xref: /openbmc/qemu/block.c (revision 5540cb97f711d191bae6cde89a03a8a9c306b638)
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 copy it to dest. Otherwise, build a
156    path to it by considering it is relative to base_path. URL are
157    supported. */
158 void path_combine(char *dest, int dest_size,
159                   const char *base_path,
160                   const char *filename)
161 {
162     const char *p, *p1;
163     int len;
164 
165     if (dest_size <= 0)
166         return;
167     if (path_is_absolute(filename)) {
168         pstrcpy(dest, dest_size, filename);
169     } else {
170         const char *protocol_stripped = NULL;
171 
172         if (path_has_protocol(base_path)) {
173             protocol_stripped = strchr(base_path, ':');
174             if (protocol_stripped) {
175                 protocol_stripped++;
176             }
177         }
178         p = protocol_stripped ?: base_path;
179 
180         p1 = strrchr(base_path, '/');
181 #ifdef _WIN32
182         {
183             const char *p2;
184             p2 = strrchr(base_path, '\\');
185             if (!p1 || p2 > p1)
186                 p1 = p2;
187         }
188 #endif
189         if (p1)
190             p1++;
191         else
192             p1 = base_path;
193         if (p1 > p)
194             p = p1;
195         len = p - base_path;
196         if (len > dest_size - 1)
197             len = dest_size - 1;
198         memcpy(dest, base_path, len);
199         dest[len] = '\0';
200         pstrcat(dest, dest_size, filename);
201     }
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 /* TODO Remove (deprecated since 2.11)
270  * Block drivers are not supposed to automatically change bs->read_only.
271  * Instead, they should just check whether they can provide what the user
272  * explicitly requested and error out if read-write is requested, but they can
273  * only provide read-only access. */
274 int bdrv_set_read_only(BlockDriverState *bs, bool read_only, Error **errp)
275 {
276     int ret = 0;
277 
278     ret = bdrv_can_set_read_only(bs, read_only, false, errp);
279     if (ret < 0) {
280         return ret;
281     }
282 
283     bs->read_only = read_only;
284     return 0;
285 }
286 
287 void bdrv_get_full_backing_filename_from_filename(const char *backed,
288                                                   const char *backing,
289                                                   char *dest, size_t sz,
290                                                   Error **errp)
291 {
292     if (backing[0] == '\0' || path_has_protocol(backing) ||
293         path_is_absolute(backing))
294     {
295         pstrcpy(dest, sz, backing);
296     } else if (backed[0] == '\0' || strstart(backed, "json:", NULL)) {
297         error_setg(errp, "Cannot use relative backing file names for '%s'",
298                    backed);
299     } else {
300         path_combine(dest, sz, backed, backing);
301     }
302 }
303 
304 void bdrv_get_full_backing_filename(BlockDriverState *bs, char *dest, size_t sz,
305                                     Error **errp)
306 {
307     char *backed = bs->exact_filename[0] ? bs->exact_filename : bs->filename;
308 
309     bdrv_get_full_backing_filename_from_filename(backed, bs->backing_file,
310                                                  dest, sz, errp);
311 }
312 
313 void bdrv_register(BlockDriver *bdrv)
314 {
315     QLIST_INSERT_HEAD(&bdrv_drivers, bdrv, list);
316 }
317 
318 BlockDriverState *bdrv_new(void)
319 {
320     BlockDriverState *bs;
321     int i;
322 
323     bs = g_new0(BlockDriverState, 1);
324     QLIST_INIT(&bs->dirty_bitmaps);
325     for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
326         QLIST_INIT(&bs->op_blockers[i]);
327     }
328     notifier_with_return_list_init(&bs->before_write_notifiers);
329     qemu_co_mutex_init(&bs->reqs_lock);
330     qemu_mutex_init(&bs->dirty_bitmap_mutex);
331     bs->refcnt = 1;
332     bs->aio_context = qemu_get_aio_context();
333 
334     qemu_co_queue_init(&bs->flush_queue);
335 
336     for (i = 0; i < bdrv_drain_all_count; i++) {
337         bdrv_drained_begin(bs);
338     }
339 
340     QTAILQ_INSERT_TAIL(&all_bdrv_states, bs, bs_list);
341 
342     return bs;
343 }
344 
345 static BlockDriver *bdrv_do_find_format(const char *format_name)
346 {
347     BlockDriver *drv1;
348 
349     QLIST_FOREACH(drv1, &bdrv_drivers, list) {
350         if (!strcmp(drv1->format_name, format_name)) {
351             return drv1;
352         }
353     }
354 
355     return NULL;
356 }
357 
358 BlockDriver *bdrv_find_format(const char *format_name)
359 {
360     BlockDriver *drv1;
361     int i;
362 
363     drv1 = bdrv_do_find_format(format_name);
364     if (drv1) {
365         return drv1;
366     }
367 
368     /* The driver isn't registered, maybe we need to load a module */
369     for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); ++i) {
370         if (!strcmp(block_driver_modules[i].format_name, format_name)) {
371             block_module_load_one(block_driver_modules[i].library_name);
372             break;
373         }
374     }
375 
376     return bdrv_do_find_format(format_name);
377 }
378 
379 int bdrv_is_whitelisted(BlockDriver *drv, bool read_only)
380 {
381     static const char *whitelist_rw[] = {
382         CONFIG_BDRV_RW_WHITELIST
383     };
384     static const char *whitelist_ro[] = {
385         CONFIG_BDRV_RO_WHITELIST
386     };
387     const char **p;
388 
389     if (!whitelist_rw[0] && !whitelist_ro[0]) {
390         return 1;               /* no whitelist, anything goes */
391     }
392 
393     for (p = whitelist_rw; *p; p++) {
394         if (!strcmp(drv->format_name, *p)) {
395             return 1;
396         }
397     }
398     if (read_only) {
399         for (p = whitelist_ro; *p; p++) {
400             if (!strcmp(drv->format_name, *p)) {
401                 return 1;
402             }
403         }
404     }
405     return 0;
406 }
407 
408 bool bdrv_uses_whitelist(void)
409 {
410     return use_bdrv_whitelist;
411 }
412 
413 typedef struct CreateCo {
414     BlockDriver *drv;
415     char *filename;
416     QemuOpts *opts;
417     int ret;
418     Error *err;
419 } CreateCo;
420 
421 static void coroutine_fn bdrv_create_co_entry(void *opaque)
422 {
423     Error *local_err = NULL;
424     int ret;
425 
426     CreateCo *cco = opaque;
427     assert(cco->drv);
428 
429     ret = cco->drv->bdrv_co_create_opts(cco->filename, cco->opts, &local_err);
430     error_propagate(&cco->err, local_err);
431     cco->ret = ret;
432 }
433 
434 int bdrv_create(BlockDriver *drv, const char* filename,
435                 QemuOpts *opts, Error **errp)
436 {
437     int ret;
438 
439     Coroutine *co;
440     CreateCo cco = {
441         .drv = drv,
442         .filename = g_strdup(filename),
443         .opts = opts,
444         .ret = NOT_DONE,
445         .err = NULL,
446     };
447 
448     if (!drv->bdrv_co_create_opts) {
449         error_setg(errp, "Driver '%s' does not support image creation", drv->format_name);
450         ret = -ENOTSUP;
451         goto out;
452     }
453 
454     if (qemu_in_coroutine()) {
455         /* Fast-path if already in coroutine context */
456         bdrv_create_co_entry(&cco);
457     } else {
458         co = qemu_coroutine_create(bdrv_create_co_entry, &cco);
459         qemu_coroutine_enter(co);
460         while (cco.ret == NOT_DONE) {
461             aio_poll(qemu_get_aio_context(), true);
462         }
463     }
464 
465     ret = cco.ret;
466     if (ret < 0) {
467         if (cco.err) {
468             error_propagate(errp, cco.err);
469         } else {
470             error_setg_errno(errp, -ret, "Could not create image");
471         }
472     }
473 
474 out:
475     g_free(cco.filename);
476     return ret;
477 }
478 
479 int bdrv_create_file(const char *filename, QemuOpts *opts, Error **errp)
480 {
481     BlockDriver *drv;
482     Error *local_err = NULL;
483     int ret;
484 
485     drv = bdrv_find_protocol(filename, true, errp);
486     if (drv == NULL) {
487         return -ENOENT;
488     }
489 
490     ret = bdrv_create(drv, filename, opts, &local_err);
491     error_propagate(errp, local_err);
492     return ret;
493 }
494 
495 /**
496  * Try to get @bs's logical and physical block size.
497  * On success, store them in @bsz struct and return 0.
498  * On failure return -errno.
499  * @bs must not be empty.
500  */
501 int bdrv_probe_blocksizes(BlockDriverState *bs, BlockSizes *bsz)
502 {
503     BlockDriver *drv = bs->drv;
504 
505     if (drv && drv->bdrv_probe_blocksizes) {
506         return drv->bdrv_probe_blocksizes(bs, bsz);
507     } else if (drv && drv->is_filter && bs->file) {
508         return bdrv_probe_blocksizes(bs->file->bs, bsz);
509     }
510 
511     return -ENOTSUP;
512 }
513 
514 /**
515  * Try to get @bs's geometry (cyls, heads, sectors).
516  * On success, store them in @geo struct and return 0.
517  * On failure return -errno.
518  * @bs must not be empty.
519  */
520 int bdrv_probe_geometry(BlockDriverState *bs, HDGeometry *geo)
521 {
522     BlockDriver *drv = bs->drv;
523 
524     if (drv && drv->bdrv_probe_geometry) {
525         return drv->bdrv_probe_geometry(bs, geo);
526     } else if (drv && drv->is_filter && bs->file) {
527         return bdrv_probe_geometry(bs->file->bs, geo);
528     }
529 
530     return -ENOTSUP;
531 }
532 
533 /*
534  * Create a uniquely-named empty temporary file.
535  * Return 0 upon success, otherwise a negative errno value.
536  */
537 int get_tmp_filename(char *filename, int size)
538 {
539 #ifdef _WIN32
540     char temp_dir[MAX_PATH];
541     /* GetTempFileName requires that its output buffer (4th param)
542        have length MAX_PATH or greater.  */
543     assert(size >= MAX_PATH);
544     return (GetTempPath(MAX_PATH, temp_dir)
545             && GetTempFileName(temp_dir, "qem", 0, filename)
546             ? 0 : -GetLastError());
547 #else
548     int fd;
549     const char *tmpdir;
550     tmpdir = getenv("TMPDIR");
551     if (!tmpdir) {
552         tmpdir = "/var/tmp";
553     }
554     if (snprintf(filename, size, "%s/vl.XXXXXX", tmpdir) >= size) {
555         return -EOVERFLOW;
556     }
557     fd = mkstemp(filename);
558     if (fd < 0) {
559         return -errno;
560     }
561     if (close(fd) != 0) {
562         unlink(filename);
563         return -errno;
564     }
565     return 0;
566 #endif
567 }
568 
569 /*
570  * Detect host devices. By convention, /dev/cdrom[N] is always
571  * recognized as a host CDROM.
572  */
573 static BlockDriver *find_hdev_driver(const char *filename)
574 {
575     int score_max = 0, score;
576     BlockDriver *drv = NULL, *d;
577 
578     QLIST_FOREACH(d, &bdrv_drivers, list) {
579         if (d->bdrv_probe_device) {
580             score = d->bdrv_probe_device(filename);
581             if (score > score_max) {
582                 score_max = score;
583                 drv = d;
584             }
585         }
586     }
587 
588     return drv;
589 }
590 
591 static BlockDriver *bdrv_do_find_protocol(const char *protocol)
592 {
593     BlockDriver *drv1;
594 
595     QLIST_FOREACH(drv1, &bdrv_drivers, list) {
596         if (drv1->protocol_name && !strcmp(drv1->protocol_name, protocol)) {
597             return drv1;
598         }
599     }
600 
601     return NULL;
602 }
603 
604 BlockDriver *bdrv_find_protocol(const char *filename,
605                                 bool allow_protocol_prefix,
606                                 Error **errp)
607 {
608     BlockDriver *drv1;
609     char protocol[128];
610     int len;
611     const char *p;
612     int i;
613 
614     /* TODO Drivers without bdrv_file_open must be specified explicitly */
615 
616     /*
617      * XXX(hch): we really should not let host device detection
618      * override an explicit protocol specification, but moving this
619      * later breaks access to device names with colons in them.
620      * Thanks to the brain-dead persistent naming schemes on udev-
621      * based Linux systems those actually are quite common.
622      */
623     drv1 = find_hdev_driver(filename);
624     if (drv1) {
625         return drv1;
626     }
627 
628     if (!path_has_protocol(filename) || !allow_protocol_prefix) {
629         return &bdrv_file;
630     }
631 
632     p = strchr(filename, ':');
633     assert(p != NULL);
634     len = p - filename;
635     if (len > sizeof(protocol) - 1)
636         len = sizeof(protocol) - 1;
637     memcpy(protocol, filename, len);
638     protocol[len] = '\0';
639 
640     drv1 = bdrv_do_find_protocol(protocol);
641     if (drv1) {
642         return drv1;
643     }
644 
645     for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); ++i) {
646         if (block_driver_modules[i].protocol_name &&
647             !strcmp(block_driver_modules[i].protocol_name, protocol)) {
648             block_module_load_one(block_driver_modules[i].library_name);
649             break;
650         }
651     }
652 
653     drv1 = bdrv_do_find_protocol(protocol);
654     if (!drv1) {
655         error_setg(errp, "Unknown protocol '%s'", protocol);
656     }
657     return drv1;
658 }
659 
660 /*
661  * Guess image format by probing its contents.
662  * This is not a good idea when your image is raw (CVE-2008-2004), but
663  * we do it anyway for backward compatibility.
664  *
665  * @buf         contains the image's first @buf_size bytes.
666  * @buf_size    is the buffer size in bytes (generally BLOCK_PROBE_BUF_SIZE,
667  *              but can be smaller if the image file is smaller)
668  * @filename    is its filename.
669  *
670  * For all block drivers, call the bdrv_probe() method to get its
671  * probing score.
672  * Return the first block driver with the highest probing score.
673  */
674 BlockDriver *bdrv_probe_all(const uint8_t *buf, int buf_size,
675                             const char *filename)
676 {
677     int score_max = 0, score;
678     BlockDriver *drv = NULL, *d;
679 
680     QLIST_FOREACH(d, &bdrv_drivers, list) {
681         if (d->bdrv_probe) {
682             score = d->bdrv_probe(buf, buf_size, filename);
683             if (score > score_max) {
684                 score_max = score;
685                 drv = d;
686             }
687         }
688     }
689 
690     return drv;
691 }
692 
693 static int find_image_format(BlockBackend *file, const char *filename,
694                              BlockDriver **pdrv, Error **errp)
695 {
696     BlockDriver *drv;
697     uint8_t buf[BLOCK_PROBE_BUF_SIZE];
698     int ret = 0;
699 
700     /* Return the raw BlockDriver * to scsi-generic devices or empty drives */
701     if (blk_is_sg(file) || !blk_is_inserted(file) || blk_getlength(file) == 0) {
702         *pdrv = &bdrv_raw;
703         return ret;
704     }
705 
706     ret = blk_pread(file, 0, buf, sizeof(buf));
707     if (ret < 0) {
708         error_setg_errno(errp, -ret, "Could not read image for determining its "
709                          "format");
710         *pdrv = NULL;
711         return ret;
712     }
713 
714     drv = bdrv_probe_all(buf, ret, filename);
715     if (!drv) {
716         error_setg(errp, "Could not determine image format: No compatible "
717                    "driver found");
718         ret = -ENOENT;
719     }
720     *pdrv = drv;
721     return ret;
722 }
723 
724 /**
725  * Set the current 'total_sectors' value
726  * Return 0 on success, -errno on error.
727  */
728 int refresh_total_sectors(BlockDriverState *bs, int64_t hint)
729 {
730     BlockDriver *drv = bs->drv;
731 
732     if (!drv) {
733         return -ENOMEDIUM;
734     }
735 
736     /* Do not attempt drv->bdrv_getlength() on scsi-generic devices */
737     if (bdrv_is_sg(bs))
738         return 0;
739 
740     /* query actual device if possible, otherwise just trust the hint */
741     if (drv->bdrv_getlength) {
742         int64_t length = drv->bdrv_getlength(bs);
743         if (length < 0) {
744             return length;
745         }
746         hint = DIV_ROUND_UP(length, BDRV_SECTOR_SIZE);
747     }
748 
749     bs->total_sectors = hint;
750     return 0;
751 }
752 
753 /**
754  * Combines a QDict of new block driver @options with any missing options taken
755  * from @old_options, so that leaving out an option defaults to its old value.
756  */
757 static void bdrv_join_options(BlockDriverState *bs, QDict *options,
758                               QDict *old_options)
759 {
760     if (bs->drv && bs->drv->bdrv_join_options) {
761         bs->drv->bdrv_join_options(options, old_options);
762     } else {
763         qdict_join(options, old_options, false);
764     }
765 }
766 
767 /**
768  * Set open flags for a given discard mode
769  *
770  * Return 0 on success, -1 if the discard mode was invalid.
771  */
772 int bdrv_parse_discard_flags(const char *mode, int *flags)
773 {
774     *flags &= ~BDRV_O_UNMAP;
775 
776     if (!strcmp(mode, "off") || !strcmp(mode, "ignore")) {
777         /* do nothing */
778     } else if (!strcmp(mode, "on") || !strcmp(mode, "unmap")) {
779         *flags |= BDRV_O_UNMAP;
780     } else {
781         return -1;
782     }
783 
784     return 0;
785 }
786 
787 /**
788  * Set open flags for a given cache mode
789  *
790  * Return 0 on success, -1 if the cache mode was invalid.
791  */
792 int bdrv_parse_cache_mode(const char *mode, int *flags, bool *writethrough)
793 {
794     *flags &= ~BDRV_O_CACHE_MASK;
795 
796     if (!strcmp(mode, "off") || !strcmp(mode, "none")) {
797         *writethrough = false;
798         *flags |= BDRV_O_NOCACHE;
799     } else if (!strcmp(mode, "directsync")) {
800         *writethrough = true;
801         *flags |= BDRV_O_NOCACHE;
802     } else if (!strcmp(mode, "writeback")) {
803         *writethrough = false;
804     } else if (!strcmp(mode, "unsafe")) {
805         *writethrough = false;
806         *flags |= BDRV_O_NO_FLUSH;
807     } else if (!strcmp(mode, "writethrough")) {
808         *writethrough = true;
809     } else {
810         return -1;
811     }
812 
813     return 0;
814 }
815 
816 static char *bdrv_child_get_parent_desc(BdrvChild *c)
817 {
818     BlockDriverState *parent = c->opaque;
819     return g_strdup(bdrv_get_device_or_node_name(parent));
820 }
821 
822 static void bdrv_child_cb_drained_begin(BdrvChild *child)
823 {
824     BlockDriverState *bs = child->opaque;
825     bdrv_do_drained_begin_quiesce(bs, NULL, false);
826 }
827 
828 static bool bdrv_child_cb_drained_poll(BdrvChild *child)
829 {
830     BlockDriverState *bs = child->opaque;
831     return bdrv_drain_poll(bs, false, NULL, false);
832 }
833 
834 static void bdrv_child_cb_drained_end(BdrvChild *child)
835 {
836     BlockDriverState *bs = child->opaque;
837     bdrv_drained_end(bs);
838 }
839 
840 static void bdrv_child_cb_attach(BdrvChild *child)
841 {
842     BlockDriverState *bs = child->opaque;
843     bdrv_apply_subtree_drain(child, bs);
844 }
845 
846 static void bdrv_child_cb_detach(BdrvChild *child)
847 {
848     BlockDriverState *bs = child->opaque;
849     bdrv_unapply_subtree_drain(child, bs);
850 }
851 
852 static int bdrv_child_cb_inactivate(BdrvChild *child)
853 {
854     BlockDriverState *bs = child->opaque;
855     assert(bs->open_flags & BDRV_O_INACTIVE);
856     return 0;
857 }
858 
859 /*
860  * Returns the options and flags that a temporary snapshot should get, based on
861  * the originally requested flags (the originally requested image will have
862  * flags like a backing file)
863  */
864 static void bdrv_temp_snapshot_options(int *child_flags, QDict *child_options,
865                                        int parent_flags, QDict *parent_options)
866 {
867     *child_flags = (parent_flags & ~BDRV_O_SNAPSHOT) | BDRV_O_TEMPORARY;
868 
869     /* For temporary files, unconditional cache=unsafe is fine */
870     qdict_set_default_str(child_options, BDRV_OPT_CACHE_DIRECT, "off");
871     qdict_set_default_str(child_options, BDRV_OPT_CACHE_NO_FLUSH, "on");
872 
873     /* Copy the read-only option from the parent */
874     qdict_copy_default(child_options, parent_options, BDRV_OPT_READ_ONLY);
875 
876     /* aio=native doesn't work for cache.direct=off, so disable it for the
877      * temporary snapshot */
878     *child_flags &= ~BDRV_O_NATIVE_AIO;
879 }
880 
881 /*
882  * Returns the options and flags that bs->file should get if a protocol driver
883  * is expected, based on the given options and flags for the parent BDS
884  */
885 static void bdrv_inherited_options(int *child_flags, QDict *child_options,
886                                    int parent_flags, QDict *parent_options)
887 {
888     int flags = parent_flags;
889 
890     /* Enable protocol handling, disable format probing for bs->file */
891     flags |= BDRV_O_PROTOCOL;
892 
893     /* If the cache mode isn't explicitly set, inherit direct and no-flush from
894      * the parent. */
895     qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_DIRECT);
896     qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_NO_FLUSH);
897     qdict_copy_default(child_options, parent_options, BDRV_OPT_FORCE_SHARE);
898 
899     /* Inherit the read-only option from the parent if it's not set */
900     qdict_copy_default(child_options, parent_options, BDRV_OPT_READ_ONLY);
901 
902     /* Our block drivers take care to send flushes and respect unmap policy,
903      * so we can default to enable both on lower layers regardless of the
904      * corresponding parent options. */
905     qdict_set_default_str(child_options, BDRV_OPT_DISCARD, "unmap");
906 
907     /* Clear flags that only apply to the top layer */
908     flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_COPY_ON_READ |
909                BDRV_O_NO_IO);
910 
911     *child_flags = flags;
912 }
913 
914 const BdrvChildRole child_file = {
915     .parent_is_bds   = true,
916     .get_parent_desc = bdrv_child_get_parent_desc,
917     .inherit_options = bdrv_inherited_options,
918     .drained_begin   = bdrv_child_cb_drained_begin,
919     .drained_poll    = bdrv_child_cb_drained_poll,
920     .drained_end     = bdrv_child_cb_drained_end,
921     .attach          = bdrv_child_cb_attach,
922     .detach          = bdrv_child_cb_detach,
923     .inactivate      = bdrv_child_cb_inactivate,
924 };
925 
926 /*
927  * Returns the options and flags that bs->file should get if the use of formats
928  * (and not only protocols) is permitted for it, based on the given options and
929  * flags for the parent BDS
930  */
931 static void bdrv_inherited_fmt_options(int *child_flags, QDict *child_options,
932                                        int parent_flags, QDict *parent_options)
933 {
934     child_file.inherit_options(child_flags, child_options,
935                                parent_flags, parent_options);
936 
937     *child_flags &= ~(BDRV_O_PROTOCOL | BDRV_O_NO_IO);
938 }
939 
940 const BdrvChildRole child_format = {
941     .parent_is_bds   = true,
942     .get_parent_desc = bdrv_child_get_parent_desc,
943     .inherit_options = bdrv_inherited_fmt_options,
944     .drained_begin   = bdrv_child_cb_drained_begin,
945     .drained_poll    = bdrv_child_cb_drained_poll,
946     .drained_end     = bdrv_child_cb_drained_end,
947     .attach          = bdrv_child_cb_attach,
948     .detach          = bdrv_child_cb_detach,
949     .inactivate      = bdrv_child_cb_inactivate,
950 };
951 
952 static void bdrv_backing_attach(BdrvChild *c)
953 {
954     BlockDriverState *parent = c->opaque;
955     BlockDriverState *backing_hd = c->bs;
956 
957     assert(!parent->backing_blocker);
958     error_setg(&parent->backing_blocker,
959                "node is used as backing hd of '%s'",
960                bdrv_get_device_or_node_name(parent));
961 
962     parent->open_flags &= ~BDRV_O_NO_BACKING;
963     pstrcpy(parent->backing_file, sizeof(parent->backing_file),
964             backing_hd->filename);
965     pstrcpy(parent->backing_format, sizeof(parent->backing_format),
966             backing_hd->drv ? backing_hd->drv->format_name : "");
967 
968     bdrv_op_block_all(backing_hd, parent->backing_blocker);
969     /* Otherwise we won't be able to commit or stream */
970     bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_COMMIT_TARGET,
971                     parent->backing_blocker);
972     bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_STREAM,
973                     parent->backing_blocker);
974     /*
975      * We do backup in 3 ways:
976      * 1. drive backup
977      *    The target bs is new opened, and the source is top BDS
978      * 2. blockdev backup
979      *    Both the source and the target are top BDSes.
