xref: /openbmc/qemu/block.c (revision dd873966)
1 /*
2  * QEMU System Emulator block driver
3  *
4  * Copyright (c) 2003 Fabrice Bellard
5  *
6  * Permission is hereby granted, free of charge, to any person obtaining a copy
7  * of this software and associated documentation files (the "Software"), to deal
8  * in the Software without restriction, including without limitation the rights
9  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
10  * copies of the Software, and to permit persons to whom the Software is
11  * furnished to do so, subject to the following conditions:
12  *
13  * The above copyright notice and this permission notice shall be included in
14  * all copies or substantial portions of the Software.
15  *
16  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
22  * THE SOFTWARE.
23  */
24 #include "qemu/osdep.h"
25 #include "block/trace.h"
26 #include "block/block_int.h"
27 #include "block/blockjob.h"
28 #include "block/nbd.h"
29 #include "qemu/error-report.h"
30 #include "module_block.h"
31 #include "qemu/module.h"
32 #include "qapi/qmp/qerror.h"
33 #include "qapi/qmp/qbool.h"
34 #include "qapi/qmp/qjson.h"
35 #include "sysemu/block-backend.h"
36 #include "sysemu/sysemu.h"
37 #include "qemu/notify.h"
38 #include "qemu/coroutine.h"
39 #include "block/qapi.h"
40 #include "qmp-commands.h"
41 #include "qemu/timer.h"
42 #include "qapi-event.h"
43 #include "qemu/cutils.h"
44 #include "qemu/id.h"
45 
46 #ifdef CONFIG_BSD
47 #include <sys/ioctl.h>
48 #include <sys/queue.h>
49 #ifndef __DragonFly__
50 #include <sys/disk.h>
51 #endif
52 #endif
53 
54 #ifdef _WIN32
55 #include <windows.h>
56 #endif
57 
58 #define NOT_DONE 0x7fffffff /* used while emulated sync operation in progress */
59 
60 static QTAILQ_HEAD(, BlockDriverState) graph_bdrv_states =
61     QTAILQ_HEAD_INITIALIZER(graph_bdrv_states);
62 
63 static QTAILQ_HEAD(, BlockDriverState) all_bdrv_states =
64     QTAILQ_HEAD_INITIALIZER(all_bdrv_states);
65 
66 static QLIST_HEAD(, BlockDriver) bdrv_drivers =
67     QLIST_HEAD_INITIALIZER(bdrv_drivers);
68 
69 static BlockDriverState *bdrv_open_inherit(const char *filename,
70                                            const char *reference,
71                                            QDict *options, int flags,
72                                            BlockDriverState *parent,
73                                            const BdrvChildRole *child_role,
74                                            Error **errp);
75 
76 /* If non-zero, use only whitelisted block drivers */
77 static int use_bdrv_whitelist;
78 
79 #ifdef _WIN32
80 static int is_windows_drive_prefix(const char *filename)
81 {
82     return (((filename[0] >= 'a' && filename[0] <= 'z') ||
83              (filename[0] >= 'A' && filename[0] <= 'Z')) &&
84             filename[1] == ':');
85 }
86 
87 int is_windows_drive(const char *filename)
88 {
89     if (is_windows_drive_prefix(filename) &&
90         filename[2] == '\0')
91         return 1;
92     if (strstart(filename, "\\\\.\\", NULL) ||
93         strstart(filename, "//./", NULL))
94         return 1;
95     return 0;
96 }
97 #endif
98 
99 size_t bdrv_opt_mem_align(BlockDriverState *bs)
100 {
101     if (!bs || !bs->drv) {
102         /* page size or 4k (hdd sector size) should be on the safe side */
103         return MAX(4096, getpagesize());
104     }
105 
106     return bs->bl.opt_mem_alignment;
107 }
108 
109 size_t bdrv_min_mem_align(BlockDriverState *bs)
110 {
111     if (!bs || !bs->drv) {
112         /* page size or 4k (hdd sector size) should be on the safe side */
113         return MAX(4096, getpagesize());
114     }
115 
116     return bs->bl.min_mem_alignment;
117 }
118 
119 /* check if the path starts with "<protocol>:" */
120 int path_has_protocol(const char *path)
121 {
122     const char *p;
123 
124 #ifdef _WIN32
125     if (is_windows_drive(path) ||
126         is_windows_drive_prefix(path)) {
127         return 0;
128     }
129     p = path + strcspn(path, ":/\\");
130 #else
131     p = path + strcspn(path, ":/");
132 #endif
133 
134     return *p == ':';
135 }
136 
137 int path_is_absolute(const char *path)
138 {
139 #ifdef _WIN32
140     /* specific case for names like: "\\.\d:" */
141     if (is_windows_drive(path) || is_windows_drive_prefix(path)) {
142         return 1;
143     }
144     return (*path == '/' || *path == '\\');
145 #else
146     return (*path == '/');
147 #endif
148 }
149 
150 /* if filename is absolute, just copy it to dest. Otherwise, build a
151    path to it by considering it is relative to base_path. URL are
152    supported. */
153 void path_combine(char *dest, int dest_size,
154                   const char *base_path,
155                   const char *filename)
156 {
157     const char *p, *p1;
158     int len;
159 
160     if (dest_size <= 0)
161         return;
162     if (path_is_absolute(filename)) {
163         pstrcpy(dest, dest_size, filename);
164     } else {
165         const char *protocol_stripped = NULL;
166 
167         if (path_has_protocol(base_path)) {
168             protocol_stripped = strchr(base_path, ':');
169             if (protocol_stripped) {
170                 protocol_stripped++;
171             }
172         }
173         p = protocol_stripped ?: base_path;
174 
175         p1 = strrchr(base_path, '/');
176 #ifdef _WIN32
177         {
178             const char *p2;
179             p2 = strrchr(base_path, '\\');
180             if (!p1 || p2 > p1)
181                 p1 = p2;
182         }
183 #endif
184         if (p1)
185             p1++;
186         else
187             p1 = base_path;
188         if (p1 > p)
189             p = p1;
190         len = p - base_path;
191         if (len > dest_size - 1)
192             len = dest_size - 1;
193         memcpy(dest, base_path, len);
194         dest[len] = '\0';
195         pstrcat(dest, dest_size, filename);
196     }
197 }
198 
199 /*
200  * Helper function for bdrv_parse_filename() implementations to remove optional
201  * protocol prefixes (especially "file:") from a filename and for putting the
202  * stripped filename into the options QDict if there is such a prefix.
203  */
204 void bdrv_parse_filename_strip_prefix(const char *filename, const char *prefix,
205                                       QDict *options)
206 {
207     if (strstart(filename, prefix, &filename)) {
208         /* Stripping the explicit protocol prefix may result in a protocol
209          * prefix being (wrongly) detected (if the filename contains a colon) */
210         if (path_has_protocol(filename)) {
211             QString *fat_filename;
212 
213             /* This means there is some colon before the first slash; therefore,
214              * this cannot be an absolute path */
215             assert(!path_is_absolute(filename));
216 
217             /* And we can thus fix the protocol detection issue by prefixing it
218              * by "./" */
219             fat_filename = qstring_from_str("./");
220             qstring_append(fat_filename, filename);
221 
222             assert(!path_has_protocol(qstring_get_str(fat_filename)));
223 
224             qdict_put(options, "filename", fat_filename);
225         } else {
226             /* If no protocol prefix was detected, we can use the shortened
227              * filename as-is */
228             qdict_put_str(options, "filename", filename);
229         }
230     }
231 }
232 
233 
234 /* Returns whether the image file is opened as read-only. Note that this can
235  * return false and writing to the image file is still not possible because the
236  * image is inactivated. */
237 bool bdrv_is_read_only(BlockDriverState *bs)
238 {
239     return bs->read_only;
240 }
241 
242 int bdrv_can_set_read_only(BlockDriverState *bs, bool read_only,
243                            bool ignore_allow_rdw, Error **errp)
244 {
245     /* Do not set read_only if copy_on_read is enabled */
246     if (bs->copy_on_read && read_only) {
247         error_setg(errp, "Can't set node '%s' to r/o with copy-on-read enabled",
248                    bdrv_get_device_or_node_name(bs));
249         return -EINVAL;
250     }
251 
252     /* Do not clear read_only if it is prohibited */
253     if (!read_only && !(bs->open_flags & BDRV_O_ALLOW_RDWR) &&
254         !ignore_allow_rdw)
255     {
256         error_setg(errp, "Node '%s' is read only",
257                    bdrv_get_device_or_node_name(bs));
258         return -EPERM;
259     }
260 
261     return 0;
262 }
263 
264 /* TODO Remove (deprecated since 2.11)
265  * Block drivers are not supposed to automatically change bs->read_only.
266  * Instead, they should just check whether they can provide what the user
267  * explicitly requested and error out if read-write is requested, but they can
268  * only provide read-only access. */
269 int bdrv_set_read_only(BlockDriverState *bs, bool read_only, Error **errp)
270 {
271     int ret = 0;
272 
273     ret = bdrv_can_set_read_only(bs, read_only, false, errp);
274     if (ret < 0) {
275         return ret;
276     }
277 
278     bs->read_only = read_only;
279     return 0;
280 }
281 
282 void bdrv_get_full_backing_filename_from_filename(const char *backed,
283                                                   const char *backing,
284                                                   char *dest, size_t sz,
285                                                   Error **errp)
286 {
287     if (backing[0] == '\0' || path_has_protocol(backing) ||
288         path_is_absolute(backing))
289     {
290         pstrcpy(dest, sz, backing);
291     } else if (backed[0] == '\0' || strstart(backed, "json:", NULL)) {
292         error_setg(errp, "Cannot use relative backing file names for '%s'",
293                    backed);
294     } else {
295         path_combine(dest, sz, backed, backing);
296     }
297 }
298 
299 void bdrv_get_full_backing_filename(BlockDriverState *bs, char *dest, size_t sz,
300                                     Error **errp)
301 {
302     char *backed = bs->exact_filename[0] ? bs->exact_filename : bs->filename;
303 
304     bdrv_get_full_backing_filename_from_filename(backed, bs->backing_file,
305                                                  dest, sz, errp);
306 }
307 
308 void bdrv_register(BlockDriver *bdrv)
309 {
310     QLIST_INSERT_HEAD(&bdrv_drivers, bdrv, list);
311 }
312 
313 BlockDriverState *bdrv_new(void)
314 {
315     BlockDriverState *bs;
316     int i;
317 
318     bs = g_new0(BlockDriverState, 1);
319     QLIST_INIT(&bs->dirty_bitmaps);
320     for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
321         QLIST_INIT(&bs->op_blockers[i]);
322     }
323     notifier_with_return_list_init(&bs->before_write_notifiers);
324     qemu_co_mutex_init(&bs->reqs_lock);
325     qemu_mutex_init(&bs->dirty_bitmap_mutex);
326     bs->refcnt = 1;
327     bs->aio_context = qemu_get_aio_context();
328 
329     qemu_co_queue_init(&bs->flush_queue);
330 
331     QTAILQ_INSERT_TAIL(&all_bdrv_states, bs, bs_list);
332 
333     return bs;
334 }
335 
336 static BlockDriver *bdrv_do_find_format(const char *format_name)
337 {
338     BlockDriver *drv1;
339 
340     QLIST_FOREACH(drv1, &bdrv_drivers, list) {
341         if (!strcmp(drv1->format_name, format_name)) {
342             return drv1;
343         }
344     }
345 
346     return NULL;
347 }
348 
349 BlockDriver *bdrv_find_format(const char *format_name)
350 {
351     BlockDriver *drv1;
352     int i;
353 
354     drv1 = bdrv_do_find_format(format_name);
355     if (drv1) {
356         return drv1;
357     }
358 
359     /* The driver isn't registered, maybe we need to load a module */
360     for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); ++i) {
361         if (!strcmp(block_driver_modules[i].format_name, format_name)) {
362             block_module_load_one(block_driver_modules[i].library_name);
363             break;
364         }
365     }
366 
367     return bdrv_do_find_format(format_name);
368 }
369 
370 static int bdrv_is_whitelisted(BlockDriver *drv, bool read_only)
371 {
372     static const char *whitelist_rw[] = {
373         CONFIG_BDRV_RW_WHITELIST
374     };
375     static const char *whitelist_ro[] = {
376         CONFIG_BDRV_RO_WHITELIST
377     };
378     const char **p;
379 
380     if (!whitelist_rw[0] && !whitelist_ro[0]) {
381         return 1;               /* no whitelist, anything goes */
382     }
383 
384     for (p = whitelist_rw; *p; p++) {
385         if (!strcmp(drv->format_name, *p)) {
386             return 1;
387         }
388     }
389     if (read_only) {
390         for (p = whitelist_ro; *p; p++) {
391             if (!strcmp(drv->format_name, *p)) {
392                 return 1;
393             }
394         }
395     }
396     return 0;
397 }
398 
399 bool bdrv_uses_whitelist(void)
400 {
401     return use_bdrv_whitelist;
402 }
403 
404 typedef struct CreateCo {
405     BlockDriver *drv;
406     char *filename;
407     QemuOpts *opts;
408     int ret;
409     Error *err;
410 } CreateCo;
411 
412 static void coroutine_fn bdrv_create_co_entry(void *opaque)
413 {
414     Error *local_err = NULL;
415     int ret;
416 
417     CreateCo *cco = opaque;
418     assert(cco->drv);
419 
420     ret = cco->drv->bdrv_create(cco->filename, cco->opts, &local_err);
421     error_propagate(&cco->err, local_err);
422     cco->ret = ret;
423 }
424 
425 int bdrv_create(BlockDriver *drv, const char* filename,
426                 QemuOpts *opts, Error **errp)
427 {
428     int ret;
429 
430     Coroutine *co;
431     CreateCo cco = {
432         .drv = drv,
433         .filename = g_strdup(filename),
434         .opts = opts,
435         .ret = NOT_DONE,
436         .err = NULL,
437     };
438 
439     if (!drv->bdrv_create) {
440         error_setg(errp, "Driver '%s' does not support image creation", drv->format_name);
441         ret = -ENOTSUP;
442         goto out;
443     }
444 
445     if (qemu_in_coroutine()) {
446         /* Fast-path if already in coroutine context */
447         bdrv_create_co_entry(&cco);
448     } else {
449         co = qemu_coroutine_create(bdrv_create_co_entry, &cco);
450         qemu_coroutine_enter(co);
451         while (cco.ret == NOT_DONE) {
452             aio_poll(qemu_get_aio_context(), true);
453         }
454     }
455 
456     ret = cco.ret;
457     if (ret < 0) {
458         if (cco.err) {
459             error_propagate(errp, cco.err);
460         } else {
461             error_setg_errno(errp, -ret, "Could not create image");
462         }
463     }
464 
465 out:
466     g_free(cco.filename);
467     return ret;
468 }
469 
470 int bdrv_create_file(const char *filename, QemuOpts *opts, Error **errp)
471 {
472     BlockDriver *drv;
473     Error *local_err = NULL;
474     int ret;
475 
476     drv = bdrv_find_protocol(filename, true, errp);
477     if (drv == NULL) {
478         return -ENOENT;
479     }
480 
481     ret = bdrv_create(drv, filename, opts, &local_err);
482     error_propagate(errp, local_err);
483     return ret;
484 }
485 
486 /**
487  * Try to get @bs's logical and physical block size.
488  * On success, store them in @bsz struct and return 0.
489  * On failure return -errno.
490  * @bs must not be empty.
491  */
492 int bdrv_probe_blocksizes(BlockDriverState *bs, BlockSizes *bsz)
493 {
494     BlockDriver *drv = bs->drv;
495 
496     if (drv && drv->bdrv_probe_blocksizes) {
497         return drv->bdrv_probe_blocksizes(bs, bsz);
498     } else if (drv && drv->is_filter && bs->file) {
499         return bdrv_probe_blocksizes(bs->file->bs, bsz);
500     }
501 
502     return -ENOTSUP;
503 }
504 
505 /**
506  * Try to get @bs's geometry (cyls, heads, sectors).
507  * On success, store them in @geo struct and return 0.
508  * On failure return -errno.
509  * @bs must not be empty.
510  */
511 int bdrv_probe_geometry(BlockDriverState *bs, HDGeometry *geo)
512 {
513     BlockDriver *drv = bs->drv;
514 
515     if (drv && drv->bdrv_probe_geometry) {
516         return drv->bdrv_probe_geometry(bs, geo);
517     } else if (drv && drv->is_filter && bs->file) {
518         return bdrv_probe_geometry(bs->file->bs, geo);
519     }
520 
521     return -ENOTSUP;
522 }
523 
524 /*
525  * Create a uniquely-named empty temporary file.
526  * Return 0 upon success, otherwise a negative errno value.
527  */
528 int get_tmp_filename(char *filename, int size)
529 {
530 #ifdef _WIN32
531     char temp_dir[MAX_PATH];
532     /* GetTempFileName requires that its output buffer (4th param)
533        have length MAX_PATH or greater.  */
534     assert(size >= MAX_PATH);
535     return (GetTempPath(MAX_PATH, temp_dir)
536             && GetTempFileName(temp_dir, "qem", 0, filename)
537             ? 0 : -GetLastError());
538 #else
539     int fd;
540     const char *tmpdir;
541     tmpdir = getenv("TMPDIR");
542     if (!tmpdir) {
543         tmpdir = "/var/tmp";
544     }
545     if (snprintf(filename, size, "%s/vl.XXXXXX", tmpdir) >= size) {
546         return -EOVERFLOW;
547     }
548     fd = mkstemp(filename);
549     if (fd < 0) {
550         return -errno;
551     }
552     if (close(fd) != 0) {
553         unlink(filename);
554         return -errno;
555     }
556     return 0;
557 #endif
558 }
559 
560 /*
561  * Detect host devices. By convention, /dev/cdrom[N] is always
562  * recognized as a host CDROM.
563  */
564 static BlockDriver *find_hdev_driver(const char *filename)
565 {
566     int score_max = 0, score;
567     BlockDriver *drv = NULL, *d;
568 
569     QLIST_FOREACH(d, &bdrv_drivers, list) {
570         if (d->bdrv_probe_device) {
571             score = d->bdrv_probe_device(filename);
572             if (score > score_max) {
573                 score_max = score;
574                 drv = d;
575             }
576         }
577     }
578 
579     return drv;
580 }
581 
582 static BlockDriver *bdrv_do_find_protocol(const char *protocol)
583 {
584     BlockDriver *drv1;
585 
586     QLIST_FOREACH(drv1, &bdrv_drivers, list) {
587         if (drv1->protocol_name && !strcmp(drv1->protocol_name, protocol)) {
588             return drv1;
589         }
590     }
591 
592     return NULL;
593 }
594 
595 BlockDriver *bdrv_find_protocol(const char *filename,
596                                 bool allow_protocol_prefix,
597                                 Error **errp)
598 {
599     BlockDriver *drv1;
600     char protocol[128];
601     int len;
602     const char *p;
603     int i;
604 
605     /* TODO Drivers without bdrv_file_open must be specified explicitly */
606 
607     /*
608      * XXX(hch): we really should not let host device detection
609      * override an explicit protocol specification, but moving this
610      * later breaks access to device names with colons in them.
611      * Thanks to the brain-dead persistent naming schemes on udev-
612      * based Linux systems those actually are quite common.
613      */
614     drv1 = find_hdev_driver(filename);
615     if (drv1) {
616         return drv1;
617     }
618 
619     if (!path_has_protocol(filename) || !allow_protocol_prefix) {
620         return &bdrv_file;
621     }
622 
623     p = strchr(filename, ':');
624     assert(p != NULL);
625     len = p - filename;
626     if (len > sizeof(protocol) - 1)
627         len = sizeof(protocol) - 1;
628     memcpy(protocol, filename, len);
629     protocol[len] = '\0';
630 
631     drv1 = bdrv_do_find_protocol(protocol);
632     if (drv1) {
633         return drv1;
634     }
635 
636     for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); ++i) {
637         if (block_driver_modules[i].protocol_name &&
638             !strcmp(block_driver_modules[i].protocol_name, protocol)) {
639             block_module_load_one(block_driver_modules[i].library_name);
640             break;
641         }
642     }
643 
644     drv1 = bdrv_do_find_protocol(protocol);
645     if (!drv1) {
646         error_setg(errp, "Unknown protocol '%s'", protocol);
647     }
648     return drv1;
649 }
650 
651 /*
652  * Guess image format by probing its contents.
653  * This is not a good idea when your image is raw (CVE-2008-2004), but
654  * we do it anyway for backward compatibility.
655  *
656  * @buf         contains the image's first @buf_size bytes.
657  * @buf_size    is the buffer size in bytes (generally BLOCK_PROBE_BUF_SIZE,
658  *              but can be smaller if the image file is smaller)
659  * @filename    is its filename.
660  *
661  * For all block drivers, call the bdrv_probe() method to get its
662  * probing score.
663  * Return the first block driver with the highest probing score.
664  */
665 BlockDriver *bdrv_probe_all(const uint8_t *buf, int buf_size,
666                             const char *filename)
667 {
668     int score_max = 0, score;
669     BlockDriver *drv = NULL, *d;
670 
671     QLIST_FOREACH(d, &bdrv_drivers, list) {
672         if (d->bdrv_probe) {
673             score = d->bdrv_probe(buf, buf_size, filename);
674             if (score > score_max) {
675                 score_max = score;
676                 drv = d;
677             }
678         }
679     }
680 
681     return drv;
682 }
683 
684 static int find_image_format(BlockBackend *file, const char *filename,
685                              BlockDriver **pdrv, Error **errp)
686 {
687     BlockDriver *drv;
688     uint8_t buf[BLOCK_PROBE_BUF_SIZE];
689     int ret = 0;
690 
691     /* Return the raw BlockDriver * to scsi-generic devices or empty drives */
692     if (blk_is_sg(file) || !blk_is_inserted(file) || blk_getlength(file) == 0) {
693         *pdrv = &bdrv_raw;
694         return ret;
695     }
696 
697     ret = blk_pread(file, 0, buf, sizeof(buf));
698     if (ret < 0) {
699         error_setg_errno(errp, -ret, "Could not read image for determining its "
700                          "format");
701         *pdrv = NULL;
702         return ret;
703     }
704 
705     drv = bdrv_probe_all(buf, ret, filename);
706     if (!drv) {
707         error_setg(errp, "Could not determine image format: No compatible "
708                    "driver found");
709         ret = -ENOENT;
710     }
711     *pdrv = drv;
712     return ret;
713 }
714 
715 /**
716  * Set the current 'total_sectors' value
717  * Return 0 on success, -errno on error.
