = qemu_opt_get_size(opts, BLOCK_OPT_SIZE, img_size);
5336 if (backing_file && !(flags & BDRV_O_NO_BACKING)) {
5337 BlockDriverState *bs;
5338 char *full_backing;
5339 int back_flags;
5340 QDict *backing_options = NULL;
5341
5342 full_backing =
5343 bdrv_get_full_backing_filename_from_filename(filename, backing_file,
5344 &local_err);
5345 if (local_err) {
5346 goto out;
5347 }
5348 assert(full_backing);
5349
5350 /* backing files always opened read-only */
5351 back_flags = flags;
5352 back_flags &= ~(BDRV_O_RDWR | BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING);
5353
5354 backing_options = qdict_new();
5355 if (backing_fmt) {
5356 qdict_put_str(backing_options, "driver", backing_fmt);
5357 }
5358 qdict_put_bool(backing_options, BDRV_OPT_FORCE_SHARE, true);
5359
5360 bs = bdrv_open(full_backing, NULL, backing_options, back_flags,
5361 &local_err);
5362 g_free(full_backing);
5363 if (!bs && size != -1) {
5364 /* Couldn't open BS, but we have a size, so it's nonfatal */
5365 warn_reportf_err(local_err,
5366 "Could not verify backing image. "
5367 "This may become an error in future versions.\n");
5368 local_err = NULL;
5369 } else if (!bs) {
5370 /* Couldn't open bs, do not have size */
5371 error_append_hint(&local_err,
5372 "Could not open backing image to determine size.\n");
5373 goto out;
5374 } else {
5375 if (size == -1) {
5376 /* Opened BS, have no size */
5377 size = bdrv_getlength(bs);
5378 if (size < 0) {
5379 error_setg_errno(errp, -size, "Could not get size of '%s'",
5380 backing_file);
5381 bdrv_unref(bs);
5382 goto out;
5383 }
5384 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, size, &error_abort);
5385 }
5386 bdrv_unref(bs);
5387 }
5388 } /* (backing_file && !(flags & BDRV_O_NO_BACKING)) */
5389
5390 if (size == -1) {
5391 error_setg(errp, "Image creation needs a size parameter");
5392 goto out;
5393 }
5394
5395 if (!quiet) {
5396 printf("Formatting '%s', fmt=%s ", filename, fmt);
5397 qemu_opts_print(opts, " ");
5398 puts("");
5399 }
5400
5401 ret = bdrv_create(drv, filename, opts, &local_err);
5402
5403 if (ret == -EFBIG) {
5404 /* This is generally a better message than whatever the driver would
5405 * deliver (especially because of the cluster_size_hint), since that
5406 * is most probably not much different from "image too large". */
5407 const char *cluster_size_hint = "";
5408 if (qemu_opt_get_size(opts, BLOCK_OPT_CLUSTER_SIZE, 0)) {
5409 cluster_size_hint = " (try using a larger cluster size)";
5410 }
5411 error_setg(errp, "The image size is too large for file format '%s'"
5412 "%s", fmt, cluster_size_hint);
5413 error_free(local_err);
5414 local_err = NULL;
5415 }
5416
5417 out:
5418 qemu_opts_del(opts);
5419 qemu_opts_free(create_opts);
5420 error_propagate(errp, local_err);
5421 }
5422
5423 AioContext *bdrv_get_aio_context(BlockDriverState *bs)
5424 {
5425 return bs ? bs->aio_context : qemu_get_aio_context();
5426 }
5427
5428 void bdrv_coroutine_enter(BlockDriverState *bs, Coroutine *co)
5429 {
5430 aio_co_enter(bdrv_get_aio_context(bs), co);
5431 }
5432
5433 static void bdrv_do_remove_aio_context_notifier(BdrvAioNotifier *ban)
5434 {
5435 QLIST_REMOVE(ban, list);
5436 g_free(ban);
5437 }
5438
5439 void bdrv_detach_aio_context(BlockDriverState *bs)
5440 {
5441 BdrvAioNotifier *baf, *baf_tmp;
5442 BdrvChild *child;
5443
5444 if (!bs->drv) {
5445 return;
5446 }
5447
5448 assert(!bs->walking_aio_notifiers);
5449 bs->walking_aio_notifiers = true;
5450 QLIST_FOREACH_SAFE(baf, &bs->aio_notifiers, list, baf_tmp) {
5451 if (baf->deleted) {
5452 bdrv_do_remove_aio_context_notifier(baf);
5453 } else {
5454 baf->detach_aio_context(baf->opaque);
5455 }
5456 }
5457 /* Never mind iterating again to check for ->deleted. bdrv_close() will
5458 * remove remaining aio notifiers if we aren't called again.