980      * 3. internal backup(used for block replication)
981      *    Both the source and the target are backing file
982      *
983      * In case 1 and 2, neither the source nor the target is the backing file.
984      * In case 3, we will block the top BDS, so there is only one block job
985      * for the top BDS and its backing chain.
986      */
987     bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_BACKUP_SOURCE,
988                     parent->backing_blocker);
989     bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_BACKUP_TARGET,
990                     parent->backing_blocker);
991 
992     bdrv_child_cb_attach(c);
993 }
994 
995 static void bdrv_backing_detach(BdrvChild *c)
996 {
997     BlockDriverState *parent = c->opaque;
998 
999     assert(parent->backing_blocker);
1000     bdrv_op_unblock_all(c->bs, parent->backing_blocker);
1001     error_free(parent->backing_blocker);
1002     parent->backing_blocker = NULL;
1003 
1004     bdrv_child_cb_detach(c);
1005 }
1006 
1007 /*
1008  * Returns the options and flags that bs->backing should get, based on the
1009  * given options and flags for the parent BDS
1010  */
1011 static void bdrv_backing_options(int *child_flags, QDict *child_options,
1012                                  int parent_flags, QDict *parent_options)
1013 {
1014     int flags = parent_flags;
1015 
1016     /* The cache mode is inherited unmodified for backing files; except WCE,
1017      * which is only applied on the top level (BlockBackend) */
1018     qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_DIRECT);
1019     qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_NO_FLUSH);
1020     qdict_copy_default(child_options, parent_options, BDRV_OPT_FORCE_SHARE);
1021 
1022     /* backing files always opened read-only */
1023     qdict_set_default_str(child_options, BDRV_OPT_READ_ONLY, "on");
1024     flags &= ~BDRV_O_COPY_ON_READ;
1025 
1026     /* snapshot=on is handled on the top layer */
1027     flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_TEMPORARY);
1028 
1029     *child_flags = flags;
1030 }
1031 
1032 static int bdrv_backing_update_filename(BdrvChild *c, BlockDriverState *base,
1033                                         const char *filename, Error **errp)
1034 {
1035     BlockDriverState *parent = c->opaque;
1036     int orig_flags = bdrv_get_flags(parent);
1037     int ret;
1038 
1039     if (!(orig_flags & BDRV_O_RDWR)) {
1040         ret = bdrv_reopen(parent, orig_flags | BDRV_O_RDWR, errp);
1041         if (ret < 0) {
1042             return ret;
1043         }
1044     }
1045 
1046     ret = bdrv_change_backing_file(parent, filename,
1047                                    base->drv ? base->drv->format_name : "");
1048     if (ret < 0) {
1049         error_setg_errno(errp, -ret, "Could not update backing file link");
1050     }
1051 
1052     if (!(orig_flags & BDRV_O_RDWR)) {
1053         bdrv_reopen(parent, orig_flags, NULL);
1054     }
1055 
1056     return ret;
1057 }
1058 
1059 const BdrvChildRole child_backing = {
1060     .parent_is_bds   = true,
1061     .get_parent_desc = bdrv_child_get_parent_desc,
1062     .attach          = bdrv_backing_attach,
1063     .detach          = bdrv_backing_detach,
1064     .inherit_options = bdrv_backing_options,
1065     .drained_begin   = bdrv_child_cb_drained_begin,
1066     .drained_poll    = bdrv_child_cb_drained_poll,
1067     .drained_end     = bdrv_child_cb_drained_end,
1068     .inactivate      = bdrv_child_cb_inactivate,
1069     .update_filename = bdrv_backing_update_filename,
1070 };
1071 
1072 static int bdrv_open_flags(BlockDriverState *bs, int flags)
1073 {
1074     int open_flags = flags;
1075 
1076     /*
1077      * Clear flags that are internal to the block layer before opening the
1078      * image.
1079      */
1080     open_flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_PROTOCOL);
1081 
1082     /*
1083      * Snapshots should be writable.
1084      */
1085     if (flags & BDRV_O_TEMPORARY) {
1086         open_flags |= BDRV_O_RDWR;
1087     }
1088 
1089     return open_flags;
1090 }
1091 
1092 static void update_flags_from_options(int *flags, QemuOpts *opts)
1093 {
1094     *flags &= ~BDRV_O_CACHE_MASK;
1095 
1096     assert(qemu_opt_find(opts, BDRV_OPT_CACHE_NO_FLUSH));
1097     if (qemu_opt_get_bool(opts, BDRV_OPT_CACHE_NO_FLUSH, false)) {
1098         *flags |= BDRV_O_NO_FLUSH;
1099     }
1100 
1101     assert(qemu_opt_find(opts, BDRV_OPT_CACHE_DIRECT));
1102     if (qemu_opt_get_bool(opts, BDRV_OPT_CACHE_DIRECT, false)) {
1103         *flags |= BDRV_O_NOCACHE;
1104     }
1105 
1106     *flags &= ~BDRV_O_RDWR;
1107 
1108     assert(qemu_opt_find(opts, BDRV_OPT_READ_ONLY));
1109     if (!qemu_opt_get_bool(opts, BDRV_OPT_READ_ONLY, false)) {
1110         *flags |= BDRV_O_RDWR;
1111     }
1112 
1113 }
1114 
1115 static void update_options_from_flags(QDict *options, int flags)
1116 {
1117     if (!qdict_haskey(options, BDRV_OPT_CACHE_DIRECT)) {
1118         qdict_put_bool(options, BDRV_OPT_CACHE_DIRECT, flags & BDRV_O_NOCACHE);
1119     }
1120     if (!qdict_haskey(options, BDRV_OPT_CACHE_NO_FLUSH)) {
1121         qdict_put_bool(options, BDRV_OPT_CACHE_NO_FLUSH,
1122                        flags & BDRV_O_NO_FLUSH);
1123     }
1124     if (!qdict_haskey(options, BDRV_OPT_READ_ONLY)) {
1125         qdict_put_bool(options, BDRV_OPT_READ_ONLY, !(flags & BDRV_O_RDWR));
1126     }
1127 }
1128 
1129 static void bdrv_assign_node_name(BlockDriverState *bs,
1130                                   const char *node_name,
1131                                   Error **errp)
1132 {
1133     char *gen_node_name = NULL;
1134 
1135     if (!node_name) {
1136         node_name = gen_node_name = id_generate(ID_BLOCK);
1137     } else if (!id_wellformed(node_name)) {
1138         /*
1139          * Check for empty string or invalid characters, but not if it is
1140          * generated (generated names use characters not available to the user)
1141          */
1142         error_setg(errp, "Invalid node name");
1143         return;
1144     }
1145 
1146     /* takes care of avoiding namespaces collisions */
1147     if (blk_by_name(node_name)) {
1148         error_setg(errp, "node-name=%s is conflicting with a device id",
1149                    node_name);
1150         goto out;
1151     }
1152 
1153     /* takes care of avoiding duplicates node names */
1154     if (bdrv_find_node(node_name)) {
1155         error_setg(errp, "Duplicate node name");
1156         goto out;
1157     }
1158 
1159     /* Make sure that the node name isn't truncated */
1160     if (strlen(node_name) >= sizeof(bs->node_name)) {
1161         error_setg(errp, "Node name too long");
1162         goto out;
1163     }
1164 
1165     /* copy node name into the bs and insert it into the graph list */
1166     pstrcpy(bs->node_name, sizeof(bs->node_name), node_name);
1167     QTAILQ_INSERT_TAIL(&graph_bdrv_states, bs, node_list);
1168 out:
1169     g_free(gen_node_name);
1170 }
1171 
1172 static int bdrv_open_driver(BlockDriverState *bs, BlockDriver *drv,
1173                             const char *node_name, QDict *options,
1174                             int open_flags, Error **errp)
1175 {
1176     Error *local_err = NULL;
1177     int i, ret;
1178 
1179     bdrv_assign_node_name(bs, node_name, &local_err);
1180     if (local_err) {
1181         error_propagate(errp, local_err);
1182         return -EINVAL;
1183     }
1184 
1185     bs->drv = drv;
1186     bs->read_only = !(bs->open_flags & BDRV_O_RDWR);
1187     bs->opaque = g_malloc0(drv->instance_size);
1188 
1189     if (drv->bdrv_file_open) {
1190         assert(!drv->bdrv_needs_filename || bs->filename[0]);
1191         ret = drv->bdrv_file_open(bs, options, open_flags, &local_err);
1192     } else if (drv->bdrv_open) {
1193         ret = drv->bdrv_open(bs, options, open_flags, &local_err);
1194     } else {
1195         ret = 0;
1196     }
1197 
1198     if (ret < 0) {
1199         if (local_err) {
1200             error_propagate(errp, local_err);
1201         } else if (bs->filename[0]) {
1202             error_setg_errno(errp, -ret, "Could not open '%s'", bs->filename);
1203         } else {
1204             error_setg_errno(errp, -ret, "Could not open image");
1205         }
1206         goto open_failed;
1207     }
1208 
1209     ret = refresh_total_sectors(bs, bs->total_sectors);
1210     if (ret < 0) {
1211         error_setg_errno(errp, -ret, "Could not refresh total sector count");
1212         return ret;
1213     }
1214 
1215     bdrv_refresh_limits(bs, &local_err);
1216     if (local_err) {
1217         error_propagate(errp, local_err);
1218         return -EINVAL;
1219     }
1220 
1221     assert(bdrv_opt_mem_align(bs) != 0);
1222     assert(bdrv_min_mem_align(bs) != 0);
1223     assert(is_power_of_2(bs->bl.request_alignment));
1224 
1225     for (i = 0; i < bs->quiesce_counter; i++) {
1226         if (drv->bdrv_co_drain_begin) {
1227             drv->bdrv_co_drain_begin(bs);
1228         }
1229     }
1230 
1231     return 0;
1232 open_failed:
1233     bs->drv = NULL;
1234     if (bs->file != NULL) {
1235         bdrv_unref_child(bs, bs->file);
1236         bs->file = NULL;
1237     }
1238     g_free(bs->opaque);
1239     bs->opaque = NULL;
1240     return ret;
1241 }
1242 
1243 BlockDriverState *bdrv_new_open_driver(BlockDriver *drv, const char *node_name,
1244                                        int flags, Error **errp)
1245 {
1246     BlockDriverState *bs;
1247     int ret;
1248 
1249     bs = bdrv_new();
1250     bs->open_flags = flags;
1251     bs->explicit_options = qdict_new();
1252     bs->options = qdict_new();
1253     bs->opaque = NULL;
1254 
1255     update_options_from_flags(bs->options, flags);
1256 
1257     ret = bdrv_open_driver(bs, drv, node_name, bs->options, flags, errp);
1258     if (ret < 0) {
1259         qobject_unref(bs->explicit_options);
1260         bs->explicit_options = NULL;
1261         qobject_unref(bs->options);
1262         bs->options = NULL;
1263         bdrv_unref(bs);
1264         return NULL;
1265     }
1266 
1267     return bs;
1268 }
1269 
1270 QemuOptsList bdrv_runtime_opts = {
1271     .name = "bdrv_common",
1272     .head = QTAILQ_HEAD_INITIALIZER(bdrv_runtime_opts.head),
1273     .desc = {
1274         {
1275             .name = "node-name",
1276             .type = QEMU_OPT_STRING,
1277             .help = "Node name of the block device node",
1278         },
1279         {
1280             .name = "driver",
1281             .type = QEMU_OPT_STRING,
1282             .help = "Block driver to use for the node",
1283         },
1284         {
1285             .name = BDRV_OPT_CACHE_DIRECT,
1286             .type = QEMU_OPT_BOOL,
1287             .help = "Bypass software writeback cache on the host",
1288         },
1289         {
1290             .name = BDRV_OPT_CACHE_NO_FLUSH,
1291             .type = QEMU_OPT_BOOL,
1292             .help = "Ignore flush requests",
1293         },
1294         {
1295             .name = BDRV_OPT_READ_ONLY,
1296             .type = QEMU_OPT_BOOL,
1297             .help = "Node is opened in read-only mode",
1298         },
1299         {
1300             .name = "detect-zeroes",
1301             .type = QEMU_OPT_STRING,
1302             .help = "try to optimize zero writes (off, on, unmap)",
1303         },
1304         {
1305             .name = "discard",
1306             .type = QEMU_OPT_STRING,
1307             .help = "discard operation (ignore/off, unmap/on)",
1308         },
1309         {
1310             .name = BDRV_OPT_FORCE_SHARE,
1311             .type = QEMU_OPT_BOOL,
1312             .help = "always accept other writers (default: off)",
1313         },
1314         { /* end of list */ }
1315     },
1316 };
1317 
1318 /*
1319  * Common part for opening disk images and files
1320  *
1321  * Removes all processed options from *options.
1322  */
1323 static int bdrv_open_common(BlockDriverState *bs, BlockBackend *file,
1324                             QDict *options, Error **errp)
1325 {
1326     int ret, open_flags;
1327     const char *filename;
1328     const char *driver_name = NULL;
1329     const char *node_name = NULL;
1330     const char *discard;
1331     const char *detect_zeroes;
1332     QemuOpts *opts;
1333     BlockDriver *drv;
1334     Error *local_err = NULL;
1335 
1336     assert(bs->file == NULL);
1337     assert(options != NULL && bs->options != options);
1338 
1339     opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
1340     qemu_opts_absorb_qdict(opts, options, &local_err);
1341     if (local_err) {
1342         error_propagate(errp, local_err);
1343         ret = -EINVAL;
1344         goto fail_opts;
1345     }
1346 
1347     update_flags_from_options(&bs->open_flags, opts);
1348 
1349     driver_name = qemu_opt_get(opts, "driver");
1350     drv = bdrv_find_format(driver_name);
1351     assert(drv != NULL);
1352 
1353     bs->force_share = qemu_opt_get_bool(opts, BDRV_OPT_FORCE_SHARE, false);
1354 
1355     if (bs->force_share && (bs->open_flags & BDRV_O_RDWR)) {
1356         error_setg(errp,
1357                    BDRV_OPT_FORCE_SHARE
1358                    "=on can only be used with read-only images");
1359         ret = -EINVAL;
1360         goto fail_opts;
1361     }
1362 
1363     if (file != NULL) {
1364         filename = blk_bs(file)->filename;
1365     } else {
1366         /*
1367          * Caution: while qdict_get_try_str() is fine, getting
1368          * non-string types would require more care.  When @options
1369          * come from -blockdev or blockdev_add, its members are typed
1370          * according to the QAPI schema, but when they come from
1371          * -drive, they're all QString.
1372          */
1373         filename = qdict_get_try_str(options, "filename");
1374     }
1375 
1376     if (drv->bdrv_needs_filename && (!filename || !filename[0])) {
1377         error_setg(errp, "The '%s' block driver requires a file name",
1378                    drv->format_name);
1379         ret = -EINVAL;
1380         goto fail_opts;
1381     }
1382 
1383     trace_bdrv_open_common(bs, filename ?: "", bs->open_flags,
1384                            drv->format_name);
1385 
1386     bs->read_only = !(bs->open_flags & BDRV_O_RDWR);
1387 
1388     if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv, bs->read_only)) {
1389         error_setg(errp,
1390                    !bs->read_only && bdrv_is_whitelisted(drv, true)
1391                         ? "Driver '%s' can only be used for read-only devices"
1392                         : "Driver '%s' is not whitelisted",
1393                    drv->format_name);
1394         ret = -ENOTSUP;
1395         goto fail_opts;
1396     }
1397 
1398     /* bdrv_new() and bdrv_close() make it so */
1399     assert(atomic_read(&bs->copy_on_read) == 0);
1400 
1401     if (bs->open_flags & BDRV_O_COPY_ON_READ) {
1402         if (!bs->read_only) {
1403             bdrv_enable_copy_on_read(bs);
1404         } else {
1405             error_setg(errp, "Can't use copy-on-read on read-only device");
1406             ret = -EINVAL;
1407             goto fail_opts;
1408         }
1409     }
1410 
1411     discard = qemu_opt_get(opts, "discard");
1412     if (discard != NULL) {
1413         if (bdrv_parse_discard_flags(discard, &bs->open_flags) != 0) {
1414             error_setg(errp, "Invalid discard option");
1415             ret = -EINVAL;
1416             goto fail_opts;
1417         }
1418     }
1419 
1420     detect_zeroes = qemu_opt_get(opts, "detect-zeroes");
1421     if (detect_zeroes) {
1422         BlockdevDetectZeroesOptions value =
1423             qapi_enum_parse(&BlockdevDetectZeroesOptions_lookup,
1424                             detect_zeroes,
1425                             BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF,
1426                             &local_err);
1427         if (local_err) {
1428             error_propagate(errp, local_err);
1429             ret = -EINVAL;
1430             goto fail_opts;
1431         }
1432 
1433         if (value == BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP &&
1434             !(bs->open_flags & BDRV_O_UNMAP))
1435         {
1436             error_setg(errp, "setting detect-zeroes to unmap is not allowed "
1437                              "without setting discard operation to unmap");
1438             ret = -EINVAL;
1439             goto fail_opts;
1440         }
1441 
1442         bs->detect_zeroes = value;
1443     }
1444 
1445     if (filename != NULL) {
1446         pstrcpy(bs->filename, sizeof(bs->filename), filename);
1447     } else {
1448         bs->filename[0] = '\0';
1449     }
1450     pstrcpy(bs->exact_filename, sizeof(bs->exact_filename), bs->filename);
1451 
1452     /* Open the image, either directly or using a protocol */
1453     open_flags = bdrv_open_flags(bs, bs->open_flags);
1454     node_name = qemu_opt_get(opts, "node-name");
1455 
1456     assert(!drv->bdrv_file_open || file == NULL);
1457     ret = bdrv_open_driver(bs, drv, node_name, options, open_flags, errp);
1458     if (ret < 0) {
1459         goto fail_opts;
1460     }
1461 
1462     qemu_opts_del(opts);
1463     return 0;
1464 
1465 fail_opts:
1466     qemu_opts_del(opts);
1467     return ret;
1468 }
1469 
1470 static QDict *parse_json_filename(const char *filename, Error **errp)
1471 {
1472     QObject *options_obj;
1473     QDict *options;
1474     int ret;
1475 
1476     ret = strstart(filename, "json:", &filename);
1477     assert(ret);
1478 
1479     options_obj = qobject_from_json(filename, errp);
1480     if (!options_obj) {
1481         error_prepend(errp, "Could not parse the JSON options: ");
1482         return NULL;
1483     }
1484 
1485     options = qobject_to(QDict, options_obj);
1486     if (!options) {
1487         qobject_unref(options_obj);
1488         error_setg(errp, "Invalid JSON object given");
1489         return NULL;
1490     }
1491 
1492     qdict_flatten(options);
1493 
1494     return options;
1495 }
1496 
1497 static void parse_json_protocol(QDict *options, const char **pfilename,
1498                                 Error **errp)
1499 {
1500     QDict *json_options;
1501     Error *local_err = NULL;
1502 
1503     /* Parse json: pseudo-protocol */
1504     if (!*pfilename || !g_str_has_prefix(*pfilename, "json:")) {
1505         return;
1506     }
1507 
1508     json_options = parse_json_filename(*pfilename, &local_err);
1509     if (local_err) {
1510         error_propagate(errp, local_err);
1511         return;
1512     }
1513 
1514     /* Options given in the filename have lower priority than options
1515      * specified directly */
1516     qdict_join(options, json_options, false);
1517     qobject_unref(json_options);
1518     *pfilename = NULL;
1519 }
1520 
1521 /*
1522  * Fills in default options for opening images and converts the legacy
1523  * filename/flags pair to option QDict entries.
1524  * The BDRV_O_PROTOCOL flag in *flags will be set or cleared accordingly if a
1525  * block driver has been specified explicitly.
1526  */
1527 static int bdrv_fill_options(QDict **options, const char *filename,
1528                              int *flags, Error **errp)
1529 {
1530     const char *drvname;
1531     bool protocol = *flags & BDRV_O_PROTOCOL;
1532     bool parse_filename = false;
1533     BlockDriver *drv = NULL;
1534     Error *local_err = NULL;
1535 
1536     /*
1537      * Caution: while qdict_get_try_str() is fine, getting non-string
1538      * types would require more care.  When @options come from
1539      * -blockdev or blockdev_add, its members are typed according to
1540      * the QAPI schema, but when they come from -drive, they're all
1541      * QString.
