xref: /openbmc/qemu/block.c (revision 7562f907)
1 /*
2  * QEMU System Emulator block driver
3  *
4  * Copyright (c) 2003 Fabrice Bellard
5  *
6  * Permission is hereby granted, free of charge, to any person obtaining a copy
7  * of this software and associated documentation files (the "Software"), to deal
8  * in the Software without restriction, including without limitation the rights
9  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
10  * copies of the Software, and to permit persons to whom the Software is
11  * furnished to do so, subject to the following conditions:
12  *
13  * The above copyright notice and this permission notice shall be included in
14  * all copies or substantial portions of the Software.
15  *
16  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
22  * THE SOFTWARE.
23  */
24 #include "qemu/osdep.h"
25 #include "block/trace.h"
26 #include "block/block_int.h"
27 #include "block/blockjob.h"
28 #include "block/nbd.h"
29 #include "qemu/error-report.h"
30 #include "module_block.h"
31 #include "qemu/module.h"
32 #include "qapi/qmp/qerror.h"
33 #include "qapi/qmp/qbool.h"
34 #include "qapi/qmp/qjson.h"
35 #include "sysemu/block-backend.h"
36 #include "sysemu/sysemu.h"
37 #include "qemu/notify.h"
38 #include "qemu/coroutine.h"
39 #include "block/qapi.h"
40 #include "qmp-commands.h"
41 #include "qemu/timer.h"
42 #include "qapi-event.h"
43 #include "qemu/cutils.h"
44 #include "qemu/id.h"
45 #include "qapi/util.h"
46 
47 #ifdef CONFIG_BSD
48 #include <sys/ioctl.h>
49 #include <sys/queue.h>
50 #ifndef __DragonFly__
51 #include <sys/disk.h>
52 #endif
53 #endif
54 
55 #ifdef _WIN32
56 #include <windows.h>
57 #endif
58 
59 #define NOT_DONE 0x7fffffff /* used while emulated sync operation in progress */
60 
61 static QTAILQ_HEAD(, BlockDriverState) graph_bdrv_states =
62     QTAILQ_HEAD_INITIALIZER(graph_bdrv_states);
63 
64 static QTAILQ_HEAD(, BlockDriverState) all_bdrv_states =
65     QTAILQ_HEAD_INITIALIZER(all_bdrv_states);
66 
67 static QLIST_HEAD(, BlockDriver) bdrv_drivers =
68     QLIST_HEAD_INITIALIZER(bdrv_drivers);
69 
70 static BlockDriverState *bdrv_open_inherit(const char *filename,
71                                            const char *reference,
72                                            QDict *options, int flags,
73                                            BlockDriverState *parent,
74                                            const BdrvChildRole *child_role,
75                                            Error **errp);
76 
77 /* If non-zero, use only whitelisted block drivers */
78 static int use_bdrv_whitelist;
79 
80 #ifdef _WIN32
81 static int is_windows_drive_prefix(const char *filename)
82 {
83     return (((filename[0] >= 'a' && filename[0] <= 'z') ||
84              (filename[0] >= 'A' && filename[0] <= 'Z')) &&
85             filename[1] == ':');
86 }
87 
88 int is_windows_drive(const char *filename)
89 {
90     if (is_windows_drive_prefix(filename) &&
91         filename[2] == '\0')
92         return 1;
93     if (strstart(filename, "\\\\.\\", NULL) ||
94         strstart(filename, "//./", NULL))
95         return 1;
96     return 0;
97 }
98 #endif
99 
100 size_t bdrv_opt_mem_align(BlockDriverState *bs)
101 {
102     if (!bs || !bs->drv) {
103         /* page size or 4k (hdd sector size) should be on the safe side */
104         return MAX(4096, getpagesize());
105     }
106 
107     return bs->bl.opt_mem_alignment;
108 }
109 
110 size_t bdrv_min_mem_align(BlockDriverState *bs)
111 {
112     if (!bs || !bs->drv) {
113         /* page size or 4k (hdd sector size) should be on the safe side */
114         return MAX(4096, getpagesize());
115     }
116 
117     return bs->bl.min_mem_alignment;
118 }
119 
120 /* check if the path starts with "<protocol>:" */
121 int path_has_protocol(const char *path)
122 {
123     const char *p;
124 
125 #ifdef _WIN32
126     if (is_windows_drive(path) ||
127         is_windows_drive_prefix(path)) {
128         return 0;
129     }
130     p = path + strcspn(path, ":/\\");
131 #else
132     p = path + strcspn(path, ":/");
133 #endif
134 
135     return *p == ':';
136 }
137 
138 int path_is_absolute(const char *path)
139 {
140 #ifdef _WIN32
141     /* specific case for names like: "\\.\d:" */
142     if (is_windows_drive(path) || is_windows_drive_prefix(path)) {
143         return 1;
144     }
145     return (*path == '/' || *path == '\\');
146 #else
147     return (*path == '/');
148 #endif
149 }
150 
151 /* if filename is absolute, just copy it to dest. Otherwise, build a
152    path to it by considering it is relative to base_path. URL are
153    supported. */
154 void path_combine(char *dest, int dest_size,
155                   const char *base_path,
156                   const char *filename)
157 {
158     const char *p, *p1;
159     int len;
160 
161     if (dest_size <= 0)
162         return;
163     if (path_is_absolute(filename)) {
164         pstrcpy(dest, dest_size, filename);
165     } else {
166         p = strchr(base_path, ':');
167         if (p)
168             p++;
169         else
170             p = base_path;
171         p1 = strrchr(base_path, '/');
172 #ifdef _WIN32
173         {
174             const char *p2;
175             p2 = strrchr(base_path, '\\');
176             if (!p1 || p2 > p1)
177                 p1 = p2;
178         }
179 #endif
180         if (p1)
181             p1++;
182         else
183             p1 = base_path;
184         if (p1 > p)
185             p = p1;
186         len = p - base_path;
187         if (len > dest_size - 1)
188             len = dest_size - 1;
189         memcpy(dest, base_path, len);
190         dest[len] = '\0';
191         pstrcat(dest, dest_size, filename);
192     }
193 }
194 
195 void bdrv_get_full_backing_filename_from_filename(const char *backed,
196                                                   const char *backing,
197                                                   char *dest, size_t sz,
198                                                   Error **errp)
199 {
200     if (backing[0] == '\0' || path_has_protocol(backing) ||
201         path_is_absolute(backing))
202     {
203         pstrcpy(dest, sz, backing);
204     } else if (backed[0] == '\0' || strstart(backed, "json:", NULL)) {
205         error_setg(errp, "Cannot use relative backing file names for '%s'",
206                    backed);
207     } else {
208         path_combine(dest, sz, backed, backing);
209     }
210 }
211 
212 void bdrv_get_full_backing_filename(BlockDriverState *bs, char *dest, size_t sz,
213                                     Error **errp)
214 {
215     char *backed = bs->exact_filename[0] ? bs->exact_filename : bs->filename;
216 
217     bdrv_get_full_backing_filename_from_filename(backed, bs->backing_file,
218                                                  dest, sz, errp);
219 }
220 
221 void bdrv_register(BlockDriver *bdrv)
222 {
223     QLIST_INSERT_HEAD(&bdrv_drivers, bdrv, list);
224 }
225 
226 BlockDriverState *bdrv_new(void)
227 {
228     BlockDriverState *bs;
229     int i;
230 
231     bs = g_new0(BlockDriverState, 1);
232     QLIST_INIT(&bs->dirty_bitmaps);
233     for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
234         QLIST_INIT(&bs->op_blockers[i]);
235     }
236     notifier_with_return_list_init(&bs->before_write_notifiers);
237     bs->refcnt = 1;
238     bs->aio_context = qemu_get_aio_context();
239 
240     qemu_co_queue_init(&bs->flush_queue);
241 
242     QTAILQ_INSERT_TAIL(&all_bdrv_states, bs, bs_list);
243 
244     return bs;
245 }
246 
247 static BlockDriver *bdrv_do_find_format(const char *format_name)
248 {
249     BlockDriver *drv1;
250 
251     QLIST_FOREACH(drv1, &bdrv_drivers, list) {
252         if (!strcmp(drv1->format_name, format_name)) {
253             return drv1;
254         }
255     }
256 
257     return NULL;
258 }
259 
260 BlockDriver *bdrv_find_format(const char *format_name)
261 {
262     BlockDriver *drv1;
263     int i;
264 
265     drv1 = bdrv_do_find_format(format_name);
266     if (drv1) {
267         return drv1;
268     }
269 
270     /* The driver isn't registered, maybe we need to load a module */
271     for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); ++i) {
272         if (!strcmp(block_driver_modules[i].format_name, format_name)) {
273             block_module_load_one(block_driver_modules[i].library_name);
274             break;
275         }
276     }
277 
278     return bdrv_do_find_format(format_name);
279 }
280 
281 static int bdrv_is_whitelisted(BlockDriver *drv, bool read_only)
282 {
283     static const char *whitelist_rw[] = {
284         CONFIG_BDRV_RW_WHITELIST
285     };
286     static const char *whitelist_ro[] = {
287         CONFIG_BDRV_RO_WHITELIST
288     };
289     const char **p;
290 
291     if (!whitelist_rw[0] && !whitelist_ro[0]) {
292         return 1;               /* no whitelist, anything goes */
293     }
294 
295     for (p = whitelist_rw; *p; p++) {
296         if (!strcmp(drv->format_name, *p)) {
297             return 1;
298         }
299     }
300     if (read_only) {
301         for (p = whitelist_ro; *p; p++) {
302             if (!strcmp(drv->format_name, *p)) {
303                 return 1;
304             }
305         }
306     }
307     return 0;
308 }
309 
310 bool bdrv_uses_whitelist(void)
311 {
312     return use_bdrv_whitelist;
313 }
314 
315 typedef struct CreateCo {
316     BlockDriver *drv;
317     char *filename;
318     QemuOpts *opts;
319     int ret;
320     Error *err;
321 } CreateCo;
322 
323 static void coroutine_fn bdrv_create_co_entry(void *opaque)
324 {
325     Error *local_err = NULL;
326     int ret;
327 
328     CreateCo *cco = opaque;
329     assert(cco->drv);
330 
331     ret = cco->drv->bdrv_create(cco->filename, cco->opts, &local_err);
332     error_propagate(&cco->err, local_err);
333     cco->ret = ret;
334 }
335 
336 int bdrv_create(BlockDriver *drv, const char* filename,
337                 QemuOpts *opts, Error **errp)
338 {
339     int ret;
340 
341     Coroutine *co;
342     CreateCo cco = {
343         .drv = drv,
344         .filename = g_strdup(filename),
345         .opts = opts,
346         .ret = NOT_DONE,
347         .err = NULL,
348     };
349 
350     if (!drv->bdrv_create) {
351         error_setg(errp, "Driver '%s' does not support image creation", drv->format_name);
352         ret = -ENOTSUP;
353         goto out;
354     }
355 
356     if (qemu_in_coroutine()) {
357         /* Fast-path if already in coroutine context */
358         bdrv_create_co_entry(&cco);
359     } else {
360         co = qemu_coroutine_create(bdrv_create_co_entry, &cco);
361         qemu_coroutine_enter(co);
362         while (cco.ret == NOT_DONE) {
363             aio_poll(qemu_get_aio_context(), true);
364         }
365     }
366 
367     ret = cco.ret;
368     if (ret < 0) {
369         if (cco.err) {
370             error_propagate(errp, cco.err);
371         } else {
372             error_setg_errno(errp, -ret, "Could not create image");
373         }
374     }
375 
376 out:
377     g_free(cco.filename);
378     return ret;
379 }
380 
381 int bdrv_create_file(const char *filename, QemuOpts *opts, Error **errp)
382 {
383     BlockDriver *drv;
384     Error *local_err = NULL;
385     int ret;
386 
387     drv = bdrv_find_protocol(filename, true, errp);
388     if (drv == NULL) {
389         return -ENOENT;
390     }
391 
392     ret = bdrv_create(drv, filename, opts, &local_err);
393     error_propagate(errp, local_err);
394     return ret;
395 }
396 
397 /**
398  * Try to get @bs's logical and physical block size.
399  * On success, store them in @bsz struct and return 0.
400  * On failure return -errno.
401  * @bs must not be empty.
402  */
403 int bdrv_probe_blocksizes(BlockDriverState *bs, BlockSizes *bsz)
404 {
405     BlockDriver *drv = bs->drv;
406 
407     if (drv && drv->bdrv_probe_blocksizes) {
408         return drv->bdrv_probe_blocksizes(bs, bsz);
409     }
410 
411     return -ENOTSUP;
412 }
413 
414 /**
415  * Try to get @bs's geometry (cyls, heads, sectors).
416  * On success, store them in @geo struct and return 0.
417  * On failure return -errno.
418  * @bs must not be empty.
419  */
420 int bdrv_probe_geometry(BlockDriverState *bs, HDGeometry *geo)
421 {
422     BlockDriver *drv = bs->drv;
423 
424     if (drv && drv->bdrv_probe_geometry) {
425         return drv->bdrv_probe_geometry(bs, geo);
426     }
427 
428     return -ENOTSUP;
429 }
430 
431 /*
432  * Create a uniquely-named empty temporary file.
433  * Return 0 upon success, otherwise a negative errno value.
434  */
435 int get_tmp_filename(char *filename, int size)
436 {
437 #ifdef _WIN32
438     char temp_dir[MAX_PATH];
439     /* GetTempFileName requires that its output buffer (4th param)
440        have length MAX_PATH or greater.  */
441     assert(size >= MAX_PATH);
442     return (GetTempPath(MAX_PATH, temp_dir)
443             && GetTempFileName(temp_dir, "qem", 0, filename)
444             ? 0 : -GetLastError());
445 #else
446     int fd;
447     const char *tmpdir;
448     tmpdir = getenv("TMPDIR");
449     if (!tmpdir) {
450         tmpdir = "/var/tmp";
451     }
452     if (snprintf(filename, size, "%s/vl.XXXXXX", tmpdir) >= size) {
453         return -EOVERFLOW;
454     }
455     fd = mkstemp(filename);
456     if (fd < 0) {
457         return -errno;
458     }
459     if (close(fd) != 0) {
460         unlink(filename);
461         return -errno;
462     }
463     return 0;
464 #endif
465 }
466 
467 /*
468  * Detect host devices. By convention, /dev/cdrom[N] is always
469  * recognized as a host CDROM.
470  */
471 static BlockDriver *find_hdev_driver(const char *filename)
472 {
473     int score_max = 0, score;
474     BlockDriver *drv = NULL, *d;
475 
476     QLIST_FOREACH(d, &bdrv_drivers, list) {
477         if (d->bdrv_probe_device) {
478             score = d->bdrv_probe_device(filename);
479             if (score > score_max) {
480                 score_max = score;
481                 drv = d;
482             }
483         }
484     }
485 
486     return drv;
487 }
488 
489 static BlockDriver *bdrv_do_find_protocol(const char *protocol)
490 {
491     BlockDriver *drv1;
492 
493     QLIST_FOREACH(drv1, &bdrv_drivers, list) {
494         if (drv1->protocol_name && !strcmp(drv1->protocol_name, protocol)) {
495             return drv1;
496         }
497     }
498 
499     return NULL;
500 }
501 
502 BlockDriver *bdrv_find_protocol(const char *filename,
503                                 bool allow_protocol_prefix,
504                                 Error **errp)
505 {
506     BlockDriver *drv1;
507     char protocol[128];
508     int len;
509     const char *p;
510     int i;
511 
512     /* TODO Drivers without bdrv_file_open must be specified explicitly */
513 
514     /*
515      * XXX(hch): we really should not let host device detection
516      * override an explicit protocol specification, but moving this
517      * later breaks access to device names with colons in them.
518      * Thanks to the brain-dead persistent naming schemes on udev-
519      * based Linux systems those actually are quite common.
520      */
521     drv1 = find_hdev_driver(filename);
522     if (drv1) {
523         return drv1;
524     }
525 
526     if (!path_has_protocol(filename) || !allow_protocol_prefix) {
527         return &bdrv_file;
528     }
529 
530     p = strchr(filename, ':');
531     assert(p != NULL);
532     len = p - filename;
533     if (len > sizeof(protocol) - 1)
534         len = sizeof(protocol) - 1;
535     memcpy(protocol, filename, len);
536     protocol[len] = '\0';
537 
538     drv1 = bdrv_do_find_protocol(protocol);
539     if (drv1) {
540         return drv1;
541     }
542 
543     for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); ++i) {
544         if (block_driver_modules[i].protocol_name &&
545             !strcmp(block_driver_modules[i].protocol_name, protocol)) {
546             block_module_load_one(block_driver_modules[i].library_name);
547             break;
548         }
549     }
550 
551     drv1 = bdrv_do_find_protocol(protocol);
552     if (!drv1) {
553         error_setg(errp, "Unknown protocol '%s'", protocol);
554     }
555     return drv1;
556 }
557 
558 /*
559  * Guess image format by probing its contents.
560  * This is not a good idea when your image is raw (CVE-2008-2004), but
561  * we do it anyway for backward compatibility.
562  *
563  * @buf         contains the image's first @buf_size bytes.
564  * @buf_size    is the buffer size in bytes (generally BLOCK_PROBE_BUF_SIZE,
565  *              but can be smaller if the image file is smaller)
566  * @filename    is its filename.
567  *
568  * For all block drivers, call the bdrv_probe() method to get its
569  * probing score.
570  * Return the first block driver with the highest probing score.