5459 */
5460 bs->walking_aio_notifiers = false;
5461
5462 if (bs->drv->bdrv_detach_aio_context) {
5463 bs->drv->bdrv_detach_aio_context(bs);
5464 }
5465 QLIST_FOREACH(child, &bs->children, next) {
5466 bdrv_detach_aio_context(child->bs);
5467 }
5468
5469 if (bs->quiesce_counter) {
5470 aio_enable_external(bs->aio_context);
5471 }
5472 bs->aio_context = NULL;
5473 }
5474
5475 void bdrv_attach_aio_context(BlockDriverState *bs,
5476 AioContext *new_context)
5477 {
5478 BdrvAioNotifier *ban, *ban_tmp;
5479 BdrvChild *child;
5480
5481 if (!bs->drv) {
5482 return;
5483 }
5484
5485 if (bs->quiesce_counter) {
5486 aio_disable_external(new_context);
5487 }
5488
5489 bs->aio_context = new_context;
5490
5491 QLIST_FOREACH(child, &bs->children, next) {
5492 bdrv_attach_aio_context(child->bs, new_context);
5493 }
5494 if (bs->drv->bdrv_attach_aio_context) {
5495 bs->drv->bdrv_attach_aio_context(bs, new_context);
5496 }
5497
5498 assert(!bs->walking_aio_notifiers);
5499 bs->walking_aio_notifiers = true;
5500 QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_tmp) {
5501 if (ban->deleted) {
5502 bdrv_do_remove_aio_context_notifier(ban);
5503 } else {
5504 ban->attached_aio_context(new_context, ban->opaque);
5505 }
5506 }
5507 bs->walking_aio_notifiers = false;
5508 }
5509
5510 /* The caller must own the AioContext lock for the old AioContext of bs, but it
5511 * must not own the AioContext lock for new_context (unless new_context is
5512 * the same as the current context of bs). */
5513 void bdrv_set_aio_context(BlockDriverState *bs, AioContext *new_context)
5514 {
5515 if (bdrv_get_aio_context(bs) == new_context) {
5516 return;
5517 }
5518
5519 bdrv_drained_begin(bs);
5520 bdrv_detach_aio_context(bs);
5521
5522 /* This function executes in the old AioContext so acquire the new one in
5523 * case it runs in a different thread.
5524 */
5525 aio_context_acquire(new_context);
5526 bdrv_attach_aio_context(bs, new_context);
5527 bdrv_drained_end(bs);
5528 aio_context_release(new_context);
5529 }
5530
5531 void bdrv_add_aio_context_notifier(BlockDriverState *bs,
5532 void (*attached_aio_context)(AioContext *new_context, void *opaque),
5533 void (*detach_aio_context)(void *opaque), void *opaque)
5534 {
5535 BdrvAioNotifier *ban = g_new(BdrvAioNotifier, 1);
5536 *ban = (BdrvAioNotifier){
5537 .attached_aio_context = attached_aio_context,
5538 .detach_aio_context = detach_aio_context,
5539 .opaque = opaque
5540 };
5541
5542 QLIST_INSERT_HEAD(&bs->aio_notifiers, ban, list);
5543 }
5544
5545 void bdrv_remove_aio_context_notifier(BlockDriverState *bs,
5546 void (*attached_aio_context)(AioContext *,
5547 void *),
5548 void (*detach_aio_context)(void *),
5549 void *opaque)
5550 {
5551 BdrvAioNotifier *ban, *ban_next;
5552
5553 QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_next) {
5554 if (ban->attached_aio_context == attached_aio_context &&
5555 ban->detach_aio_context == detach_aio_context &&
5556 ban->opaque == opaque &&
5557 ban->deleted == false)
5558 {
5559 if (bs->walking_aio_notifiers) {
5560 ban->deleted = true;
5561 } else {
5562 bdrv_do_remove_aio_context_notifier(ban);
5563 }
5564 return;
5565 }
5566 }
5567
5568 abort();
5569 }
5570
5571 int bdrv_amend_options(BlockDriverState *bs, QemuOpts *opts,
5572 BlockDriverAmendStatusCB *status_cb, void *cb_opaque,
5573 Error **errp)
5574 {
5575 if (!bs->drv) {
5576 error_setg(errp, "Node is ejected");
5577 return -ENOMEDIUM;
5578 }
5579 if (!bs->drv->bdrv_amend_options) {
5580 error_setg(errp, "Block driver '%s' does not support option amendment",
5581 bs->drv->format_name);
5582 return -ENOTSUP;
5583 }
5584 return bs->drv->bdrv_amend_options(bs, opts, status_cb, cb_opaque, errp);
5585 }
5586
5587 /* This function will be called by the bdrv_recurse_is_first_non_filter method
5588 * of block filter and by bdrv_is_first_non_filter.