718  */
719 static int refresh_total_sectors(BlockDriverState *bs, int64_t hint)
720 {
721     BlockDriver *drv = bs->drv;
722 
723     if (!drv) {
724         return -ENOMEDIUM;
725     }
726 
727     /* Do not attempt drv->bdrv_getlength() on scsi-generic devices */
728     if (bdrv_is_sg(bs))
729         return 0;
730 
731     /* query actual device if possible, otherwise just trust the hint */
732     if (drv->bdrv_getlength) {
733         int64_t length = drv->bdrv_getlength(bs);
734         if (length < 0) {
735             return length;
736         }
737         hint = DIV_ROUND_UP(length, BDRV_SECTOR_SIZE);
738     }
739 
740     bs->total_sectors = hint;
741     return 0;
742 }
743 
744 /**
745  * Combines a QDict of new block driver @options with any missing options taken
746  * from @old_options, so that leaving out an option defaults to its old value.
747  */
748 static void bdrv_join_options(BlockDriverState *bs, QDict *options,
749                               QDict *old_options)
750 {
751     if (bs->drv && bs->drv->bdrv_join_options) {
752         bs->drv->bdrv_join_options(options, old_options);
753     } else {
754         qdict_join(options, old_options, false);
755     }
756 }
757 
758 /**
759  * Set open flags for a given discard mode
760  *
761  * Return 0 on success, -1 if the discard mode was invalid.
762  */
763 int bdrv_parse_discard_flags(const char *mode, int *flags)
764 {
765     *flags &= ~BDRV_O_UNMAP;
766 
767     if (!strcmp(mode, "off") || !strcmp(mode, "ignore")) {
768         /* do nothing */
769     } else if (!strcmp(mode, "on") || !strcmp(mode, "unmap")) {
770         *flags |= BDRV_O_UNMAP;
771     } else {
772         return -1;
773     }
774 
775     return 0;
776 }
777 
778 /**
779  * Set open flags for a given cache mode
780  *
781  * Return 0 on success, -1 if the cache mode was invalid.
782  */
783 int bdrv_parse_cache_mode(const char *mode, int *flags, bool *writethrough)
784 {
785     *flags &= ~BDRV_O_CACHE_MASK;
786 
787     if (!strcmp(mode, "off") || !strcmp(mode, "none")) {
788         *writethrough = false;
789         *flags |= BDRV_O_NOCACHE;
790     } else if (!strcmp(mode, "directsync")) {
791         *writethrough = true;
792         *flags |= BDRV_O_NOCACHE;
793     } else if (!strcmp(mode, "writeback")) {
794         *writethrough = false;
795     } else if (!strcmp(mode, "unsafe")) {
796         *writethrough = false;
797         *flags |= BDRV_O_NO_FLUSH;
798     } else if (!strcmp(mode, "writethrough")) {
799         *writethrough = true;
800     } else {
801         return -1;
802     }
803 
804     return 0;
805 }
806 
807 static char *bdrv_child_get_parent_desc(BdrvChild *c)
808 {
809     BlockDriverState *parent = c->opaque;
810     return g_strdup(bdrv_get_device_or_node_name(parent));
811 }
812 
813 static void bdrv_child_cb_drained_begin(BdrvChild *child)
814 {
815     BlockDriverState *bs = child->opaque;
816     bdrv_drained_begin(bs);
817 }
818 
819 static void bdrv_child_cb_drained_end(BdrvChild *child)
820 {
821     BlockDriverState *bs = child->opaque;
822     bdrv_drained_end(bs);
823 }
824 
825 static int bdrv_child_cb_inactivate(BdrvChild *child)
826 {
827     BlockDriverState *bs = child->opaque;
828     assert(bs->open_flags & BDRV_O_INACTIVE);
829     return 0;
830 }
831 
832 /*
833  * Returns the options and flags that a temporary snapshot should get, based on
834  * the originally requested flags (the originally requested image will have
835  * flags like a backing file)
836  */
837 static void bdrv_temp_snapshot_options(int *child_flags, QDict *child_options,
838                                        int parent_flags, QDict *parent_options)
839 {
840     *child_flags = (parent_flags & ~BDRV_O_SNAPSHOT) | BDRV_O_TEMPORARY;
841 
842     /* For temporary files, unconditional cache=unsafe is fine */
843     qdict_set_default_str(child_options, BDRV_OPT_CACHE_DIRECT, "off");
844     qdict_set_default_str(child_options, BDRV_OPT_CACHE_NO_FLUSH, "on");
845 
846     /* Copy the read-only option from the parent */
847     qdict_copy_default(child_options, parent_options, BDRV_OPT_READ_ONLY);
848 
849     /* aio=native doesn't work for cache.direct=off, so disable it for the
850      * temporary snapshot */
851     *child_flags &= ~BDRV_O_NATIVE_AIO;
852 }
853 
854 /*
855  * Returns the options and flags that bs->file should get if a protocol driver
856  * is expected, based on the given options and flags for the parent BDS
857  */
858 static void bdrv_inherited_options(int *child_flags, QDict *child_options,
859                                    int parent_flags, QDict *parent_options)
860 {
861     int flags = parent_flags;
862 
863     /* Enable protocol handling, disable format probing for bs->file */
864     flags |= BDRV_O_PROTOCOL;
865 
866     /* If the cache mode isn't explicitly set, inherit direct and no-flush from
867      * the parent. */
868     qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_DIRECT);
869     qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_NO_FLUSH);
870     qdict_copy_default(child_options, parent_options, BDRV_OPT_FORCE_SHARE);
871 
872     /* Inherit the read-only option from the parent if it's not set */
873     qdict_copy_default(child_options, parent_options, BDRV_OPT_READ_ONLY);
874 
875     /* Our block drivers take care to send flushes and respect unmap policy,
876      * so we can default to enable both on lower layers regardless of the
877      * corresponding parent options. */
878     qdict_set_default_str(child_options, BDRV_OPT_DISCARD, "unmap");
879 
880     /* Clear flags that only apply to the top layer */
881     flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_COPY_ON_READ |
882                BDRV_O_NO_IO);
883 
884     *child_flags = flags;
885 }
886 
887 const BdrvChildRole child_file = {
888     .get_parent_desc = bdrv_child_get_parent_desc,
889     .inherit_options = bdrv_inherited_options,
890     .drained_begin   = bdrv_child_cb_drained_begin,
891     .drained_end     = bdrv_child_cb_drained_end,
892     .inactivate      = bdrv_child_cb_inactivate,
893 };
894 
895 /*
896  * Returns the options and flags that bs->file should get if the use of formats
897  * (and not only protocols) is permitted for it, based on the given options and
898  * flags for the parent BDS
899  */
900 static void bdrv_inherited_fmt_options(int *child_flags, QDict *child_options,
901                                        int parent_flags, QDict *parent_options)
902 {
903     child_file.inherit_options(child_flags, child_options,
904                                parent_flags, parent_options);
905 
906     *child_flags &= ~(BDRV_O_PROTOCOL | BDRV_O_NO_IO);
907 }
908 
909 const BdrvChildRole child_format = {
910     .get_parent_desc = bdrv_child_get_parent_desc,
911     .inherit_options = bdrv_inherited_fmt_options,
912     .drained_begin   = bdrv_child_cb_drained_begin,
913     .drained_end     = bdrv_child_cb_drained_end,
914     .inactivate      = bdrv_child_cb_inactivate,
915 };
916 
917 static void bdrv_backing_attach(BdrvChild *c)
918 {
919     BlockDriverState *parent = c->opaque;
920     BlockDriverState *backing_hd = c->bs;
921 
922     assert(!parent->backing_blocker);
923     error_setg(&parent->backing_blocker,
924                "node is used as backing hd of '%s'",
925                bdrv_get_device_or_node_name(parent));
926 
927     parent->open_flags &= ~BDRV_O_NO_BACKING;
928     pstrcpy(parent->backing_file, sizeof(parent->backing_file),
929             backing_hd->filename);
930     pstrcpy(parent->backing_format, sizeof(parent->backing_format),
931             backing_hd->drv ? backing_hd->drv->format_name : "");
932 
933     bdrv_op_block_all(backing_hd, parent->backing_blocker);
934     /* Otherwise we won't be able to commit or stream */
935     bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_COMMIT_TARGET,
936                     parent->backing_blocker);
937     bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_STREAM,
938                     parent->backing_blocker);
939     /*
940      * We do backup in 3 ways:
941      * 1. drive backup
942      *    The target bs is new opened, and the source is top BDS
943      * 2. blockdev backup
944      *    Both the source and the target are top BDSes.
945      * 3. internal backup(used for block replication)
946      *    Both the source and the target are backing file
947      *
948      * In case 1 and 2, neither the source nor the target is the backing file.
949      * In case 3, we will block the top BDS, so there is only one block job
950      * for the top BDS and its backing chain.
951      */
952     bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_BACKUP_SOURCE,
953                     parent->backing_blocker);
954     bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_BACKUP_TARGET,
955                     parent->backing_blocker);
956 }
957 
958 static void bdrv_backing_detach(BdrvChild *c)
959 {
960     BlockDriverState *parent = c->opaque;
961 
962     assert(parent->backing_blocker);
963     bdrv_op_unblock_all(c->bs, parent->backing_blocker);
964     error_free(parent->backing_blocker);
965     parent->backing_blocker = NULL;
966 }
967 
968 /*
969  * Returns the options and flags that bs->backing should get, based on the
970  * given options and flags for the parent BDS
971  */
972 static void bdrv_backing_options(int *child_flags, QDict *child_options,
973                                  int parent_flags, QDict *parent_options)
974 {
975     int flags = parent_flags;
976 
977     /* The cache mode is inherited unmodified for backing files; except WCE,
978      * which is only applied on the top level (BlockBackend) */
979     qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_DIRECT);
980     qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_NO_FLUSH);
981     qdict_copy_default(child_options, parent_options, BDRV_OPT_FORCE_SHARE);
982 
983     /* backing files always opened read-only */
984     qdict_set_default_str(child_options, BDRV_OPT_READ_ONLY, "on");
985     flags &= ~BDRV_O_COPY_ON_READ;
986 
987     /* snapshot=on is handled on the top layer */
988     flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_TEMPORARY);
989 
990     *child_flags = flags;
991 }
992 
993 static int bdrv_backing_update_filename(BdrvChild *c, BlockDriverState *base,
994                                         const char *filename, Error **errp)
995 {
996     BlockDriverState *parent = c->opaque;
997     int orig_flags = bdrv_get_flags(parent);
998     int ret;
999 
1000     if (!(orig_flags & BDRV_O_RDWR)) {
1001         ret = bdrv_reopen(parent, orig_flags | BDRV_O_RDWR, errp);
1002         if (ret < 0) {
1003             return ret;
1004         }
1005     }
1006 
1007     ret = bdrv_change_backing_file(parent, filename,
1008                                    base->drv ? base->drv->format_name : "");
1009     if (ret < 0) {
1010         error_setg_errno(errp, -ret, "Could not update backing file link");
1011     }
1012 
1013     if (!(orig_flags & BDRV_O_RDWR)) {
1014         bdrv_reopen(parent, orig_flags, NULL);
1015     }
1016 
1017     return ret;
1018 }
1019 
1020 const BdrvChildRole child_backing = {
1021     .get_parent_desc = bdrv_child_get_parent_desc,
1022     .attach          = bdrv_backing_attach,
1023     .detach          = bdrv_backing_detach,
1024     .inherit_options = bdrv_backing_options,
1025     .drained_begin   = bdrv_child_cb_drained_begin,
1026     .drained_end     = bdrv_child_cb_drained_end,
1027     .inactivate      = bdrv_child_cb_inactivate,
1028     .update_filename = bdrv_backing_update_filename,
1029 };
1030 
1031 static int bdrv_open_flags(BlockDriverState *bs, int flags)
1032 {
1033     int open_flags = flags;
1034 
1035     /*
1036      * Clear flags that are internal to the block layer before opening the
1037      * image.
1038      */
1039     open_flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_PROTOCOL);
1040 
1041     /*
1042      * Snapshots should be writable.
1043      */
1044     if (flags & BDRV_O_TEMPORARY) {
1045         open_flags |= BDRV_O_RDWR;
1046     }
1047 
1048     return open_flags;
1049 }
1050 
1051 static void update_flags_from_options(int *flags, QemuOpts *opts)
1052 {
1053     *flags &= ~BDRV_O_CACHE_MASK;
1054 
1055     assert(qemu_opt_find(opts, BDRV_OPT_CACHE_NO_FLUSH));
1056     if (qemu_opt_get_bool(opts, BDRV_OPT_CACHE_NO_FLUSH, false)) {
1057         *flags |= BDRV_O_NO_FLUSH;
1058     }
1059 
1060     assert(qemu_opt_find(opts, BDRV_OPT_CACHE_DIRECT));
1061     if (qemu_opt_get_bool(opts, BDRV_OPT_CACHE_DIRECT, false)) {
1062         *flags |= BDRV_O_NOCACHE;
1063     }
1064 
1065     *flags &= ~BDRV_O_RDWR;
1066 
1067     assert(qemu_opt_find(opts, BDRV_OPT_READ_ONLY));
1068     if (!qemu_opt_get_bool(opts, BDRV_OPT_READ_ONLY, false)) {
1069         *flags |= BDRV_O_RDWR;
1070     }
1071 
1072 }
1073 
1074 static void update_options_from_flags(QDict *options, int flags)
1075 {
1076     if (!qdict_haskey(options, BDRV_OPT_CACHE_DIRECT)) {
1077         qdict_put_bool(options, BDRV_OPT_CACHE_DIRECT, flags & BDRV_O_NOCACHE);
1078     }
1079     if (!qdict_haskey(options, BDRV_OPT_CACHE_NO_FLUSH)) {
1080         qdict_put_bool(options, BDRV_OPT_CACHE_NO_FLUSH,
1081                        flags & BDRV_O_NO_FLUSH);
1082     }
1083     if (!qdict_haskey(options, BDRV_OPT_READ_ONLY)) {
1084         qdict_put_bool(options, BDRV_OPT_READ_ONLY, !(flags & BDRV_O_RDWR));
1085     }
1086 }
1087 
1088 static void bdrv_assign_node_name(BlockDriverState *bs,
1089                                   const char *node_name,
1090                                   Error **errp)
1091 {
1092     char *gen_node_name = NULL;
1093 
1094     if (!node_name) {
1095         node_name = gen_node_name = id_generate(ID_BLOCK);
1096     } else if (!id_wellformed(node_name)) {
1097         /*
1098          * Check for empty string or invalid characters, but not if it is
1099          * generated (generated names use characters not available to the user)
1100          */
1101         error_setg(errp, "Invalid node name");
1102         return;
1103     }
1104 
1105     /* takes care of avoiding namespaces collisions */
1106     if (blk_by_name(node_name)) {
1107         error_setg(errp, "node-name=%s is conflicting with a device id",
1108                    node_name);
1109         goto out;
1110     }
1111 
1112     /* takes care of avoiding duplicates node names */
1113     if (bdrv_find_node(node_name)) {
1114         error_setg(errp, "Duplicate node name");
1115         goto out;
1116     }
1117 
1118     /* copy node name into the bs and insert it into the graph list */
1119     pstrcpy(bs->node_name, sizeof(bs->node_name), node_name);
1120     QTAILQ_INSERT_TAIL(&graph_bdrv_states, bs, node_list);
1121 out:
1122     g_free(gen_node_name);
1123 }
1124 
1125 static int bdrv_open_driver(BlockDriverState *bs, BlockDriver *drv,
1126                             const char *node_name, QDict *options,
1127                             int open_flags, Error **errp)
1128 {
1129     Error *local_err = NULL;
1130     int ret;
1131 
1132     bdrv_assign_node_name(bs, node_name, &local_err);
1133     if (local_err) {
1134         error_propagate(errp, local_err);
1135         return -EINVAL;
1136     }
1137 
1138     bs->drv = drv;
1139     bs->read_only = !(bs->open_flags & BDRV_O_RDWR);
1140     bs->opaque = g_malloc0(drv->instance_size);
1141 
1142     if (drv->bdrv_file_open) {
1143         assert(!drv->bdrv_needs_filename || bs->filename[0]);
1144         ret = drv->bdrv_file_open(bs, options, open_flags, &local_err);
1145     } else if (drv->bdrv_open) {
1146         ret = drv->bdrv_open(bs, options, open_flags, &local_err);
1147     } else {
1148         ret = 0;
1149     }
1150 
1151     if (ret < 0) {
1152         if (local_err) {
1153             error_propagate(errp, local_err);
1154         } else if (bs->filename[0]) {
1155             error_setg_errno(errp, -ret, "Could not open '%s'", bs->filename);
1156         } else {
1157             error_setg_errno(errp, -ret, "Could not open image");
1158         }
1159         goto open_failed;
1160     }
1161 
1162     ret = refresh_total_sectors(bs, bs->total_sectors);
1163     if (ret < 0) {
1164         error_setg_errno(errp, -ret, "Could not refresh total sector count");
1165         return ret;
1166     }
1167 
1168     bdrv_refresh_limits(bs, &local_err);
1169     if (local_err) {
1170         error_propagate(errp, local_err);
1171         return -EINVAL;
1172     }
1173 
1174     assert(bdrv_opt_mem_align(bs) != 0);
1175     assert(bdrv_min_mem_align(bs) != 0);
1176     assert(is_power_of_2(bs->bl.request_alignment));
1177 
1178     return 0;
1179 open_failed:
1180     bs->drv = NULL;
1181     if (bs->file != NULL) {
1182         bdrv_unref_child(bs, bs->file);
1183         bs->file = NULL;
1184     }
1185     g_free(bs->opaque);
1186     bs->opaque = NULL;
1187     return ret;
1188 }
1189 
1190 BlockDriverState *bdrv_new_open_driver(BlockDriver *drv, const char *node_name,
1191                                        int flags, Error **errp)
1192 {
1193     BlockDriverState *bs;
1194     int ret;
1195 
1196     bs = bdrv_new();
1197     bs->open_flags = flags;
1198     bs->explicit_options = qdict_new();
1199     bs->options = qdict_new();
1200     bs->opaque = NULL;
1201 
1202     update_options_from_flags(bs->options, flags);
1203 
1204     ret = bdrv_open_driver(bs, drv, node_name, bs->options, flags, errp);
1205     if (ret < 0) {
1206         QDECREF(bs->explicit_options);
1207         bs->explicit_options = NULL;
1208         QDECREF(bs->options);
1209         bs->options = NULL;
1210         bdrv_unref(bs);
1211         return NULL;
1212     }
1213 
1214     return bs;
1215 }
1216 
1217 QemuOptsList bdrv_runtime_opts = {
1218     .name = "bdrv_common",
1219     .head = QTAILQ_HEAD_INITIALIZER(bdrv_runtime_opts.head),
1220     .desc = {
1221         {
1222             .name = "node-name",
1223             .type = QEMU_OPT_STRING,
1224             .help = "Node name of the block device node",
1225         },
1226         {
1227             .name = "driver",
1228             .type = QEMU_OPT_STRING,
1229             .help = "Block driver to use for the node",
1230         },
1231         {
1232             .name = BDRV_OPT_CACHE_DIRECT,
1233             .type = QEMU_OPT_BOOL,
1234             .help = "Bypass software writeback cache on the host",
1235         },
1236         {
1237             .name = BDRV_OPT_CACHE_NO_FLUSH,
1238             .type = QEMU_OPT_BOOL,
1239             .help = "Ignore flush requests",
1240         },
1241         {
1242             .name = BDRV_OPT_READ_ONLY,
1243             .type = QEMU_OPT_BOOL,
1244             .help = "Node is opened in read-only mode",
1245         },
1246         {
1247             .name = "detect-zeroes",
1248             .type = QEMU_OPT_STRING,
1249             .help = "try to optimize zero writes (off, on, unmap)",
1250         },
1251         {
1252             .name = "discard",
1253             .type = QEMU_OPT_STRING,
1254             .help = "discard operation (ignore/off, unmap/on)",
1255         },
1256         {
1257             .name = BDRV_OPT_FORCE_SHARE,
1258             .type = QEMU_OPT_BOOL,
1259             .help = "always accept other writers (default: off)",
1260         },
1261         { /* end of list */ }
1262     },
1263 };
1264 
1265 /*
1266  * Common part for opening disk images and files
1267  *
1268  * Removes all processed options from *options.
1269  */
1270 static int bdrv_open_common(BlockDriverState *bs, BlockBackend *file,
1271                             QDict *options, Error **errp)
1272 {
1273     int ret, open_flags;
1274     const char *filename;
1275     const char *driver_name = NULL;
1276     const char *node_name = NULL;
1277     const char *discard;
1278     const char *detect_zeroes;
1279     QemuOpts *opts;
1280     BlockDriver *drv;
1281     Error *local_err = NULL;
1282 
1283     assert(bs->file == NULL);
1284     assert(options != NULL && bs->options != options);
1285 
1286     opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
1287     qemu_opts_absorb_qdict(opts, options, &local_err);
1288     if (local_err) {
1289         error_propagate(errp, local_err);
1290         ret = -EINVAL;
1291         goto fail_opts;
1292     }
1293 
1294     update_flags_from_options(&bs->open_flags, opts);
1295 
1296     driver_name = qemu_opt_get(opts, "driver");
1297     drv = bdrv_find_format(driver_name);
1298     assert(drv != NULL);
1299 
1300     bs->force_share = qemu_opt_get_bool(opts, BDRV_OPT_FORCE_SHARE, false);
1301 
1302     if (bs->force_share && (bs->open_flags & BDRV_O_RDWR)) {
1303         error_setg(errp,
1304                    BDRV_OPT_FORCE_SHARE
1305                    "=on can only be used with read-only images");
1306         ret = -EINVAL;
1307         goto fail_opts;
1308     }
1309 
1310     if (file != NULL) {
1311         filename = blk_bs(file)->filename;
1312     } else {
1313         /*
1314          * Caution: while qdict_get_try_str() is fine, getting
1315          * non-string types would require more care.  When @options
1316          * come from -blockdev or blockdev_add, its members are typed
1317          * according to the QAPI schema, but when they come from
1318          * -drive, they're all QString.