571  */
572 BlockDriver *bdrv_probe_all(const uint8_t *buf, int buf_size,
573                             const char *filename)
574 {
575     int score_max = 0, score;
576     BlockDriver *drv = NULL, *d;
577 
578     QLIST_FOREACH(d, &bdrv_drivers, list) {
579         if (d->bdrv_probe) {
580             score = d->bdrv_probe(buf, buf_size, filename);
581             if (score > score_max) {
582                 score_max = score;
583                 drv = d;
584             }
585         }
586     }
587 
588     return drv;
589 }
590 
591 static int find_image_format(BlockBackend *file, const char *filename,
592                              BlockDriver **pdrv, Error **errp)
593 {
594     BlockDriver *drv;
595     uint8_t buf[BLOCK_PROBE_BUF_SIZE];
596     int ret = 0;
597 
598     /* Return the raw BlockDriver * to scsi-generic devices or empty drives */
599     if (blk_is_sg(file) || !blk_is_inserted(file) || blk_getlength(file) == 0) {
600         *pdrv = &bdrv_raw;
601         return ret;
602     }
603 
604     ret = blk_pread(file, 0, buf, sizeof(buf));
605     if (ret < 0) {
606         error_setg_errno(errp, -ret, "Could not read image for determining its "
607                          "format");
608         *pdrv = NULL;
609         return ret;
610     }
611 
612     drv = bdrv_probe_all(buf, ret, filename);
613     if (!drv) {
614         error_setg(errp, "Could not determine image format: No compatible "
615                    "driver found");
616         ret = -ENOENT;
617     }
618     *pdrv = drv;
619     return ret;
620 }
621 
622 /**
623  * Set the current 'total_sectors' value
624  * Return 0 on success, -errno on error.
625  */
626 static int refresh_total_sectors(BlockDriverState *bs, int64_t hint)
627 {
628     BlockDriver *drv = bs->drv;
629 
630     /* Do not attempt drv->bdrv_getlength() on scsi-generic devices */
631     if (bdrv_is_sg(bs))
632         return 0;
633 
634     /* query actual device if possible, otherwise just trust the hint */
635     if (drv->bdrv_getlength) {
636         int64_t length = drv->bdrv_getlength(bs);
637         if (length < 0) {
638             return length;
639         }
640         hint = DIV_ROUND_UP(length, BDRV_SECTOR_SIZE);
641     }
642 
643     bs->total_sectors = hint;
644     return 0;
645 }
646 
647 /**
648  * Combines a QDict of new block driver @options with any missing options taken
649  * from @old_options, so that leaving out an option defaults to its old value.
650  */
651 static void bdrv_join_options(BlockDriverState *bs, QDict *options,
652                               QDict *old_options)
653 {
654     if (bs->drv && bs->drv->bdrv_join_options) {
655         bs->drv->bdrv_join_options(options, old_options);
656     } else {
657         qdict_join(options, old_options, false);
658     }
659 }
660 
661 /**
662  * Set open flags for a given discard mode
663  *
664  * Return 0 on success, -1 if the discard mode was invalid.
665  */
666 int bdrv_parse_discard_flags(const char *mode, int *flags)
667 {
668     *flags &= ~BDRV_O_UNMAP;
669 
670     if (!strcmp(mode, "off") || !strcmp(mode, "ignore")) {
671         /* do nothing */
672     } else if (!strcmp(mode, "on") || !strcmp(mode, "unmap")) {
673         *flags |= BDRV_O_UNMAP;
674     } else {
675         return -1;
676     }
677 
678     return 0;
679 }
680 
681 /**
682  * Set open flags for a given cache mode
683  *
684  * Return 0 on success, -1 if the cache mode was invalid.
685  */
686 int bdrv_parse_cache_mode(const char *mode, int *flags, bool *writethrough)
687 {
688     *flags &= ~BDRV_O_CACHE_MASK;
689 
690     if (!strcmp(mode, "off") || !strcmp(mode, "none")) {
691         *writethrough = false;
692         *flags |= BDRV_O_NOCACHE;
693     } else if (!strcmp(mode, "directsync")) {
694         *writethrough = true;
695         *flags |= BDRV_O_NOCACHE;
696     } else if (!strcmp(mode, "writeback")) {
697         *writethrough = false;
698     } else if (!strcmp(mode, "unsafe")) {
699         *writethrough = false;
700         *flags |= BDRV_O_NO_FLUSH;
701     } else if (!strcmp(mode, "writethrough")) {
702         *writethrough = true;
703     } else {
704         return -1;
705     }
706 
707     return 0;
708 }
709 
710 static void bdrv_child_cb_drained_begin(BdrvChild *child)
711 {
712     BlockDriverState *bs = child->opaque;
713     bdrv_drained_begin(bs);
714 }
715 
716 static void bdrv_child_cb_drained_end(BdrvChild *child)
717 {
718     BlockDriverState *bs = child->opaque;
719     bdrv_drained_end(bs);
720 }
721 
722 /*
723  * Returns the options and flags that a temporary snapshot should get, based on
724  * the originally requested flags (the originally requested image will have
725  * flags like a backing file)
726  */
727 static void bdrv_temp_snapshot_options(int *child_flags, QDict *child_options,
728                                        int parent_flags, QDict *parent_options)
729 {
730     *child_flags = (parent_flags & ~BDRV_O_SNAPSHOT) | BDRV_O_TEMPORARY;
731 
732     /* For temporary files, unconditional cache=unsafe is fine */
733     qdict_set_default_str(child_options, BDRV_OPT_CACHE_DIRECT, "off");
734     qdict_set_default_str(child_options, BDRV_OPT_CACHE_NO_FLUSH, "on");
735 
736     /* Copy the read-only option from the parent */
737     qdict_copy_default(child_options, parent_options, BDRV_OPT_READ_ONLY);
738 
739     /* aio=native doesn't work for cache.direct=off, so disable it for the
740      * temporary snapshot */
741     *child_flags &= ~BDRV_O_NATIVE_AIO;
742 }
743 
744 /*
745  * Returns the options and flags that bs->file should get if a protocol driver
746  * is expected, based on the given options and flags for the parent BDS
747  */
748 static void bdrv_inherited_options(int *child_flags, QDict *child_options,
749                                    int parent_flags, QDict *parent_options)
750 {
751     int flags = parent_flags;
752 
753     /* Enable protocol handling, disable format probing for bs->file */
754     flags |= BDRV_O_PROTOCOL;
755 
756     /* If the cache mode isn't explicitly set, inherit direct and no-flush from
757      * the parent. */
758     qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_DIRECT);
759     qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_NO_FLUSH);
760 
761     /* Inherit the read-only option from the parent if it's not set */
762     qdict_copy_default(child_options, parent_options, BDRV_OPT_READ_ONLY);
763 
764     /* Our block drivers take care to send flushes and respect unmap policy,
765      * so we can default to enable both on lower layers regardless of the
766      * corresponding parent options. */
767     qdict_set_default_str(child_options, BDRV_OPT_DISCARD, "unmap");
768 
769     /* Clear flags that only apply to the top layer */
770     flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_COPY_ON_READ |
771                BDRV_O_NO_IO);
772 
773     *child_flags = flags;
774 }
775 
776 const BdrvChildRole child_file = {
777     .inherit_options = bdrv_inherited_options,
778     .drained_begin   = bdrv_child_cb_drained_begin,
779     .drained_end     = bdrv_child_cb_drained_end,
780 };
781 
782 /*
783  * Returns the options and flags that bs->file should get if the use of formats
784  * (and not only protocols) is permitted for it, based on the given options and
785  * flags for the parent BDS
786  */
787 static void bdrv_inherited_fmt_options(int *child_flags, QDict *child_options,
788                                        int parent_flags, QDict *parent_options)
789 {
790     child_file.inherit_options(child_flags, child_options,
791                                parent_flags, parent_options);
792 
793     *child_flags &= ~(BDRV_O_PROTOCOL | BDRV_O_NO_IO);
794 }
795 
796 const BdrvChildRole child_format = {
797     .inherit_options = bdrv_inherited_fmt_options,
798     .drained_begin   = bdrv_child_cb_drained_begin,
799     .drained_end     = bdrv_child_cb_drained_end,
800 };
801 
802 /*
803  * Returns the options and flags that bs->backing should get, based on the
804  * given options and flags for the parent BDS
805  */
806 static void bdrv_backing_options(int *child_flags, QDict *child_options,
807                                  int parent_flags, QDict *parent_options)
808 {
809     int flags = parent_flags;
810 
811     /* The cache mode is inherited unmodified for backing files; except WCE,
812      * which is only applied on the top level (BlockBackend) */
813     qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_DIRECT);
814     qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_NO_FLUSH);
815 
816     /* backing files always opened read-only */
817     qdict_set_default_str(child_options, BDRV_OPT_READ_ONLY, "on");
818     flags &= ~BDRV_O_COPY_ON_READ;
819 
820     /* snapshot=on is handled on the top layer */
821     flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_TEMPORARY);
822 
823     *child_flags = flags;
824 }
825 
826 static const BdrvChildRole child_backing = {
827     .inherit_options = bdrv_backing_options,
828     .drained_begin   = bdrv_child_cb_drained_begin,
829     .drained_end     = bdrv_child_cb_drained_end,
830 };
831 
832 static int bdrv_open_flags(BlockDriverState *bs, int flags)
833 {
834     int open_flags = flags;
835 
836     /*
837      * Clear flags that are internal to the block layer before opening the
838      * image.
839      */
840     open_flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_PROTOCOL);
841 
842     /*
843      * Snapshots should be writable.
844      */
845     if (flags & BDRV_O_TEMPORARY) {
846         open_flags |= BDRV_O_RDWR;
847     }
848 
849     return open_flags;
850 }
851 
852 static void update_flags_from_options(int *flags, QemuOpts *opts)
853 {
854     *flags &= ~BDRV_O_CACHE_MASK;
855 
856     assert(qemu_opt_find(opts, BDRV_OPT_CACHE_NO_FLUSH));
857     if (qemu_opt_get_bool(opts, BDRV_OPT_CACHE_NO_FLUSH, false)) {
858         *flags |= BDRV_O_NO_FLUSH;
859     }
860 
861     assert(qemu_opt_find(opts, BDRV_OPT_CACHE_DIRECT));
862     if (qemu_opt_get_bool(opts, BDRV_OPT_CACHE_DIRECT, false)) {
863         *flags |= BDRV_O_NOCACHE;
864     }
865 
866     *flags &= ~BDRV_O_RDWR;
867 
868     assert(qemu_opt_find(opts, BDRV_OPT_READ_ONLY));
869     if (!qemu_opt_get_bool(opts, BDRV_OPT_READ_ONLY, false)) {
870         *flags |= BDRV_O_RDWR;
871     }
872 
873 }
874 
875 static void update_options_from_flags(QDict *options, int flags)
876 {
877     if (!qdict_haskey(options, BDRV_OPT_CACHE_DIRECT)) {
878         qdict_put(options, BDRV_OPT_CACHE_DIRECT,
879                   qbool_from_bool(flags & BDRV_O_NOCACHE));
880     }
881     if (!qdict_haskey(options, BDRV_OPT_CACHE_NO_FLUSH)) {
882         qdict_put(options, BDRV_OPT_CACHE_NO_FLUSH,
883                   qbool_from_bool(flags & BDRV_O_NO_FLUSH));
884     }
885     if (!qdict_haskey(options, BDRV_OPT_READ_ONLY)) {
886         qdict_put(options, BDRV_OPT_READ_ONLY,
887                   qbool_from_bool(!(flags & BDRV_O_RDWR)));
888     }
889 }
890 
891 static void bdrv_assign_node_name(BlockDriverState *bs,
892                                   const char *node_name,
893                                   Error **errp)
894 {
895     char *gen_node_name = NULL;
896 
897     if (!node_name) {
898         node_name = gen_node_name = id_generate(ID_BLOCK);
899     } else if (!id_wellformed(node_name)) {
900         /*
901          * Check for empty string or invalid characters, but not if it is
902          * generated (generated names use characters not available to the user)
903          */
904         error_setg(errp, "Invalid node name");
905         return;
906     }
907 
908     /* takes care of avoiding namespaces collisions */
909     if (blk_by_name(node_name)) {
910         error_setg(errp, "node-name=%s is conflicting with a device id",
911                    node_name);
912         goto out;
913     }
914 
915     /* takes care of avoiding duplicates node names */
916     if (bdrv_find_node(node_name)) {
917         error_setg(errp, "Duplicate node name");
918         goto out;
919     }
920 
921     /* copy node name into the bs and insert it into the graph list */
922     pstrcpy(bs->node_name, sizeof(bs->node_name), node_name);
923     QTAILQ_INSERT_TAIL(&graph_bdrv_states, bs, node_list);
924 out:
925     g_free(gen_node_name);
926 }
927 
928 static int bdrv_open_driver(BlockDriverState *bs, BlockDriver *drv,
929                             const char *node_name, QDict *options,
930                             int open_flags, Error **errp)
931 {
932     Error *local_err = NULL;
933     int ret;
934 
935     bdrv_assign_node_name(bs, node_name, &local_err);
936     if (local_err) {
937         error_propagate(errp, local_err);
938         return -EINVAL;
939     }
940 
941     bs->drv = drv;
942     bs->read_only = !(bs->open_flags & BDRV_O_RDWR);
943     bs->opaque = g_malloc0(drv->instance_size);
944 
945     if (drv->bdrv_file_open) {
946         assert(!drv->bdrv_needs_filename || bs->filename[0]);
947         ret = drv->bdrv_file_open(bs, options, open_flags, &local_err);
948     } else if (drv->bdrv_open) {
949         ret = drv->bdrv_open(bs, options, open_flags, &local_err);
950     } else {
951         ret = 0;
952     }
953 
954     if (ret < 0) {
955         if (local_err) {
956             error_propagate(errp, local_err);
957         } else if (bs->filename[0]) {
958             error_setg_errno(errp, -ret, "Could not open '%s'", bs->filename);
959         } else {
960             error_setg_errno(errp, -ret, "Could not open image");
961         }
962         goto free_and_fail;
963     }
964 
965     ret = refresh_total_sectors(bs, bs->total_sectors);
966     if (ret < 0) {
967         error_setg_errno(errp, -ret, "Could not refresh total sector count");
968         goto free_and_fail;
969     }
970 
971     bdrv_refresh_limits(bs, &local_err);
972     if (local_err) {
973         error_propagate(errp, local_err);
974         ret = -EINVAL;
975         goto free_and_fail;
976     }
977 
978     assert(bdrv_opt_mem_align(bs) != 0);
979     assert(bdrv_min_mem_align(bs) != 0);
980     assert(is_power_of_2(bs->bl.request_alignment));
981 
982     return 0;
983 
984 free_and_fail:
985     /* FIXME Close bs first if already opened*/
986     g_free(bs->opaque);
987     bs->opaque = NULL;
988     bs->drv = NULL;
989     return ret;
990 }
991 
992 BlockDriverState *bdrv_new_open_driver(BlockDriver *drv, const char *node_name,
993                                        int flags, Error **errp)
994 {
995     BlockDriverState *bs;
996     int ret;
997 
998     bs = bdrv_new();
999     bs->open_flags = flags;
1000     bs->explicit_options = qdict_new();
1001     bs->options = qdict_new();
1002     bs->opaque = NULL;
1003 
1004     update_options_from_flags(bs->options, flags);
1005 
1006     ret = bdrv_open_driver(bs, drv, node_name, bs->options, flags, errp);
1007     if (ret < 0) {
1008         QDECREF(bs->explicit_options);
1009         QDECREF(bs->options);
1010         bdrv_unref(bs);
1011         return NULL;
1012     }
1013 
1014     return bs;
1015 }
1016 
1017 QemuOptsList bdrv_runtime_opts = {
1018     .name = "bdrv_common",
1019     .head = QTAILQ_HEAD_INITIALIZER(bdrv_runtime_opts.head),
1020     .desc = {
1021         {
1022             .name = "node-name",
1023             .type = QEMU_OPT_STRING,
1024             .help = "Node name of the block device node",
1025         },
1026         {
1027             .name = "driver",
1028             .type = QEMU_OPT_STRING,
1029             .help = "Block driver to use for the node",
1030         },
1031         {
1032             .name = BDRV_OPT_CACHE_DIRECT,
1033             .type = QEMU_OPT_BOOL,
1034             .help = "Bypass software writeback cache on the host",
1035         },
1036         {
1037             .name = BDRV_OPT_CACHE_NO_FLUSH,
1038             .type = QEMU_OPT_BOOL,
1039             .help = "Ignore flush requests",
1040         },
1041         {
1042             .name = BDRV_OPT_READ_ONLY,
1043             .type = QEMU_OPT_BOOL,
1044             .help = "Node is opened in read-only mode",
1045         },
1046         {
1047             .name = "detect-zeroes",
1048             .type = QEMU_OPT_STRING,
1049             .help = "try to optimize zero writes (off, on, unmap)",
1050         },
1051         {
1052             .name = "discard",
1053             .type = QEMU_OPT_STRING,
1054             .help = "discard operation (ignore/off, unmap/on)",
1055         },
1056         { /* end of list */ }
1057     },
1058 };
1059 
1060 /*
1061  * Common part for opening disk images and files
1062  *
1063  * Removes all processed options from *options.