1542      */
1543     drvname = qdict_get_try_str(*options, "driver");
1544     if (drvname) {
1545         drv = bdrv_find_format(drvname);
1546         if (!drv) {
1547             error_setg(errp, "Unknown driver '%s'", drvname);
1548             return -ENOENT;
1549         }
1550         /* If the user has explicitly specified the driver, this choice should
1551          * override the BDRV_O_PROTOCOL flag */
1552         protocol = drv->bdrv_file_open;
1553     }
1554 
1555     if (protocol) {
1556         *flags |= BDRV_O_PROTOCOL;
1557     } else {
1558         *flags &= ~BDRV_O_PROTOCOL;
1559     }
1560 
1561     /* Translate cache options from flags into options */
1562     update_options_from_flags(*options, *flags);
1563 
1564     /* Fetch the file name from the options QDict if necessary */
1565     if (protocol && filename) {
1566         if (!qdict_haskey(*options, "filename")) {
1567             qdict_put_str(*options, "filename", filename);
1568             parse_filename = true;
1569         } else {
1570             error_setg(errp, "Can't specify 'file' and 'filename' options at "
1571                              "the same time");
1572             return -EINVAL;
1573         }
1574     }
1575 
1576     /* Find the right block driver */
1577     /* See cautionary note on accessing @options above */
1578     filename = qdict_get_try_str(*options, "filename");
1579 
1580     if (!drvname && protocol) {
1581         if (filename) {
1582             drv = bdrv_find_protocol(filename, parse_filename, errp);
1583             if (!drv) {
1584                 return -EINVAL;
1585             }
1586 
1587             drvname = drv->format_name;
1588             qdict_put_str(*options, "driver", drvname);
1589         } else {
1590             error_setg(errp, "Must specify either driver or file");
1591             return -EINVAL;
1592         }
1593     }
1594 
1595     assert(drv || !protocol);
1596 
1597     /* Driver-specific filename parsing */
1598     if (drv && drv->bdrv_parse_filename && parse_filename) {
1599         drv->bdrv_parse_filename(filename, *options, &local_err);
1600         if (local_err) {
1601             error_propagate(errp, local_err);
1602             return -EINVAL;
1603         }
1604 
1605         if (!drv->bdrv_needs_filename) {
1606             qdict_del(*options, "filename");
1607         }
1608     }
1609 
1610     return 0;
1611 }
1612 
1613 static int bdrv_child_check_perm(BdrvChild *c, BlockReopenQueue *q,
1614                                  uint64_t perm, uint64_t shared,
1615                                  GSList *ignore_children, Error **errp);
1616 static void bdrv_child_abort_perm_update(BdrvChild *c);
1617 static void bdrv_child_set_perm(BdrvChild *c, uint64_t perm, uint64_t shared);
1618 
1619 typedef struct BlockReopenQueueEntry {
1620      bool prepared;
1621      BDRVReopenState state;
1622      QSIMPLEQ_ENTRY(BlockReopenQueueEntry) entry;
1623 } BlockReopenQueueEntry;
1624 
1625 /*
1626  * Return the flags that @bs will have after the reopens in @q have
1627  * successfully completed. If @q is NULL (or @bs is not contained in @q),
1628  * return the current flags.
1629  */
1630 static int bdrv_reopen_get_flags(BlockReopenQueue *q, BlockDriverState *bs)
1631 {
1632     BlockReopenQueueEntry *entry;
1633 
1634     if (q != NULL) {
1635         QSIMPLEQ_FOREACH(entry, q, entry) {
1636             if (entry->state.bs == bs) {
1637                 return entry->state.flags;
1638             }
1639         }
1640     }
1641 
1642     return bs->open_flags;
1643 }
1644 
1645 /* Returns whether the image file can be written to after the reopen queue @q
1646  * has been successfully applied, or right now if @q is NULL. */
1647 static bool bdrv_is_writable_after_reopen(BlockDriverState *bs,
1648                                           BlockReopenQueue *q)
1649 {
1650     int flags = bdrv_reopen_get_flags(q, bs);
1651 
1652     return (flags & (BDRV_O_RDWR | BDRV_O_INACTIVE)) == BDRV_O_RDWR;
1653 }
1654 
1655 /*
1656  * Return whether the BDS can be written to.  This is not necessarily
1657  * the same as !bdrv_is_read_only(bs), as inactivated images may not
1658  * be written to but do not count as read-only images.
1659  */
1660 bool bdrv_is_writable(BlockDriverState *bs)
1661 {
1662     return bdrv_is_writable_after_reopen(bs, NULL);
1663 }
1664 
1665 static void bdrv_child_perm(BlockDriverState *bs, BlockDriverState *child_bs,
1666                             BdrvChild *c, const BdrvChildRole *role,
1667                             BlockReopenQueue *reopen_queue,
1668                             uint64_t parent_perm, uint64_t parent_shared,
1669                             uint64_t *nperm, uint64_t *nshared)
1670 {
1671     if (bs->drv && bs->drv->bdrv_child_perm) {
1672         bs->drv->bdrv_child_perm(bs, c, role, reopen_queue,
1673                                  parent_perm, parent_shared,
1674                                  nperm, nshared);
1675     }
1676     /* TODO Take force_share from reopen_queue */
1677     if (child_bs && child_bs->force_share) {
1678         *nshared = BLK_PERM_ALL;
1679     }
1680 }
1681 
1682 /*
1683  * Check whether permissions on this node can be changed in a way that
1684  * @cumulative_perms and @cumulative_shared_perms are the new cumulative
1685  * permissions of all its parents. This involves checking whether all necessary
1686  * permission changes to child nodes can be performed.
1687  *
1688  * A call to this function must always be followed by a call to bdrv_set_perm()
1689  * or bdrv_abort_perm_update().
1690  */
1691 static int bdrv_check_perm(BlockDriverState *bs, BlockReopenQueue *q,
1692                            uint64_t cumulative_perms,
1693                            uint64_t cumulative_shared_perms,
1694                            GSList *ignore_children, Error **errp)
1695 {
1696     BlockDriver *drv = bs->drv;
1697     BdrvChild *c;
1698     int ret;
1699 
1700     /* Write permissions never work with read-only images */
1701     if ((cumulative_perms & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)) &&
1702         !bdrv_is_writable_after_reopen(bs, q))
1703     {
1704         error_setg(errp, "Block node is read-only");
1705         return -EPERM;
1706     }
1707 
1708     /* Check this node */
1709     if (!drv) {
1710         return 0;
1711     }
1712 
1713     if (drv->bdrv_check_perm) {
1714         return drv->bdrv_check_perm(bs, cumulative_perms,
1715                                     cumulative_shared_perms, errp);
1716     }
1717 
1718     /* Drivers that never have children can omit .bdrv_child_perm() */
1719     if (!drv->bdrv_child_perm) {
1720         assert(QLIST_EMPTY(&bs->children));
1721         return 0;
1722     }
1723 
1724     /* Check all children */
1725     QLIST_FOREACH(c, &bs->children, next) {
1726         uint64_t cur_perm, cur_shared;
1727         bdrv_child_perm(bs, c->bs, c, c->role, q,
1728                         cumulative_perms, cumulative_shared_perms,
1729                         &cur_perm, &cur_shared);
1730         ret = bdrv_child_check_perm(c, q, cur_perm, cur_shared,
1731                                     ignore_children, errp);
1732         if (ret < 0) {
1733             return ret;
1734         }
1735     }
1736 
1737     return 0;
1738 }
1739 
1740 /*
1741  * Notifies drivers that after a previous bdrv_check_perm() call, the
1742  * permission update is not performed and any preparations made for it (e.g.
1743  * taken file locks) need to be undone.
1744  *
1745  * This function recursively notifies all child nodes.
1746  */
1747 static void bdrv_abort_perm_update(BlockDriverState *bs)
1748 {
1749     BlockDriver *drv = bs->drv;
1750     BdrvChild *c;
1751 
1752     if (!drv) {
1753         return;
1754     }
1755 
1756     if (drv->bdrv_abort_perm_update) {
1757         drv->bdrv_abort_perm_update(bs);
1758     }
1759 
1760     QLIST_FOREACH(c, &bs->children, next) {
1761         bdrv_child_abort_perm_update(c);
1762     }
1763 }
1764 
1765 static void bdrv_set_perm(BlockDriverState *bs, uint64_t cumulative_perms,
1766                           uint64_t cumulative_shared_perms)
1767 {
1768     BlockDriver *drv = bs->drv;
1769     BdrvChild *c;
1770 
1771     if (!drv) {
1772         return;
1773     }
1774 
1775     /* Update this node */
1776     if (drv->bdrv_set_perm) {
1777         drv->bdrv_set_perm(bs, cumulative_perms, cumulative_shared_perms);
1778     }
1779 
1780     /* Drivers that never have children can omit .bdrv_child_perm() */
1781     if (!drv->bdrv_child_perm) {
1782         assert(QLIST_EMPTY(&bs->children));
1783         return;
1784     }
1785 
1786     /* Update all children */
1787     QLIST_FOREACH(c, &bs->children, next) {
1788         uint64_t cur_perm, cur_shared;
1789         bdrv_child_perm(bs, c->bs, c, c->role, NULL,
1790                         cumulative_perms, cumulative_shared_perms,
1791                         &cur_perm, &cur_shared);
1792         bdrv_child_set_perm(c, cur_perm, cur_shared);
1793     }
1794 }
1795 
1796 static void bdrv_get_cumulative_perm(BlockDriverState *bs, uint64_t *perm,
1797                                      uint64_t *shared_perm)
1798 {
1799     BdrvChild *c;
1800     uint64_t cumulative_perms = 0;
1801     uint64_t cumulative_shared_perms = BLK_PERM_ALL;
1802 
1803     QLIST_FOREACH(c, &bs->parents, next_parent) {
1804         cumulative_perms |= c->perm;
1805         cumulative_shared_perms &= c->shared_perm;
1806     }
1807 
1808     *perm = cumulative_perms;
1809     *shared_perm = cumulative_shared_perms;
1810 }
1811 
1812 static char *bdrv_child_user_desc(BdrvChild *c)
1813 {
1814     if (c->role->get_parent_desc) {
1815         return c->role->get_parent_desc(c);
1816     }
1817 
1818     return g_strdup("another user");
1819 }
1820 
1821 char *bdrv_perm_names(uint64_t perm)
1822 {
1823     struct perm_name {
1824         uint64_t perm;
1825         const char *name;
1826     } permissions[] = {
1827         { BLK_PERM_CONSISTENT_READ, "consistent read" },
1828         { BLK_PERM_WRITE,           "write" },
1829         { BLK_PERM_WRITE_UNCHANGED, "write unchanged" },
1830         { BLK_PERM_RESIZE,          "resize" },
1831         { BLK_PERM_GRAPH_MOD,       "change children" },
1832         { 0, NULL }
1833     };
1834 
1835     char *result = g_strdup("");
1836     struct perm_name *p;
1837 
1838     for (p = permissions; p->name; p++) {
1839         if (perm & p->perm) {
1840             char *old = result;
1841             result = g_strdup_printf("%s%s%s", old, *old ? ", " : "", p->name);
1842             g_free(old);
1843         }
1844     }
1845 
1846     return result;
1847 }
1848 
1849 /*
1850  * Checks whether a new reference to @bs can be added if the new user requires
1851  * @new_used_perm/@new_shared_perm as its permissions. If @ignore_children is
1852  * set, the BdrvChild objects in this list are ignored in the calculations;
1853  * this allows checking permission updates for an existing reference.
1854  *
1855  * Needs to be followed by a call to either bdrv_set_perm() or
1856  * bdrv_abort_perm_update(). */
1857 static int bdrv_check_update_perm(BlockDriverState *bs, BlockReopenQueue *q,
1858                                   uint64_t new_used_perm,
1859                                   uint64_t new_shared_perm,
1860                                   GSList *ignore_children, Error **errp)
1861 {
1862     BdrvChild *c;
1863     uint64_t cumulative_perms = new_used_perm;
1864     uint64_t cumulative_shared_perms = new_shared_perm;
1865 
1866     /* There is no reason why anyone couldn't tolerate write_unchanged */
1867     assert(new_shared_perm & BLK_PERM_WRITE_UNCHANGED);
1868 
1869     QLIST_FOREACH(c, &bs->parents, next_parent) {
1870         if (g_slist_find(ignore_children, c)) {
1871             continue;
1872         }
1873 
1874         if ((new_used_perm & c->shared_perm) != new_used_perm) {
1875             char *user = bdrv_child_user_desc(c);
1876             char *perm_names = bdrv_perm_names(new_used_perm & ~c->shared_perm);
1877             error_setg(errp, "Conflicts with use by %s as '%s', which does not "
1878                              "allow '%s' on %s",
1879                        user, c->name, perm_names, bdrv_get_node_name(c->bs));
1880             g_free(user);
1881             g_free(perm_names);
1882             return -EPERM;
1883         }
1884 
1885         if ((c->perm & new_shared_perm) != c->perm) {
1886             char *user = bdrv_child_user_desc(c);
1887             char *perm_names = bdrv_perm_names(c->perm & ~new_shared_perm);
1888             error_setg(errp, "Conflicts with use by %s as '%s', which uses "
1889                              "'%s' on %s",
1890                        user, c->name, perm_names, bdrv_get_node_name(c->bs));
1891             g_free(user);
1892             g_free(perm_names);
1893             return -EPERM;
1894         }
1895 
1896         cumulative_perms |= c->perm;
1897         cumulative_shared_perms &= c->shared_perm;
1898     }
1899 
1900     return bdrv_check_perm(bs, q, cumulative_perms, cumulative_shared_perms,
1901                            ignore_children, errp);
1902 }
1903 
1904 /* Needs to be followed by a call to either bdrv_child_set_perm() or
1905  * bdrv_child_abort_perm_update(). */
1906 static int bdrv_child_check_perm(BdrvChild *c, BlockReopenQueue *q,
1907                                  uint64_t perm, uint64_t shared,
1908                                  GSList *ignore_children, Error **errp)
1909 {
1910     int ret;
1911 
1912     ignore_children = g_slist_prepend(g_slist_copy(ignore_children), c);
1913     ret = bdrv_check_update_perm(c->bs, q, perm, shared, ignore_children, errp);
1914     g_slist_free(ignore_children);
1915 
1916     return ret;
1917 }
1918 
1919 static void bdrv_child_set_perm(BdrvChild *c, uint64_t perm, uint64_t shared)
1920 {
1921     uint64_t cumulative_perms, cumulative_shared_perms;
1922 
1923     c->perm = perm;
1924     c->shared_perm = shared;
1925 
1926     bdrv_get_cumulative_perm(c->bs, &cumulative_perms,
1927                              &cumulative_shared_perms);
1928     bdrv_set_perm(c->bs, cumulative_perms, cumulative_shared_perms);
1929 }
1930 
1931 static void bdrv_child_abort_perm_update(BdrvChild *c)
1932 {
1933     bdrv_abort_perm_update(c->bs);
1934 }
1935 
1936 int bdrv_child_try_set_perm(BdrvChild *c, uint64_t perm, uint64_t shared,
1937                             Error **errp)
1938 {
1939     int ret;
1940 
1941     ret = bdrv_child_check_perm(c, NULL, perm, shared, NULL, errp);
1942     if (ret < 0) {
1943         bdrv_child_abort_perm_update(c);
1944         return ret;
1945     }
1946 
1947     bdrv_child_set_perm(c, perm, shared);
1948 
1949     return 0;
1950 }
1951 
1952 void bdrv_filter_default_perms(BlockDriverState *bs, BdrvChild *c,
1953                                const BdrvChildRole *role,
1954                                BlockReopenQueue *reopen_queue,
1955                                uint64_t perm, uint64_t shared,
1956                                uint64_t *nperm, uint64_t *nshared)
1957 {
1958     if (c == NULL) {
1959         *nperm = perm & DEFAULT_PERM_PASSTHROUGH;
1960         *nshared = (shared & DEFAULT_PERM_PASSTHROUGH) | DEFAULT_PERM_UNCHANGED;
1961         return;
1962     }
1963 
1964     *nperm = (perm & DEFAULT_PERM_PASSTHROUGH) |
1965              (c->perm & DEFAULT_PERM_UNCHANGED);
1966     *nshared = (shared & DEFAULT_PERM_PASSTHROUGH) |
1967                (c->shared_perm & DEFAULT_PERM_UNCHANGED);
1968 }
1969 
1970 void bdrv_format_default_perms(BlockDriverState *bs, BdrvChild *c,
1971                                const BdrvChildRole *role,
1972                                BlockReopenQueue *reopen_queue,
1973                                uint64_t perm, uint64_t shared,
1974                                uint64_t *nperm, uint64_t *nshared)
1975 {
1976     bool backing = (role == &child_backing);
1977     assert(role == &child_backing || role == &child_file);
1978 
1979     if (!backing) {
1980         int flags = bdrv_reopen_get_flags(reopen_queue, bs);
1981 
1982         /* Apart from the modifications below, the same permissions are
1983          * forwarded and left alone as for filters */
1984         bdrv_filter_default_perms(bs, c, role, reopen_queue, perm, shared,
1985                                   &perm, &shared);
1986 
1987         /* Format drivers may touch metadata even if the guest doesn't write */
1988         if (bdrv_is_writable_after_reopen(bs, reopen_queue)) {
1989             perm |= BLK_PERM_WRITE | BLK_PERM_RESIZE;
1990         }
1991 
1992         /* bs->file always needs to be consistent because of the metadata. We
1993          * can never allow other users to resize or write to it. */
1994         if (!(flags & BDRV_O_NO_IO)) {
1995             perm |= BLK_PERM_CONSISTENT_READ;
1996         }
1997         shared &= ~(BLK_PERM_WRITE | BLK_PERM_RESIZE);
1998     } else {
1999         /* We want consistent read from backing files if the parent needs it.
2000          * No other operations are performed on backing files. */
2001         perm &= BLK_PERM_CONSISTENT_READ;
2002 
2003         /* If the parent can deal with changing data, we're okay with a
2004          * writable and resizable backing file. */
2005         /* TODO Require !(perm & BLK_PERM_CONSISTENT_READ), too? */
2006         if (shared & BLK_PERM_WRITE) {
2007             shared = BLK_PERM_WRITE | BLK_PERM_RESIZE;
2008         } else {
2009             shared = 0;
2010         }
2011 
2012         shared |= BLK_PERM_CONSISTENT_READ | BLK_PERM_GRAPH_MOD |
2013                   BLK_PERM_WRITE_UNCHANGED;
2014     }
2015 
2016     if (bs->open_flags & BDRV_O_INACTIVE) {
2017         shared |= BLK_PERM_WRITE | BLK_PERM_RESIZE;
2018     }
2019 
2020     *nperm = perm;
2021     *nshared = shared;
2022 }
2023 
2024 static void bdrv_replace_child_noperm(BdrvChild *child,
2025                                       BlockDriverState *new_bs)
2026 {
2027     BlockDriverState *old_bs = child->bs;
2028     int i;
2029 
2030     if (old_bs && new_bs) {
2031         assert(bdrv_get_aio_context(old_bs) == bdrv_get_aio_context(new_bs));
2032     }
2033     if (old_bs) {
2034         /* Detach first so that the recursive drain sections coming from @child
2035          * are already gone and we only end the drain sections that came from
2036          * elsewhere. */
2037         if (child->role->detach) {
2038             child->role->detach(child);
2039         }
2040         if (old_bs->quiesce_counter && child->role->drained_end) {
2041             int num = old_bs->quiesce_counter;
2042             if (child->role->parent_is_bds) {
2043                 num -= bdrv_drain_all_count;
2044             }
2045             assert(num >= 0);
2046             for (i = 0; i < num; i++) {
2047                 child->role->drained_end(child);
2048             }
2049         }
2050         QLIST_REMOVE(child, next_parent);
2051     }
2052 
2053     child->bs = new_bs;
2054 
2055     if (new_bs) {
2056         QLIST_INSERT_HEAD(&new_bs->parents, child, next_parent);
2057         if (new_bs->quiesce_counter && child->role->drained_begin) {
2058             int num = new_bs->quiesce_counter;
2059             if (child->role->parent_is_bds) {
2060                 num -= bdrv_drain_all_count;
2061             }
2062             assert(num >= 0);
2063             for (i = 0; i < num; i++) {
2064                 bdrv_parent_drained_begin_single(child, true);
2065             }
2066         }
2067 
2068         /* Attach only after starting new drained sections, so that recursive
2069          * drain sections coming from @child don't get an extra .drained_begin
2070          * callback. */
2071         if (child->role->attach) {
2072             child->role->attach(child);
2073         }
2074     }
2075 }
2076 
2077 /*
2078  * Updates @child to change its reference to point to @new_bs, including
2079  * checking and applying the necessary permisson updates both to the old node
2080  * and to @new_bs.
2081  *
2082  * NULL is passed as @new_bs for removing the reference before freeing @child.
2083  *
2084  * If @new_bs is not NULL, bdrv_check_perm() must be called beforehand, as this
2085  * function uses bdrv_set_perm() to update the permissions according to the new
2086  * reference that @new_bs gets.
2087  */
2088 static void bdrv_replace_child(BdrvChild *child, BlockDriverState *new_bs)
2089 {
2090     BlockDriverState *old_bs = child->bs;
2091     uint64_t perm, shared_perm;
2092 
2093     bdrv_replace_child_noperm(child, new_bs);
2094 
2095     if (old_bs) {
2096         /* Update permissions for old node. This is guaranteed to succeed
2097          * because we're just taking a parent away, so we're loosening
2098          * restrictions. */
2099         bdrv_get_cumulative_perm(old_bs, &perm, &shared_perm);
2100         bdrv_check_perm(old_bs, NULL, perm, shared_perm, NULL, &error_abort);
2101         bdrv_set_perm(old_bs, perm, shared_perm);
2102     }
2103 
2104     if (new_bs) {
2105         bdrv_get_cumulative_perm(new_bs, &perm, &shared_perm);
2106         bdrv_set_perm(new_bs, perm, shared_perm);
2107     }
2108 }
2109 
2110 BdrvChild *bdrv_root_attach_child(BlockDriverState *child_bs,
2111                                   const char *child_name,
2112                                   const BdrvChildRole *child_role,
2113                                   uint64_t perm, uint64_t shared_perm,
2114                                   void *opaque, Error **errp)
2115 {
2116     BdrvChild *child;
2117     int ret;
2118 
2119     ret = bdrv_check_update_perm(child_bs, NULL, perm, shared_perm, NULL, errp);
2120     if (ret < 0) {
2121         bdrv_abort_perm_update(child_bs);
2122         return NULL;
2123     }
2124 
2125     child = g_new(BdrvChild, 1);
2126     *child = (BdrvChild) {
2127         .bs             = NULL,
2128         .name           = g_strdup(child_name),
2129         .role           = child_role,
2130         .perm           = perm,
2131         .shared_perm    = shared_perm,
2132         .opaque         = opaque,
2133     };
2134 
2135     /* This performs the matching bdrv_set_perm() for the above check. */
2136     bdrv_replace_child(child, child_bs);
2137 
2138     return child;
2139 }
2140 
2141 BdrvChild *bdrv_attach_child(BlockDriverState *parent_bs,
2142                              BlockDriverState *child_bs,
2143                              const char *child_name,
2144                              const BdrvChildRole *child_role,
2145                              Error **errp)
2146 {
2147     BdrvChild *child;
2148     uint64_t perm, shared_perm;
2149 
2150     bdrv_get_cumulative_perm(parent_bs, &perm, &shared_perm);
2151 
2152     assert(parent_bs->drv);
2153     assert(bdrv_get_aio_context(parent_bs) == bdrv_get_aio_context(child_bs));
2154     bdrv_child_perm(parent_bs, child_bs, NULL, child_role, NULL,
2155                     perm, shared_perm, &perm, &shared_perm);
2156 
2157     child = bdrv_root_attach_child(child_bs, child_name, child_role,
2158                                    perm, shared_perm, parent_bs, errp);
2159     if (child == NULL) {
2160         return NULL;
2161     }
2162 
2163     QLIST_INSERT_HEAD(&parent_bs->children, child, next);
2164     return child;
2165 }
2166 
2167 static void bdrv_detach_child(BdrvChild *child)
2168 {
2169     if (child->next.le_prev) {
2170         QLIST_REMOVE(child, next);
2171         child->next.le_prev = NULL;
2172     }
2173 
2174     bdrv_replace_child(child, NULL);
2175 
2176     g_free(child->name);
2177     g_free(child);
2178 }
2179 
2180 void bdrv_root_unref_child(BdrvChild *child)
2181 {
2182     BlockDriverState *child_bs;
2183 
2184     child_bs = child->bs;
2185     bdrv_detach_child(child);
2186     bdrv_unref(child_bs);
2187 }
2188 
2189 void bdrv_unref_child(BlockDriverState *parent, BdrvChild *child)
2190 {
2191     if (child == NULL) {
2192         return;
2193     }
2194 
2195     if (child->bs->inherits_from == parent) {
2196         BdrvChild *c;
2197 
2198         /* Remove inherits_from only when the last reference between parent and
2199          * child->bs goes away. */
2200         QLIST_FOREACH(c, &parent->children, next) {
2201             if (c != child && c->bs == child->bs) {
2202                 break;
2203             }
2204         }
2205         if (c == NULL) {
2206             child->bs->inherits_from = NULL;
2207         }
2208     }
2209 
2210     bdrv_root_unref_child(child);
2211 }
2212 
2213 
2214 static void bdrv_parent_cb_change_media(BlockDriverState *bs, bool load)
2215 {
2216     BdrvChild *c;
2217     QLIST_FOREACH(c, &bs->parents, next_parent) {
2218         if (c->role->change_media) {
2219             c->role->change_media(c, load);
2220         }
2221     }
2222 }
2223 
2224 /*
2225  * Sets the backing file link of a BDS. A new reference is created; callers
2226  * which don't need their own reference any more must call bdrv_unref().