1319          */
1320         filename = qdict_get_try_str(options, "filename");
1321     }
1322 
1323     if (drv->bdrv_needs_filename && (!filename || !filename[0])) {
1324         error_setg(errp, "The '%s' block driver requires a file name",
1325                    drv->format_name);
1326         ret = -EINVAL;
1327         goto fail_opts;
1328     }
1329 
1330     trace_bdrv_open_common(bs, filename ?: "", bs->open_flags,
1331                            drv->format_name);
1332 
1333     bs->read_only = !(bs->open_flags & BDRV_O_RDWR);
1334 
1335     if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv, bs->read_only)) {
1336         error_setg(errp,
1337                    !bs->read_only && bdrv_is_whitelisted(drv, true)
1338                         ? "Driver '%s' can only be used for read-only devices"
1339                         : "Driver '%s' is not whitelisted",
1340                    drv->format_name);
1341         ret = -ENOTSUP;
1342         goto fail_opts;
1343     }
1344 
1345     /* bdrv_new() and bdrv_close() make it so */
1346     assert(atomic_read(&bs->copy_on_read) == 0);
1347 
1348     if (bs->open_flags & BDRV_O_COPY_ON_READ) {
1349         if (!bs->read_only) {
1350             bdrv_enable_copy_on_read(bs);
1351         } else {
1352             error_setg(errp, "Can't use copy-on-read on read-only device");
1353             ret = -EINVAL;
1354             goto fail_opts;
1355         }
1356     }
1357 
1358     discard = qemu_opt_get(opts, "discard");
1359     if (discard != NULL) {
1360         if (bdrv_parse_discard_flags(discard, &bs->open_flags) != 0) {
1361             error_setg(errp, "Invalid discard option");
1362             ret = -EINVAL;
1363             goto fail_opts;
1364         }
1365     }
1366 
1367     detect_zeroes = qemu_opt_get(opts, "detect-zeroes");
1368     if (detect_zeroes) {
1369         BlockdevDetectZeroesOptions value =
1370             qapi_enum_parse(&BlockdevDetectZeroesOptions_lookup,
1371                             detect_zeroes,
1372                             BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF,
1373                             &local_err);
1374         if (local_err) {
1375             error_propagate(errp, local_err);
1376             ret = -EINVAL;
1377             goto fail_opts;
1378         }
1379 
1380         if (value == BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP &&
1381             !(bs->open_flags & BDRV_O_UNMAP))
1382         {
1383             error_setg(errp, "setting detect-zeroes to unmap is not allowed "
1384                              "without setting discard operation to unmap");
1385             ret = -EINVAL;
1386             goto fail_opts;
1387         }
1388 
1389         bs->detect_zeroes = value;
1390     }
1391 
1392     if (filename != NULL) {
1393         pstrcpy(bs->filename, sizeof(bs->filename), filename);
1394     } else {
1395         bs->filename[0] = '\0';
1396     }
1397     pstrcpy(bs->exact_filename, sizeof(bs->exact_filename), bs->filename);
1398 
1399     /* Open the image, either directly or using a protocol */
1400     open_flags = bdrv_open_flags(bs, bs->open_flags);
1401     node_name = qemu_opt_get(opts, "node-name");
1402 
1403     assert(!drv->bdrv_file_open || file == NULL);
1404     ret = bdrv_open_driver(bs, drv, node_name, options, open_flags, errp);
1405     if (ret < 0) {
1406         goto fail_opts;
1407     }
1408 
1409     qemu_opts_del(opts);
1410     return 0;
1411 
1412 fail_opts:
1413     qemu_opts_del(opts);
1414     return ret;
1415 }
1416 
1417 static QDict *parse_json_filename(const char *filename, Error **errp)
1418 {
1419     QObject *options_obj;
1420     QDict *options;
1421     int ret;
1422 
1423     ret = strstart(filename, "json:", &filename);
1424     assert(ret);
1425 
1426     options_obj = qobject_from_json(filename, errp);
1427     if (!options_obj) {
1428         /* Work around qobject_from_json() lossage TODO fix that */
1429         if (errp && !*errp) {
1430             error_setg(errp, "Could not parse the JSON options");
1431             return NULL;
1432         }
1433         error_prepend(errp, "Could not parse the JSON options: ");
1434         return NULL;
1435     }
1436 
1437     options = qobject_to_qdict(options_obj);
1438     if (!options) {
1439         qobject_decref(options_obj);
1440         error_setg(errp, "Invalid JSON object given");
1441         return NULL;
1442     }
1443 
1444     qdict_flatten(options);
1445 
1446     return options;
1447 }
1448 
1449 static void parse_json_protocol(QDict *options, const char **pfilename,
1450                                 Error **errp)
1451 {
1452     QDict *json_options;
1453     Error *local_err = NULL;
1454 
1455     /* Parse json: pseudo-protocol */
1456     if (!*pfilename || !g_str_has_prefix(*pfilename, "json:")) {
1457         return;
1458     }
1459 
1460     json_options = parse_json_filename(*pfilename, &local_err);
1461     if (local_err) {
1462         error_propagate(errp, local_err);
1463         return;
1464     }
1465 
1466     /* Options given in the filename have lower priority than options
1467      * specified directly */
1468     qdict_join(options, json_options, false);
1469     QDECREF(json_options);
1470     *pfilename = NULL;
1471 }
1472 
1473 /*
1474  * Fills in default options for opening images and converts the legacy
1475  * filename/flags pair to option QDict entries.
1476  * The BDRV_O_PROTOCOL flag in *flags will be set or cleared accordingly if a
1477  * block driver has been specified explicitly.
1478  */
1479 static int bdrv_fill_options(QDict **options, const char *filename,
1480                              int *flags, Error **errp)
1481 {
1482     const char *drvname;
1483     bool protocol = *flags & BDRV_O_PROTOCOL;
1484     bool parse_filename = false;
1485     BlockDriver *drv = NULL;
1486     Error *local_err = NULL;
1487 
1488     /*
1489      * Caution: while qdict_get_try_str() is fine, getting non-string
1490      * types would require more care.  When @options come from
1491      * -blockdev or blockdev_add, its members are typed according to
1492      * the QAPI schema, but when they come from -drive, they're all
1493      * QString.
1494      */
1495     drvname = qdict_get_try_str(*options, "driver");
1496     if (drvname) {
1497         drv = bdrv_find_format(drvname);
1498         if (!drv) {
1499             error_setg(errp, "Unknown driver '%s'", drvname);
1500             return -ENOENT;
1501         }
1502         /* If the user has explicitly specified the driver, this choice should
1503          * override the BDRV_O_PROTOCOL flag */
1504         protocol = drv->bdrv_file_open;
1505     }
1506 
1507     if (protocol) {
1508         *flags |= BDRV_O_PROTOCOL;
1509     } else {
1510         *flags &= ~BDRV_O_PROTOCOL;
1511     }
1512 
1513     /* Translate cache options from flags into options */
1514     update_options_from_flags(*options, *flags);
1515 
1516     /* Fetch the file name from the options QDict if necessary */
1517     if (protocol && filename) {
1518         if (!qdict_haskey(*options, "filename")) {
1519             qdict_put_str(*options, "filename", filename);
1520             parse_filename = true;
1521         } else {
1522             error_setg(errp, "Can't specify 'file' and 'filename' options at "
1523                              "the same time");
1524             return -EINVAL;
1525         }
1526     }
1527 
1528     /* Find the right block driver */
1529     /* See cautionary note on accessing @options above */
1530     filename = qdict_get_try_str(*options, "filename");
1531 
1532     if (!drvname && protocol) {
1533         if (filename) {
1534             drv = bdrv_find_protocol(filename, parse_filename, errp);
1535             if (!drv) {
1536                 return -EINVAL;
1537             }
1538 
1539             drvname = drv->format_name;
1540             qdict_put_str(*options, "driver", drvname);
1541         } else {
1542             error_setg(errp, "Must specify either driver or file");
1543             return -EINVAL;
1544         }
1545     }
1546 
1547     assert(drv || !protocol);
1548 
1549     /* Driver-specific filename parsing */
1550     if (drv && drv->bdrv_parse_filename && parse_filename) {
1551         drv->bdrv_parse_filename(filename, *options, &local_err);
1552         if (local_err) {
1553             error_propagate(errp, local_err);
1554             return -EINVAL;
1555         }
1556 
1557         if (!drv->bdrv_needs_filename) {
1558             qdict_del(*options, "filename");
1559         }
1560     }
1561 
1562     return 0;
1563 }
1564 
1565 static int bdrv_child_check_perm(BdrvChild *c, BlockReopenQueue *q,
1566                                  uint64_t perm, uint64_t shared,
1567                                  GSList *ignore_children, Error **errp);
1568 static void bdrv_child_abort_perm_update(BdrvChild *c);
1569 static void bdrv_child_set_perm(BdrvChild *c, uint64_t perm, uint64_t shared);
1570 
1571 typedef struct BlockReopenQueueEntry {
1572      bool prepared;
1573      BDRVReopenState state;
1574      QSIMPLEQ_ENTRY(BlockReopenQueueEntry) entry;
1575 } BlockReopenQueueEntry;
1576 
1577 /*
1578  * Return the flags that @bs will have after the reopens in @q have
1579  * successfully completed. If @q is NULL (or @bs is not contained in @q),
1580  * return the current flags.
1581  */
1582 static int bdrv_reopen_get_flags(BlockReopenQueue *q, BlockDriverState *bs)
1583 {
1584     BlockReopenQueueEntry *entry;
1585 
1586     if (q != NULL) {
1587         QSIMPLEQ_FOREACH(entry, q, entry) {
1588             if (entry->state.bs == bs) {
1589                 return entry->state.flags;
1590             }
1591         }
1592     }
1593 
1594     return bs->open_flags;
1595 }
1596 
1597 /* Returns whether the image file can be written to after the reopen queue @q
1598  * has been successfully applied, or right now if @q is NULL. */
1599 static bool bdrv_is_writable(BlockDriverState *bs, BlockReopenQueue *q)
1600 {
1601     int flags = bdrv_reopen_get_flags(q, bs);
1602 
1603     return (flags & (BDRV_O_RDWR | BDRV_O_INACTIVE)) == BDRV_O_RDWR;
1604 }
1605 
1606 static void bdrv_child_perm(BlockDriverState *bs, BlockDriverState *child_bs,
1607                             BdrvChild *c, const BdrvChildRole *role,
1608                             BlockReopenQueue *reopen_queue,
1609                             uint64_t parent_perm, uint64_t parent_shared,
1610                             uint64_t *nperm, uint64_t *nshared)
1611 {
1612     if (bs->drv && bs->drv->bdrv_child_perm) {
1613         bs->drv->bdrv_child_perm(bs, c, role, reopen_queue,
1614                                  parent_perm, parent_shared,
1615                                  nperm, nshared);
1616     }
1617     /* TODO Take force_share from reopen_queue */
1618     if (child_bs && child_bs->force_share) {
1619         *nshared = BLK_PERM_ALL;
1620     }
1621 }
1622 
1623 /*
1624  * Check whether permissions on this node can be changed in a way that
1625  * @cumulative_perms and @cumulative_shared_perms are the new cumulative
1626  * permissions of all its parents. This involves checking whether all necessary
1627  * permission changes to child nodes can be performed.
1628  *
1629  * A call to this function must always be followed by a call to bdrv_set_perm()
1630  * or bdrv_abort_perm_update().
1631  */
1632 static int bdrv_check_perm(BlockDriverState *bs, BlockReopenQueue *q,
1633                            uint64_t cumulative_perms,
1634                            uint64_t cumulative_shared_perms,
1635                            GSList *ignore_children, Error **errp)
1636 {
1637     BlockDriver *drv = bs->drv;
1638     BdrvChild *c;
1639     int ret;
1640 
1641     /* Write permissions never work with read-only images */
1642     if ((cumulative_perms & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)) &&
1643         !bdrv_is_writable(bs, q))
1644     {
1645         error_setg(errp, "Block node is read-only");
1646         return -EPERM;
1647     }
1648 
1649     /* Check this node */
1650     if (!drv) {
1651         return 0;
1652     }
1653 
1654     if (drv->bdrv_check_perm) {
1655         return drv->bdrv_check_perm(bs, cumulative_perms,
1656                                     cumulative_shared_perms, errp);
1657     }
1658 
1659     /* Drivers that never have children can omit .bdrv_child_perm() */
1660     if (!drv->bdrv_child_perm) {
1661         assert(QLIST_EMPTY(&bs->children));
1662         return 0;
1663     }
1664 
1665     /* Check all children */
1666     QLIST_FOREACH(c, &bs->children, next) {
1667         uint64_t cur_perm, cur_shared;
1668         bdrv_child_perm(bs, c->bs, c, c->role, q,
1669                         cumulative_perms, cumulative_shared_perms,
1670                         &cur_perm, &cur_shared);
1671         ret = bdrv_child_check_perm(c, q, cur_perm, cur_shared,
1672                                     ignore_children, errp);
1673         if (ret < 0) {
1674             return ret;
1675         }
1676     }
1677 
1678     return 0;
1679 }
1680 
1681 /*
1682  * Notifies drivers that after a previous bdrv_check_perm() call, the
1683  * permission update is not performed and any preparations made for it (e.g.
1684  * taken file locks) need to be undone.
1685  *
1686  * This function recursively notifies all child nodes.
1687  */
1688 static void bdrv_abort_perm_update(BlockDriverState *bs)
1689 {
1690     BlockDriver *drv = bs->drv;
1691     BdrvChild *c;
1692 
1693     if (!drv) {
1694         return;
1695     }
1696 
1697     if (drv->bdrv_abort_perm_update) {
1698         drv->bdrv_abort_perm_update(bs);
1699     }
1700 
1701     QLIST_FOREACH(c, &bs->children, next) {
1702         bdrv_child_abort_perm_update(c);
1703     }
1704 }
1705 
1706 static void bdrv_set_perm(BlockDriverState *bs, uint64_t cumulative_perms,
1707                           uint64_t cumulative_shared_perms)
1708 {
1709     BlockDriver *drv = bs->drv;
1710     BdrvChild *c;
1711 
1712     if (!drv) {
1713         return;
1714     }
1715 
1716     /* Update this node */
1717     if (drv->bdrv_set_perm) {
1718         drv->bdrv_set_perm(bs, cumulative_perms, cumulative_shared_perms);
1719     }
1720 
1721     /* Drivers that never have children can omit .bdrv_child_perm() */
1722     if (!drv->bdrv_child_perm) {
1723         assert(QLIST_EMPTY(&bs->children));
1724         return;
1725     }
1726 
1727     /* Update all children */
1728     QLIST_FOREACH(c, &bs->children, next) {
1729         uint64_t cur_perm, cur_shared;
1730         bdrv_child_perm(bs, c->bs, c, c->role, NULL,
1731                         cumulative_perms, cumulative_shared_perms,
1732                         &cur_perm, &cur_shared);
1733         bdrv_child_set_perm(c, cur_perm, cur_shared);
1734     }
1735 }
1736 
1737 static void bdrv_get_cumulative_perm(BlockDriverState *bs, uint64_t *perm,
1738                                      uint64_t *shared_perm)
1739 {
1740     BdrvChild *c;
1741     uint64_t cumulative_perms = 0;
1742     uint64_t cumulative_shared_perms = BLK_PERM_ALL;
1743 
1744     QLIST_FOREACH(c, &bs->parents, next_parent) {
1745         cumulative_perms |= c->perm;
1746         cumulative_shared_perms &= c->shared_perm;
1747     }
1748 
1749     *perm = cumulative_perms;
1750     *shared_perm = cumulative_shared_perms;
1751 }
1752 
1753 static char *bdrv_child_user_desc(BdrvChild *c)
1754 {
1755     if (c->role->get_parent_desc) {
1756         return c->role->get_parent_desc(c);
1757     }
1758 
1759     return g_strdup("another user");
1760 }
1761 
1762 char *bdrv_perm_names(uint64_t perm)
1763 {
1764     struct perm_name {
1765         uint64_t perm;
1766         const char *name;
1767     } permissions[] = {
1768         { BLK_PERM_CONSISTENT_READ, "consistent read" },
1769         { BLK_PERM_WRITE,           "write" },
1770         { BLK_PERM_WRITE_UNCHANGED, "write unchanged" },
1771         { BLK_PERM_RESIZE,          "resize" },
1772         { BLK_PERM_GRAPH_MOD,       "change children" },
1773         { 0, NULL }
1774     };
1775 
1776     char *result = g_strdup("");
1777     struct perm_name *p;
1778 
1779     for (p = permissions; p->name; p++) {
1780         if (perm & p->perm) {
1781             char *old = result;
1782             result = g_strdup_printf("%s%s%s", old, *old ? ", " : "", p->name);
1783             g_free(old);
1784         }
1785     }
1786 
1787     return result;
1788 }
1789 
1790 /*
1791  * Checks whether a new reference to @bs can be added if the new user requires
1792  * @new_used_perm/@new_shared_perm as its permissions. If @ignore_children is
1793  * set, the BdrvChild objects in this list are ignored in the calculations;
1794  * this allows checking permission updates for an existing reference.
1795  *
1796  * Needs to be followed by a call to either bdrv_set_perm() or
1797  * bdrv_abort_perm_update(). */
1798 static int bdrv_check_update_perm(BlockDriverState *bs, BlockReopenQueue *q,
1799                                   uint64_t new_used_perm,
1800                                   uint64_t new_shared_perm,
1801                                   GSList *ignore_children, Error **errp)
1802 {
1803     BdrvChild *c;
1804     uint64_t cumulative_perms = new_used_perm;
1805     uint64_t cumulative_shared_perms = new_shared_perm;
1806 
1807     /* There is no reason why anyone couldn't tolerate write_unchanged */
1808     assert(new_shared_perm & BLK_PERM_WRITE_UNCHANGED);
1809 
1810     QLIST_FOREACH(c, &bs->parents, next_parent) {
1811         if (g_slist_find(ignore_children, c)) {
1812             continue;
1813         }
1814 
1815         if ((new_used_perm & c->shared_perm) != new_used_perm) {
1816             char *user = bdrv_child_user_desc(c);
1817             char *perm_names = bdrv_perm_names(new_used_perm & ~c->shared_perm);
1818             error_setg(errp, "Conflicts with use by %s as '%s', which does not "
1819                              "allow '%s' on %s",
1820                        user, c->name, perm_names, bdrv_get_node_name(c->bs));
1821             g_free(user);
1822             g_free(perm_names);
1823             return -EPERM;
1824         }
1825 
1826         if ((c->perm & new_shared_perm) != c->perm) {
1827             char *user = bdrv_child_user_desc(c);
1828             char *perm_names = bdrv_perm_names(c->perm & ~new_shared_perm);
1829             error_setg(errp, "Conflicts with use by %s as '%s', which uses "
1830                              "'%s' on %s",
1831                        user, c->name, perm_names, bdrv_get_node_name(c->bs));
1832             g_free(user);
1833             g_free(perm_names);
1834             return -EPERM;
1835         }
1836 
1837         cumulative_perms |= c->perm;
1838         cumulative_shared_perms &= c->shared_perm;
1839     }
1840 
1841     return bdrv_check_perm(bs, q, cumulative_perms, cumulative_shared_perms,
1842                            ignore_children, errp);
1843 }
1844 
1845 /* Needs to be followed by a call to either bdrv_child_set_perm() or
1846  * bdrv_child_abort_perm_update(). */
1847 static int bdrv_child_check_perm(BdrvChild *c, BlockReopenQueue *q,
1848                                  uint64_t perm, uint64_t shared,
1849                                  GSList *ignore_children, Error **errp)
1850 {
1851     int ret;
1852 
1853     ignore_children = g_slist_prepend(g_slist_copy(ignore_children), c);
1854     ret = bdrv_check_update_perm(c->bs, q, perm, shared, ignore_children, errp);
1855     g_slist_free(ignore_children);
1856 
1857     return ret;
1858 }
1859 
1860 static void bdrv_child_set_perm(BdrvChild *c, uint64_t perm, uint64_t shared)
1861 {
1862     uint64_t cumulative_perms, cumulative_shared_perms;
1863 
1864     c->perm = perm;
1865     c->shared_perm = shared;
1866 
1867     bdrv_get_cumulative_perm(c->bs, &cumulative_perms,
1868                              &cumulative_shared_perms);
1869     bdrv_set_perm(c->bs, cumulative_perms, cumulative_shared_perms);
1870 }
1871 
1872 static void bdrv_child_abort_perm_update(BdrvChild *c)
1873 {
1874     bdrv_abort_perm_update(c->bs);
1875 }
1876 
1877 int bdrv_child_try_set_perm(BdrvChild *c, uint64_t perm, uint64_t shared,
1878                             Error **errp)
1879 {
1880     int ret;
1881 
1882     ret = bdrv_child_check_perm(c, NULL, perm, shared, NULL, errp);
1883     if (ret < 0) {
1884         bdrv_child_abort_perm_update(c);
1885         return ret;
1886     }
1887 
1888     bdrv_child_set_perm(c, perm, shared);
1889 
1890     return 0;
1891 }
1892 
1893 #define DEFAULT_PERM_PASSTHROUGH (BLK_PERM_CONSISTENT_READ \
1894                                  | BLK_PERM_WRITE \
1895                                  | BLK_PERM_WRITE_UNCHANGED \
1896                                  | BLK_PERM_RESIZE)
1897 #define DEFAULT_PERM_UNCHANGED (BLK_PERM_ALL & ~DEFAULT_PERM_PASSTHROUGH)
1898 
1899 void bdrv_filter_default_perms(BlockDriverState *bs, BdrvChild *c,
1900                                const BdrvChildRole *role,
1901                                BlockReopenQueue *reopen_queue,
1902                                uint64_t perm, uint64_t shared,
1903                                uint64_t *nperm, uint64_t *nshared)
1904 {
1905     if (c == NULL) {
1906         *nperm = perm & DEFAULT_PERM_PASSTHROUGH;
1907         *nshared = (shared & DEFAULT_PERM_PASSTHROUGH) | DEFAULT_PERM_UNCHANGED;
1908         return;
1909     }
1910 
1911     *nperm = (perm & DEFAULT_PERM_PASSTHROUGH) |
1912              (c->perm & DEFAULT_PERM_UNCHANGED);
1913     *nshared = (shared & DEFAULT_PERM_PASSTHROUGH) |
1914                (c->shared_perm & DEFAULT_PERM_UNCHANGED);
1915 }
1916 
1917 void bdrv_format_default_perms(BlockDriverState *bs, BdrvChild *c,
1918                                const BdrvChildRole *role,
1919                                BlockReopenQueue *reopen_queue,
1920                                uint64_t perm, uint64_t shared,
1921                                uint64_t *nperm, uint64_t *nshared)
1922 {
1923     bool backing = (role == &child_backing);
1924     assert(role == &child_backing || role == &child_file);
1925 
1926     if (!backing) {
1927         /* Apart from the modifications below, the same permissions are
1928          * forwarded and left alone as for filters */
1929         bdrv_filter_default_perms(bs, c, role, reopen_queue, perm, shared,
1930                                   &perm, &shared);
1931 
1932         /* Format drivers may touch metadata even if the guest doesn't write */
1933         if (bdrv_is_writable(bs, reopen_queue)) {
1934             perm |= BLK_PERM_WRITE | BLK_PERM_RESIZE;
1935         }
1936 
1937         /* bs->file always needs to be consistent because of the metadata. We
1938          * can never allow other users to resize or write to it. */
1939         perm |= BLK_PERM_CONSISTENT_READ;
1940         shared &= ~(BLK_PERM_WRITE | BLK_PERM_RESIZE);
1941     } else {
1942         /* We want consistent read from backing files if the parent needs it.