1064  */
1065 static int bdrv_open_common(BlockDriverState *bs, BlockBackend *file,
1066                             QDict *options, Error **errp)
1067 {
1068     int ret, open_flags;
1069     const char *filename;
1070     const char *driver_name = NULL;
1071     const char *node_name = NULL;
1072     const char *discard;
1073     const char *detect_zeroes;
1074     QemuOpts *opts;
1075     BlockDriver *drv;
1076     Error *local_err = NULL;
1077 
1078     assert(bs->file == NULL);
1079     assert(options != NULL && bs->options != options);
1080 
1081     opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
1082     qemu_opts_absorb_qdict(opts, options, &local_err);
1083     if (local_err) {
1084         error_propagate(errp, local_err);
1085         ret = -EINVAL;
1086         goto fail_opts;
1087     }
1088 
1089     update_flags_from_options(&bs->open_flags, opts);
1090 
1091     driver_name = qemu_opt_get(opts, "driver");
1092     drv = bdrv_find_format(driver_name);
1093     assert(drv != NULL);
1094 
1095     if (file != NULL) {
1096         filename = blk_bs(file)->filename;
1097     } else {
1098         filename = qdict_get_try_str(options, "filename");
1099     }
1100 
1101     if (drv->bdrv_needs_filename && !filename) {
1102         error_setg(errp, "The '%s' block driver requires a file name",
1103                    drv->format_name);
1104         ret = -EINVAL;
1105         goto fail_opts;
1106     }
1107 
1108     trace_bdrv_open_common(bs, filename ?: "", bs->open_flags,
1109                            drv->format_name);
1110 
1111     bs->read_only = !(bs->open_flags & BDRV_O_RDWR);
1112 
1113     if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv, bs->read_only)) {
1114         error_setg(errp,
1115                    !bs->read_only && bdrv_is_whitelisted(drv, true)
1116                         ? "Driver '%s' can only be used for read-only devices"
1117                         : "Driver '%s' is not whitelisted",
1118                    drv->format_name);
1119         ret = -ENOTSUP;
1120         goto fail_opts;
1121     }
1122 
1123     assert(bs->copy_on_read == 0); /* bdrv_new() and bdrv_close() make it so */
1124     if (bs->open_flags & BDRV_O_COPY_ON_READ) {
1125         if (!bs->read_only) {
1126             bdrv_enable_copy_on_read(bs);
1127         } else {
1128             error_setg(errp, "Can't use copy-on-read on read-only device");
1129             ret = -EINVAL;
1130             goto fail_opts;
1131         }
1132     }
1133 
1134     discard = qemu_opt_get(opts, "discard");
1135     if (discard != NULL) {
1136         if (bdrv_parse_discard_flags(discard, &bs->open_flags) != 0) {
1137             error_setg(errp, "Invalid discard option");
1138             ret = -EINVAL;
1139             goto fail_opts;
1140         }
1141     }
1142 
1143     detect_zeroes = qemu_opt_get(opts, "detect-zeroes");
1144     if (detect_zeroes) {
1145         BlockdevDetectZeroesOptions value =
1146             qapi_enum_parse(BlockdevDetectZeroesOptions_lookup,
1147                             detect_zeroes,
1148                             BLOCKDEV_DETECT_ZEROES_OPTIONS__MAX,
1149                             BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF,
1150                             &local_err);
1151         if (local_err) {
1152             error_propagate(errp, local_err);
1153             ret = -EINVAL;
1154             goto fail_opts;
1155         }
1156 
1157         if (value == BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP &&
1158             !(bs->open_flags & BDRV_O_UNMAP))
1159         {
1160             error_setg(errp, "setting detect-zeroes to unmap is not allowed "
1161                              "without setting discard operation to unmap");
1162             ret = -EINVAL;
1163             goto fail_opts;
1164         }
1165 
1166         bs->detect_zeroes = value;
1167     }
1168 
1169     if (filename != NULL) {
1170         pstrcpy(bs->filename, sizeof(bs->filename), filename);
1171     } else {
1172         bs->filename[0] = '\0';
1173     }
1174     pstrcpy(bs->exact_filename, sizeof(bs->exact_filename), bs->filename);
1175 
1176     /* Open the image, either directly or using a protocol */
1177     open_flags = bdrv_open_flags(bs, bs->open_flags);
1178     node_name = qemu_opt_get(opts, "node-name");
1179 
1180     assert(!drv->bdrv_file_open || file == NULL);
1181     ret = bdrv_open_driver(bs, drv, node_name, options, open_flags, errp);
1182     if (ret < 0) {
1183         goto fail_opts;
1184     }
1185 
1186     qemu_opts_del(opts);
1187     return 0;
1188 
1189 fail_opts:
1190     qemu_opts_del(opts);
1191     return ret;
1192 }
1193 
1194 static QDict *parse_json_filename(const char *filename, Error **errp)
1195 {
1196     QObject *options_obj;
1197     QDict *options;
1198     int ret;
1199 
1200     ret = strstart(filename, "json:", &filename);
1201     assert(ret);
1202 
1203     options_obj = qobject_from_json(filename);
1204     if (!options_obj) {
1205         error_setg(errp, "Could not parse the JSON options");
1206         return NULL;
1207     }
1208 
1209     options = qobject_to_qdict(options_obj);
1210     if (!options) {
1211         qobject_decref(options_obj);
1212         error_setg(errp, "Invalid JSON object given");
1213         return NULL;
1214     }
1215 
1216     qdict_flatten(options);
1217 
1218     return options;
1219 }
1220 
1221 static void parse_json_protocol(QDict *options, const char **pfilename,
1222                                 Error **errp)
1223 {
1224     QDict *json_options;
1225     Error *local_err = NULL;
1226 
1227     /* Parse json: pseudo-protocol */
1228     if (!*pfilename || !g_str_has_prefix(*pfilename, "json:")) {
1229         return;
1230     }
1231 
1232     json_options = parse_json_filename(*pfilename, &local_err);
1233     if (local_err) {
1234         error_propagate(errp, local_err);
1235         return;
1236     }
1237 
1238     /* Options given in the filename have lower priority than options
1239      * specified directly */
1240     qdict_join(options, json_options, false);
1241     QDECREF(json_options);
1242     *pfilename = NULL;
1243 }
1244 
1245 /*
1246  * Fills in default options for opening images and converts the legacy
1247  * filename/flags pair to option QDict entries.
1248  * The BDRV_O_PROTOCOL flag in *flags will be set or cleared accordingly if a
1249  * block driver has been specified explicitly.
1250  */
1251 static int bdrv_fill_options(QDict **options, const char *filename,
1252                              int *flags, Error **errp)
1253 {
1254     const char *drvname;
1255     bool protocol = *flags & BDRV_O_PROTOCOL;
1256     bool parse_filename = false;
1257     BlockDriver *drv = NULL;
1258     Error *local_err = NULL;
1259 
1260     drvname = qdict_get_try_str(*options, "driver");
1261     if (drvname) {
1262         drv = bdrv_find_format(drvname);
1263         if (!drv) {
1264             error_setg(errp, "Unknown driver '%s'", drvname);
1265             return -ENOENT;
1266         }
1267         /* If the user has explicitly specified the driver, this choice should
1268          * override the BDRV_O_PROTOCOL flag */
1269         protocol = drv->bdrv_file_open;
1270     }
1271 
1272     if (protocol) {
1273         *flags |= BDRV_O_PROTOCOL;
1274     } else {
1275         *flags &= ~BDRV_O_PROTOCOL;
1276     }
1277 
1278     /* Translate cache options from flags into options */
1279     update_options_from_flags(*options, *flags);
1280 
1281     /* Fetch the file name from the options QDict if necessary */
1282     if (protocol && filename) {
1283         if (!qdict_haskey(*options, "filename")) {
1284             qdict_put(*options, "filename", qstring_from_str(filename));
1285             parse_filename = true;
1286         } else {
1287             error_setg(errp, "Can't specify 'file' and 'filename' options at "
1288                              "the same time");
1289             return -EINVAL;
1290         }
1291     }
1292 
1293     /* Find the right block driver */
1294     filename = qdict_get_try_str(*options, "filename");
1295 
1296     if (!drvname && protocol) {
1297         if (filename) {
1298             drv = bdrv_find_protocol(filename, parse_filename, errp);
1299             if (!drv) {
1300                 return -EINVAL;
1301             }
1302 
1303             drvname = drv->format_name;
1304             qdict_put(*options, "driver", qstring_from_str(drvname));
1305         } else {
1306             error_setg(errp, "Must specify either driver or file");
1307             return -EINVAL;
1308         }
1309     }
1310 
1311     assert(drv || !protocol);
1312 
1313     /* Driver-specific filename parsing */
1314     if (drv && drv->bdrv_parse_filename && parse_filename) {
1315         drv->bdrv_parse_filename(filename, *options, &local_err);
1316         if (local_err) {
1317             error_propagate(errp, local_err);
1318             return -EINVAL;
1319         }
1320 
1321         if (!drv->bdrv_needs_filename) {
1322             qdict_del(*options, "filename");
1323         }
1324     }
1325 
1326     return 0;
1327 }
1328 
1329 static void bdrv_replace_child(BdrvChild *child, BlockDriverState *new_bs)
1330 {
1331     BlockDriverState *old_bs = child->bs;
1332 
1333     if (old_bs) {
1334         if (old_bs->quiesce_counter && child->role->drained_end) {
1335             child->role->drained_end(child);
1336         }
1337         QLIST_REMOVE(child, next_parent);
1338     }
1339 
1340     child->bs = new_bs;
1341 
1342     if (new_bs) {
1343         QLIST_INSERT_HEAD(&new_bs->parents, child, next_parent);
1344         if (new_bs->quiesce_counter && child->role->drained_begin) {
1345             child->role->drained_begin(child);
1346         }
1347     }
1348 }
1349 
1350 BdrvChild *bdrv_root_attach_child(BlockDriverState *child_bs,
1351                                   const char *child_name,
1352                                   const BdrvChildRole *child_role,
1353                                   void *opaque)
1354 {
1355     BdrvChild *child = g_new(BdrvChild, 1);
1356     *child = (BdrvChild) {
1357         .bs     = NULL,
1358         .name   = g_strdup(child_name),
1359         .role   = child_role,
1360         .opaque = opaque,
1361     };
1362 
1363     bdrv_replace_child(child, child_bs);
1364 
1365     return child;
1366 }
1367 
1368 BdrvChild *bdrv_attach_child(BlockDriverState *parent_bs,
1369                              BlockDriverState *child_bs,
1370                              const char *child_name,
1371                              const BdrvChildRole *child_role)
1372 {
1373     BdrvChild *child = bdrv_root_attach_child(child_bs, child_name, child_role,
1374                                               parent_bs);
1375     QLIST_INSERT_HEAD(&parent_bs->children, child, next);
1376     return child;
1377 }
1378 
1379 static void bdrv_detach_child(BdrvChild *child)
1380 {
1381     if (child->next.le_prev) {
1382         QLIST_REMOVE(child, next);
1383         child->next.le_prev = NULL;
1384     }
1385 
1386     bdrv_replace_child(child, NULL);
1387 
1388     g_free(child->name);
1389     g_free(child);
1390 }
1391 
1392 void bdrv_root_unref_child(BdrvChild *child)
1393 {
1394     BlockDriverState *child_bs;
1395 
1396     child_bs = child->bs;
1397     bdrv_detach_child(child);
1398     bdrv_unref(child_bs);
1399 }
1400 
1401 void bdrv_unref_child(BlockDriverState *parent, BdrvChild *child)
1402 {
1403     if (child == NULL) {
1404         return;
1405     }
1406 
1407     if (child->bs->inherits_from == parent) {
1408         BdrvChild *c;
1409 
1410         /* Remove inherits_from only when the last reference between parent and
1411          * child->bs goes away. */
1412         QLIST_FOREACH(c, &parent->children, next) {
1413             if (c != child && c->bs == child->bs) {
1414                 break;
1415             }
1416         }
1417         if (c == NULL) {
1418             child->bs->inherits_from = NULL;
1419         }
1420     }
1421 
1422     bdrv_root_unref_child(child);
1423 }
1424 
1425 
1426 static void bdrv_parent_cb_change_media(BlockDriverState *bs, bool load)
1427 {
1428     BdrvChild *c;
1429     QLIST_FOREACH(c, &bs->parents, next_parent) {
1430         if (c->role->change_media) {
1431             c->role->change_media(c, load);
1432         }
1433     }
1434 }
1435 
1436 static void bdrv_parent_cb_resize(BlockDriverState *bs)
1437 {
1438     BdrvChild *c;
1439     QLIST_FOREACH(c, &bs->parents, next_parent) {
1440         if (c->role->resize) {
1441             c->role->resize(c);
1442         }
1443     }
1444 }
1445 
1446 /*
1447  * Sets the backing file link of a BDS. A new reference is created; callers
1448  * which don't need their own reference any more must call bdrv_unref().
1449  */
1450 void bdrv_set_backing_hd(BlockDriverState *bs, BlockDriverState *backing_hd)
1451 {
1452     if (backing_hd) {
1453         bdrv_ref(backing_hd);
1454     }
1455 
1456     if (bs->backing) {
1457         assert(bs->backing_blocker);
1458         bdrv_op_unblock_all(bs->backing->bs, bs->backing_blocker);
1459         bdrv_unref_child(bs, bs->backing);
1460     } else if (backing_hd) {
1461         error_setg(&bs->backing_blocker,
1462                    "node is used as backing hd of '%s'",
1463                    bdrv_get_device_or_node_name(bs));
1464     }
1465 
1466     if (!backing_hd) {
1467         error_free(bs->backing_blocker);
1468         bs->backing_blocker = NULL;
1469         bs->backing = NULL;
1470         goto out;
1471     }
1472     bs->backing = bdrv_attach_child(bs, backing_hd, "backing", &child_backing);
1473     bs->open_flags &= ~BDRV_O_NO_BACKING;
1474     pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_hd->filename);
1475     pstrcpy(bs->backing_format, sizeof(bs->backing_format),
1476             backing_hd->drv ? backing_hd->drv->format_name : "");
1477 
1478     bdrv_op_block_all(backing_hd, bs->backing_blocker);
1479     /* Otherwise we won't be able to commit or stream */
1480     bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_COMMIT_TARGET,
1481                     bs->backing_blocker);
1482     bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_STREAM,
1483                     bs->backing_blocker);
1484     /*
1485      * We do backup in 3 ways:
1486      * 1. drive backup
1487      *    The target bs is new opened, and the source is top BDS
1488      * 2. blockdev backup
1489      *    Both the source and the target are top BDSes.
1490      * 3. internal backup(used for block replication)
1491      *    Both the source and the target are backing file
1492      *
1493      * In case 1 and 2, neither the source nor the target is the backing file.
1494      * In case 3, we will block the top BDS, so there is only one block job
1495      * for the top BDS and its backing chain.
1496      */
1497     bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_BACKUP_SOURCE,
1498                     bs->backing_blocker);
1499     bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_BACKUP_TARGET,
1500                     bs->backing_blocker);
1501 out:
1502     bdrv_refresh_limits(bs, NULL);
1503 }
1504 
1505 /*
1506  * Opens the backing file for a BlockDriverState if not yet open
1507  *
1508  * bdref_key specifies the key for the image's BlockdevRef in the options QDict.
1509  * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
1510  * itself, all options starting with "${bdref_key}." are considered part of the
1511  * BlockdevRef.
1512  *
1513  * TODO Can this be unified with bdrv_open_image()?
1514  */
1515 int bdrv_open_backing_file(BlockDriverState *bs, QDict *parent_options,
1516                            const char *bdref_key, Error **errp)
1517 {
1518     char *backing_filename = g_malloc0(PATH_MAX);
1519     char *bdref_key_dot;
1520     const char *reference = NULL;
1521     int ret = 0;
1522     BlockDriverState *backing_hd;
1523     QDict *options;
1524     QDict *tmp_parent_options = NULL;
1525     Error *local_err = NULL;
1526 
1527     if (bs->backing != NULL) {
1528         goto free_exit;
1529     }
1530 
1531     /* NULL means an empty set of options */
1532     if (parent_options == NULL) {
1533         tmp_parent_options = qdict_new();
1534         parent_options = tmp_parent_options;
1535     }
1536 
1537     bs->open_flags &= ~BDRV_O_NO_BACKING;
1538 
1539     bdref_key_dot = g_strdup_printf("%s.", bdref_key);
1540     qdict_extract_subqdict(parent_options, &options, bdref_key_dot);
1541     g_free(bdref_key_dot);
1542 
1543     reference = qdict_get_try_str(parent_options, bdref_key);
1544     if (reference || qdict_haskey(options, "file.filename")) {
1545         backing_filename[0] = '\0';
1546     } else if (bs->backing_file[0] == '\0' && qdict_size(options) == 0) {
1547         QDECREF(options);
1548         goto free_exit;
1549     } else {
1550         bdrv_get_full_backing_filename(bs, backing_filename, PATH_MAX,
1551                                        &local_err);
1552         if (local_err) {
1553             ret = -EINVAL;
1554             error_propagate(errp, local_err);
1555             QDECREF(options);
1556             goto free_exit;
1557         }
1558     }
1559 
1560     if (!bs->drv || !bs->drv->supports_backing) {
1561         ret = -EINVAL;
1562         error_setg(errp, "Driver doesn't support backing files");
1563         QDECREF(options);
1564         goto free_exit;
1565     }
1566 
1567     if (bs->backing_format[0] != '\0' && !qdict_haskey(options, "driver")) {
1568         qdict_put(options, "driver", qstring_from_str(bs->backing_format));
1569     }
1570 
1571     backing_hd = bdrv_open_inherit(*backing_filename ? backing_filename : NULL,
1572                                    reference, options, 0, bs, &child_backing,
1573                                    errp);
1574     if (!backing_hd) {
1575         bs->open_flags |= BDRV_O_NO_BACKING;
1576         error_prepend(errp, "Could not open backing file: ");
1577         ret = -EINVAL;
1578         goto free_exit;
1579     }
1580 
1581     /* Hook up the backing file link; drop our reference, bs owns the
1582      * backing_hd reference now */
1583     bdrv_set_backing_hd(bs, backing_hd);
1584     bdrv_unref(backing_hd);
1585 
1586     qdict_del(parent_options, bdref_key);
1587 
1588 free_exit:
1589     g_free(backing_filename);
1590     QDECREF(tmp_parent_options);
1591     return ret;
1592 }
1593 
1594 static BlockDriverState *
1595 bdrv_open_child_bs(const char *filename, QDict *options, const char *bdref_key,
1596                    BlockDriverState *parent, const BdrvChildRole *child_role,
1597                    bool allow_none, Error **errp)
1598 {
1599     BlockDriverState *bs = NULL;
1600     QDict *image_options;
1601     char *bdref_key_dot;
1602     const char *reference;
1603 
1604     assert(child_role != NULL);
1605 
1606     bdref_key_dot = g_strdup_printf("%s.", bdref_key);
1607     qdict_extract_subqdict(options, &image_options, bdref_key_dot);
1608     g_free(bdref_key_dot);
1609 
1610     reference = qdict_get_try_str(options, bdref_key);
1611     if (!filename && !reference && !qdict_size(image_options)) {
1612         if (!allow_none) {
1613             error_setg(errp, "A block device must be specified for \"%s\"",
1614                        bdref_key);
1615         }
1616         QDECREF(image_options);
1617         goto done;
1618     }
1619 
1620     bs = bdrv_open_inherit(filename, reference, image_options, 0,
1621                            parent, child_role, errp);
1622     if (!bs) {
1623         goto done;
1624     }
1625 
1626 done:
1627     qdict_del(options, bdref_key);
1628     return bs;
1629 }
1630 
1631 /*
1632  * Opens a disk image whose options are given as BlockdevRef in another block
1633  * device's options.