5589 * It is used to test if the given bs is the candidate or recurse more in the
5590 * node graph.
5591 */
5592 bool bdrv_recurse_is_first_non_filter(BlockDriverState *bs,
5593 BlockDriverState *candidate)
5594 {
5595 /* return false if basic checks fails */
5596 if (!bs || !bs->drv) {
5597 return false;
5598 }
5599
5600 /* the code reached a non block filter driver -> check if the bs is
5601 * the same as the candidate. It's the recursion termination condition.
5602 */
5603 if (!bs->drv->is_filter) {
5604 return bs == candidate;
5605 }
5606 /* Down this path the driver is a block filter driver */
5607
5608 /* If the block filter recursion method is defined use it to recurse down
5609 * the node graph.
5610 */
5611 if (bs->drv->bdrv_recurse_is_first_non_filter) {
5612 return bs->drv->bdrv_recurse_is_first_non_filter(bs, candidate);
5613 }
5614
5615 /* the driver is a block filter but don't allow to recurse -> return false
5616 */
5617 return false;
5618 }
5619
5620 /* This function checks if the candidate is the first non filter bs down it's
5621 * bs chain. Since we don't have pointers to parents it explore all bs chains
5622 * from the top. Some filters can choose not to pass down the recursion.
5623 */
5624 bool bdrv_is_first_non_filter(BlockDriverState *candidate)
5625 {
5626 BlockDriverState *bs;
5627 BdrvNextIterator it;
5628
5629 /* walk down the bs forest recursively */
5630 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
5631 bool perm;
5632
5633 /* try to recurse in this top level bs */
5634 perm = bdrv_recurse_is_first_non_filter(bs, candidate);
5635
5636 /* candidate is the first non filter */
5637 if (perm) {
5638 bdrv_next_cleanup(&it);
5639 return true;
5640 }
5641 }
5642
5643 return false;
5644 }
5645
5646 BlockDriverState *check_to_replace_node(BlockDriverState *parent_bs,
5647 const char *node_name, Error **errp)
5648 {
5649 BlockDriverState *to_replace_bs = bdrv_find_node(node_name);
5650 AioContext *aio_context;
5651
5652 if (!to_replace_bs) {
5653 error_setg(errp, "Node name '%s' not found", node_name);
5654 return NULL;
5655 }
5656
5657 aio_context = bdrv_get_aio_context(to_replace_bs);
5658 aio_context_acquire(aio_context);
5659
5660 if (bdrv_op_is_blocked(to_replace_bs, BLOCK_OP_TYPE_REPLACE, errp)) {
5661 to_replace_bs = NULL;
5662 goto out;
5663 }
5664
5665 /* We don't want arbitrary node of the BDS chain to be replaced only the top
5666 * most non filter in order to prevent data corruption.
5667 * Another benefit is that this tests exclude backing files which are
5668 * blocked by the backing blockers.