2227  */
2228 void bdrv_set_backing_hd(BlockDriverState *bs, BlockDriverState *backing_hd,
2229                          Error **errp)
2230 {
2231     if (backing_hd) {
2232         bdrv_ref(backing_hd);
2233     }
2234 
2235     if (bs->backing) {
2236         bdrv_unref_child(bs, bs->backing);
2237     }
2238 
2239     if (!backing_hd) {
2240         bs->backing = NULL;
2241         goto out;
2242     }
2243 
2244     bs->backing = bdrv_attach_child(bs, backing_hd, "backing", &child_backing,
2245                                     errp);
2246     if (!bs->backing) {
2247         bdrv_unref(backing_hd);
2248     }
2249 
2250     bdrv_refresh_filename(bs);
2251 
2252 out:
2253     bdrv_refresh_limits(bs, NULL);
2254 }
2255 
2256 /*
2257  * Opens the backing file for a BlockDriverState if not yet open
2258  *
2259  * bdref_key specifies the key for the image's BlockdevRef in the options QDict.
2260  * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
2261  * itself, all options starting with "${bdref_key}." are considered part of the
2262  * BlockdevRef.
2263  *
2264  * TODO Can this be unified with bdrv_open_image()?
2265  */
2266 int bdrv_open_backing_file(BlockDriverState *bs, QDict *parent_options,
2267                            const char *bdref_key, Error **errp)
2268 {
2269     char *backing_filename = g_malloc0(PATH_MAX);
2270     char *bdref_key_dot;
2271     const char *reference = NULL;
2272     int ret = 0;
2273     BlockDriverState *backing_hd;
2274     QDict *options;
2275     QDict *tmp_parent_options = NULL;
2276     Error *local_err = NULL;
2277 
2278     if (bs->backing != NULL) {
2279         goto free_exit;
2280     }
2281 
2282     /* NULL means an empty set of options */
2283     if (parent_options == NULL) {
2284         tmp_parent_options = qdict_new();
2285         parent_options = tmp_parent_options;
2286     }
2287 
2288     bs->open_flags &= ~BDRV_O_NO_BACKING;
2289 
2290     bdref_key_dot = g_strdup_printf("%s.", bdref_key);
2291     qdict_extract_subqdict(parent_options, &options, bdref_key_dot);
2292     g_free(bdref_key_dot);
2293 
2294     /*
2295      * Caution: while qdict_get_try_str() is fine, getting non-string
2296      * types would require more care.  When @parent_options come from
2297      * -blockdev or blockdev_add, its members are typed according to
2298      * the QAPI schema, but when they come from -drive, they're all
2299      * QString.
2300      */
2301     reference = qdict_get_try_str(parent_options, bdref_key);
2302     if (reference || qdict_haskey(options, "file.filename")) {
2303         backing_filename[0] = '\0';
2304     } else if (bs->backing_file[0] == '\0' && qdict_size(options) == 0) {
2305         qobject_unref(options);
2306         goto free_exit;
2307     } else {
2308         bdrv_get_full_backing_filename(bs, backing_filename, PATH_MAX,
2309                                        &local_err);
2310         if (local_err) {
2311             ret = -EINVAL;
2312             error_propagate(errp, local_err);
2313             qobject_unref(options);
2314             goto free_exit;
2315         }
2316     }
2317 
2318     if (!bs->drv || !bs->drv->supports_backing) {
2319         ret = -EINVAL;
2320         error_setg(errp, "Driver doesn't support backing files");
2321         qobject_unref(options);
2322         goto free_exit;
2323     }
2324 
2325     if (!reference &&
2326         bs->backing_format[0] != '\0' && !qdict_haskey(options, "driver")) {
2327         qdict_put_str(options, "driver", bs->backing_format);
2328     }
2329 
2330     backing_hd = bdrv_open_inherit(*backing_filename ? backing_filename : NULL,
2331                                    reference, options, 0, bs, &child_backing,
2332                                    errp);
2333     if (!backing_hd) {
2334         bs->open_flags |= BDRV_O_NO_BACKING;
2335         error_prepend(errp, "Could not open backing file: ");
2336         ret = -EINVAL;
2337         goto free_exit;
2338     }
2339     bdrv_set_aio_context(backing_hd, bdrv_get_aio_context(bs));
2340 
2341     /* Hook up the backing file link; drop our reference, bs owns the
2342      * backing_hd reference now */
2343     bdrv_set_backing_hd(bs, backing_hd, &local_err);
2344     bdrv_unref(backing_hd);
2345     if (local_err) {
2346         error_propagate(errp, local_err);
2347         ret = -EINVAL;
2348         goto free_exit;
2349     }
2350 
2351     qdict_del(parent_options, bdref_key);
2352 
2353 free_exit:
2354     g_free(backing_filename);
2355     qobject_unref(tmp_parent_options);
2356     return ret;
2357 }
2358 
2359 static BlockDriverState *
2360 bdrv_open_child_bs(const char *filename, QDict *options, const char *bdref_key,
2361                    BlockDriverState *parent, const BdrvChildRole *child_role,
2362                    bool allow_none, Error **errp)
2363 {
2364     BlockDriverState *bs = NULL;
2365     QDict *image_options;
2366     char *bdref_key_dot;
2367     const char *reference;
2368 
2369     assert(child_role != NULL);
2370 
2371     bdref_key_dot = g_strdup_printf("%s.", bdref_key);
2372     qdict_extract_subqdict(options, &image_options, bdref_key_dot);
2373     g_free(bdref_key_dot);
2374 
2375     /*
2376      * Caution: while qdict_get_try_str() is fine, getting non-string
2377      * types would require more care.  When @options come from
2378      * -blockdev or blockdev_add, its members are typed according to
2379      * the QAPI schema, but when they come from -drive, they're all
2380      * QString.
2381      */
2382     reference = qdict_get_try_str(options, bdref_key);
2383     if (!filename && !reference && !qdict_size(image_options)) {
2384         if (!allow_none) {
2385             error_setg(errp, "A block device must be specified for \"%s\"",
2386                        bdref_key);
2387         }
2388         qobject_unref(image_options);
2389         goto done;
2390     }
2391 
2392     bs = bdrv_open_inherit(filename, reference, image_options, 0,
2393                            parent, child_role, errp);
2394     if (!bs) {
2395         goto done;
2396     }
2397 
2398 done:
2399     qdict_del(options, bdref_key);
2400     return bs;
2401 }
2402 
2403 /*
2404  * Opens a disk image whose options are given as BlockdevRef in another block
2405  * device's options.
2406  *
2407  * If allow_none is true, no image will be opened if filename is false and no
2408  * BlockdevRef is given. NULL will be returned, but errp remains unset.
2409  *
2410  * bdrev_key specifies the key for the image's BlockdevRef in the options QDict.
2411  * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
2412  * itself, all options starting with "${bdref_key}." are considered part of the
2413  * BlockdevRef.
2414  *
2415  * The BlockdevRef will be removed from the options QDict.
2416  */
2417 BdrvChild *bdrv_open_child(const char *filename,
2418                            QDict *options, const char *bdref_key,
2419                            BlockDriverState *parent,
2420                            const BdrvChildRole *child_role,
2421                            bool allow_none, Error **errp)
2422 {
2423     BdrvChild *c;
2424     BlockDriverState *bs;
2425 
2426     bs = bdrv_open_child_bs(filename, options, bdref_key, parent, child_role,
2427                             allow_none, errp);
2428     if (bs == NULL) {
2429         return NULL;
2430     }
2431 
2432     c = bdrv_attach_child(parent, bs, bdref_key, child_role, errp);
2433     if (!c) {
2434         bdrv_unref(bs);
2435         return NULL;
2436     }
2437 
2438     return c;
2439 }
2440 
2441 /* TODO Future callers may need to specify parent/child_role in order for
2442  * option inheritance to work. Existing callers use it for the root node. */
2443 BlockDriverState *bdrv_open_blockdev_ref(BlockdevRef *ref, Error **errp)
2444 {
2445     BlockDriverState *bs = NULL;
2446     Error *local_err = NULL;
2447     QObject *obj = NULL;
2448     QDict *qdict = NULL;
2449     const char *reference = NULL;
2450     Visitor *v = NULL;
2451 
2452     if (ref->type == QTYPE_QSTRING) {
2453         reference = ref->u.reference;
2454     } else {
2455         BlockdevOptions *options = &ref->u.definition;
2456         assert(ref->type == QTYPE_QDICT);
2457 
2458         v = qobject_output_visitor_new(&obj);
2459         visit_type_BlockdevOptions(v, NULL, &options, &local_err);
2460         if (local_err) {
2461             error_propagate(errp, local_err);
2462             goto fail;
2463         }
2464         visit_complete(v, &obj);
2465 
2466         qdict = qobject_to(QDict, obj);
2467         qdict_flatten(qdict);
2468 
2469         /* bdrv_open_inherit() defaults to the values in bdrv_flags (for
2470          * compatibility with other callers) rather than what we want as the
2471          * real defaults. Apply the defaults here instead. */
2472         qdict_set_default_str(qdict, BDRV_OPT_CACHE_DIRECT, "off");
2473         qdict_set_default_str(qdict, BDRV_OPT_CACHE_NO_FLUSH, "off");
2474         qdict_set_default_str(qdict, BDRV_OPT_READ_ONLY, "off");
2475     }
2476 
2477     bs = bdrv_open_inherit(NULL, reference, qdict, 0, NULL, NULL, errp);
2478     obj = NULL;
2479 
2480 fail:
2481     qobject_unref(obj);
2482     visit_free(v);
2483     return bs;
2484 }
2485 
2486 static BlockDriverState *bdrv_append_temp_snapshot(BlockDriverState *bs,
2487                                                    int flags,
2488                                                    QDict *snapshot_options,
2489                                                    Error **errp)
2490 {
2491     /* TODO: extra byte is a hack to ensure MAX_PATH space on Windows. */
2492     char *tmp_filename = g_malloc0(PATH_MAX + 1);
2493     int64_t total_size;
2494     QemuOpts *opts = NULL;
2495     BlockDriverState *bs_snapshot = NULL;
2496     Error *local_err = NULL;
2497     int ret;
2498 
2499     /* if snapshot, we create a temporary backing file and open it
2500        instead of opening 'filename' directly */
2501 
2502     /* Get the required size from the image */
2503     total_size = bdrv_getlength(bs);
2504     if (total_size < 0) {
2505         error_setg_errno(errp, -total_size, "Could not get image size");
2506         goto out;
2507     }
2508 
2509     /* Create the temporary image */
2510     ret = get_tmp_filename(tmp_filename, PATH_MAX + 1);
2511     if (ret < 0) {
2512         error_setg_errno(errp, -ret, "Could not get temporary filename");
2513         goto out;
2514     }
2515 
2516     opts = qemu_opts_create(bdrv_qcow2.create_opts, NULL, 0,
2517                             &error_abort);
2518     qemu_opt_set_number(opts, BLOCK_OPT_SIZE, total_size, &error_abort);
2519     ret = bdrv_create(&bdrv_qcow2, tmp_filename, opts, errp);
2520     qemu_opts_del(opts);
2521     if (ret < 0) {
2522         error_prepend(errp, "Could not create temporary overlay '%s': ",
2523                       tmp_filename);
2524         goto out;
2525     }
2526 
2527     /* Prepare options QDict for the temporary file */
2528     qdict_put_str(snapshot_options, "file.driver", "file");
2529     qdict_put_str(snapshot_options, "file.filename", tmp_filename);
2530     qdict_put_str(snapshot_options, "driver", "qcow2");
2531 
2532     bs_snapshot = bdrv_open(NULL, NULL, snapshot_options, flags, errp);
2533     snapshot_options = NULL;
2534     if (!bs_snapshot) {
2535         goto out;
2536     }
2537 
2538     /* bdrv_append() consumes a strong reference to bs_snapshot
2539      * (i.e. it will call bdrv_unref() on it) even on error, so in
2540      * order to be able to return one, we have to increase
2541      * bs_snapshot's refcount here */
2542     bdrv_ref(bs_snapshot);
2543     bdrv_append(bs_snapshot, bs, &local_err);
2544     if (local_err) {
2545         error_propagate(errp, local_err);
2546         bs_snapshot = NULL;
2547         goto out;
2548     }
2549 
2550 out:
2551     qobject_unref(snapshot_options);
2552     g_free(tmp_filename);
2553     return bs_snapshot;
2554 }
2555 
2556 /*
2557  * Opens a disk image (raw, qcow2, vmdk, ...)
2558  *
2559  * options is a QDict of options to pass to the block drivers, or NULL for an
2560  * empty set of options. The reference to the QDict belongs to the block layer
2561  * after the call (even on failure), so if the caller intends to reuse the
2562  * dictionary, it needs to use qobject_ref() before calling bdrv_open.
2563  *
2564  * If *pbs is NULL, a new BDS will be created with a pointer to it stored there.
2565  * If it is not NULL, the referenced BDS will be reused.
2566  *
2567  * The reference parameter may be used to specify an existing block device which
2568  * should be opened. If specified, neither options nor a filename may be given,
2569  * nor can an existing BDS be reused (that is, *pbs has to be NULL).
2570  */
2571 static BlockDriverState *bdrv_open_inherit(const char *filename,
2572                                            const char *reference,
2573                                            QDict *options, int flags,
2574                                            BlockDriverState *parent,
2575                                            const BdrvChildRole *child_role,
2576                                            Error **errp)
2577 {
2578     int ret;
2579     BlockBackend *file = NULL;
2580     BlockDriverState *bs;
2581     BlockDriver *drv = NULL;
2582     BdrvChild *child;
2583     const char *drvname;
2584     const char *backing;
2585     Error *local_err = NULL;
2586     QDict *snapshot_options = NULL;
2587     int snapshot_flags = 0;
2588 
2589     assert(!child_role || !flags);
2590     assert(!child_role == !parent);
2591 
2592     if (reference) {
2593         bool options_non_empty = options ? qdict_size(options) : false;
2594         qobject_unref(options);
2595 
2596         if (filename || options_non_empty) {
2597             error_setg(errp, "Cannot reference an existing block device with "
2598                        "additional options or a new filename");
2599             return NULL;
2600         }
2601 
2602         bs = bdrv_lookup_bs(reference, reference, errp);
2603         if (!bs) {
2604             return NULL;
2605         }
2606 
2607         bdrv_ref(bs);
2608         return bs;
2609     }
2610 
2611     bs = bdrv_new();
2612 
2613     /* NULL means an empty set of options */
2614     if (options == NULL) {
2615         options = qdict_new();
2616     }
2617 
2618     /* json: syntax counts as explicit options, as if in the QDict */
2619     parse_json_protocol(options, &filename, &local_err);
2620     if (local_err) {
2621         goto fail;
2622     }
2623 
2624     bs->explicit_options = qdict_clone_shallow(options);
2625 
2626     if (child_role) {
2627         bs->inherits_from = parent;
2628         child_role->inherit_options(&flags, options,
2629                                     parent->open_flags, parent->options);
2630     }
2631 
2632     ret = bdrv_fill_options(&options, filename, &flags, &local_err);
2633     if (local_err) {
2634         goto fail;
2635     }
2636 
2637     /*
2638      * Set the BDRV_O_RDWR and BDRV_O_ALLOW_RDWR flags.
2639      * Caution: getting a boolean member of @options requires care.
2640      * When @options come from -blockdev or blockdev_add, members are
2641      * typed according to the QAPI schema, but when they come from
2642      * -drive, they're all QString.
2643      */
2644     if (g_strcmp0(qdict_get_try_str(options, BDRV_OPT_READ_ONLY), "on") &&
2645         !qdict_get_try_bool(options, BDRV_OPT_READ_ONLY, false)) {
2646         flags |= (BDRV_O_RDWR | BDRV_O_ALLOW_RDWR);
2647     } else {
2648         flags &= ~BDRV_O_RDWR;
2649     }
2650 
2651     if (flags & BDRV_O_SNAPSHOT) {
2652         snapshot_options = qdict_new();
2653         bdrv_temp_snapshot_options(&snapshot_flags, snapshot_options,
2654                                    flags, options);
2655         /* Let bdrv_backing_options() override "read-only" */
2656         qdict_del(options, BDRV_OPT_READ_ONLY);
2657         bdrv_backing_options(&flags, options, flags, options);
2658     }
2659 
2660     bs->open_flags = flags;
2661     bs->options = options;
2662     options = qdict_clone_shallow(options);
2663 
2664     /* Find the right image format driver */
2665     /* See cautionary note on accessing @options above */
2666     drvname = qdict_get_try_str(options, "driver");
2667     if (drvname) {
2668         drv = bdrv_find_format(drvname);
2669         if (!drv) {
2670             error_setg(errp, "Unknown driver: '%s'", drvname);
2671             goto fail;
2672         }
2673     }
2674 
2675     assert(drvname || !(flags & BDRV_O_PROTOCOL));
2676 
2677     /* See cautionary note on accessing @options above */
2678     backing = qdict_get_try_str(options, "backing");
2679     if (qobject_to(QNull, qdict_get(options, "backing")) != NULL ||
2680         (backing && *backing == '\0'))
2681     {
2682         if (backing) {
2683             warn_report("Use of \"backing\": \"\" is deprecated; "
2684                         "use \"backing\": null instead");
2685         }
2686         flags |= BDRV_O_NO_BACKING;
2687         qdict_del(options, "backing");
2688     }
2689 
2690     /* Open image file without format layer. This BlockBackend is only used for
2691      * probing, the block drivers will do their own bdrv_open_child() for the
2692      * same BDS, which is why we put the node name back into options. */
2693     if ((flags & BDRV_O_PROTOCOL) == 0) {
2694         BlockDriverState *file_bs;
2695 
2696         file_bs = bdrv_open_child_bs(filename, options, "file", bs,
2697                                      &child_file, true, &local_err);
2698         if (local_err) {
2699             goto fail;
2700         }
2701         if (file_bs != NULL) {
2702             /* Not requesting BLK_PERM_CONSISTENT_READ because we're only
2703              * looking at the header to guess the image format. This works even
2704              * in cases where a guest would not see a consistent state. */
2705             file = blk_new(0, BLK_PERM_ALL);
2706             blk_insert_bs(file, file_bs, &local_err);
2707             bdrv_unref(file_bs);
2708             if (local_err) {
2709                 goto fail;
2710             }
2711 
2712             qdict_put_str(options, "file", bdrv_get_node_name(file_bs));
2713         }
2714     }
2715 
2716     /* Image format probing */
2717     bs->probed = !drv;
2718     if (!drv && file) {
2719         ret = find_image_format(file, filename, &drv, &local_err);
2720         if (ret < 0) {
2721             goto fail;
2722         }
2723         /*
2724          * This option update would logically belong in bdrv_fill_options(),
2725          * but we first need to open bs->file for the probing to work, while
2726          * opening bs->file already requires the (mostly) final set of options
2727          * so that cache mode etc. can be inherited.
2728          *
2729          * Adding the driver later is somewhat ugly, but it's not an option
2730          * that would ever be inherited, so it's correct. We just need to make
2731          * sure to update both bs->options (which has the full effective
2732          * options for bs) and options (which has file.* already removed).