1943          * No other operations are performed on backing files. */
1944         perm &= BLK_PERM_CONSISTENT_READ;
1945 
1946         /* If the parent can deal with changing data, we're okay with a
1947          * writable and resizable backing file. */
1948         /* TODO Require !(perm & BLK_PERM_CONSISTENT_READ), too? */
1949         if (shared & BLK_PERM_WRITE) {
1950             shared = BLK_PERM_WRITE | BLK_PERM_RESIZE;
1951         } else {
1952             shared = 0;
1953         }
1954 
1955         shared |= BLK_PERM_CONSISTENT_READ | BLK_PERM_GRAPH_MOD |
1956                   BLK_PERM_WRITE_UNCHANGED;
1957     }
1958 
1959     if (bs->open_flags & BDRV_O_INACTIVE) {
1960         shared |= BLK_PERM_WRITE | BLK_PERM_RESIZE;
1961     }
1962 
1963     *nperm = perm;
1964     *nshared = shared;
1965 }
1966 
1967 static void bdrv_replace_child_noperm(BdrvChild *child,
1968                                       BlockDriverState *new_bs)
1969 {
1970     BlockDriverState *old_bs = child->bs;
1971 
1972     if (old_bs && new_bs) {
1973         assert(bdrv_get_aio_context(old_bs) == bdrv_get_aio_context(new_bs));
1974     }
1975     if (old_bs) {
1976         if (old_bs->quiesce_counter && child->role->drained_end) {
1977             child->role->drained_end(child);
1978         }
1979         if (child->role->detach) {
1980             child->role->detach(child);
1981         }
1982         QLIST_REMOVE(child, next_parent);
1983     }
1984 
1985     child->bs = new_bs;
1986 
1987     if (new_bs) {
1988         QLIST_INSERT_HEAD(&new_bs->parents, child, next_parent);
1989         if (new_bs->quiesce_counter && child->role->drained_begin) {
1990             child->role->drained_begin(child);
1991         }
1992 
1993         if (child->role->attach) {
1994             child->role->attach(child);
1995         }
1996     }
1997 }
1998 
1999 /*
2000  * Updates @child to change its reference to point to @new_bs, including
2001  * checking and applying the necessary permisson updates both to the old node
2002  * and to @new_bs.
2003  *
2004  * NULL is passed as @new_bs for removing the reference before freeing @child.
2005  *
2006  * If @new_bs is not NULL, bdrv_check_perm() must be called beforehand, as this
2007  * function uses bdrv_set_perm() to update the permissions according to the new
2008  * reference that @new_bs gets.
2009  */
2010 static void bdrv_replace_child(BdrvChild *child, BlockDriverState *new_bs)
2011 {
2012     BlockDriverState *old_bs = child->bs;
2013     uint64_t perm, shared_perm;
2014 
2015     bdrv_replace_child_noperm(child, new_bs);
2016 
2017     if (old_bs) {
2018         /* Update permissions for old node. This is guaranteed to succeed
2019          * because we're just taking a parent away, so we're loosening
2020          * restrictions. */
2021         bdrv_get_cumulative_perm(old_bs, &perm, &shared_perm);
2022         bdrv_check_perm(old_bs, NULL, perm, shared_perm, NULL, &error_abort);
2023         bdrv_set_perm(old_bs, perm, shared_perm);
2024     }
2025 
2026     if (new_bs) {
2027         bdrv_get_cumulative_perm(new_bs, &perm, &shared_perm);
2028         bdrv_set_perm(new_bs, perm, shared_perm);
2029     }
2030 }
2031 
2032 BdrvChild *bdrv_root_attach_child(BlockDriverState *child_bs,
2033                                   const char *child_name,
2034                                   const BdrvChildRole *child_role,
2035                                   uint64_t perm, uint64_t shared_perm,
2036                                   void *opaque, Error **errp)
2037 {
2038     BdrvChild *child;
2039     int ret;
2040 
2041     ret = bdrv_check_update_perm(child_bs, NULL, perm, shared_perm, NULL, errp);
2042     if (ret < 0) {
2043         bdrv_abort_perm_update(child_bs);
2044         return NULL;
2045     }
2046 
2047     child = g_new(BdrvChild, 1);
2048     *child = (BdrvChild) {
2049         .bs             = NULL,
2050         .name           = g_strdup(child_name),
2051         .role           = child_role,
2052         .perm           = perm,
2053         .shared_perm    = shared_perm,
2054         .opaque         = opaque,
2055     };
2056 
2057     /* This performs the matching bdrv_set_perm() for the above check. */
2058     bdrv_replace_child(child, child_bs);
2059 
2060     return child;
2061 }
2062 
2063 BdrvChild *bdrv_attach_child(BlockDriverState *parent_bs,
2064                              BlockDriverState *child_bs,
2065                              const char *child_name,
2066                              const BdrvChildRole *child_role,
2067                              Error **errp)
2068 {
2069     BdrvChild *child;
2070     uint64_t perm, shared_perm;
2071 
2072     bdrv_get_cumulative_perm(parent_bs, &perm, &shared_perm);
2073 
2074     assert(parent_bs->drv);
2075     assert(bdrv_get_aio_context(parent_bs) == bdrv_get_aio_context(child_bs));
2076     bdrv_child_perm(parent_bs, child_bs, NULL, child_role, NULL,
2077                     perm, shared_perm, &perm, &shared_perm);
2078 
2079     child = bdrv_root_attach_child(child_bs, child_name, child_role,
2080                                    perm, shared_perm, parent_bs, errp);
2081     if (child == NULL) {
2082         return NULL;
2083     }
2084 
2085     QLIST_INSERT_HEAD(&parent_bs->children, child, next);
2086     return child;
2087 }
2088 
2089 static void bdrv_detach_child(BdrvChild *child)
2090 {
2091     if (child->next.le_prev) {
2092         QLIST_REMOVE(child, next);
2093         child->next.le_prev = NULL;
2094     }
2095 
2096     bdrv_replace_child(child, NULL);
2097 
2098     g_free(child->name);
2099     g_free(child);
2100 }
2101 
2102 void bdrv_root_unref_child(BdrvChild *child)
2103 {
2104     BlockDriverState *child_bs;
2105 
2106     child_bs = child->bs;
2107     bdrv_detach_child(child);
2108     bdrv_unref(child_bs);
2109 }
2110 
2111 void bdrv_unref_child(BlockDriverState *parent, BdrvChild *child)
2112 {
2113     if (child == NULL) {
2114         return;
2115     }
2116 
2117     if (child->bs->inherits_from == parent) {
2118         BdrvChild *c;
2119 
2120         /* Remove inherits_from only when the last reference between parent and
2121          * child->bs goes away. */
2122         QLIST_FOREACH(c, &parent->children, next) {
2123             if (c != child && c->bs == child->bs) {
2124                 break;
2125             }
2126         }
2127         if (c == NULL) {
2128             child->bs->inherits_from = NULL;
2129         }
2130     }
2131 
2132     bdrv_root_unref_child(child);
2133 }
2134 
2135 
2136 static void bdrv_parent_cb_change_media(BlockDriverState *bs, bool load)
2137 {
2138     BdrvChild *c;
2139     QLIST_FOREACH(c, &bs->parents, next_parent) {
2140         if (c->role->change_media) {
2141             c->role->change_media(c, load);
2142         }
2143     }
2144 }
2145 
2146 static void bdrv_parent_cb_resize(BlockDriverState *bs)
2147 {
2148     BdrvChild *c;
2149     QLIST_FOREACH(c, &bs->parents, next_parent) {
2150         if (c->role->resize) {
2151             c->role->resize(c);
2152         }
2153     }
2154 }
2155 
2156 /*
2157  * Sets the backing file link of a BDS. A new reference is created; callers
2158  * which don't need their own reference any more must call bdrv_unref().
2159  */
2160 void bdrv_set_backing_hd(BlockDriverState *bs, BlockDriverState *backing_hd,
2161                          Error **errp)
2162 {
2163     if (backing_hd) {
2164         bdrv_ref(backing_hd);
2165     }
2166 
2167     if (bs->backing) {
2168         bdrv_unref_child(bs, bs->backing);
2169     }
2170 
2171     if (!backing_hd) {
2172         bs->backing = NULL;
2173         goto out;
2174     }
2175 
2176     bs->backing = bdrv_attach_child(bs, backing_hd, "backing", &child_backing,
2177                                     errp);
2178     if (!bs->backing) {
2179         bdrv_unref(backing_hd);
2180     }
2181 
2182     bdrv_refresh_filename(bs);
2183 
2184 out:
2185     bdrv_refresh_limits(bs, NULL);
2186 }
2187 
2188 /*
2189  * Opens the backing file for a BlockDriverState if not yet open
2190  *
2191  * bdref_key specifies the key for the image's BlockdevRef in the options QDict.
2192  * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
2193  * itself, all options starting with "${bdref_key}." are considered part of the
2194  * BlockdevRef.
2195  *
2196  * TODO Can this be unified with bdrv_open_image()?
2197  */
2198 int bdrv_open_backing_file(BlockDriverState *bs, QDict *parent_options,
2199                            const char *bdref_key, Error **errp)
2200 {
2201     char *backing_filename = g_malloc0(PATH_MAX);
2202     char *bdref_key_dot;
2203     const char *reference = NULL;
2204     int ret = 0;
2205     BlockDriverState *backing_hd;
2206     QDict *options;
2207     QDict *tmp_parent_options = NULL;
2208     Error *local_err = NULL;
2209 
2210     if (bs->backing != NULL) {
2211         goto free_exit;
2212     }
2213 
2214     /* NULL means an empty set of options */
2215     if (parent_options == NULL) {
2216         tmp_parent_options = qdict_new();
2217         parent_options = tmp_parent_options;
2218     }
2219 
2220     bs->open_flags &= ~BDRV_O_NO_BACKING;
2221 
2222     bdref_key_dot = g_strdup_printf("%s.", bdref_key);
2223     qdict_extract_subqdict(parent_options, &options, bdref_key_dot);
2224     g_free(bdref_key_dot);
2225 
2226     /*
2227      * Caution: while qdict_get_try_str() is fine, getting non-string
2228      * types would require more care.  When @parent_options come from
2229      * -blockdev or blockdev_add, its members are typed according to
2230      * the QAPI schema, but when they come from -drive, they're all
2231      * QString.
2232      */
2233     reference = qdict_get_try_str(parent_options, bdref_key);
2234     if (reference || qdict_haskey(options, "file.filename")) {
2235         backing_filename[0] = '\0';
2236     } else if (bs->backing_file[0] == '\0' && qdict_size(options) == 0) {
2237         QDECREF(options);
2238         goto free_exit;
2239     } else {
2240         bdrv_get_full_backing_filename(bs, backing_filename, PATH_MAX,
2241                                        &local_err);
2242         if (local_err) {
2243             ret = -EINVAL;
2244             error_propagate(errp, local_err);
2245             QDECREF(options);
2246             goto free_exit;
2247         }
2248     }
2249 
2250     if (!bs->drv || !bs->drv->supports_backing) {
2251         ret = -EINVAL;
2252         error_setg(errp, "Driver doesn't support backing files");
2253         QDECREF(options);
2254         goto free_exit;
2255     }
2256 
2257     if (!reference &&
2258         bs->backing_format[0] != '\0' && !qdict_haskey(options, "driver")) {
2259         qdict_put_str(options, "driver", bs->backing_format);
2260     }
2261 
2262     backing_hd = bdrv_open_inherit(*backing_filename ? backing_filename : NULL,
2263                                    reference, options, 0, bs, &child_backing,
2264                                    errp);
2265     if (!backing_hd) {
2266         bs->open_flags |= BDRV_O_NO_BACKING;
2267         error_prepend(errp, "Could not open backing file: ");
2268         ret = -EINVAL;
2269         goto free_exit;
2270     }
2271     bdrv_set_aio_context(backing_hd, bdrv_get_aio_context(bs));
2272 
2273     /* Hook up the backing file link; drop our reference, bs owns the
2274      * backing_hd reference now */
2275     bdrv_set_backing_hd(bs, backing_hd, &local_err);
2276     bdrv_unref(backing_hd);
2277     if (local_err) {
2278         error_propagate(errp, local_err);
2279         ret = -EINVAL;
2280         goto free_exit;
2281     }
2282 
2283     qdict_del(parent_options, bdref_key);
2284 
2285 free_exit:
2286     g_free(backing_filename);
2287     QDECREF(tmp_parent_options);
2288     return ret;
2289 }
2290 
2291 static BlockDriverState *
2292 bdrv_open_child_bs(const char *filename, QDict *options, const char *bdref_key,
2293                    BlockDriverState *parent, const BdrvChildRole *child_role,
2294                    bool allow_none, Error **errp)
2295 {
2296     BlockDriverState *bs = NULL;
2297     QDict *image_options;
2298     char *bdref_key_dot;
2299     const char *reference;
2300 
2301     assert(child_role != NULL);
2302 
2303     bdref_key_dot = g_strdup_printf("%s.", bdref_key);
2304     qdict_extract_subqdict(options, &image_options, bdref_key_dot);
2305     g_free(bdref_key_dot);
2306 
2307     /*
2308      * Caution: while qdict_get_try_str() is fine, getting non-string
2309      * types would require more care.  When @options come from
2310      * -blockdev or blockdev_add, its members are typed according to
2311      * the QAPI schema, but when they come from -drive, they're all
2312      * QString.
2313      */
2314     reference = qdict_get_try_str(options, bdref_key);
2315     if (!filename && !reference && !qdict_size(image_options)) {
2316         if (!allow_none) {
2317             error_setg(errp, "A block device must be specified for \"%s\"",
2318                        bdref_key);
2319         }
2320         QDECREF(image_options);
2321         goto done;
2322     }
2323 
2324     bs = bdrv_open_inherit(filename, reference, image_options, 0,
2325                            parent, child_role, errp);
2326     if (!bs) {
2327         goto done;
2328     }
2329 
2330 done:
2331     qdict_del(options, bdref_key);
2332     return bs;
2333 }
2334 
2335 /*
2336  * Opens a disk image whose options are given as BlockdevRef in another block
2337  * device's options.
2338  *
2339  * If allow_none is true, no image will be opened if filename is false and no
2340  * BlockdevRef is given. NULL will be returned, but errp remains unset.
2341  *
2342  * bdrev_key specifies the key for the image's BlockdevRef in the options QDict.
2343  * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
2344  * itself, all options starting with "${bdref_key}." are considered part of the
2345  * BlockdevRef.
2346  *
2347  * The BlockdevRef will be removed from the options QDict.
2348  */
2349 BdrvChild *bdrv_open_child(const char *filename,
2350                            QDict *options, const char *bdref_key,
2351                            BlockDriverState *parent,
2352                            const BdrvChildRole *child_role,
2353                            bool allow_none, Error **errp)
2354 {
2355     BdrvChild *c;
2356     BlockDriverState *bs;
2357 
2358     bs = bdrv_open_child_bs(filename, options, bdref_key, parent, child_role,
2359                             allow_none, errp);
2360     if (bs == NULL) {
2361         return NULL;
2362     }
2363 
2364     c = bdrv_attach_child(parent, bs, bdref_key, child_role, errp);
2365     if (!c) {
2366         bdrv_unref(bs);
2367         return NULL;
2368     }
2369 
2370     return c;
2371 }
2372 
2373 static BlockDriverState *bdrv_append_temp_snapshot(BlockDriverState *bs,
2374                                                    int flags,
2375                                                    QDict *snapshot_options,
2376                                                    Error **errp)
2377 {
2378     /* TODO: extra byte is a hack to ensure MAX_PATH space on Windows. */
2379     char *tmp_filename = g_malloc0(PATH_MAX + 1);
2380     int64_t total_size;
2381     QemuOpts *opts = NULL;
2382     BlockDriverState *bs_snapshot = NULL;
2383     Error *local_err = NULL;
2384     int ret;
2385 
2386     /* if snapshot, we create a temporary backing file and open it
2387        instead of opening 'filename' directly */
2388 
2389     /* Get the required size from the image */
2390     total_size = bdrv_getlength(bs);
2391     if (total_size < 0) {
2392         error_setg_errno(errp, -total_size, "Could not get image size");
2393         goto out;
2394     }
2395 
2396     /* Create the temporary image */
2397     ret = get_tmp_filename(tmp_filename, PATH_MAX + 1);
2398     if (ret < 0) {
2399         error_setg_errno(errp, -ret, "Could not get temporary filename");
2400         goto out;
2401     }
2402 
2403     opts = qemu_opts_create(bdrv_qcow2.create_opts, NULL, 0,
2404                             &error_abort);
2405     qemu_opt_set_number(opts, BLOCK_OPT_SIZE, total_size, &error_abort);
2406     ret = bdrv_create(&bdrv_qcow2, tmp_filename, opts, errp);
2407     qemu_opts_del(opts);
2408     if (ret < 0) {
2409         error_prepend(errp, "Could not create temporary overlay '%s': ",
2410                       tmp_filename);
2411         goto out;
2412     }
2413 
2414     /* Prepare options QDict for the temporary file */
2415     qdict_put_str(snapshot_options, "file.driver", "file");
2416     qdict_put_str(snapshot_options, "file.filename", tmp_filename);
2417     qdict_put_str(snapshot_options, "driver", "qcow2");
2418 
2419     bs_snapshot = bdrv_open(NULL, NULL, snapshot_options, flags, errp);
2420     snapshot_options = NULL;
2421     if (!bs_snapshot) {
2422         goto out;
2423     }
2424 
2425     /* bdrv_append() consumes a strong reference to bs_snapshot
2426      * (i.e. it will call bdrv_unref() on it) even on error, so in
2427      * order to be able to return one, we have to increase
2428      * bs_snapshot's refcount here */
2429     bdrv_ref(bs_snapshot);
2430     bdrv_append(bs_snapshot, bs, &local_err);
2431     if (local_err) {
2432         error_propagate(errp, local_err);
2433         bs_snapshot = NULL;
2434         goto out;
2435     }
2436 
2437 out:
2438     QDECREF(snapshot_options);
2439     g_free(tmp_filename);
2440     return bs_snapshot;
2441 }
2442 
2443 /*
2444  * Opens a disk image (raw, qcow2, vmdk, ...)
2445  *
2446  * options is a QDict of options to pass to the block drivers, or NULL for an
2447  * empty set of options. The reference to the QDict belongs to the block layer
2448  * after the call (even on failure), so if the caller intends to reuse the
2449  * dictionary, it needs to use QINCREF() before calling bdrv_open.
2450  *
2451  * If *pbs is NULL, a new BDS will be created with a pointer to it stored there.
2452  * If it is not NULL, the referenced BDS will be reused.
2453  *
2454  * The reference parameter may be used to specify an existing block device which
2455  * should be opened. If specified, neither options nor a filename may be given,
2456  * nor can an existing BDS be reused (that is, *pbs has to be NULL).
2457  */
2458 static BlockDriverState *bdrv_open_inherit(const char *filename,
2459                                            const char *reference,
2460                                            QDict *options, int flags,
2461                                            BlockDriverState *parent,
2462                                            const BdrvChildRole *child_role,
2463                                            Error **errp)
2464 {
2465     int ret;
2466     BlockBackend *file = NULL;
2467     BlockDriverState *bs;
2468     BlockDriver *drv = NULL;
2469     const char *drvname;
2470     const char *backing;
2471     Error *local_err = NULL;
2472     QDict *snapshot_options = NULL;
2473     int snapshot_flags = 0;
2474 
2475     assert(!child_role || !flags);
2476     assert(!child_role == !parent);
2477 
2478     if (reference) {
2479         bool options_non_empty = options ? qdict_size(options) : false;
2480         QDECREF(options);
2481 
2482         if (filename || options_non_empty) {
2483             error_setg(errp, "Cannot reference an existing block device with "
2484                        "additional options or a new filename");
2485             return NULL;
2486         }
2487 
2488         bs = bdrv_lookup_bs(reference, reference, errp);
2489         if (!bs) {
2490             return NULL;
2491         }
2492 
2493         bdrv_ref(bs);
2494         return bs;
2495     }
2496 
2497     bs = bdrv_new();
2498 
2499     /* NULL means an empty set of options */
2500     if (options == NULL) {
2501         options = qdict_new();
2502     }
2503 
2504     /* json: syntax counts as explicit options, as if in the QDict */
2505     parse_json_protocol(options, &filename, &local_err);
2506     if (local_err) {
2507         goto fail;
2508     }
2509 
2510     bs->explicit_options = qdict_clone_shallow(options);
2511 
2512     if (child_role) {
2513         bs->inherits_from = parent;
2514         child_role->inherit_options(&flags, options,
2515                                     parent->open_flags, parent->options);
2516     }
2517 
2518     ret = bdrv_fill_options(&options, filename, &flags, &local_err);
2519     if (local_err) {
2520         goto fail;
2521     }
2522 
2523     /*
2524      * Set the BDRV_O_RDWR and BDRV_O_ALLOW_RDWR flags.
2525      * Caution: getting a boolean member of @options requires care.
2526      * When @options come from -blockdev or blockdev_add, members are
2527      * typed according to the QAPI schema, but when they come from
2528      * -drive, they're all QString.