5669 */
5670 if (!bdrv_recurse_is_first_non_filter(parent_bs, to_replace_bs)) {
5671 error_setg(errp, "Only top most non filter can be replaced");
5672 to_replace_bs = NULL;
5673 goto out;
5674 }
5675
5676 out:
5677 aio_context_release(aio_context);
5678 return to_replace_bs;
5679 }
5680
5681 /**
5682 * Iterates through the list of runtime option keys that are said to
5683 * be "strong" for a BDS. An option is called "strong" if it changes
5684 * a BDS's data. For example, the null block driver's "size" and
5685 * "read-zeroes" options are strong, but its "latency-ns" option is
5686 * not.
5687 *
5688 * If a key returned by this function ends with a dot, all options
5689 * starting with that prefix are strong.
5690 */
5691 static const char *const *strong_options(BlockDriverState *bs,
5692 const char *const *curopt)
5693 {
5694 static const char *const global_options[] = {
5695 "driver", "filename", NULL
5696 };
5697
5698 if (!curopt) {
5699 return &global_options[0];
5700 }
5701
5702 curopt++;
5703 if (curopt == &global_options[ARRAY_SIZE(global_options) - 1] && bs->drv) {
5704 curopt = bs->drv->strong_runtime_opts;
5705 }
5706
5707 return (curopt && *curopt) ? curopt : NULL;
5708 }
5709
5710 /**
5711 * Copies all strong runtime options from bs->options to the given
5712 * QDict. The set of strong option keys is determined by invoking
5713 * strong_options().
5714 *
5715 * Returns true iff any strong option was present in bs->options (and
5716 * thus copied to the target QDict) with the exception of "filename"
5717 * and "driver". The caller is expected to use this value to decide
5718 * whether the existence of strong options prevents the generation of
5719 * a plain filename.
5720 */
5721 static bool append_strong_runtime_options(QDict *d, BlockDriverState *bs)
5722 {
5723 bool found_any = false;
5724 const char *const *option_name = NULL;
5725
5726 if (!bs->drv) {
5727 return false;
5728 }
5729
5730 while ((option_name = strong_options(bs, option_name))) {
5731 bool option_given = false;
5732
5733 assert(strlen(*option_name) > 0);
5734 if ((*option_name)[strlen(*option_name) - 1] != '.') {
5735 QObject *entry = qdict_get(bs->options, *option_name);
5736 if (!entry) {
5737 continue;
5738 }
5739
5740 qdict_put_obj(d, *option_name, qobject_ref(entry));
5741 option_given = true;
5742 } else {
5743 const QDictEntry *entry;
5744 for (entry = qdict_first(bs->options); entry;
5745 entry = qdict_next(bs->options, entry))
5746 {
5747 if (strstart(qdict_entry_key(entry), *option_name, NULL)) {
5748 qdict_put_obj(d, qdict_entry_key(entry),
5749 qobject_ref(qdict_entry_value(entry)));
5750 option_given = true;
5751 }
5752 }
5753 }
5754
5755 /* While "driver" and "filename" need to be included in a JSON filename,
5756 * their existence does not prohibit generation of a plain filename. */
5757 if (!found_any && option_given &&
5758 strcmp(*option_name, "driver") && strcmp(*option_name, "filename"))
5759 {
5760 found_any = true;
5761 }
5762 }
5763
5764 if (!qdict_haskey(d, "driver")) {
5765 /* Drivers created with bdrv_new_open_driver() may not have a
5766 * @driver option. Add it here. */
5767 qdict_put_str(d, "driver", bs->drv->format_name);
5768 }
5769
5770 return found_any;
5771 }
5772
5773 /* Note: This function may return false positives; it may return true
5774 * even if opening the backing file specified by bs's image header
5775 * would result in exactly bs->backing. */
5776 static bool bdrv_backing_overridden(BlockDriverState *bs)
5777 {
5778 if (bs->backing) {
5779 return strcmp(bs->auto_backing_file,
5780 bs->backing->bs->filename);
5781 } else {
5782 /* No backing BDS, so if the image header reports any backing
5783 * file, it must have been suppressed */
5784 return bs->auto_backing_file[0] != '\0';
5785 }
5786 }
5787
5788 /* Updates the following BDS fields:
5789 * - exact_filename: A filename which may be used for opening a block device
5790 * which (mostly) equals the given BDS (even without any
5791 * other options; so reading and writing must return the same
5792 * results, but caching etc. may be different)
5793 * - full_open_options: Options which, when given when opening a block device
5794 * (without a filename), result in a BDS (mostly)
5795 * equalling the given one
5796 * - filename: If exact_filename is set, it is copied here. Otherwise,
5797 * full_open_options is converted to a JSON object, prefixed with
5798 * "json:" (for use through the JSON pseudo protocol) and put here.