2733          */
2734         qdict_put_str(bs->options, "driver", drv->format_name);
2735         qdict_put_str(options, "driver", drv->format_name);
2736     } else if (!drv) {
2737         error_setg(errp, "Must specify either driver or file");
2738         goto fail;
2739     }
2740 
2741     /* BDRV_O_PROTOCOL must be set iff a protocol BDS is about to be created */
2742     assert(!!(flags & BDRV_O_PROTOCOL) == !!drv->bdrv_file_open);
2743     /* file must be NULL if a protocol BDS is about to be created
2744      * (the inverse results in an error message from bdrv_open_common()) */
2745     assert(!(flags & BDRV_O_PROTOCOL) || !file);
2746 
2747     /* Open the image */
2748     ret = bdrv_open_common(bs, file, options, &local_err);
2749     if (ret < 0) {
2750         goto fail;
2751     }
2752 
2753     if (file) {
2754         blk_unref(file);
2755         file = NULL;
2756     }
2757 
2758     /* If there is a backing file, use it */
2759     if ((flags & BDRV_O_NO_BACKING) == 0) {
2760         ret = bdrv_open_backing_file(bs, options, "backing", &local_err);
2761         if (ret < 0) {
2762             goto close_and_fail;
2763         }
2764     }
2765 
2766     /* Remove all children options from bs->options and bs->explicit_options */
2767     QLIST_FOREACH(child, &bs->children, next) {
2768         char *child_key_dot;
2769         child_key_dot = g_strdup_printf("%s.", child->name);
2770         qdict_extract_subqdict(bs->explicit_options, NULL, child_key_dot);
2771         qdict_extract_subqdict(bs->options, NULL, child_key_dot);
2772         g_free(child_key_dot);
2773     }
2774 
2775     bdrv_refresh_filename(bs);
2776 
2777     /* Check if any unknown options were used */
2778     if (qdict_size(options) != 0) {
2779         const QDictEntry *entry = qdict_first(options);
2780         if (flags & BDRV_O_PROTOCOL) {
2781             error_setg(errp, "Block protocol '%s' doesn't support the option "
2782                        "'%s'", drv->format_name, entry->key);
2783         } else {
2784             error_setg(errp,
2785                        "Block format '%s' does not support the option '%s'",
2786                        drv->format_name, entry->key);
2787         }
2788 
2789         goto close_and_fail;
2790     }
2791 
2792     bdrv_parent_cb_change_media(bs, true);
2793 
2794     qobject_unref(options);
2795 
2796     /* For snapshot=on, create a temporary qcow2 overlay. bs points to the
2797      * temporary snapshot afterwards. */
2798     if (snapshot_flags) {
2799         BlockDriverState *snapshot_bs;
2800         snapshot_bs = bdrv_append_temp_snapshot(bs, snapshot_flags,
2801                                                 snapshot_options, &local_err);
2802         snapshot_options = NULL;
2803         if (local_err) {
2804             goto close_and_fail;
2805         }
2806         /* We are not going to return bs but the overlay on top of it
2807          * (snapshot_bs); thus, we have to drop the strong reference to bs
2808          * (which we obtained by calling bdrv_new()). bs will not be deleted,
2809          * though, because the overlay still has a reference to it. */
2810         bdrv_unref(bs);
2811         bs = snapshot_bs;
2812     }
2813 
2814     return bs;
2815 
2816 fail:
2817     blk_unref(file);
2818     qobject_unref(snapshot_options);
2819     qobject_unref(bs->explicit_options);
2820     qobject_unref(bs->options);
2821     qobject_unref(options);
2822     bs->options = NULL;
2823     bs->explicit_options = NULL;
2824     bdrv_unref(bs);
2825     error_propagate(errp, local_err);
2826     return NULL;
2827 
2828 close_and_fail:
2829     bdrv_unref(bs);
2830     qobject_unref(snapshot_options);
2831     qobject_unref(options);
2832     error_propagate(errp, local_err);
2833     return NULL;
2834 }
2835 
2836 BlockDriverState *bdrv_open(const char *filename, const char *reference,
2837                             QDict *options, int flags, Error **errp)
2838 {
2839     return bdrv_open_inherit(filename, reference, options, flags, NULL,
2840                              NULL, errp);
2841 }
2842 
2843 /*
2844  * Adds a BlockDriverState to a simple queue for an atomic, transactional
2845  * reopen of multiple devices.
2846  *
2847  * bs_queue can either be an existing BlockReopenQueue that has had QSIMPLE_INIT
2848  * already performed, or alternatively may be NULL a new BlockReopenQueue will
2849  * be created and initialized. This newly created BlockReopenQueue should be
2850  * passed back in for subsequent calls that are intended to be of the same
2851  * atomic 'set'.
2852  *
2853  * bs is the BlockDriverState to add to the reopen queue.
2854  *
2855  * options contains the changed options for the associated bs
2856  * (the BlockReopenQueue takes ownership)
2857  *
2858  * flags contains the open flags for the associated bs
2859  *
2860  * returns a pointer to bs_queue, which is either the newly allocated
2861  * bs_queue, or the existing bs_queue being used.
2862  *
2863  * bs must be drained between bdrv_reopen_queue() and bdrv_reopen_multiple().
2864  */
2865 static BlockReopenQueue *bdrv_reopen_queue_child(BlockReopenQueue *bs_queue,
2866                                                  BlockDriverState *bs,
2867                                                  QDict *options,
2868                                                  int flags,
2869                                                  const BdrvChildRole *role,
2870                                                  QDict *parent_options,
2871                                                  int parent_flags)
2872 {
2873     assert(bs != NULL);
2874 
2875     BlockReopenQueueEntry *bs_entry;
2876     BdrvChild *child;
2877     QDict *old_options, *explicit_options;
2878 
2879     /* Make sure that the caller remembered to use a drained section. This is
2880      * important to avoid graph changes between the recursive queuing here and
2881      * bdrv_reopen_multiple(). */
2882     assert(bs->quiesce_counter > 0);
2883 
2884     if (bs_queue == NULL) {
2885         bs_queue = g_new0(BlockReopenQueue, 1);
2886         QSIMPLEQ_INIT(bs_queue);
2887     }
2888 
2889     if (!options) {
2890         options = qdict_new();
2891     }
2892 
2893     /* Check if this BlockDriverState is already in the queue */
2894     QSIMPLEQ_FOREACH(bs_entry, bs_queue, entry) {
2895         if (bs == bs_entry->state.bs) {
2896             break;
2897         }
2898     }
2899 
2900     /*
2901      * Precedence of options:
2902      * 1. Explicitly passed in options (highest)
2903      * 2. Set in flags (only for top level)
2904      * 3. Retained from explicitly set options of bs
2905      * 4. Inherited from parent node
2906      * 5. Retained from effective options of bs
2907      */
2908 
2909     if (!parent_options) {
2910         /*
2911          * Any setting represented by flags is always updated. If the
2912          * corresponding QDict option is set, it takes precedence. Otherwise
2913          * the flag is translated into a QDict option. The old setting of bs is
2914          * not considered.
2915          */
2916         update_options_from_flags(options, flags);
2917     }
2918 
2919     /* Old explicitly set values (don't overwrite by inherited value) */
2920     if (bs_entry) {
2921         old_options = qdict_clone_shallow(bs_entry->state.explicit_options);
2922     } else {
2923         old_options = qdict_clone_shallow(bs->explicit_options);
2924     }
2925     bdrv_join_options(bs, options, old_options);
2926     qobject_unref(old_options);
2927 
2928     explicit_options = qdict_clone_shallow(options);
2929 
2930     /* Inherit from parent node */
2931     if (parent_options) {
2932         QemuOpts *opts;
2933         QDict *options_copy;
2934         assert(!flags);
2935         role->inherit_options(&flags, options, parent_flags, parent_options);
2936         options_copy = qdict_clone_shallow(options);
2937         opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
2938         qemu_opts_absorb_qdict(opts, options_copy, NULL);
2939         update_flags_from_options(&flags, opts);
2940         qemu_opts_del(opts);
2941         qobject_unref(options_copy);
2942     }
2943 
2944     /* Old values are used for options that aren't set yet */
2945     old_options = qdict_clone_shallow(bs->options);
2946     bdrv_join_options(bs, options, old_options);
2947     qobject_unref(old_options);
2948 
2949     /* bdrv_open_inherit() sets and clears some additional flags internally */
2950     flags &= ~BDRV_O_PROTOCOL;
2951     if (flags & BDRV_O_RDWR) {
2952         flags |= BDRV_O_ALLOW_RDWR;
2953     }
2954 
2955     if (!bs_entry) {
2956         bs_entry = g_new0(BlockReopenQueueEntry, 1);
2957         QSIMPLEQ_INSERT_TAIL(bs_queue, bs_entry, entry);
2958     } else {
2959         qobject_unref(bs_entry->state.options);
2960         qobject_unref(bs_entry->state.explicit_options);
2961     }
2962 
2963     bs_entry->state.bs = bs;
2964     bs_entry->state.options = options;
2965     bs_entry->state.explicit_options = explicit_options;
2966     bs_entry->state.flags = flags;
2967 
2968     /* This needs to be overwritten in bdrv_reopen_prepare() */
2969     bs_entry->state.perm = UINT64_MAX;
2970     bs_entry->state.shared_perm = 0;
2971 
2972     QLIST_FOREACH(child, &bs->children, next) {
2973         QDict *new_child_options;
2974         char *child_key_dot;
2975 
2976         /* reopen can only change the options of block devices that were
2977          * implicitly created and inherited options. For other (referenced)
2978          * block devices, a syntax like "backing.foo" results in an error. */
2979         if (child->bs->inherits_from != bs) {
2980             continue;
2981         }
2982 
2983         child_key_dot = g_strdup_printf("%s.", child->name);
2984         qdict_extract_subqdict(explicit_options, NULL, child_key_dot);
2985         qdict_extract_subqdict(options, &new_child_options, child_key_dot);
2986         g_free(child_key_dot);
2987 
2988         bdrv_reopen_queue_child(bs_queue, child->bs, new_child_options, 0,
2989                                 child->role, options, flags);
2990     }
2991 
2992     return bs_queue;
2993 }
2994 
2995 BlockReopenQueue *bdrv_reopen_queue(BlockReopenQueue *bs_queue,
2996                                     BlockDriverState *bs,
2997                                     QDict *options, int flags)
2998 {
2999     return bdrv_reopen_queue_child(bs_queue, bs, options, flags,
3000                                    NULL, NULL, 0);
3001 }
3002 
3003 /*
3004  * Reopen multiple BlockDriverStates atomically & transactionally.
3005  *
3006  * The queue passed in (bs_queue) must have been built up previous
3007  * via bdrv_reopen_queue().
3008  *
3009  * Reopens all BDS specified in the queue, with the appropriate
3010  * flags.  All devices are prepared for reopen, and failure of any
3011  * device will cause all device changes to be abandoned, and intermediate
3012  * data cleaned up.
3013  *
3014  * If all devices prepare successfully, then the changes are committed
3015  * to all devices.
3016  *
3017  * All affected nodes must be drained between bdrv_reopen_queue() and
3018  * bdrv_reopen_multiple().
3019  */
3020 int bdrv_reopen_multiple(AioContext *ctx, BlockReopenQueue *bs_queue, Error **errp)
3021 {
3022     int ret = -1;
3023     BlockReopenQueueEntry *bs_entry, *next;
3024     Error *local_err = NULL;
3025 
3026     assert(bs_queue != NULL);
3027 
3028     QSIMPLEQ_FOREACH(bs_entry, bs_queue, entry) {
3029         assert(bs_entry->state.bs->quiesce_counter > 0);
3030         if (bdrv_reopen_prepare(&bs_entry->state, bs_queue, &local_err)) {
3031             error_propagate(errp, local_err);
3032             goto cleanup;
3033         }
3034         bs_entry->prepared = true;
3035     }
3036 
3037     /* If we reach this point, we have success and just need to apply the
3038      * changes
3039      */
3040     QSIMPLEQ_FOREACH(bs_entry, bs_queue, entry) {
3041         bdrv_reopen_commit(&bs_entry->state);
3042     }
3043 
3044     ret = 0;
3045 
3046 cleanup:
3047     QSIMPLEQ_FOREACH_SAFE(bs_entry, bs_queue, entry, next) {
3048         if (ret) {
3049             if (bs_entry->prepared) {
3050                 bdrv_reopen_abort(&bs_entry->state);
3051             }
3052             qobject_unref(bs_entry->state.explicit_options);
3053             qobject_unref(bs_entry->state.options);
3054         }
3055         g_free(bs_entry);
3056     }
3057     g_free(bs_queue);
3058 
3059     return ret;
3060 }
3061 
3062 
3063 /* Reopen a single BlockDriverState with the specified flags. */
3064 int bdrv_reopen(BlockDriverState *bs, int bdrv_flags, Error **errp)
3065 {
3066     int ret = -1;
3067     Error *local_err = NULL;
3068     BlockReopenQueue *queue;
3069 
3070     bdrv_subtree_drained_begin(bs);
3071 
3072     queue = bdrv_reopen_queue(NULL, bs, NULL, bdrv_flags);
3073     ret = bdrv_reopen_multiple(bdrv_get_aio_context(bs), queue, &local_err);
3074     if (local_err != NULL) {
3075         error_propagate(errp, local_err);
3076     }
3077 
3078     bdrv_subtree_drained_end(bs);
3079 
3080     return ret;
3081 }
3082 
3083 static BlockReopenQueueEntry *find_parent_in_reopen_queue(BlockReopenQueue *q,
3084                                                           BdrvChild *c)
3085 {
3086     BlockReopenQueueEntry *entry;
3087 
3088     QSIMPLEQ_FOREACH(entry, q, entry) {
3089         BlockDriverState *bs = entry->state.bs;
3090         BdrvChild *child;
3091 
3092         QLIST_FOREACH(child, &bs->children, next) {
3093             if (child == c) {
3094                 return entry;
3095             }
3096         }
3097     }
3098 
3099     return NULL;
3100 }
3101 
3102 static void bdrv_reopen_perm(BlockReopenQueue *q, BlockDriverState *bs,
3103                              uint64_t *perm, uint64_t *shared)
3104 {
3105     BdrvChild *c;
3106     BlockReopenQueueEntry *parent;
3107     uint64_t cumulative_perms = 0;
3108     uint64_t cumulative_shared_perms = BLK_PERM_ALL;
3109 
3110     QLIST_FOREACH(c, &bs->parents, next_parent) {
3111         parent = find_parent_in_reopen_queue(q, c);
3112         if (!parent) {
3113             cumulative_perms |= c->perm;
3114             cumulative_shared_perms &= c->shared_perm;
3115         } else {
3116             uint64_t nperm, nshared;
3117 
3118             bdrv_child_perm(parent->state.bs, bs, c, c->role, q,
3119                             parent->state.perm, parent->state.shared_perm,
3120                             &nperm, &nshared);
3121 
3122             cumulative_perms |= nperm;
3123             cumulative_shared_perms &= nshared;
3124         }
3125     }
3126     *perm = cumulative_perms;
3127     *shared = cumulative_shared_perms;
3128 }
3129 
3130 /*
3131  * Prepares a BlockDriverState for reopen. All changes are staged in the
3132  * 'opaque' field of the BDRVReopenState, which is used and allocated by
3133  * the block driver layer .bdrv_reopen_prepare()
3134  *
3135  * bs is the BlockDriverState to reopen
3136  * flags are the new open flags
3137  * queue is the reopen queue
3138  *
3139  * Returns 0 on success, non-zero on error.  On error errp will be set
3140  * as well.
3141  *
3142  * On failure, bdrv_reopen_abort() will be called to clean up any data.
3143  * It is the responsibility of the caller to then call the abort() or
3144  * commit() for any other BDS that have been left in a prepare() state
3145  *
3146  */
3147 int bdrv_reopen_prepare(BDRVReopenState *reopen_state, BlockReopenQueue *queue,
3148                         Error **errp)
3149 {
3150     int ret = -1;
3151     Error *local_err = NULL;
3152     BlockDriver *drv;
3153     QemuOpts *opts;
3154     QDict *orig_reopen_opts;
3155     const char *value;
3156     bool read_only;
3157 
3158     assert(reopen_state != NULL);
3159     assert(reopen_state->bs->drv != NULL);
3160     drv = reopen_state->bs->drv;
3161 
3162     /* This function and each driver's bdrv_reopen_prepare() remove
3163      * entries from reopen_state->options as they are processed, so
3164      * we need to make a copy of the original QDict. */
3165     orig_reopen_opts = qdict_clone_shallow(reopen_state->options);
3166 
3167     /* Process generic block layer options */
3168     opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
3169     qemu_opts_absorb_qdict(opts, reopen_state->options, &local_err);
3170     if (local_err) {
3171         error_propagate(errp, local_err);
3172         ret = -EINVAL;
3173         goto error;
3174     }
3175 
3176     update_flags_from_options(&reopen_state->flags, opts);
3177 
3178     /* node-name and driver must be unchanged. Put them back into the QDict, so
3179      * that they are checked at the end of this function. */
3180     value = qemu_opt_get(opts, "node-name");
3181     if (value) {
3182         qdict_put_str(reopen_state->options, "node-name", value);
3183     }
3184 
3185     value = qemu_opt_get(opts, "driver");
3186     if (value) {
3187         qdict_put_str(reopen_state->options, "driver", value);
3188     }
3189 
3190     /* If we are to stay read-only, do not allow permission change
3191      * to r/w. Attempting to set to r/w may fail if either BDRV_O_ALLOW_RDWR is
3192      * not set, or if the BDS still has copy_on_read enabled */
3193     read_only = !(reopen_state->flags & BDRV_O_RDWR);
3194     ret = bdrv_can_set_read_only(reopen_state->bs, read_only, true, &local_err);
3195     if (local_err) {
3196         error_propagate(errp, local_err);
3197         goto error;
3198     }
3199 
3200     /* Calculate required permissions after reopening */
3201     bdrv_reopen_perm(queue, reopen_state->bs,
3202                      &reopen_state->perm, &reopen_state->shared_perm);
3203 
3204     ret = bdrv_flush(reopen_state->bs);
3205     if (ret) {
3206         error_setg_errno(errp, -ret, "Error flushing drive");
3207         goto error;
3208     }
3209 
3210     if (drv->bdrv_reopen_prepare) {
3211         ret = drv->bdrv_reopen_prepare(reopen_state, queue, &local_err);
3212         if (ret) {
3213             if (local_err != NULL) {
3214                 error_propagate(errp, local_err);
3215             } else {
3216                 error_setg(errp, "failed while preparing to reopen image '%s'",
3217                            reopen_state->bs->filename);
3218             }
3219             goto error;
3220         }
3221     } else {
3222         /* It is currently mandatory to have a bdrv_reopen_prepare()
3223          * handler for each supported drv. */
3224         error_setg(errp, "Block format '%s' used by node '%s' "
3225                    "does not support reopening files", drv->format_name,
3226                    bdrv_get_device_or_node_name(reopen_state->bs));
3227         ret = -1;
3228         goto error;
3229     }
3230 
3231     /* Options that are not handled are only okay if they are unchanged
3232      * compared to the old state. It is expected that some options are only
3233      * used for the initial open, but not reopen (e.g. filename) */
3234     if (qdict_size(reopen_state->options)) {
3235         const QDictEntry *entry = qdict_first(reopen_state->options);
3236 
3237         do {
3238             QObject *new = entry->value;
3239             QObject *old = qdict_get(reopen_state->bs->options, entry->key);
3240 
3241             /*
3242              * TODO: When using -drive to specify blockdev options, all values
3243              * will be strings; however, when using -blockdev, blockdev-add or
3244              * filenames using the json:{} pseudo-protocol, they will be
3245              * correctly typed.
3246              * In contrast, reopening options are (currently) always strings
3247              * (because you can only specify them through qemu-io; all other
3248              * callers do not specify any options).
3249              * Therefore, when using anything other than -drive to create a BDS,
3250              * this cannot detect non-string options as unchanged, because
3251              * qobject_is_equal() always returns false for objects of different
3252              * type.  In the future, this should be remedied by correctly typing
3253              * all options.  For now, this is not too big of an issue because
3254              * the user can simply omit options which cannot be changed anyway,
3255              * so they will stay unchanged.
3256              */
3257             if (!qobject_is_equal(new, old)) {
3258                 error_setg(errp, "Cannot change the option '%s'", entry->key);
3259                 ret = -EINVAL;
3260                 goto error;
3261             }
3262         } while ((entry = qdict_next(reopen_state->options, entry)));
3263     }
3264 
3265     ret = bdrv_check_perm(reopen_state->bs, queue, reopen_state->perm,
3266                           reopen_state->shared_perm, NULL, errp);
3267     if (ret < 0) {
3268         goto error;
3269     }
3270 
3271     ret = 0;
3272 
3273     /* Restore the original reopen_state->options QDict */
3274     qobject_unref(reopen_state->options);
3275     reopen_state->options = qobject_ref(orig_reopen_opts);
3276 
3277 error:
3278     qemu_opts_del(opts);
3279     qobject_unref(orig_reopen_opts);
3280     return ret;
3281 }
3282 
3283 /*
3284  * Takes the staged changes for the reopen from bdrv_reopen_prepare(), and
3285  * makes them final by swapping the staging BlockDriverState contents into
3286  * the active BlockDriverState contents.
3287  */
3288 void bdrv_reopen_commit(BDRVReopenState *reopen_state)
3289 {
3290     BlockDriver *drv;
3291     BlockDriverState *bs;
3292     bool old_can_write, new_can_write;
3293 
3294     assert(reopen_state != NULL);
3295     bs = reopen_state->bs;
3296     drv = bs->drv;
3297     assert(drv != NULL);
3298 
3299     old_can_write =
3300         !bdrv_is_read_only(bs) && !(bdrv_get_flags(bs) & BDRV_O_INACTIVE);
3301 
3302     /* If there are any driver level actions to take */
3303     if (drv->bdrv_reopen_commit) {
3304         drv->bdrv_reopen_commit(reopen_state);
3305     }
3306 
3307     /* set BDS specific flags now */
3308     qobject_unref(bs->explicit_options);
3309     qobject_unref(bs->options);
3310 
3311     bs->explicit_options   = reopen_state->explicit_options;
3312     bs->options            = reopen_state->options;
3313     bs->open_flags         = reopen_state->flags;
3314     bs->read_only = !(reopen_state->flags & BDRV_O_RDWR);
3315 
3316     bdrv_refresh_limits(bs, NULL);
3317 
3318     bdrv_set_perm(reopen_state->bs, reopen_state->perm,
3319                   reopen_state->shared_perm);
3320 
3321     new_can_write =
3322         !bdrv_is_read_only(bs) && !(bdrv_get_flags(bs) & BDRV_O_INACTIVE);
3323     if (!old_can_write && new_can_write && drv->bdrv_reopen_bitmaps_rw) {
3324         Error *local_err = NULL;
3325         if (drv->bdrv_reopen_bitmaps_rw(bs, &local_err) < 0) {
3326             /* This is not fatal, bitmaps just left read-only, so all following
3327              * writes will fail. User can remove read-only bitmaps to unblock
3328              * writes.