1634  *
1635  * If allow_none is true, no image will be opened if filename is false and no
1636  * BlockdevRef is given. NULL will be returned, but errp remains unset.
1637  *
1638  * bdrev_key specifies the key for the image's BlockdevRef in the options QDict.
1639  * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
1640  * itself, all options starting with "${bdref_key}." are considered part of the
1641  * BlockdevRef.
1642  *
1643  * The BlockdevRef will be removed from the options QDict.
1644  */
1645 BdrvChild *bdrv_open_child(const char *filename,
1646                            QDict *options, const char *bdref_key,
1647                            BlockDriverState *parent,
1648                            const BdrvChildRole *child_role,
1649                            bool allow_none, Error **errp)
1650 {
1651     BlockDriverState *bs;
1652 
1653     bs = bdrv_open_child_bs(filename, options, bdref_key, parent, child_role,
1654                             allow_none, errp);
1655     if (bs == NULL) {
1656         return NULL;
1657     }
1658 
1659     return bdrv_attach_child(parent, bs, bdref_key, child_role);
1660 }
1661 
1662 static BlockDriverState *bdrv_append_temp_snapshot(BlockDriverState *bs,
1663                                                    int flags,
1664                                                    QDict *snapshot_options,
1665                                                    Error **errp)
1666 {
1667     /* TODO: extra byte is a hack to ensure MAX_PATH space on Windows. */
1668     char *tmp_filename = g_malloc0(PATH_MAX + 1);
1669     int64_t total_size;
1670     QemuOpts *opts = NULL;
1671     BlockDriverState *bs_snapshot;
1672     int ret;
1673 
1674     /* if snapshot, we create a temporary backing file and open it
1675        instead of opening 'filename' directly */
1676 
1677     /* Get the required size from the image */
1678     total_size = bdrv_getlength(bs);
1679     if (total_size < 0) {
1680         error_setg_errno(errp, -total_size, "Could not get image size");
1681         goto out;
1682     }
1683 
1684     /* Create the temporary image */
1685     ret = get_tmp_filename(tmp_filename, PATH_MAX + 1);
1686     if (ret < 0) {
1687         error_setg_errno(errp, -ret, "Could not get temporary filename");
1688         goto out;
1689     }
1690 
1691     opts = qemu_opts_create(bdrv_qcow2.create_opts, NULL, 0,
1692                             &error_abort);
1693     qemu_opt_set_number(opts, BLOCK_OPT_SIZE, total_size, &error_abort);
1694     ret = bdrv_create(&bdrv_qcow2, tmp_filename, opts, errp);
1695     qemu_opts_del(opts);
1696     if (ret < 0) {
1697         error_prepend(errp, "Could not create temporary overlay '%s': ",
1698                       tmp_filename);
1699         goto out;
1700     }
1701 
1702     /* Prepare options QDict for the temporary file */
1703     qdict_put(snapshot_options, "file.driver",
1704               qstring_from_str("file"));
1705     qdict_put(snapshot_options, "file.filename",
1706               qstring_from_str(tmp_filename));
1707     qdict_put(snapshot_options, "driver",
1708               qstring_from_str("qcow2"));
1709 
1710     bs_snapshot = bdrv_open(NULL, NULL, snapshot_options, flags, errp);
1711     snapshot_options = NULL;
1712     if (!bs_snapshot) {
1713         ret = -EINVAL;
1714         goto out;
1715     }
1716 
1717     /* bdrv_append() consumes a strong reference to bs_snapshot (i.e. it will
1718      * call bdrv_unref() on it), so in order to be able to return one, we have
1719      * to increase bs_snapshot's refcount here */
1720     bdrv_ref(bs_snapshot);
1721     bdrv_append(bs_snapshot, bs);
1722 
1723     g_free(tmp_filename);
1724     return bs_snapshot;
1725 
1726 out:
1727     QDECREF(snapshot_options);
1728     g_free(tmp_filename);
1729     return NULL;
1730 }
1731 
1732 /*
1733  * Opens a disk image (raw, qcow2, vmdk, ...)
1734  *
1735  * options is a QDict of options to pass to the block drivers, or NULL for an
1736  * empty set of options. The reference to the QDict belongs to the block layer
1737  * after the call (even on failure), so if the caller intends to reuse the
1738  * dictionary, it needs to use QINCREF() before calling bdrv_open.
1739  *
1740  * If *pbs is NULL, a new BDS will be created with a pointer to it stored there.
1741  * If it is not NULL, the referenced BDS will be reused.
1742  *
1743  * The reference parameter may be used to specify an existing block device which
1744  * should be opened. If specified, neither options nor a filename may be given,
1745  * nor can an existing BDS be reused (that is, *pbs has to be NULL).
1746  */
1747 static BlockDriverState *bdrv_open_inherit(const char *filename,
1748                                            const char *reference,
1749                                            QDict *options, int flags,
1750                                            BlockDriverState *parent,
1751                                            const BdrvChildRole *child_role,
1752                                            Error **errp)
1753 {
1754     int ret;
1755     BlockBackend *file = NULL;
1756     BlockDriverState *bs;
1757     BlockDriver *drv = NULL;
1758     const char *drvname;
1759     const char *backing;
1760     Error *local_err = NULL;
1761     QDict *snapshot_options = NULL;
1762     int snapshot_flags = 0;
1763 
1764     assert(!child_role || !flags);
1765     assert(!child_role == !parent);
1766 
1767     if (reference) {
1768         bool options_non_empty = options ? qdict_size(options) : false;
1769         QDECREF(options);
1770 
1771         if (filename || options_non_empty) {
1772             error_setg(errp, "Cannot reference an existing block device with "
1773                        "additional options or a new filename");
1774             return NULL;
1775         }
1776 
1777         bs = bdrv_lookup_bs(reference, reference, errp);
1778         if (!bs) {
1779             return NULL;
1780         }
1781 
1782         bdrv_ref(bs);
1783         return bs;
1784     }
1785 
1786     bs = bdrv_new();
1787 
1788     /* NULL means an empty set of options */
1789     if (options == NULL) {
1790         options = qdict_new();
1791     }
1792 
1793     /* json: syntax counts as explicit options, as if in the QDict */
1794     parse_json_protocol(options, &filename, &local_err);
1795     if (local_err) {
1796         goto fail;
1797     }
1798 
1799     bs->explicit_options = qdict_clone_shallow(options);
1800 
1801     if (child_role) {
1802         bs->inherits_from = parent;
1803         child_role->inherit_options(&flags, options,
1804                                     parent->open_flags, parent->options);
1805     }
1806 
1807     ret = bdrv_fill_options(&options, filename, &flags, &local_err);
1808     if (local_err) {
1809         goto fail;
1810     }
1811 
1812     /* Set the BDRV_O_RDWR and BDRV_O_ALLOW_RDWR flags.
1813      * FIXME: we're parsing the QDict to avoid having to create a
1814      * QemuOpts just for this, but neither option is optimal. */
1815     if (g_strcmp0(qdict_get_try_str(options, BDRV_OPT_READ_ONLY), "on") &&
1816         !qdict_get_try_bool(options, BDRV_OPT_READ_ONLY, false)) {
1817         flags |= (BDRV_O_RDWR | BDRV_O_ALLOW_RDWR);
1818     } else {
1819         flags &= ~BDRV_O_RDWR;
1820     }
1821 
1822     if (flags & BDRV_O_SNAPSHOT) {
1823         snapshot_options = qdict_new();
1824         bdrv_temp_snapshot_options(&snapshot_flags, snapshot_options,
1825                                    flags, options);
1826         /* Let bdrv_backing_options() override "read-only" */
1827         qdict_del(options, BDRV_OPT_READ_ONLY);
1828         bdrv_backing_options(&flags, options, flags, options);
1829     }
1830 
1831     bs->open_flags = flags;
1832     bs->options = options;
1833     options = qdict_clone_shallow(options);
1834 
1835     /* Find the right image format driver */
1836     drvname = qdict_get_try_str(options, "driver");
1837     if (drvname) {
1838         drv = bdrv_find_format(drvname);
1839         if (!drv) {
1840             error_setg(errp, "Unknown driver: '%s'", drvname);
1841             goto fail;
1842         }
1843     }
1844 
1845     assert(drvname || !(flags & BDRV_O_PROTOCOL));
1846 
1847     backing = qdict_get_try_str(options, "backing");
1848     if (backing && *backing == '\0') {
1849         flags |= BDRV_O_NO_BACKING;
1850         qdict_del(options, "backing");
1851     }
1852 
1853     /* Open image file without format layer. This BlockBackend is only used for
1854      * probing, the block drivers will do their own bdrv_open_child() for the
1855      * same BDS, which is why we put the node name back into options. */
1856     if ((flags & BDRV_O_PROTOCOL) == 0) {
1857         BlockDriverState *file_bs;
1858 
1859         file_bs = bdrv_open_child_bs(filename, options, "file", bs,
1860                                      &child_file, true, &local_err);
1861         if (local_err) {
1862             goto fail;
1863         }
1864         if (file_bs != NULL) {
1865             file = blk_new();
1866             blk_insert_bs(file, file_bs);
1867             bdrv_unref(file_bs);
1868 
1869             qdict_put(options, "file",
1870                       qstring_from_str(bdrv_get_node_name(file_bs)));
1871         }
1872     }
1873 
1874     /* Image format probing */
1875     bs->probed = !drv;
1876     if (!drv && file) {
1877         ret = find_image_format(file, filename, &drv, &local_err);
1878         if (ret < 0) {
1879             goto fail;
1880         }
1881         /*
1882          * This option update would logically belong in bdrv_fill_options(),
1883          * but we first need to open bs->file for the probing to work, while
1884          * opening bs->file already requires the (mostly) final set of options
1885          * so that cache mode etc. can be inherited.
1886          *
1887          * Adding the driver later is somewhat ugly, but it's not an option
1888          * that would ever be inherited, so it's correct. We just need to make
1889          * sure to update both bs->options (which has the full effective
1890          * options for bs) and options (which has file.* already removed).
1891          */
1892         qdict_put(bs->options, "driver", qstring_from_str(drv->format_name));
1893         qdict_put(options, "driver", qstring_from_str(drv->format_name));
1894     } else if (!drv) {
1895         error_setg(errp, "Must specify either driver or file");
1896         goto fail;
1897     }
1898 
1899     /* BDRV_O_PROTOCOL must be set iff a protocol BDS is about to be created */
1900     assert(!!(flags & BDRV_O_PROTOCOL) == !!drv->bdrv_file_open);
1901     /* file must be NULL if a protocol BDS is about to be created
1902      * (the inverse results in an error message from bdrv_open_common()) */
1903     assert(!(flags & BDRV_O_PROTOCOL) || !file);
1904 
1905     /* Open the image */
1906     ret = bdrv_open_common(bs, file, options, &local_err);
1907     if (ret < 0) {
1908         goto fail;
1909     }
1910 
1911     if (file) {
1912         blk_unref(file);
1913         file = NULL;
1914     }
1915 
1916     /* If there is a backing file, use it */
1917     if ((flags & BDRV_O_NO_BACKING) == 0) {
1918         ret = bdrv_open_backing_file(bs, options, "backing", &local_err);
1919         if (ret < 0) {
1920             goto close_and_fail;
1921         }
1922     }
1923 
1924     bdrv_refresh_filename(bs);
1925 
1926     /* Check if any unknown options were used */
1927     if (qdict_size(options) != 0) {
1928         const QDictEntry *entry = qdict_first(options);
1929         if (flags & BDRV_O_PROTOCOL) {
1930             error_setg(errp, "Block protocol '%s' doesn't support the option "
1931                        "'%s'", drv->format_name, entry->key);
1932         } else {
1933             error_setg(errp,
1934                        "Block format '%s' does not support the option '%s'",
1935                        drv->format_name, entry->key);
1936         }
1937 
1938         goto close_and_fail;
1939     }
1940 
1941     if (!bdrv_key_required(bs)) {
1942         bdrv_parent_cb_change_media(bs, true);
1943     } else if (!runstate_check(RUN_STATE_PRELAUNCH)
1944                && !runstate_check(RUN_STATE_INMIGRATE)
1945                && !runstate_check(RUN_STATE_PAUSED)) { /* HACK */
1946         error_setg(errp,
1947                    "Guest must be stopped for opening of encrypted image");
1948         goto close_and_fail;
1949     }
1950 
1951     QDECREF(options);
1952 
1953     /* For snapshot=on, create a temporary qcow2 overlay. bs points to the
1954      * temporary snapshot afterwards. */
1955     if (snapshot_flags) {
1956         BlockDriverState *snapshot_bs;
1957         snapshot_bs = bdrv_append_temp_snapshot(bs, snapshot_flags,
1958                                                 snapshot_options, &local_err);
1959         snapshot_options = NULL;
1960         if (local_err) {
1961             goto close_and_fail;
1962         }
1963         /* We are not going to return bs but the overlay on top of it
1964          * (snapshot_bs); thus, we have to drop the strong reference to bs
1965          * (which we obtained by calling bdrv_new()). bs will not be deleted,
1966          * though, because the overlay still has a reference to it. */
1967         bdrv_unref(bs);
1968         bs = snapshot_bs;
1969     }
1970 
1971     return bs;
1972 
1973 fail:
1974     blk_unref(file);
1975     if (bs->file != NULL) {
1976         bdrv_unref_child(bs, bs->file);
1977     }
1978     QDECREF(snapshot_options);
1979     QDECREF(bs->explicit_options);
1980     QDECREF(bs->options);
1981     QDECREF(options);
1982     bs->options = NULL;
1983     bdrv_unref(bs);
1984     error_propagate(errp, local_err);
1985     return NULL;
1986 
1987 close_and_fail:
1988     bdrv_unref(bs);
1989     QDECREF(snapshot_options);
1990     QDECREF(options);
1991     error_propagate(errp, local_err);
1992     return NULL;
1993 }
1994 
1995 BlockDriverState *bdrv_open(const char *filename, const char *reference,
1996                             QDict *options, int flags, Error **errp)
1997 {
1998     return bdrv_open_inherit(filename, reference, options, flags, NULL,
1999                              NULL, errp);
2000 }
2001 
2002 typedef struct BlockReopenQueueEntry {
2003      bool prepared;
2004      BDRVReopenState state;
2005      QSIMPLEQ_ENTRY(BlockReopenQueueEntry) entry;
2006 } BlockReopenQueueEntry;
2007 
2008 /*
2009  * Adds a BlockDriverState to a simple queue for an atomic, transactional
2010  * reopen of multiple devices.
2011  *
2012  * bs_queue can either be an existing BlockReopenQueue that has had QSIMPLE_INIT
2013  * already performed, or alternatively may be NULL a new BlockReopenQueue will
2014  * be created and initialized. This newly created BlockReopenQueue should be
2015  * passed back in for subsequent calls that are intended to be of the same
2016  * atomic 'set'.
2017  *
2018  * bs is the BlockDriverState to add to the reopen queue.
2019  *
2020  * options contains the changed options for the associated bs
2021  * (the BlockReopenQueue takes ownership)
2022  *
2023  * flags contains the open flags for the associated bs
2024  *
2025  * returns a pointer to bs_queue, which is either the newly allocated
2026  * bs_queue, or the existing bs_queue being used.
2027  *
2028  */
2029 static BlockReopenQueue *bdrv_reopen_queue_child(BlockReopenQueue *bs_queue,
2030                                                  BlockDriverState *bs,
2031                                                  QDict *options,
2032                                                  int flags,
2033                                                  const BdrvChildRole *role,
2034                                                  QDict *parent_options,
2035                                                  int parent_flags)
2036 {
2037     assert(bs != NULL);
2038 
2039     BlockReopenQueueEntry *bs_entry;
2040     BdrvChild *child;
2041     QDict *old_options, *explicit_options;
2042 
2043     if (bs_queue == NULL) {
2044         bs_queue = g_new0(BlockReopenQueue, 1);
2045         QSIMPLEQ_INIT(bs_queue);
2046     }
2047 
2048     if (!options) {
2049         options = qdict_new();
2050     }
2051 
2052     /* Check if this BlockDriverState is already in the queue */
2053     QSIMPLEQ_FOREACH(bs_entry, bs_queue, entry) {
2054         if (bs == bs_entry->state.bs) {
2055             break;
2056         }
2057     }
2058 
2059     /*
2060      * Precedence of options:
2061      * 1. Explicitly passed in options (highest)
2062      * 2. Set in flags (only for top level)
2063      * 3. Retained from explicitly set options of bs
2064      * 4. Inherited from parent node
2065      * 5. Retained from effective options of bs
2066      */
2067 
2068     if (!parent_options) {
2069         /*
2070          * Any setting represented by flags is always updated. If the
2071          * corresponding QDict option is set, it takes precedence. Otherwise
2072          * the flag is translated into a QDict option. The old setting of bs is
2073          * not considered.