5799 */
5800 void bdrv_refresh_filename(BlockDriverState *bs)
5801 {
5802 BlockDriver *drv = bs->drv;
5803 BdrvChild *child;
5804 QDict *opts;
5805 bool backing_overridden;
5806 bool generate_json_filename; /* Whether our default implementation should
5807 fill exact_filename (false) or not (true) */
5808
5809 if (!drv) {
5810 return;
5811 }
5812
5813 /* This BDS's file name may depend on any of its children's file names, so
5814 * refresh those first */
5815 QLIST_FOREACH(child, &bs->children, next) {
5816 bdrv_refresh_filename(child->bs);
5817 }
5818
5819 if (bs->implicit) {
5820 /* For implicit nodes, just copy everything from the single child */
5821 child = QLIST_FIRST(&bs->children);
5822 assert(QLIST_NEXT(child, next) == NULL);
5823
5824 pstrcpy(bs->exact_filename, sizeof(bs->exact_filename),
5825 child->bs->exact_filename);
5826 pstrcpy(bs->filename, sizeof(bs->filename), child->bs->filename);
5827
5828 bs->full_open_options = qobject_ref(child->bs->full_open_options);
5829
5830 return;
5831 }
5832
5833 backing_overridden = bdrv_backing_overridden(bs);
5834
5835 if (bs->open_flags & BDRV_O_NO_IO) {
5836 /* Without I/O, the backing file does not change anything.
5837 * Therefore, in such a case (primarily qemu-img), we can
5838 * pretend the backing file has not been overridden even if
5839 * it technically has been. */
5840 backing_overridden = false;
5841 }
5842
5843 /* Gather the options QDict */
5844 opts = qdict_new();
5845 generate_json_filename = append_strong_runtime_options(opts, bs);
5846 generate_json_filename |= backing_overridden;
5847
5848 if (drv->bdrv_gather_child_options) {
5849 /* Some block drivers may not want to present all of their children's
5850 * options, or name them differently from BdrvChild.name */
5851 drv->bdrv_gather_child_options(bs, opts, backing_overridden);
5852 } else {
5853 QLIST_FOREACH(child, &bs->children, next) {
5854 if (child->role == &child_backing && !backing_overridden) {
5855 /* We can skip the backing BDS if it has not been overridden */
5856 continue;
5857 }
5858
5859 qdict_put(opts, child->name,
5860 qobject_ref(child->bs->full_open_options));
5861 }
5862
5863 if (backing_overridden && !bs->backing) {
5864 /* Force no backing file */
5865 qdict_put_null(opts, "backing");
5866 }
5867 }
5868
5869 qobject_unref(bs->full_open_options);
5870 bs->full_open_options = opts;
5871
5872 if (drv->bdrv_refresh_filename) {
5873 /* Obsolete information is of no use here, so drop the old file name
5874 * information before refreshing it */
5875 bs->exact_filename[0] = '\0';
5876
5877 drv->bdrv_refresh_filename(bs);
5878 } else if (bs->file) {
5879 /* Try to reconstruct valid information from the underlying file */
5880
5881 bs->exact_filename[0] = '\0';
5882
5883 /*
5884 * We can use the underlying file's filename if:
5885 * - it has a filename,
5886 * - the file is a protocol BDS, and
5887 * - opening that file (as this BDS's format) will automatically create
5888 * the BDS tree we have right now, that is:
5889 * - the user did not significantly change this BDS's behavior with
5890 * some explicit (strong) options
5891 * - no non-file child of this BDS has been overridden by the user
5892 * Both of these conditions are represented by generate_json_filename.