3329              */
3330             error_reportf_err(local_err,
3331                               "%s: Failed to make dirty bitmaps writable: ",
3332                               bdrv_get_node_name(bs));
3333         }
3334     }
3335 }
3336 
3337 /*
3338  * Abort the reopen, and delete and free the staged changes in
3339  * reopen_state
3340  */
3341 void bdrv_reopen_abort(BDRVReopenState *reopen_state)
3342 {
3343     BlockDriver *drv;
3344 
3345     assert(reopen_state != NULL);
3346     drv = reopen_state->bs->drv;
3347     assert(drv != NULL);
3348 
3349     if (drv->bdrv_reopen_abort) {
3350         drv->bdrv_reopen_abort(reopen_state);
3351     }
3352 
3353     bdrv_abort_perm_update(reopen_state->bs);
3354 }
3355 
3356 
3357 static void bdrv_close(BlockDriverState *bs)
3358 {
3359     BdrvAioNotifier *ban, *ban_next;
3360     BdrvChild *child, *next;
3361 
3362     assert(!bs->job);
3363     assert(!bs->refcnt);
3364 
3365     bdrv_drained_begin(bs); /* complete I/O */
3366     bdrv_flush(bs);
3367     bdrv_drain(bs); /* in case flush left pending I/O */
3368 
3369     if (bs->drv) {
3370         if (bs->drv->bdrv_close) {
3371             bs->drv->bdrv_close(bs);
3372         }
3373         bs->drv = NULL;
3374     }
3375 
3376     bdrv_set_backing_hd(bs, NULL, &error_abort);
3377 
3378     if (bs->file != NULL) {
3379         bdrv_unref_child(bs, bs->file);
3380         bs->file = NULL;
3381     }
3382 
3383     QLIST_FOREACH_SAFE(child, &bs->children, next, next) {
3384         /* TODO Remove bdrv_unref() from drivers' close function and use
3385          * bdrv_unref_child() here */
3386         if (child->bs->inherits_from == bs) {
3387             child->bs->inherits_from = NULL;
3388         }
3389         bdrv_detach_child(child);
3390     }
3391 
3392     g_free(bs->opaque);
3393     bs->opaque = NULL;
3394     atomic_set(&bs->copy_on_read, 0);
3395     bs->backing_file[0] = '\0';
3396     bs->backing_format[0] = '\0';
3397     bs->total_sectors = 0;
3398     bs->encrypted = false;
3399     bs->sg = false;
3400     qobject_unref(bs->options);
3401     qobject_unref(bs->explicit_options);
3402     bs->options = NULL;
3403     bs->explicit_options = NULL;
3404     qobject_unref(bs->full_open_options);
3405     bs->full_open_options = NULL;
3406 
3407     bdrv_release_named_dirty_bitmaps(bs);
3408     assert(QLIST_EMPTY(&bs->dirty_bitmaps));
3409 
3410     QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_next) {
3411         g_free(ban);
3412     }
3413     QLIST_INIT(&bs->aio_notifiers);
3414     bdrv_drained_end(bs);
3415 }
3416 
3417 void bdrv_close_all(void)
3418 {
3419     assert(job_next(NULL) == NULL);
3420     nbd_export_close_all();
3421 
3422     /* Drop references from requests still in flight, such as canceled block
3423      * jobs whose AIO context has not been polled yet */
3424     bdrv_drain_all();
3425 
3426     blk_remove_all_bs();
3427     blockdev_close_all_bdrv_states();
3428 
3429     assert(QTAILQ_EMPTY(&all_bdrv_states));
3430 }
3431 
3432 static bool should_update_child(BdrvChild *c, BlockDriverState *to)
3433 {
3434     BdrvChild *to_c;
3435 
3436     if (c->role->stay_at_node) {
3437         return false;
3438     }
3439 
3440     /* If the child @c belongs to the BDS @to, replacing the current
3441      * c->bs by @to would mean to create a loop.
3442      *
3443      * Such a case occurs when appending a BDS to a backing chain.
3444      * For instance, imagine the following chain:
3445      *
3446      *   guest device -> node A -> further backing chain...
3447      *
3448      * Now we create a new BDS B which we want to put on top of this
3449      * chain, so we first attach A as its backing node:
3450      *
3451      *                   node B
3452      *                     |
3453      *                     v
3454      *   guest device -> node A -> further backing chain...
3455      *
3456      * Finally we want to replace A by B.  When doing that, we want to
3457      * replace all pointers to A by pointers to B -- except for the
3458      * pointer from B because (1) that would create a loop, and (2)
3459      * that pointer should simply stay intact:
3460      *
3461      *   guest device -> node B
3462      *                     |
3463      *                     v
3464      *                   node A -> further backing chain...
3465      *
3466      * In general, when replacing a node A (c->bs) by a node B (@to),
3467      * if A is a child of B, that means we cannot replace A by B there
3468      * because that would create a loop.  Silently detaching A from B
3469      * is also not really an option.  So overall just leaving A in
3470      * place there is the most sensible choice. */
3471     QLIST_FOREACH(to_c, &to->children, next) {
3472         if (to_c == c) {
3473             return false;
3474         }
3475     }
3476 
3477     return true;
3478 }
3479 
3480 void bdrv_replace_node(BlockDriverState *from, BlockDriverState *to,
3481                        Error **errp)
3482 {
3483     BdrvChild *c, *next;
3484     GSList *list = NULL, *p;
3485     uint64_t old_perm, old_shared;
3486     uint64_t perm = 0, shared = BLK_PERM_ALL;
3487     int ret;
3488 
3489     assert(!atomic_read(&from->in_flight));
3490     assert(!atomic_read(&to->in_flight));
3491 
3492     /* Make sure that @from doesn't go away until we have successfully attached
3493      * all of its parents to @to. */
3494     bdrv_ref(from);
3495 
3496     /* Put all parents into @list and calculate their cumulative permissions */
3497     QLIST_FOREACH_SAFE(c, &from->parents, next_parent, next) {
3498         assert(c->bs == from);
3499         if (!should_update_child(c, to)) {
3500             continue;
3501         }
3502         list = g_slist_prepend(list, c);
3503         perm |= c->perm;
3504         shared &= c->shared_perm;
3505     }
3506 
3507     /* Check whether the required permissions can be granted on @to, ignoring
3508      * all BdrvChild in @list so that they can't block themselves. */
3509     ret = bdrv_check_update_perm(to, NULL, perm, shared, list, errp);
3510     if (ret < 0) {
3511         bdrv_abort_perm_update(to);
3512         goto out;
3513     }
3514 
3515     /* Now actually perform the change. We performed the permission check for
3516      * all elements of @list at once, so set the permissions all at once at the
3517      * very end. */
3518     for (p = list; p != NULL; p = p->next) {
3519         c = p->data;
3520 
3521         bdrv_ref(to);
3522         bdrv_replace_child_noperm(c, to);
3523         bdrv_unref(from);
3524     }
3525 
3526     bdrv_get_cumulative_perm(to, &old_perm, &old_shared);
3527     bdrv_set_perm(to, old_perm | perm, old_shared | shared);
3528 
3529 out:
3530     g_slist_free(list);
3531     bdrv_unref(from);
3532 }
3533 
3534 /*
3535  * Add new bs contents at the top of an image chain while the chain is
3536  * live, while keeping required fields on the top layer.
3537  *
3538  * This will modify the BlockDriverState fields, and swap contents
3539  * between bs_new and bs_top. Both bs_new and bs_top are modified.
3540  *
3541  * bs_new must not be attached to a BlockBackend.
3542  *
3543  * This function does not create any image files.
3544  *
3545  * bdrv_append() takes ownership of a bs_new reference and unrefs it because
3546  * that's what the callers commonly need. bs_new will be referenced by the old
3547  * parents of bs_top after bdrv_append() returns. If the caller needs to keep a
3548  * reference of its own, it must call bdrv_ref().
3549  */
3550 void bdrv_append(BlockDriverState *bs_new, BlockDriverState *bs_top,
3551                  Error **errp)
3552 {
3553     Error *local_err = NULL;
3554 
3555     bdrv_set_backing_hd(bs_new, bs_top, &local_err);
3556     if (local_err) {
3557         error_propagate(errp, local_err);
3558         goto out;
3559     }
3560 
3561     bdrv_replace_node(bs_top, bs_new, &local_err);
3562     if (local_err) {
3563         error_propagate(errp, local_err);
3564         bdrv_set_backing_hd(bs_new, NULL, &error_abort);
3565         goto out;
3566     }
3567 
3568     /* bs_new is now referenced by its new parents, we don't need the
3569      * additional reference any more. */
3570 out:
3571     bdrv_unref(bs_new);
3572 }
3573 
3574 static void bdrv_delete(BlockDriverState *bs)
3575 {
3576     assert(!bs->job);
3577     assert(bdrv_op_blocker_is_empty(bs));
3578     assert(!bs->refcnt);
3579 
3580     bdrv_close(bs);
3581 
3582     /* remove from list, if necessary */
3583     if (bs->node_name[0] != '\0') {
3584         QTAILQ_REMOVE(&graph_bdrv_states, bs, node_list);
3585     }
3586     QTAILQ_REMOVE(&all_bdrv_states, bs, bs_list);
3587 
3588     g_free(bs);
3589 }
3590 
3591 /*
3592  * Run consistency checks on an image
3593  *
3594  * Returns 0 if the check could be completed (it doesn't mean that the image is
3595  * free of errors) or -errno when an internal error occurred. The results of the
3596  * check are stored in res.
3597  */
3598 static int coroutine_fn bdrv_co_check(BlockDriverState *bs,
3599                                       BdrvCheckResult *res, BdrvCheckMode fix)
3600 {
3601     if (bs->drv == NULL) {
3602         return -ENOMEDIUM;
3603     }
3604     if (bs->drv->bdrv_co_check == NULL) {
3605         return -ENOTSUP;
3606     }
3607 
3608     memset(res, 0, sizeof(*res));
3609     return bs->drv->bdrv_co_check(bs, res, fix);
3610 }
3611 
3612 typedef struct CheckCo {
3613     BlockDriverState *bs;
3614     BdrvCheckResult *res;
3615     BdrvCheckMode fix;
3616     int ret;
3617 } CheckCo;
3618 
3619 static void bdrv_check_co_entry(void *opaque)
3620 {
3621     CheckCo *cco = opaque;
3622     cco->ret = bdrv_co_check(cco->bs, cco->res, cco->fix);
3623 }
3624 
3625 int bdrv_check(BlockDriverState *bs,
3626                BdrvCheckResult *res, BdrvCheckMode fix)
3627 {
3628     Coroutine *co;
3629     CheckCo cco = {
3630         .bs = bs,
3631         .res = res,
3632         .ret = -EINPROGRESS,
3633         .fix = fix,
3634     };
3635 
3636     if (qemu_in_coroutine()) {
3637         /* Fast-path if already in coroutine context */
3638         bdrv_check_co_entry(&cco);
3639     } else {
3640         co = qemu_coroutine_create(bdrv_check_co_entry, &cco);
3641         qemu_coroutine_enter(co);
3642         BDRV_POLL_WHILE(bs, cco.ret == -EINPROGRESS);
3643     }
3644 
3645     return cco.ret;
3646 }
3647 
3648 /*
3649  * Return values:
3650  * 0        - success
3651  * -EINVAL  - backing format specified, but no file
3652  * -ENOSPC  - can't update the backing file because no space is left in the
3653  *            image file header
3654  * -ENOTSUP - format driver doesn't support changing the backing file
3655  */
3656 int bdrv_change_backing_file(BlockDriverState *bs,
3657     const char *backing_file, const char *backing_fmt)
3658 {
3659     BlockDriver *drv = bs->drv;
3660     int ret;
3661 
3662     if (!drv) {
3663         return -ENOMEDIUM;
3664     }
3665 
3666     /* Backing file format doesn't make sense without a backing file */
3667     if (backing_fmt && !backing_file) {
3668         return -EINVAL;
3669     }
3670 
3671     if (drv->bdrv_change_backing_file != NULL) {
3672         ret = drv->bdrv_change_backing_file(bs, backing_file, backing_fmt);
3673     } else {
3674         ret = -ENOTSUP;
3675     }
3676 
3677     if (ret == 0) {
3678         pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_file ?: "");
3679         pstrcpy(bs->backing_format, sizeof(bs->backing_format), backing_fmt ?: "");
3680     }
3681     return ret;
3682 }
3683 
3684 /*
3685  * Finds the image layer in the chain that has 'bs' as its backing file.
3686  *
3687  * active is the current topmost image.
3688  *
3689  * Returns NULL if bs is not found in active's image chain,
3690  * or if active == bs.
3691  *
3692  * Returns the bottommost base image if bs == NULL.
3693  */
3694 BlockDriverState *bdrv_find_overlay(BlockDriverState *active,
3695                                     BlockDriverState *bs)
3696 {
3697     while (active && bs != backing_bs(active)) {
3698         active = backing_bs(active);
3699     }
3700 
3701     return active;
3702 }
3703 
3704 /* Given a BDS, searches for the base layer. */
3705 BlockDriverState *bdrv_find_base(BlockDriverState *bs)
3706 {
3707     return bdrv_find_overlay(bs, NULL);
3708 }
3709 
3710 /*
3711  * Drops images above 'base' up to and including 'top', and sets the image
3712  * above 'top' to have base as its backing file.
3713  *
3714  * Requires that the overlay to 'top' is opened r/w, so that the backing file
3715  * information in 'bs' can be properly updated.
3716  *
3717  * E.g., this will convert the following chain:
3718  * bottom <- base <- intermediate <- top <- active
3719  *
3720  * to
3721  *
3722  * bottom <- base <- active
3723  *
3724  * It is allowed for bottom==base, in which case it converts:
3725  *
3726  * base <- intermediate <- top <- active
3727  *
3728  * to
3729  *
3730  * base <- active
3731  *
3732  * If backing_file_str is non-NULL, it will be used when modifying top's
3733  * overlay image metadata.
3734  *
3735  * Error conditions:
3736  *  if active == top, that is considered an error
3737  *
3738  */
3739 int bdrv_drop_intermediate(BlockDriverState *top, BlockDriverState *base,
3740                            const char *backing_file_str)
3741 {
3742     BdrvChild *c, *next;
3743     Error *local_err = NULL;
3744     int ret = -EIO;
3745 
3746     bdrv_ref(top);
3747 
3748     if (!top->drv || !base->drv) {
3749         goto exit;
3750     }
3751 
3752     /* Make sure that base is in the backing chain of top */
3753     if (!bdrv_chain_contains(top, base)) {
3754         goto exit;
3755     }
3756 
3757     /* success - we can delete the intermediate states, and link top->base */
3758     /* TODO Check graph modification op blockers (BLK_PERM_GRAPH_MOD) once
3759      * we've figured out how they should work. */
3760     backing_file_str = backing_file_str ? backing_file_str : base->filename;
3761 
3762     QLIST_FOREACH_SAFE(c, &top->parents, next_parent, next) {
3763         /* Check whether we are allowed to switch c from top to base */
3764         GSList *ignore_children = g_slist_prepend(NULL, c);
3765         bdrv_check_update_perm(base, NULL, c->perm, c->shared_perm,
3766                                ignore_children, &local_err);
3767         g_slist_free(ignore_children);
3768         if (local_err) {
3769             ret = -EPERM;
3770             error_report_err(local_err);
3771             goto exit;
3772         }
3773 
3774         /* If so, update the backing file path in the image file */
3775         if (c->role->update_filename) {
3776             ret = c->role->update_filename(c, base, backing_file_str,
3777                                            &local_err);
3778             if (ret < 0) {
3779                 bdrv_abort_perm_update(base);
3780                 error_report_err(local_err);
3781                 goto exit;
3782             }
3783         }
3784 
3785         /* Do the actual switch in the in-memory graph.
3786          * Completes bdrv_check_update_perm() transaction internally. */
3787         bdrv_ref(base);
3788         bdrv_replace_child(c, base);
3789         bdrv_unref(top);
3790     }
3791 
3792     ret = 0;
3793 exit:
3794     bdrv_unref(top);
3795     return ret;
3796 }
3797 
3798 /**
3799  * Length of a allocated file in bytes. Sparse files are counted by actual
3800  * allocated space. Return < 0 if error or unknown.
3801  */
3802 int64_t bdrv_get_allocated_file_size(BlockDriverState *bs)
3803 {
3804     BlockDriver *drv = bs->drv;
3805     if (!drv) {
3806         return -ENOMEDIUM;
3807     }
3808     if (drv->bdrv_get_allocated_file_size) {
3809         return drv->bdrv_get_allocated_file_size(bs);
3810     }
3811     if (bs->file) {
3812         return bdrv_get_allocated_file_size(bs->file->bs);
3813     }
3814     return -ENOTSUP;
3815 }
3816 
3817 /*
3818  * bdrv_measure:
3819  * @drv: Format driver
3820  * @opts: Creation options for new image
3821  * @in_bs: Existing image containing data for new image (may be NULL)
3822  * @errp: Error object
3823  * Returns: A #BlockMeasureInfo (free using qapi_free_BlockMeasureInfo())
3824  *          or NULL on error
3825  *
3826  * Calculate file size required to create a new image.
3827  *
3828  * If @in_bs is given then space for allocated clusters and zero clusters
3829  * from that image are included in the calculation.  If @opts contains a
3830  * backing file that is shared by @in_bs then backing clusters may be omitted
3831  * from the calculation.
3832  *
3833  * If @in_bs is NULL then the calculation includes no allocated clusters
3834  * unless a preallocation option is given in @opts.
3835  *
3836  * Note that @in_bs may use a different BlockDriver from @drv.
3837  *
3838  * If an error occurs the @errp pointer is set.
3839  */
3840 BlockMeasureInfo *bdrv_measure(BlockDriver *drv, QemuOpts *opts,
3841                                BlockDriverState *in_bs, Error **errp)
3842 {
3843     if (!drv->bdrv_measure) {
3844         error_setg(errp, "Block driver '%s' does not support size measurement",
3845                    drv->format_name);
3846         return NULL;
3847     }
3848 
3849     return drv->bdrv_measure(opts, in_bs, errp);
3850 }
3851 
3852 /**
3853  * Return number of sectors on success, -errno on error.
3854  */
3855 int64_t bdrv_nb_sectors(BlockDriverState *bs)
3856 {
3857     BlockDriver *drv = bs->drv;
3858 
3859     if (!drv)
3860         return -ENOMEDIUM;
3861 
3862     if (drv->has_variable_length) {
3863         int ret = refresh_total_sectors(bs, bs->total_sectors);
3864         if (ret < 0) {
3865             return ret;
3866         }
3867     }
3868     return bs->total_sectors;
3869 }
3870 
3871 /**
3872  * Return length in bytes on success, -errno on error.
3873  * The length is always a multiple of BDRV_SECTOR_SIZE.