2529      */
2530     if (g_strcmp0(qdict_get_try_str(options, BDRV_OPT_READ_ONLY), "on") &&
2531         !qdict_get_try_bool(options, BDRV_OPT_READ_ONLY, false)) {
2532         flags |= (BDRV_O_RDWR | BDRV_O_ALLOW_RDWR);
2533     } else {
2534         flags &= ~BDRV_O_RDWR;
2535     }
2536 
2537     if (flags & BDRV_O_SNAPSHOT) {
2538         snapshot_options = qdict_new();
2539         bdrv_temp_snapshot_options(&snapshot_flags, snapshot_options,
2540                                    flags, options);
2541         /* Let bdrv_backing_options() override "read-only" */
2542         qdict_del(options, BDRV_OPT_READ_ONLY);
2543         bdrv_backing_options(&flags, options, flags, options);
2544     }
2545 
2546     bs->open_flags = flags;
2547     bs->options = options;
2548     options = qdict_clone_shallow(options);
2549 
2550     /* Find the right image format driver */
2551     /* See cautionary note on accessing @options above */
2552     drvname = qdict_get_try_str(options, "driver");
2553     if (drvname) {
2554         drv = bdrv_find_format(drvname);
2555         if (!drv) {
2556             error_setg(errp, "Unknown driver: '%s'", drvname);
2557             goto fail;
2558         }
2559     }
2560 
2561     assert(drvname || !(flags & BDRV_O_PROTOCOL));
2562 
2563     /* See cautionary note on accessing @options above */
2564     backing = qdict_get_try_str(options, "backing");
2565     if (backing && *backing == '\0') {
2566         flags |= BDRV_O_NO_BACKING;
2567         qdict_del(options, "backing");
2568     }
2569 
2570     /* Open image file without format layer. This BlockBackend is only used for
2571      * probing, the block drivers will do their own bdrv_open_child() for the
2572      * same BDS, which is why we put the node name back into options. */
2573     if ((flags & BDRV_O_PROTOCOL) == 0) {
2574         BlockDriverState *file_bs;
2575 
2576         file_bs = bdrv_open_child_bs(filename, options, "file", bs,
2577                                      &child_file, true, &local_err);
2578         if (local_err) {
2579             goto fail;
2580         }
2581         if (file_bs != NULL) {
2582             /* Not requesting BLK_PERM_CONSISTENT_READ because we're only
2583              * looking at the header to guess the image format. This works even
2584              * in cases where a guest would not see a consistent state. */
2585             file = blk_new(0, BLK_PERM_ALL);
2586             blk_insert_bs(file, file_bs, &local_err);
2587             bdrv_unref(file_bs);
2588             if (local_err) {
2589                 goto fail;
2590             }
2591 
2592             qdict_put_str(options, "file", bdrv_get_node_name(file_bs));
2593         }
2594     }
2595 
2596     /* Image format probing */
2597     bs->probed = !drv;
2598     if (!drv && file) {
2599         ret = find_image_format(file, filename, &drv, &local_err);
2600         if (ret < 0) {
2601             goto fail;
2602         }
2603         /*
2604          * This option update would logically belong in bdrv_fill_options(),
2605          * but we first need to open bs->file for the probing to work, while
2606          * opening bs->file already requires the (mostly) final set of options
2607          * so that cache mode etc. can be inherited.
2608          *
2609          * Adding the driver later is somewhat ugly, but it's not an option
2610          * that would ever be inherited, so it's correct. We just need to make
2611          * sure to update both bs->options (which has the full effective
2612          * options for bs) and options (which has file.* already removed).
2613          */
2614         qdict_put_str(bs->options, "driver", drv->format_name);
2615         qdict_put_str(options, "driver", drv->format_name);
2616     } else if (!drv) {
2617         error_setg(errp, "Must specify either driver or file");
2618         goto fail;
2619     }
2620 
2621     /* BDRV_O_PROTOCOL must be set iff a protocol BDS is about to be created */
2622     assert(!!(flags & BDRV_O_PROTOCOL) == !!drv->bdrv_file_open);
2623     /* file must be NULL if a protocol BDS is about to be created
2624      * (the inverse results in an error message from bdrv_open_common()) */
2625     assert(!(flags & BDRV_O_PROTOCOL) || !file);
2626 
2627     /* Open the image */
2628     ret = bdrv_open_common(bs, file, options, &local_err);
2629     if (ret < 0) {
2630         goto fail;
2631     }
2632 
2633     if (file) {
2634         blk_unref(file);
2635         file = NULL;
2636     }
2637 
2638     /* If there is a backing file, use it */
2639     if ((flags & BDRV_O_NO_BACKING) == 0) {
2640         ret = bdrv_open_backing_file(bs, options, "backing", &local_err);
2641         if (ret < 0) {
2642             goto close_and_fail;
2643         }
2644     }
2645 
2646     bdrv_refresh_filename(bs);
2647 
2648     /* Check if any unknown options were used */
2649     if (qdict_size(options) != 0) {
2650         const QDictEntry *entry = qdict_first(options);
2651         if (flags & BDRV_O_PROTOCOL) {
2652             error_setg(errp, "Block protocol '%s' doesn't support the option "
2653                        "'%s'", drv->format_name, entry->key);
2654         } else {
2655             error_setg(errp,
2656                        "Block format '%s' does not support the option '%s'",
2657                        drv->format_name, entry->key);
2658         }
2659 
2660         goto close_and_fail;
2661     }
2662 
2663     bdrv_parent_cb_change_media(bs, true);
2664 
2665     QDECREF(options);
2666 
2667     /* For snapshot=on, create a temporary qcow2 overlay. bs points to the
2668      * temporary snapshot afterwards. */
2669     if (snapshot_flags) {
2670         BlockDriverState *snapshot_bs;
2671         snapshot_bs = bdrv_append_temp_snapshot(bs, snapshot_flags,
2672                                                 snapshot_options, &local_err);
2673         snapshot_options = NULL;
2674         if (local_err) {
2675             goto close_and_fail;
2676         }
2677         /* We are not going to return bs but the overlay on top of it
2678          * (snapshot_bs); thus, we have to drop the strong reference to bs
2679          * (which we obtained by calling bdrv_new()). bs will not be deleted,
2680          * though, because the overlay still has a reference to it. */
2681         bdrv_unref(bs);
2682         bs = snapshot_bs;
2683     }
2684 
2685     return bs;
2686 
2687 fail:
2688     blk_unref(file);
2689     QDECREF(snapshot_options);
2690     QDECREF(bs->explicit_options);
2691     QDECREF(bs->options);
2692     QDECREF(options);
2693     bs->options = NULL;
2694     bs->explicit_options = NULL;
2695     bdrv_unref(bs);
2696     error_propagate(errp, local_err);
2697     return NULL;
2698 
2699 close_and_fail:
2700     bdrv_unref(bs);
2701     QDECREF(snapshot_options);
2702     QDECREF(options);
2703     error_propagate(errp, local_err);
2704     return NULL;
2705 }
2706 
2707 BlockDriverState *bdrv_open(const char *filename, const char *reference,
2708                             QDict *options, int flags, Error **errp)
2709 {
2710     return bdrv_open_inherit(filename, reference, options, flags, NULL,
2711                              NULL, errp);
2712 }
2713 
2714 /*
2715  * Adds a BlockDriverState to a simple queue for an atomic, transactional
2716  * reopen of multiple devices.
2717  *
2718  * bs_queue can either be an existing BlockReopenQueue that has had QSIMPLE_INIT
2719  * already performed, or alternatively may be NULL a new BlockReopenQueue will
2720  * be created and initialized. This newly created BlockReopenQueue should be
2721  * passed back in for subsequent calls that are intended to be of the same
2722  * atomic 'set'.
2723  *
2724  * bs is the BlockDriverState to add to the reopen queue.
2725  *
2726  * options contains the changed options for the associated bs
2727  * (the BlockReopenQueue takes ownership)
2728  *
2729  * flags contains the open flags for the associated bs
2730  *
2731  * returns a pointer to bs_queue, which is either the newly allocated
2732  * bs_queue, or the existing bs_queue being used.
2733  *
2734  */
2735 static BlockReopenQueue *bdrv_reopen_queue_child(BlockReopenQueue *bs_queue,
2736                                                  BlockDriverState *bs,
2737                                                  QDict *options,
2738                                                  int flags,
2739                                                  const BdrvChildRole *role,
2740                                                  QDict *parent_options,
2741                                                  int parent_flags)
2742 {
2743     assert(bs != NULL);
2744 
2745     BlockReopenQueueEntry *bs_entry;
2746     BdrvChild *child;
2747     QDict *old_options, *explicit_options;
2748 
2749     if (bs_queue == NULL) {
2750         bs_queue = g_new0(BlockReopenQueue, 1);
2751         QSIMPLEQ_INIT(bs_queue);
2752     }
2753 
2754     if (!options) {
2755         options = qdict_new();
2756     }
2757 
2758     /* Check if this BlockDriverState is already in the queue */
2759     QSIMPLEQ_FOREACH(bs_entry, bs_queue, entry) {
2760         if (bs == bs_entry->state.bs) {
2761             break;
2762         }
2763     }
2764 
2765     /*
2766      * Precedence of options:
2767      * 1. Explicitly passed in options (highest)
2768      * 2. Set in flags (only for top level)
2769      * 3. Retained from explicitly set options of bs
2770      * 4. Inherited from parent node
2771      * 5. Retained from effective options of bs
2772      */
2773 
2774     if (!parent_options) {
2775         /*
2776          * Any setting represented by flags is always updated. If the
2777          * corresponding QDict option is set, it takes precedence. Otherwise
2778          * the flag is translated into a QDict option. The old setting of bs is
2779          * not considered.
2780          */
2781         update_options_from_flags(options, flags);
2782     }
2783 
2784     /* Old explicitly set values (don't overwrite by inherited value) */
2785     if (bs_entry) {
2786         old_options = qdict_clone_shallow(bs_entry->state.explicit_options);
2787     } else {
2788         old_options = qdict_clone_shallow(bs->explicit_options);
2789     }
2790     bdrv_join_options(bs, options, old_options);
2791     QDECREF(old_options);
2792 
2793     explicit_options = qdict_clone_shallow(options);
2794 
2795     /* Inherit from parent node */
2796     if (parent_options) {
2797         assert(!flags);
2798         role->inherit_options(&flags, options, parent_flags, parent_options);
2799     }
2800 
2801     /* Old values are used for options that aren't set yet */
2802     old_options = qdict_clone_shallow(bs->options);
2803     bdrv_join_options(bs, options, old_options);
2804     QDECREF(old_options);
2805 
2806     /* bdrv_open_inherit() sets and clears some additional flags internally */
2807     flags &= ~BDRV_O_PROTOCOL;
2808     if (flags & BDRV_O_RDWR) {
2809         flags |= BDRV_O_ALLOW_RDWR;
2810     }
2811 
2812     if (!bs_entry) {
2813         bs_entry = g_new0(BlockReopenQueueEntry, 1);
2814         QSIMPLEQ_INSERT_TAIL(bs_queue, bs_entry, entry);
2815     } else {
2816         QDECREF(bs_entry->state.options);
2817         QDECREF(bs_entry->state.explicit_options);
2818     }
2819 
2820     bs_entry->state.bs = bs;
2821     bs_entry->state.options = options;
2822     bs_entry->state.explicit_options = explicit_options;
2823     bs_entry->state.flags = flags;
2824 
2825     /* This needs to be overwritten in bdrv_reopen_prepare() */
2826     bs_entry->state.perm = UINT64_MAX;
2827     bs_entry->state.shared_perm = 0;
2828 
2829     QLIST_FOREACH(child, &bs->children, next) {
2830         QDict *new_child_options;
2831         char *child_key_dot;
2832 
2833         /* reopen can only change the options of block devices that were
2834          * implicitly created and inherited options. For other (referenced)
2835          * block devices, a syntax like "backing.foo" results in an error. */
2836         if (child->bs->inherits_from != bs) {
2837             continue;
2838         }
2839 
2840         child_key_dot = g_strdup_printf("%s.", child->name);
2841         qdict_extract_subqdict(options, &new_child_options, child_key_dot);
2842         g_free(child_key_dot);
2843 
2844         bdrv_reopen_queue_child(bs_queue, child->bs, new_child_options, 0,
2845                                 child->role, options, flags);
2846     }
2847 
2848     return bs_queue;
2849 }
2850 
2851 BlockReopenQueue *bdrv_reopen_queue(BlockReopenQueue *bs_queue,
2852                                     BlockDriverState *bs,
2853                                     QDict *options, int flags)
2854 {
2855     return bdrv_reopen_queue_child(bs_queue, bs, options, flags,
2856                                    NULL, NULL, 0);
2857 }
2858 
2859 /*
2860  * Reopen multiple BlockDriverStates atomically & transactionally.
2861  *
2862  * The queue passed in (bs_queue) must have been built up previous
2863  * via bdrv_reopen_queue().
2864  *
2865  * Reopens all BDS specified in the queue, with the appropriate
2866  * flags.  All devices are prepared for reopen, and failure of any
2867  * device will cause all device changes to be abandonded, and intermediate
2868  * data cleaned up.
2869  *
2870  * If all devices prepare successfully, then the changes are committed
2871  * to all devices.
2872  *
2873  */
2874 int bdrv_reopen_multiple(AioContext *ctx, BlockReopenQueue *bs_queue, Error **errp)
2875 {
2876     int ret = -1;
2877     BlockReopenQueueEntry *bs_entry, *next;
2878     Error *local_err = NULL;
2879 
2880     assert(bs_queue != NULL);
2881 
2882     aio_context_release(ctx);
2883     bdrv_drain_all_begin();
2884     aio_context_acquire(ctx);
2885 
2886     QSIMPLEQ_FOREACH(bs_entry, bs_queue, entry) {
2887         if (bdrv_reopen_prepare(&bs_entry->state, bs_queue, &local_err)) {
2888             error_propagate(errp, local_err);
2889             goto cleanup;
2890         }
2891         bs_entry->prepared = true;
2892     }
2893 
2894     /* If we reach this point, we have success and just need to apply the
2895      * changes
2896      */
2897     QSIMPLEQ_FOREACH(bs_entry, bs_queue, entry) {
2898         bdrv_reopen_commit(&bs_entry->state);
2899     }
2900 
2901     ret = 0;
2902 
2903 cleanup:
2904     QSIMPLEQ_FOREACH_SAFE(bs_entry, bs_queue, entry, next) {
2905         if (ret && bs_entry->prepared) {
2906             bdrv_reopen_abort(&bs_entry->state);
2907         } else if (ret) {
2908             QDECREF(bs_entry->state.explicit_options);
2909         }
2910         QDECREF(bs_entry->state.options);
2911         g_free(bs_entry);
2912     }
2913     g_free(bs_queue);
2914 
2915     bdrv_drain_all_end();
2916 
2917     return ret;
2918 }
2919 
2920 
2921 /* Reopen a single BlockDriverState with the specified flags. */
2922 int bdrv_reopen(BlockDriverState *bs, int bdrv_flags, Error **errp)
2923 {
2924     int ret = -1;
2925     Error *local_err = NULL;
2926     BlockReopenQueue *queue = bdrv_reopen_queue(NULL, bs, NULL, bdrv_flags);
2927 
2928     ret = bdrv_reopen_multiple(bdrv_get_aio_context(bs), queue, &local_err);
2929     if (local_err != NULL) {
2930         error_propagate(errp, local_err);
2931     }
2932     return ret;
2933 }
2934 
2935 static BlockReopenQueueEntry *find_parent_in_reopen_queue(BlockReopenQueue *q,
2936                                                           BdrvChild *c)
2937 {
2938     BlockReopenQueueEntry *entry;
2939 
2940     QSIMPLEQ_FOREACH(entry, q, entry) {
2941         BlockDriverState *bs = entry->state.bs;
2942         BdrvChild *child;
2943 
2944         QLIST_FOREACH(child, &bs->children, next) {
2945             if (child == c) {
2946                 return entry;
2947             }
2948         }
2949     }
2950 
2951     return NULL;
2952 }
2953 
2954 static void bdrv_reopen_perm(BlockReopenQueue *q, BlockDriverState *bs,
2955                              uint64_t *perm, uint64_t *shared)
2956 {
2957     BdrvChild *c;
2958     BlockReopenQueueEntry *parent;
2959     uint64_t cumulative_perms = 0;
2960     uint64_t cumulative_shared_perms = BLK_PERM_ALL;
2961 
2962     QLIST_FOREACH(c, &bs->parents, next_parent) {
2963         parent = find_parent_in_reopen_queue(q, c);
2964         if (!parent) {
2965             cumulative_perms |= c->perm;
2966             cumulative_shared_perms &= c->shared_perm;
2967         } else {
2968             uint64_t nperm, nshared;
2969 
2970             bdrv_child_perm(parent->state.bs, bs, c, c->role, q,
2971                             parent->state.perm, parent->state.shared_perm,
2972                             &nperm, &nshared);
2973 
2974             cumulative_perms |= nperm;
2975             cumulative_shared_perms &= nshared;
2976         }
2977     }
2978     *perm = cumulative_perms;
2979     *shared = cumulative_shared_perms;
2980 }
2981 
2982 /*
2983  * Prepares a BlockDriverState for reopen. All changes are staged in the
2984  * 'opaque' field of the BDRVReopenState, which is used and allocated by
2985  * the block driver layer .bdrv_reopen_prepare()
2986  *
2987  * bs is the BlockDriverState to reopen
2988  * flags are the new open flags
2989  * queue is the reopen queue
2990  *
2991  * Returns 0 on success, non-zero on error.  On error errp will be set
2992  * as well.
2993  *
2994  * On failure, bdrv_reopen_abort() will be called to clean up any data.
2995  * It is the responsibility of the caller to then call the abort() or
2996  * commit() for any other BDS that have been left in a prepare() state
2997  *
2998  */
2999 int bdrv_reopen_prepare(BDRVReopenState *reopen_state, BlockReopenQueue *queue,
3000                         Error **errp)
3001 {
3002     int ret = -1;
3003     Error *local_err = NULL;
3004     BlockDriver *drv;
3005     QemuOpts *opts;
3006     const char *value;
3007     bool read_only;
3008 
3009     assert(reopen_state != NULL);
3010     assert(reopen_state->bs->drv != NULL);
3011     drv = reopen_state->bs->drv;
3012 
3013     /* Process generic block layer options */
3014     opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
3015     qemu_opts_absorb_qdict(opts, reopen_state->options, &local_err);
3016     if (local_err) {
3017         error_propagate(errp, local_err);
3018         ret = -EINVAL;
3019         goto error;
3020     }
3021 
3022     update_flags_from_options(&reopen_state->flags, opts);
3023 
3024     /* node-name and driver must be unchanged. Put them back into the QDict, so
3025      * that they are checked at the end of this function. */
3026     value = qemu_opt_get(opts, "node-name");
3027     if (value) {
3028         qdict_put_str(reopen_state->options, "node-name", value);
3029     }
3030 
3031     value = qemu_opt_get(opts, "driver");
3032     if (value) {
3033         qdict_put_str(reopen_state->options, "driver", value);
3034     }
3035 
3036     /* If we are to stay read-only, do not allow permission change
3037      * to r/w. Attempting to set to r/w may fail if either BDRV_O_ALLOW_RDWR is
3038      * not set, or if the BDS still has copy_on_read enabled */
3039     read_only = !(reopen_state->flags & BDRV_O_RDWR);
3040     ret = bdrv_can_set_read_only(reopen_state->bs, read_only, true, &local_err);
3041     if (local_err) {
3042         error_propagate(errp, local_err);
3043         goto error;
3044     }
3045 
3046     /* Calculate required permissions after reopening */
3047     bdrv_reopen_perm(queue, reopen_state->bs,
3048                      &reopen_state->perm, &reopen_state->shared_perm);
3049 
3050     ret = bdrv_flush(reopen_state->bs);
3051     if (ret) {
3052         error_setg_errno(errp, -ret, "Error flushing drive");
3053         goto error;
3054     }
3055 
3056     if (drv->bdrv_reopen_prepare) {
3057         ret = drv->bdrv_reopen_prepare(reopen_state, queue, &local_err);
3058         if (ret) {
3059             if (local_err != NULL) {
3060                 error_propagate(errp, local_err);
3061             } else {
3062                 error_setg(errp, "failed while preparing to reopen image '%s'",
3063                            reopen_state->bs->filename);
3064             }
3065             goto error;
3066         }
3067     } else {
3068         /* It is currently mandatory to have a bdrv_reopen_prepare()
3069          * handler for each supported drv. */
3070         error_setg(errp, "Block format '%s' used by node '%s' "
3071                    "does not support reopening files", drv->format_name,
3072                    bdrv_get_device_or_node_name(reopen_state->bs));
3073         ret = -1;
3074         goto error;
3075     }
3076 
3077     /* Options that are not handled are only okay if they are unchanged
3078      * compared to the old state. It is expected that some options are only
3079      * used for the initial open, but not reopen (e.g. filename) */
3080     if (qdict_size(reopen_state->options)) {
3081         const QDictEntry *entry = qdict_first(reopen_state->options);
3082 
3083         do {
3084             QObject *new = entry->value;
3085             QObject *old = qdict_get(reopen_state->bs->options, entry->key);
3086 
3087             /*
3088              * TODO: When using -drive to specify blockdev options, all values
3089              * will be strings; however, when using -blockdev, blockdev-add or
3090              * filenames using the json:{} pseudo-protocol, they will be
3091              * correctly typed.
3092              * In contrast, reopening options are (currently) always strings
3093              * (because you can only specify them through qemu-io; all other
3094              * callers do not specify any options).
3095              * Therefore, when using anything other than -drive to create a BDS,
3096              * this cannot detect non-string options as unchanged, because
3097              * qobject_is_equal() always returns false for objects of different
3098              * type.  In the future, this should be remedied by correctly typing
3099              * all options.  For now, this is not too big of an issue because
3100              * the user can simply omit options which cannot be changed anyway,
3101              * so they will stay unchanged.
3102              */
3103             if (!qobject_is_equal(new, old)) {
3104                 error_setg(errp, "Cannot change the option '%s'", entry->key);
3105                 ret = -EINVAL;
3106                 goto error;
3107             }
3108         } while ((entry = qdict_next(reopen_state->options, entry)));
3109     }
3110 
3111     ret = bdrv_check_perm(reopen_state->bs, queue, reopen_state->perm,
3112                           reopen_state->shared_perm, NULL, errp);
3113     if (ret < 0) {
3114         goto error;
3115     }
3116 
3117     ret = 0;
3118 
3119 error:
3120     qemu_opts_del(opts);
3121     return ret;
3122 }
3123 
3124 /*
3125  * Takes the staged changes for the reopen from bdrv_reopen_prepare(), and
3126  * makes them final by swapping the staging BlockDriverState contents into
3127  * the active BlockDriverState contents.
3128  */
3129 void bdrv_reopen_commit(BDRVReopenState *reopen_state)
3130 {
3131     BlockDriver *drv;
3132     BlockDriverState *bs;
3133     bool old_can_write, new_can_write;
3134 
3135     assert(reopen_state != NULL);
3136     bs = reopen_state->bs;
3137     drv = bs->drv;
3138     assert(drv != NULL);
3139 
3140     old_can_write =
3141         !bdrv_is_read_only(bs) && !(bdrv_get_flags(bs) & BDRV_O_INACTIVE);
3142 
3143     /* If there are any driver level actions to take */
3144     if (drv->bdrv_reopen_commit) {
3145         drv->bdrv_reopen_commit(reopen_state);
3146     }
3147 
3148     /* set BDS specific flags now */
3149     QDECREF(bs->explicit_options);
3150 
3151     bs->explicit_options   = reopen_state->explicit_options;
3152     bs->open_flags         = reopen_state->flags;
3153     bs->read_only = !(reopen_state->flags & BDRV_O_RDWR);
3154 
3155     bdrv_refresh_limits(bs, NULL);
3156 
3157     bdrv_set_perm(reopen_state->bs, reopen_state->perm,
3158                   reopen_state->shared_perm);
3159 
3160     new_can_write =
3161         !bdrv_is_read_only(bs) && !(bdrv_get_flags(bs) & BDRV_O_INACTIVE);
3162     if (!old_can_write && new_can_write && drv->bdrv_reopen_bitmaps_rw) {
3163         Error *local_err = NULL;
3164         if (drv->bdrv_reopen_bitmaps_rw(bs, &local_err) < 0) {
3165             /* This is not fatal, bitmaps just left read-only, so all following
3166              * writes will fail. User can remove read-only bitmaps to unblock
3167              * writes.