2074          */
2075         update_options_from_flags(options, flags);
2076     }
2077 
2078     /* Old explicitly set values (don't overwrite by inherited value) */
2079     if (bs_entry) {
2080         old_options = qdict_clone_shallow(bs_entry->state.explicit_options);
2081     } else {
2082         old_options = qdict_clone_shallow(bs->explicit_options);
2083     }
2084     bdrv_join_options(bs, options, old_options);
2085     QDECREF(old_options);
2086 
2087     explicit_options = qdict_clone_shallow(options);
2088 
2089     /* Inherit from parent node */
2090     if (parent_options) {
2091         assert(!flags);
2092         role->inherit_options(&flags, options, parent_flags, parent_options);
2093     }
2094 
2095     /* Old values are used for options that aren't set yet */
2096     old_options = qdict_clone_shallow(bs->options);
2097     bdrv_join_options(bs, options, old_options);
2098     QDECREF(old_options);
2099 
2100     /* bdrv_open() masks this flag out */
2101     flags &= ~BDRV_O_PROTOCOL;
2102 
2103     QLIST_FOREACH(child, &bs->children, next) {
2104         QDict *new_child_options;
2105         char *child_key_dot;
2106 
2107         /* reopen can only change the options of block devices that were
2108          * implicitly created and inherited options. For other (referenced)
2109          * block devices, a syntax like "backing.foo" results in an error. */
2110         if (child->bs->inherits_from != bs) {
2111             continue;
2112         }
2113 
2114         child_key_dot = g_strdup_printf("%s.", child->name);
2115         qdict_extract_subqdict(options, &new_child_options, child_key_dot);
2116         g_free(child_key_dot);
2117 
2118         bdrv_reopen_queue_child(bs_queue, child->bs, new_child_options, 0,
2119                                 child->role, options, flags);
2120     }
2121 
2122     if (!bs_entry) {
2123         bs_entry = g_new0(BlockReopenQueueEntry, 1);
2124         QSIMPLEQ_INSERT_TAIL(bs_queue, bs_entry, entry);
2125     } else {
2126         QDECREF(bs_entry->state.options);
2127         QDECREF(bs_entry->state.explicit_options);
2128     }
2129 
2130     bs_entry->state.bs = bs;
2131     bs_entry->state.options = options;
2132     bs_entry->state.explicit_options = explicit_options;
2133     bs_entry->state.flags = flags;
2134 
2135     return bs_queue;
2136 }
2137 
2138 BlockReopenQueue *bdrv_reopen_queue(BlockReopenQueue *bs_queue,
2139                                     BlockDriverState *bs,
2140                                     QDict *options, int flags)
2141 {
2142     return bdrv_reopen_queue_child(bs_queue, bs, options, flags,
2143                                    NULL, NULL, 0);
2144 }
2145 
2146 /*
2147  * Reopen multiple BlockDriverStates atomically & transactionally.
2148  *
2149  * The queue passed in (bs_queue) must have been built up previous
2150  * via bdrv_reopen_queue().
2151  *
2152  * Reopens all BDS specified in the queue, with the appropriate
2153  * flags.  All devices are prepared for reopen, and failure of any
2154  * device will cause all device changes to be abandonded, and intermediate
2155  * data cleaned up.
2156  *
2157  * If all devices prepare successfully, then the changes are committed
2158  * to all devices.
2159  *
2160  */
2161 int bdrv_reopen_multiple(AioContext *ctx, BlockReopenQueue *bs_queue, Error **errp)
2162 {
2163     int ret = -1;
2164     BlockReopenQueueEntry *bs_entry, *next;
2165     Error *local_err = NULL;
2166 
2167     assert(bs_queue != NULL);
2168 
2169     aio_context_release(ctx);
2170     bdrv_drain_all_begin();
2171     aio_context_acquire(ctx);
2172 
2173     QSIMPLEQ_FOREACH(bs_entry, bs_queue, entry) {
2174         if (bdrv_reopen_prepare(&bs_entry->state, bs_queue, &local_err)) {
2175             error_propagate(errp, local_err);
2176             goto cleanup;
2177         }
2178         bs_entry->prepared = true;
2179     }
2180 
2181     /* If we reach this point, we have success and just need to apply the
2182      * changes
2183      */
2184     QSIMPLEQ_FOREACH(bs_entry, bs_queue, entry) {
2185         bdrv_reopen_commit(&bs_entry->state);
2186     }
2187 
2188     ret = 0;
2189 
2190 cleanup:
2191     QSIMPLEQ_FOREACH_SAFE(bs_entry, bs_queue, entry, next) {
2192         if (ret && bs_entry->prepared) {
2193             bdrv_reopen_abort(&bs_entry->state);
2194         } else if (ret) {
2195             QDECREF(bs_entry->state.explicit_options);
2196         }
2197         QDECREF(bs_entry->state.options);
2198         g_free(bs_entry);
2199     }
2200     g_free(bs_queue);
2201 
2202     bdrv_drain_all_end();
2203 
2204     return ret;
2205 }
2206 
2207 
2208 /* Reopen a single BlockDriverState with the specified flags. */
2209 int bdrv_reopen(BlockDriverState *bs, int bdrv_flags, Error **errp)
2210 {
2211     int ret = -1;
2212     Error *local_err = NULL;
2213     BlockReopenQueue *queue = bdrv_reopen_queue(NULL, bs, NULL, bdrv_flags);
2214 
2215     ret = bdrv_reopen_multiple(bdrv_get_aio_context(bs), queue, &local_err);
2216     if (local_err != NULL) {
2217         error_propagate(errp, local_err);
2218     }
2219     return ret;
2220 }
2221 
2222 
2223 /*
2224  * Prepares a BlockDriverState for reopen. All changes are staged in the
2225  * 'opaque' field of the BDRVReopenState, which is used and allocated by
2226  * the block driver layer .bdrv_reopen_prepare()
2227  *
2228  * bs is the BlockDriverState to reopen
2229  * flags are the new open flags
2230  * queue is the reopen queue
2231  *
2232  * Returns 0 on success, non-zero on error.  On error errp will be set
2233  * as well.
2234  *
2235  * On failure, bdrv_reopen_abort() will be called to clean up any data.
2236  * It is the responsibility of the caller to then call the abort() or
2237  * commit() for any other BDS that have been left in a prepare() state
2238  *
2239  */
2240 int bdrv_reopen_prepare(BDRVReopenState *reopen_state, BlockReopenQueue *queue,
2241                         Error **errp)
2242 {
2243     int ret = -1;
2244     Error *local_err = NULL;
2245     BlockDriver *drv;
2246     QemuOpts *opts;
2247     const char *value;
2248 
2249     assert(reopen_state != NULL);
2250     assert(reopen_state->bs->drv != NULL);
2251     drv = reopen_state->bs->drv;
2252 
2253     /* Process generic block layer options */
2254     opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
2255     qemu_opts_absorb_qdict(opts, reopen_state->options, &local_err);
2256     if (local_err) {
2257         error_propagate(errp, local_err);
2258         ret = -EINVAL;
2259         goto error;
2260     }
2261 
2262     update_flags_from_options(&reopen_state->flags, opts);
2263 
2264     /* node-name and driver must be unchanged. Put them back into the QDict, so
2265      * that they are checked at the end of this function. */
2266     value = qemu_opt_get(opts, "node-name");
2267     if (value) {
2268         qdict_put(reopen_state->options, "node-name", qstring_from_str(value));
2269     }
2270 
2271     value = qemu_opt_get(opts, "driver");
2272     if (value) {
2273         qdict_put(reopen_state->options, "driver", qstring_from_str(value));
2274     }
2275 
2276     /* if we are to stay read-only, do not allow permission change
2277      * to r/w */
2278     if (!(reopen_state->bs->open_flags & BDRV_O_ALLOW_RDWR) &&
2279         reopen_state->flags & BDRV_O_RDWR) {
2280         error_setg(errp, "Node '%s' is read only",
2281                    bdrv_get_device_or_node_name(reopen_state->bs));
2282         goto error;
2283     }
2284 
2285 
2286     ret = bdrv_flush(reopen_state->bs);
2287     if (ret) {
2288         error_setg_errno(errp, -ret, "Error flushing drive");
2289         goto error;
2290     }
2291 
2292     if (drv->bdrv_reopen_prepare) {
2293         ret = drv->bdrv_reopen_prepare(reopen_state, queue, &local_err);
2294         if (ret) {
2295             if (local_err != NULL) {
2296                 error_propagate(errp, local_err);
2297             } else {
2298                 error_setg(errp, "failed while preparing to reopen image '%s'",
2299                            reopen_state->bs->filename);
2300             }
2301             goto error;
2302         }
2303     } else {
2304         /* It is currently mandatory to have a bdrv_reopen_prepare()
2305          * handler for each supported drv. */
2306         error_setg(errp, "Block format '%s' used by node '%s' "
2307                    "does not support reopening files", drv->format_name,
2308                    bdrv_get_device_or_node_name(reopen_state->bs));
2309         ret = -1;
2310         goto error;
2311     }
2312 
2313     /* Options that are not handled are only okay if they are unchanged
2314      * compared to the old state. It is expected that some options are only
2315      * used for the initial open, but not reopen (e.g. filename) */
2316     if (qdict_size(reopen_state->options)) {
2317         const QDictEntry *entry = qdict_first(reopen_state->options);
2318 
2319         do {
2320             QString *new_obj = qobject_to_qstring(entry->value);
2321             const char *new = qstring_get_str(new_obj);
2322             const char *old = qdict_get_try_str(reopen_state->bs->options,
2323                                                 entry->key);
2324 
2325             if (!old || strcmp(new, old)) {
2326                 error_setg(errp, "Cannot change the option '%s'", entry->key);
2327                 ret = -EINVAL;
2328                 goto error;
2329             }
2330         } while ((entry = qdict_next(reopen_state->options, entry)));
2331     }
2332 
2333     ret = 0;
2334 
2335 error:
2336     qemu_opts_del(opts);
2337     return ret;
2338 }
2339 
2340 /*
2341  * Takes the staged changes for the reopen from bdrv_reopen_prepare(), and
2342  * makes them final by swapping the staging BlockDriverState contents into
2343  * the active BlockDriverState contents.
2344  */
2345 void bdrv_reopen_commit(BDRVReopenState *reopen_state)
2346 {
2347     BlockDriver *drv;
2348 
2349     assert(reopen_state != NULL);
2350     drv = reopen_state->bs->drv;
2351     assert(drv != NULL);
2352 
2353     /* If there are any driver level actions to take */
2354     if (drv->bdrv_reopen_commit) {
2355         drv->bdrv_reopen_commit(reopen_state);
2356     }
2357 
2358     /* set BDS specific flags now */
2359     QDECREF(reopen_state->bs->explicit_options);
2360 
2361     reopen_state->bs->explicit_options   = reopen_state->explicit_options;
2362     reopen_state->bs->open_flags         = reopen_state->flags;
2363     reopen_state->bs->read_only = !(reopen_state->flags & BDRV_O_RDWR);
2364 
2365     bdrv_refresh_limits(reopen_state->bs, NULL);
2366 }
2367 
2368 /*
2369  * Abort the reopen, and delete and free the staged changes in
2370  * reopen_state
2371  */
2372 void bdrv_reopen_abort(BDRVReopenState *reopen_state)
2373 {
2374     BlockDriver *drv;
2375 
2376     assert(reopen_state != NULL);
2377     drv = reopen_state->bs->drv;
2378     assert(drv != NULL);
2379 
2380     if (drv->bdrv_reopen_abort) {
2381         drv->bdrv_reopen_abort(reopen_state);
2382     }
2383 
2384     QDECREF(reopen_state->explicit_options);
2385 }
2386 
2387 
2388 static void bdrv_close(BlockDriverState *bs)
2389 {
2390     BdrvAioNotifier *ban, *ban_next;
2391 
2392     assert(!bs->job);
2393     assert(!bs->refcnt);
2394 
2395     bdrv_drained_begin(bs); /* complete I/O */
2396     bdrv_flush(bs);
2397     bdrv_drain(bs); /* in case flush left pending I/O */
2398 
2399     bdrv_release_named_dirty_bitmaps(bs);
2400     assert(QLIST_EMPTY(&bs->dirty_bitmaps));
2401 
2402     if (bs->drv) {
2403         BdrvChild *child, *next;
2404 
2405         bs->drv->bdrv_close(bs);
2406         bs->drv = NULL;
2407 
2408         bdrv_set_backing_hd(bs, NULL);
2409 
2410         if (bs->file != NULL) {
2411             bdrv_unref_child(bs, bs->file);
2412             bs->file = NULL;
2413         }
2414 
2415         QLIST_FOREACH_SAFE(child, &bs->children, next, next) {
2416             /* TODO Remove bdrv_unref() from drivers' close function and use
2417              * bdrv_unref_child() here */
2418             if (child->bs->inherits_from == bs) {
2419                 child->bs->inherits_from = NULL;
2420             }
2421             bdrv_detach_child(child);
2422         }
2423 
2424         g_free(bs->opaque);
2425         bs->opaque = NULL;
2426         bs->copy_on_read = 0;
2427         bs->backing_file[0] = '\0';
2428         bs->backing_format[0] = '\0';
2429         bs->total_sectors = 0;
2430         bs->encrypted = false;
2431         bs->valid_key = false;
2432         bs->sg = false;
2433         QDECREF(bs->options);
2434         QDECREF(bs->explicit_options);
2435         bs->options = NULL;
2436         QDECREF(bs->full_open_options);
2437         bs->full_open_options = NULL;
2438     }
2439 
2440     QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_next) {
2441         g_free(ban);
2442     }
2443     QLIST_INIT(&bs->aio_notifiers);
2444     bdrv_drained_end(bs);
2445 }
2446 
2447 void bdrv_close_all(void)
2448 {
2449     block_job_cancel_sync_all();
2450     nbd_export_close_all();
2451 
2452     /* Drop references from requests still in flight, such as canceled block
2453      * jobs whose AIO context has not been polled yet */
2454     bdrv_drain_all();
2455 
2456     blk_remove_all_bs();
2457     blockdev_close_all_bdrv_states();
2458 
2459     assert(QTAILQ_EMPTY(&all_bdrv_states));
2460 }
2461 
2462 static void change_parent_backing_link(BlockDriverState *from,
2463                                        BlockDriverState *to)
2464 {
2465     BdrvChild *c, *next, *to_c;
2466 
2467     QLIST_FOREACH_SAFE(c, &from->parents, next_parent, next) {
2468         if (c->role == &child_backing) {
2469             /* @from is generally not allowed to be a backing file, except for
2470              * when @to is the overlay. In that case, @from may not be replaced
2471              * by @to as @to's backing node. */
2472             QLIST_FOREACH(to_c, &to->children, next) {
2473                 if (to_c == c) {
2474                     break;
2475                 }
2476             }
2477             if (to_c) {
2478                 continue;
2479             }
2480         }
2481 
2482         assert(c->role != &child_backing);
2483         bdrv_ref(to);
2484         bdrv_replace_child(c, to);
2485         bdrv_unref(from);
2486     }
2487 }
2488 
2489 /*
2490  * Add new bs contents at the top of an image chain while the chain is
2491  * live, while keeping required fields on the top layer.
2492  *
2493  * This will modify the BlockDriverState fields, and swap contents
2494  * between bs_new and bs_top. Both bs_new and bs_top are modified.
2495  *
2496  * bs_new must not be attached to a BlockBackend.
2497  *
2498  * This function does not create any image files.
2499  *
2500  * bdrv_append() takes ownership of a bs_new reference and unrefs it because
2501  * that's what the callers commonly need. bs_new will be referenced by the old
2502  * parents of bs_top after bdrv_append() returns. If the caller needs to keep a
2503  * reference of its own, it must call bdrv_ref().
2504  */
2505 void bdrv_append(BlockDriverState *bs_new, BlockDriverState *bs_top)
2506 {
2507     assert(!bdrv_requests_pending(bs_top));
2508     assert(!bdrv_requests_pending(bs_new));
2509 
2510     bdrv_ref(bs_top);
2511 
2512     change_parent_backing_link(bs_top, bs_new);
2513     bdrv_set_backing_hd(bs_new, bs_top);
2514     bdrv_unref(bs_top);
2515 
2516     /* bs_new is now referenced by its new parents, we don't need the
2517      * additional reference any more. */
2518     bdrv_unref(bs_new);
2519 }
2520 
2521 void bdrv_replace_in_backing_chain(BlockDriverState *old, BlockDriverState *new)
2522 {
2523     assert(!bdrv_requests_pending(old));
2524     assert(!bdrv_requests_pending(new));
2525 
2526     bdrv_ref(old);
2527 
2528     change_parent_backing_link(old, new);
2529 
2530     bdrv_unref(old);
2531 }
2532 
2533 static void bdrv_delete(BlockDriverState *bs)
2534 {
2535     assert(!bs->job);
2536     assert(bdrv_op_blocker_is_empty(bs));
2537     assert(!bs->refcnt);
2538 
2539     bdrv_close(bs);
2540 
2541     /* remove from list, if necessary */
2542     if (bs->node_name[0] != '\0') {
2543         QTAILQ_REMOVE(&graph_bdrv_states, bs, node_list);
2544     }
2545     QTAILQ_REMOVE(&all_bdrv_states, bs, bs_list);
2546 
2547     g_free(bs);
2548 }
2549 
2550 /*
2551  * Run consistency checks on an image
2552  *
2553  * Returns 0 if the check could be completed (it doesn't mean that the image is
2554  * free of errors) or -errno when an internal error occurred. The results of the
2555  * check are stored in res.
2556  */
2557 int bdrv_check(BlockDriverState *bs, BdrvCheckResult *res, BdrvCheckMode fix)
2558 {
2559     if (bs->drv == NULL) {
2560         return -ENOMEDIUM;
2561     }
2562     if (bs->drv->bdrv_check == NULL) {
2563         return -ENOTSUP;
2564     }
2565 
2566     memset(res, 0, sizeof(*res));
2567     return bs->drv->bdrv_check(bs, res, fix);
2568 }
2569 
2570 /*
2571  * Return values:
2572  * 0        - success
2573  * -EINVAL  - backing format specified, but no file
2574  * -ENOSPC  - can't update the backing file because no space is left in the
2575  *            image file header
2576  * -ENOTSUP - format driver doesn't support changing the backing file
2577  */
2578 int bdrv_change_backing_file(BlockDriverState *bs,
2579     const char *backing_file, const char *backing_fmt)
2580 {
2581     BlockDriver *drv = bs->drv;
2582     int ret;
2583 
2584     /* Backing file format doesn't make sense without a backing file */
2585     if (backing_fmt && !backing_file) {
2586         return -EINVAL;
2587     }
2588 
2589     if (drv->bdrv_change_backing_file != NULL) {
2590         ret = drv->bdrv_change_backing_file(bs, backing_file, backing_fmt);
2591     } else {
2592         ret = -ENOTSUP;
2593     }
2594 
2595     if (ret == 0) {
2596         pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_file ?: "");
2597         pstrcpy(bs->backing_format, sizeof(bs->backing_format), backing_fmt ?: "");
2598     }
2599     return ret;
2600 }
2601 
2602 /*
2603  * Finds the image layer in the chain that has 'bs' as its backing file.