3874  */
3875 int64_t bdrv_getlength(BlockDriverState *bs)
3876 {
3877     int64_t ret = bdrv_nb_sectors(bs);
3878 
3879     ret = ret > INT64_MAX / BDRV_SECTOR_SIZE ? -EFBIG : ret;
3880     return ret < 0 ? ret : ret * BDRV_SECTOR_SIZE;
3881 }
3882 
3883 /* return 0 as number of sectors if no device present or error */
3884 void bdrv_get_geometry(BlockDriverState *bs, uint64_t *nb_sectors_ptr)
3885 {
3886     int64_t nb_sectors = bdrv_nb_sectors(bs);
3887 
3888     *nb_sectors_ptr = nb_sectors < 0 ? 0 : nb_sectors;
3889 }
3890 
3891 bool bdrv_is_sg(BlockDriverState *bs)
3892 {
3893     return bs->sg;
3894 }
3895 
3896 bool bdrv_is_encrypted(BlockDriverState *bs)
3897 {
3898     if (bs->backing && bs->backing->bs->encrypted) {
3899         return true;
3900     }
3901     return bs->encrypted;
3902 }
3903 
3904 const char *bdrv_get_format_name(BlockDriverState *bs)
3905 {
3906     return bs->drv ? bs->drv->format_name : NULL;
3907 }
3908 
3909 static int qsort_strcmp(const void *a, const void *b)
3910 {
3911     return strcmp(*(char *const *)a, *(char *const *)b);
3912 }
3913 
3914 void bdrv_iterate_format(void (*it)(void *opaque, const char *name),
3915                          void *opaque)
3916 {
3917     BlockDriver *drv;
3918     int count = 0;
3919     int i;
3920     const char **formats = NULL;
3921 
3922     QLIST_FOREACH(drv, &bdrv_drivers, list) {
3923         if (drv->format_name) {
3924             bool found = false;
3925             int i = count;
3926             while (formats && i && !found) {
3927                 found = !strcmp(formats[--i], drv->format_name);
3928             }
3929 
3930             if (!found) {
3931                 formats = g_renew(const char *, formats, count + 1);
3932                 formats[count++] = drv->format_name;
3933             }
3934         }
3935     }
3936 
3937     for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); i++) {
3938         const char *format_name = block_driver_modules[i].format_name;
3939 
3940         if (format_name) {
3941             bool found = false;
3942             int j = count;
3943 
3944             while (formats && j && !found) {
3945                 found = !strcmp(formats[--j], format_name);
3946             }
3947 
3948             if (!found) {
3949                 formats = g_renew(const char *, formats, count + 1);
3950                 formats[count++] = format_name;
3951             }
3952         }
3953     }
3954 
3955     qsort(formats, count, sizeof(formats[0]), qsort_strcmp);
3956 
3957     for (i = 0; i < count; i++) {
3958         it(opaque, formats[i]);
3959     }
3960 
3961     g_free(formats);
3962 }
3963 
3964 /* This function is to find a node in the bs graph */
3965 BlockDriverState *bdrv_find_node(const char *node_name)
3966 {
3967     BlockDriverState *bs;
3968 
3969     assert(node_name);
3970 
3971     QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
3972         if (!strcmp(node_name, bs->node_name)) {
3973             return bs;
3974         }
3975     }
3976     return NULL;
3977 }
3978 
3979 /* Put this QMP function here so it can access the static graph_bdrv_states. */
3980 BlockDeviceInfoList *bdrv_named_nodes_list(Error **errp)
3981 {
3982     BlockDeviceInfoList *list, *entry;
3983     BlockDriverState *bs;
3984 
3985     list = NULL;
3986     QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
3987         BlockDeviceInfo *info = bdrv_block_device_info(NULL, bs, errp);
3988         if (!info) {
3989             qapi_free_BlockDeviceInfoList(list);
3990             return NULL;
3991         }
3992         entry = g_malloc0(sizeof(*entry));
3993         entry->value = info;
3994         entry->next = list;
3995         list = entry;
3996     }
3997 
3998     return list;
3999 }
4000 
4001 BlockDriverState *bdrv_lookup_bs(const char *device,
4002                                  const char *node_name,
4003                                  Error **errp)
4004 {
4005     BlockBackend *blk;
4006     BlockDriverState *bs;
4007 
4008     if (device) {
4009         blk = blk_by_name(device);
4010 
4011         if (blk) {
4012             bs = blk_bs(blk);
4013             if (!bs) {
4014                 error_setg(errp, "Device '%s' has no medium", device);
4015             }
4016 
4017             return bs;
4018         }
4019     }
4020 
4021     if (node_name) {
4022         bs = bdrv_find_node(node_name);
4023 
4024         if (bs) {
4025             return bs;
4026         }
4027     }
4028 
4029     error_setg(errp, "Cannot find device=%s nor node_name=%s",
4030                      device ? device : "",
4031                      node_name ? node_name : "");
4032     return NULL;
4033 }
4034 
4035 /* If 'base' is in the same chain as 'top', return true. Otherwise,
4036  * return false.  If either argument is NULL, return false. */
4037 bool bdrv_chain_contains(BlockDriverState *top, BlockDriverState *base)
4038 {
4039     while (top && top != base) {
4040         top = backing_bs(top);
4041     }
4042 
4043     return top != NULL;
4044 }
4045 
4046 BlockDriverState *bdrv_next_node(BlockDriverState *bs)
4047 {
4048     if (!bs) {
4049         return QTAILQ_FIRST(&graph_bdrv_states);
4050     }
4051     return QTAILQ_NEXT(bs, node_list);
4052 }
4053 
4054 BlockDriverState *bdrv_next_all_states(BlockDriverState *bs)
4055 {
4056     if (!bs) {
4057         return QTAILQ_FIRST(&all_bdrv_states);
4058     }
4059     return QTAILQ_NEXT(bs, bs_list);
4060 }
4061 
4062 const char *bdrv_get_node_name(const BlockDriverState *bs)
4063 {
4064     return bs->node_name;
4065 }
4066 
4067 const char *bdrv_get_parent_name(const BlockDriverState *bs)
4068 {
4069     BdrvChild *c;
4070     const char *name;
4071 
4072     /* If multiple parents have a name, just pick the first one. */
4073     QLIST_FOREACH(c, &bs->parents, next_parent) {
4074         if (c->role->get_name) {
4075             name = c->role->get_name(c);
4076             if (name && *name) {
4077                 return name;
4078             }
4079         }
4080     }
4081 
4082     return NULL;
4083 }
4084 
4085 /* TODO check what callers really want: bs->node_name or blk_name() */
4086 const char *bdrv_get_device_name(const BlockDriverState *bs)
4087 {
4088     return bdrv_get_parent_name(bs) ?: "";
4089 }
4090 
4091 /* This can be used to identify nodes that might not have a device
4092  * name associated. Since node and device names live in the same
4093  * namespace, the result is unambiguous. The exception is if both are
4094  * absent, then this returns an empty (non-null) string. */
4095 const char *bdrv_get_device_or_node_name(const BlockDriverState *bs)
4096 {
4097     return bdrv_get_parent_name(bs) ?: bs->node_name;
4098 }
4099 
4100 int bdrv_get_flags(BlockDriverState *bs)
4101 {
4102     return bs->open_flags;
4103 }
4104 
4105 int bdrv_has_zero_init_1(BlockDriverState *bs)
4106 {
4107     return 1;
4108 }
4109 
4110 int bdrv_has_zero_init(BlockDriverState *bs)
4111 {
4112     if (!bs->drv) {
4113         return 0;
4114     }
4115 
4116     /* If BS is a copy on write image, it is initialized to
4117        the contents of the base image, which may not be zeroes.  */
4118     if (bs->backing) {
4119         return 0;
4120     }
4121     if (bs->drv->bdrv_has_zero_init) {
4122         return bs->drv->bdrv_has_zero_init(bs);
4123     }
4124     if (bs->file && bs->drv->is_filter) {
4125         return bdrv_has_zero_init(bs->file->bs);
4126     }
4127 
4128     /* safe default */
4129     return 0;
4130 }
4131 
4132 bool bdrv_unallocated_blocks_are_zero(BlockDriverState *bs)
4133 {
4134     BlockDriverInfo bdi;
4135 
4136     if (bs->backing) {
4137         return false;
4138     }
4139 
4140     if (bdrv_get_info(bs, &bdi) == 0) {
4141         return bdi.unallocated_blocks_are_zero;
4142     }
4143 
4144     return false;
4145 }
4146 
4147 bool bdrv_can_write_zeroes_with_unmap(BlockDriverState *bs)
4148 {
4149     if (!(bs->open_flags & BDRV_O_UNMAP)) {
4150         return false;
4151     }
4152 
4153     return bs->supported_zero_flags & BDRV_REQ_MAY_UNMAP;
4154 }
4155 
4156 const char *bdrv_get_encrypted_filename(BlockDriverState *bs)
4157 {
4158     if (bs->backing && bs->backing->bs->encrypted)
4159         return bs->backing_file;
4160     else if (bs->encrypted)
4161         return bs->filename;
4162     else
4163         return NULL;
4164 }
4165 
4166 void bdrv_get_backing_filename(BlockDriverState *bs,
4167                                char *filename, int filename_size)
4168 {
4169     pstrcpy(filename, filename_size, bs->backing_file);
4170 }
4171 
4172 int bdrv_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
4173 {
4174     BlockDriver *drv = bs->drv;
4175     /* if bs->drv == NULL, bs is closed, so there's nothing to do here */
4176     if (!drv) {
4177         return -ENOMEDIUM;
4178     }
4179     if (!drv->bdrv_get_info) {
4180         if (bs->file && drv->is_filter) {
4181             return bdrv_get_info(bs->file->bs, bdi);
4182         }
4183         return -ENOTSUP;
4184     }
4185     memset(bdi, 0, sizeof(*bdi));
4186     return drv->bdrv_get_info(bs, bdi);
4187 }
4188 
4189 ImageInfoSpecific *bdrv_get_specific_info(BlockDriverState *bs)
4190 {
4191     BlockDriver *drv = bs->drv;
4192     if (drv && drv->bdrv_get_specific_info) {
4193         return drv->bdrv_get_specific_info(bs);
4194     }
4195     return NULL;
4196 }
4197 
4198 void bdrv_debug_event(BlockDriverState *bs, BlkdebugEvent event)
4199 {
4200     if (!bs || !bs->drv || !bs->drv->bdrv_debug_event) {
4201         return;
4202     }
4203 
4204     bs->drv->bdrv_debug_event(bs, event);
4205 }
4206 
4207 int bdrv_debug_breakpoint(BlockDriverState *bs, const char *event,
4208                           const char *tag)
4209 {
4210     while (bs && bs->drv && !bs->drv->bdrv_debug_breakpoint) {
4211         bs = bs->file ? bs->file->bs : NULL;
4212     }
4213 
4214     if (bs && bs->drv && bs->drv->bdrv_debug_breakpoint) {
4215         return bs->drv->bdrv_debug_breakpoint(bs, event, tag);
4216     }
4217 
4218     return -ENOTSUP;
4219 }
4220 
4221 int bdrv_debug_remove_breakpoint(BlockDriverState *bs, const char *tag)
4222 {
4223     while (bs && bs->drv && !bs->drv->bdrv_debug_remove_breakpoint) {
4224         bs = bs->file ? bs->file->bs : NULL;
4225     }
4226 
4227     if (bs && bs->drv && bs->drv->bdrv_debug_remove_breakpoint) {
4228         return bs->drv->bdrv_debug_remove_breakpoint(bs, tag);
4229     }
4230 
4231     return -ENOTSUP;
4232 }
4233 
4234 int bdrv_debug_resume(BlockDriverState *bs, const char *tag)
4235 {
4236     while (bs && (!bs->drv || !bs->drv->bdrv_debug_resume)) {
4237         bs = bs->file ? bs->file->bs : NULL;
4238     }
4239 
4240     if (bs && bs->drv && bs->drv->bdrv_debug_resume) {
4241         return bs->drv->bdrv_debug_resume(bs, tag);
4242     }
4243 
4244     return -ENOTSUP;
4245 }
4246 
4247 bool bdrv_debug_is_suspended(BlockDriverState *bs, const char *tag)
4248 {
4249     while (bs && bs->drv && !bs->drv->bdrv_debug_is_suspended) {
4250         bs = bs->file ? bs->file->bs : NULL;
4251     }
4252 
4253     if (bs && bs->drv && bs->drv->bdrv_debug_is_suspended) {
4254         return bs->drv->bdrv_debug_is_suspended(bs, tag);
4255     }
4256 
4257     return false;
4258 }
4259 
4260 /* backing_file can either be relative, or absolute, or a protocol.  If it is
4261  * relative, it must be relative to the chain.  So, passing in bs->filename
4262  * from a BDS as backing_file should not be done, as that may be relative to
4263  * the CWD rather than the chain. */
4264 BlockDriverState *bdrv_find_backing_image(BlockDriverState *bs,
4265         const char *backing_file)
4266 {
4267     char *filename_full = NULL;
4268     char *backing_file_full = NULL;
4269     char *filename_tmp = NULL;
4270     int is_protocol = 0;
4271     BlockDriverState *curr_bs = NULL;
4272     BlockDriverState *retval = NULL;
4273     Error *local_error = NULL;
4274 
4275     if (!bs || !bs->drv || !backing_file) {
4276         return NULL;
4277     }
4278 
4279     filename_full     = g_malloc(PATH_MAX);
4280     backing_file_full = g_malloc(PATH_MAX);
4281     filename_tmp      = g_malloc(PATH_MAX);
4282 
4283     is_protocol = path_has_protocol(backing_file);
4284 
4285     for (curr_bs = bs; curr_bs->backing; curr_bs = curr_bs->backing->bs) {
4286 
4287         /* If either of the filename paths is actually a protocol, then
4288          * compare unmodified paths; otherwise make paths relative */
4289         if (is_protocol || path_has_protocol(curr_bs->backing_file)) {
4290             if (strcmp(backing_file, curr_bs->backing_file) == 0) {
4291                 retval = curr_bs->backing->bs;
4292                 break;
4293             }
4294             /* Also check against the full backing filename for the image */
4295             bdrv_get_full_backing_filename(curr_bs, backing_file_full, PATH_MAX,
4296                                            &local_error);
4297             if (local_error == NULL) {
4298                 if (strcmp(backing_file, backing_file_full) == 0) {
4299                     retval = curr_bs->backing->bs;
4300                     break;
4301                 }
4302             } else {
4303                 error_free(local_error);
4304                 local_error = NULL;
4305             }
4306         } else {
4307             /* If not an absolute filename path, make it relative to the current
4308              * image's filename path */
4309             path_combine(filename_tmp, PATH_MAX, curr_bs->filename,
4310                          backing_file);
4311 
4312             /* We are going to compare absolute pathnames */
4313             if (!realpath(filename_tmp, filename_full)) {
4314                 continue;
4315             }
4316 
4317             /* We need to make sure the backing filename we are comparing against
4318              * is relative to the current image filename (or absolute) */
4319             path_combine(filename_tmp, PATH_MAX, curr_bs->filename,
4320                          curr_bs->backing_file);
4321 
4322             if (!realpath(filename_tmp, backing_file_full)) {
4323                 continue;
4324             }
4325 
4326             if (strcmp(backing_file_full, filename_full) == 0) {
4327                 retval = curr_bs->backing->bs;
4328                 break;
4329             }
4330         }
4331     }
4332 
4333     g_free(filename_full);
4334     g_free(backing_file_full);
4335     g_free(filename_tmp);
4336     return retval;
4337 }
4338 
4339 void bdrv_init(void)
4340 {
4341     module_call_init(MODULE_INIT_BLOCK);
4342 }
4343 
4344 void bdrv_init_with_whitelist(void)
4345 {
4346     use_bdrv_whitelist = 1;
4347     bdrv_init();
4348 }
4349 
4350 static void coroutine_fn bdrv_co_invalidate_cache(BlockDriverState *bs,
4351                                                   Error **errp)
4352 {
4353     BdrvChild *child, *parent;
4354     uint64_t perm, shared_perm;
4355     Error *local_err = NULL;
4356     int ret;
4357 
4358     if (!bs->drv)  {
4359         return;
4360     }
4361 
4362     if (!(bs->open_flags & BDRV_O_INACTIVE)) {
4363         return;
4364     }
4365 
4366     QLIST_FOREACH(child, &bs->children, next) {
4367         bdrv_co_invalidate_cache(child->bs, &local_err);
4368         if (local_err) {
4369             error_propagate(errp, local_err);
4370             return;
4371         }
4372     }
4373 
4374     /*
4375      * Update permissions, they may differ for inactive nodes.
4376      *
4377      * Note that the required permissions of inactive images are always a
4378      * subset of the permissions required after activating the image. This
4379      * allows us to just get the permissions upfront without restricting
4380      * drv->bdrv_invalidate_cache().
4381      *
4382      * It also means that in error cases, we don't have to try and revert to
4383      * the old permissions (which is an operation that could fail, too). We can
4384      * just keep the extended permissions for the next time that an activation
4385      * of the image is tried.
4386      */
4387     bs->open_flags &= ~BDRV_O_INACTIVE;
4388     bdrv_get_cumulative_perm(bs, &perm, &shared_perm);
4389     ret = bdrv_check_perm(bs, NULL, perm, shared_perm, NULL, &local_err);
4390     if (ret < 0) {
4391         bs->open_flags |= BDRV_O_INACTIVE;
4392         error_propagate(errp, local_err);
4393         return;
4394     }
4395     bdrv_set_perm(bs, perm, shared_perm);
4396 
4397     if (bs->drv->bdrv_co_invalidate_cache) {
4398         bs->drv->bdrv_co_invalidate_cache(bs, &local_err);
4399         if (local_err) {
4400             bs->open_flags |= BDRV_O_INACTIVE;
4401             error_propagate(errp, local_err);
4402             return;
4403         }
4404     }
4405 
4406     ret = refresh_total_sectors(bs, bs->total_sectors);
4407     if (ret < 0) {
4408         bs->open_flags |= BDRV_O_INACTIVE;
4409         error_setg_errno(errp, -ret, "Could not refresh total sector count");
4410         return;
4411     }
4412 
4413     QLIST_FOREACH(parent, &bs->parents, next_parent) {
4414         if (parent->role->activate) {
4415             parent->role->activate(parent, &local_err);
4416             if (local_err) {
4417                 error_propagate(errp, local_err);
4418                 return;
4419             }
4420         }
4421     }
4422 }
4423 
4424 typedef struct InvalidateCacheCo {
4425     BlockDriverState *bs;
4426     Error **errp;
4427     bool done;
4428 } InvalidateCacheCo;
4429 
4430 static void coroutine_fn bdrv_invalidate_cache_co_entry(void *opaque)
4431 {
4432     InvalidateCacheCo *ico = opaque;
4433     bdrv_co_invalidate_cache(ico->bs, ico->errp);
4434     ico->done = true;
4435 }
4436 
4437 void bdrv_invalidate_cache(BlockDriverState *bs, Error **errp)
4438 {
4439     Coroutine *co;
4440     InvalidateCacheCo ico = {
4441         .bs = bs,
4442         .done = false,
4443         .errp = errp
4444     };
4445 
4446     if (qemu_in_coroutine()) {
4447         /* Fast-path if already in coroutine context */
4448         bdrv_invalidate_cache_co_entry(&ico);
4449     } else {
4450         co = qemu_coroutine_create(bdrv_invalidate_cache_co_entry, &ico);
4451         qemu_coroutine_enter(co);
4452         BDRV_POLL_WHILE(bs, !ico.done);
4453     }
4454 }
4455 
4456 void bdrv_invalidate_cache_all(Error **errp)
4457 {
4458     BlockDriverState *bs;
4459     Error *local_err = NULL;
4460     BdrvNextIterator it;
4461 
4462     for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
4463         AioContext *aio_context = bdrv_get_aio_context(bs);
4464 
4465         aio_context_acquire(aio_context);
4466         bdrv_invalidate_cache(bs, &local_err);
4467         aio_context_release(aio_context);
4468         if (local_err) {
4469             error_propagate(errp, local_err);
4470             bdrv_next_cleanup(&it);
4471             return;
4472         }
4473     }
4474 }
4475 
4476 static int bdrv_inactivate_recurse(BlockDriverState *bs,
4477                                    bool setting_flag)
4478 {
4479     BdrvChild *child, *parent;
4480     int ret;
4481 
4482     if (!bs->drv) {
4483         return -ENOMEDIUM;
4484     }
4485 
4486     if (!setting_flag && bs->drv->bdrv_inactivate) {
4487         ret = bs->drv->bdrv_inactivate(bs);
4488         if (ret < 0) {
4489             return ret;
4490         }
4491     }
4492 
4493     if (setting_flag && !(bs->open_flags & BDRV_O_INACTIVE)) {
4494         uint64_t perm, shared_perm;
4495 
4496         QLIST_FOREACH(parent, &bs->parents, next_parent) {
4497             if (parent->role->inactivate) {
4498                 ret = parent->role->inactivate(parent);
4499                 if (ret < 0) {
4500                     return ret;
4501                 }
4502             }
4503         }
4504 
4505         bs->open_flags |= BDRV_O_INACTIVE;
4506 
4507         /* Update permissions, they may differ for inactive nodes */
4508         bdrv_get_cumulative_perm(bs, &perm, &shared_perm);
4509         bdrv_check_perm(bs, NULL, perm, shared_perm, NULL, &error_abort);
4510         bdrv_set_perm(bs, perm, shared_perm);
4511     }
4512 
4513     QLIST_FOREACH(child, &bs->children, next) {
4514         ret = bdrv_inactivate_recurse(child->bs, setting_flag);
4515         if (ret < 0) {
4516             return ret;
4517         }
4518     }
4519 
4520     /* At this point persistent bitmaps should be already stored by the format
4521      * driver */
4522     bdrv_release_persistent_dirty_bitmaps(bs);
4523 
4524     return 0;
4525 }
4526 
4527 int bdrv_inactivate_all(void)
4528 {
4529     BlockDriverState *bs = NULL;
4530     BdrvNextIterator it;
4531     int ret = 0;
4532     int pass;
4533     GSList *aio_ctxs = NULL, *ctx;
4534 
4535     for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
4536         AioContext *aio_context = bdrv_get_aio_context(bs);
4537 
4538         if (!g_slist_find(aio_ctxs, aio_context)) {
4539             aio_ctxs = g_slist_prepend(aio_ctxs, aio_context);
4540             aio_context_acquire(aio_context);
4541         }
4542     }
4543 
4544     /* We do two passes of inactivation. The first pass calls to drivers'
4545      * .bdrv_inactivate callbacks recursively so all cache is flushed to disk;
4546      * the second pass sets the BDRV_O_INACTIVE flag so that no further write
4547      * is allowed. */
4548     for (pass = 0; pass < 2; pass++) {
4549         for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
4550             ret = bdrv_inactivate_recurse(bs, pass);
4551             if (ret < 0) {
4552                 bdrv_next_cleanup(&it);
4553                 goto out;
4554             }
4555         }
4556     }
4557 
4558 out:
4559     for (ctx = aio_ctxs; ctx != NULL; ctx = ctx->next) {
4560         AioContext *aio_context = ctx->data;
4561         aio_context_release(aio_context);
4562     }
4563     g_slist_free(aio_ctxs);
4564 
4565     return ret;
4566 }
4567 
4568 /**************************************************************/
4569 /* removable device support */
4570 
4571 /**
4572  * Return TRUE if the media is present
4573  */
4574 bool bdrv_is_inserted(BlockDriverState *bs)
4575 {
4576     BlockDriver *drv = bs->drv;
4577     BdrvChild *child;
4578 
4579     if (!drv) {
4580         return false;
4581     }
4582     if (drv->bdrv_is_inserted) {
4583         return drv->bdrv_is_inserted(bs);
4584     }
4585     QLIST_FOREACH(child, &bs->children, next) {
4586         if (!bdrv_is_inserted(child->bs)) {
4587             return false;
4588         }
4589     }
4590     return true;
4591 }
4592 
4593 /**
4594  * If eject_flag is TRUE, eject the media. Otherwise, close the tray
4595  */
4596 void bdrv_eject(BlockDriverState *bs, bool eject_flag)
4597 {
4598     BlockDriver *drv = bs->drv;
4599 
4600     if (drv && drv->bdrv_eject) {
4601         drv->bdrv_eject(bs, eject_flag);
4602     }
4603 }
4604 
4605 /**
4606  * Lock or unlock the media (if it is locked, the user won't be able
4607  * to eject it manually).
4608  */
4609 void bdrv_lock_medium(BlockDriverState *bs, bool locked)
4610 {
4611     BlockDriver *drv = bs->drv;
4612 
4613     trace_bdrv_lock_medium(bs, locked);
4614 
4615     if (drv && drv->bdrv_lock_medium) {
4616         drv->bdrv_lock_medium(bs, locked);
4617     }
4618 }
4619 
4620 /* Get a reference to bs */
4621 void bdrv_ref(BlockDriverState *bs)
4622 {
4623     bs->refcnt++;
4624 }
4625 
4626 /* Release a previously grabbed reference to bs.