3168              */
3169             error_reportf_err(local_err,
3170                               "%s: Failed to make dirty bitmaps writable: ",
3171                               bdrv_get_node_name(bs));
3172         }
3173     }
3174 }
3175 
3176 /*
3177  * Abort the reopen, and delete and free the staged changes in
3178  * reopen_state
3179  */
3180 void bdrv_reopen_abort(BDRVReopenState *reopen_state)
3181 {
3182     BlockDriver *drv;
3183 
3184     assert(reopen_state != NULL);
3185     drv = reopen_state->bs->drv;
3186     assert(drv != NULL);
3187 
3188     if (drv->bdrv_reopen_abort) {
3189         drv->bdrv_reopen_abort(reopen_state);
3190     }
3191 
3192     QDECREF(reopen_state->explicit_options);
3193 
3194     bdrv_abort_perm_update(reopen_state->bs);
3195 }
3196 
3197 
3198 static void bdrv_close(BlockDriverState *bs)
3199 {
3200     BdrvAioNotifier *ban, *ban_next;
3201     BdrvChild *child, *next;
3202 
3203     assert(!bs->job);
3204     assert(!bs->refcnt);
3205 
3206     bdrv_drained_begin(bs); /* complete I/O */
3207     bdrv_flush(bs);
3208     bdrv_drain(bs); /* in case flush left pending I/O */
3209 
3210     if (bs->drv) {
3211         bs->drv->bdrv_close(bs);
3212         bs->drv = NULL;
3213     }
3214 
3215     bdrv_set_backing_hd(bs, NULL, &error_abort);
3216 
3217     if (bs->file != NULL) {
3218         bdrv_unref_child(bs, bs->file);
3219         bs->file = NULL;
3220     }
3221 
3222     QLIST_FOREACH_SAFE(child, &bs->children, next, next) {
3223         /* TODO Remove bdrv_unref() from drivers' close function and use
3224          * bdrv_unref_child() here */
3225         if (child->bs->inherits_from == bs) {
3226             child->bs->inherits_from = NULL;
3227         }
3228         bdrv_detach_child(child);
3229     }
3230 
3231     g_free(bs->opaque);
3232     bs->opaque = NULL;
3233     atomic_set(&bs->copy_on_read, 0);
3234     bs->backing_file[0] = '\0';
3235     bs->backing_format[0] = '\0';
3236     bs->total_sectors = 0;
3237     bs->encrypted = false;
3238     bs->sg = false;
3239     QDECREF(bs->options);
3240     QDECREF(bs->explicit_options);
3241     bs->options = NULL;
3242     bs->explicit_options = NULL;
3243     QDECREF(bs->full_open_options);
3244     bs->full_open_options = NULL;
3245 
3246     bdrv_release_named_dirty_bitmaps(bs);
3247     assert(QLIST_EMPTY(&bs->dirty_bitmaps));
3248 
3249     QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_next) {
3250         g_free(ban);
3251     }
3252     QLIST_INIT(&bs->aio_notifiers);
3253     bdrv_drained_end(bs);
3254 }
3255 
3256 void bdrv_close_all(void)
3257 {
3258     block_job_cancel_sync_all();
3259     nbd_export_close_all();
3260 
3261     /* Drop references from requests still in flight, such as canceled block
3262      * jobs whose AIO context has not been polled yet */
3263     bdrv_drain_all();
3264 
3265     blk_remove_all_bs();
3266     blockdev_close_all_bdrv_states();
3267 
3268     assert(QTAILQ_EMPTY(&all_bdrv_states));
3269 }
3270 
3271 static bool should_update_child(BdrvChild *c, BlockDriverState *to)
3272 {
3273     BdrvChild *to_c;
3274 
3275     if (c->role->stay_at_node) {
3276         return false;
3277     }
3278 
3279     if (c->role == &child_backing) {
3280         /* If @from is a backing file of @to, ignore the child to avoid
3281          * creating a loop. We only want to change the pointer of other
3282          * parents. */
3283         QLIST_FOREACH(to_c, &to->children, next) {
3284             if (to_c == c) {
3285                 break;
3286             }
3287         }
3288         if (to_c) {
3289             return false;
3290         }
3291     }
3292 
3293     return true;
3294 }
3295 
3296 void bdrv_replace_node(BlockDriverState *from, BlockDriverState *to,
3297                        Error **errp)
3298 {
3299     BdrvChild *c, *next;
3300     GSList *list = NULL, *p;
3301     uint64_t old_perm, old_shared;
3302     uint64_t perm = 0, shared = BLK_PERM_ALL;
3303     int ret;
3304 
3305     assert(!atomic_read(&from->in_flight));
3306     assert(!atomic_read(&to->in_flight));
3307 
3308     /* Make sure that @from doesn't go away until we have successfully attached
3309      * all of its parents to @to. */
3310     bdrv_ref(from);
3311 
3312     /* Put all parents into @list and calculate their cumulative permissions */
3313     QLIST_FOREACH_SAFE(c, &from->parents, next_parent, next) {
3314         if (!should_update_child(c, to)) {
3315             continue;
3316         }
3317         list = g_slist_prepend(list, c);
3318         perm |= c->perm;
3319         shared &= c->shared_perm;
3320     }
3321 
3322     /* Check whether the required permissions can be granted on @to, ignoring
3323      * all BdrvChild in @list so that they can't block themselves. */
3324     ret = bdrv_check_update_perm(to, NULL, perm, shared, list, errp);
3325     if (ret < 0) {
3326         bdrv_abort_perm_update(to);
3327         goto out;
3328     }
3329 
3330     /* Now actually perform the change. We performed the permission check for
3331      * all elements of @list at once, so set the permissions all at once at the
3332      * very end. */
3333     for (p = list; p != NULL; p = p->next) {
3334         c = p->data;
3335 
3336         bdrv_ref(to);
3337         bdrv_replace_child_noperm(c, to);
3338         bdrv_unref(from);
3339     }
3340 
3341     bdrv_get_cumulative_perm(to, &old_perm, &old_shared);
3342     bdrv_set_perm(to, old_perm | perm, old_shared | shared);
3343 
3344 out:
3345     g_slist_free(list);
3346     bdrv_unref(from);
3347 }
3348 
3349 /*
3350  * Add new bs contents at the top of an image chain while the chain is
3351  * live, while keeping required fields on the top layer.
3352  *
3353  * This will modify the BlockDriverState fields, and swap contents
3354  * between bs_new and bs_top. Both bs_new and bs_top are modified.
3355  *
3356  * bs_new must not be attached to a BlockBackend.
3357  *
3358  * This function does not create any image files.
3359  *
3360  * bdrv_append() takes ownership of a bs_new reference and unrefs it because
3361  * that's what the callers commonly need. bs_new will be referenced by the old
3362  * parents of bs_top after bdrv_append() returns. If the caller needs to keep a
3363  * reference of its own, it must call bdrv_ref().
3364  */
3365 void bdrv_append(BlockDriverState *bs_new, BlockDriverState *bs_top,
3366                  Error **errp)
3367 {
3368     Error *local_err = NULL;
3369 
3370     bdrv_set_backing_hd(bs_new, bs_top, &local_err);
3371     if (local_err) {
3372         error_propagate(errp, local_err);
3373         goto out;
3374     }
3375 
3376     bdrv_replace_node(bs_top, bs_new, &local_err);
3377     if (local_err) {
3378         error_propagate(errp, local_err);
3379         bdrv_set_backing_hd(bs_new, NULL, &error_abort);
3380         goto out;
3381     }
3382 
3383     /* bs_new is now referenced by its new parents, we don't need the
3384      * additional reference any more. */
3385 out:
3386     bdrv_unref(bs_new);
3387 }
3388 
3389 static void bdrv_delete(BlockDriverState *bs)
3390 {
3391     assert(!bs->job);
3392     assert(bdrv_op_blocker_is_empty(bs));
3393     assert(!bs->refcnt);
3394 
3395     bdrv_close(bs);
3396 
3397     /* remove from list, if necessary */
3398     if (bs->node_name[0] != '\0') {
3399         QTAILQ_REMOVE(&graph_bdrv_states, bs, node_list);
3400     }
3401     QTAILQ_REMOVE(&all_bdrv_states, bs, bs_list);
3402 
3403     g_free(bs);
3404 }
3405 
3406 /*
3407  * Run consistency checks on an image
3408  *
3409  * Returns 0 if the check could be completed (it doesn't mean that the image is
3410  * free of errors) or -errno when an internal error occurred. The results of the
3411  * check are stored in res.
3412  */
3413 int bdrv_check(BlockDriverState *bs, BdrvCheckResult *res, BdrvCheckMode fix)
3414 {
3415     if (bs->drv == NULL) {
3416         return -ENOMEDIUM;
3417     }
3418     if (bs->drv->bdrv_check == NULL) {
3419         return -ENOTSUP;
3420     }
3421 
3422     memset(res, 0, sizeof(*res));
3423     return bs->drv->bdrv_check(bs, res, fix);
3424 }
3425 
3426 /*
3427  * Return values:
3428  * 0        - success
3429  * -EINVAL  - backing format specified, but no file
3430  * -ENOSPC  - can't update the backing file because no space is left in the
3431  *            image file header
3432  * -ENOTSUP - format driver doesn't support changing the backing file
3433  */
3434 int bdrv_change_backing_file(BlockDriverState *bs,
3435     const char *backing_file, const char *backing_fmt)
3436 {
3437     BlockDriver *drv = bs->drv;
3438     int ret;
3439 
3440     if (!drv) {
3441         return -ENOMEDIUM;
3442     }
3443 
3444     /* Backing file format doesn't make sense without a backing file */
3445     if (backing_fmt && !backing_file) {
3446         return -EINVAL;
3447     }
3448 
3449     if (drv->bdrv_change_backing_file != NULL) {
3450         ret = drv->bdrv_change_backing_file(bs, backing_file, backing_fmt);
3451     } else {
3452         ret = -ENOTSUP;
3453     }
3454 
3455     if (ret == 0) {
3456         pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_file ?: "");
3457         pstrcpy(bs->backing_format, sizeof(bs->backing_format), backing_fmt ?: "");
3458     }
3459     return ret;
3460 }
3461 
3462 /*
3463  * Finds the image layer in the chain that has 'bs' as its backing file.
3464  *
3465  * active is the current topmost image.
3466  *
3467  * Returns NULL if bs is not found in active's image chain,
3468  * or if active == bs.
3469  *
3470  * Returns the bottommost base image if bs == NULL.
3471  */
3472 BlockDriverState *bdrv_find_overlay(BlockDriverState *active,
3473                                     BlockDriverState *bs)
3474 {
3475     while (active && bs != backing_bs(active)) {
3476         active = backing_bs(active);
3477     }
3478 
3479     return active;
3480 }
3481 
3482 /* Given a BDS, searches for the base layer. */
3483 BlockDriverState *bdrv_find_base(BlockDriverState *bs)
3484 {
3485     return bdrv_find_overlay(bs, NULL);
3486 }
3487 
3488 /*
3489  * Drops images above 'base' up to and including 'top', and sets the image
3490  * above 'top' to have base as its backing file.
3491  *
3492  * Requires that the overlay to 'top' is opened r/w, so that the backing file
3493  * information in 'bs' can be properly updated.
3494  *
3495  * E.g., this will convert the following chain:
3496  * bottom <- base <- intermediate <- top <- active
3497  *
3498  * to
3499  *
3500  * bottom <- base <- active
3501  *
3502  * It is allowed for bottom==base, in which case it converts:
3503  *
3504  * base <- intermediate <- top <- active
3505  *
3506  * to
3507  *
3508  * base <- active
3509  *
3510  * If backing_file_str is non-NULL, it will be used when modifying top's
3511  * overlay image metadata.
3512  *
3513  * Error conditions:
3514  *  if active == top, that is considered an error
3515  *
3516  */
3517 int bdrv_drop_intermediate(BlockDriverState *top, BlockDriverState *base,
3518                            const char *backing_file_str)
3519 {
3520     BdrvChild *c, *next;
3521     Error *local_err = NULL;
3522     int ret = -EIO;
3523 
3524     bdrv_ref(top);
3525 
3526     if (!top->drv || !base->drv) {
3527         goto exit;
3528     }
3529 
3530     /* Make sure that base is in the backing chain of top */
3531     if (!bdrv_chain_contains(top, base)) {
3532         goto exit;
3533     }
3534 
3535     /* success - we can delete the intermediate states, and link top->base */
3536     /* TODO Check graph modification op blockers (BLK_PERM_GRAPH_MOD) once
3537      * we've figured out how they should work. */
3538     backing_file_str = backing_file_str ? backing_file_str : base->filename;
3539 
3540     QLIST_FOREACH_SAFE(c, &top->parents, next_parent, next) {
3541         /* Check whether we are allowed to switch c from top to base */
3542         GSList *ignore_children = g_slist_prepend(NULL, c);
3543         bdrv_check_update_perm(base, NULL, c->perm, c->shared_perm,
3544                                ignore_children, &local_err);
3545         if (local_err) {
3546             ret = -EPERM;
3547             error_report_err(local_err);
3548             goto exit;
3549         }
3550         g_slist_free(ignore_children);
3551 
3552         /* If so, update the backing file path in the image file */
3553         if (c->role->update_filename) {
3554             ret = c->role->update_filename(c, base, backing_file_str,
3555                                            &local_err);
3556             if (ret < 0) {
3557                 bdrv_abort_perm_update(base);
3558                 error_report_err(local_err);
3559                 goto exit;
3560             }
3561         }
3562 
3563         /* Do the actual switch in the in-memory graph.
3564          * Completes bdrv_check_update_perm() transaction internally. */
3565         bdrv_ref(base);
3566         bdrv_replace_child(c, base);
3567         bdrv_unref(top);
3568     }
3569 
3570     ret = 0;
3571 exit:
3572     bdrv_unref(top);
3573     return ret;
3574 }
3575 
3576 /**
3577  * Truncate file to 'offset' bytes (needed only for file protocols)
3578  */
3579 int bdrv_truncate(BdrvChild *child, int64_t offset, PreallocMode prealloc,
3580                   Error **errp)
3581 {
3582     BlockDriverState *bs = child->bs;
3583     BlockDriver *drv = bs->drv;
3584     int ret;
3585 
3586     assert(child->perm & BLK_PERM_RESIZE);
3587 
3588     /* if bs->drv == NULL, bs is closed, so there's nothing to do here */
3589     if (!drv) {
3590         error_setg(errp, "No medium inserted");
3591         return -ENOMEDIUM;
3592     }
3593     if (!drv->bdrv_truncate) {
3594         if (bs->file && drv->is_filter) {
3595             return bdrv_truncate(bs->file, offset, prealloc, errp);
3596         }
3597         error_setg(errp, "Image format driver does not support resize");
3598         return -ENOTSUP;
3599     }
3600     if (bs->read_only) {
3601         error_setg(errp, "Image is read-only");
3602         return -EACCES;
3603     }
3604 
3605     assert(!(bs->open_flags & BDRV_O_INACTIVE));
3606 
3607     ret = drv->bdrv_truncate(bs, offset, prealloc, errp);
3608     if (ret < 0) {
3609         return ret;
3610     }
3611     ret = refresh_total_sectors(bs, offset >> BDRV_SECTOR_BITS);
3612     if (ret < 0) {
3613         error_setg_errno(errp, -ret, "Could not refresh total sector count");
3614     } else {
3615         offset = bs->total_sectors * BDRV_SECTOR_SIZE;
3616     }
3617     bdrv_dirty_bitmap_truncate(bs, offset);
3618     bdrv_parent_cb_resize(bs);
3619     atomic_inc(&bs->write_gen);
3620     return ret;
3621 }
3622 
3623 /**
3624  * Length of a allocated file in bytes. Sparse files are counted by actual
3625  * allocated space. Return < 0 if error or unknown.
3626  */
3627 int64_t bdrv_get_allocated_file_size(BlockDriverState *bs)
3628 {
3629     BlockDriver *drv = bs->drv;
3630     if (!drv) {
3631         return -ENOMEDIUM;
3632     }
3633     if (drv->bdrv_get_allocated_file_size) {
3634         return drv->bdrv_get_allocated_file_size(bs);
3635     }
3636     if (bs->file) {
3637         return bdrv_get_allocated_file_size(bs->file->bs);
3638     }
3639     return -ENOTSUP;
3640 }
3641 
3642 /*
3643  * bdrv_measure:
3644  * @drv: Format driver
3645  * @opts: Creation options for new image
3646  * @in_bs: Existing image containing data for new image (may be NULL)
3647  * @errp: Error object
3648  * Returns: A #BlockMeasureInfo (free using qapi_free_BlockMeasureInfo())
3649  *          or NULL on error
3650  *
3651  * Calculate file size required to create a new image.
3652  *
3653  * If @in_bs is given then space for allocated clusters and zero clusters
3654  * from that image are included in the calculation.  If @opts contains a
3655  * backing file that is shared by @in_bs then backing clusters may be omitted
3656  * from the calculation.
3657  *
3658  * If @in_bs is NULL then the calculation includes no allocated clusters
3659  * unless a preallocation option is given in @opts.
3660  *
3661  * Note that @in_bs may use a different BlockDriver from @drv.
3662  *
3663  * If an error occurs the @errp pointer is set.
3664  */
3665 BlockMeasureInfo *bdrv_measure(BlockDriver *drv, QemuOpts *opts,
3666                                BlockDriverState *in_bs, Error **errp)
3667 {
3668     if (!drv->bdrv_measure) {
3669         error_setg(errp, "Block driver '%s' does not support size measurement",
3670                    drv->format_name);
3671         return NULL;
3672     }
3673 
3674     return drv->bdrv_measure(opts, in_bs, errp);
3675 }
3676 
3677 /**
3678  * Return number of sectors on success, -errno on error.
3679  */
3680 int64_t bdrv_nb_sectors(BlockDriverState *bs)
3681 {
3682     BlockDriver *drv = bs->drv;
3683 
3684     if (!drv)
3685         return -ENOMEDIUM;
3686 
3687     if (drv->has_variable_length) {
3688         int ret = refresh_total_sectors(bs, bs->total_sectors);
3689         if (ret < 0) {
3690             return ret;
3691         }
3692     }
3693     return bs->total_sectors;
3694 }
3695 
3696 /**
3697  * Return length in bytes on success, -errno on error.
3698  * The length is always a multiple of BDRV_SECTOR_SIZE.