2604  *
2605  * active is the current topmost image.
2606  *
2607  * Returns NULL if bs is not found in active's image chain,
2608  * or if active == bs.
2609  *
2610  * Returns the bottommost base image if bs == NULL.
2611  */
2612 BlockDriverState *bdrv_find_overlay(BlockDriverState *active,
2613                                     BlockDriverState *bs)
2614 {
2615     while (active && bs != backing_bs(active)) {
2616         active = backing_bs(active);
2617     }
2618 
2619     return active;
2620 }
2621 
2622 /* Given a BDS, searches for the base layer. */
2623 BlockDriverState *bdrv_find_base(BlockDriverState *bs)
2624 {
2625     return bdrv_find_overlay(bs, NULL);
2626 }
2627 
2628 /*
2629  * Drops images above 'base' up to and including 'top', and sets the image
2630  * above 'top' to have base as its backing file.
2631  *
2632  * Requires that the overlay to 'top' is opened r/w, so that the backing file
2633  * information in 'bs' can be properly updated.
2634  *
2635  * E.g., this will convert the following chain:
2636  * bottom <- base <- intermediate <- top <- active
2637  *
2638  * to
2639  *
2640  * bottom <- base <- active
2641  *
2642  * It is allowed for bottom==base, in which case it converts:
2643  *
2644  * base <- intermediate <- top <- active
2645  *
2646  * to
2647  *
2648  * base <- active
2649  *
2650  * If backing_file_str is non-NULL, it will be used when modifying top's
2651  * overlay image metadata.
2652  *
2653  * Error conditions:
2654  *  if active == top, that is considered an error
2655  *
2656  */
2657 int bdrv_drop_intermediate(BlockDriverState *active, BlockDriverState *top,
2658                            BlockDriverState *base, const char *backing_file_str)
2659 {
2660     BlockDriverState *new_top_bs = NULL;
2661     int ret = -EIO;
2662 
2663     if (!top->drv || !base->drv) {
2664         goto exit;
2665     }
2666 
2667     new_top_bs = bdrv_find_overlay(active, top);
2668 
2669     if (new_top_bs == NULL) {
2670         /* we could not find the image above 'top', this is an error */
2671         goto exit;
2672     }
2673 
2674     /* special case of new_top_bs->backing->bs already pointing to base - nothing
2675      * to do, no intermediate images */
2676     if (backing_bs(new_top_bs) == base) {
2677         ret = 0;
2678         goto exit;
2679     }
2680 
2681     /* Make sure that base is in the backing chain of top */
2682     if (!bdrv_chain_contains(top, base)) {
2683         goto exit;
2684     }
2685 
2686     /* success - we can delete the intermediate states, and link top->base */
2687     backing_file_str = backing_file_str ? backing_file_str : base->filename;
2688     ret = bdrv_change_backing_file(new_top_bs, backing_file_str,
2689                                    base->drv ? base->drv->format_name : "");
2690     if (ret) {
2691         goto exit;
2692     }
2693     bdrv_set_backing_hd(new_top_bs, base);
2694 
2695     ret = 0;
2696 exit:
2697     return ret;
2698 }
2699 
2700 /**
2701  * Truncate file to 'offset' bytes (needed only for file protocols)
2702  */
2703 int bdrv_truncate(BdrvChild *child, int64_t offset)
2704 {
2705     BlockDriverState *bs = child->bs;
2706     BlockDriver *drv = bs->drv;
2707     int ret;
2708     if (!drv)
2709         return -ENOMEDIUM;
2710     if (!drv->bdrv_truncate)
2711         return -ENOTSUP;
2712     if (bs->read_only)
2713         return -EACCES;
2714 
2715     ret = drv->bdrv_truncate(bs, offset);
2716     if (ret == 0) {
2717         ret = refresh_total_sectors(bs, offset >> BDRV_SECTOR_BITS);
2718         bdrv_dirty_bitmap_truncate(bs);
2719         bdrv_parent_cb_resize(bs);
2720         ++bs->write_gen;
2721     }
2722     return ret;
2723 }
2724 
2725 /**
2726  * Length of a allocated file in bytes. Sparse files are counted by actual
2727  * allocated space. Return < 0 if error or unknown.
2728  */
2729 int64_t bdrv_get_allocated_file_size(BlockDriverState *bs)
2730 {
2731     BlockDriver *drv = bs->drv;
2732     if (!drv) {
2733         return -ENOMEDIUM;
2734     }
2735     if (drv->bdrv_get_allocated_file_size) {
2736         return drv->bdrv_get_allocated_file_size(bs);
2737     }
2738     if (bs->file) {
2739         return bdrv_get_allocated_file_size(bs->file->bs);
2740     }
2741     return -ENOTSUP;
2742 }
2743 
2744 /**
2745  * Return number of sectors on success, -errno on error.
2746  */
2747 int64_t bdrv_nb_sectors(BlockDriverState *bs)
2748 {
2749     BlockDriver *drv = bs->drv;
2750 
2751     if (!drv)
2752         return -ENOMEDIUM;
2753 
2754     if (drv->has_variable_length) {
2755         int ret = refresh_total_sectors(bs, bs->total_sectors);
2756         if (ret < 0) {
2757             return ret;
2758         }
2759     }
2760     return bs->total_sectors;
2761 }
2762 
2763 /**
2764  * Return length in bytes on success, -errno on error.
2765  * The length is always a multiple of BDRV_SECTOR_SIZE.
2766  */
2767 int64_t bdrv_getlength(BlockDriverState *bs)
2768 {
2769     int64_t ret = bdrv_nb_sectors(bs);
2770 
2771     ret = ret > INT64_MAX / BDRV_SECTOR_SIZE ? -EFBIG : ret;
2772     return ret < 0 ? ret : ret * BDRV_SECTOR_SIZE;
2773 }
2774 
2775 /* return 0 as number of sectors if no device present or error */
2776 void bdrv_get_geometry(BlockDriverState *bs, uint64_t *nb_sectors_ptr)
2777 {
2778     int64_t nb_sectors = bdrv_nb_sectors(bs);
2779 
2780     *nb_sectors_ptr = nb_sectors < 0 ? 0 : nb_sectors;
2781 }
2782 
2783 bool bdrv_is_read_only(BlockDriverState *bs)
2784 {
2785     return bs->read_only;
2786 }
2787 
2788 bool bdrv_is_sg(BlockDriverState *bs)
2789 {
2790     return bs->sg;
2791 }
2792 
2793 bool bdrv_is_encrypted(BlockDriverState *bs)
2794 {
2795     if (bs->backing && bs->backing->bs->encrypted) {
2796         return true;
2797     }
2798     return bs->encrypted;
2799 }
2800 
2801 bool bdrv_key_required(BlockDriverState *bs)
2802 {
2803     BdrvChild *backing = bs->backing;
2804 
2805     if (backing && backing->bs->encrypted && !backing->bs->valid_key) {
2806         return true;
2807     }
2808     return (bs->encrypted && !bs->valid_key);
2809 }
2810 
2811 int bdrv_set_key(BlockDriverState *bs, const char *key)
2812 {
2813     int ret;
2814     if (bs->backing && bs->backing->bs->encrypted) {
2815         ret = bdrv_set_key(bs->backing->bs, key);
2816         if (ret < 0)
2817             return ret;
2818         if (!bs->encrypted)
2819             return 0;
2820     }
2821     if (!bs->encrypted) {
2822         return -EINVAL;
2823     } else if (!bs->drv || !bs->drv->bdrv_set_key) {
2824         return -ENOMEDIUM;
2825     }
2826     ret = bs->drv->bdrv_set_key(bs, key);
2827     if (ret < 0) {
2828         bs->valid_key = false;
2829     } else if (!bs->valid_key) {
2830         /* call the change callback now, we skipped it on open */
2831         bs->valid_key = true;
2832         bdrv_parent_cb_change_media(bs, true);
2833     }
2834     return ret;
2835 }
2836 
2837 /*
2838  * Provide an encryption key for @bs.
2839  * If @key is non-null:
2840  *     If @bs is not encrypted, fail.
2841  *     Else if the key is invalid, fail.
2842  *     Else set @bs's key to @key, replacing the existing key, if any.
2843  * If @key is null:
2844  *     If @bs is encrypted and still lacks a key, fail.
2845  *     Else do nothing.
2846  * On failure, store an error object through @errp if non-null.
2847  */
2848 void bdrv_add_key(BlockDriverState *bs, const char *key, Error **errp)
2849 {
2850     if (key) {
2851         if (!bdrv_is_encrypted(bs)) {
2852             error_setg(errp, "Node '%s' is not encrypted",
2853                       bdrv_get_device_or_node_name(bs));
2854         } else if (bdrv_set_key(bs, key) < 0) {
2855             error_setg(errp, QERR_INVALID_PASSWORD);
2856         }
2857     } else {
2858         if (bdrv_key_required(bs)) {
2859             error_set(errp, ERROR_CLASS_DEVICE_ENCRYPTED,
2860                       "'%s' (%s) is encrypted",
2861                       bdrv_get_device_or_node_name(bs),
2862                       bdrv_get_encrypted_filename(bs));
2863         }
2864     }
2865 }
2866 
2867 const char *bdrv_get_format_name(BlockDriverState *bs)
2868 {
2869     return bs->drv ? bs->drv->format_name : NULL;
2870 }
2871 
2872 static int qsort_strcmp(const void *a, const void *b)
2873 {
2874     return strcmp(*(char *const *)a, *(char *const *)b);
2875 }
2876 
2877 void bdrv_iterate_format(void (*it)(void *opaque, const char *name),
2878                          void *opaque)
2879 {
2880     BlockDriver *drv;
2881     int count = 0;
2882     int i;
2883     const char **formats = NULL;
2884 
2885     QLIST_FOREACH(drv, &bdrv_drivers, list) {
2886         if (drv->format_name) {
2887             bool found = false;
2888             int i = count;
2889             while (formats && i && !found) {
2890                 found = !strcmp(formats[--i], drv->format_name);
2891             }
2892 
2893             if (!found) {
2894                 formats = g_renew(const char *, formats, count + 1);
2895                 formats[count++] = drv->format_name;
2896             }
2897         }
2898     }
2899 
2900     for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); i++) {
2901         const char *format_name = block_driver_modules[i].format_name;
2902 
2903         if (format_name) {
2904             bool found = false;
2905             int j = count;
2906 
2907             while (formats && j && !found) {
2908                 found = !strcmp(formats[--j], format_name);
2909             }
2910 
2911             if (!found) {
2912                 formats = g_renew(const char *, formats, count + 1);
2913                 formats[count++] = format_name;
2914             }
2915         }
2916     }
2917 
2918     qsort(formats, count, sizeof(formats[0]), qsort_strcmp);
2919 
2920     for (i = 0; i < count; i++) {
2921         it(opaque, formats[i]);
2922     }
2923 
2924     g_free(formats);
2925 }
2926 
2927 /* This function is to find a node in the bs graph */
2928 BlockDriverState *bdrv_find_node(const char *node_name)
2929 {
2930     BlockDriverState *bs;
2931 
2932     assert(node_name);
2933 
2934     QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
2935         if (!strcmp(node_name, bs->node_name)) {
2936             return bs;
2937         }
2938     }
2939     return NULL;
2940 }
2941 
2942 /* Put this QMP function here so it can access the static graph_bdrv_states. */
2943 BlockDeviceInfoList *bdrv_named_nodes_list(Error **errp)
2944 {
2945     BlockDeviceInfoList *list, *entry;
2946     BlockDriverState *bs;
2947 
2948     list = NULL;
2949     QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
2950         BlockDeviceInfo *info = bdrv_block_device_info(NULL, bs, errp);
2951         if (!info) {
2952             qapi_free_BlockDeviceInfoList(list);
2953             return NULL;
2954         }
2955         entry = g_malloc0(sizeof(*entry));
2956         entry->value = info;
2957         entry->next = list;
2958         list = entry;
2959     }
2960 
2961     return list;
2962 }
2963 
2964 BlockDriverState *bdrv_lookup_bs(const char *device,
2965                                  const char *node_name,
2966                                  Error **errp)
2967 {
2968     BlockBackend *blk;
2969     BlockDriverState *bs;
2970 
2971     if (device) {
2972         blk = blk_by_name(device);
2973 
2974         if (blk) {
2975             bs = blk_bs(blk);
2976             if (!bs) {
2977                 error_setg(errp, "Device '%s' has no medium", device);
2978             }
2979 
2980             return bs;
2981         }
2982     }
2983 
2984     if (node_name) {
2985         bs = bdrv_find_node(node_name);
2986 
2987         if (bs) {
2988             return bs;
2989         }
2990     }
2991 
2992     error_setg(errp, "Cannot find device=%s nor node_name=%s",
2993                      device ? device : "",
2994                      node_name ? node_name : "");
2995     return NULL;
2996 }
2997 
2998 /* If 'base' is in the same chain as 'top', return true. Otherwise,
2999  * return false.  If either argument is NULL, return false. */
3000 bool bdrv_chain_contains(BlockDriverState *top, BlockDriverState *base)
3001 {
3002     while (top && top != base) {
3003         top = backing_bs(top);
3004     }
3005 
3006     return top != NULL;
3007 }
3008 
3009 BlockDriverState *bdrv_next_node(BlockDriverState *bs)
3010 {
3011     if (!bs) {
3012         return QTAILQ_FIRST(&graph_bdrv_states);
3013     }
3014     return QTAILQ_NEXT(bs, node_list);
3015 }
3016 
3017 const char *bdrv_get_node_name(const BlockDriverState *bs)
3018 {
3019     return bs->node_name;
3020 }
3021 
3022 const char *bdrv_get_parent_name(const BlockDriverState *bs)
3023 {
3024     BdrvChild *c;
3025     const char *name;
3026 
3027     /* If multiple parents have a name, just pick the first one. */
3028     QLIST_FOREACH(c, &bs->parents, next_parent) {
3029         if (c->role->get_name) {
3030             name = c->role->get_name(c);
3031             if (name && *name) {
3032                 return name;
3033             }
3034         }
3035     }
3036 
3037     return NULL;
3038 }
3039 
3040 /* TODO check what callers really want: bs->node_name or blk_name() */
3041 const char *bdrv_get_device_name(const BlockDriverState *bs)
3042 {
3043     return bdrv_get_parent_name(bs) ?: "";
3044 }
3045 
3046 /* This can be used to identify nodes that might not have a device
3047  * name associated. Since node and device names live in the same
3048  * namespace, the result is unambiguous. The exception is if both are
3049  * absent, then this returns an empty (non-null) string. */
3050 const char *bdrv_get_device_or_node_name(const BlockDriverState *bs)
3051 {
3052     return bdrv_get_parent_name(bs) ?: bs->node_name;
3053 }
3054 
3055 int bdrv_get_flags(BlockDriverState *bs)
3056 {
3057     return bs->open_flags;
3058 }
3059 
3060 int bdrv_has_zero_init_1(BlockDriverState *bs)
3061 {
3062     return 1;
3063 }
3064 
3065 int bdrv_has_zero_init(BlockDriverState *bs)
3066 {
3067     assert(bs->drv);
3068 
3069     /* If BS is a copy on write image, it is initialized to
3070        the contents of the base image, which may not be zeroes.  */
3071     if (bs->backing) {
3072         return 0;
3073     }
3074     if (bs->drv->bdrv_has_zero_init) {
3075         return bs->drv->bdrv_has_zero_init(bs);
3076     }
3077 
3078     /* safe default */
3079     return 0;
3080 }
3081 
3082 bool bdrv_unallocated_blocks_are_zero(BlockDriverState *bs)
3083 {
3084     BlockDriverInfo bdi;
3085 
3086     if (bs->backing) {
3087         return false;
3088     }
3089 
3090     if (bdrv_get_info(bs, &bdi) == 0) {
3091         return bdi.unallocated_blocks_are_zero;
3092     }
3093 
3094     return false;
3095 }
3096 
3097 bool bdrv_can_write_zeroes_with_unmap(BlockDriverState *bs)
3098 {
3099     BlockDriverInfo bdi;
3100 
3101     if (!(bs->open_flags & BDRV_O_UNMAP)) {
3102         return false;
3103     }
3104 
3105     if (bdrv_get_info(bs, &bdi) == 0) {
3106         return bdi.can_write_zeroes_with_unmap;
3107     }
3108 
3109     return false;
3110 }
3111 
3112 const char *bdrv_get_encrypted_filename(BlockDriverState *bs)
3113 {
3114     if (bs->backing && bs->backing->bs->encrypted)
3115         return bs->backing_file;
3116     else if (bs->encrypted)
3117         return bs->filename;
3118     else
3119         return NULL;
3120 }
3121 
3122 void bdrv_get_backing_filename(BlockDriverState *bs,
3123                                char *filename, int filename_size)
3124 {
3125     pstrcpy(filename, filename_size, bs->backing_file);
3126 }
3127 
3128 int bdrv_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
3129 {
3130     BlockDriver *drv = bs->drv;
3131     if (!drv)
3132         return -ENOMEDIUM;
3133     if (!drv->bdrv_get_info)
3134         return -ENOTSUP;
3135     memset(bdi, 0, sizeof(*bdi));
3136     return drv->bdrv_get_info(bs, bdi);
3137 }
3138 
3139 ImageInfoSpecific *bdrv_get_specific_info(BlockDriverState *bs)
3140 {
3141     BlockDriver *drv = bs->drv;
3142     if (drv && drv->bdrv_get_specific_info) {
3143         return drv->bdrv_get_specific_info(bs);
3144     }
3145     return NULL;
3146 }
3147 
3148 void bdrv_debug_event(BlockDriverState *bs, BlkdebugEvent event)
3149 {
3150     if (!bs || !bs->drv || !bs->drv->bdrv_debug_event) {
3151         return;
3152     }
3153 
3154     bs->drv->bdrv_debug_event(bs, event);
3155 }
3156 
3157 int bdrv_debug_breakpoint(BlockDriverState *bs, const char *event,
3158                           const char *tag)
3159 {
3160     while (bs && bs->drv && !bs->drv->bdrv_debug_breakpoint) {
3161         bs = bs->file ? bs->file->bs : NULL;
3162     }
3163 
3164     if (bs && bs->drv && bs->drv->bdrv_debug_breakpoint) {
3165         return bs->drv->bdrv_debug_breakpoint(bs, event, tag);
3166     }
3167 
3168     return -ENOTSUP;
3169 }
3170 
3171 int bdrv_debug_remove_breakpoint(BlockDriverState *bs, const char *tag)
3172 {
3173     while (bs && bs->drv && !bs->drv->bdrv_debug_remove_breakpoint) {
3174         bs = bs->file ? bs->file->bs : NULL;
3175     }
3176 
3177     if (bs && bs->drv && bs->drv->bdrv_debug_remove_breakpoint) {
3178         return bs->drv->bdrv_debug_remove_breakpoint(bs, tag);
3179     }
3180 
3181     return -ENOTSUP;
3182 }
3183 
3184 int bdrv_debug_resume(BlockDriverState *bs, const char *tag)
3185 {
3186     while (bs && (!bs->drv || !bs->drv->bdrv_debug_resume)) {
3187         bs = bs->file ? bs->file->bs : NULL;
3188     }
3189 
3190     if (bs && bs->drv && bs->drv->bdrv_debug_resume) {
3191         return bs->drv->bdrv_debug_resume(bs, tag);
3192     }
3193 
3194     return -ENOTSUP;
3195 }
3196 
3197 bool bdrv_debug_is_suspended(BlockDriverState *bs, const char *tag)
3198 {
3199     while (bs && bs->drv && !bs->drv->bdrv_debug_is_suspended) {
3200         bs = bs->file ? bs->file->bs : NULL;
3201     }
3202 
3203     if (bs && bs->drv && bs->drv->bdrv_debug_is_suspended) {
3204         return bs->drv->bdrv_debug_is_suspended(bs, tag);
3205     }
3206 
3207     return false;
3208 }
3209 
3210 /* backing_file can either be relative, or absolute, or a protocol.  If it is
3211  * relative, it must be relative to the chain.  So, passing in bs->filename
3212  * from a BDS as backing_file should not be done, as that may be relative to
3213  * the CWD rather than the chain. */
3214 BlockDriverState *bdrv_find_backing_image(BlockDriverState *bs,
3215         const char *backing_file)
3216 {
3217     char *filename_full = NULL;