4627  * If after releasing, reference count is zero, the BlockDriverState is
4628  * deleted. */
4629 void bdrv_unref(BlockDriverState *bs)
4630 {
4631     if (!bs) {
4632         return;
4633     }
4634     assert(bs->refcnt > 0);
4635     if (--bs->refcnt == 0) {
4636         bdrv_delete(bs);
4637     }
4638 }
4639 
4640 struct BdrvOpBlocker {
4641     Error *reason;
4642     QLIST_ENTRY(BdrvOpBlocker) list;
4643 };
4644 
4645 bool bdrv_op_is_blocked(BlockDriverState *bs, BlockOpType op, Error **errp)
4646 {
4647     BdrvOpBlocker *blocker;
4648     assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
4649     if (!QLIST_EMPTY(&bs->op_blockers[op])) {
4650         blocker = QLIST_FIRST(&bs->op_blockers[op]);
4651         error_propagate(errp, error_copy(blocker->reason));
4652         error_prepend(errp, "Node '%s' is busy: ",
4653                       bdrv_get_device_or_node_name(bs));
4654         return true;
4655     }
4656     return false;
4657 }
4658 
4659 void bdrv_op_block(BlockDriverState *bs, BlockOpType op, Error *reason)
4660 {
4661     BdrvOpBlocker *blocker;
4662     assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
4663 
4664     blocker = g_new0(BdrvOpBlocker, 1);
4665     blocker->reason = reason;
4666     QLIST_INSERT_HEAD(&bs->op_blockers[op], blocker, list);
4667 }
4668 
4669 void bdrv_op_unblock(BlockDriverState *bs, BlockOpType op, Error *reason)
4670 {
4671     BdrvOpBlocker *blocker, *next;
4672     assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
4673     QLIST_FOREACH_SAFE(blocker, &bs->op_blockers[op], list, next) {
4674         if (blocker->reason == reason) {
4675             QLIST_REMOVE(blocker, list);
4676             g_free(blocker);
4677         }
4678     }
4679 }
4680 
4681 void bdrv_op_block_all(BlockDriverState *bs, Error *reason)
4682 {
4683     int i;
4684     for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
4685         bdrv_op_block(bs, i, reason);
4686     }
4687 }
4688 
4689 void bdrv_op_unblock_all(BlockDriverState *bs, Error *reason)
4690 {
4691     int i;
4692     for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
4693         bdrv_op_unblock(bs, i, reason);
4694     }
4695 }
4696 
4697 bool bdrv_op_blocker_is_empty(BlockDriverState *bs)
4698 {
4699     int i;
4700 
4701     for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
4702         if (!QLIST_EMPTY(&bs->op_blockers[i])) {
4703             return false;
4704         }
4705     }
4706     return true;
4707 }
4708 
4709 void bdrv_img_create(const char *filename, const char *fmt,
4710                      const char *base_filename, const char *base_fmt,
4711                      char *options, uint64_t img_size, int flags, bool quiet,
4712                      Error **errp)
4713 {
4714     QemuOptsList *create_opts = NULL;
4715     QemuOpts *opts = NULL;
4716     const char *backing_fmt, *backing_file;
4717     int64_t size;
4718     BlockDriver *drv, *proto_drv;
4719     Error *local_err = NULL;
4720     int ret = 0;
4721 
4722     /* Find driver and parse its options */
4723     drv = bdrv_find_format(fmt);
4724     if (!drv) {
4725         error_setg(errp, "Unknown file format '%s'", fmt);
4726         return;
4727     }
4728 
4729     proto_drv = bdrv_find_protocol(filename, true, errp);
4730     if (!proto_drv) {
4731         return;
4732     }
4733 
4734     if (!drv->create_opts) {
4735         error_setg(errp, "Format driver '%s' does not support image creation",
4736                    drv->format_name);
4737         return;
4738     }
4739 
4740     if (!proto_drv->create_opts) {
4741         error_setg(errp, "Protocol driver '%s' does not support image creation",
4742                    proto_drv->format_name);
4743         return;
4744     }
4745 
4746     create_opts = qemu_opts_append(create_opts, drv->create_opts);
4747     create_opts = qemu_opts_append(create_opts, proto_drv->create_opts);
4748 
4749     /* Create parameter list with default values */
4750     opts = qemu_opts_create(create_opts, NULL, 0, &error_abort);
4751     qemu_opt_set_number(opts, BLOCK_OPT_SIZE, img_size, &error_abort);
4752 
4753     /* Parse -o options */
4754     if (options) {
4755         qemu_opts_do_parse(opts, options, NULL, &local_err);
4756         if (local_err) {
4757             error_report_err(local_err);
4758             local_err = NULL;
4759             error_setg(errp, "Invalid options for file format '%s'", fmt);
4760             goto out;
4761         }
4762     }
4763 
4764     if (base_filename) {
4765         qemu_opt_set(opts, BLOCK_OPT_BACKING_FILE, base_filename, &local_err);
4766         if (local_err) {
4767             error_setg(errp, "Backing file not supported for file format '%s'",
4768                        fmt);
4769             goto out;
4770         }
4771     }
4772 
4773     if (base_fmt) {
4774         qemu_opt_set(opts, BLOCK_OPT_BACKING_FMT, base_fmt, &local_err);
4775         if (local_err) {
4776             error_setg(errp, "Backing file format not supported for file "
4777                              "format '%s'", fmt);
4778             goto out;
4779         }
4780     }
4781 
4782     backing_file = qemu_opt_get(opts, BLOCK_OPT_BACKING_FILE);
4783     if (backing_file) {
4784         if (!strcmp(filename, backing_file)) {
4785             error_setg(errp, "Error: Trying to create an image with the "
4786                              "same filename as the backing file");
4787             goto out;
4788         }
4789     }
4790 
4791     backing_fmt = qemu_opt_get(opts, BLOCK_OPT_BACKING_FMT);
4792 
4793     /* The size for the image must always be specified, unless we have a backing
4794      * file and we have not been forbidden from opening it. */
4795     size = qemu_opt_get_size(opts, BLOCK_OPT_SIZE, img_size);
4796     if (backing_file && !(flags & BDRV_O_NO_BACKING)) {
4797         BlockDriverState *bs;
4798         char *full_backing = g_new0(char, PATH_MAX);
4799         int back_flags;
4800         QDict *backing_options = NULL;
4801 
4802         bdrv_get_full_backing_filename_from_filename(filename, backing_file,
4803                                                      full_backing, PATH_MAX,
4804                                                      &local_err);
4805         if (local_err) {
4806             g_free(full_backing);
4807             goto out;
4808         }
4809 
4810         /* backing files always opened read-only */
4811         back_flags = flags;
4812         back_flags &= ~(BDRV_O_RDWR | BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING);
4813 
4814         backing_options = qdict_new();
4815         if (backing_fmt) {
4816             qdict_put_str(backing_options, "driver", backing_fmt);
4817         }
4818         qdict_put_bool(backing_options, BDRV_OPT_FORCE_SHARE, true);
4819 
4820         bs = bdrv_open(full_backing, NULL, backing_options, back_flags,
4821                        &local_err);
4822         g_free(full_backing);
4823         if (!bs && size != -1) {
4824             /* Couldn't open BS, but we have a size, so it's nonfatal */
4825             warn_reportf_err(local_err,
4826                             "Could not verify backing image. "
4827                             "This may become an error in future versions.\n");
4828             local_err = NULL;
4829         } else if (!bs) {
4830             /* Couldn't open bs, do not have size */
4831             error_append_hint(&local_err,
4832                               "Could not open backing image to determine size.\n");
4833             goto out;
4834         } else {
4835             if (size == -1) {
4836                 /* Opened BS, have no size */
4837                 size = bdrv_getlength(bs);
4838                 if (size < 0) {
4839                     error_setg_errno(errp, -size, "Could not get size of '%s'",
4840                                      backing_file);
4841                     bdrv_unref(bs);
4842                     goto out;
4843                 }
4844                 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, size, &error_abort);
4845             }
4846             bdrv_unref(bs);
4847         }
4848     } /* (backing_file && !(flags & BDRV_O_NO_BACKING)) */
4849 
4850     if (size == -1) {
4851         error_setg(errp, "Image creation needs a size parameter");
4852         goto out;
4853     }
4854 
4855     if (!quiet) {
4856         printf("Formatting '%s', fmt=%s ", filename, fmt);
4857         qemu_opts_print(opts, " ");
4858         puts("");
4859     }
4860 
4861     ret = bdrv_create(drv, filename, opts, &local_err);
4862 
4863     if (ret == -EFBIG) {
4864         /* This is generally a better message than whatever the driver would
4865          * deliver (especially because of the cluster_size_hint), since that
4866          * is most probably not much different from "image too large". */
4867         const char *cluster_size_hint = "";
4868         if (qemu_opt_get_size(opts, BLOCK_OPT_CLUSTER_SIZE, 0)) {
4869             cluster_size_hint = " (try using a larger cluster size)";
4870         }
4871         error_setg(errp, "The image size is too large for file format '%s'"
4872                    "%s", fmt, cluster_size_hint);
4873         error_free(local_err);
4874         local_err = NULL;
4875     }
4876 
4877 out:
4878     qemu_opts_del(opts);
4879     qemu_opts_free(create_opts);
4880     error_propagate(errp, local_err);
4881 }
4882 
4883 AioContext *bdrv_get_aio_context(BlockDriverState *bs)
4884 {
4885     return bs ? bs->aio_context : qemu_get_aio_context();
4886 }
4887 
4888 AioWait *bdrv_get_aio_wait(BlockDriverState *bs)
4889 {
4890     return bs ? &bs->wait : NULL;
4891 }
4892 
4893 void bdrv_coroutine_enter(BlockDriverState *bs, Coroutine *co)
4894 {
4895     aio_co_enter(bdrv_get_aio_context(bs), co);
4896 }
4897 
4898 static void bdrv_do_remove_aio_context_notifier(BdrvAioNotifier *ban)
4899 {
4900     QLIST_REMOVE(ban, list);
4901     g_free(ban);
4902 }
4903 
4904 void bdrv_detach_aio_context(BlockDriverState *bs)
4905 {
4906     BdrvAioNotifier *baf, *baf_tmp;
4907     BdrvChild *child;
4908 
4909     if (!bs->drv) {
4910         return;
4911     }
4912 
4913     assert(!bs->walking_aio_notifiers);
4914     bs->walking_aio_notifiers = true;
4915     QLIST_FOREACH_SAFE(baf, &bs->aio_notifiers, list, baf_tmp) {
4916         if (baf->deleted) {
4917             bdrv_do_remove_aio_context_notifier(baf);
4918         } else {
4919             baf->detach_aio_context(baf->opaque);
4920         }
4921     }
4922     /* Never mind iterating again to check for ->deleted.  bdrv_close() will
4923      * remove remaining aio notifiers if we aren't called again.
4924      */
4925     bs->walking_aio_notifiers = false;
4926 
4927     if (bs->drv->bdrv_detach_aio_context) {
4928         bs->drv->bdrv_detach_aio_context(bs);
4929     }
4930     QLIST_FOREACH(child, &bs->children, next) {
4931         bdrv_detach_aio_context(child->bs);
4932     }
4933 
4934     bs->aio_context = NULL;
4935 }
4936 
4937 void bdrv_attach_aio_context(BlockDriverState *bs,
4938                              AioContext *new_context)
4939 {
4940     BdrvAioNotifier *ban, *ban_tmp;
4941     BdrvChild *child;
4942 
4943     if (!bs->drv) {
4944         return;
4945     }
4946 
4947     bs->aio_context = new_context;
4948 
4949     QLIST_FOREACH(child, &bs->children, next) {
4950         bdrv_attach_aio_context(child->bs, new_context);
4951     }
4952     if (bs->drv->bdrv_attach_aio_context) {
4953         bs->drv->bdrv_attach_aio_context(bs, new_context);
4954     }
4955 
4956     assert(!bs->walking_aio_notifiers);
4957     bs->walking_aio_notifiers = true;
4958     QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_tmp) {
4959         if (ban->deleted) {
4960             bdrv_do_remove_aio_context_notifier(ban);
4961         } else {
4962             ban->attached_aio_context(new_context, ban->opaque);
4963         }
4964     }
4965     bs->walking_aio_notifiers = false;
4966 }
4967 
4968 void bdrv_set_aio_context(BlockDriverState *bs, AioContext *new_context)
4969 {
4970     AioContext *ctx = bdrv_get_aio_context(bs);
4971 
4972     aio_disable_external(ctx);
4973     bdrv_parent_drained_begin(bs, NULL, false);
4974     bdrv_drain(bs); /* ensure there are no in-flight requests */
4975 
4976     while (aio_poll(ctx, false)) {
4977         /* wait for all bottom halves to execute */
4978     }
4979 
4980     bdrv_detach_aio_context(bs);
4981 
4982     /* This function executes in the old AioContext so acquire the new one in
4983      * case it runs in a different thread.
4984      */
4985     aio_context_acquire(new_context);
4986     bdrv_attach_aio_context(bs, new_context);
4987     bdrv_parent_drained_end(bs, NULL, false);
4988     aio_enable_external(ctx);
4989     aio_context_release(new_context);
4990 }
4991 
4992 void bdrv_add_aio_context_notifier(BlockDriverState *bs,
4993         void (*attached_aio_context)(AioContext *new_context, void *opaque),
4994         void (*detach_aio_context)(void *opaque), void *opaque)
4995 {
4996     BdrvAioNotifier *ban = g_new(BdrvAioNotifier, 1);
4997     *ban = (BdrvAioNotifier){
4998         .attached_aio_context = attached_aio_context,
4999         .detach_aio_context   = detach_aio_context,
5000         .opaque               = opaque
5001     };
5002 
5003     QLIST_INSERT_HEAD(&bs->aio_notifiers, ban, list);
5004 }
5005 
5006 void bdrv_remove_aio_context_notifier(BlockDriverState *bs,
5007                                       void (*attached_aio_context)(AioContext *,
5008                                                                    void *),
5009                                       void (*detach_aio_context)(void *),
5010                                       void *opaque)
5011 {
5012     BdrvAioNotifier *ban, *ban_next;
5013 
5014     QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_next) {
5015         if (ban->attached_aio_context == attached_aio_context &&
5016             ban->detach_aio_context   == detach_aio_context   &&
5017             ban->opaque               == opaque               &&
5018             ban->deleted              == false)
5019         {
5020             if (bs->walking_aio_notifiers) {
5021                 ban->deleted = true;
5022             } else {
5023                 bdrv_do_remove_aio_context_notifier(ban);
5024             }
5025             return;
5026         }
5027     }
5028 
5029     abort();
5030 }
5031 
5032 int bdrv_amend_options(BlockDriverState *bs, QemuOpts *opts,
5033                        BlockDriverAmendStatusCB *status_cb, void *cb_opaque,
5034                        Error **errp)
5035 {
5036     if (!bs->drv) {
5037         error_setg(errp, "Node is ejected");
5038         return -ENOMEDIUM;
5039     }
5040     if (!bs->drv->bdrv_amend_options) {
5041         error_setg(errp, "Block driver '%s' does not support option amendment",
5042                    bs->drv->format_name);
5043         return -ENOTSUP;
5044     }
5045     return bs->drv->bdrv_amend_options(bs, opts, status_cb, cb_opaque, errp);
5046 }
5047 
5048 /* This function will be called by the bdrv_recurse_is_first_non_filter method
5049  * of block filter and by bdrv_is_first_non_filter.
5050  * It is used to test if the given bs is the candidate or recurse more in the
5051  * node graph.
5052  */
5053 bool bdrv_recurse_is_first_non_filter(BlockDriverState *bs,
5054                                       BlockDriverState *candidate)
5055 {
5056     /* return false if basic checks fails */
5057     if (!bs || !bs->drv) {
5058         return false;
5059     }
5060 
5061     /* the code reached a non block filter driver -> check if the bs is
5062      * the same as the candidate. It's the recursion termination condition.
5063      */
5064     if (!bs->drv->is_filter) {
5065         return bs == candidate;
5066     }
5067     /* Down this path the driver is a block filter driver */
5068 
5069     /* If the block filter recursion method is defined use it to recurse down
5070      * the node graph.
5071      */
5072     if (bs->drv->bdrv_recurse_is_first_non_filter) {
5073         return bs->drv->bdrv_recurse_is_first_non_filter(bs, candidate);
5074     }
5075 
5076     /* the driver is a block filter but don't allow to recurse -> return false
5077      */
5078     return false;
5079 }
5080 
5081 /* This function checks if the candidate is the first non filter bs down it's
5082  * bs chain. Since we don't have pointers to parents it explore all bs chains
5083  * from the top. Some filters can choose not to pass down the recursion.
5084  */
5085 bool bdrv_is_first_non_filter(BlockDriverState *candidate)
5086 {
5087     BlockDriverState *bs;
5088     BdrvNextIterator it;
5089 
5090     /* walk down the bs forest recursively */
5091     for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
5092         bool perm;
5093 
5094         /* try to recurse in this top level bs */
5095         perm = bdrv_recurse_is_first_non_filter(bs, candidate);
5096 
5097         /* candidate is the first non filter */
5098         if (perm) {
5099             bdrv_next_cleanup(&it);
5100             return true;
5101         }
5102     }
5103 
5104     return false;
5105 }
5106 
5107 BlockDriverState *check_to_replace_node(BlockDriverState *parent_bs,
5108                                         const char *node_name, Error **errp)
5109 {
5110     BlockDriverState *to_replace_bs = bdrv_find_node(node_name);
5111     AioContext *aio_context;
5112 
5113     if (!to_replace_bs) {
5114         error_setg(errp, "Node name '%s' not found", node_name);
5115         return NULL;
5116     }
5117 
5118     aio_context = bdrv_get_aio_context(to_replace_bs);
5119     aio_context_acquire(aio_context);
5120 
5121     if (bdrv_op_is_blocked(to_replace_bs, BLOCK_OP_TYPE_REPLACE, errp)) {
5122         to_replace_bs = NULL;
5123         goto out;
5124     }
5125 
5126     /* We don't want arbitrary node of the BDS chain to be replaced only the top
5127      * most non filter in order to prevent data corruption.
5128      * Another benefit is that this tests exclude backing files which are
5129      * blocked by the backing blockers.
5130      */
5131     if (!bdrv_recurse_is_first_non_filter(parent_bs, to_replace_bs)) {
5132         error_setg(errp, "Only top most non filter can be replaced");
5133         to_replace_bs = NULL;
5134         goto out;
5135     }
5136 
5137 out:
5138     aio_context_release(aio_context);
5139     return to_replace_bs;
5140 }
5141 
5142 static bool append_open_options(QDict *d, BlockDriverState *bs)
5143 {
5144     const QDictEntry *entry;
5145     QemuOptDesc *desc;
5146     BdrvChild *child;
5147     bool found_any = false;
5148 
5149     for (entry = qdict_first(bs->options); entry;
5150          entry = qdict_next(bs->options, entry))
5151     {
5152         /* Exclude node-name references to children */
5153         QLIST_FOREACH(child, &bs->children, next) {
5154             if (!strcmp(entry->key, child->name)) {
5155                 break;
5156             }
5157         }
5158         if (child) {
5159             continue;
5160         }
5161 
5162         /* And exclude all non-driver-specific options */
5163         for (desc = bdrv_runtime_opts.desc; desc->name; desc++) {
5164             if (!strcmp(qdict_entry_key(entry), desc->name)) {
5165                 break;
5166             }
5167         }
5168         if (desc->name) {
5169             continue;
5170         }
5171 
5172         qdict_put_obj(d, qdict_entry_key(entry),
5173                       qobject_ref(qdict_entry_value(entry)));
5174         found_any = true;
5175     }
5176 
5177     return found_any;
5178 }
5179 
5180 /* Updates the following BDS fields:
5181  *  - exact_filename: A filename which may be used for opening a block device
5182  *                    which (mostly) equals the given BDS (even without any
5183  *                    other options; so reading and writing must return the same
5184  *                    results, but caching etc. may be different)
5185  *  - full_open_options: Options which, when given when opening a block device
5186  *                       (without a filename), result in a BDS (mostly)
5187  *                       equalling the given one
5188  *  - filename: If exact_filename is set, it is copied here. Otherwise,
5189  *              full_open_options is converted to a JSON object, prefixed with
5190  *              "json:" (for use through the JSON pseudo protocol) and put here.
5191  */
5192 void bdrv_refresh_filename(BlockDriverState *bs)
5193 {
5194     BlockDriver *drv = bs->drv;
5195     QDict *opts;
5196 
5197     if (!drv) {
5198         return;
5199     }
5200 
5201     /* This BDS's file name will most probably depend on its file's name, so
5202      * refresh that first */
5203     if (bs->file) {
5204         bdrv_refresh_filename(bs->file->bs);
5205     }
5206 
5207     if (drv->bdrv_refresh_filename) {
5208         /* Obsolete information is of no use here, so drop the old file name
5209          * information before refreshing it */
5210         bs->exact_filename[0] = '\0';
5211         if (bs->full_open_options) {
5212             qobject_unref(bs->full_open_options);
5213             bs->full_open_options = NULL;
5214         }
5215 
5216         opts = qdict_new();
5217         append_open_options(opts, bs);
5218         drv->bdrv_refresh_filename(bs, opts);
5219         qobject_unref(opts);
5220     } else if (bs->file) {
5221         /* Try to reconstruct valid information from the underlying file */
5222         bool has_open_options;
5223 
5224         bs->exact_filename[0] = '\0';
5225         if (bs->full_open_options) {
5226             qobject_unref(bs->full_open_options);
5227             bs->full_open_options = NULL;
5228         }
5229 
5230         opts = qdict_new();
5231         has_open_options = append_open_options(opts, bs);
5232 
5233         /* If no specific options have been given for this BDS, the filename of
5234          * the underlying file should suffice for this one as well */
5235         if (bs->file->bs->exact_filename[0] && !has_open_options) {
5236             strcpy(bs->exact_filename, bs->file->bs->exact_filename);
5237         }
5238         /* Reconstructing the full options QDict is simple for most format block
5239          * drivers, as long as the full options are known for the underlying
5240          * file BDS. The full options QDict of that file BDS should somehow
5241          * contain a representation of the filename, therefore the following
5242          * suffices without querying the (exact_)filename of this BDS. */
5243         if (bs->file->bs->full_open_options) {
5244             qdict_put_str(opts, "driver", drv->format_name);
5245             qdict_put(opts, "file",
5246                       qobject_ref(bs->file->bs->full_open_options));
5247 
5248             bs->full_open_options = opts;
5249         } else {
5250             qobject_unref(opts);
5251         }
5252     } else if (!bs->full_open_options && qdict_size(bs->options)) {
5253         /* There is no underlying file BDS (at least referenced by BDS.file),
5254          * so the full options QDict should be equal to the options given
5255          * specifically for this block device when it was opened (plus the
5256          * driver specification).
5257          * Because those options don't change, there is no need to update
5258          * full_open_options when it's already set. */
5259 
5260         opts = qdict_new();
5261         append_open_options(opts, bs);
5262         qdict_put_str(opts, "driver", drv->format_name);
5263 
5264         if (bs->exact_filename[0]) {
5265             /* This may not work for all block protocol drivers (some may
5266              * require this filename to be parsed), but we have to find some
5267              * default solution here, so just include it. If some block driver
5268              * does not support pure options without any filename at all or
5269              * needs some special format of the options QDict, it needs to
5270              * implement the driver-specific bdrv_refresh_filename() function.
5271              */
5272             qdict_put_str(opts, "filename", bs->exact_filename);
5273         }
5274 
5275         bs->full_open_options = opts;
5276     }
5277 
5278     if (bs->exact_filename[0]) {
5279         pstrcpy(bs->filename, sizeof(bs->filename), bs->exact_filename);
5280     } else if (bs->full_open_options) {
5281         QString *json = qobject_to_json(QOBJECT(bs->full_open_options));
5282         snprintf(bs->filename, sizeof(bs->filename), "json:%s",
5283                  qstring_get_str(json));
5284         qobject_unref(json);
5285     }
5286 }
5287 
5288 /*
5289  * Hot add/remove a BDS's child. So the user can take a child offline when
5290  * it is broken and take a new child online
5291  */
5292 void bdrv_add_child(BlockDriverState *parent_bs, BlockDriverState *child_bs,
5293                     Error **errp)
5294 {
5295 
5296     if (!parent_bs->drv || !parent_bs->drv->bdrv_add_child) {
5297         error_setg(errp, "The node %s does not support adding a child",
5298                    bdrv_get_device_or_node_name(parent_bs));
5299         return;
5300     }
5301 
5302     if (!QLIST_EMPTY(&child_bs->parents)) {
5303         error_setg(errp, "The node %s already has a parent",
5304                    child_bs->node_name);
5305         return;
5306     }
5307 
5308     parent_bs->drv->bdrv_add_child(parent_bs, child_bs, errp);
5309 }
5310 
5311 void bdrv_del_child(BlockDriverState *parent_bs, BdrvChild *child, Error **errp)
5312 {
5313     BdrvChild *tmp;
5314 
5315     if (!parent_bs->drv || !parent_bs->drv->bdrv_del_child) {
5316         error_setg(errp, "The node %s does not support removing a child",
5317                    bdrv_get_device_or_node_name(parent_bs));
5318         return;
5319     }
5320 
5321     QLIST_FOREACH(tmp, &parent_bs->children, next) {
5322         if (tmp == child) {
5323             break;
5324         }
5325     }
5326 
5327     if (!tmp) {
5328         error_setg(errp, "The node %s does not have a child named %s",
5329                    bdrv_get_device_or_node_name(parent_bs),
5330                    bdrv_get_device_or_node_name(child->bs));
5331         return;
5332     }
5333 
5334     parent_bs->drv->bdrv_del_child(parent_bs, child, errp);
5335 }
5336 
5337 bool bdrv_can_store_new_dirty_bitmap(BlockDriverState *bs, const char *name,
5338                                      uint32_t granularity, Error **errp)
5339 {
5340     BlockDriver *drv = bs->drv;
5341 
5342     if (!drv) {
5343         error_setg_errno(errp, ENOMEDIUM,
5344                          "Can't store persistent bitmaps to %s",
5345                          bdrv_get_device_or_node_name(bs));
5346         return false;
5347     }
5348 
5349     if (!drv->bdrv_can_store_new_dirty_bitmap) {
5350         error_setg_errno(errp, ENOTSUP,
5351                          "Can't store persistent bitmaps to %s",
5352                          bdrv_get_device_or_node_name(bs));
5353         return false;
5354     }
5355 
5356     return drv->bdrv_can_store_new_dirty_bitmap(bs, name, granularity, errp);
5357 }
5358