3699  */
3700 int64_t bdrv_getlength(BlockDriverState *bs)
3701 {
3702     int64_t ret = bdrv_nb_sectors(bs);
3703 
3704     ret = ret > INT64_MAX / BDRV_SECTOR_SIZE ? -EFBIG : ret;
3705     return ret < 0 ? ret : ret * BDRV_SECTOR_SIZE;
3706 }
3707 
3708 /* return 0 as number of sectors if no device present or error */
3709 void bdrv_get_geometry(BlockDriverState *bs, uint64_t *nb_sectors_ptr)
3710 {
3711     int64_t nb_sectors = bdrv_nb_sectors(bs);
3712 
3713     *nb_sectors_ptr = nb_sectors < 0 ? 0 : nb_sectors;
3714 }
3715 
3716 bool bdrv_is_sg(BlockDriverState *bs)
3717 {
3718     return bs->sg;
3719 }
3720 
3721 bool bdrv_is_encrypted(BlockDriverState *bs)
3722 {
3723     if (bs->backing && bs->backing->bs->encrypted) {
3724         return true;
3725     }
3726     return bs->encrypted;
3727 }
3728 
3729 const char *bdrv_get_format_name(BlockDriverState *bs)
3730 {
3731     return bs->drv ? bs->drv->format_name : NULL;
3732 }
3733 
3734 static int qsort_strcmp(const void *a, const void *b)
3735 {
3736     return strcmp(*(char *const *)a, *(char *const *)b);
3737 }
3738 
3739 void bdrv_iterate_format(void (*it)(void *opaque, const char *name),
3740                          void *opaque)
3741 {
3742     BlockDriver *drv;
3743     int count = 0;
3744     int i;
3745     const char **formats = NULL;
3746 
3747     QLIST_FOREACH(drv, &bdrv_drivers, list) {
3748         if (drv->format_name) {
3749             bool found = false;
3750             int i = count;
3751             while (formats && i && !found) {
3752                 found = !strcmp(formats[--i], drv->format_name);
3753             }
3754 
3755             if (!found) {
3756                 formats = g_renew(const char *, formats, count + 1);
3757                 formats[count++] = drv->format_name;
3758             }
3759         }
3760     }
3761 
3762     for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); i++) {
3763         const char *format_name = block_driver_modules[i].format_name;
3764 
3765         if (format_name) {
3766             bool found = false;
3767             int j = count;
3768 
3769             while (formats && j && !found) {
3770                 found = !strcmp(formats[--j], format_name);
3771             }
3772 
3773             if (!found) {
3774                 formats = g_renew(const char *, formats, count + 1);
3775                 formats[count++] = format_name;
3776             }
3777         }
3778     }
3779 
3780     qsort(formats, count, sizeof(formats[0]), qsort_strcmp);
3781 
3782     for (i = 0; i < count; i++) {
3783         it(opaque, formats[i]);
3784     }
3785 
3786     g_free(formats);
3787 }
3788 
3789 /* This function is to find a node in the bs graph */
3790 BlockDriverState *bdrv_find_node(const char *node_name)
3791 {
3792     BlockDriverState *bs;
3793 
3794     assert(node_name);
3795 
3796     QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
3797         if (!strcmp(node_name, bs->node_name)) {
3798             return bs;
3799         }
3800     }
3801     return NULL;
3802 }
3803 
3804 /* Put this QMP function here so it can access the static graph_bdrv_states. */
3805 BlockDeviceInfoList *bdrv_named_nodes_list(Error **errp)
3806 {
3807     BlockDeviceInfoList *list, *entry;
3808     BlockDriverState *bs;
3809 
3810     list = NULL;
3811     QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
3812         BlockDeviceInfo *info = bdrv_block_device_info(NULL, bs, errp);
3813         if (!info) {
3814             qapi_free_BlockDeviceInfoList(list);
3815             return NULL;
3816         }
3817         entry = g_malloc0(sizeof(*entry));
3818         entry->value = info;
3819         entry->next = list;
3820         list = entry;
3821     }
3822 
3823     return list;
3824 }
3825 
3826 BlockDriverState *bdrv_lookup_bs(const char *device,
3827                                  const char *node_name,
3828                                  Error **errp)
3829 {
3830     BlockBackend *blk;
3831     BlockDriverState *bs;
3832 
3833     if (device) {
3834         blk = blk_by_name(device);
3835 
3836         if (blk) {
3837             bs = blk_bs(blk);
3838             if (!bs) {
3839                 error_setg(errp, "Device '%s' has no medium", device);
3840             }
3841 
3842             return bs;
3843         }
3844     }
3845 
3846     if (node_name) {
3847         bs = bdrv_find_node(node_name);
3848 
3849         if (bs) {
3850             return bs;
3851         }
3852     }
3853 
3854     error_setg(errp, "Cannot find device=%s nor node_name=%s",
3855                      device ? device : "",
3856                      node_name ? node_name : "");
3857     return NULL;
3858 }
3859 
3860 /* If 'base' is in the same chain as 'top', return true. Otherwise,
3861  * return false.  If either argument is NULL, return false. */
3862 bool bdrv_chain_contains(BlockDriverState *top, BlockDriverState *base)
3863 {
3864     while (top && top != base) {
3865         top = backing_bs(top);
3866     }
3867 
3868     return top != NULL;
3869 }
3870 
3871 BlockDriverState *bdrv_next_node(BlockDriverState *bs)
3872 {
3873     if (!bs) {
3874         return QTAILQ_FIRST(&graph_bdrv_states);
3875     }
3876     return QTAILQ_NEXT(bs, node_list);
3877 }
3878 
3879 const char *bdrv_get_node_name(const BlockDriverState *bs)
3880 {
3881     return bs->node_name;
3882 }
3883 
3884 const char *bdrv_get_parent_name(const BlockDriverState *bs)
3885 {
3886     BdrvChild *c;
3887     const char *name;
3888 
3889     /* If multiple parents have a name, just pick the first one. */
3890     QLIST_FOREACH(c, &bs->parents, next_parent) {
3891         if (c->role->get_name) {
3892             name = c->role->get_name(c);
3893             if (name && *name) {
3894                 return name;
3895             }
3896         }
3897     }
3898 
3899     return NULL;
3900 }
3901 
3902 /* TODO check what callers really want: bs->node_name or blk_name() */
3903 const char *bdrv_get_device_name(const BlockDriverState *bs)
3904 {
3905     return bdrv_get_parent_name(bs) ?: "";
3906 }
3907 
3908 /* This can be used to identify nodes that might not have a device
3909  * name associated. Since node and device names live in the same
3910  * namespace, the result is unambiguous. The exception is if both are
3911  * absent, then this returns an empty (non-null) string. */
3912 const char *bdrv_get_device_or_node_name(const BlockDriverState *bs)
3913 {
3914     return bdrv_get_parent_name(bs) ?: bs->node_name;
3915 }
3916 
3917 int bdrv_get_flags(BlockDriverState *bs)
3918 {
3919     return bs->open_flags;
3920 }
3921 
3922 int bdrv_has_zero_init_1(BlockDriverState *bs)
3923 {
3924     return 1;
3925 }
3926 
3927 int bdrv_has_zero_init(BlockDriverState *bs)
3928 {
3929     if (!bs->drv) {
3930         return 0;
3931     }
3932 
3933     /* If BS is a copy on write image, it is initialized to
3934        the contents of the base image, which may not be zeroes.  */
3935     if (bs->backing) {
3936         return 0;
3937     }
3938     if (bs->drv->bdrv_has_zero_init) {
3939         return bs->drv->bdrv_has_zero_init(bs);
3940     }
3941     if (bs->file && bs->drv->is_filter) {
3942         return bdrv_has_zero_init(bs->file->bs);
3943     }
3944 
3945     /* safe default */
3946     return 0;
3947 }
3948 
3949 bool bdrv_unallocated_blocks_are_zero(BlockDriverState *bs)
3950 {
3951     BlockDriverInfo bdi;
3952 
3953     if (bs->backing) {
3954         return false;
3955     }
3956 
3957     if (bdrv_get_info(bs, &bdi) == 0) {
3958         return bdi.unallocated_blocks_are_zero;
3959     }
3960 
3961     return false;
3962 }
3963 
3964 bool bdrv_can_write_zeroes_with_unmap(BlockDriverState *bs)
3965 {
3966     BlockDriverInfo bdi;
3967 
3968     if (!(bs->open_flags & BDRV_O_UNMAP)) {
3969         return false;
3970     }
3971 
3972     if (bdrv_get_info(bs, &bdi) == 0) {
3973         return bdi.can_write_zeroes_with_unmap;
3974     }
3975 
3976     return false;
3977 }
3978 
3979 const char *bdrv_get_encrypted_filename(BlockDriverState *bs)
3980 {
3981     if (bs->backing && bs->backing->bs->encrypted)
3982         return bs->backing_file;
3983     else if (bs->encrypted)
3984         return bs->filename;
3985     else
3986         return NULL;
3987 }
3988 
3989 void bdrv_get_backing_filename(BlockDriverState *bs,
3990                                char *filename, int filename_size)
3991 {
3992     pstrcpy(filename, filename_size, bs->backing_file);
3993 }
3994 
3995 int bdrv_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
3996 {
3997     BlockDriver *drv = bs->drv;
3998     /* if bs->drv == NULL, bs is closed, so there's nothing to do here */
3999     if (!drv) {
4000         return -ENOMEDIUM;
4001     }
4002     if (!drv->bdrv_get_info) {
4003         if (bs->file && drv->is_filter) {
4004             return bdrv_get_info(bs->file->bs, bdi);
4005         }
4006         return -ENOTSUP;
4007     }
4008     memset(bdi, 0, sizeof(*bdi));
4009     return drv->bdrv_get_info(bs, bdi);
4010 }
4011 
4012 ImageInfoSpecific *bdrv_get_specific_info(BlockDriverState *bs)
4013 {
4014     BlockDriver *drv = bs->drv;
4015     if (drv && drv->bdrv_get_specific_info) {
4016         return drv->bdrv_get_specific_info(bs);
4017     }
4018     return NULL;
4019 }
4020 
4021 void bdrv_debug_event(BlockDriverState *bs, BlkdebugEvent event)
4022 {
4023     if (!bs || !bs->drv || !bs->drv->bdrv_debug_event) {
4024         return;
4025     }
4026 
4027     bs->drv->bdrv_debug_event(bs, event);
4028 }
4029 
4030 int bdrv_debug_breakpoint(BlockDriverState *bs, const char *event,
4031                           const char *tag)
4032 {
4033     while (bs && bs->drv && !bs->drv->bdrv_debug_breakpoint) {
4034         bs = bs->file ? bs->file->bs : NULL;
4035     }
4036 
4037     if (bs && bs->drv && bs->drv->bdrv_debug_breakpoint) {
4038         return bs->drv->bdrv_debug_breakpoint(bs, event, tag);
4039     }
4040 
4041     return -ENOTSUP;
4042 }
4043 
4044 int bdrv_debug_remove_breakpoint(BlockDriverState *bs, const char *tag)
4045 {
4046     while (bs && bs->drv && !bs->drv->bdrv_debug_remove_breakpoint) {
4047         bs = bs->file ? bs->file->bs : NULL;
4048     }
4049 
4050     if (bs && bs->drv && bs->drv->bdrv_debug_remove_breakpoint) {
4051         return bs->drv->bdrv_debug_remove_breakpoint(bs, tag);
4052     }
4053 
4054     return -ENOTSUP;
4055 }
4056 
4057 int bdrv_debug_resume(BlockDriverState *bs, const char *tag)
4058 {
4059     while (bs && (!bs->drv || !bs->drv->bdrv_debug_resume)) {
4060         bs = bs->file ? bs->file->bs : NULL;
4061     }
4062 
4063     if (bs && bs->drv && bs->drv->bdrv_debug_resume) {
4064         return bs->drv->bdrv_debug_resume(bs, tag);
4065     }
4066 
4067     return -ENOTSUP;
4068 }
4069 
4070 bool bdrv_debug_is_suspended(BlockDriverState *bs, const char *tag)
4071 {
4072     while (bs && bs->drv && !bs->drv->bdrv_debug_is_suspended) {
4073         bs = bs->file ? bs->file->bs : NULL;
4074     }
4075 
4076     if (bs && bs->drv && bs->drv->bdrv_debug_is_suspended) {
4077         return bs->drv->bdrv_debug_is_suspended(bs, tag);
4078     }
4079 
4080     return false;
4081 }
4082 
4083 /* backing_file can either be relative, or absolute, or a protocol.  If it is
4084  * relative, it must be relative to the chain.  So, passing in bs->filename
4085  * from a BDS as backing_file should not be done, as that may be relative to
4086  * the CWD rather than the chain. */
4087 BlockDriverState *bdrv_find_backing_image(BlockDriverState *bs,
4088         const char *backing_file)
4089 {
4090     char *filename_full = NULL;
4091     char *backing_file_full = NULL;
4092     char *filename_tmp = NULL;
4093     int is_protocol = 0;
4094     BlockDriverState *curr_bs = NULL;
4095     BlockDriverState *retval = NULL;
4096     Error *local_error = NULL;
4097 
4098     if (!bs || !bs->drv || !backing_file) {
4099         return NULL;
4100     }
4101 
4102     filename_full     = g_malloc(PATH_MAX);
4103     backing_file_full = g_malloc(PATH_MAX);
4104     filename_tmp      = g_malloc(PATH_MAX);
4105 
4106     is_protocol = path_has_protocol(backing_file);
4107 
4108     for (curr_bs = bs; curr_bs->backing; curr_bs = curr_bs->backing->bs) {
4109 
4110         /* If either of the filename paths is actually a protocol, then
4111          * compare unmodified paths; otherwise make paths relative */
4112         if (is_protocol || path_has_protocol(curr_bs->backing_file)) {
4113             if (strcmp(backing_file, curr_bs->backing_file) == 0) {
4114                 retval = curr_bs->backing->bs;
4115                 break;
4116             }
4117             /* Also check against the full backing filename for the image */
4118             bdrv_get_full_backing_filename(curr_bs, backing_file_full, PATH_MAX,
4119                                            &local_error);
4120             if (local_error == NULL) {
4121                 if (strcmp(backing_file, backing_file_full) == 0) {
4122                     retval = curr_bs->backing->bs;
4123                     break;
4124                 }
4125             } else {
4126                 error_free(local_error);
4127                 local_error = NULL;
4128             }
4129         } else {
4130             /* If not an absolute filename path, make it relative to the current
4131              * image's filename path */
4132             path_combine(filename_tmp, PATH_MAX, curr_bs->filename,
4133                          backing_file);
4134 
4135             /* We are going to compare absolute pathnames */
4136             if (!realpath(filename_tmp, filename_full)) {
4137                 continue;
4138             }
4139 
4140             /* We need to make sure the backing filename we are comparing against
4141              * is relative to the current image filename (or absolute) */
4142             path_combine(filename_tmp, PATH_MAX, curr_bs->filename,
4143                          curr_bs->backing_file);
4144 
4145             if (!realpath(filename_tmp, backing_file_full)) {
4146                 continue;
4147             }
4148 
4149             if (strcmp(backing_file_full, filename_full) == 0) {
4150                 retval = curr_bs->backing->bs;
4151                 break;
4152             }
4153         }
4154     }
4155 
4156     g_free(filename_full);
4157     g_free(backing_file_full);
4158     g_free(filename_tmp);
4159     return retval;
4160 }
4161 
4162 void bdrv_init(void)
4163 {
4164     module_call_init(MODULE_INIT_BLOCK);
4165 }
4166 
4167 void bdrv_init_with_whitelist(void)
4168 {
4169     use_bdrv_whitelist = 1;
4170     bdrv_init();
4171 }
4172 
4173 void bdrv_invalidate_cache(BlockDriverState *bs, Error **errp)
4174 {
4175     BdrvChild *child, *parent;
4176     uint64_t perm, shared_perm;
4177     Error *local_err = NULL;
4178     int ret;
4179 
4180     if (!bs->drv)  {
4181         return;
4182     }
4183 
4184     if (!(bs->open_flags & BDRV_O_INACTIVE)) {
4185         return;
4186     }
4187 
4188     QLIST_FOREACH(child, &bs->children, next) {
4189         bdrv_invalidate_cache(child->bs, &local_err);
4190         if (local_err) {
4191             error_propagate(errp, local_err);
4192             return;
4193         }
4194     }
4195 
4196     /*
4197      * Update permissions, they may differ for inactive nodes.
4198      *
4199      * Note that the required permissions of inactive images are always a
4200      * subset of the permissions required after activating the image. This
4201      * allows us to just get the permissions upfront without restricting
4202      * drv->bdrv_invalidate_cache().
4203      *
4204      * It also means that in error cases, we don't have to try and revert to
4205      * the old permissions (which is an operation that could fail, too). We can
4206      * just keep the extended permissions for the next time that an activation
4207      * of the image is tried.
4208      */
4209     bs->open_flags &= ~BDRV_O_INACTIVE;
4210     bdrv_get_cumulative_perm(bs, &perm, &shared_perm);
4211     ret = bdrv_check_perm(bs, NULL, perm, shared_perm, NULL, &local_err);
4212     if (ret < 0) {
4213         bs->open_flags |= BDRV_O_INACTIVE;
4214         error_propagate(errp, local_err);
4215         return;
4216     }
4217     bdrv_set_perm(bs, perm, shared_perm);
4218 
4219     if (bs->drv->bdrv_invalidate_cache) {
4220         bs->drv->bdrv_invalidate_cache(bs, &local_err);
4221         if (local_err) {
4222             bs->open_flags |= BDRV_O_INACTIVE;
4223             error_propagate(errp, local_err);
4224             return;
4225         }
4226     }
4227 
4228     ret = refresh_total_sectors(bs, bs->total_sectors);
4229     if (ret < 0) {
4230         bs->open_flags |= BDRV_O_INACTIVE;
4231         error_setg_errno(errp, -ret, "Could not refresh total sector count");
4232         return;
4233     }
4234 
4235     QLIST_FOREACH(parent, &bs->parents, next_parent) {
4236         if (parent->role->activate) {
4237             parent->role->activate(parent, &local_err);
4238             if (local_err) {
4239                 error_propagate(errp, local_err);
4240                 return;
4241             }
4242         }
4243     }
4244 }
4245 
4246 void bdrv_invalidate_cache_all(Error **errp)
4247 {
4248     BlockDriverState *bs;
4249     Error *local_err = NULL;
4250     BdrvNextIterator it;
4251 
4252     for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
4253         AioContext *aio_context = bdrv_get_aio_context(bs);
4254 
4255         aio_context_acquire(aio_context);
4256         bdrv_invalidate_cache(bs, &local_err);
4257         aio_context_release(aio_context);
4258         if (local_err) {
4259             error_propagate(errp, local_err);
4260             bdrv_next_cleanup(&it);
4261             return;
4262         }
4263     }
4264 }
4265 
4266 static int bdrv_inactivate_recurse(BlockDriverState *bs,
4267                                    bool setting_flag)
4268 {
4269     BdrvChild *child, *parent;
4270     int ret;
4271 
4272     if (!bs->drv) {
4273         return -ENOMEDIUM;
4274     }
4275 
4276     if (!setting_flag && bs->drv->bdrv_inactivate) {
4277         ret = bs->drv->bdrv_inactivate(bs);
4278         if (ret < 0) {
4279             return ret;
4280         }
4281     }
4282 
4283     if (setting_flag && !(bs->open_flags & BDRV_O_INACTIVE)) {
4284         uint64_t perm, shared_perm;
4285 
4286         QLIST_FOREACH(parent, &bs->parents, next_parent) {
4287             if (parent->role->inactivate) {
4288                 ret = parent->role->inactivate(parent);
4289                 if (ret < 0) {
4290                     return ret;
4291                 }
4292             }
4293         }
4294 
4295         bs->open_flags |= BDRV_O_INACTIVE;
4296 
4297         /* Update permissions, they may differ for inactive nodes */
4298         bdrv_get_cumulative_perm(bs, &perm, &shared_perm);
4299         bdrv_check_perm(bs, NULL, perm, shared_perm, NULL, &error_abort);
4300         bdrv_set_perm(bs, perm, shared_perm);
4301     }
4302 
4303     QLIST_FOREACH(child, &bs->children, next) {
4304         ret = bdrv_inactivate_recurse(child->bs, setting_flag);
4305         if (ret < 0) {
4306             return ret;
4307         }
4308     }
4309 
4310     /* At this point persistent bitmaps should be already stored by the format
4311      * driver */
4312     bdrv_release_persistent_dirty_bitmaps(bs);
4313 
4314     return 0;
4315 }
4316 
4317 int bdrv_inactivate_all(void)
4318 {
4319     BlockDriverState *bs = NULL;
4320     BdrvNextIterator it;
4321     int ret = 0;
4322     int pass;
4323 
4324     for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
4325         aio_context_acquire(bdrv_get_aio_context(bs));
4326     }
4327 
4328     /* We do two passes of inactivation. The first pass calls to drivers'
4329      * .bdrv_inactivate callbacks recursively so all cache is flushed to disk;
4330      * the second pass sets the BDRV_O_INACTIVE flag so that no further write
4331      * is allowed. */
4332     for (pass = 0; pass < 2; pass++) {
4333         for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
4334             ret = bdrv_inactivate_recurse(bs, pass);
4335             if (ret < 0) {
4336                 bdrv_next_cleanup(&it);
4337                 goto out;
4338             }
4339         }
4340     }
4341 
4342 out:
4343     for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
4344         aio_context_release(bdrv_get_aio_context(bs));
4345     }
4346 
4347     return ret;
4348 }
4349 
4350 /**************************************************************/
4351 /* removable device support */
4352 
4353 /**
4354  * Return TRUE if the media is present
4355  */
4356 bool bdrv_is_inserted(BlockDriverState *bs)
4357 {
4358     BlockDriver *drv = bs->drv;
4359     BdrvChild *child;
4360 
4361     if (!drv) {
4362         return false;
4363     }
4364     if (drv->bdrv_is_inserted) {
4365         return drv->bdrv_is_inserted(bs);
4366     }
4367     QLIST_FOREACH(child, &bs->children, next) {
4368         if (!bdrv_is_inserted(child->bs)) {
4369             return false;
4370         }
4371     }
4372     return true;
4373 }
4374 
4375 /**
4376  * If eject_flag is TRUE, eject the media. Otherwise, close the tray
4377  */
4378 void bdrv_eject(BlockDriverState *bs, bool eject_flag)
4379 {
4380     BlockDriver *drv = bs->drv;
4381 
4382     if (drv && drv->bdrv_eject) {
4383         drv->bdrv_eject(bs, eject_flag);
4384     }
4385 }
4386 
4387 /**
4388  * Lock or unlock the media (if it is locked, the user won't be able
4389  * to eject it manually).
4390  */
4391 void bdrv_lock_medium(BlockDriverState *bs, bool locked)
4392 {
4393     BlockDriver *drv = bs->drv;
4394 
4395     trace_bdrv_lock_medium(bs, locked);
4396 
4397     if (drv && drv->bdrv_lock_medium) {
4398         drv->bdrv_lock_medium(bs, locked);
4399     }
4400 }
4401 
4402 /* Get a reference to bs */
4403 void bdrv_ref(BlockDriverState *bs)
4404 {
4405     bs->refcnt++;
4406 }
4407 
4408 /* Release a previously grabbed reference to bs.