3218     char *backing_file_full = NULL;
3219     char *filename_tmp = NULL;
3220     int is_protocol = 0;
3221     BlockDriverState *curr_bs = NULL;
3222     BlockDriverState *retval = NULL;
3223     Error *local_error = NULL;
3224 
3225     if (!bs || !bs->drv || !backing_file) {
3226         return NULL;
3227     }
3228 
3229     filename_full     = g_malloc(PATH_MAX);
3230     backing_file_full = g_malloc(PATH_MAX);
3231     filename_tmp      = g_malloc(PATH_MAX);
3232 
3233     is_protocol = path_has_protocol(backing_file);
3234 
3235     for (curr_bs = bs; curr_bs->backing; curr_bs = curr_bs->backing->bs) {
3236 
3237         /* If either of the filename paths is actually a protocol, then
3238          * compare unmodified paths; otherwise make paths relative */
3239         if (is_protocol || path_has_protocol(curr_bs->backing_file)) {
3240             if (strcmp(backing_file, curr_bs->backing_file) == 0) {
3241                 retval = curr_bs->backing->bs;
3242                 break;
3243             }
3244             /* Also check against the full backing filename for the image */
3245             bdrv_get_full_backing_filename(curr_bs, backing_file_full, PATH_MAX,
3246                                            &local_error);
3247             if (local_error == NULL) {
3248                 if (strcmp(backing_file, backing_file_full) == 0) {
3249                     retval = curr_bs->backing->bs;
3250                     break;
3251                 }
3252             } else {
3253                 error_free(local_error);
3254                 local_error = NULL;
3255             }
3256         } else {
3257             /* If not an absolute filename path, make it relative to the current
3258              * image's filename path */
3259             path_combine(filename_tmp, PATH_MAX, curr_bs->filename,
3260                          backing_file);
3261 
3262             /* We are going to compare absolute pathnames */
3263             if (!realpath(filename_tmp, filename_full)) {
3264                 continue;
3265             }
3266 
3267             /* We need to make sure the backing filename we are comparing against
3268              * is relative to the current image filename (or absolute) */
3269             path_combine(filename_tmp, PATH_MAX, curr_bs->filename,
3270                          curr_bs->backing_file);
3271 
3272             if (!realpath(filename_tmp, backing_file_full)) {
3273                 continue;
3274             }
3275 
3276             if (strcmp(backing_file_full, filename_full) == 0) {
3277                 retval = curr_bs->backing->bs;
3278                 break;
3279             }
3280         }
3281     }
3282 
3283     g_free(filename_full);
3284     g_free(backing_file_full);
3285     g_free(filename_tmp);
3286     return retval;
3287 }
3288 
3289 int bdrv_get_backing_file_depth(BlockDriverState *bs)
3290 {
3291     if (!bs->drv) {
3292         return 0;
3293     }
3294 
3295     if (!bs->backing) {
3296         return 0;
3297     }
3298 
3299     return 1 + bdrv_get_backing_file_depth(bs->backing->bs);
3300 }
3301 
3302 void bdrv_init(void)
3303 {
3304     module_call_init(MODULE_INIT_BLOCK);
3305 }
3306 
3307 void bdrv_init_with_whitelist(void)
3308 {
3309     use_bdrv_whitelist = 1;
3310     bdrv_init();
3311 }
3312 
3313 void bdrv_invalidate_cache(BlockDriverState *bs, Error **errp)
3314 {
3315     BdrvChild *child;
3316     Error *local_err = NULL;
3317     int ret;
3318 
3319     if (!bs->drv)  {
3320         return;
3321     }
3322 
3323     if (!(bs->open_flags & BDRV_O_INACTIVE)) {
3324         return;
3325     }
3326 
3327     QLIST_FOREACH(child, &bs->children, next) {
3328         bdrv_invalidate_cache(child->bs, &local_err);
3329         if (local_err) {
3330             error_propagate(errp, local_err);
3331             return;
3332         }
3333     }
3334 
3335     bs->open_flags &= ~BDRV_O_INACTIVE;
3336     if (bs->drv->bdrv_invalidate_cache) {
3337         bs->drv->bdrv_invalidate_cache(bs, &local_err);
3338         if (local_err) {
3339             bs->open_flags |= BDRV_O_INACTIVE;
3340             error_propagate(errp, local_err);
3341             return;
3342         }
3343     }
3344 
3345     ret = refresh_total_sectors(bs, bs->total_sectors);
3346     if (ret < 0) {
3347         bs->open_flags |= BDRV_O_INACTIVE;
3348         error_setg_errno(errp, -ret, "Could not refresh total sector count");
3349         return;
3350     }
3351 }
3352 
3353 void bdrv_invalidate_cache_all(Error **errp)
3354 {
3355     BlockDriverState *bs;
3356     Error *local_err = NULL;
3357     BdrvNextIterator it;
3358 
3359     for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
3360         AioContext *aio_context = bdrv_get_aio_context(bs);
3361 
3362         aio_context_acquire(aio_context);
3363         bdrv_invalidate_cache(bs, &local_err);
3364         aio_context_release(aio_context);
3365         if (local_err) {
3366             error_propagate(errp, local_err);
3367             return;
3368         }
3369     }
3370 }
3371 
3372 static int bdrv_inactivate_recurse(BlockDriverState *bs,
3373                                    bool setting_flag)
3374 {
3375     BdrvChild *child;
3376     int ret;
3377 
3378     if (!setting_flag && bs->drv->bdrv_inactivate) {
3379         ret = bs->drv->bdrv_inactivate(bs);
3380         if (ret < 0) {
3381             return ret;
3382         }
3383     }
3384 
3385     QLIST_FOREACH(child, &bs->children, next) {
3386         ret = bdrv_inactivate_recurse(child->bs, setting_flag);
3387         if (ret < 0) {
3388             return ret;
3389         }
3390     }
3391 
3392     if (setting_flag) {
3393         bs->open_flags |= BDRV_O_INACTIVE;
3394     }
3395     return 0;
3396 }
3397 
3398 int bdrv_inactivate_all(void)
3399 {
3400     BlockDriverState *bs = NULL;
3401     BdrvNextIterator it;
3402     int ret = 0;
3403     int pass;
3404 
3405     for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
3406         aio_context_acquire(bdrv_get_aio_context(bs));
3407     }
3408 
3409     /* We do two passes of inactivation. The first pass calls to drivers'
3410      * .bdrv_inactivate callbacks recursively so all cache is flushed to disk;
3411      * the second pass sets the BDRV_O_INACTIVE flag so that no further write
3412      * is allowed. */
3413     for (pass = 0; pass < 2; pass++) {
3414         for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
3415             ret = bdrv_inactivate_recurse(bs, pass);
3416             if (ret < 0) {
3417                 goto out;
3418             }
3419         }
3420     }
3421 
3422 out:
3423     for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
3424         aio_context_release(bdrv_get_aio_context(bs));
3425     }
3426 
3427     return ret;
3428 }
3429 
3430 /**************************************************************/
3431 /* removable device support */
3432 
3433 /**
3434  * Return TRUE if the media is present
3435  */
3436 bool bdrv_is_inserted(BlockDriverState *bs)
3437 {
3438     BlockDriver *drv = bs->drv;
3439     BdrvChild *child;
3440 
3441     if (!drv) {
3442         return false;
3443     }
3444     if (drv->bdrv_is_inserted) {
3445         return drv->bdrv_is_inserted(bs);
3446     }
3447     QLIST_FOREACH(child, &bs->children, next) {
3448         if (!bdrv_is_inserted(child->bs)) {
3449             return false;
3450         }
3451     }
3452     return true;
3453 }
3454 
3455 /**
3456  * Return whether the media changed since the last call to this
3457  * function, or -ENOTSUP if we don't know.  Most drivers don't know.
3458  */
3459 int bdrv_media_changed(BlockDriverState *bs)
3460 {
3461     BlockDriver *drv = bs->drv;
3462 
3463     if (drv && drv->bdrv_media_changed) {
3464         return drv->bdrv_media_changed(bs);
3465     }
3466     return -ENOTSUP;
3467 }
3468 
3469 /**
3470  * If eject_flag is TRUE, eject the media. Otherwise, close the tray
3471  */
3472 void bdrv_eject(BlockDriverState *bs, bool eject_flag)
3473 {
3474     BlockDriver *drv = bs->drv;
3475 
3476     if (drv && drv->bdrv_eject) {
3477         drv->bdrv_eject(bs, eject_flag);
3478     }
3479 }
3480 
3481 /**
3482  * Lock or unlock the media (if it is locked, the user won't be able
3483  * to eject it manually).
3484  */
3485 void bdrv_lock_medium(BlockDriverState *bs, bool locked)
3486 {
3487     BlockDriver *drv = bs->drv;
3488 
3489     trace_bdrv_lock_medium(bs, locked);
3490 
3491     if (drv && drv->bdrv_lock_medium) {
3492         drv->bdrv_lock_medium(bs, locked);
3493     }
3494 }
3495 
3496 /* Get a reference to bs */
3497 void bdrv_ref(BlockDriverState *bs)
3498 {
3499     bs->refcnt++;
3500 }
3501 
3502 /* Release a previously grabbed reference to bs.
3503  * If after releasing, reference count is zero, the BlockDriverState is
3504  * deleted. */
3505 void bdrv_unref(BlockDriverState *bs)
3506 {
3507     if (!bs) {
3508         return;
3509     }
3510     assert(bs->refcnt > 0);
3511     if (--bs->refcnt == 0) {
3512         bdrv_delete(bs);
3513     }
3514 }
3515 
3516 struct BdrvOpBlocker {
3517     Error *reason;
3518     QLIST_ENTRY(BdrvOpBlocker) list;
3519 };
3520 
3521 bool bdrv_op_is_blocked(BlockDriverState *bs, BlockOpType op, Error **errp)
3522 {
3523     BdrvOpBlocker *blocker;
3524     assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
3525     if (!QLIST_EMPTY(&bs->op_blockers[op])) {
3526         blocker = QLIST_FIRST(&bs->op_blockers[op]);
3527         if (errp) {
3528             *errp = error_copy(blocker->reason);
3529             error_prepend(errp, "Node '%s' is busy: ",
3530                           bdrv_get_device_or_node_name(bs));
3531         }
3532         return true;
3533     }
3534     return false;
3535 }
3536 
3537 void bdrv_op_block(BlockDriverState *bs, BlockOpType op, Error *reason)
3538 {
3539     BdrvOpBlocker *blocker;
3540     assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
3541 
3542     blocker = g_new0(BdrvOpBlocker, 1);
3543     blocker->reason = reason;
3544     QLIST_INSERT_HEAD(&bs->op_blockers[op], blocker, list);
3545 }
3546 
3547 void bdrv_op_unblock(BlockDriverState *bs, BlockOpType op, Error *reason)
3548 {
3549     BdrvOpBlocker *blocker, *next;
3550     assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
3551     QLIST_FOREACH_SAFE(blocker, &bs->op_blockers[op], list, next) {
3552         if (blocker->reason == reason) {
3553             QLIST_REMOVE(blocker, list);
3554             g_free(blocker);
3555         }
3556     }
3557 }
3558 
3559 void bdrv_op_block_all(BlockDriverState *bs, Error *reason)
3560 {
3561     int i;
3562     for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
3563         bdrv_op_block(bs, i, reason);
3564     }
3565 }
3566 
3567 void bdrv_op_unblock_all(BlockDriverState *bs, Error *reason)
3568 {
3569     int i;
3570     for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
3571         bdrv_op_unblock(bs, i, reason);
3572     }
3573 }
3574 
3575 bool bdrv_op_blocker_is_empty(BlockDriverState *bs)
3576 {
3577     int i;
3578 
3579     for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
3580         if (!QLIST_EMPTY(&bs->op_blockers[i])) {
3581             return false;
3582         }
3583     }
3584     return true;
3585 }
3586 
3587 void bdrv_img_create(const char *filename, const char *fmt,
3588                      const char *base_filename, const char *base_fmt,
3589                      char *options, uint64_t img_size, int flags,
3590                      Error **errp, bool quiet)
3591 {
3592     QemuOptsList *create_opts = NULL;
3593     QemuOpts *opts = NULL;
3594     const char *backing_fmt, *backing_file;
3595     int64_t size;
3596     BlockDriver *drv, *proto_drv;
3597     Error *local_err = NULL;
3598     int ret = 0;
3599 
3600     /* Find driver and parse its options */
3601     drv = bdrv_find_format(fmt);
3602     if (!drv) {
3603         error_setg(errp, "Unknown file format '%s'", fmt);
3604         return;
3605     }
3606 
3607     proto_drv = bdrv_find_protocol(filename, true, errp);
3608     if (!proto_drv) {
3609         return;
3610     }
3611 
3612     if (!drv->create_opts) {
3613         error_setg(errp, "Format driver '%s' does not support image creation",
3614                    drv->format_name);
3615         return;
3616     }
3617 
3618     if (!proto_drv->create_opts) {
3619         error_setg(errp, "Protocol driver '%s' does not support image creation",
3620                    proto_drv->format_name);
3621         return;
3622     }
3623 
3624     create_opts = qemu_opts_append(create_opts, drv->create_opts);
3625     create_opts = qemu_opts_append(create_opts, proto_drv->create_opts);
3626 
3627     /* Create parameter list with default values */
3628     opts = qemu_opts_create(create_opts, NULL, 0, &error_abort);
3629     qemu_opt_set_number(opts, BLOCK_OPT_SIZE, img_size, &error_abort);
3630 
3631     /* Parse -o options */
3632     if (options) {
3633         qemu_opts_do_parse(opts, options, NULL, &local_err);
3634         if (local_err) {
3635             error_report_err(local_err);
3636             local_err = NULL;
3637             error_setg(errp, "Invalid options for file format '%s'", fmt);
3638             goto out;
3639         }
3640     }
3641 
3642     if (base_filename) {
3643         qemu_opt_set(opts, BLOCK_OPT_BACKING_FILE, base_filename, &local_err);
3644         if (local_err) {
3645             error_setg(errp, "Backing file not supported for file format '%s'",
3646                        fmt);
3647             goto out;
3648         }
3649     }
3650 
3651     if (base_fmt) {
3652         qemu_opt_set(opts, BLOCK_OPT_BACKING_FMT, base_fmt, &local_err);
3653         if (local_err) {
3654             error_setg(errp, "Backing file format not supported for file "
3655                              "format '%s'", fmt);
3656             goto out;
3657         }
3658     }
3659 
3660     backing_file = qemu_opt_get(opts, BLOCK_OPT_BACKING_FILE);
3661     if (backing_file) {
3662         if (!strcmp(filename, backing_file)) {
3663             error_setg(errp, "Error: Trying to create an image with the "
3664                              "same filename as the backing file");
3665             goto out;
3666         }
3667     }
3668 
3669     backing_fmt = qemu_opt_get(opts, BLOCK_OPT_BACKING_FMT);
3670 
3671     // The size for the image must always be specified, with one exception:
3672     // If we are using a backing file, we can obtain the size from there
3673     size = qemu_opt_get_size(opts, BLOCK_OPT_SIZE, 0);
3674     if (size == -1) {
3675         if (backing_file) {
3676             BlockDriverState *bs;
3677             char *full_backing = g_new0(char, PATH_MAX);
3678             int64_t size;
3679             int back_flags;
3680             QDict *backing_options = NULL;
3681 
3682             bdrv_get_full_backing_filename_from_filename(filename, backing_file,
3683                                                          full_backing, PATH_MAX,
3684                                                          &local_err);
3685             if (local_err) {
3686                 g_free(full_backing);
3687                 goto out;
3688             }
3689 
3690             /* backing files always opened read-only */
3691             back_flags = flags;
3692             back_flags &= ~(BDRV_O_RDWR | BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING);
3693 
3694             if (backing_fmt) {
3695                 backing_options = qdict_new();
3696                 qdict_put(backing_options, "driver",
3697                           qstring_from_str(backing_fmt));
3698             }
3699 
3700             bs = bdrv_open(full_backing, NULL, backing_options, back_flags,
3701                            &local_err);
3702             g_free(full_backing);
3703             if (!bs) {
3704                 goto out;
3705             }
3706             size = bdrv_getlength(bs);
3707             if (size < 0) {
3708                 error_setg_errno(errp, -size, "Could not get size of '%s'",
3709                                  backing_file);
3710                 bdrv_unref(bs);
3711                 goto out;
3712             }
3713 
3714             qemu_opt_set_number(opts, BLOCK_OPT_SIZE, size, &error_abort);
3715 
3716             bdrv_unref(bs);
3717         } else {
3718             error_setg(errp, "Image creation needs a size parameter");
3719             goto out;
3720         }
3721     }
3722 
3723     if (!quiet) {
3724         printf("Formatting '%s', fmt=%s ", filename, fmt);
3725         qemu_opts_print(opts, " ");
3726         puts("");
3727     }
3728 
3729     ret = bdrv_create(drv, filename, opts, &local_err);
3730 
3731     if (ret == -EFBIG) {
3732         /* This is generally a better message than whatever the driver would
3733          * deliver (especially because of the cluster_size_hint), since that
3734          * is most probably not much different from "image too large". */
3735         const char *cluster_size_hint = "";
3736         if (qemu_opt_get_size(opts, BLOCK_OPT_CLUSTER_SIZE, 0)) {
3737             cluster_size_hint = " (try using a larger cluster size)";
3738         }
3739         error_setg(errp, "The image size is too large for file format '%s'"
3740                    "%s", fmt, cluster_size_hint);
3741         error_free(local_err);
3742         local_err = NULL;
3743     }
3744 
3745 out:
3746     qemu_opts_del(opts);
3747     qemu_opts_free(create_opts);
3748     error_propagate(errp, local_err);
3749 }
3750 
3751 AioContext *bdrv_get_aio_context(BlockDriverState *bs)
3752 {
3753     return bs->aio_context;
3754 }
3755 
3756 static void bdrv_do_remove_aio_context_notifier(BdrvAioNotifier *ban)
3757 {
3758     QLIST_REMOVE(ban, list);
3759     g_free(ban);
3760 }
3761 
3762 void bdrv_detach_aio_context(BlockDriverState *bs)
3763 {
3764     BdrvAioNotifier *baf, *baf_tmp;
3765     BdrvChild *child;
3766 
3767     if (!bs->drv) {
3768         return;
3769     }
3770 
3771     assert(!bs->walking_aio_notifiers);
3772     bs->walking_aio_notifiers = true;
3773     QLIST_FOREACH_SAFE(baf, &bs->aio_notifiers, list, baf_tmp) {
3774         if (baf->deleted) {
3775             bdrv_do_remove_aio_context_notifier(baf);
3776         } else {
3777             baf->detach_aio_context(baf->opaque);
3778         }
3779     }
3780     /* Never mind iterating again to check for ->deleted.  bdrv_close() will
3781      * remove remaining aio notifiers if we aren't called again.