5893 */
5894 if (bs->file->bs->exact_filename[0] &&
5895 bs->file->bs->drv->bdrv_file_open &&
5896 !generate_json_filename)
5897 {
5898 strcpy(bs->exact_filename, bs->file->bs->exact_filename);
5899 }
5900 }
5901
5902 if (bs->exact_filename[0]) {
5903 pstrcpy(bs->filename, sizeof(bs->filename), bs->exact_filename);
5904 } else {
5905 QString *json = qobject_to_json(QOBJECT(bs->full_open_options));
5906 snprintf(bs->filename, sizeof(bs->filename), "json:%s",
5907 qstring_get_str(json));
5908 qobject_unref(json);
5909 }
5910 }
5911
5912 char *bdrv_dirname(BlockDriverState *bs, Error **errp)
5913 {
5914 BlockDriver *drv = bs->drv;
5915
5916 if (!drv) {
5917 error_setg(errp, "Node '%s' is ejected", bs->node_name);
5918 return NULL;
5919 }
5920
5921 if (drv->bdrv_dirname) {
5922 return drv->bdrv_dirname(bs, errp);
5923 }
5924
5925 if (bs->file) {
5926 return bdrv_dirname(bs->file->bs, errp);
5927 }
5928
5929 bdrv_refresh_filename(bs);
5930 if (bs->exact_filename[0] != '\0') {
5931 return path_combine(bs->exact_filename, "");
5932 }
5933
5934 error_setg(errp, "Cannot generate a base directory for %s nodes",
5935 drv->format_name);
5936 return NULL;
5937 }
5938
5939 /*
5940 * Hot add/remove a BDS's child. So the user can take a child offline when
5941 * it is broken and take a new child online
5942 */
5943 void bdrv_add_child(BlockDriverState *parent_bs, BlockDriverState *child_bs,
5944 Error **errp)
5945 {
5946
5947 if (!parent_bs->drv || !parent_bs->drv->bdrv_add_child) {
5948 error_setg(errp, "The node %s does not support adding a child",
5949 bdrv_get_device_or_node_name(parent_bs));
5950 return;
5951 }
5952
5953 if (!QLIST_EMPTY(&child_bs->parents)) {
5954 error_setg(errp, "The node %s already has a parent",
5955 child_bs->node_name);
5956 return;
5957 }
5958
5959 parent_bs->drv->bdrv_add_child(parent_bs, child_bs, errp);
5960 }
5961
5962 void bdrv_del_child(BlockDriverState *parent_bs, BdrvChild *child, Error **errp)
5963 {
5964 BdrvChild *tmp;
5965
5966 if (!parent_bs->drv || !parent_bs->drv->bdrv_del_child) {
5967 error_setg(errp, "The node %s does not support removing a child",
5968 bdrv_get_device_or_node_name(parent_bs));
5969 return;
5970 }
5971
5972 QLIST_FOREACH(tmp, &parent_bs->children, next) {
5973 if (tmp == child) {
5974 break;
5975 }
5976 }
5977
5978 if (!tmp) {
5979 error_setg(errp, "The node %s does not have a child named %s",
5980 bdrv_get_device_or_node_name(parent_bs),
5981 bdrv_get_device_or_node_name(child->bs));
5982 return;
5983 }
5984
5985 parent_bs->drv->bdrv_del_child(parent_bs, child, errp);
5986 }
5987
5988 bool bdrv_can_store_new_dirty_bitmap(BlockDriverState *bs, const char *name,
5989 uint32_t granularity, Error **errp)
5990 {
5991 BlockDriver *drv = bs->drv;
5992
5993 if (!drv) {
5994 error_setg_errno(errp, ENOMEDIUM,
5995 "Can't store persistent bitmaps to %s",
5996 bdrv_get_device_or_node_name(bs));
5997 return false;
5998 }
5999
6000 if (!drv->bdrv_can_store_new_dirty_bitmap) {
6001 error_setg_errno(errp, ENOTSUP,
6002 "Can't store persistent bitmaps to %s",
6003 bdrv_get_device_or_node_name(bs));
6004 return false;
6005 }
6006
6007 return drv->bdrv_can_store_new_dirty_bitmap(bs, name, granularity, errp);
6008 }