4409  * If after releasing, reference count is zero, the BlockDriverState is
4410  * deleted. */
4411 void bdrv_unref(BlockDriverState *bs)
4412 {
4413     if (!bs) {
4414         return;
4415     }
4416     assert(bs->refcnt > 0);
4417     if (--bs->refcnt == 0) {
4418         bdrv_delete(bs);
4419     }
4420 }
4421 
4422 struct BdrvOpBlocker {
4423     Error *reason;
4424     QLIST_ENTRY(BdrvOpBlocker) list;
4425 };
4426 
4427 bool bdrv_op_is_blocked(BlockDriverState *bs, BlockOpType op, Error **errp)
4428 {
4429     BdrvOpBlocker *blocker;
4430     assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
4431     if (!QLIST_EMPTY(&bs->op_blockers[op])) {
4432         blocker = QLIST_FIRST(&bs->op_blockers[op]);
4433         error_propagate(errp, error_copy(blocker->reason));
4434         error_prepend(errp, "Node '%s' is busy: ",
4435                       bdrv_get_device_or_node_name(bs));
4436         return true;
4437     }
4438     return false;
4439 }
4440 
4441 void bdrv_op_block(BlockDriverState *bs, BlockOpType op, Error *reason)
4442 {
4443     BdrvOpBlocker *blocker;
4444     assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
4445 
4446     blocker = g_new0(BdrvOpBlocker, 1);
4447     blocker->reason = reason;
4448     QLIST_INSERT_HEAD(&bs->op_blockers[op], blocker, list);
4449 }
4450 
4451 void bdrv_op_unblock(BlockDriverState *bs, BlockOpType op, Error *reason)
4452 {
4453     BdrvOpBlocker *blocker, *next;
4454     assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
4455     QLIST_FOREACH_SAFE(blocker, &bs->op_blockers[op], list, next) {
4456         if (blocker->reason == reason) {
4457             QLIST_REMOVE(blocker, list);
4458             g_free(blocker);
4459         }
4460     }
4461 }
4462 
4463 void bdrv_op_block_all(BlockDriverState *bs, Error *reason)
4464 {
4465     int i;
4466     for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
4467         bdrv_op_block(bs, i, reason);
4468     }
4469 }
4470 
4471 void bdrv_op_unblock_all(BlockDriverState *bs, Error *reason)
4472 {
4473     int i;
4474     for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
4475         bdrv_op_unblock(bs, i, reason);
4476     }
4477 }
4478 
4479 bool bdrv_op_blocker_is_empty(BlockDriverState *bs)
4480 {
4481     int i;
4482 
4483     for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
4484         if (!QLIST_EMPTY(&bs->op_blockers[i])) {
4485             return false;
4486         }
4487     }
4488     return true;
4489 }
4490 
4491 void bdrv_img_create(const char *filename, const char *fmt,
4492                      const char *base_filename, const char *base_fmt,
4493                      char *options, uint64_t img_size, int flags, bool quiet,
4494                      Error **errp)
4495 {
4496     QemuOptsList *create_opts = NULL;
4497     QemuOpts *opts = NULL;
4498     const char *backing_fmt, *backing_file;
4499     int64_t size;
4500     BlockDriver *drv, *proto_drv;
4501     Error *local_err = NULL;
4502     int ret = 0;
4503 
4504     /* Find driver and parse its options */
4505     drv = bdrv_find_format(fmt);
4506     if (!drv) {
4507         error_setg(errp, "Unknown file format '%s'", fmt);
4508         return;
4509     }
4510 
4511     proto_drv = bdrv_find_protocol(filename, true, errp);
4512     if (!proto_drv) {
4513         return;
4514     }
4515 
4516     if (!drv->create_opts) {
4517         error_setg(errp, "Format driver '%s' does not support image creation",
4518                    drv->format_name);
4519         return;
4520     }
4521 
4522     if (!proto_drv->create_opts) {
4523         error_setg(errp, "Protocol driver '%s' does not support image creation",
4524                    proto_drv->format_name);
4525         return;
4526     }
4527 
4528     create_opts = qemu_opts_append(create_opts, drv->create_opts);
4529     create_opts = qemu_opts_append(create_opts, proto_drv->create_opts);
4530 
4531     /* Create parameter list with default values */
4532     opts = qemu_opts_create(create_opts, NULL, 0, &error_abort);
4533     qemu_opt_set_number(opts, BLOCK_OPT_SIZE, img_size, &error_abort);
4534 
4535     /* Parse -o options */
4536     if (options) {
4537         qemu_opts_do_parse(opts, options, NULL, &local_err);
4538         if (local_err) {
4539             error_report_err(local_err);
4540             local_err = NULL;
4541             error_setg(errp, "Invalid options for file format '%s'", fmt);
4542             goto out;
4543         }
4544     }
4545 
4546     if (base_filename) {
4547         qemu_opt_set(opts, BLOCK_OPT_BACKING_FILE, base_filename, &local_err);
4548         if (local_err) {
4549             error_setg(errp, "Backing file not supported for file format '%s'",
4550                        fmt);
4551             goto out;
4552         }
4553     }
4554 
4555     if (base_fmt) {
4556         qemu_opt_set(opts, BLOCK_OPT_BACKING_FMT, base_fmt, &local_err);
4557         if (local_err) {
4558             error_setg(errp, "Backing file format not supported for file "
4559                              "format '%s'", fmt);
4560             goto out;
4561         }
4562     }
4563 
4564     backing_file = qemu_opt_get(opts, BLOCK_OPT_BACKING_FILE);
4565     if (backing_file) {
4566         if (!strcmp(filename, backing_file)) {
4567             error_setg(errp, "Error: Trying to create an image with the "
4568                              "same filename as the backing file");
4569             goto out;
4570         }
4571     }
4572 
4573     backing_fmt = qemu_opt_get(opts, BLOCK_OPT_BACKING_FMT);
4574 
4575     /* The size for the image must always be specified, unless we have a backing
4576      * file and we have not been forbidden from opening it. */
4577     size = qemu_opt_get_size(opts, BLOCK_OPT_SIZE, img_size);
4578     if (backing_file && !(flags & BDRV_O_NO_BACKING)) {
4579         BlockDriverState *bs;
4580         char *full_backing = g_new0(char, PATH_MAX);
4581         int back_flags;
4582         QDict *backing_options = NULL;
4583 
4584         bdrv_get_full_backing_filename_from_filename(filename, backing_file,
4585                                                      full_backing, PATH_MAX,
4586                                                      &local_err);
4587         if (local_err) {
4588             g_free(full_backing);
4589             goto out;
4590         }
4591 
4592         /* backing files always opened read-only */
4593         back_flags = flags;
4594         back_flags &= ~(BDRV_O_RDWR | BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING);
4595 
4596         if (backing_fmt) {
4597             backing_options = qdict_new();
4598             qdict_put_str(backing_options, "driver", backing_fmt);
4599         }
4600 
4601         bs = bdrv_open(full_backing, NULL, backing_options, back_flags,
4602                        &local_err);
4603         g_free(full_backing);
4604         if (!bs && size != -1) {
4605             /* Couldn't open BS, but we have a size, so it's nonfatal */
4606             warn_reportf_err(local_err,
4607                             "Could not verify backing image. "
4608                             "This may become an error in future versions.\n");
4609             local_err = NULL;
4610         } else if (!bs) {
4611             /* Couldn't open bs, do not have size */
4612             error_append_hint(&local_err,
4613                               "Could not open backing image to determine size.\n");
4614             goto out;
4615         } else {
4616             if (size == -1) {
4617                 /* Opened BS, have no size */
4618                 size = bdrv_getlength(bs);
4619                 if (size < 0) {
4620                     error_setg_errno(errp, -size, "Could not get size of '%s'",
4621                                      backing_file);
4622                     bdrv_unref(bs);
4623                     goto out;
4624                 }
4625                 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, size, &error_abort);
4626             }
4627             bdrv_unref(bs);
4628         }
4629     } /* (backing_file && !(flags & BDRV_O_NO_BACKING)) */
4630 
4631     if (size == -1) {
4632         error_setg(errp, "Image creation needs a size parameter");
4633         goto out;
4634     }
4635 
4636     if (!quiet) {
4637         printf("Formatting '%s', fmt=%s ", filename, fmt);
4638         qemu_opts_print(opts, " ");
4639         puts("");
4640     }
4641 
4642     ret = bdrv_create(drv, filename, opts, &local_err);
4643 
4644     if (ret == -EFBIG) {
4645         /* This is generally a better message than whatever the driver would
4646          * deliver (especially because of the cluster_size_hint), since that
4647          * is most probably not much different from "image too large". */
4648         const char *cluster_size_hint = "";
4649         if (qemu_opt_get_size(opts, BLOCK_OPT_CLUSTER_SIZE, 0)) {
4650             cluster_size_hint = " (try using a larger cluster size)";
4651         }
4652         error_setg(errp, "The image size is too large for file format '%s'"
4653                    "%s", fmt, cluster_size_hint);
4654         error_free(local_err);
4655         local_err = NULL;
4656     }
4657 
4658 out:
4659     qemu_opts_del(opts);
4660     qemu_opts_free(create_opts);
4661     error_propagate(errp, local_err);
4662 }
4663 
4664 AioContext *bdrv_get_aio_context(BlockDriverState *bs)
4665 {
4666     return bs->aio_context;
4667 }
4668 
4669 void bdrv_coroutine_enter(BlockDriverState *bs, Coroutine *co)
4670 {
4671     aio_co_enter(bdrv_get_aio_context(bs), co);
4672 }
4673 
4674 static void bdrv_do_remove_aio_context_notifier(BdrvAioNotifier *ban)
4675 {
4676     QLIST_REMOVE(ban, list);
4677     g_free(ban);
4678 }
4679 
4680 void bdrv_detach_aio_context(BlockDriverState *bs)
4681 {
4682     BdrvAioNotifier *baf, *baf_tmp;
4683     BdrvChild *child;
4684 
4685     if (!bs->drv) {
4686         return;
4687     }
4688 
4689     assert(!bs->walking_aio_notifiers);
4690     bs->walking_aio_notifiers = true;
4691     QLIST_FOREACH_SAFE(baf, &bs->aio_notifiers, list, baf_tmp) {
4692         if (baf->deleted) {
4693             bdrv_do_remove_aio_context_notifier(baf);
4694         } else {
4695             baf->detach_aio_context(baf->opaque);
4696         }
4697     }
4698     /* Never mind iterating again to check for ->deleted.  bdrv_close() will
4699      * remove remaining aio notifiers if we aren't called again.
4700      */
4701     bs->walking_aio_notifiers = false;
4702 
4703     if (bs->drv->bdrv_detach_aio_context) {
4704         bs->drv->bdrv_detach_aio_context(bs);
4705     }
4706     QLIST_FOREACH(child, &bs->children, next) {
4707         bdrv_detach_aio_context(child->bs);
4708     }
4709 
4710     bs->aio_context = NULL;
4711 }
4712 
4713 void bdrv_attach_aio_context(BlockDriverState *bs,
4714                              AioContext *new_context)
4715 {
4716     BdrvAioNotifier *ban, *ban_tmp;
4717     BdrvChild *child;
4718 
4719     if (!bs->drv) {
4720         return;
4721     }
4722 
4723     bs->aio_context = new_context;
4724 
4725     QLIST_FOREACH(child, &bs->children, next) {
4726         bdrv_attach_aio_context(child->bs, new_context);
4727     }
4728     if (bs->drv->bdrv_attach_aio_context) {
4729         bs->drv->bdrv_attach_aio_context(bs, new_context);
4730     }
4731 
4732     assert(!bs->walking_aio_notifiers);
4733     bs->walking_aio_notifiers = true;
4734     QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_tmp) {
4735         if (ban->deleted) {
4736             bdrv_do_remove_aio_context_notifier(ban);
4737         } else {
4738             ban->attached_aio_context(new_context, ban->opaque);
4739         }
4740     }
4741     bs->walking_aio_notifiers = false;
4742 }
4743 
4744 void bdrv_set_aio_context(BlockDriverState *bs, AioContext *new_context)
4745 {
4746     AioContext *ctx = bdrv_get_aio_context(bs);
4747 
4748     aio_disable_external(ctx);
4749     bdrv_parent_drained_begin(bs);
4750     bdrv_drain(bs); /* ensure there are no in-flight requests */
4751 
4752     while (aio_poll(ctx, false)) {
4753         /* wait for all bottom halves to execute */
4754     }
4755 
4756     bdrv_detach_aio_context(bs);
4757 
4758     /* This function executes in the old AioContext so acquire the new one in
4759      * case it runs in a different thread.
4760      */
4761     aio_context_acquire(new_context);
4762     bdrv_attach_aio_context(bs, new_context);
4763     bdrv_parent_drained_end(bs);
4764     aio_enable_external(ctx);
4765     aio_context_release(new_context);
4766 }
4767 
4768 void bdrv_add_aio_context_notifier(BlockDriverState *bs,
4769         void (*attached_aio_context)(AioContext *new_context, void *opaque),
4770         void (*detach_aio_context)(void *opaque), void *opaque)
4771 {
4772     BdrvAioNotifier *ban = g_new(BdrvAioNotifier, 1);
4773     *ban = (BdrvAioNotifier){
4774         .attached_aio_context = attached_aio_context,
4775         .detach_aio_context   = detach_aio_context,
4776         .opaque               = opaque
4777     };
4778 
4779     QLIST_INSERT_HEAD(&bs->aio_notifiers, ban, list);
4780 }
4781 
4782 void bdrv_remove_aio_context_notifier(BlockDriverState *bs,
4783                                       void (*attached_aio_context)(AioContext *,
4784                                                                    void *),
4785                                       void (*detach_aio_context)(void *),
4786                                       void *opaque)
4787 {
4788     BdrvAioNotifier *ban, *ban_next;
4789 
4790     QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_next) {
4791         if (ban->attached_aio_context == attached_aio_context &&
4792             ban->detach_aio_context   == detach_aio_context   &&
4793             ban->opaque               == opaque               &&
4794             ban->deleted              == false)
4795         {
4796             if (bs->walking_aio_notifiers) {
4797                 ban->deleted = true;
4798             } else {
4799                 bdrv_do_remove_aio_context_notifier(ban);
4800             }
4801             return;
4802         }
4803     }
4804 
4805     abort();
4806 }
4807 
4808 int bdrv_amend_options(BlockDriverState *bs, QemuOpts *opts,
4809                        BlockDriverAmendStatusCB *status_cb, void *cb_opaque)
4810 {
4811     if (!bs->drv) {
4812         return -ENOMEDIUM;
4813     }
4814     if (!bs->drv->bdrv_amend_options) {
4815         return -ENOTSUP;
4816     }
4817     return bs->drv->bdrv_amend_options(bs, opts, status_cb, cb_opaque);
4818 }
4819 
4820 /* This function will be called by the bdrv_recurse_is_first_non_filter method
4821  * of block filter and by bdrv_is_first_non_filter.
4822  * It is used to test if the given bs is the candidate or recurse more in the
4823  * node graph.
4824  */
4825 bool bdrv_recurse_is_first_non_filter(BlockDriverState *bs,
4826                                       BlockDriverState *candidate)
4827 {
4828     /* return false if basic checks fails */
4829     if (!bs || !bs->drv) {
4830         return false;
4831     }
4832 
4833     /* the code reached a non block filter driver -> check if the bs is
4834      * the same as the candidate. It's the recursion termination condition.
4835      */
4836     if (!bs->drv->is_filter) {
4837         return bs == candidate;
4838     }
4839     /* Down this path the driver is a block filter driver */
4840 
4841     /* If the block filter recursion method is defined use it to recurse down
4842      * the node graph.
4843      */
4844     if (bs->drv->bdrv_recurse_is_first_non_filter) {
4845         return bs->drv->bdrv_recurse_is_first_non_filter(bs, candidate);
4846     }
4847 
4848     /* the driver is a block filter but don't allow to recurse -> return false
4849      */
4850     return false;
4851 }
4852 
4853 /* This function checks if the candidate is the first non filter bs down it's
4854  * bs chain. Since we don't have pointers to parents it explore all bs chains
4855  * from the top. Some filters can choose not to pass down the recursion.
4856  */
4857 bool bdrv_is_first_non_filter(BlockDriverState *candidate)
4858 {
4859     BlockDriverState *bs;
4860     BdrvNextIterator it;
4861 
4862     /* walk down the bs forest recursively */
4863     for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
4864         bool perm;
4865 
4866         /* try to recurse in this top level bs */
4867         perm = bdrv_recurse_is_first_non_filter(bs, candidate);
4868 
4869         /* candidate is the first non filter */
4870         if (perm) {
4871             bdrv_next_cleanup(&it);
4872             return true;
4873         }
4874     }
4875 
4876     return false;
4877 }
4878 
4879 BlockDriverState *check_to_replace_node(BlockDriverState *parent_bs,
4880                                         const char *node_name, Error **errp)
4881 {
4882     BlockDriverState *to_replace_bs = bdrv_find_node(node_name);
4883     AioContext *aio_context;
4884 
4885     if (!to_replace_bs) {
4886         error_setg(errp, "Node name '%s' not found", node_name);
4887         return NULL;
4888     }
4889 
4890     aio_context = bdrv_get_aio_context(to_replace_bs);
4891     aio_context_acquire(aio_context);
4892 
4893     if (bdrv_op_is_blocked(to_replace_bs, BLOCK_OP_TYPE_REPLACE, errp)) {
4894         to_replace_bs = NULL;
4895         goto out;
4896     }
4897 
4898     /* We don't want arbitrary node of the BDS chain to be replaced only the top
4899      * most non filter in order to prevent data corruption.
4900      * Another benefit is that this tests exclude backing files which are
4901      * blocked by the backing blockers.
4902      */
4903     if (!bdrv_recurse_is_first_non_filter(parent_bs, to_replace_bs)) {
4904         error_setg(errp, "Only top most non filter can be replaced");
4905         to_replace_bs = NULL;
4906         goto out;
4907     }
4908 
4909 out:
4910     aio_context_release(aio_context);
4911     return to_replace_bs;
4912 }
4913 
4914 static bool append_open_options(QDict *d, BlockDriverState *bs)
4915 {
4916     const QDictEntry *entry;
4917     QemuOptDesc *desc;
4918     BdrvChild *child;
4919     bool found_any = false;
4920     const char *p;
4921 
4922     for (entry = qdict_first(bs->options); entry;
4923          entry = qdict_next(bs->options, entry))
4924     {
4925         /* Exclude options for children */
4926         QLIST_FOREACH(child, &bs->children, next) {
4927             if (strstart(qdict_entry_key(entry), child->name, &p)
4928                 && (!*p || *p == '.'))
4929             {
4930                 break;
4931             }
4932         }
4933         if (child) {
4934             continue;
4935         }
4936 
4937         /* And exclude all non-driver-specific options */
4938         for (desc = bdrv_runtime_opts.desc; desc->name; desc++) {
4939             if (!strcmp(qdict_entry_key(entry), desc->name)) {
4940                 break;
4941             }
4942         }
4943         if (desc->name) {
4944             continue;
4945         }
4946 
4947         qobject_incref(qdict_entry_value(entry));
4948         qdict_put_obj(d, qdict_entry_key(entry), qdict_entry_value(entry));
4949         found_any = true;
4950     }
4951 
4952     return found_any;
4953 }
4954 
4955 /* Updates the following BDS fields:
4956  *  - exact_filename: A filename which may be used for opening a block device
4957  *                    which (mostly) equals the given BDS (even without any
4958  *                    other options; so reading and writing must return the same
4959  *                    results, but caching etc. may be different)
4960  *  - full_open_options: Options which, when given when opening a block device
4961  *                       (without a filename), result in a BDS (mostly)
4962  *                       equalling the given one
4963  *  - filename: If exact_filename is set, it is copied here. Otherwise,
4964  *              full_open_options is converted to a JSON object, prefixed with
4965  *              "json:" (for use through the JSON pseudo protocol) and put here.
4966  */
4967 void bdrv_refresh_filename(BlockDriverState *bs)
4968 {
4969     BlockDriver *drv = bs->drv;
4970     QDict *opts;
4971 
4972     if (!drv) {
4973         return;
4974     }
4975 
4976     /* This BDS's file name will most probably depend on its file's name, so
4977      * refresh that first */
4978     if (bs->file) {
4979         bdrv_refresh_filename(bs->file->bs);
4980     }
4981 
4982     if (drv->bdrv_refresh_filename) {
4983         /* Obsolete information is of no use here, so drop the old file name
4984          * information before refreshing it */
4985         bs->exact_filename[0] = '\0';
4986         if (bs->full_open_options) {
4987             QDECREF(bs->full_open_options);
4988             bs->full_open_options = NULL;
4989         }
4990 
4991         opts = qdict_new();
4992         append_open_options(opts, bs);
4993         drv->bdrv_refresh_filename(bs, opts);
4994         QDECREF(opts);
4995     } else if (bs->file) {
4996         /* Try to reconstruct valid information from the underlying file */
4997         bool has_open_options;
4998 
4999         bs->exact_filename[0] = '\0';
5000         if (bs->full_open_options) {
5001             QDECREF(bs->full_open_options);
5002             bs->full_open_options = NULL;
5003         }
5004 
5005         opts = qdict_new();
5006         has_open_options = append_open_options(opts, bs);
5007 
5008         /* If no specific options have been given for this BDS, the filename of
5009          * the underlying file should suffice for this one as well */
5010         if (bs->file->bs->exact_filename[0] && !has_open_options) {
5011             strcpy(bs->exact_filename, bs->file->bs->exact_filename);
5012         }
5013         /* Reconstructing the full options QDict is simple for most format block
5014          * drivers, as long as the full options are known for the underlying
5015          * file BDS. The full options QDict of that file BDS should somehow
5016          * contain a representation of the filename, therefore the following
5017          * suffices without querying the (exact_)filename of this BDS. */
5018         if (bs->file->bs->full_open_options) {
5019             qdict_put_str(opts, "driver", drv->format_name);
5020             QINCREF(bs->file->bs->full_open_options);
5021             qdict_put(opts, "file", bs->file->bs->full_open_options);
5022 
5023             bs->full_open_options = opts;
5024         } else {
5025             QDECREF(opts);
5026         }
5027     } else if (!bs->full_open_options && qdict_size(bs->options)) {
5028         /* There is no underlying file BDS (at least referenced by BDS.file),
5029          * so the full options QDict should be equal to the options given
5030          * specifically for this block device when it was opened (plus the
5031          * driver specification).
5032          * Because those options don't change, there is no need to update
5033          * full_open_options when it's already set. */
5034 
5035         opts = qdict_new();
5036         append_open_options(opts, bs);
5037         qdict_put_str(opts, "driver", drv->format_name);
5038 
5039         if (bs->exact_filename[0]) {
5040             /* This may not work for all block protocol drivers (some may
5041              * require this filename to be parsed), but we have to find some
5042              * default solution here, so just include it. If some block driver
5043              * does not support pure options without any filename at all or
5044              * needs some special format of the options QDict, it needs to
5045              * implement the driver-specific bdrv_refresh_filename() function.
5046              */
5047             qdict_put_str(opts, "filename", bs->exact_filename);
5048         }
5049 
5050         bs->full_open_options = opts;
5051     }
5052 
5053     if (bs->exact_filename[0]) {
5054         pstrcpy(bs->filename, sizeof(bs->filename), bs->exact_filename);
5055     } else if (bs->full_open_options) {
5056         QString *json = qobject_to_json(QOBJECT(bs->full_open_options));
5057         snprintf(bs->filename, sizeof(bs->filename), "json:%s",
5058                  qstring_get_str(json));
5059         QDECREF(json);
5060     }
5061 }
5062 
5063 /*
5064  * Hot add/remove a BDS's child. So the user can take a child offline when
5065  * it is broken and take a new child online
5066  */
5067 void bdrv_add_child(BlockDriverState *parent_bs, BlockDriverState *child_bs,
5068                     Error **errp)
5069 {
5070 
5071     if (!parent_bs->drv || !parent_bs->drv->bdrv_add_child) {
5072         error_setg(errp, "The node %s does not support adding a child",
5073                    bdrv_get_device_or_node_name(parent_bs));
5074         return;
5075     }
5076 
5077     if (!QLIST_EMPTY(&child_bs->parents)) {
5078         error_setg(errp, "The node %s already has a parent",
5079                    child_bs->node_name);
5080         return;
5081     }
5082 
5083     parent_bs->drv->bdrv_add_child(parent_bs, child_bs, errp);
5084 }
5085 
5086 void bdrv_del_child(BlockDriverState *parent_bs, BdrvChild *child, Error **errp)
5087 {
5088     BdrvChild *tmp;
5089 
5090     if (!parent_bs->drv || !parent_bs->drv->bdrv_del_child) {
5091         error_setg(errp, "The node %s does not support removing a child",
5092                    bdrv_get_device_or_node_name(parent_bs));
5093         return;
5094     }
5095 
5096     QLIST_FOREACH(tmp, &parent_bs->children, next) {
5097         if (tmp == child) {
5098             break;
5099         }
5100     }
5101 
5102     if (!tmp) {
5103         error_setg(errp, "The node %s does not have a child named %s",
5104                    bdrv_get_device_or_node_name(parent_bs),
5105                    bdrv_get_device_or_node_name(child->bs));
5106         return;
5107     }
5108 
5109     parent_bs->drv->bdrv_del_child(parent_bs, child, errp);
5110 }
5111 
5112 bool bdrv_can_store_new_dirty_bitmap(BlockDriverState *bs, const char *name,
5113                                      uint32_t granularity, Error **errp)
5114 {
5115     BlockDriver *drv = bs->drv;
5116 
5117     if (!drv) {
5118         error_setg_errno(errp, ENOMEDIUM,
5119                          "Can't store persistent bitmaps to %s",
5120                          bdrv_get_device_or_node_name(bs));
5121         return false;
5122     }
5123 
5124     if (!drv->bdrv_can_store_new_dirty_bitmap) {
5125         error_setg_errno(errp, ENOTSUP,
5126                          "Can't store persistent bitmaps to %s",
5127                          bdrv_get_device_or_node_name(bs));
5128         return false;
5129     }
5130 
5131     return drv->bdrv_can_store_new_dirty_bitmap(bs, name, granularity, errp);
5132 }
5133