3782      */
3783     bs->walking_aio_notifiers = false;
3784 
3785     if (bs->drv->bdrv_detach_aio_context) {
3786         bs->drv->bdrv_detach_aio_context(bs);
3787     }
3788     QLIST_FOREACH(child, &bs->children, next) {
3789         bdrv_detach_aio_context(child->bs);
3790     }
3791 
3792     bs->aio_context = NULL;
3793 }
3794 
3795 void bdrv_attach_aio_context(BlockDriverState *bs,
3796                              AioContext *new_context)
3797 {
3798     BdrvAioNotifier *ban, *ban_tmp;
3799     BdrvChild *child;
3800 
3801     if (!bs->drv) {
3802         return;
3803     }
3804 
3805     bs->aio_context = new_context;
3806 
3807     QLIST_FOREACH(child, &bs->children, next) {
3808         bdrv_attach_aio_context(child->bs, new_context);
3809     }
3810     if (bs->drv->bdrv_attach_aio_context) {
3811         bs->drv->bdrv_attach_aio_context(bs, new_context);
3812     }
3813 
3814     assert(!bs->walking_aio_notifiers);
3815     bs->walking_aio_notifiers = true;
3816     QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_tmp) {
3817         if (ban->deleted) {
3818             bdrv_do_remove_aio_context_notifier(ban);
3819         } else {
3820             ban->attached_aio_context(new_context, ban->opaque);
3821         }
3822     }
3823     bs->walking_aio_notifiers = false;
3824 }
3825 
3826 void bdrv_set_aio_context(BlockDriverState *bs, AioContext *new_context)
3827 {
3828     bdrv_drain(bs); /* ensure there are no in-flight requests */
3829 
3830     bdrv_detach_aio_context(bs);
3831 
3832     /* This function executes in the old AioContext so acquire the new one in
3833      * case it runs in a different thread.
3834      */
3835     aio_context_acquire(new_context);
3836     bdrv_attach_aio_context(bs, new_context);
3837     aio_context_release(new_context);
3838 }
3839 
3840 void bdrv_add_aio_context_notifier(BlockDriverState *bs,
3841         void (*attached_aio_context)(AioContext *new_context, void *opaque),
3842         void (*detach_aio_context)(void *opaque), void *opaque)
3843 {
3844     BdrvAioNotifier *ban = g_new(BdrvAioNotifier, 1);
3845     *ban = (BdrvAioNotifier){
3846         .attached_aio_context = attached_aio_context,
3847         .detach_aio_context   = detach_aio_context,
3848         .opaque               = opaque
3849     };
3850 
3851     QLIST_INSERT_HEAD(&bs->aio_notifiers, ban, list);
3852 }
3853 
3854 void bdrv_remove_aio_context_notifier(BlockDriverState *bs,
3855                                       void (*attached_aio_context)(AioContext *,
3856                                                                    void *),
3857                                       void (*detach_aio_context)(void *),
3858                                       void *opaque)
3859 {
3860     BdrvAioNotifier *ban, *ban_next;
3861 
3862     QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_next) {
3863         if (ban->attached_aio_context == attached_aio_context &&
3864             ban->detach_aio_context   == detach_aio_context   &&
3865             ban->opaque               == opaque               &&
3866             ban->deleted              == false)
3867         {
3868             if (bs->walking_aio_notifiers) {
3869                 ban->deleted = true;
3870             } else {
3871                 bdrv_do_remove_aio_context_notifier(ban);
3872             }
3873             return;
3874         }
3875     }
3876 
3877     abort();
3878 }
3879 
3880 int bdrv_amend_options(BlockDriverState *bs, QemuOpts *opts,
3881                        BlockDriverAmendStatusCB *status_cb, void *cb_opaque)
3882 {
3883     if (!bs->drv->bdrv_amend_options) {
3884         return -ENOTSUP;
3885     }
3886     return bs->drv->bdrv_amend_options(bs, opts, status_cb, cb_opaque);
3887 }
3888 
3889 /* This function will be called by the bdrv_recurse_is_first_non_filter method
3890  * of block filter and by bdrv_is_first_non_filter.
3891  * It is used to test if the given bs is the candidate or recurse more in the
3892  * node graph.
3893  */
3894 bool bdrv_recurse_is_first_non_filter(BlockDriverState *bs,
3895                                       BlockDriverState *candidate)
3896 {
3897     /* return false if basic checks fails */
3898     if (!bs || !bs->drv) {
3899         return false;
3900     }
3901 
3902     /* the code reached a non block filter driver -> check if the bs is
3903      * the same as the candidate. It's the recursion termination condition.
3904      */
3905     if (!bs->drv->is_filter) {
3906         return bs == candidate;
3907     }
3908     /* Down this path the driver is a block filter driver */
3909 
3910     /* If the block filter recursion method is defined use it to recurse down
3911      * the node graph.
3912      */
3913     if (bs->drv->bdrv_recurse_is_first_non_filter) {
3914         return bs->drv->bdrv_recurse_is_first_non_filter(bs, candidate);
3915     }
3916 
3917     /* the driver is a block filter but don't allow to recurse -> return false
3918      */
3919     return false;
3920 }
3921 
3922 /* This function checks if the candidate is the first non filter bs down it's
3923  * bs chain. Since we don't have pointers to parents it explore all bs chains
3924  * from the top. Some filters can choose not to pass down the recursion.
3925  */
3926 bool bdrv_is_first_non_filter(BlockDriverState *candidate)
3927 {
3928     BlockDriverState *bs;
3929     BdrvNextIterator it;
3930 
3931     /* walk down the bs forest recursively */
3932     for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
3933         bool perm;
3934 
3935         /* try to recurse in this top level bs */
3936         perm = bdrv_recurse_is_first_non_filter(bs, candidate);
3937 
3938         /* candidate is the first non filter */
3939         if (perm) {
3940             return true;
3941         }
3942     }
3943 
3944     return false;
3945 }
3946 
3947 BlockDriverState *check_to_replace_node(BlockDriverState *parent_bs,
3948                                         const char *node_name, Error **errp)
3949 {
3950     BlockDriverState *to_replace_bs = bdrv_find_node(node_name);
3951     AioContext *aio_context;
3952 
3953     if (!to_replace_bs) {
3954         error_setg(errp, "Node name '%s' not found", node_name);
3955         return NULL;
3956     }
3957 
3958     aio_context = bdrv_get_aio_context(to_replace_bs);
3959     aio_context_acquire(aio_context);
3960 
3961     if (bdrv_op_is_blocked(to_replace_bs, BLOCK_OP_TYPE_REPLACE, errp)) {
3962         to_replace_bs = NULL;
3963         goto out;
3964     }
3965 
3966     /* We don't want arbitrary node of the BDS chain to be replaced only the top
3967      * most non filter in order to prevent data corruption.
3968      * Another benefit is that this tests exclude backing files which are
3969      * blocked by the backing blockers.
3970      */
3971     if (!bdrv_recurse_is_first_non_filter(parent_bs, to_replace_bs)) {
3972         error_setg(errp, "Only top most non filter can be replaced");
3973         to_replace_bs = NULL;
3974         goto out;
3975     }
3976 
3977 out:
3978     aio_context_release(aio_context);
3979     return to_replace_bs;
3980 }
3981 
3982 static bool append_open_options(QDict *d, BlockDriverState *bs)
3983 {
3984     const QDictEntry *entry;
3985     QemuOptDesc *desc;
3986     BdrvChild *child;
3987     bool found_any = false;
3988     const char *p;
3989 
3990     for (entry = qdict_first(bs->options); entry;
3991          entry = qdict_next(bs->options, entry))
3992     {
3993         /* Exclude options for children */
3994         QLIST_FOREACH(child, &bs->children, next) {
3995             if (strstart(qdict_entry_key(entry), child->name, &p)
3996                 && (!*p || *p == '.'))
3997             {
3998                 break;
3999             }
4000         }
4001         if (child) {
4002             continue;
4003         }
4004 
4005         /* And exclude all non-driver-specific options */
4006         for (desc = bdrv_runtime_opts.desc; desc->name; desc++) {
4007             if (!strcmp(qdict_entry_key(entry), desc->name)) {
4008                 break;
4009             }
4010         }
4011         if (desc->name) {
4012             continue;
4013         }
4014 
4015         qobject_incref(qdict_entry_value(entry));
4016         qdict_put_obj(d, qdict_entry_key(entry), qdict_entry_value(entry));
4017         found_any = true;
4018     }
4019 
4020     return found_any;
4021 }
4022 
4023 /* Updates the following BDS fields:
4024  *  - exact_filename: A filename which may be used for opening a block device
4025  *                    which (mostly) equals the given BDS (even without any
4026  *                    other options; so reading and writing must return the same
4027  *                    results, but caching etc. may be different)
4028  *  - full_open_options: Options which, when given when opening a block device
4029  *                       (without a filename), result in a BDS (mostly)
4030  *                       equalling the given one
4031  *  - filename: If exact_filename is set, it is copied here. Otherwise,
4032  *              full_open_options is converted to a JSON object, prefixed with
4033  *              "json:" (for use through the JSON pseudo protocol) and put here.
4034  */
4035 void bdrv_refresh_filename(BlockDriverState *bs)
4036 {
4037     BlockDriver *drv = bs->drv;
4038     QDict *opts;
4039 
4040     if (!drv) {
4041         return;
4042     }
4043 
4044     /* This BDS's file name will most probably depend on its file's name, so
4045      * refresh that first */
4046     if (bs->file) {
4047         bdrv_refresh_filename(bs->file->bs);
4048     }
4049 
4050     if (drv->bdrv_refresh_filename) {
4051         /* Obsolete information is of no use here, so drop the old file name
4052          * information before refreshing it */
4053         bs->exact_filename[0] = '\0';
4054         if (bs->full_open_options) {
4055             QDECREF(bs->full_open_options);
4056             bs->full_open_options = NULL;
4057         }
4058 
4059         opts = qdict_new();
4060         append_open_options(opts, bs);
4061         drv->bdrv_refresh_filename(bs, opts);
4062         QDECREF(opts);
4063     } else if (bs->file) {
4064         /* Try to reconstruct valid information from the underlying file */
4065         bool has_open_options;
4066 
4067         bs->exact_filename[0] = '\0';
4068         if (bs->full_open_options) {
4069             QDECREF(bs->full_open_options);
4070             bs->full_open_options = NULL;
4071         }
4072 
4073         opts = qdict_new();
4074         has_open_options = append_open_options(opts, bs);
4075 
4076         /* If no specific options have been given for this BDS, the filename of
4077          * the underlying file should suffice for this one as well */
4078         if (bs->file->bs->exact_filename[0] && !has_open_options) {
4079             strcpy(bs->exact_filename, bs->file->bs->exact_filename);
4080         }
4081         /* Reconstructing the full options QDict is simple for most format block
4082          * drivers, as long as the full options are known for the underlying
4083          * file BDS. The full options QDict of that file BDS should somehow
4084          * contain a representation of the filename, therefore the following
4085          * suffices without querying the (exact_)filename of this BDS. */
4086         if (bs->file->bs->full_open_options) {
4087             qdict_put_obj(opts, "driver",
4088                           QOBJECT(qstring_from_str(drv->format_name)));
4089             QINCREF(bs->file->bs->full_open_options);
4090             qdict_put_obj(opts, "file",
4091                           QOBJECT(bs->file->bs->full_open_options));
4092 
4093             bs->full_open_options = opts;
4094         } else {
4095             QDECREF(opts);
4096         }
4097     } else if (!bs->full_open_options && qdict_size(bs->options)) {
4098         /* There is no underlying file BDS (at least referenced by BDS.file),
4099          * so the full options QDict should be equal to the options given
4100          * specifically for this block device when it was opened (plus the
4101          * driver specification).
4102          * Because those options don't change, there is no need to update
4103          * full_open_options when it's already set. */
4104 
4105         opts = qdict_new();
4106         append_open_options(opts, bs);
4107         qdict_put_obj(opts, "driver",
4108                       QOBJECT(qstring_from_str(drv->format_name)));
4109 
4110         if (bs->exact_filename[0]) {
4111             /* This may not work for all block protocol drivers (some may
4112              * require this filename to be parsed), but we have to find some
4113              * default solution here, so just include it. If some block driver
4114              * does not support pure options without any filename at all or
4115              * needs some special format of the options QDict, it needs to
4116              * implement the driver-specific bdrv_refresh_filename() function.
4117              */
4118             qdict_put_obj(opts, "filename",
4119                           QOBJECT(qstring_from_str(bs->exact_filename)));
4120         }
4121 
4122         bs->full_open_options = opts;
4123     }
4124 
4125     if (bs->exact_filename[0]) {
4126         pstrcpy(bs->filename, sizeof(bs->filename), bs->exact_filename);
4127     } else if (bs->full_open_options) {
4128         QString *json = qobject_to_json(QOBJECT(bs->full_open_options));
4129         snprintf(bs->filename, sizeof(bs->filename), "json:%s",
4130                  qstring_get_str(json));
4131         QDECREF(json);
4132     }
4133 }
4134 
4135 /*
4136  * Hot add/remove a BDS's child. So the user can take a child offline when
4137  * it is broken and take a new child online
4138  */
4139 void bdrv_add_child(BlockDriverState *parent_bs, BlockDriverState *child_bs,
4140                     Error **errp)
4141 {
4142 
4143     if (!parent_bs->drv || !parent_bs->drv->bdrv_add_child) {
4144         error_setg(errp, "The node %s does not support adding a child",
4145                    bdrv_get_device_or_node_name(parent_bs));
4146         return;
4147     }
4148 
4149     if (!QLIST_EMPTY(&child_bs->parents)) {
4150         error_setg(errp, "The node %s already has a parent",
4151                    child_bs->node_name);
4152         return;
4153     }
4154 
4155     parent_bs->drv->bdrv_add_child(parent_bs, child_bs, errp);
4156 }
4157 
4158 void bdrv_del_child(BlockDriverState *parent_bs, BdrvChild *child, Error **errp)
4159 {
4160     BdrvChild *tmp;
4161 
4162     if (!parent_bs->drv || !parent_bs->drv->bdrv_del_child) {
4163         error_setg(errp, "The node %s does not support removing a child",
4164                    bdrv_get_device_or_node_name(parent_bs));
4165         return;
4166     }
4167 
4168     QLIST_FOREACH(tmp, &parent_bs->children, next) {
4169         if (tmp == child) {
4170             break;
4171         }
4172     }
4173 
4174     if (!tmp) {
4175         error_setg(errp, "The node %s does not have a child named %s",
4176                    bdrv_get_device_or_node_name(parent_bs),
4177                    bdrv_get_device_or_node_name(child->bs));
4178         return;
4179     }
4180 
4181     parent_bs->drv->bdrv_del_child(parent_bs, child, errp);
4182 }
4183